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 * Create an empty transaction with no reservation. This is a defensive
267 * mechanism for routines that query metadata without actually modifying
268 * them -- if the metadata being queried is somehow cross-linked (think a
269 * btree block pointer that points higher in the tree), we risk deadlock.
270 * However, blocks grabbed as part of a transaction can be re-grabbed.
271 * The verifiers will notice the corrupt block and the operation will fail
272 * back to userspace without deadlocking.
274 * Note the zero-length reservation; this transaction MUST be cancelled
275 * without any dirty data.
278 xfs_trans_alloc_empty(
279 struct xfs_mount *mp,
280 struct xfs_trans **tpp)
282 struct xfs_trans_res resv = {0};
284 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
288 * Record the indicated change to the given field for application
289 * to the file system's superblock when the transaction commits.
290 * For now, just store the change in the transaction structure.
292 * Mark the transaction structure to indicate that the superblock
293 * needs to be updated before committing.
295 * Because we may not be keeping track of allocated/free inodes and
296 * used filesystem blocks in the superblock, we do not mark the
297 * superblock dirty in this transaction if we modify these fields.
298 * We still need to update the transaction deltas so that they get
299 * applied to the incore superblock, but we don't want them to
300 * cause the superblock to get locked and logged if these are the
301 * only fields in the superblock that the transaction modifies.
309 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
310 xfs_mount_t *mp = tp->t_mountp;
313 case XFS_TRANS_SB_ICOUNT:
314 tp->t_icount_delta += delta;
315 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
316 flags &= ~XFS_TRANS_SB_DIRTY;
318 case XFS_TRANS_SB_IFREE:
319 tp->t_ifree_delta += delta;
320 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
321 flags &= ~XFS_TRANS_SB_DIRTY;
323 case XFS_TRANS_SB_FDBLOCKS:
325 * Track the number of blocks allocated in the
326 * transaction. Make sure it does not exceed the
330 tp->t_blk_res_used += (uint)-delta;
331 ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
333 tp->t_fdblocks_delta += delta;
334 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
335 flags &= ~XFS_TRANS_SB_DIRTY;
337 case XFS_TRANS_SB_RES_FDBLOCKS:
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.
343 tp->t_res_fdblocks_delta += delta;
344 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
345 flags &= ~XFS_TRANS_SB_DIRTY;
347 case XFS_TRANS_SB_FREXTENTS:
349 * Track the number of blocks allocated in the
350 * transaction. Make sure it does not exceed the
354 tp->t_rtx_res_used += (uint)-delta;
355 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
357 tp->t_frextents_delta += delta;
359 case XFS_TRANS_SB_RES_FREXTENTS:
361 * The allocation has already been applied to the
362 * in-core superblock's counter. This should only
363 * be applied to the on-disk superblock.
366 tp->t_res_frextents_delta += delta;
368 case XFS_TRANS_SB_DBLOCKS:
370 tp->t_dblocks_delta += delta;
372 case XFS_TRANS_SB_AGCOUNT:
374 tp->t_agcount_delta += delta;
376 case XFS_TRANS_SB_IMAXPCT:
377 tp->t_imaxpct_delta += delta;
379 case XFS_TRANS_SB_REXTSIZE:
380 tp->t_rextsize_delta += delta;
382 case XFS_TRANS_SB_RBMBLOCKS:
383 tp->t_rbmblocks_delta += delta;
385 case XFS_TRANS_SB_RBLOCKS:
386 tp->t_rblocks_delta += delta;
388 case XFS_TRANS_SB_REXTENTS:
389 tp->t_rextents_delta += delta;
391 case XFS_TRANS_SB_REXTSLOG:
392 tp->t_rextslog_delta += delta;
399 tp->t_flags |= flags;
403 * xfs_trans_apply_sb_deltas() is called from the commit code
404 * to bring the superblock buffer into the current transaction
405 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
407 * For now we just look at each field allowed to change and change
411 xfs_trans_apply_sb_deltas(
418 bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
419 sbp = XFS_BUF_TO_SBP(bp);
422 * Check that superblock mods match the mods made to AGF counters.
424 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
425 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
426 tp->t_ag_btree_delta));
429 * Only update the superblock counters if we are logging them
431 if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
432 if (tp->t_icount_delta)
433 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
434 if (tp->t_ifree_delta)
435 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
436 if (tp->t_fdblocks_delta)
437 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
438 if (tp->t_res_fdblocks_delta)
439 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
442 if (tp->t_frextents_delta)
443 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
444 if (tp->t_res_frextents_delta)
445 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
447 if (tp->t_dblocks_delta) {
448 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
451 if (tp->t_agcount_delta) {
452 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
455 if (tp->t_imaxpct_delta) {
456 sbp->sb_imax_pct += tp->t_imaxpct_delta;
459 if (tp->t_rextsize_delta) {
460 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
463 if (tp->t_rbmblocks_delta) {
464 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
467 if (tp->t_rblocks_delta) {
468 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
471 if (tp->t_rextents_delta) {
472 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
475 if (tp->t_rextslog_delta) {
476 sbp->sb_rextslog += tp->t_rextslog_delta;
480 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
483 * Log the whole thing, the fields are noncontiguous.
485 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
488 * Since all the modifiable fields are contiguous, we
489 * can get away with this.
491 xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
492 offsetof(xfs_dsb_t, sb_frextents) +
493 sizeof(sbp->sb_frextents) - 1);
501 int8_t counter = *field;
517 int32_t counter = *field;
533 int64_t counter = *field;
545 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
546 * and apply superblock counter changes to the in-core superblock. The
547 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
548 * applied to the in-core superblock. The idea is that that has already been
551 * If we are not logging superblock counters, then the inode allocated/free and
552 * used block counts are not updated in the on disk superblock. In this case,
553 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
554 * still need to update the incore superblock with the changes.
557 xfs_trans_unreserve_and_mod_sb(
558 struct xfs_trans *tp)
560 struct xfs_mount *mp = tp->t_mountp;
561 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
562 int64_t blkdelta = 0;
563 int64_t rtxdelta = 0;
565 int64_t ifreedelta = 0;
568 /* calculate deltas */
569 if (tp->t_blk_res > 0)
570 blkdelta = tp->t_blk_res;
571 if ((tp->t_fdblocks_delta != 0) &&
572 (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
573 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
574 blkdelta += tp->t_fdblocks_delta;
576 if (tp->t_rtx_res > 0)
577 rtxdelta = tp->t_rtx_res;
578 if ((tp->t_frextents_delta != 0) &&
579 (tp->t_flags & XFS_TRANS_SB_DIRTY))
580 rtxdelta += tp->t_frextents_delta;
582 if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
583 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
584 idelta = tp->t_icount_delta;
585 ifreedelta = tp->t_ifree_delta;
588 /* apply the per-cpu counters */
590 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
596 error = xfs_mod_icount(mp, idelta);
598 goto out_undo_fdblocks;
602 error = xfs_mod_ifree(mp, ifreedelta);
604 goto out_undo_icount;
607 if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
610 /* apply remaining deltas */
611 spin_lock(&mp->m_sb_lock);
613 error = xfs_sb_mod64(&mp->m_sb.sb_frextents, rtxdelta);
618 if (tp->t_dblocks_delta != 0) {
619 error = xfs_sb_mod64(&mp->m_sb.sb_dblocks, tp->t_dblocks_delta);
621 goto out_undo_frextents;
623 if (tp->t_agcount_delta != 0) {
624 error = xfs_sb_mod32(&mp->m_sb.sb_agcount, tp->t_agcount_delta);
626 goto out_undo_dblocks;
628 if (tp->t_imaxpct_delta != 0) {
629 error = xfs_sb_mod8(&mp->m_sb.sb_imax_pct, tp->t_imaxpct_delta);
631 goto out_undo_agcount;
633 if (tp->t_rextsize_delta != 0) {
634 error = xfs_sb_mod32(&mp->m_sb.sb_rextsize,
635 tp->t_rextsize_delta);
637 goto out_undo_imaxpct;
639 if (tp->t_rbmblocks_delta != 0) {
640 error = xfs_sb_mod32(&mp->m_sb.sb_rbmblocks,
641 tp->t_rbmblocks_delta);
643 goto out_undo_rextsize;
645 if (tp->t_rblocks_delta != 0) {
646 error = xfs_sb_mod64(&mp->m_sb.sb_rblocks, tp->t_rblocks_delta);
648 goto out_undo_rbmblocks;
650 if (tp->t_rextents_delta != 0) {
651 error = xfs_sb_mod64(&mp->m_sb.sb_rextents,
652 tp->t_rextents_delta);
654 goto out_undo_rblocks;
656 if (tp->t_rextslog_delta != 0) {
657 error = xfs_sb_mod8(&mp->m_sb.sb_rextslog,
658 tp->t_rextslog_delta);
660 goto out_undo_rextents;
662 spin_unlock(&mp->m_sb_lock);
666 if (tp->t_rextents_delta)
667 xfs_sb_mod64(&mp->m_sb.sb_rextents, -tp->t_rextents_delta);
669 if (tp->t_rblocks_delta)
670 xfs_sb_mod64(&mp->m_sb.sb_rblocks, -tp->t_rblocks_delta);
672 if (tp->t_rbmblocks_delta)
673 xfs_sb_mod32(&mp->m_sb.sb_rbmblocks, -tp->t_rbmblocks_delta);
675 if (tp->t_rextsize_delta)
676 xfs_sb_mod32(&mp->m_sb.sb_rextsize, -tp->t_rextsize_delta);
678 if (tp->t_rextsize_delta)
679 xfs_sb_mod8(&mp->m_sb.sb_imax_pct, -tp->t_imaxpct_delta);
681 if (tp->t_agcount_delta)
682 xfs_sb_mod32(&mp->m_sb.sb_agcount, -tp->t_agcount_delta);
684 if (tp->t_dblocks_delta)
685 xfs_sb_mod64(&mp->m_sb.sb_dblocks, -tp->t_dblocks_delta);
688 xfs_sb_mod64(&mp->m_sb.sb_frextents, -rtxdelta);
690 spin_unlock(&mp->m_sb_lock);
692 xfs_mod_ifree(mp, -ifreedelta);
695 xfs_mod_icount(mp, -idelta);
698 xfs_mod_fdblocks(mp, -blkdelta, rsvd);
705 * Add the given log item to the transaction's list of log items.
707 * The log item will now point to its new descriptor with its li_desc field.
711 struct xfs_trans *tp,
712 struct xfs_log_item *lip)
714 struct xfs_log_item_desc *lidp;
716 ASSERT(lip->li_mountp == tp->t_mountp);
717 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
719 lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
721 lidp->lid_item = lip;
723 list_add_tail(&lidp->lid_trans, &tp->t_items);
729 xfs_trans_free_item_desc(
730 struct xfs_log_item_desc *lidp)
732 list_del_init(&lidp->lid_trans);
733 kmem_zone_free(xfs_log_item_desc_zone, lidp);
737 * Unlink and free the given descriptor.
741 struct xfs_log_item *lip)
743 xfs_trans_free_item_desc(lip->li_desc);
748 * Unlock all of the items of a transaction and free all the descriptors
749 * of that transaction.
752 xfs_trans_free_items(
753 struct xfs_trans *tp,
754 xfs_lsn_t commit_lsn,
757 struct xfs_log_item_desc *lidp, *next;
759 list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
760 struct xfs_log_item *lip = lidp->lid_item;
764 if (commit_lsn != NULLCOMMITLSN)
765 lip->li_ops->iop_committing(lip, commit_lsn);
767 lip->li_flags |= XFS_LI_ABORTED;
768 lip->li_ops->iop_unlock(lip);
770 xfs_trans_free_item_desc(lidp);
775 xfs_log_item_batch_insert(
776 struct xfs_ail *ailp,
777 struct xfs_ail_cursor *cur,
778 struct xfs_log_item **log_items,
780 xfs_lsn_t commit_lsn)
784 spin_lock(&ailp->xa_lock);
785 /* xfs_trans_ail_update_bulk drops ailp->xa_lock */
786 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
788 for (i = 0; i < nr_items; i++) {
789 struct xfs_log_item *lip = log_items[i];
791 lip->li_ops->iop_unpin(lip, 0);
796 * Bulk operation version of xfs_trans_committed that takes a log vector of
797 * items to insert into the AIL. This uses bulk AIL insertion techniques to
798 * minimise lock traffic.
800 * If we are called with the aborted flag set, it is because a log write during
801 * a CIL checkpoint commit has failed. In this case, all the items in the
802 * checkpoint have already gone through iop_commited and iop_unlock, which
803 * means that checkpoint commit abort handling is treated exactly the same
804 * as an iclog write error even though we haven't started any IO yet. Hence in
805 * this case all we need to do is iop_committed processing, followed by an
806 * iop_unpin(aborted) call.
808 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
809 * at the end of the AIL, the insert cursor avoids the need to walk
810 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
811 * call. This saves a lot of needless list walking and is a net win, even
812 * though it slightly increases that amount of AIL lock traffic to set it up
816 xfs_trans_committed_bulk(
817 struct xfs_ail *ailp,
818 struct xfs_log_vec *log_vector,
819 xfs_lsn_t commit_lsn,
822 #define LOG_ITEM_BATCH_SIZE 32
823 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
824 struct xfs_log_vec *lv;
825 struct xfs_ail_cursor cur;
828 spin_lock(&ailp->xa_lock);
829 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
830 spin_unlock(&ailp->xa_lock);
832 /* unpin all the log items */
833 for (lv = log_vector; lv; lv = lv->lv_next ) {
834 struct xfs_log_item *lip = lv->lv_item;
838 lip->li_flags |= XFS_LI_ABORTED;
839 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
841 /* item_lsn of -1 means the item needs no further processing */
842 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
846 * if we are aborting the operation, no point in inserting the
847 * object into the AIL as we are in a shutdown situation.
850 ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount));
851 lip->li_ops->iop_unpin(lip, 1);
855 if (item_lsn != commit_lsn) {
858 * Not a bulk update option due to unusual item_lsn.
859 * Push into AIL immediately, rechecking the lsn once
860 * we have the ail lock. Then unpin the item. This does
861 * not affect the AIL cursor the bulk insert path is
864 spin_lock(&ailp->xa_lock);
865 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
866 xfs_trans_ail_update(ailp, lip, item_lsn);
868 spin_unlock(&ailp->xa_lock);
869 lip->li_ops->iop_unpin(lip, 0);
873 /* Item is a candidate for bulk AIL insert. */
874 log_items[i++] = lv->lv_item;
875 if (i >= LOG_ITEM_BATCH_SIZE) {
876 xfs_log_item_batch_insert(ailp, &cur, log_items,
877 LOG_ITEM_BATCH_SIZE, commit_lsn);
882 /* make sure we insert the remainder! */
884 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
886 spin_lock(&ailp->xa_lock);
887 xfs_trans_ail_cursor_done(&cur);
888 spin_unlock(&ailp->xa_lock);
892 * Commit the given transaction to the log.
894 * XFS disk error handling mechanism is not based on a typical
895 * transaction abort mechanism. Logically after the filesystem
896 * gets marked 'SHUTDOWN', we can't let any new transactions
897 * be durable - ie. committed to disk - because some metadata might
898 * be inconsistent. In such cases, this returns an error, and the
899 * caller may assume that all locked objects joined to the transaction
900 * have already been unlocked as if the commit had succeeded.
901 * Do not reference the transaction structure after this call.
905 struct xfs_trans *tp,
908 struct xfs_mount *mp = tp->t_mountp;
909 xfs_lsn_t commit_lsn = -1;
911 int sync = tp->t_flags & XFS_TRANS_SYNC;
914 * If there is nothing to be logged by the transaction,
915 * then unlock all of the items associated with the
916 * transaction and free the transaction structure.
917 * Also make sure to return any reserved blocks to
920 if (!(tp->t_flags & XFS_TRANS_DIRTY))
923 if (XFS_FORCED_SHUTDOWN(mp)) {
928 ASSERT(tp->t_ticket != NULL);
931 * If we need to update the superblock, then do it now.
933 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
934 xfs_trans_apply_sb_deltas(tp);
935 xfs_trans_apply_dquot_deltas(tp);
937 xfs_log_commit_cil(mp, tp, &commit_lsn, regrant);
939 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
943 * If the transaction needs to be synchronous, then force the
944 * log out now and wait for it.
947 error = _xfs_log_force_lsn(mp, commit_lsn, XFS_LOG_SYNC, NULL);
948 XFS_STATS_INC(mp, xs_trans_sync);
950 XFS_STATS_INC(mp, xs_trans_async);
956 xfs_trans_unreserve_and_mod_sb(tp);
959 * It is indeed possible for the transaction to be not dirty but
960 * the dqinfo portion to be. All that means is that we have some
961 * (non-persistent) quota reservations that need to be unreserved.
963 xfs_trans_unreserve_and_mod_dquots(tp);
965 commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, regrant);
966 if (commit_lsn == -1 && !error)
969 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
970 xfs_trans_free_items(tp, NULLCOMMITLSN, !!error);
973 XFS_STATS_INC(mp, xs_trans_empty);
979 struct xfs_trans *tp)
981 return __xfs_trans_commit(tp, false);
985 * Unlock all of the transaction's items and free the transaction.
986 * The transaction must not have modified any of its items, because
987 * there is no way to restore them to their previous state.
989 * If the transaction has made a log reservation, make sure to release
994 struct xfs_trans *tp)
996 struct xfs_mount *mp = tp->t_mountp;
997 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
1000 * See if the caller is relying on us to shut down the
1001 * filesystem. This happens in paths where we detect
1002 * corruption and decide to give up.
1004 if (dirty && !XFS_FORCED_SHUTDOWN(mp)) {
1005 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1006 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1009 if (!dirty && !XFS_FORCED_SHUTDOWN(mp)) {
1010 struct xfs_log_item_desc *lidp;
1012 list_for_each_entry(lidp, &tp->t_items, lid_trans)
1013 ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
1016 xfs_trans_unreserve_and_mod_sb(tp);
1017 xfs_trans_unreserve_and_mod_dquots(tp);
1020 xfs_log_done(mp, tp->t_ticket, NULL, false);
1022 /* mark this thread as no longer being in a transaction */
1023 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1025 xfs_trans_free_items(tp, NULLCOMMITLSN, dirty);
1030 * Roll from one trans in the sequence of PERMANENT transactions to
1031 * the next: permanent transactions are only flushed out when
1032 * committed with xfs_trans_commit(), but we still want as soon
1033 * as possible to let chunks of it go to the log. So we commit the
1034 * chunk we've been working on and get a new transaction to continue.
1038 struct xfs_trans **tpp,
1039 struct xfs_inode *dp,
1042 struct xfs_trans *trans;
1043 struct xfs_trans_res tres;
1049 * Ensure that the inode is always logged.
1053 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
1056 * Copy the critical parameters from one trans to the next.
1058 tres.tr_logres = trans->t_log_res;
1059 tres.tr_logcount = trans->t_log_count;
1060 *tpp = xfs_trans_dup(trans);
1063 * Commit the current transaction.
1064 * If this commit failed, then it'd just unlock those items that
1065 * are not marked ihold. That also means that a filesystem shutdown
1066 * is in progress. The caller takes the responsibility to cancel
1067 * the duplicate transaction that gets returned.
1069 error = __xfs_trans_commit(trans, true);
1077 * Reserve space in the log for th next transaction.
1078 * This also pushes items in the "AIL", the list of logged items,
1079 * out to disk if they are taking up space at the tail of the log
1080 * that we want to use. This requires that either nothing be locked
1081 * across this call, or that anything that is locked be logged in
1082 * the prior and the next transactions.
1084 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1085 error = xfs_trans_reserve(trans, &tres, 0, 0);
1087 * Ensure that the inode is in the new transaction and locked.
1093 xfs_trans_ijoin(trans, dp, 0);
1099 struct xfs_trans **tpp,
1100 struct xfs_inode *dp)
1103 return __xfs_trans_roll(tpp, dp, &committed);