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"
26 #include "xfs_alloc.h"
27 #include "xfs_quota.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_ialloc.h"
34 #include "xfs_itable.h"
35 #include "xfs_rtalloc.h"
36 #include "xfs_error.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_trans_space.h"
41 #include "xfs_utils.h"
43 #include "xfs_trace.h"
46 * The global quota manager. There is only one of these for the entire
47 * system, _not_ one per file system. XQM keeps track of the overall
48 * quota functionality, including maintaining the freelist and hash
51 struct mutex xfs_Gqm_lock
;
52 struct xfs_qm
*xfs_Gqm
;
55 kmem_zone_t
*qm_dqzone
;
56 kmem_zone_t
*qm_dqtrxzone
;
58 STATIC
void xfs_qm_list_init(xfs_dqlist_t
*, char *, int);
59 STATIC
void xfs_qm_list_destroy(xfs_dqlist_t
*);
61 STATIC
int xfs_qm_init_quotainos(xfs_mount_t
*);
62 STATIC
int xfs_qm_init_quotainfo(xfs_mount_t
*);
63 STATIC
int xfs_qm_shake(struct shrinker
*, struct shrink_control
*);
65 static struct shrinker xfs_qm_shaker
= {
66 .shrink
= xfs_qm_shake
,
67 .seeks
= DEFAULT_SEEKS
,
71 * Initialize the XQM structure.
72 * Note that there is not one quota manager per file system.
74 STATIC
struct xfs_qm
*
77 xfs_dqhash_t
*udqhash
, *gdqhash
;
83 * Initialize the dquot hash tables.
85 udqhash
= kmem_zalloc_greedy(&hsize
,
86 XFS_QM_HASHSIZE_LOW
* sizeof(xfs_dqhash_t
),
87 XFS_QM_HASHSIZE_HIGH
* sizeof(xfs_dqhash_t
));
91 gdqhash
= kmem_zalloc_large(hsize
);
93 goto out_free_udqhash
;
95 hsize
/= sizeof(xfs_dqhash_t
);
98 xqm
= kmem_zalloc(sizeof(xfs_qm_t
), KM_SLEEP
);
99 xqm
->qm_dqhashmask
= hsize
- 1;
100 xqm
->qm_usr_dqhtable
= udqhash
;
101 xqm
->qm_grp_dqhtable
= gdqhash
;
102 ASSERT(xqm
->qm_usr_dqhtable
!= NULL
);
103 ASSERT(xqm
->qm_grp_dqhtable
!= NULL
);
105 for (i
= 0; i
< hsize
; i
++) {
106 xfs_qm_list_init(&(xqm
->qm_usr_dqhtable
[i
]), "uxdqh", i
);
107 xfs_qm_list_init(&(xqm
->qm_grp_dqhtable
[i
]), "gxdqh", i
);
111 * Freelist of all dquots of all file systems
113 INIT_LIST_HEAD(&xqm
->qm_dqfrlist
);
114 xqm
->qm_dqfrlist_cnt
= 0;
115 mutex_init(&xqm
->qm_dqfrlist_lock
);
118 * dquot zone. we register our own low-memory callback.
121 xqm
->qm_dqzone
= kmem_zone_init(sizeof(xfs_dquot_t
),
123 qm_dqzone
= xqm
->qm_dqzone
;
125 xqm
->qm_dqzone
= qm_dqzone
;
127 register_shrinker(&xfs_qm_shaker
);
130 * The t_dqinfo portion of transactions.
133 xqm
->qm_dqtrxzone
= kmem_zone_init(sizeof(xfs_dquot_acct_t
),
135 qm_dqtrxzone
= xqm
->qm_dqtrxzone
;
137 xqm
->qm_dqtrxzone
= qm_dqtrxzone
;
139 atomic_set(&xqm
->qm_totaldquots
, 0);
140 xqm
->qm_dqfree_ratio
= XFS_QM_DQFREE_RATIO
;
145 kmem_free_large(udqhash
);
151 * Destroy the global quota manager when its reference count goes to zero.
160 ASSERT(xqm
->qm_nrefs
== 0);
162 unregister_shrinker(&xfs_qm_shaker
);
164 mutex_lock(&xqm
->qm_dqfrlist_lock
);
165 ASSERT(list_empty(&xqm
->qm_dqfrlist
));
166 mutex_unlock(&xqm
->qm_dqfrlist_lock
);
168 hsize
= xqm
->qm_dqhashmask
+ 1;
169 for (i
= 0; i
< hsize
; i
++) {
170 xfs_qm_list_destroy(&(xqm
->qm_usr_dqhtable
[i
]));
171 xfs_qm_list_destroy(&(xqm
->qm_grp_dqhtable
[i
]));
173 kmem_free_large(xqm
->qm_usr_dqhtable
);
174 kmem_free_large(xqm
->qm_grp_dqhtable
);
175 xqm
->qm_usr_dqhtable
= NULL
;
176 xqm
->qm_grp_dqhtable
= NULL
;
177 xqm
->qm_dqhashmask
= 0;
183 * Called at mount time to let XQM know that another file system is
184 * starting quotas. This isn't crucial information as the individual mount
185 * structures are pretty independent, but it helps the XQM keep a
186 * global view of what's going on.
190 xfs_qm_hold_quotafs_ref(
191 struct xfs_mount
*mp
)
194 * Need to lock the xfs_Gqm structure for things like this. For example,
195 * the structure could disappear between the entry to this routine and
196 * a HOLD operation if not locked.
198 mutex_lock(&xfs_Gqm_lock
);
201 xfs_Gqm
= xfs_Gqm_init();
203 mutex_unlock(&xfs_Gqm_lock
);
209 * We can keep a list of all filesystems with quotas mounted for
210 * debugging and statistical purposes, but ...
211 * Just take a reference and get out.
214 mutex_unlock(&xfs_Gqm_lock
);
221 * Release the reference that a filesystem took at mount time,
222 * so that we know when we need to destroy the entire quota manager.
226 xfs_qm_rele_quotafs_ref(
227 struct xfs_mount
*mp
)
230 ASSERT(xfs_Gqm
->qm_nrefs
> 0);
233 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
236 mutex_lock(&xfs_Gqm_lock
);
237 if (--xfs_Gqm
->qm_nrefs
== 0) {
238 xfs_qm_destroy(xfs_Gqm
);
241 mutex_unlock(&xfs_Gqm_lock
);
245 * Just destroy the quotainfo structure.
249 struct xfs_mount
*mp
)
251 if (mp
->m_quotainfo
) {
252 xfs_qm_dqpurge_all(mp
, XFS_QMOPT_QUOTALL
);
253 xfs_qm_destroy_quotainfo(mp
);
259 * This is called from xfs_mountfs to start quotas and initialize all
260 * necessary data structures like quotainfo. This is also responsible for
261 * running a quotacheck as necessary. We are guaranteed that the superblock
262 * is consistently read in at this point.
264 * If we fail here, the mount will continue with quota turned off. We don't
265 * need to inidicate success or failure at all.
275 * If quotas on realtime volumes is not supported, we disable
276 * quotas immediately.
278 if (mp
->m_sb
.sb_rextents
) {
279 xfs_notice(mp
, "Cannot turn on quotas for realtime filesystem");
284 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
287 * Allocate the quotainfo structure inside the mount struct, and
288 * create quotainode(s), and change/rev superblock if necessary.
290 error
= xfs_qm_init_quotainfo(mp
);
293 * We must turn off quotas.
295 ASSERT(mp
->m_quotainfo
== NULL
);
300 * If any of the quotas are not consistent, do a quotacheck.
302 if (XFS_QM_NEED_QUOTACHECK(mp
)) {
303 error
= xfs_qm_quotacheck(mp
);
305 /* Quotacheck failed and disabled quotas. */
310 * If one type of quotas is off, then it will lose its
311 * quotachecked status, since we won't be doing accounting for
314 if (!XFS_IS_UQUOTA_ON(mp
))
315 mp
->m_qflags
&= ~XFS_UQUOTA_CHKD
;
316 if (!(XFS_IS_GQUOTA_ON(mp
) || XFS_IS_PQUOTA_ON(mp
)))
317 mp
->m_qflags
&= ~XFS_OQUOTA_CHKD
;
321 * We actually don't have to acquire the m_sb_lock at all.
322 * This can only be called from mount, and that's single threaded. XXX
324 spin_lock(&mp
->m_sb_lock
);
325 sbf
= mp
->m_sb
.sb_qflags
;
326 mp
->m_sb
.sb_qflags
= mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
;
327 spin_unlock(&mp
->m_sb_lock
);
329 if (sbf
!= (mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
)) {
330 if (xfs_qm_write_sb_changes(mp
, XFS_SB_QFLAGS
)) {
332 * We could only have been turning quotas off.
333 * We aren't in very good shape actually because
334 * the incore structures are convinced that quotas are
335 * off, but the on disk superblock doesn't know that !
337 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp
)));
338 xfs_alert(mp
, "%s: Superblock update failed!",
344 xfs_warn(mp
, "Failed to initialize disk quotas.");
350 * Called from the vfsops layer.
353 xfs_qm_unmount_quotas(
357 * Release the dquots that root inode, et al might be holding,
358 * before we flush quotas and blow away the quotainfo structure.
360 ASSERT(mp
->m_rootip
);
361 xfs_qm_dqdetach(mp
->m_rootip
);
363 xfs_qm_dqdetach(mp
->m_rbmip
);
365 xfs_qm_dqdetach(mp
->m_rsumip
);
368 * Release the quota inodes.
370 if (mp
->m_quotainfo
) {
371 if (mp
->m_quotainfo
->qi_uquotaip
) {
372 IRELE(mp
->m_quotainfo
->qi_uquotaip
);
373 mp
->m_quotainfo
->qi_uquotaip
= NULL
;
375 if (mp
->m_quotainfo
->qi_gquotaip
) {
376 IRELE(mp
->m_quotainfo
->qi_gquotaip
);
377 mp
->m_quotainfo
->qi_gquotaip
= NULL
;
383 * Flush all dquots of the given file system to disk. The dquots are
384 * _not_ purged from memory here, just their data written to disk.
388 struct xfs_mount
*mp
)
390 struct xfs_quotainfo
*q
= mp
->m_quotainfo
;
392 struct xfs_dquot
*dqp
;
398 mutex_lock(&q
->qi_dqlist_lock
);
399 list_for_each_entry(dqp
, &q
->qi_dqlist
, q_mplist
) {
401 if ((dqp
->dq_flags
& XFS_DQ_FREEING
) ||
402 !XFS_DQ_IS_DIRTY(dqp
)) {
407 /* XXX a sentinel would be better */
408 recl
= q
->qi_dqreclaims
;
409 if (!xfs_dqflock_nowait(dqp
)) {
411 * If we can't grab the flush lock then check
412 * to see if the dquot has been flushed delayed
413 * write. If so, grab its buffer and send it
414 * out immediately. We'll be able to acquire
415 * the flush lock when the I/O completes.
417 xfs_dqflock_pushbuf_wait(dqp
);
420 * Let go of the mplist lock. We don't want to hold it
421 * across a disk write.
423 mutex_unlock(&q
->qi_dqlist_lock
);
424 error
= xfs_qm_dqflush(dqp
, 0);
429 mutex_lock(&q
->qi_dqlist_lock
);
430 if (recl
!= q
->qi_dqreclaims
) {
431 mutex_unlock(&q
->qi_dqlist_lock
);
432 /* XXX restart limit */
437 mutex_unlock(&q
->qi_dqlist_lock
);
443 * Release the group dquot pointers the user dquots may be
444 * carrying around as a hint. mplist is locked on entry and exit.
447 xfs_qm_detach_gdquots(
448 struct xfs_mount
*mp
)
450 struct xfs_quotainfo
*q
= mp
->m_quotainfo
;
451 struct xfs_dquot
*dqp
, *gdqp
;
455 ASSERT(mutex_is_locked(&q
->qi_dqlist_lock
));
456 list_for_each_entry(dqp
, &q
->qi_dqlist
, q_mplist
) {
458 if (dqp
->dq_flags
& XFS_DQ_FREEING
) {
460 mutex_unlock(&q
->qi_dqlist_lock
);
462 mutex_lock(&q
->qi_dqlist_lock
);
465 if ((gdqp
= dqp
->q_gdquot
)) {
467 dqp
->q_gdquot
= NULL
;
473 * Can't hold the mplist lock across a dqput.
474 * XXXmust convert to marker based iterations here.
476 nrecl
= q
->qi_dqreclaims
;
477 mutex_unlock(&q
->qi_dqlist_lock
);
480 mutex_lock(&q
->qi_dqlist_lock
);
481 if (nrecl
!= q
->qi_dqreclaims
)
488 * Go through all the incore dquots of this file system and take them
489 * off the mplist and hashlist, if the dquot type matches the dqtype
490 * parameter. This is used when turning off quota accounting for
491 * users and/or groups, as well as when the filesystem is unmounting.
495 struct xfs_mount
*mp
,
498 struct xfs_quotainfo
*q
= mp
->m_quotainfo
;
499 struct xfs_dquot
*dqp
, *n
;
502 LIST_HEAD (dispose_list
);
507 dqtype
= (flags
& XFS_QMOPT_UQUOTA
) ? XFS_DQ_USER
: 0;
508 dqtype
|= (flags
& XFS_QMOPT_PQUOTA
) ? XFS_DQ_PROJ
: 0;
509 dqtype
|= (flags
& XFS_QMOPT_GQUOTA
) ? XFS_DQ_GROUP
: 0;
511 mutex_lock(&q
->qi_dqlist_lock
);
514 * In the first pass through all incore dquots of this filesystem,
515 * we release the group dquot pointers the user dquots may be
516 * carrying around as a hint. We need to do this irrespective of
517 * what's being turned off.
519 xfs_qm_detach_gdquots(mp
);
522 * Try to get rid of all of the unwanted dquots.
524 list_for_each_entry_safe(dqp
, n
, &q
->qi_dqlist
, q_mplist
) {
526 if ((dqp
->dq_flags
& dqtype
) != 0 &&
527 !(dqp
->dq_flags
& XFS_DQ_FREEING
)) {
528 if (dqp
->q_nrefs
== 0) {
529 dqp
->dq_flags
|= XFS_DQ_FREEING
;
530 list_move_tail(&dqp
->q_mplist
, &dispose_list
);
536 mutex_unlock(&q
->qi_dqlist_lock
);
538 list_for_each_entry_safe(dqp
, n
, &dispose_list
, q_mplist
)
552 * Purge the dquot cache.
553 * None of the dquots should really be busy at this point.
555 if (mp
->m_quotainfo
) {
556 while ((ndquots
= xfs_qm_dqpurge_int(mp
, flags
))) {
569 xfs_dquot_t
*udqhint
, /* hint */
570 xfs_dquot_t
**IO_idqpp
)
575 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
579 * See if we already have it in the inode itself. IO_idqpp is
580 * &i_udquot or &i_gdquot. This made the code look weird, but
581 * made the logic a lot simpler.
585 trace_xfs_dqattach_found(dqp
);
590 * udqhint is the i_udquot field in inode, and is non-NULL only
591 * when the type arg is group/project. Its purpose is to save a
592 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
596 ASSERT(type
== XFS_DQ_GROUP
|| type
== XFS_DQ_PROJ
);
600 * No need to take dqlock to look at the id.
602 * The ID can't change until it gets reclaimed, and it won't
603 * be reclaimed as long as we have a ref from inode and we
606 dqp
= udqhint
->q_gdquot
;
607 if (dqp
&& be32_to_cpu(dqp
->q_core
.d_id
) == id
) {
608 ASSERT(*IO_idqpp
== NULL
);
610 *IO_idqpp
= xfs_qm_dqhold(dqp
);
611 xfs_dqunlock(udqhint
);
616 * We can't hold a dquot lock when we call the dqget code.
617 * We'll deadlock in no time, because of (not conforming to)
618 * lock ordering - the inodelock comes before any dquot lock,
619 * and we may drop and reacquire the ilock in xfs_qm_dqget().
621 xfs_dqunlock(udqhint
);
625 * Find the dquot from somewhere. This bumps the
626 * reference count of dquot and returns it locked.
627 * This can return ENOENT if dquot didn't exist on
628 * disk and we didn't ask it to allocate;
629 * ESRCH if quotas got turned off suddenly.
631 error
= xfs_qm_dqget(ip
->i_mount
, ip
, id
, type
,
632 doalloc
| XFS_QMOPT_DOWARN
, &dqp
);
636 trace_xfs_dqattach_get(dqp
);
639 * dqget may have dropped and re-acquired the ilock, but it guarantees
640 * that the dquot returned is the one that should go in the inode.
649 * Given a udquot and gdquot, attach a ptr to the group dquot in the
650 * udquot as a hint for future lookups.
653 xfs_qm_dqattach_grouphint(
666 udq
->q_gdquot
= NULL
;
670 udq
->q_gdquot
= xfs_qm_dqhold(gdq
);
677 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
679 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
680 * Inode may get unlocked and relocked in here, and the caller must deal with
684 xfs_qm_dqattach_locked(
688 xfs_mount_t
*mp
= ip
->i_mount
;
692 if (!XFS_IS_QUOTA_RUNNING(mp
) ||
693 !XFS_IS_QUOTA_ON(mp
) ||
694 !XFS_NOT_DQATTACHED(mp
, ip
) ||
695 ip
->i_ino
== mp
->m_sb
.sb_uquotino
||
696 ip
->i_ino
== mp
->m_sb
.sb_gquotino
)
699 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
701 if (XFS_IS_UQUOTA_ON(mp
)) {
702 error
= xfs_qm_dqattach_one(ip
, ip
->i_d
.di_uid
, XFS_DQ_USER
,
703 flags
& XFS_QMOPT_DQALLOC
,
704 NULL
, &ip
->i_udquot
);
710 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
711 if (XFS_IS_OQUOTA_ON(mp
)) {
712 error
= XFS_IS_GQUOTA_ON(mp
) ?
713 xfs_qm_dqattach_one(ip
, ip
->i_d
.di_gid
, XFS_DQ_GROUP
,
714 flags
& XFS_QMOPT_DQALLOC
,
715 ip
->i_udquot
, &ip
->i_gdquot
) :
716 xfs_qm_dqattach_one(ip
, xfs_get_projid(ip
), XFS_DQ_PROJ
,
717 flags
& XFS_QMOPT_DQALLOC
,
718 ip
->i_udquot
, &ip
->i_gdquot
);
720 * Don't worry about the udquot that we may have
721 * attached above. It'll get detached, if not already.
729 * Attach this group quota to the user quota as a hint.
730 * This WON'T, in general, result in a thrash.
733 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
734 ASSERT(ip
->i_udquot
);
735 ASSERT(ip
->i_gdquot
);
738 * We do not have i_udquot locked at this point, but this check
739 * is OK since we don't depend on the i_gdquot to be accurate
740 * 100% all the time. It is just a hint, and this will
741 * succeed in general.
743 if (ip
->i_udquot
->q_gdquot
!= ip
->i_gdquot
)
744 xfs_qm_dqattach_grouphint(ip
->i_udquot
, ip
->i_gdquot
);
750 if (XFS_IS_UQUOTA_ON(mp
))
751 ASSERT(ip
->i_udquot
);
752 if (XFS_IS_OQUOTA_ON(mp
))
753 ASSERT(ip
->i_gdquot
);
755 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
762 struct xfs_inode
*ip
,
767 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
768 error
= xfs_qm_dqattach_locked(ip
, flags
);
769 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
775 * Release dquots (and their references) if any.
776 * The inode should be locked EXCL except when this's called by
783 if (!(ip
->i_udquot
|| ip
->i_gdquot
))
786 trace_xfs_dquot_dqdetach(ip
);
788 ASSERT(ip
->i_ino
!= ip
->i_mount
->m_sb
.sb_uquotino
);
789 ASSERT(ip
->i_ino
!= ip
->i_mount
->m_sb
.sb_gquotino
);
791 xfs_qm_dqrele(ip
->i_udquot
);
795 xfs_qm_dqrele(ip
->i_gdquot
);
801 * The hash chains and the mplist use the same xfs_dqhash structure as
802 * their list head, but we can take the mplist qh_lock and one of the
803 * hash qh_locks at the same time without any problem as they aren't
806 static struct lock_class_key xfs_quota_mplist_class
;
809 * This initializes all the quota information that's kept in the
813 xfs_qm_init_quotainfo(
816 xfs_quotainfo_t
*qinf
;
820 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
823 * Tell XQM that we exist as soon as possible.
825 if ((error
= xfs_qm_hold_quotafs_ref(mp
))) {
829 qinf
= mp
->m_quotainfo
= kmem_zalloc(sizeof(xfs_quotainfo_t
), KM_SLEEP
);
832 * See if quotainodes are setup, and if not, allocate them,
833 * and change the superblock accordingly.
835 if ((error
= xfs_qm_init_quotainos(mp
))) {
837 mp
->m_quotainfo
= NULL
;
841 INIT_LIST_HEAD(&qinf
->qi_dqlist
);
842 mutex_init(&qinf
->qi_dqlist_lock
);
843 lockdep_set_class(&qinf
->qi_dqlist_lock
, &xfs_quota_mplist_class
);
845 qinf
->qi_dqreclaims
= 0;
847 /* mutex used to serialize quotaoffs */
848 mutex_init(&qinf
->qi_quotaofflock
);
850 /* Precalc some constants */
851 qinf
->qi_dqchunklen
= XFS_FSB_TO_BB(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
852 ASSERT(qinf
->qi_dqchunklen
);
853 qinf
->qi_dqperchunk
= BBTOB(qinf
->qi_dqchunklen
);
854 do_div(qinf
->qi_dqperchunk
, sizeof(xfs_dqblk_t
));
856 mp
->m_qflags
|= (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_CHKD
);
859 * We try to get the limits from the superuser's limits fields.
860 * This is quite hacky, but it is standard quota practice.
861 * We look at the USR dquot with id == 0 first, but if user quotas
862 * are not enabled we goto the GRP dquot with id == 0.
863 * We don't really care to keep separate default limits for user
864 * and group quotas, at least not at this point.
866 error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)0,
867 XFS_IS_UQUOTA_RUNNING(mp
) ? XFS_DQ_USER
:
868 (XFS_IS_GQUOTA_RUNNING(mp
) ? XFS_DQ_GROUP
:
870 XFS_QMOPT_DQSUSER
|XFS_QMOPT_DOWARN
,
873 xfs_disk_dquot_t
*ddqp
= &dqp
->q_core
;
876 * The warnings and timers set the grace period given to
877 * a user or group before he or she can not perform any
878 * more writing. If it is zero, a default is used.
880 qinf
->qi_btimelimit
= ddqp
->d_btimer
?
881 be32_to_cpu(ddqp
->d_btimer
) : XFS_QM_BTIMELIMIT
;
882 qinf
->qi_itimelimit
= ddqp
->d_itimer
?
883 be32_to_cpu(ddqp
->d_itimer
) : XFS_QM_ITIMELIMIT
;
884 qinf
->qi_rtbtimelimit
= ddqp
->d_rtbtimer
?
885 be32_to_cpu(ddqp
->d_rtbtimer
) : XFS_QM_RTBTIMELIMIT
;
886 qinf
->qi_bwarnlimit
= ddqp
->d_bwarns
?
887 be16_to_cpu(ddqp
->d_bwarns
) : XFS_QM_BWARNLIMIT
;
888 qinf
->qi_iwarnlimit
= ddqp
->d_iwarns
?
889 be16_to_cpu(ddqp
->d_iwarns
) : XFS_QM_IWARNLIMIT
;
890 qinf
->qi_rtbwarnlimit
= ddqp
->d_rtbwarns
?
891 be16_to_cpu(ddqp
->d_rtbwarns
) : XFS_QM_RTBWARNLIMIT
;
892 qinf
->qi_bhardlimit
= be64_to_cpu(ddqp
->d_blk_hardlimit
);
893 qinf
->qi_bsoftlimit
= be64_to_cpu(ddqp
->d_blk_softlimit
);
894 qinf
->qi_ihardlimit
= be64_to_cpu(ddqp
->d_ino_hardlimit
);
895 qinf
->qi_isoftlimit
= be64_to_cpu(ddqp
->d_ino_softlimit
);
896 qinf
->qi_rtbhardlimit
= be64_to_cpu(ddqp
->d_rtb_hardlimit
);
897 qinf
->qi_rtbsoftlimit
= be64_to_cpu(ddqp
->d_rtb_softlimit
);
900 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
901 * we don't want this dquot cached. We haven't done a
902 * quotacheck yet, and quotacheck doesn't like incore dquots.
904 xfs_qm_dqdestroy(dqp
);
906 qinf
->qi_btimelimit
= XFS_QM_BTIMELIMIT
;
907 qinf
->qi_itimelimit
= XFS_QM_ITIMELIMIT
;
908 qinf
->qi_rtbtimelimit
= XFS_QM_RTBTIMELIMIT
;
909 qinf
->qi_bwarnlimit
= XFS_QM_BWARNLIMIT
;
910 qinf
->qi_iwarnlimit
= XFS_QM_IWARNLIMIT
;
911 qinf
->qi_rtbwarnlimit
= XFS_QM_RTBWARNLIMIT
;
919 * Gets called when unmounting a filesystem or when all quotas get
921 * This purges the quota inodes, destroys locks and frees itself.
924 xfs_qm_destroy_quotainfo(
929 qi
= mp
->m_quotainfo
;
931 ASSERT(xfs_Gqm
!= NULL
);
934 * Release the reference that XQM kept, so that we know
935 * when the XQM structure should be freed. We cannot assume
936 * that xfs_Gqm is non-null after this point.
938 xfs_qm_rele_quotafs_ref(mp
);
940 ASSERT(list_empty(&qi
->qi_dqlist
));
941 mutex_destroy(&qi
->qi_dqlist_lock
);
943 if (qi
->qi_uquotaip
) {
944 IRELE(qi
->qi_uquotaip
);
945 qi
->qi_uquotaip
= NULL
; /* paranoia */
947 if (qi
->qi_gquotaip
) {
948 IRELE(qi
->qi_gquotaip
);
949 qi
->qi_gquotaip
= NULL
;
951 mutex_destroy(&qi
->qi_quotaofflock
);
953 mp
->m_quotainfo
= NULL
;
958 /* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
967 mutex_init(&list
->qh_lock
);
968 INIT_LIST_HEAD(&list
->qh_list
);
969 list
->qh_version
= 0;
977 mutex_destroy(&(list
->qh_lock
));
981 * Create an inode and return with a reference already taken, but unlocked
982 * This is how we create quota inodes
995 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_QINOCREATE
);
996 if ((error
= xfs_trans_reserve(tp
,
997 XFS_QM_QINOCREATE_SPACE_RES(mp
),
998 XFS_CREATE_LOG_RES(mp
), 0,
999 XFS_TRANS_PERM_LOG_RES
,
1000 XFS_CREATE_LOG_COUNT
))) {
1001 xfs_trans_cancel(tp
, 0);
1005 error
= xfs_dir_ialloc(&tp
, NULL
, S_IFREG
, 1, 0, 0, 1, ip
, &committed
);
1007 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1013 * Make the changes in the superblock, and log those too.
1014 * sbfields arg may contain fields other than *QUOTINO;
1015 * VERSIONNUM for example.
1017 spin_lock(&mp
->m_sb_lock
);
1018 if (flags
& XFS_QMOPT_SBVERSION
) {
1019 ASSERT(!xfs_sb_version_hasquota(&mp
->m_sb
));
1020 ASSERT((sbfields
& (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1021 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
)) ==
1022 (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1023 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
));
1025 xfs_sb_version_addquota(&mp
->m_sb
);
1026 mp
->m_sb
.sb_uquotino
= NULLFSINO
;
1027 mp
->m_sb
.sb_gquotino
= NULLFSINO
;
1029 /* qflags will get updated _after_ quotacheck */
1030 mp
->m_sb
.sb_qflags
= 0;
1032 if (flags
& XFS_QMOPT_UQUOTA
)
1033 mp
->m_sb
.sb_uquotino
= (*ip
)->i_ino
;
1035 mp
->m_sb
.sb_gquotino
= (*ip
)->i_ino
;
1036 spin_unlock(&mp
->m_sb_lock
);
1037 xfs_mod_sb(tp
, sbfields
);
1039 if ((error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
))) {
1040 xfs_alert(mp
, "%s failed (error %d)!", __func__
, error
);
1048 xfs_qm_reset_dqcounts(
1054 xfs_disk_dquot_t
*ddq
;
1057 trace_xfs_reset_dqcounts(bp
, _RET_IP_
);
1060 * Reset all counters and timers. They'll be
1061 * started afresh by xfs_qm_quotacheck.
1064 j
= XFS_FSB_TO_B(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
1065 do_div(j
, sizeof(xfs_dqblk_t
));
1066 ASSERT(mp
->m_quotainfo
->qi_dqperchunk
== j
);
1069 for (j
= 0; j
< mp
->m_quotainfo
->qi_dqperchunk
; j
++) {
1071 * Do a sanity check, and if needed, repair the dqblk. Don't
1072 * output any warnings because it's perfectly possible to
1073 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1075 (void) xfs_qm_dqcheck(mp
, ddq
, id
+j
, type
, XFS_QMOPT_DQREPAIR
,
1079 ddq
->d_rtbcount
= 0;
1082 ddq
->d_rtbtimer
= 0;
1085 ddq
->d_rtbwarns
= 0;
1086 ddq
= (xfs_disk_dquot_t
*) ((xfs_dqblk_t
*)ddq
+ 1);
1095 xfs_filblks_t blkcnt
,
1103 type
= flags
& XFS_QMOPT_UQUOTA
? XFS_DQ_USER
:
1104 (flags
& XFS_QMOPT_PQUOTA
? XFS_DQ_PROJ
: XFS_DQ_GROUP
);
1108 * Blkcnt arg can be a very big number, and might even be
1109 * larger than the log itself. So, we have to break it up into
1110 * manageable-sized transactions.
1111 * Note that we don't start a permanent transaction here; we might
1112 * not be able to get a log reservation for the whole thing up front,
1113 * and we don't really care to either, because we just discard
1114 * everything if we were to crash in the middle of this loop.
1117 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
,
1118 XFS_FSB_TO_DADDR(mp
, bno
),
1119 mp
->m_quotainfo
->qi_dqchunklen
, 0, &bp
);
1123 xfs_qm_reset_dqcounts(mp
, bp
, firstid
, type
);
1124 xfs_buf_delwri_queue(bp
);
1127 * goto the next block.
1130 firstid
+= mp
->m_quotainfo
->qi_dqperchunk
;
1136 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1137 * caller supplied function for every chunk of dquots that we find.
1145 xfs_bmbt_irec_t
*map
;
1146 int i
, nmaps
; /* number of map entries */
1147 int error
; /* return value */
1148 xfs_fileoff_t lblkno
;
1149 xfs_filblks_t maxlblkcnt
;
1151 xfs_fsblock_t rablkno
;
1152 xfs_filblks_t rablkcnt
;
1156 * This looks racy, but we can't keep an inode lock across a
1157 * trans_reserve. But, this gets called during quotacheck, and that
1158 * happens only at mount time which is single threaded.
1160 if (qip
->i_d
.di_nblocks
== 0)
1163 map
= kmem_alloc(XFS_DQITER_MAP_SIZE
* sizeof(*map
), KM_SLEEP
);
1166 maxlblkcnt
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1168 nmaps
= XFS_DQITER_MAP_SIZE
;
1170 * We aren't changing the inode itself. Just changing
1171 * some of its data. No new blocks are added here, and
1172 * the inode is never added to the transaction.
1174 xfs_ilock(qip
, XFS_ILOCK_SHARED
);
1175 error
= xfs_bmapi_read(qip
, lblkno
, maxlblkcnt
- lblkno
,
1177 xfs_iunlock(qip
, XFS_ILOCK_SHARED
);
1181 ASSERT(nmaps
<= XFS_DQITER_MAP_SIZE
);
1182 for (i
= 0; i
< nmaps
; i
++) {
1183 ASSERT(map
[i
].br_startblock
!= DELAYSTARTBLOCK
);
1184 ASSERT(map
[i
].br_blockcount
);
1187 lblkno
+= map
[i
].br_blockcount
;
1189 if (map
[i
].br_startblock
== HOLESTARTBLOCK
)
1192 firstid
= (xfs_dqid_t
) map
[i
].br_startoff
*
1193 mp
->m_quotainfo
->qi_dqperchunk
;
1195 * Do a read-ahead on the next extent.
1197 if ((i
+1 < nmaps
) &&
1198 (map
[i
+1].br_startblock
!= HOLESTARTBLOCK
)) {
1199 rablkcnt
= map
[i
+1].br_blockcount
;
1200 rablkno
= map
[i
+1].br_startblock
;
1201 while (rablkcnt
--) {
1202 xfs_buf_readahead(mp
->m_ddev_targp
,
1203 XFS_FSB_TO_DADDR(mp
, rablkno
),
1204 mp
->m_quotainfo
->qi_dqchunklen
);
1209 * Iterate thru all the blks in the extent and
1210 * reset the counters of all the dquots inside them.
1212 if ((error
= xfs_qm_dqiter_bufs(mp
,
1214 map
[i
].br_startblock
,
1215 map
[i
].br_blockcount
,
1223 } while (nmaps
> 0);
1231 * Called by dqusage_adjust in doing a quotacheck.
1233 * Given the inode, and a dquot id this updates both the incore dqout as well
1234 * as the buffer copy. This is so that once the quotacheck is done, we can
1235 * just log all the buffers, as opposed to logging numerous updates to
1236 * individual dquots.
1239 xfs_qm_quotacheck_dqadjust(
1240 struct xfs_inode
*ip
,
1246 struct xfs_mount
*mp
= ip
->i_mount
;
1247 struct xfs_dquot
*dqp
;
1250 error
= xfs_qm_dqget(mp
, ip
, id
, type
,
1251 XFS_QMOPT_DQALLOC
| XFS_QMOPT_DOWARN
, &dqp
);
1254 * Shouldn't be able to turn off quotas here.
1256 ASSERT(error
!= ESRCH
);
1257 ASSERT(error
!= ENOENT
);
1261 trace_xfs_dqadjust(dqp
);
1264 * Adjust the inode count and the block count to reflect this inode's
1267 be64_add_cpu(&dqp
->q_core
.d_icount
, 1);
1268 dqp
->q_res_icount
++;
1270 be64_add_cpu(&dqp
->q_core
.d_bcount
, nblks
);
1271 dqp
->q_res_bcount
+= nblks
;
1274 be64_add_cpu(&dqp
->q_core
.d_rtbcount
, rtblks
);
1275 dqp
->q_res_rtbcount
+= rtblks
;
1279 * Set default limits, adjust timers (since we changed usages)
1281 * There are no timers for the default values set in the root dquot.
1283 if (dqp
->q_core
.d_id
) {
1284 xfs_qm_adjust_dqlimits(mp
, &dqp
->q_core
);
1285 xfs_qm_adjust_dqtimers(mp
, &dqp
->q_core
);
1288 dqp
->dq_flags
|= XFS_DQ_DIRTY
;
1296 xfs_qcnt_t
*O_rtblks
)
1298 xfs_filblks_t rtblks
; /* total rt blks */
1299 xfs_extnum_t idx
; /* extent record index */
1300 xfs_ifork_t
*ifp
; /* inode fork pointer */
1301 xfs_extnum_t nextents
; /* number of extent entries */
1304 ASSERT(XFS_IS_REALTIME_INODE(ip
));
1305 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1306 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1307 if ((error
= xfs_iread_extents(NULL
, ip
, XFS_DATA_FORK
)))
1311 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
1312 for (idx
= 0; idx
< nextents
; idx
++)
1313 rtblks
+= xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp
, idx
));
1314 *O_rtblks
= (xfs_qcnt_t
)rtblks
;
1319 * callback routine supplied to bulkstat(). Given an inumber, find its
1320 * dquots and update them to account for resources taken by that inode.
1324 xfs_qm_dqusage_adjust(
1325 xfs_mount_t
*mp
, /* mount point for filesystem */
1326 xfs_ino_t ino
, /* inode number to get data for */
1327 void __user
*buffer
, /* not used */
1328 int ubsize
, /* not used */
1329 int *ubused
, /* not used */
1330 int *res
) /* result code value */
1333 xfs_qcnt_t nblks
, rtblks
= 0;
1336 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1339 * rootino must have its resources accounted for, not so with the quota
1342 if (ino
== mp
->m_sb
.sb_uquotino
|| ino
== mp
->m_sb
.sb_gquotino
) {
1343 *res
= BULKSTAT_RV_NOTHING
;
1344 return XFS_ERROR(EINVAL
);
1348 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1349 * interface expects the inode to be exclusively locked because that's
1350 * the case in all other instances. It's OK that we do this because
1351 * quotacheck is done only at mount time.
1353 error
= xfs_iget(mp
, NULL
, ino
, 0, XFS_ILOCK_EXCL
, &ip
);
1355 *res
= BULKSTAT_RV_NOTHING
;
1359 ASSERT(ip
->i_delayed_blks
== 0);
1361 if (XFS_IS_REALTIME_INODE(ip
)) {
1363 * Walk thru the extent list and count the realtime blocks.
1365 error
= xfs_qm_get_rtblks(ip
, &rtblks
);
1370 nblks
= (xfs_qcnt_t
)ip
->i_d
.di_nblocks
- rtblks
;
1373 * Add the (disk blocks and inode) resources occupied by this
1374 * inode to its dquots. We do this adjustment in the incore dquot,
1375 * and also copy the changes to its buffer.
1376 * We don't care about putting these changes in a transaction
1377 * envelope because if we crash in the middle of a 'quotacheck'
1378 * we have to start from the beginning anyway.
1379 * Once we're done, we'll log all the dquot bufs.
1381 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1382 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1384 if (XFS_IS_UQUOTA_ON(mp
)) {
1385 error
= xfs_qm_quotacheck_dqadjust(ip
, ip
->i_d
.di_uid
,
1386 XFS_DQ_USER
, nblks
, rtblks
);
1391 if (XFS_IS_GQUOTA_ON(mp
)) {
1392 error
= xfs_qm_quotacheck_dqadjust(ip
, ip
->i_d
.di_gid
,
1393 XFS_DQ_GROUP
, nblks
, rtblks
);
1398 if (XFS_IS_PQUOTA_ON(mp
)) {
1399 error
= xfs_qm_quotacheck_dqadjust(ip
, xfs_get_projid(ip
),
1400 XFS_DQ_PROJ
, nblks
, rtblks
);
1405 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1407 *res
= BULKSTAT_RV_DIDONE
;
1411 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1413 *res
= BULKSTAT_RV_GIVEUP
;
1418 * Walk thru all the filesystem inodes and construct a consistent view
1419 * of the disk quota world. If the quotacheck fails, disable quotas.
1425 int done
, count
, error
;
1428 xfs_inode_t
*uip
, *gip
;
1436 ASSERT(mp
->m_quotainfo
->qi_uquotaip
|| mp
->m_quotainfo
->qi_gquotaip
);
1437 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1440 * There should be no cached dquots. The (simplistic) quotacheck
1441 * algorithm doesn't like that.
1443 ASSERT(list_empty(&mp
->m_quotainfo
->qi_dqlist
));
1445 xfs_notice(mp
, "Quotacheck needed: Please wait.");
1448 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1449 * their counters to zero. We need a clean slate.
1450 * We don't log our changes till later.
1452 uip
= mp
->m_quotainfo
->qi_uquotaip
;
1454 error
= xfs_qm_dqiterate(mp
, uip
, XFS_QMOPT_UQUOTA
);
1457 flags
|= XFS_UQUOTA_CHKD
;
1460 gip
= mp
->m_quotainfo
->qi_gquotaip
;
1462 error
= xfs_qm_dqiterate(mp
, gip
, XFS_IS_GQUOTA_ON(mp
) ?
1463 XFS_QMOPT_GQUOTA
: XFS_QMOPT_PQUOTA
);
1466 flags
|= XFS_OQUOTA_CHKD
;
1471 * Iterate thru all the inodes in the file system,
1472 * adjusting the corresponding dquot counters in core.
1474 error
= xfs_bulkstat(mp
, &lastino
, &count
,
1475 xfs_qm_dqusage_adjust
,
1476 structsz
, NULL
, &done
);
1483 * We've made all the changes that we need to make incore.
1484 * Flush them down to disk buffers if everything was updated
1488 error
= xfs_qm_dqflush_all(mp
);
1491 * We can get this error if we couldn't do a dquot allocation inside
1492 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1493 * dirty dquots that might be cached, we just want to get rid of them
1494 * and turn quotaoff. The dquots won't be attached to any of the inodes
1495 * at this point (because we intentionally didn't in dqget_noattach).
1498 xfs_qm_dqpurge_all(mp
, XFS_QMOPT_QUOTALL
);
1503 * We didn't log anything, because if we crashed, we'll have to
1504 * start the quotacheck from scratch anyway. However, we must make
1505 * sure that our dquot changes are secure before we put the
1506 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1509 xfs_flush_buftarg(mp
->m_ddev_targp
, 1);
1512 * If one type of quotas is off, then it will lose its
1513 * quotachecked status, since we won't be doing accounting for
1514 * that type anymore.
1516 mp
->m_qflags
&= ~(XFS_OQUOTA_CHKD
| XFS_UQUOTA_CHKD
);
1517 mp
->m_qflags
|= flags
;
1522 "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1525 * We must turn off quotas.
1527 ASSERT(mp
->m_quotainfo
!= NULL
);
1528 ASSERT(xfs_Gqm
!= NULL
);
1529 xfs_qm_destroy_quotainfo(mp
);
1530 if (xfs_mount_reset_sbqflags(mp
)) {
1532 "Quotacheck: Failed to reset quota flags.");
1535 xfs_notice(mp
, "Quotacheck: Done.");
1540 * This is called after the superblock has been read in and we're ready to
1541 * iget the quota inodes.
1544 xfs_qm_init_quotainos(
1547 xfs_inode_t
*uip
, *gip
;
1552 ASSERT(mp
->m_quotainfo
);
1558 * Get the uquota and gquota inodes
1560 if (xfs_sb_version_hasquota(&mp
->m_sb
)) {
1561 if (XFS_IS_UQUOTA_ON(mp
) &&
1562 mp
->m_sb
.sb_uquotino
!= NULLFSINO
) {
1563 ASSERT(mp
->m_sb
.sb_uquotino
> 0);
1564 if ((error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_uquotino
,
1566 return XFS_ERROR(error
);
1568 if (XFS_IS_OQUOTA_ON(mp
) &&
1569 mp
->m_sb
.sb_gquotino
!= NULLFSINO
) {
1570 ASSERT(mp
->m_sb
.sb_gquotino
> 0);
1571 if ((error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_gquotino
,
1575 return XFS_ERROR(error
);
1579 flags
|= XFS_QMOPT_SBVERSION
;
1580 sbflags
|= (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1581 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
);
1585 * Create the two inodes, if they don't exist already. The changes
1586 * made above will get added to a transaction and logged in one of
1587 * the qino_alloc calls below. If the device is readonly,
1588 * temporarily switch to read-write to do this.
1590 if (XFS_IS_UQUOTA_ON(mp
) && uip
== NULL
) {
1591 if ((error
= xfs_qm_qino_alloc(mp
, &uip
,
1592 sbflags
| XFS_SB_UQUOTINO
,
1593 flags
| XFS_QMOPT_UQUOTA
)))
1594 return XFS_ERROR(error
);
1596 flags
&= ~XFS_QMOPT_SBVERSION
;
1598 if (XFS_IS_OQUOTA_ON(mp
) && gip
== NULL
) {
1599 flags
|= (XFS_IS_GQUOTA_ON(mp
) ?
1600 XFS_QMOPT_GQUOTA
: XFS_QMOPT_PQUOTA
);
1601 error
= xfs_qm_qino_alloc(mp
, &gip
,
1602 sbflags
| XFS_SB_GQUOTINO
, flags
);
1607 return XFS_ERROR(error
);
1611 mp
->m_quotainfo
->qi_uquotaip
= uip
;
1612 mp
->m_quotainfo
->qi_gquotaip
= gip
;
1620 * Pop the least recently used dquot off the freelist and recycle it.
1622 STATIC
struct xfs_dquot
*
1623 xfs_qm_dqreclaim_one(void)
1625 struct xfs_dquot
*dqp
;
1628 mutex_lock(&xfs_Gqm
->qm_dqfrlist_lock
);
1630 list_for_each_entry(dqp
, &xfs_Gqm
->qm_dqfrlist
, q_freelist
) {
1631 struct xfs_mount
*mp
= dqp
->q_mount
;
1633 if (!xfs_dqlock_nowait(dqp
))
1637 * This dquot has already been grabbed by dqlookup.
1638 * Remove it from the freelist and try again.
1641 trace_xfs_dqreclaim_want(dqp
);
1642 XQM_STATS_INC(xqmstats
.xs_qm_dqwants
);
1644 list_del_init(&dqp
->q_freelist
);
1645 xfs_Gqm
->qm_dqfrlist_cnt
--;
1650 ASSERT(dqp
->q_hash
);
1651 ASSERT(!list_empty(&dqp
->q_mplist
));
1654 * Try to grab the flush lock. If this dquot is in the process
1655 * of getting flushed to disk, we don't want to reclaim it.
1657 if (!xfs_dqflock_nowait(dqp
))
1661 * We have the flush lock so we know that this is not in the
1662 * process of being flushed. So, if this is dirty, flush it
1663 * DELWRI so that we don't get a freelist infested with
1666 if (XFS_DQ_IS_DIRTY(dqp
)) {
1669 trace_xfs_dqreclaim_dirty(dqp
);
1672 * We flush it delayed write, so don't bother
1673 * releasing the freelist lock.
1675 error
= xfs_qm_dqflush(dqp
, SYNC_TRYLOCK
);
1677 xfs_warn(mp
, "%s: dquot %p flush failed",
1685 * Prevent lookup now that we are going to reclaim the dquot.
1686 * Once XFS_DQ_FREEING is set lookup won't touch the dquot,
1687 * thus we can drop the lock now.
1689 dqp
->dq_flags
|= XFS_DQ_FREEING
;
1692 mutex_lock(&dqp
->q_hash
->qh_lock
);
1693 list_del_init(&dqp
->q_hashlist
);
1694 dqp
->q_hash
->qh_version
++;
1695 mutex_unlock(&dqp
->q_hash
->qh_lock
);
1697 mutex_lock(&mp
->m_quotainfo
->qi_dqlist_lock
);
1698 list_del_init(&dqp
->q_mplist
);
1699 mp
->m_quotainfo
->qi_dquots
--;
1700 mp
->m_quotainfo
->qi_dqreclaims
++;
1701 mutex_unlock(&mp
->m_quotainfo
->qi_dqlist_lock
);
1703 ASSERT(dqp
->q_nrefs
== 0);
1704 list_del_init(&dqp
->q_freelist
);
1705 xfs_Gqm
->qm_dqfrlist_cnt
--;
1707 mutex_unlock(&xfs_Gqm
->qm_dqfrlist_lock
);
1711 if (restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
1716 mutex_unlock(&xfs_Gqm
->qm_dqfrlist_lock
);
1721 * Traverse the freelist of dquots and attempt to reclaim a maximum of
1722 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
1723 * favor the lookup function ...
1726 xfs_qm_shake_freelist(
1735 while (nreclaimed
< howmany
) {
1736 dqp
= xfs_qm_dqreclaim_one();
1739 xfs_qm_dqdestroy(dqp
);
1746 * The kmem_shake interface is invoked when memory is running low.
1751 struct shrinker
*shrink
,
1752 struct shrink_control
*sc
)
1754 int ndqused
, nfree
, n
;
1755 gfp_t gfp_mask
= sc
->gfp_mask
;
1757 if (!kmem_shake_allow(gfp_mask
))
1762 nfree
= xfs_Gqm
->qm_dqfrlist_cnt
; /* free dquots */
1763 /* incore dquots in all f/s's */
1764 ndqused
= atomic_read(&xfs_Gqm
->qm_totaldquots
) - nfree
;
1766 ASSERT(ndqused
>= 0);
1768 if (nfree
<= ndqused
&& nfree
< ndquot
)
1771 ndqused
*= xfs_Gqm
->qm_dqfree_ratio
; /* target # of free dquots */
1772 n
= nfree
- ndqused
- ndquot
; /* # over target */
1774 return xfs_qm_shake_freelist(MAX(nfree
, n
));
1778 /*------------------------------------------------------------------*/
1781 * Return a new incore dquot. Depending on the number of
1782 * dquots in the system, we either allocate a new one on the kernel heap,
1783 * or reclaim a free one.
1784 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
1785 * to reclaim an existing one from the freelist.
1788 xfs_qm_dqalloc_incore(
1789 xfs_dquot_t
**O_dqpp
)
1794 * Check against high water mark to see if we want to pop
1795 * a nincompoop dquot off the freelist.
1797 if (atomic_read(&xfs_Gqm
->qm_totaldquots
) >= ndquot
) {
1799 * Try to recycle a dquot from the freelist.
1801 if ((dqp
= xfs_qm_dqreclaim_one())) {
1802 XQM_STATS_INC(xqmstats
.xs_qm_dqreclaims
);
1804 * Just zero the core here. The rest will get
1805 * reinitialized by caller. XXX we shouldn't even
1808 memset(&dqp
->q_core
, 0, sizeof(dqp
->q_core
));
1812 XQM_STATS_INC(xqmstats
.xs_qm_dqreclaim_misses
);
1816 * Allocate a brand new dquot on the kernel heap and return it
1817 * to the caller to initialize.
1819 ASSERT(xfs_Gqm
->qm_dqzone
!= NULL
);
1820 *O_dqpp
= kmem_zone_zalloc(xfs_Gqm
->qm_dqzone
, KM_SLEEP
);
1821 atomic_inc(&xfs_Gqm
->qm_totaldquots
);
1828 * Start a transaction and write the incore superblock changes to
1829 * disk. flags parameter indicates which fields have changed.
1832 xfs_qm_write_sb_changes(
1839 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
1840 if ((error
= xfs_trans_reserve(tp
, 0,
1841 mp
->m_sb
.sb_sectsize
+ 128, 0,
1843 XFS_DEFAULT_LOG_COUNT
))) {
1844 xfs_trans_cancel(tp
, 0);
1848 xfs_mod_sb(tp
, flags
);
1849 error
= xfs_trans_commit(tp
, 0);
1855 /* --------------- utility functions for vnodeops ---------------- */
1859 * Given an inode, a uid, gid and prid make sure that we have
1860 * allocated relevant dquot(s) on disk, and that we won't exceed inode
1861 * quotas by creating this file.
1862 * This also attaches dquot(s) to the given inode after locking it,
1863 * and returns the dquots corresponding to the uid and/or gid.
1865 * in : inode (unlocked)
1866 * out : udquot, gdquot with references taken and unlocked
1870 struct xfs_inode
*ip
,
1875 struct xfs_dquot
**O_udqpp
,
1876 struct xfs_dquot
**O_gdqpp
)
1878 struct xfs_mount
*mp
= ip
->i_mount
;
1879 struct xfs_dquot
*uq
, *gq
;
1883 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
1886 lockflags
= XFS_ILOCK_EXCL
;
1887 xfs_ilock(ip
, lockflags
);
1889 if ((flags
& XFS_QMOPT_INHERIT
) && XFS_INHERIT_GID(ip
))
1890 gid
= ip
->i_d
.di_gid
;
1893 * Attach the dquot(s) to this inode, doing a dquot allocation
1894 * if necessary. The dquot(s) will not be locked.
1896 if (XFS_NOT_DQATTACHED(mp
, ip
)) {
1897 error
= xfs_qm_dqattach_locked(ip
, XFS_QMOPT_DQALLOC
);
1899 xfs_iunlock(ip
, lockflags
);
1905 if ((flags
& XFS_QMOPT_UQUOTA
) && XFS_IS_UQUOTA_ON(mp
)) {
1906 if (ip
->i_d
.di_uid
!= uid
) {
1908 * What we need is the dquot that has this uid, and
1909 * if we send the inode to dqget, the uid of the inode
1910 * takes priority over what's sent in the uid argument.
1911 * We must unlock inode here before calling dqget if
1912 * we're not sending the inode, because otherwise
1913 * we'll deadlock by doing trans_reserve while
1916 xfs_iunlock(ip
, lockflags
);
1917 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
) uid
,
1922 ASSERT(error
!= ENOENT
);
1926 * Get the ilock in the right order.
1929 lockflags
= XFS_ILOCK_SHARED
;
1930 xfs_ilock(ip
, lockflags
);
1933 * Take an extra reference, because we'll return
1936 ASSERT(ip
->i_udquot
);
1937 uq
= xfs_qm_dqhold(ip
->i_udquot
);
1940 if ((flags
& XFS_QMOPT_GQUOTA
) && XFS_IS_GQUOTA_ON(mp
)) {
1941 if (ip
->i_d
.di_gid
!= gid
) {
1942 xfs_iunlock(ip
, lockflags
);
1943 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)gid
,
1950 ASSERT(error
!= ENOENT
);
1954 lockflags
= XFS_ILOCK_SHARED
;
1955 xfs_ilock(ip
, lockflags
);
1957 ASSERT(ip
->i_gdquot
);
1958 gq
= xfs_qm_dqhold(ip
->i_gdquot
);
1960 } else if ((flags
& XFS_QMOPT_PQUOTA
) && XFS_IS_PQUOTA_ON(mp
)) {
1961 if (xfs_get_projid(ip
) != prid
) {
1962 xfs_iunlock(ip
, lockflags
);
1963 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)prid
,
1970 ASSERT(error
!= ENOENT
);
1974 lockflags
= XFS_ILOCK_SHARED
;
1975 xfs_ilock(ip
, lockflags
);
1977 ASSERT(ip
->i_gdquot
);
1978 gq
= xfs_qm_dqhold(ip
->i_gdquot
);
1982 trace_xfs_dquot_dqalloc(ip
);
1984 xfs_iunlock(ip
, lockflags
);
1997 * Actually transfer ownership, and do dquot modifications.
1998 * These were already reserved.
2004 xfs_dquot_t
**IO_olddq
,
2007 xfs_dquot_t
*prevdq
;
2008 uint bfield
= XFS_IS_REALTIME_INODE(ip
) ?
2009 XFS_TRANS_DQ_RTBCOUNT
: XFS_TRANS_DQ_BCOUNT
;
2012 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2013 ASSERT(XFS_IS_QUOTA_RUNNING(ip
->i_mount
));
2018 ASSERT(prevdq
!= newdq
);
2020 xfs_trans_mod_dquot(tp
, prevdq
, bfield
, -(ip
->i_d
.di_nblocks
));
2021 xfs_trans_mod_dquot(tp
, prevdq
, XFS_TRANS_DQ_ICOUNT
, -1);
2023 /* the sparkling new dquot */
2024 xfs_trans_mod_dquot(tp
, newdq
, bfield
, ip
->i_d
.di_nblocks
);
2025 xfs_trans_mod_dquot(tp
, newdq
, XFS_TRANS_DQ_ICOUNT
, 1);
2028 * Take an extra reference, because the inode is going to keep
2029 * this dquot pointer even after the trans_commit.
2031 *IO_olddq
= xfs_qm_dqhold(newdq
);
2037 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
2040 xfs_qm_vop_chown_reserve(
2047 xfs_mount_t
*mp
= ip
->i_mount
;
2048 uint delblks
, blkflags
, prjflags
= 0;
2049 xfs_dquot_t
*unresudq
, *unresgdq
, *delblksudq
, *delblksgdq
;
2053 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2054 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
2056 delblks
= ip
->i_delayed_blks
;
2057 delblksudq
= delblksgdq
= unresudq
= unresgdq
= NULL
;
2058 blkflags
= XFS_IS_REALTIME_INODE(ip
) ?
2059 XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
;
2061 if (XFS_IS_UQUOTA_ON(mp
) && udqp
&&
2062 ip
->i_d
.di_uid
!= (uid_t
)be32_to_cpu(udqp
->q_core
.d_id
)) {
2065 * If there are delayed allocation blocks, then we have to
2066 * unreserve those from the old dquot, and add them to the
2070 ASSERT(ip
->i_udquot
);
2071 unresudq
= ip
->i_udquot
;
2074 if (XFS_IS_OQUOTA_ON(ip
->i_mount
) && gdqp
) {
2075 if (XFS_IS_PQUOTA_ON(ip
->i_mount
) &&
2076 xfs_get_projid(ip
) != be32_to_cpu(gdqp
->q_core
.d_id
))
2077 prjflags
= XFS_QMOPT_ENOSPC
;
2080 (XFS_IS_GQUOTA_ON(ip
->i_mount
) &&
2081 ip
->i_d
.di_gid
!= be32_to_cpu(gdqp
->q_core
.d_id
))) {
2084 ASSERT(ip
->i_gdquot
);
2085 unresgdq
= ip
->i_gdquot
;
2090 if ((error
= xfs_trans_reserve_quota_bydquots(tp
, ip
->i_mount
,
2091 delblksudq
, delblksgdq
, ip
->i_d
.di_nblocks
, 1,
2092 flags
| blkflags
| prjflags
)))
2096 * Do the delayed blks reservations/unreservations now. Since, these
2097 * are done without the help of a transaction, if a reservation fails
2098 * its previous reservations won't be automatically undone by trans
2099 * code. So, we have to do it manually here.
2103 * Do the reservations first. Unreservation can't fail.
2105 ASSERT(delblksudq
|| delblksgdq
);
2106 ASSERT(unresudq
|| unresgdq
);
2107 if ((error
= xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
2108 delblksudq
, delblksgdq
, (xfs_qcnt_t
)delblks
, 0,
2109 flags
| blkflags
| prjflags
)))
2111 xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
2112 unresudq
, unresgdq
, -((xfs_qcnt_t
)delblks
), 0,
2120 xfs_qm_vop_rename_dqattach(
2121 struct xfs_inode
**i_tab
)
2123 struct xfs_mount
*mp
= i_tab
[0]->i_mount
;
2126 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
2129 for (i
= 0; (i
< 4 && i_tab
[i
]); i
++) {
2130 struct xfs_inode
*ip
= i_tab
[i
];
2134 * Watch out for duplicate entries in the table.
2136 if (i
== 0 || ip
!= i_tab
[i
-1]) {
2137 if (XFS_NOT_DQATTACHED(mp
, ip
)) {
2138 error
= xfs_qm_dqattach(ip
, 0);
2148 xfs_qm_vop_create_dqattach(
2149 struct xfs_trans
*tp
,
2150 struct xfs_inode
*ip
,
2151 struct xfs_dquot
*udqp
,
2152 struct xfs_dquot
*gdqp
)
2154 struct xfs_mount
*mp
= tp
->t_mountp
;
2156 if (!XFS_IS_QUOTA_RUNNING(mp
) || !XFS_IS_QUOTA_ON(mp
))
2159 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2160 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
2163 ASSERT(ip
->i_udquot
== NULL
);
2164 ASSERT(XFS_IS_UQUOTA_ON(mp
));
2165 ASSERT(ip
->i_d
.di_uid
== be32_to_cpu(udqp
->q_core
.d_id
));
2167 ip
->i_udquot
= xfs_qm_dqhold(udqp
);
2168 xfs_trans_mod_dquot(tp
, udqp
, XFS_TRANS_DQ_ICOUNT
, 1);
2171 ASSERT(ip
->i_gdquot
== NULL
);
2172 ASSERT(XFS_IS_OQUOTA_ON(mp
));
2173 ASSERT((XFS_IS_GQUOTA_ON(mp
) ?
2174 ip
->i_d
.di_gid
: xfs_get_projid(ip
)) ==
2175 be32_to_cpu(gdqp
->q_core
.d_id
));
2177 ip
->i_gdquot
= xfs_qm_dqhold(gdqp
);
2178 xfs_trans_mod_dquot(tp
, gdqp
, XFS_TRANS_DQ_ICOUNT
, 1);