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"
34 #include "xfs_defer.h"
36 kmem_zone_t *xfs_trans_zone;
37 kmem_zone_t *xfs_log_item_desc_zone;
39 #if defined(CONFIG_TRACEPOINTS)
41 xfs_trans_trace_reservations(
44 struct xfs_trans_res resv;
45 struct xfs_trans_res *res;
46 struct xfs_trans_res *end_res;
49 res = (struct xfs_trans_res *)M_RES(mp);
50 end_res = (struct xfs_trans_res *)(M_RES(mp) + 1);
51 for (i = 0; res < end_res; i++, res++)
52 trace_xfs_trans_resv_calc(mp, i, res);
53 xfs_log_get_max_trans_res(mp, &resv);
54 trace_xfs_trans_resv_calc(mp, -1, &resv);
57 # define xfs_trans_trace_reservations(mp)
61 * Initialize the precomputed transaction reservation values
62 * in the mount structure.
68 xfs_trans_resv_calc(mp, M_RES(mp));
69 xfs_trans_trace_reservations(mp);
73 * Free the transaction structure. If there is more clean up
74 * to do when the structure is freed, add it here.
80 xfs_extent_busy_sort(&tp->t_busy);
81 xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
83 trace_xfs_trans_free(tp, _RET_IP_);
84 atomic_dec(&tp->t_mountp->m_active_trans);
85 if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
86 sb_end_intwrite(tp->t_mountp->m_super);
87 xfs_trans_free_dqinfo(tp);
88 kmem_zone_free(xfs_trans_zone, tp);
92 * This is called to create a new transaction which will share the
93 * permanent log reservation of the given transaction. The remaining
94 * unused block and rt extent reservations are also inherited. This
95 * implies that the original transaction is no longer allowed to allocate
96 * blocks. Locks and log items, however, are no inherited. They must
97 * be added to the new transaction explicitly.
99 STATIC struct xfs_trans *
101 struct xfs_trans *tp)
103 struct xfs_trans *ntp;
105 trace_xfs_trans_dup(tp, _RET_IP_);
107 ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
110 * Initialize the new transaction structure.
112 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
113 ntp->t_mountp = tp->t_mountp;
114 INIT_LIST_HEAD(&ntp->t_items);
115 INIT_LIST_HEAD(&ntp->t_busy);
117 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
118 ASSERT(tp->t_ticket != NULL);
120 ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
121 (tp->t_flags & XFS_TRANS_RESERVE) |
122 (tp->t_flags & XFS_TRANS_NO_WRITECOUNT);
123 /* We gave our writer reference to the new transaction */
124 tp->t_flags |= XFS_TRANS_NO_WRITECOUNT;
125 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
127 ASSERT(tp->t_blk_res >= tp->t_blk_res_used);
128 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
129 tp->t_blk_res = tp->t_blk_res_used;
131 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
132 tp->t_rtx_res = tp->t_rtx_res_used;
133 ntp->t_pflags = tp->t_pflags;
134 ntp->t_agfl_dfops = tp->t_agfl_dfops;
136 xfs_trans_dup_dqinfo(tp, ntp);
138 atomic_inc(&tp->t_mountp->m_active_trans);
143 * This is called to reserve free disk blocks and log space for the
144 * given transaction. This must be done before allocating any resources
145 * within the transaction.
147 * This will return ENOSPC if there are not enough blocks available.
148 * It will sleep waiting for available log space.
149 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
150 * is used by long running transactions. If any one of the reservations
151 * fails then they will all be backed out.
153 * This does not do quota reservations. That typically is done by the
158 struct xfs_trans *tp,
159 struct xfs_trans_res *resp,
164 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
166 /* Mark this thread as being in a transaction */
167 current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
170 * Attempt to reserve the needed disk blocks by decrementing
171 * the number needed from the number available. This will
172 * fail if the count would go below zero.
175 error = xfs_mod_fdblocks(tp->t_mountp, -((int64_t)blocks), rsvd);
177 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
180 tp->t_blk_res += blocks;
184 * Reserve the log space needed for this transaction.
186 if (resp->tr_logres > 0) {
187 bool permanent = false;
189 ASSERT(tp->t_log_res == 0 ||
190 tp->t_log_res == resp->tr_logres);
191 ASSERT(tp->t_log_count == 0 ||
192 tp->t_log_count == resp->tr_logcount);
194 if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
195 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
198 ASSERT(tp->t_ticket == NULL);
199 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
202 if (tp->t_ticket != NULL) {
203 ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
204 error = xfs_log_regrant(tp->t_mountp, tp->t_ticket);
206 error = xfs_log_reserve(tp->t_mountp,
209 &tp->t_ticket, XFS_TRANSACTION,
216 tp->t_log_res = resp->tr_logres;
217 tp->t_log_count = resp->tr_logcount;
221 * Attempt to reserve the needed realtime extents by decrementing
222 * the number needed from the number available. This will
223 * fail if the count would go below zero.
226 error = xfs_mod_frextents(tp->t_mountp, -((int64_t)rtextents));
231 tp->t_rtx_res += rtextents;
237 * Error cases jump to one of these labels to undo any
238 * reservations which have already been performed.
241 if (resp->tr_logres > 0) {
242 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, false);
245 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
250 xfs_mod_fdblocks(tp->t_mountp, (int64_t)blocks, rsvd);
254 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
261 struct xfs_mount *mp,
262 struct xfs_trans_res *resp,
266 struct xfs_trans **tpp)
268 struct xfs_trans *tp;
271 if (!(flags & XFS_TRANS_NO_WRITECOUNT))
272 sb_start_intwrite(mp->m_super);
274 WARN_ON(mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
275 atomic_inc(&mp->m_active_trans);
277 tp = kmem_zone_zalloc(xfs_trans_zone,
278 (flags & XFS_TRANS_NOFS) ? KM_NOFS : KM_SLEEP);
279 tp->t_magic = XFS_TRANS_HEADER_MAGIC;
282 INIT_LIST_HEAD(&tp->t_items);
283 INIT_LIST_HEAD(&tp->t_busy);
285 error = xfs_trans_reserve(tp, resp, blocks, rtextents);
287 xfs_trans_cancel(tp);
291 trace_xfs_trans_alloc(tp, _RET_IP_);
298 * Create an empty transaction with no reservation. This is a defensive
299 * mechanism for routines that query metadata without actually modifying
300 * them -- if the metadata being queried is somehow cross-linked (think a
301 * btree block pointer that points higher in the tree), we risk deadlock.
302 * However, blocks grabbed as part of a transaction can be re-grabbed.
303 * The verifiers will notice the corrupt block and the operation will fail
304 * back to userspace without deadlocking.
306 * Note the zero-length reservation; this transaction MUST be cancelled
307 * without any dirty data.
310 xfs_trans_alloc_empty(
311 struct xfs_mount *mp,
312 struct xfs_trans **tpp)
314 struct xfs_trans_res resv = {0};
316 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
320 * Record the indicated change to the given field for application
321 * to the file system's superblock when the transaction commits.
322 * For now, just store the change in the transaction structure.
324 * Mark the transaction structure to indicate that the superblock
325 * needs to be updated before committing.
327 * Because we may not be keeping track of allocated/free inodes and
328 * used filesystem blocks in the superblock, we do not mark the
329 * superblock dirty in this transaction if we modify these fields.
330 * We still need to update the transaction deltas so that they get
331 * applied to the incore superblock, but we don't want them to
332 * cause the superblock to get locked and logged if these are the
333 * only fields in the superblock that the transaction modifies.
341 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
342 xfs_mount_t *mp = tp->t_mountp;
345 case XFS_TRANS_SB_ICOUNT:
346 tp->t_icount_delta += delta;
347 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
348 flags &= ~XFS_TRANS_SB_DIRTY;
350 case XFS_TRANS_SB_IFREE:
351 tp->t_ifree_delta += delta;
352 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
353 flags &= ~XFS_TRANS_SB_DIRTY;
355 case XFS_TRANS_SB_FDBLOCKS:
357 * Track the number of blocks allocated in the transaction.
358 * Make sure it does not exceed the number reserved. If so,
359 * shutdown as this can lead to accounting inconsistency.
362 tp->t_blk_res_used += (uint)-delta;
363 if (tp->t_blk_res_used > tp->t_blk_res)
364 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
366 tp->t_fdblocks_delta += delta;
367 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
368 flags &= ~XFS_TRANS_SB_DIRTY;
370 case XFS_TRANS_SB_RES_FDBLOCKS:
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.
376 tp->t_res_fdblocks_delta += delta;
377 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
378 flags &= ~XFS_TRANS_SB_DIRTY;
380 case XFS_TRANS_SB_FREXTENTS:
382 * Track the number of blocks allocated in the
383 * transaction. Make sure it does not exceed the
387 tp->t_rtx_res_used += (uint)-delta;
388 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
390 tp->t_frextents_delta += delta;
392 case XFS_TRANS_SB_RES_FREXTENTS:
394 * The allocation has already been applied to the
395 * in-core superblock's counter. This should only
396 * be applied to the on-disk superblock.
399 tp->t_res_frextents_delta += delta;
401 case XFS_TRANS_SB_DBLOCKS:
403 tp->t_dblocks_delta += delta;
405 case XFS_TRANS_SB_AGCOUNT:
407 tp->t_agcount_delta += delta;
409 case XFS_TRANS_SB_IMAXPCT:
410 tp->t_imaxpct_delta += delta;
412 case XFS_TRANS_SB_REXTSIZE:
413 tp->t_rextsize_delta += delta;
415 case XFS_TRANS_SB_RBMBLOCKS:
416 tp->t_rbmblocks_delta += delta;
418 case XFS_TRANS_SB_RBLOCKS:
419 tp->t_rblocks_delta += delta;
421 case XFS_TRANS_SB_REXTENTS:
422 tp->t_rextents_delta += delta;
424 case XFS_TRANS_SB_REXTSLOG:
425 tp->t_rextslog_delta += delta;
432 tp->t_flags |= flags;
436 * xfs_trans_apply_sb_deltas() is called from the commit code
437 * to bring the superblock buffer into the current transaction
438 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
440 * For now we just look at each field allowed to change and change
444 xfs_trans_apply_sb_deltas(
451 bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
452 sbp = XFS_BUF_TO_SBP(bp);
455 * Check that superblock mods match the mods made to AGF counters.
457 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
458 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
459 tp->t_ag_btree_delta));
462 * Only update the superblock counters if we are logging them
464 if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
465 if (tp->t_icount_delta)
466 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
467 if (tp->t_ifree_delta)
468 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
469 if (tp->t_fdblocks_delta)
470 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
471 if (tp->t_res_fdblocks_delta)
472 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
475 if (tp->t_frextents_delta)
476 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
477 if (tp->t_res_frextents_delta)
478 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
480 if (tp->t_dblocks_delta) {
481 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
484 if (tp->t_agcount_delta) {
485 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
488 if (tp->t_imaxpct_delta) {
489 sbp->sb_imax_pct += tp->t_imaxpct_delta;
492 if (tp->t_rextsize_delta) {
493 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
496 if (tp->t_rbmblocks_delta) {
497 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
500 if (tp->t_rblocks_delta) {
501 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
504 if (tp->t_rextents_delta) {
505 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
508 if (tp->t_rextslog_delta) {
509 sbp->sb_rextslog += tp->t_rextslog_delta;
513 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
516 * Log the whole thing, the fields are noncontiguous.
518 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
521 * Since all the modifiable fields are contiguous, we
522 * can get away with this.
524 xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
525 offsetof(xfs_dsb_t, sb_frextents) +
526 sizeof(sbp->sb_frextents) - 1);
534 int8_t counter = *field;
550 int32_t counter = *field;
566 int64_t counter = *field;
578 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
579 * and apply superblock counter changes to the in-core superblock. The
580 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
581 * applied to the in-core superblock. The idea is that that has already been
584 * If we are not logging superblock counters, then the inode allocated/free and
585 * used block counts are not updated in the on disk superblock. In this case,
586 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
587 * still need to update the incore superblock with the changes.
590 xfs_trans_unreserve_and_mod_sb(
591 struct xfs_trans *tp)
593 struct xfs_mount *mp = tp->t_mountp;
594 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
595 int64_t blkdelta = 0;
596 int64_t rtxdelta = 0;
598 int64_t ifreedelta = 0;
601 /* calculate deltas */
602 if (tp->t_blk_res > 0)
603 blkdelta = tp->t_blk_res;
604 if ((tp->t_fdblocks_delta != 0) &&
605 (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
606 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
607 blkdelta += tp->t_fdblocks_delta;
609 if (tp->t_rtx_res > 0)
610 rtxdelta = tp->t_rtx_res;
611 if ((tp->t_frextents_delta != 0) &&
612 (tp->t_flags & XFS_TRANS_SB_DIRTY))
613 rtxdelta += tp->t_frextents_delta;
615 if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
616 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
617 idelta = tp->t_icount_delta;
618 ifreedelta = tp->t_ifree_delta;
621 /* apply the per-cpu counters */
623 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
629 error = xfs_mod_icount(mp, idelta);
631 goto out_undo_fdblocks;
635 error = xfs_mod_ifree(mp, ifreedelta);
637 goto out_undo_icount;
640 if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
643 /* apply remaining deltas */
644 spin_lock(&mp->m_sb_lock);
646 error = xfs_sb_mod64(&mp->m_sb.sb_frextents, rtxdelta);
651 if (tp->t_dblocks_delta != 0) {
652 error = xfs_sb_mod64(&mp->m_sb.sb_dblocks, tp->t_dblocks_delta);
654 goto out_undo_frextents;
656 if (tp->t_agcount_delta != 0) {
657 error = xfs_sb_mod32(&mp->m_sb.sb_agcount, tp->t_agcount_delta);
659 goto out_undo_dblocks;
661 if (tp->t_imaxpct_delta != 0) {
662 error = xfs_sb_mod8(&mp->m_sb.sb_imax_pct, tp->t_imaxpct_delta);
664 goto out_undo_agcount;
666 if (tp->t_rextsize_delta != 0) {
667 error = xfs_sb_mod32(&mp->m_sb.sb_rextsize,
668 tp->t_rextsize_delta);
670 goto out_undo_imaxpct;
672 if (tp->t_rbmblocks_delta != 0) {
673 error = xfs_sb_mod32(&mp->m_sb.sb_rbmblocks,
674 tp->t_rbmblocks_delta);
676 goto out_undo_rextsize;
678 if (tp->t_rblocks_delta != 0) {
679 error = xfs_sb_mod64(&mp->m_sb.sb_rblocks, tp->t_rblocks_delta);
681 goto out_undo_rbmblocks;
683 if (tp->t_rextents_delta != 0) {
684 error = xfs_sb_mod64(&mp->m_sb.sb_rextents,
685 tp->t_rextents_delta);
687 goto out_undo_rblocks;
689 if (tp->t_rextslog_delta != 0) {
690 error = xfs_sb_mod8(&mp->m_sb.sb_rextslog,
691 tp->t_rextslog_delta);
693 goto out_undo_rextents;
695 spin_unlock(&mp->m_sb_lock);
699 if (tp->t_rextents_delta)
700 xfs_sb_mod64(&mp->m_sb.sb_rextents, -tp->t_rextents_delta);
702 if (tp->t_rblocks_delta)
703 xfs_sb_mod64(&mp->m_sb.sb_rblocks, -tp->t_rblocks_delta);
705 if (tp->t_rbmblocks_delta)
706 xfs_sb_mod32(&mp->m_sb.sb_rbmblocks, -tp->t_rbmblocks_delta);
708 if (tp->t_rextsize_delta)
709 xfs_sb_mod32(&mp->m_sb.sb_rextsize, -tp->t_rextsize_delta);
711 if (tp->t_rextsize_delta)
712 xfs_sb_mod8(&mp->m_sb.sb_imax_pct, -tp->t_imaxpct_delta);
714 if (tp->t_agcount_delta)
715 xfs_sb_mod32(&mp->m_sb.sb_agcount, -tp->t_agcount_delta);
717 if (tp->t_dblocks_delta)
718 xfs_sb_mod64(&mp->m_sb.sb_dblocks, -tp->t_dblocks_delta);
721 xfs_sb_mod64(&mp->m_sb.sb_frextents, -rtxdelta);
723 spin_unlock(&mp->m_sb_lock);
725 xfs_mod_ifree(mp, -ifreedelta);
728 xfs_mod_icount(mp, -idelta);
731 xfs_mod_fdblocks(mp, -blkdelta, rsvd);
738 * Add the given log item to the transaction's list of log items.
740 * The log item will now point to its new descriptor with its li_desc field.
744 struct xfs_trans *tp,
745 struct xfs_log_item *lip)
747 struct xfs_log_item_desc *lidp;
749 ASSERT(lip->li_mountp == tp->t_mountp);
750 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
752 lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
754 lidp->lid_item = lip;
756 list_add_tail(&lidp->lid_trans, &tp->t_items);
760 trace_xfs_trans_add_item(tp, _RET_IP_);
764 xfs_trans_free_item_desc(
765 struct xfs_log_item_desc *lidp)
767 list_del_init(&lidp->lid_trans);
768 kmem_zone_free(xfs_log_item_desc_zone, lidp);
772 * Unlink and free the given descriptor.
776 struct xfs_log_item *lip)
778 xfs_trans_free_item_desc(lip->li_desc);
783 * Unlock all of the items of a transaction and free all the descriptors
784 * of that transaction.
787 xfs_trans_free_items(
788 struct xfs_trans *tp,
789 xfs_lsn_t commit_lsn,
792 struct xfs_log_item_desc *lidp, *next;
794 trace_xfs_trans_free_items(tp, _RET_IP_);
796 list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
797 struct xfs_log_item *lip = lidp->lid_item;
801 if (commit_lsn != NULLCOMMITLSN)
802 lip->li_ops->iop_committing(lip, commit_lsn);
804 set_bit(XFS_LI_ABORTED, &lip->li_flags);
805 lip->li_ops->iop_unlock(lip);
807 xfs_trans_free_item_desc(lidp);
812 xfs_log_item_batch_insert(
813 struct xfs_ail *ailp,
814 struct xfs_ail_cursor *cur,
815 struct xfs_log_item **log_items,
817 xfs_lsn_t commit_lsn)
821 spin_lock(&ailp->ail_lock);
822 /* xfs_trans_ail_update_bulk drops ailp->ail_lock */
823 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
825 for (i = 0; i < nr_items; i++) {
826 struct xfs_log_item *lip = log_items[i];
828 lip->li_ops->iop_unpin(lip, 0);
833 * Bulk operation version of xfs_trans_committed that takes a log vector of
834 * items to insert into the AIL. This uses bulk AIL insertion techniques to
835 * minimise lock traffic.
837 * If we are called with the aborted flag set, it is because a log write during
838 * a CIL checkpoint commit has failed. In this case, all the items in the
839 * checkpoint have already gone through iop_commited and iop_unlock, which
840 * means that checkpoint commit abort handling is treated exactly the same
841 * as an iclog write error even though we haven't started any IO yet. Hence in
842 * this case all we need to do is iop_committed processing, followed by an
843 * iop_unpin(aborted) call.
845 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
846 * at the end of the AIL, the insert cursor avoids the need to walk
847 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
848 * call. This saves a lot of needless list walking and is a net win, even
849 * though it slightly increases that amount of AIL lock traffic to set it up
853 xfs_trans_committed_bulk(
854 struct xfs_ail *ailp,
855 struct xfs_log_vec *log_vector,
856 xfs_lsn_t commit_lsn,
859 #define LOG_ITEM_BATCH_SIZE 32
860 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
861 struct xfs_log_vec *lv;
862 struct xfs_ail_cursor cur;
865 spin_lock(&ailp->ail_lock);
866 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
867 spin_unlock(&ailp->ail_lock);
869 /* unpin all the log items */
870 for (lv = log_vector; lv; lv = lv->lv_next ) {
871 struct xfs_log_item *lip = lv->lv_item;
875 set_bit(XFS_LI_ABORTED, &lip->li_flags);
876 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
878 /* item_lsn of -1 means the item needs no further processing */
879 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
883 * if we are aborting the operation, no point in inserting the
884 * object into the AIL as we are in a shutdown situation.
887 ASSERT(XFS_FORCED_SHUTDOWN(ailp->ail_mount));
888 lip->li_ops->iop_unpin(lip, 1);
892 if (item_lsn != commit_lsn) {
895 * Not a bulk update option due to unusual item_lsn.
896 * Push into AIL immediately, rechecking the lsn once
897 * we have the ail lock. Then unpin the item. This does
898 * not affect the AIL cursor the bulk insert path is
901 spin_lock(&ailp->ail_lock);
902 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
903 xfs_trans_ail_update(ailp, lip, item_lsn);
905 spin_unlock(&ailp->ail_lock);
906 lip->li_ops->iop_unpin(lip, 0);
910 /* Item is a candidate for bulk AIL insert. */
911 log_items[i++] = lv->lv_item;
912 if (i >= LOG_ITEM_BATCH_SIZE) {
913 xfs_log_item_batch_insert(ailp, &cur, log_items,
914 LOG_ITEM_BATCH_SIZE, commit_lsn);
919 /* make sure we insert the remainder! */
921 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
923 spin_lock(&ailp->ail_lock);
924 xfs_trans_ail_cursor_done(&cur);
925 spin_unlock(&ailp->ail_lock);
929 * Commit the given transaction to the log.
931 * XFS disk error handling mechanism is not based on a typical
932 * transaction abort mechanism. Logically after the filesystem
933 * gets marked 'SHUTDOWN', we can't let any new transactions
934 * be durable - ie. committed to disk - because some metadata might
935 * be inconsistent. In such cases, this returns an error, and the
936 * caller may assume that all locked objects joined to the transaction
937 * have already been unlocked as if the commit had succeeded.
938 * Do not reference the transaction structure after this call.
942 struct xfs_trans *tp,
945 struct xfs_mount *mp = tp->t_mountp;
946 xfs_lsn_t commit_lsn = -1;
948 int sync = tp->t_flags & XFS_TRANS_SYNC;
950 ASSERT(!tp->t_agfl_dfops ||
951 !xfs_defer_has_unfinished_work(tp->t_agfl_dfops) || regrant);
953 trace_xfs_trans_commit(tp, _RET_IP_);
956 * If there is nothing to be logged by the transaction,
957 * then unlock all of the items associated with the
958 * transaction and free the transaction structure.
959 * Also make sure to return any reserved blocks to
962 if (!(tp->t_flags & XFS_TRANS_DIRTY))
965 if (XFS_FORCED_SHUTDOWN(mp)) {
970 ASSERT(tp->t_ticket != NULL);
973 * If we need to update the superblock, then do it now.
975 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
976 xfs_trans_apply_sb_deltas(tp);
977 xfs_trans_apply_dquot_deltas(tp);
979 xfs_log_commit_cil(mp, tp, &commit_lsn, regrant);
981 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
985 * If the transaction needs to be synchronous, then force the
986 * log out now and wait for it.
989 error = xfs_log_force_lsn(mp, commit_lsn, XFS_LOG_SYNC, NULL);
990 XFS_STATS_INC(mp, xs_trans_sync);
992 XFS_STATS_INC(mp, xs_trans_async);
998 xfs_trans_unreserve_and_mod_sb(tp);
1001 * It is indeed possible for the transaction to be not dirty but
1002 * the dqinfo portion to be. All that means is that we have some
1003 * (non-persistent) quota reservations that need to be unreserved.
1005 xfs_trans_unreserve_and_mod_dquots(tp);
1007 commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, regrant);
1008 if (commit_lsn == -1 && !error)
1010 tp->t_ticket = NULL;
1012 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
1013 xfs_trans_free_items(tp, NULLCOMMITLSN, !!error);
1016 XFS_STATS_INC(mp, xs_trans_empty);
1022 struct xfs_trans *tp)
1024 return __xfs_trans_commit(tp, false);
1028 * Unlock all of the transaction's items and free the transaction.
1029 * The transaction must not have modified any of its items, because
1030 * there is no way to restore them to their previous state.
1032 * If the transaction has made a log reservation, make sure to release
1037 struct xfs_trans *tp)
1039 struct xfs_mount *mp = tp->t_mountp;
1040 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
1042 trace_xfs_trans_cancel(tp, _RET_IP_);
1045 * See if the caller is relying on us to shut down the
1046 * filesystem. This happens in paths where we detect
1047 * corruption and decide to give up.
1049 if (dirty && !XFS_FORCED_SHUTDOWN(mp)) {
1050 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1051 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1054 if (!dirty && !XFS_FORCED_SHUTDOWN(mp)) {
1055 struct xfs_log_item_desc *lidp;
1057 list_for_each_entry(lidp, &tp->t_items, lid_trans)
1058 ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
1061 xfs_trans_unreserve_and_mod_sb(tp);
1062 xfs_trans_unreserve_and_mod_dquots(tp);
1065 xfs_log_done(mp, tp->t_ticket, NULL, false);
1066 tp->t_ticket = NULL;
1069 /* mark this thread as no longer being in a transaction */
1070 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
1072 xfs_trans_free_items(tp, NULLCOMMITLSN, dirty);
1077 * Roll from one trans in the sequence of PERMANENT transactions to
1078 * the next: permanent transactions are only flushed out when
1079 * committed with xfs_trans_commit(), but we still want as soon
1080 * as possible to let chunks of it go to the log. So we commit the
1081 * chunk we've been working on and get a new transaction to continue.
1085 struct xfs_trans **tpp)
1087 struct xfs_trans *trans = *tpp;
1088 struct xfs_trans_res tres;
1091 trace_xfs_trans_roll(trans, _RET_IP_);
1094 * Copy the critical parameters from one trans to the next.
1096 tres.tr_logres = trans->t_log_res;
1097 tres.tr_logcount = trans->t_log_count;
1099 *tpp = xfs_trans_dup(trans);
1102 * Commit the current transaction.
1103 * If this commit failed, then it'd just unlock those items that
1104 * are not marked ihold. That also means that a filesystem shutdown
1105 * is in progress. The caller takes the responsibility to cancel
1106 * the duplicate transaction that gets returned.
1108 error = __xfs_trans_commit(trans, true);
1113 * Reserve space in the log for the next transaction.
1114 * This also pushes items in the "AIL", the list of logged items,
1115 * out to disk if they are taking up space at the tail of the log
1116 * that we want to use. This requires that either nothing be locked
1117 * across this call, or that anything that is locked be logged in
1118 * the prior and the next transactions.
1120 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1121 return xfs_trans_reserve(*tpp, &tres, 0, 0);