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_trace.h"
20 #include "xfs_error.h"
21 #include "xfs_defer.h"
23 kmem_zone_t *xfs_trans_zone;
25 #if defined(CONFIG_TRACEPOINTS)
27 xfs_trans_trace_reservations(
30 struct xfs_trans_res resv;
31 struct xfs_trans_res *res;
32 struct xfs_trans_res *end_res;
35 res = (struct xfs_trans_res *)M_RES(mp);
36 end_res = (struct xfs_trans_res *)(M_RES(mp) + 1);
37 for (i = 0; res < end_res; i++, res++)
38 trace_xfs_trans_resv_calc(mp, i, res);
39 xfs_log_get_max_trans_res(mp, &resv);
40 trace_xfs_trans_resv_calc(mp, -1, &resv);
43 # define xfs_trans_trace_reservations(mp)
47 * Initialize the precomputed transaction reservation values
48 * in the mount structure.
54 xfs_trans_resv_calc(mp, M_RES(mp));
55 xfs_trans_trace_reservations(mp);
59 * Free the transaction structure. If there is more clean up
60 * to do when the structure is freed, add it here.
66 xfs_extent_busy_sort(&tp->t_busy);
67 xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
69 trace_xfs_trans_free(tp, _RET_IP_);
70 atomic_dec(&tp->t_mountp->m_active_trans);
71 if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
72 sb_end_intwrite(tp->t_mountp->m_super);
73 xfs_trans_free_dqinfo(tp);
74 kmem_zone_free(xfs_trans_zone, tp);
78 * This is called to create a new transaction which will share the
79 * permanent log reservation of the given transaction. The remaining
80 * unused block and rt extent reservations are also inherited. This
81 * implies that the original transaction is no longer allowed to allocate
82 * blocks. Locks and log items, however, are no inherited. They must
83 * be added to the new transaction explicitly.
85 STATIC struct xfs_trans *
89 struct xfs_trans *ntp;
91 trace_xfs_trans_dup(tp, _RET_IP_);
93 ntp = kmem_zone_zalloc(xfs_trans_zone, 0);
96 * Initialize the new transaction structure.
98 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
99 ntp->t_mountp = tp->t_mountp;
100 INIT_LIST_HEAD(&ntp->t_items);
101 INIT_LIST_HEAD(&ntp->t_busy);
102 INIT_LIST_HEAD(&ntp->t_dfops);
103 ntp->t_firstblock = NULLFSBLOCK;
105 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
106 ASSERT(tp->t_ticket != NULL);
108 ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
109 (tp->t_flags & XFS_TRANS_RESERVE) |
110 (tp->t_flags & XFS_TRANS_NO_WRITECOUNT);
111 /* We gave our writer reference to the new transaction */
112 tp->t_flags |= XFS_TRANS_NO_WRITECOUNT;
113 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
115 ASSERT(tp->t_blk_res >= tp->t_blk_res_used);
116 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
117 tp->t_blk_res = tp->t_blk_res_used;
119 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
120 tp->t_rtx_res = tp->t_rtx_res_used;
121 ntp->t_pflags = tp->t_pflags;
123 /* move deferred ops over to the new tp */
124 xfs_defer_move(ntp, tp);
126 xfs_trans_dup_dqinfo(tp, ntp);
128 atomic_inc(&tp->t_mountp->m_active_trans);
133 * This is called to reserve free disk blocks and log space for the
134 * given transaction. This must be done before allocating any resources
135 * within the transaction.
137 * This will return ENOSPC if there are not enough blocks available.
138 * It will sleep waiting for available log space.
139 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
140 * is used by long running transactions. If any one of the reservations
141 * fails then they will all be backed out.
143 * This does not do quota reservations. That typically is done by the
148 struct xfs_trans *tp,
149 struct xfs_trans_res *resp,
154 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
156 /* Mark this thread as being in a transaction */
157 current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
160 * Attempt to reserve the needed disk blocks by decrementing
161 * the number needed from the number available. This will
162 * fail if the count would go below zero.
165 error = xfs_mod_fdblocks(tp->t_mountp, -((int64_t)blocks), rsvd);
167 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
170 tp->t_blk_res += blocks;
174 * Reserve the log space needed for this transaction.
176 if (resp->tr_logres > 0) {
177 bool permanent = false;
179 ASSERT(tp->t_log_res == 0 ||
180 tp->t_log_res == resp->tr_logres);
181 ASSERT(tp->t_log_count == 0 ||
182 tp->t_log_count == resp->tr_logcount);
184 if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
185 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
188 ASSERT(tp->t_ticket == NULL);
189 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
192 if (tp->t_ticket != NULL) {
193 ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
194 error = xfs_log_regrant(tp->t_mountp, tp->t_ticket);
196 error = xfs_log_reserve(tp->t_mountp,
199 &tp->t_ticket, XFS_TRANSACTION,
206 tp->t_log_res = resp->tr_logres;
207 tp->t_log_count = resp->tr_logcount;
211 * Attempt to reserve the needed realtime extents by decrementing
212 * the number needed from the number available. This will
213 * fail if the count would go below zero.
216 error = xfs_mod_frextents(tp->t_mountp, -((int64_t)rtextents));
221 tp->t_rtx_res += rtextents;
227 * Error cases jump to one of these labels to undo any
228 * reservations which have already been performed.
231 if (resp->tr_logres > 0) {
232 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, false);
235 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
240 xfs_mod_fdblocks(tp->t_mountp, (int64_t)blocks, rsvd);
244 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
251 struct xfs_mount *mp,
252 struct xfs_trans_res *resp,
256 struct xfs_trans **tpp)
258 struct xfs_trans *tp;
262 * Allocate the handle before we do our freeze accounting and setting up
263 * GFP_NOFS allocation context so that we avoid lockdep false positives
264 * by doing GFP_KERNEL allocations inside sb_start_intwrite().
266 tp = kmem_zone_zalloc(xfs_trans_zone, 0);
267 if (!(flags & XFS_TRANS_NO_WRITECOUNT))
268 sb_start_intwrite(mp->m_super);
271 * Zero-reservation ("empty") transactions can't modify anything, so
272 * they're allowed to run while we're frozen.
274 WARN_ON(resp->tr_logres > 0 &&
275 mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
276 atomic_inc(&mp->m_active_trans);
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_firstblock = NULLFSBLOCK;
286 error = xfs_trans_reserve(tp, resp, blocks, rtextents);
288 xfs_trans_cancel(tp);
292 trace_xfs_trans_alloc(tp, _RET_IP_);
299 * Create an empty transaction with no reservation. This is a defensive
300 * mechanism for routines that query metadata without actually modifying
301 * them -- if the metadata being queried is somehow cross-linked (think a
302 * btree block pointer that points higher in the tree), we risk deadlock.
303 * However, blocks grabbed as part of a transaction can be re-grabbed.
304 * The verifiers will notice the corrupt block and the operation will fail
305 * back to userspace without deadlocking.
307 * Note the zero-length reservation; this transaction MUST be cancelled
308 * without any dirty data.
311 xfs_trans_alloc_empty(
312 struct xfs_mount *mp,
313 struct xfs_trans **tpp)
315 struct xfs_trans_res resv = {0};
317 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
321 * Record the indicated change to the given field for application
322 * to the file system's superblock when the transaction commits.
323 * For now, just store the change in the transaction structure.
325 * Mark the transaction structure to indicate that the superblock
326 * needs to be updated before committing.
328 * Because we may not be keeping track of allocated/free inodes and
329 * used filesystem blocks in the superblock, we do not mark the
330 * superblock dirty in this transaction if we modify these fields.
331 * We still need to update the transaction deltas so that they get
332 * applied to the incore superblock, but we don't want them to
333 * cause the superblock to get locked and logged if these are the
334 * only fields in the superblock that the transaction modifies.
342 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
343 xfs_mount_t *mp = tp->t_mountp;
346 case XFS_TRANS_SB_ICOUNT:
347 tp->t_icount_delta += delta;
348 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
349 flags &= ~XFS_TRANS_SB_DIRTY;
351 case XFS_TRANS_SB_IFREE:
352 tp->t_ifree_delta += delta;
353 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
354 flags &= ~XFS_TRANS_SB_DIRTY;
356 case XFS_TRANS_SB_FDBLOCKS:
358 * Track the number of blocks allocated in the transaction.
359 * Make sure it does not exceed the number reserved. If so,
360 * shutdown as this can lead to accounting inconsistency.
363 tp->t_blk_res_used += (uint)-delta;
364 if (tp->t_blk_res_used > tp->t_blk_res)
365 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
367 tp->t_fdblocks_delta += delta;
368 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
369 flags &= ~XFS_TRANS_SB_DIRTY;
371 case XFS_TRANS_SB_RES_FDBLOCKS:
373 * The allocation has already been applied to the
374 * in-core superblock's counter. This should only
375 * be applied to the on-disk superblock.
377 tp->t_res_fdblocks_delta += delta;
378 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
379 flags &= ~XFS_TRANS_SB_DIRTY;
381 case XFS_TRANS_SB_FREXTENTS:
383 * Track the number of blocks allocated in the
384 * transaction. Make sure it does not exceed the
388 tp->t_rtx_res_used += (uint)-delta;
389 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
391 tp->t_frextents_delta += delta;
393 case XFS_TRANS_SB_RES_FREXTENTS:
395 * The allocation has already been applied to the
396 * in-core superblock's counter. This should only
397 * be applied to the on-disk superblock.
400 tp->t_res_frextents_delta += delta;
402 case XFS_TRANS_SB_DBLOCKS:
404 tp->t_dblocks_delta += delta;
406 case XFS_TRANS_SB_AGCOUNT:
408 tp->t_agcount_delta += delta;
410 case XFS_TRANS_SB_IMAXPCT:
411 tp->t_imaxpct_delta += delta;
413 case XFS_TRANS_SB_REXTSIZE:
414 tp->t_rextsize_delta += delta;
416 case XFS_TRANS_SB_RBMBLOCKS:
417 tp->t_rbmblocks_delta += delta;
419 case XFS_TRANS_SB_RBLOCKS:
420 tp->t_rblocks_delta += delta;
422 case XFS_TRANS_SB_REXTENTS:
423 tp->t_rextents_delta += delta;
425 case XFS_TRANS_SB_REXTSLOG:
426 tp->t_rextslog_delta += delta;
433 tp->t_flags |= flags;
437 * xfs_trans_apply_sb_deltas() is called from the commit code
438 * to bring the superblock buffer into the current transaction
439 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
441 * For now we just look at each field allowed to change and change
445 xfs_trans_apply_sb_deltas(
452 bp = xfs_trans_getsb(tp, tp->t_mountp);
453 sbp = XFS_BUF_TO_SBP(bp);
456 * Check that superblock mods match the mods made to AGF counters.
458 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
459 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
460 tp->t_ag_btree_delta));
463 * Only update the superblock counters if we are logging them
465 if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
466 if (tp->t_icount_delta)
467 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
468 if (tp->t_ifree_delta)
469 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
470 if (tp->t_fdblocks_delta)
471 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
472 if (tp->t_res_fdblocks_delta)
473 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
476 if (tp->t_frextents_delta)
477 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
478 if (tp->t_res_frextents_delta)
479 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
481 if (tp->t_dblocks_delta) {
482 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
485 if (tp->t_agcount_delta) {
486 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
489 if (tp->t_imaxpct_delta) {
490 sbp->sb_imax_pct += tp->t_imaxpct_delta;
493 if (tp->t_rextsize_delta) {
494 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
497 if (tp->t_rbmblocks_delta) {
498 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
501 if (tp->t_rblocks_delta) {
502 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
505 if (tp->t_rextents_delta) {
506 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
509 if (tp->t_rextslog_delta) {
510 sbp->sb_rextslog += tp->t_rextslog_delta;
514 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
517 * Log the whole thing, the fields are noncontiguous.
519 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
522 * Since all the modifiable fields are contiguous, we
523 * can get away with this.
525 xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
526 offsetof(xfs_dsb_t, sb_frextents) +
527 sizeof(sbp->sb_frextents) - 1);
535 int8_t counter = *field;
551 int32_t counter = *field;
567 int64_t counter = *field;
579 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
580 * and apply superblock counter changes to the in-core superblock. The
581 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
582 * applied to the in-core superblock. The idea is that that has already been
585 * If we are not logging superblock counters, then the inode allocated/free and
586 * used block counts are not updated in the on disk superblock. In this case,
587 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
588 * still need to update the incore superblock with the changes.
591 xfs_trans_unreserve_and_mod_sb(
592 struct xfs_trans *tp)
594 struct xfs_mount *mp = tp->t_mountp;
595 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
596 int64_t blkdelta = 0;
597 int64_t rtxdelta = 0;
599 int64_t ifreedelta = 0;
602 /* calculate deltas */
603 if (tp->t_blk_res > 0)
604 blkdelta = tp->t_blk_res;
605 if ((tp->t_fdblocks_delta != 0) &&
606 (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
607 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
608 blkdelta += tp->t_fdblocks_delta;
610 if (tp->t_rtx_res > 0)
611 rtxdelta = tp->t_rtx_res;
612 if ((tp->t_frextents_delta != 0) &&
613 (tp->t_flags & XFS_TRANS_SB_DIRTY))
614 rtxdelta += tp->t_frextents_delta;
616 if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
617 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
618 idelta = tp->t_icount_delta;
619 ifreedelta = tp->t_ifree_delta;
622 /* apply the per-cpu counters */
624 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
630 error = xfs_mod_icount(mp, idelta);
632 goto out_undo_fdblocks;
636 error = xfs_mod_ifree(mp, ifreedelta);
638 goto out_undo_icount;
641 if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
644 /* apply remaining deltas */
645 spin_lock(&mp->m_sb_lock);
647 error = xfs_sb_mod64(&mp->m_sb.sb_frextents, rtxdelta);
652 if (tp->t_dblocks_delta != 0) {
653 error = xfs_sb_mod64(&mp->m_sb.sb_dblocks, tp->t_dblocks_delta);
655 goto out_undo_frextents;
657 if (tp->t_agcount_delta != 0) {
658 error = xfs_sb_mod32(&mp->m_sb.sb_agcount, tp->t_agcount_delta);
660 goto out_undo_dblocks;
662 if (tp->t_imaxpct_delta != 0) {
663 error = xfs_sb_mod8(&mp->m_sb.sb_imax_pct, tp->t_imaxpct_delta);
665 goto out_undo_agcount;
667 if (tp->t_rextsize_delta != 0) {
668 error = xfs_sb_mod32(&mp->m_sb.sb_rextsize,
669 tp->t_rextsize_delta);
671 goto out_undo_imaxpct;
673 if (tp->t_rbmblocks_delta != 0) {
674 error = xfs_sb_mod32(&mp->m_sb.sb_rbmblocks,
675 tp->t_rbmblocks_delta);
677 goto out_undo_rextsize;
679 if (tp->t_rblocks_delta != 0) {
680 error = xfs_sb_mod64(&mp->m_sb.sb_rblocks, tp->t_rblocks_delta);
682 goto out_undo_rbmblocks;
684 if (tp->t_rextents_delta != 0) {
685 error = xfs_sb_mod64(&mp->m_sb.sb_rextents,
686 tp->t_rextents_delta);
688 goto out_undo_rblocks;
690 if (tp->t_rextslog_delta != 0) {
691 error = xfs_sb_mod8(&mp->m_sb.sb_rextslog,
692 tp->t_rextslog_delta);
694 goto out_undo_rextents;
696 spin_unlock(&mp->m_sb_lock);
700 if (tp->t_rextents_delta)
701 xfs_sb_mod64(&mp->m_sb.sb_rextents, -tp->t_rextents_delta);
703 if (tp->t_rblocks_delta)
704 xfs_sb_mod64(&mp->m_sb.sb_rblocks, -tp->t_rblocks_delta);
706 if (tp->t_rbmblocks_delta)
707 xfs_sb_mod32(&mp->m_sb.sb_rbmblocks, -tp->t_rbmblocks_delta);
709 if (tp->t_rextsize_delta)
710 xfs_sb_mod32(&mp->m_sb.sb_rextsize, -tp->t_rextsize_delta);
712 if (tp->t_rextsize_delta)
713 xfs_sb_mod8(&mp->m_sb.sb_imax_pct, -tp->t_imaxpct_delta);
715 if (tp->t_agcount_delta)
716 xfs_sb_mod32(&mp->m_sb.sb_agcount, -tp->t_agcount_delta);
718 if (tp->t_dblocks_delta)
719 xfs_sb_mod64(&mp->m_sb.sb_dblocks, -tp->t_dblocks_delta);
722 xfs_sb_mod64(&mp->m_sb.sb_frextents, -rtxdelta);
724 spin_unlock(&mp->m_sb_lock);
726 xfs_mod_ifree(mp, -ifreedelta);
729 xfs_mod_icount(mp, -idelta);
732 xfs_mod_fdblocks(mp, -blkdelta, rsvd);
738 /* Add the given log item to the transaction's list of log items. */
741 struct xfs_trans *tp,
742 struct xfs_log_item *lip)
744 ASSERT(lip->li_mountp == tp->t_mountp);
745 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
746 ASSERT(list_empty(&lip->li_trans));
747 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->li_flags));
749 list_add_tail(&lip->li_trans, &tp->t_items);
750 trace_xfs_trans_add_item(tp, _RET_IP_);
754 * Unlink the log item from the transaction. the log item is no longer
755 * considered dirty in this transaction, as the linked transaction has
756 * finished, either by abort or commit completion.
760 struct xfs_log_item *lip)
762 clear_bit(XFS_LI_DIRTY, &lip->li_flags);
763 list_del_init(&lip->li_trans);
766 /* Detach and unlock all of the items in a transaction */
768 xfs_trans_free_items(
769 struct xfs_trans *tp,
772 struct xfs_log_item *lip, *next;
774 trace_xfs_trans_free_items(tp, _RET_IP_);
776 list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
777 xfs_trans_del_item(lip);
779 set_bit(XFS_LI_ABORTED, &lip->li_flags);
780 if (lip->li_ops->iop_release)
781 lip->li_ops->iop_release(lip);
786 xfs_log_item_batch_insert(
787 struct xfs_ail *ailp,
788 struct xfs_ail_cursor *cur,
789 struct xfs_log_item **log_items,
791 xfs_lsn_t commit_lsn)
795 spin_lock(&ailp->ail_lock);
796 /* xfs_trans_ail_update_bulk drops ailp->ail_lock */
797 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
799 for (i = 0; i < nr_items; i++) {
800 struct xfs_log_item *lip = log_items[i];
802 if (lip->li_ops->iop_unpin)
803 lip->li_ops->iop_unpin(lip, 0);
808 * Bulk operation version of xfs_trans_committed that takes a log vector of
809 * items to insert into the AIL. This uses bulk AIL insertion techniques to
810 * minimise lock traffic.
812 * If we are called with the aborted flag set, it is because a log write during
813 * a CIL checkpoint commit has failed. In this case, all the items in the
814 * checkpoint have already gone through iop_committed and iop_committing, which
815 * means that checkpoint commit abort handling is treated exactly the same
816 * as an iclog write error even though we haven't started any IO yet. Hence in
817 * this case all we need to do is iop_committed processing, followed by an
818 * iop_unpin(aborted) call.
820 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
821 * at the end of the AIL, the insert cursor avoids the need to walk
822 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
823 * call. This saves a lot of needless list walking and is a net win, even
824 * though it slightly increases that amount of AIL lock traffic to set it up
828 xfs_trans_committed_bulk(
829 struct xfs_ail *ailp,
830 struct xfs_log_vec *log_vector,
831 xfs_lsn_t commit_lsn,
834 #define LOG_ITEM_BATCH_SIZE 32
835 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
836 struct xfs_log_vec *lv;
837 struct xfs_ail_cursor cur;
840 spin_lock(&ailp->ail_lock);
841 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
842 spin_unlock(&ailp->ail_lock);
844 /* unpin all the log items */
845 for (lv = log_vector; lv; lv = lv->lv_next ) {
846 struct xfs_log_item *lip = lv->lv_item;
850 set_bit(XFS_LI_ABORTED, &lip->li_flags);
852 if (lip->li_ops->flags & XFS_ITEM_RELEASE_WHEN_COMMITTED) {
853 lip->li_ops->iop_release(lip);
857 if (lip->li_ops->iop_committed)
858 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
860 item_lsn = commit_lsn;
862 /* item_lsn of -1 means the item needs no further processing */
863 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
867 * if we are aborting the operation, no point in inserting the
868 * object into the AIL as we are in a shutdown situation.
871 ASSERT(XFS_FORCED_SHUTDOWN(ailp->ail_mount));
872 if (lip->li_ops->iop_unpin)
873 lip->li_ops->iop_unpin(lip, 1);
877 if (item_lsn != commit_lsn) {
880 * Not a bulk update option due to unusual item_lsn.
881 * Push into AIL immediately, rechecking the lsn once
882 * we have the ail lock. Then unpin the item. This does
883 * not affect the AIL cursor the bulk insert path is
886 spin_lock(&ailp->ail_lock);
887 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
888 xfs_trans_ail_update(ailp, lip, item_lsn);
890 spin_unlock(&ailp->ail_lock);
891 if (lip->li_ops->iop_unpin)
892 lip->li_ops->iop_unpin(lip, 0);
896 /* Item is a candidate for bulk AIL insert. */
897 log_items[i++] = lv->lv_item;
898 if (i >= LOG_ITEM_BATCH_SIZE) {
899 xfs_log_item_batch_insert(ailp, &cur, log_items,
900 LOG_ITEM_BATCH_SIZE, commit_lsn);
905 /* make sure we insert the remainder! */
907 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
909 spin_lock(&ailp->ail_lock);
910 xfs_trans_ail_cursor_done(&cur);
911 spin_unlock(&ailp->ail_lock);
915 * Commit the given transaction to the log.
917 * XFS disk error handling mechanism is not based on a typical
918 * transaction abort mechanism. Logically after the filesystem
919 * gets marked 'SHUTDOWN', we can't let any new transactions
920 * be durable - ie. committed to disk - because some metadata might
921 * be inconsistent. In such cases, this returns an error, and the
922 * caller may assume that all locked objects joined to the transaction
923 * have already been unlocked as if the commit had succeeded.
924 * Do not reference the transaction structure after this call.
928 struct xfs_trans *tp,
931 struct xfs_mount *mp = tp->t_mountp;
932 xfs_lsn_t commit_lsn = -1;
934 int sync = tp->t_flags & XFS_TRANS_SYNC;
936 trace_xfs_trans_commit(tp, _RET_IP_);
939 * Finish deferred items on final commit. Only permanent transactions
940 * should ever have deferred ops.
942 WARN_ON_ONCE(!list_empty(&tp->t_dfops) &&
943 !(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
944 if (!regrant && (tp->t_flags & XFS_TRANS_PERM_LOG_RES)) {
945 error = xfs_defer_finish_noroll(&tp);
951 * If there is nothing to be logged by the transaction,
952 * then unlock all of the items associated with the
953 * transaction and free the transaction structure.
954 * Also make sure to return any reserved blocks to
957 if (!(tp->t_flags & XFS_TRANS_DIRTY))
960 if (XFS_FORCED_SHUTDOWN(mp)) {
965 ASSERT(tp->t_ticket != NULL);
968 * If we need to update the superblock, then do it now.
970 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
971 xfs_trans_apply_sb_deltas(tp);
972 xfs_trans_apply_dquot_deltas(tp);
974 xfs_log_commit_cil(mp, tp, &commit_lsn, regrant);
976 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
980 * If the transaction needs to be synchronous, then force the
981 * log out now and wait for it.
984 error = xfs_log_force_lsn(mp, commit_lsn, XFS_LOG_SYNC, NULL);
985 XFS_STATS_INC(mp, xs_trans_sync);
987 XFS_STATS_INC(mp, xs_trans_async);
993 xfs_trans_unreserve_and_mod_sb(tp);
996 * It is indeed possible for the transaction to be not dirty but
997 * the dqinfo portion to be. All that means is that we have some
998 * (non-persistent) quota reservations that need to be unreserved.
1000 xfs_trans_unreserve_and_mod_dquots(tp);
1002 commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, regrant);
1003 if (commit_lsn == -1 && !error)
1005 tp->t_ticket = NULL;
1007 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
1008 xfs_trans_free_items(tp, !!error);
1011 XFS_STATS_INC(mp, xs_trans_empty);
1017 struct xfs_trans *tp)
1019 return __xfs_trans_commit(tp, false);
1023 * Unlock all of the transaction's items and free the transaction.
1024 * The transaction must not have modified any of its items, because
1025 * there is no way to restore them to their previous state.
1027 * If the transaction has made a log reservation, make sure to release
1032 struct xfs_trans *tp)
1034 struct xfs_mount *mp = tp->t_mountp;
1035 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
1037 trace_xfs_trans_cancel(tp, _RET_IP_);
1039 if (tp->t_flags & XFS_TRANS_PERM_LOG_RES)
1040 xfs_defer_cancel(tp);
1043 * See if the caller is relying on us to shut down the
1044 * filesystem. This happens in paths where we detect
1045 * corruption and decide to give up.
1047 if (dirty && !XFS_FORCED_SHUTDOWN(mp)) {
1048 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1049 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1052 if (!dirty && !XFS_FORCED_SHUTDOWN(mp)) {
1053 struct xfs_log_item *lip;
1055 list_for_each_entry(lip, &tp->t_items, li_trans)
1056 ASSERT(!(lip->li_type == XFS_LI_EFD));
1059 xfs_trans_unreserve_and_mod_sb(tp);
1060 xfs_trans_unreserve_and_mod_dquots(tp);
1063 xfs_log_done(mp, tp->t_ticket, NULL, false);
1064 tp->t_ticket = NULL;
1067 /* mark this thread as no longer being in a transaction */
1068 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
1070 xfs_trans_free_items(tp, dirty);
1075 * Roll from one trans in the sequence of PERMANENT transactions to
1076 * the next: permanent transactions are only flushed out when
1077 * committed with xfs_trans_commit(), but we still want as soon
1078 * as possible to let chunks of it go to the log. So we commit the
1079 * chunk we've been working on and get a new transaction to continue.
1083 struct xfs_trans **tpp)
1085 struct xfs_trans *trans = *tpp;
1086 struct xfs_trans_res tres;
1089 trace_xfs_trans_roll(trans, _RET_IP_);
1092 * Copy the critical parameters from one trans to the next.
1094 tres.tr_logres = trans->t_log_res;
1095 tres.tr_logcount = trans->t_log_count;
1097 *tpp = xfs_trans_dup(trans);
1100 * Commit the current transaction.
1101 * If this commit failed, then it'd just unlock those items that
1102 * are not marked ihold. That also means that a filesystem shutdown
1103 * is in progress. The caller takes the responsibility to cancel
1104 * the duplicate transaction that gets returned.
1106 error = __xfs_trans_commit(trans, true);
1111 * Reserve space in the log for the next transaction.
1112 * This also pushes items in the "AIL", the list of logged items,
1113 * out to disk if they are taking up space at the tail of the log
1114 * that we want to use. This requires that either nothing be locked
1115 * across this call, or that anything that is locked be logged in
1116 * the prior and the next transactions.
1118 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1119 return xfs_trans_reserve(*tpp, &tres, 0, 0);