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"
27 #include "xfs_rtbitmap.h"
29 struct kmem_cache *xfs_trans_cache;
31 #if defined(CONFIG_TRACEPOINTS)
33 xfs_trans_trace_reservations(
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);
46 # define xfs_trans_trace_reservations(mp)
50 * Initialize the precomputed transaction reservation values
51 * in the mount structure.
57 xfs_trans_resv_calc(mp, M_RES(mp));
58 xfs_trans_trace_reservations(mp);
62 * Free the transaction structure. If there is more clean up
63 * to do when the structure is freed, add it here.
69 xfs_extent_busy_sort(&tp->t_busy);
70 xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
72 trace_xfs_trans_free(tp, _RET_IP_);
73 xfs_trans_clear_context(tp);
74 if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
75 sb_end_intwrite(tp->t_mountp->m_super);
76 xfs_trans_free_dqinfo(tp);
77 kmem_cache_free(xfs_trans_cache, tp);
81 * This is called to create a new transaction which will share the
82 * permanent log reservation of the given transaction. The remaining
83 * unused block and rt extent reservations are also inherited. This
84 * implies that the original transaction is no longer allowed to allocate
85 * blocks. Locks and log items, however, are no inherited. They must
86 * be added to the new transaction explicitly.
88 STATIC struct xfs_trans *
92 struct xfs_trans *ntp;
94 trace_xfs_trans_dup(tp, _RET_IP_);
96 ntp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
99 * Initialize the new transaction structure.
101 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
102 ntp->t_mountp = tp->t_mountp;
103 INIT_LIST_HEAD(&ntp->t_items);
104 INIT_LIST_HEAD(&ntp->t_busy);
105 INIT_LIST_HEAD(&ntp->t_dfops);
106 ntp->t_highest_agno = NULLAGNUMBER;
108 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
109 ASSERT(tp->t_ticket != NULL);
111 ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
112 (tp->t_flags & XFS_TRANS_RESERVE) |
113 (tp->t_flags & XFS_TRANS_NO_WRITECOUNT) |
114 (tp->t_flags & XFS_TRANS_RES_FDBLKS);
115 /* We gave our writer reference to the new transaction */
116 tp->t_flags |= XFS_TRANS_NO_WRITECOUNT;
117 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
119 ASSERT(tp->t_blk_res >= tp->t_blk_res_used);
120 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
121 tp->t_blk_res = tp->t_blk_res_used;
123 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
124 tp->t_rtx_res = tp->t_rtx_res_used;
126 xfs_trans_switch_context(tp, ntp);
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;
161 * Attempt to reserve the needed disk blocks by decrementing
162 * the number needed from the number available. This will
163 * fail if the count would go below zero.
166 error = xfs_mod_fdblocks(mp, -((int64_t)blocks), rsvd);
169 tp->t_blk_res += blocks;
173 * Reserve the log space needed for this transaction.
175 if (resp->tr_logres > 0) {
176 bool permanent = false;
178 ASSERT(tp->t_log_res == 0 ||
179 tp->t_log_res == resp->tr_logres);
180 ASSERT(tp->t_log_count == 0 ||
181 tp->t_log_count == resp->tr_logcount);
183 if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
184 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
187 ASSERT(tp->t_ticket == NULL);
188 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
191 if (tp->t_ticket != NULL) {
192 ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
193 error = xfs_log_regrant(mp, tp->t_ticket);
195 error = xfs_log_reserve(mp, resp->tr_logres,
197 &tp->t_ticket, permanent);
203 tp->t_log_res = resp->tr_logres;
204 tp->t_log_count = resp->tr_logcount;
208 * Attempt to reserve the needed realtime extents by decrementing
209 * the number needed from the number available. This will
210 * fail if the count would go below zero.
213 error = xfs_mod_frextents(mp, -((int64_t)rtextents));
218 tp->t_rtx_res += rtextents;
224 * Error cases jump to one of these labels to undo any
225 * reservations which have already been performed.
228 if (resp->tr_logres > 0) {
229 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
232 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
237 xfs_mod_fdblocks(mp, (int64_t)blocks, rsvd);
245 struct xfs_mount *mp,
246 struct xfs_trans_res *resp,
250 struct xfs_trans **tpp)
252 struct xfs_trans *tp;
253 bool want_retry = true;
257 * Allocate the handle before we do our freeze accounting and setting up
258 * GFP_NOFS allocation context so that we avoid lockdep false positives
259 * by doing GFP_KERNEL allocations inside sb_start_intwrite().
262 tp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
263 if (!(flags & XFS_TRANS_NO_WRITECOUNT))
264 sb_start_intwrite(mp->m_super);
265 xfs_trans_set_context(tp);
268 * Zero-reservation ("empty") transactions can't modify anything, so
269 * they're allowed to run while we're frozen.
271 WARN_ON(resp->tr_logres > 0 &&
272 mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
273 ASSERT(!(flags & XFS_TRANS_RES_FDBLKS) ||
274 xfs_has_lazysbcount(mp));
276 tp->t_magic = XFS_TRANS_HEADER_MAGIC;
279 INIT_LIST_HEAD(&tp->t_items);
280 INIT_LIST_HEAD(&tp->t_busy);
281 INIT_LIST_HEAD(&tp->t_dfops);
282 tp->t_highest_agno = NULLAGNUMBER;
284 error = xfs_trans_reserve(tp, resp, blocks, rtextents);
285 if (error == -ENOSPC && want_retry) {
286 xfs_trans_cancel(tp);
289 * We weren't able to reserve enough space for the transaction.
290 * Flush the other speculative space allocations to free space.
291 * Do not perform a synchronous scan because callers can hold
294 error = xfs_blockgc_flush_all(mp);
301 xfs_trans_cancel(tp);
305 trace_xfs_trans_alloc(tp, _RET_IP_);
312 * Create an empty transaction with no reservation. This is a defensive
313 * mechanism for routines that query metadata without actually modifying them --
314 * if the metadata being queried is somehow cross-linked (think a btree block
315 * pointer that points higher in the tree), we risk deadlock. However, blocks
316 * grabbed as part of a transaction can be re-grabbed. The verifiers will
317 * notice the corrupt block and the operation will fail back to userspace
318 * without deadlocking.
320 * Note the zero-length reservation; this transaction MUST be cancelled without
323 * Callers should obtain freeze protection to avoid a conflict with fs freezing
324 * where we can be grabbing buffers at the same time that freeze is trying to
325 * drain the buffer LRU list.
328 xfs_trans_alloc_empty(
329 struct xfs_mount *mp,
330 struct xfs_trans **tpp)
332 struct xfs_trans_res resv = {0};
334 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
338 * Record the indicated change to the given field for application
339 * to the file system's superblock when the transaction commits.
340 * For now, just store the change in the transaction structure.
342 * Mark the transaction structure to indicate that the superblock
343 * needs to be updated before committing.
345 * Because we may not be keeping track of allocated/free inodes and
346 * used filesystem blocks in the superblock, we do not mark the
347 * superblock dirty in this transaction if we modify these fields.
348 * We still need to update the transaction deltas so that they get
349 * applied to the incore superblock, but we don't want them to
350 * cause the superblock to get locked and logged if these are the
351 * only fields in the superblock that the transaction modifies.
359 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
360 xfs_mount_t *mp = tp->t_mountp;
363 case XFS_TRANS_SB_ICOUNT:
364 tp->t_icount_delta += delta;
365 if (xfs_has_lazysbcount(mp))
366 flags &= ~XFS_TRANS_SB_DIRTY;
368 case XFS_TRANS_SB_IFREE:
369 tp->t_ifree_delta += delta;
370 if (xfs_has_lazysbcount(mp))
371 flags &= ~XFS_TRANS_SB_DIRTY;
373 case XFS_TRANS_SB_FDBLOCKS:
375 * Track the number of blocks allocated in the transaction.
376 * Make sure it does not exceed the number reserved. If so,
377 * shutdown as this can lead to accounting inconsistency.
380 tp->t_blk_res_used += (uint)-delta;
381 if (tp->t_blk_res_used > tp->t_blk_res)
382 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
383 } else if (delta > 0 && (tp->t_flags & XFS_TRANS_RES_FDBLKS)) {
384 int64_t blkres_delta;
387 * Return freed blocks directly to the reservation
388 * instead of the global pool, being careful not to
389 * overflow the trans counter. This is used to preserve
390 * reservation across chains of transaction rolls that
391 * repeatedly free and allocate blocks.
393 blkres_delta = min_t(int64_t, delta,
394 UINT_MAX - tp->t_blk_res);
395 tp->t_blk_res += blkres_delta;
396 delta -= blkres_delta;
398 tp->t_fdblocks_delta += delta;
399 if (xfs_has_lazysbcount(mp))
400 flags &= ~XFS_TRANS_SB_DIRTY;
402 case XFS_TRANS_SB_RES_FDBLOCKS:
404 * The allocation has already been applied to the
405 * in-core superblock's counter. This should only
406 * be applied to the on-disk superblock.
408 tp->t_res_fdblocks_delta += delta;
409 if (xfs_has_lazysbcount(mp))
410 flags &= ~XFS_TRANS_SB_DIRTY;
412 case XFS_TRANS_SB_FREXTENTS:
414 * Track the number of blocks allocated in the
415 * transaction. Make sure it does not exceed the
419 tp->t_rtx_res_used += (uint)-delta;
420 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
422 tp->t_frextents_delta += delta;
424 case XFS_TRANS_SB_RES_FREXTENTS:
426 * The allocation has already been applied to the
427 * in-core superblock's counter. This should only
428 * be applied to the on-disk superblock.
431 tp->t_res_frextents_delta += delta;
433 case XFS_TRANS_SB_DBLOCKS:
434 tp->t_dblocks_delta += delta;
436 case XFS_TRANS_SB_AGCOUNT:
438 tp->t_agcount_delta += delta;
440 case XFS_TRANS_SB_IMAXPCT:
441 tp->t_imaxpct_delta += delta;
443 case XFS_TRANS_SB_REXTSIZE:
444 tp->t_rextsize_delta += delta;
446 case XFS_TRANS_SB_RBMBLOCKS:
447 tp->t_rbmblocks_delta += delta;
449 case XFS_TRANS_SB_RBLOCKS:
450 tp->t_rblocks_delta += delta;
452 case XFS_TRANS_SB_REXTENTS:
453 tp->t_rextents_delta += delta;
455 case XFS_TRANS_SB_REXTSLOG:
456 tp->t_rextslog_delta += delta;
463 tp->t_flags |= flags;
467 * xfs_trans_apply_sb_deltas() is called from the commit code
468 * to bring the superblock buffer into the current transaction
469 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
471 * For now we just look at each field allowed to change and change
475 xfs_trans_apply_sb_deltas(
482 bp = xfs_trans_getsb(tp);
486 * Only update the superblock counters if we are logging them
488 if (!xfs_has_lazysbcount((tp->t_mountp))) {
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);
500 * Updating frextents requires careful handling because it does not
501 * behave like the lazysb counters because we cannot rely on log
502 * recovery in older kenels to recompute the value from the rtbitmap.
503 * This means that the ondisk frextents must be consistent with the
506 * Therefore, log the frextents change to the ondisk superblock and
507 * update the incore superblock so that future calls to xfs_log_sb
508 * write the correct value ondisk.
510 * Don't touch m_frextents because it includes incore reservations,
511 * and those are handled by the unreserve function.
513 if (tp->t_frextents_delta || tp->t_res_frextents_delta) {
514 struct xfs_mount *mp = tp->t_mountp;
517 rtxdelta = tp->t_frextents_delta + tp->t_res_frextents_delta;
519 spin_lock(&mp->m_sb_lock);
520 be64_add_cpu(&sbp->sb_frextents, rtxdelta);
521 mp->m_sb.sb_frextents += rtxdelta;
522 spin_unlock(&mp->m_sb_lock);
525 if (tp->t_dblocks_delta) {
526 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
529 if (tp->t_agcount_delta) {
530 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
533 if (tp->t_imaxpct_delta) {
534 sbp->sb_imax_pct += tp->t_imaxpct_delta;
537 if (tp->t_rextsize_delta) {
538 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
541 if (tp->t_rbmblocks_delta) {
542 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
545 if (tp->t_rblocks_delta) {
546 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
549 if (tp->t_rextents_delta) {
550 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
553 if (tp->t_rextslog_delta) {
554 sbp->sb_rextslog += tp->t_rextslog_delta;
558 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
561 * Log the whole thing, the fields are noncontiguous.
563 xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
566 * Since all the modifiable fields are contiguous, we
567 * can get away with this.
569 xfs_trans_log_buf(tp, bp, offsetof(struct xfs_dsb, sb_icount),
570 offsetof(struct xfs_dsb, sb_frextents) +
571 sizeof(sbp->sb_frextents) - 1);
575 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations and
576 * apply superblock counter changes to the in-core superblock. The
577 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
578 * applied to the in-core superblock. The idea is that that has already been
581 * If we are not logging superblock counters, then the inode allocated/free and
582 * used block counts are not updated in the on disk superblock. In this case,
583 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
584 * still need to update the incore superblock with the changes.
586 * Deltas for the inode count are +/-64, hence we use a large batch size of 128
587 * so we don't need to take the counter lock on every update.
589 #define XFS_ICOUNT_BATCH 128
592 xfs_trans_unreserve_and_mod_sb(
593 struct xfs_trans *tp)
595 struct xfs_mount *mp = tp->t_mountp;
596 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
597 int64_t blkdelta = 0;
598 int64_t rtxdelta = 0;
600 int64_t ifreedelta = 0;
603 /* calculate deltas */
604 if (tp->t_blk_res > 0)
605 blkdelta = tp->t_blk_res;
606 if ((tp->t_fdblocks_delta != 0) &&
607 (xfs_has_lazysbcount(mp) ||
608 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
609 blkdelta += tp->t_fdblocks_delta;
611 if (tp->t_rtx_res > 0)
612 rtxdelta = tp->t_rtx_res;
613 if ((tp->t_frextents_delta != 0) &&
614 (tp->t_flags & XFS_TRANS_SB_DIRTY))
615 rtxdelta += tp->t_frextents_delta;
617 if (xfs_has_lazysbcount(mp) ||
618 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
619 idelta = tp->t_icount_delta;
620 ifreedelta = tp->t_ifree_delta;
623 /* apply the per-cpu counters */
625 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
630 percpu_counter_add_batch(&mp->m_icount, idelta,
634 percpu_counter_add(&mp->m_ifree, ifreedelta);
637 error = xfs_mod_frextents(mp, rtxdelta);
641 if (!(tp->t_flags & XFS_TRANS_SB_DIRTY))
644 /* apply remaining deltas */
645 spin_lock(&mp->m_sb_lock);
646 mp->m_sb.sb_fdblocks += tp->t_fdblocks_delta + tp->t_res_fdblocks_delta;
647 mp->m_sb.sb_icount += idelta;
648 mp->m_sb.sb_ifree += ifreedelta;
650 * Do not touch sb_frextents here because we are dealing with incore
651 * reservation. sb_frextents is not part of the lazy sb counters so it
652 * must be consistent with the ondisk rtbitmap and must never include
653 * incore reservations.
655 mp->m_sb.sb_dblocks += tp->t_dblocks_delta;
656 mp->m_sb.sb_agcount += tp->t_agcount_delta;
657 mp->m_sb.sb_imax_pct += tp->t_imaxpct_delta;
658 mp->m_sb.sb_rextsize += tp->t_rextsize_delta;
659 if (tp->t_rextsize_delta) {
660 mp->m_rtxblklog = log2_if_power2(mp->m_sb.sb_rextsize);
661 mp->m_rtxblkmask = mask64_if_power2(mp->m_sb.sb_rextsize);
663 mp->m_sb.sb_rbmblocks += tp->t_rbmblocks_delta;
664 mp->m_sb.sb_rblocks += tp->t_rblocks_delta;
665 mp->m_sb.sb_rextents += tp->t_rextents_delta;
666 mp->m_sb.sb_rextslog += tp->t_rextslog_delta;
667 spin_unlock(&mp->m_sb_lock);
670 * Debug checks outside of the spinlock so they don't lock up the
671 * machine if they fail.
673 ASSERT(mp->m_sb.sb_imax_pct >= 0);
674 ASSERT(mp->m_sb.sb_rextslog >= 0);
678 /* Add the given log item to the transaction's list of log items. */
681 struct xfs_trans *tp,
682 struct xfs_log_item *lip)
684 ASSERT(lip->li_log == tp->t_mountp->m_log);
685 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
686 ASSERT(list_empty(&lip->li_trans));
687 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->li_flags));
689 list_add_tail(&lip->li_trans, &tp->t_items);
690 trace_xfs_trans_add_item(tp, _RET_IP_);
694 * Unlink the log item from the transaction. the log item is no longer
695 * considered dirty in this transaction, as the linked transaction has
696 * finished, either by abort or commit completion.
700 struct xfs_log_item *lip)
702 clear_bit(XFS_LI_DIRTY, &lip->li_flags);
703 list_del_init(&lip->li_trans);
706 /* Detach and unlock all of the items in a transaction */
708 xfs_trans_free_items(
709 struct xfs_trans *tp,
712 struct xfs_log_item *lip, *next;
714 trace_xfs_trans_free_items(tp, _RET_IP_);
716 list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
717 xfs_trans_del_item(lip);
719 set_bit(XFS_LI_ABORTED, &lip->li_flags);
720 if (lip->li_ops->iop_release)
721 lip->li_ops->iop_release(lip);
726 xfs_log_item_batch_insert(
727 struct xfs_ail *ailp,
728 struct xfs_ail_cursor *cur,
729 struct xfs_log_item **log_items,
731 xfs_lsn_t commit_lsn)
735 spin_lock(&ailp->ail_lock);
736 /* xfs_trans_ail_update_bulk drops ailp->ail_lock */
737 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
739 for (i = 0; i < nr_items; i++) {
740 struct xfs_log_item *lip = log_items[i];
742 if (lip->li_ops->iop_unpin)
743 lip->li_ops->iop_unpin(lip, 0);
748 * Bulk operation version of xfs_trans_committed that takes a log vector of
749 * items to insert into the AIL. This uses bulk AIL insertion techniques to
750 * minimise lock traffic.
752 * If we are called with the aborted flag set, it is because a log write during
753 * a CIL checkpoint commit has failed. In this case, all the items in the
754 * checkpoint have already gone through iop_committed and iop_committing, which
755 * means that checkpoint commit abort handling is treated exactly the same
756 * as an iclog write error even though we haven't started any IO yet. Hence in
757 * this case all we need to do is iop_committed processing, followed by an
758 * iop_unpin(aborted) call.
760 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
761 * at the end of the AIL, the insert cursor avoids the need to walk
762 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
763 * call. This saves a lot of needless list walking and is a net win, even
764 * though it slightly increases that amount of AIL lock traffic to set it up
768 xfs_trans_committed_bulk(
769 struct xfs_ail *ailp,
770 struct list_head *lv_chain,
771 xfs_lsn_t commit_lsn,
774 #define LOG_ITEM_BATCH_SIZE 32
775 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
776 struct xfs_log_vec *lv;
777 struct xfs_ail_cursor cur;
780 spin_lock(&ailp->ail_lock);
781 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
782 spin_unlock(&ailp->ail_lock);
784 /* unpin all the log items */
785 list_for_each_entry(lv, lv_chain, lv_list) {
786 struct xfs_log_item *lip = lv->lv_item;
790 set_bit(XFS_LI_ABORTED, &lip->li_flags);
792 if (lip->li_ops->flags & XFS_ITEM_RELEASE_WHEN_COMMITTED) {
793 lip->li_ops->iop_release(lip);
797 if (lip->li_ops->iop_committed)
798 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
800 item_lsn = commit_lsn;
802 /* item_lsn of -1 means the item needs no further processing */
803 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
807 * if we are aborting the operation, no point in inserting the
808 * object into the AIL as we are in a shutdown situation.
811 ASSERT(xlog_is_shutdown(ailp->ail_log));
812 if (lip->li_ops->iop_unpin)
813 lip->li_ops->iop_unpin(lip, 1);
817 if (item_lsn != commit_lsn) {
820 * Not a bulk update option due to unusual item_lsn.
821 * Push into AIL immediately, rechecking the lsn once
822 * we have the ail lock. Then unpin the item. This does
823 * not affect the AIL cursor the bulk insert path is
826 spin_lock(&ailp->ail_lock);
827 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
828 xfs_trans_ail_update(ailp, lip, item_lsn);
830 spin_unlock(&ailp->ail_lock);
831 if (lip->li_ops->iop_unpin)
832 lip->li_ops->iop_unpin(lip, 0);
836 /* Item is a candidate for bulk AIL insert. */
837 log_items[i++] = lv->lv_item;
838 if (i >= LOG_ITEM_BATCH_SIZE) {
839 xfs_log_item_batch_insert(ailp, &cur, log_items,
840 LOG_ITEM_BATCH_SIZE, commit_lsn);
845 /* make sure we insert the remainder! */
847 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
849 spin_lock(&ailp->ail_lock);
850 xfs_trans_ail_cursor_done(&cur);
851 spin_unlock(&ailp->ail_lock);
855 * Sort transaction items prior to running precommit operations. This will
856 * attempt to order the items such that they will always be locked in the same
857 * order. Items that have no sort function are moved to the end of the list
858 * and so are locked last.
860 * This may need refinement as different types of objects add sort functions.
862 * Function is more complex than it needs to be because we are comparing 64 bit
863 * values and the function only returns 32 bit values.
866 xfs_trans_precommit_sort(
868 const struct list_head *a,
869 const struct list_head *b)
871 struct xfs_log_item *lia = container_of(a,
872 struct xfs_log_item, li_trans);
873 struct xfs_log_item *lib = container_of(b,
874 struct xfs_log_item, li_trans);
878 * If both items are non-sortable, leave them alone. If only one is
879 * sortable, move the non-sortable item towards the end of the list.
881 if (!lia->li_ops->iop_sort && !lib->li_ops->iop_sort)
883 if (!lia->li_ops->iop_sort)
885 if (!lib->li_ops->iop_sort)
888 diff = lia->li_ops->iop_sort(lia) - lib->li_ops->iop_sort(lib);
897 * Run transaction precommit functions.
899 * If there is an error in any of the callouts, then stop immediately and
900 * trigger a shutdown to abort the transaction. There is no recovery possible
901 * from errors at this point as the transaction is dirty....
904 xfs_trans_run_precommits(
905 struct xfs_trans *tp)
907 struct xfs_mount *mp = tp->t_mountp;
908 struct xfs_log_item *lip, *n;
912 * Sort the item list to avoid ABBA deadlocks with other transactions
913 * running precommit operations that lock multiple shared items such as
914 * inode cluster buffers.
916 list_sort(NULL, &tp->t_items, xfs_trans_precommit_sort);
919 * Precommit operations can remove the log item from the transaction
920 * if the log item exists purely to delay modifications until they
921 * can be ordered against other operations. Hence we have to use
922 * list_for_each_entry_safe() here.
924 list_for_each_entry_safe(lip, n, &tp->t_items, li_trans) {
925 if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
927 if (lip->li_ops->iop_precommit) {
928 error = lip->li_ops->iop_precommit(tp, lip);
934 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
939 * Commit the given transaction to the log.
941 * XFS disk error handling mechanism is not based on a typical
942 * transaction abort mechanism. Logically after the filesystem
943 * gets marked 'SHUTDOWN', we can't let any new transactions
944 * be durable - ie. committed to disk - because some metadata might
945 * be inconsistent. In such cases, this returns an error, and the
946 * caller may assume that all locked objects joined to the transaction
947 * have already been unlocked as if the commit had succeeded.
948 * Do not reference the transaction structure after this call.
952 struct xfs_trans *tp,
955 struct xfs_mount *mp = tp->t_mountp;
956 struct xlog *log = mp->m_log;
957 xfs_csn_t commit_seq = 0;
959 int sync = tp->t_flags & XFS_TRANS_SYNC;
961 trace_xfs_trans_commit(tp, _RET_IP_);
963 error = xfs_trans_run_precommits(tp);
965 if (tp->t_flags & XFS_TRANS_PERM_LOG_RES)
966 xfs_defer_cancel(tp);
971 * Finish deferred items on final commit. Only permanent transactions
972 * should ever have deferred ops.
974 WARN_ON_ONCE(!list_empty(&tp->t_dfops) &&
975 !(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
976 if (!regrant && (tp->t_flags & XFS_TRANS_PERM_LOG_RES)) {
977 error = xfs_defer_finish_noroll(&tp);
981 /* Run precommits from final tx in defer chain. */
982 error = xfs_trans_run_precommits(tp);
988 * If there is nothing to be logged by the transaction,
989 * then unlock all of the items associated with the
990 * transaction and free the transaction structure.
991 * Also make sure to return any reserved blocks to
994 if (!(tp->t_flags & XFS_TRANS_DIRTY))
998 * We must check against log shutdown here because we cannot abort log
999 * items and leave them dirty, inconsistent and unpinned in memory while
1000 * the log is active. This leaves them open to being written back to
1001 * disk, and that will lead to on-disk corruption.
1003 if (xlog_is_shutdown(log)) {
1008 ASSERT(tp->t_ticket != NULL);
1011 * If we need to update the superblock, then do it now.
1013 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
1014 xfs_trans_apply_sb_deltas(tp);
1015 xfs_trans_apply_dquot_deltas(tp);
1017 xlog_cil_commit(log, tp, &commit_seq, regrant);
1022 * If the transaction needs to be synchronous, then force the
1023 * log out now and wait for it.
1026 error = xfs_log_force_seq(mp, commit_seq, XFS_LOG_SYNC, NULL);
1027 XFS_STATS_INC(mp, xs_trans_sync);
1029 XFS_STATS_INC(mp, xs_trans_async);
1035 xfs_trans_unreserve_and_mod_sb(tp);
1038 * It is indeed possible for the transaction to be not dirty but
1039 * the dqinfo portion to be. All that means is that we have some
1040 * (non-persistent) quota reservations that need to be unreserved.
1042 xfs_trans_unreserve_and_mod_dquots(tp);
1044 if (regrant && !xlog_is_shutdown(log))
1045 xfs_log_ticket_regrant(log, tp->t_ticket);
1047 xfs_log_ticket_ungrant(log, tp->t_ticket);
1048 tp->t_ticket = NULL;
1050 xfs_trans_free_items(tp, !!error);
1053 XFS_STATS_INC(mp, xs_trans_empty);
1059 struct xfs_trans *tp)
1061 return __xfs_trans_commit(tp, false);
1065 * Unlock all of the transaction's items and free the transaction. If the
1066 * transaction is dirty, we must shut down the filesystem because there is no
1067 * way to restore them to their previous state.
1069 * If the transaction has made a log reservation, make sure to release it as
1072 * This is a high level function (equivalent to xfs_trans_commit()) and so can
1073 * be called after the transaction has effectively been aborted due to the mount
1074 * being shut down. However, if the mount has not been shut down and the
1075 * transaction is dirty we will shut the mount down and, in doing so, that
1076 * guarantees that the log is shut down, too. Hence we don't need to be as
1077 * careful with shutdown state and dirty items here as we need to be in
1078 * xfs_trans_commit().
1082 struct xfs_trans *tp)
1084 struct xfs_mount *mp = tp->t_mountp;
1085 struct xlog *log = mp->m_log;
1086 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
1088 trace_xfs_trans_cancel(tp, _RET_IP_);
1091 * It's never valid to cancel a transaction with deferred ops attached,
1092 * because the transaction is effectively dirty. Complain about this
1093 * loudly before freeing the in-memory defer items and shutting down the
1096 if (!list_empty(&tp->t_dfops)) {
1097 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1099 xfs_defer_cancel(tp);
1103 * See if the caller is relying on us to shut down the filesystem. We
1104 * only want an error report if there isn't already a shutdown in
1105 * progress, so we only need to check against the mount shutdown state
1108 if (dirty && !xfs_is_shutdown(mp)) {
1109 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1110 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1113 /* Log items need to be consistent until the log is shut down. */
1114 if (!dirty && !xlog_is_shutdown(log)) {
1115 struct xfs_log_item *lip;
1117 list_for_each_entry(lip, &tp->t_items, li_trans)
1118 ASSERT(!xlog_item_is_intent_done(lip));
1121 xfs_trans_unreserve_and_mod_sb(tp);
1122 xfs_trans_unreserve_and_mod_dquots(tp);
1125 xfs_log_ticket_ungrant(log, tp->t_ticket);
1126 tp->t_ticket = NULL;
1129 xfs_trans_free_items(tp, dirty);
1134 * Roll from one trans in the sequence of PERMANENT transactions to
1135 * the next: permanent transactions are only flushed out when
1136 * committed with xfs_trans_commit(), but we still want as soon
1137 * as possible to let chunks of it go to the log. So we commit the
1138 * chunk we've been working on and get a new transaction to continue.
1142 struct xfs_trans **tpp)
1144 struct xfs_trans *trans = *tpp;
1145 struct xfs_trans_res tres;
1148 trace_xfs_trans_roll(trans, _RET_IP_);
1151 * Copy the critical parameters from one trans to the next.
1153 tres.tr_logres = trans->t_log_res;
1154 tres.tr_logcount = trans->t_log_count;
1156 *tpp = xfs_trans_dup(trans);
1159 * Commit the current transaction.
1160 * If this commit failed, then it'd just unlock those items that
1161 * are not marked ihold. That also means that a filesystem shutdown
1162 * is in progress. The caller takes the responsibility to cancel
1163 * the duplicate transaction that gets returned.
1165 error = __xfs_trans_commit(trans, true);
1170 * Reserve space in the log for the next transaction.
1171 * This also pushes items in the "AIL", the list of logged items,
1172 * out to disk if they are taking up space at the tail of the log
1173 * that we want to use. This requires that either nothing be locked
1174 * across this call, or that anything that is locked be logged in
1175 * the prior and the next transactions.
1177 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1178 return xfs_trans_reserve(*tpp, &tres, 0, 0);
1182 * Allocate an transaction, lock and join the inode to it, and reserve quota.
1184 * The caller must ensure that the on-disk dquots attached to this inode have
1185 * already been allocated and initialized. The caller is responsible for
1186 * releasing ILOCK_EXCL if a new transaction is returned.
1189 xfs_trans_alloc_inode(
1190 struct xfs_inode *ip,
1191 struct xfs_trans_res *resv,
1192 unsigned int dblocks,
1193 unsigned int rblocks,
1195 struct xfs_trans **tpp)
1197 struct xfs_trans *tp;
1198 struct xfs_mount *mp = ip->i_mount;
1199 bool retried = false;
1203 error = xfs_trans_alloc(mp, resv, dblocks,
1204 xfs_extlen_to_rtxlen(mp, rblocks),
1205 force ? XFS_TRANS_RESERVE : 0, &tp);
1209 xfs_ilock(ip, XFS_ILOCK_EXCL);
1210 xfs_trans_ijoin(tp, ip, 0);
1212 error = xfs_qm_dqattach_locked(ip, false);
1214 /* Caller should have allocated the dquots! */
1215 ASSERT(error != -ENOENT);
1219 error = xfs_trans_reserve_quota_nblks(tp, ip, dblocks, rblocks, force);
1220 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1221 xfs_trans_cancel(tp);
1222 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1223 xfs_blockgc_free_quota(ip, 0);
1234 xfs_trans_cancel(tp);
1235 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1240 * Try to reserve more blocks for a transaction.
1242 * This is for callers that need to attach resources to a transaction, scan
1243 * those resources to determine the space reservation requirements, and then
1244 * modify the attached resources. In other words, online repair. This can
1245 * fail due to ENOSPC, so the caller must be able to cancel the transaction
1246 * without shutting down the fs.
1249 xfs_trans_reserve_more(
1250 struct xfs_trans *tp,
1251 unsigned int blocks,
1252 unsigned int rtextents)
1254 struct xfs_trans_res resv = { };
1256 return xfs_trans_reserve(tp, &resv, blocks, rtextents);
1260 * Try to reserve more blocks and file quota for a transaction. Same
1261 * conditions of usage as xfs_trans_reserve_more.
1264 xfs_trans_reserve_more_inode(
1265 struct xfs_trans *tp,
1266 struct xfs_inode *ip,
1267 unsigned int dblocks,
1268 unsigned int rblocks,
1271 struct xfs_trans_res resv = { };
1272 struct xfs_mount *mp = ip->i_mount;
1273 unsigned int rtx = xfs_extlen_to_rtxlen(mp, rblocks);
1276 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1278 error = xfs_trans_reserve(tp, &resv, dblocks, rtx);
1282 if (!XFS_IS_QUOTA_ON(mp) || xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
1285 if (tp->t_flags & XFS_TRANS_RESERVE)
1288 error = xfs_trans_reserve_quota_nblks(tp, ip, dblocks, rblocks,
1293 /* Quota failed, give back the new reservation. */
1294 xfs_mod_fdblocks(mp, dblocks, tp->t_flags & XFS_TRANS_RESERVE);
1295 tp->t_blk_res -= dblocks;
1296 xfs_mod_frextents(mp, rtx);
1297 tp->t_rtx_res -= rtx;
1302 * Allocate an transaction in preparation for inode creation by reserving quota
1303 * against the given dquots. Callers are not required to hold any inode locks.
1306 xfs_trans_alloc_icreate(
1307 struct xfs_mount *mp,
1308 struct xfs_trans_res *resv,
1309 struct xfs_dquot *udqp,
1310 struct xfs_dquot *gdqp,
1311 struct xfs_dquot *pdqp,
1312 unsigned int dblocks,
1313 struct xfs_trans **tpp)
1315 struct xfs_trans *tp;
1316 bool retried = false;
1320 error = xfs_trans_alloc(mp, resv, dblocks, 0, 0, &tp);
1324 error = xfs_trans_reserve_quota_icreate(tp, udqp, gdqp, pdqp, dblocks);
1325 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1326 xfs_trans_cancel(tp);
1327 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1332 xfs_trans_cancel(tp);
1341 * Allocate an transaction, lock and join the inode to it, and reserve quota
1342 * in preparation for inode attribute changes that include uid, gid, or prid
1345 * The caller must ensure that the on-disk dquots attached to this inode have
1346 * already been allocated and initialized. The ILOCK will be dropped when the
1347 * transaction is committed or cancelled.
1350 xfs_trans_alloc_ichange(
1351 struct xfs_inode *ip,
1352 struct xfs_dquot *new_udqp,
1353 struct xfs_dquot *new_gdqp,
1354 struct xfs_dquot *new_pdqp,
1356 struct xfs_trans **tpp)
1358 struct xfs_trans *tp;
1359 struct xfs_mount *mp = ip->i_mount;
1360 struct xfs_dquot *udqp;
1361 struct xfs_dquot *gdqp;
1362 struct xfs_dquot *pdqp;
1363 bool retried = false;
1367 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1371 xfs_ilock(ip, XFS_ILOCK_EXCL);
1372 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1374 error = xfs_qm_dqattach_locked(ip, false);
1376 /* Caller should have allocated the dquots! */
1377 ASSERT(error != -ENOENT);
1382 * For each quota type, skip quota reservations if the inode's dquots
1383 * now match the ones that came from the caller, or the caller didn't
1384 * pass one in. The inode's dquots can change if we drop the ILOCK to
1385 * perform a blockgc scan, so we must preserve the caller's arguments.
1387 udqp = (new_udqp != ip->i_udquot) ? new_udqp : NULL;
1388 gdqp = (new_gdqp != ip->i_gdquot) ? new_gdqp : NULL;
1389 pdqp = (new_pdqp != ip->i_pdquot) ? new_pdqp : NULL;
1390 if (udqp || gdqp || pdqp) {
1391 unsigned int qflags = XFS_QMOPT_RES_REGBLKS;
1394 qflags |= XFS_QMOPT_FORCE_RES;
1397 * Reserve enough quota to handle blocks on disk and reserved
1398 * for a delayed allocation. We'll actually transfer the
1399 * delalloc reservation between dquots at chown time, even
1400 * though that part is only semi-transactional.
1402 error = xfs_trans_reserve_quota_bydquots(tp, mp, udqp, gdqp,
1403 pdqp, ip->i_nblocks + ip->i_delayed_blks,
1405 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1406 xfs_trans_cancel(tp);
1407 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1419 xfs_trans_cancel(tp);
1424 * Allocate an transaction, lock and join the directory and child inodes to it,
1425 * and reserve quota for a directory update. If there isn't sufficient space,
1426 * @dblocks will be set to zero for a reservationless directory update and
1427 * @nospace_error will be set to a negative errno describing the space
1428 * constraint we hit.
1430 * The caller must ensure that the on-disk dquots attached to this inode have
1431 * already been allocated and initialized. The ILOCKs will be dropped when the
1432 * transaction is committed or cancelled.
1435 xfs_trans_alloc_dir(
1436 struct xfs_inode *dp,
1437 struct xfs_trans_res *resv,
1438 struct xfs_inode *ip,
1439 unsigned int *dblocks,
1440 struct xfs_trans **tpp,
1443 struct xfs_trans *tp;
1444 struct xfs_mount *mp = ip->i_mount;
1445 unsigned int resblks;
1446 bool retried = false;
1452 error = xfs_trans_alloc(mp, resv, resblks, 0, 0, &tp);
1453 if (error == -ENOSPC) {
1454 *nospace_error = error;
1456 error = xfs_trans_alloc(mp, resv, resblks, 0, 0, &tp);
1461 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
1463 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1464 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1466 error = xfs_qm_dqattach_locked(dp, false);
1468 /* Caller should have allocated the dquots! */
1469 ASSERT(error != -ENOENT);
1473 error = xfs_qm_dqattach_locked(ip, false);
1475 /* Caller should have allocated the dquots! */
1476 ASSERT(error != -ENOENT);
1483 error = xfs_trans_reserve_quota_nblks(tp, dp, resblks, 0, false);
1484 if (error == -EDQUOT || error == -ENOSPC) {
1486 xfs_trans_cancel(tp);
1487 xfs_blockgc_free_quota(dp, 0);
1492 *nospace_error = error;
1505 xfs_trans_cancel(tp);