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
20 #include "xfs_format.h"
21 #include "xfs_log_format.h"
22 #include "xfs_shared.h"
23 #include "xfs_trans_resv.h"
27 #include "xfs_mount.h"
28 #include "xfs_inode.h"
30 #include "xfs_bmap_util.h"
31 #include "xfs_alloc.h"
32 #include "xfs_quota.h"
33 #include "xfs_error.h"
34 #include "xfs_trans.h"
35 #include "xfs_buf_item.h"
36 #include "xfs_trans_space.h"
37 #include "xfs_trans_priv.h"
39 #include "xfs_cksum.h"
40 #include "xfs_trace.h"
42 #include "xfs_bmap_btree.h"
49 * dquot->q_qlock (xfs_dqlock() and friends)
50 * dquot->q_flush (xfs_dqflock() and friends)
53 * If two dquots need to be locked the order is user before group/project,
54 * otherwise by the lowest id first, see xfs_dqlock2.
58 xfs_buftarg_t *xfs_dqerror_target;
61 int xfs_dqerror_mod = 33;
64 struct kmem_zone *xfs_qm_dqtrxzone;
65 static struct kmem_zone *xfs_qm_dqzone;
67 static struct lock_class_key xfs_dquot_group_class;
68 static struct lock_class_key xfs_dquot_project_class;
71 * This is called to free all the memory associated with a dquot
77 ASSERT(list_empty(&dqp->q_lru));
79 mutex_destroy(&dqp->q_qlock);
80 kmem_zone_free(xfs_qm_dqzone, dqp);
82 XFS_STATS_DEC(xs_qm_dquot);
86 * If default limits are in force, push them into the dquot now.
87 * We overwrite the dquot limits only if they are zero and this
88 * is not the root dquot.
91 xfs_qm_adjust_dqlimits(
95 struct xfs_quotainfo *q = mp->m_quotainfo;
96 struct xfs_disk_dquot *d = &dq->q_core;
101 if (q->qi_bsoftlimit && !d->d_blk_softlimit) {
102 d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit);
105 if (q->qi_bhardlimit && !d->d_blk_hardlimit) {
106 d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit);
109 if (q->qi_isoftlimit && !d->d_ino_softlimit)
110 d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit);
111 if (q->qi_ihardlimit && !d->d_ino_hardlimit)
112 d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit);
113 if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit)
114 d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit);
115 if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit)
116 d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit);
119 xfs_dquot_set_prealloc_limits(dq);
123 * Check the limits and timers of a dquot and start or reset timers
125 * This gets called even when quota enforcement is OFF, which makes our
126 * life a little less complicated. (We just don't reject any quota
127 * reservations in that case, when enforcement is off).
128 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
130 * In contrast, warnings are a little different in that they don't
131 * 'automatically' get started when limits get exceeded. They do
132 * get reset to zero, however, when we find the count to be under
133 * the soft limit (they are only ever set non-zero via userspace).
136 xfs_qm_adjust_dqtimers(
143 if (d->d_blk_hardlimit)
144 ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
145 be64_to_cpu(d->d_blk_hardlimit));
146 if (d->d_ino_hardlimit)
147 ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
148 be64_to_cpu(d->d_ino_hardlimit));
149 if (d->d_rtb_hardlimit)
150 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
151 be64_to_cpu(d->d_rtb_hardlimit));
155 if ((d->d_blk_softlimit &&
156 (be64_to_cpu(d->d_bcount) >
157 be64_to_cpu(d->d_blk_softlimit))) ||
158 (d->d_blk_hardlimit &&
159 (be64_to_cpu(d->d_bcount) >
160 be64_to_cpu(d->d_blk_hardlimit)))) {
161 d->d_btimer = cpu_to_be32(get_seconds() +
162 mp->m_quotainfo->qi_btimelimit);
167 if ((!d->d_blk_softlimit ||
168 (be64_to_cpu(d->d_bcount) <=
169 be64_to_cpu(d->d_blk_softlimit))) &&
170 (!d->d_blk_hardlimit ||
171 (be64_to_cpu(d->d_bcount) <=
172 be64_to_cpu(d->d_blk_hardlimit)))) {
178 if ((d->d_ino_softlimit &&
179 (be64_to_cpu(d->d_icount) >
180 be64_to_cpu(d->d_ino_softlimit))) ||
181 (d->d_ino_hardlimit &&
182 (be64_to_cpu(d->d_icount) >
183 be64_to_cpu(d->d_ino_hardlimit)))) {
184 d->d_itimer = cpu_to_be32(get_seconds() +
185 mp->m_quotainfo->qi_itimelimit);
190 if ((!d->d_ino_softlimit ||
191 (be64_to_cpu(d->d_icount) <=
192 be64_to_cpu(d->d_ino_softlimit))) &&
193 (!d->d_ino_hardlimit ||
194 (be64_to_cpu(d->d_icount) <=
195 be64_to_cpu(d->d_ino_hardlimit)))) {
200 if (!d->d_rtbtimer) {
201 if ((d->d_rtb_softlimit &&
202 (be64_to_cpu(d->d_rtbcount) >
203 be64_to_cpu(d->d_rtb_softlimit))) ||
204 (d->d_rtb_hardlimit &&
205 (be64_to_cpu(d->d_rtbcount) >
206 be64_to_cpu(d->d_rtb_hardlimit)))) {
207 d->d_rtbtimer = cpu_to_be32(get_seconds() +
208 mp->m_quotainfo->qi_rtbtimelimit);
213 if ((!d->d_rtb_softlimit ||
214 (be64_to_cpu(d->d_rtbcount) <=
215 be64_to_cpu(d->d_rtb_softlimit))) &&
216 (!d->d_rtb_hardlimit ||
217 (be64_to_cpu(d->d_rtbcount) <=
218 be64_to_cpu(d->d_rtb_hardlimit)))) {
225 * initialize a buffer full of dquots and log the whole thing
228 xfs_qm_init_dquot_blk(
235 struct xfs_quotainfo *q = mp->m_quotainfo;
240 ASSERT(xfs_buf_islocked(bp));
245 * ID of the first dquot in the block - id's are zero based.
247 curid = id - (id % q->qi_dqperchunk);
249 memset(d, 0, BBTOB(q->qi_dqchunklen));
250 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
251 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
252 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
253 d->dd_diskdq.d_id = cpu_to_be32(curid);
254 d->dd_diskdq.d_flags = type;
255 if (xfs_sb_version_hascrc(&mp->m_sb)) {
256 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid);
257 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
262 xfs_trans_dquot_buf(tp, bp,
263 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
264 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
265 XFS_BLF_GDQUOT_BUF)));
266 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
270 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
271 * watermarks correspond to the soft and hard limits by default. If a soft limit
272 * is not specified, we use 95% of the hard limit.
275 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
279 dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit);
280 dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit);
281 if (!dqp->q_prealloc_lo_wmark) {
282 dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
283 do_div(dqp->q_prealloc_lo_wmark, 100);
284 dqp->q_prealloc_lo_wmark *= 95;
287 space = dqp->q_prealloc_hi_wmark;
290 dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
291 dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
292 dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
296 * Allocate a block and fill it with dquots.
297 * This is called when the bmapi finds a hole.
305 xfs_fileoff_t offset_fsb,
308 xfs_fsblock_t firstblock;
309 xfs_bmap_free_t flist;
311 int nmaps, error, committed;
313 xfs_trans_t *tp = *tpp;
317 trace_xfs_dqalloc(dqp);
320 * Initialize the bmap freelist prior to calling bmapi code.
322 xfs_bmap_init(&flist, &firstblock);
323 xfs_ilock(quotip, XFS_ILOCK_EXCL);
325 * Return if this type of quotas is turned off while we didn't
328 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
329 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
333 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
335 error = xfs_bmapi_write(tp, quotip, offset_fsb,
336 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
337 &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp),
338 &map, &nmaps, &flist);
341 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
343 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
344 (map.br_startblock != HOLESTARTBLOCK));
347 * Keep track of the blkno to save a lookup later
349 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
351 /* now we can just get the buffer (there's nothing to read yet) */
352 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
354 mp->m_quotainfo->qi_dqchunklen,
357 error = xfs_buf_geterror(bp);
360 bp->b_ops = &xfs_dquot_buf_ops;
363 * Make a chunk of dquots out of this buffer and log
366 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
367 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
370 * xfs_bmap_finish() may commit the current transaction and
371 * start a second transaction if the freelist is not empty.
373 * Since we still want to modify this buffer, we need to
374 * ensure that the buffer is not released on commit of
375 * the first transaction and ensure the buffer is added to the
376 * second transaction.
378 * If there is only one transaction then don't stop the buffer
379 * from being released when it commits later on.
382 xfs_trans_bhold(tp, bp);
384 if ((error = xfs_bmap_finish(tpp, &flist, &committed))) {
390 xfs_trans_bjoin(tp, bp);
392 xfs_trans_bhold_release(tp, bp);
399 xfs_bmap_cancel(&flist);
401 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
408 struct xfs_mount *mp,
409 struct xfs_trans *tp,
410 struct xfs_dquot *dqp,
412 struct xfs_buf **bpp)
415 struct xfs_disk_dquot *ddq;
420 * Read the buffer without verification so we get the corrupted
421 * buffer returned to us. make sure we verify it on write, though.
423 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno,
424 mp->m_quotainfo->qi_dqchunklen,
428 ASSERT(*bpp == NULL);
429 return XFS_ERROR(error);
431 (*bpp)->b_ops = &xfs_dquot_buf_ops;
433 ASSERT(xfs_buf_islocked(*bpp));
434 d = (struct xfs_dqblk *)(*bpp)->b_addr;
436 /* Do the actual repair of dquots in this buffer */
437 for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) {
438 ddq = &d[i].dd_diskdq;
439 error = xfs_dqcheck(mp, ddq, firstid + i,
440 dqp->dq_flags & XFS_DQ_ALLTYPES,
441 XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair");
443 /* repair failed, we're screwed */
444 xfs_trans_brelse(tp, *bpp);
445 return XFS_ERROR(EIO);
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,
465 struct xfs_bmbt_irec map;
466 int nmaps = 1, error;
468 struct xfs_inode *quotip = xfs_dq_to_quota_inode(dqp);
469 struct xfs_mount *mp = dqp->q_mount;
470 xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id);
471 struct xfs_trans *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_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
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,
530 0, &bp, &xfs_dquot_buf_ops);
532 if (error == EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) {
533 xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff *
534 mp->m_quotainfo->qi_dqperchunk;
536 error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp);
541 return XFS_ERROR(error);
545 ASSERT(xfs_buf_islocked(bp));
547 *O_ddpp = bp->b_addr + dqp->q_bufoffset;
554 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
555 * and release the buffer immediately.
557 * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed.
561 struct xfs_mount *mp,
565 struct xfs_dquot **O_dqpp)
567 struct xfs_dquot *dqp;
568 struct xfs_disk_dquot *ddqp;
570 struct xfs_trans *tp = NULL;
575 dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP);
577 dqp->dq_flags = type;
578 dqp->q_core.d_id = cpu_to_be32(id);
580 INIT_LIST_HEAD(&dqp->q_lru);
581 mutex_init(&dqp->q_qlock);
582 init_waitqueue_head(&dqp->q_pinwait);
585 * Because we want to use a counting completion, complete
586 * the flush completion once to allow a single access to
587 * the flush completion without blocking.
589 init_completion(&dqp->q_flush);
590 complete(&dqp->q_flush);
593 * Make sure group quotas have a different lock class than user
598 /* uses the default lock class */
601 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
604 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
611 XFS_STATS_INC(xs_qm_dquot);
613 trace_xfs_dqread(dqp);
615 if (flags & XFS_QMOPT_DQALLOC) {
616 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
617 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_attrsetm,
618 XFS_QM_DQALLOC_SPACE_RES(mp), 0);
621 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
625 * get a pointer to the on-disk dquot and the buffer containing it
626 * dqp already knows its own type (GROUP/USER).
628 error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags);
631 * This can happen if quotas got turned off (ESRCH),
632 * or if the dquot didn't exist on disk and we ask to
635 trace_xfs_dqread_fail(dqp);
636 cancelflags |= XFS_TRANS_ABORT;
640 /* copy everything from disk dquot to the incore dquot */
641 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
642 xfs_qm_dquot_logitem_init(dqp);
645 * Reservation counters are defined as reservation plus current usage
646 * to avoid having to add every time.
648 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
649 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
650 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
652 /* initialize the dquot speculative prealloc thresholds */
653 xfs_dquot_set_prealloc_limits(dqp);
655 /* Mark the buf so that this will stay incore a little longer */
656 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
659 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
660 * So we need to release with xfs_trans_brelse().
661 * The strategy here is identical to that of inodes; we lock
662 * the dquot in xfs_qm_dqget() before making it accessible to
663 * others. This is because dquots, like inodes, need a good level of
664 * concurrency, and we don't want to take locks on the entire buffers
665 * for dquot accesses.
666 * Note also that the dquot buffer may even be dirty at this point, if
667 * this particular dquot was repaired. We still aren't afraid to
668 * brelse it because we have the changes incore.
670 ASSERT(xfs_buf_islocked(bp));
671 xfs_trans_brelse(tp, bp);
674 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
684 xfs_trans_cancel(tp, cancelflags);
686 xfs_qm_dqdestroy(dqp);
692 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
693 * a locked dquot, doing an allocation (if requested) as needed.
694 * When both an inode and an id are given, the inode's id takes precedence.
695 * That is, if the id changes while we don't hold the ilock inside this
696 * function, the new dquot is returned, not necessarily the one requested
697 * in the id argument.
702 xfs_inode_t *ip, /* locked inode (optional) */
703 xfs_dqid_t id, /* uid/projid/gid depending on type */
704 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
705 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
706 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
708 struct xfs_quotainfo *qi = mp->m_quotainfo;
709 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
710 struct xfs_dquot *dqp;
713 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
714 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
715 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
716 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
721 if (xfs_do_dqerror) {
722 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
723 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
724 xfs_debug(mp, "Returning error in dqget");
729 ASSERT(type == XFS_DQ_USER ||
730 type == XFS_DQ_PROJ ||
731 type == XFS_DQ_GROUP);
733 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
734 ASSERT(xfs_inode_dquot(ip, type) == NULL);
739 mutex_lock(&qi->qi_tree_lock);
740 dqp = radix_tree_lookup(tree, id);
743 if (dqp->dq_flags & XFS_DQ_FREEING) {
745 mutex_unlock(&qi->qi_tree_lock);
746 trace_xfs_dqget_freeing(dqp);
752 mutex_unlock(&qi->qi_tree_lock);
754 trace_xfs_dqget_hit(dqp);
755 XFS_STATS_INC(xs_qm_dqcachehits);
759 mutex_unlock(&qi->qi_tree_lock);
760 XFS_STATS_INC(xs_qm_dqcachemisses);
763 * Dquot cache miss. We don't want to keep the inode lock across
764 * a (potential) disk read. Also we don't want to deal with the lock
765 * ordering between quotainode and this inode. OTOH, dropping the inode
766 * lock here means dealing with a chown that can happen before
767 * we re-acquire the lock.
770 xfs_iunlock(ip, XFS_ILOCK_EXCL);
772 error = xfs_qm_dqread(mp, id, type, flags, &dqp);
775 xfs_ilock(ip, XFS_ILOCK_EXCL);
782 * A dquot could be attached to this inode by now, since
783 * we had dropped the ilock.
785 if (xfs_this_quota_on(mp, type)) {
786 struct xfs_dquot *dqp1;
788 dqp1 = xfs_inode_dquot(ip, type);
790 xfs_qm_dqdestroy(dqp);
796 /* inode stays locked on return */
797 xfs_qm_dqdestroy(dqp);
798 return XFS_ERROR(ESRCH);
802 mutex_lock(&qi->qi_tree_lock);
803 error = -radix_tree_insert(tree, id, dqp);
804 if (unlikely(error)) {
805 WARN_ON(error != EEXIST);
808 * Duplicate found. Just throw away the new dquot and start
811 mutex_unlock(&qi->qi_tree_lock);
812 trace_xfs_dqget_dup(dqp);
813 xfs_qm_dqdestroy(dqp);
814 XFS_STATS_INC(xs_qm_dquot_dups);
819 * We return a locked dquot to the caller, with a reference taken
825 mutex_unlock(&qi->qi_tree_lock);
828 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
829 trace_xfs_dqget_miss(dqp);
837 struct xfs_dquot *dqp)
839 struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
840 struct xfs_dquot *gdqp;
841 struct xfs_dquot *pdqp;
843 trace_xfs_dqput_free(dqp);
845 if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
846 XFS_STATS_INC(xs_qm_dquot_unused);
849 * If we just added a udquot to the freelist, then we want to release
850 * the gdquot/pdquot reference that it (probably) has. Otherwise it'll
851 * keep the gdquot/pdquot from getting reclaimed.
853 gdqp = dqp->q_gdquot;
856 dqp->q_gdquot = NULL;
859 pdqp = dqp->q_pdquot;
862 dqp->q_pdquot = NULL;
867 * If we had a group/project quota hint, release it now.
876 * Release a reference to the dquot (decrement ref-count) and unlock it.
878 * If there is a group quota attached to this dquot, carefully release that
879 * too without tripping over deadlocks'n'stuff.
883 struct xfs_dquot *dqp)
885 ASSERT(dqp->q_nrefs > 0);
886 ASSERT(XFS_DQ_IS_LOCKED(dqp));
888 trace_xfs_dqput(dqp);
890 if (--dqp->q_nrefs > 0)
893 xfs_qm_dqput_final(dqp);
897 * Release a dquot. Flush it if dirty, then dqput() it.
898 * dquot must not be locked.
907 trace_xfs_dqrele(dqp);
911 * We don't care to flush it if the dquot is dirty here.
912 * That will create stutters that we want to avoid.
913 * Instead we do a delayed write when we try to reclaim
914 * a dirty dquot. Also xfs_sync will take part of the burden...
920 * This is the dquot flushing I/O completion routine. It is called
921 * from interrupt level when the buffer containing the dquot is
922 * flushed to disk. It is responsible for removing the dquot logitem
923 * from the AIL if it has not been re-logged, and unlocking the dquot's
924 * flush lock. This behavior is very similar to that of inodes..
929 struct xfs_log_item *lip)
931 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
932 xfs_dquot_t *dqp = qip->qli_dquot;
933 struct xfs_ail *ailp = lip->li_ailp;
936 * We only want to pull the item from the AIL if its
937 * location in the log has not changed since we started the flush.
938 * Thus, we only bother if the dquot's lsn has
939 * not changed. First we check the lsn outside the lock
940 * since it's cheaper, and then we recheck while
941 * holding the lock before removing the dquot from the AIL.
943 if ((lip->li_flags & XFS_LI_IN_AIL) &&
944 lip->li_lsn == qip->qli_flush_lsn) {
946 /* xfs_trans_ail_delete() drops the AIL lock. */
947 spin_lock(&ailp->xa_lock);
948 if (lip->li_lsn == qip->qli_flush_lsn)
949 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
951 spin_unlock(&ailp->xa_lock);
955 * Release the dq's flush lock since we're done with it.
961 * Write a modified dquot to disk.
962 * The dquot must be locked and the flush lock too taken by caller.
963 * The flush lock will not be unlocked until the dquot reaches the disk,
964 * but the dquot is free to be unlocked and modified by the caller
965 * in the interim. Dquot is still locked on return. This behavior is
966 * identical to that of inodes.
970 struct xfs_dquot *dqp,
971 struct xfs_buf **bpp)
973 struct xfs_mount *mp = dqp->q_mount;
975 struct xfs_disk_dquot *ddqp;
978 ASSERT(XFS_DQ_IS_LOCKED(dqp));
979 ASSERT(!completion_done(&dqp->q_flush));
981 trace_xfs_dqflush(dqp);
985 xfs_qm_dqunpin_wait(dqp);
988 * This may have been unpinned because the filesystem is shutting
989 * down forcibly. If that's the case we must not write this dquot
990 * to disk, because the log record didn't make it to disk.
992 * We also have to remove the log item from the AIL in this case,
993 * as we wait for an emptry AIL as part of the unmount process.
995 if (XFS_FORCED_SHUTDOWN(mp)) {
996 struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
997 dqp->dq_flags &= ~XFS_DQ_DIRTY;
999 spin_lock(&mp->m_ail->xa_lock);
1000 if (lip->li_flags & XFS_LI_IN_AIL)
1001 xfs_trans_ail_delete(mp->m_ail, lip,
1002 SHUTDOWN_CORRUPT_INCORE);
1004 spin_unlock(&mp->m_ail->xa_lock);
1005 error = XFS_ERROR(EIO);
1010 * Get the buffer containing the on-disk dquot
1012 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1013 mp->m_quotainfo->qi_dqchunklen, 0, &bp, NULL);
1018 * Calculate the location of the dquot inside the buffer.
1020 ddqp = bp->b_addr + dqp->q_bufoffset;
1023 * A simple sanity check in case we got a corrupted dquot..
1025 error = xfs_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
1026 XFS_QMOPT_DOWARN, "dqflush (incore copy)");
1030 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1031 return XFS_ERROR(EIO);
1034 /* This is the only portion of data that needs to persist */
1035 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
1038 * Clear the dirty field and remember the flush lsn for later use.
1040 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1042 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1043 &dqp->q_logitem.qli_item.li_lsn);
1046 * copy the lsn into the on-disk dquot now while we have the in memory
1047 * dquot here. This can't be done later in the write verifier as we
1048 * can't get access to the log item at that point in time.
1050 * We also calculate the CRC here so that the on-disk dquot in the
1051 * buffer always has a valid CRC. This ensures there is no possibility
1052 * of a dquot without an up-to-date CRC getting to disk.
1054 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1055 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp;
1057 dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1058 xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk),
1063 * Attach an iodone routine so that we can remove this dquot from the
1064 * AIL and release the flush lock once the dquot is synced to disk.
1066 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1067 &dqp->q_logitem.qli_item);
1070 * If the buffer is pinned then push on the log so we won't
1071 * get stuck waiting in the write for too long.
1073 if (xfs_buf_ispinned(bp)) {
1074 trace_xfs_dqflush_force(dqp);
1075 xfs_log_force(mp, 0);
1078 trace_xfs_dqflush_done(dqp);
1084 return XFS_ERROR(EIO);
1088 * Lock two xfs_dquot structures.
1090 * To avoid deadlocks we always lock the quota structure with
1091 * the lowerd id first.
1100 if (be32_to_cpu(d1->q_core.d_id) >
1101 be32_to_cpu(d2->q_core.d_id)) {
1102 mutex_lock(&d2->q_qlock);
1103 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1105 mutex_lock(&d1->q_qlock);
1106 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1109 mutex_lock(&d1->q_qlock);
1111 mutex_lock(&d2->q_qlock);
1119 kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
1124 kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
1125 if (!xfs_qm_dqtrxzone)
1126 goto out_free_dqzone;
1131 kmem_zone_destroy(xfs_qm_dqzone);
1139 kmem_zone_destroy(xfs_qm_dqtrxzone);
1140 kmem_zone_destroy(xfs_qm_dqzone);