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"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_btree.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_alloc.h"
43 #include "xfs_quota.h"
44 #include "xfs_trans_priv.h"
45 #include "xfs_trans_space.h"
46 #include "xfs_inode_item.h"
49 STATIC
void xfs_trans_apply_sb_deltas(xfs_trans_t
*);
50 STATIC uint
xfs_trans_count_vecs(xfs_trans_t
*);
51 STATIC
void xfs_trans_fill_vecs(xfs_trans_t
*, xfs_log_iovec_t
*);
52 STATIC
void xfs_trans_uncommit(xfs_trans_t
*, uint
);
53 STATIC
void xfs_trans_committed(xfs_trans_t
*, int);
54 STATIC
void xfs_trans_chunk_committed(xfs_log_item_chunk_t
*, xfs_lsn_t
, int);
55 STATIC
void xfs_trans_free(xfs_trans_t
*);
57 kmem_zone_t
*xfs_trans_zone
;
61 * Reservation functions here avoid a huge stack in xfs_trans_init
62 * due to register overflow from temporaries in the calculations.
66 xfs_calc_write_reservation(xfs_mount_t
*mp
)
68 return XFS_CALC_WRITE_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
72 xfs_calc_itruncate_reservation(xfs_mount_t
*mp
)
74 return XFS_CALC_ITRUNCATE_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
78 xfs_calc_rename_reservation(xfs_mount_t
*mp
)
80 return XFS_CALC_RENAME_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
84 xfs_calc_link_reservation(xfs_mount_t
*mp
)
86 return XFS_CALC_LINK_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
90 xfs_calc_remove_reservation(xfs_mount_t
*mp
)
92 return XFS_CALC_REMOVE_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
96 xfs_calc_symlink_reservation(xfs_mount_t
*mp
)
98 return XFS_CALC_SYMLINK_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
102 xfs_calc_create_reservation(xfs_mount_t
*mp
)
104 return XFS_CALC_CREATE_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
108 xfs_calc_mkdir_reservation(xfs_mount_t
*mp
)
110 return XFS_CALC_MKDIR_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
114 xfs_calc_ifree_reservation(xfs_mount_t
*mp
)
116 return XFS_CALC_IFREE_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
120 xfs_calc_ichange_reservation(xfs_mount_t
*mp
)
122 return XFS_CALC_ICHANGE_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
126 xfs_calc_growdata_reservation(xfs_mount_t
*mp
)
128 return XFS_CALC_GROWDATA_LOG_RES(mp
);
132 xfs_calc_growrtalloc_reservation(xfs_mount_t
*mp
)
134 return XFS_CALC_GROWRTALLOC_LOG_RES(mp
);
138 xfs_calc_growrtzero_reservation(xfs_mount_t
*mp
)
140 return XFS_CALC_GROWRTZERO_LOG_RES(mp
);
144 xfs_calc_growrtfree_reservation(xfs_mount_t
*mp
)
146 return XFS_CALC_GROWRTFREE_LOG_RES(mp
);
150 xfs_calc_swrite_reservation(xfs_mount_t
*mp
)
152 return XFS_CALC_SWRITE_LOG_RES(mp
);
156 xfs_calc_writeid_reservation(xfs_mount_t
*mp
)
158 return XFS_CALC_WRITEID_LOG_RES(mp
);
162 xfs_calc_addafork_reservation(xfs_mount_t
*mp
)
164 return XFS_CALC_ADDAFORK_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
168 xfs_calc_attrinval_reservation(xfs_mount_t
*mp
)
170 return XFS_CALC_ATTRINVAL_LOG_RES(mp
);
174 xfs_calc_attrset_reservation(xfs_mount_t
*mp
)
176 return XFS_CALC_ATTRSET_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
180 xfs_calc_attrrm_reservation(xfs_mount_t
*mp
)
182 return XFS_CALC_ATTRRM_LOG_RES(mp
) + XFS_DQUOT_LOGRES(mp
);
186 xfs_calc_clear_agi_bucket_reservation(xfs_mount_t
*mp
)
188 return XFS_CALC_CLEAR_AGI_BUCKET_LOG_RES(mp
);
192 * Initialize the precomputed transaction reservation values
193 * in the mount structure.
199 xfs_trans_reservations_t
*resp
;
201 resp
= &(mp
->m_reservations
);
202 resp
->tr_write
= xfs_calc_write_reservation(mp
);
203 resp
->tr_itruncate
= xfs_calc_itruncate_reservation(mp
);
204 resp
->tr_rename
= xfs_calc_rename_reservation(mp
);
205 resp
->tr_link
= xfs_calc_link_reservation(mp
);
206 resp
->tr_remove
= xfs_calc_remove_reservation(mp
);
207 resp
->tr_symlink
= xfs_calc_symlink_reservation(mp
);
208 resp
->tr_create
= xfs_calc_create_reservation(mp
);
209 resp
->tr_mkdir
= xfs_calc_mkdir_reservation(mp
);
210 resp
->tr_ifree
= xfs_calc_ifree_reservation(mp
);
211 resp
->tr_ichange
= xfs_calc_ichange_reservation(mp
);
212 resp
->tr_growdata
= xfs_calc_growdata_reservation(mp
);
213 resp
->tr_swrite
= xfs_calc_swrite_reservation(mp
);
214 resp
->tr_writeid
= xfs_calc_writeid_reservation(mp
);
215 resp
->tr_addafork
= xfs_calc_addafork_reservation(mp
);
216 resp
->tr_attrinval
= xfs_calc_attrinval_reservation(mp
);
217 resp
->tr_attrset
= xfs_calc_attrset_reservation(mp
);
218 resp
->tr_attrrm
= xfs_calc_attrrm_reservation(mp
);
219 resp
->tr_clearagi
= xfs_calc_clear_agi_bucket_reservation(mp
);
220 resp
->tr_growrtalloc
= xfs_calc_growrtalloc_reservation(mp
);
221 resp
->tr_growrtzero
= xfs_calc_growrtzero_reservation(mp
);
222 resp
->tr_growrtfree
= xfs_calc_growrtfree_reservation(mp
);
226 * This routine is called to allocate a transaction structure.
227 * The type parameter indicates the type of the transaction. These
228 * are enumerated in xfs_trans.h.
230 * Dynamically allocate the transaction structure from the transaction
231 * zone, initialize it, and return it to the caller.
238 xfs_wait_for_freeze(mp
, SB_FREEZE_TRANS
);
239 return _xfs_trans_alloc(mp
, type
, KM_SLEEP
);
250 atomic_inc(&mp
->m_active_trans
);
252 tp
= kmem_zone_zalloc(xfs_trans_zone
, memflags
);
253 tp
->t_magic
= XFS_TRANS_MAGIC
;
256 tp
->t_items_free
= XFS_LIC_NUM_SLOTS
;
257 tp
->t_busy_free
= XFS_LBC_NUM_SLOTS
;
258 xfs_lic_init(&(tp
->t_items
));
259 XFS_LBC_INIT(&(tp
->t_busy
));
264 * This is called to create a new transaction which will share the
265 * permanent log reservation of the given transaction. The remaining
266 * unused block and rt extent reservations are also inherited. This
267 * implies that the original transaction is no longer allowed to allocate
268 * blocks. Locks and log items, however, are no inherited. They must
269 * be added to the new transaction explicitly.
277 ntp
= kmem_zone_zalloc(xfs_trans_zone
, KM_SLEEP
);
280 * Initialize the new transaction structure.
282 ntp
->t_magic
= XFS_TRANS_MAGIC
;
283 ntp
->t_type
= tp
->t_type
;
284 ntp
->t_mountp
= tp
->t_mountp
;
285 ntp
->t_items_free
= XFS_LIC_NUM_SLOTS
;
286 ntp
->t_busy_free
= XFS_LBC_NUM_SLOTS
;
287 xfs_lic_init(&(ntp
->t_items
));
288 XFS_LBC_INIT(&(ntp
->t_busy
));
290 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
291 ASSERT(tp
->t_ticket
!= NULL
);
293 ntp
->t_flags
= XFS_TRANS_PERM_LOG_RES
| (tp
->t_flags
& XFS_TRANS_RESERVE
);
294 ntp
->t_ticket
= xfs_log_ticket_get(tp
->t_ticket
);
295 ntp
->t_blk_res
= tp
->t_blk_res
- tp
->t_blk_res_used
;
296 tp
->t_blk_res
= tp
->t_blk_res_used
;
297 ntp
->t_rtx_res
= tp
->t_rtx_res
- tp
->t_rtx_res_used
;
298 tp
->t_rtx_res
= tp
->t_rtx_res_used
;
299 ntp
->t_pflags
= tp
->t_pflags
;
301 xfs_trans_dup_dqinfo(tp
, ntp
);
303 atomic_inc(&tp
->t_mountp
->m_active_trans
);
308 * This is called to reserve free disk blocks and log space for the
309 * given transaction. This must be done before allocating any resources
310 * within the transaction.
312 * This will return ENOSPC if there are not enough blocks available.
313 * It will sleep waiting for available log space.
314 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
315 * is used by long running transactions. If any one of the reservations
316 * fails then they will all be backed out.
318 * This does not do quota reservations. That typically is done by the
332 int rsvd
= (tp
->t_flags
& XFS_TRANS_RESERVE
) != 0;
334 /* Mark this thread as being in a transaction */
335 current_set_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
338 * Attempt to reserve the needed disk blocks by decrementing
339 * the number needed from the number available. This will
340 * fail if the count would go below zero.
343 error
= xfs_mod_incore_sb(tp
->t_mountp
, XFS_SBS_FDBLOCKS
,
344 -((int64_t)blocks
), rsvd
);
346 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
347 return (XFS_ERROR(ENOSPC
));
349 tp
->t_blk_res
+= blocks
;
353 * Reserve the log space needed for this transaction.
356 ASSERT((tp
->t_log_res
== 0) || (tp
->t_log_res
== logspace
));
357 ASSERT((tp
->t_log_count
== 0) ||
358 (tp
->t_log_count
== logcount
));
359 if (flags
& XFS_TRANS_PERM_LOG_RES
) {
360 log_flags
= XFS_LOG_PERM_RESERV
;
361 tp
->t_flags
|= XFS_TRANS_PERM_LOG_RES
;
363 ASSERT(tp
->t_ticket
== NULL
);
364 ASSERT(!(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
));
368 error
= xfs_log_reserve(tp
->t_mountp
, logspace
, logcount
,
370 XFS_TRANSACTION
, log_flags
, tp
->t_type
);
374 tp
->t_log_res
= logspace
;
375 tp
->t_log_count
= logcount
;
379 * Attempt to reserve the needed realtime extents by decrementing
380 * the number needed from the number available. This will
381 * fail if the count would go below zero.
384 error
= xfs_mod_incore_sb(tp
->t_mountp
, XFS_SBS_FREXTENTS
,
385 -((int64_t)rtextents
), rsvd
);
387 error
= XFS_ERROR(ENOSPC
);
390 tp
->t_rtx_res
+= rtextents
;
396 * Error cases jump to one of these labels to undo any
397 * reservations which have already been performed.
401 if (flags
& XFS_TRANS_PERM_LOG_RES
) {
402 log_flags
= XFS_LOG_REL_PERM_RESERV
;
406 xfs_log_done(tp
->t_mountp
, tp
->t_ticket
, NULL
, log_flags
);
409 tp
->t_flags
&= ~XFS_TRANS_PERM_LOG_RES
;
414 (void) xfs_mod_incore_sb(tp
->t_mountp
, XFS_SBS_FDBLOCKS
,
415 (int64_t)blocks
, rsvd
);
419 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
426 * Record the indicated change to the given field for application
427 * to the file system's superblock when the transaction commits.
428 * For now, just store the change in the transaction structure.
430 * Mark the transaction structure to indicate that the superblock
431 * needs to be updated before committing.
433 * Because we may not be keeping track of allocated/free inodes and
434 * used filesystem blocks in the superblock, we do not mark the
435 * superblock dirty in this transaction if we modify these fields.
436 * We still need to update the transaction deltas so that they get
437 * applied to the incore superblock, but we don't want them to
438 * cause the superblock to get locked and logged if these are the
439 * only fields in the superblock that the transaction modifies.
447 uint32_t flags
= (XFS_TRANS_DIRTY
|XFS_TRANS_SB_DIRTY
);
448 xfs_mount_t
*mp
= tp
->t_mountp
;
451 case XFS_TRANS_SB_ICOUNT
:
452 tp
->t_icount_delta
+= delta
;
453 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
454 flags
&= ~XFS_TRANS_SB_DIRTY
;
456 case XFS_TRANS_SB_IFREE
:
457 tp
->t_ifree_delta
+= delta
;
458 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
459 flags
&= ~XFS_TRANS_SB_DIRTY
;
461 case XFS_TRANS_SB_FDBLOCKS
:
463 * Track the number of blocks allocated in the
464 * transaction. Make sure it does not exceed the
468 tp
->t_blk_res_used
+= (uint
)-delta
;
469 ASSERT(tp
->t_blk_res_used
<= tp
->t_blk_res
);
471 tp
->t_fdblocks_delta
+= delta
;
472 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
473 flags
&= ~XFS_TRANS_SB_DIRTY
;
475 case XFS_TRANS_SB_RES_FDBLOCKS
:
477 * The allocation has already been applied to the
478 * in-core superblock's counter. This should only
479 * be applied to the on-disk superblock.
482 tp
->t_res_fdblocks_delta
+= delta
;
483 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
484 flags
&= ~XFS_TRANS_SB_DIRTY
;
486 case XFS_TRANS_SB_FREXTENTS
:
488 * Track the number of blocks allocated in the
489 * transaction. Make sure it does not exceed the
493 tp
->t_rtx_res_used
+= (uint
)-delta
;
494 ASSERT(tp
->t_rtx_res_used
<= tp
->t_rtx_res
);
496 tp
->t_frextents_delta
+= delta
;
498 case XFS_TRANS_SB_RES_FREXTENTS
:
500 * The allocation has already been applied to the
501 * in-core superblock's counter. This should only
502 * be applied to the on-disk superblock.
505 tp
->t_res_frextents_delta
+= delta
;
507 case XFS_TRANS_SB_DBLOCKS
:
509 tp
->t_dblocks_delta
+= delta
;
511 case XFS_TRANS_SB_AGCOUNT
:
513 tp
->t_agcount_delta
+= delta
;
515 case XFS_TRANS_SB_IMAXPCT
:
516 tp
->t_imaxpct_delta
+= delta
;
518 case XFS_TRANS_SB_REXTSIZE
:
519 tp
->t_rextsize_delta
+= delta
;
521 case XFS_TRANS_SB_RBMBLOCKS
:
522 tp
->t_rbmblocks_delta
+= delta
;
524 case XFS_TRANS_SB_RBLOCKS
:
525 tp
->t_rblocks_delta
+= delta
;
527 case XFS_TRANS_SB_REXTENTS
:
528 tp
->t_rextents_delta
+= delta
;
530 case XFS_TRANS_SB_REXTSLOG
:
531 tp
->t_rextslog_delta
+= delta
;
538 tp
->t_flags
|= flags
;
542 * xfs_trans_apply_sb_deltas() is called from the commit code
543 * to bring the superblock buffer into the current transaction
544 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
546 * For now we just look at each field allowed to change and change
550 xfs_trans_apply_sb_deltas(
557 bp
= xfs_trans_getsb(tp
, tp
->t_mountp
, 0);
558 sbp
= XFS_BUF_TO_SBP(bp
);
561 * Check that superblock mods match the mods made to AGF counters.
563 ASSERT((tp
->t_fdblocks_delta
+ tp
->t_res_fdblocks_delta
) ==
564 (tp
->t_ag_freeblks_delta
+ tp
->t_ag_flist_delta
+
565 tp
->t_ag_btree_delta
));
568 * Only update the superblock counters if we are logging them
570 if (!xfs_sb_version_haslazysbcount(&(tp
->t_mountp
->m_sb
))) {
571 if (tp
->t_icount_delta
)
572 be64_add_cpu(&sbp
->sb_icount
, tp
->t_icount_delta
);
573 if (tp
->t_ifree_delta
)
574 be64_add_cpu(&sbp
->sb_ifree
, tp
->t_ifree_delta
);
575 if (tp
->t_fdblocks_delta
)
576 be64_add_cpu(&sbp
->sb_fdblocks
, tp
->t_fdblocks_delta
);
577 if (tp
->t_res_fdblocks_delta
)
578 be64_add_cpu(&sbp
->sb_fdblocks
, tp
->t_res_fdblocks_delta
);
581 if (tp
->t_frextents_delta
)
582 be64_add_cpu(&sbp
->sb_frextents
, tp
->t_frextents_delta
);
583 if (tp
->t_res_frextents_delta
)
584 be64_add_cpu(&sbp
->sb_frextents
, tp
->t_res_frextents_delta
);
586 if (tp
->t_dblocks_delta
) {
587 be64_add_cpu(&sbp
->sb_dblocks
, tp
->t_dblocks_delta
);
590 if (tp
->t_agcount_delta
) {
591 be32_add_cpu(&sbp
->sb_agcount
, tp
->t_agcount_delta
);
594 if (tp
->t_imaxpct_delta
) {
595 sbp
->sb_imax_pct
+= tp
->t_imaxpct_delta
;
598 if (tp
->t_rextsize_delta
) {
599 be32_add_cpu(&sbp
->sb_rextsize
, tp
->t_rextsize_delta
);
602 if (tp
->t_rbmblocks_delta
) {
603 be32_add_cpu(&sbp
->sb_rbmblocks
, tp
->t_rbmblocks_delta
);
606 if (tp
->t_rblocks_delta
) {
607 be64_add_cpu(&sbp
->sb_rblocks
, tp
->t_rblocks_delta
);
610 if (tp
->t_rextents_delta
) {
611 be64_add_cpu(&sbp
->sb_rextents
, tp
->t_rextents_delta
);
614 if (tp
->t_rextslog_delta
) {
615 sbp
->sb_rextslog
+= tp
->t_rextslog_delta
;
621 * Log the whole thing, the fields are noncontiguous.
623 xfs_trans_log_buf(tp
, bp
, 0, sizeof(xfs_dsb_t
) - 1);
626 * Since all the modifiable fields are contiguous, we
627 * can get away with this.
629 xfs_trans_log_buf(tp
, bp
, offsetof(xfs_dsb_t
, sb_icount
),
630 offsetof(xfs_dsb_t
, sb_frextents
) +
631 sizeof(sbp
->sb_frextents
) - 1);
635 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
636 * and apply superblock counter changes to the in-core superblock. The
637 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
638 * applied to the in-core superblock. The idea is that that has already been
641 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
642 * However, we have to ensure that we only modify each superblock field only
643 * once because the application of the delta values may not be atomic. That can
644 * lead to ENOSPC races occurring if we have two separate modifcations of the
645 * free space counter to put back the entire reservation and then take away
648 * If we are not logging superblock counters, then the inode allocated/free and
649 * used block counts are not updated in the on disk superblock. In this case,
650 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
651 * still need to update the incore superblock with the changes.
654 xfs_trans_unreserve_and_mod_sb(
657 xfs_mod_sb_t msb
[14]; /* If you add cases, add entries */
659 xfs_mount_t
*mp
= tp
->t_mountp
;
663 int64_t blkdelta
= 0;
664 int64_t rtxdelta
= 0;
667 rsvd
= (tp
->t_flags
& XFS_TRANS_RESERVE
) != 0;
669 /* calculate free blocks delta */
670 if (tp
->t_blk_res
> 0)
671 blkdelta
= tp
->t_blk_res
;
673 if ((tp
->t_fdblocks_delta
!= 0) &&
674 (xfs_sb_version_haslazysbcount(&mp
->m_sb
) ||
675 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)))
676 blkdelta
+= tp
->t_fdblocks_delta
;
679 msbp
->msb_field
= XFS_SBS_FDBLOCKS
;
680 msbp
->msb_delta
= blkdelta
;
684 /* calculate free realtime extents delta */
685 if (tp
->t_rtx_res
> 0)
686 rtxdelta
= tp
->t_rtx_res
;
688 if ((tp
->t_frextents_delta
!= 0) &&
689 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
))
690 rtxdelta
+= tp
->t_frextents_delta
;
693 msbp
->msb_field
= XFS_SBS_FREXTENTS
;
694 msbp
->msb_delta
= rtxdelta
;
698 /* apply remaining deltas */
700 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) ||
701 (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)) {
702 if (tp
->t_icount_delta
!= 0) {
703 msbp
->msb_field
= XFS_SBS_ICOUNT
;
704 msbp
->msb_delta
= tp
->t_icount_delta
;
707 if (tp
->t_ifree_delta
!= 0) {
708 msbp
->msb_field
= XFS_SBS_IFREE
;
709 msbp
->msb_delta
= tp
->t_ifree_delta
;
714 if (tp
->t_flags
& XFS_TRANS_SB_DIRTY
) {
715 if (tp
->t_dblocks_delta
!= 0) {
716 msbp
->msb_field
= XFS_SBS_DBLOCKS
;
717 msbp
->msb_delta
= tp
->t_dblocks_delta
;
720 if (tp
->t_agcount_delta
!= 0) {
721 msbp
->msb_field
= XFS_SBS_AGCOUNT
;
722 msbp
->msb_delta
= tp
->t_agcount_delta
;
725 if (tp
->t_imaxpct_delta
!= 0) {
726 msbp
->msb_field
= XFS_SBS_IMAX_PCT
;
727 msbp
->msb_delta
= tp
->t_imaxpct_delta
;
730 if (tp
->t_rextsize_delta
!= 0) {
731 msbp
->msb_field
= XFS_SBS_REXTSIZE
;
732 msbp
->msb_delta
= tp
->t_rextsize_delta
;
735 if (tp
->t_rbmblocks_delta
!= 0) {
736 msbp
->msb_field
= XFS_SBS_RBMBLOCKS
;
737 msbp
->msb_delta
= tp
->t_rbmblocks_delta
;
740 if (tp
->t_rblocks_delta
!= 0) {
741 msbp
->msb_field
= XFS_SBS_RBLOCKS
;
742 msbp
->msb_delta
= tp
->t_rblocks_delta
;
745 if (tp
->t_rextents_delta
!= 0) {
746 msbp
->msb_field
= XFS_SBS_REXTENTS
;
747 msbp
->msb_delta
= tp
->t_rextents_delta
;
750 if (tp
->t_rextslog_delta
!= 0) {
751 msbp
->msb_field
= XFS_SBS_REXTSLOG
;
752 msbp
->msb_delta
= tp
->t_rextslog_delta
;
758 * If we need to change anything, do it.
761 error
= xfs_mod_incore_sb_batch(tp
->t_mountp
, msb
,
762 (uint
)(msbp
- msb
), rsvd
);
771 * Commit the given transaction to the log a/synchronously.
773 * XFS disk error handling mechanism is not based on a typical
774 * transaction abort mechanism. Logically after the filesystem
775 * gets marked 'SHUTDOWN', we can't let any new transactions
776 * be durable - ie. committed to disk - because some metadata might
777 * be inconsistent. In such cases, this returns an error, and the
778 * caller may assume that all locked objects joined to the transaction
779 * have already been unlocked as if the commit had succeeded.
780 * Do not reference the transaction structure after this call.
789 xfs_log_iovec_t
*log_vector
;
792 xfs_lsn_t commit_lsn
;
797 #define XFS_TRANS_LOGVEC_COUNT 16
798 xfs_log_iovec_t log_vector_fast
[XFS_TRANS_LOGVEC_COUNT
];
799 struct xlog_in_core
*commit_iclog
;
805 * Determine whether this commit is releasing a permanent
806 * log reservation or not.
808 if (flags
& XFS_TRANS_RELEASE_LOG_RES
) {
809 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
810 log_flags
= XFS_LOG_REL_PERM_RESERV
;
817 * If there is nothing to be logged by the transaction,
818 * then unlock all of the items associated with the
819 * transaction and free the transaction structure.
820 * Also make sure to return any reserved blocks to
824 shutdown
= XFS_FORCED_SHUTDOWN(mp
) ? EIO
: 0;
825 if (!(tp
->t_flags
& XFS_TRANS_DIRTY
) || shutdown
) {
826 xfs_trans_unreserve_and_mod_sb(tp
);
828 * It is indeed possible for the transaction to be
829 * not dirty but the dqinfo portion to be. All that
830 * means is that we have some (non-persistent) quota
831 * reservations that need to be unreserved.
833 xfs_trans_unreserve_and_mod_dquots(tp
);
835 commit_lsn
= xfs_log_done(mp
, tp
->t_ticket
,
837 if (commit_lsn
== -1 && !shutdown
)
838 shutdown
= XFS_ERROR(EIO
);
840 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
841 xfs_trans_free_items(tp
, shutdown
? XFS_TRANS_ABORT
: 0);
842 xfs_trans_free_busy(tp
);
844 XFS_STATS_INC(xs_trans_empty
);
847 ASSERT(tp
->t_ticket
!= NULL
);
850 * If we need to update the superblock, then do it now.
852 if (tp
->t_flags
& XFS_TRANS_SB_DIRTY
)
853 xfs_trans_apply_sb_deltas(tp
);
854 xfs_trans_apply_dquot_deltas(tp
);
857 * Ask each log item how many log_vector entries it will
858 * need so we can figure out how many to allocate.
859 * Try to avoid the kmem_alloc() call in the common case
860 * by using a vector from the stack when it fits.
862 nvec
= xfs_trans_count_vecs(tp
);
864 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
866 } else if (nvec
<= XFS_TRANS_LOGVEC_COUNT
) {
867 log_vector
= log_vector_fast
;
869 log_vector
= (xfs_log_iovec_t
*)kmem_alloc(nvec
*
870 sizeof(xfs_log_iovec_t
),
875 * Fill in the log_vector and pin the logged items, and
876 * then write the transaction to the log.
878 xfs_trans_fill_vecs(tp
, log_vector
);
880 error
= xfs_log_write(mp
, log_vector
, nvec
, tp
->t_ticket
, &(tp
->t_lsn
));
883 * The transaction is committed incore here, and can go out to disk
884 * at any time after this call. However, all the items associated
885 * with the transaction are still locked and pinned in memory.
887 commit_lsn
= xfs_log_done(mp
, tp
->t_ticket
, &commit_iclog
, log_flags
);
889 tp
->t_commit_lsn
= commit_lsn
;
890 if (nvec
> XFS_TRANS_LOGVEC_COUNT
) {
891 kmem_free(log_vector
);
895 * If we got a log write error. Unpin the logitems that we
896 * had pinned, clean up, free trans structure, and return error.
898 if (error
|| commit_lsn
== -1) {
899 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
900 xfs_trans_uncommit(tp
, flags
|XFS_TRANS_ABORT
);
901 return XFS_ERROR(EIO
);
905 * Once the transaction has committed, unused
906 * reservations need to be released and changes to
907 * the superblock need to be reflected in the in-core
908 * version. Do that now.
910 xfs_trans_unreserve_and_mod_sb(tp
);
912 sync
= tp
->t_flags
& XFS_TRANS_SYNC
;
915 * Tell the LM to call the transaction completion routine
916 * when the log write with LSN commit_lsn completes (e.g.
917 * when the transaction commit really hits the on-disk log).
918 * After this call we cannot reference tp, because the call
919 * can happen at any time and the call will free the transaction
920 * structure pointed to by tp. The only case where we call
921 * the completion routine (xfs_trans_committed) directly is
922 * if the log is turned off on a debug kernel or we're
923 * running in simulation mode (the log is explicitly turned
926 tp
->t_logcb
.cb_func
= (void(*)(void*, int))xfs_trans_committed
;
927 tp
->t_logcb
.cb_arg
= tp
;
930 * We need to pass the iclog buffer which was used for the
931 * transaction commit record into this function, and attach
932 * the callback to it. The callback must be attached before
933 * the items are unlocked to avoid racing with other threads
934 * waiting for an item to unlock.
936 shutdown
= xfs_log_notify(mp
, commit_iclog
, &(tp
->t_logcb
));
939 * Mark this thread as no longer being in a transaction
941 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
944 * Once all the items of the transaction have been copied
945 * to the in core log and the callback is attached, the
946 * items can be unlocked.
948 * This will free descriptors pointing to items which were
949 * not logged since there is nothing more to do with them.
950 * For items which were logged, we will keep pointers to them
951 * so they can be unpinned after the transaction commits to disk.
952 * This will also stamp each modified meta-data item with
953 * the commit lsn of this transaction for dependency tracking
956 xfs_trans_unlock_items(tp
, commit_lsn
);
959 * If we detected a log error earlier, finish committing
960 * the transaction now (unpin log items, etc).
962 * Order is critical here, to avoid using the transaction
963 * pointer after its been freed (by xfs_trans_committed
964 * either here now, or as a callback). We cannot do this
965 * step inside xfs_log_notify as was done earlier because
969 xfs_trans_committed(tp
, XFS_LI_ABORTED
);
972 * Now that the xfs_trans_committed callback has been attached,
973 * and the items are released we can finally allow the iclog to
976 error
= xfs_log_release_iclog(mp
, commit_iclog
);
979 * If the transaction needs to be synchronous, then force the
980 * log out now and wait for it.
984 error
= _xfs_log_force_lsn(mp
, commit_lsn
,
985 XFS_LOG_SYNC
, log_flushed
);
987 XFS_STATS_INC(xs_trans_sync
);
989 XFS_STATS_INC(xs_trans_async
);
997 * Total up the number of log iovecs needed to commit this
998 * transaction. The transaction itself needs one for the
999 * transaction header. Ask each dirty item in turn how many
1000 * it needs to get the total.
1003 xfs_trans_count_vecs(
1007 xfs_log_item_desc_t
*lidp
;
1010 lidp
= xfs_trans_first_item(tp
);
1011 ASSERT(lidp
!= NULL
);
1013 /* In the non-debug case we need to start bailing out if we
1014 * didn't find a log_item here, return zero and let trans_commit
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
);
1028 lidp
->lid_size
= IOP_SIZE(lidp
->lid_item
);
1029 nvecs
+= lidp
->lid_size
;
1030 lidp
= xfs_trans_next_item(tp
, lidp
);
1037 * Called from the trans_commit code when we notice that
1038 * the filesystem is in the middle of a forced shutdown.
1045 xfs_log_item_desc_t
*lidp
;
1047 for (lidp
= xfs_trans_first_item(tp
);
1049 lidp
= xfs_trans_next_item(tp
, lidp
)) {
1051 * Unpin all but those that aren't dirty.
1053 if (lidp
->lid_flags
& XFS_LID_DIRTY
)
1054 IOP_UNPIN_REMOVE(lidp
->lid_item
, tp
);
1057 xfs_trans_unreserve_and_mod_sb(tp
);
1058 xfs_trans_unreserve_and_mod_dquots(tp
);
1060 xfs_trans_free_items(tp
, flags
);
1061 xfs_trans_free_busy(tp
);
1066 * Fill in the vector with pointers to data to be logged
1067 * by this transaction. The transaction header takes
1068 * the first vector, and then each dirty item takes the
1069 * number of vectors it indicated it needed in xfs_trans_count_vecs().
1071 * As each item fills in the entries it needs, also pin the item
1072 * so that it cannot be flushed out until the log write completes.
1075 xfs_trans_fill_vecs(
1077 xfs_log_iovec_t
*log_vector
)
1079 xfs_log_item_desc_t
*lidp
;
1080 xfs_log_iovec_t
*vecp
;
1084 * Skip over the entry for the transaction header, we'll
1085 * fill that in at the end.
1087 vecp
= log_vector
+ 1; /* pointer arithmetic */
1090 lidp
= xfs_trans_first_item(tp
);
1091 ASSERT(lidp
!= NULL
);
1092 while (lidp
!= NULL
) {
1094 * Skip items which aren't dirty in this transaction.
1096 if (!(lidp
->lid_flags
& XFS_LID_DIRTY
)) {
1097 lidp
= xfs_trans_next_item(tp
, lidp
);
1101 * The item may be marked dirty but not log anything.
1102 * This can be used to get called when a transaction
1105 if (lidp
->lid_size
) {
1108 IOP_FORMAT(lidp
->lid_item
, vecp
);
1109 vecp
+= lidp
->lid_size
; /* pointer arithmetic */
1110 IOP_PIN(lidp
->lid_item
);
1111 lidp
= xfs_trans_next_item(tp
, lidp
);
1115 * Now that we've counted the number of items in this
1116 * transaction, fill in the transaction header.
1118 tp
->t_header
.th_magic
= XFS_TRANS_HEADER_MAGIC
;
1119 tp
->t_header
.th_type
= tp
->t_type
;
1120 tp
->t_header
.th_num_items
= nitems
;
1121 log_vector
->i_addr
= (xfs_caddr_t
)&tp
->t_header
;
1122 log_vector
->i_len
= sizeof(xfs_trans_header_t
);
1123 log_vector
->i_type
= XLOG_REG_TYPE_TRANSHDR
;
1128 * Unlock all of the transaction's items and free the transaction.
1129 * The transaction must not have modified any of its items, because
1130 * there is no way to restore them to their previous state.
1132 * If the transaction has made a log reservation, make sure to release
1142 xfs_log_item_chunk_t
*licp
;
1143 xfs_log_item_desc_t
*lidp
;
1144 xfs_log_item_t
*lip
;
1147 xfs_mount_t
*mp
= tp
->t_mountp
;
1150 * See if the caller is being too lazy to figure out if
1151 * the transaction really needs an abort.
1153 if ((flags
& XFS_TRANS_ABORT
) && !(tp
->t_flags
& XFS_TRANS_DIRTY
))
1154 flags
&= ~XFS_TRANS_ABORT
;
1156 * See if the caller is relying on us to shut down the
1157 * filesystem. This happens in paths where we detect
1158 * corruption and decide to give up.
1160 if ((tp
->t_flags
& XFS_TRANS_DIRTY
) && !XFS_FORCED_SHUTDOWN(mp
)) {
1161 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW
, mp
);
1162 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1165 if (!(flags
& XFS_TRANS_ABORT
)) {
1166 licp
= &(tp
->t_items
);
1167 while (licp
!= NULL
) {
1168 lidp
= licp
->lic_descs
;
1169 for (i
= 0; i
< licp
->lic_unused
; i
++, lidp
++) {
1170 if (xfs_lic_isfree(licp
, i
)) {
1174 lip
= lidp
->lid_item
;
1175 if (!XFS_FORCED_SHUTDOWN(mp
))
1176 ASSERT(!(lip
->li_type
== XFS_LI_EFD
));
1178 licp
= licp
->lic_next
;
1182 xfs_trans_unreserve_and_mod_sb(tp
);
1183 xfs_trans_unreserve_and_mod_dquots(tp
);
1186 if (flags
& XFS_TRANS_RELEASE_LOG_RES
) {
1187 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1188 log_flags
= XFS_LOG_REL_PERM_RESERV
;
1192 xfs_log_done(mp
, tp
->t_ticket
, NULL
, log_flags
);
1195 /* mark this thread as no longer being in a transaction */
1196 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
1198 xfs_trans_free_items(tp
, flags
);
1199 xfs_trans_free_busy(tp
);
1205 * Free the transaction structure. If there is more clean up
1206 * to do when the structure is freed, add it here.
1212 atomic_dec(&tp
->t_mountp
->m_active_trans
);
1213 xfs_trans_free_dqinfo(tp
);
1214 kmem_zone_free(xfs_trans_zone
, tp
);
1218 * Roll from one trans in the sequence of PERMANENT transactions to
1219 * the next: permanent transactions are only flushed out when
1220 * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
1221 * as possible to let chunks of it go to the log. So we commit the
1222 * chunk we've been working on and get a new transaction to continue.
1226 struct xfs_trans
**tpp
,
1227 struct xfs_inode
*dp
)
1229 struct xfs_trans
*trans
;
1230 unsigned int logres
, count
;
1234 * Ensure that the inode is always logged.
1237 xfs_trans_log_inode(trans
, dp
, XFS_ILOG_CORE
);
1240 * Copy the critical parameters from one trans to the next.
1242 logres
= trans
->t_log_res
;
1243 count
= trans
->t_log_count
;
1244 *tpp
= xfs_trans_dup(trans
);
1247 * Commit the current transaction.
1248 * If this commit failed, then it'd just unlock those items that
1249 * are not marked ihold. That also means that a filesystem shutdown
1250 * is in progress. The caller takes the responsibility to cancel
1251 * the duplicate transaction that gets returned.
1253 error
= xfs_trans_commit(trans
, 0);
1260 * transaction commit worked ok so we can drop the extra ticket
1261 * reference that we gained in xfs_trans_dup()
1263 xfs_log_ticket_put(trans
->t_ticket
);
1267 * Reserve space in the log for th next transaction.
1268 * This also pushes items in the "AIL", the list of logged items,
1269 * out to disk if they are taking up space at the tail of the log
1270 * that we want to use. This requires that either nothing be locked
1271 * across this call, or that anything that is locked be logged in
1272 * the prior and the next transactions.
1274 error
= xfs_trans_reserve(trans
, 0, logres
, 0,
1275 XFS_TRANS_PERM_LOG_RES
, count
);
1277 * Ensure that the inode is in the new transaction and locked.
1282 xfs_trans_ijoin(trans
, dp
, XFS_ILOCK_EXCL
);
1283 xfs_trans_ihold(trans
, dp
);
1288 * THIS SHOULD BE REWRITTEN TO USE xfs_trans_next_item().
1290 * This is typically called by the LM when a transaction has been fully
1291 * committed to disk. It needs to unpin the items which have
1292 * been logged by the transaction and update their positions
1293 * in the AIL if necessary.
1294 * This also gets called when the transactions didn't get written out
1295 * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then.
1297 * Call xfs_trans_chunk_committed() to process the items in
1301 xfs_trans_committed(
1305 xfs_log_item_chunk_t
*licp
;
1306 xfs_log_item_chunk_t
*next_licp
;
1307 xfs_log_busy_chunk_t
*lbcp
;
1308 xfs_log_busy_slot_t
*lbsp
;
1312 * Call the transaction's completion callback if there
1315 if (tp
->t_callback
!= NULL
) {
1316 tp
->t_callback(tp
, tp
->t_callarg
);
1320 * Special case the chunk embedded in the transaction.
1322 licp
= &(tp
->t_items
);
1323 if (!(xfs_lic_are_all_free(licp
))) {
1324 xfs_trans_chunk_committed(licp
, tp
->t_lsn
, abortflag
);
1328 * Process the items in each chunk in turn.
1330 licp
= licp
->lic_next
;
1331 while (licp
!= NULL
) {
1332 ASSERT(!xfs_lic_are_all_free(licp
));
1333 xfs_trans_chunk_committed(licp
, tp
->t_lsn
, abortflag
);
1334 next_licp
= licp
->lic_next
;
1340 * Clear all the per-AG busy list items listed in this transaction
1343 while (lbcp
!= NULL
) {
1344 for (i
= 0, lbsp
= lbcp
->lbc_busy
; i
< lbcp
->lbc_unused
; i
++, lbsp
++) {
1345 if (!XFS_LBC_ISFREE(lbcp
, i
)) {
1346 xfs_alloc_clear_busy(tp
, lbsp
->lbc_ag
,
1350 lbcp
= lbcp
->lbc_next
;
1352 xfs_trans_free_busy(tp
);
1355 * That's it for the transaction structure. Free it.
1361 * This is called to perform the commit processing for each
1362 * item described by the given chunk.
1364 * The commit processing consists of unlocking items which were
1365 * held locked with the SYNC_UNLOCK attribute, calling the committed
1366 * routine of each logged item, updating the item's position in the AIL
1367 * if necessary, and unpinning each item. If the committed routine
1368 * returns -1, then do nothing further with the item because it
1369 * may have been freed.
1371 * Since items are unlocked when they are copied to the incore
1372 * log, it is possible for two transactions to be completing
1373 * and manipulating the same item simultaneously. The AIL lock
1374 * will protect the lsn field of each item. The value of this
1375 * field can never go backwards.
1377 * We unpin the items after repositioning them in the AIL, because
1378 * otherwise they could be immediately flushed and we'd have to race
1379 * with the flusher trying to pull the item from the AIL as we add it.
1382 xfs_trans_chunk_committed(
1383 xfs_log_item_chunk_t
*licp
,
1387 xfs_log_item_desc_t
*lidp
;
1388 xfs_log_item_t
*lip
;
1392 lidp
= licp
->lic_descs
;
1393 for (i
= 0; i
< licp
->lic_unused
; i
++, lidp
++) {
1394 struct xfs_ail
*ailp
;
1396 if (xfs_lic_isfree(licp
, i
)) {
1400 lip
= lidp
->lid_item
;
1402 lip
->li_flags
|= XFS_LI_ABORTED
;
1405 * Send in the ABORTED flag to the COMMITTED routine
1406 * so that it knows whether the transaction was aborted
1409 item_lsn
= IOP_COMMITTED(lip
, lsn
);
1412 * If the committed routine returns -1, make
1413 * no more references to the item.
1415 if (XFS_LSN_CMP(item_lsn
, (xfs_lsn_t
)-1) == 0) {
1420 * If the returned lsn is greater than what it
1421 * contained before, update the location of the
1422 * item in the AIL. If it is not, then do nothing.
1423 * Items can never move backwards in the AIL.
1425 * While the new lsn should usually be greater, it
1426 * is possible that a later transaction completing
1427 * simultaneously with an earlier one using the
1428 * same item could complete first with a higher lsn.
1429 * This would cause the earlier transaction to fail
1432 ailp
= lip
->li_ailp
;
1433 spin_lock(&ailp
->xa_lock
);
1434 if (XFS_LSN_CMP(item_lsn
, lip
->li_lsn
) > 0) {
1436 * This will set the item's lsn to item_lsn
1437 * and update the position of the item in
1440 * xfs_trans_ail_update() drops the AIL lock.
1442 xfs_trans_ail_update(ailp
, lip
, item_lsn
);
1444 spin_unlock(&ailp
->xa_lock
);
1448 * Now that we've repositioned the item in the AIL,
1449 * unpin it so it can be flushed. Pass information
1450 * about buffer stale state down from the log item
1451 * flags, if anyone else stales the buffer we do not
1452 * want to pay any attention to it.
1454 IOP_UNPIN(lip
, lidp
->lid_flags
& XFS_LID_BUF_STALE
);