2 * Copyright (c) 2000-2003 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
22 #include "xfs_trans.h"
25 #include "xfs_alloc.h"
26 #include "xfs_quota.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_inode.h"
31 #include "xfs_rtalloc.h"
32 #include "xfs_error.h"
33 #include "xfs_itable.h"
35 #include "xfs_buf_item.h"
36 #include "xfs_trans_space.h"
37 #include "xfs_trans_priv.h"
39 #include "xfs_trace.h"
46 * dquot->q_qlock (xfs_dqlock() and friends)
47 * dquot->q_flush (xfs_dqflock() and friends)
50 * If two dquots need to be locked the order is user before group/project,
51 * otherwise by the lowest id first, see xfs_dqlock2.
55 xfs_buftarg_t
*xfs_dqerror_target
;
58 int xfs_dqerror_mod
= 33;
61 struct kmem_zone
*xfs_qm_dqtrxzone
;
62 static struct kmem_zone
*xfs_qm_dqzone
;
64 static struct lock_class_key xfs_dquot_other_class
;
67 * This is called to free all the memory associated with a dquot
73 ASSERT(list_empty(&dqp
->q_lru
));
75 mutex_destroy(&dqp
->q_qlock
);
76 kmem_zone_free(xfs_qm_dqzone
, dqp
);
78 XFS_STATS_DEC(xs_qm_dquot
);
82 * If default limits are in force, push them into the dquot now.
83 * We overwrite the dquot limits only if they are zero and this
84 * is not the root dquot.
87 xfs_qm_adjust_dqlimits(
91 xfs_quotainfo_t
*q
= mp
->m_quotainfo
;
95 if (q
->qi_bsoftlimit
&& !d
->d_blk_softlimit
)
96 d
->d_blk_softlimit
= cpu_to_be64(q
->qi_bsoftlimit
);
97 if (q
->qi_bhardlimit
&& !d
->d_blk_hardlimit
)
98 d
->d_blk_hardlimit
= cpu_to_be64(q
->qi_bhardlimit
);
99 if (q
->qi_isoftlimit
&& !d
->d_ino_softlimit
)
100 d
->d_ino_softlimit
= cpu_to_be64(q
->qi_isoftlimit
);
101 if (q
->qi_ihardlimit
&& !d
->d_ino_hardlimit
)
102 d
->d_ino_hardlimit
= cpu_to_be64(q
->qi_ihardlimit
);
103 if (q
->qi_rtbsoftlimit
&& !d
->d_rtb_softlimit
)
104 d
->d_rtb_softlimit
= cpu_to_be64(q
->qi_rtbsoftlimit
);
105 if (q
->qi_rtbhardlimit
&& !d
->d_rtb_hardlimit
)
106 d
->d_rtb_hardlimit
= cpu_to_be64(q
->qi_rtbhardlimit
);
110 * Check the limits and timers of a dquot and start or reset timers
112 * This gets called even when quota enforcement is OFF, which makes our
113 * life a little less complicated. (We just don't reject any quota
114 * reservations in that case, when enforcement is off).
115 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
117 * In contrast, warnings are a little different in that they don't
118 * 'automatically' get started when limits get exceeded. They do
119 * get reset to zero, however, when we find the count to be under
120 * the soft limit (they are only ever set non-zero via userspace).
123 xfs_qm_adjust_dqtimers(
130 if (d
->d_blk_hardlimit
)
131 ASSERT(be64_to_cpu(d
->d_blk_softlimit
) <=
132 be64_to_cpu(d
->d_blk_hardlimit
));
133 if (d
->d_ino_hardlimit
)
134 ASSERT(be64_to_cpu(d
->d_ino_softlimit
) <=
135 be64_to_cpu(d
->d_ino_hardlimit
));
136 if (d
->d_rtb_hardlimit
)
137 ASSERT(be64_to_cpu(d
->d_rtb_softlimit
) <=
138 be64_to_cpu(d
->d_rtb_hardlimit
));
142 if ((d
->d_blk_softlimit
&&
143 (be64_to_cpu(d
->d_bcount
) >
144 be64_to_cpu(d
->d_blk_softlimit
))) ||
145 (d
->d_blk_hardlimit
&&
146 (be64_to_cpu(d
->d_bcount
) >
147 be64_to_cpu(d
->d_blk_hardlimit
)))) {
148 d
->d_btimer
= cpu_to_be32(get_seconds() +
149 mp
->m_quotainfo
->qi_btimelimit
);
154 if ((!d
->d_blk_softlimit
||
155 (be64_to_cpu(d
->d_bcount
) <=
156 be64_to_cpu(d
->d_blk_softlimit
))) &&
157 (!d
->d_blk_hardlimit
||
158 (be64_to_cpu(d
->d_bcount
) <=
159 be64_to_cpu(d
->d_blk_hardlimit
)))) {
165 if ((d
->d_ino_softlimit
&&
166 (be64_to_cpu(d
->d_icount
) >
167 be64_to_cpu(d
->d_ino_softlimit
))) ||
168 (d
->d_ino_hardlimit
&&
169 (be64_to_cpu(d
->d_icount
) >
170 be64_to_cpu(d
->d_ino_hardlimit
)))) {
171 d
->d_itimer
= cpu_to_be32(get_seconds() +
172 mp
->m_quotainfo
->qi_itimelimit
);
177 if ((!d
->d_ino_softlimit
||
178 (be64_to_cpu(d
->d_icount
) <=
179 be64_to_cpu(d
->d_ino_softlimit
))) &&
180 (!d
->d_ino_hardlimit
||
181 (be64_to_cpu(d
->d_icount
) <=
182 be64_to_cpu(d
->d_ino_hardlimit
)))) {
187 if (!d
->d_rtbtimer
) {
188 if ((d
->d_rtb_softlimit
&&
189 (be64_to_cpu(d
->d_rtbcount
) >
190 be64_to_cpu(d
->d_rtb_softlimit
))) ||
191 (d
->d_rtb_hardlimit
&&
192 (be64_to_cpu(d
->d_rtbcount
) >
193 be64_to_cpu(d
->d_rtb_hardlimit
)))) {
194 d
->d_rtbtimer
= cpu_to_be32(get_seconds() +
195 mp
->m_quotainfo
->qi_rtbtimelimit
);
200 if ((!d
->d_rtb_softlimit
||
201 (be64_to_cpu(d
->d_rtbcount
) <=
202 be64_to_cpu(d
->d_rtb_softlimit
))) &&
203 (!d
->d_rtb_hardlimit
||
204 (be64_to_cpu(d
->d_rtbcount
) <=
205 be64_to_cpu(d
->d_rtb_hardlimit
)))) {
212 * initialize a buffer full of dquots and log the whole thing
215 xfs_qm_init_dquot_blk(
222 struct xfs_quotainfo
*q
= mp
->m_quotainfo
;
227 ASSERT(xfs_buf_islocked(bp
));
232 * ID of the first dquot in the block - id's are zero based.
234 curid
= id
- (id
% q
->qi_dqperchunk
);
236 memset(d
, 0, BBTOB(q
->qi_dqchunklen
));
237 for (i
= 0; i
< q
->qi_dqperchunk
; i
++, d
++, curid
++) {
238 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
239 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
240 d
->dd_diskdq
.d_id
= cpu_to_be32(curid
);
241 d
->dd_diskdq
.d_flags
= type
;
244 xfs_trans_dquot_buf(tp
, bp
,
245 (type
& XFS_DQ_USER
? XFS_BLF_UDQUOT_BUF
:
246 ((type
& XFS_DQ_PROJ
) ? XFS_BLF_PDQUOT_BUF
:
247 XFS_BLF_GDQUOT_BUF
)));
248 xfs_trans_log_buf(tp
, bp
, 0, BBTOB(q
->qi_dqchunklen
) - 1);
252 xfs_dquot_buf_verify(
255 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
256 struct xfs_dqblk
*d
= (struct xfs_dqblk
*)bp
->b_addr
;
257 struct xfs_disk_dquot
*ddq
;
262 * On the first read of the buffer, verify that each dquot is valid.
263 * We don't know what the id of the dquot is supposed to be, just that
264 * they should be increasing monotonically within the buffer. If the
265 * first id is corrupt, then it will fail on the second dquot in the
266 * buffer so corruptions could point to the wrong dquot in this case.
268 for (i
= 0; i
< mp
->m_quotainfo
->qi_dqperchunk
; i
++) {
271 ddq
= &d
[i
].dd_diskdq
;
274 id
= be32_to_cpu(ddq
->d_id
);
276 error
= xfs_qm_dqcheck(mp
, ddq
, id
+ i
, 0, XFS_QMOPT_DOWARN
,
277 "xfs_dquot_read_verify");
279 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, d
);
280 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
287 xfs_dquot_buf_read_verify(
290 xfs_dquot_buf_verify(bp
);
294 xfs_dquot_buf_write_verify(
297 xfs_dquot_buf_verify(bp
);
300 const struct xfs_buf_ops xfs_dquot_buf_ops
= {
301 .verify_read
= xfs_dquot_buf_read_verify
,
302 .verify_write
= xfs_dquot_buf_write_verify
,
306 * Allocate a block and fill it with dquots.
307 * This is called when the bmapi finds a hole.
315 xfs_fileoff_t offset_fsb
,
318 xfs_fsblock_t firstblock
;
319 xfs_bmap_free_t flist
;
321 int nmaps
, error
, committed
;
323 xfs_trans_t
*tp
= *tpp
;
327 trace_xfs_dqalloc(dqp
);
330 * Initialize the bmap freelist prior to calling bmapi code.
332 xfs_bmap_init(&flist
, &firstblock
);
333 xfs_ilock(quotip
, XFS_ILOCK_EXCL
);
335 * Return if this type of quotas is turned off while we didn't
338 if (!xfs_this_quota_on(dqp
->q_mount
, dqp
->dq_flags
)) {
339 xfs_iunlock(quotip
, XFS_ILOCK_EXCL
);
343 xfs_trans_ijoin(tp
, quotip
, XFS_ILOCK_EXCL
);
345 error
= xfs_bmapi_write(tp
, quotip
, offset_fsb
,
346 XFS_DQUOT_CLUSTER_SIZE_FSB
, XFS_BMAPI_METADATA
,
347 &firstblock
, XFS_QM_DQALLOC_SPACE_RES(mp
),
348 &map
, &nmaps
, &flist
);
351 ASSERT(map
.br_blockcount
== XFS_DQUOT_CLUSTER_SIZE_FSB
);
353 ASSERT((map
.br_startblock
!= DELAYSTARTBLOCK
) &&
354 (map
.br_startblock
!= HOLESTARTBLOCK
));
357 * Keep track of the blkno to save a lookup later
359 dqp
->q_blkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
);
361 /* now we can just get the buffer (there's nothing to read yet) */
362 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
364 mp
->m_quotainfo
->qi_dqchunklen
,
367 error
= xfs_buf_geterror(bp
);
370 bp
->b_ops
= &xfs_dquot_buf_ops
;
373 * Make a chunk of dquots out of this buffer and log
376 xfs_qm_init_dquot_blk(tp
, mp
, be32_to_cpu(dqp
->q_core
.d_id
),
377 dqp
->dq_flags
& XFS_DQ_ALLTYPES
, bp
);
380 * xfs_bmap_finish() may commit the current transaction and
381 * start a second transaction if the freelist is not empty.
383 * Since we still want to modify this buffer, we need to
384 * ensure that the buffer is not released on commit of
385 * the first transaction and ensure the buffer is added to the
386 * second transaction.
388 * If there is only one transaction then don't stop the buffer
389 * from being released when it commits later on.
392 xfs_trans_bhold(tp
, bp
);
394 if ((error
= xfs_bmap_finish(tpp
, &flist
, &committed
))) {
400 xfs_trans_bjoin(tp
, bp
);
402 xfs_trans_bhold_release(tp
, bp
);
409 xfs_bmap_cancel(&flist
);
411 xfs_iunlock(quotip
, XFS_ILOCK_EXCL
);
417 struct xfs_mount
*mp
,
418 struct xfs_trans
*tp
,
419 struct xfs_dquot
*dqp
,
421 struct xfs_buf
**bpp
)
424 struct xfs_disk_dquot
*ddq
;
429 * Read the buffer without verification so we get the corrupted
430 * buffer returned to us. make sure we verify it on write, though.
432 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, dqp
->q_blkno
,
433 mp
->m_quotainfo
->qi_dqchunklen
,
437 ASSERT(*bpp
== NULL
);
438 return XFS_ERROR(error
);
440 (*bpp
)->b_ops
= &xfs_dquot_buf_ops
;
442 ASSERT(xfs_buf_islocked(*bpp
));
443 d
= (struct xfs_dqblk
*)(*bpp
)->b_addr
;
445 /* Do the actual repair of dquots in this buffer */
446 for (i
= 0; i
< mp
->m_quotainfo
->qi_dqperchunk
; i
++) {
447 ddq
= &d
[i
].dd_diskdq
;
448 error
= xfs_qm_dqcheck(mp
, ddq
, firstid
+ i
,
449 dqp
->dq_flags
& XFS_DQ_ALLTYPES
,
450 XFS_QMOPT_DQREPAIR
, "xfs_qm_dqrepair");
452 /* repair failed, we're screwed */
453 xfs_trans_brelse(tp
, *bpp
);
454 return XFS_ERROR(EIO
);
462 * Maps a dquot to the buffer containing its on-disk version.
463 * This returns a ptr to the buffer containing the on-disk dquot
464 * in the bpp param, and a ptr to the on-disk dquot within that buffer
470 xfs_disk_dquot_t
**O_ddpp
,
475 int nmaps
= 1, error
;
477 xfs_inode_t
*quotip
= XFS_DQ_TO_QIP(dqp
);
478 xfs_mount_t
*mp
= dqp
->q_mount
;
479 xfs_dqid_t id
= be32_to_cpu(dqp
->q_core
.d_id
);
480 xfs_trans_t
*tp
= (tpp
? *tpp
: NULL
);
482 dqp
->q_fileoffset
= (xfs_fileoff_t
)id
/ mp
->m_quotainfo
->qi_dqperchunk
;
484 xfs_ilock(quotip
, XFS_ILOCK_SHARED
);
485 if (!xfs_this_quota_on(dqp
->q_mount
, dqp
->dq_flags
)) {
487 * Return if this type of quotas is turned off while we
488 * didn't have the quota inode lock.
490 xfs_iunlock(quotip
, XFS_ILOCK_SHARED
);
495 * Find the block map; no allocations yet
497 error
= xfs_bmapi_read(quotip
, dqp
->q_fileoffset
,
498 XFS_DQUOT_CLUSTER_SIZE_FSB
, &map
, &nmaps
, 0);
500 xfs_iunlock(quotip
, XFS_ILOCK_SHARED
);
505 ASSERT(map
.br_blockcount
== 1);
508 * Offset of dquot in the (fixed sized) dquot chunk.
510 dqp
->q_bufoffset
= (id
% mp
->m_quotainfo
->qi_dqperchunk
) *
513 ASSERT(map
.br_startblock
!= DELAYSTARTBLOCK
);
514 if (map
.br_startblock
== HOLESTARTBLOCK
) {
516 * We don't allocate unless we're asked to
518 if (!(flags
& XFS_QMOPT_DQALLOC
))
522 error
= xfs_qm_dqalloc(tpp
, mp
, dqp
, quotip
,
523 dqp
->q_fileoffset
, &bp
);
528 trace_xfs_dqtobp_read(dqp
);
531 * store the blkno etc so that we don't have to do the
532 * mapping all the time
534 dqp
->q_blkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
);
536 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
538 mp
->m_quotainfo
->qi_dqchunklen
,
539 0, &bp
, &xfs_dquot_buf_ops
);
541 if (error
== EFSCORRUPTED
&& (flags
& XFS_QMOPT_DQREPAIR
)) {
542 xfs_dqid_t firstid
= (xfs_dqid_t
)map
.br_startoff
*
543 mp
->m_quotainfo
->qi_dqperchunk
;
545 error
= xfs_qm_dqrepair(mp
, tp
, dqp
, firstid
, &bp
);
550 return XFS_ERROR(error
);
554 ASSERT(xfs_buf_islocked(bp
));
556 *O_ddpp
= bp
->b_addr
+ dqp
->q_bufoffset
;
563 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
564 * and release the buffer immediately.
566 * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed.
570 struct xfs_mount
*mp
,
574 struct xfs_dquot
**O_dqpp
)
576 struct xfs_dquot
*dqp
;
577 struct xfs_disk_dquot
*ddqp
;
579 struct xfs_trans
*tp
= NULL
;
584 dqp
= kmem_zone_zalloc(xfs_qm_dqzone
, KM_SLEEP
);
586 dqp
->dq_flags
= type
;
587 dqp
->q_core
.d_id
= cpu_to_be32(id
);
589 INIT_LIST_HEAD(&dqp
->q_lru
);
590 mutex_init(&dqp
->q_qlock
);
591 init_waitqueue_head(&dqp
->q_pinwait
);
594 * Because we want to use a counting completion, complete
595 * the flush completion once to allow a single access to
596 * the flush completion without blocking.
598 init_completion(&dqp
->q_flush
);
599 complete(&dqp
->q_flush
);
602 * Make sure group quotas have a different lock class than user
605 if (!(type
& XFS_DQ_USER
))
606 lockdep_set_class(&dqp
->q_qlock
, &xfs_dquot_other_class
);
608 XFS_STATS_INC(xs_qm_dquot
);
610 trace_xfs_dqread(dqp
);
612 if (flags
& XFS_QMOPT_DQALLOC
) {
613 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_DQALLOC
);
614 error
= xfs_trans_reserve(tp
, XFS_QM_DQALLOC_SPACE_RES(mp
),
615 XFS_WRITE_LOG_RES(mp
) +
617 * Round the chunklen up to the next multiple
618 * of 128 (buf log item chunk size)).
620 BBTOB(mp
->m_quotainfo
->qi_dqchunklen
) - 1 + 128,
622 XFS_TRANS_PERM_LOG_RES
,
623 XFS_WRITE_LOG_COUNT
);
626 cancelflags
= XFS_TRANS_RELEASE_LOG_RES
;
630 * get a pointer to the on-disk dquot and the buffer containing it
631 * dqp already knows its own type (GROUP/USER).
633 error
= xfs_qm_dqtobp(&tp
, dqp
, &ddqp
, &bp
, flags
);
636 * This can happen if quotas got turned off (ESRCH),
637 * or if the dquot didn't exist on disk and we ask to
640 trace_xfs_dqread_fail(dqp
);
641 cancelflags
|= XFS_TRANS_ABORT
;
645 /* copy everything from disk dquot to the incore dquot */
646 memcpy(&dqp
->q_core
, ddqp
, sizeof(xfs_disk_dquot_t
));
647 xfs_qm_dquot_logitem_init(dqp
);
650 * Reservation counters are defined as reservation plus current usage
651 * to avoid having to add every time.
653 dqp
->q_res_bcount
= be64_to_cpu(ddqp
->d_bcount
);
654 dqp
->q_res_icount
= be64_to_cpu(ddqp
->d_icount
);
655 dqp
->q_res_rtbcount
= be64_to_cpu(ddqp
->d_rtbcount
);
657 /* Mark the buf so that this will stay incore a little longer */
658 xfs_buf_set_ref(bp
, XFS_DQUOT_REF
);
661 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
662 * So we need to release with xfs_trans_brelse().
663 * The strategy here is identical to that of inodes; we lock
664 * the dquot in xfs_qm_dqget() before making it accessible to
665 * others. This is because dquots, like inodes, need a good level of
666 * concurrency, and we don't want to take locks on the entire buffers
667 * for dquot accesses.
668 * Note also that the dquot buffer may even be dirty at this point, if
669 * this particular dquot was repaired. We still aren't afraid to
670 * brelse it because we have the changes incore.
672 ASSERT(xfs_buf_islocked(bp
));
673 xfs_trans_brelse(tp
, bp
);
676 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
686 xfs_trans_cancel(tp
, cancelflags
);
688 xfs_qm_dqdestroy(dqp
);
694 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
695 * a locked dquot, doing an allocation (if requested) as needed.
696 * When both an inode and an id are given, the inode's id takes precedence.
697 * That is, if the id changes while we don't hold the ilock inside this
698 * function, the new dquot is returned, not necessarily the one requested
699 * in the id argument.
704 xfs_inode_t
*ip
, /* locked inode (optional) */
705 xfs_dqid_t id
, /* uid/projid/gid depending on type */
706 uint type
, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
707 uint flags
, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
708 xfs_dquot_t
**O_dqpp
) /* OUT : locked incore dquot */
710 struct xfs_quotainfo
*qi
= mp
->m_quotainfo
;
711 struct radix_tree_root
*tree
= XFS_DQUOT_TREE(qi
, type
);
712 struct xfs_dquot
*dqp
;
715 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
716 if ((! XFS_IS_UQUOTA_ON(mp
) && type
== XFS_DQ_USER
) ||
717 (! XFS_IS_PQUOTA_ON(mp
) && type
== XFS_DQ_PROJ
) ||
718 (! XFS_IS_GQUOTA_ON(mp
) && type
== XFS_DQ_GROUP
)) {
723 if (xfs_do_dqerror
) {
724 if ((xfs_dqerror_target
== mp
->m_ddev_targp
) &&
725 (xfs_dqreq_num
++ % xfs_dqerror_mod
) == 0) {
726 xfs_debug(mp
, "Returning error in dqget");
731 ASSERT(type
== XFS_DQ_USER
||
732 type
== XFS_DQ_PROJ
||
733 type
== XFS_DQ_GROUP
);
735 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
736 ASSERT(xfs_inode_dquot(ip
, type
) == NULL
);
741 mutex_lock(&qi
->qi_tree_lock
);
742 dqp
= radix_tree_lookup(tree
, id
);
745 if (dqp
->dq_flags
& XFS_DQ_FREEING
) {
747 mutex_unlock(&qi
->qi_tree_lock
);
748 trace_xfs_dqget_freeing(dqp
);
754 mutex_unlock(&qi
->qi_tree_lock
);
756 trace_xfs_dqget_hit(dqp
);
757 XFS_STATS_INC(xs_qm_dqcachehits
);
761 mutex_unlock(&qi
->qi_tree_lock
);
762 XFS_STATS_INC(xs_qm_dqcachemisses
);
765 * Dquot cache miss. We don't want to keep the inode lock across
766 * a (potential) disk read. Also we don't want to deal with the lock
767 * ordering between quotainode and this inode. OTOH, dropping the inode
768 * lock here means dealing with a chown that can happen before
769 * we re-acquire the lock.
772 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
774 error
= xfs_qm_dqread(mp
, id
, type
, flags
, &dqp
);
777 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
784 * A dquot could be attached to this inode by now, since
785 * we had dropped the ilock.
787 if (xfs_this_quota_on(mp
, type
)) {
788 struct xfs_dquot
*dqp1
;
790 dqp1
= xfs_inode_dquot(ip
, type
);
792 xfs_qm_dqdestroy(dqp
);
798 /* inode stays locked on return */
799 xfs_qm_dqdestroy(dqp
);
800 return XFS_ERROR(ESRCH
);
804 mutex_lock(&qi
->qi_tree_lock
);
805 error
= -radix_tree_insert(tree
, id
, dqp
);
806 if (unlikely(error
)) {
807 WARN_ON(error
!= EEXIST
);
810 * Duplicate found. Just throw away the new dquot and start
813 mutex_unlock(&qi
->qi_tree_lock
);
814 trace_xfs_dqget_dup(dqp
);
815 xfs_qm_dqdestroy(dqp
);
816 XFS_STATS_INC(xs_qm_dquot_dups
);
821 * We return a locked dquot to the caller, with a reference taken
827 mutex_unlock(&qi
->qi_tree_lock
);
830 ASSERT((ip
== NULL
) || xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
831 trace_xfs_dqget_miss(dqp
);
839 struct xfs_dquot
*dqp
)
841 struct xfs_quotainfo
*qi
= dqp
->q_mount
->m_quotainfo
;
842 struct xfs_dquot
*gdqp
;
844 trace_xfs_dqput_free(dqp
);
846 mutex_lock(&qi
->qi_lru_lock
);
847 if (list_empty(&dqp
->q_lru
)) {
848 list_add_tail(&dqp
->q_lru
, &qi
->qi_lru_list
);
850 XFS_STATS_INC(xs_qm_dquot_unused
);
852 mutex_unlock(&qi
->qi_lru_lock
);
855 * If we just added a udquot to the freelist, then we want to release
856 * the gdquot reference that it (probably) has. Otherwise it'll keep
857 * the gdquot from getting reclaimed.
859 gdqp
= dqp
->q_gdquot
;
862 dqp
->q_gdquot
= NULL
;
867 * If we had a group quota hint, release it now.
874 * Release a reference to the dquot (decrement ref-count) and unlock it.
876 * If there is a group quota attached to this dquot, carefully release that
877 * too without tripping over deadlocks'n'stuff.
881 struct xfs_dquot
*dqp
)
883 ASSERT(dqp
->q_nrefs
> 0);
884 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
886 trace_xfs_dqput(dqp
);
888 if (--dqp
->q_nrefs
> 0)
891 xfs_qm_dqput_final(dqp
);
895 * Release a dquot. Flush it if dirty, then dqput() it.
896 * dquot must not be locked.
905 trace_xfs_dqrele(dqp
);
909 * We don't care to flush it if the dquot is dirty here.
910 * That will create stutters that we want to avoid.
911 * Instead we do a delayed write when we try to reclaim
912 * a dirty dquot. Also xfs_sync will take part of the burden...
918 * This is the dquot flushing I/O completion routine. It is called
919 * from interrupt level when the buffer containing the dquot is
920 * flushed to disk. It is responsible for removing the dquot logitem
921 * from the AIL if it has not been re-logged, and unlocking the dquot's
922 * flush lock. This behavior is very similar to that of inodes..
927 struct xfs_log_item
*lip
)
929 xfs_dq_logitem_t
*qip
= (struct xfs_dq_logitem
*)lip
;
930 xfs_dquot_t
*dqp
= qip
->qli_dquot
;
931 struct xfs_ail
*ailp
= lip
->li_ailp
;
934 * We only want to pull the item from the AIL if its
935 * location in the log has not changed since we started the flush.
936 * Thus, we only bother if the dquot's lsn has
937 * not changed. First we check the lsn outside the lock
938 * since it's cheaper, and then we recheck while
939 * holding the lock before removing the dquot from the AIL.
941 if ((lip
->li_flags
& XFS_LI_IN_AIL
) &&
942 lip
->li_lsn
== qip
->qli_flush_lsn
) {
944 /* xfs_trans_ail_delete() drops the AIL lock. */
945 spin_lock(&ailp
->xa_lock
);
946 if (lip
->li_lsn
== qip
->qli_flush_lsn
)
947 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_CORRUPT_INCORE
);
949 spin_unlock(&ailp
->xa_lock
);
953 * Release the dq's flush lock since we're done with it.
959 * Write a modified dquot to disk.
960 * The dquot must be locked and the flush lock too taken by caller.
961 * The flush lock will not be unlocked until the dquot reaches the disk,
962 * but the dquot is free to be unlocked and modified by the caller
963 * in the interim. Dquot is still locked on return. This behavior is
964 * identical to that of inodes.
968 struct xfs_dquot
*dqp
,
969 struct xfs_buf
**bpp
)
971 struct xfs_mount
*mp
= dqp
->q_mount
;
973 struct xfs_disk_dquot
*ddqp
;
976 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
977 ASSERT(!completion_done(&dqp
->q_flush
));
979 trace_xfs_dqflush(dqp
);
983 xfs_qm_dqunpin_wait(dqp
);
986 * This may have been unpinned because the filesystem is shutting
987 * down forcibly. If that's the case we must not write this dquot
988 * to disk, because the log record didn't make it to disk.
990 * We also have to remove the log item from the AIL in this case,
991 * as we wait for an emptry AIL as part of the unmount process.
993 if (XFS_FORCED_SHUTDOWN(mp
)) {
994 struct xfs_log_item
*lip
= &dqp
->q_logitem
.qli_item
;
995 dqp
->dq_flags
&= ~XFS_DQ_DIRTY
;
997 spin_lock(&mp
->m_ail
->xa_lock
);
998 if (lip
->li_flags
& XFS_LI_IN_AIL
)
999 xfs_trans_ail_delete(mp
->m_ail
, lip
,
1000 SHUTDOWN_CORRUPT_INCORE
);
1002 spin_unlock(&mp
->m_ail
->xa_lock
);
1003 error
= XFS_ERROR(EIO
);
1008 * Get the buffer containing the on-disk dquot
1010 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dqp
->q_blkno
,
1011 mp
->m_quotainfo
->qi_dqchunklen
, 0, &bp
, NULL
);
1016 * Calculate the location of the dquot inside the buffer.
1018 ddqp
= bp
->b_addr
+ dqp
->q_bufoffset
;
1021 * A simple sanity check in case we got a corrupted dquot..
1023 error
= xfs_qm_dqcheck(mp
, &dqp
->q_core
, be32_to_cpu(ddqp
->d_id
), 0,
1024 XFS_QMOPT_DOWARN
, "dqflush (incore copy)");
1028 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1029 return XFS_ERROR(EIO
);
1032 /* This is the only portion of data that needs to persist */
1033 memcpy(ddqp
, &dqp
->q_core
, sizeof(xfs_disk_dquot_t
));
1036 * Clear the dirty field and remember the flush lsn for later use.
1038 dqp
->dq_flags
&= ~XFS_DQ_DIRTY
;
1040 xfs_trans_ail_copy_lsn(mp
->m_ail
, &dqp
->q_logitem
.qli_flush_lsn
,
1041 &dqp
->q_logitem
.qli_item
.li_lsn
);
1044 * Attach an iodone routine so that we can remove this dquot from the
1045 * AIL and release the flush lock once the dquot is synced to disk.
1047 xfs_buf_attach_iodone(bp
, xfs_qm_dqflush_done
,
1048 &dqp
->q_logitem
.qli_item
);
1051 * If the buffer is pinned then push on the log so we won't
1052 * get stuck waiting in the write for too long.
1054 if (xfs_buf_ispinned(bp
)) {
1055 trace_xfs_dqflush_force(dqp
);
1056 xfs_log_force(mp
, 0);
1059 trace_xfs_dqflush_done(dqp
);
1065 return XFS_ERROR(EIO
);
1069 * Lock two xfs_dquot structures.
1071 * To avoid deadlocks we always lock the quota structure with
1072 * the lowerd id first.
1081 if (be32_to_cpu(d1
->q_core
.d_id
) >
1082 be32_to_cpu(d2
->q_core
.d_id
)) {
1083 mutex_lock(&d2
->q_qlock
);
1084 mutex_lock_nested(&d1
->q_qlock
, XFS_QLOCK_NESTED
);
1086 mutex_lock(&d1
->q_qlock
);
1087 mutex_lock_nested(&d2
->q_qlock
, XFS_QLOCK_NESTED
);
1090 mutex_lock(&d1
->q_qlock
);
1092 mutex_lock(&d2
->q_qlock
);
1100 kmem_zone_init(sizeof(struct xfs_dquot
), "xfs_dquot");
1105 kmem_zone_init(sizeof(struct xfs_dquot_acct
), "xfs_dqtrx");
1106 if (!xfs_qm_dqtrxzone
)
1107 goto out_free_dqzone
;
1112 kmem_zone_destroy(xfs_qm_dqzone
);
1120 kmem_zone_destroy(xfs_qm_dqtrxzone
);
1121 kmem_zone_destroy(xfs_qm_dqzone
);