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"
25 #include "xfs_mount.h"
26 #include "xfs_defer.h"
27 #include "xfs_inode.h"
29 #include "xfs_bmap_util.h"
30 #include "xfs_alloc.h"
31 #include "xfs_quota.h"
32 #include "xfs_error.h"
33 #include "xfs_trans.h"
34 #include "xfs_buf_item.h"
35 #include "xfs_trans_space.h"
36 #include "xfs_trans_priv.h"
38 #include "xfs_cksum.h"
39 #include "xfs_trace.h"
41 #include "xfs_bmap_btree.h"
48 * dquot->q_qlock (xfs_dqlock() and friends)
49 * dquot->q_flush (xfs_dqflock() and friends)
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.
56 struct kmem_zone *xfs_qm_dqtrxzone;
57 static struct kmem_zone *xfs_qm_dqzone;
59 static struct lock_class_key xfs_dquot_group_class;
60 static struct lock_class_key xfs_dquot_project_class;
63 * This is called to free all the memory associated with a dquot
69 ASSERT(list_empty(&dqp->q_lru));
71 kmem_free(dqp->q_logitem.qli_item.li_lv_shadow);
72 mutex_destroy(&dqp->q_qlock);
74 XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot);
75 kmem_zone_free(xfs_qm_dqzone, dqp);
79 * If default limits are in force, push them into the dquot now.
80 * We overwrite the dquot limits only if they are zero and this
81 * is not the root dquot.
84 xfs_qm_adjust_dqlimits(
88 struct xfs_quotainfo *q = mp->m_quotainfo;
89 struct xfs_disk_dquot *d = &dq->q_core;
90 struct xfs_def_quota *defq;
94 defq = xfs_get_defquota(dq, q);
96 if (defq->bsoftlimit && !d->d_blk_softlimit) {
97 d->d_blk_softlimit = cpu_to_be64(defq->bsoftlimit);
100 if (defq->bhardlimit && !d->d_blk_hardlimit) {
101 d->d_blk_hardlimit = cpu_to_be64(defq->bhardlimit);
104 if (defq->isoftlimit && !d->d_ino_softlimit)
105 d->d_ino_softlimit = cpu_to_be64(defq->isoftlimit);
106 if (defq->ihardlimit && !d->d_ino_hardlimit)
107 d->d_ino_hardlimit = cpu_to_be64(defq->ihardlimit);
108 if (defq->rtbsoftlimit && !d->d_rtb_softlimit)
109 d->d_rtb_softlimit = cpu_to_be64(defq->rtbsoftlimit);
110 if (defq->rtbhardlimit && !d->d_rtb_hardlimit)
111 d->d_rtb_hardlimit = cpu_to_be64(defq->rtbhardlimit);
114 xfs_dquot_set_prealloc_limits(dq);
118 * Check the limits and timers of a dquot and start or reset timers
120 * This gets called even when quota enforcement is OFF, which makes our
121 * life a little less complicated. (We just don't reject any quota
122 * reservations in that case, when enforcement is off).
123 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
125 * In contrast, warnings are a little different in that they don't
126 * 'automatically' get started when limits get exceeded. They do
127 * get reset to zero, however, when we find the count to be under
128 * the soft limit (they are only ever set non-zero via userspace).
131 xfs_qm_adjust_dqtimers(
138 if (d->d_blk_hardlimit)
139 ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
140 be64_to_cpu(d->d_blk_hardlimit));
141 if (d->d_ino_hardlimit)
142 ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
143 be64_to_cpu(d->d_ino_hardlimit));
144 if (d->d_rtb_hardlimit)
145 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
146 be64_to_cpu(d->d_rtb_hardlimit));
150 if ((d->d_blk_softlimit &&
151 (be64_to_cpu(d->d_bcount) >
152 be64_to_cpu(d->d_blk_softlimit))) ||
153 (d->d_blk_hardlimit &&
154 (be64_to_cpu(d->d_bcount) >
155 be64_to_cpu(d->d_blk_hardlimit)))) {
156 d->d_btimer = cpu_to_be32(get_seconds() +
157 mp->m_quotainfo->qi_btimelimit);
162 if ((!d->d_blk_softlimit ||
163 (be64_to_cpu(d->d_bcount) <=
164 be64_to_cpu(d->d_blk_softlimit))) &&
165 (!d->d_blk_hardlimit ||
166 (be64_to_cpu(d->d_bcount) <=
167 be64_to_cpu(d->d_blk_hardlimit)))) {
173 if ((d->d_ino_softlimit &&
174 (be64_to_cpu(d->d_icount) >
175 be64_to_cpu(d->d_ino_softlimit))) ||
176 (d->d_ino_hardlimit &&
177 (be64_to_cpu(d->d_icount) >
178 be64_to_cpu(d->d_ino_hardlimit)))) {
179 d->d_itimer = cpu_to_be32(get_seconds() +
180 mp->m_quotainfo->qi_itimelimit);
185 if ((!d->d_ino_softlimit ||
186 (be64_to_cpu(d->d_icount) <=
187 be64_to_cpu(d->d_ino_softlimit))) &&
188 (!d->d_ino_hardlimit ||
189 (be64_to_cpu(d->d_icount) <=
190 be64_to_cpu(d->d_ino_hardlimit)))) {
195 if (!d->d_rtbtimer) {
196 if ((d->d_rtb_softlimit &&
197 (be64_to_cpu(d->d_rtbcount) >
198 be64_to_cpu(d->d_rtb_softlimit))) ||
199 (d->d_rtb_hardlimit &&
200 (be64_to_cpu(d->d_rtbcount) >
201 be64_to_cpu(d->d_rtb_hardlimit)))) {
202 d->d_rtbtimer = cpu_to_be32(get_seconds() +
203 mp->m_quotainfo->qi_rtbtimelimit);
208 if ((!d->d_rtb_softlimit ||
209 (be64_to_cpu(d->d_rtbcount) <=
210 be64_to_cpu(d->d_rtb_softlimit))) &&
211 (!d->d_rtb_hardlimit ||
212 (be64_to_cpu(d->d_rtbcount) <=
213 be64_to_cpu(d->d_rtb_hardlimit)))) {
220 * initialize a buffer full of dquots and log the whole thing
223 xfs_qm_init_dquot_blk(
230 struct xfs_quotainfo *q = mp->m_quotainfo;
236 ASSERT(xfs_buf_islocked(bp));
241 * ID of the first dquot in the block - id's are zero based.
243 curid = id - (id % q->qi_dqperchunk);
244 memset(d, 0, BBTOB(q->qi_dqchunklen));
245 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
246 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
247 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
248 d->dd_diskdq.d_id = cpu_to_be32(curid);
249 d->dd_diskdq.d_flags = type;
250 if (xfs_sb_version_hascrc(&mp->m_sb)) {
251 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
252 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
257 xfs_trans_dquot_buf(tp, bp,
258 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
259 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
260 XFS_BLF_GDQUOT_BUF)));
261 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
265 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
266 * watermarks correspond to the soft and hard limits by default. If a soft limit
267 * is not specified, we use 95% of the hard limit.
270 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
274 dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit);
275 dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit);
276 if (!dqp->q_prealloc_lo_wmark) {
277 dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
278 do_div(dqp->q_prealloc_lo_wmark, 100);
279 dqp->q_prealloc_lo_wmark *= 95;
282 space = dqp->q_prealloc_hi_wmark;
285 dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
286 dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
287 dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
291 * Allocate a block and fill it with dquots.
292 * This is called when the bmapi finds a hole.
300 xfs_fileoff_t offset_fsb,
303 xfs_fsblock_t firstblock;
304 struct xfs_defer_ops dfops;
308 xfs_trans_t *tp = *tpp;
312 trace_xfs_dqalloc(dqp);
315 * Initialize the bmap freelist prior to calling bmapi code.
317 xfs_defer_init(&dfops, &firstblock);
318 xfs_ilock(quotip, XFS_ILOCK_EXCL);
320 * Return if this type of quotas is turned off while we didn't
323 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
324 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
328 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
330 error = xfs_bmapi_write(tp, quotip, offset_fsb,
331 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
332 &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp),
333 &map, &nmaps, &dfops);
336 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
338 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
339 (map.br_startblock != HOLESTARTBLOCK));
342 * Keep track of the blkno to save a lookup later
344 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
346 /* now we can just get the buffer (there's nothing to read yet) */
347 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
349 mp->m_quotainfo->qi_dqchunklen,
355 bp->b_ops = &xfs_dquot_buf_ops;
358 * Make a chunk of dquots out of this buffer and log
361 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
362 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
365 * xfs_defer_finish() may commit the current transaction and
366 * start a second transaction if the freelist is not empty.
368 * Since we still want to modify this buffer, we need to
369 * ensure that the buffer is not released on commit of
370 * the first transaction and ensure the buffer is added to the
371 * second transaction.
373 * If there is only one transaction then don't stop the buffer
374 * from being released when it commits later on.
377 xfs_trans_bhold(tp, bp);
379 error = xfs_defer_finish(tpp, &dfops);
383 /* Transaction was committed? */
386 xfs_trans_bjoin(tp, bp);
388 xfs_trans_bhold_release(tp, bp);
395 xfs_defer_cancel(&dfops);
397 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
404 struct xfs_mount *mp,
405 struct xfs_trans *tp,
406 struct xfs_dquot *dqp,
408 struct xfs_buf **bpp)
411 struct xfs_disk_dquot *ddq;
416 * Read the buffer without verification so we get the corrupted
417 * buffer returned to us. make sure we verify it on write, though.
419 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno,
420 mp->m_quotainfo->qi_dqchunklen,
424 ASSERT(*bpp == NULL);
427 (*bpp)->b_ops = &xfs_dquot_buf_ops;
429 ASSERT(xfs_buf_islocked(*bpp));
430 d = (struct xfs_dqblk *)(*bpp)->b_addr;
432 /* Do the actual repair of dquots in this buffer */
433 for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) {
434 ddq = &d[i].dd_diskdq;
435 error = xfs_dqcheck(mp, ddq, firstid + i,
436 dqp->dq_flags & XFS_DQ_ALLTYPES,
437 XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair");
439 /* repair failed, we're screwed */
440 xfs_trans_brelse(tp, *bpp);
449 * Maps a dquot to the buffer containing its on-disk version.
450 * This returns a ptr to the buffer containing the on-disk dquot
451 * in the bpp param, and a ptr to the on-disk dquot within that buffer
457 xfs_disk_dquot_t **O_ddpp,
461 struct xfs_bmbt_irec map;
462 int nmaps = 1, error;
464 struct xfs_inode *quotip;
465 struct xfs_mount *mp = dqp->q_mount;
466 xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id);
467 struct xfs_trans *tp = (tpp ? *tpp : NULL);
470 quotip = xfs_quota_inode(dqp->q_mount, dqp->dq_flags);
471 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
473 lock_mode = xfs_ilock_data_map_shared(quotip);
474 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
476 * Return if this type of quotas is turned off while we
477 * didn't have the quota inode lock.
479 xfs_iunlock(quotip, lock_mode);
484 * Find the block map; no allocations yet
486 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
487 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
489 xfs_iunlock(quotip, lock_mode);
494 ASSERT(map.br_blockcount == 1);
497 * Offset of dquot in the (fixed sized) dquot chunk.
499 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
502 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
503 if (map.br_startblock == HOLESTARTBLOCK) {
505 * We don't allocate unless we're asked to
507 if (!(flags & XFS_QMOPT_DQALLOC))
511 error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
512 dqp->q_fileoffset, &bp);
517 trace_xfs_dqtobp_read(dqp);
520 * store the blkno etc so that we don't have to do the
521 * mapping all the time
523 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
525 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
527 mp->m_quotainfo->qi_dqchunklen,
528 0, &bp, &xfs_dquot_buf_ops);
530 if (error == -EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) {
531 xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff *
532 mp->m_quotainfo->qi_dqperchunk;
534 error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp);
543 ASSERT(xfs_buf_islocked(bp));
545 *O_ddpp = bp->b_addr + dqp->q_bufoffset;
552 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
553 * and release the buffer immediately.
555 * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed.
559 struct xfs_mount *mp,
563 struct xfs_dquot **O_dqpp)
565 struct xfs_dquot *dqp;
566 struct xfs_disk_dquot *ddqp;
568 struct xfs_trans *tp = NULL;
571 dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP);
573 dqp->dq_flags = type;
574 dqp->q_core.d_id = cpu_to_be32(id);
576 INIT_LIST_HEAD(&dqp->q_lru);
577 mutex_init(&dqp->q_qlock);
578 init_waitqueue_head(&dqp->q_pinwait);
581 * Because we want to use a counting completion, complete
582 * the flush completion once to allow a single access to
583 * the flush completion without blocking.
585 init_completion(&dqp->q_flush);
586 complete(&dqp->q_flush);
589 * Make sure group quotas have a different lock class than user
594 /* uses the default lock class */
597 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
600 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
607 XFS_STATS_INC(mp, xs_qm_dquot);
609 trace_xfs_dqread(dqp);
611 if (flags & XFS_QMOPT_DQALLOC) {
612 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
613 XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
619 * get a pointer to the on-disk dquot and the buffer containing it
620 * dqp already knows its own type (GROUP/USER).
622 error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags);
625 * This can happen if quotas got turned off (ESRCH),
626 * or if the dquot didn't exist on disk and we ask to
629 trace_xfs_dqread_fail(dqp);
633 /* copy everything from disk dquot to the incore dquot */
634 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
635 xfs_qm_dquot_logitem_init(dqp);
638 * Reservation counters are defined as reservation plus current usage
639 * to avoid having to add every time.
641 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
642 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
643 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
645 /* initialize the dquot speculative prealloc thresholds */
646 xfs_dquot_set_prealloc_limits(dqp);
648 /* Mark the buf so that this will stay incore a little longer */
649 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
652 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
653 * So we need to release with xfs_trans_brelse().
654 * The strategy here is identical to that of inodes; we lock
655 * the dquot in xfs_qm_dqget() before making it accessible to
656 * others. This is because dquots, like inodes, need a good level of
657 * concurrency, and we don't want to take locks on the entire buffers
658 * for dquot accesses.
659 * Note also that the dquot buffer may even be dirty at this point, if
660 * this particular dquot was repaired. We still aren't afraid to
661 * brelse it because we have the changes incore.
663 ASSERT(xfs_buf_islocked(bp));
664 xfs_trans_brelse(tp, bp);
667 error = xfs_trans_commit(tp);
677 xfs_trans_cancel(tp);
679 xfs_qm_dqdestroy(dqp);
685 * Advance to the next id in the current chunk, or if at the
686 * end of the chunk, skip ahead to first id in next allocated chunk
687 * using the SEEK_DATA interface.
691 struct xfs_mount *mp,
695 struct xfs_inode *quotip = xfs_quota_inode(mp, type);
696 xfs_dqid_t next_id = *id + 1; /* simple advance */
698 struct xfs_bmbt_irec got;
699 struct xfs_iext_cursor cur;
703 /* If we'd wrap past the max ID, stop */
707 /* If new ID is within the current chunk, advancing it sufficed */
708 if (next_id % mp->m_quotainfo->qi_dqperchunk) {
713 /* Nope, next_id is now past the current chunk, so find the next one */
714 start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
716 lock_flags = xfs_ilock_data_map_shared(quotip);
717 if (!(quotip->i_df.if_flags & XFS_IFEXTENTS)) {
718 error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
723 if (xfs_iext_lookup_extent(quotip, "ip->i_df, start, &cur, &got)) {
724 /* contiguous chunk, bump startoff for the id calculation */
725 if (got.br_startoff < start)
726 got.br_startoff = start;
727 *id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
732 xfs_iunlock(quotip, lock_flags);
738 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
739 * a locked dquot, doing an allocation (if requested) as needed.
740 * When both an inode and an id are given, the inode's id takes precedence.
741 * That is, if the id changes while we don't hold the ilock inside this
742 * function, the new dquot is returned, not necessarily the one requested
743 * in the id argument.
748 xfs_inode_t *ip, /* locked inode (optional) */
749 xfs_dqid_t id, /* uid/projid/gid depending on type */
750 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
751 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
752 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
754 struct xfs_quotainfo *qi = mp->m_quotainfo;
755 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
756 struct xfs_dquot *dqp;
759 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
760 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
761 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
762 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
766 ASSERT(type == XFS_DQ_USER ||
767 type == XFS_DQ_PROJ ||
768 type == XFS_DQ_GROUP);
770 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
771 ASSERT(xfs_inode_dquot(ip, type) == NULL);
775 mutex_lock(&qi->qi_tree_lock);
776 dqp = radix_tree_lookup(tree, id);
779 if (dqp->dq_flags & XFS_DQ_FREEING) {
781 mutex_unlock(&qi->qi_tree_lock);
782 trace_xfs_dqget_freeing(dqp);
787 /* uninit / unused quota found in radix tree, keep looking */
788 if (flags & XFS_QMOPT_DQNEXT) {
789 if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
791 mutex_unlock(&qi->qi_tree_lock);
792 error = xfs_dq_get_next_id(mp, type, &id);
800 mutex_unlock(&qi->qi_tree_lock);
802 trace_xfs_dqget_hit(dqp);
803 XFS_STATS_INC(mp, xs_qm_dqcachehits);
807 mutex_unlock(&qi->qi_tree_lock);
808 XFS_STATS_INC(mp, xs_qm_dqcachemisses);
811 * Dquot cache miss. We don't want to keep the inode lock across
812 * a (potential) disk read. Also we don't want to deal with the lock
813 * ordering between quotainode and this inode. OTOH, dropping the inode
814 * lock here means dealing with a chown that can happen before
815 * we re-acquire the lock.
818 xfs_iunlock(ip, XFS_ILOCK_EXCL);
820 error = xfs_qm_dqread(mp, id, type, flags, &dqp);
823 xfs_ilock(ip, XFS_ILOCK_EXCL);
825 /* If we are asked to find next active id, keep looking */
826 if (error == -ENOENT && (flags & XFS_QMOPT_DQNEXT)) {
827 error = xfs_dq_get_next_id(mp, type, &id);
837 * A dquot could be attached to this inode by now, since
838 * we had dropped the ilock.
840 if (xfs_this_quota_on(mp, type)) {
841 struct xfs_dquot *dqp1;
843 dqp1 = xfs_inode_dquot(ip, type);
845 xfs_qm_dqdestroy(dqp);
851 /* inode stays locked on return */
852 xfs_qm_dqdestroy(dqp);
857 mutex_lock(&qi->qi_tree_lock);
858 error = radix_tree_insert(tree, id, dqp);
859 if (unlikely(error)) {
860 WARN_ON(error != -EEXIST);
863 * Duplicate found. Just throw away the new dquot and start
866 mutex_unlock(&qi->qi_tree_lock);
867 trace_xfs_dqget_dup(dqp);
868 xfs_qm_dqdestroy(dqp);
869 XFS_STATS_INC(mp, xs_qm_dquot_dups);
874 * We return a locked dquot to the caller, with a reference taken
880 mutex_unlock(&qi->qi_tree_lock);
882 /* If we are asked to find next active id, keep looking */
883 if (flags & XFS_QMOPT_DQNEXT) {
884 if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
886 error = xfs_dq_get_next_id(mp, type, &id);
894 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
895 trace_xfs_dqget_miss(dqp);
901 * Release a reference to the dquot (decrement ref-count) and unlock it.
903 * If there is a group quota attached to this dquot, carefully release that
904 * too without tripping over deadlocks'n'stuff.
908 struct xfs_dquot *dqp)
910 ASSERT(dqp->q_nrefs > 0);
911 ASSERT(XFS_DQ_IS_LOCKED(dqp));
913 trace_xfs_dqput(dqp);
915 if (--dqp->q_nrefs == 0) {
916 struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
917 trace_xfs_dqput_free(dqp);
919 if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
920 XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
926 * Release a dquot. Flush it if dirty, then dqput() it.
927 * dquot must not be locked.
936 trace_xfs_dqrele(dqp);
940 * We don't care to flush it if the dquot is dirty here.
941 * That will create stutters that we want to avoid.
942 * Instead we do a delayed write when we try to reclaim
943 * a dirty dquot. Also xfs_sync will take part of the burden...
949 * This is the dquot flushing I/O completion routine. It is called
950 * from interrupt level when the buffer containing the dquot is
951 * flushed to disk. It is responsible for removing the dquot logitem
952 * from the AIL if it has not been re-logged, and unlocking the dquot's
953 * flush lock. This behavior is very similar to that of inodes..
958 struct xfs_log_item *lip)
960 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
961 xfs_dquot_t *dqp = qip->qli_dquot;
962 struct xfs_ail *ailp = lip->li_ailp;
965 * We only want to pull the item from the AIL if its
966 * location in the log has not changed since we started the flush.
967 * Thus, we only bother if the dquot's lsn has
968 * not changed. First we check the lsn outside the lock
969 * since it's cheaper, and then we recheck while
970 * holding the lock before removing the dquot from the AIL.
972 if ((lip->li_flags & XFS_LI_IN_AIL) &&
973 lip->li_lsn == qip->qli_flush_lsn) {
975 /* xfs_trans_ail_delete() drops the AIL lock. */
976 spin_lock(&ailp->xa_lock);
977 if (lip->li_lsn == qip->qli_flush_lsn)
978 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
980 spin_unlock(&ailp->xa_lock);
984 * Release the dq's flush lock since we're done with it.
990 * Write a modified dquot to disk.
991 * The dquot must be locked and the flush lock too taken by caller.
992 * The flush lock will not be unlocked until the dquot reaches the disk,
993 * but the dquot is free to be unlocked and modified by the caller
994 * in the interim. Dquot is still locked on return. This behavior is
995 * identical to that of inodes.
999 struct xfs_dquot *dqp,
1000 struct xfs_buf **bpp)
1002 struct xfs_mount *mp = dqp->q_mount;
1004 struct xfs_disk_dquot *ddqp;
1007 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1008 ASSERT(!completion_done(&dqp->q_flush));
1010 trace_xfs_dqflush(dqp);
1014 xfs_qm_dqunpin_wait(dqp);
1017 * This may have been unpinned because the filesystem is shutting
1018 * down forcibly. If that's the case we must not write this dquot
1019 * to disk, because the log record didn't make it to disk.
1021 * We also have to remove the log item from the AIL in this case,
1022 * as we wait for an emptry AIL as part of the unmount process.
1024 if (XFS_FORCED_SHUTDOWN(mp)) {
1025 struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
1026 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1028 xfs_trans_ail_remove(lip, SHUTDOWN_CORRUPT_INCORE);
1035 * Get the buffer containing the on-disk dquot
1037 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1038 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1039 &xfs_dquot_buf_ops);
1044 * Calculate the location of the dquot inside the buffer.
1046 ddqp = bp->b_addr + dqp->q_bufoffset;
1049 * A simple sanity check in case we got a corrupted dquot..
1051 error = xfs_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
1052 XFS_QMOPT_DOWARN, "dqflush (incore copy)");
1056 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1060 /* This is the only portion of data that needs to persist */
1061 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
1064 * Clear the dirty field and remember the flush lsn for later use.
1066 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1068 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1069 &dqp->q_logitem.qli_item.li_lsn);
1072 * copy the lsn into the on-disk dquot now while we have the in memory
1073 * dquot here. This can't be done later in the write verifier as we
1074 * can't get access to the log item at that point in time.
1076 * We also calculate the CRC here so that the on-disk dquot in the
1077 * buffer always has a valid CRC. This ensures there is no possibility
1078 * of a dquot without an up-to-date CRC getting to disk.
1080 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1081 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp;
1083 dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1084 xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk),
1089 * Attach an iodone routine so that we can remove this dquot from the
1090 * AIL and release the flush lock once the dquot is synced to disk.
1092 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1093 &dqp->q_logitem.qli_item);
1096 * If the buffer is pinned then push on the log so we won't
1097 * get stuck waiting in the write for too long.
1099 if (xfs_buf_ispinned(bp)) {
1100 trace_xfs_dqflush_force(dqp);
1101 xfs_log_force(mp, 0);
1104 trace_xfs_dqflush_done(dqp);
1114 * Lock two xfs_dquot structures.
1116 * To avoid deadlocks we always lock the quota structure with
1117 * the lowerd id first.
1126 if (be32_to_cpu(d1->q_core.d_id) >
1127 be32_to_cpu(d2->q_core.d_id)) {
1128 mutex_lock(&d2->q_qlock);
1129 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1131 mutex_lock(&d1->q_qlock);
1132 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1135 mutex_lock(&d1->q_qlock);
1137 mutex_lock(&d2->q_qlock);
1145 kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
1150 kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
1151 if (!xfs_qm_dqtrxzone)
1152 goto out_free_dqzone;
1157 kmem_zone_destroy(xfs_qm_dqzone);
1165 kmem_zone_destroy(xfs_qm_dqtrxzone);
1166 kmem_zone_destroy(xfs_qm_dqzone);