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_inode.h"
15 #include "xfs_extent_busy.h"
16 #include "xfs_quota.h"
17 #include "xfs_trans.h"
18 #include "xfs_trans_priv.h"
20 #include "xfs_trace.h"
21 #include "xfs_error.h"
22 #include "xfs_defer.h"
24 kmem_zone_t *xfs_trans_zone;
26 #if defined(CONFIG_TRACEPOINTS)
28 xfs_trans_trace_reservations(
31 struct xfs_trans_res resv;
32 struct xfs_trans_res *res;
33 struct xfs_trans_res *end_res;
36 res = (struct xfs_trans_res *)M_RES(mp);
37 end_res = (struct xfs_trans_res *)(M_RES(mp) + 1);
38 for (i = 0; res < end_res; i++, res++)
39 trace_xfs_trans_resv_calc(mp, i, res);
40 xfs_log_get_max_trans_res(mp, &resv);
41 trace_xfs_trans_resv_calc(mp, -1, &resv);
44 # define xfs_trans_trace_reservations(mp)
48 * Initialize the precomputed transaction reservation values
49 * in the mount structure.
55 xfs_trans_resv_calc(mp, M_RES(mp));
56 xfs_trans_trace_reservations(mp);
60 * Free the transaction structure. If there is more clean up
61 * to do when the structure is freed, add it here.
67 xfs_extent_busy_sort(&tp->t_busy);
68 xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
70 trace_xfs_trans_free(tp, _RET_IP_);
71 atomic_dec(&tp->t_mountp->m_active_trans);
72 if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
73 sb_end_intwrite(tp->t_mountp->m_super);
74 xfs_trans_free_dqinfo(tp);
75 kmem_zone_free(xfs_trans_zone, tp);
79 * This is called to create a new transaction which will share the
80 * permanent log reservation of the given transaction. The remaining
81 * unused block and rt extent reservations are also inherited. This
82 * implies that the original transaction is no longer allowed to allocate
83 * blocks. Locks and log items, however, are no inherited. They must
84 * be added to the new transaction explicitly.
86 STATIC struct xfs_trans *
90 struct xfs_trans *ntp;
92 trace_xfs_trans_dup(tp, _RET_IP_);
94 ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
97 * Initialize the new transaction structure.
99 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
100 ntp->t_mountp = tp->t_mountp;
101 INIT_LIST_HEAD(&ntp->t_items);
102 INIT_LIST_HEAD(&ntp->t_busy);
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;
122 ntp->t_dfops = tp->t_dfops;
124 xfs_trans_dup_dqinfo(tp, ntp);
126 atomic_inc(&tp->t_mountp->m_active_trans);
131 * This is called to reserve free disk blocks and log space for the
132 * given transaction. This must be done before allocating any resources
133 * within the transaction.
135 * This will return ENOSPC if there are not enough blocks available.
136 * It will sleep waiting for available log space.
137 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
138 * is used by long running transactions. If any one of the reservations
139 * fails then they will all be backed out.
141 * This does not do quota reservations. That typically is done by the
146 struct xfs_trans *tp,
147 struct xfs_trans_res *resp,
152 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
154 /* Mark this thread as being in a transaction */
155 current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
158 * Attempt to reserve the needed disk blocks by decrementing
159 * the number needed from the number available. This will
160 * fail if the count would go below zero.
163 error = xfs_mod_fdblocks(tp->t_mountp, -((int64_t)blocks), rsvd);
165 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
168 tp->t_blk_res += blocks;
172 * Reserve the log space needed for this transaction.
174 if (resp->tr_logres > 0) {
175 bool permanent = false;
177 ASSERT(tp->t_log_res == 0 ||
178 tp->t_log_res == resp->tr_logres);
179 ASSERT(tp->t_log_count == 0 ||
180 tp->t_log_count == resp->tr_logcount);
182 if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
183 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
186 ASSERT(tp->t_ticket == NULL);
187 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
190 if (tp->t_ticket != NULL) {
191 ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
192 error = xfs_log_regrant(tp->t_mountp, tp->t_ticket);
194 error = xfs_log_reserve(tp->t_mountp,
197 &tp->t_ticket, XFS_TRANSACTION,
204 tp->t_log_res = resp->tr_logres;
205 tp->t_log_count = resp->tr_logcount;
209 * Attempt to reserve the needed realtime extents by decrementing
210 * the number needed from the number available. This will
211 * fail if the count would go below zero.
214 error = xfs_mod_frextents(tp->t_mountp, -((int64_t)rtextents));
219 tp->t_rtx_res += rtextents;
225 * Error cases jump to one of these labels to undo any
226 * reservations which have already been performed.
229 if (resp->tr_logres > 0) {
230 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, false);
233 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
238 xfs_mod_fdblocks(tp->t_mountp, (int64_t)blocks, rsvd);
242 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
249 struct xfs_mount *mp,
250 struct xfs_trans_res *resp,
254 struct xfs_trans **tpp)
256 struct xfs_trans *tp;
259 if (!(flags & XFS_TRANS_NO_WRITECOUNT))
260 sb_start_intwrite(mp->m_super);
263 * Zero-reservation ("empty") transactions can't modify anything, so
264 * they're allowed to run while we're frozen.
266 WARN_ON(resp->tr_logres > 0 &&
267 mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
268 atomic_inc(&mp->m_active_trans);
270 tp = kmem_zone_zalloc(xfs_trans_zone,
271 (flags & XFS_TRANS_NOFS) ? KM_NOFS : KM_SLEEP);
272 tp->t_magic = XFS_TRANS_HEADER_MAGIC;
275 INIT_LIST_HEAD(&tp->t_items);
276 INIT_LIST_HEAD(&tp->t_busy);
277 tp->t_firstblock = NULLFSBLOCK;
279 error = xfs_trans_reserve(tp, resp, blocks, rtextents);
281 xfs_trans_cancel(tp);
285 trace_xfs_trans_alloc(tp, _RET_IP_);
292 * Create an empty transaction with no reservation. This is a defensive
293 * mechanism for routines that query metadata without actually modifying
294 * them -- if the metadata being queried is somehow cross-linked (think a
295 * btree block pointer that points higher in the tree), we risk deadlock.
296 * However, blocks grabbed as part of a transaction can be re-grabbed.
297 * The verifiers will notice the corrupt block and the operation will fail
298 * back to userspace without deadlocking.
300 * Note the zero-length reservation; this transaction MUST be cancelled
301 * without any dirty data.
304 xfs_trans_alloc_empty(
305 struct xfs_mount *mp,
306 struct xfs_trans **tpp)
308 struct xfs_trans_res resv = {0};
310 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
314 * Record the indicated change to the given field for application
315 * to the file system's superblock when the transaction commits.
316 * For now, just store the change in the transaction structure.
318 * Mark the transaction structure to indicate that the superblock
319 * needs to be updated before committing.
321 * Because we may not be keeping track of allocated/free inodes and
322 * used filesystem blocks in the superblock, we do not mark the
323 * superblock dirty in this transaction if we modify these fields.
324 * We still need to update the transaction deltas so that they get
325 * applied to the incore superblock, but we don't want them to
326 * cause the superblock to get locked and logged if these are the
327 * only fields in the superblock that the transaction modifies.
335 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
336 xfs_mount_t *mp = tp->t_mountp;
339 case XFS_TRANS_SB_ICOUNT:
340 tp->t_icount_delta += delta;
341 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
342 flags &= ~XFS_TRANS_SB_DIRTY;
344 case XFS_TRANS_SB_IFREE:
345 tp->t_ifree_delta += delta;
346 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
347 flags &= ~XFS_TRANS_SB_DIRTY;
349 case XFS_TRANS_SB_FDBLOCKS:
351 * Track the number of blocks allocated in the transaction.
352 * Make sure it does not exceed the number reserved. If so,
353 * shutdown as this can lead to accounting inconsistency.
356 tp->t_blk_res_used += (uint)-delta;
357 if (tp->t_blk_res_used > tp->t_blk_res)
358 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
360 tp->t_fdblocks_delta += delta;
361 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
362 flags &= ~XFS_TRANS_SB_DIRTY;
364 case XFS_TRANS_SB_RES_FDBLOCKS:
366 * The allocation has already been applied to the
367 * in-core superblock's counter. This should only
368 * be applied to the on-disk superblock.
370 tp->t_res_fdblocks_delta += delta;
371 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
372 flags &= ~XFS_TRANS_SB_DIRTY;
374 case XFS_TRANS_SB_FREXTENTS:
376 * Track the number of blocks allocated in the
377 * transaction. Make sure it does not exceed the
381 tp->t_rtx_res_used += (uint)-delta;
382 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
384 tp->t_frextents_delta += delta;
386 case XFS_TRANS_SB_RES_FREXTENTS:
388 * The allocation has already been applied to the
389 * in-core superblock's counter. This should only
390 * be applied to the on-disk superblock.
393 tp->t_res_frextents_delta += delta;
395 case XFS_TRANS_SB_DBLOCKS:
397 tp->t_dblocks_delta += delta;
399 case XFS_TRANS_SB_AGCOUNT:
401 tp->t_agcount_delta += delta;
403 case XFS_TRANS_SB_IMAXPCT:
404 tp->t_imaxpct_delta += delta;
406 case XFS_TRANS_SB_REXTSIZE:
407 tp->t_rextsize_delta += delta;
409 case XFS_TRANS_SB_RBMBLOCKS:
410 tp->t_rbmblocks_delta += delta;
412 case XFS_TRANS_SB_RBLOCKS:
413 tp->t_rblocks_delta += delta;
415 case XFS_TRANS_SB_REXTENTS:
416 tp->t_rextents_delta += delta;
418 case XFS_TRANS_SB_REXTSLOG:
419 tp->t_rextslog_delta += delta;
426 tp->t_flags |= flags;
430 * xfs_trans_apply_sb_deltas() is called from the commit code
431 * to bring the superblock buffer into the current transaction
432 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
434 * For now we just look at each field allowed to change and change
438 xfs_trans_apply_sb_deltas(
445 bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
446 sbp = XFS_BUF_TO_SBP(bp);
449 * Check that superblock mods match the mods made to AGF counters.
451 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
452 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
453 tp->t_ag_btree_delta));
456 * Only update the superblock counters if we are logging them
458 if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
459 if (tp->t_icount_delta)
460 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
461 if (tp->t_ifree_delta)
462 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
463 if (tp->t_fdblocks_delta)
464 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
465 if (tp->t_res_fdblocks_delta)
466 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
469 if (tp->t_frextents_delta)
470 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
471 if (tp->t_res_frextents_delta)
472 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
474 if (tp->t_dblocks_delta) {
475 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
478 if (tp->t_agcount_delta) {
479 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
482 if (tp->t_imaxpct_delta) {
483 sbp->sb_imax_pct += tp->t_imaxpct_delta;
486 if (tp->t_rextsize_delta) {
487 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
490 if (tp->t_rbmblocks_delta) {
491 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
494 if (tp->t_rblocks_delta) {
495 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
498 if (tp->t_rextents_delta) {
499 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
502 if (tp->t_rextslog_delta) {
503 sbp->sb_rextslog += tp->t_rextslog_delta;
507 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
510 * Log the whole thing, the fields are noncontiguous.
512 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
515 * Since all the modifiable fields are contiguous, we
516 * can get away with this.
518 xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
519 offsetof(xfs_dsb_t, sb_frextents) +
520 sizeof(sbp->sb_frextents) - 1);
528 int8_t counter = *field;
544 int32_t counter = *field;
560 int64_t counter = *field;
572 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
573 * and apply superblock counter changes to the in-core superblock. The
574 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
575 * applied to the in-core superblock. The idea is that that has already been
578 * If we are not logging superblock counters, then the inode allocated/free and
579 * used block counts are not updated in the on disk superblock. In this case,
580 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
581 * still need to update the incore superblock with the changes.
584 xfs_trans_unreserve_and_mod_sb(
585 struct xfs_trans *tp)
587 struct xfs_mount *mp = tp->t_mountp;
588 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
589 int64_t blkdelta = 0;
590 int64_t rtxdelta = 0;
592 int64_t ifreedelta = 0;
595 /* calculate deltas */
596 if (tp->t_blk_res > 0)
597 blkdelta = tp->t_blk_res;
598 if ((tp->t_fdblocks_delta != 0) &&
599 (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
600 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
601 blkdelta += tp->t_fdblocks_delta;
603 if (tp->t_rtx_res > 0)
604 rtxdelta = tp->t_rtx_res;
605 if ((tp->t_frextents_delta != 0) &&
606 (tp->t_flags & XFS_TRANS_SB_DIRTY))
607 rtxdelta += tp->t_frextents_delta;
609 if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
610 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
611 idelta = tp->t_icount_delta;
612 ifreedelta = tp->t_ifree_delta;
615 /* apply the per-cpu counters */
617 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
623 error = xfs_mod_icount(mp, idelta);
625 goto out_undo_fdblocks;
629 error = xfs_mod_ifree(mp, ifreedelta);
631 goto out_undo_icount;
634 if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
637 /* apply remaining deltas */
638 spin_lock(&mp->m_sb_lock);
640 error = xfs_sb_mod64(&mp->m_sb.sb_frextents, rtxdelta);
645 if (tp->t_dblocks_delta != 0) {
646 error = xfs_sb_mod64(&mp->m_sb.sb_dblocks, tp->t_dblocks_delta);
648 goto out_undo_frextents;
650 if (tp->t_agcount_delta != 0) {
651 error = xfs_sb_mod32(&mp->m_sb.sb_agcount, tp->t_agcount_delta);
653 goto out_undo_dblocks;
655 if (tp->t_imaxpct_delta != 0) {
656 error = xfs_sb_mod8(&mp->m_sb.sb_imax_pct, tp->t_imaxpct_delta);
658 goto out_undo_agcount;
660 if (tp->t_rextsize_delta != 0) {
661 error = xfs_sb_mod32(&mp->m_sb.sb_rextsize,
662 tp->t_rextsize_delta);
664 goto out_undo_imaxpct;
666 if (tp->t_rbmblocks_delta != 0) {
667 error = xfs_sb_mod32(&mp->m_sb.sb_rbmblocks,
668 tp->t_rbmblocks_delta);
670 goto out_undo_rextsize;
672 if (tp->t_rblocks_delta != 0) {
673 error = xfs_sb_mod64(&mp->m_sb.sb_rblocks, tp->t_rblocks_delta);
675 goto out_undo_rbmblocks;
677 if (tp->t_rextents_delta != 0) {
678 error = xfs_sb_mod64(&mp->m_sb.sb_rextents,
679 tp->t_rextents_delta);
681 goto out_undo_rblocks;
683 if (tp->t_rextslog_delta != 0) {
684 error = xfs_sb_mod8(&mp->m_sb.sb_rextslog,
685 tp->t_rextslog_delta);
687 goto out_undo_rextents;
689 spin_unlock(&mp->m_sb_lock);
693 if (tp->t_rextents_delta)
694 xfs_sb_mod64(&mp->m_sb.sb_rextents, -tp->t_rextents_delta);
696 if (tp->t_rblocks_delta)
697 xfs_sb_mod64(&mp->m_sb.sb_rblocks, -tp->t_rblocks_delta);
699 if (tp->t_rbmblocks_delta)
700 xfs_sb_mod32(&mp->m_sb.sb_rbmblocks, -tp->t_rbmblocks_delta);
702 if (tp->t_rextsize_delta)
703 xfs_sb_mod32(&mp->m_sb.sb_rextsize, -tp->t_rextsize_delta);
705 if (tp->t_rextsize_delta)
706 xfs_sb_mod8(&mp->m_sb.sb_imax_pct, -tp->t_imaxpct_delta);
708 if (tp->t_agcount_delta)
709 xfs_sb_mod32(&mp->m_sb.sb_agcount, -tp->t_agcount_delta);
711 if (tp->t_dblocks_delta)
712 xfs_sb_mod64(&mp->m_sb.sb_dblocks, -tp->t_dblocks_delta);
715 xfs_sb_mod64(&mp->m_sb.sb_frextents, -rtxdelta);
717 spin_unlock(&mp->m_sb_lock);
719 xfs_mod_ifree(mp, -ifreedelta);
722 xfs_mod_icount(mp, -idelta);
725 xfs_mod_fdblocks(mp, -blkdelta, rsvd);
731 /* Add the given log item to the transaction's list of log items. */
734 struct xfs_trans *tp,
735 struct xfs_log_item *lip)
737 ASSERT(lip->li_mountp == tp->t_mountp);
738 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
739 ASSERT(list_empty(&lip->li_trans));
740 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->li_flags));
742 list_add_tail(&lip->li_trans, &tp->t_items);
743 trace_xfs_trans_add_item(tp, _RET_IP_);
747 * Unlink the log item from the transaction. the log item is no longer
748 * considered dirty in this transaction, as the linked transaction has
749 * finished, either by abort or commit completion.
753 struct xfs_log_item *lip)
755 clear_bit(XFS_LI_DIRTY, &lip->li_flags);
756 list_del_init(&lip->li_trans);
759 /* Detach and unlock all of the items in a transaction */
761 xfs_trans_free_items(
762 struct xfs_trans *tp,
763 xfs_lsn_t commit_lsn,
766 struct xfs_log_item *lip, *next;
768 trace_xfs_trans_free_items(tp, _RET_IP_);
770 list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
771 xfs_trans_del_item(lip);
772 if (commit_lsn != NULLCOMMITLSN)
773 lip->li_ops->iop_committing(lip, commit_lsn);
775 set_bit(XFS_LI_ABORTED, &lip->li_flags);
776 lip->li_ops->iop_unlock(lip);
781 xfs_log_item_batch_insert(
782 struct xfs_ail *ailp,
783 struct xfs_ail_cursor *cur,
784 struct xfs_log_item **log_items,
786 xfs_lsn_t commit_lsn)
790 spin_lock(&ailp->ail_lock);
791 /* xfs_trans_ail_update_bulk drops ailp->ail_lock */
792 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
794 for (i = 0; i < nr_items; i++) {
795 struct xfs_log_item *lip = log_items[i];
797 lip->li_ops->iop_unpin(lip, 0);
802 * Bulk operation version of xfs_trans_committed that takes a log vector of
803 * items to insert into the AIL. This uses bulk AIL insertion techniques to
804 * minimise lock traffic.
806 * If we are called with the aborted flag set, it is because a log write during
807 * a CIL checkpoint commit has failed. In this case, all the items in the
808 * checkpoint have already gone through iop_commited and iop_unlock, which
809 * means that checkpoint commit abort handling is treated exactly the same
810 * as an iclog write error even though we haven't started any IO yet. Hence in
811 * this case all we need to do is iop_committed processing, followed by an
812 * iop_unpin(aborted) call.
814 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
815 * at the end of the AIL, the insert cursor avoids the need to walk
816 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
817 * call. This saves a lot of needless list walking and is a net win, even
818 * though it slightly increases that amount of AIL lock traffic to set it up
822 xfs_trans_committed_bulk(
823 struct xfs_ail *ailp,
824 struct xfs_log_vec *log_vector,
825 xfs_lsn_t commit_lsn,
828 #define LOG_ITEM_BATCH_SIZE 32
829 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
830 struct xfs_log_vec *lv;
831 struct xfs_ail_cursor cur;
834 spin_lock(&ailp->ail_lock);
835 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
836 spin_unlock(&ailp->ail_lock);
838 /* unpin all the log items */
839 for (lv = log_vector; lv; lv = lv->lv_next ) {
840 struct xfs_log_item *lip = lv->lv_item;
844 set_bit(XFS_LI_ABORTED, &lip->li_flags);
845 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
847 /* item_lsn of -1 means the item needs no further processing */
848 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
852 * if we are aborting the operation, no point in inserting the
853 * object into the AIL as we are in a shutdown situation.
856 ASSERT(XFS_FORCED_SHUTDOWN(ailp->ail_mount));
857 lip->li_ops->iop_unpin(lip, 1);
861 if (item_lsn != commit_lsn) {
864 * Not a bulk update option due to unusual item_lsn.
865 * Push into AIL immediately, rechecking the lsn once
866 * we have the ail lock. Then unpin the item. This does
867 * not affect the AIL cursor the bulk insert path is
870 spin_lock(&ailp->ail_lock);
871 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
872 xfs_trans_ail_update(ailp, lip, item_lsn);
874 spin_unlock(&ailp->ail_lock);
875 lip->li_ops->iop_unpin(lip, 0);
879 /* Item is a candidate for bulk AIL insert. */
880 log_items[i++] = lv->lv_item;
881 if (i >= LOG_ITEM_BATCH_SIZE) {
882 xfs_log_item_batch_insert(ailp, &cur, log_items,
883 LOG_ITEM_BATCH_SIZE, commit_lsn);
888 /* make sure we insert the remainder! */
890 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
892 spin_lock(&ailp->ail_lock);
893 xfs_trans_ail_cursor_done(&cur);
894 spin_unlock(&ailp->ail_lock);
898 * Commit the given transaction to the log.
900 * XFS disk error handling mechanism is not based on a typical
901 * transaction abort mechanism. Logically after the filesystem
902 * gets marked 'SHUTDOWN', we can't let any new transactions
903 * be durable - ie. committed to disk - because some metadata might
904 * be inconsistent. In such cases, this returns an error, and the
905 * caller may assume that all locked objects joined to the transaction
906 * have already been unlocked as if the commit had succeeded.
907 * Do not reference the transaction structure after this call.
911 struct xfs_trans *tp,
914 struct xfs_mount *mp = tp->t_mountp;
915 xfs_lsn_t commit_lsn = -1;
917 int sync = tp->t_flags & XFS_TRANS_SYNC;
919 ASSERT(!tp->t_dfops ||
920 !xfs_defer_has_unfinished_work(tp->t_dfops) || regrant);
922 trace_xfs_trans_commit(tp, _RET_IP_);
925 * If there is nothing to be logged by the transaction,
926 * then unlock all of the items associated with the
927 * transaction and free the transaction structure.
928 * Also make sure to return any reserved blocks to
931 if (!(tp->t_flags & XFS_TRANS_DIRTY))
934 if (XFS_FORCED_SHUTDOWN(mp)) {
939 ASSERT(tp->t_ticket != NULL);
942 * If we need to update the superblock, then do it now.
944 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
945 xfs_trans_apply_sb_deltas(tp);
946 xfs_trans_apply_dquot_deltas(tp);
948 xfs_log_commit_cil(mp, tp, &commit_lsn, regrant);
950 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
954 * If the transaction needs to be synchronous, then force the
955 * log out now and wait for it.
958 error = xfs_log_force_lsn(mp, commit_lsn, XFS_LOG_SYNC, NULL);
959 XFS_STATS_INC(mp, xs_trans_sync);
961 XFS_STATS_INC(mp, xs_trans_async);
967 xfs_trans_unreserve_and_mod_sb(tp);
970 * It is indeed possible for the transaction to be not dirty but
971 * the dqinfo portion to be. All that means is that we have some
972 * (non-persistent) quota reservations that need to be unreserved.
974 xfs_trans_unreserve_and_mod_dquots(tp);
976 commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, regrant);
977 if (commit_lsn == -1 && !error)
981 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
982 xfs_trans_free_items(tp, NULLCOMMITLSN, !!error);
985 XFS_STATS_INC(mp, xs_trans_empty);
991 struct xfs_trans *tp)
993 return __xfs_trans_commit(tp, false);
997 * Unlock all of the transaction's items and free the transaction.
998 * The transaction must not have modified any of its items, because
999 * there is no way to restore them to their previous state.
1001 * If the transaction has made a log reservation, make sure to release
1006 struct xfs_trans *tp)
1008 struct xfs_mount *mp = tp->t_mountp;
1009 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
1011 trace_xfs_trans_cancel(tp, _RET_IP_);
1014 * See if the caller is relying on us to shut down the
1015 * filesystem. This happens in paths where we detect
1016 * corruption and decide to give up.
1018 if (dirty && !XFS_FORCED_SHUTDOWN(mp)) {
1019 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1020 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1023 if (!dirty && !XFS_FORCED_SHUTDOWN(mp)) {
1024 struct xfs_log_item *lip;
1026 list_for_each_entry(lip, &tp->t_items, li_trans)
1027 ASSERT(!(lip->li_type == XFS_LI_EFD));
1030 xfs_trans_unreserve_and_mod_sb(tp);
1031 xfs_trans_unreserve_and_mod_dquots(tp);
1034 xfs_log_done(mp, tp->t_ticket, NULL, false);
1035 tp->t_ticket = NULL;
1038 /* mark this thread as no longer being in a transaction */
1039 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
1041 xfs_trans_free_items(tp, NULLCOMMITLSN, dirty);
1046 * Roll from one trans in the sequence of PERMANENT transactions to
1047 * the next: permanent transactions are only flushed out when
1048 * committed with xfs_trans_commit(), but we still want as soon
1049 * as possible to let chunks of it go to the log. So we commit the
1050 * chunk we've been working on and get a new transaction to continue.
1054 struct xfs_trans **tpp)
1056 struct xfs_trans *trans = *tpp;
1057 struct xfs_trans_res tres;
1060 trace_xfs_trans_roll(trans, _RET_IP_);
1063 * Copy the critical parameters from one trans to the next.
1065 tres.tr_logres = trans->t_log_res;
1066 tres.tr_logcount = trans->t_log_count;
1068 *tpp = xfs_trans_dup(trans);
1071 * Commit the current transaction.
1072 * If this commit failed, then it'd just unlock those items that
1073 * are not marked ihold. That also means that a filesystem shutdown
1074 * is in progress. The caller takes the responsibility to cancel
1075 * the duplicate transaction that gets returned.
1077 error = __xfs_trans_commit(trans, true);
1082 * Reserve space in the log for the next transaction.
1083 * This also pushes items in the "AIL", the list of logged items,
1084 * out to disk if they are taking up space at the tail of the log
1085 * that we want to use. This requires that either nothing be locked
1086 * across this call, or that anything that is locked be logged in
1087 * the prior and the next transactions.
1089 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1090 return xfs_trans_reserve(*tpp, &tres, 0, 0);