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"
25 #include "xfs_mount.h"
26 #include "xfs_inode.h"
27 #include "xfs_extent_busy.h"
28 #include "xfs_quota.h"
29 #include "xfs_trans.h"
30 #include "xfs_trans_priv.h"
32 #include "xfs_trace.h"
33 #include "xfs_error.h"
35 kmem_zone_t
*xfs_trans_zone
;
36 kmem_zone_t
*xfs_log_item_desc_zone
;
39 * Initialize the precomputed transaction reservation values
40 * in the mount structure.
46 xfs_trans_resv_calc(mp
, M_RES(mp
));
50 * Free the transaction structure. If there is more clean up
51 * to do when the structure is freed, add it here.
57 xfs_extent_busy_sort(&tp
->t_busy
);
58 xfs_extent_busy_clear(tp
->t_mountp
, &tp
->t_busy
, false);
60 atomic_dec(&tp
->t_mountp
->m_active_trans
);
61 if (!(tp
->t_flags
& XFS_TRANS_NO_WRITECOUNT
))
62 sb_end_intwrite(tp
->t_mountp
->m_super
);
63 xfs_trans_free_dqinfo(tp
);
64 kmem_zone_free(xfs_trans_zone
, tp
);
68 * This is called to create a new transaction which will share the
69 * permanent log reservation of the given transaction. The remaining
70 * unused block and rt extent reservations are also inherited. This
71 * implies that the original transaction is no longer allowed to allocate
72 * blocks. Locks and log items, however, are no inherited. They must
73 * be added to the new transaction explicitly.
81 ntp
= kmem_zone_zalloc(xfs_trans_zone
, KM_SLEEP
);
84 * Initialize the new transaction structure.
86 ntp
->t_magic
= XFS_TRANS_HEADER_MAGIC
;
87 ntp
->t_mountp
= tp
->t_mountp
;
88 INIT_LIST_HEAD(&ntp
->t_items
);
89 INIT_LIST_HEAD(&ntp
->t_busy
);
91 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
92 ASSERT(tp
->t_ticket
!= NULL
);
94 ntp
->t_flags
= XFS_TRANS_PERM_LOG_RES
|
95 (tp
->t_flags
& XFS_TRANS_RESERVE
) |
96 (tp
->t_flags
& XFS_TRANS_NO_WRITECOUNT
);
97 /* We gave our writer reference to the new transaction */
98 tp
->t_flags
|= XFS_TRANS_NO_WRITECOUNT
;
99 ntp
->t_ticket
= xfs_log_ticket_get(tp
->t_ticket
);
100 ntp
->t_blk_res
= tp
->t_blk_res
- tp
->t_blk_res_used
;
101 tp
->t_blk_res
= tp
->t_blk_res_used
;
102 ntp
->t_rtx_res
= tp
->t_rtx_res
- tp
->t_rtx_res_used
;
103 tp
->t_rtx_res
= tp
->t_rtx_res_used
;
104 ntp
->t_pflags
= tp
->t_pflags
;
106 xfs_trans_dup_dqinfo(tp
, ntp
);
108 atomic_inc(&tp
->t_mountp
->m_active_trans
);
113 * This is called to reserve free disk blocks and log space for the
114 * given transaction. This must be done before allocating any resources
115 * within the transaction.
117 * This will return ENOSPC if there are not enough blocks available.
118 * It will sleep waiting for available log space.
119 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
120 * is used by long running transactions. If any one of the reservations
121 * fails then they will all be backed out.
123 * This does not do quota reservations. That typically is done by the
128 struct xfs_trans
*tp
,
129 struct xfs_trans_res
*resp
,
134 bool rsvd
= (tp
->t_flags
& XFS_TRANS_RESERVE
) != 0;
136 /* Mark this thread as being in a transaction */
137 current_set_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
140 * Attempt to reserve the needed disk blocks by decrementing
141 * the number needed from the number available. This will
142 * fail if the count would go below zero.
145 error
= xfs_mod_fdblocks(tp
->t_mountp
, -((int64_t)blocks
), rsvd
);
147 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
150 tp
->t_blk_res
+= blocks
;
154 * Reserve the log space needed for this transaction.
156 if (resp
->tr_logres
> 0) {
157 bool permanent
= false;
159 ASSERT(tp
->t_log_res
== 0 ||
160 tp
->t_log_res
== resp
->tr_logres
);
161 ASSERT(tp
->t_log_count
== 0 ||
162 tp
->t_log_count
== resp
->tr_logcount
);
164 if (resp
->tr_logflags
& XFS_TRANS_PERM_LOG_RES
) {
165 tp
->t_flags
|= XFS_TRANS_PERM_LOG_RES
;
168 ASSERT(tp
->t_ticket
== NULL
);
169 ASSERT(!(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
));
172 if (tp
->t_ticket
!= NULL
) {
173 ASSERT(resp
->tr_logflags
& XFS_TRANS_PERM_LOG_RES
);
174 error
= xfs_log_regrant(tp
->t_mountp
, tp
->t_ticket
);
176 error
= xfs_log_reserve(tp
->t_mountp
,
179 &tp
->t_ticket
, XFS_TRANSACTION
,
186 tp
->t_log_res
= resp
->tr_logres
;
187 tp
->t_log_count
= resp
->tr_logcount
;
191 * Attempt to reserve the needed realtime extents by decrementing
192 * the number needed from the number available. This will
193 * fail if the count would go below zero.
196 error
= xfs_mod_frextents(tp
->t_mountp
, -((int64_t)rtextents
));
201 tp
->t_rtx_res
+= rtextents
;
207 * Error cases jump to one of these labels to undo any
208 * reservations which have already been performed.
211 if (resp
->tr_logres
> 0) {
212 xfs_log_done(tp
->t_mountp
, tp
->t_ticket
, NULL
, false);
215 tp
->t_flags
&= ~XFS_TRANS_PERM_LOG_RES
;
220 xfs_mod_fdblocks(tp
->t_mountp
, (int64_t)blocks
, rsvd
);
224 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
231 struct xfs_mount
*mp
,
232 struct xfs_trans_res
*resp
,
236 struct xfs_trans
**tpp
)
238 struct xfs_trans
*tp
;
241 if (!(flags
& XFS_TRANS_NO_WRITECOUNT
))
242 sb_start_intwrite(mp
->m_super
);
244 WARN_ON(mp
->m_super
->s_writers
.frozen
== SB_FREEZE_COMPLETE
);
245 atomic_inc(&mp
->m_active_trans
);
247 tp
= kmem_zone_zalloc(xfs_trans_zone
,
248 (flags
& XFS_TRANS_NOFS
) ? KM_NOFS
: KM_SLEEP
);
249 tp
->t_magic
= XFS_TRANS_HEADER_MAGIC
;
252 INIT_LIST_HEAD(&tp
->t_items
);
253 INIT_LIST_HEAD(&tp
->t_busy
);
255 error
= xfs_trans_reserve(tp
, resp
, blocks
, rtextents
);
257 xfs_trans_cancel(tp
);
266 * Record the indicated change to the given field for application
267 * to the file system's superblock when the transaction commits.
268 * For now, just store the change in the transaction structure.
270 * Mark the transaction structure to indicate that the superblock
271 * needs to be updated before committing.
273 * Because we may not be keeping track of allocated/free inodes and
274 * used filesystem blocks in the superblock, we do not mark the
275 * superblock dirty in this transaction if we modify these fields.
276 * We still need to update the transaction deltas so that they get
277 * applied to the incore superblock, but we don't want them to
278 * cause the superblock to get locked and logged if these are the
279 * only fields in the superblock that the transaction modifies.
287 uint32_t flags
= (XFS_TRANS_DIRTY
|XFS_TRANS_SB_DIRTY
);
288 xfs_mount_t
*mp
= tp
->t_mountp
;
291 case XFS_TRANS_SB_ICOUNT
:
292 tp
->t_icount_delta
+= delta
;
293 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
294 flags
&= ~XFS_TRANS_SB_DIRTY
;
296 case XFS_TRANS_SB_IFREE
:
297 tp
->t_ifree_delta
+= delta
;
298 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
299 flags
&= ~XFS_TRANS_SB_DIRTY
;
301 case XFS_TRANS_SB_FDBLOCKS
:
303 * Track the number of blocks allocated in the
304 * transaction. Make sure it does not exceed the
308 tp
->t_blk_res_used
+= (uint
)-delta
;
309 ASSERT(tp
->t_blk_res_used
<= tp
->t_blk_res
);
311 tp
->t_fdblocks_delta
+= delta
;
312 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
313 flags
&= ~XFS_TRANS_SB_DIRTY
;
315 case XFS_TRANS_SB_RES_FDBLOCKS
:
317 * The allocation has already been applied to the
318 * in-core superblock's counter. This should only
319 * be applied to the on-disk superblock.
321 tp
->t_res_fdblocks_delta
+= delta
;
322 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
323 flags
&= ~XFS_TRANS_SB_DIRTY
;
325 case XFS_TRANS_SB_FREXTENTS
:
327 * Track the number of blocks allocated in the
328 * transaction. Make sure it does not exceed the
332 tp
->t_rtx_res_used
+= (uint
)-delta
;
333 ASSERT(tp
->t_rtx_res_used
<= tp
->t_rtx_res
);
335 tp
->t_frextents_delta
+= delta
;
337 case XFS_TRANS_SB_RES_FREXTENTS
:
339 * The allocation has already been applied to the
340 * in-core superblock's counter. This should only
341 * be applied to the on-disk superblock.
344 tp
->t_res_frextents_delta
+= delta
;
346 case XFS_TRANS_SB_DBLOCKS
:
348 tp
->t_dblocks_delta
+= delta
;
350 case XFS_TRANS_SB_AGCOUNT
:
352 tp
->t_agcount_delta
+= delta
;
354 case XFS_TRANS_SB_IMAXPCT
:
355 tp
->t_imaxpct_delta
+= delta
;
357 case XFS_TRANS_SB_REXTSIZE
:
358 tp
->t_rextsize_delta
+= delta
;
360 case XFS_TRANS_SB_RBMBLOCKS
:
361 tp
->t_rbmblocks_delta
+= delta
;
363 case XFS_TRANS_SB_RBLOCKS
:
364 tp
->t_rblocks_delta
+= delta
;
366 case XFS_TRANS_SB_REXTENTS
:
367 tp
->t_rextents_delta
+= delta
;
369 case XFS_TRANS_SB_REXTSLOG
:
370 tp
->t_rextslog_delta
+= delta
;
377 tp
->t_flags
|= flags
;
381 * xfs_trans_apply_sb_deltas() is called from the commit code
382 * to bring the superblock buffer into the current transaction
383 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
385 * For now we just look at each field allowed to change and change
389 xfs_trans_apply_sb_deltas(
396 bp
= xfs_trans_getsb(tp
, tp
->t_mountp
, 0);
397 sbp
= XFS_BUF_TO_SBP(bp
);
400 * Check that superblock mods match the mods made to AGF counters.
402 ASSERT((tp
->t_fdblocks_delta
+ tp
->t_res_fdblocks_delta
) ==
403 (tp
->t_ag_freeblks_delta
+ tp
->t_ag_flist_delta
+
404 tp
->t_ag_btree_delta
));
407 * Only update the superblock counters if we are logging them
409 if (!xfs_sb_version_haslazysbcount(&(tp
->t_mountp
->m_sb
))) {
410 if (tp
->t_icount_delta
)
411 be64_add_cpu(&sbp
->sb_icount
, tp
->t_icount_delta
);
412 if (tp
->t_ifree_delta
)
413 be64_add_cpu(&sbp
->sb_ifree
, tp
->t_ifree_delta
);
414 if (tp
->t_fdblocks_delta
)
415 be64_add_cpu(&sbp
->sb_fdblocks
, tp
->t_fdblocks_delta
);
416 if (tp
->t_res_fdblocks_delta
)
417 be64_add_cpu(&sbp
->sb_fdblocks
, tp
->t_res_fdblocks_delta
);
420 if (tp
->t_frextents_delta
)
421 be64_add_cpu(&sbp
->sb_frextents
, tp
->t_frextents_delta
);
422 if (tp
->t_res_frextents_delta
)
423 be64_add_cpu(&sbp
->sb_frextents
, tp
->t_res_frextents_delta
);
425 if (tp
->t_dblocks_delta
) {
426 be64_add_cpu(&sbp
->sb_dblocks
, tp
->t_dblocks_delta
);
429 if (tp
->t_agcount_delta
) {
430 be32_add_cpu(&sbp
->sb_agcount
, tp
->t_agcount_delta
);
433 if (tp
->t_imaxpct_delta
) {
434 sbp
->sb_imax_pct
+= tp
->t_imaxpct_delta
;
437 if (tp
->t_rextsize_delta
) {
438 be32_add_cpu(&sbp
->sb_rextsize
, tp
->t_rextsize_delta
);
441 if (tp
->t_rbmblocks_delta
) {
442 be32_add_cpu(&sbp
->sb_rbmblocks
, tp
->t_rbmblocks_delta
);
445 if (tp
->t_rblocks_delta
) {
446 be64_add_cpu(&sbp
->sb_rblocks
, tp
->t_rblocks_delta
);
449 if (tp
->t_rextents_delta
) {
450 be64_add_cpu(&sbp
->sb_rextents
, tp
->t_rextents_delta
);
453 if (tp
->t_rextslog_delta
) {
454 sbp
->sb_rextslog
+= tp
->t_rextslog_delta
;
458 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_SB_BUF
);
461 * Log the whole thing, the fields are noncontiguous.
463 xfs_trans_log_buf(tp
, bp
, 0, sizeof(xfs_dsb_t
) - 1);
466 * Since all the modifiable fields are contiguous, we
467 * can get away with this.
469 xfs_trans_log_buf(tp
, bp
, offsetof(xfs_dsb_t
, sb_icount
),
470 offsetof(xfs_dsb_t
, sb_frextents
) +
471 sizeof(sbp
->sb_frextents
) - 1);
479 int8_t counter
= *field
;
495 int32_t counter
= *field
;
511 int64_t counter
= *field
;
523 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
524 * and apply superblock counter changes to the in-core superblock. The
525 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
526 * applied to the in-core superblock. The idea is that that has already been
529 * If we are not logging superblock counters, then the inode allocated/free and
530 * used block counts are not updated in the on disk superblock. In this case,
531 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
532 * still need to update the incore superblock with the changes.
535 xfs_trans_unreserve_and_mod_sb(
536 struct xfs_trans
*tp
)
538 struct xfs_mount
*mp
= tp
->t_mountp
;
539 bool rsvd
= (tp
->t_flags
& XFS_TRANS_RESERVE
) != 0;
540 int64_t blkdelta
= 0;
541 int64_t rtxdelta
= 0;
543 int64_t ifreedelta
= 0;
546 /* calculate deltas */
547 if (tp
->t_blk_res
> 0)
548 blkdelta
= tp
->t_blk_res
;
549 if ((tp
->t_fdblocks_delta
!= 0) &&
550 (xfs_sb_version_haslazysbcount(&mp
->m_sb
) ||
551 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)))
552 blkdelta
+= tp
->t_fdblocks_delta
;
554 if (tp
->t_rtx_res
> 0)
555 rtxdelta
= tp
->t_rtx_res
;
556 if ((tp
->t_frextents_delta
!= 0) &&
557 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
))
558 rtxdelta
+= tp
->t_frextents_delta
;
560 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) ||
561 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)) {
562 idelta
= tp
->t_icount_delta
;
563 ifreedelta
= tp
->t_ifree_delta
;
566 /* apply the per-cpu counters */
568 error
= xfs_mod_fdblocks(mp
, blkdelta
, rsvd
);
574 error
= xfs_mod_icount(mp
, idelta
);
576 goto out_undo_fdblocks
;
580 error
= xfs_mod_ifree(mp
, ifreedelta
);
582 goto out_undo_icount
;
585 if (rtxdelta
== 0 && !(tp
->t_flags
& XFS_TRANS_SB_DIRTY
))
588 /* apply remaining deltas */
589 spin_lock(&mp
->m_sb_lock
);
591 error
= xfs_sb_mod64(&mp
->m_sb
.sb_frextents
, rtxdelta
);
596 if (tp
->t_dblocks_delta
!= 0) {
597 error
= xfs_sb_mod64(&mp
->m_sb
.sb_dblocks
, tp
->t_dblocks_delta
);
599 goto out_undo_frextents
;
601 if (tp
->t_agcount_delta
!= 0) {
602 error
= xfs_sb_mod32(&mp
->m_sb
.sb_agcount
, tp
->t_agcount_delta
);
604 goto out_undo_dblocks
;
606 if (tp
->t_imaxpct_delta
!= 0) {
607 error
= xfs_sb_mod8(&mp
->m_sb
.sb_imax_pct
, tp
->t_imaxpct_delta
);
609 goto out_undo_agcount
;
611 if (tp
->t_rextsize_delta
!= 0) {
612 error
= xfs_sb_mod32(&mp
->m_sb
.sb_rextsize
,
613 tp
->t_rextsize_delta
);
615 goto out_undo_imaxpct
;
617 if (tp
->t_rbmblocks_delta
!= 0) {
618 error
= xfs_sb_mod32(&mp
->m_sb
.sb_rbmblocks
,
619 tp
->t_rbmblocks_delta
);
621 goto out_undo_rextsize
;
623 if (tp
->t_rblocks_delta
!= 0) {
624 error
= xfs_sb_mod64(&mp
->m_sb
.sb_rblocks
, tp
->t_rblocks_delta
);
626 goto out_undo_rbmblocks
;
628 if (tp
->t_rextents_delta
!= 0) {
629 error
= xfs_sb_mod64(&mp
->m_sb
.sb_rextents
,
630 tp
->t_rextents_delta
);
632 goto out_undo_rblocks
;
634 if (tp
->t_rextslog_delta
!= 0) {
635 error
= xfs_sb_mod8(&mp
->m_sb
.sb_rextslog
,
636 tp
->t_rextslog_delta
);
638 goto out_undo_rextents
;
640 spin_unlock(&mp
->m_sb_lock
);
644 if (tp
->t_rextents_delta
)
645 xfs_sb_mod64(&mp
->m_sb
.sb_rextents
, -tp
->t_rextents_delta
);
647 if (tp
->t_rblocks_delta
)
648 xfs_sb_mod64(&mp
->m_sb
.sb_rblocks
, -tp
->t_rblocks_delta
);
650 if (tp
->t_rbmblocks_delta
)
651 xfs_sb_mod32(&mp
->m_sb
.sb_rbmblocks
, -tp
->t_rbmblocks_delta
);
653 if (tp
->t_rextsize_delta
)
654 xfs_sb_mod32(&mp
->m_sb
.sb_rextsize
, -tp
->t_rextsize_delta
);
656 if (tp
->t_rextsize_delta
)
657 xfs_sb_mod8(&mp
->m_sb
.sb_imax_pct
, -tp
->t_imaxpct_delta
);
659 if (tp
->t_agcount_delta
)
660 xfs_sb_mod32(&mp
->m_sb
.sb_agcount
, -tp
->t_agcount_delta
);
662 if (tp
->t_dblocks_delta
)
663 xfs_sb_mod64(&mp
->m_sb
.sb_dblocks
, -tp
->t_dblocks_delta
);
666 xfs_sb_mod64(&mp
->m_sb
.sb_frextents
, -rtxdelta
);
668 spin_unlock(&mp
->m_sb_lock
);
670 xfs_mod_ifree(mp
, -ifreedelta
);
673 xfs_mod_icount(mp
, -idelta
);
676 xfs_mod_fdblocks(mp
, -blkdelta
, rsvd
);
683 * Add the given log item to the transaction's list of log items.
685 * The log item will now point to its new descriptor with its li_desc field.
689 struct xfs_trans
*tp
,
690 struct xfs_log_item
*lip
)
692 struct xfs_log_item_desc
*lidp
;
694 ASSERT(lip
->li_mountp
== tp
->t_mountp
);
695 ASSERT(lip
->li_ailp
== tp
->t_mountp
->m_ail
);
697 lidp
= kmem_zone_zalloc(xfs_log_item_desc_zone
, KM_SLEEP
| KM_NOFS
);
699 lidp
->lid_item
= lip
;
701 list_add_tail(&lidp
->lid_trans
, &tp
->t_items
);
707 xfs_trans_free_item_desc(
708 struct xfs_log_item_desc
*lidp
)
710 list_del_init(&lidp
->lid_trans
);
711 kmem_zone_free(xfs_log_item_desc_zone
, lidp
);
715 * Unlink and free the given descriptor.
719 struct xfs_log_item
*lip
)
721 xfs_trans_free_item_desc(lip
->li_desc
);
726 * Unlock all of the items of a transaction and free all the descriptors
727 * of that transaction.
730 xfs_trans_free_items(
731 struct xfs_trans
*tp
,
732 xfs_lsn_t commit_lsn
,
735 struct xfs_log_item_desc
*lidp
, *next
;
737 list_for_each_entry_safe(lidp
, next
, &tp
->t_items
, lid_trans
) {
738 struct xfs_log_item
*lip
= lidp
->lid_item
;
742 if (commit_lsn
!= NULLCOMMITLSN
)
743 lip
->li_ops
->iop_committing(lip
, commit_lsn
);
745 lip
->li_flags
|= XFS_LI_ABORTED
;
746 lip
->li_ops
->iop_unlock(lip
);
748 xfs_trans_free_item_desc(lidp
);
753 xfs_log_item_batch_insert(
754 struct xfs_ail
*ailp
,
755 struct xfs_ail_cursor
*cur
,
756 struct xfs_log_item
**log_items
,
758 xfs_lsn_t commit_lsn
)
762 spin_lock(&ailp
->xa_lock
);
763 /* xfs_trans_ail_update_bulk drops ailp->xa_lock */
764 xfs_trans_ail_update_bulk(ailp
, cur
, log_items
, nr_items
, commit_lsn
);
766 for (i
= 0; i
< nr_items
; i
++) {
767 struct xfs_log_item
*lip
= log_items
[i
];
769 lip
->li_ops
->iop_unpin(lip
, 0);
774 * Bulk operation version of xfs_trans_committed that takes a log vector of
775 * items to insert into the AIL. This uses bulk AIL insertion techniques to
776 * minimise lock traffic.
778 * If we are called with the aborted flag set, it is because a log write during
779 * a CIL checkpoint commit has failed. In this case, all the items in the
780 * checkpoint have already gone through iop_commited and iop_unlock, which
781 * means that checkpoint commit abort handling is treated exactly the same
782 * as an iclog write error even though we haven't started any IO yet. Hence in
783 * this case all we need to do is iop_committed processing, followed by an
784 * iop_unpin(aborted) call.
786 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
787 * at the end of the AIL, the insert cursor avoids the need to walk
788 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
789 * call. This saves a lot of needless list walking and is a net win, even
790 * though it slightly increases that amount of AIL lock traffic to set it up
794 xfs_trans_committed_bulk(
795 struct xfs_ail
*ailp
,
796 struct xfs_log_vec
*log_vector
,
797 xfs_lsn_t commit_lsn
,
800 #define LOG_ITEM_BATCH_SIZE 32
801 struct xfs_log_item
*log_items
[LOG_ITEM_BATCH_SIZE
];
802 struct xfs_log_vec
*lv
;
803 struct xfs_ail_cursor cur
;
806 spin_lock(&ailp
->xa_lock
);
807 xfs_trans_ail_cursor_last(ailp
, &cur
, commit_lsn
);
808 spin_unlock(&ailp
->xa_lock
);
810 /* unpin all the log items */
811 for (lv
= log_vector
; lv
; lv
= lv
->lv_next
) {
812 struct xfs_log_item
*lip
= lv
->lv_item
;
816 lip
->li_flags
|= XFS_LI_ABORTED
;
817 item_lsn
= lip
->li_ops
->iop_committed(lip
, commit_lsn
);
819 /* item_lsn of -1 means the item needs no further processing */
820 if (XFS_LSN_CMP(item_lsn
, (xfs_lsn_t
)-1) == 0)
824 * if we are aborting the operation, no point in inserting the
825 * object into the AIL as we are in a shutdown situation.
828 ASSERT(XFS_FORCED_SHUTDOWN(ailp
->xa_mount
));
829 lip
->li_ops
->iop_unpin(lip
, 1);
833 if (item_lsn
!= commit_lsn
) {
836 * Not a bulk update option due to unusual item_lsn.
837 * Push into AIL immediately, rechecking the lsn once
838 * we have the ail lock. Then unpin the item. This does
839 * not affect the AIL cursor the bulk insert path is
842 spin_lock(&ailp
->xa_lock
);
843 if (XFS_LSN_CMP(item_lsn
, lip
->li_lsn
) > 0)
844 xfs_trans_ail_update(ailp
, lip
, item_lsn
);
846 spin_unlock(&ailp
->xa_lock
);
847 lip
->li_ops
->iop_unpin(lip
, 0);
851 /* Item is a candidate for bulk AIL insert. */
852 log_items
[i
++] = lv
->lv_item
;
853 if (i
>= LOG_ITEM_BATCH_SIZE
) {
854 xfs_log_item_batch_insert(ailp
, &cur
, log_items
,
855 LOG_ITEM_BATCH_SIZE
, commit_lsn
);
860 /* make sure we insert the remainder! */
862 xfs_log_item_batch_insert(ailp
, &cur
, log_items
, i
, commit_lsn
);
864 spin_lock(&ailp
->xa_lock
);
865 xfs_trans_ail_cursor_done(&cur
);
866 spin_unlock(&ailp
->xa_lock
);
870 * Commit the given transaction to the log.
872 * XFS disk error handling mechanism is not based on a typical
873 * transaction abort mechanism. Logically after the filesystem
874 * gets marked 'SHUTDOWN', we can't let any new transactions
875 * be durable - ie. committed to disk - because some metadata might
876 * be inconsistent. In such cases, this returns an error, and the
877 * caller may assume that all locked objects joined to the transaction
878 * have already been unlocked as if the commit had succeeded.
879 * Do not reference the transaction structure after this call.
883 struct xfs_trans
*tp
,
886 struct xfs_mount
*mp
= tp
->t_mountp
;
887 xfs_lsn_t commit_lsn
= -1;
889 int sync
= tp
->t_flags
& XFS_TRANS_SYNC
;
892 * If there is nothing to be logged by the transaction,
893 * then unlock all of the items associated with the
894 * transaction and free the transaction structure.
895 * Also make sure to return any reserved blocks to
898 if (!(tp
->t_flags
& XFS_TRANS_DIRTY
))
901 if (XFS_FORCED_SHUTDOWN(mp
)) {
906 ASSERT(tp
->t_ticket
!= NULL
);
909 * If we need to update the superblock, then do it now.
911 if (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)
912 xfs_trans_apply_sb_deltas(tp
);
913 xfs_trans_apply_dquot_deltas(tp
);
915 xfs_log_commit_cil(mp
, tp
, &commit_lsn
, regrant
);
917 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
921 * If the transaction needs to be synchronous, then force the
922 * log out now and wait for it.
925 error
= _xfs_log_force_lsn(mp
, commit_lsn
, XFS_LOG_SYNC
, NULL
);
926 XFS_STATS_INC(mp
, xs_trans_sync
);
928 XFS_STATS_INC(mp
, xs_trans_async
);
934 xfs_trans_unreserve_and_mod_sb(tp
);
937 * It is indeed possible for the transaction to be not dirty but
938 * the dqinfo portion to be. All that means is that we have some
939 * (non-persistent) quota reservations that need to be unreserved.
941 xfs_trans_unreserve_and_mod_dquots(tp
);
943 commit_lsn
= xfs_log_done(mp
, tp
->t_ticket
, NULL
, regrant
);
944 if (commit_lsn
== -1 && !error
)
947 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
948 xfs_trans_free_items(tp
, NULLCOMMITLSN
, !!error
);
951 XFS_STATS_INC(mp
, xs_trans_empty
);
957 struct xfs_trans
*tp
)
959 return __xfs_trans_commit(tp
, false);
963 * Unlock all of the transaction's items and free the transaction.
964 * The transaction must not have modified any of its items, because
965 * there is no way to restore them to their previous state.
967 * If the transaction has made a log reservation, make sure to release
972 struct xfs_trans
*tp
)
974 struct xfs_mount
*mp
= tp
->t_mountp
;
975 bool dirty
= (tp
->t_flags
& XFS_TRANS_DIRTY
);
978 * See if the caller is relying on us to shut down the
979 * filesystem. This happens in paths where we detect
980 * corruption and decide to give up.
982 if (dirty
&& !XFS_FORCED_SHUTDOWN(mp
)) {
983 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW
, mp
);
984 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
987 if (!dirty
&& !XFS_FORCED_SHUTDOWN(mp
)) {
988 struct xfs_log_item_desc
*lidp
;
990 list_for_each_entry(lidp
, &tp
->t_items
, lid_trans
)
991 ASSERT(!(lidp
->lid_item
->li_type
== XFS_LI_EFD
));
994 xfs_trans_unreserve_and_mod_sb(tp
);
995 xfs_trans_unreserve_and_mod_dquots(tp
);
998 xfs_log_done(mp
, tp
->t_ticket
, NULL
, false);
1000 /* mark this thread as no longer being in a transaction */
1001 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
1003 xfs_trans_free_items(tp
, NULLCOMMITLSN
, dirty
);
1008 * Roll from one trans in the sequence of PERMANENT transactions to
1009 * the next: permanent transactions are only flushed out when
1010 * committed with xfs_trans_commit(), but we still want as soon
1011 * as possible to let chunks of it go to the log. So we commit the
1012 * chunk we've been working on and get a new transaction to continue.
1016 struct xfs_trans
**tpp
,
1017 struct xfs_inode
*dp
,
1020 struct xfs_trans
*trans
;
1021 struct xfs_trans_res tres
;
1027 * 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
, true);
1055 * Reserve space in the log for th next transaction.
1056 * This also pushes items in the "AIL", the list of logged items,
1057 * out to disk if they are taking up space at the tail of the log
1058 * that we want to use. This requires that either nothing be locked
1059 * across this call, or that anything that is locked be logged in
1060 * the prior and the next transactions.
1062 tres
.tr_logflags
= XFS_TRANS_PERM_LOG_RES
;
1063 error
= xfs_trans_reserve(trans
, &tres
, 0, 0);
1065 * Ensure that the inode is in the new transaction and locked.
1071 xfs_trans_ijoin(trans
, dp
, 0);
1077 struct xfs_trans
**tpp
,
1078 struct xfs_inode
*dp
)
1081 return __xfs_trans_roll(tpp
, dp
, &committed
);