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_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_extent_busy.h"
15 #include "xfs_quota.h"
16 #include "xfs_trans.h"
17 #include "xfs_trans_priv.h"
19 #include "xfs_log_priv.h"
20 #include "xfs_trace.h"
21 #include "xfs_error.h"
22 #include "xfs_defer.h"
23 #include "xfs_inode.h"
24 #include "xfs_dquot_item.h"
25 #include "xfs_dquot.h"
26 #include "xfs_icache.h"
28 struct kmem_cache *xfs_trans_cache;
30 #if defined(CONFIG_TRACEPOINTS)
32 xfs_trans_trace_reservations(
35 struct xfs_trans_res *res;
36 struct xfs_trans_res *end_res;
39 res = (struct xfs_trans_res *)M_RES(mp);
40 end_res = (struct xfs_trans_res *)(M_RES(mp) + 1);
41 for (i = 0; res < end_res; i++, res++)
42 trace_xfs_trans_resv_calc(mp, i, res);
45 # define xfs_trans_trace_reservations(mp)
49 * Initialize the precomputed transaction reservation values
50 * in the mount structure.
56 xfs_trans_resv_calc(mp, M_RES(mp));
57 xfs_trans_trace_reservations(mp);
61 * Free the transaction structure. If there is more clean up
62 * to do when the structure is freed, add it here.
68 xfs_extent_busy_sort(&tp->t_busy);
69 xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
71 trace_xfs_trans_free(tp, _RET_IP_);
72 xfs_trans_clear_context(tp);
73 if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
74 sb_end_intwrite(tp->t_mountp->m_super);
75 xfs_trans_free_dqinfo(tp);
76 kmem_cache_free(xfs_trans_cache, tp);
80 * This is called to create a new transaction which will share the
81 * permanent log reservation of the given transaction. The remaining
82 * unused block and rt extent reservations are also inherited. This
83 * implies that the original transaction is no longer allowed to allocate
84 * blocks. Locks and log items, however, are no inherited. They must
85 * be added to the new transaction explicitly.
87 STATIC struct xfs_trans *
91 struct xfs_trans *ntp;
93 trace_xfs_trans_dup(tp, _RET_IP_);
95 ntp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
98 * Initialize the new transaction structure.
100 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
101 ntp->t_mountp = tp->t_mountp;
102 INIT_LIST_HEAD(&ntp->t_items);
103 INIT_LIST_HEAD(&ntp->t_busy);
104 INIT_LIST_HEAD(&ntp->t_dfops);
105 ntp->t_highest_agno = NULLAGNUMBER;
107 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
108 ASSERT(tp->t_ticket != NULL);
110 ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
111 (tp->t_flags & XFS_TRANS_RESERVE) |
112 (tp->t_flags & XFS_TRANS_NO_WRITECOUNT) |
113 (tp->t_flags & XFS_TRANS_RES_FDBLKS);
114 /* We gave our writer reference to the new transaction */
115 tp->t_flags |= XFS_TRANS_NO_WRITECOUNT;
116 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
118 ASSERT(tp->t_blk_res >= tp->t_blk_res_used);
119 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
120 tp->t_blk_res = tp->t_blk_res_used;
122 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
123 tp->t_rtx_res = tp->t_rtx_res_used;
125 xfs_trans_switch_context(tp, ntp);
127 /* move deferred ops over to the new tp */
128 xfs_defer_move(ntp, tp);
130 xfs_trans_dup_dqinfo(tp, ntp);
135 * This is called to reserve free disk blocks and log space for the
136 * given transaction. This must be done before allocating any resources
137 * within the transaction.
139 * This will return ENOSPC if there are not enough blocks available.
140 * It will sleep waiting for available log space.
141 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
142 * is used by long running transactions. If any one of the reservations
143 * fails then they will all be backed out.
145 * This does not do quota reservations. That typically is done by the
150 struct xfs_trans *tp,
151 struct xfs_trans_res *resp,
155 struct xfs_mount *mp = tp->t_mountp;
157 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
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);
168 tp->t_blk_res += blocks;
172 * Reserve the log space needed for this transaction.
174 if (resp->tr_logres > 0) {
175 bool permanent = false;
177 ASSERT(tp->t_log_res == 0 ||
178 tp->t_log_res == resp->tr_logres);
179 ASSERT(tp->t_log_count == 0 ||
180 tp->t_log_count == resp->tr_logcount);
182 if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
183 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
186 ASSERT(tp->t_ticket == NULL);
187 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
190 if (tp->t_ticket != NULL) {
191 ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
192 error = xfs_log_regrant(mp, tp->t_ticket);
194 error = xfs_log_reserve(mp, resp->tr_logres,
196 &tp->t_ticket, permanent);
202 tp->t_log_res = resp->tr_logres;
203 tp->t_log_count = resp->tr_logcount;
207 * Attempt to reserve the needed realtime extents by decrementing
208 * the number needed from the number available. This will
209 * fail if the count would go below zero.
212 error = xfs_mod_frextents(mp, -((int64_t)rtextents));
217 tp->t_rtx_res += rtextents;
223 * Error cases jump to one of these labels to undo any
224 * reservations which have already been performed.
227 if (resp->tr_logres > 0) {
228 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
231 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
236 xfs_mod_fdblocks(mp, (int64_t)blocks, rsvd);
244 struct xfs_mount *mp,
245 struct xfs_trans_res *resp,
249 struct xfs_trans **tpp)
251 struct xfs_trans *tp;
252 bool want_retry = true;
256 * Allocate the handle before we do our freeze accounting and setting up
257 * GFP_NOFS allocation context so that we avoid lockdep false positives
258 * by doing GFP_KERNEL allocations inside sb_start_intwrite().
261 tp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
262 if (!(flags & XFS_TRANS_NO_WRITECOUNT))
263 sb_start_intwrite(mp->m_super);
264 xfs_trans_set_context(tp);
267 * Zero-reservation ("empty") transactions can't modify anything, so
268 * they're allowed to run while we're frozen.
270 WARN_ON(resp->tr_logres > 0 &&
271 mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
272 ASSERT(!(flags & XFS_TRANS_RES_FDBLKS) ||
273 xfs_has_lazysbcount(mp));
275 tp->t_magic = XFS_TRANS_HEADER_MAGIC;
278 INIT_LIST_HEAD(&tp->t_items);
279 INIT_LIST_HEAD(&tp->t_busy);
280 INIT_LIST_HEAD(&tp->t_dfops);
281 tp->t_highest_agno = NULLAGNUMBER;
283 error = xfs_trans_reserve(tp, resp, blocks, rtextents);
284 if (error == -ENOSPC && want_retry) {
285 xfs_trans_cancel(tp);
288 * We weren't able to reserve enough space for the transaction.
289 * Flush the other speculative space allocations to free space.
290 * Do not perform a synchronous scan because callers can hold
293 error = xfs_blockgc_flush_all(mp);
300 xfs_trans_cancel(tp);
304 trace_xfs_trans_alloc(tp, _RET_IP_);
311 * Create an empty transaction with no reservation. This is a defensive
312 * mechanism for routines that query metadata without actually modifying them --
313 * if the metadata being queried is somehow cross-linked (think a btree block
314 * pointer that points higher in the tree), we risk deadlock. However, blocks
315 * grabbed as part of a transaction can be re-grabbed. The verifiers will
316 * notice the corrupt block and the operation will fail back to userspace
317 * without deadlocking.
319 * Note the zero-length reservation; this transaction MUST be cancelled without
322 * Callers should obtain freeze protection to avoid a conflict with fs freezing
323 * where we can be grabbing buffers at the same time that freeze is trying to
324 * drain the buffer LRU list.
327 xfs_trans_alloc_empty(
328 struct xfs_mount *mp,
329 struct xfs_trans **tpp)
331 struct xfs_trans_res resv = {0};
333 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
337 * Record the indicated change to the given field for application
338 * to the file system's superblock when the transaction commits.
339 * For now, just store the change in the transaction structure.
341 * Mark the transaction structure to indicate that the superblock
342 * needs to be updated before committing.
344 * Because we may not be keeping track of allocated/free inodes and
345 * used filesystem blocks in the superblock, we do not mark the
346 * superblock dirty in this transaction if we modify these fields.
347 * We still need to update the transaction deltas so that they get
348 * applied to the incore superblock, but we don't want them to
349 * cause the superblock to get locked and logged if these are the
350 * only fields in the superblock that the transaction modifies.
358 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
359 xfs_mount_t *mp = tp->t_mountp;
362 case XFS_TRANS_SB_ICOUNT:
363 tp->t_icount_delta += delta;
364 if (xfs_has_lazysbcount(mp))
365 flags &= ~XFS_TRANS_SB_DIRTY;
367 case XFS_TRANS_SB_IFREE:
368 tp->t_ifree_delta += delta;
369 if (xfs_has_lazysbcount(mp))
370 flags &= ~XFS_TRANS_SB_DIRTY;
372 case XFS_TRANS_SB_FDBLOCKS:
374 * Track the number of blocks allocated in the transaction.
375 * Make sure it does not exceed the number reserved. If so,
376 * shutdown as this can lead to accounting inconsistency.
379 tp->t_blk_res_used += (uint)-delta;
380 if (tp->t_blk_res_used > tp->t_blk_res)
381 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
382 } else if (delta > 0 && (tp->t_flags & XFS_TRANS_RES_FDBLKS)) {
383 int64_t blkres_delta;
386 * Return freed blocks directly to the reservation
387 * instead of the global pool, being careful not to
388 * overflow the trans counter. This is used to preserve
389 * reservation across chains of transaction rolls that
390 * repeatedly free and allocate blocks.
392 blkres_delta = min_t(int64_t, delta,
393 UINT_MAX - tp->t_blk_res);
394 tp->t_blk_res += blkres_delta;
395 delta -= blkres_delta;
397 tp->t_fdblocks_delta += delta;
398 if (xfs_has_lazysbcount(mp))
399 flags &= ~XFS_TRANS_SB_DIRTY;
401 case XFS_TRANS_SB_RES_FDBLOCKS:
403 * The allocation has already been applied to the
404 * in-core superblock's counter. This should only
405 * be applied to the on-disk superblock.
407 tp->t_res_fdblocks_delta += delta;
408 if (xfs_has_lazysbcount(mp))
409 flags &= ~XFS_TRANS_SB_DIRTY;
411 case XFS_TRANS_SB_FREXTENTS:
413 * Track the number of blocks allocated in the
414 * transaction. Make sure it does not exceed the
418 tp->t_rtx_res_used += (uint)-delta;
419 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
421 tp->t_frextents_delta += delta;
423 case XFS_TRANS_SB_RES_FREXTENTS:
425 * The allocation has already been applied to the
426 * in-core superblock's counter. This should only
427 * be applied to the on-disk superblock.
430 tp->t_res_frextents_delta += delta;
432 case XFS_TRANS_SB_DBLOCKS:
433 tp->t_dblocks_delta += delta;
435 case XFS_TRANS_SB_AGCOUNT:
437 tp->t_agcount_delta += delta;
439 case XFS_TRANS_SB_IMAXPCT:
440 tp->t_imaxpct_delta += delta;
442 case XFS_TRANS_SB_REXTSIZE:
443 tp->t_rextsize_delta += delta;
445 case XFS_TRANS_SB_RBMBLOCKS:
446 tp->t_rbmblocks_delta += delta;
448 case XFS_TRANS_SB_RBLOCKS:
449 tp->t_rblocks_delta += delta;
451 case XFS_TRANS_SB_REXTENTS:
452 tp->t_rextents_delta += delta;
454 case XFS_TRANS_SB_REXTSLOG:
455 tp->t_rextslog_delta += delta;
462 tp->t_flags |= flags;
466 * xfs_trans_apply_sb_deltas() is called from the commit code
467 * to bring the superblock buffer into the current transaction
468 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
470 * For now we just look at each field allowed to change and change
474 xfs_trans_apply_sb_deltas(
481 bp = xfs_trans_getsb(tp);
485 * Only update the superblock counters if we are logging them
487 if (!xfs_has_lazysbcount((tp->t_mountp))) {
488 if (tp->t_icount_delta)
489 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
490 if (tp->t_ifree_delta)
491 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
492 if (tp->t_fdblocks_delta)
493 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
494 if (tp->t_res_fdblocks_delta)
495 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
499 * Updating frextents requires careful handling because it does not
500 * behave like the lazysb counters because we cannot rely on log
501 * recovery in older kenels to recompute the value from the rtbitmap.
502 * This means that the ondisk frextents must be consistent with the
505 * Therefore, log the frextents change to the ondisk superblock and
506 * update the incore superblock so that future calls to xfs_log_sb
507 * write the correct value ondisk.
509 * Don't touch m_frextents because it includes incore reservations,
510 * and those are handled by the unreserve function.
512 if (tp->t_frextents_delta || tp->t_res_frextents_delta) {
513 struct xfs_mount *mp = tp->t_mountp;
516 rtxdelta = tp->t_frextents_delta + tp->t_res_frextents_delta;
518 spin_lock(&mp->m_sb_lock);
519 be64_add_cpu(&sbp->sb_frextents, rtxdelta);
520 mp->m_sb.sb_frextents += rtxdelta;
521 spin_unlock(&mp->m_sb_lock);
524 if (tp->t_dblocks_delta) {
525 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
528 if (tp->t_agcount_delta) {
529 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
532 if (tp->t_imaxpct_delta) {
533 sbp->sb_imax_pct += tp->t_imaxpct_delta;
536 if (tp->t_rextsize_delta) {
537 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
540 if (tp->t_rbmblocks_delta) {
541 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
544 if (tp->t_rblocks_delta) {
545 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
548 if (tp->t_rextents_delta) {
549 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
552 if (tp->t_rextslog_delta) {
553 sbp->sb_rextslog += tp->t_rextslog_delta;
557 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
560 * Log the whole thing, the fields are noncontiguous.
562 xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
565 * Since all the modifiable fields are contiguous, we
566 * can get away with this.
568 xfs_trans_log_buf(tp, bp, offsetof(struct xfs_dsb, sb_icount),
569 offsetof(struct xfs_dsb, sb_frextents) +
570 sizeof(sbp->sb_frextents) - 1);
574 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations and
575 * apply superblock counter changes to the in-core superblock. The
576 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
577 * applied to the in-core superblock. The idea is that that has already been
580 * If we are not logging superblock counters, then the inode allocated/free and
581 * used block counts are not updated in the on disk superblock. In this case,
582 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
583 * still need to update the incore superblock with the changes.
585 * Deltas for the inode count are +/-64, hence we use a large batch size of 128
586 * so we don't need to take the counter lock on every update.
588 #define XFS_ICOUNT_BATCH 128
591 xfs_trans_unreserve_and_mod_sb(
592 struct xfs_trans *tp)
594 struct xfs_mount *mp = tp->t_mountp;
595 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
596 int64_t blkdelta = 0;
597 int64_t rtxdelta = 0;
599 int64_t ifreedelta = 0;
602 /* calculate deltas */
603 if (tp->t_blk_res > 0)
604 blkdelta = tp->t_blk_res;
605 if ((tp->t_fdblocks_delta != 0) &&
606 (xfs_has_lazysbcount(mp) ||
607 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
608 blkdelta += tp->t_fdblocks_delta;
610 if (tp->t_rtx_res > 0)
611 rtxdelta = tp->t_rtx_res;
612 if ((tp->t_frextents_delta != 0) &&
613 (tp->t_flags & XFS_TRANS_SB_DIRTY))
614 rtxdelta += tp->t_frextents_delta;
616 if (xfs_has_lazysbcount(mp) ||
617 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
618 idelta = tp->t_icount_delta;
619 ifreedelta = tp->t_ifree_delta;
622 /* apply the per-cpu counters */
624 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
629 percpu_counter_add_batch(&mp->m_icount, idelta,
633 percpu_counter_add(&mp->m_ifree, ifreedelta);
636 error = xfs_mod_frextents(mp, rtxdelta);
640 if (!(tp->t_flags & XFS_TRANS_SB_DIRTY))
643 /* apply remaining deltas */
644 spin_lock(&mp->m_sb_lock);
645 mp->m_sb.sb_fdblocks += tp->t_fdblocks_delta + tp->t_res_fdblocks_delta;
646 mp->m_sb.sb_icount += idelta;
647 mp->m_sb.sb_ifree += ifreedelta;
649 * Do not touch sb_frextents here because we are dealing with incore
650 * reservation. sb_frextents is not part of the lazy sb counters so it
651 * must be consistent with the ondisk rtbitmap and must never include
652 * incore reservations.
654 mp->m_sb.sb_dblocks += tp->t_dblocks_delta;
655 mp->m_sb.sb_agcount += tp->t_agcount_delta;
656 mp->m_sb.sb_imax_pct += tp->t_imaxpct_delta;
657 mp->m_sb.sb_rextsize += tp->t_rextsize_delta;
658 mp->m_sb.sb_rbmblocks += tp->t_rbmblocks_delta;
659 mp->m_sb.sb_rblocks += tp->t_rblocks_delta;
660 mp->m_sb.sb_rextents += tp->t_rextents_delta;
661 mp->m_sb.sb_rextslog += tp->t_rextslog_delta;
662 spin_unlock(&mp->m_sb_lock);
665 * Debug checks outside of the spinlock so they don't lock up the
666 * machine if they fail.
668 ASSERT(mp->m_sb.sb_imax_pct >= 0);
669 ASSERT(mp->m_sb.sb_rextslog >= 0);
673 /* Add the given log item to the transaction's list of log items. */
676 struct xfs_trans *tp,
677 struct xfs_log_item *lip)
679 ASSERT(lip->li_log == tp->t_mountp->m_log);
680 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
681 ASSERT(list_empty(&lip->li_trans));
682 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->li_flags));
684 list_add_tail(&lip->li_trans, &tp->t_items);
685 trace_xfs_trans_add_item(tp, _RET_IP_);
689 * Unlink the log item from the transaction. the log item is no longer
690 * considered dirty in this transaction, as the linked transaction has
691 * finished, either by abort or commit completion.
695 struct xfs_log_item *lip)
697 clear_bit(XFS_LI_DIRTY, &lip->li_flags);
698 list_del_init(&lip->li_trans);
701 /* Detach and unlock all of the items in a transaction */
703 xfs_trans_free_items(
704 struct xfs_trans *tp,
707 struct xfs_log_item *lip, *next;
709 trace_xfs_trans_free_items(tp, _RET_IP_);
711 list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
712 xfs_trans_del_item(lip);
714 set_bit(XFS_LI_ABORTED, &lip->li_flags);
715 if (lip->li_ops->iop_release)
716 lip->li_ops->iop_release(lip);
721 xfs_log_item_batch_insert(
722 struct xfs_ail *ailp,
723 struct xfs_ail_cursor *cur,
724 struct xfs_log_item **log_items,
726 xfs_lsn_t commit_lsn)
730 spin_lock(&ailp->ail_lock);
731 /* xfs_trans_ail_update_bulk drops ailp->ail_lock */
732 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
734 for (i = 0; i < nr_items; i++) {
735 struct xfs_log_item *lip = log_items[i];
737 if (lip->li_ops->iop_unpin)
738 lip->li_ops->iop_unpin(lip, 0);
743 * Bulk operation version of xfs_trans_committed that takes a log vector of
744 * items to insert into the AIL. This uses bulk AIL insertion techniques to
745 * minimise lock traffic.
747 * If we are called with the aborted flag set, it is because a log write during
748 * a CIL checkpoint commit has failed. In this case, all the items in the
749 * checkpoint have already gone through iop_committed and iop_committing, which
750 * means that checkpoint commit abort handling is treated exactly the same
751 * as an iclog write error even though we haven't started any IO yet. Hence in
752 * this case all we need to do is iop_committed processing, followed by an
753 * iop_unpin(aborted) call.
755 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
756 * at the end of the AIL, the insert cursor avoids the need to walk
757 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
758 * call. This saves a lot of needless list walking and is a net win, even
759 * though it slightly increases that amount of AIL lock traffic to set it up
763 xfs_trans_committed_bulk(
764 struct xfs_ail *ailp,
765 struct list_head *lv_chain,
766 xfs_lsn_t commit_lsn,
769 #define LOG_ITEM_BATCH_SIZE 32
770 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
771 struct xfs_log_vec *lv;
772 struct xfs_ail_cursor cur;
775 spin_lock(&ailp->ail_lock);
776 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
777 spin_unlock(&ailp->ail_lock);
779 /* unpin all the log items */
780 list_for_each_entry(lv, lv_chain, lv_list) {
781 struct xfs_log_item *lip = lv->lv_item;
785 set_bit(XFS_LI_ABORTED, &lip->li_flags);
787 if (lip->li_ops->flags & XFS_ITEM_RELEASE_WHEN_COMMITTED) {
788 lip->li_ops->iop_release(lip);
792 if (lip->li_ops->iop_committed)
793 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
795 item_lsn = commit_lsn;
797 /* item_lsn of -1 means the item needs no further processing */
798 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
802 * if we are aborting the operation, no point in inserting the
803 * object into the AIL as we are in a shutdown situation.
806 ASSERT(xlog_is_shutdown(ailp->ail_log));
807 if (lip->li_ops->iop_unpin)
808 lip->li_ops->iop_unpin(lip, 1);
812 if (item_lsn != commit_lsn) {
815 * Not a bulk update option due to unusual item_lsn.
816 * Push into AIL immediately, rechecking the lsn once
817 * we have the ail lock. Then unpin the item. This does
818 * not affect the AIL cursor the bulk insert path is
821 spin_lock(&ailp->ail_lock);
822 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
823 xfs_trans_ail_update(ailp, lip, item_lsn);
825 spin_unlock(&ailp->ail_lock);
826 if (lip->li_ops->iop_unpin)
827 lip->li_ops->iop_unpin(lip, 0);
831 /* Item is a candidate for bulk AIL insert. */
832 log_items[i++] = lv->lv_item;
833 if (i >= LOG_ITEM_BATCH_SIZE) {
834 xfs_log_item_batch_insert(ailp, &cur, log_items,
835 LOG_ITEM_BATCH_SIZE, commit_lsn);
840 /* make sure we insert the remainder! */
842 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
844 spin_lock(&ailp->ail_lock);
845 xfs_trans_ail_cursor_done(&cur);
846 spin_unlock(&ailp->ail_lock);
850 * Sort transaction items prior to running precommit operations. This will
851 * attempt to order the items such that they will always be locked in the same
852 * order. Items that have no sort function are moved to the end of the list
853 * and so are locked last.
855 * This may need refinement as different types of objects add sort functions.
857 * Function is more complex than it needs to be because we are comparing 64 bit
858 * values and the function only returns 32 bit values.
861 xfs_trans_precommit_sort(
863 const struct list_head *a,
864 const struct list_head *b)
866 struct xfs_log_item *lia = container_of(a,
867 struct xfs_log_item, li_trans);
868 struct xfs_log_item *lib = container_of(b,
869 struct xfs_log_item, li_trans);
873 * If both items are non-sortable, leave them alone. If only one is
874 * sortable, move the non-sortable item towards the end of the list.
876 if (!lia->li_ops->iop_sort && !lib->li_ops->iop_sort)
878 if (!lia->li_ops->iop_sort)
880 if (!lib->li_ops->iop_sort)
883 diff = lia->li_ops->iop_sort(lia) - lib->li_ops->iop_sort(lib);
892 * Run transaction precommit functions.
894 * If there is an error in any of the callouts, then stop immediately and
895 * trigger a shutdown to abort the transaction. There is no recovery possible
896 * from errors at this point as the transaction is dirty....
899 xfs_trans_run_precommits(
900 struct xfs_trans *tp)
902 struct xfs_mount *mp = tp->t_mountp;
903 struct xfs_log_item *lip, *n;
907 * Sort the item list to avoid ABBA deadlocks with other transactions
908 * running precommit operations that lock multiple shared items such as
909 * inode cluster buffers.
911 list_sort(NULL, &tp->t_items, xfs_trans_precommit_sort);
914 * Precommit operations can remove the log item from the transaction
915 * if the log item exists purely to delay modifications until they
916 * can be ordered against other operations. Hence we have to use
917 * list_for_each_entry_safe() here.
919 list_for_each_entry_safe(lip, n, &tp->t_items, li_trans) {
920 if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
922 if (lip->li_ops->iop_precommit) {
923 error = lip->li_ops->iop_precommit(tp, lip);
929 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
934 * Commit the given transaction to the log.
936 * XFS disk error handling mechanism is not based on a typical
937 * transaction abort mechanism. Logically after the filesystem
938 * gets marked 'SHUTDOWN', we can't let any new transactions
939 * be durable - ie. committed to disk - because some metadata might
940 * be inconsistent. In such cases, this returns an error, and the
941 * caller may assume that all locked objects joined to the transaction
942 * have already been unlocked as if the commit had succeeded.
943 * Do not reference the transaction structure after this call.
947 struct xfs_trans *tp,
950 struct xfs_mount *mp = tp->t_mountp;
951 struct xlog *log = mp->m_log;
952 xfs_csn_t commit_seq = 0;
954 int sync = tp->t_flags & XFS_TRANS_SYNC;
956 trace_xfs_trans_commit(tp, _RET_IP_);
958 error = xfs_trans_run_precommits(tp);
960 if (tp->t_flags & XFS_TRANS_PERM_LOG_RES)
961 xfs_defer_cancel(tp);
966 * Finish deferred items on final commit. Only permanent transactions
967 * should ever have deferred ops.
969 WARN_ON_ONCE(!list_empty(&tp->t_dfops) &&
970 !(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
971 if (!regrant && (tp->t_flags & XFS_TRANS_PERM_LOG_RES)) {
972 error = xfs_defer_finish_noroll(&tp);
976 /* Run precommits from final tx in defer chain. */
977 error = xfs_trans_run_precommits(tp);
983 * If there is nothing to be logged by the transaction,
984 * then unlock all of the items associated with the
985 * transaction and free the transaction structure.
986 * Also make sure to return any reserved blocks to
989 if (!(tp->t_flags & XFS_TRANS_DIRTY))
993 * We must check against log shutdown here because we cannot abort log
994 * items and leave them dirty, inconsistent and unpinned in memory while
995 * the log is active. This leaves them open to being written back to
996 * disk, and that will lead to on-disk corruption.
998 if (xlog_is_shutdown(log)) {
1003 ASSERT(tp->t_ticket != NULL);
1006 * If we need to update the superblock, then do it now.
1008 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
1009 xfs_trans_apply_sb_deltas(tp);
1010 xfs_trans_apply_dquot_deltas(tp);
1012 xlog_cil_commit(log, tp, &commit_seq, regrant);
1017 * If the transaction needs to be synchronous, then force the
1018 * log out now and wait for it.
1021 error = xfs_log_force_seq(mp, commit_seq, XFS_LOG_SYNC, NULL);
1022 XFS_STATS_INC(mp, xs_trans_sync);
1024 XFS_STATS_INC(mp, xs_trans_async);
1030 xfs_trans_unreserve_and_mod_sb(tp);
1033 * It is indeed possible for the transaction to be not dirty but
1034 * the dqinfo portion to be. All that means is that we have some
1035 * (non-persistent) quota reservations that need to be unreserved.
1037 xfs_trans_unreserve_and_mod_dquots(tp);
1039 if (regrant && !xlog_is_shutdown(log))
1040 xfs_log_ticket_regrant(log, tp->t_ticket);
1042 xfs_log_ticket_ungrant(log, tp->t_ticket);
1043 tp->t_ticket = NULL;
1045 xfs_trans_free_items(tp, !!error);
1048 XFS_STATS_INC(mp, xs_trans_empty);
1054 struct xfs_trans *tp)
1056 return __xfs_trans_commit(tp, false);
1060 * Unlock all of the transaction's items and free the transaction. If the
1061 * transaction is dirty, we must shut down the filesystem because there is no
1062 * way to restore them to their previous state.
1064 * If the transaction has made a log reservation, make sure to release it as
1067 * This is a high level function (equivalent to xfs_trans_commit()) and so can
1068 * be called after the transaction has effectively been aborted due to the mount
1069 * being shut down. However, if the mount has not been shut down and the
1070 * transaction is dirty we will shut the mount down and, in doing so, that
1071 * guarantees that the log is shut down, too. Hence we don't need to be as
1072 * careful with shutdown state and dirty items here as we need to be in
1073 * xfs_trans_commit().
1077 struct xfs_trans *tp)
1079 struct xfs_mount *mp = tp->t_mountp;
1080 struct xlog *log = mp->m_log;
1081 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
1083 trace_xfs_trans_cancel(tp, _RET_IP_);
1086 * It's never valid to cancel a transaction with deferred ops attached,
1087 * because the transaction is effectively dirty. Complain about this
1088 * loudly before freeing the in-memory defer items and shutting down the
1091 if (!list_empty(&tp->t_dfops)) {
1092 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1094 xfs_defer_cancel(tp);
1098 * See if the caller is relying on us to shut down the filesystem. We
1099 * only want an error report if there isn't already a shutdown in
1100 * progress, so we only need to check against the mount shutdown state
1103 if (dirty && !xfs_is_shutdown(mp)) {
1104 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1105 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1108 /* Log items need to be consistent until the log is shut down. */
1109 if (!dirty && !xlog_is_shutdown(log)) {
1110 struct xfs_log_item *lip;
1112 list_for_each_entry(lip, &tp->t_items, li_trans)
1113 ASSERT(!xlog_item_is_intent_done(lip));
1116 xfs_trans_unreserve_and_mod_sb(tp);
1117 xfs_trans_unreserve_and_mod_dquots(tp);
1120 xfs_log_ticket_ungrant(log, tp->t_ticket);
1121 tp->t_ticket = NULL;
1124 xfs_trans_free_items(tp, dirty);
1129 * Roll from one trans in the sequence of PERMANENT transactions to
1130 * the next: permanent transactions are only flushed out when
1131 * committed with xfs_trans_commit(), but we still want as soon
1132 * as possible to let chunks of it go to the log. So we commit the
1133 * chunk we've been working on and get a new transaction to continue.
1137 struct xfs_trans **tpp)
1139 struct xfs_trans *trans = *tpp;
1140 struct xfs_trans_res tres;
1143 trace_xfs_trans_roll(trans, _RET_IP_);
1146 * Copy the critical parameters from one trans to the next.
1148 tres.tr_logres = trans->t_log_res;
1149 tres.tr_logcount = trans->t_log_count;
1151 *tpp = xfs_trans_dup(trans);
1154 * Commit the current transaction.
1155 * If this commit failed, then it'd just unlock those items that
1156 * are not marked ihold. That also means that a filesystem shutdown
1157 * is in progress. The caller takes the responsibility to cancel
1158 * the duplicate transaction that gets returned.
1160 error = __xfs_trans_commit(trans, true);
1165 * Reserve space in the log for the next transaction.
1166 * This also pushes items in the "AIL", the list of logged items,
1167 * out to disk if they are taking up space at the tail of the log
1168 * that we want to use. This requires that either nothing be locked
1169 * across this call, or that anything that is locked be logged in
1170 * the prior and the next transactions.
1172 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1173 return xfs_trans_reserve(*tpp, &tres, 0, 0);
1177 * Allocate an transaction, lock and join the inode to it, and reserve quota.
1179 * The caller must ensure that the on-disk dquots attached to this inode have
1180 * already been allocated and initialized. The caller is responsible for
1181 * releasing ILOCK_EXCL if a new transaction is returned.
1184 xfs_trans_alloc_inode(
1185 struct xfs_inode *ip,
1186 struct xfs_trans_res *resv,
1187 unsigned int dblocks,
1188 unsigned int rblocks,
1190 struct xfs_trans **tpp)
1192 struct xfs_trans *tp;
1193 struct xfs_mount *mp = ip->i_mount;
1194 bool retried = false;
1198 error = xfs_trans_alloc(mp, resv, dblocks,
1199 rblocks / mp->m_sb.sb_rextsize,
1200 force ? XFS_TRANS_RESERVE : 0, &tp);
1204 xfs_ilock(ip, XFS_ILOCK_EXCL);
1205 xfs_trans_ijoin(tp, ip, 0);
1207 error = xfs_qm_dqattach_locked(ip, false);
1209 /* Caller should have allocated the dquots! */
1210 ASSERT(error != -ENOENT);
1214 error = xfs_trans_reserve_quota_nblks(tp, ip, dblocks, rblocks, force);
1215 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1216 xfs_trans_cancel(tp);
1217 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1218 xfs_blockgc_free_quota(ip, 0);
1229 xfs_trans_cancel(tp);
1230 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1235 * Allocate an transaction in preparation for inode creation by reserving quota
1236 * against the given dquots. Callers are not required to hold any inode locks.
1239 xfs_trans_alloc_icreate(
1240 struct xfs_mount *mp,
1241 struct xfs_trans_res *resv,
1242 struct xfs_dquot *udqp,
1243 struct xfs_dquot *gdqp,
1244 struct xfs_dquot *pdqp,
1245 unsigned int dblocks,
1246 struct xfs_trans **tpp)
1248 struct xfs_trans *tp;
1249 bool retried = false;
1253 error = xfs_trans_alloc(mp, resv, dblocks, 0, 0, &tp);
1257 error = xfs_trans_reserve_quota_icreate(tp, udqp, gdqp, pdqp, dblocks);
1258 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1259 xfs_trans_cancel(tp);
1260 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1265 xfs_trans_cancel(tp);
1274 * Allocate an transaction, lock and join the inode to it, and reserve quota
1275 * in preparation for inode attribute changes that include uid, gid, or prid
1278 * The caller must ensure that the on-disk dquots attached to this inode have
1279 * already been allocated and initialized. The ILOCK will be dropped when the
1280 * transaction is committed or cancelled.
1283 xfs_trans_alloc_ichange(
1284 struct xfs_inode *ip,
1285 struct xfs_dquot *new_udqp,
1286 struct xfs_dquot *new_gdqp,
1287 struct xfs_dquot *new_pdqp,
1289 struct xfs_trans **tpp)
1291 struct xfs_trans *tp;
1292 struct xfs_mount *mp = ip->i_mount;
1293 struct xfs_dquot *udqp;
1294 struct xfs_dquot *gdqp;
1295 struct xfs_dquot *pdqp;
1296 bool retried = false;
1300 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1304 xfs_ilock(ip, XFS_ILOCK_EXCL);
1305 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1307 error = xfs_qm_dqattach_locked(ip, false);
1309 /* Caller should have allocated the dquots! */
1310 ASSERT(error != -ENOENT);
1315 * For each quota type, skip quota reservations if the inode's dquots
1316 * now match the ones that came from the caller, or the caller didn't
1317 * pass one in. The inode's dquots can change if we drop the ILOCK to
1318 * perform a blockgc scan, so we must preserve the caller's arguments.
1320 udqp = (new_udqp != ip->i_udquot) ? new_udqp : NULL;
1321 gdqp = (new_gdqp != ip->i_gdquot) ? new_gdqp : NULL;
1322 pdqp = (new_pdqp != ip->i_pdquot) ? new_pdqp : NULL;
1323 if (udqp || gdqp || pdqp) {
1324 unsigned int qflags = XFS_QMOPT_RES_REGBLKS;
1327 qflags |= XFS_QMOPT_FORCE_RES;
1330 * Reserve enough quota to handle blocks on disk and reserved
1331 * for a delayed allocation. We'll actually transfer the
1332 * delalloc reservation between dquots at chown time, even
1333 * though that part is only semi-transactional.
1335 error = xfs_trans_reserve_quota_bydquots(tp, mp, udqp, gdqp,
1336 pdqp, ip->i_nblocks + ip->i_delayed_blks,
1338 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1339 xfs_trans_cancel(tp);
1340 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1352 xfs_trans_cancel(tp);
1357 * Allocate an transaction, lock and join the directory and child inodes to it,
1358 * and reserve quota for a directory update. If there isn't sufficient space,
1359 * @dblocks will be set to zero for a reservationless directory update and
1360 * @nospace_error will be set to a negative errno describing the space
1361 * constraint we hit.
1363 * The caller must ensure that the on-disk dquots attached to this inode have
1364 * already been allocated and initialized. The ILOCKs will be dropped when the
1365 * transaction is committed or cancelled.
1368 xfs_trans_alloc_dir(
1369 struct xfs_inode *dp,
1370 struct xfs_trans_res *resv,
1371 struct xfs_inode *ip,
1372 unsigned int *dblocks,
1373 struct xfs_trans **tpp,
1376 struct xfs_trans *tp;
1377 struct xfs_mount *mp = ip->i_mount;
1378 unsigned int resblks;
1379 bool retried = false;
1385 error = xfs_trans_alloc(mp, resv, resblks, 0, 0, &tp);
1386 if (error == -ENOSPC) {
1387 *nospace_error = error;
1389 error = xfs_trans_alloc(mp, resv, resblks, 0, 0, &tp);
1394 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
1396 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1397 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1399 error = xfs_qm_dqattach_locked(dp, false);
1401 /* Caller should have allocated the dquots! */
1402 ASSERT(error != -ENOENT);
1406 error = xfs_qm_dqattach_locked(ip, false);
1408 /* Caller should have allocated the dquots! */
1409 ASSERT(error != -ENOENT);
1416 error = xfs_trans_reserve_quota_nblks(tp, dp, resblks, 0, false);
1417 if (error == -EDQUOT || error == -ENOSPC) {
1419 xfs_trans_cancel(tp);
1420 xfs_blockgc_free_quota(dp, 0);
1425 *nospace_error = error;
1438 xfs_trans_cancel(tp);