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_dmapi.h"
29 #include "xfs_quota.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_btree.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_itable.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
46 #include "xfs_buf_item.h"
47 #include "xfs_trans_space.h"
48 #include "xfs_utils.h"
50 #include "xfs_trace.h"
53 * The global quota manager. There is only one of these for the entire
54 * system, _not_ one per file system. XQM keeps track of the overall
55 * quota functionality, including maintaining the freelist and hash
58 struct mutex xfs_Gqm_lock;
59 struct xfs_qm *xfs_Gqm;
62 kmem_zone_t *qm_dqzone;
63 kmem_zone_t *qm_dqtrxzone;
65 static cred_t xfs_zerocr;
67 STATIC void xfs_qm_list_init(xfs_dqlist_t *, char *, int);
68 STATIC void xfs_qm_list_destroy(xfs_dqlist_t *);
70 STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
71 STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
72 STATIC int xfs_qm_shake(int, gfp_t);
74 static struct shrinker xfs_qm_shaker = {
75 .shrink = xfs_qm_shake,
76 .seeks = DEFAULT_SEEKS,
80 extern struct mutex qcheck_lock;
85 xfs_qm_dquot_list_print(
91 list_for_each_entry(dqp, &mp->m_quotainfo->qi_dqlist_lock, qi_mplist) {
92 cmn_err(CE_DEBUG, " %d. \"%d (%s)\" "
93 "bcnt = %lld, icnt = %lld, refs = %d",
94 i++, be32_to_cpu(dqp->q_core.d_id),
95 DQFLAGTO_TYPESTR(dqp),
96 (long long)be64_to_cpu(dqp->q_core.d_bcount),
97 (long long)be64_to_cpu(dqp->q_core.d_icount),
102 static void xfs_qm_dquot_list_print(struct xfs_mount *mp) { }
106 * Initialize the XQM structure.
107 * Note that there is not one quota manager per file system.
109 STATIC struct xfs_qm *
112 xfs_dqhash_t *udqhash, *gdqhash;
118 * Initialize the dquot hash tables.
120 udqhash = kmem_zalloc_greedy(&hsize,
121 XFS_QM_HASHSIZE_LOW * sizeof(xfs_dqhash_t),
122 XFS_QM_HASHSIZE_HIGH * sizeof(xfs_dqhash_t));
126 gdqhash = kmem_zalloc_large(hsize);
128 goto out_free_udqhash;
130 hsize /= sizeof(xfs_dqhash_t);
133 xqm = kmem_zalloc(sizeof(xfs_qm_t), KM_SLEEP);
134 xqm->qm_dqhashmask = hsize - 1;
135 xqm->qm_usr_dqhtable = udqhash;
136 xqm->qm_grp_dqhtable = gdqhash;
137 ASSERT(xqm->qm_usr_dqhtable != NULL);
138 ASSERT(xqm->qm_grp_dqhtable != NULL);
140 for (i = 0; i < hsize; i++) {
141 xfs_qm_list_init(&(xqm->qm_usr_dqhtable[i]), "uxdqh", i);
142 xfs_qm_list_init(&(xqm->qm_grp_dqhtable[i]), "gxdqh", i);
146 * Freelist of all dquots of all file systems
148 INIT_LIST_HEAD(&xqm->qm_dqfrlist);
149 xqm->qm_dqfrlist_cnt = 0;
150 mutex_init(&xqm->qm_dqfrlist_lock);
153 * dquot zone. we register our own low-memory callback.
156 xqm->qm_dqzone = kmem_zone_init(sizeof(xfs_dquot_t),
158 qm_dqzone = xqm->qm_dqzone;
160 xqm->qm_dqzone = qm_dqzone;
162 register_shrinker(&xfs_qm_shaker);
165 * The t_dqinfo portion of transactions.
168 xqm->qm_dqtrxzone = kmem_zone_init(sizeof(xfs_dquot_acct_t),
170 qm_dqtrxzone = xqm->qm_dqtrxzone;
172 xqm->qm_dqtrxzone = qm_dqtrxzone;
174 atomic_set(&xqm->qm_totaldquots, 0);
175 xqm->qm_dqfree_ratio = XFS_QM_DQFREE_RATIO;
178 mutex_init(&qcheck_lock);
183 kmem_free_large(udqhash);
189 * Destroy the global quota manager when its reference count goes to zero.
195 struct xfs_dquot *dqp, *n;
199 ASSERT(xqm->qm_nrefs == 0);
200 unregister_shrinker(&xfs_qm_shaker);
201 hsize = xqm->qm_dqhashmask + 1;
202 for (i = 0; i < hsize; i++) {
203 xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
204 xfs_qm_list_destroy(&(xqm->qm_grp_dqhtable[i]));
206 kmem_free_large(xqm->qm_usr_dqhtable);
207 kmem_free_large(xqm->qm_grp_dqhtable);
208 xqm->qm_usr_dqhtable = NULL;
209 xqm->qm_grp_dqhtable = NULL;
210 xqm->qm_dqhashmask = 0;
213 mutex_lock(&xqm->qm_dqfrlist_lock);
214 list_for_each_entry_safe(dqp, n, &xqm->qm_dqfrlist, q_freelist) {
217 cmn_err(CE_DEBUG, "FREELIST destroy 0x%p", dqp);
219 list_del_init(&dqp->q_freelist);
220 xfs_Gqm->qm_dqfrlist_cnt--;
222 xfs_qm_dqdestroy(dqp);
224 mutex_unlock(&xqm->qm_dqfrlist_lock);
225 mutex_destroy(&xqm->qm_dqfrlist_lock);
227 mutex_destroy(&qcheck_lock);
233 * Called at mount time to let XQM know that another file system is
234 * starting quotas. This isn't crucial information as the individual mount
235 * structures are pretty independent, but it helps the XQM keep a
236 * global view of what's going on.
240 xfs_qm_hold_quotafs_ref(
241 struct xfs_mount *mp)
244 * Need to lock the xfs_Gqm structure for things like this. For example,
245 * the structure could disappear between the entry to this routine and
246 * a HOLD operation if not locked.
248 mutex_lock(&xfs_Gqm_lock);
251 xfs_Gqm = xfs_Gqm_init();
257 * We can keep a list of all filesystems with quotas mounted for
258 * debugging and statistical purposes, but ...
259 * Just take a reference and get out.
262 mutex_unlock(&xfs_Gqm_lock);
269 * Release the reference that a filesystem took at mount time,
270 * so that we know when we need to destroy the entire quota manager.
274 xfs_qm_rele_quotafs_ref(
275 struct xfs_mount *mp)
277 xfs_dquot_t *dqp, *n;
280 ASSERT(xfs_Gqm->qm_nrefs > 0);
283 * Go thru the freelist and destroy all inactive dquots.
285 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
287 list_for_each_entry_safe(dqp, n, &xfs_Gqm->qm_dqfrlist, q_freelist) {
289 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
290 ASSERT(dqp->q_mount == NULL);
291 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
292 ASSERT(list_empty(&dqp->q_hashlist));
293 ASSERT(list_empty(&dqp->q_mplist));
294 list_del_init(&dqp->q_freelist);
295 xfs_Gqm->qm_dqfrlist_cnt--;
297 xfs_qm_dqdestroy(dqp);
302 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
305 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
308 mutex_lock(&xfs_Gqm_lock);
309 if (--xfs_Gqm->qm_nrefs == 0) {
310 xfs_qm_destroy(xfs_Gqm);
313 mutex_unlock(&xfs_Gqm_lock);
317 * Just destroy the quotainfo structure.
321 struct xfs_mount *mp)
323 if (mp->m_quotainfo) {
324 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
325 xfs_qm_destroy_quotainfo(mp);
331 * This is called from xfs_mountfs to start quotas and initialize all
332 * necessary data structures like quotainfo. This is also responsible for
333 * running a quotacheck as necessary. We are guaranteed that the superblock
334 * is consistently read in at this point.
336 * If we fail here, the mount will continue with quota turned off. We don't
337 * need to inidicate success or failure at all.
347 * If quotas on realtime volumes is not supported, we disable
348 * quotas immediately.
350 if (mp->m_sb.sb_rextents) {
352 "Cannot turn on quotas for realtime filesystem %s",
358 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
361 * Allocate the quotainfo structure inside the mount struct, and
362 * create quotainode(s), and change/rev superblock if necessary.
364 error = xfs_qm_init_quotainfo(mp);
367 * We must turn off quotas.
369 ASSERT(mp->m_quotainfo == NULL);
374 * If any of the quotas are not consistent, do a quotacheck.
376 if (XFS_QM_NEED_QUOTACHECK(mp)) {
377 error = xfs_qm_quotacheck(mp);
379 /* Quotacheck failed and disabled quotas. */
384 * If one type of quotas is off, then it will lose its
385 * quotachecked status, since we won't be doing accounting for
388 if (!XFS_IS_UQUOTA_ON(mp))
389 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
390 if (!(XFS_IS_GQUOTA_ON(mp) || XFS_IS_PQUOTA_ON(mp)))
391 mp->m_qflags &= ~XFS_OQUOTA_CHKD;
395 * We actually don't have to acquire the m_sb_lock at all.
396 * This can only be called from mount, and that's single threaded. XXX
398 spin_lock(&mp->m_sb_lock);
399 sbf = mp->m_sb.sb_qflags;
400 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
401 spin_unlock(&mp->m_sb_lock);
403 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
404 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
406 * We could only have been turning quotas off.
407 * We aren't in very good shape actually because
408 * the incore structures are convinced that quotas are
409 * off, but the on disk superblock doesn't know that !
411 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
412 xfs_fs_cmn_err(CE_ALERT, mp,
413 "XFS mount_quotas: Superblock update failed!");
418 xfs_fs_cmn_err(CE_WARN, mp,
419 "Failed to initialize disk quotas.");
424 if (XFS_IS_QUOTA_ON(mp))
425 xfs_qm_internalqcheck(mp);
430 * Called from the vfsops layer.
433 xfs_qm_unmount_quotas(
437 * Release the dquots that root inode, et al might be holding,
438 * before we flush quotas and blow away the quotainfo structure.
440 ASSERT(mp->m_rootip);
441 xfs_qm_dqdetach(mp->m_rootip);
443 xfs_qm_dqdetach(mp->m_rbmip);
445 xfs_qm_dqdetach(mp->m_rsumip);
448 * Release the quota inodes.
450 if (mp->m_quotainfo) {
451 if (mp->m_quotainfo->qi_uquotaip) {
452 IRELE(mp->m_quotainfo->qi_uquotaip);
453 mp->m_quotainfo->qi_uquotaip = NULL;
455 if (mp->m_quotainfo->qi_gquotaip) {
456 IRELE(mp->m_quotainfo->qi_gquotaip);
457 mp->m_quotainfo->qi_gquotaip = NULL;
463 * Flush all dquots of the given file system to disk. The dquots are
464 * _not_ purged from memory here, just their data written to disk.
468 struct xfs_mount *mp,
471 struct xfs_quotainfo *q = mp->m_quotainfo;
473 struct xfs_dquot *dqp;
481 mutex_lock(&q->qi_dqlist_lock);
482 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
484 if (! XFS_DQ_IS_DIRTY(dqp)) {
489 /* XXX a sentinel would be better */
490 recl = q->qi_dqreclaims;
491 if (!xfs_dqflock_nowait(dqp)) {
493 * If we can't grab the flush lock then check
494 * to see if the dquot has been flushed delayed
495 * write. If so, grab its buffer and send it
496 * out immediately. We'll be able to acquire
497 * the flush lock when the I/O completes.
499 xfs_qm_dqflock_pushbuf_wait(dqp);
502 * Let go of the mplist lock. We don't want to hold it
503 * across a disk write.
505 mutex_unlock(&q->qi_dqlist_lock);
506 error = xfs_qm_dqflush(dqp, sync_mode);
511 mutex_lock(&q->qi_dqlist_lock);
512 if (recl != q->qi_dqreclaims) {
513 mutex_unlock(&q->qi_dqlist_lock);
514 /* XXX restart limit */
519 mutex_unlock(&q->qi_dqlist_lock);
524 * Release the group dquot pointers the user dquots may be
525 * carrying around as a hint. mplist is locked on entry and exit.
528 xfs_qm_detach_gdquots(
529 struct xfs_mount *mp)
531 struct xfs_quotainfo *q = mp->m_quotainfo;
532 struct xfs_dquot *dqp, *gdqp;
536 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
537 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
539 if ((gdqp = dqp->q_gdquot)) {
541 dqp->q_gdquot = NULL;
547 * Can't hold the mplist lock across a dqput.
548 * XXXmust convert to marker based iterations here.
550 nrecl = q->qi_dqreclaims;
551 mutex_unlock(&q->qi_dqlist_lock);
554 mutex_lock(&q->qi_dqlist_lock);
555 if (nrecl != q->qi_dqreclaims)
562 * Go through all the incore dquots of this file system and take them
563 * off the mplist and hashlist, if the dquot type matches the dqtype
564 * parameter. This is used when turning off quota accounting for
565 * users and/or groups, as well as when the filesystem is unmounting.
569 struct xfs_mount *mp,
572 struct xfs_quotainfo *q = mp->m_quotainfo;
573 struct xfs_dquot *dqp, *n;
581 dqtype = (flags & XFS_QMOPT_UQUOTA) ? XFS_DQ_USER : 0;
582 dqtype |= (flags & XFS_QMOPT_PQUOTA) ? XFS_DQ_PROJ : 0;
583 dqtype |= (flags & XFS_QMOPT_GQUOTA) ? XFS_DQ_GROUP : 0;
585 mutex_lock(&q->qi_dqlist_lock);
588 * In the first pass through all incore dquots of this filesystem,
589 * we release the group dquot pointers the user dquots may be
590 * carrying around as a hint. We need to do this irrespective of
591 * what's being turned off.
593 xfs_qm_detach_gdquots(mp);
597 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
599 * Try to get rid of all of the unwanted dquots. The idea is to
600 * get them off mplist and hashlist, but leave them on freelist.
602 list_for_each_entry_safe(dqp, n, &q->qi_dqlist, q_mplist) {
604 * It's OK to look at the type without taking dqlock here.
605 * We're holding the mplist lock here, and that's needed for
608 if ((dqp->dq_flags & dqtype) == 0)
611 if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
612 nrecl = q->qi_dqreclaims;
613 mutex_unlock(&q->qi_dqlist_lock);
614 mutex_lock(&dqp->q_hash->qh_lock);
615 mutex_lock(&q->qi_dqlist_lock);
618 * XXXTheoretically, we can get into a very long
619 * ping pong game here.
620 * No one can be adding dquots to the mplist at
621 * this point, but somebody might be taking things off.
623 if (nrecl != q->qi_dqreclaims) {
624 mutex_unlock(&dqp->q_hash->qh_lock);
630 * Take the dquot off the mplist and hashlist. It may remain on
631 * freelist in INACTIVE state.
633 nmisses += xfs_qm_dqpurge(dqp);
635 mutex_unlock(&q->qi_dqlist_lock);
647 * Purge the dquot cache.
648 * None of the dquots should really be busy at this point.
650 if (mp->m_quotainfo) {
651 while ((ndquots = xfs_qm_dqpurge_int(mp, flags))) {
664 xfs_dquot_t *udqhint, /* hint */
665 xfs_dquot_t **IO_idqpp)
670 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
674 * See if we already have it in the inode itself. IO_idqpp is
675 * &i_udquot or &i_gdquot. This made the code look weird, but
676 * made the logic a lot simpler.
680 trace_xfs_dqattach_found(dqp);
685 * udqhint is the i_udquot field in inode, and is non-NULL only
686 * when the type arg is group/project. Its purpose is to save a
687 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
691 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
695 * No need to take dqlock to look at the id.
697 * The ID can't change until it gets reclaimed, and it won't
698 * be reclaimed as long as we have a ref from inode and we
701 dqp = udqhint->q_gdquot;
702 if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
705 ASSERT(*IO_idqpp == NULL);
709 xfs_dqunlock(udqhint);
714 * We can't hold a dquot lock when we call the dqget code.
715 * We'll deadlock in no time, because of (not conforming to)
716 * lock ordering - the inodelock comes before any dquot lock,
717 * and we may drop and reacquire the ilock in xfs_qm_dqget().
719 xfs_dqunlock(udqhint);
723 * Find the dquot from somewhere. This bumps the
724 * reference count of dquot and returns it locked.
725 * This can return ENOENT if dquot didn't exist on
726 * disk and we didn't ask it to allocate;
727 * ESRCH if quotas got turned off suddenly.
729 error = xfs_qm_dqget(ip->i_mount, ip, id, type, XFS_QMOPT_DOWARN, &dqp);
733 trace_xfs_dqattach_get(dqp);
736 * dqget may have dropped and re-acquired the ilock, but it guarantees
737 * that the dquot returned is the one that should go in the inode.
746 * Given a udquot and gdquot, attach a ptr to the group dquot in the
747 * udquot as a hint for future lookups. The idea sounds simple, but the
748 * execution isn't, because the udquot might have a group dquot attached
749 * already and getting rid of that gets us into lock ordering constraints.
750 * The process is complicated more by the fact that the dquots may or may not
751 * be locked on entry.
754 xfs_qm_dqattach_grouphint(
762 if ((tmp = udq->q_gdquot)) {
768 udq->q_gdquot = NULL;
770 * We can't keep any dqlocks when calling dqrele,
771 * because the freelist lock comes before dqlocks.
775 * we took a hard reference once upon a time in dqget,
776 * so give it back when the udquot no longer points at it
777 * dqput() does the unlocking of the dquot.
785 ASSERT(XFS_DQ_IS_LOCKED(udq));
789 ASSERT(XFS_DQ_IS_LOCKED(udq));
790 ASSERT(XFS_DQ_IS_LOCKED(gdq));
792 * Somebody could have attached a gdquot here,
793 * when we dropped the uqlock. If so, just do nothing.
795 if (udq->q_gdquot == NULL) {
806 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
808 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
809 * Inode may get unlocked and relocked in here, and the caller must deal with
813 xfs_qm_dqattach_locked(
817 xfs_mount_t *mp = ip->i_mount;
821 if (!XFS_IS_QUOTA_RUNNING(mp) ||
822 !XFS_IS_QUOTA_ON(mp) ||
823 !XFS_NOT_DQATTACHED(mp, ip) ||
824 ip->i_ino == mp->m_sb.sb_uquotino ||
825 ip->i_ino == mp->m_sb.sb_gquotino)
828 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
830 if (XFS_IS_UQUOTA_ON(mp)) {
831 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
832 flags & XFS_QMOPT_DQALLOC,
833 NULL, &ip->i_udquot);
839 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
840 if (XFS_IS_OQUOTA_ON(mp)) {
841 error = XFS_IS_GQUOTA_ON(mp) ?
842 xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
843 flags & XFS_QMOPT_DQALLOC,
844 ip->i_udquot, &ip->i_gdquot) :
845 xfs_qm_dqattach_one(ip, ip->i_d.di_projid, XFS_DQ_PROJ,
846 flags & XFS_QMOPT_DQALLOC,
847 ip->i_udquot, &ip->i_gdquot);
849 * Don't worry about the udquot that we may have
850 * attached above. It'll get detached, if not already.
858 * Attach this group quota to the user quota as a hint.
859 * This WON'T, in general, result in a thrash.
862 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
863 ASSERT(ip->i_udquot);
864 ASSERT(ip->i_gdquot);
867 * We may or may not have the i_udquot locked at this point,
868 * but this check is OK since we don't depend on the i_gdquot to
869 * be accurate 100% all the time. It is just a hint, and this
870 * will succeed in general.
872 if (ip->i_udquot->q_gdquot == ip->i_gdquot)
875 * Attach i_gdquot to the gdquot hint inside the i_udquot.
877 xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot);
883 if (XFS_IS_UQUOTA_ON(mp))
884 ASSERT(ip->i_udquot);
885 if (XFS_IS_OQUOTA_ON(mp))
886 ASSERT(ip->i_gdquot);
888 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
895 struct xfs_inode *ip,
900 xfs_ilock(ip, XFS_ILOCK_EXCL);
901 error = xfs_qm_dqattach_locked(ip, flags);
902 xfs_iunlock(ip, XFS_ILOCK_EXCL);
908 * Release dquots (and their references) if any.
909 * The inode should be locked EXCL except when this's called by
916 if (!(ip->i_udquot || ip->i_gdquot))
919 trace_xfs_dquot_dqdetach(ip);
921 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
922 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
924 xfs_qm_dqrele(ip->i_udquot);
928 xfs_qm_dqrele(ip->i_gdquot);
935 struct xfs_mount *mp,
938 struct xfs_quotainfo *q = mp->m_quotainfo;
940 struct xfs_dquot *dqp;
943 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
949 mutex_lock(&q->qi_dqlist_lock);
951 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
952 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
953 * when we have the mplist lock, we know that dquots will be consistent
954 * as long as we have it locked.
956 if (!XFS_IS_QUOTA_ON(mp)) {
957 mutex_unlock(&q->qi_dqlist_lock);
960 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
961 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
963 * If this is vfs_sync calling, then skip the dquots that
964 * don't 'seem' to be dirty. ie. don't acquire dqlock.
965 * This is very similar to what xfs_sync does with inodes.
967 if (flags & SYNC_TRYLOCK) {
968 if (!XFS_DQ_IS_DIRTY(dqp))
970 if (!xfs_qm_dqlock_nowait(dqp))
977 * Now, find out for sure if this dquot is dirty or not.
979 if (! XFS_DQ_IS_DIRTY(dqp)) {
984 /* XXX a sentinel would be better */
985 recl = q->qi_dqreclaims;
986 if (!xfs_dqflock_nowait(dqp)) {
987 if (flags & SYNC_TRYLOCK) {
992 * If we can't grab the flush lock then if the caller
993 * really wanted us to give this our best shot, so
994 * see if we can give a push to the buffer before we wait
995 * on the flush lock. At this point, we know that
996 * even though the dquot is being flushed,
997 * it has (new) dirty data.
999 xfs_qm_dqflock_pushbuf_wait(dqp);
1002 * Let go of the mplist lock. We don't want to hold it
1003 * across a disk write
1005 mutex_unlock(&q->qi_dqlist_lock);
1006 error = xfs_qm_dqflush(dqp, flags);
1008 if (error && XFS_FORCED_SHUTDOWN(mp))
1009 return 0; /* Need to prevent umount failure */
1013 mutex_lock(&q->qi_dqlist_lock);
1014 if (recl != q->qi_dqreclaims) {
1015 if (++restarts >= XFS_QM_SYNC_MAX_RESTARTS)
1018 mutex_unlock(&q->qi_dqlist_lock);
1023 mutex_unlock(&q->qi_dqlist_lock);
1028 * The hash chains and the mplist use the same xfs_dqhash structure as
1029 * their list head, but we can take the mplist qh_lock and one of the
1030 * hash qh_locks at the same time without any problem as they aren't
1033 static struct lock_class_key xfs_quota_mplist_class;
1036 * This initializes all the quota information that's kept in the
1040 xfs_qm_init_quotainfo(
1043 xfs_quotainfo_t *qinf;
1047 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1050 * Tell XQM that we exist as soon as possible.
1052 if ((error = xfs_qm_hold_quotafs_ref(mp))) {
1056 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
1059 * See if quotainodes are setup, and if not, allocate them,
1060 * and change the superblock accordingly.
1062 if ((error = xfs_qm_init_quotainos(mp))) {
1064 mp->m_quotainfo = NULL;
1068 INIT_LIST_HEAD(&qinf->qi_dqlist);
1069 mutex_init(&qinf->qi_dqlist_lock);
1070 lockdep_set_class(&qinf->qi_dqlist_lock, &xfs_quota_mplist_class);
1072 qinf->qi_dqreclaims = 0;
1074 /* mutex used to serialize quotaoffs */
1075 mutex_init(&qinf->qi_quotaofflock);
1077 /* Precalc some constants */
1078 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1079 ASSERT(qinf->qi_dqchunklen);
1080 qinf->qi_dqperchunk = BBTOB(qinf->qi_dqchunklen);
1081 do_div(qinf->qi_dqperchunk, sizeof(xfs_dqblk_t));
1083 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
1086 * We try to get the limits from the superuser's limits fields.
1087 * This is quite hacky, but it is standard quota practice.
1088 * We look at the USR dquot with id == 0 first, but if user quotas
1089 * are not enabled we goto the GRP dquot with id == 0.
1090 * We don't really care to keep separate default limits for user
1091 * and group quotas, at least not at this point.
1093 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)0,
1094 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
1095 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
1097 XFS_QMOPT_DQSUSER|XFS_QMOPT_DOWARN,
1100 xfs_disk_dquot_t *ddqp = &dqp->q_core;
1103 * The warnings and timers set the grace period given to
1104 * a user or group before he or she can not perform any
1105 * more writing. If it is zero, a default is used.
1107 qinf->qi_btimelimit = ddqp->d_btimer ?
1108 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
1109 qinf->qi_itimelimit = ddqp->d_itimer ?
1110 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
1111 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
1112 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
1113 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
1114 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
1115 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
1116 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
1117 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
1118 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
1119 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
1120 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
1121 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
1122 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
1123 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
1124 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
1127 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1128 * we don't want this dquot cached. We haven't done a
1129 * quotacheck yet, and quotacheck doesn't like incore dquots.
1131 xfs_qm_dqdestroy(dqp);
1133 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
1134 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
1135 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
1136 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
1137 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
1138 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
1146 * Gets called when unmounting a filesystem or when all quotas get
1148 * This purges the quota inodes, destroys locks and frees itself.
1151 xfs_qm_destroy_quotainfo(
1154 xfs_quotainfo_t *qi;
1156 qi = mp->m_quotainfo;
1158 ASSERT(xfs_Gqm != NULL);
1161 * Release the reference that XQM kept, so that we know
1162 * when the XQM structure should be freed. We cannot assume
1163 * that xfs_Gqm is non-null after this point.
1165 xfs_qm_rele_quotafs_ref(mp);
1167 ASSERT(list_empty(&qi->qi_dqlist));
1168 mutex_destroy(&qi->qi_dqlist_lock);
1170 if (qi->qi_uquotaip) {
1171 IRELE(qi->qi_uquotaip);
1172 qi->qi_uquotaip = NULL; /* paranoia */
1174 if (qi->qi_gquotaip) {
1175 IRELE(qi->qi_gquotaip);
1176 qi->qi_gquotaip = NULL;
1178 mutex_destroy(&qi->qi_quotaofflock);
1180 mp->m_quotainfo = NULL;
1185 /* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1194 mutex_init(&list->qh_lock);
1195 INIT_LIST_HEAD(&list->qh_list);
1196 list->qh_version = 0;
1197 list->qh_nelems = 0;
1201 xfs_qm_list_destroy(
1204 mutex_destroy(&(list->qh_lock));
1209 * Stripped down version of dqattach. This doesn't attach, or even look at the
1210 * dquots attached to the inode. The rationale is that there won't be any
1211 * attached at the time this is called from quotacheck.
1214 xfs_qm_dqget_noattach(
1216 xfs_dquot_t **O_udqpp,
1217 xfs_dquot_t **O_gdqpp)
1221 xfs_dquot_t *udqp, *gdqp;
1223 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1228 if (XFS_IS_UQUOTA_ON(mp)) {
1229 ASSERT(ip->i_udquot == NULL);
1231 * We want the dquot allocated if it doesn't exist.
1233 if ((error = xfs_qm_dqget(mp, ip, ip->i_d.di_uid, XFS_DQ_USER,
1234 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN,
1237 * Shouldn't be able to turn off quotas here.
1239 ASSERT(error != ESRCH);
1240 ASSERT(error != ENOENT);
1246 if (XFS_IS_OQUOTA_ON(mp)) {
1247 ASSERT(ip->i_gdquot == NULL);
1250 error = XFS_IS_GQUOTA_ON(mp) ?
1251 xfs_qm_dqget(mp, ip,
1252 ip->i_d.di_gid, XFS_DQ_GROUP,
1253 XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
1255 xfs_qm_dqget(mp, ip,
1256 ip->i_d.di_projid, XFS_DQ_PROJ,
1257 XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
1261 xfs_qm_dqrele(udqp);
1262 ASSERT(error != ESRCH);
1263 ASSERT(error != ENOENT);
1268 /* Reacquire the locks in the right order */
1270 if (! xfs_qm_dqlock_nowait(udqp)) {
1282 if (udqp) ASSERT(XFS_DQ_IS_LOCKED(udqp));
1283 if (gdqp) ASSERT(XFS_DQ_IS_LOCKED(gdqp));
1289 * Create an inode and return with a reference already taken, but unlocked
1290 * This is how we create quota inodes
1303 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
1304 if ((error = xfs_trans_reserve(tp,
1305 XFS_QM_QINOCREATE_SPACE_RES(mp),
1306 XFS_CREATE_LOG_RES(mp), 0,
1307 XFS_TRANS_PERM_LOG_RES,
1308 XFS_CREATE_LOG_COUNT))) {
1309 xfs_trans_cancel(tp, 0);
1313 if ((error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0,
1314 &xfs_zerocr, 0, 1, ip, &committed))) {
1315 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1321 * Keep an extra reference to this quota inode. This inode is
1322 * locked exclusively and joined to the transaction already.
1324 ASSERT(xfs_isilocked(*ip, XFS_ILOCK_EXCL));
1328 * Make the changes in the superblock, and log those too.
1329 * sbfields arg may contain fields other than *QUOTINO;
1330 * VERSIONNUM for example.
1332 spin_lock(&mp->m_sb_lock);
1333 if (flags & XFS_QMOPT_SBVERSION) {
1334 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1335 unsigned oldv = mp->m_sb.sb_versionnum;
1337 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1338 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1339 XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
1340 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1341 XFS_SB_GQUOTINO | XFS_SB_QFLAGS));
1343 xfs_sb_version_addquota(&mp->m_sb);
1344 mp->m_sb.sb_uquotino = NULLFSINO;
1345 mp->m_sb.sb_gquotino = NULLFSINO;
1347 /* qflags will get updated _after_ quotacheck */
1348 mp->m_sb.sb_qflags = 0;
1349 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1351 "Old superblock version %x, converting to %x.",
1352 oldv, mp->m_sb.sb_versionnum);
1355 if (flags & XFS_QMOPT_UQUOTA)
1356 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1358 mp->m_sb.sb_gquotino = (*ip)->i_ino;
1359 spin_unlock(&mp->m_sb_lock);
1360 xfs_mod_sb(tp, sbfields);
1362 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
1363 xfs_fs_cmn_err(CE_ALERT, mp, "XFS qino_alloc failed!");
1371 xfs_qm_reset_dqcounts(
1377 xfs_disk_dquot_t *ddq;
1380 trace_xfs_reset_dqcounts(bp, _RET_IP_);
1383 * Reset all counters and timers. They'll be
1384 * started afresh by xfs_qm_quotacheck.
1387 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1388 do_div(j, sizeof(xfs_dqblk_t));
1389 ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
1391 ddq = (xfs_disk_dquot_t *)XFS_BUF_PTR(bp);
1392 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
1394 * Do a sanity check, and if needed, repair the dqblk. Don't
1395 * output any warnings because it's perfectly possible to
1396 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1398 (void) xfs_qm_dqcheck(ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1402 ddq->d_rtbcount = 0;
1405 ddq->d_rtbtimer = 0;
1408 ddq->d_rtbwarns = 0;
1409 ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
1418 xfs_filblks_t blkcnt,
1429 incr = (blkcnt > XFS_QM_MAX_DQCLUSTER_LOGSZ) ?
1430 XFS_QM_MAX_DQCLUSTER_LOGSZ : blkcnt;
1431 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1432 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1436 * Blkcnt arg can be a very big number, and might even be
1437 * larger than the log itself. So, we have to break it up into
1438 * manageable-sized transactions.
1439 * Note that we don't start a permanent transaction here; we might
1440 * not be able to get a log reservation for the whole thing up front,
1441 * and we don't really care to either, because we just discard
1442 * everything if we were to crash in the middle of this loop.
1445 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1446 XFS_FSB_TO_DADDR(mp, bno),
1447 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
1451 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1452 xfs_bdwrite(mp, bp);
1454 * goto the next block.
1457 firstid += mp->m_quotainfo->qi_dqperchunk;
1463 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1464 * caller supplied function for every chunk of dquots that we find.
1472 xfs_bmbt_irec_t *map;
1473 int i, nmaps; /* number of map entries */
1474 int error; /* return value */
1475 xfs_fileoff_t lblkno;
1476 xfs_filblks_t maxlblkcnt;
1478 xfs_fsblock_t rablkno;
1479 xfs_filblks_t rablkcnt;
1483 * This looks racy, but we can't keep an inode lock across a
1484 * trans_reserve. But, this gets called during quotacheck, and that
1485 * happens only at mount time which is single threaded.
1487 if (qip->i_d.di_nblocks == 0)
1490 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1493 maxlblkcnt = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1495 nmaps = XFS_DQITER_MAP_SIZE;
1497 * We aren't changing the inode itself. Just changing
1498 * some of its data. No new blocks are added here, and
1499 * the inode is never added to the transaction.
1501 xfs_ilock(qip, XFS_ILOCK_SHARED);
1502 error = xfs_bmapi(NULL, qip, lblkno,
1503 maxlblkcnt - lblkno,
1506 0, map, &nmaps, NULL, NULL);
1507 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1511 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1512 for (i = 0; i < nmaps; i++) {
1513 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1514 ASSERT(map[i].br_blockcount);
1517 lblkno += map[i].br_blockcount;
1519 if (map[i].br_startblock == HOLESTARTBLOCK)
1522 firstid = (xfs_dqid_t) map[i].br_startoff *
1523 mp->m_quotainfo->qi_dqperchunk;
1525 * Do a read-ahead on the next extent.
1527 if ((i+1 < nmaps) &&
1528 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1529 rablkcnt = map[i+1].br_blockcount;
1530 rablkno = map[i+1].br_startblock;
1531 while (rablkcnt--) {
1532 xfs_baread(mp->m_ddev_targp,
1533 XFS_FSB_TO_DADDR(mp, rablkno),
1534 mp->m_quotainfo->qi_dqchunklen);
1539 * Iterate thru all the blks in the extent and
1540 * reset the counters of all the dquots inside them.
1542 if ((error = xfs_qm_dqiter_bufs(mp,
1544 map[i].br_startblock,
1545 map[i].br_blockcount,
1553 } while (nmaps > 0);
1561 * Called by dqusage_adjust in doing a quotacheck.
1562 * Given the inode, and a dquot (either USR or GRP, doesn't matter),
1563 * this updates its incore copy as well as the buffer copy. This is
1564 * so that once the quotacheck is done, we can just log all the buffers,
1565 * as opposed to logging numerous updates to individual dquots.
1568 xfs_qm_quotacheck_dqadjust(
1573 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1575 trace_xfs_dqadjust(dqp);
1578 * Adjust the inode count and the block count to reflect this inode's
1581 be64_add_cpu(&dqp->q_core.d_icount, 1);
1582 dqp->q_res_icount++;
1584 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1585 dqp->q_res_bcount += nblks;
1588 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1589 dqp->q_res_rtbcount += rtblks;
1593 * Set default limits, adjust timers (since we changed usages)
1595 * There are no timers for the default values set in the root dquot.
1597 if (dqp->q_core.d_id) {
1598 xfs_qm_adjust_dqlimits(dqp->q_mount, &dqp->q_core);
1599 xfs_qm_adjust_dqtimers(dqp->q_mount, &dqp->q_core);
1602 dqp->dq_flags |= XFS_DQ_DIRTY;
1608 xfs_qcnt_t *O_rtblks)
1610 xfs_filblks_t rtblks; /* total rt blks */
1611 xfs_extnum_t idx; /* extent record index */
1612 xfs_ifork_t *ifp; /* inode fork pointer */
1613 xfs_extnum_t nextents; /* number of extent entries */
1616 ASSERT(XFS_IS_REALTIME_INODE(ip));
1617 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1618 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1619 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1623 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1624 for (idx = 0; idx < nextents; idx++)
1625 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1626 *O_rtblks = (xfs_qcnt_t)rtblks;
1631 * callback routine supplied to bulkstat(). Given an inumber, find its
1632 * dquots and update them to account for resources taken by that inode.
1636 xfs_qm_dqusage_adjust(
1637 xfs_mount_t *mp, /* mount point for filesystem */
1638 xfs_ino_t ino, /* inode number to get data for */
1639 void __user *buffer, /* not used */
1640 int ubsize, /* not used */
1641 void *private_data, /* not used */
1642 xfs_daddr_t bno, /* starting block of inode cluster */
1643 int *ubused, /* not used */
1644 void *dip, /* on-disk inode pointer (not used) */
1645 int *res) /* result code value */
1648 xfs_dquot_t *udqp, *gdqp;
1649 xfs_qcnt_t nblks, rtblks;
1652 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1655 * rootino must have its resources accounted for, not so with the quota
1658 if (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino) {
1659 *res = BULKSTAT_RV_NOTHING;
1660 return XFS_ERROR(EINVAL);
1664 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1665 * interface expects the inode to be exclusively locked because that's
1666 * the case in all other instances. It's OK that we do this because
1667 * quotacheck is done only at mount time.
1669 if ((error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip, bno))) {
1670 *res = BULKSTAT_RV_NOTHING;
1675 * Obtain the locked dquots. In case of an error (eg. allocation
1676 * fails for ENOSPC), we return the negative of the error number
1677 * to bulkstat, so that it can get propagated to quotacheck() and
1678 * making us disable quotas for the file system.
1680 if ((error = xfs_qm_dqget_noattach(ip, &udqp, &gdqp))) {
1681 xfs_iput(ip, XFS_ILOCK_EXCL);
1682 *res = BULKSTAT_RV_GIVEUP;
1687 if (! XFS_IS_REALTIME_INODE(ip)) {
1688 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks;
1691 * Walk thru the extent list and count the realtime blocks.
1693 if ((error = xfs_qm_get_rtblks(ip, &rtblks))) {
1694 xfs_iput(ip, XFS_ILOCK_EXCL);
1699 *res = BULKSTAT_RV_GIVEUP;
1702 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1704 ASSERT(ip->i_delayed_blks == 0);
1707 * We can't release the inode while holding its dquot locks.
1708 * The inode can go into inactive and might try to acquire the dquotlocks.
1709 * So, just unlock here and do a vn_rele at the end.
1711 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1714 * Add the (disk blocks and inode) resources occupied by this
1715 * inode to its dquots. We do this adjustment in the incore dquot,
1716 * and also copy the changes to its buffer.
1717 * We don't care about putting these changes in a transaction
1718 * envelope because if we crash in the middle of a 'quotacheck'
1719 * we have to start from the beginning anyway.
1720 * Once we're done, we'll log all the dquot bufs.
1722 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1723 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1725 if (XFS_IS_UQUOTA_ON(mp)) {
1727 xfs_qm_quotacheck_dqadjust(udqp, nblks, rtblks);
1730 if (XFS_IS_OQUOTA_ON(mp)) {
1732 xfs_qm_quotacheck_dqadjust(gdqp, nblks, rtblks);
1736 * Now release the inode. This will send it to 'inactive', and
1737 * possibly even free blocks.
1744 *res = BULKSTAT_RV_DIDONE;
1749 * Walk thru all the filesystem inodes and construct a consistent view
1750 * of the disk quota world. If the quotacheck fails, disable quotas.
1756 int done, count, error;
1759 xfs_inode_t *uip, *gip;
1767 ASSERT(mp->m_quotainfo->qi_uquotaip || mp->m_quotainfo->qi_gquotaip);
1768 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1771 * There should be no cached dquots. The (simplistic) quotacheck
1772 * algorithm doesn't like that.
1774 ASSERT(list_empty(&mp->m_quotainfo->qi_dqlist));
1776 cmn_err(CE_NOTE, "XFS quotacheck %s: Please wait.", mp->m_fsname);
1779 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1780 * their counters to zero. We need a clean slate.
1781 * We don't log our changes till later.
1783 uip = mp->m_quotainfo->qi_uquotaip;
1785 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA);
1788 flags |= XFS_UQUOTA_CHKD;
1791 gip = mp->m_quotainfo->qi_gquotaip;
1793 error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
1794 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1797 flags |= XFS_OQUOTA_CHKD;
1802 * Iterate thru all the inodes in the file system,
1803 * adjusting the corresponding dquot counters in core.
1805 if ((error = xfs_bulkstat(mp, &lastino, &count,
1806 xfs_qm_dqusage_adjust, NULL,
1807 structsz, NULL, BULKSTAT_FG_IGET, &done)))
1813 * We've made all the changes that we need to make incore.
1814 * Flush them down to disk buffers if everything was updated
1818 error = xfs_qm_dqflush_all(mp, 0);
1821 * We can get this error if we couldn't do a dquot allocation inside
1822 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1823 * dirty dquots that might be cached, we just want to get rid of them
1824 * and turn quotaoff. The dquots won't be attached to any of the inodes
1825 * at this point (because we intentionally didn't in dqget_noattach).
1828 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1833 * We didn't log anything, because if we crashed, we'll have to
1834 * start the quotacheck from scratch anyway. However, we must make
1835 * sure that our dquot changes are secure before we put the
1836 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1839 XFS_bflush(mp->m_ddev_targp);
1842 * If one type of quotas is off, then it will lose its
1843 * quotachecked status, since we won't be doing accounting for
1844 * that type anymore.
1846 mp->m_qflags &= ~(XFS_OQUOTA_CHKD | XFS_UQUOTA_CHKD);
1847 mp->m_qflags |= flags;
1849 xfs_qm_dquot_list_print(mp);
1853 cmn_err(CE_WARN, "XFS quotacheck %s: Unsuccessful (Error %d): "
1854 "Disabling quotas.",
1855 mp->m_fsname, error);
1857 * We must turn off quotas.
1859 ASSERT(mp->m_quotainfo != NULL);
1860 ASSERT(xfs_Gqm != NULL);
1861 xfs_qm_destroy_quotainfo(mp);
1862 if (xfs_mount_reset_sbqflags(mp)) {
1863 cmn_err(CE_WARN, "XFS quotacheck %s: "
1864 "Failed to reset quota flags.", mp->m_fsname);
1867 cmn_err(CE_NOTE, "XFS quotacheck %s: Done.", mp->m_fsname);
1873 * This is called after the superblock has been read in and we're ready to
1874 * iget the quota inodes.
1877 xfs_qm_init_quotainos(
1880 xfs_inode_t *uip, *gip;
1885 ASSERT(mp->m_quotainfo);
1891 * Get the uquota and gquota inodes
1893 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1894 if (XFS_IS_UQUOTA_ON(mp) &&
1895 mp->m_sb.sb_uquotino != NULLFSINO) {
1896 ASSERT(mp->m_sb.sb_uquotino > 0);
1897 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1899 return XFS_ERROR(error);
1901 if (XFS_IS_OQUOTA_ON(mp) &&
1902 mp->m_sb.sb_gquotino != NULLFSINO) {
1903 ASSERT(mp->m_sb.sb_gquotino > 0);
1904 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1908 return XFS_ERROR(error);
1912 flags |= XFS_QMOPT_SBVERSION;
1913 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1914 XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
1918 * Create the two inodes, if they don't exist already. The changes
1919 * made above will get added to a transaction and logged in one of
1920 * the qino_alloc calls below. If the device is readonly,
1921 * temporarily switch to read-write to do this.
1923 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1924 if ((error = xfs_qm_qino_alloc(mp, &uip,
1925 sbflags | XFS_SB_UQUOTINO,
1926 flags | XFS_QMOPT_UQUOTA)))
1927 return XFS_ERROR(error);
1929 flags &= ~XFS_QMOPT_SBVERSION;
1931 if (XFS_IS_OQUOTA_ON(mp) && gip == NULL) {
1932 flags |= (XFS_IS_GQUOTA_ON(mp) ?
1933 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1934 error = xfs_qm_qino_alloc(mp, &gip,
1935 sbflags | XFS_SB_GQUOTINO, flags);
1940 return XFS_ERROR(error);
1944 mp->m_quotainfo->qi_uquotaip = uip;
1945 mp->m_quotainfo->qi_gquotaip = gip;
1953 * Just pop the least recently used dquot off the freelist and
1954 * recycle it. The returned dquot is locked.
1956 STATIC xfs_dquot_t *
1957 xfs_qm_dqreclaim_one(void)
1959 xfs_dquot_t *dqpout;
1966 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
1968 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1970 list_for_each_entry(dqp, &xfs_Gqm->qm_dqfrlist, q_freelist) {
1971 struct xfs_mount *mp = dqp->q_mount;
1975 * We are racing with dqlookup here. Naturally we don't
1976 * want to reclaim a dquot that lookup wants. We release the
1977 * freelist lock and start over, so that lookup will grab
1978 * both the dquot and the freelistlock.
1980 if (dqp->dq_flags & XFS_DQ_WANT) {
1981 ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE));
1983 trace_xfs_dqreclaim_want(dqp);
1986 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1987 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
1989 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
1994 * If the dquot is inactive, we are assured that it is
1995 * not on the mplist or the hashlist, and that makes our
1998 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
2000 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
2001 ASSERT(list_empty(&dqp->q_hashlist));
2002 ASSERT(list_empty(&dqp->q_mplist));
2003 list_del_init(&dqp->q_freelist);
2004 xfs_Gqm->qm_dqfrlist_cnt--;
2007 XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
2011 ASSERT(dqp->q_hash);
2012 ASSERT(!list_empty(&dqp->q_mplist));
2015 * Try to grab the flush lock. If this dquot is in the process of
2016 * getting flushed to disk, we don't want to reclaim it.
2018 if (!xfs_dqflock_nowait(dqp)) {
2024 * We have the flush lock so we know that this is not in the
2025 * process of being flushed. So, if this is dirty, flush it
2026 * DELWRI so that we don't get a freelist infested with
2029 if (XFS_DQ_IS_DIRTY(dqp)) {
2032 trace_xfs_dqreclaim_dirty(dqp);
2035 * We flush it delayed write, so don't bother
2036 * releasing the freelist lock.
2038 error = xfs_qm_dqflush(dqp, 0);
2040 xfs_fs_cmn_err(CE_WARN, mp,
2041 "xfs_qm_dqreclaim: dquot %p flush failed", dqp);
2043 xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
2048 * We're trying to get the hashlock out of order. This races
2049 * with dqlookup; so, we giveup and goto the next dquot if
2050 * we couldn't get the hashlock. This way, we won't starve
2051 * a dqlookup process that holds the hashlock that is
2052 * waiting for the freelist lock.
2054 if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
2060 * This races with dquot allocation code as well as dqflush_all
2061 * and reclaim code. So, if we failed to grab the mplist lock,
2062 * giveup everything and start over.
2064 if (!mutex_trylock(&mp->m_quotainfo->qi_dqlist_lock)) {
2066 mutex_unlock(&dqp->q_hash->qh_lock);
2069 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
2070 if (restarts++ >= XFS_QM_RECLAIM_MAX_RESTARTS)
2075 ASSERT(dqp->q_nrefs == 0);
2076 list_del_init(&dqp->q_mplist);
2077 mp->m_quotainfo->qi_dquots--;
2078 mp->m_quotainfo->qi_dqreclaims++;
2079 list_del_init(&dqp->q_hashlist);
2080 dqp->q_hash->qh_version++;
2081 list_del_init(&dqp->q_freelist);
2082 xfs_Gqm->qm_dqfrlist_cnt--;
2084 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
2085 mutex_unlock(&dqp->q_hash->qh_lock);
2091 if (restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
2094 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
2099 * Traverse the freelist of dquots and attempt to reclaim a maximum of
2100 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
2101 * favor the lookup function ...
2104 xfs_qm_shake_freelist(
2113 while (nreclaimed < howmany) {
2114 dqp = xfs_qm_dqreclaim_one();
2117 xfs_qm_dqdestroy(dqp);
2124 * The kmem_shake interface is invoked when memory is running low.
2128 xfs_qm_shake(int nr_to_scan, gfp_t gfp_mask)
2130 int ndqused, nfree, n;
2132 if (!kmem_shake_allow(gfp_mask))
2137 nfree = xfs_Gqm->qm_dqfrlist_cnt; /* free dquots */
2138 /* incore dquots in all f/s's */
2139 ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;
2141 ASSERT(ndqused >= 0);
2143 if (nfree <= ndqused && nfree < ndquot)
2146 ndqused *= xfs_Gqm->qm_dqfree_ratio; /* target # of free dquots */
2147 n = nfree - ndqused - ndquot; /* # over target */
2149 return xfs_qm_shake_freelist(MAX(nfree, n));
2153 /*------------------------------------------------------------------*/
2156 * Return a new incore dquot. Depending on the number of
2157 * dquots in the system, we either allocate a new one on the kernel heap,
2158 * or reclaim a free one.
2159 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2160 * to reclaim an existing one from the freelist.
2163 xfs_qm_dqalloc_incore(
2164 xfs_dquot_t **O_dqpp)
2169 * Check against high water mark to see if we want to pop
2170 * a nincompoop dquot off the freelist.
2172 if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
2174 * Try to recycle a dquot from the freelist.
2176 if ((dqp = xfs_qm_dqreclaim_one())) {
2177 XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
2179 * Just zero the core here. The rest will get
2180 * reinitialized by caller. XXX we shouldn't even
2183 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
2187 XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
2191 * Allocate a brand new dquot on the kernel heap and return it
2192 * to the caller to initialize.
2194 ASSERT(xfs_Gqm->qm_dqzone != NULL);
2195 *O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
2196 atomic_inc(&xfs_Gqm->qm_totaldquots);
2203 * Start a transaction and write the incore superblock changes to
2204 * disk. flags parameter indicates which fields have changed.
2207 xfs_qm_write_sb_changes(
2215 cmn_err(CE_NOTE, "Writing superblock quota changes :%s", mp->m_fsname);
2217 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
2218 if ((error = xfs_trans_reserve(tp, 0,
2219 mp->m_sb.sb_sectsize + 128, 0,
2221 XFS_DEFAULT_LOG_COUNT))) {
2222 xfs_trans_cancel(tp, 0);
2226 xfs_mod_sb(tp, flags);
2227 error = xfs_trans_commit(tp, 0);
2233 /* --------------- utility functions for vnodeops ---------------- */
2237 * Given an inode, a uid and gid (from cred_t) make sure that we have
2238 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2239 * quotas by creating this file.
2240 * This also attaches dquot(s) to the given inode after locking it,
2241 * and returns the dquots corresponding to the uid and/or gid.
2243 * in : inode (unlocked)
2244 * out : udquot, gdquot with references taken and unlocked
2248 struct xfs_inode *ip,
2253 struct xfs_dquot **O_udqpp,
2254 struct xfs_dquot **O_gdqpp)
2256 struct xfs_mount *mp = ip->i_mount;
2257 struct xfs_dquot *uq, *gq;
2261 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2264 lockflags = XFS_ILOCK_EXCL;
2265 xfs_ilock(ip, lockflags);
2267 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
2268 gid = ip->i_d.di_gid;
2271 * Attach the dquot(s) to this inode, doing a dquot allocation
2272 * if necessary. The dquot(s) will not be locked.
2274 if (XFS_NOT_DQATTACHED(mp, ip)) {
2275 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
2277 xfs_iunlock(ip, lockflags);
2283 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
2284 if (ip->i_d.di_uid != uid) {
2286 * What we need is the dquot that has this uid, and
2287 * if we send the inode to dqget, the uid of the inode
2288 * takes priority over what's sent in the uid argument.
2289 * We must unlock inode here before calling dqget if
2290 * we're not sending the inode, because otherwise
2291 * we'll deadlock by doing trans_reserve while
2294 xfs_iunlock(ip, lockflags);
2295 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
2300 ASSERT(error != ENOENT);
2304 * Get the ilock in the right order.
2307 lockflags = XFS_ILOCK_SHARED;
2308 xfs_ilock(ip, lockflags);
2311 * Take an extra reference, because we'll return
2314 ASSERT(ip->i_udquot);
2321 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
2322 if (ip->i_d.di_gid != gid) {
2323 xfs_iunlock(ip, lockflags);
2324 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
2331 ASSERT(error != ENOENT);
2335 lockflags = XFS_ILOCK_SHARED;
2336 xfs_ilock(ip, lockflags);
2338 ASSERT(ip->i_gdquot);
2344 } else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
2345 if (ip->i_d.di_projid != prid) {
2346 xfs_iunlock(ip, lockflags);
2347 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
2354 ASSERT(error != ENOENT);
2358 lockflags = XFS_ILOCK_SHARED;
2359 xfs_ilock(ip, lockflags);
2361 ASSERT(ip->i_gdquot);
2369 trace_xfs_dquot_dqalloc(ip);
2371 xfs_iunlock(ip, lockflags);
2384 * Actually transfer ownership, and do dquot modifications.
2385 * These were already reserved.
2391 xfs_dquot_t **IO_olddq,
2394 xfs_dquot_t *prevdq;
2395 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
2396 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
2399 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2400 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
2405 ASSERT(prevdq != newdq);
2407 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
2408 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
2410 /* the sparkling new dquot */
2411 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
2412 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
2415 * Take an extra reference, because the inode
2416 * is going to keep this dquot pointer even
2417 * after the trans_commit.
2421 xfs_dqunlock(newdq);
2428 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
2431 xfs_qm_vop_chown_reserve(
2438 xfs_mount_t *mp = ip->i_mount;
2439 uint delblks, blkflags, prjflags = 0;
2440 xfs_dquot_t *unresudq, *unresgdq, *delblksudq, *delblksgdq;
2444 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2445 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2447 delblks = ip->i_delayed_blks;
2448 delblksudq = delblksgdq = unresudq = unresgdq = NULL;
2449 blkflags = XFS_IS_REALTIME_INODE(ip) ?
2450 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
2452 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2453 ip->i_d.di_uid != (uid_t)be32_to_cpu(udqp->q_core.d_id)) {
2456 * If there are delayed allocation blocks, then we have to
2457 * unreserve those from the old dquot, and add them to the
2461 ASSERT(ip->i_udquot);
2462 unresudq = ip->i_udquot;
2465 if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
2466 if (XFS_IS_PQUOTA_ON(ip->i_mount) &&
2467 ip->i_d.di_projid != be32_to_cpu(gdqp->q_core.d_id))
2468 prjflags = XFS_QMOPT_ENOSPC;
2471 (XFS_IS_GQUOTA_ON(ip->i_mount) &&
2472 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id))) {
2475 ASSERT(ip->i_gdquot);
2476 unresgdq = ip->i_gdquot;
2481 if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2482 delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
2483 flags | blkflags | prjflags)))
2487 * Do the delayed blks reservations/unreservations now. Since, these
2488 * are done without the help of a transaction, if a reservation fails
2489 * its previous reservations won't be automatically undone by trans
2490 * code. So, we have to do it manually here.
2494 * Do the reservations first. Unreservation can't fail.
2496 ASSERT(delblksudq || delblksgdq);
2497 ASSERT(unresudq || unresgdq);
2498 if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2499 delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
2500 flags | blkflags | prjflags)))
2502 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2503 unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
2511 xfs_qm_vop_rename_dqattach(
2512 struct xfs_inode **i_tab)
2514 struct xfs_mount *mp = i_tab[0]->i_mount;
2517 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2520 for (i = 0; (i < 4 && i_tab[i]); i++) {
2521 struct xfs_inode *ip = i_tab[i];
2525 * Watch out for duplicate entries in the table.
2527 if (i == 0 || ip != i_tab[i-1]) {
2528 if (XFS_NOT_DQATTACHED(mp, ip)) {
2529 error = xfs_qm_dqattach(ip, 0);
2539 xfs_qm_vop_create_dqattach(
2540 struct xfs_trans *tp,
2541 struct xfs_inode *ip,
2542 struct xfs_dquot *udqp,
2543 struct xfs_dquot *gdqp)
2545 struct xfs_mount *mp = tp->t_mountp;
2547 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2550 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2551 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2557 ASSERT(ip->i_udquot == NULL);
2558 ip->i_udquot = udqp;
2559 ASSERT(XFS_IS_UQUOTA_ON(mp));
2560 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
2561 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2567 ASSERT(ip->i_gdquot == NULL);
2568 ip->i_gdquot = gdqp;
2569 ASSERT(XFS_IS_OQUOTA_ON(mp));
2570 ASSERT((XFS_IS_GQUOTA_ON(mp) ?
2571 ip->i_d.di_gid : ip->i_d.di_projid) ==
2572 be32_to_cpu(gdqp->q_core.d_id));
2573 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);