2 * Copyright (c) 2000-2002 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
38 #include "xfs_trans.h"
43 #include "xfs_alloc.h"
44 #include "xfs_dmapi.h"
45 #include "xfs_quota.h"
46 #include "xfs_mount.h"
47 #include "xfs_alloc_btree.h"
48 #include "xfs_bmap_btree.h"
49 #include "xfs_ialloc_btree.h"
50 #include "xfs_btree.h"
51 #include "xfs_ialloc.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode.h"
59 #include "xfs_rtalloc.h"
60 #include "xfs_error.h"
61 #include "xfs_itable.h"
67 #include "xfs_buf_item.h"
68 #include "xfs_trans_space.h"
69 #include "xfs_utils.h"
74 * The global quota manager. There is only one of these for the entire
75 * system, _not_ one per file system. XQM keeps track of the overall
76 * quota functionality, including maintaining the freelist and hash
80 struct xfs_qm
*xfs_Gqm
;
82 kmem_zone_t
*qm_dqzone
;
83 kmem_zone_t
*qm_dqtrxzone
;
84 kmem_shaker_t xfs_qm_shaker
;
86 STATIC
void xfs_qm_list_init(xfs_dqlist_t
*, char *, int);
87 STATIC
void xfs_qm_list_destroy(xfs_dqlist_t
*);
88 STATIC
int xfs_qm_quotacheck(xfs_mount_t
*);
90 STATIC
int xfs_qm_init_quotainos(xfs_mount_t
*);
91 STATIC
int xfs_qm_shake(int, unsigned int);
94 extern mutex_t qcheck_lock
;
98 #define XQM_LIST_PRINT(l, NXT, title) \
100 xfs_dquot_t *dqp; int i = 0; \
101 cmn_err(CE_DEBUG, "%s (#%d)", title, (int) (l)->qh_nelems); \
102 for (dqp = (l)->qh_next; dqp != NULL; dqp = dqp->NXT) { \
103 cmn_err(CE_DEBUG, " %d. \"%d (%s)\" " \
104 "bcnt = %d, icnt = %d, refs = %d", \
105 ++i, (int) INT_GET(dqp->q_core.d_id, ARCH_CONVERT), \
106 DQFLAGTO_TYPESTR(dqp), \
107 (int) INT_GET(dqp->q_core.d_bcount, ARCH_CONVERT), \
108 (int) INT_GET(dqp->q_core.d_icount, ARCH_CONVERT), \
109 (int) dqp->q_nrefs); } \
112 #define XQM_LIST_PRINT(l, NXT, title) do { } while (0)
116 * Initialize the XQM structure.
117 * Note that there is not one quota manager per file system.
119 STATIC
struct xfs_qm
*
125 xqm
= kmem_zalloc(sizeof(xfs_qm_t
), KM_SLEEP
);
129 * Initialize the dquot hash tables.
131 hsize
= (DQUOT_HASH_HEURISTIC
< XFS_QM_NCSIZE_THRESHOLD
) ?
132 XFS_QM_HASHSIZE_LOW
: XFS_QM_HASHSIZE_HIGH
;
133 xqm
->qm_dqhashmask
= hsize
- 1;
135 xqm
->qm_usr_dqhtable
= (xfs_dqhash_t
*)kmem_zalloc(hsize
*
136 sizeof(xfs_dqhash_t
),
138 xqm
->qm_grp_dqhtable
= (xfs_dqhash_t
*)kmem_zalloc(hsize
*
139 sizeof(xfs_dqhash_t
),
141 ASSERT(xqm
->qm_usr_dqhtable
!= NULL
);
142 ASSERT(xqm
->qm_grp_dqhtable
!= NULL
);
144 for (i
= 0; i
< hsize
; i
++) {
145 xfs_qm_list_init(&(xqm
->qm_usr_dqhtable
[i
]), "uxdqh", i
);
146 xfs_qm_list_init(&(xqm
->qm_grp_dqhtable
[i
]), "gxdqh", i
);
150 * Freelist of all dquots of all file systems
152 xfs_qm_freelist_init(&(xqm
->qm_dqfreelist
));
155 * dquot zone. we register our own low-memory callback.
158 xqm
->qm_dqzone
= kmem_zone_init(sizeof(xfs_dquot_t
),
160 qm_dqzone
= xqm
->qm_dqzone
;
162 xqm
->qm_dqzone
= qm_dqzone
;
164 xfs_qm_shaker
= kmem_shake_register(xfs_qm_shake
);
167 * The t_dqinfo portion of transactions.
170 xqm
->qm_dqtrxzone
= kmem_zone_init(sizeof(xfs_dquot_acct_t
),
172 qm_dqtrxzone
= xqm
->qm_dqtrxzone
;
174 xqm
->qm_dqtrxzone
= qm_dqtrxzone
;
176 atomic_set(&xqm
->qm_totaldquots
, 0);
177 xqm
->qm_dqfree_ratio
= XFS_QM_DQFREE_RATIO
;
180 mutex_init(&qcheck_lock
, MUTEX_DEFAULT
, "qchk");
186 * Destroy the global quota manager when its reference count goes to zero.
195 ASSERT(xqm
->qm_nrefs
== 0);
196 kmem_shake_deregister(xfs_qm_shaker
);
197 hsize
= xqm
->qm_dqhashmask
+ 1;
198 for (i
= 0; i
< hsize
; i
++) {
199 xfs_qm_list_destroy(&(xqm
->qm_usr_dqhtable
[i
]));
200 xfs_qm_list_destroy(&(xqm
->qm_grp_dqhtable
[i
]));
202 kmem_free(xqm
->qm_usr_dqhtable
, hsize
* sizeof(xfs_dqhash_t
));
203 kmem_free(xqm
->qm_grp_dqhtable
, hsize
* sizeof(xfs_dqhash_t
));
204 xqm
->qm_usr_dqhtable
= NULL
;
205 xqm
->qm_grp_dqhtable
= NULL
;
206 xqm
->qm_dqhashmask
= 0;
207 xfs_qm_freelist_destroy(&(xqm
->qm_dqfreelist
));
209 mutex_destroy(&qcheck_lock
);
211 kmem_free(xqm
, sizeof(xfs_qm_t
));
215 * Called at mount time to let XQM know that another file system is
216 * starting quotas. This isn't crucial information as the individual mount
217 * structures are pretty independent, but it helps the XQM keep a
218 * global view of what's going on.
222 xfs_qm_hold_quotafs_ref(
223 struct xfs_mount
*mp
)
226 * Need to lock the xfs_Gqm structure for things like this. For example,
227 * the structure could disappear between the entry to this routine and
228 * a HOLD operation if not locked.
230 XFS_QM_LOCK(xfs_Gqm
);
233 xfs_Gqm
= xfs_Gqm_init();
235 * We can keep a list of all filesystems with quotas mounted for
236 * debugging and statistical purposes, but ...
237 * Just take a reference and get out.
239 XFS_QM_HOLD(xfs_Gqm
);
240 XFS_QM_UNLOCK(xfs_Gqm
);
247 * Release the reference that a filesystem took at mount time,
248 * so that we know when we need to destroy the entire quota manager.
252 xfs_qm_rele_quotafs_ref(
253 struct xfs_mount
*mp
)
255 xfs_dquot_t
*dqp
, *nextdqp
;
258 ASSERT(xfs_Gqm
->qm_nrefs
> 0);
261 * Go thru the freelist and destroy all inactive dquots.
263 xfs_qm_freelist_lock(xfs_Gqm
);
265 for (dqp
= xfs_Gqm
->qm_dqfreelist
.qh_next
;
266 dqp
!= (xfs_dquot_t
*)&(xfs_Gqm
->qm_dqfreelist
); ) {
268 nextdqp
= dqp
->dq_flnext
;
269 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
270 ASSERT(dqp
->q_mount
== NULL
);
271 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
272 ASSERT(dqp
->HL_PREVP
== NULL
);
273 ASSERT(dqp
->MPL_PREVP
== NULL
);
274 XQM_FREELIST_REMOVE(dqp
);
276 xfs_qm_dqdestroy(dqp
);
282 xfs_qm_freelist_unlock(xfs_Gqm
);
285 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
288 XFS_QM_LOCK(xfs_Gqm
);
289 XFS_QM_RELE(xfs_Gqm
);
290 if (xfs_Gqm
->qm_nrefs
== 0) {
291 xfs_qm_destroy(xfs_Gqm
);
294 XFS_QM_UNLOCK(xfs_Gqm
);
298 * This is called at mount time from xfs_mountfs to initialize the quotainfo
299 * structure and start the global quotamanager (xfs_Gqm) if it hasn't done
300 * so already. Note that the superblock has not been read in yet.
303 xfs_qm_mount_quotainit(
308 * User or group quotas has to be on.
310 ASSERT(flags
& (XFSMNT_UQUOTA
| XFSMNT_GQUOTA
));
313 * Initialize the flags in the mount structure. From this point
314 * onwards we look at m_qflags to figure out if quotas's ON/OFF, etc.
315 * Note that we enforce nothing if accounting is off.
316 * ie. XFSMNT_*QUOTA must be ON for XFSMNT_*QUOTAENF.
317 * It isn't necessary to take the quotaoff lock to do this; this is
320 if (flags
& XFSMNT_UQUOTA
) {
321 mp
->m_qflags
|= (XFS_UQUOTA_ACCT
| XFS_UQUOTA_ACTIVE
);
322 if (flags
& XFSMNT_UQUOTAENF
)
323 mp
->m_qflags
|= XFS_UQUOTA_ENFD
;
325 if (flags
& XFSMNT_GQUOTA
) {
326 mp
->m_qflags
|= (XFS_GQUOTA_ACCT
| XFS_GQUOTA_ACTIVE
);
327 if (flags
& XFSMNT_GQUOTAENF
)
328 mp
->m_qflags
|= XFS_GQUOTA_ENFD
;
333 * Just destroy the quotainfo structure.
336 xfs_qm_unmount_quotadestroy(
340 xfs_qm_destroy_quotainfo(mp
);
345 * This is called from xfs_mountfs to start quotas and initialize all
346 * necessary data structures like quotainfo. This is also responsible for
347 * running a quotacheck as necessary. We are guaranteed that the superblock
348 * is consistently read in at this point.
359 * If a file system had quotas running earlier, but decided to
360 * mount without -o quota/uquota/gquota options, revoke the
361 * quotachecked license, and bail out.
363 if (! XFS_IS_QUOTA_ON(mp
) &&
364 (mp
->m_sb
.sb_qflags
& (XFS_UQUOTA_ACCT
|XFS_GQUOTA_ACCT
))) {
370 * If quotas on realtime volumes is not supported, we disable
371 * quotas immediately.
373 if (mp
->m_sb
.sb_rextents
) {
375 "Cannot turn on quotas for realtime filesystem %s",
381 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
382 cmn_err(CE_NOTE
, "Attempting to turn on disk quotas.");
385 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
387 * Allocate the quotainfo structure inside the mount struct, and
388 * create quotainode(s), and change/rev superblock if necessary.
390 if ((error
= xfs_qm_init_quotainfo(mp
))) {
392 * We must turn off quotas.
394 ASSERT(mp
->m_quotainfo
== NULL
);
399 * If any of the quotas are not consistent, do a quotacheck.
401 if (XFS_QM_NEED_QUOTACHECK(mp
)) {
403 cmn_err(CE_NOTE
, "Doing a quotacheck. Please wait.");
405 if ((error
= xfs_qm_quotacheck(mp
))) {
406 cmn_err(CE_WARN
, "Quotacheck unsuccessful (Error %d): "
410 * We must turn off quotas.
412 ASSERT(mp
->m_quotainfo
!= NULL
);
413 ASSERT(xfs_Gqm
!= NULL
);
414 xfs_qm_destroy_quotainfo(mp
);
419 cmn_err(CE_NOTE
, "Done quotacheck.");
424 * We actually don't have to acquire the SB_LOCK at all.
425 * This can only be called from mount, and that's single threaded. XXX
428 sbf
= mp
->m_sb
.sb_qflags
;
429 mp
->m_sb
.sb_qflags
= mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
;
430 XFS_SB_UNLOCK(mp
, s
);
432 if (sbf
!= (mp
->m_qflags
& XFS_MOUNT_QUOTA_ALL
)) {
433 if (xfs_qm_write_sb_changes(mp
, XFS_SB_QFLAGS
)) {
435 * We could only have been turning quotas off.
436 * We aren't in very good shape actually because
437 * the incore structures are convinced that quotas are
438 * off, but the on disk superblock doesn't know that !
440 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp
)));
441 xfs_fs_cmn_err(CE_ALERT
, mp
,
442 "XFS mount_quotas: Superblock update failed!");
447 xfs_fs_cmn_err(CE_WARN
, mp
,
448 "Failed to initialize disk quotas.");
450 return XFS_ERROR(error
);
454 * Called from the vfsops layer.
457 xfs_qm_unmount_quotas(
460 xfs_inode_t
*uqp
, *gqp
;
464 * Release the dquots that root inode, et al might be holding,
465 * before we flush quotas and blow away the quotainfo structure.
467 ASSERT(mp
->m_rootip
);
468 xfs_qm_dqdetach(mp
->m_rootip
);
470 xfs_qm_dqdetach(mp
->m_rbmip
);
472 xfs_qm_dqdetach(mp
->m_rsumip
);
475 * Flush out the quota inodes.
478 if (mp
->m_quotainfo
) {
479 if ((uqp
= mp
->m_quotainfo
->qi_uquotaip
) != NULL
) {
480 xfs_ilock(uqp
, XFS_ILOCK_EXCL
);
482 error
= xfs_iflush(uqp
, XFS_IFLUSH_SYNC
);
483 xfs_iunlock(uqp
, XFS_ILOCK_EXCL
);
484 if (unlikely(error
== EFSCORRUPTED
)) {
485 XFS_ERROR_REPORT("xfs_qm_unmount_quotas(1)",
486 XFS_ERRLEVEL_LOW
, mp
);
490 if ((gqp
= mp
->m_quotainfo
->qi_gquotaip
) != NULL
) {
491 xfs_ilock(gqp
, XFS_ILOCK_EXCL
);
493 error
= xfs_iflush(gqp
, XFS_IFLUSH_SYNC
);
494 xfs_iunlock(gqp
, XFS_ILOCK_EXCL
);
495 if (unlikely(error
== EFSCORRUPTED
)) {
496 XFS_ERROR_REPORT("xfs_qm_unmount_quotas(2)",
497 XFS_ERRLEVEL_LOW
, mp
);
503 XFS_PURGE_INODE(uqp
);
504 mp
->m_quotainfo
->qi_uquotaip
= NULL
;
507 XFS_PURGE_INODE(gqp
);
508 mp
->m_quotainfo
->qi_gquotaip
= NULL
;
511 return XFS_ERROR(error
);
515 * Flush all dquots of the given file system to disk. The dquots are
516 * _not_ purged from memory here, just their data written to disk.
528 if (mp
->m_quotainfo
== NULL
)
532 xfs_qm_mplist_lock(mp
);
533 FOREACH_DQUOT_IN_MP(dqp
, mp
) {
535 if (! XFS_DQ_IS_DIRTY(dqp
)) {
539 xfs_dqtrace_entry(dqp
, "FLUSHALL: DQDIRTY");
540 /* XXX a sentinel would be better */
541 recl
= XFS_QI_MPLRECLAIMS(mp
);
542 if (! xfs_qm_dqflock_nowait(dqp
)) {
544 * If we can't grab the flush lock then check
545 * to see if the dquot has been flushed delayed
546 * write. If so, grab its buffer and send it
547 * out immediately. We'll be able to acquire
548 * the flush lock when the I/O completes.
550 xfs_qm_dqflock_pushbuf_wait(dqp
);
553 * Let go of the mplist lock. We don't want to hold it
554 * across a disk write.
556 xfs_qm_mplist_unlock(mp
);
557 error
= xfs_qm_dqflush(dqp
, flags
);
562 xfs_qm_mplist_lock(mp
);
563 if (recl
!= XFS_QI_MPLRECLAIMS(mp
)) {
564 xfs_qm_mplist_unlock(mp
);
565 /* XXX restart limit */
570 xfs_qm_mplist_unlock(mp
);
575 * Release the group dquot pointers the user dquots may be
576 * carrying around as a hint. mplist is locked on entry and exit.
579 xfs_qm_detach_gdquots(
582 xfs_dquot_t
*dqp
, *gdqp
;
586 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp
));
587 dqp
= XFS_QI_MPLNEXT(mp
);
590 if ((gdqp
= dqp
->q_gdquot
)) {
592 dqp
->q_gdquot
= NULL
;
598 * Can't hold the mplist lock across a dqput.
599 * XXXmust convert to marker based iterations here.
601 nrecl
= XFS_QI_MPLRECLAIMS(mp
);
602 xfs_qm_mplist_unlock(mp
);
605 xfs_qm_mplist_lock(mp
);
606 if (nrecl
!= XFS_QI_MPLRECLAIMS(mp
))
614 * Go through all the incore dquots of this file system and take them
615 * off the mplist and hashlist, if the dquot type matches the dqtype
616 * parameter. This is used when turning off quota accounting for
617 * users and/or groups, as well as when the filesystem is unmounting.
622 uint flags
) /* QUOTAOFF/UMOUNTING/UQUOTA/GQUOTA */
627 xfs_dquot_t
*nextdqp
;
630 if (mp
->m_quotainfo
== NULL
)
633 dqtype
= (flags
& XFS_QMOPT_UQUOTA
) ? XFS_DQ_USER
: 0;
634 dqtype
|= (flags
& XFS_QMOPT_GQUOTA
) ? XFS_DQ_GROUP
: 0;
636 xfs_qm_mplist_lock(mp
);
639 * In the first pass through all incore dquots of this filesystem,
640 * we release the group dquot pointers the user dquots may be
641 * carrying around as a hint. We need to do this irrespective of
642 * what's being turned off.
644 xfs_qm_detach_gdquots(mp
);
648 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp
));
650 * Try to get rid of all of the unwanted dquots. The idea is to
651 * get them off mplist and hashlist, but leave them on freelist.
653 dqp
= XFS_QI_MPLNEXT(mp
);
656 * It's OK to look at the type without taking dqlock here.
657 * We're holding the mplist lock here, and that's needed for
660 if ((dqp
->dq_flags
& dqtype
) == 0) {
665 if (! xfs_qm_dqhashlock_nowait(dqp
)) {
666 nrecl
= XFS_QI_MPLRECLAIMS(mp
);
667 xfs_qm_mplist_unlock(mp
);
668 XFS_DQ_HASH_LOCK(dqp
->q_hash
);
669 xfs_qm_mplist_lock(mp
);
672 * XXXTheoretically, we can get into a very long
673 * ping pong game here.
674 * No one can be adding dquots to the mplist at
675 * this point, but somebody might be taking things off.
677 if (nrecl
!= XFS_QI_MPLRECLAIMS(mp
)) {
678 XFS_DQ_HASH_UNLOCK(dqp
->q_hash
);
684 * Take the dquot off the mplist and hashlist. It may remain on
685 * freelist in INACTIVE state.
687 nextdqp
= dqp
->MPL_NEXT
;
688 nmisses
+= xfs_qm_dqpurge(dqp
, flags
);
691 xfs_qm_mplist_unlock(mp
);
703 * Purge the dquot cache.
704 * None of the dquots should really be busy at this point.
706 if (mp
->m_quotainfo
) {
707 while ((ndquots
= xfs_qm_dqpurge_int(mp
, flags
))) {
721 xfs_dquot_t
*udqhint
, /* hint */
722 xfs_dquot_t
**IO_idqpp
)
727 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
730 * See if we already have it in the inode itself. IO_idqpp is
731 * &i_udquot or &i_gdquot. This made the code look weird, but
732 * made the logic a lot simpler.
734 if ((dqp
= *IO_idqpp
)) {
737 xfs_dqtrace_entry(dqp
, "DQATTACH: found in ip");
742 * udqhint is the i_udquot field in inode, and is non-NULL only
743 * when the type arg is XFS_DQ_GROUP. Its purpose is to save a
744 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
747 ASSERT(!udqhint
|| type
== XFS_DQ_GROUP
);
748 if (udqhint
&& !dolock
)
752 * No need to take dqlock to look at the id.
753 * The ID can't change until it gets reclaimed, and it won't
754 * be reclaimed as long as we have a ref from inode and we hold
758 (dqp
= udqhint
->q_gdquot
) &&
759 (INT_GET(dqp
->q_core
.d_id
, ARCH_CONVERT
) == id
)) {
760 ASSERT(XFS_DQ_IS_LOCKED(udqhint
));
763 ASSERT(*IO_idqpp
== NULL
);
767 xfs_dqunlock(udqhint
);
772 * We can't hold a dquot lock when we call the dqget code.
773 * We'll deadlock in no time, because of (not conforming to)
774 * lock ordering - the inodelock comes before any dquot lock,
775 * and we may drop and reacquire the ilock in xfs_qm_dqget().
778 xfs_dqunlock(udqhint
);
780 * Find the dquot from somewhere. This bumps the
781 * reference count of dquot and returns it locked.
782 * This can return ENOENT if dquot didn't exist on
783 * disk and we didn't ask it to allocate;
784 * ESRCH if quotas got turned off suddenly.
786 if ((error
= xfs_qm_dqget(ip
->i_mount
, ip
, id
, type
,
787 doalloc
|XFS_QMOPT_DOWARN
, &dqp
))) {
788 if (udqhint
&& dolock
)
793 xfs_dqtrace_entry(dqp
, "DQATTACH: found by dqget");
795 * dqget may have dropped and re-acquired the ilock, but it guarantees
796 * that the dquot returned is the one that should go in the inode.
800 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
810 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
811 if (! xfs_qm_dqlock_nowait(udqhint
)) {
820 ASSERT(XFS_DQ_IS_LOCKED(udqhint
));
824 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
832 * Given a udquot and gdquot, attach a ptr to the group dquot in the
833 * udquot as a hint for future lookups. The idea sounds simple, but the
834 * execution isn't, because the udquot might have a group dquot attached
835 * already and getting rid of that gets us into lock ordering contraints.
836 * The process is complicated more by the fact that the dquots may or may not
837 * be locked on entry.
840 xfs_qm_dqattach_grouphint(
849 ASSERT(XFS_DQ_IS_LOCKED(udq
));
850 ASSERT(XFS_DQ_IS_LOCKED(gdq
));
856 if ((tmp
= udq
->q_gdquot
)) {
863 udq
->q_gdquot
= NULL
;
865 * We can't keep any dqlocks when calling dqrele,
866 * because the freelist lock comes before dqlocks.
872 * we took a hard reference once upon a time in dqget,
873 * so give it back when the udquot no longer points at it
874 * dqput() does the unlocking of the dquot.
882 ASSERT(XFS_DQ_IS_LOCKED(udq
));
888 ASSERT(XFS_DQ_IS_LOCKED(udq
));
889 ASSERT(XFS_DQ_IS_LOCKED(gdq
));
891 * Somebody could have attached a gdquot here,
892 * when we dropped the uqlock. If so, just do nothing.
894 if (udq
->q_gdquot
== NULL
) {
906 * Given a locked inode, attach dquot(s) to it, taking UQUOTAON / GQUOTAON
908 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
909 * If XFS_QMOPT_DQLOCK, the dquot(s) will be returned locked. This option pretty
910 * much made this code a complete mess, but it has been pretty useful.
911 * If XFS_QMOPT_ILOCKED, then inode sent is already locked EXCL.
912 * Inode may get unlocked and relocked in here, and the caller must deal with
920 xfs_mount_t
*mp
= ip
->i_mount
;
924 if ((! XFS_IS_QUOTA_ON(mp
)) ||
925 (! XFS_NOT_DQATTACHED(mp
, ip
)) ||
926 (ip
->i_ino
== mp
->m_sb
.sb_uquotino
) ||
927 (ip
->i_ino
== mp
->m_sb
.sb_gquotino
))
930 ASSERT((flags
& XFS_QMOPT_ILOCKED
) == 0 ||
931 XFS_ISLOCKED_INODE_EXCL(ip
));
933 if (! (flags
& XFS_QMOPT_ILOCKED
))
934 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
936 if (XFS_IS_UQUOTA_ON(mp
)) {
937 error
= xfs_qm_dqattach_one(ip
, ip
->i_d
.di_uid
, XFS_DQ_USER
,
938 flags
& XFS_QMOPT_DQALLOC
,
939 flags
& XFS_QMOPT_DQLOCK
,
940 NULL
, &ip
->i_udquot
);
945 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
946 if (XFS_IS_GQUOTA_ON(mp
)) {
947 error
= xfs_qm_dqattach_one(ip
, ip
->i_d
.di_gid
, XFS_DQ_GROUP
,
948 flags
& XFS_QMOPT_DQALLOC
,
949 flags
& XFS_QMOPT_DQLOCK
,
950 ip
->i_udquot
, &ip
->i_gdquot
);
952 * Don't worry about the udquot that we may have
953 * attached above. It'll get detached, if not already.
961 * Attach this group quota to the user quota as a hint.
962 * This WON'T, in general, result in a thrash.
965 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
966 ASSERT(ip
->i_udquot
);
967 ASSERT(ip
->i_gdquot
);
970 * We may or may not have the i_udquot locked at this point,
971 * but this check is OK since we don't depend on the i_gdquot to
972 * be accurate 100% all the time. It is just a hint, and this
973 * will succeed in general.
975 if (ip
->i_udquot
->q_gdquot
== ip
->i_gdquot
)
978 * Attach i_gdquot to the gdquot hint inside the i_udquot.
980 xfs_qm_dqattach_grouphint(ip
->i_udquot
, ip
->i_gdquot
,
981 flags
& XFS_QMOPT_DQLOCK
);
989 if (flags
& XFS_QMOPT_DQLOCK
)
990 ASSERT(XFS_DQ_IS_LOCKED(ip
->i_udquot
));
993 if (flags
& XFS_QMOPT_DQLOCK
)
994 ASSERT(XFS_DQ_IS_LOCKED(ip
->i_gdquot
));
996 if (XFS_IS_UQUOTA_ON(mp
))
997 ASSERT(ip
->i_udquot
);
998 if (XFS_IS_GQUOTA_ON(mp
))
999 ASSERT(ip
->i_gdquot
);
1003 if (! (flags
& XFS_QMOPT_ILOCKED
))
1004 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1008 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
1014 * Release dquots (and their references) if any.
1015 * The inode should be locked EXCL except when this's called by
1022 if (!(ip
->i_udquot
|| ip
->i_gdquot
))
1025 ASSERT(ip
->i_ino
!= ip
->i_mount
->m_sb
.sb_uquotino
);
1026 ASSERT(ip
->i_ino
!= ip
->i_mount
->m_sb
.sb_gquotino
);
1028 xfs_dqtrace_entry_ino(ip
->i_udquot
, "DQDETTACH", ip
);
1030 xfs_qm_dqrele(ip
->i_udquot
);
1031 ip
->i_udquot
= NULL
;
1034 xfs_qm_dqrele(ip
->i_gdquot
);
1035 ip
->i_gdquot
= NULL
;
1040 * This is called by VFS_SYNC and flags arg determines the caller,
1041 * and its motives, as done in xfs_sync.
1043 * vfs_sync: SYNC_FSDATA|SYNC_ATTR|SYNC_BDFLUSH 0x31
1044 * syscall sync: SYNC_FSDATA|SYNC_ATTR|SYNC_DELWRI 0x25
1045 * umountroot : SYNC_WAIT | SYNC_CLOSE | SYNC_ATTR | SYNC_FSDATA
1061 * We won't block unless we are asked to.
1063 nowait
= (boolean_t
)(flags
& SYNC_BDFLUSH
|| (flags
& SYNC_WAIT
) == 0);
1066 xfs_qm_mplist_lock(mp
);
1068 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
1069 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
1070 * when we have the mplist lock, we know that dquots will be consistent
1071 * as long as we have it locked.
1073 if (! XFS_IS_QUOTA_ON(mp
)) {
1074 xfs_qm_mplist_unlock(mp
);
1077 FOREACH_DQUOT_IN_MP(dqp
, mp
) {
1079 * If this is vfs_sync calling, then skip the dquots that
1080 * don't 'seem' to be dirty. ie. don't acquire dqlock.
1081 * This is very similar to what xfs_sync does with inodes.
1083 if (flags
& SYNC_BDFLUSH
) {
1084 if (! XFS_DQ_IS_DIRTY(dqp
))
1090 * Try to acquire the dquot lock. We are NOT out of
1091 * lock order, but we just don't want to wait for this
1092 * lock, unless somebody wanted us to.
1094 if (! xfs_qm_dqlock_nowait(dqp
))
1101 * Now, find out for sure if this dquot is dirty or not.
1103 if (! XFS_DQ_IS_DIRTY(dqp
)) {
1108 /* XXX a sentinel would be better */
1109 recl
= XFS_QI_MPLRECLAIMS(mp
);
1110 if (! xfs_qm_dqflock_nowait(dqp
)) {
1116 * If we can't grab the flush lock then if the caller
1117 * really wanted us to give this our best shot,
1118 * see if we can give a push to the buffer before we wait
1119 * on the flush lock. At this point, we know that
1120 * eventhough the dquot is being flushed,
1121 * it has (new) dirty data.
1123 xfs_qm_dqflock_pushbuf_wait(dqp
);
1126 * Let go of the mplist lock. We don't want to hold it
1127 * across a disk write
1129 flush_flags
= (nowait
) ? XFS_QMOPT_DELWRI
: XFS_QMOPT_SYNC
;
1130 xfs_qm_mplist_unlock(mp
);
1131 xfs_dqtrace_entry(dqp
, "XQM_SYNC: DQFLUSH");
1132 error
= xfs_qm_dqflush(dqp
, flush_flags
);
1134 if (error
&& XFS_FORCED_SHUTDOWN(mp
))
1135 return(0); /* Need to prevent umount failure */
1139 xfs_qm_mplist_lock(mp
);
1140 if (recl
!= XFS_QI_MPLRECLAIMS(mp
)) {
1141 if (++restarts
>= XFS_QM_SYNC_MAX_RESTARTS
)
1144 xfs_qm_mplist_unlock(mp
);
1149 xfs_qm_mplist_unlock(mp
);
1155 * This initializes all the quota information that's kept in the
1159 xfs_qm_init_quotainfo(
1162 xfs_quotainfo_t
*qinf
;
1166 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1169 * Tell XQM that we exist as soon as possible.
1171 if ((error
= xfs_qm_hold_quotafs_ref(mp
))) {
1175 qinf
= mp
->m_quotainfo
= kmem_zalloc(sizeof(xfs_quotainfo_t
), KM_SLEEP
);
1178 * See if quotainodes are setup, and if not, allocate them,
1179 * and change the superblock accordingly.
1181 if ((error
= xfs_qm_init_quotainos(mp
))) {
1182 kmem_free(qinf
, sizeof(xfs_quotainfo_t
));
1183 mp
->m_quotainfo
= NULL
;
1187 spinlock_init(&qinf
->qi_pinlock
, "xfs_qinf_pin");
1188 xfs_qm_list_init(&qinf
->qi_dqlist
, "mpdqlist", 0);
1189 qinf
->qi_dqreclaims
= 0;
1191 /* mutex used to serialize quotaoffs */
1192 mutex_init(&qinf
->qi_quotaofflock
, MUTEX_DEFAULT
, "qoff");
1194 /* Precalc some constants */
1195 qinf
->qi_dqchunklen
= XFS_FSB_TO_BB(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
1196 ASSERT(qinf
->qi_dqchunklen
);
1197 qinf
->qi_dqperchunk
= BBTOB(qinf
->qi_dqchunklen
);
1198 do_div(qinf
->qi_dqperchunk
, sizeof(xfs_dqblk_t
));
1200 mp
->m_qflags
|= (mp
->m_sb
.sb_qflags
& XFS_ALL_QUOTA_CHKD
);
1203 * We try to get the limits from the superuser's limits fields.
1204 * This is quite hacky, but it is standard quota practice.
1205 * We look at the USR dquot with id == 0 first, but if user quotas
1206 * are not enabled we goto the GRP dquot with id == 0.
1207 * We don't really care to keep separate default limits for user
1208 * and group quotas, at least not at this point.
1210 error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)0,
1211 (XFS_IS_UQUOTA_RUNNING(mp
)) ?
1212 XFS_DQ_USER
: XFS_DQ_GROUP
,
1213 XFS_QMOPT_DQSUSER
|XFS_QMOPT_DOWARN
,
1216 xfs_disk_dquot_t
*ddqp
= &dqp
->q_core
;
1219 * The warnings and timers set the grace period given to
1220 * a user or group before he or she can not perform any
1221 * more writing. If it is zero, a default is used.
1223 qinf
->qi_btimelimit
=
1224 INT_GET(ddqp
->d_btimer
, ARCH_CONVERT
) ?
1225 INT_GET(ddqp
->d_btimer
, ARCH_CONVERT
) :
1227 qinf
->qi_itimelimit
=
1228 INT_GET(ddqp
->d_itimer
, ARCH_CONVERT
) ?
1229 INT_GET(ddqp
->d_itimer
, ARCH_CONVERT
) :
1231 qinf
->qi_rtbtimelimit
=
1232 INT_GET(ddqp
->d_rtbtimer
, ARCH_CONVERT
) ?
1233 INT_GET(ddqp
->d_rtbtimer
, ARCH_CONVERT
) :
1234 XFS_QM_RTBTIMELIMIT
;
1235 qinf
->qi_bwarnlimit
=
1236 INT_GET(ddqp
->d_bwarns
, ARCH_CONVERT
) ?
1237 INT_GET(ddqp
->d_bwarns
, ARCH_CONVERT
) :
1239 qinf
->qi_iwarnlimit
=
1240 INT_GET(ddqp
->d_iwarns
, ARCH_CONVERT
) ?
1241 INT_GET(ddqp
->d_iwarns
, ARCH_CONVERT
) :
1243 qinf
->qi_bhardlimit
=
1244 INT_GET(ddqp
->d_blk_hardlimit
, ARCH_CONVERT
);
1245 qinf
->qi_bsoftlimit
=
1246 INT_GET(ddqp
->d_blk_softlimit
, ARCH_CONVERT
);
1247 qinf
->qi_ihardlimit
=
1248 INT_GET(ddqp
->d_ino_hardlimit
, ARCH_CONVERT
);
1249 qinf
->qi_isoftlimit
=
1250 INT_GET(ddqp
->d_ino_softlimit
, ARCH_CONVERT
);
1251 qinf
->qi_rtbhardlimit
=
1252 INT_GET(ddqp
->d_rtb_hardlimit
, ARCH_CONVERT
);
1253 qinf
->qi_rtbsoftlimit
=
1254 INT_GET(ddqp
->d_rtb_softlimit
, ARCH_CONVERT
);
1257 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1258 * we don't want this dquot cached. We haven't done a
1259 * quotacheck yet, and quotacheck doesn't like incore dquots.
1261 xfs_qm_dqdestroy(dqp
);
1263 qinf
->qi_btimelimit
= XFS_QM_BTIMELIMIT
;
1264 qinf
->qi_itimelimit
= XFS_QM_ITIMELIMIT
;
1265 qinf
->qi_rtbtimelimit
= XFS_QM_RTBTIMELIMIT
;
1266 qinf
->qi_bwarnlimit
= XFS_QM_BWARNLIMIT
;
1267 qinf
->qi_iwarnlimit
= XFS_QM_IWARNLIMIT
;
1275 * Gets called when unmounting a filesystem or when all quotas get
1277 * This purges the quota inodes, destroys locks and frees itself.
1280 xfs_qm_destroy_quotainfo(
1283 xfs_quotainfo_t
*qi
;
1285 qi
= mp
->m_quotainfo
;
1287 ASSERT(xfs_Gqm
!= NULL
);
1290 * Release the reference that XQM kept, so that we know
1291 * when the XQM structure should be freed. We cannot assume
1292 * that xfs_Gqm is non-null after this point.
1294 xfs_qm_rele_quotafs_ref(mp
);
1296 spinlock_destroy(&qi
->qi_pinlock
);
1297 xfs_qm_list_destroy(&qi
->qi_dqlist
);
1299 if (qi
->qi_uquotaip
) {
1300 XFS_PURGE_INODE(qi
->qi_uquotaip
);
1301 qi
->qi_uquotaip
= NULL
; /* paranoia */
1303 if (qi
->qi_gquotaip
) {
1304 XFS_PURGE_INODE(qi
->qi_gquotaip
);
1305 qi
->qi_gquotaip
= NULL
;
1307 mutex_destroy(&qi
->qi_quotaofflock
);
1308 kmem_free(qi
, sizeof(xfs_quotainfo_t
));
1309 mp
->m_quotainfo
= NULL
;
1314 /* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1323 mutex_init(&list
->qh_lock
, MUTEX_DEFAULT
, str
);
1324 list
->qh_next
= NULL
;
1325 list
->qh_version
= 0;
1326 list
->qh_nelems
= 0;
1330 xfs_qm_list_destroy(
1333 mutex_destroy(&(list
->qh_lock
));
1338 * Stripped down version of dqattach. This doesn't attach, or even look at the
1339 * dquots attached to the inode. The rationale is that there won't be any
1340 * attached at the time this is called from quotacheck.
1343 xfs_qm_dqget_noattach(
1345 xfs_dquot_t
**O_udqpp
,
1346 xfs_dquot_t
**O_gdqpp
)
1350 xfs_dquot_t
*udqp
, *gdqp
;
1352 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
1357 if (XFS_IS_UQUOTA_ON(mp
)) {
1358 ASSERT(ip
->i_udquot
== NULL
);
1360 * We want the dquot allocated if it doesn't exist.
1362 if ((error
= xfs_qm_dqget(mp
, ip
, ip
->i_d
.di_uid
, XFS_DQ_USER
,
1363 XFS_QMOPT_DQALLOC
| XFS_QMOPT_DOWARN
,
1366 * Shouldn't be able to turn off quotas here.
1368 ASSERT(error
!= ESRCH
);
1369 ASSERT(error
!= ENOENT
);
1375 if (XFS_IS_GQUOTA_ON(mp
)) {
1376 ASSERT(ip
->i_gdquot
== NULL
);
1379 if ((error
= xfs_qm_dqget(mp
, ip
, ip
->i_d
.di_gid
, XFS_DQ_GROUP
,
1380 XFS_QMOPT_DQALLOC
|XFS_QMOPT_DOWARN
,
1383 xfs_qm_dqrele(udqp
);
1384 ASSERT(error
!= ESRCH
);
1385 ASSERT(error
!= ENOENT
);
1390 /* Reacquire the locks in the right order */
1392 if (! xfs_qm_dqlock_nowait(udqp
)) {
1404 if (udqp
) ASSERT(XFS_DQ_IS_LOCKED(udqp
));
1405 if (gdqp
) ASSERT(XFS_DQ_IS_LOCKED(gdqp
));
1411 * Create an inode and return with a reference already taken, but unlocked
1412 * This is how we create quota inodes
1427 tp
= xfs_trans_alloc(mp
,XFS_TRANS_QM_QINOCREATE
);
1428 if ((error
= xfs_trans_reserve(tp
,
1429 XFS_QM_QINOCREATE_SPACE_RES(mp
),
1430 XFS_CREATE_LOG_RES(mp
), 0,
1431 XFS_TRANS_PERM_LOG_RES
,
1432 XFS_CREATE_LOG_COUNT
))) {
1433 xfs_trans_cancel(tp
, 0);
1436 memset(&zerocr
, 0, sizeof(zerocr
));
1438 if ((error
= xfs_dir_ialloc(&tp
, mp
->m_rootip
, S_IFREG
, 1, 0,
1439 &zerocr
, 0, 1, ip
, &committed
))) {
1440 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1446 * Keep an extra reference to this quota inode. This inode is
1447 * locked exclusively and joined to the transaction already.
1449 ASSERT(XFS_ISLOCKED_INODE_EXCL(*ip
));
1450 VN_HOLD(XFS_ITOV((*ip
)));
1453 * Make the changes in the superblock, and log those too.
1454 * sbfields arg may contain fields other than *QUOTINO;
1455 * VERSIONNUM for example.
1457 s
= XFS_SB_LOCK(mp
);
1458 if (flags
& XFS_QMOPT_SBVERSION
) {
1459 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1460 unsigned oldv
= mp
->m_sb
.sb_versionnum
;
1462 ASSERT(!XFS_SB_VERSION_HASQUOTA(&mp
->m_sb
));
1463 ASSERT((sbfields
& (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1464 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
)) ==
1465 (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
1466 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
));
1468 XFS_SB_VERSION_ADDQUOTA(&mp
->m_sb
);
1469 mp
->m_sb
.sb_uquotino
= NULLFSINO
;
1470 mp
->m_sb
.sb_gquotino
= NULLFSINO
;
1472 /* qflags will get updated _after_ quotacheck */
1473 mp
->m_sb
.sb_qflags
= 0;
1474 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1476 "Old superblock version %x, converting to %x.",
1477 oldv
, mp
->m_sb
.sb_versionnum
);
1480 if (flags
& XFS_QMOPT_UQUOTA
)
1481 mp
->m_sb
.sb_uquotino
= (*ip
)->i_ino
;
1483 mp
->m_sb
.sb_gquotino
= (*ip
)->i_ino
;
1484 XFS_SB_UNLOCK(mp
, s
);
1485 xfs_mod_sb(tp
, sbfields
);
1487 if ((error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
,
1489 xfs_fs_cmn_err(CE_ALERT
, mp
, "XFS qino_alloc failed!");
1497 xfs_qm_reset_dqcounts(
1503 xfs_disk_dquot_t
*ddq
;
1506 xfs_buftrace("RESET DQUOTS", bp
);
1508 * Reset all counters and timers. They'll be
1509 * started afresh by xfs_qm_quotacheck.
1512 j
= XFS_FSB_TO_B(mp
, XFS_DQUOT_CLUSTER_SIZE_FSB
);
1513 do_div(j
, sizeof(xfs_dqblk_t
));
1514 ASSERT(XFS_QM_DQPERBLK(mp
) == j
);
1516 ddq
= (xfs_disk_dquot_t
*)XFS_BUF_PTR(bp
);
1517 for (j
= 0; j
< XFS_QM_DQPERBLK(mp
); j
++) {
1519 * Do a sanity check, and if needed, repair the dqblk. Don't
1520 * output any warnings because it's perfectly possible to
1521 * find unitialized dquot blks. See comment in xfs_qm_dqcheck.
1523 (void) xfs_qm_dqcheck(ddq
, id
+j
, type
, XFS_QMOPT_DQREPAIR
,
1525 INT_SET(ddq
->d_bcount
, ARCH_CONVERT
, 0ULL);
1526 INT_SET(ddq
->d_icount
, ARCH_CONVERT
, 0ULL);
1527 INT_SET(ddq
->d_rtbcount
, ARCH_CONVERT
, 0ULL);
1528 INT_SET(ddq
->d_btimer
, ARCH_CONVERT
, (time_t)0);
1529 INT_SET(ddq
->d_itimer
, ARCH_CONVERT
, (time_t)0);
1530 INT_SET(ddq
->d_bwarns
, ARCH_CONVERT
, 0UL);
1531 INT_SET(ddq
->d_iwarns
, ARCH_CONVERT
, 0UL);
1532 ddq
= (xfs_disk_dquot_t
*) ((xfs_dqblk_t
*)ddq
+ 1);
1543 xfs_filblks_t blkcnt
,
1553 incr
= (blkcnt
> XFS_QM_MAX_DQCLUSTER_LOGSZ
) ?
1554 XFS_QM_MAX_DQCLUSTER_LOGSZ
: blkcnt
;
1558 * Blkcnt arg can be a very big number, and might even be
1559 * larger than the log itself. So, we have to break it up into
1560 * manageable-sized transactions.
1561 * Note that we don't start a permanent transaction here; we might
1562 * not be able to get a log reservation for the whole thing up front,
1563 * and we don't really care to either, because we just discard
1564 * everything if we were to crash in the middle of this loop.
1567 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
,
1568 XFS_FSB_TO_DADDR(mp
, bno
),
1569 (int)XFS_QI_DQCHUNKLEN(mp
), 0, &bp
);
1573 (void) xfs_qm_reset_dqcounts(mp
, bp
, firstid
,
1574 flags
& XFS_QMOPT_UQUOTA
?
1575 XFS_DQ_USER
: XFS_DQ_GROUP
);
1576 xfs_bdwrite(mp
, bp
);
1578 * goto the next block.
1581 firstid
+= XFS_QM_DQPERBLK(mp
);
1587 * Iterate over all allocated USR/GRP dquots in the system, calling a
1588 * caller supplied function for every chunk of dquots that we find.
1596 xfs_bmbt_irec_t
*map
;
1597 int i
, nmaps
; /* number of map entries */
1598 int error
; /* return value */
1599 xfs_fileoff_t lblkno
;
1600 xfs_filblks_t maxlblkcnt
;
1602 xfs_fsblock_t rablkno
;
1603 xfs_filblks_t rablkcnt
;
1607 * This looks racey, but we can't keep an inode lock across a
1608 * trans_reserve. But, this gets called during quotacheck, and that
1609 * happens only at mount time which is single threaded.
1611 if (qip
->i_d
.di_nblocks
== 0)
1614 map
= kmem_alloc(XFS_DQITER_MAP_SIZE
* sizeof(*map
), KM_SLEEP
);
1617 maxlblkcnt
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1619 nmaps
= XFS_DQITER_MAP_SIZE
;
1621 * We aren't changing the inode itself. Just changing
1622 * some of its data. No new blocks are added here, and
1623 * the inode is never added to the transaction.
1625 xfs_ilock(qip
, XFS_ILOCK_SHARED
);
1626 error
= xfs_bmapi(NULL
, qip
, lblkno
,
1627 maxlblkcnt
- lblkno
,
1630 0, map
, &nmaps
, NULL
);
1631 xfs_iunlock(qip
, XFS_ILOCK_SHARED
);
1635 ASSERT(nmaps
<= XFS_DQITER_MAP_SIZE
);
1636 for (i
= 0; i
< nmaps
; i
++) {
1637 ASSERT(map
[i
].br_startblock
!= DELAYSTARTBLOCK
);
1638 ASSERT(map
[i
].br_blockcount
);
1641 lblkno
+= map
[i
].br_blockcount
;
1643 if (map
[i
].br_startblock
== HOLESTARTBLOCK
)
1646 firstid
= (xfs_dqid_t
) map
[i
].br_startoff
*
1647 XFS_QM_DQPERBLK(mp
);
1649 * Do a read-ahead on the next extent.
1651 if ((i
+1 < nmaps
) &&
1652 (map
[i
+1].br_startblock
!= HOLESTARTBLOCK
)) {
1653 rablkcnt
= map
[i
+1].br_blockcount
;
1654 rablkno
= map
[i
+1].br_startblock
;
1655 while (rablkcnt
--) {
1656 xfs_baread(mp
->m_ddev_targp
,
1657 XFS_FSB_TO_DADDR(mp
, rablkno
),
1658 (int)XFS_QI_DQCHUNKLEN(mp
));
1663 * Iterate thru all the blks in the extent and
1664 * reset the counters of all the dquots inside them.
1666 if ((error
= xfs_qm_dqiter_bufs(mp
,
1668 map
[i
].br_startblock
,
1669 map
[i
].br_blockcount
,
1677 } while (nmaps
> 0);
1679 kmem_free(map
, XFS_DQITER_MAP_SIZE
* sizeof(*map
));
1685 * Called by dqusage_adjust in doing a quotacheck.
1686 * Given the inode, and a dquot (either USR or GRP, doesn't matter),
1687 * this updates its incore copy as well as the buffer copy. This is
1688 * so that once the quotacheck is done, we can just log all the buffers,
1689 * as opposed to logging numerous updates to individual dquots.
1692 xfs_qm_quotacheck_dqadjust(
1697 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
1698 xfs_dqtrace_entry(dqp
, "QCHECK DQADJUST");
1700 * Adjust the inode count and the block count to reflect this inode's
1703 INT_MOD(dqp
->q_core
.d_icount
, ARCH_CONVERT
, +1);
1704 dqp
->q_res_icount
++;
1706 INT_MOD(dqp
->q_core
.d_bcount
, ARCH_CONVERT
, nblks
);
1707 dqp
->q_res_bcount
+= nblks
;
1710 INT_MOD(dqp
->q_core
.d_rtbcount
, ARCH_CONVERT
, rtblks
);
1711 dqp
->q_res_rtbcount
+= rtblks
;
1715 * Set default limits, adjust timers (since we changed usages)
1717 if (! XFS_IS_SUSER_DQUOT(dqp
)) {
1718 xfs_qm_adjust_dqlimits(dqp
->q_mount
, &dqp
->q_core
);
1719 xfs_qm_adjust_dqtimers(dqp
->q_mount
, &dqp
->q_core
);
1722 dqp
->dq_flags
|= XFS_DQ_DIRTY
;
1728 xfs_qcnt_t
*O_rtblks
)
1730 xfs_filblks_t rtblks
; /* total rt blks */
1731 xfs_ifork_t
*ifp
; /* inode fork pointer */
1732 xfs_extnum_t nextents
; /* number of extent entries */
1733 xfs_bmbt_rec_t
*base
; /* base of extent array */
1734 xfs_bmbt_rec_t
*ep
; /* pointer to an extent entry */
1737 ASSERT(XFS_IS_REALTIME_INODE(ip
));
1738 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1739 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1740 if ((error
= xfs_iread_extents(NULL
, ip
, XFS_DATA_FORK
)))
1744 nextents
= ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
);
1745 base
= &ifp
->if_u1
.if_extents
[0];
1746 for (ep
= base
; ep
< &base
[nextents
]; ep
++)
1747 rtblks
+= xfs_bmbt_get_blockcount(ep
);
1748 *O_rtblks
= (xfs_qcnt_t
)rtblks
;
1753 * callback routine supplied to bulkstat(). Given an inumber, find its
1754 * dquots and update them to account for resources taken by that inode.
1758 xfs_qm_dqusage_adjust(
1759 xfs_mount_t
*mp
, /* mount point for filesystem */
1760 xfs_ino_t ino
, /* inode number to get data for */
1761 void __user
*buffer
, /* not used */
1762 int ubsize
, /* not used */
1763 void *private_data
, /* not used */
1764 xfs_daddr_t bno
, /* starting block of inode cluster */
1765 int *ubused
, /* not used */
1766 void *dip
, /* on-disk inode pointer (not used) */
1767 int *res
) /* result code value */
1770 xfs_dquot_t
*udqp
, *gdqp
;
1771 xfs_qcnt_t nblks
, rtblks
;
1774 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1777 * rootino must have its resources accounted for, not so with the quota
1780 if (ino
== mp
->m_sb
.sb_uquotino
|| ino
== mp
->m_sb
.sb_gquotino
) {
1781 *res
= BULKSTAT_RV_NOTHING
;
1782 return XFS_ERROR(EINVAL
);
1786 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1787 * interface expects the inode to be exclusively locked because that's
1788 * the case in all other instances. It's OK that we do this because
1789 * quotacheck is done only at mount time.
1791 if ((error
= xfs_iget(mp
, NULL
, ino
, XFS_ILOCK_EXCL
, &ip
, bno
))) {
1792 *res
= BULKSTAT_RV_NOTHING
;
1796 if (ip
->i_d
.di_mode
== 0) {
1797 xfs_iput_new(ip
, XFS_ILOCK_EXCL
);
1798 *res
= BULKSTAT_RV_NOTHING
;
1799 return XFS_ERROR(ENOENT
);
1803 * Obtain the locked dquots. In case of an error (eg. allocation
1804 * fails for ENOSPC), we return the negative of the error number
1805 * to bulkstat, so that it can get propagated to quotacheck() and
1806 * making us disable quotas for the file system.
1808 if ((error
= xfs_qm_dqget_noattach(ip
, &udqp
, &gdqp
))) {
1809 xfs_iput(ip
, XFS_ILOCK_EXCL
);
1810 *res
= BULKSTAT_RV_GIVEUP
;
1815 if (! XFS_IS_REALTIME_INODE(ip
)) {
1816 nblks
= (xfs_qcnt_t
)ip
->i_d
.di_nblocks
;
1819 * Walk thru the extent list and count the realtime blocks.
1821 if ((error
= xfs_qm_get_rtblks(ip
, &rtblks
))) {
1822 xfs_iput(ip
, XFS_ILOCK_EXCL
);
1827 *res
= BULKSTAT_RV_GIVEUP
;
1830 nblks
= (xfs_qcnt_t
)ip
->i_d
.di_nblocks
- rtblks
;
1832 ASSERT(ip
->i_delayed_blks
== 0);
1835 * We can't release the inode while holding its dquot locks.
1836 * The inode can go into inactive and might try to acquire the dquotlocks.
1837 * So, just unlock here and do a vn_rele at the end.
1839 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1842 * Add the (disk blocks and inode) resources occupied by this
1843 * inode to its dquots. We do this adjustment in the incore dquot,
1844 * and also copy the changes to its buffer.
1845 * We don't care about putting these changes in a transaction
1846 * envelope because if we crash in the middle of a 'quotacheck'
1847 * we have to start from the beginning anyway.
1848 * Once we're done, we'll log all the dquot bufs.
1850 * The *QUOTA_ON checks below may look pretty racey, but quotachecks
1851 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1853 if (XFS_IS_UQUOTA_ON(mp
)) {
1855 xfs_qm_quotacheck_dqadjust(udqp
, nblks
, rtblks
);
1858 if (XFS_IS_GQUOTA_ON(mp
)) {
1860 xfs_qm_quotacheck_dqadjust(gdqp
, nblks
, rtblks
);
1864 * Now release the inode. This will send it to 'inactive', and
1865 * possibly even free blocks.
1867 VN_RELE(XFS_ITOV(ip
));
1872 *res
= BULKSTAT_RV_DIDONE
;
1877 * Walk thru all the filesystem inodes and construct a consistent view
1878 * of the disk quota world.
1884 int done
, count
, error
;
1887 xfs_inode_t
*uip
, *gip
;
1895 ASSERT(XFS_QI_UQIP(mp
) || XFS_QI_GQIP(mp
));
1896 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
1899 * There should be no cached dquots. The (simplistic) quotacheck
1900 * algorithm doesn't like that.
1902 ASSERT(XFS_QI_MPLNDQUOTS(mp
) == 0);
1904 cmn_err(CE_NOTE
, "XFS quotacheck %s: Please wait.", mp
->m_fsname
);
1907 * First we go thru all the dquots on disk, USR and GRP, and reset
1908 * their counters to zero. We need a clean slate.
1909 * We don't log our changes till later.
1911 if ((uip
= XFS_QI_UQIP(mp
))) {
1912 if ((error
= xfs_qm_dqiterate(mp
, uip
, XFS_QMOPT_UQUOTA
)))
1914 flags
|= XFS_UQUOTA_CHKD
;
1917 if ((gip
= XFS_QI_GQIP(mp
))) {
1918 if ((error
= xfs_qm_dqiterate(mp
, gip
, XFS_QMOPT_GQUOTA
)))
1920 flags
|= XFS_GQUOTA_CHKD
;
1925 * Iterate thru all the inodes in the file system,
1926 * adjusting the corresponding dquot counters in core.
1928 if ((error
= xfs_bulkstat(mp
, &lastino
, &count
,
1929 xfs_qm_dqusage_adjust
, NULL
,
1931 BULKSTAT_FG_IGET
|BULKSTAT_FG_VFSLOCKED
,
1938 * We can get this error if we couldn't do a dquot allocation inside
1939 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1940 * dirty dquots that might be cached, we just want to get rid of them
1941 * and turn quotaoff. The dquots won't be attached to any of the inodes
1942 * at this point (because we intentionally didn't in dqget_noattach).
1945 xfs_qm_dqpurge_all(mp
,
1946 XFS_QMOPT_UQUOTA
|XFS_QMOPT_GQUOTA
|
1947 XFS_QMOPT_QUOTAOFF
);
1951 * We've made all the changes that we need to make incore.
1952 * Now flush_them down to disk buffers.
1954 xfs_qm_dqflush_all(mp
, XFS_QMOPT_DELWRI
);
1957 * We didn't log anything, because if we crashed, we'll have to
1958 * start the quotacheck from scratch anyway. However, we must make
1959 * sure that our dquot changes are secure before we put the
1960 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1963 XFS_bflush(mp
->m_ddev_targp
);
1966 * If one type of quotas is off, then it will lose its
1967 * quotachecked status, since we won't be doing accounting for
1968 * that type anymore.
1970 mp
->m_qflags
&= ~(XFS_GQUOTA_CHKD
| XFS_UQUOTA_CHKD
);
1971 mp
->m_qflags
|= flags
;
1973 XQM_LIST_PRINT(&(XFS_QI_MPL_LIST(mp
)), MPL_NEXT
, "++++ Mp list +++");
1976 cmn_err(CE_NOTE
, "XFS quotacheck %s: Done.", mp
->m_fsname
);
1981 * This is called after the superblock has been read in and we're ready to
1982 * iget the quota inodes.
1985 xfs_qm_init_quotainos(
1988 xfs_inode_t
*uip
, *gip
;
1993 ASSERT(mp
->m_quotainfo
);
1999 * Get the uquota and gquota inodes
2001 if (XFS_SB_VERSION_HASQUOTA(&mp
->m_sb
)) {
2002 if (XFS_IS_UQUOTA_ON(mp
) &&
2003 mp
->m_sb
.sb_uquotino
!= NULLFSINO
) {
2004 ASSERT(mp
->m_sb
.sb_uquotino
> 0);
2005 if ((error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_uquotino
,
2007 return XFS_ERROR(error
);
2009 if (XFS_IS_GQUOTA_ON(mp
) &&
2010 mp
->m_sb
.sb_gquotino
!= NULLFSINO
) {
2011 ASSERT(mp
->m_sb
.sb_gquotino
> 0);
2012 if ((error
= xfs_iget(mp
, NULL
, mp
->m_sb
.sb_gquotino
,
2015 VN_RELE(XFS_ITOV(uip
));
2016 return XFS_ERROR(error
);
2020 flags
|= XFS_QMOPT_SBVERSION
;
2021 sbflags
|= (XFS_SB_VERSIONNUM
| XFS_SB_UQUOTINO
|
2022 XFS_SB_GQUOTINO
| XFS_SB_QFLAGS
);
2026 * Create the two inodes, if they don't exist already. The changes
2027 * made above will get added to a transaction and logged in one of
2028 * the qino_alloc calls below. If the device is readonly,
2029 * temporarily switch to read-write to do this.
2031 if (XFS_IS_UQUOTA_ON(mp
) && uip
== NULL
) {
2032 if ((error
= xfs_qm_qino_alloc(mp
, &uip
,
2033 sbflags
| XFS_SB_UQUOTINO
,
2034 flags
| XFS_QMOPT_UQUOTA
)))
2035 return XFS_ERROR(error
);
2037 flags
&= ~XFS_QMOPT_SBVERSION
;
2039 if (XFS_IS_GQUOTA_ON(mp
) && gip
== NULL
) {
2040 if ((error
= xfs_qm_qino_alloc(mp
, &gip
,
2041 sbflags
| XFS_SB_GQUOTINO
,
2042 flags
| XFS_QMOPT_GQUOTA
))) {
2044 VN_RELE(XFS_ITOV(uip
));
2046 return XFS_ERROR(error
);
2050 XFS_QI_UQIP(mp
) = uip
;
2051 XFS_QI_GQIP(mp
) = gip
;
2058 * Traverse the freelist of dquots and attempt to reclaim a maximum of
2059 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
2060 * favor the lookup function ...
2061 * XXXsup merge this with qm_reclaim_one().
2064 xfs_qm_shake_freelist(
2069 xfs_dquot_t
*dqp
, *nextdqp
;
2081 cmn_err(CE_DEBUG
, "Shake free 0x%x", howmany
);
2083 /* lock order is : hashchainlock, freelistlock, mplistlock */
2085 xfs_qm_freelist_lock(xfs_Gqm
);
2087 for (dqp
= xfs_Gqm
->qm_dqfreelist
.qh_next
;
2088 ((dqp
!= (xfs_dquot_t
*) &xfs_Gqm
->qm_dqfreelist
) &&
2089 nreclaimed
< howmany
); ) {
2093 * We are racing with dqlookup here. Naturally we don't
2094 * want to reclaim a dquot that lookup wants.
2096 if (dqp
->dq_flags
& XFS_DQ_WANT
) {
2098 xfs_qm_freelist_unlock(xfs_Gqm
);
2099 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2100 return (nreclaimed
);
2101 XQM_STATS_INC(xqmstats
.xs_qm_dqwants
);
2106 * If the dquot is inactive, we are assured that it is
2107 * not on the mplist or the hashlist, and that makes our
2110 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
2111 ASSERT(dqp
->q_mount
== NULL
);
2112 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
2113 ASSERT(dqp
->HL_PREVP
== NULL
);
2114 ASSERT(dqp
->MPL_PREVP
== NULL
);
2115 XQM_STATS_INC(xqmstats
.xs_qm_dqinact_reclaims
);
2116 nextdqp
= dqp
->dq_flnext
;
2120 ASSERT(dqp
->MPL_PREVP
);
2122 * Try to grab the flush lock. If this dquot is in the process of
2123 * getting flushed to disk, we don't want to reclaim it.
2125 if (! xfs_qm_dqflock_nowait(dqp
)) {
2127 dqp
= dqp
->dq_flnext
;
2132 * We have the flush lock so we know that this is not in the
2133 * process of being flushed. So, if this is dirty, flush it
2134 * DELWRI so that we don't get a freelist infested with
2137 if (XFS_DQ_IS_DIRTY(dqp
)) {
2138 xfs_dqtrace_entry(dqp
, "DQSHAKE: DQDIRTY");
2140 * We flush it delayed write, so don't bother
2141 * releasing the mplock.
2143 (void) xfs_qm_dqflush(dqp
, XFS_QMOPT_DELWRI
);
2144 xfs_dqunlock(dqp
); /* dqflush unlocks dqflock */
2145 dqp
= dqp
->dq_flnext
;
2149 * We're trying to get the hashlock out of order. This races
2150 * with dqlookup; so, we giveup and goto the next dquot if
2151 * we couldn't get the hashlock. This way, we won't starve
2152 * a dqlookup process that holds the hashlock that is
2153 * waiting for the freelist lock.
2155 if (! xfs_qm_dqhashlock_nowait(dqp
)) {
2158 dqp
= dqp
->dq_flnext
;
2162 * This races with dquot allocation code as well as dqflush_all
2163 * and reclaim code. So, if we failed to grab the mplist lock,
2164 * giveup everything and start over.
2168 if (! xfs_qm_mplist_nowait(dqp
->q_mount
)) {
2169 /* XXX put a sentinel so that we can come back here */
2172 XFS_DQ_HASH_UNLOCK(hash
);
2173 xfs_qm_freelist_unlock(xfs_Gqm
);
2174 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2175 return (nreclaimed
);
2178 xfs_dqtrace_entry(dqp
, "DQSHAKE: UNLINKING");
2180 cmn_err(CE_DEBUG
, "Shake 0x%p, ID 0x%x\n",
2181 dqp
, INT_GET(dqp
->q_core
.d_id
, ARCH_CONVERT
));
2183 ASSERT(dqp
->q_nrefs
== 0);
2184 nextdqp
= dqp
->dq_flnext
;
2185 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp
->q_mount
)), dqp
);
2186 XQM_HASHLIST_REMOVE(hash
, dqp
);
2188 xfs_qm_mplist_unlock(dqp
->q_mount
);
2189 XFS_DQ_HASH_UNLOCK(hash
);
2192 XQM_FREELIST_REMOVE(dqp
);
2195 XQM_STATS_INC(xqmstats
.xs_qm_dqshake_reclaims
);
2196 xfs_qm_dqdestroy(dqp
);
2199 xfs_qm_freelist_unlock(xfs_Gqm
);
2200 return (nreclaimed
);
2205 * The kmem_shake interface is invoked when memory is running low.
2209 xfs_qm_shake(int nr_to_scan
, unsigned int gfp_mask
)
2211 int ndqused
, nfree
, n
;
2213 if (!kmem_shake_allow(gfp_mask
))
2218 nfree
= xfs_Gqm
->qm_dqfreelist
.qh_nelems
; /* free dquots */
2219 /* incore dquots in all f/s's */
2220 ndqused
= atomic_read(&xfs_Gqm
->qm_totaldquots
) - nfree
;
2222 ASSERT(ndqused
>= 0);
2224 if (nfree
<= ndqused
&& nfree
< ndquot
)
2227 ndqused
*= xfs_Gqm
->qm_dqfree_ratio
; /* target # of free dquots */
2228 n
= nfree
- ndqused
- ndquot
; /* # over target */
2230 return xfs_qm_shake_freelist(MAX(nfree
, n
));
2235 * Just pop the least recently used dquot off the freelist and
2236 * recycle it. The returned dquot is locked.
2238 STATIC xfs_dquot_t
*
2239 xfs_qm_dqreclaim_one(void)
2241 xfs_dquot_t
*dqpout
;
2250 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
2252 xfs_qm_freelist_lock(xfs_Gqm
);
2254 FOREACH_DQUOT_IN_FREELIST(dqp
, &(xfs_Gqm
->qm_dqfreelist
)) {
2258 * We are racing with dqlookup here. Naturally we don't
2259 * want to reclaim a dquot that lookup wants. We release the
2260 * freelist lock and start over, so that lookup will grab
2261 * both the dquot and the freelistlock.
2263 if (dqp
->dq_flags
& XFS_DQ_WANT
) {
2264 ASSERT(! (dqp
->dq_flags
& XFS_DQ_INACTIVE
));
2265 xfs_dqtrace_entry(dqp
, "DQRECLAIM: DQWANT");
2267 xfs_qm_freelist_unlock(xfs_Gqm
);
2268 if (++restarts
>= XFS_QM_RECLAIM_MAX_RESTARTS
)
2270 XQM_STATS_INC(xqmstats
.xs_qm_dqwants
);
2275 * If the dquot is inactive, we are assured that it is
2276 * not on the mplist or the hashlist, and that makes our
2279 if (dqp
->dq_flags
& XFS_DQ_INACTIVE
) {
2280 ASSERT(dqp
->q_mount
== NULL
);
2281 ASSERT(! XFS_DQ_IS_DIRTY(dqp
));
2282 ASSERT(dqp
->HL_PREVP
== NULL
);
2283 ASSERT(dqp
->MPL_PREVP
== NULL
);
2284 XQM_FREELIST_REMOVE(dqp
);
2287 XQM_STATS_INC(xqmstats
.xs_qm_dqinact_reclaims
);
2291 ASSERT(dqp
->q_hash
);
2292 ASSERT(dqp
->MPL_PREVP
);
2295 * Try to grab the flush lock. If this dquot is in the process of
2296 * getting flushed to disk, we don't want to reclaim it.
2298 if (! xfs_qm_dqflock_nowait(dqp
)) {
2304 * We have the flush lock so we know that this is not in the
2305 * process of being flushed. So, if this is dirty, flush it
2306 * DELWRI so that we don't get a freelist infested with
2309 if (XFS_DQ_IS_DIRTY(dqp
)) {
2310 xfs_dqtrace_entry(dqp
, "DQRECLAIM: DQDIRTY");
2312 * We flush it delayed write, so don't bother
2313 * releasing the freelist lock.
2315 (void) xfs_qm_dqflush(dqp
, XFS_QMOPT_DELWRI
);
2316 xfs_dqunlock(dqp
); /* dqflush unlocks dqflock */
2320 if (! xfs_qm_mplist_nowait(dqp
->q_mount
)) {
2326 if (! xfs_qm_dqhashlock_nowait(dqp
))
2329 ASSERT(dqp
->q_nrefs
== 0);
2330 xfs_dqtrace_entry(dqp
, "DQRECLAIM: UNLINKING");
2331 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp
->q_mount
)), dqp
);
2332 XQM_HASHLIST_REMOVE(dqp
->q_hash
, dqp
);
2333 XQM_FREELIST_REMOVE(dqp
);
2335 XFS_DQ_HASH_UNLOCK(dqp
->q_hash
);
2337 xfs_qm_mplist_unlock(dqp
->q_mount
);
2344 xfs_qm_freelist_unlock(xfs_Gqm
);
2349 /*------------------------------------------------------------------*/
2352 * Return a new incore dquot. Depending on the number of
2353 * dquots in the system, we either allocate a new one on the kernel heap,
2354 * or reclaim a free one.
2355 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2356 * to reclaim an existing one from the freelist.
2359 xfs_qm_dqalloc_incore(
2360 xfs_dquot_t
**O_dqpp
)
2365 * Check against high water mark to see if we want to pop
2366 * a nincompoop dquot off the freelist.
2368 if (atomic_read(&xfs_Gqm
->qm_totaldquots
) >= ndquot
) {
2370 * Try to recycle a dquot from the freelist.
2372 if ((dqp
= xfs_qm_dqreclaim_one())) {
2373 XQM_STATS_INC(xqmstats
.xs_qm_dqreclaims
);
2375 * Just zero the core here. The rest will get
2376 * reinitialized by caller. XXX we shouldn't even
2379 memset(&dqp
->q_core
, 0, sizeof(dqp
->q_core
));
2383 XQM_STATS_INC(xqmstats
.xs_qm_dqreclaim_misses
);
2387 * Allocate a brand new dquot on the kernel heap and return it
2388 * to the caller to initialize.
2390 ASSERT(xfs_Gqm
->qm_dqzone
!= NULL
);
2391 *O_dqpp
= kmem_zone_zalloc(xfs_Gqm
->qm_dqzone
, KM_SLEEP
);
2392 atomic_inc(&xfs_Gqm
->qm_totaldquots
);
2399 * Start a transaction and write the incore superblock changes to
2400 * disk. flags parameter indicates which fields have changed.
2403 xfs_qm_write_sb_changes(
2411 cmn_err(CE_NOTE
, "Writing superblock quota changes :%s", mp
->m_fsname
);
2413 tp
= xfs_trans_alloc(mp
, XFS_TRANS_QM_SBCHANGE
);
2414 if ((error
= xfs_trans_reserve(tp
, 0,
2415 mp
->m_sb
.sb_sectsize
+ 128, 0,
2417 XFS_DEFAULT_LOG_COUNT
))) {
2418 xfs_trans_cancel(tp
, 0);
2422 xfs_mod_sb(tp
, flags
);
2423 (void) xfs_trans_commit(tp
, 0, NULL
);
2429 /* --------------- utility functions for vnodeops ---------------- */
2433 * Given an inode, a uid and gid (from cred_t) make sure that we have
2434 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2435 * quotas by creating this file.
2436 * This also attaches dquot(s) to the given inode after locking it,
2437 * and returns the dquots corresponding to the uid and/or gid.
2439 * in : inode (unlocked)
2440 * out : udquot, gdquot with references taken and unlocked
2449 xfs_dquot_t
**O_udqpp
,
2450 xfs_dquot_t
**O_gdqpp
)
2453 xfs_dquot_t
*uq
, *gq
;
2456 if (!XFS_IS_QUOTA_ON(mp
))
2459 lockflags
= XFS_ILOCK_EXCL
;
2460 xfs_ilock(ip
, lockflags
);
2462 if ((flags
& XFS_QMOPT_INHERIT
) &&
2463 XFS_INHERIT_GID(ip
, XFS_MTOVFS(mp
)))
2464 gid
= ip
->i_d
.di_gid
;
2467 * Attach the dquot(s) to this inode, doing a dquot allocation
2468 * if necessary. The dquot(s) will not be locked.
2470 if (XFS_NOT_DQATTACHED(mp
, ip
)) {
2471 if ((error
= xfs_qm_dqattach(ip
, XFS_QMOPT_DQALLOC
|
2472 XFS_QMOPT_ILOCKED
))) {
2473 xfs_iunlock(ip
, lockflags
);
2479 if ((flags
& XFS_QMOPT_UQUOTA
) &&
2480 XFS_IS_UQUOTA_ON(mp
)) {
2481 if (ip
->i_d
.di_uid
!= uid
) {
2483 * What we need is the dquot that has this uid, and
2484 * if we send the inode to dqget, the uid of the inode
2485 * takes priority over what's sent in the uid argument.
2486 * We must unlock inode here before calling dqget if
2487 * we're not sending the inode, because otherwise
2488 * we'll deadlock by doing trans_reserve while
2491 xfs_iunlock(ip
, lockflags
);
2492 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
) uid
,
2497 ASSERT(error
!= ENOENT
);
2501 * Get the ilock in the right order.
2504 lockflags
= XFS_ILOCK_SHARED
;
2505 xfs_ilock(ip
, lockflags
);
2508 * Take an extra reference, because we'll return
2511 ASSERT(ip
->i_udquot
);
2518 if ((flags
& XFS_QMOPT_GQUOTA
) &&
2519 XFS_IS_GQUOTA_ON(mp
)) {
2520 if (ip
->i_d
.di_gid
!= gid
) {
2521 xfs_iunlock(ip
, lockflags
);
2522 if ((error
= xfs_qm_dqget(mp
, NULL
, (xfs_dqid_t
)gid
,
2529 ASSERT(error
!= ENOENT
);
2533 lockflags
= XFS_ILOCK_SHARED
;
2534 xfs_ilock(ip
, lockflags
);
2536 ASSERT(ip
->i_gdquot
);
2544 xfs_dqtrace_entry_ino(uq
, "DQALLOC", ip
);
2546 xfs_iunlock(ip
, lockflags
);
2559 * Actually transfer ownership, and do dquot modifications.
2560 * These were already reserved.
2566 xfs_dquot_t
**IO_olddq
,
2569 xfs_dquot_t
*prevdq
;
2570 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
2571 ASSERT(XFS_IS_QUOTA_RUNNING(ip
->i_mount
));
2576 ASSERT(prevdq
!= newdq
);
2578 xfs_trans_mod_dquot(tp
, prevdq
,
2579 XFS_TRANS_DQ_BCOUNT
,
2580 -(ip
->i_d
.di_nblocks
));
2581 xfs_trans_mod_dquot(tp
, prevdq
,
2582 XFS_TRANS_DQ_ICOUNT
,
2585 /* the sparkling new dquot */
2586 xfs_trans_mod_dquot(tp
, newdq
,
2587 XFS_TRANS_DQ_BCOUNT
,
2588 ip
->i_d
.di_nblocks
);
2589 xfs_trans_mod_dquot(tp
, newdq
,
2590 XFS_TRANS_DQ_ICOUNT
,
2594 * Take an extra reference, because the inode
2595 * is going to keep this dquot pointer even
2596 * after the trans_commit.
2600 xfs_dqunlock(newdq
);
2607 * Quota reservations for setattr(AT_UID|AT_GID).
2610 xfs_qm_vop_chown_reserve(
2620 xfs_dquot_t
*unresudq
, *unresgdq
, *delblksudq
, *delblksgdq
;
2622 ASSERT(XFS_ISLOCKED_INODE(ip
));
2624 ASSERT(XFS_IS_QUOTA_RUNNING(mp
));
2626 delblks
= ip
->i_delayed_blks
;
2627 delblksudq
= delblksgdq
= unresudq
= unresgdq
= NULL
;
2629 if (XFS_IS_UQUOTA_ON(mp
) && udqp
&&
2630 ip
->i_d
.di_uid
!= (uid_t
)INT_GET(udqp
->q_core
.d_id
, ARCH_CONVERT
)) {
2633 * If there are delayed allocation blocks, then we have to
2634 * unreserve those from the old dquot, and add them to the
2638 ASSERT(ip
->i_udquot
);
2639 unresudq
= ip
->i_udquot
;
2642 if (XFS_IS_GQUOTA_ON(ip
->i_mount
) && gdqp
&&
2643 ip
->i_d
.di_gid
!= INT_GET(gdqp
->q_core
.d_id
, ARCH_CONVERT
)) {
2646 ASSERT(ip
->i_gdquot
);
2647 unresgdq
= ip
->i_gdquot
;
2651 if ((error
= xfs_trans_reserve_quota_bydquots(tp
, ip
->i_mount
,
2652 delblksudq
, delblksgdq
, ip
->i_d
.di_nblocks
, 1,
2653 flags
| XFS_QMOPT_RES_REGBLKS
)))
2657 * Do the delayed blks reservations/unreservations now. Since, these
2658 * are done without the help of a transaction, if a reservation fails
2659 * its previous reservations won't be automatically undone by trans
2660 * code. So, we have to do it manually here.
2664 * Do the reservations first. Unreservation can't fail.
2666 ASSERT(delblksudq
|| delblksgdq
);
2667 ASSERT(unresudq
|| unresgdq
);
2668 if ((error
= xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
2669 delblksudq
, delblksgdq
, (xfs_qcnt_t
)delblks
, 0,
2670 flags
| XFS_QMOPT_RES_REGBLKS
)))
2672 xfs_trans_reserve_quota_bydquots(NULL
, ip
->i_mount
,
2673 unresudq
, unresgdq
, -((xfs_qcnt_t
)delblks
), 0,
2674 XFS_QMOPT_RES_REGBLKS
);
2681 xfs_qm_vop_rename_dqattach(
2682 xfs_inode_t
**i_tab
)
2690 if (! XFS_IS_QUOTA_ON(ip
->i_mount
))
2693 if (XFS_NOT_DQATTACHED(ip
->i_mount
, ip
)) {
2694 error
= xfs_qm_dqattach(ip
, 0);
2698 for (i
= 1; (i
< 4 && i_tab
[i
]); i
++) {
2700 * Watch out for duplicate entries in the table.
2702 if ((ip
= i_tab
[i
]) != i_tab
[i
-1]) {
2703 if (XFS_NOT_DQATTACHED(ip
->i_mount
, ip
)) {
2704 error
= xfs_qm_dqattach(ip
, 0);
2714 xfs_qm_vop_dqattach_and_dqmod_newinode(
2720 if (!XFS_IS_QUOTA_ON(tp
->t_mountp
))
2723 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip
));
2724 ASSERT(XFS_IS_QUOTA_RUNNING(tp
->t_mountp
));
2730 ASSERT(ip
->i_udquot
== NULL
);
2731 ip
->i_udquot
= udqp
;
2732 ASSERT(ip
->i_d
.di_uid
== INT_GET(udqp
->q_core
.d_id
, ARCH_CONVERT
));
2733 xfs_trans_mod_dquot(tp
, udqp
, XFS_TRANS_DQ_ICOUNT
, 1);
2739 ASSERT(ip
->i_gdquot
== NULL
);
2740 ip
->i_gdquot
= gdqp
;
2741 ASSERT(ip
->i_d
.di_gid
== INT_GET(gdqp
->q_core
.d_id
, ARCH_CONVERT
));
2742 xfs_trans_mod_dquot(tp
, gdqp
, XFS_TRANS_DQ_ICOUNT
, 1);
2746 /* ------------- list stuff -----------------*/
2748 xfs_qm_freelist_init(xfs_frlist_t
*ql
)
2750 ql
->qh_next
= ql
->qh_prev
= (xfs_dquot_t
*) ql
;
2751 mutex_init(&ql
->qh_lock
, MUTEX_DEFAULT
, "dqf");
2757 xfs_qm_freelist_destroy(xfs_frlist_t
*ql
)
2759 xfs_dquot_t
*dqp
, *nextdqp
;
2761 mutex_lock(&ql
->qh_lock
, PINOD
);
2762 for (dqp
= ql
->qh_next
;
2763 dqp
!= (xfs_dquot_t
*)ql
; ) {
2765 nextdqp
= dqp
->dq_flnext
;
2767 cmn_err(CE_DEBUG
, "FREELIST destroy 0x%p", dqp
);
2769 XQM_FREELIST_REMOVE(dqp
);
2771 xfs_qm_dqdestroy(dqp
);
2775 * Don't bother about unlocking.
2777 mutex_destroy(&ql
->qh_lock
);
2779 ASSERT(ql
->qh_nelems
== 0);
2783 xfs_qm_freelist_insert(xfs_frlist_t
*ql
, xfs_dquot_t
*dq
)
2785 dq
->dq_flnext
= ql
->qh_next
;
2786 dq
->dq_flprev
= (xfs_dquot_t
*)ql
;
2788 dq
->dq_flnext
->dq_flprev
= dq
;
2789 xfs_Gqm
->qm_dqfreelist
.qh_nelems
++;
2790 xfs_Gqm
->qm_dqfreelist
.qh_version
++;
2794 xfs_qm_freelist_unlink(xfs_dquot_t
*dq
)
2796 xfs_dquot_t
*next
= dq
->dq_flnext
;
2797 xfs_dquot_t
*prev
= dq
->dq_flprev
;
2799 next
->dq_flprev
= prev
;
2800 prev
->dq_flnext
= next
;
2801 dq
->dq_flnext
= dq
->dq_flprev
= dq
;
2802 xfs_Gqm
->qm_dqfreelist
.qh_nelems
--;
2803 xfs_Gqm
->qm_dqfreelist
.qh_version
++;
2807 xfs_qm_freelist_append(xfs_frlist_t
*ql
, xfs_dquot_t
*dq
)
2809 xfs_qm_freelist_insert((xfs_frlist_t
*)ql
->qh_prev
, dq
);
2813 xfs_qm_dqhashlock_nowait(
2818 locked
= mutex_trylock(&((dqp
)->q_hash
->qh_lock
));
2823 xfs_qm_freelist_lock_nowait(
2828 locked
= mutex_trylock(&(xqm
->qm_dqfreelist
.qh_lock
));
2833 xfs_qm_mplist_nowait(
2838 ASSERT(mp
->m_quotainfo
);
2839 locked
= mutex_trylock(&(XFS_QI_MPLLOCK(mp
)));