2
* Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
5
* This program is free software; you can redistribute it and/or
6
* modify it under the terms of the GNU General Public License as
7
* published by the Free Software Foundation.
9
* This program is distributed in the hope that it would be useful,
10
* but WITHOUT ANY WARRANTY; without even the implied warranty of
11
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12
* GNU General Public License for more details.
14
* You should have received a copy of the GNU General Public License
15
* along with this program; if not, write the Free Software Foundation,
16
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20
#include "xfs_types.h"
24
#include "xfs_trans.h"
27
#include "xfs_mount.h"
28
#include "xfs_bmap_btree.h"
29
#include "xfs_alloc_btree.h"
30
#include "xfs_ialloc_btree.h"
31
#include "xfs_dinode.h"
32
#include "xfs_inode.h"
33
#include "xfs_btree.h"
34
#include "xfs_alloc.h"
35
#include "xfs_error.h"
36
#include "xfs_trace.h"
39
#define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b)))
41
#define XFSA_FIXUP_BNO_OK 1
42
#define XFSA_FIXUP_CNT_OK 2
44
STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *);
45
STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *);
46
STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *);
47
STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *,
48
xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *);
49
STATIC void xfs_alloc_busy_trim(struct xfs_alloc_arg *,
50
xfs_agblock_t, xfs_extlen_t, xfs_agblock_t *, xfs_extlen_t *);
53
* Lookup the record equal to [bno, len] in the btree given by cur.
55
STATIC int /* error */
57
struct xfs_btree_cur *cur, /* btree cursor */
58
xfs_agblock_t bno, /* starting block of extent */
59
xfs_extlen_t len, /* length of extent */
60
int *stat) /* success/failure */
62
cur->bc_rec.a.ar_startblock = bno;
63
cur->bc_rec.a.ar_blockcount = len;
64
return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
68
* Lookup the first record greater than or equal to [bno, len]
69
* in the btree given by cur.
71
STATIC int /* error */
73
struct xfs_btree_cur *cur, /* btree cursor */
74
xfs_agblock_t bno, /* starting block of extent */
75
xfs_extlen_t len, /* length of extent */
76
int *stat) /* success/failure */
78
cur->bc_rec.a.ar_startblock = bno;
79
cur->bc_rec.a.ar_blockcount = len;
80
return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
84
* Lookup the first record less than or equal to [bno, len]
85
* in the btree given by cur.
89
struct xfs_btree_cur *cur, /* btree cursor */
90
xfs_agblock_t bno, /* starting block of extent */
91
xfs_extlen_t len, /* length of extent */
92
int *stat) /* success/failure */
94
cur->bc_rec.a.ar_startblock = bno;
95
cur->bc_rec.a.ar_blockcount = len;
96
return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
100
* Update the record referred to by cur to the value given
102
* This either works (return 0) or gets an EFSCORRUPTED error.
104
STATIC int /* error */
106
struct xfs_btree_cur *cur, /* btree cursor */
107
xfs_agblock_t bno, /* starting block of extent */
108
xfs_extlen_t len) /* length of extent */
110
union xfs_btree_rec rec;
112
rec.alloc.ar_startblock = cpu_to_be32(bno);
113
rec.alloc.ar_blockcount = cpu_to_be32(len);
114
return xfs_btree_update(cur, &rec);
118
* Get the data from the pointed-to record.
122
struct xfs_btree_cur *cur, /* btree cursor */
123
xfs_agblock_t *bno, /* output: starting block of extent */
124
xfs_extlen_t *len, /* output: length of extent */
125
int *stat) /* output: success/failure */
127
union xfs_btree_rec *rec;
130
error = xfs_btree_get_rec(cur, &rec, stat);
131
if (!error && *stat == 1) {
132
*bno = be32_to_cpu(rec->alloc.ar_startblock);
133
*len = be32_to_cpu(rec->alloc.ar_blockcount);
139
* Compute aligned version of the found extent.
140
* Takes alignment and min length into account.
143
xfs_alloc_compute_aligned(
144
xfs_alloc_arg_t *args, /* allocation argument structure */
145
xfs_agblock_t foundbno, /* starting block in found extent */
146
xfs_extlen_t foundlen, /* length in found extent */
147
xfs_agblock_t *resbno, /* result block number */
148
xfs_extlen_t *reslen) /* result length */
153
/* Trim busy sections out of found extent */
154
xfs_alloc_busy_trim(args, foundbno, foundlen, &bno, &len);
156
if (args->alignment > 1 && len >= args->minlen) {
157
xfs_agblock_t aligned_bno = roundup(bno, args->alignment);
158
xfs_extlen_t diff = aligned_bno - bno;
160
*resbno = aligned_bno;
161
*reslen = diff >= len ? 0 : len - diff;
169
* Compute best start block and diff for "near" allocations.
170
* freelen >= wantlen already checked by caller.
172
STATIC xfs_extlen_t /* difference value (absolute) */
173
xfs_alloc_compute_diff(
174
xfs_agblock_t wantbno, /* target starting block */
175
xfs_extlen_t wantlen, /* target length */
176
xfs_extlen_t alignment, /* target alignment */
177
xfs_agblock_t freebno, /* freespace's starting block */
178
xfs_extlen_t freelen, /* freespace's length */
179
xfs_agblock_t *newbnop) /* result: best start block from free */
181
xfs_agblock_t freeend; /* end of freespace extent */
182
xfs_agblock_t newbno1; /* return block number */
183
xfs_agblock_t newbno2; /* other new block number */
184
xfs_extlen_t newlen1=0; /* length with newbno1 */
185
xfs_extlen_t newlen2=0; /* length with newbno2 */
186
xfs_agblock_t wantend; /* end of target extent */
188
ASSERT(freelen >= wantlen);
189
freeend = freebno + freelen;
190
wantend = wantbno + wantlen;
191
if (freebno >= wantbno) {
192
if ((newbno1 = roundup(freebno, alignment)) >= freeend)
193
newbno1 = NULLAGBLOCK;
194
} else if (freeend >= wantend && alignment > 1) {
195
newbno1 = roundup(wantbno, alignment);
196
newbno2 = newbno1 - alignment;
197
if (newbno1 >= freeend)
198
newbno1 = NULLAGBLOCK;
200
newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1);
201
if (newbno2 < freebno)
202
newbno2 = NULLAGBLOCK;
204
newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2);
205
if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) {
206
if (newlen1 < newlen2 ||
207
(newlen1 == newlen2 &&
208
XFS_ABSDIFF(newbno1, wantbno) >
209
XFS_ABSDIFF(newbno2, wantbno)))
211
} else if (newbno2 != NULLAGBLOCK)
213
} else if (freeend >= wantend) {
215
} else if (alignment > 1) {
216
newbno1 = roundup(freeend - wantlen, alignment);
217
if (newbno1 > freeend - wantlen &&
218
newbno1 - alignment >= freebno)
219
newbno1 -= alignment;
220
else if (newbno1 >= freeend)
221
newbno1 = NULLAGBLOCK;
223
newbno1 = freeend - wantlen;
225
return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno);
229
* Fix up the length, based on mod and prod.
230
* len should be k * prod + mod for some k.
231
* If len is too small it is returned unchanged.
232
* If len hits maxlen it is left alone.
236
xfs_alloc_arg_t *args) /* allocation argument structure */
241
ASSERT(args->mod < args->prod);
243
ASSERT(rlen >= args->minlen);
244
ASSERT(rlen <= args->maxlen);
245
if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen ||
246
(args->mod == 0 && rlen < args->prod))
248
k = rlen % args->prod;
252
if ((int)(rlen = rlen - k - args->mod) < (int)args->minlen)
255
if ((int)(rlen = rlen - args->prod - (args->mod - k)) <
259
ASSERT(rlen >= args->minlen);
260
ASSERT(rlen <= args->maxlen);
265
* Fix up length if there is too little space left in the a.g.
266
* Return 1 if ok, 0 if too little, should give up.
269
xfs_alloc_fix_minleft(
270
xfs_alloc_arg_t *args) /* allocation argument structure */
272
xfs_agf_t *agf; /* a.g. freelist header */
273
int diff; /* free space difference */
275
if (args->minleft == 0)
277
agf = XFS_BUF_TO_AGF(args->agbp);
278
diff = be32_to_cpu(agf->agf_freeblks)
279
- args->len - args->minleft;
282
args->len += diff; /* shrink the allocated space */
283
if (args->len >= args->minlen)
285
args->agbno = NULLAGBLOCK;
290
* Update the two btrees, logically removing from freespace the extent
291
* starting at rbno, rlen blocks. The extent is contained within the
292
* actual (current) free extent fbno for flen blocks.
293
* Flags are passed in indicating whether the cursors are set to the
296
STATIC int /* error code */
297
xfs_alloc_fixup_trees(
298
xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */
299
xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */
300
xfs_agblock_t fbno, /* starting block of free extent */
301
xfs_extlen_t flen, /* length of free extent */
302
xfs_agblock_t rbno, /* starting block of returned extent */
303
xfs_extlen_t rlen, /* length of returned extent */
304
int flags) /* flags, XFSA_FIXUP_... */
306
int error; /* error code */
307
int i; /* operation results */
308
xfs_agblock_t nfbno1; /* first new free startblock */
309
xfs_agblock_t nfbno2; /* second new free startblock */
310
xfs_extlen_t nflen1=0; /* first new free length */
311
xfs_extlen_t nflen2=0; /* second new free length */
314
* Look up the record in the by-size tree if necessary.
316
if (flags & XFSA_FIXUP_CNT_OK) {
318
if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i)))
320
XFS_WANT_CORRUPTED_RETURN(
321
i == 1 && nfbno1 == fbno && nflen1 == flen);
324
if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i)))
326
XFS_WANT_CORRUPTED_RETURN(i == 1);
329
* Look up the record in the by-block tree if necessary.
331
if (flags & XFSA_FIXUP_BNO_OK) {
333
if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i)))
335
XFS_WANT_CORRUPTED_RETURN(
336
i == 1 && nfbno1 == fbno && nflen1 == flen);
339
if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i)))
341
XFS_WANT_CORRUPTED_RETURN(i == 1);
345
if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) {
346
struct xfs_btree_block *bnoblock;
347
struct xfs_btree_block *cntblock;
349
bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_bufs[0]);
350
cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_bufs[0]);
352
XFS_WANT_CORRUPTED_RETURN(
353
bnoblock->bb_numrecs == cntblock->bb_numrecs);
358
* Deal with all four cases: the allocated record is contained
359
* within the freespace record, so we can have new freespace
360
* at either (or both) end, or no freespace remaining.
362
if (rbno == fbno && rlen == flen)
363
nfbno1 = nfbno2 = NULLAGBLOCK;
364
else if (rbno == fbno) {
365
nfbno1 = rbno + rlen;
366
nflen1 = flen - rlen;
367
nfbno2 = NULLAGBLOCK;
368
} else if (rbno + rlen == fbno + flen) {
370
nflen1 = flen - rlen;
371
nfbno2 = NULLAGBLOCK;
374
nflen1 = rbno - fbno;
375
nfbno2 = rbno + rlen;
376
nflen2 = (fbno + flen) - nfbno2;
379
* Delete the entry from the by-size btree.
381
if ((error = xfs_btree_delete(cnt_cur, &i)))
383
XFS_WANT_CORRUPTED_RETURN(i == 1);
385
* Add new by-size btree entry(s).
387
if (nfbno1 != NULLAGBLOCK) {
388
if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i)))
390
XFS_WANT_CORRUPTED_RETURN(i == 0);
391
if ((error = xfs_btree_insert(cnt_cur, &i)))
393
XFS_WANT_CORRUPTED_RETURN(i == 1);
395
if (nfbno2 != NULLAGBLOCK) {
396
if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i)))
398
XFS_WANT_CORRUPTED_RETURN(i == 0);
399
if ((error = xfs_btree_insert(cnt_cur, &i)))
401
XFS_WANT_CORRUPTED_RETURN(i == 1);
404
* Fix up the by-block btree entry(s).
406
if (nfbno1 == NULLAGBLOCK) {
408
* No remaining freespace, just delete the by-block tree entry.
410
if ((error = xfs_btree_delete(bno_cur, &i)))
412
XFS_WANT_CORRUPTED_RETURN(i == 1);
415
* Update the by-block entry to start later|be shorter.
417
if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1)))
420
if (nfbno2 != NULLAGBLOCK) {
422
* 2 resulting free entries, need to add one.
424
if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i)))
426
XFS_WANT_CORRUPTED_RETURN(i == 0);
427
if ((error = xfs_btree_insert(bno_cur, &i)))
429
XFS_WANT_CORRUPTED_RETURN(i == 1);
435
* Read in the allocation group free block array.
437
STATIC int /* error */
439
xfs_mount_t *mp, /* mount point structure */
440
xfs_trans_t *tp, /* transaction pointer */
441
xfs_agnumber_t agno, /* allocation group number */
442
xfs_buf_t **bpp) /* buffer for the ag free block array */
444
xfs_buf_t *bp; /* return value */
447
ASSERT(agno != NULLAGNUMBER);
448
error = xfs_trans_read_buf(
449
mp, tp, mp->m_ddev_targp,
450
XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
451
XFS_FSS_TO_BB(mp, 1), 0, &bp);
454
ASSERT(!xfs_buf_geterror(bp));
455
xfs_buf_set_ref(bp, XFS_AGFL_REF);
461
xfs_alloc_update_counters(
462
struct xfs_trans *tp,
463
struct xfs_perag *pag,
464
struct xfs_buf *agbp,
467
struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp);
469
pag->pagf_freeblks += len;
470
be32_add_cpu(&agf->agf_freeblks, len);
472
xfs_trans_agblocks_delta(tp, len);
473
if (unlikely(be32_to_cpu(agf->agf_freeblks) >
474
be32_to_cpu(agf->agf_length)))
477
xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
482
* Allocation group level functions.
486
* Allocate a variable extent in the allocation group agno.
487
* Type and bno are used to determine where in the allocation group the
489
* Extent's length (returned in *len) will be between minlen and maxlen,
490
* and of the form k * prod + mod unless there's nothing that large.
491
* Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
493
STATIC int /* error */
494
xfs_alloc_ag_vextent(
495
xfs_alloc_arg_t *args) /* argument structure for allocation */
499
ASSERT(args->minlen > 0);
500
ASSERT(args->maxlen > 0);
501
ASSERT(args->minlen <= args->maxlen);
502
ASSERT(args->mod < args->prod);
503
ASSERT(args->alignment > 0);
505
* Branch to correct routine based on the type.
508
switch (args->type) {
509
case XFS_ALLOCTYPE_THIS_AG:
510
error = xfs_alloc_ag_vextent_size(args);
512
case XFS_ALLOCTYPE_NEAR_BNO:
513
error = xfs_alloc_ag_vextent_near(args);
515
case XFS_ALLOCTYPE_THIS_BNO:
516
error = xfs_alloc_ag_vextent_exact(args);
523
if (error || args->agbno == NULLAGBLOCK)
526
ASSERT(args->len >= args->minlen);
527
ASSERT(args->len <= args->maxlen);
528
ASSERT(!args->wasfromfl || !args->isfl);
529
ASSERT(args->agbno % args->alignment == 0);
531
if (!args->wasfromfl) {
532
error = xfs_alloc_update_counters(args->tp, args->pag,
534
-((long)(args->len)));
538
ASSERT(!xfs_alloc_busy_search(args->mp, args->agno,
539
args->agbno, args->len));
543
xfs_trans_mod_sb(args->tp, args->wasdel ?
544
XFS_TRANS_SB_RES_FDBLOCKS :
545
XFS_TRANS_SB_FDBLOCKS,
546
-((long)(args->len)));
549
XFS_STATS_INC(xs_allocx);
550
XFS_STATS_ADD(xs_allocb, args->len);
555
* Allocate a variable extent at exactly agno/bno.
556
* Extent's length (returned in *len) will be between minlen and maxlen,
557
* and of the form k * prod + mod unless there's nothing that large.
558
* Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
560
STATIC int /* error */
561
xfs_alloc_ag_vextent_exact(
562
xfs_alloc_arg_t *args) /* allocation argument structure */
564
xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */
565
xfs_btree_cur_t *cnt_cur;/* by count btree cursor */
567
xfs_agblock_t fbno; /* start block of found extent */
568
xfs_extlen_t flen; /* length of found extent */
569
xfs_agblock_t tbno; /* start block of trimmed extent */
570
xfs_extlen_t tlen; /* length of trimmed extent */
571
xfs_agblock_t tend; /* end block of trimmed extent */
572
int i; /* success/failure of operation */
574
ASSERT(args->alignment == 1);
577
* Allocate/initialize a cursor for the by-number freespace btree.
579
bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
580
args->agno, XFS_BTNUM_BNO);
583
* Lookup bno and minlen in the btree (minlen is irrelevant, really).
584
* Look for the closest free block <= bno, it must contain bno
585
* if any free block does.
587
error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i);
594
* Grab the freespace record.
596
error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i);
599
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
600
ASSERT(fbno <= args->agbno);
603
* Check for overlapping busy extents.
605
xfs_alloc_busy_trim(args, fbno, flen, &tbno, &tlen);
608
* Give up if the start of the extent is busy, or the freespace isn't
609
* long enough for the minimum request.
611
if (tbno > args->agbno)
613
if (tlen < args->minlen)
616
if (tend < args->agbno + args->minlen)
620
* End of extent will be smaller of the freespace end and the
621
* maximal requested end.
623
* Fix the length according to mod and prod if given.
625
args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen)
627
xfs_alloc_fix_len(args);
628
if (!xfs_alloc_fix_minleft(args))
631
ASSERT(args->agbno + args->len <= tend);
634
* We are allocating agbno for args->len
635
* Allocate/initialize a cursor for the by-size btree.
637
cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
638
args->agno, XFS_BTNUM_CNT);
639
ASSERT(args->agbno + args->len <=
640
be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
641
error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno,
642
args->len, XFSA_FIXUP_BNO_OK);
644
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
648
xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
649
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
652
trace_xfs_alloc_exact_done(args);
656
/* Didn't find it, return null. */
657
xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
658
args->agbno = NULLAGBLOCK;
659
trace_xfs_alloc_exact_notfound(args);
663
xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
664
trace_xfs_alloc_exact_error(args);
669
* Search the btree in a given direction via the search cursor and compare
670
* the records found against the good extent we've already found.
673
xfs_alloc_find_best_extent(
674
struct xfs_alloc_arg *args, /* allocation argument structure */
675
struct xfs_btree_cur **gcur, /* good cursor */
676
struct xfs_btree_cur **scur, /* searching cursor */
677
xfs_agblock_t gdiff, /* difference for search comparison */
678
xfs_agblock_t *sbno, /* extent found by search */
679
xfs_extlen_t *slen, /* extent length */
680
xfs_agblock_t *sbnoa, /* aligned extent found by search */
681
xfs_extlen_t *slena, /* aligned extent length */
682
int dir) /* 0 = search right, 1 = search left */
689
/* The good extent is perfect, no need to search. */
694
* Look until we find a better one, run out of space or run off the end.
697
error = xfs_alloc_get_rec(*scur, sbno, slen, &i);
700
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
701
xfs_alloc_compute_aligned(args, *sbno, *slen, sbnoa, slena);
704
* The good extent is closer than this one.
707
if (*sbnoa >= args->agbno + gdiff)
710
if (*sbnoa <= args->agbno - gdiff)
715
* Same distance, compare length and pick the best.
717
if (*slena >= args->minlen) {
718
args->len = XFS_EXTLEN_MIN(*slena, args->maxlen);
719
xfs_alloc_fix_len(args);
721
sdiff = xfs_alloc_compute_diff(args->agbno, args->len,
722
args->alignment, *sbnoa,
726
* Choose closer size and invalidate other cursor.
734
error = xfs_btree_increment(*scur, 0, &i);
736
error = xfs_btree_decrement(*scur, 0, &i);
742
xfs_btree_del_cursor(*scur, XFS_BTREE_NOERROR);
747
xfs_btree_del_cursor(*gcur, XFS_BTREE_NOERROR);
752
/* caller invalidates cursors */
757
* Allocate a variable extent near bno in the allocation group agno.
758
* Extent's length (returned in len) will be between minlen and maxlen,
759
* and of the form k * prod + mod unless there's nothing that large.
760
* Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
762
STATIC int /* error */
763
xfs_alloc_ag_vextent_near(
764
xfs_alloc_arg_t *args) /* allocation argument structure */
766
xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */
767
xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */
768
xfs_btree_cur_t *cnt_cur; /* cursor for count btree */
769
xfs_agblock_t gtbno; /* start bno of right side entry */
770
xfs_agblock_t gtbnoa; /* aligned ... */
771
xfs_extlen_t gtdiff; /* difference to right side entry */
772
xfs_extlen_t gtlen; /* length of right side entry */
773
xfs_extlen_t gtlena; /* aligned ... */
774
xfs_agblock_t gtnew; /* useful start bno of right side */
775
int error; /* error code */
776
int i; /* result code, temporary */
777
int j; /* result code, temporary */
778
xfs_agblock_t ltbno; /* start bno of left side entry */
779
xfs_agblock_t ltbnoa; /* aligned ... */
780
xfs_extlen_t ltdiff; /* difference to left side entry */
781
xfs_extlen_t ltlen; /* length of left side entry */
782
xfs_extlen_t ltlena; /* aligned ... */
783
xfs_agblock_t ltnew; /* useful start bno of left side */
784
xfs_extlen_t rlen; /* length of returned extent */
786
#if defined(DEBUG) && defined(__KERNEL__)
788
* Randomly don't execute the first algorithm.
790
int dofirst; /* set to do first algorithm */
792
dofirst = random32() & 1;
803
* Get a cursor for the by-size btree.
805
cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
806
args->agno, XFS_BTNUM_CNT);
809
* See if there are any free extents as big as maxlen.
811
if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i)))
814
* If none, then pick up the last entry in the tree unless the
818
if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno,
821
if (i == 0 || ltlen == 0) {
822
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
823
trace_xfs_alloc_near_noentry(args);
832
* If the requested extent is large wrt the freespaces available
833
* in this a.g., then the cursor will be pointing to a btree entry
834
* near the right edge of the tree. If it's in the last btree leaf
835
* block, then we just examine all the entries in that block
836
* that are big enough, and pick the best one.
837
* This is written as a while loop so we can break out of it,
838
* but we never loop back to the top.
840
while (xfs_btree_islastblock(cnt_cur, 0)) {
844
xfs_agblock_t bnew=0;
846
#if defined(DEBUG) && defined(__KERNEL__)
851
* Start from the entry that lookup found, sequence through
852
* all larger free blocks. If we're actually pointing at a
853
* record smaller than maxlen, go to the start of this block,
854
* and skip all those smaller than minlen.
856
if (ltlen || args->alignment > 1) {
857
cnt_cur->bc_ptrs[0] = 1;
859
if ((error = xfs_alloc_get_rec(cnt_cur, <bno,
862
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
863
if (ltlen >= args->minlen)
865
if ((error = xfs_btree_increment(cnt_cur, 0, &i)))
868
ASSERT(ltlen >= args->minlen);
872
i = cnt_cur->bc_ptrs[0];
873
for (j = 1, blen = 0, bdiff = 0;
874
!error && j && (blen < args->maxlen || bdiff > 0);
875
error = xfs_btree_increment(cnt_cur, 0, &j)) {
877
* For each entry, decide if it's better than
878
* the previous best entry.
880
if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
882
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
883
xfs_alloc_compute_aligned(args, ltbno, ltlen,
885
if (ltlena < args->minlen)
887
args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
888
xfs_alloc_fix_len(args);
889
ASSERT(args->len >= args->minlen);
890
if (args->len < blen)
892
ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
893
args->alignment, ltbnoa, ltlena, <new);
894
if (ltnew != NULLAGBLOCK &&
895
(args->len > blen || ltdiff < bdiff)) {
899
besti = cnt_cur->bc_ptrs[0];
903
* It didn't work. We COULD be in a case where
904
* there's a good record somewhere, so try again.
909
* Point at the best entry, and retrieve it again.
911
cnt_cur->bc_ptrs[0] = besti;
912
if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
914
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
915
ASSERT(ltbno + ltlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
917
if (!xfs_alloc_fix_minleft(args)) {
918
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
919
trace_xfs_alloc_near_nominleft(args);
924
* We are allocating starting at bnew for blen blocks.
927
ASSERT(bnew >= ltbno);
928
ASSERT(bnew + blen <= ltbno + ltlen);
930
* Set up a cursor for the by-bno tree.
932
bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp,
933
args->agbp, args->agno, XFS_BTNUM_BNO);
935
* Fix up the btree entries.
937
if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno,
938
ltlen, bnew, blen, XFSA_FIXUP_CNT_OK)))
940
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
941
xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
943
trace_xfs_alloc_near_first(args);
948
* Search in the by-bno tree to the left and to the right
949
* simultaneously, until in each case we find a space big enough,
950
* or run into the edge of the tree. When we run into the edge,
951
* we deallocate that cursor.
952
* If both searches succeed, we compare the two spaces and pick
954
* With alignment, it's possible for both to fail; the upper
955
* level algorithm that picks allocation groups for allocations
956
* is not supposed to do this.
959
* Allocate and initialize the cursor for the leftward search.
961
bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
962
args->agno, XFS_BTNUM_BNO);
964
* Lookup <= bno to find the leftward search's starting point.
966
if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i)))
970
* Didn't find anything; use this cursor for the rightward
973
bno_cur_gt = bno_cur_lt;
977
* Found something. Duplicate the cursor for the rightward search.
979
else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt)))
982
* Increment the cursor, so we will point at the entry just right
983
* of the leftward entry if any, or to the leftmost entry.
985
if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
989
* It failed, there are no rightward entries.
991
xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR);
995
* Loop going left with the leftward cursor, right with the
996
* rightward cursor, until either both directions give up or
997
* we find an entry at least as big as minlen.
1001
if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i)))
1003
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1004
xfs_alloc_compute_aligned(args, ltbno, ltlen,
1006
if (ltlena >= args->minlen)
1008
if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i)))
1011
xfs_btree_del_cursor(bno_cur_lt,
1017
if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i)))
1019
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1020
xfs_alloc_compute_aligned(args, gtbno, gtlen,
1022
if (gtlena >= args->minlen)
1024
if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
1027
xfs_btree_del_cursor(bno_cur_gt,
1032
} while (bno_cur_lt || bno_cur_gt);
1035
* Got both cursors still active, need to find better entry.
1037
if (bno_cur_lt && bno_cur_gt) {
1038
if (ltlena >= args->minlen) {
1040
* Left side is good, look for a right side entry.
1042
args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1043
xfs_alloc_fix_len(args);
1044
ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1045
args->alignment, ltbnoa, ltlena, <new);
1047
error = xfs_alloc_find_best_extent(args,
1048
&bno_cur_lt, &bno_cur_gt,
1049
ltdiff, >bno, >len,
1051
0 /* search right */);
1053
ASSERT(gtlena >= args->minlen);
1056
* Right side is good, look for a left side entry.
1058
args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen);
1059
xfs_alloc_fix_len(args);
1060
gtdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1061
args->alignment, gtbnoa, gtlena, >new);
1063
error = xfs_alloc_find_best_extent(args,
1064
&bno_cur_gt, &bno_cur_lt,
1065
gtdiff, <bno, <len,
1067
1 /* search left */);
1075
* If we couldn't get anything, give up.
1077
if (bno_cur_lt == NULL && bno_cur_gt == NULL) {
1079
trace_xfs_alloc_near_busy(args);
1080
xfs_log_force(args->mp, XFS_LOG_SYNC);
1084
trace_xfs_alloc_size_neither(args);
1085
args->agbno = NULLAGBLOCK;
1090
* At this point we have selected a freespace entry, either to the
1091
* left or to the right. If it's on the right, copy all the
1092
* useful variables to the "left" set so we only have one
1093
* copy of this code.
1096
bno_cur_lt = bno_cur_gt;
1107
* Fix up the length and compute the useful address.
1109
args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1110
xfs_alloc_fix_len(args);
1111
if (!xfs_alloc_fix_minleft(args)) {
1112
trace_xfs_alloc_near_nominleft(args);
1113
xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1114
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1118
(void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment,
1119
ltbnoa, ltlena, <new);
1120
ASSERT(ltnew >= ltbno);
1121
ASSERT(ltnew + rlen <= ltbnoa + ltlena);
1122
ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
1123
args->agbno = ltnew;
1125
if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen,
1126
ltnew, rlen, XFSA_FIXUP_BNO_OK)))
1130
trace_xfs_alloc_near_greater(args);
1132
trace_xfs_alloc_near_lesser(args);
1134
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1135
xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1139
trace_xfs_alloc_near_error(args);
1140
if (cnt_cur != NULL)
1141
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1142
if (bno_cur_lt != NULL)
1143
xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR);
1144
if (bno_cur_gt != NULL)
1145
xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR);
1150
* Allocate a variable extent anywhere in the allocation group agno.
1151
* Extent's length (returned in len) will be between minlen and maxlen,
1152
* and of the form k * prod + mod unless there's nothing that large.
1153
* Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1155
STATIC int /* error */
1156
xfs_alloc_ag_vextent_size(
1157
xfs_alloc_arg_t *args) /* allocation argument structure */
1159
xfs_btree_cur_t *bno_cur; /* cursor for bno btree */
1160
xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */
1161
int error; /* error result */
1162
xfs_agblock_t fbno; /* start of found freespace */
1163
xfs_extlen_t flen; /* length of found freespace */
1164
int i; /* temp status variable */
1165
xfs_agblock_t rbno; /* returned block number */
1166
xfs_extlen_t rlen; /* length of returned extent */
1171
* Allocate and initialize a cursor for the by-size btree.
1173
cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1174
args->agno, XFS_BTNUM_CNT);
1178
* Look for an entry >= maxlen+alignment-1 blocks.
1180
if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1181
args->maxlen + args->alignment - 1, &i)))
1185
* If none or we have busy extents that we cannot allocate from, then
1186
* we have to settle for a smaller extent. In the case that there are
1187
* no large extents, this will return the last entry in the tree unless
1188
* the tree is empty. In the case that there are only busy large
1189
* extents, this will return the largest small extent unless there
1190
* are no smaller extents available.
1192
if (!i || forced > 1) {
1193
error = xfs_alloc_ag_vextent_small(args, cnt_cur,
1197
if (i == 0 || flen == 0) {
1198
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1199
trace_xfs_alloc_size_noentry(args);
1203
xfs_alloc_compute_aligned(args, fbno, flen, &rbno, &rlen);
1206
* Search for a non-busy extent that is large enough.
1207
* If we are at low space, don't check, or if we fall of
1208
* the end of the btree, turn off the busy check and
1212
error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i);
1215
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1217
xfs_alloc_compute_aligned(args, fbno, flen,
1220
if (rlen >= args->maxlen)
1223
error = xfs_btree_increment(cnt_cur, 0, &i);
1228
* Our only valid extents must have been busy.
1229
* Make it unbusy by forcing the log out and
1230
* retrying. If we've been here before, forcing
1231
* the log isn't making the extents available,
1232
* which means they have probably been freed in
1233
* this transaction. In that case, we have to
1234
* give up on them and we'll attempt a minlen
1235
* allocation the next time around.
1237
xfs_btree_del_cursor(cnt_cur,
1239
trace_xfs_alloc_size_busy(args);
1241
xfs_log_force(args->mp, XFS_LOG_SYNC);
1248
* In the first case above, we got the last entry in the
1249
* by-size btree. Now we check to see if the space hits maxlen
1250
* once aligned; if not, we search left for something better.
1251
* This can't happen in the second case above.
1253
rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1254
XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1255
(rlen <= flen && rbno + rlen <= fbno + flen), error0);
1256
if (rlen < args->maxlen) {
1257
xfs_agblock_t bestfbno;
1258
xfs_extlen_t bestflen;
1259
xfs_agblock_t bestrbno;
1260
xfs_extlen_t bestrlen;
1267
if ((error = xfs_btree_decrement(cnt_cur, 0, &i)))
1271
if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1274
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1275
if (flen < bestrlen)
1277
xfs_alloc_compute_aligned(args, fbno, flen,
1279
rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1280
XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1281
(rlen <= flen && rbno + rlen <= fbno + flen),
1283
if (rlen > bestrlen) {
1288
if (rlen == args->maxlen)
1292
if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1295
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1301
args->wasfromfl = 0;
1303
* Fix up the length.
1306
if (rlen < args->minlen) {
1308
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1309
trace_xfs_alloc_size_busy(args);
1310
xfs_log_force(args->mp, XFS_LOG_SYNC);
1315
xfs_alloc_fix_len(args);
1317
if (!xfs_alloc_fix_minleft(args))
1320
XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0);
1322
* Allocate and initialize a cursor for the by-block tree.
1324
bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1325
args->agno, XFS_BTNUM_BNO);
1326
if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1327
rbno, rlen, XFSA_FIXUP_CNT_OK)))
1329
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1330
xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1331
cnt_cur = bno_cur = NULL;
1334
XFS_WANT_CORRUPTED_GOTO(
1335
args->agbno + args->len <=
1336
be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1338
trace_xfs_alloc_size_done(args);
1342
trace_xfs_alloc_size_error(args);
1344
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1346
xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1350
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1351
trace_xfs_alloc_size_nominleft(args);
1352
args->agbno = NULLAGBLOCK;
1357
* Deal with the case where only small freespaces remain.
1358
* Either return the contents of the last freespace record,
1359
* or allocate space from the freelist if there is nothing in the tree.
1361
STATIC int /* error */
1362
xfs_alloc_ag_vextent_small(
1363
xfs_alloc_arg_t *args, /* allocation argument structure */
1364
xfs_btree_cur_t *ccur, /* by-size cursor */
1365
xfs_agblock_t *fbnop, /* result block number */
1366
xfs_extlen_t *flenp, /* result length */
1367
int *stat) /* status: 0-freelist, 1-normal/none */
1374
if ((error = xfs_btree_decrement(ccur, 0, &i)))
1377
if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i)))
1379
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1382
* Nothing in the btree, try the freelist. Make sure
1383
* to respect minleft even when pulling from the
1386
else if (args->minlen == 1 && args->alignment == 1 && !args->isfl &&
1387
(be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount)
1389
error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0);
1392
if (fbno != NULLAGBLOCK) {
1393
xfs_alloc_busy_reuse(args->mp, args->agno, fbno, 1,
1396
if (args->userdata) {
1399
bp = xfs_btree_get_bufs(args->mp, args->tp,
1400
args->agno, fbno, 0);
1401
xfs_trans_binval(args->tp, bp);
1405
XFS_WANT_CORRUPTED_GOTO(
1406
args->agbno + args->len <=
1407
be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1409
args->wasfromfl = 1;
1410
trace_xfs_alloc_small_freelist(args);
1415
* Nothing in the freelist.
1421
* Can't allocate from the freelist for some reason.
1428
* Can't do the allocation, give up.
1430
if (flen < args->minlen) {
1431
args->agbno = NULLAGBLOCK;
1432
trace_xfs_alloc_small_notenough(args);
1438
trace_xfs_alloc_small_done(args);
1442
trace_xfs_alloc_small_error(args);
1447
* Free the extent starting at agno/bno for length.
1449
STATIC int /* error */
1451
xfs_trans_t *tp, /* transaction pointer */
1452
xfs_buf_t *agbp, /* buffer for a.g. freelist header */
1453
xfs_agnumber_t agno, /* allocation group number */
1454
xfs_agblock_t bno, /* starting block number */
1455
xfs_extlen_t len, /* length of extent */
1456
int isfl) /* set if is freelist blocks - no sb acctg */
1458
xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */
1459
xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */
1460
int error; /* error return value */
1461
xfs_agblock_t gtbno; /* start of right neighbor block */
1462
xfs_extlen_t gtlen; /* length of right neighbor block */
1463
int haveleft; /* have a left neighbor block */
1464
int haveright; /* have a right neighbor block */
1465
int i; /* temp, result code */
1466
xfs_agblock_t ltbno; /* start of left neighbor block */
1467
xfs_extlen_t ltlen; /* length of left neighbor block */
1468
xfs_mount_t *mp; /* mount point struct for filesystem */
1469
xfs_agblock_t nbno; /* new starting block of freespace */
1470
xfs_extlen_t nlen; /* new length of freespace */
1471
xfs_perag_t *pag; /* per allocation group data */
1475
* Allocate and initialize a cursor for the by-block btree.
1477
bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO);
1480
* Look for a neighboring block on the left (lower block numbers)
1481
* that is contiguous with this space.
1483
if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1487
* There is a block to our left.
1489
if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i)))
1491
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1493
* It's not contiguous, though.
1495
if (ltbno + ltlen < bno)
1499
* If this failure happens the request to free this
1500
* space was invalid, it's (partly) already free.
1503
XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0);
1507
* Look for a neighboring block on the right (higher block numbers)
1508
* that is contiguous with this space.
1510
if ((error = xfs_btree_increment(bno_cur, 0, &haveright)))
1514
* There is a block to our right.
1516
if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i)))
1518
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1520
* It's not contiguous, though.
1522
if (bno + len < gtbno)
1526
* If this failure happens the request to free this
1527
* space was invalid, it's (partly) already free.
1530
XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0);
1534
* Now allocate and initialize a cursor for the by-size tree.
1536
cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT);
1538
* Have both left and right contiguous neighbors.
1539
* Merge all three into a single free block.
1541
if (haveleft && haveright) {
1543
* Delete the old by-size entry on the left.
1545
if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1547
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1548
if ((error = xfs_btree_delete(cnt_cur, &i)))
1550
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1552
* Delete the old by-size entry on the right.
1554
if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1556
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1557
if ((error = xfs_btree_delete(cnt_cur, &i)))
1559
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1561
* Delete the old by-block entry for the right block.
1563
if ((error = xfs_btree_delete(bno_cur, &i)))
1565
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1567
* Move the by-block cursor back to the left neighbor.
1569
if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
1571
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1574
* Check that this is the right record: delete didn't
1575
* mangle the cursor.
1578
xfs_agblock_t xxbno;
1581
if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
1584
XFS_WANT_CORRUPTED_GOTO(
1585
i == 1 && xxbno == ltbno && xxlen == ltlen,
1590
* Update remaining by-block entry to the new, joined block.
1593
nlen = len + ltlen + gtlen;
1594
if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1598
* Have only a left contiguous neighbor.
1599
* Merge it together with the new freespace.
1601
else if (haveleft) {
1603
* Delete the old by-size entry on the left.
1605
if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1607
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1608
if ((error = xfs_btree_delete(cnt_cur, &i)))
1610
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1612
* Back up the by-block cursor to the left neighbor, and
1613
* update its length.
1615
if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
1617
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1620
if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1624
* Have only a right contiguous neighbor.
1625
* Merge it together with the new freespace.
1627
else if (haveright) {
1629
* Delete the old by-size entry on the right.
1631
if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1633
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1634
if ((error = xfs_btree_delete(cnt_cur, &i)))
1636
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1638
* Update the starting block and length of the right
1639
* neighbor in the by-block tree.
1643
if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1647
* No contiguous neighbors.
1648
* Insert the new freespace into the by-block tree.
1653
if ((error = xfs_btree_insert(bno_cur, &i)))
1655
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1657
xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1660
* In all cases we need to insert the new freespace in the by-size tree.
1662
if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
1664
XFS_WANT_CORRUPTED_GOTO(i == 0, error0);
1665
if ((error = xfs_btree_insert(cnt_cur, &i)))
1667
XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1668
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1672
* Update the freespace totals in the ag and superblock.
1674
pag = xfs_perag_get(mp, agno);
1675
error = xfs_alloc_update_counters(tp, pag, agbp, len);
1681
xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len);
1682
XFS_STATS_INC(xs_freex);
1683
XFS_STATS_ADD(xs_freeb, len);
1685
trace_xfs_free_extent(mp, agno, bno, len, isfl, haveleft, haveright);
1690
trace_xfs_free_extent(mp, agno, bno, len, isfl, -1, -1);
1692
xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1694
xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1699
* Visible (exported) allocation/free functions.
1700
* Some of these are used just by xfs_alloc_btree.c and this file.
1704
* Compute and fill in value of m_ag_maxlevels.
1707
xfs_alloc_compute_maxlevels(
1708
xfs_mount_t *mp) /* file system mount structure */
1716
maxleafents = (mp->m_sb.sb_agblocks + 1) / 2;
1717
minleafrecs = mp->m_alloc_mnr[0];
1718
minnoderecs = mp->m_alloc_mnr[1];
1719
maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1720
for (level = 1; maxblocks > 1; level++)
1721
maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1722
mp->m_ag_maxlevels = level;
1726
* Find the length of the longest extent in an AG.
1729
xfs_alloc_longest_free_extent(
1730
struct xfs_mount *mp,
1731
struct xfs_perag *pag)
1733
xfs_extlen_t need, delta = 0;
1735
need = XFS_MIN_FREELIST_PAG(pag, mp);
1736
if (need > pag->pagf_flcount)
1737
delta = need - pag->pagf_flcount;
1739
if (pag->pagf_longest > delta)
1740
return pag->pagf_longest - delta;
1741
return pag->pagf_flcount > 0 || pag->pagf_longest > 0;
1745
* Decide whether to use this allocation group for this allocation.
1746
* If so, fix up the btree freelist's size.
1748
STATIC int /* error */
1749
xfs_alloc_fix_freelist(
1750
xfs_alloc_arg_t *args, /* allocation argument structure */
1751
int flags) /* XFS_ALLOC_FLAG_... */
1753
xfs_buf_t *agbp; /* agf buffer pointer */
1754
xfs_agf_t *agf; /* a.g. freespace structure pointer */
1755
xfs_buf_t *agflbp;/* agfl buffer pointer */
1756
xfs_agblock_t bno; /* freelist block */
1757
xfs_extlen_t delta; /* new blocks needed in freelist */
1758
int error; /* error result code */
1759
xfs_extlen_t longest;/* longest extent in allocation group */
1760
xfs_mount_t *mp; /* file system mount point structure */
1761
xfs_extlen_t need; /* total blocks needed in freelist */
1762
xfs_perag_t *pag; /* per-ag information structure */
1763
xfs_alloc_arg_t targs; /* local allocation arguments */
1764
xfs_trans_t *tp; /* transaction pointer */
1770
if (!pag->pagf_init) {
1771
if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1774
if (!pag->pagf_init) {
1775
ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1776
ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1784
* If this is a metadata preferred pag and we are user data
1785
* then try somewhere else if we are not being asked to
1786
* try harder at this point
1788
if (pag->pagf_metadata && args->userdata &&
1789
(flags & XFS_ALLOC_FLAG_TRYLOCK)) {
1790
ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1795
if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1797
* If it looks like there isn't a long enough extent, or enough
1798
* total blocks, reject it.
1800
need = XFS_MIN_FREELIST_PAG(pag, mp);
1801
longest = xfs_alloc_longest_free_extent(mp, pag);
1802
if ((args->minlen + args->alignment + args->minalignslop - 1) >
1804
((int)(pag->pagf_freeblks + pag->pagf_flcount -
1805
need - args->total) < (int)args->minleft)) {
1807
xfs_trans_brelse(tp, agbp);
1814
* Get the a.g. freespace buffer.
1815
* Can fail if we're not blocking on locks, and it's held.
1818
if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1822
ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1823
ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1829
* Figure out how many blocks we should have in the freelist.
1831
agf = XFS_BUF_TO_AGF(agbp);
1832
need = XFS_MIN_FREELIST(agf, mp);
1834
* If there isn't enough total or single-extent, reject it.
1836
if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1837
delta = need > be32_to_cpu(agf->agf_flcount) ?
1838
(need - be32_to_cpu(agf->agf_flcount)) : 0;
1839
longest = be32_to_cpu(agf->agf_longest);
1840
longest = (longest > delta) ? (longest - delta) :
1841
(be32_to_cpu(agf->agf_flcount) > 0 || longest > 0);
1842
if ((args->minlen + args->alignment + args->minalignslop - 1) >
1844
((int)(be32_to_cpu(agf->agf_freeblks) +
1845
be32_to_cpu(agf->agf_flcount) - need - args->total) <
1846
(int)args->minleft)) {
1847
xfs_trans_brelse(tp, agbp);
1853
* Make the freelist shorter if it's too long.
1855
while (be32_to_cpu(agf->agf_flcount) > need) {
1858
error = xfs_alloc_get_freelist(tp, agbp, &bno, 0);
1861
if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1)))
1863
bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0);
1864
xfs_trans_binval(tp, bp);
1867
* Initialize the args structure.
1872
targs.agno = args->agno;
1873
targs.mod = targs.minleft = targs.wasdel = targs.userdata =
1874
targs.minalignslop = 0;
1875
targs.alignment = targs.minlen = targs.prod = targs.isfl = 1;
1876
targs.type = XFS_ALLOCTYPE_THIS_AG;
1878
if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp)))
1881
* Make the freelist longer if it's too short.
1883
while (be32_to_cpu(agf->agf_flcount) < need) {
1885
targs.maxlen = need - be32_to_cpu(agf->agf_flcount);
1887
* Allocate as many blocks as possible at once.
1889
if ((error = xfs_alloc_ag_vextent(&targs))) {
1890
xfs_trans_brelse(tp, agflbp);
1894
* Stop if we run out. Won't happen if callers are obeying
1895
* the restrictions correctly. Can happen for free calls
1896
* on a completely full ag.
1898
if (targs.agbno == NULLAGBLOCK) {
1899
if (flags & XFS_ALLOC_FLAG_FREEING)
1901
xfs_trans_brelse(tp, agflbp);
1906
* Put each allocated block on the list.
1908
for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
1909
error = xfs_alloc_put_freelist(tp, agbp,
1915
xfs_trans_brelse(tp, agflbp);
1921
* Get a block from the freelist.
1922
* Returns with the buffer for the block gotten.
1925
xfs_alloc_get_freelist(
1926
xfs_trans_t *tp, /* transaction pointer */
1927
xfs_buf_t *agbp, /* buffer containing the agf structure */
1928
xfs_agblock_t *bnop, /* block address retrieved from freelist */
1929
int btreeblk) /* destination is a AGF btree */
1931
xfs_agf_t *agf; /* a.g. freespace structure */
1932
xfs_agfl_t *agfl; /* a.g. freelist structure */
1933
xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */
1934
xfs_agblock_t bno; /* block number returned */
1937
xfs_mount_t *mp; /* mount structure */
1938
xfs_perag_t *pag; /* per allocation group data */
1940
agf = XFS_BUF_TO_AGF(agbp);
1942
* Freelist is empty, give up.
1944
if (!agf->agf_flcount) {
1945
*bnop = NULLAGBLOCK;
1949
* Read the array of free blocks.
1952
if ((error = xfs_alloc_read_agfl(mp, tp,
1953
be32_to_cpu(agf->agf_seqno), &agflbp)))
1955
agfl = XFS_BUF_TO_AGFL(agflbp);
1957
* Get the block number and update the data structures.
1959
bno = be32_to_cpu(agfl->agfl_bno[be32_to_cpu(agf->agf_flfirst)]);
1960
be32_add_cpu(&agf->agf_flfirst, 1);
1961
xfs_trans_brelse(tp, agflbp);
1962
if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp))
1963
agf->agf_flfirst = 0;
1965
pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
1966
be32_add_cpu(&agf->agf_flcount, -1);
1967
xfs_trans_agflist_delta(tp, -1);
1968
pag->pagf_flcount--;
1971
logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT;
1973
be32_add_cpu(&agf->agf_btreeblks, 1);
1974
pag->pagf_btreeblks++;
1975
logflags |= XFS_AGF_BTREEBLKS;
1978
xfs_alloc_log_agf(tp, agbp, logflags);
1985
* Log the given fields from the agf structure.
1989
xfs_trans_t *tp, /* transaction pointer */
1990
xfs_buf_t *bp, /* buffer for a.g. freelist header */
1991
int fields) /* mask of fields to be logged (XFS_AGF_...) */
1993
int first; /* first byte offset */
1994
int last; /* last byte offset */
1995
static const short offsets[] = {
1996
offsetof(xfs_agf_t, agf_magicnum),
1997
offsetof(xfs_agf_t, agf_versionnum),
1998
offsetof(xfs_agf_t, agf_seqno),
1999
offsetof(xfs_agf_t, agf_length),
2000
offsetof(xfs_agf_t, agf_roots[0]),
2001
offsetof(xfs_agf_t, agf_levels[0]),
2002
offsetof(xfs_agf_t, agf_flfirst),
2003
offsetof(xfs_agf_t, agf_fllast),
2004
offsetof(xfs_agf_t, agf_flcount),
2005
offsetof(xfs_agf_t, agf_freeblks),
2006
offsetof(xfs_agf_t, agf_longest),
2007
offsetof(xfs_agf_t, agf_btreeblks),
2011
trace_xfs_agf(tp->t_mountp, XFS_BUF_TO_AGF(bp), fields, _RET_IP_);
2013
xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2014
xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2018
* Interface for inode allocation to force the pag data to be initialized.
2021
xfs_alloc_pagf_init(
2022
xfs_mount_t *mp, /* file system mount structure */
2023
xfs_trans_t *tp, /* transaction pointer */
2024
xfs_agnumber_t agno, /* allocation group number */
2025
int flags) /* XFS_ALLOC_FLAGS_... */
2030
if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp)))
2033
xfs_trans_brelse(tp, bp);
2038
* Put the block on the freelist for the allocation group.
2041
xfs_alloc_put_freelist(
2042
xfs_trans_t *tp, /* transaction pointer */
2043
xfs_buf_t *agbp, /* buffer for a.g. freelist header */
2044
xfs_buf_t *agflbp,/* buffer for a.g. free block array */
2045
xfs_agblock_t bno, /* block being freed */
2046
int btreeblk) /* block came from a AGF btree */
2048
xfs_agf_t *agf; /* a.g. freespace structure */
2049
xfs_agfl_t *agfl; /* a.g. free block array */
2050
__be32 *blockp;/* pointer to array entry */
2053
xfs_mount_t *mp; /* mount structure */
2054
xfs_perag_t *pag; /* per allocation group data */
2056
agf = XFS_BUF_TO_AGF(agbp);
2059
if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp,
2060
be32_to_cpu(agf->agf_seqno), &agflbp)))
2062
agfl = XFS_BUF_TO_AGFL(agflbp);
2063
be32_add_cpu(&agf->agf_fllast, 1);
2064
if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp))
2065
agf->agf_fllast = 0;
2067
pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
2068
be32_add_cpu(&agf->agf_flcount, 1);
2069
xfs_trans_agflist_delta(tp, 1);
2070
pag->pagf_flcount++;
2072
logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT;
2074
be32_add_cpu(&agf->agf_btreeblks, -1);
2075
pag->pagf_btreeblks--;
2076
logflags |= XFS_AGF_BTREEBLKS;
2080
xfs_alloc_log_agf(tp, agbp, logflags);
2082
ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp));
2083
blockp = &agfl->agfl_bno[be32_to_cpu(agf->agf_fllast)];
2084
*blockp = cpu_to_be32(bno);
2085
xfs_alloc_log_agf(tp, agbp, logflags);
2086
xfs_trans_log_buf(tp, agflbp,
2087
(int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl),
2088
(int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl +
2089
sizeof(xfs_agblock_t) - 1));
2094
* Read in the allocation group header (free/alloc section).
2098
struct xfs_mount *mp, /* mount point structure */
2099
struct xfs_trans *tp, /* transaction pointer */
2100
xfs_agnumber_t agno, /* allocation group number */
2101
int flags, /* XFS_BUF_ */
2102
struct xfs_buf **bpp) /* buffer for the ag freelist header */
2104
struct xfs_agf *agf; /* ag freelist header */
2105
int agf_ok; /* set if agf is consistent */
2108
ASSERT(agno != NULLAGNUMBER);
2109
error = xfs_trans_read_buf(
2110
mp, tp, mp->m_ddev_targp,
2111
XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
2112
XFS_FSS_TO_BB(mp, 1), flags, bpp);
2118
ASSERT(!(*bpp)->b_error);
2119
agf = XFS_BUF_TO_AGF(*bpp);
2122
* Validate the magic number of the agf block.
2125
agf->agf_magicnum == cpu_to_be32(XFS_AGF_MAGIC) &&
2126
XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) &&
2127
be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) &&
2128
be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) &&
2129
be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) &&
2130
be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp) &&
2131
be32_to_cpu(agf->agf_seqno) == agno;
2132
if (xfs_sb_version_haslazysbcount(&mp->m_sb))
2133
agf_ok = agf_ok && be32_to_cpu(agf->agf_btreeblks) <=
2134
be32_to_cpu(agf->agf_length);
2135
if (unlikely(XFS_TEST_ERROR(!agf_ok, mp, XFS_ERRTAG_ALLOC_READ_AGF,
2136
XFS_RANDOM_ALLOC_READ_AGF))) {
2137
XFS_CORRUPTION_ERROR("xfs_alloc_read_agf",
2138
XFS_ERRLEVEL_LOW, mp, agf);
2139
xfs_trans_brelse(tp, *bpp);
2140
return XFS_ERROR(EFSCORRUPTED);
2142
xfs_buf_set_ref(*bpp, XFS_AGF_REF);
2147
* Read in the allocation group header (free/alloc section).
2151
struct xfs_mount *mp, /* mount point structure */
2152
struct xfs_trans *tp, /* transaction pointer */
2153
xfs_agnumber_t agno, /* allocation group number */
2154
int flags, /* XFS_ALLOC_FLAG_... */
2155
struct xfs_buf **bpp) /* buffer for the ag freelist header */
2157
struct xfs_agf *agf; /* ag freelist header */
2158
struct xfs_perag *pag; /* per allocation group data */
2161
ASSERT(agno != NULLAGNUMBER);
2163
error = xfs_read_agf(mp, tp, agno,
2164
(flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0,
2170
ASSERT(!(*bpp)->b_error);
2172
agf = XFS_BUF_TO_AGF(*bpp);
2173
pag = xfs_perag_get(mp, agno);
2174
if (!pag->pagf_init) {
2175
pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
2176
pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks);
2177
pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
2178
pag->pagf_longest = be32_to_cpu(agf->agf_longest);
2179
pag->pagf_levels[XFS_BTNUM_BNOi] =
2180
be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
2181
pag->pagf_levels[XFS_BTNUM_CNTi] =
2182
be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
2183
spin_lock_init(&pag->pagb_lock);
2184
pag->pagb_count = 0;
2185
pag->pagb_tree = RB_ROOT;
2189
else if (!XFS_FORCED_SHUTDOWN(mp)) {
2190
ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
2191
ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks));
2192
ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
2193
ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
2194
ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
2195
be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
2196
ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
2197
be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
2205
* Allocate an extent (variable-size).
2206
* Depending on the allocation type, we either look in a single allocation
2207
* group or loop over the allocation groups to find the result.
2211
xfs_alloc_arg_t *args) /* allocation argument structure */
2213
xfs_agblock_t agsize; /* allocation group size */
2215
int flags; /* XFS_ALLOC_FLAG_... locking flags */
2216
xfs_extlen_t minleft;/* minimum left value, temp copy */
2217
xfs_mount_t *mp; /* mount structure pointer */
2218
xfs_agnumber_t sagno; /* starting allocation group number */
2219
xfs_alloctype_t type; /* input allocation type */
2222
xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */
2225
type = args->otype = args->type;
2226
args->agbno = NULLAGBLOCK;
2228
* Just fix this up, for the case where the last a.g. is shorter
2229
* (or there's only one a.g.) and the caller couldn't easily figure
2230
* that out (xfs_bmap_alloc).
2232
agsize = mp->m_sb.sb_agblocks;
2233
if (args->maxlen > agsize)
2234
args->maxlen = agsize;
2235
if (args->alignment == 0)
2236
args->alignment = 1;
2237
ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount);
2238
ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize);
2239
ASSERT(args->minlen <= args->maxlen);
2240
ASSERT(args->minlen <= agsize);
2241
ASSERT(args->mod < args->prod);
2242
if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount ||
2243
XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize ||
2244
args->minlen > args->maxlen || args->minlen > agsize ||
2245
args->mod >= args->prod) {
2246
args->fsbno = NULLFSBLOCK;
2247
trace_xfs_alloc_vextent_badargs(args);
2250
minleft = args->minleft;
2253
case XFS_ALLOCTYPE_THIS_AG:
2254
case XFS_ALLOCTYPE_NEAR_BNO:
2255
case XFS_ALLOCTYPE_THIS_BNO:
2257
* These three force us into a single a.g.
2259
args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2260
args->pag = xfs_perag_get(mp, args->agno);
2262
error = xfs_alloc_fix_freelist(args, 0);
2263
args->minleft = minleft;
2265
trace_xfs_alloc_vextent_nofix(args);
2269
trace_xfs_alloc_vextent_noagbp(args);
2272
args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2273
if ((error = xfs_alloc_ag_vextent(args)))
2276
case XFS_ALLOCTYPE_START_BNO:
2278
* Try near allocation first, then anywhere-in-ag after
2279
* the first a.g. fails.
2281
if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) &&
2282
(mp->m_flags & XFS_MOUNT_32BITINODES)) {
2283
args->fsbno = XFS_AGB_TO_FSB(mp,
2284
((mp->m_agfrotor / rotorstep) %
2285
mp->m_sb.sb_agcount), 0);
2288
args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2289
args->type = XFS_ALLOCTYPE_NEAR_BNO;
2291
case XFS_ALLOCTYPE_ANY_AG:
2292
case XFS_ALLOCTYPE_START_AG:
2293
case XFS_ALLOCTYPE_FIRST_AG:
2295
* Rotate through the allocation groups looking for a winner.
2297
if (type == XFS_ALLOCTYPE_ANY_AG) {
2299
* Start with the last place we left off.
2301
args->agno = sagno = (mp->m_agfrotor / rotorstep) %
2302
mp->m_sb.sb_agcount;
2303
args->type = XFS_ALLOCTYPE_THIS_AG;
2304
flags = XFS_ALLOC_FLAG_TRYLOCK;
2305
} else if (type == XFS_ALLOCTYPE_FIRST_AG) {
2307
* Start with allocation group given by bno.
2309
args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2310
args->type = XFS_ALLOCTYPE_THIS_AG;
2314
if (type == XFS_ALLOCTYPE_START_AG)
2315
args->type = XFS_ALLOCTYPE_THIS_AG;
2317
* Start with the given allocation group.
2319
args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2320
flags = XFS_ALLOC_FLAG_TRYLOCK;
2323
* Loop over allocation groups twice; first time with
2324
* trylock set, second time without.
2327
args->pag = xfs_perag_get(mp, args->agno);
2328
if (no_min) args->minleft = 0;
2329
error = xfs_alloc_fix_freelist(args, flags);
2330
args->minleft = minleft;
2332
trace_xfs_alloc_vextent_nofix(args);
2336
* If we get a buffer back then the allocation will fly.
2339
if ((error = xfs_alloc_ag_vextent(args)))
2344
trace_xfs_alloc_vextent_loopfailed(args);
2347
* Didn't work, figure out the next iteration.
2349
if (args->agno == sagno &&
2350
type == XFS_ALLOCTYPE_START_BNO)
2351
args->type = XFS_ALLOCTYPE_THIS_AG;
2353
* For the first allocation, we can try any AG to get
2354
* space. However, if we already have allocated a
2355
* block, we don't want to try AGs whose number is below
2356
* sagno. Otherwise, we may end up with out-of-order
2357
* locking of AGF, which might cause deadlock.
2359
if (++(args->agno) == mp->m_sb.sb_agcount) {
2360
if (args->firstblock != NULLFSBLOCK)
2366
* Reached the starting a.g., must either be done
2367
* or switch to non-trylock mode.
2369
if (args->agno == sagno) {
2371
args->agbno = NULLAGBLOCK;
2372
trace_xfs_alloc_vextent_allfailed(args);
2379
if (type == XFS_ALLOCTYPE_START_BNO) {
2380
args->agbno = XFS_FSB_TO_AGBNO(mp,
2382
args->type = XFS_ALLOCTYPE_NEAR_BNO;
2386
xfs_perag_put(args->pag);
2388
if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) {
2389
if (args->agno == sagno)
2390
mp->m_agfrotor = (mp->m_agfrotor + 1) %
2391
(mp->m_sb.sb_agcount * rotorstep);
2393
mp->m_agfrotor = (args->agno * rotorstep + 1) %
2394
(mp->m_sb.sb_agcount * rotorstep);
2401
if (args->agbno == NULLAGBLOCK)
2402
args->fsbno = NULLFSBLOCK;
2404
args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
2406
ASSERT(args->len >= args->minlen);
2407
ASSERT(args->len <= args->maxlen);
2408
ASSERT(args->agbno % args->alignment == 0);
2409
XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno),
2413
xfs_perag_put(args->pag);
2416
xfs_perag_put(args->pag);
2422
* Just break up the extent address and hand off to xfs_free_ag_extent
2423
* after fixing up the freelist.
2427
xfs_trans_t *tp, /* transaction pointer */
2428
xfs_fsblock_t bno, /* starting block number of extent */
2429
xfs_extlen_t len) /* length of extent */
2431
xfs_alloc_arg_t args;
2435
memset(&args, 0, sizeof(xfs_alloc_arg_t));
2437
args.mp = tp->t_mountp;
2440
* validate that the block number is legal - the enables us to detect
2441
* and handle a silent filesystem corruption rather than crashing.
2443
args.agno = XFS_FSB_TO_AGNO(args.mp, bno);
2444
if (args.agno >= args.mp->m_sb.sb_agcount)
2445
return EFSCORRUPTED;
2447
args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno);
2448
if (args.agbno >= args.mp->m_sb.sb_agblocks)
2449
return EFSCORRUPTED;
2451
args.pag = xfs_perag_get(args.mp, args.agno);
2454
error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING);
2458
/* validate the extent size is legal now we have the agf locked */
2459
if (args.agbno + len >
2460
be32_to_cpu(XFS_BUF_TO_AGF(args.agbp)->agf_length)) {
2461
error = EFSCORRUPTED;
2465
error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0);
2467
xfs_alloc_busy_insert(tp, args.agno, args.agbno, len, 0);
2469
xfs_perag_put(args.pag);
2474
xfs_alloc_busy_insert(
2475
struct xfs_trans *tp,
2476
xfs_agnumber_t agno,
2481
struct xfs_busy_extent *new;
2482
struct xfs_busy_extent *busyp;
2483
struct xfs_perag *pag;
2484
struct rb_node **rbp;
2485
struct rb_node *parent = NULL;
2487
new = kmem_zalloc(sizeof(struct xfs_busy_extent), KM_MAYFAIL);
2490
* No Memory! Since it is now not possible to track the free
2491
* block, make this a synchronous transaction to insure that
2492
* the block is not reused before this transaction commits.
2494
trace_xfs_alloc_busy_enomem(tp->t_mountp, agno, bno, len);
2495
xfs_trans_set_sync(tp);
2502
INIT_LIST_HEAD(&new->list);
2505
/* trace before insert to be able to see failed inserts */
2506
trace_xfs_alloc_busy(tp->t_mountp, agno, bno, len);
2508
pag = xfs_perag_get(tp->t_mountp, new->agno);
2509
spin_lock(&pag->pagb_lock);
2510
rbp = &pag->pagb_tree.rb_node;
2513
busyp = rb_entry(parent, struct xfs_busy_extent, rb_node);
2515
if (new->bno < busyp->bno) {
2516
rbp = &(*rbp)->rb_left;
2517
ASSERT(new->bno + new->length <= busyp->bno);
2518
} else if (new->bno > busyp->bno) {
2519
rbp = &(*rbp)->rb_right;
2520
ASSERT(bno >= busyp->bno + busyp->length);
2526
rb_link_node(&new->rb_node, parent, rbp);
2527
rb_insert_color(&new->rb_node, &pag->pagb_tree);
2529
list_add(&new->list, &tp->t_busy);
2530
spin_unlock(&pag->pagb_lock);
2535
* Search for a busy extent within the range of the extent we are about to
2536
* allocate. You need to be holding the busy extent tree lock when calling
2537
* xfs_alloc_busy_search(). This function returns 0 for no overlapping busy
2538
* extent, -1 for an overlapping but not exact busy extent, and 1 for an exact
2539
* match. This is done so that a non-zero return indicates an overlap that
2540
* will require a synchronous transaction, but it can still be
2541
* used to distinguish between a partial or exact match.
2544
xfs_alloc_busy_search(
2545
struct xfs_mount *mp,
2546
xfs_agnumber_t agno,
2550
struct xfs_perag *pag;
2551
struct rb_node *rbp;
2552
struct xfs_busy_extent *busyp;
2555
pag = xfs_perag_get(mp, agno);
2556
spin_lock(&pag->pagb_lock);
2558
rbp = pag->pagb_tree.rb_node;
2560
/* find closest start bno overlap */
2562
busyp = rb_entry(rbp, struct xfs_busy_extent, rb_node);
2563
if (bno < busyp->bno) {
2564
/* may overlap, but exact start block is lower */
2565
if (bno + len > busyp->bno)
2568
} else if (bno > busyp->bno) {
2569
/* may overlap, but exact start block is higher */
2570
if (bno < busyp->bno + busyp->length)
2572
rbp = rbp->rb_right;
2574
/* bno matches busyp, length determines exact match */
2575
match = (busyp->length == len) ? 1 : -1;
2579
spin_unlock(&pag->pagb_lock);
2585
* The found free extent [fbno, fend] overlaps part or all of the given busy
2586
* extent. If the overlap covers the beginning, the end, or all of the busy
2587
* extent, the overlapping portion can be made unbusy and used for the
2588
* allocation. We can't split a busy extent because we can't modify a
2589
* transaction/CIL context busy list, but we can update an entries block
2592
* Returns true if the extent can safely be reused, or false if the search
2593
* needs to be restarted.
2596
xfs_alloc_busy_update_extent(
2597
struct xfs_mount *mp,
2598
struct xfs_perag *pag,
2599
struct xfs_busy_extent *busyp,
2604
xfs_agblock_t fend = fbno + flen;
2605
xfs_agblock_t bbno = busyp->bno;
2606
xfs_agblock_t bend = bbno + busyp->length;
2609
* This extent is currently being discarded. Give the thread
2610
* performing the discard a chance to mark the extent unbusy
2613
if (busyp->flags & XFS_ALLOC_BUSY_DISCARDED) {
2614
spin_unlock(&pag->pagb_lock);
2616
spin_lock(&pag->pagb_lock);
2621
* If there is a busy extent overlapping a user allocation, we have
2622
* no choice but to force the log and retry the search.
2624
* Fortunately this does not happen during normal operation, but
2625
* only if the filesystem is very low on space and has to dip into
2626
* the AGFL for normal allocations.
2631
if (bbno < fbno && bend > fend) {
2635
* +BBBBBBBBBBBBBBBBB+
2641
* We would have to split the busy extent to be able to track
2642
* it correct, which we cannot do because we would have to
2643
* modify the list of busy extents attached to the transaction
2644
* or CIL context, which is immutable.
2646
* Force out the log to clear the busy extent and retry the
2650
} else if (bbno >= fbno && bend <= fend) {
2654
* +BBBBBBBBBBBBBBBBB+
2655
* +-----------------+
2660
* +BBBBBBBBBBBBBBBBB+
2661
* +--------------------------+
2666
* +BBBBBBBBBBBBBBBBB+
2667
* +--------------------------+
2672
* +BBBBBBBBBBBBBBBBB+
2673
* +-----------------------------------+
2679
* The busy extent is fully covered by the extent we are
2680
* allocating, and can simply be removed from the rbtree.
2681
* However we cannot remove it from the immutable list
2682
* tracking busy extents in the transaction or CIL context,
2683
* so set the length to zero to mark it invalid.
2685
* We also need to restart the busy extent search from the
2686
* tree root, because erasing the node can rearrange the
2689
rb_erase(&busyp->rb_node, &pag->pagb_tree);
2692
} else if (fend < bend) {
2696
* +BBBBBBBBBBBBBBBBB+
2702
* +BBBBBBBBBBBBBBBBB+
2703
* +------------------+
2708
} else if (bbno < fbno) {
2712
* +BBBBBBBBBBBBBBBBB+
2718
* +BBBBBBBBBBBBBBBBB+
2719
* +----------------------+
2722
busyp->length = fbno - busyp->bno;
2727
trace_xfs_alloc_busy_reuse(mp, pag->pag_agno, fbno, flen);
2731
spin_unlock(&pag->pagb_lock);
2732
xfs_log_force(mp, XFS_LOG_SYNC);
2733
trace_xfs_alloc_busy_force(mp, pag->pag_agno, fbno, flen);
2734
spin_lock(&pag->pagb_lock);
2740
* For a given extent [fbno, flen], make sure we can reuse it safely.
2743
xfs_alloc_busy_reuse(
2744
struct xfs_mount *mp,
2745
xfs_agnumber_t agno,
2750
struct xfs_perag *pag;
2751
struct rb_node *rbp;
2755
pag = xfs_perag_get(mp, agno);
2756
spin_lock(&pag->pagb_lock);
2758
rbp = pag->pagb_tree.rb_node;
2760
struct xfs_busy_extent *busyp =
2761
rb_entry(rbp, struct xfs_busy_extent, rb_node);
2762
xfs_agblock_t bbno = busyp->bno;
2763
xfs_agblock_t bend = bbno + busyp->length;
2765
if (fbno + flen <= bbno) {
2768
} else if (fbno >= bend) {
2769
rbp = rbp->rb_right;
2773
if (!xfs_alloc_busy_update_extent(mp, pag, busyp, fbno, flen,
2777
spin_unlock(&pag->pagb_lock);
2782
* For a given extent [fbno, flen], search the busy extent list to find a
2783
* subset of the extent that is not busy. If *rlen is smaller than
2784
* args->minlen no suitable extent could be found, and the higher level
2785
* code needs to force out the log and retry the allocation.
2788
xfs_alloc_busy_trim(
2789
struct xfs_alloc_arg *args,
2792
xfs_agblock_t *rbno,
2797
struct rb_node *rbp;
2801
spin_lock(&args->pag->pagb_lock);
2805
rbp = args->pag->pagb_tree.rb_node;
2806
while (rbp && flen >= args->minlen) {
2807
struct xfs_busy_extent *busyp =
2808
rb_entry(rbp, struct xfs_busy_extent, rb_node);
2809
xfs_agblock_t fend = fbno + flen;
2810
xfs_agblock_t bbno = busyp->bno;
2811
xfs_agblock_t bend = bbno + busyp->length;
2816
} else if (fbno >= bend) {
2817
rbp = rbp->rb_right;
2822
* If this is a metadata allocation, try to reuse the busy
2823
* extent instead of trimming the allocation.
2825
if (!args->userdata &&
2826
!(busyp->flags & XFS_ALLOC_BUSY_DISCARDED)) {
2827
if (!xfs_alloc_busy_update_extent(args->mp, args->pag,
2840
* +BBBBBBBBBBBBBBBBB+
2846
* +BBBBBBBBBBBBBBBBB+
2852
* +BBBBBBBBBBBBBBBBB+
2858
* +BBBBBBBBBBBBBBBBB+
2859
* +-----------------+
2862
* No unbusy region in extent, return failure.
2870
* +BBBBBBBBBBBBBBBBB+
2871
* +----------------------+
2876
* +BBBBBBBBBBBBBBBBB+
2877
* +--------------------------+
2880
* Needs to be trimmed to:
2885
} else if (bend >= fend) {
2891
* +BBBBBBBBBBBBBBBBB+
2892
* +------------------+
2897
* +BBBBBBBBBBBBBBBBB+
2898
* +--------------------------+
2901
* Needs to be trimmed to:
2907
/* middle overlap */
2912
* +BBBBBBBBBBBBBBBBB+
2913
* +-----------------------------------+
2916
* Can be trimmed to:
2917
* +-------+ OR +-------+
2918
* fbno fend fbno fend
2920
* Backward allocation leads to significant
2921
* fragmentation of directories, which degrades
2922
* directory performance, therefore we always want to
2923
* choose the option that produces forward allocation
2925
* Preferring the lower bno extent will make the next
2926
* request use "fend" as the start of the next
2927
* allocation; if the segment is no longer busy at
2928
* that point, we'll get a contiguous allocation, but
2929
* even if it is still busy, we will get a forward
2931
* We try to avoid choosing the segment at "bend",
2932
* because that can lead to the next allocation
2933
* taking the segment at "fbno", which would be a
2934
* backward allocation. We only use the segment at
2935
* "fbno" if it is much larger than the current
2936
* requested size, because in that case there's a
2937
* good chance subsequent allocations will be
2940
if (bbno - fbno >= args->maxlen) {
2941
/* left candidate fits perfect */
2943
} else if (fend - bend >= args->maxlen * 4) {
2944
/* right candidate has enough free space */
2946
} else if (bbno - fbno >= args->minlen) {
2947
/* left candidate fits minimum requirement */
2956
spin_unlock(&args->pag->pagb_lock);
2958
if (fbno != bno || flen != len) {
2959
trace_xfs_alloc_busy_trim(args->mp, args->agno, bno, len,
2967
* Return a zero extent length as failure indications. All callers
2968
* re-check if the trimmed extent satisfies the minlen requirement.
2970
spin_unlock(&args->pag->pagb_lock);
2971
trace_xfs_alloc_busy_trim(args->mp, args->agno, bno, len, fbno, 0);
2977
xfs_alloc_busy_clear_one(
2978
struct xfs_mount *mp,
2979
struct xfs_perag *pag,
2980
struct xfs_busy_extent *busyp)
2982
if (busyp->length) {
2983
trace_xfs_alloc_busy_clear(mp, busyp->agno, busyp->bno,
2985
rb_erase(&busyp->rb_node, &pag->pagb_tree);
2988
list_del_init(&busyp->list);
2993
* Remove all extents on the passed in list from the busy extents tree.
2994
* If do_discard is set skip extents that need to be discarded, and mark
2995
* these as undergoing a discard operation instead.
2998
xfs_alloc_busy_clear(
2999
struct xfs_mount *mp,
3000
struct list_head *list,
3003
struct xfs_busy_extent *busyp, *n;
3004
struct xfs_perag *pag = NULL;
3005
xfs_agnumber_t agno = NULLAGNUMBER;
3007
list_for_each_entry_safe(busyp, n, list, list) {
3008
if (busyp->agno != agno) {
3010
spin_unlock(&pag->pagb_lock);
3013
pag = xfs_perag_get(mp, busyp->agno);
3014
spin_lock(&pag->pagb_lock);
3018
if (do_discard && busyp->length &&
3019
!(busyp->flags & XFS_ALLOC_BUSY_SKIP_DISCARD))
3020
busyp->flags = XFS_ALLOC_BUSY_DISCARDED;
3022
xfs_alloc_busy_clear_one(mp, pag, busyp);
3026
spin_unlock(&pag->pagb_lock);
3032
* Callback for list_sort to sort busy extents by the AG they reside in.
3035
xfs_busy_extent_ag_cmp(
3037
struct list_head *a,
3038
struct list_head *b)
3040
return container_of(a, struct xfs_busy_extent, list)->agno -
3041
container_of(b, struct xfs_busy_extent, list)->agno;