2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * Copyright (C) 2010 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_format.h"
23 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_error.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_ialloc.h"
36 #include "xfs_alloc.h"
37 #include "xfs_extent_busy.h"
39 #include "xfs_quota.h"
41 #include "xfs_trans_priv.h"
42 #include "xfs_trans_space.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_log_priv.h"
45 #include "xfs_buf_item.h"
46 #include "xfs_trace.h"
48 kmem_zone_t
*xfs_trans_zone
;
49 kmem_zone_t
*xfs_log_item_desc_zone
;
52 * Initialize the precomputed transaction reservation values
53 * in the mount structure.
59 xfs_trans_resv_calc(mp
, &mp
->m_resv
);
63 * This routine is called to allocate a transaction structure.
64 * The type parameter indicates the type of the transaction. These
65 * are enumerated in xfs_trans.h.
67 * Dynamically allocate the transaction structure from the transaction
68 * zone, initialize it, and return it to the caller.
77 sb_start_intwrite(mp
->m_super
);
78 tp
= _xfs_trans_alloc(mp
, type
, KM_SLEEP
);
79 tp
->t_flags
|= XFS_TRANS_FREEZE_PROT
;
87 xfs_km_flags_t memflags
)
91 WARN_ON(mp
->m_super
->s_writers
.frozen
== SB_FREEZE_COMPLETE
);
92 atomic_inc(&mp
->m_active_trans
);
94 tp
= kmem_zone_zalloc(xfs_trans_zone
, memflags
);
95 tp
->t_magic
= XFS_TRANS_HEADER_MAGIC
;
98 INIT_LIST_HEAD(&tp
->t_items
);
99 INIT_LIST_HEAD(&tp
->t_busy
);
104 * Free the transaction structure. If there is more clean up
105 * to do when the structure is freed, add it here.
109 struct xfs_trans
*tp
)
111 xfs_extent_busy_sort(&tp
->t_busy
);
112 xfs_extent_busy_clear(tp
->t_mountp
, &tp
->t_busy
, false);
114 atomic_dec(&tp
->t_mountp
->m_active_trans
);
115 if (tp
->t_flags
& XFS_TRANS_FREEZE_PROT
)
116 sb_end_intwrite(tp
->t_mountp
->m_super
);
117 xfs_trans_free_dqinfo(tp
);
118 kmem_zone_free(xfs_trans_zone
, tp
);
122 * This is called to create a new transaction which will share the
123 * permanent log reservation of the given transaction. The remaining
124 * unused block and rt extent reservations are also inherited. This
125 * implies that the original transaction is no longer allowed to allocate
126 * blocks. Locks and log items, however, are no inherited. They must
127 * be added to the new transaction explicitly.
135 ntp
= kmem_zone_zalloc(xfs_trans_zone
, KM_SLEEP
);
138 * Initialize the new transaction structure.
140 ntp
->t_magic
= XFS_TRANS_HEADER_MAGIC
;
141 ntp
->t_type
= tp
->t_type
;
142 ntp
->t_mountp
= tp
->t_mountp
;
143 INIT_LIST_HEAD(&ntp
->t_items
);
144 INIT_LIST_HEAD(&ntp
->t_busy
);
146 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
147 ASSERT(tp
->t_ticket
!= NULL
);
149 ntp
->t_flags
= XFS_TRANS_PERM_LOG_RES
|
150 (tp
->t_flags
& XFS_TRANS_RESERVE
) |
151 (tp
->t_flags
& XFS_TRANS_FREEZE_PROT
);
152 /* We gave our writer reference to the new transaction */
153 tp
->t_flags
&= ~XFS_TRANS_FREEZE_PROT
;
154 ntp
->t_ticket
= xfs_log_ticket_get(tp
->t_ticket
);
155 ntp
->t_blk_res
= tp
->t_blk_res
- tp
->t_blk_res_used
;
156 tp
->t_blk_res
= tp
->t_blk_res_used
;
157 ntp
->t_rtx_res
= tp
->t_rtx_res
- tp
->t_rtx_res_used
;
158 tp
->t_rtx_res
= tp
->t_rtx_res_used
;
159 ntp
->t_pflags
= tp
->t_pflags
;
161 xfs_trans_dup_dqinfo(tp
, ntp
);
163 atomic_inc(&tp
->t_mountp
->m_active_trans
);
168 * This is called to reserve free disk blocks and log space for the
169 * given transaction. This must be done before allocating any resources
170 * within the transaction.
172 * This will return ENOSPC if there are not enough blocks available.
173 * It will sleep waiting for available log space.
174 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
175 * is used by long running transactions. If any one of the reservations
176 * fails then they will all be backed out.
178 * This does not do quota reservations. That typically is done by the
191 int rsvd
= (tp
->t_flags
& XFS_TRANS_RESERVE
) != 0;
193 /* Mark this thread as being in a transaction */
194 current_set_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
197 * Attempt to reserve the needed disk blocks by decrementing
198 * the number needed from the number available. This will
199 * fail if the count would go below zero.
202 error
= xfs_icsb_modify_counters(tp
->t_mountp
, XFS_SBS_FDBLOCKS
,
203 -((int64_t)blocks
), rsvd
);
205 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
206 return (XFS_ERROR(ENOSPC
));
208 tp
->t_blk_res
+= blocks
;
212 * Reserve the log space needed for this transaction.
215 bool permanent
= false;
217 ASSERT(tp
->t_log_res
== 0 || tp
->t_log_res
== logspace
);
218 ASSERT(tp
->t_log_count
== 0 || tp
->t_log_count
== logcount
);
220 if (flags
& XFS_TRANS_PERM_LOG_RES
) {
221 tp
->t_flags
|= XFS_TRANS_PERM_LOG_RES
;
224 ASSERT(tp
->t_ticket
== NULL
);
225 ASSERT(!(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
));
228 if (tp
->t_ticket
!= NULL
) {
229 ASSERT(flags
& XFS_TRANS_PERM_LOG_RES
);
230 error
= xfs_log_regrant(tp
->t_mountp
, tp
->t_ticket
);
232 error
= xfs_log_reserve(tp
->t_mountp
, logspace
,
233 logcount
, &tp
->t_ticket
,
234 XFS_TRANSACTION
, permanent
,
241 tp
->t_log_res
= logspace
;
242 tp
->t_log_count
= logcount
;
246 * Attempt to reserve the needed realtime extents by decrementing
247 * the number needed from the number available. This will
248 * fail if the count would go below zero.
251 error
= xfs_mod_incore_sb(tp
->t_mountp
, XFS_SBS_FREXTENTS
,
252 -((int64_t)rtextents
), rsvd
);
254 error
= XFS_ERROR(ENOSPC
);
257 tp
->t_rtx_res
+= rtextents
;
263 * Error cases jump to one of these labels to undo any
264 * reservations which have already been performed.
270 if (flags
& XFS_TRANS_PERM_LOG_RES
) {
271 log_flags
= XFS_LOG_REL_PERM_RESERV
;
275 xfs_log_done(tp
->t_mountp
, tp
->t_ticket
, NULL
, log_flags
);
278 tp
->t_flags
&= ~XFS_TRANS_PERM_LOG_RES
;
283 xfs_icsb_modify_counters(tp
->t_mountp
, XFS_SBS_FDBLOCKS
,
284 (int64_t)blocks
, rsvd
);
288 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
294 * Record the indicated change to the given field for application
295 * to the file system's superblock when the transaction commits.
296 * For now, just store the change in the transaction structure.
298 * Mark the transaction structure to indicate that the superblock
299 * needs to be updated before committing.
301 * Because we may not be keeping track of allocated/free inodes and
302 * used filesystem blocks in the superblock, we do not mark the
303 * superblock dirty in this transaction if we modify these fields.
304 * We still need to update the transaction deltas so that they get
305 * applied to the incore superblock, but we don't want them to
306 * cause the superblock to get locked and logged if these are the
307 * only fields in the superblock that the transaction modifies.
315 uint32_t flags
= (XFS_TRANS_DIRTY
|XFS_TRANS_SB_DIRTY
);
316 xfs_mount_t
*mp
= tp
->t_mountp
;
319 case XFS_TRANS_SB_ICOUNT
:
320 tp
->t_icount_delta
+= delta
;
321 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
322 flags
&= ~XFS_TRANS_SB_DIRTY
;
324 case XFS_TRANS_SB_IFREE
:
325 tp
->t_ifree_delta
+= delta
;
326 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
327 flags
&= ~XFS_TRANS_SB_DIRTY
;
329 case XFS_TRANS_SB_FDBLOCKS
:
331 * Track the number of blocks allocated in the
332 * transaction. Make sure it does not exceed the
336 tp
->t_blk_res_used
+= (uint
)-delta
;
337 ASSERT(tp
->t_blk_res_used
<= tp
->t_blk_res
);
339 tp
->t_fdblocks_delta
+= delta
;
340 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
341 flags
&= ~XFS_TRANS_SB_DIRTY
;
343 case XFS_TRANS_SB_RES_FDBLOCKS
:
345 * The allocation has already been applied to the
346 * in-core superblock's counter. This should only
347 * be applied to the on-disk superblock.
350 tp
->t_res_fdblocks_delta
+= delta
;
351 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
352 flags
&= ~XFS_TRANS_SB_DIRTY
;
354 case XFS_TRANS_SB_FREXTENTS
:
356 * Track the number of blocks allocated in the
357 * transaction. Make sure it does not exceed the
361 tp
->t_rtx_res_used
+= (uint
)-delta
;
362 ASSERT(tp
->t_rtx_res_used
<= tp
->t_rtx_res
);
364 tp
->t_frextents_delta
+= delta
;
366 case XFS_TRANS_SB_RES_FREXTENTS
:
368 * The allocation has already been applied to the
369 * in-core superblock's counter. This should only
370 * be applied to the on-disk superblock.
373 tp
->t_res_frextents_delta
+= delta
;
375 case XFS_TRANS_SB_DBLOCKS
:
377 tp
->t_dblocks_delta
+= delta
;
379 case XFS_TRANS_SB_AGCOUNT
:
381 tp
->t_agcount_delta
+= delta
;
383 case XFS_TRANS_SB_IMAXPCT
:
384 tp
->t_imaxpct_delta
+= delta
;
386 case XFS_TRANS_SB_REXTSIZE
:
387 tp
->t_rextsize_delta
+= delta
;
389 case XFS_TRANS_SB_RBMBLOCKS
:
390 tp
->t_rbmblocks_delta
+= delta
;
392 case XFS_TRANS_SB_RBLOCKS
:
393 tp
->t_rblocks_delta
+= delta
;
395 case XFS_TRANS_SB_REXTENTS
:
396 tp
->t_rextents_delta
+= delta
;
398 case XFS_TRANS_SB_REXTSLOG
:
399 tp
->t_rextslog_delta
+= delta
;
406 tp
->t_flags
|= flags
;
410 * xfs_trans_apply_sb_deltas() is called from the commit code
411 * to bring the superblock buffer into the current transaction
412 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
414 * For now we just look at each field allowed to change and change
418 xfs_trans_apply_sb_deltas(
425 bp
= xfs_trans_getsb(tp
, tp
->t_mountp
, 0);
426 sbp
= XFS_BUF_TO_SBP(bp
);
429 * Check that superblock mods match the mods made to AGF counters.
431 ASSERT((tp
->t_fdblocks_delta
+ tp
->t_res_fdblocks_delta
) ==
432 (tp
->t_ag_freeblks_delta
+ tp
->t_ag_flist_delta
+
433 tp
->t_ag_btree_delta
));
436 * Only update the superblock counters if we are logging them
438 if (!xfs_sb_version_haslazysbcount(&(tp
->t_mountp
->m_sb
))) {
439 if (tp
->t_icount_delta
)
440 be64_add_cpu(&sbp
->sb_icount
, tp
->t_icount_delta
);
441 if (tp
->t_ifree_delta
)
442 be64_add_cpu(&sbp
->sb_ifree
, tp
->t_ifree_delta
);
443 if (tp
->t_fdblocks_delta
)
444 be64_add_cpu(&sbp
->sb_fdblocks
, tp
->t_fdblocks_delta
);
445 if (tp
->t_res_fdblocks_delta
)
446 be64_add_cpu(&sbp
->sb_fdblocks
, tp
->t_res_fdblocks_delta
);
449 if (tp
->t_frextents_delta
)
450 be64_add_cpu(&sbp
->sb_frextents
, tp
->t_frextents_delta
);
451 if (tp
->t_res_frextents_delta
)
452 be64_add_cpu(&sbp
->sb_frextents
, tp
->t_res_frextents_delta
);
454 if (tp
->t_dblocks_delta
) {
455 be64_add_cpu(&sbp
->sb_dblocks
, tp
->t_dblocks_delta
);
458 if (tp
->t_agcount_delta
) {
459 be32_add_cpu(&sbp
->sb_agcount
, tp
->t_agcount_delta
);
462 if (tp
->t_imaxpct_delta
) {
463 sbp
->sb_imax_pct
+= tp
->t_imaxpct_delta
;
466 if (tp
->t_rextsize_delta
) {
467 be32_add_cpu(&sbp
->sb_rextsize
, tp
->t_rextsize_delta
);
470 if (tp
->t_rbmblocks_delta
) {
471 be32_add_cpu(&sbp
->sb_rbmblocks
, tp
->t_rbmblocks_delta
);
474 if (tp
->t_rblocks_delta
) {
475 be64_add_cpu(&sbp
->sb_rblocks
, tp
->t_rblocks_delta
);
478 if (tp
->t_rextents_delta
) {
479 be64_add_cpu(&sbp
->sb_rextents
, tp
->t_rextents_delta
);
482 if (tp
->t_rextslog_delta
) {
483 sbp
->sb_rextslog
+= tp
->t_rextslog_delta
;
489 * Log the whole thing, the fields are noncontiguous.
491 xfs_trans_log_buf(tp
, bp
, 0, sizeof(xfs_dsb_t
) - 1);
494 * Since all the modifiable fields are contiguous, we
495 * can get away with this.
497 xfs_trans_log_buf(tp
, bp
, offsetof(xfs_dsb_t
, sb_icount
),
498 offsetof(xfs_dsb_t
, sb_frextents
) +
499 sizeof(sbp
->sb_frextents
) - 1);
503 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
504 * and apply superblock counter changes to the in-core superblock. The
505 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
506 * applied to the in-core superblock. The idea is that that has already been
509 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
510 * However, we have to ensure that we only modify each superblock field only
511 * once because the application of the delta values may not be atomic. That can
512 * lead to ENOSPC races occurring if we have two separate modifcations of the
513 * free space counter to put back the entire reservation and then take away
516 * If we are not logging superblock counters, then the inode allocated/free and
517 * used block counts are not updated in the on disk superblock. In this case,
518 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
519 * still need to update the incore superblock with the changes.
522 xfs_trans_unreserve_and_mod_sb(
525 xfs_mod_sb_t msb
[9]; /* If you add cases, add entries */
527 xfs_mount_t
*mp
= tp
->t_mountp
;
531 int64_t blkdelta
= 0;
532 int64_t rtxdelta
= 0;
534 int64_t ifreedelta
= 0;
537 rsvd
= (tp
->t_flags
& XFS_TRANS_RESERVE
) != 0;
539 /* calculate deltas */
540 if (tp
->t_blk_res
> 0)
541 blkdelta
= tp
->t_blk_res
;
542 if ((tp
->t_fdblocks_delta
!= 0) &&
543 (xfs_sb_version_haslazysbcount(&mp
->m_sb
) ||
544 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)))
545 blkdelta
+= tp
->t_fdblocks_delta
;
547 if (tp
->t_rtx_res
> 0)
548 rtxdelta
= tp
->t_rtx_res
;
549 if ((tp
->t_frextents_delta
!= 0) &&
550 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
))
551 rtxdelta
+= tp
->t_frextents_delta
;
553 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) ||
554 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)) {
555 idelta
= tp
->t_icount_delta
;
556 ifreedelta
= tp
->t_ifree_delta
;
559 /* apply the per-cpu counters */
561 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
568 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_ICOUNT
,
571 goto out_undo_fdblocks
;
575 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_IFREE
,
578 goto out_undo_icount
;
581 /* apply remaining deltas */
583 msbp
->msb_field
= XFS_SBS_FREXTENTS
;
584 msbp
->msb_delta
= rtxdelta
;
588 if (tp
->t_flags
& XFS_TRANS_SB_DIRTY
) {
589 if (tp
->t_dblocks_delta
!= 0) {
590 msbp
->msb_field
= XFS_SBS_DBLOCKS
;
591 msbp
->msb_delta
= tp
->t_dblocks_delta
;
594 if (tp
->t_agcount_delta
!= 0) {
595 msbp
->msb_field
= XFS_SBS_AGCOUNT
;
596 msbp
->msb_delta
= tp
->t_agcount_delta
;
599 if (tp
->t_imaxpct_delta
!= 0) {
600 msbp
->msb_field
= XFS_SBS_IMAX_PCT
;
601 msbp
->msb_delta
= tp
->t_imaxpct_delta
;
604 if (tp
->t_rextsize_delta
!= 0) {
605 msbp
->msb_field
= XFS_SBS_REXTSIZE
;
606 msbp
->msb_delta
= tp
->t_rextsize_delta
;
609 if (tp
->t_rbmblocks_delta
!= 0) {
610 msbp
->msb_field
= XFS_SBS_RBMBLOCKS
;
611 msbp
->msb_delta
= tp
->t_rbmblocks_delta
;
614 if (tp
->t_rblocks_delta
!= 0) {
615 msbp
->msb_field
= XFS_SBS_RBLOCKS
;
616 msbp
->msb_delta
= tp
->t_rblocks_delta
;
619 if (tp
->t_rextents_delta
!= 0) {
620 msbp
->msb_field
= XFS_SBS_REXTENTS
;
621 msbp
->msb_delta
= tp
->t_rextents_delta
;
624 if (tp
->t_rextslog_delta
!= 0) {
625 msbp
->msb_field
= XFS_SBS_REXTSLOG
;
626 msbp
->msb_delta
= tp
->t_rextslog_delta
;
632 * If we need to change anything, do it.
635 error
= xfs_mod_incore_sb_batch(tp
->t_mountp
, msb
,
636 (uint
)(msbp
- msb
), rsvd
);
638 goto out_undo_ifreecount
;
645 xfs_icsb_modify_counters(mp
, XFS_SBS_IFREE
, -ifreedelta
, rsvd
);
648 xfs_icsb_modify_counters(mp
, XFS_SBS_ICOUNT
, -idelta
, rsvd
);
651 xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
, -blkdelta
, rsvd
);
658 * Add the given log item to the transaction's list of log items.
660 * The log item will now point to its new descriptor with its li_desc field.
664 struct xfs_trans
*tp
,
665 struct xfs_log_item
*lip
)
667 struct xfs_log_item_desc
*lidp
;
669 ASSERT(lip
->li_mountp
== tp
->t_mountp
);
670 ASSERT(lip
->li_ailp
== tp
->t_mountp
->m_ail
);
672 lidp
= kmem_zone_zalloc(xfs_log_item_desc_zone
, KM_SLEEP
| KM_NOFS
);
674 lidp
->lid_item
= lip
;
676 list_add_tail(&lidp
->lid_trans
, &tp
->t_items
);
682 xfs_trans_free_item_desc(
683 struct xfs_log_item_desc
*lidp
)
685 list_del_init(&lidp
->lid_trans
);
686 kmem_zone_free(xfs_log_item_desc_zone
, lidp
);
690 * Unlink and free the given descriptor.
694 struct xfs_log_item
*lip
)
696 xfs_trans_free_item_desc(lip
->li_desc
);
701 * Unlock all of the items of a transaction and free all the descriptors
702 * of that transaction.
705 xfs_trans_free_items(
706 struct xfs_trans
*tp
,
707 xfs_lsn_t commit_lsn
,
710 struct xfs_log_item_desc
*lidp
, *next
;
712 list_for_each_entry_safe(lidp
, next
, &tp
->t_items
, lid_trans
) {
713 struct xfs_log_item
*lip
= lidp
->lid_item
;
717 if (commit_lsn
!= NULLCOMMITLSN
)
718 IOP_COMMITTING(lip
, commit_lsn
);
719 if (flags
& XFS_TRANS_ABORT
)
720 lip
->li_flags
|= XFS_LI_ABORTED
;
723 xfs_trans_free_item_desc(lidp
);
728 xfs_log_item_batch_insert(
729 struct xfs_ail
*ailp
,
730 struct xfs_ail_cursor
*cur
,
731 struct xfs_log_item
**log_items
,
733 xfs_lsn_t commit_lsn
)
737 spin_lock(&ailp
->xa_lock
);
738 /* xfs_trans_ail_update_bulk drops ailp->xa_lock */
739 xfs_trans_ail_update_bulk(ailp
, cur
, log_items
, nr_items
, commit_lsn
);
741 for (i
= 0; i
< nr_items
; i
++)
742 IOP_UNPIN(log_items
[i
], 0);
746 * Bulk operation version of xfs_trans_committed that takes a log vector of
747 * items to insert into the AIL. This uses bulk AIL insertion techniques to
748 * minimise lock traffic.
750 * If we are called with the aborted flag set, it is because a log write during
751 * a CIL checkpoint commit has failed. In this case, all the items in the
752 * checkpoint have already gone through IOP_COMMITED and IOP_UNLOCK, which
753 * means that checkpoint commit abort handling is treated exactly the same
754 * as an iclog write error even though we haven't started any IO yet. Hence in
755 * this case all we need to do is IOP_COMMITTED processing, followed by an
756 * IOP_UNPIN(aborted) call.
758 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
759 * at the end of the AIL, the insert cursor avoids the need to walk
760 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
761 * call. This saves a lot of needless list walking and is a net win, even
762 * though it slightly increases that amount of AIL lock traffic to set it up
766 xfs_trans_committed_bulk(
767 struct xfs_ail
*ailp
,
768 struct xfs_log_vec
*log_vector
,
769 xfs_lsn_t commit_lsn
,
772 #define LOG_ITEM_BATCH_SIZE 32
773 struct xfs_log_item
*log_items
[LOG_ITEM_BATCH_SIZE
];
774 struct xfs_log_vec
*lv
;
775 struct xfs_ail_cursor cur
;
778 spin_lock(&ailp
->xa_lock
);
779 xfs_trans_ail_cursor_last(ailp
, &cur
, commit_lsn
);
780 spin_unlock(&ailp
->xa_lock
);
782 /* unpin all the log items */
783 for (lv
= log_vector
; lv
; lv
= lv
->lv_next
) {
784 struct xfs_log_item
*lip
= lv
->lv_item
;
788 lip
->li_flags
|= XFS_LI_ABORTED
;
789 item_lsn
= IOP_COMMITTED(lip
, commit_lsn
);
791 /* item_lsn of -1 means the item needs no further processing */
792 if (XFS_LSN_CMP(item_lsn
, (xfs_lsn_t
)-1) == 0)
796 * if we are aborting the operation, no point in inserting the
797 * object into the AIL as we are in a shutdown situation.
800 ASSERT(XFS_FORCED_SHUTDOWN(ailp
->xa_mount
));
805 if (item_lsn
!= commit_lsn
) {
808 * Not a bulk update option due to unusual item_lsn.
809 * Push into AIL immediately, rechecking the lsn once
810 * we have the ail lock. Then unpin the item. This does
811 * not affect the AIL cursor the bulk insert path is
814 spin_lock(&ailp
->xa_lock
);
815 if (XFS_LSN_CMP(item_lsn
, lip
->li_lsn
) > 0)
816 xfs_trans_ail_update(ailp
, lip
, item_lsn
);
818 spin_unlock(&ailp
->xa_lock
);
823 /* Item is a candidate for bulk AIL insert. */
824 log_items
[i
++] = lv
->lv_item
;
825 if (i
>= LOG_ITEM_BATCH_SIZE
) {
826 xfs_log_item_batch_insert(ailp
, &cur
, log_items
,
827 LOG_ITEM_BATCH_SIZE
, commit_lsn
);
832 /* make sure we insert the remainder! */
834 xfs_log_item_batch_insert(ailp
, &cur
, log_items
, i
, commit_lsn
);
836 spin_lock(&ailp
->xa_lock
);
837 xfs_trans_ail_cursor_done(ailp
, &cur
);
838 spin_unlock(&ailp
->xa_lock
);
842 * Commit the given transaction to the log.
844 * XFS disk error handling mechanism is not based on a typical
845 * transaction abort mechanism. Logically after the filesystem
846 * gets marked 'SHUTDOWN', we can't let any new transactions
847 * be durable - ie. committed to disk - because some metadata might
848 * be inconsistent. In such cases, this returns an error, and the
849 * caller may assume that all locked objects joined to the transaction
850 * have already been unlocked as if the commit had succeeded.
851 * Do not reference the transaction structure after this call.
855 struct xfs_trans
*tp
,
858 struct xfs_mount
*mp
= tp
->t_mountp
;
859 xfs_lsn_t commit_lsn
= -1;
862 int sync
= tp
->t_flags
& XFS_TRANS_SYNC
;
865 * Determine whether this commit is releasing a permanent
866 * log reservation or not.
868 if (flags
& XFS_TRANS_RELEASE_LOG_RES
) {
869 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
870 log_flags
= XFS_LOG_REL_PERM_RESERV
;
874 * If there is nothing to be logged by the transaction,
875 * then unlock all of the items associated with the
876 * transaction and free the transaction structure.
877 * Also make sure to return any reserved blocks to
880 if (!(tp
->t_flags
& XFS_TRANS_DIRTY
))
883 if (XFS_FORCED_SHUTDOWN(mp
)) {
884 error
= XFS_ERROR(EIO
);
888 ASSERT(tp
->t_ticket
!= NULL
);
891 * If we need to update the superblock, then do it now.
893 if (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)
894 xfs_trans_apply_sb_deltas(tp
);
895 xfs_trans_apply_dquot_deltas(tp
);
897 error
= xfs_log_commit_cil(mp
, tp
, &commit_lsn
, flags
);
898 if (error
== ENOMEM
) {
899 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
900 error
= XFS_ERROR(EIO
);
904 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
908 * If the transaction needs to be synchronous, then force the
909 * log out now and wait for it.
913 error
= _xfs_log_force_lsn(mp
, commit_lsn
,
916 XFS_STATS_INC(xs_trans_sync
);
918 XFS_STATS_INC(xs_trans_async
);
924 xfs_trans_unreserve_and_mod_sb(tp
);
927 * It is indeed possible for the transaction to be not dirty but
928 * the dqinfo portion to be. All that means is that we have some
929 * (non-persistent) quota reservations that need to be unreserved.
931 xfs_trans_unreserve_and_mod_dquots(tp
);
933 commit_lsn
= xfs_log_done(mp
, tp
->t_ticket
, NULL
, log_flags
);
934 if (commit_lsn
== -1 && !error
)
935 error
= XFS_ERROR(EIO
);
937 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
938 xfs_trans_free_items(tp
, NULLCOMMITLSN
, error
? XFS_TRANS_ABORT
: 0);
941 XFS_STATS_INC(xs_trans_empty
);
946 * Unlock all of the transaction's items and free the transaction.
947 * The transaction must not have modified any of its items, because
948 * there is no way to restore them to their previous state.
950 * If the transaction has made a log reservation, make sure to release
959 xfs_mount_t
*mp
= tp
->t_mountp
;
962 * See if the caller is being too lazy to figure out if
963 * the transaction really needs an abort.
965 if ((flags
& XFS_TRANS_ABORT
) && !(tp
->t_flags
& XFS_TRANS_DIRTY
))
966 flags
&= ~XFS_TRANS_ABORT
;
968 * See if the caller is relying on us to shut down the
969 * filesystem. This happens in paths where we detect
970 * corruption and decide to give up.
972 if ((tp
->t_flags
& XFS_TRANS_DIRTY
) && !XFS_FORCED_SHUTDOWN(mp
)) {
973 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW
, mp
);
974 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
977 if (!(flags
& XFS_TRANS_ABORT
) && !XFS_FORCED_SHUTDOWN(mp
)) {
978 struct xfs_log_item_desc
*lidp
;
980 list_for_each_entry(lidp
, &tp
->t_items
, lid_trans
)
981 ASSERT(!(lidp
->lid_item
->li_type
== XFS_LI_EFD
));
984 xfs_trans_unreserve_and_mod_sb(tp
);
985 xfs_trans_unreserve_and_mod_dquots(tp
);
988 if (flags
& XFS_TRANS_RELEASE_LOG_RES
) {
989 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
990 log_flags
= XFS_LOG_REL_PERM_RESERV
;
994 xfs_log_done(mp
, tp
->t_ticket
, NULL
, log_flags
);
997 /* mark this thread as no longer being in a transaction */
998 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
1000 xfs_trans_free_items(tp
, NULLCOMMITLSN
, flags
);
1005 * Roll from one trans in the sequence of PERMANENT transactions to
1006 * the next: permanent transactions are only flushed out when
1007 * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
1008 * as possible to let chunks of it go to the log. So we commit the
1009 * chunk we've been working on and get a new transaction to continue.
1013 struct xfs_trans
**tpp
,
1014 struct xfs_inode
*dp
)
1016 struct xfs_trans
*trans
;
1017 unsigned int logres
, count
;
1021 * Ensure that the inode is always logged.
1024 xfs_trans_log_inode(trans
, dp
, XFS_ILOG_CORE
);
1027 * Copy the critical parameters from one trans to the next.
1029 logres
= trans
->t_log_res
;
1030 count
= trans
->t_log_count
;
1031 *tpp
= xfs_trans_dup(trans
);
1034 * Commit the current transaction.
1035 * If this commit failed, then it'd just unlock those items that
1036 * are not marked ihold. That also means that a filesystem shutdown
1037 * is in progress. The caller takes the responsibility to cancel
1038 * the duplicate transaction that gets returned.
1040 error
= xfs_trans_commit(trans
, 0);
1047 * transaction commit worked ok so we can drop the extra ticket
1048 * reference that we gained in xfs_trans_dup()
1050 xfs_log_ticket_put(trans
->t_ticket
);
1054 * Reserve space in the log for th next transaction.
1055 * This also pushes items in the "AIL", the list of logged items,
1056 * out to disk if they are taking up space at the tail of the log
1057 * that we want to use. This requires that either nothing be locked
1058 * across this call, or that anything that is locked be logged in
1059 * the prior and the next transactions.
1061 error
= xfs_trans_reserve(trans
, 0, logres
, 0,
1062 XFS_TRANS_PERM_LOG_RES
, count
);
1064 * Ensure that the inode is in the new transaction and locked.
1069 xfs_trans_ijoin(trans
, dp
, 0);