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"
23 #include "xfs_inode.h"
24 #include "xfs_dquot_item.h"
25 #include "xfs_dquot.h"
26 #include "xfs_icache.h"
28 kmem_zone_t *xfs_trans_zone;
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 if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
76 sb_end_intwrite(tp->t_mountp->m_super);
77 xfs_trans_free_dqinfo(tp);
78 kmem_cache_free(xfs_trans_zone, tp);
82 * This is called to create a new transaction which will share the
83 * permanent log reservation of the given transaction. The remaining
84 * unused block and rt extent reservations are also inherited. This
85 * implies that the original transaction is no longer allowed to allocate
86 * blocks. Locks and log items, however, are no inherited. They must
87 * be added to the new transaction explicitly.
89 STATIC struct xfs_trans *
93 struct xfs_trans *ntp;
95 trace_xfs_trans_dup(tp, _RET_IP_);
97 ntp = kmem_cache_zalloc(xfs_trans_zone, GFP_KERNEL | __GFP_NOFAIL);
100 * Initialize the new transaction structure.
102 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
103 ntp->t_mountp = tp->t_mountp;
104 INIT_LIST_HEAD(&ntp->t_items);
105 INIT_LIST_HEAD(&ntp->t_busy);
106 INIT_LIST_HEAD(&ntp->t_dfops);
107 ntp->t_firstblock = NULLFSBLOCK;
109 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
110 ASSERT(tp->t_ticket != NULL);
112 ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
113 (tp->t_flags & XFS_TRANS_RESERVE) |
114 (tp->t_flags & XFS_TRANS_NO_WRITECOUNT) |
115 (tp->t_flags & XFS_TRANS_RES_FDBLKS);
116 /* We gave our writer reference to the new transaction */
117 tp->t_flags |= XFS_TRANS_NO_WRITECOUNT;
118 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
120 ASSERT(tp->t_blk_res >= tp->t_blk_res_used);
121 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
122 tp->t_blk_res = tp->t_blk_res_used;
124 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
125 tp->t_rtx_res = tp->t_rtx_res_used;
126 ntp->t_pflags = tp->t_pflags;
128 /* move deferred ops over to the new tp */
129 xfs_defer_move(ntp, tp);
131 xfs_trans_dup_dqinfo(tp, ntp);
136 * This is called to reserve free disk blocks and log space for the
137 * given transaction. This must be done before allocating any resources
138 * within the transaction.
140 * This will return ENOSPC if there are not enough blocks available.
141 * It will sleep waiting for available log space.
142 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
143 * is used by long running transactions. If any one of the reservations
144 * fails then they will all be backed out.
146 * This does not do quota reservations. That typically is done by the
151 struct xfs_trans *tp,
152 struct xfs_trans_res *resp,
156 struct xfs_mount *mp = tp->t_mountp;
158 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
160 /* Mark this thread as being in a transaction */
161 current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
164 * Attempt to reserve the needed disk blocks by decrementing
165 * the number needed from the number available. This will
166 * fail if the count would go below zero.
169 error = xfs_mod_fdblocks(mp, -((int64_t)blocks), rsvd);
171 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
174 tp->t_blk_res += blocks;
178 * Reserve the log space needed for this transaction.
180 if (resp->tr_logres > 0) {
181 bool permanent = false;
183 ASSERT(tp->t_log_res == 0 ||
184 tp->t_log_res == resp->tr_logres);
185 ASSERT(tp->t_log_count == 0 ||
186 tp->t_log_count == resp->tr_logcount);
188 if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
189 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
192 ASSERT(tp->t_ticket == NULL);
193 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
196 if (tp->t_ticket != NULL) {
197 ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
198 error = xfs_log_regrant(mp, tp->t_ticket);
200 error = xfs_log_reserve(mp,
203 &tp->t_ticket, XFS_TRANSACTION,
210 tp->t_log_res = resp->tr_logres;
211 tp->t_log_count = resp->tr_logcount;
215 * Attempt to reserve the needed realtime extents by decrementing
216 * the number needed from the number available. This will
217 * fail if the count would go below zero.
220 error = xfs_mod_frextents(mp, -((int64_t)rtextents));
225 tp->t_rtx_res += rtextents;
231 * Error cases jump to one of these labels to undo any
232 * reservations which have already been performed.
235 if (resp->tr_logres > 0) {
236 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
239 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
244 xfs_mod_fdblocks(mp, (int64_t)blocks, rsvd);
248 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
255 struct xfs_mount *mp,
256 struct xfs_trans_res *resp,
260 struct xfs_trans **tpp)
262 struct xfs_trans *tp;
266 * Allocate the handle before we do our freeze accounting and setting up
267 * GFP_NOFS allocation context so that we avoid lockdep false positives
268 * by doing GFP_KERNEL allocations inside sb_start_intwrite().
270 tp = kmem_cache_zalloc(xfs_trans_zone, GFP_KERNEL | __GFP_NOFAIL);
271 if (!(flags & XFS_TRANS_NO_WRITECOUNT))
272 sb_start_intwrite(mp->m_super);
275 * Zero-reservation ("empty") transactions can't modify anything, so
276 * they're allowed to run while we're frozen.
278 WARN_ON(resp->tr_logres > 0 &&
279 mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
280 ASSERT(!(flags & XFS_TRANS_RES_FDBLKS) ||
281 xfs_sb_version_haslazysbcount(&mp->m_sb));
283 tp->t_magic = XFS_TRANS_HEADER_MAGIC;
286 INIT_LIST_HEAD(&tp->t_items);
287 INIT_LIST_HEAD(&tp->t_busy);
288 INIT_LIST_HEAD(&tp->t_dfops);
289 tp->t_firstblock = NULLFSBLOCK;
291 error = xfs_trans_reserve(tp, resp, blocks, rtextents);
293 xfs_trans_cancel(tp);
297 trace_xfs_trans_alloc(tp, _RET_IP_);
304 * Create an empty transaction with no reservation. This is a defensive
305 * mechanism for routines that query metadata without actually modifying them --
306 * if the metadata being queried is somehow cross-linked (think a btree block
307 * pointer that points higher in the tree), we risk deadlock. However, blocks
308 * grabbed as part of a transaction can be re-grabbed. The verifiers will
309 * notice the corrupt block and the operation will fail back to userspace
310 * without deadlocking.
312 * Note the zero-length reservation; this transaction MUST be cancelled without
315 * Callers should obtain freeze protection to avoid a conflict with fs freezing
316 * where we can be grabbing buffers at the same time that freeze is trying to
317 * drain the buffer LRU list.
320 xfs_trans_alloc_empty(
321 struct xfs_mount *mp,
322 struct xfs_trans **tpp)
324 struct xfs_trans_res resv = {0};
326 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
330 * Record the indicated change to the given field for application
331 * to the file system's superblock when the transaction commits.
332 * For now, just store the change in the transaction structure.
334 * Mark the transaction structure to indicate that the superblock
335 * needs to be updated before committing.
337 * Because we may not be keeping track of allocated/free inodes and
338 * used filesystem blocks in the superblock, we do not mark the
339 * superblock dirty in this transaction if we modify these fields.
340 * We still need to update the transaction deltas so that they get
341 * applied to the incore superblock, but we don't want them to
342 * cause the superblock to get locked and logged if these are the
343 * only fields in the superblock that the transaction modifies.
351 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
352 xfs_mount_t *mp = tp->t_mountp;
355 case XFS_TRANS_SB_ICOUNT:
356 tp->t_icount_delta += delta;
357 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
358 flags &= ~XFS_TRANS_SB_DIRTY;
360 case XFS_TRANS_SB_IFREE:
361 tp->t_ifree_delta += delta;
362 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
363 flags &= ~XFS_TRANS_SB_DIRTY;
365 case XFS_TRANS_SB_FDBLOCKS:
367 * Track the number of blocks allocated in the transaction.
368 * Make sure it does not exceed the number reserved. If so,
369 * shutdown as this can lead to accounting inconsistency.
372 tp->t_blk_res_used += (uint)-delta;
373 if (tp->t_blk_res_used > tp->t_blk_res)
374 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
375 } else if (delta > 0 && (tp->t_flags & XFS_TRANS_RES_FDBLKS)) {
376 int64_t blkres_delta;
379 * Return freed blocks directly to the reservation
380 * instead of the global pool, being careful not to
381 * overflow the trans counter. This is used to preserve
382 * reservation across chains of transaction rolls that
383 * repeatedly free and allocate blocks.
385 blkres_delta = min_t(int64_t, delta,
386 UINT_MAX - tp->t_blk_res);
387 tp->t_blk_res += blkres_delta;
388 delta -= blkres_delta;
390 tp->t_fdblocks_delta += delta;
391 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
392 flags &= ~XFS_TRANS_SB_DIRTY;
394 case XFS_TRANS_SB_RES_FDBLOCKS:
396 * The allocation has already been applied to the
397 * in-core superblock's counter. This should only
398 * be applied to the on-disk superblock.
400 tp->t_res_fdblocks_delta += delta;
401 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
402 flags &= ~XFS_TRANS_SB_DIRTY;
404 case XFS_TRANS_SB_FREXTENTS:
406 * Track the number of blocks allocated in the
407 * transaction. Make sure it does not exceed the
411 tp->t_rtx_res_used += (uint)-delta;
412 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
414 tp->t_frextents_delta += delta;
416 case XFS_TRANS_SB_RES_FREXTENTS:
418 * The allocation has already been applied to the
419 * in-core superblock's counter. This should only
420 * be applied to the on-disk superblock.
423 tp->t_res_frextents_delta += delta;
425 case XFS_TRANS_SB_DBLOCKS:
427 tp->t_dblocks_delta += delta;
429 case XFS_TRANS_SB_AGCOUNT:
431 tp->t_agcount_delta += delta;
433 case XFS_TRANS_SB_IMAXPCT:
434 tp->t_imaxpct_delta += delta;
436 case XFS_TRANS_SB_REXTSIZE:
437 tp->t_rextsize_delta += delta;
439 case XFS_TRANS_SB_RBMBLOCKS:
440 tp->t_rbmblocks_delta += delta;
442 case XFS_TRANS_SB_RBLOCKS:
443 tp->t_rblocks_delta += delta;
445 case XFS_TRANS_SB_REXTENTS:
446 tp->t_rextents_delta += delta;
448 case XFS_TRANS_SB_REXTSLOG:
449 tp->t_rextslog_delta += delta;
456 tp->t_flags |= flags;
460 * xfs_trans_apply_sb_deltas() is called from the commit code
461 * to bring the superblock buffer into the current transaction
462 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
464 * For now we just look at each field allowed to change and change
468 xfs_trans_apply_sb_deltas(
475 bp = xfs_trans_getsb(tp);
479 * Check that superblock mods match the mods made to AGF counters.
481 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
482 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
483 tp->t_ag_btree_delta));
486 * Only update the superblock counters if we are logging them
488 if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
489 if (tp->t_icount_delta)
490 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
491 if (tp->t_ifree_delta)
492 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
493 if (tp->t_fdblocks_delta)
494 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
495 if (tp->t_res_fdblocks_delta)
496 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
499 if (tp->t_frextents_delta)
500 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
501 if (tp->t_res_frextents_delta)
502 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
504 if (tp->t_dblocks_delta) {
505 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
508 if (tp->t_agcount_delta) {
509 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
512 if (tp->t_imaxpct_delta) {
513 sbp->sb_imax_pct += tp->t_imaxpct_delta;
516 if (tp->t_rextsize_delta) {
517 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
520 if (tp->t_rbmblocks_delta) {
521 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
524 if (tp->t_rblocks_delta) {
525 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
528 if (tp->t_rextents_delta) {
529 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
532 if (tp->t_rextslog_delta) {
533 sbp->sb_rextslog += tp->t_rextslog_delta;
537 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
540 * Log the whole thing, the fields are noncontiguous.
542 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
545 * Since all the modifiable fields are contiguous, we
546 * can get away with this.
548 xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
549 offsetof(xfs_dsb_t, sb_frextents) +
550 sizeof(sbp->sb_frextents) - 1);
554 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations and
555 * apply superblock counter changes to the in-core superblock. The
556 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
557 * applied to the in-core superblock. The idea is that that has already been
560 * If we are not logging superblock counters, then the inode allocated/free and
561 * used block counts are not updated in the on disk superblock. In this case,
562 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
563 * still need to update the incore superblock with the changes.
565 * Deltas for the inode count are +/-64, hence we use a large batch size of 128
566 * so we don't need to take the counter lock on every update.
568 #define XFS_ICOUNT_BATCH 128
571 xfs_trans_unreserve_and_mod_sb(
572 struct xfs_trans *tp)
574 struct xfs_mount *mp = tp->t_mountp;
575 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
576 int64_t blkdelta = 0;
577 int64_t rtxdelta = 0;
579 int64_t ifreedelta = 0;
582 /* calculate deltas */
583 if (tp->t_blk_res > 0)
584 blkdelta = tp->t_blk_res;
585 if ((tp->t_fdblocks_delta != 0) &&
586 (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
587 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
588 blkdelta += tp->t_fdblocks_delta;
590 if (tp->t_rtx_res > 0)
591 rtxdelta = tp->t_rtx_res;
592 if ((tp->t_frextents_delta != 0) &&
593 (tp->t_flags & XFS_TRANS_SB_DIRTY))
594 rtxdelta += tp->t_frextents_delta;
596 if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
597 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
598 idelta = tp->t_icount_delta;
599 ifreedelta = tp->t_ifree_delta;
602 /* apply the per-cpu counters */
604 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
609 percpu_counter_add_batch(&mp->m_icount, idelta,
612 ASSERT(__percpu_counter_compare(&mp->m_icount, 0,
613 XFS_ICOUNT_BATCH) >= 0);
617 percpu_counter_add(&mp->m_ifree, ifreedelta);
619 ASSERT(percpu_counter_compare(&mp->m_ifree, 0) >= 0);
622 if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
625 /* apply remaining deltas */
626 spin_lock(&mp->m_sb_lock);
627 mp->m_sb.sb_frextents += rtxdelta;
628 mp->m_sb.sb_dblocks += tp->t_dblocks_delta;
629 mp->m_sb.sb_agcount += tp->t_agcount_delta;
630 mp->m_sb.sb_imax_pct += tp->t_imaxpct_delta;
631 mp->m_sb.sb_rextsize += tp->t_rextsize_delta;
632 mp->m_sb.sb_rbmblocks += tp->t_rbmblocks_delta;
633 mp->m_sb.sb_rblocks += tp->t_rblocks_delta;
634 mp->m_sb.sb_rextents += tp->t_rextents_delta;
635 mp->m_sb.sb_rextslog += tp->t_rextslog_delta;
636 spin_unlock(&mp->m_sb_lock);
639 * Debug checks outside of the spinlock so they don't lock up the
640 * machine if they fail.
642 ASSERT(mp->m_sb.sb_imax_pct >= 0);
643 ASSERT(mp->m_sb.sb_rextslog >= 0);
647 /* Add the given log item to the transaction's list of log items. */
650 struct xfs_trans *tp,
651 struct xfs_log_item *lip)
653 ASSERT(lip->li_mountp == tp->t_mountp);
654 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
655 ASSERT(list_empty(&lip->li_trans));
656 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->li_flags));
658 list_add_tail(&lip->li_trans, &tp->t_items);
659 trace_xfs_trans_add_item(tp, _RET_IP_);
663 * Unlink the log item from the transaction. the log item is no longer
664 * considered dirty in this transaction, as the linked transaction has
665 * finished, either by abort or commit completion.
669 struct xfs_log_item *lip)
671 clear_bit(XFS_LI_DIRTY, &lip->li_flags);
672 list_del_init(&lip->li_trans);
675 /* Detach and unlock all of the items in a transaction */
677 xfs_trans_free_items(
678 struct xfs_trans *tp,
681 struct xfs_log_item *lip, *next;
683 trace_xfs_trans_free_items(tp, _RET_IP_);
685 list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
686 xfs_trans_del_item(lip);
688 set_bit(XFS_LI_ABORTED, &lip->li_flags);
689 if (lip->li_ops->iop_release)
690 lip->li_ops->iop_release(lip);
695 xfs_log_item_batch_insert(
696 struct xfs_ail *ailp,
697 struct xfs_ail_cursor *cur,
698 struct xfs_log_item **log_items,
700 xfs_lsn_t commit_lsn)
704 spin_lock(&ailp->ail_lock);
705 /* xfs_trans_ail_update_bulk drops ailp->ail_lock */
706 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
708 for (i = 0; i < nr_items; i++) {
709 struct xfs_log_item *lip = log_items[i];
711 if (lip->li_ops->iop_unpin)
712 lip->li_ops->iop_unpin(lip, 0);
717 * Bulk operation version of xfs_trans_committed that takes a log vector of
718 * items to insert into the AIL. This uses bulk AIL insertion techniques to
719 * minimise lock traffic.
721 * If we are called with the aborted flag set, it is because a log write during
722 * a CIL checkpoint commit has failed. In this case, all the items in the
723 * checkpoint have already gone through iop_committed and iop_committing, which
724 * means that checkpoint commit abort handling is treated exactly the same
725 * as an iclog write error even though we haven't started any IO yet. Hence in
726 * this case all we need to do is iop_committed processing, followed by an
727 * iop_unpin(aborted) call.
729 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
730 * at the end of the AIL, the insert cursor avoids the need to walk
731 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
732 * call. This saves a lot of needless list walking and is a net win, even
733 * though it slightly increases that amount of AIL lock traffic to set it up
737 xfs_trans_committed_bulk(
738 struct xfs_ail *ailp,
739 struct xfs_log_vec *log_vector,
740 xfs_lsn_t commit_lsn,
743 #define LOG_ITEM_BATCH_SIZE 32
744 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
745 struct xfs_log_vec *lv;
746 struct xfs_ail_cursor cur;
749 spin_lock(&ailp->ail_lock);
750 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
751 spin_unlock(&ailp->ail_lock);
753 /* unpin all the log items */
754 for (lv = log_vector; lv; lv = lv->lv_next ) {
755 struct xfs_log_item *lip = lv->lv_item;
759 set_bit(XFS_LI_ABORTED, &lip->li_flags);
761 if (lip->li_ops->flags & XFS_ITEM_RELEASE_WHEN_COMMITTED) {
762 lip->li_ops->iop_release(lip);
766 if (lip->li_ops->iop_committed)
767 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
769 item_lsn = commit_lsn;
771 /* item_lsn of -1 means the item needs no further processing */
772 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
776 * if we are aborting the operation, no point in inserting the
777 * object into the AIL as we are in a shutdown situation.
780 ASSERT(XFS_FORCED_SHUTDOWN(ailp->ail_mount));
781 if (lip->li_ops->iop_unpin)
782 lip->li_ops->iop_unpin(lip, 1);
786 if (item_lsn != commit_lsn) {
789 * Not a bulk update option due to unusual item_lsn.
790 * Push into AIL immediately, rechecking the lsn once
791 * we have the ail lock. Then unpin the item. This does
792 * not affect the AIL cursor the bulk insert path is
795 spin_lock(&ailp->ail_lock);
796 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
797 xfs_trans_ail_update(ailp, lip, item_lsn);
799 spin_unlock(&ailp->ail_lock);
800 if (lip->li_ops->iop_unpin)
801 lip->li_ops->iop_unpin(lip, 0);
805 /* Item is a candidate for bulk AIL insert. */
806 log_items[i++] = lv->lv_item;
807 if (i >= LOG_ITEM_BATCH_SIZE) {
808 xfs_log_item_batch_insert(ailp, &cur, log_items,
809 LOG_ITEM_BATCH_SIZE, commit_lsn);
814 /* make sure we insert the remainder! */
816 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
818 spin_lock(&ailp->ail_lock);
819 xfs_trans_ail_cursor_done(&cur);
820 spin_unlock(&ailp->ail_lock);
824 * Commit the given transaction to the log.
826 * XFS disk error handling mechanism is not based on a typical
827 * transaction abort mechanism. Logically after the filesystem
828 * gets marked 'SHUTDOWN', we can't let any new transactions
829 * be durable - ie. committed to disk - because some metadata might
830 * be inconsistent. In such cases, this returns an error, and the
831 * caller may assume that all locked objects joined to the transaction
832 * have already been unlocked as if the commit had succeeded.
833 * Do not reference the transaction structure after this call.
837 struct xfs_trans *tp,
840 struct xfs_mount *mp = tp->t_mountp;
841 xfs_lsn_t commit_lsn = -1;
843 int sync = tp->t_flags & XFS_TRANS_SYNC;
845 trace_xfs_trans_commit(tp, _RET_IP_);
848 * Finish deferred items on final commit. Only permanent transactions
849 * should ever have deferred ops.
851 WARN_ON_ONCE(!list_empty(&tp->t_dfops) &&
852 !(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
853 if (!regrant && (tp->t_flags & XFS_TRANS_PERM_LOG_RES)) {
854 error = xfs_defer_finish_noroll(&tp);
860 * If there is nothing to be logged by the transaction,
861 * then unlock all of the items associated with the
862 * transaction and free the transaction structure.
863 * Also make sure to return any reserved blocks to
866 if (!(tp->t_flags & XFS_TRANS_DIRTY))
869 if (XFS_FORCED_SHUTDOWN(mp)) {
874 ASSERT(tp->t_ticket != NULL);
877 * If we need to update the superblock, then do it now.
879 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
880 xfs_trans_apply_sb_deltas(tp);
881 xfs_trans_apply_dquot_deltas(tp);
883 xfs_log_commit_cil(mp, tp, &commit_lsn, regrant);
885 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
889 * If the transaction needs to be synchronous, then force the
890 * log out now and wait for it.
893 error = xfs_log_force_lsn(mp, commit_lsn, XFS_LOG_SYNC, NULL);
894 XFS_STATS_INC(mp, xs_trans_sync);
896 XFS_STATS_INC(mp, xs_trans_async);
902 xfs_trans_unreserve_and_mod_sb(tp);
905 * It is indeed possible for the transaction to be not dirty but
906 * the dqinfo portion to be. All that means is that we have some
907 * (non-persistent) quota reservations that need to be unreserved.
909 xfs_trans_unreserve_and_mod_dquots(tp);
911 if (regrant && !XLOG_FORCED_SHUTDOWN(mp->m_log))
912 xfs_log_ticket_regrant(mp->m_log, tp->t_ticket);
914 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
917 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
918 xfs_trans_free_items(tp, !!error);
921 XFS_STATS_INC(mp, xs_trans_empty);
927 struct xfs_trans *tp)
929 return __xfs_trans_commit(tp, false);
933 * Unlock all of the transaction's items and free the transaction.
934 * The transaction must not have modified any of its items, because
935 * there is no way to restore them to their previous state.
937 * If the transaction has made a log reservation, make sure to release
942 struct xfs_trans *tp)
944 struct xfs_mount *mp = tp->t_mountp;
945 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
947 trace_xfs_trans_cancel(tp, _RET_IP_);
949 if (tp->t_flags & XFS_TRANS_PERM_LOG_RES)
950 xfs_defer_cancel(tp);
953 * See if the caller is relying on us to shut down the
954 * filesystem. This happens in paths where we detect
955 * corruption and decide to give up.
957 if (dirty && !XFS_FORCED_SHUTDOWN(mp)) {
958 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
959 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
962 if (!dirty && !XFS_FORCED_SHUTDOWN(mp)) {
963 struct xfs_log_item *lip;
965 list_for_each_entry(lip, &tp->t_items, li_trans)
966 ASSERT(!xlog_item_is_intent_done(lip));
969 xfs_trans_unreserve_and_mod_sb(tp);
970 xfs_trans_unreserve_and_mod_dquots(tp);
973 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
977 /* mark this thread as no longer being in a transaction */
978 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
980 xfs_trans_free_items(tp, dirty);
985 * Roll from one trans in the sequence of PERMANENT transactions to
986 * the next: permanent transactions are only flushed out when
987 * committed with xfs_trans_commit(), but we still want as soon
988 * as possible to let chunks of it go to the log. So we commit the
989 * chunk we've been working on and get a new transaction to continue.
993 struct xfs_trans **tpp)
995 struct xfs_trans *trans = *tpp;
996 struct xfs_trans_res tres;
999 trace_xfs_trans_roll(trans, _RET_IP_);
1002 * Copy the critical parameters from one trans to the next.
1004 tres.tr_logres = trans->t_log_res;
1005 tres.tr_logcount = trans->t_log_count;
1007 *tpp = xfs_trans_dup(trans);
1010 * Commit the current transaction.
1011 * If this commit failed, then it'd just unlock those items that
1012 * are not marked ihold. That also means that a filesystem shutdown
1013 * is in progress. The caller takes the responsibility to cancel
1014 * the duplicate transaction that gets returned.
1016 error = __xfs_trans_commit(trans, true);
1021 * Reserve space in the log for the next transaction.
1022 * This also pushes items in the "AIL", the list of logged items,
1023 * out to disk if they are taking up space at the tail of the log
1024 * that we want to use. This requires that either nothing be locked
1025 * across this call, or that anything that is locked be logged in
1026 * the prior and the next transactions.
1028 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1029 return xfs_trans_reserve(*tpp, &tres, 0, 0);
1033 * Allocate an transaction, lock and join the inode to it, and reserve quota.
1035 * The caller must ensure that the on-disk dquots attached to this inode have
1036 * already been allocated and initialized. The caller is responsible for
1037 * releasing ILOCK_EXCL if a new transaction is returned.
1040 xfs_trans_alloc_inode(
1041 struct xfs_inode *ip,
1042 struct xfs_trans_res *resv,
1043 unsigned int dblocks,
1044 unsigned int rblocks,
1046 struct xfs_trans **tpp)
1048 struct xfs_trans *tp;
1049 struct xfs_mount *mp = ip->i_mount;
1050 bool retried = false;
1054 error = xfs_trans_alloc(mp, resv, dblocks,
1055 rblocks / mp->m_sb.sb_rextsize,
1056 force ? XFS_TRANS_RESERVE : 0, &tp);
1060 xfs_ilock(ip, XFS_ILOCK_EXCL);
1061 xfs_trans_ijoin(tp, ip, 0);
1063 error = xfs_qm_dqattach_locked(ip, false);
1065 /* Caller should have allocated the dquots! */
1066 ASSERT(error != -ENOENT);
1070 error = xfs_trans_reserve_quota_nblks(tp, ip, dblocks, rblocks, force);
1071 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1072 xfs_trans_cancel(tp);
1073 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1074 xfs_blockgc_free_quota(ip, 0);
1085 xfs_trans_cancel(tp);
1086 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1091 * Allocate an transaction in preparation for inode creation by reserving quota
1092 * against the given dquots. Callers are not required to hold any inode locks.
1095 xfs_trans_alloc_icreate(
1096 struct xfs_mount *mp,
1097 struct xfs_trans_res *resv,
1098 struct xfs_dquot *udqp,
1099 struct xfs_dquot *gdqp,
1100 struct xfs_dquot *pdqp,
1101 unsigned int dblocks,
1102 struct xfs_trans **tpp)
1104 struct xfs_trans *tp;
1105 bool retried = false;
1109 error = xfs_trans_alloc(mp, resv, dblocks, 0, 0, &tp);
1113 error = xfs_trans_reserve_quota_icreate(tp, udqp, gdqp, pdqp, dblocks);
1114 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1115 xfs_trans_cancel(tp);
1116 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1121 xfs_trans_cancel(tp);
1130 * Allocate an transaction, lock and join the inode to it, and reserve quota
1131 * in preparation for inode attribute changes that include uid, gid, or prid
1134 * The caller must ensure that the on-disk dquots attached to this inode have
1135 * already been allocated and initialized. The ILOCK will be dropped when the
1136 * transaction is committed or cancelled.
1139 xfs_trans_alloc_ichange(
1140 struct xfs_inode *ip,
1141 struct xfs_dquot *new_udqp,
1142 struct xfs_dquot *new_gdqp,
1143 struct xfs_dquot *new_pdqp,
1145 struct xfs_trans **tpp)
1147 struct xfs_trans *tp;
1148 struct xfs_mount *mp = ip->i_mount;
1149 struct xfs_dquot *udqp;
1150 struct xfs_dquot *gdqp;
1151 struct xfs_dquot *pdqp;
1152 bool retried = false;
1156 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1160 xfs_ilock(ip, XFS_ILOCK_EXCL);
1161 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1163 error = xfs_qm_dqattach_locked(ip, false);
1165 /* Caller should have allocated the dquots! */
1166 ASSERT(error != -ENOENT);
1171 * For each quota type, skip quota reservations if the inode's dquots
1172 * now match the ones that came from the caller, or the caller didn't
1173 * pass one in. The inode's dquots can change if we drop the ILOCK to
1174 * perform a blockgc scan, so we must preserve the caller's arguments.
1176 udqp = (new_udqp != ip->i_udquot) ? new_udqp : NULL;
1177 gdqp = (new_gdqp != ip->i_gdquot) ? new_gdqp : NULL;
1178 pdqp = (new_pdqp != ip->i_pdquot) ? new_pdqp : NULL;
1179 if (udqp || gdqp || pdqp) {
1180 unsigned int qflags = XFS_QMOPT_RES_REGBLKS;
1183 qflags |= XFS_QMOPT_FORCE_RES;
1186 * Reserve enough quota to handle blocks on disk and reserved
1187 * for a delayed allocation. We'll actually transfer the
1188 * delalloc reservation between dquots at chown time, even
1189 * though that part is only semi-transactional.
1191 error = xfs_trans_reserve_quota_bydquots(tp, mp, udqp, gdqp,
1192 pdqp, ip->i_d.di_nblocks + ip->i_delayed_blks,
1194 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1195 xfs_trans_cancel(tp);
1196 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1208 xfs_trans_cancel(tp);