2 * Copyright (c) 2000-2003 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
38 #include "xfs_trans.h"
43 #include "xfs_dmapi.h"
44 #include "xfs_mount.h"
45 #include "xfs_error.h"
46 #include "xfs_trans_priv.h"
47 #include "xfs_alloc_btree.h"
48 #include "xfs_bmap_btree.h"
49 #include "xfs_ialloc_btree.h"
50 #include "xfs_btree.h"
51 #include "xfs_ialloc.h"
52 #include "xfs_alloc.h"
53 #include "xfs_attr_sf.h"
54 #include "xfs_dir_sf.h"
55 #include "xfs_dir2_sf.h"
56 #include "xfs_dinode.h"
57 #include "xfs_inode.h"
59 #include "xfs_da_btree.h"
60 #include "xfs_quota.h"
61 #include "xfs_trans_space.h"
64 STATIC void xfs_trans_apply_sb_deltas(xfs_trans_t *);
65 STATIC uint xfs_trans_count_vecs(xfs_trans_t *);
66 STATIC void xfs_trans_fill_vecs(xfs_trans_t *, xfs_log_iovec_t *);
67 STATIC void xfs_trans_uncommit(xfs_trans_t *, uint);
68 STATIC void xfs_trans_committed(xfs_trans_t *, int);
69 STATIC void xfs_trans_chunk_committed(xfs_log_item_chunk_t *, xfs_lsn_t, int);
70 STATIC void xfs_trans_free(xfs_trans_t *);
72 kmem_zone_t *xfs_trans_zone;
76 * Initialize the precomputed transaction reservation values
77 * in the mount structure.
83 xfs_trans_reservations_t *resp;
85 resp = &(mp->m_reservations);
87 (uint)(XFS_CALC_WRITE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
89 (uint)(XFS_CALC_ITRUNCATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
91 (uint)(XFS_CALC_RENAME_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
92 resp->tr_link = (uint)XFS_CALC_LINK_LOG_RES(mp);
94 (uint)(XFS_CALC_REMOVE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
96 (uint)(XFS_CALC_SYMLINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
98 (uint)(XFS_CALC_CREATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
100 (uint)(XFS_CALC_MKDIR_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
102 (uint)(XFS_CALC_IFREE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
104 (uint)(XFS_CALC_ICHANGE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
105 resp->tr_growdata = (uint)XFS_CALC_GROWDATA_LOG_RES(mp);
106 resp->tr_swrite = (uint)XFS_CALC_SWRITE_LOG_RES(mp);
107 resp->tr_writeid = (uint)XFS_CALC_WRITEID_LOG_RES(mp);
109 (uint)(XFS_CALC_ADDAFORK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
110 resp->tr_attrinval = (uint)XFS_CALC_ATTRINVAL_LOG_RES(mp);
112 (uint)(XFS_CALC_ATTRSET_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
114 (uint)(XFS_CALC_ATTRRM_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
115 resp->tr_clearagi = (uint)XFS_CALC_CLEAR_AGI_BUCKET_LOG_RES(mp);
116 resp->tr_growrtalloc = (uint)XFS_CALC_GROWRTALLOC_LOG_RES(mp);
117 resp->tr_growrtzero = (uint)XFS_CALC_GROWRTZERO_LOG_RES(mp);
118 resp->tr_growrtfree = (uint)XFS_CALC_GROWRTFREE_LOG_RES(mp);
122 * This routine is called to allocate a transaction structure.
123 * The type parameter indicates the type of the transaction. These
124 * are enumerated in xfs_trans.h.
126 * Dynamically allocate the transaction structure from the transaction
127 * zone, initialize it, and return it to the caller.
134 vfs_check_frozen(XFS_MTOVFS(mp)->vfs_super, SB_FREEZE_TRANS);
135 atomic_inc(&mp->m_active_trans);
137 return (_xfs_trans_alloc(mp, type));
148 ASSERT(xfs_trans_zone != NULL);
149 tp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
152 * Initialize the transaction structure.
154 tp->t_magic = XFS_TRANS_MAGIC;
157 tp->t_items_free = XFS_LIC_NUM_SLOTS;
158 tp->t_busy_free = XFS_LBC_NUM_SLOTS;
159 XFS_LIC_INIT(&(tp->t_items));
160 XFS_LBC_INIT(&(tp->t_busy));
166 * This is called to create a new transaction which will share the
167 * permanent log reservation of the given transaction. The remaining
168 * unused block and rt extent reservations are also inherited. This
169 * implies that the original transaction is no longer allowed to allocate
170 * blocks. Locks and log items, however, are no inherited. They must
171 * be added to the new transaction explicitly.
179 ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
182 * Initialize the new transaction structure.
184 ntp->t_magic = XFS_TRANS_MAGIC;
185 ntp->t_type = tp->t_type;
186 ntp->t_mountp = tp->t_mountp;
187 ntp->t_items_free = XFS_LIC_NUM_SLOTS;
188 ntp->t_busy_free = XFS_LBC_NUM_SLOTS;
189 XFS_LIC_INIT(&(ntp->t_items));
190 XFS_LBC_INIT(&(ntp->t_busy));
192 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
194 #if defined(XLOG_NOLOG) || defined(DEBUG)
195 ASSERT(!xlog_debug || tp->t_ticket != NULL);
197 ASSERT(tp->t_ticket != NULL);
199 ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE);
200 ntp->t_ticket = tp->t_ticket;
201 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
202 tp->t_blk_res = tp->t_blk_res_used;
203 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
204 tp->t_rtx_res = tp->t_rtx_res_used;
205 PFLAGS_DUP(&tp->t_pflags, &ntp->t_pflags);
207 XFS_TRANS_DUP_DQINFO(tp->t_mountp, tp, ntp);
209 atomic_inc(&tp->t_mountp->m_active_trans);
214 * This is called to reserve free disk blocks and log space for the
215 * given transaction. This must be done before allocating any resources
216 * within the transaction.
218 * This will return ENOSPC if there are not enough blocks available.
219 * It will sleep waiting for available log space.
220 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
221 * is used by long running transactions. If any one of the reservations
222 * fails then they will all be backed out.
224 * This does not do quota reservations. That typically is done by the
241 rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
243 /* Mark this thread as being in a transaction */
244 PFLAGS_SET_FSTRANS(&tp->t_pflags);
247 * Attempt to reserve the needed disk blocks by decrementing
248 * the number needed from the number available. This will
249 * fail if the count would go below zero.
252 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
255 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
256 return (XFS_ERROR(ENOSPC));
258 tp->t_blk_res += blocks;
262 * Reserve the log space needed for this transaction.
265 ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace));
266 ASSERT((tp->t_log_count == 0) ||
267 (tp->t_log_count == logcount));
268 if (flags & XFS_TRANS_PERM_LOG_RES) {
269 log_flags = XFS_LOG_PERM_RESERV;
270 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
272 ASSERT(tp->t_ticket == NULL);
273 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
277 error = xfs_log_reserve(tp->t_mountp, logspace, logcount,
279 XFS_TRANSACTION, log_flags);
283 tp->t_log_res = logspace;
284 tp->t_log_count = logcount;
288 * Attempt to reserve the needed realtime extents by decrementing
289 * the number needed from the number available. This will
290 * fail if the count would go below zero.
293 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
296 error = XFS_ERROR(ENOSPC);
299 tp->t_rtx_res += rtextents;
305 * Error cases jump to one of these labels to undo any
306 * reservations which have already been performed.
310 if (flags & XFS_TRANS_PERM_LOG_RES) {
311 log_flags = XFS_LOG_REL_PERM_RESERV;
315 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
318 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
323 (void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
328 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
335 * This is called to set the a callback to be called when the given
336 * transaction is committed to disk. The transaction pointer and the
337 * argument pointer will be passed to the callback routine.
339 * Only one callback can be associated with any single transaction.
344 xfs_trans_callback_t callback,
347 ASSERT(tp->t_callback == NULL);
348 tp->t_callback = callback;
354 * Record the indicated change to the given field for application
355 * to the file system's superblock when the transaction commits.
356 * For now, just store the change in the transaction structure.
358 * Mark the transaction structure to indicate that the superblock
359 * needs to be updated before committing.
369 case XFS_TRANS_SB_ICOUNT:
370 tp->t_icount_delta += delta;
372 case XFS_TRANS_SB_IFREE:
373 tp->t_ifree_delta += delta;
375 case XFS_TRANS_SB_FDBLOCKS:
377 * Track the number of blocks allocated in the
378 * transaction. Make sure it does not exceed the
382 tp->t_blk_res_used += (uint)-delta;
383 ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
385 tp->t_fdblocks_delta += delta;
387 case XFS_TRANS_SB_RES_FDBLOCKS:
389 * The allocation has already been applied to the
390 * in-core superblock's counter. This should only
391 * be applied to the on-disk superblock.
394 tp->t_res_fdblocks_delta += delta;
396 case XFS_TRANS_SB_FREXTENTS:
398 * Track the number of blocks allocated in the
399 * transaction. Make sure it does not exceed the
403 tp->t_rtx_res_used += (uint)-delta;
404 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
406 tp->t_frextents_delta += delta;
408 case XFS_TRANS_SB_RES_FREXTENTS:
410 * The allocation has already been applied to the
411 * in-core superblocks's counter. This should only
412 * be applied to the on-disk superblock.
415 tp->t_res_frextents_delta += delta;
417 case XFS_TRANS_SB_DBLOCKS:
419 tp->t_dblocks_delta += delta;
421 case XFS_TRANS_SB_AGCOUNT:
423 tp->t_agcount_delta += delta;
425 case XFS_TRANS_SB_IMAXPCT:
426 tp->t_imaxpct_delta += delta;
428 case XFS_TRANS_SB_REXTSIZE:
429 tp->t_rextsize_delta += delta;
431 case XFS_TRANS_SB_RBMBLOCKS:
432 tp->t_rbmblocks_delta += delta;
434 case XFS_TRANS_SB_RBLOCKS:
435 tp->t_rblocks_delta += delta;
437 case XFS_TRANS_SB_REXTENTS:
438 tp->t_rextents_delta += delta;
440 case XFS_TRANS_SB_REXTSLOG:
441 tp->t_rextslog_delta += delta;
448 tp->t_flags |= (XFS_TRANS_SB_DIRTY | XFS_TRANS_DIRTY);
452 * xfs_trans_apply_sb_deltas() is called from the commit code
453 * to bring the superblock buffer into the current transaction
454 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
456 * For now we just look at each field allowed to change and change
460 xfs_trans_apply_sb_deltas(
467 bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
468 sbp = XFS_BUF_TO_SBP(bp);
471 * Check that superblock mods match the mods made to AGF counters.
473 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
474 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
475 tp->t_ag_btree_delta));
477 if (tp->t_icount_delta != 0) {
478 INT_MOD(sbp->sb_icount, ARCH_CONVERT, tp->t_icount_delta);
480 if (tp->t_ifree_delta != 0) {
481 INT_MOD(sbp->sb_ifree, ARCH_CONVERT, tp->t_ifree_delta);
484 if (tp->t_fdblocks_delta != 0) {
485 INT_MOD(sbp->sb_fdblocks, ARCH_CONVERT, tp->t_fdblocks_delta);
487 if (tp->t_res_fdblocks_delta != 0) {
488 INT_MOD(sbp->sb_fdblocks, ARCH_CONVERT, tp->t_res_fdblocks_delta);
491 if (tp->t_frextents_delta != 0) {
492 INT_MOD(sbp->sb_frextents, ARCH_CONVERT, tp->t_frextents_delta);
494 if (tp->t_dblocks_delta != 0) {
495 INT_MOD(sbp->sb_dblocks, ARCH_CONVERT, tp->t_dblocks_delta);
498 if (tp->t_agcount_delta != 0) {
499 INT_MOD(sbp->sb_agcount, ARCH_CONVERT, tp->t_agcount_delta);
502 if (tp->t_imaxpct_delta != 0) {
503 INT_MOD(sbp->sb_imax_pct, ARCH_CONVERT, tp->t_imaxpct_delta);
506 if (tp->t_rextsize_delta != 0) {
507 INT_MOD(sbp->sb_rextsize, ARCH_CONVERT, tp->t_rextsize_delta);
510 if (tp->t_rbmblocks_delta != 0) {
511 INT_MOD(sbp->sb_rbmblocks, ARCH_CONVERT, tp->t_rbmblocks_delta);
514 if (tp->t_rblocks_delta != 0) {
515 INT_MOD(sbp->sb_rblocks, ARCH_CONVERT, tp->t_rblocks_delta);
518 if (tp->t_rextents_delta != 0) {
519 INT_MOD(sbp->sb_rextents, ARCH_CONVERT, tp->t_rextents_delta);
522 if (tp->t_rextslog_delta != 0) {
523 INT_MOD(sbp->sb_rextslog, ARCH_CONVERT, tp->t_rextslog_delta);
529 * Log the whole thing, the fields are discontiguous.
531 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_sb_t) - 1);
534 * Since all the modifiable fields are contiguous, we
535 * can get away with this.
537 xfs_trans_log_buf(tp, bp, offsetof(xfs_sb_t, sb_icount),
538 offsetof(xfs_sb_t, sb_frextents) +
539 sizeof(sbp->sb_frextents) - 1);
541 XFS_MTOVFS(tp->t_mountp)->vfs_super->s_dirt = 1;
545 * xfs_trans_unreserve_and_mod_sb() is called to release unused
546 * reservations and apply superblock counter changes to the in-core
549 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
552 xfs_trans_unreserve_and_mod_sb(
555 xfs_mod_sb_t msb[14]; /* If you add cases, add entries */
562 rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
565 * Release any reserved blocks. Any that were allocated
566 * will be taken back again by fdblocks_delta below.
568 if (tp->t_blk_res > 0) {
569 msbp->msb_field = XFS_SBS_FDBLOCKS;
570 msbp->msb_delta = tp->t_blk_res;
575 * Release any reserved real time extents . Any that were
576 * allocated will be taken back again by frextents_delta below.
578 if (tp->t_rtx_res > 0) {
579 msbp->msb_field = XFS_SBS_FREXTENTS;
580 msbp->msb_delta = tp->t_rtx_res;
585 * Apply any superblock modifications to the in-core version.
586 * The t_res_fdblocks_delta and t_res_frextents_delta fields are
587 * explicity NOT applied to the in-core superblock.
588 * The idea is that that has already been done.
590 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
591 if (tp->t_icount_delta != 0) {
592 msbp->msb_field = XFS_SBS_ICOUNT;
593 msbp->msb_delta = (int)tp->t_icount_delta;
596 if (tp->t_ifree_delta != 0) {
597 msbp->msb_field = XFS_SBS_IFREE;
598 msbp->msb_delta = (int)tp->t_ifree_delta;
601 if (tp->t_fdblocks_delta != 0) {
602 msbp->msb_field = XFS_SBS_FDBLOCKS;
603 msbp->msb_delta = (int)tp->t_fdblocks_delta;
606 if (tp->t_frextents_delta != 0) {
607 msbp->msb_field = XFS_SBS_FREXTENTS;
608 msbp->msb_delta = (int)tp->t_frextents_delta;
611 if (tp->t_dblocks_delta != 0) {
612 msbp->msb_field = XFS_SBS_DBLOCKS;
613 msbp->msb_delta = (int)tp->t_dblocks_delta;
616 if (tp->t_agcount_delta != 0) {
617 msbp->msb_field = XFS_SBS_AGCOUNT;
618 msbp->msb_delta = (int)tp->t_agcount_delta;
621 if (tp->t_imaxpct_delta != 0) {
622 msbp->msb_field = XFS_SBS_IMAX_PCT;
623 msbp->msb_delta = (int)tp->t_imaxpct_delta;
626 if (tp->t_rextsize_delta != 0) {
627 msbp->msb_field = XFS_SBS_REXTSIZE;
628 msbp->msb_delta = (int)tp->t_rextsize_delta;
631 if (tp->t_rbmblocks_delta != 0) {
632 msbp->msb_field = XFS_SBS_RBMBLOCKS;
633 msbp->msb_delta = (int)tp->t_rbmblocks_delta;
636 if (tp->t_rblocks_delta != 0) {
637 msbp->msb_field = XFS_SBS_RBLOCKS;
638 msbp->msb_delta = (int)tp->t_rblocks_delta;
641 if (tp->t_rextents_delta != 0) {
642 msbp->msb_field = XFS_SBS_REXTENTS;
643 msbp->msb_delta = (int)tp->t_rextents_delta;
646 if (tp->t_rextslog_delta != 0) {
647 msbp->msb_field = XFS_SBS_REXTSLOG;
648 msbp->msb_delta = (int)tp->t_rextslog_delta;
654 * If we need to change anything, do it.
657 error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
658 (uint)(msbp - msb), rsvd);
667 * Commit the given transaction to the log a/synchronously.
669 * XFS disk error handling mechanism is not based on a typical
670 * transaction abort mechanism. Logically after the filesystem
671 * gets marked 'SHUTDOWN', we can't let any new transactions
672 * be durable - ie. committed to disk - because some metadata might
673 * be inconsistent. In such cases, this returns an error, and the
674 * caller may assume that all locked objects joined to the transaction
675 * have already been unlocked as if the commit had succeeded.
676 * Do not reference the transaction structure after this call.
683 xfs_lsn_t *commit_lsn_p)
685 xfs_log_iovec_t *log_vector;
688 xfs_lsn_t commit_lsn;
693 #define XFS_TRANS_LOGVEC_COUNT 16
694 xfs_log_iovec_t log_vector_fast[XFS_TRANS_LOGVEC_COUNT];
695 #if defined(XLOG_NOLOG) || defined(DEBUG)
696 static xfs_lsn_t trans_lsn = 1;
704 * Determine whether this commit is releasing a permanent
705 * log reservation or not.
707 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
708 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
709 log_flags = XFS_LOG_REL_PERM_RESERV;
716 * If there is nothing to be logged by the transaction,
717 * then unlock all of the items associated with the
718 * transaction and free the transaction structure.
719 * Also make sure to return any reserved blocks to
723 shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0;
724 if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) {
725 xfs_trans_unreserve_and_mod_sb(tp);
727 * It is indeed possible for the transaction to be
728 * not dirty but the dqinfo portion to be. All that
729 * means is that we have some (non-persistent) quota
730 * reservations that need to be unreserved.
732 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(mp, tp);
734 commit_lsn = xfs_log_done(mp, tp->t_ticket,
736 if (commit_lsn == -1 && !shutdown)
737 shutdown = XFS_ERROR(EIO);
739 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
740 xfs_trans_free_items(tp, shutdown? XFS_TRANS_ABORT : 0);
741 xfs_trans_free_busy(tp);
743 XFS_STATS_INC(xs_trans_empty);
745 *commit_lsn_p = commit_lsn;
748 #if defined(XLOG_NOLOG) || defined(DEBUG)
749 ASSERT(!xlog_debug || tp->t_ticket != NULL);
751 ASSERT(tp->t_ticket != NULL);
755 * If we need to update the superblock, then do it now.
757 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
758 xfs_trans_apply_sb_deltas(tp);
760 XFS_TRANS_APPLY_DQUOT_DELTAS(mp, tp);
763 * Ask each log item how many log_vector entries it will
764 * need so we can figure out how many to allocate.
765 * Try to avoid the kmem_alloc() call in the common case
766 * by using a vector from the stack when it fits.
768 nvec = xfs_trans_count_vecs(tp);
771 xfs_force_shutdown(mp, XFS_LOG_IO_ERROR);
776 if (nvec <= XFS_TRANS_LOGVEC_COUNT) {
777 log_vector = log_vector_fast;
779 log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec *
780 sizeof(xfs_log_iovec_t),
785 * Fill in the log_vector and pin the logged items, and
786 * then write the transaction to the log.
788 xfs_trans_fill_vecs(tp, log_vector);
791 * Ignore errors here. xfs_log_done would do the right thing.
792 * We need to put the ticket, etc. away.
794 error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket,
797 #if defined(XLOG_NOLOG) || defined(DEBUG)
799 commit_lsn = xfs_log_done(mp, tp->t_ticket,
800 &commit_iclog, log_flags);
803 tp->t_lsn = trans_lsn++;
807 * This is the regular case. At this point (after the call finishes),
808 * the transaction is committed incore and could go out to disk at
809 * any time. However, all the items associated with the transaction
810 * are still locked and pinned in memory.
812 commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);
815 tp->t_commit_lsn = commit_lsn;
816 if (nvec > XFS_TRANS_LOGVEC_COUNT) {
817 kmem_free(log_vector, nvec * sizeof(xfs_log_iovec_t));
821 *commit_lsn_p = commit_lsn;
824 * If we got a log write error. Unpin the logitems that we
825 * had pinned, clean up, free trans structure, and return error.
827 if (error || commit_lsn == -1) {
828 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
829 xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT);
830 return XFS_ERROR(EIO);
834 * Once the transaction has committed, unused
835 * reservations need to be released and changes to
836 * the superblock need to be reflected in the in-core
837 * version. Do that now.
839 xfs_trans_unreserve_and_mod_sb(tp);
841 sync = tp->t_flags & XFS_TRANS_SYNC;
844 * Tell the LM to call the transaction completion routine
845 * when the log write with LSN commit_lsn completes (e.g.
846 * when the transaction commit really hits the on-disk log).
847 * After this call we cannot reference tp, because the call
848 * can happen at any time and the call will free the transaction
849 * structure pointed to by tp. The only case where we call
850 * the completion routine (xfs_trans_committed) directly is
851 * if the log is turned off on a debug kernel or we're
852 * running in simulation mode (the log is explicitly turned
855 tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed;
856 tp->t_logcb.cb_arg = tp;
859 * We need to pass the iclog buffer which was used for the
860 * transaction commit record into this function, and attach
861 * the callback to it. The callback must be attached before
862 * the items are unlocked to avoid racing with other threads
863 * waiting for an item to unlock.
865 shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb));
868 * Mark this thread as no longer being in a transaction
870 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
873 * Once all the items of the transaction have been copied
874 * to the in core log and the callback is attached, the
875 * items can be unlocked.
877 * This will free descriptors pointing to items which were
878 * not logged since there is nothing more to do with them.
879 * For items which were logged, we will keep pointers to them
880 * so they can be unpinned after the transaction commits to disk.
881 * This will also stamp each modified meta-data item with
882 * the commit lsn of this transaction for dependency tracking
885 xfs_trans_unlock_items(tp, commit_lsn);
888 * If we detected a log error earlier, finish committing
889 * the transaction now (unpin log items, etc).
891 * Order is critical here, to avoid using the transaction
892 * pointer after its been freed (by xfs_trans_committed
893 * either here now, or as a callback). We cannot do this
894 * step inside xfs_log_notify as was done earlier because
898 xfs_trans_committed(tp, XFS_LI_ABORTED);
901 * Now that the xfs_trans_committed callback has been attached,
902 * and the items are released we can finally allow the iclog to
905 error = xfs_log_release_iclog(mp, commit_iclog);
908 * If the transaction needs to be synchronous, then force the
909 * log out now and wait for it.
913 error = xfs_log_force(mp, commit_lsn,
914 XFS_LOG_FORCE | XFS_LOG_SYNC);
915 XFS_STATS_INC(xs_trans_sync);
917 XFS_STATS_INC(xs_trans_async);
925 * Total up the number of log iovecs needed to commit this
926 * transaction. The transaction itself needs one for the
927 * transaction header. Ask each dirty item in turn how many
928 * it needs to get the total.
931 xfs_trans_count_vecs(
935 xfs_log_item_desc_t *lidp;
938 lidp = xfs_trans_first_item(tp);
939 ASSERT(lidp != NULL);
941 /* In the non-debug case we need to start bailing out if we
942 * didn't find a log_item here, return zero and let trans_commit
948 while (lidp != NULL) {
950 * Skip items which aren't dirty in this transaction.
952 if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
953 lidp = xfs_trans_next_item(tp, lidp);
956 lidp->lid_size = IOP_SIZE(lidp->lid_item);
957 nvecs += lidp->lid_size;
958 lidp = xfs_trans_next_item(tp, lidp);
965 * Called from the trans_commit code when we notice that
966 * the filesystem is in the middle of a forced shutdown.
973 xfs_log_item_desc_t *lidp;
975 for (lidp = xfs_trans_first_item(tp);
977 lidp = xfs_trans_next_item(tp, lidp)) {
979 * Unpin all but those that aren't dirty.
981 if (lidp->lid_flags & XFS_LID_DIRTY)
982 IOP_UNPIN_REMOVE(lidp->lid_item, tp);
985 xfs_trans_unreserve_and_mod_sb(tp);
986 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(tp->t_mountp, tp);
988 xfs_trans_free_items(tp, flags);
989 xfs_trans_free_busy(tp);
994 * Fill in the vector with pointers to data to be logged
995 * by this transaction. The transaction header takes
996 * the first vector, and then each dirty item takes the
997 * number of vectors it indicated it needed in xfs_trans_count_vecs().
999 * As each item fills in the entries it needs, also pin the item
1000 * so that it cannot be flushed out until the log write completes.
1003 xfs_trans_fill_vecs(
1005 xfs_log_iovec_t *log_vector)
1007 xfs_log_item_desc_t *lidp;
1008 xfs_log_iovec_t *vecp;
1012 * Skip over the entry for the transaction header, we'll
1013 * fill that in at the end.
1015 vecp = log_vector + 1; /* pointer arithmetic */
1018 lidp = xfs_trans_first_item(tp);
1019 ASSERT(lidp != NULL);
1020 while (lidp != NULL) {
1022 * Skip items which aren't dirty in this transaction.
1024 if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
1025 lidp = xfs_trans_next_item(tp, lidp);
1029 * The item may be marked dirty but not log anything.
1030 * This can be used to get called when a transaction
1033 if (lidp->lid_size) {
1036 IOP_FORMAT(lidp->lid_item, vecp);
1037 vecp += lidp->lid_size; /* pointer arithmetic */
1038 IOP_PIN(lidp->lid_item);
1039 lidp = xfs_trans_next_item(tp, lidp);
1043 * Now that we've counted the number of items in this
1044 * transaction, fill in the transaction header.
1046 tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC;
1047 tp->t_header.th_type = tp->t_type;
1048 tp->t_header.th_num_items = nitems;
1049 log_vector->i_addr = (xfs_caddr_t)&tp->t_header;
1050 log_vector->i_len = sizeof(xfs_trans_header_t);
1055 * Unlock all of the transaction's items and free the transaction.
1056 * The transaction must not have modified any of its items, because
1057 * there is no way to restore them to their previous state.
1059 * If the transaction has made a log reservation, make sure to release
1069 xfs_log_item_chunk_t *licp;
1070 xfs_log_item_desc_t *lidp;
1071 xfs_log_item_t *lip;
1076 * See if the caller is being too lazy to figure out if
1077 * the transaction really needs an abort.
1079 if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1080 flags &= ~XFS_TRANS_ABORT;
1082 * See if the caller is relying on us to shut down the
1083 * filesystem. This happens in paths where we detect
1084 * corruption and decide to give up.
1086 if ((tp->t_flags & XFS_TRANS_DIRTY) &&
1087 !XFS_FORCED_SHUTDOWN(tp->t_mountp))
1088 xfs_force_shutdown(tp->t_mountp, XFS_CORRUPT_INCORE);
1090 if (!(flags & XFS_TRANS_ABORT)) {
1091 licp = &(tp->t_items);
1092 while (licp != NULL) {
1093 lidp = licp->lic_descs;
1094 for (i = 0; i < licp->lic_unused; i++, lidp++) {
1095 if (XFS_LIC_ISFREE(licp, i)) {
1099 lip = lidp->lid_item;
1100 if (!XFS_FORCED_SHUTDOWN(tp->t_mountp))
1101 ASSERT(!(lip->li_type == XFS_LI_EFD));
1103 licp = licp->lic_next;
1107 xfs_trans_unreserve_and_mod_sb(tp);
1108 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(tp->t_mountp, tp);
1111 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1112 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1113 log_flags = XFS_LOG_REL_PERM_RESERV;
1117 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
1120 /* mark this thread as no longer being in a transaction */
1121 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
1123 xfs_trans_free_items(tp, flags);
1124 xfs_trans_free_busy(tp);
1130 * Free the transaction structure. If there is more clean up
1131 * to do when the structure is freed, add it here.
1137 atomic_dec(&tp->t_mountp->m_active_trans);
1138 XFS_TRANS_FREE_DQINFO(tp->t_mountp, tp);
1139 kmem_zone_free(xfs_trans_zone, tp);
1144 * THIS SHOULD BE REWRITTEN TO USE xfs_trans_next_item().
1146 * This is typically called by the LM when a transaction has been fully
1147 * committed to disk. It needs to unpin the items which have
1148 * been logged by the transaction and update their positions
1149 * in the AIL if necessary.
1150 * This also gets called when the transactions didn't get written out
1151 * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then.
1153 * Call xfs_trans_chunk_committed() to process the items in
1157 xfs_trans_committed(
1161 xfs_log_item_chunk_t *licp;
1162 xfs_log_item_chunk_t *next_licp;
1163 xfs_log_busy_chunk_t *lbcp;
1164 xfs_log_busy_slot_t *lbsp;
1168 * Call the transaction's completion callback if there
1171 if (tp->t_callback != NULL) {
1172 tp->t_callback(tp, tp->t_callarg);
1176 * Special case the chunk embedded in the transaction.
1178 licp = &(tp->t_items);
1179 if (!(XFS_LIC_ARE_ALL_FREE(licp))) {
1180 xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
1184 * Process the items in each chunk in turn.
1186 licp = licp->lic_next;
1187 while (licp != NULL) {
1188 ASSERT(!XFS_LIC_ARE_ALL_FREE(licp));
1189 xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
1190 next_licp = licp->lic_next;
1191 kmem_free(licp, sizeof(xfs_log_item_chunk_t));
1196 * Clear all the per-AG busy list items listed in this transaction
1199 while (lbcp != NULL) {
1200 for (i = 0, lbsp = lbcp->lbc_busy; i < lbcp->lbc_unused; i++, lbsp++) {
1201 if (!XFS_LBC_ISFREE(lbcp, i)) {
1202 xfs_alloc_clear_busy(tp, lbsp->lbc_ag,
1206 lbcp = lbcp->lbc_next;
1208 xfs_trans_free_busy(tp);
1211 * That's it for the transaction structure. Free it.
1217 * This is called to perform the commit processing for each
1218 * item described by the given chunk.
1220 * The commit processing consists of unlocking items which were
1221 * held locked with the SYNC_UNLOCK attribute, calling the committed
1222 * routine of each logged item, updating the item's position in the AIL
1223 * if necessary, and unpinning each item. If the committed routine
1224 * returns -1, then do nothing further with the item because it
1225 * may have been freed.
1227 * Since items are unlocked when they are copied to the incore
1228 * log, it is possible for two transactions to be completing
1229 * and manipulating the same item simultaneously. The AIL lock
1230 * will protect the lsn field of each item. The value of this
1231 * field can never go backwards.
1233 * We unpin the items after repositioning them in the AIL, because
1234 * otherwise they could be immediately flushed and we'd have to race
1235 * with the flusher trying to pull the item from the AIL as we add it.
1238 xfs_trans_chunk_committed(
1239 xfs_log_item_chunk_t *licp,
1243 xfs_log_item_desc_t *lidp;
1244 xfs_log_item_t *lip;
1246 struct xfs_mount *mp;
1250 lidp = licp->lic_descs;
1251 for (i = 0; i < licp->lic_unused; i++, lidp++) {
1252 if (XFS_LIC_ISFREE(licp, i)) {
1256 lip = lidp->lid_item;
1258 lip->li_flags |= XFS_LI_ABORTED;
1261 * Send in the ABORTED flag to the COMMITTED routine
1262 * so that it knows whether the transaction was aborted
1265 item_lsn = IOP_COMMITTED(lip, lsn);
1268 * If the committed routine returns -1, make
1269 * no more references to the item.
1271 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) {
1276 * If the returned lsn is greater than what it
1277 * contained before, update the location of the
1278 * item in the AIL. If it is not, then do nothing.
1279 * Items can never move backwards in the AIL.
1281 * While the new lsn should usually be greater, it
1282 * is possible that a later transaction completing
1283 * simultaneously with an earlier one using the
1284 * same item could complete first with a higher lsn.
1285 * This would cause the earlier transaction to fail
1288 mp = lip->li_mountp;
1290 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) {
1292 * This will set the item's lsn to item_lsn
1293 * and update the position of the item in
1296 * xfs_trans_update_ail() drops the AIL lock.
1298 xfs_trans_update_ail(mp, lip, item_lsn, s);
1304 * Now that we've repositioned the item in the AIL,
1305 * unpin it so it can be flushed. Pass information
1306 * about buffer stale state down from the log item
1307 * flags, if anyone else stales the buffer we do not
1308 * want to pay any attention to it.
1310 IOP_UNPIN(lip, lidp->lid_flags & XFS_LID_BUF_STALE);