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"
48 * dquot->q_qlock (xfs_dqlock() and friends)
49 * dquot->q_flush (xfs_dqflock() and friends)
50 * xfs_Gqm->qm_dqfrlist_lock
52 * If two dquots need to be locked the order is user before group/project,
53 * otherwise by the lowest id first, see xfs_dqlock2.
57 xfs_buftarg_t *xfs_dqerror_target;
60 int xfs_dqerror_mod = 33;
63 static struct lock_class_key xfs_dquot_other_class;
66 * Allocate and initialize a dquot. We don't always allocate fresh memory;
67 * we try to reclaim a free dquot if the number of incore dquots are above
69 * The only field inside the core that gets initialized at this point
70 * is the d_id field. The idea is to fill in the entire q_core
71 * when we read in the on disk dquot.
80 boolean_t brandnewdquot;
82 brandnewdquot = xfs_qm_dqalloc_incore(&dqp);
84 dqp->q_core.d_id = cpu_to_be32(id);
88 * No need to re-initialize these if this is a reclaimed dquot.
91 INIT_LIST_HEAD(&dqp->q_freelist);
92 mutex_init(&dqp->q_qlock);
93 init_waitqueue_head(&dqp->q_pinwait);
96 * Because we want to use a counting completion, complete
97 * the flush completion once to allow a single access to
98 * the flush completion without blocking.
100 init_completion(&dqp->q_flush);
101 complete(&dqp->q_flush);
103 trace_xfs_dqinit(dqp);
106 * Only the q_core portion was zeroed in dqreclaim_one().
107 * So, we need to reset others.
111 INIT_LIST_HEAD(&dqp->q_mplist);
112 INIT_LIST_HEAD(&dqp->q_hashlist);
113 dqp->q_bufoffset = 0;
114 dqp->q_fileoffset = 0;
115 dqp->q_transp = NULL;
116 dqp->q_gdquot = NULL;
117 dqp->q_res_bcount = 0;
118 dqp->q_res_icount = 0;
119 dqp->q_res_rtbcount = 0;
120 atomic_set(&dqp->q_pincount, 0);
122 ASSERT(list_empty(&dqp->q_freelist));
124 trace_xfs_dqreuse(dqp);
128 * In either case we need to make sure group quotas have a different
129 * lock class than user quotas, to make sure lockdep knows we can
130 * locks of one of each at the same time.
132 if (!(type & XFS_DQ_USER))
133 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class);
136 * log item gets initialized later
142 * This is called to free all the memory associated with a dquot
148 ASSERT(list_empty(&dqp->q_freelist));
150 mutex_destroy(&dqp->q_qlock);
151 kmem_zone_free(xfs_Gqm->qm_dqzone, dqp);
153 atomic_dec(&xfs_Gqm->qm_totaldquots);
157 * This is what a 'fresh' dquot inside a dquot chunk looks like on disk.
166 * Caller has zero'd the entire dquot 'chunk' already.
168 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
169 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
170 d->dd_diskdq.d_id = cpu_to_be32(id);
171 d->dd_diskdq.d_flags = type;
175 * If default limits are in force, push them into the dquot now.
176 * We overwrite the dquot limits only if they are zero and this
177 * is not the root dquot.
180 xfs_qm_adjust_dqlimits(
184 xfs_quotainfo_t *q = mp->m_quotainfo;
188 if (q->qi_bsoftlimit && !d->d_blk_softlimit)
189 d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit);
190 if (q->qi_bhardlimit && !d->d_blk_hardlimit)
191 d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit);
192 if (q->qi_isoftlimit && !d->d_ino_softlimit)
193 d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit);
194 if (q->qi_ihardlimit && !d->d_ino_hardlimit)
195 d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit);
196 if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit)
197 d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit);
198 if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit)
199 d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit);
203 * Check the limits and timers of a dquot and start or reset timers
205 * This gets called even when quota enforcement is OFF, which makes our
206 * life a little less complicated. (We just don't reject any quota
207 * reservations in that case, when enforcement is off).
208 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
210 * In contrast, warnings are a little different in that they don't
211 * 'automatically' get started when limits get exceeded. They do
212 * get reset to zero, however, when we find the count to be under
213 * the soft limit (they are only ever set non-zero via userspace).
216 xfs_qm_adjust_dqtimers(
223 if (d->d_blk_hardlimit)
224 ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
225 be64_to_cpu(d->d_blk_hardlimit));
226 if (d->d_ino_hardlimit)
227 ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
228 be64_to_cpu(d->d_ino_hardlimit));
229 if (d->d_rtb_hardlimit)
230 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
231 be64_to_cpu(d->d_rtb_hardlimit));
235 if ((d->d_blk_softlimit &&
236 (be64_to_cpu(d->d_bcount) >=
237 be64_to_cpu(d->d_blk_softlimit))) ||
238 (d->d_blk_hardlimit &&
239 (be64_to_cpu(d->d_bcount) >=
240 be64_to_cpu(d->d_blk_hardlimit)))) {
241 d->d_btimer = cpu_to_be32(get_seconds() +
242 mp->m_quotainfo->qi_btimelimit);
247 if ((!d->d_blk_softlimit ||
248 (be64_to_cpu(d->d_bcount) <
249 be64_to_cpu(d->d_blk_softlimit))) &&
250 (!d->d_blk_hardlimit ||
251 (be64_to_cpu(d->d_bcount) <
252 be64_to_cpu(d->d_blk_hardlimit)))) {
258 if ((d->d_ino_softlimit &&
259 (be64_to_cpu(d->d_icount) >=
260 be64_to_cpu(d->d_ino_softlimit))) ||
261 (d->d_ino_hardlimit &&
262 (be64_to_cpu(d->d_icount) >=
263 be64_to_cpu(d->d_ino_hardlimit)))) {
264 d->d_itimer = cpu_to_be32(get_seconds() +
265 mp->m_quotainfo->qi_itimelimit);
270 if ((!d->d_ino_softlimit ||
271 (be64_to_cpu(d->d_icount) <
272 be64_to_cpu(d->d_ino_softlimit))) &&
273 (!d->d_ino_hardlimit ||
274 (be64_to_cpu(d->d_icount) <
275 be64_to_cpu(d->d_ino_hardlimit)))) {
280 if (!d->d_rtbtimer) {
281 if ((d->d_rtb_softlimit &&
282 (be64_to_cpu(d->d_rtbcount) >=
283 be64_to_cpu(d->d_rtb_softlimit))) ||
284 (d->d_rtb_hardlimit &&
285 (be64_to_cpu(d->d_rtbcount) >=
286 be64_to_cpu(d->d_rtb_hardlimit)))) {
287 d->d_rtbtimer = cpu_to_be32(get_seconds() +
288 mp->m_quotainfo->qi_rtbtimelimit);
293 if ((!d->d_rtb_softlimit ||
294 (be64_to_cpu(d->d_rtbcount) <
295 be64_to_cpu(d->d_rtb_softlimit))) &&
296 (!d->d_rtb_hardlimit ||
297 (be64_to_cpu(d->d_rtbcount) <
298 be64_to_cpu(d->d_rtb_hardlimit)))) {
305 * initialize a buffer full of dquots and log the whole thing
308 xfs_qm_init_dquot_blk(
315 struct xfs_quotainfo *q = mp->m_quotainfo;
320 ASSERT(xfs_buf_islocked(bp));
325 * ID of the first dquot in the block - id's are zero based.
327 curid = id - (id % q->qi_dqperchunk);
329 memset(d, 0, BBTOB(q->qi_dqchunklen));
330 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++)
331 xfs_qm_dqinit_core(curid, type, d);
332 xfs_trans_dquot_buf(tp, bp,
333 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
334 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
335 XFS_BLF_GDQUOT_BUF)));
336 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
342 * Allocate a block and fill it with dquots.
343 * This is called when the bmapi finds a hole.
351 xfs_fileoff_t offset_fsb,
354 xfs_fsblock_t firstblock;
355 xfs_bmap_free_t flist;
357 int nmaps, error, committed;
359 xfs_trans_t *tp = *tpp;
363 trace_xfs_dqalloc(dqp);
366 * Initialize the bmap freelist prior to calling bmapi code.
368 xfs_bmap_init(&flist, &firstblock);
369 xfs_ilock(quotip, XFS_ILOCK_EXCL);
371 * Return if this type of quotas is turned off while we didn't
374 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
375 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
379 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
381 error = xfs_bmapi_write(tp, quotip, offset_fsb,
382 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
383 &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp),
384 &map, &nmaps, &flist);
387 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
389 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
390 (map.br_startblock != HOLESTARTBLOCK));
393 * Keep track of the blkno to save a lookup later
395 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
397 /* now we can just get the buffer (there's nothing to read yet) */
398 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
400 mp->m_quotainfo->qi_dqchunklen,
403 error = xfs_buf_geterror(bp);
408 * Make a chunk of dquots out of this buffer and log
411 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
412 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
415 * xfs_bmap_finish() may commit the current transaction and
416 * start a second transaction if the freelist is not empty.
418 * Since we still want to modify this buffer, we need to
419 * ensure that the buffer is not released on commit of
420 * the first transaction and ensure the buffer is added to the
421 * second transaction.
423 * If there is only one transaction then don't stop the buffer
424 * from being released when it commits later on.
427 xfs_trans_bhold(tp, bp);
429 if ((error = xfs_bmap_finish(tpp, &flist, &committed))) {
435 xfs_trans_bjoin(tp, bp);
437 xfs_trans_bhold_release(tp, bp);
444 xfs_bmap_cancel(&flist);
446 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
452 * Maps a dquot to the buffer containing its on-disk version.
453 * This returns a ptr to the buffer containing the on-disk dquot
454 * in the bpp param, and a ptr to the on-disk dquot within that buffer
460 xfs_disk_dquot_t **O_ddpp,
465 int nmaps = 1, error;
467 xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp);
468 xfs_mount_t *mp = dqp->q_mount;
469 xfs_disk_dquot_t *ddq;
470 xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id);
471 xfs_trans_t *tp = (tpp ? *tpp : NULL);
473 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
475 xfs_ilock(quotip, XFS_ILOCK_SHARED);
476 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
478 * Return if this type of quotas is turned off while we
479 * didn't have the quota inode lock.
481 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
486 * Find the block map; no allocations yet
488 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
489 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
491 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
496 ASSERT(map.br_blockcount == 1);
499 * Offset of dquot in the (fixed sized) dquot chunk.
501 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
504 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
505 if (map.br_startblock == HOLESTARTBLOCK) {
507 * We don't allocate unless we're asked to
509 if (!(flags & XFS_QMOPT_DQALLOC))
513 error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
514 dqp->q_fileoffset, &bp);
519 trace_xfs_dqtobp_read(dqp);
522 * store the blkno etc so that we don't have to do the
523 * mapping all the time
525 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
527 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
529 mp->m_quotainfo->qi_dqchunklen,
532 return XFS_ERROR(error);
535 ASSERT(xfs_buf_islocked(bp));
538 * calculate the location of the dquot inside the buffer.
540 ddq = bp->b_addr + dqp->q_bufoffset;
543 * A simple sanity check in case we got a corrupted dquot...
545 error = xfs_qm_dqcheck(mp, ddq, id, dqp->dq_flags & XFS_DQ_ALLTYPES,
546 flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN),
549 if (!(flags & XFS_QMOPT_DQREPAIR)) {
550 xfs_trans_brelse(tp, bp);
551 return XFS_ERROR(EIO);
563 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
564 * and release the buffer immediately.
572 xfs_dquot_t *dqp, /* dquot to get filled in */
575 xfs_disk_dquot_t *ddqp;
582 trace_xfs_dqread(dqp);
585 * get a pointer to the on-disk dquot and the buffer containing it
586 * dqp already knows its own type (GROUP/USER).
588 if ((error = xfs_qm_dqtobp(tpp, dqp, &ddqp, &bp, flags))) {
593 /* copy everything from disk dquot to the incore dquot */
594 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
595 ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);
596 xfs_qm_dquot_logitem_init(dqp);
599 * Reservation counters are defined as reservation plus current usage
600 * to avoid having to add every time.
602 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
603 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
604 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
606 /* Mark the buf so that this will stay incore a little longer */
607 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
610 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
611 * So we need to release with xfs_trans_brelse().
612 * The strategy here is identical to that of inodes; we lock
613 * the dquot in xfs_qm_dqget() before making it accessible to
614 * others. This is because dquots, like inodes, need a good level of
615 * concurrency, and we don't want to take locks on the entire buffers
616 * for dquot accesses.
617 * Note also that the dquot buffer may even be dirty at this point, if
618 * this particular dquot was repaired. We still aren't afraid to
619 * brelse it because we have the changes incore.
621 ASSERT(xfs_buf_islocked(bp));
622 xfs_trans_brelse(tp, bp);
629 * allocate an incore dquot from the kernel heap,
630 * and fill its core with quota information kept on disk.
631 * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk
632 * if it wasn't already allocated.
637 xfs_dqid_t id, /* gid or uid, depending on type */
638 uint type, /* UDQUOT or GDQUOT */
639 uint flags, /* DQALLOC, DQREPAIR */
640 xfs_dquot_t **O_dqpp)/* OUT : incore dquot, not locked */
647 dqp = xfs_qm_dqinit(mp, id, type);
649 if (flags & XFS_QMOPT_DQALLOC) {
650 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
651 error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp),
652 XFS_WRITE_LOG_RES(mp) +
653 BBTOB(mp->m_quotainfo->qi_dqchunklen) - 1 +
656 XFS_TRANS_PERM_LOG_RES,
657 XFS_WRITE_LOG_COUNT);
662 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
666 * Read it from disk; xfs_dqread() takes care of
667 * all the necessary initialization of dquot's fields (locks, etc)
669 if ((error = xfs_qm_dqread(&tp, id, dqp, flags))) {
671 * This can happen if quotas got turned off (ESRCH),
672 * or if the dquot didn't exist on disk and we ask to
675 trace_xfs_dqread_fail(dqp);
676 cancelflags |= XFS_TRANS_ABORT;
680 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES)))
690 xfs_trans_cancel(tp, cancelflags);
692 xfs_qm_dqdestroy(dqp);
698 * Lookup a dquot in the incore dquot hashtable. We keep two separate
699 * hashtables for user and group dquots; and, these are global tables
700 * inside the XQM, not per-filesystem tables.
701 * The hash chain must be locked by caller, and it is left locked
702 * on return. Returning dquot is locked.
709 xfs_dquot_t **O_dqpp)
713 ASSERT(mutex_is_locked(&qh->qh_lock));
716 * Traverse the hashchain looking for a match
718 list_for_each_entry(dqp, &qh->qh_list, q_hashlist) {
720 * We already have the hashlock. We don't need the
721 * dqlock to look at the id field of the dquot, since the
722 * id can't be modified without the hashlock anyway.
724 if (be32_to_cpu(dqp->q_core.d_id) != id || dqp->q_mount != mp)
727 trace_xfs_dqlookup_found(dqp);
730 if (dqp->dq_flags & XFS_DQ_FREEING) {
739 * move the dquot to the front of the hashchain
741 list_move(&dqp->q_hashlist, &qh->qh_list);
742 trace_xfs_dqlookup_done(dqp);
752 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
753 * a locked dquot, doing an allocation (if requested) as needed.
754 * When both an inode and an id are given, the inode's id takes precedence.
755 * That is, if the id changes while we don't hold the ilock inside this
756 * function, the new dquot is returned, not necessarily the one requested
757 * in the id argument.
762 xfs_inode_t *ip, /* locked inode (optional) */
763 xfs_dqid_t id, /* uid/projid/gid depending on type */
764 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
765 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
766 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
773 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
774 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
775 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
776 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
779 h = XFS_DQ_HASH(mp, id, type);
782 if (xfs_do_dqerror) {
783 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
784 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
785 xfs_debug(mp, "Returning error in dqget");
790 ASSERT(type == XFS_DQ_USER ||
791 type == XFS_DQ_PROJ ||
792 type == XFS_DQ_GROUP);
794 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
795 if (type == XFS_DQ_USER)
796 ASSERT(ip->i_udquot == NULL);
798 ASSERT(ip->i_gdquot == NULL);
803 mutex_lock(&h->qh_lock);
806 * Look in the cache (hashtable).
807 * The chain is kept locked during lookup.
809 switch (xfs_qm_dqlookup(mp, id, h, O_dqpp)) {
811 XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
812 mutex_unlock(&h->qh_lock);
816 XQM_STATS_INC(xqmstats.xs_qm_dqcachehits);
818 * The dquot was found, moved to the front of the chain,
819 * taken off the freelist if it was on it, and locked
820 * at this point. Just unlock the hashchain and return.
823 ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp));
824 mutex_unlock(&h->qh_lock);
825 trace_xfs_dqget_hit(*O_dqpp);
826 return 0; /* success */
828 XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses);
833 * Dquot cache miss. We don't want to keep the inode lock across
834 * a (potential) disk read. Also we don't want to deal with the lock
835 * ordering between quotainode and this inode. OTOH, dropping the inode
836 * lock here means dealing with a chown that can happen before
837 * we re-acquire the lock.
840 xfs_iunlock(ip, XFS_ILOCK_EXCL);
842 * Save the hashchain version stamp, and unlock the chain, so that
843 * we don't keep the lock across a disk read
845 version = h->qh_version;
846 mutex_unlock(&h->qh_lock);
849 * Allocate the dquot on the kernel heap, and read the ondisk
850 * portion off the disk. Also, do all the necessary initialization
851 * This can return ENOENT if dquot didn't exist on disk and we didn't
852 * ask it to allocate; ESRCH if quotas got turned off suddenly.
854 if ((error = xfs_qm_idtodq(mp, id, type,
855 flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR|
859 xfs_ilock(ip, XFS_ILOCK_EXCL);
864 * See if this is mount code calling to look at the overall quota limits
865 * which are stored in the id == 0 user or group's dquot.
866 * Since we may not have done a quotacheck by this point, just return
867 * the dquot without attaching it to any hashtables, lists, etc, or even
868 * taking a reference.
869 * The caller must dqdestroy this once done.
871 if (flags & XFS_QMOPT_DQSUSER) {
878 * Dquot lock comes after hashlock in the lock ordering
881 xfs_ilock(ip, XFS_ILOCK_EXCL);
884 * A dquot could be attached to this inode by now, since
885 * we had dropped the ilock.
887 if (type == XFS_DQ_USER) {
888 if (!XFS_IS_UQUOTA_ON(mp)) {
889 /* inode stays locked on return */
890 xfs_qm_dqdestroy(dqp);
891 return XFS_ERROR(ESRCH);
894 xfs_qm_dqdestroy(dqp);
900 if (!XFS_IS_OQUOTA_ON(mp)) {
901 /* inode stays locked on return */
902 xfs_qm_dqdestroy(dqp);
903 return XFS_ERROR(ESRCH);
906 xfs_qm_dqdestroy(dqp);
915 * Hashlock comes after ilock in lock order
917 mutex_lock(&h->qh_lock);
918 if (version != h->qh_version) {
921 * Now, see if somebody else put the dquot in the
922 * hashtable before us. This can happen because we didn't
923 * keep the hashchain lock. We don't have to worry about
924 * lock order between the two dquots here since dqp isn't
925 * on any findable lists yet.
927 switch (xfs_qm_dqlookup(mp, id, h, &tmpdqp)) {
931 * Duplicate found, either in cache or on its way out.
932 * Just throw away the new dquot and start over.
935 xfs_qm_dqput(tmpdqp);
936 mutex_unlock(&h->qh_lock);
937 xfs_qm_dqdestroy(dqp);
938 XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
946 * Put the dquot at the beginning of the hash-chain and mp's list
947 * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock ..
949 ASSERT(mutex_is_locked(&h->qh_lock));
951 list_add(&dqp->q_hashlist, &h->qh_list);
955 * Attach this dquot to this filesystem's list of all dquots,
956 * kept inside the mount structure in m_quotainfo field
958 mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
961 * We return a locked dquot to the caller, with a reference taken
966 list_add(&dqp->q_mplist, &mp->m_quotainfo->qi_dqlist);
967 mp->m_quotainfo->qi_dquots++;
968 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
969 mutex_unlock(&h->qh_lock);
971 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
972 trace_xfs_dqget_miss(dqp);
979 * Release a reference to the dquot (decrement ref-count)
980 * and unlock it. If there is a group quota attached to this
981 * dquot, carefully release that too without tripping over
986 struct xfs_dquot *dqp)
988 struct xfs_dquot *gdqp;
990 ASSERT(dqp->q_nrefs > 0);
991 ASSERT(XFS_DQ_IS_LOCKED(dqp));
993 trace_xfs_dqput(dqp);
996 if (--dqp->q_nrefs > 0) {
1001 trace_xfs_dqput_free(dqp);
1003 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1004 if (list_empty(&dqp->q_freelist)) {
1005 list_add_tail(&dqp->q_freelist, &xfs_Gqm->qm_dqfrlist);
1006 xfs_Gqm->qm_dqfrlist_cnt++;
1008 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1011 * If we just added a udquot to the freelist, then we want to release
1012 * the gdquot reference that it (probably) has. Otherwise it'll keep
1013 * the gdquot from getting reclaimed.
1015 gdqp = dqp->q_gdquot;
1018 dqp->q_gdquot = NULL;
1023 * If we had a group quota hint, release it now.
1032 * Release a dquot. Flush it if dirty, then dqput() it.
1033 * dquot must not be locked.
1042 trace_xfs_dqrele(dqp);
1046 * We don't care to flush it if the dquot is dirty here.
1047 * That will create stutters that we want to avoid.
1048 * Instead we do a delayed write when we try to reclaim
1049 * a dirty dquot. Also xfs_sync will take part of the burden...
1055 * This is the dquot flushing I/O completion routine. It is called
1056 * from interrupt level when the buffer containing the dquot is
1057 * flushed to disk. It is responsible for removing the dquot logitem
1058 * from the AIL if it has not been re-logged, and unlocking the dquot's
1059 * flush lock. This behavior is very similar to that of inodes..
1062 xfs_qm_dqflush_done(
1064 struct xfs_log_item *lip)
1066 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
1067 xfs_dquot_t *dqp = qip->qli_dquot;
1068 struct xfs_ail *ailp = lip->li_ailp;
1071 * We only want to pull the item from the AIL if its
1072 * location in the log has not changed since we started the flush.
1073 * Thus, we only bother if the dquot's lsn has
1074 * not changed. First we check the lsn outside the lock
1075 * since it's cheaper, and then we recheck while
1076 * holding the lock before removing the dquot from the AIL.
1078 if ((lip->li_flags & XFS_LI_IN_AIL) &&
1079 lip->li_lsn == qip->qli_flush_lsn) {
1081 /* xfs_trans_ail_delete() drops the AIL lock. */
1082 spin_lock(&ailp->xa_lock);
1083 if (lip->li_lsn == qip->qli_flush_lsn)
1084 xfs_trans_ail_delete(ailp, lip);
1086 spin_unlock(&ailp->xa_lock);
1090 * Release the dq's flush lock since we're done with it.
1096 * Write a modified dquot to disk.
1097 * The dquot must be locked and the flush lock too taken by caller.
1098 * The flush lock will not be unlocked until the dquot reaches the disk,
1099 * but the dquot is free to be unlocked and modified by the caller
1100 * in the interim. Dquot is still locked on return. This behavior is
1101 * identical to that of inodes.
1108 struct xfs_mount *mp = dqp->q_mount;
1110 struct xfs_disk_dquot *ddqp;
1113 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1114 ASSERT(!completion_done(&dqp->q_flush));
1116 trace_xfs_dqflush(dqp);
1119 * If not dirty, or it's pinned and we are not supposed to block, nada.
1121 if (!XFS_DQ_IS_DIRTY(dqp) ||
1122 ((flags & SYNC_TRYLOCK) && atomic_read(&dqp->q_pincount) > 0)) {
1126 xfs_qm_dqunpin_wait(dqp);
1129 * This may have been unpinned because the filesystem is shutting
1130 * down forcibly. If that's the case we must not write this dquot
1131 * to disk, because the log record didn't make it to disk!
1133 if (XFS_FORCED_SHUTDOWN(mp)) {
1134 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1136 return XFS_ERROR(EIO);
1140 * Get the buffer containing the on-disk dquot
1142 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1143 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
1145 ASSERT(error != ENOENT);
1151 * Calculate the location of the dquot inside the buffer.
1153 ddqp = bp->b_addr + dqp->q_bufoffset;
1156 * A simple sanity check in case we got a corrupted dquot..
1158 error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
1159 XFS_QMOPT_DOWARN, "dqflush (incore copy)");
1163 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1164 return XFS_ERROR(EIO);
1167 /* This is the only portion of data that needs to persist */
1168 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
1171 * Clear the dirty field and remember the flush lsn for later use.
1173 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1175 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1176 &dqp->q_logitem.qli_item.li_lsn);
1179 * Attach an iodone routine so that we can remove this dquot from the
1180 * AIL and release the flush lock once the dquot is synced to disk.
1182 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1183 &dqp->q_logitem.qli_item);
1186 * If the buffer is pinned then push on the log so we won't
1187 * get stuck waiting in the write for too long.
1189 if (xfs_buf_ispinned(bp)) {
1190 trace_xfs_dqflush_force(dqp);
1191 xfs_log_force(mp, 0);
1194 if (flags & SYNC_WAIT)
1195 error = xfs_bwrite(bp);
1197 xfs_buf_delwri_queue(bp);
1201 trace_xfs_dqflush_done(dqp);
1204 * dqp is still locked, but caller is free to unlock it now.
1214 xfs_dqunlock_nonotify(dqp);
1215 if (dqp->q_logitem.qli_dquot == dqp) {
1216 xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_ailp,
1217 &dqp->q_logitem.qli_item);
1222 * Lock two xfs_dquot structures.
1224 * To avoid deadlocks we always lock the quota structure with
1225 * the lowerd id first.
1234 if (be32_to_cpu(d1->q_core.d_id) >
1235 be32_to_cpu(d2->q_core.d_id)) {
1236 mutex_lock(&d2->q_qlock);
1237 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1239 mutex_lock(&d1->q_qlock);
1240 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1243 mutex_lock(&d1->q_qlock);
1245 mutex_lock(&d2->q_qlock);
1250 * Take a dquot out of the mount's dqlist as well as the hashlist. This is
1251 * called via unmount as well as quotaoff, and the purge will always succeed.
1255 struct xfs_dquot *dqp)
1257 struct xfs_mount *mp = dqp->q_mount;
1258 struct xfs_dqhash *qh = dqp->q_hash;
1263 * If we're turning off quotas, we have to make sure that, for
1264 * example, we don't delete quota disk blocks while dquots are
1265 * in the process of getting written to those disk blocks.
1266 * This dquot might well be on AIL, and we can't leave it there
1267 * if we're turning off quotas. Basically, we need this flush
1268 * lock, and are willing to block on it.
1270 if (!xfs_dqflock_nowait(dqp)) {
1272 * Block on the flush lock after nudging dquot buffer,
1275 xfs_dqflock_pushbuf_wait(dqp);
1279 * If we are turning this type of quotas off, we don't care
1280 * about the dirty metadata sitting in this dquot. OTOH, if
1281 * we're unmounting, we do care, so we flush it and wait.
1283 if (XFS_DQ_IS_DIRTY(dqp)) {
1287 * We don't care about getting disk errors here. We need
1288 * to purge this dquot anyway, so we go ahead regardless.
1290 error = xfs_qm_dqflush(dqp, SYNC_WAIT);
1292 xfs_warn(mp, "%s: dquot %p flush failed",
1297 ASSERT(atomic_read(&dqp->q_pincount) == 0);
1298 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1299 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
1304 mutex_lock(&qh->qh_lock);
1305 list_del_init(&dqp->q_hashlist);
1307 mutex_unlock(&qh->qh_lock);
1309 mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
1310 list_del_init(&dqp->q_mplist);
1311 mp->m_quotainfo->qi_dqreclaims++;
1312 mp->m_quotainfo->qi_dquots--;
1313 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
1316 * We move dquots to the freelist as soon as their reference count
1317 * hits zero, so it really should be on the freelist here.
1319 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1320 ASSERT(!list_empty(&dqp->q_freelist));
1321 list_del_init(&dqp->q_freelist);
1322 xfs_Gqm->qm_dqfrlist_cnt--;
1323 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1325 xfs_qm_dqdestroy(dqp);
1329 * Give the buffer a little push if it is incore and
1330 * wait on the flush lock.
1333 xfs_dqflock_pushbuf_wait(
1336 xfs_mount_t *mp = dqp->q_mount;
1340 * Check to see if the dquot has been flushed delayed
1341 * write. If so, grab its buffer and send it
1342 * out immediately. We'll be able to acquire
1343 * the flush lock when the I/O completes.
1345 bp = xfs_incore(mp->m_ddev_targp, dqp->q_blkno,
1346 mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK);
1350 if (XFS_BUF_ISDELAYWRITE(bp)) {
1351 if (xfs_buf_ispinned(bp))
1352 xfs_log_force(mp, 0);
1353 xfs_buf_delwri_promote(bp);
1354 wake_up_process(bp->b_target->bt_task);