2 * Copyright (c) 2000-2004 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 STATIC 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 *);
89 STATIC void xfs_qm_freelist_init(xfs_frlist_t *);
90 STATIC void xfs_qm_freelist_destroy(xfs_frlist_t *);
91 STATIC int xfs_qm_mplist_nowait(xfs_mount_t *);
92 STATIC int xfs_qm_dqhashlock_nowait(xfs_dquot_t *);
94 STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
95 STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
96 STATIC int xfs_qm_shake(int, unsigned int);
99 extern mutex_t qcheck_lock;
103 #define XQM_LIST_PRINT(l, NXT, title) \
105 xfs_dquot_t *dqp; int i = 0; \
106 cmn_err(CE_DEBUG, "%s (#%d)", title, (int) (l)->qh_nelems); \
107 for (dqp = (l)->qh_next; dqp != NULL; dqp = dqp->NXT) { \
108 cmn_err(CE_DEBUG, " %d. \"%d (%s)\" " \
109 "bcnt = %d, icnt = %d, refs = %d", \
110 ++i, (int) INT_GET(dqp->q_core.d_id, ARCH_CONVERT), \
111 DQFLAGTO_TYPESTR(dqp), \
112 (int) INT_GET(dqp->q_core.d_bcount, ARCH_CONVERT), \
113 (int) INT_GET(dqp->q_core.d_icount, ARCH_CONVERT), \
114 (int) dqp->q_nrefs); } \
117 #define XQM_LIST_PRINT(l, NXT, title) do { } while (0)
121 * Initialize the XQM structure.
122 * Note that there is not one quota manager per file system.
124 STATIC struct xfs_qm *
130 xqm = kmem_zalloc(sizeof(xfs_qm_t), KM_SLEEP);
134 * Initialize the dquot hash tables.
136 hsize = (DQUOT_HASH_HEURISTIC < XFS_QM_NCSIZE_THRESHOLD) ?
137 XFS_QM_HASHSIZE_LOW : XFS_QM_HASHSIZE_HIGH;
138 xqm->qm_dqhashmask = hsize - 1;
140 xqm->qm_usr_dqhtable = (xfs_dqhash_t *)kmem_zalloc(hsize *
141 sizeof(xfs_dqhash_t),
143 xqm->qm_grp_dqhtable = (xfs_dqhash_t *)kmem_zalloc(hsize *
144 sizeof(xfs_dqhash_t),
146 ASSERT(xqm->qm_usr_dqhtable != NULL);
147 ASSERT(xqm->qm_grp_dqhtable != NULL);
149 for (i = 0; i < hsize; i++) {
150 xfs_qm_list_init(&(xqm->qm_usr_dqhtable[i]), "uxdqh", i);
151 xfs_qm_list_init(&(xqm->qm_grp_dqhtable[i]), "gxdqh", i);
155 * Freelist of all dquots of all file systems
157 xfs_qm_freelist_init(&(xqm->qm_dqfreelist));
160 * dquot zone. we register our own low-memory callback.
163 xqm->qm_dqzone = kmem_zone_init(sizeof(xfs_dquot_t),
165 qm_dqzone = xqm->qm_dqzone;
167 xqm->qm_dqzone = qm_dqzone;
169 xfs_qm_shaker = kmem_shake_register(xfs_qm_shake);
172 * The t_dqinfo portion of transactions.
175 xqm->qm_dqtrxzone = kmem_zone_init(sizeof(xfs_dquot_acct_t),
177 qm_dqtrxzone = xqm->qm_dqtrxzone;
179 xqm->qm_dqtrxzone = qm_dqtrxzone;
181 atomic_set(&xqm->qm_totaldquots, 0);
182 xqm->qm_dqfree_ratio = XFS_QM_DQFREE_RATIO;
185 mutex_init(&qcheck_lock, MUTEX_DEFAULT, "qchk");
191 * Destroy the global quota manager when its reference count goes to zero.
200 ASSERT(xqm->qm_nrefs == 0);
201 kmem_shake_deregister(xfs_qm_shaker);
202 hsize = xqm->qm_dqhashmask + 1;
203 for (i = 0; i < hsize; i++) {
204 xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
205 xfs_qm_list_destroy(&(xqm->qm_grp_dqhtable[i]));
207 kmem_free(xqm->qm_usr_dqhtable, hsize * sizeof(xfs_dqhash_t));
208 kmem_free(xqm->qm_grp_dqhtable, hsize * sizeof(xfs_dqhash_t));
209 xqm->qm_usr_dqhtable = NULL;
210 xqm->qm_grp_dqhtable = NULL;
211 xqm->qm_dqhashmask = 0;
212 xfs_qm_freelist_destroy(&(xqm->qm_dqfreelist));
214 mutex_destroy(&qcheck_lock);
216 kmem_free(xqm, sizeof(xfs_qm_t));
220 * Called at mount time to let XQM know that another file system is
221 * starting quotas. This isn't crucial information as the individual mount
222 * structures are pretty independent, but it helps the XQM keep a
223 * global view of what's going on.
227 xfs_qm_hold_quotafs_ref(
228 struct xfs_mount *mp)
231 * Need to lock the xfs_Gqm structure for things like this. For example,
232 * the structure could disappear between the entry to this routine and
233 * a HOLD operation if not locked.
235 XFS_QM_LOCK(xfs_Gqm);
238 xfs_Gqm = xfs_Gqm_init();
240 * We can keep a list of all filesystems with quotas mounted for
241 * debugging and statistical purposes, but ...
242 * Just take a reference and get out.
244 XFS_QM_HOLD(xfs_Gqm);
245 XFS_QM_UNLOCK(xfs_Gqm);
252 * Release the reference that a filesystem took at mount time,
253 * so that we know when we need to destroy the entire quota manager.
257 xfs_qm_rele_quotafs_ref(
258 struct xfs_mount *mp)
260 xfs_dquot_t *dqp, *nextdqp;
263 ASSERT(xfs_Gqm->qm_nrefs > 0);
266 * Go thru the freelist and destroy all inactive dquots.
268 xfs_qm_freelist_lock(xfs_Gqm);
270 for (dqp = xfs_Gqm->qm_dqfreelist.qh_next;
271 dqp != (xfs_dquot_t *)&(xfs_Gqm->qm_dqfreelist); ) {
273 nextdqp = dqp->dq_flnext;
274 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
275 ASSERT(dqp->q_mount == NULL);
276 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
277 ASSERT(dqp->HL_PREVP == NULL);
278 ASSERT(dqp->MPL_PREVP == NULL);
279 XQM_FREELIST_REMOVE(dqp);
281 xfs_qm_dqdestroy(dqp);
287 xfs_qm_freelist_unlock(xfs_Gqm);
290 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
293 XFS_QM_LOCK(xfs_Gqm);
294 XFS_QM_RELE(xfs_Gqm);
295 if (xfs_Gqm->qm_nrefs == 0) {
296 xfs_qm_destroy(xfs_Gqm);
299 XFS_QM_UNLOCK(xfs_Gqm);
303 * This is called at mount time from xfs_mountfs to initialize the quotainfo
304 * structure and start the global quotamanager (xfs_Gqm) if it hasn't done
305 * so already. Note that the superblock has not been read in yet.
308 xfs_qm_mount_quotainit(
313 * User or group quotas has to be on.
315 ASSERT(flags & (XFSMNT_UQUOTA | XFSMNT_GQUOTA));
318 * Initialize the flags in the mount structure. From this point
319 * onwards we look at m_qflags to figure out if quotas's ON/OFF, etc.
320 * Note that we enforce nothing if accounting is off.
321 * ie. XFSMNT_*QUOTA must be ON for XFSMNT_*QUOTAENF.
322 * It isn't necessary to take the quotaoff lock to do this; this is
325 if (flags & XFSMNT_UQUOTA) {
326 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
327 if (flags & XFSMNT_UQUOTAENF)
328 mp->m_qflags |= XFS_UQUOTA_ENFD;
330 if (flags & XFSMNT_GQUOTA) {
331 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
332 if (flags & XFSMNT_GQUOTAENF)
333 mp->m_qflags |= XFS_GQUOTA_ENFD;
338 * Just destroy the quotainfo structure.
341 xfs_qm_unmount_quotadestroy(
345 xfs_qm_destroy_quotainfo(mp);
350 * This is called from xfs_mountfs to start quotas and initialize all
351 * necessary data structures like quotainfo. This is also responsible for
352 * running a quotacheck as necessary. We are guaranteed that the superblock
353 * is consistently read in at this point.
365 * If a file system had quotas running earlier, but decided to
366 * mount without -o quota/uquota/gquota options, revoke the
367 * quotachecked license, and bail out.
369 if (! XFS_IS_QUOTA_ON(mp) &&
370 (mp->m_sb.sb_qflags & (XFS_UQUOTA_ACCT|XFS_GQUOTA_ACCT))) {
376 * If quotas on realtime volumes is not supported, we disable
377 * quotas immediately.
379 if (mp->m_sb.sb_rextents) {
381 "Cannot turn on quotas for realtime filesystem %s",
387 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
388 cmn_err(CE_NOTE, "Attempting to turn on disk quotas.");
391 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
393 * Allocate the quotainfo structure inside the mount struct, and
394 * create quotainode(s), and change/rev superblock if necessary.
396 if ((error = xfs_qm_init_quotainfo(mp))) {
398 * We must turn off quotas.
400 ASSERT(mp->m_quotainfo == NULL);
405 * If any of the quotas are not consistent, do a quotacheck.
407 if (XFS_QM_NEED_QUOTACHECK(mp) &&
408 !(mfsi_flags & XFS_MFSI_NO_QUOTACHECK)) {
410 cmn_err(CE_NOTE, "Doing a quotacheck. Please wait.");
412 if ((error = xfs_qm_quotacheck(mp))) {
413 /* Quotacheck has failed and quotas have
416 return XFS_ERROR(error);
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, 0, 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. If the quotacheck fails, disable quotas.
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 +++");
1977 cmn_err(CE_WARN, "XFS quotacheck %s: Unsuccessful (Error %d): "
1978 "Disabling quotas.",
1979 mp->m_fsname, error);
1981 * We must turn off quotas.
1983 ASSERT(mp->m_quotainfo != NULL);
1984 ASSERT(xfs_Gqm != NULL);
1985 xfs_qm_destroy_quotainfo(mp);
1986 xfs_mount_reset_sbqflags(mp);
1988 cmn_err(CE_NOTE, "XFS quotacheck %s: Done.", mp->m_fsname);
1994 * This is called after the superblock has been read in and we're ready to
1995 * iget the quota inodes.
1998 xfs_qm_init_quotainos(
2001 xfs_inode_t *uip, *gip;
2006 ASSERT(mp->m_quotainfo);
2012 * Get the uquota and gquota inodes
2014 if (XFS_SB_VERSION_HASQUOTA(&mp->m_sb)) {
2015 if (XFS_IS_UQUOTA_ON(mp) &&
2016 mp->m_sb.sb_uquotino != NULLFSINO) {
2017 ASSERT(mp->m_sb.sb_uquotino > 0);
2018 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
2020 return XFS_ERROR(error);
2022 if (XFS_IS_GQUOTA_ON(mp) &&
2023 mp->m_sb.sb_gquotino != NULLFSINO) {
2024 ASSERT(mp->m_sb.sb_gquotino > 0);
2025 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
2028 VN_RELE(XFS_ITOV(uip));
2029 return XFS_ERROR(error);
2033 flags |= XFS_QMOPT_SBVERSION;
2034 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
2035 XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
2039 * Create the two inodes, if they don't exist already. The changes
2040 * made above will get added to a transaction and logged in one of
2041 * the qino_alloc calls below. If the device is readonly,
2042 * temporarily switch to read-write to do this.
2044 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
2045 if ((error = xfs_qm_qino_alloc(mp, &uip,
2046 sbflags | XFS_SB_UQUOTINO,
2047 flags | XFS_QMOPT_UQUOTA)))
2048 return XFS_ERROR(error);
2050 flags &= ~XFS_QMOPT_SBVERSION;
2052 if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
2053 if ((error = xfs_qm_qino_alloc(mp, &gip,
2054 sbflags | XFS_SB_GQUOTINO,
2055 flags | XFS_QMOPT_GQUOTA))) {
2057 VN_RELE(XFS_ITOV(uip));
2059 return XFS_ERROR(error);
2063 XFS_QI_UQIP(mp) = uip;
2064 XFS_QI_GQIP(mp) = gip;
2071 * Traverse the freelist of dquots and attempt to reclaim a maximum of
2072 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
2073 * favor the lookup function ...
2074 * XXXsup merge this with qm_reclaim_one().
2077 xfs_qm_shake_freelist(
2082 xfs_dquot_t *dqp, *nextdqp;
2094 cmn_err(CE_DEBUG, "Shake free 0x%x", howmany);
2096 /* lock order is : hashchainlock, freelistlock, mplistlock */
2098 xfs_qm_freelist_lock(xfs_Gqm);
2100 for (dqp = xfs_Gqm->qm_dqfreelist.qh_next;
2101 ((dqp != (xfs_dquot_t *) &xfs_Gqm->qm_dqfreelist) &&
2102 nreclaimed < howmany); ) {
2106 * We are racing with dqlookup here. Naturally we don't
2107 * want to reclaim a dquot that lookup wants.
2109 if (dqp->dq_flags & XFS_DQ_WANT) {
2111 xfs_qm_freelist_unlock(xfs_Gqm);
2112 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
2113 return (nreclaimed);
2114 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
2119 * If the dquot is inactive, we are assured that it is
2120 * not on the mplist or the hashlist, and that makes our
2123 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
2124 ASSERT(dqp->q_mount == NULL);
2125 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
2126 ASSERT(dqp->HL_PREVP == NULL);
2127 ASSERT(dqp->MPL_PREVP == NULL);
2128 XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
2129 nextdqp = dqp->dq_flnext;
2133 ASSERT(dqp->MPL_PREVP);
2135 * Try to grab the flush lock. If this dquot is in the process of
2136 * getting flushed to disk, we don't want to reclaim it.
2138 if (! xfs_qm_dqflock_nowait(dqp)) {
2140 dqp = dqp->dq_flnext;
2145 * We have the flush lock so we know that this is not in the
2146 * process of being flushed. So, if this is dirty, flush it
2147 * DELWRI so that we don't get a freelist infested with
2150 if (XFS_DQ_IS_DIRTY(dqp)) {
2151 xfs_dqtrace_entry(dqp, "DQSHAKE: DQDIRTY");
2153 * We flush it delayed write, so don't bother
2154 * releasing the mplock.
2156 (void) xfs_qm_dqflush(dqp, XFS_QMOPT_DELWRI);
2157 xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
2158 dqp = dqp->dq_flnext;
2162 * We're trying to get the hashlock out of order. This races
2163 * with dqlookup; so, we giveup and goto the next dquot if
2164 * we couldn't get the hashlock. This way, we won't starve
2165 * a dqlookup process that holds the hashlock that is
2166 * waiting for the freelist lock.
2168 if (! xfs_qm_dqhashlock_nowait(dqp)) {
2171 dqp = dqp->dq_flnext;
2175 * This races with dquot allocation code as well as dqflush_all
2176 * and reclaim code. So, if we failed to grab the mplist lock,
2177 * giveup everything and start over.
2181 if (! xfs_qm_mplist_nowait(dqp->q_mount)) {
2182 /* XXX put a sentinel so that we can come back here */
2185 XFS_DQ_HASH_UNLOCK(hash);
2186 xfs_qm_freelist_unlock(xfs_Gqm);
2187 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
2188 return (nreclaimed);
2191 xfs_dqtrace_entry(dqp, "DQSHAKE: UNLINKING");
2193 cmn_err(CE_DEBUG, "Shake 0x%p, ID 0x%x\n",
2194 dqp, INT_GET(dqp->q_core.d_id, ARCH_CONVERT));
2196 ASSERT(dqp->q_nrefs == 0);
2197 nextdqp = dqp->dq_flnext;
2198 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp->q_mount)), dqp);
2199 XQM_HASHLIST_REMOVE(hash, dqp);
2201 xfs_qm_mplist_unlock(dqp->q_mount);
2202 XFS_DQ_HASH_UNLOCK(hash);
2205 XQM_FREELIST_REMOVE(dqp);
2208 XQM_STATS_INC(xqmstats.xs_qm_dqshake_reclaims);
2209 xfs_qm_dqdestroy(dqp);
2212 xfs_qm_freelist_unlock(xfs_Gqm);
2213 return (nreclaimed);
2218 * The kmem_shake interface is invoked when memory is running low.
2222 xfs_qm_shake(int nr_to_scan, unsigned int gfp_mask)
2224 int ndqused, nfree, n;
2226 if (!kmem_shake_allow(gfp_mask))
2231 nfree = xfs_Gqm->qm_dqfreelist.qh_nelems; /* free dquots */
2232 /* incore dquots in all f/s's */
2233 ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;
2235 ASSERT(ndqused >= 0);
2237 if (nfree <= ndqused && nfree < ndquot)
2240 ndqused *= xfs_Gqm->qm_dqfree_ratio; /* target # of free dquots */
2241 n = nfree - ndqused - ndquot; /* # over target */
2243 return xfs_qm_shake_freelist(MAX(nfree, n));
2248 * Just pop the least recently used dquot off the freelist and
2249 * recycle it. The returned dquot is locked.
2251 STATIC xfs_dquot_t *
2252 xfs_qm_dqreclaim_one(void)
2254 xfs_dquot_t *dqpout;
2263 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
2265 xfs_qm_freelist_lock(xfs_Gqm);
2267 FOREACH_DQUOT_IN_FREELIST(dqp, &(xfs_Gqm->qm_dqfreelist)) {
2271 * We are racing with dqlookup here. Naturally we don't
2272 * want to reclaim a dquot that lookup wants. We release the
2273 * freelist lock and start over, so that lookup will grab
2274 * both the dquot and the freelistlock.
2276 if (dqp->dq_flags & XFS_DQ_WANT) {
2277 ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE));
2278 xfs_dqtrace_entry(dqp, "DQRECLAIM: DQWANT");
2280 xfs_qm_freelist_unlock(xfs_Gqm);
2281 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
2283 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
2288 * If the dquot is inactive, we are assured that it is
2289 * not on the mplist or the hashlist, and that makes our
2292 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
2293 ASSERT(dqp->q_mount == NULL);
2294 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
2295 ASSERT(dqp->HL_PREVP == NULL);
2296 ASSERT(dqp->MPL_PREVP == NULL);
2297 XQM_FREELIST_REMOVE(dqp);
2300 XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
2304 ASSERT(dqp->q_hash);
2305 ASSERT(dqp->MPL_PREVP);
2308 * Try to grab the flush lock. If this dquot is in the process of
2309 * getting flushed to disk, we don't want to reclaim it.
2311 if (! xfs_qm_dqflock_nowait(dqp)) {
2317 * We have the flush lock so we know that this is not in the
2318 * process of being flushed. So, if this is dirty, flush it
2319 * DELWRI so that we don't get a freelist infested with
2322 if (XFS_DQ_IS_DIRTY(dqp)) {
2323 xfs_dqtrace_entry(dqp, "DQRECLAIM: DQDIRTY");
2325 * We flush it delayed write, so don't bother
2326 * releasing the freelist lock.
2328 (void) xfs_qm_dqflush(dqp, XFS_QMOPT_DELWRI);
2329 xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
2333 if (! xfs_qm_mplist_nowait(dqp->q_mount)) {
2339 if (! xfs_qm_dqhashlock_nowait(dqp))
2342 ASSERT(dqp->q_nrefs == 0);
2343 xfs_dqtrace_entry(dqp, "DQRECLAIM: UNLINKING");
2344 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp->q_mount)), dqp);
2345 XQM_HASHLIST_REMOVE(dqp->q_hash, dqp);
2346 XQM_FREELIST_REMOVE(dqp);
2348 XFS_DQ_HASH_UNLOCK(dqp->q_hash);
2350 xfs_qm_mplist_unlock(dqp->q_mount);
2357 xfs_qm_freelist_unlock(xfs_Gqm);
2362 /*------------------------------------------------------------------*/
2365 * Return a new incore dquot. Depending on the number of
2366 * dquots in the system, we either allocate a new one on the kernel heap,
2367 * or reclaim a free one.
2368 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2369 * to reclaim an existing one from the freelist.
2372 xfs_qm_dqalloc_incore(
2373 xfs_dquot_t **O_dqpp)
2378 * Check against high water mark to see if we want to pop
2379 * a nincompoop dquot off the freelist.
2381 if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
2383 * Try to recycle a dquot from the freelist.
2385 if ((dqp = xfs_qm_dqreclaim_one())) {
2386 XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
2388 * Just zero the core here. The rest will get
2389 * reinitialized by caller. XXX we shouldn't even
2392 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
2396 XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
2400 * Allocate a brand new dquot on the kernel heap and return it
2401 * to the caller to initialize.
2403 ASSERT(xfs_Gqm->qm_dqzone != NULL);
2404 *O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
2405 atomic_inc(&xfs_Gqm->qm_totaldquots);
2412 * Start a transaction and write the incore superblock changes to
2413 * disk. flags parameter indicates which fields have changed.
2416 xfs_qm_write_sb_changes(
2424 cmn_err(CE_NOTE, "Writing superblock quota changes :%s", mp->m_fsname);
2426 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
2427 if ((error = xfs_trans_reserve(tp, 0,
2428 mp->m_sb.sb_sectsize + 128, 0,
2430 XFS_DEFAULT_LOG_COUNT))) {
2431 xfs_trans_cancel(tp, 0);
2435 xfs_mod_sb(tp, flags);
2436 (void) xfs_trans_commit(tp, 0, NULL);
2442 /* --------------- utility functions for vnodeops ---------------- */
2446 * Given an inode, a uid and gid (from cred_t) make sure that we have
2447 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2448 * quotas by creating this file.
2449 * This also attaches dquot(s) to the given inode after locking it,
2450 * and returns the dquots corresponding to the uid and/or gid.
2452 * in : inode (unlocked)
2453 * out : udquot, gdquot with references taken and unlocked
2462 xfs_dquot_t **O_udqpp,
2463 xfs_dquot_t **O_gdqpp)
2466 xfs_dquot_t *uq, *gq;
2469 if (!XFS_IS_QUOTA_ON(mp))
2472 lockflags = XFS_ILOCK_EXCL;
2473 xfs_ilock(ip, lockflags);
2475 if ((flags & XFS_QMOPT_INHERIT) &&
2476 XFS_INHERIT_GID(ip, XFS_MTOVFS(mp)))
2477 gid = ip->i_d.di_gid;
2480 * Attach the dquot(s) to this inode, doing a dquot allocation
2481 * if necessary. The dquot(s) will not be locked.
2483 if (XFS_NOT_DQATTACHED(mp, ip)) {
2484 if ((error = xfs_qm_dqattach(ip, XFS_QMOPT_DQALLOC |
2485 XFS_QMOPT_ILOCKED))) {
2486 xfs_iunlock(ip, lockflags);
2492 if ((flags & XFS_QMOPT_UQUOTA) &&
2493 XFS_IS_UQUOTA_ON(mp)) {
2494 if (ip->i_d.di_uid != uid) {
2496 * What we need is the dquot that has this uid, and
2497 * if we send the inode to dqget, the uid of the inode
2498 * takes priority over what's sent in the uid argument.
2499 * We must unlock inode here before calling dqget if
2500 * we're not sending the inode, because otherwise
2501 * we'll deadlock by doing trans_reserve while
2504 xfs_iunlock(ip, lockflags);
2505 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
2510 ASSERT(error != ENOENT);
2514 * Get the ilock in the right order.
2517 lockflags = XFS_ILOCK_SHARED;
2518 xfs_ilock(ip, lockflags);
2521 * Take an extra reference, because we'll return
2524 ASSERT(ip->i_udquot);
2531 if ((flags & XFS_QMOPT_GQUOTA) &&
2532 XFS_IS_GQUOTA_ON(mp)) {
2533 if (ip->i_d.di_gid != gid) {
2534 xfs_iunlock(ip, lockflags);
2535 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
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 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
2584 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
2589 ASSERT(prevdq != newdq);
2591 xfs_trans_mod_dquot(tp, prevdq,
2592 XFS_TRANS_DQ_BCOUNT,
2593 -(ip->i_d.di_nblocks));
2594 xfs_trans_mod_dquot(tp, prevdq,
2595 XFS_TRANS_DQ_ICOUNT,
2598 /* the sparkling new dquot */
2599 xfs_trans_mod_dquot(tp, newdq,
2600 XFS_TRANS_DQ_BCOUNT,
2601 ip->i_d.di_nblocks);
2602 xfs_trans_mod_dquot(tp, newdq,
2603 XFS_TRANS_DQ_ICOUNT,
2607 * Take an extra reference, because the inode
2608 * is going to keep this dquot pointer even
2609 * after the trans_commit.
2613 xfs_dqunlock(newdq);
2620 * Quota reservations for setattr(AT_UID|AT_GID).
2623 xfs_qm_vop_chown_reserve(
2633 xfs_dquot_t *unresudq, *unresgdq, *delblksudq, *delblksgdq;
2635 ASSERT(XFS_ISLOCKED_INODE(ip));
2637 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2639 delblks = ip->i_delayed_blks;
2640 delblksudq = delblksgdq = unresudq = unresgdq = NULL;
2642 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2643 ip->i_d.di_uid != (uid_t)INT_GET(udqp->q_core.d_id, ARCH_CONVERT)) {
2646 * If there are delayed allocation blocks, then we have to
2647 * unreserve those from the old dquot, and add them to the
2651 ASSERT(ip->i_udquot);
2652 unresudq = ip->i_udquot;
2655 if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp &&
2656 ip->i_d.di_gid != INT_GET(gdqp->q_core.d_id, ARCH_CONVERT)) {
2659 ASSERT(ip->i_gdquot);
2660 unresgdq = ip->i_gdquot;
2664 if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2665 delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
2666 flags | XFS_QMOPT_RES_REGBLKS)))
2670 * Do the delayed blks reservations/unreservations now. Since, these
2671 * are done without the help of a transaction, if a reservation fails
2672 * its previous reservations won't be automatically undone by trans
2673 * code. So, we have to do it manually here.
2677 * Do the reservations first. Unreservation can't fail.
2679 ASSERT(delblksudq || delblksgdq);
2680 ASSERT(unresudq || unresgdq);
2681 if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2682 delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
2683 flags | XFS_QMOPT_RES_REGBLKS)))
2685 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2686 unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
2687 XFS_QMOPT_RES_REGBLKS);
2694 xfs_qm_vop_rename_dqattach(
2695 xfs_inode_t **i_tab)
2703 if (! XFS_IS_QUOTA_ON(ip->i_mount))
2706 if (XFS_NOT_DQATTACHED(ip->i_mount, ip)) {
2707 error = xfs_qm_dqattach(ip, 0);
2711 for (i = 1; (i < 4 && i_tab[i]); i++) {
2713 * Watch out for duplicate entries in the table.
2715 if ((ip = i_tab[i]) != i_tab[i-1]) {
2716 if (XFS_NOT_DQATTACHED(ip->i_mount, ip)) {
2717 error = xfs_qm_dqattach(ip, 0);
2727 xfs_qm_vop_dqattach_and_dqmod_newinode(
2733 if (!XFS_IS_QUOTA_ON(tp->t_mountp))
2736 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
2737 ASSERT(XFS_IS_QUOTA_RUNNING(tp->t_mountp));
2743 ASSERT(ip->i_udquot == NULL);
2744 ip->i_udquot = udqp;
2745 ASSERT(ip->i_d.di_uid == INT_GET(udqp->q_core.d_id, ARCH_CONVERT));
2746 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2752 ASSERT(ip->i_gdquot == NULL);
2753 ip->i_gdquot = gdqp;
2754 ASSERT(ip->i_d.di_gid == INT_GET(gdqp->q_core.d_id, ARCH_CONVERT));
2755 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
2759 /* ------------- list stuff -----------------*/
2761 xfs_qm_freelist_init(xfs_frlist_t *ql)
2763 ql->qh_next = ql->qh_prev = (xfs_dquot_t *) ql;
2764 mutex_init(&ql->qh_lock, MUTEX_DEFAULT, "dqf");
2770 xfs_qm_freelist_destroy(xfs_frlist_t *ql)
2772 xfs_dquot_t *dqp, *nextdqp;
2774 mutex_lock(&ql->qh_lock, PINOD);
2775 for (dqp = ql->qh_next;
2776 dqp != (xfs_dquot_t *)ql; ) {
2778 nextdqp = dqp->dq_flnext;
2780 cmn_err(CE_DEBUG, "FREELIST destroy 0x%p", dqp);
2782 XQM_FREELIST_REMOVE(dqp);
2784 xfs_qm_dqdestroy(dqp);
2788 * Don't bother about unlocking.
2790 mutex_destroy(&ql->qh_lock);
2792 ASSERT(ql->qh_nelems == 0);
2796 xfs_qm_freelist_insert(xfs_frlist_t *ql, xfs_dquot_t *dq)
2798 dq->dq_flnext = ql->qh_next;
2799 dq->dq_flprev = (xfs_dquot_t *)ql;
2801 dq->dq_flnext->dq_flprev = dq;
2802 xfs_Gqm->qm_dqfreelist.qh_nelems++;
2803 xfs_Gqm->qm_dqfreelist.qh_version++;
2807 xfs_qm_freelist_unlink(xfs_dquot_t *dq)
2809 xfs_dquot_t *next = dq->dq_flnext;
2810 xfs_dquot_t *prev = dq->dq_flprev;
2812 next->dq_flprev = prev;
2813 prev->dq_flnext = next;
2814 dq->dq_flnext = dq->dq_flprev = dq;
2815 xfs_Gqm->qm_dqfreelist.qh_nelems--;
2816 xfs_Gqm->qm_dqfreelist.qh_version++;
2820 xfs_qm_freelist_append(xfs_frlist_t *ql, xfs_dquot_t *dq)
2822 xfs_qm_freelist_insert((xfs_frlist_t *)ql->qh_prev, dq);
2826 xfs_qm_dqhashlock_nowait(
2831 locked = mutex_trylock(&((dqp)->q_hash->qh_lock));
2836 xfs_qm_freelist_lock_nowait(
2841 locked = mutex_trylock(&(xqm->qm_dqfreelist.qh_lock));
2846 xfs_qm_mplist_nowait(
2851 ASSERT(mp->m_quotainfo);
2852 locked = mutex_trylock(&(XFS_QI_MPLLOCK(mp)));