« back to all changes in this revision
Viewing changes to sys/ufs/ffs/ffs_softdep.c
- browse files at revision 12
- compare with another revision
- download diff
- download tarball
- view history from revision 12
- Committer: Package Import Robot
- Author(s): Robert Millan
- Date: 2013-11-29 14:21:12 UTC
- mfrom: (11.1.1 experimental)
- Revision ID: package-import@ubuntu.com-20131129142112-5tz5g7a4b6prb0dt
Tags: 9.2-2
* Avoid kfreebsd-kernel-headers versions prior to ino_t fix.
* Remove gratuitous dependency on libtermcap / libtinfo.
* Remove gratuitous dependency on libtermcap / libtinfo.
- files added:
- .pc/fsck_getopt.diff
- .pc/fsck_getopt.diff/sbin
- .pc/fsck_getopt.diff/sbin/fsck_ffs
- .pc/makefiles.diff
- .pc/makefiles.diff/lib
- .pc/makefiles.diff/lib/libufs
- .pc/makefiles.diff/sbin
- .pc/makefiles.diff/sbin/badsect
- .pc/makefiles.diff/sbin/dumpfs
- .pc/makefiles.diff/sbin/ffsinfo
- .pc/makefiles.diff/sbin/fsck_ffs
- .pc/makefiles.diff/sbin/fsdb
- .pc/makefiles.diff/sbin/growfs
- .pc/makefiles.diff/sbin/newfs
- .pc/makefiles.diff/sbin/tunefs
- files removed:
- .pc/00_include.patch
- .pc/00_include.patch/sys
- .pc/00_include.patch/sys/sys
- .pc/00_libport.patch
- .pc/00_libport.patch/lib
- .pc/00_libport.patch/lib/port
- .pc/00_mount.patch
- .pc/00_mount.patch/sys
- .pc/00_mount.patch/sys/sys
- .pc/00_param.patch
- .pc/00_param.patch/sys
- .pc/00_param.patch/sys/sys
- .pc/00_portable_ufs_types.patch
- .pc/00_portable_ufs_types.patch/sys
- .pc/00_portable_ufs_types.patch/sys/ufs
- .pc/00_portable_ufs_types.patch/sys/ufs/ffs
- .pc/01_libufs.patch
- .pc/01_libufs.patch/lib
- .pc/01_libufs.patch/lib/libufs
- .pc/01_libufs.patch/sys
- .pc/01_libufs.patch/sys/ufs
- .pc/01_libufs.patch/sys/ufs/ffs
- .pc/01_libufs.patch/sys/ufs/ufs
- .pc/02_bsdlabel.ufs.patch
- .pc/02_bsdlabel.ufs.patch/sbin
- .pc/02_bsdlabel.ufs.patch/sbin/bsdlabel
- .pc/02_dumpfs.ufs.patch
- .pc/02_dumpfs.ufs.patch/sbin
- .pc/02_dumpfs.ufs.patch/sbin/dumpfs
- .pc/02_fsck.ufs.patch
- .pc/02_fsck.ufs.patch/sbin
- .pc/02_fsck.ufs.patch/sbin/fsck_ffs
- .pc/02_fsdb.ufs.patch
- .pc/02_fsdb.ufs.patch/sbin
- .pc/02_fsdb.ufs.patch/sbin/fsdb
- .pc/02_growfs.ufs.patch
- .pc/02_growfs.ufs.patch/sbin
- .pc/02_growfs.ufs.patch/sbin/growfs
- .pc/02_mkfs.ufs.patch
- .pc/02_mkfs.ufs.patch/sbin
- .pc/02_mkfs.ufs.patch/sbin/newfs
- .pc/02_tunefs.ufs.patch
- .pc/02_tunefs.ufs.patch/sbin
- .pc/02_tunefs.ufs.patch/sbin/tunefs
- .pc/03_ufsmount.patch
- .pc/03_ufsmount.patch/sys
- .pc/03_ufsmount.patch/sys/ufs
- .pc/03_ufsmount.patch/sys/ufs/ufs
- lib/port
- sbin/bsdlabel
- sbin/dump
- sbin/fsck_ffs/SMM.doc
- sbin/sunlabel
- sys/sys
- files modified:
- .pc/04_avoid_diocgdelete.patch/lib/libufs/block.c
added
removed
sys/ufs/ffs/ffs_softdep.c
1
1
/*-
2
* Copyright 1998, 2000 Marshall Kirk McKusick. All Rights Reserved.
2
* Copyright 1998, 2000 Marshall Kirk McKusick.
3
* Copyright 2009, 2010 Jeffrey W. Roberson <jeff@FreeBSD.org>
4
* All rights reserved.
3
5
*
4
6
* The soft updates code is derived from the appendix of a University
5
7
* of Michigan technical report (Gregory R. Ganger and Yale N. Patt,
23
25
* notice, this list of conditions and the following disclaimer in the
24
26
* documentation and/or other materials provided with the distribution.
25
27
*
26
* THIS SOFTWARE IS PROVIDED BY MARSHALL KIRK MCKUSICK ``AS IS'' AND ANY
27
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
28
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
29
* DISCLAIMED. IN NO EVENT SHALL MARSHALL KIRK MCKUSICK BE LIABLE FOR
30
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36
* SUCH DAMAGE.
28
* THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
29
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
30
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
31
* IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
32
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
33
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
34
* OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
35
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR
36
* TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
37
* USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37
38
*
38
39
* from: @(#)ffs_softdep.c 9.59 (McKusick) 6/21/00
39
40
*/
42
43
__FBSDID("$FreeBSD$");
43
44
44
45
#include "opt_ffs.h"
46
#include "opt_quota.h"
45
47
#include "opt_ddb.h"
46
48
47
49
/*
58
60
#include <sys/buf.h>
59
61
#include <sys/kdb.h>
60
62
#include <sys/kthread.h>
63
#include <sys/ktr.h>
64
#include <sys/limits.h>
61
65
#include <sys/lock.h>
62
66
#include <sys/malloc.h>
63
67
#include <sys/mount.h>
64
68
#include <sys/mutex.h>
69
#include <sys/namei.h>
70
#include <sys/priv.h>
65
71
#include <sys/proc.h>
66
72
#include <sys/stat.h>
67
73
#include <sys/sysctl.h>
68
74
#include <sys/syslog.h>
69
75
#include <sys/vnode.h>
70
76
#include <sys/conf.h>
77
71
78
#include <ufs/ufs/dir.h>
72
79
#include <ufs/ufs/extattr.h>
73
80
#include <ufs/ufs/quota.h>
79
86
#include <ufs/ufs/ufs_extern.h>
80
87
81
88
#include <vm/vm.h>
89
#include <vm/vm_extern.h>
90
#include <vm/vm_object.h>
91
92
#include <geom/geom.h>
82
93
83
94
#include <ddb/ddb.h>
84
95
96
#define KTR_SUJ 0 /* Define to KTR_SPARE. */
97
85
98
#ifndef SOFTUPDATES
86
99
87
100
int
120
133
}
121
134
122
135
void
123
softdep_setup_inomapdep(bp, ip, newinum)
136
softdep_unmount(mp)
137
struct mount *mp;
138
{
139
140
}
141
142
void
143
softdep_setup_sbupdate(ump, fs, bp)
144
struct ufsmount *ump;
145
struct fs *fs;
146
struct buf *bp;
147
{
148
}
149
150
void
151
softdep_setup_inomapdep(bp, ip, newinum, mode)
124
152
struct buf *bp;
125
153
struct inode *ip;
126
154
ino_t newinum;
155
int mode;
127
156
{
128
157
129
158
panic("softdep_setup_inomapdep called");
130
159
}
131
160
132
161
void
133
softdep_setup_blkmapdep(bp, mp, newblkno)
162
softdep_setup_blkmapdep(bp, mp, newblkno, frags, oldfrags)
134
163
struct buf *bp;
135
164
struct mount *mp;
136
165
ufs2_daddr_t newblkno;
166
int frags;
167
int oldfrags;
137
168
{
138
169
139
170
panic("softdep_setup_blkmapdep called");
194
225
}
195
226
196
227
void
228
softdep_journal_freeblocks(ip, cred, length, flags)
229
struct inode *ip;
230
struct ucred *cred;
231
off_t length;
232
int flags;
233
{
234
235
panic("softdep_journal_freeblocks called");
236
}
237
238
void
239
softdep_journal_fsync(ip)
240
struct inode *ip;
241
{
242
243
panic("softdep_journal_fsync called");
244
}
245
246
void
197
247
softdep_setup_freeblocks(ip, length, flags)
198
248
struct inode *ip;
199
249
off_t length;
227
277
}
228
278
229
279
void
230
softdep_change_directoryentry_offset(dp, base, oldloc, newloc, entrysize)
280
softdep_change_directoryentry_offset(bp, dp, base, oldloc, newloc, entrysize)
281
struct buf *bp;
231
282
struct inode *dp;
232
283
caddr_t base;
233
284
caddr_t oldloc;
262
313
}
263
314
264
315
void
316
softdep_setup_blkfree(mp, bp, blkno, frags, wkhd)
317
struct mount *mp;
318
struct buf *bp;
319
ufs2_daddr_t blkno;
320
int frags;
321
struct workhead *wkhd;
322
{
323
324
panic("%s called", __FUNCTION__);
325
}
326
327
void
328
softdep_setup_inofree(mp, bp, ino, wkhd)
329
struct mount *mp;
330
struct buf *bp;
331
ino_t ino;
332
struct workhead *wkhd;
333
{
334
335
panic("%s called", __FUNCTION__);
336
}
337
338
void
339
softdep_setup_unlink(dp, ip)
340
struct inode *dp;
341
struct inode *ip;
342
{
343
344
panic("%s called", __FUNCTION__);
345
}
346
347
void
348
softdep_setup_link(dp, ip)
349
struct inode *dp;
350
struct inode *ip;
351
{
352
353
panic("%s called", __FUNCTION__);
354
}
355
356
void
357
softdep_revert_link(dp, ip)
358
struct inode *dp;
359
struct inode *ip;
360
{
361
362
panic("%s called", __FUNCTION__);
363
}
364
365
void
366
softdep_setup_rmdir(dp, ip)
367
struct inode *dp;
368
struct inode *ip;
369
{
370
371
panic("%s called", __FUNCTION__);
372
}
373
374
void
375
softdep_revert_rmdir(dp, ip)
376
struct inode *dp;
377
struct inode *ip;
378
{
379
380
panic("%s called", __FUNCTION__);
381
}
382
383
void
384
softdep_setup_create(dp, ip)
385
struct inode *dp;
386
struct inode *ip;
387
{
388
389
panic("%s called", __FUNCTION__);
390
}
391
392
void
393
softdep_revert_create(dp, ip)
394
struct inode *dp;
395
struct inode *ip;
396
{
397
398
panic("%s called", __FUNCTION__);
399
}
400
401
void
402
softdep_setup_mkdir(dp, ip)
403
struct inode *dp;
404
struct inode *ip;
405
{
406
407
panic("%s called", __FUNCTION__);
408
}
409
410
void
411
softdep_revert_mkdir(dp, ip)
412
struct inode *dp;
413
struct inode *ip;
414
{
415
416
panic("%s called", __FUNCTION__);
417
}
418
419
void
420
softdep_setup_dotdot_link(dp, ip)
421
struct inode *dp;
422
struct inode *ip;
423
{
424
425
panic("%s called", __FUNCTION__);
426
}
427
428
int
429
softdep_prealloc(vp, waitok)
430
struct vnode *vp;
431
int waitok;
432
{
433
434
panic("%s called", __FUNCTION__);
435
436
return (0);
437
}
438
439
int
440
softdep_journal_lookup(mp, vpp)
441
struct mount *mp;
442
struct vnode **vpp;
443
{
444
445
return (ENOENT);
446
}
447
448
void
265
449
softdep_change_linkcnt(ip)
266
450
struct inode *ip;
267
451
{
322
506
}
323
507
324
508
int
509
softdep_sync_buf(struct vnode *vp, struct buf *bp, int waitfor)
510
{
511
512
return (0);
513
}
514
515
int
325
516
softdep_slowdown(vp)
326
517
struct vnode *vp;
327
518
{
338
529
}
339
530
340
531
int
341
softdep_request_cleanup(fs, vp)
532
softdep_request_cleanup(fs, vp, cred, resource)
342
533
struct fs *fs;
343
534
struct vnode *vp;
535
struct ucred *cred;
536
int resource;
344
537
{
345
538
346
539
return (0);
398
591
*softdepactiveaccp = 0;
399
592
}
400
593
594
void
595
softdep_buf_append(bp, wkhd)
596
struct buf *bp;
597
struct workhead *wkhd;
598
{
599
600
panic("softdep_buf_appendwork called");
601
}
602
603
void
604
softdep_inode_append(ip, cred, wkhd)
605
struct inode *ip;
606
struct ucred *cred;
607
struct workhead *wkhd;
608
{
609
610
panic("softdep_inode_appendwork called");
611
}
612
613
void
614
softdep_freework(wkhd)
615
struct workhead *wkhd;
616
{
617
618
panic("softdep_freework called");
619
}
620
401
621
#else
622
623
FEATURE(softupdates, "FFS soft-updates support");
624
402
625
/*
403
626
* These definitions need to be adapted to the system to which
404
627
* this file is being ported.
405
628
*/
406
/*
407
* malloc types defined for the softdep system.
408
*/
409
static MALLOC_DEFINE(M_PAGEDEP, "pagedep","File page dependencies");
410
static MALLOC_DEFINE(M_INODEDEP, "inodedep","Inode dependencies");
411
static MALLOC_DEFINE(M_NEWBLK, "newblk","New block allocation");
412
static MALLOC_DEFINE(M_BMSAFEMAP, "bmsafemap","Block or frag allocated from cyl group map");
413
static MALLOC_DEFINE(M_ALLOCDIRECT, "allocdirect","Block or frag dependency for an inode");
414
static MALLOC_DEFINE(M_INDIRDEP, "indirdep","Indirect block dependencies");
415
static MALLOC_DEFINE(M_ALLOCINDIR, "allocindir","Block dependency for an indirect block");
416
static MALLOC_DEFINE(M_FREEFRAG, "freefrag","Previously used frag for an inode");
417
static MALLOC_DEFINE(M_FREEBLKS, "freeblks","Blocks freed from an inode");
418
static MALLOC_DEFINE(M_FREEFILE, "freefile","Inode deallocated");
419
static MALLOC_DEFINE(M_DIRADD, "diradd","New directory entry");
420
static MALLOC_DEFINE(M_MKDIR, "mkdir","New directory");
421
static MALLOC_DEFINE(M_DIRREM, "dirrem","Directory entry deleted");
422
static MALLOC_DEFINE(M_NEWDIRBLK, "newdirblk","Unclaimed new directory block");
423
static MALLOC_DEFINE(M_SAVEDINO, "savedino","Saved inodes");
424
629
425
#define M_SOFTDEP_FLAGS (M_WAITOK | M_USE_RESERVE)
630
#define M_SOFTDEP_FLAGS (M_WAITOK)
426
631
427
632
#define D_PAGEDEP 0
428
633
#define D_INODEDEP 1
429
#define D_NEWBLK 2
430
#define D_BMSAFEMAP 3
634
#define D_BMSAFEMAP 2
635
#define D_NEWBLK 3
431
636
#define D_ALLOCDIRECT 4
432
637
#define D_INDIRDEP 5
433
638
#define D_ALLOCINDIR 6
438
643
#define D_MKDIR 11
439
644
#define D_DIRREM 12
440
645
#define D_NEWDIRBLK 13
441
#define D_LAST D_NEWDIRBLK
646
#define D_FREEWORK 14
647
#define D_FREEDEP 15
648
#define D_JADDREF 16
649
#define D_JREMREF 17
650
#define D_JMVREF 18
651
#define D_JNEWBLK 19
652
#define D_JFREEBLK 20
653
#define D_JFREEFRAG 21
654
#define D_JSEG 22
655
#define D_JSEGDEP 23
656
#define D_SBDEP 24
657
#define D_JTRUNC 25
658
#define D_JFSYNC 26
659
#define D_SENTINEL 27
660
#define D_LAST D_SENTINEL
661
662
unsigned long dep_current[D_LAST + 1];
663
unsigned long dep_total[D_LAST + 1];
664
unsigned long dep_write[D_LAST + 1];
665
666
667
static SYSCTL_NODE(_debug, OID_AUTO, softdep, CTLFLAG_RW, 0,
668
"soft updates stats");
669
static SYSCTL_NODE(_debug_softdep, OID_AUTO, total, CTLFLAG_RW, 0,
670
"total dependencies allocated");
671
static SYSCTL_NODE(_debug_softdep, OID_AUTO, current, CTLFLAG_RW, 0,
672
"current dependencies allocated");
673
static SYSCTL_NODE(_debug_softdep, OID_AUTO, write, CTLFLAG_RW, 0,
674
"current dependencies written");
675
676
#define SOFTDEP_TYPE(type, str, long) \
677
static MALLOC_DEFINE(M_ ## type, #str, long); \
678
SYSCTL_ULONG(_debug_softdep_total, OID_AUTO, str, CTLFLAG_RD, \
679
&dep_total[D_ ## type], 0, ""); \
680
SYSCTL_ULONG(_debug_softdep_current, OID_AUTO, str, CTLFLAG_RD, \
681
&dep_current[D_ ## type], 0, ""); \
682
SYSCTL_ULONG(_debug_softdep_write, OID_AUTO, str, CTLFLAG_RD, \
683
&dep_write[D_ ## type], 0, "");
684
685
SOFTDEP_TYPE(PAGEDEP, pagedep, "File page dependencies");
686
SOFTDEP_TYPE(INODEDEP, inodedep, "Inode dependencies");
687
SOFTDEP_TYPE(BMSAFEMAP, bmsafemap,
688
"Block or frag allocated from cyl group map");
689
SOFTDEP_TYPE(NEWBLK, newblk, "New block or frag allocation dependency");
690
SOFTDEP_TYPE(ALLOCDIRECT, allocdirect, "Block or frag dependency for an inode");
691
SOFTDEP_TYPE(INDIRDEP, indirdep, "Indirect block dependencies");
692
SOFTDEP_TYPE(ALLOCINDIR, allocindir, "Block dependency for an indirect block");
693
SOFTDEP_TYPE(FREEFRAG, freefrag, "Previously used frag for an inode");
694
SOFTDEP_TYPE(FREEBLKS, freeblks, "Blocks freed from an inode");
695
SOFTDEP_TYPE(FREEFILE, freefile, "Inode deallocated");
696
SOFTDEP_TYPE(DIRADD, diradd, "New directory entry");
697
SOFTDEP_TYPE(MKDIR, mkdir, "New directory");
698
SOFTDEP_TYPE(DIRREM, dirrem, "Directory entry deleted");
699
SOFTDEP_TYPE(NEWDIRBLK, newdirblk, "Unclaimed new directory block");
700
SOFTDEP_TYPE(FREEWORK, freework, "free an inode block");
701
SOFTDEP_TYPE(FREEDEP, freedep, "track a block free");
702
SOFTDEP_TYPE(JADDREF, jaddref, "Journal inode ref add");
703
SOFTDEP_TYPE(JREMREF, jremref, "Journal inode ref remove");
704
SOFTDEP_TYPE(JMVREF, jmvref, "Journal inode ref move");
705
SOFTDEP_TYPE(JNEWBLK, jnewblk, "Journal new block");
706
SOFTDEP_TYPE(JFREEBLK, jfreeblk, "Journal free block");
707
SOFTDEP_TYPE(JFREEFRAG, jfreefrag, "Journal free frag");
708
SOFTDEP_TYPE(JSEG, jseg, "Journal segment");
709
SOFTDEP_TYPE(JSEGDEP, jsegdep, "Journal segment complete");
710
SOFTDEP_TYPE(SBDEP, sbdep, "Superblock write dependency");
711
SOFTDEP_TYPE(JTRUNC, jtrunc, "Journal inode truncation");
712
SOFTDEP_TYPE(JFSYNC, jfsync, "Journal fsync complete");
713
714
static MALLOC_DEFINE(M_SENTINEL, "sentinel", "Worklist sentinel");
715
716
static MALLOC_DEFINE(M_SAVEDINO, "savedino", "Saved inodes");
717
static MALLOC_DEFINE(M_JBLOCKS, "jblocks", "Journal block locations");
442
718
443
719
/*
444
720
* translate from workitem type to memory type
447
723
static struct malloc_type *memtype[] = {
448
724
M_PAGEDEP,
449
725
M_INODEDEP,
726
M_BMSAFEMAP,
450
727
M_NEWBLK,
451
M_BMSAFEMAP,
452
728
M_ALLOCDIRECT,
453
729
M_INDIRDEP,
454
730
M_ALLOCINDIR,
458
734
M_DIRADD,
459
735
M_MKDIR,
460
736
M_DIRREM,
461
M_NEWDIRBLK
737
M_NEWDIRBLK,
738
M_FREEWORK,
739
M_FREEDEP,
740
M_JADDREF,
741
M_JREMREF,
742
M_JMVREF,
743
M_JNEWBLK,
744
M_JFREEBLK,
745
M_JFREEFRAG,
746
M_JSEG,
747
M_JSEGDEP,
748
M_SBDEP,
749
M_JTRUNC,
750
M_JFSYNC,
751
M_SENTINEL
462
752
};
463
753
754
static LIST_HEAD(mkdirlist, mkdir) mkdirlisthd;
755
464
756
#define DtoM(type) (memtype[type])
465
757
466
758
/*
467
759
* Names of malloc types.
468
760
*/
469
761
#define TYPENAME(type) \
470
((unsigned)(type) < D_LAST ? memtype[type]->ks_shortdesc : "???")
762
((unsigned)(type) <= D_LAST ? memtype[type]->ks_shortdesc : "???")
471
763
/*
472
764
* End system adaptation definitions.
473
765
*/
474
766
767
#define DOTDOT_OFFSET offsetof(struct dirtemplate, dotdot_ino)
768
#define DOT_OFFSET offsetof(struct dirtemplate, dot_ino)
769
475
770
/*
476
771
* Forward declarations.
477
772
*/
478
773
struct inodedep_hashhead;
479
774
struct newblk_hashhead;
480
775
struct pagedep_hashhead;
776
struct bmsafemap_hashhead;
777
778
/*
779
* Private journaling structures.
780
*/
781
struct jblocks {
782
struct jseglst jb_segs; /* TAILQ of current segments. */
783
struct jseg *jb_writeseg; /* Next write to complete. */
784
struct jseg *jb_oldestseg; /* Oldest segment with valid entries. */
785
struct jextent *jb_extent; /* Extent array. */
786
uint64_t jb_nextseq; /* Next sequence number. */
787
uint64_t jb_oldestwrseq; /* Oldest written sequence number. */
788
uint8_t jb_needseg; /* Need a forced segment. */
789
uint8_t jb_suspended; /* Did journal suspend writes? */
790
int jb_avail; /* Available extents. */
791
int jb_used; /* Last used extent. */
792
int jb_head; /* Allocator head. */
793
int jb_off; /* Allocator extent offset. */
794
int jb_blocks; /* Total disk blocks covered. */
795
int jb_free; /* Total disk blocks free. */
796
int jb_min; /* Minimum free space. */
797
int jb_low; /* Low on space. */
798
int jb_age; /* Insertion time of oldest rec. */
799
};
800
801
struct jextent {
802
ufs2_daddr_t je_daddr; /* Disk block address. */
803
int je_blocks; /* Disk block count. */
804
};
481
805
482
806
/*
483
807
* Internal function prototypes.
487
811
static struct buf *getdirtybuf(struct buf *, struct mtx *, int);
488
812
static void clear_remove(struct thread *);
489
813
static void clear_inodedeps(struct thread *);
814
static void unlinked_inodedep(struct mount *, struct inodedep *);
815
static void clear_unlinked_inodedep(struct inodedep *);
816
static struct inodedep *first_unlinked_inodedep(struct ufsmount *);
490
817
static int flush_pagedep_deps(struct vnode *, struct mount *,
491
818
struct diraddhd *);
492
static int flush_inodedep_deps(struct mount *, ino_t);
819
static int free_pagedep(struct pagedep *);
820
static int flush_newblk_dep(struct vnode *, struct mount *, ufs_lbn_t);
821
static int flush_inodedep_deps(struct vnode *, struct mount *, ino_t);
493
822
static int flush_deplist(struct allocdirectlst *, int, int *);
823
static int sync_cgs(struct mount *, int);
494
824
static int handle_written_filepage(struct pagedep *, struct buf *);
825
static int handle_written_sbdep(struct sbdep *, struct buf *);
826
static void initiate_write_sbdep(struct sbdep *);
495
827
static void diradd_inode_written(struct diradd *, struct inodedep *);
828
static int handle_written_indirdep(struct indirdep *, struct buf *,
829
struct buf**);
496
830
static int handle_written_inodeblock(struct inodedep *, struct buf *);
497
static void handle_allocdirect_partdone(struct allocdirect *);
831
static int jnewblk_rollforward(struct jnewblk *, struct fs *, struct cg *,
832
uint8_t *);
833
static int handle_written_bmsafemap(struct bmsafemap *, struct buf *);
834
static void handle_written_jaddref(struct jaddref *);
835
static void handle_written_jremref(struct jremref *);
836
static void handle_written_jseg(struct jseg *, struct buf *);
837
static void handle_written_jnewblk(struct jnewblk *);
838
static void handle_written_jblkdep(struct jblkdep *);
839
static void handle_written_jfreefrag(struct jfreefrag *);
840
static void complete_jseg(struct jseg *);
841
static void complete_jsegs(struct jseg *);
842
static void jseg_write(struct ufsmount *ump, struct jseg *, uint8_t *);
843
static void jaddref_write(struct jaddref *, struct jseg *, uint8_t *);
844
static void jremref_write(struct jremref *, struct jseg *, uint8_t *);
845
static void jmvref_write(struct jmvref *, struct jseg *, uint8_t *);
846
static void jtrunc_write(struct jtrunc *, struct jseg *, uint8_t *);
847
static void jfsync_write(struct jfsync *, struct jseg *, uint8_t *data);
848
static void jnewblk_write(struct jnewblk *, struct jseg *, uint8_t *);
849
static void jfreeblk_write(struct jfreeblk *, struct jseg *, uint8_t *);
850
static void jfreefrag_write(struct jfreefrag *, struct jseg *, uint8_t *);
851
static inline void inoref_write(struct inoref *, struct jseg *,
852
struct jrefrec *);
853
static void handle_allocdirect_partdone(struct allocdirect *,
854
struct workhead *);
855
static struct jnewblk *cancel_newblk(struct newblk *, struct worklist *,
856
struct workhead *);
857
static void indirdep_complete(struct indirdep *);
858
static int indirblk_lookup(struct mount *, ufs2_daddr_t);
859
static void indirblk_insert(struct freework *);
860
static void indirblk_remove(struct freework *);
498
861
static void handle_allocindir_partdone(struct allocindir *);
499
862
static void initiate_write_filepage(struct pagedep *, struct buf *);
863
static void initiate_write_indirdep(struct indirdep*, struct buf *);
500
864
static void handle_written_mkdir(struct mkdir *, int);
865
static int jnewblk_rollback(struct jnewblk *, struct fs *, struct cg *,
866
uint8_t *);
867
static void initiate_write_bmsafemap(struct bmsafemap *, struct buf *);
501
868
static void initiate_write_inodeblock_ufs1(struct inodedep *, struct buf *);
502
869
static void initiate_write_inodeblock_ufs2(struct inodedep *, struct buf *);
503
870
static void handle_workitem_freefile(struct freefile *);
504
static void handle_workitem_remove(struct dirrem *, struct vnode *);
871
static int handle_workitem_remove(struct dirrem *, int);
505
872
static struct dirrem *newdirrem(struct buf *, struct inode *,
506
873
struct inode *, int, struct dirrem **);
507
static void free_diradd(struct diradd *);
508
static void free_allocindir(struct allocindir *, struct inodedep *);
874
static struct indirdep *indirdep_lookup(struct mount *, struct inode *,
875
struct buf *);
876
static void cancel_indirdep(struct indirdep *, struct buf *,
877
struct freeblks *);
878
static void free_indirdep(struct indirdep *);
879
static void free_diradd(struct diradd *, struct workhead *);
880
static void merge_diradd(struct inodedep *, struct diradd *);
881
static void complete_diradd(struct diradd *);
882
static struct diradd *diradd_lookup(struct pagedep *, int);
883
static struct jremref *cancel_diradd_dotdot(struct inode *, struct dirrem *,
884
struct jremref *);
885
static struct jremref *cancel_mkdir_dotdot(struct inode *, struct dirrem *,
886
struct jremref *);
887
static void cancel_diradd(struct diradd *, struct dirrem *, struct jremref *,
888
struct jremref *, struct jremref *);
889
static void dirrem_journal(struct dirrem *, struct jremref *, struct jremref *,
890
struct jremref *);
891
static void cancel_allocindir(struct allocindir *, struct buf *bp,
892
struct freeblks *, int);
893
static int setup_trunc_indir(struct freeblks *, struct inode *,
894
ufs_lbn_t, ufs_lbn_t, ufs2_daddr_t);
895
static void complete_trunc_indir(struct freework *);
896
static void trunc_indirdep(struct indirdep *, struct freeblks *, struct buf *,
897
int);
898
static void complete_mkdir(struct mkdir *);
509
899
static void free_newdirblk(struct newdirblk *);
510
static int indir_trunc(struct freeblks *, ufs2_daddr_t, int, ufs_lbn_t,
511
ufs2_daddr_t *);
512
static void deallocate_dependencies(struct buf *, struct inodedep *);
513
static void free_allocdirect(struct allocdirectlst *,
514
struct allocdirect *, int);
900
static void free_jremref(struct jremref *);
901
static void free_jaddref(struct jaddref *);
902
static void free_jsegdep(struct jsegdep *);
903
static void free_jsegs(struct jblocks *);
904
static void rele_jseg(struct jseg *);
905
static void free_jseg(struct jseg *, struct jblocks *);
906
static void free_jnewblk(struct jnewblk *);
907
static void free_jblkdep(struct jblkdep *);
908
static void free_jfreefrag(struct jfreefrag *);
909
static void free_freedep(struct freedep *);
910
static void journal_jremref(struct dirrem *, struct jremref *,
911
struct inodedep *);
912
static void cancel_jnewblk(struct jnewblk *, struct workhead *);
913
static int cancel_jaddref(struct jaddref *, struct inodedep *,
914
struct workhead *);
915
static void cancel_jfreefrag(struct jfreefrag *);
916
static inline void setup_freedirect(struct freeblks *, struct inode *,
917
int, int);
918
static inline void setup_freeext(struct freeblks *, struct inode *, int, int);
919
static inline void setup_freeindir(struct freeblks *, struct inode *, int,
920
ufs_lbn_t, int);
921
static inline struct freeblks *newfreeblks(struct mount *, struct inode *);
922
static void freeblks_free(struct ufsmount *, struct freeblks *, int);
923
static void indir_trunc(struct freework *, ufs2_daddr_t, ufs_lbn_t);
924
ufs2_daddr_t blkcount(struct fs *, ufs2_daddr_t, off_t);
925
static int trunc_check_buf(struct buf *, int *, ufs_lbn_t, int, int);
926
static void trunc_dependencies(struct inode *, struct freeblks *, ufs_lbn_t,
927
int, int);
928
static void trunc_pages(struct inode *, off_t, ufs2_daddr_t, int);
929
static int cancel_pagedep(struct pagedep *, struct freeblks *, int);
930
static int deallocate_dependencies(struct buf *, struct freeblks *, int);
931
static void newblk_freefrag(struct newblk*);
932
static void free_newblk(struct newblk *);
933
static void cancel_allocdirect(struct allocdirectlst *,
934
struct allocdirect *, struct freeblks *);
515
935
static int check_inode_unwritten(struct inodedep *);
516
936
static int free_inodedep(struct inodedep *);
517
static void handle_workitem_freeblocks(struct freeblks *, int);
937
static void freework_freeblock(struct freework *);
938
static void freework_enqueue(struct freework *);
939
static int handle_workitem_freeblocks(struct freeblks *, int);
940
static int handle_complete_freeblocks(struct freeblks *, int);
941
static void handle_workitem_indirblk(struct freework *);
942
static void handle_written_freework(struct freework *);
518
943
static void merge_inode_lists(struct allocdirectlst *,struct allocdirectlst *);
519
static void setup_allocindir_phase2(struct buf *, struct inode *,
520
struct allocindir *);
944
static struct worklist *jnewblk_merge(struct worklist *, struct worklist *,
945
struct workhead *);
946
static struct freefrag *setup_allocindir_phase2(struct buf *, struct inode *,
947
struct inodedep *, struct allocindir *, ufs_lbn_t);
521
948
static struct allocindir *newallocindir(struct inode *, int, ufs2_daddr_t,
522
ufs2_daddr_t);
949
ufs2_daddr_t, ufs_lbn_t);
523
950
static void handle_workitem_freefrag(struct freefrag *);
524
static struct freefrag *newfreefrag(struct inode *, ufs2_daddr_t, long);
951
static struct freefrag *newfreefrag(struct inode *, ufs2_daddr_t, long,
952
ufs_lbn_t);
525
953
static void allocdirect_merge(struct allocdirectlst *,
526
954
struct allocdirect *, struct allocdirect *);
527
static struct bmsafemap *bmsafemap_lookup(struct mount *, struct buf *);
528
static int newblk_find(struct newblk_hashhead *, struct fs *, ufs2_daddr_t,
529
struct newblk **);
530
static int newblk_lookup(struct fs *, ufs2_daddr_t, int, struct newblk **);
955
static struct freefrag *allocindir_merge(struct allocindir *,
956
struct allocindir *);
957
static int bmsafemap_find(struct bmsafemap_hashhead *, struct mount *, int,
958
struct bmsafemap **);
959
static struct bmsafemap *bmsafemap_lookup(struct mount *, struct buf *,
960
int cg, struct bmsafemap *);
961
static int newblk_find(struct newblk_hashhead *, struct mount *, ufs2_daddr_t,
962
int, struct newblk **);
963
static int newblk_lookup(struct mount *, ufs2_daddr_t, int, struct newblk **);
531
964
static int inodedep_find(struct inodedep_hashhead *, struct fs *, ino_t,
532
965
struct inodedep **);
533
966
static int inodedep_lookup(struct mount *, ino_t, int, struct inodedep **);
534
static int pagedep_lookup(struct inode *, ufs_lbn_t, int, struct pagedep **);
967
static int pagedep_lookup(struct mount *, struct buf *bp, ino_t, ufs_lbn_t,
968
int, struct pagedep **);
535
969
static int pagedep_find(struct pagedep_hashhead *, ino_t, ufs_lbn_t,
536
970
struct mount *mp, int, struct pagedep **);
537
971
static void pause_timer(void *);
538
972
static int request_cleanup(struct mount *, int);
539
static int process_worklist_item(struct mount *, int);
540
static void add_to_worklist(struct worklist *);
973
static int process_worklist_item(struct mount *, int, int);
974
static void process_removes(struct vnode *);
975
static void process_truncates(struct vnode *);
976
static void jwork_move(struct workhead *, struct workhead *);
977
static void jwork_insert(struct workhead *, struct jsegdep *);
978
static void add_to_worklist(struct worklist *, int);
979
static void wake_worklist(struct worklist *);
980
static void wait_worklist(struct worklist *, char *);
981
static void remove_from_worklist(struct worklist *);
541
982
static void softdep_flush(void);
983
static void softdep_flushjournal(struct mount *);
542
984
static int softdep_speedup(void);
985
static void worklist_speedup(void);
986
static int journal_mount(struct mount *, struct fs *, struct ucred *);
987
static void journal_unmount(struct mount *);
988
static int journal_space(struct ufsmount *, int);
989
static void journal_suspend(struct ufsmount *);
990
static int journal_unsuspend(struct ufsmount *ump);
991
static void softdep_prelink(struct vnode *, struct vnode *);
992
static void add_to_journal(struct worklist *);
993
static void remove_from_journal(struct worklist *);
994
static void softdep_process_journal(struct mount *, struct worklist *, int);
995
static struct jremref *newjremref(struct dirrem *, struct inode *,
996
struct inode *ip, off_t, nlink_t);
997
static struct jaddref *newjaddref(struct inode *, ino_t, off_t, int16_t,
998
uint16_t);
999
static inline void newinoref(struct inoref *, ino_t, ino_t, off_t, nlink_t,
1000
uint16_t);
1001
static inline struct jsegdep *inoref_jseg(struct inoref *);
1002
static struct jmvref *newjmvref(struct inode *, ino_t, off_t, off_t);
1003
static struct jfreeblk *newjfreeblk(struct freeblks *, ufs_lbn_t,
1004
ufs2_daddr_t, int);
1005
static struct jtrunc *newjtrunc(struct freeblks *, off_t, int);
1006
static void move_newblock_dep(struct jaddref *, struct inodedep *);
1007
static void cancel_jfreeblk(struct freeblks *, ufs2_daddr_t);
1008
static struct jfreefrag *newjfreefrag(struct freefrag *, struct inode *,
1009
ufs2_daddr_t, long, ufs_lbn_t);
1010
static struct freework *newfreework(struct ufsmount *, struct freeblks *,
1011
struct freework *, ufs_lbn_t, ufs2_daddr_t, int, int, int);
1012
static int jwait(struct worklist *, int);
1013
static struct inodedep *inodedep_lookup_ip(struct inode *);
1014
static int bmsafemap_backgroundwrite(struct bmsafemap *, struct buf *);
1015
static struct freefile *handle_bufwait(struct inodedep *, struct workhead *);
1016
static void handle_jwork(struct workhead *);
1017
static struct mkdir *setup_newdir(struct diradd *, ino_t, ino_t, struct buf *,
1018
struct mkdir **);
1019
static struct jblocks *jblocks_create(void);
1020
static ufs2_daddr_t jblocks_alloc(struct jblocks *, int, int *);
1021
static void jblocks_free(struct jblocks *, struct mount *, int);
1022
static void jblocks_destroy(struct jblocks *);
1023
static void jblocks_add(struct jblocks *, ufs2_daddr_t, int);
543
1024
544
1025
/*
545
1026
* Exported softdep operations.
556
1037
#define ACQUIRE_LOCK(lk) mtx_lock(lk)
557
1038
#define FREE_LOCK(lk) mtx_unlock(lk)
558
1039
559
#define BUF_AREC(bp) ((bp)->b_lock.lock_object.lo_flags |= LO_RECURSABLE)
560
#define BUF_NOREC(bp) ((bp)->b_lock.lock_object.lo_flags &= ~LO_RECURSABLE)
1040
#define BUF_AREC(bp) lockallowrecurse(&(bp)->b_lock)
1041
#define BUF_NOREC(bp) lockdisablerecurse(&(bp)->b_lock)
561
1042
562
1043
/*
563
1044
* Worklist queue management.
572
1053
(item)->wk_state &= ~ONWORKLIST; \
573
1054
LIST_REMOVE(item, wk_list); \
574
1055
} while (0)
1056
#define WORKLIST_INSERT_UNLOCKED WORKLIST_INSERT
1057
#define WORKLIST_REMOVE_UNLOCKED WORKLIST_REMOVE
1058
575
1059
#else /* DEBUG */
576
static void worklist_insert(struct workhead *, struct worklist *);
577
static void worklist_remove(struct worklist *);
1060
static void worklist_insert(struct workhead *, struct worklist *, int);
1061
static void worklist_remove(struct worklist *, int);
578
1062
579
#define WORKLIST_INSERT(head, item) worklist_insert(head, item)
580
#define WORKLIST_REMOVE(item) worklist_remove(item)
1063
#define WORKLIST_INSERT(head, item) worklist_insert(head, item, 1)
1064
#define WORKLIST_INSERT_UNLOCKED(head, item) worklist_insert(head, item, 0)
1065
#define WORKLIST_REMOVE(item) worklist_remove(item, 1)
1066
#define WORKLIST_REMOVE_UNLOCKED(item) worklist_remove(item, 0)
581
1067
582
1068
static void
583
worklist_insert(head, item)
1069
worklist_insert(head, item, locked)
584
1070
struct workhead *head;
585
1071
struct worklist *item;
1072
int locked;
586
1073
{
587
1074
588
mtx_assert(&lk, MA_OWNED);
1075
if (locked)
1076
mtx_assert(&lk, MA_OWNED);
589
1077
if (item->wk_state & ONWORKLIST)
590
panic("worklist_insert: already on list");
1078
panic("worklist_insert: %p %s(0x%X) already on list",
1079
item, TYPENAME(item->wk_type), item->wk_state);
591
1080
item->wk_state |= ONWORKLIST;
592
1081
LIST_INSERT_HEAD(head, item, wk_list);
593
1082
}
594
1083
595
1084
static void
596
worklist_remove(item)
1085
worklist_remove(item, locked)
597
1086
struct worklist *item;
1087
int locked;
598
1088
{
599
1089
600
mtx_assert(&lk, MA_OWNED);
1090
if (locked)
1091
mtx_assert(&lk, MA_OWNED);
601
1092
if ((item->wk_state & ONWORKLIST) == 0)
602
panic("worklist_remove: not on list");
1093
panic("worklist_remove: %p %s(0x%X) not on list",
1094
item, TYPENAME(item->wk_type), item->wk_state);
603
1095
item->wk_state &= ~ONWORKLIST;
604
1096
LIST_REMOVE(item, wk_list);
605
1097
}
606
1098
#endif /* DEBUG */
607
1099
608
1100
/*
1101
* Merge two jsegdeps keeping only the oldest one as newer references
1102
* can't be discarded until after older references.
1103
*/
1104
static inline struct jsegdep *
1105
jsegdep_merge(struct jsegdep *one, struct jsegdep *two)
1106
{
1107
struct jsegdep *swp;
1108
1109
if (two == NULL)
1110
return (one);
1111
1112
if (one->jd_seg->js_seq > two->jd_seg->js_seq) {
1113
swp = one;
1114
one = two;
1115
two = swp;
1116
}
1117
WORKLIST_REMOVE(&two->jd_list);
1118
free_jsegdep(two);
1119
1120
return (one);
1121
}
1122
1123
/*
1124
* If two freedeps are compatible free one to reduce list size.
1125
*/
1126
static inline struct freedep *
1127
freedep_merge(struct freedep *one, struct freedep *two)
1128
{
1129
if (two == NULL)
1130
return (one);
1131
1132
if (one->fd_freework == two->fd_freework) {
1133
WORKLIST_REMOVE(&two->fd_list);
1134
free_freedep(two);
1135
}
1136
return (one);
1137
}
1138
1139
/*
1140
* Move journal work from one list to another. Duplicate freedeps and
1141
* jsegdeps are coalesced to keep the lists as small as possible.
1142
*/
1143
static void
1144
jwork_move(dst, src)
1145
struct workhead *dst;
1146
struct workhead *src;
1147
{
1148
struct freedep *freedep;
1149
struct jsegdep *jsegdep;
1150
struct worklist *wkn;
1151
struct worklist *wk;
1152
1153
KASSERT(dst != src,
1154
("jwork_move: dst == src"));
1155
freedep = NULL;
1156
jsegdep = NULL;
1157
LIST_FOREACH_SAFE(wk, dst, wk_list, wkn) {
1158
if (wk->wk_type == D_JSEGDEP)
1159
jsegdep = jsegdep_merge(WK_JSEGDEP(wk), jsegdep);
1160
if (wk->wk_type == D_FREEDEP)
1161
freedep = freedep_merge(WK_FREEDEP(wk), freedep);
1162
}
1163
1164
mtx_assert(&lk, MA_OWNED);
1165
while ((wk = LIST_FIRST(src)) != NULL) {
1166
WORKLIST_REMOVE(wk);
1167
WORKLIST_INSERT(dst, wk);
1168
if (wk->wk_type == D_JSEGDEP) {
1169
jsegdep = jsegdep_merge(WK_JSEGDEP(wk), jsegdep);
1170
continue;
1171
}
1172
if (wk->wk_type == D_FREEDEP)
1173
freedep = freedep_merge(WK_FREEDEP(wk), freedep);
1174
}
1175
}
1176
1177
static void
1178
jwork_insert(dst, jsegdep)
1179
struct workhead *dst;
1180
struct jsegdep *jsegdep;
1181
{
1182
struct jsegdep *jsegdepn;
1183
struct worklist *wk;
1184
1185
LIST_FOREACH(wk, dst, wk_list)
1186
if (wk->wk_type == D_JSEGDEP)
1187
break;
1188
if (wk == NULL) {
1189
WORKLIST_INSERT(dst, &jsegdep->jd_list);
1190
return;
1191
}
1192
jsegdepn = WK_JSEGDEP(wk);
1193
if (jsegdep->jd_seg->js_seq < jsegdepn->jd_seg->js_seq) {
1194
WORKLIST_REMOVE(wk);
1195
free_jsegdep(jsegdepn);
1196
WORKLIST_INSERT(dst, &jsegdep->jd_list);
1197
} else
1198
free_jsegdep(jsegdep);
1199
}
1200
1201
/*
609
1202
* Routines for tracking and managing workitems.
610
1203
*/
611
1204
static void workitem_free(struct worklist *, int);
623
1216
624
1217
#ifdef DEBUG
625
1218
if (item->wk_state & ONWORKLIST)
626
panic("workitem_free: still on list");
1219
panic("workitem_free: %s(0x%X) still on list",
1220
TYPENAME(item->wk_type), item->wk_state);
627
1221
if (item->wk_type != type)
628
panic("workitem_free: type mismatch");
1222
panic("workitem_free: type mismatch %s != %s",
1223
TYPENAME(item->wk_type), TYPENAME(type));
629
1224
#endif
1225
if (item->wk_state & IOWAITING)
1226
wakeup(item);
630
1227
ump = VFSTOUFS(item->wk_mp);
631
1228
if (--ump->softdep_deps == 0 && ump->softdep_req)
632
1229
wakeup(&ump->softdep_deps);
1230
dep_current[type]--;
633
1231
free(item, DtoM(type));
634
1232
}
635
1233
639
1237
int type;
640
1238
struct mount *mp;
641
1239
{
1240
struct ufsmount *ump;
1241
642
1242
item->wk_type = type;
643
1243
item->wk_mp = mp;
644
1244
item->wk_state = 0;
1245
1246
ump = VFSTOUFS(mp);
645
1247
ACQUIRE_LOCK(&lk);
646
VFSTOUFS(mp)->softdep_deps++;
647
VFSTOUFS(mp)->softdep_accdeps++;
1248
dep_current[type]++;
1249
dep_total[type]++;
1250
ump->softdep_deps++;
1251
ump->softdep_accdeps++;
648
1252
FREE_LOCK(&lk);
649
1253
}
650
1254
659
1263
static struct callout softdep_callout;
660
1264
static int req_pending;
661
1265
static int req_clear_inodedeps; /* syncer process flush some inodedeps */
662
#define FLUSH_INODES 1
663
1266
static int req_clear_remove; /* syncer process flush some freeblks */
664
#define FLUSH_REMOVE 2
665
#define FLUSH_REMOVE_WAIT 3
666
static long num_freeblkdep; /* number of freeblks workitems allocated */
1267
static int softdep_flushcache = 0; /* Should we do BIO_FLUSH? */
667
1268
668
1269
/*
669
1270
* runtime statistics
678
1279
static int stat_inode_bitmap; /* bufs redirtied as inode bitmap not written */
679
1280
static int stat_direct_blk_ptrs;/* bufs redirtied as direct ptrs not written */
680
1281
static int stat_dir_entry; /* bufs redirtied as dir entry cannot write */
1282
static int stat_jaddref; /* bufs redirtied as ino bitmap can not write */
1283
static int stat_jnewblk; /* bufs redirtied as blk bitmap can not write */
1284
static int stat_journal_min; /* Times hit journal min threshold */
1285
static int stat_journal_low; /* Times hit journal low threshold */
1286
static int stat_journal_wait; /* Times blocked in jwait(). */
1287
static int stat_jwait_filepage; /* Times blocked in jwait() for filepage. */
1288
static int stat_jwait_freeblks; /* Times blocked in jwait() for freeblks. */
1289
static int stat_jwait_inode; /* Times blocked in jwait() for inodes. */
1290
static int stat_jwait_newblk; /* Times blocked in jwait() for newblks. */
1291
static int stat_cleanup_high_delay; /* Maximum cleanup delay (in ticks) */
1292
static int stat_cleanup_blkrequests; /* Number of block cleanup requests */
1293
static int stat_cleanup_inorequests; /* Number of inode cleanup requests */
1294
static int stat_cleanup_retries; /* Number of cleanups that needed to flush */
1295
static int stat_cleanup_failures; /* Number of cleanup requests that failed */
681
1296
682
SYSCTL_INT(_debug, OID_AUTO, max_softdeps, CTLFLAG_RW, &max_softdeps, 0, "");
683
SYSCTL_INT(_debug, OID_AUTO, tickdelay, CTLFLAG_RW, &tickdelay, 0, "");
684
SYSCTL_INT(_debug, OID_AUTO, maxindirdeps, CTLFLAG_RW, &maxindirdeps, 0, "");
685
SYSCTL_INT(_debug, OID_AUTO, worklist_push, CTLFLAG_RW, &stat_worklist_push, 0,"");
686
SYSCTL_INT(_debug, OID_AUTO, blk_limit_push, CTLFLAG_RW, &stat_blk_limit_push, 0,"");
687
SYSCTL_INT(_debug, OID_AUTO, ino_limit_push, CTLFLAG_RW, &stat_ino_limit_push, 0,"");
688
SYSCTL_INT(_debug, OID_AUTO, blk_limit_hit, CTLFLAG_RW, &stat_blk_limit_hit, 0, "");
689
SYSCTL_INT(_debug, OID_AUTO, ino_limit_hit, CTLFLAG_RW, &stat_ino_limit_hit, 0, "");
690
SYSCTL_INT(_debug, OID_AUTO, sync_limit_hit, CTLFLAG_RW, &stat_sync_limit_hit, 0, "");
691
SYSCTL_INT(_debug, OID_AUTO, indir_blk_ptrs, CTLFLAG_RW, &stat_indir_blk_ptrs, 0, "");
692
SYSCTL_INT(_debug, OID_AUTO, inode_bitmap, CTLFLAG_RW, &stat_inode_bitmap, 0, "");
693
SYSCTL_INT(_debug, OID_AUTO, direct_blk_ptrs, CTLFLAG_RW, &stat_direct_blk_ptrs, 0, "");
694
SYSCTL_INT(_debug, OID_AUTO, dir_entry, CTLFLAG_RW, &stat_dir_entry, 0, "");
695
/* SYSCTL_INT(_debug, OID_AUTO, worklist_num, CTLFLAG_RD, &softdep_on_worklist, 0, ""); */
1297
SYSCTL_INT(_debug_softdep, OID_AUTO, max_softdeps, CTLFLAG_RW,
1298
&max_softdeps, 0, "");
1299
SYSCTL_INT(_debug_softdep, OID_AUTO, tickdelay, CTLFLAG_RW,
1300
&tickdelay, 0, "");
1301
SYSCTL_INT(_debug_softdep, OID_AUTO, maxindirdeps, CTLFLAG_RW,
1302
&maxindirdeps, 0, "");
1303
SYSCTL_INT(_debug_softdep, OID_AUTO, worklist_push, CTLFLAG_RW,
1304
&stat_worklist_push, 0,"");
1305
SYSCTL_INT(_debug_softdep, OID_AUTO, blk_limit_push, CTLFLAG_RW,
1306
&stat_blk_limit_push, 0,"");
1307
SYSCTL_INT(_debug_softdep, OID_AUTO, ino_limit_push, CTLFLAG_RW,
1308
&stat_ino_limit_push, 0,"");
1309
SYSCTL_INT(_debug_softdep, OID_AUTO, blk_limit_hit, CTLFLAG_RW,
1310
&stat_blk_limit_hit, 0, "");
1311
SYSCTL_INT(_debug_softdep, OID_AUTO, ino_limit_hit, CTLFLAG_RW,
1312
&stat_ino_limit_hit, 0, "");
1313
SYSCTL_INT(_debug_softdep, OID_AUTO, sync_limit_hit, CTLFLAG_RW,
1314
&stat_sync_limit_hit, 0, "");
1315
SYSCTL_INT(_debug_softdep, OID_AUTO, indir_blk_ptrs, CTLFLAG_RW,
1316
&stat_indir_blk_ptrs, 0, "");
1317
SYSCTL_INT(_debug_softdep, OID_AUTO, inode_bitmap, CTLFLAG_RW,
1318
&stat_inode_bitmap, 0, "");
1319
SYSCTL_INT(_debug_softdep, OID_AUTO, direct_blk_ptrs, CTLFLAG_RW,
1320
&stat_direct_blk_ptrs, 0, "");
1321
SYSCTL_INT(_debug_softdep, OID_AUTO, dir_entry, CTLFLAG_RW,
1322
&stat_dir_entry, 0, "");
1323
SYSCTL_INT(_debug_softdep, OID_AUTO, jaddref_rollback, CTLFLAG_RW,
1324
&stat_jaddref, 0, "");
1325
SYSCTL_INT(_debug_softdep, OID_AUTO, jnewblk_rollback, CTLFLAG_RW,
1326
&stat_jnewblk, 0, "");
1327
SYSCTL_INT(_debug_softdep, OID_AUTO, journal_low, CTLFLAG_RW,
1328
&stat_journal_low, 0, "");
1329
SYSCTL_INT(_debug_softdep, OID_AUTO, journal_min, CTLFLAG_RW,
1330
&stat_journal_min, 0, "");
1331
SYSCTL_INT(_debug_softdep, OID_AUTO, journal_wait, CTLFLAG_RW,
1332
&stat_journal_wait, 0, "");
1333
SYSCTL_INT(_debug_softdep, OID_AUTO, jwait_filepage, CTLFLAG_RW,
1334
&stat_jwait_filepage, 0, "");
1335
SYSCTL_INT(_debug_softdep, OID_AUTO, jwait_freeblks, CTLFLAG_RW,
1336
&stat_jwait_freeblks, 0, "");
1337
SYSCTL_INT(_debug_softdep, OID_AUTO, jwait_inode, CTLFLAG_RW,
1338
&stat_jwait_inode, 0, "");
1339
SYSCTL_INT(_debug_softdep, OID_AUTO, jwait_newblk, CTLFLAG_RW,
1340
&stat_jwait_newblk, 0, "");
1341
SYSCTL_INT(_debug_softdep, OID_AUTO, cleanup_blkrequests, CTLFLAG_RW,
1342
&stat_cleanup_blkrequests, 0, "");
1343
SYSCTL_INT(_debug_softdep, OID_AUTO, cleanup_inorequests, CTLFLAG_RW,
1344
&stat_cleanup_inorequests, 0, "");
1345
SYSCTL_INT(_debug_softdep, OID_AUTO, cleanup_high_delay, CTLFLAG_RW,
1346
&stat_cleanup_high_delay, 0, "");
1347
SYSCTL_INT(_debug_softdep, OID_AUTO, cleanup_retries, CTLFLAG_RW,
1348
&stat_cleanup_retries, 0, "");
1349
SYSCTL_INT(_debug_softdep, OID_AUTO, cleanup_failures, CTLFLAG_RW,
1350
&stat_cleanup_failures, 0, "");
1351
SYSCTL_INT(_debug_softdep, OID_AUTO, flushcache, CTLFLAG_RW,
1352
&softdep_flushcache, 0, "");
696
1353
697
1354
SYSCTL_DECL(_vfs_ffs);
698
1355
1356
LIST_HEAD(bmsafemap_hashhead, bmsafemap) *bmsafemap_hashtbl;
1357
static u_long bmsafemap_hash; /* size of hash table - 1 */
1358
699
1359
static int compute_summary_at_mount = 0; /* Whether to recompute the summary at mount time */
700
1360
SYSCTL_INT(_vfs_ffs, OID_AUTO, compute_summary_at_mount, CTLFLAG_RW,
701
1361
&compute_summary_at_mount, 0, "Recompute summary at mount");
717
1377
struct ufsmount *ump;
718
1378
struct thread *td;
719
1379
int remaining;
1380
int progress;
720
1381
int vfslocked;
721
1382
722
1383
td = curthread;
741
1402
}
742
1403
FREE_LOCK(&lk);
743
1404
VFS_UNLOCK_GIANT(vfslocked);
744
remaining = 0;
1405
remaining = progress = 0;
745
1406
mtx_lock(&mountlist_mtx);
746
1407
for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
747
1408
nmp = TAILQ_NEXT(mp, mnt_list);
748
if ((mp->mnt_flag & MNT_SOFTDEP) == 0)
1409
if (MOUNTEDSOFTDEP(mp) == 0)
749
1410
continue;
750
1411
if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
751
1412
continue;
752
1413
vfslocked = VFS_LOCK_GIANT(mp);
753
softdep_process_worklist(mp, 0);
1414
progress += softdep_process_worklist(mp, 0);
754
1415
ump = VFSTOUFS(mp);
755
remaining += ump->softdep_on_worklist -
756
ump->softdep_on_worklist_inprogress;
1416
remaining += ump->softdep_on_worklist;
757
1417
VFS_UNLOCK_GIANT(vfslocked);
758
1418
mtx_lock(&mountlist_mtx);
759
1419
nmp = TAILQ_NEXT(mp, mnt_list);
760
1420
vfs_unbusy(mp);
761
1421
}
762
1422
mtx_unlock(&mountlist_mtx);
763
if (remaining)
1423
if (remaining && progress)
764
1424
continue;
765
1425
ACQUIRE_LOCK(&lk);
766
1426
if (!req_pending)
770
1430
}
771
1431
}
772
1432
773
static int
774
softdep_speedup(void)
1433
static void
1434
worklist_speedup(void)
775
1435
{
776
777
1436
mtx_assert(&lk, MA_OWNED);
778
1437
if (req_pending == 0) {
779
1438
req_pending = 1;
780
1439
wakeup(&req_pending);
781
1440
}
782
1441
}
1442
1443
static int
1444
softdep_speedup(void)
1445
{
1446
1447
worklist_speedup();
1448
bd_speedup();
783
1449
return speedup_syncer();
784
1450
}
785
1451
790
1456
* The following routine is the only one that removes items
791
1457
* and does so in order from first to last.
792
1458
*/
1459
1460
#define WK_HEAD 0x0001 /* Add to HEAD. */
1461
#define WK_NODELAY 0x0002 /* Process immediately. */
1462
793
1463
static void
794
add_to_worklist(wk)
1464
add_to_worklist(wk, flags)
795
1465
struct worklist *wk;
1466
int flags;
796
1467
{
797
1468
struct ufsmount *ump;
798
1469
799
1470
mtx_assert(&lk, MA_OWNED);
800
1471
ump = VFSTOUFS(wk->wk_mp);
801
1472
if (wk->wk_state & ONWORKLIST)
802
panic("add_to_worklist: already on list");
1473
panic("add_to_worklist: %s(0x%X) already on list",
1474
TYPENAME(wk->wk_type), wk->wk_state);
803
1475
wk->wk_state |= ONWORKLIST;
804
if (LIST_EMPTY(&ump->softdep_workitem_pending))
805
LIST_INSERT_HEAD(&ump->softdep_workitem_pending, wk, wk_list);
806
else
1476
if (ump->softdep_on_worklist == 0) {
1477
LIST_INSERT_HEAD(&ump->softdep_workitem_pending, wk, wk_list);
1478
ump->softdep_worklist_tail = wk;
1479
} else if (flags & WK_HEAD) {
1480
LIST_INSERT_HEAD(&ump->softdep_workitem_pending, wk, wk_list);
1481
} else {
807
1482
LIST_INSERT_AFTER(ump->softdep_worklist_tail, wk, wk_list);
808
ump->softdep_worklist_tail = wk;
1483
ump->softdep_worklist_tail = wk;
1484
}
809
1485
ump->softdep_on_worklist += 1;
1486
if (flags & WK_NODELAY)
1487
worklist_speedup();
1488
}
1489
1490
/*
1491
* Remove the item to be processed. If we are removing the last
1492
* item on the list, we need to recalculate the tail pointer.
1493
*/
1494
static void
1495
remove_from_worklist(wk)
1496
struct worklist *wk;
1497
{
1498
struct ufsmount *ump;
1499
1500
ump = VFSTOUFS(wk->wk_mp);
1501
WORKLIST_REMOVE(wk);
1502
if (ump->softdep_worklist_tail == wk)
1503
ump->softdep_worklist_tail =
1504
(struct worklist *)wk->wk_list.le_prev;
1505
ump->softdep_on_worklist -= 1;
1506
}
1507
1508
static void
1509
wake_worklist(wk)
1510
struct worklist *wk;
1511
{
1512
if (wk->wk_state & IOWAITING) {
1513
wk->wk_state &= ~IOWAITING;
1514
wakeup(wk);
1515
}
1516
}
1517
1518
static void
1519
wait_worklist(wk, wmesg)
1520
struct worklist *wk;
1521
char *wmesg;
1522
{
1523
1524
wk->wk_state |= IOWAITING;
1525
msleep(wk, &lk, PVM, wmesg, 0);
810
1526
}
811
1527
812
1528
/*
824
1540
int full;
825
1541
{
826
1542
struct thread *td = curthread;
827
int cnt, matchcnt, loopcount;
1543
int cnt, matchcnt;
828
1544
struct ufsmount *ump;
829
1545
long starttime;
830
1546
836
1552
matchcnt = 0;
837
1553
ump = VFSTOUFS(mp);
838
1554
ACQUIRE_LOCK(&lk);
839
loopcount = 1;
840
1555
starttime = time_second;
1556
softdep_process_journal(mp, NULL, full?MNT_WAIT:0);
841
1557
while (ump->softdep_on_worklist > 0) {
842
if ((cnt = process_worklist_item(mp, 0)) == -1)
1558
if ((cnt = process_worklist_item(mp, 10, LK_NOWAIT)) == 0)
843
1559
break;
844
1560
else
845
1561
matchcnt += cnt;
860
1576
* We do not generally want to stop for buffer space, but if
861
1577
* we are really being a buffer hog, we will stop and wait.
862
1578
*/
863
if (loopcount++ % 128 == 0) {
1579
if (should_yield()) {
864
1580
FREE_LOCK(&lk);
865
uio_yield();
1581
kern_yield(PRI_UNCHANGED);
866
1582
bwillwrite();
867
1583
ACQUIRE_LOCK(&lk);
868
1584
}
871
1587
* second. Otherwise the other mountpoints may get
872
1588
* excessively backlogged.
873
1589
*/
874
if (!full && starttime != time_second) {
875
matchcnt = -1;
1590
if (!full && starttime != time_second)
876
1591
break;
877
}
878
1592
}
1593
if (full == 0)
1594
journal_unsuspend(ump);
879
1595
FREE_LOCK(&lk);
880
1596
return (matchcnt);
881
1597
}
882
1598
883
1599
/*
1600
* Process all removes associated with a vnode if we are running out of
1601
* journal space. Any other process which attempts to flush these will
1602
* be unable as we have the vnodes locked.
1603
*/
1604
static void
1605
process_removes(vp)
1606
struct vnode *vp;
1607
{
1608
struct inodedep *inodedep;
1609
struct dirrem *dirrem;
1610
struct mount *mp;
1611
ino_t inum;
1612
1613
mtx_assert(&lk, MA_OWNED);
1614
1615
mp = vp->v_mount;
1616
inum = VTOI(vp)->i_number;
1617
for (;;) {
1618
top:
1619
if (inodedep_lookup(mp, inum, 0, &inodedep) == 0)
1620
return;
1621
LIST_FOREACH(dirrem, &inodedep->id_dirremhd, dm_inonext) {
1622
/*
1623
* If another thread is trying to lock this vnode
1624
* it will fail but we must wait for it to do so
1625
* before we can proceed.
1626
*/
1627
if (dirrem->dm_state & INPROGRESS) {
1628
wait_worklist(&dirrem->dm_list, "pwrwait");
1629
goto top;
1630
}
1631
if ((dirrem->dm_state & (COMPLETE | ONWORKLIST)) ==
1632
(COMPLETE | ONWORKLIST))
1633
break;
1634
}
1635
if (dirrem == NULL)
1636
return;
1637
remove_from_worklist(&dirrem->dm_list);
1638
FREE_LOCK(&lk);
1639
if (vn_start_secondary_write(NULL, &mp, V_NOWAIT))
1640
panic("process_removes: suspended filesystem");
1641
handle_workitem_remove(dirrem, 0);
1642
vn_finished_secondary_write(mp);
1643
ACQUIRE_LOCK(&lk);
1644
}
1645
}
1646
1647
/*
1648
* Process all truncations associated with a vnode if we are running out
1649
* of journal space. This is called when the vnode lock is already held
1650
* and no other process can clear the truncation. This function returns
1651
* a value greater than zero if it did any work.
1652
*/
1653
static void
1654
process_truncates(vp)
1655
struct vnode *vp;
1656
{
1657
struct inodedep *inodedep;
1658
struct freeblks *freeblks;
1659
struct mount *mp;
1660
ino_t inum;
1661
int cgwait;
1662
1663
mtx_assert(&lk, MA_OWNED);
1664
1665
mp = vp->v_mount;
1666
inum = VTOI(vp)->i_number;
1667
for (;;) {
1668
if (inodedep_lookup(mp, inum, 0, &inodedep) == 0)
1669
return;
1670
cgwait = 0;
1671
TAILQ_FOREACH(freeblks, &inodedep->id_freeblklst, fb_next) {
1672
/* Journal entries not yet written. */
1673
if (!LIST_EMPTY(&freeblks->fb_jblkdephd)) {
1674
jwait(&LIST_FIRST(
1675
&freeblks->fb_jblkdephd)->jb_list,
1676
MNT_WAIT);
1677
break;
1678
}
1679
/* Another thread is executing this item. */
1680
if (freeblks->fb_state & INPROGRESS) {
1681
wait_worklist(&freeblks->fb_list, "ptrwait");
1682
break;
1683
}
1684
/* Freeblks is waiting on a inode write. */
1685
if ((freeblks->fb_state & COMPLETE) == 0) {
1686
FREE_LOCK(&lk);
1687
ffs_update(vp, 1);
1688
ACQUIRE_LOCK(&lk);
1689
break;
1690
}
1691
if ((freeblks->fb_state & (ALLCOMPLETE | ONWORKLIST)) ==
1692
(ALLCOMPLETE | ONWORKLIST)) {
1693
remove_from_worklist(&freeblks->fb_list);
1694
freeblks->fb_state |= INPROGRESS;
1695
FREE_LOCK(&lk);
1696
if (vn_start_secondary_write(NULL, &mp,
1697
V_NOWAIT))
1698
panic("process_truncates: "
1699
"suspended filesystem");
1700
handle_workitem_freeblocks(freeblks, 0);
1701
vn_finished_secondary_write(mp);
1702
ACQUIRE_LOCK(&lk);
1703
break;
1704
}
1705
if (freeblks->fb_cgwait)
1706
cgwait++;
1707
}
1708
if (cgwait) {
1709
FREE_LOCK(&lk);
1710
sync_cgs(mp, MNT_WAIT);
1711
ffs_sync_snap(mp, MNT_WAIT);
1712
ACQUIRE_LOCK(&lk);
1713
continue;
1714
}
1715
if (freeblks == NULL)
1716
break;
1717
}
1718
return;
1719
}
1720
1721
/*
884
1722
* Process one item on the worklist.
885
1723
*/
886
1724
static int
887
process_worklist_item(mp, flags)
1725
process_worklist_item(mp, target, flags)
888
1726
struct mount *mp;
1727
int target;
889
1728
int flags;
890
1729
{
891
struct worklist *wk, *wkend;
1730
struct worklist sentinel;
1731
struct worklist *wk;
892
1732
struct ufsmount *ump;
893
struct vnode *vp;
894
int matchcnt = 0;
1733
int matchcnt;
1734
int error;
895
1735
896
1736
mtx_assert(&lk, MA_OWNED);
897
1737
KASSERT(mp != NULL, ("process_worklist_item: NULL mp"));
902
1742
*/
903
1743
if (curthread->td_pflags & TDP_COWINPROGRESS)
904
1744
return (-1);
905
/*
906
* Normally we just process each item on the worklist in order.
907
* However, if we are in a situation where we cannot lock any
908
* inodes, we have to skip over any dirrem requests whose
909
* vnodes are resident and locked.
910
*/
1745
PHOLD(curproc); /* Don't let the stack go away. */
911
1746
ump = VFSTOUFS(mp);
912
vp = NULL;
913
LIST_FOREACH(wk, &ump->softdep_workitem_pending, wk_list) {
1747
matchcnt = 0;
1748
sentinel.wk_mp = NULL;
1749
sentinel.wk_type = D_SENTINEL;
1750
LIST_INSERT_HEAD(&ump->softdep_workitem_pending, &sentinel, wk_list);
1751
for (wk = LIST_NEXT(&sentinel, wk_list); wk != NULL;
1752
wk = LIST_NEXT(&sentinel, wk_list)) {
1753
if (wk->wk_type == D_SENTINEL) {
1754
LIST_REMOVE(&sentinel, wk_list);
1755
LIST_INSERT_AFTER(wk, &sentinel, wk_list);
1756
continue;
1757
}
914
1758
if (wk->wk_state & INPROGRESS)
915
continue;
916
if ((flags & LK_NOWAIT) == 0 || wk->wk_type != D_DIRREM)
917
break;
1759
panic("process_worklist_item: %p already in progress.",
1760
wk);
918
1761
wk->wk_state |= INPROGRESS;
919
ump->softdep_on_worklist_inprogress++;
1762
remove_from_worklist(wk);
920
1763
FREE_LOCK(&lk);
921
ffs_vgetf(mp, WK_DIRREM(wk)->dm_oldinum,
922
LK_NOWAIT | LK_EXCLUSIVE, &vp, FFSV_FORCEINSMQ);
1764
if (vn_start_secondary_write(NULL, &mp, V_NOWAIT))
1765
panic("process_worklist_item: suspended filesystem");
1766
switch (wk->wk_type) {
1767
case D_DIRREM:
1768
/* removal of a directory entry */
1769
error = handle_workitem_remove(WK_DIRREM(wk), flags);
1770
break;
1771
1772
case D_FREEBLKS:
1773
/* releasing blocks and/or fragments from a file */
1774
error = handle_workitem_freeblocks(WK_FREEBLKS(wk),
1775
flags);
1776
break;
1777
1778
case D_FREEFRAG:
1779
/* releasing a fragment when replaced as a file grows */
1780
handle_workitem_freefrag(WK_FREEFRAG(wk));
1781
error = 0;
1782
break;
1783
1784
case D_FREEFILE:
1785
/* releasing an inode when its link count drops to 0 */
1786
handle_workitem_freefile(WK_FREEFILE(wk));
1787
error = 0;
1788
break;
1789
1790
default:
1791
panic("%s_process_worklist: Unknown type %s",
1792
"softdep", TYPENAME(wk->wk_type));
1793
/* NOTREACHED */
1794
}
1795
vn_finished_secondary_write(mp);
923
1796
ACQUIRE_LOCK(&lk);
1797
if (error == 0) {
1798
if (++matchcnt == target)
1799
break;
1800
continue;
1801
}
1802
/*
1803
* We have to retry the worklist item later. Wake up any
1804
* waiters who may be able to complete it immediately and
1805
* add the item back to the head so we don't try to execute
1806
* it again.
1807
*/
924
1808
wk->wk_state &= ~INPROGRESS;
925
ump->softdep_on_worklist_inprogress--;
926
if (vp != NULL)
927
break;
928
}
929
if (wk == 0)
930
return (-1);
931
/*
932
* Remove the item to be processed. If we are removing the last
933
* item on the list, we need to recalculate the tail pointer.
934
* As this happens rarely and usually when the list is short,
935
* we just run down the list to find it rather than tracking it
936
* in the above loop.
937
*/
938
WORKLIST_REMOVE(wk);
939
if (wk == ump->softdep_worklist_tail) {
940
LIST_FOREACH(wkend, &ump->softdep_workitem_pending, wk_list)
941
if (LIST_NEXT(wkend, wk_list) == NULL)
942
break;
943
ump->softdep_worklist_tail = wkend;
944
}
945
ump->softdep_on_worklist -= 1;
946
FREE_LOCK(&lk);
947
if (vn_start_secondary_write(NULL, &mp, V_NOWAIT))
948
panic("process_worklist_item: suspended filesystem");
949
matchcnt++;
950
switch (wk->wk_type) {
951
952
case D_DIRREM:
953
/* removal of a directory entry */
954
handle_workitem_remove(WK_DIRREM(wk), vp);
955
break;
956
957
case D_FREEBLKS:
958
/* releasing blocks and/or fragments from a file */
959
handle_workitem_freeblocks(WK_FREEBLKS(wk), flags & LK_NOWAIT);
960
break;
961
962
case D_FREEFRAG:
963
/* releasing a fragment when replaced as a file grows */
964
handle_workitem_freefrag(WK_FREEFRAG(wk));
965
break;
966
967
case D_FREEFILE:
968
/* releasing an inode when its link count drops to 0 */
969
handle_workitem_freefile(WK_FREEFILE(wk));
970
break;
971
972
default:
973
panic("%s_process_worklist: Unknown type %s",
974
"softdep", TYPENAME(wk->wk_type));
975
/* NOTREACHED */
976
}
977
vn_finished_secondary_write(mp);
978
ACQUIRE_LOCK(&lk);
1809
wake_worklist(wk);
1810
add_to_worklist(wk, WK_HEAD);
1811
}
1812
LIST_REMOVE(&sentinel, wk_list);
1813
/* Sentinal could've become the tail from remove_from_worklist. */
1814
if (ump->softdep_worklist_tail == &sentinel)
1815
ump->softdep_worklist_tail =
1816
(struct worklist *)sentinel.wk_list.le_prev;
1817
PRELE(curproc);
979
1818
return (matchcnt);
980
1819
}
981
1820
982
1821
/*
983
1822
* Move dependencies from one buffer to another.
984
1823
*/
985
void
1824
int
986
1825
softdep_move_dependencies(oldbp, newbp)
987
1826
struct buf *oldbp;
988
1827
struct buf *newbp;
989
1828
{
990
1829
struct worklist *wk, *wktail;
1830
int dirty;
991
1831
992
if (!LIST_EMPTY(&newbp->b_dep))
993
panic("softdep_move_dependencies: need merge code");
994
wktail = 0;
1832
dirty = 0;
1833
wktail = NULL;
995
1834
ACQUIRE_LOCK(&lk);
996
1835
while ((wk = LIST_FIRST(&oldbp->b_dep)) != NULL) {
997
1836
LIST_REMOVE(wk, wk_list);
1837
if (wk->wk_type == D_BMSAFEMAP &&
1838
bmsafemap_backgroundwrite(WK_BMSAFEMAP(wk), newbp))
1839
dirty = 1;
998
1840
if (wktail == 0)
999
1841
LIST_INSERT_HEAD(&newbp->b_dep, wk, wk_list);
1000
1842
else
1002
1844
wktail = wk;
1003
1845
}
1004
1846
FREE_LOCK(&lk);
1847
1848
return (dirty);
1005
1849
}
1006
1850
1007
1851
/*
1073
1917
int flags;
1074
1918
struct thread *td;
1075
1919
{
1076
int error, depcount, loopcnt, retry_flush_count, retry;
1920
#ifdef QUOTA
1921
struct ufsmount *ump;
1922
int i;
1923
#endif
1924
int error, early, depcount, loopcnt, retry_flush_count, retry;
1925
int morework;
1077
1926
1078
1927
loopcnt = 10;
1079
1928
retry_flush_count = 3;
1091
1940
* Do another flush in case any vnodes were brought in
1092
1941
* as part of the cleanup operations.
1093
1942
*/
1094
if ((error = ffs_flushfiles(oldmnt, flags, td)) != 0)
1943
early = retry_flush_count == 1 || (oldmnt->mnt_kern_flag &
1944
MNTK_UNMOUNT) == 0 ? 0 : EARLYFLUSH;
1945
if ((error = ffs_flushfiles(oldmnt, flags | early, td)) != 0)
1095
1946
break;
1096
1947
if ((error = softdep_flushworklist(oldmnt, &depcount, td)) != 0 ||
1097
1948
depcount == 0)
1115
1966
MNT_ILOCK(oldmnt);
1116
1967
KASSERT((oldmnt->mnt_kern_flag & MNTK_NOINSMNTQ) != 0,
1117
1968
("softdep_flushfiles: !MNTK_NOINSMNTQ"));
1118
if (oldmnt->mnt_nvnodelistsize > 0) {
1969
morework = oldmnt->mnt_nvnodelistsize > 0;
1970
#ifdef QUOTA
1971
ump = VFSTOUFS(oldmnt);
1972
UFS_LOCK(ump);
1973
for (i = 0; i < MAXQUOTAS; i++) {
1974
if (ump->um_quotas[i] != NULLVP)
1975
morework = 1;
1976
}
1977
UFS_UNLOCK(ump);
1978
#endif
1979
if (morework) {
1119
1980
if (--retry_flush_count > 0) {
1120
1981
retry = 1;
1121
1982
loopcnt = 3;
1175
2036
{
1176
2037
struct pagedep *pagedep;
1177
2038
1178
LIST_FOREACH(pagedep, pagedephd, pd_hash)
1179
if (ino == pagedep->pd_ino &&
1180
lbn == pagedep->pd_lbn &&
1181
mp == pagedep->pd_list.wk_mp)
1182
break;
1183
if (pagedep) {
1184
*pagedeppp = pagedep;
1185
if ((flags & DEPALLOC) != 0 &&
1186
(pagedep->pd_state & ONWORKLIST) == 0)
1187
return (0);
1188
return (1);
2039
LIST_FOREACH(pagedep, pagedephd, pd_hash) {
2040
if (ino == pagedep->pd_ino && lbn == pagedep->pd_lbn &&
2041
mp == pagedep->pd_list.wk_mp) {
2042
*pagedeppp = pagedep;
2043
return (1);
2044
}
1189
2045
}
1190
2046
*pagedeppp = NULL;
1191
2047
return (0);
1192
2048
}
1193
2049
/*
1194
* Look up a pagedep. Return 1 if found, 0 if not found or found
1195
* when asked to allocate but not associated with any buffer.
2050
* Look up a pagedep. Return 1 if found, 0 otherwise.
1196
2051
* If not found, allocate if DEPALLOC flag is passed.
1197
2052
* Found or allocated entry is returned in pagedeppp.
1198
2053
* This routine must be called with splbio interrupts blocked.
1199
2054
*/
1200
2055
static int
1201
pagedep_lookup(ip, lbn, flags, pagedeppp)
1202
struct inode *ip;
2056
pagedep_lookup(mp, bp, ino, lbn, flags, pagedeppp)
2057
struct mount *mp;
2058
struct buf *bp;
2059
ino_t ino;
1203
2060
ufs_lbn_t lbn;
1204
2061
int flags;
1205
2062
struct pagedep **pagedeppp;
1206
2063
{
1207
2064
struct pagedep *pagedep;
1208
2065
struct pagedep_hashhead *pagedephd;
1209
struct mount *mp;
2066
struct worklist *wk;
1210
2067
int ret;
1211
2068
int i;
1212
2069
1213
2070
mtx_assert(&lk, MA_OWNED);
1214
mp = ITOV(ip)->v_mount;
1215
pagedephd = PAGEDEP_HASH(mp, ip->i_number, lbn);
1216
1217
ret = pagedep_find(pagedephd, ip->i_number, lbn, mp, flags, pagedeppp);
1218
if (*pagedeppp || (flags & DEPALLOC) == 0)
1219
return (ret);
2071
if (bp) {
2072
LIST_FOREACH(wk, &bp->b_dep, wk_list) {
2073
if (wk->wk_type == D_PAGEDEP) {
2074
*pagedeppp = WK_PAGEDEP(wk);
2075
return (1);
2076
}
2077
}
2078
}
2079
pagedephd = PAGEDEP_HASH(mp, ino, lbn);
2080
ret = pagedep_find(pagedephd, ino, lbn, mp, flags, pagedeppp);
2081
if (ret) {
2082
if (((*pagedeppp)->pd_state & ONWORKLIST) == 0 && bp)
2083
WORKLIST_INSERT(&bp->b_dep, &(*pagedeppp)->pd_list);
2084
return (1);
2085
}
2086
if ((flags & DEPALLOC) == 0)
2087
return (0);
1220
2088
FREE_LOCK(&lk);
1221
2089
pagedep = malloc(sizeof(struct pagedep),
1222
2090
M_PAGEDEP, M_SOFTDEP_FLAGS|M_ZERO);
1223
2091
workitem_alloc(&pagedep->pd_list, D_PAGEDEP, mp);
1224
2092
ACQUIRE_LOCK(&lk);
1225
ret = pagedep_find(pagedephd, ip->i_number, lbn, mp, flags, pagedeppp);
2093
ret = pagedep_find(pagedephd, ino, lbn, mp, flags, pagedeppp);
1226
2094
if (*pagedeppp) {
2095
/*
2096
* This should never happen since we only create pagedeps
2097
* with the vnode lock held. Could be an assert.
2098
*/
1227
2099
WORKITEM_FREE(pagedep, D_PAGEDEP);
1228
2100
return (ret);
1229
2101
}
1230
pagedep->pd_ino = ip->i_number;
2102
pagedep->pd_ino = ino;
1231
2103
pagedep->pd_lbn = lbn;
1232
2104
LIST_INIT(&pagedep->pd_dirremhd);
1233
2105
LIST_INIT(&pagedep->pd_pendinghd);
1234
2106
for (i = 0; i < DAHASHSZ; i++)
1235
2107
LIST_INIT(&pagedep->pd_diraddhd[i]);
1236
2108
LIST_INSERT_HEAD(pagedephd, pagedep, pd_hash);
2109
WORKLIST_INSERT(&bp->b_dep, &pagedep->pd_list);
1237
2110
*pagedeppp = pagedep;
1238
2111
return (0);
1239
2112
}
1243
2116
*/
1244
2117
LIST_HEAD(inodedep_hashhead, inodedep) *inodedep_hashtbl;
1245
2118
static u_long inodedep_hash; /* size of hash table - 1 */
1246
static long num_inodedep; /* number of inodedep allocated */
1247
2119
#define INODEDEP_HASH(fs, inum) \
1248
2120
(&inodedep_hashtbl[((((register_t)(fs)) >> 13) + (inum)) & inodedep_hash])
1249
2121
1295
2167
/*
1296
2168
* If we are over our limit, try to improve the situation.
1297
2169
*/
1298
if (num_inodedep > max_softdeps && (flags & NODELAY) == 0)
2170
if (dep_current[D_INODEDEP] > max_softdeps && (flags & NODELAY) == 0)
1299
2171
request_cleanup(mp, FLUSH_INODES);
1300
2172
FREE_LOCK(&lk);
1301
2173
inodedep = malloc(sizeof(struct inodedep),
1306
2178
WORKITEM_FREE(inodedep, D_INODEDEP);
1307
2179
return (1);
1308
2180
}
1309
num_inodedep += 1;
1310
2181
inodedep->id_fs = fs;
1311
2182
inodedep->id_ino = inum;
1312
2183
inodedep->id_state = ALLCOMPLETE;
1314
2185
inodedep->id_savedino1 = NULL;
1315
2186
inodedep->id_savedsize = -1;
1316
2187
inodedep->id_savedextsize = -1;
1317
inodedep->id_buf = NULL;
2188
inodedep->id_savednlink = -1;
2189
inodedep->id_bmsafemap = NULL;
2190
inodedep->id_mkdiradd = NULL;
2191
LIST_INIT(&inodedep->id_dirremhd);
1318
2192
LIST_INIT(&inodedep->id_pendinghd);
1319
2193
LIST_INIT(&inodedep->id_inowait);
1320
2194
LIST_INIT(&inodedep->id_bufwait);
2195
TAILQ_INIT(&inodedep->id_inoreflst);
1321
2196
TAILQ_INIT(&inodedep->id_inoupdt);
1322
2197
TAILQ_INIT(&inodedep->id_newinoupdt);
1323
2198
TAILQ_INIT(&inodedep->id_extupdt);
1324
2199
TAILQ_INIT(&inodedep->id_newextupdt);
2200
TAILQ_INIT(&inodedep->id_freeblklst);
1325
2201
LIST_INSERT_HEAD(inodedephd, inodedep, id_hash);
1326
2202
*inodedeppp = inodedep;
1327
2203
return (0);
1336
2212
(&newblk_hashtbl[((((register_t)(fs)) >> 13) + (inum)) & newblk_hash])
1337
2213
1338
2214
static int
1339
newblk_find(newblkhd, fs, newblkno, newblkpp)
2215
newblk_find(newblkhd, mp, newblkno, flags, newblkpp)
1340
2216
struct newblk_hashhead *newblkhd;
1341
struct fs *fs;
2217
struct mount *mp;
1342
2218
ufs2_daddr_t newblkno;
2219
int flags;
1343
2220
struct newblk **newblkpp;
1344
2221
{
1345
2222
struct newblk *newblk;
1346
2223
1347
LIST_FOREACH(newblk, newblkhd, nb_hash)
1348
if (newblkno == newblk->nb_newblkno && fs == newblk->nb_fs)
1349
break;
2224
LIST_FOREACH(newblk, newblkhd, nb_hash) {
2225
if (newblkno != newblk->nb_newblkno)
2226
continue;
2227
if (mp != newblk->nb_list.wk_mp)
2228
continue;
2229
/*
2230
* If we're creating a new dependency don't match those that
2231
* have already been converted to allocdirects. This is for
2232
* a frag extend.
2233
*/
2234
if ((flags & DEPALLOC) && newblk->nb_list.wk_type != D_NEWBLK)
2235
continue;
2236
break;
2237
}
1350
2238
if (newblk) {
1351
2239
*newblkpp = newblk;
1352
2240
return (1);
1361
2249
* Found or allocated entry is returned in newblkpp.
1362
2250
*/
1363
2251
static int
1364
newblk_lookup(fs, newblkno, flags, newblkpp)
1365
struct fs *fs;
2252
newblk_lookup(mp, newblkno, flags, newblkpp)
2253
struct mount *mp;
1366
2254
ufs2_daddr_t newblkno;
1367
2255
int flags;
1368
2256
struct newblk **newblkpp;
1370
2258
struct newblk *newblk;
1371
2259
struct newblk_hashhead *newblkhd;
1372
2260
1373
newblkhd = NEWBLK_HASH(fs, newblkno);
1374
if (newblk_find(newblkhd, fs, newblkno, newblkpp))
2261
newblkhd = NEWBLK_HASH(VFSTOUFS(mp)->um_fs, newblkno);
2262
if (newblk_find(newblkhd, mp, newblkno, flags, newblkpp))
1375
2263
return (1);
1376
2264
if ((flags & DEPALLOC) == 0)
1377
2265
return (0);
1378
2266
FREE_LOCK(&lk);
1379
newblk = malloc(sizeof(struct newblk),
1380
M_NEWBLK, M_SOFTDEP_FLAGS);
2267
newblk = malloc(sizeof(union allblk), M_NEWBLK,
2268
M_SOFTDEP_FLAGS | M_ZERO);
2269
workitem_alloc(&newblk->nb_list, D_NEWBLK, mp);
1381
2270
ACQUIRE_LOCK(&lk);
1382
if (newblk_find(newblkhd, fs, newblkno, newblkpp)) {
1383
free(newblk, M_NEWBLK);
2271
if (newblk_find(newblkhd, mp, newblkno, flags, newblkpp)) {
2272
WORKITEM_FREE(newblk, D_NEWBLK);
1384
2273
return (1);
1385
2274
}
1386
newblk->nb_state = 0;
1387
newblk->nb_fs = fs;
2275
newblk->nb_freefrag = NULL;
2276
LIST_INIT(&newblk->nb_indirdeps);
2277
LIST_INIT(&newblk->nb_newdirblk);
2278
LIST_INIT(&newblk->nb_jwork);
2279
newblk->nb_state = ATTACHED;
1388
2280
newblk->nb_newblkno = newblkno;
1389
2281
LIST_INSERT_HEAD(newblkhd, newblk, nb_hash);
1390
2282
*newblkpp = newblk;
1392
2284
}
1393
2285
1394
2286
/*
2287
* Structures and routines associated with freed indirect block caching.
2288
*/
2289
struct freeworklst *indir_hashtbl;
2290
u_long indir_hash; /* size of hash table - 1 */
2291
#define INDIR_HASH(mp, blkno) \
2292
(&indir_hashtbl[((((register_t)(mp)) >> 13) + (blkno)) & indir_hash])
2293
2294
/*
2295
* Lookup an indirect block in the indir hash table. The freework is
2296
* removed and potentially freed. The caller must do a blocking journal
2297
* write before writing to the blkno.
2298
*/
2299
static int
2300
indirblk_lookup(mp, blkno)
2301
struct mount *mp;
2302
ufs2_daddr_t blkno;
2303
{
2304
struct freework *freework;
2305
struct freeworklst *wkhd;
2306
2307
wkhd = INDIR_HASH(mp, blkno);
2308
TAILQ_FOREACH(freework, wkhd, fw_next) {
2309
if (freework->fw_blkno != blkno)
2310
continue;
2311
if (freework->fw_list.wk_mp != mp)
2312
continue;
2313
indirblk_remove(freework);
2314
return (1);
2315
}
2316
return (0);
2317
}
2318
2319
/*
2320
* Insert an indirect block represented by freework into the indirblk
2321
* hash table so that it may prevent the block from being re-used prior
2322
* to the journal being written.
2323
*/
2324
static void
2325
indirblk_insert(freework)
2326
struct freework *freework;
2327
{
2328
struct jblocks *jblocks;
2329
struct jseg *jseg;
2330
2331
jblocks = VFSTOUFS(freework->fw_list.wk_mp)->softdep_jblocks;
2332
jseg = TAILQ_LAST(&jblocks->jb_segs, jseglst);
2333
if (jseg == NULL)
2334
return;
2335
2336
LIST_INSERT_HEAD(&jseg->js_indirs, freework, fw_segs);
2337
TAILQ_INSERT_HEAD(INDIR_HASH(freework->fw_list.wk_mp,
2338
freework->fw_blkno), freework, fw_next);
2339
freework->fw_state &= ~DEPCOMPLETE;
2340
}
2341
2342
static void
2343
indirblk_remove(freework)
2344
struct freework *freework;
2345
{
2346
2347
LIST_REMOVE(freework, fw_segs);
2348
TAILQ_REMOVE(INDIR_HASH(freework->fw_list.wk_mp,
2349
freework->fw_blkno), freework, fw_next);
2350
freework->fw_state |= DEPCOMPLETE;
2351
if ((freework->fw_state & ALLCOMPLETE) == ALLCOMPLETE)
2352
WORKITEM_FREE(freework, D_FREEWORK);
2353
}
2354
2355
/*
1395
2356
* Executed during filesystem system initialization before
1396
2357
* mounting any filesystems.
1397
2358
*/
1398
2359
void
1399
2360
softdep_initialize()
1400
2361
{
2362
int i;
1401
2363
1402
2364
LIST_INIT(&mkdirlisthd);
1403
2365
max_softdeps = desiredvnodes * 4;
1404
pagedep_hashtbl = hashinit(desiredvnodes / 5, M_PAGEDEP,
1405
&pagedep_hash);
2366
pagedep_hashtbl = hashinit(desiredvnodes / 5, M_PAGEDEP, &pagedep_hash);
1406
2367
inodedep_hashtbl = hashinit(desiredvnodes, M_INODEDEP, &inodedep_hash);
1407
newblk_hashtbl = hashinit(64, M_NEWBLK, &newblk_hash);
2368
newblk_hashtbl = hashinit(desiredvnodes / 5, M_NEWBLK, &newblk_hash);
2369
bmsafemap_hashtbl = hashinit(1024, M_BMSAFEMAP, &bmsafemap_hash);
2370
i = 1 << (ffs(desiredvnodes / 10) - 1);
2371
indir_hashtbl = malloc(i * sizeof(indir_hashtbl[0]), M_FREEWORK,
2372
M_WAITOK);
2373
indir_hash = i - 1;
2374
for (i = 0; i <= indir_hash; i++)
2375
TAILQ_INIT(&indir_hashtbl[i]);
1408
2376
1409
2377
/* initialise bioops hack */
1410
2378
bioops.io_start = softdep_disk_io_initiation;
1428
2396
hashdestroy(pagedep_hashtbl, M_PAGEDEP, pagedep_hash);
1429
2397
hashdestroy(inodedep_hashtbl, M_INODEDEP, inodedep_hash);
1430
2398
hashdestroy(newblk_hashtbl, M_NEWBLK, newblk_hash);
2399
hashdestroy(bmsafemap_hashtbl, M_BMSAFEMAP, bmsafemap_hash);
2400
free(indir_hashtbl, M_FREEWORK);
1431
2401
}
1432
2402
1433
2403
/*
1451
2421
mp->mnt_flag = (mp->mnt_flag & ~MNT_ASYNC) | MNT_SOFTDEP;
1452
2422
if ((mp->mnt_kern_flag & MNTK_SOFTDEP) == 0) {
1453
2423
mp->mnt_kern_flag = (mp->mnt_kern_flag & ~MNTK_ASYNC) |
1454
MNTK_SOFTDEP;
1455
mp->mnt_noasync++;
2424
MNTK_SOFTDEP | MNTK_NOASYNC;
1456
2425
}
1457
2426
MNT_IUNLOCK(mp);
1458
2427
ump = VFSTOUFS(mp);
1459
2428
LIST_INIT(&ump->softdep_workitem_pending);
2429
LIST_INIT(&ump->softdep_journal_pending);
2430
TAILQ_INIT(&ump->softdep_unlinked);
2431
LIST_INIT(&ump->softdep_dirtycg);
1460
2432
ump->softdep_worklist_tail = NULL;
1461
2433
ump->softdep_on_worklist = 0;
1462
2434
ump->softdep_deps = 0;
2435
if ((fs->fs_flags & FS_SUJ) &&
2436
(error = journal_mount(mp, fs, cred)) != 0) {
2437
printf("Failed to start journal: %d\n", error);
2438
return (error);
2439
}
1463
2440
/*
1464
2441
* When doing soft updates, the counters in the
1465
2442
* superblock may have gotten out of sync. Recomputation
1493
2470
return (0);
1494
2471
}
1495
2472
2473
void
2474
softdep_unmount(mp)
2475
struct mount *mp;
2476
{
2477
2478
MNT_ILOCK(mp);
2479
mp->mnt_flag &= ~MNT_SOFTDEP;
2480
if (MOUNTEDSUJ(mp) == 0) {
2481
MNT_IUNLOCK(mp);
2482
return;
2483
}
2484
mp->mnt_flag &= ~MNT_SUJ;
2485
MNT_IUNLOCK(mp);
2486
journal_unmount(mp);
2487
}
2488
2489
static struct jblocks *
2490
jblocks_create(void)
2491
{
2492
struct jblocks *jblocks;
2493
2494
jblocks = malloc(sizeof(*jblocks), M_JBLOCKS, M_WAITOK | M_ZERO);
2495
TAILQ_INIT(&jblocks->jb_segs);
2496
jblocks->jb_avail = 10;
2497
jblocks->jb_extent = malloc(sizeof(struct jextent) * jblocks->jb_avail,
2498
M_JBLOCKS, M_WAITOK | M_ZERO);
2499
2500
return (jblocks);
2501
}
2502
2503
static ufs2_daddr_t
2504
jblocks_alloc(jblocks, bytes, actual)
2505
struct jblocks *jblocks;
2506
int bytes;
2507
int *actual;
2508
{
2509
ufs2_daddr_t daddr;
2510
struct jextent *jext;
2511
int freecnt;
2512
int blocks;
2513
2514
blocks = bytes / DEV_BSIZE;
2515
jext = &jblocks->jb_extent[jblocks->jb_head];
2516
freecnt = jext->je_blocks - jblocks->jb_off;
2517
if (freecnt == 0) {
2518
jblocks->jb_off = 0;
2519
if (++jblocks->jb_head > jblocks->jb_used)
2520
jblocks->jb_head = 0;
2521
jext = &jblocks->jb_extent[jblocks->jb_head];
2522
freecnt = jext->je_blocks;
2523
}
2524
if (freecnt > blocks)
2525
freecnt = blocks;
2526
*actual = freecnt * DEV_BSIZE;
2527
daddr = jext->je_daddr + jblocks->jb_off;
2528
jblocks->jb_off += freecnt;
2529
jblocks->jb_free -= freecnt;
2530
2531
return (daddr);
2532
}
2533
2534
static void
2535
jblocks_free(jblocks, mp, bytes)
2536
struct jblocks *jblocks;
2537
struct mount *mp;
2538
int bytes;
2539
{
2540
2541
jblocks->jb_free += bytes / DEV_BSIZE;
2542
if (jblocks->jb_suspended)
2543
worklist_speedup();
2544
wakeup(jblocks);
2545
}
2546
2547
static void
2548
jblocks_destroy(jblocks)
2549
struct jblocks *jblocks;
2550
{
2551
2552
if (jblocks->jb_extent)
2553
free(jblocks->jb_extent, M_JBLOCKS);
2554
free(jblocks, M_JBLOCKS);
2555
}
2556
2557
static void
2558
jblocks_add(jblocks, daddr, blocks)
2559
struct jblocks *jblocks;
2560
ufs2_daddr_t daddr;
2561
int blocks;
2562
{
2563
struct jextent *jext;
2564
2565
jblocks->jb_blocks += blocks;
2566
jblocks->jb_free += blocks;
2567
jext = &jblocks->jb_extent[jblocks->jb_used];
2568
/* Adding the first block. */
2569
if (jext->je_daddr == 0) {
2570
jext->je_daddr = daddr;
2571
jext->je_blocks = blocks;
2572
return;
2573
}
2574
/* Extending the last extent. */
2575
if (jext->je_daddr + jext->je_blocks == daddr) {
2576
jext->je_blocks += blocks;
2577
return;
2578
}
2579
/* Adding a new extent. */
2580
if (++jblocks->jb_used == jblocks->jb_avail) {
2581
jblocks->jb_avail *= 2;
2582
jext = malloc(sizeof(struct jextent) * jblocks->jb_avail,
2583
M_JBLOCKS, M_WAITOK | M_ZERO);
2584
memcpy(jext, jblocks->jb_extent,
2585
sizeof(struct jextent) * jblocks->jb_used);
2586
free(jblocks->jb_extent, M_JBLOCKS);
2587
jblocks->jb_extent = jext;
2588
}
2589
jext = &jblocks->jb_extent[jblocks->jb_used];
2590
jext->je_daddr = daddr;
2591
jext->je_blocks = blocks;
2592
return;
2593
}
2594
2595
int
2596
softdep_journal_lookup(mp, vpp)
2597
struct mount *mp;
2598
struct vnode **vpp;
2599
{
2600
struct componentname cnp;
2601
struct vnode *dvp;
2602
ino_t sujournal;
2603
int error;
2604
2605
error = VFS_VGET(mp, ROOTINO, LK_EXCLUSIVE, &dvp);
2606
if (error)
2607
return (error);
2608
bzero(&cnp, sizeof(cnp));
2609
cnp.cn_nameiop = LOOKUP;
2610
cnp.cn_flags = ISLASTCN;
2611
cnp.cn_thread = curthread;
2612
cnp.cn_cred = curthread->td_ucred;
2613
cnp.cn_pnbuf = SUJ_FILE;
2614
cnp.cn_nameptr = SUJ_FILE;
2615
cnp.cn_namelen = strlen(SUJ_FILE);
2616
error = ufs_lookup_ino(dvp, NULL, &cnp, &sujournal);
2617
vput(dvp);
2618
if (error != 0)
2619
return (error);
2620
error = VFS_VGET(mp, sujournal, LK_EXCLUSIVE, vpp);
2621
return (error);
2622
}
2623
2624
/*
2625
* Open and verify the journal file.
2626
*/
2627
static int
2628
journal_mount(mp, fs, cred)
2629
struct mount *mp;
2630
struct fs *fs;
2631
struct ucred *cred;
2632
{
2633
struct jblocks *jblocks;
2634
struct vnode *vp;
2635
struct inode *ip;
2636
ufs2_daddr_t blkno;
2637
int bcount;
2638
int error;
2639
int i;
2640
2641
error = softdep_journal_lookup(mp, &vp);
2642
if (error != 0) {
2643
printf("Failed to find journal. Use tunefs to create one\n");
2644
return (error);
2645
}
2646
ip = VTOI(vp);
2647
if (ip->i_size < SUJ_MIN) {
2648
error = ENOSPC;
2649
goto out;
2650
}
2651
bcount = lblkno(fs, ip->i_size); /* Only use whole blocks. */
2652
jblocks = jblocks_create();
2653
for (i = 0; i < bcount; i++) {
2654
error = ufs_bmaparray(vp, i, &blkno, NULL, NULL, NULL);
2655
if (error)
2656
break;
2657
jblocks_add(jblocks, blkno, fsbtodb(fs, fs->fs_frag));
2658
}
2659
if (error) {
2660
jblocks_destroy(jblocks);
2661
goto out;
2662
}
2663
jblocks->jb_low = jblocks->jb_free / 3; /* Reserve 33%. */
2664
jblocks->jb_min = jblocks->jb_free / 10; /* Suspend at 10%. */
2665
VFSTOUFS(mp)->softdep_jblocks = jblocks;
2666
out:
2667
if (error == 0) {
2668
MNT_ILOCK(mp);
2669
mp->mnt_flag |= MNT_SUJ;
2670
mp->mnt_flag &= ~MNT_SOFTDEP;
2671
MNT_IUNLOCK(mp);
2672
/*
2673
* Only validate the journal contents if the
2674
* filesystem is clean, otherwise we write the logs
2675
* but they'll never be used. If the filesystem was
2676
* still dirty when we mounted it the journal is
2677
* invalid and a new journal can only be valid if it
2678
* starts from a clean mount.
2679
*/
2680
if (fs->fs_clean) {
2681
DIP_SET(ip, i_modrev, fs->fs_mtime);
2682
ip->i_flags |= IN_MODIFIED;
2683
ffs_update(vp, 1);
2684
}
2685
}
2686
vput(vp);
2687
return (error);
2688
}
2689
2690
static void
2691
journal_unmount(mp)
2692
struct mount *mp;
2693
{
2694
struct ufsmount *ump;
2695
2696
ump = VFSTOUFS(mp);
2697
if (ump->softdep_jblocks)
2698
jblocks_destroy(ump->softdep_jblocks);
2699
ump->softdep_jblocks = NULL;
2700
}
2701
2702
/*
2703
* Called when a journal record is ready to be written. Space is allocated
2704
* and the journal entry is created when the journal is flushed to stable
2705
* store.
2706
*/
2707
static void
2708
add_to_journal(wk)
2709
struct worklist *wk;
2710
{
2711
struct ufsmount *ump;
2712
2713
mtx_assert(&lk, MA_OWNED);
2714
ump = VFSTOUFS(wk->wk_mp);
2715
if (wk->wk_state & ONWORKLIST)
2716
panic("add_to_journal: %s(0x%X) already on list",
2717
TYPENAME(wk->wk_type), wk->wk_state);
2718
wk->wk_state |= ONWORKLIST | DEPCOMPLETE;
2719
if (LIST_EMPTY(&ump->softdep_journal_pending)) {
2720
ump->softdep_jblocks->jb_age = ticks;
2721
LIST_INSERT_HEAD(&ump->softdep_journal_pending, wk, wk_list);
2722
} else
2723
LIST_INSERT_AFTER(ump->softdep_journal_tail, wk, wk_list);
2724
ump->softdep_journal_tail = wk;
2725
ump->softdep_on_journal += 1;
2726
}
2727
2728
/*
2729
* Remove an arbitrary item for the journal worklist maintain the tail
2730
* pointer. This happens when a new operation obviates the need to
2731
* journal an old operation.
2732
*/
2733
static void
2734
remove_from_journal(wk)
2735
struct worklist *wk;
2736
{
2737
struct ufsmount *ump;
2738
2739
mtx_assert(&lk, MA_OWNED);
2740
ump = VFSTOUFS(wk->wk_mp);
2741
#ifdef SUJ_DEBUG
2742
{
2743
struct worklist *wkn;
2744
2745
LIST_FOREACH(wkn, &ump->softdep_journal_pending, wk_list)
2746
if (wkn == wk)
2747
break;
2748
if (wkn == NULL)
2749
panic("remove_from_journal: %p is not in journal", wk);
2750
}
2751
#endif
2752
/*
2753
* We emulate a TAILQ to save space in most structures which do not
2754
* require TAILQ semantics. Here we must update the tail position
2755
* when removing the tail which is not the final entry. This works
2756
* only if the worklist linkage are at the beginning of the structure.
2757
*/
2758
if (ump->softdep_journal_tail == wk)
2759
ump->softdep_journal_tail =
2760
(struct worklist *)wk->wk_list.le_prev;
2761
2762
WORKLIST_REMOVE(wk);
2763
ump->softdep_on_journal -= 1;
2764
}
2765
2766
/*
2767
* Check for journal space as well as dependency limits so the prelink
2768
* code can throttle both journaled and non-journaled filesystems.
2769
* Threshold is 0 for low and 1 for min.
2770
*/
2771
static int
2772
journal_space(ump, thresh)
2773
struct ufsmount *ump;
2774
int thresh;
2775
{
2776
struct jblocks *jblocks;
2777
int avail;
2778
2779
jblocks = ump->softdep_jblocks;
2780
if (jblocks == NULL)
2781
return (1);
2782
/*
2783
* We use a tighter restriction here to prevent request_cleanup()
2784
* running in threads from running into locks we currently hold.
2785
*/
2786
if (dep_current[D_INODEDEP] > (max_softdeps / 10) * 9)
2787
return (0);
2788
if (thresh)
2789
thresh = jblocks->jb_min;
2790
else
2791
thresh = jblocks->jb_low;
2792
avail = (ump->softdep_on_journal * JREC_SIZE) / DEV_BSIZE;
2793
avail = jblocks->jb_free - avail;
2794
2795
return (avail > thresh);
2796
}
2797
2798
static void
2799
journal_suspend(ump)
2800
struct ufsmount *ump;
2801
{
2802
struct jblocks *jblocks;
2803
struct mount *mp;
2804
2805
mp = UFSTOVFS(ump);
2806
jblocks = ump->softdep_jblocks;
2807
MNT_ILOCK(mp);
2808
if ((mp->mnt_kern_flag & MNTK_SUSPEND) == 0) {
2809
stat_journal_min++;
2810
mp->mnt_kern_flag |= MNTK_SUSPEND;
2811
mp->mnt_susp_owner = FIRST_THREAD_IN_PROC(softdepproc);
2812
}
2813
jblocks->jb_suspended = 1;
2814
MNT_IUNLOCK(mp);
2815
}
2816
2817
static int
2818
journal_unsuspend(struct ufsmount *ump)
2819
{
2820
struct jblocks *jblocks;
2821
struct mount *mp;
2822
2823
mp = UFSTOVFS(ump);
2824
jblocks = ump->softdep_jblocks;
2825
2826
if (jblocks != NULL && jblocks->jb_suspended &&
2827
journal_space(ump, jblocks->jb_min)) {
2828
jblocks->jb_suspended = 0;
2829
FREE_LOCK(&lk);
2830
mp->mnt_susp_owner = curthread;
2831
vfs_write_resume(mp);
2832
ACQUIRE_LOCK(&lk);
2833
return (1);
2834
}
2835
return (0);
2836
}
2837
2838
/*
2839
* Called before any allocation function to be certain that there is
2840
* sufficient space in the journal prior to creating any new records.
2841
* Since in the case of block allocation we may have multiple locked
2842
* buffers at the time of the actual allocation we can not block
2843
* when the journal records are created. Doing so would create a deadlock
2844
* if any of these buffers needed to be flushed to reclaim space. Instead
2845
* we require a sufficiently large amount of available space such that
2846
* each thread in the system could have passed this allocation check and
2847
* still have sufficient free space. With 20% of a minimum journal size
2848
* of 1MB we have 6553 records available.
2849
*/
2850
int
2851
softdep_prealloc(vp, waitok)
2852
struct vnode *vp;
2853
int waitok;
2854
{
2855
struct ufsmount *ump;
2856
2857
/*
2858
* Nothing to do if we are not running journaled soft updates.
2859
* If we currently hold the snapshot lock, we must avoid handling
2860
* other resources that could cause deadlock.
2861
*/
2862
if (DOINGSUJ(vp) == 0 || IS_SNAPSHOT(VTOI(vp)))
2863
return (0);
2864
ump = VFSTOUFS(vp->v_mount);
2865
ACQUIRE_LOCK(&lk);
2866
if (journal_space(ump, 0)) {
2867
FREE_LOCK(&lk);
2868
return (0);
2869
}
2870
stat_journal_low++;
2871
FREE_LOCK(&lk);
2872
if (waitok == MNT_NOWAIT)
2873
return (ENOSPC);
2874
/*
2875
* Attempt to sync this vnode once to flush any journal
2876
* work attached to it.
2877
*/
2878
if ((curthread->td_pflags & TDP_COWINPROGRESS) == 0)
2879
ffs_syncvnode(vp, waitok, 0);
2880
ACQUIRE_LOCK(&lk);
2881
process_removes(vp);
2882
process_truncates(vp);
2883
if (journal_space(ump, 0) == 0) {
2884
softdep_speedup();
2885
if (journal_space(ump, 1) == 0)
2886
journal_suspend(ump);
2887
}
2888
FREE_LOCK(&lk);
2889
2890
return (0);
2891
}
2892
2893
/*
2894
* Before adjusting a link count on a vnode verify that we have sufficient
2895
* journal space. If not, process operations that depend on the currently
2896
* locked pair of vnodes to try to flush space as the syncer, buf daemon,
2897
* and softdep flush threads can not acquire these locks to reclaim space.
2898
*/
2899
static void
2900
softdep_prelink(dvp, vp)
2901
struct vnode *dvp;
2902
struct vnode *vp;
2903
{
2904
struct ufsmount *ump;
2905
2906
ump = VFSTOUFS(dvp->v_mount);
2907
mtx_assert(&lk, MA_OWNED);
2908
/*
2909
* Nothing to do if we have sufficient journal space.
2910
* If we currently hold the snapshot lock, we must avoid
2911
* handling other resources that could cause deadlock.
2912
*/
2913
if (journal_space(ump, 0) || (vp && IS_SNAPSHOT(VTOI(vp))))
2914
return;
2915
stat_journal_low++;
2916
FREE_LOCK(&lk);
2917
if (vp)
2918
ffs_syncvnode(vp, MNT_NOWAIT, 0);
2919
ffs_syncvnode(dvp, MNT_WAIT, 0);
2920
ACQUIRE_LOCK(&lk);
2921
/* Process vp before dvp as it may create .. removes. */
2922
if (vp) {
2923
process_removes(vp);
2924
process_truncates(vp);
2925
}
2926
process_removes(dvp);
2927
process_truncates(dvp);
2928
softdep_speedup();
2929
process_worklist_item(UFSTOVFS(ump), 2, LK_NOWAIT);
2930
if (journal_space(ump, 0) == 0) {
2931
softdep_speedup();
2932
if (journal_space(ump, 1) == 0)
2933
journal_suspend(ump);
2934
}
2935
}
2936
2937
static void
2938
jseg_write(ump, jseg, data)
2939
struct ufsmount *ump;
2940
struct jseg *jseg;
2941
uint8_t *data;
2942
{
2943
struct jsegrec *rec;
2944
2945
rec = (struct jsegrec *)data;
2946
rec->jsr_seq = jseg->js_seq;
2947
rec->jsr_oldest = jseg->js_oldseq;
2948
rec->jsr_cnt = jseg->js_cnt;
2949
rec->jsr_blocks = jseg->js_size / ump->um_devvp->v_bufobj.bo_bsize;
2950
rec->jsr_crc = 0;
2951
rec->jsr_time = ump->um_fs->fs_mtime;
2952
}
2953
2954
static inline void
2955
inoref_write(inoref, jseg, rec)
2956
struct inoref *inoref;
2957
struct jseg *jseg;
2958
struct jrefrec *rec;
2959
{
2960
2961
inoref->if_jsegdep->jd_seg = jseg;
2962
rec->jr_ino = inoref->if_ino;
2963
rec->jr_parent = inoref->if_parent;
2964
rec->jr_nlink = inoref->if_nlink;
2965
rec->jr_mode = inoref->if_mode;
2966
rec->jr_diroff = inoref->if_diroff;
2967
}
2968
2969
static void
2970
jaddref_write(jaddref, jseg, data)
2971
struct jaddref *jaddref;
2972
struct jseg *jseg;
2973
uint8_t *data;
2974
{
2975
struct jrefrec *rec;
2976
2977
rec = (struct jrefrec *)data;
2978
rec->jr_op = JOP_ADDREF;
2979
inoref_write(&jaddref->ja_ref, jseg, rec);
2980
}
2981
2982
static void
2983
jremref_write(jremref, jseg, data)
2984
struct jremref *jremref;
2985
struct jseg *jseg;
2986
uint8_t *data;
2987
{
2988
struct jrefrec *rec;
2989
2990
rec = (struct jrefrec *)data;
2991
rec->jr_op = JOP_REMREF;
2992
inoref_write(&jremref->jr_ref, jseg, rec);
2993
}
2994
2995
static void
2996
jmvref_write(jmvref, jseg, data)
2997
struct jmvref *jmvref;
2998
struct jseg *jseg;
2999
uint8_t *data;
3000
{
3001
struct jmvrec *rec;
3002
3003
rec = (struct jmvrec *)data;
3004
rec->jm_op = JOP_MVREF;
3005
rec->jm_ino = jmvref->jm_ino;
3006
rec->jm_parent = jmvref->jm_parent;
3007
rec->jm_oldoff = jmvref->jm_oldoff;
3008
rec->jm_newoff = jmvref->jm_newoff;
3009
}
3010
3011
static void
3012
jnewblk_write(jnewblk, jseg, data)
3013
struct jnewblk *jnewblk;
3014
struct jseg *jseg;
3015
uint8_t *data;
3016
{
3017
struct jblkrec *rec;
3018
3019
jnewblk->jn_jsegdep->jd_seg = jseg;
3020
rec = (struct jblkrec *)data;
3021
rec->jb_op = JOP_NEWBLK;
3022
rec->jb_ino = jnewblk->jn_ino;
3023
rec->jb_blkno = jnewblk->jn_blkno;
3024
rec->jb_lbn = jnewblk->jn_lbn;
3025
rec->jb_frags = jnewblk->jn_frags;
3026
rec->jb_oldfrags = jnewblk->jn_oldfrags;
3027
}
3028
3029
static void
3030
jfreeblk_write(jfreeblk, jseg, data)
3031
struct jfreeblk *jfreeblk;
3032
struct jseg *jseg;
3033
uint8_t *data;
3034
{
3035
struct jblkrec *rec;
3036
3037
jfreeblk->jf_dep.jb_jsegdep->jd_seg = jseg;
3038
rec = (struct jblkrec *)data;
3039
rec->jb_op = JOP_FREEBLK;
3040
rec->jb_ino = jfreeblk->jf_ino;
3041
rec->jb_blkno = jfreeblk->jf_blkno;
3042
rec->jb_lbn = jfreeblk->jf_lbn;
3043
rec->jb_frags = jfreeblk->jf_frags;
3044
rec->jb_oldfrags = 0;
3045
}
3046
3047
static void
3048
jfreefrag_write(jfreefrag, jseg, data)
3049
struct jfreefrag *jfreefrag;
3050
struct jseg *jseg;
3051
uint8_t *data;
3052
{
3053
struct jblkrec *rec;
3054
3055
jfreefrag->fr_jsegdep->jd_seg = jseg;
3056
rec = (struct jblkrec *)data;
3057
rec->jb_op = JOP_FREEBLK;
3058
rec->jb_ino = jfreefrag->fr_ino;
3059
rec->jb_blkno = jfreefrag->fr_blkno;
3060
rec->jb_lbn = jfreefrag->fr_lbn;
3061
rec->jb_frags = jfreefrag->fr_frags;
3062
rec->jb_oldfrags = 0;
3063
}
3064
3065
static void
3066
jtrunc_write(jtrunc, jseg, data)
3067
struct jtrunc *jtrunc;
3068
struct jseg *jseg;
3069
uint8_t *data;
3070
{
3071
struct jtrncrec *rec;
3072
3073
jtrunc->jt_dep.jb_jsegdep->jd_seg = jseg;
3074
rec = (struct jtrncrec *)data;
3075
rec->jt_op = JOP_TRUNC;
3076
rec->jt_ino = jtrunc->jt_ino;
3077
rec->jt_size = jtrunc->jt_size;
3078
rec->jt_extsize = jtrunc->jt_extsize;
3079
}
3080
3081
static void
3082
jfsync_write(jfsync, jseg, data)
3083
struct jfsync *jfsync;
3084
struct jseg *jseg;
3085
uint8_t *data;
3086
{
3087
struct jtrncrec *rec;
3088
3089
rec = (struct jtrncrec *)data;
3090
rec->jt_op = JOP_SYNC;
3091
rec->jt_ino = jfsync->jfs_ino;
3092
rec->jt_size = jfsync->jfs_size;
3093
rec->jt_extsize = jfsync->jfs_extsize;
3094
}
3095
3096
static void
3097
softdep_flushjournal(mp)
3098
struct mount *mp;
3099
{
3100
struct jblocks *jblocks;
3101
struct ufsmount *ump;
3102
3103
if (MOUNTEDSUJ(mp) == 0)
3104
return;
3105
ump = VFSTOUFS(mp);
3106
jblocks = ump->softdep_jblocks;
3107
ACQUIRE_LOCK(&lk);
3108
while (ump->softdep_on_journal) {
3109
jblocks->jb_needseg = 1;
3110
softdep_process_journal(mp, NULL, MNT_WAIT);
3111
}
3112
FREE_LOCK(&lk);
3113
}
3114
3115
static void softdep_synchronize_completed(struct bio *);
3116
static void softdep_synchronize(struct bio *, struct ufsmount *, void *);
3117
3118
static void
3119
softdep_synchronize_completed(bp)
3120
struct bio *bp;
3121
{
3122
struct jseg *oldest;
3123
struct jseg *jseg;
3124
3125
/*
3126
* caller1 marks the last segment written before we issued the
3127
* synchronize cache.
3128
*/
3129
jseg = bp->bio_caller1;
3130
oldest = NULL;
3131
ACQUIRE_LOCK(&lk);
3132
/*
3133
* Mark all the journal entries waiting on the synchronize cache
3134
* as completed so they may continue on.
3135
*/
3136
while (jseg != NULL && (jseg->js_state & COMPLETE) == 0) {
3137
jseg->js_state |= COMPLETE;
3138
oldest = jseg;
3139
jseg = TAILQ_PREV(jseg, jseglst, js_next);
3140
}
3141
/*
3142
* Restart deferred journal entry processing from the oldest
3143
* completed jseg.
3144
*/
3145
if (oldest)
3146
complete_jsegs(oldest);
3147
3148
FREE_LOCK(&lk);
3149
g_destroy_bio(bp);
3150
}
3151
3152
/*
3153
* Send BIO_FLUSH/SYNCHRONIZE CACHE to the device to enforce write ordering
3154
* barriers. The journal must be written prior to any blocks that depend
3155
* on it and the journal can not be released until the blocks have be
3156
* written. This code handles both barriers simultaneously.
3157
*/
3158
static void
3159
softdep_synchronize(bp, ump, caller1)
3160
struct bio *bp;
3161
struct ufsmount *ump;
3162
void *caller1;
3163
{
3164
3165
bp->bio_cmd = BIO_FLUSH;
3166
bp->bio_flags |= BIO_ORDERED;
3167
bp->bio_data = NULL;
3168
bp->bio_offset = ump->um_cp->provider->mediasize;
3169
bp->bio_length = 0;
3170
bp->bio_done = softdep_synchronize_completed;
3171
bp->bio_caller1 = caller1;
3172
g_io_request(bp,
3173
(struct g_consumer *)ump->um_devvp->v_bufobj.bo_private);
3174
}
3175
3176
/*
3177
* Flush some journal records to disk.
3178
*/
3179
static void
3180
softdep_process_journal(mp, needwk, flags)
3181
struct mount *mp;
3182
struct worklist *needwk;
3183
int flags;
3184
{
3185
struct jblocks *jblocks;
3186
struct ufsmount *ump;
3187
struct worklist *wk;
3188
struct jseg *jseg;
3189
struct buf *bp;
3190
struct bio *bio;
3191
uint8_t *data;
3192
struct fs *fs;
3193
int shouldflush;
3194
int segwritten;
3195
int jrecmin; /* Minimum records per block. */
3196
int jrecmax; /* Maximum records per block. */
3197
int size;
3198
int cnt;
3199
int off;
3200
int devbsize;
3201
3202
if (MOUNTEDSUJ(mp) == 0)
3203
return;
3204
shouldflush = softdep_flushcache;
3205
bio = NULL;
3206
jseg = NULL;
3207
ump = VFSTOUFS(mp);
3208
fs = ump->um_fs;
3209
jblocks = ump->softdep_jblocks;
3210
devbsize = ump->um_devvp->v_bufobj.bo_bsize;
3211
/*
3212
* We write anywhere between a disk block and fs block. The upper
3213
* bound is picked to prevent buffer cache fragmentation and limit
3214
* processing time per I/O.
3215
*/
3216
jrecmin = (devbsize / JREC_SIZE) - 1; /* -1 for seg header */
3217
jrecmax = (fs->fs_bsize / devbsize) * jrecmin;
3218
segwritten = 0;
3219
for (;;) {
3220
cnt = ump->softdep_on_journal;
3221
/*
3222
* Criteria for writing a segment:
3223
* 1) We have a full block.
3224
* 2) We're called from jwait() and haven't found the
3225
* journal item yet.
3226
* 3) Always write if needseg is set.
3227
* 4) If we are called from process_worklist and have
3228
* not yet written anything we write a partial block
3229
* to enforce a 1 second maximum latency on journal
3230
* entries.
3231
*/
3232
if (cnt < (jrecmax - 1) && needwk == NULL &&
3233
jblocks->jb_needseg == 0 && (segwritten || cnt == 0))
3234
break;
3235
cnt++;
3236
/*
3237
* Verify some free journal space. softdep_prealloc() should
3238
* guarantee that we don't run out so this is indicative of
3239
* a problem with the flow control. Try to recover
3240
* gracefully in any event.
3241
*/
3242
while (jblocks->jb_free == 0) {
3243
if (flags != MNT_WAIT)
3244
break;
3245
printf("softdep: Out of journal space!\n");
3246
softdep_speedup();
3247
msleep(jblocks, &lk, PRIBIO, "jblocks", hz);
3248
}
3249
FREE_LOCK(&lk);
3250
jseg = malloc(sizeof(*jseg), M_JSEG, M_SOFTDEP_FLAGS);
3251
workitem_alloc(&jseg->js_list, D_JSEG, mp);
3252
LIST_INIT(&jseg->js_entries);
3253
LIST_INIT(&jseg->js_indirs);
3254
jseg->js_state = ATTACHED;
3255
if (shouldflush == 0)
3256
jseg->js_state |= COMPLETE;
3257
else if (bio == NULL)
3258
bio = g_alloc_bio();
3259
jseg->js_jblocks = jblocks;
3260
bp = geteblk(fs->fs_bsize, 0);
3261
ACQUIRE_LOCK(&lk);
3262
/*
3263
* If there was a race while we were allocating the block
3264
* and jseg the entry we care about was likely written.
3265
* We bail out in both the WAIT and NOWAIT case and assume
3266
* the caller will loop if the entry it cares about is
3267
* not written.
3268
*/
3269
cnt = ump->softdep_on_journal;
3270
if (cnt + jblocks->jb_needseg == 0 || jblocks->jb_free == 0) {
3271
bp->b_flags |= B_INVAL | B_NOCACHE;
3272
WORKITEM_FREE(jseg, D_JSEG);
3273
FREE_LOCK(&lk);
3274
brelse(bp);
3275
ACQUIRE_LOCK(&lk);
3276
break;
3277
}
3278
/*
3279
* Calculate the disk block size required for the available
3280
* records rounded to the min size.
3281
*/
3282
if (cnt == 0)
3283
size = devbsize;
3284
else if (cnt < jrecmax)
3285
size = howmany(cnt, jrecmin) * devbsize;
3286
else
3287
size = fs->fs_bsize;
3288
/*
3289
* Allocate a disk block for this journal data and account
3290
* for truncation of the requested size if enough contiguous
3291
* space was not available.
3292
*/
3293
bp->b_blkno = jblocks_alloc(jblocks, size, &size);
3294
bp->b_lblkno = bp->b_blkno;
3295
bp->b_offset = bp->b_blkno * DEV_BSIZE;
3296
bp->b_bcount = size;
3297
bp->b_bufobj = &ump->um_devvp->v_bufobj;
3298
bp->b_flags &= ~B_INVAL;
3299
bp->b_flags |= B_VALIDSUSPWRT | B_NOCOPY;
3300
/*
3301
* Initialize our jseg with cnt records. Assign the next
3302
* sequence number to it and link it in-order.
3303
*/
3304
cnt = MIN(cnt, (size / devbsize) * jrecmin);
3305
jseg->js_buf = bp;
3306
jseg->js_cnt = cnt;
3307
jseg->js_refs = cnt + 1; /* Self ref. */
3308
jseg->js_size = size;
3309
jseg->js_seq = jblocks->jb_nextseq++;
3310
if (jblocks->jb_oldestseg == NULL)
3311
jblocks->jb_oldestseg = jseg;
3312
jseg->js_oldseq = jblocks->jb_oldestseg->js_seq;
3313
TAILQ_INSERT_TAIL(&jblocks->jb_segs, jseg, js_next);
3314
if (jblocks->jb_writeseg == NULL)
3315
jblocks->jb_writeseg = jseg;
3316
/*
3317
* Start filling in records from the pending list.
3318
*/
3319
data = bp->b_data;
3320
off = 0;
3321
while ((wk = LIST_FIRST(&ump->softdep_journal_pending))
3322
!= NULL) {
3323
if (cnt == 0)
3324
break;
3325
/* Place a segment header on every device block. */
3326
if ((off % devbsize) == 0) {
3327
jseg_write(ump, jseg, data);
3328
off += JREC_SIZE;
3329
data = bp->b_data + off;
3330
}
3331
if (wk == needwk)
3332
needwk = NULL;
3333
remove_from_journal(wk);
3334
wk->wk_state |= INPROGRESS;
3335
WORKLIST_INSERT(&jseg->js_entries, wk);
3336
switch (wk->wk_type) {
3337
case D_JADDREF:
3338
jaddref_write(WK_JADDREF(wk), jseg, data);
3339
break;
3340
case D_JREMREF:
3341
jremref_write(WK_JREMREF(wk), jseg, data);
3342
break;
3343
case D_JMVREF:
3344
jmvref_write(WK_JMVREF(wk), jseg, data);
3345
break;
3346
case D_JNEWBLK:
3347
jnewblk_write(WK_JNEWBLK(wk), jseg, data);
3348
break;
3349
case D_JFREEBLK:
3350
jfreeblk_write(WK_JFREEBLK(wk), jseg, data);
3351
break;
3352
case D_JFREEFRAG:
3353
jfreefrag_write(WK_JFREEFRAG(wk), jseg, data);
3354
break;
3355
case D_JTRUNC:
3356
jtrunc_write(WK_JTRUNC(wk), jseg, data);
3357
break;
3358
case D_JFSYNC:
3359
jfsync_write(WK_JFSYNC(wk), jseg, data);
3360
break;
3361
default:
3362
panic("process_journal: Unknown type %s",
3363
TYPENAME(wk->wk_type));
3364
/* NOTREACHED */
3365
}
3366
off += JREC_SIZE;
3367
data = bp->b_data + off;
3368
cnt--;
3369
}
3370
/*
3371
* Write this one buffer and continue.
3372
*/
3373
segwritten = 1;
3374
jblocks->jb_needseg = 0;
3375
WORKLIST_INSERT(&bp->b_dep, &jseg->js_list);
3376
FREE_LOCK(&lk);
3377
BO_LOCK(bp->b_bufobj);
3378
bgetvp(ump->um_devvp, bp);
3379
BO_UNLOCK(bp->b_bufobj);
3380
/*
3381
* We only do the blocking wait once we find the journal
3382
* entry we're looking for.
3383
*/
3384
if (needwk == NULL && flags == MNT_WAIT)
3385
bwrite(bp);
3386
else
3387
bawrite(bp);
3388
ACQUIRE_LOCK(&lk);
3389
}
3390
/*
3391
* If we wrote a segment issue a synchronize cache so the journal
3392
* is reflected on disk before the data is written. Since reclaiming
3393
* journal space also requires writing a journal record this
3394
* process also enforces a barrier before reclamation.
3395
*/
3396
if (segwritten && shouldflush) {
3397
softdep_synchronize(bio, ump,
3398
TAILQ_LAST(&jblocks->jb_segs, jseglst));
3399
} else if (bio)
3400
g_destroy_bio(bio);
3401
/*
3402
* If we've suspended the filesystem because we ran out of journal
3403
* space either try to sync it here to make some progress or
3404
* unsuspend it if we already have.
3405
*/
3406
if (flags == 0 && jblocks->jb_suspended) {
3407
if (journal_unsuspend(ump))
3408
return;
3409
FREE_LOCK(&lk);
3410
VFS_SYNC(mp, MNT_NOWAIT);
3411
ffs_sbupdate(ump, MNT_WAIT, 0);
3412
ACQUIRE_LOCK(&lk);
3413
}
3414
}
3415
3416
/*
3417
* Complete a jseg, allowing all dependencies awaiting journal writes
3418
* to proceed. Each journal dependency also attaches a jsegdep to dependent
3419
* structures so that the journal segment can be freed to reclaim space.
3420
*/
3421
static void
3422
complete_jseg(jseg)
3423
struct jseg *jseg;
3424
{
3425
struct worklist *wk;
3426
struct jmvref *jmvref;
3427
int waiting;
3428
#ifdef INVARIANTS
3429
int i = 0;
3430
#endif
3431
3432
while ((wk = LIST_FIRST(&jseg->js_entries)) != NULL) {
3433
WORKLIST_REMOVE(wk);
3434
waiting = wk->wk_state & IOWAITING;
3435
wk->wk_state &= ~(INPROGRESS | IOWAITING);
3436
wk->wk_state |= COMPLETE;
3437
KASSERT(i++ < jseg->js_cnt,
3438
("handle_written_jseg: overflow %d >= %d",
3439
i - 1, jseg->js_cnt));
3440
switch (wk->wk_type) {
3441
case D_JADDREF:
3442
handle_written_jaddref(WK_JADDREF(wk));
3443
break;
3444
case D_JREMREF:
3445
handle_written_jremref(WK_JREMREF(wk));
3446
break;
3447
case D_JMVREF:
3448
rele_jseg(jseg); /* No jsegdep. */
3449
jmvref = WK_JMVREF(wk);
3450
LIST_REMOVE(jmvref, jm_deps);
3451
if ((jmvref->jm_pagedep->pd_state & ONWORKLIST) == 0)
3452
free_pagedep(jmvref->jm_pagedep);
3453
WORKITEM_FREE(jmvref, D_JMVREF);
3454
break;
3455
case D_JNEWBLK:
3456
handle_written_jnewblk(WK_JNEWBLK(wk));
3457
break;
3458
case D_JFREEBLK:
3459
handle_written_jblkdep(&WK_JFREEBLK(wk)->jf_dep);
3460
break;
3461
case D_JTRUNC:
3462
handle_written_jblkdep(&WK_JTRUNC(wk)->jt_dep);
3463
break;
3464
case D_JFSYNC:
3465
rele_jseg(jseg); /* No jsegdep. */
3466
WORKITEM_FREE(wk, D_JFSYNC);
3467
break;
3468
case D_JFREEFRAG:
3469
handle_written_jfreefrag(WK_JFREEFRAG(wk));
3470
break;
3471
default:
3472
panic("handle_written_jseg: Unknown type %s",
3473
TYPENAME(wk->wk_type));
3474
/* NOTREACHED */
3475
}
3476
if (waiting)
3477
wakeup(wk);
3478
}
3479
/* Release the self reference so the structure may be freed. */
3480
rele_jseg(jseg);
3481
}
3482
3483
/*
3484
* Determine which jsegs are ready for completion processing. Waits for
3485
* synchronize cache to complete as well as forcing in-order completion
3486
* of journal entries.
3487
*/
3488
static void
3489
complete_jsegs(jseg)
3490
struct jseg *jseg;
3491
{
3492
struct jblocks *jblocks;
3493
struct jseg *jsegn;
3494
3495
jblocks = jseg->js_jblocks;
3496
/*
3497
* Don't allow out of order completions. If this isn't the first
3498
* block wait for it to write before we're done.
3499
*/
3500
if (jseg != jblocks->jb_writeseg)
3501
return;
3502
/* Iterate through available jsegs processing their entries. */
3503
while (jseg && (jseg->js_state & ALLCOMPLETE) == ALLCOMPLETE) {
3504
jblocks->jb_oldestwrseq = jseg->js_oldseq;
3505
jsegn = TAILQ_NEXT(jseg, js_next);
3506
complete_jseg(jseg);
3507
jseg = jsegn;
3508
}
3509
jblocks->jb_writeseg = jseg;
3510
/*
3511
* Attempt to free jsegs now that oldestwrseq may have advanced.
3512
*/
3513
free_jsegs(jblocks);
3514
}
3515
3516
/*
3517
* Mark a jseg as DEPCOMPLETE and throw away the buffer. Attempt to handle
3518
* the final completions.
3519
*/
3520
static void
3521
handle_written_jseg(jseg, bp)
3522
struct jseg *jseg;
3523
struct buf *bp;
3524
{
3525
3526
if (jseg->js_refs == 0)
3527
panic("handle_written_jseg: No self-reference on %p", jseg);
3528
jseg->js_state |= DEPCOMPLETE;
3529
/*
3530
* We'll never need this buffer again, set flags so it will be
3531
* discarded.
3532
*/
3533
bp->b_flags |= B_INVAL | B_NOCACHE;
3534
complete_jsegs(jseg);
3535
}
3536
3537
static inline struct jsegdep *
3538
inoref_jseg(inoref)
3539
struct inoref *inoref;
3540
{
3541
struct jsegdep *jsegdep;
3542
3543
jsegdep = inoref->if_jsegdep;
3544
inoref->if_jsegdep = NULL;
3545
3546
return (jsegdep);
3547
}
3548
3549
/*
3550
* Called once a jremref has made it to stable store. The jremref is marked
3551
* complete and we attempt to free it. Any pagedeps writes sleeping waiting
3552
* for the jremref to complete will be awoken by free_jremref.
3553
*/
3554
static void
3555
handle_written_jremref(jremref)
3556
struct jremref *jremref;
3557
{
3558
struct inodedep *inodedep;
3559
struct jsegdep *jsegdep;
3560
struct dirrem *dirrem;
3561
3562
/* Grab the jsegdep. */
3563
jsegdep = inoref_jseg(&jremref->jr_ref);
3564
/*
3565
* Remove us from the inoref list.
3566
*/
3567
if (inodedep_lookup(jremref->jr_list.wk_mp, jremref->jr_ref.if_ino,
3568
0, &inodedep) == 0)
3569
panic("handle_written_jremref: Lost inodedep");
3570
TAILQ_REMOVE(&inodedep->id_inoreflst, &jremref->jr_ref, if_deps);
3571
/*
3572
* Complete the dirrem.
3573
*/
3574
dirrem = jremref->jr_dirrem;
3575
jremref->jr_dirrem = NULL;
3576
LIST_REMOVE(jremref, jr_deps);
3577
jsegdep->jd_state |= jremref->jr_state & MKDIR_PARENT;
3578
jwork_insert(&dirrem->dm_jwork, jsegdep);
3579
if (LIST_EMPTY(&dirrem->dm_jremrefhd) &&
3580
(dirrem->dm_state & COMPLETE) != 0)
3581
add_to_worklist(&dirrem->dm_list, 0);
3582
free_jremref(jremref);
3583
}
3584
3585
/*
3586
* Called once a jaddref has made it to stable store. The dependency is
3587
* marked complete and any dependent structures are added to the inode
3588
* bufwait list to be completed as soon as it is written. If a bitmap write
3589
* depends on this entry we move the inode into the inodedephd of the
3590
* bmsafemap dependency and attempt to remove the jaddref from the bmsafemap.
3591
*/
3592
static void
3593
handle_written_jaddref(jaddref)
3594
struct jaddref *jaddref;
3595
{
3596
struct jsegdep *jsegdep;
3597
struct inodedep *inodedep;
3598
struct diradd *diradd;
3599
struct mkdir *mkdir;
3600
3601
/* Grab the jsegdep. */
3602
jsegdep = inoref_jseg(&jaddref->ja_ref);
3603
mkdir = NULL;
3604
diradd = NULL;
3605
if (inodedep_lookup(jaddref->ja_list.wk_mp, jaddref->ja_ino,
3606
0, &inodedep) == 0)
3607
panic("handle_written_jaddref: Lost inodedep.");
3608
if (jaddref->ja_diradd == NULL)
3609
panic("handle_written_jaddref: No dependency");
3610
if (jaddref->ja_diradd->da_list.wk_type == D_DIRADD) {
3611
diradd = jaddref->ja_diradd;
3612
WORKLIST_INSERT(&inodedep->id_bufwait, &diradd->da_list);
3613
} else if (jaddref->ja_state & MKDIR_PARENT) {
3614
mkdir = jaddref->ja_mkdir;
3615
WORKLIST_INSERT(&inodedep->id_bufwait, &mkdir->md_list);
3616
} else if (jaddref->ja_state & MKDIR_BODY)
3617
mkdir = jaddref->ja_mkdir;
3618
else
3619
panic("handle_written_jaddref: Unknown dependency %p",
3620
jaddref->ja_diradd);
3621
jaddref->ja_diradd = NULL; /* also clears ja_mkdir */
3622
/*
3623
* Remove us from the inode list.
3624
*/
3625
TAILQ_REMOVE(&inodedep->id_inoreflst, &jaddref->ja_ref, if_deps);
3626
/*
3627
* The mkdir may be waiting on the jaddref to clear before freeing.
3628
*/
3629
if (mkdir) {
3630
KASSERT(mkdir->md_list.wk_type == D_MKDIR,
3631
("handle_written_jaddref: Incorrect type for mkdir %s",
3632
TYPENAME(mkdir->md_list.wk_type)));
3633
mkdir->md_jaddref = NULL;
3634
diradd = mkdir->md_diradd;
3635
mkdir->md_state |= DEPCOMPLETE;
3636
complete_mkdir(mkdir);
3637
}
3638
jwork_insert(&diradd->da_jwork, jsegdep);
3639
if (jaddref->ja_state & NEWBLOCK) {
3640
inodedep->id_state |= ONDEPLIST;
3641
LIST_INSERT_HEAD(&inodedep->id_bmsafemap->sm_inodedephd,
3642
inodedep, id_deps);
3643
}
3644
free_jaddref(jaddref);
3645
}
3646
3647
/*
3648
* Called once a jnewblk journal is written. The allocdirect or allocindir
3649
* is placed in the bmsafemap to await notification of a written bitmap. If
3650
* the operation was canceled we add the segdep to the appropriate
3651
* dependency to free the journal space once the canceling operation
3652
* completes.
3653
*/
3654
static void
3655
handle_written_jnewblk(jnewblk)
3656
struct jnewblk *jnewblk;
3657
{
3658
struct bmsafemap *bmsafemap;
3659
struct freefrag *freefrag;
3660
struct freework *freework;
3661
struct jsegdep *jsegdep;
3662
struct newblk *newblk;
3663
3664
/* Grab the jsegdep. */
3665
jsegdep = jnewblk->jn_jsegdep;
3666
jnewblk->jn_jsegdep = NULL;
3667
if (jnewblk->jn_dep == NULL)
3668
panic("handle_written_jnewblk: No dependency for the segdep.");
3669
switch (jnewblk->jn_dep->wk_type) {
3670
case D_NEWBLK:
3671
case D_ALLOCDIRECT:
3672
case D_ALLOCINDIR:
3673
/*
3674
* Add the written block to the bmsafemap so it can
3675
* be notified when the bitmap is on disk.
3676
*/
3677
newblk = WK_NEWBLK(jnewblk->jn_dep);
3678
newblk->nb_jnewblk = NULL;
3679
if ((newblk->nb_state & GOINGAWAY) == 0) {
3680
bmsafemap = newblk->nb_bmsafemap;
3681
newblk->nb_state |= ONDEPLIST;
3682
LIST_INSERT_HEAD(&bmsafemap->sm_newblkhd, newblk,
3683
nb_deps);
3684
}
3685
jwork_insert(&newblk->nb_jwork, jsegdep);
3686
break;
3687
case D_FREEFRAG:
3688
/*
3689
* A newblock being removed by a freefrag when replaced by
3690
* frag extension.
3691
*/
3692
freefrag = WK_FREEFRAG(jnewblk->jn_dep);
3693
freefrag->ff_jdep = NULL;
3694
jwork_insert(&freefrag->ff_jwork, jsegdep);
3695
break;
3696
case D_FREEWORK:
3697
/*
3698
* A direct block was removed by truncate.
3699
*/
3700
freework = WK_FREEWORK(jnewblk->jn_dep);
3701
freework->fw_jnewblk = NULL;
3702
jwork_insert(&freework->fw_freeblks->fb_jwork, jsegdep);
3703
break;
3704
default:
3705
panic("handle_written_jnewblk: Unknown type %d.",
3706
jnewblk->jn_dep->wk_type);
3707
}
3708
jnewblk->jn_dep = NULL;
3709
free_jnewblk(jnewblk);
3710
}
3711
3712
/*
3713
* Cancel a jfreefrag that won't be needed, probably due to colliding with
3714
* an in-flight allocation that has not yet been committed. Divorce us
3715
* from the freefrag and mark it DEPCOMPLETE so that it may be added
3716
* to the worklist.
3717
*/
3718
static void
3719
cancel_jfreefrag(jfreefrag)
3720
struct jfreefrag *jfreefrag;
3721
{
3722
struct freefrag *freefrag;
3723
3724
if (jfreefrag->fr_jsegdep) {
3725
free_jsegdep(jfreefrag->fr_jsegdep);
3726
jfreefrag->fr_jsegdep = NULL;
3727
}
3728
freefrag = jfreefrag->fr_freefrag;
3729
jfreefrag->fr_freefrag = NULL;
3730
free_jfreefrag(jfreefrag);
3731
freefrag->ff_state |= DEPCOMPLETE;
3732
CTR1(KTR_SUJ, "cancel_jfreefrag: blkno %jd", freefrag->ff_blkno);
3733
}
3734
3735
/*
3736
* Free a jfreefrag when the parent freefrag is rendered obsolete.
3737
*/
3738
static void
3739
free_jfreefrag(jfreefrag)
3740
struct jfreefrag *jfreefrag;
3741
{
3742
3743
if (jfreefrag->fr_state & INPROGRESS)
3744
WORKLIST_REMOVE(&jfreefrag->fr_list);
3745
else if (jfreefrag->fr_state & ONWORKLIST)
3746
remove_from_journal(&jfreefrag->fr_list);
3747
if (jfreefrag->fr_freefrag != NULL)
3748
panic("free_jfreefrag: Still attached to a freefrag.");
3749
WORKITEM_FREE(jfreefrag, D_JFREEFRAG);
3750
}
3751
3752
/*
3753
* Called when the journal write for a jfreefrag completes. The parent
3754
* freefrag is added to the worklist if this completes its dependencies.
3755
*/
3756
static void
3757
handle_written_jfreefrag(jfreefrag)
3758
struct jfreefrag *jfreefrag;
3759
{
3760
struct jsegdep *jsegdep;
3761
struct freefrag *freefrag;
3762
3763
/* Grab the jsegdep. */
3764
jsegdep = jfreefrag->fr_jsegdep;
3765
jfreefrag->fr_jsegdep = NULL;
3766
freefrag = jfreefrag->fr_freefrag;
3767
if (freefrag == NULL)
3768
panic("handle_written_jfreefrag: No freefrag.");
3769
freefrag->ff_state |= DEPCOMPLETE;
3770
freefrag->ff_jdep = NULL;
3771
jwork_insert(&freefrag->ff_jwork, jsegdep);
3772
if ((freefrag->ff_state & ALLCOMPLETE) == ALLCOMPLETE)
3773
add_to_worklist(&freefrag->ff_list, 0);
3774
jfreefrag->fr_freefrag = NULL;
3775
free_jfreefrag(jfreefrag);
3776
}
3777
3778
/*
3779
* Called when the journal write for a jfreeblk completes. The jfreeblk
3780
* is removed from the freeblks list of pending journal writes and the
3781
* jsegdep is moved to the freeblks jwork to be completed when all blocks
3782
* have been reclaimed.
3783
*/
3784
static void
3785
handle_written_jblkdep(jblkdep)
3786
struct jblkdep *jblkdep;
3787
{
3788
struct freeblks *freeblks;
3789
struct jsegdep *jsegdep;
3790
3791
/* Grab the jsegdep. */
3792
jsegdep = jblkdep->jb_jsegdep;
3793
jblkdep->jb_jsegdep = NULL;
3794
freeblks = jblkdep->jb_freeblks;
3795
LIST_REMOVE(jblkdep, jb_deps);
3796
jwork_insert(&freeblks->fb_jwork, jsegdep);
3797
/*
3798
* If the freeblks is all journaled, we can add it to the worklist.
3799
*/
3800
if (LIST_EMPTY(&freeblks->fb_jblkdephd) &&
3801
(freeblks->fb_state & ALLCOMPLETE) == ALLCOMPLETE)
3802
add_to_worklist(&freeblks->fb_list, WK_NODELAY);
3803
3804
free_jblkdep(jblkdep);
3805
}
3806
3807
static struct jsegdep *
3808
newjsegdep(struct worklist *wk)
3809
{
3810
struct jsegdep *jsegdep;
3811
3812
jsegdep = malloc(sizeof(*jsegdep), M_JSEGDEP, M_SOFTDEP_FLAGS);
3813
workitem_alloc(&jsegdep->jd_list, D_JSEGDEP, wk->wk_mp);
3814
jsegdep->jd_seg = NULL;
3815
3816
return (jsegdep);
3817
}
3818
3819
static struct jmvref *
3820
newjmvref(dp, ino, oldoff, newoff)
3821
struct inode *dp;
3822
ino_t ino;
3823
off_t oldoff;
3824
off_t newoff;
3825
{
3826
struct jmvref *jmvref;
3827
3828
jmvref = malloc(sizeof(*jmvref), M_JMVREF, M_SOFTDEP_FLAGS);
3829
workitem_alloc(&jmvref->jm_list, D_JMVREF, UFSTOVFS(dp->i_ump));
3830
jmvref->jm_list.wk_state = ATTACHED | DEPCOMPLETE;
3831
jmvref->jm_parent = dp->i_number;
3832
jmvref->jm_ino = ino;
3833
jmvref->jm_oldoff = oldoff;
3834
jmvref->jm_newoff = newoff;
3835
3836
return (jmvref);
3837
}
3838
3839
/*
3840
* Allocate a new jremref that tracks the removal of ip from dp with the
3841
* directory entry offset of diroff. Mark the entry as ATTACHED and
3842
* DEPCOMPLETE as we have all the information required for the journal write
3843
* and the directory has already been removed from the buffer. The caller
3844
* is responsible for linking the jremref into the pagedep and adding it
3845
* to the journal to write. The MKDIR_PARENT flag is set if we're doing
3846
* a DOTDOT addition so handle_workitem_remove() can properly assign
3847
* the jsegdep when we're done.
3848
*/
3849
static struct jremref *
3850
newjremref(struct dirrem *dirrem, struct inode *dp, struct inode *ip,
3851
off_t diroff, nlink_t nlink)
3852
{
3853
struct jremref *jremref;
3854
3855
jremref = malloc(sizeof(*jremref), M_JREMREF, M_SOFTDEP_FLAGS);
3856
workitem_alloc(&jremref->jr_list, D_JREMREF, UFSTOVFS(dp->i_ump));
3857
jremref->jr_state = ATTACHED;
3858
newinoref(&jremref->jr_ref, ip->i_number, dp->i_number, diroff,
3859
nlink, ip->i_mode);
3860
jremref->jr_dirrem = dirrem;
3861
3862
return (jremref);
3863
}
3864
3865
static inline void
3866
newinoref(struct inoref *inoref, ino_t ino, ino_t parent, off_t diroff,
3867
nlink_t nlink, uint16_t mode)
3868
{
3869
3870
inoref->if_jsegdep = newjsegdep(&inoref->if_list);
3871
inoref->if_diroff = diroff;
3872
inoref->if_ino = ino;
3873
inoref->if_parent = parent;
3874
inoref->if_nlink = nlink;
3875
inoref->if_mode = mode;
3876
}
3877
3878
/*
3879
* Allocate a new jaddref to track the addition of ino to dp at diroff. The
3880
* directory offset may not be known until later. The caller is responsible
3881
* adding the entry to the journal when this information is available. nlink
3882
* should be the link count prior to the addition and mode is only required
3883
* to have the correct FMT.
3884
*/
3885
static struct jaddref *
3886
newjaddref(struct inode *dp, ino_t ino, off_t diroff, int16_t nlink,
3887
uint16_t mode)
3888
{
3889
struct jaddref *jaddref;
3890
3891
jaddref = malloc(sizeof(*jaddref), M_JADDREF, M_SOFTDEP_FLAGS);
3892
workitem_alloc(&jaddref->ja_list, D_JADDREF, UFSTOVFS(dp->i_ump));
3893
jaddref->ja_state = ATTACHED;
3894
jaddref->ja_mkdir = NULL;
3895
newinoref(&jaddref->ja_ref, ino, dp->i_number, diroff, nlink, mode);
3896
3897
return (jaddref);
3898
}
3899
3900
/*
3901
* Create a new free dependency for a freework. The caller is responsible
3902
* for adjusting the reference count when it has the lock held. The freedep
3903
* will track an outstanding bitmap write that will ultimately clear the
3904
* freework to continue.
3905
*/
3906
static struct freedep *
3907
newfreedep(struct freework *freework)
3908
{
3909
struct freedep *freedep;
3910
3911
freedep = malloc(sizeof(*freedep), M_FREEDEP, M_SOFTDEP_FLAGS);
3912
workitem_alloc(&freedep->fd_list, D_FREEDEP, freework->fw_list.wk_mp);
3913
freedep->fd_freework = freework;
3914
3915
return (freedep);
3916
}
3917
3918
/*
3919
* Free a freedep structure once the buffer it is linked to is written. If
3920
* this is the last reference to the freework schedule it for completion.
3921
*/
3922
static void
3923
free_freedep(freedep)
3924
struct freedep *freedep;
3925
{
3926
struct freework *freework;
3927
3928
freework = freedep->fd_freework;
3929
freework->fw_freeblks->fb_cgwait--;
3930
if (--freework->fw_ref == 0)
3931
freework_enqueue(freework);
3932
WORKITEM_FREE(freedep, D_FREEDEP);
3933
}
3934
3935
/*
3936
* Allocate a new freework structure that may be a level in an indirect
3937
* when parent is not NULL or a top level block when it is. The top level
3938
* freework structures are allocated without lk held and before the freeblks
3939
* is visible outside of softdep_setup_freeblocks().
3940
*/
3941
static struct freework *
3942
newfreework(ump, freeblks, parent, lbn, nb, frags, off, journal)
3943
struct ufsmount *ump;
3944
struct freeblks *freeblks;
3945
struct freework *parent;
3946
ufs_lbn_t lbn;
3947
ufs2_daddr_t nb;
3948
int frags;
3949
int off;
3950
int journal;
3951
{
3952
struct freework *freework;
3953
3954
freework = malloc(sizeof(*freework), M_FREEWORK, M_SOFTDEP_FLAGS);
3955
workitem_alloc(&freework->fw_list, D_FREEWORK, freeblks->fb_list.wk_mp);
3956
freework->fw_state = ATTACHED;
3957
freework->fw_jnewblk = NULL;
3958
freework->fw_freeblks = freeblks;
3959
freework->fw_parent = parent;
3960
freework->fw_lbn = lbn;
3961
freework->fw_blkno = nb;
3962
freework->fw_frags = frags;
3963
freework->fw_indir = NULL;
3964
freework->fw_ref = (MOUNTEDSUJ(UFSTOVFS(ump)) == 0 || lbn >= -NXADDR)
3965
? 0 : NINDIR(ump->um_fs) + 1;
3966
freework->fw_start = freework->fw_off = off;
3967
if (journal)
3968
newjfreeblk(freeblks, lbn, nb, frags);
3969
if (parent == NULL) {
3970
ACQUIRE_LOCK(&lk);
3971
WORKLIST_INSERT(&freeblks->fb_freeworkhd, &freework->fw_list);
3972
freeblks->fb_ref++;
3973
FREE_LOCK(&lk);
3974
}
3975
3976
return (freework);
3977
}
3978
3979
/*
3980
* Eliminate a jfreeblk for a block that does not need journaling.
3981
*/
3982
static void
3983
cancel_jfreeblk(freeblks, blkno)
3984
struct freeblks *freeblks;
3985
ufs2_daddr_t blkno;
3986
{
3987
struct jfreeblk *jfreeblk;
3988
struct jblkdep *jblkdep;
3989
3990
LIST_FOREACH(jblkdep, &freeblks->fb_jblkdephd, jb_deps) {
3991
if (jblkdep->jb_list.wk_type != D_JFREEBLK)
3992
continue;
3993
jfreeblk = WK_JFREEBLK(&jblkdep->jb_list);
3994
if (jfreeblk->jf_blkno == blkno)
3995
break;
3996
}
3997
if (jblkdep == NULL)
3998
return;
3999
CTR1(KTR_SUJ, "cancel_jfreeblk: blkno %jd", blkno);
4000
free_jsegdep(jblkdep->jb_jsegdep);
4001
LIST_REMOVE(jblkdep, jb_deps);
4002
WORKITEM_FREE(jfreeblk, D_JFREEBLK);
4003
}
4004
4005
/*
4006
* Allocate a new jfreeblk to journal top level block pointer when truncating
4007
* a file. The caller must add this to the worklist when lk is held.
4008
*/
4009
static struct jfreeblk *
4010
newjfreeblk(freeblks, lbn, blkno, frags)
4011
struct freeblks *freeblks;
4012
ufs_lbn_t lbn;
4013
ufs2_daddr_t blkno;
4014
int frags;
4015
{
4016
struct jfreeblk *jfreeblk;
4017
4018
jfreeblk = malloc(sizeof(*jfreeblk), M_JFREEBLK, M_SOFTDEP_FLAGS);
4019
workitem_alloc(&jfreeblk->jf_dep.jb_list, D_JFREEBLK,
4020
freeblks->fb_list.wk_mp);
4021
jfreeblk->jf_dep.jb_jsegdep = newjsegdep(&jfreeblk->jf_dep.jb_list);
4022
jfreeblk->jf_dep.jb_freeblks = freeblks;
4023
jfreeblk->jf_ino = freeblks->fb_inum;
4024
jfreeblk->jf_lbn = lbn;
4025
jfreeblk->jf_blkno = blkno;
4026
jfreeblk->jf_frags = frags;
4027
LIST_INSERT_HEAD(&freeblks->fb_jblkdephd, &jfreeblk->jf_dep, jb_deps);
4028
4029
return (jfreeblk);
4030
}
4031
4032
/*
4033
* Allocate a new jtrunc to track a partial truncation.
4034
*/
4035
static struct jtrunc *
4036
newjtrunc(freeblks, size, extsize)
4037
struct freeblks *freeblks;
4038
off_t size;
4039
int extsize;
4040
{
4041
struct jtrunc *jtrunc;
4042
4043
jtrunc = malloc(sizeof(*jtrunc), M_JTRUNC, M_SOFTDEP_FLAGS);
4044
workitem_alloc(&jtrunc->jt_dep.jb_list, D_JTRUNC,
4045
freeblks->fb_list.wk_mp);
4046
jtrunc->jt_dep.jb_jsegdep = newjsegdep(&jtrunc->jt_dep.jb_list);
4047
jtrunc->jt_dep.jb_freeblks = freeblks;
4048
jtrunc->jt_ino = freeblks->fb_inum;
4049
jtrunc->jt_size = size;
4050
jtrunc->jt_extsize = extsize;
4051
LIST_INSERT_HEAD(&freeblks->fb_jblkdephd, &jtrunc->jt_dep, jb_deps);
4052
4053
return (jtrunc);
4054
}
4055
4056
/*
4057
* If we're canceling a new bitmap we have to search for another ref
4058
* to move into the bmsafemap dep. This might be better expressed
4059
* with another structure.
4060
*/
4061
static void
4062
move_newblock_dep(jaddref, inodedep)
4063
struct jaddref *jaddref;
4064
struct inodedep *inodedep;
4065
{
4066
struct inoref *inoref;
4067
struct jaddref *jaddrefn;
4068
4069
jaddrefn = NULL;
4070
for (inoref = TAILQ_NEXT(&jaddref->ja_ref, if_deps); inoref;
4071
inoref = TAILQ_NEXT(inoref, if_deps)) {
4072
if ((jaddref->ja_state & NEWBLOCK) &&
4073
inoref->if_list.wk_type == D_JADDREF) {
4074
jaddrefn = (struct jaddref *)inoref;
4075
break;
4076
}
4077
}
4078
if (jaddrefn == NULL)
4079
return;
4080
jaddrefn->ja_state &= ~(ATTACHED | UNDONE);
4081
jaddrefn->ja_state |= jaddref->ja_state &
4082
(ATTACHED | UNDONE | NEWBLOCK);
4083
jaddref->ja_state &= ~(ATTACHED | UNDONE | NEWBLOCK);
4084
jaddref->ja_state |= ATTACHED;
4085
LIST_REMOVE(jaddref, ja_bmdeps);
4086
LIST_INSERT_HEAD(&inodedep->id_bmsafemap->sm_jaddrefhd, jaddrefn,
4087
ja_bmdeps);
4088
}
4089
4090
/*
4091
* Cancel a jaddref either before it has been written or while it is being
4092
* written. This happens when a link is removed before the add reaches
4093
* the disk. The jaddref dependency is kept linked into the bmsafemap
4094
* and inode to prevent the link count or bitmap from reaching the disk
4095
* until handle_workitem_remove() re-adjusts the counts and bitmaps as
4096
* required.
4097
*
4098
* Returns 1 if the canceled addref requires journaling of the remove and
4099
* 0 otherwise.
4100
*/
4101
static int
4102
cancel_jaddref(jaddref, inodedep, wkhd)
4103
struct jaddref *jaddref;
4104
struct inodedep *inodedep;
4105
struct workhead *wkhd;
4106
{
4107
struct inoref *inoref;
4108
struct jsegdep *jsegdep;
4109
int needsj;
4110
4111
KASSERT((jaddref->ja_state & COMPLETE) == 0,
4112
("cancel_jaddref: Canceling complete jaddref"));
4113
if (jaddref->ja_state & (INPROGRESS | COMPLETE))
4114
needsj = 1;
4115
else
4116
needsj = 0;
4117
if (inodedep == NULL)
4118
if (inodedep_lookup(jaddref->ja_list.wk_mp, jaddref->ja_ino,
4119
0, &inodedep) == 0)
4120
panic("cancel_jaddref: Lost inodedep");
4121
/*
4122
* We must adjust the nlink of any reference operation that follows
4123
* us so that it is consistent with the in-memory reference. This
4124
* ensures that inode nlink rollbacks always have the correct link.
4125
*/
4126
if (needsj == 0) {
4127
for (inoref = TAILQ_NEXT(&jaddref->ja_ref, if_deps); inoref;
4128
inoref = TAILQ_NEXT(inoref, if_deps)) {
4129
if (inoref->if_state & GOINGAWAY)
4130
break;
4131
inoref->if_nlink--;
4132
}
4133
}
4134
jsegdep = inoref_jseg(&jaddref->ja_ref);
4135
if (jaddref->ja_state & NEWBLOCK)
4136
move_newblock_dep(jaddref, inodedep);
4137
wake_worklist(&jaddref->ja_list);
4138
jaddref->ja_mkdir = NULL;
4139
if (jaddref->ja_state & INPROGRESS) {
4140
jaddref->ja_state &= ~INPROGRESS;
4141
WORKLIST_REMOVE(&jaddref->ja_list);
4142
jwork_insert(wkhd, jsegdep);
4143
} else {
4144
free_jsegdep(jsegdep);
4145
if (jaddref->ja_state & DEPCOMPLETE)
4146
remove_from_journal(&jaddref->ja_list);
4147
}
4148
jaddref->ja_state |= (GOINGAWAY | DEPCOMPLETE);
4149
/*
4150
* Leave NEWBLOCK jaddrefs on the inodedep so handle_workitem_remove
4151
* can arrange for them to be freed with the bitmap. Otherwise we
4152
* no longer need this addref attached to the inoreflst and it
4153
* will incorrectly adjust nlink if we leave it.
4154
*/
4155
if ((jaddref->ja_state & NEWBLOCK) == 0) {
4156
TAILQ_REMOVE(&inodedep->id_inoreflst, &jaddref->ja_ref,
4157
if_deps);
4158
jaddref->ja_state |= COMPLETE;
4159
free_jaddref(jaddref);
4160
return (needsj);
4161
}
4162
/*
4163
* Leave the head of the list for jsegdeps for fast merging.
4164
*/
4165
if (LIST_FIRST(wkhd) != NULL) {
4166
jaddref->ja_state |= ONWORKLIST;
4167
LIST_INSERT_AFTER(LIST_FIRST(wkhd), &jaddref->ja_list, wk_list);
4168
} else
4169
WORKLIST_INSERT(wkhd, &jaddref->ja_list);
4170
4171
return (needsj);
4172
}
4173
4174
/*
4175
* Attempt to free a jaddref structure when some work completes. This
4176
* should only succeed once the entry is written and all dependencies have
4177
* been notified.
4178
*/
4179
static void
4180
free_jaddref(jaddref)
4181
struct jaddref *jaddref;
4182
{
4183
4184
if ((jaddref->ja_state & ALLCOMPLETE) != ALLCOMPLETE)
4185
return;
4186
if (jaddref->ja_ref.if_jsegdep)
4187
panic("free_jaddref: segdep attached to jaddref %p(0x%X)\n",
4188
jaddref, jaddref->ja_state);
4189
if (jaddref->ja_state & NEWBLOCK)
4190
LIST_REMOVE(jaddref, ja_bmdeps);
4191
if (jaddref->ja_state & (INPROGRESS | ONWORKLIST))
4192
panic("free_jaddref: Bad state %p(0x%X)",
4193
jaddref, jaddref->ja_state);
4194
if (jaddref->ja_mkdir != NULL)
4195
panic("free_jaddref: Work pending, 0x%X\n", jaddref->ja_state);
4196
WORKITEM_FREE(jaddref, D_JADDREF);
4197
}
4198
4199
/*
4200
* Free a jremref structure once it has been written or discarded.
4201
*/
4202
static void
4203
free_jremref(jremref)
4204
struct jremref *jremref;
4205
{
4206
4207
if (jremref->jr_ref.if_jsegdep)
4208
free_jsegdep(jremref->jr_ref.if_jsegdep);
4209
if (jremref->jr_state & INPROGRESS)
4210
panic("free_jremref: IO still pending");
4211
WORKITEM_FREE(jremref, D_JREMREF);
4212
}
4213
4214
/*
4215
* Free a jnewblk structure.
4216
*/
4217
static void
4218
free_jnewblk(jnewblk)
4219
struct jnewblk *jnewblk;
4220
{
4221
4222
if ((jnewblk->jn_state & ALLCOMPLETE) != ALLCOMPLETE)
4223
return;
4224
LIST_REMOVE(jnewblk, jn_deps);
4225
if (jnewblk->jn_dep != NULL)
4226
panic("free_jnewblk: Dependency still attached.");
4227
WORKITEM_FREE(jnewblk, D_JNEWBLK);
4228
}
4229
4230
/*
4231
* Cancel a jnewblk which has been been made redundant by frag extension.
4232
*/
4233
static void
4234
cancel_jnewblk(jnewblk, wkhd)
4235
struct jnewblk *jnewblk;
4236
struct workhead *wkhd;
4237
{
4238
struct jsegdep *jsegdep;
4239
4240
CTR1(KTR_SUJ, "cancel_jnewblk: blkno %jd", jnewblk->jn_blkno);
4241
jsegdep = jnewblk->jn_jsegdep;
4242
if (jnewblk->jn_jsegdep == NULL || jnewblk->jn_dep == NULL)
4243
panic("cancel_jnewblk: Invalid state");
4244
jnewblk->jn_jsegdep = NULL;
4245
jnewblk->jn_dep = NULL;
4246
jnewblk->jn_state |= GOINGAWAY;
4247
if (jnewblk->jn_state & INPROGRESS) {
4248
jnewblk->jn_state &= ~INPROGRESS;
4249
WORKLIST_REMOVE(&jnewblk->jn_list);
4250
jwork_insert(wkhd, jsegdep);
4251
} else {
4252
free_jsegdep(jsegdep);
4253
remove_from_journal(&jnewblk->jn_list);
4254
}
4255
wake_worklist(&jnewblk->jn_list);
4256
WORKLIST_INSERT(wkhd, &jnewblk->jn_list);
4257
}
4258
4259
static void
4260
free_jblkdep(jblkdep)
4261
struct jblkdep *jblkdep;
4262
{
4263
4264
if (jblkdep->jb_list.wk_type == D_JFREEBLK)
4265
WORKITEM_FREE(jblkdep, D_JFREEBLK);
4266
else if (jblkdep->jb_list.wk_type == D_JTRUNC)
4267
WORKITEM_FREE(jblkdep, D_JTRUNC);
4268
else
4269
panic("free_jblkdep: Unexpected type %s",
4270
TYPENAME(jblkdep->jb_list.wk_type));
4271
}
4272
4273
/*
4274
* Free a single jseg once it is no longer referenced in memory or on
4275
* disk. Reclaim journal blocks and dependencies waiting for the segment
4276
* to disappear.
4277
*/
4278
static void
4279
free_jseg(jseg, jblocks)
4280
struct jseg *jseg;
4281
struct jblocks *jblocks;
4282
{
4283
struct freework *freework;
4284
4285
/*
4286
* Free freework structures that were lingering to indicate freed
4287
* indirect blocks that forced journal write ordering on reallocate.
4288
*/
4289
while ((freework = LIST_FIRST(&jseg->js_indirs)) != NULL)
4290
indirblk_remove(freework);
4291
if (jblocks->jb_oldestseg == jseg)
4292
jblocks->jb_oldestseg = TAILQ_NEXT(jseg, js_next);
4293
TAILQ_REMOVE(&jblocks->jb_segs, jseg, js_next);
4294
jblocks_free(jblocks, jseg->js_list.wk_mp, jseg->js_size);
4295
KASSERT(LIST_EMPTY(&jseg->js_entries),
4296
("free_jseg: Freed jseg has valid entries."));
4297
WORKITEM_FREE(jseg, D_JSEG);
4298
}
4299
4300
/*
4301
* Free all jsegs that meet the criteria for being reclaimed and update
4302
* oldestseg.
4303
*/
4304
static void
4305
free_jsegs(jblocks)
4306
struct jblocks *jblocks;
4307
{
4308
struct jseg *jseg;
4309
4310
/*
4311
* Free only those jsegs which have none allocated before them to
4312
* preserve the journal space ordering.
4313
*/
4314
while ((jseg = TAILQ_FIRST(&jblocks->jb_segs)) != NULL) {
4315
/*
4316
* Only reclaim space when nothing depends on this journal
4317
* set and another set has written that it is no longer
4318
* valid.
4319
*/
4320
if (jseg->js_refs != 0) {
4321
jblocks->jb_oldestseg = jseg;
4322
return;
4323
}
4324
if ((jseg->js_state & ALLCOMPLETE) != ALLCOMPLETE)
4325
break;
4326
if (jseg->js_seq > jblocks->jb_oldestwrseq)
4327
break;
4328
/*
4329
* We can free jsegs that didn't write entries when
4330
* oldestwrseq == js_seq.
4331
*/
4332
if (jseg->js_seq == jblocks->jb_oldestwrseq &&
4333
jseg->js_cnt != 0)
4334
break;
4335
free_jseg(jseg, jblocks);
4336
}
4337
/*
4338
* If we exited the loop above we still must discover the
4339
* oldest valid segment.
4340
*/
4341
if (jseg)
4342
for (jseg = jblocks->jb_oldestseg; jseg != NULL;
4343
jseg = TAILQ_NEXT(jseg, js_next))
4344
if (jseg->js_refs != 0)
4345
break;
4346
jblocks->jb_oldestseg = jseg;
4347
/*
4348
* The journal has no valid records but some jsegs may still be
4349
* waiting on oldestwrseq to advance. We force a small record
4350
* out to permit these lingering records to be reclaimed.
4351
*/
4352
if (jblocks->jb_oldestseg == NULL && !TAILQ_EMPTY(&jblocks->jb_segs))
4353
jblocks->jb_needseg = 1;
4354
}
4355
4356
/*
4357
* Release one reference to a jseg and free it if the count reaches 0. This
4358
* should eventually reclaim journal space as well.
4359
*/
4360
static void
4361
rele_jseg(jseg)
4362
struct jseg *jseg;
4363
{
4364
4365
KASSERT(jseg->js_refs > 0,
4366
("free_jseg: Invalid refcnt %d", jseg->js_refs));
4367
if (--jseg->js_refs != 0)
4368
return;
4369
free_jsegs(jseg->js_jblocks);
4370
}
4371
4372
/*
4373
* Release a jsegdep and decrement the jseg count.
4374
*/
4375
static void
4376
free_jsegdep(jsegdep)
4377
struct jsegdep *jsegdep;
4378
{
4379
4380
if (jsegdep->jd_seg)
4381
rele_jseg(jsegdep->jd_seg);
4382
WORKITEM_FREE(jsegdep, D_JSEGDEP);
4383
}
4384
4385
/*
4386
* Wait for a journal item to make it to disk. Initiate journal processing
4387
* if required.
4388
*/
4389
static int
4390
jwait(wk, waitfor)
4391
struct worklist *wk;
4392
int waitfor;
4393
{
4394
4395
/*
4396
* Blocking journal waits cause slow synchronous behavior. Record
4397
* stats on the frequency of these blocking operations.
4398
*/
4399
if (waitfor == MNT_WAIT) {
4400
stat_journal_wait++;
4401
switch (wk->wk_type) {
4402
case D_JREMREF:
4403
case D_JMVREF:
4404
stat_jwait_filepage++;
4405
break;
4406
case D_JTRUNC:
4407
case D_JFREEBLK:
4408
stat_jwait_freeblks++;
4409
break;
4410
case D_JNEWBLK:
4411
stat_jwait_newblk++;
4412
break;
4413
case D_JADDREF:
4414
stat_jwait_inode++;
4415
break;
4416
default:
4417
break;
4418
}
4419
}
4420
/*
4421
* If IO has not started we process the journal. We can't mark the
4422
* worklist item as IOWAITING because we drop the lock while
4423
* processing the journal and the worklist entry may be freed after
4424
* this point. The caller may call back in and re-issue the request.
4425
*/
4426
if ((wk->wk_state & INPROGRESS) == 0) {
4427
softdep_process_journal(wk->wk_mp, wk, waitfor);
4428
if (waitfor != MNT_WAIT)
4429
return (EBUSY);
4430
return (0);
4431
}
4432
if (waitfor != MNT_WAIT)
4433
return (EBUSY);
4434
wait_worklist(wk, "jwait");
4435
return (0);
4436
}
4437
4438
/*
4439
* Lookup an inodedep based on an inode pointer and set the nlinkdelta as
4440
* appropriate. This is a convenience function to reduce duplicate code
4441
* for the setup and revert functions below.
4442
*/
4443
static struct inodedep *
4444
inodedep_lookup_ip(ip)
4445
struct inode *ip;
4446
{
4447
struct inodedep *inodedep;
4448
int dflags;
4449
4450
KASSERT(ip->i_nlink >= ip->i_effnlink,
4451
("inodedep_lookup_ip: bad delta"));
4452
dflags = DEPALLOC;
4453
if (IS_SNAPSHOT(ip))
4454
dflags |= NODELAY;
4455
(void) inodedep_lookup(UFSTOVFS(ip->i_ump), ip->i_number, dflags,
4456
&inodedep);
4457
inodedep->id_nlinkdelta = ip->i_nlink - ip->i_effnlink;
4458
KASSERT((inodedep->id_state & UNLINKED) == 0, ("inode unlinked"));
4459
4460
return (inodedep);
4461
}
4462
4463
/*
4464
* Called prior to creating a new inode and linking it to a directory. The
4465
* jaddref structure must already be allocated by softdep_setup_inomapdep
4466
* and it is discovered here so we can initialize the mode and update
4467
* nlinkdelta.
4468
*/
4469
void
4470
softdep_setup_create(dp, ip)
4471
struct inode *dp;
4472
struct inode *ip;
4473
{
4474
struct inodedep *inodedep;
4475
struct jaddref *jaddref;
4476
struct vnode *dvp;
4477
4478
KASSERT(ip->i_nlink == 1,
4479
("softdep_setup_create: Invalid link count."));
4480
dvp = ITOV(dp);
4481
ACQUIRE_LOCK(&lk);
4482
inodedep = inodedep_lookup_ip(ip);
4483
if (DOINGSUJ(dvp)) {
4484
jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst,
4485
inoreflst);
4486
KASSERT(jaddref != NULL && jaddref->ja_parent == dp->i_number,
4487
("softdep_setup_create: No addref structure present."));
4488
}
4489
softdep_prelink(dvp, NULL);
4490
FREE_LOCK(&lk);
4491
}
4492
4493
/*
4494
* Create a jaddref structure to track the addition of a DOTDOT link when
4495
* we are reparenting an inode as part of a rename. This jaddref will be
4496
* found by softdep_setup_directory_change. Adjusts nlinkdelta for
4497
* non-journaling softdep.
4498
*/
4499
void
4500
softdep_setup_dotdot_link(dp, ip)
4501
struct inode *dp;
4502
struct inode *ip;
4503
{
4504
struct inodedep *inodedep;
4505
struct jaddref *jaddref;
4506
struct vnode *dvp;
4507
struct vnode *vp;
4508
4509
dvp = ITOV(dp);
4510
vp = ITOV(ip);
4511
jaddref = NULL;
4512
/*
4513
* We don't set MKDIR_PARENT as this is not tied to a mkdir and
4514
* is used as a normal link would be.
4515
*/
4516
if (DOINGSUJ(dvp))
4517
jaddref = newjaddref(ip, dp->i_number, DOTDOT_OFFSET,
4518
dp->i_effnlink - 1, dp->i_mode);
4519
ACQUIRE_LOCK(&lk);
4520
inodedep = inodedep_lookup_ip(dp);
4521
if (jaddref)
4522
TAILQ_INSERT_TAIL(&inodedep->id_inoreflst, &jaddref->ja_ref,
4523
if_deps);
4524
softdep_prelink(dvp, ITOV(ip));
4525
FREE_LOCK(&lk);
4526
}
4527
4528
/*
4529
* Create a jaddref structure to track a new link to an inode. The directory
4530
* offset is not known until softdep_setup_directory_add or
4531
* softdep_setup_directory_change. Adjusts nlinkdelta for non-journaling
4532
* softdep.
4533
*/
4534
void
4535
softdep_setup_link(dp, ip)
4536
struct inode *dp;
4537
struct inode *ip;
4538
{
4539
struct inodedep *inodedep;
4540
struct jaddref *jaddref;
4541
struct vnode *dvp;
4542
4543
dvp = ITOV(dp);
4544
jaddref = NULL;
4545
if (DOINGSUJ(dvp))
4546
jaddref = newjaddref(dp, ip->i_number, 0, ip->i_effnlink - 1,
4547
ip->i_mode);
4548
ACQUIRE_LOCK(&lk);
4549
inodedep = inodedep_lookup_ip(ip);
4550
if (jaddref)
4551
TAILQ_INSERT_TAIL(&inodedep->id_inoreflst, &jaddref->ja_ref,
4552
if_deps);
4553
softdep_prelink(dvp, ITOV(ip));
4554
FREE_LOCK(&lk);
4555
}
4556
4557
/*
4558
* Called to create the jaddref structures to track . and .. references as
4559
* well as lookup and further initialize the incomplete jaddref created
4560
* by softdep_setup_inomapdep when the inode was allocated. Adjusts
4561
* nlinkdelta for non-journaling softdep.
4562
*/
4563
void
4564
softdep_setup_mkdir(dp, ip)
4565
struct inode *dp;
4566
struct inode *ip;
4567
{
4568
struct inodedep *inodedep;
4569
struct jaddref *dotdotaddref;
4570
struct jaddref *dotaddref;
4571
struct jaddref *jaddref;
4572
struct vnode *dvp;
4573
4574
dvp = ITOV(dp);
4575
dotaddref = dotdotaddref = NULL;
4576
if (DOINGSUJ(dvp)) {
4577
dotaddref = newjaddref(ip, ip->i_number, DOT_OFFSET, 1,
4578
ip->i_mode);
4579
dotaddref->ja_state |= MKDIR_BODY;
4580
dotdotaddref = newjaddref(ip, dp->i_number, DOTDOT_OFFSET,
4581
dp->i_effnlink - 1, dp->i_mode);
4582
dotdotaddref->ja_state |= MKDIR_PARENT;
4583
}
4584
ACQUIRE_LOCK(&lk);
4585
inodedep = inodedep_lookup_ip(ip);
4586
if (DOINGSUJ(dvp)) {
4587
jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst,
4588
inoreflst);
4589
KASSERT(jaddref != NULL,
4590
("softdep_setup_mkdir: No addref structure present."));
4591
KASSERT(jaddref->ja_parent == dp->i_number,
4592
("softdep_setup_mkdir: bad parent %d",
4593
jaddref->ja_parent));
4594
TAILQ_INSERT_BEFORE(&jaddref->ja_ref, &dotaddref->ja_ref,
4595
if_deps);
4596
}
4597
inodedep = inodedep_lookup_ip(dp);
4598
if (DOINGSUJ(dvp))
4599
TAILQ_INSERT_TAIL(&inodedep->id_inoreflst,
4600
&dotdotaddref->ja_ref, if_deps);
4601
softdep_prelink(ITOV(dp), NULL);
4602
FREE_LOCK(&lk);
4603
}
4604
4605
/*
4606
* Called to track nlinkdelta of the inode and parent directories prior to
4607
* unlinking a directory.
4608
*/
4609
void
4610
softdep_setup_rmdir(dp, ip)
4611
struct inode *dp;
4612
struct inode *ip;
4613
{
4614
struct vnode *dvp;
4615
4616
dvp = ITOV(dp);
4617
ACQUIRE_LOCK(&lk);
4618
(void) inodedep_lookup_ip(ip);
4619
(void) inodedep_lookup_ip(dp);
4620
softdep_prelink(dvp, ITOV(ip));
4621
FREE_LOCK(&lk);
4622
}
4623
4624
/*
4625
* Called to track nlinkdelta of the inode and parent directories prior to
4626
* unlink.
4627
*/
4628
void
4629
softdep_setup_unlink(dp, ip)
4630
struct inode *dp;
4631
struct inode *ip;
4632
{
4633
struct vnode *dvp;
4634
4635
dvp = ITOV(dp);
4636
ACQUIRE_LOCK(&lk);
4637
(void) inodedep_lookup_ip(ip);
4638
(void) inodedep_lookup_ip(dp);
4639
softdep_prelink(dvp, ITOV(ip));
4640
FREE_LOCK(&lk);
4641
}
4642
4643
/*
4644
* Called to release the journal structures created by a failed non-directory
4645
* creation. Adjusts nlinkdelta for non-journaling softdep.
4646
*/
4647
void
4648
softdep_revert_create(dp, ip)
4649
struct inode *dp;
4650
struct inode *ip;
4651
{
4652
struct inodedep *inodedep;
4653
struct jaddref *jaddref;
4654
struct vnode *dvp;
4655
4656
dvp = ITOV(dp);
4657
ACQUIRE_LOCK(&lk);
4658
inodedep = inodedep_lookup_ip(ip);
4659
if (DOINGSUJ(dvp)) {
4660
jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst,
4661
inoreflst);
4662
KASSERT(jaddref->ja_parent == dp->i_number,
4663
("softdep_revert_create: addref parent mismatch"));
4664
cancel_jaddref(jaddref, inodedep, &inodedep->id_inowait);
4665
}
4666
FREE_LOCK(&lk);
4667
}
4668
4669
/*
4670
* Called to release the journal structures created by a failed dotdot link
4671
* creation. Adjusts nlinkdelta for non-journaling softdep.
4672
*/
4673
void
4674
softdep_revert_dotdot_link(dp, ip)
4675
struct inode *dp;
4676
struct inode *ip;
4677
{
4678
struct inodedep *inodedep;
4679
struct jaddref *jaddref;
4680
struct vnode *dvp;
4681
4682
dvp = ITOV(dp);
4683
ACQUIRE_LOCK(&lk);
4684
inodedep = inodedep_lookup_ip(dp);
4685
if (DOINGSUJ(dvp)) {
4686
jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst,
4687
inoreflst);
4688
KASSERT(jaddref->ja_parent == ip->i_number,
4689
("softdep_revert_dotdot_link: addref parent mismatch"));
4690
cancel_jaddref(jaddref, inodedep, &inodedep->id_inowait);
4691
}
4692
FREE_LOCK(&lk);
4693
}
4694
4695
/*
4696
* Called to release the journal structures created by a failed link
4697
* addition. Adjusts nlinkdelta for non-journaling softdep.
4698
*/
4699
void
4700
softdep_revert_link(dp, ip)
4701
struct inode *dp;
4702
struct inode *ip;
4703
{
4704
struct inodedep *inodedep;
4705
struct jaddref *jaddref;
4706
struct vnode *dvp;
4707
4708
dvp = ITOV(dp);
4709
ACQUIRE_LOCK(&lk);
4710
inodedep = inodedep_lookup_ip(ip);
4711
if (DOINGSUJ(dvp)) {
4712
jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst,
4713
inoreflst);
4714
KASSERT(jaddref->ja_parent == dp->i_number,
4715
("softdep_revert_link: addref parent mismatch"));
4716
cancel_jaddref(jaddref, inodedep, &inodedep->id_inowait);
4717
}
4718
FREE_LOCK(&lk);
4719
}
4720
4721
/*
4722
* Called to release the journal structures created by a failed mkdir
4723
* attempt. Adjusts nlinkdelta for non-journaling softdep.
4724
*/
4725
void
4726
softdep_revert_mkdir(dp, ip)
4727
struct inode *dp;
4728
struct inode *ip;
4729
{
4730
struct inodedep *inodedep;
4731
struct jaddref *jaddref;
4732
struct jaddref *dotaddref;
4733
struct vnode *dvp;
4734
4735
dvp = ITOV(dp);
4736
4737
ACQUIRE_LOCK(&lk);
4738
inodedep = inodedep_lookup_ip(dp);
4739
if (DOINGSUJ(dvp)) {
4740
jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst,
4741
inoreflst);
4742
KASSERT(jaddref->ja_parent == ip->i_number,
4743
("softdep_revert_mkdir: dotdot addref parent mismatch"));
4744
cancel_jaddref(jaddref, inodedep, &inodedep->id_inowait);
4745
}
4746
inodedep = inodedep_lookup_ip(ip);
4747
if (DOINGSUJ(dvp)) {
4748
jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst,
4749
inoreflst);
4750
KASSERT(jaddref->ja_parent == dp->i_number,
4751
("softdep_revert_mkdir: addref parent mismatch"));
4752
dotaddref = (struct jaddref *)TAILQ_PREV(&jaddref->ja_ref,
4753
inoreflst, if_deps);
4754
cancel_jaddref(jaddref, inodedep, &inodedep->id_inowait);
4755
KASSERT(dotaddref->ja_parent == ip->i_number,
4756
("softdep_revert_mkdir: dot addref parent mismatch"));
4757
cancel_jaddref(dotaddref, inodedep, &inodedep->id_inowait);
4758
}
4759
FREE_LOCK(&lk);
4760
}
4761
4762
/*
4763
* Called to correct nlinkdelta after a failed rmdir.
4764
*/
4765
void
4766
softdep_revert_rmdir(dp, ip)
4767
struct inode *dp;
4768
struct inode *ip;
4769
{
4770
4771
ACQUIRE_LOCK(&lk);
4772
(void) inodedep_lookup_ip(ip);
4773
(void) inodedep_lookup_ip(dp);
4774
FREE_LOCK(&lk);
4775
}
4776
1496
4777
/*
1497
4778
* Protecting the freemaps (or bitmaps).
1498
4779
*
1529
4810
* Called just after updating the cylinder group block to allocate an inode.
1530
4811
*/
1531
4812
void
1532
softdep_setup_inomapdep(bp, ip, newinum)
4813
softdep_setup_inomapdep(bp, ip, newinum, mode)
1533
4814
struct buf *bp; /* buffer for cylgroup block with inode map */
1534
4815
struct inode *ip; /* inode related to allocation */
1535
4816
ino_t newinum; /* new inode number being allocated */
4817
int mode;
1536
4818
{
1537
4819
struct inodedep *inodedep;
1538
4820
struct bmsafemap *bmsafemap;
4821
struct jaddref *jaddref;
4822
struct mount *mp;
4823
struct fs *fs;
4824
4825
mp = UFSTOVFS(ip->i_ump);
4826
fs = ip->i_ump->um_fs;
4827
jaddref = NULL;
4828
4829
/*
4830
* Allocate the journal reference add structure so that the bitmap
4831
* can be dependent on it.
4832
*/
4833
if (MOUNTEDSUJ(mp)) {
4834
jaddref = newjaddref(ip, newinum, 0, 0, mode);
4835
jaddref->ja_state |= NEWBLOCK;
4836
}
1539
4837
1540
4838
/*
1541
4839
* Create a dependency for the newly allocated inode.
1542
4840
* Panic if it already exists as something is seriously wrong.
1543
4841
* Otherwise add it to the dependency list for the buffer holding
1544
4842
* the cylinder group map from which it was allocated.
4843
*
4844
* We have to preallocate a bmsafemap entry in case it is needed
4845
* in bmsafemap_lookup since once we allocate the inodedep, we
4846
* have to finish initializing it before we can FREE_LOCK().
4847
* By preallocating, we avoid FREE_LOCK() while doing a malloc
4848
* in bmsafemap_lookup. We cannot call bmsafemap_lookup before
4849
* creating the inodedep as it can be freed during the time
4850
* that we FREE_LOCK() while allocating the inodedep. We must
4851
* call workitem_alloc() before entering the locked section as
4852
* it also acquires the lock and we must avoid trying doing so
4853
* recursively.
1545
4854
*/
4855
bmsafemap = malloc(sizeof(struct bmsafemap),
4856
M_BMSAFEMAP, M_SOFTDEP_FLAGS);
4857
workitem_alloc(&bmsafemap->sm_list, D_BMSAFEMAP, mp);
1546
4858
ACQUIRE_LOCK(&lk);
1547
if ((inodedep_lookup(UFSTOVFS(ip->i_ump), newinum, DEPALLOC|NODELAY,
1548
&inodedep)))
1549
panic("softdep_setup_inomapdep: dependency for new inode "
1550
"already exists");
1551
inodedep->id_buf = bp;
4859
if ((inodedep_lookup(mp, newinum, DEPALLOC | NODELAY, &inodedep)))
4860
panic("softdep_setup_inomapdep: dependency %p for new"
4861
"inode already exists", inodedep);
4862
bmsafemap = bmsafemap_lookup(mp, bp, ino_to_cg(fs, newinum), bmsafemap);
4863
if (jaddref) {
4864
LIST_INSERT_HEAD(&bmsafemap->sm_jaddrefhd, jaddref, ja_bmdeps);
4865
TAILQ_INSERT_TAIL(&inodedep->id_inoreflst, &jaddref->ja_ref,
4866
if_deps);
4867
} else {
4868
inodedep->id_state |= ONDEPLIST;
4869
LIST_INSERT_HEAD(&bmsafemap->sm_inodedephd, inodedep, id_deps);
4870
}
4871
inodedep->id_bmsafemap = bmsafemap;
1552
4872
inodedep->id_state &= ~DEPCOMPLETE;
1553
bmsafemap = bmsafemap_lookup(inodedep->id_list.wk_mp, bp);
1554
LIST_INSERT_HEAD(&bmsafemap->sm_inodedephd, inodedep, id_deps);
1555
4873
FREE_LOCK(&lk);
1556
4874
}
1557
4875
1560
4878
* allocate block or fragment.
1561
4879
*/
1562
4880
void
1563
softdep_setup_blkmapdep(bp, mp, newblkno)
4881
softdep_setup_blkmapdep(bp, mp, newblkno, frags, oldfrags)
1564
4882
struct buf *bp; /* buffer for cylgroup block with block map */
1565
4883
struct mount *mp; /* filesystem doing allocation */
1566
4884
ufs2_daddr_t newblkno; /* number of newly allocated block */
4885
int frags; /* Number of fragments. */
4886
int oldfrags; /* Previous number of fragments for extend. */
1567
4887
{
1568
4888
struct newblk *newblk;
1569
4889
struct bmsafemap *bmsafemap;
4890
struct jnewblk *jnewblk;
1570
4891
struct fs *fs;
1571
4892
1572
4893
fs = VFSTOUFS(mp)->um_fs;
4894
jnewblk = NULL;
1573
4895
/*
1574
4896
* Create a dependency for the newly allocated block.
1575
4897
* Add it to the dependency list for the buffer holding
1576
4898
* the cylinder group map from which it was allocated.
1577
4899
*/
4900
if (MOUNTEDSUJ(mp)) {
4901
jnewblk = malloc(sizeof(*jnewblk), M_JNEWBLK, M_SOFTDEP_FLAGS);
4902
workitem_alloc(&jnewblk->jn_list, D_JNEWBLK, mp);
4903
jnewblk->jn_jsegdep = newjsegdep(&jnewblk->jn_list);
4904
jnewblk->jn_state = ATTACHED;
4905
jnewblk->jn_blkno = newblkno;
4906
jnewblk->jn_frags = frags;
4907
jnewblk->jn_oldfrags = oldfrags;
4908
#ifdef SUJ_DEBUG
4909
{
4910
struct cg *cgp;
4911
uint8_t *blksfree;
4912
long bno;
4913
int i;
4914
4915
cgp = (struct cg *)bp->b_data;
4916
blksfree = cg_blksfree(cgp);
4917
bno = dtogd(fs, jnewblk->jn_blkno);
4918
for (i = jnewblk->jn_oldfrags; i < jnewblk->jn_frags;
4919
i++) {
4920
if (isset(blksfree, bno + i))
4921
panic("softdep_setup_blkmapdep: "
4922
"free fragment %d from %d-%d "
4923
"state 0x%X dep %p", i,
4924
jnewblk->jn_oldfrags,
4925
jnewblk->jn_frags,
4926
jnewblk->jn_state,
4927
jnewblk->jn_dep);
4928
}
4929
}
4930
#endif
4931
}
4932
4933
CTR3(KTR_SUJ,
4934
"softdep_setup_blkmapdep: blkno %jd frags %d oldfrags %d",
4935
newblkno, frags, oldfrags);
1578
4936
ACQUIRE_LOCK(&lk);
1579
if (newblk_lookup(fs, newblkno, DEPALLOC, &newblk) != 0)
4937
if (newblk_lookup(mp, newblkno, DEPALLOC, &newblk) != 0)
1580
4938
panic("softdep_setup_blkmapdep: found block");
1581
newblk->nb_bmsafemap = bmsafemap = bmsafemap_lookup(mp, bp);
1582
LIST_INSERT_HEAD(&bmsafemap->sm_newblkhd, newblk, nb_deps);
4939
newblk->nb_bmsafemap = bmsafemap = bmsafemap_lookup(mp, bp,
4940
dtog(fs, newblkno), NULL);
4941
if (jnewblk) {
4942
jnewblk->jn_dep = (struct worklist *)newblk;
4943
LIST_INSERT_HEAD(&bmsafemap->sm_jnewblkhd, jnewblk, jn_deps);
4944
} else {
4945
newblk->nb_state |= ONDEPLIST;
4946
LIST_INSERT_HEAD(&bmsafemap->sm_newblkhd, newblk, nb_deps);
4947
}
4948
newblk->nb_bmsafemap = bmsafemap;
4949
newblk->nb_jnewblk = jnewblk;
1583
4950
FREE_LOCK(&lk);
1584
4951
}
1585
4952
4953
#define BMSAFEMAP_HASH(fs, cg) \
4954
(&bmsafemap_hashtbl[((((register_t)(fs)) >> 13) + (cg)) & bmsafemap_hash])
4955
4956
static int
4957
bmsafemap_find(bmsafemaphd, mp, cg, bmsafemapp)
4958
struct bmsafemap_hashhead *bmsafemaphd;
4959
struct mount *mp;
4960
int cg;
4961
struct bmsafemap **bmsafemapp;
4962
{
4963
struct bmsafemap *bmsafemap;
4964
4965
LIST_FOREACH(bmsafemap, bmsafemaphd, sm_hash)
4966
if (bmsafemap->sm_list.wk_mp == mp && bmsafemap->sm_cg == cg)
4967
break;
4968
if (bmsafemap) {
4969
*bmsafemapp = bmsafemap;
4970
return (1);
4971
}
4972
*bmsafemapp = NULL;
4973
4974
return (0);
4975
}
4976
1586
4977
/*
1587
4978
* Find the bmsafemap associated with a cylinder group buffer.
1588
4979
* If none exists, create one. The buffer must be locked when
1589
4980
* this routine is called and this routine must be called with
1590
* splbio interrupts blocked.
4981
* the softdep lock held. To avoid giving up the lock while
4982
* allocating a new bmsafemap, a preallocated bmsafemap may be
4983
* provided. If it is provided but not needed, it is freed.
1591
4984
*/
1592
4985
static struct bmsafemap *
1593
bmsafemap_lookup(mp, bp)
4986
bmsafemap_lookup(mp, bp, cg, newbmsafemap)
1594
4987
struct mount *mp;
1595
4988
struct buf *bp;
4989
int cg;
4990
struct bmsafemap *newbmsafemap;
1596
4991
{
1597
struct bmsafemap *bmsafemap;
4992
struct bmsafemap_hashhead *bmsafemaphd;
4993
struct bmsafemap *bmsafemap, *collision;
1598
4994
struct worklist *wk;
4995
struct fs *fs;
1599
4996
1600
4997
mtx_assert(&lk, MA_OWNED);
1601
LIST_FOREACH(wk, &bp->b_dep, wk_list)
1602
if (wk->wk_type == D_BMSAFEMAP)
4998
KASSERT(bp != NULL, ("bmsafemap_lookup: missing buffer"));
4999
LIST_FOREACH(wk, &bp->b_dep, wk_list) {
5000
if (wk->wk_type == D_BMSAFEMAP) {
5001
if (newbmsafemap)
5002
WORKITEM_FREE(newbmsafemap, D_BMSAFEMAP);
1603
5003
return (WK_BMSAFEMAP(wk));
1604
FREE_LOCK(&lk);
1605
bmsafemap = malloc(sizeof(struct bmsafemap),
1606
M_BMSAFEMAP, M_SOFTDEP_FLAGS);
1607
workitem_alloc(&bmsafemap->sm_list, D_BMSAFEMAP, mp);
5004
}
5005
}
5006
fs = VFSTOUFS(mp)->um_fs;
5007
bmsafemaphd = BMSAFEMAP_HASH(fs, cg);
5008
if (bmsafemap_find(bmsafemaphd, mp, cg, &bmsafemap) == 1) {
5009
if (newbmsafemap)
5010
WORKITEM_FREE(newbmsafemap, D_BMSAFEMAP);
5011
return (bmsafemap);
5012
}
5013
if (newbmsafemap) {
5014
bmsafemap = newbmsafemap;
5015
} else {
5016
FREE_LOCK(&lk);
5017
bmsafemap = malloc(sizeof(struct bmsafemap),
5018
M_BMSAFEMAP, M_SOFTDEP_FLAGS);
5019
workitem_alloc(&bmsafemap->sm_list, D_BMSAFEMAP, mp);
5020
ACQUIRE_LOCK(&lk);
5021
}
1608
5022
bmsafemap->sm_buf = bp;
1609
LIST_INIT(&bmsafemap->sm_allocdirecthd);
1610
LIST_INIT(&bmsafemap->sm_allocindirhd);
1611
5023
LIST_INIT(&bmsafemap->sm_inodedephd);
5024
LIST_INIT(&bmsafemap->sm_inodedepwr);
1612
5025
LIST_INIT(&bmsafemap->sm_newblkhd);
1613
ACQUIRE_LOCK(&lk);
5026
LIST_INIT(&bmsafemap->sm_newblkwr);
5027
LIST_INIT(&bmsafemap->sm_jaddrefhd);
5028
LIST_INIT(&bmsafemap->sm_jnewblkhd);
5029
LIST_INIT(&bmsafemap->sm_freehd);
5030
LIST_INIT(&bmsafemap->sm_freewr);
5031
if (bmsafemap_find(bmsafemaphd, mp, cg, &collision) == 1) {
5032
WORKITEM_FREE(bmsafemap, D_BMSAFEMAP);
5033
return (collision);
5034
}
5035
bmsafemap->sm_cg = cg;
5036
LIST_INSERT_HEAD(bmsafemaphd, bmsafemap, sm_hash);
5037
LIST_INSERT_HEAD(&VFSTOUFS(mp)->softdep_dirtycg, bmsafemap, sm_next);
1614
5038
WORKLIST_INSERT(&bp->b_dep, &bmsafemap->sm_list);
1615
5039
return (bmsafemap);
1616
5040
}
1645
5069
* unreferenced fragments.
1646
5070
*/
1647
5071
void
1648
softdep_setup_allocdirect(ip, lbn, newblkno, oldblkno, newsize, oldsize, bp)
5072
softdep_setup_allocdirect(ip, off, newblkno, oldblkno, newsize, oldsize, bp)
1649
5073
struct inode *ip; /* inode to which block is being added */
1650
ufs_lbn_t lbn; /* block pointer within inode */
5074
ufs_lbn_t off; /* block pointer within inode */
1651
5075
ufs2_daddr_t newblkno; /* disk block number being added */
1652
5076
ufs2_daddr_t oldblkno; /* previous block number, 0 unless frag */
1653
5077
long newsize; /* size of new block */
1656
5080
{
1657
5081
struct allocdirect *adp, *oldadp;
1658
5082
struct allocdirectlst *adphead;
1659
struct bmsafemap *bmsafemap;
5083
struct freefrag *freefrag;
1660
5084
struct inodedep *inodedep;
1661
5085
struct pagedep *pagedep;
5086
struct jnewblk *jnewblk;
1662
5087
struct newblk *newblk;
1663
5088
struct mount *mp;
5089
ufs_lbn_t lbn;
1664
5090
5091
lbn = bp->b_lblkno;
1665
5092
mp = UFSTOVFS(ip->i_ump);
1666
adp = malloc(sizeof(struct allocdirect),
1667
M_ALLOCDIRECT, M_SOFTDEP_FLAGS|M_ZERO);
1668
workitem_alloc(&adp->ad_list, D_ALLOCDIRECT, mp);
1669
adp->ad_lbn = lbn;
1670
adp->ad_newblkno = newblkno;
1671
adp->ad_oldblkno = oldblkno;
1672
adp->ad_newsize = newsize;
1673
adp->ad_oldsize = oldsize;
1674
adp->ad_state = ATTACHED;
1675
LIST_INIT(&adp->ad_newdirblk);
1676
if (newblkno == oldblkno)
1677
adp->ad_freefrag = NULL;
5093
if (oldblkno && oldblkno != newblkno)
5094
freefrag = newfreefrag(ip, oldblkno, oldsize, lbn);
1678
5095
else
1679
adp->ad_freefrag = newfreefrag(ip, oldblkno, oldsize);
5096
freefrag = NULL;
1680
5097
5098
CTR6(KTR_SUJ,
5099
"softdep_setup_allocdirect: ino %d blkno %jd oldblkno %jd "
5100
"off %jd newsize %ld oldsize %d",
5101
ip->i_number, newblkno, oldblkno, off, newsize, oldsize);
1681
5102
ACQUIRE_LOCK(&lk);
1682
if (lbn >= NDADDR) {
5103
if (off >= NDADDR) {
5104
if (lbn > 0)
5105
panic("softdep_setup_allocdirect: bad lbn %jd, off %jd",
5106
lbn, off);
1683
5107
/* allocating an indirect block */
1684
5108
if (oldblkno != 0)
1685
5109
panic("softdep_setup_allocdirect: non-zero indir");
1686
5110
} else {
5111
if (off != lbn)
5112
panic("softdep_setup_allocdirect: lbn %jd != off %jd",
5113
lbn, off);
1687
5114
/*
1688
5115
* Allocating a direct block.
1689
5116
*
1691
5118
* allocate an associated pagedep to track additions and
1692
5119
* deletions.
1693
5120
*/
1694
if ((ip->i_mode & IFMT) == IFDIR &&
1695
pagedep_lookup(ip, lbn, DEPALLOC, &pagedep) == 0)
1696
WORKLIST_INSERT(&bp->b_dep, &pagedep->pd_list);
5121
if ((ip->i_mode & IFMT) == IFDIR)
5122
pagedep_lookup(mp, bp, ip->i_number, off, DEPALLOC,
5123
&pagedep);
1697
5124
}
1698
if (newblk_lookup(ip->i_fs, newblkno, 0, &newblk) == 0)
5125
if (newblk_lookup(mp, newblkno, 0, &newblk) == 0)
1699
5126
panic("softdep_setup_allocdirect: lost block");
1700
if (newblk->nb_state == DEPCOMPLETE) {
1701
adp->ad_state |= DEPCOMPLETE;
1702
adp->ad_buf = NULL;
1703
} else {
1704
bmsafemap = newblk->nb_bmsafemap;
1705
adp->ad_buf = bmsafemap->sm_buf;
1706
LIST_REMOVE(newblk, nb_deps);
1707
LIST_INSERT_HEAD(&bmsafemap->sm_allocdirecthd, adp, ad_deps);
5127
KASSERT(newblk->nb_list.wk_type == D_NEWBLK,
5128
("softdep_setup_allocdirect: newblk already initialized"));
5129
/*
5130
* Convert the newblk to an allocdirect.
5131
*/
5132
newblk->nb_list.wk_type = D_ALLOCDIRECT;
5133
adp = (struct allocdirect *)newblk;
5134
newblk->nb_freefrag = freefrag;
5135
adp->ad_offset = off;
5136
adp->ad_oldblkno = oldblkno;
5137
adp->ad_newsize = newsize;
5138
adp->ad_oldsize = oldsize;
5139
5140
/*
5141
* Finish initializing the journal.
5142
*/
5143
if ((jnewblk = newblk->nb_jnewblk) != NULL) {
5144
jnewblk->jn_ino = ip->i_number;
5145
jnewblk->jn_lbn = lbn;
5146
add_to_journal(&jnewblk->jn_list);
1708
5147
}
1709
LIST_REMOVE(newblk, nb_hash);
1710
free(newblk, M_NEWBLK);
1711
5148
if (freefrag && freefrag->ff_jdep != NULL &&
5149
freefrag->ff_jdep->wk_type == D_JFREEFRAG)
5150
add_to_journal(freefrag->ff_jdep);
1712
5151
inodedep_lookup(mp, ip->i_number, DEPALLOC | NODELAY, &inodedep);
1713
5152
adp->ad_inodedep = inodedep;
1714
WORKLIST_INSERT(&bp->b_dep, &adp->ad_list);
5153
5154
WORKLIST_INSERT(&bp->b_dep, &newblk->nb_list);
1715
5155
/*
1716
5156
* The list of allocdirects must be kept in sorted and ascending
1717
5157
* order so that the rollback routines can quickly determine the
1726
5166
*/
1727
5167
adphead = &inodedep->id_newinoupdt;
1728
5168
oldadp = TAILQ_LAST(adphead, allocdirectlst);
1729
if (oldadp == NULL || oldadp->ad_lbn <= lbn) {
5169
if (oldadp == NULL || oldadp->ad_offset <= off) {
1730
5170
/* insert at end of list */
1731
5171
TAILQ_INSERT_TAIL(adphead, adp, ad_next);
1732
if (oldadp != NULL && oldadp->ad_lbn == lbn)
5172
if (oldadp != NULL && oldadp->ad_offset == off)
1733
5173
allocdirect_merge(adphead, adp, oldadp);
1734
5174
FREE_LOCK(&lk);
1735
5175
return;
1736
5176
}
1737
5177
TAILQ_FOREACH(oldadp, adphead, ad_next) {
1738
if (oldadp->ad_lbn >= lbn)
5178
if (oldadp->ad_offset >= off)
1739
5179
break;
1740
5180
}
1741
5181
if (oldadp == NULL)
1742
5182
panic("softdep_setup_allocdirect: lost entry");
1743
5183
/* insert in middle of list */
1744
5184
TAILQ_INSERT_BEFORE(oldadp, adp, ad_next);
1745
if (oldadp->ad_lbn == lbn)
5185
if (oldadp->ad_offset == off)
1746
5186
allocdirect_merge(adphead, adp, oldadp);
5187
1747
5188
FREE_LOCK(&lk);
1748
5189
}
1749
5190
1750
5191
/*
5192
* Merge a newer and older journal record to be stored either in a
5193
* newblock or freefrag. This handles aggregating journal records for
5194
* fragment allocation into a second record as well as replacing a
5195
* journal free with an aborted journal allocation. A segment for the
5196
* oldest record will be placed on wkhd if it has been written. If not
5197
* the segment for the newer record will suffice.
5198
*/
5199
static struct worklist *
5200
jnewblk_merge(new, old, wkhd)
5201
struct worklist *new;
5202
struct worklist *old;
5203
struct workhead *wkhd;
5204
{
5205
struct jnewblk *njnewblk;
5206
struct jnewblk *jnewblk;
5207
5208
/* Handle NULLs to simplify callers. */
5209
if (new == NULL)
5210
return (old);
5211
if (old == NULL)
5212
return (new);
5213
/* Replace a jfreefrag with a jnewblk. */
5214
if (new->wk_type == D_JFREEFRAG) {
5215
if (WK_JNEWBLK(old)->jn_blkno != WK_JFREEFRAG(new)->fr_blkno)
5216
panic("jnewblk_merge: blkno mismatch: %p, %p",
5217
old, new);
5218
cancel_jfreefrag(WK_JFREEFRAG(new));
5219
return (old);
5220
}
5221
if (old->wk_type != D_JNEWBLK || new->wk_type != D_JNEWBLK)
5222
panic("jnewblk_merge: Bad type: old %d new %d\n",
5223
old->wk_type, new->wk_type);
5224
/*
5225
* Handle merging of two jnewblk records that describe
5226
* different sets of fragments in the same block.
5227
*/
5228
jnewblk = WK_JNEWBLK(old);
5229
njnewblk = WK_JNEWBLK(new);
5230
if (jnewblk->jn_blkno != njnewblk->jn_blkno)
5231
panic("jnewblk_merge: Merging disparate blocks.");
5232
/*
5233
* The record may be rolled back in the cg.
5234
*/
5235
if (jnewblk->jn_state & UNDONE) {
5236
jnewblk->jn_state &= ~UNDONE;
5237
njnewblk->jn_state |= UNDONE;
5238
njnewblk->jn_state &= ~ATTACHED;
5239
}
5240
/*
5241
* We modify the newer addref and free the older so that if neither
5242
* has been written the most up-to-date copy will be on disk. If
5243
* both have been written but rolled back we only temporarily need
5244
* one of them to fix the bits when the cg write completes.
5245
*/
5246
jnewblk->jn_state |= ATTACHED | COMPLETE;
5247
njnewblk->jn_oldfrags = jnewblk->jn_oldfrags;
5248
cancel_jnewblk(jnewblk, wkhd);
5249
WORKLIST_REMOVE(&jnewblk->jn_list);
5250
free_jnewblk(jnewblk);
5251
return (new);
5252
}
5253
5254
/*
1751
5255
* Replace an old allocdirect dependency with a newer one.
1752
5256
* This routine must be called with splbio interrupts blocked.
1753
5257
*/
1759
5263
{
1760
5264
struct worklist *wk;
1761
5265
struct freefrag *freefrag;
1762
struct newdirblk *newdirblk;
1763
5266
5267
freefrag = NULL;
1764
5268
mtx_assert(&lk, MA_OWNED);
1765
5269
if (newadp->ad_oldblkno != oldadp->ad_newblkno ||
1766
5270
newadp->ad_oldsize != oldadp->ad_newsize ||
1767
newadp->ad_lbn >= NDADDR)
5271
newadp->ad_offset >= NDADDR)
1768
5272
panic("%s %jd != new %jd || old size %ld != new %ld",
1769
5273
"allocdirect_merge: old blkno",
1770
5274
(intmax_t)newadp->ad_oldblkno,
1779
5283
* This action is done by swapping the freefrag dependencies.
1780
5284
* The new dependency gains the old one's freefrag, and the
1781
5285
* old one gets the new one and then immediately puts it on
1782
* the worklist when it is freed by free_allocdirect. It is
5286
* the worklist when it is freed by free_newblk. It is
1783
5287
* not possible to do this swap when the old dependency had a
1784
5288
* non-zero size but no previous fragment to free. This condition
1785
5289
* arises when the new block is an extension of the old block.
1788
5292
* the old dependency, so cannot legitimately be freed until the
1789
5293
* conditions for the new dependency are fulfilled.
1790
5294
*/
5295
freefrag = newadp->ad_freefrag;
1791
5296
if (oldadp->ad_freefrag != NULL || oldadp->ad_oldblkno == 0) {
1792
freefrag = newadp->ad_freefrag;
1793
5297
newadp->ad_freefrag = oldadp->ad_freefrag;
1794
5298
oldadp->ad_freefrag = freefrag;
1795
5299
}
1798
5302
* move it from the old allocdirect to the new allocdirect.
1799
5303
*/
1800
5304
if ((wk = LIST_FIRST(&oldadp->ad_newdirblk)) != NULL) {
1801
newdirblk = WK_NEWDIRBLK(wk);
1802
WORKLIST_REMOVE(&newdirblk->db_list);
5305
WORKLIST_REMOVE(wk);
1803
5306
if (!LIST_EMPTY(&oldadp->ad_newdirblk))
1804
5307
panic("allocdirect_merge: extra newdirblk");
1805
WORKLIST_INSERT(&newadp->ad_newdirblk, &newdirblk->db_list);
1806
}
1807
free_allocdirect(adphead, oldadp, 0);
1808
}
1809
1810
/*
1811
* Allocate a new freefrag structure if needed.
5308
WORKLIST_INSERT(&newadp->ad_newdirblk, wk);
5309
}
5310
TAILQ_REMOVE(adphead, oldadp, ad_next);
5311
/*
5312
* We need to move any journal dependencies over to the freefrag
5313
* that releases this block if it exists. Otherwise we are
5314
* extending an existing block and we'll wait until that is
5315
* complete to release the journal space and extend the
5316
* new journal to cover this old space as well.
5317
*/
5318
if (freefrag == NULL) {
5319
if (oldadp->ad_newblkno != newadp->ad_newblkno)
5320
panic("allocdirect_merge: %jd != %jd",
5321
oldadp->ad_newblkno, newadp->ad_newblkno);
5322
newadp->ad_block.nb_jnewblk = (struct jnewblk *)
5323
jnewblk_merge(&newadp->ad_block.nb_jnewblk->jn_list,
5324
&oldadp->ad_block.nb_jnewblk->jn_list,
5325
&newadp->ad_block.nb_jwork);
5326
oldadp->ad_block.nb_jnewblk = NULL;
5327
cancel_newblk(&oldadp->ad_block, NULL,
5328
&newadp->ad_block.nb_jwork);
5329
} else {
5330
wk = (struct worklist *) cancel_newblk(&oldadp->ad_block,
5331
&freefrag->ff_list, &freefrag->ff_jwork);
5332
freefrag->ff_jdep = jnewblk_merge(freefrag->ff_jdep, wk,
5333
&freefrag->ff_jwork);
5334
}
5335
free_newblk(&oldadp->ad_block);
5336
}
5337
5338
/*
5339
* Allocate a jfreefrag structure to journal a single block free.
5340
*/
5341
static struct jfreefrag *
5342
newjfreefrag(freefrag, ip, blkno, size, lbn)
5343
struct freefrag *freefrag;
5344
struct inode *ip;
5345
ufs2_daddr_t blkno;
5346
long size;
5347
ufs_lbn_t lbn;
5348
{
5349
struct jfreefrag *jfreefrag;
5350
struct fs *fs;
5351
5352
fs = ip->i_fs;
5353
jfreefrag = malloc(sizeof(struct jfreefrag), M_JFREEFRAG,
5354
M_SOFTDEP_FLAGS);
5355
workitem_alloc(&jfreefrag->fr_list, D_JFREEFRAG, UFSTOVFS(ip->i_ump));
5356
jfreefrag->fr_jsegdep = newjsegdep(&jfreefrag->fr_list);
5357
jfreefrag->fr_state = ATTACHED | DEPCOMPLETE;
5358
jfreefrag->fr_ino = ip->i_number;
5359
jfreefrag->fr_lbn = lbn;
5360
jfreefrag->fr_blkno = blkno;
5361
jfreefrag->fr_frags = numfrags(fs, size);
5362
jfreefrag->fr_freefrag = freefrag;
5363
5364
return (jfreefrag);
5365
}
5366
5367
/*
5368
* Allocate a new freefrag structure.
1812
5369
*/
1813
5370
static struct freefrag *
1814
newfreefrag(ip, blkno, size)
5371
newfreefrag(ip, blkno, size, lbn)
1815
5372
struct inode *ip;
1816
5373
ufs2_daddr_t blkno;
1817
5374
long size;
5375
ufs_lbn_t lbn;
1818
5376
{
1819
5377
struct freefrag *freefrag;
1820
5378
struct fs *fs;
1821
5379
1822
if (blkno == 0)
1823
return (NULL);
5380
CTR4(KTR_SUJ, "newfreefrag: ino %d blkno %jd size %ld lbn %jd",
5381
ip->i_number, blkno, size, lbn);
1824
5382
fs = ip->i_fs;
1825
5383
if (fragnum(fs, blkno) + numfrags(fs, size) > fs->fs_frag)
1826
5384
panic("newfreefrag: frag size");
1827
5385
freefrag = malloc(sizeof(struct freefrag),
1828
M_FREEFRAG, M_SOFTDEP_FLAGS);
5386
M_FREEFRAG, M_SOFTDEP_FLAGS);
1829
5387
workitem_alloc(&freefrag->ff_list, D_FREEFRAG, UFSTOVFS(ip->i_ump));
5388
freefrag->ff_state = ATTACHED;
5389
LIST_INIT(&freefrag->ff_jwork);
1830
5390
freefrag->ff_inum = ip->i_number;
5391
freefrag->ff_vtype = ITOV(ip)->v_type;
1831
5392
freefrag->ff_blkno = blkno;
1832
5393
freefrag->ff_fragsize = size;
5394
5395
if (MOUNTEDSUJ(UFSTOVFS(ip->i_ump))) {
5396
freefrag->ff_jdep = (struct worklist *)
5397
newjfreefrag(freefrag, ip, blkno, size, lbn);
5398
} else {
5399
freefrag->ff_state |= DEPCOMPLETE;
5400
freefrag->ff_jdep = NULL;
5401
}
5402
1833
5403
return (freefrag);
1834
5404
}
1835
5405
1842
5412
struct freefrag *freefrag;
1843
5413
{
1844
5414
struct ufsmount *ump = VFSTOUFS(freefrag->ff_list.wk_mp);
5415
struct workhead wkhd;
1845
5416
5417
CTR3(KTR_SUJ,
5418
"handle_workitem_freefrag: ino %d blkno %jd size %ld",
5419
freefrag->ff_inum, freefrag->ff_blkno, freefrag->ff_fragsize);
5420
/*
5421
* It would be illegal to add new completion items to the
5422
* freefrag after it was schedule to be done so it must be
5423
* safe to modify the list head here.
5424
*/
5425
LIST_INIT(&wkhd);
5426
ACQUIRE_LOCK(&lk);
5427
LIST_SWAP(&freefrag->ff_jwork, &wkhd, worklist, wk_list);
5428
/*
5429
* If the journal has not been written we must cancel it here.
5430
*/
5431
if (freefrag->ff_jdep) {
5432
if (freefrag->ff_jdep->wk_type != D_JNEWBLK)
5433
panic("handle_workitem_freefrag: Unexpected type %d\n",
5434
freefrag->ff_jdep->wk_type);
5435
cancel_jnewblk(WK_JNEWBLK(freefrag->ff_jdep), &wkhd);
5436
}
5437
FREE_LOCK(&lk);
1846
5438
ffs_blkfree(ump, ump->um_fs, ump->um_devvp, freefrag->ff_blkno,
1847
freefrag->ff_fragsize, freefrag->ff_inum);
5439
freefrag->ff_fragsize, freefrag->ff_inum, freefrag->ff_vtype, &wkhd);
1848
5440
ACQUIRE_LOCK(&lk);
1849
5441
WORKITEM_FREE(freefrag, D_FREEFRAG);
1850
5442
FREE_LOCK(&lk);
1856
5448
* See the description of softdep_setup_allocdirect above for details.
1857
5449
*/
1858
5450
void
1859
softdep_setup_allocext(ip, lbn, newblkno, oldblkno, newsize, oldsize, bp)
5451
softdep_setup_allocext(ip, off, newblkno, oldblkno, newsize, oldsize, bp)
1860
5452
struct inode *ip;
1861
ufs_lbn_t lbn;
5453
ufs_lbn_t off;
1862
5454
ufs2_daddr_t newblkno;
1863
5455
ufs2_daddr_t oldblkno;
1864
5456
long newsize;
1867
5459
{
1868
5460
struct allocdirect *adp, *oldadp;
1869
5461
struct allocdirectlst *adphead;
1870
struct bmsafemap *bmsafemap;
5462
struct freefrag *freefrag;
1871
5463
struct inodedep *inodedep;
5464
struct jnewblk *jnewblk;
1872
5465
struct newblk *newblk;
1873
5466
struct mount *mp;
1874
5467
ufs_lbn_t lbn;
5468
5469
if (off >= NXADDR)
5470
panic("softdep_setup_allocext: lbn %lld > NXADDR",
5471
(long long)off);
5472
5473
lbn = bp->b_lblkno;
1875
5474
mp = UFSTOVFS(ip->i_ump);
1876
adp = malloc(sizeof(struct allocdirect),
1877
M_ALLOCDIRECT, M_SOFTDEP_FLAGS|M_ZERO);
1878
workitem_alloc(&adp->ad_list, D_ALLOCDIRECT, mp);
1879
adp->ad_lbn = lbn;
1880
adp->ad_newblkno = newblkno;
1881
adp->ad_oldblkno = oldblkno;
1882
adp->ad_newsize = newsize;
1883
adp->ad_oldsize = oldsize;
1884
adp->ad_state = ATTACHED | EXTDATA;
1885
LIST_INIT(&adp->ad_newdirblk);
1886
if (newblkno == oldblkno)
1887
adp->ad_freefrag = NULL;
5475
if (oldblkno && oldblkno != newblkno)
5476
freefrag = newfreefrag(ip, oldblkno, oldsize, lbn);
1888
5477
else
1889
adp->ad_freefrag = newfreefrag(ip, oldblkno, oldsize);
5478
freefrag = NULL;
1890
5479
1891
5480
ACQUIRE_LOCK(&lk);
1892
if (newblk_lookup(ip->i_fs, newblkno, 0, &newblk) == 0)
5481
if (newblk_lookup(mp, newblkno, 0, &newblk) == 0)
1893
5482
panic("softdep_setup_allocext: lost block");
5483
KASSERT(newblk->nb_list.wk_type == D_NEWBLK,
5484
("softdep_setup_allocext: newblk already initialized"));
5485
/*
5486
* Convert the newblk to an allocdirect.
5487
*/
5488
newblk->nb_list.wk_type = D_ALLOCDIRECT;
5489
adp = (struct allocdirect *)newblk;
5490
newblk->nb_freefrag = freefrag;
5491
adp->ad_offset = off;
5492
adp->ad_oldblkno = oldblkno;
5493
adp->ad_newsize = newsize;
5494
adp->ad_oldsize = oldsize;
5495
adp->ad_state |= EXTDATA;
1894
5496
5497
/*
5498
* Finish initializing the journal.
5499
*/
5500
if ((jnewblk = newblk->nb_jnewblk) != NULL) {
5501
jnewblk->jn_ino = ip->i_number;
5502
jnewblk->jn_lbn = lbn;
5503
add_to_journal(&jnewblk->jn_list);
5504
}
5505
if (freefrag && freefrag->ff_jdep != NULL &&
5506
freefrag->ff_jdep->wk_type == D_JFREEFRAG)
5507
add_to_journal(freefrag->ff_jdep);
1895
5508
inodedep_lookup(mp, ip->i_number, DEPALLOC | NODELAY, &inodedep);
1896
5509
adp->ad_inodedep = inodedep;
1897
5510
1898
if (newblk->nb_state == DEPCOMPLETE) {
1899
adp->ad_state |= DEPCOMPLETE;
1900
adp->ad_buf = NULL;
1901
} else {
1902
bmsafemap = newblk->nb_bmsafemap;
1903
adp->ad_buf = bmsafemap->sm_buf;
1904
LIST_REMOVE(newblk, nb_deps);
1905
LIST_INSERT_HEAD(&bmsafemap->sm_allocdirecthd, adp, ad_deps);
1906
}
1907
LIST_REMOVE(newblk, nb_hash);
1908
free(newblk, M_NEWBLK);
1909
1910
WORKLIST_INSERT(&bp->b_dep, &adp->ad_list);
1911
if (lbn >= NXADDR)
1912
panic("softdep_setup_allocext: lbn %lld > NXADDR",
1913
(long long)lbn);
5511
WORKLIST_INSERT(&bp->b_dep, &newblk->nb_list);
1914
5512
/*
1915
5513
* The list of allocdirects must be kept in sorted and ascending
1916
5514
* order so that the rollback routines can quickly determine the
1925
5523
*/
1926
5524
adphead = &inodedep->id_newextupdt;
1927
5525
oldadp = TAILQ_LAST(adphead, allocdirectlst);
1928
if (oldadp == NULL || oldadp->ad_lbn <= lbn) {
5526
if (oldadp == NULL || oldadp->ad_offset <= off) {
1929
5527
/* insert at end of list */
1930
5528
TAILQ_INSERT_TAIL(adphead, adp, ad_next);
1931
if (oldadp != NULL && oldadp->ad_lbn == lbn)
5529
if (oldadp != NULL && oldadp->ad_offset == off)
1932
5530
allocdirect_merge(adphead, adp, oldadp);
1933
5531
FREE_LOCK(&lk);
1934
5532
return;
1935
5533
}
1936
5534
TAILQ_FOREACH(oldadp, adphead, ad_next) {
1937
if (oldadp->ad_lbn >= lbn)
5535
if (oldadp->ad_offset >= off)
1938
5536
break;
1939
5537
}
1940
5538
if (oldadp == NULL)
1941
5539
panic("softdep_setup_allocext: lost entry");
1942
5540
/* insert in middle of list */
1943
5541
TAILQ_INSERT_BEFORE(oldadp, adp, ad_next);
1944
if (oldadp->ad_lbn == lbn)
5542
if (oldadp->ad_offset == off)
1945
5543
allocdirect_merge(adphead, adp, oldadp);
1946
5544
FREE_LOCK(&lk);
1947
5545
}
1975
5573
* Allocate a new allocindir structure.
1976
5574
*/
1977
5575
static struct allocindir *
1978
newallocindir(ip, ptrno, newblkno, oldblkno)
5576
newallocindir(ip, ptrno, newblkno, oldblkno, lbn)
1979
5577
struct inode *ip; /* inode for file being extended */
1980
5578
int ptrno; /* offset of pointer in indirect block */
1981
5579
ufs2_daddr_t newblkno; /* disk block number being added */
1982
5580
ufs2_daddr_t oldblkno; /* previous block number, 0 if none */
5581
ufs_lbn_t lbn;
1983
5582
{
5583
struct newblk *newblk;
1984
5584
struct allocindir *aip;
5585
struct freefrag *freefrag;
5586
struct jnewblk *jnewblk;
1985
5587
1986
aip = malloc(sizeof(struct allocindir),
1987
M_ALLOCINDIR, M_SOFTDEP_FLAGS|M_ZERO);
1988
workitem_alloc(&aip->ai_list, D_ALLOCINDIR, UFSTOVFS(ip->i_ump));
1989
aip->ai_state = ATTACHED;
5588
if (oldblkno)
5589
freefrag = newfreefrag(ip, oldblkno, ip->i_fs->fs_bsize, lbn);
5590
else
5591
freefrag = NULL;
5592
ACQUIRE_LOCK(&lk);
5593
if (newblk_lookup(UFSTOVFS(ip->i_ump), newblkno, 0, &newblk) == 0)
5594
panic("new_allocindir: lost block");
5595
KASSERT(newblk->nb_list.wk_type == D_NEWBLK,
5596
("newallocindir: newblk already initialized"));
5597
newblk->nb_list.wk_type = D_ALLOCINDIR;
5598
newblk->nb_freefrag = freefrag;
5599
aip = (struct allocindir *)newblk;
1990
5600
aip->ai_offset = ptrno;
1991
aip->ai_newblkno = newblkno;
1992
5601
aip->ai_oldblkno = oldblkno;
1993
aip->ai_freefrag = newfreefrag(ip, oldblkno, ip->i_fs->fs_bsize);
5602
aip->ai_lbn = lbn;
5603
if ((jnewblk = newblk->nb_jnewblk) != NULL) {
5604
jnewblk->jn_ino = ip->i_number;
5605
jnewblk->jn_lbn = lbn;
5606
add_to_journal(&jnewblk->jn_list);
5607
}
5608
if (freefrag && freefrag->ff_jdep != NULL &&
5609
freefrag->ff_jdep->wk_type == D_JFREEFRAG)
5610
add_to_journal(freefrag->ff_jdep);
1994
5611
return (aip);
1995
5612
}
1996
5613
2008
5625
ufs2_daddr_t oldblkno; /* previous block number, 0 if none */
2009
5626
struct buf *nbp; /* buffer holding allocated page */
2010
5627
{
5628
struct inodedep *inodedep;
5629
struct freefrag *freefrag;
2011
5630
struct allocindir *aip;
2012
5631
struct pagedep *pagedep;
5632
struct mount *mp;
5633
int dflags;
2013
5634
5635
if (lbn != nbp->b_lblkno)
5636
panic("softdep_setup_allocindir_page: lbn %jd != lblkno %jd",
5637
lbn, bp->b_lblkno);
5638
CTR4(KTR_SUJ,
5639
"softdep_setup_allocindir_page: ino %d blkno %jd oldblkno %jd "
5640
"lbn %jd", ip->i_number, newblkno, oldblkno, lbn);
2014
5641
ASSERT_VOP_LOCKED(ITOV(ip), "softdep_setup_allocindir_page");
2015
aip = newallocindir(ip, ptrno, newblkno, oldblkno);
2016
ACQUIRE_LOCK(&lk);
5642
mp = UFSTOVFS(ip->i_ump);
5643
aip = newallocindir(ip, ptrno, newblkno, oldblkno, lbn);
5644
dflags = DEPALLOC;
5645
if (IS_SNAPSHOT(ip))
5646
dflags |= NODELAY;
5647
(void) inodedep_lookup(mp, ip->i_number, dflags, &inodedep);
2017
5648
/*
2018
5649
* If we are allocating a directory page, then we must
2019
5650
* allocate an associated pagedep to track additions and
2020
5651
* deletions.
2021
5652
*/
2022
if ((ip->i_mode & IFMT) == IFDIR &&
2023
pagedep_lookup(ip, lbn, DEPALLOC, &pagedep) == 0)
2024
WORKLIST_INSERT(&nbp->b_dep, &pagedep->pd_list);
2025
WORKLIST_INSERT(&nbp->b_dep, &aip->ai_list);
2026
setup_allocindir_phase2(bp, ip, aip);
5653
if ((ip->i_mode & IFMT) == IFDIR)
5654
pagedep_lookup(mp, nbp, ip->i_number, lbn, DEPALLOC, &pagedep);
5655
WORKLIST_INSERT(&nbp->b_dep, &aip->ai_block.nb_list);
5656
freefrag = setup_allocindir_phase2(bp, ip, inodedep, aip, lbn);
2027
5657
FREE_LOCK(&lk);
5658
if (freefrag)
5659
handle_workitem_freefrag(freefrag);
2028
5660
}
2029
5661
2030
5662
/*
2039
5671
int ptrno; /* offset of pointer in indirect block */
2040
5672
ufs2_daddr_t newblkno; /* disk block number being added */
2041
5673
{
5674
struct inodedep *inodedep;
2042
5675
struct allocindir *aip;
5676
ufs_lbn_t lbn;
5677
int dflags;
2043
5678
5679
CTR3(KTR_SUJ,
5680
"softdep_setup_allocindir_meta: ino %d blkno %jd ptrno %d",
5681
ip->i_number, newblkno, ptrno);
5682
lbn = nbp->b_lblkno;
2044
5683
ASSERT_VOP_LOCKED(ITOV(ip), "softdep_setup_allocindir_meta");
2045
aip = newallocindir(ip, ptrno, newblkno, 0);
2046
ACQUIRE_LOCK(&lk);
2047
WORKLIST_INSERT(&nbp->b_dep, &aip->ai_list);
2048
setup_allocindir_phase2(bp, ip, aip);
5684
aip = newallocindir(ip, ptrno, newblkno, 0, lbn);
5685
dflags = DEPALLOC;
5686
if (IS_SNAPSHOT(ip))
5687
dflags |= NODELAY;
5688
inodedep_lookup(UFSTOVFS(ip->i_ump), ip->i_number, dflags, &inodedep);
5689
WORKLIST_INSERT(&nbp->b_dep, &aip->ai_block.nb_list);
5690
if (setup_allocindir_phase2(bp, ip, inodedep, aip, lbn))
5691
panic("softdep_setup_allocindir_meta: Block already existed");
2049
5692
FREE_LOCK(&lk);
2050
5693
}
2051
5694
2052
/*
2053
* Called to finish the allocation of the "aip" allocated
2054
* by one of the two routines above.
2055
*/
2056
static void
2057
setup_allocindir_phase2(bp, ip, aip)
2058
struct buf *bp; /* in-memory copy of the indirect block */
2059
struct inode *ip; /* inode for file being extended */
2060
struct allocindir *aip; /* allocindir allocated by the above routines */
2061
{
2062
struct worklist *wk;
5695
static void
5696
indirdep_complete(indirdep)
5697
struct indirdep *indirdep;
5698
{
5699
struct allocindir *aip;
5700
5701
LIST_REMOVE(indirdep, ir_next);
5702
indirdep->ir_state |= DEPCOMPLETE;
5703
5704
while ((aip = LIST_FIRST(&indirdep->ir_completehd)) != NULL) {
5705
LIST_REMOVE(aip, ai_next);
5706
free_newblk(&aip->ai_block);
5707
}
5708
/*
5709
* If this indirdep is not attached to a buf it was simply waiting
5710
* on completion to clear completehd. free_indirdep() asserts
5711
* that nothing is dangling.
5712
*/
5713
if ((indirdep->ir_state & ONWORKLIST) == 0)
5714
free_indirdep(indirdep);
5715
}
5716
5717
static struct indirdep *
5718
indirdep_lookup(mp, ip, bp)
5719
struct mount *mp;
5720
struct inode *ip;
5721
struct buf *bp;
5722
{
2063
5723
struct indirdep *indirdep, *newindirdep;
2064
struct bmsafemap *bmsafemap;
2065
struct allocindir *oldaip;
2066
struct freefrag *freefrag;
2067
5724
struct newblk *newblk;
5725
struct worklist *wk;
5726
struct fs *fs;
2068
5727
ufs2_daddr_t blkno;
2069
5728
2070
5729
mtx_assert(&lk, MA_OWNED);
2071
if (bp->b_lblkno >= 0)
2072
panic("setup_allocindir_phase2: not indir blk");
2073
for (indirdep = NULL, newindirdep = NULL; ; ) {
5730
indirdep = NULL;
5731
newindirdep = NULL;
5732
fs = ip->i_fs;
5733
for (;;) {
2074
5734
LIST_FOREACH(wk, &bp->b_dep, wk_list) {
2075
5735
if (wk->wk_type != D_INDIRDEP)
2076
5736
continue;
2077
5737
indirdep = WK_INDIRDEP(wk);
2078
5738
break;
2079
5739
}
2080
if (indirdep == NULL && newindirdep) {
2081
indirdep = newindirdep;
2082
WORKLIST_INSERT(&bp->b_dep, &indirdep->ir_list);
2083
newindirdep = NULL;
2084
}
2085
if (indirdep) {
2086
if (newblk_lookup(ip->i_fs, aip->ai_newblkno, 0,
2087
&newblk) == 0)
2088
panic("setup_allocindir: lost block");
2089
if (newblk->nb_state == DEPCOMPLETE) {
2090
aip->ai_state |= DEPCOMPLETE;
2091
aip->ai_buf = NULL;
2092
} else {
2093
bmsafemap = newblk->nb_bmsafemap;
2094
aip->ai_buf = bmsafemap->sm_buf;
2095
LIST_REMOVE(newblk, nb_deps);
2096
LIST_INSERT_HEAD(&bmsafemap->sm_allocindirhd,
2097
aip, ai_deps);
2098
}
2099
LIST_REMOVE(newblk, nb_hash);
2100
free(newblk, M_NEWBLK);
2101
aip->ai_indirdep = indirdep;
2102
/*
2103
* Check to see if there is an existing dependency
2104
* for this block. If there is, merge the old
2105
* dependency into the new one.
2106
*/
2107
if (aip->ai_oldblkno == 0)
2108
oldaip = NULL;
2109
else
2110
2111
LIST_FOREACH(oldaip, &indirdep->ir_deplisthd, ai_next)
2112
if (oldaip->ai_offset == aip->ai_offset)
2113
break;
2114
freefrag = NULL;
2115
if (oldaip != NULL) {
2116
if (oldaip->ai_newblkno != aip->ai_oldblkno)
2117
panic("setup_allocindir_phase2: blkno");
2118
aip->ai_oldblkno = oldaip->ai_oldblkno;
2119
freefrag = aip->ai_freefrag;
2120
aip->ai_freefrag = oldaip->ai_freefrag;
2121
oldaip->ai_freefrag = NULL;
2122
free_allocindir(oldaip, NULL);
2123
}
2124
LIST_INSERT_HEAD(&indirdep->ir_deplisthd, aip, ai_next);
2125
if (ip->i_ump->um_fstype == UFS1)
2126
((ufs1_daddr_t *)indirdep->ir_savebp->b_data)
2127
[aip->ai_offset] = aip->ai_oldblkno;
2128
else
2129
((ufs2_daddr_t *)indirdep->ir_savebp->b_data)
2130
[aip->ai_offset] = aip->ai_oldblkno;
2131
FREE_LOCK(&lk);
2132
if (freefrag != NULL)
2133
handle_workitem_freefrag(freefrag);
2134
} else
2135
FREE_LOCK(&lk);
2136
if (newindirdep) {
2137
newindirdep->ir_savebp->b_flags |= B_INVAL | B_NOCACHE;
2138
brelse(newindirdep->ir_savebp);
2139
ACQUIRE_LOCK(&lk);
2140
WORKITEM_FREE((caddr_t)newindirdep, D_INDIRDEP);
2141
if (indirdep)
2142
break;
2143
FREE_LOCK(&lk);
2144
}
2145
if (indirdep) {
2146
ACQUIRE_LOCK(&lk);
5740
/* Found on the buffer worklist, no new structure to free. */
5741
if (indirdep != NULL && newindirdep == NULL)
5742
return (indirdep);
5743
if (indirdep != NULL && newindirdep != NULL)
5744
panic("indirdep_lookup: simultaneous create");
5745
/* None found on the buffer and a new structure is ready. */
5746
if (indirdep == NULL && newindirdep != NULL)
2147
5747
break;
2148
}
5748
/* None found and no new structure available. */
5749
FREE_LOCK(&lk);
2149
5750
newindirdep = malloc(sizeof(struct indirdep),
2150
M_INDIRDEP, M_SOFTDEP_FLAGS);
2151
workitem_alloc(&newindirdep->ir_list, D_INDIRDEP,
2152
UFSTOVFS(ip->i_ump));
5751
M_INDIRDEP, M_SOFTDEP_FLAGS);
5752
workitem_alloc(&newindirdep->ir_list, D_INDIRDEP, mp);
2153
5753
newindirdep->ir_state = ATTACHED;
2154
5754
if (ip->i_ump->um_fstype == UFS1)
2155
5755
newindirdep->ir_state |= UFS1FMT;
5756
TAILQ_INIT(&newindirdep->ir_trunc);
5757
newindirdep->ir_saveddata = NULL;
2156
5758
LIST_INIT(&newindirdep->ir_deplisthd);
2157
5759
LIST_INIT(&newindirdep->ir_donehd);
5760
LIST_INIT(&newindirdep->ir_writehd);
5761
LIST_INIT(&newindirdep->ir_completehd);
2158
5762
if (bp->b_blkno == bp->b_lblkno) {
2159
5763
ufs_bmaparray(bp->b_vp, bp->b_lblkno, &blkno, bp,
2160
5764
NULL, NULL);
2161
5765
bp->b_blkno = blkno;
2162
5766
}
5767
newindirdep->ir_freeblks = NULL;
2163
5768
newindirdep->ir_savebp =
2164
5769
getblk(ip->i_devvp, bp->b_blkno, bp->b_bcount, 0, 0, 0);
5770
newindirdep->ir_bp = bp;
2165
5771
BUF_KERNPROC(newindirdep->ir_savebp);
2166
5772
bcopy(bp->b_data, newindirdep->ir_savebp->b_data, bp->b_bcount);
2167
5773
ACQUIRE_LOCK(&lk);
2168
5774
}
5775
indirdep = newindirdep;
5776
WORKLIST_INSERT(&bp->b_dep, &indirdep->ir_list);
5777
/*
5778
* If the block is not yet allocated we don't set DEPCOMPLETE so
5779
* that we don't free dependencies until the pointers are valid.
5780
* This could search b_dep for D_ALLOCDIRECT/D_ALLOCINDIR rather
5781
* than using the hash.
5782
*/
5783
if (newblk_lookup(mp, dbtofsb(fs, bp->b_blkno), 0, &newblk))
5784
LIST_INSERT_HEAD(&newblk->nb_indirdeps, indirdep, ir_next);
5785
else
5786
indirdep->ir_state |= DEPCOMPLETE;
5787
return (indirdep);
5788
}
5789
5790
/*
5791
* Called to finish the allocation of the "aip" allocated
5792
* by one of the two routines above.
5793
*/
5794
static struct freefrag *
5795
setup_allocindir_phase2(bp, ip, inodedep, aip, lbn)
5796
struct buf *bp; /* in-memory copy of the indirect block */
5797
struct inode *ip; /* inode for file being extended */
5798
struct inodedep *inodedep; /* Inodedep for ip */
5799
struct allocindir *aip; /* allocindir allocated by the above routines */
5800
ufs_lbn_t lbn; /* Logical block number for this block. */
5801
{
5802
struct fs *fs;
5803
struct indirdep *indirdep;
5804
struct allocindir *oldaip;
5805
struct freefrag *freefrag;
5806
struct mount *mp;
5807
5808
mtx_assert(&lk, MA_OWNED);
5809
mp = UFSTOVFS(ip->i_ump);
5810
fs = ip->i_fs;
5811
if (bp->b_lblkno >= 0)
5812
panic("setup_allocindir_phase2: not indir blk");
5813
KASSERT(aip->ai_offset >= 0 && aip->ai_offset < NINDIR(fs),
5814
("setup_allocindir_phase2: Bad offset %d", aip->ai_offset));
5815
indirdep = indirdep_lookup(mp, ip, bp);
5816
KASSERT(indirdep->ir_savebp != NULL,
5817
("setup_allocindir_phase2 NULL ir_savebp"));
5818
aip->ai_indirdep = indirdep;
5819
/*
5820
* Check for an unwritten dependency for this indirect offset. If
5821
* there is, merge the old dependency into the new one. This happens
5822
* as a result of reallocblk only.
5823
*/
5824
freefrag = NULL;
5825
if (aip->ai_oldblkno != 0) {
5826
LIST_FOREACH(oldaip, &indirdep->ir_deplisthd, ai_next) {
5827
if (oldaip->ai_offset == aip->ai_offset) {
5828
freefrag = allocindir_merge(aip, oldaip);
5829
goto done;
5830
}
5831
}
5832
LIST_FOREACH(oldaip, &indirdep->ir_donehd, ai_next) {
5833
if (oldaip->ai_offset == aip->ai_offset) {
5834
freefrag = allocindir_merge(aip, oldaip);
5835
goto done;
5836
}
5837
}
5838
}
5839
done:
5840
LIST_INSERT_HEAD(&indirdep->ir_deplisthd, aip, ai_next);
5841
return (freefrag);
5842
}
5843
5844
/*
5845
* Merge two allocindirs which refer to the same block. Move newblock
5846
* dependencies and setup the freefrags appropriately.
5847
*/
5848
static struct freefrag *
5849
allocindir_merge(aip, oldaip)
5850
struct allocindir *aip;
5851
struct allocindir *oldaip;
5852
{
5853
struct freefrag *freefrag;
5854
struct worklist *wk;
5855
5856
if (oldaip->ai_newblkno != aip->ai_oldblkno)
5857
panic("allocindir_merge: blkno");
5858
aip->ai_oldblkno = oldaip->ai_oldblkno;
5859
freefrag = aip->ai_freefrag;
5860
aip->ai_freefrag = oldaip->ai_freefrag;
5861
oldaip->ai_freefrag = NULL;
5862
KASSERT(freefrag != NULL, ("setup_allocindir_phase2: No freefrag"));
5863
/*
5864
* If we are tracking a new directory-block allocation,
5865
* move it from the old allocindir to the new allocindir.
5866
*/
5867
if ((wk = LIST_FIRST(&oldaip->ai_newdirblk)) != NULL) {
5868
WORKLIST_REMOVE(wk);
5869
if (!LIST_EMPTY(&oldaip->ai_newdirblk))
5870
panic("allocindir_merge: extra newdirblk");
5871
WORKLIST_INSERT(&aip->ai_newdirblk, wk);
5872
}
5873
/*
5874
* We can skip journaling for this freefrag and just complete
5875
* any pending journal work for the allocindir that is being
5876
* removed after the freefrag completes.
5877
*/
5878
if (freefrag->ff_jdep)
5879
cancel_jfreefrag(WK_JFREEFRAG(freefrag->ff_jdep));
5880
LIST_REMOVE(oldaip, ai_next);
5881
freefrag->ff_jdep = (struct worklist *)cancel_newblk(&oldaip->ai_block,
5882
&freefrag->ff_list, &freefrag->ff_jwork);
5883
free_newblk(&oldaip->ai_block);
5884
5885
return (freefrag);
5886
}
5887
5888
static inline void
5889
setup_freedirect(freeblks, ip, i, needj)
5890
struct freeblks *freeblks;
5891
struct inode *ip;
5892
int i;
5893
int needj;
5894
{
5895
ufs2_daddr_t blkno;
5896
int frags;
5897
5898
blkno = DIP(ip, i_db[i]);
5899
if (blkno == 0)
5900
return;
5901
DIP_SET(ip, i_db[i], 0);
5902
frags = sblksize(ip->i_fs, ip->i_size, i);
5903
frags = numfrags(ip->i_fs, frags);
5904
newfreework(ip->i_ump, freeblks, NULL, i, blkno, frags, 0, needj);
5905
}
5906
5907
static inline void
5908
setup_freeext(freeblks, ip, i, needj)
5909
struct freeblks *freeblks;
5910
struct inode *ip;
5911
int i;
5912
int needj;
5913
{
5914
ufs2_daddr_t blkno;
5915
int frags;
5916
5917
blkno = ip->i_din2->di_extb[i];
5918
if (blkno == 0)
5919
return;
5920
ip->i_din2->di_extb[i] = 0;
5921
frags = sblksize(ip->i_fs, ip->i_din2->di_extsize, i);
5922
frags = numfrags(ip->i_fs, frags);
5923
newfreework(ip->i_ump, freeblks, NULL, -1 - i, blkno, frags, 0, needj);
5924
}
5925
5926
static inline void
5927
setup_freeindir(freeblks, ip, i, lbn, needj)
5928
struct freeblks *freeblks;
5929
struct inode *ip;
5930
int i;
5931
ufs_lbn_t lbn;
5932
int needj;
5933
{
5934
ufs2_daddr_t blkno;
5935
5936
blkno = DIP(ip, i_ib[i]);
5937
if (blkno == 0)
5938
return;
5939
DIP_SET(ip, i_ib[i], 0);
5940
newfreework(ip->i_ump, freeblks, NULL, lbn, blkno, ip->i_fs->fs_frag,
5941
0, needj);
5942
}
5943
5944
static inline struct freeblks *
5945
newfreeblks(mp, ip)
5946
struct mount *mp;
5947
struct inode *ip;
5948
{
5949
struct freeblks *freeblks;
5950
5951
freeblks = malloc(sizeof(struct freeblks),
5952
M_FREEBLKS, M_SOFTDEP_FLAGS|M_ZERO);
5953
workitem_alloc(&freeblks->fb_list, D_FREEBLKS, mp);
5954
LIST_INIT(&freeblks->fb_jblkdephd);
5955
LIST_INIT(&freeblks->fb_jwork);
5956
freeblks->fb_ref = 0;
5957
freeblks->fb_cgwait = 0;
5958
freeblks->fb_state = ATTACHED;
5959
freeblks->fb_uid = ip->i_uid;
5960
freeblks->fb_inum = ip->i_number;
5961
freeblks->fb_vtype = ITOV(ip)->v_type;
5962
freeblks->fb_modrev = DIP(ip, i_modrev);
5963
freeblks->fb_devvp = ip->i_devvp;
5964
freeblks->fb_chkcnt = 0;
5965
freeblks->fb_len = 0;
5966
5967
return (freeblks);
5968
}
5969
5970
static void
5971
trunc_indirdep(indirdep, freeblks, bp, off)
5972
struct indirdep *indirdep;
5973
struct freeblks *freeblks;
5974
struct buf *bp;
5975
int off;
5976
{
5977
struct allocindir *aip, *aipn;
5978
5979
/*
5980
* The first set of allocindirs won't be in savedbp.
5981
*/
5982
LIST_FOREACH_SAFE(aip, &indirdep->ir_deplisthd, ai_next, aipn)
5983
if (aip->ai_offset > off)
5984
cancel_allocindir(aip, bp, freeblks, 1);
5985
LIST_FOREACH_SAFE(aip, &indirdep->ir_donehd, ai_next, aipn)
5986
if (aip->ai_offset > off)
5987
cancel_allocindir(aip, bp, freeblks, 1);
5988
/*
5989
* These will exist in savedbp.
5990
*/
5991
LIST_FOREACH_SAFE(aip, &indirdep->ir_writehd, ai_next, aipn)
5992
if (aip->ai_offset > off)
5993
cancel_allocindir(aip, NULL, freeblks, 0);
5994
LIST_FOREACH_SAFE(aip, &indirdep->ir_completehd, ai_next, aipn)
5995
if (aip->ai_offset > off)
5996
cancel_allocindir(aip, NULL, freeblks, 0);
5997
}
5998
5999
/*
6000
* Follow the chain of indirects down to lastlbn creating a freework
6001
* structure for each. This will be used to start indir_trunc() at
6002
* the right offset and create the journal records for the parrtial
6003
* truncation. A second step will handle the truncated dependencies.
6004
*/
6005
static int
6006
setup_trunc_indir(freeblks, ip, lbn, lastlbn, blkno)
6007
struct freeblks *freeblks;
6008
struct inode *ip;
6009
ufs_lbn_t lbn;
6010
ufs_lbn_t lastlbn;
6011
ufs2_daddr_t blkno;
6012
{
6013
struct indirdep *indirdep;
6014
struct indirdep *indirn;
6015
struct freework *freework;
6016
struct newblk *newblk;
6017
struct mount *mp;
6018
struct buf *bp;
6019
uint8_t *start;
6020
uint8_t *end;
6021
ufs_lbn_t lbnadd;
6022
int level;
6023
int error;
6024
int off;
6025
6026
6027
freework = NULL;
6028
if (blkno == 0)
6029
return (0);
6030
mp = freeblks->fb_list.wk_mp;
6031
bp = getblk(ITOV(ip), lbn, mp->mnt_stat.f_iosize, 0, 0, 0);
6032
if ((bp->b_flags & B_CACHE) == 0) {
6033
bp->b_blkno = blkptrtodb(VFSTOUFS(mp), blkno);
6034
bp->b_iocmd = BIO_READ;
6035
bp->b_flags &= ~B_INVAL;
6036
bp->b_ioflags &= ~BIO_ERROR;
6037
vfs_busy_pages(bp, 0);
6038
bp->b_iooffset = dbtob(bp->b_blkno);
6039
bstrategy(bp);
6040
curthread->td_ru.ru_inblock++;
6041
error = bufwait(bp);
6042
if (error) {
6043
brelse(bp);
6044
return (error);
6045
}
6046
}
6047
level = lbn_level(lbn);
6048
lbnadd = lbn_offset(ip->i_fs, level);
6049
/*
6050
* Compute the offset of the last block we want to keep. Store
6051
* in the freework the first block we want to completely free.
6052
*/
6053
off = (lastlbn - -(lbn + level)) / lbnadd;
6054
if (off + 1 == NINDIR(ip->i_fs))
6055
goto nowork;
6056
freework = newfreework(ip->i_ump, freeblks, NULL, lbn, blkno, 0, off+1,
6057
0);
6058
/*
6059
* Link the freework into the indirdep. This will prevent any new
6060
* allocations from proceeding until we are finished with the
6061
* truncate and the block is written.
6062
*/
6063
ACQUIRE_LOCK(&lk);
6064
indirdep = indirdep_lookup(mp, ip, bp);
6065
if (indirdep->ir_freeblks)
6066
panic("setup_trunc_indir: indirdep already truncated.");
6067
TAILQ_INSERT_TAIL(&indirdep->ir_trunc, freework, fw_next);
6068
freework->fw_indir = indirdep;
6069
/*
6070
* Cancel any allocindirs that will not make it to disk.
6071
* We have to do this for all copies of the indirdep that
6072
* live on this newblk.
6073
*/
6074
if ((indirdep->ir_state & DEPCOMPLETE) == 0) {
6075
newblk_lookup(mp, dbtofsb(ip->i_fs, bp->b_blkno), 0, &newblk);
6076
LIST_FOREACH(indirn, &newblk->nb_indirdeps, ir_next)
6077
trunc_indirdep(indirn, freeblks, bp, off);
6078
} else
6079
trunc_indirdep(indirdep, freeblks, bp, off);
6080
FREE_LOCK(&lk);
6081
/*
6082
* Creation is protected by the buf lock. The saveddata is only
6083
* needed if a full truncation follows a partial truncation but it
6084
* is difficult to allocate in that case so we fetch it anyway.
6085
*/
6086
if (indirdep->ir_saveddata == NULL)
6087
indirdep->ir_saveddata = malloc(bp->b_bcount, M_INDIRDEP,
6088
M_SOFTDEP_FLAGS);
6089
nowork:
6090
/* Fetch the blkno of the child and the zero start offset. */
6091
if (ip->i_ump->um_fstype == UFS1) {
6092
blkno = ((ufs1_daddr_t *)bp->b_data)[off];
6093
start = (uint8_t *)&((ufs1_daddr_t *)bp->b_data)[off+1];
6094
} else {
6095
blkno = ((ufs2_daddr_t *)bp->b_data)[off];
6096
start = (uint8_t *)&((ufs2_daddr_t *)bp->b_data)[off+1];
6097
}
6098
if (freework) {
6099
/* Zero the truncated pointers. */
6100
end = bp->b_data + bp->b_bcount;
6101
bzero(start, end - start);
6102
bdwrite(bp);
6103
} else
6104
bqrelse(bp);
6105
if (level == 0)
6106
return (0);
6107
lbn++; /* adjust level */
6108
lbn -= (off * lbnadd);
6109
return setup_trunc_indir(freeblks, ip, lbn, lastlbn, blkno);
6110
}
6111
6112
/*
6113
* Complete the partial truncation of an indirect block setup by
6114
* setup_trunc_indir(). This zeros the truncated pointers in the saved
6115
* copy and writes them to disk before the freeblks is allowed to complete.
6116
*/
6117
static void
6118
complete_trunc_indir(freework)
6119
struct freework *freework;
6120
{
6121
struct freework *fwn;
6122
struct indirdep *indirdep;
6123
struct buf *bp;
6124
uintptr_t start;
6125
int count;
6126
6127
indirdep = freework->fw_indir;
6128
for (;;) {
6129
bp = indirdep->ir_bp;
6130
/* See if the block was discarded. */
6131
if (bp == NULL)
6132
break;
6133
/* Inline part of getdirtybuf(). We dont want bremfree. */
6134
if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0)
6135
break;
6136
if (BUF_LOCK(bp,
6137
LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, &lk) == 0)
6138
BUF_UNLOCK(bp);
6139
ACQUIRE_LOCK(&lk);
6140
}
6141
mtx_assert(&lk, MA_OWNED);
6142
freework->fw_state |= DEPCOMPLETE;
6143
TAILQ_REMOVE(&indirdep->ir_trunc, freework, fw_next);
6144
/*
6145
* Zero the pointers in the saved copy.
6146
*/
6147
if (indirdep->ir_state & UFS1FMT)
6148
start = sizeof(ufs1_daddr_t);
6149
else
6150
start = sizeof(ufs2_daddr_t);
6151
start *= freework->fw_start;
6152
count = indirdep->ir_savebp->b_bcount - start;
6153
start += (uintptr_t)indirdep->ir_savebp->b_data;
6154
bzero((char *)start, count);
6155
/*
6156
* We need to start the next truncation in the list if it has not
6157
* been started yet.
6158
*/
6159
fwn = TAILQ_FIRST(&indirdep->ir_trunc);
6160
if (fwn != NULL) {
6161
if (fwn->fw_freeblks == indirdep->ir_freeblks)
6162
TAILQ_REMOVE(&indirdep->ir_trunc, fwn, fw_next);
6163
if ((fwn->fw_state & ONWORKLIST) == 0)
6164
freework_enqueue(fwn);
6165
}
6166
/*
6167
* If bp is NULL the block was fully truncated, restore
6168
* the saved block list otherwise free it if it is no
6169
* longer needed.
6170
*/
6171
if (TAILQ_EMPTY(&indirdep->ir_trunc)) {
6172
if (bp == NULL)
6173
bcopy(indirdep->ir_saveddata,
6174
indirdep->ir_savebp->b_data,
6175
indirdep->ir_savebp->b_bcount);
6176
free(indirdep->ir_saveddata, M_INDIRDEP);
6177
indirdep->ir_saveddata = NULL;
6178
}
6179
/*
6180
* When bp is NULL there is a full truncation pending. We
6181
* must wait for this full truncation to be journaled before
6182
* we can release this freework because the disk pointers will
6183
* never be written as zero.
6184
*/
6185
if (bp == NULL) {
6186
if (LIST_EMPTY(&indirdep->ir_freeblks->fb_jblkdephd))
6187
handle_written_freework(freework);
6188
else
6189
WORKLIST_INSERT(&indirdep->ir_freeblks->fb_freeworkhd,
6190
&freework->fw_list);
6191
} else {
6192
/* Complete when the real copy is written. */
6193
WORKLIST_INSERT(&bp->b_dep, &freework->fw_list);
6194
BUF_UNLOCK(bp);
6195
}
6196
}
6197
6198
/*
6199
* Calculate the number of blocks we are going to release where datablocks
6200
* is the current total and length is the new file size.
6201
*/
6202
ufs2_daddr_t
6203
blkcount(fs, datablocks, length)
6204
struct fs *fs;
6205
ufs2_daddr_t datablocks;
6206
off_t length;
6207
{
6208
off_t totblks, numblks;
6209
6210
totblks = 0;
6211
numblks = howmany(length, fs->fs_bsize);
6212
if (numblks <= NDADDR) {
6213
totblks = howmany(length, fs->fs_fsize);
6214
goto out;
6215
}
6216
totblks = blkstofrags(fs, numblks);
6217
numblks -= NDADDR;
6218
/*
6219
* Count all single, then double, then triple indirects required.
6220
* Subtracting one indirects worth of blocks for each pass
6221
* acknowledges one of each pointed to by the inode.
6222
*/
6223
for (;;) {
6224
totblks += blkstofrags(fs, howmany(numblks, NINDIR(fs)));
6225
numblks -= NINDIR(fs);
6226
if (numblks <= 0)
6227
break;
6228
numblks = howmany(numblks, NINDIR(fs));
6229
}
6230
out:
6231
totblks = fsbtodb(fs, totblks);
6232
/*
6233
* Handle sparse files. We can't reclaim more blocks than the inode
6234
* references. We will correct it later in handle_complete_freeblks()
6235
* when we know the real count.
6236
*/
6237
if (totblks > datablocks)
6238
return (0);
6239
return (datablocks - totblks);
6240
}
6241
6242
/*
6243
* Handle freeblocks for journaled softupdate filesystems.
6244
*
6245
* Contrary to normal softupdates, we must preserve the block pointers in
6246
* indirects until their subordinates are free. This is to avoid journaling
6247
* every block that is freed which may consume more space than the journal
6248
* itself. The recovery program will see the free block journals at the
6249
* base of the truncated area and traverse them to reclaim space. The
6250
* pointers in the inode may be cleared immediately after the journal
6251
* records are written because each direct and indirect pointer in the
6252
* inode is recorded in a journal. This permits full truncation to proceed
6253
* asynchronously. The write order is journal -> inode -> cgs -> indirects.
6254
*
6255
* The algorithm is as follows:
6256
* 1) Traverse the in-memory state and create journal entries to release
6257
* the relevant blocks and full indirect trees.
6258
* 2) Traverse the indirect block chain adding partial truncation freework
6259
* records to indirects in the path to lastlbn. The freework will
6260
* prevent new allocation dependencies from being satisfied in this
6261
* indirect until the truncation completes.
6262
* 3) Read and lock the inode block, performing an update with the new size
6263
* and pointers. This prevents truncated data from becoming valid on
6264
* disk through step 4.
6265
* 4) Reap unsatisfied dependencies that are beyond the truncated area,
6266
* eliminate journal work for those records that do not require it.
6267
* 5) Schedule the journal records to be written followed by the inode block.
6268
* 6) Allocate any necessary frags for the end of file.
6269
* 7) Zero any partially truncated blocks.
6270
*
6271
* From this truncation proceeds asynchronously using the freework and
6272
* indir_trunc machinery. The file will not be extended again into a
6273
* partially truncated indirect block until all work is completed but
6274
* the normal dependency mechanism ensures that it is rolled back/forward
6275
* as appropriate. Further truncation may occur without delay and is
6276
* serialized in indir_trunc().
6277
*/
6278
void
6279
softdep_journal_freeblocks(ip, cred, length, flags)
6280
struct inode *ip; /* The inode whose length is to be reduced */
6281
struct ucred *cred;
6282
off_t length; /* The new length for the file */
6283
int flags; /* IO_EXT and/or IO_NORMAL */
6284
{
6285
struct freeblks *freeblks, *fbn;
6286
struct worklist *wk, *wkn;
6287
struct inodedep *inodedep;
6288
struct jblkdep *jblkdep;
6289
struct allocdirect *adp, *adpn;
6290
struct fs *fs;
6291
struct buf *bp;
6292
struct vnode *vp;
6293
struct mount *mp;
6294
ufs2_daddr_t extblocks, datablocks;
6295
ufs_lbn_t tmpval, lbn, lastlbn;
6296
int frags, lastoff, iboff, allocblock, needj, dflags, error, i;
6297
6298
fs = ip->i_fs;
6299
mp = UFSTOVFS(ip->i_ump);
6300
vp = ITOV(ip);
6301
needj = 1;
6302
iboff = -1;
6303
allocblock = 0;
6304
extblocks = 0;
6305
datablocks = 0;
6306
frags = 0;
6307
freeblks = newfreeblks(mp, ip);
6308
ACQUIRE_LOCK(&lk);
6309
/*
6310
* If we're truncating a removed file that will never be written
6311
* we don't need to journal the block frees. The canceled journals
6312
* for the allocations will suffice.
6313
*/
6314
dflags = DEPALLOC;
6315
if (IS_SNAPSHOT(ip))
6316
dflags |= NODELAY;
6317
inodedep_lookup(mp, ip->i_number, dflags, &inodedep);
6318
if ((inodedep->id_state & (UNLINKED | DEPCOMPLETE)) == UNLINKED &&
6319
length == 0)
6320
needj = 0;
6321
CTR3(KTR_SUJ, "softdep_journal_freeblks: ip %d length %ld needj %d",
6322
ip->i_number, length, needj);
6323
FREE_LOCK(&lk);
6324
/*
6325
* Calculate the lbn that we are truncating to. This results in -1
6326
* if we're truncating the 0 bytes. So it is the last lbn we want
6327
* to keep, not the first lbn we want to truncate.
6328
*/
6329
lastlbn = lblkno(fs, length + fs->fs_bsize - 1) - 1;
6330
lastoff = blkoff(fs, length);
6331
/*
6332
* Compute frags we are keeping in lastlbn. 0 means all.
6333
*/
6334
if (lastlbn >= 0 && lastlbn < NDADDR) {
6335
frags = fragroundup(fs, lastoff);
6336
/* adp offset of last valid allocdirect. */
6337
iboff = lastlbn;
6338
} else if (lastlbn > 0)
6339
iboff = NDADDR;
6340
if (fs->fs_magic == FS_UFS2_MAGIC)
6341
extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
6342
/*
6343
* Handle normal data blocks and indirects. This section saves
6344
* values used after the inode update to complete frag and indirect
6345
* truncation.
6346
*/
6347
if ((flags & IO_NORMAL) != 0) {
6348
/*
6349
* Handle truncation of whole direct and indirect blocks.
6350
*/
6351
for (i = iboff + 1; i < NDADDR; i++)
6352
setup_freedirect(freeblks, ip, i, needj);
6353
for (i = 0, tmpval = NINDIR(fs), lbn = NDADDR; i < NIADDR;
6354
i++, lbn += tmpval, tmpval *= NINDIR(fs)) {
6355
/* Release a whole indirect tree. */
6356
if (lbn > lastlbn) {
6357
setup_freeindir(freeblks, ip, i, -lbn -i,
6358
needj);
6359
continue;
6360
}
6361
iboff = i + NDADDR;
6362
/*
6363
* Traverse partially truncated indirect tree.
6364
*/
6365
if (lbn <= lastlbn && lbn + tmpval - 1 > lastlbn)
6366
setup_trunc_indir(freeblks, ip, -lbn - i,
6367
lastlbn, DIP(ip, i_ib[i]));
6368
}
6369
/*
6370
* Handle partial truncation to a frag boundary.
6371
*/
6372
if (frags) {
6373
ufs2_daddr_t blkno;
6374
long oldfrags;
6375
6376
oldfrags = blksize(fs, ip, lastlbn);
6377
blkno = DIP(ip, i_db[lastlbn]);
6378
if (blkno && oldfrags != frags) {
6379
oldfrags -= frags;
6380
oldfrags = numfrags(ip->i_fs, oldfrags);
6381
blkno += numfrags(ip->i_fs, frags);
6382
newfreework(ip->i_ump, freeblks, NULL, lastlbn,
6383
blkno, oldfrags, 0, needj);
6384
} else if (blkno == 0)
6385
allocblock = 1;
6386
}
6387
/*
6388
* Add a journal record for partial truncate if we are
6389
* handling indirect blocks. Non-indirects need no extra
6390
* journaling.
6391
*/
6392
if (length != 0 && lastlbn >= NDADDR) {
6393
ip->i_flag |= IN_TRUNCATED;
6394
newjtrunc(freeblks, length, 0);
6395
}
6396
ip->i_size = length;
6397
DIP_SET(ip, i_size, ip->i_size);
6398
datablocks = DIP(ip, i_blocks) - extblocks;
6399
if (length != 0)
6400
datablocks = blkcount(ip->i_fs, datablocks, length);
6401
freeblks->fb_len = length;
6402
}
6403
if ((flags & IO_EXT) != 0) {
6404
for (i = 0; i < NXADDR; i++)
6405
setup_freeext(freeblks, ip, i, needj);
6406
ip->i_din2->di_extsize = 0;
6407
datablocks += extblocks;
6408
}
6409
#ifdef QUOTA
6410
/* Reference the quotas in case the block count is wrong in the end. */
6411
quotaref(vp, freeblks->fb_quota);
6412
(void) chkdq(ip, -datablocks, NOCRED, 0);
6413
#endif
6414
freeblks->fb_chkcnt = -datablocks;
6415
UFS_LOCK(ip->i_ump);
6416
fs->fs_pendingblocks += datablocks;
6417
UFS_UNLOCK(ip->i_ump);
6418
DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - datablocks);
6419
/*
6420
* Handle truncation of incomplete alloc direct dependencies. We
6421
* hold the inode block locked to prevent incomplete dependencies
6422
* from reaching the disk while we are eliminating those that
6423
* have been truncated. This is a partially inlined ffs_update().
6424
*/
6425
ufs_itimes(vp);
6426
ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED);
6427
error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
6428
(int)fs->fs_bsize, cred, &bp);
6429
if (error) {
6430
brelse(bp);
6431
softdep_error("softdep_journal_freeblocks", error);
6432
return;
6433
}
6434
if (bp->b_bufsize == fs->fs_bsize)
6435
bp->b_flags |= B_CLUSTEROK;
6436
softdep_update_inodeblock(ip, bp, 0);
6437
if (ip->i_ump->um_fstype == UFS1)
6438
*((struct ufs1_dinode *)bp->b_data +
6439
ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
6440
else
6441
*((struct ufs2_dinode *)bp->b_data +
6442
ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
6443
ACQUIRE_LOCK(&lk);
6444
(void) inodedep_lookup(mp, ip->i_number, dflags, &inodedep);
6445
if ((inodedep->id_state & IOSTARTED) != 0)
6446
panic("softdep_setup_freeblocks: inode busy");
6447
/*
6448
* Add the freeblks structure to the list of operations that
6449
* must await the zero'ed inode being written to disk. If we
6450
* still have a bitmap dependency (needj), then the inode
6451
* has never been written to disk, so we can process the
6452
* freeblks below once we have deleted the dependencies.
6453
*/
6454
if (needj)
6455
WORKLIST_INSERT(&bp->b_dep, &freeblks->fb_list);
6456
else
6457
freeblks->fb_state |= COMPLETE;
6458
if ((flags & IO_NORMAL) != 0) {
6459
TAILQ_FOREACH_SAFE(adp, &inodedep->id_inoupdt, ad_next, adpn) {
6460
if (adp->ad_offset > iboff)
6461
cancel_allocdirect(&inodedep->id_inoupdt, adp,
6462
freeblks);
6463
/*
6464
* Truncate the allocdirect. We could eliminate
6465
* or modify journal records as well.
6466
*/
6467
else if (adp->ad_offset == iboff && frags)
6468
adp->ad_newsize = frags;
6469
}
6470
}
6471
if ((flags & IO_EXT) != 0)
6472
while ((adp = TAILQ_FIRST(&inodedep->id_extupdt)) != 0)
6473
cancel_allocdirect(&inodedep->id_extupdt, adp,
6474
freeblks);
6475
/*
6476
* Scan the bufwait list for newblock dependencies that will never
6477
* make it to disk.
6478
*/
6479
LIST_FOREACH_SAFE(wk, &inodedep->id_bufwait, wk_list, wkn) {
6480
if (wk->wk_type != D_ALLOCDIRECT)
6481
continue;
6482
adp = WK_ALLOCDIRECT(wk);
6483
if (((flags & IO_NORMAL) != 0 && (adp->ad_offset > iboff)) ||
6484
((flags & IO_EXT) != 0 && (adp->ad_state & EXTDATA))) {
6485
cancel_jfreeblk(freeblks, adp->ad_newblkno);
6486
cancel_newblk(WK_NEWBLK(wk), NULL, &freeblks->fb_jwork);
6487
WORKLIST_INSERT(&freeblks->fb_freeworkhd, wk);
6488
}
6489
}
6490
/*
6491
* Add journal work.
6492
*/
6493
LIST_FOREACH(jblkdep, &freeblks->fb_jblkdephd, jb_deps)
6494
add_to_journal(&jblkdep->jb_list);
6495
FREE_LOCK(&lk);
6496
bdwrite(bp);
6497
/*
6498
* Truncate dependency structures beyond length.
6499
*/
6500
trunc_dependencies(ip, freeblks, lastlbn, frags, flags);
6501
/*
6502
* This is only set when we need to allocate a fragment because
6503
* none existed at the end of a frag-sized file. It handles only
6504
* allocating a new, zero filled block.
6505
*/
6506
if (allocblock) {
6507
ip->i_size = length - lastoff;
6508
DIP_SET(ip, i_size, ip->i_size);
6509
error = UFS_BALLOC(vp, length - 1, 1, cred, BA_CLRBUF, &bp);
6510
if (error != 0) {
6511
softdep_error("softdep_journal_freeblks", error);
6512
return;
6513
}
6514
ip->i_size = length;
6515
DIP_SET(ip, i_size, length);
6516
ip->i_flag |= IN_CHANGE | IN_UPDATE;
6517
allocbuf(bp, frags);
6518
ffs_update(vp, 0);
6519
bawrite(bp);
6520
} else if (lastoff != 0 && vp->v_type != VDIR) {
6521
int size;
6522
6523
/*
6524
* Zero the end of a truncated frag or block.
6525
*/
6526
size = sblksize(fs, length, lastlbn);
6527
error = bread(vp, lastlbn, size, cred, &bp);
6528
if (error) {
6529
softdep_error("softdep_journal_freeblks", error);
6530
return;
6531
}
6532
bzero((char *)bp->b_data + lastoff, size - lastoff);
6533
bawrite(bp);
6534
6535
}
6536
ACQUIRE_LOCK(&lk);
6537
inodedep_lookup(mp, ip->i_number, dflags, &inodedep);
6538
TAILQ_INSERT_TAIL(&inodedep->id_freeblklst, freeblks, fb_next);
6539
freeblks->fb_state |= DEPCOMPLETE | ONDEPLIST;
6540
/*
6541
* We zero earlier truncations so they don't erroneously
6542
* update i_blocks.
6543
*/
6544
if (freeblks->fb_len == 0 && (flags & IO_NORMAL) != 0)
6545
TAILQ_FOREACH(fbn, &inodedep->id_freeblklst, fb_next)
6546
fbn->fb_len = 0;
6547
if ((freeblks->fb_state & ALLCOMPLETE) == ALLCOMPLETE &&
6548
LIST_EMPTY(&freeblks->fb_jblkdephd))
6549
freeblks->fb_state |= INPROGRESS;
6550
else
6551
freeblks = NULL;
6552
FREE_LOCK(&lk);
6553
if (freeblks)
6554
handle_workitem_freeblocks(freeblks, 0);
6555
trunc_pages(ip, length, extblocks, flags);
6556
6557
}
6558
6559
/*
6560
* Flush a JOP_SYNC to the journal.
6561
*/
6562
void
6563
softdep_journal_fsync(ip)
6564
struct inode *ip;
6565
{
6566
struct jfsync *jfsync;
6567
6568
if ((ip->i_flag & IN_TRUNCATED) == 0)
6569
return;
6570
ip->i_flag &= ~IN_TRUNCATED;
6571
jfsync = malloc(sizeof(*jfsync), M_JFSYNC, M_SOFTDEP_FLAGS | M_ZERO);
6572
workitem_alloc(&jfsync->jfs_list, D_JFSYNC, UFSTOVFS(ip->i_ump));
6573
jfsync->jfs_size = ip->i_size;
6574
jfsync->jfs_ino = ip->i_number;
6575
ACQUIRE_LOCK(&lk);
6576
add_to_journal(&jfsync->jfs_list);
6577
jwait(&jfsync->jfs_list, MNT_WAIT);
6578
FREE_LOCK(&lk);
2169
6579
}
2170
6580
2171
6581
/*
2203
6613
off_t length; /* The new length for the file */
2204
6614
int flags; /* IO_EXT and/or IO_NORMAL */
2205
6615
{
6616
struct ufs1_dinode *dp1;
6617
struct ufs2_dinode *dp2;
2206
6618
struct freeblks *freeblks;
2207
6619
struct inodedep *inodedep;
2208
6620
struct allocdirect *adp;
2209
struct bufobj *bo;
2210
struct vnode *vp;
2211
6621
struct buf *bp;
2212
6622
struct fs *fs;
2213
6623
ufs2_daddr_t extblocks, datablocks;
2214
6624
struct mount *mp;
2215
int i, delay, error;
6625
int i, delay, error, dflags;
6626
ufs_lbn_t tmpval;
6627
ufs_lbn_t lbn;
2216
6628
6629
CTR2(KTR_SUJ, "softdep_setup_freeblks: ip %d length %ld",
6630
ip->i_number, length);
2217
6631
fs = ip->i_fs;
2218
6632
mp = UFSTOVFS(ip->i_ump);
2219
6633
if (length != 0)
2220
6634
panic("softdep_setup_freeblocks: non-zero length");
2221
freeblks = malloc(sizeof(struct freeblks),
2222
M_FREEBLKS, M_SOFTDEP_FLAGS|M_ZERO);
2223
workitem_alloc(&freeblks->fb_list, D_FREEBLKS, mp);
2224
freeblks->fb_state = ATTACHED;
2225
freeblks->fb_uid = ip->i_uid;
2226
freeblks->fb_previousinum = ip->i_number;
2227
freeblks->fb_devvp = ip->i_devvp;
2228
ACQUIRE_LOCK(&lk);
2229
num_freeblkdep++;
2230
FREE_LOCK(&lk);
6635
freeblks = newfreeblks(mp, ip);
2231
6636
extblocks = 0;
6637
datablocks = 0;
2232
6638
if (fs->fs_magic == FS_UFS2_MAGIC)
2233
6639
extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
2234
datablocks = DIP(ip, i_blocks) - extblocks;
2235
if ((flags & IO_NORMAL) == 0) {
2236
freeblks->fb_oldsize = 0;
2237
freeblks->fb_chkcnt = 0;
2238
} else {
2239
freeblks->fb_oldsize = ip->i_size;
6640
if ((flags & IO_NORMAL) != 0) {
6641
for (i = 0; i < NDADDR; i++)
6642
setup_freedirect(freeblks, ip, i, 0);
6643
for (i = 0, tmpval = NINDIR(fs), lbn = NDADDR; i < NIADDR;
6644
i++, lbn += tmpval, tmpval *= NINDIR(fs))
6645
setup_freeindir(freeblks, ip, i, -lbn -i, 0);
2240
6646
ip->i_size = 0;
2241
6647
DIP_SET(ip, i_size, 0);
2242
freeblks->fb_chkcnt = datablocks;
2243
for (i = 0; i < NDADDR; i++) {
2244
freeblks->fb_dblks[i] = DIP(ip, i_db[i]);
2245
DIP_SET(ip, i_db[i], 0);
2246
}
2247
for (i = 0; i < NIADDR; i++) {
2248
freeblks->fb_iblks[i] = DIP(ip, i_ib[i]);
2249
DIP_SET(ip, i_ib[i], 0);
2250
}
2251
/*
2252
* If the file was removed, then the space being freed was
2253
* accounted for then (see softdep_releasefile()). If the
2254
* file is merely being truncated, then we account for it now.
2255
*/
2256
if ((ip->i_flag & IN_SPACECOUNTED) == 0) {
2257
UFS_LOCK(ip->i_ump);
2258
fs->fs_pendingblocks += datablocks;
2259
UFS_UNLOCK(ip->i_ump);
2260
}
6648
datablocks = DIP(ip, i_blocks) - extblocks;
2261
6649
}
2262
if ((flags & IO_EXT) == 0) {
2263
freeblks->fb_oldextsize = 0;
2264
} else {
2265
freeblks->fb_oldextsize = ip->i_din2->di_extsize;
6650
if ((flags & IO_EXT) != 0) {
6651
for (i = 0; i < NXADDR; i++)
6652
setup_freeext(freeblks, ip, i, 0);
2266
6653
ip->i_din2->di_extsize = 0;
2267
freeblks->fb_chkcnt += extblocks;
2268
for (i = 0; i < NXADDR; i++) {
2269
freeblks->fb_eblks[i] = ip->i_din2->di_extb[i];
2270
ip->i_din2->di_extb[i] = 0;
2271
}
6654
datablocks += extblocks;
2272
6655
}
2273
DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - freeblks->fb_chkcnt);
6656
#ifdef QUOTA
6657
/* Reference the quotas in case the block count is wrong in the end. */
6658
quotaref(ITOV(ip), freeblks->fb_quota);
6659
(void) chkdq(ip, -datablocks, NOCRED, 0);
6660
#endif
6661
freeblks->fb_chkcnt = -datablocks;
6662
UFS_LOCK(ip->i_ump);
6663
fs->fs_pendingblocks += datablocks;
6664
UFS_UNLOCK(ip->i_ump);
6665
DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - datablocks);
2274
6666
/*
2275
6667
* Push the zero'ed inode to to its disk buffer so that we are free
2276
6668
* to delete its dependencies below. Once the dependencies are gone
2282
6674
brelse(bp);
2283
6675
softdep_error("softdep_setup_freeblocks", error);
2284
6676
}
2285
if (ip->i_ump->um_fstype == UFS1)
2286
*((struct ufs1_dinode *)bp->b_data +
2287
ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
2288
else
2289
*((struct ufs2_dinode *)bp->b_data +
2290
ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
6677
if (ip->i_ump->um_fstype == UFS1) {
6678
dp1 = ((struct ufs1_dinode *)bp->b_data +
6679
ino_to_fsbo(fs, ip->i_number));
6680
ip->i_din1->di_freelink = dp1->di_freelink;
6681
*dp1 = *ip->i_din1;
6682
} else {
6683
dp2 = ((struct ufs2_dinode *)bp->b_data +
6684
ino_to_fsbo(fs, ip->i_number));
6685
ip->i_din2->di_freelink = dp2->di_freelink;
6686
*dp2 = *ip->i_din2;
6687
}
2291
6688
/*
2292
6689
* Find and eliminate any inode dependencies.
2293
6690
*/
2294
6691
ACQUIRE_LOCK(&lk);
2295
(void) inodedep_lookup(mp, ip->i_number, DEPALLOC, &inodedep);
6692
dflags = DEPALLOC;
6693
if (IS_SNAPSHOT(ip))
6694
dflags |= NODELAY;
6695
(void) inodedep_lookup(mp, ip->i_number, dflags, &inodedep);
2296
6696
if ((inodedep->id_state & IOSTARTED) != 0)
2297
6697
panic("softdep_setup_freeblocks: inode busy");
2298
6698
/*
2304
6704
*/
2305
6705
delay = (inodedep->id_state & DEPCOMPLETE);
2306
6706
if (delay)
2307
WORKLIST_INSERT(&inodedep->id_bufwait, &freeblks->fb_list);
6707
WORKLIST_INSERT(&bp->b_dep, &freeblks->fb_list);
6708
else
6709
freeblks->fb_state |= COMPLETE;
2308
6710
/*
2309
6711
* Because the file length has been truncated to zero, any
2310
6712
* pending block allocation dependency structures associated
2318
6720
merge_inode_lists(&inodedep->id_newinoupdt,
2319
6721
&inodedep->id_inoupdt);
2320
6722
while ((adp = TAILQ_FIRST(&inodedep->id_inoupdt)) != 0)
2321
free_allocdirect(&inodedep->id_inoupdt, adp, delay);
6723
cancel_allocdirect(&inodedep->id_inoupdt, adp,
6724
freeblks);
2322
6725
}
2323
6726
if (flags & IO_EXT) {
2324
6727
merge_inode_lists(&inodedep->id_newextupdt,
2325
6728
&inodedep->id_extupdt);
2326
6729
while ((adp = TAILQ_FIRST(&inodedep->id_extupdt)) != 0)
2327
free_allocdirect(&inodedep->id_extupdt, adp, delay);
6730
cancel_allocdirect(&inodedep->id_extupdt, adp,
6731
freeblks);
2328
6732
}
2329
6733
FREE_LOCK(&lk);
2330
6734
bdwrite(bp);
6735
trunc_dependencies(ip, freeblks, -1, 0, flags);
6736
ACQUIRE_LOCK(&lk);
6737
if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) != 0)
6738
(void) free_inodedep(inodedep);
6739
freeblks->fb_state |= DEPCOMPLETE;
6740
/*
6741
* If the inode with zeroed block pointers is now on disk
6742
* we can start freeing blocks.
6743
*/
6744
if ((freeblks->fb_state & ALLCOMPLETE) == ALLCOMPLETE)
6745
freeblks->fb_state |= INPROGRESS;
6746
else
6747
freeblks = NULL;
6748
FREE_LOCK(&lk);
6749
if (freeblks)
6750
handle_workitem_freeblocks(freeblks, 0);
6751
trunc_pages(ip, length, extblocks, flags);
6752
}
6753
6754
/*
6755
* Eliminate pages from the page cache that back parts of this inode and
6756
* adjust the vnode pager's idea of our size. This prevents stale data
6757
* from hanging around in the page cache.
6758
*/
6759
static void
6760
trunc_pages(ip, length, extblocks, flags)
6761
struct inode *ip;
6762
off_t length;
6763
ufs2_daddr_t extblocks;
6764
int flags;
6765
{
6766
struct vnode *vp;
6767
struct fs *fs;
6768
ufs_lbn_t lbn;
6769
off_t end, extend;
6770
6771
vp = ITOV(ip);
6772
fs = ip->i_fs;
6773
extend = OFF_TO_IDX(lblktosize(fs, -extblocks));
6774
if ((flags & IO_EXT) != 0)
6775
vn_pages_remove(vp, extend, 0);
6776
if ((flags & IO_NORMAL) == 0)
6777
return;
6778
BO_LOCK(&vp->v_bufobj);
6779
drain_output(vp);
6780
BO_UNLOCK(&vp->v_bufobj);
6781
/*
6782
* The vnode pager eliminates file pages we eliminate indirects
6783
* below.
6784
*/
6785
vnode_pager_setsize(vp, length);
6786
/*
6787
* Calculate the end based on the last indirect we want to keep. If
6788
* the block extends into indirects we can just use the negative of
6789
* its lbn. Doubles and triples exist at lower numbers so we must
6790
* be careful not to remove those, if they exist. double and triple
6791
* indirect lbns do not overlap with others so it is not important
6792
* to verify how many levels are required.
6793
*/
6794
lbn = lblkno(fs, length);
6795
if (lbn >= NDADDR) {
6796
/* Calculate the virtual lbn of the triple indirect. */
6797
lbn = -lbn - (NIADDR - 1);
6798
end = OFF_TO_IDX(lblktosize(fs, lbn));
6799
} else
6800
end = extend;
6801
vn_pages_remove(vp, OFF_TO_IDX(OFF_MAX), end);
6802
}
6803
6804
/*
6805
* See if the buf bp is in the range eliminated by truncation.
6806
*/
6807
static int
6808
trunc_check_buf(bp, blkoffp, lastlbn, lastoff, flags)
6809
struct buf *bp;
6810
int *blkoffp;
6811
ufs_lbn_t lastlbn;
6812
int lastoff;
6813
int flags;
6814
{
6815
ufs_lbn_t lbn;
6816
6817
*blkoffp = 0;
6818
/* Only match ext/normal blocks as appropriate. */
6819
if (((flags & IO_EXT) == 0 && (bp->b_xflags & BX_ALTDATA)) ||
6820
((flags & IO_NORMAL) == 0 && (bp->b_xflags & BX_ALTDATA) == 0))
6821
return (0);
6822
/* ALTDATA is always a full truncation. */
6823
if ((bp->b_xflags & BX_ALTDATA) != 0)
6824
return (1);
6825
/* -1 is full truncation. */
6826
if (lastlbn == -1)
6827
return (1);
6828
/*
6829
* If this is a partial truncate we only want those
6830
* blocks and indirect blocks that cover the range
6831
* we're after.
6832
*/
6833
lbn = bp->b_lblkno;
6834
if (lbn < 0)
6835
lbn = -(lbn + lbn_level(lbn));
6836
if (lbn < lastlbn)
6837
return (0);
6838
/* Here we only truncate lblkno if it's partial. */
6839
if (lbn == lastlbn) {
6840
if (lastoff == 0)
6841
return (0);
6842
*blkoffp = lastoff;
6843
}
6844
return (1);
6845
}
6846
6847
/*
6848
* Eliminate any dependencies that exist in memory beyond lblkno:off
6849
*/
6850
static void
6851
trunc_dependencies(ip, freeblks, lastlbn, lastoff, flags)
6852
struct inode *ip;
6853
struct freeblks *freeblks;
6854
ufs_lbn_t lastlbn;
6855
int lastoff;
6856
int flags;
6857
{
6858
struct bufobj *bo;
6859
struct vnode *vp;
6860
struct buf *bp;
6861
struct fs *fs;
6862
int blkoff;
6863
2331
6864
/*
2332
6865
* We must wait for any I/O in progress to finish so that
2333
6866
* all potential buffers on the dirty list will be visible.
2334
6867
* Once they are all there, walk the list and get rid of
2335
6868
* any dependencies.
2336
6869
*/
6870
fs = ip->i_fs;
2337
6871
vp = ITOV(ip);
2338
6872
bo = &vp->v_bufobj;
2339
6873
BO_LOCK(bo);
2340
6874
drain_output(vp);
6875
TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
6876
bp->b_vflags &= ~BV_SCANNED;
2341
6877
restart:
2342
6878
TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
2343
if (((flags & IO_EXT) == 0 && (bp->b_xflags & BX_ALTDATA)) ||
2344
((flags & IO_NORMAL) == 0 &&
2345
(bp->b_xflags & BX_ALTDATA) == 0))
2346
continue;
6879
if (bp->b_vflags & BV_SCANNED)
6880
continue;
6881
if (!trunc_check_buf(bp, &blkoff, lastlbn, lastoff, flags)) {
6882
bp->b_vflags |= BV_SCANNED;
6883
continue;
6884
}
2347
6885
if ((bp = getdirtybuf(bp, BO_MTX(bo), MNT_WAIT)) == NULL)
2348
6886
goto restart;
2349
6887
BO_UNLOCK(bo);
2350
ACQUIRE_LOCK(&lk);
2351
(void) inodedep_lookup(mp, ip->i_number, 0, &inodedep);
2352
deallocate_dependencies(bp, inodedep);
2353
FREE_LOCK(&lk);
2354
bp->b_flags |= B_INVAL | B_NOCACHE;
2355
brelse(bp);
6888
if (deallocate_dependencies(bp, freeblks, blkoff))
6889
bqrelse(bp);
6890
else
6891
brelse(bp);
2356
6892
BO_LOCK(bo);
2357
6893
goto restart;
2358
6894
}
6895
/*
6896
* Now do the work of vtruncbuf while also matching indirect blocks.
6897
*/
6898
TAILQ_FOREACH(bp, &bo->bo_clean.bv_hd, b_bobufs)
6899
bp->b_vflags &= ~BV_SCANNED;
6900
cleanrestart:
6901
TAILQ_FOREACH(bp, &bo->bo_clean.bv_hd, b_bobufs) {
6902
if (bp->b_vflags & BV_SCANNED)
6903
continue;
6904
if (!trunc_check_buf(bp, &blkoff, lastlbn, lastoff, flags)) {
6905
bp->b_vflags |= BV_SCANNED;
6906
continue;
6907
}
6908
if (BUF_LOCK(bp,
6909
LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
6910
BO_MTX(bo)) == ENOLCK) {
6911
BO_LOCK(bo);
6912
goto cleanrestart;
6913
}
6914
bp->b_vflags |= BV_SCANNED;
6915
BO_LOCK(bo);
6916
bremfree(bp);
6917
BO_UNLOCK(bo);
6918
if (blkoff != 0) {
6919
allocbuf(bp, blkoff);
6920
bqrelse(bp);
6921
} else {
6922
bp->b_flags |= B_INVAL | B_NOCACHE | B_RELBUF;
6923
brelse(bp);
6924
}
6925
BO_LOCK(bo);
6926
goto cleanrestart;
6927
}
6928
drain_output(vp);
2359
6929
BO_UNLOCK(bo);
2360
ACQUIRE_LOCK(&lk);
2361
if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) != 0)
2362
(void) free_inodedep(inodedep);
2363
2364
if(delay) {
2365
freeblks->fb_state |= DEPCOMPLETE;
6930
}
6931
6932
static int
6933
cancel_pagedep(pagedep, freeblks, blkoff)
6934
struct pagedep *pagedep;
6935
struct freeblks *freeblks;
6936
int blkoff;
6937
{
6938
struct jremref *jremref;
6939
struct jmvref *jmvref;
6940
struct dirrem *dirrem, *tmp;
6941
int i;
6942
6943
/*
6944
* Copy any directory remove dependencies to the list
6945
* to be processed after the freeblks proceeds. If
6946
* directory entry never made it to disk they
6947
* can be dumped directly onto the work list.
6948
*/
6949
LIST_FOREACH_SAFE(dirrem, &pagedep->pd_dirremhd, dm_next, tmp) {
6950
/* Skip this directory removal if it is intended to remain. */
6951
if (dirrem->dm_offset < blkoff)
6952
continue;
2366
6953
/*
2367
* If the inode with zeroed block pointers is now on disk
2368
* we can start freeing blocks. Add freeblks to the worklist
2369
* instead of calling handle_workitem_freeblocks directly as
2370
* it is more likely that additional IO is needed to complete
2371
* the request here than in the !delay case.
2372
*/
2373
if ((freeblks->fb_state & ALLCOMPLETE) == ALLCOMPLETE)
2374
add_to_worklist(&freeblks->fb_list);
2375
}
6954
* If there are any dirrems we wait for the journal write
6955
* to complete and then restart the buf scan as the lock
6956
* has been dropped.
6957
*/
6958
while ((jremref = LIST_FIRST(&dirrem->dm_jremrefhd)) != NULL) {
6959
jwait(&jremref->jr_list, MNT_WAIT);
6960
return (ERESTART);
6961
}
6962
LIST_REMOVE(dirrem, dm_next);
6963
dirrem->dm_dirinum = pagedep->pd_ino;
6964
WORKLIST_INSERT(&freeblks->fb_freeworkhd, &dirrem->dm_list);
6965
}
6966
while ((jmvref = LIST_FIRST(&pagedep->pd_jmvrefhd)) != NULL) {
6967
jwait(&jmvref->jm_list, MNT_WAIT);
6968
return (ERESTART);
6969
}
6970
/*
6971
* When we're partially truncating a pagedep we just want to flush
6972
* journal entries and return. There can not be any adds in the
6973
* truncated portion of the directory and newblk must remain if
6974
* part of the block remains.
6975
*/
6976
if (blkoff != 0) {
6977
struct diradd *dap;
2376
6978
2377
FREE_LOCK(&lk);
6979
LIST_FOREACH(dap, &pagedep->pd_pendinghd, da_pdlist)
6980
if (dap->da_offset > blkoff)
6981
panic("cancel_pagedep: diradd %p off %d > %d",
6982
dap, dap->da_offset, blkoff);
6983
for (i = 0; i < DAHASHSZ; i++)
6984
LIST_FOREACH(dap, &pagedep->pd_diraddhd[i], da_pdlist)
6985
if (dap->da_offset > blkoff)
6986
panic("cancel_pagedep: diradd %p off %d > %d",
6987
dap, dap->da_offset, blkoff);
6988
return (0);
6989
}
2378
6990
/*
2379
* If the inode has never been written to disk (delay == 0),
2380
* then we can process the freeblks now that we have deleted
2381
* the dependencies.
6991
* There should be no directory add dependencies present
6992
* as the directory could not be truncated until all
6993
* children were removed.
2382
6994
*/
2383
if (!delay)
2384
handle_workitem_freeblocks(freeblks, 0);
6995
KASSERT(LIST_FIRST(&pagedep->pd_pendinghd) == NULL,
6996
("deallocate_dependencies: pendinghd != NULL"));
6997
for (i = 0; i < DAHASHSZ; i++)
6998
KASSERT(LIST_FIRST(&pagedep->pd_diraddhd[i]) == NULL,
6999
("deallocate_dependencies: diraddhd != NULL"));
7000
if ((pagedep->pd_state & NEWBLOCK) != 0)
7001
free_newdirblk(pagedep->pd_newdirblk);
7002
if (free_pagedep(pagedep) == 0)
7003
panic("Failed to free pagedep %p", pagedep);
7004
return (0);
2385
7005
}
2386
7006
2387
7007
/*
2391
7011
* its associated dependencies. The mutex is held so that other I/O's
2392
7012
* associated with related dependencies do not occur.
2393
7013
*/
2394
static void
2395
deallocate_dependencies(bp, inodedep)
7014
static int
7015
deallocate_dependencies(bp, freeblks, off)
2396
7016
struct buf *bp;
2397
struct inodedep *inodedep;
7017
struct freeblks *freeblks;
7018
int off;
2398
7019
{
2399
struct worklist *wk;
2400
7020
struct indirdep *indirdep;
2401
struct allocindir *aip;
2402
7021
struct pagedep *pagedep;
2403
struct dirrem *dirrem;
2404
struct diradd *dap;
2405
int i;
7022
struct allocdirect *adp;
7023
struct worklist *wk, *wkn;
2406
7024
2407
mtx_assert(&lk, MA_OWNED);
2408
while ((wk = LIST_FIRST(&bp->b_dep)) != NULL) {
7025
ACQUIRE_LOCK(&lk);
7026
LIST_FOREACH_SAFE(wk, &bp->b_dep, wk_list, wkn) {
2409
7027
switch (wk->wk_type) {
2410
2411
7028
case D_INDIRDEP:
2412
7029
indirdep = WK_INDIRDEP(wk);
2413
/*
2414
* None of the indirect pointers will ever be visible,
2415
* so they can simply be tossed. GOINGAWAY ensures
2416
* that allocated pointers will be saved in the buffer
2417
* cache until they are freed. Note that they will
2418
* only be able to be found by their physical address
2419
* since the inode mapping the logical address will
2420
* be gone. The save buffer used for the safe copy
2421
* was allocated in setup_allocindir_phase2 using
2422
* the physical address so it could be used for this
2423
* purpose. Hence we swap the safe copy with the real
2424
* copy, allowing the safe copy to be freed and holding
2425
* on to the real copy for later use in indir_trunc.
2426
*/
2427
if (indirdep->ir_state & GOINGAWAY)
2428
panic("deallocate_dependencies: already gone");
2429
indirdep->ir_state |= GOINGAWAY;
2430
VFSTOUFS(bp->b_vp->v_mount)->um_numindirdeps += 1;
2431
while ((aip = LIST_FIRST(&indirdep->ir_deplisthd)) != 0)
2432
free_allocindir(aip, inodedep);
2433
7030
if (bp->b_lblkno >= 0 ||
2434
7031
bp->b_blkno != indirdep->ir_savebp->b_lblkno)
2435
7032
panic("deallocate_dependencies: not indir");
2436
bcopy(bp->b_data, indirdep->ir_savebp->b_data,
2437
bp->b_bcount);
2438
WORKLIST_REMOVE(wk);
2439
WORKLIST_INSERT(&indirdep->ir_savebp->b_dep, wk);
7033
cancel_indirdep(indirdep, bp, freeblks);
2440
7034
continue;
2441
7035
2442
7036
case D_PAGEDEP:
2443
7037
pagedep = WK_PAGEDEP(wk);
2444
/*
2445
* None of the directory additions will ever be
2446
* visible, so they can simply be tossed.
2447
*/
2448
for (i = 0; i < DAHASHSZ; i++)
2449
while ((dap =
2450
LIST_FIRST(&pagedep->pd_diraddhd[i])))
2451
free_diradd(dap);
2452
while ((dap = LIST_FIRST(&pagedep->pd_pendinghd)) != 0)
2453
free_diradd(dap);
2454
/*
2455
* Copy any directory remove dependencies to the list
2456
* to be processed after the zero'ed inode is written.
2457
* If the inode has already been written, then they
2458
* can be dumped directly onto the work list.
2459
*/
2460
LIST_FOREACH(dirrem, &pagedep->pd_dirremhd, dm_next) {
2461
LIST_REMOVE(dirrem, dm_next);
2462
dirrem->dm_dirinum = pagedep->pd_ino;
2463
if (inodedep == NULL ||
2464
(inodedep->id_state & ALLCOMPLETE) ==
2465
ALLCOMPLETE)
2466
add_to_worklist(&dirrem->dm_list);
2467
else
2468
WORKLIST_INSERT(&inodedep->id_bufwait,
2469
&dirrem->dm_list);
2470
}
2471
if ((pagedep->pd_state & NEWBLOCK) != 0) {
2472
LIST_FOREACH(wk, &inodedep->id_bufwait, wk_list)
2473
if (wk->wk_type == D_NEWDIRBLK &&
2474
WK_NEWDIRBLK(wk)->db_pagedep ==
2475
pagedep)
2476
break;
2477
if (wk != NULL) {
2478
WORKLIST_REMOVE(wk);
2479
free_newdirblk(WK_NEWDIRBLK(wk));
2480
} else
2481
panic("deallocate_dependencies: "
2482
"lost pagedep");
2483
}
2484
WORKLIST_REMOVE(&pagedep->pd_list);
2485
LIST_REMOVE(pagedep, pd_hash);
2486
WORKITEM_FREE(pagedep, D_PAGEDEP);
7038
if (cancel_pagedep(pagedep, freeblks, off)) {
7039
FREE_LOCK(&lk);
7040
return (ERESTART);
7041
}
2487
7042
continue;
2488
7043
2489
7044
case D_ALLOCINDIR:
2490
free_allocindir(WK_ALLOCINDIR(wk), inodedep);
7045
/*
7046
* Simply remove the allocindir, we'll find it via
7047
* the indirdep where we can clear pointers if
7048
* needed.
7049
*/
7050
WORKLIST_REMOVE(wk);
2491
7051
continue;
2492
7052
7053
case D_FREEWORK:
7054
/*
7055
* A truncation is waiting for the zero'd pointers
7056
* to be written. It can be freed when the freeblks
7057
* is journaled.
7058
*/
7059
WORKLIST_REMOVE(wk);
7060
wk->wk_state |= ONDEPLIST;
7061
WORKLIST_INSERT(&freeblks->fb_freeworkhd, wk);
7062
break;
7063
2493
7064
case D_ALLOCDIRECT:
2494
case D_INODEDEP:
7065
adp = WK_ALLOCDIRECT(wk);
7066
if (off != 0)
7067
continue;
7068
/* FALLTHROUGH */
7069
default:
2495
7070
panic("deallocate_dependencies: Unexpected type %s",
2496
7071
TYPENAME(wk->wk_type));
2497
7072
/* NOTREACHED */
2498
2499
default:
2500
panic("deallocate_dependencies: Unknown type %s",
2501
TYPENAME(wk->wk_type));
2502
/* NOTREACHED */
2503
7073
}
2504
7074
}
7075
FREE_LOCK(&lk);
7076
/*
7077
* Don't throw away this buf, we were partially truncating and
7078
* some deps may always remain.
7079
*/
7080
if (off) {
7081
allocbuf(bp, off);
7082
bp->b_vflags |= BV_SCANNED;
7083
return (EBUSY);
7084
}
7085
bp->b_flags |= B_INVAL | B_NOCACHE;
7086
7087
return (0);
2505
7088
}
2506
7089
2507
7090
/*
2508
* Free an allocdirect. Generate a new freefrag work request if appropriate.
2509
* This routine must be called with splbio interrupts blocked.
7091
* An allocdirect is being canceled due to a truncate. We must make sure
7092
* the journal entry is released in concert with the blkfree that releases
7093
* the storage. Completed journal entries must not be released until the
7094
* space is no longer pointed to by the inode or in the bitmap.
2510
7095
*/
2511
7096
static void
2512
free_allocdirect(adphead, adp, delay)
7097
cancel_allocdirect(adphead, adp, freeblks)
2513
7098
struct allocdirectlst *adphead;
2514
7099
struct allocdirect *adp;
2515
int delay;
7100
struct freeblks *freeblks;
2516
7101
{
2517
struct newdirblk *newdirblk;
7102
struct freework *freework;
7103
struct newblk *newblk;
2518
7104
struct worklist *wk;
2519
7105
2520
mtx_assert(&lk, MA_OWNED);
2521
if ((adp->ad_state & DEPCOMPLETE) == 0)
2522
LIST_REMOVE(adp, ad_deps);
2523
7106
TAILQ_REMOVE(adphead, adp, ad_next);
2524
if ((adp->ad_state & COMPLETE) == 0)
2525
WORKLIST_REMOVE(&adp->ad_list);
2526
if (adp->ad_freefrag != NULL) {
2527
if (delay)
2528
WORKLIST_INSERT(&adp->ad_inodedep->id_bufwait,
2529
&adp->ad_freefrag->ff_list);
2530
else
2531
add_to_worklist(&adp->ad_freefrag->ff_list);
2532
}
2533
if ((wk = LIST_FIRST(&adp->ad_newdirblk)) != NULL) {
2534
newdirblk = WK_NEWDIRBLK(wk);
2535
WORKLIST_REMOVE(&newdirblk->db_list);
2536
if (!LIST_EMPTY(&adp->ad_newdirblk))
2537
panic("free_allocdirect: extra newdirblk");
2538
if (delay)
2539
WORKLIST_INSERT(&adp->ad_inodedep->id_bufwait,
2540
&newdirblk->db_list);
2541
else
2542
free_newdirblk(newdirblk);
2543
}
2544
WORKITEM_FREE(adp, D_ALLOCDIRECT);
7107
newblk = (struct newblk *)adp;
7108
freework = NULL;
7109
/*
7110
* Find the correct freework structure.
7111
*/
7112
LIST_FOREACH(wk, &freeblks->fb_freeworkhd, wk_list) {
7113
if (wk->wk_type != D_FREEWORK)
7114
continue;
7115
freework = WK_FREEWORK(wk);
7116
if (freework->fw_blkno == newblk->nb_newblkno)
7117
break;
7118
}
7119
if (freework == NULL)
7120
panic("cancel_allocdirect: Freework not found");
7121
/*
7122
* If a newblk exists at all we still have the journal entry that
7123
* initiated the allocation so we do not need to journal the free.
7124
*/
7125
cancel_jfreeblk(freeblks, freework->fw_blkno);
7126
/*
7127
* If the journal hasn't been written the jnewblk must be passed
7128
* to the call to ffs_blkfree that reclaims the space. We accomplish
7129
* this by linking the journal dependency into the freework to be
7130
* freed when freework_freeblock() is called. If the journal has
7131
* been written we can simply reclaim the journal space when the
7132
* freeblks work is complete.
7133
*/
7134
freework->fw_jnewblk = cancel_newblk(newblk, &freework->fw_list,
7135
&freeblks->fb_jwork);
7136
WORKLIST_INSERT(&freeblks->fb_freeworkhd, &newblk->nb_list);
7137
}
7138
7139
7140
/*
7141
* Cancel a new block allocation. May be an indirect or direct block. We
7142
* remove it from various lists and return any journal record that needs to
7143
* be resolved by the caller.
7144
*
7145
* A special consideration is made for indirects which were never pointed
7146
* at on disk and will never be found once this block is released.
7147
*/
7148
static struct jnewblk *
7149
cancel_newblk(newblk, wk, wkhd)
7150
struct newblk *newblk;
7151
struct worklist *wk;
7152
struct workhead *wkhd;
7153
{
7154
struct jnewblk *jnewblk;
7155
7156
CTR1(KTR_SUJ, "cancel_newblk: blkno %jd", newblk->nb_newblkno);
7157
7158
newblk->nb_state |= GOINGAWAY;
7159
/*
7160
* Previously we traversed the completedhd on each indirdep
7161
* attached to this newblk to cancel them and gather journal
7162
* work. Since we need only the oldest journal segment and
7163
* the lowest point on the tree will always have the oldest
7164
* journal segment we are free to release the segments
7165
* of any subordinates and may leave the indirdep list to
7166
* indirdep_complete() when this newblk is freed.
7167
*/
7168
if (newblk->nb_state & ONDEPLIST) {
7169
newblk->nb_state &= ~ONDEPLIST;
7170
LIST_REMOVE(newblk, nb_deps);
7171
}
7172
if (newblk->nb_state & ONWORKLIST)
7173
WORKLIST_REMOVE(&newblk->nb_list);
7174
/*
7175
* If the journal entry hasn't been written we save a pointer to
7176
* the dependency that frees it until it is written or the
7177
* superseding operation completes.
7178
*/
7179
jnewblk = newblk->nb_jnewblk;
7180
if (jnewblk != NULL && wk != NULL) {
7181
newblk->nb_jnewblk = NULL;
7182
jnewblk->jn_dep = wk;
7183
}
7184
if (!LIST_EMPTY(&newblk->nb_jwork))
7185
jwork_move(wkhd, &newblk->nb_jwork);
7186
/*
7187
* When truncating we must free the newdirblk early to remove
7188
* the pagedep from the hash before returning.
7189
*/
7190
if ((wk = LIST_FIRST(&newblk->nb_newdirblk)) != NULL)
7191
free_newdirblk(WK_NEWDIRBLK(wk));
7192
if (!LIST_EMPTY(&newblk->nb_newdirblk))
7193
panic("cancel_newblk: extra newdirblk");
7194
7195
return (jnewblk);
7196
}
7197
7198
/*
7199
* Schedule the freefrag associated with a newblk to be released once
7200
* the pointers are written and the previous block is no longer needed.
7201
*/
7202
static void
7203
newblk_freefrag(newblk)
7204
struct newblk *newblk;
7205
{
7206
struct freefrag *freefrag;
7207
7208
if (newblk->nb_freefrag == NULL)
7209
return;
7210
freefrag = newblk->nb_freefrag;
7211
newblk->nb_freefrag = NULL;
7212
freefrag->ff_state |= COMPLETE;
7213
if ((freefrag->ff_state & ALLCOMPLETE) == ALLCOMPLETE)
7214
add_to_worklist(&freefrag->ff_list, 0);
7215
}
7216
7217
/*
7218
* Free a newblk. Generate a new freefrag work request if appropriate.
7219
* This must be called after the inode pointer and any direct block pointers
7220
* are valid or fully removed via truncate or frag extension.
7221
*/
7222
static void
7223
free_newblk(newblk)
7224
struct newblk *newblk;
7225
{
7226
struct indirdep *indirdep;
7227
struct worklist *wk;
7228
7229
KASSERT(newblk->nb_jnewblk == NULL,
7230
("free_newblk; jnewblk %p still attached", newblk->nb_jnewblk));
7231
mtx_assert(&lk, MA_OWNED);
7232
newblk_freefrag(newblk);
7233
if (newblk->nb_state & ONDEPLIST)
7234
LIST_REMOVE(newblk, nb_deps);
7235
if (newblk->nb_state & ONWORKLIST)
7236
WORKLIST_REMOVE(&newblk->nb_list);
7237
LIST_REMOVE(newblk, nb_hash);
7238
if ((wk = LIST_FIRST(&newblk->nb_newdirblk)) != NULL)
7239
free_newdirblk(WK_NEWDIRBLK(wk));
7240
if (!LIST_EMPTY(&newblk->nb_newdirblk))
7241
panic("free_newblk: extra newdirblk");
7242
while ((indirdep = LIST_FIRST(&newblk->nb_indirdeps)) != NULL)
7243
indirdep_complete(indirdep);
7244
handle_jwork(&newblk->nb_jwork);
7245
newblk->nb_list.wk_type = D_NEWBLK;
7246
WORKITEM_FREE(newblk, D_NEWBLK);
2545
7247
}
2546
7248
2547
7249
/*
2554
7256
{
2555
7257
struct pagedep *pagedep;
2556
7258
struct diradd *dap;
2557
int i;
7259
struct worklist *wk;
2558
7260
2559
7261
mtx_assert(&lk, MA_OWNED);
7262
WORKLIST_REMOVE(&newdirblk->db_list);
2560
7263
/*
2561
7264
* If the pagedep is still linked onto the directory buffer
2562
7265
* dependency chain, then some of the entries on the
2569
7272
*/
2570
7273
pagedep = newdirblk->db_pagedep;
2571
7274
pagedep->pd_state &= ~NEWBLOCK;
2572
if ((pagedep->pd_state & ONWORKLIST) == 0)
7275
if ((pagedep->pd_state & ONWORKLIST) == 0) {
2573
7276
while ((dap = LIST_FIRST(&pagedep->pd_pendinghd)) != NULL)
2574
free_diradd(dap);
2575
/*
2576
* If no dependencies remain, the pagedep will be freed.
2577
*/
2578
for (i = 0; i < DAHASHSZ; i++)
2579
if (!LIST_EMPTY(&pagedep->pd_diraddhd[i]))
2580
break;
2581
if (i == DAHASHSZ && (pagedep->pd_state & ONWORKLIST) == 0) {
2582
LIST_REMOVE(pagedep, pd_hash);
2583
WORKITEM_FREE(pagedep, D_PAGEDEP);
7277
free_diradd(dap, NULL);
7278
/*
7279
* If no dependencies remain, the pagedep will be freed.
7280
*/
7281
free_pagedep(pagedep);
7282
}
7283
/* Should only ever be one item in the list. */
7284
while ((wk = LIST_FIRST(&newdirblk->db_mkdir)) != NULL) {
7285
WORKLIST_REMOVE(wk);
7286
handle_written_mkdir(WK_MKDIR(wk), MKDIR_BODY);
2584
7287
}
2585
7288
WORKITEM_FREE(newdirblk, D_NEWDIRBLK);
2586
7289
}
2598
7301
struct inode *ip = VTOI(pvp);
2599
7302
struct inodedep *inodedep;
2600
7303
struct freefile *freefile;
7304
struct freeblks *freeblks;
2601
7305
2602
7306
/*
2603
7307
* This sets up the inode de-allocation dependency.
2608
7312
freefile->fx_mode = mode;
2609
7313
freefile->fx_oldinum = ino;
2610
7314
freefile->fx_devvp = ip->i_devvp;
2611
if ((ip->i_flag & IN_SPACECOUNTED) == 0) {
2612
UFS_LOCK(ip->i_ump);
2613
ip->i_fs->fs_pendinginodes += 1;
2614
UFS_UNLOCK(ip->i_ump);
2615
}
7315
LIST_INIT(&freefile->fx_jwork);
7316
UFS_LOCK(ip->i_ump);
7317
ip->i_fs->fs_pendinginodes += 1;
7318
UFS_UNLOCK(ip->i_ump);
2616
7319
2617
7320
/*
2618
7321
* If the inodedep does not exist, then the zero'ed inode has
2619
7322
* been written to disk. If the allocated inode has never been
2620
7323
* written to disk, then the on-disk inode is zero'ed. In either
2621
* case we can free the file immediately.
7324
* case we can free the file immediately. If the journal was
7325
* canceled before being written the inode will never make it to
7326
* disk and we must send the canceled journal entrys to
7327
* ffs_freefile() to be cleared in conjunction with the bitmap.
7328
* Any blocks waiting on the inode to write can be safely freed
7329
* here as it will never been written.
2622
7330
*/
2623
7331
ACQUIRE_LOCK(&lk);
2624
if (inodedep_lookup(pvp->v_mount, ino, 0, &inodedep) == 0 ||
2625
check_inode_unwritten(inodedep)) {
7332
inodedep_lookup(pvp->v_mount, ino, 0, &inodedep);
7333
if (inodedep) {
7334
/*
7335
* Clear out freeblks that no longer need to reference
7336
* this inode.
7337
*/
7338
while ((freeblks =
7339
TAILQ_FIRST(&inodedep->id_freeblklst)) != NULL) {
7340
TAILQ_REMOVE(&inodedep->id_freeblklst, freeblks,
7341
fb_next);
7342
freeblks->fb_state &= ~ONDEPLIST;
7343
}
7344
/*
7345
* Remove this inode from the unlinked list.
7346
*/
7347
if (inodedep->id_state & UNLINKED) {
7348
/*
7349
* Save the journal work to be freed with the bitmap
7350
* before we clear UNLINKED. Otherwise it can be lost
7351
* if the inode block is written.
7352
*/
7353
handle_bufwait(inodedep, &freefile->fx_jwork);
7354
clear_unlinked_inodedep(inodedep);
7355
/* Re-acquire inodedep as we've dropped lk. */
7356
inodedep_lookup(pvp->v_mount, ino, 0, &inodedep);
7357
}
7358
}
7359
if (inodedep == NULL || check_inode_unwritten(inodedep)) {
2626
7360
FREE_LOCK(&lk);
2627
7361
handle_workitem_freefile(freefile);
2628
7362
return;
2629
7363
}
7364
if ((inodedep->id_state & DEPCOMPLETE) == 0)
7365
inodedep->id_state |= GOINGAWAY;
2630
7366
WORKLIST_INSERT(&inodedep->id_inowait, &freefile->fx_list);
2631
7367
FREE_LOCK(&lk);
2632
7368
if (ip->i_number == ino)
2654
7390
{
2655
7391
2656
7392
mtx_assert(&lk, MA_OWNED);
2657
if ((inodedep->id_state & DEPCOMPLETE) != 0 ||
7393
7394
if ((inodedep->id_state & (DEPCOMPLETE | UNLINKED)) != 0 ||
7395
!LIST_EMPTY(&inodedep->id_dirremhd) ||
2658
7396
!LIST_EMPTY(&inodedep->id_pendinghd) ||
2659
7397
!LIST_EMPTY(&inodedep->id_bufwait) ||
2660
7398
!LIST_EMPTY(&inodedep->id_inowait) ||
7399
!TAILQ_EMPTY(&inodedep->id_inoreflst) ||
2661
7400
!TAILQ_EMPTY(&inodedep->id_inoupdt) ||
2662
7401
!TAILQ_EMPTY(&inodedep->id_newinoupdt) ||
2663
7402
!TAILQ_EMPTY(&inodedep->id_extupdt) ||
2664
7403
!TAILQ_EMPTY(&inodedep->id_newextupdt) ||
7404
!TAILQ_EMPTY(&inodedep->id_freeblklst) ||
7405
inodedep->id_mkdiradd != NULL ||
2665
7406
inodedep->id_nlinkdelta != 0)
2666
7407
return (0);
2667
2668
7408
/*
2669
7409
* Another process might be in initiate_write_inodeblock_ufs[12]
2670
7410
* trying to allocate memory without holding "Softdep Lock".
2673
7413
inodedep->id_savedino1 == NULL)
2674
7414
return (0);
2675
7415
7416
if (inodedep->id_state & ONDEPLIST)
7417
LIST_REMOVE(inodedep, id_deps);
7418
inodedep->id_state &= ~ONDEPLIST;
2676
7419
inodedep->id_state |= ALLCOMPLETE;
2677
LIST_REMOVE(inodedep, id_deps);
2678
inodedep->id_buf = NULL;
7420
inodedep->id_bmsafemap = NULL;
2679
7421
if (inodedep->id_state & ONWORKLIST)
2680
7422
WORKLIST_REMOVE(&inodedep->id_list);
2681
7423
if (inodedep->id_savedino1 != NULL) {
2696
7438
{
2697
7439
2698
7440
mtx_assert(&lk, MA_OWNED);
2699
if ((inodedep->id_state & ONWORKLIST) != 0 ||
7441
if ((inodedep->id_state & (ONWORKLIST | UNLINKED)) != 0 ||
2700
7442
(inodedep->id_state & ALLCOMPLETE) != ALLCOMPLETE ||
7443
!LIST_EMPTY(&inodedep->id_dirremhd) ||
2701
7444
!LIST_EMPTY(&inodedep->id_pendinghd) ||
2702
7445
!LIST_EMPTY(&inodedep->id_bufwait) ||
2703
7446
!LIST_EMPTY(&inodedep->id_inowait) ||
7447
!TAILQ_EMPTY(&inodedep->id_inoreflst) ||
2704
7448
!TAILQ_EMPTY(&inodedep->id_inoupdt) ||
2705
7449
!TAILQ_EMPTY(&inodedep->id_newinoupdt) ||
2706
7450
!TAILQ_EMPTY(&inodedep->id_extupdt) ||
2707
7451
!TAILQ_EMPTY(&inodedep->id_newextupdt) ||
2708
inodedep->id_nlinkdelta != 0 || inodedep->id_savedino1 != NULL)
7452
!TAILQ_EMPTY(&inodedep->id_freeblklst) ||
7453
inodedep->id_mkdiradd != NULL ||
7454
inodedep->id_nlinkdelta != 0 ||
7455
inodedep->id_savedino1 != NULL)
2709
7456
return (0);
7457
if (inodedep->id_state & ONDEPLIST)
7458
LIST_REMOVE(inodedep, id_deps);
2710
7459
LIST_REMOVE(inodedep, id_hash);
2711
7460
WORKITEM_FREE(inodedep, D_INODEDEP);
2712
num_inodedep -= 1;
2713
7461
return (1);
2714
7462
}
2715
7463
2716
7464
/*
7465
* Free the block referenced by a freework structure. The parent freeblks
7466
* structure is released and completed when the final cg bitmap reaches
7467
* the disk. This routine may be freeing a jnewblk which never made it to
7468
* disk in which case we do not have to wait as the operation is undone
7469
* in memory immediately.
7470
*/
7471
static void
7472
freework_freeblock(freework)
7473
struct freework *freework;
7474
{
7475
struct freeblks *freeblks;
7476
struct jnewblk *jnewblk;
7477
struct ufsmount *ump;
7478
struct workhead wkhd;
7479
struct fs *fs;
7480
int bsize;
7481
int needj;
7482
7483
mtx_assert(&lk, MA_OWNED);
7484
/*
7485
* Handle partial truncate separately.
7486
*/
7487
if (freework->fw_indir) {
7488
complete_trunc_indir(freework);
7489
return;
7490
}
7491
freeblks = freework->fw_freeblks;
7492
ump = VFSTOUFS(freeblks->fb_list.wk_mp);
7493
fs = ump->um_fs;
7494
needj = MOUNTEDSUJ(freeblks->fb_list.wk_mp) != 0;
7495
bsize = lfragtosize(fs, freework->fw_frags);
7496
LIST_INIT(&wkhd);
7497
/*
7498
* DEPCOMPLETE is cleared in indirblk_insert() if the block lives
7499
* on the indirblk hashtable and prevents premature freeing.
7500
*/
7501
freework->fw_state |= DEPCOMPLETE;
7502
/*
7503
* SUJ needs to wait for the segment referencing freed indirect
7504
* blocks to expire so that we know the checker will not confuse
7505
* a re-allocated indirect block with its old contents.
7506
*/
7507
if (needj && freework->fw_lbn <= -NDADDR)
7508
indirblk_insert(freework);
7509
/*
7510
* If we are canceling an existing jnewblk pass it to the free
7511
* routine, otherwise pass the freeblk which will ultimately
7512
* release the freeblks. If we're not journaling, we can just
7513
* free the freeblks immediately.
7514
*/
7515
jnewblk = freework->fw_jnewblk;
7516
if (jnewblk != NULL) {
7517
cancel_jnewblk(jnewblk, &wkhd);
7518
needj = 0;
7519
} else if (needj) {
7520
freework->fw_state |= DELAYEDFREE;
7521
freeblks->fb_cgwait++;
7522
WORKLIST_INSERT(&wkhd, &freework->fw_list);
7523
}
7524
FREE_LOCK(&lk);
7525
freeblks_free(ump, freeblks, btodb(bsize));
7526
CTR4(KTR_SUJ,
7527
"freework_freeblock: ino %d blkno %jd lbn %jd size %ld",
7528
freeblks->fb_inum, freework->fw_blkno, freework->fw_lbn, bsize);
7529
ffs_blkfree(ump, fs, freeblks->fb_devvp, freework->fw_blkno, bsize,
7530
freeblks->fb_inum, freeblks->fb_vtype, &wkhd);
7531
ACQUIRE_LOCK(&lk);
7532
/*
7533
* The jnewblk will be discarded and the bits in the map never
7534
* made it to disk. We can immediately free the freeblk.
7535
*/
7536
if (needj == 0)
7537
handle_written_freework(freework);
7538
}
7539
7540
/*
7541
* We enqueue freework items that need processing back on the freeblks and
7542
* add the freeblks to the worklist. This makes it easier to find all work
7543
* required to flush a truncation in process_truncates().
7544
*/
7545
static void
7546
freework_enqueue(freework)
7547
struct freework *freework;
7548
{
7549
struct freeblks *freeblks;
7550
7551
freeblks = freework->fw_freeblks;
7552
if ((freework->fw_state & INPROGRESS) == 0)
7553
WORKLIST_INSERT(&freeblks->fb_freeworkhd, &freework->fw_list);
7554
if ((freeblks->fb_state &
7555
(ONWORKLIST | INPROGRESS | ALLCOMPLETE)) == ALLCOMPLETE &&
7556
LIST_EMPTY(&freeblks->fb_jblkdephd))
7557
add_to_worklist(&freeblks->fb_list, WK_NODELAY);
7558
}
7559
7560
/*
7561
* Start, continue, or finish the process of freeing an indirect block tree.
7562
* The free operation may be paused at any point with fw_off containing the
7563
* offset to restart from. This enables us to implement some flow control
7564
* for large truncates which may fan out and generate a huge number of
7565
* dependencies.
7566
*/
7567
static void
7568
handle_workitem_indirblk(freework)
7569
struct freework *freework;
7570
{
7571
struct freeblks *freeblks;
7572
struct ufsmount *ump;
7573
struct fs *fs;
7574
7575
freeblks = freework->fw_freeblks;
7576
ump = VFSTOUFS(freeblks->fb_list.wk_mp);
7577
fs = ump->um_fs;
7578
if (freework->fw_state & DEPCOMPLETE) {
7579
handle_written_freework(freework);
7580
return;
7581
}
7582
if (freework->fw_off == NINDIR(fs)) {
7583
freework_freeblock(freework);
7584
return;
7585
}
7586
freework->fw_state |= INPROGRESS;
7587
FREE_LOCK(&lk);
7588
indir_trunc(freework, fsbtodb(fs, freework->fw_blkno),
7589
freework->fw_lbn);
7590
ACQUIRE_LOCK(&lk);
7591
}
7592
7593
/*
7594
* Called when a freework structure attached to a cg buf is written. The
7595
* ref on either the parent or the freeblks structure is released and
7596
* the freeblks is added back to the worklist if there is more work to do.
7597
*/
7598
static void
7599
handle_written_freework(freework)
7600
struct freework *freework;
7601
{
7602
struct freeblks *freeblks;
7603
struct freework *parent;
7604
7605
freeblks = freework->fw_freeblks;
7606
parent = freework->fw_parent;
7607
if (freework->fw_state & DELAYEDFREE)
7608
freeblks->fb_cgwait--;
7609
freework->fw_state |= COMPLETE;
7610
if ((freework->fw_state & ALLCOMPLETE) == ALLCOMPLETE)
7611
WORKITEM_FREE(freework, D_FREEWORK);
7612
if (parent) {
7613
if (--parent->fw_ref == 0)
7614
freework_enqueue(parent);
7615
return;
7616
}
7617
if (--freeblks->fb_ref != 0)
7618
return;
7619
if ((freeblks->fb_state & (ALLCOMPLETE | ONWORKLIST | INPROGRESS)) ==
7620
ALLCOMPLETE && LIST_EMPTY(&freeblks->fb_jblkdephd))
7621
add_to_worklist(&freeblks->fb_list, WK_NODELAY);
7622
}
7623
7624
/*
2717
7625
* This workitem routine performs the block de-allocation.
2718
7626
* The workitem is added to the pending list after the updated
2719
7627
* inode block has been written to disk. As mentioned above,
2721
7629
* to the number of blocks allocated for the file) are also
2722
7630
* performed in this function.
2723
7631
*/
2724
static void
7632
static int
2725
7633
handle_workitem_freeblocks(freeblks, flags)
2726
7634
struct freeblks *freeblks;
2727
7635
int flags;
2728
7636
{
7637
struct freework *freework;
7638
struct newblk *newblk;
7639
struct allocindir *aip;
7640
struct ufsmount *ump;
7641
struct worklist *wk;
7642
7643
KASSERT(LIST_EMPTY(&freeblks->fb_jblkdephd),
7644
("handle_workitem_freeblocks: Journal entries not written."));
7645
ump = VFSTOUFS(freeblks->fb_list.wk_mp);
7646
ACQUIRE_LOCK(&lk);
7647
while ((wk = LIST_FIRST(&freeblks->fb_freeworkhd)) != NULL) {
7648
WORKLIST_REMOVE(wk);
7649
switch (wk->wk_type) {
7650
case D_DIRREM:
7651
wk->wk_state |= COMPLETE;
7652
add_to_worklist(wk, 0);
7653
continue;
7654
7655
case D_ALLOCDIRECT:
7656
free_newblk(WK_NEWBLK(wk));
7657
continue;
7658
7659
case D_ALLOCINDIR:
7660
aip = WK_ALLOCINDIR(wk);
7661
freework = NULL;
7662
if (aip->ai_state & DELAYEDFREE) {
7663
FREE_LOCK(&lk);
7664
freework = newfreework(ump, freeblks, NULL,
7665
aip->ai_lbn, aip->ai_newblkno,
7666
ump->um_fs->fs_frag, 0, 0);
7667
ACQUIRE_LOCK(&lk);
7668
}
7669
newblk = WK_NEWBLK(wk);
7670
if (newblk->nb_jnewblk) {
7671
freework->fw_jnewblk = newblk->nb_jnewblk;
7672
newblk->nb_jnewblk->jn_dep = &freework->fw_list;
7673
newblk->nb_jnewblk = NULL;
7674
}
7675
free_newblk(newblk);
7676
continue;
7677
7678
case D_FREEWORK:
7679
freework = WK_FREEWORK(wk);
7680
if (freework->fw_lbn <= -NDADDR)
7681
handle_workitem_indirblk(freework);
7682
else
7683
freework_freeblock(freework);
7684
continue;
7685
default:
7686
panic("handle_workitem_freeblocks: Unknown type %s",
7687
TYPENAME(wk->wk_type));
7688
}
7689
}
7690
if (freeblks->fb_ref != 0) {
7691
freeblks->fb_state &= ~INPROGRESS;
7692
wake_worklist(&freeblks->fb_list);
7693
freeblks = NULL;
7694
}
7695
FREE_LOCK(&lk);
7696
if (freeblks)
7697
return handle_complete_freeblocks(freeblks, flags);
7698
return (0);
7699
}
7700
7701
/*
7702
* Handle completion of block free via truncate. This allows fs_pending
7703
* to track the actual free block count more closely than if we only updated
7704
* it at the end. We must be careful to handle cases where the block count
7705
* on free was incorrect.
7706
*/
7707
static void
7708
freeblks_free(ump, freeblks, blocks)
7709
struct ufsmount *ump;
7710
struct freeblks *freeblks;
7711
int blocks;
7712
{
7713
struct fs *fs;
7714
ufs2_daddr_t remain;
7715
7716
UFS_LOCK(ump);
7717
remain = -freeblks->fb_chkcnt;
7718
freeblks->fb_chkcnt += blocks;
7719
if (remain > 0) {
7720
if (remain < blocks)
7721
blocks = remain;
7722
fs = ump->um_fs;
7723
fs->fs_pendingblocks -= blocks;
7724
}
7725
UFS_UNLOCK(ump);
7726
}
7727
7728
/*
7729
* Once all of the freework workitems are complete we can retire the
7730
* freeblocks dependency and any journal work awaiting completion. This
7731
* can not be called until all other dependencies are stable on disk.
7732
*/
7733
static int
7734
handle_complete_freeblocks(freeblks, flags)
7735
struct freeblks *freeblks;
7736
int flags;
7737
{
7738
struct inodedep *inodedep;
2729
7739
struct inode *ip;
2730
7740
struct vnode *vp;
2731
7741
struct fs *fs;
2732
7742
struct ufsmount *ump;
2733
int i, nblocks, level, bsize;
2734
ufs2_daddr_t bn, blocksreleased = 0;
2735
int error, allerror = 0;
2736
ufs_lbn_t baselbns[NIADDR], tmpval;
2737
int fs_pendingblocks;
7743
ufs2_daddr_t spare;
2738
7744
2739
7745
ump = VFSTOUFS(freeblks->fb_list.wk_mp);
2740
7746
fs = ump->um_fs;
2741
fs_pendingblocks = 0;
2742
tmpval = 1;
2743
baselbns[0] = NDADDR;
2744
for (i = 1; i < NIADDR; i++) {
2745
tmpval *= NINDIR(fs);
2746
baselbns[i] = baselbns[i - 1] + tmpval;
2747
}
2748
nblocks = btodb(fs->fs_bsize);
2749
blocksreleased = 0;
2750
/*
2751
* Release all extended attribute blocks or frags.
2752
*/
2753
if (freeblks->fb_oldextsize > 0) {
2754
for (i = (NXADDR - 1); i >= 0; i--) {
2755
if ((bn = freeblks->fb_eblks[i]) == 0)
2756
continue;
2757
bsize = sblksize(fs, freeblks->fb_oldextsize, i);
2758
ffs_blkfree(ump, fs, freeblks->fb_devvp, bn, bsize,
2759
freeblks->fb_previousinum);
2760
blocksreleased += btodb(bsize);
2761
}
2762
}
2763
/*
2764
* Release all data blocks or frags.
2765
*/
2766
if (freeblks->fb_oldsize > 0) {
2767
/*
2768
* Indirect blocks first.
2769
*/
2770
for (level = (NIADDR - 1); level >= 0; level--) {
2771
if ((bn = freeblks->fb_iblks[level]) == 0)
2772
continue;
2773
if ((error = indir_trunc(freeblks, fsbtodb(fs, bn),
2774
level, baselbns[level], &blocksreleased)) != 0)
2775
allerror = error;
2776
ffs_blkfree(ump, fs, freeblks->fb_devvp, bn,
2777
fs->fs_bsize, freeblks->fb_previousinum);
2778
fs_pendingblocks += nblocks;
2779
blocksreleased += nblocks;
2780
}
2781
/*
2782
* All direct blocks or frags.
2783
*/
2784
for (i = (NDADDR - 1); i >= 0; i--) {
2785
if ((bn = freeblks->fb_dblks[i]) == 0)
2786
continue;
2787
bsize = sblksize(fs, freeblks->fb_oldsize, i);
2788
ffs_blkfree(ump, fs, freeblks->fb_devvp, bn, bsize,
2789
freeblks->fb_previousinum);
2790
fs_pendingblocks += btodb(bsize);
2791
blocksreleased += btodb(bsize);
2792
}
2793
}
2794
UFS_LOCK(ump);
2795
fs->fs_pendingblocks -= fs_pendingblocks;
2796
UFS_UNLOCK(ump);
2797
/*
2798
* If we still have not finished background cleanup, then check
2799
* to see if the block count needs to be adjusted.
2800
*/
2801
if (freeblks->fb_chkcnt != blocksreleased &&
2802
(fs->fs_flags & FS_UNCLEAN) != 0 &&
2803
ffs_vgetf(freeblks->fb_list.wk_mp, freeblks->fb_previousinum,
2804
(flags & LK_NOWAIT) | LK_EXCLUSIVE, &vp, FFSV_FORCEINSMQ)
2805
== 0) {
7747
flags = LK_EXCLUSIVE | flags;
7748
spare = freeblks->fb_chkcnt;
7749
7750
/*
7751
* If we did not release the expected number of blocks we may have
7752
* to adjust the inode block count here. Only do so if it wasn't
7753
* a truncation to zero and the modrev still matches.
7754
*/
7755
if (spare && freeblks->fb_len != 0) {
7756
if (ffs_vgetf(freeblks->fb_list.wk_mp, freeblks->fb_inum,
7757
flags, &vp, FFSV_FORCEINSMQ) != 0)
7758
return (EBUSY);
2806
7759
ip = VTOI(vp);
2807
DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + \
2808
freeblks->fb_chkcnt - blocksreleased);
2809
ip->i_flag |= IN_CHANGE;
7760
if (DIP(ip, i_modrev) == freeblks->fb_modrev) {
7761
DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - spare);
7762
ip->i_flag |= IN_CHANGE;
7763
/*
7764
* We must wait so this happens before the
7765
* journal is reclaimed.
7766
*/
7767
ffs_update(vp, 1);
7768
}
2810
7769
vput(vp);
2811
7770
}
2812
2813
#ifdef INVARIANTS
2814
if (freeblks->fb_chkcnt != blocksreleased &&
2815
((fs->fs_flags & FS_UNCLEAN) == 0 || (flags & LK_NOWAIT) != 0))
2816
printf("handle_workitem_freeblocks: block count\n");
2817
if (allerror)
2818
softdep_error("handle_workitem_freeblks", allerror);
2819
#endif /* INVARIANTS */
2820
7771
if (spare < 0) {
7772
UFS_LOCK(ump);
7773
fs->fs_pendingblocks += spare;
7774
UFS_UNLOCK(ump);
7775
}
7776
#ifdef QUOTA
7777
/* Handle spare. */
7778
if (spare)
7779
quotaadj(freeblks->fb_quota, ump, -spare);
7780
quotarele(freeblks->fb_quota);
7781
#endif
2821
7782
ACQUIRE_LOCK(&lk);
7783
if (freeblks->fb_state & ONDEPLIST) {
7784
inodedep_lookup(freeblks->fb_list.wk_mp, freeblks->fb_inum,
7785
0, &inodedep);
7786
TAILQ_REMOVE(&inodedep->id_freeblklst, freeblks, fb_next);
7787
freeblks->fb_state &= ~ONDEPLIST;
7788
if (TAILQ_EMPTY(&inodedep->id_freeblklst))
7789
free_inodedep(inodedep);
7790
}
7791
/*
7792
* All of the freeblock deps must be complete prior to this call
7793
* so it's now safe to complete earlier outstanding journal entries.
7794
*/
7795
handle_jwork(&freeblks->fb_jwork);
2822
7796
WORKITEM_FREE(freeblks, D_FREEBLKS);
2823
num_freeblkdep--;
2824
7797
FREE_LOCK(&lk);
7798
return (0);
2825
7799
}
2826
7800
2827
7801
/*
2828
* Release blocks associated with the inode ip and stored in the indirect
7802
* Release blocks associated with the freeblks and stored in the indirect
2829
7803
* block dbn. If level is greater than SINGLE, the block is an indirect block
2830
7804
* and recursive calls to indirtrunc must be used to cleanse other indirect
2831
7805
* blocks.
7806
*
7807
* This handles partial and complete truncation of blocks. Partial is noted
7808
* with goingaway == 0. In this case the freework is completed after the
7809
* zero'd indirects are written to disk. For full truncation the freework
7810
* is completed after the block is freed.
2832
7811
*/
2833
static int
2834
indir_trunc(freeblks, dbn, level, lbn, countp)
2835
struct freeblks *freeblks;
7812
static void
7813
indir_trunc(freework, dbn, lbn)
7814
struct freework *freework;
2836
7815
ufs2_daddr_t dbn;
2837
int level;
2838
7816
ufs_lbn_t lbn;
2839
ufs2_daddr_t *countp;
2840
7817
{
7818
struct freework *nfreework;
7819
struct workhead wkhd;
7820
struct freeblks *freeblks;
2841
7821
struct buf *bp;
2842
7822
struct fs *fs;
2843
struct worklist *wk;
2844
7823
struct indirdep *indirdep;
2845
7824
struct ufsmount *ump;
2846
7825
ufs1_daddr_t *bap1 = 0;
2847
ufs2_daddr_t nb, *bap2 = 0;
2848
ufs_lbn_t lbnadd;
7826
ufs2_daddr_t nb, nnb, *bap2 = 0;
7827
ufs_lbn_t lbnadd, nlbn;
2849
7828
int i, nblocks, ufs1fmt;
2850
int error, allerror = 0;
2851
int fs_pendingblocks;
7829
int freedblocks;
7830
int goingaway;
7831
int freedeps;
7832
int needj;
7833
int level;
7834
int cnt;
2852
7835
7836
freeblks = freework->fw_freeblks;
2853
7837
ump = VFSTOUFS(freeblks->fb_list.wk_mp);
2854
7838
fs = ump->um_fs;
2855
fs_pendingblocks = 0;
2856
lbnadd = 1;
2857
for (i = level; i > 0; i--)
2858
lbnadd *= NINDIR(fs);
2859
7839
/*
2860
* Get buffer of block pointers to be freed. This routine is not
2861
* called until the zero'ed inode has been written, so it is safe
2862
* to free blocks as they are encountered. Because the inode has
2863
* been zero'ed, calls to bmap on these blocks will fail. So, we
2864
* have to use the on-disk address and the block device for the
2865
* filesystem to look them up. If the file was deleted before its
2866
* indirect blocks were all written to disk, the routine that set
2867
* us up (deallocate_dependencies) will have arranged to leave
2868
* a complete copy of the indirect block in memory for our use.
2869
* Otherwise we have to read the blocks in from the disk.
7840
* Get buffer of block pointers to be freed. There are three cases:
7841
*
7842
* 1) Partial truncate caches the indirdep pointer in the freework
7843
* which provides us a back copy to the save bp which holds the
7844
* pointers we want to clear. When this completes the zero
7845
* pointers are written to the real copy.
7846
* 2) The indirect is being completely truncated, cancel_indirdep()
7847
* eliminated the real copy and placed the indirdep on the saved
7848
* copy. The indirdep and buf are discarded when this completes.
7849
* 3) The indirect was not in memory, we read a copy off of the disk
7850
* using the devvp and drop and invalidate the buffer when we're
7851
* done.
2870
7852
*/
2871
#ifdef notyet
2872
bp = getblk(freeblks->fb_devvp, dbn, (int)fs->fs_bsize, 0, 0,
2873
GB_NOCREAT);
2874
#else
2875
bp = incore(&freeblks->fb_devvp->v_bufobj, dbn);
2876
#endif
7853
goingaway = 1;
7854
indirdep = NULL;
7855
if (freework->fw_indir != NULL) {
7856
goingaway = 0;
7857
indirdep = freework->fw_indir;
7858
bp = indirdep->ir_savebp;
7859
if (bp == NULL || bp->b_blkno != dbn)
7860
panic("indir_trunc: Bad saved buf %p blkno %jd",
7861
bp, (intmax_t)dbn);
7862
} else if ((bp = incore(&freeblks->fb_devvp->v_bufobj, dbn)) != NULL) {
7863
/*
7864
* The lock prevents the buf dep list from changing and
7865
* indirects on devvp should only ever have one dependency.
7866
*/
7867
indirdep = WK_INDIRDEP(LIST_FIRST(&bp->b_dep));
7868
if (indirdep == NULL || (indirdep->ir_state & GOINGAWAY) == 0)
7869
panic("indir_trunc: Bad indirdep %p from buf %p",
7870
indirdep, bp);
7871
} else if (bread(freeblks->fb_devvp, dbn, (int)fs->fs_bsize,
7872
NOCRED, &bp) != 0) {
7873
brelse(bp);
7874
return;
7875
}
2877
7876
ACQUIRE_LOCK(&lk);
2878
if (bp != NULL && (wk = LIST_FIRST(&bp->b_dep)) != NULL) {
2879
if (wk->wk_type != D_INDIRDEP ||
2880
(indirdep = WK_INDIRDEP(wk))->ir_savebp != bp ||
2881
(indirdep->ir_state & GOINGAWAY) == 0)
2882
panic("indir_trunc: lost indirdep");
2883
WORKLIST_REMOVE(wk);
2884
WORKITEM_FREE(indirdep, D_INDIRDEP);
2885
if (!LIST_EMPTY(&bp->b_dep))
2886
panic("indir_trunc: dangling dep");
2887
ump->um_numindirdeps -= 1;
2888
FREE_LOCK(&lk);
2889
} else {
2890
#ifdef notyet
2891
if (bp)
2892
brelse(bp);
2893
#endif
2894
FREE_LOCK(&lk);
2895
error = bread(freeblks->fb_devvp, dbn, (int)fs->fs_bsize,
2896
NOCRED, &bp);
2897
if (error) {
2898
brelse(bp);
2899
return (error);
2900
}
2901
}
7877
/* Protects against a race with complete_trunc_indir(). */
7878
freework->fw_state &= ~INPROGRESS;
2902
7879
/*
2903
* Recursively free indirect blocks.
7880
* If we have an indirdep we need to enforce the truncation order
7881
* and discard it when it is complete.
2904
7882
*/
7883
if (indirdep) {
7884
if (freework != TAILQ_FIRST(&indirdep->ir_trunc) &&
7885
!TAILQ_EMPTY(&indirdep->ir_trunc)) {
7886
/*
7887
* Add the complete truncate to the list on the
7888
* indirdep to enforce in-order processing.
7889
*/
7890
if (freework->fw_indir == NULL)
7891
TAILQ_INSERT_TAIL(&indirdep->ir_trunc,
7892
freework, fw_next);
7893
FREE_LOCK(&lk);
7894
return;
7895
}
7896
/*
7897
* If we're goingaway, free the indirdep. Otherwise it will
7898
* linger until the write completes.
7899
*/
7900
if (goingaway) {
7901
free_indirdep(indirdep);
7902
ump->um_numindirdeps -= 1;
7903
}
7904
}
7905
FREE_LOCK(&lk);
7906
/* Initialize pointers depending on block size. */
2905
7907
if (ump->um_fstype == UFS1) {
7908
bap1 = (ufs1_daddr_t *)bp->b_data;
7909
nb = bap1[freework->fw_off];
2906
7910
ufs1fmt = 1;
2907
bap1 = (ufs1_daddr_t *)bp->b_data;
2908
7911
} else {
7912
bap2 = (ufs2_daddr_t *)bp->b_data;
7913
nb = bap2[freework->fw_off];
2909
7914
ufs1fmt = 0;
2910
bap2 = (ufs2_daddr_t *)bp->b_data;
2911
7915
}
7916
level = lbn_level(lbn);
7917
needj = MOUNTEDSUJ(UFSTOVFS(ump)) != 0;
7918
lbnadd = lbn_offset(fs, level);
2912
7919
nblocks = btodb(fs->fs_bsize);
2913
for (i = NINDIR(fs) - 1; i >= 0; i--) {
2914
if (ufs1fmt)
2915
nb = bap1[i];
2916
else
2917
nb = bap2[i];
7920
nfreework = freework;
7921
freedeps = 0;
7922
cnt = 0;
7923
/*
7924
* Reclaim blocks. Traverses into nested indirect levels and
7925
* arranges for the current level to be freed when subordinates
7926
* are free when journaling.
7927
*/
7928
for (i = freework->fw_off; i < NINDIR(fs); i++, nb = nnb) {
7929
if (i != NINDIR(fs) - 1) {
7930
if (ufs1fmt)
7931
nnb = bap1[i+1];
7932
else
7933
nnb = bap2[i+1];
7934
} else
7935
nnb = 0;
2918
7936
if (nb == 0)
2919
7937
continue;
7938
cnt++;
2920
7939
if (level != 0) {
2921
if ((error = indir_trunc(freeblks, fsbtodb(fs, nb),
2922
level - 1, lbn + (i * lbnadd), countp)) != 0)
2923
allerror = error;
7940
nlbn = (lbn + 1) - (i * lbnadd);
7941
if (needj != 0) {
7942
nfreework = newfreework(ump, freeblks, freework,
7943
nlbn, nb, fs->fs_frag, 0, 0);
7944
freedeps++;
7945
}
7946
indir_trunc(nfreework, fsbtodb(fs, nb), nlbn);
7947
} else {
7948
struct freedep *freedep;
7949
7950
/*
7951
* Attempt to aggregate freedep dependencies for
7952
* all blocks being released to the same CG.
7953
*/
7954
LIST_INIT(&wkhd);
7955
if (needj != 0 &&
7956
(nnb == 0 || (dtog(fs, nb) != dtog(fs, nnb)))) {
7957
freedep = newfreedep(freework);
7958
WORKLIST_INSERT_UNLOCKED(&wkhd,
7959
&freedep->fd_list);
7960
freedeps++;
7961
}
7962
CTR3(KTR_SUJ,
7963
"indir_trunc: ino %d blkno %jd size %ld",
7964
freeblks->fb_inum, nb, fs->fs_bsize);
7965
ffs_blkfree(ump, fs, freeblks->fb_devvp, nb,
7966
fs->fs_bsize, freeblks->fb_inum,
7967
freeblks->fb_vtype, &wkhd);
2924
7968
}
2925
ffs_blkfree(ump, fs, freeblks->fb_devvp, nb, fs->fs_bsize,
2926
freeblks->fb_previousinum);
2927
fs_pendingblocks += nblocks;
2928
*countp += nblocks;
2929
}
2930
UFS_LOCK(ump);
2931
fs->fs_pendingblocks -= fs_pendingblocks;
2932
UFS_UNLOCK(ump);
2933
bp->b_flags |= B_INVAL | B_NOCACHE;
2934
brelse(bp);
2935
return (allerror);
7969
}
7970
if (goingaway) {
7971
bp->b_flags |= B_INVAL | B_NOCACHE;
7972
brelse(bp);
7973
}
7974
freedblocks = 0;
7975
if (level == 0)
7976
freedblocks = (nblocks * cnt);
7977
if (needj == 0)
7978
freedblocks += nblocks;
7979
freeblks_free(ump, freeblks, freedblocks);
7980
/*
7981
* If we are journaling set up the ref counts and offset so this
7982
* indirect can be completed when its children are free.
7983
*/
7984
if (needj) {
7985
ACQUIRE_LOCK(&lk);
7986
freework->fw_off = i;
7987
freework->fw_ref += freedeps;
7988
freework->fw_ref -= NINDIR(fs) + 1;
7989
if (level == 0)
7990
freeblks->fb_cgwait += freedeps;
7991
if (freework->fw_ref == 0)
7992
freework_freeblock(freework);
7993
FREE_LOCK(&lk);
7994
return;
7995
}
7996
/*
7997
* If we're not journaling we can free the indirect now.
7998
*/
7999
dbn = dbtofsb(fs, dbn);
8000
CTR3(KTR_SUJ,
8001
"indir_trunc 2: ino %d blkno %jd size %ld",
8002
freeblks->fb_inum, dbn, fs->fs_bsize);
8003
ffs_blkfree(ump, fs, freeblks->fb_devvp, dbn, fs->fs_bsize,
8004
freeblks->fb_inum, freeblks->fb_vtype, NULL);
8005
/* Non SUJ softdep does single-threaded truncations. */
8006
if (freework->fw_blkno == dbn) {
8007
freework->fw_state |= ALLCOMPLETE;
8008
ACQUIRE_LOCK(&lk);
8009
handle_written_freework(freework);
8010
FREE_LOCK(&lk);
8011
}
8012
return;
2936
8013
}
2937
8014
2938
8015
/*
2939
* Free an allocindir.
2940
* This routine must be called with splbio interrupts blocked.
8016
* Cancel an allocindir when it is removed via truncation. When bp is not
8017
* NULL the indirect never appeared on disk and is scheduled to be freed
8018
* independently of the indir so we can more easily track journal work.
2941
8019
*/
2942
8020
static void
2943
free_allocindir(aip, inodedep)
8021
cancel_allocindir(aip, bp, freeblks, trunc)
2944
8022
struct allocindir *aip;
2945
struct inodedep *inodedep;
8023
struct buf *bp;
8024
struct freeblks *freeblks;
8025
int trunc;
2946
8026
{
8027
struct indirdep *indirdep;
2947
8028
struct freefrag *freefrag;
8029
struct newblk *newblk;
2948
8030
2949
mtx_assert(&lk, MA_OWNED);
2950
if ((aip->ai_state & DEPCOMPLETE) == 0)
2951
LIST_REMOVE(aip, ai_deps);
2952
if (aip->ai_state & ONWORKLIST)
2953
WORKLIST_REMOVE(&aip->ai_list);
8031
newblk = (struct newblk *)aip;
2954
8032
LIST_REMOVE(aip, ai_next);
2955
if ((freefrag = aip->ai_freefrag) != NULL) {
8033
/*
8034
* We must eliminate the pointer in bp if it must be freed on its
8035
* own due to partial truncate or pending journal work.
8036
*/
8037
if (bp && (trunc || newblk->nb_jnewblk)) {
8038
/*
8039
* Clear the pointer and mark the aip to be freed
8040
* directly if it never existed on disk.
8041
*/
8042
aip->ai_state |= DELAYEDFREE;
8043
indirdep = aip->ai_indirdep;
8044
if (indirdep->ir_state & UFS1FMT)
8045
((ufs1_daddr_t *)bp->b_data)[aip->ai_offset] = 0;
8046
else
8047
((ufs2_daddr_t *)bp->b_data)[aip->ai_offset] = 0;
8048
}
8049
/*
8050
* When truncating the previous pointer will be freed via
8051
* savedbp. Eliminate the freefrag which would dup free.
8052
*/
8053
if (trunc && (freefrag = newblk->nb_freefrag) != NULL) {
8054
newblk->nb_freefrag = NULL;
8055
if (freefrag->ff_jdep)
8056
cancel_jfreefrag(
8057
WK_JFREEFRAG(freefrag->ff_jdep));
8058
jwork_move(&freeblks->fb_jwork, &freefrag->ff_jwork);
8059
WORKITEM_FREE(freefrag, D_FREEFRAG);
8060
}
8061
/*
8062
* If the journal hasn't been written the jnewblk must be passed
8063
* to the call to ffs_blkfree that reclaims the space. We accomplish
8064
* this by leaving the journal dependency on the newblk to be freed
8065
* when a freework is created in handle_workitem_freeblocks().
8066
*/
8067
cancel_newblk(newblk, NULL, &freeblks->fb_jwork);
8068
WORKLIST_INSERT(&freeblks->fb_freeworkhd, &newblk->nb_list);
8069
}
8070
8071
/*
8072
* Create the mkdir dependencies for . and .. in a new directory. Link them
8073
* in to a newdirblk so any subsequent additions are tracked properly. The
8074
* caller is responsible for adding the mkdir1 dependency to the journal
8075
* and updating id_mkdiradd. This function returns with lk held.
8076
*/
8077
static struct mkdir *
8078
setup_newdir(dap, newinum, dinum, newdirbp, mkdirp)
8079
struct diradd *dap;
8080
ino_t newinum;
8081
ino_t dinum;
8082
struct buf *newdirbp;
8083
struct mkdir **mkdirp;
8084
{
8085
struct newblk *newblk;
8086
struct pagedep *pagedep;
8087
struct inodedep *inodedep;
8088
struct newdirblk *newdirblk = 0;
8089
struct mkdir *mkdir1, *mkdir2;
8090
struct worklist *wk;
8091
struct jaddref *jaddref;
8092
struct mount *mp;
8093
8094
mp = dap->da_list.wk_mp;
8095
newdirblk = malloc(sizeof(struct newdirblk), M_NEWDIRBLK,
8096
M_SOFTDEP_FLAGS);
8097
workitem_alloc(&newdirblk->db_list, D_NEWDIRBLK, mp);
8098
LIST_INIT(&newdirblk->db_mkdir);
8099
mkdir1 = malloc(sizeof(struct mkdir), M_MKDIR, M_SOFTDEP_FLAGS);
8100
workitem_alloc(&mkdir1->md_list, D_MKDIR, mp);
8101
mkdir1->md_state = ATTACHED | MKDIR_BODY;
8102
mkdir1->md_diradd = dap;
8103
mkdir1->md_jaddref = NULL;
8104
mkdir2 = malloc(sizeof(struct mkdir), M_MKDIR, M_SOFTDEP_FLAGS);
8105
workitem_alloc(&mkdir2->md_list, D_MKDIR, mp);
8106
mkdir2->md_state = ATTACHED | MKDIR_PARENT;
8107
mkdir2->md_diradd = dap;
8108
mkdir2->md_jaddref = NULL;
8109
if (MOUNTEDSUJ(mp) == 0) {
8110
mkdir1->md_state |= DEPCOMPLETE;
8111
mkdir2->md_state |= DEPCOMPLETE;
8112
}
8113
/*
8114
* Dependency on "." and ".." being written to disk.
8115
*/
8116
mkdir1->md_buf = newdirbp;
8117
ACQUIRE_LOCK(&lk);
8118
LIST_INSERT_HEAD(&mkdirlisthd, mkdir1, md_mkdirs);
8119
/*
8120
* We must link the pagedep, allocdirect, and newdirblk for
8121
* the initial file page so the pointer to the new directory
8122
* is not written until the directory contents are live and
8123
* any subsequent additions are not marked live until the
8124
* block is reachable via the inode.
8125
*/
8126
if (pagedep_lookup(mp, newdirbp, newinum, 0, 0, &pagedep) == 0)
8127
panic("setup_newdir: lost pagedep");
8128
LIST_FOREACH(wk, &newdirbp->b_dep, wk_list)
8129
if (wk->wk_type == D_ALLOCDIRECT)
8130
break;
8131
if (wk == NULL)
8132
panic("setup_newdir: lost allocdirect");
8133
if (pagedep->pd_state & NEWBLOCK)
8134
panic("setup_newdir: NEWBLOCK already set");
8135
newblk = WK_NEWBLK(wk);
8136
pagedep->pd_state |= NEWBLOCK;
8137
pagedep->pd_newdirblk = newdirblk;
8138
newdirblk->db_pagedep = pagedep;
8139
WORKLIST_INSERT(&newblk->nb_newdirblk, &newdirblk->db_list);
8140
WORKLIST_INSERT(&newdirblk->db_mkdir, &mkdir1->md_list);
8141
/*
8142
* Look up the inodedep for the parent directory so that we
8143
* can link mkdir2 into the pending dotdot jaddref or
8144
* the inode write if there is none. If the inode is
8145
* ALLCOMPLETE and no jaddref is present all dependencies have
8146
* been satisfied and mkdir2 can be freed.
8147
*/
8148
inodedep_lookup(mp, dinum, 0, &inodedep);
8149
if (MOUNTEDSUJ(mp)) {
2956
8150
if (inodedep == NULL)
2957
add_to_worklist(&freefrag->ff_list);
2958
else
2959
WORKLIST_INSERT(&inodedep->id_bufwait,
2960
&freefrag->ff_list);
8151
panic("setup_newdir: Lost parent.");
8152
jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst,
8153
inoreflst);
8154
KASSERT(jaddref != NULL && jaddref->ja_parent == newinum &&
8155
(jaddref->ja_state & MKDIR_PARENT),
8156
("setup_newdir: bad dotdot jaddref %p", jaddref));
8157
LIST_INSERT_HEAD(&mkdirlisthd, mkdir2, md_mkdirs);
8158
mkdir2->md_jaddref = jaddref;
8159
jaddref->ja_mkdir = mkdir2;
8160
} else if (inodedep == NULL ||
8161
(inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) {
8162
dap->da_state &= ~MKDIR_PARENT;
8163
WORKITEM_FREE(mkdir2, D_MKDIR);
8164
mkdir2 = NULL;
8165
} else {
8166
LIST_INSERT_HEAD(&mkdirlisthd, mkdir2, md_mkdirs);
8167
WORKLIST_INSERT(&inodedep->id_bufwait, &mkdir2->md_list);
2961
8168
}
2962
WORKITEM_FREE(aip, D_ALLOCINDIR);
8169
*mkdirp = mkdir2;
8170
8171
return (mkdir1);
2963
8172
}
2964
8173
2965
8174
/*
2998
8207
ufs_lbn_t lbn; /* block in directory containing new entry */
2999
8208
struct fs *fs;
3000
8209
struct diradd *dap;
3001
struct allocdirect *adp;
8210
struct newblk *newblk;
3002
8211
struct pagedep *pagedep;
3003
8212
struct inodedep *inodedep;
3004
8213
struct newdirblk *newdirblk = 0;
3005
8214
struct mkdir *mkdir1, *mkdir2;
8215
struct jaddref *jaddref;
3006
8216
struct mount *mp;
8217
int isindir;
3007
8218
3008
8219
/*
3009
8220
* Whiteouts have no dependencies.
3013
8224
bdwrite(newdirbp);
3014
8225
return (0);
3015
8226
}
8227
jaddref = NULL;
8228
mkdir1 = mkdir2 = NULL;
3016
8229
mp = UFSTOVFS(dp->i_ump);
3017
8230
fs = dp->i_fs;
3018
8231
lbn = lblkno(fs, diroffset);
3023
8236
dap->da_offset = offset;
3024
8237
dap->da_newinum = newinum;
3025
8238
dap->da_state = ATTACHED;
3026
if (isnewblk && lbn < NDADDR && fragoff(fs, diroffset) == 0) {
8239
LIST_INIT(&dap->da_jwork);
8240
isindir = bp->b_lblkno >= NDADDR;
8241
if (isnewblk &&
8242
(isindir ? blkoff(fs, diroffset) : fragoff(fs, diroffset)) == 0) {
3027
8243
newdirblk = malloc(sizeof(struct newdirblk),
3028
8244
M_NEWDIRBLK, M_SOFTDEP_FLAGS);
3029
8245
workitem_alloc(&newdirblk->db_list, D_NEWDIRBLK, mp);
8246
LIST_INIT(&newdirblk->db_mkdir);
3030
8247
}
8248
/*
8249
* If we're creating a new directory setup the dependencies and set
8250
* the dap state to wait for them. Otherwise it's COMPLETE and
8251
* we can move on.
8252
*/
3031
8253
if (newdirbp == NULL) {
3032
8254
dap->da_state |= DEPCOMPLETE;
3033
8255
ACQUIRE_LOCK(&lk);
3034
8256
} else {
3035
8257
dap->da_state |= MKDIR_BODY | MKDIR_PARENT;
3036
mkdir1 = malloc(sizeof(struct mkdir), M_MKDIR,
3037
M_SOFTDEP_FLAGS);
3038
workitem_alloc(&mkdir1->md_list, D_MKDIR, mp);
3039
mkdir1->md_state = MKDIR_BODY;
3040
mkdir1->md_diradd = dap;
3041
mkdir2 = malloc(sizeof(struct mkdir), M_MKDIR,
3042
M_SOFTDEP_FLAGS);
3043
workitem_alloc(&mkdir2->md_list, D_MKDIR, mp);
3044
mkdir2->md_state = MKDIR_PARENT;
3045
mkdir2->md_diradd = dap;
3046
/*
3047
* Dependency on "." and ".." being written to disk.
3048
*/
3049
mkdir1->md_buf = newdirbp;
3050
ACQUIRE_LOCK(&lk);
3051
LIST_INSERT_HEAD(&mkdirlisthd, mkdir1, md_mkdirs);
3052
WORKLIST_INSERT(&newdirbp->b_dep, &mkdir1->md_list);
3053
FREE_LOCK(&lk);
3054
bdwrite(newdirbp);
3055
/*
3056
* Dependency on link count increase for parent directory
3057
*/
3058
ACQUIRE_LOCK(&lk);
3059
if (inodedep_lookup(mp, dp->i_number, 0, &inodedep) == 0
3060
|| (inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) {
3061
dap->da_state &= ~MKDIR_PARENT;
3062
WORKITEM_FREE(mkdir2, D_MKDIR);
3063
} else {
3064
LIST_INSERT_HEAD(&mkdirlisthd, mkdir2, md_mkdirs);
3065
WORKLIST_INSERT(&inodedep->id_bufwait,&mkdir2->md_list);
3066
}
8258
mkdir1 = setup_newdir(dap, newinum, dp->i_number, newdirbp,
8259
&mkdir2);
3067
8260
}
3068
8261
/*
3069
8262
* Link into parent directory pagedep to await its being written.
3070
8263
*/
3071
if (pagedep_lookup(dp, lbn, DEPALLOC, &pagedep) == 0)
3072
WORKLIST_INSERT(&bp->b_dep, &pagedep->pd_list);
8264
pagedep_lookup(mp, bp, dp->i_number, lbn, DEPALLOC, &pagedep);
8265
#ifdef DEBUG
8266
if (diradd_lookup(pagedep, offset) != NULL)
8267
panic("softdep_setup_directory_add: %p already at off %d\n",
8268
diradd_lookup(pagedep, offset), offset);
8269
#endif
3073
8270
dap->da_pagedep = pagedep;
3074
8271
LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(offset)], dap,
3075
8272
da_pdlist);
8273
inodedep_lookup(mp, newinum, DEPALLOC | NODELAY, &inodedep);
3076
8274
/*
3077
* Link into its inodedep. Put it on the id_bufwait list if the inode
3078
* is not yet written. If it is written, do the post-inode write
3079
* processing to put it on the id_pendinghd list.
8275
* If we're journaling, link the diradd into the jaddref so it
8276
* may be completed after the journal entry is written. Otherwise,
8277
* link the diradd into its inodedep. If the inode is not yet
8278
* written place it on the bufwait list, otherwise do the post-inode
8279
* write processing to put it on the id_pendinghd list.
3080
8280
*/
3081
(void) inodedep_lookup(mp, newinum, DEPALLOC, &inodedep);
3082
if ((inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE)
8281
if (MOUNTEDSUJ(mp)) {
8282
jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst,
8283
inoreflst);
8284
KASSERT(jaddref != NULL && jaddref->ja_parent == dp->i_number,
8285
("softdep_setup_directory_add: bad jaddref %p", jaddref));
8286
jaddref->ja_diroff = diroffset;
8287
jaddref->ja_diradd = dap;
8288
add_to_journal(&jaddref->ja_list);
8289
} else if ((inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE)
3083
8290
diradd_inode_written(dap, inodedep);
3084
8291
else
3085
8292
WORKLIST_INSERT(&inodedep->id_bufwait, &dap->da_list);
3086
if (isnewblk) {
3087
/*
3088
* Directories growing into indirect blocks are rare
3089
* enough and the frequency of new block allocation
3090
* in those cases even more rare, that we choose not
3091
* to bother tracking them. Rather we simply force the
3092
* new directory entry to disk.
3093
*/
3094
if (lbn >= NDADDR) {
3095
FREE_LOCK(&lk);
3096
/*
3097
* We only have a new allocation when at the
3098
* beginning of a new block, not when we are
3099
* expanding into an existing block.
3100
*/
3101
if (blkoff(fs, diroffset) == 0)
3102
return (1);
3103
return (0);
3104
}
3105
/*
3106
* We only have a new allocation when at the beginning
3107
* of a new fragment, not when we are expanding into an
3108
* existing fragment. Also, there is nothing to do if we
3109
* are already tracking this block.
3110
*/
3111
if (fragoff(fs, diroffset) != 0) {
3112
FREE_LOCK(&lk);
3113
return (0);
3114
}
8293
/*
8294
* Add the journal entries for . and .. links now that the primary
8295
* link is written.
8296
*/
8297
if (mkdir1 != NULL && MOUNTEDSUJ(mp)) {
8298
jaddref = (struct jaddref *)TAILQ_PREV(&jaddref->ja_ref,
8299
inoreflst, if_deps);
8300
KASSERT(jaddref != NULL &&
8301
jaddref->ja_ino == jaddref->ja_parent &&
8302
(jaddref->ja_state & MKDIR_BODY),
8303
("softdep_setup_directory_add: bad dot jaddref %p",
8304
jaddref));
8305
mkdir1->md_jaddref = jaddref;
8306
jaddref->ja_mkdir = mkdir1;
8307
/*
8308
* It is important that the dotdot journal entry
8309
* is added prior to the dot entry since dot writes
8310
* both the dot and dotdot links. These both must
8311
* be added after the primary link for the journal
8312
* to remain consistent.
8313
*/
8314
add_to_journal(&mkdir2->md_jaddref->ja_list);
8315
add_to_journal(&jaddref->ja_list);
8316
}
8317
/*
8318
* If we are adding a new directory remember this diradd so that if
8319
* we rename it we can keep the dot and dotdot dependencies. If
8320
* we are adding a new name for an inode that has a mkdiradd we
8321
* must be in rename and we have to move the dot and dotdot
8322
* dependencies to this new name. The old name is being orphaned
8323
* soon.
8324
*/
8325
if (mkdir1 != NULL) {
8326
if (inodedep->id_mkdiradd != NULL)
8327
panic("softdep_setup_directory_add: Existing mkdir");
8328
inodedep->id_mkdiradd = dap;
8329
} else if (inodedep->id_mkdiradd)
8330
merge_diradd(inodedep, dap);
8331
if (newdirblk) {
8332
/*
8333
* There is nothing to do if we are already tracking
8334
* this block.
8335
*/
3115
8336
if ((pagedep->pd_state & NEWBLOCK) != 0) {
3116
8337
WORKITEM_FREE(newdirblk, D_NEWDIRBLK);
3117
8338
FREE_LOCK(&lk);
3118
8339
return (0);
3119
8340
}
3120
/*
3121
* Find our associated allocdirect and have it track us.
3122
*/
3123
if (inodedep_lookup(mp, dp->i_number, 0, &inodedep) == 0)
3124
panic("softdep_setup_directory_add: lost inodedep");
3125
adp = TAILQ_LAST(&inodedep->id_newinoupdt, allocdirectlst);
3126
if (adp == NULL || adp->ad_lbn != lbn)
8341
if (newblk_lookup(mp, dbtofsb(fs, bp->b_blkno), 0, &newblk)
8342
== 0)
3127
8343
panic("softdep_setup_directory_add: lost entry");
8344
WORKLIST_INSERT(&newblk->nb_newdirblk, &newdirblk->db_list);
3128
8345
pagedep->pd_state |= NEWBLOCK;
8346
pagedep->pd_newdirblk = newdirblk;
3129
8347
newdirblk->db_pagedep = pagedep;
3130
WORKLIST_INSERT(&adp->ad_newdirblk, &newdirblk->db_list);
8348
FREE_LOCK(&lk);
8349
/*
8350
* If we extended into an indirect signal direnter to sync.
8351
*/
8352
if (isindir)
8353
return (1);
8354
return (0);
3131
8355
}
3132
8356
FREE_LOCK(&lk);
3133
8357
return (0);
3141
8365
* occur while the move is in progress.
3142
8366
*/
3143
8367
void
3144
softdep_change_directoryentry_offset(dp, base, oldloc, newloc, entrysize)
8368
softdep_change_directoryentry_offset(bp, dp, base, oldloc, newloc, entrysize)
8369
struct buf *bp; /* Buffer holding directory block. */
3145
8370
struct inode *dp; /* inode for directory */
3146
8371
caddr_t base; /* address of dp->i_offset */
3147
8372
caddr_t oldloc; /* address of old directory location */
3150
8375
{
3151
8376
int offset, oldoffset, newoffset;
3152
8377
struct pagedep *pagedep;
8378
struct jmvref *jmvref;
3153
8379
struct diradd *dap;
8380
struct direct *de;
8381
struct mount *mp;
3154
8382
ufs_lbn_t lbn;
8383
int flags;
3155
8384
3156
ACQUIRE_LOCK(&lk);
8385
mp = UFSTOVFS(dp->i_ump);
8386
de = (struct direct *)oldloc;
8387
jmvref = NULL;
8388
flags = 0;
8389
/*
8390
* Moves are always journaled as it would be too complex to
8391
* determine if any affected adds or removes are present in the
8392
* journal.
8393
*/
8394
if (MOUNTEDSUJ(mp)) {
8395
flags = DEPALLOC;
8396
jmvref = newjmvref(dp, de->d_ino,
8397
dp->i_offset + (oldloc - base),
8398
dp->i_offset + (newloc - base));
8399
}
3157
8400
lbn = lblkno(dp->i_fs, dp->i_offset);
3158
8401
offset = blkoff(dp->i_fs, dp->i_offset);
3159
if (pagedep_lookup(dp, lbn, 0, &pagedep) == 0)
3160
goto done;
3161
8402
oldoffset = offset + (oldloc - base);
3162
8403
newoffset = offset + (newloc - base);
3163
3164
LIST_FOREACH(dap, &pagedep->pd_diraddhd[DIRADDHASH(oldoffset)], da_pdlist) {
3165
if (dap->da_offset != oldoffset)
3166
continue;
8404
ACQUIRE_LOCK(&lk);
8405
if (pagedep_lookup(mp, bp, dp->i_number, lbn, flags, &pagedep) == 0)
8406
goto done;
8407
dap = diradd_lookup(pagedep, oldoffset);
8408
if (dap) {
3167
8409
dap->da_offset = newoffset;
3168
if (DIRADDHASH(newoffset) == DIRADDHASH(oldoffset))
3169
break;
3170
LIST_REMOVE(dap, da_pdlist);
3171
LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(newoffset)],
3172
dap, da_pdlist);
3173
break;
3174
}
3175
if (dap == NULL) {
3176
3177
LIST_FOREACH(dap, &pagedep->pd_pendinghd, da_pdlist) {
3178
if (dap->da_offset == oldoffset) {
3179
dap->da_offset = newoffset;
3180
break;
3181
}
8410
newoffset = DIRADDHASH(newoffset);
8411
oldoffset = DIRADDHASH(oldoffset);
8412
if ((dap->da_state & ALLCOMPLETE) != ALLCOMPLETE &&
8413
newoffset != oldoffset) {
8414
LIST_REMOVE(dap, da_pdlist);
8415
LIST_INSERT_HEAD(&pagedep->pd_diraddhd[newoffset],
8416
dap, da_pdlist);
3182
8417
}
3183
8418
}
3184
8419
done:
8420
if (jmvref) {
8421
jmvref->jm_pagedep = pagedep;
8422
LIST_INSERT_HEAD(&pagedep->pd_jmvrefhd, jmvref, jm_deps);
8423
add_to_journal(&jmvref->jm_list);
8424
}
3185
8425
bcopy(oldloc, newloc, entrysize);
3186
8426
FREE_LOCK(&lk);
3187
8427
}
3188
8428
3189
8429
/*
8430
* Move the mkdir dependencies and journal work from one diradd to another
8431
* when renaming a directory. The new name must depend on the mkdir deps
8432
* completing as the old name did. Directories can only have one valid link
8433
* at a time so one must be canonical.
8434
*/
8435
static void
8436
merge_diradd(inodedep, newdap)
8437
struct inodedep *inodedep;
8438
struct diradd *newdap;
8439
{
8440
struct diradd *olddap;
8441
struct mkdir *mkdir, *nextmd;
8442
short state;
8443
8444
olddap = inodedep->id_mkdiradd;
8445
inodedep->id_mkdiradd = newdap;
8446
if ((olddap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0) {
8447
newdap->da_state &= ~DEPCOMPLETE;
8448
for (mkdir = LIST_FIRST(&mkdirlisthd); mkdir; mkdir = nextmd) {
8449
nextmd = LIST_NEXT(mkdir, md_mkdirs);
8450
if (mkdir->md_diradd != olddap)
8451
continue;
8452
mkdir->md_diradd = newdap;
8453
state = mkdir->md_state & (MKDIR_PARENT | MKDIR_BODY);
8454
newdap->da_state |= state;
8455
olddap->da_state &= ~state;
8456
if ((olddap->da_state &
8457
(MKDIR_PARENT | MKDIR_BODY)) == 0)
8458
break;
8459
}
8460
if ((olddap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0)
8461
panic("merge_diradd: unfound ref");
8462
}
8463
/*
8464
* Any mkdir related journal items are not safe to be freed until
8465
* the new name is stable.
8466
*/
8467
jwork_move(&newdap->da_jwork, &olddap->da_jwork);
8468
olddap->da_state |= DEPCOMPLETE;
8469
complete_diradd(olddap);
8470
}
8471
8472
/*
8473
* Move the diradd to the pending list when all diradd dependencies are
8474
* complete.
8475
*/
8476
static void
8477
complete_diradd(dap)
8478
struct diradd *dap;
8479
{
8480
struct pagedep *pagedep;
8481
8482
if ((dap->da_state & ALLCOMPLETE) == ALLCOMPLETE) {
8483
if (dap->da_state & DIRCHG)
8484
pagedep = dap->da_previous->dm_pagedep;
8485
else
8486
pagedep = dap->da_pagedep;
8487
LIST_REMOVE(dap, da_pdlist);
8488
LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist);
8489
}
8490
}
8491
8492
/*
8493
* Cancel a diradd when a dirrem overlaps with it. We must cancel the journal
8494
* add entries and conditonally journal the remove.
8495
*/
8496
static void
8497
cancel_diradd(dap, dirrem, jremref, dotremref, dotdotremref)
8498
struct diradd *dap;
8499
struct dirrem *dirrem;
8500
struct jremref *jremref;
8501
struct jremref *dotremref;
8502
struct jremref *dotdotremref;
8503
{
8504
struct inodedep *inodedep;
8505
struct jaddref *jaddref;
8506
struct inoref *inoref;
8507
struct mkdir *mkdir;
8508
8509
/*
8510
* If no remove references were allocated we're on a non-journaled
8511
* filesystem and can skip the cancel step.
8512
*/
8513
if (jremref == NULL) {
8514
free_diradd(dap, NULL);
8515
return;
8516
}
8517
/*
8518
* Cancel the primary name an free it if it does not require
8519
* journaling.
8520
*/
8521
if (inodedep_lookup(dap->da_list.wk_mp, dap->da_newinum,
8522
0, &inodedep) != 0) {
8523
/* Abort the addref that reference this diradd. */
8524
TAILQ_FOREACH(inoref, &inodedep->id_inoreflst, if_deps) {
8525
if (inoref->if_list.wk_type != D_JADDREF)
8526
continue;
8527
jaddref = (struct jaddref *)inoref;
8528
if (jaddref->ja_diradd != dap)
8529
continue;
8530
if (cancel_jaddref(jaddref, inodedep,
8531
&dirrem->dm_jwork) == 0) {
8532
free_jremref(jremref);
8533
jremref = NULL;
8534
}
8535
break;
8536
}
8537
}
8538
/*
8539
* Cancel subordinate names and free them if they do not require
8540
* journaling.
8541
*/
8542
if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0) {
8543
LIST_FOREACH(mkdir, &mkdirlisthd, md_mkdirs) {
8544
if (mkdir->md_diradd != dap)
8545
continue;
8546
if ((jaddref = mkdir->md_jaddref) == NULL)
8547
continue;
8548
mkdir->md_jaddref = NULL;
8549
if (mkdir->md_state & MKDIR_PARENT) {
8550
if (cancel_jaddref(jaddref, NULL,
8551
&dirrem->dm_jwork) == 0) {
8552
free_jremref(dotdotremref);
8553
dotdotremref = NULL;
8554
}
8555
} else {
8556
if (cancel_jaddref(jaddref, inodedep,
8557
&dirrem->dm_jwork) == 0) {
8558
free_jremref(dotremref);
8559
dotremref = NULL;
8560
}
8561
}
8562
}
8563
}
8564
8565
if (jremref)
8566
journal_jremref(dirrem, jremref, inodedep);
8567
if (dotremref)
8568
journal_jremref(dirrem, dotremref, inodedep);
8569
if (dotdotremref)
8570
journal_jremref(dirrem, dotdotremref, NULL);
8571
jwork_move(&dirrem->dm_jwork, &dap->da_jwork);
8572
free_diradd(dap, &dirrem->dm_jwork);
8573
}
8574
8575
/*
3190
8576
* Free a diradd dependency structure. This routine must be called
3191
8577
* with splbio interrupts blocked.
3192
8578
*/
3193
8579
static void
3194
free_diradd(dap)
8580
free_diradd(dap, wkhd)
3195
8581
struct diradd *dap;
8582
struct workhead *wkhd;
3196
8583
{
3197
8584
struct dirrem *dirrem;
3198
8585
struct pagedep *pagedep;
3200
8587
struct mkdir *mkdir, *nextmd;
3201
8588
3202
8589
mtx_assert(&lk, MA_OWNED);
3203
WORKLIST_REMOVE(&dap->da_list);
3204
8590
LIST_REMOVE(dap, da_pdlist);
8591
if (dap->da_state & ONWORKLIST)
8592
WORKLIST_REMOVE(&dap->da_list);
3205
8593
if ((dap->da_state & DIRCHG) == 0) {
3206
8594
pagedep = dap->da_pagedep;
3207
8595
} else {
3208
8596
dirrem = dap->da_previous;
3209
8597
pagedep = dirrem->dm_pagedep;
3210
8598
dirrem->dm_dirinum = pagedep->pd_ino;
3211
add_to_worklist(&dirrem->dm_list);
8599
dirrem->dm_state |= COMPLETE;
8600
if (LIST_EMPTY(&dirrem->dm_jremrefhd))
8601
add_to_worklist(&dirrem->dm_list, 0);
3212
8602
}
3213
8603
if (inodedep_lookup(pagedep->pd_list.wk_mp, dap->da_newinum,
3214
8604
0, &inodedep) != 0)
3215
(void) free_inodedep(inodedep);
8605
if (inodedep->id_mkdiradd == dap)
8606
inodedep->id_mkdiradd = NULL;
3216
8607
if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0) {
3217
8608
for (mkdir = LIST_FIRST(&mkdirlisthd); mkdir; mkdir = nextmd) {
3218
8609
nextmd = LIST_NEXT(mkdir, md_mkdirs);
3219
8610
if (mkdir->md_diradd != dap)
3220
8611
continue;
3221
dap->da_state &= ~mkdir->md_state;
3222
WORKLIST_REMOVE(&mkdir->md_list);
8612
dap->da_state &=
8613
~(mkdir->md_state & (MKDIR_PARENT | MKDIR_BODY));
3223
8614
LIST_REMOVE(mkdir, md_mkdirs);
8615
if (mkdir->md_state & ONWORKLIST)
8616
WORKLIST_REMOVE(&mkdir->md_list);
8617
if (mkdir->md_jaddref != NULL)
8618
panic("free_diradd: Unexpected jaddref");
3224
8619
WORKITEM_FREE(mkdir, D_MKDIR);
8620
if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) == 0)
8621
break;
3225
8622
}
3226
8623
if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0)
3227
8624
panic("free_diradd: unfound ref");
3228
8625
}
8626
if (inodedep)
8627
free_inodedep(inodedep);
8628
/*
8629
* Free any journal segments waiting for the directory write.
8630
*/
8631
handle_jwork(&dap->da_jwork);
3229
8632
WORKITEM_FREE(dap, D_DIRADD);
3230
8633
}
3231
8634
3254
8657
int isrmdir; /* indicates if doing RMDIR */
3255
8658
{
3256
8659
struct dirrem *dirrem, *prevdirrem;
8660
struct inodedep *inodedep;
8661
int direct;
3257
8662
3258
8663
/*
3259
* Allocate a new dirrem if appropriate and ACQUIRE_LOCK.
8664
* Allocate a new dirrem if appropriate and ACQUIRE_LOCK. We want
8665
* newdirrem() to setup the full directory remove which requires
8666
* isrmdir > 1.
3260
8667
*/
3261
8668
dirrem = newdirrem(bp, dp, ip, isrmdir, &prevdirrem);
8669
/*
8670
* Add the dirrem to the inodedep's pending remove list for quick
8671
* discovery later.
8672
*/
8673
if (inodedep_lookup(UFSTOVFS(ip->i_ump), ip->i_number, 0,
8674
&inodedep) == 0)
8675
panic("softdep_setup_remove: Lost inodedep.");
8676
KASSERT((inodedep->id_state & UNLINKED) == 0, ("inode unlinked"));
8677
dirrem->dm_state |= ONDEPLIST;
8678
LIST_INSERT_HEAD(&inodedep->id_dirremhd, dirrem, dm_inonext);
3262
8679
3263
8680
/*
3264
8681
* If the COMPLETE flag is clear, then there were no active
3280
8697
LIST_INSERT_HEAD(&dirrem->dm_pagedep->pd_dirremhd,
3281
8698
prevdirrem, dm_next);
3282
8699
dirrem->dm_dirinum = dirrem->dm_pagedep->pd_ino;
8700
direct = LIST_EMPTY(&dirrem->dm_jremrefhd);
3283
8701
FREE_LOCK(&lk);
3284
handle_workitem_remove(dirrem, NULL);
3285
}
8702
if (direct)
8703
handle_workitem_remove(dirrem, 0);
8704
}
8705
}
8706
8707
/*
8708
* Check for an entry matching 'offset' on both the pd_dirraddhd list and the
8709
* pd_pendinghd list of a pagedep.
8710
*/
8711
static struct diradd *
8712
diradd_lookup(pagedep, offset)
8713
struct pagedep *pagedep;
8714
int offset;
8715
{
8716
struct diradd *dap;
8717
8718
LIST_FOREACH(dap, &pagedep->pd_diraddhd[DIRADDHASH(offset)], da_pdlist)
8719
if (dap->da_offset == offset)
8720
return (dap);
8721
LIST_FOREACH(dap, &pagedep->pd_pendinghd, da_pdlist)
8722
if (dap->da_offset == offset)
8723
return (dap);
8724
return (NULL);
8725
}
8726
8727
/*
8728
* Search for a .. diradd dependency in a directory that is being removed.
8729
* If the directory was renamed to a new parent we have a diradd rather
8730
* than a mkdir for the .. entry. We need to cancel it now before
8731
* it is found in truncate().
8732
*/
8733
static struct jremref *
8734
cancel_diradd_dotdot(ip, dirrem, jremref)
8735
struct inode *ip;
8736
struct dirrem *dirrem;
8737
struct jremref *jremref;
8738
{
8739
struct pagedep *pagedep;
8740
struct diradd *dap;
8741
struct worklist *wk;
8742
8743
if (pagedep_lookup(UFSTOVFS(ip->i_ump), NULL, ip->i_number, 0, 0,
8744
&pagedep) == 0)
8745
return (jremref);
8746
dap = diradd_lookup(pagedep, DOTDOT_OFFSET);
8747
if (dap == NULL)
8748
return (jremref);
8749
cancel_diradd(dap, dirrem, jremref, NULL, NULL);
8750
/*
8751
* Mark any journal work as belonging to the parent so it is freed
8752
* with the .. reference.
8753
*/
8754
LIST_FOREACH(wk, &dirrem->dm_jwork, wk_list)
8755
wk->wk_state |= MKDIR_PARENT;
8756
return (NULL);
8757
}
8758
8759
/*
8760
* Cancel the MKDIR_PARENT mkdir component of a diradd when we're going to
8761
* replace it with a dirrem/diradd pair as a result of re-parenting a
8762
* directory. This ensures that we don't simultaneously have a mkdir and
8763
* a diradd for the same .. entry.
8764
*/
8765
static struct jremref *
8766
cancel_mkdir_dotdot(ip, dirrem, jremref)
8767
struct inode *ip;
8768
struct dirrem *dirrem;
8769
struct jremref *jremref;
8770
{
8771
struct inodedep *inodedep;
8772
struct jaddref *jaddref;
8773
struct mkdir *mkdir;
8774
struct diradd *dap;
8775
8776
if (inodedep_lookup(UFSTOVFS(ip->i_ump), ip->i_number, 0,
8777
&inodedep) == 0)
8778
return (jremref);
8779
dap = inodedep->id_mkdiradd;
8780
if (dap == NULL || (dap->da_state & MKDIR_PARENT) == 0)
8781
return (jremref);
8782
for (mkdir = LIST_FIRST(&mkdirlisthd); mkdir;
8783
mkdir = LIST_NEXT(mkdir, md_mkdirs))
8784
if (mkdir->md_diradd == dap && mkdir->md_state & MKDIR_PARENT)
8785
break;
8786
if (mkdir == NULL)
8787
panic("cancel_mkdir_dotdot: Unable to find mkdir\n");
8788
if ((jaddref = mkdir->md_jaddref) != NULL) {
8789
mkdir->md_jaddref = NULL;
8790
jaddref->ja_state &= ~MKDIR_PARENT;
8791
if (inodedep_lookup(UFSTOVFS(ip->i_ump), jaddref->ja_ino, 0,
8792
&inodedep) == 0)
8793
panic("cancel_mkdir_dotdot: Lost parent inodedep");
8794
if (cancel_jaddref(jaddref, inodedep, &dirrem->dm_jwork)) {
8795
journal_jremref(dirrem, jremref, inodedep);
8796
jremref = NULL;
8797
}
8798
}
8799
if (mkdir->md_state & ONWORKLIST)
8800
WORKLIST_REMOVE(&mkdir->md_list);
8801
mkdir->md_state |= ALLCOMPLETE;
8802
complete_mkdir(mkdir);
8803
return (jremref);
8804
}
8805
8806
static void
8807
journal_jremref(dirrem, jremref, inodedep)
8808
struct dirrem *dirrem;
8809
struct jremref *jremref;
8810
struct inodedep *inodedep;
8811
{
8812
8813
if (inodedep == NULL)
8814
if (inodedep_lookup(jremref->jr_list.wk_mp,
8815
jremref->jr_ref.if_ino, 0, &inodedep) == 0)
8816
panic("journal_jremref: Lost inodedep");
8817
LIST_INSERT_HEAD(&dirrem->dm_jremrefhd, jremref, jr_deps);
8818
TAILQ_INSERT_TAIL(&inodedep->id_inoreflst, &jremref->jr_ref, if_deps);
8819
add_to_journal(&jremref->jr_list);
8820
}
8821
8822
static void
8823
dirrem_journal(dirrem, jremref, dotremref, dotdotremref)
8824
struct dirrem *dirrem;
8825
struct jremref *jremref;
8826
struct jremref *dotremref;
8827
struct jremref *dotdotremref;
8828
{
8829
struct inodedep *inodedep;
8830
8831
8832
if (inodedep_lookup(jremref->jr_list.wk_mp, jremref->jr_ref.if_ino, 0,
8833
&inodedep) == 0)
8834
panic("dirrem_journal: Lost inodedep");
8835
journal_jremref(dirrem, jremref, inodedep);
8836
if (dotremref)
8837
journal_jremref(dirrem, dotremref, inodedep);
8838
if (dotdotremref)
8839
journal_jremref(dirrem, dotdotremref, NULL);
3286
8840
}
3287
8841
3288
8842
/*
3289
8843
* Allocate a new dirrem if appropriate and return it along with
3290
8844
* its associated pagedep. Called without a lock, returns with lock.
3291
8845
*/
3292
static long num_dirrem; /* number of dirrem allocated */
3293
8846
static struct dirrem *
3294
8847
newdirrem(bp, dp, ip, isrmdir, prevdirremp)
3295
8848
struct buf *bp; /* buffer containing directory block */
3303
8856
struct diradd *dap;
3304
8857
struct dirrem *dirrem;
3305
8858
struct pagedep *pagedep;
8859
struct jremref *jremref;
8860
struct jremref *dotremref;
8861
struct jremref *dotdotremref;
8862
struct vnode *dvp;
3306
8863
3307
8864
/*
3308
8865
* Whiteouts have no deletion dependencies.
3309
8866
*/
3310
8867
if (ip == NULL)
3311
8868
panic("newdirrem: whiteout");
8869
dvp = ITOV(dp);
3312
8870
/*
3313
8871
* If we are over our limit, try to improve the situation.
3314
8872
* Limiting the number of dirrem structures will also limit
3315
8873
* the number of freefile and freeblks structures.
3316
8874
*/
3317
8875
ACQUIRE_LOCK(&lk);
3318
if (!(ip->i_flags & SF_SNAPSHOT) && num_dirrem > max_softdeps / 2)
3319
(void) request_cleanup(ITOV(dp)->v_mount, FLUSH_REMOVE);
3320
num_dirrem += 1;
8876
if (!IS_SNAPSHOT(ip) && dep_current[D_DIRREM] > max_softdeps / 2)
8877
(void) request_cleanup(ITOV(dp)->v_mount, FLUSH_BLOCKS);
3321
8878
FREE_LOCK(&lk);
3322
8879
dirrem = malloc(sizeof(struct dirrem),
3323
8880
M_DIRREM, M_SOFTDEP_FLAGS|M_ZERO);
3324
workitem_alloc(&dirrem->dm_list, D_DIRREM, ITOV(dp)->v_mount);
8881
workitem_alloc(&dirrem->dm_list, D_DIRREM, dvp->v_mount);
8882
LIST_INIT(&dirrem->dm_jremrefhd);
8883
LIST_INIT(&dirrem->dm_jwork);
3325
8884
dirrem->dm_state = isrmdir ? RMDIR : 0;
3326
8885
dirrem->dm_oldinum = ip->i_number;
3327
8886
*prevdirremp = NULL;
3328
8887
/*
8888
* Allocate remove reference structures to track journal write
8889
* dependencies. We will always have one for the link and
8890
* when doing directories we will always have one more for dot.
8891
* When renaming a directory we skip the dotdot link change so
8892
* this is not needed.
8893
*/
8894
jremref = dotremref = dotdotremref = NULL;
8895
if (DOINGSUJ(dvp)) {
8896
if (isrmdir) {
8897
jremref = newjremref(dirrem, dp, ip, dp->i_offset,
8898
ip->i_effnlink + 2);
8899
dotremref = newjremref(dirrem, ip, ip, DOT_OFFSET,
8900
ip->i_effnlink + 1);
8901
dotdotremref = newjremref(dirrem, ip, dp, DOTDOT_OFFSET,
8902
dp->i_effnlink + 1);
8903
dotdotremref->jr_state |= MKDIR_PARENT;
8904
} else
8905
jremref = newjremref(dirrem, dp, ip, dp->i_offset,
8906
ip->i_effnlink + 1);
8907
}
3329
8908
ACQUIRE_LOCK(&lk);
3330
8909
lbn = lblkno(dp->i_fs, dp->i_offset);
3331
8910
offset = blkoff(dp->i_fs, dp->i_offset);
3332
if (pagedep_lookup(dp, lbn, DEPALLOC, &pagedep) == 0)
3333
WORKLIST_INSERT(&bp->b_dep, &pagedep->pd_list);
8911
pagedep_lookup(UFSTOVFS(dp->i_ump), bp, dp->i_number, lbn, DEPALLOC,
8912
&pagedep);
3334
8913
dirrem->dm_pagedep = pagedep;
8914
dirrem->dm_offset = offset;
8915
/*
8916
* If we're renaming a .. link to a new directory, cancel any
8917
* existing MKDIR_PARENT mkdir. If it has already been canceled
8918
* the jremref is preserved for any potential diradd in this
8919
* location. This can not coincide with a rmdir.
8920
*/
8921
if (dp->i_offset == DOTDOT_OFFSET) {
8922
if (isrmdir)
8923
panic("newdirrem: .. directory change during remove?");
8924
jremref = cancel_mkdir_dotdot(dp, dirrem, jremref);
8925
}
8926
/*
8927
* If we're removing a directory search for the .. dependency now and
8928
* cancel it. Any pending journal work will be added to the dirrem
8929
* to be completed when the workitem remove completes.
8930
*/
8931
if (isrmdir)
8932
dotdotremref = cancel_diradd_dotdot(ip, dirrem, dotdotremref);
3335
8933
/*
3336
8934
* Check for a diradd dependency for the same directory entry.
3337
8935
* If present, then both dependencies become obsolete and can
3338
* be de-allocated. Check for an entry on both the pd_dirraddhd
3339
* list and the pd_pendinghd list.
8936
* be de-allocated.
3340
8937
*/
3341
3342
LIST_FOREACH(dap, &pagedep->pd_diraddhd[DIRADDHASH(offset)], da_pdlist)
3343
if (dap->da_offset == offset)
3344
break;
8938
dap = diradd_lookup(pagedep, offset);
3345
8939
if (dap == NULL) {
3346
3347
LIST_FOREACH(dap, &pagedep->pd_pendinghd, da_pdlist)
3348
if (dap->da_offset == offset)
3349
break;
3350
if (dap == NULL)
3351
return (dirrem);
8940
/*
8941
* Link the jremref structures into the dirrem so they are
8942
* written prior to the pagedep.
8943
*/
8944
if (jremref)
8945
dirrem_journal(dirrem, jremref, dotremref,
8946
dotdotremref);
8947
return (dirrem);
3352
8948
}
3353
8949
/*
3354
8950
* Must be ATTACHED at this point.
3373
8969
* Mark it COMPLETE so we can delete its inode immediately.
3374
8970
*/
3375
8971
dirrem->dm_state |= COMPLETE;
3376
free_diradd(dap);
8972
cancel_diradd(dap, dirrem, jremref, dotremref, dotdotremref);
8973
#ifdef SUJ_DEBUG
8974
if (isrmdir == 0) {
8975
struct worklist *wk;
8976
8977
LIST_FOREACH(wk, &dirrem->dm_jwork, wk_list)
8978
if (wk->wk_state & (MKDIR_BODY | MKDIR_PARENT))
8979
panic("bad wk %p (0x%X)\n", wk, wk->wk_state);
8980
}
8981
#endif
8982
3377
8983
return (dirrem);
3378
8984
}
3379
8985
3407
9013
struct dirrem *dirrem, *prevdirrem;
3408
9014
struct pagedep *pagedep;
3409
9015
struct inodedep *inodedep;
9016
struct jaddref *jaddref;
3410
9017
struct mount *mp;
3411
9018
3412
9019
offset = blkoff(dp->i_fs, dp->i_offset);
3422
9029
dap->da_state = DIRCHG | ATTACHED | DEPCOMPLETE;
3423
9030
dap->da_offset = offset;
3424
9031
dap->da_newinum = newinum;
9032
LIST_INIT(&dap->da_jwork);
3425
9033
}
3426
9034
3427
9035
/*
3454
9062
dm_next);
3455
9063
} else {
3456
9064
dirrem->dm_dirinum = pagedep->pd_ino;
3457
add_to_worklist(&dirrem->dm_list);
9065
if (LIST_EMPTY(&dirrem->dm_jremrefhd))
9066
add_to_worklist(&dirrem->dm_list, 0);
3458
9067
}
3459
9068
FREE_LOCK(&lk);
3460
9069
return;
3461
9070
}
9071
/*
9072
* Add the dirrem to the inodedep's pending remove list for quick
9073
* discovery later. A valid nlinkdelta ensures that this lookup
9074
* will not fail.
9075
*/
9076
if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) == 0)
9077
panic("softdep_setup_directory_change: Lost inodedep.");
9078
dirrem->dm_state |= ONDEPLIST;
9079
LIST_INSERT_HEAD(&inodedep->id_dirremhd, dirrem, dm_inonext);
3462
9080
3463
9081
/*
3464
9082
* If the COMPLETE flag is clear, then there were no active
3483
9101
dap->da_pagedep = pagedep;
3484
9102
}
3485
9103
dirrem->dm_dirinum = pagedep->pd_ino;
3486
add_to_worklist(&dirrem->dm_list);
9104
if (LIST_EMPTY(&dirrem->dm_jremrefhd))
9105
add_to_worklist(&dirrem->dm_list, 0);
3487
9106
}
3488
9107
/*
3489
* Link into its inodedep. Put it on the id_bufwait list if the inode
3490
* is not yet written. If it is written, do the post-inode write
3491
* processing to put it on the id_pendinghd list.
9108
* Lookup the jaddref for this journal entry. We must finish
9109
* initializing it and make the diradd write dependent on it.
9110
* If we're not journaling, put it on the id_bufwait list if the
9111
* inode is not yet written. If it is written, do the post-inode
9112
* write processing to put it on the id_pendinghd list.
3492
9113
*/
3493
if (inodedep_lookup(mp, newinum, DEPALLOC, &inodedep) == 0 ||
3494
(inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) {
9114
inodedep_lookup(mp, newinum, DEPALLOC | NODELAY, &inodedep);
9115
if (MOUNTEDSUJ(mp)) {
9116
jaddref = (struct jaddref *)TAILQ_LAST(&inodedep->id_inoreflst,
9117
inoreflst);
9118
KASSERT(jaddref != NULL && jaddref->ja_parent == dp->i_number,
9119
("softdep_setup_directory_change: bad jaddref %p",
9120
jaddref));
9121
jaddref->ja_diroff = dp->i_offset;
9122
jaddref->ja_diradd = dap;
9123
LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(offset)],
9124
dap, da_pdlist);
9125
add_to_journal(&jaddref->ja_list);
9126
} else if ((inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) {
3495
9127
dap->da_state |= COMPLETE;
3496
9128
LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist);
3497
9129
WORKLIST_INSERT(&inodedep->id_pendinghd, &dap->da_list);
3500
9132
dap, da_pdlist);
3501
9133
WORKLIST_INSERT(&inodedep->id_bufwait, &dap->da_list);
3502
9134
}
9135
/*
9136
* If we're making a new name for a directory that has not been
9137
* committed when need to move the dot and dotdot references to
9138
* this new name.
9139
*/
9140
if (inodedep->id_mkdiradd && dp->i_offset != DOTDOT_OFFSET)
9141
merge_diradd(inodedep, dap);
3503
9142
FREE_LOCK(&lk);
3504
9143
}
3505
9144
3514
9153
struct inode *ip; /* the inode with the increased link count */
3515
9154
{
3516
9155
struct inodedep *inodedep;
9156
int dflags;
3517
9157
3518
9158
ACQUIRE_LOCK(&lk);
3519
(void) inodedep_lookup(UFSTOVFS(ip->i_ump), ip->i_number,
3520
DEPALLOC, &inodedep);
9159
dflags = DEPALLOC;
9160
if (IS_SNAPSHOT(ip))
9161
dflags |= NODELAY;
9162
inodedep_lookup(UFSTOVFS(ip->i_ump), ip->i_number, dflags, &inodedep);
3521
9163
if (ip->i_nlink < ip->i_effnlink)
3522
9164
panic("softdep_change_linkcnt: bad delta");
3523
9165
inodedep->id_nlinkdelta = ip->i_nlink - ip->i_effnlink;
3525
9167
}
3526
9168
3527
9169
/*
3528
* Called when the effective link count and the reference count
3529
* on an inode drops to zero. At this point there are no names
3530
* referencing the file in the filesystem and no active file
3531
* references. The space associated with the file will be freed
3532
* as soon as the necessary soft dependencies are cleared.
9170
* Attach a sbdep dependency to the superblock buf so that we can keep
9171
* track of the head of the linked list of referenced but unlinked inodes.
3533
9172
*/
3534
9173
void
3535
softdep_releasefile(ip)
3536
struct inode *ip; /* inode with the zero effective link count */
9174
softdep_setup_sbupdate(ump, fs, bp)
9175
struct ufsmount *ump;
9176
struct fs *fs;
9177
struct buf *bp;
3537
9178
{
3538
struct inodedep *inodedep;
3539
struct fs *fs;
3540
int extblocks;
9179
struct sbdep *sbdep;
9180
struct worklist *wk;
3541
9181
3542
if (ip->i_effnlink > 0)
3543
panic("softdep_releasefile: file still referenced");
3544
/*
3545
* We may be called several times as the on-disk link count
3546
* drops to zero. We only want to account for the space once.
3547
*/
3548
if (ip->i_flag & IN_SPACECOUNTED)
3549
return;
3550
/*
3551
* We have to deactivate a snapshot otherwise copyonwrites may
3552
* add blocks and the cleanup may remove blocks after we have
3553
* tried to account for them.
3554
*/
3555
if ((ip->i_flags & SF_SNAPSHOT) != 0)
3556
ffs_snapremove(ITOV(ip));
3557
/*
3558
* If we are tracking an nlinkdelta, we have to also remember
3559
* whether we accounted for the freed space yet.
3560
*/
9182
if (MOUNTEDSUJ(UFSTOVFS(ump)) == 0)
9183
return;
9184
LIST_FOREACH(wk, &bp->b_dep, wk_list)
9185
if (wk->wk_type == D_SBDEP)
9186
break;
9187
if (wk != NULL)
9188
return;
9189
sbdep = malloc(sizeof(struct sbdep), M_SBDEP, M_SOFTDEP_FLAGS);
9190
workitem_alloc(&sbdep->sb_list, D_SBDEP, UFSTOVFS(ump));
9191
sbdep->sb_fs = fs;
9192
sbdep->sb_ump = ump;
3561
9193
ACQUIRE_LOCK(&lk);
3562
if ((inodedep_lookup(UFSTOVFS(ip->i_ump), ip->i_number, 0, &inodedep)))
3563
inodedep->id_state |= SPACECOUNTED;
9194
WORKLIST_INSERT(&bp->b_dep, &sbdep->sb_list);
3564
9195
FREE_LOCK(&lk);
3565
fs = ip->i_fs;
3566
extblocks = 0;
3567
if (fs->fs_magic == FS_UFS2_MAGIC)
3568
extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
3569
UFS_LOCK(ip->i_ump);
3570
ip->i_fs->fs_pendingblocks += DIP(ip, i_blocks) - extblocks;
3571
ip->i_fs->fs_pendinginodes += 1;
3572
UFS_UNLOCK(ip->i_ump);
3573
ip->i_flag |= IN_SPACECOUNTED;
9196
}
9197
9198
/*
9199
* Return the first unlinked inodedep which is ready to be the head of the
9200
* list. The inodedep and all those after it must have valid next pointers.
9201
*/
9202
static struct inodedep *
9203
first_unlinked_inodedep(ump)
9204
struct ufsmount *ump;
9205
{
9206
struct inodedep *inodedep;
9207
struct inodedep *idp;
9208
9209
mtx_assert(&lk, MA_OWNED);
9210
for (inodedep = TAILQ_LAST(&ump->softdep_unlinked, inodedeplst);
9211
inodedep; inodedep = idp) {
9212
if ((inodedep->id_state & UNLINKNEXT) == 0)
9213
return (NULL);
9214
idp = TAILQ_PREV(inodedep, inodedeplst, id_unlinked);
9215
if (idp == NULL || (idp->id_state & UNLINKNEXT) == 0)
9216
break;
9217
if ((inodedep->id_state & UNLINKPREV) == 0)
9218
break;
9219
}
9220
return (inodedep);
9221
}
9222
9223
/*
9224
* Set the sujfree unlinked head pointer prior to writing a superblock.
9225
*/
9226
static void
9227
initiate_write_sbdep(sbdep)
9228
struct sbdep *sbdep;
9229
{
9230
struct inodedep *inodedep;
9231
struct fs *bpfs;
9232
struct fs *fs;
9233
9234
bpfs = sbdep->sb_fs;
9235
fs = sbdep->sb_ump->um_fs;
9236
inodedep = first_unlinked_inodedep(sbdep->sb_ump);
9237
if (inodedep) {
9238
fs->fs_sujfree = inodedep->id_ino;
9239
inodedep->id_state |= UNLINKPREV;
9240
} else
9241
fs->fs_sujfree = 0;
9242
bpfs->fs_sujfree = fs->fs_sujfree;
9243
}
9244
9245
/*
9246
* After a superblock is written determine whether it must be written again
9247
* due to a changing unlinked list head.
9248
*/
9249
static int
9250
handle_written_sbdep(sbdep, bp)
9251
struct sbdep *sbdep;
9252
struct buf *bp;
9253
{
9254
struct inodedep *inodedep;
9255
struct mount *mp;
9256
struct fs *fs;
9257
9258
mtx_assert(&lk, MA_OWNED);
9259
fs = sbdep->sb_fs;
9260
mp = UFSTOVFS(sbdep->sb_ump);
9261
/*
9262
* If the superblock doesn't match the in-memory list start over.
9263
*/
9264
inodedep = first_unlinked_inodedep(sbdep->sb_ump);
9265
if ((inodedep && fs->fs_sujfree != inodedep->id_ino) ||
9266
(inodedep == NULL && fs->fs_sujfree != 0)) {
9267
bdirty(bp);
9268
return (1);
9269
}
9270
WORKITEM_FREE(sbdep, D_SBDEP);
9271
if (fs->fs_sujfree == 0)
9272
return (0);
9273
/*
9274
* Now that we have a record of this inode in stable store allow it
9275
* to be written to free up pending work. Inodes may see a lot of
9276
* write activity after they are unlinked which we must not hold up.
9277
*/
9278
for (; inodedep != NULL; inodedep = TAILQ_NEXT(inodedep, id_unlinked)) {
9279
if ((inodedep->id_state & UNLINKLINKS) != UNLINKLINKS)
9280
panic("handle_written_sbdep: Bad inodedep %p (0x%X)",
9281
inodedep, inodedep->id_state);
9282
if (inodedep->id_state & UNLINKONLIST)
9283
break;
9284
inodedep->id_state |= DEPCOMPLETE | UNLINKONLIST;
9285
}
9286
9287
return (0);
9288
}
9289
9290
/*
9291
* Mark an inodedep as unlinked and insert it into the in-memory unlinked list.
9292
*/
9293
static void
9294
unlinked_inodedep(mp, inodedep)
9295
struct mount *mp;
9296
struct inodedep *inodedep;
9297
{
9298
struct ufsmount *ump;
9299
9300
mtx_assert(&lk, MA_OWNED);
9301
if (MOUNTEDSUJ(mp) == 0)
9302
return;
9303
ump = VFSTOUFS(mp);
9304
ump->um_fs->fs_fmod = 1;
9305
if (inodedep->id_state & UNLINKED)
9306
panic("unlinked_inodedep: %p already unlinked\n", inodedep);
9307
inodedep->id_state |= UNLINKED;
9308
TAILQ_INSERT_HEAD(&ump->softdep_unlinked, inodedep, id_unlinked);
9309
}
9310
9311
/*
9312
* Remove an inodedep from the unlinked inodedep list. This may require
9313
* disk writes if the inode has made it that far.
9314
*/
9315
static void
9316
clear_unlinked_inodedep(inodedep)
9317
struct inodedep *inodedep;
9318
{
9319
struct ufsmount *ump;
9320
struct inodedep *idp;
9321
struct inodedep *idn;
9322
struct fs *fs;
9323
struct buf *bp;
9324
ino_t ino;
9325
ino_t nino;
9326
ino_t pino;
9327
int error;
9328
9329
ump = VFSTOUFS(inodedep->id_list.wk_mp);
9330
fs = ump->um_fs;
9331
ino = inodedep->id_ino;
9332
error = 0;
9333
for (;;) {
9334
mtx_assert(&lk, MA_OWNED);
9335
KASSERT((inodedep->id_state & UNLINKED) != 0,
9336
("clear_unlinked_inodedep: inodedep %p not unlinked",
9337
inodedep));
9338
/*
9339
* If nothing has yet been written simply remove us from
9340
* the in memory list and return. This is the most common
9341
* case where handle_workitem_remove() loses the final
9342
* reference.
9343
*/
9344
if ((inodedep->id_state & UNLINKLINKS) == 0)
9345
break;
9346
/*
9347
* If we have a NEXT pointer and no PREV pointer we can simply
9348
* clear NEXT's PREV and remove ourselves from the list. Be
9349
* careful not to clear PREV if the superblock points at
9350
* next as well.
9351
*/
9352
idn = TAILQ_NEXT(inodedep, id_unlinked);
9353
if ((inodedep->id_state & UNLINKLINKS) == UNLINKNEXT) {
9354
if (idn && fs->fs_sujfree != idn->id_ino)
9355
idn->id_state &= ~UNLINKPREV;
9356
break;
9357
}
9358
/*
9359
* Here we have an inodedep which is actually linked into
9360
* the list. We must remove it by forcing a write to the
9361
* link before us, whether it be the superblock or an inode.
9362
* Unfortunately the list may change while we're waiting
9363
* on the buf lock for either resource so we must loop until
9364
* we lock the right one. If both the superblock and an
9365
* inode point to this inode we must clear the inode first
9366
* followed by the superblock.
9367
*/
9368
idp = TAILQ_PREV(inodedep, inodedeplst, id_unlinked);
9369
pino = 0;
9370
if (idp && (idp->id_state & UNLINKNEXT))
9371
pino = idp->id_ino;
9372
FREE_LOCK(&lk);
9373
if (pino == 0) {
9374
bp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
9375
(int)fs->fs_sbsize, 0, 0, 0);
9376
} else {
9377
error = bread(ump->um_devvp,
9378
fsbtodb(fs, ino_to_fsba(fs, pino)),
9379
(int)fs->fs_bsize, NOCRED, &bp);
9380
if (error)
9381
brelse(bp);
9382
}
9383
ACQUIRE_LOCK(&lk);
9384
if (error)
9385
break;
9386
/* If the list has changed restart the loop. */
9387
idp = TAILQ_PREV(inodedep, inodedeplst, id_unlinked);
9388
nino = 0;
9389
if (idp && (idp->id_state & UNLINKNEXT))
9390
nino = idp->id_ino;
9391
if (nino != pino ||
9392
(inodedep->id_state & UNLINKPREV) != UNLINKPREV) {
9393
FREE_LOCK(&lk);
9394
brelse(bp);
9395
ACQUIRE_LOCK(&lk);
9396
continue;
9397
}
9398
nino = 0;
9399
idn = TAILQ_NEXT(inodedep, id_unlinked);
9400
if (idn)
9401
nino = idn->id_ino;
9402
/*
9403
* Remove us from the in memory list. After this we cannot
9404
* access the inodedep.
9405
*/
9406
KASSERT((inodedep->id_state & UNLINKED) != 0,
9407
("clear_unlinked_inodedep: inodedep %p not unlinked",
9408
inodedep));
9409
inodedep->id_state &= ~(UNLINKED | UNLINKLINKS | UNLINKONLIST);
9410
TAILQ_REMOVE(&ump->softdep_unlinked, inodedep, id_unlinked);
9411
FREE_LOCK(&lk);
9412
/*
9413
* The predecessor's next pointer is manually updated here
9414
* so that the NEXT flag is never cleared for an element
9415
* that is in the list.
9416
*/
9417
if (pino == 0) {
9418
bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
9419
ffs_oldfscompat_write((struct fs *)bp->b_data, ump);
9420
softdep_setup_sbupdate(ump, (struct fs *)bp->b_data,
9421
bp);
9422
} else if (fs->fs_magic == FS_UFS1_MAGIC)
9423
((struct ufs1_dinode *)bp->b_data +
9424
ino_to_fsbo(fs, pino))->di_freelink = nino;
9425
else
9426
((struct ufs2_dinode *)bp->b_data +
9427
ino_to_fsbo(fs, pino))->di_freelink = nino;
9428
/*
9429
* If the bwrite fails we have no recourse to recover. The
9430
* filesystem is corrupted already.
9431
*/
9432
bwrite(bp);
9433
ACQUIRE_LOCK(&lk);
9434
/*
9435
* If the superblock pointer still needs to be cleared force
9436
* a write here.
9437
*/
9438
if (fs->fs_sujfree == ino) {
9439
FREE_LOCK(&lk);
9440
bp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
9441
(int)fs->fs_sbsize, 0, 0, 0);
9442
bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
9443
ffs_oldfscompat_write((struct fs *)bp->b_data, ump);
9444
softdep_setup_sbupdate(ump, (struct fs *)bp->b_data,
9445
bp);
9446
bwrite(bp);
9447
ACQUIRE_LOCK(&lk);
9448
}
9449
9450
if (fs->fs_sujfree != ino)
9451
return;
9452
panic("clear_unlinked_inodedep: Failed to clear free head");
9453
}
9454
if (inodedep->id_ino == fs->fs_sujfree)
9455
panic("clear_unlinked_inodedep: Freeing head of free list");
9456
inodedep->id_state &= ~(UNLINKED | UNLINKLINKS | UNLINKONLIST);
9457
TAILQ_REMOVE(&ump->softdep_unlinked, inodedep, id_unlinked);
9458
return;
3574
9459
}
3575
9460
3576
9461
/*
3577
9462
* This workitem decrements the inode's link count.
3578
9463
* If the link count reaches zero, the file is removed.
3579
9464
*/
3580
static void
3581
handle_workitem_remove(dirrem, xp)
9465
static int
9466
handle_workitem_remove(dirrem, flags)
3582
9467
struct dirrem *dirrem;
3583
struct vnode *xp;
9468
int flags;
3584
9469
{
3585
struct thread *td = curthread;
3586
9470
struct inodedep *inodedep;
9471
struct workhead dotdotwk;
9472
struct worklist *wk;
9473
struct ufsmount *ump;
9474
struct mount *mp;
3587
9475
struct vnode *vp;
3588
9476
struct inode *ip;
3589
9477
ino_t oldinum;
3590
int error;
3591
9478
3592
if ((vp = xp) == NULL &&
3593
(error = ffs_vgetf(dirrem->dm_list.wk_mp,
3594
dirrem->dm_oldinum, LK_EXCLUSIVE, &vp, FFSV_FORCEINSMQ)) != 0) {
3595
softdep_error("handle_workitem_remove: vget", error);
3596
return;
3597
}
9479
if (dirrem->dm_state & ONWORKLIST)
9480
panic("handle_workitem_remove: dirrem %p still on worklist",
9481
dirrem);
9482
oldinum = dirrem->dm_oldinum;
9483
mp = dirrem->dm_list.wk_mp;
9484
ump = VFSTOUFS(mp);
9485
flags |= LK_EXCLUSIVE;
9486
if (ffs_vgetf(mp, oldinum, flags, &vp, FFSV_FORCEINSMQ) != 0)
9487
return (EBUSY);
3598
9488
ip = VTOI(vp);
3599
9489
ACQUIRE_LOCK(&lk);
3600
if ((inodedep_lookup(dirrem->dm_list.wk_mp,
3601
dirrem->dm_oldinum, 0, &inodedep)) == 0)
9490
if ((inodedep_lookup(mp, oldinum, 0, &inodedep)) == 0)
3602
9491
panic("handle_workitem_remove: lost inodedep");
9492
if (dirrem->dm_state & ONDEPLIST)
9493
LIST_REMOVE(dirrem, dm_inonext);
9494
KASSERT(LIST_EMPTY(&dirrem->dm_jremrefhd),
9495
("handle_workitem_remove: Journal entries not written."));
9496
9497
/*
9498
* Move all dependencies waiting on the remove to complete
9499
* from the dirrem to the inode inowait list to be completed
9500
* after the inode has been updated and written to disk. Any
9501
* marked MKDIR_PARENT are saved to be completed when the .. ref
9502
* is removed.
9503
*/
9504
LIST_INIT(&dotdotwk);
9505
while ((wk = LIST_FIRST(&dirrem->dm_jwork)) != NULL) {
9506
WORKLIST_REMOVE(wk);
9507
if (wk->wk_state & MKDIR_PARENT) {
9508
wk->wk_state &= ~MKDIR_PARENT;
9509
WORKLIST_INSERT(&dotdotwk, wk);
9510
continue;
9511
}
9512
WORKLIST_INSERT(&inodedep->id_inowait, wk);
9513
}
9514
LIST_SWAP(&dirrem->dm_jwork, &dotdotwk, worklist, wk_list);
3603
9515
/*
3604
9516
* Normal file deletion.
3605
9517
*/
3609
9521
ip->i_flag |= IN_CHANGE;
3610
9522
if (ip->i_nlink < ip->i_effnlink)
3611
9523
panic("handle_workitem_remove: bad file delta");
9524
if (ip->i_nlink == 0)
9525
unlinked_inodedep(mp, inodedep);
3612
9526
inodedep->id_nlinkdelta = ip->i_nlink - ip->i_effnlink;
3613
num_dirrem -= 1;
9527
KASSERT(LIST_EMPTY(&dirrem->dm_jwork),
9528
("handle_workitem_remove: worklist not empty. %s",
9529
TYPENAME(LIST_FIRST(&dirrem->dm_jwork)->wk_type)));
3614
9530
WORKITEM_FREE(dirrem, D_DIRREM);
3615
9531
FREE_LOCK(&lk);
3616
vput(vp);
3617
return;
9532
goto out;
3618
9533
}
3619
9534
/*
3620
9535
* Directory deletion. Decrement reference count for both the
3621
9536
* just deleted parent directory entry and the reference for ".".
3622
* Next truncate the directory to length zero. When the
3623
* truncation completes, arrange to have the reference count on
3624
* the parent decremented to account for the loss of "..".
9537
* Arrange to have the reference count on the parent decremented
9538
* to account for the loss of "..".
3625
9539
*/
3626
9540
ip->i_nlink -= 2;
3627
9541
DIP_SET(ip, i_nlink, ip->i_nlink);
3628
9542
ip->i_flag |= IN_CHANGE;
3629
9543
if (ip->i_nlink < ip->i_effnlink)
3630
9544
panic("handle_workitem_remove: bad dir delta");
9545
if (ip->i_nlink == 0)
9546
unlinked_inodedep(mp, inodedep);
3631
9547
inodedep->id_nlinkdelta = ip->i_nlink - ip->i_effnlink;
3632
FREE_LOCK(&lk);
3633
if ((error = ffs_truncate(vp, (off_t)0, 0, td->td_ucred, td)) != 0)
3634
softdep_error("handle_workitem_remove: truncate", error);
3635
ACQUIRE_LOCK(&lk);
3636
9548
/*
3637
9549
* Rename a directory to a new parent. Since, we are both deleting
3638
9550
* and creating a new directory entry, the link count on the new
3639
9551
* directory should not change. Thus we skip the followup dirrem.
3640
9552
*/
3641
9553
if (dirrem->dm_state & DIRCHG) {
3642
num_dirrem -= 1;
9554
KASSERT(LIST_EMPTY(&dirrem->dm_jwork),
9555
("handle_workitem_remove: DIRCHG and worklist not empty."));
3643
9556
WORKITEM_FREE(dirrem, D_DIRREM);
3644
9557
FREE_LOCK(&lk);
3645
vput(vp);
3646
return;
9558
goto out;
3647
9559
}
3648
/*
3649
* If the inodedep does not exist, then the zero'ed inode has
3650
* been written to disk. If the allocated inode has never been
3651
* written to disk, then the on-disk inode is zero'ed. In either
3652
* case we can remove the file immediately.
3653
*/
3654
dirrem->dm_state = 0;
3655
oldinum = dirrem->dm_oldinum;
9560
dirrem->dm_state = ONDEPLIST;
3656
9561
dirrem->dm_oldinum = dirrem->dm_dirinum;
3657
if (inodedep_lookup(dirrem->dm_list.wk_mp, oldinum,
3658
0, &inodedep) == 0 || check_inode_unwritten(inodedep)) {
3659
if (xp != NULL)
3660
add_to_worklist(&dirrem->dm_list);
9562
/*
9563
* Place the dirrem on the parent's diremhd list.
9564
*/
9565
if (inodedep_lookup(mp, dirrem->dm_oldinum, 0, &inodedep) == 0)
9566
panic("handle_workitem_remove: lost dir inodedep");
9567
LIST_INSERT_HEAD(&inodedep->id_dirremhd, dirrem, dm_inonext);
9568
/*
9569
* If the allocated inode has never been written to disk, then
9570
* the on-disk inode is zero'ed and we can remove the file
9571
* immediately. When journaling if the inode has been marked
9572
* unlinked and not DEPCOMPLETE we know it can never be written.
9573
*/
9574
inodedep_lookup(mp, oldinum, 0, &inodedep);
9575
if (inodedep == NULL ||
9576
(inodedep->id_state & (DEPCOMPLETE | UNLINKED)) == UNLINKED ||
9577
check_inode_unwritten(inodedep)) {
3661
9578
FREE_LOCK(&lk);
3662
9579
vput(vp);
3663
if (xp == NULL)
3664
handle_workitem_remove(dirrem, NULL);
3665
return;
9580
return handle_workitem_remove(dirrem, flags);
3666
9581
}
3667
9582
WORKLIST_INSERT(&inodedep->id_inowait, &dirrem->dm_list);
3668
9583
FREE_LOCK(&lk);
3669
9584
ip->i_flag |= IN_CHANGE;
9585
out:
3670
9586
ffs_update(vp, 0);
3671
9587
vput(vp);
9588
return (0);
3672
9589
}
3673
9590
3674
9591
/*
3689
9606
handle_workitem_freefile(freefile)
3690
9607
struct freefile *freefile;
3691
9608
{
9609
struct workhead wkhd;
3692
9610
struct fs *fs;
3693
9611
struct inodedep *idp;
3694
9612
struct ufsmount *ump;
3701
9619
error = inodedep_lookup(UFSTOVFS(ump), freefile->fx_oldinum, 0, &idp);
3702
9620
FREE_LOCK(&lk);
3703
9621
if (error)
3704
panic("handle_workitem_freefile: inodedep survived");
9622
panic("handle_workitem_freefile: inodedep %p survived", idp);
3705
9623
#endif
3706
9624
UFS_LOCK(ump);
3707
9625
fs->fs_pendinginodes -= 1;
3708
9626
UFS_UNLOCK(ump);
9627
LIST_INIT(&wkhd);
9628
LIST_SWAP(&freefile->fx_jwork, &wkhd, worklist, wk_list);
3709
9629
if ((error = ffs_freefile(ump, fs, freefile->fx_devvp,
3710
freefile->fx_oldinum, freefile->fx_mode)) != 0)
9630
freefile->fx_oldinum, freefile->fx_mode, &wkhd)) != 0)
3711
9631
softdep_error("handle_workitem_freefile", error);
3712
9632
ACQUIRE_LOCK(&lk);
3713
9633
WORKITEM_FREE(freefile, D_FREEFILE);
3757
9677
{
3758
9678
struct worklist *wk;
3759
9679
struct worklist marker;
3760
struct indirdep *indirdep;
3761
9680
struct inodedep *inodedep;
9681
struct freeblks *freeblks;
9682
struct jblkdep *jblkdep;
9683
struct newblk *newblk;
3762
9684
3763
9685
/*
3764
9686
* We only care about write operations. There should never
3767
9689
if (bp->b_iocmd != BIO_WRITE)
3768
9690
panic("softdep_disk_io_initiation: not write");
3769
9691
9692
if (bp->b_vflags & BV_BKGRDINPROG)
9693
panic("softdep_disk_io_initiation: Writing buffer with "
9694
"background write in progress: %p", bp);
9695
3770
9696
marker.wk_type = D_LAST + 1; /* Not a normal workitem */
3771
9697
PHOLD(curproc); /* Don't swap out kernel stack */
3772
9698
3792
9718
continue;
3793
9719
3794
9720
case D_INDIRDEP:
3795
indirdep = WK_INDIRDEP(wk);
3796
if (indirdep->ir_state & GOINGAWAY)
3797
panic("disk_io_initiation: indirdep gone");
3798
/*
3799
* If there are no remaining dependencies, this
3800
* will be writing the real pointers, so the
3801
* dependency can be freed.
3802
*/
3803
if (LIST_EMPTY(&indirdep->ir_deplisthd)) {
3804
struct buf *bp;
3805
3806
bp = indirdep->ir_savebp;
3807
bp->b_flags |= B_INVAL | B_NOCACHE;
3808
/* inline expand WORKLIST_REMOVE(wk); */
3809
wk->wk_state &= ~ONWORKLIST;
3810
LIST_REMOVE(wk, wk_list);
3811
WORKITEM_FREE(indirdep, D_INDIRDEP);
3812
FREE_LOCK(&lk);
3813
brelse(bp);
3814
ACQUIRE_LOCK(&lk);
3815
continue;
3816
}
3817
/*
3818
* Replace up-to-date version with safe version.
3819
*/
3820
FREE_LOCK(&lk);
3821
indirdep->ir_saveddata = malloc(bp->b_bcount,
3822
M_INDIRDEP, M_SOFTDEP_FLAGS);
3823
ACQUIRE_LOCK(&lk);
3824
indirdep->ir_state &= ~ATTACHED;
3825
indirdep->ir_state |= UNDONE;
3826
bcopy(bp->b_data, indirdep->ir_saveddata, bp->b_bcount);
3827
bcopy(indirdep->ir_savebp->b_data, bp->b_data,
3828
bp->b_bcount);
9721
initiate_write_indirdep(WK_INDIRDEP(wk), bp);
3829
9722
continue;
3830
9723
3831
case D_MKDIR:
3832
9724
case D_BMSAFEMAP:
9725
initiate_write_bmsafemap(WK_BMSAFEMAP(wk), bp);
9726
continue;
9727
9728
case D_JSEG:
9729
WK_JSEG(wk)->js_buf = NULL;
9730
continue;
9731
9732
case D_FREEBLKS:
9733
freeblks = WK_FREEBLKS(wk);
9734
jblkdep = LIST_FIRST(&freeblks->fb_jblkdephd);
9735
/*
9736
* We have to wait for the freeblks to be journaled
9737
* before we can write an inodeblock with updated
9738
* pointers. Be careful to arrange the marker so
9739
* we revisit the freeblks if it's not removed by
9740
* the first jwait().
9741
*/
9742
if (jblkdep != NULL) {
9743
LIST_REMOVE(&marker, wk_list);
9744
LIST_INSERT_BEFORE(wk, &marker, wk_list);
9745
jwait(&jblkdep->jb_list, MNT_WAIT);
9746
}
9747
continue;
3833
9748
case D_ALLOCDIRECT:
3834
9749
case D_ALLOCINDIR:
9750
/*
9751
* We have to wait for the jnewblk to be journaled
9752
* before we can write to a block if the contents
9753
* may be confused with an earlier file's indirect
9754
* at recovery time. Handle the marker as described
9755
* above.
9756
*/
9757
newblk = WK_NEWBLK(wk);
9758
if (newblk->nb_jnewblk != NULL &&
9759
indirblk_lookup(newblk->nb_list.wk_mp,
9760
newblk->nb_newblkno)) {
9761
LIST_REMOVE(&marker, wk_list);
9762
LIST_INSERT_BEFORE(wk, &marker, wk_list);
9763
jwait(&newblk->nb_jnewblk->jn_list, MNT_WAIT);
9764
}
9765
continue;
9766
9767
case D_SBDEP:
9768
initiate_write_sbdep(WK_SBDEP(wk));
9769
continue;
9770
9771
case D_MKDIR:
9772
case D_FREEWORK:
9773
case D_FREEDEP:
9774
case D_JSEGDEP:
3835
9775
continue;
3836
9776
3837
9777
default:
3855
9795
struct pagedep *pagedep;
3856
9796
struct buf *bp;
3857
9797
{
9798
struct jremref *jremref;
9799
struct jmvref *jmvref;
9800
struct dirrem *dirrem;
3858
9801
struct diradd *dap;
3859
9802
struct direct *ep;
3860
9803
int i;
3869
9812
return;
3870
9813
}
3871
9814
pagedep->pd_state |= IOSTARTED;
9815
/*
9816
* Wait for all journal remove dependencies to hit the disk.
9817
* We can not allow any potentially conflicting directory adds
9818
* to be visible before removes and rollback is too difficult.
9819
* lk may be dropped and re-acquired, however we hold the buf
9820
* locked so the dependency can not go away.
9821
*/
9822
LIST_FOREACH(dirrem, &pagedep->pd_dirremhd, dm_next)
9823
while ((jremref = LIST_FIRST(&dirrem->dm_jremrefhd)) != NULL)
9824
jwait(&jremref->jr_list, MNT_WAIT);
9825
while ((jmvref = LIST_FIRST(&pagedep->pd_jmvrefhd)) != NULL)
9826
jwait(&jmvref->jm_list, MNT_WAIT);
3872
9827
for (i = 0; i < DAHASHSZ; i++) {
3873
9828
LIST_FOREACH(dap, &pagedep->pd_diraddhd[i], da_pdlist) {
3874
9829
ep = (struct direct *)
3905
9860
struct allocdirect *adp, *lastadp;
3906
9861
struct ufs1_dinode *dp;
3907
9862
struct ufs1_dinode *sip;
9863
struct inoref *inoref;
3908
9864
struct fs *fs;
3909
9865
ufs_lbn_t i;
3910
9866
#ifdef INVARIANTS
3918
9874
fs = inodedep->id_fs;
3919
9875
dp = (struct ufs1_dinode *)bp->b_data +
3920
9876
ino_to_fsbo(fs, inodedep->id_ino);
9877
9878
/*
9879
* If we're on the unlinked list but have not yet written our
9880
* next pointer initialize it here.
9881
*/
9882
if ((inodedep->id_state & (UNLINKED | UNLINKNEXT)) == UNLINKED) {
9883
struct inodedep *inon;
9884
9885
inon = TAILQ_NEXT(inodedep, id_unlinked);
9886
dp->di_freelink = inon ? inon->id_ino : 0;
9887
}
3921
9888
/*
3922
9889
* If the bitmap is not yet written, then the allocated
3923
9890
* inode cannot be written to disk.
3933
9900
*inodedep->id_savedino1 = *dp;
3934
9901
bzero((caddr_t)dp, sizeof(struct ufs1_dinode));
3935
9902
dp->di_gen = inodedep->id_savedino1->di_gen;
9903
dp->di_freelink = inodedep->id_savedino1->di_freelink;
3936
9904
return;
3937
9905
}
3938
9906
/*
3940
9908
*/
3941
9909
inodedep->id_savedsize = dp->di_size;
3942
9910
inodedep->id_savedextsize = 0;
3943
if (TAILQ_EMPTY(&inodedep->id_inoupdt))
9911
inodedep->id_savednlink = dp->di_nlink;
9912
if (TAILQ_EMPTY(&inodedep->id_inoupdt) &&
9913
TAILQ_EMPTY(&inodedep->id_inoreflst))
3944
9914
return;
3945
9915
/*
9916
* Revert the link count to that of the first unwritten journal entry.
9917
*/
9918
inoref = TAILQ_FIRST(&inodedep->id_inoreflst);
9919
if (inoref)
9920
dp->di_nlink = inoref->if_nlink;
9921
/*
3946
9922
* Set the dependencies to busy.
3947
9923
*/
3948
9924
for (deplist = 0, adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp;
3949
9925
adp = TAILQ_NEXT(adp, ad_next)) {
3950
9926
#ifdef INVARIANTS
3951
if (deplist != 0 && prevlbn >= adp->ad_lbn)
9927
if (deplist != 0 && prevlbn >= adp->ad_offset)
3952
9928
panic("softdep_write_inodeblock: lbn order");
3953
prevlbn = adp->ad_lbn;
3954
if (adp->ad_lbn < NDADDR &&
3955
dp->di_db[adp->ad_lbn] != adp->ad_newblkno)
9929
prevlbn = adp->ad_offset;
9930
if (adp->ad_offset < NDADDR &&
9931
dp->di_db[adp->ad_offset] != adp->ad_newblkno)
3956
9932
panic("%s: direct pointer #%jd mismatch %d != %jd",
3957
9933
"softdep_write_inodeblock",
3958
(intmax_t)adp->ad_lbn,
3959
dp->di_db[adp->ad_lbn],
9934
(intmax_t)adp->ad_offset,
9935
dp->di_db[adp->ad_offset],
3960
9936
(intmax_t)adp->ad_newblkno);
3961
if (adp->ad_lbn >= NDADDR &&
3962
dp->di_ib[adp->ad_lbn - NDADDR] != adp->ad_newblkno)
9937
if (adp->ad_offset >= NDADDR &&
9938
dp->di_ib[adp->ad_offset - NDADDR] != adp->ad_newblkno)
3963
9939
panic("%s: indirect pointer #%jd mismatch %d != %jd",
3964
9940
"softdep_write_inodeblock",
3965
(intmax_t)adp->ad_lbn - NDADDR,
3966
dp->di_ib[adp->ad_lbn - NDADDR],
9941
(intmax_t)adp->ad_offset - NDADDR,
9942
dp->di_ib[adp->ad_offset - NDADDR],
3967
9943
(intmax_t)adp->ad_newblkno);
3968
deplist |= 1 << adp->ad_lbn;
9944
deplist |= 1 << adp->ad_offset;
3969
9945
if ((adp->ad_state & ATTACHED) == 0)
3970
9946
panic("softdep_write_inodeblock: Unknown state 0x%x",
3971
9947
adp->ad_state);
3981
9957
*/
3982
9958
for (lastadp = NULL, adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp;
3983
9959
lastadp = adp, adp = TAILQ_NEXT(adp, ad_next)) {
3984
if (adp->ad_lbn >= NDADDR)
9960
if (adp->ad_offset >= NDADDR)
3985
9961
break;
3986
dp->di_db[adp->ad_lbn] = adp->ad_oldblkno;
9962
dp->di_db[adp->ad_offset] = adp->ad_oldblkno;
3987
9963
/* keep going until hitting a rollback to a frag */
3988
9964
if (adp->ad_oldsize == 0 || adp->ad_oldsize == fs->fs_bsize)
3989
9965
continue;
3990
dp->di_size = fs->fs_bsize * adp->ad_lbn + adp->ad_oldsize;
3991
for (i = adp->ad_lbn + 1; i < NDADDR; i++) {
9966
dp->di_size = fs->fs_bsize * adp->ad_offset + adp->ad_oldsize;
9967
for (i = adp->ad_offset + 1; i < NDADDR; i++) {
3992
9968
#ifdef INVARIANTS
3993
9969
if (dp->di_db[i] != 0 && (deplist & (1 << i)) == 0)
3994
9970
panic("softdep_write_inodeblock: lost dep1");
4012
9988
* we already checked for fragments in the loop above.
4013
9989
*/
4014
9990
if (lastadp != NULL &&
4015
dp->di_size <= (lastadp->ad_lbn + 1) * fs->fs_bsize) {
4016
for (i = lastadp->ad_lbn; i >= 0; i--)
9991
dp->di_size <= (lastadp->ad_offset + 1) * fs->fs_bsize) {
9992
for (i = lastadp->ad_offset; i >= 0; i--)
4017
9993
if (dp->di_db[i] != 0)
4018
9994
break;
4019
9995
dp->di_size = (i + 1) * fs->fs_bsize;
4030
10006
* postpone fsck, we are stuck with this argument.
4031
10007
*/
4032
10008
for (; adp; adp = TAILQ_NEXT(adp, ad_next))
4033
dp->di_ib[adp->ad_lbn - NDADDR] = 0;
10009
dp->di_ib[adp->ad_offset - NDADDR] = 0;
4034
10010
}
4035
10011
4036
10012
/*
4051
10027
struct allocdirect *adp, *lastadp;
4052
10028
struct ufs2_dinode *dp;
4053
10029
struct ufs2_dinode *sip;
10030
struct inoref *inoref;
4054
10031
struct fs *fs;
4055
10032
ufs_lbn_t i;
4056
10033
#ifdef INVARIANTS
4064
10041
fs = inodedep->id_fs;
4065
10042
dp = (struct ufs2_dinode *)bp->b_data +
4066
10043
ino_to_fsbo(fs, inodedep->id_ino);
10044
10045
/*
10046
* If we're on the unlinked list but have not yet written our
10047
* next pointer initialize it here.
10048
*/
10049
if ((inodedep->id_state & (UNLINKED | UNLINKNEXT)) == UNLINKED) {
10050
struct inodedep *inon;
10051
10052
inon = TAILQ_NEXT(inodedep, id_unlinked);
10053
dp->di_freelink = inon ? inon->id_ino : 0;
10054
}
4067
10055
/*
4068
10056
* If the bitmap is not yet written, then the allocated
4069
10057
* inode cannot be written to disk.
4079
10067
*inodedep->id_savedino2 = *dp;
4080
10068
bzero((caddr_t)dp, sizeof(struct ufs2_dinode));
4081
10069
dp->di_gen = inodedep->id_savedino2->di_gen;
10070
dp->di_freelink = inodedep->id_savedino2->di_freelink;
4082
10071
return;
4083
10072
}
4084
10073
/*
4086
10075
*/
4087
10076
inodedep->id_savedsize = dp->di_size;
4088
10077
inodedep->id_savedextsize = dp->di_extsize;
10078
inodedep->id_savednlink = dp->di_nlink;
4089
10079
if (TAILQ_EMPTY(&inodedep->id_inoupdt) &&
4090
TAILQ_EMPTY(&inodedep->id_extupdt))
10080
TAILQ_EMPTY(&inodedep->id_extupdt) &&
10081
TAILQ_EMPTY(&inodedep->id_inoreflst))
4091
10082
return;
4092
10083
/*
10084
* Revert the link count to that of the first unwritten journal entry.
10085
*/
10086
inoref = TAILQ_FIRST(&inodedep->id_inoreflst);
10087
if (inoref)
10088
dp->di_nlink = inoref->if_nlink;
10089
10090
/*
4093
10091
* Set the ext data dependencies to busy.
4094
10092
*/
4095
10093
for (deplist = 0, adp = TAILQ_FIRST(&inodedep->id_extupdt); adp;
4096
10094
adp = TAILQ_NEXT(adp, ad_next)) {
4097
10095
#ifdef INVARIANTS
4098
if (deplist != 0 && prevlbn >= adp->ad_lbn)
10096
if (deplist != 0 && prevlbn >= adp->ad_offset)
4099
10097
panic("softdep_write_inodeblock: lbn order");
4100
prevlbn = adp->ad_lbn;
4101
if (dp->di_extb[adp->ad_lbn] != adp->ad_newblkno)
10098
prevlbn = adp->ad_offset;
10099
if (dp->di_extb[adp->ad_offset] != adp->ad_newblkno)
4102
10100
panic("%s: direct pointer #%jd mismatch %jd != %jd",
4103
10101
"softdep_write_inodeblock",
4104
(intmax_t)adp->ad_lbn,
4105
(intmax_t)dp->di_extb[adp->ad_lbn],
10102
(intmax_t)adp->ad_offset,
10103
(intmax_t)dp->di_extb[adp->ad_offset],
4106
10104
(intmax_t)adp->ad_newblkno);
4107
deplist |= 1 << adp->ad_lbn;
10105
deplist |= 1 << adp->ad_offset;
4108
10106
if ((adp->ad_state & ATTACHED) == 0)
4109
10107
panic("softdep_write_inodeblock: Unknown state 0x%x",
4110
10108
adp->ad_state);
4120
10118
*/
4121
10119
for (lastadp = NULL, adp = TAILQ_FIRST(&inodedep->id_extupdt); adp;
4122
10120
lastadp = adp, adp = TAILQ_NEXT(adp, ad_next)) {
4123
dp->di_extb[adp->ad_lbn] = adp->ad_oldblkno;
10121
dp->di_extb[adp->ad_offset] = adp->ad_oldblkno;
4124
10122
/* keep going until hitting a rollback to a frag */
4125
10123
if (adp->ad_oldsize == 0 || adp->ad_oldsize == fs->fs_bsize)
4126
10124
continue;
4127
dp->di_extsize = fs->fs_bsize * adp->ad_lbn + adp->ad_oldsize;
4128
for (i = adp->ad_lbn + 1; i < NXADDR; i++) {
10125
dp->di_extsize = fs->fs_bsize * adp->ad_offset + adp->ad_oldsize;
10126
for (i = adp->ad_offset + 1; i < NXADDR; i++) {
4129
10127
#ifdef INVARIANTS
4130
10128
if (dp->di_extb[i] != 0 && (deplist & (1 << i)) == 0)
4131
10129
panic("softdep_write_inodeblock: lost dep1");
4142
10140
* we already checked for fragments in the loop above.
4143
10141
*/
4144
10142
if (lastadp != NULL &&
4145
dp->di_extsize <= (lastadp->ad_lbn + 1) * fs->fs_bsize) {
4146
for (i = lastadp->ad_lbn; i >= 0; i--)
10143
dp->di_extsize <= (lastadp->ad_offset + 1) * fs->fs_bsize) {
10144
for (i = lastadp->ad_offset; i >= 0; i--)
4147
10145
if (dp->di_extb[i] != 0)
4148
10146
break;
4149
10147
dp->di_extsize = (i + 1) * fs->fs_bsize;
4154
10152
for (deplist = 0, adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp;
4155
10153
adp = TAILQ_NEXT(adp, ad_next)) {
4156
10154
#ifdef INVARIANTS
4157
if (deplist != 0 && prevlbn >= adp->ad_lbn)
10155
if (deplist != 0 && prevlbn >= adp->ad_offset)
4158
10156
panic("softdep_write_inodeblock: lbn order");
4159
prevlbn = adp->ad_lbn;
4160
if (adp->ad_lbn < NDADDR &&
4161
dp->di_db[adp->ad_lbn] != adp->ad_newblkno)
10157
if ((adp->ad_state & ATTACHED) == 0)
10158
panic("inodedep %p and adp %p not attached", inodedep, adp);
10159
prevlbn = adp->ad_offset;
10160
if (adp->ad_offset < NDADDR &&
10161
dp->di_db[adp->ad_offset] != adp->ad_newblkno)
4162
10162
panic("%s: direct pointer #%jd mismatch %jd != %jd",
4163
10163
"softdep_write_inodeblock",
4164
(intmax_t)adp->ad_lbn,
4165
(intmax_t)dp->di_db[adp->ad_lbn],
10164
(intmax_t)adp->ad_offset,
10165
(intmax_t)dp->di_db[adp->ad_offset],
4166
10166
(intmax_t)adp->ad_newblkno);
4167
if (adp->ad_lbn >= NDADDR &&
4168
dp->di_ib[adp->ad_lbn - NDADDR] != adp->ad_newblkno)
10167
if (adp->ad_offset >= NDADDR &&
10168
dp->di_ib[adp->ad_offset - NDADDR] != adp->ad_newblkno)
4169
10169
panic("%s indirect pointer #%jd mismatch %jd != %jd",
4170
10170
"softdep_write_inodeblock:",
4171
(intmax_t)adp->ad_lbn - NDADDR,
4172
(intmax_t)dp->di_ib[adp->ad_lbn - NDADDR],
10171
(intmax_t)adp->ad_offset - NDADDR,
10172
(intmax_t)dp->di_ib[adp->ad_offset - NDADDR],
4173
10173
(intmax_t)adp->ad_newblkno);
4174
deplist |= 1 << adp->ad_lbn;
10174
deplist |= 1 << adp->ad_offset;
4175
10175
if ((adp->ad_state & ATTACHED) == 0)
4176
10176
panic("softdep_write_inodeblock: Unknown state 0x%x",
4177
10177
adp->ad_state);
4187
10187
*/
4188
10188
for (lastadp = NULL, adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp;
4189
10189
lastadp = adp, adp = TAILQ_NEXT(adp, ad_next)) {
4190
if (adp->ad_lbn >= NDADDR)
10190
if (adp->ad_offset >= NDADDR)
4191
10191
break;
4192
dp->di_db[adp->ad_lbn] = adp->ad_oldblkno;
10192
dp->di_db[adp->ad_offset] = adp->ad_oldblkno;
4193
10193
/* keep going until hitting a rollback to a frag */
4194
10194
if (adp->ad_oldsize == 0 || adp->ad_oldsize == fs->fs_bsize)
4195
10195
continue;
4196
dp->di_size = fs->fs_bsize * adp->ad_lbn + adp->ad_oldsize;
4197
for (i = adp->ad_lbn + 1; i < NDADDR; i++) {
10196
dp->di_size = fs->fs_bsize * adp->ad_offset + adp->ad_oldsize;
10197
for (i = adp->ad_offset + 1; i < NDADDR; i++) {
4198
10198
#ifdef INVARIANTS
4199
10199
if (dp->di_db[i] != 0 && (deplist & (1 << i)) == 0)
4200
10200
panic("softdep_write_inodeblock: lost dep2");
4218
10218
* we already checked for fragments in the loop above.
4219
10219
*/
4220
10220
if (lastadp != NULL &&
4221
dp->di_size <= (lastadp->ad_lbn + 1) * fs->fs_bsize) {
4222
for (i = lastadp->ad_lbn; i >= 0; i--)
10221
dp->di_size <= (lastadp->ad_offset + 1) * fs->fs_bsize) {
10222
for (i = lastadp->ad_offset; i >= 0; i--)
4223
10223
if (dp->di_db[i] != 0)
4224
10224
break;
4225
10225
dp->di_size = (i + 1) * fs->fs_bsize;
4236
10236
* postpone fsck, we are stuck with this argument.
4237
10237
*/
4238
10238
for (; adp; adp = TAILQ_NEXT(adp, ad_next))
4239
dp->di_ib[adp->ad_lbn - NDADDR] = 0;
10239
dp->di_ib[adp->ad_offset - NDADDR] = 0;
10240
}
10241
10242
/*
10243
* Cancel an indirdep as a result of truncation. Release all of the
10244
* children allocindirs and place their journal work on the appropriate
10245
* list.
10246
*/
10247
static void
10248
cancel_indirdep(indirdep, bp, freeblks)
10249
struct indirdep *indirdep;
10250
struct buf *bp;
10251
struct freeblks *freeblks;
10252
{
10253
struct allocindir *aip;
10254
10255
/*
10256
* None of the indirect pointers will ever be visible,
10257
* so they can simply be tossed. GOINGAWAY ensures
10258
* that allocated pointers will be saved in the buffer
10259
* cache until they are freed. Note that they will
10260
* only be able to be found by their physical address
10261
* since the inode mapping the logical address will
10262
* be gone. The save buffer used for the safe copy
10263
* was allocated in setup_allocindir_phase2 using
10264
* the physical address so it could be used for this
10265
* purpose. Hence we swap the safe copy with the real
10266
* copy, allowing the safe copy to be freed and holding
10267
* on to the real copy for later use in indir_trunc.
10268
*/
10269
if (indirdep->ir_state & GOINGAWAY)
10270
panic("cancel_indirdep: already gone");
10271
if ((indirdep->ir_state & DEPCOMPLETE) == 0) {
10272
indirdep->ir_state |= DEPCOMPLETE;
10273
LIST_REMOVE(indirdep, ir_next);
10274
}
10275
indirdep->ir_state |= GOINGAWAY;
10276
VFSTOUFS(indirdep->ir_list.wk_mp)->um_numindirdeps += 1;
10277
/*
10278
* Pass in bp for blocks still have journal writes
10279
* pending so we can cancel them on their own.
10280
*/
10281
while ((aip = LIST_FIRST(&indirdep->ir_deplisthd)) != 0)
10282
cancel_allocindir(aip, bp, freeblks, 0);
10283
while ((aip = LIST_FIRST(&indirdep->ir_donehd)) != 0)
10284
cancel_allocindir(aip, NULL, freeblks, 0);
10285
while ((aip = LIST_FIRST(&indirdep->ir_writehd)) != 0)
10286
cancel_allocindir(aip, NULL, freeblks, 0);
10287
while ((aip = LIST_FIRST(&indirdep->ir_completehd)) != 0)
10288
cancel_allocindir(aip, NULL, freeblks, 0);
10289
/*
10290
* If there are pending partial truncations we need to keep the
10291
* old block copy around until they complete. This is because
10292
* the current b_data is not a perfect superset of the available
10293
* blocks.
10294
*/
10295
if (TAILQ_EMPTY(&indirdep->ir_trunc))
10296
bcopy(bp->b_data, indirdep->ir_savebp->b_data, bp->b_bcount);
10297
else
10298
bcopy(bp->b_data, indirdep->ir_saveddata, bp->b_bcount);
10299
WORKLIST_REMOVE(&indirdep->ir_list);
10300
WORKLIST_INSERT(&indirdep->ir_savebp->b_dep, &indirdep->ir_list);
10301
indirdep->ir_bp = NULL;
10302
indirdep->ir_freeblks = freeblks;
10303
}
10304
10305
/*
10306
* Free an indirdep once it no longer has new pointers to track.
10307
*/
10308
static void
10309
free_indirdep(indirdep)
10310
struct indirdep *indirdep;
10311
{
10312
10313
KASSERT(TAILQ_EMPTY(&indirdep->ir_trunc),
10314
("free_indirdep: Indir trunc list not empty."));
10315
KASSERT(LIST_EMPTY(&indirdep->ir_completehd),
10316
("free_indirdep: Complete head not empty."));
10317
KASSERT(LIST_EMPTY(&indirdep->ir_writehd),
10318
("free_indirdep: write head not empty."));
10319
KASSERT(LIST_EMPTY(&indirdep->ir_donehd),
10320
("free_indirdep: done head not empty."));
10321
KASSERT(LIST_EMPTY(&indirdep->ir_deplisthd),
10322
("free_indirdep: deplist head not empty."));
10323
KASSERT((indirdep->ir_state & DEPCOMPLETE),
10324
("free_indirdep: %p still on newblk list.", indirdep));
10325
KASSERT(indirdep->ir_saveddata == NULL,
10326
("free_indirdep: %p still has saved data.", indirdep));
10327
if (indirdep->ir_state & ONWORKLIST)
10328
WORKLIST_REMOVE(&indirdep->ir_list);
10329
WORKITEM_FREE(indirdep, D_INDIRDEP);
10330
}
10331
10332
/*
10333
* Called before a write to an indirdep. This routine is responsible for
10334
* rolling back pointers to a safe state which includes only those
10335
* allocindirs which have been completed.
10336
*/
10337
static void
10338
initiate_write_indirdep(indirdep, bp)
10339
struct indirdep *indirdep;
10340
struct buf *bp;
10341
{
10342
10343
indirdep->ir_state |= IOSTARTED;
10344
if (indirdep->ir_state & GOINGAWAY)
10345
panic("disk_io_initiation: indirdep gone");
10346
/*
10347
* If there are no remaining dependencies, this will be writing
10348
* the real pointers.
10349
*/
10350
if (LIST_EMPTY(&indirdep->ir_deplisthd) &&
10351
TAILQ_EMPTY(&indirdep->ir_trunc))
10352
return;
10353
/*
10354
* Replace up-to-date version with safe version.
10355
*/
10356
if (indirdep->ir_saveddata == NULL) {
10357
FREE_LOCK(&lk);
10358
indirdep->ir_saveddata = malloc(bp->b_bcount, M_INDIRDEP,
10359
M_SOFTDEP_FLAGS);
10360
ACQUIRE_LOCK(&lk);
10361
}
10362
indirdep->ir_state &= ~ATTACHED;
10363
indirdep->ir_state |= UNDONE;
10364
bcopy(bp->b_data, indirdep->ir_saveddata, bp->b_bcount);
10365
bcopy(indirdep->ir_savebp->b_data, bp->b_data,
10366
bp->b_bcount);
10367
}
10368
10369
/*
10370
* Called when an inode has been cleared in a cg bitmap. This finally
10371
* eliminates any canceled jaddrefs
10372
*/
10373
void
10374
softdep_setup_inofree(mp, bp, ino, wkhd)
10375
struct mount *mp;
10376
struct buf *bp;
10377
ino_t ino;
10378
struct workhead *wkhd;
10379
{
10380
struct worklist *wk, *wkn;
10381
struct inodedep *inodedep;
10382
uint8_t *inosused;
10383
struct cg *cgp;
10384
struct fs *fs;
10385
10386
ACQUIRE_LOCK(&lk);
10387
fs = VFSTOUFS(mp)->um_fs;
10388
cgp = (struct cg *)bp->b_data;
10389
inosused = cg_inosused(cgp);
10390
if (isset(inosused, ino % fs->fs_ipg))
10391
panic("softdep_setup_inofree: inode %d not freed.", ino);
10392
if (inodedep_lookup(mp, ino, 0, &inodedep))
10393
panic("softdep_setup_inofree: ino %d has existing inodedep %p",
10394
ino, inodedep);
10395
if (wkhd) {
10396
LIST_FOREACH_SAFE(wk, wkhd, wk_list, wkn) {
10397
if (wk->wk_type != D_JADDREF)
10398
continue;
10399
WORKLIST_REMOVE(wk);
10400
/*
10401
* We can free immediately even if the jaddref
10402
* isn't attached in a background write as now
10403
* the bitmaps are reconciled.
10404
*/
10405
wk->wk_state |= COMPLETE | ATTACHED;
10406
free_jaddref(WK_JADDREF(wk));
10407
}
10408
jwork_move(&bp->b_dep, wkhd);
10409
}
10410
FREE_LOCK(&lk);
10411
}
10412
10413
10414
/*
10415
* Called via ffs_blkfree() after a set of frags has been cleared from a cg
10416
* map. Any dependencies waiting for the write to clear are added to the
10417
* buf's list and any jnewblks that are being canceled are discarded
10418
* immediately.
10419
*/
10420
void
10421
softdep_setup_blkfree(mp, bp, blkno, frags, wkhd)
10422
struct mount *mp;
10423
struct buf *bp;
10424
ufs2_daddr_t blkno;
10425
int frags;
10426
struct workhead *wkhd;
10427
{
10428
struct bmsafemap *bmsafemap;
10429
struct jnewblk *jnewblk;
10430
struct worklist *wk;
10431
struct fs *fs;
10432
#ifdef SUJ_DEBUG
10433
uint8_t *blksfree;
10434
struct cg *cgp;
10435
ufs2_daddr_t jstart;
10436
ufs2_daddr_t jend;
10437
ufs2_daddr_t end;
10438
long bno;
10439
int i;
10440
#endif
10441
10442
CTR3(KTR_SUJ,
10443
"softdep_setup_blkfree: blkno %jd frags %d wk head %p",
10444
blkno, frags, wkhd);
10445
10446
ACQUIRE_LOCK(&lk);
10447
/* Lookup the bmsafemap so we track when it is dirty. */
10448
fs = VFSTOUFS(mp)->um_fs;
10449
bmsafemap = bmsafemap_lookup(mp, bp, dtog(fs, blkno), NULL);
10450
/*
10451
* Detach any jnewblks which have been canceled. They must linger
10452
* until the bitmap is cleared again by ffs_blkfree() to prevent
10453
* an unjournaled allocation from hitting the disk.
10454
*/
10455
if (wkhd) {
10456
while ((wk = LIST_FIRST(wkhd)) != NULL) {
10457
CTR2(KTR_SUJ,
10458
"softdep_setup_blkfree: blkno %jd wk type %d",
10459
blkno, wk->wk_type);
10460
WORKLIST_REMOVE(wk);
10461
if (wk->wk_type != D_JNEWBLK) {
10462
WORKLIST_INSERT(&bmsafemap->sm_freehd, wk);
10463
continue;
10464
}
10465
jnewblk = WK_JNEWBLK(wk);
10466
KASSERT(jnewblk->jn_state & GOINGAWAY,
10467
("softdep_setup_blkfree: jnewblk not canceled."));
10468
#ifdef SUJ_DEBUG
10469
/*
10470
* Assert that this block is free in the bitmap
10471
* before we discard the jnewblk.
10472
*/
10473
cgp = (struct cg *)bp->b_data;
10474
blksfree = cg_blksfree(cgp);
10475
bno = dtogd(fs, jnewblk->jn_blkno);
10476
for (i = jnewblk->jn_oldfrags;
10477
i < jnewblk->jn_frags; i++) {
10478
if (isset(blksfree, bno + i))
10479
continue;
10480
panic("softdep_setup_blkfree: not free");
10481
}
10482
#endif
10483
/*
10484
* Even if it's not attached we can free immediately
10485
* as the new bitmap is correct.
10486
*/
10487
wk->wk_state |= COMPLETE | ATTACHED;
10488
free_jnewblk(jnewblk);
10489
}
10490
}
10491
10492
#ifdef SUJ_DEBUG
10493
/*
10494
* Assert that we are not freeing a block which has an outstanding
10495
* allocation dependency.
10496
*/
10497
fs = VFSTOUFS(mp)->um_fs;
10498
bmsafemap = bmsafemap_lookup(mp, bp, dtog(fs, blkno), NULL);
10499
end = blkno + frags;
10500
LIST_FOREACH(jnewblk, &bmsafemap->sm_jnewblkhd, jn_deps) {
10501
/*
10502
* Don't match against blocks that will be freed when the
10503
* background write is done.
10504
*/
10505
if ((jnewblk->jn_state & (ATTACHED | COMPLETE | DEPCOMPLETE)) ==
10506
(COMPLETE | DEPCOMPLETE))
10507
continue;
10508
jstart = jnewblk->jn_blkno + jnewblk->jn_oldfrags;
10509
jend = jnewblk->jn_blkno + jnewblk->jn_frags;
10510
if ((blkno >= jstart && blkno < jend) ||
10511
(end > jstart && end <= jend)) {
10512
printf("state 0x%X %jd - %d %d dep %p\n",
10513
jnewblk->jn_state, jnewblk->jn_blkno,
10514
jnewblk->jn_oldfrags, jnewblk->jn_frags,
10515
jnewblk->jn_dep);
10516
panic("softdep_setup_blkfree: "
10517
"%jd-%jd(%d) overlaps with %jd-%jd",
10518
blkno, end, frags, jstart, jend);
10519
}
10520
}
10521
#endif
10522
FREE_LOCK(&lk);
10523
}
10524
10525
/*
10526
* Revert a block allocation when the journal record that describes it
10527
* is not yet written.
10528
*/
10529
int
10530
jnewblk_rollback(jnewblk, fs, cgp, blksfree)
10531
struct jnewblk *jnewblk;
10532
struct fs *fs;
10533
struct cg *cgp;
10534
uint8_t *blksfree;
10535
{
10536
ufs1_daddr_t fragno;
10537
long cgbno, bbase;
10538
int frags, blk;
10539
int i;
10540
10541
frags = 0;
10542
cgbno = dtogd(fs, jnewblk->jn_blkno);
10543
/*
10544
* We have to test which frags need to be rolled back. We may
10545
* be operating on a stale copy when doing background writes.
10546
*/
10547
for (i = jnewblk->jn_oldfrags; i < jnewblk->jn_frags; i++)
10548
if (isclr(blksfree, cgbno + i))
10549
frags++;
10550
if (frags == 0)
10551
return (0);
10552
/*
10553
* This is mostly ffs_blkfree() sans some validation and
10554
* superblock updates.
10555
*/
10556
if (frags == fs->fs_frag) {
10557
fragno = fragstoblks(fs, cgbno);
10558
ffs_setblock(fs, blksfree, fragno);
10559
ffs_clusteracct(fs, cgp, fragno, 1);
10560
cgp->cg_cs.cs_nbfree++;
10561
} else {
10562
cgbno += jnewblk->jn_oldfrags;
10563
bbase = cgbno - fragnum(fs, cgbno);
10564
/* Decrement the old frags. */
10565
blk = blkmap(fs, blksfree, bbase);
10566
ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
10567
/* Deallocate the fragment */
10568
for (i = 0; i < frags; i++)
10569
setbit(blksfree, cgbno + i);
10570
cgp->cg_cs.cs_nffree += frags;
10571
/* Add back in counts associated with the new frags */
10572
blk = blkmap(fs, blksfree, bbase);
10573
ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
10574
/* If a complete block has been reassembled, account for it. */
10575
fragno = fragstoblks(fs, bbase);
10576
if (ffs_isblock(fs, blksfree, fragno)) {
10577
cgp->cg_cs.cs_nffree -= fs->fs_frag;
10578
ffs_clusteracct(fs, cgp, fragno, 1);
10579
cgp->cg_cs.cs_nbfree++;
10580
}
10581
}
10582
stat_jnewblk++;
10583
jnewblk->jn_state &= ~ATTACHED;
10584
jnewblk->jn_state |= UNDONE;
10585
10586
return (frags);
10587
}
10588
10589
static void
10590
initiate_write_bmsafemap(bmsafemap, bp)
10591
struct bmsafemap *bmsafemap;
10592
struct buf *bp; /* The cg block. */
10593
{
10594
struct jaddref *jaddref;
10595
struct jnewblk *jnewblk;
10596
uint8_t *inosused;
10597
uint8_t *blksfree;
10598
struct cg *cgp;
10599
struct fs *fs;
10600
ino_t ino;
10601
10602
if (bmsafemap->sm_state & IOSTARTED)
10603
return;
10604
bmsafemap->sm_state |= IOSTARTED;
10605
/*
10606
* Clear any inode allocations which are pending journal writes.
10607
*/
10608
if (LIST_FIRST(&bmsafemap->sm_jaddrefhd) != NULL) {
10609
cgp = (struct cg *)bp->b_data;
10610
fs = VFSTOUFS(bmsafemap->sm_list.wk_mp)->um_fs;
10611
inosused = cg_inosused(cgp);
10612
LIST_FOREACH(jaddref, &bmsafemap->sm_jaddrefhd, ja_bmdeps) {
10613
ino = jaddref->ja_ino % fs->fs_ipg;
10614
if (isset(inosused, ino)) {
10615
if ((jaddref->ja_mode & IFMT) == IFDIR)
10616
cgp->cg_cs.cs_ndir--;
10617
cgp->cg_cs.cs_nifree++;
10618
clrbit(inosused, ino);
10619
jaddref->ja_state &= ~ATTACHED;
10620
jaddref->ja_state |= UNDONE;
10621
stat_jaddref++;
10622
} else
10623
panic("initiate_write_bmsafemap: inode %d "
10624
"marked free", jaddref->ja_ino);
10625
}
10626
}
10627
/*
10628
* Clear any block allocations which are pending journal writes.
10629
*/
10630
if (LIST_FIRST(&bmsafemap->sm_jnewblkhd) != NULL) {
10631
cgp = (struct cg *)bp->b_data;
10632
fs = VFSTOUFS(bmsafemap->sm_list.wk_mp)->um_fs;
10633
blksfree = cg_blksfree(cgp);
10634
LIST_FOREACH(jnewblk, &bmsafemap->sm_jnewblkhd, jn_deps) {
10635
if (jnewblk_rollback(jnewblk, fs, cgp, blksfree))
10636
continue;
10637
panic("initiate_write_bmsafemap: block %jd "
10638
"marked free", jnewblk->jn_blkno);
10639
}
10640
}
10641
/*
10642
* Move allocation lists to the written lists so they can be
10643
* cleared once the block write is complete.
10644
*/
10645
LIST_SWAP(&bmsafemap->sm_inodedephd, &bmsafemap->sm_inodedepwr,
10646
inodedep, id_deps);
10647
LIST_SWAP(&bmsafemap->sm_newblkhd, &bmsafemap->sm_newblkwr,
10648
newblk, nb_deps);
10649
LIST_SWAP(&bmsafemap->sm_freehd, &bmsafemap->sm_freewr, worklist,
10650
wk_list);
4240
10651
}
4241
10652
4242
10653
/*
4246
10657
* a request completion). It should be called early in this
4247
10658
* procedure, before the block is made available to other
4248
10659
* processes or other routines are called.
10660
*
4249
10661
*/
4250
10662
static void
4251
10663
softdep_disk_write_complete(bp)
4254
10666
struct worklist *wk;
4255
10667
struct worklist *owk;
4256
10668
struct workhead reattach;
4257
struct newblk *newblk;
4258
struct allocindir *aip;
4259
struct allocdirect *adp;
4260
struct indirdep *indirdep;
4261
struct inodedep *inodedep;
4262
struct bmsafemap *bmsafemap;
10669
struct freeblks *freeblks;
10670
struct buf *sbp;
4263
10671
4264
10672
/*
4265
10673
* If an error occurred while doing the write, then the data
4271
10679
/*
4272
10680
* This lock must not be released anywhere in this code segment.
4273
10681
*/
4274
ACQUIRE_LOCK(&lk);
10682
sbp = NULL;
4275
10683
owk = NULL;
10684
ACQUIRE_LOCK(&lk);
4276
10685
while ((wk = LIST_FIRST(&bp->b_dep)) != NULL) {
4277
10686
WORKLIST_REMOVE(wk);
10687
dep_write[wk->wk_type]++;
4278
10688
if (wk == owk)
4279
10689
panic("duplicate worklist: %p\n", wk);
4280
10690
owk = wk;
4291
10701
continue;
4292
10702
4293
10703
case D_BMSAFEMAP:
4294
bmsafemap = WK_BMSAFEMAP(wk);
4295
while ((newblk = LIST_FIRST(&bmsafemap->sm_newblkhd))) {
4296
newblk->nb_state |= DEPCOMPLETE;
4297
newblk->nb_bmsafemap = NULL;
4298
LIST_REMOVE(newblk, nb_deps);
4299
}
4300
while ((adp =
4301
LIST_FIRST(&bmsafemap->sm_allocdirecthd))) {
4302
adp->ad_state |= DEPCOMPLETE;
4303
adp->ad_buf = NULL;
4304
LIST_REMOVE(adp, ad_deps);
4305
handle_allocdirect_partdone(adp);
4306
}
4307
while ((aip =
4308
LIST_FIRST(&bmsafemap->sm_allocindirhd))) {
4309
aip->ai_state |= DEPCOMPLETE;
4310
aip->ai_buf = NULL;
4311
LIST_REMOVE(aip, ai_deps);
4312
handle_allocindir_partdone(aip);
4313
}
4314
while ((inodedep =
4315
LIST_FIRST(&bmsafemap->sm_inodedephd)) != NULL) {
4316
inodedep->id_state |= DEPCOMPLETE;
4317
LIST_REMOVE(inodedep, id_deps);
4318
inodedep->id_buf = NULL;
4319
}
4320
WORKITEM_FREE(bmsafemap, D_BMSAFEMAP);
10704
if (handle_written_bmsafemap(WK_BMSAFEMAP(wk), bp))
10705
WORKLIST_INSERT(&reattach, wk);
4321
10706
continue;
4322
10707
4323
10708
case D_MKDIR:
4325
10710
continue;
4326
10711
4327
10712
case D_ALLOCDIRECT:
4328
adp = WK_ALLOCDIRECT(wk);
4329
adp->ad_state |= COMPLETE;
4330
handle_allocdirect_partdone(adp);
10713
wk->wk_state |= COMPLETE;
10714
handle_allocdirect_partdone(WK_ALLOCDIRECT(wk), NULL);
4331
10715
continue;
4332
10716
4333
10717
case D_ALLOCINDIR:
4334
aip = WK_ALLOCINDIR(wk);
4335
aip->ai_state |= COMPLETE;
4336
handle_allocindir_partdone(aip);
10718
wk->wk_state |= COMPLETE;
10719
handle_allocindir_partdone(WK_ALLOCINDIR(wk));
4337
10720
continue;
4338
10721
4339
10722
case D_INDIRDEP:
4340
indirdep = WK_INDIRDEP(wk);
4341
if (indirdep->ir_state & GOINGAWAY)
4342
panic("disk_write_complete: indirdep gone");
4343
bcopy(indirdep->ir_saveddata, bp->b_data, bp->b_bcount);
4344
free(indirdep->ir_saveddata, M_INDIRDEP);
4345
indirdep->ir_saveddata = 0;
4346
indirdep->ir_state &= ~UNDONE;
4347
indirdep->ir_state |= ATTACHED;
4348
while ((aip = LIST_FIRST(&indirdep->ir_donehd)) != 0) {
4349
handle_allocindir_partdone(aip);
4350
if (aip == LIST_FIRST(&indirdep->ir_donehd))
4351
panic("disk_write_complete: not gone");
4352
}
4353
WORKLIST_INSERT(&reattach, wk);
4354
if ((bp->b_flags & B_DELWRI) == 0)
4355
stat_indir_blk_ptrs++;
4356
bdirty(bp);
10723
if (handle_written_indirdep(WK_INDIRDEP(wk), bp, &sbp))
10724
WORKLIST_INSERT(&reattach, wk);
10725
continue;
10726
10727
case D_FREEBLKS:
10728
wk->wk_state |= COMPLETE;
10729
freeblks = WK_FREEBLKS(wk);
10730
if ((wk->wk_state & ALLCOMPLETE) == ALLCOMPLETE &&
10731
LIST_EMPTY(&freeblks->fb_jblkdephd))
10732
add_to_worklist(wk, WK_NODELAY);
10733
continue;
10734
10735
case D_FREEWORK:
10736
handle_written_freework(WK_FREEWORK(wk));
10737
break;
10738
10739
case D_JSEGDEP:
10740
free_jsegdep(WK_JSEGDEP(wk));
10741
continue;
10742
10743
case D_JSEG:
10744
handle_written_jseg(WK_JSEG(wk), bp);
10745
continue;
10746
10747
case D_SBDEP:
10748
if (handle_written_sbdep(WK_SBDEP(wk), bp))
10749
WORKLIST_INSERT(&reattach, wk);
10750
continue;
10751
10752
case D_FREEDEP:
10753
free_freedep(WK_FREEDEP(wk));
4357
10754
continue;
4358
10755
4359
10756
default:
4370
10767
WORKLIST_INSERT(&bp->b_dep, wk);
4371
10768
}
4372
10769
FREE_LOCK(&lk);
10770
if (sbp)
10771
brelse(sbp);
4373
10772
}
4374
10773
4375
10774
/*
4378
10777
* splbio interrupts blocked.
4379
10778
*/
4380
10779
static void
4381
handle_allocdirect_partdone(adp)
10780
handle_allocdirect_partdone(adp, wkhd)
4382
10781
struct allocdirect *adp; /* the completed allocdirect */
10782
struct workhead *wkhd; /* Work to do when inode is writtne. */
4383
10783
{
4384
10784
struct allocdirectlst *listhead;
4385
10785
struct allocdirect *listadp;
4386
10786
struct inodedep *inodedep;
4387
long bsize, delay;
10787
long bsize;
4388
10788
4389
10789
if ((adp->ad_state & ALLCOMPLETE) != ALLCOMPLETE)
4390
10790
return;
4391
if (adp->ad_buf != NULL)
4392
panic("handle_allocdirect_partdone: dangling dep");
4393
10791
/*
4394
10792
* The on-disk inode cannot claim to be any larger than the last
4395
10793
* fragment that has been written. Otherwise, the on-disk inode
4439
10837
return;
4440
10838
}
4441
10839
/*
4442
* If we have found the just finished dependency, then free
10840
* If we have found the just finished dependency, then queue
4443
10841
* it along with anything that follows it that is complete.
4444
* If the inode still has a bitmap dependency, then it has
4445
* never been written to disk, hence the on-disk inode cannot
4446
* reference the old fragment so we can free it without delay.
10842
* Since the pointer has not yet been written in the inode
10843
* as the dependency prevents it, place the allocdirect on the
10844
* bufwait list where it will be freed once the pointer is
10845
* valid.
4447
10846
*/
4448
delay = (inodedep->id_state & DEPCOMPLETE);
10847
if (wkhd == NULL)
10848
wkhd = &inodedep->id_bufwait;
4449
10849
for (; adp; adp = listadp) {
4450
10850
listadp = TAILQ_NEXT(adp, ad_next);
4451
10851
if ((adp->ad_state & ALLCOMPLETE) != ALLCOMPLETE)
4452
10852
return;
4453
free_allocdirect(listhead, adp, delay);
10853
TAILQ_REMOVE(listhead, adp, ad_next);
10854
WORKLIST_INSERT(wkhd, &adp->ad_block.nb_list);
4454
10855
}
4455
10856
}
4456
10857
4457
10858
/*
4458
* Called from within softdep_disk_write_complete above. Note that
4459
* this routine is always called from interrupt level with further
4460
* splbio interrupts blocked.
10859
* Called from within softdep_disk_write_complete above. This routine
10860
* completes successfully written allocindirs.
4461
10861
*/
4462
10862
static void
4463
10863
handle_allocindir_partdone(aip)
4467
10867
4468
10868
if ((aip->ai_state & ALLCOMPLETE) != ALLCOMPLETE)
4469
10869
return;
4470
if (aip->ai_buf != NULL)
4471
panic("handle_allocindir_partdone: dangling dependency");
4472
10870
indirdep = aip->ai_indirdep;
4473
if (indirdep->ir_state & UNDONE) {
4474
LIST_REMOVE(aip, ai_next);
10871
LIST_REMOVE(aip, ai_next);
10872
/*
10873
* Don't set a pointer while the buffer is undergoing IO or while
10874
* we have active truncations.
10875
*/
10876
if (indirdep->ir_state & UNDONE || !TAILQ_EMPTY(&indirdep->ir_trunc)) {
4475
10877
LIST_INSERT_HEAD(&indirdep->ir_donehd, aip, ai_next);
4476
10878
return;
4477
10879
}
4481
10883
else
4482
10884
((ufs2_daddr_t *)indirdep->ir_savebp->b_data)[aip->ai_offset] =
4483
10885
aip->ai_newblkno;
4484
LIST_REMOVE(aip, ai_next);
4485
if (aip->ai_freefrag != NULL)
4486
add_to_worklist(&aip->ai_freefrag->ff_list);
4487
WORKITEM_FREE(aip, D_ALLOCINDIR);
4488
}
4489
10886
/*
10887
* Await the pointer write before freeing the allocindir.
10888
*/
10889
LIST_INSERT_HEAD(&indirdep->ir_writehd, aip, ai_next);
10890
}
10891
10892
/*
10893
* Release segments held on a jwork list.
10894
*/
10895
static void
10896
handle_jwork(wkhd)
10897
struct workhead *wkhd;
10898
{
10899
struct worklist *wk;
10900
10901
while ((wk = LIST_FIRST(wkhd)) != NULL) {
10902
WORKLIST_REMOVE(wk);
10903
switch (wk->wk_type) {
10904
case D_JSEGDEP:
10905
free_jsegdep(WK_JSEGDEP(wk));
10906
continue;
10907
case D_FREEDEP:
10908
free_freedep(WK_FREEDEP(wk));
10909
continue;
10910
case D_FREEFRAG:
10911
rele_jseg(WK_JSEG(WK_FREEFRAG(wk)->ff_jdep));
10912
WORKITEM_FREE(wk, D_FREEFRAG);
10913
continue;
10914
case D_FREEWORK:
10915
handle_written_freework(WK_FREEWORK(wk));
10916
continue;
10917
default:
10918
panic("handle_jwork: Unknown type %s\n",
10919
TYPENAME(wk->wk_type));
10920
}
10921
}
10922
}
10923
10924
/*
10925
* Handle the bufwait list on an inode when it is safe to release items
10926
* held there. This normally happens after an inode block is written but
10927
* may be delayed and handled later if there are pending journal items that
10928
* are not yet safe to be released.
10929
*/
10930
static struct freefile *
10931
handle_bufwait(inodedep, refhd)
10932
struct inodedep *inodedep;
10933
struct workhead *refhd;
10934
{
10935
struct jaddref *jaddref;
10936
struct freefile *freefile;
10937
struct worklist *wk;
10938
10939
freefile = NULL;
10940
while ((wk = LIST_FIRST(&inodedep->id_bufwait)) != NULL) {
10941
WORKLIST_REMOVE(wk);
10942
switch (wk->wk_type) {
10943
case D_FREEFILE:
10944
/*
10945
* We defer adding freefile to the worklist
10946
* until all other additions have been made to
10947
* ensure that it will be done after all the
10948
* old blocks have been freed.
10949
*/
10950
if (freefile != NULL)
10951
panic("handle_bufwait: freefile");
10952
freefile = WK_FREEFILE(wk);
10953
continue;
10954
10955
case D_MKDIR:
10956
handle_written_mkdir(WK_MKDIR(wk), MKDIR_PARENT);
10957
continue;
10958
10959
case D_DIRADD:
10960
diradd_inode_written(WK_DIRADD(wk), inodedep);
10961
continue;
10962
10963
case D_FREEFRAG:
10964
wk->wk_state |= COMPLETE;
10965
if ((wk->wk_state & ALLCOMPLETE) == ALLCOMPLETE)
10966
add_to_worklist(wk, 0);
10967
continue;
10968
10969
case D_DIRREM:
10970
wk->wk_state |= COMPLETE;
10971
add_to_worklist(wk, 0);
10972
continue;
10973
10974
case D_ALLOCDIRECT:
10975
case D_ALLOCINDIR:
10976
free_newblk(WK_NEWBLK(wk));
10977
continue;
10978
10979
case D_JNEWBLK:
10980
wk->wk_state |= COMPLETE;
10981
free_jnewblk(WK_JNEWBLK(wk));
10982
continue;
10983
10984
/*
10985
* Save freed journal segments and add references on
10986
* the supplied list which will delay their release
10987
* until the cg bitmap is cleared on disk.
10988
*/
10989
case D_JSEGDEP:
10990
if (refhd == NULL)
10991
free_jsegdep(WK_JSEGDEP(wk));
10992
else
10993
WORKLIST_INSERT(refhd, wk);
10994
continue;
10995
10996
case D_JADDREF:
10997
jaddref = WK_JADDREF(wk);
10998
TAILQ_REMOVE(&inodedep->id_inoreflst, &jaddref->ja_ref,
10999
if_deps);
11000
/*
11001
* Transfer any jaddrefs to the list to be freed with
11002
* the bitmap if we're handling a removed file.
11003
*/
11004
if (refhd == NULL) {
11005
wk->wk_state |= COMPLETE;
11006
free_jaddref(jaddref);
11007
} else
11008
WORKLIST_INSERT(refhd, wk);
11009
continue;
11010
11011
default:
11012
panic("handle_bufwait: Unknown type %p(%s)",
11013
wk, TYPENAME(wk->wk_type));
11014
/* NOTREACHED */
11015
}
11016
}
11017
return (freefile);
11018
}
4490
11019
/*
4491
11020
* Called from within softdep_disk_write_complete above to restore
4492
11021
* in-memory inode block contents to their most up-to-date state. Note
4498
11027
struct inodedep *inodedep;
4499
11028
struct buf *bp; /* buffer containing the inode block */
4500
11029
{
4501
struct worklist *wk, *filefree;
11030
struct freefile *freefile;
4502
11031
struct allocdirect *adp, *nextadp;
4503
11032
struct ufs1_dinode *dp1 = NULL;
4504
11033
struct ufs2_dinode *dp2 = NULL;
11034
struct workhead wkhd;
4505
11035
int hadchanges, fstype;
11036
ino_t freelink;
4506
11037
11038
LIST_INIT(&wkhd);
11039
hadchanges = 0;
11040
freefile = NULL;
4507
11041
if ((inodedep->id_state & IOSTARTED) == 0)
4508
11042
panic("handle_written_inodeblock: not started");
4509
11043
inodedep->id_state &= ~IOSTARTED;
4511
11045
fstype = UFS1;
4512
11046
dp1 = (struct ufs1_dinode *)bp->b_data +
4513
11047
ino_to_fsbo(inodedep->id_fs, inodedep->id_ino);
11048
freelink = dp1->di_freelink;
4514
11049
} else {
4515
11050
fstype = UFS2;
4516
11051
dp2 = (struct ufs2_dinode *)bp->b_data +
4517
11052
ino_to_fsbo(inodedep->id_fs, inodedep->id_ino);
11053
freelink = dp2->di_freelink;
11054
}
11055
/*
11056
* Leave this inodeblock dirty until it's in the list.
11057
*/
11058
if ((inodedep->id_state & (UNLINKED | UNLINKONLIST)) == UNLINKED) {
11059
struct inodedep *inon;
11060
11061
inon = TAILQ_NEXT(inodedep, id_unlinked);
11062
if ((inon == NULL && freelink == 0) ||
11063
(inon && inon->id_ino == freelink)) {
11064
if (inon)
11065
inon->id_state |= UNLINKPREV;
11066
inodedep->id_state |= UNLINKNEXT;
11067
}
11068
hadchanges = 1;
4518
11069
}
4519
11070
/*
4520
11071
* If we had to rollback the inode allocation because of
4524
11075
* corresponding updates written to disk.
4525
11076
*/
4526
11077
if (inodedep->id_savedino1 != NULL) {
11078
hadchanges = 1;
4527
11079
if (fstype == UFS1)
4528
11080
*dp1 = *inodedep->id_savedino1;
4529
11081
else
4533
11085
if ((bp->b_flags & B_DELWRI) == 0)
4534
11086
stat_inode_bitmap++;
4535
11087
bdirty(bp);
11088
/*
11089
* If the inode is clear here and GOINGAWAY it will never
11090
* be written. Process the bufwait and clear any pending
11091
* work which may include the freefile.
11092
*/
11093
if (inodedep->id_state & GOINGAWAY)
11094
goto bufwait;
4536
11095
return (1);
4537
11096
}
4538
11097
inodedep->id_state |= COMPLETE;
4540
11099
* Roll forward anything that had to be rolled back before
4541
11100
* the inode could be updated.
4542
11101
*/
4543
hadchanges = 0;
4544
11102
for (adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp; adp = nextadp) {
4545
11103
nextadp = TAILQ_NEXT(adp, ad_next);
4546
11104
if (adp->ad_state & ATTACHED)
4547
11105
panic("handle_written_inodeblock: new entry");
4548
11106
if (fstype == UFS1) {
4549
if (adp->ad_lbn < NDADDR) {
4550
if (dp1->di_db[adp->ad_lbn]!=adp->ad_oldblkno)
11107
if (adp->ad_offset < NDADDR) {
11108
if (dp1->di_db[adp->ad_offset]!=adp->ad_oldblkno)
4551
11109
panic("%s %s #%jd mismatch %d != %jd",
4552
11110
"handle_written_inodeblock:",
4553
11111
"direct pointer",
4554
(intmax_t)adp->ad_lbn,
4555
dp1->di_db[adp->ad_lbn],
11112
(intmax_t)adp->ad_offset,
11113
dp1->di_db[adp->ad_offset],
4556
11114
(intmax_t)adp->ad_oldblkno);
4557
dp1->di_db[adp->ad_lbn] = adp->ad_newblkno;
11115
dp1->di_db[adp->ad_offset] = adp->ad_newblkno;
4558
11116
} else {
4559
if (dp1->di_ib[adp->ad_lbn - NDADDR] != 0)
11117
if (dp1->di_ib[adp->ad_offset - NDADDR] != 0)
4560
11118
panic("%s: %s #%jd allocated as %d",
4561
11119
"handle_written_inodeblock",
4562
11120
"indirect pointer",
4563
(intmax_t)adp->ad_lbn - NDADDR,
4564
dp1->di_ib[adp->ad_lbn - NDADDR]);
4565
dp1->di_ib[adp->ad_lbn - NDADDR] =
11121
(intmax_t)adp->ad_offset - NDADDR,
11122
dp1->di_ib[adp->ad_offset - NDADDR]);
11123
dp1->di_ib[adp->ad_offset - NDADDR] =
4566
11124
adp->ad_newblkno;
4567
11125
}
4568
11126
} else {
4569
if (adp->ad_lbn < NDADDR) {
4570
if (dp2->di_db[adp->ad_lbn]!=adp->ad_oldblkno)
11127
if (adp->ad_offset < NDADDR) {
11128
if (dp2->di_db[adp->ad_offset]!=adp->ad_oldblkno)
4571
11129
panic("%s: %s #%jd %s %jd != %jd",
4572
11130
"handle_written_inodeblock",
4573
11131
"direct pointer",
4574
(intmax_t)adp->ad_lbn, "mismatch",
4575
(intmax_t)dp2->di_db[adp->ad_lbn],
11132
(intmax_t)adp->ad_offset, "mismatch",
11133
(intmax_t)dp2->di_db[adp->ad_offset],
4576
11134
(intmax_t)adp->ad_oldblkno);
4577
dp2->di_db[adp->ad_lbn] = adp->ad_newblkno;
11135
dp2->di_db[adp->ad_offset] = adp->ad_newblkno;
4578
11136
} else {
4579
if (dp2->di_ib[adp->ad_lbn - NDADDR] != 0)
11137
if (dp2->di_ib[adp->ad_offset - NDADDR] != 0)
4580
11138
panic("%s: %s #%jd allocated as %jd",
4581
11139
"handle_written_inodeblock",
4582
11140
"indirect pointer",
4583
(intmax_t)adp->ad_lbn - NDADDR,
11141
(intmax_t)adp->ad_offset - NDADDR,
4584
11142
(intmax_t)
4585
dp2->di_ib[adp->ad_lbn - NDADDR]);
4586
dp2->di_ib[adp->ad_lbn - NDADDR] =
11143
dp2->di_ib[adp->ad_offset - NDADDR]);
11144
dp2->di_ib[adp->ad_offset - NDADDR] =
4587
11145
adp->ad_newblkno;
4588
11146
}
4589
11147
}
4595
11153
nextadp = TAILQ_NEXT(adp, ad_next);
4596
11154
if (adp->ad_state & ATTACHED)
4597
11155
panic("handle_written_inodeblock: new entry");
4598
if (dp2->di_extb[adp->ad_lbn] != adp->ad_oldblkno)
11156
if (dp2->di_extb[adp->ad_offset] != adp->ad_oldblkno)
4599
11157
panic("%s: direct pointers #%jd %s %jd != %jd",
4600
11158
"handle_written_inodeblock",
4601
(intmax_t)adp->ad_lbn, "mismatch",
4602
(intmax_t)dp2->di_extb[adp->ad_lbn],
11159
(intmax_t)adp->ad_offset, "mismatch",
11160
(intmax_t)dp2->di_extb[adp->ad_offset],
4603
11161
(intmax_t)adp->ad_oldblkno);
4604
dp2->di_extb[adp->ad_lbn] = adp->ad_newblkno;
11162
dp2->di_extb[adp->ad_offset] = adp->ad_newblkno;
4605
11163
adp->ad_state &= ~UNDONE;
4606
11164
adp->ad_state |= ATTACHED;
4607
11165
hadchanges = 1;
4613
11171
*/
4614
11172
if (inodedep->id_savedsize == -1 || inodedep->id_savedextsize == -1)
4615
11173
panic("handle_written_inodeblock: bad size");
11174
if (inodedep->id_savednlink > LINK_MAX)
11175
panic("handle_written_inodeblock: Invalid link count "
11176
"%d for inodedep %p", inodedep->id_savednlink, inodedep);
4616
11177
if (fstype == UFS1) {
11178
if (dp1->di_nlink != inodedep->id_savednlink) {
11179
dp1->di_nlink = inodedep->id_savednlink;
11180
hadchanges = 1;
11181
}
4617
11182
if (dp1->di_size != inodedep->id_savedsize) {
4618
11183
dp1->di_size = inodedep->id_savedsize;
4619
11184
hadchanges = 1;
4620
11185
}
4621
11186
} else {
11187
if (dp2->di_nlink != inodedep->id_savednlink) {
11188
dp2->di_nlink = inodedep->id_savednlink;
11189
hadchanges = 1;
11190
}
4622
11191
if (dp2->di_size != inodedep->id_savedsize) {
4623
11192
dp2->di_size = inodedep->id_savedsize;
4624
11193
hadchanges = 1;
4630
11199
}
4631
11200
inodedep->id_savedsize = -1;
4632
11201
inodedep->id_savedextsize = -1;
11202
inodedep->id_savednlink = -1;
4633
11203
/*
4634
11204
* If there were any rollbacks in the inode block, then it must be
4635
11205
* marked dirty so that its will eventually get written back in
4637
11207
*/
4638
11208
if (hadchanges)
4639
11209
bdirty(bp);
11210
bufwait:
4640
11211
/*
4641
11212
* Process any allocdirects that completed during the update.
4642
11213
*/
4643
11214
if ((adp = TAILQ_FIRST(&inodedep->id_inoupdt)) != NULL)
4644
handle_allocdirect_partdone(adp);
11215
handle_allocdirect_partdone(adp, &wkhd);
4645
11216
if ((adp = TAILQ_FIRST(&inodedep->id_extupdt)) != NULL)
4646
handle_allocdirect_partdone(adp);
11217
handle_allocdirect_partdone(adp, &wkhd);
4647
11218
/*
4648
11219
* Process deallocations that were held pending until the
4649
11220
* inode had been written to disk. Freeing of the inode
4650
11221
* is delayed until after all blocks have been freed to
4651
11222
* avoid creation of new <vfsid, inum, lbn> triples
4652
* before the old ones have been deleted.
11223
* before the old ones have been deleted. Completely
11224
* unlinked inodes are not processed until the unlinked
11225
* inode list is written or the last reference is removed.
4653
11226
*/
4654
filefree = NULL;
4655
while ((wk = LIST_FIRST(&inodedep->id_bufwait)) != NULL) {
4656
WORKLIST_REMOVE(wk);
4657
switch (wk->wk_type) {
4658
4659
case D_FREEFILE:
4660
/*
4661
* We defer adding filefree to the worklist until
4662
* all other additions have been made to ensure
4663
* that it will be done after all the old blocks
4664
* have been freed.
4665
*/
4666
if (filefree != NULL)
4667
panic("handle_written_inodeblock: filefree");
4668
filefree = wk;
4669
continue;
4670
4671
case D_MKDIR:
4672
handle_written_mkdir(WK_MKDIR(wk), MKDIR_PARENT);
4673
continue;
4674
4675
case D_DIRADD:
4676
diradd_inode_written(WK_DIRADD(wk), inodedep);
4677
continue;
4678
4679
case D_FREEBLKS:
4680
wk->wk_state |= COMPLETE;
4681
if ((wk->wk_state & ALLCOMPLETE) != ALLCOMPLETE)
4682
continue;
4683
/* -- fall through -- */
4684
case D_FREEFRAG:
4685
case D_DIRREM:
4686
add_to_worklist(wk);
4687
continue;
4688
4689
case D_NEWDIRBLK:
4690
free_newdirblk(WK_NEWDIRBLK(wk));
4691
continue;
4692
4693
default:
4694
panic("handle_written_inodeblock: Unknown type %s",
4695
TYPENAME(wk->wk_type));
4696
/* NOTREACHED */
11227
if ((inodedep->id_state & (UNLINKED | UNLINKONLIST)) != UNLINKED) {
11228
freefile = handle_bufwait(inodedep, NULL);
11229
if (freefile && !LIST_EMPTY(&wkhd)) {
11230
WORKLIST_INSERT(&wkhd, &freefile->fx_list);
11231
freefile = NULL;
4697
11232
}
4698
11233
}
4699
if (filefree != NULL) {
11234
/*
11235
* Move rolled forward dependency completions to the bufwait list
11236
* now that those that were already written have been processed.
11237
*/
11238
if (!LIST_EMPTY(&wkhd) && hadchanges == 0)
11239
panic("handle_written_inodeblock: bufwait but no changes");
11240
jwork_move(&inodedep->id_bufwait, &wkhd);
11241
11242
if (freefile != NULL) {
11243
/*
11244
* If the inode is goingaway it was never written. Fake up
11245
* the state here so free_inodedep() can succeed.
11246
*/
11247
if (inodedep->id_state & GOINGAWAY)
11248
inodedep->id_state |= COMPLETE | DEPCOMPLETE;
4700
11249
if (free_inodedep(inodedep) == 0)
4701
panic("handle_written_inodeblock: live inodedep");
4702
add_to_worklist(filefree);
11250
panic("handle_written_inodeblock: live inodedep %p",
11251
inodedep);
11252
add_to_worklist(&freefile->fx_list, 0);
4703
11253
return (0);
4704
11254
}
4705
11255
4707
11257
* If no outstanding dependencies, free it.
4708
11258
*/
4709
11259
if (free_inodedep(inodedep) ||
4710
(TAILQ_FIRST(&inodedep->id_inoupdt) == 0 &&
4711
TAILQ_FIRST(&inodedep->id_extupdt) == 0))
11260
(TAILQ_FIRST(&inodedep->id_inoreflst) == 0 &&
11261
TAILQ_FIRST(&inodedep->id_inoupdt) == 0 &&
11262
TAILQ_FIRST(&inodedep->id_extupdt) == 0 &&
11263
LIST_FIRST(&inodedep->id_bufwait) == 0))
4712
11264
return (0);
4713
11265
return (hadchanges);
4714
11266
}
4715
11267
11268
static int
11269
handle_written_indirdep(indirdep, bp, bpp)
11270
struct indirdep *indirdep;
11271
struct buf *bp;
11272
struct buf **bpp;
11273
{
11274
struct allocindir *aip;
11275
struct buf *sbp;
11276
int chgs;
11277
11278
if (indirdep->ir_state & GOINGAWAY)
11279
panic("handle_written_indirdep: indirdep gone");
11280
if ((indirdep->ir_state & IOSTARTED) == 0)
11281
panic("handle_written_indirdep: IO not started");
11282
chgs = 0;
11283
/*
11284
* If there were rollbacks revert them here.
11285
*/
11286
if (indirdep->ir_saveddata) {
11287
bcopy(indirdep->ir_saveddata, bp->b_data, bp->b_bcount);
11288
if (TAILQ_EMPTY(&indirdep->ir_trunc)) {
11289
free(indirdep->ir_saveddata, M_INDIRDEP);
11290
indirdep->ir_saveddata = NULL;
11291
}
11292
chgs = 1;
11293
}
11294
indirdep->ir_state &= ~(UNDONE | IOSTARTED);
11295
indirdep->ir_state |= ATTACHED;
11296
/*
11297
* Move allocindirs with written pointers to the completehd if
11298
* the indirdep's pointer is not yet written. Otherwise
11299
* free them here.
11300
*/
11301
while ((aip = LIST_FIRST(&indirdep->ir_writehd)) != 0) {
11302
LIST_REMOVE(aip, ai_next);
11303
if ((indirdep->ir_state & DEPCOMPLETE) == 0) {
11304
LIST_INSERT_HEAD(&indirdep->ir_completehd, aip,
11305
ai_next);
11306
newblk_freefrag(&aip->ai_block);
11307
continue;
11308
}
11309
free_newblk(&aip->ai_block);
11310
}
11311
/*
11312
* Move allocindirs that have finished dependency processing from
11313
* the done list to the write list after updating the pointers.
11314
*/
11315
if (TAILQ_EMPTY(&indirdep->ir_trunc)) {
11316
while ((aip = LIST_FIRST(&indirdep->ir_donehd)) != 0) {
11317
handle_allocindir_partdone(aip);
11318
if (aip == LIST_FIRST(&indirdep->ir_donehd))
11319
panic("disk_write_complete: not gone");
11320
chgs = 1;
11321
}
11322
}
11323
/*
11324
* Preserve the indirdep if there were any changes or if it is not
11325
* yet valid on disk.
11326
*/
11327
if (chgs) {
11328
stat_indir_blk_ptrs++;
11329
bdirty(bp);
11330
return (1);
11331
}
11332
/*
11333
* If there were no changes we can discard the savedbp and detach
11334
* ourselves from the buf. We are only carrying completed pointers
11335
* in this case.
11336
*/
11337
sbp = indirdep->ir_savebp;
11338
sbp->b_flags |= B_INVAL | B_NOCACHE;
11339
indirdep->ir_savebp = NULL;
11340
indirdep->ir_bp = NULL;
11341
if (*bpp != NULL)
11342
panic("handle_written_indirdep: bp already exists.");
11343
*bpp = sbp;
11344
/*
11345
* The indirdep may not be freed until its parent points at it.
11346
*/
11347
if (indirdep->ir_state & DEPCOMPLETE)
11348
free_indirdep(indirdep);
11349
11350
return (0);
11351
}
11352
4716
11353
/*
4717
11354
* Process a diradd entry after its dependent inode has been written.
4718
11355
* This routine must be called with splbio interrupts blocked.
4722
11359
struct diradd *dap;
4723
11360
struct inodedep *inodedep;
4724
11361
{
4725
struct pagedep *pagedep;
4726
11362
4727
11363
dap->da_state |= COMPLETE;
4728
if ((dap->da_state & ALLCOMPLETE) == ALLCOMPLETE) {
4729
if (dap->da_state & DIRCHG)
4730
pagedep = dap->da_previous->dm_pagedep;
4731
else
4732
pagedep = dap->da_pagedep;
4733
LIST_REMOVE(dap, da_pdlist);
4734
LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist);
4735
}
11364
complete_diradd(dap);
4736
11365
WORKLIST_INSERT(&inodedep->id_pendinghd, &dap->da_list);
4737
11366
}
4738
11367
4739
11368
/*
11369
* Returns true if the bmsafemap will have rollbacks when written. Must
11370
* only be called with lk and the buf lock on the cg held.
11371
*/
11372
static int
11373
bmsafemap_backgroundwrite(bmsafemap, bp)
11374
struct bmsafemap *bmsafemap;
11375
struct buf *bp;
11376
{
11377
int dirty;
11378
11379
dirty = !LIST_EMPTY(&bmsafemap->sm_jaddrefhd) |
11380
!LIST_EMPTY(&bmsafemap->sm_jnewblkhd);
11381
/*
11382
* If we're initiating a background write we need to process the
11383
* rollbacks as they exist now, not as they exist when IO starts.
11384
* No other consumers will look at the contents of the shadowed
11385
* buf so this is safe to do here.
11386
*/
11387
if (bp->b_xflags & BX_BKGRDMARKER)
11388
initiate_write_bmsafemap(bmsafemap, bp);
11389
11390
return (dirty);
11391
}
11392
11393
/*
11394
* Re-apply an allocation when a cg write is complete.
11395
*/
11396
static int
11397
jnewblk_rollforward(jnewblk, fs, cgp, blksfree)
11398
struct jnewblk *jnewblk;
11399
struct fs *fs;
11400
struct cg *cgp;
11401
uint8_t *blksfree;
11402
{
11403
ufs1_daddr_t fragno;
11404
ufs2_daddr_t blkno;
11405
long cgbno, bbase;
11406
int frags, blk;
11407
int i;
11408
11409
frags = 0;
11410
cgbno = dtogd(fs, jnewblk->jn_blkno);
11411
for (i = jnewblk->jn_oldfrags; i < jnewblk->jn_frags; i++) {
11412
if (isclr(blksfree, cgbno + i))
11413
panic("jnewblk_rollforward: re-allocated fragment");
11414
frags++;
11415
}
11416
if (frags == fs->fs_frag) {
11417
blkno = fragstoblks(fs, cgbno);
11418
ffs_clrblock(fs, blksfree, (long)blkno);
11419
ffs_clusteracct(fs, cgp, blkno, -1);
11420
cgp->cg_cs.cs_nbfree--;
11421
} else {
11422
bbase = cgbno - fragnum(fs, cgbno);
11423
cgbno += jnewblk->jn_oldfrags;
11424
/* If a complete block had been reassembled, account for it. */
11425
fragno = fragstoblks(fs, bbase);
11426
if (ffs_isblock(fs, blksfree, fragno)) {
11427
cgp->cg_cs.cs_nffree += fs->fs_frag;
11428
ffs_clusteracct(fs, cgp, fragno, -1);
11429
cgp->cg_cs.cs_nbfree--;
11430
}
11431
/* Decrement the old frags. */
11432
blk = blkmap(fs, blksfree, bbase);
11433
ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
11434
/* Allocate the fragment */
11435
for (i = 0; i < frags; i++)
11436
clrbit(blksfree, cgbno + i);
11437
cgp->cg_cs.cs_nffree -= frags;
11438
/* Add back in counts associated with the new frags */
11439
blk = blkmap(fs, blksfree, bbase);
11440
ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
11441
}
11442
return (frags);
11443
}
11444
11445
/*
11446
* Complete a write to a bmsafemap structure. Roll forward any bitmap
11447
* changes if it's not a background write. Set all written dependencies
11448
* to DEPCOMPLETE and free the structure if possible.
11449
*/
11450
static int
11451
handle_written_bmsafemap(bmsafemap, bp)
11452
struct bmsafemap *bmsafemap;
11453
struct buf *bp;
11454
{
11455
struct newblk *newblk;
11456
struct inodedep *inodedep;
11457
struct jaddref *jaddref, *jatmp;
11458
struct jnewblk *jnewblk, *jntmp;
11459
struct ufsmount *ump;
11460
uint8_t *inosused;
11461
uint8_t *blksfree;
11462
struct cg *cgp;
11463
struct fs *fs;
11464
ino_t ino;
11465
int chgs;
11466
11467
if ((bmsafemap->sm_state & IOSTARTED) == 0)
11468
panic("initiate_write_bmsafemap: Not started\n");
11469
ump = VFSTOUFS(bmsafemap->sm_list.wk_mp);
11470
chgs = 0;
11471
bmsafemap->sm_state &= ~IOSTARTED;
11472
/*
11473
* Release journal work that was waiting on the write.
11474
*/
11475
handle_jwork(&bmsafemap->sm_freewr);
11476
11477
/*
11478
* Restore unwritten inode allocation pending jaddref writes.
11479
*/
11480
if (!LIST_EMPTY(&bmsafemap->sm_jaddrefhd)) {
11481
cgp = (struct cg *)bp->b_data;
11482
fs = VFSTOUFS(bmsafemap->sm_list.wk_mp)->um_fs;
11483
inosused = cg_inosused(cgp);
11484
LIST_FOREACH_SAFE(jaddref, &bmsafemap->sm_jaddrefhd,
11485
ja_bmdeps, jatmp) {
11486
if ((jaddref->ja_state & UNDONE) == 0)
11487
continue;
11488
ino = jaddref->ja_ino % fs->fs_ipg;
11489
if (isset(inosused, ino))
11490
panic("handle_written_bmsafemap: "
11491
"re-allocated inode");
11492
if ((bp->b_xflags & BX_BKGRDMARKER) == 0) {
11493
if ((jaddref->ja_mode & IFMT) == IFDIR)
11494
cgp->cg_cs.cs_ndir++;
11495
cgp->cg_cs.cs_nifree--;
11496
setbit(inosused, ino);
11497
chgs = 1;
11498
}
11499
jaddref->ja_state &= ~UNDONE;
11500
jaddref->ja_state |= ATTACHED;
11501
free_jaddref(jaddref);
11502
}
11503
}
11504
/*
11505
* Restore any block allocations which are pending journal writes.
11506
*/
11507
if (LIST_FIRST(&bmsafemap->sm_jnewblkhd) != NULL) {
11508
cgp = (struct cg *)bp->b_data;
11509
fs = VFSTOUFS(bmsafemap->sm_list.wk_mp)->um_fs;
11510
blksfree = cg_blksfree(cgp);
11511
LIST_FOREACH_SAFE(jnewblk, &bmsafemap->sm_jnewblkhd, jn_deps,
11512
jntmp) {
11513
if ((jnewblk->jn_state & UNDONE) == 0)
11514
continue;
11515
if ((bp->b_xflags & BX_BKGRDMARKER) == 0 &&
11516
jnewblk_rollforward(jnewblk, fs, cgp, blksfree))
11517
chgs = 1;
11518
jnewblk->jn_state &= ~(UNDONE | NEWBLOCK);
11519
jnewblk->jn_state |= ATTACHED;
11520
free_jnewblk(jnewblk);
11521
}
11522
}
11523
while ((newblk = LIST_FIRST(&bmsafemap->sm_newblkwr))) {
11524
newblk->nb_state |= DEPCOMPLETE;
11525
newblk->nb_state &= ~ONDEPLIST;
11526
newblk->nb_bmsafemap = NULL;
11527
LIST_REMOVE(newblk, nb_deps);
11528
if (newblk->nb_list.wk_type == D_ALLOCDIRECT)
11529
handle_allocdirect_partdone(
11530
WK_ALLOCDIRECT(&newblk->nb_list), NULL);
11531
else if (newblk->nb_list.wk_type == D_ALLOCINDIR)
11532
handle_allocindir_partdone(
11533
WK_ALLOCINDIR(&newblk->nb_list));
11534
else if (newblk->nb_list.wk_type != D_NEWBLK)
11535
panic("handle_written_bmsafemap: Unexpected type: %s",
11536
TYPENAME(newblk->nb_list.wk_type));
11537
}
11538
while ((inodedep = LIST_FIRST(&bmsafemap->sm_inodedepwr)) != NULL) {
11539
inodedep->id_state |= DEPCOMPLETE;
11540
inodedep->id_state &= ~ONDEPLIST;
11541
LIST_REMOVE(inodedep, id_deps);
11542
inodedep->id_bmsafemap = NULL;
11543
}
11544
LIST_REMOVE(bmsafemap, sm_next);
11545
if (chgs == 0 && LIST_EMPTY(&bmsafemap->sm_jaddrefhd) &&
11546
LIST_EMPTY(&bmsafemap->sm_jnewblkhd) &&
11547
LIST_EMPTY(&bmsafemap->sm_newblkhd) &&
11548
LIST_EMPTY(&bmsafemap->sm_inodedephd) &&
11549
LIST_EMPTY(&bmsafemap->sm_freehd)) {
11550
LIST_REMOVE(bmsafemap, sm_hash);
11551
WORKITEM_FREE(bmsafemap, D_BMSAFEMAP);
11552
return (0);
11553
}
11554
LIST_INSERT_HEAD(&ump->softdep_dirtycg, bmsafemap, sm_next);
11555
bdirty(bp);
11556
return (1);
11557
}
11558
11559
/*
11560
* Try to free a mkdir dependency.
11561
*/
11562
static void
11563
complete_mkdir(mkdir)
11564
struct mkdir *mkdir;
11565
{
11566
struct diradd *dap;
11567
11568
if ((mkdir->md_state & ALLCOMPLETE) != ALLCOMPLETE)
11569
return;
11570
LIST_REMOVE(mkdir, md_mkdirs);
11571
dap = mkdir->md_diradd;
11572
dap->da_state &= ~(mkdir->md_state & (MKDIR_PARENT | MKDIR_BODY));
11573
if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) == 0) {
11574
dap->da_state |= DEPCOMPLETE;
11575
complete_diradd(dap);
11576
}
11577
WORKITEM_FREE(mkdir, D_MKDIR);
11578
}
11579
11580
/*
4740
11581
* Handle the completion of a mkdir dependency.
4741
11582
*/
4742
11583
static void
4744
11585
struct mkdir *mkdir;
4745
11586
int type;
4746
11587
{
4747
struct diradd *dap;
4748
struct pagedep *pagedep;
4749
11588
4750
if (mkdir->md_state != type)
11589
if ((mkdir->md_state & (MKDIR_PARENT | MKDIR_BODY)) != type)
4751
11590
panic("handle_written_mkdir: bad type");
4752
dap = mkdir->md_diradd;
4753
dap->da_state &= ~type;
4754
if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) == 0)
4755
dap->da_state |= DEPCOMPLETE;
4756
if ((dap->da_state & ALLCOMPLETE) == ALLCOMPLETE) {
4757
if (dap->da_state & DIRCHG)
4758
pagedep = dap->da_previous->dm_pagedep;
4759
else
4760
pagedep = dap->da_pagedep;
4761
LIST_REMOVE(dap, da_pdlist);
4762
LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist);
4763
}
4764
LIST_REMOVE(mkdir, md_mkdirs);
4765
WORKITEM_FREE(mkdir, D_MKDIR);
11591
mkdir->md_state |= COMPLETE;
11592
complete_mkdir(mkdir);
11593
}
11594
11595
static int
11596
free_pagedep(pagedep)
11597
struct pagedep *pagedep;
11598
{
11599
int i;
11600
11601
if (pagedep->pd_state & NEWBLOCK)
11602
return (0);
11603
if (!LIST_EMPTY(&pagedep->pd_dirremhd))
11604
return (0);
11605
for (i = 0; i < DAHASHSZ; i++)
11606
if (!LIST_EMPTY(&pagedep->pd_diraddhd[i]))
11607
return (0);
11608
if (!LIST_EMPTY(&pagedep->pd_pendinghd))
11609
return (0);
11610
if (!LIST_EMPTY(&pagedep->pd_jmvrefhd))
11611
return (0);
11612
if (pagedep->pd_state & ONWORKLIST)
11613
WORKLIST_REMOVE(&pagedep->pd_list);
11614
LIST_REMOVE(pagedep, pd_hash);
11615
WORKITEM_FREE(pagedep, D_PAGEDEP);
11616
11617
return (1);
4766
11618
}
4767
11619
4768
11620
/*
4790
11642
*/
4791
11643
while ((dirrem = LIST_FIRST(&pagedep->pd_dirremhd)) != NULL) {
4792
11644
LIST_REMOVE(dirrem, dm_next);
11645
dirrem->dm_state |= COMPLETE;
4793
11646
dirrem->dm_dirinum = pagedep->pd_ino;
4794
add_to_worklist(&dirrem->dm_list);
11647
KASSERT(LIST_EMPTY(&dirrem->dm_jremrefhd),
11648
("handle_written_filepage: Journal entries not written."));
11649
add_to_worklist(&dirrem->dm_list, 0);
4795
11650
}
4796
11651
/*
4797
11652
* Free any directory additions that have been committed.
4800
11655
*/
4801
11656
if ((pagedep->pd_state & NEWBLOCK) == 0)
4802
11657
while ((dap = LIST_FIRST(&pagedep->pd_pendinghd)) != NULL)
4803
free_diradd(dap);
11658
free_diradd(dap, NULL);
4804
11659
/*
4805
11660
* Uncommitted directory entries must be restored.
4806
11661
*/
4845
11700
* Otherwise it will remain to track any new entries on
4846
11701
* the page in case they are fsync'ed.
4847
11702
*/
4848
if ((pagedep->pd_state & NEWBLOCK) == 0) {
4849
LIST_REMOVE(pagedep, pd_hash);
4850
WORKITEM_FREE(pagedep, D_PAGEDEP);
4851
}
11703
free_pagedep(pagedep);
4852
11704
return (0);
4853
11705
}
4854
11706
4880
11732
*/
4881
11733
ip->i_effnlink = ip->i_nlink;
4882
11734
ACQUIRE_LOCK(&lk);
4883
if (inodedep_lookup(UFSTOVFS(ip->i_ump),
4884
ip->i_number, 0, &inodedep) == 0) {
11735
if (inodedep_lookup(UFSTOVFS(ip->i_ump), ip->i_number, 0,
11736
&inodedep) == 0) {
4885
11737
FREE_LOCK(&lk);
4886
11738
return;
4887
11739
}
4888
11740
ip->i_effnlink -= inodedep->id_nlinkdelta;
4889
if (inodedep->id_state & SPACECOUNTED)
4890
ip->i_flag |= IN_SPACECOUNTED;
4891
11741
FREE_LOCK(&lk);
4892
11742
}
4893
11743
4908
11758
int waitfor; /* nonzero => update must be allowed */
4909
11759
{
4910
11760
struct inodedep *inodedep;
11761
struct inoref *inoref;
4911
11762
struct worklist *wk;
4912
11763
struct mount *mp;
4913
11764
struct buf *ibp;
11765
struct fs *fs;
4914
11766
int error;
4915
11767
11768
mp = UFSTOVFS(ip->i_ump);
11769
fs = ip->i_fs;
11770
/*
11771
* Preserve the freelink that is on disk. clear_unlinked_inodedep()
11772
* does not have access to the in-core ip so must write directly into
11773
* the inode block buffer when setting freelink.
11774
*/
11775
if (fs->fs_magic == FS_UFS1_MAGIC)
11776
DIP_SET(ip, i_freelink, ((struct ufs1_dinode *)bp->b_data +
11777
ino_to_fsbo(fs, ip->i_number))->di_freelink);
11778
else
11779
DIP_SET(ip, i_freelink, ((struct ufs2_dinode *)bp->b_data +
11780
ino_to_fsbo(fs, ip->i_number))->di_freelink);
4916
11781
/*
4917
11782
* If the effective link count is not equal to the actual link
4918
11783
* count, then we must track the difference in an inodedep while
4920
11785
* if there is no existing inodedep, then there are no dependencies
4921
11786
* to track.
4922
11787
*/
4923
mp = UFSTOVFS(ip->i_ump);
4924
11788
ACQUIRE_LOCK(&lk);
11789
again:
4925
11790
if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) == 0) {
4926
11791
FREE_LOCK(&lk);
4927
11792
if (ip->i_effnlink != ip->i_nlink)
4931
11796
if (inodedep->id_nlinkdelta != ip->i_nlink - ip->i_effnlink)
4932
11797
panic("softdep_update_inodeblock: bad delta");
4933
11798
/*
11799
* If we're flushing all dependencies we must also move any waiting
11800
* for journal writes onto the bufwait list prior to I/O.
11801
*/
11802
if (waitfor) {
11803
TAILQ_FOREACH(inoref, &inodedep->id_inoreflst, if_deps) {
11804
if ((inoref->if_state & (DEPCOMPLETE | GOINGAWAY))
11805
== DEPCOMPLETE) {
11806
jwait(&inoref->if_list, MNT_WAIT);
11807
goto again;
11808
}
11809
}
11810
}
11811
/*
4934
11812
* Changes have been initiated. Anything depending on these
4935
11813
* changes cannot occur until this inode has been written.
4936
11814
*/
4945
11823
*/
4946
11824
merge_inode_lists(&inodedep->id_newinoupdt, &inodedep->id_inoupdt);
4947
11825
if (!TAILQ_EMPTY(&inodedep->id_inoupdt))
4948
handle_allocdirect_partdone(TAILQ_FIRST(&inodedep->id_inoupdt));
11826
handle_allocdirect_partdone(TAILQ_FIRST(&inodedep->id_inoupdt),
11827
NULL);
4949
11828
merge_inode_lists(&inodedep->id_newextupdt, &inodedep->id_extupdt);
4950
11829
if (!TAILQ_EMPTY(&inodedep->id_extupdt))
4951
handle_allocdirect_partdone(TAILQ_FIRST(&inodedep->id_extupdt));
11830
handle_allocdirect_partdone(TAILQ_FIRST(&inodedep->id_extupdt),
11831
NULL);
4952
11832
/*
4953
11833
* Now that the inode has been pushed into the buffer, the
4954
11834
* operations dependent on the inode being written to disk
4971
11851
return;
4972
11852
}
4973
11853
retry:
4974
if ((inodedep->id_state & DEPCOMPLETE) != 0) {
11854
if ((inodedep->id_state & (DEPCOMPLETE | GOINGAWAY)) != 0) {
4975
11855
FREE_LOCK(&lk);
4976
11856
return;
4977
11857
}
4978
ibp = inodedep->id_buf;
11858
ibp = inodedep->id_bmsafemap->sm_buf;
4979
11859
ibp = getdirtybuf(ibp, &lk, MNT_WAIT);
4980
11860
if (ibp == NULL) {
4981
11861
/*
5007
11887
5008
11888
newadp = TAILQ_FIRST(newlisthead);
5009
11889
for (listadp = TAILQ_FIRST(oldlisthead); listadp && newadp;) {
5010
if (listadp->ad_lbn < newadp->ad_lbn) {
11890
if (listadp->ad_offset < newadp->ad_offset) {
5011
11891
listadp = TAILQ_NEXT(listadp, ad_next);
5012
11892
continue;
5013
11893
}
5014
11894
TAILQ_REMOVE(newlisthead, newadp, ad_next);
5015
11895
TAILQ_INSERT_BEFORE(listadp, newadp, ad_next);
5016
if (listadp->ad_lbn == newadp->ad_lbn) {
11896
if (listadp->ad_offset == newadp->ad_offset) {
5017
11897
allocdirect_merge(oldlisthead, newadp,
5018
11898
listadp);
5019
11899
listadp = newadp;
5036
11916
{
5037
11917
struct inodedep *inodedep;
5038
11918
struct pagedep *pagedep;
11919
struct inoref *inoref;
5039
11920
struct worklist *wk;
5040
11921
struct diradd *dap;
5041
11922
struct mount *mp;
5052
11933
fs = ip->i_fs;
5053
11934
mp = vp->v_mount;
5054
11935
ACQUIRE_LOCK(&lk);
11936
restart:
5055
11937
if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) == 0) {
5056
11938
FREE_LOCK(&lk);
5057
11939
return (0);
5058
11940
}
11941
TAILQ_FOREACH(inoref, &inodedep->id_inoreflst, if_deps) {
11942
if ((inoref->if_state & (DEPCOMPLETE | GOINGAWAY))
11943
== DEPCOMPLETE) {
11944
jwait(&inoref->if_list, MNT_WAIT);
11945
goto restart;
11946
}
11947
}
5059
11948
if (!LIST_EMPTY(&inodedep->id_inowait) ||
5060
!LIST_EMPTY(&inodedep->id_bufwait) ||
5061
11949
!TAILQ_EMPTY(&inodedep->id_extupdt) ||
5062
11950
!TAILQ_EMPTY(&inodedep->id_newextupdt) ||
5063
11951
!TAILQ_EMPTY(&inodedep->id_inoupdt) ||
5064
11952
!TAILQ_EMPTY(&inodedep->id_newinoupdt))
5065
panic("softdep_fsync: pending ops");
11953
panic("softdep_fsync: pending ops %p", inodedep);
5066
11954
for (error = 0, flushparent = 0; ; ) {
5067
11955
if ((wk = LIST_FIRST(&inodedep->id_pendinghd)) == NULL)
5068
11956
break;
5165
12053
pagedep_new_block = pagedep->pd_state & NEWBLOCK;
5166
12054
FREE_LOCK(&lk);
5167
12055
locked = 0;
5168
if (pagedep_new_block &&
5169
(error = ffs_syncvnode(pvp, MNT_WAIT))) {
12056
if (pagedep_new_block && (error =
12057
ffs_syncvnode(pvp, MNT_WAIT, 0))) {
5170
12058
vput(pvp);
5171
12059
return (error);
5172
12060
}
5199
12087
* Flush all the dirty bitmaps associated with the block device
5200
12088
* before flushing the rest of the dirty blocks so as to reduce
5201
12089
* the number of dependencies that will have to be rolled back.
12090
*
12091
* XXX Unused?
5202
12092
*/
5203
12093
void
5204
12094
softdep_fsync_mountdev(vp)
5245
12135
}
5246
12136
5247
12137
/*
12138
* Sync all cylinder groups that were dirty at the time this function is
12139
* called. Newly dirtied cgs will be inserted before the sentinel. This
12140
* is used to flush freedep activity that may be holding up writes to a
12141
* indirect block.
12142
*/
12143
static int
12144
sync_cgs(mp, waitfor)
12145
struct mount *mp;
12146
int waitfor;
12147
{
12148
struct bmsafemap *bmsafemap;
12149
struct bmsafemap *sentinel;
12150
struct ufsmount *ump;
12151
struct buf *bp;
12152
int error;
12153
12154
sentinel = malloc(sizeof(*sentinel), M_BMSAFEMAP, M_ZERO | M_WAITOK);
12155
sentinel->sm_cg = -1;
12156
ump = VFSTOUFS(mp);
12157
error = 0;
12158
ACQUIRE_LOCK(&lk);
12159
LIST_INSERT_HEAD(&ump->softdep_dirtycg, sentinel, sm_next);
12160
for (bmsafemap = LIST_NEXT(sentinel, sm_next); bmsafemap != NULL;
12161
bmsafemap = LIST_NEXT(sentinel, sm_next)) {
12162
/* Skip sentinels and cgs with no work to release. */
12163
if (bmsafemap->sm_cg == -1 ||
12164
(LIST_EMPTY(&bmsafemap->sm_freehd) &&
12165
LIST_EMPTY(&bmsafemap->sm_freewr))) {
12166
LIST_REMOVE(sentinel, sm_next);
12167
LIST_INSERT_AFTER(bmsafemap, sentinel, sm_next);
12168
continue;
12169
}
12170
/*
12171
* If we don't get the lock and we're waiting try again, if
12172
* not move on to the next buf and try to sync it.
12173
*/
12174
bp = getdirtybuf(bmsafemap->sm_buf, &lk, waitfor);
12175
if (bp == NULL && waitfor == MNT_WAIT)
12176
continue;
12177
LIST_REMOVE(sentinel, sm_next);
12178
LIST_INSERT_AFTER(bmsafemap, sentinel, sm_next);
12179
if (bp == NULL)
12180
continue;
12181
FREE_LOCK(&lk);
12182
if (waitfor == MNT_NOWAIT)
12183
bawrite(bp);
12184
else
12185
error = bwrite(bp);
12186
ACQUIRE_LOCK(&lk);
12187
if (error)
12188
break;
12189
}
12190
LIST_REMOVE(sentinel, sm_next);
12191
FREE_LOCK(&lk);
12192
free(sentinel, M_BMSAFEMAP);
12193
return (error);
12194
}
12195
12196
/*
5248
12197
* This routine is called when we are trying to synchronously flush a
5249
12198
* file. This routine must eliminate any filesystem metadata dependencies
5250
* so that the syncing routine can succeed by pushing the dirty blocks
5251
* associated with the file. If any I/O errors occur, they are returned.
12199
* so that the syncing routine can succeed.
5252
12200
*/
5253
12201
int
5254
12202
softdep_sync_metadata(struct vnode *vp)
5255
12203
{
12204
int error;
12205
12206
/*
12207
* Ensure that any direct block dependencies have been cleared,
12208
* truncations are started, and inode references are journaled.
12209
*/
12210
ACQUIRE_LOCK(&lk);
12211
/*
12212
* Write all journal records to prevent rollbacks on devvp.
12213
*/
12214
if (vp->v_type == VCHR)
12215
softdep_flushjournal(vp->v_mount);
12216
error = flush_inodedep_deps(vp, vp->v_mount, VTOI(vp)->i_number);
12217
/*
12218
* Ensure that all truncates are written so we won't find deps on
12219
* indirect blocks.
12220
*/
12221
process_truncates(vp);
12222
FREE_LOCK(&lk);
12223
12224
return (error);
12225
}
12226
12227
/*
12228
* This routine is called when we are attempting to sync a buf with
12229
* dependencies. If waitfor is MNT_NOWAIT it attempts to schedule any
12230
* other IO it can but returns EBUSY if the buffer is not yet able to
12231
* be written. Dependencies which will not cause rollbacks will always
12232
* return 0.
12233
*/
12234
int
12235
softdep_sync_buf(struct vnode *vp, struct buf *bp, int waitfor)
12236
{
12237
struct indirdep *indirdep;
5256
12238
struct pagedep *pagedep;
5257
struct allocdirect *adp;
5258
12239
struct allocindir *aip;
5259
struct buf *bp, *nbp;
12240
struct newblk *newblk;
12241
struct buf *nbp;
5260
12242
struct worklist *wk;
5261
struct bufobj *bo;
5262
int i, error, waitfor;
5263
5264
if (!DOINGSOFTDEP(vp))
5265
return (0);
5266
/*
5267
* Ensure that any direct block dependencies have been cleared.
5268
*/
5269
ACQUIRE_LOCK(&lk);
5270
if ((error = flush_inodedep_deps(vp->v_mount, VTOI(vp)->i_number))) {
5271
FREE_LOCK(&lk);
5272
return (error);
5273
}
5274
FREE_LOCK(&lk);
5275
/*
5276
* For most files, the only metadata dependencies are the
5277
* cylinder group maps that allocate their inode or blocks.
5278
* The block allocation dependencies can be found by traversing
5279
* the dependency lists for any buffers that remain on their
5280
* dirty buffer list. The inode allocation dependency will
5281
* be resolved when the inode is updated with MNT_WAIT.
5282
* This work is done in two passes. The first pass grabs most
5283
* of the buffers and begins asynchronously writing them. The
5284
* only way to wait for these asynchronous writes is to sleep
5285
* on the filesystem vnode which may stay busy for a long time
5286
* if the filesystem is active. So, instead, we make a second
5287
* pass over the dependencies blocking on each write. In the
5288
* usual case we will be blocking against a write that we
5289
* initiated, so when it is done the dependency will have been
5290
* resolved. Thus the second pass is expected to end quickly.
5291
*/
5292
waitfor = MNT_NOWAIT;
5293
bo = &vp->v_bufobj;
5294
5295
top:
5296
/*
5297
* We must wait for any I/O in progress to finish so that
5298
* all potential buffers on the dirty list will be visible.
5299
*/
5300
BO_LOCK(bo);
5301
drain_output(vp);
5302
while ((bp = TAILQ_FIRST(&bo->bo_dirty.bv_hd)) != NULL) {
5303
bp = getdirtybuf(bp, BO_MTX(bo), MNT_WAIT);
5304
if (bp)
5305
break;
5306
}
5307
BO_UNLOCK(bo);
5308
if (bp == NULL)
5309
return (0);
5310
loop:
5311
/* While syncing snapshots, we must allow recursive lookups */
5312
BUF_AREC(bp);
12243
int i, error;
12244
12245
/*
12246
* For VCHR we just don't want to force flush any dependencies that
12247
* will cause rollbacks.
12248
*/
12249
if (vp->v_type == VCHR) {
12250
if (waitfor == MNT_NOWAIT && softdep_count_dependencies(bp, 0))
12251
return (EBUSY);
12252
return (0);
12253
}
5313
12254
ACQUIRE_LOCK(&lk);
5314
12255
/*
5315
12256
* As we hold the buffer locked, none of its dependencies
5316
12257
* will disappear.
5317
12258
*/
12259
error = 0;
12260
top:
5318
12261
LIST_FOREACH(wk, &bp->b_dep, wk_list) {
5319
12262
switch (wk->wk_type) {
5320
12263
5321
12264
case D_ALLOCDIRECT:
5322
adp = WK_ALLOCDIRECT(wk);
5323
if (adp->ad_state & DEPCOMPLETE)
5324
continue;
5325
nbp = adp->ad_buf;
5326
nbp = getdirtybuf(nbp, &lk, waitfor);
5327
if (nbp == NULL)
5328
continue;
5329
FREE_LOCK(&lk);
5330
if (waitfor == MNT_NOWAIT) {
5331
bawrite(nbp);
5332
} else if ((error = bwrite(nbp)) != 0) {
5333
break;
5334
}
5335
ACQUIRE_LOCK(&lk);
5336
continue;
5337
5338
12265
case D_ALLOCINDIR:
5339
aip = WK_ALLOCINDIR(wk);
5340
if (aip->ai_state & DEPCOMPLETE)
12266
newblk = WK_NEWBLK(wk);
12267
if (newblk->nb_jnewblk != NULL) {
12268
if (waitfor == MNT_NOWAIT) {
12269
error = EBUSY;
12270
goto out_unlock;
12271
}
12272
jwait(&newblk->nb_jnewblk->jn_list, waitfor);
12273
goto top;
12274
}
12275
if (newblk->nb_state & DEPCOMPLETE ||
12276
waitfor == MNT_NOWAIT)
5341
12277
continue;
5342
nbp = aip->ai_buf;
12278
nbp = newblk->nb_bmsafemap->sm_buf;
5343
12279
nbp = getdirtybuf(nbp, &lk, waitfor);
5344
12280
if (nbp == NULL)
5345
continue;
12281
goto top;
5346
12282
FREE_LOCK(&lk);
5347
if (waitfor == MNT_NOWAIT) {
5348
bawrite(nbp);
5349
} else if ((error = bwrite(nbp)) != 0) {
5350
break;
5351
}
12283
if ((error = bwrite(nbp)) != 0)
12284
goto out;
5352
12285
ACQUIRE_LOCK(&lk);
5353
12286
continue;
5354
12287
5355
12288
case D_INDIRDEP:
12289
indirdep = WK_INDIRDEP(wk);
12290
if (waitfor == MNT_NOWAIT) {
12291
if (!TAILQ_EMPTY(&indirdep->ir_trunc) ||
12292
!LIST_EMPTY(&indirdep->ir_deplisthd)) {
12293
error = EBUSY;
12294
goto out_unlock;
12295
}
12296
}
12297
if (!TAILQ_EMPTY(&indirdep->ir_trunc))
12298
panic("softdep_sync_buf: truncation pending.");
5356
12299
restart:
5357
5358
LIST_FOREACH(aip, &WK_INDIRDEP(wk)->ir_deplisthd, ai_next) {
5359
if (aip->ai_state & DEPCOMPLETE)
12300
LIST_FOREACH(aip, &indirdep->ir_deplisthd, ai_next) {
12301
newblk = (struct newblk *)aip;
12302
if (newblk->nb_jnewblk != NULL) {
12303
jwait(&newblk->nb_jnewblk->jn_list,
12304
waitfor);
12305
goto restart;
12306
}
12307
if (newblk->nb_state & DEPCOMPLETE)
5360
12308
continue;
5361
nbp = aip->ai_buf;
5362
nbp = getdirtybuf(nbp, &lk, MNT_WAIT);
12309
nbp = newblk->nb_bmsafemap->sm_buf;
12310
nbp = getdirtybuf(nbp, &lk, waitfor);
5363
12311
if (nbp == NULL)
5364
12312
goto restart;
5365
12313
FREE_LOCK(&lk);
5366
if ((error = bwrite(nbp)) != 0) {
5367
goto loop_end;
5368
}
12314
if ((error = bwrite(nbp)) != 0)
12315
goto out;
5369
12316
ACQUIRE_LOCK(&lk);
5370
12317
goto restart;
5371
12318
}
5372
12319
continue;
5373
12320
5374
case D_INODEDEP:
5375
if ((error = flush_inodedep_deps(wk->wk_mp,
5376
WK_INODEDEP(wk)->id_ino)) != 0) {
5377
FREE_LOCK(&lk);
5378
break;
5379
}
5380
continue;
5381
5382
12321
case D_PAGEDEP:
5383
12322
/*
12323
* Only flush directory entries in synchronous passes.
12324
*/
12325
if (waitfor != MNT_WAIT) {
12326
error = EBUSY;
12327
goto out_unlock;
12328
}
12329
/*
12330
* While syncing snapshots, we must allow recursive
12331
* lookups.
12332
*/
12333
BUF_AREC(bp);
12334
/*
5384
12335
* We are trying to sync a directory that may
5385
12336
* have dependencies on both its own metadata
5386
12337
* and/or dependencies on the inodes of any
5391
12342
for (i = 0; i < DAHASHSZ; i++) {
5392
12343
if (LIST_FIRST(&pagedep->pd_diraddhd[i]) == 0)
5393
12344
continue;
5394
if ((error =
5395
flush_pagedep_deps(vp, wk->wk_mp,
5396
&pagedep->pd_diraddhd[i]))) {
5397
FREE_LOCK(&lk);
5398
goto loop_end;
12345
if ((error = flush_pagedep_deps(vp, wk->wk_mp,
12346
&pagedep->pd_diraddhd[i]))) {
12347
BUF_NOREC(bp);
12348
goto out_unlock;
5399
12349
}
5400
12350
}
5401
continue;
5402
5403
case D_MKDIR:
5404
/*
5405
* This case should never happen if the vnode has
5406
* been properly sync'ed. However, if this function
5407
* is used at a place where the vnode has not yet
5408
* been sync'ed, this dependency can show up. So,
5409
* rather than panic, just flush it.
5410
*/
5411
nbp = WK_MKDIR(wk)->md_buf;
5412
nbp = getdirtybuf(nbp, &lk, waitfor);
5413
if (nbp == NULL)
5414
continue;
5415
FREE_LOCK(&lk);
5416
if (waitfor == MNT_NOWAIT) {
5417
bawrite(nbp);
5418
} else if ((error = bwrite(nbp)) != 0) {
5419
break;
5420
}
5421
ACQUIRE_LOCK(&lk);
5422
continue;
5423
5424
case D_BMSAFEMAP:
5425
/*
5426
* This case should never happen if the vnode has
5427
* been properly sync'ed. However, if this function
5428
* is used at a place where the vnode has not yet
5429
* been sync'ed, this dependency can show up. So,
5430
* rather than panic, just flush it.
5431
*/
5432
nbp = WK_BMSAFEMAP(wk)->sm_buf;
5433
nbp = getdirtybuf(nbp, &lk, waitfor);
5434
if (nbp == NULL)
5435
continue;
5436
FREE_LOCK(&lk);
5437
if (waitfor == MNT_NOWAIT) {
5438
bawrite(nbp);
5439
} else if ((error = bwrite(nbp)) != 0) {
5440
break;
5441
}
5442
ACQUIRE_LOCK(&lk);
12351
BUF_NOREC(bp);
12352
continue;
12353
12354
case D_FREEWORK:
12355
case D_FREEDEP:
12356
case D_JSEGDEP:
12357
case D_JNEWBLK:
5443
12358
continue;
5444
12359
5445
12360
default:
5446
panic("softdep_sync_metadata: Unknown type %s",
12361
panic("softdep_sync_buf: Unknown type %s",
5447
12362
TYPENAME(wk->wk_type));
5448
12363
/* NOTREACHED */
5449
12364
}
5450
loop_end:
5451
/* We reach here only in error and unlocked */
5452
if (error == 0)
5453
panic("softdep_sync_metadata: zero error");
5454
BUF_NOREC(bp);
5455
bawrite(bp);
5456
return (error);
5457
12365
}
12366
out_unlock:
5458
12367
FREE_LOCK(&lk);
5459
BO_LOCK(bo);
5460
while ((nbp = TAILQ_NEXT(bp, b_bobufs)) != NULL) {
5461
nbp = getdirtybuf(nbp, BO_MTX(bo), MNT_WAIT);
5462
if (nbp)
5463
break;
5464
}
5465
BO_UNLOCK(bo);
5466
BUF_NOREC(bp);
5467
bawrite(bp);
5468
if (nbp != NULL) {
5469
bp = nbp;
5470
goto loop;
5471
}
5472
/*
5473
* The brief unlock is to allow any pent up dependency
5474
* processing to be done. Then proceed with the second pass.
5475
*/
5476
if (waitfor == MNT_NOWAIT) {
5477
waitfor = MNT_WAIT;
5478
goto top;
5479
}
5480
5481
/*
5482
* If we have managed to get rid of all the dirty buffers,
5483
* then we are done. For certain directories and block
5484
* devices, we may need to do further work.
5485
*
5486
* We must wait for any I/O in progress to finish so that
5487
* all potential buffers on the dirty list will be visible.
5488
*/
5489
BO_LOCK(bo);
5490
drain_output(vp);
5491
BO_UNLOCK(bo);
5492
return (0);
12368
out:
12369
return (error);
5493
12370
}
5494
12371
5495
12372
/*
5497
12374
* Called with splbio blocked.
5498
12375
*/
5499
12376
static int
5500
flush_inodedep_deps(mp, ino)
12377
flush_inodedep_deps(vp, mp, ino)
12378
struct vnode *vp;
5501
12379
struct mount *mp;
5502
12380
ino_t ino;
5503
12381
{
5504
12382
struct inodedep *inodedep;
12383
struct inoref *inoref;
5505
12384
int error, waitfor;
5506
12385
5507
12386
/*
5522
12401
return (error);
5523
12402
FREE_LOCK(&lk);
5524
12403
ACQUIRE_LOCK(&lk);
12404
restart:
5525
12405
if (inodedep_lookup(mp, ino, 0, &inodedep) == 0)
5526
12406
return (0);
12407
TAILQ_FOREACH(inoref, &inodedep->id_inoreflst, if_deps) {
12408
if ((inoref->if_state & (DEPCOMPLETE | GOINGAWAY))
12409
== DEPCOMPLETE) {
12410
jwait(&inoref->if_list, MNT_WAIT);
12411
goto restart;
12412
}
12413
}
5527
12414
if (flush_deplist(&inodedep->id_inoupdt, waitfor, &error) ||
5528
12415
flush_deplist(&inodedep->id_newinoupdt, waitfor, &error) ||
5529
12416
flush_deplist(&inodedep->id_extupdt, waitfor, &error) ||
5555
12442
int *errorp;
5556
12443
{
5557
12444
struct allocdirect *adp;
12445
struct newblk *newblk;
5558
12446
struct buf *bp;
5559
12447
5560
12448
mtx_assert(&lk, MA_OWNED);
5561
12449
TAILQ_FOREACH(adp, listhead, ad_next) {
5562
if (adp->ad_state & DEPCOMPLETE)
12450
newblk = (struct newblk *)adp;
12451
if (newblk->nb_jnewblk != NULL) {
12452
jwait(&newblk->nb_jnewblk->jn_list, MNT_WAIT);
12453
return (1);
12454
}
12455
if (newblk->nb_state & DEPCOMPLETE)
5563
12456
continue;
5564
bp = adp->ad_buf;
12457
bp = newblk->nb_bmsafemap->sm_buf;
5565
12458
bp = getdirtybuf(bp, &lk, waitfor);
5566
12459
if (bp == NULL) {
5567
12460
if (waitfor == MNT_NOWAIT)
5569
12462
return (1);
5570
12463
}
5571
12464
FREE_LOCK(&lk);
5572
if (waitfor == MNT_NOWAIT) {
12465
if (waitfor == MNT_NOWAIT)
5573
12466
bawrite(bp);
5574
} else if ((*errorp = bwrite(bp)) != 0) {
5575
ACQUIRE_LOCK(&lk);
5576
return (1);
5577
}
12467
else
12468
*errorp = bwrite(bp);
5578
12469
ACQUIRE_LOCK(&lk);
5579
12470
return (1);
5580
12471
}
5582
12473
}
5583
12474
5584
12475
/*
12476
* Flush dependencies associated with an allocdirect block.
12477
*/
12478
static int
12479
flush_newblk_dep(vp, mp, lbn)
12480
struct vnode *vp;
12481
struct mount *mp;
12482
ufs_lbn_t lbn;
12483
{
12484
struct newblk *newblk;
12485
struct bufobj *bo;
12486
struct inode *ip;
12487
struct buf *bp;
12488
ufs2_daddr_t blkno;
12489
int error;
12490
12491
error = 0;
12492
bo = &vp->v_bufobj;
12493
ip = VTOI(vp);
12494
blkno = DIP(ip, i_db[lbn]);
12495
if (blkno == 0)
12496
panic("flush_newblk_dep: Missing block");
12497
ACQUIRE_LOCK(&lk);
12498
/*
12499
* Loop until all dependencies related to this block are satisfied.
12500
* We must be careful to restart after each sleep in case a write
12501
* completes some part of this process for us.
12502
*/
12503
for (;;) {
12504
if (newblk_lookup(mp, blkno, 0, &newblk) == 0) {
12505
FREE_LOCK(&lk);
12506
break;
12507
}
12508
if (newblk->nb_list.wk_type != D_ALLOCDIRECT)
12509
panic("flush_newblk_deps: Bad newblk %p", newblk);
12510
/*
12511
* Flush the journal.
12512
*/
12513
if (newblk->nb_jnewblk != NULL) {
12514
jwait(&newblk->nb_jnewblk->jn_list, MNT_WAIT);
12515
continue;
12516
}
12517
/*
12518
* Write the bitmap dependency.
12519
*/
12520
if ((newblk->nb_state & DEPCOMPLETE) == 0) {
12521
bp = newblk->nb_bmsafemap->sm_buf;
12522
bp = getdirtybuf(bp, &lk, MNT_WAIT);
12523
if (bp == NULL)
12524
continue;
12525
FREE_LOCK(&lk);
12526
error = bwrite(bp);
12527
if (error)
12528
break;
12529
ACQUIRE_LOCK(&lk);
12530
continue;
12531
}
12532
/*
12533
* Write the buffer.
12534
*/
12535
FREE_LOCK(&lk);
12536
BO_LOCK(bo);
12537
bp = gbincore(bo, lbn);
12538
if (bp != NULL) {
12539
error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
12540
LK_INTERLOCK, BO_MTX(bo));
12541
if (error == ENOLCK) {
12542
ACQUIRE_LOCK(&lk);
12543
continue; /* Slept, retry */
12544
}
12545
if (error != 0)
12546
break; /* Failed */
12547
if (bp->b_flags & B_DELWRI) {
12548
bremfree(bp);
12549
error = bwrite(bp);
12550
if (error)
12551
break;
12552
} else
12553
BUF_UNLOCK(bp);
12554
} else
12555
BO_UNLOCK(bo);
12556
/*
12557
* We have to wait for the direct pointers to
12558
* point at the newdirblk before the dependency
12559
* will go away.
12560
*/
12561
error = ffs_update(vp, 1);
12562
if (error)
12563
break;
12564
ACQUIRE_LOCK(&lk);
12565
}
12566
return (error);
12567
}
12568
12569
/*
5585
12570
* Eliminate a pagedep dependency by flushing out all its diradd dependencies.
5586
12571
* Called with splbio blocked.
5587
12572
*/
5592
12577
struct diraddhd *diraddhdp;
5593
12578
{
5594
12579
struct inodedep *inodedep;
12580
struct inoref *inoref;
5595
12581
struct ufsmount *ump;
5596
12582
struct diradd *dap;
5597
12583
struct vnode *vp;
5598
struct bufobj *bo;
5599
12584
int error = 0;
5600
12585
struct buf *bp;
5601
12586
ino_t inum;
5602
struct worklist *wk;
5603
12587
5604
12588
ump = VFSTOUFS(mp);
12589
restart:
5605
12590
while ((dap = LIST_FIRST(diraddhdp)) != NULL) {
5606
12591
/*
5607
12592
* Flush ourselves if this directory entry
5623
12608
/*
5624
12609
* A newly allocated directory must have its "." and
5625
12610
* ".." entries written out before its name can be
5626
* committed in its parent. We do not want or need
5627
* the full semantics of a synchronous ffs_syncvnode as
5628
* that may end up here again, once for each directory
5629
* level in the filesystem. Instead, we push the blocks
5630
* and wait for them to clear. We have to fsync twice
5631
* because the first call may choose to defer blocks
5632
* that still have dependencies, but deferral will
5633
* happen at most once.
12611
* committed in its parent.
5634
12612
*/
5635
12613
inum = dap->da_newinum;
12614
if (inodedep_lookup(UFSTOVFS(ump), inum, 0, &inodedep) == 0)
12615
panic("flush_pagedep_deps: lost inode1");
12616
/*
12617
* Wait for any pending journal adds to complete so we don't
12618
* cause rollbacks while syncing.
12619
*/
12620
TAILQ_FOREACH(inoref, &inodedep->id_inoreflst, if_deps) {
12621
if ((inoref->if_state & (DEPCOMPLETE | GOINGAWAY))
12622
== DEPCOMPLETE) {
12623
jwait(&inoref->if_list, MNT_WAIT);
12624
goto restart;
12625
}
12626
}
5636
12627
if (dap->da_state & MKDIR_BODY) {
5637
12628
FREE_LOCK(&lk);
5638
12629
if ((error = ffs_vgetf(mp, inum, LK_EXCLUSIVE, &vp,
5639
12630
FFSV_FORCEINSMQ)))
5640
12631
break;
5641
if ((error=ffs_syncvnode(vp, MNT_NOWAIT)) ||
5642
(error=ffs_syncvnode(vp, MNT_NOWAIT))) {
5643
vput(vp);
5644
break;
5645
}
5646
bo = &vp->v_bufobj;
5647
BO_LOCK(bo);
5648
drain_output(vp);
12632
error = flush_newblk_dep(vp, mp, 0);
5649
12633
/*
5650
* If first block is still dirty with a D_MKDIR
5651
* dependency then it needs to be written now.
12634
* If we still have the dependency we might need to
12635
* update the vnode to sync the new link count to
12636
* disk.
5652
12637
*/
5653
for (;;) {
5654
error = 0;
5655
bp = gbincore(bo, 0);
5656
if (bp == NULL)
5657
break; /* First block not present */
5658
error = BUF_LOCK(bp,
5659
LK_EXCLUSIVE |
5660
LK_SLEEPFAIL |
5661
LK_INTERLOCK,
5662
BO_MTX(bo));
5663
BO_LOCK(bo);
5664
if (error == ENOLCK)
5665
continue; /* Slept, retry */
5666
if (error != 0)
5667
break; /* Failed */
5668
if ((bp->b_flags & B_DELWRI) == 0) {
5669
BUF_UNLOCK(bp);
5670
break; /* Buffer not dirty */
5671
}
5672
for (wk = LIST_FIRST(&bp->b_dep);
5673
wk != NULL;
5674
wk = LIST_NEXT(wk, wk_list))
5675
if (wk->wk_type == D_MKDIR)
5676
break;
5677
if (wk == NULL)
5678
BUF_UNLOCK(bp); /* Dependency gone */
5679
else {
5680
/*
5681
* D_MKDIR dependency remains,
5682
* must write buffer to stable
5683
* storage.
5684
*/
5685
BO_UNLOCK(bo);
5686
bremfree(bp);
5687
error = bwrite(bp);
5688
BO_LOCK(bo);
5689
}
5690
break;
5691
}
5692
BO_UNLOCK(bo);
12638
if (error == 0 && dap == LIST_FIRST(diraddhdp))
12639
error = ffs_update(vp, 1);
5693
12640
vput(vp);
5694
12641
if (error != 0)
5695
break; /* Flushing of first block failed */
12642
break;
5696
12643
ACQUIRE_LOCK(&lk);
5697
12644
/*
5698
12645
* If that cleared dependencies, go on to next.
5699
12646
*/
5700
12647
if (dap != LIST_FIRST(diraddhdp))
5701
12648
continue;
5702
if (dap->da_state & MKDIR_BODY)
5703
panic("flush_pagedep_deps: MKDIR_BODY");
12649
if (dap->da_state & MKDIR_BODY) {
12650
inodedep_lookup(UFSTOVFS(ump), inum, 0,
12651
&inodedep);
12652
panic("flush_pagedep_deps: MKDIR_BODY "
12653
"inodedep %p dap %p vp %p",
12654
inodedep, dap, vp);
12655
}
5704
12656
}
5705
12657
/*
5706
12658
* Flush the inode on which the directory entry depends.
5719
12671
* If the inode still has bitmap dependencies,
5720
12672
* push them to disk.
5721
12673
*/
5722
if ((inodedep->id_state & DEPCOMPLETE) == 0) {
5723
bp = inodedep->id_buf;
12674
if ((inodedep->id_state & (DEPCOMPLETE | GOINGAWAY)) == 0) {
12675
bp = inodedep->id_bmsafemap->sm_buf;
5724
12676
bp = getdirtybuf(bp, &lk, MNT_WAIT);
5725
12677
if (bp == NULL)
5726
12678
goto retry;
5733
12685
}
5734
12686
/*
5735
12687
* If the inode is still sitting in a buffer waiting
5736
* to be written, push it to disk.
12688
* to be written or waiting for the link count to be
12689
* adjusted update it here to flush it to disk.
5737
12690
*/
5738
FREE_LOCK(&lk);
5739
if ((error = bread(ump->um_devvp,
5740
fsbtodb(ump->um_fs, ino_to_fsba(ump->um_fs, inum)),
5741
(int)ump->um_fs->fs_bsize, NOCRED, &bp)) != 0) {
5742
brelse(bp);
5743
break;
12691
if (dap == LIST_FIRST(diraddhdp)) {
12692
FREE_LOCK(&lk);
12693
if ((error = ffs_vgetf(mp, inum, LK_EXCLUSIVE, &vp,
12694
FFSV_FORCEINSMQ)))
12695
break;
12696
error = ffs_update(vp, 1);
12697
vput(vp);
12698
if (error)
12699
break;
12700
ACQUIRE_LOCK(&lk);
5744
12701
}
5745
if ((error = bwrite(bp)) != 0)
5746
break;
5747
ACQUIRE_LOCK(&lk);
5748
12702
/*
5749
12703
* If we have failed to get rid of all the dependencies
5750
12704
* then something is seriously wrong.
5751
12705
*/
5752
if (dap == LIST_FIRST(diraddhdp))
5753
panic("flush_pagedep_deps: flush failed");
12706
if (dap == LIST_FIRST(diraddhdp)) {
12707
inodedep_lookup(UFSTOVFS(ump), inum, 0, &inodedep);
12708
panic("flush_pagedep_deps: failed to flush "
12709
"inodedep %p ino %d dap %p", inodedep, inum, dap);
12710
}
5754
12711
}
5755
12712
if (error)
5756
12713
ACQUIRE_LOCK(&lk);
5768
12725
softdep_slowdown(vp)
5769
12726
struct vnode *vp;
5770
12727
{
12728
struct ufsmount *ump;
12729
int jlow;
5771
12730
int max_softdeps_hard;
5772
12731
5773
12732
ACQUIRE_LOCK(&lk);
12733
jlow = 0;
12734
/*
12735
* Check for journal space if needed.
12736
*/
12737
if (DOINGSUJ(vp)) {
12738
ump = VFSTOUFS(vp->v_mount);
12739
if (journal_space(ump, 0) == 0)
12740
jlow = 1;
12741
}
5774
12742
max_softdeps_hard = max_softdeps * 11 / 10;
5775
if (num_dirrem < max_softdeps_hard / 2 &&
5776
num_inodedep < max_softdeps_hard &&
12743
if (dep_current[D_DIRREM] < max_softdeps_hard / 2 &&
12744
dep_current[D_INODEDEP] < max_softdeps_hard &&
5777
12745
VFSTOUFS(vp->v_mount)->um_numindirdeps < maxindirdeps &&
5778
num_freeblkdep < max_softdeps_hard) {
12746
dep_current[D_FREEBLKS] < max_softdeps_hard && jlow == 0) {
5779
12747
FREE_LOCK(&lk);
5780
12748
return (0);
5781
12749
}
5782
if (VFSTOUFS(vp->v_mount)->um_numindirdeps >= maxindirdeps)
12750
if (VFSTOUFS(vp->v_mount)->um_numindirdeps >= maxindirdeps || jlow)
5783
12751
softdep_speedup();
5784
12752
stat_sync_limit_hit += 1;
5785
12753
FREE_LOCK(&lk);
12754
if (DOINGSUJ(vp))
12755
return (0);
5786
12756
return (1);
5787
12757
}
5788
12758
5789
12759
/*
5790
12760
* Called by the allocation routines when they are about to fail
5791
* in the hope that we can free up some disk space.
12761
* in the hope that we can free up the requested resource (inodes
12762
* or disk space).
5792
12763
*
5793
12764
* First check to see if the work list has anything on it. If it has,
5794
* clean up entries until we successfully free some space. Because this
5795
* process holds inodes locked, we cannot handle any remove requests
5796
* that might block on a locked inode as that could lead to deadlock.
5797
* If the worklist yields no free space, encourage the syncer daemon
5798
* to help us. In no event will we try for longer than tickdelay seconds.
12765
* clean up entries until we successfully free the requested resource.
12766
* Because this process holds inodes locked, we cannot handle any remove
12767
* requests that might block on a locked inode as that could lead to
12768
* deadlock. If the worklist yields none of the requested resource,
12769
* start syncing out vnodes to free up the needed space.
5799
12770
*/
5800
12771
int
5801
softdep_request_cleanup(fs, vp)
12772
softdep_request_cleanup(fs, vp, cred, resource)
5802
12773
struct fs *fs;
5803
12774
struct vnode *vp;
12775
struct ucred *cred;
12776
int resource;
5804
12777
{
5805
12778
struct ufsmount *ump;
12779
struct mount *mp;
12780
struct vnode *lvp, *mvp;
5806
12781
long starttime;
5807
12782
ufs2_daddr_t needed;
5808
12783
int error;
5809
12784
5810
ump = VTOI(vp)->i_ump;
12785
/*
12786
* If we are being called because of a process doing a
12787
* copy-on-write, then it is not safe to process any
12788
* worklist items as we will recurse into the copyonwrite
12789
* routine. This will result in an incoherent snapshot.
12790
* If the vnode that we hold is a snapshot, we must avoid
12791
* handling other resources that could cause deadlock.
12792
*/
12793
if ((curthread->td_pflags & TDP_COWINPROGRESS) || IS_SNAPSHOT(VTOI(vp)))
12794
return (0);
12795
12796
if (resource == FLUSH_BLOCKS_WAIT)
12797
stat_cleanup_blkrequests += 1;
12798
else
12799
stat_cleanup_inorequests += 1;
12800
12801
mp = vp->v_mount;
12802
ump = VFSTOUFS(mp);
5811
12803
mtx_assert(UFS_MTX(ump), MA_OWNED);
5812
needed = fs->fs_cstotal.cs_nbfree + fs->fs_contigsumsize;
5813
starttime = time_second + tickdelay;
5814
/*
5815
* If we are being called because of a process doing a
5816
* copy-on-write, then it is not safe to update the vnode
5817
* as we may recurse into the copy-on-write routine.
5818
*/
5819
if (!(curthread->td_pflags & TDP_COWINPROGRESS)) {
5820
UFS_UNLOCK(ump);
5821
error = ffs_update(vp, 1);
5822
UFS_LOCK(ump);
5823
if (error != 0)
5824
return (0);
5825
}
5826
while (fs->fs_pendingblocks > 0 && fs->fs_cstotal.cs_nbfree <= needed) {
5827
if (time_second > starttime)
5828
return (0);
5829
UFS_UNLOCK(ump);
12804
UFS_UNLOCK(ump);
12805
error = ffs_update(vp, 1);
12806
if (error != 0) {
12807
UFS_LOCK(ump);
12808
return (0);
12809
}
12810
/*
12811
* If we are in need of resources, consider pausing for
12812
* tickdelay to give ourselves some breathing room.
12813
*/
12814
ACQUIRE_LOCK(&lk);
12815
process_removes(vp);
12816
process_truncates(vp);
12817
request_cleanup(UFSTOVFS(ump), resource);
12818
FREE_LOCK(&lk);
12819
/*
12820
* Now clean up at least as many resources as we will need.
12821
*
12822
* When requested to clean up inodes, the number that are needed
12823
* is set by the number of simultaneous writers (mnt_writeopcount)
12824
* plus a bit of slop (2) in case some more writers show up while
12825
* we are cleaning.
12826
*
12827
* When requested to free up space, the amount of space that
12828
* we need is enough blocks to allocate a full-sized segment
12829
* (fs_contigsumsize). The number of such segments that will
12830
* be needed is set by the number of simultaneous writers
12831
* (mnt_writeopcount) plus a bit of slop (2) in case some more
12832
* writers show up while we are cleaning.
12833
*
12834
* Additionally, if we are unpriviledged and allocating space,
12835
* we need to ensure that we clean up enough blocks to get the
12836
* needed number of blocks over the threshhold of the minimum
12837
* number of blocks required to be kept free by the filesystem
12838
* (fs_minfree).
12839
*/
12840
if (resource == FLUSH_INODES_WAIT) {
12841
needed = vp->v_mount->mnt_writeopcount + 2;
12842
} else if (resource == FLUSH_BLOCKS_WAIT) {
12843
needed = (vp->v_mount->mnt_writeopcount + 2) *
12844
fs->fs_contigsumsize;
12845
if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0))
12846
needed += fragstoblks(fs,
12847
roundup((fs->fs_dsize * fs->fs_minfree / 100) -
12848
fs->fs_cstotal.cs_nffree, fs->fs_frag));
12849
} else {
12850
UFS_LOCK(ump);
12851
printf("softdep_request_cleanup: Unknown resource type %d\n",
12852
resource);
12853
return (0);
12854
}
12855
starttime = time_second;
12856
retry:
12857
if ((resource == FLUSH_BLOCKS_WAIT && ump->softdep_on_worklist > 0 &&
12858
fs->fs_cstotal.cs_nbfree <= needed) ||
12859
(resource == FLUSH_INODES_WAIT && fs->fs_pendinginodes > 0 &&
12860
fs->fs_cstotal.cs_nifree <= needed)) {
5830
12861
ACQUIRE_LOCK(&lk);
5831
12862
if (ump->softdep_on_worklist > 0 &&
5832
process_worklist_item(UFSTOVFS(ump), LK_NOWAIT) != -1) {
12863
process_worklist_item(UFSTOVFS(ump),
12864
ump->softdep_on_worklist, LK_NOWAIT) != 0)
5833
12865
stat_worklist_push += 1;
5834
FREE_LOCK(&lk);
5835
UFS_LOCK(ump);
5836
continue;
5837
}
5838
request_cleanup(UFSTOVFS(ump), FLUSH_REMOVE_WAIT);
5839
12866
FREE_LOCK(&lk);
5840
UFS_LOCK(ump);
5841
}
12867
}
12868
/*
12869
* If we still need resources and there are no more worklist
12870
* entries to process to obtain them, we have to start flushing
12871
* the dirty vnodes to force the release of additional requests
12872
* to the worklist that we can then process to reap addition
12873
* resources. We walk the vnodes associated with the mount point
12874
* until we get the needed worklist requests that we can reap.
12875
*/
12876
if ((resource == FLUSH_BLOCKS_WAIT &&
12877
fs->fs_cstotal.cs_nbfree <= needed) ||
12878
(resource == FLUSH_INODES_WAIT && fs->fs_pendinginodes > 0 &&
12879
fs->fs_cstotal.cs_nifree <= needed)) {
12880
MNT_VNODE_FOREACH_ALL(lvp, mp, mvp) {
12881
if (TAILQ_FIRST(&lvp->v_bufobj.bo_dirty.bv_hd) == 0) {
12882
VI_UNLOCK(lvp);
12883
continue;
12884
}
12885
if (vget(lvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_NOWAIT,
12886
curthread))
12887
continue;
12888
if (lvp->v_vflag & VV_NOSYNC) { /* unlinked */
12889
vput(lvp);
12890
continue;
12891
}
12892
(void) ffs_syncvnode(lvp, MNT_NOWAIT, 0);
12893
vput(lvp);
12894
}
12895
lvp = ump->um_devvp;
12896
if (vn_lock(lvp, LK_EXCLUSIVE | LK_NOWAIT) == 0) {
12897
VOP_FSYNC(lvp, MNT_NOWAIT, curthread);
12898
VOP_UNLOCK(lvp, 0);
12899
}
12900
if (ump->softdep_on_worklist > 0) {
12901
stat_cleanup_retries += 1;
12902
goto retry;
12903
}
12904
stat_cleanup_failures += 1;
12905
}
12906
if (time_second - starttime > stat_cleanup_high_delay)
12907
stat_cleanup_high_delay = time_second - starttime;
12908
UFS_LOCK(ump);
5842
12909
return (1);
5843
12910
}
5844
12911
5872
12939
*/
5873
12940
if (ump->softdep_on_worklist > max_softdeps / 10) {
5874
12941
td->td_pflags |= TDP_SOFTDEP;
5875
process_worklist_item(mp, LK_NOWAIT);
5876
process_worklist_item(mp, LK_NOWAIT);
12942
process_worklist_item(mp, 2, LK_NOWAIT);
5877
12943
td->td_pflags &= ~TDP_SOFTDEP;
5878
12944
stat_worklist_push += 2;
5879
12945
return(1);
5882
12948
* Next, we attempt to speed up the syncer process. If that
5883
12949
* is successful, then we allow the process to continue.
5884
12950
*/
5885
if (softdep_speedup() && resource != FLUSH_REMOVE_WAIT)
12951
if (softdep_speedup() &&
12952
resource != FLUSH_BLOCKS_WAIT &&
12953
resource != FLUSH_INODES_WAIT)
5886
12954
return(0);
5887
12955
/*
5888
12956
* If we are resource constrained on inode dependencies, try
5897
12965
switch (resource) {
5898
12966
5899
12967
case FLUSH_INODES:
12968
case FLUSH_INODES_WAIT:
5900
12969
stat_ino_limit_push += 1;
5901
12970
req_clear_inodedeps += 1;
5902
12971
stat_countp = &stat_ino_limit_hit;
5903
12972
break;
5904
12973
5905
case FLUSH_REMOVE:
5906
case FLUSH_REMOVE_WAIT:
12974
case FLUSH_BLOCKS:
12975
case FLUSH_BLOCKS_WAIT:
5907
12976
stat_blk_limit_push += 1;
5908
12977
req_clear_remove += 1;
5909
12978
stat_countp = &stat_blk_limit_hit;
5965
13034
5966
13035
mtx_assert(&lk, MA_OWNED);
5967
13036
5968
for (cnt = 0; cnt < pagedep_hash; cnt++) {
13037
for (cnt = 0; cnt <= pagedep_hash; cnt++) {
5969
13038
pagedephd = &pagedep_hashtbl[next++];
5970
if (next >= pagedep_hash)
13039
if (next > pagedep_hash)
5971
13040
next = 0;
5972
13041
LIST_FOREACH(pagedep, pagedephd, pd_hash) {
5973
13042
if (LIST_EMPTY(&pagedep->pd_dirremhd))
5991
13060
softdep_error("clear_remove: vget", error);
5992
13061
goto finish_write;
5993
13062
}
5994
if ((error = ffs_syncvnode(vp, MNT_NOWAIT)))
13063
if ((error = ffs_syncvnode(vp, MNT_NOWAIT, 0)))
5995
13064
softdep_error("clear_remove: fsync", error);
5996
13065
bo = &vp->v_bufobj;
5997
13066
BO_LOCK(bo);
6029
13098
* We will then gather up all the inodes in its block
6030
13099
* that have dependencies and flush them out.
6031
13100
*/
6032
for (cnt = 0; cnt < inodedep_hash; cnt++) {
13101
for (cnt = 0; cnt <= inodedep_hash; cnt++) {
6033
13102
inodedephd = &inodedep_hashtbl[next++];
6034
if (next >= inodedep_hash)
13103
if (next > inodedep_hash)
6035
13104
next = 0;
6036
13105
if ((inodedep = LIST_FIRST(inodedephd)) != NULL)
6037
13106
break;
6074
13143
}
6075
13144
vfs_unbusy(mp);
6076
13145
if (ino == lastino) {
6077
if ((error = ffs_syncvnode(vp, MNT_WAIT)))
13146
if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)))
6078
13147
softdep_error("clear_inodedeps: fsync1", error);
6079
13148
} else {
6080
if ((error = ffs_syncvnode(vp, MNT_NOWAIT)))
13149
if ((error = ffs_syncvnode(vp, MNT_NOWAIT, 0)))
6081
13150
softdep_error("clear_inodedeps: fsync2", error);
6082
13151
BO_LOCK(&vp->v_bufobj);
6083
13152
drain_output(vp);
6089
13158
}
6090
13159
}
6091
13160
13161
void
13162
softdep_buf_append(bp, wkhd)
13163
struct buf *bp;
13164
struct workhead *wkhd;
13165
{
13166
struct worklist *wk;
13167
13168
ACQUIRE_LOCK(&lk);
13169
while ((wk = LIST_FIRST(wkhd)) != NULL) {
13170
WORKLIST_REMOVE(wk);
13171
WORKLIST_INSERT(&bp->b_dep, wk);
13172
}
13173
FREE_LOCK(&lk);
13174
13175
}
13176
13177
void
13178
softdep_inode_append(ip, cred, wkhd)
13179
struct inode *ip;
13180
struct ucred *cred;
13181
struct workhead *wkhd;
13182
{
13183
struct buf *bp;
13184
struct fs *fs;
13185
int error;
13186
13187
fs = ip->i_fs;
13188
error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
13189
(int)fs->fs_bsize, cred, &bp);
13190
if (error) {
13191
bqrelse(bp);
13192
softdep_freework(wkhd);
13193
return;
13194
}
13195
softdep_buf_append(bp, wkhd);
13196
bqrelse(bp);
13197
}
13198
13199
void
13200
softdep_freework(wkhd)
13201
struct workhead *wkhd;
13202
{
13203
13204
ACQUIRE_LOCK(&lk);
13205
handle_jwork(wkhd);
13206
FREE_LOCK(&lk);
13207
}
13208
6092
13209
/*
6093
13210
* Function to determine if the buffer has outstanding dependencies
6094
13211
* that will cause a roll-back if the buffer is written. If wantcount
6100
13217
int wantcount;
6101
13218
{
6102
13219
struct worklist *wk;
13220
struct bmsafemap *bmsafemap;
13221
struct freework *freework;
6103
13222
struct inodedep *inodedep;
6104
13223
struct indirdep *indirdep;
13224
struct freeblks *freeblks;
6105
13225
struct allocindir *aip;
6106
13226
struct pagedep *pagedep;
13227
struct dirrem *dirrem;
13228
struct newblk *newblk;
13229
struct mkdir *mkdir;
6107
13230
struct diradd *dap;
6108
13231
int i, retval;
6109
13232
6132
13255
if (!wantcount)
6133
13256
goto out;
6134
13257
}
13258
if (TAILQ_FIRST(&inodedep->id_inoreflst)) {
13259
/* Add reference dependency. */
13260
retval += 1;
13261
if (!wantcount)
13262
goto out;
13263
}
6135
13264
continue;
6136
13265
6137
13266
case D_INDIRDEP:
6138
13267
indirdep = WK_INDIRDEP(wk);
6139
13268
13269
TAILQ_FOREACH(freework, &indirdep->ir_trunc, fw_next) {
13270
/* indirect truncation dependency */
13271
retval += 1;
13272
if (!wantcount)
13273
goto out;
13274
}
13275
6140
13276
LIST_FOREACH(aip, &indirdep->ir_deplisthd, ai_next) {
6141
13277
/* indirect block pointer dependency */
6142
13278
retval += 1;
6147
13283
6148
13284
case D_PAGEDEP:
6149
13285
pagedep = WK_PAGEDEP(wk);
13286
LIST_FOREACH(dirrem, &pagedep->pd_dirremhd, dm_next) {
13287
if (LIST_FIRST(&dirrem->dm_jremrefhd)) {
13288
/* Journal remove ref dependency. */
13289
retval += 1;
13290
if (!wantcount)
13291
goto out;
13292
}
13293
}
6150
13294
for (i = 0; i < DAHASHSZ; i++) {
6151
13295
6152
13296
LIST_FOREACH(dap, &pagedep->pd_diraddhd[i], da_pdlist) {
6159
13303
continue;
6160
13304
6161
13305
case D_BMSAFEMAP:
13306
bmsafemap = WK_BMSAFEMAP(wk);
13307
if (LIST_FIRST(&bmsafemap->sm_jaddrefhd)) {
13308
/* Add reference dependency. */
13309
retval += 1;
13310
if (!wantcount)
13311
goto out;
13312
}
13313
if (LIST_FIRST(&bmsafemap->sm_jnewblkhd)) {
13314
/* Allocate block dependency. */
13315
retval += 1;
13316
if (!wantcount)
13317
goto out;
13318
}
13319
continue;
13320
13321
case D_FREEBLKS:
13322
freeblks = WK_FREEBLKS(wk);
13323
if (LIST_FIRST(&freeblks->fb_jblkdephd)) {
13324
/* Freeblk journal dependency. */
13325
retval += 1;
13326
if (!wantcount)
13327
goto out;
13328
}
13329
continue;
13330
6162
13331
case D_ALLOCDIRECT:
6163
13332
case D_ALLOCINDIR:
13333
newblk = WK_NEWBLK(wk);
13334
if (newblk->nb_jnewblk) {
13335
/* Journal allocate dependency. */
13336
retval += 1;
13337
if (!wantcount)
13338
goto out;
13339
}
13340
continue;
13341
6164
13342
case D_MKDIR:
13343
mkdir = WK_MKDIR(wk);
13344
if (mkdir->md_jaddref) {
13345
/* Journal reference dependency. */
13346
retval += 1;
13347
if (!wantcount)
13348
goto out;
13349
}
13350
continue;
13351
13352
case D_FREEWORK:
13353
case D_FREEDEP:
13354
case D_JSEGDEP:
13355
case D_JSEG:
13356
case D_SBDEP:
6165
13357
/* never a dependency on these blocks */
6166
13358
continue;
6167
13359
6168
13360
default:
6169
panic("softdep_check_for_rollback: Unexpected type %s",
13361
panic("softdep_count_dependencies: Unexpected type %s",
6170
13362
TYPENAME(wk->wk_type));
6171
13363
/* NOTREACHED */
6172
13364
}
6382
13574
6383
13575
#ifdef DDB
6384
13576
13577
static void
13578
inodedep_print(struct inodedep *inodedep, int verbose)
13579
{
13580
db_printf("%p fs %p st %x ino %jd inoblk %jd delta %d nlink %d"
13581
" saveino %p\n",
13582
inodedep, inodedep->id_fs, inodedep->id_state,
13583
(intmax_t)inodedep->id_ino,
13584
(intmax_t)fsbtodb(inodedep->id_fs,
13585
ino_to_fsba(inodedep->id_fs, inodedep->id_ino)),
13586
inodedep->id_nlinkdelta, inodedep->id_savednlink,
13587
inodedep->id_savedino1);
13588
13589
if (verbose == 0)
13590
return;
13591
13592
db_printf("\tpendinghd %p, bufwait %p, inowait %p, inoreflst %p, "
13593
"mkdiradd %p\n",
13594
LIST_FIRST(&inodedep->id_pendinghd),
13595
LIST_FIRST(&inodedep->id_bufwait),
13596
LIST_FIRST(&inodedep->id_inowait),
13597
TAILQ_FIRST(&inodedep->id_inoreflst),
13598
inodedep->id_mkdiradd);
13599
db_printf("\tinoupdt %p, newinoupdt %p, extupdt %p, newextupdt %p\n",
13600
TAILQ_FIRST(&inodedep->id_inoupdt),
13601
TAILQ_FIRST(&inodedep->id_newinoupdt),
13602
TAILQ_FIRST(&inodedep->id_extupdt),
13603
TAILQ_FIRST(&inodedep->id_newextupdt));
13604
}
13605
13606
DB_SHOW_COMMAND(inodedep, db_show_inodedep)
13607
{
13608
13609
if (have_addr == 0) {
13610
db_printf("Address required\n");
13611
return;
13612
}
13613
inodedep_print((struct inodedep*)addr, 1);
13614
}
13615
6385
13616
DB_SHOW_COMMAND(inodedeps, db_show_inodedeps)
6386
13617
{
6387
13618
struct inodedep_hashhead *inodedephd;
6395
13626
LIST_FOREACH(inodedep, inodedephd, id_hash) {
6396
13627
if (fs != NULL && fs != inodedep->id_fs)
6397
13628
continue;
6398
db_printf("%p fs %p st %x ino %jd inoblk %jd\n",
6399
inodedep, inodedep->id_fs, inodedep->id_state,
6400
(intmax_t)inodedep->id_ino,
6401
(intmax_t)fsbtodb(inodedep->id_fs,
6402
ino_to_fsba(inodedep->id_fs, inodedep->id_ino)));
13629
inodedep_print(inodedep, 0);
6403
13630
}
6404
13631
}
6405
13632
}
6406
13633
13634
DB_SHOW_COMMAND(worklist, db_show_worklist)
13635
{
13636
struct worklist *wk;
13637
13638
if (have_addr == 0) {
13639
db_printf("Address required\n");
13640
return;
13641
}
13642
wk = (struct worklist *)addr;
13643
printf("worklist: %p type %s state 0x%X\n",
13644
wk, TYPENAME(wk->wk_type), wk->wk_state);
13645
}
13646
13647
DB_SHOW_COMMAND(workhead, db_show_workhead)
13648
{
13649
struct workhead *wkhd;
13650
struct worklist *wk;
13651
int i;
13652
13653
if (have_addr == 0) {
13654
db_printf("Address required\n");
13655
return;
13656
}
13657
wkhd = (struct workhead *)addr;
13658
wk = LIST_FIRST(wkhd);
13659
for (i = 0; i < 100 && wk != NULL; i++, wk = LIST_NEXT(wk, wk_list))
13660
db_printf("worklist: %p type %s state 0x%X",
13661
wk, TYPENAME(wk->wk_type), wk->wk_state);
13662
if (i == 100)
13663
db_printf("workhead overflow");
13664
printf("\n");
13665
}
13666
13667
13668
DB_SHOW_COMMAND(mkdirs, db_show_mkdirs)
13669
{
13670
struct jaddref *jaddref;
13671
struct diradd *diradd;
13672
struct mkdir *mkdir;
13673
13674
LIST_FOREACH(mkdir, &mkdirlisthd, md_mkdirs) {
13675
diradd = mkdir->md_diradd;
13676
db_printf("mkdir: %p state 0x%X dap %p state 0x%X",
13677
mkdir, mkdir->md_state, diradd, diradd->da_state);
13678
if ((jaddref = mkdir->md_jaddref) != NULL)
13679
db_printf(" jaddref %p jaddref state 0x%X",
13680
jaddref, jaddref->ja_state);
13681
db_printf("\n");
13682
}
13683
}
13684
6407
13685
#endif /* DDB */
6408
13686
6409
13687
#endif /* SOFTUPDATES */
Loggerhead is a web-based interface for Breezy
Version: 2.0.1