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_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
27 #include "xfs_mount.h"
28 #include "xfs_error.h"
29 #include "xfs_da_format.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_btree.h"
36 #include "xfs_ialloc.h"
37 #include "xfs_alloc.h"
38 #include "xfs_extent_busy.h"
40 #include "xfs_quota.h"
41 #include "xfs_trans.h"
42 #include "xfs_trans_priv.h"
43 #include "xfs_trans_space.h"
45 #include "xfs_inode_item.h"
47 #include "xfs_log_priv.h"
48 #include "xfs_buf_item.h"
49 #include "xfs_trace.h"
51 kmem_zone_t
*xfs_trans_zone
;
52 kmem_zone_t
*xfs_log_item_desc_zone
;
55 * Initialize the precomputed transaction reservation values
56 * in the mount structure.
62 xfs_trans_resv_calc(mp
, M_RES(mp
));
66 * This routine is called to allocate a transaction structure.
67 * The type parameter indicates the type of the transaction. These
68 * are enumerated in xfs_trans.h.
70 * Dynamically allocate the transaction structure from the transaction
71 * zone, initialize it, and return it to the caller.
80 sb_start_intwrite(mp
->m_super
);
81 tp
= _xfs_trans_alloc(mp
, type
, KM_SLEEP
);
82 tp
->t_flags
|= XFS_TRANS_FREEZE_PROT
;
90 xfs_km_flags_t memflags
)
94 WARN_ON(mp
->m_super
->s_writers
.frozen
== SB_FREEZE_COMPLETE
);
95 atomic_inc(&mp
->m_active_trans
);
97 tp
= kmem_zone_zalloc(xfs_trans_zone
, memflags
);
98 tp
->t_magic
= XFS_TRANS_HEADER_MAGIC
;
101 INIT_LIST_HEAD(&tp
->t_items
);
102 INIT_LIST_HEAD(&tp
->t_busy
);
107 * Free the transaction structure. If there is more clean up
108 * to do when the structure is freed, add it here.
112 struct xfs_trans
*tp
)
114 xfs_extent_busy_sort(&tp
->t_busy
);
115 xfs_extent_busy_clear(tp
->t_mountp
, &tp
->t_busy
, false);
117 atomic_dec(&tp
->t_mountp
->m_active_trans
);
118 if (tp
->t_flags
& XFS_TRANS_FREEZE_PROT
)
119 sb_end_intwrite(tp
->t_mountp
->m_super
);
120 xfs_trans_free_dqinfo(tp
);
121 kmem_zone_free(xfs_trans_zone
, tp
);
125 * This is called to create a new transaction which will share the
126 * permanent log reservation of the given transaction. The remaining
127 * unused block and rt extent reservations are also inherited. This
128 * implies that the original transaction is no longer allowed to allocate
129 * blocks. Locks and log items, however, are no inherited. They must
130 * be added to the new transaction explicitly.
138 ntp
= kmem_zone_zalloc(xfs_trans_zone
, KM_SLEEP
);
141 * Initialize the new transaction structure.
143 ntp
->t_magic
= XFS_TRANS_HEADER_MAGIC
;
144 ntp
->t_type
= tp
->t_type
;
145 ntp
->t_mountp
= tp
->t_mountp
;
146 INIT_LIST_HEAD(&ntp
->t_items
);
147 INIT_LIST_HEAD(&ntp
->t_busy
);
149 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
150 ASSERT(tp
->t_ticket
!= NULL
);
152 ntp
->t_flags
= XFS_TRANS_PERM_LOG_RES
|
153 (tp
->t_flags
& XFS_TRANS_RESERVE
) |
154 (tp
->t_flags
& XFS_TRANS_FREEZE_PROT
);
155 /* We gave our writer reference to the new transaction */
156 tp
->t_flags
&= ~XFS_TRANS_FREEZE_PROT
;
157 ntp
->t_ticket
= xfs_log_ticket_get(tp
->t_ticket
);
158 ntp
->t_blk_res
= tp
->t_blk_res
- tp
->t_blk_res_used
;
159 tp
->t_blk_res
= tp
->t_blk_res_used
;
160 ntp
->t_rtx_res
= tp
->t_rtx_res
- tp
->t_rtx_res_used
;
161 tp
->t_rtx_res
= tp
->t_rtx_res_used
;
162 ntp
->t_pflags
= tp
->t_pflags
;
164 xfs_trans_dup_dqinfo(tp
, ntp
);
166 atomic_inc(&tp
->t_mountp
->m_active_trans
);
171 * This is called to reserve free disk blocks and log space for the
172 * given transaction. This must be done before allocating any resources
173 * within the transaction.
175 * This will return ENOSPC if there are not enough blocks available.
176 * It will sleep waiting for available log space.
177 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
178 * is used by long running transactions. If any one of the reservations
179 * fails then they will all be backed out.
181 * This does not do quota reservations. That typically is done by the
186 struct xfs_trans
*tp
,
187 struct xfs_trans_res
*resp
,
192 int rsvd
= (tp
->t_flags
& XFS_TRANS_RESERVE
) != 0;
194 /* Mark this thread as being in a transaction */
195 current_set_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
198 * Attempt to reserve the needed disk blocks by decrementing
199 * the number needed from the number available. This will
200 * fail if the count would go below zero.
203 error
= xfs_icsb_modify_counters(tp
->t_mountp
, XFS_SBS_FDBLOCKS
,
204 -((int64_t)blocks
), rsvd
);
206 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
207 return (XFS_ERROR(ENOSPC
));
209 tp
->t_blk_res
+= blocks
;
213 * Reserve the log space needed for this transaction.
215 if (resp
->tr_logres
> 0) {
216 bool permanent
= false;
218 ASSERT(tp
->t_log_res
== 0 ||
219 tp
->t_log_res
== resp
->tr_logres
);
220 ASSERT(tp
->t_log_count
== 0 ||
221 tp
->t_log_count
== resp
->tr_logcount
);
223 if (resp
->tr_logflags
& XFS_TRANS_PERM_LOG_RES
) {
224 tp
->t_flags
|= XFS_TRANS_PERM_LOG_RES
;
227 ASSERT(tp
->t_ticket
== NULL
);
228 ASSERT(!(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
));
231 if (tp
->t_ticket
!= NULL
) {
232 ASSERT(resp
->tr_logflags
& XFS_TRANS_PERM_LOG_RES
);
233 error
= xfs_log_regrant(tp
->t_mountp
, tp
->t_ticket
);
235 error
= xfs_log_reserve(tp
->t_mountp
,
238 &tp
->t_ticket
, XFS_TRANSACTION
,
239 permanent
, tp
->t_type
);
245 tp
->t_log_res
= resp
->tr_logres
;
246 tp
->t_log_count
= resp
->tr_logcount
;
250 * Attempt to reserve the needed realtime extents by decrementing
251 * the number needed from the number available. This will
252 * fail if the count would go below zero.
255 error
= xfs_mod_incore_sb(tp
->t_mountp
, XFS_SBS_FREXTENTS
,
256 -((int64_t)rtextents
), rsvd
);
258 error
= XFS_ERROR(ENOSPC
);
261 tp
->t_rtx_res
+= rtextents
;
267 * Error cases jump to one of these labels to undo any
268 * reservations which have already been performed.
271 if (resp
->tr_logres
> 0) {
274 if (resp
->tr_logflags
& XFS_TRANS_PERM_LOG_RES
) {
275 log_flags
= XFS_LOG_REL_PERM_RESERV
;
279 xfs_log_done(tp
->t_mountp
, tp
->t_ticket
, NULL
, log_flags
);
282 tp
->t_flags
&= ~XFS_TRANS_PERM_LOG_RES
;
287 xfs_icsb_modify_counters(tp
->t_mountp
, XFS_SBS_FDBLOCKS
,
288 (int64_t)blocks
, rsvd
);
292 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
298 * Record the indicated change to the given field for application
299 * to the file system's superblock when the transaction commits.
300 * For now, just store the change in the transaction structure.
302 * Mark the transaction structure to indicate that the superblock
303 * needs to be updated before committing.
305 * Because we may not be keeping track of allocated/free inodes and
306 * used filesystem blocks in the superblock, we do not mark the
307 * superblock dirty in this transaction if we modify these fields.
308 * We still need to update the transaction deltas so that they get
309 * applied to the incore superblock, but we don't want them to
310 * cause the superblock to get locked and logged if these are the
311 * only fields in the superblock that the transaction modifies.
319 uint32_t flags
= (XFS_TRANS_DIRTY
|XFS_TRANS_SB_DIRTY
);
320 xfs_mount_t
*mp
= tp
->t_mountp
;
323 case XFS_TRANS_SB_ICOUNT
:
324 tp
->t_icount_delta
+= delta
;
325 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
326 flags
&= ~XFS_TRANS_SB_DIRTY
;
328 case XFS_TRANS_SB_IFREE
:
329 tp
->t_ifree_delta
+= delta
;
330 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
331 flags
&= ~XFS_TRANS_SB_DIRTY
;
333 case XFS_TRANS_SB_FDBLOCKS
:
335 * Track the number of blocks allocated in the
336 * transaction. Make sure it does not exceed the
340 tp
->t_blk_res_used
+= (uint
)-delta
;
341 ASSERT(tp
->t_blk_res_used
<= tp
->t_blk_res
);
343 tp
->t_fdblocks_delta
+= delta
;
344 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
345 flags
&= ~XFS_TRANS_SB_DIRTY
;
347 case XFS_TRANS_SB_RES_FDBLOCKS
:
349 * The allocation has already been applied to the
350 * in-core superblock's counter. This should only
351 * be applied to the on-disk superblock.
354 tp
->t_res_fdblocks_delta
+= delta
;
355 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
356 flags
&= ~XFS_TRANS_SB_DIRTY
;
358 case XFS_TRANS_SB_FREXTENTS
:
360 * Track the number of blocks allocated in the
361 * transaction. Make sure it does not exceed the
365 tp
->t_rtx_res_used
+= (uint
)-delta
;
366 ASSERT(tp
->t_rtx_res_used
<= tp
->t_rtx_res
);
368 tp
->t_frextents_delta
+= delta
;
370 case XFS_TRANS_SB_RES_FREXTENTS
:
372 * The allocation has already been applied to the
373 * in-core superblock's counter. This should only
374 * be applied to the on-disk superblock.
377 tp
->t_res_frextents_delta
+= delta
;
379 case XFS_TRANS_SB_DBLOCKS
:
381 tp
->t_dblocks_delta
+= delta
;
383 case XFS_TRANS_SB_AGCOUNT
:
385 tp
->t_agcount_delta
+= delta
;
387 case XFS_TRANS_SB_IMAXPCT
:
388 tp
->t_imaxpct_delta
+= delta
;
390 case XFS_TRANS_SB_REXTSIZE
:
391 tp
->t_rextsize_delta
+= delta
;
393 case XFS_TRANS_SB_RBMBLOCKS
:
394 tp
->t_rbmblocks_delta
+= delta
;
396 case XFS_TRANS_SB_RBLOCKS
:
397 tp
->t_rblocks_delta
+= delta
;
399 case XFS_TRANS_SB_REXTENTS
:
400 tp
->t_rextents_delta
+= delta
;
402 case XFS_TRANS_SB_REXTSLOG
:
403 tp
->t_rextslog_delta
+= delta
;
410 tp
->t_flags
|= flags
;
414 * xfs_trans_apply_sb_deltas() is called from the commit code
415 * to bring the superblock buffer into the current transaction
416 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
418 * For now we just look at each field allowed to change and change
422 xfs_trans_apply_sb_deltas(
429 bp
= xfs_trans_getsb(tp
, tp
->t_mountp
, 0);
430 sbp
= XFS_BUF_TO_SBP(bp
);
433 * Check that superblock mods match the mods made to AGF counters.
435 ASSERT((tp
->t_fdblocks_delta
+ tp
->t_res_fdblocks_delta
) ==
436 (tp
->t_ag_freeblks_delta
+ tp
->t_ag_flist_delta
+
437 tp
->t_ag_btree_delta
));
440 * Only update the superblock counters if we are logging them
442 if (!xfs_sb_version_haslazysbcount(&(tp
->t_mountp
->m_sb
))) {
443 if (tp
->t_icount_delta
)
444 be64_add_cpu(&sbp
->sb_icount
, tp
->t_icount_delta
);
445 if (tp
->t_ifree_delta
)
446 be64_add_cpu(&sbp
->sb_ifree
, tp
->t_ifree_delta
);
447 if (tp
->t_fdblocks_delta
)
448 be64_add_cpu(&sbp
->sb_fdblocks
, tp
->t_fdblocks_delta
);
449 if (tp
->t_res_fdblocks_delta
)
450 be64_add_cpu(&sbp
->sb_fdblocks
, tp
->t_res_fdblocks_delta
);
453 if (tp
->t_frextents_delta
)
454 be64_add_cpu(&sbp
->sb_frextents
, tp
->t_frextents_delta
);
455 if (tp
->t_res_frextents_delta
)
456 be64_add_cpu(&sbp
->sb_frextents
, tp
->t_res_frextents_delta
);
458 if (tp
->t_dblocks_delta
) {
459 be64_add_cpu(&sbp
->sb_dblocks
, tp
->t_dblocks_delta
);
462 if (tp
->t_agcount_delta
) {
463 be32_add_cpu(&sbp
->sb_agcount
, tp
->t_agcount_delta
);
466 if (tp
->t_imaxpct_delta
) {
467 sbp
->sb_imax_pct
+= tp
->t_imaxpct_delta
;
470 if (tp
->t_rextsize_delta
) {
471 be32_add_cpu(&sbp
->sb_rextsize
, tp
->t_rextsize_delta
);
474 if (tp
->t_rbmblocks_delta
) {
475 be32_add_cpu(&sbp
->sb_rbmblocks
, tp
->t_rbmblocks_delta
);
478 if (tp
->t_rblocks_delta
) {
479 be64_add_cpu(&sbp
->sb_rblocks
, tp
->t_rblocks_delta
);
482 if (tp
->t_rextents_delta
) {
483 be64_add_cpu(&sbp
->sb_rextents
, tp
->t_rextents_delta
);
486 if (tp
->t_rextslog_delta
) {
487 sbp
->sb_rextslog
+= tp
->t_rextslog_delta
;
493 * Log the whole thing, the fields are noncontiguous.
495 xfs_trans_log_buf(tp
, bp
, 0, sizeof(xfs_dsb_t
) - 1);
498 * Since all the modifiable fields are contiguous, we
499 * can get away with this.
501 xfs_trans_log_buf(tp
, bp
, offsetof(xfs_dsb_t
, sb_icount
),
502 offsetof(xfs_dsb_t
, sb_frextents
) +
503 sizeof(sbp
->sb_frextents
) - 1);
507 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
508 * and apply superblock counter changes to the in-core superblock. The
509 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
510 * applied to the in-core superblock. The idea is that that has already been
513 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
514 * However, we have to ensure that we only modify each superblock field only
515 * once because the application of the delta values may not be atomic. That can
516 * lead to ENOSPC races occurring if we have two separate modifcations of the
517 * free space counter to put back the entire reservation and then take away
520 * If we are not logging superblock counters, then the inode allocated/free and
521 * used block counts are not updated in the on disk superblock. In this case,
522 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
523 * still need to update the incore superblock with the changes.
526 xfs_trans_unreserve_and_mod_sb(
529 xfs_mod_sb_t msb
[9]; /* If you add cases, add entries */
531 xfs_mount_t
*mp
= tp
->t_mountp
;
535 int64_t blkdelta
= 0;
536 int64_t rtxdelta
= 0;
538 int64_t ifreedelta
= 0;
541 rsvd
= (tp
->t_flags
& XFS_TRANS_RESERVE
) != 0;
543 /* calculate deltas */
544 if (tp
->t_blk_res
> 0)
545 blkdelta
= tp
->t_blk_res
;
546 if ((tp
->t_fdblocks_delta
!= 0) &&
547 (xfs_sb_version_haslazysbcount(&mp
->m_sb
) ||
548 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)))
549 blkdelta
+= tp
->t_fdblocks_delta
;
551 if (tp
->t_rtx_res
> 0)
552 rtxdelta
= tp
->t_rtx_res
;
553 if ((tp
->t_frextents_delta
!= 0) &&
554 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
))
555 rtxdelta
+= tp
->t_frextents_delta
;
557 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) ||
558 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)) {
559 idelta
= tp
->t_icount_delta
;
560 ifreedelta
= tp
->t_ifree_delta
;
563 /* apply the per-cpu counters */
565 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
572 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_ICOUNT
,
575 goto out_undo_fdblocks
;
579 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_IFREE
,
582 goto out_undo_icount
;
585 /* apply remaining deltas */
587 msbp
->msb_field
= XFS_SBS_FREXTENTS
;
588 msbp
->msb_delta
= rtxdelta
;
592 if (tp
->t_flags
& XFS_TRANS_SB_DIRTY
) {
593 if (tp
->t_dblocks_delta
!= 0) {
594 msbp
->msb_field
= XFS_SBS_DBLOCKS
;
595 msbp
->msb_delta
= tp
->t_dblocks_delta
;
598 if (tp
->t_agcount_delta
!= 0) {
599 msbp
->msb_field
= XFS_SBS_AGCOUNT
;
600 msbp
->msb_delta
= tp
->t_agcount_delta
;
603 if (tp
->t_imaxpct_delta
!= 0) {
604 msbp
->msb_field
= XFS_SBS_IMAX_PCT
;
605 msbp
->msb_delta
= tp
->t_imaxpct_delta
;
608 if (tp
->t_rextsize_delta
!= 0) {
609 msbp
->msb_field
= XFS_SBS_REXTSIZE
;
610 msbp
->msb_delta
= tp
->t_rextsize_delta
;
613 if (tp
->t_rbmblocks_delta
!= 0) {
614 msbp
->msb_field
= XFS_SBS_RBMBLOCKS
;
615 msbp
->msb_delta
= tp
->t_rbmblocks_delta
;
618 if (tp
->t_rblocks_delta
!= 0) {
619 msbp
->msb_field
= XFS_SBS_RBLOCKS
;
620 msbp
->msb_delta
= tp
->t_rblocks_delta
;
623 if (tp
->t_rextents_delta
!= 0) {
624 msbp
->msb_field
= XFS_SBS_REXTENTS
;
625 msbp
->msb_delta
= tp
->t_rextents_delta
;
628 if (tp
->t_rextslog_delta
!= 0) {
629 msbp
->msb_field
= XFS_SBS_REXTSLOG
;
630 msbp
->msb_delta
= tp
->t_rextslog_delta
;
636 * If we need to change anything, do it.
639 error
= xfs_mod_incore_sb_batch(tp
->t_mountp
, msb
,
640 (uint
)(msbp
- msb
), rsvd
);
642 goto out_undo_ifreecount
;
649 xfs_icsb_modify_counters(mp
, XFS_SBS_IFREE
, -ifreedelta
, rsvd
);
652 xfs_icsb_modify_counters(mp
, XFS_SBS_ICOUNT
, -idelta
, rsvd
);
655 xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
, -blkdelta
, rsvd
);
662 * Add the given log item to the transaction's list of log items.
664 * The log item will now point to its new descriptor with its li_desc field.
668 struct xfs_trans
*tp
,
669 struct xfs_log_item
*lip
)
671 struct xfs_log_item_desc
*lidp
;
673 ASSERT(lip
->li_mountp
== tp
->t_mountp
);
674 ASSERT(lip
->li_ailp
== tp
->t_mountp
->m_ail
);
676 lidp
= kmem_zone_zalloc(xfs_log_item_desc_zone
, KM_SLEEP
| KM_NOFS
);
678 lidp
->lid_item
= lip
;
680 list_add_tail(&lidp
->lid_trans
, &tp
->t_items
);
686 xfs_trans_free_item_desc(
687 struct xfs_log_item_desc
*lidp
)
689 list_del_init(&lidp
->lid_trans
);
690 kmem_zone_free(xfs_log_item_desc_zone
, lidp
);
694 * Unlink and free the given descriptor.
698 struct xfs_log_item
*lip
)
700 xfs_trans_free_item_desc(lip
->li_desc
);
705 * Unlock all of the items of a transaction and free all the descriptors
706 * of that transaction.
709 xfs_trans_free_items(
710 struct xfs_trans
*tp
,
711 xfs_lsn_t commit_lsn
,
714 struct xfs_log_item_desc
*lidp
, *next
;
716 list_for_each_entry_safe(lidp
, next
, &tp
->t_items
, lid_trans
) {
717 struct xfs_log_item
*lip
= lidp
->lid_item
;
721 if (commit_lsn
!= NULLCOMMITLSN
)
722 lip
->li_ops
->iop_committing(lip
, commit_lsn
);
723 if (flags
& XFS_TRANS_ABORT
)
724 lip
->li_flags
|= XFS_LI_ABORTED
;
725 lip
->li_ops
->iop_unlock(lip
);
727 xfs_trans_free_item_desc(lidp
);
732 xfs_log_item_batch_insert(
733 struct xfs_ail
*ailp
,
734 struct xfs_ail_cursor
*cur
,
735 struct xfs_log_item
**log_items
,
737 xfs_lsn_t commit_lsn
)
741 spin_lock(&ailp
->xa_lock
);
742 /* xfs_trans_ail_update_bulk drops ailp->xa_lock */
743 xfs_trans_ail_update_bulk(ailp
, cur
, log_items
, nr_items
, commit_lsn
);
745 for (i
= 0; i
< nr_items
; i
++) {
746 struct xfs_log_item
*lip
= log_items
[i
];
748 lip
->li_ops
->iop_unpin(lip
, 0);
753 * Bulk operation version of xfs_trans_committed that takes a log vector of
754 * items to insert into the AIL. This uses bulk AIL insertion techniques to
755 * minimise lock traffic.
757 * If we are called with the aborted flag set, it is because a log write during
758 * a CIL checkpoint commit has failed. In this case, all the items in the
759 * checkpoint have already gone through iop_commited and iop_unlock, which
760 * means that checkpoint commit abort handling is treated exactly the same
761 * as an iclog write error even though we haven't started any IO yet. Hence in
762 * this case all we need to do is iop_committed processing, followed by an
763 * iop_unpin(aborted) call.
765 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
766 * at the end of the AIL, the insert cursor avoids the need to walk
767 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
768 * call. This saves a lot of needless list walking and is a net win, even
769 * though it slightly increases that amount of AIL lock traffic to set it up
773 xfs_trans_committed_bulk(
774 struct xfs_ail
*ailp
,
775 struct xfs_log_vec
*log_vector
,
776 xfs_lsn_t commit_lsn
,
779 #define LOG_ITEM_BATCH_SIZE 32
780 struct xfs_log_item
*log_items
[LOG_ITEM_BATCH_SIZE
];
781 struct xfs_log_vec
*lv
;
782 struct xfs_ail_cursor cur
;
785 spin_lock(&ailp
->xa_lock
);
786 xfs_trans_ail_cursor_last(ailp
, &cur
, commit_lsn
);
787 spin_unlock(&ailp
->xa_lock
);
789 /* unpin all the log items */
790 for (lv
= log_vector
; lv
; lv
= lv
->lv_next
) {
791 struct xfs_log_item
*lip
= lv
->lv_item
;
795 lip
->li_flags
|= XFS_LI_ABORTED
;
796 item_lsn
= lip
->li_ops
->iop_committed(lip
, commit_lsn
);
798 /* item_lsn of -1 means the item needs no further processing */
799 if (XFS_LSN_CMP(item_lsn
, (xfs_lsn_t
)-1) == 0)
803 * if we are aborting the operation, no point in inserting the
804 * object into the AIL as we are in a shutdown situation.
807 ASSERT(XFS_FORCED_SHUTDOWN(ailp
->xa_mount
));
808 lip
->li_ops
->iop_unpin(lip
, 1);
812 if (item_lsn
!= commit_lsn
) {
815 * Not a bulk update option due to unusual item_lsn.
816 * Push into AIL immediately, rechecking the lsn once
817 * we have the ail lock. Then unpin the item. This does
818 * not affect the AIL cursor the bulk insert path is
821 spin_lock(&ailp
->xa_lock
);
822 if (XFS_LSN_CMP(item_lsn
, lip
->li_lsn
) > 0)
823 xfs_trans_ail_update(ailp
, lip
, item_lsn
);
825 spin_unlock(&ailp
->xa_lock
);
826 lip
->li_ops
->iop_unpin(lip
, 0);
830 /* Item is a candidate for bulk AIL insert. */
831 log_items
[i
++] = lv
->lv_item
;
832 if (i
>= LOG_ITEM_BATCH_SIZE
) {
833 xfs_log_item_batch_insert(ailp
, &cur
, log_items
,
834 LOG_ITEM_BATCH_SIZE
, commit_lsn
);
839 /* make sure we insert the remainder! */
841 xfs_log_item_batch_insert(ailp
, &cur
, log_items
, i
, commit_lsn
);
843 spin_lock(&ailp
->xa_lock
);
844 xfs_trans_ail_cursor_done(ailp
, &cur
);
845 spin_unlock(&ailp
->xa_lock
);
849 * Commit the given transaction to the log.
851 * XFS disk error handling mechanism is not based on a typical
852 * transaction abort mechanism. Logically after the filesystem
853 * gets marked 'SHUTDOWN', we can't let any new transactions
854 * be durable - ie. committed to disk - because some metadata might
855 * be inconsistent. In such cases, this returns an error, and the
856 * caller may assume that all locked objects joined to the transaction
857 * have already been unlocked as if the commit had succeeded.
858 * Do not reference the transaction structure after this call.
862 struct xfs_trans
*tp
,
865 struct xfs_mount
*mp
= tp
->t_mountp
;
866 xfs_lsn_t commit_lsn
= -1;
869 int sync
= tp
->t_flags
& XFS_TRANS_SYNC
;
872 * Determine whether this commit is releasing a permanent
873 * log reservation or not.
875 if (flags
& XFS_TRANS_RELEASE_LOG_RES
) {
876 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
877 log_flags
= XFS_LOG_REL_PERM_RESERV
;
881 * If there is nothing to be logged by the transaction,
882 * then unlock all of the items associated with the
883 * transaction and free the transaction structure.
884 * Also make sure to return any reserved blocks to
887 if (!(tp
->t_flags
& XFS_TRANS_DIRTY
))
890 if (XFS_FORCED_SHUTDOWN(mp
)) {
891 error
= XFS_ERROR(EIO
);
895 ASSERT(tp
->t_ticket
!= NULL
);
898 * If we need to update the superblock, then do it now.
900 if (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)
901 xfs_trans_apply_sb_deltas(tp
);
902 xfs_trans_apply_dquot_deltas(tp
);
904 error
= xfs_log_commit_cil(mp
, tp
, &commit_lsn
, flags
);
905 if (error
== ENOMEM
) {
906 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
907 error
= XFS_ERROR(EIO
);
911 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
915 * If the transaction needs to be synchronous, then force the
916 * log out now and wait for it.
920 error
= _xfs_log_force_lsn(mp
, commit_lsn
,
923 XFS_STATS_INC(xs_trans_sync
);
925 XFS_STATS_INC(xs_trans_async
);
931 xfs_trans_unreserve_and_mod_sb(tp
);
934 * It is indeed possible for the transaction to be not dirty but
935 * the dqinfo portion to be. All that means is that we have some
936 * (non-persistent) quota reservations that need to be unreserved.
938 xfs_trans_unreserve_and_mod_dquots(tp
);
940 commit_lsn
= xfs_log_done(mp
, tp
->t_ticket
, NULL
, log_flags
);
941 if (commit_lsn
== -1 && !error
)
942 error
= XFS_ERROR(EIO
);
944 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
945 xfs_trans_free_items(tp
, NULLCOMMITLSN
, error
? XFS_TRANS_ABORT
: 0);
948 XFS_STATS_INC(xs_trans_empty
);
953 * Unlock all of the transaction's items and free the transaction.
954 * The transaction must not have modified any of its items, because
955 * there is no way to restore them to their previous state.
957 * If the transaction has made a log reservation, make sure to release
966 xfs_mount_t
*mp
= tp
->t_mountp
;
969 * See if the caller is being too lazy to figure out if
970 * the transaction really needs an abort.
972 if ((flags
& XFS_TRANS_ABORT
) && !(tp
->t_flags
& XFS_TRANS_DIRTY
))
973 flags
&= ~XFS_TRANS_ABORT
;
975 * See if the caller is relying on us to shut down the
976 * filesystem. This happens in paths where we detect
977 * corruption and decide to give up.
979 if ((tp
->t_flags
& XFS_TRANS_DIRTY
) && !XFS_FORCED_SHUTDOWN(mp
)) {
980 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW
, mp
);
981 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
984 if (!(flags
& XFS_TRANS_ABORT
) && !XFS_FORCED_SHUTDOWN(mp
)) {
985 struct xfs_log_item_desc
*lidp
;
987 list_for_each_entry(lidp
, &tp
->t_items
, lid_trans
)
988 ASSERT(!(lidp
->lid_item
->li_type
== XFS_LI_EFD
));
991 xfs_trans_unreserve_and_mod_sb(tp
);
992 xfs_trans_unreserve_and_mod_dquots(tp
);
995 if (flags
& XFS_TRANS_RELEASE_LOG_RES
) {
996 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
997 log_flags
= XFS_LOG_REL_PERM_RESERV
;
1001 xfs_log_done(mp
, tp
->t_ticket
, NULL
, log_flags
);
1004 /* mark this thread as no longer being in a transaction */
1005 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
1007 xfs_trans_free_items(tp
, NULLCOMMITLSN
, flags
);
1012 * Roll from one trans in the sequence of PERMANENT transactions to
1013 * the next: permanent transactions are only flushed out when
1014 * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
1015 * as possible to let chunks of it go to the log. So we commit the
1016 * chunk we've been working on and get a new transaction to continue.
1020 struct xfs_trans
**tpp
,
1021 struct xfs_inode
*dp
)
1023 struct xfs_trans
*trans
;
1024 struct xfs_trans_res tres
;
1028 * Ensure that the inode is always logged.
1031 xfs_trans_log_inode(trans
, dp
, XFS_ILOG_CORE
);
1034 * Copy the critical parameters from one trans to the next.
1036 tres
.tr_logres
= trans
->t_log_res
;
1037 tres
.tr_logcount
= trans
->t_log_count
;
1038 *tpp
= xfs_trans_dup(trans
);
1041 * Commit the current transaction.
1042 * If this commit failed, then it'd just unlock those items that
1043 * are not marked ihold. That also means that a filesystem shutdown
1044 * is in progress. The caller takes the responsibility to cancel
1045 * the duplicate transaction that gets returned.
1047 error
= xfs_trans_commit(trans
, 0);
1054 * transaction commit worked ok so we can drop the extra ticket
1055 * reference that we gained in xfs_trans_dup()
1057 xfs_log_ticket_put(trans
->t_ticket
);
1061 * Reserve space in the log for th next transaction.
1062 * This also pushes items in the "AIL", the list of logged items,
1063 * out to disk if they are taking up space at the tail of the log
1064 * that we want to use. This requires that either nothing be locked
1065 * across this call, or that anything that is locked be logged in
1066 * the prior and the next transactions.
1068 tres
.tr_logflags
= XFS_TRANS_PERM_LOG_RES
;
1069 error
= xfs_trans_reserve(trans
, &tres
, 0, 0);
1071 * Ensure that the inode is in the new transaction and locked.
1076 xfs_trans_ijoin(trans
, dp
, 0);