2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "xfs_trans.h"
26 #include "xfs_alloc.h"
27 #include "xfs_quota.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_ialloc.h"
34 #include "xfs_itable.h"
35 #include "xfs_rtalloc.h"
36 #include "xfs_error.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_trans_space.h"
41 #include "xfs_utils.h"
43 #include "xfs_trace.h"
46 * The global quota manager. There is only one of these for the entire
47 * system, _not_ one per file system. XQM keeps track of the overall
48 * quota functionality, including maintaining the freelist and hash
51 struct mutex xfs_Gqm_lock;
52 struct xfs_qm *xfs_Gqm;
55 kmem_zone_t *qm_dqzone;
56 kmem_zone_t *qm_dqtrxzone;
58 static cred_t xfs_zerocr;
60 STATIC void xfs_qm_list_init(xfs_dqlist_t *, char *, int);
61 STATIC void xfs_qm_list_destroy(xfs_dqlist_t *);
63 STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
64 STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
65 STATIC int xfs_qm_shake(struct shrinker *, int, gfp_t);
67 static struct shrinker xfs_qm_shaker = {
68 .shrink = xfs_qm_shake,
69 .seeks = DEFAULT_SEEKS,
73 extern struct mutex qcheck_lock;
78 xfs_qm_dquot_list_print(
84 list_for_each_entry(dqp, &mp->m_quotainfo->qi_dqlist_lock, qi_mplist) {
85 cmn_err(CE_DEBUG, " %d. \"%d (%s)\" "
86 "bcnt = %lld, icnt = %lld, refs = %d",
87 i++, be32_to_cpu(dqp->q_core.d_id),
88 DQFLAGTO_TYPESTR(dqp),
89 (long long)be64_to_cpu(dqp->q_core.d_bcount),
90 (long long)be64_to_cpu(dqp->q_core.d_icount),
95 static void xfs_qm_dquot_list_print(struct xfs_mount *mp) { }
99 * Initialize the XQM structure.
100 * Note that there is not one quota manager per file system.
102 STATIC struct xfs_qm *
105 xfs_dqhash_t *udqhash, *gdqhash;
111 * Initialize the dquot hash tables.
113 udqhash = kmem_zalloc_greedy(&hsize,
114 XFS_QM_HASHSIZE_LOW * sizeof(xfs_dqhash_t),
115 XFS_QM_HASHSIZE_HIGH * sizeof(xfs_dqhash_t));
119 gdqhash = kmem_zalloc_large(hsize);
121 goto out_free_udqhash;
123 hsize /= sizeof(xfs_dqhash_t);
126 xqm = kmem_zalloc(sizeof(xfs_qm_t), KM_SLEEP);
127 xqm->qm_dqhashmask = hsize - 1;
128 xqm->qm_usr_dqhtable = udqhash;
129 xqm->qm_grp_dqhtable = gdqhash;
130 ASSERT(xqm->qm_usr_dqhtable != NULL);
131 ASSERT(xqm->qm_grp_dqhtable != NULL);
133 for (i = 0; i < hsize; i++) {
134 xfs_qm_list_init(&(xqm->qm_usr_dqhtable[i]), "uxdqh", i);
135 xfs_qm_list_init(&(xqm->qm_grp_dqhtable[i]), "gxdqh", i);
139 * Freelist of all dquots of all file systems
141 INIT_LIST_HEAD(&xqm->qm_dqfrlist);
142 xqm->qm_dqfrlist_cnt = 0;
143 mutex_init(&xqm->qm_dqfrlist_lock);
146 * dquot zone. we register our own low-memory callback.
149 xqm->qm_dqzone = kmem_zone_init(sizeof(xfs_dquot_t),
151 qm_dqzone = xqm->qm_dqzone;
153 xqm->qm_dqzone = qm_dqzone;
155 register_shrinker(&xfs_qm_shaker);
158 * The t_dqinfo portion of transactions.
161 xqm->qm_dqtrxzone = kmem_zone_init(sizeof(xfs_dquot_acct_t),
163 qm_dqtrxzone = xqm->qm_dqtrxzone;
165 xqm->qm_dqtrxzone = qm_dqtrxzone;
167 atomic_set(&xqm->qm_totaldquots, 0);
168 xqm->qm_dqfree_ratio = XFS_QM_DQFREE_RATIO;
171 mutex_init(&qcheck_lock);
176 kmem_free_large(udqhash);
182 * Destroy the global quota manager when its reference count goes to zero.
188 struct xfs_dquot *dqp, *n;
192 ASSERT(xqm->qm_nrefs == 0);
193 unregister_shrinker(&xfs_qm_shaker);
194 hsize = xqm->qm_dqhashmask + 1;
195 for (i = 0; i < hsize; i++) {
196 xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
197 xfs_qm_list_destroy(&(xqm->qm_grp_dqhtable[i]));
199 kmem_free_large(xqm->qm_usr_dqhtable);
200 kmem_free_large(xqm->qm_grp_dqhtable);
201 xqm->qm_usr_dqhtable = NULL;
202 xqm->qm_grp_dqhtable = NULL;
203 xqm->qm_dqhashmask = 0;
206 mutex_lock(&xqm->qm_dqfrlist_lock);
207 list_for_each_entry_safe(dqp, n, &xqm->qm_dqfrlist, q_freelist) {
210 cmn_err(CE_DEBUG, "FREELIST destroy 0x%p", dqp);
212 list_del_init(&dqp->q_freelist);
213 xfs_Gqm->qm_dqfrlist_cnt--;
215 xfs_qm_dqdestroy(dqp);
217 mutex_unlock(&xqm->qm_dqfrlist_lock);
218 mutex_destroy(&xqm->qm_dqfrlist_lock);
220 mutex_destroy(&qcheck_lock);
226 * Called at mount time to let XQM know that another file system is
227 * starting quotas. This isn't crucial information as the individual mount
228 * structures are pretty independent, but it helps the XQM keep a
229 * global view of what's going on.
233 xfs_qm_hold_quotafs_ref(
234 struct xfs_mount *mp)
237 * Need to lock the xfs_Gqm structure for things like this. For example,
238 * the structure could disappear between the entry to this routine and
239 * a HOLD operation if not locked.
241 mutex_lock(&xfs_Gqm_lock);
244 xfs_Gqm = xfs_Gqm_init();
246 mutex_unlock(&xfs_Gqm_lock);
252 * We can keep a list of all filesystems with quotas mounted for
253 * debugging and statistical purposes, but ...
254 * Just take a reference and get out.
257 mutex_unlock(&xfs_Gqm_lock);
264 * Release the reference that a filesystem took at mount time,
265 * so that we know when we need to destroy the entire quota manager.
269 xfs_qm_rele_quotafs_ref(
270 struct xfs_mount *mp)
272 xfs_dquot_t *dqp, *n;
275 ASSERT(xfs_Gqm->qm_nrefs > 0);
278 * Go thru the freelist and destroy all inactive dquots.
280 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
282 list_for_each_entry_safe(dqp, n, &xfs_Gqm->qm_dqfrlist, q_freelist) {
284 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
285 ASSERT(dqp->q_mount == NULL);
286 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
287 ASSERT(list_empty(&dqp->q_hashlist));
288 ASSERT(list_empty(&dqp->q_mplist));
289 list_del_init(&dqp->q_freelist);
290 xfs_Gqm->qm_dqfrlist_cnt--;
292 xfs_qm_dqdestroy(dqp);
297 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
300 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
303 mutex_lock(&xfs_Gqm_lock);
304 if (--xfs_Gqm->qm_nrefs == 0) {
305 xfs_qm_destroy(xfs_Gqm);
308 mutex_unlock(&xfs_Gqm_lock);
312 * Just destroy the quotainfo structure.
316 struct xfs_mount *mp)
318 if (mp->m_quotainfo) {
319 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
320 xfs_qm_destroy_quotainfo(mp);
326 * This is called from xfs_mountfs to start quotas and initialize all
327 * necessary data structures like quotainfo. This is also responsible for
328 * running a quotacheck as necessary. We are guaranteed that the superblock
329 * is consistently read in at this point.
331 * If we fail here, the mount will continue with quota turned off. We don't
332 * need to inidicate success or failure at all.
342 * If quotas on realtime volumes is not supported, we disable
343 * quotas immediately.
345 if (mp->m_sb.sb_rextents) {
347 "Cannot turn on quotas for realtime filesystem %s",
353 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
356 * Allocate the quotainfo structure inside the mount struct, and
357 * create quotainode(s), and change/rev superblock if necessary.
359 error = xfs_qm_init_quotainfo(mp);
362 * We must turn off quotas.
364 ASSERT(mp->m_quotainfo == NULL);
369 * If any of the quotas are not consistent, do a quotacheck.
371 if (XFS_QM_NEED_QUOTACHECK(mp)) {
372 error = xfs_qm_quotacheck(mp);
374 /* Quotacheck failed and disabled quotas. */
379 * If one type of quotas is off, then it will lose its
380 * quotachecked status, since we won't be doing accounting for
383 if (!XFS_IS_UQUOTA_ON(mp))
384 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
385 if (!(XFS_IS_GQUOTA_ON(mp) || XFS_IS_PQUOTA_ON(mp)))
386 mp->m_qflags &= ~XFS_OQUOTA_CHKD;
390 * We actually don't have to acquire the m_sb_lock at all.
391 * This can only be called from mount, and that's single threaded. XXX
393 spin_lock(&mp->m_sb_lock);
394 sbf = mp->m_sb.sb_qflags;
395 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
396 spin_unlock(&mp->m_sb_lock);
398 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
399 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
401 * We could only have been turning quotas off.
402 * We aren't in very good shape actually because
403 * the incore structures are convinced that quotas are
404 * off, but the on disk superblock doesn't know that !
406 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
407 xfs_fs_cmn_err(CE_ALERT, mp,
408 "XFS mount_quotas: Superblock update failed!");
413 xfs_fs_cmn_err(CE_WARN, mp,
414 "Failed to initialize disk quotas.");
419 if (XFS_IS_QUOTA_ON(mp))
420 xfs_qm_internalqcheck(mp);
425 * Called from the vfsops layer.
428 xfs_qm_unmount_quotas(
432 * Release the dquots that root inode, et al might be holding,
433 * before we flush quotas and blow away the quotainfo structure.
435 ASSERT(mp->m_rootip);
436 xfs_qm_dqdetach(mp->m_rootip);
438 xfs_qm_dqdetach(mp->m_rbmip);
440 xfs_qm_dqdetach(mp->m_rsumip);
443 * Release the quota inodes.
445 if (mp->m_quotainfo) {
446 if (mp->m_quotainfo->qi_uquotaip) {
447 IRELE(mp->m_quotainfo->qi_uquotaip);
448 mp->m_quotainfo->qi_uquotaip = NULL;
450 if (mp->m_quotainfo->qi_gquotaip) {
451 IRELE(mp->m_quotainfo->qi_gquotaip);
452 mp->m_quotainfo->qi_gquotaip = NULL;
458 * Flush all dquots of the given file system to disk. The dquots are
459 * _not_ purged from memory here, just their data written to disk.
463 struct xfs_mount *mp,
466 struct xfs_quotainfo *q = mp->m_quotainfo;
468 struct xfs_dquot *dqp;
476 mutex_lock(&q->qi_dqlist_lock);
477 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
479 if (! XFS_DQ_IS_DIRTY(dqp)) {
484 /* XXX a sentinel would be better */
485 recl = q->qi_dqreclaims;
486 if (!xfs_dqflock_nowait(dqp)) {
488 * If we can't grab the flush lock then check
489 * to see if the dquot has been flushed delayed
490 * write. If so, grab its buffer and send it
491 * out immediately. We'll be able to acquire
492 * the flush lock when the I/O completes.
494 xfs_qm_dqflock_pushbuf_wait(dqp);
497 * Let go of the mplist lock. We don't want to hold it
498 * across a disk write.
500 mutex_unlock(&q->qi_dqlist_lock);
501 error = xfs_qm_dqflush(dqp, sync_mode);
506 mutex_lock(&q->qi_dqlist_lock);
507 if (recl != q->qi_dqreclaims) {
508 mutex_unlock(&q->qi_dqlist_lock);
509 /* XXX restart limit */
514 mutex_unlock(&q->qi_dqlist_lock);
519 * Release the group dquot pointers the user dquots may be
520 * carrying around as a hint. mplist is locked on entry and exit.
523 xfs_qm_detach_gdquots(
524 struct xfs_mount *mp)
526 struct xfs_quotainfo *q = mp->m_quotainfo;
527 struct xfs_dquot *dqp, *gdqp;
531 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
532 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
534 if ((gdqp = dqp->q_gdquot)) {
536 dqp->q_gdquot = NULL;
542 * Can't hold the mplist lock across a dqput.
543 * XXXmust convert to marker based iterations here.
545 nrecl = q->qi_dqreclaims;
546 mutex_unlock(&q->qi_dqlist_lock);
549 mutex_lock(&q->qi_dqlist_lock);
550 if (nrecl != q->qi_dqreclaims)
557 * Go through all the incore dquots of this file system and take them
558 * off the mplist and hashlist, if the dquot type matches the dqtype
559 * parameter. This is used when turning off quota accounting for
560 * users and/or groups, as well as when the filesystem is unmounting.
564 struct xfs_mount *mp,
567 struct xfs_quotainfo *q = mp->m_quotainfo;
568 struct xfs_dquot *dqp, *n;
576 dqtype = (flags & XFS_QMOPT_UQUOTA) ? XFS_DQ_USER : 0;
577 dqtype |= (flags & XFS_QMOPT_PQUOTA) ? XFS_DQ_PROJ : 0;
578 dqtype |= (flags & XFS_QMOPT_GQUOTA) ? XFS_DQ_GROUP : 0;
580 mutex_lock(&q->qi_dqlist_lock);
583 * In the first pass through all incore dquots of this filesystem,
584 * we release the group dquot pointers the user dquots may be
585 * carrying around as a hint. We need to do this irrespective of
586 * what's being turned off.
588 xfs_qm_detach_gdquots(mp);
592 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
594 * Try to get rid of all of the unwanted dquots. The idea is to
595 * get them off mplist and hashlist, but leave them on freelist.
597 list_for_each_entry_safe(dqp, n, &q->qi_dqlist, q_mplist) {
599 * It's OK to look at the type without taking dqlock here.
600 * We're holding the mplist lock here, and that's needed for
603 if ((dqp->dq_flags & dqtype) == 0)
606 if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
607 nrecl = q->qi_dqreclaims;
608 mutex_unlock(&q->qi_dqlist_lock);
609 mutex_lock(&dqp->q_hash->qh_lock);
610 mutex_lock(&q->qi_dqlist_lock);
613 * XXXTheoretically, we can get into a very long
614 * ping pong game here.
615 * No one can be adding dquots to the mplist at
616 * this point, but somebody might be taking things off.
618 if (nrecl != q->qi_dqreclaims) {
619 mutex_unlock(&dqp->q_hash->qh_lock);
625 * Take the dquot off the mplist and hashlist. It may remain on
626 * freelist in INACTIVE state.
628 nmisses += xfs_qm_dqpurge(dqp);
630 mutex_unlock(&q->qi_dqlist_lock);
642 * Purge the dquot cache.
643 * None of the dquots should really be busy at this point.
645 if (mp->m_quotainfo) {
646 while ((ndquots = xfs_qm_dqpurge_int(mp, flags))) {
659 xfs_dquot_t *udqhint, /* hint */
660 xfs_dquot_t **IO_idqpp)
665 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
669 * See if we already have it in the inode itself. IO_idqpp is
670 * &i_udquot or &i_gdquot. This made the code look weird, but
671 * made the logic a lot simpler.
675 trace_xfs_dqattach_found(dqp);
680 * udqhint is the i_udquot field in inode, and is non-NULL only
681 * when the type arg is group/project. Its purpose is to save a
682 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
686 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
690 * No need to take dqlock to look at the id.
692 * The ID can't change until it gets reclaimed, and it won't
693 * be reclaimed as long as we have a ref from inode and we
696 dqp = udqhint->q_gdquot;
697 if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
700 ASSERT(*IO_idqpp == NULL);
704 xfs_dqunlock(udqhint);
709 * We can't hold a dquot lock when we call the dqget code.
710 * We'll deadlock in no time, because of (not conforming to)
711 * lock ordering - the inodelock comes before any dquot lock,
712 * and we may drop and reacquire the ilock in xfs_qm_dqget().
714 xfs_dqunlock(udqhint);
718 * Find the dquot from somewhere. This bumps the
719 * reference count of dquot and returns it locked.
720 * This can return ENOENT if dquot didn't exist on
721 * disk and we didn't ask it to allocate;
722 * ESRCH if quotas got turned off suddenly.
724 error = xfs_qm_dqget(ip->i_mount, ip, id, type, XFS_QMOPT_DOWARN, &dqp);
728 trace_xfs_dqattach_get(dqp);
731 * dqget may have dropped and re-acquired the ilock, but it guarantees
732 * that the dquot returned is the one that should go in the inode.
741 * Given a udquot and gdquot, attach a ptr to the group dquot in the
742 * udquot as a hint for future lookups. The idea sounds simple, but the
743 * execution isn't, because the udquot might have a group dquot attached
744 * already and getting rid of that gets us into lock ordering constraints.
745 * The process is complicated more by the fact that the dquots may or may not
746 * be locked on entry.
749 xfs_qm_dqattach_grouphint(
757 if ((tmp = udq->q_gdquot)) {
763 udq->q_gdquot = NULL;
765 * We can't keep any dqlocks when calling dqrele,
766 * because the freelist lock comes before dqlocks.
770 * we took a hard reference once upon a time in dqget,
771 * so give it back when the udquot no longer points at it
772 * dqput() does the unlocking of the dquot.
780 ASSERT(XFS_DQ_IS_LOCKED(udq));
784 ASSERT(XFS_DQ_IS_LOCKED(udq));
785 ASSERT(XFS_DQ_IS_LOCKED(gdq));
787 * Somebody could have attached a gdquot here,
788 * when we dropped the uqlock. If so, just do nothing.
790 if (udq->q_gdquot == NULL) {
801 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
803 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
804 * Inode may get unlocked and relocked in here, and the caller must deal with
808 xfs_qm_dqattach_locked(
812 xfs_mount_t *mp = ip->i_mount;
816 if (!XFS_IS_QUOTA_RUNNING(mp) ||
817 !XFS_IS_QUOTA_ON(mp) ||
818 !XFS_NOT_DQATTACHED(mp, ip) ||
819 ip->i_ino == mp->m_sb.sb_uquotino ||
820 ip->i_ino == mp->m_sb.sb_gquotino)
823 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
825 if (XFS_IS_UQUOTA_ON(mp)) {
826 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
827 flags & XFS_QMOPT_DQALLOC,
828 NULL, &ip->i_udquot);
834 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
835 if (XFS_IS_OQUOTA_ON(mp)) {
836 error = XFS_IS_GQUOTA_ON(mp) ?
837 xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
838 flags & XFS_QMOPT_DQALLOC,
839 ip->i_udquot, &ip->i_gdquot) :
840 xfs_qm_dqattach_one(ip, ip->i_d.di_projid, XFS_DQ_PROJ,
841 flags & XFS_QMOPT_DQALLOC,
842 ip->i_udquot, &ip->i_gdquot);
844 * Don't worry about the udquot that we may have
845 * attached above. It'll get detached, if not already.
853 * Attach this group quota to the user quota as a hint.
854 * This WON'T, in general, result in a thrash.
857 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
858 ASSERT(ip->i_udquot);
859 ASSERT(ip->i_gdquot);
862 * We may or may not have the i_udquot locked at this point,
863 * but this check is OK since we don't depend on the i_gdquot to
864 * be accurate 100% all the time. It is just a hint, and this
865 * will succeed in general.
867 if (ip->i_udquot->q_gdquot == ip->i_gdquot)
870 * Attach i_gdquot to the gdquot hint inside the i_udquot.
872 xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot);
878 if (XFS_IS_UQUOTA_ON(mp))
879 ASSERT(ip->i_udquot);
880 if (XFS_IS_OQUOTA_ON(mp))
881 ASSERT(ip->i_gdquot);
883 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
890 struct xfs_inode *ip,
895 xfs_ilock(ip, XFS_ILOCK_EXCL);
896 error = xfs_qm_dqattach_locked(ip, flags);
897 xfs_iunlock(ip, XFS_ILOCK_EXCL);
903 * Release dquots (and their references) if any.
904 * The inode should be locked EXCL except when this's called by
911 if (!(ip->i_udquot || ip->i_gdquot))
914 trace_xfs_dquot_dqdetach(ip);
916 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
917 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
919 xfs_qm_dqrele(ip->i_udquot);
923 xfs_qm_dqrele(ip->i_gdquot);
930 struct xfs_mount *mp,
933 struct xfs_quotainfo *q = mp->m_quotainfo;
935 struct xfs_dquot *dqp;
938 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
944 mutex_lock(&q->qi_dqlist_lock);
946 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
947 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
948 * when we have the mplist lock, we know that dquots will be consistent
949 * as long as we have it locked.
951 if (!XFS_IS_QUOTA_ON(mp)) {
952 mutex_unlock(&q->qi_dqlist_lock);
955 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
956 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
958 * If this is vfs_sync calling, then skip the dquots that
959 * don't 'seem' to be dirty. ie. don't acquire dqlock.
960 * This is very similar to what xfs_sync does with inodes.
962 if (flags & SYNC_TRYLOCK) {
963 if (!XFS_DQ_IS_DIRTY(dqp))
965 if (!xfs_qm_dqlock_nowait(dqp))
972 * Now, find out for sure if this dquot is dirty or not.
974 if (! XFS_DQ_IS_DIRTY(dqp)) {
979 /* XXX a sentinel would be better */
980 recl = q->qi_dqreclaims;
981 if (!xfs_dqflock_nowait(dqp)) {
982 if (flags & SYNC_TRYLOCK) {
987 * If we can't grab the flush lock then if the caller
988 * really wanted us to give this our best shot, so
989 * see if we can give a push to the buffer before we wait
990 * on the flush lock. At this point, we know that
991 * even though the dquot is being flushed,
992 * it has (new) dirty data.
994 xfs_qm_dqflock_pushbuf_wait(dqp);
997 * Let go of the mplist lock. We don't want to hold it
998 * across a disk write
1000 mutex_unlock(&q->qi_dqlist_lock);
1001 error = xfs_qm_dqflush(dqp, flags);
1003 if (error && XFS_FORCED_SHUTDOWN(mp))
1004 return 0; /* Need to prevent umount failure */
1008 mutex_lock(&q->qi_dqlist_lock);
1009 if (recl != q->qi_dqreclaims) {
1010 if (++restarts >= XFS_QM_SYNC_MAX_RESTARTS)
1013 mutex_unlock(&q->qi_dqlist_lock);
1018 mutex_unlock(&q->qi_dqlist_lock);
1023 * The hash chains and the mplist use the same xfs_dqhash structure as
1024 * their list head, but we can take the mplist qh_lock and one of the
1025 * hash qh_locks at the same time without any problem as they aren't
1028 static struct lock_class_key xfs_quota_mplist_class;
1031 * This initializes all the quota information that's kept in the
1035 xfs_qm_init_quotainfo(
1038 xfs_quotainfo_t *qinf;
1042 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1045 * Tell XQM that we exist as soon as possible.
1047 if ((error = xfs_qm_hold_quotafs_ref(mp))) {
1051 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
1054 * See if quotainodes are setup, and if not, allocate them,
1055 * and change the superblock accordingly.
1057 if ((error = xfs_qm_init_quotainos(mp))) {
1059 mp->m_quotainfo = NULL;
1063 INIT_LIST_HEAD(&qinf->qi_dqlist);
1064 mutex_init(&qinf->qi_dqlist_lock);
1065 lockdep_set_class(&qinf->qi_dqlist_lock, &xfs_quota_mplist_class);
1067 qinf->qi_dqreclaims = 0;
1069 /* mutex used to serialize quotaoffs */
1070 mutex_init(&qinf->qi_quotaofflock);
1072 /* Precalc some constants */
1073 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1074 ASSERT(qinf->qi_dqchunklen);
1075 qinf->qi_dqperchunk = BBTOB(qinf->qi_dqchunklen);
1076 do_div(qinf->qi_dqperchunk, sizeof(xfs_dqblk_t));
1078 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
1081 * We try to get the limits from the superuser's limits fields.
1082 * This is quite hacky, but it is standard quota practice.
1083 * We look at the USR dquot with id == 0 first, but if user quotas
1084 * are not enabled we goto the GRP dquot with id == 0.
1085 * We don't really care to keep separate default limits for user
1086 * and group quotas, at least not at this point.
1088 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)0,
1089 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
1090 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
1092 XFS_QMOPT_DQSUSER|XFS_QMOPT_DOWARN,
1095 xfs_disk_dquot_t *ddqp = &dqp->q_core;
1098 * The warnings and timers set the grace period given to
1099 * a user or group before he or she can not perform any
1100 * more writing. If it is zero, a default is used.
1102 qinf->qi_btimelimit = ddqp->d_btimer ?
1103 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
1104 qinf->qi_itimelimit = ddqp->d_itimer ?
1105 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
1106 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
1107 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
1108 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
1109 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
1110 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
1111 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
1112 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
1113 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
1114 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
1115 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
1116 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
1117 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
1118 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
1119 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
1122 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1123 * we don't want this dquot cached. We haven't done a
1124 * quotacheck yet, and quotacheck doesn't like incore dquots.
1126 xfs_qm_dqdestroy(dqp);
1128 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
1129 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
1130 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
1131 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
1132 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
1133 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
1141 * Gets called when unmounting a filesystem or when all quotas get
1143 * This purges the quota inodes, destroys locks and frees itself.
1146 xfs_qm_destroy_quotainfo(
1149 xfs_quotainfo_t *qi;
1151 qi = mp->m_quotainfo;
1153 ASSERT(xfs_Gqm != NULL);
1156 * Release the reference that XQM kept, so that we know
1157 * when the XQM structure should be freed. We cannot assume
1158 * that xfs_Gqm is non-null after this point.
1160 xfs_qm_rele_quotafs_ref(mp);
1162 ASSERT(list_empty(&qi->qi_dqlist));
1163 mutex_destroy(&qi->qi_dqlist_lock);
1165 if (qi->qi_uquotaip) {
1166 IRELE(qi->qi_uquotaip);
1167 qi->qi_uquotaip = NULL; /* paranoia */
1169 if (qi->qi_gquotaip) {
1170 IRELE(qi->qi_gquotaip);
1171 qi->qi_gquotaip = NULL;
1173 mutex_destroy(&qi->qi_quotaofflock);
1175 mp->m_quotainfo = NULL;
1180 /* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1189 mutex_init(&list->qh_lock);
1190 INIT_LIST_HEAD(&list->qh_list);
1191 list->qh_version = 0;
1192 list->qh_nelems = 0;
1196 xfs_qm_list_destroy(
1199 mutex_destroy(&(list->qh_lock));
1204 * Stripped down version of dqattach. This doesn't attach, or even look at the
1205 * dquots attached to the inode. The rationale is that there won't be any
1206 * attached at the time this is called from quotacheck.
1209 xfs_qm_dqget_noattach(
1211 xfs_dquot_t **O_udqpp,
1212 xfs_dquot_t **O_gdqpp)
1216 xfs_dquot_t *udqp, *gdqp;
1218 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1223 if (XFS_IS_UQUOTA_ON(mp)) {
1224 ASSERT(ip->i_udquot == NULL);
1226 * We want the dquot allocated if it doesn't exist.
1228 if ((error = xfs_qm_dqget(mp, ip, ip->i_d.di_uid, XFS_DQ_USER,
1229 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN,
1232 * Shouldn't be able to turn off quotas here.
1234 ASSERT(error != ESRCH);
1235 ASSERT(error != ENOENT);
1241 if (XFS_IS_OQUOTA_ON(mp)) {
1242 ASSERT(ip->i_gdquot == NULL);
1245 error = XFS_IS_GQUOTA_ON(mp) ?
1246 xfs_qm_dqget(mp, ip,
1247 ip->i_d.di_gid, XFS_DQ_GROUP,
1248 XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
1250 xfs_qm_dqget(mp, ip,
1251 ip->i_d.di_projid, XFS_DQ_PROJ,
1252 XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
1256 xfs_qm_dqrele(udqp);
1257 ASSERT(error != ESRCH);
1258 ASSERT(error != ENOENT);
1263 /* Reacquire the locks in the right order */
1265 if (! xfs_qm_dqlock_nowait(udqp)) {
1277 if (udqp) ASSERT(XFS_DQ_IS_LOCKED(udqp));
1278 if (gdqp) ASSERT(XFS_DQ_IS_LOCKED(gdqp));
1284 * Create an inode and return with a reference already taken, but unlocked
1285 * This is how we create quota inodes
1298 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
1299 if ((error = xfs_trans_reserve(tp,
1300 XFS_QM_QINOCREATE_SPACE_RES(mp),
1301 XFS_CREATE_LOG_RES(mp), 0,
1302 XFS_TRANS_PERM_LOG_RES,
1303 XFS_CREATE_LOG_COUNT))) {
1304 xfs_trans_cancel(tp, 0);
1308 if ((error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0,
1309 &xfs_zerocr, 0, 1, ip, &committed))) {
1310 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1316 * Keep an extra reference to this quota inode. This inode is
1317 * locked exclusively and joined to the transaction already.
1319 ASSERT(xfs_isilocked(*ip, XFS_ILOCK_EXCL));
1323 * Make the changes in the superblock, and log those too.
1324 * sbfields arg may contain fields other than *QUOTINO;
1325 * VERSIONNUM for example.
1327 spin_lock(&mp->m_sb_lock);
1328 if (flags & XFS_QMOPT_SBVERSION) {
1329 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1330 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1331 XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
1332 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1333 XFS_SB_GQUOTINO | XFS_SB_QFLAGS));
1335 xfs_sb_version_addquota(&mp->m_sb);
1336 mp->m_sb.sb_uquotino = NULLFSINO;
1337 mp->m_sb.sb_gquotino = NULLFSINO;
1339 /* qflags will get updated _after_ quotacheck */
1340 mp->m_sb.sb_qflags = 0;
1342 if (flags & XFS_QMOPT_UQUOTA)
1343 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1345 mp->m_sb.sb_gquotino = (*ip)->i_ino;
1346 spin_unlock(&mp->m_sb_lock);
1347 xfs_mod_sb(tp, sbfields);
1349 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
1350 xfs_fs_cmn_err(CE_ALERT, mp, "XFS qino_alloc failed!");
1358 xfs_qm_reset_dqcounts(
1364 xfs_disk_dquot_t *ddq;
1367 trace_xfs_reset_dqcounts(bp, _RET_IP_);
1370 * Reset all counters and timers. They'll be
1371 * started afresh by xfs_qm_quotacheck.
1374 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1375 do_div(j, sizeof(xfs_dqblk_t));
1376 ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
1378 ddq = (xfs_disk_dquot_t *)XFS_BUF_PTR(bp);
1379 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
1381 * Do a sanity check, and if needed, repair the dqblk. Don't
1382 * output any warnings because it's perfectly possible to
1383 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1385 (void) xfs_qm_dqcheck(ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1389 ddq->d_rtbcount = 0;
1392 ddq->d_rtbtimer = 0;
1395 ddq->d_rtbwarns = 0;
1396 ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
1405 xfs_filblks_t blkcnt,
1416 incr = (blkcnt > XFS_QM_MAX_DQCLUSTER_LOGSZ) ?
1417 XFS_QM_MAX_DQCLUSTER_LOGSZ : blkcnt;
1418 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1419 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1423 * Blkcnt arg can be a very big number, and might even be
1424 * larger than the log itself. So, we have to break it up into
1425 * manageable-sized transactions.
1426 * Note that we don't start a permanent transaction here; we might
1427 * not be able to get a log reservation for the whole thing up front,
1428 * and we don't really care to either, because we just discard
1429 * everything if we were to crash in the middle of this loop.
1432 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1433 XFS_FSB_TO_DADDR(mp, bno),
1434 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
1438 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1439 xfs_bdwrite(mp, bp);
1441 * goto the next block.
1444 firstid += mp->m_quotainfo->qi_dqperchunk;
1450 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1451 * caller supplied function for every chunk of dquots that we find.
1459 xfs_bmbt_irec_t *map;
1460 int i, nmaps; /* number of map entries */
1461 int error; /* return value */
1462 xfs_fileoff_t lblkno;
1463 xfs_filblks_t maxlblkcnt;
1465 xfs_fsblock_t rablkno;
1466 xfs_filblks_t rablkcnt;
1470 * This looks racy, but we can't keep an inode lock across a
1471 * trans_reserve. But, this gets called during quotacheck, and that
1472 * happens only at mount time which is single threaded.
1474 if (qip->i_d.di_nblocks == 0)
1477 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1480 maxlblkcnt = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1482 nmaps = XFS_DQITER_MAP_SIZE;
1484 * We aren't changing the inode itself. Just changing
1485 * some of its data. No new blocks are added here, and
1486 * the inode is never added to the transaction.
1488 xfs_ilock(qip, XFS_ILOCK_SHARED);
1489 error = xfs_bmapi(NULL, qip, lblkno,
1490 maxlblkcnt - lblkno,
1493 0, map, &nmaps, NULL);
1494 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1498 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1499 for (i = 0; i < nmaps; i++) {
1500 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1501 ASSERT(map[i].br_blockcount);
1504 lblkno += map[i].br_blockcount;
1506 if (map[i].br_startblock == HOLESTARTBLOCK)
1509 firstid = (xfs_dqid_t) map[i].br_startoff *
1510 mp->m_quotainfo->qi_dqperchunk;
1512 * Do a read-ahead on the next extent.
1514 if ((i+1 < nmaps) &&
1515 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1516 rablkcnt = map[i+1].br_blockcount;
1517 rablkno = map[i+1].br_startblock;
1518 while (rablkcnt--) {
1519 xfs_baread(mp->m_ddev_targp,
1520 XFS_FSB_TO_DADDR(mp, rablkno),
1521 mp->m_quotainfo->qi_dqchunklen);
1526 * Iterate thru all the blks in the extent and
1527 * reset the counters of all the dquots inside them.
1529 if ((error = xfs_qm_dqiter_bufs(mp,
1531 map[i].br_startblock,
1532 map[i].br_blockcount,
1540 } while (nmaps > 0);
1548 * Called by dqusage_adjust in doing a quotacheck.
1549 * Given the inode, and a dquot (either USR or GRP, doesn't matter),
1550 * this updates its incore copy as well as the buffer copy. This is
1551 * so that once the quotacheck is done, we can just log all the buffers,
1552 * as opposed to logging numerous updates to individual dquots.
1555 xfs_qm_quotacheck_dqadjust(
1560 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1562 trace_xfs_dqadjust(dqp);
1565 * Adjust the inode count and the block count to reflect this inode's
1568 be64_add_cpu(&dqp->q_core.d_icount, 1);
1569 dqp->q_res_icount++;
1571 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1572 dqp->q_res_bcount += nblks;
1575 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1576 dqp->q_res_rtbcount += rtblks;
1580 * Set default limits, adjust timers (since we changed usages)
1582 * There are no timers for the default values set in the root dquot.
1584 if (dqp->q_core.d_id) {
1585 xfs_qm_adjust_dqlimits(dqp->q_mount, &dqp->q_core);
1586 xfs_qm_adjust_dqtimers(dqp->q_mount, &dqp->q_core);
1589 dqp->dq_flags |= XFS_DQ_DIRTY;
1595 xfs_qcnt_t *O_rtblks)
1597 xfs_filblks_t rtblks; /* total rt blks */
1598 xfs_extnum_t idx; /* extent record index */
1599 xfs_ifork_t *ifp; /* inode fork pointer */
1600 xfs_extnum_t nextents; /* number of extent entries */
1603 ASSERT(XFS_IS_REALTIME_INODE(ip));
1604 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1605 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1606 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1610 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1611 for (idx = 0; idx < nextents; idx++)
1612 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1613 *O_rtblks = (xfs_qcnt_t)rtblks;
1618 * callback routine supplied to bulkstat(). Given an inumber, find its
1619 * dquots and update them to account for resources taken by that inode.
1623 xfs_qm_dqusage_adjust(
1624 xfs_mount_t *mp, /* mount point for filesystem */
1625 xfs_ino_t ino, /* inode number to get data for */
1626 void __user *buffer, /* not used */
1627 int ubsize, /* not used */
1628 int *ubused, /* not used */
1629 int *res) /* result code value */
1632 xfs_dquot_t *udqp, *gdqp;
1633 xfs_qcnt_t nblks, rtblks;
1636 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1639 * rootino must have its resources accounted for, not so with the quota
1642 if (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino) {
1643 *res = BULKSTAT_RV_NOTHING;
1644 return XFS_ERROR(EINVAL);
1648 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1649 * interface expects the inode to be exclusively locked because that's
1650 * the case in all other instances. It's OK that we do this because
1651 * quotacheck is done only at mount time.
1653 if ((error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip))) {
1654 *res = BULKSTAT_RV_NOTHING;
1659 * Obtain the locked dquots. In case of an error (eg. allocation
1660 * fails for ENOSPC), we return the negative of the error number
1661 * to bulkstat, so that it can get propagated to quotacheck() and
1662 * making us disable quotas for the file system.
1664 if ((error = xfs_qm_dqget_noattach(ip, &udqp, &gdqp))) {
1665 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1667 *res = BULKSTAT_RV_GIVEUP;
1672 if (! XFS_IS_REALTIME_INODE(ip)) {
1673 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks;
1676 * Walk thru the extent list and count the realtime blocks.
1678 if ((error = xfs_qm_get_rtblks(ip, &rtblks))) {
1679 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1685 *res = BULKSTAT_RV_GIVEUP;
1688 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1690 ASSERT(ip->i_delayed_blks == 0);
1693 * We can't release the inode while holding its dquot locks.
1694 * The inode can go into inactive and might try to acquire the dquotlocks.
1695 * So, just unlock here and do a vn_rele at the end.
1697 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1700 * Add the (disk blocks and inode) resources occupied by this
1701 * inode to its dquots. We do this adjustment in the incore dquot,
1702 * and also copy the changes to its buffer.
1703 * We don't care about putting these changes in a transaction
1704 * envelope because if we crash in the middle of a 'quotacheck'
1705 * we have to start from the beginning anyway.
1706 * Once we're done, we'll log all the dquot bufs.
1708 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1709 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1711 if (XFS_IS_UQUOTA_ON(mp)) {
1713 xfs_qm_quotacheck_dqadjust(udqp, nblks, rtblks);
1716 if (XFS_IS_OQUOTA_ON(mp)) {
1718 xfs_qm_quotacheck_dqadjust(gdqp, nblks, rtblks);
1722 * Now release the inode. This will send it to 'inactive', and
1723 * possibly even free blocks.
1730 *res = BULKSTAT_RV_DIDONE;
1735 * Walk thru all the filesystem inodes and construct a consistent view
1736 * of the disk quota world. If the quotacheck fails, disable quotas.
1742 int done, count, error;
1745 xfs_inode_t *uip, *gip;
1753 ASSERT(mp->m_quotainfo->qi_uquotaip || mp->m_quotainfo->qi_gquotaip);
1754 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1757 * There should be no cached dquots. The (simplistic) quotacheck
1758 * algorithm doesn't like that.
1760 ASSERT(list_empty(&mp->m_quotainfo->qi_dqlist));
1762 cmn_err(CE_NOTE, "XFS quotacheck %s: Please wait.", mp->m_fsname);
1765 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1766 * their counters to zero. We need a clean slate.
1767 * We don't log our changes till later.
1769 uip = mp->m_quotainfo->qi_uquotaip;
1771 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA);
1774 flags |= XFS_UQUOTA_CHKD;
1777 gip = mp->m_quotainfo->qi_gquotaip;
1779 error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
1780 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1783 flags |= XFS_OQUOTA_CHKD;
1788 * Iterate thru all the inodes in the file system,
1789 * adjusting the corresponding dquot counters in core.
1791 error = xfs_bulkstat(mp, &lastino, &count,
1792 xfs_qm_dqusage_adjust,
1793 structsz, NULL, &done);
1800 * We've made all the changes that we need to make incore.
1801 * Flush them down to disk buffers if everything was updated
1805 error = xfs_qm_dqflush_all(mp, 0);
1808 * We can get this error if we couldn't do a dquot allocation inside
1809 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1810 * dirty dquots that might be cached, we just want to get rid of them
1811 * and turn quotaoff. The dquots won't be attached to any of the inodes
1812 * at this point (because we intentionally didn't in dqget_noattach).
1815 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1820 * We didn't log anything, because if we crashed, we'll have to
1821 * start the quotacheck from scratch anyway. However, we must make
1822 * sure that our dquot changes are secure before we put the
1823 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1826 XFS_bflush(mp->m_ddev_targp);
1829 * If one type of quotas is off, then it will lose its
1830 * quotachecked status, since we won't be doing accounting for
1831 * that type anymore.
1833 mp->m_qflags &= ~(XFS_OQUOTA_CHKD | XFS_UQUOTA_CHKD);
1834 mp->m_qflags |= flags;
1836 xfs_qm_dquot_list_print(mp);
1840 cmn_err(CE_WARN, "XFS quotacheck %s: Unsuccessful (Error %d): "
1841 "Disabling quotas.",
1842 mp->m_fsname, error);
1844 * We must turn off quotas.
1846 ASSERT(mp->m_quotainfo != NULL);
1847 ASSERT(xfs_Gqm != NULL);
1848 xfs_qm_destroy_quotainfo(mp);
1849 if (xfs_mount_reset_sbqflags(mp)) {
1850 cmn_err(CE_WARN, "XFS quotacheck %s: "
1851 "Failed to reset quota flags.", mp->m_fsname);
1854 cmn_err(CE_NOTE, "XFS quotacheck %s: Done.", mp->m_fsname);
1860 * This is called after the superblock has been read in and we're ready to
1861 * iget the quota inodes.
1864 xfs_qm_init_quotainos(
1867 xfs_inode_t *uip, *gip;
1872 ASSERT(mp->m_quotainfo);
1878 * Get the uquota and gquota inodes
1880 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1881 if (XFS_IS_UQUOTA_ON(mp) &&
1882 mp->m_sb.sb_uquotino != NULLFSINO) {
1883 ASSERT(mp->m_sb.sb_uquotino > 0);
1884 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1886 return XFS_ERROR(error);
1888 if (XFS_IS_OQUOTA_ON(mp) &&
1889 mp->m_sb.sb_gquotino != NULLFSINO) {
1890 ASSERT(mp->m_sb.sb_gquotino > 0);
1891 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1895 return XFS_ERROR(error);
1899 flags |= XFS_QMOPT_SBVERSION;
1900 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1901 XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
1905 * Create the two inodes, if they don't exist already. The changes
1906 * made above will get added to a transaction and logged in one of
1907 * the qino_alloc calls below. If the device is readonly,
1908 * temporarily switch to read-write to do this.
1910 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1911 if ((error = xfs_qm_qino_alloc(mp, &uip,
1912 sbflags | XFS_SB_UQUOTINO,
1913 flags | XFS_QMOPT_UQUOTA)))
1914 return XFS_ERROR(error);
1916 flags &= ~XFS_QMOPT_SBVERSION;
1918 if (XFS_IS_OQUOTA_ON(mp) && gip == NULL) {
1919 flags |= (XFS_IS_GQUOTA_ON(mp) ?
1920 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1921 error = xfs_qm_qino_alloc(mp, &gip,
1922 sbflags | XFS_SB_GQUOTINO, flags);
1927 return XFS_ERROR(error);
1931 mp->m_quotainfo->qi_uquotaip = uip;
1932 mp->m_quotainfo->qi_gquotaip = gip;
1940 * Just pop the least recently used dquot off the freelist and
1941 * recycle it. The returned dquot is locked.
1943 STATIC xfs_dquot_t *
1944 xfs_qm_dqreclaim_one(void)
1946 xfs_dquot_t *dqpout;
1953 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
1955 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1957 list_for_each_entry(dqp, &xfs_Gqm->qm_dqfrlist, q_freelist) {
1958 struct xfs_mount *mp = dqp->q_mount;
1962 * We are racing with dqlookup here. Naturally we don't
1963 * want to reclaim a dquot that lookup wants. We release the
1964 * freelist lock and start over, so that lookup will grab
1965 * both the dquot and the freelistlock.
1967 if (dqp->dq_flags & XFS_DQ_WANT) {
1968 ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE));
1970 trace_xfs_dqreclaim_want(dqp);
1973 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1974 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
1976 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
1981 * If the dquot is inactive, we are assured that it is
1982 * not on the mplist or the hashlist, and that makes our
1985 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
1987 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
1988 ASSERT(list_empty(&dqp->q_hashlist));
1989 ASSERT(list_empty(&dqp->q_mplist));
1990 list_del_init(&dqp->q_freelist);
1991 xfs_Gqm->qm_dqfrlist_cnt--;
1994 XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
1998 ASSERT(dqp->q_hash);
1999 ASSERT(!list_empty(&dqp->q_mplist));
2002 * Try to grab the flush lock. If this dquot is in the process of
2003 * getting flushed to disk, we don't want to reclaim it.
2005 if (!xfs_dqflock_nowait(dqp)) {
2011 * We have the flush lock so we know that this is not in the
2012 * process of being flushed. So, if this is dirty, flush it
2013 * DELWRI so that we don't get a freelist infested with
2016 if (XFS_DQ_IS_DIRTY(dqp)) {
2019 trace_xfs_dqreclaim_dirty(dqp);
2022 * We flush it delayed write, so don't bother
2023 * releasing the freelist lock.
2025 error = xfs_qm_dqflush(dqp, 0);
2027 xfs_fs_cmn_err(CE_WARN, mp,
2028 "xfs_qm_dqreclaim: dquot %p flush failed", dqp);
2030 xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
2035 * We're trying to get the hashlock out of order. This races
2036 * with dqlookup; so, we giveup and goto the next dquot if
2037 * we couldn't get the hashlock. This way, we won't starve
2038 * a dqlookup process that holds the hashlock that is
2039 * waiting for the freelist lock.
2041 if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
2047 * This races with dquot allocation code as well as dqflush_all
2048 * and reclaim code. So, if we failed to grab the mplist lock,
2049 * giveup everything and start over.
2051 if (!mutex_trylock(&mp->m_quotainfo->qi_dqlist_lock)) {
2053 mutex_unlock(&dqp->q_hash->qh_lock);
2056 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
2057 if (restarts++ >= XFS_QM_RECLAIM_MAX_RESTARTS)
2062 ASSERT(dqp->q_nrefs == 0);
2063 list_del_init(&dqp->q_mplist);
2064 mp->m_quotainfo->qi_dquots--;
2065 mp->m_quotainfo->qi_dqreclaims++;
2066 list_del_init(&dqp->q_hashlist);
2067 dqp->q_hash->qh_version++;
2068 list_del_init(&dqp->q_freelist);
2069 xfs_Gqm->qm_dqfrlist_cnt--;
2071 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
2072 mutex_unlock(&dqp->q_hash->qh_lock);
2078 if (restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
2081 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
2086 * Traverse the freelist of dquots and attempt to reclaim a maximum of
2087 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
2088 * favor the lookup function ...
2091 xfs_qm_shake_freelist(
2100 while (nreclaimed < howmany) {
2101 dqp = xfs_qm_dqreclaim_one();
2104 xfs_qm_dqdestroy(dqp);
2111 * The kmem_shake interface is invoked when memory is running low.
2116 struct shrinker *shrink,
2120 int ndqused, nfree, n;
2122 if (!kmem_shake_allow(gfp_mask))
2127 nfree = xfs_Gqm->qm_dqfrlist_cnt; /* free dquots */
2128 /* incore dquots in all f/s's */
2129 ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;
2131 ASSERT(ndqused >= 0);
2133 if (nfree <= ndqused && nfree < ndquot)
2136 ndqused *= xfs_Gqm->qm_dqfree_ratio; /* target # of free dquots */
2137 n = nfree - ndqused - ndquot; /* # over target */
2139 return xfs_qm_shake_freelist(MAX(nfree, n));
2143 /*------------------------------------------------------------------*/
2146 * Return a new incore dquot. Depending on the number of
2147 * dquots in the system, we either allocate a new one on the kernel heap,
2148 * or reclaim a free one.
2149 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2150 * to reclaim an existing one from the freelist.
2153 xfs_qm_dqalloc_incore(
2154 xfs_dquot_t **O_dqpp)
2159 * Check against high water mark to see if we want to pop
2160 * a nincompoop dquot off the freelist.
2162 if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
2164 * Try to recycle a dquot from the freelist.
2166 if ((dqp = xfs_qm_dqreclaim_one())) {
2167 XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
2169 * Just zero the core here. The rest will get
2170 * reinitialized by caller. XXX we shouldn't even
2173 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
2177 XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
2181 * Allocate a brand new dquot on the kernel heap and return it
2182 * to the caller to initialize.
2184 ASSERT(xfs_Gqm->qm_dqzone != NULL);
2185 *O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
2186 atomic_inc(&xfs_Gqm->qm_totaldquots);
2193 * Start a transaction and write the incore superblock changes to
2194 * disk. flags parameter indicates which fields have changed.
2197 xfs_qm_write_sb_changes(
2205 cmn_err(CE_NOTE, "Writing superblock quota changes :%s", mp->m_fsname);
2207 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
2208 if ((error = xfs_trans_reserve(tp, 0,
2209 mp->m_sb.sb_sectsize + 128, 0,
2211 XFS_DEFAULT_LOG_COUNT))) {
2212 xfs_trans_cancel(tp, 0);
2216 xfs_mod_sb(tp, flags);
2217 error = xfs_trans_commit(tp, 0);
2223 /* --------------- utility functions for vnodeops ---------------- */
2227 * Given an inode, a uid and gid (from cred_t) make sure that we have
2228 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2229 * quotas by creating this file.
2230 * This also attaches dquot(s) to the given inode after locking it,
2231 * and returns the dquots corresponding to the uid and/or gid.
2233 * in : inode (unlocked)
2234 * out : udquot, gdquot with references taken and unlocked
2238 struct xfs_inode *ip,
2243 struct xfs_dquot **O_udqpp,
2244 struct xfs_dquot **O_gdqpp)
2246 struct xfs_mount *mp = ip->i_mount;
2247 struct xfs_dquot *uq, *gq;
2251 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2254 lockflags = XFS_ILOCK_EXCL;
2255 xfs_ilock(ip, lockflags);
2257 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
2258 gid = ip->i_d.di_gid;
2261 * Attach the dquot(s) to this inode, doing a dquot allocation
2262 * if necessary. The dquot(s) will not be locked.
2264 if (XFS_NOT_DQATTACHED(mp, ip)) {
2265 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
2267 xfs_iunlock(ip, lockflags);
2273 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
2274 if (ip->i_d.di_uid != uid) {
2276 * What we need is the dquot that has this uid, and
2277 * if we send the inode to dqget, the uid of the inode
2278 * takes priority over what's sent in the uid argument.
2279 * We must unlock inode here before calling dqget if
2280 * we're not sending the inode, because otherwise
2281 * we'll deadlock by doing trans_reserve while
2284 xfs_iunlock(ip, lockflags);
2285 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
2290 ASSERT(error != ENOENT);
2294 * Get the ilock in the right order.
2297 lockflags = XFS_ILOCK_SHARED;
2298 xfs_ilock(ip, lockflags);
2301 * Take an extra reference, because we'll return
2304 ASSERT(ip->i_udquot);
2311 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
2312 if (ip->i_d.di_gid != gid) {
2313 xfs_iunlock(ip, lockflags);
2314 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
2321 ASSERT(error != ENOENT);
2325 lockflags = XFS_ILOCK_SHARED;
2326 xfs_ilock(ip, lockflags);
2328 ASSERT(ip->i_gdquot);
2334 } else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
2335 if (ip->i_d.di_projid != prid) {
2336 xfs_iunlock(ip, lockflags);
2337 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
2344 ASSERT(error != ENOENT);
2348 lockflags = XFS_ILOCK_SHARED;
2349 xfs_ilock(ip, lockflags);
2351 ASSERT(ip->i_gdquot);
2359 trace_xfs_dquot_dqalloc(ip);
2361 xfs_iunlock(ip, lockflags);
2374 * Actually transfer ownership, and do dquot modifications.
2375 * These were already reserved.
2381 xfs_dquot_t **IO_olddq,
2384 xfs_dquot_t *prevdq;
2385 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
2386 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
2389 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2390 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
2395 ASSERT(prevdq != newdq);
2397 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
2398 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
2400 /* the sparkling new dquot */
2401 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
2402 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
2405 * Take an extra reference, because the inode
2406 * is going to keep this dquot pointer even
2407 * after the trans_commit.
2411 xfs_dqunlock(newdq);
2418 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
2421 xfs_qm_vop_chown_reserve(
2428 xfs_mount_t *mp = ip->i_mount;
2429 uint delblks, blkflags, prjflags = 0;
2430 xfs_dquot_t *unresudq, *unresgdq, *delblksudq, *delblksgdq;
2434 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2435 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2437 delblks = ip->i_delayed_blks;
2438 delblksudq = delblksgdq = unresudq = unresgdq = NULL;
2439 blkflags = XFS_IS_REALTIME_INODE(ip) ?
2440 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
2442 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2443 ip->i_d.di_uid != (uid_t)be32_to_cpu(udqp->q_core.d_id)) {
2446 * If there are delayed allocation blocks, then we have to
2447 * unreserve those from the old dquot, and add them to the
2451 ASSERT(ip->i_udquot);
2452 unresudq = ip->i_udquot;
2455 if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
2456 if (XFS_IS_PQUOTA_ON(ip->i_mount) &&
2457 ip->i_d.di_projid != be32_to_cpu(gdqp->q_core.d_id))
2458 prjflags = XFS_QMOPT_ENOSPC;
2461 (XFS_IS_GQUOTA_ON(ip->i_mount) &&
2462 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id))) {
2465 ASSERT(ip->i_gdquot);
2466 unresgdq = ip->i_gdquot;
2471 if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2472 delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
2473 flags | blkflags | prjflags)))
2477 * Do the delayed blks reservations/unreservations now. Since, these
2478 * are done without the help of a transaction, if a reservation fails
2479 * its previous reservations won't be automatically undone by trans
2480 * code. So, we have to do it manually here.
2484 * Do the reservations first. Unreservation can't fail.
2486 ASSERT(delblksudq || delblksgdq);
2487 ASSERT(unresudq || unresgdq);
2488 if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2489 delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
2490 flags | blkflags | prjflags)))
2492 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2493 unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
2501 xfs_qm_vop_rename_dqattach(
2502 struct xfs_inode **i_tab)
2504 struct xfs_mount *mp = i_tab[0]->i_mount;
2507 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2510 for (i = 0; (i < 4 && i_tab[i]); i++) {
2511 struct xfs_inode *ip = i_tab[i];
2515 * Watch out for duplicate entries in the table.
2517 if (i == 0 || ip != i_tab[i-1]) {
2518 if (XFS_NOT_DQATTACHED(mp, ip)) {
2519 error = xfs_qm_dqattach(ip, 0);
2529 xfs_qm_vop_create_dqattach(
2530 struct xfs_trans *tp,
2531 struct xfs_inode *ip,
2532 struct xfs_dquot *udqp,
2533 struct xfs_dquot *gdqp)
2535 struct xfs_mount *mp = tp->t_mountp;
2537 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2540 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2541 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2547 ASSERT(ip->i_udquot == NULL);
2548 ip->i_udquot = udqp;
2549 ASSERT(XFS_IS_UQUOTA_ON(mp));
2550 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
2551 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2557 ASSERT(ip->i_gdquot == NULL);
2558 ip->i_gdquot = gdqp;
2559 ASSERT(XFS_IS_OQUOTA_ON(mp));
2560 ASSERT((XFS_IS_GQUOTA_ON(mp) ?
2561 ip->i_d.di_gid : ip->i_d.di_projid) ==
2562 be32_to_cpu(gdqp->q_core.d_id));
2563 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);