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
24 #include "xfs_trans.h"
28 #include "xfs_alloc.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_quota.h"
31 #include "xfs_mount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_btree.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_itable.h"
42 #include "xfs_rtalloc.h"
43 #include "xfs_error.h"
48 #include "xfs_buf_item.h"
49 #include "xfs_trans_space.h"
50 #include "xfs_utils.h"
54 * The global quota manager. There is only one of these for the entire
55 * system, _not_ one per file system. XQM keeps track of the overall
56 * quota functionality, including maintaining the freelist and hash
60 struct xfs_qm
*xfs_Gqm
;
63 kmem_zone_t
*qm_dqzone
;
64 kmem_zone_t
*qm_dqtrxzone
;
65 static kmem_shaker_t xfs_qm_shaker
;
67 static cred_t xfs_zerocr
;
68 static xfs_inode_t xfs_zeroino
;
70 STATIC
void xfs_qm_list_init(xfs_dqlist_t
*, char *, int);
71 STATIC
void xfs_qm_list_destroy(xfs_dqlist_t
*);
73 STATIC
void xfs_qm_freelist_init(xfs_frlist_t
*);
74 STATIC
void xfs_qm_freelist_destroy(xfs_frlist_t
*);
75 STATIC
int xfs_qm_mplist_nowait(xfs_mount_t
*);
76 STATIC
int xfs_qm_dqhashlock_nowait(xfs_dquot_t
*);
78 STATIC
int xfs_qm_init_quotainos(xfs_mount_t
*);
79 STATIC
int xfs_qm_init_quotainfo(xfs_mount_t
*);
80 STATIC
int xfs_qm_shake(int, gfp_t
);
83 extern mutex_t qcheck_lock
;
87 #define XQM_LIST_PRINT(l, NXT, title) \
89 xfs_dquot_t *dqp; int i = 0; \
90 cmn_err(CE_DEBUG, "%s (#%d)", title, (int) (l)->qh_nelems); \
91 for (dqp = (l)->qh_next; dqp != NULL; dqp = dqp->NXT) { \
92 cmn_err(CE_DEBUG, " %d. \"%d (%s)\" " \
93 "bcnt = %d, icnt = %d, refs = %d", \
94 ++i, (int) be32_to_cpu(dqp->q_core.d_id), \
95 DQFLAGTO_TYPESTR(dqp), \
96 (int) be64_to_cpu(dqp->q_core.d_bcount), \
97 (int) be64_to_cpu(dqp->q_core.d_icount), \
98 (int) dqp->q_nrefs); } \
101 #define XQM_LIST_PRINT(l, NXT, title) do { } while (0)
105 * Initialize the XQM structure.
106 * Note that there is not one quota manager per file system.
108 STATIC
struct xfs_qm
*
111 xfs_dqhash_t
*udqhash
, *gdqhash
;
117 * Initialize the dquot hash tables.
119 udqhash
= kmem_zalloc_greedy(&hsize
,
120 XFS_QM_HASHSIZE_LOW
, XFS_QM_HASHSIZE_HIGH
,
121 KM_SLEEP
| KM_MAYFAIL
| KM_LARGE
);
122 gdqhash
= kmem_zalloc(hsize
, KM_SLEEP
| KM_LARGE
);
123 hsize
/= sizeof(xfs_dqhash_t
);
126 xqm
= kmem_zalloc(sizeof(xfs_qm_t
), KM_SLEEP
);
127 xqm
->qm_dqhashmask
= hsize
- 1;
128 xqm
->qm_usr_dqhtable
= udqhash
;
129 xqm
->qm_grp_dqhtable
= gdqhash
;
130 ASSERT(xqm
->qm_usr_dqhtable
!= NULL
);
131 ASSERT(xqm
->qm_grp_dqhtable
!= NULL
);
133 for (i
= 0; i
< hsize
; i
++) {
134 xfs_qm_list_init(&(xqm
->qm_usr_dqhtable
[i
]), "uxdqh", i
);
135 xfs_qm_list_init(&(xqm
->qm_grp_dqhtable
[i
]), "gxdqh", i
);
139 * Freelist of all dquots of all file systems
141 xfs_qm_freelist_init(&(xqm
->qm_dqfreelist
));
144 * dquot zone. we register our own low-memory callback.
147 xqm
->qm_dqzone
= kmem_zone_init(sizeof(xfs_dquot_t
),
149 qm_dqzone
= xqm
->qm_dqzone
;
151 xqm
->qm_dqzone
= qm_dqzone
;
153 xfs_qm_shaker
= kmem_shake_register(xfs_qm_shake
);
156 * The t_dqinfo portion of transactions.
159 xqm
->qm_dqtrxzone
= kmem_zone_init(sizeof(xfs_dquot_acct_t
),
161 qm_dqtrxzone
= xqm
->qm_dqtrxzone
;
163 xqm
->qm_dqtrxzone
= qm_dqtrxzone
;
165 atomic_set(&xqm
->qm_totaldquots
, 0);
166 xqm
->qm_dqfree_ratio
= XFS_QM_DQFREE_RATIO
;
169 mutex_init(&qcheck_lock
);
175 * Destroy the global quota manager when its reference count goes to zero.
184 ASSERT(xqm
->qm_nrefs
== 0);
185 kmem_shake_deregister(xfs_qm_shaker
);
186 hsize
= xqm
->qm_dqhashmask
+ 1;
187 for (i
= 0; i
< hsize
; i
++) {
188 xfs_qm_list_destroy(&(xqm
->qm_usr_dqhtable
[i
]));
189 xfs_qm_list_destroy(&(xqm
->qm_grp_dqhtable
[i
]));
191 kmem_free(xqm
->qm_usr_dqhtable
, hsize
* sizeof(xfs_dqhash_t
));
192 kmem_free(xqm
->qm_grp_dqhtable
, hsize
* sizeof(xfs_dqhash_t
));
193 xqm
->qm_usr_dqhtable
= NULL
;
194 xqm
->qm_grp_dqhtable
= NULL
;
195 xqm
->qm_dqhashmask
= 0;
196 xfs_qm_freelist_destroy(&(xqm
->qm_dqfreelist
));
198 mutex_destroy(&qcheck_lock
);
200 kmem_free(xqm
, sizeof(xfs_qm_t
));
204 * Called at mount time to let XQM know that another file system is
205 * starting quotas. This isn't crucial information as the individual mount
206 * structures are pretty independent, but it helps the XQM keep a
207 * global view of what's going on.
211 xfs_qm_hold_quotafs_ref(
212 struct xfs_mount
*mp
)
215 * Need to lock the xfs_Gqm structure for things like this. For example,
216 * the structure could disappear between the entry to this routine and
217 * a HOLD operation if not locked.
219 XFS_QM_LOCK(xfs_Gqm
);
222 xfs_Gqm
= xfs_Gqm_init();
224 * We can keep a list of all filesystems with quotas mounted for
225 * debugging and statistical purposes, but ...
226 * Just take a reference and get out.
228 XFS_QM_HOLD(xfs_Gqm
);
229 XFS_QM_UNLOCK(xfs_Gqm
);
236 * Release the reference that a filesystem took at mount time,
237 * so that we know when we need to destroy the entire quota manager.
241 xfs_qm_rele_quotafs_ref(
242 struct xfs_mount
*mp
)
244 xfs_dquot_t
*dqp
, *nextdqp
;
247 ASSERT(xfs_Gqm
->qm_nrefs
> 0);
250 * Go thru the freelist and destroy all inactive dquots.
252 xfs_qm_freelist_lock(xfs_Gqm
);
254 for (dqp
= xfs_Gqm
->qm_dqfreelist
.qh_next
;
255 dqp
!= (xfs_dquot_t
*)&(xfs_Gqm
->qm_dqfreelist
); ) {
257 nextdqp
= dqp
->dq_flnext
;
258 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
259 ASSERT(dqp
->q_mount
== NULL
);
260 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
261 ASSERT(dqp
->HL_PREVP
== NULL
);
262 ASSERT(dqp
->MPL_PREVP
== NULL
);
263 XQM_FREELIST_REMOVE(dqp
);
265 xfs_qm_dqdestroy(dqp
);
271 xfs_qm_freelist_unlock(xfs_Gqm
);
274 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
277 XFS_QM_LOCK(xfs_Gqm
);
278 XFS_QM_RELE(xfs_Gqm
);
279 if (xfs_Gqm
->qm_nrefs
== 0) {
280 xfs_qm_destroy(xfs_Gqm
);
283 XFS_QM_UNLOCK(xfs_Gqm
);
287 * This is called at mount time from xfs_mountfs to initialize the quotainfo
288 * structure and start the global quota manager (xfs_Gqm) if it hasn't done
289 * so already. Note that the superblock has not been read in yet.
292 xfs_qm_mount_quotainit(
297 * User, projects or group quotas has to be on.
299 ASSERT(flags
& (XFSMNT_UQUOTA
| XFSMNT_PQUOTA
| XFSMNT_GQUOTA
));
302 * Initialize the flags in the mount structure. From this point
303 * onwards we look at m_qflags to figure out if quotas's ON/OFF, etc.
304 * Note that we enforce nothing if accounting is off.
305 * ie. XFSMNT_*QUOTA must be ON for XFSMNT_*QUOTAENF.
306 * It isn't necessary to take the quotaoff lock to do this; this is
309 if (flags
& XFSMNT_UQUOTA
) {
310 mp
->m_qflags
|= (XFS_UQUOTA_ACCT
| XFS_UQUOTA_ACTIVE
);
311 if (flags
& XFSMNT_UQUOTAENF
)
312 mp
->m_qflags
|= XFS_UQUOTA_ENFD
;
314 if (flags
& XFSMNT_GQUOTA
) {
315 mp
->m_qflags
|= (XFS_GQUOTA_ACCT
| XFS_GQUOTA_ACTIVE
);
316 if (flags
& XFSMNT_GQUOTAENF
)
317 mp
->m_qflags
|= XFS_OQUOTA_ENFD
;
318 } else if (flags
& XFSMNT_PQUOTA
) {
319 mp
->m_qflags
|= (XFS_PQUOTA_ACCT
| XFS_PQUOTA_ACTIVE
);
320 if (flags
& XFSMNT_PQUOTAENF
)
321 mp
->m_qflags
|= XFS_OQUOTA_ENFD
;
326 * Just destroy the quotainfo structure.
329 xfs_qm_unmount_quotadestroy(
333 xfs_qm_destroy_quotainfo(mp
);
338 * This is called from xfs_mountfs to start quotas and initialize all
339 * necessary data structures like quotainfo. This is also responsible for
340 * running a quotacheck as necessary. We are guaranteed that the superblock
341 * is consistently read in at this point.
354 * If quotas on realtime volumes is not supported, we disable
355 * quotas immediately.
357 if (mp
->m_sb
.sb_rextents
) {
359 "Cannot turn on quotas for realtime filesystem %s",
365 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
368 * Allocate the quotainfo structure inside the mount struct, and
369 * create quotainode(s), and change/rev superblock if necessary.
371 if ((error
= xfs_qm_init_quotainfo(mp
))) {
373 * We must turn off quotas.
375 ASSERT(mp
->m_quotainfo
== NULL
);
380 * If any of the quotas are not consistent, do a quotacheck.
382 if (XFS_QM_NEED_QUOTACHECK(mp
) &&
383 !(mfsi_flags
& XFS_MFSI_NO_QUOTACHECK
)) {
384 if ((error
= xfs_qm_quotacheck(mp
))) {
385 /* Quotacheck has failed and quotas have
388 return XFS_ERROR(error
);
392 * If one type of quotas is off, then it will lose its
393 * quotachecked status, since we won't be doing accounting for
396 if (!XFS_IS_UQUOTA_ON(mp
)) {
397 mp
->m_qflags
&= ~XFS_UQUOTA_CHKD
;
399 if (!(XFS_IS_GQUOTA_ON(mp
) || XFS_IS_PQUOTA_ON(mp
))) {
400 mp
->m_qflags
&= ~XFS_OQUOTA_CHKD
;
405 * We actually don't have to acquire the SB_LOCK at all.
406 * This can only be called from mount, and that's single threaded. XXX
409 sbf
= mp
->m_sb
.sb_qflags
;
410 mp
->m_sb
.sb_qflags
= mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
;
411 XFS_SB_UNLOCK(mp
, s
);
413 if (sbf
!= (mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
)) {
414 if (xfs_qm_write_sb_changes(mp
, XFS_SB_QFLAGS
)) {
416 * We could only have been turning quotas off.
417 * We aren't in very good shape actually because
418 * the incore structures are convinced that quotas are
419 * off, but the on disk superblock doesn't know that !
421 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp
)));
422 xfs_fs_cmn_err(CE_ALERT
, mp
,
423 "XFS mount_quotas: Superblock update failed!");
428 xfs_fs_cmn_err(CE_WARN
, mp
,
429 "Failed to initialize disk quotas.");
431 return XFS_ERROR(error
);
435 * Called from the vfsops layer.
438 xfs_qm_unmount_quotas(
441 xfs_inode_t
*uqp
, *gqp
;
445 * Release the dquots that root inode, et al might be holding,
446 * before we flush quotas and blow away the quotainfo structure.
448 ASSERT(mp
->m_rootip
);
449 xfs_qm_dqdetach(mp
->m_rootip
);
451 xfs_qm_dqdetach(mp
->m_rbmip
);
453 xfs_qm_dqdetach(mp
->m_rsumip
);
456 * Flush out the quota inodes.
459 if (mp
->m_quotainfo
) {
460 if ((uqp
= mp
->m_quotainfo
->qi_uquotaip
) != NULL
) {
461 xfs_ilock(uqp
, XFS_ILOCK_EXCL
);
463 error
= xfs_iflush(uqp
, XFS_IFLUSH_SYNC
);
464 xfs_iunlock(uqp
, XFS_ILOCK_EXCL
);
465 if (unlikely(error
== EFSCORRUPTED
)) {
466 XFS_ERROR_REPORT("xfs_qm_unmount_quotas(1)",
467 XFS_ERRLEVEL_LOW
, mp
);
471 if ((gqp
= mp
->m_quotainfo
->qi_gquotaip
) != NULL
) {
472 xfs_ilock(gqp
, XFS_ILOCK_EXCL
);
474 error
= xfs_iflush(gqp
, XFS_IFLUSH_SYNC
);
475 xfs_iunlock(gqp
, XFS_ILOCK_EXCL
);
476 if (unlikely(error
== EFSCORRUPTED
)) {
477 XFS_ERROR_REPORT("xfs_qm_unmount_quotas(2)",
478 XFS_ERRLEVEL_LOW
, mp
);
484 XFS_PURGE_INODE(uqp
);
485 mp
->m_quotainfo
->qi_uquotaip
= NULL
;
488 XFS_PURGE_INODE(gqp
);
489 mp
->m_quotainfo
->qi_gquotaip
= NULL
;
492 return XFS_ERROR(error
);
496 * Flush all dquots of the given file system to disk. The dquots are
497 * _not_ purged from memory here, just their data written to disk.
509 if (mp
->m_quotainfo
== NULL
)
513 xfs_qm_mplist_lock(mp
);
514 FOREACH_DQUOT_IN_MP(dqp
, mp
) {
516 if (! XFS_DQ_IS_DIRTY(dqp
)) {
520 xfs_dqtrace_entry(dqp
, "FLUSHALL: DQDIRTY");
521 /* XXX a sentinel would be better */
522 recl
= XFS_QI_MPLRECLAIMS(mp
);
523 if (! xfs_qm_dqflock_nowait(dqp
)) {
525 * If we can't grab the flush lock then check
526 * to see if the dquot has been flushed delayed
527 * write. If so, grab its buffer and send it
528 * out immediately. We'll be able to acquire
529 * the flush lock when the I/O completes.
531 xfs_qm_dqflock_pushbuf_wait(dqp
);
534 * Let go of the mplist lock. We don't want to hold it
535 * across a disk write.
537 xfs_qm_mplist_unlock(mp
);
538 error
= xfs_qm_dqflush(dqp
, flags
);
543 xfs_qm_mplist_lock(mp
);
544 if (recl
!= XFS_QI_MPLRECLAIMS(mp
)) {
545 xfs_qm_mplist_unlock(mp
);
546 /* XXX restart limit */
551 xfs_qm_mplist_unlock(mp
);
556 * Release the group dquot pointers the user dquots may be
557 * carrying around as a hint. mplist is locked on entry and exit.
560 xfs_qm_detach_gdquots(
563 xfs_dquot_t
*dqp
, *gdqp
;
567 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp
));
568 dqp
= XFS_QI_MPLNEXT(mp
);
571 if ((gdqp
= dqp
->q_gdquot
)) {
573 dqp
->q_gdquot
= NULL
;
579 * Can't hold the mplist lock across a dqput.
580 * XXXmust convert to marker based iterations here.
582 nrecl
= XFS_QI_MPLRECLAIMS(mp
);
583 xfs_qm_mplist_unlock(mp
);
586 xfs_qm_mplist_lock(mp
);
587 if (nrecl
!= XFS_QI_MPLRECLAIMS(mp
))
595 * Go through all the incore dquots of this file system and take them
596 * off the mplist and hashlist, if the dquot type matches the dqtype
597 * parameter. This is used when turning off quota accounting for
598 * users and/or groups, as well as when the filesystem is unmounting.
603 uint flags
) /* QUOTAOFF/UMOUNTING/UQUOTA/PQUOTA/GQUOTA */
608 xfs_dquot_t
*nextdqp
;
611 if (mp
->m_quotainfo
== NULL
)
614 dqtype
= (flags
& XFS_QMOPT_UQUOTA
) ? XFS_DQ_USER
: 0;
615 dqtype
|= (flags
& XFS_QMOPT_PQUOTA
) ? XFS_DQ_PROJ
: 0;
616 dqtype
|= (flags
& XFS_QMOPT_GQUOTA
) ? XFS_DQ_GROUP
: 0;
618 xfs_qm_mplist_lock(mp
);
621 * In the first pass through all incore dquots of this filesystem,
622 * we release the group dquot pointers the user dquots may be
623 * carrying around as a hint. We need to do this irrespective of
624 * what's being turned off.
626 xfs_qm_detach_gdquots(mp
);
630 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp
));
632 * Try to get rid of all of the unwanted dquots. The idea is to
633 * get them off mplist and hashlist, but leave them on freelist.
635 dqp
= XFS_QI_MPLNEXT(mp
);
638 * It's OK to look at the type without taking dqlock here.
639 * We're holding the mplist lock here, and that's needed for
642 if ((dqp
->dq_flags
& dqtype
) == 0) {
647 if (! xfs_qm_dqhashlock_nowait(dqp
)) {
648 nrecl
= XFS_QI_MPLRECLAIMS(mp
);
649 xfs_qm_mplist_unlock(mp
);
650 XFS_DQ_HASH_LOCK(dqp
->q_hash
);
651 xfs_qm_mplist_lock(mp
);
654 * XXXTheoretically, we can get into a very long
655 * ping pong game here.
656 * No one can be adding dquots to the mplist at
657 * this point, but somebody might be taking things off.
659 if (nrecl
!= XFS_QI_MPLRECLAIMS(mp
)) {
660 XFS_DQ_HASH_UNLOCK(dqp
->q_hash
);
666 * Take the dquot off the mplist and hashlist. It may remain on
667 * freelist in INACTIVE state.
669 nextdqp
= dqp
->MPL_NEXT
;
670 nmisses
+= xfs_qm_dqpurge(dqp
, flags
);
673 xfs_qm_mplist_unlock(mp
);
685 * Purge the dquot cache.
686 * None of the dquots should really be busy at this point.
688 if (mp
->m_quotainfo
) {
689 while ((ndquots
= xfs_qm_dqpurge_int(mp
, flags
))) {
703 xfs_dquot_t
*udqhint
, /* hint */
704 xfs_dquot_t
**IO_idqpp
)
709 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
712 * See if we already have it in the inode itself. IO_idqpp is
713 * &i_udquot or &i_gdquot. This made the code look weird, but
714 * made the logic a lot simpler.
716 if ((dqp
= *IO_idqpp
)) {
719 xfs_dqtrace_entry(dqp
, "DQATTACH: found in ip");
724 * udqhint is the i_udquot field in inode, and is non-NULL only
725 * when the type arg is group/project. Its purpose is to save a
726 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
729 ASSERT(!udqhint
|| type
== XFS_DQ_GROUP
|| type
== XFS_DQ_PROJ
);
730 if (udqhint
&& !dolock
)
734 * No need to take dqlock to look at the id.
735 * The ID can't change until it gets reclaimed, and it won't
736 * be reclaimed as long as we have a ref from inode and we hold
740 (dqp
= udqhint
->q_gdquot
) &&
741 (be32_to_cpu(dqp
->q_core
.d_id
) == id
)) {
742 ASSERT(XFS_DQ_IS_LOCKED(udqhint
));
745 ASSERT(*IO_idqpp
== NULL
);
749 xfs_dqunlock(udqhint
);
754 * We can't hold a dquot lock when we call the dqget code.
755 * We'll deadlock in no time, because of (not conforming to)
756 * lock ordering - the inodelock comes before any dquot lock,
757 * and we may drop and reacquire the ilock in xfs_qm_dqget().
760 xfs_dqunlock(udqhint
);
762 * Find the dquot from somewhere. This bumps the
763 * reference count of dquot and returns it locked.
764 * This can return ENOENT if dquot didn't exist on
765 * disk and we didn't ask it to allocate;
766 * ESRCH if quotas got turned off suddenly.
768 if ((error
= xfs_qm_dqget(ip
->i_mount
, ip
, id
, type
,
769 doalloc
|XFS_QMOPT_DOWARN
, &dqp
))) {
770 if (udqhint
&& dolock
)
775 xfs_dqtrace_entry(dqp
, "DQATTACH: found by dqget");
777 * dqget may have dropped and re-acquired the ilock, but it guarantees
778 * that the dquot returned is the one that should go in the inode.
782 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
792 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
793 if (! xfs_qm_dqlock_nowait(udqhint
)) {
802 ASSERT(XFS_DQ_IS_LOCKED(udqhint
));
806 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
814 * Given a udquot and gdquot, attach a ptr to the group dquot in the
815 * udquot as a hint for future lookups. The idea sounds simple, but the
816 * execution isn't, because the udquot might have a group dquot attached
817 * already and getting rid of that gets us into lock ordering constraints.
818 * The process is complicated more by the fact that the dquots may or may not
819 * be locked on entry.
822 xfs_qm_dqattach_grouphint(
831 ASSERT(XFS_DQ_IS_LOCKED(udq
));
832 ASSERT(XFS_DQ_IS_LOCKED(gdq
));
838 if ((tmp
= udq
->q_gdquot
)) {
845 udq
->q_gdquot
= NULL
;
847 * We can't keep any dqlocks when calling dqrele,
848 * because the freelist lock comes before dqlocks.
854 * we took a hard reference once upon a time in dqget,
855 * so give it back when the udquot no longer points at it
856 * dqput() does the unlocking of the dquot.
864 ASSERT(XFS_DQ_IS_LOCKED(udq
));
870 ASSERT(XFS_DQ_IS_LOCKED(udq
));
871 ASSERT(XFS_DQ_IS_LOCKED(gdq
));
873 * Somebody could have attached a gdquot here,
874 * when we dropped the uqlock. If so, just do nothing.
876 if (udq
->q_gdquot
== NULL
) {
888 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
890 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
891 * If XFS_QMOPT_DQLOCK, the dquot(s) will be returned locked. This option pretty
892 * much made this code a complete mess, but it has been pretty useful.
893 * If XFS_QMOPT_ILOCKED, then inode sent is already locked EXCL.
894 * Inode may get unlocked and relocked in here, and the caller must deal with
902 xfs_mount_t
*mp
= ip
->i_mount
;
906 if ((! XFS_IS_QUOTA_ON(mp
)) ||
907 (! XFS_NOT_DQATTACHED(mp
, ip
)) ||
908 (ip
->i_ino
== mp
->m_sb
.sb_uquotino
) ||
909 (ip
->i_ino
== mp
->m_sb
.sb_gquotino
))
912 ASSERT((flags
& XFS_QMOPT_ILOCKED
) == 0 ||
913 XFS_ISLOCKED_INODE_EXCL(ip
));
915 if (! (flags
& XFS_QMOPT_ILOCKED
))
916 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
918 if (XFS_IS_UQUOTA_ON(mp
)) {
919 error
= xfs_qm_dqattach_one(ip
, ip
->i_d
.di_uid
, XFS_DQ_USER
,
920 flags
& XFS_QMOPT_DQALLOC
,
921 flags
& XFS_QMOPT_DQLOCK
,
922 NULL
, &ip
->i_udquot
);
927 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
928 if (XFS_IS_OQUOTA_ON(mp
)) {
929 error
= XFS_IS_GQUOTA_ON(mp
) ?
930 xfs_qm_dqattach_one(ip
, ip
->i_d
.di_gid
, XFS_DQ_GROUP
,
931 flags
& XFS_QMOPT_DQALLOC
,
932 flags
& XFS_QMOPT_DQLOCK
,
933 ip
->i_udquot
, &ip
->i_gdquot
) :
934 xfs_qm_dqattach_one(ip
, ip
->i_d
.di_projid
, XFS_DQ_PROJ
,
935 flags
& XFS_QMOPT_DQALLOC
,
936 flags
& XFS_QMOPT_DQLOCK
,
937 ip
->i_udquot
, &ip
->i_gdquot
);
939 * Don't worry about the udquot that we may have
940 * attached above. It'll get detached, if not already.
948 * Attach this group quota to the user quota as a hint.
949 * This WON'T, in general, result in a thrash.
952 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
953 ASSERT(ip
->i_udquot
);
954 ASSERT(ip
->i_gdquot
);
957 * We may or may not have the i_udquot locked at this point,
958 * but this check is OK since we don't depend on the i_gdquot to
959 * be accurate 100% all the time. It is just a hint, and this
960 * will succeed in general.
962 if (ip
->i_udquot
->q_gdquot
== ip
->i_gdquot
)
965 * Attach i_gdquot to the gdquot hint inside the i_udquot.
967 xfs_qm_dqattach_grouphint(ip
->i_udquot
, ip
->i_gdquot
,
968 flags
& XFS_QMOPT_DQLOCK
);
976 if (flags
& XFS_QMOPT_DQLOCK
)
977 ASSERT(XFS_DQ_IS_LOCKED(ip
->i_udquot
));
980 if (flags
& XFS_QMOPT_DQLOCK
)
981 ASSERT(XFS_DQ_IS_LOCKED(ip
->i_gdquot
));
983 if (XFS_IS_UQUOTA_ON(mp
))
984 ASSERT(ip
->i_udquot
);
985 if (XFS_IS_OQUOTA_ON(mp
))
986 ASSERT(ip
->i_gdquot
);
990 if (! (flags
& XFS_QMOPT_ILOCKED
))
991 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
995 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
1001 * Release dquots (and their references) if any.
1002 * The inode should be locked EXCL except when this's called by
1009 if (!(ip
->i_udquot
|| ip
->i_gdquot
))
1012 ASSERT(ip
->i_ino
!= ip
->i_mount
->m_sb
.sb_uquotino
);
1013 ASSERT(ip
->i_ino
!= ip
->i_mount
->m_sb
.sb_gquotino
);
1015 xfs_dqtrace_entry_ino(ip
->i_udquot
, "DQDETTACH", ip
);
1016 xfs_qm_dqrele(ip
->i_udquot
);
1017 ip
->i_udquot
= NULL
;
1020 xfs_dqtrace_entry_ino(ip
->i_gdquot
, "DQDETTACH", ip
);
1021 xfs_qm_dqrele(ip
->i_gdquot
);
1022 ip
->i_gdquot
= NULL
;
1027 * This is called by VFS_SYNC and flags arg determines the caller,
1028 * and its motives, as done in xfs_sync.
1030 * vfs_sync: SYNC_FSDATA|SYNC_ATTR|SYNC_BDFLUSH 0x31
1031 * syscall sync: SYNC_FSDATA|SYNC_ATTR|SYNC_DELWRI 0x25
1032 * umountroot : SYNC_WAIT | SYNC_CLOSE | SYNC_ATTR | SYNC_FSDATA
1048 * We won't block unless we are asked to.
1050 nowait
= (boolean_t
)(flags
& SYNC_BDFLUSH
|| (flags
& SYNC_WAIT
) == 0);
1053 xfs_qm_mplist_lock(mp
);
1055 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
1056 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
1057 * when we have the mplist lock, we know that dquots will be consistent
1058 * as long as we have it locked.
1060 if (! XFS_IS_QUOTA_ON(mp
)) {
1061 xfs_qm_mplist_unlock(mp
);
1064 FOREACH_DQUOT_IN_MP(dqp
, mp
) {
1066 * If this is vfs_sync calling, then skip the dquots that
1067 * don't 'seem' to be dirty. ie. don't acquire dqlock.
1068 * This is very similar to what xfs_sync does with inodes.
1070 if (flags
& SYNC_BDFLUSH
) {
1071 if (! XFS_DQ_IS_DIRTY(dqp
))
1077 * Try to acquire the dquot lock. We are NOT out of
1078 * lock order, but we just don't want to wait for this
1079 * lock, unless somebody wanted us to.
1081 if (! xfs_qm_dqlock_nowait(dqp
))
1088 * Now, find out for sure if this dquot is dirty or not.
1090 if (! XFS_DQ_IS_DIRTY(dqp
)) {
1095 /* XXX a sentinel would be better */
1096 recl
= XFS_QI_MPLRECLAIMS(mp
);
1097 if (! xfs_qm_dqflock_nowait(dqp
)) {
1103 * If we can't grab the flush lock then if the caller
1104 * really wanted us to give this our best shot, so
1105 * see if we can give a push to the buffer before we wait
1106 * on the flush lock. At this point, we know that
1107 * even though the dquot is being flushed,
1108 * it has (new) dirty data.
1110 xfs_qm_dqflock_pushbuf_wait(dqp
);
1113 * Let go of the mplist lock. We don't want to hold it
1114 * across a disk write
1116 flush_flags
= (nowait
) ? XFS_QMOPT_DELWRI
: XFS_QMOPT_SYNC
;
1117 xfs_qm_mplist_unlock(mp
);
1118 xfs_dqtrace_entry(dqp
, "XQM_SYNC: DQFLUSH");
1119 error
= xfs_qm_dqflush(dqp
, flush_flags
);
1121 if (error
&& XFS_FORCED_SHUTDOWN(mp
))
1122 return 0; /* Need to prevent umount failure */
1126 xfs_qm_mplist_lock(mp
);
1127 if (recl
!= XFS_QI_MPLRECLAIMS(mp
)) {
1128 if (++restarts
>= XFS_QM_SYNC_MAX_RESTARTS
)
1131 xfs_qm_mplist_unlock(mp
);
1136 xfs_qm_mplist_unlock(mp
);
1142 * This initializes all the quota information that's kept in the
1146 xfs_qm_init_quotainfo(
1149 xfs_quotainfo_t
*qinf
;
1153 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1156 * Tell XQM that we exist as soon as possible.
1158 if ((error
= xfs_qm_hold_quotafs_ref(mp
))) {
1162 qinf
= mp
->m_quotainfo
= kmem_zalloc(sizeof(xfs_quotainfo_t
), KM_SLEEP
);
1165 * See if quotainodes are setup, and if not, allocate them,
1166 * and change the superblock accordingly.
1168 if ((error
= xfs_qm_init_quotainos(mp
))) {
1169 kmem_free(qinf
, sizeof(xfs_quotainfo_t
));
1170 mp
->m_quotainfo
= NULL
;
1174 spinlock_init(&qinf
->qi_pinlock
, "xfs_qinf_pin");
1175 xfs_qm_list_init(&qinf
->qi_dqlist
, "mpdqlist", 0);
1176 qinf
->qi_dqreclaims
= 0;
1178 /* mutex used to serialize quotaoffs */
1179 mutex_init(&qinf
->qi_quotaofflock
);
1181 /* Precalc some constants */
1182 qinf
->qi_dqchunklen
= XFS_FSB_TO_BB(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
1183 ASSERT(qinf
->qi_dqchunklen
);
1184 qinf
->qi_dqperchunk
= BBTOB(qinf
->qi_dqchunklen
);
1185 do_div(qinf
->qi_dqperchunk
, sizeof(xfs_dqblk_t
));
1187 mp
->m_qflags
|= (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_CHKD
);
1190 * We try to get the limits from the superuser's limits fields.
1191 * This is quite hacky, but it is standard quota practice.
1192 * We look at the USR dquot with id == 0 first, but if user quotas
1193 * are not enabled we goto the GRP dquot with id == 0.
1194 * We don't really care to keep separate default limits for user
1195 * and group quotas, at least not at this point.
1197 error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)0,
1198 XFS_IS_UQUOTA_RUNNING(mp
) ? XFS_DQ_USER
:
1199 (XFS_IS_GQUOTA_RUNNING(mp
) ? XFS_DQ_GROUP
:
1201 XFS_QMOPT_DQSUSER
|XFS_QMOPT_DOWARN
,
1204 xfs_disk_dquot_t
*ddqp
= &dqp
->q_core
;
1207 * The warnings and timers set the grace period given to
1208 * a user or group before he or she can not perform any
1209 * more writing. If it is zero, a default is used.
1211 qinf
->qi_btimelimit
= ddqp
->d_btimer
?
1212 be32_to_cpu(ddqp
->d_btimer
) : XFS_QM_BTIMELIMIT
;
1213 qinf
->qi_itimelimit
= ddqp
->d_itimer
?
1214 be32_to_cpu(ddqp
->d_itimer
) : XFS_QM_ITIMELIMIT
;
1215 qinf
->qi_rtbtimelimit
= ddqp
->d_rtbtimer
?
1216 be32_to_cpu(ddqp
->d_rtbtimer
) : XFS_QM_RTBTIMELIMIT
;
1217 qinf
->qi_bwarnlimit
= ddqp
->d_bwarns
?
1218 be16_to_cpu(ddqp
->d_bwarns
) : XFS_QM_BWARNLIMIT
;
1219 qinf
->qi_iwarnlimit
= ddqp
->d_iwarns
?
1220 be16_to_cpu(ddqp
->d_iwarns
) : XFS_QM_IWARNLIMIT
;
1221 qinf
->qi_rtbwarnlimit
= ddqp
->d_rtbwarns
?
1222 be16_to_cpu(ddqp
->d_rtbwarns
) : XFS_QM_RTBWARNLIMIT
;
1223 qinf
->qi_bhardlimit
= be64_to_cpu(ddqp
->d_blk_hardlimit
);
1224 qinf
->qi_bsoftlimit
= be64_to_cpu(ddqp
->d_blk_softlimit
);
1225 qinf
->qi_ihardlimit
= be64_to_cpu(ddqp
->d_ino_hardlimit
);
1226 qinf
->qi_isoftlimit
= be64_to_cpu(ddqp
->d_ino_softlimit
);
1227 qinf
->qi_rtbhardlimit
= be64_to_cpu(ddqp
->d_rtb_hardlimit
);
1228 qinf
->qi_rtbsoftlimit
= be64_to_cpu(ddqp
->d_rtb_softlimit
);
1231 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1232 * we don't want this dquot cached. We haven't done a
1233 * quotacheck yet, and quotacheck doesn't like incore dquots.
1235 xfs_qm_dqdestroy(dqp
);
1237 qinf
->qi_btimelimit
= XFS_QM_BTIMELIMIT
;
1238 qinf
->qi_itimelimit
= XFS_QM_ITIMELIMIT
;
1239 qinf
->qi_rtbtimelimit
= XFS_QM_RTBTIMELIMIT
;
1240 qinf
->qi_bwarnlimit
= XFS_QM_BWARNLIMIT
;
1241 qinf
->qi_iwarnlimit
= XFS_QM_IWARNLIMIT
;
1242 qinf
->qi_rtbwarnlimit
= XFS_QM_RTBWARNLIMIT
;
1250 * Gets called when unmounting a filesystem or when all quotas get
1252 * This purges the quota inodes, destroys locks and frees itself.
1255 xfs_qm_destroy_quotainfo(
1258 xfs_quotainfo_t
*qi
;
1260 qi
= mp
->m_quotainfo
;
1262 ASSERT(xfs_Gqm
!= NULL
);
1265 * Release the reference that XQM kept, so that we know
1266 * when the XQM structure should be freed. We cannot assume
1267 * that xfs_Gqm is non-null after this point.
1269 xfs_qm_rele_quotafs_ref(mp
);
1271 spinlock_destroy(&qi
->qi_pinlock
);
1272 xfs_qm_list_destroy(&qi
->qi_dqlist
);
1274 if (qi
->qi_uquotaip
) {
1275 XFS_PURGE_INODE(qi
->qi_uquotaip
);
1276 qi
->qi_uquotaip
= NULL
; /* paranoia */
1278 if (qi
->qi_gquotaip
) {
1279 XFS_PURGE_INODE(qi
->qi_gquotaip
);
1280 qi
->qi_gquotaip
= NULL
;
1282 mutex_destroy(&qi
->qi_quotaofflock
);
1283 kmem_free(qi
, sizeof(xfs_quotainfo_t
));
1284 mp
->m_quotainfo
= NULL
;
1289 /* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1298 mutex_init(&list
->qh_lock
);
1299 list
->qh_next
= NULL
;
1300 list
->qh_version
= 0;
1301 list
->qh_nelems
= 0;
1305 xfs_qm_list_destroy(
1308 mutex_destroy(&(list
->qh_lock
));
1313 * Stripped down version of dqattach. This doesn't attach, or even look at the
1314 * dquots attached to the inode. The rationale is that there won't be any
1315 * attached at the time this is called from quotacheck.
1318 xfs_qm_dqget_noattach(
1320 xfs_dquot_t
**O_udqpp
,
1321 xfs_dquot_t
**O_gdqpp
)
1325 xfs_dquot_t
*udqp
, *gdqp
;
1327 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
1332 if (XFS_IS_UQUOTA_ON(mp
)) {
1333 ASSERT(ip
->i_udquot
== NULL
);
1335 * We want the dquot allocated if it doesn't exist.
1337 if ((error
= xfs_qm_dqget(mp
, ip
, ip
->i_d
.di_uid
, XFS_DQ_USER
,
1338 XFS_QMOPT_DQALLOC
| XFS_QMOPT_DOWARN
,
1341 * Shouldn't be able to turn off quotas here.
1343 ASSERT(error
!= ESRCH
);
1344 ASSERT(error
!= ENOENT
);
1350 if (XFS_IS_OQUOTA_ON(mp
)) {
1351 ASSERT(ip
->i_gdquot
== NULL
);
1354 error
= XFS_IS_GQUOTA_ON(mp
) ?
1355 xfs_qm_dqget(mp
, ip
,
1356 ip
->i_d
.di_gid
, XFS_DQ_GROUP
,
1357 XFS_QMOPT_DQALLOC
|XFS_QMOPT_DOWARN
,
1359 xfs_qm_dqget(mp
, ip
,
1360 ip
->i_d
.di_projid
, XFS_DQ_PROJ
,
1361 XFS_QMOPT_DQALLOC
|XFS_QMOPT_DOWARN
,
1365 xfs_qm_dqrele(udqp
);
1366 ASSERT(error
!= ESRCH
);
1367 ASSERT(error
!= ENOENT
);
1372 /* Reacquire the locks in the right order */
1374 if (! xfs_qm_dqlock_nowait(udqp
)) {
1386 if (udqp
) ASSERT(XFS_DQ_IS_LOCKED(udqp
));
1387 if (gdqp
) ASSERT(XFS_DQ_IS_LOCKED(gdqp
));
1393 * Create an inode and return with a reference already taken, but unlocked
1394 * This is how we create quota inodes
1408 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_QINOCREATE
);
1409 if ((error
= xfs_trans_reserve(tp
,
1410 XFS_QM_QINOCREATE_SPACE_RES(mp
),
1411 XFS_CREATE_LOG_RES(mp
), 0,
1412 XFS_TRANS_PERM_LOG_RES
,
1413 XFS_CREATE_LOG_COUNT
))) {
1414 xfs_trans_cancel(tp
, 0);
1418 if ((error
= xfs_dir_ialloc(&tp
, &xfs_zeroino
, S_IFREG
, 1, 0,
1419 &xfs_zerocr
, 0, 1, ip
, &committed
))) {
1420 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1426 * Keep an extra reference to this quota inode. This inode is
1427 * locked exclusively and joined to the transaction already.
1429 ASSERT(XFS_ISLOCKED_INODE_EXCL(*ip
));
1430 VN_HOLD(XFS_ITOV((*ip
)));
1433 * Make the changes in the superblock, and log those too.
1434 * sbfields arg may contain fields other than *QUOTINO;
1435 * VERSIONNUM for example.
1437 s
= XFS_SB_LOCK(mp
);
1438 if (flags
& XFS_QMOPT_SBVERSION
) {
1439 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1440 unsigned oldv
= mp
->m_sb
.sb_versionnum
;
1442 ASSERT(!XFS_SB_VERSION_HASQUOTA(&mp
->m_sb
));
1443 ASSERT((sbfields
& (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1444 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
)) ==
1445 (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1446 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
));
1448 XFS_SB_VERSION_ADDQUOTA(&mp
->m_sb
);
1449 mp
->m_sb
.sb_uquotino
= NULLFSINO
;
1450 mp
->m_sb
.sb_gquotino
= NULLFSINO
;
1452 /* qflags will get updated _after_ quotacheck */
1453 mp
->m_sb
.sb_qflags
= 0;
1454 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1456 "Old superblock version %x, converting to %x.",
1457 oldv
, mp
->m_sb
.sb_versionnum
);
1460 if (flags
& XFS_QMOPT_UQUOTA
)
1461 mp
->m_sb
.sb_uquotino
= (*ip
)->i_ino
;
1463 mp
->m_sb
.sb_gquotino
= (*ip
)->i_ino
;
1464 XFS_SB_UNLOCK(mp
, s
);
1465 xfs_mod_sb(tp
, sbfields
);
1467 if ((error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
))) {
1468 xfs_fs_cmn_err(CE_ALERT
, mp
, "XFS qino_alloc failed!");
1476 xfs_qm_reset_dqcounts(
1482 xfs_disk_dquot_t
*ddq
;
1485 xfs_buftrace("RESET DQUOTS", bp
);
1487 * Reset all counters and timers. They'll be
1488 * started afresh by xfs_qm_quotacheck.
1491 j
= XFS_FSB_TO_B(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
1492 do_div(j
, sizeof(xfs_dqblk_t
));
1493 ASSERT(XFS_QM_DQPERBLK(mp
) == j
);
1495 ddq
= (xfs_disk_dquot_t
*)XFS_BUF_PTR(bp
);
1496 for (j
= 0; j
< XFS_QM_DQPERBLK(mp
); j
++) {
1498 * Do a sanity check, and if needed, repair the dqblk. Don't
1499 * output any warnings because it's perfectly possible to
1500 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1502 (void) xfs_qm_dqcheck(ddq
, id
+j
, type
, XFS_QMOPT_DQREPAIR
,
1506 ddq
->d_rtbcount
= 0;
1509 ddq
->d_rtbtimer
= 0;
1512 ddq
->d_rtbwarns
= 0;
1513 ddq
= (xfs_disk_dquot_t
*) ((xfs_dqblk_t
*)ddq
+ 1);
1524 xfs_filblks_t blkcnt
,
1535 incr
= (blkcnt
> XFS_QM_MAX_DQCLUSTER_LOGSZ
) ?
1536 XFS_QM_MAX_DQCLUSTER_LOGSZ
: blkcnt
;
1537 type
= flags
& XFS_QMOPT_UQUOTA
? XFS_DQ_USER
:
1538 (flags
& XFS_QMOPT_PQUOTA
? XFS_DQ_PROJ
: XFS_DQ_GROUP
);
1542 * Blkcnt arg can be a very big number, and might even be
1543 * larger than the log itself. So, we have to break it up into
1544 * manageable-sized transactions.
1545 * Note that we don't start a permanent transaction here; we might
1546 * not be able to get a log reservation for the whole thing up front,
1547 * and we don't really care to either, because we just discard
1548 * everything if we were to crash in the middle of this loop.
1551 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
,
1552 XFS_FSB_TO_DADDR(mp
, bno
),
1553 (int)XFS_QI_DQCHUNKLEN(mp
), 0, &bp
);
1557 (void) xfs_qm_reset_dqcounts(mp
, bp
, firstid
, type
);
1558 xfs_bdwrite(mp
, bp
);
1560 * goto the next block.
1563 firstid
+= XFS_QM_DQPERBLK(mp
);
1569 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1570 * caller supplied function for every chunk of dquots that we find.
1578 xfs_bmbt_irec_t
*map
;
1579 int i
, nmaps
; /* number of map entries */
1580 int error
; /* return value */
1581 xfs_fileoff_t lblkno
;
1582 xfs_filblks_t maxlblkcnt
;
1584 xfs_fsblock_t rablkno
;
1585 xfs_filblks_t rablkcnt
;
1589 * This looks racy, but we can't keep an inode lock across a
1590 * trans_reserve. But, this gets called during quotacheck, and that
1591 * happens only at mount time which is single threaded.
1593 if (qip
->i_d
.di_nblocks
== 0)
1596 map
= kmem_alloc(XFS_DQITER_MAP_SIZE
* sizeof(*map
), KM_SLEEP
);
1599 maxlblkcnt
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1601 nmaps
= XFS_DQITER_MAP_SIZE
;
1603 * We aren't changing the inode itself. Just changing
1604 * some of its data. No new blocks are added here, and
1605 * the inode is never added to the transaction.
1607 xfs_ilock(qip
, XFS_ILOCK_SHARED
);
1608 error
= xfs_bmapi(NULL
, qip
, lblkno
,
1609 maxlblkcnt
- lblkno
,
1612 0, map
, &nmaps
, NULL
, NULL
);
1613 xfs_iunlock(qip
, XFS_ILOCK_SHARED
);
1617 ASSERT(nmaps
<= XFS_DQITER_MAP_SIZE
);
1618 for (i
= 0; i
< nmaps
; i
++) {
1619 ASSERT(map
[i
].br_startblock
!= DELAYSTARTBLOCK
);
1620 ASSERT(map
[i
].br_blockcount
);
1623 lblkno
+= map
[i
].br_blockcount
;
1625 if (map
[i
].br_startblock
== HOLESTARTBLOCK
)
1628 firstid
= (xfs_dqid_t
) map
[i
].br_startoff
*
1629 XFS_QM_DQPERBLK(mp
);
1631 * Do a read-ahead on the next extent.
1633 if ((i
+1 < nmaps
) &&
1634 (map
[i
+1].br_startblock
!= HOLESTARTBLOCK
)) {
1635 rablkcnt
= map
[i
+1].br_blockcount
;
1636 rablkno
= map
[i
+1].br_startblock
;
1637 while (rablkcnt
--) {
1638 xfs_baread(mp
->m_ddev_targp
,
1639 XFS_FSB_TO_DADDR(mp
, rablkno
),
1640 (int)XFS_QI_DQCHUNKLEN(mp
));
1645 * Iterate thru all the blks in the extent and
1646 * reset the counters of all the dquots inside them.
1648 if ((error
= xfs_qm_dqiter_bufs(mp
,
1650 map
[i
].br_startblock
,
1651 map
[i
].br_blockcount
,
1659 } while (nmaps
> 0);
1661 kmem_free(map
, XFS_DQITER_MAP_SIZE
* sizeof(*map
));
1667 * Called by dqusage_adjust in doing a quotacheck.
1668 * Given the inode, and a dquot (either USR or GRP, doesn't matter),
1669 * this updates its incore copy as well as the buffer copy. This is
1670 * so that once the quotacheck is done, we can just log all the buffers,
1671 * as opposed to logging numerous updates to individual dquots.
1674 xfs_qm_quotacheck_dqadjust(
1679 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1680 xfs_dqtrace_entry(dqp
, "QCHECK DQADJUST");
1682 * Adjust the inode count and the block count to reflect this inode's
1685 be64_add(&dqp
->q_core
.d_icount
, 1);
1686 dqp
->q_res_icount
++;
1688 be64_add(&dqp
->q_core
.d_bcount
, nblks
);
1689 dqp
->q_res_bcount
+= nblks
;
1692 be64_add(&dqp
->q_core
.d_rtbcount
, rtblks
);
1693 dqp
->q_res_rtbcount
+= rtblks
;
1697 * Set default limits, adjust timers (since we changed usages)
1699 if (! XFS_IS_SUSER_DQUOT(dqp
)) {
1700 xfs_qm_adjust_dqlimits(dqp
->q_mount
, &dqp
->q_core
);
1701 xfs_qm_adjust_dqtimers(dqp
->q_mount
, &dqp
->q_core
);
1704 dqp
->dq_flags
|= XFS_DQ_DIRTY
;
1710 xfs_qcnt_t
*O_rtblks
)
1712 xfs_filblks_t rtblks
; /* total rt blks */
1713 xfs_extnum_t idx
; /* extent record index */
1714 xfs_ifork_t
*ifp
; /* inode fork pointer */
1715 xfs_extnum_t nextents
; /* number of extent entries */
1716 xfs_bmbt_rec_t
*ep
; /* pointer to an extent entry */
1719 ASSERT(XFS_IS_REALTIME_INODE(ip
));
1720 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1721 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1722 if ((error
= xfs_iread_extents(NULL
, ip
, XFS_DATA_FORK
)))
1726 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
1727 for (idx
= 0; idx
< nextents
; idx
++) {
1728 ep
= xfs_iext_get_ext(ifp
, idx
);
1729 rtblks
+= xfs_bmbt_get_blockcount(ep
);
1731 *O_rtblks
= (xfs_qcnt_t
)rtblks
;
1736 * callback routine supplied to bulkstat(). Given an inumber, find its
1737 * dquots and update them to account for resources taken by that inode.
1741 xfs_qm_dqusage_adjust(
1742 xfs_mount_t
*mp
, /* mount point for filesystem */
1743 xfs_ino_t ino
, /* inode number to get data for */
1744 void __user
*buffer
, /* not used */
1745 int ubsize
, /* not used */
1746 void *private_data
, /* not used */
1747 xfs_daddr_t bno
, /* starting block of inode cluster */
1748 int *ubused
, /* not used */
1749 void *dip
, /* on-disk inode pointer (not used) */
1750 int *res
) /* result code value */
1753 xfs_dquot_t
*udqp
, *gdqp
;
1754 xfs_qcnt_t nblks
, rtblks
;
1757 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1760 * rootino must have its resources accounted for, not so with the quota
1763 if (ino
== mp
->m_sb
.sb_uquotino
|| ino
== mp
->m_sb
.sb_gquotino
) {
1764 *res
= BULKSTAT_RV_NOTHING
;
1765 return XFS_ERROR(EINVAL
);
1769 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1770 * interface expects the inode to be exclusively locked because that's
1771 * the case in all other instances. It's OK that we do this because
1772 * quotacheck is done only at mount time.
1774 if ((error
= xfs_iget(mp
, NULL
, ino
, 0, XFS_ILOCK_EXCL
, &ip
, bno
))) {
1775 *res
= BULKSTAT_RV_NOTHING
;
1779 if (ip
->i_d
.di_mode
== 0) {
1780 xfs_iput_new(ip
, XFS_ILOCK_EXCL
);
1781 *res
= BULKSTAT_RV_NOTHING
;
1782 return XFS_ERROR(ENOENT
);
1786 * Obtain the locked dquots. In case of an error (eg. allocation
1787 * fails for ENOSPC), we return the negative of the error number
1788 * to bulkstat, so that it can get propagated to quotacheck() and
1789 * making us disable quotas for the file system.
1791 if ((error
= xfs_qm_dqget_noattach(ip
, &udqp
, &gdqp
))) {
1792 xfs_iput(ip
, XFS_ILOCK_EXCL
);
1793 *res
= BULKSTAT_RV_GIVEUP
;
1798 if (! XFS_IS_REALTIME_INODE(ip
)) {
1799 nblks
= (xfs_qcnt_t
)ip
->i_d
.di_nblocks
;
1802 * Walk thru the extent list and count the realtime blocks.
1804 if ((error
= xfs_qm_get_rtblks(ip
, &rtblks
))) {
1805 xfs_iput(ip
, XFS_ILOCK_EXCL
);
1810 *res
= BULKSTAT_RV_GIVEUP
;
1813 nblks
= (xfs_qcnt_t
)ip
->i_d
.di_nblocks
- rtblks
;
1815 ASSERT(ip
->i_delayed_blks
== 0);
1818 * We can't release the inode while holding its dquot locks.
1819 * The inode can go into inactive and might try to acquire the dquotlocks.
1820 * So, just unlock here and do a vn_rele at the end.
1822 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1825 * Add the (disk blocks and inode) resources occupied by this
1826 * inode to its dquots. We do this adjustment in the incore dquot,
1827 * and also copy the changes to its buffer.
1828 * We don't care about putting these changes in a transaction
1829 * envelope because if we crash in the middle of a 'quotacheck'
1830 * we have to start from the beginning anyway.
1831 * Once we're done, we'll log all the dquot bufs.
1833 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1834 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1836 if (XFS_IS_UQUOTA_ON(mp
)) {
1838 xfs_qm_quotacheck_dqadjust(udqp
, nblks
, rtblks
);
1841 if (XFS_IS_OQUOTA_ON(mp
)) {
1843 xfs_qm_quotacheck_dqadjust(gdqp
, nblks
, rtblks
);
1847 * Now release the inode. This will send it to 'inactive', and
1848 * possibly even free blocks.
1850 VN_RELE(XFS_ITOV(ip
));
1855 *res
= BULKSTAT_RV_DIDONE
;
1860 * Walk thru all the filesystem inodes and construct a consistent view
1861 * of the disk quota world. If the quotacheck fails, disable quotas.
1867 int done
, count
, error
;
1870 xfs_inode_t
*uip
, *gip
;
1878 ASSERT(XFS_QI_UQIP(mp
) || XFS_QI_GQIP(mp
));
1879 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1882 * There should be no cached dquots. The (simplistic) quotacheck
1883 * algorithm doesn't like that.
1885 ASSERT(XFS_QI_MPLNDQUOTS(mp
) == 0);
1887 cmn_err(CE_NOTE
, "XFS quotacheck %s: Please wait.", mp
->m_fsname
);
1890 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1891 * their counters to zero. We need a clean slate.
1892 * We don't log our changes till later.
1894 if ((uip
= XFS_QI_UQIP(mp
))) {
1895 if ((error
= xfs_qm_dqiterate(mp
, uip
, XFS_QMOPT_UQUOTA
)))
1897 flags
|= XFS_UQUOTA_CHKD
;
1900 if ((gip
= XFS_QI_GQIP(mp
))) {
1901 if ((error
= xfs_qm_dqiterate(mp
, gip
, XFS_IS_GQUOTA_ON(mp
) ?
1902 XFS_QMOPT_GQUOTA
: XFS_QMOPT_PQUOTA
)))
1904 flags
|= XFS_OQUOTA_CHKD
;
1909 * Iterate thru all the inodes in the file system,
1910 * adjusting the corresponding dquot counters in core.
1912 if ((error
= xfs_bulkstat(mp
, &lastino
, &count
,
1913 xfs_qm_dqusage_adjust
, NULL
,
1914 structsz
, NULL
, BULKSTAT_FG_IGET
, &done
)))
1920 * We can get this error if we couldn't do a dquot allocation inside
1921 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1922 * dirty dquots that might be cached, we just want to get rid of them
1923 * and turn quotaoff. The dquots won't be attached to any of the inodes
1924 * at this point (because we intentionally didn't in dqget_noattach).
1927 xfs_qm_dqpurge_all(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_QUOTAOFF
);
1931 * We've made all the changes that we need to make incore.
1932 * Now flush_them down to disk buffers.
1934 xfs_qm_dqflush_all(mp
, XFS_QMOPT_DELWRI
);
1937 * We didn't log anything, because if we crashed, we'll have to
1938 * start the quotacheck from scratch anyway. However, we must make
1939 * sure that our dquot changes are secure before we put the
1940 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1943 XFS_bflush(mp
->m_ddev_targp
);
1946 * If one type of quotas is off, then it will lose its
1947 * quotachecked status, since we won't be doing accounting for
1948 * that type anymore.
1950 mp
->m_qflags
&= ~(XFS_OQUOTA_CHKD
| XFS_UQUOTA_CHKD
);
1951 mp
->m_qflags
|= flags
;
1953 XQM_LIST_PRINT(&(XFS_QI_MPL_LIST(mp
)), MPL_NEXT
, "++++ Mp list +++");
1957 cmn_err(CE_WARN
, "XFS quotacheck %s: Unsuccessful (Error %d): "
1958 "Disabling quotas.",
1959 mp
->m_fsname
, error
);
1961 * We must turn off quotas.
1963 ASSERT(mp
->m_quotainfo
!= NULL
);
1964 ASSERT(xfs_Gqm
!= NULL
);
1965 xfs_qm_destroy_quotainfo(mp
);
1966 (void)xfs_mount_reset_sbqflags(mp
);
1968 cmn_err(CE_NOTE
, "XFS quotacheck %s: Done.", mp
->m_fsname
);
1974 * This is called after the superblock has been read in and we're ready to
1975 * iget the quota inodes.
1978 xfs_qm_init_quotainos(
1981 xfs_inode_t
*uip
, *gip
;
1986 ASSERT(mp
->m_quotainfo
);
1992 * Get the uquota and gquota inodes
1994 if (XFS_SB_VERSION_HASQUOTA(&mp
->m_sb
)) {
1995 if (XFS_IS_UQUOTA_ON(mp
) &&
1996 mp
->m_sb
.sb_uquotino
!= NULLFSINO
) {
1997 ASSERT(mp
->m_sb
.sb_uquotino
> 0);
1998 if ((error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_uquotino
,
2000 return XFS_ERROR(error
);
2002 if (XFS_IS_OQUOTA_ON(mp
) &&
2003 mp
->m_sb
.sb_gquotino
!= NULLFSINO
) {
2004 ASSERT(mp
->m_sb
.sb_gquotino
> 0);
2005 if ((error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_gquotino
,
2008 VN_RELE(XFS_ITOV(uip
));
2009 return XFS_ERROR(error
);
2013 flags
|= XFS_QMOPT_SBVERSION
;
2014 sbflags
|= (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
2015 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
);
2019 * Create the two inodes, if they don't exist already. The changes
2020 * made above will get added to a transaction and logged in one of
2021 * the qino_alloc calls below. If the device is readonly,
2022 * temporarily switch to read-write to do this.
2024 if (XFS_IS_UQUOTA_ON(mp
) && uip
== NULL
) {
2025 if ((error
= xfs_qm_qino_alloc(mp
, &uip
,
2026 sbflags
| XFS_SB_UQUOTINO
,
2027 flags
| XFS_QMOPT_UQUOTA
)))
2028 return XFS_ERROR(error
);
2030 flags
&= ~XFS_QMOPT_SBVERSION
;
2032 if (XFS_IS_OQUOTA_ON(mp
) && gip
== NULL
) {
2033 flags
|= (XFS_IS_GQUOTA_ON(mp
) ?
2034 XFS_QMOPT_GQUOTA
: XFS_QMOPT_PQUOTA
);
2035 error
= xfs_qm_qino_alloc(mp
, &gip
,
2036 sbflags
| XFS_SB_GQUOTINO
, flags
);
2039 VN_RELE(XFS_ITOV(uip
));
2041 return XFS_ERROR(error
);
2045 XFS_QI_UQIP(mp
) = uip
;
2046 XFS_QI_GQIP(mp
) = gip
;
2053 * Traverse the freelist of dquots and attempt to reclaim a maximum of
2054 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
2055 * favor the lookup function ...
2056 * XXXsup merge this with qm_reclaim_one().
2059 xfs_qm_shake_freelist(
2064 xfs_dquot_t
*dqp
, *nextdqp
;
2076 cmn_err(CE_DEBUG
, "Shake free 0x%x", howmany
);
2078 /* lock order is : hashchainlock, freelistlock, mplistlock */
2080 xfs_qm_freelist_lock(xfs_Gqm
);
2082 for (dqp
= xfs_Gqm
->qm_dqfreelist
.qh_next
;
2083 ((dqp
!= (xfs_dquot_t
*) &xfs_Gqm
->qm_dqfreelist
) &&
2084 nreclaimed
< howmany
); ) {
2088 * We are racing with dqlookup here. Naturally we don't
2089 * want to reclaim a dquot that lookup wants.
2091 if (dqp
->dq_flags
& XFS_DQ_WANT
) {
2093 xfs_qm_freelist_unlock(xfs_Gqm
);
2094 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2096 XQM_STATS_INC(xqmstats
.xs_qm_dqwants
);
2101 * If the dquot is inactive, we are assured that it is
2102 * not on the mplist or the hashlist, and that makes our
2105 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
2106 ASSERT(dqp
->q_mount
== NULL
);
2107 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
2108 ASSERT(dqp
->HL_PREVP
== NULL
);
2109 ASSERT(dqp
->MPL_PREVP
== NULL
);
2110 XQM_STATS_INC(xqmstats
.xs_qm_dqinact_reclaims
);
2111 nextdqp
= dqp
->dq_flnext
;
2115 ASSERT(dqp
->MPL_PREVP
);
2117 * Try to grab the flush lock. If this dquot is in the process of
2118 * getting flushed to disk, we don't want to reclaim it.
2120 if (! xfs_qm_dqflock_nowait(dqp
)) {
2122 dqp
= dqp
->dq_flnext
;
2127 * We have the flush lock so we know that this is not in the
2128 * process of being flushed. So, if this is dirty, flush it
2129 * DELWRI so that we don't get a freelist infested with
2132 if (XFS_DQ_IS_DIRTY(dqp
)) {
2133 xfs_dqtrace_entry(dqp
, "DQSHAKE: DQDIRTY");
2135 * We flush it delayed write, so don't bother
2136 * releasing the mplock.
2138 (void) xfs_qm_dqflush(dqp
, XFS_QMOPT_DELWRI
);
2139 xfs_dqunlock(dqp
); /* dqflush unlocks dqflock */
2140 dqp
= dqp
->dq_flnext
;
2144 * We're trying to get the hashlock out of order. This races
2145 * with dqlookup; so, we giveup and goto the next dquot if
2146 * we couldn't get the hashlock. This way, we won't starve
2147 * a dqlookup process that holds the hashlock that is
2148 * waiting for the freelist lock.
2150 if (! xfs_qm_dqhashlock_nowait(dqp
)) {
2153 dqp
= dqp
->dq_flnext
;
2157 * This races with dquot allocation code as well as dqflush_all
2158 * and reclaim code. So, if we failed to grab the mplist lock,
2159 * giveup everything and start over.
2163 if (! xfs_qm_mplist_nowait(dqp
->q_mount
)) {
2164 /* XXX put a sentinel so that we can come back here */
2167 XFS_DQ_HASH_UNLOCK(hash
);
2168 xfs_qm_freelist_unlock(xfs_Gqm
);
2169 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2173 xfs_dqtrace_entry(dqp
, "DQSHAKE: UNLINKING");
2175 cmn_err(CE_DEBUG
, "Shake 0x%p, ID 0x%x\n",
2176 dqp
, be32_to_cpu(dqp
->q_core
.d_id
));
2178 ASSERT(dqp
->q_nrefs
== 0);
2179 nextdqp
= dqp
->dq_flnext
;
2180 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp
->q_mount
)), dqp
);
2181 XQM_HASHLIST_REMOVE(hash
, dqp
);
2183 xfs_qm_mplist_unlock(dqp
->q_mount
);
2184 XFS_DQ_HASH_UNLOCK(hash
);
2187 XQM_FREELIST_REMOVE(dqp
);
2190 XQM_STATS_INC(xqmstats
.xs_qm_dqshake_reclaims
);
2191 xfs_qm_dqdestroy(dqp
);
2194 xfs_qm_freelist_unlock(xfs_Gqm
);
2200 * The kmem_shake interface is invoked when memory is running low.
2204 xfs_qm_shake(int nr_to_scan
, gfp_t gfp_mask
)
2206 int ndqused
, nfree
, n
;
2208 if (!kmem_shake_allow(gfp_mask
))
2213 nfree
= xfs_Gqm
->qm_dqfreelist
.qh_nelems
; /* free dquots */
2214 /* incore dquots in all f/s's */
2215 ndqused
= atomic_read(&xfs_Gqm
->qm_totaldquots
) - nfree
;
2217 ASSERT(ndqused
>= 0);
2219 if (nfree
<= ndqused
&& nfree
< ndquot
)
2222 ndqused
*= xfs_Gqm
->qm_dqfree_ratio
; /* target # of free dquots */
2223 n
= nfree
- ndqused
- ndquot
; /* # over target */
2225 return xfs_qm_shake_freelist(MAX(nfree
, n
));
2230 * Just pop the least recently used dquot off the freelist and
2231 * recycle it. The returned dquot is locked.
2233 STATIC xfs_dquot_t
*
2234 xfs_qm_dqreclaim_one(void)
2236 xfs_dquot_t
*dqpout
;
2245 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
2247 xfs_qm_freelist_lock(xfs_Gqm
);
2249 FOREACH_DQUOT_IN_FREELIST(dqp
, &(xfs_Gqm
->qm_dqfreelist
)) {
2253 * We are racing with dqlookup here. Naturally we don't
2254 * want to reclaim a dquot that lookup wants. We release the
2255 * freelist lock and start over, so that lookup will grab
2256 * both the dquot and the freelistlock.
2258 if (dqp
->dq_flags
& XFS_DQ_WANT
) {
2259 ASSERT(! (dqp
->dq_flags
& XFS_DQ_INACTIVE
));
2260 xfs_dqtrace_entry(dqp
, "DQRECLAIM: DQWANT");
2262 xfs_qm_freelist_unlock(xfs_Gqm
);
2263 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2265 XQM_STATS_INC(xqmstats
.xs_qm_dqwants
);
2270 * If the dquot is inactive, we are assured that it is
2271 * not on the mplist or the hashlist, and that makes our
2274 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
2275 ASSERT(dqp
->q_mount
== NULL
);
2276 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
2277 ASSERT(dqp
->HL_PREVP
== NULL
);
2278 ASSERT(dqp
->MPL_PREVP
== NULL
);
2279 XQM_FREELIST_REMOVE(dqp
);
2282 XQM_STATS_INC(xqmstats
.xs_qm_dqinact_reclaims
);
2286 ASSERT(dqp
->q_hash
);
2287 ASSERT(dqp
->MPL_PREVP
);
2290 * Try to grab the flush lock. If this dquot is in the process of
2291 * getting flushed to disk, we don't want to reclaim it.
2293 if (! xfs_qm_dqflock_nowait(dqp
)) {
2299 * We have the flush lock so we know that this is not in the
2300 * process of being flushed. So, if this is dirty, flush it
2301 * DELWRI so that we don't get a freelist infested with
2304 if (XFS_DQ_IS_DIRTY(dqp
)) {
2305 xfs_dqtrace_entry(dqp
, "DQRECLAIM: DQDIRTY");
2307 * We flush it delayed write, so don't bother
2308 * releasing the freelist lock.
2310 (void) xfs_qm_dqflush(dqp
, XFS_QMOPT_DELWRI
);
2311 xfs_dqunlock(dqp
); /* dqflush unlocks dqflock */
2315 if (! xfs_qm_mplist_nowait(dqp
->q_mount
)) {
2321 if (! xfs_qm_dqhashlock_nowait(dqp
))
2324 ASSERT(dqp
->q_nrefs
== 0);
2325 xfs_dqtrace_entry(dqp
, "DQRECLAIM: UNLINKING");
2326 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp
->q_mount
)), dqp
);
2327 XQM_HASHLIST_REMOVE(dqp
->q_hash
, dqp
);
2328 XQM_FREELIST_REMOVE(dqp
);
2330 XFS_DQ_HASH_UNLOCK(dqp
->q_hash
);
2332 xfs_qm_mplist_unlock(dqp
->q_mount
);
2339 xfs_qm_freelist_unlock(xfs_Gqm
);
2344 /*------------------------------------------------------------------*/
2347 * Return a new incore dquot. Depending on the number of
2348 * dquots in the system, we either allocate a new one on the kernel heap,
2349 * or reclaim a free one.
2350 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2351 * to reclaim an existing one from the freelist.
2354 xfs_qm_dqalloc_incore(
2355 xfs_dquot_t
**O_dqpp
)
2360 * Check against high water mark to see if we want to pop
2361 * a nincompoop dquot off the freelist.
2363 if (atomic_read(&xfs_Gqm
->qm_totaldquots
) >= ndquot
) {
2365 * Try to recycle a dquot from the freelist.
2367 if ((dqp
= xfs_qm_dqreclaim_one())) {
2368 XQM_STATS_INC(xqmstats
.xs_qm_dqreclaims
);
2370 * Just zero the core here. The rest will get
2371 * reinitialized by caller. XXX we shouldn't even
2374 memset(&dqp
->q_core
, 0, sizeof(dqp
->q_core
));
2378 XQM_STATS_INC(xqmstats
.xs_qm_dqreclaim_misses
);
2382 * Allocate a brand new dquot on the kernel heap and return it
2383 * to the caller to initialize.
2385 ASSERT(xfs_Gqm
->qm_dqzone
!= NULL
);
2386 *O_dqpp
= kmem_zone_zalloc(xfs_Gqm
->qm_dqzone
, KM_SLEEP
);
2387 atomic_inc(&xfs_Gqm
->qm_totaldquots
);
2394 * Start a transaction and write the incore superblock changes to
2395 * disk. flags parameter indicates which fields have changed.
2398 xfs_qm_write_sb_changes(
2406 cmn_err(CE_NOTE
, "Writing superblock quota changes :%s", mp
->m_fsname
);
2408 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
2409 if ((error
= xfs_trans_reserve(tp
, 0,
2410 mp
->m_sb
.sb_sectsize
+ 128, 0,
2412 XFS_DEFAULT_LOG_COUNT
))) {
2413 xfs_trans_cancel(tp
, 0);
2417 xfs_mod_sb(tp
, flags
);
2418 (void) xfs_trans_commit(tp
, 0);
2424 /* --------------- utility functions for vnodeops ---------------- */
2428 * Given an inode, a uid and gid (from cred_t) make sure that we have
2429 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2430 * quotas by creating this file.
2431 * This also attaches dquot(s) to the given inode after locking it,
2432 * and returns the dquots corresponding to the uid and/or gid.
2434 * in : inode (unlocked)
2435 * out : udquot, gdquot with references taken and unlocked
2445 xfs_dquot_t
**O_udqpp
,
2446 xfs_dquot_t
**O_gdqpp
)
2449 xfs_dquot_t
*uq
, *gq
;
2452 if (!XFS_IS_QUOTA_ON(mp
))
2455 lockflags
= XFS_ILOCK_EXCL
;
2456 xfs_ilock(ip
, lockflags
);
2458 if ((flags
& XFS_QMOPT_INHERIT
) &&
2459 XFS_INHERIT_GID(ip
, XFS_MTOVFS(mp
)))
2460 gid
= ip
->i_d
.di_gid
;
2463 * Attach the dquot(s) to this inode, doing a dquot allocation
2464 * if necessary. The dquot(s) will not be locked.
2466 if (XFS_NOT_DQATTACHED(mp
, ip
)) {
2467 if ((error
= xfs_qm_dqattach(ip
, XFS_QMOPT_DQALLOC
|
2468 XFS_QMOPT_ILOCKED
))) {
2469 xfs_iunlock(ip
, lockflags
);
2475 if ((flags
& XFS_QMOPT_UQUOTA
) && XFS_IS_UQUOTA_ON(mp
)) {
2476 if (ip
->i_d
.di_uid
!= uid
) {
2478 * What we need is the dquot that has this uid, and
2479 * if we send the inode to dqget, the uid of the inode
2480 * takes priority over what's sent in the uid argument.
2481 * We must unlock inode here before calling dqget if
2482 * we're not sending the inode, because otherwise
2483 * we'll deadlock by doing trans_reserve while
2486 xfs_iunlock(ip
, lockflags
);
2487 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
) uid
,
2492 ASSERT(error
!= ENOENT
);
2496 * Get the ilock in the right order.
2499 lockflags
= XFS_ILOCK_SHARED
;
2500 xfs_ilock(ip
, lockflags
);
2503 * Take an extra reference, because we'll return
2506 ASSERT(ip
->i_udquot
);
2513 if ((flags
& XFS_QMOPT_GQUOTA
) && XFS_IS_GQUOTA_ON(mp
)) {
2514 if (ip
->i_d
.di_gid
!= gid
) {
2515 xfs_iunlock(ip
, lockflags
);
2516 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)gid
,
2523 ASSERT(error
!= ENOENT
);
2527 lockflags
= XFS_ILOCK_SHARED
;
2528 xfs_ilock(ip
, lockflags
);
2530 ASSERT(ip
->i_gdquot
);
2536 } else if ((flags
& XFS_QMOPT_PQUOTA
) && XFS_IS_PQUOTA_ON(mp
)) {
2537 if (ip
->i_d
.di_projid
!= prid
) {
2538 xfs_iunlock(ip
, lockflags
);
2539 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)prid
,
2546 ASSERT(error
!= ENOENT
);
2550 lockflags
= XFS_ILOCK_SHARED
;
2551 xfs_ilock(ip
, lockflags
);
2553 ASSERT(ip
->i_gdquot
);
2561 xfs_dqtrace_entry_ino(uq
, "DQALLOC", ip
);
2563 xfs_iunlock(ip
, lockflags
);
2576 * Actually transfer ownership, and do dquot modifications.
2577 * These were already reserved.
2583 xfs_dquot_t
**IO_olddq
,
2586 xfs_dquot_t
*prevdq
;
2587 uint bfield
= XFS_IS_REALTIME_INODE(ip
) ?
2588 XFS_TRANS_DQ_RTBCOUNT
: XFS_TRANS_DQ_BCOUNT
;
2590 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
2591 ASSERT(XFS_IS_QUOTA_RUNNING(ip
->i_mount
));
2596 ASSERT(prevdq
!= newdq
);
2598 xfs_trans_mod_dquot(tp
, prevdq
, bfield
, -(ip
->i_d
.di_nblocks
));
2599 xfs_trans_mod_dquot(tp
, prevdq
, XFS_TRANS_DQ_ICOUNT
, -1);
2601 /* the sparkling new dquot */
2602 xfs_trans_mod_dquot(tp
, newdq
, bfield
, ip
->i_d
.di_nblocks
);
2603 xfs_trans_mod_dquot(tp
, newdq
, XFS_TRANS_DQ_ICOUNT
, 1);
2606 * Take an extra reference, because the inode
2607 * is going to keep this dquot pointer even
2608 * after the trans_commit.
2612 xfs_dqunlock(newdq
);
2619 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
2622 xfs_qm_vop_chown_reserve(
2631 uint delblks
, blkflags
, prjflags
= 0;
2632 xfs_dquot_t
*unresudq
, *unresgdq
, *delblksudq
, *delblksgdq
;
2634 ASSERT(XFS_ISLOCKED_INODE(ip
));
2636 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
2638 delblks
= ip
->i_delayed_blks
;
2639 delblksudq
= delblksgdq
= unresudq
= unresgdq
= NULL
;
2640 blkflags
= XFS_IS_REALTIME_INODE(ip
) ?
2641 XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
;
2643 if (XFS_IS_UQUOTA_ON(mp
) && udqp
&&
2644 ip
->i_d
.di_uid
!= (uid_t
)be32_to_cpu(udqp
->q_core
.d_id
)) {
2647 * If there are delayed allocation blocks, then we have to
2648 * unreserve those from the old dquot, and add them to the
2652 ASSERT(ip
->i_udquot
);
2653 unresudq
= ip
->i_udquot
;
2656 if (XFS_IS_OQUOTA_ON(ip
->i_mount
) && gdqp
) {
2657 if (XFS_IS_PQUOTA_ON(ip
->i_mount
) &&
2658 ip
->i_d
.di_projid
!= be32_to_cpu(gdqp
->q_core
.d_id
))
2659 prjflags
= XFS_QMOPT_ENOSPC
;
2662 (XFS_IS_GQUOTA_ON(ip
->i_mount
) &&
2663 ip
->i_d
.di_gid
!= be32_to_cpu(gdqp
->q_core
.d_id
))) {
2666 ASSERT(ip
->i_gdquot
);
2667 unresgdq
= ip
->i_gdquot
;
2672 if ((error
= xfs_trans_reserve_quota_bydquots(tp
, ip
->i_mount
,
2673 delblksudq
, delblksgdq
, ip
->i_d
.di_nblocks
, 1,
2674 flags
| blkflags
| prjflags
)))
2678 * Do the delayed blks reservations/unreservations now. Since, these
2679 * are done without the help of a transaction, if a reservation fails
2680 * its previous reservations won't be automatically undone by trans
2681 * code. So, we have to do it manually here.
2685 * Do the reservations first. Unreservation can't fail.
2687 ASSERT(delblksudq
|| delblksgdq
);
2688 ASSERT(unresudq
|| unresgdq
);
2689 if ((error
= xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
2690 delblksudq
, delblksgdq
, (xfs_qcnt_t
)delblks
, 0,
2691 flags
| blkflags
| prjflags
)))
2693 xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
2694 unresudq
, unresgdq
, -((xfs_qcnt_t
)delblks
), 0,
2702 xfs_qm_vop_rename_dqattach(
2703 xfs_inode_t
**i_tab
)
2711 if (! XFS_IS_QUOTA_ON(ip
->i_mount
))
2714 if (XFS_NOT_DQATTACHED(ip
->i_mount
, ip
)) {
2715 error
= xfs_qm_dqattach(ip
, 0);
2719 for (i
= 1; (i
< 4 && i_tab
[i
]); i
++) {
2721 * Watch out for duplicate entries in the table.
2723 if ((ip
= i_tab
[i
]) != i_tab
[i
-1]) {
2724 if (XFS_NOT_DQATTACHED(ip
->i_mount
, ip
)) {
2725 error
= xfs_qm_dqattach(ip
, 0);
2735 xfs_qm_vop_dqattach_and_dqmod_newinode(
2741 if (!XFS_IS_QUOTA_ON(tp
->t_mountp
))
2744 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
2745 ASSERT(XFS_IS_QUOTA_RUNNING(tp
->t_mountp
));
2751 ASSERT(ip
->i_udquot
== NULL
);
2752 ip
->i_udquot
= udqp
;
2753 ASSERT(XFS_IS_UQUOTA_ON(tp
->t_mountp
));
2754 ASSERT(ip
->i_d
.di_uid
== be32_to_cpu(udqp
->q_core
.d_id
));
2755 xfs_trans_mod_dquot(tp
, udqp
, XFS_TRANS_DQ_ICOUNT
, 1);
2761 ASSERT(ip
->i_gdquot
== NULL
);
2762 ip
->i_gdquot
= gdqp
;
2763 ASSERT(XFS_IS_OQUOTA_ON(tp
->t_mountp
));
2764 ASSERT((XFS_IS_GQUOTA_ON(tp
->t_mountp
) ?
2765 ip
->i_d
.di_gid
: ip
->i_d
.di_projid
) ==
2766 be32_to_cpu(gdqp
->q_core
.d_id
));
2767 xfs_trans_mod_dquot(tp
, gdqp
, XFS_TRANS_DQ_ICOUNT
, 1);
2771 /* ------------- list stuff -----------------*/
2773 xfs_qm_freelist_init(xfs_frlist_t
*ql
)
2775 ql
->qh_next
= ql
->qh_prev
= (xfs_dquot_t
*) ql
;
2776 mutex_init(&ql
->qh_lock
);
2782 xfs_qm_freelist_destroy(xfs_frlist_t
*ql
)
2784 xfs_dquot_t
*dqp
, *nextdqp
;
2786 mutex_lock(&ql
->qh_lock
);
2787 for (dqp
= ql
->qh_next
;
2788 dqp
!= (xfs_dquot_t
*)ql
; ) {
2790 nextdqp
= dqp
->dq_flnext
;
2792 cmn_err(CE_DEBUG
, "FREELIST destroy 0x%p", dqp
);
2794 XQM_FREELIST_REMOVE(dqp
);
2796 xfs_qm_dqdestroy(dqp
);
2799 mutex_unlock(&ql
->qh_lock
);
2800 mutex_destroy(&ql
->qh_lock
);
2802 ASSERT(ql
->qh_nelems
== 0);
2806 xfs_qm_freelist_insert(xfs_frlist_t
*ql
, xfs_dquot_t
*dq
)
2808 dq
->dq_flnext
= ql
->qh_next
;
2809 dq
->dq_flprev
= (xfs_dquot_t
*)ql
;
2811 dq
->dq_flnext
->dq_flprev
= dq
;
2812 xfs_Gqm
->qm_dqfreelist
.qh_nelems
++;
2813 xfs_Gqm
->qm_dqfreelist
.qh_version
++;
2817 xfs_qm_freelist_unlink(xfs_dquot_t
*dq
)
2819 xfs_dquot_t
*next
= dq
->dq_flnext
;
2820 xfs_dquot_t
*prev
= dq
->dq_flprev
;
2822 next
->dq_flprev
= prev
;
2823 prev
->dq_flnext
= next
;
2824 dq
->dq_flnext
= dq
->dq_flprev
= dq
;
2825 xfs_Gqm
->qm_dqfreelist
.qh_nelems
--;
2826 xfs_Gqm
->qm_dqfreelist
.qh_version
++;
2830 xfs_qm_freelist_append(xfs_frlist_t
*ql
, xfs_dquot_t
*dq
)
2832 xfs_qm_freelist_insert((xfs_frlist_t
*)ql
->qh_prev
, dq
);
2836 xfs_qm_dqhashlock_nowait(
2841 locked
= mutex_trylock(&((dqp
)->q_hash
->qh_lock
));
2846 xfs_qm_freelist_lock_nowait(
2851 locked
= mutex_trylock(&(xqm
->qm_dqfreelist
.qh_lock
));
2856 xfs_qm_mplist_nowait(
2861 ASSERT(mp
->m_quotainfo
);
2862 locked
= mutex_trylock(&(XFS_QI_MPLLOCK(mp
)));