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
23 #include "xfs_trans.h"
27 #include "xfs_alloc.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_quota.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_btree.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_itable.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
47 #include "xfs_buf_item.h"
48 #include "xfs_trans_space.h"
49 #include "xfs_utils.h"
53 * The global quota manager. There is only one of these for the entire
54 * system, _not_ one per file system. XQM keeps track of the overall
55 * quota functionality, including maintaining the freelist and hash
58 struct mutex xfs_Gqm_lock
;
59 struct xfs_qm
*xfs_Gqm
;
62 kmem_zone_t
*qm_dqzone
;
63 kmem_zone_t
*qm_dqtrxzone
;
65 static cred_t xfs_zerocr
;
67 STATIC
void xfs_qm_list_init(xfs_dqlist_t
*, char *, int);
68 STATIC
void xfs_qm_list_destroy(xfs_dqlist_t
*);
70 STATIC
void xfs_qm_freelist_init(xfs_frlist_t
*);
71 STATIC
void xfs_qm_freelist_destroy(xfs_frlist_t
*);
73 STATIC
int xfs_qm_init_quotainos(xfs_mount_t
*);
74 STATIC
int xfs_qm_init_quotainfo(xfs_mount_t
*);
75 STATIC
int xfs_qm_shake(int, gfp_t
);
77 static struct shrinker xfs_qm_shaker
= {
78 .shrink
= xfs_qm_shake
,
79 .seeks
= DEFAULT_SEEKS
,
83 extern struct mutex qcheck_lock
;
87 #define XQM_LIST_PRINT(l, NXT, title) \
89 xfs_dquot_t *dqp; int i = 0; \
90 cmn_err(CE_DEBUG, "%s (#%d)", title, (int) (l)->qh_nelems); \
91 for (dqp = (l)->qh_next; dqp != NULL; dqp = dqp->NXT) { \
92 cmn_err(CE_DEBUG, " %d. \"%d (%s)\" " \
93 "bcnt = %d, icnt = %d, refs = %d", \
94 ++i, (int) be32_to_cpu(dqp->q_core.d_id), \
95 DQFLAGTO_TYPESTR(dqp), \
96 (int) be64_to_cpu(dqp->q_core.d_bcount), \
97 (int) be64_to_cpu(dqp->q_core.d_icount), \
98 (int) dqp->q_nrefs); } \
101 #define XQM_LIST_PRINT(l, NXT, title) do { } while (0)
105 * Initialize the XQM structure.
106 * Note that there is not one quota manager per file system.
108 STATIC
struct xfs_qm
*
111 xfs_dqhash_t
*udqhash
, *gdqhash
;
117 * Initialize the dquot hash tables.
119 udqhash
= kmem_zalloc_greedy(&hsize
,
120 XFS_QM_HASHSIZE_LOW
* sizeof(xfs_dqhash_t
),
121 XFS_QM_HASHSIZE_HIGH
* sizeof(xfs_dqhash_t
),
122 KM_SLEEP
| KM_MAYFAIL
| KM_LARGE
);
123 gdqhash
= kmem_zalloc(hsize
, KM_SLEEP
| KM_LARGE
);
124 hsize
/= sizeof(xfs_dqhash_t
);
127 xqm
= kmem_zalloc(sizeof(xfs_qm_t
), KM_SLEEP
);
128 xqm
->qm_dqhashmask
= hsize
- 1;
129 xqm
->qm_usr_dqhtable
= udqhash
;
130 xqm
->qm_grp_dqhtable
= gdqhash
;
131 ASSERT(xqm
->qm_usr_dqhtable
!= NULL
);
132 ASSERT(xqm
->qm_grp_dqhtable
!= NULL
);
134 for (i
= 0; i
< hsize
; i
++) {
135 xfs_qm_list_init(&(xqm
->qm_usr_dqhtable
[i
]), "uxdqh", i
);
136 xfs_qm_list_init(&(xqm
->qm_grp_dqhtable
[i
]), "gxdqh", i
);
140 * Freelist of all dquots of all file systems
142 xfs_qm_freelist_init(&(xqm
->qm_dqfreelist
));
145 * dquot zone. we register our own low-memory callback.
148 xqm
->qm_dqzone
= kmem_zone_init(sizeof(xfs_dquot_t
),
150 qm_dqzone
= xqm
->qm_dqzone
;
152 xqm
->qm_dqzone
= qm_dqzone
;
154 register_shrinker(&xfs_qm_shaker
);
157 * The t_dqinfo portion of transactions.
160 xqm
->qm_dqtrxzone
= kmem_zone_init(sizeof(xfs_dquot_acct_t
),
162 qm_dqtrxzone
= xqm
->qm_dqtrxzone
;
164 xqm
->qm_dqtrxzone
= qm_dqtrxzone
;
166 atomic_set(&xqm
->qm_totaldquots
, 0);
167 xqm
->qm_dqfree_ratio
= XFS_QM_DQFREE_RATIO
;
170 mutex_init(&qcheck_lock
);
176 * Destroy the global quota manager when its reference count goes to zero.
185 ASSERT(xqm
->qm_nrefs
== 0);
186 unregister_shrinker(&xfs_qm_shaker
);
187 hsize
= xqm
->qm_dqhashmask
+ 1;
188 for (i
= 0; i
< hsize
; i
++) {
189 xfs_qm_list_destroy(&(xqm
->qm_usr_dqhtable
[i
]));
190 xfs_qm_list_destroy(&(xqm
->qm_grp_dqhtable
[i
]));
192 kmem_free(xqm
->qm_usr_dqhtable
);
193 kmem_free(xqm
->qm_grp_dqhtable
);
194 xqm
->qm_usr_dqhtable
= NULL
;
195 xqm
->qm_grp_dqhtable
= NULL
;
196 xqm
->qm_dqhashmask
= 0;
197 xfs_qm_freelist_destroy(&(xqm
->qm_dqfreelist
));
199 mutex_destroy(&qcheck_lock
);
205 * Called at mount time to let XQM know that another file system is
206 * starting quotas. This isn't crucial information as the individual mount
207 * structures are pretty independent, but it helps the XQM keep a
208 * global view of what's going on.
212 xfs_qm_hold_quotafs_ref(
213 struct xfs_mount
*mp
)
216 * Need to lock the xfs_Gqm structure for things like this. For example,
217 * the structure could disappear between the entry to this routine and
218 * a HOLD operation if not locked.
220 mutex_lock(&xfs_Gqm_lock
);
223 xfs_Gqm
= xfs_Gqm_init();
225 * We can keep a list of all filesystems with quotas mounted for
226 * debugging and statistical purposes, but ...
227 * Just take a reference and get out.
230 mutex_unlock(&xfs_Gqm_lock
);
237 * Release the reference that a filesystem took at mount time,
238 * so that we know when we need to destroy the entire quota manager.
242 xfs_qm_rele_quotafs_ref(
243 struct xfs_mount
*mp
)
245 xfs_dquot_t
*dqp
, *nextdqp
;
248 ASSERT(xfs_Gqm
->qm_nrefs
> 0);
251 * Go thru the freelist and destroy all inactive dquots.
253 xfs_qm_freelist_lock(xfs_Gqm
);
255 for (dqp
= xfs_Gqm
->qm_dqfreelist
.qh_next
;
256 dqp
!= (xfs_dquot_t
*)&(xfs_Gqm
->qm_dqfreelist
); ) {
258 nextdqp
= dqp
->dq_flnext
;
259 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
260 ASSERT(dqp
->q_mount
== NULL
);
261 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
262 ASSERT(dqp
->HL_PREVP
== NULL
);
263 ASSERT(dqp
->MPL_PREVP
== NULL
);
264 XQM_FREELIST_REMOVE(dqp
);
266 xfs_qm_dqdestroy(dqp
);
272 xfs_qm_freelist_unlock(xfs_Gqm
);
275 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
278 mutex_lock(&xfs_Gqm_lock
);
279 if (--xfs_Gqm
->qm_nrefs
== 0) {
280 xfs_qm_destroy(xfs_Gqm
);
283 mutex_unlock(&xfs_Gqm_lock
);
287 * Just destroy the quotainfo structure.
291 struct xfs_mount
*mp
)
293 if (mp
->m_quotainfo
) {
294 xfs_qm_dqpurge_all(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_UMOUNTING
);
295 xfs_qm_destroy_quotainfo(mp
);
301 * This is called from xfs_mountfs to start quotas and initialize all
302 * necessary data structures like quotainfo. This is also responsible for
303 * running a quotacheck as necessary. We are guaranteed that the superblock
304 * is consistently read in at this point.
306 * If we fail here, the mount will continue with quota turned off. We don't
307 * need to inidicate success or failure at all.
317 * If quotas on realtime volumes is not supported, we disable
318 * quotas immediately.
320 if (mp
->m_sb
.sb_rextents
) {
322 "Cannot turn on quotas for realtime filesystem %s",
328 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
331 * Allocate the quotainfo structure inside the mount struct, and
332 * create quotainode(s), and change/rev superblock if necessary.
334 error
= xfs_qm_init_quotainfo(mp
);
337 * We must turn off quotas.
339 ASSERT(mp
->m_quotainfo
== NULL
);
344 * If any of the quotas are not consistent, do a quotacheck.
346 if (XFS_QM_NEED_QUOTACHECK(mp
)) {
347 error
= xfs_qm_quotacheck(mp
);
349 /* Quotacheck failed and disabled quotas. */
354 * If one type of quotas is off, then it will lose its
355 * quotachecked status, since we won't be doing accounting for
358 if (!XFS_IS_UQUOTA_ON(mp
))
359 mp
->m_qflags
&= ~XFS_UQUOTA_CHKD
;
360 if (!(XFS_IS_GQUOTA_ON(mp
) || XFS_IS_PQUOTA_ON(mp
)))
361 mp
->m_qflags
&= ~XFS_OQUOTA_CHKD
;
365 * We actually don't have to acquire the m_sb_lock at all.
366 * This can only be called from mount, and that's single threaded. XXX
368 spin_lock(&mp
->m_sb_lock
);
369 sbf
= mp
->m_sb
.sb_qflags
;
370 mp
->m_sb
.sb_qflags
= mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
;
371 spin_unlock(&mp
->m_sb_lock
);
373 if (sbf
!= (mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
)) {
374 if (xfs_qm_write_sb_changes(mp
, XFS_SB_QFLAGS
)) {
376 * We could only have been turning quotas off.
377 * We aren't in very good shape actually because
378 * the incore structures are convinced that quotas are
379 * off, but the on disk superblock doesn't know that !
381 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp
)));
382 xfs_fs_cmn_err(CE_ALERT
, mp
,
383 "XFS mount_quotas: Superblock update failed!");
388 xfs_fs_cmn_err(CE_WARN
, mp
,
389 "Failed to initialize disk quotas.");
394 if (XFS_IS_QUOTA_ON(mp
))
395 xfs_qm_internalqcheck(mp
);
400 * Called from the vfsops layer.
403 xfs_qm_unmount_quotas(
407 * Release the dquots that root inode, et al might be holding,
408 * before we flush quotas and blow away the quotainfo structure.
410 ASSERT(mp
->m_rootip
);
411 xfs_qm_dqdetach(mp
->m_rootip
);
413 xfs_qm_dqdetach(mp
->m_rbmip
);
415 xfs_qm_dqdetach(mp
->m_rsumip
);
418 * Release the quota inodes.
420 if (mp
->m_quotainfo
) {
421 if (mp
->m_quotainfo
->qi_uquotaip
) {
422 IRELE(mp
->m_quotainfo
->qi_uquotaip
);
423 mp
->m_quotainfo
->qi_uquotaip
= NULL
;
425 if (mp
->m_quotainfo
->qi_gquotaip
) {
426 IRELE(mp
->m_quotainfo
->qi_gquotaip
);
427 mp
->m_quotainfo
->qi_gquotaip
= NULL
;
433 * Flush all dquots of the given file system to disk. The dquots are
434 * _not_ purged from memory here, just their data written to disk.
446 if (mp
->m_quotainfo
== NULL
)
450 xfs_qm_mplist_lock(mp
);
451 FOREACH_DQUOT_IN_MP(dqp
, mp
) {
453 if (! XFS_DQ_IS_DIRTY(dqp
)) {
457 xfs_dqtrace_entry(dqp
, "FLUSHALL: DQDIRTY");
458 /* XXX a sentinel would be better */
459 recl
= XFS_QI_MPLRECLAIMS(mp
);
460 if (!xfs_dqflock_nowait(dqp
)) {
462 * If we can't grab the flush lock then check
463 * to see if the dquot has been flushed delayed
464 * write. If so, grab its buffer and send it
465 * out immediately. We'll be able to acquire
466 * the flush lock when the I/O completes.
468 xfs_qm_dqflock_pushbuf_wait(dqp
);
471 * Let go of the mplist lock. We don't want to hold it
472 * across a disk write.
474 xfs_qm_mplist_unlock(mp
);
475 error
= xfs_qm_dqflush(dqp
, flags
);
480 xfs_qm_mplist_lock(mp
);
481 if (recl
!= XFS_QI_MPLRECLAIMS(mp
)) {
482 xfs_qm_mplist_unlock(mp
);
483 /* XXX restart limit */
488 xfs_qm_mplist_unlock(mp
);
493 * Release the group dquot pointers the user dquots may be
494 * carrying around as a hint. mplist is locked on entry and exit.
497 xfs_qm_detach_gdquots(
500 xfs_dquot_t
*dqp
, *gdqp
;
504 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp
));
505 dqp
= XFS_QI_MPLNEXT(mp
);
508 if ((gdqp
= dqp
->q_gdquot
)) {
510 dqp
->q_gdquot
= NULL
;
516 * Can't hold the mplist lock across a dqput.
517 * XXXmust convert to marker based iterations here.
519 nrecl
= XFS_QI_MPLRECLAIMS(mp
);
520 xfs_qm_mplist_unlock(mp
);
523 xfs_qm_mplist_lock(mp
);
524 if (nrecl
!= XFS_QI_MPLRECLAIMS(mp
))
532 * Go through all the incore dquots of this file system and take them
533 * off the mplist and hashlist, if the dquot type matches the dqtype
534 * parameter. This is used when turning off quota accounting for
535 * users and/or groups, as well as when the filesystem is unmounting.
540 uint flags
) /* QUOTAOFF/UMOUNTING/UQUOTA/PQUOTA/GQUOTA */
545 xfs_dquot_t
*nextdqp
;
548 if (mp
->m_quotainfo
== NULL
)
551 dqtype
= (flags
& XFS_QMOPT_UQUOTA
) ? XFS_DQ_USER
: 0;
552 dqtype
|= (flags
& XFS_QMOPT_PQUOTA
) ? XFS_DQ_PROJ
: 0;
553 dqtype
|= (flags
& XFS_QMOPT_GQUOTA
) ? XFS_DQ_GROUP
: 0;
555 xfs_qm_mplist_lock(mp
);
558 * In the first pass through all incore dquots of this filesystem,
559 * we release the group dquot pointers the user dquots may be
560 * carrying around as a hint. We need to do this irrespective of
561 * what's being turned off.
563 xfs_qm_detach_gdquots(mp
);
567 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp
));
569 * Try to get rid of all of the unwanted dquots. The idea is to
570 * get them off mplist and hashlist, but leave them on freelist.
572 dqp
= XFS_QI_MPLNEXT(mp
);
575 * It's OK to look at the type without taking dqlock here.
576 * We're holding the mplist lock here, and that's needed for
579 if ((dqp
->dq_flags
& dqtype
) == 0) {
584 if (!mutex_trylock(&dqp
->q_hash
->qh_lock
)) {
585 nrecl
= XFS_QI_MPLRECLAIMS(mp
);
586 xfs_qm_mplist_unlock(mp
);
587 mutex_lock(&dqp
->q_hash
->qh_lock
);
588 xfs_qm_mplist_lock(mp
);
591 * XXXTheoretically, we can get into a very long
592 * ping pong game here.
593 * No one can be adding dquots to the mplist at
594 * this point, but somebody might be taking things off.
596 if (nrecl
!= XFS_QI_MPLRECLAIMS(mp
)) {
597 mutex_unlock(&dqp
->q_hash
->qh_lock
);
603 * Take the dquot off the mplist and hashlist. It may remain on
604 * freelist in INACTIVE state.
606 nextdqp
= dqp
->MPL_NEXT
;
607 nmisses
+= xfs_qm_dqpurge(dqp
);
610 xfs_qm_mplist_unlock(mp
);
622 * Purge the dquot cache.
623 * None of the dquots should really be busy at this point.
625 if (mp
->m_quotainfo
) {
626 while ((ndquots
= xfs_qm_dqpurge_int(mp
, flags
))) {
639 xfs_dquot_t
*udqhint
, /* hint */
640 xfs_dquot_t
**IO_idqpp
)
645 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
649 * See if we already have it in the inode itself. IO_idqpp is
650 * &i_udquot or &i_gdquot. This made the code look weird, but
651 * made the logic a lot simpler.
655 xfs_dqtrace_entry(dqp
, "DQATTACH: found in ip");
660 * udqhint is the i_udquot field in inode, and is non-NULL only
661 * when the type arg is group/project. Its purpose is to save a
662 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
666 ASSERT(type
== XFS_DQ_GROUP
|| type
== XFS_DQ_PROJ
);
670 * No need to take dqlock to look at the id.
672 * The ID can't change until it gets reclaimed, and it won't
673 * be reclaimed as long as we have a ref from inode and we
676 dqp
= udqhint
->q_gdquot
;
677 if (dqp
&& be32_to_cpu(dqp
->q_core
.d_id
) == id
) {
680 ASSERT(*IO_idqpp
== NULL
);
684 xfs_dqunlock(udqhint
);
689 * We can't hold a dquot lock when we call the dqget code.
690 * We'll deadlock in no time, because of (not conforming to)
691 * lock ordering - the inodelock comes before any dquot lock,
692 * and we may drop and reacquire the ilock in xfs_qm_dqget().
694 xfs_dqunlock(udqhint
);
698 * Find the dquot from somewhere. This bumps the
699 * reference count of dquot and returns it locked.
700 * This can return ENOENT if dquot didn't exist on
701 * disk and we didn't ask it to allocate;
702 * ESRCH if quotas got turned off suddenly.
704 error
= xfs_qm_dqget(ip
->i_mount
, ip
, id
, type
, XFS_QMOPT_DOWARN
, &dqp
);
708 xfs_dqtrace_entry(dqp
, "DQATTACH: found by dqget");
711 * dqget may have dropped and re-acquired the ilock, but it guarantees
712 * that the dquot returned is the one that should go in the inode.
721 * Given a udquot and gdquot, attach a ptr to the group dquot in the
722 * udquot as a hint for future lookups. The idea sounds simple, but the
723 * execution isn't, because the udquot might have a group dquot attached
724 * already and getting rid of that gets us into lock ordering constraints.
725 * The process is complicated more by the fact that the dquots may or may not
726 * be locked on entry.
729 xfs_qm_dqattach_grouphint(
737 if ((tmp
= udq
->q_gdquot
)) {
743 udq
->q_gdquot
= NULL
;
745 * We can't keep any dqlocks when calling dqrele,
746 * because the freelist lock comes before dqlocks.
750 * we took a hard reference once upon a time in dqget,
751 * so give it back when the udquot no longer points at it
752 * dqput() does the unlocking of the dquot.
760 ASSERT(XFS_DQ_IS_LOCKED(udq
));
764 ASSERT(XFS_DQ_IS_LOCKED(udq
));
765 ASSERT(XFS_DQ_IS_LOCKED(gdq
));
767 * Somebody could have attached a gdquot here,
768 * when we dropped the uqlock. If so, just do nothing.
770 if (udq
->q_gdquot
== NULL
) {
781 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
783 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
784 * Inode may get unlocked and relocked in here, and the caller must deal with
788 xfs_qm_dqattach_locked(
792 xfs_mount_t
*mp
= ip
->i_mount
;
796 if (!XFS_IS_QUOTA_RUNNING(mp
) ||
797 !XFS_IS_QUOTA_ON(mp
) ||
798 !XFS_NOT_DQATTACHED(mp
, ip
) ||
799 ip
->i_ino
== mp
->m_sb
.sb_uquotino
||
800 ip
->i_ino
== mp
->m_sb
.sb_gquotino
)
803 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
805 if (XFS_IS_UQUOTA_ON(mp
)) {
806 error
= xfs_qm_dqattach_one(ip
, ip
->i_d
.di_uid
, XFS_DQ_USER
,
807 flags
& XFS_QMOPT_DQALLOC
,
808 NULL
, &ip
->i_udquot
);
814 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
815 if (XFS_IS_OQUOTA_ON(mp
)) {
816 error
= XFS_IS_GQUOTA_ON(mp
) ?
817 xfs_qm_dqattach_one(ip
, ip
->i_d
.di_gid
, XFS_DQ_GROUP
,
818 flags
& XFS_QMOPT_DQALLOC
,
819 ip
->i_udquot
, &ip
->i_gdquot
) :
820 xfs_qm_dqattach_one(ip
, ip
->i_d
.di_projid
, XFS_DQ_PROJ
,
821 flags
& XFS_QMOPT_DQALLOC
,
822 ip
->i_udquot
, &ip
->i_gdquot
);
824 * Don't worry about the udquot that we may have
825 * attached above. It'll get detached, if not already.
833 * Attach this group quota to the user quota as a hint.
834 * This WON'T, in general, result in a thrash.
837 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
838 ASSERT(ip
->i_udquot
);
839 ASSERT(ip
->i_gdquot
);
842 * We may or may not have the i_udquot locked at this point,
843 * but this check is OK since we don't depend on the i_gdquot to
844 * be accurate 100% all the time. It is just a hint, and this
845 * will succeed in general.
847 if (ip
->i_udquot
->q_gdquot
== ip
->i_gdquot
)
850 * Attach i_gdquot to the gdquot hint inside the i_udquot.
852 xfs_qm_dqattach_grouphint(ip
->i_udquot
, ip
->i_gdquot
);
858 if (XFS_IS_UQUOTA_ON(mp
))
859 ASSERT(ip
->i_udquot
);
860 if (XFS_IS_OQUOTA_ON(mp
))
861 ASSERT(ip
->i_gdquot
);
863 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
870 struct xfs_inode
*ip
,
875 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
876 error
= xfs_qm_dqattach_locked(ip
, flags
);
877 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
883 * Release dquots (and their references) if any.
884 * The inode should be locked EXCL except when this's called by
891 if (!(ip
->i_udquot
|| ip
->i_gdquot
))
894 ASSERT(ip
->i_ino
!= ip
->i_mount
->m_sb
.sb_uquotino
);
895 ASSERT(ip
->i_ino
!= ip
->i_mount
->m_sb
.sb_gquotino
);
897 xfs_dqtrace_entry_ino(ip
->i_udquot
, "DQDETTACH", ip
);
898 xfs_qm_dqrele(ip
->i_udquot
);
902 xfs_dqtrace_entry_ino(ip
->i_gdquot
, "DQDETTACH", ip
);
903 xfs_qm_dqrele(ip
->i_gdquot
);
909 * This is called to sync quotas. We can be told to use non-blocking
910 * semantics by either the SYNC_BDFLUSH flag or the absence of the
924 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
929 * We won't block unless we are asked to.
931 nowait
= (boolean_t
)(flags
& SYNC_BDFLUSH
|| (flags
& SYNC_WAIT
) == 0);
934 xfs_qm_mplist_lock(mp
);
936 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
937 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
938 * when we have the mplist lock, we know that dquots will be consistent
939 * as long as we have it locked.
941 if (! XFS_IS_QUOTA_ON(mp
)) {
942 xfs_qm_mplist_unlock(mp
);
945 FOREACH_DQUOT_IN_MP(dqp
, mp
) {
947 * If this is vfs_sync calling, then skip the dquots that
948 * don't 'seem' to be dirty. ie. don't acquire dqlock.
949 * This is very similar to what xfs_sync does with inodes.
951 if (flags
& SYNC_BDFLUSH
) {
952 if (! XFS_DQ_IS_DIRTY(dqp
))
958 * Try to acquire the dquot lock. We are NOT out of
959 * lock order, but we just don't want to wait for this
960 * lock, unless somebody wanted us to.
962 if (! xfs_qm_dqlock_nowait(dqp
))
969 * Now, find out for sure if this dquot is dirty or not.
971 if (! XFS_DQ_IS_DIRTY(dqp
)) {
976 /* XXX a sentinel would be better */
977 recl
= XFS_QI_MPLRECLAIMS(mp
);
978 if (!xfs_dqflock_nowait(dqp
)) {
984 * If we can't grab the flush lock then if the caller
985 * really wanted us to give this our best shot, so
986 * see if we can give a push to the buffer before we wait
987 * on the flush lock. At this point, we know that
988 * even though the dquot is being flushed,
989 * it has (new) dirty data.
991 xfs_qm_dqflock_pushbuf_wait(dqp
);
994 * Let go of the mplist lock. We don't want to hold it
995 * across a disk write
997 flush_flags
= (nowait
) ? XFS_QMOPT_DELWRI
: XFS_QMOPT_SYNC
;
998 xfs_qm_mplist_unlock(mp
);
999 xfs_dqtrace_entry(dqp
, "XQM_SYNC: DQFLUSH");
1000 error
= xfs_qm_dqflush(dqp
, flush_flags
);
1002 if (error
&& XFS_FORCED_SHUTDOWN(mp
))
1003 return 0; /* Need to prevent umount failure */
1007 xfs_qm_mplist_lock(mp
);
1008 if (recl
!= XFS_QI_MPLRECLAIMS(mp
)) {
1009 if (++restarts
>= XFS_QM_SYNC_MAX_RESTARTS
)
1012 xfs_qm_mplist_unlock(mp
);
1017 xfs_qm_mplist_unlock(mp
);
1022 * The hash chains and the mplist use the same xfs_dqhash structure as
1023 * their list head, but we can take the mplist qh_lock and one of the
1024 * hash qh_locks at the same time without any problem as they aren't
1027 static struct lock_class_key xfs_quota_mplist_class
;
1030 * This initializes all the quota information that's kept in the
1034 xfs_qm_init_quotainfo(
1037 xfs_quotainfo_t
*qinf
;
1041 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1044 * Tell XQM that we exist as soon as possible.
1046 if ((error
= xfs_qm_hold_quotafs_ref(mp
))) {
1050 qinf
= mp
->m_quotainfo
= kmem_zalloc(sizeof(xfs_quotainfo_t
), KM_SLEEP
);
1053 * See if quotainodes are setup, and if not, allocate them,
1054 * and change the superblock accordingly.
1056 if ((error
= xfs_qm_init_quotainos(mp
))) {
1058 mp
->m_quotainfo
= NULL
;
1062 xfs_qm_list_init(&qinf
->qi_dqlist
, "mpdqlist", 0);
1063 lockdep_set_class(&qinf
->qi_dqlist
.qh_lock
, &xfs_quota_mplist_class
);
1065 qinf
->qi_dqreclaims
= 0;
1067 /* mutex used to serialize quotaoffs */
1068 mutex_init(&qinf
->qi_quotaofflock
);
1070 /* Precalc some constants */
1071 qinf
->qi_dqchunklen
= XFS_FSB_TO_BB(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
1072 ASSERT(qinf
->qi_dqchunklen
);
1073 qinf
->qi_dqperchunk
= BBTOB(qinf
->qi_dqchunklen
);
1074 do_div(qinf
->qi_dqperchunk
, sizeof(xfs_dqblk_t
));
1076 mp
->m_qflags
|= (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_CHKD
);
1079 * We try to get the limits from the superuser's limits fields.
1080 * This is quite hacky, but it is standard quota practice.
1081 * We look at the USR dquot with id == 0 first, but if user quotas
1082 * are not enabled we goto the GRP dquot with id == 0.
1083 * We don't really care to keep separate default limits for user
1084 * and group quotas, at least not at this point.
1086 error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)0,
1087 XFS_IS_UQUOTA_RUNNING(mp
) ? XFS_DQ_USER
:
1088 (XFS_IS_GQUOTA_RUNNING(mp
) ? XFS_DQ_GROUP
:
1090 XFS_QMOPT_DQSUSER
|XFS_QMOPT_DOWARN
,
1093 xfs_disk_dquot_t
*ddqp
= &dqp
->q_core
;
1096 * The warnings and timers set the grace period given to
1097 * a user or group before he or she can not perform any
1098 * more writing. If it is zero, a default is used.
1100 qinf
->qi_btimelimit
= ddqp
->d_btimer
?
1101 be32_to_cpu(ddqp
->d_btimer
) : XFS_QM_BTIMELIMIT
;
1102 qinf
->qi_itimelimit
= ddqp
->d_itimer
?
1103 be32_to_cpu(ddqp
->d_itimer
) : XFS_QM_ITIMELIMIT
;
1104 qinf
->qi_rtbtimelimit
= ddqp
->d_rtbtimer
?
1105 be32_to_cpu(ddqp
->d_rtbtimer
) : XFS_QM_RTBTIMELIMIT
;
1106 qinf
->qi_bwarnlimit
= ddqp
->d_bwarns
?
1107 be16_to_cpu(ddqp
->d_bwarns
) : XFS_QM_BWARNLIMIT
;
1108 qinf
->qi_iwarnlimit
= ddqp
->d_iwarns
?
1109 be16_to_cpu(ddqp
->d_iwarns
) : XFS_QM_IWARNLIMIT
;
1110 qinf
->qi_rtbwarnlimit
= ddqp
->d_rtbwarns
?
1111 be16_to_cpu(ddqp
->d_rtbwarns
) : XFS_QM_RTBWARNLIMIT
;
1112 qinf
->qi_bhardlimit
= be64_to_cpu(ddqp
->d_blk_hardlimit
);
1113 qinf
->qi_bsoftlimit
= be64_to_cpu(ddqp
->d_blk_softlimit
);
1114 qinf
->qi_ihardlimit
= be64_to_cpu(ddqp
->d_ino_hardlimit
);
1115 qinf
->qi_isoftlimit
= be64_to_cpu(ddqp
->d_ino_softlimit
);
1116 qinf
->qi_rtbhardlimit
= be64_to_cpu(ddqp
->d_rtb_hardlimit
);
1117 qinf
->qi_rtbsoftlimit
= be64_to_cpu(ddqp
->d_rtb_softlimit
);
1120 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1121 * we don't want this dquot cached. We haven't done a
1122 * quotacheck yet, and quotacheck doesn't like incore dquots.
1124 xfs_qm_dqdestroy(dqp
);
1126 qinf
->qi_btimelimit
= XFS_QM_BTIMELIMIT
;
1127 qinf
->qi_itimelimit
= XFS_QM_ITIMELIMIT
;
1128 qinf
->qi_rtbtimelimit
= XFS_QM_RTBTIMELIMIT
;
1129 qinf
->qi_bwarnlimit
= XFS_QM_BWARNLIMIT
;
1130 qinf
->qi_iwarnlimit
= XFS_QM_IWARNLIMIT
;
1131 qinf
->qi_rtbwarnlimit
= XFS_QM_RTBWARNLIMIT
;
1139 * Gets called when unmounting a filesystem or when all quotas get
1141 * This purges the quota inodes, destroys locks and frees itself.
1144 xfs_qm_destroy_quotainfo(
1147 xfs_quotainfo_t
*qi
;
1149 qi
= mp
->m_quotainfo
;
1151 ASSERT(xfs_Gqm
!= NULL
);
1154 * Release the reference that XQM kept, so that we know
1155 * when the XQM structure should be freed. We cannot assume
1156 * that xfs_Gqm is non-null after this point.
1158 xfs_qm_rele_quotafs_ref(mp
);
1160 xfs_qm_list_destroy(&qi
->qi_dqlist
);
1162 if (qi
->qi_uquotaip
) {
1163 IRELE(qi
->qi_uquotaip
);
1164 qi
->qi_uquotaip
= NULL
; /* paranoia */
1166 if (qi
->qi_gquotaip
) {
1167 IRELE(qi
->qi_gquotaip
);
1168 qi
->qi_gquotaip
= NULL
;
1170 mutex_destroy(&qi
->qi_quotaofflock
);
1172 mp
->m_quotainfo
= NULL
;
1177 /* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1186 mutex_init(&list
->qh_lock
);
1187 list
->qh_next
= NULL
;
1188 list
->qh_version
= 0;
1189 list
->qh_nelems
= 0;
1193 xfs_qm_list_destroy(
1196 mutex_destroy(&(list
->qh_lock
));
1201 * Stripped down version of dqattach. This doesn't attach, or even look at the
1202 * dquots attached to the inode. The rationale is that there won't be any
1203 * attached at the time this is called from quotacheck.
1206 xfs_qm_dqget_noattach(
1208 xfs_dquot_t
**O_udqpp
,
1209 xfs_dquot_t
**O_gdqpp
)
1213 xfs_dquot_t
*udqp
, *gdqp
;
1215 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1220 if (XFS_IS_UQUOTA_ON(mp
)) {
1221 ASSERT(ip
->i_udquot
== NULL
);
1223 * We want the dquot allocated if it doesn't exist.
1225 if ((error
= xfs_qm_dqget(mp
, ip
, ip
->i_d
.di_uid
, XFS_DQ_USER
,
1226 XFS_QMOPT_DQALLOC
| XFS_QMOPT_DOWARN
,
1229 * Shouldn't be able to turn off quotas here.
1231 ASSERT(error
!= ESRCH
);
1232 ASSERT(error
!= ENOENT
);
1238 if (XFS_IS_OQUOTA_ON(mp
)) {
1239 ASSERT(ip
->i_gdquot
== NULL
);
1242 error
= XFS_IS_GQUOTA_ON(mp
) ?
1243 xfs_qm_dqget(mp
, ip
,
1244 ip
->i_d
.di_gid
, XFS_DQ_GROUP
,
1245 XFS_QMOPT_DQALLOC
|XFS_QMOPT_DOWARN
,
1247 xfs_qm_dqget(mp
, ip
,
1248 ip
->i_d
.di_projid
, XFS_DQ_PROJ
,
1249 XFS_QMOPT_DQALLOC
|XFS_QMOPT_DOWARN
,
1253 xfs_qm_dqrele(udqp
);
1254 ASSERT(error
!= ESRCH
);
1255 ASSERT(error
!= ENOENT
);
1260 /* Reacquire the locks in the right order */
1262 if (! xfs_qm_dqlock_nowait(udqp
)) {
1274 if (udqp
) ASSERT(XFS_DQ_IS_LOCKED(udqp
));
1275 if (gdqp
) ASSERT(XFS_DQ_IS_LOCKED(gdqp
));
1281 * Create an inode and return with a reference already taken, but unlocked
1282 * This is how we create quota inodes
1295 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_QINOCREATE
);
1296 if ((error
= xfs_trans_reserve(tp
,
1297 XFS_QM_QINOCREATE_SPACE_RES(mp
),
1298 XFS_CREATE_LOG_RES(mp
), 0,
1299 XFS_TRANS_PERM_LOG_RES
,
1300 XFS_CREATE_LOG_COUNT
))) {
1301 xfs_trans_cancel(tp
, 0);
1305 if ((error
= xfs_dir_ialloc(&tp
, NULL
, S_IFREG
, 1, 0,
1306 &xfs_zerocr
, 0, 1, ip
, &committed
))) {
1307 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1313 * Keep an extra reference to this quota inode. This inode is
1314 * locked exclusively and joined to the transaction already.
1316 ASSERT(xfs_isilocked(*ip
, XFS_ILOCK_EXCL
));
1320 * Make the changes in the superblock, and log those too.
1321 * sbfields arg may contain fields other than *QUOTINO;
1322 * VERSIONNUM for example.
1324 spin_lock(&mp
->m_sb_lock
);
1325 if (flags
& XFS_QMOPT_SBVERSION
) {
1326 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1327 unsigned oldv
= mp
->m_sb
.sb_versionnum
;
1329 ASSERT(!xfs_sb_version_hasquota(&mp
->m_sb
));
1330 ASSERT((sbfields
& (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1331 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
)) ==
1332 (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1333 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
));
1335 xfs_sb_version_addquota(&mp
->m_sb
);
1336 mp
->m_sb
.sb_uquotino
= NULLFSINO
;
1337 mp
->m_sb
.sb_gquotino
= NULLFSINO
;
1339 /* qflags will get updated _after_ quotacheck */
1340 mp
->m_sb
.sb_qflags
= 0;
1341 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1343 "Old superblock version %x, converting to %x.",
1344 oldv
, mp
->m_sb
.sb_versionnum
);
1347 if (flags
& XFS_QMOPT_UQUOTA
)
1348 mp
->m_sb
.sb_uquotino
= (*ip
)->i_ino
;
1350 mp
->m_sb
.sb_gquotino
= (*ip
)->i_ino
;
1351 spin_unlock(&mp
->m_sb_lock
);
1352 xfs_mod_sb(tp
, sbfields
);
1354 if ((error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
))) {
1355 xfs_fs_cmn_err(CE_ALERT
, mp
, "XFS qino_alloc failed!");
1363 xfs_qm_reset_dqcounts(
1369 xfs_disk_dquot_t
*ddq
;
1372 xfs_buftrace("RESET DQUOTS", bp
);
1374 * Reset all counters and timers. They'll be
1375 * started afresh by xfs_qm_quotacheck.
1378 j
= XFS_FSB_TO_B(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
1379 do_div(j
, sizeof(xfs_dqblk_t
));
1380 ASSERT(XFS_QM_DQPERBLK(mp
) == j
);
1382 ddq
= (xfs_disk_dquot_t
*)XFS_BUF_PTR(bp
);
1383 for (j
= 0; j
< XFS_QM_DQPERBLK(mp
); j
++) {
1385 * Do a sanity check, and if needed, repair the dqblk. Don't
1386 * output any warnings because it's perfectly possible to
1387 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1389 (void) xfs_qm_dqcheck(ddq
, id
+j
, type
, XFS_QMOPT_DQREPAIR
,
1393 ddq
->d_rtbcount
= 0;
1396 ddq
->d_rtbtimer
= 0;
1399 ddq
->d_rtbwarns
= 0;
1400 ddq
= (xfs_disk_dquot_t
*) ((xfs_dqblk_t
*)ddq
+ 1);
1409 xfs_filblks_t blkcnt
,
1420 incr
= (blkcnt
> XFS_QM_MAX_DQCLUSTER_LOGSZ
) ?
1421 XFS_QM_MAX_DQCLUSTER_LOGSZ
: blkcnt
;
1422 type
= flags
& XFS_QMOPT_UQUOTA
? XFS_DQ_USER
:
1423 (flags
& XFS_QMOPT_PQUOTA
? XFS_DQ_PROJ
: XFS_DQ_GROUP
);
1427 * Blkcnt arg can be a very big number, and might even be
1428 * larger than the log itself. So, we have to break it up into
1429 * manageable-sized transactions.
1430 * Note that we don't start a permanent transaction here; we might
1431 * not be able to get a log reservation for the whole thing up front,
1432 * and we don't really care to either, because we just discard
1433 * everything if we were to crash in the middle of this loop.
1436 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
,
1437 XFS_FSB_TO_DADDR(mp
, bno
),
1438 (int)XFS_QI_DQCHUNKLEN(mp
), 0, &bp
);
1442 xfs_qm_reset_dqcounts(mp
, bp
, firstid
, type
);
1443 xfs_bdwrite(mp
, bp
);
1445 * goto the next block.
1448 firstid
+= XFS_QM_DQPERBLK(mp
);
1454 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1455 * caller supplied function for every chunk of dquots that we find.
1463 xfs_bmbt_irec_t
*map
;
1464 int i
, nmaps
; /* number of map entries */
1465 int error
; /* return value */
1466 xfs_fileoff_t lblkno
;
1467 xfs_filblks_t maxlblkcnt
;
1469 xfs_fsblock_t rablkno
;
1470 xfs_filblks_t rablkcnt
;
1474 * This looks racy, but we can't keep an inode lock across a
1475 * trans_reserve. But, this gets called during quotacheck, and that
1476 * happens only at mount time which is single threaded.
1478 if (qip
->i_d
.di_nblocks
== 0)
1481 map
= kmem_alloc(XFS_DQITER_MAP_SIZE
* sizeof(*map
), KM_SLEEP
);
1484 maxlblkcnt
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1486 nmaps
= XFS_DQITER_MAP_SIZE
;
1488 * We aren't changing the inode itself. Just changing
1489 * some of its data. No new blocks are added here, and
1490 * the inode is never added to the transaction.
1492 xfs_ilock(qip
, XFS_ILOCK_SHARED
);
1493 error
= xfs_bmapi(NULL
, qip
, lblkno
,
1494 maxlblkcnt
- lblkno
,
1497 0, map
, &nmaps
, NULL
, NULL
);
1498 xfs_iunlock(qip
, XFS_ILOCK_SHARED
);
1502 ASSERT(nmaps
<= XFS_DQITER_MAP_SIZE
);
1503 for (i
= 0; i
< nmaps
; i
++) {
1504 ASSERT(map
[i
].br_startblock
!= DELAYSTARTBLOCK
);
1505 ASSERT(map
[i
].br_blockcount
);
1508 lblkno
+= map
[i
].br_blockcount
;
1510 if (map
[i
].br_startblock
== HOLESTARTBLOCK
)
1513 firstid
= (xfs_dqid_t
) map
[i
].br_startoff
*
1514 XFS_QM_DQPERBLK(mp
);
1516 * Do a read-ahead on the next extent.
1518 if ((i
+1 < nmaps
) &&
1519 (map
[i
+1].br_startblock
!= HOLESTARTBLOCK
)) {
1520 rablkcnt
= map
[i
+1].br_blockcount
;
1521 rablkno
= map
[i
+1].br_startblock
;
1522 while (rablkcnt
--) {
1523 xfs_baread(mp
->m_ddev_targp
,
1524 XFS_FSB_TO_DADDR(mp
, rablkno
),
1525 (int)XFS_QI_DQCHUNKLEN(mp
));
1530 * Iterate thru all the blks in the extent and
1531 * reset the counters of all the dquots inside them.
1533 if ((error
= xfs_qm_dqiter_bufs(mp
,
1535 map
[i
].br_startblock
,
1536 map
[i
].br_blockcount
,
1544 } while (nmaps
> 0);
1552 * Called by dqusage_adjust in doing a quotacheck.
1553 * Given the inode, and a dquot (either USR or GRP, doesn't matter),
1554 * this updates its incore copy as well as the buffer copy. This is
1555 * so that once the quotacheck is done, we can just log all the buffers,
1556 * as opposed to logging numerous updates to individual dquots.
1559 xfs_qm_quotacheck_dqadjust(
1564 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1565 xfs_dqtrace_entry(dqp
, "QCHECK DQADJUST");
1567 * Adjust the inode count and the block count to reflect this inode's
1570 be64_add_cpu(&dqp
->q_core
.d_icount
, 1);
1571 dqp
->q_res_icount
++;
1573 be64_add_cpu(&dqp
->q_core
.d_bcount
, nblks
);
1574 dqp
->q_res_bcount
+= nblks
;
1577 be64_add_cpu(&dqp
->q_core
.d_rtbcount
, rtblks
);
1578 dqp
->q_res_rtbcount
+= rtblks
;
1582 * Set default limits, adjust timers (since we changed usages)
1584 if (! XFS_IS_SUSER_DQUOT(dqp
)) {
1585 xfs_qm_adjust_dqlimits(dqp
->q_mount
, &dqp
->q_core
);
1586 xfs_qm_adjust_dqtimers(dqp
->q_mount
, &dqp
->q_core
);
1589 dqp
->dq_flags
|= XFS_DQ_DIRTY
;
1595 xfs_qcnt_t
*O_rtblks
)
1597 xfs_filblks_t rtblks
; /* total rt blks */
1598 xfs_extnum_t idx
; /* extent record index */
1599 xfs_ifork_t
*ifp
; /* inode fork pointer */
1600 xfs_extnum_t nextents
; /* number of extent entries */
1603 ASSERT(XFS_IS_REALTIME_INODE(ip
));
1604 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1605 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1606 if ((error
= xfs_iread_extents(NULL
, ip
, XFS_DATA_FORK
)))
1610 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
1611 for (idx
= 0; idx
< nextents
; idx
++)
1612 rtblks
+= xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp
, idx
));
1613 *O_rtblks
= (xfs_qcnt_t
)rtblks
;
1618 * callback routine supplied to bulkstat(). Given an inumber, find its
1619 * dquots and update them to account for resources taken by that inode.
1623 xfs_qm_dqusage_adjust(
1624 xfs_mount_t
*mp
, /* mount point for filesystem */
1625 xfs_ino_t ino
, /* inode number to get data for */
1626 void __user
*buffer
, /* not used */
1627 int ubsize
, /* not used */
1628 void *private_data
, /* not used */
1629 xfs_daddr_t bno
, /* starting block of inode cluster */
1630 int *ubused
, /* not used */
1631 void *dip
, /* on-disk inode pointer (not used) */
1632 int *res
) /* result code value */
1635 xfs_dquot_t
*udqp
, *gdqp
;
1636 xfs_qcnt_t nblks
, rtblks
;
1639 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1642 * rootino must have its resources accounted for, not so with the quota
1645 if (ino
== mp
->m_sb
.sb_uquotino
|| ino
== mp
->m_sb
.sb_gquotino
) {
1646 *res
= BULKSTAT_RV_NOTHING
;
1647 return XFS_ERROR(EINVAL
);
1651 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1652 * interface expects the inode to be exclusively locked because that's
1653 * the case in all other instances. It's OK that we do this because
1654 * quotacheck is done only at mount time.
1656 if ((error
= xfs_iget(mp
, NULL
, ino
, 0, XFS_ILOCK_EXCL
, &ip
, bno
))) {
1657 *res
= BULKSTAT_RV_NOTHING
;
1662 * Obtain the locked dquots. In case of an error (eg. allocation
1663 * fails for ENOSPC), we return the negative of the error number
1664 * to bulkstat, so that it can get propagated to quotacheck() and
1665 * making us disable quotas for the file system.
1667 if ((error
= xfs_qm_dqget_noattach(ip
, &udqp
, &gdqp
))) {
1668 xfs_iput(ip
, XFS_ILOCK_EXCL
);
1669 *res
= BULKSTAT_RV_GIVEUP
;
1674 if (! XFS_IS_REALTIME_INODE(ip
)) {
1675 nblks
= (xfs_qcnt_t
)ip
->i_d
.di_nblocks
;
1678 * Walk thru the extent list and count the realtime blocks.
1680 if ((error
= xfs_qm_get_rtblks(ip
, &rtblks
))) {
1681 xfs_iput(ip
, XFS_ILOCK_EXCL
);
1686 *res
= BULKSTAT_RV_GIVEUP
;
1689 nblks
= (xfs_qcnt_t
)ip
->i_d
.di_nblocks
- rtblks
;
1691 ASSERT(ip
->i_delayed_blks
== 0);
1694 * We can't release the inode while holding its dquot locks.
1695 * The inode can go into inactive and might try to acquire the dquotlocks.
1696 * So, just unlock here and do a vn_rele at the end.
1698 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1701 * Add the (disk blocks and inode) resources occupied by this
1702 * inode to its dquots. We do this adjustment in the incore dquot,
1703 * and also copy the changes to its buffer.
1704 * We don't care about putting these changes in a transaction
1705 * envelope because if we crash in the middle of a 'quotacheck'
1706 * we have to start from the beginning anyway.
1707 * Once we're done, we'll log all the dquot bufs.
1709 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1710 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1712 if (XFS_IS_UQUOTA_ON(mp
)) {
1714 xfs_qm_quotacheck_dqadjust(udqp
, nblks
, rtblks
);
1717 if (XFS_IS_OQUOTA_ON(mp
)) {
1719 xfs_qm_quotacheck_dqadjust(gdqp
, nblks
, rtblks
);
1723 * Now release the inode. This will send it to 'inactive', and
1724 * possibly even free blocks.
1731 *res
= BULKSTAT_RV_DIDONE
;
1736 * Walk thru all the filesystem inodes and construct a consistent view
1737 * of the disk quota world. If the quotacheck fails, disable quotas.
1743 int done
, count
, error
;
1746 xfs_inode_t
*uip
, *gip
;
1754 ASSERT(XFS_QI_UQIP(mp
) || XFS_QI_GQIP(mp
));
1755 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1758 * There should be no cached dquots. The (simplistic) quotacheck
1759 * algorithm doesn't like that.
1761 ASSERT(XFS_QI_MPLNDQUOTS(mp
) == 0);
1763 cmn_err(CE_NOTE
, "XFS quotacheck %s: Please wait.", mp
->m_fsname
);
1766 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1767 * their counters to zero. We need a clean slate.
1768 * We don't log our changes till later.
1770 if ((uip
= XFS_QI_UQIP(mp
))) {
1771 if ((error
= xfs_qm_dqiterate(mp
, uip
, XFS_QMOPT_UQUOTA
)))
1773 flags
|= XFS_UQUOTA_CHKD
;
1776 if ((gip
= XFS_QI_GQIP(mp
))) {
1777 if ((error
= xfs_qm_dqiterate(mp
, gip
, XFS_IS_GQUOTA_ON(mp
) ?
1778 XFS_QMOPT_GQUOTA
: XFS_QMOPT_PQUOTA
)))
1780 flags
|= XFS_OQUOTA_CHKD
;
1785 * Iterate thru all the inodes in the file system,
1786 * adjusting the corresponding dquot counters in core.
1788 if ((error
= xfs_bulkstat(mp
, &lastino
, &count
,
1789 xfs_qm_dqusage_adjust
, NULL
,
1790 structsz
, NULL
, BULKSTAT_FG_IGET
, &done
)))
1796 * We've made all the changes that we need to make incore.
1797 * Flush them down to disk buffers if everything was updated
1801 error
= xfs_qm_dqflush_all(mp
, XFS_QMOPT_DELWRI
);
1804 * We can get this error if we couldn't do a dquot allocation inside
1805 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1806 * dirty dquots that might be cached, we just want to get rid of them
1807 * and turn quotaoff. The dquots won't be attached to any of the inodes
1808 * at this point (because we intentionally didn't in dqget_noattach).
1811 xfs_qm_dqpurge_all(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_QUOTAOFF
);
1816 * We didn't log anything, because if we crashed, we'll have to
1817 * start the quotacheck from scratch anyway. However, we must make
1818 * sure that our dquot changes are secure before we put the
1819 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1822 XFS_bflush(mp
->m_ddev_targp
);
1825 * If one type of quotas is off, then it will lose its
1826 * quotachecked status, since we won't be doing accounting for
1827 * that type anymore.
1829 mp
->m_qflags
&= ~(XFS_OQUOTA_CHKD
| XFS_UQUOTA_CHKD
);
1830 mp
->m_qflags
|= flags
;
1832 XQM_LIST_PRINT(&(XFS_QI_MPL_LIST(mp
)), MPL_NEXT
, "++++ Mp list +++");
1836 cmn_err(CE_WARN
, "XFS quotacheck %s: Unsuccessful (Error %d): "
1837 "Disabling quotas.",
1838 mp
->m_fsname
, error
);
1840 * We must turn off quotas.
1842 ASSERT(mp
->m_quotainfo
!= NULL
);
1843 ASSERT(xfs_Gqm
!= NULL
);
1844 xfs_qm_destroy_quotainfo(mp
);
1845 if (xfs_mount_reset_sbqflags(mp
)) {
1846 cmn_err(CE_WARN
, "XFS quotacheck %s: "
1847 "Failed to reset quota flags.", mp
->m_fsname
);
1850 cmn_err(CE_NOTE
, "XFS quotacheck %s: Done.", mp
->m_fsname
);
1856 * This is called after the superblock has been read in and we're ready to
1857 * iget the quota inodes.
1860 xfs_qm_init_quotainos(
1863 xfs_inode_t
*uip
, *gip
;
1868 ASSERT(mp
->m_quotainfo
);
1874 * Get the uquota and gquota inodes
1876 if (xfs_sb_version_hasquota(&mp
->m_sb
)) {
1877 if (XFS_IS_UQUOTA_ON(mp
) &&
1878 mp
->m_sb
.sb_uquotino
!= NULLFSINO
) {
1879 ASSERT(mp
->m_sb
.sb_uquotino
> 0);
1880 if ((error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_uquotino
,
1882 return XFS_ERROR(error
);
1884 if (XFS_IS_OQUOTA_ON(mp
) &&
1885 mp
->m_sb
.sb_gquotino
!= NULLFSINO
) {
1886 ASSERT(mp
->m_sb
.sb_gquotino
> 0);
1887 if ((error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_gquotino
,
1891 return XFS_ERROR(error
);
1895 flags
|= XFS_QMOPT_SBVERSION
;
1896 sbflags
|= (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1897 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
);
1901 * Create the two inodes, if they don't exist already. The changes
1902 * made above will get added to a transaction and logged in one of
1903 * the qino_alloc calls below. If the device is readonly,
1904 * temporarily switch to read-write to do this.
1906 if (XFS_IS_UQUOTA_ON(mp
) && uip
== NULL
) {
1907 if ((error
= xfs_qm_qino_alloc(mp
, &uip
,
1908 sbflags
| XFS_SB_UQUOTINO
,
1909 flags
| XFS_QMOPT_UQUOTA
)))
1910 return XFS_ERROR(error
);
1912 flags
&= ~XFS_QMOPT_SBVERSION
;
1914 if (XFS_IS_OQUOTA_ON(mp
) && gip
== NULL
) {
1915 flags
|= (XFS_IS_GQUOTA_ON(mp
) ?
1916 XFS_QMOPT_GQUOTA
: XFS_QMOPT_PQUOTA
);
1917 error
= xfs_qm_qino_alloc(mp
, &gip
,
1918 sbflags
| XFS_SB_GQUOTINO
, flags
);
1923 return XFS_ERROR(error
);
1927 XFS_QI_UQIP(mp
) = uip
;
1928 XFS_QI_GQIP(mp
) = gip
;
1935 * Traverse the freelist of dquots and attempt to reclaim a maximum of
1936 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
1937 * favor the lookup function ...
1938 * XXXsup merge this with qm_reclaim_one().
1941 xfs_qm_shake_freelist(
1946 xfs_dquot_t
*dqp
, *nextdqp
;
1958 cmn_err(CE_DEBUG
, "Shake free 0x%x", howmany
);
1960 /* lock order is : hashchainlock, freelistlock, mplistlock */
1962 xfs_qm_freelist_lock(xfs_Gqm
);
1964 for (dqp
= xfs_Gqm
->qm_dqfreelist
.qh_next
;
1965 ((dqp
!= (xfs_dquot_t
*) &xfs_Gqm
->qm_dqfreelist
) &&
1966 nreclaimed
< howmany
); ) {
1970 * We are racing with dqlookup here. Naturally we don't
1971 * want to reclaim a dquot that lookup wants.
1973 if (dqp
->dq_flags
& XFS_DQ_WANT
) {
1975 xfs_qm_freelist_unlock(xfs_Gqm
);
1976 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
1978 XQM_STATS_INC(xqmstats
.xs_qm_dqwants
);
1983 * If the dquot is inactive, we are assured that it is
1984 * not on the mplist or the hashlist, and that makes our
1987 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
1988 ASSERT(dqp
->q_mount
== NULL
);
1989 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
1990 ASSERT(dqp
->HL_PREVP
== NULL
);
1991 ASSERT(dqp
->MPL_PREVP
== NULL
);
1992 XQM_STATS_INC(xqmstats
.xs_qm_dqinact_reclaims
);
1993 nextdqp
= dqp
->dq_flnext
;
1997 ASSERT(dqp
->MPL_PREVP
);
1999 * Try to grab the flush lock. If this dquot is in the process of
2000 * getting flushed to disk, we don't want to reclaim it.
2002 if (!xfs_dqflock_nowait(dqp
)) {
2004 dqp
= dqp
->dq_flnext
;
2009 * We have the flush lock so we know that this is not in the
2010 * process of being flushed. So, if this is dirty, flush it
2011 * DELWRI so that we don't get a freelist infested with
2014 if (XFS_DQ_IS_DIRTY(dqp
)) {
2016 xfs_dqtrace_entry(dqp
, "DQSHAKE: DQDIRTY");
2018 * We flush it delayed write, so don't bother
2019 * releasing the mplock.
2021 error
= xfs_qm_dqflush(dqp
, XFS_QMOPT_DELWRI
);
2023 xfs_fs_cmn_err(CE_WARN
, dqp
->q_mount
,
2024 "xfs_qm_dqflush_all: dquot %p flush failed", dqp
);
2026 xfs_dqunlock(dqp
); /* dqflush unlocks dqflock */
2027 dqp
= dqp
->dq_flnext
;
2031 * We're trying to get the hashlock out of order. This races
2032 * with dqlookup; so, we giveup and goto the next dquot if
2033 * we couldn't get the hashlock. This way, we won't starve
2034 * a dqlookup process that holds the hashlock that is
2035 * waiting for the freelist lock.
2037 if (!mutex_trylock(&dqp
->q_hash
->qh_lock
)) {
2040 dqp
= dqp
->dq_flnext
;
2044 * This races with dquot allocation code as well as dqflush_all
2045 * and reclaim code. So, if we failed to grab the mplist lock,
2046 * giveup everything and start over.
2050 if (! xfs_qm_mplist_nowait(dqp
->q_mount
)) {
2051 /* XXX put a sentinel so that we can come back here */
2054 mutex_unlock(&hash
->qh_lock
);
2055 xfs_qm_freelist_unlock(xfs_Gqm
);
2056 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2060 xfs_dqtrace_entry(dqp
, "DQSHAKE: UNLINKING");
2062 cmn_err(CE_DEBUG
, "Shake 0x%p, ID 0x%x\n",
2063 dqp
, be32_to_cpu(dqp
->q_core
.d_id
));
2065 ASSERT(dqp
->q_nrefs
== 0);
2066 nextdqp
= dqp
->dq_flnext
;
2067 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp
->q_mount
)), dqp
);
2068 XQM_HASHLIST_REMOVE(hash
, dqp
);
2070 xfs_qm_mplist_unlock(dqp
->q_mount
);
2071 mutex_unlock(&hash
->qh_lock
);
2074 XQM_FREELIST_REMOVE(dqp
);
2077 XQM_STATS_INC(xqmstats
.xs_qm_dqshake_reclaims
);
2078 xfs_qm_dqdestroy(dqp
);
2081 xfs_qm_freelist_unlock(xfs_Gqm
);
2087 * The kmem_shake interface is invoked when memory is running low.
2091 xfs_qm_shake(int nr_to_scan
, gfp_t gfp_mask
)
2093 int ndqused
, nfree
, n
;
2095 if (!kmem_shake_allow(gfp_mask
))
2100 nfree
= xfs_Gqm
->qm_dqfreelist
.qh_nelems
; /* free dquots */
2101 /* incore dquots in all f/s's */
2102 ndqused
= atomic_read(&xfs_Gqm
->qm_totaldquots
) - nfree
;
2104 ASSERT(ndqused
>= 0);
2106 if (nfree
<= ndqused
&& nfree
< ndquot
)
2109 ndqused
*= xfs_Gqm
->qm_dqfree_ratio
; /* target # of free dquots */
2110 n
= nfree
- ndqused
- ndquot
; /* # over target */
2112 return xfs_qm_shake_freelist(MAX(nfree
, n
));
2117 * Just pop the least recently used dquot off the freelist and
2118 * recycle it. The returned dquot is locked.
2120 STATIC xfs_dquot_t
*
2121 xfs_qm_dqreclaim_one(void)
2123 xfs_dquot_t
*dqpout
;
2132 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
2134 xfs_qm_freelist_lock(xfs_Gqm
);
2136 FOREACH_DQUOT_IN_FREELIST(dqp
, &(xfs_Gqm
->qm_dqfreelist
)) {
2140 * We are racing with dqlookup here. Naturally we don't
2141 * want to reclaim a dquot that lookup wants. We release the
2142 * freelist lock and start over, so that lookup will grab
2143 * both the dquot and the freelistlock.
2145 if (dqp
->dq_flags
& XFS_DQ_WANT
) {
2146 ASSERT(! (dqp
->dq_flags
& XFS_DQ_INACTIVE
));
2147 xfs_dqtrace_entry(dqp
, "DQRECLAIM: DQWANT");
2149 xfs_qm_freelist_unlock(xfs_Gqm
);
2150 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2152 XQM_STATS_INC(xqmstats
.xs_qm_dqwants
);
2157 * If the dquot is inactive, we are assured that it is
2158 * not on the mplist or the hashlist, and that makes our
2161 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
2162 ASSERT(dqp
->q_mount
== NULL
);
2163 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
2164 ASSERT(dqp
->HL_PREVP
== NULL
);
2165 ASSERT(dqp
->MPL_PREVP
== NULL
);
2166 XQM_FREELIST_REMOVE(dqp
);
2169 XQM_STATS_INC(xqmstats
.xs_qm_dqinact_reclaims
);
2173 ASSERT(dqp
->q_hash
);
2174 ASSERT(dqp
->MPL_PREVP
);
2177 * Try to grab the flush lock. If this dquot is in the process of
2178 * getting flushed to disk, we don't want to reclaim it.
2180 if (!xfs_dqflock_nowait(dqp
)) {
2186 * We have the flush lock so we know that this is not in the
2187 * process of being flushed. So, if this is dirty, flush it
2188 * DELWRI so that we don't get a freelist infested with
2191 if (XFS_DQ_IS_DIRTY(dqp
)) {
2193 xfs_dqtrace_entry(dqp
, "DQRECLAIM: DQDIRTY");
2195 * We flush it delayed write, so don't bother
2196 * releasing the freelist lock.
2198 error
= xfs_qm_dqflush(dqp
, XFS_QMOPT_DELWRI
);
2200 xfs_fs_cmn_err(CE_WARN
, dqp
->q_mount
,
2201 "xfs_qm_dqreclaim: dquot %p flush failed", dqp
);
2203 xfs_dqunlock(dqp
); /* dqflush unlocks dqflock */
2207 if (! xfs_qm_mplist_nowait(dqp
->q_mount
)) {
2213 if (!mutex_trylock(&dqp
->q_hash
->qh_lock
))
2216 ASSERT(dqp
->q_nrefs
== 0);
2217 xfs_dqtrace_entry(dqp
, "DQRECLAIM: UNLINKING");
2218 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp
->q_mount
)), dqp
);
2219 XQM_HASHLIST_REMOVE(dqp
->q_hash
, dqp
);
2220 XQM_FREELIST_REMOVE(dqp
);
2222 mutex_unlock(&dqp
->q_hash
->qh_lock
);
2224 xfs_qm_mplist_unlock(dqp
->q_mount
);
2231 xfs_qm_freelist_unlock(xfs_Gqm
);
2236 /*------------------------------------------------------------------*/
2239 * Return a new incore dquot. Depending on the number of
2240 * dquots in the system, we either allocate a new one on the kernel heap,
2241 * or reclaim a free one.
2242 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2243 * to reclaim an existing one from the freelist.
2246 xfs_qm_dqalloc_incore(
2247 xfs_dquot_t
**O_dqpp
)
2252 * Check against high water mark to see if we want to pop
2253 * a nincompoop dquot off the freelist.
2255 if (atomic_read(&xfs_Gqm
->qm_totaldquots
) >= ndquot
) {
2257 * Try to recycle a dquot from the freelist.
2259 if ((dqp
= xfs_qm_dqreclaim_one())) {
2260 XQM_STATS_INC(xqmstats
.xs_qm_dqreclaims
);
2262 * Just zero the core here. The rest will get
2263 * reinitialized by caller. XXX we shouldn't even
2266 memset(&dqp
->q_core
, 0, sizeof(dqp
->q_core
));
2270 XQM_STATS_INC(xqmstats
.xs_qm_dqreclaim_misses
);
2274 * Allocate a brand new dquot on the kernel heap and return it
2275 * to the caller to initialize.
2277 ASSERT(xfs_Gqm
->qm_dqzone
!= NULL
);
2278 *O_dqpp
= kmem_zone_zalloc(xfs_Gqm
->qm_dqzone
, KM_SLEEP
);
2279 atomic_inc(&xfs_Gqm
->qm_totaldquots
);
2286 * Start a transaction and write the incore superblock changes to
2287 * disk. flags parameter indicates which fields have changed.
2290 xfs_qm_write_sb_changes(
2298 cmn_err(CE_NOTE
, "Writing superblock quota changes :%s", mp
->m_fsname
);
2300 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
2301 if ((error
= xfs_trans_reserve(tp
, 0,
2302 mp
->m_sb
.sb_sectsize
+ 128, 0,
2304 XFS_DEFAULT_LOG_COUNT
))) {
2305 xfs_trans_cancel(tp
, 0);
2309 xfs_mod_sb(tp
, flags
);
2310 error
= xfs_trans_commit(tp
, 0);
2316 /* --------------- utility functions for vnodeops ---------------- */
2320 * Given an inode, a uid and gid (from cred_t) make sure that we have
2321 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2322 * quotas by creating this file.
2323 * This also attaches dquot(s) to the given inode after locking it,
2324 * and returns the dquots corresponding to the uid and/or gid.
2326 * in : inode (unlocked)
2327 * out : udquot, gdquot with references taken and unlocked
2331 struct xfs_inode
*ip
,
2336 struct xfs_dquot
**O_udqpp
,
2337 struct xfs_dquot
**O_gdqpp
)
2339 struct xfs_mount
*mp
= ip
->i_mount
;
2340 struct xfs_dquot
*uq
, *gq
;
2344 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
2347 lockflags
= XFS_ILOCK_EXCL
;
2348 xfs_ilock(ip
, lockflags
);
2350 if ((flags
& XFS_QMOPT_INHERIT
) && XFS_INHERIT_GID(ip
))
2351 gid
= ip
->i_d
.di_gid
;
2354 * Attach the dquot(s) to this inode, doing a dquot allocation
2355 * if necessary. The dquot(s) will not be locked.
2357 if (XFS_NOT_DQATTACHED(mp
, ip
)) {
2358 error
= xfs_qm_dqattach_locked(ip
, XFS_QMOPT_DQALLOC
);
2360 xfs_iunlock(ip
, lockflags
);
2366 if ((flags
& XFS_QMOPT_UQUOTA
) && XFS_IS_UQUOTA_ON(mp
)) {
2367 if (ip
->i_d
.di_uid
!= uid
) {
2369 * What we need is the dquot that has this uid, and
2370 * if we send the inode to dqget, the uid of the inode
2371 * takes priority over what's sent in the uid argument.
2372 * We must unlock inode here before calling dqget if
2373 * we're not sending the inode, because otherwise
2374 * we'll deadlock by doing trans_reserve while
2377 xfs_iunlock(ip
, lockflags
);
2378 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
) uid
,
2383 ASSERT(error
!= ENOENT
);
2387 * Get the ilock in the right order.
2390 lockflags
= XFS_ILOCK_SHARED
;
2391 xfs_ilock(ip
, lockflags
);
2394 * Take an extra reference, because we'll return
2397 ASSERT(ip
->i_udquot
);
2404 if ((flags
& XFS_QMOPT_GQUOTA
) && XFS_IS_GQUOTA_ON(mp
)) {
2405 if (ip
->i_d
.di_gid
!= gid
) {
2406 xfs_iunlock(ip
, lockflags
);
2407 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)gid
,
2414 ASSERT(error
!= ENOENT
);
2418 lockflags
= XFS_ILOCK_SHARED
;
2419 xfs_ilock(ip
, lockflags
);
2421 ASSERT(ip
->i_gdquot
);
2427 } else if ((flags
& XFS_QMOPT_PQUOTA
) && XFS_IS_PQUOTA_ON(mp
)) {
2428 if (ip
->i_d
.di_projid
!= prid
) {
2429 xfs_iunlock(ip
, lockflags
);
2430 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)prid
,
2437 ASSERT(error
!= ENOENT
);
2441 lockflags
= XFS_ILOCK_SHARED
;
2442 xfs_ilock(ip
, lockflags
);
2444 ASSERT(ip
->i_gdquot
);
2452 xfs_dqtrace_entry_ino(uq
, "DQALLOC", ip
);
2454 xfs_iunlock(ip
, lockflags
);
2467 * Actually transfer ownership, and do dquot modifications.
2468 * These were already reserved.
2474 xfs_dquot_t
**IO_olddq
,
2477 xfs_dquot_t
*prevdq
;
2478 uint bfield
= XFS_IS_REALTIME_INODE(ip
) ?
2479 XFS_TRANS_DQ_RTBCOUNT
: XFS_TRANS_DQ_BCOUNT
;
2482 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2483 ASSERT(XFS_IS_QUOTA_RUNNING(ip
->i_mount
));
2488 ASSERT(prevdq
!= newdq
);
2490 xfs_trans_mod_dquot(tp
, prevdq
, bfield
, -(ip
->i_d
.di_nblocks
));
2491 xfs_trans_mod_dquot(tp
, prevdq
, XFS_TRANS_DQ_ICOUNT
, -1);
2493 /* the sparkling new dquot */
2494 xfs_trans_mod_dquot(tp
, newdq
, bfield
, ip
->i_d
.di_nblocks
);
2495 xfs_trans_mod_dquot(tp
, newdq
, XFS_TRANS_DQ_ICOUNT
, 1);
2498 * Take an extra reference, because the inode
2499 * is going to keep this dquot pointer even
2500 * after the trans_commit.
2504 xfs_dqunlock(newdq
);
2511 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
2514 xfs_qm_vop_chown_reserve(
2521 xfs_mount_t
*mp
= ip
->i_mount
;
2522 uint delblks
, blkflags
, prjflags
= 0;
2523 xfs_dquot_t
*unresudq
, *unresgdq
, *delblksudq
, *delblksgdq
;
2527 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2528 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
2530 delblks
= ip
->i_delayed_blks
;
2531 delblksudq
= delblksgdq
= unresudq
= unresgdq
= NULL
;
2532 blkflags
= XFS_IS_REALTIME_INODE(ip
) ?
2533 XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
;
2535 if (XFS_IS_UQUOTA_ON(mp
) && udqp
&&
2536 ip
->i_d
.di_uid
!= (uid_t
)be32_to_cpu(udqp
->q_core
.d_id
)) {
2539 * If there are delayed allocation blocks, then we have to
2540 * unreserve those from the old dquot, and add them to the
2544 ASSERT(ip
->i_udquot
);
2545 unresudq
= ip
->i_udquot
;
2548 if (XFS_IS_OQUOTA_ON(ip
->i_mount
) && gdqp
) {
2549 if (XFS_IS_PQUOTA_ON(ip
->i_mount
) &&
2550 ip
->i_d
.di_projid
!= be32_to_cpu(gdqp
->q_core
.d_id
))
2551 prjflags
= XFS_QMOPT_ENOSPC
;
2554 (XFS_IS_GQUOTA_ON(ip
->i_mount
) &&
2555 ip
->i_d
.di_gid
!= be32_to_cpu(gdqp
->q_core
.d_id
))) {
2558 ASSERT(ip
->i_gdquot
);
2559 unresgdq
= ip
->i_gdquot
;
2564 if ((error
= xfs_trans_reserve_quota_bydquots(tp
, ip
->i_mount
,
2565 delblksudq
, delblksgdq
, ip
->i_d
.di_nblocks
, 1,
2566 flags
| blkflags
| prjflags
)))
2570 * Do the delayed blks reservations/unreservations now. Since, these
2571 * are done without the help of a transaction, if a reservation fails
2572 * its previous reservations won't be automatically undone by trans
2573 * code. So, we have to do it manually here.
2577 * Do the reservations first. Unreservation can't fail.
2579 ASSERT(delblksudq
|| delblksgdq
);
2580 ASSERT(unresudq
|| unresgdq
);
2581 if ((error
= xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
2582 delblksudq
, delblksgdq
, (xfs_qcnt_t
)delblks
, 0,
2583 flags
| blkflags
| prjflags
)))
2585 xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
2586 unresudq
, unresgdq
, -((xfs_qcnt_t
)delblks
), 0,
2594 xfs_qm_vop_rename_dqattach(
2595 struct xfs_inode
**i_tab
)
2597 struct xfs_mount
*mp
= i_tab
[0]->i_mount
;
2600 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
2603 for (i
= 0; (i
< 4 && i_tab
[i
]); i
++) {
2604 struct xfs_inode
*ip
= i_tab
[i
];
2608 * Watch out for duplicate entries in the table.
2610 if (i
== 0 || ip
!= i_tab
[i
-1]) {
2611 if (XFS_NOT_DQATTACHED(mp
, ip
)) {
2612 error
= xfs_qm_dqattach(ip
, 0);
2622 xfs_qm_vop_create_dqattach(
2623 struct xfs_trans
*tp
,
2624 struct xfs_inode
*ip
,
2625 struct xfs_dquot
*udqp
,
2626 struct xfs_dquot
*gdqp
)
2628 struct xfs_mount
*mp
= tp
->t_mountp
;
2630 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
2633 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2634 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
2640 ASSERT(ip
->i_udquot
== NULL
);
2641 ip
->i_udquot
= udqp
;
2642 ASSERT(XFS_IS_UQUOTA_ON(mp
));
2643 ASSERT(ip
->i_d
.di_uid
== be32_to_cpu(udqp
->q_core
.d_id
));
2644 xfs_trans_mod_dquot(tp
, udqp
, XFS_TRANS_DQ_ICOUNT
, 1);
2650 ASSERT(ip
->i_gdquot
== NULL
);
2651 ip
->i_gdquot
= gdqp
;
2652 ASSERT(XFS_IS_OQUOTA_ON(mp
));
2653 ASSERT((XFS_IS_GQUOTA_ON(mp
) ?
2654 ip
->i_d
.di_gid
: ip
->i_d
.di_projid
) ==
2655 be32_to_cpu(gdqp
->q_core
.d_id
));
2656 xfs_trans_mod_dquot(tp
, gdqp
, XFS_TRANS_DQ_ICOUNT
, 1);
2660 /* ------------- list stuff -----------------*/
2662 xfs_qm_freelist_init(xfs_frlist_t
*ql
)
2664 ql
->qh_next
= ql
->qh_prev
= (xfs_dquot_t
*) ql
;
2665 mutex_init(&ql
->qh_lock
);
2671 xfs_qm_freelist_destroy(xfs_frlist_t
*ql
)
2673 xfs_dquot_t
*dqp
, *nextdqp
;
2675 mutex_lock(&ql
->qh_lock
);
2676 for (dqp
= ql
->qh_next
;
2677 dqp
!= (xfs_dquot_t
*)ql
; ) {
2679 nextdqp
= dqp
->dq_flnext
;
2681 cmn_err(CE_DEBUG
, "FREELIST destroy 0x%p", dqp
);
2683 XQM_FREELIST_REMOVE(dqp
);
2685 xfs_qm_dqdestroy(dqp
);
2688 mutex_unlock(&ql
->qh_lock
);
2689 mutex_destroy(&ql
->qh_lock
);
2691 ASSERT(ql
->qh_nelems
== 0);
2695 xfs_qm_freelist_insert(xfs_frlist_t
*ql
, xfs_dquot_t
*dq
)
2697 dq
->dq_flnext
= ql
->qh_next
;
2698 dq
->dq_flprev
= (xfs_dquot_t
*)ql
;
2700 dq
->dq_flnext
->dq_flprev
= dq
;
2701 xfs_Gqm
->qm_dqfreelist
.qh_nelems
++;
2702 xfs_Gqm
->qm_dqfreelist
.qh_version
++;
2706 xfs_qm_freelist_unlink(xfs_dquot_t
*dq
)
2708 xfs_dquot_t
*next
= dq
->dq_flnext
;
2709 xfs_dquot_t
*prev
= dq
->dq_flprev
;
2711 next
->dq_flprev
= prev
;
2712 prev
->dq_flnext
= next
;
2713 dq
->dq_flnext
= dq
->dq_flprev
= dq
;
2714 xfs_Gqm
->qm_dqfreelist
.qh_nelems
--;
2715 xfs_Gqm
->qm_dqfreelist
.qh_version
++;
2719 xfs_qm_freelist_append(xfs_frlist_t
*ql
, xfs_dquot_t
*dq
)
2721 xfs_qm_freelist_insert((xfs_frlist_t
*)ql
->qh_prev
, dq
);