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 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_zone, 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_zone, 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_zone, 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_sb_version_haslazysbcount(&mp->m_sb));
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 error = xfs_blockgc_free_space(mp, NULL);
306 xfs_trans_cancel(tp);
310 trace_xfs_trans_alloc(tp, _RET_IP_);
317 * Create an empty transaction with no reservation. This is a defensive
318 * mechanism for routines that query metadata without actually modifying them --
319 * if the metadata being queried is somehow cross-linked (think a btree block
320 * pointer that points higher in the tree), we risk deadlock. However, blocks
321 * grabbed as part of a transaction can be re-grabbed. The verifiers will
322 * notice the corrupt block and the operation will fail back to userspace
323 * without deadlocking.
325 * Note the zero-length reservation; this transaction MUST be cancelled without
328 * Callers should obtain freeze protection to avoid a conflict with fs freezing
329 * where we can be grabbing buffers at the same time that freeze is trying to
330 * drain the buffer LRU list.
333 xfs_trans_alloc_empty(
334 struct xfs_mount *mp,
335 struct xfs_trans **tpp)
337 struct xfs_trans_res resv = {0};
339 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
343 * Record the indicated change to the given field for application
344 * to the file system's superblock when the transaction commits.
345 * For now, just store the change in the transaction structure.
347 * Mark the transaction structure to indicate that the superblock
348 * needs to be updated before committing.
350 * Because we may not be keeping track of allocated/free inodes and
351 * used filesystem blocks in the superblock, we do not mark the
352 * superblock dirty in this transaction if we modify these fields.
353 * We still need to update the transaction deltas so that they get
354 * applied to the incore superblock, but we don't want them to
355 * cause the superblock to get locked and logged if these are the
356 * only fields in the superblock that the transaction modifies.
364 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
365 xfs_mount_t *mp = tp->t_mountp;
368 case XFS_TRANS_SB_ICOUNT:
369 tp->t_icount_delta += delta;
370 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
371 flags &= ~XFS_TRANS_SB_DIRTY;
373 case XFS_TRANS_SB_IFREE:
374 tp->t_ifree_delta += delta;
375 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
376 flags &= ~XFS_TRANS_SB_DIRTY;
378 case XFS_TRANS_SB_FDBLOCKS:
380 * Track the number of blocks allocated in the transaction.
381 * Make sure it does not exceed the number reserved. If so,
382 * shutdown as this can lead to accounting inconsistency.
385 tp->t_blk_res_used += (uint)-delta;
386 if (tp->t_blk_res_used > tp->t_blk_res)
387 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
388 } else if (delta > 0 && (tp->t_flags & XFS_TRANS_RES_FDBLKS)) {
389 int64_t blkres_delta;
392 * Return freed blocks directly to the reservation
393 * instead of the global pool, being careful not to
394 * overflow the trans counter. This is used to preserve
395 * reservation across chains of transaction rolls that
396 * repeatedly free and allocate blocks.
398 blkres_delta = min_t(int64_t, delta,
399 UINT_MAX - tp->t_blk_res);
400 tp->t_blk_res += blkres_delta;
401 delta -= blkres_delta;
403 tp->t_fdblocks_delta += delta;
404 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
405 flags &= ~XFS_TRANS_SB_DIRTY;
407 case XFS_TRANS_SB_RES_FDBLOCKS:
409 * The allocation has already been applied to the
410 * in-core superblock's counter. This should only
411 * be applied to the on-disk superblock.
413 tp->t_res_fdblocks_delta += delta;
414 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
415 flags &= ~XFS_TRANS_SB_DIRTY;
417 case XFS_TRANS_SB_FREXTENTS:
419 * Track the number of blocks allocated in the
420 * transaction. Make sure it does not exceed the
424 tp->t_rtx_res_used += (uint)-delta;
425 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
427 tp->t_frextents_delta += delta;
429 case XFS_TRANS_SB_RES_FREXTENTS:
431 * The allocation has already been applied to the
432 * in-core superblock's counter. This should only
433 * be applied to the on-disk superblock.
436 tp->t_res_frextents_delta += delta;
438 case XFS_TRANS_SB_DBLOCKS:
439 tp->t_dblocks_delta += delta;
441 case XFS_TRANS_SB_AGCOUNT:
443 tp->t_agcount_delta += delta;
445 case XFS_TRANS_SB_IMAXPCT:
446 tp->t_imaxpct_delta += delta;
448 case XFS_TRANS_SB_REXTSIZE:
449 tp->t_rextsize_delta += delta;
451 case XFS_TRANS_SB_RBMBLOCKS:
452 tp->t_rbmblocks_delta += delta;
454 case XFS_TRANS_SB_RBLOCKS:
455 tp->t_rblocks_delta += delta;
457 case XFS_TRANS_SB_REXTENTS:
458 tp->t_rextents_delta += delta;
460 case XFS_TRANS_SB_REXTSLOG:
461 tp->t_rextslog_delta += delta;
468 tp->t_flags |= flags;
472 * xfs_trans_apply_sb_deltas() is called from the commit code
473 * to bring the superblock buffer into the current transaction
474 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
476 * For now we just look at each field allowed to change and change
480 xfs_trans_apply_sb_deltas(
487 bp = xfs_trans_getsb(tp);
491 * Only update the superblock counters if we are logging them
493 if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
494 if (tp->t_icount_delta)
495 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
496 if (tp->t_ifree_delta)
497 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
498 if (tp->t_fdblocks_delta)
499 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
500 if (tp->t_res_fdblocks_delta)
501 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
504 if (tp->t_frextents_delta)
505 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
506 if (tp->t_res_frextents_delta)
507 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
509 if (tp->t_dblocks_delta) {
510 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
513 if (tp->t_agcount_delta) {
514 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
517 if (tp->t_imaxpct_delta) {
518 sbp->sb_imax_pct += tp->t_imaxpct_delta;
521 if (tp->t_rextsize_delta) {
522 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
525 if (tp->t_rbmblocks_delta) {
526 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
529 if (tp->t_rblocks_delta) {
530 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
533 if (tp->t_rextents_delta) {
534 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
537 if (tp->t_rextslog_delta) {
538 sbp->sb_rextslog += tp->t_rextslog_delta;
542 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
545 * Log the whole thing, the fields are noncontiguous.
547 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
550 * Since all the modifiable fields are contiguous, we
551 * can get away with this.
553 xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
554 offsetof(xfs_dsb_t, sb_frextents) +
555 sizeof(sbp->sb_frextents) - 1);
559 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations and
560 * apply superblock counter changes to the in-core superblock. The
561 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
562 * applied to the in-core superblock. The idea is that that has already been
565 * If we are not logging superblock counters, then the inode allocated/free and
566 * used block counts are not updated in the on disk superblock. In this case,
567 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
568 * still need to update the incore superblock with the changes.
570 * Deltas for the inode count are +/-64, hence we use a large batch size of 128
571 * so we don't need to take the counter lock on every update.
573 #define XFS_ICOUNT_BATCH 128
576 xfs_trans_unreserve_and_mod_sb(
577 struct xfs_trans *tp)
579 struct xfs_mount *mp = tp->t_mountp;
580 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
581 int64_t blkdelta = 0;
582 int64_t rtxdelta = 0;
584 int64_t ifreedelta = 0;
587 /* calculate deltas */
588 if (tp->t_blk_res > 0)
589 blkdelta = tp->t_blk_res;
590 if ((tp->t_fdblocks_delta != 0) &&
591 (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
592 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
593 blkdelta += tp->t_fdblocks_delta;
595 if (tp->t_rtx_res > 0)
596 rtxdelta = tp->t_rtx_res;
597 if ((tp->t_frextents_delta != 0) &&
598 (tp->t_flags & XFS_TRANS_SB_DIRTY))
599 rtxdelta += tp->t_frextents_delta;
601 if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
602 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
603 idelta = tp->t_icount_delta;
604 ifreedelta = tp->t_ifree_delta;
607 /* apply the per-cpu counters */
609 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
614 percpu_counter_add_batch(&mp->m_icount, idelta,
618 percpu_counter_add(&mp->m_ifree, ifreedelta);
620 if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
623 /* apply remaining deltas */
624 spin_lock(&mp->m_sb_lock);
625 mp->m_sb.sb_fdblocks += tp->t_fdblocks_delta + tp->t_res_fdblocks_delta;
626 mp->m_sb.sb_icount += idelta;
627 mp->m_sb.sb_ifree += ifreedelta;
628 mp->m_sb.sb_frextents += rtxdelta;
629 mp->m_sb.sb_dblocks += tp->t_dblocks_delta;
630 mp->m_sb.sb_agcount += tp->t_agcount_delta;
631 mp->m_sb.sb_imax_pct += tp->t_imaxpct_delta;
632 mp->m_sb.sb_rextsize += tp->t_rextsize_delta;
633 mp->m_sb.sb_rbmblocks += tp->t_rbmblocks_delta;
634 mp->m_sb.sb_rblocks += tp->t_rblocks_delta;
635 mp->m_sb.sb_rextents += tp->t_rextents_delta;
636 mp->m_sb.sb_rextslog += tp->t_rextslog_delta;
637 spin_unlock(&mp->m_sb_lock);
640 * Debug checks outside of the spinlock so they don't lock up the
641 * machine if they fail.
643 ASSERT(mp->m_sb.sb_imax_pct >= 0);
644 ASSERT(mp->m_sb.sb_rextslog >= 0);
648 /* Add the given log item to the transaction's list of log items. */
651 struct xfs_trans *tp,
652 struct xfs_log_item *lip)
654 ASSERT(lip->li_mountp == tp->t_mountp);
655 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
656 ASSERT(list_empty(&lip->li_trans));
657 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->li_flags));
659 list_add_tail(&lip->li_trans, &tp->t_items);
660 trace_xfs_trans_add_item(tp, _RET_IP_);
664 * Unlink the log item from the transaction. the log item is no longer
665 * considered dirty in this transaction, as the linked transaction has
666 * finished, either by abort or commit completion.
670 struct xfs_log_item *lip)
672 clear_bit(XFS_LI_DIRTY, &lip->li_flags);
673 list_del_init(&lip->li_trans);
676 /* Detach and unlock all of the items in a transaction */
678 xfs_trans_free_items(
679 struct xfs_trans *tp,
682 struct xfs_log_item *lip, *next;
684 trace_xfs_trans_free_items(tp, _RET_IP_);
686 list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
687 xfs_trans_del_item(lip);
689 set_bit(XFS_LI_ABORTED, &lip->li_flags);
690 if (lip->li_ops->iop_release)
691 lip->li_ops->iop_release(lip);
696 xfs_log_item_batch_insert(
697 struct xfs_ail *ailp,
698 struct xfs_ail_cursor *cur,
699 struct xfs_log_item **log_items,
701 xfs_lsn_t commit_lsn)
705 spin_lock(&ailp->ail_lock);
706 /* xfs_trans_ail_update_bulk drops ailp->ail_lock */
707 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
709 for (i = 0; i < nr_items; i++) {
710 struct xfs_log_item *lip = log_items[i];
712 if (lip->li_ops->iop_unpin)
713 lip->li_ops->iop_unpin(lip, 0);
718 * Bulk operation version of xfs_trans_committed that takes a log vector of
719 * items to insert into the AIL. This uses bulk AIL insertion techniques to
720 * minimise lock traffic.
722 * If we are called with the aborted flag set, it is because a log write during
723 * a CIL checkpoint commit has failed. In this case, all the items in the
724 * checkpoint have already gone through iop_committed and iop_committing, which
725 * means that checkpoint commit abort handling is treated exactly the same
726 * as an iclog write error even though we haven't started any IO yet. Hence in
727 * this case all we need to do is iop_committed processing, followed by an
728 * iop_unpin(aborted) call.
730 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
731 * at the end of the AIL, the insert cursor avoids the need to walk
732 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
733 * call. This saves a lot of needless list walking and is a net win, even
734 * though it slightly increases that amount of AIL lock traffic to set it up
738 xfs_trans_committed_bulk(
739 struct xfs_ail *ailp,
740 struct xfs_log_vec *log_vector,
741 xfs_lsn_t commit_lsn,
744 #define LOG_ITEM_BATCH_SIZE 32
745 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
746 struct xfs_log_vec *lv;
747 struct xfs_ail_cursor cur;
750 spin_lock(&ailp->ail_lock);
751 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
752 spin_unlock(&ailp->ail_lock);
754 /* unpin all the log items */
755 for (lv = log_vector; lv; lv = lv->lv_next ) {
756 struct xfs_log_item *lip = lv->lv_item;
760 set_bit(XFS_LI_ABORTED, &lip->li_flags);
762 if (lip->li_ops->flags & XFS_ITEM_RELEASE_WHEN_COMMITTED) {
763 lip->li_ops->iop_release(lip);
767 if (lip->li_ops->iop_committed)
768 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
770 item_lsn = commit_lsn;
772 /* item_lsn of -1 means the item needs no further processing */
773 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
777 * if we are aborting the operation, no point in inserting the
778 * object into the AIL as we are in a shutdown situation.
781 ASSERT(XFS_FORCED_SHUTDOWN(ailp->ail_mount));
782 if (lip->li_ops->iop_unpin)
783 lip->li_ops->iop_unpin(lip, 1);
787 if (item_lsn != commit_lsn) {
790 * Not a bulk update option due to unusual item_lsn.
791 * Push into AIL immediately, rechecking the lsn once
792 * we have the ail lock. Then unpin the item. This does
793 * not affect the AIL cursor the bulk insert path is
796 spin_lock(&ailp->ail_lock);
797 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
798 xfs_trans_ail_update(ailp, lip, item_lsn);
800 spin_unlock(&ailp->ail_lock);
801 if (lip->li_ops->iop_unpin)
802 lip->li_ops->iop_unpin(lip, 0);
806 /* Item is a candidate for bulk AIL insert. */
807 log_items[i++] = lv->lv_item;
808 if (i >= LOG_ITEM_BATCH_SIZE) {
809 xfs_log_item_batch_insert(ailp, &cur, log_items,
810 LOG_ITEM_BATCH_SIZE, commit_lsn);
815 /* make sure we insert the remainder! */
817 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
819 spin_lock(&ailp->ail_lock);
820 xfs_trans_ail_cursor_done(&cur);
821 spin_unlock(&ailp->ail_lock);
825 * Commit the given transaction to the log.
827 * XFS disk error handling mechanism is not based on a typical
828 * transaction abort mechanism. Logically after the filesystem
829 * gets marked 'SHUTDOWN', we can't let any new transactions
830 * be durable - ie. committed to disk - because some metadata might
831 * be inconsistent. In such cases, this returns an error, and the
832 * caller may assume that all locked objects joined to the transaction
833 * have already been unlocked as if the commit had succeeded.
834 * Do not reference the transaction structure after this call.
838 struct xfs_trans *tp,
841 struct xfs_mount *mp = tp->t_mountp;
842 xfs_csn_t commit_seq = 0;
844 int sync = tp->t_flags & XFS_TRANS_SYNC;
846 trace_xfs_trans_commit(tp, _RET_IP_);
849 * Finish deferred items on final commit. Only permanent transactions
850 * should ever have deferred ops.
852 WARN_ON_ONCE(!list_empty(&tp->t_dfops) &&
853 !(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
854 if (!regrant && (tp->t_flags & XFS_TRANS_PERM_LOG_RES)) {
855 error = xfs_defer_finish_noroll(&tp);
861 * If there is nothing to be logged by the transaction,
862 * then unlock all of the items associated with the
863 * transaction and free the transaction structure.
864 * Also make sure to return any reserved blocks to
867 if (!(tp->t_flags & XFS_TRANS_DIRTY))
870 if (XFS_FORCED_SHUTDOWN(mp)) {
875 ASSERT(tp->t_ticket != NULL);
878 * If we need to update the superblock, then do it now.
880 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
881 xfs_trans_apply_sb_deltas(tp);
882 xfs_trans_apply_dquot_deltas(tp);
884 xlog_cil_commit(mp->m_log, tp, &commit_seq, regrant);
889 * If the transaction needs to be synchronous, then force the
890 * log out now and wait for it.
893 error = xfs_log_force_seq(mp, commit_seq, 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 xfs_trans_free_items(tp, !!error);
920 XFS_STATS_INC(mp, xs_trans_empty);
926 struct xfs_trans *tp)
928 return __xfs_trans_commit(tp, false);
932 * Unlock all of the transaction's items and free the transaction.
933 * The transaction must not have modified any of its items, because
934 * there is no way to restore them to their previous state.
936 * If the transaction has made a log reservation, make sure to release
941 struct xfs_trans *tp)
943 struct xfs_mount *mp = tp->t_mountp;
944 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
946 trace_xfs_trans_cancel(tp, _RET_IP_);
948 if (tp->t_flags & XFS_TRANS_PERM_LOG_RES)
949 xfs_defer_cancel(tp);
952 * See if the caller is relying on us to shut down the
953 * filesystem. This happens in paths where we detect
954 * corruption and decide to give up.
956 if (dirty && !XFS_FORCED_SHUTDOWN(mp)) {
957 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
958 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
961 if (!dirty && !XFS_FORCED_SHUTDOWN(mp)) {
962 struct xfs_log_item *lip;
964 list_for_each_entry(lip, &tp->t_items, li_trans)
965 ASSERT(!xlog_item_is_intent_done(lip));
968 xfs_trans_unreserve_and_mod_sb(tp);
969 xfs_trans_unreserve_and_mod_dquots(tp);
972 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
976 xfs_trans_free_items(tp, dirty);
981 * Roll from one trans in the sequence of PERMANENT transactions to
982 * the next: permanent transactions are only flushed out when
983 * committed with xfs_trans_commit(), but we still want as soon
984 * as possible to let chunks of it go to the log. So we commit the
985 * chunk we've been working on and get a new transaction to continue.
989 struct xfs_trans **tpp)
991 struct xfs_trans *trans = *tpp;
992 struct xfs_trans_res tres;
995 trace_xfs_trans_roll(trans, _RET_IP_);
998 * Copy the critical parameters from one trans to the next.
1000 tres.tr_logres = trans->t_log_res;
1001 tres.tr_logcount = trans->t_log_count;
1003 *tpp = xfs_trans_dup(trans);
1006 * Commit the current transaction.
1007 * If this commit failed, then it'd just unlock those items that
1008 * are not marked ihold. That also means that a filesystem shutdown
1009 * is in progress. The caller takes the responsibility to cancel
1010 * the duplicate transaction that gets returned.
1012 error = __xfs_trans_commit(trans, true);
1017 * Reserve space in the log for the next transaction.
1018 * This also pushes items in the "AIL", the list of logged items,
1019 * out to disk if they are taking up space at the tail of the log
1020 * that we want to use. This requires that either nothing be locked
1021 * across this call, or that anything that is locked be logged in
1022 * the prior and the next transactions.
1024 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1025 return xfs_trans_reserve(*tpp, &tres, 0, 0);
1029 * Allocate an transaction, lock and join the inode to it, and reserve quota.
1031 * The caller must ensure that the on-disk dquots attached to this inode have
1032 * already been allocated and initialized. The caller is responsible for
1033 * releasing ILOCK_EXCL if a new transaction is returned.
1036 xfs_trans_alloc_inode(
1037 struct xfs_inode *ip,
1038 struct xfs_trans_res *resv,
1039 unsigned int dblocks,
1040 unsigned int rblocks,
1042 struct xfs_trans **tpp)
1044 struct xfs_trans *tp;
1045 struct xfs_mount *mp = ip->i_mount;
1046 bool retried = false;
1050 error = xfs_trans_alloc(mp, resv, dblocks,
1051 rblocks / mp->m_sb.sb_rextsize,
1052 force ? XFS_TRANS_RESERVE : 0, &tp);
1056 xfs_ilock(ip, XFS_ILOCK_EXCL);
1057 xfs_trans_ijoin(tp, ip, 0);
1059 error = xfs_qm_dqattach_locked(ip, false);
1061 /* Caller should have allocated the dquots! */
1062 ASSERT(error != -ENOENT);
1066 error = xfs_trans_reserve_quota_nblks(tp, ip, dblocks, rblocks, force);
1067 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1068 xfs_trans_cancel(tp);
1069 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1070 xfs_blockgc_free_quota(ip, 0);
1081 xfs_trans_cancel(tp);
1082 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1087 * Allocate an transaction in preparation for inode creation by reserving quota
1088 * against the given dquots. Callers are not required to hold any inode locks.
1091 xfs_trans_alloc_icreate(
1092 struct xfs_mount *mp,
1093 struct xfs_trans_res *resv,
1094 struct xfs_dquot *udqp,
1095 struct xfs_dquot *gdqp,
1096 struct xfs_dquot *pdqp,
1097 unsigned int dblocks,
1098 struct xfs_trans **tpp)
1100 struct xfs_trans *tp;
1101 bool retried = false;
1105 error = xfs_trans_alloc(mp, resv, dblocks, 0, 0, &tp);
1109 error = xfs_trans_reserve_quota_icreate(tp, udqp, gdqp, pdqp, dblocks);
1110 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1111 xfs_trans_cancel(tp);
1112 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1117 xfs_trans_cancel(tp);
1126 * Allocate an transaction, lock and join the inode to it, and reserve quota
1127 * in preparation for inode attribute changes that include uid, gid, or prid
1130 * The caller must ensure that the on-disk dquots attached to this inode have
1131 * already been allocated and initialized. The ILOCK will be dropped when the
1132 * transaction is committed or cancelled.
1135 xfs_trans_alloc_ichange(
1136 struct xfs_inode *ip,
1137 struct xfs_dquot *new_udqp,
1138 struct xfs_dquot *new_gdqp,
1139 struct xfs_dquot *new_pdqp,
1141 struct xfs_trans **tpp)
1143 struct xfs_trans *tp;
1144 struct xfs_mount *mp = ip->i_mount;
1145 struct xfs_dquot *udqp;
1146 struct xfs_dquot *gdqp;
1147 struct xfs_dquot *pdqp;
1148 bool retried = false;
1152 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1156 xfs_ilock(ip, XFS_ILOCK_EXCL);
1157 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1159 error = xfs_qm_dqattach_locked(ip, false);
1161 /* Caller should have allocated the dquots! */
1162 ASSERT(error != -ENOENT);
1167 * For each quota type, skip quota reservations if the inode's dquots
1168 * now match the ones that came from the caller, or the caller didn't
1169 * pass one in. The inode's dquots can change if we drop the ILOCK to
1170 * perform a blockgc scan, so we must preserve the caller's arguments.
1172 udqp = (new_udqp != ip->i_udquot) ? new_udqp : NULL;
1173 gdqp = (new_gdqp != ip->i_gdquot) ? new_gdqp : NULL;
1174 pdqp = (new_pdqp != ip->i_pdquot) ? new_pdqp : NULL;
1175 if (udqp || gdqp || pdqp) {
1176 unsigned int qflags = XFS_QMOPT_RES_REGBLKS;
1179 qflags |= XFS_QMOPT_FORCE_RES;
1182 * Reserve enough quota to handle blocks on disk and reserved
1183 * for a delayed allocation. We'll actually transfer the
1184 * delalloc reservation between dquots at chown time, even
1185 * though that part is only semi-transactional.
1187 error = xfs_trans_reserve_quota_bydquots(tp, mp, udqp, gdqp,
1188 pdqp, ip->i_nblocks + ip->i_delayed_blks,
1190 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1191 xfs_trans_cancel(tp);
1192 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1204 xfs_trans_cancel(tp);