2 * Copyright (c) 2000-2005 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_ialloc_btree.h"
30 #include "xfs_dinode.h"
31 #include "xfs_inode.h"
32 #include "xfs_ialloc.h"
33 #include "xfs_itable.h"
34 #include "xfs_rtalloc.h"
35 #include "xfs_error.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_trans_space.h"
40 #include "xfs_utils.h"
42 #include "xfs_trace.h"
43 #include "xfs_icache.h"
44 #include "xfs_cksum.h"
47 * The global quota manager. There is only one of these for the entire
48 * system, _not_ one per file system. XQM keeps track of the overall
49 * quota functionality, including maintaining the freelist and hash
52 STATIC
int xfs_qm_init_quotainos(xfs_mount_t
*);
53 STATIC
int xfs_qm_init_quotainfo(xfs_mount_t
*);
54 STATIC
int xfs_qm_shake(struct shrinker
*, struct shrink_control
*);
57 * We use the batch lookup interface to iterate over the dquots as it
58 * currently is the only interface into the radix tree code that allows
59 * fuzzy lookups instead of exact matches. Holding the lock over multiple
60 * operations is fine as all callers are used either during mount/umount
63 #define XFS_DQ_LOOKUP_BATCH 32
69 int (*execute
)(struct xfs_dquot
*dqp
, void *data
),
72 struct xfs_quotainfo
*qi
= mp
->m_quotainfo
;
73 struct radix_tree_root
*tree
= xfs_dquot_tree(qi
, type
);
85 struct xfs_dquot
*batch
[XFS_DQ_LOOKUP_BATCH
];
89 mutex_lock(&qi
->qi_tree_lock
);
90 nr_found
= radix_tree_gang_lookup(tree
, (void **)batch
,
91 next_index
, XFS_DQ_LOOKUP_BATCH
);
93 mutex_unlock(&qi
->qi_tree_lock
);
97 for (i
= 0; i
< nr_found
; i
++) {
98 struct xfs_dquot
*dqp
= batch
[i
];
100 next_index
= be32_to_cpu(dqp
->q_core
.d_id
) + 1;
102 error
= execute(batch
[i
], data
);
103 if (error
== EAGAIN
) {
107 if (error
&& last_error
!= EFSCORRUPTED
)
111 mutex_unlock(&qi
->qi_tree_lock
);
113 /* bail out if the filesystem is corrupted. */
114 if (last_error
== EFSCORRUPTED
) {
130 * Purge a dquot from all tracking data structures and free it.
134 struct xfs_dquot
*dqp
,
137 struct xfs_mount
*mp
= dqp
->q_mount
;
138 struct xfs_quotainfo
*qi
= mp
->m_quotainfo
;
139 struct xfs_dquot
*gdqp
= NULL
;
142 if ((dqp
->dq_flags
& XFS_DQ_FREEING
) || dqp
->q_nrefs
!= 0) {
148 * If this quota has a group hint attached, prepare for releasing it
151 gdqp
= dqp
->q_gdquot
;
154 dqp
->q_gdquot
= NULL
;
157 dqp
->dq_flags
|= XFS_DQ_FREEING
;
162 * If we are turning this type of quotas off, we don't care
163 * about the dirty metadata sitting in this dquot. OTOH, if
164 * we're unmounting, we do care, so we flush it and wait.
166 if (XFS_DQ_IS_DIRTY(dqp
)) {
167 struct xfs_buf
*bp
= NULL
;
171 * We don't care about getting disk errors here. We need
172 * to purge this dquot anyway, so we go ahead regardless.
174 error
= xfs_qm_dqflush(dqp
, &bp
);
176 xfs_warn(mp
, "%s: dquot %p flush failed",
179 error
= xfs_bwrite(bp
);
185 ASSERT(atomic_read(&dqp
->q_pincount
) == 0);
186 ASSERT(XFS_FORCED_SHUTDOWN(mp
) ||
187 !(dqp
->q_logitem
.qli_item
.li_flags
& XFS_LI_IN_AIL
));
192 radix_tree_delete(xfs_dquot_tree(qi
, dqp
->q_core
.d_flags
),
193 be32_to_cpu(dqp
->q_core
.d_id
));
197 * We move dquots to the freelist as soon as their reference count
198 * hits zero, so it really should be on the freelist here.
200 mutex_lock(&qi
->qi_lru_lock
);
201 ASSERT(!list_empty(&dqp
->q_lru
));
202 list_del_init(&dqp
->q_lru
);
204 XFS_STATS_DEC(xs_qm_dquot_unused
);
205 mutex_unlock(&qi
->qi_lru_lock
);
207 xfs_qm_dqdestroy(dqp
);
215 * Purge the dquot cache.
219 struct xfs_mount
*mp
,
222 if (flags
& XFS_QMOPT_UQUOTA
)
223 xfs_qm_dquot_walk(mp
, XFS_DQ_USER
, xfs_qm_dqpurge
, NULL
);
224 if (flags
& XFS_QMOPT_GQUOTA
)
225 xfs_qm_dquot_walk(mp
, XFS_DQ_GROUP
, xfs_qm_dqpurge
, NULL
);
226 if (flags
& XFS_QMOPT_PQUOTA
)
227 xfs_qm_dquot_walk(mp
, XFS_DQ_PROJ
, xfs_qm_dqpurge
, NULL
);
231 * Just destroy the quotainfo structure.
235 struct xfs_mount
*mp
)
237 if (mp
->m_quotainfo
) {
238 xfs_qm_dqpurge_all(mp
, XFS_QMOPT_QUOTALL
);
239 xfs_qm_destroy_quotainfo(mp
);
245 * This is called from xfs_mountfs to start quotas and initialize all
246 * necessary data structures like quotainfo. This is also responsible for
247 * running a quotacheck as necessary. We are guaranteed that the superblock
248 * is consistently read in at this point.
250 * If we fail here, the mount will continue with quota turned off. We don't
251 * need to inidicate success or failure at all.
261 * If quotas on realtime volumes is not supported, we disable
262 * quotas immediately.
264 if (mp
->m_sb
.sb_rextents
) {
265 xfs_notice(mp
, "Cannot turn on quotas for realtime filesystem");
270 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
273 * Allocate the quotainfo structure inside the mount struct, and
274 * create quotainode(s), and change/rev superblock if necessary.
276 error
= xfs_qm_init_quotainfo(mp
);
279 * We must turn off quotas.
281 ASSERT(mp
->m_quotainfo
== NULL
);
286 * If any of the quotas are not consistent, do a quotacheck.
288 if (XFS_QM_NEED_QUOTACHECK(mp
)) {
289 error
= xfs_qm_quotacheck(mp
);
291 /* Quotacheck failed and disabled quotas. */
296 * If one type of quotas is off, then it will lose its
297 * quotachecked status, since we won't be doing accounting for
300 if (!XFS_IS_UQUOTA_ON(mp
))
301 mp
->m_qflags
&= ~XFS_UQUOTA_CHKD
;
302 if (!(XFS_IS_GQUOTA_ON(mp
) || XFS_IS_PQUOTA_ON(mp
)))
303 mp
->m_qflags
&= ~XFS_OQUOTA_CHKD
;
307 * We actually don't have to acquire the m_sb_lock at all.
308 * This can only be called from mount, and that's single threaded. XXX
310 spin_lock(&mp
->m_sb_lock
);
311 sbf
= mp
->m_sb
.sb_qflags
;
312 mp
->m_sb
.sb_qflags
= mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
;
313 spin_unlock(&mp
->m_sb_lock
);
315 if (sbf
!= (mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
)) {
316 if (xfs_qm_write_sb_changes(mp
, XFS_SB_QFLAGS
)) {
318 * We could only have been turning quotas off.
319 * We aren't in very good shape actually because
320 * the incore structures are convinced that quotas are
321 * off, but the on disk superblock doesn't know that !
323 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp
)));
324 xfs_alert(mp
, "%s: Superblock update failed!",
330 xfs_warn(mp
, "Failed to initialize disk quotas.");
336 * Called from the vfsops layer.
339 xfs_qm_unmount_quotas(
343 * Release the dquots that root inode, et al might be holding,
344 * before we flush quotas and blow away the quotainfo structure.
346 ASSERT(mp
->m_rootip
);
347 xfs_qm_dqdetach(mp
->m_rootip
);
349 xfs_qm_dqdetach(mp
->m_rbmip
);
351 xfs_qm_dqdetach(mp
->m_rsumip
);
354 * Release the quota inodes.
356 if (mp
->m_quotainfo
) {
357 if (mp
->m_quotainfo
->qi_uquotaip
) {
358 IRELE(mp
->m_quotainfo
->qi_uquotaip
);
359 mp
->m_quotainfo
->qi_uquotaip
= NULL
;
361 if (mp
->m_quotainfo
->qi_gquotaip
) {
362 IRELE(mp
->m_quotainfo
->qi_gquotaip
);
363 mp
->m_quotainfo
->qi_gquotaip
= NULL
;
374 xfs_dquot_t
*udqhint
, /* hint */
375 xfs_dquot_t
**IO_idqpp
)
380 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
384 * See if we already have it in the inode itself. IO_idqpp is
385 * &i_udquot or &i_gdquot. This made the code look weird, but
386 * made the logic a lot simpler.
390 trace_xfs_dqattach_found(dqp
);
395 * udqhint is the i_udquot field in inode, and is non-NULL only
396 * when the type arg is group/project. Its purpose is to save a
397 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
401 ASSERT(type
== XFS_DQ_GROUP
|| type
== XFS_DQ_PROJ
);
405 * No need to take dqlock to look at the id.
407 * The ID can't change until it gets reclaimed, and it won't
408 * be reclaimed as long as we have a ref from inode and we
411 dqp
= udqhint
->q_gdquot
;
412 if (dqp
&& be32_to_cpu(dqp
->q_core
.d_id
) == id
) {
413 ASSERT(*IO_idqpp
== NULL
);
415 *IO_idqpp
= xfs_qm_dqhold(dqp
);
416 xfs_dqunlock(udqhint
);
421 * We can't hold a dquot lock when we call the dqget code.
422 * We'll deadlock in no time, because of (not conforming to)
423 * lock ordering - the inodelock comes before any dquot lock,
424 * and we may drop and reacquire the ilock in xfs_qm_dqget().
426 xfs_dqunlock(udqhint
);
430 * Find the dquot from somewhere. This bumps the
431 * reference count of dquot and returns it locked.
432 * This can return ENOENT if dquot didn't exist on
433 * disk and we didn't ask it to allocate;
434 * ESRCH if quotas got turned off suddenly.
436 error
= xfs_qm_dqget(ip
->i_mount
, ip
, id
, type
,
437 doalloc
| XFS_QMOPT_DOWARN
, &dqp
);
441 trace_xfs_dqattach_get(dqp
);
444 * dqget may have dropped and re-acquired the ilock, but it guarantees
445 * that the dquot returned is the one that should go in the inode.
454 * Given a udquot and gdquot, attach a ptr to the group dquot in the
455 * udquot as a hint for future lookups.
458 xfs_qm_dqattach_grouphint(
471 udq
->q_gdquot
= NULL
;
475 udq
->q_gdquot
= xfs_qm_dqhold(gdq
);
481 xfs_qm_need_dqattach(
482 struct xfs_inode
*ip
)
484 struct xfs_mount
*mp
= ip
->i_mount
;
486 if (!XFS_IS_QUOTA_RUNNING(mp
))
488 if (!XFS_IS_QUOTA_ON(mp
))
490 if (!XFS_NOT_DQATTACHED(mp
, ip
))
492 if (xfs_is_quota_inode(&mp
->m_sb
, ip
->i_ino
))
498 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
500 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
501 * Inode may get unlocked and relocked in here, and the caller must deal with
505 xfs_qm_dqattach_locked(
509 xfs_mount_t
*mp
= ip
->i_mount
;
513 if (!xfs_qm_need_dqattach(ip
))
516 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
518 if (XFS_IS_UQUOTA_ON(mp
)) {
519 error
= xfs_qm_dqattach_one(ip
, ip
->i_d
.di_uid
, XFS_DQ_USER
,
520 flags
& XFS_QMOPT_DQALLOC
,
521 NULL
, &ip
->i_udquot
);
527 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
528 if (XFS_IS_OQUOTA_ON(mp
)) {
529 error
= XFS_IS_GQUOTA_ON(mp
) ?
530 xfs_qm_dqattach_one(ip
, ip
->i_d
.di_gid
, XFS_DQ_GROUP
,
531 flags
& XFS_QMOPT_DQALLOC
,
532 ip
->i_udquot
, &ip
->i_gdquot
) :
533 xfs_qm_dqattach_one(ip
, xfs_get_projid(ip
), XFS_DQ_PROJ
,
534 flags
& XFS_QMOPT_DQALLOC
,
535 ip
->i_udquot
, &ip
->i_gdquot
);
537 * Don't worry about the udquot that we may have
538 * attached above. It'll get detached, if not already.
546 * Attach this group quota to the user quota as a hint.
547 * This WON'T, in general, result in a thrash.
550 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
551 ASSERT(ip
->i_udquot
);
552 ASSERT(ip
->i_gdquot
);
555 * We do not have i_udquot locked at this point, but this check
556 * is OK since we don't depend on the i_gdquot to be accurate
557 * 100% all the time. It is just a hint, and this will
558 * succeed in general.
560 if (ip
->i_udquot
->q_gdquot
!= ip
->i_gdquot
)
561 xfs_qm_dqattach_grouphint(ip
->i_udquot
, ip
->i_gdquot
);
567 if (XFS_IS_UQUOTA_ON(mp
))
568 ASSERT(ip
->i_udquot
);
569 if (XFS_IS_OQUOTA_ON(mp
))
570 ASSERT(ip
->i_gdquot
);
572 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
579 struct xfs_inode
*ip
,
584 if (!xfs_qm_need_dqattach(ip
))
587 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
588 error
= xfs_qm_dqattach_locked(ip
, flags
);
589 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
595 * Release dquots (and their references) if any.
596 * The inode should be locked EXCL except when this's called by
603 if (!(ip
->i_udquot
|| ip
->i_gdquot
))
606 trace_xfs_dquot_dqdetach(ip
);
608 ASSERT(!xfs_is_quota_inode(&ip
->i_mount
->m_sb
, ip
->i_ino
));
610 xfs_qm_dqrele(ip
->i_udquot
);
614 xfs_qm_dqrele(ip
->i_gdquot
);
620 xfs_qm_calc_dquots_per_chunk(
621 struct xfs_mount
*mp
,
622 unsigned int nbblks
) /* basic block units */
624 unsigned int ndquots
;
627 ndquots
= BBTOB(nbblks
);
628 do_div(ndquots
, sizeof(xfs_dqblk_t
));
634 * This initializes all the quota information that's kept in the
638 xfs_qm_init_quotainfo(
641 xfs_quotainfo_t
*qinf
;
645 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
647 qinf
= mp
->m_quotainfo
= kmem_zalloc(sizeof(xfs_quotainfo_t
), KM_SLEEP
);
650 * See if quotainodes are setup, and if not, allocate them,
651 * and change the superblock accordingly.
653 if ((error
= xfs_qm_init_quotainos(mp
))) {
655 mp
->m_quotainfo
= NULL
;
659 INIT_RADIX_TREE(&qinf
->qi_uquota_tree
, GFP_NOFS
);
660 INIT_RADIX_TREE(&qinf
->qi_gquota_tree
, GFP_NOFS
);
661 mutex_init(&qinf
->qi_tree_lock
);
663 INIT_LIST_HEAD(&qinf
->qi_lru_list
);
664 qinf
->qi_lru_count
= 0;
665 mutex_init(&qinf
->qi_lru_lock
);
667 /* mutex used to serialize quotaoffs */
668 mutex_init(&qinf
->qi_quotaofflock
);
670 /* Precalc some constants */
671 qinf
->qi_dqchunklen
= XFS_FSB_TO_BB(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
672 qinf
->qi_dqperchunk
= xfs_qm_calc_dquots_per_chunk(mp
,
673 qinf
->qi_dqchunklen
);
675 mp
->m_qflags
|= (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_CHKD
);
678 * We try to get the limits from the superuser's limits fields.
679 * This is quite hacky, but it is standard quota practice.
681 * We look at the USR dquot with id == 0 first, but if user quotas
682 * are not enabled we goto the GRP dquot with id == 0.
683 * We don't really care to keep separate default limits for user
684 * and group quotas, at least not at this point.
686 * Since we may not have done a quotacheck by this point, just read
687 * the dquot without attaching it to any hashtables or lists.
689 error
= xfs_qm_dqread(mp
, 0,
690 XFS_IS_UQUOTA_RUNNING(mp
) ? XFS_DQ_USER
:
691 (XFS_IS_GQUOTA_RUNNING(mp
) ? XFS_DQ_GROUP
:
693 XFS_QMOPT_DOWARN
, &dqp
);
695 xfs_disk_dquot_t
*ddqp
= &dqp
->q_core
;
698 * The warnings and timers set the grace period given to
699 * a user or group before he or she can not perform any
700 * more writing. If it is zero, a default is used.
702 qinf
->qi_btimelimit
= ddqp
->d_btimer
?
703 be32_to_cpu(ddqp
->d_btimer
) : XFS_QM_BTIMELIMIT
;
704 qinf
->qi_itimelimit
= ddqp
->d_itimer
?
705 be32_to_cpu(ddqp
->d_itimer
) : XFS_QM_ITIMELIMIT
;
706 qinf
->qi_rtbtimelimit
= ddqp
->d_rtbtimer
?
707 be32_to_cpu(ddqp
->d_rtbtimer
) : XFS_QM_RTBTIMELIMIT
;
708 qinf
->qi_bwarnlimit
= ddqp
->d_bwarns
?
709 be16_to_cpu(ddqp
->d_bwarns
) : XFS_QM_BWARNLIMIT
;
710 qinf
->qi_iwarnlimit
= ddqp
->d_iwarns
?
711 be16_to_cpu(ddqp
->d_iwarns
) : XFS_QM_IWARNLIMIT
;
712 qinf
->qi_rtbwarnlimit
= ddqp
->d_rtbwarns
?
713 be16_to_cpu(ddqp
->d_rtbwarns
) : XFS_QM_RTBWARNLIMIT
;
714 qinf
->qi_bhardlimit
= be64_to_cpu(ddqp
->d_blk_hardlimit
);
715 qinf
->qi_bsoftlimit
= be64_to_cpu(ddqp
->d_blk_softlimit
);
716 qinf
->qi_ihardlimit
= be64_to_cpu(ddqp
->d_ino_hardlimit
);
717 qinf
->qi_isoftlimit
= be64_to_cpu(ddqp
->d_ino_softlimit
);
718 qinf
->qi_rtbhardlimit
= be64_to_cpu(ddqp
->d_rtb_hardlimit
);
719 qinf
->qi_rtbsoftlimit
= be64_to_cpu(ddqp
->d_rtb_softlimit
);
721 xfs_qm_dqdestroy(dqp
);
723 qinf
->qi_btimelimit
= XFS_QM_BTIMELIMIT
;
724 qinf
->qi_itimelimit
= XFS_QM_ITIMELIMIT
;
725 qinf
->qi_rtbtimelimit
= XFS_QM_RTBTIMELIMIT
;
726 qinf
->qi_bwarnlimit
= XFS_QM_BWARNLIMIT
;
727 qinf
->qi_iwarnlimit
= XFS_QM_IWARNLIMIT
;
728 qinf
->qi_rtbwarnlimit
= XFS_QM_RTBWARNLIMIT
;
731 qinf
->qi_shrinker
.shrink
= xfs_qm_shake
;
732 qinf
->qi_shrinker
.seeks
= DEFAULT_SEEKS
;
733 register_shrinker(&qinf
->qi_shrinker
);
739 * Gets called when unmounting a filesystem or when all quotas get
741 * This purges the quota inodes, destroys locks and frees itself.
744 xfs_qm_destroy_quotainfo(
749 qi
= mp
->m_quotainfo
;
752 unregister_shrinker(&qi
->qi_shrinker
);
754 if (qi
->qi_uquotaip
) {
755 IRELE(qi
->qi_uquotaip
);
756 qi
->qi_uquotaip
= NULL
; /* paranoia */
758 if (qi
->qi_gquotaip
) {
759 IRELE(qi
->qi_gquotaip
);
760 qi
->qi_gquotaip
= NULL
;
762 mutex_destroy(&qi
->qi_quotaofflock
);
764 mp
->m_quotainfo
= NULL
;
768 * Create an inode and return with a reference already taken, but unlocked
769 * This is how we create quota inodes
782 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_QINOCREATE
);
783 if ((error
= xfs_trans_reserve(tp
,
784 XFS_QM_QINOCREATE_SPACE_RES(mp
),
785 XFS_CREATE_LOG_RES(mp
), 0,
786 XFS_TRANS_PERM_LOG_RES
,
787 XFS_CREATE_LOG_COUNT
))) {
788 xfs_trans_cancel(tp
, 0);
792 error
= xfs_dir_ialloc(&tp
, NULL
, S_IFREG
, 1, 0, 0, 1, ip
, &committed
);
794 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
800 * Make the changes in the superblock, and log those too.
801 * sbfields arg may contain fields other than *QUOTINO;
802 * VERSIONNUM for example.
804 spin_lock(&mp
->m_sb_lock
);
805 if (flags
& XFS_QMOPT_SBVERSION
) {
806 ASSERT(!xfs_sb_version_hasquota(&mp
->m_sb
));
807 ASSERT((sbfields
& (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
808 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
)) ==
809 (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
810 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
));
812 xfs_sb_version_addquota(&mp
->m_sb
);
813 mp
->m_sb
.sb_uquotino
= NULLFSINO
;
814 mp
->m_sb
.sb_gquotino
= NULLFSINO
;
816 /* qflags will get updated _after_ quotacheck */
817 mp
->m_sb
.sb_qflags
= 0;
819 if (flags
& XFS_QMOPT_UQUOTA
)
820 mp
->m_sb
.sb_uquotino
= (*ip
)->i_ino
;
822 mp
->m_sb
.sb_gquotino
= (*ip
)->i_ino
;
823 spin_unlock(&mp
->m_sb_lock
);
824 xfs_mod_sb(tp
, sbfields
);
826 if ((error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
))) {
827 xfs_alert(mp
, "%s failed (error %d)!", __func__
, error
);
835 xfs_qm_reset_dqcounts(
841 struct xfs_dqblk
*dqb
;
844 trace_xfs_reset_dqcounts(bp
, _RET_IP_
);
847 * Reset all counters and timers. They'll be
848 * started afresh by xfs_qm_quotacheck.
851 j
= XFS_FSB_TO_B(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
852 do_div(j
, sizeof(xfs_dqblk_t
));
853 ASSERT(mp
->m_quotainfo
->qi_dqperchunk
== j
);
856 for (j
= 0; j
< mp
->m_quotainfo
->qi_dqperchunk
; j
++) {
857 struct xfs_disk_dquot
*ddq
;
859 ddq
= (struct xfs_disk_dquot
*)&dqb
[j
];
862 * Do a sanity check, and if needed, repair the dqblk. Don't
863 * output any warnings because it's perfectly possible to
864 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
866 (void) xfs_qm_dqcheck(mp
, ddq
, id
+j
, type
, XFS_QMOPT_DQREPAIR
,
878 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
879 xfs_update_cksum((char *)&dqb
[j
],
880 sizeof(struct xfs_dqblk
),
888 struct xfs_mount
*mp
,
891 xfs_filblks_t blkcnt
,
893 struct list_head
*buffer_list
)
900 type
= flags
& XFS_QMOPT_UQUOTA
? XFS_DQ_USER
:
901 (flags
& XFS_QMOPT_PQUOTA
? XFS_DQ_PROJ
: XFS_DQ_GROUP
);
905 * Blkcnt arg can be a very big number, and might even be
906 * larger than the log itself. So, we have to break it up into
907 * manageable-sized transactions.
908 * Note that we don't start a permanent transaction here; we might
909 * not be able to get a log reservation for the whole thing up front,
910 * and we don't really care to either, because we just discard
911 * everything if we were to crash in the middle of this loop.
914 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
,
915 XFS_FSB_TO_DADDR(mp
, bno
),
916 mp
->m_quotainfo
->qi_dqchunklen
, 0, &bp
,
920 * CRC and validation errors will return a EFSCORRUPTED here. If
921 * this occurs, re-read without CRC validation so that we can
922 * repair the damage via xfs_qm_reset_dqcounts(). This process
923 * will leave a trace in the log indicating corruption has
926 if (error
== EFSCORRUPTED
) {
927 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
,
928 XFS_FSB_TO_DADDR(mp
, bno
),
929 mp
->m_quotainfo
->qi_dqchunklen
, 0, &bp
,
936 xfs_qm_reset_dqcounts(mp
, bp
, firstid
, type
);
937 xfs_buf_delwri_queue(bp
, buffer_list
);
940 /* goto the next block. */
942 firstid
+= mp
->m_quotainfo
->qi_dqperchunk
;
949 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
950 * caller supplied function for every chunk of dquots that we find.
954 struct xfs_mount
*mp
,
955 struct xfs_inode
*qip
,
957 struct list_head
*buffer_list
)
959 struct xfs_bmbt_irec
*map
;
960 int i
, nmaps
; /* number of map entries */
961 int error
; /* return value */
962 xfs_fileoff_t lblkno
;
963 xfs_filblks_t maxlblkcnt
;
965 xfs_fsblock_t rablkno
;
966 xfs_filblks_t rablkcnt
;
970 * This looks racy, but we can't keep an inode lock across a
971 * trans_reserve. But, this gets called during quotacheck, and that
972 * happens only at mount time which is single threaded.
974 if (qip
->i_d
.di_nblocks
== 0)
977 map
= kmem_alloc(XFS_DQITER_MAP_SIZE
* sizeof(*map
), KM_SLEEP
);
980 maxlblkcnt
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
982 nmaps
= XFS_DQITER_MAP_SIZE
;
984 * We aren't changing the inode itself. Just changing
985 * some of its data. No new blocks are added here, and
986 * the inode is never added to the transaction.
988 xfs_ilock(qip
, XFS_ILOCK_SHARED
);
989 error
= xfs_bmapi_read(qip
, lblkno
, maxlblkcnt
- lblkno
,
991 xfs_iunlock(qip
, XFS_ILOCK_SHARED
);
995 ASSERT(nmaps
<= XFS_DQITER_MAP_SIZE
);
996 for (i
= 0; i
< nmaps
; i
++) {
997 ASSERT(map
[i
].br_startblock
!= DELAYSTARTBLOCK
);
998 ASSERT(map
[i
].br_blockcount
);
1001 lblkno
+= map
[i
].br_blockcount
;
1003 if (map
[i
].br_startblock
== HOLESTARTBLOCK
)
1006 firstid
= (xfs_dqid_t
) map
[i
].br_startoff
*
1007 mp
->m_quotainfo
->qi_dqperchunk
;
1009 * Do a read-ahead on the next extent.
1011 if ((i
+1 < nmaps
) &&
1012 (map
[i
+1].br_startblock
!= HOLESTARTBLOCK
)) {
1013 rablkcnt
= map
[i
+1].br_blockcount
;
1014 rablkno
= map
[i
+1].br_startblock
;
1015 while (rablkcnt
--) {
1016 xfs_buf_readahead(mp
->m_ddev_targp
,
1017 XFS_FSB_TO_DADDR(mp
, rablkno
),
1018 mp
->m_quotainfo
->qi_dqchunklen
,
1024 * Iterate thru all the blks in the extent and
1025 * reset the counters of all the dquots inside them.
1027 error
= xfs_qm_dqiter_bufs(mp
, firstid
,
1028 map
[i
].br_startblock
,
1029 map
[i
].br_blockcount
,
1030 flags
, buffer_list
);
1034 } while (nmaps
> 0);
1042 * Called by dqusage_adjust in doing a quotacheck.
1044 * Given the inode, and a dquot id this updates both the incore dqout as well
1045 * as the buffer copy. This is so that once the quotacheck is done, we can
1046 * just log all the buffers, as opposed to logging numerous updates to
1047 * individual dquots.
1050 xfs_qm_quotacheck_dqadjust(
1051 struct xfs_inode
*ip
,
1057 struct xfs_mount
*mp
= ip
->i_mount
;
1058 struct xfs_dquot
*dqp
;
1061 error
= xfs_qm_dqget(mp
, ip
, id
, type
,
1062 XFS_QMOPT_DQALLOC
| XFS_QMOPT_DOWARN
, &dqp
);
1065 * Shouldn't be able to turn off quotas here.
1067 ASSERT(error
!= ESRCH
);
1068 ASSERT(error
!= ENOENT
);
1072 trace_xfs_dqadjust(dqp
);
1075 * Adjust the inode count and the block count to reflect this inode's
1078 be64_add_cpu(&dqp
->q_core
.d_icount
, 1);
1079 dqp
->q_res_icount
++;
1081 be64_add_cpu(&dqp
->q_core
.d_bcount
, nblks
);
1082 dqp
->q_res_bcount
+= nblks
;
1085 be64_add_cpu(&dqp
->q_core
.d_rtbcount
, rtblks
);
1086 dqp
->q_res_rtbcount
+= rtblks
;
1090 * Set default limits, adjust timers (since we changed usages)
1092 * There are no timers for the default values set in the root dquot.
1094 if (dqp
->q_core
.d_id
) {
1095 xfs_qm_adjust_dqlimits(mp
, dqp
);
1096 xfs_qm_adjust_dqtimers(mp
, &dqp
->q_core
);
1099 dqp
->dq_flags
|= XFS_DQ_DIRTY
;
1107 xfs_qcnt_t
*O_rtblks
)
1109 xfs_filblks_t rtblks
; /* total rt blks */
1110 xfs_extnum_t idx
; /* extent record index */
1111 xfs_ifork_t
*ifp
; /* inode fork pointer */
1112 xfs_extnum_t nextents
; /* number of extent entries */
1115 ASSERT(XFS_IS_REALTIME_INODE(ip
));
1116 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1117 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1118 if ((error
= xfs_iread_extents(NULL
, ip
, XFS_DATA_FORK
)))
1122 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
1123 for (idx
= 0; idx
< nextents
; idx
++)
1124 rtblks
+= xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp
, idx
));
1125 *O_rtblks
= (xfs_qcnt_t
)rtblks
;
1130 * callback routine supplied to bulkstat(). Given an inumber, find its
1131 * dquots and update them to account for resources taken by that inode.
1135 xfs_qm_dqusage_adjust(
1136 xfs_mount_t
*mp
, /* mount point for filesystem */
1137 xfs_ino_t ino
, /* inode number to get data for */
1138 void __user
*buffer
, /* not used */
1139 int ubsize
, /* not used */
1140 int *ubused
, /* not used */
1141 int *res
) /* result code value */
1144 xfs_qcnt_t nblks
, rtblks
= 0;
1147 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1150 * rootino must have its resources accounted for, not so with the quota
1153 if (xfs_is_quota_inode(&mp
->m_sb
, ino
)) {
1154 *res
= BULKSTAT_RV_NOTHING
;
1155 return XFS_ERROR(EINVAL
);
1159 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1160 * interface expects the inode to be exclusively locked because that's
1161 * the case in all other instances. It's OK that we do this because
1162 * quotacheck is done only at mount time.
1164 error
= xfs_iget(mp
, NULL
, ino
, 0, XFS_ILOCK_EXCL
, &ip
);
1166 *res
= BULKSTAT_RV_NOTHING
;
1170 ASSERT(ip
->i_delayed_blks
== 0);
1172 if (XFS_IS_REALTIME_INODE(ip
)) {
1174 * Walk thru the extent list and count the realtime blocks.
1176 error
= xfs_qm_get_rtblks(ip
, &rtblks
);
1181 nblks
= (xfs_qcnt_t
)ip
->i_d
.di_nblocks
- rtblks
;
1184 * Add the (disk blocks and inode) resources occupied by this
1185 * inode to its dquots. We do this adjustment in the incore dquot,
1186 * and also copy the changes to its buffer.
1187 * We don't care about putting these changes in a transaction
1188 * envelope because if we crash in the middle of a 'quotacheck'
1189 * we have to start from the beginning anyway.
1190 * Once we're done, we'll log all the dquot bufs.
1192 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1193 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1195 if (XFS_IS_UQUOTA_ON(mp
)) {
1196 error
= xfs_qm_quotacheck_dqadjust(ip
, ip
->i_d
.di_uid
,
1197 XFS_DQ_USER
, nblks
, rtblks
);
1202 if (XFS_IS_GQUOTA_ON(mp
)) {
1203 error
= xfs_qm_quotacheck_dqadjust(ip
, ip
->i_d
.di_gid
,
1204 XFS_DQ_GROUP
, nblks
, rtblks
);
1209 if (XFS_IS_PQUOTA_ON(mp
)) {
1210 error
= xfs_qm_quotacheck_dqadjust(ip
, xfs_get_projid(ip
),
1211 XFS_DQ_PROJ
, nblks
, rtblks
);
1216 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1218 *res
= BULKSTAT_RV_DIDONE
;
1222 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1224 *res
= BULKSTAT_RV_GIVEUP
;
1230 struct xfs_dquot
*dqp
,
1233 struct list_head
*buffer_list
= data
;
1234 struct xfs_buf
*bp
= NULL
;
1238 if (dqp
->dq_flags
& XFS_DQ_FREEING
)
1240 if (!XFS_DQ_IS_DIRTY(dqp
))
1244 error
= xfs_qm_dqflush(dqp
, &bp
);
1248 xfs_buf_delwri_queue(bp
, buffer_list
);
1256 * Walk thru all the filesystem inodes and construct a consistent view
1257 * of the disk quota world. If the quotacheck fails, disable quotas.
1263 int done
, count
, error
, error2
;
1267 LIST_HEAD (buffer_list
);
1268 struct xfs_inode
*uip
= mp
->m_quotainfo
->qi_uquotaip
;
1269 struct xfs_inode
*gip
= mp
->m_quotainfo
->qi_gquotaip
;
1277 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1279 xfs_notice(mp
, "Quotacheck needed: Please wait.");
1282 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1283 * their counters to zero. We need a clean slate.
1284 * We don't log our changes till later.
1287 error
= xfs_qm_dqiterate(mp
, uip
, XFS_QMOPT_UQUOTA
,
1291 flags
|= XFS_UQUOTA_CHKD
;
1295 error
= xfs_qm_dqiterate(mp
, gip
, XFS_IS_GQUOTA_ON(mp
) ?
1296 XFS_QMOPT_GQUOTA
: XFS_QMOPT_PQUOTA
,
1300 flags
|= XFS_OQUOTA_CHKD
;
1305 * Iterate thru all the inodes in the file system,
1306 * adjusting the corresponding dquot counters in core.
1308 error
= xfs_bulkstat(mp
, &lastino
, &count
,
1309 xfs_qm_dqusage_adjust
,
1310 structsz
, NULL
, &done
);
1317 * We've made all the changes that we need to make incore. Flush them
1318 * down to disk buffers if everything was updated successfully.
1320 if (XFS_IS_UQUOTA_ON(mp
)) {
1321 error
= xfs_qm_dquot_walk(mp
, XFS_DQ_USER
, xfs_qm_flush_one
,
1324 if (XFS_IS_GQUOTA_ON(mp
)) {
1325 error2
= xfs_qm_dquot_walk(mp
, XFS_DQ_GROUP
, xfs_qm_flush_one
,
1330 if (XFS_IS_PQUOTA_ON(mp
)) {
1331 error2
= xfs_qm_dquot_walk(mp
, XFS_DQ_PROJ
, xfs_qm_flush_one
,
1337 error2
= xfs_buf_delwri_submit(&buffer_list
);
1342 * We can get this error if we couldn't do a dquot allocation inside
1343 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1344 * dirty dquots that might be cached, we just want to get rid of them
1345 * and turn quotaoff. The dquots won't be attached to any of the inodes
1346 * at this point (because we intentionally didn't in dqget_noattach).
1349 xfs_qm_dqpurge_all(mp
, XFS_QMOPT_QUOTALL
);
1354 * If one type of quotas is off, then it will lose its
1355 * quotachecked status, since we won't be doing accounting for
1356 * that type anymore.
1358 mp
->m_qflags
&= ~XFS_ALL_QUOTA_CHKD
;
1359 mp
->m_qflags
|= flags
;
1362 while (!list_empty(&buffer_list
)) {
1363 struct xfs_buf
*bp
=
1364 list_first_entry(&buffer_list
, struct xfs_buf
, b_list
);
1365 list_del_init(&bp
->b_list
);
1371 "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1374 * We must turn off quotas.
1376 ASSERT(mp
->m_quotainfo
!= NULL
);
1377 xfs_qm_destroy_quotainfo(mp
);
1378 if (xfs_mount_reset_sbqflags(mp
)) {
1380 "Quotacheck: Failed to reset quota flags.");
1383 xfs_notice(mp
, "Quotacheck: Done.");
1388 * This is called after the superblock has been read in and we're ready to
1389 * iget the quota inodes.
1392 xfs_qm_init_quotainos(
1395 struct xfs_inode
*uip
= NULL
;
1396 struct xfs_inode
*gip
= NULL
;
1398 __int64_t sbflags
= 0;
1401 ASSERT(mp
->m_quotainfo
);
1404 * Get the uquota and gquota inodes
1406 if (xfs_sb_version_hasquota(&mp
->m_sb
)) {
1407 if (XFS_IS_UQUOTA_ON(mp
) &&
1408 mp
->m_sb
.sb_uquotino
!= NULLFSINO
) {
1409 ASSERT(mp
->m_sb
.sb_uquotino
> 0);
1410 error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_uquotino
,
1413 return XFS_ERROR(error
);
1415 if (XFS_IS_OQUOTA_ON(mp
) &&
1416 mp
->m_sb
.sb_gquotino
!= NULLFSINO
) {
1417 ASSERT(mp
->m_sb
.sb_gquotino
> 0);
1418 error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_gquotino
,
1424 flags
|= XFS_QMOPT_SBVERSION
;
1425 sbflags
|= (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1426 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
);
1430 * Create the two inodes, if they don't exist already. The changes
1431 * made above will get added to a transaction and logged in one of
1432 * the qino_alloc calls below. If the device is readonly,
1433 * temporarily switch to read-write to do this.
1435 if (XFS_IS_UQUOTA_ON(mp
) && uip
== NULL
) {
1436 error
= xfs_qm_qino_alloc(mp
, &uip
,
1437 sbflags
| XFS_SB_UQUOTINO
,
1438 flags
| XFS_QMOPT_UQUOTA
);
1442 flags
&= ~XFS_QMOPT_SBVERSION
;
1444 if (XFS_IS_OQUOTA_ON(mp
) && gip
== NULL
) {
1445 flags
|= (XFS_IS_GQUOTA_ON(mp
) ?
1446 XFS_QMOPT_GQUOTA
: XFS_QMOPT_PQUOTA
);
1447 error
= xfs_qm_qino_alloc(mp
, &gip
,
1448 sbflags
| XFS_SB_GQUOTINO
, flags
);
1453 mp
->m_quotainfo
->qi_uquotaip
= uip
;
1454 mp
->m_quotainfo
->qi_gquotaip
= gip
;
1463 return XFS_ERROR(error
);
1468 struct xfs_dquot
*dqp
)
1470 struct xfs_mount
*mp
= dqp
->q_mount
;
1471 struct xfs_quotainfo
*qi
= mp
->m_quotainfo
;
1473 mutex_lock(&qi
->qi_tree_lock
);
1474 radix_tree_delete(xfs_dquot_tree(qi
, dqp
->q_core
.d_flags
),
1475 be32_to_cpu(dqp
->q_core
.d_id
));
1478 mutex_unlock(&qi
->qi_tree_lock
);
1480 xfs_qm_dqdestroy(dqp
);
1484 xfs_qm_dqreclaim_one(
1485 struct xfs_dquot
*dqp
,
1486 struct list_head
*buffer_list
,
1487 struct list_head
*dispose_list
)
1489 struct xfs_mount
*mp
= dqp
->q_mount
;
1490 struct xfs_quotainfo
*qi
= mp
->m_quotainfo
;
1493 if (!xfs_dqlock_nowait(dqp
))
1497 * This dquot has acquired a reference in the meantime remove it from
1498 * the freelist and try again.
1503 trace_xfs_dqreclaim_want(dqp
);
1504 XFS_STATS_INC(xs_qm_dqwants
);
1506 list_del_init(&dqp
->q_lru
);
1508 XFS_STATS_DEC(xs_qm_dquot_unused
);
1513 * Try to grab the flush lock. If this dquot is in the process of
1514 * getting flushed to disk, we don't want to reclaim it.
1516 if (!xfs_dqflock_nowait(dqp
))
1517 goto out_unlock_move_tail
;
1519 if (XFS_DQ_IS_DIRTY(dqp
)) {
1520 struct xfs_buf
*bp
= NULL
;
1522 trace_xfs_dqreclaim_dirty(dqp
);
1524 error
= xfs_qm_dqflush(dqp
, &bp
);
1526 xfs_warn(mp
, "%s: dquot %p flush failed",
1528 goto out_unlock_move_tail
;
1531 xfs_buf_delwri_queue(bp
, buffer_list
);
1534 * Give the dquot another try on the freelist, as the
1535 * flushing will take some time.
1537 goto out_unlock_move_tail
;
1542 * Prevent lookups now that we are past the point of no return.
1544 dqp
->dq_flags
|= XFS_DQ_FREEING
;
1547 ASSERT(dqp
->q_nrefs
== 0);
1548 list_move_tail(&dqp
->q_lru
, dispose_list
);
1550 XFS_STATS_DEC(xs_qm_dquot_unused
);
1552 trace_xfs_dqreclaim_done(dqp
);
1553 XFS_STATS_INC(xs_qm_dqreclaims
);
1557 * Move the dquot to the tail of the list so that we don't spin on it.
1559 out_unlock_move_tail
:
1562 list_move_tail(&dqp
->q_lru
, &qi
->qi_lru_list
);
1563 trace_xfs_dqreclaim_busy(dqp
);
1564 XFS_STATS_INC(xs_qm_dqreclaim_misses
);
1569 struct shrinker
*shrink
,
1570 struct shrink_control
*sc
)
1572 struct xfs_quotainfo
*qi
=
1573 container_of(shrink
, struct xfs_quotainfo
, qi_shrinker
);
1574 int nr_to_scan
= sc
->nr_to_scan
;
1575 LIST_HEAD (buffer_list
);
1576 LIST_HEAD (dispose_list
);
1577 struct xfs_dquot
*dqp
;
1580 if ((sc
->gfp_mask
& (__GFP_FS
|__GFP_WAIT
)) != (__GFP_FS
|__GFP_WAIT
))
1585 mutex_lock(&qi
->qi_lru_lock
);
1586 while (!list_empty(&qi
->qi_lru_list
)) {
1587 if (nr_to_scan
-- <= 0)
1589 dqp
= list_first_entry(&qi
->qi_lru_list
, struct xfs_dquot
,
1591 xfs_qm_dqreclaim_one(dqp
, &buffer_list
, &dispose_list
);
1593 mutex_unlock(&qi
->qi_lru_lock
);
1595 error
= xfs_buf_delwri_submit(&buffer_list
);
1597 xfs_warn(NULL
, "%s: dquot reclaim failed", __func__
);
1599 while (!list_empty(&dispose_list
)) {
1600 dqp
= list_first_entry(&dispose_list
, struct xfs_dquot
, q_lru
);
1601 list_del_init(&dqp
->q_lru
);
1602 xfs_qm_dqfree_one(dqp
);
1606 return (qi
->qi_lru_count
/ 100) * sysctl_vfs_cache_pressure
;
1610 * Start a transaction and write the incore superblock changes to
1611 * disk. flags parameter indicates which fields have changed.
1614 xfs_qm_write_sb_changes(
1621 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
1622 error
= xfs_trans_reserve(tp
, 0, XFS_QM_SBCHANGE_LOG_RES(mp
),
1623 0, 0, XFS_DEFAULT_LOG_COUNT
);
1625 xfs_trans_cancel(tp
, 0);
1629 xfs_mod_sb(tp
, flags
);
1630 error
= xfs_trans_commit(tp
, 0);
1636 /* --------------- utility functions for vnodeops ---------------- */
1640 * Given an inode, a uid, gid and prid make sure that we have
1641 * allocated relevant dquot(s) on disk, and that we won't exceed inode
1642 * quotas by creating this file.
1643 * This also attaches dquot(s) to the given inode after locking it,
1644 * and returns the dquots corresponding to the uid and/or gid.
1646 * in : inode (unlocked)
1647 * out : udquot, gdquot with references taken and unlocked
1651 struct xfs_inode
*ip
,
1656 struct xfs_dquot
**O_udqpp
,
1657 struct xfs_dquot
**O_gdqpp
)
1659 struct xfs_mount
*mp
= ip
->i_mount
;
1660 struct xfs_dquot
*uq
= NULL
;
1661 struct xfs_dquot
*gq
= NULL
;
1665 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
1668 lockflags
= XFS_ILOCK_EXCL
;
1669 xfs_ilock(ip
, lockflags
);
1671 if ((flags
& XFS_QMOPT_INHERIT
) && XFS_INHERIT_GID(ip
))
1672 gid
= ip
->i_d
.di_gid
;
1675 * Attach the dquot(s) to this inode, doing a dquot allocation
1676 * if necessary. The dquot(s) will not be locked.
1678 if (XFS_NOT_DQATTACHED(mp
, ip
)) {
1679 error
= xfs_qm_dqattach_locked(ip
, XFS_QMOPT_DQALLOC
);
1681 xfs_iunlock(ip
, lockflags
);
1686 if ((flags
& XFS_QMOPT_UQUOTA
) && XFS_IS_UQUOTA_ON(mp
)) {
1687 if (ip
->i_d
.di_uid
!= uid
) {
1689 * What we need is the dquot that has this uid, and
1690 * if we send the inode to dqget, the uid of the inode
1691 * takes priority over what's sent in the uid argument.
1692 * We must unlock inode here before calling dqget if
1693 * we're not sending the inode, because otherwise
1694 * we'll deadlock by doing trans_reserve while
1697 xfs_iunlock(ip
, lockflags
);
1698 error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
) uid
,
1704 ASSERT(error
!= ENOENT
);
1708 * Get the ilock in the right order.
1711 lockflags
= XFS_ILOCK_SHARED
;
1712 xfs_ilock(ip
, lockflags
);
1715 * Take an extra reference, because we'll return
1718 ASSERT(ip
->i_udquot
);
1719 uq
= xfs_qm_dqhold(ip
->i_udquot
);
1722 if ((flags
& XFS_QMOPT_GQUOTA
) && XFS_IS_GQUOTA_ON(mp
)) {
1723 if (ip
->i_d
.di_gid
!= gid
) {
1724 xfs_iunlock(ip
, lockflags
);
1725 error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)gid
,
1731 ASSERT(error
!= ENOENT
);
1735 lockflags
= XFS_ILOCK_SHARED
;
1736 xfs_ilock(ip
, lockflags
);
1738 ASSERT(ip
->i_gdquot
);
1739 gq
= xfs_qm_dqhold(ip
->i_gdquot
);
1741 } else if ((flags
& XFS_QMOPT_PQUOTA
) && XFS_IS_PQUOTA_ON(mp
)) {
1742 if (xfs_get_projid(ip
) != prid
) {
1743 xfs_iunlock(ip
, lockflags
);
1744 error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)prid
,
1750 ASSERT(error
!= ENOENT
);
1754 lockflags
= XFS_ILOCK_SHARED
;
1755 xfs_ilock(ip
, lockflags
);
1757 ASSERT(ip
->i_gdquot
);
1758 gq
= xfs_qm_dqhold(ip
->i_gdquot
);
1762 trace_xfs_dquot_dqalloc(ip
);
1764 xfs_iunlock(ip
, lockflags
);
1782 * Actually transfer ownership, and do dquot modifications.
1783 * These were already reserved.
1789 xfs_dquot_t
**IO_olddq
,
1792 xfs_dquot_t
*prevdq
;
1793 uint bfield
= XFS_IS_REALTIME_INODE(ip
) ?
1794 XFS_TRANS_DQ_RTBCOUNT
: XFS_TRANS_DQ_BCOUNT
;
1797 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1798 ASSERT(XFS_IS_QUOTA_RUNNING(ip
->i_mount
));
1803 ASSERT(prevdq
!= newdq
);
1805 xfs_trans_mod_dquot(tp
, prevdq
, bfield
, -(ip
->i_d
.di_nblocks
));
1806 xfs_trans_mod_dquot(tp
, prevdq
, XFS_TRANS_DQ_ICOUNT
, -1);
1808 /* the sparkling new dquot */
1809 xfs_trans_mod_dquot(tp
, newdq
, bfield
, ip
->i_d
.di_nblocks
);
1810 xfs_trans_mod_dquot(tp
, newdq
, XFS_TRANS_DQ_ICOUNT
, 1);
1813 * Take an extra reference, because the inode is going to keep
1814 * this dquot pointer even after the trans_commit.
1816 *IO_olddq
= xfs_qm_dqhold(newdq
);
1822 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1825 xfs_qm_vop_chown_reserve(
1826 struct xfs_trans
*tp
,
1827 struct xfs_inode
*ip
,
1828 struct xfs_dquot
*udqp
,
1829 struct xfs_dquot
*gdqp
,
1832 struct xfs_mount
*mp
= ip
->i_mount
;
1833 uint delblks
, blkflags
, prjflags
= 0;
1834 struct xfs_dquot
*udq_unres
= NULL
;
1835 struct xfs_dquot
*gdq_unres
= NULL
;
1836 struct xfs_dquot
*udq_delblks
= NULL
;
1837 struct xfs_dquot
*gdq_delblks
= NULL
;
1841 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
1842 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1844 delblks
= ip
->i_delayed_blks
;
1845 blkflags
= XFS_IS_REALTIME_INODE(ip
) ?
1846 XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
;
1848 if (XFS_IS_UQUOTA_ON(mp
) && udqp
&&
1849 ip
->i_d
.di_uid
!= (uid_t
)be32_to_cpu(udqp
->q_core
.d_id
)) {
1852 * If there are delayed allocation blocks, then we have to
1853 * unreserve those from the old dquot, and add them to the
1857 ASSERT(ip
->i_udquot
);
1858 udq_unres
= ip
->i_udquot
;
1861 if (XFS_IS_OQUOTA_ON(ip
->i_mount
) && gdqp
) {
1862 if (XFS_IS_PQUOTA_ON(ip
->i_mount
) &&
1863 xfs_get_projid(ip
) != be32_to_cpu(gdqp
->q_core
.d_id
))
1864 prjflags
= XFS_QMOPT_ENOSPC
;
1867 (XFS_IS_GQUOTA_ON(ip
->i_mount
) &&
1868 ip
->i_d
.di_gid
!= be32_to_cpu(gdqp
->q_core
.d_id
))) {
1871 ASSERT(ip
->i_gdquot
);
1872 gdq_unres
= ip
->i_gdquot
;
1877 error
= xfs_trans_reserve_quota_bydquots(tp
, ip
->i_mount
,
1878 udq_delblks
, gdq_delblks
, ip
->i_d
.di_nblocks
, 1,
1879 flags
| blkflags
| prjflags
);
1884 * Do the delayed blks reservations/unreservations now. Since, these
1885 * are done without the help of a transaction, if a reservation fails
1886 * its previous reservations won't be automatically undone by trans
1887 * code. So, we have to do it manually here.
1891 * Do the reservations first. Unreservation can't fail.
1893 ASSERT(udq_delblks
|| gdq_delblks
);
1894 ASSERT(udq_unres
|| gdq_unres
);
1895 error
= xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
1896 udq_delblks
, gdq_delblks
, (xfs_qcnt_t
)delblks
, 0,
1897 flags
| blkflags
| prjflags
);
1900 xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
1901 udq_unres
, gdq_unres
, -((xfs_qcnt_t
)delblks
), 0,
1909 xfs_qm_vop_rename_dqattach(
1910 struct xfs_inode
**i_tab
)
1912 struct xfs_mount
*mp
= i_tab
[0]->i_mount
;
1915 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
1918 for (i
= 0; (i
< 4 && i_tab
[i
]); i
++) {
1919 struct xfs_inode
*ip
= i_tab
[i
];
1923 * Watch out for duplicate entries in the table.
1925 if (i
== 0 || ip
!= i_tab
[i
-1]) {
1926 if (XFS_NOT_DQATTACHED(mp
, ip
)) {
1927 error
= xfs_qm_dqattach(ip
, 0);
1937 xfs_qm_vop_create_dqattach(
1938 struct xfs_trans
*tp
,
1939 struct xfs_inode
*ip
,
1940 struct xfs_dquot
*udqp
,
1941 struct xfs_dquot
*gdqp
)
1943 struct xfs_mount
*mp
= tp
->t_mountp
;
1945 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
1948 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1949 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1952 ASSERT(ip
->i_udquot
== NULL
);
1953 ASSERT(XFS_IS_UQUOTA_ON(mp
));
1954 ASSERT(ip
->i_d
.di_uid
== be32_to_cpu(udqp
->q_core
.d_id
));
1956 ip
->i_udquot
= xfs_qm_dqhold(udqp
);
1957 xfs_trans_mod_dquot(tp
, udqp
, XFS_TRANS_DQ_ICOUNT
, 1);
1960 ASSERT(ip
->i_gdquot
== NULL
);
1961 ASSERT(XFS_IS_OQUOTA_ON(mp
));
1962 ASSERT((XFS_IS_GQUOTA_ON(mp
) ?
1963 ip
->i_d
.di_gid
: xfs_get_projid(ip
)) ==
1964 be32_to_cpu(gdqp
->q_core
.d_id
));
1966 ip
->i_gdquot
= xfs_qm_dqhold(gdqp
);
1967 xfs_trans_mod_dquot(tp
, gdqp
, XFS_TRANS_DQ_ICOUNT
, 1);