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"
22 #include "xfs_shared.h"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_error.h"
29 #include "xfs_da_btree.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"
42 #include "xfs_trans_priv.h"
43 #include "xfs_trans_space.h"
44 #include "xfs_inode_item.h"
45 #include "xfs_log_priv.h"
46 #include "xfs_buf_item.h"
47 #include "xfs_trace.h"
49 kmem_zone_t
*xfs_trans_zone
;
50 kmem_zone_t
*xfs_log_item_desc_zone
;
53 * Initialize the precomputed transaction reservation values
54 * in the mount structure.
60 xfs_trans_resv_calc(mp
, M_RES(mp
));
64 * This routine is called to allocate a transaction structure.
65 * The type parameter indicates the type of the transaction. These
66 * are enumerated in xfs_trans.h.
68 * Dynamically allocate the transaction structure from the transaction
69 * zone, initialize it, and return it to the caller.
78 sb_start_intwrite(mp
->m_super
);
79 tp
= _xfs_trans_alloc(mp
, type
, KM_SLEEP
);
80 tp
->t_flags
|= XFS_TRANS_FREEZE_PROT
;
88 xfs_km_flags_t memflags
)
92 WARN_ON(mp
->m_super
->s_writers
.frozen
== SB_FREEZE_COMPLETE
);
93 atomic_inc(&mp
->m_active_trans
);
95 tp
= kmem_zone_zalloc(xfs_trans_zone
, memflags
);
96 tp
->t_magic
= XFS_TRANS_HEADER_MAGIC
;
99 INIT_LIST_HEAD(&tp
->t_items
);
100 INIT_LIST_HEAD(&tp
->t_busy
);
105 * Free the transaction structure. If there is more clean up
106 * to do when the structure is freed, add it here.
110 struct xfs_trans
*tp
)
112 xfs_extent_busy_sort(&tp
->t_busy
);
113 xfs_extent_busy_clear(tp
->t_mountp
, &tp
->t_busy
, false);
115 atomic_dec(&tp
->t_mountp
->m_active_trans
);
116 if (tp
->t_flags
& XFS_TRANS_FREEZE_PROT
)
117 sb_end_intwrite(tp
->t_mountp
->m_super
);
118 xfs_trans_free_dqinfo(tp
);
119 kmem_zone_free(xfs_trans_zone
, tp
);
123 * This is called to create a new transaction which will share the
124 * permanent log reservation of the given transaction. The remaining
125 * unused block and rt extent reservations are also inherited. This
126 * implies that the original transaction is no longer allowed to allocate
127 * blocks. Locks and log items, however, are no inherited. They must
128 * be added to the new transaction explicitly.
136 ntp
= kmem_zone_zalloc(xfs_trans_zone
, KM_SLEEP
);
139 * Initialize the new transaction structure.
141 ntp
->t_magic
= XFS_TRANS_HEADER_MAGIC
;
142 ntp
->t_type
= tp
->t_type
;
143 ntp
->t_mountp
= tp
->t_mountp
;
144 INIT_LIST_HEAD(&ntp
->t_items
);
145 INIT_LIST_HEAD(&ntp
->t_busy
);
147 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
148 ASSERT(tp
->t_ticket
!= NULL
);
150 ntp
->t_flags
= XFS_TRANS_PERM_LOG_RES
|
151 (tp
->t_flags
& XFS_TRANS_RESERVE
) |
152 (tp
->t_flags
& XFS_TRANS_FREEZE_PROT
);
153 /* We gave our writer reference to the new transaction */
154 tp
->t_flags
&= ~XFS_TRANS_FREEZE_PROT
;
155 ntp
->t_ticket
= xfs_log_ticket_get(tp
->t_ticket
);
156 ntp
->t_blk_res
= tp
->t_blk_res
- tp
->t_blk_res_used
;
157 tp
->t_blk_res
= tp
->t_blk_res_used
;
158 ntp
->t_rtx_res
= tp
->t_rtx_res
- tp
->t_rtx_res_used
;
159 tp
->t_rtx_res
= tp
->t_rtx_res_used
;
160 ntp
->t_pflags
= tp
->t_pflags
;
162 xfs_trans_dup_dqinfo(tp
, ntp
);
164 atomic_inc(&tp
->t_mountp
->m_active_trans
);
169 * This is called to reserve free disk blocks and log space for the
170 * given transaction. This must be done before allocating any resources
171 * within the transaction.
173 * This will return ENOSPC if there are not enough blocks available.
174 * It will sleep waiting for available log space.
175 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
176 * is used by long running transactions. If any one of the reservations
177 * fails then they will all be backed out.
179 * This does not do quota reservations. That typically is done by the
184 struct xfs_trans
*tp
,
185 struct xfs_trans_res
*resp
,
190 int rsvd
= (tp
->t_flags
& XFS_TRANS_RESERVE
) != 0;
192 /* Mark this thread as being in a transaction */
193 current_set_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
196 * Attempt to reserve the needed disk blocks by decrementing
197 * the number needed from the number available. This will
198 * fail if the count would go below zero.
201 error
= xfs_icsb_modify_counters(tp
->t_mountp
, XFS_SBS_FDBLOCKS
,
202 -((int64_t)blocks
), rsvd
);
204 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
205 return (XFS_ERROR(ENOSPC
));
207 tp
->t_blk_res
+= blocks
;
211 * Reserve the log space needed for this transaction.
213 if (resp
->tr_logres
> 0) {
214 bool permanent
= false;
216 ASSERT(tp
->t_log_res
== 0 ||
217 tp
->t_log_res
== resp
->tr_logres
);
218 ASSERT(tp
->t_log_count
== 0 ||
219 tp
->t_log_count
== resp
->tr_logcount
);
221 if (resp
->tr_logflags
& XFS_TRANS_PERM_LOG_RES
) {
222 tp
->t_flags
|= XFS_TRANS_PERM_LOG_RES
;
225 ASSERT(tp
->t_ticket
== NULL
);
226 ASSERT(!(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
));
229 if (tp
->t_ticket
!= NULL
) {
230 ASSERT(resp
->tr_logflags
& XFS_TRANS_PERM_LOG_RES
);
231 error
= xfs_log_regrant(tp
->t_mountp
, tp
->t_ticket
);
233 error
= xfs_log_reserve(tp
->t_mountp
,
236 &tp
->t_ticket
, XFS_TRANSACTION
,
237 permanent
, tp
->t_type
);
243 tp
->t_log_res
= resp
->tr_logres
;
244 tp
->t_log_count
= resp
->tr_logcount
;
248 * Attempt to reserve the needed realtime extents by decrementing
249 * the number needed from the number available. This will
250 * fail if the count would go below zero.
253 error
= xfs_mod_incore_sb(tp
->t_mountp
, XFS_SBS_FREXTENTS
,
254 -((int64_t)rtextents
), rsvd
);
256 error
= XFS_ERROR(ENOSPC
);
259 tp
->t_rtx_res
+= rtextents
;
265 * Error cases jump to one of these labels to undo any
266 * reservations which have already been performed.
269 if (resp
->tr_logres
> 0) {
272 if (resp
->tr_logflags
& XFS_TRANS_PERM_LOG_RES
) {
273 log_flags
= XFS_LOG_REL_PERM_RESERV
;
277 xfs_log_done(tp
->t_mountp
, tp
->t_ticket
, NULL
, log_flags
);
280 tp
->t_flags
&= ~XFS_TRANS_PERM_LOG_RES
;
285 xfs_icsb_modify_counters(tp
->t_mountp
, XFS_SBS_FDBLOCKS
,
286 (int64_t)blocks
, rsvd
);
290 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
296 * Record the indicated change to the given field for application
297 * to the file system's superblock when the transaction commits.
298 * For now, just store the change in the transaction structure.
300 * Mark the transaction structure to indicate that the superblock
301 * needs to be updated before committing.
303 * Because we may not be keeping track of allocated/free inodes and
304 * used filesystem blocks in the superblock, we do not mark the
305 * superblock dirty in this transaction if we modify these fields.
306 * We still need to update the transaction deltas so that they get
307 * applied to the incore superblock, but we don't want them to
308 * cause the superblock to get locked and logged if these are the
309 * only fields in the superblock that the transaction modifies.
317 uint32_t flags
= (XFS_TRANS_DIRTY
|XFS_TRANS_SB_DIRTY
);
318 xfs_mount_t
*mp
= tp
->t_mountp
;
321 case XFS_TRANS_SB_ICOUNT
:
322 tp
->t_icount_delta
+= delta
;
323 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
324 flags
&= ~XFS_TRANS_SB_DIRTY
;
326 case XFS_TRANS_SB_IFREE
:
327 tp
->t_ifree_delta
+= delta
;
328 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
329 flags
&= ~XFS_TRANS_SB_DIRTY
;
331 case XFS_TRANS_SB_FDBLOCKS
:
333 * Track the number of blocks allocated in the
334 * transaction. Make sure it does not exceed the
338 tp
->t_blk_res_used
+= (uint
)-delta
;
339 ASSERT(tp
->t_blk_res_used
<= tp
->t_blk_res
);
341 tp
->t_fdblocks_delta
+= delta
;
342 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
343 flags
&= ~XFS_TRANS_SB_DIRTY
;
345 case XFS_TRANS_SB_RES_FDBLOCKS
:
347 * The allocation has already been applied to the
348 * in-core superblock's counter. This should only
349 * be applied to the on-disk superblock.
352 tp
->t_res_fdblocks_delta
+= delta
;
353 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
354 flags
&= ~XFS_TRANS_SB_DIRTY
;
356 case XFS_TRANS_SB_FREXTENTS
:
358 * Track the number of blocks allocated in the
359 * transaction. Make sure it does not exceed the
363 tp
->t_rtx_res_used
+= (uint
)-delta
;
364 ASSERT(tp
->t_rtx_res_used
<= tp
->t_rtx_res
);
366 tp
->t_frextents_delta
+= delta
;
368 case XFS_TRANS_SB_RES_FREXTENTS
:
370 * The allocation has already been applied to the
371 * in-core superblock's counter. This should only
372 * be applied to the on-disk superblock.
375 tp
->t_res_frextents_delta
+= delta
;
377 case XFS_TRANS_SB_DBLOCKS
:
379 tp
->t_dblocks_delta
+= delta
;
381 case XFS_TRANS_SB_AGCOUNT
:
383 tp
->t_agcount_delta
+= delta
;
385 case XFS_TRANS_SB_IMAXPCT
:
386 tp
->t_imaxpct_delta
+= delta
;
388 case XFS_TRANS_SB_REXTSIZE
:
389 tp
->t_rextsize_delta
+= delta
;
391 case XFS_TRANS_SB_RBMBLOCKS
:
392 tp
->t_rbmblocks_delta
+= delta
;
394 case XFS_TRANS_SB_RBLOCKS
:
395 tp
->t_rblocks_delta
+= delta
;
397 case XFS_TRANS_SB_REXTENTS
:
398 tp
->t_rextents_delta
+= delta
;
400 case XFS_TRANS_SB_REXTSLOG
:
401 tp
->t_rextslog_delta
+= delta
;
408 tp
->t_flags
|= flags
;
412 * xfs_trans_apply_sb_deltas() is called from the commit code
413 * to bring the superblock buffer into the current transaction
414 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
416 * For now we just look at each field allowed to change and change
420 xfs_trans_apply_sb_deltas(
427 bp
= xfs_trans_getsb(tp
, tp
->t_mountp
, 0);
428 sbp
= XFS_BUF_TO_SBP(bp
);
431 * Check that superblock mods match the mods made to AGF counters.
433 ASSERT((tp
->t_fdblocks_delta
+ tp
->t_res_fdblocks_delta
) ==
434 (tp
->t_ag_freeblks_delta
+ tp
->t_ag_flist_delta
+
435 tp
->t_ag_btree_delta
));
438 * Only update the superblock counters if we are logging them
440 if (!xfs_sb_version_haslazysbcount(&(tp
->t_mountp
->m_sb
))) {
441 if (tp
->t_icount_delta
)
442 be64_add_cpu(&sbp
->sb_icount
, tp
->t_icount_delta
);
443 if (tp
->t_ifree_delta
)
444 be64_add_cpu(&sbp
->sb_ifree
, tp
->t_ifree_delta
);
445 if (tp
->t_fdblocks_delta
)
446 be64_add_cpu(&sbp
->sb_fdblocks
, tp
->t_fdblocks_delta
);
447 if (tp
->t_res_fdblocks_delta
)
448 be64_add_cpu(&sbp
->sb_fdblocks
, tp
->t_res_fdblocks_delta
);
451 if (tp
->t_frextents_delta
)
452 be64_add_cpu(&sbp
->sb_frextents
, tp
->t_frextents_delta
);
453 if (tp
->t_res_frextents_delta
)
454 be64_add_cpu(&sbp
->sb_frextents
, tp
->t_res_frextents_delta
);
456 if (tp
->t_dblocks_delta
) {
457 be64_add_cpu(&sbp
->sb_dblocks
, tp
->t_dblocks_delta
);
460 if (tp
->t_agcount_delta
) {
461 be32_add_cpu(&sbp
->sb_agcount
, tp
->t_agcount_delta
);
464 if (tp
->t_imaxpct_delta
) {
465 sbp
->sb_imax_pct
+= tp
->t_imaxpct_delta
;
468 if (tp
->t_rextsize_delta
) {
469 be32_add_cpu(&sbp
->sb_rextsize
, tp
->t_rextsize_delta
);
472 if (tp
->t_rbmblocks_delta
) {
473 be32_add_cpu(&sbp
->sb_rbmblocks
, tp
->t_rbmblocks_delta
);
476 if (tp
->t_rblocks_delta
) {
477 be64_add_cpu(&sbp
->sb_rblocks
, tp
->t_rblocks_delta
);
480 if (tp
->t_rextents_delta
) {
481 be64_add_cpu(&sbp
->sb_rextents
, tp
->t_rextents_delta
);
484 if (tp
->t_rextslog_delta
) {
485 sbp
->sb_rextslog
+= tp
->t_rextslog_delta
;
491 * Log the whole thing, the fields are noncontiguous.
493 xfs_trans_log_buf(tp
, bp
, 0, sizeof(xfs_dsb_t
) - 1);
496 * Since all the modifiable fields are contiguous, we
497 * can get away with this.
499 xfs_trans_log_buf(tp
, bp
, offsetof(xfs_dsb_t
, sb_icount
),
500 offsetof(xfs_dsb_t
, sb_frextents
) +
501 sizeof(sbp
->sb_frextents
) - 1);
505 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
506 * and apply superblock counter changes to the in-core superblock. The
507 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
508 * applied to the in-core superblock. The idea is that that has already been
511 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
512 * However, we have to ensure that we only modify each superblock field only
513 * once because the application of the delta values may not be atomic. That can
514 * lead to ENOSPC races occurring if we have two separate modifcations of the
515 * free space counter to put back the entire reservation and then take away
518 * If we are not logging superblock counters, then the inode allocated/free and
519 * used block counts are not updated in the on disk superblock. In this case,
520 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
521 * still need to update the incore superblock with the changes.
524 xfs_trans_unreserve_and_mod_sb(
527 xfs_mod_sb_t msb
[9]; /* If you add cases, add entries */
529 xfs_mount_t
*mp
= tp
->t_mountp
;
533 int64_t blkdelta
= 0;
534 int64_t rtxdelta
= 0;
536 int64_t ifreedelta
= 0;
539 rsvd
= (tp
->t_flags
& XFS_TRANS_RESERVE
) != 0;
541 /* calculate deltas */
542 if (tp
->t_blk_res
> 0)
543 blkdelta
= tp
->t_blk_res
;
544 if ((tp
->t_fdblocks_delta
!= 0) &&
545 (xfs_sb_version_haslazysbcount(&mp
->m_sb
) ||
546 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)))
547 blkdelta
+= tp
->t_fdblocks_delta
;
549 if (tp
->t_rtx_res
> 0)
550 rtxdelta
= tp
->t_rtx_res
;
551 if ((tp
->t_frextents_delta
!= 0) &&
552 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
))
553 rtxdelta
+= tp
->t_frextents_delta
;
555 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) ||
556 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)) {
557 idelta
= tp
->t_icount_delta
;
558 ifreedelta
= tp
->t_ifree_delta
;
561 /* apply the per-cpu counters */
563 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
570 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_ICOUNT
,
573 goto out_undo_fdblocks
;
577 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_IFREE
,
580 goto out_undo_icount
;
583 /* apply remaining deltas */
585 msbp
->msb_field
= XFS_SBS_FREXTENTS
;
586 msbp
->msb_delta
= rtxdelta
;
590 if (tp
->t_flags
& XFS_TRANS_SB_DIRTY
) {
591 if (tp
->t_dblocks_delta
!= 0) {
592 msbp
->msb_field
= XFS_SBS_DBLOCKS
;
593 msbp
->msb_delta
= tp
->t_dblocks_delta
;
596 if (tp
->t_agcount_delta
!= 0) {
597 msbp
->msb_field
= XFS_SBS_AGCOUNT
;
598 msbp
->msb_delta
= tp
->t_agcount_delta
;
601 if (tp
->t_imaxpct_delta
!= 0) {
602 msbp
->msb_field
= XFS_SBS_IMAX_PCT
;
603 msbp
->msb_delta
= tp
->t_imaxpct_delta
;
606 if (tp
->t_rextsize_delta
!= 0) {
607 msbp
->msb_field
= XFS_SBS_REXTSIZE
;
608 msbp
->msb_delta
= tp
->t_rextsize_delta
;
611 if (tp
->t_rbmblocks_delta
!= 0) {
612 msbp
->msb_field
= XFS_SBS_RBMBLOCKS
;
613 msbp
->msb_delta
= tp
->t_rbmblocks_delta
;
616 if (tp
->t_rblocks_delta
!= 0) {
617 msbp
->msb_field
= XFS_SBS_RBLOCKS
;
618 msbp
->msb_delta
= tp
->t_rblocks_delta
;
621 if (tp
->t_rextents_delta
!= 0) {
622 msbp
->msb_field
= XFS_SBS_REXTENTS
;
623 msbp
->msb_delta
= tp
->t_rextents_delta
;
626 if (tp
->t_rextslog_delta
!= 0) {
627 msbp
->msb_field
= XFS_SBS_REXTSLOG
;
628 msbp
->msb_delta
= tp
->t_rextslog_delta
;
634 * If we need to change anything, do it.
637 error
= xfs_mod_incore_sb_batch(tp
->t_mountp
, msb
,
638 (uint
)(msbp
- msb
), rsvd
);
640 goto out_undo_ifreecount
;
647 xfs_icsb_modify_counters(mp
, XFS_SBS_IFREE
, -ifreedelta
, rsvd
);
650 xfs_icsb_modify_counters(mp
, XFS_SBS_ICOUNT
, -idelta
, rsvd
);
653 xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
, -blkdelta
, rsvd
);
660 * Add the given log item to the transaction's list of log items.
662 * The log item will now point to its new descriptor with its li_desc field.
666 struct xfs_trans
*tp
,
667 struct xfs_log_item
*lip
)
669 struct xfs_log_item_desc
*lidp
;
671 ASSERT(lip
->li_mountp
== tp
->t_mountp
);
672 ASSERT(lip
->li_ailp
== tp
->t_mountp
->m_ail
);
674 lidp
= kmem_zone_zalloc(xfs_log_item_desc_zone
, KM_SLEEP
| KM_NOFS
);
676 lidp
->lid_item
= lip
;
678 list_add_tail(&lidp
->lid_trans
, &tp
->t_items
);
684 xfs_trans_free_item_desc(
685 struct xfs_log_item_desc
*lidp
)
687 list_del_init(&lidp
->lid_trans
);
688 kmem_zone_free(xfs_log_item_desc_zone
, lidp
);
692 * Unlink and free the given descriptor.
696 struct xfs_log_item
*lip
)
698 xfs_trans_free_item_desc(lip
->li_desc
);
703 * Unlock all of the items of a transaction and free all the descriptors
704 * of that transaction.
707 xfs_trans_free_items(
708 struct xfs_trans
*tp
,
709 xfs_lsn_t commit_lsn
,
712 struct xfs_log_item_desc
*lidp
, *next
;
714 list_for_each_entry_safe(lidp
, next
, &tp
->t_items
, lid_trans
) {
715 struct xfs_log_item
*lip
= lidp
->lid_item
;
719 if (commit_lsn
!= NULLCOMMITLSN
)
720 lip
->li_ops
->iop_committing(lip
, commit_lsn
);
721 if (flags
& XFS_TRANS_ABORT
)
722 lip
->li_flags
|= XFS_LI_ABORTED
;
723 lip
->li_ops
->iop_unlock(lip
);
725 xfs_trans_free_item_desc(lidp
);
730 xfs_log_item_batch_insert(
731 struct xfs_ail
*ailp
,
732 struct xfs_ail_cursor
*cur
,
733 struct xfs_log_item
**log_items
,
735 xfs_lsn_t commit_lsn
)
739 spin_lock(&ailp
->xa_lock
);
740 /* xfs_trans_ail_update_bulk drops ailp->xa_lock */
741 xfs_trans_ail_update_bulk(ailp
, cur
, log_items
, nr_items
, commit_lsn
);
743 for (i
= 0; i
< nr_items
; i
++) {
744 struct xfs_log_item
*lip
= log_items
[i
];
746 lip
->li_ops
->iop_unpin(lip
, 0);
751 * Bulk operation version of xfs_trans_committed that takes a log vector of
752 * items to insert into the AIL. This uses bulk AIL insertion techniques to
753 * minimise lock traffic.
755 * If we are called with the aborted flag set, it is because a log write during
756 * a CIL checkpoint commit has failed. In this case, all the items in the
757 * checkpoint have already gone through iop_commited and iop_unlock, which
758 * means that checkpoint commit abort handling is treated exactly the same
759 * as an iclog write error even though we haven't started any IO yet. Hence in
760 * this case all we need to do is iop_committed processing, followed by an
761 * iop_unpin(aborted) call.
763 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
764 * at the end of the AIL, the insert cursor avoids the need to walk
765 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
766 * call. This saves a lot of needless list walking and is a net win, even
767 * though it slightly increases that amount of AIL lock traffic to set it up
771 xfs_trans_committed_bulk(
772 struct xfs_ail
*ailp
,
773 struct xfs_log_vec
*log_vector
,
774 xfs_lsn_t commit_lsn
,
777 #define LOG_ITEM_BATCH_SIZE 32
778 struct xfs_log_item
*log_items
[LOG_ITEM_BATCH_SIZE
];
779 struct xfs_log_vec
*lv
;
780 struct xfs_ail_cursor cur
;
783 spin_lock(&ailp
->xa_lock
);
784 xfs_trans_ail_cursor_last(ailp
, &cur
, commit_lsn
);
785 spin_unlock(&ailp
->xa_lock
);
787 /* unpin all the log items */
788 for (lv
= log_vector
; lv
; lv
= lv
->lv_next
) {
789 struct xfs_log_item
*lip
= lv
->lv_item
;
793 lip
->li_flags
|= XFS_LI_ABORTED
;
794 item_lsn
= lip
->li_ops
->iop_committed(lip
, commit_lsn
);
796 /* item_lsn of -1 means the item needs no further processing */
797 if (XFS_LSN_CMP(item_lsn
, (xfs_lsn_t
)-1) == 0)
801 * if we are aborting the operation, no point in inserting the
802 * object into the AIL as we are in a shutdown situation.
805 ASSERT(XFS_FORCED_SHUTDOWN(ailp
->xa_mount
));
806 lip
->li_ops
->iop_unpin(lip
, 1);
810 if (item_lsn
!= commit_lsn
) {
813 * Not a bulk update option due to unusual item_lsn.
814 * Push into AIL immediately, rechecking the lsn once
815 * we have the ail lock. Then unpin the item. This does
816 * not affect the AIL cursor the bulk insert path is
819 spin_lock(&ailp
->xa_lock
);
820 if (XFS_LSN_CMP(item_lsn
, lip
->li_lsn
) > 0)
821 xfs_trans_ail_update(ailp
, lip
, item_lsn
);
823 spin_unlock(&ailp
->xa_lock
);
824 lip
->li_ops
->iop_unpin(lip
, 0);
828 /* Item is a candidate for bulk AIL insert. */
829 log_items
[i
++] = lv
->lv_item
;
830 if (i
>= LOG_ITEM_BATCH_SIZE
) {
831 xfs_log_item_batch_insert(ailp
, &cur
, log_items
,
832 LOG_ITEM_BATCH_SIZE
, commit_lsn
);
837 /* make sure we insert the remainder! */
839 xfs_log_item_batch_insert(ailp
, &cur
, log_items
, i
, commit_lsn
);
841 spin_lock(&ailp
->xa_lock
);
842 xfs_trans_ail_cursor_done(ailp
, &cur
);
843 spin_unlock(&ailp
->xa_lock
);
847 * Commit the given transaction to the log.
849 * XFS disk error handling mechanism is not based on a typical
850 * transaction abort mechanism. Logically after the filesystem
851 * gets marked 'SHUTDOWN', we can't let any new transactions
852 * be durable - ie. committed to disk - because some metadata might
853 * be inconsistent. In such cases, this returns an error, and the
854 * caller may assume that all locked objects joined to the transaction
855 * have already been unlocked as if the commit had succeeded.
856 * Do not reference the transaction structure after this call.
860 struct xfs_trans
*tp
,
863 struct xfs_mount
*mp
= tp
->t_mountp
;
864 xfs_lsn_t commit_lsn
= -1;
867 int sync
= tp
->t_flags
& XFS_TRANS_SYNC
;
870 * Determine whether this commit is releasing a permanent
871 * log reservation or not.
873 if (flags
& XFS_TRANS_RELEASE_LOG_RES
) {
874 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
875 log_flags
= XFS_LOG_REL_PERM_RESERV
;
879 * If there is nothing to be logged by the transaction,
880 * then unlock all of the items associated with the
881 * transaction and free the transaction structure.
882 * Also make sure to return any reserved blocks to
885 if (!(tp
->t_flags
& XFS_TRANS_DIRTY
))
888 if (XFS_FORCED_SHUTDOWN(mp
)) {
889 error
= XFS_ERROR(EIO
);
893 ASSERT(tp
->t_ticket
!= NULL
);
896 * If we need to update the superblock, then do it now.
898 if (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)
899 xfs_trans_apply_sb_deltas(tp
);
900 xfs_trans_apply_dquot_deltas(tp
);
902 error
= xfs_log_commit_cil(mp
, tp
, &commit_lsn
, flags
);
903 if (error
== ENOMEM
) {
904 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
905 error
= XFS_ERROR(EIO
);
909 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
913 * If the transaction needs to be synchronous, then force the
914 * log out now and wait for it.
918 error
= _xfs_log_force_lsn(mp
, commit_lsn
,
921 XFS_STATS_INC(xs_trans_sync
);
923 XFS_STATS_INC(xs_trans_async
);
929 xfs_trans_unreserve_and_mod_sb(tp
);
932 * It is indeed possible for the transaction to be not dirty but
933 * the dqinfo portion to be. All that means is that we have some
934 * (non-persistent) quota reservations that need to be unreserved.
936 xfs_trans_unreserve_and_mod_dquots(tp
);
938 commit_lsn
= xfs_log_done(mp
, tp
->t_ticket
, NULL
, log_flags
);
939 if (commit_lsn
== -1 && !error
)
940 error
= XFS_ERROR(EIO
);
942 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
943 xfs_trans_free_items(tp
, NULLCOMMITLSN
, error
? XFS_TRANS_ABORT
: 0);
946 XFS_STATS_INC(xs_trans_empty
);
951 * Unlock all of the transaction's items and free the transaction.
952 * The transaction must not have modified any of its items, because
953 * there is no way to restore them to their previous state.
955 * If the transaction has made a log reservation, make sure to release
964 xfs_mount_t
*mp
= tp
->t_mountp
;
967 * See if the caller is being too lazy to figure out if
968 * the transaction really needs an abort.
970 if ((flags
& XFS_TRANS_ABORT
) && !(tp
->t_flags
& XFS_TRANS_DIRTY
))
971 flags
&= ~XFS_TRANS_ABORT
;
973 * See if the caller is relying on us to shut down the
974 * filesystem. This happens in paths where we detect
975 * corruption and decide to give up.
977 if ((tp
->t_flags
& XFS_TRANS_DIRTY
) && !XFS_FORCED_SHUTDOWN(mp
)) {
978 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW
, mp
);
979 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
982 if (!(flags
& XFS_TRANS_ABORT
) && !XFS_FORCED_SHUTDOWN(mp
)) {
983 struct xfs_log_item_desc
*lidp
;
985 list_for_each_entry(lidp
, &tp
->t_items
, lid_trans
)
986 ASSERT(!(lidp
->lid_item
->li_type
== XFS_LI_EFD
));
989 xfs_trans_unreserve_and_mod_sb(tp
);
990 xfs_trans_unreserve_and_mod_dquots(tp
);
993 if (flags
& XFS_TRANS_RELEASE_LOG_RES
) {
994 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
995 log_flags
= XFS_LOG_REL_PERM_RESERV
;
999 xfs_log_done(mp
, tp
->t_ticket
, NULL
, log_flags
);
1002 /* mark this thread as no longer being in a transaction */
1003 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
1005 xfs_trans_free_items(tp
, NULLCOMMITLSN
, flags
);
1010 * Roll from one trans in the sequence of PERMANENT transactions to
1011 * the next: permanent transactions are only flushed out when
1012 * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
1013 * as possible to let chunks of it go to the log. So we commit the
1014 * chunk we've been working on and get a new transaction to continue.
1018 struct xfs_trans
**tpp
,
1019 struct xfs_inode
*dp
)
1021 struct xfs_trans
*trans
;
1022 struct xfs_trans_res tres
;
1026 * Ensure that the inode is always logged.
1029 xfs_trans_log_inode(trans
, dp
, XFS_ILOG_CORE
);
1032 * Copy the critical parameters from one trans to the next.
1034 tres
.tr_logres
= trans
->t_log_res
;
1035 tres
.tr_logcount
= trans
->t_log_count
;
1036 *tpp
= xfs_trans_dup(trans
);
1039 * Commit the current transaction.
1040 * If this commit failed, then it'd just unlock those items that
1041 * are not marked ihold. That also means that a filesystem shutdown
1042 * is in progress. The caller takes the responsibility to cancel
1043 * the duplicate transaction that gets returned.
1045 error
= xfs_trans_commit(trans
, 0);
1052 * transaction commit worked ok so we can drop the extra ticket
1053 * reference that we gained in xfs_trans_dup()
1055 xfs_log_ticket_put(trans
->t_ticket
);
1059 * Reserve space in the log for th next transaction.
1060 * This also pushes items in the "AIL", the list of logged items,
1061 * out to disk if they are taking up space at the tail of the log
1062 * that we want to use. This requires that either nothing be locked
1063 * across this call, or that anything that is locked be logged in
1064 * the prior and the next transactions.
1066 tres
.tr_logflags
= XFS_TRANS_PERM_LOG_RES
;
1067 error
= xfs_trans_reserve(trans
, &tres
, 0, 0);
1069 * Ensure that the inode is in the new transaction and locked.
1074 xfs_trans_ijoin(trans
, dp
, 0);