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
23 #include "xfs_trans.h"
26 #include "xfs_alloc.h"
27 #include "xfs_quota.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_inode.h"
32 #include "xfs_rtalloc.h"
33 #include "xfs_error.h"
34 #include "xfs_itable.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_trans_space.h"
38 #include "xfs_trans_priv.h"
40 #include "xfs_trace.h"
47 dquot hash-chain lock (hashlock)
48 xqm dquot freelist lock (freelistlock
49 mount's dquot list lock (mplistlock)
50 user dquot lock - lock ordering among dquots is based on the uid or gid
51 group dquot lock - similar to udquots. Between the two dquots, the udquot
52 has to be locked first.
53 pin lock - the dquot lock must be held to take this lock.
58 xfs_buftarg_t
*xfs_dqerror_target
;
61 int xfs_dqerror_mod
= 33;
64 static struct lock_class_key xfs_dquot_other_class
;
67 * Allocate and initialize a dquot. We don't always allocate fresh memory;
68 * we try to reclaim a free dquot if the number of incore dquots are above
70 * The only field inside the core that gets initialized at this point
71 * is the d_id field. The idea is to fill in the entire q_core
72 * when we read in the on disk dquot.
81 boolean_t brandnewdquot
;
83 brandnewdquot
= xfs_qm_dqalloc_incore(&dqp
);
85 dqp
->q_core
.d_id
= cpu_to_be32(id
);
89 * No need to re-initialize these if this is a reclaimed dquot.
92 INIT_LIST_HEAD(&dqp
->q_freelist
);
93 mutex_init(&dqp
->q_qlock
);
94 init_waitqueue_head(&dqp
->q_pinwait
);
97 * Because we want to use a counting completion, complete
98 * the flush completion once to allow a single access to
99 * the flush completion without blocking.
101 init_completion(&dqp
->q_flush
);
102 complete(&dqp
->q_flush
);
104 trace_xfs_dqinit(dqp
);
107 * Only the q_core portion was zeroed in dqreclaim_one().
108 * So, we need to reset others.
112 INIT_LIST_HEAD(&dqp
->q_mplist
);
113 INIT_LIST_HEAD(&dqp
->q_hashlist
);
114 dqp
->q_bufoffset
= 0;
115 dqp
->q_fileoffset
= 0;
116 dqp
->q_transp
= NULL
;
117 dqp
->q_gdquot
= NULL
;
118 dqp
->q_res_bcount
= 0;
119 dqp
->q_res_icount
= 0;
120 dqp
->q_res_rtbcount
= 0;
121 atomic_set(&dqp
->q_pincount
, 0);
123 ASSERT(list_empty(&dqp
->q_freelist
));
125 trace_xfs_dqreuse(dqp
);
129 * In either case we need to make sure group quotas have a different
130 * lock class than user quotas, to make sure lockdep knows we can
131 * locks of one of each at the same time.
133 if (!(type
& XFS_DQ_USER
))
134 lockdep_set_class(&dqp
->q_qlock
, &xfs_dquot_other_class
);
137 * log item gets initialized later
143 * This is called to free all the memory associated with a dquot
149 ASSERT(list_empty(&dqp
->q_freelist
));
151 mutex_destroy(&dqp
->q_qlock
);
152 kmem_zone_free(xfs_Gqm
->qm_dqzone
, dqp
);
154 atomic_dec(&xfs_Gqm
->qm_totaldquots
);
158 * This is what a 'fresh' dquot inside a dquot chunk looks like on disk.
167 * Caller has zero'd the entire dquot 'chunk' already.
169 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
170 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
171 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
172 d
->dd_diskdq
.d_flags
= type
;
176 * If default limits are in force, push them into the dquot now.
177 * We overwrite the dquot limits only if they are zero and this
178 * is not the root dquot.
181 xfs_qm_adjust_dqlimits(
185 xfs_quotainfo_t
*q
= mp
->m_quotainfo
;
189 if (q
->qi_bsoftlimit
&& !d
->d_blk_softlimit
)
190 d
->d_blk_softlimit
= cpu_to_be64(q
->qi_bsoftlimit
);
191 if (q
->qi_bhardlimit
&& !d
->d_blk_hardlimit
)
192 d
->d_blk_hardlimit
= cpu_to_be64(q
->qi_bhardlimit
);
193 if (q
->qi_isoftlimit
&& !d
->d_ino_softlimit
)
194 d
->d_ino_softlimit
= cpu_to_be64(q
->qi_isoftlimit
);
195 if (q
->qi_ihardlimit
&& !d
->d_ino_hardlimit
)
196 d
->d_ino_hardlimit
= cpu_to_be64(q
->qi_ihardlimit
);
197 if (q
->qi_rtbsoftlimit
&& !d
->d_rtb_softlimit
)
198 d
->d_rtb_softlimit
= cpu_to_be64(q
->qi_rtbsoftlimit
);
199 if (q
->qi_rtbhardlimit
&& !d
->d_rtb_hardlimit
)
200 d
->d_rtb_hardlimit
= cpu_to_be64(q
->qi_rtbhardlimit
);
204 * Check the limits and timers of a dquot and start or reset timers
206 * This gets called even when quota enforcement is OFF, which makes our
207 * life a little less complicated. (We just don't reject any quota
208 * reservations in that case, when enforcement is off).
209 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
211 * In contrast, warnings are a little different in that they don't
212 * 'automatically' get started when limits get exceeded. They do
213 * get reset to zero, however, when we find the count to be under
214 * the soft limit (they are only ever set non-zero via userspace).
217 xfs_qm_adjust_dqtimers(
224 if (d
->d_blk_hardlimit
)
225 ASSERT(be64_to_cpu(d
->d_blk_softlimit
) <=
226 be64_to_cpu(d
->d_blk_hardlimit
));
227 if (d
->d_ino_hardlimit
)
228 ASSERT(be64_to_cpu(d
->d_ino_softlimit
) <=
229 be64_to_cpu(d
->d_ino_hardlimit
));
230 if (d
->d_rtb_hardlimit
)
231 ASSERT(be64_to_cpu(d
->d_rtb_softlimit
) <=
232 be64_to_cpu(d
->d_rtb_hardlimit
));
236 if ((d
->d_blk_softlimit
&&
237 (be64_to_cpu(d
->d_bcount
) >=
238 be64_to_cpu(d
->d_blk_softlimit
))) ||
239 (d
->d_blk_hardlimit
&&
240 (be64_to_cpu(d
->d_bcount
) >=
241 be64_to_cpu(d
->d_blk_hardlimit
)))) {
242 d
->d_btimer
= cpu_to_be32(get_seconds() +
243 mp
->m_quotainfo
->qi_btimelimit
);
248 if ((!d
->d_blk_softlimit
||
249 (be64_to_cpu(d
->d_bcount
) <
250 be64_to_cpu(d
->d_blk_softlimit
))) &&
251 (!d
->d_blk_hardlimit
||
252 (be64_to_cpu(d
->d_bcount
) <
253 be64_to_cpu(d
->d_blk_hardlimit
)))) {
259 if ((d
->d_ino_softlimit
&&
260 (be64_to_cpu(d
->d_icount
) >=
261 be64_to_cpu(d
->d_ino_softlimit
))) ||
262 (d
->d_ino_hardlimit
&&
263 (be64_to_cpu(d
->d_icount
) >=
264 be64_to_cpu(d
->d_ino_hardlimit
)))) {
265 d
->d_itimer
= cpu_to_be32(get_seconds() +
266 mp
->m_quotainfo
->qi_itimelimit
);
271 if ((!d
->d_ino_softlimit
||
272 (be64_to_cpu(d
->d_icount
) <
273 be64_to_cpu(d
->d_ino_softlimit
))) &&
274 (!d
->d_ino_hardlimit
||
275 (be64_to_cpu(d
->d_icount
) <
276 be64_to_cpu(d
->d_ino_hardlimit
)))) {
281 if (!d
->d_rtbtimer
) {
282 if ((d
->d_rtb_softlimit
&&
283 (be64_to_cpu(d
->d_rtbcount
) >=
284 be64_to_cpu(d
->d_rtb_softlimit
))) ||
285 (d
->d_rtb_hardlimit
&&
286 (be64_to_cpu(d
->d_rtbcount
) >=
287 be64_to_cpu(d
->d_rtb_hardlimit
)))) {
288 d
->d_rtbtimer
= cpu_to_be32(get_seconds() +
289 mp
->m_quotainfo
->qi_rtbtimelimit
);
294 if ((!d
->d_rtb_softlimit
||
295 (be64_to_cpu(d
->d_rtbcount
) <
296 be64_to_cpu(d
->d_rtb_softlimit
))) &&
297 (!d
->d_rtb_hardlimit
||
298 (be64_to_cpu(d
->d_rtbcount
) <
299 be64_to_cpu(d
->d_rtb_hardlimit
)))) {
306 * initialize a buffer full of dquots and log the whole thing
309 xfs_qm_init_dquot_blk(
316 struct xfs_quotainfo
*q
= mp
->m_quotainfo
;
321 ASSERT(xfs_buf_islocked(bp
));
326 * ID of the first dquot in the block - id's are zero based.
328 curid
= id
- (id
% q
->qi_dqperchunk
);
330 memset(d
, 0, BBTOB(q
->qi_dqchunklen
));
331 for (i
= 0; i
< q
->qi_dqperchunk
; i
++, d
++, curid
++)
332 xfs_qm_dqinit_core(curid
, type
, d
);
333 xfs_trans_dquot_buf(tp
, bp
,
334 (type
& XFS_DQ_USER
? XFS_BLF_UDQUOT_BUF
:
335 ((type
& XFS_DQ_PROJ
) ? XFS_BLF_PDQUOT_BUF
:
336 XFS_BLF_GDQUOT_BUF
)));
337 xfs_trans_log_buf(tp
, bp
, 0, BBTOB(q
->qi_dqchunklen
) - 1);
343 * Allocate a block and fill it with dquots.
344 * This is called when the bmapi finds a hole.
352 xfs_fileoff_t offset_fsb
,
355 xfs_fsblock_t firstblock
;
356 xfs_bmap_free_t flist
;
358 int nmaps
, error
, committed
;
360 xfs_trans_t
*tp
= *tpp
;
364 trace_xfs_dqalloc(dqp
);
367 * Initialize the bmap freelist prior to calling bmapi code.
369 xfs_bmap_init(&flist
, &firstblock
);
370 xfs_ilock(quotip
, XFS_ILOCK_EXCL
);
372 * Return if this type of quotas is turned off while we didn't
375 if (XFS_IS_THIS_QUOTA_OFF(dqp
)) {
376 xfs_iunlock(quotip
, XFS_ILOCK_EXCL
);
380 xfs_trans_ijoin(tp
, quotip
, XFS_ILOCK_EXCL
);
382 error
= xfs_bmapi_write(tp
, quotip
, offset_fsb
,
383 XFS_DQUOT_CLUSTER_SIZE_FSB
, XFS_BMAPI_METADATA
,
384 &firstblock
, XFS_QM_DQALLOC_SPACE_RES(mp
),
385 &map
, &nmaps
, &flist
);
388 ASSERT(map
.br_blockcount
== XFS_DQUOT_CLUSTER_SIZE_FSB
);
390 ASSERT((map
.br_startblock
!= DELAYSTARTBLOCK
) &&
391 (map
.br_startblock
!= HOLESTARTBLOCK
));
394 * Keep track of the blkno to save a lookup later
396 dqp
->q_blkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
);
398 /* now we can just get the buffer (there's nothing to read yet) */
399 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
401 mp
->m_quotainfo
->qi_dqchunklen
,
404 error
= xfs_buf_geterror(bp
);
409 * Make a chunk of dquots out of this buffer and log
412 xfs_qm_init_dquot_blk(tp
, mp
, be32_to_cpu(dqp
->q_core
.d_id
),
413 dqp
->dq_flags
& XFS_DQ_ALLTYPES
, bp
);
416 * xfs_bmap_finish() may commit the current transaction and
417 * start a second transaction if the freelist is not empty.
419 * Since we still want to modify this buffer, we need to
420 * ensure that the buffer is not released on commit of
421 * the first transaction and ensure the buffer is added to the
422 * second transaction.
424 * If there is only one transaction then don't stop the buffer
425 * from being released when it commits later on.
428 xfs_trans_bhold(tp
, bp
);
430 if ((error
= xfs_bmap_finish(tpp
, &flist
, &committed
))) {
436 xfs_trans_bjoin(tp
, bp
);
438 xfs_trans_bhold_release(tp
, bp
);
445 xfs_bmap_cancel(&flist
);
447 xfs_iunlock(quotip
, XFS_ILOCK_EXCL
);
453 * Maps a dquot to the buffer containing its on-disk version.
454 * This returns a ptr to the buffer containing the on-disk dquot
455 * in the bpp param, and a ptr to the on-disk dquot within that buffer
461 xfs_disk_dquot_t
**O_ddpp
,
466 int nmaps
= 1, error
;
468 xfs_inode_t
*quotip
= XFS_DQ_TO_QIP(dqp
);
469 xfs_mount_t
*mp
= dqp
->q_mount
;
470 xfs_disk_dquot_t
*ddq
;
471 xfs_dqid_t id
= be32_to_cpu(dqp
->q_core
.d_id
);
472 xfs_trans_t
*tp
= (tpp
? *tpp
: NULL
);
474 dqp
->q_fileoffset
= (xfs_fileoff_t
)id
/ mp
->m_quotainfo
->qi_dqperchunk
;
476 xfs_ilock(quotip
, XFS_ILOCK_SHARED
);
477 if (XFS_IS_THIS_QUOTA_OFF(dqp
)) {
479 * Return if this type of quotas is turned off while we
480 * didn't have the quota inode lock.
482 xfs_iunlock(quotip
, XFS_ILOCK_SHARED
);
487 * Find the block map; no allocations yet
489 error
= xfs_bmapi_read(quotip
, dqp
->q_fileoffset
,
490 XFS_DQUOT_CLUSTER_SIZE_FSB
, &map
, &nmaps
, 0);
492 xfs_iunlock(quotip
, XFS_ILOCK_SHARED
);
497 ASSERT(map
.br_blockcount
== 1);
500 * Offset of dquot in the (fixed sized) dquot chunk.
502 dqp
->q_bufoffset
= (id
% mp
->m_quotainfo
->qi_dqperchunk
) *
505 ASSERT(map
.br_startblock
!= DELAYSTARTBLOCK
);
506 if (map
.br_startblock
== HOLESTARTBLOCK
) {
508 * We don't allocate unless we're asked to
510 if (!(flags
& XFS_QMOPT_DQALLOC
))
514 error
= xfs_qm_dqalloc(tpp
, mp
, dqp
, quotip
,
515 dqp
->q_fileoffset
, &bp
);
520 trace_xfs_dqtobp_read(dqp
);
523 * store the blkno etc so that we don't have to do the
524 * mapping all the time
526 dqp
->q_blkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
);
528 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
530 mp
->m_quotainfo
->qi_dqchunklen
,
533 return XFS_ERROR(error
);
536 ASSERT(xfs_buf_islocked(bp
));
539 * calculate the location of the dquot inside the buffer.
541 ddq
= bp
->b_addr
+ dqp
->q_bufoffset
;
544 * A simple sanity check in case we got a corrupted dquot...
546 error
= xfs_qm_dqcheck(mp
, ddq
, id
, dqp
->dq_flags
& XFS_DQ_ALLTYPES
,
547 flags
& (XFS_QMOPT_DQREPAIR
|XFS_QMOPT_DOWARN
),
550 if (!(flags
& XFS_QMOPT_DQREPAIR
)) {
551 xfs_trans_brelse(tp
, bp
);
552 return XFS_ERROR(EIO
);
564 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
565 * and release the buffer immediately.
573 xfs_dquot_t
*dqp
, /* dquot to get filled in */
576 xfs_disk_dquot_t
*ddqp
;
583 trace_xfs_dqread(dqp
);
586 * get a pointer to the on-disk dquot and the buffer containing it
587 * dqp already knows its own type (GROUP/USER).
589 if ((error
= xfs_qm_dqtobp(tpp
, dqp
, &ddqp
, &bp
, flags
))) {
594 /* copy everything from disk dquot to the incore dquot */
595 memcpy(&dqp
->q_core
, ddqp
, sizeof(xfs_disk_dquot_t
));
596 ASSERT(be32_to_cpu(dqp
->q_core
.d_id
) == id
);
597 xfs_qm_dquot_logitem_init(dqp
);
600 * Reservation counters are defined as reservation plus current usage
601 * to avoid having to add every time.
603 dqp
->q_res_bcount
= be64_to_cpu(ddqp
->d_bcount
);
604 dqp
->q_res_icount
= be64_to_cpu(ddqp
->d_icount
);
605 dqp
->q_res_rtbcount
= be64_to_cpu(ddqp
->d_rtbcount
);
607 /* Mark the buf so that this will stay incore a little longer */
608 xfs_buf_set_ref(bp
, XFS_DQUOT_REF
);
611 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
612 * So we need to release with xfs_trans_brelse().
613 * The strategy here is identical to that of inodes; we lock
614 * the dquot in xfs_qm_dqget() before making it accessible to
615 * others. This is because dquots, like inodes, need a good level of
616 * concurrency, and we don't want to take locks on the entire buffers
617 * for dquot accesses.
618 * Note also that the dquot buffer may even be dirty at this point, if
619 * this particular dquot was repaired. We still aren't afraid to
620 * brelse it because we have the changes incore.
622 ASSERT(xfs_buf_islocked(bp
));
623 xfs_trans_brelse(tp
, bp
);
630 * allocate an incore dquot from the kernel heap,
631 * and fill its core with quota information kept on disk.
632 * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk
633 * if it wasn't already allocated.
638 xfs_dqid_t id
, /* gid or uid, depending on type */
639 uint type
, /* UDQUOT or GDQUOT */
640 uint flags
, /* DQALLOC, DQREPAIR */
641 xfs_dquot_t
**O_dqpp
)/* OUT : incore dquot, not locked */
648 dqp
= xfs_qm_dqinit(mp
, id
, type
);
650 if (flags
& XFS_QMOPT_DQALLOC
) {
651 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_DQALLOC
);
652 error
= xfs_trans_reserve(tp
, XFS_QM_DQALLOC_SPACE_RES(mp
),
653 XFS_WRITE_LOG_RES(mp
) +
654 BBTOB(mp
->m_quotainfo
->qi_dqchunklen
) - 1 +
657 XFS_TRANS_PERM_LOG_RES
,
658 XFS_WRITE_LOG_COUNT
);
663 cancelflags
= XFS_TRANS_RELEASE_LOG_RES
;
667 * Read it from disk; xfs_dqread() takes care of
668 * all the necessary initialization of dquot's fields (locks, etc)
670 if ((error
= xfs_qm_dqread(&tp
, id
, dqp
, flags
))) {
672 * This can happen if quotas got turned off (ESRCH),
673 * or if the dquot didn't exist on disk and we ask to
676 trace_xfs_dqread_fail(dqp
);
677 cancelflags
|= XFS_TRANS_ABORT
;
681 if ((error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
)))
691 xfs_trans_cancel(tp
, cancelflags
);
693 xfs_qm_dqdestroy(dqp
);
699 * Lookup a dquot in the incore dquot hashtable. We keep two separate
700 * hashtables for user and group dquots; and, these are global tables
701 * inside the XQM, not per-filesystem tables.
702 * The hash chain must be locked by caller, and it is left locked
703 * on return. Returning dquot is locked.
710 xfs_dquot_t
**O_dqpp
)
715 ASSERT(mutex_is_locked(&qh
->qh_lock
));
717 flist_locked
= B_FALSE
;
720 * Traverse the hashchain looking for a match
722 list_for_each_entry(dqp
, &qh
->qh_list
, q_hashlist
) {
724 * We already have the hashlock. We don't need the
725 * dqlock to look at the id field of the dquot, since the
726 * id can't be modified without the hashlock anyway.
728 if (be32_to_cpu(dqp
->q_core
.d_id
) == id
&& dqp
->q_mount
== mp
) {
729 trace_xfs_dqlookup_found(dqp
);
732 * All in core dquots must be on the dqlist of mp
734 ASSERT(!list_empty(&dqp
->q_mplist
));
737 if (dqp
->q_nrefs
== 0) {
738 ASSERT(!list_empty(&dqp
->q_freelist
));
739 if (!mutex_trylock(&xfs_Gqm
->qm_dqfrlist_lock
)) {
740 trace_xfs_dqlookup_want(dqp
);
743 * We may have raced with dqreclaim_one()
744 * (and lost). So, flag that we don't
745 * want the dquot to be reclaimed.
747 dqp
->dq_flags
|= XFS_DQ_WANT
;
749 mutex_lock(&xfs_Gqm
->qm_dqfrlist_lock
);
751 dqp
->dq_flags
&= ~(XFS_DQ_WANT
);
753 flist_locked
= B_TRUE
;
757 * id couldn't have changed; we had the hashlock all
760 ASSERT(be32_to_cpu(dqp
->q_core
.d_id
) == id
);
763 if (dqp
->q_nrefs
!= 0) {
764 mutex_unlock(&xfs_Gqm
->qm_dqfrlist_lock
);
765 flist_locked
= B_FALSE
;
767 /* take it off the freelist */
768 trace_xfs_dqlookup_freelist(dqp
);
769 list_del_init(&dqp
->q_freelist
);
770 xfs_Gqm
->qm_dqfrlist_cnt
--;
777 mutex_unlock(&xfs_Gqm
->qm_dqfrlist_lock
);
779 * move the dquot to the front of the hashchain
781 ASSERT(mutex_is_locked(&qh
->qh_lock
));
782 list_move(&dqp
->q_hashlist
, &qh
->qh_list
);
783 trace_xfs_dqlookup_done(dqp
);
790 ASSERT(mutex_is_locked(&qh
->qh_lock
));
795 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
796 * a locked dquot, doing an allocation (if requested) as needed.
797 * When both an inode and an id are given, the inode's id takes precedence.
798 * That is, if the id changes while we don't hold the ilock inside this
799 * function, the new dquot is returned, not necessarily the one requested
800 * in the id argument.
805 xfs_inode_t
*ip
, /* locked inode (optional) */
806 xfs_dqid_t id
, /* uid/projid/gid depending on type */
807 uint type
, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
808 uint flags
, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
809 xfs_dquot_t
**O_dqpp
) /* OUT : locked incore dquot */
816 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
817 if ((! XFS_IS_UQUOTA_ON(mp
) && type
== XFS_DQ_USER
) ||
818 (! XFS_IS_PQUOTA_ON(mp
) && type
== XFS_DQ_PROJ
) ||
819 (! XFS_IS_GQUOTA_ON(mp
) && type
== XFS_DQ_GROUP
)) {
822 h
= XFS_DQ_HASH(mp
, id
, type
);
825 if (xfs_do_dqerror
) {
826 if ((xfs_dqerror_target
== mp
->m_ddev_targp
) &&
827 (xfs_dqreq_num
++ % xfs_dqerror_mod
) == 0) {
828 xfs_debug(mp
, "Returning error in dqget");
837 ASSERT(type
== XFS_DQ_USER
||
838 type
== XFS_DQ_PROJ
||
839 type
== XFS_DQ_GROUP
);
841 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
842 if (type
== XFS_DQ_USER
)
843 ASSERT(ip
->i_udquot
== NULL
);
845 ASSERT(ip
->i_gdquot
== NULL
);
848 mutex_lock(&h
->qh_lock
);
851 * Look in the cache (hashtable).
852 * The chain is kept locked during lookup.
854 if (xfs_qm_dqlookup(mp
, id
, h
, O_dqpp
) == 0) {
855 XQM_STATS_INC(xqmstats
.xs_qm_dqcachehits
);
857 * The dquot was found, moved to the front of the chain,
858 * taken off the freelist if it was on it, and locked
859 * at this point. Just unlock the hashchain and return.
862 ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp
));
863 mutex_unlock(&h
->qh_lock
);
864 trace_xfs_dqget_hit(*O_dqpp
);
865 return (0); /* success */
867 XQM_STATS_INC(xqmstats
.xs_qm_dqcachemisses
);
870 * Dquot cache miss. We don't want to keep the inode lock across
871 * a (potential) disk read. Also we don't want to deal with the lock
872 * ordering between quotainode and this inode. OTOH, dropping the inode
873 * lock here means dealing with a chown that can happen before
874 * we re-acquire the lock.
877 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
879 * Save the hashchain version stamp, and unlock the chain, so that
880 * we don't keep the lock across a disk read
882 version
= h
->qh_version
;
883 mutex_unlock(&h
->qh_lock
);
886 * Allocate the dquot on the kernel heap, and read the ondisk
887 * portion off the disk. Also, do all the necessary initialization
888 * This can return ENOENT if dquot didn't exist on disk and we didn't
889 * ask it to allocate; ESRCH if quotas got turned off suddenly.
891 if ((error
= xfs_qm_idtodq(mp
, id
, type
,
892 flags
& (XFS_QMOPT_DQALLOC
|XFS_QMOPT_DQREPAIR
|
896 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
901 * See if this is mount code calling to look at the overall quota limits
902 * which are stored in the id == 0 user or group's dquot.
903 * Since we may not have done a quotacheck by this point, just return
904 * the dquot without attaching it to any hashtables, lists, etc, or even
905 * taking a reference.
906 * The caller must dqdestroy this once done.
908 if (flags
& XFS_QMOPT_DQSUSER
) {
915 * Dquot lock comes after hashlock in the lock ordering
918 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
921 * A dquot could be attached to this inode by now, since
922 * we had dropped the ilock.
924 if (type
== XFS_DQ_USER
) {
925 if (!XFS_IS_UQUOTA_ON(mp
)) {
926 /* inode stays locked on return */
927 xfs_qm_dqdestroy(dqp
);
928 return XFS_ERROR(ESRCH
);
931 xfs_qm_dqdestroy(dqp
);
937 if (!XFS_IS_OQUOTA_ON(mp
)) {
938 /* inode stays locked on return */
939 xfs_qm_dqdestroy(dqp
);
940 return XFS_ERROR(ESRCH
);
943 xfs_qm_dqdestroy(dqp
);
952 * Hashlock comes after ilock in lock order
954 mutex_lock(&h
->qh_lock
);
955 if (version
!= h
->qh_version
) {
958 * Now, see if somebody else put the dquot in the
959 * hashtable before us. This can happen because we didn't
960 * keep the hashchain lock. We don't have to worry about
961 * lock order between the two dquots here since dqp isn't
962 * on any findable lists yet.
964 if (xfs_qm_dqlookup(mp
, id
, h
, &tmpdqp
) == 0) {
966 * Duplicate found. Just throw away the new dquot
969 xfs_qm_dqput(tmpdqp
);
970 mutex_unlock(&h
->qh_lock
);
971 xfs_qm_dqdestroy(dqp
);
972 XQM_STATS_INC(xqmstats
.xs_qm_dquot_dups
);
978 * Put the dquot at the beginning of the hash-chain and mp's list
979 * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock ..
981 ASSERT(mutex_is_locked(&h
->qh_lock
));
983 list_add(&dqp
->q_hashlist
, &h
->qh_list
);
987 * Attach this dquot to this filesystem's list of all dquots,
988 * kept inside the mount structure in m_quotainfo field
990 mutex_lock(&mp
->m_quotainfo
->qi_dqlist_lock
);
993 * We return a locked dquot to the caller, with a reference taken
998 list_add(&dqp
->q_mplist
, &mp
->m_quotainfo
->qi_dqlist
);
999 mp
->m_quotainfo
->qi_dquots
++;
1000 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
1001 mutex_unlock(&h
->qh_lock
);
1003 ASSERT((ip
== NULL
) || xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1004 trace_xfs_dqget_miss(dqp
);
1011 * Release a reference to the dquot (decrement ref-count)
1012 * and unlock it. If there is a group quota attached to this
1013 * dquot, carefully release that too without tripping over
1014 * deadlocks'n'stuff.
1022 ASSERT(dqp
->q_nrefs
> 0);
1023 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1025 trace_xfs_dqput(dqp
);
1027 if (dqp
->q_nrefs
!= 1) {
1034 * drop the dqlock and acquire the freelist and dqlock
1035 * in the right order; but try to get it out-of-order first
1037 if (!mutex_trylock(&xfs_Gqm
->qm_dqfrlist_lock
)) {
1038 trace_xfs_dqput_wait(dqp
);
1040 mutex_lock(&xfs_Gqm
->qm_dqfrlist_lock
);
1047 /* We can't depend on nrefs being == 1 here */
1048 if (--dqp
->q_nrefs
== 0) {
1049 trace_xfs_dqput_free(dqp
);
1051 list_add_tail(&dqp
->q_freelist
, &xfs_Gqm
->qm_dqfrlist
);
1052 xfs_Gqm
->qm_dqfrlist_cnt
++;
1055 * If we just added a udquot to the freelist, then
1056 * we want to release the gdquot reference that
1057 * it (probably) has. Otherwise it'll keep the
1058 * gdquot from getting reclaimed.
1060 if ((gdqp
= dqp
->q_gdquot
)) {
1062 * Avoid a recursive dqput call
1065 dqp
->q_gdquot
= NULL
;
1071 * If we had a group quota inside the user quota as a hint,
1078 mutex_unlock(&xfs_Gqm
->qm_dqfrlist_lock
);
1082 * Release a dquot. Flush it if dirty, then dqput() it.
1083 * dquot must not be locked.
1092 trace_xfs_dqrele(dqp
);
1096 * We don't care to flush it if the dquot is dirty here.
1097 * That will create stutters that we want to avoid.
1098 * Instead we do a delayed write when we try to reclaim
1099 * a dirty dquot. Also xfs_sync will take part of the burden...
1105 * This is the dquot flushing I/O completion routine. It is called
1106 * from interrupt level when the buffer containing the dquot is
1107 * flushed to disk. It is responsible for removing the dquot logitem
1108 * from the AIL if it has not been re-logged, and unlocking the dquot's
1109 * flush lock. This behavior is very similar to that of inodes..
1112 xfs_qm_dqflush_done(
1114 struct xfs_log_item
*lip
)
1116 xfs_dq_logitem_t
*qip
= (struct xfs_dq_logitem
*)lip
;
1117 xfs_dquot_t
*dqp
= qip
->qli_dquot
;
1118 struct xfs_ail
*ailp
= lip
->li_ailp
;
1121 * We only want to pull the item from the AIL if its
1122 * location in the log has not changed since we started the flush.
1123 * Thus, we only bother if the dquot's lsn has
1124 * not changed. First we check the lsn outside the lock
1125 * since it's cheaper, and then we recheck while
1126 * holding the lock before removing the dquot from the AIL.
1128 if ((lip
->li_flags
& XFS_LI_IN_AIL
) &&
1129 lip
->li_lsn
== qip
->qli_flush_lsn
) {
1131 /* xfs_trans_ail_delete() drops the AIL lock. */
1132 spin_lock(&ailp
->xa_lock
);
1133 if (lip
->li_lsn
== qip
->qli_flush_lsn
)
1134 xfs_trans_ail_delete(ailp
, lip
);
1136 spin_unlock(&ailp
->xa_lock
);
1140 * Release the dq's flush lock since we're done with it.
1146 * Write a modified dquot to disk.
1147 * The dquot must be locked and the flush lock too taken by caller.
1148 * The flush lock will not be unlocked until the dquot reaches the disk,
1149 * but the dquot is free to be unlocked and modified by the caller
1150 * in the interim. Dquot is still locked on return. This behavior is
1151 * identical to that of inodes.
1158 struct xfs_mount
*mp
= dqp
->q_mount
;
1160 struct xfs_disk_dquot
*ddqp
;
1163 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1164 ASSERT(!completion_done(&dqp
->q_flush
));
1166 trace_xfs_dqflush(dqp
);
1169 * If not dirty, or it's pinned and we are not supposed to block, nada.
1171 if (!XFS_DQ_IS_DIRTY(dqp
) ||
1172 (!(flags
& SYNC_WAIT
) && atomic_read(&dqp
->q_pincount
) > 0)) {
1176 xfs_qm_dqunpin_wait(dqp
);
1179 * This may have been unpinned because the filesystem is shutting
1180 * down forcibly. If that's the case we must not write this dquot
1181 * to disk, because the log record didn't make it to disk!
1183 if (XFS_FORCED_SHUTDOWN(mp
)) {
1184 dqp
->dq_flags
&= ~XFS_DQ_DIRTY
;
1186 return XFS_ERROR(EIO
);
1190 * Get the buffer containing the on-disk dquot
1192 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dqp
->q_blkno
,
1193 mp
->m_quotainfo
->qi_dqchunklen
, 0, &bp
);
1195 ASSERT(error
!= ENOENT
);
1201 * Calculate the location of the dquot inside the buffer.
1203 ddqp
= bp
->b_addr
+ dqp
->q_bufoffset
;
1206 * A simple sanity check in case we got a corrupted dquot..
1208 error
= xfs_qm_dqcheck(mp
, &dqp
->q_core
, be32_to_cpu(ddqp
->d_id
), 0,
1209 XFS_QMOPT_DOWARN
, "dqflush (incore copy)");
1213 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1214 return XFS_ERROR(EIO
);
1217 /* This is the only portion of data that needs to persist */
1218 memcpy(ddqp
, &dqp
->q_core
, sizeof(xfs_disk_dquot_t
));
1221 * Clear the dirty field and remember the flush lsn for later use.
1223 dqp
->dq_flags
&= ~XFS_DQ_DIRTY
;
1225 xfs_trans_ail_copy_lsn(mp
->m_ail
, &dqp
->q_logitem
.qli_flush_lsn
,
1226 &dqp
->q_logitem
.qli_item
.li_lsn
);
1229 * Attach an iodone routine so that we can remove this dquot from the
1230 * AIL and release the flush lock once the dquot is synced to disk.
1232 xfs_buf_attach_iodone(bp
, xfs_qm_dqflush_done
,
1233 &dqp
->q_logitem
.qli_item
);
1236 * If the buffer is pinned then push on the log so we won't
1237 * get stuck waiting in the write for too long.
1239 if (xfs_buf_ispinned(bp
)) {
1240 trace_xfs_dqflush_force(dqp
);
1241 xfs_log_force(mp
, 0);
1244 if (flags
& SYNC_WAIT
)
1245 error
= xfs_bwrite(bp
);
1247 xfs_buf_delwri_queue(bp
);
1251 trace_xfs_dqflush_done(dqp
);
1254 * dqp is still locked, but caller is free to unlock it now.
1261 xfs_qm_dqlock_nowait(
1264 return mutex_trylock(&dqp
->q_qlock
);
1271 mutex_lock(&dqp
->q_qlock
);
1278 mutex_unlock(&(dqp
->q_qlock
));
1279 if (dqp
->q_logitem
.qli_dquot
== dqp
) {
1280 /* Once was dqp->q_mount, but might just have been cleared */
1281 xfs_trans_unlocked_item(dqp
->q_logitem
.qli_item
.li_ailp
,
1282 (xfs_log_item_t
*)&(dqp
->q_logitem
));
1288 xfs_dqunlock_nonotify(
1291 mutex_unlock(&(dqp
->q_qlock
));
1295 * Lock two xfs_dquot structures.
1297 * To avoid deadlocks we always lock the quota structure with
1298 * the lowerd id first.
1307 if (be32_to_cpu(d1
->q_core
.d_id
) >
1308 be32_to_cpu(d2
->q_core
.d_id
)) {
1309 mutex_lock(&d2
->q_qlock
);
1310 mutex_lock_nested(&d1
->q_qlock
, XFS_QLOCK_NESTED
);
1312 mutex_lock(&d1
->q_qlock
);
1313 mutex_lock_nested(&d2
->q_qlock
, XFS_QLOCK_NESTED
);
1316 mutex_lock(&d1
->q_qlock
);
1318 mutex_lock(&d2
->q_qlock
);
1324 * Take a dquot out of the mount's dqlist as well as the hashlist.
1325 * This is called via unmount as well as quotaoff, and the purge
1326 * will always succeed unless there are soft (temp) references
1329 * This returns 0 if it was purged, 1 if it wasn't. It's not an error code
1330 * that we're returning! XXXsup - not cool.
1337 xfs_dqhash_t
*qh
= dqp
->q_hash
;
1338 xfs_mount_t
*mp
= dqp
->q_mount
;
1340 ASSERT(mutex_is_locked(&mp
->m_quotainfo
->qi_dqlist_lock
));
1341 ASSERT(mutex_is_locked(&dqp
->q_hash
->qh_lock
));
1345 * We really can't afford to purge a dquot that is
1346 * referenced, because these are hard refs.
1347 * It shouldn't happen in general because we went thru _all_ inodes in
1348 * dqrele_all_inodes before calling this and didn't let the mountlock go.
1349 * However it is possible that we have dquots with temporary
1350 * references that are not attached to an inode. e.g. see xfs_setattr().
1352 if (dqp
->q_nrefs
!= 0) {
1354 mutex_unlock(&dqp
->q_hash
->qh_lock
);
1358 ASSERT(!list_empty(&dqp
->q_freelist
));
1361 * If we're turning off quotas, we have to make sure that, for
1362 * example, we don't delete quota disk blocks while dquots are
1363 * in the process of getting written to those disk blocks.
1364 * This dquot might well be on AIL, and we can't leave it there
1365 * if we're turning off quotas. Basically, we need this flush
1366 * lock, and are willing to block on it.
1368 if (!xfs_dqflock_nowait(dqp
)) {
1370 * Block on the flush lock after nudging dquot buffer,
1373 xfs_qm_dqflock_pushbuf_wait(dqp
);
1377 * XXXIf we're turning this type of quotas off, we don't care
1378 * about the dirty metadata sitting in this dquot. OTOH, if
1379 * we're unmounting, we do care, so we flush it and wait.
1381 if (XFS_DQ_IS_DIRTY(dqp
)) {
1384 /* dqflush unlocks dqflock */
1386 * Given that dqpurge is a very rare occurrence, it is OK
1387 * that we're holding the hashlist and mplist locks
1388 * across the disk write. But, ... XXXsup
1390 * We don't care about getting disk errors here. We need
1391 * to purge this dquot anyway, so we go ahead regardless.
1393 error
= xfs_qm_dqflush(dqp
, SYNC_WAIT
);
1395 xfs_warn(mp
, "%s: dquot %p flush failed",
1399 ASSERT(atomic_read(&dqp
->q_pincount
) == 0);
1400 ASSERT(XFS_FORCED_SHUTDOWN(mp
) ||
1401 !(dqp
->q_logitem
.qli_item
.li_flags
& XFS_LI_IN_AIL
));
1403 list_del_init(&dqp
->q_hashlist
);
1405 list_del_init(&dqp
->q_mplist
);
1406 mp
->m_quotainfo
->qi_dqreclaims
++;
1407 mp
->m_quotainfo
->qi_dquots
--;
1409 * XXX Move this to the front of the freelist, if we can get the
1412 ASSERT(!list_empty(&dqp
->q_freelist
));
1414 dqp
->q_mount
= NULL
;
1416 dqp
->dq_flags
= XFS_DQ_INACTIVE
;
1417 memset(&dqp
->q_core
, 0, sizeof(dqp
->q_core
));
1420 mutex_unlock(&qh
->qh_lock
);
1426 * Give the buffer a little push if it is incore and
1427 * wait on the flush lock.
1430 xfs_qm_dqflock_pushbuf_wait(
1433 xfs_mount_t
*mp
= dqp
->q_mount
;
1437 * Check to see if the dquot has been flushed delayed
1438 * write. If so, grab its buffer and send it
1439 * out immediately. We'll be able to acquire
1440 * the flush lock when the I/O completes.
1442 bp
= xfs_incore(mp
->m_ddev_targp
, dqp
->q_blkno
,
1443 mp
->m_quotainfo
->qi_dqchunklen
, XBF_TRYLOCK
);
1447 if (XFS_BUF_ISDELAYWRITE(bp
)) {
1448 if (xfs_buf_ispinned(bp
))
1449 xfs_log_force(mp
, 0);
1450 xfs_buf_delwri_promote(bp
);
1451 wake_up_process(bp
->b_target
->bt_task
);