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"
29 #include "xfs_alloc.h"
30 #include "xfs_dmapi.h"
31 #include "xfs_quota.h"
32 #include "xfs_mount.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir_sf.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_btree.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_itable.h"
44 #include "xfs_rtalloc.h"
45 #include "xfs_error.h"
52 #include "xfs_buf_item.h"
53 #include "xfs_trans_space.h"
54 #include "xfs_utils.h"
58 * The global quota manager. There is only one of these for the entire
59 * system, _not_ one per file system. XQM keeps track of the overall
60 * quota functionality, including maintaining the freelist and hash
64 struct xfs_qm
*xfs_Gqm
;
67 kmem_zone_t
*qm_dqzone
;
68 kmem_zone_t
*qm_dqtrxzone
;
69 STATIC kmem_shaker_t xfs_qm_shaker
;
71 STATIC cred_t xfs_zerocr
;
72 STATIC xfs_inode_t xfs_zeroino
;
74 STATIC
void xfs_qm_list_init(xfs_dqlist_t
*, char *, int);
75 STATIC
void xfs_qm_list_destroy(xfs_dqlist_t
*);
77 STATIC
void xfs_qm_freelist_init(xfs_frlist_t
*);
78 STATIC
void xfs_qm_freelist_destroy(xfs_frlist_t
*);
79 STATIC
int xfs_qm_mplist_nowait(xfs_mount_t
*);
80 STATIC
int xfs_qm_dqhashlock_nowait(xfs_dquot_t
*);
82 STATIC
int xfs_qm_init_quotainos(xfs_mount_t
*);
83 STATIC
int xfs_qm_init_quotainfo(xfs_mount_t
*);
84 STATIC
int xfs_qm_shake(int, gfp_t
);
87 extern mutex_t qcheck_lock
;
91 #define XQM_LIST_PRINT(l, NXT, title) \
93 xfs_dquot_t *dqp; int i = 0; \
94 cmn_err(CE_DEBUG, "%s (#%d)", title, (int) (l)->qh_nelems); \
95 for (dqp = (l)->qh_next; dqp != NULL; dqp = dqp->NXT) { \
96 cmn_err(CE_DEBUG, " %d. \"%d (%s)\" " \
97 "bcnt = %d, icnt = %d, refs = %d", \
98 ++i, (int) be32_to_cpu(dqp->q_core.d_id), \
99 DQFLAGTO_TYPESTR(dqp), \
100 (int) be64_to_cpu(dqp->q_core.d_bcount), \
101 (int) be64_to_cpu(dqp->q_core.d_icount), \
102 (int) dqp->q_nrefs); } \
105 #define XQM_LIST_PRINT(l, NXT, title) do { } while (0)
109 * Initialize the XQM structure.
110 * Note that there is not one quota manager per file system.
112 STATIC
struct xfs_qm
*
115 xfs_dqhash_t
*udqhash
, *gdqhash
;
117 uint i
, hsize
, flags
= KM_SLEEP
| KM_MAYFAIL
;
120 * Initialize the dquot hash tables.
122 hsize
= XFS_QM_HASHSIZE_HIGH
;
123 while (!(udqhash
= kmem_zalloc(hsize
* sizeof(xfs_dqhash_t
), flags
))) {
124 if ((hsize
>>= 1) <= XFS_QM_HASHSIZE_LOW
)
127 gdqhash
= kmem_zalloc(hsize
* sizeof(xfs_dqhash_t
), KM_SLEEP
);
130 xqm
= kmem_zalloc(sizeof(xfs_qm_t
), KM_SLEEP
);
131 xqm
->qm_dqhashmask
= hsize
- 1;
132 xqm
->qm_usr_dqhtable
= udqhash
;
133 xqm
->qm_grp_dqhtable
= gdqhash
;
134 ASSERT(xqm
->qm_usr_dqhtable
!= NULL
);
135 ASSERT(xqm
->qm_grp_dqhtable
!= NULL
);
137 for (i
= 0; i
< hsize
; i
++) {
138 xfs_qm_list_init(&(xqm
->qm_usr_dqhtable
[i
]), "uxdqh", i
);
139 xfs_qm_list_init(&(xqm
->qm_grp_dqhtable
[i
]), "gxdqh", i
);
143 * Freelist of all dquots of all file systems
145 xfs_qm_freelist_init(&(xqm
->qm_dqfreelist
));
148 * dquot zone. we register our own low-memory callback.
151 xqm
->qm_dqzone
= kmem_zone_init(sizeof(xfs_dquot_t
),
153 qm_dqzone
= xqm
->qm_dqzone
;
155 xqm
->qm_dqzone
= qm_dqzone
;
157 xfs_qm_shaker
= kmem_shake_register(xfs_qm_shake
);
160 * The t_dqinfo portion of transactions.
163 xqm
->qm_dqtrxzone
= kmem_zone_init(sizeof(xfs_dquot_acct_t
),
165 qm_dqtrxzone
= xqm
->qm_dqtrxzone
;
167 xqm
->qm_dqtrxzone
= qm_dqtrxzone
;
169 atomic_set(&xqm
->qm_totaldquots
, 0);
170 xqm
->qm_dqfree_ratio
= XFS_QM_DQFREE_RATIO
;
173 mutex_init(&qcheck_lock
);
179 * Destroy the global quota manager when its reference count goes to zero.
188 ASSERT(xqm
->qm_nrefs
== 0);
189 kmem_shake_deregister(xfs_qm_shaker
);
190 hsize
= xqm
->qm_dqhashmask
+ 1;
191 for (i
= 0; i
< hsize
; i
++) {
192 xfs_qm_list_destroy(&(xqm
->qm_usr_dqhtable
[i
]));
193 xfs_qm_list_destroy(&(xqm
->qm_grp_dqhtable
[i
]));
195 kmem_free(xqm
->qm_usr_dqhtable
, hsize
* sizeof(xfs_dqhash_t
));
196 kmem_free(xqm
->qm_grp_dqhtable
, hsize
* sizeof(xfs_dqhash_t
));
197 xqm
->qm_usr_dqhtable
= NULL
;
198 xqm
->qm_grp_dqhtable
= NULL
;
199 xqm
->qm_dqhashmask
= 0;
200 xfs_qm_freelist_destroy(&(xqm
->qm_dqfreelist
));
202 mutex_destroy(&qcheck_lock
);
204 kmem_free(xqm
, sizeof(xfs_qm_t
));
208 * Called at mount time to let XQM know that another file system is
209 * starting quotas. This isn't crucial information as the individual mount
210 * structures are pretty independent, but it helps the XQM keep a
211 * global view of what's going on.
215 xfs_qm_hold_quotafs_ref(
216 struct xfs_mount
*mp
)
219 * Need to lock the xfs_Gqm structure for things like this. For example,
220 * the structure could disappear between the entry to this routine and
221 * a HOLD operation if not locked.
223 XFS_QM_LOCK(xfs_Gqm
);
226 xfs_Gqm
= xfs_Gqm_init();
228 * We can keep a list of all filesystems with quotas mounted for
229 * debugging and statistical purposes, but ...
230 * Just take a reference and get out.
232 XFS_QM_HOLD(xfs_Gqm
);
233 XFS_QM_UNLOCK(xfs_Gqm
);
240 * Release the reference that a filesystem took at mount time,
241 * so that we know when we need to destroy the entire quota manager.
245 xfs_qm_rele_quotafs_ref(
246 struct xfs_mount
*mp
)
248 xfs_dquot_t
*dqp
, *nextdqp
;
251 ASSERT(xfs_Gqm
->qm_nrefs
> 0);
254 * Go thru the freelist and destroy all inactive dquots.
256 xfs_qm_freelist_lock(xfs_Gqm
);
258 for (dqp
= xfs_Gqm
->qm_dqfreelist
.qh_next
;
259 dqp
!= (xfs_dquot_t
*)&(xfs_Gqm
->qm_dqfreelist
); ) {
261 nextdqp
= dqp
->dq_flnext
;
262 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
263 ASSERT(dqp
->q_mount
== NULL
);
264 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
265 ASSERT(dqp
->HL_PREVP
== NULL
);
266 ASSERT(dqp
->MPL_PREVP
== NULL
);
267 XQM_FREELIST_REMOVE(dqp
);
269 xfs_qm_dqdestroy(dqp
);
275 xfs_qm_freelist_unlock(xfs_Gqm
);
278 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
281 XFS_QM_LOCK(xfs_Gqm
);
282 XFS_QM_RELE(xfs_Gqm
);
283 if (xfs_Gqm
->qm_nrefs
== 0) {
284 xfs_qm_destroy(xfs_Gqm
);
287 XFS_QM_UNLOCK(xfs_Gqm
);
291 * This is called at mount time from xfs_mountfs to initialize the quotainfo
292 * structure and start the global quota manager (xfs_Gqm) if it hasn't done
293 * so already. Note that the superblock has not been read in yet.
296 xfs_qm_mount_quotainit(
301 * User, projects or group quotas has to be on.
303 ASSERT(flags
& (XFSMNT_UQUOTA
| XFSMNT_PQUOTA
| XFSMNT_GQUOTA
));
306 * Initialize the flags in the mount structure. From this point
307 * onwards we look at m_qflags to figure out if quotas's ON/OFF, etc.
308 * Note that we enforce nothing if accounting is off.
309 * ie. XFSMNT_*QUOTA must be ON for XFSMNT_*QUOTAENF.
310 * It isn't necessary to take the quotaoff lock to do this; this is
313 if (flags
& XFSMNT_UQUOTA
) {
314 mp
->m_qflags
|= (XFS_UQUOTA_ACCT
| XFS_UQUOTA_ACTIVE
);
315 if (flags
& XFSMNT_UQUOTAENF
)
316 mp
->m_qflags
|= XFS_UQUOTA_ENFD
;
318 if (flags
& XFSMNT_GQUOTA
) {
319 mp
->m_qflags
|= (XFS_GQUOTA_ACCT
| XFS_GQUOTA_ACTIVE
);
320 if (flags
& XFSMNT_GQUOTAENF
)
321 mp
->m_qflags
|= XFS_OQUOTA_ENFD
;
322 } else if (flags
& XFSMNT_PQUOTA
) {
323 mp
->m_qflags
|= (XFS_PQUOTA_ACCT
| XFS_PQUOTA_ACTIVE
);
324 if (flags
& XFSMNT_PQUOTAENF
)
325 mp
->m_qflags
|= XFS_OQUOTA_ENFD
;
330 * Just destroy the quotainfo structure.
333 xfs_qm_unmount_quotadestroy(
337 xfs_qm_destroy_quotainfo(mp
);
342 * This is called from xfs_mountfs to start quotas and initialize all
343 * necessary data structures like quotainfo. This is also responsible for
344 * running a quotacheck as necessary. We are guaranteed that the superblock
345 * is consistently read in at this point.
358 * If quotas on realtime volumes is not supported, we disable
359 * quotas immediately.
361 if (mp
->m_sb
.sb_rextents
) {
363 "Cannot turn on quotas for realtime filesystem %s",
369 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
372 * Allocate the quotainfo structure inside the mount struct, and
373 * create quotainode(s), and change/rev superblock if necessary.
375 if ((error
= xfs_qm_init_quotainfo(mp
))) {
377 * We must turn off quotas.
379 ASSERT(mp
->m_quotainfo
== NULL
);
384 * If any of the quotas are not consistent, do a quotacheck.
386 if (XFS_QM_NEED_QUOTACHECK(mp
) &&
387 !(mfsi_flags
& XFS_MFSI_NO_QUOTACHECK
)) {
388 if ((error
= xfs_qm_quotacheck(mp
))) {
389 /* Quotacheck has failed and quotas have
392 return XFS_ERROR(error
);
398 * We actually don't have to acquire the SB_LOCK at all.
399 * This can only be called from mount, and that's single threaded. XXX
402 sbf
= mp
->m_sb
.sb_qflags
;
403 mp
->m_sb
.sb_qflags
= mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
;
404 XFS_SB_UNLOCK(mp
, s
);
406 if (sbf
!= (mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
)) {
407 if (xfs_qm_write_sb_changes(mp
, XFS_SB_QFLAGS
)) {
409 * We could only have been turning quotas off.
410 * We aren't in very good shape actually because
411 * the incore structures are convinced that quotas are
412 * off, but the on disk superblock doesn't know that !
414 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp
)));
415 xfs_fs_cmn_err(CE_ALERT
, mp
,
416 "XFS mount_quotas: Superblock update failed!");
421 xfs_fs_cmn_err(CE_WARN
, mp
,
422 "Failed to initialize disk quotas.");
424 return XFS_ERROR(error
);
428 * Called from the vfsops layer.
431 xfs_qm_unmount_quotas(
434 xfs_inode_t
*uqp
, *gqp
;
438 * Release the dquots that root inode, et al might be holding,
439 * before we flush quotas and blow away the quotainfo structure.
441 ASSERT(mp
->m_rootip
);
442 xfs_qm_dqdetach(mp
->m_rootip
);
444 xfs_qm_dqdetach(mp
->m_rbmip
);
446 xfs_qm_dqdetach(mp
->m_rsumip
);
449 * Flush out the quota inodes.
452 if (mp
->m_quotainfo
) {
453 if ((uqp
= mp
->m_quotainfo
->qi_uquotaip
) != NULL
) {
454 xfs_ilock(uqp
, XFS_ILOCK_EXCL
);
456 error
= xfs_iflush(uqp
, XFS_IFLUSH_SYNC
);
457 xfs_iunlock(uqp
, XFS_ILOCK_EXCL
);
458 if (unlikely(error
== EFSCORRUPTED
)) {
459 XFS_ERROR_REPORT("xfs_qm_unmount_quotas(1)",
460 XFS_ERRLEVEL_LOW
, mp
);
464 if ((gqp
= mp
->m_quotainfo
->qi_gquotaip
) != NULL
) {
465 xfs_ilock(gqp
, XFS_ILOCK_EXCL
);
467 error
= xfs_iflush(gqp
, XFS_IFLUSH_SYNC
);
468 xfs_iunlock(gqp
, XFS_ILOCK_EXCL
);
469 if (unlikely(error
== EFSCORRUPTED
)) {
470 XFS_ERROR_REPORT("xfs_qm_unmount_quotas(2)",
471 XFS_ERRLEVEL_LOW
, mp
);
477 XFS_PURGE_INODE(uqp
);
478 mp
->m_quotainfo
->qi_uquotaip
= NULL
;
481 XFS_PURGE_INODE(gqp
);
482 mp
->m_quotainfo
->qi_gquotaip
= NULL
;
485 return XFS_ERROR(error
);
489 * Flush all dquots of the given file system to disk. The dquots are
490 * _not_ purged from memory here, just their data written to disk.
502 if (mp
->m_quotainfo
== NULL
)
506 xfs_qm_mplist_lock(mp
);
507 FOREACH_DQUOT_IN_MP(dqp
, mp
) {
509 if (! XFS_DQ_IS_DIRTY(dqp
)) {
513 xfs_dqtrace_entry(dqp
, "FLUSHALL: DQDIRTY");
514 /* XXX a sentinel would be better */
515 recl
= XFS_QI_MPLRECLAIMS(mp
);
516 if (! xfs_qm_dqflock_nowait(dqp
)) {
518 * If we can't grab the flush lock then check
519 * to see if the dquot has been flushed delayed
520 * write. If so, grab its buffer and send it
521 * out immediately. We'll be able to acquire
522 * the flush lock when the I/O completes.
524 xfs_qm_dqflock_pushbuf_wait(dqp
);
527 * Let go of the mplist lock. We don't want to hold it
528 * across a disk write.
530 xfs_qm_mplist_unlock(mp
);
531 error
= xfs_qm_dqflush(dqp
, flags
);
536 xfs_qm_mplist_lock(mp
);
537 if (recl
!= XFS_QI_MPLRECLAIMS(mp
)) {
538 xfs_qm_mplist_unlock(mp
);
539 /* XXX restart limit */
544 xfs_qm_mplist_unlock(mp
);
549 * Release the group dquot pointers the user dquots may be
550 * carrying around as a hint. mplist is locked on entry and exit.
553 xfs_qm_detach_gdquots(
556 xfs_dquot_t
*dqp
, *gdqp
;
560 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp
));
561 dqp
= XFS_QI_MPLNEXT(mp
);
564 if ((gdqp
= dqp
->q_gdquot
)) {
566 dqp
->q_gdquot
= NULL
;
572 * Can't hold the mplist lock across a dqput.
573 * XXXmust convert to marker based iterations here.
575 nrecl
= XFS_QI_MPLRECLAIMS(mp
);
576 xfs_qm_mplist_unlock(mp
);
579 xfs_qm_mplist_lock(mp
);
580 if (nrecl
!= XFS_QI_MPLRECLAIMS(mp
))
588 * Go through all the incore dquots of this file system and take them
589 * off the mplist and hashlist, if the dquot type matches the dqtype
590 * parameter. This is used when turning off quota accounting for
591 * users and/or groups, as well as when the filesystem is unmounting.
596 uint flags
) /* QUOTAOFF/UMOUNTING/UQUOTA/PQUOTA/GQUOTA */
601 xfs_dquot_t
*nextdqp
;
604 if (mp
->m_quotainfo
== NULL
)
607 dqtype
= (flags
& XFS_QMOPT_UQUOTA
) ? XFS_DQ_USER
: 0;
608 dqtype
|= (flags
& XFS_QMOPT_PQUOTA
) ? XFS_DQ_PROJ
: 0;
609 dqtype
|= (flags
& XFS_QMOPT_GQUOTA
) ? XFS_DQ_GROUP
: 0;
611 xfs_qm_mplist_lock(mp
);
614 * In the first pass through all incore dquots of this filesystem,
615 * we release the group dquot pointers the user dquots may be
616 * carrying around as a hint. We need to do this irrespective of
617 * what's being turned off.
619 xfs_qm_detach_gdquots(mp
);
623 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp
));
625 * Try to get rid of all of the unwanted dquots. The idea is to
626 * get them off mplist and hashlist, but leave them on freelist.
628 dqp
= XFS_QI_MPLNEXT(mp
);
631 * It's OK to look at the type without taking dqlock here.
632 * We're holding the mplist lock here, and that's needed for
635 if ((dqp
->dq_flags
& dqtype
) == 0) {
640 if (! xfs_qm_dqhashlock_nowait(dqp
)) {
641 nrecl
= XFS_QI_MPLRECLAIMS(mp
);
642 xfs_qm_mplist_unlock(mp
);
643 XFS_DQ_HASH_LOCK(dqp
->q_hash
);
644 xfs_qm_mplist_lock(mp
);
647 * XXXTheoretically, we can get into a very long
648 * ping pong game here.
649 * No one can be adding dquots to the mplist at
650 * this point, but somebody might be taking things off.
652 if (nrecl
!= XFS_QI_MPLRECLAIMS(mp
)) {
653 XFS_DQ_HASH_UNLOCK(dqp
->q_hash
);
659 * Take the dquot off the mplist and hashlist. It may remain on
660 * freelist in INACTIVE state.
662 nextdqp
= dqp
->MPL_NEXT
;
663 nmisses
+= xfs_qm_dqpurge(dqp
, flags
);
666 xfs_qm_mplist_unlock(mp
);
678 * Purge the dquot cache.
679 * None of the dquots should really be busy at this point.
681 if (mp
->m_quotainfo
) {
682 while ((ndquots
= xfs_qm_dqpurge_int(mp
, flags
))) {
696 xfs_dquot_t
*udqhint
, /* hint */
697 xfs_dquot_t
**IO_idqpp
)
702 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
705 * See if we already have it in the inode itself. IO_idqpp is
706 * &i_udquot or &i_gdquot. This made the code look weird, but
707 * made the logic a lot simpler.
709 if ((dqp
= *IO_idqpp
)) {
712 xfs_dqtrace_entry(dqp
, "DQATTACH: found in ip");
717 * udqhint is the i_udquot field in inode, and is non-NULL only
718 * when the type arg is group/project. Its purpose is to save a
719 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
722 ASSERT(!udqhint
|| type
== XFS_DQ_GROUP
|| type
== XFS_DQ_PROJ
);
723 if (udqhint
&& !dolock
)
727 * No need to take dqlock to look at the id.
728 * The ID can't change until it gets reclaimed, and it won't
729 * be reclaimed as long as we have a ref from inode and we hold
733 (dqp
= udqhint
->q_gdquot
) &&
734 (be32_to_cpu(dqp
->q_core
.d_id
) == id
)) {
735 ASSERT(XFS_DQ_IS_LOCKED(udqhint
));
738 ASSERT(*IO_idqpp
== NULL
);
742 xfs_dqunlock(udqhint
);
747 * We can't hold a dquot lock when we call the dqget code.
748 * We'll deadlock in no time, because of (not conforming to)
749 * lock ordering - the inodelock comes before any dquot lock,
750 * and we may drop and reacquire the ilock in xfs_qm_dqget().
753 xfs_dqunlock(udqhint
);
755 * Find the dquot from somewhere. This bumps the
756 * reference count of dquot and returns it locked.
757 * This can return ENOENT if dquot didn't exist on
758 * disk and we didn't ask it to allocate;
759 * ESRCH if quotas got turned off suddenly.
761 if ((error
= xfs_qm_dqget(ip
->i_mount
, ip
, id
, type
,
762 doalloc
|XFS_QMOPT_DOWARN
, &dqp
))) {
763 if (udqhint
&& dolock
)
768 xfs_dqtrace_entry(dqp
, "DQATTACH: found by dqget");
770 * dqget may have dropped and re-acquired the ilock, but it guarantees
771 * that the dquot returned is the one that should go in the inode.
775 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
785 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
786 if (! xfs_qm_dqlock_nowait(udqhint
)) {
795 ASSERT(XFS_DQ_IS_LOCKED(udqhint
));
799 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
807 * Given a udquot and gdquot, attach a ptr to the group dquot in the
808 * udquot as a hint for future lookups. The idea sounds simple, but the
809 * execution isn't, because the udquot might have a group dquot attached
810 * already and getting rid of that gets us into lock ordering constraints.
811 * The process is complicated more by the fact that the dquots may or may not
812 * be locked on entry.
815 xfs_qm_dqattach_grouphint(
824 ASSERT(XFS_DQ_IS_LOCKED(udq
));
825 ASSERT(XFS_DQ_IS_LOCKED(gdq
));
831 if ((tmp
= udq
->q_gdquot
)) {
838 udq
->q_gdquot
= NULL
;
840 * We can't keep any dqlocks when calling dqrele,
841 * because the freelist lock comes before dqlocks.
847 * we took a hard reference once upon a time in dqget,
848 * so give it back when the udquot no longer points at it
849 * dqput() does the unlocking of the dquot.
857 ASSERT(XFS_DQ_IS_LOCKED(udq
));
863 ASSERT(XFS_DQ_IS_LOCKED(udq
));
864 ASSERT(XFS_DQ_IS_LOCKED(gdq
));
866 * Somebody could have attached a gdquot here,
867 * when we dropped the uqlock. If so, just do nothing.
869 if (udq
->q_gdquot
== NULL
) {
881 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
883 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
884 * If XFS_QMOPT_DQLOCK, the dquot(s) will be returned locked. This option pretty
885 * much made this code a complete mess, but it has been pretty useful.
886 * If XFS_QMOPT_ILOCKED, then inode sent is already locked EXCL.
887 * Inode may get unlocked and relocked in here, and the caller must deal with
895 xfs_mount_t
*mp
= ip
->i_mount
;
899 if ((! XFS_IS_QUOTA_ON(mp
)) ||
900 (! XFS_NOT_DQATTACHED(mp
, ip
)) ||
901 (ip
->i_ino
== mp
->m_sb
.sb_uquotino
) ||
902 (ip
->i_ino
== mp
->m_sb
.sb_gquotino
))
905 ASSERT((flags
& XFS_QMOPT_ILOCKED
) == 0 ||
906 XFS_ISLOCKED_INODE_EXCL(ip
));
908 if (! (flags
& XFS_QMOPT_ILOCKED
))
909 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
911 if (XFS_IS_UQUOTA_ON(mp
)) {
912 error
= xfs_qm_dqattach_one(ip
, ip
->i_d
.di_uid
, XFS_DQ_USER
,
913 flags
& XFS_QMOPT_DQALLOC
,
914 flags
& XFS_QMOPT_DQLOCK
,
915 NULL
, &ip
->i_udquot
);
920 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
921 if (XFS_IS_OQUOTA_ON(mp
)) {
922 error
= XFS_IS_GQUOTA_ON(mp
) ?
923 xfs_qm_dqattach_one(ip
, ip
->i_d
.di_gid
, XFS_DQ_GROUP
,
924 flags
& XFS_QMOPT_DQALLOC
,
925 flags
& XFS_QMOPT_DQLOCK
,
926 ip
->i_udquot
, &ip
->i_gdquot
) :
927 xfs_qm_dqattach_one(ip
, ip
->i_d
.di_projid
, XFS_DQ_PROJ
,
928 flags
& XFS_QMOPT_DQALLOC
,
929 flags
& XFS_QMOPT_DQLOCK
,
930 ip
->i_udquot
, &ip
->i_gdquot
);
932 * Don't worry about the udquot that we may have
933 * attached above. It'll get detached, if not already.
941 * Attach this group quota to the user quota as a hint.
942 * This WON'T, in general, result in a thrash.
945 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
946 ASSERT(ip
->i_udquot
);
947 ASSERT(ip
->i_gdquot
);
950 * We may or may not have the i_udquot locked at this point,
951 * but this check is OK since we don't depend on the i_gdquot to
952 * be accurate 100% all the time. It is just a hint, and this
953 * will succeed in general.
955 if (ip
->i_udquot
->q_gdquot
== ip
->i_gdquot
)
958 * Attach i_gdquot to the gdquot hint inside the i_udquot.
960 xfs_qm_dqattach_grouphint(ip
->i_udquot
, ip
->i_gdquot
,
961 flags
& XFS_QMOPT_DQLOCK
);
969 if (flags
& XFS_QMOPT_DQLOCK
)
970 ASSERT(XFS_DQ_IS_LOCKED(ip
->i_udquot
));
973 if (flags
& XFS_QMOPT_DQLOCK
)
974 ASSERT(XFS_DQ_IS_LOCKED(ip
->i_gdquot
));
976 if (XFS_IS_UQUOTA_ON(mp
))
977 ASSERT(ip
->i_udquot
);
978 if (XFS_IS_OQUOTA_ON(mp
))
979 ASSERT(ip
->i_gdquot
);
983 if (! (flags
& XFS_QMOPT_ILOCKED
))
984 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
988 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
994 * Release dquots (and their references) if any.
995 * The inode should be locked EXCL except when this's called by
1002 if (!(ip
->i_udquot
|| ip
->i_gdquot
))
1005 ASSERT(ip
->i_ino
!= ip
->i_mount
->m_sb
.sb_uquotino
);
1006 ASSERT(ip
->i_ino
!= ip
->i_mount
->m_sb
.sb_gquotino
);
1008 xfs_dqtrace_entry_ino(ip
->i_udquot
, "DQDETTACH", ip
);
1009 xfs_qm_dqrele(ip
->i_udquot
);
1010 ip
->i_udquot
= NULL
;
1013 xfs_dqtrace_entry_ino(ip
->i_gdquot
, "DQDETTACH", ip
);
1014 xfs_qm_dqrele(ip
->i_gdquot
);
1015 ip
->i_gdquot
= NULL
;
1020 * This is called by VFS_SYNC and flags arg determines the caller,
1021 * and its motives, as done in xfs_sync.
1023 * vfs_sync: SYNC_FSDATA|SYNC_ATTR|SYNC_BDFLUSH 0x31
1024 * syscall sync: SYNC_FSDATA|SYNC_ATTR|SYNC_DELWRI 0x25
1025 * umountroot : SYNC_WAIT | SYNC_CLOSE | SYNC_ATTR | SYNC_FSDATA
1041 * We won't block unless we are asked to.
1043 nowait
= (boolean_t
)(flags
& SYNC_BDFLUSH
|| (flags
& SYNC_WAIT
) == 0);
1046 xfs_qm_mplist_lock(mp
);
1048 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
1049 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
1050 * when we have the mplist lock, we know that dquots will be consistent
1051 * as long as we have it locked.
1053 if (! XFS_IS_QUOTA_ON(mp
)) {
1054 xfs_qm_mplist_unlock(mp
);
1057 FOREACH_DQUOT_IN_MP(dqp
, mp
) {
1059 * If this is vfs_sync calling, then skip the dquots that
1060 * don't 'seem' to be dirty. ie. don't acquire dqlock.
1061 * This is very similar to what xfs_sync does with inodes.
1063 if (flags
& SYNC_BDFLUSH
) {
1064 if (! XFS_DQ_IS_DIRTY(dqp
))
1070 * Try to acquire the dquot lock. We are NOT out of
1071 * lock order, but we just don't want to wait for this
1072 * lock, unless somebody wanted us to.
1074 if (! xfs_qm_dqlock_nowait(dqp
))
1081 * Now, find out for sure if this dquot is dirty or not.
1083 if (! XFS_DQ_IS_DIRTY(dqp
)) {
1088 /* XXX a sentinel would be better */
1089 recl
= XFS_QI_MPLRECLAIMS(mp
);
1090 if (! xfs_qm_dqflock_nowait(dqp
)) {
1096 * If we can't grab the flush lock then if the caller
1097 * really wanted us to give this our best shot, so
1098 * see if we can give a push to the buffer before we wait
1099 * on the flush lock. At this point, we know that
1100 * even though the dquot is being flushed,
1101 * it has (new) dirty data.
1103 xfs_qm_dqflock_pushbuf_wait(dqp
);
1106 * Let go of the mplist lock. We don't want to hold it
1107 * across a disk write
1109 flush_flags
= (nowait
) ? XFS_QMOPT_DELWRI
: XFS_QMOPT_SYNC
;
1110 xfs_qm_mplist_unlock(mp
);
1111 xfs_dqtrace_entry(dqp
, "XQM_SYNC: DQFLUSH");
1112 error
= xfs_qm_dqflush(dqp
, flush_flags
);
1114 if (error
&& XFS_FORCED_SHUTDOWN(mp
))
1115 return 0; /* Need to prevent umount failure */
1119 xfs_qm_mplist_lock(mp
);
1120 if (recl
!= XFS_QI_MPLRECLAIMS(mp
)) {
1121 if (++restarts
>= XFS_QM_SYNC_MAX_RESTARTS
)
1124 xfs_qm_mplist_unlock(mp
);
1129 xfs_qm_mplist_unlock(mp
);
1135 * This initializes all the quota information that's kept in the
1139 xfs_qm_init_quotainfo(
1142 xfs_quotainfo_t
*qinf
;
1146 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1149 * Tell XQM that we exist as soon as possible.
1151 if ((error
= xfs_qm_hold_quotafs_ref(mp
))) {
1155 qinf
= mp
->m_quotainfo
= kmem_zalloc(sizeof(xfs_quotainfo_t
), KM_SLEEP
);
1158 * See if quotainodes are setup, and if not, allocate them,
1159 * and change the superblock accordingly.
1161 if ((error
= xfs_qm_init_quotainos(mp
))) {
1162 kmem_free(qinf
, sizeof(xfs_quotainfo_t
));
1163 mp
->m_quotainfo
= NULL
;
1167 spinlock_init(&qinf
->qi_pinlock
, "xfs_qinf_pin");
1168 xfs_qm_list_init(&qinf
->qi_dqlist
, "mpdqlist", 0);
1169 qinf
->qi_dqreclaims
= 0;
1171 /* mutex used to serialize quotaoffs */
1172 mutex_init(&qinf
->qi_quotaofflock
);
1174 /* Precalc some constants */
1175 qinf
->qi_dqchunklen
= XFS_FSB_TO_BB(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
1176 ASSERT(qinf
->qi_dqchunklen
);
1177 qinf
->qi_dqperchunk
= BBTOB(qinf
->qi_dqchunklen
);
1178 do_div(qinf
->qi_dqperchunk
, sizeof(xfs_dqblk_t
));
1180 mp
->m_qflags
|= (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_CHKD
);
1183 * We try to get the limits from the superuser's limits fields.
1184 * This is quite hacky, but it is standard quota practice.
1185 * We look at the USR dquot with id == 0 first, but if user quotas
1186 * are not enabled we goto the GRP dquot with id == 0.
1187 * We don't really care to keep separate default limits for user
1188 * and group quotas, at least not at this point.
1190 error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)0,
1191 XFS_IS_UQUOTA_RUNNING(mp
) ? XFS_DQ_USER
:
1192 (XFS_IS_GQUOTA_RUNNING(mp
) ? XFS_DQ_GROUP
:
1194 XFS_QMOPT_DQSUSER
|XFS_QMOPT_DOWARN
,
1197 xfs_disk_dquot_t
*ddqp
= &dqp
->q_core
;
1200 * The warnings and timers set the grace period given to
1201 * a user or group before he or she can not perform any
1202 * more writing. If it is zero, a default is used.
1204 qinf
->qi_btimelimit
= ddqp
->d_btimer
?
1205 be32_to_cpu(ddqp
->d_btimer
) : XFS_QM_BTIMELIMIT
;
1206 qinf
->qi_itimelimit
= ddqp
->d_itimer
?
1207 be32_to_cpu(ddqp
->d_itimer
) : XFS_QM_ITIMELIMIT
;
1208 qinf
->qi_rtbtimelimit
= ddqp
->d_rtbtimer
?
1209 be32_to_cpu(ddqp
->d_rtbtimer
) : XFS_QM_RTBTIMELIMIT
;
1210 qinf
->qi_bwarnlimit
= ddqp
->d_bwarns
?
1211 be16_to_cpu(ddqp
->d_bwarns
) : XFS_QM_BWARNLIMIT
;
1212 qinf
->qi_iwarnlimit
= ddqp
->d_iwarns
?
1213 be16_to_cpu(ddqp
->d_iwarns
) : XFS_QM_IWARNLIMIT
;
1214 qinf
->qi_rtbwarnlimit
= ddqp
->d_rtbwarns
?
1215 be16_to_cpu(ddqp
->d_rtbwarns
) : XFS_QM_RTBWARNLIMIT
;
1216 qinf
->qi_bhardlimit
= be64_to_cpu(ddqp
->d_blk_hardlimit
);
1217 qinf
->qi_bsoftlimit
= be64_to_cpu(ddqp
->d_blk_softlimit
);
1218 qinf
->qi_ihardlimit
= be64_to_cpu(ddqp
->d_ino_hardlimit
);
1219 qinf
->qi_isoftlimit
= be64_to_cpu(ddqp
->d_ino_softlimit
);
1220 qinf
->qi_rtbhardlimit
= be64_to_cpu(ddqp
->d_rtb_hardlimit
);
1221 qinf
->qi_rtbsoftlimit
= be64_to_cpu(ddqp
->d_rtb_softlimit
);
1224 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1225 * we don't want this dquot cached. We haven't done a
1226 * quotacheck yet, and quotacheck doesn't like incore dquots.
1228 xfs_qm_dqdestroy(dqp
);
1230 qinf
->qi_btimelimit
= XFS_QM_BTIMELIMIT
;
1231 qinf
->qi_itimelimit
= XFS_QM_ITIMELIMIT
;
1232 qinf
->qi_rtbtimelimit
= XFS_QM_RTBTIMELIMIT
;
1233 qinf
->qi_bwarnlimit
= XFS_QM_BWARNLIMIT
;
1234 qinf
->qi_iwarnlimit
= XFS_QM_IWARNLIMIT
;
1235 qinf
->qi_rtbwarnlimit
= XFS_QM_RTBWARNLIMIT
;
1243 * Gets called when unmounting a filesystem or when all quotas get
1245 * This purges the quota inodes, destroys locks and frees itself.
1248 xfs_qm_destroy_quotainfo(
1251 xfs_quotainfo_t
*qi
;
1253 qi
= mp
->m_quotainfo
;
1255 ASSERT(xfs_Gqm
!= NULL
);
1258 * Release the reference that XQM kept, so that we know
1259 * when the XQM structure should be freed. We cannot assume
1260 * that xfs_Gqm is non-null after this point.
1262 xfs_qm_rele_quotafs_ref(mp
);
1264 spinlock_destroy(&qi
->qi_pinlock
);
1265 xfs_qm_list_destroy(&qi
->qi_dqlist
);
1267 if (qi
->qi_uquotaip
) {
1268 XFS_PURGE_INODE(qi
->qi_uquotaip
);
1269 qi
->qi_uquotaip
= NULL
; /* paranoia */
1271 if (qi
->qi_gquotaip
) {
1272 XFS_PURGE_INODE(qi
->qi_gquotaip
);
1273 qi
->qi_gquotaip
= NULL
;
1275 mutex_destroy(&qi
->qi_quotaofflock
);
1276 kmem_free(qi
, sizeof(xfs_quotainfo_t
));
1277 mp
->m_quotainfo
= NULL
;
1282 /* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1291 mutex_init(&list
->qh_lock
);
1292 list
->qh_next
= NULL
;
1293 list
->qh_version
= 0;
1294 list
->qh_nelems
= 0;
1298 xfs_qm_list_destroy(
1301 mutex_destroy(&(list
->qh_lock
));
1306 * Stripped down version of dqattach. This doesn't attach, or even look at the
1307 * dquots attached to the inode. The rationale is that there won't be any
1308 * attached at the time this is called from quotacheck.
1311 xfs_qm_dqget_noattach(
1313 xfs_dquot_t
**O_udqpp
,
1314 xfs_dquot_t
**O_gdqpp
)
1318 xfs_dquot_t
*udqp
, *gdqp
;
1320 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
1325 if (XFS_IS_UQUOTA_ON(mp
)) {
1326 ASSERT(ip
->i_udquot
== NULL
);
1328 * We want the dquot allocated if it doesn't exist.
1330 if ((error
= xfs_qm_dqget(mp
, ip
, ip
->i_d
.di_uid
, XFS_DQ_USER
,
1331 XFS_QMOPT_DQALLOC
| XFS_QMOPT_DOWARN
,
1334 * Shouldn't be able to turn off quotas here.
1336 ASSERT(error
!= ESRCH
);
1337 ASSERT(error
!= ENOENT
);
1343 if (XFS_IS_OQUOTA_ON(mp
)) {
1344 ASSERT(ip
->i_gdquot
== NULL
);
1347 error
= XFS_IS_GQUOTA_ON(mp
) ?
1348 xfs_qm_dqget(mp
, ip
,
1349 ip
->i_d
.di_gid
, XFS_DQ_GROUP
,
1350 XFS_QMOPT_DQALLOC
|XFS_QMOPT_DOWARN
,
1352 xfs_qm_dqget(mp
, ip
,
1353 ip
->i_d
.di_projid
, XFS_DQ_PROJ
,
1354 XFS_QMOPT_DQALLOC
|XFS_QMOPT_DOWARN
,
1358 xfs_qm_dqrele(udqp
);
1359 ASSERT(error
!= ESRCH
);
1360 ASSERT(error
!= ENOENT
);
1365 /* Reacquire the locks in the right order */
1367 if (! xfs_qm_dqlock_nowait(udqp
)) {
1379 if (udqp
) ASSERT(XFS_DQ_IS_LOCKED(udqp
));
1380 if (gdqp
) ASSERT(XFS_DQ_IS_LOCKED(gdqp
));
1386 * Create an inode and return with a reference already taken, but unlocked
1387 * This is how we create quota inodes
1401 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_QINOCREATE
);
1402 if ((error
= xfs_trans_reserve(tp
,
1403 XFS_QM_QINOCREATE_SPACE_RES(mp
),
1404 XFS_CREATE_LOG_RES(mp
), 0,
1405 XFS_TRANS_PERM_LOG_RES
,
1406 XFS_CREATE_LOG_COUNT
))) {
1407 xfs_trans_cancel(tp
, 0);
1411 if ((error
= xfs_dir_ialloc(&tp
, &xfs_zeroino
, S_IFREG
, 1, 0,
1412 &xfs_zerocr
, 0, 1, ip
, &committed
))) {
1413 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1419 * Keep an extra reference to this quota inode. This inode is
1420 * locked exclusively and joined to the transaction already.
1422 ASSERT(XFS_ISLOCKED_INODE_EXCL(*ip
));
1423 VN_HOLD(XFS_ITOV((*ip
)));
1426 * Make the changes in the superblock, and log those too.
1427 * sbfields arg may contain fields other than *QUOTINO;
1428 * VERSIONNUM for example.
1430 s
= XFS_SB_LOCK(mp
);
1431 if (flags
& XFS_QMOPT_SBVERSION
) {
1432 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1433 unsigned oldv
= mp
->m_sb
.sb_versionnum
;
1435 ASSERT(!XFS_SB_VERSION_HASQUOTA(&mp
->m_sb
));
1436 ASSERT((sbfields
& (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1437 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
)) ==
1438 (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1439 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
));
1441 XFS_SB_VERSION_ADDQUOTA(&mp
->m_sb
);
1442 mp
->m_sb
.sb_uquotino
= NULLFSINO
;
1443 mp
->m_sb
.sb_gquotino
= NULLFSINO
;
1445 /* qflags will get updated _after_ quotacheck */
1446 mp
->m_sb
.sb_qflags
= 0;
1447 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1449 "Old superblock version %x, converting to %x.",
1450 oldv
, mp
->m_sb
.sb_versionnum
);
1453 if (flags
& XFS_QMOPT_UQUOTA
)
1454 mp
->m_sb
.sb_uquotino
= (*ip
)->i_ino
;
1456 mp
->m_sb
.sb_gquotino
= (*ip
)->i_ino
;
1457 XFS_SB_UNLOCK(mp
, s
);
1458 xfs_mod_sb(tp
, sbfields
);
1460 if ((error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
,
1462 xfs_fs_cmn_err(CE_ALERT
, mp
, "XFS qino_alloc failed!");
1470 xfs_qm_reset_dqcounts(
1476 xfs_disk_dquot_t
*ddq
;
1479 xfs_buftrace("RESET DQUOTS", bp
);
1481 * Reset all counters and timers. They'll be
1482 * started afresh by xfs_qm_quotacheck.
1485 j
= XFS_FSB_TO_B(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
1486 do_div(j
, sizeof(xfs_dqblk_t
));
1487 ASSERT(XFS_QM_DQPERBLK(mp
) == j
);
1489 ddq
= (xfs_disk_dquot_t
*)XFS_BUF_PTR(bp
);
1490 for (j
= 0; j
< XFS_QM_DQPERBLK(mp
); j
++) {
1492 * Do a sanity check, and if needed, repair the dqblk. Don't
1493 * output any warnings because it's perfectly possible to
1494 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1496 (void) xfs_qm_dqcheck(ddq
, id
+j
, type
, XFS_QMOPT_DQREPAIR
,
1500 ddq
->d_rtbcount
= 0;
1503 ddq
->d_rtbtimer
= 0;
1506 ddq
->d_rtbwarns
= 0;
1507 ddq
= (xfs_disk_dquot_t
*) ((xfs_dqblk_t
*)ddq
+ 1);
1518 xfs_filblks_t blkcnt
,
1529 incr
= (blkcnt
> XFS_QM_MAX_DQCLUSTER_LOGSZ
) ?
1530 XFS_QM_MAX_DQCLUSTER_LOGSZ
: blkcnt
;
1531 type
= flags
& XFS_QMOPT_UQUOTA
? XFS_DQ_USER
:
1532 (flags
& XFS_QMOPT_PQUOTA
? XFS_DQ_PROJ
: XFS_DQ_GROUP
);
1536 * Blkcnt arg can be a very big number, and might even be
1537 * larger than the log itself. So, we have to break it up into
1538 * manageable-sized transactions.
1539 * Note that we don't start a permanent transaction here; we might
1540 * not be able to get a log reservation for the whole thing up front,
1541 * and we don't really care to either, because we just discard
1542 * everything if we were to crash in the middle of this loop.
1545 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
,
1546 XFS_FSB_TO_DADDR(mp
, bno
),
1547 (int)XFS_QI_DQCHUNKLEN(mp
), 0, &bp
);
1551 (void) xfs_qm_reset_dqcounts(mp
, bp
, firstid
, type
);
1552 xfs_bdwrite(mp
, bp
);
1554 * goto the next block.
1557 firstid
+= XFS_QM_DQPERBLK(mp
);
1563 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1564 * caller supplied function for every chunk of dquots that we find.
1572 xfs_bmbt_irec_t
*map
;
1573 int i
, nmaps
; /* number of map entries */
1574 int error
; /* return value */
1575 xfs_fileoff_t lblkno
;
1576 xfs_filblks_t maxlblkcnt
;
1578 xfs_fsblock_t rablkno
;
1579 xfs_filblks_t rablkcnt
;
1583 * This looks racy, but we can't keep an inode lock across a
1584 * trans_reserve. But, this gets called during quotacheck, and that
1585 * happens only at mount time which is single threaded.
1587 if (qip
->i_d
.di_nblocks
== 0)
1590 map
= kmem_alloc(XFS_DQITER_MAP_SIZE
* sizeof(*map
), KM_SLEEP
);
1593 maxlblkcnt
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1595 nmaps
= XFS_DQITER_MAP_SIZE
;
1597 * We aren't changing the inode itself. Just changing
1598 * some of its data. No new blocks are added here, and
1599 * the inode is never added to the transaction.
1601 xfs_ilock(qip
, XFS_ILOCK_SHARED
);
1602 error
= xfs_bmapi(NULL
, qip
, lblkno
,
1603 maxlblkcnt
- lblkno
,
1606 0, map
, &nmaps
, NULL
, NULL
);
1607 xfs_iunlock(qip
, XFS_ILOCK_SHARED
);
1611 ASSERT(nmaps
<= XFS_DQITER_MAP_SIZE
);
1612 for (i
= 0; i
< nmaps
; i
++) {
1613 ASSERT(map
[i
].br_startblock
!= DELAYSTARTBLOCK
);
1614 ASSERT(map
[i
].br_blockcount
);
1617 lblkno
+= map
[i
].br_blockcount
;
1619 if (map
[i
].br_startblock
== HOLESTARTBLOCK
)
1622 firstid
= (xfs_dqid_t
) map
[i
].br_startoff
*
1623 XFS_QM_DQPERBLK(mp
);
1625 * Do a read-ahead on the next extent.
1627 if ((i
+1 < nmaps
) &&
1628 (map
[i
+1].br_startblock
!= HOLESTARTBLOCK
)) {
1629 rablkcnt
= map
[i
+1].br_blockcount
;
1630 rablkno
= map
[i
+1].br_startblock
;
1631 while (rablkcnt
--) {
1632 xfs_baread(mp
->m_ddev_targp
,
1633 XFS_FSB_TO_DADDR(mp
, rablkno
),
1634 (int)XFS_QI_DQCHUNKLEN(mp
));
1639 * Iterate thru all the blks in the extent and
1640 * reset the counters of all the dquots inside them.
1642 if ((error
= xfs_qm_dqiter_bufs(mp
,
1644 map
[i
].br_startblock
,
1645 map
[i
].br_blockcount
,
1653 } while (nmaps
> 0);
1655 kmem_free(map
, XFS_DQITER_MAP_SIZE
* sizeof(*map
));
1661 * Called by dqusage_adjust in doing a quotacheck.
1662 * Given the inode, and a dquot (either USR or GRP, doesn't matter),
1663 * this updates its incore copy as well as the buffer copy. This is
1664 * so that once the quotacheck is done, we can just log all the buffers,
1665 * as opposed to logging numerous updates to individual dquots.
1668 xfs_qm_quotacheck_dqadjust(
1673 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1674 xfs_dqtrace_entry(dqp
, "QCHECK DQADJUST");
1676 * Adjust the inode count and the block count to reflect this inode's
1679 be64_add(&dqp
->q_core
.d_icount
, 1);
1680 dqp
->q_res_icount
++;
1682 be64_add(&dqp
->q_core
.d_bcount
, nblks
);
1683 dqp
->q_res_bcount
+= nblks
;
1686 be64_add(&dqp
->q_core
.d_rtbcount
, rtblks
);
1687 dqp
->q_res_rtbcount
+= rtblks
;
1691 * Set default limits, adjust timers (since we changed usages)
1693 if (! XFS_IS_SUSER_DQUOT(dqp
)) {
1694 xfs_qm_adjust_dqlimits(dqp
->q_mount
, &dqp
->q_core
);
1695 xfs_qm_adjust_dqtimers(dqp
->q_mount
, &dqp
->q_core
);
1698 dqp
->dq_flags
|= XFS_DQ_DIRTY
;
1704 xfs_qcnt_t
*O_rtblks
)
1706 xfs_filblks_t rtblks
; /* total rt blks */
1707 xfs_extnum_t idx
; /* extent record index */
1708 xfs_ifork_t
*ifp
; /* inode fork pointer */
1709 xfs_extnum_t nextents
; /* number of extent entries */
1710 xfs_bmbt_rec_t
*ep
; /* pointer to an extent entry */
1713 ASSERT(XFS_IS_REALTIME_INODE(ip
));
1714 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1715 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1716 if ((error
= xfs_iread_extents(NULL
, ip
, XFS_DATA_FORK
)))
1720 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
1721 for (idx
= 0; idx
< nextents
; idx
++) {
1722 ep
= xfs_iext_get_ext(ifp
, idx
);
1723 rtblks
+= xfs_bmbt_get_blockcount(ep
);
1725 *O_rtblks
= (xfs_qcnt_t
)rtblks
;
1730 * callback routine supplied to bulkstat(). Given an inumber, find its
1731 * dquots and update them to account for resources taken by that inode.
1735 xfs_qm_dqusage_adjust(
1736 xfs_mount_t
*mp
, /* mount point for filesystem */
1737 xfs_ino_t ino
, /* inode number to get data for */
1738 void __user
*buffer
, /* not used */
1739 int ubsize
, /* not used */
1740 void *private_data
, /* not used */
1741 xfs_daddr_t bno
, /* starting block of inode cluster */
1742 int *ubused
, /* not used */
1743 void *dip
, /* on-disk inode pointer (not used) */
1744 int *res
) /* result code value */
1747 xfs_dquot_t
*udqp
, *gdqp
;
1748 xfs_qcnt_t nblks
, rtblks
;
1751 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1754 * rootino must have its resources accounted for, not so with the quota
1757 if (ino
== mp
->m_sb
.sb_uquotino
|| ino
== mp
->m_sb
.sb_gquotino
) {
1758 *res
= BULKSTAT_RV_NOTHING
;
1759 return XFS_ERROR(EINVAL
);
1763 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1764 * interface expects the inode to be exclusively locked because that's
1765 * the case in all other instances. It's OK that we do this because
1766 * quotacheck is done only at mount time.
1768 if ((error
= xfs_iget(mp
, NULL
, ino
, 0, XFS_ILOCK_EXCL
, &ip
, bno
))) {
1769 *res
= BULKSTAT_RV_NOTHING
;
1773 if (ip
->i_d
.di_mode
== 0) {
1774 xfs_iput_new(ip
, XFS_ILOCK_EXCL
);
1775 *res
= BULKSTAT_RV_NOTHING
;
1776 return XFS_ERROR(ENOENT
);
1780 * Obtain the locked dquots. In case of an error (eg. allocation
1781 * fails for ENOSPC), we return the negative of the error number
1782 * to bulkstat, so that it can get propagated to quotacheck() and
1783 * making us disable quotas for the file system.
1785 if ((error
= xfs_qm_dqget_noattach(ip
, &udqp
, &gdqp
))) {
1786 xfs_iput(ip
, XFS_ILOCK_EXCL
);
1787 *res
= BULKSTAT_RV_GIVEUP
;
1792 if (! XFS_IS_REALTIME_INODE(ip
)) {
1793 nblks
= (xfs_qcnt_t
)ip
->i_d
.di_nblocks
;
1796 * Walk thru the extent list and count the realtime blocks.
1798 if ((error
= xfs_qm_get_rtblks(ip
, &rtblks
))) {
1799 xfs_iput(ip
, XFS_ILOCK_EXCL
);
1804 *res
= BULKSTAT_RV_GIVEUP
;
1807 nblks
= (xfs_qcnt_t
)ip
->i_d
.di_nblocks
- rtblks
;
1809 ASSERT(ip
->i_delayed_blks
== 0);
1812 * We can't release the inode while holding its dquot locks.
1813 * The inode can go into inactive and might try to acquire the dquotlocks.
1814 * So, just unlock here and do a vn_rele at the end.
1816 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1819 * Add the (disk blocks and inode) resources occupied by this
1820 * inode to its dquots. We do this adjustment in the incore dquot,
1821 * and also copy the changes to its buffer.
1822 * We don't care about putting these changes in a transaction
1823 * envelope because if we crash in the middle of a 'quotacheck'
1824 * we have to start from the beginning anyway.
1825 * Once we're done, we'll log all the dquot bufs.
1827 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1828 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1830 if (XFS_IS_UQUOTA_ON(mp
)) {
1832 xfs_qm_quotacheck_dqadjust(udqp
, nblks
, rtblks
);
1835 if (XFS_IS_OQUOTA_ON(mp
)) {
1837 xfs_qm_quotacheck_dqadjust(gdqp
, nblks
, rtblks
);
1841 * Now release the inode. This will send it to 'inactive', and
1842 * possibly even free blocks.
1844 VN_RELE(XFS_ITOV(ip
));
1849 *res
= BULKSTAT_RV_DIDONE
;
1854 * Walk thru all the filesystem inodes and construct a consistent view
1855 * of the disk quota world. If the quotacheck fails, disable quotas.
1861 int done
, count
, error
;
1864 xfs_inode_t
*uip
, *gip
;
1872 ASSERT(XFS_QI_UQIP(mp
) || XFS_QI_GQIP(mp
));
1873 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1876 * There should be no cached dquots. The (simplistic) quotacheck
1877 * algorithm doesn't like that.
1879 ASSERT(XFS_QI_MPLNDQUOTS(mp
) == 0);
1881 cmn_err(CE_NOTE
, "XFS quotacheck %s: Please wait.", mp
->m_fsname
);
1884 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1885 * their counters to zero. We need a clean slate.
1886 * We don't log our changes till later.
1888 if ((uip
= XFS_QI_UQIP(mp
))) {
1889 if ((error
= xfs_qm_dqiterate(mp
, uip
, XFS_QMOPT_UQUOTA
)))
1891 flags
|= XFS_UQUOTA_CHKD
;
1894 if ((gip
= XFS_QI_GQIP(mp
))) {
1895 if ((error
= xfs_qm_dqiterate(mp
, gip
, XFS_IS_GQUOTA_ON(mp
) ?
1896 XFS_QMOPT_GQUOTA
: XFS_QMOPT_PQUOTA
)))
1898 flags
|= XFS_OQUOTA_CHKD
;
1903 * Iterate thru all the inodes in the file system,
1904 * adjusting the corresponding dquot counters in core.
1906 if ((error
= xfs_bulkstat(mp
, &lastino
, &count
,
1907 xfs_qm_dqusage_adjust
, NULL
,
1909 BULKSTAT_FG_IGET
|BULKSTAT_FG_VFSLOCKED
,
1916 * We can get this error if we couldn't do a dquot allocation inside
1917 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1918 * dirty dquots that might be cached, we just want to get rid of them
1919 * and turn quotaoff. The dquots won't be attached to any of the inodes
1920 * at this point (because we intentionally didn't in dqget_noattach).
1923 xfs_qm_dqpurge_all(mp
, XFS_QMOPT_QUOTALL
| XFS_QMOPT_QUOTAOFF
);
1927 * We've made all the changes that we need to make incore.
1928 * Now flush_them down to disk buffers.
1930 xfs_qm_dqflush_all(mp
, XFS_QMOPT_DELWRI
);
1933 * We didn't log anything, because if we crashed, we'll have to
1934 * start the quotacheck from scratch anyway. However, we must make
1935 * sure that our dquot changes are secure before we put the
1936 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1939 XFS_bflush(mp
->m_ddev_targp
);
1942 * If one type of quotas is off, then it will lose its
1943 * quotachecked status, since we won't be doing accounting for
1944 * that type anymore.
1946 mp
->m_qflags
&= ~(XFS_OQUOTA_CHKD
| XFS_UQUOTA_CHKD
);
1947 mp
->m_qflags
|= flags
;
1949 XQM_LIST_PRINT(&(XFS_QI_MPL_LIST(mp
)), MPL_NEXT
, "++++ Mp list +++");
1953 cmn_err(CE_WARN
, "XFS quotacheck %s: Unsuccessful (Error %d): "
1954 "Disabling quotas.",
1955 mp
->m_fsname
, error
);
1957 * We must turn off quotas.
1959 ASSERT(mp
->m_quotainfo
!= NULL
);
1960 ASSERT(xfs_Gqm
!= NULL
);
1961 xfs_qm_destroy_quotainfo(mp
);
1962 (void)xfs_mount_reset_sbqflags(mp
);
1964 cmn_err(CE_NOTE
, "XFS quotacheck %s: Done.", mp
->m_fsname
);
1970 * This is called after the superblock has been read in and we're ready to
1971 * iget the quota inodes.
1974 xfs_qm_init_quotainos(
1977 xfs_inode_t
*uip
, *gip
;
1982 ASSERT(mp
->m_quotainfo
);
1988 * Get the uquota and gquota inodes
1990 if (XFS_SB_VERSION_HASQUOTA(&mp
->m_sb
)) {
1991 if (XFS_IS_UQUOTA_ON(mp
) &&
1992 mp
->m_sb
.sb_uquotino
!= NULLFSINO
) {
1993 ASSERT(mp
->m_sb
.sb_uquotino
> 0);
1994 if ((error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_uquotino
,
1996 return XFS_ERROR(error
);
1998 if (XFS_IS_OQUOTA_ON(mp
) &&
1999 mp
->m_sb
.sb_gquotino
!= NULLFSINO
) {
2000 ASSERT(mp
->m_sb
.sb_gquotino
> 0);
2001 if ((error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_gquotino
,
2004 VN_RELE(XFS_ITOV(uip
));
2005 return XFS_ERROR(error
);
2009 flags
|= XFS_QMOPT_SBVERSION
;
2010 sbflags
|= (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
2011 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
);
2015 * Create the two inodes, if they don't exist already. The changes
2016 * made above will get added to a transaction and logged in one of
2017 * the qino_alloc calls below. If the device is readonly,
2018 * temporarily switch to read-write to do this.
2020 if (XFS_IS_UQUOTA_ON(mp
) && uip
== NULL
) {
2021 if ((error
= xfs_qm_qino_alloc(mp
, &uip
,
2022 sbflags
| XFS_SB_UQUOTINO
,
2023 flags
| XFS_QMOPT_UQUOTA
)))
2024 return XFS_ERROR(error
);
2026 flags
&= ~XFS_QMOPT_SBVERSION
;
2028 if (XFS_IS_OQUOTA_ON(mp
) && gip
== NULL
) {
2029 flags
|= (XFS_IS_GQUOTA_ON(mp
) ?
2030 XFS_QMOPT_GQUOTA
: XFS_QMOPT_PQUOTA
);
2031 error
= xfs_qm_qino_alloc(mp
, &gip
,
2032 sbflags
| XFS_SB_GQUOTINO
, flags
);
2035 VN_RELE(XFS_ITOV(uip
));
2037 return XFS_ERROR(error
);
2041 XFS_QI_UQIP(mp
) = uip
;
2042 XFS_QI_GQIP(mp
) = gip
;
2049 * Traverse the freelist of dquots and attempt to reclaim a maximum of
2050 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
2051 * favor the lookup function ...
2052 * XXXsup merge this with qm_reclaim_one().
2055 xfs_qm_shake_freelist(
2060 xfs_dquot_t
*dqp
, *nextdqp
;
2072 cmn_err(CE_DEBUG
, "Shake free 0x%x", howmany
);
2074 /* lock order is : hashchainlock, freelistlock, mplistlock */
2076 xfs_qm_freelist_lock(xfs_Gqm
);
2078 for (dqp
= xfs_Gqm
->qm_dqfreelist
.qh_next
;
2079 ((dqp
!= (xfs_dquot_t
*) &xfs_Gqm
->qm_dqfreelist
) &&
2080 nreclaimed
< howmany
); ) {
2084 * We are racing with dqlookup here. Naturally we don't
2085 * want to reclaim a dquot that lookup wants.
2087 if (dqp
->dq_flags
& XFS_DQ_WANT
) {
2089 xfs_qm_freelist_unlock(xfs_Gqm
);
2090 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2092 XQM_STATS_INC(xqmstats
.xs_qm_dqwants
);
2097 * If the dquot is inactive, we are assured that it is
2098 * not on the mplist or the hashlist, and that makes our
2101 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
2102 ASSERT(dqp
->q_mount
== NULL
);
2103 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
2104 ASSERT(dqp
->HL_PREVP
== NULL
);
2105 ASSERT(dqp
->MPL_PREVP
== NULL
);
2106 XQM_STATS_INC(xqmstats
.xs_qm_dqinact_reclaims
);
2107 nextdqp
= dqp
->dq_flnext
;
2111 ASSERT(dqp
->MPL_PREVP
);
2113 * Try to grab the flush lock. If this dquot is in the process of
2114 * getting flushed to disk, we don't want to reclaim it.
2116 if (! xfs_qm_dqflock_nowait(dqp
)) {
2118 dqp
= dqp
->dq_flnext
;
2123 * We have the flush lock so we know that this is not in the
2124 * process of being flushed. So, if this is dirty, flush it
2125 * DELWRI so that we don't get a freelist infested with
2128 if (XFS_DQ_IS_DIRTY(dqp
)) {
2129 xfs_dqtrace_entry(dqp
, "DQSHAKE: DQDIRTY");
2131 * We flush it delayed write, so don't bother
2132 * releasing the mplock.
2134 (void) xfs_qm_dqflush(dqp
, XFS_QMOPT_DELWRI
);
2135 xfs_dqunlock(dqp
); /* dqflush unlocks dqflock */
2136 dqp
= dqp
->dq_flnext
;
2140 * We're trying to get the hashlock out of order. This races
2141 * with dqlookup; so, we giveup and goto the next dquot if
2142 * we couldn't get the hashlock. This way, we won't starve
2143 * a dqlookup process that holds the hashlock that is
2144 * waiting for the freelist lock.
2146 if (! xfs_qm_dqhashlock_nowait(dqp
)) {
2149 dqp
= dqp
->dq_flnext
;
2153 * This races with dquot allocation code as well as dqflush_all
2154 * and reclaim code. So, if we failed to grab the mplist lock,
2155 * giveup everything and start over.
2159 if (! xfs_qm_mplist_nowait(dqp
->q_mount
)) {
2160 /* XXX put a sentinel so that we can come back here */
2163 XFS_DQ_HASH_UNLOCK(hash
);
2164 xfs_qm_freelist_unlock(xfs_Gqm
);
2165 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2169 xfs_dqtrace_entry(dqp
, "DQSHAKE: UNLINKING");
2171 cmn_err(CE_DEBUG
, "Shake 0x%p, ID 0x%x\n",
2172 dqp
, be32_to_cpu(dqp
->q_core
.d_id
));
2174 ASSERT(dqp
->q_nrefs
== 0);
2175 nextdqp
= dqp
->dq_flnext
;
2176 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp
->q_mount
)), dqp
);
2177 XQM_HASHLIST_REMOVE(hash
, dqp
);
2179 xfs_qm_mplist_unlock(dqp
->q_mount
);
2180 XFS_DQ_HASH_UNLOCK(hash
);
2183 XQM_FREELIST_REMOVE(dqp
);
2186 XQM_STATS_INC(xqmstats
.xs_qm_dqshake_reclaims
);
2187 xfs_qm_dqdestroy(dqp
);
2190 xfs_qm_freelist_unlock(xfs_Gqm
);
2196 * The kmem_shake interface is invoked when memory is running low.
2200 xfs_qm_shake(int nr_to_scan
, gfp_t gfp_mask
)
2202 int ndqused
, nfree
, n
;
2204 if (!kmem_shake_allow(gfp_mask
))
2209 nfree
= xfs_Gqm
->qm_dqfreelist
.qh_nelems
; /* free dquots */
2210 /* incore dquots in all f/s's */
2211 ndqused
= atomic_read(&xfs_Gqm
->qm_totaldquots
) - nfree
;
2213 ASSERT(ndqused
>= 0);
2215 if (nfree
<= ndqused
&& nfree
< ndquot
)
2218 ndqused
*= xfs_Gqm
->qm_dqfree_ratio
; /* target # of free dquots */
2219 n
= nfree
- ndqused
- ndquot
; /* # over target */
2221 return xfs_qm_shake_freelist(MAX(nfree
, n
));
2226 * Just pop the least recently used dquot off the freelist and
2227 * recycle it. The returned dquot is locked.
2229 STATIC xfs_dquot_t
*
2230 xfs_qm_dqreclaim_one(void)
2232 xfs_dquot_t
*dqpout
;
2241 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
2243 xfs_qm_freelist_lock(xfs_Gqm
);
2245 FOREACH_DQUOT_IN_FREELIST(dqp
, &(xfs_Gqm
->qm_dqfreelist
)) {
2249 * We are racing with dqlookup here. Naturally we don't
2250 * want to reclaim a dquot that lookup wants. We release the
2251 * freelist lock and start over, so that lookup will grab
2252 * both the dquot and the freelistlock.
2254 if (dqp
->dq_flags
& XFS_DQ_WANT
) {
2255 ASSERT(! (dqp
->dq_flags
& XFS_DQ_INACTIVE
));
2256 xfs_dqtrace_entry(dqp
, "DQRECLAIM: DQWANT");
2258 xfs_qm_freelist_unlock(xfs_Gqm
);
2259 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2261 XQM_STATS_INC(xqmstats
.xs_qm_dqwants
);
2266 * If the dquot is inactive, we are assured that it is
2267 * not on the mplist or the hashlist, and that makes our
2270 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
2271 ASSERT(dqp
->q_mount
== NULL
);
2272 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
2273 ASSERT(dqp
->HL_PREVP
== NULL
);
2274 ASSERT(dqp
->MPL_PREVP
== NULL
);
2275 XQM_FREELIST_REMOVE(dqp
);
2278 XQM_STATS_INC(xqmstats
.xs_qm_dqinact_reclaims
);
2282 ASSERT(dqp
->q_hash
);
2283 ASSERT(dqp
->MPL_PREVP
);
2286 * Try to grab the flush lock. If this dquot is in the process of
2287 * getting flushed to disk, we don't want to reclaim it.
2289 if (! xfs_qm_dqflock_nowait(dqp
)) {
2295 * We have the flush lock so we know that this is not in the
2296 * process of being flushed. So, if this is dirty, flush it
2297 * DELWRI so that we don't get a freelist infested with
2300 if (XFS_DQ_IS_DIRTY(dqp
)) {
2301 xfs_dqtrace_entry(dqp
, "DQRECLAIM: DQDIRTY");
2303 * We flush it delayed write, so don't bother
2304 * releasing the freelist lock.
2306 (void) xfs_qm_dqflush(dqp
, XFS_QMOPT_DELWRI
);
2307 xfs_dqunlock(dqp
); /* dqflush unlocks dqflock */
2311 if (! xfs_qm_mplist_nowait(dqp
->q_mount
)) {
2317 if (! xfs_qm_dqhashlock_nowait(dqp
))
2320 ASSERT(dqp
->q_nrefs
== 0);
2321 xfs_dqtrace_entry(dqp
, "DQRECLAIM: UNLINKING");
2322 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp
->q_mount
)), dqp
);
2323 XQM_HASHLIST_REMOVE(dqp
->q_hash
, dqp
);
2324 XQM_FREELIST_REMOVE(dqp
);
2326 XFS_DQ_HASH_UNLOCK(dqp
->q_hash
);
2328 xfs_qm_mplist_unlock(dqp
->q_mount
);
2335 xfs_qm_freelist_unlock(xfs_Gqm
);
2340 /*------------------------------------------------------------------*/
2343 * Return a new incore dquot. Depending on the number of
2344 * dquots in the system, we either allocate a new one on the kernel heap,
2345 * or reclaim a free one.
2346 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2347 * to reclaim an existing one from the freelist.
2350 xfs_qm_dqalloc_incore(
2351 xfs_dquot_t
**O_dqpp
)
2356 * Check against high water mark to see if we want to pop
2357 * a nincompoop dquot off the freelist.
2359 if (atomic_read(&xfs_Gqm
->qm_totaldquots
) >= ndquot
) {
2361 * Try to recycle a dquot from the freelist.
2363 if ((dqp
= xfs_qm_dqreclaim_one())) {
2364 XQM_STATS_INC(xqmstats
.xs_qm_dqreclaims
);
2366 * Just zero the core here. The rest will get
2367 * reinitialized by caller. XXX we shouldn't even
2370 memset(&dqp
->q_core
, 0, sizeof(dqp
->q_core
));
2374 XQM_STATS_INC(xqmstats
.xs_qm_dqreclaim_misses
);
2378 * Allocate a brand new dquot on the kernel heap and return it
2379 * to the caller to initialize.
2381 ASSERT(xfs_Gqm
->qm_dqzone
!= NULL
);
2382 *O_dqpp
= kmem_zone_zalloc(xfs_Gqm
->qm_dqzone
, KM_SLEEP
);
2383 atomic_inc(&xfs_Gqm
->qm_totaldquots
);
2390 * Start a transaction and write the incore superblock changes to
2391 * disk. flags parameter indicates which fields have changed.
2394 xfs_qm_write_sb_changes(
2402 cmn_err(CE_NOTE
, "Writing superblock quota changes :%s", mp
->m_fsname
);
2404 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
2405 if ((error
= xfs_trans_reserve(tp
, 0,
2406 mp
->m_sb
.sb_sectsize
+ 128, 0,
2408 XFS_DEFAULT_LOG_COUNT
))) {
2409 xfs_trans_cancel(tp
, 0);
2413 xfs_mod_sb(tp
, flags
);
2414 (void) xfs_trans_commit(tp
, 0, NULL
);
2420 /* --------------- utility functions for vnodeops ---------------- */
2424 * Given an inode, a uid and gid (from cred_t) make sure that we have
2425 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2426 * quotas by creating this file.
2427 * This also attaches dquot(s) to the given inode after locking it,
2428 * and returns the dquots corresponding to the uid and/or gid.
2430 * in : inode (unlocked)
2431 * out : udquot, gdquot with references taken and unlocked
2441 xfs_dquot_t
**O_udqpp
,
2442 xfs_dquot_t
**O_gdqpp
)
2445 xfs_dquot_t
*uq
, *gq
;
2448 if (!XFS_IS_QUOTA_ON(mp
))
2451 lockflags
= XFS_ILOCK_EXCL
;
2452 xfs_ilock(ip
, lockflags
);
2454 if ((flags
& XFS_QMOPT_INHERIT
) &&
2455 XFS_INHERIT_GID(ip
, XFS_MTOVFS(mp
)))
2456 gid
= ip
->i_d
.di_gid
;
2459 * Attach the dquot(s) to this inode, doing a dquot allocation
2460 * if necessary. The dquot(s) will not be locked.
2462 if (XFS_NOT_DQATTACHED(mp
, ip
)) {
2463 if ((error
= xfs_qm_dqattach(ip
, XFS_QMOPT_DQALLOC
|
2464 XFS_QMOPT_ILOCKED
))) {
2465 xfs_iunlock(ip
, lockflags
);
2471 if ((flags
& XFS_QMOPT_UQUOTA
) && XFS_IS_UQUOTA_ON(mp
)) {
2472 if (ip
->i_d
.di_uid
!= uid
) {
2474 * What we need is the dquot that has this uid, and
2475 * if we send the inode to dqget, the uid of the inode
2476 * takes priority over what's sent in the uid argument.
2477 * We must unlock inode here before calling dqget if
2478 * we're not sending the inode, because otherwise
2479 * we'll deadlock by doing trans_reserve while
2482 xfs_iunlock(ip
, lockflags
);
2483 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
) uid
,
2488 ASSERT(error
!= ENOENT
);
2492 * Get the ilock in the right order.
2495 lockflags
= XFS_ILOCK_SHARED
;
2496 xfs_ilock(ip
, lockflags
);
2499 * Take an extra reference, because we'll return
2502 ASSERT(ip
->i_udquot
);
2509 if ((flags
& XFS_QMOPT_GQUOTA
) && XFS_IS_GQUOTA_ON(mp
)) {
2510 if (ip
->i_d
.di_gid
!= gid
) {
2511 xfs_iunlock(ip
, lockflags
);
2512 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)gid
,
2519 ASSERT(error
!= ENOENT
);
2523 lockflags
= XFS_ILOCK_SHARED
;
2524 xfs_ilock(ip
, lockflags
);
2526 ASSERT(ip
->i_gdquot
);
2532 } else if ((flags
& XFS_QMOPT_PQUOTA
) && XFS_IS_PQUOTA_ON(mp
)) {
2533 if (ip
->i_d
.di_projid
!= prid
) {
2534 xfs_iunlock(ip
, lockflags
);
2535 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)prid
,
2542 ASSERT(error
!= ENOENT
);
2546 lockflags
= XFS_ILOCK_SHARED
;
2547 xfs_ilock(ip
, lockflags
);
2549 ASSERT(ip
->i_gdquot
);
2557 xfs_dqtrace_entry_ino(uq
, "DQALLOC", ip
);
2559 xfs_iunlock(ip
, lockflags
);
2572 * Actually transfer ownership, and do dquot modifications.
2573 * These were already reserved.
2579 xfs_dquot_t
**IO_olddq
,
2582 xfs_dquot_t
*prevdq
;
2583 uint bfield
= XFS_IS_REALTIME_INODE(ip
) ?
2584 XFS_TRANS_DQ_RTBCOUNT
: XFS_TRANS_DQ_BCOUNT
;
2586 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
2587 ASSERT(XFS_IS_QUOTA_RUNNING(ip
->i_mount
));
2592 ASSERT(prevdq
!= newdq
);
2594 xfs_trans_mod_dquot(tp
, prevdq
, bfield
, -(ip
->i_d
.di_nblocks
));
2595 xfs_trans_mod_dquot(tp
, prevdq
, XFS_TRANS_DQ_ICOUNT
, -1);
2597 /* the sparkling new dquot */
2598 xfs_trans_mod_dquot(tp
, newdq
, bfield
, ip
->i_d
.di_nblocks
);
2599 xfs_trans_mod_dquot(tp
, newdq
, XFS_TRANS_DQ_ICOUNT
, 1);
2602 * Take an extra reference, because the inode
2603 * is going to keep this dquot pointer even
2604 * after the trans_commit.
2608 xfs_dqunlock(newdq
);
2615 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
2618 xfs_qm_vop_chown_reserve(
2627 uint delblks
, blkflags
, prjflags
= 0;
2628 xfs_dquot_t
*unresudq
, *unresgdq
, *delblksudq
, *delblksgdq
;
2630 ASSERT(XFS_ISLOCKED_INODE(ip
));
2632 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
2634 delblks
= ip
->i_delayed_blks
;
2635 delblksudq
= delblksgdq
= unresudq
= unresgdq
= NULL
;
2636 blkflags
= XFS_IS_REALTIME_INODE(ip
) ?
2637 XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
;
2639 if (XFS_IS_UQUOTA_ON(mp
) && udqp
&&
2640 ip
->i_d
.di_uid
!= (uid_t
)be32_to_cpu(udqp
->q_core
.d_id
)) {
2643 * If there are delayed allocation blocks, then we have to
2644 * unreserve those from the old dquot, and add them to the
2648 ASSERT(ip
->i_udquot
);
2649 unresudq
= ip
->i_udquot
;
2652 if (XFS_IS_OQUOTA_ON(ip
->i_mount
) && gdqp
) {
2653 if (XFS_IS_PQUOTA_ON(ip
->i_mount
) &&
2654 ip
->i_d
.di_projid
!= be32_to_cpu(gdqp
->q_core
.d_id
))
2655 prjflags
= XFS_QMOPT_ENOSPC
;
2658 (XFS_IS_GQUOTA_ON(ip
->i_mount
) &&
2659 ip
->i_d
.di_gid
!= be32_to_cpu(gdqp
->q_core
.d_id
))) {
2662 ASSERT(ip
->i_gdquot
);
2663 unresgdq
= ip
->i_gdquot
;
2668 if ((error
= xfs_trans_reserve_quota_bydquots(tp
, ip
->i_mount
,
2669 delblksudq
, delblksgdq
, ip
->i_d
.di_nblocks
, 1,
2670 flags
| blkflags
| prjflags
)))
2674 * Do the delayed blks reservations/unreservations now. Since, these
2675 * are done without the help of a transaction, if a reservation fails
2676 * its previous reservations won't be automatically undone by trans
2677 * code. So, we have to do it manually here.
2681 * Do the reservations first. Unreservation can't fail.
2683 ASSERT(delblksudq
|| delblksgdq
);
2684 ASSERT(unresudq
|| unresgdq
);
2685 if ((error
= xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
2686 delblksudq
, delblksgdq
, (xfs_qcnt_t
)delblks
, 0,
2687 flags
| blkflags
| prjflags
)))
2689 xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
2690 unresudq
, unresgdq
, -((xfs_qcnt_t
)delblks
), 0,
2698 xfs_qm_vop_rename_dqattach(
2699 xfs_inode_t
**i_tab
)
2707 if (! XFS_IS_QUOTA_ON(ip
->i_mount
))
2710 if (XFS_NOT_DQATTACHED(ip
->i_mount
, ip
)) {
2711 error
= xfs_qm_dqattach(ip
, 0);
2715 for (i
= 1; (i
< 4 && i_tab
[i
]); i
++) {
2717 * Watch out for duplicate entries in the table.
2719 if ((ip
= i_tab
[i
]) != i_tab
[i
-1]) {
2720 if (XFS_NOT_DQATTACHED(ip
->i_mount
, ip
)) {
2721 error
= xfs_qm_dqattach(ip
, 0);
2731 xfs_qm_vop_dqattach_and_dqmod_newinode(
2737 if (!XFS_IS_QUOTA_ON(tp
->t_mountp
))
2740 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
2741 ASSERT(XFS_IS_QUOTA_RUNNING(tp
->t_mountp
));
2747 ASSERT(ip
->i_udquot
== NULL
);
2748 ip
->i_udquot
= udqp
;
2749 ASSERT(XFS_IS_UQUOTA_ON(tp
->t_mountp
));
2750 ASSERT(ip
->i_d
.di_uid
== be32_to_cpu(udqp
->q_core
.d_id
));
2751 xfs_trans_mod_dquot(tp
, udqp
, XFS_TRANS_DQ_ICOUNT
, 1);
2757 ASSERT(ip
->i_gdquot
== NULL
);
2758 ip
->i_gdquot
= gdqp
;
2759 ASSERT(XFS_IS_OQUOTA_ON(tp
->t_mountp
));
2760 ASSERT((XFS_IS_GQUOTA_ON(tp
->t_mountp
) ?
2761 ip
->i_d
.di_gid
: ip
->i_d
.di_projid
) ==
2762 be32_to_cpu(gdqp
->q_core
.d_id
));
2763 xfs_trans_mod_dquot(tp
, gdqp
, XFS_TRANS_DQ_ICOUNT
, 1);
2767 /* ------------- list stuff -----------------*/
2769 xfs_qm_freelist_init(xfs_frlist_t
*ql
)
2771 ql
->qh_next
= ql
->qh_prev
= (xfs_dquot_t
*) ql
;
2772 mutex_init(&ql
->qh_lock
);
2778 xfs_qm_freelist_destroy(xfs_frlist_t
*ql
)
2780 xfs_dquot_t
*dqp
, *nextdqp
;
2782 mutex_lock(&ql
->qh_lock
);
2783 for (dqp
= ql
->qh_next
;
2784 dqp
!= (xfs_dquot_t
*)ql
; ) {
2786 nextdqp
= dqp
->dq_flnext
;
2788 cmn_err(CE_DEBUG
, "FREELIST destroy 0x%p", dqp
);
2790 XQM_FREELIST_REMOVE(dqp
);
2792 xfs_qm_dqdestroy(dqp
);
2795 mutex_unlock(&ql
->qh_lock
);
2796 mutex_destroy(&ql
->qh_lock
);
2798 ASSERT(ql
->qh_nelems
== 0);
2802 xfs_qm_freelist_insert(xfs_frlist_t
*ql
, xfs_dquot_t
*dq
)
2804 dq
->dq_flnext
= ql
->qh_next
;
2805 dq
->dq_flprev
= (xfs_dquot_t
*)ql
;
2807 dq
->dq_flnext
->dq_flprev
= dq
;
2808 xfs_Gqm
->qm_dqfreelist
.qh_nelems
++;
2809 xfs_Gqm
->qm_dqfreelist
.qh_version
++;
2813 xfs_qm_freelist_unlink(xfs_dquot_t
*dq
)
2815 xfs_dquot_t
*next
= dq
->dq_flnext
;
2816 xfs_dquot_t
*prev
= dq
->dq_flprev
;
2818 next
->dq_flprev
= prev
;
2819 prev
->dq_flnext
= next
;
2820 dq
->dq_flnext
= dq
->dq_flprev
= dq
;
2821 xfs_Gqm
->qm_dqfreelist
.qh_nelems
--;
2822 xfs_Gqm
->qm_dqfreelist
.qh_version
++;
2826 xfs_qm_freelist_append(xfs_frlist_t
*ql
, xfs_dquot_t
*dq
)
2828 xfs_qm_freelist_insert((xfs_frlist_t
*)ql
->qh_prev
, dq
);
2832 xfs_qm_dqhashlock_nowait(
2837 locked
= mutex_trylock(&((dqp
)->q_hash
->qh_lock
));
2842 xfs_qm_freelist_lock_nowait(
2847 locked
= mutex_trylock(&(xqm
->qm_dqfreelist
.qh_lock
));
2852 xfs_qm_mplist_nowait(
2857 ASSERT(mp
->m_quotainfo
);
2858 locked
= mutex_trylock(&(XFS_QI_MPLLOCK(mp
)));