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_inode.h"
28 #include "xfs_bmap_util.h"
29 #include "xfs_alloc.h"
30 #include "xfs_quota.h"
31 #include "xfs_error.h"
32 #include "xfs_trans.h"
33 #include "xfs_buf_item.h"
34 #include "xfs_trans_space.h"
35 #include "xfs_trans_priv.h"
37 #include "xfs_cksum.h"
38 #include "xfs_trace.h"
40 #include "xfs_bmap_btree.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_group_class
;
66 static struct lock_class_key xfs_dquot_project_class
;
69 * This is called to free all the memory associated with a dquot
75 ASSERT(list_empty(&dqp
->q_lru
));
77 kmem_free(dqp
->q_logitem
.qli_item
.li_lv_shadow
);
78 mutex_destroy(&dqp
->q_qlock
);
80 XFS_STATS_DEC(dqp
->q_mount
, xs_qm_dquot
);
81 kmem_zone_free(xfs_qm_dqzone
, dqp
);
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
;
96 struct xfs_def_quota
*defq
;
100 defq
= xfs_get_defquota(dq
, q
);
102 if (defq
->bsoftlimit
&& !d
->d_blk_softlimit
) {
103 d
->d_blk_softlimit
= cpu_to_be64(defq
->bsoftlimit
);
106 if (defq
->bhardlimit
&& !d
->d_blk_hardlimit
) {
107 d
->d_blk_hardlimit
= cpu_to_be64(defq
->bhardlimit
);
110 if (defq
->isoftlimit
&& !d
->d_ino_softlimit
)
111 d
->d_ino_softlimit
= cpu_to_be64(defq
->isoftlimit
);
112 if (defq
->ihardlimit
&& !d
->d_ino_hardlimit
)
113 d
->d_ino_hardlimit
= cpu_to_be64(defq
->ihardlimit
);
114 if (defq
->rtbsoftlimit
&& !d
->d_rtb_softlimit
)
115 d
->d_rtb_softlimit
= cpu_to_be64(defq
->rtbsoftlimit
);
116 if (defq
->rtbhardlimit
&& !d
->d_rtb_hardlimit
)
117 d
->d_rtb_hardlimit
= cpu_to_be64(defq
->rtbhardlimit
);
120 xfs_dquot_set_prealloc_limits(dq
);
124 * Check the limits and timers of a dquot and start or reset timers
126 * This gets called even when quota enforcement is OFF, which makes our
127 * life a little less complicated. (We just don't reject any quota
128 * reservations in that case, when enforcement is off).
129 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
131 * In contrast, warnings are a little different in that they don't
132 * 'automatically' get started when limits get exceeded. They do
133 * get reset to zero, however, when we find the count to be under
134 * the soft limit (they are only ever set non-zero via userspace).
137 xfs_qm_adjust_dqtimers(
144 if (d
->d_blk_hardlimit
)
145 ASSERT(be64_to_cpu(d
->d_blk_softlimit
) <=
146 be64_to_cpu(d
->d_blk_hardlimit
));
147 if (d
->d_ino_hardlimit
)
148 ASSERT(be64_to_cpu(d
->d_ino_softlimit
) <=
149 be64_to_cpu(d
->d_ino_hardlimit
));
150 if (d
->d_rtb_hardlimit
)
151 ASSERT(be64_to_cpu(d
->d_rtb_softlimit
) <=
152 be64_to_cpu(d
->d_rtb_hardlimit
));
156 if ((d
->d_blk_softlimit
&&
157 (be64_to_cpu(d
->d_bcount
) >
158 be64_to_cpu(d
->d_blk_softlimit
))) ||
159 (d
->d_blk_hardlimit
&&
160 (be64_to_cpu(d
->d_bcount
) >
161 be64_to_cpu(d
->d_blk_hardlimit
)))) {
162 d
->d_btimer
= cpu_to_be32(get_seconds() +
163 mp
->m_quotainfo
->qi_btimelimit
);
168 if ((!d
->d_blk_softlimit
||
169 (be64_to_cpu(d
->d_bcount
) <=
170 be64_to_cpu(d
->d_blk_softlimit
))) &&
171 (!d
->d_blk_hardlimit
||
172 (be64_to_cpu(d
->d_bcount
) <=
173 be64_to_cpu(d
->d_blk_hardlimit
)))) {
179 if ((d
->d_ino_softlimit
&&
180 (be64_to_cpu(d
->d_icount
) >
181 be64_to_cpu(d
->d_ino_softlimit
))) ||
182 (d
->d_ino_hardlimit
&&
183 (be64_to_cpu(d
->d_icount
) >
184 be64_to_cpu(d
->d_ino_hardlimit
)))) {
185 d
->d_itimer
= cpu_to_be32(get_seconds() +
186 mp
->m_quotainfo
->qi_itimelimit
);
191 if ((!d
->d_ino_softlimit
||
192 (be64_to_cpu(d
->d_icount
) <=
193 be64_to_cpu(d
->d_ino_softlimit
))) &&
194 (!d
->d_ino_hardlimit
||
195 (be64_to_cpu(d
->d_icount
) <=
196 be64_to_cpu(d
->d_ino_hardlimit
)))) {
201 if (!d
->d_rtbtimer
) {
202 if ((d
->d_rtb_softlimit
&&
203 (be64_to_cpu(d
->d_rtbcount
) >
204 be64_to_cpu(d
->d_rtb_softlimit
))) ||
205 (d
->d_rtb_hardlimit
&&
206 (be64_to_cpu(d
->d_rtbcount
) >
207 be64_to_cpu(d
->d_rtb_hardlimit
)))) {
208 d
->d_rtbtimer
= cpu_to_be32(get_seconds() +
209 mp
->m_quotainfo
->qi_rtbtimelimit
);
214 if ((!d
->d_rtb_softlimit
||
215 (be64_to_cpu(d
->d_rtbcount
) <=
216 be64_to_cpu(d
->d_rtb_softlimit
))) &&
217 (!d
->d_rtb_hardlimit
||
218 (be64_to_cpu(d
->d_rtbcount
) <=
219 be64_to_cpu(d
->d_rtb_hardlimit
)))) {
226 * initialize a buffer full of dquots and log the whole thing
229 xfs_qm_init_dquot_blk(
236 struct xfs_quotainfo
*q
= mp
->m_quotainfo
;
242 ASSERT(xfs_buf_islocked(bp
));
247 * ID of the first dquot in the block - id's are zero based.
249 curid
= id
- (id
% q
->qi_dqperchunk
);
250 memset(d
, 0, BBTOB(q
->qi_dqchunklen
));
251 for (i
= 0; i
< q
->qi_dqperchunk
; i
++, d
++, curid
++) {
252 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
253 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
254 d
->dd_diskdq
.d_id
= cpu_to_be32(curid
);
255 d
->dd_diskdq
.d_flags
= type
;
256 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
257 uuid_copy(&d
->dd_uuid
, &mp
->m_sb
.sb_meta_uuid
);
258 xfs_update_cksum((char *)d
, sizeof(struct xfs_dqblk
),
263 xfs_trans_dquot_buf(tp
, bp
,
264 (type
& XFS_DQ_USER
? XFS_BLF_UDQUOT_BUF
:
265 ((type
& XFS_DQ_PROJ
) ? XFS_BLF_PDQUOT_BUF
:
266 XFS_BLF_GDQUOT_BUF
)));
267 xfs_trans_log_buf(tp
, bp
, 0, BBTOB(q
->qi_dqchunklen
) - 1);
271 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
272 * watermarks correspond to the soft and hard limits by default. If a soft limit
273 * is not specified, we use 95% of the hard limit.
276 xfs_dquot_set_prealloc_limits(struct xfs_dquot
*dqp
)
280 dqp
->q_prealloc_hi_wmark
= be64_to_cpu(dqp
->q_core
.d_blk_hardlimit
);
281 dqp
->q_prealloc_lo_wmark
= be64_to_cpu(dqp
->q_core
.d_blk_softlimit
);
282 if (!dqp
->q_prealloc_lo_wmark
) {
283 dqp
->q_prealloc_lo_wmark
= dqp
->q_prealloc_hi_wmark
;
284 do_div(dqp
->q_prealloc_lo_wmark
, 100);
285 dqp
->q_prealloc_lo_wmark
*= 95;
288 space
= dqp
->q_prealloc_hi_wmark
;
291 dqp
->q_low_space
[XFS_QLOWSP_1_PCNT
] = space
;
292 dqp
->q_low_space
[XFS_QLOWSP_3_PCNT
] = space
* 3;
293 dqp
->q_low_space
[XFS_QLOWSP_5_PCNT
] = space
* 5;
297 * Allocate a block and fill it with dquots.
298 * This is called when the bmapi finds a hole.
306 xfs_fileoff_t offset_fsb
,
309 xfs_fsblock_t firstblock
;
310 xfs_bmap_free_t flist
;
314 xfs_trans_t
*tp
= *tpp
;
318 trace_xfs_dqalloc(dqp
);
321 * Initialize the bmap freelist prior to calling bmapi code.
323 xfs_bmap_init(&flist
, &firstblock
);
324 xfs_ilock(quotip
, XFS_ILOCK_EXCL
);
326 * Return if this type of quotas is turned off while we didn't
329 if (!xfs_this_quota_on(dqp
->q_mount
, dqp
->dq_flags
)) {
330 xfs_iunlock(quotip
, XFS_ILOCK_EXCL
);
334 xfs_trans_ijoin(tp
, quotip
, XFS_ILOCK_EXCL
);
336 error
= xfs_bmapi_write(tp
, quotip
, offset_fsb
,
337 XFS_DQUOT_CLUSTER_SIZE_FSB
, XFS_BMAPI_METADATA
,
338 &firstblock
, XFS_QM_DQALLOC_SPACE_RES(mp
),
339 &map
, &nmaps
, &flist
);
342 ASSERT(map
.br_blockcount
== XFS_DQUOT_CLUSTER_SIZE_FSB
);
344 ASSERT((map
.br_startblock
!= DELAYSTARTBLOCK
) &&
345 (map
.br_startblock
!= HOLESTARTBLOCK
));
348 * Keep track of the blkno to save a lookup later
350 dqp
->q_blkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
);
352 /* now we can just get the buffer (there's nothing to read yet) */
353 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
355 mp
->m_quotainfo
->qi_dqchunklen
,
361 bp
->b_ops
= &xfs_dquot_buf_ops
;
364 * Make a chunk of dquots out of this buffer and log
367 xfs_qm_init_dquot_blk(tp
, mp
, be32_to_cpu(dqp
->q_core
.d_id
),
368 dqp
->dq_flags
& XFS_DQ_ALLTYPES
, bp
);
371 * xfs_bmap_finish() may commit the current transaction and
372 * start a second transaction if the freelist is not empty.
374 * Since we still want to modify this buffer, we need to
375 * ensure that the buffer is not released on commit of
376 * the first transaction and ensure the buffer is added to the
377 * second transaction.
379 * If there is only one transaction then don't stop the buffer
380 * from being released when it commits later on.
383 xfs_trans_bhold(tp
, bp
);
385 error
= xfs_bmap_finish(tpp
, &flist
, NULL
);
389 /* Transaction was committed? */
392 xfs_trans_bjoin(tp
, bp
);
394 xfs_trans_bhold_release(tp
, bp
);
401 xfs_bmap_cancel(&flist
);
403 xfs_iunlock(quotip
, XFS_ILOCK_EXCL
);
410 struct xfs_mount
*mp
,
411 struct xfs_trans
*tp
,
412 struct xfs_dquot
*dqp
,
414 struct xfs_buf
**bpp
)
417 struct xfs_disk_dquot
*ddq
;
422 * Read the buffer without verification so we get the corrupted
423 * buffer returned to us. make sure we verify it on write, though.
425 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, dqp
->q_blkno
,
426 mp
->m_quotainfo
->qi_dqchunklen
,
430 ASSERT(*bpp
== NULL
);
433 (*bpp
)->b_ops
= &xfs_dquot_buf_ops
;
435 ASSERT(xfs_buf_islocked(*bpp
));
436 d
= (struct xfs_dqblk
*)(*bpp
)->b_addr
;
438 /* Do the actual repair of dquots in this buffer */
439 for (i
= 0; i
< mp
->m_quotainfo
->qi_dqperchunk
; i
++) {
440 ddq
= &d
[i
].dd_diskdq
;
441 error
= xfs_dqcheck(mp
, ddq
, firstid
+ i
,
442 dqp
->dq_flags
& XFS_DQ_ALLTYPES
,
443 XFS_QMOPT_DQREPAIR
, "xfs_qm_dqrepair");
445 /* repair failed, we're screwed */
446 xfs_trans_brelse(tp
, *bpp
);
455 * Maps a dquot to the buffer containing its on-disk version.
456 * This returns a ptr to the buffer containing the on-disk dquot
457 * in the bpp param, and a ptr to the on-disk dquot within that buffer
463 xfs_disk_dquot_t
**O_ddpp
,
467 struct xfs_bmbt_irec map
;
468 int nmaps
= 1, error
;
470 struct xfs_inode
*quotip
;
471 struct xfs_mount
*mp
= dqp
->q_mount
;
472 xfs_dqid_t id
= be32_to_cpu(dqp
->q_core
.d_id
);
473 struct xfs_trans
*tp
= (tpp
? *tpp
: NULL
);
476 quotip
= xfs_quota_inode(dqp
->q_mount
, dqp
->dq_flags
);
477 dqp
->q_fileoffset
= (xfs_fileoff_t
)id
/ mp
->m_quotainfo
->qi_dqperchunk
;
479 lock_mode
= xfs_ilock_data_map_shared(quotip
);
480 if (!xfs_this_quota_on(dqp
->q_mount
, dqp
->dq_flags
)) {
482 * Return if this type of quotas is turned off while we
483 * didn't have the quota inode lock.
485 xfs_iunlock(quotip
, lock_mode
);
490 * Find the block map; no allocations yet
492 error
= xfs_bmapi_read(quotip
, dqp
->q_fileoffset
,
493 XFS_DQUOT_CLUSTER_SIZE_FSB
, &map
, &nmaps
, 0);
495 xfs_iunlock(quotip
, lock_mode
);
500 ASSERT(map
.br_blockcount
== 1);
503 * Offset of dquot in the (fixed sized) dquot chunk.
505 dqp
->q_bufoffset
= (id
% mp
->m_quotainfo
->qi_dqperchunk
) *
508 ASSERT(map
.br_startblock
!= DELAYSTARTBLOCK
);
509 if (map
.br_startblock
== HOLESTARTBLOCK
) {
511 * We don't allocate unless we're asked to
513 if (!(flags
& XFS_QMOPT_DQALLOC
))
517 error
= xfs_qm_dqalloc(tpp
, mp
, dqp
, quotip
,
518 dqp
->q_fileoffset
, &bp
);
523 trace_xfs_dqtobp_read(dqp
);
526 * store the blkno etc so that we don't have to do the
527 * mapping all the time
529 dqp
->q_blkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
);
531 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
533 mp
->m_quotainfo
->qi_dqchunklen
,
534 0, &bp
, &xfs_dquot_buf_ops
);
536 if (error
== -EFSCORRUPTED
&& (flags
& XFS_QMOPT_DQREPAIR
)) {
537 xfs_dqid_t firstid
= (xfs_dqid_t
)map
.br_startoff
*
538 mp
->m_quotainfo
->qi_dqperchunk
;
540 error
= xfs_qm_dqrepair(mp
, tp
, dqp
, firstid
, &bp
);
549 ASSERT(xfs_buf_islocked(bp
));
551 *O_ddpp
= bp
->b_addr
+ dqp
->q_bufoffset
;
558 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
559 * and release the buffer immediately.
561 * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed.
565 struct xfs_mount
*mp
,
569 struct xfs_dquot
**O_dqpp
)
571 struct xfs_dquot
*dqp
;
572 struct xfs_disk_dquot
*ddqp
;
574 struct xfs_trans
*tp
= NULL
;
577 dqp
= kmem_zone_zalloc(xfs_qm_dqzone
, KM_SLEEP
);
579 dqp
->dq_flags
= type
;
580 dqp
->q_core
.d_id
= cpu_to_be32(id
);
582 INIT_LIST_HEAD(&dqp
->q_lru
);
583 mutex_init(&dqp
->q_qlock
);
584 init_waitqueue_head(&dqp
->q_pinwait
);
587 * Because we want to use a counting completion, complete
588 * the flush completion once to allow a single access to
589 * the flush completion without blocking.
591 init_completion(&dqp
->q_flush
);
592 complete(&dqp
->q_flush
);
595 * Make sure group quotas have a different lock class than user
600 /* uses the default lock class */
603 lockdep_set_class(&dqp
->q_qlock
, &xfs_dquot_group_class
);
606 lockdep_set_class(&dqp
->q_qlock
, &xfs_dquot_project_class
);
613 XFS_STATS_INC(mp
, xs_qm_dquot
);
615 trace_xfs_dqread(dqp
);
617 if (flags
& XFS_QMOPT_DQALLOC
) {
618 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_qm_dqalloc
,
619 XFS_QM_DQALLOC_SPACE_RES(mp
), 0, 0, &tp
);
625 * get a pointer to the on-disk dquot and the buffer containing it
626 * dqp already knows its own type (GROUP/USER).
628 error
= xfs_qm_dqtobp(&tp
, dqp
, &ddqp
, &bp
, flags
);
631 * This can happen if quotas got turned off (ESRCH),
632 * or if the dquot didn't exist on disk and we ask to
635 trace_xfs_dqread_fail(dqp
);
639 /* copy everything from disk dquot to the incore dquot */
640 memcpy(&dqp
->q_core
, ddqp
, sizeof(xfs_disk_dquot_t
));
641 xfs_qm_dquot_logitem_init(dqp
);
644 * Reservation counters are defined as reservation plus current usage
645 * to avoid having to add every time.
647 dqp
->q_res_bcount
= be64_to_cpu(ddqp
->d_bcount
);
648 dqp
->q_res_icount
= be64_to_cpu(ddqp
->d_icount
);
649 dqp
->q_res_rtbcount
= be64_to_cpu(ddqp
->d_rtbcount
);
651 /* initialize the dquot speculative prealloc thresholds */
652 xfs_dquot_set_prealloc_limits(dqp
);
654 /* Mark the buf so that this will stay incore a little longer */
655 xfs_buf_set_ref(bp
, XFS_DQUOT_REF
);
658 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
659 * So we need to release with xfs_trans_brelse().
660 * The strategy here is identical to that of inodes; we lock
661 * the dquot in xfs_qm_dqget() before making it accessible to
662 * others. This is because dquots, like inodes, need a good level of
663 * concurrency, and we don't want to take locks on the entire buffers
664 * for dquot accesses.
665 * Note also that the dquot buffer may even be dirty at this point, if
666 * this particular dquot was repaired. We still aren't afraid to
667 * brelse it because we have the changes incore.
669 ASSERT(xfs_buf_islocked(bp
));
670 xfs_trans_brelse(tp
, bp
);
673 error
= xfs_trans_commit(tp
);
683 xfs_trans_cancel(tp
);
685 xfs_qm_dqdestroy(dqp
);
691 * Advance to the next id in the current chunk, or if at the
692 * end of the chunk, skip ahead to first id in next allocated chunk
693 * using the SEEK_DATA interface.
702 struct xfs_inode
*quotip
;
712 /* If new ID is within the current chunk, advancing it sufficed */
713 if (next_id
% mp
->m_quotainfo
->qi_dqperchunk
) {
718 /* Nope, next_id is now past the current chunk, so find the next one */
719 start
= (xfs_fsblock_t
)next_id
/ mp
->m_quotainfo
->qi_dqperchunk
;
721 quotip
= xfs_quota_inode(mp
, type
);
722 lock
= xfs_ilock_data_map_shared(quotip
);
724 offset
= __xfs_seek_hole_data(VFS_I(quotip
), XFS_FSB_TO_B(mp
, start
),
729 xfs_iunlock(quotip
, lock
);
731 /* -ENXIO is essentially "no more data" */
733 return (error
== -ENXIO
? -ENOENT
: error
);
735 /* Convert next data offset back to a quota id */
736 *id
= XFS_B_TO_FSB(mp
, offset
) * mp
->m_quotainfo
->qi_dqperchunk
;
741 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
742 * a locked dquot, doing an allocation (if requested) as needed.
743 * When both an inode and an id are given, the inode's id takes precedence.
744 * That is, if the id changes while we don't hold the ilock inside this
745 * function, the new dquot is returned, not necessarily the one requested
746 * in the id argument.
751 xfs_inode_t
*ip
, /* locked inode (optional) */
752 xfs_dqid_t id
, /* uid/projid/gid depending on type */
753 uint type
, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
754 uint flags
, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
755 xfs_dquot_t
**O_dqpp
) /* OUT : locked incore dquot */
757 struct xfs_quotainfo
*qi
= mp
->m_quotainfo
;
758 struct radix_tree_root
*tree
= xfs_dquot_tree(qi
, type
);
759 struct xfs_dquot
*dqp
;
763 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
764 if ((! XFS_IS_UQUOTA_ON(mp
) && type
== XFS_DQ_USER
) ||
765 (! XFS_IS_PQUOTA_ON(mp
) && type
== XFS_DQ_PROJ
) ||
766 (! XFS_IS_GQUOTA_ON(mp
) && type
== XFS_DQ_GROUP
)) {
771 if (xfs_do_dqerror
) {
772 if ((xfs_dqerror_target
== mp
->m_ddev_targp
) &&
773 (xfs_dqreq_num
++ % xfs_dqerror_mod
) == 0) {
774 xfs_debug(mp
, "Returning error in dqget");
779 ASSERT(type
== XFS_DQ_USER
||
780 type
== XFS_DQ_PROJ
||
781 type
== XFS_DQ_GROUP
);
783 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
784 ASSERT(xfs_inode_dquot(ip
, type
) == NULL
);
788 /* Get the end of the quota file if we need it */
789 if (flags
& XFS_QMOPT_DQNEXT
) {
790 struct xfs_inode
*quotip
;
794 quotip
= xfs_quota_inode(mp
, type
);
795 lock_mode
= xfs_ilock_data_map_shared(quotip
);
796 error
= xfs_bmap_last_offset(quotip
, &last
, XFS_DATA_FORK
);
797 xfs_iunlock(quotip
, lock_mode
);
800 eof
= XFS_FSB_TO_B(mp
, last
);
804 mutex_lock(&qi
->qi_tree_lock
);
805 dqp
= radix_tree_lookup(tree
, id
);
808 if (dqp
->dq_flags
& XFS_DQ_FREEING
) {
810 mutex_unlock(&qi
->qi_tree_lock
);
811 trace_xfs_dqget_freeing(dqp
);
816 /* uninit / unused quota found in radix tree, keep looking */
817 if (flags
& XFS_QMOPT_DQNEXT
) {
818 if (XFS_IS_DQUOT_UNINITIALIZED(dqp
)) {
820 mutex_unlock(&qi
->qi_tree_lock
);
821 error
= xfs_dq_get_next_id(mp
, type
, &id
, eof
);
829 mutex_unlock(&qi
->qi_tree_lock
);
831 trace_xfs_dqget_hit(dqp
);
832 XFS_STATS_INC(mp
, xs_qm_dqcachehits
);
836 mutex_unlock(&qi
->qi_tree_lock
);
837 XFS_STATS_INC(mp
, xs_qm_dqcachemisses
);
840 * Dquot cache miss. We don't want to keep the inode lock across
841 * a (potential) disk read. Also we don't want to deal with the lock
842 * ordering between quotainode and this inode. OTOH, dropping the inode
843 * lock here means dealing with a chown that can happen before
844 * we re-acquire the lock.
847 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
849 error
= xfs_qm_dqread(mp
, id
, type
, flags
, &dqp
);
852 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
854 /* If we are asked to find next active id, keep looking */
855 if (error
== -ENOENT
&& (flags
& XFS_QMOPT_DQNEXT
)) {
856 error
= xfs_dq_get_next_id(mp
, type
, &id
, eof
);
866 * A dquot could be attached to this inode by now, since
867 * we had dropped the ilock.
869 if (xfs_this_quota_on(mp
, type
)) {
870 struct xfs_dquot
*dqp1
;
872 dqp1
= xfs_inode_dquot(ip
, type
);
874 xfs_qm_dqdestroy(dqp
);
880 /* inode stays locked on return */
881 xfs_qm_dqdestroy(dqp
);
886 mutex_lock(&qi
->qi_tree_lock
);
887 error
= radix_tree_insert(tree
, id
, dqp
);
888 if (unlikely(error
)) {
889 WARN_ON(error
!= -EEXIST
);
892 * Duplicate found. Just throw away the new dquot and start
895 mutex_unlock(&qi
->qi_tree_lock
);
896 trace_xfs_dqget_dup(dqp
);
897 xfs_qm_dqdestroy(dqp
);
898 XFS_STATS_INC(mp
, xs_qm_dquot_dups
);
903 * We return a locked dquot to the caller, with a reference taken
909 mutex_unlock(&qi
->qi_tree_lock
);
911 /* If we are asked to find next active id, keep looking */
912 if (flags
& XFS_QMOPT_DQNEXT
) {
913 if (XFS_IS_DQUOT_UNINITIALIZED(dqp
)) {
915 error
= xfs_dq_get_next_id(mp
, type
, &id
, eof
);
923 ASSERT((ip
== NULL
) || xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
924 trace_xfs_dqget_miss(dqp
);
930 * Release a reference to the dquot (decrement ref-count) and unlock it.
932 * If there is a group quota attached to this dquot, carefully release that
933 * too without tripping over deadlocks'n'stuff.
937 struct xfs_dquot
*dqp
)
939 ASSERT(dqp
->q_nrefs
> 0);
940 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
942 trace_xfs_dqput(dqp
);
944 if (--dqp
->q_nrefs
== 0) {
945 struct xfs_quotainfo
*qi
= dqp
->q_mount
->m_quotainfo
;
946 trace_xfs_dqput_free(dqp
);
948 if (list_lru_add(&qi
->qi_lru
, &dqp
->q_lru
))
949 XFS_STATS_INC(dqp
->q_mount
, xs_qm_dquot_unused
);
955 * Release a dquot. Flush it if dirty, then dqput() it.
956 * dquot must not be locked.
965 trace_xfs_dqrele(dqp
);
969 * We don't care to flush it if the dquot is dirty here.
970 * That will create stutters that we want to avoid.
971 * Instead we do a delayed write when we try to reclaim
972 * a dirty dquot. Also xfs_sync will take part of the burden...
978 * This is the dquot flushing I/O completion routine. It is called
979 * from interrupt level when the buffer containing the dquot is
980 * flushed to disk. It is responsible for removing the dquot logitem
981 * from the AIL if it has not been re-logged, and unlocking the dquot's
982 * flush lock. This behavior is very similar to that of inodes..
987 struct xfs_log_item
*lip
)
989 xfs_dq_logitem_t
*qip
= (struct xfs_dq_logitem
*)lip
;
990 xfs_dquot_t
*dqp
= qip
->qli_dquot
;
991 struct xfs_ail
*ailp
= lip
->li_ailp
;
994 * We only want to pull the item from the AIL if its
995 * location in the log has not changed since we started the flush.
996 * Thus, we only bother if the dquot's lsn has
997 * not changed. First we check the lsn outside the lock
998 * since it's cheaper, and then we recheck while
999 * holding the lock before removing the dquot from the AIL.
1001 if ((lip
->li_flags
& XFS_LI_IN_AIL
) &&
1002 lip
->li_lsn
== qip
->qli_flush_lsn
) {
1004 /* xfs_trans_ail_delete() drops the AIL lock. */
1005 spin_lock(&ailp
->xa_lock
);
1006 if (lip
->li_lsn
== qip
->qli_flush_lsn
)
1007 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_CORRUPT_INCORE
);
1009 spin_unlock(&ailp
->xa_lock
);
1013 * Release the dq's flush lock since we're done with it.
1019 * Write a modified dquot to disk.
1020 * The dquot must be locked and the flush lock too taken by caller.
1021 * The flush lock will not be unlocked until the dquot reaches the disk,
1022 * but the dquot is free to be unlocked and modified by the caller
1023 * in the interim. Dquot is still locked on return. This behavior is
1024 * identical to that of inodes.
1028 struct xfs_dquot
*dqp
,
1029 struct xfs_buf
**bpp
)
1031 struct xfs_mount
*mp
= dqp
->q_mount
;
1033 struct xfs_disk_dquot
*ddqp
;
1036 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1037 ASSERT(!completion_done(&dqp
->q_flush
));
1039 trace_xfs_dqflush(dqp
);
1043 xfs_qm_dqunpin_wait(dqp
);
1046 * This may have been unpinned because the filesystem is shutting
1047 * down forcibly. If that's the case we must not write this dquot
1048 * to disk, because the log record didn't make it to disk.
1050 * We also have to remove the log item from the AIL in this case,
1051 * as we wait for an emptry AIL as part of the unmount process.
1053 if (XFS_FORCED_SHUTDOWN(mp
)) {
1054 struct xfs_log_item
*lip
= &dqp
->q_logitem
.qli_item
;
1055 dqp
->dq_flags
&= ~XFS_DQ_DIRTY
;
1057 xfs_trans_ail_remove(lip
, SHUTDOWN_CORRUPT_INCORE
);
1064 * Get the buffer containing the on-disk dquot
1066 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dqp
->q_blkno
,
1067 mp
->m_quotainfo
->qi_dqchunklen
, 0, &bp
,
1068 &xfs_dquot_buf_ops
);
1073 * Calculate the location of the dquot inside the buffer.
1075 ddqp
= bp
->b_addr
+ dqp
->q_bufoffset
;
1078 * A simple sanity check in case we got a corrupted dquot..
1080 error
= xfs_dqcheck(mp
, &dqp
->q_core
, be32_to_cpu(ddqp
->d_id
), 0,
1081 XFS_QMOPT_DOWARN
, "dqflush (incore copy)");
1085 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1089 /* This is the only portion of data that needs to persist */
1090 memcpy(ddqp
, &dqp
->q_core
, sizeof(xfs_disk_dquot_t
));
1093 * Clear the dirty field and remember the flush lsn for later use.
1095 dqp
->dq_flags
&= ~XFS_DQ_DIRTY
;
1097 xfs_trans_ail_copy_lsn(mp
->m_ail
, &dqp
->q_logitem
.qli_flush_lsn
,
1098 &dqp
->q_logitem
.qli_item
.li_lsn
);
1101 * copy the lsn into the on-disk dquot now while we have the in memory
1102 * dquot here. This can't be done later in the write verifier as we
1103 * can't get access to the log item at that point in time.
1105 * We also calculate the CRC here so that the on-disk dquot in the
1106 * buffer always has a valid CRC. This ensures there is no possibility
1107 * of a dquot without an up-to-date CRC getting to disk.
1109 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
1110 struct xfs_dqblk
*dqb
= (struct xfs_dqblk
*)ddqp
;
1112 dqb
->dd_lsn
= cpu_to_be64(dqp
->q_logitem
.qli_item
.li_lsn
);
1113 xfs_update_cksum((char *)dqb
, sizeof(struct xfs_dqblk
),
1118 * Attach an iodone routine so that we can remove this dquot from the
1119 * AIL and release the flush lock once the dquot is synced to disk.
1121 xfs_buf_attach_iodone(bp
, xfs_qm_dqflush_done
,
1122 &dqp
->q_logitem
.qli_item
);
1125 * If the buffer is pinned then push on the log so we won't
1126 * get stuck waiting in the write for too long.
1128 if (xfs_buf_ispinned(bp
)) {
1129 trace_xfs_dqflush_force(dqp
);
1130 xfs_log_force(mp
, 0);
1133 trace_xfs_dqflush_done(dqp
);
1143 * Lock two xfs_dquot structures.
1145 * To avoid deadlocks we always lock the quota structure with
1146 * the lowerd id first.
1155 if (be32_to_cpu(d1
->q_core
.d_id
) >
1156 be32_to_cpu(d2
->q_core
.d_id
)) {
1157 mutex_lock(&d2
->q_qlock
);
1158 mutex_lock_nested(&d1
->q_qlock
, XFS_QLOCK_NESTED
);
1160 mutex_lock(&d1
->q_qlock
);
1161 mutex_lock_nested(&d2
->q_qlock
, XFS_QLOCK_NESTED
);
1164 mutex_lock(&d1
->q_qlock
);
1166 mutex_lock(&d2
->q_qlock
);
1174 kmem_zone_init(sizeof(struct xfs_dquot
), "xfs_dquot");
1179 kmem_zone_init(sizeof(struct xfs_dquot_acct
), "xfs_dqtrx");
1180 if (!xfs_qm_dqtrxzone
)
1181 goto out_free_dqzone
;
1186 kmem_zone_destroy(xfs_qm_dqzone
);
1194 kmem_zone_destroy(xfs_qm_dqtrxzone
);
1195 kmem_zone_destroy(xfs_qm_dqzone
);