2
* Copyright (c) 2000 Silicon Graphics, Inc. All Rights Reserved.
4
* This program is free software; you can redistribute it and/or modify it
5
* under the terms of version 2 of the GNU General Public License as
6
* published by the Free Software Foundation.
8
* This program is distributed in the hope that it would be useful, but
9
* WITHOUT ANY WARRANTY; without even the implied warranty of
10
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12
* Further, this software is distributed without any warranty that it is
13
* free of the rightful claim of any third person regarding infringement
14
* or the like. Any license provided herein, whether implied or
15
* otherwise, applies only to this software file. Patent licenses, if
16
* any, provided herein do not apply to combinations of this program with
17
* other software, or any other product whatsoever.
19
* You should have received a copy of the GNU General Public License along
20
* with this program; if not, write the Free Software Foundation, Inc., 59
21
* Temple Place - Suite 330, Boston MA 02111-1307, USA.
23
* Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24
* Mountain View, CA 94043, or:
28
* For further information regarding this notice, see:
30
* http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
32
#ifndef __XFS_LOG_PRIV_H__
33
#define __XFS_LOG_PRIV_H__
35
#if defined(XFS_ALL_TRACE)
46
struct xfs_buf_cancel;
50
* Macros, structures, prototypes for internal log manager use.
53
#define XLOG_NUM_ICLOGS 2
54
#define XLOG_MAX_ICLOGS 8
55
#define XLOG_CALLBACK_SIZE 10
56
#define XLOG_HEADER_MAGIC_NUM 0xFEEDbabe /* Illegal cycle number */
57
#define XLOG_RECORD_BSIZE (16*1024) /* eventually 32k */
58
#define XLOG_MAX_RECORD_BSIZE (32*1024)
59
#define XLOG_RECORD_BSHIFT 14 /* 16384 == 1 << 14 */
60
#define XLOG_MAX_RECORD_BSHIFT 15 /* 32k == 1 << 15 */
61
#if XFS_WANT_FUNCS || (XFS_WANT_SPACE && XFSSO_XLOG_BTOLRBB)
62
int xlog_btolrbb(int b);
63
#define XLOG_BTOLRBB(b) xlog_btolrbb(b)
65
#define XLOG_BTOLRBB(b) (((b)+XLOG_RECORD_BSIZE-1) >> XLOG_RECORD_BSHIFT)
68
#define XLOG_HEADER_SIZE 512
74
#define ASSIGN_LSN_CYCLE(lsn,cycle,arch) \
75
INT_SET(((uint *)&(lsn))[LSN_FIELD_CYCLE(arch)], arch, (cycle));
76
#define ASSIGN_LSN_BLOCK(lsn,block,arch) \
77
INT_SET(((uint *)&(lsn))[LSN_FIELD_BLOCK(arch)], arch, (block));
78
#define ASSIGN_ANY_LSN(lsn,cycle,block,arch) \
80
ASSIGN_LSN_CYCLE(lsn,cycle,arch); \
81
ASSIGN_LSN_BLOCK(lsn,block,arch); \
83
#define ASSIGN_LSN(lsn,log,arch) \
84
ASSIGN_ANY_LSN(lsn,(log)->l_curr_cycle,(log)->l_curr_block,arch);
86
#define XLOG_SET(f,b) (((f) & (b)) == (b))
88
#define GET_CYCLE(ptr, arch) \
89
(INT_GET(*(uint *)(ptr), arch) == XLOG_HEADER_MAGIC_NUM ? \
90
INT_GET(*((uint *)(ptr)+1), arch) : \
91
INT_GET(*(uint *)(ptr), arch) \
94
#define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
99
* get client id from packed copy.
101
* this hack is here because the xlog_pack code copies four bytes
102
* of xlog_op_header containing the fields oh_clientid, oh_flags
103
* and oh_res2 into the packed copy.
105
* later on this four byte chunk is treated as an int and the
106
* client id is pulled out.
108
* this has endian issues, of course.
111
#if __BYTE_ORDER == __LITTLE_ENDIAN
112
#define GET_CLIENT_ID(i,arch) \
115
#define GET_CLIENT_ID(i,arch) \
119
#if XFS_WANT_FUNCS || (XFS_WANT_SPACE && XFSSO_XLOG_GRANT_SUB_SPACE)
120
void xlog_grant_sub_space(struct log *log, int bytes, int type);
121
#define XLOG_GRANT_SUB_SPACE(log,bytes,type) \
122
xlog_grant_sub_space(log,bytes,type)
124
#define XLOG_GRANT_SUB_SPACE(log,bytes,type) \
127
(log)->l_grant_write_bytes -= (bytes); \
128
if ((log)->l_grant_write_bytes < 0) { \
129
(log)->l_grant_write_bytes += (log)->l_logsize; \
130
(log)->l_grant_write_cycle--; \
133
(log)->l_grant_reserve_bytes -= (bytes); \
134
if ((log)->l_grant_reserve_bytes < 0) { \
135
(log)->l_grant_reserve_bytes += (log)->l_logsize;\
136
(log)->l_grant_reserve_cycle--; \
141
#if XFS_WANT_FUNCS || (XFS_WANT_SPACE && XFSSO_XLOG_GRANT_ADD_SPACE)
142
void xlog_grant_add_space(struct log *log, int bytes, int type);
143
#define XLOG_GRANT_ADD_SPACE(log,bytes,type) \
144
xlog_grant_add_space(log,bytes,type)
146
#define XLOG_GRANT_ADD_SPACE(log,bytes,type) \
149
(log)->l_grant_write_bytes += (bytes); \
150
if ((log)->l_grant_write_bytes > (log)->l_logsize) { \
151
(log)->l_grant_write_bytes -= (log)->l_logsize; \
152
(log)->l_grant_write_cycle++; \
155
(log)->l_grant_reserve_bytes += (bytes); \
156
if ((log)->l_grant_reserve_bytes > (log)->l_logsize) { \
157
(log)->l_grant_reserve_bytes -= (log)->l_logsize;\
158
(log)->l_grant_reserve_cycle++; \
163
#define XLOG_INS_TICKETQ(q,tic) \
166
(tic)->t_next = (q); \
167
(tic)->t_prev = (q)->t_prev; \
168
(q)->t_prev->t_next = (tic); \
169
(q)->t_prev = (tic); \
171
(tic)->t_prev = (tic)->t_next = (tic); \
174
(tic)->t_flags |= XLOG_TIC_IN_Q; \
176
#define XLOG_DEL_TICKETQ(q,tic) \
178
if ((tic) == (tic)->t_next) { \
181
(q) = (tic)->t_next; \
182
(tic)->t_next->t_prev = (tic)->t_prev; \
183
(tic)->t_prev->t_next = (tic)->t_next; \
185
(tic)->t_next = (tic)->t_prev = NULL; \
186
(tic)->t_flags &= ~XLOG_TIC_IN_Q; \
190
#define GRANT_LOCK(log) mutex_spinlock(&(log)->l_grant_lock)
191
#define GRANT_UNLOCK(log, s) mutex_spinunlock(&(log)->l_grant_lock, s)
192
#define LOG_LOCK(log) mutex_spinlock(&(log)->l_icloglock)
193
#define LOG_UNLOCK(log, s) mutex_spinunlock(&(log)->l_icloglock, s)
195
#define xlog_panic(s) {cmn_err(CE_PANIC, s); }
196
#define xlog_exit(s) {cmn_err(CE_PANIC, s); }
197
#define xlog_warn(s) {cmn_err(CE_WARN, s); }
202
#define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
203
#define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
204
#define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
205
#define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
206
#define XLOG_STATE_DO_CALLBACK \
207
0x0010 /* Process callback functions */
208
#define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
209
#define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
210
#define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
211
#define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
212
#define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
213
#endif /* __KERNEL__ */
216
* Flags to log operation header
218
* The first write of a new transaction will be preceded with a start
219
* record, XLOG_START_TRANS. Once a transaction is committed, a commit
220
* record is written, XLOG_COMMIT_TRANS. If a single region can not fit into
221
* the remainder of the current active in-core log, it is split up into
222
* multiple regions. Each partial region will be marked with a
223
* XLOG_CONTINUE_TRANS until the last one, which gets marked with XLOG_END_TRANS.
226
#define XLOG_START_TRANS 0x01 /* Start a new transaction */
227
#define XLOG_COMMIT_TRANS 0x02 /* Commit this transaction */
228
#define XLOG_CONTINUE_TRANS 0x04 /* Cont this trans into new region */
229
#define XLOG_WAS_CONT_TRANS 0x08 /* Cont this trans into new region */
230
#define XLOG_END_TRANS 0x10 /* End a continued transaction */
231
#define XLOG_UNMOUNT_TRANS 0x20 /* Unmount a filesystem transaction */
232
#define XLOG_SKIP_TRANS (XLOG_COMMIT_TRANS | XLOG_CONTINUE_TRANS | \
233
XLOG_WAS_CONT_TRANS | XLOG_END_TRANS | \
238
* Flags to log ticket
240
#define XLOG_TIC_INITED 0x1 /* has been initialized */
241
#define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
242
#define XLOG_TIC_IN_Q 0x4
243
#endif /* __KERNEL__ */
245
#define XLOG_UNMOUNT_TYPE 0x556e /* Un for Unmount */
248
* Flags for log structure
250
#define XLOG_CHKSUM_MISMATCH 0x1 /* used only during recovery */
251
#define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
252
#define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
253
#define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
255
typedef __uint32_t xlog_tid_t;
260
* Below are states for covering allocation transactions.
261
* By covering, we mean changing the h_tail_lsn in the last on-disk
262
* log write such that no allocation transactions will be re-done during
263
* recovery after a system crash. Recovery starts at the last on-disk
266
* These states are used to insert dummy log entries to cover
267
* space allocation transactions which can undo non-transactional changes
268
* after a crash. Writes to a file with space
269
* already allocated do not result in any transactions. Allocations
270
* might include space beyond the EOF. So if we just push the EOF a
271
* little, the last transaction for the file could contain the wrong
272
* size. If there is no file system activity, after an allocation
273
* transaction, and the system crashes, the allocation transaction
274
* will get replayed and the file will be truncated. This could
275
* be hours/days/... after the allocation occurred.
277
* The fix for this is to do two dummy transactions when the
278
* system is idle. We need two dummy transaction because the h_tail_lsn
279
* in the log record header needs to point beyond the last possible
280
* non-dummy transaction. The first dummy changes the h_tail_lsn to
281
* the first transaction before the dummy. The second dummy causes
282
* h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
284
* These dummy transactions get committed when everything
285
* is idle (after there has been some activity).
287
* There are 5 states used to control this.
289
* IDLE -- no logging has been done on the file system or
290
* we are done covering previous transactions.
291
* NEED -- logging has occurred and we need a dummy transaction
292
* when the log becomes idle.
293
* DONE -- we were in the NEED state and have committed a dummy
295
* NEED2 -- we detected that a dummy transaction has gone to the
296
* on disk log with no other transactions.
297
* DONE2 -- we committed a dummy transaction when in the NEED2 state.
299
* There are two places where we switch states:
301
* 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
302
* We commit the dummy transaction and switch to DONE or DONE2,
303
* respectively. In all other states, we don't do anything.
305
* 2.) When we finish writing the on-disk log (xlog_state_clean_log).
307
* No matter what state we are in, if this isn't the dummy
308
* transaction going out, the next state is NEED.
309
* So, if we aren't in the DONE or DONE2 states, the next state
310
* is NEED. We can't be finishing a write of the dummy record
311
* unless it was committed and the state switched to DONE or DONE2.
313
* If we are in the DONE state and this was a write of the
314
* dummy transaction, we move to NEED2.
316
* If we are in the DONE2 state and this was a write of the
317
* dummy transaction, we move to IDLE.
320
* Writing only one dummy transaction can get appended to
321
* one file space allocation. When this happens, the log recovery
322
* code replays the space allocation and a file could be truncated.
323
* This is why we have the NEED2 and DONE2 states before going idle.
326
#define XLOG_STATE_COVER_IDLE 0
327
#define XLOG_STATE_COVER_NEED 1
328
#define XLOG_STATE_COVER_DONE 2
329
#define XLOG_STATE_COVER_NEED2 3
330
#define XLOG_STATE_COVER_DONE2 4
332
#define XLOG_COVER_OPS 5
334
typedef struct xlog_ticket {
335
sv_t t_sema; /* sleep on this semaphore :20 */
336
struct xlog_ticket *t_next; /* : 4 */
337
struct xlog_ticket *t_prev; /* : 4 */
338
xlog_tid_t t_tid; /* transaction identifier : 4 */
339
int t_curr_res; /* current reservation in bytes : 4 */
340
int t_unit_res; /* unit reservation in bytes : 4 */
341
__uint8_t t_ocnt; /* original count : 1 */
342
__uint8_t t_cnt; /* current count : 1 */
343
__uint8_t t_clientid; /* who does this belong to; : 1 */
344
__uint8_t t_flags; /* properties of reservation : 1 */
349
typedef struct xlog_op_header {
350
xlog_tid_t oh_tid; /* transaction id of operation : 4 b */
351
int oh_len; /* bytes in data region : 2 b */
352
__uint8_t oh_clientid; /* who sent me this : 1 b */
353
__uint8_t oh_flags; /* : 1 b */
354
ushort oh_res2; /* 32 bit align : 2 b */
358
/* valid values for h_fmt */
359
#define XLOG_FMT_UNKNOWN 0
360
#define XLOG_FMT_LINUX_LE 1
361
#define XLOG_FMT_LINUX_BE 2
362
#define XLOG_FMT_IRIX_BE 3
365
#if __BYTE_ORDER == __LITTLE_ENDIAN
366
#define XLOG_FMT XLOG_FMT_LINUX_LE
368
#if __BYTE_ORDER == __BIG_ENDIAN
369
#define XLOG_FMT XLOG_FMT_LINUX_BE
371
#error unknown byte order
375
typedef struct xlog_rec_header {
376
uint h_magicno; /* log record (LR) identifier : 4 */
377
uint h_cycle; /* write cycle of log : 4 */
378
int h_version; /* LR version : 4 */
379
int h_len; /* len in bytes; should be 64-bit aligned: 4 */
380
xfs_lsn_t h_lsn; /* lsn of this LR : 8 */
381
xfs_lsn_t h_tail_lsn; /* lsn of 1st LR w/ buffers not committed: 8 */
382
uint h_chksum; /* may not be used; non-zero if used : 4 */
383
int h_prev_block; /* block number to previous LR : 4 */
384
int h_num_logops; /* number of log operations in this LR : 4 */
385
uint h_cycle_data[XLOG_MAX_RECORD_BSIZE / BBSIZE];
387
int h_fmt; /* format of log record : 4 */
388
uuid_t h_fs_uuid; /* uuid of FS : 16 */
393
* - A log record header is 512 bytes. There is plenty of room to grow the
394
* xlog_rec_header_t into the reserved space.
395
* - ic_data follows, so a write to disk can start at the beginning of
397
* - ic_forcesema is used to implement synchronous forcing of the iclog to disk.
398
* - ic_next is the pointer to the next iclog in the ring.
399
* - ic_bp is a pointer to the buffer used to write this incore log to disk.
400
* - ic_log is a pointer back to the global log structure.
401
* - ic_callback is a linked list of callback function/argument pairs to be
402
* called after an iclog finishes writing.
403
* - ic_size is the full size of the header plus data.
404
* - ic_offset is the current number of bytes written to in this iclog.
405
* - ic_refcnt is bumped when someone is writing to the log.
406
* - ic_state is the state of the iclog.
408
typedef struct xlog_iclog_fields {
410
struct xlog_in_core *ic_next;
411
struct xlog_in_core *ic_prev;
412
struct xfs_buf *ic_bp;
414
xfs_log_callback_t *ic_callback;
415
xfs_log_callback_t **ic_callback_tail;
417
struct ktrace *ic_trace;
425
} xlog_iclog_fields_t;
427
typedef struct xlog_in_core2 {
429
xlog_rec_header_t hic_header;
430
char hic_sector[XLOG_HEADER_SIZE];
432
__uint8_t hic_data[1];
435
typedef struct xlog_in_core {
436
xlog_iclog_fields_t hic_fields;
437
xlog_in_core_2_t *hic_data;
441
* Defines to save our code from this glop.
443
#define ic_forcesema hic_fields.ic_forcesema
444
#define ic_next hic_fields.ic_next
445
#define ic_prev hic_fields.ic_prev
446
#define ic_bp hic_fields.ic_bp
447
#define ic_log hic_fields.ic_log
448
#define ic_callback hic_fields.ic_callback
449
#define ic_callback_tail hic_fields.ic_callback_tail
450
#define ic_trace hic_fields.ic_trace
451
#define ic_size hic_fields.ic_size
452
#define ic_offset hic_fields.ic_offset
453
#define ic_refcnt hic_fields.ic_refcnt
454
#define ic_roundoff hic_fields.ic_roundoff
455
#define ic_bwritecnt hic_fields.ic_bwritecnt
456
#define ic_state hic_fields.ic_state
457
#define ic_header hic_data->ic_h.hic_header
458
#define ic_data hic_data->hic_data
461
* The reservation head lsn is not made up of a cycle number and block number.
462
* Instead, it uses a cycle number and byte number. Logs don't expect to
463
* overflow 31 bits worth of byte offset, so using a byte number will mean
464
* that round off problems won't occur when releasing partial reservations.
467
/* The following block of fields are changed while holding icloglock */
468
sema_t l_flushsema; /* iclog flushing semaphore */
469
int l_flushcnt; /* # of procs waiting on this sema */
470
int l_ticket_cnt; /* free ticket count */
471
int l_ticket_tcnt; /* total ticket count */
472
int l_covered_state;/* state of "covering disk log entries" */
473
xlog_ticket_t *l_freelist; /* free list of tickets */
474
xlog_ticket_t *l_unmount_free;/* kmem_free these addresses */
475
xlog_ticket_t *l_tail; /* free list of tickets */
476
xlog_in_core_t *l_iclog; /* head log queue */
477
lock_t l_icloglock; /* grab to change iclog state */
478
xfs_lsn_t l_tail_lsn; /* lsn of 1st LR w/ unflush buffers */
479
xfs_lsn_t l_last_sync_lsn;/* lsn of last LR on disk */
480
struct xfs_mount *l_mp; /* mount point */
481
struct xfs_buf *l_xbuf; /* extra buffer for log wrapping */
482
dev_t l_dev; /* dev_t of log */
483
xfs_daddr_t l_logBBstart; /* start block of log */
484
int l_logsize; /* size of log in bytes */
485
int l_logBBsize; /* size of log in 512 byte chunks */
486
int l_roundoff; /* round off error of all iclogs */
487
int l_curr_cycle; /* Cycle number of log writes */
488
int l_prev_cycle; /* Cycle # b4 last block increment */
489
int l_curr_block; /* current logical block of log */
490
int l_prev_block; /* previous logical block of log */
491
int l_iclog_size; /* size of log in bytes */
492
int l_iclog_size_log;/* log power size of log */
493
int l_iclog_bufs; /* number of iclog buffers */
495
/* The following field are used for debugging; need to hold icloglock */
496
char *l_iclog_bak[XLOG_MAX_ICLOGS];
498
/* The following block of fields are changed while holding grant_lock */
499
lock_t l_grant_lock; /* protects below fields */
500
xlog_ticket_t *l_reserve_headq; /* */
501
xlog_ticket_t *l_write_headq; /* */
502
int l_grant_reserve_cycle; /* */
503
int l_grant_reserve_bytes; /* */
504
int l_grant_write_cycle; /* */
505
int l_grant_write_bytes; /* */
507
/* The following fields don't need locking */
509
struct ktrace *l_trace;
510
struct ktrace *l_grant_trace;
513
uint l_quotaoffs_flag;/* XFS_DQ_*, if QUOTAOFFs found */
514
struct xfs_buf_cancel **l_buf_cancel_table;
518
/* common routines */
519
extern xfs_lsn_t xlog_assign_tail_lsn(struct xfs_mount *mp,
520
xlog_in_core_t *iclog);
521
extern int xlog_find_head(xlog_t *log, xfs_daddr_t *head_blk);
522
extern int xlog_find_tail(xlog_t *log,
523
xfs_daddr_t *head_blk,
524
xfs_daddr_t *tail_blk,
526
extern int xlog_print_find_oldest(xlog_t *log, xfs_daddr_t *last_blk);
527
extern int xlog_recover(xlog_t *log, int readonly);
528
extern int xlog_recover_finish(xlog_t *log, int mfsi_flags);
529
extern void xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog);
530
extern struct xfs_buf *xlog_get_bp(int,xfs_mount_t *);
531
extern void xlog_put_bp(struct xfs_buf *);
532
extern int xlog_bread(xlog_t *, xfs_daddr_t blkno, int bblks, struct xfs_buf *bp);
533
extern void xlog_recover_process_iunlinks(xlog_t *log);
535
#define XLOG_TRACE_GRAB_FLUSH 1
536
#define XLOG_TRACE_REL_FLUSH 2
537
#define XLOG_TRACE_SLEEP_FLUSH 3
538
#define XLOG_TRACE_WAKE_FLUSH 4
540
#endif /* __KERNEL__ */
542
#endif /* __XFS_LOG_PRIV_H__ */