1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
4 * Copyright (C) 2010 Red Hat, Inc.
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_log_priv.h"
13 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_extent_busy.h"
16 #include "xfs_quota.h"
17 #include "xfs_trans.h"
18 #include "xfs_trans_priv.h"
20 #include "xfs_trace.h"
21 #include "xfs_error.h"
22 #include "xfs_defer.h"
24 kmem_zone_t *xfs_trans_zone;
26 #if defined(CONFIG_TRACEPOINTS)
28 xfs_trans_trace_reservations(
31 struct xfs_trans_res resv;
32 struct xfs_trans_res *res;
33 struct xfs_trans_res *end_res;
36 res = (struct xfs_trans_res *)M_RES(mp);
37 end_res = (struct xfs_trans_res *)(M_RES(mp) + 1);
38 for (i = 0; res < end_res; i++, res++)
39 trace_xfs_trans_resv_calc(mp, i, res);
40 xfs_log_get_max_trans_res(mp, &resv);
41 trace_xfs_trans_resv_calc(mp, -1, &resv);
44 # define xfs_trans_trace_reservations(mp)
48 * Initialize the precomputed transaction reservation values
49 * in the mount structure.
55 xfs_trans_resv_calc(mp, M_RES(mp));
56 xfs_trans_trace_reservations(mp);
60 * Free the transaction structure. If there is more clean up
61 * to do when the structure is freed, add it here.
67 xfs_extent_busy_sort(&tp->t_busy);
68 xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
70 trace_xfs_trans_free(tp, _RET_IP_);
71 if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
72 sb_end_intwrite(tp->t_mountp->m_super);
73 xfs_trans_free_dqinfo(tp);
74 kmem_cache_free(xfs_trans_zone, tp);
78 * This is called to create a new transaction which will share the
79 * permanent log reservation of the given transaction. The remaining
80 * unused block and rt extent reservations are also inherited. This
81 * implies that the original transaction is no longer allowed to allocate
82 * blocks. Locks and log items, however, are no inherited. They must
83 * be added to the new transaction explicitly.
85 STATIC struct xfs_trans *
89 struct xfs_trans *ntp;
91 trace_xfs_trans_dup(tp, _RET_IP_);
93 ntp = kmem_cache_zalloc(xfs_trans_zone, GFP_KERNEL | __GFP_NOFAIL);
96 * Initialize the new transaction structure.
98 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
99 ntp->t_mountp = tp->t_mountp;
100 INIT_LIST_HEAD(&ntp->t_items);
101 INIT_LIST_HEAD(&ntp->t_busy);
102 INIT_LIST_HEAD(&ntp->t_dfops);
103 ntp->t_firstblock = NULLFSBLOCK;
105 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
106 ASSERT(tp->t_ticket != NULL);
108 ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
109 (tp->t_flags & XFS_TRANS_RESERVE) |
110 (tp->t_flags & XFS_TRANS_NO_WRITECOUNT) |
111 (tp->t_flags & XFS_TRANS_RES_FDBLKS);
112 /* We gave our writer reference to the new transaction */
113 tp->t_flags |= XFS_TRANS_NO_WRITECOUNT;
114 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
116 ASSERT(tp->t_blk_res >= tp->t_blk_res_used);
117 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
118 tp->t_blk_res = tp->t_blk_res_used;
120 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
121 tp->t_rtx_res = tp->t_rtx_res_used;
122 ntp->t_pflags = tp->t_pflags;
124 /* move deferred ops over to the new tp */
125 xfs_defer_move(ntp, tp);
127 xfs_trans_dup_dqinfo(tp, ntp);
132 * This is called to reserve free disk blocks and log space for the
133 * given transaction. This must be done before allocating any resources
134 * within the transaction.
136 * This will return ENOSPC if there are not enough blocks available.
137 * It will sleep waiting for available log space.
138 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
139 * is used by long running transactions. If any one of the reservations
140 * fails then they will all be backed out.
142 * This does not do quota reservations. That typically is done by the
147 struct xfs_trans *tp,
148 struct xfs_trans_res *resp,
152 struct xfs_mount *mp = tp->t_mountp;
154 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
156 /* Mark this thread as being in a transaction */
157 current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
160 * Attempt to reserve the needed disk blocks by decrementing
161 * the number needed from the number available. This will
162 * fail if the count would go below zero.
165 error = xfs_mod_fdblocks(mp, -((int64_t)blocks), rsvd);
167 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
170 tp->t_blk_res += blocks;
174 * Reserve the log space needed for this transaction.
176 if (resp->tr_logres > 0) {
177 bool permanent = false;
179 ASSERT(tp->t_log_res == 0 ||
180 tp->t_log_res == resp->tr_logres);
181 ASSERT(tp->t_log_count == 0 ||
182 tp->t_log_count == resp->tr_logcount);
184 if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
185 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
188 ASSERT(tp->t_ticket == NULL);
189 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
192 if (tp->t_ticket != NULL) {
193 ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
194 error = xfs_log_regrant(mp, tp->t_ticket);
196 error = xfs_log_reserve(mp,
199 &tp->t_ticket, XFS_TRANSACTION,
206 tp->t_log_res = resp->tr_logres;
207 tp->t_log_count = resp->tr_logcount;
211 * Attempt to reserve the needed realtime extents by decrementing
212 * the number needed from the number available. This will
213 * fail if the count would go below zero.
216 error = xfs_mod_frextents(mp, -((int64_t)rtextents));
221 tp->t_rtx_res += rtextents;
227 * Error cases jump to one of these labels to undo any
228 * reservations which have already been performed.
231 if (resp->tr_logres > 0) {
232 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
235 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
240 xfs_mod_fdblocks(mp, (int64_t)blocks, rsvd);
244 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
251 struct xfs_mount *mp,
252 struct xfs_trans_res *resp,
256 struct xfs_trans **tpp)
258 struct xfs_trans *tp;
262 * Allocate the handle before we do our freeze accounting and setting up
263 * GFP_NOFS allocation context so that we avoid lockdep false positives
264 * by doing GFP_KERNEL allocations inside sb_start_intwrite().
266 tp = kmem_cache_zalloc(xfs_trans_zone, GFP_KERNEL | __GFP_NOFAIL);
267 if (!(flags & XFS_TRANS_NO_WRITECOUNT))
268 sb_start_intwrite(mp->m_super);
271 * Zero-reservation ("empty") transactions can't modify anything, so
272 * they're allowed to run while we're frozen.
274 WARN_ON(resp->tr_logres > 0 &&
275 mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
276 ASSERT(!(flags & XFS_TRANS_RES_FDBLKS) ||
277 xfs_sb_version_haslazysbcount(&mp->m_sb));
279 tp->t_magic = XFS_TRANS_HEADER_MAGIC;
282 INIT_LIST_HEAD(&tp->t_items);
283 INIT_LIST_HEAD(&tp->t_busy);
284 INIT_LIST_HEAD(&tp->t_dfops);
285 tp->t_firstblock = NULLFSBLOCK;
287 error = xfs_trans_reserve(tp, resp, blocks, rtextents);
289 xfs_trans_cancel(tp);
293 trace_xfs_trans_alloc(tp, _RET_IP_);
300 * Create an empty transaction with no reservation. This is a defensive
301 * mechanism for routines that query metadata without actually modifying them --
302 * if the metadata being queried is somehow cross-linked (think a btree block
303 * pointer that points higher in the tree), we risk deadlock. However, blocks
304 * grabbed as part of a transaction can be re-grabbed. The verifiers will
305 * notice the corrupt block and the operation will fail back to userspace
306 * without deadlocking.
308 * Note the zero-length reservation; this transaction MUST be cancelled without
311 * Callers should obtain freeze protection to avoid a conflict with fs freezing
312 * where we can be grabbing buffers at the same time that freeze is trying to
313 * drain the buffer LRU list.
316 xfs_trans_alloc_empty(
317 struct xfs_mount *mp,
318 struct xfs_trans **tpp)
320 struct xfs_trans_res resv = {0};
322 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
326 * Record the indicated change to the given field for application
327 * to the file system's superblock when the transaction commits.
328 * For now, just store the change in the transaction structure.
330 * Mark the transaction structure to indicate that the superblock
331 * needs to be updated before committing.
333 * Because we may not be keeping track of allocated/free inodes and
334 * used filesystem blocks in the superblock, we do not mark the
335 * superblock dirty in this transaction if we modify these fields.
336 * We still need to update the transaction deltas so that they get
337 * applied to the incore superblock, but we don't want them to
338 * cause the superblock to get locked and logged if these are the
339 * only fields in the superblock that the transaction modifies.
347 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
348 xfs_mount_t *mp = tp->t_mountp;
351 case XFS_TRANS_SB_ICOUNT:
352 tp->t_icount_delta += delta;
353 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
354 flags &= ~XFS_TRANS_SB_DIRTY;
356 case XFS_TRANS_SB_IFREE:
357 tp->t_ifree_delta += delta;
358 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
359 flags &= ~XFS_TRANS_SB_DIRTY;
361 case XFS_TRANS_SB_FDBLOCKS:
363 * Track the number of blocks allocated in the transaction.
364 * Make sure it does not exceed the number reserved. If so,
365 * shutdown as this can lead to accounting inconsistency.
368 tp->t_blk_res_used += (uint)-delta;
369 if (tp->t_blk_res_used > tp->t_blk_res)
370 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
371 } else if (delta > 0 && (tp->t_flags & XFS_TRANS_RES_FDBLKS)) {
372 int64_t blkres_delta;
375 * Return freed blocks directly to the reservation
376 * instead of the global pool, being careful not to
377 * overflow the trans counter. This is used to preserve
378 * reservation across chains of transaction rolls that
379 * repeatedly free and allocate blocks.
381 blkres_delta = min_t(int64_t, delta,
382 UINT_MAX - tp->t_blk_res);
383 tp->t_blk_res += blkres_delta;
384 delta -= blkres_delta;
386 tp->t_fdblocks_delta += delta;
387 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
388 flags &= ~XFS_TRANS_SB_DIRTY;
390 case XFS_TRANS_SB_RES_FDBLOCKS:
392 * The allocation has already been applied to the
393 * in-core superblock's counter. This should only
394 * be applied to the on-disk superblock.
396 tp->t_res_fdblocks_delta += delta;
397 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
398 flags &= ~XFS_TRANS_SB_DIRTY;
400 case XFS_TRANS_SB_FREXTENTS:
402 * Track the number of blocks allocated in the
403 * transaction. Make sure it does not exceed the
407 tp->t_rtx_res_used += (uint)-delta;
408 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
410 tp->t_frextents_delta += delta;
412 case XFS_TRANS_SB_RES_FREXTENTS:
414 * The allocation has already been applied to the
415 * in-core superblock's counter. This should only
416 * be applied to the on-disk superblock.
419 tp->t_res_frextents_delta += delta;
421 case XFS_TRANS_SB_DBLOCKS:
423 tp->t_dblocks_delta += delta;
425 case XFS_TRANS_SB_AGCOUNT:
427 tp->t_agcount_delta += delta;
429 case XFS_TRANS_SB_IMAXPCT:
430 tp->t_imaxpct_delta += delta;
432 case XFS_TRANS_SB_REXTSIZE:
433 tp->t_rextsize_delta += delta;
435 case XFS_TRANS_SB_RBMBLOCKS:
436 tp->t_rbmblocks_delta += delta;
438 case XFS_TRANS_SB_RBLOCKS:
439 tp->t_rblocks_delta += delta;
441 case XFS_TRANS_SB_REXTENTS:
442 tp->t_rextents_delta += delta;
444 case XFS_TRANS_SB_REXTSLOG:
445 tp->t_rextslog_delta += delta;
452 tp->t_flags |= flags;
456 * xfs_trans_apply_sb_deltas() is called from the commit code
457 * to bring the superblock buffer into the current transaction
458 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
460 * For now we just look at each field allowed to change and change
464 xfs_trans_apply_sb_deltas(
471 bp = xfs_trans_getsb(tp, tp->t_mountp);
475 * Check that superblock mods match the mods made to AGF counters.
477 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
478 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
479 tp->t_ag_btree_delta));
482 * Only update the superblock counters if we are logging them
484 if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
485 if (tp->t_icount_delta)
486 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
487 if (tp->t_ifree_delta)
488 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
489 if (tp->t_fdblocks_delta)
490 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
491 if (tp->t_res_fdblocks_delta)
492 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
495 if (tp->t_frextents_delta)
496 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
497 if (tp->t_res_frextents_delta)
498 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
500 if (tp->t_dblocks_delta) {
501 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
504 if (tp->t_agcount_delta) {
505 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
508 if (tp->t_imaxpct_delta) {
509 sbp->sb_imax_pct += tp->t_imaxpct_delta;
512 if (tp->t_rextsize_delta) {
513 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
516 if (tp->t_rbmblocks_delta) {
517 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
520 if (tp->t_rblocks_delta) {
521 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
524 if (tp->t_rextents_delta) {
525 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
528 if (tp->t_rextslog_delta) {
529 sbp->sb_rextslog += tp->t_rextslog_delta;
533 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
536 * Log the whole thing, the fields are noncontiguous.
538 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
541 * Since all the modifiable fields are contiguous, we
542 * can get away with this.
544 xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
545 offsetof(xfs_dsb_t, sb_frextents) +
546 sizeof(sbp->sb_frextents) - 1);
550 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations and
551 * apply superblock counter changes to the in-core superblock. The
552 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
553 * applied to the in-core superblock. The idea is that that has already been
556 * If we are not logging superblock counters, then the inode allocated/free and
557 * used block counts are not updated in the on disk superblock. In this case,
558 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
559 * still need to update the incore superblock with the changes.
561 * Deltas for the inode count are +/-64, hence we use a large batch size of 128
562 * so we don't need to take the counter lock on every update.
564 #define XFS_ICOUNT_BATCH 128
567 xfs_trans_unreserve_and_mod_sb(
568 struct xfs_trans *tp)
570 struct xfs_mount *mp = tp->t_mountp;
571 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
572 int64_t blkdelta = 0;
573 int64_t rtxdelta = 0;
575 int64_t ifreedelta = 0;
578 /* calculate deltas */
579 if (tp->t_blk_res > 0)
580 blkdelta = tp->t_blk_res;
581 if ((tp->t_fdblocks_delta != 0) &&
582 (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
583 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
584 blkdelta += tp->t_fdblocks_delta;
586 if (tp->t_rtx_res > 0)
587 rtxdelta = tp->t_rtx_res;
588 if ((tp->t_frextents_delta != 0) &&
589 (tp->t_flags & XFS_TRANS_SB_DIRTY))
590 rtxdelta += tp->t_frextents_delta;
592 if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
593 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
594 idelta = tp->t_icount_delta;
595 ifreedelta = tp->t_ifree_delta;
598 /* apply the per-cpu counters */
600 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
605 percpu_counter_add_batch(&mp->m_icount, idelta,
608 ASSERT(__percpu_counter_compare(&mp->m_icount, 0,
609 XFS_ICOUNT_BATCH) >= 0);
613 percpu_counter_add(&mp->m_ifree, ifreedelta);
615 ASSERT(percpu_counter_compare(&mp->m_ifree, 0) >= 0);
618 if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
621 /* apply remaining deltas */
622 spin_lock(&mp->m_sb_lock);
623 mp->m_sb.sb_frextents += rtxdelta;
624 mp->m_sb.sb_dblocks += tp->t_dblocks_delta;
625 mp->m_sb.sb_agcount += tp->t_agcount_delta;
626 mp->m_sb.sb_imax_pct += tp->t_imaxpct_delta;
627 mp->m_sb.sb_rextsize += tp->t_rextsize_delta;
628 mp->m_sb.sb_rbmblocks += tp->t_rbmblocks_delta;
629 mp->m_sb.sb_rblocks += tp->t_rblocks_delta;
630 mp->m_sb.sb_rextents += tp->t_rextents_delta;
631 mp->m_sb.sb_rextslog += tp->t_rextslog_delta;
632 spin_unlock(&mp->m_sb_lock);
635 * Debug checks outside of the spinlock so they don't lock up the
636 * machine if they fail.
638 ASSERT(mp->m_sb.sb_imax_pct >= 0);
639 ASSERT(mp->m_sb.sb_rextslog >= 0);
643 /* Add the given log item to the transaction's list of log items. */
646 struct xfs_trans *tp,
647 struct xfs_log_item *lip)
649 ASSERT(lip->li_mountp == tp->t_mountp);
650 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
651 ASSERT(list_empty(&lip->li_trans));
652 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->li_flags));
654 list_add_tail(&lip->li_trans, &tp->t_items);
655 trace_xfs_trans_add_item(tp, _RET_IP_);
659 * Unlink the log item from the transaction. the log item is no longer
660 * considered dirty in this transaction, as the linked transaction has
661 * finished, either by abort or commit completion.
665 struct xfs_log_item *lip)
667 clear_bit(XFS_LI_DIRTY, &lip->li_flags);
668 list_del_init(&lip->li_trans);
671 /* Detach and unlock all of the items in a transaction */
673 xfs_trans_free_items(
674 struct xfs_trans *tp,
677 struct xfs_log_item *lip, *next;
679 trace_xfs_trans_free_items(tp, _RET_IP_);
681 list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
682 xfs_trans_del_item(lip);
684 set_bit(XFS_LI_ABORTED, &lip->li_flags);
685 if (lip->li_ops->iop_release)
686 lip->li_ops->iop_release(lip);
691 xfs_log_item_batch_insert(
692 struct xfs_ail *ailp,
693 struct xfs_ail_cursor *cur,
694 struct xfs_log_item **log_items,
696 xfs_lsn_t commit_lsn)
700 spin_lock(&ailp->ail_lock);
701 /* xfs_trans_ail_update_bulk drops ailp->ail_lock */
702 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
704 for (i = 0; i < nr_items; i++) {
705 struct xfs_log_item *lip = log_items[i];
707 if (lip->li_ops->iop_unpin)
708 lip->li_ops->iop_unpin(lip, 0);
713 * Bulk operation version of xfs_trans_committed that takes a log vector of
714 * items to insert into the AIL. This uses bulk AIL insertion techniques to
715 * minimise lock traffic.
717 * If we are called with the aborted flag set, it is because a log write during
718 * a CIL checkpoint commit has failed. In this case, all the items in the
719 * checkpoint have already gone through iop_committed and iop_committing, which
720 * means that checkpoint commit abort handling is treated exactly the same
721 * as an iclog write error even though we haven't started any IO yet. Hence in
722 * this case all we need to do is iop_committed processing, followed by an
723 * iop_unpin(aborted) call.
725 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
726 * at the end of the AIL, the insert cursor avoids the need to walk
727 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
728 * call. This saves a lot of needless list walking and is a net win, even
729 * though it slightly increases that amount of AIL lock traffic to set it up
733 xfs_trans_committed_bulk(
734 struct xfs_ail *ailp,
735 struct xfs_log_vec *log_vector,
736 xfs_lsn_t commit_lsn,
739 #define LOG_ITEM_BATCH_SIZE 32
740 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
741 struct xfs_log_vec *lv;
742 struct xfs_ail_cursor cur;
745 spin_lock(&ailp->ail_lock);
746 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
747 spin_unlock(&ailp->ail_lock);
749 /* unpin all the log items */
750 for (lv = log_vector; lv; lv = lv->lv_next ) {
751 struct xfs_log_item *lip = lv->lv_item;
755 set_bit(XFS_LI_ABORTED, &lip->li_flags);
757 if (lip->li_ops->flags & XFS_ITEM_RELEASE_WHEN_COMMITTED) {
758 lip->li_ops->iop_release(lip);
762 if (lip->li_ops->iop_committed)
763 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
765 item_lsn = commit_lsn;
767 /* item_lsn of -1 means the item needs no further processing */
768 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
772 * if we are aborting the operation, no point in inserting the
773 * object into the AIL as we are in a shutdown situation.
776 ASSERT(XFS_FORCED_SHUTDOWN(ailp->ail_mount));
777 if (lip->li_ops->iop_unpin)
778 lip->li_ops->iop_unpin(lip, 1);
782 if (item_lsn != commit_lsn) {
785 * Not a bulk update option due to unusual item_lsn.
786 * Push into AIL immediately, rechecking the lsn once
787 * we have the ail lock. Then unpin the item. This does
788 * not affect the AIL cursor the bulk insert path is
791 spin_lock(&ailp->ail_lock);
792 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
793 xfs_trans_ail_update(ailp, lip, item_lsn);
795 spin_unlock(&ailp->ail_lock);
796 if (lip->li_ops->iop_unpin)
797 lip->li_ops->iop_unpin(lip, 0);
801 /* Item is a candidate for bulk AIL insert. */
802 log_items[i++] = lv->lv_item;
803 if (i >= LOG_ITEM_BATCH_SIZE) {
804 xfs_log_item_batch_insert(ailp, &cur, log_items,
805 LOG_ITEM_BATCH_SIZE, commit_lsn);
810 /* make sure we insert the remainder! */
812 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
814 spin_lock(&ailp->ail_lock);
815 xfs_trans_ail_cursor_done(&cur);
816 spin_unlock(&ailp->ail_lock);
820 * Commit the given transaction to the log.
822 * XFS disk error handling mechanism is not based on a typical
823 * transaction abort mechanism. Logically after the filesystem
824 * gets marked 'SHUTDOWN', we can't let any new transactions
825 * be durable - ie. committed to disk - because some metadata might
826 * be inconsistent. In such cases, this returns an error, and the
827 * caller may assume that all locked objects joined to the transaction
828 * have already been unlocked as if the commit had succeeded.
829 * Do not reference the transaction structure after this call.
833 struct xfs_trans *tp,
836 struct xfs_mount *mp = tp->t_mountp;
837 xfs_lsn_t commit_lsn = -1;
839 int sync = tp->t_flags & XFS_TRANS_SYNC;
841 trace_xfs_trans_commit(tp, _RET_IP_);
844 * Finish deferred items on final commit. Only permanent transactions
845 * should ever have deferred ops.
847 WARN_ON_ONCE(!list_empty(&tp->t_dfops) &&
848 !(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
849 if (!regrant && (tp->t_flags & XFS_TRANS_PERM_LOG_RES)) {
850 error = xfs_defer_finish_noroll(&tp);
856 * If there is nothing to be logged by the transaction,
857 * then unlock all of the items associated with the
858 * transaction and free the transaction structure.
859 * Also make sure to return any reserved blocks to
862 if (!(tp->t_flags & XFS_TRANS_DIRTY))
865 if (XFS_FORCED_SHUTDOWN(mp)) {
870 ASSERT(tp->t_ticket != NULL);
873 * If we need to update the superblock, then do it now.
875 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
876 xfs_trans_apply_sb_deltas(tp);
877 xfs_trans_apply_dquot_deltas(tp);
879 xfs_log_commit_cil(mp, tp, &commit_lsn, regrant);
881 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
885 * If the transaction needs to be synchronous, then force the
886 * log out now and wait for it.
889 error = xfs_log_force_lsn(mp, commit_lsn, XFS_LOG_SYNC, NULL);
890 XFS_STATS_INC(mp, xs_trans_sync);
892 XFS_STATS_INC(mp, xs_trans_async);
898 xfs_trans_unreserve_and_mod_sb(tp);
901 * It is indeed possible for the transaction to be not dirty but
902 * the dqinfo portion to be. All that means is that we have some
903 * (non-persistent) quota reservations that need to be unreserved.
905 xfs_trans_unreserve_and_mod_dquots(tp);
907 if (regrant && !XLOG_FORCED_SHUTDOWN(mp->m_log))
908 xfs_log_ticket_regrant(mp->m_log, tp->t_ticket);
910 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
913 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
914 xfs_trans_free_items(tp, !!error);
917 XFS_STATS_INC(mp, xs_trans_empty);
923 struct xfs_trans *tp)
925 return __xfs_trans_commit(tp, false);
929 * Unlock all of the transaction's items and free the transaction.
930 * The transaction must not have modified any of its items, because
931 * there is no way to restore them to their previous state.
933 * If the transaction has made a log reservation, make sure to release
938 struct xfs_trans *tp)
940 struct xfs_mount *mp = tp->t_mountp;
941 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
943 trace_xfs_trans_cancel(tp, _RET_IP_);
945 if (tp->t_flags & XFS_TRANS_PERM_LOG_RES)
946 xfs_defer_cancel(tp);
949 * See if the caller is relying on us to shut down the
950 * filesystem. This happens in paths where we detect
951 * corruption and decide to give up.
953 if (dirty && !XFS_FORCED_SHUTDOWN(mp)) {
954 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
955 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
958 if (!dirty && !XFS_FORCED_SHUTDOWN(mp)) {
959 struct xfs_log_item *lip;
961 list_for_each_entry(lip, &tp->t_items, li_trans)
962 ASSERT(!(lip->li_type == XFS_LI_EFD));
965 xfs_trans_unreserve_and_mod_sb(tp);
966 xfs_trans_unreserve_and_mod_dquots(tp);
969 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
973 /* mark this thread as no longer being in a transaction */
974 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
976 xfs_trans_free_items(tp, dirty);
981 * Roll from one trans in the sequence of PERMANENT transactions to
982 * the next: permanent transactions are only flushed out when
983 * committed with xfs_trans_commit(), but we still want as soon
984 * as possible to let chunks of it go to the log. So we commit the
985 * chunk we've been working on and get a new transaction to continue.
989 struct xfs_trans **tpp)
991 struct xfs_trans *trans = *tpp;
992 struct xfs_trans_res tres;
995 trace_xfs_trans_roll(trans, _RET_IP_);
998 * Copy the critical parameters from one trans to the next.
1000 tres.tr_logres = trans->t_log_res;
1001 tres.tr_logcount = trans->t_log_count;
1003 *tpp = xfs_trans_dup(trans);
1006 * Commit the current transaction.
1007 * If this commit failed, then it'd just unlock those items that
1008 * are not marked ihold. That also means that a filesystem shutdown
1009 * is in progress. The caller takes the responsibility to cancel
1010 * the duplicate transaction that gets returned.
1012 error = __xfs_trans_commit(trans, true);
1017 * Reserve space in the log for the next transaction.
1018 * This also pushes items in the "AIL", the list of logged items,
1019 * out to disk if they are taking up space at the tail of the log
1020 * that we want to use. This requires that either nothing be locked
1021 * across this call, or that anything that is locked be logged in
1022 * the prior and the next transactions.
1024 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1025 return xfs_trans_reserve(*tpp, &tres, 0, 0);