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"
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_cksum.h"
41 #include "xfs_trace.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.
57 xfs_buftarg_t *xfs_dqerror_target;
60 int xfs_dqerror_mod = 33;
63 struct kmem_zone *xfs_qm_dqtrxzone;
64 static struct kmem_zone *xfs_qm_dqzone;
66 static struct lock_class_key xfs_dquot_other_class;
69 * This is called to free all the memory associated with a dquot
75 ASSERT(list_empty(&dqp->q_lru));
77 mutex_destroy(&dqp->q_qlock);
78 kmem_zone_free(xfs_qm_dqzone, dqp);
80 XFS_STATS_DEC(xs_qm_dquot);
84 * If default limits are in force, push them into the dquot now.
85 * We overwrite the dquot limits only if they are zero and this
86 * is not the root dquot.
89 xfs_qm_adjust_dqlimits(
93 struct xfs_quotainfo *q = mp->m_quotainfo;
94 struct xfs_disk_dquot *d = &dq->q_core;
99 if (q->qi_bsoftlimit && !d->d_blk_softlimit) {
100 d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit);
103 if (q->qi_bhardlimit && !d->d_blk_hardlimit) {
104 d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit);
107 if (q->qi_isoftlimit && !d->d_ino_softlimit)
108 d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit);
109 if (q->qi_ihardlimit && !d->d_ino_hardlimit)
110 d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit);
111 if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit)
112 d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit);
113 if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit)
114 d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit);
117 xfs_dquot_set_prealloc_limits(dq);
121 * Check the limits and timers of a dquot and start or reset timers
123 * This gets called even when quota enforcement is OFF, which makes our
124 * life a little less complicated. (We just don't reject any quota
125 * reservations in that case, when enforcement is off).
126 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
128 * In contrast, warnings are a little different in that they don't
129 * 'automatically' get started when limits get exceeded. They do
130 * get reset to zero, however, when we find the count to be under
131 * the soft limit (they are only ever set non-zero via userspace).
134 xfs_qm_adjust_dqtimers(
141 if (d->d_blk_hardlimit)
142 ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
143 be64_to_cpu(d->d_blk_hardlimit));
144 if (d->d_ino_hardlimit)
145 ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
146 be64_to_cpu(d->d_ino_hardlimit));
147 if (d->d_rtb_hardlimit)
148 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
149 be64_to_cpu(d->d_rtb_hardlimit));
153 if ((d->d_blk_softlimit &&
154 (be64_to_cpu(d->d_bcount) >
155 be64_to_cpu(d->d_blk_softlimit))) ||
156 (d->d_blk_hardlimit &&
157 (be64_to_cpu(d->d_bcount) >
158 be64_to_cpu(d->d_blk_hardlimit)))) {
159 d->d_btimer = cpu_to_be32(get_seconds() +
160 mp->m_quotainfo->qi_btimelimit);
165 if ((!d->d_blk_softlimit ||
166 (be64_to_cpu(d->d_bcount) <=
167 be64_to_cpu(d->d_blk_softlimit))) &&
168 (!d->d_blk_hardlimit ||
169 (be64_to_cpu(d->d_bcount) <=
170 be64_to_cpu(d->d_blk_hardlimit)))) {
176 if ((d->d_ino_softlimit &&
177 (be64_to_cpu(d->d_icount) >
178 be64_to_cpu(d->d_ino_softlimit))) ||
179 (d->d_ino_hardlimit &&
180 (be64_to_cpu(d->d_icount) >
181 be64_to_cpu(d->d_ino_hardlimit)))) {
182 d->d_itimer = cpu_to_be32(get_seconds() +
183 mp->m_quotainfo->qi_itimelimit);
188 if ((!d->d_ino_softlimit ||
189 (be64_to_cpu(d->d_icount) <=
190 be64_to_cpu(d->d_ino_softlimit))) &&
191 (!d->d_ino_hardlimit ||
192 (be64_to_cpu(d->d_icount) <=
193 be64_to_cpu(d->d_ino_hardlimit)))) {
198 if (!d->d_rtbtimer) {
199 if ((d->d_rtb_softlimit &&
200 (be64_to_cpu(d->d_rtbcount) >
201 be64_to_cpu(d->d_rtb_softlimit))) ||
202 (d->d_rtb_hardlimit &&
203 (be64_to_cpu(d->d_rtbcount) >
204 be64_to_cpu(d->d_rtb_hardlimit)))) {
205 d->d_rtbtimer = cpu_to_be32(get_seconds() +
206 mp->m_quotainfo->qi_rtbtimelimit);
211 if ((!d->d_rtb_softlimit ||
212 (be64_to_cpu(d->d_rtbcount) <=
213 be64_to_cpu(d->d_rtb_softlimit))) &&
214 (!d->d_rtb_hardlimit ||
215 (be64_to_cpu(d->d_rtbcount) <=
216 be64_to_cpu(d->d_rtb_hardlimit)))) {
223 * initialize a buffer full of dquots and log the whole thing
226 xfs_qm_init_dquot_blk(
233 struct xfs_quotainfo *q = mp->m_quotainfo;
238 ASSERT(xfs_buf_islocked(bp));
243 * ID of the first dquot in the block - id's are zero based.
245 curid = id - (id % q->qi_dqperchunk);
247 memset(d, 0, BBTOB(q->qi_dqchunklen));
248 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
249 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
250 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
251 d->dd_diskdq.d_id = cpu_to_be32(curid);
252 d->dd_diskdq.d_flags = type;
253 if (xfs_sb_version_hascrc(&mp->m_sb)) {
254 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid);
255 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
260 xfs_trans_dquot_buf(tp, bp,
261 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
262 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
263 XFS_BLF_GDQUOT_BUF)));
264 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
268 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
269 * watermarks correspond to the soft and hard limits by default. If a soft limit
270 * is not specified, we use 95% of the hard limit.
273 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
277 dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit);
278 dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit);
279 if (!dqp->q_prealloc_lo_wmark) {
280 dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
281 do_div(dqp->q_prealloc_lo_wmark, 100);
282 dqp->q_prealloc_lo_wmark *= 95;
285 space = dqp->q_prealloc_hi_wmark;
288 dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
289 dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
290 dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
294 xfs_dquot_buf_verify_crc(
295 struct xfs_mount *mp,
298 struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr;
302 if (!xfs_sb_version_hascrc(&mp->m_sb))
306 * if we are in log recovery, the quota subsystem has not been
307 * initialised so we have no quotainfo structure. In that case, we need
308 * to manually calculate the number of dquots in the buffer.
311 ndquots = mp->m_quotainfo->qi_dqperchunk;
313 ndquots = xfs_qm_calc_dquots_per_chunk(mp,
314 XFS_BB_TO_FSB(mp, bp->b_length));
316 for (i = 0; i < ndquots; i++, d++) {
317 if (!xfs_verify_cksum((char *)d, sizeof(struct xfs_dqblk),
320 if (!uuid_equal(&d->dd_uuid, &mp->m_sb.sb_uuid))
327 xfs_dquot_buf_verify(
328 struct xfs_mount *mp,
331 struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr;
337 * if we are in log recovery, the quota subsystem has not been
338 * initialised so we have no quotainfo structure. In that case, we need
339 * to manually calculate the number of dquots in the buffer.
342 ndquots = mp->m_quotainfo->qi_dqperchunk;
344 ndquots = xfs_qm_calc_dquots_per_chunk(mp, bp->b_length);
347 * On the first read of the buffer, verify that each dquot is valid.
348 * We don't know what the id of the dquot is supposed to be, just that
349 * they should be increasing monotonically within the buffer. If the
350 * first id is corrupt, then it will fail on the second dquot in the
351 * buffer so corruptions could point to the wrong dquot in this case.
353 for (i = 0; i < ndquots; i++) {
354 struct xfs_disk_dquot *ddq;
357 ddq = &d[i].dd_diskdq;
360 id = be32_to_cpu(ddq->d_id);
362 error = xfs_qm_dqcheck(mp, ddq, id + i, 0, XFS_QMOPT_DOWARN,
363 "xfs_dquot_buf_verify");
371 xfs_dquot_buf_read_verify(
374 struct xfs_mount *mp = bp->b_target->bt_mount;
376 if (!xfs_dquot_buf_verify_crc(mp, bp) || !xfs_dquot_buf_verify(mp, bp)) {
377 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
378 xfs_buf_ioerror(bp, EFSCORRUPTED);
383 * we don't calculate the CRC here as that is done when the dquot is flushed to
384 * the buffer after the update is done. This ensures that the dquot in the
385 * buffer always has an up-to-date CRC value.
388 xfs_dquot_buf_write_verify(
391 struct xfs_mount *mp = bp->b_target->bt_mount;
393 if (!xfs_dquot_buf_verify(mp, bp)) {
394 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
395 xfs_buf_ioerror(bp, EFSCORRUPTED);
400 const struct xfs_buf_ops xfs_dquot_buf_ops = {
401 .verify_read = xfs_dquot_buf_read_verify,
402 .verify_write = xfs_dquot_buf_write_verify,
406 * Allocate a block and fill it with dquots.
407 * This is called when the bmapi finds a hole.
415 xfs_fileoff_t offset_fsb,
418 xfs_fsblock_t firstblock;
419 xfs_bmap_free_t flist;
421 int nmaps, error, committed;
423 xfs_trans_t *tp = *tpp;
427 trace_xfs_dqalloc(dqp);
430 * Initialize the bmap freelist prior to calling bmapi code.
432 xfs_bmap_init(&flist, &firstblock);
433 xfs_ilock(quotip, XFS_ILOCK_EXCL);
435 * Return if this type of quotas is turned off while we didn't
438 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
439 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
443 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
445 error = xfs_bmapi_write(tp, quotip, offset_fsb,
446 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
447 &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp),
448 &map, &nmaps, &flist);
451 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
453 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
454 (map.br_startblock != HOLESTARTBLOCK));
457 * Keep track of the blkno to save a lookup later
459 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
461 /* now we can just get the buffer (there's nothing to read yet) */
462 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
464 mp->m_quotainfo->qi_dqchunklen,
467 error = xfs_buf_geterror(bp);
470 bp->b_ops = &xfs_dquot_buf_ops;
473 * Make a chunk of dquots out of this buffer and log
476 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
477 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
480 * xfs_bmap_finish() may commit the current transaction and
481 * start a second transaction if the freelist is not empty.
483 * Since we still want to modify this buffer, we need to
484 * ensure that the buffer is not released on commit of
485 * the first transaction and ensure the buffer is added to the
486 * second transaction.
488 * If there is only one transaction then don't stop the buffer
489 * from being released when it commits later on.
492 xfs_trans_bhold(tp, bp);
494 if ((error = xfs_bmap_finish(tpp, &flist, &committed))) {
500 xfs_trans_bjoin(tp, bp);
502 xfs_trans_bhold_release(tp, bp);
509 xfs_bmap_cancel(&flist);
511 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
517 struct xfs_mount *mp,
518 struct xfs_trans *tp,
519 struct xfs_dquot *dqp,
521 struct xfs_buf **bpp)
524 struct xfs_disk_dquot *ddq;
529 * Read the buffer without verification so we get the corrupted
530 * buffer returned to us. make sure we verify it on write, though.
532 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno,
533 mp->m_quotainfo->qi_dqchunklen,
537 ASSERT(*bpp == NULL);
538 return XFS_ERROR(error);
540 (*bpp)->b_ops = &xfs_dquot_buf_ops;
542 ASSERT(xfs_buf_islocked(*bpp));
543 d = (struct xfs_dqblk *)(*bpp)->b_addr;
545 /* Do the actual repair of dquots in this buffer */
546 for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) {
547 ddq = &d[i].dd_diskdq;
548 error = xfs_qm_dqcheck(mp, ddq, firstid + i,
549 dqp->dq_flags & XFS_DQ_ALLTYPES,
550 XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair");
552 /* repair failed, we're screwed */
553 xfs_trans_brelse(tp, *bpp);
554 return XFS_ERROR(EIO);
562 * Maps a dquot to the buffer containing its on-disk version.
563 * This returns a ptr to the buffer containing the on-disk dquot
564 * in the bpp param, and a ptr to the on-disk dquot within that buffer
570 xfs_disk_dquot_t **O_ddpp,
574 struct xfs_bmbt_irec map;
575 int nmaps = 1, error;
577 struct xfs_inode *quotip = xfs_dq_to_quota_inode(dqp);
578 struct xfs_mount *mp = dqp->q_mount;
579 xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id);
580 struct xfs_trans *tp = (tpp ? *tpp : NULL);
582 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
584 xfs_ilock(quotip, XFS_ILOCK_SHARED);
585 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
587 * Return if this type of quotas is turned off while we
588 * didn't have the quota inode lock.
590 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
595 * Find the block map; no allocations yet
597 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
598 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
600 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
605 ASSERT(map.br_blockcount == 1);
608 * Offset of dquot in the (fixed sized) dquot chunk.
610 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
613 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
614 if (map.br_startblock == HOLESTARTBLOCK) {
616 * We don't allocate unless we're asked to
618 if (!(flags & XFS_QMOPT_DQALLOC))
622 error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
623 dqp->q_fileoffset, &bp);
628 trace_xfs_dqtobp_read(dqp);
631 * store the blkno etc so that we don't have to do the
632 * mapping all the time
634 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
636 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
638 mp->m_quotainfo->qi_dqchunklen,
639 0, &bp, &xfs_dquot_buf_ops);
641 if (error == EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) {
642 xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff *
643 mp->m_quotainfo->qi_dqperchunk;
645 error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp);
650 return XFS_ERROR(error);
654 ASSERT(xfs_buf_islocked(bp));
656 *O_ddpp = bp->b_addr + dqp->q_bufoffset;
663 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
664 * and release the buffer immediately.
666 * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed.
670 struct xfs_mount *mp,
674 struct xfs_dquot **O_dqpp)
676 struct xfs_dquot *dqp;
677 struct xfs_disk_dquot *ddqp;
679 struct xfs_trans *tp = NULL;
684 dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP);
686 dqp->dq_flags = type;
687 dqp->q_core.d_id = cpu_to_be32(id);
689 INIT_LIST_HEAD(&dqp->q_lru);
690 mutex_init(&dqp->q_qlock);
691 init_waitqueue_head(&dqp->q_pinwait);
694 * Because we want to use a counting completion, complete
695 * the flush completion once to allow a single access to
696 * the flush completion without blocking.
698 init_completion(&dqp->q_flush);
699 complete(&dqp->q_flush);
702 * Make sure group quotas have a different lock class than user
705 if (!(type & XFS_DQ_USER))
706 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class);
708 XFS_STATS_INC(xs_qm_dquot);
710 trace_xfs_dqread(dqp);
712 if (flags & XFS_QMOPT_DQALLOC) {
713 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
714 error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp),
715 XFS_QM_DQALLOC_LOG_RES(mp), 0,
716 XFS_TRANS_PERM_LOG_RES,
717 XFS_WRITE_LOG_COUNT);
720 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
724 * get a pointer to the on-disk dquot and the buffer containing it
725 * dqp already knows its own type (GROUP/USER).
727 error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags);
730 * This can happen if quotas got turned off (ESRCH),
731 * or if the dquot didn't exist on disk and we ask to
734 trace_xfs_dqread_fail(dqp);
735 cancelflags |= XFS_TRANS_ABORT;
739 /* copy everything from disk dquot to the incore dquot */
740 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
741 xfs_qm_dquot_logitem_init(dqp);
744 * Reservation counters are defined as reservation plus current usage
745 * to avoid having to add every time.
747 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
748 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
749 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
751 /* initialize the dquot speculative prealloc thresholds */
752 xfs_dquot_set_prealloc_limits(dqp);
754 /* Mark the buf so that this will stay incore a little longer */
755 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
758 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
759 * So we need to release with xfs_trans_brelse().
760 * The strategy here is identical to that of inodes; we lock
761 * the dquot in xfs_qm_dqget() before making it accessible to
762 * others. This is because dquots, like inodes, need a good level of
763 * concurrency, and we don't want to take locks on the entire buffers
764 * for dquot accesses.
765 * Note also that the dquot buffer may even be dirty at this point, if
766 * this particular dquot was repaired. We still aren't afraid to
767 * brelse it because we have the changes incore.
769 ASSERT(xfs_buf_islocked(bp));
770 xfs_trans_brelse(tp, bp);
773 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
783 xfs_trans_cancel(tp, cancelflags);
785 xfs_qm_dqdestroy(dqp);
791 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
792 * a locked dquot, doing an allocation (if requested) as needed.
793 * When both an inode and an id are given, the inode's id takes precedence.
794 * That is, if the id changes while we don't hold the ilock inside this
795 * function, the new dquot is returned, not necessarily the one requested
796 * in the id argument.
801 xfs_inode_t *ip, /* locked inode (optional) */
802 xfs_dqid_t id, /* uid/projid/gid depending on type */
803 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
804 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
805 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
807 struct xfs_quotainfo *qi = mp->m_quotainfo;
808 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
809 struct xfs_dquot *dqp;
812 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
813 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
814 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
815 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
820 if (xfs_do_dqerror) {
821 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
822 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
823 xfs_debug(mp, "Returning error in dqget");
828 ASSERT(type == XFS_DQ_USER ||
829 type == XFS_DQ_PROJ ||
830 type == XFS_DQ_GROUP);
832 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
833 ASSERT(xfs_inode_dquot(ip, type) == NULL);
838 mutex_lock(&qi->qi_tree_lock);
839 dqp = radix_tree_lookup(tree, id);
842 if (dqp->dq_flags & XFS_DQ_FREEING) {
844 mutex_unlock(&qi->qi_tree_lock);
845 trace_xfs_dqget_freeing(dqp);
851 mutex_unlock(&qi->qi_tree_lock);
853 trace_xfs_dqget_hit(dqp);
854 XFS_STATS_INC(xs_qm_dqcachehits);
858 mutex_unlock(&qi->qi_tree_lock);
859 XFS_STATS_INC(xs_qm_dqcachemisses);
862 * Dquot cache miss. We don't want to keep the inode lock across
863 * a (potential) disk read. Also we don't want to deal with the lock
864 * ordering between quotainode and this inode. OTOH, dropping the inode
865 * lock here means dealing with a chown that can happen before
866 * we re-acquire the lock.
869 xfs_iunlock(ip, XFS_ILOCK_EXCL);
871 error = xfs_qm_dqread(mp, id, type, flags, &dqp);
874 xfs_ilock(ip, XFS_ILOCK_EXCL);
881 * A dquot could be attached to this inode by now, since
882 * we had dropped the ilock.
884 if (xfs_this_quota_on(mp, type)) {
885 struct xfs_dquot *dqp1;
887 dqp1 = xfs_inode_dquot(ip, type);
889 xfs_qm_dqdestroy(dqp);
895 /* inode stays locked on return */
896 xfs_qm_dqdestroy(dqp);
897 return XFS_ERROR(ESRCH);
901 mutex_lock(&qi->qi_tree_lock);
902 error = -radix_tree_insert(tree, id, dqp);
903 if (unlikely(error)) {
904 WARN_ON(error != EEXIST);
907 * Duplicate found. Just throw away the new dquot and start
910 mutex_unlock(&qi->qi_tree_lock);
911 trace_xfs_dqget_dup(dqp);
912 xfs_qm_dqdestroy(dqp);
913 XFS_STATS_INC(xs_qm_dquot_dups);
918 * We return a locked dquot to the caller, with a reference taken
924 mutex_unlock(&qi->qi_tree_lock);
927 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
928 trace_xfs_dqget_miss(dqp);
936 struct xfs_dquot *dqp)
938 struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
939 struct xfs_dquot *gdqp;
940 struct xfs_dquot *pdqp;
942 trace_xfs_dqput_free(dqp);
944 mutex_lock(&qi->qi_lru_lock);
945 if (list_empty(&dqp->q_lru)) {
946 list_add_tail(&dqp->q_lru, &qi->qi_lru_list);
948 XFS_STATS_INC(xs_qm_dquot_unused);
950 mutex_unlock(&qi->qi_lru_lock);
953 * If we just added a udquot to the freelist, then we want to release
954 * the gdquot/pdquot reference that it (probably) has. Otherwise it'll
955 * keep the gdquot/pdquot from getting reclaimed.
957 gdqp = dqp->q_gdquot;
960 dqp->q_gdquot = NULL;
963 pdqp = dqp->q_pdquot;
966 dqp->q_pdquot = NULL;
971 * If we had a group/project quota hint, release it now.
980 * Release a reference to the dquot (decrement ref-count) and unlock it.
982 * If there is a group quota attached to this dquot, carefully release that
983 * too without tripping over deadlocks'n'stuff.
987 struct xfs_dquot *dqp)
989 ASSERT(dqp->q_nrefs > 0);
990 ASSERT(XFS_DQ_IS_LOCKED(dqp));
992 trace_xfs_dqput(dqp);
994 if (--dqp->q_nrefs > 0)
997 xfs_qm_dqput_final(dqp);
1001 * Release a dquot. Flush it if dirty, then dqput() it.
1002 * dquot must not be locked.
1011 trace_xfs_dqrele(dqp);
1015 * We don't care to flush it if the dquot is dirty here.
1016 * That will create stutters that we want to avoid.
1017 * Instead we do a delayed write when we try to reclaim
1018 * a dirty dquot. Also xfs_sync will take part of the burden...
1024 * This is the dquot flushing I/O completion routine. It is called
1025 * from interrupt level when the buffer containing the dquot is
1026 * flushed to disk. It is responsible for removing the dquot logitem
1027 * from the AIL if it has not been re-logged, and unlocking the dquot's
1028 * flush lock. This behavior is very similar to that of inodes..
1031 xfs_qm_dqflush_done(
1033 struct xfs_log_item *lip)
1035 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
1036 xfs_dquot_t *dqp = qip->qli_dquot;
1037 struct xfs_ail *ailp = lip->li_ailp;
1040 * We only want to pull the item from the AIL if its
1041 * location in the log has not changed since we started the flush.
1042 * Thus, we only bother if the dquot's lsn has
1043 * not changed. First we check the lsn outside the lock
1044 * since it's cheaper, and then we recheck while
1045 * holding the lock before removing the dquot from the AIL.
1047 if ((lip->li_flags & XFS_LI_IN_AIL) &&
1048 lip->li_lsn == qip->qli_flush_lsn) {
1050 /* xfs_trans_ail_delete() drops the AIL lock. */
1051 spin_lock(&ailp->xa_lock);
1052 if (lip->li_lsn == qip->qli_flush_lsn)
1053 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
1055 spin_unlock(&ailp->xa_lock);
1059 * Release the dq's flush lock since we're done with it.
1065 * Write a modified dquot to disk.
1066 * The dquot must be locked and the flush lock too taken by caller.
1067 * The flush lock will not be unlocked until the dquot reaches the disk,
1068 * but the dquot is free to be unlocked and modified by the caller
1069 * in the interim. Dquot is still locked on return. This behavior is
1070 * identical to that of inodes.
1074 struct xfs_dquot *dqp,
1075 struct xfs_buf **bpp)
1077 struct xfs_mount *mp = dqp->q_mount;
1079 struct xfs_disk_dquot *ddqp;
1082 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1083 ASSERT(!completion_done(&dqp->q_flush));
1085 trace_xfs_dqflush(dqp);
1089 xfs_qm_dqunpin_wait(dqp);
1092 * This may have been unpinned because the filesystem is shutting
1093 * down forcibly. If that's the case we must not write this dquot
1094 * to disk, because the log record didn't make it to disk.
1096 * We also have to remove the log item from the AIL in this case,
1097 * as we wait for an emptry AIL as part of the unmount process.
1099 if (XFS_FORCED_SHUTDOWN(mp)) {
1100 struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
1101 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1103 spin_lock(&mp->m_ail->xa_lock);
1104 if (lip->li_flags & XFS_LI_IN_AIL)
1105 xfs_trans_ail_delete(mp->m_ail, lip,
1106 SHUTDOWN_CORRUPT_INCORE);
1108 spin_unlock(&mp->m_ail->xa_lock);
1109 error = XFS_ERROR(EIO);
1114 * Get the buffer containing the on-disk dquot
1116 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1117 mp->m_quotainfo->qi_dqchunklen, 0, &bp, NULL);
1122 * Calculate the location of the dquot inside the buffer.
1124 ddqp = bp->b_addr + dqp->q_bufoffset;
1127 * A simple sanity check in case we got a corrupted dquot..
1129 error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
1130 XFS_QMOPT_DOWARN, "dqflush (incore copy)");
1134 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1135 return XFS_ERROR(EIO);
1138 /* This is the only portion of data that needs to persist */
1139 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
1142 * Clear the dirty field and remember the flush lsn for later use.
1144 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1146 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1147 &dqp->q_logitem.qli_item.li_lsn);
1150 * copy the lsn into the on-disk dquot now while we have the in memory
1151 * dquot here. This can't be done later in the write verifier as we
1152 * can't get access to the log item at that point in time.
1154 * We also calculate the CRC here so that the on-disk dquot in the
1155 * buffer always has a valid CRC. This ensures there is no possibility
1156 * of a dquot without an up-to-date CRC getting to disk.
1158 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1159 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp;
1161 dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1162 xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk),
1167 * Attach an iodone routine so that we can remove this dquot from the
1168 * AIL and release the flush lock once the dquot is synced to disk.
1170 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1171 &dqp->q_logitem.qli_item);
1174 * If the buffer is pinned then push on the log so we won't
1175 * get stuck waiting in the write for too long.
1177 if (xfs_buf_ispinned(bp)) {
1178 trace_xfs_dqflush_force(dqp);
1179 xfs_log_force(mp, 0);
1182 trace_xfs_dqflush_done(dqp);
1188 return XFS_ERROR(EIO);
1192 * Lock two xfs_dquot structures.
1194 * To avoid deadlocks we always lock the quota structure with
1195 * the lowerd id first.
1204 if (be32_to_cpu(d1->q_core.d_id) >
1205 be32_to_cpu(d2->q_core.d_id)) {
1206 mutex_lock(&d2->q_qlock);
1207 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1209 mutex_lock(&d1->q_qlock);
1210 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1213 mutex_lock(&d1->q_qlock);
1215 mutex_lock(&d2->q_qlock);
1223 kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
1228 kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
1229 if (!xfs_qm_dqtrxzone)
1230 goto out_free_dqzone;
1235 kmem_zone_destroy(xfs_qm_dqzone);
1243 kmem_zone_destroy(xfs_qm_dqtrxzone);
1244 kmem_zone_destroy(xfs_qm_dqzone);