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
22 #include "xfs_trans.h"
25 #include "xfs_alloc.h"
26 #include "xfs_quota.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_inode.h"
31 #include "xfs_rtalloc.h"
32 #include "xfs_error.h"
33 #include "xfs_itable.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"
47 * dquot->q_qlock (xfs_dqlock() and friends)
48 * dquot->q_flush (xfs_dqflock() and friends)
51 * If two dquots need to be locked the order is user before group/project,
52 * otherwise by the lowest id first, see xfs_dqlock2.
56 xfs_buftarg_t *xfs_dqerror_target;
59 int xfs_dqerror_mod = 33;
62 struct kmem_zone *xfs_qm_dqtrxzone;
63 static struct kmem_zone *xfs_qm_dqzone;
65 static struct lock_class_key xfs_dquot_other_class;
68 * This is called to free all the memory associated with a dquot
74 ASSERT(list_empty(&dqp->q_lru));
76 mutex_destroy(&dqp->q_qlock);
77 kmem_zone_free(xfs_qm_dqzone, dqp);
79 XFS_STATS_DEC(xs_qm_dquot);
83 * If default limits are in force, push them into the dquot now.
84 * We overwrite the dquot limits only if they are zero and this
85 * is not the root dquot.
88 xfs_qm_adjust_dqlimits(
92 struct xfs_quotainfo *q = mp->m_quotainfo;
93 struct xfs_disk_dquot *d = &dq->q_core;
98 if (q->qi_bsoftlimit && !d->d_blk_softlimit) {
99 d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit);
102 if (q->qi_bhardlimit && !d->d_blk_hardlimit) {
103 d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit);
106 if (q->qi_isoftlimit && !d->d_ino_softlimit)
107 d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit);
108 if (q->qi_ihardlimit && !d->d_ino_hardlimit)
109 d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit);
110 if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit)
111 d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit);
112 if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit)
113 d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit);
116 xfs_dquot_set_prealloc_limits(dq);
120 * Check the limits and timers of a dquot and start or reset timers
122 * This gets called even when quota enforcement is OFF, which makes our
123 * life a little less complicated. (We just don't reject any quota
124 * reservations in that case, when enforcement is off).
125 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
127 * In contrast, warnings are a little different in that they don't
128 * 'automatically' get started when limits get exceeded. They do
129 * get reset to zero, however, when we find the count to be under
130 * the soft limit (they are only ever set non-zero via userspace).
133 xfs_qm_adjust_dqtimers(
140 if (d->d_blk_hardlimit)
141 ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
142 be64_to_cpu(d->d_blk_hardlimit));
143 if (d->d_ino_hardlimit)
144 ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
145 be64_to_cpu(d->d_ino_hardlimit));
146 if (d->d_rtb_hardlimit)
147 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
148 be64_to_cpu(d->d_rtb_hardlimit));
152 if ((d->d_blk_softlimit &&
153 (be64_to_cpu(d->d_bcount) >
154 be64_to_cpu(d->d_blk_softlimit))) ||
155 (d->d_blk_hardlimit &&
156 (be64_to_cpu(d->d_bcount) >
157 be64_to_cpu(d->d_blk_hardlimit)))) {
158 d->d_btimer = cpu_to_be32(get_seconds() +
159 mp->m_quotainfo->qi_btimelimit);
164 if ((!d->d_blk_softlimit ||
165 (be64_to_cpu(d->d_bcount) <=
166 be64_to_cpu(d->d_blk_softlimit))) &&
167 (!d->d_blk_hardlimit ||
168 (be64_to_cpu(d->d_bcount) <=
169 be64_to_cpu(d->d_blk_hardlimit)))) {
175 if ((d->d_ino_softlimit &&
176 (be64_to_cpu(d->d_icount) >
177 be64_to_cpu(d->d_ino_softlimit))) ||
178 (d->d_ino_hardlimit &&
179 (be64_to_cpu(d->d_icount) >
180 be64_to_cpu(d->d_ino_hardlimit)))) {
181 d->d_itimer = cpu_to_be32(get_seconds() +
182 mp->m_quotainfo->qi_itimelimit);
187 if ((!d->d_ino_softlimit ||
188 (be64_to_cpu(d->d_icount) <=
189 be64_to_cpu(d->d_ino_softlimit))) &&
190 (!d->d_ino_hardlimit ||
191 (be64_to_cpu(d->d_icount) <=
192 be64_to_cpu(d->d_ino_hardlimit)))) {
197 if (!d->d_rtbtimer) {
198 if ((d->d_rtb_softlimit &&
199 (be64_to_cpu(d->d_rtbcount) >
200 be64_to_cpu(d->d_rtb_softlimit))) ||
201 (d->d_rtb_hardlimit &&
202 (be64_to_cpu(d->d_rtbcount) >
203 be64_to_cpu(d->d_rtb_hardlimit)))) {
204 d->d_rtbtimer = cpu_to_be32(get_seconds() +
205 mp->m_quotainfo->qi_rtbtimelimit);
210 if ((!d->d_rtb_softlimit ||
211 (be64_to_cpu(d->d_rtbcount) <=
212 be64_to_cpu(d->d_rtb_softlimit))) &&
213 (!d->d_rtb_hardlimit ||
214 (be64_to_cpu(d->d_rtbcount) <=
215 be64_to_cpu(d->d_rtb_hardlimit)))) {
222 * initialize a buffer full of dquots and log the whole thing
225 xfs_qm_init_dquot_blk(
232 struct xfs_quotainfo *q = mp->m_quotainfo;
237 ASSERT(xfs_buf_islocked(bp));
242 * ID of the first dquot in the block - id's are zero based.
244 curid = id - (id % q->qi_dqperchunk);
246 memset(d, 0, BBTOB(q->qi_dqchunklen));
247 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
248 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
249 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
250 d->dd_diskdq.d_id = cpu_to_be32(curid);
251 d->dd_diskdq.d_flags = type;
252 if (xfs_sb_version_hascrc(&mp->m_sb))
253 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid);
256 xfs_trans_dquot_buf(tp, bp,
257 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
258 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
259 XFS_BLF_GDQUOT_BUF)));
260 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
264 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
265 * watermarks correspond to the soft and hard limits by default. If a soft limit
266 * is not specified, we use 95% of the hard limit.
269 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
273 dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit);
274 dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit);
275 if (!dqp->q_prealloc_lo_wmark) {
276 dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
277 do_div(dqp->q_prealloc_lo_wmark, 100);
278 dqp->q_prealloc_lo_wmark *= 95;
281 space = dqp->q_prealloc_hi_wmark;
284 dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
285 dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
286 dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
290 xfs_dquot_buf_calc_crc(
291 struct xfs_mount *mp,
294 struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr;
297 if (!xfs_sb_version_hascrc(&mp->m_sb))
300 for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++, d++) {
301 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
302 offsetof(struct xfs_dqblk, dd_crc));
307 xfs_dquot_buf_verify_crc(
308 struct xfs_mount *mp,
311 struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr;
315 if (!xfs_sb_version_hascrc(&mp->m_sb))
319 * if we are in log recovery, the quota subsystem has not been
320 * initialised so we have no quotainfo structure. In that case, we need
321 * to manually calculate the number of dquots in the buffer.
324 ndquots = mp->m_quotainfo->qi_dqperchunk;
326 ndquots = xfs_qm_calc_dquots_per_chunk(mp,
327 XFS_BB_TO_FSB(mp, bp->b_length));
329 for (i = 0; i < ndquots; i++, d++) {
330 if (!xfs_verify_cksum((char *)d, sizeof(struct xfs_dqblk),
331 offsetof(struct xfs_dqblk, dd_crc)))
333 if (!uuid_equal(&d->dd_uuid, &mp->m_sb.sb_uuid))
341 xfs_dquot_buf_verify(
342 struct xfs_mount *mp,
345 struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr;
351 * if we are in log recovery, the quota subsystem has not been
352 * initialised so we have no quotainfo structure. In that case, we need
353 * to manually calculate the number of dquots in the buffer.
356 ndquots = mp->m_quotainfo->qi_dqperchunk;
358 ndquots = xfs_qm_calc_dquots_per_chunk(mp, bp->b_length);
361 * On the first read of the buffer, verify that each dquot is valid.
362 * We don't know what the id of the dquot is supposed to be, just that
363 * they should be increasing monotonically within the buffer. If the
364 * first id is corrupt, then it will fail on the second dquot in the
365 * buffer so corruptions could point to the wrong dquot in this case.
367 for (i = 0; i < ndquots; i++) {
368 struct xfs_disk_dquot *ddq;
371 ddq = &d[i].dd_diskdq;
374 id = be32_to_cpu(ddq->d_id);
376 error = xfs_qm_dqcheck(mp, ddq, id + i, 0, XFS_QMOPT_DOWARN,
377 "xfs_dquot_buf_verify");
385 xfs_dquot_buf_read_verify(
388 struct xfs_mount *mp = bp->b_target->bt_mount;
390 if (!xfs_dquot_buf_verify_crc(mp, bp) || !xfs_dquot_buf_verify(mp, bp)) {
391 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
392 xfs_buf_ioerror(bp, EFSCORRUPTED);
397 xfs_dquot_buf_write_verify(
400 struct xfs_mount *mp = bp->b_target->bt_mount;
402 if (!xfs_dquot_buf_verify(mp, bp)) {
403 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
404 xfs_buf_ioerror(bp, EFSCORRUPTED);
407 xfs_dquot_buf_calc_crc(mp, bp);
410 const struct xfs_buf_ops xfs_dquot_buf_ops = {
411 .verify_read = xfs_dquot_buf_read_verify,
412 .verify_write = xfs_dquot_buf_write_verify,
416 * Allocate a block and fill it with dquots.
417 * This is called when the bmapi finds a hole.
425 xfs_fileoff_t offset_fsb,
428 xfs_fsblock_t firstblock;
429 xfs_bmap_free_t flist;
431 int nmaps, error, committed;
433 xfs_trans_t *tp = *tpp;
437 trace_xfs_dqalloc(dqp);
440 * Initialize the bmap freelist prior to calling bmapi code.
442 xfs_bmap_init(&flist, &firstblock);
443 xfs_ilock(quotip, XFS_ILOCK_EXCL);
445 * Return if this type of quotas is turned off while we didn't
448 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
449 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
453 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
455 error = xfs_bmapi_write(tp, quotip, offset_fsb,
456 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
457 &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp),
458 &map, &nmaps, &flist);
461 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
463 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
464 (map.br_startblock != HOLESTARTBLOCK));
467 * Keep track of the blkno to save a lookup later
469 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
471 /* now we can just get the buffer (there's nothing to read yet) */
472 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
474 mp->m_quotainfo->qi_dqchunklen,
477 error = xfs_buf_geterror(bp);
480 bp->b_ops = &xfs_dquot_buf_ops;
483 * Make a chunk of dquots out of this buffer and log
486 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
487 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
490 * xfs_bmap_finish() may commit the current transaction and
491 * start a second transaction if the freelist is not empty.
493 * Since we still want to modify this buffer, we need to
494 * ensure that the buffer is not released on commit of
495 * the first transaction and ensure the buffer is added to the
496 * second transaction.
498 * If there is only one transaction then don't stop the buffer
499 * from being released when it commits later on.
502 xfs_trans_bhold(tp, bp);
504 if ((error = xfs_bmap_finish(tpp, &flist, &committed))) {
510 xfs_trans_bjoin(tp, bp);
512 xfs_trans_bhold_release(tp, bp);
519 xfs_bmap_cancel(&flist);
521 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
527 struct xfs_mount *mp,
528 struct xfs_trans *tp,
529 struct xfs_dquot *dqp,
531 struct xfs_buf **bpp)
534 struct xfs_disk_dquot *ddq;
539 * Read the buffer without verification so we get the corrupted
540 * buffer returned to us. make sure we verify it on write, though.
542 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno,
543 mp->m_quotainfo->qi_dqchunklen,
547 ASSERT(*bpp == NULL);
548 return XFS_ERROR(error);
550 (*bpp)->b_ops = &xfs_dquot_buf_ops;
552 ASSERT(xfs_buf_islocked(*bpp));
553 d = (struct xfs_dqblk *)(*bpp)->b_addr;
555 /* Do the actual repair of dquots in this buffer */
556 for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) {
557 ddq = &d[i].dd_diskdq;
558 error = xfs_qm_dqcheck(mp, ddq, firstid + i,
559 dqp->dq_flags & XFS_DQ_ALLTYPES,
560 XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair");
562 /* repair failed, we're screwed */
563 xfs_trans_brelse(tp, *bpp);
564 return XFS_ERROR(EIO);
572 * Maps a dquot to the buffer containing its on-disk version.
573 * This returns a ptr to the buffer containing the on-disk dquot
574 * in the bpp param, and a ptr to the on-disk dquot within that buffer
580 xfs_disk_dquot_t **O_ddpp,
585 int nmaps = 1, error;
587 xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp);
588 xfs_mount_t *mp = dqp->q_mount;
589 xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id);
590 xfs_trans_t *tp = (tpp ? *tpp : NULL);
592 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
594 xfs_ilock(quotip, XFS_ILOCK_SHARED);
595 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
597 * Return if this type of quotas is turned off while we
598 * didn't have the quota inode lock.
600 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
605 * Find the block map; no allocations yet
607 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
608 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
610 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
615 ASSERT(map.br_blockcount == 1);
618 * Offset of dquot in the (fixed sized) dquot chunk.
620 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
623 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
624 if (map.br_startblock == HOLESTARTBLOCK) {
626 * We don't allocate unless we're asked to
628 if (!(flags & XFS_QMOPT_DQALLOC))
632 error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
633 dqp->q_fileoffset, &bp);
638 trace_xfs_dqtobp_read(dqp);
641 * store the blkno etc so that we don't have to do the
642 * mapping all the time
644 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
646 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
648 mp->m_quotainfo->qi_dqchunklen,
649 0, &bp, &xfs_dquot_buf_ops);
651 if (error == EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) {
652 xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff *
653 mp->m_quotainfo->qi_dqperchunk;
655 error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp);
660 return XFS_ERROR(error);
664 ASSERT(xfs_buf_islocked(bp));
666 *O_ddpp = bp->b_addr + dqp->q_bufoffset;
673 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
674 * and release the buffer immediately.
676 * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed.
680 struct xfs_mount *mp,
684 struct xfs_dquot **O_dqpp)
686 struct xfs_dquot *dqp;
687 struct xfs_disk_dquot *ddqp;
689 struct xfs_trans *tp = NULL;
694 dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP);
696 dqp->dq_flags = type;
697 dqp->q_core.d_id = cpu_to_be32(id);
699 INIT_LIST_HEAD(&dqp->q_lru);
700 mutex_init(&dqp->q_qlock);
701 init_waitqueue_head(&dqp->q_pinwait);
704 * Because we want to use a counting completion, complete
705 * the flush completion once to allow a single access to
706 * the flush completion without blocking.
708 init_completion(&dqp->q_flush);
709 complete(&dqp->q_flush);
712 * Make sure group quotas have a different lock class than user
715 if (!(type & XFS_DQ_USER))
716 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class);
718 XFS_STATS_INC(xs_qm_dquot);
720 trace_xfs_dqread(dqp);
722 if (flags & XFS_QMOPT_DQALLOC) {
723 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
724 error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp),
725 XFS_QM_DQALLOC_LOG_RES(mp), 0,
726 XFS_TRANS_PERM_LOG_RES,
727 XFS_WRITE_LOG_COUNT);
730 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
734 * get a pointer to the on-disk dquot and the buffer containing it
735 * dqp already knows its own type (GROUP/USER).
737 error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags);
740 * This can happen if quotas got turned off (ESRCH),
741 * or if the dquot didn't exist on disk and we ask to
744 trace_xfs_dqread_fail(dqp);
745 cancelflags |= XFS_TRANS_ABORT;
749 /* copy everything from disk dquot to the incore dquot */
750 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
751 xfs_qm_dquot_logitem_init(dqp);
754 * Reservation counters are defined as reservation plus current usage
755 * to avoid having to add every time.
757 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
758 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
759 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
761 /* initialize the dquot speculative prealloc thresholds */
762 xfs_dquot_set_prealloc_limits(dqp);
764 /* Mark the buf so that this will stay incore a little longer */
765 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
768 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
769 * So we need to release with xfs_trans_brelse().
770 * The strategy here is identical to that of inodes; we lock
771 * the dquot in xfs_qm_dqget() before making it accessible to
772 * others. This is because dquots, like inodes, need a good level of
773 * concurrency, and we don't want to take locks on the entire buffers
774 * for dquot accesses.
775 * Note also that the dquot buffer may even be dirty at this point, if
776 * this particular dquot was repaired. We still aren't afraid to
777 * brelse it because we have the changes incore.
779 ASSERT(xfs_buf_islocked(bp));
780 xfs_trans_brelse(tp, bp);
783 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
793 xfs_trans_cancel(tp, cancelflags);
795 xfs_qm_dqdestroy(dqp);
801 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
802 * a locked dquot, doing an allocation (if requested) as needed.
803 * When both an inode and an id are given, the inode's id takes precedence.
804 * That is, if the id changes while we don't hold the ilock inside this
805 * function, the new dquot is returned, not necessarily the one requested
806 * in the id argument.
811 xfs_inode_t *ip, /* locked inode (optional) */
812 xfs_dqid_t id, /* uid/projid/gid depending on type */
813 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
814 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
815 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
817 struct xfs_quotainfo *qi = mp->m_quotainfo;
818 struct radix_tree_root *tree = XFS_DQUOT_TREE(qi, type);
819 struct xfs_dquot *dqp;
822 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
823 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
824 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
825 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
830 if (xfs_do_dqerror) {
831 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
832 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
833 xfs_debug(mp, "Returning error in dqget");
838 ASSERT(type == XFS_DQ_USER ||
839 type == XFS_DQ_PROJ ||
840 type == XFS_DQ_GROUP);
842 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
843 ASSERT(xfs_inode_dquot(ip, type) == NULL);
848 mutex_lock(&qi->qi_tree_lock);
849 dqp = radix_tree_lookup(tree, id);
852 if (dqp->dq_flags & XFS_DQ_FREEING) {
854 mutex_unlock(&qi->qi_tree_lock);
855 trace_xfs_dqget_freeing(dqp);
861 mutex_unlock(&qi->qi_tree_lock);
863 trace_xfs_dqget_hit(dqp);
864 XFS_STATS_INC(xs_qm_dqcachehits);
868 mutex_unlock(&qi->qi_tree_lock);
869 XFS_STATS_INC(xs_qm_dqcachemisses);
872 * Dquot cache miss. We don't want to keep the inode lock across
873 * a (potential) disk read. Also we don't want to deal with the lock
874 * ordering between quotainode and this inode. OTOH, dropping the inode
875 * lock here means dealing with a chown that can happen before
876 * we re-acquire the lock.
879 xfs_iunlock(ip, XFS_ILOCK_EXCL);
881 error = xfs_qm_dqread(mp, id, type, flags, &dqp);
884 xfs_ilock(ip, XFS_ILOCK_EXCL);
891 * A dquot could be attached to this inode by now, since
892 * we had dropped the ilock.
894 if (xfs_this_quota_on(mp, type)) {
895 struct xfs_dquot *dqp1;
897 dqp1 = xfs_inode_dquot(ip, type);
899 xfs_qm_dqdestroy(dqp);
905 /* inode stays locked on return */
906 xfs_qm_dqdestroy(dqp);
907 return XFS_ERROR(ESRCH);
911 mutex_lock(&qi->qi_tree_lock);
912 error = -radix_tree_insert(tree, id, dqp);
913 if (unlikely(error)) {
914 WARN_ON(error != EEXIST);
917 * Duplicate found. Just throw away the new dquot and start
920 mutex_unlock(&qi->qi_tree_lock);
921 trace_xfs_dqget_dup(dqp);
922 xfs_qm_dqdestroy(dqp);
923 XFS_STATS_INC(xs_qm_dquot_dups);
928 * We return a locked dquot to the caller, with a reference taken
934 mutex_unlock(&qi->qi_tree_lock);
937 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
938 trace_xfs_dqget_miss(dqp);
946 struct xfs_dquot *dqp)
948 struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
949 struct xfs_dquot *gdqp;
951 trace_xfs_dqput_free(dqp);
953 mutex_lock(&qi->qi_lru_lock);
954 if (list_empty(&dqp->q_lru)) {
955 list_add_tail(&dqp->q_lru, &qi->qi_lru_list);
957 XFS_STATS_INC(xs_qm_dquot_unused);
959 mutex_unlock(&qi->qi_lru_lock);
962 * If we just added a udquot to the freelist, then we want to release
963 * the gdquot reference that it (probably) has. Otherwise it'll keep
964 * the gdquot from getting reclaimed.
966 gdqp = dqp->q_gdquot;
969 dqp->q_gdquot = NULL;
974 * If we had a group quota hint, release it now.
981 * Release a reference to the dquot (decrement ref-count) and unlock it.
983 * If there is a group quota attached to this dquot, carefully release that
984 * too without tripping over deadlocks'n'stuff.
988 struct xfs_dquot *dqp)
990 ASSERT(dqp->q_nrefs > 0);
991 ASSERT(XFS_DQ_IS_LOCKED(dqp));
993 trace_xfs_dqput(dqp);
995 if (--dqp->q_nrefs > 0)
998 xfs_qm_dqput_final(dqp);
1002 * Release a dquot. Flush it if dirty, then dqput() it.
1003 * dquot must not be locked.
1012 trace_xfs_dqrele(dqp);
1016 * We don't care to flush it if the dquot is dirty here.
1017 * That will create stutters that we want to avoid.
1018 * Instead we do a delayed write when we try to reclaim
1019 * a dirty dquot. Also xfs_sync will take part of the burden...
1025 * This is the dquot flushing I/O completion routine. It is called
1026 * from interrupt level when the buffer containing the dquot is
1027 * flushed to disk. It is responsible for removing the dquot logitem
1028 * from the AIL if it has not been re-logged, and unlocking the dquot's
1029 * flush lock. This behavior is very similar to that of inodes..
1032 xfs_qm_dqflush_done(
1034 struct xfs_log_item *lip)
1036 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
1037 xfs_dquot_t *dqp = qip->qli_dquot;
1038 struct xfs_ail *ailp = lip->li_ailp;
1041 * We only want to pull the item from the AIL if its
1042 * location in the log has not changed since we started the flush.
1043 * Thus, we only bother if the dquot's lsn has
1044 * not changed. First we check the lsn outside the lock
1045 * since it's cheaper, and then we recheck while
1046 * holding the lock before removing the dquot from the AIL.
1048 if ((lip->li_flags & XFS_LI_IN_AIL) &&
1049 lip->li_lsn == qip->qli_flush_lsn) {
1051 /* xfs_trans_ail_delete() drops the AIL lock. */
1052 spin_lock(&ailp->xa_lock);
1053 if (lip->li_lsn == qip->qli_flush_lsn)
1054 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
1056 spin_unlock(&ailp->xa_lock);
1060 * Release the dq's flush lock since we're done with it.
1066 * Write a modified dquot to disk.
1067 * The dquot must be locked and the flush lock too taken by caller.
1068 * The flush lock will not be unlocked until the dquot reaches the disk,
1069 * but the dquot is free to be unlocked and modified by the caller
1070 * in the interim. Dquot is still locked on return. This behavior is
1071 * identical to that of inodes.
1075 struct xfs_dquot *dqp,
1076 struct xfs_buf **bpp)
1078 struct xfs_mount *mp = dqp->q_mount;
1080 struct xfs_disk_dquot *ddqp;
1083 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1084 ASSERT(!completion_done(&dqp->q_flush));
1086 trace_xfs_dqflush(dqp);
1090 xfs_qm_dqunpin_wait(dqp);
1093 * This may have been unpinned because the filesystem is shutting
1094 * down forcibly. If that's the case we must not write this dquot
1095 * to disk, because the log record didn't make it to disk.
1097 * We also have to remove the log item from the AIL in this case,
1098 * as we wait for an emptry AIL as part of the unmount process.
1100 if (XFS_FORCED_SHUTDOWN(mp)) {
1101 struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
1102 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1104 spin_lock(&mp->m_ail->xa_lock);
1105 if (lip->li_flags & XFS_LI_IN_AIL)
1106 xfs_trans_ail_delete(mp->m_ail, lip,
1107 SHUTDOWN_CORRUPT_INCORE);
1109 spin_unlock(&mp->m_ail->xa_lock);
1110 error = XFS_ERROR(EIO);
1115 * Get the buffer containing the on-disk dquot
1117 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1118 mp->m_quotainfo->qi_dqchunklen, 0, &bp, NULL);
1123 * Calculate the location of the dquot inside the buffer.
1125 ddqp = bp->b_addr + dqp->q_bufoffset;
1128 * A simple sanity check in case we got a corrupted dquot..
1130 error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
1131 XFS_QMOPT_DOWARN, "dqflush (incore copy)");
1135 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1136 return XFS_ERROR(EIO);
1139 /* This is the only portion of data that needs to persist */
1140 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
1143 * Clear the dirty field and remember the flush lsn for later use.
1145 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1147 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1148 &dqp->q_logitem.qli_item.li_lsn);
1151 * copy the lsn into the on-disk dquot now while we have the in memory
1152 * dquot here. This can't be done later in the write verifier as we
1153 * can't get access to the log item at that point in time.
1155 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1156 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp;
1158 dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1162 * Attach an iodone routine so that we can remove this dquot from the
1163 * AIL and release the flush lock once the dquot is synced to disk.
1165 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1166 &dqp->q_logitem.qli_item);
1169 * If the buffer is pinned then push on the log so we won't
1170 * get stuck waiting in the write for too long.
1172 if (xfs_buf_ispinned(bp)) {
1173 trace_xfs_dqflush_force(dqp);
1174 xfs_log_force(mp, 0);
1177 trace_xfs_dqflush_done(dqp);
1183 return XFS_ERROR(EIO);
1187 * Lock two xfs_dquot structures.
1189 * To avoid deadlocks we always lock the quota structure with
1190 * the lowerd id first.
1199 if (be32_to_cpu(d1->q_core.d_id) >
1200 be32_to_cpu(d2->q_core.d_id)) {
1201 mutex_lock(&d2->q_qlock);
1202 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1204 mutex_lock(&d1->q_qlock);
1205 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1208 mutex_lock(&d1->q_qlock);
1210 mutex_lock(&d2->q_qlock);
1218 kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
1223 kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
1224 if (!xfs_qm_dqtrxzone)
1225 goto out_free_dqzone;
1230 kmem_zone_destroy(xfs_qm_dqzone);
1238 kmem_zone_destroy(xfs_qm_dqtrxzone);
1239 kmem_zone_destroy(xfs_qm_dqzone);