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 sv_destroy(&dqp
->q_pinwait
);
153 kmem_zone_free(xfs_Gqm
->qm_dqzone
, dqp
);
155 atomic_dec(&xfs_Gqm
->qm_totaldquots
);
159 * This is what a 'fresh' dquot inside a dquot chunk looks like on disk.
168 * Caller has zero'd the entire dquot 'chunk' already.
170 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
171 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
172 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
173 d
->dd_diskdq
.d_flags
= type
;
177 * If default limits are in force, push them into the dquot now.
178 * We overwrite the dquot limits only if they are zero and this
179 * is not the root dquot.
182 xfs_qm_adjust_dqlimits(
186 xfs_quotainfo_t
*q
= mp
->m_quotainfo
;
190 if (q
->qi_bsoftlimit
&& !d
->d_blk_softlimit
)
191 d
->d_blk_softlimit
= cpu_to_be64(q
->qi_bsoftlimit
);
192 if (q
->qi_bhardlimit
&& !d
->d_blk_hardlimit
)
193 d
->d_blk_hardlimit
= cpu_to_be64(q
->qi_bhardlimit
);
194 if (q
->qi_isoftlimit
&& !d
->d_ino_softlimit
)
195 d
->d_ino_softlimit
= cpu_to_be64(q
->qi_isoftlimit
);
196 if (q
->qi_ihardlimit
&& !d
->d_ino_hardlimit
)
197 d
->d_ino_hardlimit
= cpu_to_be64(q
->qi_ihardlimit
);
198 if (q
->qi_rtbsoftlimit
&& !d
->d_rtb_softlimit
)
199 d
->d_rtb_softlimit
= cpu_to_be64(q
->qi_rtbsoftlimit
);
200 if (q
->qi_rtbhardlimit
&& !d
->d_rtb_hardlimit
)
201 d
->d_rtb_hardlimit
= cpu_to_be64(q
->qi_rtbhardlimit
);
205 * Check the limits and timers of a dquot and start or reset timers
207 * This gets called even when quota enforcement is OFF, which makes our
208 * life a little less complicated. (We just don't reject any quota
209 * reservations in that case, when enforcement is off).
210 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
212 * In contrast, warnings are a little different in that they don't
213 * 'automatically' get started when limits get exceeded. They do
214 * get reset to zero, however, when we find the count to be under
215 * the soft limit (they are only ever set non-zero via userspace).
218 xfs_qm_adjust_dqtimers(
225 if (d
->d_blk_hardlimit
)
226 ASSERT(be64_to_cpu(d
->d_blk_softlimit
) <=
227 be64_to_cpu(d
->d_blk_hardlimit
));
228 if (d
->d_ino_hardlimit
)
229 ASSERT(be64_to_cpu(d
->d_ino_softlimit
) <=
230 be64_to_cpu(d
->d_ino_hardlimit
));
231 if (d
->d_rtb_hardlimit
)
232 ASSERT(be64_to_cpu(d
->d_rtb_softlimit
) <=
233 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_ISBUSY(bp
));
322 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
324 d
= (xfs_dqblk_t
*)XFS_BUF_PTR(bp
);
327 * ID of the first dquot in the block - id's are zero based.
329 curid
= id
- (id
% q
->qi_dqperchunk
);
331 memset(d
, 0, BBTOB(q
->qi_dqchunklen
));
332 for (i
= 0; i
< q
->qi_dqperchunk
; i
++, d
++, curid
++)
333 xfs_qm_dqinit_core(curid
, type
, d
);
334 xfs_trans_dquot_buf(tp
, bp
,
335 (type
& XFS_DQ_USER
? XFS_BLF_UDQUOT_BUF
:
336 ((type
& XFS_DQ_PROJ
) ? XFS_BLF_PDQUOT_BUF
:
337 XFS_BLF_GDQUOT_BUF
)));
338 xfs_trans_log_buf(tp
, bp
, 0, BBTOB(q
->qi_dqchunklen
) - 1);
344 * Allocate a block and fill it with dquots.
345 * This is called when the bmapi finds a hole.
353 xfs_fileoff_t offset_fsb
,
356 xfs_fsblock_t firstblock
;
357 xfs_bmap_free_t flist
;
359 int nmaps
, error
, committed
;
361 xfs_trans_t
*tp
= *tpp
;
365 trace_xfs_dqalloc(dqp
);
368 * Initialize the bmap freelist prior to calling bmapi code.
370 xfs_bmap_init(&flist
, &firstblock
);
371 xfs_ilock(quotip
, XFS_ILOCK_EXCL
);
373 * Return if this type of quotas is turned off while we didn't
376 if (XFS_IS_THIS_QUOTA_OFF(dqp
)) {
377 xfs_iunlock(quotip
, XFS_ILOCK_EXCL
);
381 xfs_trans_ijoin_ref(tp
, quotip
, XFS_ILOCK_EXCL
);
383 if ((error
= xfs_bmapi(tp
, quotip
,
384 offset_fsb
, XFS_DQUOT_CLUSTER_SIZE_FSB
,
385 XFS_BMAPI_METADATA
| XFS_BMAPI_WRITE
,
387 XFS_QM_DQALLOC_SPACE_RES(mp
),
388 &map
, &nmaps
, &flist
))) {
391 ASSERT(map
.br_blockcount
== XFS_DQUOT_CLUSTER_SIZE_FSB
);
393 ASSERT((map
.br_startblock
!= DELAYSTARTBLOCK
) &&
394 (map
.br_startblock
!= HOLESTARTBLOCK
));
397 * Keep track of the blkno to save a lookup later
399 dqp
->q_blkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
);
401 /* now we can just get the buffer (there's nothing to read yet) */
402 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
404 mp
->m_quotainfo
->qi_dqchunklen
,
406 if (!bp
|| (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
,
470 xfs_disk_dquot_t
*ddq
;
473 xfs_trans_t
*tp
= (tpp
? *tpp
: NULL
);
476 id
= be32_to_cpu(dqp
->q_core
.d_id
);
481 * If we don't know where the dquot lives, find out.
483 if (dqp
->q_blkno
== (xfs_daddr_t
) 0) {
484 /* We use the id as an index */
485 dqp
->q_fileoffset
= (xfs_fileoff_t
)id
/
486 mp
->m_quotainfo
->qi_dqperchunk
;
488 quotip
= XFS_DQ_TO_QIP(dqp
);
489 xfs_ilock(quotip
, XFS_ILOCK_SHARED
);
491 * Return if this type of quotas is turned off while we didn't
494 if (XFS_IS_THIS_QUOTA_OFF(dqp
)) {
495 xfs_iunlock(quotip
, XFS_ILOCK_SHARED
);
499 * Find the block map; no allocations yet
501 error
= xfs_bmapi(NULL
, quotip
, dqp
->q_fileoffset
,
502 XFS_DQUOT_CLUSTER_SIZE_FSB
,
504 NULL
, 0, &map
, &nmaps
, NULL
);
506 xfs_iunlock(quotip
, XFS_ILOCK_SHARED
);
510 ASSERT(map
.br_blockcount
== 1);
513 * offset of dquot in the (fixed sized) dquot chunk.
515 dqp
->q_bufoffset
= (id
% mp
->m_quotainfo
->qi_dqperchunk
) *
517 if (map
.br_startblock
== HOLESTARTBLOCK
) {
519 * We don't allocate unless we're asked to
521 if (!(flags
& XFS_QMOPT_DQALLOC
))
525 if ((error
= xfs_qm_dqalloc(tpp
, mp
, dqp
, quotip
,
526 dqp
->q_fileoffset
, &bp
)))
532 * store the blkno etc so that we don't have to do the
533 * mapping all the time
535 dqp
->q_blkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
);
538 ASSERT(dqp
->q_blkno
!= DELAYSTARTBLOCK
);
539 ASSERT(dqp
->q_blkno
!= HOLESTARTBLOCK
);
542 * Read in the buffer, unless we've just done the allocation
543 * (in which case we already have the buf).
546 trace_xfs_dqtobp_read(dqp
);
548 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
550 mp
->m_quotainfo
->qi_dqchunklen
,
553 return XFS_ERROR(error
);
555 ASSERT(XFS_BUF_ISBUSY(bp
));
556 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
559 * calculate the location of the dquot inside the buffer.
561 ddq
= (xfs_disk_dquot_t
*)((char *)XFS_BUF_PTR(bp
) + dqp
->q_bufoffset
);
564 * A simple sanity check in case we got a corrupted dquot...
566 if (xfs_qm_dqcheck(ddq
, id
, dqp
->dq_flags
& XFS_DQ_ALLTYPES
,
567 flags
& (XFS_QMOPT_DQREPAIR
|XFS_QMOPT_DOWARN
),
569 if (!(flags
& XFS_QMOPT_DQREPAIR
)) {
570 xfs_trans_brelse(tp
, bp
);
571 return XFS_ERROR(EIO
);
573 XFS_BUF_BUSY(bp
); /* We dirtied this */
584 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
585 * and release the buffer immediately.
593 xfs_dquot_t
*dqp
, /* dquot to get filled in */
596 xfs_disk_dquot_t
*ddqp
;
603 trace_xfs_dqread(dqp
);
606 * get a pointer to the on-disk dquot and the buffer containing it
607 * dqp already knows its own type (GROUP/USER).
609 if ((error
= xfs_qm_dqtobp(tpp
, dqp
, &ddqp
, &bp
, flags
))) {
614 /* copy everything from disk dquot to the incore dquot */
615 memcpy(&dqp
->q_core
, ddqp
, sizeof(xfs_disk_dquot_t
));
616 ASSERT(be32_to_cpu(dqp
->q_core
.d_id
) == id
);
617 xfs_qm_dquot_logitem_init(dqp
);
620 * Reservation counters are defined as reservation plus current usage
621 * to avoid having to add everytime.
623 dqp
->q_res_bcount
= be64_to_cpu(ddqp
->d_bcount
);
624 dqp
->q_res_icount
= be64_to_cpu(ddqp
->d_icount
);
625 dqp
->q_res_rtbcount
= be64_to_cpu(ddqp
->d_rtbcount
);
627 /* Mark the buf so that this will stay incore a little longer */
628 XFS_BUF_SET_VTYPE_REF(bp
, B_FS_DQUOT
, XFS_DQUOT_REF
);
631 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
632 * So we need to release with xfs_trans_brelse().
633 * The strategy here is identical to that of inodes; we lock
634 * the dquot in xfs_qm_dqget() before making it accessible to
635 * others. This is because dquots, like inodes, need a good level of
636 * concurrency, and we don't want to take locks on the entire buffers
637 * for dquot accesses.
638 * Note also that the dquot buffer may even be dirty at this point, if
639 * this particular dquot was repaired. We still aren't afraid to
640 * brelse it because we have the changes incore.
642 ASSERT(XFS_BUF_ISBUSY(bp
));
643 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
644 xfs_trans_brelse(tp
, bp
);
651 * allocate an incore dquot from the kernel heap,
652 * and fill its core with quota information kept on disk.
653 * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk
654 * if it wasn't already allocated.
659 xfs_dqid_t id
, /* gid or uid, depending on type */
660 uint type
, /* UDQUOT or GDQUOT */
661 uint flags
, /* DQALLOC, DQREPAIR */
662 xfs_dquot_t
**O_dqpp
)/* OUT : incore dquot, not locked */
669 dqp
= xfs_qm_dqinit(mp
, id
, type
);
671 if (flags
& XFS_QMOPT_DQALLOC
) {
672 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_DQALLOC
);
673 error
= xfs_trans_reserve(tp
, XFS_QM_DQALLOC_SPACE_RES(mp
),
674 XFS_WRITE_LOG_RES(mp
) +
675 BBTOB(mp
->m_quotainfo
->qi_dqchunklen
) - 1 +
678 XFS_TRANS_PERM_LOG_RES
,
679 XFS_WRITE_LOG_COUNT
);
684 cancelflags
= XFS_TRANS_RELEASE_LOG_RES
;
688 * Read it from disk; xfs_dqread() takes care of
689 * all the necessary initialization of dquot's fields (locks, etc)
691 if ((error
= xfs_qm_dqread(&tp
, id
, dqp
, flags
))) {
693 * This can happen if quotas got turned off (ESRCH),
694 * or if the dquot didn't exist on disk and we ask to
697 trace_xfs_dqread_fail(dqp
);
698 cancelflags
|= XFS_TRANS_ABORT
;
702 if ((error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
)))
712 xfs_trans_cancel(tp
, cancelflags
);
714 xfs_qm_dqdestroy(dqp
);
720 * Lookup a dquot in the incore dquot hashtable. We keep two separate
721 * hashtables for user and group dquots; and, these are global tables
722 * inside the XQM, not per-filesystem tables.
723 * The hash chain must be locked by caller, and it is left locked
724 * on return. Returning dquot is locked.
731 xfs_dquot_t
**O_dqpp
)
736 ASSERT(mutex_is_locked(&qh
->qh_lock
));
738 flist_locked
= B_FALSE
;
741 * Traverse the hashchain looking for a match
743 list_for_each_entry(dqp
, &qh
->qh_list
, q_hashlist
) {
745 * We already have the hashlock. We don't need the
746 * dqlock to look at the id field of the dquot, since the
747 * id can't be modified without the hashlock anyway.
749 if (be32_to_cpu(dqp
->q_core
.d_id
) == id
&& dqp
->q_mount
== mp
) {
750 trace_xfs_dqlookup_found(dqp
);
753 * All in core dquots must be on the dqlist of mp
755 ASSERT(!list_empty(&dqp
->q_mplist
));
758 if (dqp
->q_nrefs
== 0) {
759 ASSERT(!list_empty(&dqp
->q_freelist
));
760 if (!mutex_trylock(&xfs_Gqm
->qm_dqfrlist_lock
)) {
761 trace_xfs_dqlookup_want(dqp
);
764 * We may have raced with dqreclaim_one()
765 * (and lost). So, flag that we don't
766 * want the dquot to be reclaimed.
768 dqp
->dq_flags
|= XFS_DQ_WANT
;
770 mutex_lock(&xfs_Gqm
->qm_dqfrlist_lock
);
772 dqp
->dq_flags
&= ~(XFS_DQ_WANT
);
774 flist_locked
= B_TRUE
;
778 * id couldn't have changed; we had the hashlock all
781 ASSERT(be32_to_cpu(dqp
->q_core
.d_id
) == id
);
784 if (dqp
->q_nrefs
!= 0) {
785 mutex_unlock(&xfs_Gqm
->qm_dqfrlist_lock
);
786 flist_locked
= B_FALSE
;
788 /* take it off the freelist */
789 trace_xfs_dqlookup_freelist(dqp
);
790 list_del_init(&dqp
->q_freelist
);
791 xfs_Gqm
->qm_dqfrlist_cnt
--;
798 mutex_unlock(&xfs_Gqm
->qm_dqfrlist_lock
);
800 * move the dquot to the front of the hashchain
802 ASSERT(mutex_is_locked(&qh
->qh_lock
));
803 list_move(&dqp
->q_hashlist
, &qh
->qh_list
);
804 trace_xfs_dqlookup_done(dqp
);
811 ASSERT(mutex_is_locked(&qh
->qh_lock
));
816 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
817 * a locked dquot, doing an allocation (if requested) as needed.
818 * When both an inode and an id are given, the inode's id takes precedence.
819 * That is, if the id changes while we don't hold the ilock inside this
820 * function, the new dquot is returned, not necessarily the one requested
821 * in the id argument.
826 xfs_inode_t
*ip
, /* locked inode (optional) */
827 xfs_dqid_t id
, /* uid/projid/gid depending on type */
828 uint type
, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
829 uint flags
, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
830 xfs_dquot_t
**O_dqpp
) /* OUT : locked incore dquot */
837 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
838 if ((! XFS_IS_UQUOTA_ON(mp
) && type
== XFS_DQ_USER
) ||
839 (! XFS_IS_PQUOTA_ON(mp
) && type
== XFS_DQ_PROJ
) ||
840 (! XFS_IS_GQUOTA_ON(mp
) && type
== XFS_DQ_GROUP
)) {
843 h
= XFS_DQ_HASH(mp
, id
, type
);
846 if (xfs_do_dqerror
) {
847 if ((xfs_dqerror_target
== mp
->m_ddev_targp
) &&
848 (xfs_dqreq_num
++ % xfs_dqerror_mod
) == 0) {
849 cmn_err(CE_DEBUG
, "Returning error in dqget");
858 ASSERT(type
== XFS_DQ_USER
||
859 type
== XFS_DQ_PROJ
||
860 type
== XFS_DQ_GROUP
);
862 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
863 if (type
== XFS_DQ_USER
)
864 ASSERT(ip
->i_udquot
== NULL
);
866 ASSERT(ip
->i_gdquot
== NULL
);
869 mutex_lock(&h
->qh_lock
);
872 * Look in the cache (hashtable).
873 * The chain is kept locked during lookup.
875 if (xfs_qm_dqlookup(mp
, id
, h
, O_dqpp
) == 0) {
876 XQM_STATS_INC(xqmstats
.xs_qm_dqcachehits
);
878 * The dquot was found, moved to the front of the chain,
879 * taken off the freelist if it was on it, and locked
880 * at this point. Just unlock the hashchain and return.
883 ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp
));
884 mutex_unlock(&h
->qh_lock
);
885 trace_xfs_dqget_hit(*O_dqpp
);
886 return (0); /* success */
888 XQM_STATS_INC(xqmstats
.xs_qm_dqcachemisses
);
891 * Dquot cache miss. We don't want to keep the inode lock across
892 * a (potential) disk read. Also we don't want to deal with the lock
893 * ordering between quotainode and this inode. OTOH, dropping the inode
894 * lock here means dealing with a chown that can happen before
895 * we re-acquire the lock.
898 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
900 * Save the hashchain version stamp, and unlock the chain, so that
901 * we don't keep the lock across a disk read
903 version
= h
->qh_version
;
904 mutex_unlock(&h
->qh_lock
);
907 * Allocate the dquot on the kernel heap, and read the ondisk
908 * portion off the disk. Also, do all the necessary initialization
909 * This can return ENOENT if dquot didn't exist on disk and we didn't
910 * ask it to allocate; ESRCH if quotas got turned off suddenly.
912 if ((error
= xfs_qm_idtodq(mp
, id
, type
,
913 flags
& (XFS_QMOPT_DQALLOC
|XFS_QMOPT_DQREPAIR
|
917 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
922 * See if this is mount code calling to look at the overall quota limits
923 * which are stored in the id == 0 user or group's dquot.
924 * Since we may not have done a quotacheck by this point, just return
925 * the dquot without attaching it to any hashtables, lists, etc, or even
926 * taking a reference.
927 * The caller must dqdestroy this once done.
929 if (flags
& XFS_QMOPT_DQSUSER
) {
936 * Dquot lock comes after hashlock in the lock ordering
939 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
942 * A dquot could be attached to this inode by now, since
943 * we had dropped the ilock.
945 if (type
== XFS_DQ_USER
) {
946 if (!XFS_IS_UQUOTA_ON(mp
)) {
947 /* inode stays locked on return */
948 xfs_qm_dqdestroy(dqp
);
949 return XFS_ERROR(ESRCH
);
952 xfs_qm_dqdestroy(dqp
);
958 if (!XFS_IS_OQUOTA_ON(mp
)) {
959 /* inode stays locked on return */
960 xfs_qm_dqdestroy(dqp
);
961 return XFS_ERROR(ESRCH
);
964 xfs_qm_dqdestroy(dqp
);
973 * Hashlock comes after ilock in lock order
975 mutex_lock(&h
->qh_lock
);
976 if (version
!= h
->qh_version
) {
979 * Now, see if somebody else put the dquot in the
980 * hashtable before us. This can happen because we didn't
981 * keep the hashchain lock. We don't have to worry about
982 * lock order between the two dquots here since dqp isn't
983 * on any findable lists yet.
985 if (xfs_qm_dqlookup(mp
, id
, h
, &tmpdqp
) == 0) {
987 * Duplicate found. Just throw away the new dquot
990 xfs_qm_dqput(tmpdqp
);
991 mutex_unlock(&h
->qh_lock
);
992 xfs_qm_dqdestroy(dqp
);
993 XQM_STATS_INC(xqmstats
.xs_qm_dquot_dups
);
999 * Put the dquot at the beginning of the hash-chain and mp's list
1000 * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock ..
1002 ASSERT(mutex_is_locked(&h
->qh_lock
));
1004 list_add(&dqp
->q_hashlist
, &h
->qh_list
);
1008 * Attach this dquot to this filesystem's list of all dquots,
1009 * kept inside the mount structure in m_quotainfo field
1011 mutex_lock(&mp
->m_quotainfo
->qi_dqlist_lock
);
1014 * We return a locked dquot to the caller, with a reference taken
1019 list_add(&dqp
->q_mplist
, &mp
->m_quotainfo
->qi_dqlist
);
1020 mp
->m_quotainfo
->qi_dquots
++;
1021 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
1022 mutex_unlock(&h
->qh_lock
);
1024 ASSERT((ip
== NULL
) || xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1025 trace_xfs_dqget_miss(dqp
);
1032 * Release a reference to the dquot (decrement ref-count)
1033 * and unlock it. If there is a group quota attached to this
1034 * dquot, carefully release that too without tripping over
1035 * deadlocks'n'stuff.
1043 ASSERT(dqp
->q_nrefs
> 0);
1044 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1046 trace_xfs_dqput(dqp
);
1048 if (dqp
->q_nrefs
!= 1) {
1055 * drop the dqlock and acquire the freelist and dqlock
1056 * in the right order; but try to get it out-of-order first
1058 if (!mutex_trylock(&xfs_Gqm
->qm_dqfrlist_lock
)) {
1059 trace_xfs_dqput_wait(dqp
);
1061 mutex_lock(&xfs_Gqm
->qm_dqfrlist_lock
);
1068 /* We can't depend on nrefs being == 1 here */
1069 if (--dqp
->q_nrefs
== 0) {
1070 trace_xfs_dqput_free(dqp
);
1072 list_add_tail(&dqp
->q_freelist
, &xfs_Gqm
->qm_dqfrlist
);
1073 xfs_Gqm
->qm_dqfrlist_cnt
++;
1076 * If we just added a udquot to the freelist, then
1077 * we want to release the gdquot reference that
1078 * it (probably) has. Otherwise it'll keep the
1079 * gdquot from getting reclaimed.
1081 if ((gdqp
= dqp
->q_gdquot
)) {
1083 * Avoid a recursive dqput call
1086 dqp
->q_gdquot
= NULL
;
1092 * If we had a group quota inside the user quota as a hint,
1099 mutex_unlock(&xfs_Gqm
->qm_dqfrlist_lock
);
1103 * Release a dquot. Flush it if dirty, then dqput() it.
1104 * dquot must not be locked.
1113 trace_xfs_dqrele(dqp
);
1117 * We don't care to flush it if the dquot is dirty here.
1118 * That will create stutters that we want to avoid.
1119 * Instead we do a delayed write when we try to reclaim
1120 * a dirty dquot. Also xfs_sync will take part of the burden...
1126 * This is the dquot flushing I/O completion routine. It is called
1127 * from interrupt level when the buffer containing the dquot is
1128 * flushed to disk. It is responsible for removing the dquot logitem
1129 * from the AIL if it has not been re-logged, and unlocking the dquot's
1130 * flush lock. This behavior is very similar to that of inodes..
1133 xfs_qm_dqflush_done(
1135 struct xfs_log_item
*lip
)
1137 xfs_dq_logitem_t
*qip
= (struct xfs_dq_logitem
*)lip
;
1138 xfs_dquot_t
*dqp
= qip
->qli_dquot
;
1139 struct xfs_ail
*ailp
= lip
->li_ailp
;
1142 * We only want to pull the item from the AIL if its
1143 * location in the log has not changed since we started the flush.
1144 * Thus, we only bother if the dquot's lsn has
1145 * not changed. First we check the lsn outside the lock
1146 * since it's cheaper, and then we recheck while
1147 * holding the lock before removing the dquot from the AIL.
1149 if ((lip
->li_flags
& XFS_LI_IN_AIL
) &&
1150 lip
->li_lsn
== qip
->qli_flush_lsn
) {
1152 /* xfs_trans_ail_delete() drops the AIL lock. */
1153 spin_lock(&ailp
->xa_lock
);
1154 if (lip
->li_lsn
== qip
->qli_flush_lsn
)
1155 xfs_trans_ail_delete(ailp
, lip
);
1157 spin_unlock(&ailp
->xa_lock
);
1161 * Release the dq's flush lock since we're done with it.
1167 * Write a modified dquot to disk.
1168 * The dquot must be locked and the flush lock too taken by caller.
1169 * The flush lock will not be unlocked until the dquot reaches the disk,
1170 * but the dquot is free to be unlocked and modified by the caller
1171 * in the interim. Dquot is still locked on return. This behavior is
1172 * identical to that of inodes.
1181 xfs_disk_dquot_t
*ddqp
;
1184 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1185 ASSERT(!completion_done(&dqp
->q_flush
));
1186 trace_xfs_dqflush(dqp
);
1189 * If not dirty, or it's pinned and we are not supposed to
1192 if (!XFS_DQ_IS_DIRTY(dqp
) ||
1193 (!(flags
& SYNC_WAIT
) && atomic_read(&dqp
->q_pincount
) > 0)) {
1197 xfs_qm_dqunpin_wait(dqp
);
1200 * This may have been unpinned because the filesystem is shutting
1201 * down forcibly. If that's the case we must not write this dquot
1202 * to disk, because the log record didn't make it to disk!
1204 if (XFS_FORCED_SHUTDOWN(dqp
->q_mount
)) {
1205 dqp
->dq_flags
&= ~(XFS_DQ_DIRTY
);
1207 return XFS_ERROR(EIO
);
1211 * Get the buffer containing the on-disk dquot
1212 * We don't need a transaction envelope because we know that the
1213 * the ondisk-dquot has already been allocated for.
1215 if ((error
= xfs_qm_dqtobp(NULL
, dqp
, &ddqp
, &bp
, XFS_QMOPT_DOWARN
))) {
1216 ASSERT(error
!= ENOENT
);
1218 * Quotas could have gotten turned off (ESRCH)
1224 if (xfs_qm_dqcheck(&dqp
->q_core
, be32_to_cpu(ddqp
->d_id
),
1225 0, XFS_QMOPT_DOWARN
, "dqflush (incore copy)")) {
1226 xfs_force_shutdown(dqp
->q_mount
, SHUTDOWN_CORRUPT_INCORE
);
1227 return XFS_ERROR(EIO
);
1230 /* This is the only portion of data that needs to persist */
1231 memcpy(ddqp
, &(dqp
->q_core
), sizeof(xfs_disk_dquot_t
));
1234 * Clear the dirty field and remember the flush lsn for later use.
1236 dqp
->dq_flags
&= ~(XFS_DQ_DIRTY
);
1239 xfs_trans_ail_copy_lsn(mp
->m_ail
, &dqp
->q_logitem
.qli_flush_lsn
,
1240 &dqp
->q_logitem
.qli_item
.li_lsn
);
1243 * Attach an iodone routine so that we can remove this dquot from the
1244 * AIL and release the flush lock once the dquot is synced to disk.
1246 xfs_buf_attach_iodone(bp
, xfs_qm_dqflush_done
,
1247 &dqp
->q_logitem
.qli_item
);
1250 * If the buffer is pinned then push on the log so we won't
1251 * get stuck waiting in the write for too long.
1253 if (XFS_BUF_ISPINNED(bp
)) {
1254 trace_xfs_dqflush_force(dqp
);
1255 xfs_log_force(mp
, 0);
1258 if (flags
& SYNC_WAIT
)
1259 error
= xfs_bwrite(mp
, bp
);
1261 xfs_bdwrite(mp
, bp
);
1263 trace_xfs_dqflush_done(dqp
);
1266 * dqp is still locked, but caller is free to unlock it now.
1273 xfs_qm_dqlock_nowait(
1276 return mutex_trylock(&dqp
->q_qlock
);
1283 mutex_lock(&dqp
->q_qlock
);
1290 mutex_unlock(&(dqp
->q_qlock
));
1291 if (dqp
->q_logitem
.qli_dquot
== dqp
) {
1292 /* Once was dqp->q_mount, but might just have been cleared */
1293 xfs_trans_unlocked_item(dqp
->q_logitem
.qli_item
.li_ailp
,
1294 (xfs_log_item_t
*)&(dqp
->q_logitem
));
1300 xfs_dqunlock_nonotify(
1303 mutex_unlock(&(dqp
->q_qlock
));
1307 * Lock two xfs_dquot structures.
1309 * To avoid deadlocks we always lock the quota structure with
1310 * the lowerd id first.
1319 if (be32_to_cpu(d1
->q_core
.d_id
) >
1320 be32_to_cpu(d2
->q_core
.d_id
)) {
1321 mutex_lock(&d2
->q_qlock
);
1322 mutex_lock_nested(&d1
->q_qlock
, XFS_QLOCK_NESTED
);
1324 mutex_lock(&d1
->q_qlock
);
1325 mutex_lock_nested(&d2
->q_qlock
, XFS_QLOCK_NESTED
);
1328 mutex_lock(&d1
->q_qlock
);
1330 mutex_lock(&d2
->q_qlock
);
1336 * Take a dquot out of the mount's dqlist as well as the hashlist.
1337 * This is called via unmount as well as quotaoff, and the purge
1338 * will always succeed unless there are soft (temp) references
1341 * This returns 0 if it was purged, 1 if it wasn't. It's not an error code
1342 * that we're returning! XXXsup - not cool.
1349 xfs_dqhash_t
*qh
= dqp
->q_hash
;
1350 xfs_mount_t
*mp
= dqp
->q_mount
;
1352 ASSERT(mutex_is_locked(&mp
->m_quotainfo
->qi_dqlist_lock
));
1353 ASSERT(mutex_is_locked(&dqp
->q_hash
->qh_lock
));
1357 * We really can't afford to purge a dquot that is
1358 * referenced, because these are hard refs.
1359 * It shouldn't happen in general because we went thru _all_ inodes in
1360 * dqrele_all_inodes before calling this and didn't let the mountlock go.
1361 * However it is possible that we have dquots with temporary
1362 * references that are not attached to an inode. e.g. see xfs_setattr().
1364 if (dqp
->q_nrefs
!= 0) {
1366 mutex_unlock(&dqp
->q_hash
->qh_lock
);
1370 ASSERT(!list_empty(&dqp
->q_freelist
));
1373 * If we're turning off quotas, we have to make sure that, for
1374 * example, we don't delete quota disk blocks while dquots are
1375 * in the process of getting written to those disk blocks.
1376 * This dquot might well be on AIL, and we can't leave it there
1377 * if we're turning off quotas. Basically, we need this flush
1378 * lock, and are willing to block on it.
1380 if (!xfs_dqflock_nowait(dqp
)) {
1382 * Block on the flush lock after nudging dquot buffer,
1385 xfs_qm_dqflock_pushbuf_wait(dqp
);
1389 * XXXIf we're turning this type of quotas off, we don't care
1390 * about the dirty metadata sitting in this dquot. OTOH, if
1391 * we're unmounting, we do care, so we flush it and wait.
1393 if (XFS_DQ_IS_DIRTY(dqp
)) {
1396 /* dqflush unlocks dqflock */
1398 * Given that dqpurge is a very rare occurrence, it is OK
1399 * that we're holding the hashlist and mplist locks
1400 * across the disk write. But, ... XXXsup
1402 * We don't care about getting disk errors here. We need
1403 * to purge this dquot anyway, so we go ahead regardless.
1405 error
= xfs_qm_dqflush(dqp
, SYNC_WAIT
);
1407 xfs_fs_cmn_err(CE_WARN
, mp
,
1408 "xfs_qm_dqpurge: dquot %p flush failed", dqp
);
1411 ASSERT(atomic_read(&dqp
->q_pincount
) == 0);
1412 ASSERT(XFS_FORCED_SHUTDOWN(mp
) ||
1413 !(dqp
->q_logitem
.qli_item
.li_flags
& XFS_LI_IN_AIL
));
1415 list_del_init(&dqp
->q_hashlist
);
1417 list_del_init(&dqp
->q_mplist
);
1418 mp
->m_quotainfo
->qi_dqreclaims
++;
1419 mp
->m_quotainfo
->qi_dquots
--;
1421 * XXX Move this to the front of the freelist, if we can get the
1424 ASSERT(!list_empty(&dqp
->q_freelist
));
1426 dqp
->q_mount
= NULL
;
1428 dqp
->dq_flags
= XFS_DQ_INACTIVE
;
1429 memset(&dqp
->q_core
, 0, sizeof(dqp
->q_core
));
1432 mutex_unlock(&qh
->qh_lock
);
1439 xfs_qm_dqprint(xfs_dquot_t
*dqp
)
1441 cmn_err(CE_DEBUG
, "-----------KERNEL DQUOT----------------");
1442 cmn_err(CE_DEBUG
, "---- dquotID = %d",
1443 (int)be32_to_cpu(dqp
->q_core
.d_id
));
1444 cmn_err(CE_DEBUG
, "---- type = %s", DQFLAGTO_TYPESTR(dqp
));
1445 cmn_err(CE_DEBUG
, "---- fs = 0x%p", dqp
->q_mount
);
1446 cmn_err(CE_DEBUG
, "---- blkno = 0x%x", (int) dqp
->q_blkno
);
1447 cmn_err(CE_DEBUG
, "---- boffset = 0x%x", (int) dqp
->q_bufoffset
);
1448 cmn_err(CE_DEBUG
, "---- blkhlimit = %Lu (0x%x)",
1449 be64_to_cpu(dqp
->q_core
.d_blk_hardlimit
),
1450 (int)be64_to_cpu(dqp
->q_core
.d_blk_hardlimit
));
1451 cmn_err(CE_DEBUG
, "---- blkslimit = %Lu (0x%x)",
1452 be64_to_cpu(dqp
->q_core
.d_blk_softlimit
),
1453 (int)be64_to_cpu(dqp
->q_core
.d_blk_softlimit
));
1454 cmn_err(CE_DEBUG
, "---- inohlimit = %Lu (0x%x)",
1455 be64_to_cpu(dqp
->q_core
.d_ino_hardlimit
),
1456 (int)be64_to_cpu(dqp
->q_core
.d_ino_hardlimit
));
1457 cmn_err(CE_DEBUG
, "---- inoslimit = %Lu (0x%x)",
1458 be64_to_cpu(dqp
->q_core
.d_ino_softlimit
),
1459 (int)be64_to_cpu(dqp
->q_core
.d_ino_softlimit
));
1460 cmn_err(CE_DEBUG
, "---- bcount = %Lu (0x%x)",
1461 be64_to_cpu(dqp
->q_core
.d_bcount
),
1462 (int)be64_to_cpu(dqp
->q_core
.d_bcount
));
1463 cmn_err(CE_DEBUG
, "---- icount = %Lu (0x%x)",
1464 be64_to_cpu(dqp
->q_core
.d_icount
),
1465 (int)be64_to_cpu(dqp
->q_core
.d_icount
));
1466 cmn_err(CE_DEBUG
, "---- btimer = %d",
1467 (int)be32_to_cpu(dqp
->q_core
.d_btimer
));
1468 cmn_err(CE_DEBUG
, "---- itimer = %d",
1469 (int)be32_to_cpu(dqp
->q_core
.d_itimer
));
1470 cmn_err(CE_DEBUG
, "---------------------------");
1475 * Give the buffer a little push if it is incore and
1476 * wait on the flush lock.
1479 xfs_qm_dqflock_pushbuf_wait(
1482 xfs_mount_t
*mp
= dqp
->q_mount
;
1486 * Check to see if the dquot has been flushed delayed
1487 * write. If so, grab its buffer and send it
1488 * out immediately. We'll be able to acquire
1489 * the flush lock when the I/O completes.
1491 bp
= xfs_incore(mp
->m_ddev_targp
, dqp
->q_blkno
,
1492 mp
->m_quotainfo
->qi_dqchunklen
, XBF_TRYLOCK
);
1496 if (XFS_BUF_ISDELAYWRITE(bp
)) {
1497 if (XFS_BUF_ISPINNED(bp
))
1498 xfs_log_force(mp
, 0);
1499 xfs_buf_delwri_promote(bp
);
1500 wake_up_process(bp
->b_target
->bt_task
);