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_bmap_util.h"
33 #include "xfs_rtalloc.h"
34 #include "xfs_error.h"
35 #include "xfs_itable.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_trans_space.h"
39 #include "xfs_trans_priv.h"
41 #include "xfs_cksum.h"
42 #include "xfs_trace.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_other_class;
70 * This is called to free all the memory associated with a dquot
76 ASSERT(list_empty(&dqp->q_lru));
78 mutex_destroy(&dqp->q_qlock);
79 kmem_zone_free(xfs_qm_dqzone, dqp);
81 XFS_STATS_DEC(xs_qm_dquot);
85 * If default limits are in force, push them into the dquot now.
86 * We overwrite the dquot limits only if they are zero and this
87 * is not the root dquot.
90 xfs_qm_adjust_dqlimits(
94 struct xfs_quotainfo *q = mp->m_quotainfo;
95 struct xfs_disk_dquot *d = &dq->q_core;
100 if (q->qi_bsoftlimit && !d->d_blk_softlimit) {
101 d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit);
104 if (q->qi_bhardlimit && !d->d_blk_hardlimit) {
105 d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit);
108 if (q->qi_isoftlimit && !d->d_ino_softlimit)
109 d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit);
110 if (q->qi_ihardlimit && !d->d_ino_hardlimit)
111 d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit);
112 if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit)
113 d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit);
114 if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit)
115 d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit);
118 xfs_dquot_set_prealloc_limits(dq);
122 * Check the limits and timers of a dquot and start or reset timers
124 * This gets called even when quota enforcement is OFF, which makes our
125 * life a little less complicated. (We just don't reject any quota
126 * reservations in that case, when enforcement is off).
127 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
129 * In contrast, warnings are a little different in that they don't
130 * 'automatically' get started when limits get exceeded. They do
131 * get reset to zero, however, when we find the count to be under
132 * the soft limit (they are only ever set non-zero via userspace).
135 xfs_qm_adjust_dqtimers(
142 if (d->d_blk_hardlimit)
143 ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
144 be64_to_cpu(d->d_blk_hardlimit));
145 if (d->d_ino_hardlimit)
146 ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
147 be64_to_cpu(d->d_ino_hardlimit));
148 if (d->d_rtb_hardlimit)
149 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
150 be64_to_cpu(d->d_rtb_hardlimit));
154 if ((d->d_blk_softlimit &&
155 (be64_to_cpu(d->d_bcount) >
156 be64_to_cpu(d->d_blk_softlimit))) ||
157 (d->d_blk_hardlimit &&
158 (be64_to_cpu(d->d_bcount) >
159 be64_to_cpu(d->d_blk_hardlimit)))) {
160 d->d_btimer = cpu_to_be32(get_seconds() +
161 mp->m_quotainfo->qi_btimelimit);
166 if ((!d->d_blk_softlimit ||
167 (be64_to_cpu(d->d_bcount) <=
168 be64_to_cpu(d->d_blk_softlimit))) &&
169 (!d->d_blk_hardlimit ||
170 (be64_to_cpu(d->d_bcount) <=
171 be64_to_cpu(d->d_blk_hardlimit)))) {
177 if ((d->d_ino_softlimit &&
178 (be64_to_cpu(d->d_icount) >
179 be64_to_cpu(d->d_ino_softlimit))) ||
180 (d->d_ino_hardlimit &&
181 (be64_to_cpu(d->d_icount) >
182 be64_to_cpu(d->d_ino_hardlimit)))) {
183 d->d_itimer = cpu_to_be32(get_seconds() +
184 mp->m_quotainfo->qi_itimelimit);
189 if ((!d->d_ino_softlimit ||
190 (be64_to_cpu(d->d_icount) <=
191 be64_to_cpu(d->d_ino_softlimit))) &&
192 (!d->d_ino_hardlimit ||
193 (be64_to_cpu(d->d_icount) <=
194 be64_to_cpu(d->d_ino_hardlimit)))) {
199 if (!d->d_rtbtimer) {
200 if ((d->d_rtb_softlimit &&
201 (be64_to_cpu(d->d_rtbcount) >
202 be64_to_cpu(d->d_rtb_softlimit))) ||
203 (d->d_rtb_hardlimit &&
204 (be64_to_cpu(d->d_rtbcount) >
205 be64_to_cpu(d->d_rtb_hardlimit)))) {
206 d->d_rtbtimer = cpu_to_be32(get_seconds() +
207 mp->m_quotainfo->qi_rtbtimelimit);
212 if ((!d->d_rtb_softlimit ||
213 (be64_to_cpu(d->d_rtbcount) <=
214 be64_to_cpu(d->d_rtb_softlimit))) &&
215 (!d->d_rtb_hardlimit ||
216 (be64_to_cpu(d->d_rtbcount) <=
217 be64_to_cpu(d->d_rtb_hardlimit)))) {
224 * initialize a buffer full of dquots and log the whole thing
227 xfs_qm_init_dquot_blk(
234 struct xfs_quotainfo *q = mp->m_quotainfo;
239 ASSERT(xfs_buf_islocked(bp));
244 * ID of the first dquot in the block - id's are zero based.
246 curid = id - (id % q->qi_dqperchunk);
248 memset(d, 0, BBTOB(q->qi_dqchunklen));
249 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
250 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
251 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
252 d->dd_diskdq.d_id = cpu_to_be32(curid);
253 d->dd_diskdq.d_flags = type;
254 if (xfs_sb_version_hascrc(&mp->m_sb)) {
255 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid);
256 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
261 xfs_trans_dquot_buf(tp, bp,
262 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
263 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
264 XFS_BLF_GDQUOT_BUF)));
265 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
269 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
270 * watermarks correspond to the soft and hard limits by default. If a soft limit
271 * is not specified, we use 95% of the hard limit.
274 xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
278 dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit);
279 dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit);
280 if (!dqp->q_prealloc_lo_wmark) {
281 dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
282 do_div(dqp->q_prealloc_lo_wmark, 100);
283 dqp->q_prealloc_lo_wmark *= 95;
286 space = dqp->q_prealloc_hi_wmark;
289 dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
290 dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
291 dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
295 xfs_dquot_buf_verify_crc(
296 struct xfs_mount *mp,
299 struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr;
303 if (!xfs_sb_version_hascrc(&mp->m_sb))
307 * if we are in log recovery, the quota subsystem has not been
308 * initialised so we have no quotainfo structure. In that case, we need
309 * to manually calculate the number of dquots in the buffer.
312 ndquots = mp->m_quotainfo->qi_dqperchunk;
314 ndquots = xfs_qm_calc_dquots_per_chunk(mp,
315 XFS_BB_TO_FSB(mp, bp->b_length));
317 for (i = 0; i < ndquots; i++, d++) {
318 if (!xfs_verify_cksum((char *)d, sizeof(struct xfs_dqblk),
321 if (!uuid_equal(&d->dd_uuid, &mp->m_sb.sb_uuid))
328 xfs_dquot_buf_verify(
329 struct xfs_mount *mp,
332 struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr;
338 * if we are in log recovery, the quota subsystem has not been
339 * initialised so we have no quotainfo structure. In that case, we need
340 * to manually calculate the number of dquots in the buffer.
343 ndquots = mp->m_quotainfo->qi_dqperchunk;
345 ndquots = xfs_qm_calc_dquots_per_chunk(mp, bp->b_length);
348 * On the first read of the buffer, verify that each dquot is valid.
349 * We don't know what the id of the dquot is supposed to be, just that
350 * they should be increasing monotonically within the buffer. If the
351 * first id is corrupt, then it will fail on the second dquot in the
352 * buffer so corruptions could point to the wrong dquot in this case.
354 for (i = 0; i < ndquots; i++) {
355 struct xfs_disk_dquot *ddq;
358 ddq = &d[i].dd_diskdq;
361 id = be32_to_cpu(ddq->d_id);
363 error = xfs_qm_dqcheck(mp, ddq, id + i, 0, XFS_QMOPT_DOWARN,
364 "xfs_dquot_buf_verify");
372 xfs_dquot_buf_read_verify(
375 struct xfs_mount *mp = bp->b_target->bt_mount;
377 if (!xfs_dquot_buf_verify_crc(mp, bp) || !xfs_dquot_buf_verify(mp, bp)) {
378 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
379 xfs_buf_ioerror(bp, EFSCORRUPTED);
384 * we don't calculate the CRC here as that is done when the dquot is flushed to
385 * the buffer after the update is done. This ensures that the dquot in the
386 * buffer always has an up-to-date CRC value.
389 xfs_dquot_buf_write_verify(
392 struct xfs_mount *mp = bp->b_target->bt_mount;
394 if (!xfs_dquot_buf_verify(mp, bp)) {
395 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
396 xfs_buf_ioerror(bp, EFSCORRUPTED);
401 const struct xfs_buf_ops xfs_dquot_buf_ops = {
402 .verify_read = xfs_dquot_buf_read_verify,
403 .verify_write = xfs_dquot_buf_write_verify,
407 * Allocate a block and fill it with dquots.
408 * This is called when the bmapi finds a hole.
416 xfs_fileoff_t offset_fsb,
419 xfs_fsblock_t firstblock;
420 xfs_bmap_free_t flist;
422 int nmaps, error, committed;
424 xfs_trans_t *tp = *tpp;
428 trace_xfs_dqalloc(dqp);
431 * Initialize the bmap freelist prior to calling bmapi code.
433 xfs_bmap_init(&flist, &firstblock);
434 xfs_ilock(quotip, XFS_ILOCK_EXCL);
436 * Return if this type of quotas is turned off while we didn't
439 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
440 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
444 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
446 error = xfs_bmapi_write(tp, quotip, offset_fsb,
447 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
448 &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp),
449 &map, &nmaps, &flist);
452 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
454 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
455 (map.br_startblock != HOLESTARTBLOCK));
458 * Keep track of the blkno to save a lookup later
460 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
462 /* now we can just get the buffer (there's nothing to read yet) */
463 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
465 mp->m_quotainfo->qi_dqchunklen,
468 error = xfs_buf_geterror(bp);
471 bp->b_ops = &xfs_dquot_buf_ops;
474 * Make a chunk of dquots out of this buffer and log
477 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
478 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
481 * xfs_bmap_finish() may commit the current transaction and
482 * start a second transaction if the freelist is not empty.
484 * Since we still want to modify this buffer, we need to
485 * ensure that the buffer is not released on commit of
486 * the first transaction and ensure the buffer is added to the
487 * second transaction.
489 * If there is only one transaction then don't stop the buffer
490 * from being released when it commits later on.
493 xfs_trans_bhold(tp, bp);
495 if ((error = xfs_bmap_finish(tpp, &flist, &committed))) {
501 xfs_trans_bjoin(tp, bp);
503 xfs_trans_bhold_release(tp, bp);
510 xfs_bmap_cancel(&flist);
512 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
518 struct xfs_mount *mp,
519 struct xfs_trans *tp,
520 struct xfs_dquot *dqp,
522 struct xfs_buf **bpp)
525 struct xfs_disk_dquot *ddq;
530 * Read the buffer without verification so we get the corrupted
531 * buffer returned to us. make sure we verify it on write, though.
533 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno,
534 mp->m_quotainfo->qi_dqchunklen,
538 ASSERT(*bpp == NULL);
539 return XFS_ERROR(error);
541 (*bpp)->b_ops = &xfs_dquot_buf_ops;
543 ASSERT(xfs_buf_islocked(*bpp));
544 d = (struct xfs_dqblk *)(*bpp)->b_addr;
546 /* Do the actual repair of dquots in this buffer */
547 for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) {
548 ddq = &d[i].dd_diskdq;
549 error = xfs_qm_dqcheck(mp, ddq, firstid + i,
550 dqp->dq_flags & XFS_DQ_ALLTYPES,
551 XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair");
553 /* repair failed, we're screwed */
554 xfs_trans_brelse(tp, *bpp);
555 return XFS_ERROR(EIO);
563 * Maps a dquot to the buffer containing its on-disk version.
564 * This returns a ptr to the buffer containing the on-disk dquot
565 * in the bpp param, and a ptr to the on-disk dquot within that buffer
571 xfs_disk_dquot_t **O_ddpp,
575 struct xfs_bmbt_irec map;
576 int nmaps = 1, error;
578 struct xfs_inode *quotip = xfs_dq_to_quota_inode(dqp);
579 struct xfs_mount *mp = dqp->q_mount;
580 xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id);
581 struct xfs_trans *tp = (tpp ? *tpp : NULL);
583 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
585 xfs_ilock(quotip, XFS_ILOCK_SHARED);
586 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
588 * Return if this type of quotas is turned off while we
589 * didn't have the quota inode lock.
591 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
596 * Find the block map; no allocations yet
598 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
599 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
601 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
606 ASSERT(map.br_blockcount == 1);
609 * Offset of dquot in the (fixed sized) dquot chunk.
611 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
614 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
615 if (map.br_startblock == HOLESTARTBLOCK) {
617 * We don't allocate unless we're asked to
619 if (!(flags & XFS_QMOPT_DQALLOC))
623 error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
624 dqp->q_fileoffset, &bp);
629 trace_xfs_dqtobp_read(dqp);
632 * store the blkno etc so that we don't have to do the
633 * mapping all the time
635 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
637 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
639 mp->m_quotainfo->qi_dqchunklen,
640 0, &bp, &xfs_dquot_buf_ops);
642 if (error == EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) {
643 xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff *
644 mp->m_quotainfo->qi_dqperchunk;
646 error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp);
651 return XFS_ERROR(error);
655 ASSERT(xfs_buf_islocked(bp));
657 *O_ddpp = bp->b_addr + dqp->q_bufoffset;
664 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
665 * and release the buffer immediately.
667 * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed.
671 struct xfs_mount *mp,
675 struct xfs_dquot **O_dqpp)
677 struct xfs_dquot *dqp;
678 struct xfs_disk_dquot *ddqp;
680 struct xfs_trans *tp = NULL;
685 dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP);
687 dqp->dq_flags = type;
688 dqp->q_core.d_id = cpu_to_be32(id);
690 INIT_LIST_HEAD(&dqp->q_lru);
691 mutex_init(&dqp->q_qlock);
692 init_waitqueue_head(&dqp->q_pinwait);
695 * Because we want to use a counting completion, complete
696 * the flush completion once to allow a single access to
697 * the flush completion without blocking.
699 init_completion(&dqp->q_flush);
700 complete(&dqp->q_flush);
703 * Make sure group quotas have a different lock class than user
706 if (!(type & XFS_DQ_USER))
707 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class);
709 XFS_STATS_INC(xs_qm_dquot);
711 trace_xfs_dqread(dqp);
713 if (flags & XFS_QMOPT_DQALLOC) {
714 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
715 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_attrsetm,
716 XFS_QM_DQALLOC_SPACE_RES(mp), 0);
719 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
723 * get a pointer to the on-disk dquot and the buffer containing it
724 * dqp already knows its own type (GROUP/USER).
726 error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags);
729 * This can happen if quotas got turned off (ESRCH),
730 * or if the dquot didn't exist on disk and we ask to
733 trace_xfs_dqread_fail(dqp);
734 cancelflags |= XFS_TRANS_ABORT;
738 /* copy everything from disk dquot to the incore dquot */
739 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
740 xfs_qm_dquot_logitem_init(dqp);
743 * Reservation counters are defined as reservation plus current usage
744 * to avoid having to add every time.
746 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
747 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
748 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
750 /* initialize the dquot speculative prealloc thresholds */
751 xfs_dquot_set_prealloc_limits(dqp);
753 /* Mark the buf so that this will stay incore a little longer */
754 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
757 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
758 * So we need to release with xfs_trans_brelse().
759 * The strategy here is identical to that of inodes; we lock
760 * the dquot in xfs_qm_dqget() before making it accessible to
761 * others. This is because dquots, like inodes, need a good level of
762 * concurrency, and we don't want to take locks on the entire buffers
763 * for dquot accesses.
764 * Note also that the dquot buffer may even be dirty at this point, if
765 * this particular dquot was repaired. We still aren't afraid to
766 * brelse it because we have the changes incore.
768 ASSERT(xfs_buf_islocked(bp));
769 xfs_trans_brelse(tp, bp);
772 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
782 xfs_trans_cancel(tp, cancelflags);
784 xfs_qm_dqdestroy(dqp);
790 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
791 * a locked dquot, doing an allocation (if requested) as needed.
792 * When both an inode and an id are given, the inode's id takes precedence.
793 * That is, if the id changes while we don't hold the ilock inside this
794 * function, the new dquot is returned, not necessarily the one requested
795 * in the id argument.
800 xfs_inode_t *ip, /* locked inode (optional) */
801 xfs_dqid_t id, /* uid/projid/gid depending on type */
802 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
803 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
804 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
806 struct xfs_quotainfo *qi = mp->m_quotainfo;
807 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
808 struct xfs_dquot *dqp;
811 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
812 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
813 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
814 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
819 if (xfs_do_dqerror) {
820 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
821 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
822 xfs_debug(mp, "Returning error in dqget");
827 ASSERT(type == XFS_DQ_USER ||
828 type == XFS_DQ_PROJ ||
829 type == XFS_DQ_GROUP);
831 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
832 ASSERT(xfs_inode_dquot(ip, type) == NULL);
837 mutex_lock(&qi->qi_tree_lock);
838 dqp = radix_tree_lookup(tree, id);
841 if (dqp->dq_flags & XFS_DQ_FREEING) {
843 mutex_unlock(&qi->qi_tree_lock);
844 trace_xfs_dqget_freeing(dqp);
850 mutex_unlock(&qi->qi_tree_lock);
852 trace_xfs_dqget_hit(dqp);
853 XFS_STATS_INC(xs_qm_dqcachehits);
857 mutex_unlock(&qi->qi_tree_lock);
858 XFS_STATS_INC(xs_qm_dqcachemisses);
861 * Dquot cache miss. We don't want to keep the inode lock across
862 * a (potential) disk read. Also we don't want to deal with the lock
863 * ordering between quotainode and this inode. OTOH, dropping the inode
864 * lock here means dealing with a chown that can happen before
865 * we re-acquire the lock.
868 xfs_iunlock(ip, XFS_ILOCK_EXCL);
870 error = xfs_qm_dqread(mp, id, type, flags, &dqp);
873 xfs_ilock(ip, XFS_ILOCK_EXCL);
880 * A dquot could be attached to this inode by now, since
881 * we had dropped the ilock.
883 if (xfs_this_quota_on(mp, type)) {
884 struct xfs_dquot *dqp1;
886 dqp1 = xfs_inode_dquot(ip, type);
888 xfs_qm_dqdestroy(dqp);
894 /* inode stays locked on return */
895 xfs_qm_dqdestroy(dqp);
896 return XFS_ERROR(ESRCH);
900 mutex_lock(&qi->qi_tree_lock);
901 error = -radix_tree_insert(tree, id, dqp);
902 if (unlikely(error)) {
903 WARN_ON(error != EEXIST);
906 * Duplicate found. Just throw away the new dquot and start
909 mutex_unlock(&qi->qi_tree_lock);
910 trace_xfs_dqget_dup(dqp);
911 xfs_qm_dqdestroy(dqp);
912 XFS_STATS_INC(xs_qm_dquot_dups);
917 * We return a locked dquot to the caller, with a reference taken
923 mutex_unlock(&qi->qi_tree_lock);
926 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
927 trace_xfs_dqget_miss(dqp);
935 struct xfs_dquot *dqp)
937 struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
938 struct xfs_dquot *gdqp;
939 struct xfs_dquot *pdqp;
941 trace_xfs_dqput_free(dqp);
943 mutex_lock(&qi->qi_lru_lock);
944 if (list_empty(&dqp->q_lru)) {
945 list_add_tail(&dqp->q_lru, &qi->qi_lru_list);
947 XFS_STATS_INC(xs_qm_dquot_unused);
949 mutex_unlock(&qi->qi_lru_lock);
952 * If we just added a udquot to the freelist, then we want to release
953 * the gdquot/pdquot reference that it (probably) has. Otherwise it'll
954 * keep the gdquot/pdquot from getting reclaimed.
956 gdqp = dqp->q_gdquot;
959 dqp->q_gdquot = NULL;
962 pdqp = dqp->q_pdquot;
965 dqp->q_pdquot = NULL;
970 * If we had a group/project quota hint, release it now.
979 * Release a reference to the dquot (decrement ref-count) and unlock it.
981 * If there is a group quota attached to this dquot, carefully release that
982 * too without tripping over deadlocks'n'stuff.
986 struct xfs_dquot *dqp)
988 ASSERT(dqp->q_nrefs > 0);
989 ASSERT(XFS_DQ_IS_LOCKED(dqp));
991 trace_xfs_dqput(dqp);
993 if (--dqp->q_nrefs > 0)
996 xfs_qm_dqput_final(dqp);
1000 * Release a dquot. Flush it if dirty, then dqput() it.
1001 * dquot must not be locked.
1010 trace_xfs_dqrele(dqp);
1014 * We don't care to flush it if the dquot is dirty here.
1015 * That will create stutters that we want to avoid.
1016 * Instead we do a delayed write when we try to reclaim
1017 * a dirty dquot. Also xfs_sync will take part of the burden...
1023 * This is the dquot flushing I/O completion routine. It is called
1024 * from interrupt level when the buffer containing the dquot is
1025 * flushed to disk. It is responsible for removing the dquot logitem
1026 * from the AIL if it has not been re-logged, and unlocking the dquot's
1027 * flush lock. This behavior is very similar to that of inodes..
1030 xfs_qm_dqflush_done(
1032 struct xfs_log_item *lip)
1034 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
1035 xfs_dquot_t *dqp = qip->qli_dquot;
1036 struct xfs_ail *ailp = lip->li_ailp;
1039 * We only want to pull the item from the AIL if its
1040 * location in the log has not changed since we started the flush.
1041 * Thus, we only bother if the dquot's lsn has
1042 * not changed. First we check the lsn outside the lock
1043 * since it's cheaper, and then we recheck while
1044 * holding the lock before removing the dquot from the AIL.
1046 if ((lip->li_flags & XFS_LI_IN_AIL) &&
1047 lip->li_lsn == qip->qli_flush_lsn) {
1049 /* xfs_trans_ail_delete() drops the AIL lock. */
1050 spin_lock(&ailp->xa_lock);
1051 if (lip->li_lsn == qip->qli_flush_lsn)
1052 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
1054 spin_unlock(&ailp->xa_lock);
1058 * Release the dq's flush lock since we're done with it.
1064 * Write a modified dquot to disk.
1065 * The dquot must be locked and the flush lock too taken by caller.
1066 * The flush lock will not be unlocked until the dquot reaches the disk,
1067 * but the dquot is free to be unlocked and modified by the caller
1068 * in the interim. Dquot is still locked on return. This behavior is
1069 * identical to that of inodes.
1073 struct xfs_dquot *dqp,
1074 struct xfs_buf **bpp)
1076 struct xfs_mount *mp = dqp->q_mount;
1078 struct xfs_disk_dquot *ddqp;
1081 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1082 ASSERT(!completion_done(&dqp->q_flush));
1084 trace_xfs_dqflush(dqp);
1088 xfs_qm_dqunpin_wait(dqp);
1091 * This may have been unpinned because the filesystem is shutting
1092 * down forcibly. If that's the case we must not write this dquot
1093 * to disk, because the log record didn't make it to disk.
1095 * We also have to remove the log item from the AIL in this case,
1096 * as we wait for an emptry AIL as part of the unmount process.
1098 if (XFS_FORCED_SHUTDOWN(mp)) {
1099 struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
1100 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1102 spin_lock(&mp->m_ail->xa_lock);
1103 if (lip->li_flags & XFS_LI_IN_AIL)
1104 xfs_trans_ail_delete(mp->m_ail, lip,
1105 SHUTDOWN_CORRUPT_INCORE);
1107 spin_unlock(&mp->m_ail->xa_lock);
1108 error = XFS_ERROR(EIO);
1113 * Get the buffer containing the on-disk dquot
1115 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1116 mp->m_quotainfo->qi_dqchunklen, 0, &bp, NULL);
1121 * Calculate the location of the dquot inside the buffer.
1123 ddqp = bp->b_addr + dqp->q_bufoffset;
1126 * A simple sanity check in case we got a corrupted dquot..
1128 error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
1129 XFS_QMOPT_DOWARN, "dqflush (incore copy)");
1133 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1134 return XFS_ERROR(EIO);
1137 /* This is the only portion of data that needs to persist */
1138 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
1141 * Clear the dirty field and remember the flush lsn for later use.
1143 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1145 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1146 &dqp->q_logitem.qli_item.li_lsn);
1149 * copy the lsn into the on-disk dquot now while we have the in memory
1150 * dquot here. This can't be done later in the write verifier as we
1151 * can't get access to the log item at that point in time.
1153 * We also calculate the CRC here so that the on-disk dquot in the
1154 * buffer always has a valid CRC. This ensures there is no possibility
1155 * of a dquot without an up-to-date CRC getting to disk.
1157 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1158 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp;
1160 dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1161 xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk),
1166 * Attach an iodone routine so that we can remove this dquot from the
1167 * AIL and release the flush lock once the dquot is synced to disk.
1169 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1170 &dqp->q_logitem.qli_item);
1173 * If the buffer is pinned then push on the log so we won't
1174 * get stuck waiting in the write for too long.
1176 if (xfs_buf_ispinned(bp)) {
1177 trace_xfs_dqflush_force(dqp);
1178 xfs_log_force(mp, 0);
1181 trace_xfs_dqflush_done(dqp);
1187 return XFS_ERROR(EIO);
1191 * Lock two xfs_dquot structures.
1193 * To avoid deadlocks we always lock the quota structure with
1194 * the lowerd id first.
1203 if (be32_to_cpu(d1->q_core.d_id) >
1204 be32_to_cpu(d2->q_core.d_id)) {
1205 mutex_lock(&d2->q_qlock);
1206 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1208 mutex_lock(&d1->q_qlock);
1209 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1212 mutex_lock(&d1->q_qlock);
1214 mutex_lock(&d2->q_qlock);
1222 kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
1227 kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
1228 if (!xfs_qm_dqtrxzone)
1229 goto out_free_dqzone;
1234 kmem_zone_destroy(xfs_qm_dqzone);
1242 kmem_zone_destroy(xfs_qm_dqtrxzone);
1243 kmem_zone_destroy(xfs_qm_dqzone);