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"
28 #include "xfs_rtgroup.h"
31 struct kmem_cache *xfs_trans_cache;
33 #if defined(CONFIG_TRACEPOINTS)
35 xfs_trans_trace_reservations(
38 struct xfs_trans_res *res;
39 struct xfs_trans_res *end_res;
42 res = (struct xfs_trans_res *)M_RES(mp);
43 end_res = (struct xfs_trans_res *)(M_RES(mp) + 1);
44 for (i = 0; res < end_res; i++, res++)
45 trace_xfs_trans_resv_calc(mp, i, res);
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_busy, false);
74 trace_xfs_trans_free(tp, _RET_IP_);
75 xfs_trans_clear_context(tp);
76 if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
77 sb_end_intwrite(tp->t_mountp->m_super);
78 xfs_trans_free_dqinfo(tp);
79 kmem_cache_free(xfs_trans_cache, tp);
83 * This is called to create a new transaction which will share the
84 * permanent log reservation of the given transaction. The remaining
85 * unused block and rt extent reservations are also inherited. This
86 * implies that the original transaction is no longer allowed to allocate
87 * blocks. Locks and log items, however, are no inherited. They must
88 * be added to the new transaction explicitly.
90 STATIC struct xfs_trans *
94 struct xfs_trans *ntp;
96 trace_xfs_trans_dup(tp, _RET_IP_);
98 ntp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
101 * Initialize the new transaction structure.
103 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
104 ntp->t_mountp = tp->t_mountp;
105 INIT_LIST_HEAD(&ntp->t_items);
106 INIT_LIST_HEAD(&ntp->t_busy);
107 INIT_LIST_HEAD(&ntp->t_dfops);
108 ntp->t_highest_agno = NULLAGNUMBER;
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_dec_fdblocks(mp, 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, resp->tr_logres,
199 &tp->t_ticket, permanent);
205 tp->t_log_res = resp->tr_logres;
206 tp->t_log_count = resp->tr_logcount;
210 * Attempt to reserve the needed realtime extents by decrementing
211 * the number needed from the number available. This will
212 * fail if the count would go below zero.
215 error = xfs_dec_frextents(mp, rtextents);
220 tp->t_rtx_res += rtextents;
226 * Error cases jump to one of these labels to undo any
227 * reservations which have already been performed.
230 if (resp->tr_logres > 0) {
231 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
234 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
239 xfs_add_fdblocks(mp, blocks);
247 struct xfs_mount *mp,
248 struct xfs_trans_res *resp,
252 struct xfs_trans **tpp)
254 struct xfs_trans *tp;
255 bool want_retry = true;
259 * Allocate the handle before we do our freeze accounting and setting up
260 * GFP_NOFS allocation context so that we avoid lockdep false positives
261 * by doing GFP_KERNEL allocations inside sb_start_intwrite().
264 tp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
265 if (!(flags & XFS_TRANS_NO_WRITECOUNT))
266 sb_start_intwrite(mp->m_super);
267 xfs_trans_set_context(tp);
270 * Zero-reservation ("empty") transactions can't modify anything, so
271 * they're allowed to run while we're frozen.
273 WARN_ON(resp->tr_logres > 0 &&
274 mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
275 ASSERT(!(flags & XFS_TRANS_RES_FDBLKS) ||
276 xfs_has_lazysbcount(mp));
278 tp->t_magic = XFS_TRANS_HEADER_MAGIC;
281 INIT_LIST_HEAD(&tp->t_items);
282 INIT_LIST_HEAD(&tp->t_busy);
283 INIT_LIST_HEAD(&tp->t_dfops);
284 tp->t_highest_agno = NULLAGNUMBER;
286 error = xfs_trans_reserve(tp, resp, blocks, rtextents);
287 if (error == -ENOSPC && want_retry) {
288 xfs_trans_cancel(tp);
291 * We weren't able to reserve enough space for the transaction.
292 * Flush the other speculative space allocations to free space.
293 * Do not perform a synchronous scan because callers can hold
296 error = xfs_blockgc_flush_all(mp);
303 xfs_trans_cancel(tp);
307 trace_xfs_trans_alloc(tp, _RET_IP_);
314 * Create an empty transaction with no reservation. This is a defensive
315 * mechanism for routines that query metadata without actually modifying them --
316 * if the metadata being queried is somehow cross-linked (think a btree block
317 * pointer that points higher in the tree), we risk deadlock. However, blocks
318 * grabbed as part of a transaction can be re-grabbed. The verifiers will
319 * notice the corrupt block and the operation will fail back to userspace
320 * without deadlocking.
322 * Note the zero-length reservation; this transaction MUST be cancelled without
325 * Callers should obtain freeze protection to avoid a conflict with fs freezing
326 * where we can be grabbing buffers at the same time that freeze is trying to
327 * drain the buffer LRU list.
330 xfs_trans_alloc_empty(
331 struct xfs_mount *mp,
332 struct xfs_trans **tpp)
334 struct xfs_trans_res resv = {0};
336 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
340 * Record the indicated change to the given field for application
341 * to the file system's superblock when the transaction commits.
342 * For now, just store the change in the transaction structure.
344 * Mark the transaction structure to indicate that the superblock
345 * needs to be updated before committing.
347 * Because we may not be keeping track of allocated/free inodes and
348 * used filesystem blocks in the superblock, we do not mark the
349 * superblock dirty in this transaction if we modify these fields.
350 * We still need to update the transaction deltas so that they get
351 * applied to the incore superblock, but we don't want them to
352 * cause the superblock to get locked and logged if these are the
353 * only fields in the superblock that the transaction modifies.
361 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
362 xfs_mount_t *mp = tp->t_mountp;
365 case XFS_TRANS_SB_ICOUNT:
366 tp->t_icount_delta += delta;
367 if (xfs_has_lazysbcount(mp))
368 flags &= ~XFS_TRANS_SB_DIRTY;
370 case XFS_TRANS_SB_IFREE:
371 tp->t_ifree_delta += delta;
372 if (xfs_has_lazysbcount(mp))
373 flags &= ~XFS_TRANS_SB_DIRTY;
375 case XFS_TRANS_SB_FDBLOCKS:
377 * Track the number of blocks allocated in the transaction.
378 * Make sure it does not exceed the number reserved. If so,
379 * shutdown as this can lead to accounting inconsistency.
382 tp->t_blk_res_used += (uint)-delta;
383 if (tp->t_blk_res_used > tp->t_blk_res)
384 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
385 } else if (delta > 0 && (tp->t_flags & XFS_TRANS_RES_FDBLKS)) {
386 int64_t blkres_delta;
389 * Return freed blocks directly to the reservation
390 * instead of the global pool, being careful not to
391 * overflow the trans counter. This is used to preserve
392 * reservation across chains of transaction rolls that
393 * repeatedly free and allocate blocks.
395 blkres_delta = min_t(int64_t, delta,
396 UINT_MAX - tp->t_blk_res);
397 tp->t_blk_res += blkres_delta;
398 delta -= blkres_delta;
400 tp->t_fdblocks_delta += delta;
401 if (xfs_has_lazysbcount(mp))
402 flags &= ~XFS_TRANS_SB_DIRTY;
404 case XFS_TRANS_SB_RES_FDBLOCKS:
406 * The allocation has already been applied to the
407 * in-core superblock's counter. This should only
408 * be applied to the on-disk superblock.
410 tp->t_res_fdblocks_delta += delta;
411 if (xfs_has_lazysbcount(mp))
412 flags &= ~XFS_TRANS_SB_DIRTY;
414 case XFS_TRANS_SB_FREXTENTS:
416 * Track the number of blocks allocated in the
417 * transaction. Make sure it does not exceed the
421 tp->t_rtx_res_used += (uint)-delta;
422 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
424 tp->t_frextents_delta += delta;
425 if (xfs_has_rtgroups(mp))
426 flags &= ~XFS_TRANS_SB_DIRTY;
428 case XFS_TRANS_SB_RES_FREXTENTS:
430 * The allocation has already been applied to the
431 * in-core superblock's counter. This should only
432 * be applied to the on-disk superblock.
435 tp->t_res_frextents_delta += delta;
436 if (xfs_has_rtgroups(mp))
437 flags &= ~XFS_TRANS_SB_DIRTY;
439 case XFS_TRANS_SB_DBLOCKS:
440 tp->t_dblocks_delta += delta;
442 case XFS_TRANS_SB_AGCOUNT:
444 tp->t_agcount_delta += delta;
446 case XFS_TRANS_SB_IMAXPCT:
447 tp->t_imaxpct_delta += delta;
449 case XFS_TRANS_SB_REXTSIZE:
450 tp->t_rextsize_delta += delta;
452 case XFS_TRANS_SB_RBMBLOCKS:
453 tp->t_rbmblocks_delta += delta;
455 case XFS_TRANS_SB_RBLOCKS:
456 tp->t_rblocks_delta += delta;
458 case XFS_TRANS_SB_REXTENTS:
459 tp->t_rextents_delta += delta;
461 case XFS_TRANS_SB_REXTSLOG:
462 tp->t_rextslog_delta += delta;
464 case XFS_TRANS_SB_RGCOUNT:
466 tp->t_rgcount_delta += delta;
473 tp->t_flags |= flags;
477 * xfs_trans_apply_sb_deltas() is called from the commit code
478 * to bring the superblock buffer into the current transaction
479 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
481 * For now we just look at each field allowed to change and change
485 xfs_trans_apply_sb_deltas(
492 bp = xfs_trans_getsb(tp);
496 * Only update the superblock counters if we are logging them
498 if (!xfs_has_lazysbcount((tp->t_mountp))) {
499 if (tp->t_icount_delta)
500 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
501 if (tp->t_ifree_delta)
502 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
503 if (tp->t_fdblocks_delta)
504 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
505 if (tp->t_res_fdblocks_delta)
506 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
510 * sb_frextents was added to the lazy sb counters when the rt groups
511 * feature was introduced. This is possible because we know that all
512 * kernels supporting rtgroups will also recompute frextents from the
515 * For older file systems, updating frextents requires careful handling
516 * because we cannot rely on log recovery in older kernels to recompute
517 * the value from the rtbitmap. This means that the ondisk frextents
518 * must be consistent with the rtbitmap.
520 * Therefore, log the frextents change to the ondisk superblock and
521 * update the incore superblock so that future calls to xfs_log_sb
522 * write the correct value ondisk.
524 if ((tp->t_frextents_delta || tp->t_res_frextents_delta) &&
525 !xfs_has_rtgroups(tp->t_mountp)) {
526 struct xfs_mount *mp = tp->t_mountp;
529 rtxdelta = tp->t_frextents_delta + tp->t_res_frextents_delta;
531 spin_lock(&mp->m_sb_lock);
532 be64_add_cpu(&sbp->sb_frextents, rtxdelta);
533 mp->m_sb.sb_frextents += rtxdelta;
534 spin_unlock(&mp->m_sb_lock);
537 if (tp->t_dblocks_delta) {
538 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
541 if (tp->t_agcount_delta) {
542 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
545 if (tp->t_imaxpct_delta) {
546 sbp->sb_imax_pct += tp->t_imaxpct_delta;
549 if (tp->t_rextsize_delta) {
550 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
553 * Because the ondisk sb records rtgroup size in units of rt
554 * extents, any time we update the rt extent size we have to
555 * recompute the ondisk rtgroup block log. The incore values
556 * will be recomputed in xfs_trans_unreserve_and_mod_sb.
558 if (xfs_has_rtgroups(tp->t_mountp)) {
559 sbp->sb_rgblklog = xfs_compute_rgblklog(
560 be32_to_cpu(sbp->sb_rgextents),
561 be32_to_cpu(sbp->sb_rextsize));
565 if (tp->t_rbmblocks_delta) {
566 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
569 if (tp->t_rblocks_delta) {
570 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
573 if (tp->t_rextents_delta) {
574 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
577 if (tp->t_rextslog_delta) {
578 sbp->sb_rextslog += tp->t_rextslog_delta;
581 if (tp->t_rgcount_delta) {
582 be32_add_cpu(&sbp->sb_rgcount, tp->t_rgcount_delta);
586 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
589 * Log the whole thing, the fields are noncontiguous.
591 xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
594 * Since all the modifiable fields are contiguous, we
595 * can get away with this.
597 xfs_trans_log_buf(tp, bp, offsetof(struct xfs_dsb, sb_icount),
598 offsetof(struct xfs_dsb, sb_frextents) +
599 sizeof(sbp->sb_frextents) - 1);
603 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations and
604 * apply superblock counter changes to the in-core superblock. The
605 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
606 * applied to the in-core superblock. The idea is that that has already been
609 * If we are not logging superblock counters, then the inode allocated/free and
610 * used block counts are not updated in the on disk superblock. In this case,
611 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
612 * still need to update the incore superblock with the changes.
614 * Deltas for the inode count are +/-64, hence we use a large batch size of 128
615 * so we don't need to take the counter lock on every update.
617 #define XFS_ICOUNT_BATCH 128
620 xfs_trans_unreserve_and_mod_sb(
621 struct xfs_trans *tp)
623 struct xfs_mount *mp = tp->t_mountp;
624 int64_t blkdelta = tp->t_blk_res;
625 int64_t rtxdelta = tp->t_rtx_res;
627 int64_t ifreedelta = 0;
630 * Calculate the deltas.
632 * t_fdblocks_delta and t_frextents_delta can be positive or negative:
634 * - positive values indicate blocks freed in the transaction.
635 * - negative values indicate blocks allocated in the transaction
637 * Negative values can only happen if the transaction has a block
638 * reservation that covers the allocated block. The end result is
639 * that the calculated delta values must always be positive and we
640 * can only put back previous allocated or reserved blocks here.
642 ASSERT(tp->t_blk_res || tp->t_fdblocks_delta >= 0);
643 if (xfs_has_lazysbcount(mp) || (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
644 blkdelta += tp->t_fdblocks_delta;
645 ASSERT(blkdelta >= 0);
648 ASSERT(tp->t_rtx_res || tp->t_frextents_delta >= 0);
649 if (xfs_has_rtgroups(mp) || (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
650 rtxdelta += tp->t_frextents_delta;
651 ASSERT(rtxdelta >= 0);
654 if (xfs_has_lazysbcount(mp) || (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
655 idelta = tp->t_icount_delta;
656 ifreedelta = tp->t_ifree_delta;
659 /* apply the per-cpu counters */
661 xfs_add_fdblocks(mp, blkdelta);
664 percpu_counter_add_batch(&mp->m_icount, idelta,
668 percpu_counter_add(&mp->m_ifree, ifreedelta);
671 xfs_add_frextents(mp, rtxdelta);
673 if (!(tp->t_flags & XFS_TRANS_SB_DIRTY))
676 /* apply remaining deltas */
677 spin_lock(&mp->m_sb_lock);
678 mp->m_sb.sb_fdblocks += tp->t_fdblocks_delta + tp->t_res_fdblocks_delta;
679 mp->m_sb.sb_icount += idelta;
680 mp->m_sb.sb_ifree += ifreedelta;
682 * Do not touch sb_frextents here because it is handled in
683 * xfs_trans_apply_sb_deltas for file systems where it isn't a lazy
686 mp->m_sb.sb_dblocks += tp->t_dblocks_delta;
687 mp->m_sb.sb_agcount += tp->t_agcount_delta;
688 mp->m_sb.sb_imax_pct += tp->t_imaxpct_delta;
689 if (tp->t_rextsize_delta)
690 xfs_mount_sb_set_rextsize(mp, &mp->m_sb,
691 mp->m_sb.sb_rextsize + tp->t_rextsize_delta);
692 mp->m_sb.sb_rbmblocks += tp->t_rbmblocks_delta;
693 mp->m_sb.sb_rblocks += tp->t_rblocks_delta;
694 mp->m_sb.sb_rextents += tp->t_rextents_delta;
695 mp->m_sb.sb_rextslog += tp->t_rextslog_delta;
696 mp->m_sb.sb_rgcount += tp->t_rgcount_delta;
697 spin_unlock(&mp->m_sb_lock);
700 * Debug checks outside of the spinlock so they don't lock up the
701 * machine if they fail.
703 ASSERT(mp->m_sb.sb_imax_pct >= 0);
704 ASSERT(mp->m_sb.sb_rextslog >= 0);
707 /* Add the given log item to the transaction's list of log items. */
710 struct xfs_trans *tp,
711 struct xfs_log_item *lip)
713 ASSERT(lip->li_log == tp->t_mountp->m_log);
714 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
715 ASSERT(list_empty(&lip->li_trans));
716 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->li_flags));
718 list_add_tail(&lip->li_trans, &tp->t_items);
719 trace_xfs_trans_add_item(tp, _RET_IP_);
723 * Unlink the log item from the transaction. the log item is no longer
724 * considered dirty in this transaction, as the linked transaction has
725 * finished, either by abort or commit completion.
729 struct xfs_log_item *lip)
731 clear_bit(XFS_LI_DIRTY, &lip->li_flags);
732 list_del_init(&lip->li_trans);
735 /* Detach and unlock all of the items in a transaction */
737 xfs_trans_free_items(
738 struct xfs_trans *tp,
741 struct xfs_log_item *lip, *next;
743 trace_xfs_trans_free_items(tp, _RET_IP_);
745 list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
746 xfs_trans_del_item(lip);
748 set_bit(XFS_LI_ABORTED, &lip->li_flags);
749 if (lip->li_ops->iop_release)
750 lip->li_ops->iop_release(lip);
755 * Sort transaction items prior to running precommit operations. This will
756 * attempt to order the items such that they will always be locked in the same
757 * order. Items that have no sort function are moved to the end of the list
758 * and so are locked last.
760 * This may need refinement as different types of objects add sort functions.
762 * Function is more complex than it needs to be because we are comparing 64 bit
763 * values and the function only returns 32 bit values.
766 xfs_trans_precommit_sort(
768 const struct list_head *a,
769 const struct list_head *b)
771 struct xfs_log_item *lia = container_of(a,
772 struct xfs_log_item, li_trans);
773 struct xfs_log_item *lib = container_of(b,
774 struct xfs_log_item, li_trans);
778 * If both items are non-sortable, leave them alone. If only one is
779 * sortable, move the non-sortable item towards the end of the list.
781 if (!lia->li_ops->iop_sort && !lib->li_ops->iop_sort)
783 if (!lia->li_ops->iop_sort)
785 if (!lib->li_ops->iop_sort)
788 diff = lia->li_ops->iop_sort(lia) - lib->li_ops->iop_sort(lib);
797 * Run transaction precommit functions.
799 * If there is an error in any of the callouts, then stop immediately and
800 * trigger a shutdown to abort the transaction. There is no recovery possible
801 * from errors at this point as the transaction is dirty....
804 xfs_trans_run_precommits(
805 struct xfs_trans *tp)
807 struct xfs_mount *mp = tp->t_mountp;
808 struct xfs_log_item *lip, *n;
812 * Sort the item list to avoid ABBA deadlocks with other transactions
813 * running precommit operations that lock multiple shared items such as
814 * inode cluster buffers.
816 list_sort(NULL, &tp->t_items, xfs_trans_precommit_sort);
819 * Precommit operations can remove the log item from the transaction
820 * if the log item exists purely to delay modifications until they
821 * can be ordered against other operations. Hence we have to use
822 * list_for_each_entry_safe() here.
824 list_for_each_entry_safe(lip, n, &tp->t_items, li_trans) {
825 if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
827 if (lip->li_ops->iop_precommit) {
828 error = lip->li_ops->iop_precommit(tp, lip);
834 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
839 * Commit the given transaction to the log.
841 * XFS disk error handling mechanism is not based on a typical
842 * transaction abort mechanism. Logically after the filesystem
843 * gets marked 'SHUTDOWN', we can't let any new transactions
844 * be durable - ie. committed to disk - because some metadata might
845 * be inconsistent. In such cases, this returns an error, and the
846 * caller may assume that all locked objects joined to the transaction
847 * have already been unlocked as if the commit had succeeded.
848 * Do not reference the transaction structure after this call.
852 struct xfs_trans *tp,
855 struct xfs_mount *mp = tp->t_mountp;
856 struct xlog *log = mp->m_log;
857 xfs_csn_t commit_seq = 0;
859 int sync = tp->t_flags & XFS_TRANS_SYNC;
861 trace_xfs_trans_commit(tp, _RET_IP_);
864 * Commit per-transaction changes that are not already tracked through
865 * log items. This can add dirty log items to the transaction.
867 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
868 xfs_trans_apply_sb_deltas(tp);
869 xfs_trans_apply_dquot_deltas(tp);
871 error = xfs_trans_run_precommits(tp);
876 * If there is nothing to be logged by the transaction,
877 * then unlock all of the items associated with the
878 * transaction and free the transaction structure.
879 * Also make sure to return any reserved blocks to
882 if (!(tp->t_flags & XFS_TRANS_DIRTY))
886 * We must check against log shutdown here because we cannot abort log
887 * items and leave them dirty, inconsistent and unpinned in memory while
888 * the log is active. This leaves them open to being written back to
889 * disk, and that will lead to on-disk corruption.
891 if (xlog_is_shutdown(log)) {
896 ASSERT(tp->t_ticket != NULL);
898 xlog_cil_commit(log, tp, &commit_seq, regrant);
903 * If the transaction needs to be synchronous, then force the
904 * log out now and wait for it.
907 error = xfs_log_force_seq(mp, commit_seq, XFS_LOG_SYNC, NULL);
908 XFS_STATS_INC(mp, xs_trans_sync);
910 XFS_STATS_INC(mp, xs_trans_async);
916 xfs_trans_unreserve_and_mod_sb(tp);
919 * It is indeed possible for the transaction to be not dirty but
920 * the dqinfo portion to be. All that means is that we have some
921 * (non-persistent) quota reservations that need to be unreserved.
923 xfs_trans_unreserve_and_mod_dquots(tp, true);
925 if (regrant && !xlog_is_shutdown(log))
926 xfs_log_ticket_regrant(log, tp->t_ticket);
928 xfs_log_ticket_ungrant(log, tp->t_ticket);
931 xfs_trans_free_items(tp, !!error);
934 XFS_STATS_INC(mp, xs_trans_empty);
940 struct xfs_trans *tp)
943 * Finish deferred items on final commit. Only permanent transactions
944 * should ever have deferred ops.
946 WARN_ON_ONCE(!list_empty(&tp->t_dfops) &&
947 !(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
948 if (tp->t_flags & XFS_TRANS_PERM_LOG_RES) {
949 int error = xfs_defer_finish_noroll(&tp);
951 xfs_trans_cancel(tp);
956 return __xfs_trans_commit(tp, false);
960 * Unlock all of the transaction's items and free the transaction. If the
961 * transaction is dirty, we must shut down the filesystem because there is no
962 * way to restore them to their previous state.
964 * If the transaction has made a log reservation, make sure to release it as
967 * This is a high level function (equivalent to xfs_trans_commit()) and so can
968 * be called after the transaction has effectively been aborted due to the mount
969 * being shut down. However, if the mount has not been shut down and the
970 * transaction is dirty we will shut the mount down and, in doing so, that
971 * guarantees that the log is shut down, too. Hence we don't need to be as
972 * careful with shutdown state and dirty items here as we need to be in
973 * xfs_trans_commit().
977 struct xfs_trans *tp)
979 struct xfs_mount *mp = tp->t_mountp;
980 struct xlog *log = mp->m_log;
981 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
983 trace_xfs_trans_cancel(tp, _RET_IP_);
986 * It's never valid to cancel a transaction with deferred ops attached,
987 * because the transaction is effectively dirty. Complain about this
988 * loudly before freeing the in-memory defer items and shutting down the
991 if (!list_empty(&tp->t_dfops)) {
992 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
994 xfs_defer_cancel(tp);
998 * See if the caller is relying on us to shut down the filesystem. We
999 * only want an error report if there isn't already a shutdown in
1000 * progress, so we only need to check against the mount shutdown state
1003 if (dirty && !xfs_is_shutdown(mp)) {
1004 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1005 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1008 /* Log items need to be consistent until the log is shut down. */
1009 if (!dirty && !xlog_is_shutdown(log)) {
1010 struct xfs_log_item *lip;
1012 list_for_each_entry(lip, &tp->t_items, li_trans)
1013 ASSERT(!xlog_item_is_intent_done(lip));
1016 xfs_trans_unreserve_and_mod_sb(tp);
1017 xfs_trans_unreserve_and_mod_dquots(tp, false);
1020 xfs_log_ticket_ungrant(log, tp->t_ticket);
1021 tp->t_ticket = NULL;
1024 xfs_trans_free_items(tp, dirty);
1029 * Roll from one trans in the sequence of PERMANENT transactions to
1030 * the next: permanent transactions are only flushed out when
1031 * committed with xfs_trans_commit(), but we still want as soon
1032 * as possible to let chunks of it go to the log. So we commit the
1033 * chunk we've been working on and get a new transaction to continue.
1037 struct xfs_trans **tpp)
1039 struct xfs_trans *trans = *tpp;
1040 struct xfs_trans_res tres;
1043 trace_xfs_trans_roll(trans, _RET_IP_);
1046 * Copy the critical parameters from one trans to the next.
1048 tres.tr_logres = trans->t_log_res;
1049 tres.tr_logcount = trans->t_log_count;
1051 *tpp = xfs_trans_dup(trans);
1054 * Commit the current transaction.
1055 * If this commit failed, then it'd just unlock those items that
1056 * are not marked ihold. That also means that a filesystem shutdown
1057 * is in progress. The caller takes the responsibility to cancel
1058 * the duplicate transaction that gets returned.
1060 error = __xfs_trans_commit(trans, true);
1065 * Reserve space in the log for the next transaction.
1066 * This also pushes items in the "AIL", the list of logged items,
1067 * out to disk if they are taking up space at the tail of the log
1068 * that we want to use. This requires that either nothing be locked
1069 * across this call, or that anything that is locked be logged in
1070 * the prior and the next transactions.
1072 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1073 return xfs_trans_reserve(*tpp, &tres, 0, 0);
1077 * Allocate an transaction, lock and join the inode to it, and reserve quota.
1079 * The caller must ensure that the on-disk dquots attached to this inode have
1080 * already been allocated and initialized. The caller is responsible for
1081 * releasing ILOCK_EXCL if a new transaction is returned.
1084 xfs_trans_alloc_inode(
1085 struct xfs_inode *ip,
1086 struct xfs_trans_res *resv,
1087 unsigned int dblocks,
1088 unsigned int rblocks,
1090 struct xfs_trans **tpp)
1092 struct xfs_trans *tp;
1093 struct xfs_mount *mp = ip->i_mount;
1094 bool retried = false;
1098 error = xfs_trans_alloc(mp, resv, dblocks,
1099 xfs_extlen_to_rtxlen(mp, rblocks),
1100 force ? XFS_TRANS_RESERVE : 0, &tp);
1104 xfs_ilock(ip, XFS_ILOCK_EXCL);
1105 xfs_trans_ijoin(tp, ip, 0);
1107 error = xfs_qm_dqattach_locked(ip, false);
1109 /* Caller should have allocated the dquots! */
1110 ASSERT(error != -ENOENT);
1114 error = xfs_trans_reserve_quota_nblks(tp, ip, dblocks, rblocks, force);
1115 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1116 xfs_trans_cancel(tp);
1117 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1118 xfs_blockgc_free_quota(ip, 0);
1129 xfs_trans_cancel(tp);
1130 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1135 * Try to reserve more blocks for a transaction.
1137 * This is for callers that need to attach resources to a transaction, scan
1138 * those resources to determine the space reservation requirements, and then
1139 * modify the attached resources. In other words, online repair. This can
1140 * fail due to ENOSPC, so the caller must be able to cancel the transaction
1141 * without shutting down the fs.
1144 xfs_trans_reserve_more(
1145 struct xfs_trans *tp,
1146 unsigned int blocks,
1147 unsigned int rtextents)
1149 struct xfs_trans_res resv = { };
1151 return xfs_trans_reserve(tp, &resv, blocks, rtextents);
1155 * Try to reserve more blocks and file quota for a transaction. Same
1156 * conditions of usage as xfs_trans_reserve_more.
1159 xfs_trans_reserve_more_inode(
1160 struct xfs_trans *tp,
1161 struct xfs_inode *ip,
1162 unsigned int dblocks,
1163 unsigned int rblocks,
1166 struct xfs_trans_res resv = { };
1167 struct xfs_mount *mp = ip->i_mount;
1168 unsigned int rtx = xfs_extlen_to_rtxlen(mp, rblocks);
1171 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1173 error = xfs_trans_reserve(tp, &resv, dblocks, rtx);
1177 if (!XFS_IS_QUOTA_ON(mp) || xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
1180 if (tp->t_flags & XFS_TRANS_RESERVE)
1183 error = xfs_trans_reserve_quota_nblks(tp, ip, dblocks, rblocks,
1188 /* Quota failed, give back the new reservation. */
1189 xfs_add_fdblocks(mp, dblocks);
1190 tp->t_blk_res -= dblocks;
1191 xfs_add_frextents(mp, rtx);
1192 tp->t_rtx_res -= rtx;
1197 * Allocate an transaction in preparation for inode creation by reserving quota
1198 * against the given dquots. Callers are not required to hold any inode locks.
1201 xfs_trans_alloc_icreate(
1202 struct xfs_mount *mp,
1203 struct xfs_trans_res *resv,
1204 struct xfs_dquot *udqp,
1205 struct xfs_dquot *gdqp,
1206 struct xfs_dquot *pdqp,
1207 unsigned int dblocks,
1208 struct xfs_trans **tpp)
1210 struct xfs_trans *tp;
1211 bool retried = false;
1215 error = xfs_trans_alloc(mp, resv, dblocks, 0, 0, &tp);
1219 error = xfs_trans_reserve_quota_icreate(tp, udqp, gdqp, pdqp, dblocks);
1220 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1221 xfs_trans_cancel(tp);
1222 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1227 xfs_trans_cancel(tp);
1236 * Allocate an transaction, lock and join the inode to it, and reserve quota
1237 * in preparation for inode attribute changes that include uid, gid, or prid
1240 * The caller must ensure that the on-disk dquots attached to this inode have
1241 * already been allocated and initialized. The ILOCK will be dropped when the
1242 * transaction is committed or cancelled.
1245 xfs_trans_alloc_ichange(
1246 struct xfs_inode *ip,
1247 struct xfs_dquot *new_udqp,
1248 struct xfs_dquot *new_gdqp,
1249 struct xfs_dquot *new_pdqp,
1251 struct xfs_trans **tpp)
1253 struct xfs_trans *tp;
1254 struct xfs_mount *mp = ip->i_mount;
1255 struct xfs_dquot *udqp;
1256 struct xfs_dquot *gdqp;
1257 struct xfs_dquot *pdqp;
1258 bool retried = false;
1262 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1266 xfs_ilock(ip, XFS_ILOCK_EXCL);
1267 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1269 error = xfs_qm_dqattach_locked(ip, false);
1271 /* Caller should have allocated the dquots! */
1272 ASSERT(error != -ENOENT);
1277 * For each quota type, skip quota reservations if the inode's dquots
1278 * now match the ones that came from the caller, or the caller didn't
1279 * pass one in. The inode's dquots can change if we drop the ILOCK to
1280 * perform a blockgc scan, so we must preserve the caller's arguments.
1282 udqp = (new_udqp != ip->i_udquot) ? new_udqp : NULL;
1283 gdqp = (new_gdqp != ip->i_gdquot) ? new_gdqp : NULL;
1284 pdqp = (new_pdqp != ip->i_pdquot) ? new_pdqp : NULL;
1285 if (udqp || gdqp || pdqp) {
1286 xfs_filblks_t dblocks, rblocks;
1287 unsigned int qflags = XFS_QMOPT_RES_REGBLKS;
1288 bool isrt = XFS_IS_REALTIME_INODE(ip);
1291 qflags |= XFS_QMOPT_FORCE_RES;
1294 error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
1299 xfs_inode_count_blocks(tp, ip, &dblocks, &rblocks);
1302 rblocks += ip->i_delayed_blks;
1304 dblocks += ip->i_delayed_blks;
1307 * Reserve enough quota to handle blocks on disk and reserved
1308 * for a delayed allocation. We'll actually transfer the
1309 * delalloc reservation between dquots at chown time, even
1310 * though that part is only semi-transactional.
1312 error = xfs_trans_reserve_quota_bydquots(tp, mp, udqp, gdqp,
1313 pdqp, dblocks, 1, qflags);
1314 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1315 xfs_trans_cancel(tp);
1316 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1323 /* Do the same for realtime. */
1324 qflags = XFS_QMOPT_RES_RTBLKS | (qflags & XFS_QMOPT_FORCE_RES);
1325 error = xfs_trans_reserve_quota_bydquots(tp, mp, udqp, gdqp,
1326 pdqp, rblocks, 0, qflags);
1327 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1328 xfs_trans_cancel(tp);
1329 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1341 xfs_trans_cancel(tp);
1346 * Allocate an transaction, lock and join the directory and child inodes to it,
1347 * and reserve quota for a directory update. If there isn't sufficient space,
1348 * @dblocks will be set to zero for a reservationless directory update and
1349 * @nospace_error will be set to a negative errno describing the space
1350 * constraint we hit.
1352 * The caller must ensure that the on-disk dquots attached to this inode have
1353 * already been allocated and initialized. The ILOCKs will be dropped when the
1354 * transaction is committed or cancelled.
1356 * Caller is responsible for unlocking the inodes manually upon return
1359 xfs_trans_alloc_dir(
1360 struct xfs_inode *dp,
1361 struct xfs_trans_res *resv,
1362 struct xfs_inode *ip,
1363 unsigned int *dblocks,
1364 struct xfs_trans **tpp,
1367 struct xfs_trans *tp;
1368 struct xfs_mount *mp = ip->i_mount;
1369 unsigned int resblks;
1370 bool retried = false;
1376 error = xfs_trans_alloc(mp, resv, resblks, 0, 0, &tp);
1377 if (error == -ENOSPC) {
1378 *nospace_error = error;
1380 error = xfs_trans_alloc(mp, resv, resblks, 0, 0, &tp);
1385 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
1387 xfs_trans_ijoin(tp, dp, 0);
1388 xfs_trans_ijoin(tp, ip, 0);
1390 error = xfs_qm_dqattach_locked(dp, false);
1392 /* Caller should have allocated the dquots! */
1393 ASSERT(error != -ENOENT);
1397 error = xfs_qm_dqattach_locked(ip, false);
1399 /* Caller should have allocated the dquots! */
1400 ASSERT(error != -ENOENT);
1407 error = xfs_trans_reserve_quota_nblks(tp, dp, resblks, 0, false);
1408 if (error == -EDQUOT || error == -ENOSPC) {
1410 xfs_trans_cancel(tp);
1411 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1413 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1414 xfs_blockgc_free_quota(dp, 0);
1419 *nospace_error = error;
1432 xfs_trans_cancel(tp);
1433 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1435 xfs_iunlock(ip, XFS_ILOCK_EXCL);