2 * Copyright (c) 2000-2003 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__
38 struct xfs_buf_cancel;
42 * Macros, structures, prototypes for internal log manager use.
45 #define XLOG_MIN_ICLOGS 2
46 #define XLOG_MED_ICLOGS 4
47 #define XLOG_MAX_ICLOGS 8
48 #define XLOG_CALLBACK_SIZE 10
49 #define XLOG_HEADER_MAGIC_NUM 0xFEEDbabe /* Invalid cycle number */
50 #define XLOG_VERSION_1 1
51 #define XLOG_VERSION_2 2 /* Large IClogs, Log sunit */
52 #define XLOG_VERSION_OKBITS (XLOG_VERSION_1 | XLOG_VERSION_2)
53 #define XLOG_RECORD_BSIZE (16*1024) /* eventually 32k */
54 #define XLOG_BIG_RECORD_BSIZE (32*1024) /* 32k buffers */
55 #define XLOG_MAX_RECORD_BSIZE (256*1024)
56 #define XLOG_HEADER_CYCLE_SIZE (32*1024) /* cycle data in header */
57 #define XLOG_RECORD_BSHIFT 14 /* 16384 == 1 << 14 */
58 #define XLOG_BIG_RECORD_BSHIFT 15 /* 32k == 1 << 15 */
59 #define XLOG_MAX_RECORD_BSHIFT 18 /* 256k == 1 << 18 */
60 #define XLOG_BTOLSUNIT(log, b) (((b)+(log)->l_mp->m_sb.sb_logsunit-1) / \
61 (log)->l_mp->m_sb.sb_logsunit)
62 #define XLOG_LSUNITTOB(log, su) ((su) * (log)->l_mp->m_sb.sb_logsunit)
64 #define XLOG_HEADER_SIZE 512
66 #define XLOG_REC_SHIFT(log) \
67 BTOBB(1 << (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? \
68 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
69 #define XLOG_TOTAL_REC_SHIFT(log) \
70 BTOBB(XLOG_MAX_ICLOGS << (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? \
71 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
77 #define ASSIGN_ANY_LSN_HOST(lsn,cycle,block) \
79 (lsn) = ((xfs_lsn_t)(cycle)<<32)|(block); \
81 #define ASSIGN_ANY_LSN_DISK(lsn,cycle,block) \
83 INT_SET(((uint *)&(lsn))[0], ARCH_CONVERT, (cycle)); \
84 INT_SET(((uint *)&(lsn))[1], ARCH_CONVERT, (block)); \
86 #define ASSIGN_LSN(lsn,log) \
87 ASSIGN_ANY_LSN_DISK(lsn,(log)->l_curr_cycle,(log)->l_curr_block);
89 #define XLOG_SET(f,b) (((f) & (b)) == (b))
91 #define GET_CYCLE(ptr, arch) \
92 (INT_GET(*(uint *)(ptr), arch) == XLOG_HEADER_MAGIC_NUM ? \
93 INT_GET(*((uint *)(ptr)+1), arch) : \
94 INT_GET(*(uint *)(ptr), arch) \
97 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
103 * get client id from packed copy.
105 * this hack is here because the xlog_pack code copies four bytes
106 * of xlog_op_header containing the fields oh_clientid, oh_flags
107 * and oh_res2 into the packed copy.
109 * later on this four byte chunk is treated as an int and the
110 * client id is pulled out.
112 * this has endian issues, of course.
115 #if __BYTE_ORDER == __LITTLE_ENDIAN
116 #define GET_CLIENT_ID(i,arch) \
119 #define GET_CLIENT_ID(i,arch) \
123 #if XFS_WANT_FUNCS || (XFS_WANT_SPACE && XFSSO_XLOG_GRANT_SUB_SPACE)
124 void xlog_grant_sub_space(struct log *log, int bytes, int type);
125 #define XLOG_GRANT_SUB_SPACE(log,bytes,type) \
126 xlog_grant_sub_space(log,bytes,type)
128 #define XLOG_GRANT_SUB_SPACE(log,bytes,type) \
131 (log)->l_grant_write_bytes -= (bytes); \
132 if ((log)->l_grant_write_bytes < 0) { \
133 (log)->l_grant_write_bytes += (log)->l_logsize; \
134 (log)->l_grant_write_cycle--; \
137 (log)->l_grant_reserve_bytes -= (bytes); \
138 if ((log)->l_grant_reserve_bytes < 0) { \
139 (log)->l_grant_reserve_bytes += (log)->l_logsize;\
140 (log)->l_grant_reserve_cycle--; \
145 #if XFS_WANT_FUNCS || (XFS_WANT_SPACE && XFSSO_XLOG_GRANT_ADD_SPACE)
146 void xlog_grant_add_space(struct log *log, int bytes, int type);
147 #define XLOG_GRANT_ADD_SPACE(log,bytes,type) \
148 xlog_grant_add_space(log,bytes,type)
150 #define XLOG_GRANT_ADD_SPACE(log,bytes,type) \
153 (log)->l_grant_write_bytes += (bytes); \
154 if ((log)->l_grant_write_bytes > (log)->l_logsize) { \
155 (log)->l_grant_write_bytes -= (log)->l_logsize; \
156 (log)->l_grant_write_cycle++; \
159 (log)->l_grant_reserve_bytes += (bytes); \
160 if ((log)->l_grant_reserve_bytes > (log)->l_logsize) { \
161 (log)->l_grant_reserve_bytes -= (log)->l_logsize;\
162 (log)->l_grant_reserve_cycle++; \
167 #define XLOG_INS_TICKETQ(q,tic) \
170 (tic)->t_next = (q); \
171 (tic)->t_prev = (q)->t_prev; \
172 (q)->t_prev->t_next = (tic); \
173 (q)->t_prev = (tic); \
175 (tic)->t_prev = (tic)->t_next = (tic); \
178 (tic)->t_flags |= XLOG_TIC_IN_Q; \
180 #define XLOG_DEL_TICKETQ(q,tic) \
182 if ((tic) == (tic)->t_next) { \
185 (q) = (tic)->t_next; \
186 (tic)->t_next->t_prev = (tic)->t_prev; \
187 (tic)->t_prev->t_next = (tic)->t_next; \
189 (tic)->t_next = (tic)->t_prev = NULL; \
190 (tic)->t_flags &= ~XLOG_TIC_IN_Q; \
194 #define GRANT_LOCK(log) mutex_spinlock(&(log)->l_grant_lock)
195 #define GRANT_UNLOCK(log, s) mutex_spinunlock(&(log)->l_grant_lock, s)
196 #define LOG_LOCK(log) mutex_spinlock(&(log)->l_icloglock)
197 #define LOG_UNLOCK(log, s) mutex_spinunlock(&(log)->l_icloglock, s)
199 #define xlog_panic(args...) cmn_err(CE_PANIC, ## args)
200 #define xlog_exit(args...) cmn_err(CE_PANIC, ## args)
201 #define xlog_warn(args...) cmn_err(CE_WARN, ## args)
206 #define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
207 #define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
208 #define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
209 #define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
210 #define XLOG_STATE_DO_CALLBACK \
211 0x0010 /* Process callback functions */
212 #define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
213 #define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
214 #define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
215 #define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
216 #define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
217 #endif /* __KERNEL__ */
220 * Flags to log operation header
222 * The first write of a new transaction will be preceded with a start
223 * record, XLOG_START_TRANS. Once a transaction is committed, a commit
224 * record is written, XLOG_COMMIT_TRANS. If a single region can not fit into
225 * the remainder of the current active in-core log, it is split up into
226 * multiple regions. Each partial region will be marked with a
227 * XLOG_CONTINUE_TRANS until the last one, which gets marked with XLOG_END_TRANS.
230 #define XLOG_START_TRANS 0x01 /* Start a new transaction */
231 #define XLOG_COMMIT_TRANS 0x02 /* Commit this transaction */
232 #define XLOG_CONTINUE_TRANS 0x04 /* Cont this trans into new region */
233 #define XLOG_WAS_CONT_TRANS 0x08 /* Cont this trans into new region */
234 #define XLOG_END_TRANS 0x10 /* End a continued transaction */
235 #define XLOG_UNMOUNT_TRANS 0x20 /* Unmount a filesystem transaction */
236 #define XLOG_SKIP_TRANS (XLOG_COMMIT_TRANS | XLOG_CONTINUE_TRANS | \
237 XLOG_WAS_CONT_TRANS | XLOG_END_TRANS | \
242 * Flags to log ticket
244 #define XLOG_TIC_INITED 0x1 /* has been initialized */
245 #define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
246 #define XLOG_TIC_IN_Q 0x4
247 #endif /* __KERNEL__ */
249 #define XLOG_UNMOUNT_TYPE 0x556e /* Un for Unmount */
252 * Flags for log structure
254 #define XLOG_CHKSUM_MISMATCH 0x1 /* used only during recovery */
255 #define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
256 #define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
257 #define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
259 typedef __uint32_t xlog_tid_t;
264 * Below are states for covering allocation transactions.
265 * By covering, we mean changing the h_tail_lsn in the last on-disk
266 * log write such that no allocation transactions will be re-done during
267 * recovery after a system crash. Recovery starts at the last on-disk
270 * These states are used to insert dummy log entries to cover
271 * space allocation transactions which can undo non-transactional changes
272 * after a crash. Writes to a file with space
273 * already allocated do not result in any transactions. Allocations
274 * might include space beyond the EOF. So if we just push the EOF a
275 * little, the last transaction for the file could contain the wrong
276 * size. If there is no file system activity, after an allocation
277 * transaction, and the system crashes, the allocation transaction
278 * will get replayed and the file will be truncated. This could
279 * be hours/days/... after the allocation occurred.
281 * The fix for this is to do two dummy transactions when the
282 * system is idle. We need two dummy transaction because the h_tail_lsn
283 * in the log record header needs to point beyond the last possible
284 * non-dummy transaction. The first dummy changes the h_tail_lsn to
285 * the first transaction before the dummy. The second dummy causes
286 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
288 * These dummy transactions get committed when everything
289 * is idle (after there has been some activity).
291 * There are 5 states used to control this.
293 * IDLE -- no logging has been done on the file system or
294 * we are done covering previous transactions.
295 * NEED -- logging has occurred and we need a dummy transaction
296 * when the log becomes idle.
297 * DONE -- we were in the NEED state and have committed a dummy
299 * NEED2 -- we detected that a dummy transaction has gone to the
300 * on disk log with no other transactions.
301 * DONE2 -- we committed a dummy transaction when in the NEED2 state.
303 * There are two places where we switch states:
305 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
306 * We commit the dummy transaction and switch to DONE or DONE2,
307 * respectively. In all other states, we don't do anything.
309 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
311 * No matter what state we are in, if this isn't the dummy
312 * transaction going out, the next state is NEED.
313 * So, if we aren't in the DONE or DONE2 states, the next state
314 * is NEED. We can't be finishing a write of the dummy record
315 * unless it was committed and the state switched to DONE or DONE2.
317 * If we are in the DONE state and this was a write of the
318 * dummy transaction, we move to NEED2.
320 * If we are in the DONE2 state and this was a write of the
321 * dummy transaction, we move to IDLE.
324 * Writing only one dummy transaction can get appended to
325 * one file space allocation. When this happens, the log recovery
326 * code replays the space allocation and a file could be truncated.
327 * This is why we have the NEED2 and DONE2 states before going idle.
330 #define XLOG_STATE_COVER_IDLE 0
331 #define XLOG_STATE_COVER_NEED 1
332 #define XLOG_STATE_COVER_DONE 2
333 #define XLOG_STATE_COVER_NEED2 3
334 #define XLOG_STATE_COVER_DONE2 4
336 #define XLOG_COVER_OPS 5
338 typedef struct xlog_ticket {
339 sv_t t_sema; /* sleep on this semaphore :20 */
340 struct xlog_ticket *t_next; /* : 4 */
341 struct xlog_ticket *t_prev; /* : 4 */
342 xlog_tid_t t_tid; /* transaction identifier : 4 */
343 int t_curr_res; /* current reservation in bytes : 4 */
344 int t_unit_res; /* unit reservation in bytes : 4 */
345 __uint8_t t_ocnt; /* original count : 1 */
346 __uint8_t t_cnt; /* current count : 1 */
347 __uint8_t t_clientid; /* who does this belong to; : 1 */
348 __uint8_t t_flags; /* properties of reservation : 1 */
353 typedef struct xlog_op_header {
354 xlog_tid_t oh_tid; /* transaction id of operation : 4 b */
355 int oh_len; /* bytes in data region : 4 b */
356 __uint8_t oh_clientid; /* who sent me this : 1 b */
357 __uint8_t oh_flags; /* : 1 b */
358 ushort oh_res2; /* 32 bit align : 2 b */
362 /* valid values for h_fmt */
363 #define XLOG_FMT_UNKNOWN 0
364 #define XLOG_FMT_LINUX_LE 1
365 #define XLOG_FMT_LINUX_BE 2
366 #define XLOG_FMT_IRIX_BE 3
369 #if __BYTE_ORDER == __LITTLE_ENDIAN
370 #define XLOG_FMT XLOG_FMT_LINUX_LE
372 #if __BYTE_ORDER == __BIG_ENDIAN
373 #define XLOG_FMT XLOG_FMT_LINUX_BE
375 #error unknown byte order
379 typedef struct xlog_rec_header {
380 uint h_magicno; /* log record (LR) identifier : 4 */
381 uint h_cycle; /* write cycle of log : 4 */
382 int h_version; /* LR version : 4 */
383 int h_len; /* len in bytes; should be 64-bit aligned: 4 */
384 xfs_lsn_t h_lsn; /* lsn of this LR : 8 */
385 xfs_lsn_t h_tail_lsn; /* lsn of 1st LR w/ buffers not committed: 8 */
386 uint h_chksum; /* may not be used; non-zero if used : 4 */
387 int h_prev_block; /* block number to previous LR : 4 */
388 int h_num_logops; /* number of log operations in this LR : 4 */
389 uint h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE];
391 int h_fmt; /* format of log record : 4 */
392 uuid_t h_fs_uuid; /* uuid of FS : 16 */
393 int h_size; /* iclog size : 4 */
396 typedef struct xlog_rec_ext_header {
397 uint xh_cycle; /* write cycle of log : 4 */
398 uint xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /* : 256 */
399 } xlog_rec_ext_header_t;
403 * - A log record header is 512 bytes. There is plenty of room to grow the
404 * xlog_rec_header_t into the reserved space.
405 * - ic_data follows, so a write to disk can start at the beginning of
407 * - ic_forcesema is used to implement synchronous forcing of the iclog to disk.
408 * - ic_next is the pointer to the next iclog in the ring.
409 * - ic_bp is a pointer to the buffer used to write this incore log to disk.
410 * - ic_log is a pointer back to the global log structure.
411 * - ic_callback is a linked list of callback function/argument pairs to be
412 * called after an iclog finishes writing.
413 * - ic_size is the full size of the header plus data.
414 * - ic_offset is the current number of bytes written to in this iclog.
415 * - ic_refcnt is bumped when someone is writing to the log.
416 * - ic_state is the state of the iclog.
418 typedef struct xlog_iclog_fields {
421 struct xlog_in_core *ic_next;
422 struct xlog_in_core *ic_prev;
423 struct xfs_buf *ic_bp;
425 xfs_log_callback_t *ic_callback;
426 xfs_log_callback_t **ic_callback_tail;
428 struct ktrace *ic_trace;
435 char *ic_datap; /* pointer to iclog data */
436 } xlog_iclog_fields_t;
438 typedef union xlog_in_core2 {
439 xlog_rec_header_t hic_header;
440 xlog_rec_ext_header_t hic_xheader;
441 char hic_sector[XLOG_HEADER_SIZE];
444 typedef struct xlog_in_core {
445 xlog_iclog_fields_t hic_fields;
446 xlog_in_core_2_t *hic_data;
450 * Defines to save our code from this glop.
452 #define ic_forcesema hic_fields.ic_forcesema
453 #define ic_writesema hic_fields.ic_writesema
454 #define ic_next hic_fields.ic_next
455 #define ic_prev hic_fields.ic_prev
456 #define ic_bp hic_fields.ic_bp
457 #define ic_log hic_fields.ic_log
458 #define ic_callback hic_fields.ic_callback
459 #define ic_callback_tail hic_fields.ic_callback_tail
460 #define ic_trace hic_fields.ic_trace
461 #define ic_size hic_fields.ic_size
462 #define ic_offset hic_fields.ic_offset
463 #define ic_refcnt hic_fields.ic_refcnt
464 #define ic_bwritecnt hic_fields.ic_bwritecnt
465 #define ic_state hic_fields.ic_state
466 #define ic_datap hic_fields.ic_datap
467 #define ic_header hic_data->hic_header
470 * The reservation head lsn is not made up of a cycle number and block number.
471 * Instead, it uses a cycle number and byte number. Logs don't expect to
472 * overflow 31 bits worth of byte offset, so using a byte number will mean
473 * that round off problems won't occur when releasing partial reservations.
476 /* The following block of fields are changed while holding icloglock */
477 sema_t l_flushsema; /* iclog flushing semaphore */
478 int l_flushcnt; /* # of procs waiting on this
480 int l_ticket_cnt; /* free ticket count */
481 int l_ticket_tcnt; /* total ticket count */
482 int l_covered_state;/* state of "covering disk
484 xlog_ticket_t *l_freelist; /* free list of tickets */
485 xlog_ticket_t *l_unmount_free;/* kmem_free these addresses */
486 xlog_ticket_t *l_tail; /* free list of tickets */
487 xlog_in_core_t *l_iclog; /* head log queue */
488 lock_t l_icloglock; /* grab to change iclog state */
489 xfs_lsn_t l_tail_lsn; /* lsn of 1st LR with unflushed
491 xfs_lsn_t l_last_sync_lsn;/* lsn of last LR on disk */
492 struct xfs_mount *l_mp; /* mount point */
493 struct xfs_buf *l_xbuf; /* extra buffer for log
495 struct xfs_buftarg *l_targ; /* buftarg of log */
496 xfs_daddr_t l_logBBstart; /* start block of log */
497 int l_logsize; /* size of log in bytes */
498 int l_logBBsize; /* size of log in BB chunks */
499 int l_curr_cycle; /* Cycle number of log writes */
500 int l_prev_cycle; /* Cycle number before last
502 int l_curr_block; /* current logical log block */
503 int l_prev_block; /* previous logical log block */
504 int l_iclog_size; /* size of log in bytes */
505 int l_iclog_size_log; /* log power size of log */
506 int l_iclog_bufs; /* number of iclog buffers */
508 /* The following field are used for debugging; need to hold icloglock */
509 char *l_iclog_bak[XLOG_MAX_ICLOGS];
511 /* The following block of fields are changed while holding grant_lock */
513 xlog_ticket_t *l_reserve_headq;
514 xlog_ticket_t *l_write_headq;
515 int l_grant_reserve_cycle;
516 int l_grant_reserve_bytes;
517 int l_grant_write_cycle;
518 int l_grant_write_bytes;
520 /* The following fields don't need locking */
522 struct ktrace *l_trace;
523 struct ktrace *l_grant_trace;
526 uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
527 struct xfs_buf_cancel **l_buf_cancel_table;
528 int l_iclog_hsize; /* size of iclog header */
529 int l_iclog_heads; /* # of iclog header sectors */
530 uint l_sectbb_log; /* log2 of sector size in BBs */
531 uint l_sectbb_mask; /* sector size (in BBs)
536 /* common routines */
537 extern xfs_lsn_t xlog_assign_tail_lsn(struct xfs_mount *mp);
538 extern int xlog_find_tail(xlog_t *log,
539 xfs_daddr_t *head_blk,
540 xfs_daddr_t *tail_blk,
542 extern int xlog_recover(xlog_t *log, int readonly);
543 extern int xlog_recover_finish(xlog_t *log, int mfsi_flags);
544 extern void xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog, int);
545 extern void xlog_recover_process_iunlinks(xlog_t *log);
547 extern struct xfs_buf *xlog_get_bp(xlog_t *, int);
548 extern void xlog_put_bp(struct xfs_buf *);
549 extern int xlog_bread(xlog_t *, xfs_daddr_t, int, struct xfs_buf *);
552 #define XLOG_TRACE_GRAB_FLUSH 1
553 #define XLOG_TRACE_REL_FLUSH 2
554 #define XLOG_TRACE_SLEEP_FLUSH 3
555 #define XLOG_TRACE_WAKE_FLUSH 4
557 #endif /* __KERNEL__ */
559 #endif /* __XFS_LOG_PRIV_H__ */