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_ISBUSY(bp));
322 ASSERT(xfs_buf_islocked(bp));
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,
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(NULL, quotip, dqp->q_fileoffset,
490 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
491 NULL, 0, &map, &nmaps, NULL);
493 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
498 ASSERT(map.br_blockcount == 1);
501 * Offset of dquot in the (fixed sized) dquot chunk.
503 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
506 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
507 if (map.br_startblock == HOLESTARTBLOCK) {
509 * We don't allocate unless we're asked to
511 if (!(flags & XFS_QMOPT_DQALLOC))
515 error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
516 dqp->q_fileoffset, &bp);
521 trace_xfs_dqtobp_read(dqp);
524 * store the blkno etc so that we don't have to do the
525 * mapping all the time
527 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
529 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
531 mp->m_quotainfo->qi_dqchunklen,
534 return XFS_ERROR(error);
537 ASSERT(XFS_BUF_ISBUSY(bp));
538 ASSERT(xfs_buf_islocked(bp));
541 * calculate the location of the dquot inside the buffer.
543 ddq = (struct xfs_disk_dquot *)(XFS_BUF_PTR(bp) + dqp->q_bufoffset);
546 * A simple sanity check in case we got a corrupted dquot...
548 error = xfs_qm_dqcheck(mp, ddq, id, dqp->dq_flags & XFS_DQ_ALLTYPES,
549 flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN),
552 if (!(flags & XFS_QMOPT_DQREPAIR)) {
553 xfs_trans_brelse(tp, bp);
554 return XFS_ERROR(EIO);
556 XFS_BUF_BUSY(bp); /* We dirtied this */
567 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
568 * and release the buffer immediately.
576 xfs_dquot_t *dqp, /* dquot to get filled in */
579 xfs_disk_dquot_t *ddqp;
586 trace_xfs_dqread(dqp);
589 * get a pointer to the on-disk dquot and the buffer containing it
590 * dqp already knows its own type (GROUP/USER).
592 if ((error = xfs_qm_dqtobp(tpp, dqp, &ddqp, &bp, flags))) {
597 /* copy everything from disk dquot to the incore dquot */
598 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
599 ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);
600 xfs_qm_dquot_logitem_init(dqp);
603 * Reservation counters are defined as reservation plus current usage
604 * to avoid having to add every time.
606 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
607 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
608 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
610 /* Mark the buf so that this will stay incore a little longer */
611 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DQUOT, XFS_DQUOT_REF);
614 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
615 * So we need to release with xfs_trans_brelse().
616 * The strategy here is identical to that of inodes; we lock
617 * the dquot in xfs_qm_dqget() before making it accessible to
618 * others. This is because dquots, like inodes, need a good level of
619 * concurrency, and we don't want to take locks on the entire buffers
620 * for dquot accesses.
621 * Note also that the dquot buffer may even be dirty at this point, if
622 * this particular dquot was repaired. We still aren't afraid to
623 * brelse it because we have the changes incore.
625 ASSERT(XFS_BUF_ISBUSY(bp));
626 ASSERT(xfs_buf_islocked(bp));
627 xfs_trans_brelse(tp, bp);
634 * allocate an incore dquot from the kernel heap,
635 * and fill its core with quota information kept on disk.
636 * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk
637 * if it wasn't already allocated.
642 xfs_dqid_t id, /* gid or uid, depending on type */
643 uint type, /* UDQUOT or GDQUOT */
644 uint flags, /* DQALLOC, DQREPAIR */
645 xfs_dquot_t **O_dqpp)/* OUT : incore dquot, not locked */
652 dqp = xfs_qm_dqinit(mp, id, type);
654 if (flags & XFS_QMOPT_DQALLOC) {
655 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
656 error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp),
657 XFS_WRITE_LOG_RES(mp) +
658 BBTOB(mp->m_quotainfo->qi_dqchunklen) - 1 +
661 XFS_TRANS_PERM_LOG_RES,
662 XFS_WRITE_LOG_COUNT);
667 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
671 * Read it from disk; xfs_dqread() takes care of
672 * all the necessary initialization of dquot's fields (locks, etc)
674 if ((error = xfs_qm_dqread(&tp, id, dqp, flags))) {
676 * This can happen if quotas got turned off (ESRCH),
677 * or if the dquot didn't exist on disk and we ask to
680 trace_xfs_dqread_fail(dqp);
681 cancelflags |= XFS_TRANS_ABORT;
685 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES)))
695 xfs_trans_cancel(tp, cancelflags);
697 xfs_qm_dqdestroy(dqp);
703 * Lookup a dquot in the incore dquot hashtable. We keep two separate
704 * hashtables for user and group dquots; and, these are global tables
705 * inside the XQM, not per-filesystem tables.
706 * The hash chain must be locked by caller, and it is left locked
707 * on return. Returning dquot is locked.
714 xfs_dquot_t **O_dqpp)
719 ASSERT(mutex_is_locked(&qh->qh_lock));
721 flist_locked = B_FALSE;
724 * Traverse the hashchain looking for a match
726 list_for_each_entry(dqp, &qh->qh_list, q_hashlist) {
728 * We already have the hashlock. We don't need the
729 * dqlock to look at the id field of the dquot, since the
730 * id can't be modified without the hashlock anyway.
732 if (be32_to_cpu(dqp->q_core.d_id) == id && dqp->q_mount == mp) {
733 trace_xfs_dqlookup_found(dqp);
736 * All in core dquots must be on the dqlist of mp
738 ASSERT(!list_empty(&dqp->q_mplist));
741 if (dqp->q_nrefs == 0) {
742 ASSERT(!list_empty(&dqp->q_freelist));
743 if (!mutex_trylock(&xfs_Gqm->qm_dqfrlist_lock)) {
744 trace_xfs_dqlookup_want(dqp);
747 * We may have raced with dqreclaim_one()
748 * (and lost). So, flag that we don't
749 * want the dquot to be reclaimed.
751 dqp->dq_flags |= XFS_DQ_WANT;
753 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
755 dqp->dq_flags &= ~(XFS_DQ_WANT);
757 flist_locked = B_TRUE;
761 * id couldn't have changed; we had the hashlock all
764 ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);
767 if (dqp->q_nrefs != 0) {
768 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
769 flist_locked = B_FALSE;
771 /* take it off the freelist */
772 trace_xfs_dqlookup_freelist(dqp);
773 list_del_init(&dqp->q_freelist);
774 xfs_Gqm->qm_dqfrlist_cnt--;
781 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
783 * move the dquot to the front of the hashchain
785 ASSERT(mutex_is_locked(&qh->qh_lock));
786 list_move(&dqp->q_hashlist, &qh->qh_list);
787 trace_xfs_dqlookup_done(dqp);
794 ASSERT(mutex_is_locked(&qh->qh_lock));
799 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
800 * a locked dquot, doing an allocation (if requested) as needed.
801 * When both an inode and an id are given, the inode's id takes precedence.
802 * That is, if the id changes while we don't hold the ilock inside this
803 * function, the new dquot is returned, not necessarily the one requested
804 * in the id argument.
809 xfs_inode_t *ip, /* locked inode (optional) */
810 xfs_dqid_t id, /* uid/projid/gid depending on type */
811 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
812 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
813 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
820 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
821 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
822 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
823 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
826 h = XFS_DQ_HASH(mp, id, type);
829 if (xfs_do_dqerror) {
830 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
831 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
832 xfs_debug(mp, "Returning error in dqget");
841 ASSERT(type == XFS_DQ_USER ||
842 type == XFS_DQ_PROJ ||
843 type == XFS_DQ_GROUP);
845 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
846 if (type == XFS_DQ_USER)
847 ASSERT(ip->i_udquot == NULL);
849 ASSERT(ip->i_gdquot == NULL);
852 mutex_lock(&h->qh_lock);
855 * Look in the cache (hashtable).
856 * The chain is kept locked during lookup.
858 if (xfs_qm_dqlookup(mp, id, h, O_dqpp) == 0) {
859 XQM_STATS_INC(xqmstats.xs_qm_dqcachehits);
861 * The dquot was found, moved to the front of the chain,
862 * taken off the freelist if it was on it, and locked
863 * at this point. Just unlock the hashchain and return.
866 ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp));
867 mutex_unlock(&h->qh_lock);
868 trace_xfs_dqget_hit(*O_dqpp);
869 return (0); /* success */
871 XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses);
874 * Dquot cache miss. We don't want to keep the inode lock across
875 * a (potential) disk read. Also we don't want to deal with the lock
876 * ordering between quotainode and this inode. OTOH, dropping the inode
877 * lock here means dealing with a chown that can happen before
878 * we re-acquire the lock.
881 xfs_iunlock(ip, XFS_ILOCK_EXCL);
883 * Save the hashchain version stamp, and unlock the chain, so that
884 * we don't keep the lock across a disk read
886 version = h->qh_version;
887 mutex_unlock(&h->qh_lock);
890 * Allocate the dquot on the kernel heap, and read the ondisk
891 * portion off the disk. Also, do all the necessary initialization
892 * This can return ENOENT if dquot didn't exist on disk and we didn't
893 * ask it to allocate; ESRCH if quotas got turned off suddenly.
895 if ((error = xfs_qm_idtodq(mp, id, type,
896 flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR|
900 xfs_ilock(ip, XFS_ILOCK_EXCL);
905 * See if this is mount code calling to look at the overall quota limits
906 * which are stored in the id == 0 user or group's dquot.
907 * Since we may not have done a quotacheck by this point, just return
908 * the dquot without attaching it to any hashtables, lists, etc, or even
909 * taking a reference.
910 * The caller must dqdestroy this once done.
912 if (flags & XFS_QMOPT_DQSUSER) {
919 * Dquot lock comes after hashlock in the lock ordering
922 xfs_ilock(ip, XFS_ILOCK_EXCL);
925 * A dquot could be attached to this inode by now, since
926 * we had dropped the ilock.
928 if (type == XFS_DQ_USER) {
929 if (!XFS_IS_UQUOTA_ON(mp)) {
930 /* inode stays locked on return */
931 xfs_qm_dqdestroy(dqp);
932 return XFS_ERROR(ESRCH);
935 xfs_qm_dqdestroy(dqp);
941 if (!XFS_IS_OQUOTA_ON(mp)) {
942 /* inode stays locked on return */
943 xfs_qm_dqdestroy(dqp);
944 return XFS_ERROR(ESRCH);
947 xfs_qm_dqdestroy(dqp);
956 * Hashlock comes after ilock in lock order
958 mutex_lock(&h->qh_lock);
959 if (version != h->qh_version) {
962 * Now, see if somebody else put the dquot in the
963 * hashtable before us. This can happen because we didn't
964 * keep the hashchain lock. We don't have to worry about
965 * lock order between the two dquots here since dqp isn't
966 * on any findable lists yet.
968 if (xfs_qm_dqlookup(mp, id, h, &tmpdqp) == 0) {
970 * Duplicate found. Just throw away the new dquot
973 xfs_qm_dqput(tmpdqp);
974 mutex_unlock(&h->qh_lock);
975 xfs_qm_dqdestroy(dqp);
976 XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
982 * Put the dquot at the beginning of the hash-chain and mp's list
983 * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock ..
985 ASSERT(mutex_is_locked(&h->qh_lock));
987 list_add(&dqp->q_hashlist, &h->qh_list);
991 * Attach this dquot to this filesystem's list of all dquots,
992 * kept inside the mount structure in m_quotainfo field
994 mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
997 * We return a locked dquot to the caller, with a reference taken
1002 list_add(&dqp->q_mplist, &mp->m_quotainfo->qi_dqlist);
1003 mp->m_quotainfo->qi_dquots++;
1004 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
1005 mutex_unlock(&h->qh_lock);
1007 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
1008 trace_xfs_dqget_miss(dqp);
1015 * Release a reference to the dquot (decrement ref-count)
1016 * and unlock it. If there is a group quota attached to this
1017 * dquot, carefully release that too without tripping over
1018 * deadlocks'n'stuff.
1026 ASSERT(dqp->q_nrefs > 0);
1027 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1029 trace_xfs_dqput(dqp);
1031 if (dqp->q_nrefs != 1) {
1038 * drop the dqlock and acquire the freelist and dqlock
1039 * in the right order; but try to get it out-of-order first
1041 if (!mutex_trylock(&xfs_Gqm->qm_dqfrlist_lock)) {
1042 trace_xfs_dqput_wait(dqp);
1044 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1051 /* We can't depend on nrefs being == 1 here */
1052 if (--dqp->q_nrefs == 0) {
1053 trace_xfs_dqput_free(dqp);
1055 list_add_tail(&dqp->q_freelist, &xfs_Gqm->qm_dqfrlist);
1056 xfs_Gqm->qm_dqfrlist_cnt++;
1059 * If we just added a udquot to the freelist, then
1060 * we want to release the gdquot reference that
1061 * it (probably) has. Otherwise it'll keep the
1062 * gdquot from getting reclaimed.
1064 if ((gdqp = dqp->q_gdquot)) {
1066 * Avoid a recursive dqput call
1069 dqp->q_gdquot = NULL;
1075 * If we had a group quota inside the user quota as a hint,
1082 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1086 * Release a dquot. Flush it if dirty, then dqput() it.
1087 * dquot must not be locked.
1096 trace_xfs_dqrele(dqp);
1100 * We don't care to flush it if the dquot is dirty here.
1101 * That will create stutters that we want to avoid.
1102 * Instead we do a delayed write when we try to reclaim
1103 * a dirty dquot. Also xfs_sync will take part of the burden...
1109 * This is the dquot flushing I/O completion routine. It is called
1110 * from interrupt level when the buffer containing the dquot is
1111 * flushed to disk. It is responsible for removing the dquot logitem
1112 * from the AIL if it has not been re-logged, and unlocking the dquot's
1113 * flush lock. This behavior is very similar to that of inodes..
1116 xfs_qm_dqflush_done(
1118 struct xfs_log_item *lip)
1120 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
1121 xfs_dquot_t *dqp = qip->qli_dquot;
1122 struct xfs_ail *ailp = lip->li_ailp;
1125 * We only want to pull the item from the AIL if its
1126 * location in the log has not changed since we started the flush.
1127 * Thus, we only bother if the dquot's lsn has
1128 * not changed. First we check the lsn outside the lock
1129 * since it's cheaper, and then we recheck while
1130 * holding the lock before removing the dquot from the AIL.
1132 if ((lip->li_flags & XFS_LI_IN_AIL) &&
1133 lip->li_lsn == qip->qli_flush_lsn) {
1135 /* xfs_trans_ail_delete() drops the AIL lock. */
1136 spin_lock(&ailp->xa_lock);
1137 if (lip->li_lsn == qip->qli_flush_lsn)
1138 xfs_trans_ail_delete(ailp, lip);
1140 spin_unlock(&ailp->xa_lock);
1144 * Release the dq's flush lock since we're done with it.
1150 * Write a modified dquot to disk.
1151 * The dquot must be locked and the flush lock too taken by caller.
1152 * The flush lock will not be unlocked until the dquot reaches the disk,
1153 * but the dquot is free to be unlocked and modified by the caller
1154 * in the interim. Dquot is still locked on return. This behavior is
1155 * identical to that of inodes.
1162 struct xfs_mount *mp = dqp->q_mount;
1164 struct xfs_disk_dquot *ddqp;
1167 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1168 ASSERT(!completion_done(&dqp->q_flush));
1170 trace_xfs_dqflush(dqp);
1173 * If not dirty, or it's pinned and we are not supposed to block, nada.
1175 if (!XFS_DQ_IS_DIRTY(dqp) ||
1176 (!(flags & SYNC_WAIT) && atomic_read(&dqp->q_pincount) > 0)) {
1180 xfs_qm_dqunpin_wait(dqp);
1183 * This may have been unpinned because the filesystem is shutting
1184 * down forcibly. If that's the case we must not write this dquot
1185 * to disk, because the log record didn't make it to disk!
1187 if (XFS_FORCED_SHUTDOWN(mp)) {
1188 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1190 return XFS_ERROR(EIO);
1194 * Get the buffer containing the on-disk dquot
1196 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1197 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
1199 ASSERT(error != ENOENT);
1205 * Calculate the location of the dquot inside the buffer.
1207 ddqp = (struct xfs_disk_dquot *)(XFS_BUF_PTR(bp) + dqp->q_bufoffset);
1210 * A simple sanity check in case we got a corrupted dquot..
1212 error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
1213 XFS_QMOPT_DOWARN, "dqflush (incore copy)");
1217 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1218 return XFS_ERROR(EIO);
1221 /* This is the only portion of data that needs to persist */
1222 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
1225 * Clear the dirty field and remember the flush lsn for later use.
1227 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1229 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1230 &dqp->q_logitem.qli_item.li_lsn);
1233 * Attach an iodone routine so that we can remove this dquot from the
1234 * AIL and release the flush lock once the dquot is synced to disk.
1236 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1237 &dqp->q_logitem.qli_item);
1240 * If the buffer is pinned then push on the log so we won't
1241 * get stuck waiting in the write for too long.
1243 if (XFS_BUF_ISPINNED(bp)) {
1244 trace_xfs_dqflush_force(dqp);
1245 xfs_log_force(mp, 0);
1248 if (flags & SYNC_WAIT)
1249 error = xfs_bwrite(mp, bp);
1251 xfs_bdwrite(mp, bp);
1253 trace_xfs_dqflush_done(dqp);
1256 * dqp is still locked, but caller is free to unlock it now.
1263 xfs_qm_dqlock_nowait(
1266 return mutex_trylock(&dqp->q_qlock);
1273 mutex_lock(&dqp->q_qlock);
1280 mutex_unlock(&(dqp->q_qlock));
1281 if (dqp->q_logitem.qli_dquot == dqp) {
1282 /* Once was dqp->q_mount, but might just have been cleared */
1283 xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_ailp,
1284 (xfs_log_item_t*)&(dqp->q_logitem));
1290 xfs_dqunlock_nonotify(
1293 mutex_unlock(&(dqp->q_qlock));
1297 * Lock two xfs_dquot structures.
1299 * To avoid deadlocks we always lock the quota structure with
1300 * the lowerd id first.
1309 if (be32_to_cpu(d1->q_core.d_id) >
1310 be32_to_cpu(d2->q_core.d_id)) {
1311 mutex_lock(&d2->q_qlock);
1312 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1314 mutex_lock(&d1->q_qlock);
1315 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1318 mutex_lock(&d1->q_qlock);
1320 mutex_lock(&d2->q_qlock);
1326 * Take a dquot out of the mount's dqlist as well as the hashlist.
1327 * This is called via unmount as well as quotaoff, and the purge
1328 * will always succeed unless there are soft (temp) references
1331 * This returns 0 if it was purged, 1 if it wasn't. It's not an error code
1332 * that we're returning! XXXsup - not cool.
1339 xfs_dqhash_t *qh = dqp->q_hash;
1340 xfs_mount_t *mp = dqp->q_mount;
1342 ASSERT(mutex_is_locked(&mp->m_quotainfo->qi_dqlist_lock));
1343 ASSERT(mutex_is_locked(&dqp->q_hash->qh_lock));
1347 * We really can't afford to purge a dquot that is
1348 * referenced, because these are hard refs.
1349 * It shouldn't happen in general because we went thru _all_ inodes in
1350 * dqrele_all_inodes before calling this and didn't let the mountlock go.
1351 * However it is possible that we have dquots with temporary
1352 * references that are not attached to an inode. e.g. see xfs_setattr().
1354 if (dqp->q_nrefs != 0) {
1356 mutex_unlock(&dqp->q_hash->qh_lock);
1360 ASSERT(!list_empty(&dqp->q_freelist));
1363 * If we're turning off quotas, we have to make sure that, for
1364 * example, we don't delete quota disk blocks while dquots are
1365 * in the process of getting written to those disk blocks.
1366 * This dquot might well be on AIL, and we can't leave it there
1367 * if we're turning off quotas. Basically, we need this flush
1368 * lock, and are willing to block on it.
1370 if (!xfs_dqflock_nowait(dqp)) {
1372 * Block on the flush lock after nudging dquot buffer,
1375 xfs_qm_dqflock_pushbuf_wait(dqp);
1379 * XXXIf we're turning this type of quotas off, we don't care
1380 * about the dirty metadata sitting in this dquot. OTOH, if
1381 * we're unmounting, we do care, so we flush it and wait.
1383 if (XFS_DQ_IS_DIRTY(dqp)) {
1386 /* dqflush unlocks dqflock */
1388 * Given that dqpurge is a very rare occurrence, it is OK
1389 * that we're holding the hashlist and mplist locks
1390 * across the disk write. But, ... XXXsup
1392 * We don't care about getting disk errors here. We need
1393 * to purge this dquot anyway, so we go ahead regardless.
1395 error = xfs_qm_dqflush(dqp, SYNC_WAIT);
1397 xfs_warn(mp, "%s: dquot %p flush failed",
1401 ASSERT(atomic_read(&dqp->q_pincount) == 0);
1402 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1403 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
1405 list_del_init(&dqp->q_hashlist);
1407 list_del_init(&dqp->q_mplist);
1408 mp->m_quotainfo->qi_dqreclaims++;
1409 mp->m_quotainfo->qi_dquots--;
1411 * XXX Move this to the front of the freelist, if we can get the
1414 ASSERT(!list_empty(&dqp->q_freelist));
1416 dqp->q_mount = NULL;
1418 dqp->dq_flags = XFS_DQ_INACTIVE;
1419 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
1422 mutex_unlock(&qh->qh_lock);
1428 * Give the buffer a little push if it is incore and
1429 * wait on the flush lock.
1432 xfs_qm_dqflock_pushbuf_wait(
1435 xfs_mount_t *mp = dqp->q_mount;
1439 * Check to see if the dquot has been flushed delayed
1440 * write. If so, grab its buffer and send it
1441 * out immediately. We'll be able to acquire
1442 * the flush lock when the I/O completes.
1444 bp = xfs_incore(mp->m_ddev_targp, dqp->q_blkno,
1445 mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK);
1449 if (XFS_BUF_ISDELAYWRITE(bp)) {
1450 if (XFS_BUF_ISPINNED(bp))
1451 xfs_log_force(mp, 0);
1452 xfs_buf_delwri_promote(bp);
1453 wake_up_process(bp->b_target->bt_task);