2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "xfs_trans.h"
26 #include "xfs_alloc.h"
27 #include "xfs_quota.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_ialloc.h"
34 #include "xfs_itable.h"
35 #include "xfs_rtalloc.h"
36 #include "xfs_error.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_trans_space.h"
41 #include "xfs_utils.h"
43 #include "xfs_trace.h"
46 * The global quota manager. There is only one of these for the entire
47 * system, _not_ one per file system. XQM keeps track of the overall
48 * quota functionality, including maintaining the freelist and hash
51 struct mutex xfs_Gqm_lock;
52 struct xfs_qm *xfs_Gqm;
55 kmem_zone_t *qm_dqzone;
56 kmem_zone_t *qm_dqtrxzone;
58 STATIC void xfs_qm_list_init(xfs_dqlist_t *, char *, int);
59 STATIC void xfs_qm_list_destroy(xfs_dqlist_t *);
61 STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
62 STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
63 STATIC int xfs_qm_shake(struct shrinker *, int, gfp_t);
65 static struct shrinker xfs_qm_shaker = {
66 .shrink = xfs_qm_shake,
67 .seeks = DEFAULT_SEEKS,
71 extern struct mutex qcheck_lock;
76 xfs_qm_dquot_list_print(
82 list_for_each_entry(dqp, &mp->m_quotainfo->qi_dqlist_lock, qi_mplist) {
83 cmn_err(CE_DEBUG, " %d. \"%d (%s)\" "
84 "bcnt = %lld, icnt = %lld, refs = %d",
85 i++, be32_to_cpu(dqp->q_core.d_id),
86 DQFLAGTO_TYPESTR(dqp),
87 (long long)be64_to_cpu(dqp->q_core.d_bcount),
88 (long long)be64_to_cpu(dqp->q_core.d_icount),
93 static void xfs_qm_dquot_list_print(struct xfs_mount *mp) { }
97 * Initialize the XQM structure.
98 * Note that there is not one quota manager per file system.
100 STATIC struct xfs_qm *
103 xfs_dqhash_t *udqhash, *gdqhash;
109 * Initialize the dquot hash tables.
111 udqhash = kmem_zalloc_greedy(&hsize,
112 XFS_QM_HASHSIZE_LOW * sizeof(xfs_dqhash_t),
113 XFS_QM_HASHSIZE_HIGH * sizeof(xfs_dqhash_t));
117 gdqhash = kmem_zalloc_large(hsize);
119 goto out_free_udqhash;
121 hsize /= sizeof(xfs_dqhash_t);
124 xqm = kmem_zalloc(sizeof(xfs_qm_t), KM_SLEEP);
125 xqm->qm_dqhashmask = hsize - 1;
126 xqm->qm_usr_dqhtable = udqhash;
127 xqm->qm_grp_dqhtable = gdqhash;
128 ASSERT(xqm->qm_usr_dqhtable != NULL);
129 ASSERT(xqm->qm_grp_dqhtable != NULL);
131 for (i = 0; i < hsize; i++) {
132 xfs_qm_list_init(&(xqm->qm_usr_dqhtable[i]), "uxdqh", i);
133 xfs_qm_list_init(&(xqm->qm_grp_dqhtable[i]), "gxdqh", i);
137 * Freelist of all dquots of all file systems
139 INIT_LIST_HEAD(&xqm->qm_dqfrlist);
140 xqm->qm_dqfrlist_cnt = 0;
141 mutex_init(&xqm->qm_dqfrlist_lock);
144 * dquot zone. we register our own low-memory callback.
147 xqm->qm_dqzone = kmem_zone_init(sizeof(xfs_dquot_t),
149 qm_dqzone = xqm->qm_dqzone;
151 xqm->qm_dqzone = qm_dqzone;
153 register_shrinker(&xfs_qm_shaker);
156 * The t_dqinfo portion of transactions.
159 xqm->qm_dqtrxzone = kmem_zone_init(sizeof(xfs_dquot_acct_t),
161 qm_dqtrxzone = xqm->qm_dqtrxzone;
163 xqm->qm_dqtrxzone = qm_dqtrxzone;
165 atomic_set(&xqm->qm_totaldquots, 0);
166 xqm->qm_dqfree_ratio = XFS_QM_DQFREE_RATIO;
169 mutex_init(&qcheck_lock);
174 kmem_free_large(udqhash);
180 * Destroy the global quota manager when its reference count goes to zero.
186 struct xfs_dquot *dqp, *n;
190 ASSERT(xqm->qm_nrefs == 0);
191 unregister_shrinker(&xfs_qm_shaker);
192 hsize = xqm->qm_dqhashmask + 1;
193 for (i = 0; i < hsize; i++) {
194 xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
195 xfs_qm_list_destroy(&(xqm->qm_grp_dqhtable[i]));
197 kmem_free_large(xqm->qm_usr_dqhtable);
198 kmem_free_large(xqm->qm_grp_dqhtable);
199 xqm->qm_usr_dqhtable = NULL;
200 xqm->qm_grp_dqhtable = NULL;
201 xqm->qm_dqhashmask = 0;
204 mutex_lock(&xqm->qm_dqfrlist_lock);
205 list_for_each_entry_safe(dqp, n, &xqm->qm_dqfrlist, q_freelist) {
208 cmn_err(CE_DEBUG, "FREELIST destroy 0x%p", dqp);
210 list_del_init(&dqp->q_freelist);
211 xfs_Gqm->qm_dqfrlist_cnt--;
213 xfs_qm_dqdestroy(dqp);
215 mutex_unlock(&xqm->qm_dqfrlist_lock);
216 mutex_destroy(&xqm->qm_dqfrlist_lock);
218 mutex_destroy(&qcheck_lock);
224 * Called at mount time to let XQM know that another file system is
225 * starting quotas. This isn't crucial information as the individual mount
226 * structures are pretty independent, but it helps the XQM keep a
227 * global view of what's going on.
231 xfs_qm_hold_quotafs_ref(
232 struct xfs_mount *mp)
235 * Need to lock the xfs_Gqm structure for things like this. For example,
236 * the structure could disappear between the entry to this routine and
237 * a HOLD operation if not locked.
239 mutex_lock(&xfs_Gqm_lock);
242 xfs_Gqm = xfs_Gqm_init();
244 mutex_unlock(&xfs_Gqm_lock);
250 * We can keep a list of all filesystems with quotas mounted for
251 * debugging and statistical purposes, but ...
252 * Just take a reference and get out.
255 mutex_unlock(&xfs_Gqm_lock);
262 * Release the reference that a filesystem took at mount time,
263 * so that we know when we need to destroy the entire quota manager.
267 xfs_qm_rele_quotafs_ref(
268 struct xfs_mount *mp)
270 xfs_dquot_t *dqp, *n;
273 ASSERT(xfs_Gqm->qm_nrefs > 0);
276 * Go thru the freelist and destroy all inactive dquots.
278 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
280 list_for_each_entry_safe(dqp, n, &xfs_Gqm->qm_dqfrlist, q_freelist) {
282 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
283 ASSERT(dqp->q_mount == NULL);
284 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
285 ASSERT(list_empty(&dqp->q_hashlist));
286 ASSERT(list_empty(&dqp->q_mplist));
287 list_del_init(&dqp->q_freelist);
288 xfs_Gqm->qm_dqfrlist_cnt--;
290 xfs_qm_dqdestroy(dqp);
295 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
298 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
301 mutex_lock(&xfs_Gqm_lock);
302 if (--xfs_Gqm->qm_nrefs == 0) {
303 xfs_qm_destroy(xfs_Gqm);
306 mutex_unlock(&xfs_Gqm_lock);
310 * Just destroy the quotainfo structure.
314 struct xfs_mount *mp)
316 if (mp->m_quotainfo) {
317 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
318 xfs_qm_destroy_quotainfo(mp);
324 * This is called from xfs_mountfs to start quotas and initialize all
325 * necessary data structures like quotainfo. This is also responsible for
326 * running a quotacheck as necessary. We are guaranteed that the superblock
327 * is consistently read in at this point.
329 * If we fail here, the mount will continue with quota turned off. We don't
330 * need to inidicate success or failure at all.
340 * If quotas on realtime volumes is not supported, we disable
341 * quotas immediately.
343 if (mp->m_sb.sb_rextents) {
345 "Cannot turn on quotas for realtime filesystem %s",
351 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
354 * Allocate the quotainfo structure inside the mount struct, and
355 * create quotainode(s), and change/rev superblock if necessary.
357 error = xfs_qm_init_quotainfo(mp);
360 * We must turn off quotas.
362 ASSERT(mp->m_quotainfo == NULL);
367 * If any of the quotas are not consistent, do a quotacheck.
369 if (XFS_QM_NEED_QUOTACHECK(mp)) {
370 error = xfs_qm_quotacheck(mp);
372 /* Quotacheck failed and disabled quotas. */
377 * If one type of quotas is off, then it will lose its
378 * quotachecked status, since we won't be doing accounting for
381 if (!XFS_IS_UQUOTA_ON(mp))
382 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
383 if (!(XFS_IS_GQUOTA_ON(mp) || XFS_IS_PQUOTA_ON(mp)))
384 mp->m_qflags &= ~XFS_OQUOTA_CHKD;
388 * We actually don't have to acquire the m_sb_lock at all.
389 * This can only be called from mount, and that's single threaded. XXX
391 spin_lock(&mp->m_sb_lock);
392 sbf = mp->m_sb.sb_qflags;
393 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
394 spin_unlock(&mp->m_sb_lock);
396 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
397 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
399 * We could only have been turning quotas off.
400 * We aren't in very good shape actually because
401 * the incore structures are convinced that quotas are
402 * off, but the on disk superblock doesn't know that !
404 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
405 xfs_fs_cmn_err(CE_ALERT, mp,
406 "XFS mount_quotas: Superblock update failed!");
411 xfs_fs_cmn_err(CE_WARN, mp,
412 "Failed to initialize disk quotas.");
417 if (XFS_IS_QUOTA_ON(mp))
418 xfs_qm_internalqcheck(mp);
423 * Called from the vfsops layer.
426 xfs_qm_unmount_quotas(
430 * Release the dquots that root inode, et al might be holding,
431 * before we flush quotas and blow away the quotainfo structure.
433 ASSERT(mp->m_rootip);
434 xfs_qm_dqdetach(mp->m_rootip);
436 xfs_qm_dqdetach(mp->m_rbmip);
438 xfs_qm_dqdetach(mp->m_rsumip);
441 * Release the quota inodes.
443 if (mp->m_quotainfo) {
444 if (mp->m_quotainfo->qi_uquotaip) {
445 IRELE(mp->m_quotainfo->qi_uquotaip);
446 mp->m_quotainfo->qi_uquotaip = NULL;
448 if (mp->m_quotainfo->qi_gquotaip) {
449 IRELE(mp->m_quotainfo->qi_gquotaip);
450 mp->m_quotainfo->qi_gquotaip = NULL;
456 * Flush all dquots of the given file system to disk. The dquots are
457 * _not_ purged from memory here, just their data written to disk.
461 struct xfs_mount *mp,
464 struct xfs_quotainfo *q = mp->m_quotainfo;
466 struct xfs_dquot *dqp;
474 mutex_lock(&q->qi_dqlist_lock);
475 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
477 if (! XFS_DQ_IS_DIRTY(dqp)) {
482 /* XXX a sentinel would be better */
483 recl = q->qi_dqreclaims;
484 if (!xfs_dqflock_nowait(dqp)) {
486 * If we can't grab the flush lock then check
487 * to see if the dquot has been flushed delayed
488 * write. If so, grab its buffer and send it
489 * out immediately. We'll be able to acquire
490 * the flush lock when the I/O completes.
492 xfs_qm_dqflock_pushbuf_wait(dqp);
495 * Let go of the mplist lock. We don't want to hold it
496 * across a disk write.
498 mutex_unlock(&q->qi_dqlist_lock);
499 error = xfs_qm_dqflush(dqp, sync_mode);
504 mutex_lock(&q->qi_dqlist_lock);
505 if (recl != q->qi_dqreclaims) {
506 mutex_unlock(&q->qi_dqlist_lock);
507 /* XXX restart limit */
512 mutex_unlock(&q->qi_dqlist_lock);
517 * Release the group dquot pointers the user dquots may be
518 * carrying around as a hint. mplist is locked on entry and exit.
521 xfs_qm_detach_gdquots(
522 struct xfs_mount *mp)
524 struct xfs_quotainfo *q = mp->m_quotainfo;
525 struct xfs_dquot *dqp, *gdqp;
529 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
530 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
532 if ((gdqp = dqp->q_gdquot)) {
534 dqp->q_gdquot = NULL;
540 * Can't hold the mplist lock across a dqput.
541 * XXXmust convert to marker based iterations here.
543 nrecl = q->qi_dqreclaims;
544 mutex_unlock(&q->qi_dqlist_lock);
547 mutex_lock(&q->qi_dqlist_lock);
548 if (nrecl != q->qi_dqreclaims)
555 * Go through all the incore dquots of this file system and take them
556 * off the mplist and hashlist, if the dquot type matches the dqtype
557 * parameter. This is used when turning off quota accounting for
558 * users and/or groups, as well as when the filesystem is unmounting.
562 struct xfs_mount *mp,
565 struct xfs_quotainfo *q = mp->m_quotainfo;
566 struct xfs_dquot *dqp, *n;
574 dqtype = (flags & XFS_QMOPT_UQUOTA) ? XFS_DQ_USER : 0;
575 dqtype |= (flags & XFS_QMOPT_PQUOTA) ? XFS_DQ_PROJ : 0;
576 dqtype |= (flags & XFS_QMOPT_GQUOTA) ? XFS_DQ_GROUP : 0;
578 mutex_lock(&q->qi_dqlist_lock);
581 * In the first pass through all incore dquots of this filesystem,
582 * we release the group dquot pointers the user dquots may be
583 * carrying around as a hint. We need to do this irrespective of
584 * what's being turned off.
586 xfs_qm_detach_gdquots(mp);
590 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
592 * Try to get rid of all of the unwanted dquots. The idea is to
593 * get them off mplist and hashlist, but leave them on freelist.
595 list_for_each_entry_safe(dqp, n, &q->qi_dqlist, q_mplist) {
597 * It's OK to look at the type without taking dqlock here.
598 * We're holding the mplist lock here, and that's needed for
601 if ((dqp->dq_flags & dqtype) == 0)
604 if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
605 nrecl = q->qi_dqreclaims;
606 mutex_unlock(&q->qi_dqlist_lock);
607 mutex_lock(&dqp->q_hash->qh_lock);
608 mutex_lock(&q->qi_dqlist_lock);
611 * XXXTheoretically, we can get into a very long
612 * ping pong game here.
613 * No one can be adding dquots to the mplist at
614 * this point, but somebody might be taking things off.
616 if (nrecl != q->qi_dqreclaims) {
617 mutex_unlock(&dqp->q_hash->qh_lock);
623 * Take the dquot off the mplist and hashlist. It may remain on
624 * freelist in INACTIVE state.
626 nmisses += xfs_qm_dqpurge(dqp);
628 mutex_unlock(&q->qi_dqlist_lock);
640 * Purge the dquot cache.
641 * None of the dquots should really be busy at this point.
643 if (mp->m_quotainfo) {
644 while ((ndquots = xfs_qm_dqpurge_int(mp, flags))) {
657 xfs_dquot_t *udqhint, /* hint */
658 xfs_dquot_t **IO_idqpp)
663 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
667 * See if we already have it in the inode itself. IO_idqpp is
668 * &i_udquot or &i_gdquot. This made the code look weird, but
669 * made the logic a lot simpler.
673 trace_xfs_dqattach_found(dqp);
678 * udqhint is the i_udquot field in inode, and is non-NULL only
679 * when the type arg is group/project. Its purpose is to save a
680 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
684 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
688 * No need to take dqlock to look at the id.
690 * The ID can't change until it gets reclaimed, and it won't
691 * be reclaimed as long as we have a ref from inode and we
694 dqp = udqhint->q_gdquot;
695 if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
698 ASSERT(*IO_idqpp == NULL);
702 xfs_dqunlock(udqhint);
707 * We can't hold a dquot lock when we call the dqget code.
708 * We'll deadlock in no time, because of (not conforming to)
709 * lock ordering - the inodelock comes before any dquot lock,
710 * and we may drop and reacquire the ilock in xfs_qm_dqget().
712 xfs_dqunlock(udqhint);
716 * Find the dquot from somewhere. This bumps the
717 * reference count of dquot and returns it locked.
718 * This can return ENOENT if dquot didn't exist on
719 * disk and we didn't ask it to allocate;
720 * ESRCH if quotas got turned off suddenly.
722 error = xfs_qm_dqget(ip->i_mount, ip, id, type, XFS_QMOPT_DOWARN, &dqp);
726 trace_xfs_dqattach_get(dqp);
729 * dqget may have dropped and re-acquired the ilock, but it guarantees
730 * that the dquot returned is the one that should go in the inode.
739 * Given a udquot and gdquot, attach a ptr to the group dquot in the
740 * udquot as a hint for future lookups. The idea sounds simple, but the
741 * execution isn't, because the udquot might have a group dquot attached
742 * already and getting rid of that gets us into lock ordering constraints.
743 * The process is complicated more by the fact that the dquots may or may not
744 * be locked on entry.
747 xfs_qm_dqattach_grouphint(
755 if ((tmp = udq->q_gdquot)) {
761 udq->q_gdquot = NULL;
763 * We can't keep any dqlocks when calling dqrele,
764 * because the freelist lock comes before dqlocks.
768 * we took a hard reference once upon a time in dqget,
769 * so give it back when the udquot no longer points at it
770 * dqput() does the unlocking of the dquot.
778 ASSERT(XFS_DQ_IS_LOCKED(udq));
782 ASSERT(XFS_DQ_IS_LOCKED(udq));
783 ASSERT(XFS_DQ_IS_LOCKED(gdq));
785 * Somebody could have attached a gdquot here,
786 * when we dropped the uqlock. If so, just do nothing.
788 if (udq->q_gdquot == NULL) {
799 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
801 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
802 * Inode may get unlocked and relocked in here, and the caller must deal with
806 xfs_qm_dqattach_locked(
810 xfs_mount_t *mp = ip->i_mount;
814 if (!XFS_IS_QUOTA_RUNNING(mp) ||
815 !XFS_IS_QUOTA_ON(mp) ||
816 !XFS_NOT_DQATTACHED(mp, ip) ||
817 ip->i_ino == mp->m_sb.sb_uquotino ||
818 ip->i_ino == mp->m_sb.sb_gquotino)
821 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
823 if (XFS_IS_UQUOTA_ON(mp)) {
824 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
825 flags & XFS_QMOPT_DQALLOC,
826 NULL, &ip->i_udquot);
832 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
833 if (XFS_IS_OQUOTA_ON(mp)) {
834 error = XFS_IS_GQUOTA_ON(mp) ?
835 xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
836 flags & XFS_QMOPT_DQALLOC,
837 ip->i_udquot, &ip->i_gdquot) :
838 xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
839 flags & XFS_QMOPT_DQALLOC,
840 ip->i_udquot, &ip->i_gdquot);
842 * Don't worry about the udquot that we may have
843 * attached above. It'll get detached, if not already.
851 * Attach this group quota to the user quota as a hint.
852 * This WON'T, in general, result in a thrash.
855 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
856 ASSERT(ip->i_udquot);
857 ASSERT(ip->i_gdquot);
860 * We may or may not have the i_udquot locked at this point,
861 * but this check is OK since we don't depend on the i_gdquot to
862 * be accurate 100% all the time. It is just a hint, and this
863 * will succeed in general.
865 if (ip->i_udquot->q_gdquot == ip->i_gdquot)
868 * Attach i_gdquot to the gdquot hint inside the i_udquot.
870 xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot);
876 if (XFS_IS_UQUOTA_ON(mp))
877 ASSERT(ip->i_udquot);
878 if (XFS_IS_OQUOTA_ON(mp))
879 ASSERT(ip->i_gdquot);
881 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
888 struct xfs_inode *ip,
893 xfs_ilock(ip, XFS_ILOCK_EXCL);
894 error = xfs_qm_dqattach_locked(ip, flags);
895 xfs_iunlock(ip, XFS_ILOCK_EXCL);
901 * Release dquots (and their references) if any.
902 * The inode should be locked EXCL except when this's called by
909 if (!(ip->i_udquot || ip->i_gdquot))
912 trace_xfs_dquot_dqdetach(ip);
914 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
915 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
917 xfs_qm_dqrele(ip->i_udquot);
921 xfs_qm_dqrele(ip->i_gdquot);
928 struct xfs_mount *mp,
931 struct xfs_quotainfo *q = mp->m_quotainfo;
933 struct xfs_dquot *dqp;
936 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
942 mutex_lock(&q->qi_dqlist_lock);
944 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
945 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
946 * when we have the mplist lock, we know that dquots will be consistent
947 * as long as we have it locked.
949 if (!XFS_IS_QUOTA_ON(mp)) {
950 mutex_unlock(&q->qi_dqlist_lock);
953 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
954 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
956 * If this is vfs_sync calling, then skip the dquots that
957 * don't 'seem' to be dirty. ie. don't acquire dqlock.
958 * This is very similar to what xfs_sync does with inodes.
960 if (flags & SYNC_TRYLOCK) {
961 if (!XFS_DQ_IS_DIRTY(dqp))
963 if (!xfs_qm_dqlock_nowait(dqp))
970 * Now, find out for sure if this dquot is dirty or not.
972 if (! XFS_DQ_IS_DIRTY(dqp)) {
977 /* XXX a sentinel would be better */
978 recl = q->qi_dqreclaims;
979 if (!xfs_dqflock_nowait(dqp)) {
980 if (flags & SYNC_TRYLOCK) {
985 * If we can't grab the flush lock then if the caller
986 * really wanted us to give this our best shot, so
987 * see if we can give a push to the buffer before we wait
988 * on the flush lock. At this point, we know that
989 * even though the dquot is being flushed,
990 * it has (new) dirty data.
992 xfs_qm_dqflock_pushbuf_wait(dqp);
995 * Let go of the mplist lock. We don't want to hold it
996 * across a disk write
998 mutex_unlock(&q->qi_dqlist_lock);
999 error = xfs_qm_dqflush(dqp, flags);
1001 if (error && XFS_FORCED_SHUTDOWN(mp))
1002 return 0; /* Need to prevent umount failure */
1006 mutex_lock(&q->qi_dqlist_lock);
1007 if (recl != q->qi_dqreclaims) {
1008 if (++restarts >= XFS_QM_SYNC_MAX_RESTARTS)
1011 mutex_unlock(&q->qi_dqlist_lock);
1016 mutex_unlock(&q->qi_dqlist_lock);
1021 * The hash chains and the mplist use the same xfs_dqhash structure as
1022 * their list head, but we can take the mplist qh_lock and one of the
1023 * hash qh_locks at the same time without any problem as they aren't
1026 static struct lock_class_key xfs_quota_mplist_class;
1029 * This initializes all the quota information that's kept in the
1033 xfs_qm_init_quotainfo(
1036 xfs_quotainfo_t *qinf;
1040 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1043 * Tell XQM that we exist as soon as possible.
1045 if ((error = xfs_qm_hold_quotafs_ref(mp))) {
1049 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
1052 * See if quotainodes are setup, and if not, allocate them,
1053 * and change the superblock accordingly.
1055 if ((error = xfs_qm_init_quotainos(mp))) {
1057 mp->m_quotainfo = NULL;
1061 INIT_LIST_HEAD(&qinf->qi_dqlist);
1062 mutex_init(&qinf->qi_dqlist_lock);
1063 lockdep_set_class(&qinf->qi_dqlist_lock, &xfs_quota_mplist_class);
1065 qinf->qi_dqreclaims = 0;
1067 /* mutex used to serialize quotaoffs */
1068 mutex_init(&qinf->qi_quotaofflock);
1070 /* Precalc some constants */
1071 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1072 ASSERT(qinf->qi_dqchunklen);
1073 qinf->qi_dqperchunk = BBTOB(qinf->qi_dqchunklen);
1074 do_div(qinf->qi_dqperchunk, sizeof(xfs_dqblk_t));
1076 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
1079 * We try to get the limits from the superuser's limits fields.
1080 * This is quite hacky, but it is standard quota practice.
1081 * We look at the USR dquot with id == 0 first, but if user quotas
1082 * are not enabled we goto the GRP dquot with id == 0.
1083 * We don't really care to keep separate default limits for user
1084 * and group quotas, at least not at this point.
1086 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)0,
1087 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
1088 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
1090 XFS_QMOPT_DQSUSER|XFS_QMOPT_DOWARN,
1093 xfs_disk_dquot_t *ddqp = &dqp->q_core;
1096 * The warnings and timers set the grace period given to
1097 * a user or group before he or she can not perform any
1098 * more writing. If it is zero, a default is used.
1100 qinf->qi_btimelimit = ddqp->d_btimer ?
1101 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
1102 qinf->qi_itimelimit = ddqp->d_itimer ?
1103 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
1104 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
1105 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
1106 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
1107 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
1108 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
1109 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
1110 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
1111 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
1112 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
1113 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
1114 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
1115 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
1116 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
1117 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
1120 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1121 * we don't want this dquot cached. We haven't done a
1122 * quotacheck yet, and quotacheck doesn't like incore dquots.
1124 xfs_qm_dqdestroy(dqp);
1126 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
1127 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
1128 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
1129 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
1130 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
1131 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
1139 * Gets called when unmounting a filesystem or when all quotas get
1141 * This purges the quota inodes, destroys locks and frees itself.
1144 xfs_qm_destroy_quotainfo(
1147 xfs_quotainfo_t *qi;
1149 qi = mp->m_quotainfo;
1151 ASSERT(xfs_Gqm != NULL);
1154 * Release the reference that XQM kept, so that we know
1155 * when the XQM structure should be freed. We cannot assume
1156 * that xfs_Gqm is non-null after this point.
1158 xfs_qm_rele_quotafs_ref(mp);
1160 ASSERT(list_empty(&qi->qi_dqlist));
1161 mutex_destroy(&qi->qi_dqlist_lock);
1163 if (qi->qi_uquotaip) {
1164 IRELE(qi->qi_uquotaip);
1165 qi->qi_uquotaip = NULL; /* paranoia */
1167 if (qi->qi_gquotaip) {
1168 IRELE(qi->qi_gquotaip);
1169 qi->qi_gquotaip = NULL;
1171 mutex_destroy(&qi->qi_quotaofflock);
1173 mp->m_quotainfo = NULL;
1178 /* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1187 mutex_init(&list->qh_lock);
1188 INIT_LIST_HEAD(&list->qh_list);
1189 list->qh_version = 0;
1190 list->qh_nelems = 0;
1194 xfs_qm_list_destroy(
1197 mutex_destroy(&(list->qh_lock));
1201 * Create an inode and return with a reference already taken, but unlocked
1202 * This is how we create quota inodes
1215 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
1216 if ((error = xfs_trans_reserve(tp,
1217 XFS_QM_QINOCREATE_SPACE_RES(mp),
1218 XFS_CREATE_LOG_RES(mp), 0,
1219 XFS_TRANS_PERM_LOG_RES,
1220 XFS_CREATE_LOG_COUNT))) {
1221 xfs_trans_cancel(tp, 0);
1225 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip, &committed);
1227 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1233 * Keep an extra reference to this quota inode. This inode is
1234 * locked exclusively and joined to the transaction already.
1236 ASSERT(xfs_isilocked(*ip, XFS_ILOCK_EXCL));
1240 * Make the changes in the superblock, and log those too.
1241 * sbfields arg may contain fields other than *QUOTINO;
1242 * VERSIONNUM for example.
1244 spin_lock(&mp->m_sb_lock);
1245 if (flags & XFS_QMOPT_SBVERSION) {
1246 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1247 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1248 XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
1249 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1250 XFS_SB_GQUOTINO | XFS_SB_QFLAGS));
1252 xfs_sb_version_addquota(&mp->m_sb);
1253 mp->m_sb.sb_uquotino = NULLFSINO;
1254 mp->m_sb.sb_gquotino = NULLFSINO;
1256 /* qflags will get updated _after_ quotacheck */
1257 mp->m_sb.sb_qflags = 0;
1259 if (flags & XFS_QMOPT_UQUOTA)
1260 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1262 mp->m_sb.sb_gquotino = (*ip)->i_ino;
1263 spin_unlock(&mp->m_sb_lock);
1264 xfs_mod_sb(tp, sbfields);
1266 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
1267 xfs_fs_cmn_err(CE_ALERT, mp, "XFS qino_alloc failed!");
1275 xfs_qm_reset_dqcounts(
1281 xfs_disk_dquot_t *ddq;
1284 trace_xfs_reset_dqcounts(bp, _RET_IP_);
1287 * Reset all counters and timers. They'll be
1288 * started afresh by xfs_qm_quotacheck.
1291 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1292 do_div(j, sizeof(xfs_dqblk_t));
1293 ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
1295 ddq = (xfs_disk_dquot_t *)XFS_BUF_PTR(bp);
1296 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
1298 * Do a sanity check, and if needed, repair the dqblk. Don't
1299 * output any warnings because it's perfectly possible to
1300 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1302 (void) xfs_qm_dqcheck(ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1306 ddq->d_rtbcount = 0;
1309 ddq->d_rtbtimer = 0;
1312 ddq->d_rtbwarns = 0;
1313 ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
1322 xfs_filblks_t blkcnt,
1333 incr = (blkcnt > XFS_QM_MAX_DQCLUSTER_LOGSZ) ?
1334 XFS_QM_MAX_DQCLUSTER_LOGSZ : blkcnt;
1335 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1336 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1340 * Blkcnt arg can be a very big number, and might even be
1341 * larger than the log itself. So, we have to break it up into
1342 * manageable-sized transactions.
1343 * Note that we don't start a permanent transaction here; we might
1344 * not be able to get a log reservation for the whole thing up front,
1345 * and we don't really care to either, because we just discard
1346 * everything if we were to crash in the middle of this loop.
1349 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1350 XFS_FSB_TO_DADDR(mp, bno),
1351 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
1355 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1356 xfs_bdwrite(mp, bp);
1358 * goto the next block.
1361 firstid += mp->m_quotainfo->qi_dqperchunk;
1367 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1368 * caller supplied function for every chunk of dquots that we find.
1376 xfs_bmbt_irec_t *map;
1377 int i, nmaps; /* number of map entries */
1378 int error; /* return value */
1379 xfs_fileoff_t lblkno;
1380 xfs_filblks_t maxlblkcnt;
1382 xfs_fsblock_t rablkno;
1383 xfs_filblks_t rablkcnt;
1387 * This looks racy, but we can't keep an inode lock across a
1388 * trans_reserve. But, this gets called during quotacheck, and that
1389 * happens only at mount time which is single threaded.
1391 if (qip->i_d.di_nblocks == 0)
1394 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1397 maxlblkcnt = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1399 nmaps = XFS_DQITER_MAP_SIZE;
1401 * We aren't changing the inode itself. Just changing
1402 * some of its data. No new blocks are added here, and
1403 * the inode is never added to the transaction.
1405 xfs_ilock(qip, XFS_ILOCK_SHARED);
1406 error = xfs_bmapi(NULL, qip, lblkno,
1407 maxlblkcnt - lblkno,
1410 0, map, &nmaps, NULL);
1411 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1415 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1416 for (i = 0; i < nmaps; i++) {
1417 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1418 ASSERT(map[i].br_blockcount);
1421 lblkno += map[i].br_blockcount;
1423 if (map[i].br_startblock == HOLESTARTBLOCK)
1426 firstid = (xfs_dqid_t) map[i].br_startoff *
1427 mp->m_quotainfo->qi_dqperchunk;
1429 * Do a read-ahead on the next extent.
1431 if ((i+1 < nmaps) &&
1432 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1433 rablkcnt = map[i+1].br_blockcount;
1434 rablkno = map[i+1].br_startblock;
1435 while (rablkcnt--) {
1436 xfs_buf_readahead(mp->m_ddev_targp,
1437 XFS_FSB_TO_DADDR(mp, rablkno),
1438 mp->m_quotainfo->qi_dqchunklen);
1443 * Iterate thru all the blks in the extent and
1444 * reset the counters of all the dquots inside them.
1446 if ((error = xfs_qm_dqiter_bufs(mp,
1448 map[i].br_startblock,
1449 map[i].br_blockcount,
1457 } while (nmaps > 0);
1465 * Called by dqusage_adjust in doing a quotacheck.
1467 * Given the inode, and a dquot id this updates both the incore dqout as well
1468 * as the buffer copy. This is so that once the quotacheck is done, we can
1469 * just log all the buffers, as opposed to logging numerous updates to
1470 * individual dquots.
1473 xfs_qm_quotacheck_dqadjust(
1474 struct xfs_inode *ip,
1480 struct xfs_mount *mp = ip->i_mount;
1481 struct xfs_dquot *dqp;
1484 error = xfs_qm_dqget(mp, ip, id, type,
1485 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1488 * Shouldn't be able to turn off quotas here.
1490 ASSERT(error != ESRCH);
1491 ASSERT(error != ENOENT);
1495 trace_xfs_dqadjust(dqp);
1498 * Adjust the inode count and the block count to reflect this inode's
1501 be64_add_cpu(&dqp->q_core.d_icount, 1);
1502 dqp->q_res_icount++;
1504 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1505 dqp->q_res_bcount += nblks;
1508 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1509 dqp->q_res_rtbcount += rtblks;
1513 * Set default limits, adjust timers (since we changed usages)
1515 * There are no timers for the default values set in the root dquot.
1517 if (dqp->q_core.d_id) {
1518 xfs_qm_adjust_dqlimits(mp, &dqp->q_core);
1519 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1522 dqp->dq_flags |= XFS_DQ_DIRTY;
1530 xfs_qcnt_t *O_rtblks)
1532 xfs_filblks_t rtblks; /* total rt blks */
1533 xfs_extnum_t idx; /* extent record index */
1534 xfs_ifork_t *ifp; /* inode fork pointer */
1535 xfs_extnum_t nextents; /* number of extent entries */
1538 ASSERT(XFS_IS_REALTIME_INODE(ip));
1539 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1540 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1541 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1545 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1546 for (idx = 0; idx < nextents; idx++)
1547 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1548 *O_rtblks = (xfs_qcnt_t)rtblks;
1553 * callback routine supplied to bulkstat(). Given an inumber, find its
1554 * dquots and update them to account for resources taken by that inode.
1558 xfs_qm_dqusage_adjust(
1559 xfs_mount_t *mp, /* mount point for filesystem */
1560 xfs_ino_t ino, /* inode number to get data for */
1561 void __user *buffer, /* not used */
1562 int ubsize, /* not used */
1563 int *ubused, /* not used */
1564 int *res) /* result code value */
1567 xfs_qcnt_t nblks, rtblks = 0;
1570 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1573 * rootino must have its resources accounted for, not so with the quota
1576 if (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino) {
1577 *res = BULKSTAT_RV_NOTHING;
1578 return XFS_ERROR(EINVAL);
1582 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1583 * interface expects the inode to be exclusively locked because that's
1584 * the case in all other instances. It's OK that we do this because
1585 * quotacheck is done only at mount time.
1587 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1589 *res = BULKSTAT_RV_NOTHING;
1593 ASSERT(ip->i_delayed_blks == 0);
1595 if (XFS_IS_REALTIME_INODE(ip)) {
1597 * Walk thru the extent list and count the realtime blocks.
1599 error = xfs_qm_get_rtblks(ip, &rtblks);
1604 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1607 * Add the (disk blocks and inode) resources occupied by this
1608 * inode to its dquots. We do this adjustment in the incore dquot,
1609 * and also copy the changes to its buffer.
1610 * We don't care about putting these changes in a transaction
1611 * envelope because if we crash in the middle of a 'quotacheck'
1612 * we have to start from the beginning anyway.
1613 * Once we're done, we'll log all the dquot bufs.
1615 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1616 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1618 if (XFS_IS_UQUOTA_ON(mp)) {
1619 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1620 XFS_DQ_USER, nblks, rtblks);
1625 if (XFS_IS_GQUOTA_ON(mp)) {
1626 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1627 XFS_DQ_GROUP, nblks, rtblks);
1632 if (XFS_IS_PQUOTA_ON(mp)) {
1633 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1634 XFS_DQ_PROJ, nblks, rtblks);
1639 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1641 *res = BULKSTAT_RV_DIDONE;
1645 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1647 *res = BULKSTAT_RV_GIVEUP;
1652 * Walk thru all the filesystem inodes and construct a consistent view
1653 * of the disk quota world. If the quotacheck fails, disable quotas.
1659 int done, count, error;
1662 xfs_inode_t *uip, *gip;
1670 ASSERT(mp->m_quotainfo->qi_uquotaip || mp->m_quotainfo->qi_gquotaip);
1671 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1674 * There should be no cached dquots. The (simplistic) quotacheck
1675 * algorithm doesn't like that.
1677 ASSERT(list_empty(&mp->m_quotainfo->qi_dqlist));
1679 cmn_err(CE_NOTE, "XFS quotacheck %s: Please wait.", mp->m_fsname);
1682 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1683 * their counters to zero. We need a clean slate.
1684 * We don't log our changes till later.
1686 uip = mp->m_quotainfo->qi_uquotaip;
1688 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA);
1691 flags |= XFS_UQUOTA_CHKD;
1694 gip = mp->m_quotainfo->qi_gquotaip;
1696 error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
1697 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1700 flags |= XFS_OQUOTA_CHKD;
1705 * Iterate thru all the inodes in the file system,
1706 * adjusting the corresponding dquot counters in core.
1708 error = xfs_bulkstat(mp, &lastino, &count,
1709 xfs_qm_dqusage_adjust,
1710 structsz, NULL, &done);
1717 * We've made all the changes that we need to make incore.
1718 * Flush them down to disk buffers if everything was updated
1722 error = xfs_qm_dqflush_all(mp, 0);
1725 * We can get this error if we couldn't do a dquot allocation inside
1726 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1727 * dirty dquots that might be cached, we just want to get rid of them
1728 * and turn quotaoff. The dquots won't be attached to any of the inodes
1729 * at this point (because we intentionally didn't in dqget_noattach).
1732 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1737 * We didn't log anything, because if we crashed, we'll have to
1738 * start the quotacheck from scratch anyway. However, we must make
1739 * sure that our dquot changes are secure before we put the
1740 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1743 XFS_bflush(mp->m_ddev_targp);
1746 * If one type of quotas is off, then it will lose its
1747 * quotachecked status, since we won't be doing accounting for
1748 * that type anymore.
1750 mp->m_qflags &= ~(XFS_OQUOTA_CHKD | XFS_UQUOTA_CHKD);
1751 mp->m_qflags |= flags;
1753 xfs_qm_dquot_list_print(mp);
1757 cmn_err(CE_WARN, "XFS quotacheck %s: Unsuccessful (Error %d): "
1758 "Disabling quotas.",
1759 mp->m_fsname, error);
1761 * We must turn off quotas.
1763 ASSERT(mp->m_quotainfo != NULL);
1764 ASSERT(xfs_Gqm != NULL);
1765 xfs_qm_destroy_quotainfo(mp);
1766 if (xfs_mount_reset_sbqflags(mp)) {
1767 cmn_err(CE_WARN, "XFS quotacheck %s: "
1768 "Failed to reset quota flags.", mp->m_fsname);
1771 cmn_err(CE_NOTE, "XFS quotacheck %s: Done.", mp->m_fsname);
1777 * This is called after the superblock has been read in and we're ready to
1778 * iget the quota inodes.
1781 xfs_qm_init_quotainos(
1784 xfs_inode_t *uip, *gip;
1789 ASSERT(mp->m_quotainfo);
1795 * Get the uquota and gquota inodes
1797 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1798 if (XFS_IS_UQUOTA_ON(mp) &&
1799 mp->m_sb.sb_uquotino != NULLFSINO) {
1800 ASSERT(mp->m_sb.sb_uquotino > 0);
1801 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1803 return XFS_ERROR(error);
1805 if (XFS_IS_OQUOTA_ON(mp) &&
1806 mp->m_sb.sb_gquotino != NULLFSINO) {
1807 ASSERT(mp->m_sb.sb_gquotino > 0);
1808 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1812 return XFS_ERROR(error);
1816 flags |= XFS_QMOPT_SBVERSION;
1817 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1818 XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
1822 * Create the two inodes, if they don't exist already. The changes
1823 * made above will get added to a transaction and logged in one of
1824 * the qino_alloc calls below. If the device is readonly,
1825 * temporarily switch to read-write to do this.
1827 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1828 if ((error = xfs_qm_qino_alloc(mp, &uip,
1829 sbflags | XFS_SB_UQUOTINO,
1830 flags | XFS_QMOPT_UQUOTA)))
1831 return XFS_ERROR(error);
1833 flags &= ~XFS_QMOPT_SBVERSION;
1835 if (XFS_IS_OQUOTA_ON(mp) && gip == NULL) {
1836 flags |= (XFS_IS_GQUOTA_ON(mp) ?
1837 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1838 error = xfs_qm_qino_alloc(mp, &gip,
1839 sbflags | XFS_SB_GQUOTINO, flags);
1844 return XFS_ERROR(error);
1848 mp->m_quotainfo->qi_uquotaip = uip;
1849 mp->m_quotainfo->qi_gquotaip = gip;
1857 * Just pop the least recently used dquot off the freelist and
1858 * recycle it. The returned dquot is locked.
1860 STATIC xfs_dquot_t *
1861 xfs_qm_dqreclaim_one(void)
1863 xfs_dquot_t *dqpout;
1870 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
1872 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1874 list_for_each_entry(dqp, &xfs_Gqm->qm_dqfrlist, q_freelist) {
1875 struct xfs_mount *mp = dqp->q_mount;
1879 * We are racing with dqlookup here. Naturally we don't
1880 * want to reclaim a dquot that lookup wants. We release the
1881 * freelist lock and start over, so that lookup will grab
1882 * both the dquot and the freelistlock.
1884 if (dqp->dq_flags & XFS_DQ_WANT) {
1885 ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE));
1887 trace_xfs_dqreclaim_want(dqp);
1890 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1891 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
1893 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
1898 * If the dquot is inactive, we are assured that it is
1899 * not on the mplist or the hashlist, and that makes our
1902 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
1904 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
1905 ASSERT(list_empty(&dqp->q_hashlist));
1906 ASSERT(list_empty(&dqp->q_mplist));
1907 list_del_init(&dqp->q_freelist);
1908 xfs_Gqm->qm_dqfrlist_cnt--;
1911 XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
1915 ASSERT(dqp->q_hash);
1916 ASSERT(!list_empty(&dqp->q_mplist));
1919 * Try to grab the flush lock. If this dquot is in the process of
1920 * getting flushed to disk, we don't want to reclaim it.
1922 if (!xfs_dqflock_nowait(dqp)) {
1928 * We have the flush lock so we know that this is not in the
1929 * process of being flushed. So, if this is dirty, flush it
1930 * DELWRI so that we don't get a freelist infested with
1933 if (XFS_DQ_IS_DIRTY(dqp)) {
1936 trace_xfs_dqreclaim_dirty(dqp);
1939 * We flush it delayed write, so don't bother
1940 * releasing the freelist lock.
1942 error = xfs_qm_dqflush(dqp, 0);
1944 xfs_fs_cmn_err(CE_WARN, mp,
1945 "xfs_qm_dqreclaim: dquot %p flush failed", dqp);
1947 xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
1952 * We're trying to get the hashlock out of order. This races
1953 * with dqlookup; so, we giveup and goto the next dquot if
1954 * we couldn't get the hashlock. This way, we won't starve
1955 * a dqlookup process that holds the hashlock that is
1956 * waiting for the freelist lock.
1958 if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
1964 * This races with dquot allocation code as well as dqflush_all
1965 * and reclaim code. So, if we failed to grab the mplist lock,
1966 * giveup everything and start over.
1968 if (!mutex_trylock(&mp->m_quotainfo->qi_dqlist_lock)) {
1970 mutex_unlock(&dqp->q_hash->qh_lock);
1973 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1974 if (restarts++ >= XFS_QM_RECLAIM_MAX_RESTARTS)
1979 ASSERT(dqp->q_nrefs == 0);
1980 list_del_init(&dqp->q_mplist);
1981 mp->m_quotainfo->qi_dquots--;
1982 mp->m_quotainfo->qi_dqreclaims++;
1983 list_del_init(&dqp->q_hashlist);
1984 dqp->q_hash->qh_version++;
1985 list_del_init(&dqp->q_freelist);
1986 xfs_Gqm->qm_dqfrlist_cnt--;
1988 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
1989 mutex_unlock(&dqp->q_hash->qh_lock);
1995 if (restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
1998 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
2003 * Traverse the freelist of dquots and attempt to reclaim a maximum of
2004 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
2005 * favor the lookup function ...
2008 xfs_qm_shake_freelist(
2017 while (nreclaimed < howmany) {
2018 dqp = xfs_qm_dqreclaim_one();
2021 xfs_qm_dqdestroy(dqp);
2028 * The kmem_shake interface is invoked when memory is running low.
2033 struct shrinker *shrink,
2037 int ndqused, nfree, n;
2039 if (!kmem_shake_allow(gfp_mask))
2044 nfree = xfs_Gqm->qm_dqfrlist_cnt; /* free dquots */
2045 /* incore dquots in all f/s's */
2046 ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;
2048 ASSERT(ndqused >= 0);
2050 if (nfree <= ndqused && nfree < ndquot)
2053 ndqused *= xfs_Gqm->qm_dqfree_ratio; /* target # of free dquots */
2054 n = nfree - ndqused - ndquot; /* # over target */
2056 return xfs_qm_shake_freelist(MAX(nfree, n));
2060 /*------------------------------------------------------------------*/
2063 * Return a new incore dquot. Depending on the number of
2064 * dquots in the system, we either allocate a new one on the kernel heap,
2065 * or reclaim a free one.
2066 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2067 * to reclaim an existing one from the freelist.
2070 xfs_qm_dqalloc_incore(
2071 xfs_dquot_t **O_dqpp)
2076 * Check against high water mark to see if we want to pop
2077 * a nincompoop dquot off the freelist.
2079 if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
2081 * Try to recycle a dquot from the freelist.
2083 if ((dqp = xfs_qm_dqreclaim_one())) {
2084 XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
2086 * Just zero the core here. The rest will get
2087 * reinitialized by caller. XXX we shouldn't even
2090 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
2094 XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
2098 * Allocate a brand new dquot on the kernel heap and return it
2099 * to the caller to initialize.
2101 ASSERT(xfs_Gqm->qm_dqzone != NULL);
2102 *O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
2103 atomic_inc(&xfs_Gqm->qm_totaldquots);
2110 * Start a transaction and write the incore superblock changes to
2111 * disk. flags parameter indicates which fields have changed.
2114 xfs_qm_write_sb_changes(
2122 cmn_err(CE_NOTE, "Writing superblock quota changes :%s", mp->m_fsname);
2124 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
2125 if ((error = xfs_trans_reserve(tp, 0,
2126 mp->m_sb.sb_sectsize + 128, 0,
2128 XFS_DEFAULT_LOG_COUNT))) {
2129 xfs_trans_cancel(tp, 0);
2133 xfs_mod_sb(tp, flags);
2134 error = xfs_trans_commit(tp, 0);
2140 /* --------------- utility functions for vnodeops ---------------- */
2144 * Given an inode, a uid, gid and prid make sure that we have
2145 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2146 * quotas by creating this file.
2147 * This also attaches dquot(s) to the given inode after locking it,
2148 * and returns the dquots corresponding to the uid and/or gid.
2150 * in : inode (unlocked)
2151 * out : udquot, gdquot with references taken and unlocked
2155 struct xfs_inode *ip,
2160 struct xfs_dquot **O_udqpp,
2161 struct xfs_dquot **O_gdqpp)
2163 struct xfs_mount *mp = ip->i_mount;
2164 struct xfs_dquot *uq, *gq;
2168 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2171 lockflags = XFS_ILOCK_EXCL;
2172 xfs_ilock(ip, lockflags);
2174 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
2175 gid = ip->i_d.di_gid;
2178 * Attach the dquot(s) to this inode, doing a dquot allocation
2179 * if necessary. The dquot(s) will not be locked.
2181 if (XFS_NOT_DQATTACHED(mp, ip)) {
2182 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
2184 xfs_iunlock(ip, lockflags);
2190 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
2191 if (ip->i_d.di_uid != uid) {
2193 * What we need is the dquot that has this uid, and
2194 * if we send the inode to dqget, the uid of the inode
2195 * takes priority over what's sent in the uid argument.
2196 * We must unlock inode here before calling dqget if
2197 * we're not sending the inode, because otherwise
2198 * we'll deadlock by doing trans_reserve while
2201 xfs_iunlock(ip, lockflags);
2202 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
2207 ASSERT(error != ENOENT);
2211 * Get the ilock in the right order.
2214 lockflags = XFS_ILOCK_SHARED;
2215 xfs_ilock(ip, lockflags);
2218 * Take an extra reference, because we'll return
2221 ASSERT(ip->i_udquot);
2228 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
2229 if (ip->i_d.di_gid != gid) {
2230 xfs_iunlock(ip, lockflags);
2231 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
2238 ASSERT(error != ENOENT);
2242 lockflags = XFS_ILOCK_SHARED;
2243 xfs_ilock(ip, lockflags);
2245 ASSERT(ip->i_gdquot);
2251 } else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
2252 if (xfs_get_projid(ip) != prid) {
2253 xfs_iunlock(ip, lockflags);
2254 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
2261 ASSERT(error != ENOENT);
2265 lockflags = XFS_ILOCK_SHARED;
2266 xfs_ilock(ip, lockflags);
2268 ASSERT(ip->i_gdquot);
2276 trace_xfs_dquot_dqalloc(ip);
2278 xfs_iunlock(ip, lockflags);
2291 * Actually transfer ownership, and do dquot modifications.
2292 * These were already reserved.
2298 xfs_dquot_t **IO_olddq,
2301 xfs_dquot_t *prevdq;
2302 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
2303 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
2306 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2307 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
2312 ASSERT(prevdq != newdq);
2314 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
2315 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
2317 /* the sparkling new dquot */
2318 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
2319 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
2322 * Take an extra reference, because the inode
2323 * is going to keep this dquot pointer even
2324 * after the trans_commit.
2328 xfs_dqunlock(newdq);
2335 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
2338 xfs_qm_vop_chown_reserve(
2345 xfs_mount_t *mp = ip->i_mount;
2346 uint delblks, blkflags, prjflags = 0;
2347 xfs_dquot_t *unresudq, *unresgdq, *delblksudq, *delblksgdq;
2351 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2352 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2354 delblks = ip->i_delayed_blks;
2355 delblksudq = delblksgdq = unresudq = unresgdq = NULL;
2356 blkflags = XFS_IS_REALTIME_INODE(ip) ?
2357 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
2359 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2360 ip->i_d.di_uid != (uid_t)be32_to_cpu(udqp->q_core.d_id)) {
2363 * If there are delayed allocation blocks, then we have to
2364 * unreserve those from the old dquot, and add them to the
2368 ASSERT(ip->i_udquot);
2369 unresudq = ip->i_udquot;
2372 if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
2373 if (XFS_IS_PQUOTA_ON(ip->i_mount) &&
2374 xfs_get_projid(ip) != be32_to_cpu(gdqp->q_core.d_id))
2375 prjflags = XFS_QMOPT_ENOSPC;
2378 (XFS_IS_GQUOTA_ON(ip->i_mount) &&
2379 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id))) {
2382 ASSERT(ip->i_gdquot);
2383 unresgdq = ip->i_gdquot;
2388 if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2389 delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
2390 flags | blkflags | prjflags)))
2394 * Do the delayed blks reservations/unreservations now. Since, these
2395 * are done without the help of a transaction, if a reservation fails
2396 * its previous reservations won't be automatically undone by trans
2397 * code. So, we have to do it manually here.
2401 * Do the reservations first. Unreservation can't fail.
2403 ASSERT(delblksudq || delblksgdq);
2404 ASSERT(unresudq || unresgdq);
2405 if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2406 delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
2407 flags | blkflags | prjflags)))
2409 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2410 unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
2418 xfs_qm_vop_rename_dqattach(
2419 struct xfs_inode **i_tab)
2421 struct xfs_mount *mp = i_tab[0]->i_mount;
2424 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2427 for (i = 0; (i < 4 && i_tab[i]); i++) {
2428 struct xfs_inode *ip = i_tab[i];
2432 * Watch out for duplicate entries in the table.
2434 if (i == 0 || ip != i_tab[i-1]) {
2435 if (XFS_NOT_DQATTACHED(mp, ip)) {
2436 error = xfs_qm_dqattach(ip, 0);
2446 xfs_qm_vop_create_dqattach(
2447 struct xfs_trans *tp,
2448 struct xfs_inode *ip,
2449 struct xfs_dquot *udqp,
2450 struct xfs_dquot *gdqp)
2452 struct xfs_mount *mp = tp->t_mountp;
2454 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2457 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2458 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2464 ASSERT(ip->i_udquot == NULL);
2465 ip->i_udquot = udqp;
2466 ASSERT(XFS_IS_UQUOTA_ON(mp));
2467 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
2468 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2474 ASSERT(ip->i_gdquot == NULL);
2475 ip->i_gdquot = gdqp;
2476 ASSERT(XFS_IS_OQUOTA_ON(mp));
2477 ASSERT((XFS_IS_GQUOTA_ON(mp) ?
2478 ip->i_d.di_gid : xfs_get_projid(ip)) ==
2479 be32_to_cpu(gdqp->q_core.d_id));
2480 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);