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"
46 #include "xfs_buf_item.h"
47 #include "xfs_trans_space.h"
48 #include "xfs_utils.h"
50 #include "xfs_trace.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
;
88 xfs_qm_dquot_list_print(
94 list_for_each_entry(dqp
, &mp
->m_quotainfo
->qi_dqlist_lock
, qi_mplist
) {
95 cmn_err(CE_DEBUG
, " %d. \"%d (%s)\" "
96 "bcnt = %lld, icnt = %lld, refs = %d",
97 i
++, be32_to_cpu(dqp
->q_core
.d_id
),
98 DQFLAGTO_TYPESTR(dqp
),
99 (long long)be64_to_cpu(dqp
->q_core
.d_bcount
),
100 (long long)be64_to_cpu(dqp
->q_core
.d_icount
),
105 static void xfs_qm_dquot_list_print(struct xfs_mount
*mp
) { }
109 * Initialize the XQM structure.
110 * Note that there is not one quota manager per file system.
112 STATIC
struct xfs_qm
*
115 xfs_dqhash_t
*udqhash
, *gdqhash
;
121 * Initialize the dquot hash tables.
123 udqhash
= kmem_zalloc_greedy(&hsize
,
124 XFS_QM_HASHSIZE_LOW
* sizeof(xfs_dqhash_t
),
125 XFS_QM_HASHSIZE_HIGH
* sizeof(xfs_dqhash_t
));
129 gdqhash
= kmem_zalloc_large(hsize
);
131 goto out_free_udqhash
;
133 hsize
/= sizeof(xfs_dqhash_t
);
136 xqm
= kmem_zalloc(sizeof(xfs_qm_t
), KM_SLEEP
);
137 xqm
->qm_dqhashmask
= hsize
- 1;
138 xqm
->qm_usr_dqhtable
= udqhash
;
139 xqm
->qm_grp_dqhtable
= gdqhash
;
140 ASSERT(xqm
->qm_usr_dqhtable
!= NULL
);
141 ASSERT(xqm
->qm_grp_dqhtable
!= NULL
);
143 for (i
= 0; i
< hsize
; i
++) {
144 xfs_qm_list_init(&(xqm
->qm_usr_dqhtable
[i
]), "uxdqh", i
);
145 xfs_qm_list_init(&(xqm
->qm_grp_dqhtable
[i
]), "gxdqh", i
);
149 * Freelist of all dquots of all file systems
151 xfs_qm_freelist_init(&(xqm
->qm_dqfreelist
));
154 * dquot zone. we register our own low-memory callback.
157 xqm
->qm_dqzone
= kmem_zone_init(sizeof(xfs_dquot_t
),
159 qm_dqzone
= xqm
->qm_dqzone
;
161 xqm
->qm_dqzone
= qm_dqzone
;
163 register_shrinker(&xfs_qm_shaker
);
166 * The t_dqinfo portion of transactions.
169 xqm
->qm_dqtrxzone
= kmem_zone_init(sizeof(xfs_dquot_acct_t
),
171 qm_dqtrxzone
= xqm
->qm_dqtrxzone
;
173 xqm
->qm_dqtrxzone
= qm_dqtrxzone
;
175 atomic_set(&xqm
->qm_totaldquots
, 0);
176 xqm
->qm_dqfree_ratio
= XFS_QM_DQFREE_RATIO
;
179 mutex_init(&qcheck_lock
);
184 kmem_free_large(udqhash
);
190 * Destroy the global quota manager when its reference count goes to zero.
199 ASSERT(xqm
->qm_nrefs
== 0);
200 unregister_shrinker(&xfs_qm_shaker
);
201 hsize
= xqm
->qm_dqhashmask
+ 1;
202 for (i
= 0; i
< hsize
; i
++) {
203 xfs_qm_list_destroy(&(xqm
->qm_usr_dqhtable
[i
]));
204 xfs_qm_list_destroy(&(xqm
->qm_grp_dqhtable
[i
]));
206 kmem_free_large(xqm
->qm_usr_dqhtable
);
207 kmem_free_large(xqm
->qm_grp_dqhtable
);
208 xqm
->qm_usr_dqhtable
= NULL
;
209 xqm
->qm_grp_dqhtable
= NULL
;
210 xqm
->qm_dqhashmask
= 0;
211 xfs_qm_freelist_destroy(&(xqm
->qm_dqfreelist
));
213 mutex_destroy(&qcheck_lock
);
219 * Called at mount time to let XQM know that another file system is
220 * starting quotas. This isn't crucial information as the individual mount
221 * structures are pretty independent, but it helps the XQM keep a
222 * global view of what's going on.
226 xfs_qm_hold_quotafs_ref(
227 struct xfs_mount
*mp
)
230 * Need to lock the xfs_Gqm structure for things like this. For example,
231 * the structure could disappear between the entry to this routine and
232 * a HOLD operation if not locked.
234 mutex_lock(&xfs_Gqm_lock
);
237 xfs_Gqm
= xfs_Gqm_init();
243 * We can keep a list of all filesystems with quotas mounted for
244 * debugging and statistical purposes, but ...
245 * Just take a reference and get out.
248 mutex_unlock(&xfs_Gqm_lock
);
255 * Release the reference that a filesystem took at mount time,
256 * so that we know when we need to destroy the entire quota manager.
260 xfs_qm_rele_quotafs_ref(
261 struct xfs_mount
*mp
)
263 xfs_dquot_t
*dqp
, *nextdqp
;
266 ASSERT(xfs_Gqm
->qm_nrefs
> 0);
269 * Go thru the freelist and destroy all inactive dquots.
271 xfs_qm_freelist_lock(xfs_Gqm
);
273 for (dqp
= xfs_Gqm
->qm_dqfreelist
.qh_next
;
274 dqp
!= (xfs_dquot_t
*)&(xfs_Gqm
->qm_dqfreelist
); ) {
276 nextdqp
= dqp
->dq_flnext
;
277 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
278 ASSERT(dqp
->q_mount
== NULL
);
279 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
280 ASSERT(dqp
->HL_PREVP
== NULL
);
281 ASSERT(list_empty(&dqp
->q_mplist
));
282 XQM_FREELIST_REMOVE(dqp
);
284 xfs_qm_dqdestroy(dqp
);
290 xfs_qm_freelist_unlock(xfs_Gqm
);
293 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
296 mutex_lock(&xfs_Gqm_lock
);
297 if (--xfs_Gqm
->qm_nrefs
== 0) {
298 xfs_qm_destroy(xfs_Gqm
);
301 mutex_unlock(&xfs_Gqm_lock
);
305 * Just destroy the quotainfo structure.
309 struct xfs_mount
*mp
)
311 if (mp
->m_quotainfo
) {
312 xfs_qm_dqpurge_all(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_UMOUNTING
);
313 xfs_qm_destroy_quotainfo(mp
);
319 * This is called from xfs_mountfs to start quotas and initialize all
320 * necessary data structures like quotainfo. This is also responsible for
321 * running a quotacheck as necessary. We are guaranteed that the superblock
322 * is consistently read in at this point.
324 * If we fail here, the mount will continue with quota turned off. We don't
325 * need to inidicate success or failure at all.
335 * If quotas on realtime volumes is not supported, we disable
336 * quotas immediately.
338 if (mp
->m_sb
.sb_rextents
) {
340 "Cannot turn on quotas for realtime filesystem %s",
346 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
349 * Allocate the quotainfo structure inside the mount struct, and
350 * create quotainode(s), and change/rev superblock if necessary.
352 error
= xfs_qm_init_quotainfo(mp
);
355 * We must turn off quotas.
357 ASSERT(mp
->m_quotainfo
== NULL
);
362 * If any of the quotas are not consistent, do a quotacheck.
364 if (XFS_QM_NEED_QUOTACHECK(mp
)) {
365 error
= xfs_qm_quotacheck(mp
);
367 /* Quotacheck failed and disabled quotas. */
372 * If one type of quotas is off, then it will lose its
373 * quotachecked status, since we won't be doing accounting for
376 if (!XFS_IS_UQUOTA_ON(mp
))
377 mp
->m_qflags
&= ~XFS_UQUOTA_CHKD
;
378 if (!(XFS_IS_GQUOTA_ON(mp
) || XFS_IS_PQUOTA_ON(mp
)))
379 mp
->m_qflags
&= ~XFS_OQUOTA_CHKD
;
383 * We actually don't have to acquire the m_sb_lock at all.
384 * This can only be called from mount, and that's single threaded. XXX
386 spin_lock(&mp
->m_sb_lock
);
387 sbf
= mp
->m_sb
.sb_qflags
;
388 mp
->m_sb
.sb_qflags
= mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
;
389 spin_unlock(&mp
->m_sb_lock
);
391 if (sbf
!= (mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
)) {
392 if (xfs_qm_write_sb_changes(mp
, XFS_SB_QFLAGS
)) {
394 * We could only have been turning quotas off.
395 * We aren't in very good shape actually because
396 * the incore structures are convinced that quotas are
397 * off, but the on disk superblock doesn't know that !
399 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp
)));
400 xfs_fs_cmn_err(CE_ALERT
, mp
,
401 "XFS mount_quotas: Superblock update failed!");
406 xfs_fs_cmn_err(CE_WARN
, mp
,
407 "Failed to initialize disk quotas.");
412 if (XFS_IS_QUOTA_ON(mp
))
413 xfs_qm_internalqcheck(mp
);
418 * Called from the vfsops layer.
421 xfs_qm_unmount_quotas(
425 * Release the dquots that root inode, et al might be holding,
426 * before we flush quotas and blow away the quotainfo structure.
428 ASSERT(mp
->m_rootip
);
429 xfs_qm_dqdetach(mp
->m_rootip
);
431 xfs_qm_dqdetach(mp
->m_rbmip
);
433 xfs_qm_dqdetach(mp
->m_rsumip
);
436 * Release the quota inodes.
438 if (mp
->m_quotainfo
) {
439 if (mp
->m_quotainfo
->qi_uquotaip
) {
440 IRELE(mp
->m_quotainfo
->qi_uquotaip
);
441 mp
->m_quotainfo
->qi_uquotaip
= NULL
;
443 if (mp
->m_quotainfo
->qi_gquotaip
) {
444 IRELE(mp
->m_quotainfo
->qi_gquotaip
);
445 mp
->m_quotainfo
->qi_gquotaip
= NULL
;
451 * Flush all dquots of the given file system to disk. The dquots are
452 * _not_ purged from memory here, just their data written to disk.
464 if (mp
->m_quotainfo
== NULL
)
468 mutex_lock(&mp
->m_quotainfo
->qi_dqlist_lock
);
469 list_for_each_entry(dqp
, &mp
->m_quotainfo
->qi_dqlist
, q_mplist
) {
471 if (! XFS_DQ_IS_DIRTY(dqp
)) {
476 /* XXX a sentinel would be better */
477 recl
= mp
->m_quotainfo
->qi_dqreclaims
;
478 if (!xfs_dqflock_nowait(dqp
)) {
480 * If we can't grab the flush lock then check
481 * to see if the dquot has been flushed delayed
482 * write. If so, grab its buffer and send it
483 * out immediately. We'll be able to acquire
484 * the flush lock when the I/O completes.
486 xfs_qm_dqflock_pushbuf_wait(dqp
);
489 * Let go of the mplist lock. We don't want to hold it
490 * across a disk write.
492 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
493 error
= xfs_qm_dqflush(dqp
, sync_mode
);
498 mutex_lock(&mp
->m_quotainfo
->qi_dqlist_lock
);
499 if (recl
!= mp
->m_quotainfo
->qi_dqreclaims
) {
500 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
501 /* XXX restart limit */
506 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
511 * Release the group dquot pointers the user dquots may be
512 * carrying around as a hint. mplist is locked on entry and exit.
515 xfs_qm_detach_gdquots(
518 xfs_dquot_t
*dqp
, *gdqp
;
522 ASSERT(mutex_is_locked(&mp
->m_quotainfo
->qi_dqlist_lock
));
523 list_for_each_entry(dqp
, &mp
->m_quotainfo
->qi_dqlist
, q_mplist
) {
525 if ((gdqp
= dqp
->q_gdquot
)) {
527 dqp
->q_gdquot
= NULL
;
533 * Can't hold the mplist lock across a dqput.
534 * XXXmust convert to marker based iterations here.
536 nrecl
= mp
->m_quotainfo
->qi_dqreclaims
;
537 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
540 mutex_lock(&mp
->m_quotainfo
->qi_dqlist_lock
);
541 if (nrecl
!= mp
->m_quotainfo
->qi_dqreclaims
)
548 * Go through all the incore dquots of this file system and take them
549 * off the mplist and hashlist, if the dquot type matches the dqtype
550 * parameter. This is used when turning off quota accounting for
551 * users and/or groups, as well as when the filesystem is unmounting.
556 uint flags
) /* QUOTAOFF/UMOUNTING/UQUOTA/PQUOTA/GQUOTA */
558 xfs_dquot_t
*dqp
, *n
;
563 if (mp
->m_quotainfo
== NULL
)
566 dqtype
= (flags
& XFS_QMOPT_UQUOTA
) ? XFS_DQ_USER
: 0;
567 dqtype
|= (flags
& XFS_QMOPT_PQUOTA
) ? XFS_DQ_PROJ
: 0;
568 dqtype
|= (flags
& XFS_QMOPT_GQUOTA
) ? XFS_DQ_GROUP
: 0;
570 mutex_lock(&mp
->m_quotainfo
->qi_dqlist_lock
);
573 * In the first pass through all incore dquots of this filesystem,
574 * we release the group dquot pointers the user dquots may be
575 * carrying around as a hint. We need to do this irrespective of
576 * what's being turned off.
578 xfs_qm_detach_gdquots(mp
);
582 ASSERT(mutex_is_locked(&mp
->m_quotainfo
->qi_dqlist_lock
));
584 * Try to get rid of all of the unwanted dquots. The idea is to
585 * get them off mplist and hashlist, but leave them on freelist.
587 list_for_each_entry_safe(dqp
, n
, &mp
->m_quotainfo
->qi_dqlist
, q_mplist
) {
589 * It's OK to look at the type without taking dqlock here.
590 * We're holding the mplist lock here, and that's needed for
593 if ((dqp
->dq_flags
& dqtype
) == 0)
596 if (!mutex_trylock(&dqp
->q_hash
->qh_lock
)) {
597 nrecl
= mp
->m_quotainfo
->qi_dqreclaims
;
598 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
599 mutex_lock(&dqp
->q_hash
->qh_lock
);
600 mutex_lock(&mp
->m_quotainfo
->qi_dqlist_lock
);
603 * XXXTheoretically, we can get into a very long
604 * ping pong game here.
605 * No one can be adding dquots to the mplist at
606 * this point, but somebody might be taking things off.
608 if (nrecl
!= mp
->m_quotainfo
->qi_dqreclaims
) {
609 mutex_unlock(&dqp
->q_hash
->qh_lock
);
615 * Take the dquot off the mplist and hashlist. It may remain on
616 * freelist in INACTIVE state.
618 nmisses
+= xfs_qm_dqpurge(dqp
);
620 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
632 * Purge the dquot cache.
633 * None of the dquots should really be busy at this point.
635 if (mp
->m_quotainfo
) {
636 while ((ndquots
= xfs_qm_dqpurge_int(mp
, flags
))) {
649 xfs_dquot_t
*udqhint
, /* hint */
650 xfs_dquot_t
**IO_idqpp
)
655 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
659 * See if we already have it in the inode itself. IO_idqpp is
660 * &i_udquot or &i_gdquot. This made the code look weird, but
661 * made the logic a lot simpler.
665 trace_xfs_dqattach_found(dqp
);
670 * udqhint is the i_udquot field in inode, and is non-NULL only
671 * when the type arg is group/project. Its purpose is to save a
672 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
676 ASSERT(type
== XFS_DQ_GROUP
|| type
== XFS_DQ_PROJ
);
680 * No need to take dqlock to look at the id.
682 * The ID can't change until it gets reclaimed, and it won't
683 * be reclaimed as long as we have a ref from inode and we
686 dqp
= udqhint
->q_gdquot
;
687 if (dqp
&& be32_to_cpu(dqp
->q_core
.d_id
) == id
) {
690 ASSERT(*IO_idqpp
== NULL
);
694 xfs_dqunlock(udqhint
);
699 * We can't hold a dquot lock when we call the dqget code.
700 * We'll deadlock in no time, because of (not conforming to)
701 * lock ordering - the inodelock comes before any dquot lock,
702 * and we may drop and reacquire the ilock in xfs_qm_dqget().
704 xfs_dqunlock(udqhint
);
708 * Find the dquot from somewhere. This bumps the
709 * reference count of dquot and returns it locked.
710 * This can return ENOENT if dquot didn't exist on
711 * disk and we didn't ask it to allocate;
712 * ESRCH if quotas got turned off suddenly.
714 error
= xfs_qm_dqget(ip
->i_mount
, ip
, id
, type
, XFS_QMOPT_DOWARN
, &dqp
);
718 trace_xfs_dqattach_get(dqp
);
721 * dqget may have dropped and re-acquired the ilock, but it guarantees
722 * that the dquot returned is the one that should go in the inode.
731 * Given a udquot and gdquot, attach a ptr to the group dquot in the
732 * udquot as a hint for future lookups. The idea sounds simple, but the
733 * execution isn't, because the udquot might have a group dquot attached
734 * already and getting rid of that gets us into lock ordering constraints.
735 * The process is complicated more by the fact that the dquots may or may not
736 * be locked on entry.
739 xfs_qm_dqattach_grouphint(
747 if ((tmp
= udq
->q_gdquot
)) {
753 udq
->q_gdquot
= NULL
;
755 * We can't keep any dqlocks when calling dqrele,
756 * because the freelist lock comes before dqlocks.
760 * we took a hard reference once upon a time in dqget,
761 * so give it back when the udquot no longer points at it
762 * dqput() does the unlocking of the dquot.
770 ASSERT(XFS_DQ_IS_LOCKED(udq
));
774 ASSERT(XFS_DQ_IS_LOCKED(udq
));
775 ASSERT(XFS_DQ_IS_LOCKED(gdq
));
777 * Somebody could have attached a gdquot here,
778 * when we dropped the uqlock. If so, just do nothing.
780 if (udq
->q_gdquot
== NULL
) {
791 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
793 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
794 * Inode may get unlocked and relocked in here, and the caller must deal with
798 xfs_qm_dqattach_locked(
802 xfs_mount_t
*mp
= ip
->i_mount
;
806 if (!XFS_IS_QUOTA_RUNNING(mp
) ||
807 !XFS_IS_QUOTA_ON(mp
) ||
808 !XFS_NOT_DQATTACHED(mp
, ip
) ||
809 ip
->i_ino
== mp
->m_sb
.sb_uquotino
||
810 ip
->i_ino
== mp
->m_sb
.sb_gquotino
)
813 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
815 if (XFS_IS_UQUOTA_ON(mp
)) {
816 error
= xfs_qm_dqattach_one(ip
, ip
->i_d
.di_uid
, XFS_DQ_USER
,
817 flags
& XFS_QMOPT_DQALLOC
,
818 NULL
, &ip
->i_udquot
);
824 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
825 if (XFS_IS_OQUOTA_ON(mp
)) {
826 error
= XFS_IS_GQUOTA_ON(mp
) ?
827 xfs_qm_dqattach_one(ip
, ip
->i_d
.di_gid
, XFS_DQ_GROUP
,
828 flags
& XFS_QMOPT_DQALLOC
,
829 ip
->i_udquot
, &ip
->i_gdquot
) :
830 xfs_qm_dqattach_one(ip
, ip
->i_d
.di_projid
, XFS_DQ_PROJ
,
831 flags
& XFS_QMOPT_DQALLOC
,
832 ip
->i_udquot
, &ip
->i_gdquot
);
834 * Don't worry about the udquot that we may have
835 * attached above. It'll get detached, if not already.
843 * Attach this group quota to the user quota as a hint.
844 * This WON'T, in general, result in a thrash.
847 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
848 ASSERT(ip
->i_udquot
);
849 ASSERT(ip
->i_gdquot
);
852 * We may or may not have the i_udquot locked at this point,
853 * but this check is OK since we don't depend on the i_gdquot to
854 * be accurate 100% all the time. It is just a hint, and this
855 * will succeed in general.
857 if (ip
->i_udquot
->q_gdquot
== ip
->i_gdquot
)
860 * Attach i_gdquot to the gdquot hint inside the i_udquot.
862 xfs_qm_dqattach_grouphint(ip
->i_udquot
, ip
->i_gdquot
);
868 if (XFS_IS_UQUOTA_ON(mp
))
869 ASSERT(ip
->i_udquot
);
870 if (XFS_IS_OQUOTA_ON(mp
))
871 ASSERT(ip
->i_gdquot
);
873 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
880 struct xfs_inode
*ip
,
885 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
886 error
= xfs_qm_dqattach_locked(ip
, flags
);
887 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
893 * Release dquots (and their references) if any.
894 * The inode should be locked EXCL except when this's called by
901 if (!(ip
->i_udquot
|| ip
->i_gdquot
))
904 trace_xfs_dquot_dqdetach(ip
);
906 ASSERT(ip
->i_ino
!= ip
->i_mount
->m_sb
.sb_uquotino
);
907 ASSERT(ip
->i_ino
!= ip
->i_mount
->m_sb
.sb_gquotino
);
909 xfs_qm_dqrele(ip
->i_udquot
);
913 xfs_qm_dqrele(ip
->i_gdquot
);
927 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
933 mutex_lock(&mp
->m_quotainfo
->qi_dqlist_lock
);
935 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
936 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
937 * when we have the mplist lock, we know that dquots will be consistent
938 * as long as we have it locked.
940 if (!XFS_IS_QUOTA_ON(mp
)) {
941 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
944 ASSERT(mutex_is_locked(&mp
->m_quotainfo
->qi_dqlist_lock
));
945 list_for_each_entry(dqp
, &mp
->m_quotainfo
->qi_dqlist
, q_mplist
) {
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_TRYLOCK
) {
952 if (!XFS_DQ_IS_DIRTY(dqp
))
954 if (!xfs_qm_dqlock_nowait(dqp
))
961 * Now, find out for sure if this dquot is dirty or not.
963 if (! XFS_DQ_IS_DIRTY(dqp
)) {
968 /* XXX a sentinel would be better */
969 recl
= mp
->m_quotainfo
->qi_dqreclaims
;
970 if (!xfs_dqflock_nowait(dqp
)) {
971 if (flags
& SYNC_TRYLOCK
) {
976 * If we can't grab the flush lock then if the caller
977 * really wanted us to give this our best shot, so
978 * see if we can give a push to the buffer before we wait
979 * on the flush lock. At this point, we know that
980 * even though the dquot is being flushed,
981 * it has (new) dirty data.
983 xfs_qm_dqflock_pushbuf_wait(dqp
);
986 * Let go of the mplist lock. We don't want to hold it
987 * across a disk write
989 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
990 error
= xfs_qm_dqflush(dqp
, flags
);
992 if (error
&& XFS_FORCED_SHUTDOWN(mp
))
993 return 0; /* Need to prevent umount failure */
997 mutex_lock(&mp
->m_quotainfo
->qi_dqlist_lock
);
998 if (recl
!= mp
->m_quotainfo
->qi_dqreclaims
) {
999 if (++restarts
>= XFS_QM_SYNC_MAX_RESTARTS
)
1002 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
1007 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
1012 * The hash chains and the mplist use the same xfs_dqhash structure as
1013 * their list head, but we can take the mplist qh_lock and one of the
1014 * hash qh_locks at the same time without any problem as they aren't
1017 static struct lock_class_key xfs_quota_mplist_class
;
1020 * This initializes all the quota information that's kept in the
1024 xfs_qm_init_quotainfo(
1027 xfs_quotainfo_t
*qinf
;
1031 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1034 * Tell XQM that we exist as soon as possible.
1036 if ((error
= xfs_qm_hold_quotafs_ref(mp
))) {
1040 qinf
= mp
->m_quotainfo
= kmem_zalloc(sizeof(xfs_quotainfo_t
), KM_SLEEP
);
1043 * See if quotainodes are setup, and if not, allocate them,
1044 * and change the superblock accordingly.
1046 if ((error
= xfs_qm_init_quotainos(mp
))) {
1048 mp
->m_quotainfo
= NULL
;
1052 INIT_LIST_HEAD(&qinf
->qi_dqlist
);
1053 mutex_init(&qinf
->qi_dqlist_lock
);
1054 lockdep_set_class(&qinf
->qi_dqlist_lock
, &xfs_quota_mplist_class
);
1056 qinf
->qi_dqreclaims
= 0;
1058 /* mutex used to serialize quotaoffs */
1059 mutex_init(&qinf
->qi_quotaofflock
);
1061 /* Precalc some constants */
1062 qinf
->qi_dqchunklen
= XFS_FSB_TO_BB(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
1063 ASSERT(qinf
->qi_dqchunklen
);
1064 qinf
->qi_dqperchunk
= BBTOB(qinf
->qi_dqchunklen
);
1065 do_div(qinf
->qi_dqperchunk
, sizeof(xfs_dqblk_t
));
1067 mp
->m_qflags
|= (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_CHKD
);
1070 * We try to get the limits from the superuser's limits fields.
1071 * This is quite hacky, but it is standard quota practice.
1072 * We look at the USR dquot with id == 0 first, but if user quotas
1073 * are not enabled we goto the GRP dquot with id == 0.
1074 * We don't really care to keep separate default limits for user
1075 * and group quotas, at least not at this point.
1077 error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)0,
1078 XFS_IS_UQUOTA_RUNNING(mp
) ? XFS_DQ_USER
:
1079 (XFS_IS_GQUOTA_RUNNING(mp
) ? XFS_DQ_GROUP
:
1081 XFS_QMOPT_DQSUSER
|XFS_QMOPT_DOWARN
,
1084 xfs_disk_dquot_t
*ddqp
= &dqp
->q_core
;
1087 * The warnings and timers set the grace period given to
1088 * a user or group before he or she can not perform any
1089 * more writing. If it is zero, a default is used.
1091 qinf
->qi_btimelimit
= ddqp
->d_btimer
?
1092 be32_to_cpu(ddqp
->d_btimer
) : XFS_QM_BTIMELIMIT
;
1093 qinf
->qi_itimelimit
= ddqp
->d_itimer
?
1094 be32_to_cpu(ddqp
->d_itimer
) : XFS_QM_ITIMELIMIT
;
1095 qinf
->qi_rtbtimelimit
= ddqp
->d_rtbtimer
?
1096 be32_to_cpu(ddqp
->d_rtbtimer
) : XFS_QM_RTBTIMELIMIT
;
1097 qinf
->qi_bwarnlimit
= ddqp
->d_bwarns
?
1098 be16_to_cpu(ddqp
->d_bwarns
) : XFS_QM_BWARNLIMIT
;
1099 qinf
->qi_iwarnlimit
= ddqp
->d_iwarns
?
1100 be16_to_cpu(ddqp
->d_iwarns
) : XFS_QM_IWARNLIMIT
;
1101 qinf
->qi_rtbwarnlimit
= ddqp
->d_rtbwarns
?
1102 be16_to_cpu(ddqp
->d_rtbwarns
) : XFS_QM_RTBWARNLIMIT
;
1103 qinf
->qi_bhardlimit
= be64_to_cpu(ddqp
->d_blk_hardlimit
);
1104 qinf
->qi_bsoftlimit
= be64_to_cpu(ddqp
->d_blk_softlimit
);
1105 qinf
->qi_ihardlimit
= be64_to_cpu(ddqp
->d_ino_hardlimit
);
1106 qinf
->qi_isoftlimit
= be64_to_cpu(ddqp
->d_ino_softlimit
);
1107 qinf
->qi_rtbhardlimit
= be64_to_cpu(ddqp
->d_rtb_hardlimit
);
1108 qinf
->qi_rtbsoftlimit
= be64_to_cpu(ddqp
->d_rtb_softlimit
);
1111 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1112 * we don't want this dquot cached. We haven't done a
1113 * quotacheck yet, and quotacheck doesn't like incore dquots.
1115 xfs_qm_dqdestroy(dqp
);
1117 qinf
->qi_btimelimit
= XFS_QM_BTIMELIMIT
;
1118 qinf
->qi_itimelimit
= XFS_QM_ITIMELIMIT
;
1119 qinf
->qi_rtbtimelimit
= XFS_QM_RTBTIMELIMIT
;
1120 qinf
->qi_bwarnlimit
= XFS_QM_BWARNLIMIT
;
1121 qinf
->qi_iwarnlimit
= XFS_QM_IWARNLIMIT
;
1122 qinf
->qi_rtbwarnlimit
= XFS_QM_RTBWARNLIMIT
;
1130 * Gets called when unmounting a filesystem or when all quotas get
1132 * This purges the quota inodes, destroys locks and frees itself.
1135 xfs_qm_destroy_quotainfo(
1138 xfs_quotainfo_t
*qi
;
1140 qi
= mp
->m_quotainfo
;
1142 ASSERT(xfs_Gqm
!= NULL
);
1145 * Release the reference that XQM kept, so that we know
1146 * when the XQM structure should be freed. We cannot assume
1147 * that xfs_Gqm is non-null after this point.
1149 xfs_qm_rele_quotafs_ref(mp
);
1151 ASSERT(list_empty(&qi
->qi_dqlist
));
1152 mutex_destroy(&qi
->qi_dqlist_lock
);
1154 if (qi
->qi_uquotaip
) {
1155 IRELE(qi
->qi_uquotaip
);
1156 qi
->qi_uquotaip
= NULL
; /* paranoia */
1158 if (qi
->qi_gquotaip
) {
1159 IRELE(qi
->qi_gquotaip
);
1160 qi
->qi_gquotaip
= NULL
;
1162 mutex_destroy(&qi
->qi_quotaofflock
);
1164 mp
->m_quotainfo
= NULL
;
1169 /* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1178 mutex_init(&list
->qh_lock
);
1179 list
->qh_next
= NULL
;
1180 list
->qh_version
= 0;
1181 list
->qh_nelems
= 0;
1185 xfs_qm_list_destroy(
1188 mutex_destroy(&(list
->qh_lock
));
1193 * Stripped down version of dqattach. This doesn't attach, or even look at the
1194 * dquots attached to the inode. The rationale is that there won't be any
1195 * attached at the time this is called from quotacheck.
1198 xfs_qm_dqget_noattach(
1200 xfs_dquot_t
**O_udqpp
,
1201 xfs_dquot_t
**O_gdqpp
)
1205 xfs_dquot_t
*udqp
, *gdqp
;
1207 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1212 if (XFS_IS_UQUOTA_ON(mp
)) {
1213 ASSERT(ip
->i_udquot
== NULL
);
1215 * We want the dquot allocated if it doesn't exist.
1217 if ((error
= xfs_qm_dqget(mp
, ip
, ip
->i_d
.di_uid
, XFS_DQ_USER
,
1218 XFS_QMOPT_DQALLOC
| XFS_QMOPT_DOWARN
,
1221 * Shouldn't be able to turn off quotas here.
1223 ASSERT(error
!= ESRCH
);
1224 ASSERT(error
!= ENOENT
);
1230 if (XFS_IS_OQUOTA_ON(mp
)) {
1231 ASSERT(ip
->i_gdquot
== NULL
);
1234 error
= XFS_IS_GQUOTA_ON(mp
) ?
1235 xfs_qm_dqget(mp
, ip
,
1236 ip
->i_d
.di_gid
, XFS_DQ_GROUP
,
1237 XFS_QMOPT_DQALLOC
|XFS_QMOPT_DOWARN
,
1239 xfs_qm_dqget(mp
, ip
,
1240 ip
->i_d
.di_projid
, XFS_DQ_PROJ
,
1241 XFS_QMOPT_DQALLOC
|XFS_QMOPT_DOWARN
,
1245 xfs_qm_dqrele(udqp
);
1246 ASSERT(error
!= ESRCH
);
1247 ASSERT(error
!= ENOENT
);
1252 /* Reacquire the locks in the right order */
1254 if (! xfs_qm_dqlock_nowait(udqp
)) {
1266 if (udqp
) ASSERT(XFS_DQ_IS_LOCKED(udqp
));
1267 if (gdqp
) ASSERT(XFS_DQ_IS_LOCKED(gdqp
));
1273 * Create an inode and return with a reference already taken, but unlocked
1274 * This is how we create quota inodes
1287 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_QINOCREATE
);
1288 if ((error
= xfs_trans_reserve(tp
,
1289 XFS_QM_QINOCREATE_SPACE_RES(mp
),
1290 XFS_CREATE_LOG_RES(mp
), 0,
1291 XFS_TRANS_PERM_LOG_RES
,
1292 XFS_CREATE_LOG_COUNT
))) {
1293 xfs_trans_cancel(tp
, 0);
1297 if ((error
= xfs_dir_ialloc(&tp
, NULL
, S_IFREG
, 1, 0,
1298 &xfs_zerocr
, 0, 1, ip
, &committed
))) {
1299 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1305 * Keep an extra reference to this quota inode. This inode is
1306 * locked exclusively and joined to the transaction already.
1308 ASSERT(xfs_isilocked(*ip
, XFS_ILOCK_EXCL
));
1312 * Make the changes in the superblock, and log those too.
1313 * sbfields arg may contain fields other than *QUOTINO;
1314 * VERSIONNUM for example.
1316 spin_lock(&mp
->m_sb_lock
);
1317 if (flags
& XFS_QMOPT_SBVERSION
) {
1318 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1319 unsigned oldv
= mp
->m_sb
.sb_versionnum
;
1321 ASSERT(!xfs_sb_version_hasquota(&mp
->m_sb
));
1322 ASSERT((sbfields
& (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1323 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
)) ==
1324 (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1325 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
));
1327 xfs_sb_version_addquota(&mp
->m_sb
);
1328 mp
->m_sb
.sb_uquotino
= NULLFSINO
;
1329 mp
->m_sb
.sb_gquotino
= NULLFSINO
;
1331 /* qflags will get updated _after_ quotacheck */
1332 mp
->m_sb
.sb_qflags
= 0;
1333 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1335 "Old superblock version %x, converting to %x.",
1336 oldv
, mp
->m_sb
.sb_versionnum
);
1339 if (flags
& XFS_QMOPT_UQUOTA
)
1340 mp
->m_sb
.sb_uquotino
= (*ip
)->i_ino
;
1342 mp
->m_sb
.sb_gquotino
= (*ip
)->i_ino
;
1343 spin_unlock(&mp
->m_sb_lock
);
1344 xfs_mod_sb(tp
, sbfields
);
1346 if ((error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
))) {
1347 xfs_fs_cmn_err(CE_ALERT
, mp
, "XFS qino_alloc failed!");
1355 xfs_qm_reset_dqcounts(
1361 xfs_disk_dquot_t
*ddq
;
1364 trace_xfs_reset_dqcounts(bp
, _RET_IP_
);
1367 * Reset all counters and timers. They'll be
1368 * started afresh by xfs_qm_quotacheck.
1371 j
= XFS_FSB_TO_B(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
1372 do_div(j
, sizeof(xfs_dqblk_t
));
1373 ASSERT(XFS_QM_DQPERBLK(mp
) == j
);
1375 ddq
= (xfs_disk_dquot_t
*)XFS_BUF_PTR(bp
);
1376 for (j
= 0; j
< XFS_QM_DQPERBLK(mp
); j
++) {
1378 * Do a sanity check, and if needed, repair the dqblk. Don't
1379 * output any warnings because it's perfectly possible to
1380 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1382 (void) xfs_qm_dqcheck(ddq
, id
+j
, type
, XFS_QMOPT_DQREPAIR
,
1386 ddq
->d_rtbcount
= 0;
1389 ddq
->d_rtbtimer
= 0;
1392 ddq
->d_rtbwarns
= 0;
1393 ddq
= (xfs_disk_dquot_t
*) ((xfs_dqblk_t
*)ddq
+ 1);
1402 xfs_filblks_t blkcnt
,
1413 incr
= (blkcnt
> XFS_QM_MAX_DQCLUSTER_LOGSZ
) ?
1414 XFS_QM_MAX_DQCLUSTER_LOGSZ
: blkcnt
;
1415 type
= flags
& XFS_QMOPT_UQUOTA
? XFS_DQ_USER
:
1416 (flags
& XFS_QMOPT_PQUOTA
? XFS_DQ_PROJ
: XFS_DQ_GROUP
);
1420 * Blkcnt arg can be a very big number, and might even be
1421 * larger than the log itself. So, we have to break it up into
1422 * manageable-sized transactions.
1423 * Note that we don't start a permanent transaction here; we might
1424 * not be able to get a log reservation for the whole thing up front,
1425 * and we don't really care to either, because we just discard
1426 * everything if we were to crash in the middle of this loop.
1429 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
,
1430 XFS_FSB_TO_DADDR(mp
, bno
),
1431 (int)XFS_QI_DQCHUNKLEN(mp
), 0, &bp
);
1435 xfs_qm_reset_dqcounts(mp
, bp
, firstid
, type
);
1436 xfs_bdwrite(mp
, bp
);
1438 * goto the next block.
1441 firstid
+= XFS_QM_DQPERBLK(mp
);
1447 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1448 * caller supplied function for every chunk of dquots that we find.
1456 xfs_bmbt_irec_t
*map
;
1457 int i
, nmaps
; /* number of map entries */
1458 int error
; /* return value */
1459 xfs_fileoff_t lblkno
;
1460 xfs_filblks_t maxlblkcnt
;
1462 xfs_fsblock_t rablkno
;
1463 xfs_filblks_t rablkcnt
;
1467 * This looks racy, but we can't keep an inode lock across a
1468 * trans_reserve. But, this gets called during quotacheck, and that
1469 * happens only at mount time which is single threaded.
1471 if (qip
->i_d
.di_nblocks
== 0)
1474 map
= kmem_alloc(XFS_DQITER_MAP_SIZE
* sizeof(*map
), KM_SLEEP
);
1477 maxlblkcnt
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1479 nmaps
= XFS_DQITER_MAP_SIZE
;
1481 * We aren't changing the inode itself. Just changing
1482 * some of its data. No new blocks are added here, and
1483 * the inode is never added to the transaction.
1485 xfs_ilock(qip
, XFS_ILOCK_SHARED
);
1486 error
= xfs_bmapi(NULL
, qip
, lblkno
,
1487 maxlblkcnt
- lblkno
,
1490 0, map
, &nmaps
, NULL
, NULL
);
1491 xfs_iunlock(qip
, XFS_ILOCK_SHARED
);
1495 ASSERT(nmaps
<= XFS_DQITER_MAP_SIZE
);
1496 for (i
= 0; i
< nmaps
; i
++) {
1497 ASSERT(map
[i
].br_startblock
!= DELAYSTARTBLOCK
);
1498 ASSERT(map
[i
].br_blockcount
);
1501 lblkno
+= map
[i
].br_blockcount
;
1503 if (map
[i
].br_startblock
== HOLESTARTBLOCK
)
1506 firstid
= (xfs_dqid_t
) map
[i
].br_startoff
*
1507 XFS_QM_DQPERBLK(mp
);
1509 * Do a read-ahead on the next extent.
1511 if ((i
+1 < nmaps
) &&
1512 (map
[i
+1].br_startblock
!= HOLESTARTBLOCK
)) {
1513 rablkcnt
= map
[i
+1].br_blockcount
;
1514 rablkno
= map
[i
+1].br_startblock
;
1515 while (rablkcnt
--) {
1516 xfs_baread(mp
->m_ddev_targp
,
1517 XFS_FSB_TO_DADDR(mp
, rablkno
),
1518 (int)XFS_QI_DQCHUNKLEN(mp
));
1523 * Iterate thru all the blks in the extent and
1524 * reset the counters of all the dquots inside them.
1526 if ((error
= xfs_qm_dqiter_bufs(mp
,
1528 map
[i
].br_startblock
,
1529 map
[i
].br_blockcount
,
1537 } while (nmaps
> 0);
1545 * Called by dqusage_adjust in doing a quotacheck.
1546 * Given the inode, and a dquot (either USR or GRP, doesn't matter),
1547 * this updates its incore copy as well as the buffer copy. This is
1548 * so that once the quotacheck is done, we can just log all the buffers,
1549 * as opposed to logging numerous updates to individual dquots.
1552 xfs_qm_quotacheck_dqadjust(
1557 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1559 trace_xfs_dqadjust(dqp
);
1562 * Adjust the inode count and the block count to reflect this inode's
1565 be64_add_cpu(&dqp
->q_core
.d_icount
, 1);
1566 dqp
->q_res_icount
++;
1568 be64_add_cpu(&dqp
->q_core
.d_bcount
, nblks
);
1569 dqp
->q_res_bcount
+= nblks
;
1572 be64_add_cpu(&dqp
->q_core
.d_rtbcount
, rtblks
);
1573 dqp
->q_res_rtbcount
+= rtblks
;
1577 * Set default limits, adjust timers (since we changed usages)
1579 if (! XFS_IS_SUSER_DQUOT(dqp
)) {
1580 xfs_qm_adjust_dqlimits(dqp
->q_mount
, &dqp
->q_core
);
1581 xfs_qm_adjust_dqtimers(dqp
->q_mount
, &dqp
->q_core
);
1584 dqp
->dq_flags
|= XFS_DQ_DIRTY
;
1590 xfs_qcnt_t
*O_rtblks
)
1592 xfs_filblks_t rtblks
; /* total rt blks */
1593 xfs_extnum_t idx
; /* extent record index */
1594 xfs_ifork_t
*ifp
; /* inode fork pointer */
1595 xfs_extnum_t nextents
; /* number of extent entries */
1598 ASSERT(XFS_IS_REALTIME_INODE(ip
));
1599 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1600 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1601 if ((error
= xfs_iread_extents(NULL
, ip
, XFS_DATA_FORK
)))
1605 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
1606 for (idx
= 0; idx
< nextents
; idx
++)
1607 rtblks
+= xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp
, idx
));
1608 *O_rtblks
= (xfs_qcnt_t
)rtblks
;
1613 * callback routine supplied to bulkstat(). Given an inumber, find its
1614 * dquots and update them to account for resources taken by that inode.
1618 xfs_qm_dqusage_adjust(
1619 xfs_mount_t
*mp
, /* mount point for filesystem */
1620 xfs_ino_t ino
, /* inode number to get data for */
1621 void __user
*buffer
, /* not used */
1622 int ubsize
, /* not used */
1623 void *private_data
, /* not used */
1624 xfs_daddr_t bno
, /* starting block of inode cluster */
1625 int *ubused
, /* not used */
1626 void *dip
, /* on-disk inode pointer (not used) */
1627 int *res
) /* result code value */
1630 xfs_dquot_t
*udqp
, *gdqp
;
1631 xfs_qcnt_t nblks
, rtblks
;
1634 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1637 * rootino must have its resources accounted for, not so with the quota
1640 if (ino
== mp
->m_sb
.sb_uquotino
|| ino
== mp
->m_sb
.sb_gquotino
) {
1641 *res
= BULKSTAT_RV_NOTHING
;
1642 return XFS_ERROR(EINVAL
);
1646 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1647 * interface expects the inode to be exclusively locked because that's
1648 * the case in all other instances. It's OK that we do this because
1649 * quotacheck is done only at mount time.
1651 if ((error
= xfs_iget(mp
, NULL
, ino
, 0, XFS_ILOCK_EXCL
, &ip
, bno
))) {
1652 *res
= BULKSTAT_RV_NOTHING
;
1657 * Obtain the locked dquots. In case of an error (eg. allocation
1658 * fails for ENOSPC), we return the negative of the error number
1659 * to bulkstat, so that it can get propagated to quotacheck() and
1660 * making us disable quotas for the file system.
1662 if ((error
= xfs_qm_dqget_noattach(ip
, &udqp
, &gdqp
))) {
1663 xfs_iput(ip
, XFS_ILOCK_EXCL
);
1664 *res
= BULKSTAT_RV_GIVEUP
;
1669 if (! XFS_IS_REALTIME_INODE(ip
)) {
1670 nblks
= (xfs_qcnt_t
)ip
->i_d
.di_nblocks
;
1673 * Walk thru the extent list and count the realtime blocks.
1675 if ((error
= xfs_qm_get_rtblks(ip
, &rtblks
))) {
1676 xfs_iput(ip
, XFS_ILOCK_EXCL
);
1681 *res
= BULKSTAT_RV_GIVEUP
;
1684 nblks
= (xfs_qcnt_t
)ip
->i_d
.di_nblocks
- rtblks
;
1686 ASSERT(ip
->i_delayed_blks
== 0);
1689 * We can't release the inode while holding its dquot locks.
1690 * The inode can go into inactive and might try to acquire the dquotlocks.
1691 * So, just unlock here and do a vn_rele at the end.
1693 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1696 * Add the (disk blocks and inode) resources occupied by this
1697 * inode to its dquots. We do this adjustment in the incore dquot,
1698 * and also copy the changes to its buffer.
1699 * We don't care about putting these changes in a transaction
1700 * envelope because if we crash in the middle of a 'quotacheck'
1701 * we have to start from the beginning anyway.
1702 * Once we're done, we'll log all the dquot bufs.
1704 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1705 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1707 if (XFS_IS_UQUOTA_ON(mp
)) {
1709 xfs_qm_quotacheck_dqadjust(udqp
, nblks
, rtblks
);
1712 if (XFS_IS_OQUOTA_ON(mp
)) {
1714 xfs_qm_quotacheck_dqadjust(gdqp
, nblks
, rtblks
);
1718 * Now release the inode. This will send it to 'inactive', and
1719 * possibly even free blocks.
1726 *res
= BULKSTAT_RV_DIDONE
;
1731 * Walk thru all the filesystem inodes and construct a consistent view
1732 * of the disk quota world. If the quotacheck fails, disable quotas.
1738 int done
, count
, error
;
1741 xfs_inode_t
*uip
, *gip
;
1749 ASSERT(XFS_QI_UQIP(mp
) || XFS_QI_GQIP(mp
));
1750 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1753 * There should be no cached dquots. The (simplistic) quotacheck
1754 * algorithm doesn't like that.
1756 ASSERT(list_empty(&mp
->m_quotainfo
->qi_dqlist
));
1758 cmn_err(CE_NOTE
, "XFS quotacheck %s: Please wait.", mp
->m_fsname
);
1761 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1762 * their counters to zero. We need a clean slate.
1763 * We don't log our changes till later.
1765 if ((uip
= XFS_QI_UQIP(mp
))) {
1766 if ((error
= xfs_qm_dqiterate(mp
, uip
, XFS_QMOPT_UQUOTA
)))
1768 flags
|= XFS_UQUOTA_CHKD
;
1771 if ((gip
= XFS_QI_GQIP(mp
))) {
1772 if ((error
= xfs_qm_dqiterate(mp
, gip
, XFS_IS_GQUOTA_ON(mp
) ?
1773 XFS_QMOPT_GQUOTA
: XFS_QMOPT_PQUOTA
)))
1775 flags
|= XFS_OQUOTA_CHKD
;
1780 * Iterate thru all the inodes in the file system,
1781 * adjusting the corresponding dquot counters in core.
1783 if ((error
= xfs_bulkstat(mp
, &lastino
, &count
,
1784 xfs_qm_dqusage_adjust
, NULL
,
1785 structsz
, NULL
, BULKSTAT_FG_IGET
, &done
)))
1791 * We've made all the changes that we need to make incore.
1792 * Flush them down to disk buffers if everything was updated
1796 error
= xfs_qm_dqflush_all(mp
, 0);
1799 * We can get this error if we couldn't do a dquot allocation inside
1800 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1801 * dirty dquots that might be cached, we just want to get rid of them
1802 * and turn quotaoff. The dquots won't be attached to any of the inodes
1803 * at this point (because we intentionally didn't in dqget_noattach).
1806 xfs_qm_dqpurge_all(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_QUOTAOFF
);
1811 * We didn't log anything, because if we crashed, we'll have to
1812 * start the quotacheck from scratch anyway. However, we must make
1813 * sure that our dquot changes are secure before we put the
1814 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1817 XFS_bflush(mp
->m_ddev_targp
);
1820 * If one type of quotas is off, then it will lose its
1821 * quotachecked status, since we won't be doing accounting for
1822 * that type anymore.
1824 mp
->m_qflags
&= ~(XFS_OQUOTA_CHKD
| XFS_UQUOTA_CHKD
);
1825 mp
->m_qflags
|= flags
;
1827 xfs_qm_dquot_list_print(mp
);
1831 cmn_err(CE_WARN
, "XFS quotacheck %s: Unsuccessful (Error %d): "
1832 "Disabling quotas.",
1833 mp
->m_fsname
, error
);
1835 * We must turn off quotas.
1837 ASSERT(mp
->m_quotainfo
!= NULL
);
1838 ASSERT(xfs_Gqm
!= NULL
);
1839 xfs_qm_destroy_quotainfo(mp
);
1840 if (xfs_mount_reset_sbqflags(mp
)) {
1841 cmn_err(CE_WARN
, "XFS quotacheck %s: "
1842 "Failed to reset quota flags.", mp
->m_fsname
);
1845 cmn_err(CE_NOTE
, "XFS quotacheck %s: Done.", mp
->m_fsname
);
1851 * This is called after the superblock has been read in and we're ready to
1852 * iget the quota inodes.
1855 xfs_qm_init_quotainos(
1858 xfs_inode_t
*uip
, *gip
;
1863 ASSERT(mp
->m_quotainfo
);
1869 * Get the uquota and gquota inodes
1871 if (xfs_sb_version_hasquota(&mp
->m_sb
)) {
1872 if (XFS_IS_UQUOTA_ON(mp
) &&
1873 mp
->m_sb
.sb_uquotino
!= NULLFSINO
) {
1874 ASSERT(mp
->m_sb
.sb_uquotino
> 0);
1875 if ((error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_uquotino
,
1877 return XFS_ERROR(error
);
1879 if (XFS_IS_OQUOTA_ON(mp
) &&
1880 mp
->m_sb
.sb_gquotino
!= NULLFSINO
) {
1881 ASSERT(mp
->m_sb
.sb_gquotino
> 0);
1882 if ((error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_gquotino
,
1886 return XFS_ERROR(error
);
1890 flags
|= XFS_QMOPT_SBVERSION
;
1891 sbflags
|= (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1892 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
);
1896 * Create the two inodes, if they don't exist already. The changes
1897 * made above will get added to a transaction and logged in one of
1898 * the qino_alloc calls below. If the device is readonly,
1899 * temporarily switch to read-write to do this.
1901 if (XFS_IS_UQUOTA_ON(mp
) && uip
== NULL
) {
1902 if ((error
= xfs_qm_qino_alloc(mp
, &uip
,
1903 sbflags
| XFS_SB_UQUOTINO
,
1904 flags
| XFS_QMOPT_UQUOTA
)))
1905 return XFS_ERROR(error
);
1907 flags
&= ~XFS_QMOPT_SBVERSION
;
1909 if (XFS_IS_OQUOTA_ON(mp
) && gip
== NULL
) {
1910 flags
|= (XFS_IS_GQUOTA_ON(mp
) ?
1911 XFS_QMOPT_GQUOTA
: XFS_QMOPT_PQUOTA
);
1912 error
= xfs_qm_qino_alloc(mp
, &gip
,
1913 sbflags
| XFS_SB_GQUOTINO
, flags
);
1918 return XFS_ERROR(error
);
1922 XFS_QI_UQIP(mp
) = uip
;
1923 XFS_QI_GQIP(mp
) = gip
;
1930 * Traverse the freelist of dquots and attempt to reclaim a maximum of
1931 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
1932 * favor the lookup function ...
1933 * XXXsup merge this with qm_reclaim_one().
1936 xfs_qm_shake_freelist(
1941 xfs_dquot_t
*dqp
, *nextdqp
;
1953 cmn_err(CE_DEBUG
, "Shake free 0x%x", howmany
);
1955 /* lock order is : hashchainlock, freelistlock, mplistlock */
1957 xfs_qm_freelist_lock(xfs_Gqm
);
1959 for (dqp
= xfs_Gqm
->qm_dqfreelist
.qh_next
;
1960 ((dqp
!= (xfs_dquot_t
*) &xfs_Gqm
->qm_dqfreelist
) &&
1961 nreclaimed
< howmany
); ) {
1962 struct xfs_mount
*mp
= dqp
->q_mount
;
1966 * We are racing with dqlookup here. Naturally we don't
1967 * want to reclaim a dquot that lookup wants.
1969 if (dqp
->dq_flags
& XFS_DQ_WANT
) {
1971 xfs_qm_freelist_unlock(xfs_Gqm
);
1972 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
1974 XQM_STATS_INC(xqmstats
.xs_qm_dqwants
);
1979 * If the dquot is inactive, we are assured that it is
1980 * not on the mplist or the hashlist, and that makes our
1983 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
1985 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
1986 ASSERT(dqp
->HL_PREVP
== NULL
);
1987 ASSERT(list_empty(&dqp
->q_mplist
));
1988 XQM_STATS_INC(xqmstats
.xs_qm_dqinact_reclaims
);
1989 nextdqp
= dqp
->dq_flnext
;
1993 ASSERT(!list_empty(&dqp
->q_mplist
));
1995 * Try to grab the flush lock. If this dquot is in the process of
1996 * getting flushed to disk, we don't want to reclaim it.
1998 if (!xfs_dqflock_nowait(dqp
)) {
2000 dqp
= dqp
->dq_flnext
;
2005 * We have the flush lock so we know that this is not in the
2006 * process of being flushed. So, if this is dirty, flush it
2007 * DELWRI so that we don't get a freelist infested with
2010 if (XFS_DQ_IS_DIRTY(dqp
)) {
2013 trace_xfs_dqshake_dirty(dqp
);
2016 * We flush it delayed write, so don't bother
2017 * releasing the mplock.
2019 error
= xfs_qm_dqflush(dqp
, 0);
2021 xfs_fs_cmn_err(CE_WARN
, mp
,
2022 "xfs_qm_dqflush_all: dquot %p flush failed", dqp
);
2024 xfs_dqunlock(dqp
); /* dqflush unlocks dqflock */
2025 dqp
= dqp
->dq_flnext
;
2029 * We're trying to get the hashlock out of order. This races
2030 * with dqlookup; so, we giveup and goto the next dquot if
2031 * we couldn't get the hashlock. This way, we won't starve
2032 * a dqlookup process that holds the hashlock that is
2033 * waiting for the freelist lock.
2035 if (!mutex_trylock(&dqp
->q_hash
->qh_lock
)) {
2038 dqp
= dqp
->dq_flnext
;
2042 * This races with dquot allocation code as well as dqflush_all
2043 * and reclaim code. So, if we failed to grab the mplist lock,
2044 * giveup everything and start over.
2048 if (!mutex_trylock(&mp
->m_quotainfo
->qi_dqlist_lock
)) {
2049 /* XXX put a sentinel so that we can come back here */
2052 mutex_unlock(&hash
->qh_lock
);
2053 xfs_qm_freelist_unlock(xfs_Gqm
);
2054 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2059 trace_xfs_dqshake_unlink(dqp
);
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_HASHLIST_REMOVE(hash
, dqp
);
2068 list_del_init(&dqp
->q_mplist
);
2069 mp
->m_quotainfo
->qi_dquots
--;
2070 mp
->m_quotainfo
->qi_dqreclaims
++;
2072 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
2073 mutex_unlock(&hash
->qh_lock
);
2076 XQM_FREELIST_REMOVE(dqp
);
2079 XQM_STATS_INC(xqmstats
.xs_qm_dqshake_reclaims
);
2080 xfs_qm_dqdestroy(dqp
);
2083 xfs_qm_freelist_unlock(xfs_Gqm
);
2089 * The kmem_shake interface is invoked when memory is running low.
2093 xfs_qm_shake(int nr_to_scan
, gfp_t gfp_mask
)
2095 int ndqused
, nfree
, n
;
2097 if (!kmem_shake_allow(gfp_mask
))
2102 nfree
= xfs_Gqm
->qm_dqfreelist
.qh_nelems
; /* free dquots */
2103 /* incore dquots in all f/s's */
2104 ndqused
= atomic_read(&xfs_Gqm
->qm_totaldquots
) - nfree
;
2106 ASSERT(ndqused
>= 0);
2108 if (nfree
<= ndqused
&& nfree
< ndquot
)
2111 ndqused
*= xfs_Gqm
->qm_dqfree_ratio
; /* target # of free dquots */
2112 n
= nfree
- ndqused
- ndquot
; /* # over target */
2114 return xfs_qm_shake_freelist(MAX(nfree
, n
));
2119 * Just pop the least recently used dquot off the freelist and
2120 * recycle it. The returned dquot is locked.
2122 STATIC xfs_dquot_t
*
2123 xfs_qm_dqreclaim_one(void)
2125 xfs_dquot_t
*dqpout
;
2134 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
2136 xfs_qm_freelist_lock(xfs_Gqm
);
2138 FOREACH_DQUOT_IN_FREELIST(dqp
, &(xfs_Gqm
->qm_dqfreelist
)) {
2139 struct xfs_mount
*mp
= dqp
->q_mount
;
2143 * We are racing with dqlookup here. Naturally we don't
2144 * want to reclaim a dquot that lookup wants. We release the
2145 * freelist lock and start over, so that lookup will grab
2146 * both the dquot and the freelistlock.
2148 if (dqp
->dq_flags
& XFS_DQ_WANT
) {
2149 ASSERT(! (dqp
->dq_flags
& XFS_DQ_INACTIVE
));
2151 trace_xfs_dqreclaim_want(dqp
);
2154 xfs_qm_freelist_unlock(xfs_Gqm
);
2155 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2157 XQM_STATS_INC(xqmstats
.xs_qm_dqwants
);
2162 * If the dquot is inactive, we are assured that it is
2163 * not on the mplist or the hashlist, and that makes our
2166 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
2168 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
2169 ASSERT(dqp
->HL_PREVP
== NULL
);
2170 ASSERT(list_empty(&dqp
->q_mplist
));
2171 XQM_FREELIST_REMOVE(dqp
);
2174 XQM_STATS_INC(xqmstats
.xs_qm_dqinact_reclaims
);
2178 ASSERT(dqp
->q_hash
);
2179 ASSERT(!list_empty(&dqp
->q_mplist
));
2182 * Try to grab the flush lock. If this dquot is in the process of
2183 * getting flushed to disk, we don't want to reclaim it.
2185 if (!xfs_dqflock_nowait(dqp
)) {
2191 * We have the flush lock so we know that this is not in the
2192 * process of being flushed. So, if this is dirty, flush it
2193 * DELWRI so that we don't get a freelist infested with
2196 if (XFS_DQ_IS_DIRTY(dqp
)) {
2199 trace_xfs_dqreclaim_dirty(dqp
);
2202 * We flush it delayed write, so don't bother
2203 * releasing the freelist lock.
2205 error
= xfs_qm_dqflush(dqp
, 0);
2207 xfs_fs_cmn_err(CE_WARN
, mp
,
2208 "xfs_qm_dqreclaim: dquot %p flush failed", dqp
);
2210 xfs_dqunlock(dqp
); /* dqflush unlocks dqflock */
2214 if (!mutex_trylock(&mp
->m_quotainfo
->qi_dqlist_lock
)) {
2220 if (!mutex_trylock(&dqp
->q_hash
->qh_lock
))
2223 trace_xfs_dqreclaim_unlink(dqp
);
2225 ASSERT(dqp
->q_nrefs
== 0);
2226 list_del_init(&dqp
->q_mplist
);
2227 mp
->m_quotainfo
->qi_dquots
--;
2228 mp
->m_quotainfo
->qi_dqreclaims
++;
2229 XQM_HASHLIST_REMOVE(dqp
->q_hash
, dqp
);
2230 XQM_FREELIST_REMOVE(dqp
);
2232 mutex_unlock(&dqp
->q_hash
->qh_lock
);
2234 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
2241 xfs_qm_freelist_unlock(xfs_Gqm
);
2246 /*------------------------------------------------------------------*/
2249 * Return a new incore dquot. Depending on the number of
2250 * dquots in the system, we either allocate a new one on the kernel heap,
2251 * or reclaim a free one.
2252 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2253 * to reclaim an existing one from the freelist.
2256 xfs_qm_dqalloc_incore(
2257 xfs_dquot_t
**O_dqpp
)
2262 * Check against high water mark to see if we want to pop
2263 * a nincompoop dquot off the freelist.
2265 if (atomic_read(&xfs_Gqm
->qm_totaldquots
) >= ndquot
) {
2267 * Try to recycle a dquot from the freelist.
2269 if ((dqp
= xfs_qm_dqreclaim_one())) {
2270 XQM_STATS_INC(xqmstats
.xs_qm_dqreclaims
);
2272 * Just zero the core here. The rest will get
2273 * reinitialized by caller. XXX we shouldn't even
2276 memset(&dqp
->q_core
, 0, sizeof(dqp
->q_core
));
2280 XQM_STATS_INC(xqmstats
.xs_qm_dqreclaim_misses
);
2284 * Allocate a brand new dquot on the kernel heap and return it
2285 * to the caller to initialize.
2287 ASSERT(xfs_Gqm
->qm_dqzone
!= NULL
);
2288 *O_dqpp
= kmem_zone_zalloc(xfs_Gqm
->qm_dqzone
, KM_SLEEP
);
2289 atomic_inc(&xfs_Gqm
->qm_totaldquots
);
2296 * Start a transaction and write the incore superblock changes to
2297 * disk. flags parameter indicates which fields have changed.
2300 xfs_qm_write_sb_changes(
2308 cmn_err(CE_NOTE
, "Writing superblock quota changes :%s", mp
->m_fsname
);
2310 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
2311 if ((error
= xfs_trans_reserve(tp
, 0,
2312 mp
->m_sb
.sb_sectsize
+ 128, 0,
2314 XFS_DEFAULT_LOG_COUNT
))) {
2315 xfs_trans_cancel(tp
, 0);
2319 xfs_mod_sb(tp
, flags
);
2320 error
= xfs_trans_commit(tp
, 0);
2326 /* --------------- utility functions for vnodeops ---------------- */
2330 * Given an inode, a uid and gid (from cred_t) make sure that we have
2331 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2332 * quotas by creating this file.
2333 * This also attaches dquot(s) to the given inode after locking it,
2334 * and returns the dquots corresponding to the uid and/or gid.
2336 * in : inode (unlocked)
2337 * out : udquot, gdquot with references taken and unlocked
2341 struct xfs_inode
*ip
,
2346 struct xfs_dquot
**O_udqpp
,
2347 struct xfs_dquot
**O_gdqpp
)
2349 struct xfs_mount
*mp
= ip
->i_mount
;
2350 struct xfs_dquot
*uq
, *gq
;
2354 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
2357 lockflags
= XFS_ILOCK_EXCL
;
2358 xfs_ilock(ip
, lockflags
);
2360 if ((flags
& XFS_QMOPT_INHERIT
) && XFS_INHERIT_GID(ip
))
2361 gid
= ip
->i_d
.di_gid
;
2364 * Attach the dquot(s) to this inode, doing a dquot allocation
2365 * if necessary. The dquot(s) will not be locked.
2367 if (XFS_NOT_DQATTACHED(mp
, ip
)) {
2368 error
= xfs_qm_dqattach_locked(ip
, XFS_QMOPT_DQALLOC
);
2370 xfs_iunlock(ip
, lockflags
);
2376 if ((flags
& XFS_QMOPT_UQUOTA
) && XFS_IS_UQUOTA_ON(mp
)) {
2377 if (ip
->i_d
.di_uid
!= uid
) {
2379 * What we need is the dquot that has this uid, and
2380 * if we send the inode to dqget, the uid of the inode
2381 * takes priority over what's sent in the uid argument.
2382 * We must unlock inode here before calling dqget if
2383 * we're not sending the inode, because otherwise
2384 * we'll deadlock by doing trans_reserve while
2387 xfs_iunlock(ip
, lockflags
);
2388 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
) uid
,
2393 ASSERT(error
!= ENOENT
);
2397 * Get the ilock in the right order.
2400 lockflags
= XFS_ILOCK_SHARED
;
2401 xfs_ilock(ip
, lockflags
);
2404 * Take an extra reference, because we'll return
2407 ASSERT(ip
->i_udquot
);
2414 if ((flags
& XFS_QMOPT_GQUOTA
) && XFS_IS_GQUOTA_ON(mp
)) {
2415 if (ip
->i_d
.di_gid
!= gid
) {
2416 xfs_iunlock(ip
, lockflags
);
2417 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)gid
,
2424 ASSERT(error
!= ENOENT
);
2428 lockflags
= XFS_ILOCK_SHARED
;
2429 xfs_ilock(ip
, lockflags
);
2431 ASSERT(ip
->i_gdquot
);
2437 } else if ((flags
& XFS_QMOPT_PQUOTA
) && XFS_IS_PQUOTA_ON(mp
)) {
2438 if (ip
->i_d
.di_projid
!= prid
) {
2439 xfs_iunlock(ip
, lockflags
);
2440 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)prid
,
2447 ASSERT(error
!= ENOENT
);
2451 lockflags
= XFS_ILOCK_SHARED
;
2452 xfs_ilock(ip
, lockflags
);
2454 ASSERT(ip
->i_gdquot
);
2462 trace_xfs_dquot_dqalloc(ip
);
2464 xfs_iunlock(ip
, lockflags
);
2477 * Actually transfer ownership, and do dquot modifications.
2478 * These were already reserved.
2484 xfs_dquot_t
**IO_olddq
,
2487 xfs_dquot_t
*prevdq
;
2488 uint bfield
= XFS_IS_REALTIME_INODE(ip
) ?
2489 XFS_TRANS_DQ_RTBCOUNT
: XFS_TRANS_DQ_BCOUNT
;
2492 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2493 ASSERT(XFS_IS_QUOTA_RUNNING(ip
->i_mount
));
2498 ASSERT(prevdq
!= newdq
);
2500 xfs_trans_mod_dquot(tp
, prevdq
, bfield
, -(ip
->i_d
.di_nblocks
));
2501 xfs_trans_mod_dquot(tp
, prevdq
, XFS_TRANS_DQ_ICOUNT
, -1);
2503 /* the sparkling new dquot */
2504 xfs_trans_mod_dquot(tp
, newdq
, bfield
, ip
->i_d
.di_nblocks
);
2505 xfs_trans_mod_dquot(tp
, newdq
, XFS_TRANS_DQ_ICOUNT
, 1);
2508 * Take an extra reference, because the inode
2509 * is going to keep this dquot pointer even
2510 * after the trans_commit.
2514 xfs_dqunlock(newdq
);
2521 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
2524 xfs_qm_vop_chown_reserve(
2531 xfs_mount_t
*mp
= ip
->i_mount
;
2532 uint delblks
, blkflags
, prjflags
= 0;
2533 xfs_dquot_t
*unresudq
, *unresgdq
, *delblksudq
, *delblksgdq
;
2537 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2538 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
2540 delblks
= ip
->i_delayed_blks
;
2541 delblksudq
= delblksgdq
= unresudq
= unresgdq
= NULL
;
2542 blkflags
= XFS_IS_REALTIME_INODE(ip
) ?
2543 XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
;
2545 if (XFS_IS_UQUOTA_ON(mp
) && udqp
&&
2546 ip
->i_d
.di_uid
!= (uid_t
)be32_to_cpu(udqp
->q_core
.d_id
)) {
2549 * If there are delayed allocation blocks, then we have to
2550 * unreserve those from the old dquot, and add them to the
2554 ASSERT(ip
->i_udquot
);
2555 unresudq
= ip
->i_udquot
;
2558 if (XFS_IS_OQUOTA_ON(ip
->i_mount
) && gdqp
) {
2559 if (XFS_IS_PQUOTA_ON(ip
->i_mount
) &&
2560 ip
->i_d
.di_projid
!= be32_to_cpu(gdqp
->q_core
.d_id
))
2561 prjflags
= XFS_QMOPT_ENOSPC
;
2564 (XFS_IS_GQUOTA_ON(ip
->i_mount
) &&
2565 ip
->i_d
.di_gid
!= be32_to_cpu(gdqp
->q_core
.d_id
))) {
2568 ASSERT(ip
->i_gdquot
);
2569 unresgdq
= ip
->i_gdquot
;
2574 if ((error
= xfs_trans_reserve_quota_bydquots(tp
, ip
->i_mount
,
2575 delblksudq
, delblksgdq
, ip
->i_d
.di_nblocks
, 1,
2576 flags
| blkflags
| prjflags
)))
2580 * Do the delayed blks reservations/unreservations now. Since, these
2581 * are done without the help of a transaction, if a reservation fails
2582 * its previous reservations won't be automatically undone by trans
2583 * code. So, we have to do it manually here.
2587 * Do the reservations first. Unreservation can't fail.
2589 ASSERT(delblksudq
|| delblksgdq
);
2590 ASSERT(unresudq
|| unresgdq
);
2591 if ((error
= xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
2592 delblksudq
, delblksgdq
, (xfs_qcnt_t
)delblks
, 0,
2593 flags
| blkflags
| prjflags
)))
2595 xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
2596 unresudq
, unresgdq
, -((xfs_qcnt_t
)delblks
), 0,
2604 xfs_qm_vop_rename_dqattach(
2605 struct xfs_inode
**i_tab
)
2607 struct xfs_mount
*mp
= i_tab
[0]->i_mount
;
2610 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
2613 for (i
= 0; (i
< 4 && i_tab
[i
]); i
++) {
2614 struct xfs_inode
*ip
= i_tab
[i
];
2618 * Watch out for duplicate entries in the table.
2620 if (i
== 0 || ip
!= i_tab
[i
-1]) {
2621 if (XFS_NOT_DQATTACHED(mp
, ip
)) {
2622 error
= xfs_qm_dqattach(ip
, 0);
2632 xfs_qm_vop_create_dqattach(
2633 struct xfs_trans
*tp
,
2634 struct xfs_inode
*ip
,
2635 struct xfs_dquot
*udqp
,
2636 struct xfs_dquot
*gdqp
)
2638 struct xfs_mount
*mp
= tp
->t_mountp
;
2640 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
2643 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2644 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
2650 ASSERT(ip
->i_udquot
== NULL
);
2651 ip
->i_udquot
= udqp
;
2652 ASSERT(XFS_IS_UQUOTA_ON(mp
));
2653 ASSERT(ip
->i_d
.di_uid
== be32_to_cpu(udqp
->q_core
.d_id
));
2654 xfs_trans_mod_dquot(tp
, udqp
, XFS_TRANS_DQ_ICOUNT
, 1);
2660 ASSERT(ip
->i_gdquot
== NULL
);
2661 ip
->i_gdquot
= gdqp
;
2662 ASSERT(XFS_IS_OQUOTA_ON(mp
));
2663 ASSERT((XFS_IS_GQUOTA_ON(mp
) ?
2664 ip
->i_d
.di_gid
: ip
->i_d
.di_projid
) ==
2665 be32_to_cpu(gdqp
->q_core
.d_id
));
2666 xfs_trans_mod_dquot(tp
, gdqp
, XFS_TRANS_DQ_ICOUNT
, 1);
2670 /* ------------- list stuff -----------------*/
2672 xfs_qm_freelist_init(xfs_frlist_t
*ql
)
2674 ql
->qh_next
= ql
->qh_prev
= (xfs_dquot_t
*) ql
;
2675 mutex_init(&ql
->qh_lock
);
2681 xfs_qm_freelist_destroy(xfs_frlist_t
*ql
)
2683 xfs_dquot_t
*dqp
, *nextdqp
;
2685 mutex_lock(&ql
->qh_lock
);
2686 for (dqp
= ql
->qh_next
;
2687 dqp
!= (xfs_dquot_t
*)ql
; ) {
2689 nextdqp
= dqp
->dq_flnext
;
2691 cmn_err(CE_DEBUG
, "FREELIST destroy 0x%p", dqp
);
2693 XQM_FREELIST_REMOVE(dqp
);
2695 xfs_qm_dqdestroy(dqp
);
2698 mutex_unlock(&ql
->qh_lock
);
2699 mutex_destroy(&ql
->qh_lock
);
2701 ASSERT(ql
->qh_nelems
== 0);
2705 xfs_qm_freelist_insert(xfs_frlist_t
*ql
, xfs_dquot_t
*dq
)
2707 dq
->dq_flnext
= ql
->qh_next
;
2708 dq
->dq_flprev
= (xfs_dquot_t
*)ql
;
2710 dq
->dq_flnext
->dq_flprev
= dq
;
2711 xfs_Gqm
->qm_dqfreelist
.qh_nelems
++;
2712 xfs_Gqm
->qm_dqfreelist
.qh_version
++;
2716 xfs_qm_freelist_unlink(xfs_dquot_t
*dq
)
2718 xfs_dquot_t
*next
= dq
->dq_flnext
;
2719 xfs_dquot_t
*prev
= dq
->dq_flprev
;
2721 next
->dq_flprev
= prev
;
2722 prev
->dq_flnext
= next
;
2723 dq
->dq_flnext
= dq
->dq_flprev
= dq
;
2724 xfs_Gqm
->qm_dqfreelist
.qh_nelems
--;
2725 xfs_Gqm
->qm_dqfreelist
.qh_version
++;
2729 xfs_qm_freelist_append(xfs_frlist_t
*ql
, xfs_dquot_t
*dq
)
2731 xfs_qm_freelist_insert((xfs_frlist_t
*)ql
->qh_prev
, dq
);