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 resv;
36 struct xfs_trans_res *res;
37 struct xfs_trans_res *end_res;
40 res = (struct xfs_trans_res *)M_RES(mp);
41 end_res = (struct xfs_trans_res *)(M_RES(mp) + 1);
42 for (i = 0; res < end_res; i++, res++)
43 trace_xfs_trans_resv_calc(mp, i, res);
44 xfs_log_get_max_trans_res(mp, &resv);
45 trace_xfs_trans_resv_calc(mp, -1, &resv);
48 # define xfs_trans_trace_reservations(mp)
52 * Initialize the precomputed transaction reservation values
53 * in the mount structure.
59 xfs_trans_resv_calc(mp, M_RES(mp));
60 xfs_trans_trace_reservations(mp);
64 * Free the transaction structure. If there is more clean up
65 * to do when the structure is freed, add it here.
71 xfs_extent_busy_sort(&tp->t_busy);
72 xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
74 trace_xfs_trans_free(tp, _RET_IP_);
75 xfs_trans_clear_context(tp);
76 if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
77 sb_end_intwrite(tp->t_mountp->m_super);
78 xfs_trans_free_dqinfo(tp);
79 kmem_cache_free(xfs_trans_cache, tp);
83 * This is called to create a new transaction which will share the
84 * permanent log reservation of the given transaction. The remaining
85 * unused block and rt extent reservations are also inherited. This
86 * implies that the original transaction is no longer allowed to allocate
87 * blocks. Locks and log items, however, are no inherited. They must
88 * be added to the new transaction explicitly.
90 STATIC struct xfs_trans *
94 struct xfs_trans *ntp;
96 trace_xfs_trans_dup(tp, _RET_IP_);
98 ntp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
101 * Initialize the new transaction structure.
103 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
104 ntp->t_mountp = tp->t_mountp;
105 INIT_LIST_HEAD(&ntp->t_items);
106 INIT_LIST_HEAD(&ntp->t_busy);
107 INIT_LIST_HEAD(&ntp->t_dfops);
108 ntp->t_firstblock = NULLFSBLOCK;
110 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
111 ASSERT(tp->t_ticket != NULL);
113 ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
114 (tp->t_flags & XFS_TRANS_RESERVE) |
115 (tp->t_flags & XFS_TRANS_NO_WRITECOUNT) |
116 (tp->t_flags & XFS_TRANS_RES_FDBLKS);
117 /* We gave our writer reference to the new transaction */
118 tp->t_flags |= XFS_TRANS_NO_WRITECOUNT;
119 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
121 ASSERT(tp->t_blk_res >= tp->t_blk_res_used);
122 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
123 tp->t_blk_res = tp->t_blk_res_used;
125 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
126 tp->t_rtx_res = tp->t_rtx_res_used;
128 xfs_trans_switch_context(tp, ntp);
130 /* move deferred ops over to the new tp */
131 xfs_defer_move(ntp, tp);
133 xfs_trans_dup_dqinfo(tp, ntp);
138 * This is called to reserve free disk blocks and log space for the
139 * given transaction. This must be done before allocating any resources
140 * within the transaction.
142 * This will return ENOSPC if there are not enough blocks available.
143 * It will sleep waiting for available log space.
144 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
145 * is used by long running transactions. If any one of the reservations
146 * fails then they will all be backed out.
148 * This does not do quota reservations. That typically is done by the
153 struct xfs_trans *tp,
154 struct xfs_trans_res *resp,
158 struct xfs_mount *mp = tp->t_mountp;
160 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
163 * Attempt to reserve the needed disk blocks by decrementing
164 * the number needed from the number available. This will
165 * fail if the count would go below zero.
168 error = xfs_mod_fdblocks(mp, -((int64_t)blocks), rsvd);
171 tp->t_blk_res += blocks;
175 * Reserve the log space needed for this transaction.
177 if (resp->tr_logres > 0) {
178 bool permanent = false;
180 ASSERT(tp->t_log_res == 0 ||
181 tp->t_log_res == resp->tr_logres);
182 ASSERT(tp->t_log_count == 0 ||
183 tp->t_log_count == resp->tr_logcount);
185 if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
186 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
189 ASSERT(tp->t_ticket == NULL);
190 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
193 if (tp->t_ticket != NULL) {
194 ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
195 error = xfs_log_regrant(mp, tp->t_ticket);
197 error = xfs_log_reserve(mp,
200 &tp->t_ticket, XFS_TRANSACTION,
207 tp->t_log_res = resp->tr_logres;
208 tp->t_log_count = resp->tr_logcount;
212 * Attempt to reserve the needed realtime extents by decrementing
213 * the number needed from the number available. This will
214 * fail if the count would go below zero.
217 error = xfs_mod_frextents(mp, -((int64_t)rtextents));
222 tp->t_rtx_res += rtextents;
228 * Error cases jump to one of these labels to undo any
229 * reservations which have already been performed.
232 if (resp->tr_logres > 0) {
233 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
236 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
241 xfs_mod_fdblocks(mp, (int64_t)blocks, rsvd);
249 struct xfs_mount *mp,
250 struct xfs_trans_res *resp,
254 struct xfs_trans **tpp)
256 struct xfs_trans *tp;
257 bool want_retry = true;
261 * Allocate the handle before we do our freeze accounting and setting up
262 * GFP_NOFS allocation context so that we avoid lockdep false positives
263 * by doing GFP_KERNEL allocations inside sb_start_intwrite().
266 tp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
267 if (!(flags & XFS_TRANS_NO_WRITECOUNT))
268 sb_start_intwrite(mp->m_super);
269 xfs_trans_set_context(tp);
272 * Zero-reservation ("empty") transactions can't modify anything, so
273 * they're allowed to run while we're frozen.
275 WARN_ON(resp->tr_logres > 0 &&
276 mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
277 ASSERT(!(flags & XFS_TRANS_RES_FDBLKS) ||
278 xfs_has_lazysbcount(mp));
280 tp->t_magic = XFS_TRANS_HEADER_MAGIC;
283 INIT_LIST_HEAD(&tp->t_items);
284 INIT_LIST_HEAD(&tp->t_busy);
285 INIT_LIST_HEAD(&tp->t_dfops);
286 tp->t_firstblock = NULLFSBLOCK;
288 error = xfs_trans_reserve(tp, resp, blocks, rtextents);
289 if (error == -ENOSPC && want_retry) {
290 xfs_trans_cancel(tp);
293 * We weren't able to reserve enough space for the transaction.
294 * Flush the other speculative space allocations to free space.
295 * Do not perform a synchronous scan because callers can hold
298 xfs_blockgc_flush_all(mp);
303 xfs_trans_cancel(tp);
307 trace_xfs_trans_alloc(tp, _RET_IP_);
314 * Create an empty transaction with no reservation. This is a defensive
315 * mechanism for routines that query metadata without actually modifying them --
316 * if the metadata being queried is somehow cross-linked (think a btree block
317 * pointer that points higher in the tree), we risk deadlock. However, blocks
318 * grabbed as part of a transaction can be re-grabbed. The verifiers will
319 * notice the corrupt block and the operation will fail back to userspace
320 * without deadlocking.
322 * Note the zero-length reservation; this transaction MUST be cancelled without
325 * Callers should obtain freeze protection to avoid a conflict with fs freezing
326 * where we can be grabbing buffers at the same time that freeze is trying to
327 * drain the buffer LRU list.
330 xfs_trans_alloc_empty(
331 struct xfs_mount *mp,
332 struct xfs_trans **tpp)
334 struct xfs_trans_res resv = {0};
336 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
340 * Record the indicated change to the given field for application
341 * to the file system's superblock when the transaction commits.
342 * For now, just store the change in the transaction structure.
344 * Mark the transaction structure to indicate that the superblock
345 * needs to be updated before committing.
347 * Because we may not be keeping track of allocated/free inodes and
348 * used filesystem blocks in the superblock, we do not mark the
349 * superblock dirty in this transaction if we modify these fields.
350 * We still need to update the transaction deltas so that they get
351 * applied to the incore superblock, but we don't want them to
352 * cause the superblock to get locked and logged if these are the
353 * only fields in the superblock that the transaction modifies.
361 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
362 xfs_mount_t *mp = tp->t_mountp;
365 case XFS_TRANS_SB_ICOUNT:
366 tp->t_icount_delta += delta;
367 if (xfs_has_lazysbcount(mp))
368 flags &= ~XFS_TRANS_SB_DIRTY;
370 case XFS_TRANS_SB_IFREE:
371 tp->t_ifree_delta += delta;
372 if (xfs_has_lazysbcount(mp))
373 flags &= ~XFS_TRANS_SB_DIRTY;
375 case XFS_TRANS_SB_FDBLOCKS:
377 * Track the number of blocks allocated in the transaction.
378 * Make sure it does not exceed the number reserved. If so,
379 * shutdown as this can lead to accounting inconsistency.
382 tp->t_blk_res_used += (uint)-delta;
383 if (tp->t_blk_res_used > tp->t_blk_res)
384 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
385 } else if (delta > 0 && (tp->t_flags & XFS_TRANS_RES_FDBLKS)) {
386 int64_t blkres_delta;
389 * Return freed blocks directly to the reservation
390 * instead of the global pool, being careful not to
391 * overflow the trans counter. This is used to preserve
392 * reservation across chains of transaction rolls that
393 * repeatedly free and allocate blocks.
395 blkres_delta = min_t(int64_t, delta,
396 UINT_MAX - tp->t_blk_res);
397 tp->t_blk_res += blkres_delta;
398 delta -= blkres_delta;
400 tp->t_fdblocks_delta += delta;
401 if (xfs_has_lazysbcount(mp))
402 flags &= ~XFS_TRANS_SB_DIRTY;
404 case XFS_TRANS_SB_RES_FDBLOCKS:
406 * The allocation has already been applied to the
407 * in-core superblock's counter. This should only
408 * be applied to the on-disk superblock.
410 tp->t_res_fdblocks_delta += delta;
411 if (xfs_has_lazysbcount(mp))
412 flags &= ~XFS_TRANS_SB_DIRTY;
414 case XFS_TRANS_SB_FREXTENTS:
416 * Track the number of blocks allocated in the
417 * transaction. Make sure it does not exceed the
421 tp->t_rtx_res_used += (uint)-delta;
422 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
424 tp->t_frextents_delta += delta;
426 case XFS_TRANS_SB_RES_FREXTENTS:
428 * The allocation has already been applied to the
429 * in-core superblock's counter. This should only
430 * be applied to the on-disk superblock.
433 tp->t_res_frextents_delta += delta;
435 case XFS_TRANS_SB_DBLOCKS:
436 tp->t_dblocks_delta += delta;
438 case XFS_TRANS_SB_AGCOUNT:
440 tp->t_agcount_delta += delta;
442 case XFS_TRANS_SB_IMAXPCT:
443 tp->t_imaxpct_delta += delta;
445 case XFS_TRANS_SB_REXTSIZE:
446 tp->t_rextsize_delta += delta;
448 case XFS_TRANS_SB_RBMBLOCKS:
449 tp->t_rbmblocks_delta += delta;
451 case XFS_TRANS_SB_RBLOCKS:
452 tp->t_rblocks_delta += delta;
454 case XFS_TRANS_SB_REXTENTS:
455 tp->t_rextents_delta += delta;
457 case XFS_TRANS_SB_REXTSLOG:
458 tp->t_rextslog_delta += delta;
465 tp->t_flags |= flags;
469 * xfs_trans_apply_sb_deltas() is called from the commit code
470 * to bring the superblock buffer into the current transaction
471 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
473 * For now we just look at each field allowed to change and change
477 xfs_trans_apply_sb_deltas(
484 bp = xfs_trans_getsb(tp);
488 * Only update the superblock counters if we are logging them
490 if (!xfs_has_lazysbcount((tp->t_mountp))) {
491 if (tp->t_icount_delta)
492 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
493 if (tp->t_ifree_delta)
494 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
495 if (tp->t_fdblocks_delta)
496 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
497 if (tp->t_res_fdblocks_delta)
498 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
501 if (tp->t_frextents_delta)
502 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
503 if (tp->t_res_frextents_delta)
504 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
506 if (tp->t_dblocks_delta) {
507 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
510 if (tp->t_agcount_delta) {
511 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
514 if (tp->t_imaxpct_delta) {
515 sbp->sb_imax_pct += tp->t_imaxpct_delta;
518 if (tp->t_rextsize_delta) {
519 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
522 if (tp->t_rbmblocks_delta) {
523 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
526 if (tp->t_rblocks_delta) {
527 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
530 if (tp->t_rextents_delta) {
531 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
534 if (tp->t_rextslog_delta) {
535 sbp->sb_rextslog += tp->t_rextslog_delta;
539 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
542 * Log the whole thing, the fields are noncontiguous.
544 xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
547 * Since all the modifiable fields are contiguous, we
548 * can get away with this.
550 xfs_trans_log_buf(tp, bp, offsetof(struct xfs_dsb, sb_icount),
551 offsetof(struct xfs_dsb, sb_frextents) +
552 sizeof(sbp->sb_frextents) - 1);
556 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations and
557 * apply superblock counter changes to the in-core superblock. The
558 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
559 * applied to the in-core superblock. The idea is that that has already been
562 * If we are not logging superblock counters, then the inode allocated/free and
563 * used block counts are not updated in the on disk superblock. In this case,
564 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
565 * still need to update the incore superblock with the changes.
567 * Deltas for the inode count are +/-64, hence we use a large batch size of 128
568 * so we don't need to take the counter lock on every update.
570 #define XFS_ICOUNT_BATCH 128
573 xfs_trans_unreserve_and_mod_sb(
574 struct xfs_trans *tp)
576 struct xfs_mount *mp = tp->t_mountp;
577 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
578 int64_t blkdelta = 0;
579 int64_t rtxdelta = 0;
581 int64_t ifreedelta = 0;
584 /* calculate deltas */
585 if (tp->t_blk_res > 0)
586 blkdelta = tp->t_blk_res;
587 if ((tp->t_fdblocks_delta != 0) &&
588 (xfs_has_lazysbcount(mp) ||
589 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
590 blkdelta += tp->t_fdblocks_delta;
592 if (tp->t_rtx_res > 0)
593 rtxdelta = tp->t_rtx_res;
594 if ((tp->t_frextents_delta != 0) &&
595 (tp->t_flags & XFS_TRANS_SB_DIRTY))
596 rtxdelta += tp->t_frextents_delta;
598 if (xfs_has_lazysbcount(mp) ||
599 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
600 idelta = tp->t_icount_delta;
601 ifreedelta = tp->t_ifree_delta;
604 /* apply the per-cpu counters */
606 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
611 percpu_counter_add_batch(&mp->m_icount, idelta,
615 percpu_counter_add(&mp->m_ifree, ifreedelta);
617 if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
620 /* apply remaining deltas */
621 spin_lock(&mp->m_sb_lock);
622 mp->m_sb.sb_fdblocks += tp->t_fdblocks_delta + tp->t_res_fdblocks_delta;
623 mp->m_sb.sb_icount += idelta;
624 mp->m_sb.sb_ifree += ifreedelta;
625 mp->m_sb.sb_frextents += rtxdelta;
626 mp->m_sb.sb_dblocks += tp->t_dblocks_delta;
627 mp->m_sb.sb_agcount += tp->t_agcount_delta;
628 mp->m_sb.sb_imax_pct += tp->t_imaxpct_delta;
629 mp->m_sb.sb_rextsize += tp->t_rextsize_delta;
630 mp->m_sb.sb_rbmblocks += tp->t_rbmblocks_delta;
631 mp->m_sb.sb_rblocks += tp->t_rblocks_delta;
632 mp->m_sb.sb_rextents += tp->t_rextents_delta;
633 mp->m_sb.sb_rextslog += tp->t_rextslog_delta;
634 spin_unlock(&mp->m_sb_lock);
637 * Debug checks outside of the spinlock so they don't lock up the
638 * machine if they fail.
640 ASSERT(mp->m_sb.sb_imax_pct >= 0);
641 ASSERT(mp->m_sb.sb_rextslog >= 0);
645 /* Add the given log item to the transaction's list of log items. */
648 struct xfs_trans *tp,
649 struct xfs_log_item *lip)
651 ASSERT(lip->li_log == tp->t_mountp->m_log);
652 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
653 ASSERT(list_empty(&lip->li_trans));
654 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->li_flags));
656 list_add_tail(&lip->li_trans, &tp->t_items);
657 trace_xfs_trans_add_item(tp, _RET_IP_);
661 * Unlink the log item from the transaction. the log item is no longer
662 * considered dirty in this transaction, as the linked transaction has
663 * finished, either by abort or commit completion.
667 struct xfs_log_item *lip)
669 clear_bit(XFS_LI_DIRTY, &lip->li_flags);
670 list_del_init(&lip->li_trans);
673 /* Detach and unlock all of the items in a transaction */
675 xfs_trans_free_items(
676 struct xfs_trans *tp,
679 struct xfs_log_item *lip, *next;
681 trace_xfs_trans_free_items(tp, _RET_IP_);
683 list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
684 xfs_trans_del_item(lip);
686 set_bit(XFS_LI_ABORTED, &lip->li_flags);
687 if (lip->li_ops->iop_release)
688 lip->li_ops->iop_release(lip);
693 xfs_log_item_batch_insert(
694 struct xfs_ail *ailp,
695 struct xfs_ail_cursor *cur,
696 struct xfs_log_item **log_items,
698 xfs_lsn_t commit_lsn)
702 spin_lock(&ailp->ail_lock);
703 /* xfs_trans_ail_update_bulk drops ailp->ail_lock */
704 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
706 for (i = 0; i < nr_items; i++) {
707 struct xfs_log_item *lip = log_items[i];
709 if (lip->li_ops->iop_unpin)
710 lip->li_ops->iop_unpin(lip, 0);
715 * Bulk operation version of xfs_trans_committed that takes a log vector of
716 * items to insert into the AIL. This uses bulk AIL insertion techniques to
717 * minimise lock traffic.
719 * If we are called with the aborted flag set, it is because a log write during
720 * a CIL checkpoint commit has failed. In this case, all the items in the
721 * checkpoint have already gone through iop_committed and iop_committing, which
722 * means that checkpoint commit abort handling is treated exactly the same
723 * as an iclog write error even though we haven't started any IO yet. Hence in
724 * this case all we need to do is iop_committed processing, followed by an
725 * iop_unpin(aborted) call.
727 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
728 * at the end of the AIL, the insert cursor avoids the need to walk
729 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
730 * call. This saves a lot of needless list walking and is a net win, even
731 * though it slightly increases that amount of AIL lock traffic to set it up
735 xfs_trans_committed_bulk(
736 struct xfs_ail *ailp,
737 struct xfs_log_vec *log_vector,
738 xfs_lsn_t commit_lsn,
741 #define LOG_ITEM_BATCH_SIZE 32
742 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
743 struct xfs_log_vec *lv;
744 struct xfs_ail_cursor cur;
747 spin_lock(&ailp->ail_lock);
748 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
749 spin_unlock(&ailp->ail_lock);
751 /* unpin all the log items */
752 for (lv = log_vector; lv; lv = lv->lv_next ) {
753 struct xfs_log_item *lip = lv->lv_item;
757 set_bit(XFS_LI_ABORTED, &lip->li_flags);
759 if (lip->li_ops->flags & XFS_ITEM_RELEASE_WHEN_COMMITTED) {
760 lip->li_ops->iop_release(lip);
764 if (lip->li_ops->iop_committed)
765 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
767 item_lsn = commit_lsn;
769 /* item_lsn of -1 means the item needs no further processing */
770 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
774 * if we are aborting the operation, no point in inserting the
775 * object into the AIL as we are in a shutdown situation.
778 ASSERT(xlog_is_shutdown(ailp->ail_log));
779 if (lip->li_ops->iop_unpin)
780 lip->li_ops->iop_unpin(lip, 1);
784 if (item_lsn != commit_lsn) {
787 * Not a bulk update option due to unusual item_lsn.
788 * Push into AIL immediately, rechecking the lsn once
789 * we have the ail lock. Then unpin the item. This does
790 * not affect the AIL cursor the bulk insert path is
793 spin_lock(&ailp->ail_lock);
794 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
795 xfs_trans_ail_update(ailp, lip, item_lsn);
797 spin_unlock(&ailp->ail_lock);
798 if (lip->li_ops->iop_unpin)
799 lip->li_ops->iop_unpin(lip, 0);
803 /* Item is a candidate for bulk AIL insert. */
804 log_items[i++] = lv->lv_item;
805 if (i >= LOG_ITEM_BATCH_SIZE) {
806 xfs_log_item_batch_insert(ailp, &cur, log_items,
807 LOG_ITEM_BATCH_SIZE, commit_lsn);
812 /* make sure we insert the remainder! */
814 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
816 spin_lock(&ailp->ail_lock);
817 xfs_trans_ail_cursor_done(&cur);
818 spin_unlock(&ailp->ail_lock);
822 * Commit the given transaction to the log.
824 * XFS disk error handling mechanism is not based on a typical
825 * transaction abort mechanism. Logically after the filesystem
826 * gets marked 'SHUTDOWN', we can't let any new transactions
827 * be durable - ie. committed to disk - because some metadata might
828 * be inconsistent. In such cases, this returns an error, and the
829 * caller may assume that all locked objects joined to the transaction
830 * have already been unlocked as if the commit had succeeded.
831 * Do not reference the transaction structure after this call.
835 struct xfs_trans *tp,
838 struct xfs_mount *mp = tp->t_mountp;
839 struct xlog *log = mp->m_log;
840 xfs_csn_t commit_seq = 0;
842 int sync = tp->t_flags & XFS_TRANS_SYNC;
844 trace_xfs_trans_commit(tp, _RET_IP_);
847 * Finish deferred items on final commit. Only permanent transactions
848 * should ever have deferred ops.
850 WARN_ON_ONCE(!list_empty(&tp->t_dfops) &&
851 !(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
852 if (!regrant && (tp->t_flags & XFS_TRANS_PERM_LOG_RES)) {
853 error = xfs_defer_finish_noroll(&tp);
859 * If there is nothing to be logged by the transaction,
860 * then unlock all of the items associated with the
861 * transaction and free the transaction structure.
862 * Also make sure to return any reserved blocks to
865 if (!(tp->t_flags & XFS_TRANS_DIRTY))
869 * We must check against log shutdown here because we cannot abort log
870 * items and leave them dirty, inconsistent and unpinned in memory while
871 * the log is active. This leaves them open to being written back to
872 * disk, and that will lead to on-disk corruption.
874 if (xlog_is_shutdown(log)) {
879 ASSERT(tp->t_ticket != NULL);
882 * If we need to update the superblock, then do it now.
884 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
885 xfs_trans_apply_sb_deltas(tp);
886 xfs_trans_apply_dquot_deltas(tp);
888 xlog_cil_commit(log, tp, &commit_seq, regrant);
893 * If the transaction needs to be synchronous, then force the
894 * log out now and wait for it.
897 error = xfs_log_force_seq(mp, commit_seq, XFS_LOG_SYNC, NULL);
898 XFS_STATS_INC(mp, xs_trans_sync);
900 XFS_STATS_INC(mp, xs_trans_async);
906 xfs_trans_unreserve_and_mod_sb(tp);
909 * It is indeed possible for the transaction to be not dirty but
910 * the dqinfo portion to be. All that means is that we have some
911 * (non-persistent) quota reservations that need to be unreserved.
913 xfs_trans_unreserve_and_mod_dquots(tp);
915 if (regrant && !xlog_is_shutdown(log))
916 xfs_log_ticket_regrant(log, tp->t_ticket);
918 xfs_log_ticket_ungrant(log, tp->t_ticket);
921 xfs_trans_free_items(tp, !!error);
924 XFS_STATS_INC(mp, xs_trans_empty);
930 struct xfs_trans *tp)
932 return __xfs_trans_commit(tp, false);
936 * Unlock all of the transaction's items and free the transaction. If the
937 * transaction is dirty, we must shut down the filesystem because there is no
938 * way to restore them to their previous state.
940 * If the transaction has made a log reservation, make sure to release it as
943 * This is a high level function (equivalent to xfs_trans_commit()) and so can
944 * be called after the transaction has effectively been aborted due to the mount
945 * being shut down. However, if the mount has not been shut down and the
946 * transaction is dirty we will shut the mount down and, in doing so, that
947 * guarantees that the log is shut down, too. Hence we don't need to be as
948 * careful with shutdown state and dirty items here as we need to be in
949 * xfs_trans_commit().
953 struct xfs_trans *tp)
955 struct xfs_mount *mp = tp->t_mountp;
956 struct xlog *log = mp->m_log;
957 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
959 trace_xfs_trans_cancel(tp, _RET_IP_);
962 * It's never valid to cancel a transaction with deferred ops attached,
963 * because the transaction is effectively dirty. Complain about this
964 * loudly before freeing the in-memory defer items.
966 if (!list_empty(&tp->t_dfops)) {
967 ASSERT(xfs_is_shutdown(mp) || list_empty(&tp->t_dfops));
968 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
970 xfs_defer_cancel(tp);
974 * See if the caller is relying on us to shut down the filesystem. We
975 * only want an error report if there isn't already a shutdown in
976 * progress, so we only need to check against the mount shutdown state
979 if (dirty && !xfs_is_shutdown(mp)) {
980 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
981 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
984 /* Log items need to be consistent until the log is shut down. */
985 if (!dirty && !xlog_is_shutdown(log)) {
986 struct xfs_log_item *lip;
988 list_for_each_entry(lip, &tp->t_items, li_trans)
989 ASSERT(!xlog_item_is_intent_done(lip));
992 xfs_trans_unreserve_and_mod_sb(tp);
993 xfs_trans_unreserve_and_mod_dquots(tp);
996 xfs_log_ticket_ungrant(log, tp->t_ticket);
1000 xfs_trans_free_items(tp, dirty);
1005 * Roll from one trans in the sequence of PERMANENT transactions to
1006 * the next: permanent transactions are only flushed out when
1007 * committed with xfs_trans_commit(), but we still want as soon
1008 * as possible to let chunks of it go to the log. So we commit the
1009 * chunk we've been working on and get a new transaction to continue.
1013 struct xfs_trans **tpp)
1015 struct xfs_trans *trans = *tpp;
1016 struct xfs_trans_res tres;
1019 trace_xfs_trans_roll(trans, _RET_IP_);
1022 * Copy the critical parameters from one trans to the next.
1024 tres.tr_logres = trans->t_log_res;
1025 tres.tr_logcount = trans->t_log_count;
1027 *tpp = xfs_trans_dup(trans);
1030 * Commit the current transaction.
1031 * If this commit failed, then it'd just unlock those items that
1032 * are not marked ihold. That also means that a filesystem shutdown
1033 * is in progress. The caller takes the responsibility to cancel
1034 * the duplicate transaction that gets returned.
1036 error = __xfs_trans_commit(trans, true);
1041 * Reserve space in the log for the next transaction.
1042 * This also pushes items in the "AIL", the list of logged items,
1043 * out to disk if they are taking up space at the tail of the log
1044 * that we want to use. This requires that either nothing be locked
1045 * across this call, or that anything that is locked be logged in
1046 * the prior and the next transactions.
1048 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1049 return xfs_trans_reserve(*tpp, &tres, 0, 0);
1053 * Allocate an transaction, lock and join the inode to it, and reserve quota.
1055 * The caller must ensure that the on-disk dquots attached to this inode have
1056 * already been allocated and initialized. The caller is responsible for
1057 * releasing ILOCK_EXCL if a new transaction is returned.
1060 xfs_trans_alloc_inode(
1061 struct xfs_inode *ip,
1062 struct xfs_trans_res *resv,
1063 unsigned int dblocks,
1064 unsigned int rblocks,
1066 struct xfs_trans **tpp)
1068 struct xfs_trans *tp;
1069 struct xfs_mount *mp = ip->i_mount;
1070 bool retried = false;
1074 error = xfs_trans_alloc(mp, resv, dblocks,
1075 rblocks / mp->m_sb.sb_rextsize,
1076 force ? XFS_TRANS_RESERVE : 0, &tp);
1080 xfs_ilock(ip, XFS_ILOCK_EXCL);
1081 xfs_trans_ijoin(tp, ip, 0);
1083 error = xfs_qm_dqattach_locked(ip, false);
1085 /* Caller should have allocated the dquots! */
1086 ASSERT(error != -ENOENT);
1090 error = xfs_trans_reserve_quota_nblks(tp, ip, dblocks, rblocks, force);
1091 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1092 xfs_trans_cancel(tp);
1093 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1094 xfs_blockgc_free_quota(ip, 0);
1105 xfs_trans_cancel(tp);
1106 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1111 * Allocate an transaction in preparation for inode creation by reserving quota
1112 * against the given dquots. Callers are not required to hold any inode locks.
1115 xfs_trans_alloc_icreate(
1116 struct xfs_mount *mp,
1117 struct xfs_trans_res *resv,
1118 struct xfs_dquot *udqp,
1119 struct xfs_dquot *gdqp,
1120 struct xfs_dquot *pdqp,
1121 unsigned int dblocks,
1122 struct xfs_trans **tpp)
1124 struct xfs_trans *tp;
1125 bool retried = false;
1129 error = xfs_trans_alloc(mp, resv, dblocks, 0, 0, &tp);
1133 error = xfs_trans_reserve_quota_icreate(tp, udqp, gdqp, pdqp, dblocks);
1134 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1135 xfs_trans_cancel(tp);
1136 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1141 xfs_trans_cancel(tp);
1150 * Allocate an transaction, lock and join the inode to it, and reserve quota
1151 * in preparation for inode attribute changes that include uid, gid, or prid
1154 * The caller must ensure that the on-disk dquots attached to this inode have
1155 * already been allocated and initialized. The ILOCK will be dropped when the
1156 * transaction is committed or cancelled.
1159 xfs_trans_alloc_ichange(
1160 struct xfs_inode *ip,
1161 struct xfs_dquot *new_udqp,
1162 struct xfs_dquot *new_gdqp,
1163 struct xfs_dquot *new_pdqp,
1165 struct xfs_trans **tpp)
1167 struct xfs_trans *tp;
1168 struct xfs_mount *mp = ip->i_mount;
1169 struct xfs_dquot *udqp;
1170 struct xfs_dquot *gdqp;
1171 struct xfs_dquot *pdqp;
1172 bool retried = false;
1176 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1180 xfs_ilock(ip, XFS_ILOCK_EXCL);
1181 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1183 error = xfs_qm_dqattach_locked(ip, false);
1185 /* Caller should have allocated the dquots! */
1186 ASSERT(error != -ENOENT);
1191 * For each quota type, skip quota reservations if the inode's dquots
1192 * now match the ones that came from the caller, or the caller didn't
1193 * pass one in. The inode's dquots can change if we drop the ILOCK to
1194 * perform a blockgc scan, so we must preserve the caller's arguments.
1196 udqp = (new_udqp != ip->i_udquot) ? new_udqp : NULL;
1197 gdqp = (new_gdqp != ip->i_gdquot) ? new_gdqp : NULL;
1198 pdqp = (new_pdqp != ip->i_pdquot) ? new_pdqp : NULL;
1199 if (udqp || gdqp || pdqp) {
1200 unsigned int qflags = XFS_QMOPT_RES_REGBLKS;
1203 qflags |= XFS_QMOPT_FORCE_RES;
1206 * Reserve enough quota to handle blocks on disk and reserved
1207 * for a delayed allocation. We'll actually transfer the
1208 * delalloc reservation between dquots at chown time, even
1209 * though that part is only semi-transactional.
1211 error = xfs_trans_reserve_quota_bydquots(tp, mp, udqp, gdqp,
1212 pdqp, ip->i_nblocks + ip->i_delayed_blks,
1214 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1215 xfs_trans_cancel(tp);
1216 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1228 xfs_trans_cancel(tp);
1233 * Allocate an transaction, lock and join the directory and child inodes to it,
1234 * and reserve quota for a directory update. If there isn't sufficient space,
1235 * @dblocks will be set to zero for a reservationless directory update and
1236 * @nospace_error will be set to a negative errno describing the space
1237 * constraint we hit.
1239 * The caller must ensure that the on-disk dquots attached to this inode have
1240 * already been allocated and initialized. The ILOCKs will be dropped when the
1241 * transaction is committed or cancelled.
1244 xfs_trans_alloc_dir(
1245 struct xfs_inode *dp,
1246 struct xfs_trans_res *resv,
1247 struct xfs_inode *ip,
1248 unsigned int *dblocks,
1249 struct xfs_trans **tpp,
1252 struct xfs_trans *tp;
1253 struct xfs_mount *mp = ip->i_mount;
1254 unsigned int resblks;
1255 bool retried = false;
1261 error = xfs_trans_alloc(mp, resv, resblks, 0, 0, &tp);
1262 if (error == -ENOSPC) {
1263 *nospace_error = error;
1265 error = xfs_trans_alloc(mp, resv, resblks, 0, 0, &tp);
1270 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
1272 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1273 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1275 error = xfs_qm_dqattach_locked(dp, false);
1277 /* Caller should have allocated the dquots! */
1278 ASSERT(error != -ENOENT);
1282 error = xfs_qm_dqattach_locked(ip, false);
1284 /* Caller should have allocated the dquots! */
1285 ASSERT(error != -ENOENT);
1292 error = xfs_trans_reserve_quota_nblks(tp, dp, resblks, 0, false);
1293 if (error == -EDQUOT || error == -ENOSPC) {
1295 xfs_trans_cancel(tp);
1296 xfs_blockgc_free_quota(dp, 0);
1301 *nospace_error = error;
1314 xfs_trans_cancel(tp);