2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_error.h"
32 #include "xfs_da_btree.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir_sf.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_btree.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_alloc.h"
45 #include "xfs_quota.h"
46 #include "xfs_trans_priv.h"
47 #include "xfs_trans_space.h"
50 STATIC void xfs_trans_apply_sb_deltas(xfs_trans_t *);
51 STATIC uint xfs_trans_count_vecs(xfs_trans_t *);
52 STATIC void xfs_trans_fill_vecs(xfs_trans_t *, xfs_log_iovec_t *);
53 STATIC void xfs_trans_uncommit(xfs_trans_t *, uint);
54 STATIC void xfs_trans_committed(xfs_trans_t *, int);
55 STATIC void xfs_trans_chunk_committed(xfs_log_item_chunk_t *, xfs_lsn_t, int);
56 STATIC void xfs_trans_free(xfs_trans_t *);
58 kmem_zone_t *xfs_trans_zone;
62 * Reservation functions here avoid a huge stack in xfs_trans_init
63 * due to register overflow from temporaries in the calculations.
67 xfs_calc_write_reservation(xfs_mount_t *mp)
69 return XFS_CALC_WRITE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
73 xfs_calc_itruncate_reservation(xfs_mount_t *mp)
75 return XFS_CALC_ITRUNCATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
79 xfs_calc_rename_reservation(xfs_mount_t *mp)
81 return XFS_CALC_RENAME_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
85 xfs_calc_link_reservation(xfs_mount_t *mp)
87 return XFS_CALC_LINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
91 xfs_calc_remove_reservation(xfs_mount_t *mp)
93 return XFS_CALC_REMOVE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
97 xfs_calc_symlink_reservation(xfs_mount_t *mp)
99 return XFS_CALC_SYMLINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
103 xfs_calc_create_reservation(xfs_mount_t *mp)
105 return XFS_CALC_CREATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
109 xfs_calc_mkdir_reservation(xfs_mount_t *mp)
111 return XFS_CALC_MKDIR_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
115 xfs_calc_ifree_reservation(xfs_mount_t *mp)
117 return XFS_CALC_IFREE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
121 xfs_calc_ichange_reservation(xfs_mount_t *mp)
123 return XFS_CALC_ICHANGE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
127 xfs_calc_growdata_reservation(xfs_mount_t *mp)
129 return XFS_CALC_GROWDATA_LOG_RES(mp);
133 xfs_calc_growrtalloc_reservation(xfs_mount_t *mp)
135 return XFS_CALC_GROWRTALLOC_LOG_RES(mp);
139 xfs_calc_growrtzero_reservation(xfs_mount_t *mp)
141 return XFS_CALC_GROWRTZERO_LOG_RES(mp);
145 xfs_calc_growrtfree_reservation(xfs_mount_t *mp)
147 return XFS_CALC_GROWRTFREE_LOG_RES(mp);
151 xfs_calc_swrite_reservation(xfs_mount_t *mp)
153 return XFS_CALC_SWRITE_LOG_RES(mp);
157 xfs_calc_writeid_reservation(xfs_mount_t *mp)
159 return XFS_CALC_WRITEID_LOG_RES(mp);
163 xfs_calc_addafork_reservation(xfs_mount_t *mp)
165 return XFS_CALC_ADDAFORK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
169 xfs_calc_attrinval_reservation(xfs_mount_t *mp)
171 return XFS_CALC_ATTRINVAL_LOG_RES(mp);
175 xfs_calc_attrset_reservation(xfs_mount_t *mp)
177 return XFS_CALC_ATTRSET_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
181 xfs_calc_attrrm_reservation(xfs_mount_t *mp)
183 return XFS_CALC_ATTRRM_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
187 xfs_calc_clear_agi_bucket_reservation(xfs_mount_t *mp)
189 return XFS_CALC_CLEAR_AGI_BUCKET_LOG_RES(mp);
193 * Initialize the precomputed transaction reservation values
194 * in the mount structure.
200 xfs_trans_reservations_t *resp;
202 resp = &(mp->m_reservations);
203 resp->tr_write = xfs_calc_write_reservation(mp);
204 resp->tr_itruncate = xfs_calc_itruncate_reservation(mp);
205 resp->tr_rename = xfs_calc_rename_reservation(mp);
206 resp->tr_link = xfs_calc_link_reservation(mp);
207 resp->tr_remove = xfs_calc_remove_reservation(mp);
208 resp->tr_symlink = xfs_calc_symlink_reservation(mp);
209 resp->tr_create = xfs_calc_create_reservation(mp);
210 resp->tr_mkdir = xfs_calc_mkdir_reservation(mp);
211 resp->tr_ifree = xfs_calc_ifree_reservation(mp);
212 resp->tr_ichange = xfs_calc_ichange_reservation(mp);
213 resp->tr_growdata = xfs_calc_growdata_reservation(mp);
214 resp->tr_swrite = xfs_calc_swrite_reservation(mp);
215 resp->tr_writeid = xfs_calc_writeid_reservation(mp);
216 resp->tr_addafork = xfs_calc_addafork_reservation(mp);
217 resp->tr_attrinval = xfs_calc_attrinval_reservation(mp);
218 resp->tr_attrset = xfs_calc_attrset_reservation(mp);
219 resp->tr_attrrm = xfs_calc_attrrm_reservation(mp);
220 resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp);
221 resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp);
222 resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp);
223 resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp);
227 * This routine is called to allocate a transaction structure.
228 * The type parameter indicates the type of the transaction. These
229 * are enumerated in xfs_trans.h.
231 * Dynamically allocate the transaction structure from the transaction
232 * zone, initialize it, and return it to the caller.
239 fs_check_frozen(XFS_MTOVFS(mp), SB_FREEZE_TRANS);
240 atomic_inc(&mp->m_active_trans);
242 return (_xfs_trans_alloc(mp, type));
253 ASSERT(xfs_trans_zone != NULL);
254 tp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
257 * Initialize the transaction structure.
259 tp->t_magic = XFS_TRANS_MAGIC;
262 tp->t_items_free = XFS_LIC_NUM_SLOTS;
263 tp->t_busy_free = XFS_LBC_NUM_SLOTS;
264 XFS_LIC_INIT(&(tp->t_items));
265 XFS_LBC_INIT(&(tp->t_busy));
271 * This is called to create a new transaction which will share the
272 * permanent log reservation of the given transaction. The remaining
273 * unused block and rt extent reservations are also inherited. This
274 * implies that the original transaction is no longer allowed to allocate
275 * blocks. Locks and log items, however, are no inherited. They must
276 * be added to the new transaction explicitly.
284 ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
287 * Initialize the new transaction structure.
289 ntp->t_magic = XFS_TRANS_MAGIC;
290 ntp->t_type = tp->t_type;
291 ntp->t_mountp = tp->t_mountp;
292 ntp->t_items_free = XFS_LIC_NUM_SLOTS;
293 ntp->t_busy_free = XFS_LBC_NUM_SLOTS;
294 XFS_LIC_INIT(&(ntp->t_items));
295 XFS_LBC_INIT(&(ntp->t_busy));
297 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
298 ASSERT(tp->t_ticket != NULL);
300 ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE);
301 ntp->t_ticket = tp->t_ticket;
302 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
303 tp->t_blk_res = tp->t_blk_res_used;
304 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
305 tp->t_rtx_res = tp->t_rtx_res_used;
306 ntp->t_pflags = tp->t_pflags;
308 XFS_TRANS_DUP_DQINFO(tp->t_mountp, tp, ntp);
310 atomic_inc(&tp->t_mountp->m_active_trans);
315 * This is called to reserve free disk blocks and log space for the
316 * given transaction. This must be done before allocating any resources
317 * within the transaction.
319 * This will return ENOSPC if there are not enough blocks available.
320 * It will sleep waiting for available log space.
321 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
322 * is used by long running transactions. If any one of the reservations
323 * fails then they will all be backed out.
325 * This does not do quota reservations. That typically is done by the
339 int rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
341 /* Mark this thread as being in a transaction */
342 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
345 * Attempt to reserve the needed disk blocks by decrementing
346 * the number needed from the number available. This will
347 * fail if the count would go below zero.
350 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
353 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
354 return (XFS_ERROR(ENOSPC));
356 tp->t_blk_res += blocks;
360 * Reserve the log space needed for this transaction.
363 ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace));
364 ASSERT((tp->t_log_count == 0) ||
365 (tp->t_log_count == logcount));
366 if (flags & XFS_TRANS_PERM_LOG_RES) {
367 log_flags = XFS_LOG_PERM_RESERV;
368 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
370 ASSERT(tp->t_ticket == NULL);
371 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
375 error = xfs_log_reserve(tp->t_mountp, logspace, logcount,
377 XFS_TRANSACTION, log_flags, tp->t_type);
381 tp->t_log_res = logspace;
382 tp->t_log_count = logcount;
386 * Attempt to reserve the needed realtime extents by decrementing
387 * the number needed from the number available. This will
388 * fail if the count would go below zero.
391 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
394 error = XFS_ERROR(ENOSPC);
397 tp->t_rtx_res += rtextents;
403 * Error cases jump to one of these labels to undo any
404 * reservations which have already been performed.
408 if (flags & XFS_TRANS_PERM_LOG_RES) {
409 log_flags = XFS_LOG_REL_PERM_RESERV;
413 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
416 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
421 (void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
426 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
433 * Record the indicated change to the given field for application
434 * to the file system's superblock when the transaction commits.
435 * For now, just store the change in the transaction structure.
437 * Mark the transaction structure to indicate that the superblock
438 * needs to be updated before committing.
448 case XFS_TRANS_SB_ICOUNT:
449 tp->t_icount_delta += delta;
451 case XFS_TRANS_SB_IFREE:
452 tp->t_ifree_delta += delta;
454 case XFS_TRANS_SB_FDBLOCKS:
456 * Track the number of blocks allocated in the
457 * transaction. Make sure it does not exceed the
461 tp->t_blk_res_used += (uint)-delta;
462 ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
464 tp->t_fdblocks_delta += delta;
466 case XFS_TRANS_SB_RES_FDBLOCKS:
468 * The allocation has already been applied to the
469 * in-core superblock's counter. This should only
470 * be applied to the on-disk superblock.
473 tp->t_res_fdblocks_delta += delta;
475 case XFS_TRANS_SB_FREXTENTS:
477 * Track the number of blocks allocated in the
478 * transaction. Make sure it does not exceed the
482 tp->t_rtx_res_used += (uint)-delta;
483 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
485 tp->t_frextents_delta += delta;
487 case XFS_TRANS_SB_RES_FREXTENTS:
489 * The allocation has already been applied to the
490 * in-core superblock's counter. This should only
491 * be applied to the on-disk superblock.
494 tp->t_res_frextents_delta += delta;
496 case XFS_TRANS_SB_DBLOCKS:
498 tp->t_dblocks_delta += delta;
500 case XFS_TRANS_SB_AGCOUNT:
502 tp->t_agcount_delta += delta;
504 case XFS_TRANS_SB_IMAXPCT:
505 tp->t_imaxpct_delta += delta;
507 case XFS_TRANS_SB_REXTSIZE:
508 tp->t_rextsize_delta += delta;
510 case XFS_TRANS_SB_RBMBLOCKS:
511 tp->t_rbmblocks_delta += delta;
513 case XFS_TRANS_SB_RBLOCKS:
514 tp->t_rblocks_delta += delta;
516 case XFS_TRANS_SB_REXTENTS:
517 tp->t_rextents_delta += delta;
519 case XFS_TRANS_SB_REXTSLOG:
520 tp->t_rextslog_delta += delta;
527 tp->t_flags |= (XFS_TRANS_SB_DIRTY | XFS_TRANS_DIRTY);
531 * xfs_trans_apply_sb_deltas() is called from the commit code
532 * to bring the superblock buffer into the current transaction
533 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
535 * For now we just look at each field allowed to change and change
539 xfs_trans_apply_sb_deltas(
546 bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
547 sbp = XFS_BUF_TO_SBP(bp);
550 * Check that superblock mods match the mods made to AGF counters.
552 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
553 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
554 tp->t_ag_btree_delta));
556 if (tp->t_icount_delta != 0) {
557 INT_MOD(sbp->sb_icount, ARCH_CONVERT, tp->t_icount_delta);
559 if (tp->t_ifree_delta != 0) {
560 INT_MOD(sbp->sb_ifree, ARCH_CONVERT, tp->t_ifree_delta);
563 if (tp->t_fdblocks_delta != 0) {
564 INT_MOD(sbp->sb_fdblocks, ARCH_CONVERT, tp->t_fdblocks_delta);
566 if (tp->t_res_fdblocks_delta != 0) {
567 INT_MOD(sbp->sb_fdblocks, ARCH_CONVERT, tp->t_res_fdblocks_delta);
570 if (tp->t_frextents_delta != 0) {
571 INT_MOD(sbp->sb_frextents, ARCH_CONVERT, tp->t_frextents_delta);
573 if (tp->t_res_frextents_delta != 0) {
574 INT_MOD(sbp->sb_frextents, ARCH_CONVERT, tp->t_res_frextents_delta);
576 if (tp->t_dblocks_delta != 0) {
577 INT_MOD(sbp->sb_dblocks, ARCH_CONVERT, tp->t_dblocks_delta);
580 if (tp->t_agcount_delta != 0) {
581 INT_MOD(sbp->sb_agcount, ARCH_CONVERT, tp->t_agcount_delta);
584 if (tp->t_imaxpct_delta != 0) {
585 INT_MOD(sbp->sb_imax_pct, ARCH_CONVERT, tp->t_imaxpct_delta);
588 if (tp->t_rextsize_delta != 0) {
589 INT_MOD(sbp->sb_rextsize, ARCH_CONVERT, tp->t_rextsize_delta);
592 if (tp->t_rbmblocks_delta != 0) {
593 INT_MOD(sbp->sb_rbmblocks, ARCH_CONVERT, tp->t_rbmblocks_delta);
596 if (tp->t_rblocks_delta != 0) {
597 INT_MOD(sbp->sb_rblocks, ARCH_CONVERT, tp->t_rblocks_delta);
600 if (tp->t_rextents_delta != 0) {
601 INT_MOD(sbp->sb_rextents, ARCH_CONVERT, tp->t_rextents_delta);
604 if (tp->t_rextslog_delta != 0) {
605 INT_MOD(sbp->sb_rextslog, ARCH_CONVERT, tp->t_rextslog_delta);
611 * Log the whole thing, the fields are noncontiguous.
613 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_sb_t) - 1);
616 * Since all the modifiable fields are contiguous, we
617 * can get away with this.
619 xfs_trans_log_buf(tp, bp, offsetof(xfs_sb_t, sb_icount),
620 offsetof(xfs_sb_t, sb_frextents) +
621 sizeof(sbp->sb_frextents) - 1);
623 XFS_MTOVFS(tp->t_mountp)->vfs_super->s_dirt = 1;
627 * xfs_trans_unreserve_and_mod_sb() is called to release unused
628 * reservations and apply superblock counter changes to the in-core
631 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
634 xfs_trans_unreserve_and_mod_sb(
637 xfs_mod_sb_t msb[14]; /* If you add cases, add entries */
644 rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
647 * Release any reserved blocks. Any that were allocated
648 * will be taken back again by fdblocks_delta below.
650 if (tp->t_blk_res > 0) {
651 msbp->msb_field = XFS_SBS_FDBLOCKS;
652 msbp->msb_delta = tp->t_blk_res;
657 * Release any reserved real time extents . Any that were
658 * allocated will be taken back again by frextents_delta below.
660 if (tp->t_rtx_res > 0) {
661 msbp->msb_field = XFS_SBS_FREXTENTS;
662 msbp->msb_delta = tp->t_rtx_res;
667 * Apply any superblock modifications to the in-core version.
668 * The t_res_fdblocks_delta and t_res_frextents_delta fields are
669 * explicitly NOT applied to the in-core superblock.
670 * The idea is that that has already been done.
672 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
673 if (tp->t_icount_delta != 0) {
674 msbp->msb_field = XFS_SBS_ICOUNT;
675 msbp->msb_delta = (int)tp->t_icount_delta;
678 if (tp->t_ifree_delta != 0) {
679 msbp->msb_field = XFS_SBS_IFREE;
680 msbp->msb_delta = (int)tp->t_ifree_delta;
683 if (tp->t_fdblocks_delta != 0) {
684 msbp->msb_field = XFS_SBS_FDBLOCKS;
685 msbp->msb_delta = (int)tp->t_fdblocks_delta;
688 if (tp->t_frextents_delta != 0) {
689 msbp->msb_field = XFS_SBS_FREXTENTS;
690 msbp->msb_delta = (int)tp->t_frextents_delta;
693 if (tp->t_dblocks_delta != 0) {
694 msbp->msb_field = XFS_SBS_DBLOCKS;
695 msbp->msb_delta = (int)tp->t_dblocks_delta;
698 if (tp->t_agcount_delta != 0) {
699 msbp->msb_field = XFS_SBS_AGCOUNT;
700 msbp->msb_delta = (int)tp->t_agcount_delta;
703 if (tp->t_imaxpct_delta != 0) {
704 msbp->msb_field = XFS_SBS_IMAX_PCT;
705 msbp->msb_delta = (int)tp->t_imaxpct_delta;
708 if (tp->t_rextsize_delta != 0) {
709 msbp->msb_field = XFS_SBS_REXTSIZE;
710 msbp->msb_delta = (int)tp->t_rextsize_delta;
713 if (tp->t_rbmblocks_delta != 0) {
714 msbp->msb_field = XFS_SBS_RBMBLOCKS;
715 msbp->msb_delta = (int)tp->t_rbmblocks_delta;
718 if (tp->t_rblocks_delta != 0) {
719 msbp->msb_field = XFS_SBS_RBLOCKS;
720 msbp->msb_delta = (int)tp->t_rblocks_delta;
723 if (tp->t_rextents_delta != 0) {
724 msbp->msb_field = XFS_SBS_REXTENTS;
725 msbp->msb_delta = (int)tp->t_rextents_delta;
728 if (tp->t_rextslog_delta != 0) {
729 msbp->msb_field = XFS_SBS_REXTSLOG;
730 msbp->msb_delta = (int)tp->t_rextslog_delta;
736 * If we need to change anything, do it.
739 error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
740 (uint)(msbp - msb), rsvd);
749 * Commit the given transaction to the log a/synchronously.
751 * XFS disk error handling mechanism is not based on a typical
752 * transaction abort mechanism. Logically after the filesystem
753 * gets marked 'SHUTDOWN', we can't let any new transactions
754 * be durable - ie. committed to disk - because some metadata might
755 * be inconsistent. In such cases, this returns an error, and the
756 * caller may assume that all locked objects joined to the transaction
757 * have already been unlocked as if the commit had succeeded.
758 * Do not reference the transaction structure after this call.
765 xfs_lsn_t *commit_lsn_p,
768 xfs_log_iovec_t *log_vector;
771 xfs_lsn_t commit_lsn;
776 #define XFS_TRANS_LOGVEC_COUNT 16
777 xfs_log_iovec_t log_vector_fast[XFS_TRANS_LOGVEC_COUNT];
784 * Determine whether this commit is releasing a permanent
785 * log reservation or not.
787 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
788 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
789 log_flags = XFS_LOG_REL_PERM_RESERV;
796 * If there is nothing to be logged by the transaction,
797 * then unlock all of the items associated with the
798 * transaction and free the transaction structure.
799 * Also make sure to return any reserved blocks to
803 shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0;
804 if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) {
805 xfs_trans_unreserve_and_mod_sb(tp);
807 * It is indeed possible for the transaction to be
808 * not dirty but the dqinfo portion to be. All that
809 * means is that we have some (non-persistent) quota
810 * reservations that need to be unreserved.
812 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(mp, tp);
814 commit_lsn = xfs_log_done(mp, tp->t_ticket,
816 if (commit_lsn == -1 && !shutdown)
817 shutdown = XFS_ERROR(EIO);
819 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
820 xfs_trans_free_items(tp, shutdown? XFS_TRANS_ABORT : 0);
821 xfs_trans_free_busy(tp);
823 XFS_STATS_INC(xs_trans_empty);
825 *commit_lsn_p = commit_lsn;
828 ASSERT(tp->t_ticket != NULL);
831 * If we need to update the superblock, then do it now.
833 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
834 xfs_trans_apply_sb_deltas(tp);
836 XFS_TRANS_APPLY_DQUOT_DELTAS(mp, tp);
839 * Ask each log item how many log_vector entries it will
840 * need so we can figure out how many to allocate.
841 * Try to avoid the kmem_alloc() call in the common case
842 * by using a vector from the stack when it fits.
844 nvec = xfs_trans_count_vecs(tp);
846 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
848 } else if (nvec <= XFS_TRANS_LOGVEC_COUNT) {
849 log_vector = log_vector_fast;
851 log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec *
852 sizeof(xfs_log_iovec_t),
857 * Fill in the log_vector and pin the logged items, and
858 * then write the transaction to the log.
860 xfs_trans_fill_vecs(tp, log_vector);
862 error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn));
865 * The transaction is committed incore here, and can go out to disk
866 * at any time after this call. However, all the items associated
867 * with the transaction are still locked and pinned in memory.
869 commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);
871 tp->t_commit_lsn = commit_lsn;
872 if (nvec > XFS_TRANS_LOGVEC_COUNT) {
873 kmem_free(log_vector, nvec * sizeof(xfs_log_iovec_t));
877 *commit_lsn_p = commit_lsn;
880 * If we got a log write error. Unpin the logitems that we
881 * had pinned, clean up, free trans structure, and return error.
883 if (error || commit_lsn == -1) {
884 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
885 xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT);
886 return XFS_ERROR(EIO);
890 * Once the transaction has committed, unused
891 * reservations need to be released and changes to
892 * the superblock need to be reflected in the in-core
893 * version. Do that now.
895 xfs_trans_unreserve_and_mod_sb(tp);
897 sync = tp->t_flags & XFS_TRANS_SYNC;
900 * Tell the LM to call the transaction completion routine
901 * when the log write with LSN commit_lsn completes (e.g.
902 * when the transaction commit really hits the on-disk log).
903 * After this call we cannot reference tp, because the call
904 * can happen at any time and the call will free the transaction
905 * structure pointed to by tp. The only case where we call
906 * the completion routine (xfs_trans_committed) directly is
907 * if the log is turned off on a debug kernel or we're
908 * running in simulation mode (the log is explicitly turned
911 tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed;
912 tp->t_logcb.cb_arg = tp;
915 * We need to pass the iclog buffer which was used for the
916 * transaction commit record into this function, and attach
917 * the callback to it. The callback must be attached before
918 * the items are unlocked to avoid racing with other threads
919 * waiting for an item to unlock.
921 shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb));
924 * Mark this thread as no longer being in a transaction
926 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
929 * Once all the items of the transaction have been copied
930 * to the in core log and the callback is attached, the
931 * items can be unlocked.
933 * This will free descriptors pointing to items which were
934 * not logged since there is nothing more to do with them.
935 * For items which were logged, we will keep pointers to them
936 * so they can be unpinned after the transaction commits to disk.
937 * This will also stamp each modified meta-data item with
938 * the commit lsn of this transaction for dependency tracking
941 xfs_trans_unlock_items(tp, commit_lsn);
944 * If we detected a log error earlier, finish committing
945 * the transaction now (unpin log items, etc).
947 * Order is critical here, to avoid using the transaction
948 * pointer after its been freed (by xfs_trans_committed
949 * either here now, or as a callback). We cannot do this
950 * step inside xfs_log_notify as was done earlier because
954 xfs_trans_committed(tp, XFS_LI_ABORTED);
957 * Now that the xfs_trans_committed callback has been attached,
958 * and the items are released we can finally allow the iclog to
961 error = xfs_log_release_iclog(mp, commit_iclog);
964 * If the transaction needs to be synchronous, then force the
965 * log out now and wait for it.
969 error = _xfs_log_force(mp, commit_lsn,
970 XFS_LOG_FORCE | XFS_LOG_SYNC,
973 XFS_STATS_INC(xs_trans_sync);
975 XFS_STATS_INC(xs_trans_async);
983 * Total up the number of log iovecs needed to commit this
984 * transaction. The transaction itself needs one for the
985 * transaction header. Ask each dirty item in turn how many
986 * it needs to get the total.
989 xfs_trans_count_vecs(
993 xfs_log_item_desc_t *lidp;
996 lidp = xfs_trans_first_item(tp);
997 ASSERT(lidp != NULL);
999 /* In the non-debug case we need to start bailing out if we
1000 * didn't find a log_item here, return zero and let trans_commit
1006 while (lidp != NULL) {
1008 * Skip items which aren't dirty in this transaction.
1010 if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
1011 lidp = xfs_trans_next_item(tp, lidp);
1014 lidp->lid_size = IOP_SIZE(lidp->lid_item);
1015 nvecs += lidp->lid_size;
1016 lidp = xfs_trans_next_item(tp, lidp);
1023 * Called from the trans_commit code when we notice that
1024 * the filesystem is in the middle of a forced shutdown.
1031 xfs_log_item_desc_t *lidp;
1033 for (lidp = xfs_trans_first_item(tp);
1035 lidp = xfs_trans_next_item(tp, lidp)) {
1037 * Unpin all but those that aren't dirty.
1039 if (lidp->lid_flags & XFS_LID_DIRTY)
1040 IOP_UNPIN_REMOVE(lidp->lid_item, tp);
1043 xfs_trans_unreserve_and_mod_sb(tp);
1044 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(tp->t_mountp, tp);
1046 xfs_trans_free_items(tp, flags);
1047 xfs_trans_free_busy(tp);
1052 * Fill in the vector with pointers to data to be logged
1053 * by this transaction. The transaction header takes
1054 * the first vector, and then each dirty item takes the
1055 * number of vectors it indicated it needed in xfs_trans_count_vecs().
1057 * As each item fills in the entries it needs, also pin the item
1058 * so that it cannot be flushed out until the log write completes.
1061 xfs_trans_fill_vecs(
1063 xfs_log_iovec_t *log_vector)
1065 xfs_log_item_desc_t *lidp;
1066 xfs_log_iovec_t *vecp;
1070 * Skip over the entry for the transaction header, we'll
1071 * fill that in at the end.
1073 vecp = log_vector + 1; /* pointer arithmetic */
1076 lidp = xfs_trans_first_item(tp);
1077 ASSERT(lidp != NULL);
1078 while (lidp != NULL) {
1080 * Skip items which aren't dirty in this transaction.
1082 if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
1083 lidp = xfs_trans_next_item(tp, lidp);
1087 * The item may be marked dirty but not log anything.
1088 * This can be used to get called when a transaction
1091 if (lidp->lid_size) {
1094 IOP_FORMAT(lidp->lid_item, vecp);
1095 vecp += lidp->lid_size; /* pointer arithmetic */
1096 IOP_PIN(lidp->lid_item);
1097 lidp = xfs_trans_next_item(tp, lidp);
1101 * Now that we've counted the number of items in this
1102 * transaction, fill in the transaction header.
1104 tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC;
1105 tp->t_header.th_type = tp->t_type;
1106 tp->t_header.th_num_items = nitems;
1107 log_vector->i_addr = (xfs_caddr_t)&tp->t_header;
1108 log_vector->i_len = sizeof(xfs_trans_header_t);
1109 XLOG_VEC_SET_TYPE(log_vector, XLOG_REG_TYPE_TRANSHDR);
1114 * Unlock all of the transaction's items and free the transaction.
1115 * The transaction must not have modified any of its items, because
1116 * there is no way to restore them to their previous state.
1118 * If the transaction has made a log reservation, make sure to release
1128 xfs_log_item_chunk_t *licp;
1129 xfs_log_item_desc_t *lidp;
1130 xfs_log_item_t *lip;
1133 xfs_mount_t *mp = tp->t_mountp;
1136 * See if the caller is being too lazy to figure out if
1137 * the transaction really needs an abort.
1139 if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1140 flags &= ~XFS_TRANS_ABORT;
1142 * See if the caller is relying on us to shut down the
1143 * filesystem. This happens in paths where we detect
1144 * corruption and decide to give up.
1146 if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
1147 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1148 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1151 if (!(flags & XFS_TRANS_ABORT)) {
1152 licp = &(tp->t_items);
1153 while (licp != NULL) {
1154 lidp = licp->lic_descs;
1155 for (i = 0; i < licp->lic_unused; i++, lidp++) {
1156 if (XFS_LIC_ISFREE(licp, i)) {
1160 lip = lidp->lid_item;
1161 if (!XFS_FORCED_SHUTDOWN(mp))
1162 ASSERT(!(lip->li_type == XFS_LI_EFD));
1164 licp = licp->lic_next;
1168 xfs_trans_unreserve_and_mod_sb(tp);
1169 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(mp, tp);
1172 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1173 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1174 log_flags = XFS_LOG_REL_PERM_RESERV;
1178 xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1181 /* mark this thread as no longer being in a transaction */
1182 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1184 xfs_trans_free_items(tp, flags);
1185 xfs_trans_free_busy(tp);
1191 * Free the transaction structure. If there is more clean up
1192 * to do when the structure is freed, add it here.
1198 atomic_dec(&tp->t_mountp->m_active_trans);
1199 XFS_TRANS_FREE_DQINFO(tp->t_mountp, tp);
1200 kmem_zone_free(xfs_trans_zone, tp);
1205 * THIS SHOULD BE REWRITTEN TO USE xfs_trans_next_item().
1207 * This is typically called by the LM when a transaction has been fully
1208 * committed to disk. It needs to unpin the items which have
1209 * been logged by the transaction and update their positions
1210 * in the AIL if necessary.
1211 * This also gets called when the transactions didn't get written out
1212 * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then.
1214 * Call xfs_trans_chunk_committed() to process the items in
1218 xfs_trans_committed(
1222 xfs_log_item_chunk_t *licp;
1223 xfs_log_item_chunk_t *next_licp;
1224 xfs_log_busy_chunk_t *lbcp;
1225 xfs_log_busy_slot_t *lbsp;
1229 * Call the transaction's completion callback if there
1232 if (tp->t_callback != NULL) {
1233 tp->t_callback(tp, tp->t_callarg);
1237 * Special case the chunk embedded in the transaction.
1239 licp = &(tp->t_items);
1240 if (!(XFS_LIC_ARE_ALL_FREE(licp))) {
1241 xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
1245 * Process the items in each chunk in turn.
1247 licp = licp->lic_next;
1248 while (licp != NULL) {
1249 ASSERT(!XFS_LIC_ARE_ALL_FREE(licp));
1250 xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
1251 next_licp = licp->lic_next;
1252 kmem_free(licp, sizeof(xfs_log_item_chunk_t));
1257 * Clear all the per-AG busy list items listed in this transaction
1260 while (lbcp != NULL) {
1261 for (i = 0, lbsp = lbcp->lbc_busy; i < lbcp->lbc_unused; i++, lbsp++) {
1262 if (!XFS_LBC_ISFREE(lbcp, i)) {
1263 xfs_alloc_clear_busy(tp, lbsp->lbc_ag,
1267 lbcp = lbcp->lbc_next;
1269 xfs_trans_free_busy(tp);
1272 * That's it for the transaction structure. Free it.
1278 * This is called to perform the commit processing for each
1279 * item described by the given chunk.
1281 * The commit processing consists of unlocking items which were
1282 * held locked with the SYNC_UNLOCK attribute, calling the committed
1283 * routine of each logged item, updating the item's position in the AIL
1284 * if necessary, and unpinning each item. If the committed routine
1285 * returns -1, then do nothing further with the item because it
1286 * may have been freed.
1288 * Since items are unlocked when they are copied to the incore
1289 * log, it is possible for two transactions to be completing
1290 * and manipulating the same item simultaneously. The AIL lock
1291 * will protect the lsn field of each item. The value of this
1292 * field can never go backwards.
1294 * We unpin the items after repositioning them in the AIL, because
1295 * otherwise they could be immediately flushed and we'd have to race
1296 * with the flusher trying to pull the item from the AIL as we add it.
1299 xfs_trans_chunk_committed(
1300 xfs_log_item_chunk_t *licp,
1304 xfs_log_item_desc_t *lidp;
1305 xfs_log_item_t *lip;
1307 struct xfs_mount *mp;
1311 lidp = licp->lic_descs;
1312 for (i = 0; i < licp->lic_unused; i++, lidp++) {
1313 if (XFS_LIC_ISFREE(licp, i)) {
1317 lip = lidp->lid_item;
1319 lip->li_flags |= XFS_LI_ABORTED;
1322 * Send in the ABORTED flag to the COMMITTED routine
1323 * so that it knows whether the transaction was aborted
1326 item_lsn = IOP_COMMITTED(lip, lsn);
1329 * If the committed routine returns -1, make
1330 * no more references to the item.
1332 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) {
1337 * If the returned lsn is greater than what it
1338 * contained before, update the location of the
1339 * item in the AIL. If it is not, then do nothing.
1340 * Items can never move backwards in the AIL.
1342 * While the new lsn should usually be greater, it
1343 * is possible that a later transaction completing
1344 * simultaneously with an earlier one using the
1345 * same item could complete first with a higher lsn.
1346 * This would cause the earlier transaction to fail
1349 mp = lip->li_mountp;
1351 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) {
1353 * This will set the item's lsn to item_lsn
1354 * and update the position of the item in
1357 * xfs_trans_update_ail() drops the AIL lock.
1359 xfs_trans_update_ail(mp, lip, item_lsn, s);
1365 * Now that we've repositioned the item in the AIL,
1366 * unpin it so it can be flushed. Pass information
1367 * about buffer stale state down from the log item
1368 * flags, if anyone else stales the buffer we do not
1369 * want to pay any attention to it.
1371 IOP_UNPIN(lip, lidp->lid_flags & XFS_LID_BUF_STALE);