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"
27 #include "xfs_alloc.h"
28 #include "xfs_quota.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_itable.h"
40 #include "xfs_rtalloc.h"
41 #include "xfs_error.h"
45 #include "xfs_buf_item.h"
46 #include "xfs_trans_space.h"
47 #include "xfs_utils.h"
49 #include "xfs_trace.h"
52 * The global quota manager. There is only one of these for the entire
53 * system, _not_ one per file system. XQM keeps track of the overall
54 * quota functionality, including maintaining the freelist and hash
57 struct mutex xfs_Gqm_lock;
58 struct xfs_qm *xfs_Gqm;
61 kmem_zone_t *qm_dqzone;
62 kmem_zone_t *qm_dqtrxzone;
64 static cred_t xfs_zerocr;
66 STATIC void xfs_qm_list_init(xfs_dqlist_t *, char *, int);
67 STATIC void xfs_qm_list_destroy(xfs_dqlist_t *);
69 STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
70 STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
71 STATIC int xfs_qm_shake(struct shrinker *, int, gfp_t);
73 static struct shrinker xfs_qm_shaker = {
74 .shrink = xfs_qm_shake,
75 .seeks = DEFAULT_SEEKS,
79 extern struct mutex qcheck_lock;
84 xfs_qm_dquot_list_print(
90 list_for_each_entry(dqp, &mp->m_quotainfo->qi_dqlist_lock, qi_mplist) {
91 cmn_err(CE_DEBUG, " %d. \"%d (%s)\" "
92 "bcnt = %lld, icnt = %lld, refs = %d",
93 i++, be32_to_cpu(dqp->q_core.d_id),
94 DQFLAGTO_TYPESTR(dqp),
95 (long long)be64_to_cpu(dqp->q_core.d_bcount),
96 (long long)be64_to_cpu(dqp->q_core.d_icount),
101 static void xfs_qm_dquot_list_print(struct xfs_mount *mp) { }
105 * Initialize the XQM structure.
106 * Note that there is not one quota manager per file system.
108 STATIC struct xfs_qm *
111 xfs_dqhash_t *udqhash, *gdqhash;
117 * Initialize the dquot hash tables.
119 udqhash = kmem_zalloc_greedy(&hsize,
120 XFS_QM_HASHSIZE_LOW * sizeof(xfs_dqhash_t),
121 XFS_QM_HASHSIZE_HIGH * sizeof(xfs_dqhash_t));
125 gdqhash = kmem_zalloc_large(hsize);
127 goto out_free_udqhash;
129 hsize /= sizeof(xfs_dqhash_t);
132 xqm = kmem_zalloc(sizeof(xfs_qm_t), KM_SLEEP);
133 xqm->qm_dqhashmask = hsize - 1;
134 xqm->qm_usr_dqhtable = udqhash;
135 xqm->qm_grp_dqhtable = gdqhash;
136 ASSERT(xqm->qm_usr_dqhtable != NULL);
137 ASSERT(xqm->qm_grp_dqhtable != NULL);
139 for (i = 0; i < hsize; i++) {
140 xfs_qm_list_init(&(xqm->qm_usr_dqhtable[i]), "uxdqh", i);
141 xfs_qm_list_init(&(xqm->qm_grp_dqhtable[i]), "gxdqh", i);
145 * Freelist of all dquots of all file systems
147 INIT_LIST_HEAD(&xqm->qm_dqfrlist);
148 xqm->qm_dqfrlist_cnt = 0;
149 mutex_init(&xqm->qm_dqfrlist_lock);
152 * dquot zone. we register our own low-memory callback.
155 xqm->qm_dqzone = kmem_zone_init(sizeof(xfs_dquot_t),
157 qm_dqzone = xqm->qm_dqzone;
159 xqm->qm_dqzone = qm_dqzone;
161 register_shrinker(&xfs_qm_shaker);
164 * The t_dqinfo portion of transactions.
167 xqm->qm_dqtrxzone = kmem_zone_init(sizeof(xfs_dquot_acct_t),
169 qm_dqtrxzone = xqm->qm_dqtrxzone;
171 xqm->qm_dqtrxzone = qm_dqtrxzone;
173 atomic_set(&xqm->qm_totaldquots, 0);
174 xqm->qm_dqfree_ratio = XFS_QM_DQFREE_RATIO;
177 mutex_init(&qcheck_lock);
182 kmem_free_large(udqhash);
188 * Destroy the global quota manager when its reference count goes to zero.
194 struct xfs_dquot *dqp, *n;
198 ASSERT(xqm->qm_nrefs == 0);
199 unregister_shrinker(&xfs_qm_shaker);
200 hsize = xqm->qm_dqhashmask + 1;
201 for (i = 0; i < hsize; i++) {
202 xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
203 xfs_qm_list_destroy(&(xqm->qm_grp_dqhtable[i]));
205 kmem_free_large(xqm->qm_usr_dqhtable);
206 kmem_free_large(xqm->qm_grp_dqhtable);
207 xqm->qm_usr_dqhtable = NULL;
208 xqm->qm_grp_dqhtable = NULL;
209 xqm->qm_dqhashmask = 0;
212 mutex_lock(&xqm->qm_dqfrlist_lock);
213 list_for_each_entry_safe(dqp, n, &xqm->qm_dqfrlist, q_freelist) {
216 cmn_err(CE_DEBUG, "FREELIST destroy 0x%p", dqp);
218 list_del_init(&dqp->q_freelist);
219 xfs_Gqm->qm_dqfrlist_cnt--;
221 xfs_qm_dqdestroy(dqp);
223 mutex_unlock(&xqm->qm_dqfrlist_lock);
224 mutex_destroy(&xqm->qm_dqfrlist_lock);
226 mutex_destroy(&qcheck_lock);
232 * Called at mount time to let XQM know that another file system is
233 * starting quotas. This isn't crucial information as the individual mount
234 * structures are pretty independent, but it helps the XQM keep a
235 * global view of what's going on.
239 xfs_qm_hold_quotafs_ref(
240 struct xfs_mount *mp)
243 * Need to lock the xfs_Gqm structure for things like this. For example,
244 * the structure could disappear between the entry to this routine and
245 * a HOLD operation if not locked.
247 mutex_lock(&xfs_Gqm_lock);
250 xfs_Gqm = xfs_Gqm_init();
252 mutex_unlock(&xfs_Gqm_lock);
258 * We can keep a list of all filesystems with quotas mounted for
259 * debugging and statistical purposes, but ...
260 * Just take a reference and get out.
263 mutex_unlock(&xfs_Gqm_lock);
270 * Release the reference that a filesystem took at mount time,
271 * so that we know when we need to destroy the entire quota manager.
275 xfs_qm_rele_quotafs_ref(
276 struct xfs_mount *mp)
278 xfs_dquot_t *dqp, *n;
281 ASSERT(xfs_Gqm->qm_nrefs > 0);
284 * Go thru the freelist and destroy all inactive dquots.
286 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
288 list_for_each_entry_safe(dqp, n, &xfs_Gqm->qm_dqfrlist, q_freelist) {
290 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
291 ASSERT(dqp->q_mount == NULL);
292 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
293 ASSERT(list_empty(&dqp->q_hashlist));
294 ASSERT(list_empty(&dqp->q_mplist));
295 list_del_init(&dqp->q_freelist);
296 xfs_Gqm->qm_dqfrlist_cnt--;
298 xfs_qm_dqdestroy(dqp);
303 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
306 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
309 mutex_lock(&xfs_Gqm_lock);
310 if (--xfs_Gqm->qm_nrefs == 0) {
311 xfs_qm_destroy(xfs_Gqm);
314 mutex_unlock(&xfs_Gqm_lock);
318 * Just destroy the quotainfo structure.
322 struct xfs_mount *mp)
324 if (mp->m_quotainfo) {
325 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
326 xfs_qm_destroy_quotainfo(mp);
332 * This is called from xfs_mountfs to start quotas and initialize all
333 * necessary data structures like quotainfo. This is also responsible for
334 * running a quotacheck as necessary. We are guaranteed that the superblock
335 * is consistently read in at this point.
337 * If we fail here, the mount will continue with quota turned off. We don't
338 * need to inidicate success or failure at all.
348 * If quotas on realtime volumes is not supported, we disable
349 * quotas immediately.
351 if (mp->m_sb.sb_rextents) {
353 "Cannot turn on quotas for realtime filesystem %s",
359 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
362 * Allocate the quotainfo structure inside the mount struct, and
363 * create quotainode(s), and change/rev superblock if necessary.
365 error = xfs_qm_init_quotainfo(mp);
368 * We must turn off quotas.
370 ASSERT(mp->m_quotainfo == NULL);
375 * If any of the quotas are not consistent, do a quotacheck.
377 if (XFS_QM_NEED_QUOTACHECK(mp)) {
378 error = xfs_qm_quotacheck(mp);
380 /* Quotacheck failed and disabled quotas. */
385 * If one type of quotas is off, then it will lose its
386 * quotachecked status, since we won't be doing accounting for
389 if (!XFS_IS_UQUOTA_ON(mp))
390 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
391 if (!(XFS_IS_GQUOTA_ON(mp) || XFS_IS_PQUOTA_ON(mp)))
392 mp->m_qflags &= ~XFS_OQUOTA_CHKD;
396 * We actually don't have to acquire the m_sb_lock at all.
397 * This can only be called from mount, and that's single threaded. XXX
399 spin_lock(&mp->m_sb_lock);
400 sbf = mp->m_sb.sb_qflags;
401 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
402 spin_unlock(&mp->m_sb_lock);
404 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
405 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
407 * We could only have been turning quotas off.
408 * We aren't in very good shape actually because
409 * the incore structures are convinced that quotas are
410 * off, but the on disk superblock doesn't know that !
412 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
413 xfs_fs_cmn_err(CE_ALERT, mp,
414 "XFS mount_quotas: Superblock update failed!");
419 xfs_fs_cmn_err(CE_WARN, mp,
420 "Failed to initialize disk quotas.");
425 if (XFS_IS_QUOTA_ON(mp))
426 xfs_qm_internalqcheck(mp);
431 * Called from the vfsops layer.
434 xfs_qm_unmount_quotas(
438 * Release the dquots that root inode, et al might be holding,
439 * before we flush quotas and blow away the quotainfo structure.
441 ASSERT(mp->m_rootip);
442 xfs_qm_dqdetach(mp->m_rootip);
444 xfs_qm_dqdetach(mp->m_rbmip);
446 xfs_qm_dqdetach(mp->m_rsumip);
449 * Release the quota inodes.
451 if (mp->m_quotainfo) {
452 if (mp->m_quotainfo->qi_uquotaip) {
453 IRELE(mp->m_quotainfo->qi_uquotaip);
454 mp->m_quotainfo->qi_uquotaip = NULL;
456 if (mp->m_quotainfo->qi_gquotaip) {
457 IRELE(mp->m_quotainfo->qi_gquotaip);
458 mp->m_quotainfo->qi_gquotaip = NULL;
464 * Flush all dquots of the given file system to disk. The dquots are
465 * _not_ purged from memory here, just their data written to disk.
469 struct xfs_mount *mp,
472 struct xfs_quotainfo *q = mp->m_quotainfo;
474 struct xfs_dquot *dqp;
482 mutex_lock(&q->qi_dqlist_lock);
483 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
485 if (! XFS_DQ_IS_DIRTY(dqp)) {
490 /* XXX a sentinel would be better */
491 recl = q->qi_dqreclaims;
492 if (!xfs_dqflock_nowait(dqp)) {
494 * If we can't grab the flush lock then check
495 * to see if the dquot has been flushed delayed
496 * write. If so, grab its buffer and send it
497 * out immediately. We'll be able to acquire
498 * the flush lock when the I/O completes.
500 xfs_qm_dqflock_pushbuf_wait(dqp);
503 * Let go of the mplist lock. We don't want to hold it
504 * across a disk write.
506 mutex_unlock(&q->qi_dqlist_lock);
507 error = xfs_qm_dqflush(dqp, sync_mode);
512 mutex_lock(&q->qi_dqlist_lock);
513 if (recl != q->qi_dqreclaims) {
514 mutex_unlock(&q->qi_dqlist_lock);
515 /* XXX restart limit */
520 mutex_unlock(&q->qi_dqlist_lock);
525 * Release the group dquot pointers the user dquots may be
526 * carrying around as a hint. mplist is locked on entry and exit.
529 xfs_qm_detach_gdquots(
530 struct xfs_mount *mp)
532 struct xfs_quotainfo *q = mp->m_quotainfo;
533 struct xfs_dquot *dqp, *gdqp;
537 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
538 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
540 if ((gdqp = dqp->q_gdquot)) {
542 dqp->q_gdquot = NULL;
548 * Can't hold the mplist lock across a dqput.
549 * XXXmust convert to marker based iterations here.
551 nrecl = q->qi_dqreclaims;
552 mutex_unlock(&q->qi_dqlist_lock);
555 mutex_lock(&q->qi_dqlist_lock);
556 if (nrecl != q->qi_dqreclaims)
563 * Go through all the incore dquots of this file system and take them
564 * off the mplist and hashlist, if the dquot type matches the dqtype
565 * parameter. This is used when turning off quota accounting for
566 * users and/or groups, as well as when the filesystem is unmounting.
570 struct xfs_mount *mp,
573 struct xfs_quotainfo *q = mp->m_quotainfo;
574 struct xfs_dquot *dqp, *n;
582 dqtype = (flags & XFS_QMOPT_UQUOTA) ? XFS_DQ_USER : 0;
583 dqtype |= (flags & XFS_QMOPT_PQUOTA) ? XFS_DQ_PROJ : 0;
584 dqtype |= (flags & XFS_QMOPT_GQUOTA) ? XFS_DQ_GROUP : 0;
586 mutex_lock(&q->qi_dqlist_lock);
589 * In the first pass through all incore dquots of this filesystem,
590 * we release the group dquot pointers the user dquots may be
591 * carrying around as a hint. We need to do this irrespective of
592 * what's being turned off.
594 xfs_qm_detach_gdquots(mp);
598 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
600 * Try to get rid of all of the unwanted dquots. The idea is to
601 * get them off mplist and hashlist, but leave them on freelist.
603 list_for_each_entry_safe(dqp, n, &q->qi_dqlist, q_mplist) {
605 * It's OK to look at the type without taking dqlock here.
606 * We're holding the mplist lock here, and that's needed for
609 if ((dqp->dq_flags & dqtype) == 0)
612 if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
613 nrecl = q->qi_dqreclaims;
614 mutex_unlock(&q->qi_dqlist_lock);
615 mutex_lock(&dqp->q_hash->qh_lock);
616 mutex_lock(&q->qi_dqlist_lock);
619 * XXXTheoretically, we can get into a very long
620 * ping pong game here.
621 * No one can be adding dquots to the mplist at
622 * this point, but somebody might be taking things off.
624 if (nrecl != q->qi_dqreclaims) {
625 mutex_unlock(&dqp->q_hash->qh_lock);
631 * Take the dquot off the mplist and hashlist. It may remain on
632 * freelist in INACTIVE state.
634 nmisses += xfs_qm_dqpurge(dqp);
636 mutex_unlock(&q->qi_dqlist_lock);
648 * Purge the dquot cache.
649 * None of the dquots should really be busy at this point.
651 if (mp->m_quotainfo) {
652 while ((ndquots = xfs_qm_dqpurge_int(mp, flags))) {
665 xfs_dquot_t *udqhint, /* hint */
666 xfs_dquot_t **IO_idqpp)
671 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
675 * See if we already have it in the inode itself. IO_idqpp is
676 * &i_udquot or &i_gdquot. This made the code look weird, but
677 * made the logic a lot simpler.
681 trace_xfs_dqattach_found(dqp);
686 * udqhint is the i_udquot field in inode, and is non-NULL only
687 * when the type arg is group/project. Its purpose is to save a
688 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
692 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
696 * No need to take dqlock to look at the id.
698 * The ID can't change until it gets reclaimed, and it won't
699 * be reclaimed as long as we have a ref from inode and we
702 dqp = udqhint->q_gdquot;
703 if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
706 ASSERT(*IO_idqpp == NULL);
710 xfs_dqunlock(udqhint);
715 * We can't hold a dquot lock when we call the dqget code.
716 * We'll deadlock in no time, because of (not conforming to)
717 * lock ordering - the inodelock comes before any dquot lock,
718 * and we may drop and reacquire the ilock in xfs_qm_dqget().
720 xfs_dqunlock(udqhint);
724 * Find the dquot from somewhere. This bumps the
725 * reference count of dquot and returns it locked.
726 * This can return ENOENT if dquot didn't exist on
727 * disk and we didn't ask it to allocate;
728 * ESRCH if quotas got turned off suddenly.
730 error = xfs_qm_dqget(ip->i_mount, ip, id, type, XFS_QMOPT_DOWARN, &dqp);
734 trace_xfs_dqattach_get(dqp);
737 * dqget may have dropped and re-acquired the ilock, but it guarantees
738 * that the dquot returned is the one that should go in the inode.
747 * Given a udquot and gdquot, attach a ptr to the group dquot in the
748 * udquot as a hint for future lookups. The idea sounds simple, but the
749 * execution isn't, because the udquot might have a group dquot attached
750 * already and getting rid of that gets us into lock ordering constraints.
751 * The process is complicated more by the fact that the dquots may or may not
752 * be locked on entry.
755 xfs_qm_dqattach_grouphint(
763 if ((tmp = udq->q_gdquot)) {
769 udq->q_gdquot = NULL;
771 * We can't keep any dqlocks when calling dqrele,
772 * because the freelist lock comes before dqlocks.
776 * we took a hard reference once upon a time in dqget,
777 * so give it back when the udquot no longer points at it
778 * dqput() does the unlocking of the dquot.
786 ASSERT(XFS_DQ_IS_LOCKED(udq));
790 ASSERT(XFS_DQ_IS_LOCKED(udq));
791 ASSERT(XFS_DQ_IS_LOCKED(gdq));
793 * Somebody could have attached a gdquot here,
794 * when we dropped the uqlock. If so, just do nothing.
796 if (udq->q_gdquot == NULL) {
807 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
809 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
810 * Inode may get unlocked and relocked in here, and the caller must deal with
814 xfs_qm_dqattach_locked(
818 xfs_mount_t *mp = ip->i_mount;
822 if (!XFS_IS_QUOTA_RUNNING(mp) ||
823 !XFS_IS_QUOTA_ON(mp) ||
824 !XFS_NOT_DQATTACHED(mp, ip) ||
825 ip->i_ino == mp->m_sb.sb_uquotino ||
826 ip->i_ino == mp->m_sb.sb_gquotino)
829 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
831 if (XFS_IS_UQUOTA_ON(mp)) {
832 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
833 flags & XFS_QMOPT_DQALLOC,
834 NULL, &ip->i_udquot);
840 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
841 if (XFS_IS_OQUOTA_ON(mp)) {
842 error = XFS_IS_GQUOTA_ON(mp) ?
843 xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
844 flags & XFS_QMOPT_DQALLOC,
845 ip->i_udquot, &ip->i_gdquot) :
846 xfs_qm_dqattach_one(ip, ip->i_d.di_projid, XFS_DQ_PROJ,
847 flags & XFS_QMOPT_DQALLOC,
848 ip->i_udquot, &ip->i_gdquot);
850 * Don't worry about the udquot that we may have
851 * attached above. It'll get detached, if not already.
859 * Attach this group quota to the user quota as a hint.
860 * This WON'T, in general, result in a thrash.
863 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
864 ASSERT(ip->i_udquot);
865 ASSERT(ip->i_gdquot);
868 * We may or may not have the i_udquot locked at this point,
869 * but this check is OK since we don't depend on the i_gdquot to
870 * be accurate 100% all the time. It is just a hint, and this
871 * will succeed in general.
873 if (ip->i_udquot->q_gdquot == ip->i_gdquot)
876 * Attach i_gdquot to the gdquot hint inside the i_udquot.
878 xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot);
884 if (XFS_IS_UQUOTA_ON(mp))
885 ASSERT(ip->i_udquot);
886 if (XFS_IS_OQUOTA_ON(mp))
887 ASSERT(ip->i_gdquot);
889 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
896 struct xfs_inode *ip,
901 xfs_ilock(ip, XFS_ILOCK_EXCL);
902 error = xfs_qm_dqattach_locked(ip, flags);
903 xfs_iunlock(ip, XFS_ILOCK_EXCL);
909 * Release dquots (and their references) if any.
910 * The inode should be locked EXCL except when this's called by
917 if (!(ip->i_udquot || ip->i_gdquot))
920 trace_xfs_dquot_dqdetach(ip);
922 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
923 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
925 xfs_qm_dqrele(ip->i_udquot);
929 xfs_qm_dqrele(ip->i_gdquot);
936 struct xfs_mount *mp,
939 struct xfs_quotainfo *q = mp->m_quotainfo;
941 struct xfs_dquot *dqp;
944 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
950 mutex_lock(&q->qi_dqlist_lock);
952 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
953 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
954 * when we have the mplist lock, we know that dquots will be consistent
955 * as long as we have it locked.
957 if (!XFS_IS_QUOTA_ON(mp)) {
958 mutex_unlock(&q->qi_dqlist_lock);
961 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
962 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
964 * If this is vfs_sync calling, then skip the dquots that
965 * don't 'seem' to be dirty. ie. don't acquire dqlock.
966 * This is very similar to what xfs_sync does with inodes.
968 if (flags & SYNC_TRYLOCK) {
969 if (!XFS_DQ_IS_DIRTY(dqp))
971 if (!xfs_qm_dqlock_nowait(dqp))
978 * Now, find out for sure if this dquot is dirty or not.
980 if (! XFS_DQ_IS_DIRTY(dqp)) {
985 /* XXX a sentinel would be better */
986 recl = q->qi_dqreclaims;
987 if (!xfs_dqflock_nowait(dqp)) {
988 if (flags & SYNC_TRYLOCK) {
993 * If we can't grab the flush lock then if the caller
994 * really wanted us to give this our best shot, so
995 * see if we can give a push to the buffer before we wait
996 * on the flush lock. At this point, we know that
997 * even though the dquot is being flushed,
998 * it has (new) dirty data.
1000 xfs_qm_dqflock_pushbuf_wait(dqp);
1003 * Let go of the mplist lock. We don't want to hold it
1004 * across a disk write
1006 mutex_unlock(&q->qi_dqlist_lock);
1007 error = xfs_qm_dqflush(dqp, flags);
1009 if (error && XFS_FORCED_SHUTDOWN(mp))
1010 return 0; /* Need to prevent umount failure */
1014 mutex_lock(&q->qi_dqlist_lock);
1015 if (recl != q->qi_dqreclaims) {
1016 if (++restarts >= XFS_QM_SYNC_MAX_RESTARTS)
1019 mutex_unlock(&q->qi_dqlist_lock);
1024 mutex_unlock(&q->qi_dqlist_lock);
1029 * The hash chains and the mplist use the same xfs_dqhash structure as
1030 * their list head, but we can take the mplist qh_lock and one of the
1031 * hash qh_locks at the same time without any problem as they aren't
1034 static struct lock_class_key xfs_quota_mplist_class;
1037 * This initializes all the quota information that's kept in the
1041 xfs_qm_init_quotainfo(
1044 xfs_quotainfo_t *qinf;
1048 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1051 * Tell XQM that we exist as soon as possible.
1053 if ((error = xfs_qm_hold_quotafs_ref(mp))) {
1057 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
1060 * See if quotainodes are setup, and if not, allocate them,
1061 * and change the superblock accordingly.
1063 if ((error = xfs_qm_init_quotainos(mp))) {
1065 mp->m_quotainfo = NULL;
1069 INIT_LIST_HEAD(&qinf->qi_dqlist);
1070 mutex_init(&qinf->qi_dqlist_lock);
1071 lockdep_set_class(&qinf->qi_dqlist_lock, &xfs_quota_mplist_class);
1073 qinf->qi_dqreclaims = 0;
1075 /* mutex used to serialize quotaoffs */
1076 mutex_init(&qinf->qi_quotaofflock);
1078 /* Precalc some constants */
1079 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1080 ASSERT(qinf->qi_dqchunklen);
1081 qinf->qi_dqperchunk = BBTOB(qinf->qi_dqchunklen);
1082 do_div(qinf->qi_dqperchunk, sizeof(xfs_dqblk_t));
1084 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
1087 * We try to get the limits from the superuser's limits fields.
1088 * This is quite hacky, but it is standard quota practice.
1089 * We look at the USR dquot with id == 0 first, but if user quotas
1090 * are not enabled we goto the GRP dquot with id == 0.
1091 * We don't really care to keep separate default limits for user
1092 * and group quotas, at least not at this point.
1094 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)0,
1095 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
1096 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
1098 XFS_QMOPT_DQSUSER|XFS_QMOPT_DOWARN,
1101 xfs_disk_dquot_t *ddqp = &dqp->q_core;
1104 * The warnings and timers set the grace period given to
1105 * a user or group before he or she can not perform any
1106 * more writing. If it is zero, a default is used.
1108 qinf->qi_btimelimit = ddqp->d_btimer ?
1109 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
1110 qinf->qi_itimelimit = ddqp->d_itimer ?
1111 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
1112 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
1113 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
1114 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
1115 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
1116 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
1117 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
1118 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
1119 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
1120 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
1121 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
1122 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
1123 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
1124 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
1125 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
1128 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1129 * we don't want this dquot cached. We haven't done a
1130 * quotacheck yet, and quotacheck doesn't like incore dquots.
1132 xfs_qm_dqdestroy(dqp);
1134 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
1135 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
1136 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
1137 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
1138 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
1139 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
1147 * Gets called when unmounting a filesystem or when all quotas get
1149 * This purges the quota inodes, destroys locks and frees itself.
1152 xfs_qm_destroy_quotainfo(
1155 xfs_quotainfo_t *qi;
1157 qi = mp->m_quotainfo;
1159 ASSERT(xfs_Gqm != NULL);
1162 * Release the reference that XQM kept, so that we know
1163 * when the XQM structure should be freed. We cannot assume
1164 * that xfs_Gqm is non-null after this point.
1166 xfs_qm_rele_quotafs_ref(mp);
1168 ASSERT(list_empty(&qi->qi_dqlist));
1169 mutex_destroy(&qi->qi_dqlist_lock);
1171 if (qi->qi_uquotaip) {
1172 IRELE(qi->qi_uquotaip);
1173 qi->qi_uquotaip = NULL; /* paranoia */
1175 if (qi->qi_gquotaip) {
1176 IRELE(qi->qi_gquotaip);
1177 qi->qi_gquotaip = NULL;
1179 mutex_destroy(&qi->qi_quotaofflock);
1181 mp->m_quotainfo = NULL;
1186 /* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1195 mutex_init(&list->qh_lock);
1196 INIT_LIST_HEAD(&list->qh_list);
1197 list->qh_version = 0;
1198 list->qh_nelems = 0;
1202 xfs_qm_list_destroy(
1205 mutex_destroy(&(list->qh_lock));
1210 * Stripped down version of dqattach. This doesn't attach, or even look at the
1211 * dquots attached to the inode. The rationale is that there won't be any
1212 * attached at the time this is called from quotacheck.
1215 xfs_qm_dqget_noattach(
1217 xfs_dquot_t **O_udqpp,
1218 xfs_dquot_t **O_gdqpp)
1222 xfs_dquot_t *udqp, *gdqp;
1224 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1229 if (XFS_IS_UQUOTA_ON(mp)) {
1230 ASSERT(ip->i_udquot == NULL);
1232 * We want the dquot allocated if it doesn't exist.
1234 if ((error = xfs_qm_dqget(mp, ip, ip->i_d.di_uid, XFS_DQ_USER,
1235 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN,
1238 * Shouldn't be able to turn off quotas here.
1240 ASSERT(error != ESRCH);
1241 ASSERT(error != ENOENT);
1247 if (XFS_IS_OQUOTA_ON(mp)) {
1248 ASSERT(ip->i_gdquot == NULL);
1251 error = XFS_IS_GQUOTA_ON(mp) ?
1252 xfs_qm_dqget(mp, ip,
1253 ip->i_d.di_gid, XFS_DQ_GROUP,
1254 XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
1256 xfs_qm_dqget(mp, ip,
1257 ip->i_d.di_projid, XFS_DQ_PROJ,
1258 XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
1262 xfs_qm_dqrele(udqp);
1263 ASSERT(error != ESRCH);
1264 ASSERT(error != ENOENT);
1269 /* Reacquire the locks in the right order */
1271 if (! xfs_qm_dqlock_nowait(udqp)) {
1283 if (udqp) ASSERT(XFS_DQ_IS_LOCKED(udqp));
1284 if (gdqp) ASSERT(XFS_DQ_IS_LOCKED(gdqp));
1290 * Create an inode and return with a reference already taken, but unlocked
1291 * This is how we create quota inodes
1304 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
1305 if ((error = xfs_trans_reserve(tp,
1306 XFS_QM_QINOCREATE_SPACE_RES(mp),
1307 XFS_CREATE_LOG_RES(mp), 0,
1308 XFS_TRANS_PERM_LOG_RES,
1309 XFS_CREATE_LOG_COUNT))) {
1310 xfs_trans_cancel(tp, 0);
1314 if ((error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0,
1315 &xfs_zerocr, 0, 1, ip, &committed))) {
1316 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1322 * Keep an extra reference to this quota inode. This inode is
1323 * locked exclusively and joined to the transaction already.
1325 ASSERT(xfs_isilocked(*ip, XFS_ILOCK_EXCL));
1329 * Make the changes in the superblock, and log those too.
1330 * sbfields arg may contain fields other than *QUOTINO;
1331 * VERSIONNUM for example.
1333 spin_lock(&mp->m_sb_lock);
1334 if (flags & XFS_QMOPT_SBVERSION) {
1335 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1336 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1337 XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
1338 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1339 XFS_SB_GQUOTINO | XFS_SB_QFLAGS));
1341 xfs_sb_version_addquota(&mp->m_sb);
1342 mp->m_sb.sb_uquotino = NULLFSINO;
1343 mp->m_sb.sb_gquotino = NULLFSINO;
1345 /* qflags will get updated _after_ quotacheck */
1346 mp->m_sb.sb_qflags = 0;
1348 if (flags & XFS_QMOPT_UQUOTA)
1349 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1351 mp->m_sb.sb_gquotino = (*ip)->i_ino;
1352 spin_unlock(&mp->m_sb_lock);
1353 xfs_mod_sb(tp, sbfields);
1355 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
1356 xfs_fs_cmn_err(CE_ALERT, mp, "XFS qino_alloc failed!");
1364 xfs_qm_reset_dqcounts(
1370 xfs_disk_dquot_t *ddq;
1373 trace_xfs_reset_dqcounts(bp, _RET_IP_);
1376 * Reset all counters and timers. They'll be
1377 * started afresh by xfs_qm_quotacheck.
1380 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1381 do_div(j, sizeof(xfs_dqblk_t));
1382 ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
1384 ddq = (xfs_disk_dquot_t *)XFS_BUF_PTR(bp);
1385 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
1387 * Do a sanity check, and if needed, repair the dqblk. Don't
1388 * output any warnings because it's perfectly possible to
1389 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1391 (void) xfs_qm_dqcheck(ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1395 ddq->d_rtbcount = 0;
1398 ddq->d_rtbtimer = 0;
1401 ddq->d_rtbwarns = 0;
1402 ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
1411 xfs_filblks_t blkcnt,
1422 incr = (blkcnt > XFS_QM_MAX_DQCLUSTER_LOGSZ) ?
1423 XFS_QM_MAX_DQCLUSTER_LOGSZ : blkcnt;
1424 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1425 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1429 * Blkcnt arg can be a very big number, and might even be
1430 * larger than the log itself. So, we have to break it up into
1431 * manageable-sized transactions.
1432 * Note that we don't start a permanent transaction here; we might
1433 * not be able to get a log reservation for the whole thing up front,
1434 * and we don't really care to either, because we just discard
1435 * everything if we were to crash in the middle of this loop.
1438 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1439 XFS_FSB_TO_DADDR(mp, bno),
1440 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
1444 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1445 xfs_bdwrite(mp, bp);
1447 * goto the next block.
1450 firstid += mp->m_quotainfo->qi_dqperchunk;
1456 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1457 * caller supplied function for every chunk of dquots that we find.
1465 xfs_bmbt_irec_t *map;
1466 int i, nmaps; /* number of map entries */
1467 int error; /* return value */
1468 xfs_fileoff_t lblkno;
1469 xfs_filblks_t maxlblkcnt;
1471 xfs_fsblock_t rablkno;
1472 xfs_filblks_t rablkcnt;
1476 * This looks racy, but we can't keep an inode lock across a
1477 * trans_reserve. But, this gets called during quotacheck, and that
1478 * happens only at mount time which is single threaded.
1480 if (qip->i_d.di_nblocks == 0)
1483 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1486 maxlblkcnt = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1488 nmaps = XFS_DQITER_MAP_SIZE;
1490 * We aren't changing the inode itself. Just changing
1491 * some of its data. No new blocks are added here, and
1492 * the inode is never added to the transaction.
1494 xfs_ilock(qip, XFS_ILOCK_SHARED);
1495 error = xfs_bmapi(NULL, qip, lblkno,
1496 maxlblkcnt - lblkno,
1499 0, map, &nmaps, NULL, NULL);
1500 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1504 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1505 for (i = 0; i < nmaps; i++) {
1506 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1507 ASSERT(map[i].br_blockcount);
1510 lblkno += map[i].br_blockcount;
1512 if (map[i].br_startblock == HOLESTARTBLOCK)
1515 firstid = (xfs_dqid_t) map[i].br_startoff *
1516 mp->m_quotainfo->qi_dqperchunk;
1518 * Do a read-ahead on the next extent.
1520 if ((i+1 < nmaps) &&
1521 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1522 rablkcnt = map[i+1].br_blockcount;
1523 rablkno = map[i+1].br_startblock;
1524 while (rablkcnt--) {
1525 xfs_baread(mp->m_ddev_targp,
1526 XFS_FSB_TO_DADDR(mp, rablkno),
1527 mp->m_quotainfo->qi_dqchunklen);
1532 * Iterate thru all the blks in the extent and
1533 * reset the counters of all the dquots inside them.
1535 if ((error = xfs_qm_dqiter_bufs(mp,
1537 map[i].br_startblock,
1538 map[i].br_blockcount,
1546 } while (nmaps > 0);
1554 * Called by dqusage_adjust in doing a quotacheck.
1555 * Given the inode, and a dquot (either USR or GRP, doesn't matter),
1556 * this updates its incore copy as well as the buffer copy. This is
1557 * so that once the quotacheck is done, we can just log all the buffers,
1558 * as opposed to logging numerous updates to individual dquots.
1561 xfs_qm_quotacheck_dqadjust(
1566 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1568 trace_xfs_dqadjust(dqp);
1571 * Adjust the inode count and the block count to reflect this inode's
1574 be64_add_cpu(&dqp->q_core.d_icount, 1);
1575 dqp->q_res_icount++;
1577 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1578 dqp->q_res_bcount += nblks;
1581 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1582 dqp->q_res_rtbcount += rtblks;
1586 * Set default limits, adjust timers (since we changed usages)
1588 * There are no timers for the default values set in the root dquot.
1590 if (dqp->q_core.d_id) {
1591 xfs_qm_adjust_dqlimits(dqp->q_mount, &dqp->q_core);
1592 xfs_qm_adjust_dqtimers(dqp->q_mount, &dqp->q_core);
1595 dqp->dq_flags |= XFS_DQ_DIRTY;
1601 xfs_qcnt_t *O_rtblks)
1603 xfs_filblks_t rtblks; /* total rt blks */
1604 xfs_extnum_t idx; /* extent record index */
1605 xfs_ifork_t *ifp; /* inode fork pointer */
1606 xfs_extnum_t nextents; /* number of extent entries */
1609 ASSERT(XFS_IS_REALTIME_INODE(ip));
1610 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1611 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1612 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1616 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1617 for (idx = 0; idx < nextents; idx++)
1618 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1619 *O_rtblks = (xfs_qcnt_t)rtblks;
1624 * callback routine supplied to bulkstat(). Given an inumber, find its
1625 * dquots and update them to account for resources taken by that inode.
1629 xfs_qm_dqusage_adjust(
1630 xfs_mount_t *mp, /* mount point for filesystem */
1631 xfs_ino_t ino, /* inode number to get data for */
1632 void __user *buffer, /* not used */
1633 int ubsize, /* not used */
1634 int *ubused, /* not used */
1635 int *res) /* result code value */
1638 xfs_dquot_t *udqp, *gdqp;
1639 xfs_qcnt_t nblks, rtblks;
1642 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1645 * rootino must have its resources accounted for, not so with the quota
1648 if (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino) {
1649 *res = BULKSTAT_RV_NOTHING;
1650 return XFS_ERROR(EINVAL);
1654 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1655 * interface expects the inode to be exclusively locked because that's
1656 * the case in all other instances. It's OK that we do this because
1657 * quotacheck is done only at mount time.
1659 if ((error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip))) {
1660 *res = BULKSTAT_RV_NOTHING;
1665 * Obtain the locked dquots. In case of an error (eg. allocation
1666 * fails for ENOSPC), we return the negative of the error number
1667 * to bulkstat, so that it can get propagated to quotacheck() and
1668 * making us disable quotas for the file system.
1670 if ((error = xfs_qm_dqget_noattach(ip, &udqp, &gdqp))) {
1671 xfs_iput(ip, XFS_ILOCK_EXCL);
1672 *res = BULKSTAT_RV_GIVEUP;
1677 if (! XFS_IS_REALTIME_INODE(ip)) {
1678 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks;
1681 * Walk thru the extent list and count the realtime blocks.
1683 if ((error = xfs_qm_get_rtblks(ip, &rtblks))) {
1684 xfs_iput(ip, XFS_ILOCK_EXCL);
1689 *res = BULKSTAT_RV_GIVEUP;
1692 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1694 ASSERT(ip->i_delayed_blks == 0);
1697 * We can't release the inode while holding its dquot locks.
1698 * The inode can go into inactive and might try to acquire the dquotlocks.
1699 * So, just unlock here and do a vn_rele at the end.
1701 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1704 * Add the (disk blocks and inode) resources occupied by this
1705 * inode to its dquots. We do this adjustment in the incore dquot,
1706 * and also copy the changes to its buffer.
1707 * We don't care about putting these changes in a transaction
1708 * envelope because if we crash in the middle of a 'quotacheck'
1709 * we have to start from the beginning anyway.
1710 * Once we're done, we'll log all the dquot bufs.
1712 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1713 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1715 if (XFS_IS_UQUOTA_ON(mp)) {
1717 xfs_qm_quotacheck_dqadjust(udqp, nblks, rtblks);
1720 if (XFS_IS_OQUOTA_ON(mp)) {
1722 xfs_qm_quotacheck_dqadjust(gdqp, nblks, rtblks);
1726 * Now release the inode. This will send it to 'inactive', and
1727 * possibly even free blocks.
1734 *res = BULKSTAT_RV_DIDONE;
1739 * Walk thru all the filesystem inodes and construct a consistent view
1740 * of the disk quota world. If the quotacheck fails, disable quotas.
1746 int done, count, error;
1749 xfs_inode_t *uip, *gip;
1757 ASSERT(mp->m_quotainfo->qi_uquotaip || mp->m_quotainfo->qi_gquotaip);
1758 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1761 * There should be no cached dquots. The (simplistic) quotacheck
1762 * algorithm doesn't like that.
1764 ASSERT(list_empty(&mp->m_quotainfo->qi_dqlist));
1766 cmn_err(CE_NOTE, "XFS quotacheck %s: Please wait.", mp->m_fsname);
1769 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1770 * their counters to zero. We need a clean slate.
1771 * We don't log our changes till later.
1773 uip = mp->m_quotainfo->qi_uquotaip;
1775 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA);
1778 flags |= XFS_UQUOTA_CHKD;
1781 gip = mp->m_quotainfo->qi_gquotaip;
1783 error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
1784 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1787 flags |= XFS_OQUOTA_CHKD;
1792 * Iterate thru all the inodes in the file system,
1793 * adjusting the corresponding dquot counters in core.
1795 error = xfs_bulkstat(mp, &lastino, &count,
1796 xfs_qm_dqusage_adjust,
1797 structsz, NULL, &done);
1804 * We've made all the changes that we need to make incore.
1805 * Flush them down to disk buffers if everything was updated
1809 error = xfs_qm_dqflush_all(mp, 0);
1812 * We can get this error if we couldn't do a dquot allocation inside
1813 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1814 * dirty dquots that might be cached, we just want to get rid of them
1815 * and turn quotaoff. The dquots won't be attached to any of the inodes
1816 * at this point (because we intentionally didn't in dqget_noattach).
1819 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1824 * We didn't log anything, because if we crashed, we'll have to
1825 * start the quotacheck from scratch anyway. However, we must make
1826 * sure that our dquot changes are secure before we put the
1827 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1830 XFS_bflush(mp->m_ddev_targp);
1833 * If one type of quotas is off, then it will lose its
1834 * quotachecked status, since we won't be doing accounting for
1835 * that type anymore.
1837 mp->m_qflags &= ~(XFS_OQUOTA_CHKD | XFS_UQUOTA_CHKD);
1838 mp->m_qflags |= flags;
1840 xfs_qm_dquot_list_print(mp);
1844 cmn_err(CE_WARN, "XFS quotacheck %s: Unsuccessful (Error %d): "
1845 "Disabling quotas.",
1846 mp->m_fsname, error);
1848 * We must turn off quotas.
1850 ASSERT(mp->m_quotainfo != NULL);
1851 ASSERT(xfs_Gqm != NULL);
1852 xfs_qm_destroy_quotainfo(mp);
1853 if (xfs_mount_reset_sbqflags(mp)) {
1854 cmn_err(CE_WARN, "XFS quotacheck %s: "
1855 "Failed to reset quota flags.", mp->m_fsname);
1858 cmn_err(CE_NOTE, "XFS quotacheck %s: Done.", mp->m_fsname);
1864 * This is called after the superblock has been read in and we're ready to
1865 * iget the quota inodes.
1868 xfs_qm_init_quotainos(
1871 xfs_inode_t *uip, *gip;
1876 ASSERT(mp->m_quotainfo);
1882 * Get the uquota and gquota inodes
1884 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1885 if (XFS_IS_UQUOTA_ON(mp) &&
1886 mp->m_sb.sb_uquotino != NULLFSINO) {
1887 ASSERT(mp->m_sb.sb_uquotino > 0);
1888 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1890 return XFS_ERROR(error);
1892 if (XFS_IS_OQUOTA_ON(mp) &&
1893 mp->m_sb.sb_gquotino != NULLFSINO) {
1894 ASSERT(mp->m_sb.sb_gquotino > 0);
1895 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1899 return XFS_ERROR(error);
1903 flags |= XFS_QMOPT_SBVERSION;
1904 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1905 XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
1909 * Create the two inodes, if they don't exist already. The changes
1910 * made above will get added to a transaction and logged in one of
1911 * the qino_alloc calls below. If the device is readonly,
1912 * temporarily switch to read-write to do this.
1914 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1915 if ((error = xfs_qm_qino_alloc(mp, &uip,
1916 sbflags | XFS_SB_UQUOTINO,
1917 flags | XFS_QMOPT_UQUOTA)))
1918 return XFS_ERROR(error);
1920 flags &= ~XFS_QMOPT_SBVERSION;
1922 if (XFS_IS_OQUOTA_ON(mp) && gip == NULL) {
1923 flags |= (XFS_IS_GQUOTA_ON(mp) ?
1924 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1925 error = xfs_qm_qino_alloc(mp, &gip,
1926 sbflags | XFS_SB_GQUOTINO, flags);
1931 return XFS_ERROR(error);
1935 mp->m_quotainfo->qi_uquotaip = uip;
1936 mp->m_quotainfo->qi_gquotaip = gip;
1944 * Just pop the least recently used dquot off the freelist and
1945 * recycle it. The returned dquot is locked.
1947 STATIC xfs_dquot_t *
1948 xfs_qm_dqreclaim_one(void)
1950 xfs_dquot_t *dqpout;
1957 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
1959 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1961 list_for_each_entry(dqp, &xfs_Gqm->qm_dqfrlist, q_freelist) {
1962 struct xfs_mount *mp = dqp->q_mount;
1966 * We are racing with dqlookup here. Naturally we don't
1967 * want to reclaim a dquot that lookup wants. We release the
1968 * freelist lock and start over, so that lookup will grab
1969 * both the dquot and the freelistlock.
1971 if (dqp->dq_flags & XFS_DQ_WANT) {
1972 ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE));
1974 trace_xfs_dqreclaim_want(dqp);
1977 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1978 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
1980 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
1985 * If the dquot is inactive, we are assured that it is
1986 * not on the mplist or the hashlist, and that makes our
1989 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
1991 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
1992 ASSERT(list_empty(&dqp->q_hashlist));
1993 ASSERT(list_empty(&dqp->q_mplist));
1994 list_del_init(&dqp->q_freelist);
1995 xfs_Gqm->qm_dqfrlist_cnt--;
1998 XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
2002 ASSERT(dqp->q_hash);
2003 ASSERT(!list_empty(&dqp->q_mplist));
2006 * Try to grab the flush lock. If this dquot is in the process of
2007 * getting flushed to disk, we don't want to reclaim it.
2009 if (!xfs_dqflock_nowait(dqp)) {
2015 * We have the flush lock so we know that this is not in the
2016 * process of being flushed. So, if this is dirty, flush it
2017 * DELWRI so that we don't get a freelist infested with
2020 if (XFS_DQ_IS_DIRTY(dqp)) {
2023 trace_xfs_dqreclaim_dirty(dqp);
2026 * We flush it delayed write, so don't bother
2027 * releasing the freelist lock.
2029 error = xfs_qm_dqflush(dqp, 0);
2031 xfs_fs_cmn_err(CE_WARN, mp,
2032 "xfs_qm_dqreclaim: dquot %p flush failed", dqp);
2034 xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
2039 * We're trying to get the hashlock out of order. This races
2040 * with dqlookup; so, we giveup and goto the next dquot if
2041 * we couldn't get the hashlock. This way, we won't starve
2042 * a dqlookup process that holds the hashlock that is
2043 * waiting for the freelist lock.
2045 if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
2051 * This races with dquot allocation code as well as dqflush_all
2052 * and reclaim code. So, if we failed to grab the mplist lock,
2053 * giveup everything and start over.
2055 if (!mutex_trylock(&mp->m_quotainfo->qi_dqlist_lock)) {
2057 mutex_unlock(&dqp->q_hash->qh_lock);
2060 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
2061 if (restarts++ >= XFS_QM_RECLAIM_MAX_RESTARTS)
2066 ASSERT(dqp->q_nrefs == 0);
2067 list_del_init(&dqp->q_mplist);
2068 mp->m_quotainfo->qi_dquots--;
2069 mp->m_quotainfo->qi_dqreclaims++;
2070 list_del_init(&dqp->q_hashlist);
2071 dqp->q_hash->qh_version++;
2072 list_del_init(&dqp->q_freelist);
2073 xfs_Gqm->qm_dqfrlist_cnt--;
2075 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
2076 mutex_unlock(&dqp->q_hash->qh_lock);
2082 if (restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
2085 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
2090 * Traverse the freelist of dquots and attempt to reclaim a maximum of
2091 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
2092 * favor the lookup function ...
2095 xfs_qm_shake_freelist(
2104 while (nreclaimed < howmany) {
2105 dqp = xfs_qm_dqreclaim_one();
2108 xfs_qm_dqdestroy(dqp);
2115 * The kmem_shake interface is invoked when memory is running low.
2120 struct shrinker *shrink,
2124 int ndqused, nfree, n;
2126 if (!kmem_shake_allow(gfp_mask))
2131 nfree = xfs_Gqm->qm_dqfrlist_cnt; /* free dquots */
2132 /* incore dquots in all f/s's */
2133 ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;
2135 ASSERT(ndqused >= 0);
2137 if (nfree <= ndqused && nfree < ndquot)
2140 ndqused *= xfs_Gqm->qm_dqfree_ratio; /* target # of free dquots */
2141 n = nfree - ndqused - ndquot; /* # over target */
2143 return xfs_qm_shake_freelist(MAX(nfree, n));
2147 /*------------------------------------------------------------------*/
2150 * Return a new incore dquot. Depending on the number of
2151 * dquots in the system, we either allocate a new one on the kernel heap,
2152 * or reclaim a free one.
2153 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2154 * to reclaim an existing one from the freelist.
2157 xfs_qm_dqalloc_incore(
2158 xfs_dquot_t **O_dqpp)
2163 * Check against high water mark to see if we want to pop
2164 * a nincompoop dquot off the freelist.
2166 if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
2168 * Try to recycle a dquot from the freelist.
2170 if ((dqp = xfs_qm_dqreclaim_one())) {
2171 XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
2173 * Just zero the core here. The rest will get
2174 * reinitialized by caller. XXX we shouldn't even
2177 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
2181 XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
2185 * Allocate a brand new dquot on the kernel heap and return it
2186 * to the caller to initialize.
2188 ASSERT(xfs_Gqm->qm_dqzone != NULL);
2189 *O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
2190 atomic_inc(&xfs_Gqm->qm_totaldquots);
2197 * Start a transaction and write the incore superblock changes to
2198 * disk. flags parameter indicates which fields have changed.
2201 xfs_qm_write_sb_changes(
2209 cmn_err(CE_NOTE, "Writing superblock quota changes :%s", mp->m_fsname);
2211 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
2212 if ((error = xfs_trans_reserve(tp, 0,
2213 mp->m_sb.sb_sectsize + 128, 0,
2215 XFS_DEFAULT_LOG_COUNT))) {
2216 xfs_trans_cancel(tp, 0);
2220 xfs_mod_sb(tp, flags);
2221 error = xfs_trans_commit(tp, 0);
2227 /* --------------- utility functions for vnodeops ---------------- */
2231 * Given an inode, a uid and gid (from cred_t) make sure that we have
2232 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2233 * quotas by creating this file.
2234 * This also attaches dquot(s) to the given inode after locking it,
2235 * and returns the dquots corresponding to the uid and/or gid.
2237 * in : inode (unlocked)
2238 * out : udquot, gdquot with references taken and unlocked
2242 struct xfs_inode *ip,
2247 struct xfs_dquot **O_udqpp,
2248 struct xfs_dquot **O_gdqpp)
2250 struct xfs_mount *mp = ip->i_mount;
2251 struct xfs_dquot *uq, *gq;
2255 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2258 lockflags = XFS_ILOCK_EXCL;
2259 xfs_ilock(ip, lockflags);
2261 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
2262 gid = ip->i_d.di_gid;
2265 * Attach the dquot(s) to this inode, doing a dquot allocation
2266 * if necessary. The dquot(s) will not be locked.
2268 if (XFS_NOT_DQATTACHED(mp, ip)) {
2269 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
2271 xfs_iunlock(ip, lockflags);
2277 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
2278 if (ip->i_d.di_uid != uid) {
2280 * What we need is the dquot that has this uid, and
2281 * if we send the inode to dqget, the uid of the inode
2282 * takes priority over what's sent in the uid argument.
2283 * We must unlock inode here before calling dqget if
2284 * we're not sending the inode, because otherwise
2285 * we'll deadlock by doing trans_reserve while
2288 xfs_iunlock(ip, lockflags);
2289 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
2294 ASSERT(error != ENOENT);
2298 * Get the ilock in the right order.
2301 lockflags = XFS_ILOCK_SHARED;
2302 xfs_ilock(ip, lockflags);
2305 * Take an extra reference, because we'll return
2308 ASSERT(ip->i_udquot);
2315 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
2316 if (ip->i_d.di_gid != gid) {
2317 xfs_iunlock(ip, lockflags);
2318 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
2325 ASSERT(error != ENOENT);
2329 lockflags = XFS_ILOCK_SHARED;
2330 xfs_ilock(ip, lockflags);
2332 ASSERT(ip->i_gdquot);
2338 } else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
2339 if (ip->i_d.di_projid != prid) {
2340 xfs_iunlock(ip, lockflags);
2341 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
2348 ASSERT(error != ENOENT);
2352 lockflags = XFS_ILOCK_SHARED;
2353 xfs_ilock(ip, lockflags);
2355 ASSERT(ip->i_gdquot);
2363 trace_xfs_dquot_dqalloc(ip);
2365 xfs_iunlock(ip, lockflags);
2378 * Actually transfer ownership, and do dquot modifications.
2379 * These were already reserved.
2385 xfs_dquot_t **IO_olddq,
2388 xfs_dquot_t *prevdq;
2389 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
2390 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
2393 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2394 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
2399 ASSERT(prevdq != newdq);
2401 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
2402 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
2404 /* the sparkling new dquot */
2405 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
2406 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
2409 * Take an extra reference, because the inode
2410 * is going to keep this dquot pointer even
2411 * after the trans_commit.
2415 xfs_dqunlock(newdq);
2422 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
2425 xfs_qm_vop_chown_reserve(
2432 xfs_mount_t *mp = ip->i_mount;
2433 uint delblks, blkflags, prjflags = 0;
2434 xfs_dquot_t *unresudq, *unresgdq, *delblksudq, *delblksgdq;
2438 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2439 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2441 delblks = ip->i_delayed_blks;
2442 delblksudq = delblksgdq = unresudq = unresgdq = NULL;
2443 blkflags = XFS_IS_REALTIME_INODE(ip) ?
2444 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
2446 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2447 ip->i_d.di_uid != (uid_t)be32_to_cpu(udqp->q_core.d_id)) {
2450 * If there are delayed allocation blocks, then we have to
2451 * unreserve those from the old dquot, and add them to the
2455 ASSERT(ip->i_udquot);
2456 unresudq = ip->i_udquot;
2459 if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
2460 if (XFS_IS_PQUOTA_ON(ip->i_mount) &&
2461 ip->i_d.di_projid != be32_to_cpu(gdqp->q_core.d_id))
2462 prjflags = XFS_QMOPT_ENOSPC;
2465 (XFS_IS_GQUOTA_ON(ip->i_mount) &&
2466 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id))) {
2469 ASSERT(ip->i_gdquot);
2470 unresgdq = ip->i_gdquot;
2475 if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2476 delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
2477 flags | blkflags | prjflags)))
2481 * Do the delayed blks reservations/unreservations now. Since, these
2482 * are done without the help of a transaction, if a reservation fails
2483 * its previous reservations won't be automatically undone by trans
2484 * code. So, we have to do it manually here.
2488 * Do the reservations first. Unreservation can't fail.
2490 ASSERT(delblksudq || delblksgdq);
2491 ASSERT(unresudq || unresgdq);
2492 if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2493 delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
2494 flags | blkflags | prjflags)))
2496 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2497 unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
2505 xfs_qm_vop_rename_dqattach(
2506 struct xfs_inode **i_tab)
2508 struct xfs_mount *mp = i_tab[0]->i_mount;
2511 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2514 for (i = 0; (i < 4 && i_tab[i]); i++) {
2515 struct xfs_inode *ip = i_tab[i];
2519 * Watch out for duplicate entries in the table.
2521 if (i == 0 || ip != i_tab[i-1]) {
2522 if (XFS_NOT_DQATTACHED(mp, ip)) {
2523 error = xfs_qm_dqattach(ip, 0);
2533 xfs_qm_vop_create_dqattach(
2534 struct xfs_trans *tp,
2535 struct xfs_inode *ip,
2536 struct xfs_dquot *udqp,
2537 struct xfs_dquot *gdqp)
2539 struct xfs_mount *mp = tp->t_mountp;
2541 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2544 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2545 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2551 ASSERT(ip->i_udquot == NULL);
2552 ip->i_udquot = udqp;
2553 ASSERT(XFS_IS_UQUOTA_ON(mp));
2554 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
2555 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2561 ASSERT(ip->i_gdquot == NULL);
2562 ip->i_gdquot = gdqp;
2563 ASSERT(XFS_IS_OQUOTA_ON(mp));
2564 ASSERT((XFS_IS_GQUOTA_ON(mp) ?
2565 ip->i_d.di_gid : ip->i_d.di_projid) ==
2566 be32_to_cpu(gdqp->q_core.d_id));
2567 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);