mm: CONFIG_ZONE_DEVICE stop depending on CONFIG_EXPERT
[linux-2.6/btrfs-unstable.git] / fs / xfs / xfs_qm.c
bloba60d9e2739d14a2ebcff8ee7dcedae7f5177bf3c
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
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
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_sb.h"
26 #include "xfs_mount.h"
27 #include "xfs_inode.h"
28 #include "xfs_ialloc.h"
29 #include "xfs_itable.h"
30 #include "xfs_quota.h"
31 #include "xfs_error.h"
32 #include "xfs_bmap.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_trans.h"
35 #include "xfs_trans_space.h"
36 #include "xfs_qm.h"
37 #include "xfs_trace.h"
38 #include "xfs_icache.h"
39 #include "xfs_cksum.h"
42 * The global quota manager. There is only one of these for the entire
43 * system, _not_ one per file system. XQM keeps track of the overall
44 * quota functionality, including maintaining the freelist and hash
45 * tables of dquots.
47 STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
48 STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
51 STATIC void xfs_qm_dqfree_one(struct xfs_dquot *dqp);
53 * We use the batch lookup interface to iterate over the dquots as it
54 * currently is the only interface into the radix tree code that allows
55 * fuzzy lookups instead of exact matches. Holding the lock over multiple
56 * operations is fine as all callers are used either during mount/umount
57 * or quotaoff.
59 #define XFS_DQ_LOOKUP_BATCH 32
61 STATIC int
62 xfs_qm_dquot_walk(
63 struct xfs_mount *mp,
64 int type,
65 int (*execute)(struct xfs_dquot *dqp, void *data),
66 void *data)
68 struct xfs_quotainfo *qi = mp->m_quotainfo;
69 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
70 uint32_t next_index;
71 int last_error = 0;
72 int skipped;
73 int nr_found;
75 restart:
76 skipped = 0;
77 next_index = 0;
78 nr_found = 0;
80 while (1) {
81 struct xfs_dquot *batch[XFS_DQ_LOOKUP_BATCH];
82 int error = 0;
83 int i;
85 mutex_lock(&qi->qi_tree_lock);
86 nr_found = radix_tree_gang_lookup(tree, (void **)batch,
87 next_index, XFS_DQ_LOOKUP_BATCH);
88 if (!nr_found) {
89 mutex_unlock(&qi->qi_tree_lock);
90 break;
93 for (i = 0; i < nr_found; i++) {
94 struct xfs_dquot *dqp = batch[i];
96 next_index = be32_to_cpu(dqp->q_core.d_id) + 1;
98 error = execute(batch[i], data);
99 if (error == -EAGAIN) {
100 skipped++;
101 continue;
103 if (error && last_error != -EFSCORRUPTED)
104 last_error = error;
107 mutex_unlock(&qi->qi_tree_lock);
109 /* bail out if the filesystem is corrupted. */
110 if (last_error == -EFSCORRUPTED) {
111 skipped = 0;
112 break;
116 if (skipped) {
117 delay(1);
118 goto restart;
121 return last_error;
126 * Purge a dquot from all tracking data structures and free it.
128 STATIC int
129 xfs_qm_dqpurge(
130 struct xfs_dquot *dqp,
131 void *data)
133 struct xfs_mount *mp = dqp->q_mount;
134 struct xfs_quotainfo *qi = mp->m_quotainfo;
136 xfs_dqlock(dqp);
137 if ((dqp->dq_flags & XFS_DQ_FREEING) || dqp->q_nrefs != 0) {
138 xfs_dqunlock(dqp);
139 return -EAGAIN;
142 dqp->dq_flags |= XFS_DQ_FREEING;
144 xfs_dqflock(dqp);
147 * If we are turning this type of quotas off, we don't care
148 * about the dirty metadata sitting in this dquot. OTOH, if
149 * we're unmounting, we do care, so we flush it and wait.
151 if (XFS_DQ_IS_DIRTY(dqp)) {
152 struct xfs_buf *bp = NULL;
153 int error;
156 * We don't care about getting disk errors here. We need
157 * to purge this dquot anyway, so we go ahead regardless.
159 error = xfs_qm_dqflush(dqp, &bp);
160 if (error) {
161 xfs_warn(mp, "%s: dquot %p flush failed",
162 __func__, dqp);
163 } else {
164 error = xfs_bwrite(bp);
165 xfs_buf_relse(bp);
167 xfs_dqflock(dqp);
170 ASSERT(atomic_read(&dqp->q_pincount) == 0);
171 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
172 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
174 xfs_dqfunlock(dqp);
175 xfs_dqunlock(dqp);
177 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
178 be32_to_cpu(dqp->q_core.d_id));
179 qi->qi_dquots--;
182 * We move dquots to the freelist as soon as their reference count
183 * hits zero, so it really should be on the freelist here.
185 ASSERT(!list_empty(&dqp->q_lru));
186 list_lru_del(&qi->qi_lru, &dqp->q_lru);
187 XFS_STATS_DEC(mp, xs_qm_dquot_unused);
189 xfs_qm_dqdestroy(dqp);
190 return 0;
194 * Purge the dquot cache.
196 void
197 xfs_qm_dqpurge_all(
198 struct xfs_mount *mp,
199 uint flags)
201 if (flags & XFS_QMOPT_UQUOTA)
202 xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_dqpurge, NULL);
203 if (flags & XFS_QMOPT_GQUOTA)
204 xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_dqpurge, NULL);
205 if (flags & XFS_QMOPT_PQUOTA)
206 xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_dqpurge, NULL);
210 * Just destroy the quotainfo structure.
212 void
213 xfs_qm_unmount(
214 struct xfs_mount *mp)
216 if (mp->m_quotainfo) {
217 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
218 xfs_qm_destroy_quotainfo(mp);
223 * Called from the vfsops layer.
225 void
226 xfs_qm_unmount_quotas(
227 xfs_mount_t *mp)
230 * Release the dquots that root inode, et al might be holding,
231 * before we flush quotas and blow away the quotainfo structure.
233 ASSERT(mp->m_rootip);
234 xfs_qm_dqdetach(mp->m_rootip);
235 if (mp->m_rbmip)
236 xfs_qm_dqdetach(mp->m_rbmip);
237 if (mp->m_rsumip)
238 xfs_qm_dqdetach(mp->m_rsumip);
241 * Release the quota inodes.
243 if (mp->m_quotainfo) {
244 if (mp->m_quotainfo->qi_uquotaip) {
245 IRELE(mp->m_quotainfo->qi_uquotaip);
246 mp->m_quotainfo->qi_uquotaip = NULL;
248 if (mp->m_quotainfo->qi_gquotaip) {
249 IRELE(mp->m_quotainfo->qi_gquotaip);
250 mp->m_quotainfo->qi_gquotaip = NULL;
252 if (mp->m_quotainfo->qi_pquotaip) {
253 IRELE(mp->m_quotainfo->qi_pquotaip);
254 mp->m_quotainfo->qi_pquotaip = NULL;
259 STATIC int
260 xfs_qm_dqattach_one(
261 xfs_inode_t *ip,
262 xfs_dqid_t id,
263 uint type,
264 uint doalloc,
265 xfs_dquot_t **IO_idqpp)
267 xfs_dquot_t *dqp;
268 int error;
270 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
271 error = 0;
274 * See if we already have it in the inode itself. IO_idqpp is &i_udquot
275 * or &i_gdquot. This made the code look weird, but made the logic a lot
276 * simpler.
278 dqp = *IO_idqpp;
279 if (dqp) {
280 trace_xfs_dqattach_found(dqp);
281 return 0;
285 * Find the dquot from somewhere. This bumps the reference count of
286 * dquot and returns it locked. This can return ENOENT if dquot didn't
287 * exist on disk and we didn't ask it to allocate; ESRCH if quotas got
288 * turned off suddenly.
290 error = xfs_qm_dqget(ip->i_mount, ip, id, type,
291 doalloc | XFS_QMOPT_DOWARN, &dqp);
292 if (error)
293 return error;
295 trace_xfs_dqattach_get(dqp);
298 * dqget may have dropped and re-acquired the ilock, but it guarantees
299 * that the dquot returned is the one that should go in the inode.
301 *IO_idqpp = dqp;
302 xfs_dqunlock(dqp);
303 return 0;
306 static bool
307 xfs_qm_need_dqattach(
308 struct xfs_inode *ip)
310 struct xfs_mount *mp = ip->i_mount;
312 if (!XFS_IS_QUOTA_RUNNING(mp))
313 return false;
314 if (!XFS_IS_QUOTA_ON(mp))
315 return false;
316 if (!XFS_NOT_DQATTACHED(mp, ip))
317 return false;
318 if (xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
319 return false;
320 return true;
324 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
325 * into account.
326 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
327 * Inode may get unlocked and relocked in here, and the caller must deal with
328 * the consequences.
331 xfs_qm_dqattach_locked(
332 xfs_inode_t *ip,
333 uint flags)
335 xfs_mount_t *mp = ip->i_mount;
336 int error = 0;
338 if (!xfs_qm_need_dqattach(ip))
339 return 0;
341 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
343 if (XFS_IS_UQUOTA_ON(mp) && !ip->i_udquot) {
344 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
345 flags & XFS_QMOPT_DQALLOC,
346 &ip->i_udquot);
347 if (error)
348 goto done;
349 ASSERT(ip->i_udquot);
352 if (XFS_IS_GQUOTA_ON(mp) && !ip->i_gdquot) {
353 error = xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
354 flags & XFS_QMOPT_DQALLOC,
355 &ip->i_gdquot);
356 if (error)
357 goto done;
358 ASSERT(ip->i_gdquot);
361 if (XFS_IS_PQUOTA_ON(mp) && !ip->i_pdquot) {
362 error = xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
363 flags & XFS_QMOPT_DQALLOC,
364 &ip->i_pdquot);
365 if (error)
366 goto done;
367 ASSERT(ip->i_pdquot);
370 done:
372 * Don't worry about the dquots that we may have attached before any
373 * error - they'll get detached later if it has not already been done.
375 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
376 return error;
380 xfs_qm_dqattach(
381 struct xfs_inode *ip,
382 uint flags)
384 int error;
386 if (!xfs_qm_need_dqattach(ip))
387 return 0;
389 xfs_ilock(ip, XFS_ILOCK_EXCL);
390 error = xfs_qm_dqattach_locked(ip, flags);
391 xfs_iunlock(ip, XFS_ILOCK_EXCL);
393 return error;
397 * Release dquots (and their references) if any.
398 * The inode should be locked EXCL except when this's called by
399 * xfs_ireclaim.
401 void
402 xfs_qm_dqdetach(
403 xfs_inode_t *ip)
405 if (!(ip->i_udquot || ip->i_gdquot || ip->i_pdquot))
406 return;
408 trace_xfs_dquot_dqdetach(ip);
410 ASSERT(!xfs_is_quota_inode(&ip->i_mount->m_sb, ip->i_ino));
411 if (ip->i_udquot) {
412 xfs_qm_dqrele(ip->i_udquot);
413 ip->i_udquot = NULL;
415 if (ip->i_gdquot) {
416 xfs_qm_dqrele(ip->i_gdquot);
417 ip->i_gdquot = NULL;
419 if (ip->i_pdquot) {
420 xfs_qm_dqrele(ip->i_pdquot);
421 ip->i_pdquot = NULL;
425 struct xfs_qm_isolate {
426 struct list_head buffers;
427 struct list_head dispose;
430 static enum lru_status
431 xfs_qm_dquot_isolate(
432 struct list_head *item,
433 struct list_lru_one *lru,
434 spinlock_t *lru_lock,
435 void *arg)
436 __releases(lru_lock) __acquires(lru_lock)
438 struct xfs_dquot *dqp = container_of(item,
439 struct xfs_dquot, q_lru);
440 struct xfs_qm_isolate *isol = arg;
442 if (!xfs_dqlock_nowait(dqp))
443 goto out_miss_busy;
446 * This dquot has acquired a reference in the meantime remove it from
447 * the freelist and try again.
449 if (dqp->q_nrefs) {
450 xfs_dqunlock(dqp);
451 XFS_STATS_INC(dqp->q_mount, xs_qm_dqwants);
453 trace_xfs_dqreclaim_want(dqp);
454 list_lru_isolate(lru, &dqp->q_lru);
455 XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot_unused);
456 return LRU_REMOVED;
460 * If the dquot is dirty, flush it. If it's already being flushed, just
461 * skip it so there is time for the IO to complete before we try to
462 * reclaim it again on the next LRU pass.
464 if (!xfs_dqflock_nowait(dqp)) {
465 xfs_dqunlock(dqp);
466 goto out_miss_busy;
469 if (XFS_DQ_IS_DIRTY(dqp)) {
470 struct xfs_buf *bp = NULL;
471 int error;
473 trace_xfs_dqreclaim_dirty(dqp);
475 /* we have to drop the LRU lock to flush the dquot */
476 spin_unlock(lru_lock);
478 error = xfs_qm_dqflush(dqp, &bp);
479 if (error) {
480 xfs_warn(dqp->q_mount, "%s: dquot %p flush failed",
481 __func__, dqp);
482 goto out_unlock_dirty;
485 xfs_buf_delwri_queue(bp, &isol->buffers);
486 xfs_buf_relse(bp);
487 goto out_unlock_dirty;
489 xfs_dqfunlock(dqp);
492 * Prevent lookups now that we are past the point of no return.
494 dqp->dq_flags |= XFS_DQ_FREEING;
495 xfs_dqunlock(dqp);
497 ASSERT(dqp->q_nrefs == 0);
498 list_lru_isolate_move(lru, &dqp->q_lru, &isol->dispose);
499 XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot_unused);
500 trace_xfs_dqreclaim_done(dqp);
501 XFS_STATS_INC(dqp->q_mount, xs_qm_dqreclaims);
502 return LRU_REMOVED;
504 out_miss_busy:
505 trace_xfs_dqreclaim_busy(dqp);
506 XFS_STATS_INC(dqp->q_mount, xs_qm_dqreclaim_misses);
507 return LRU_SKIP;
509 out_unlock_dirty:
510 trace_xfs_dqreclaim_busy(dqp);
511 XFS_STATS_INC(dqp->q_mount, xs_qm_dqreclaim_misses);
512 xfs_dqunlock(dqp);
513 spin_lock(lru_lock);
514 return LRU_RETRY;
517 static unsigned long
518 xfs_qm_shrink_scan(
519 struct shrinker *shrink,
520 struct shrink_control *sc)
522 struct xfs_quotainfo *qi = container_of(shrink,
523 struct xfs_quotainfo, qi_shrinker);
524 struct xfs_qm_isolate isol;
525 unsigned long freed;
526 int error;
528 if ((sc->gfp_mask & (__GFP_FS|__GFP_DIRECT_RECLAIM)) != (__GFP_FS|__GFP_DIRECT_RECLAIM))
529 return 0;
531 INIT_LIST_HEAD(&isol.buffers);
532 INIT_LIST_HEAD(&isol.dispose);
534 freed = list_lru_shrink_walk(&qi->qi_lru, sc,
535 xfs_qm_dquot_isolate, &isol);
537 error = xfs_buf_delwri_submit(&isol.buffers);
538 if (error)
539 xfs_warn(NULL, "%s: dquot reclaim failed", __func__);
541 while (!list_empty(&isol.dispose)) {
542 struct xfs_dquot *dqp;
544 dqp = list_first_entry(&isol.dispose, struct xfs_dquot, q_lru);
545 list_del_init(&dqp->q_lru);
546 xfs_qm_dqfree_one(dqp);
549 return freed;
552 static unsigned long
553 xfs_qm_shrink_count(
554 struct shrinker *shrink,
555 struct shrink_control *sc)
557 struct xfs_quotainfo *qi = container_of(shrink,
558 struct xfs_quotainfo, qi_shrinker);
560 return list_lru_shrink_count(&qi->qi_lru, sc);
563 STATIC void
564 xfs_qm_set_defquota(
565 xfs_mount_t *mp,
566 uint type,
567 xfs_quotainfo_t *qinf)
569 xfs_dquot_t *dqp;
570 struct xfs_def_quota *defq;
571 int error;
573 error = xfs_qm_dqread(mp, 0, type, XFS_QMOPT_DOWARN, &dqp);
575 if (!error) {
576 xfs_disk_dquot_t *ddqp = &dqp->q_core;
578 defq = xfs_get_defquota(dqp, qinf);
581 * Timers and warnings have been already set, let's just set the
582 * default limits for this quota type
584 defq->bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
585 defq->bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
586 defq->ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
587 defq->isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
588 defq->rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
589 defq->rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
590 xfs_qm_dqdestroy(dqp);
595 * This initializes all the quota information that's kept in the
596 * mount structure
598 STATIC int
599 xfs_qm_init_quotainfo(
600 xfs_mount_t *mp)
602 xfs_quotainfo_t *qinf;
603 int error;
604 xfs_dquot_t *dqp;
606 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
608 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
610 error = list_lru_init(&qinf->qi_lru);
611 if (error)
612 goto out_free_qinf;
615 * See if quotainodes are setup, and if not, allocate them,
616 * and change the superblock accordingly.
618 error = xfs_qm_init_quotainos(mp);
619 if (error)
620 goto out_free_lru;
622 INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
623 INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
624 INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS);
625 mutex_init(&qinf->qi_tree_lock);
627 /* mutex used to serialize quotaoffs */
628 mutex_init(&qinf->qi_quotaofflock);
630 /* Precalc some constants */
631 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
632 qinf->qi_dqperchunk = xfs_calc_dquots_per_chunk(qinf->qi_dqchunklen);
634 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
637 * We try to get the limits from the superuser's limits fields.
638 * This is quite hacky, but it is standard quota practice.
640 * Since we may not have done a quotacheck by this point, just read
641 * the dquot without attaching it to any hashtables or lists.
643 * Timers and warnings are globally set by the first timer found in
644 * user/group/proj quota types, otherwise a default value is used.
645 * This should be split into different fields per quota type.
647 error = xfs_qm_dqread(mp, 0,
648 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
649 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
650 XFS_DQ_PROJ),
651 XFS_QMOPT_DOWARN, &dqp);
653 if (!error) {
654 xfs_disk_dquot_t *ddqp = &dqp->q_core;
657 * The warnings and timers set the grace period given to
658 * a user or group before he or she can not perform any
659 * more writing. If it is zero, a default is used.
661 qinf->qi_btimelimit = ddqp->d_btimer ?
662 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
663 qinf->qi_itimelimit = ddqp->d_itimer ?
664 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
665 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
666 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
667 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
668 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
669 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
670 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
671 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
672 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
673 xfs_qm_dqdestroy(dqp);
674 } else {
675 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
676 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
677 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
678 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
679 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
680 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
683 if (XFS_IS_UQUOTA_RUNNING(mp))
684 xfs_qm_set_defquota(mp, XFS_DQ_USER, qinf);
685 if (XFS_IS_GQUOTA_RUNNING(mp))
686 xfs_qm_set_defquota(mp, XFS_DQ_GROUP, qinf);
687 if (XFS_IS_PQUOTA_RUNNING(mp))
688 xfs_qm_set_defquota(mp, XFS_DQ_PROJ, qinf);
690 qinf->qi_shrinker.count_objects = xfs_qm_shrink_count;
691 qinf->qi_shrinker.scan_objects = xfs_qm_shrink_scan;
692 qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
693 qinf->qi_shrinker.flags = SHRINKER_NUMA_AWARE;
694 register_shrinker(&qinf->qi_shrinker);
695 return 0;
697 out_free_lru:
698 list_lru_destroy(&qinf->qi_lru);
699 out_free_qinf:
700 kmem_free(qinf);
701 mp->m_quotainfo = NULL;
702 return error;
707 * Gets called when unmounting a filesystem or when all quotas get
708 * turned off.
709 * This purges the quota inodes, destroys locks and frees itself.
711 void
712 xfs_qm_destroy_quotainfo(
713 xfs_mount_t *mp)
715 xfs_quotainfo_t *qi;
717 qi = mp->m_quotainfo;
718 ASSERT(qi != NULL);
720 unregister_shrinker(&qi->qi_shrinker);
721 list_lru_destroy(&qi->qi_lru);
723 if (qi->qi_uquotaip) {
724 IRELE(qi->qi_uquotaip);
725 qi->qi_uquotaip = NULL; /* paranoia */
727 if (qi->qi_gquotaip) {
728 IRELE(qi->qi_gquotaip);
729 qi->qi_gquotaip = NULL;
731 if (qi->qi_pquotaip) {
732 IRELE(qi->qi_pquotaip);
733 qi->qi_pquotaip = NULL;
735 mutex_destroy(&qi->qi_quotaofflock);
736 kmem_free(qi);
737 mp->m_quotainfo = NULL;
741 * Create an inode and return with a reference already taken, but unlocked
742 * This is how we create quota inodes
744 STATIC int
745 xfs_qm_qino_alloc(
746 xfs_mount_t *mp,
747 xfs_inode_t **ip,
748 uint flags)
750 xfs_trans_t *tp;
751 int error;
752 int committed;
753 bool need_alloc = true;
755 *ip = NULL;
757 * With superblock that doesn't have separate pquotino, we
758 * share an inode between gquota and pquota. If the on-disk
759 * superblock has GQUOTA and the filesystem is now mounted
760 * with PQUOTA, just use sb_gquotino for sb_pquotino and
761 * vice-versa.
763 if (!xfs_sb_version_has_pquotino(&mp->m_sb) &&
764 (flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) {
765 xfs_ino_t ino = NULLFSINO;
767 if ((flags & XFS_QMOPT_PQUOTA) &&
768 (mp->m_sb.sb_gquotino != NULLFSINO)) {
769 ino = mp->m_sb.sb_gquotino;
770 ASSERT(mp->m_sb.sb_pquotino == NULLFSINO);
771 } else if ((flags & XFS_QMOPT_GQUOTA) &&
772 (mp->m_sb.sb_pquotino != NULLFSINO)) {
773 ino = mp->m_sb.sb_pquotino;
774 ASSERT(mp->m_sb.sb_gquotino == NULLFSINO);
776 if (ino != NULLFSINO) {
777 error = xfs_iget(mp, NULL, ino, 0, 0, ip);
778 if (error)
779 return error;
780 mp->m_sb.sb_gquotino = NULLFSINO;
781 mp->m_sb.sb_pquotino = NULLFSINO;
782 need_alloc = false;
786 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_create,
787 XFS_QM_QINOCREATE_SPACE_RES(mp), 0, 0, &tp);
788 if (error)
789 return error;
791 if (need_alloc) {
792 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip,
793 &committed);
794 if (error) {
795 xfs_trans_cancel(tp);
796 return error;
801 * Make the changes in the superblock, and log those too.
802 * sbfields arg may contain fields other than *QUOTINO;
803 * VERSIONNUM for example.
805 spin_lock(&mp->m_sb_lock);
806 if (flags & XFS_QMOPT_SBVERSION) {
807 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
809 xfs_sb_version_addquota(&mp->m_sb);
810 mp->m_sb.sb_uquotino = NULLFSINO;
811 mp->m_sb.sb_gquotino = NULLFSINO;
812 mp->m_sb.sb_pquotino = NULLFSINO;
814 /* qflags will get updated fully _after_ quotacheck */
815 mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT;
817 if (flags & XFS_QMOPT_UQUOTA)
818 mp->m_sb.sb_uquotino = (*ip)->i_ino;
819 else if (flags & XFS_QMOPT_GQUOTA)
820 mp->m_sb.sb_gquotino = (*ip)->i_ino;
821 else
822 mp->m_sb.sb_pquotino = (*ip)->i_ino;
823 spin_unlock(&mp->m_sb_lock);
824 xfs_log_sb(tp);
826 error = xfs_trans_commit(tp);
827 if (error) {
828 ASSERT(XFS_FORCED_SHUTDOWN(mp));
829 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
831 if (need_alloc)
832 xfs_finish_inode_setup(*ip);
833 return error;
837 STATIC void
838 xfs_qm_reset_dqcounts(
839 xfs_mount_t *mp,
840 xfs_buf_t *bp,
841 xfs_dqid_t id,
842 uint type)
844 struct xfs_dqblk *dqb;
845 int j;
847 trace_xfs_reset_dqcounts(bp, _RET_IP_);
850 * Reset all counters and timers. They'll be
851 * started afresh by xfs_qm_quotacheck.
853 #ifdef DEBUG
854 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
855 do_div(j, sizeof(xfs_dqblk_t));
856 ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
857 #endif
858 dqb = bp->b_addr;
859 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
860 struct xfs_disk_dquot *ddq;
862 ddq = (struct xfs_disk_dquot *)&dqb[j];
865 * Do a sanity check, and if needed, repair the dqblk. Don't
866 * output any warnings because it's perfectly possible to
867 * find uninitialised dquot blks. See comment in xfs_dqcheck.
869 xfs_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
870 "xfs_quotacheck");
872 * Reset type in case we are reusing group quota file for
873 * project quotas or vice versa
875 ddq->d_flags = type;
876 ddq->d_bcount = 0;
877 ddq->d_icount = 0;
878 ddq->d_rtbcount = 0;
879 ddq->d_btimer = 0;
880 ddq->d_itimer = 0;
881 ddq->d_rtbtimer = 0;
882 ddq->d_bwarns = 0;
883 ddq->d_iwarns = 0;
884 ddq->d_rtbwarns = 0;
886 if (xfs_sb_version_hascrc(&mp->m_sb)) {
887 xfs_update_cksum((char *)&dqb[j],
888 sizeof(struct xfs_dqblk),
889 XFS_DQUOT_CRC_OFF);
894 STATIC int
895 xfs_qm_dqiter_bufs(
896 struct xfs_mount *mp,
897 xfs_dqid_t firstid,
898 xfs_fsblock_t bno,
899 xfs_filblks_t blkcnt,
900 uint flags,
901 struct list_head *buffer_list)
903 struct xfs_buf *bp;
904 int error;
905 int type;
907 ASSERT(blkcnt > 0);
908 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
909 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
910 error = 0;
913 * Blkcnt arg can be a very big number, and might even be
914 * larger than the log itself. So, we have to break it up into
915 * manageable-sized transactions.
916 * Note that we don't start a permanent transaction here; we might
917 * not be able to get a log reservation for the whole thing up front,
918 * and we don't really care to either, because we just discard
919 * everything if we were to crash in the middle of this loop.
921 while (blkcnt--) {
922 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
923 XFS_FSB_TO_DADDR(mp, bno),
924 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
925 &xfs_dquot_buf_ops);
928 * CRC and validation errors will return a EFSCORRUPTED here. If
929 * this occurs, re-read without CRC validation so that we can
930 * repair the damage via xfs_qm_reset_dqcounts(). This process
931 * will leave a trace in the log indicating corruption has
932 * been detected.
934 if (error == -EFSCORRUPTED) {
935 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
936 XFS_FSB_TO_DADDR(mp, bno),
937 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
938 NULL);
941 if (error)
942 break;
945 * A corrupt buffer might not have a verifier attached, so
946 * make sure we have the correct one attached before writeback
947 * occurs.
949 bp->b_ops = &xfs_dquot_buf_ops;
950 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
951 xfs_buf_delwri_queue(bp, buffer_list);
952 xfs_buf_relse(bp);
954 /* goto the next block. */
955 bno++;
956 firstid += mp->m_quotainfo->qi_dqperchunk;
959 return error;
963 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
964 * caller supplied function for every chunk of dquots that we find.
966 STATIC int
967 xfs_qm_dqiterate(
968 struct xfs_mount *mp,
969 struct xfs_inode *qip,
970 uint flags,
971 struct list_head *buffer_list)
973 struct xfs_bmbt_irec *map;
974 int i, nmaps; /* number of map entries */
975 int error; /* return value */
976 xfs_fileoff_t lblkno;
977 xfs_filblks_t maxlblkcnt;
978 xfs_dqid_t firstid;
979 xfs_fsblock_t rablkno;
980 xfs_filblks_t rablkcnt;
982 error = 0;
984 * This looks racy, but we can't keep an inode lock across a
985 * trans_reserve. But, this gets called during quotacheck, and that
986 * happens only at mount time which is single threaded.
988 if (qip->i_d.di_nblocks == 0)
989 return 0;
991 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
993 lblkno = 0;
994 maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
995 do {
996 uint lock_mode;
998 nmaps = XFS_DQITER_MAP_SIZE;
1000 * We aren't changing the inode itself. Just changing
1001 * some of its data. No new blocks are added here, and
1002 * the inode is never added to the transaction.
1004 lock_mode = xfs_ilock_data_map_shared(qip);
1005 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
1006 map, &nmaps, 0);
1007 xfs_iunlock(qip, lock_mode);
1008 if (error)
1009 break;
1011 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1012 for (i = 0; i < nmaps; i++) {
1013 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1014 ASSERT(map[i].br_blockcount);
1017 lblkno += map[i].br_blockcount;
1019 if (map[i].br_startblock == HOLESTARTBLOCK)
1020 continue;
1022 firstid = (xfs_dqid_t) map[i].br_startoff *
1023 mp->m_quotainfo->qi_dqperchunk;
1025 * Do a read-ahead on the next extent.
1027 if ((i+1 < nmaps) &&
1028 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1029 rablkcnt = map[i+1].br_blockcount;
1030 rablkno = map[i+1].br_startblock;
1031 while (rablkcnt--) {
1032 xfs_buf_readahead(mp->m_ddev_targp,
1033 XFS_FSB_TO_DADDR(mp, rablkno),
1034 mp->m_quotainfo->qi_dqchunklen,
1035 &xfs_dquot_buf_ops);
1036 rablkno++;
1040 * Iterate thru all the blks in the extent and
1041 * reset the counters of all the dquots inside them.
1043 error = xfs_qm_dqiter_bufs(mp, firstid,
1044 map[i].br_startblock,
1045 map[i].br_blockcount,
1046 flags, buffer_list);
1047 if (error)
1048 goto out;
1050 } while (nmaps > 0);
1052 out:
1053 kmem_free(map);
1054 return error;
1058 * Called by dqusage_adjust in doing a quotacheck.
1060 * Given the inode, and a dquot id this updates both the incore dqout as well
1061 * as the buffer copy. This is so that once the quotacheck is done, we can
1062 * just log all the buffers, as opposed to logging numerous updates to
1063 * individual dquots.
1065 STATIC int
1066 xfs_qm_quotacheck_dqadjust(
1067 struct xfs_inode *ip,
1068 xfs_dqid_t id,
1069 uint type,
1070 xfs_qcnt_t nblks,
1071 xfs_qcnt_t rtblks)
1073 struct xfs_mount *mp = ip->i_mount;
1074 struct xfs_dquot *dqp;
1075 int error;
1077 error = xfs_qm_dqget(mp, ip, id, type,
1078 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1079 if (error) {
1081 * Shouldn't be able to turn off quotas here.
1083 ASSERT(error != -ESRCH);
1084 ASSERT(error != -ENOENT);
1085 return error;
1088 trace_xfs_dqadjust(dqp);
1091 * Adjust the inode count and the block count to reflect this inode's
1092 * resource usage.
1094 be64_add_cpu(&dqp->q_core.d_icount, 1);
1095 dqp->q_res_icount++;
1096 if (nblks) {
1097 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1098 dqp->q_res_bcount += nblks;
1100 if (rtblks) {
1101 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1102 dqp->q_res_rtbcount += rtblks;
1106 * Set default limits, adjust timers (since we changed usages)
1108 * There are no timers for the default values set in the root dquot.
1110 if (dqp->q_core.d_id) {
1111 xfs_qm_adjust_dqlimits(mp, dqp);
1112 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1115 dqp->dq_flags |= XFS_DQ_DIRTY;
1116 xfs_qm_dqput(dqp);
1117 return 0;
1120 STATIC int
1121 xfs_qm_get_rtblks(
1122 xfs_inode_t *ip,
1123 xfs_qcnt_t *O_rtblks)
1125 xfs_filblks_t rtblks; /* total rt blks */
1126 xfs_extnum_t idx; /* extent record index */
1127 xfs_ifork_t *ifp; /* inode fork pointer */
1128 xfs_extnum_t nextents; /* number of extent entries */
1129 int error;
1131 ASSERT(XFS_IS_REALTIME_INODE(ip));
1132 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1133 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1134 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1135 return error;
1137 rtblks = 0;
1138 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1139 for (idx = 0; idx < nextents; idx++)
1140 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1141 *O_rtblks = (xfs_qcnt_t)rtblks;
1142 return 0;
1146 * callback routine supplied to bulkstat(). Given an inumber, find its
1147 * dquots and update them to account for resources taken by that inode.
1149 /* ARGSUSED */
1150 STATIC int
1151 xfs_qm_dqusage_adjust(
1152 xfs_mount_t *mp, /* mount point for filesystem */
1153 xfs_ino_t ino, /* inode number to get data for */
1154 void __user *buffer, /* not used */
1155 int ubsize, /* not used */
1156 int *ubused, /* not used */
1157 int *res) /* result code value */
1159 xfs_inode_t *ip;
1160 xfs_qcnt_t nblks, rtblks = 0;
1161 int error;
1163 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1166 * rootino must have its resources accounted for, not so with the quota
1167 * inodes.
1169 if (xfs_is_quota_inode(&mp->m_sb, ino)) {
1170 *res = BULKSTAT_RV_NOTHING;
1171 return -EINVAL;
1175 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1176 * interface expects the inode to be exclusively locked because that's
1177 * the case in all other instances. It's OK that we do this because
1178 * quotacheck is done only at mount time.
1180 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1181 if (error) {
1182 *res = BULKSTAT_RV_NOTHING;
1183 return error;
1186 ASSERT(ip->i_delayed_blks == 0);
1188 if (XFS_IS_REALTIME_INODE(ip)) {
1190 * Walk thru the extent list and count the realtime blocks.
1192 error = xfs_qm_get_rtblks(ip, &rtblks);
1193 if (error)
1194 goto error0;
1197 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1200 * Add the (disk blocks and inode) resources occupied by this
1201 * inode to its dquots. We do this adjustment in the incore dquot,
1202 * and also copy the changes to its buffer.
1203 * We don't care about putting these changes in a transaction
1204 * envelope because if we crash in the middle of a 'quotacheck'
1205 * we have to start from the beginning anyway.
1206 * Once we're done, we'll log all the dquot bufs.
1208 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1209 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1211 if (XFS_IS_UQUOTA_ON(mp)) {
1212 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1213 XFS_DQ_USER, nblks, rtblks);
1214 if (error)
1215 goto error0;
1218 if (XFS_IS_GQUOTA_ON(mp)) {
1219 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1220 XFS_DQ_GROUP, nblks, rtblks);
1221 if (error)
1222 goto error0;
1225 if (XFS_IS_PQUOTA_ON(mp)) {
1226 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1227 XFS_DQ_PROJ, nblks, rtblks);
1228 if (error)
1229 goto error0;
1232 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1233 IRELE(ip);
1234 *res = BULKSTAT_RV_DIDONE;
1235 return 0;
1237 error0:
1238 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1239 IRELE(ip);
1240 *res = BULKSTAT_RV_GIVEUP;
1241 return error;
1244 STATIC int
1245 xfs_qm_flush_one(
1246 struct xfs_dquot *dqp,
1247 void *data)
1249 struct list_head *buffer_list = data;
1250 struct xfs_buf *bp = NULL;
1251 int error = 0;
1253 xfs_dqlock(dqp);
1254 if (dqp->dq_flags & XFS_DQ_FREEING)
1255 goto out_unlock;
1256 if (!XFS_DQ_IS_DIRTY(dqp))
1257 goto out_unlock;
1259 xfs_dqflock(dqp);
1260 error = xfs_qm_dqflush(dqp, &bp);
1261 if (error)
1262 goto out_unlock;
1264 xfs_buf_delwri_queue(bp, buffer_list);
1265 xfs_buf_relse(bp);
1266 out_unlock:
1267 xfs_dqunlock(dqp);
1268 return error;
1272 * Walk thru all the filesystem inodes and construct a consistent view
1273 * of the disk quota world. If the quotacheck fails, disable quotas.
1275 STATIC int
1276 xfs_qm_quotacheck(
1277 xfs_mount_t *mp)
1279 int done, count, error, error2;
1280 xfs_ino_t lastino;
1281 size_t structsz;
1282 uint flags;
1283 LIST_HEAD (buffer_list);
1284 struct xfs_inode *uip = mp->m_quotainfo->qi_uquotaip;
1285 struct xfs_inode *gip = mp->m_quotainfo->qi_gquotaip;
1286 struct xfs_inode *pip = mp->m_quotainfo->qi_pquotaip;
1288 count = INT_MAX;
1289 structsz = 1;
1290 lastino = 0;
1291 flags = 0;
1293 ASSERT(uip || gip || pip);
1294 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1296 xfs_notice(mp, "Quotacheck needed: Please wait.");
1299 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1300 * their counters to zero. We need a clean slate.
1301 * We don't log our changes till later.
1303 if (uip) {
1304 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA,
1305 &buffer_list);
1306 if (error)
1307 goto error_return;
1308 flags |= XFS_UQUOTA_CHKD;
1311 if (gip) {
1312 error = xfs_qm_dqiterate(mp, gip, XFS_QMOPT_GQUOTA,
1313 &buffer_list);
1314 if (error)
1315 goto error_return;
1316 flags |= XFS_GQUOTA_CHKD;
1319 if (pip) {
1320 error = xfs_qm_dqiterate(mp, pip, XFS_QMOPT_PQUOTA,
1321 &buffer_list);
1322 if (error)
1323 goto error_return;
1324 flags |= XFS_PQUOTA_CHKD;
1327 do {
1329 * Iterate thru all the inodes in the file system,
1330 * adjusting the corresponding dquot counters in core.
1332 error = xfs_bulkstat(mp, &lastino, &count,
1333 xfs_qm_dqusage_adjust,
1334 structsz, NULL, &done);
1335 if (error)
1336 break;
1338 } while (!done);
1341 * We've made all the changes that we need to make incore. Flush them
1342 * down to disk buffers if everything was updated successfully.
1344 if (XFS_IS_UQUOTA_ON(mp)) {
1345 error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one,
1346 &buffer_list);
1348 if (XFS_IS_GQUOTA_ON(mp)) {
1349 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one,
1350 &buffer_list);
1351 if (!error)
1352 error = error2;
1354 if (XFS_IS_PQUOTA_ON(mp)) {
1355 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one,
1356 &buffer_list);
1357 if (!error)
1358 error = error2;
1361 error2 = xfs_buf_delwri_submit(&buffer_list);
1362 if (!error)
1363 error = error2;
1366 * We can get this error if we couldn't do a dquot allocation inside
1367 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1368 * dirty dquots that might be cached, we just want to get rid of them
1369 * and turn quotaoff. The dquots won't be attached to any of the inodes
1370 * at this point (because we intentionally didn't in dqget_noattach).
1372 if (error) {
1373 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1374 goto error_return;
1378 * If one type of quotas is off, then it will lose its
1379 * quotachecked status, since we won't be doing accounting for
1380 * that type anymore.
1382 mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1383 mp->m_qflags |= flags;
1385 error_return:
1386 while (!list_empty(&buffer_list)) {
1387 struct xfs_buf *bp =
1388 list_first_entry(&buffer_list, struct xfs_buf, b_list);
1389 list_del_init(&bp->b_list);
1390 xfs_buf_relse(bp);
1393 if (error) {
1394 xfs_warn(mp,
1395 "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1396 error);
1398 * We must turn off quotas.
1400 ASSERT(mp->m_quotainfo != NULL);
1401 xfs_qm_destroy_quotainfo(mp);
1402 if (xfs_mount_reset_sbqflags(mp)) {
1403 xfs_warn(mp,
1404 "Quotacheck: Failed to reset quota flags.");
1406 } else
1407 xfs_notice(mp, "Quotacheck: Done.");
1408 return error;
1412 * This is called from xfs_mountfs to start quotas and initialize all
1413 * necessary data structures like quotainfo. This is also responsible for
1414 * running a quotacheck as necessary. We are guaranteed that the superblock
1415 * is consistently read in at this point.
1417 * If we fail here, the mount will continue with quota turned off. We don't
1418 * need to inidicate success or failure at all.
1420 void
1421 xfs_qm_mount_quotas(
1422 struct xfs_mount *mp)
1424 int error = 0;
1425 uint sbf;
1428 * If quotas on realtime volumes is not supported, we disable
1429 * quotas immediately.
1431 if (mp->m_sb.sb_rextents) {
1432 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
1433 mp->m_qflags = 0;
1434 goto write_changes;
1437 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1440 * Allocate the quotainfo structure inside the mount struct, and
1441 * create quotainode(s), and change/rev superblock if necessary.
1443 error = xfs_qm_init_quotainfo(mp);
1444 if (error) {
1446 * We must turn off quotas.
1448 ASSERT(mp->m_quotainfo == NULL);
1449 mp->m_qflags = 0;
1450 goto write_changes;
1453 * If any of the quotas are not consistent, do a quotacheck.
1455 if (XFS_QM_NEED_QUOTACHECK(mp)) {
1456 error = xfs_qm_quotacheck(mp);
1457 if (error) {
1458 /* Quotacheck failed and disabled quotas. */
1459 return;
1463 * If one type of quotas is off, then it will lose its
1464 * quotachecked status, since we won't be doing accounting for
1465 * that type anymore.
1467 if (!XFS_IS_UQUOTA_ON(mp))
1468 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
1469 if (!XFS_IS_GQUOTA_ON(mp))
1470 mp->m_qflags &= ~XFS_GQUOTA_CHKD;
1471 if (!XFS_IS_PQUOTA_ON(mp))
1472 mp->m_qflags &= ~XFS_PQUOTA_CHKD;
1474 write_changes:
1476 * We actually don't have to acquire the m_sb_lock at all.
1477 * This can only be called from mount, and that's single threaded. XXX
1479 spin_lock(&mp->m_sb_lock);
1480 sbf = mp->m_sb.sb_qflags;
1481 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
1482 spin_unlock(&mp->m_sb_lock);
1484 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
1485 if (xfs_sync_sb(mp, false)) {
1487 * We could only have been turning quotas off.
1488 * We aren't in very good shape actually because
1489 * the incore structures are convinced that quotas are
1490 * off, but the on disk superblock doesn't know that !
1492 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
1493 xfs_alert(mp, "%s: Superblock update failed!",
1494 __func__);
1498 if (error) {
1499 xfs_warn(mp, "Failed to initialize disk quotas.");
1500 return;
1505 * This is called after the superblock has been read in and we're ready to
1506 * iget the quota inodes.
1508 STATIC int
1509 xfs_qm_init_quotainos(
1510 xfs_mount_t *mp)
1512 struct xfs_inode *uip = NULL;
1513 struct xfs_inode *gip = NULL;
1514 struct xfs_inode *pip = NULL;
1515 int error;
1516 uint flags = 0;
1518 ASSERT(mp->m_quotainfo);
1521 * Get the uquota and gquota inodes
1523 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1524 if (XFS_IS_UQUOTA_ON(mp) &&
1525 mp->m_sb.sb_uquotino != NULLFSINO) {
1526 ASSERT(mp->m_sb.sb_uquotino > 0);
1527 error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1528 0, 0, &uip);
1529 if (error)
1530 return error;
1532 if (XFS_IS_GQUOTA_ON(mp) &&
1533 mp->m_sb.sb_gquotino != NULLFSINO) {
1534 ASSERT(mp->m_sb.sb_gquotino > 0);
1535 error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1536 0, 0, &gip);
1537 if (error)
1538 goto error_rele;
1540 if (XFS_IS_PQUOTA_ON(mp) &&
1541 mp->m_sb.sb_pquotino != NULLFSINO) {
1542 ASSERT(mp->m_sb.sb_pquotino > 0);
1543 error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino,
1544 0, 0, &pip);
1545 if (error)
1546 goto error_rele;
1548 } else {
1549 flags |= XFS_QMOPT_SBVERSION;
1553 * Create the three inodes, if they don't exist already. The changes
1554 * made above will get added to a transaction and logged in one of
1555 * the qino_alloc calls below. If the device is readonly,
1556 * temporarily switch to read-write to do this.
1558 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1559 error = xfs_qm_qino_alloc(mp, &uip,
1560 flags | XFS_QMOPT_UQUOTA);
1561 if (error)
1562 goto error_rele;
1564 flags &= ~XFS_QMOPT_SBVERSION;
1566 if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
1567 error = xfs_qm_qino_alloc(mp, &gip,
1568 flags | XFS_QMOPT_GQUOTA);
1569 if (error)
1570 goto error_rele;
1572 flags &= ~XFS_QMOPT_SBVERSION;
1574 if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) {
1575 error = xfs_qm_qino_alloc(mp, &pip,
1576 flags | XFS_QMOPT_PQUOTA);
1577 if (error)
1578 goto error_rele;
1581 mp->m_quotainfo->qi_uquotaip = uip;
1582 mp->m_quotainfo->qi_gquotaip = gip;
1583 mp->m_quotainfo->qi_pquotaip = pip;
1585 return 0;
1587 error_rele:
1588 if (uip)
1589 IRELE(uip);
1590 if (gip)
1591 IRELE(gip);
1592 if (pip)
1593 IRELE(pip);
1594 return error;
1597 STATIC void
1598 xfs_qm_dqfree_one(
1599 struct xfs_dquot *dqp)
1601 struct xfs_mount *mp = dqp->q_mount;
1602 struct xfs_quotainfo *qi = mp->m_quotainfo;
1604 mutex_lock(&qi->qi_tree_lock);
1605 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
1606 be32_to_cpu(dqp->q_core.d_id));
1608 qi->qi_dquots--;
1609 mutex_unlock(&qi->qi_tree_lock);
1611 xfs_qm_dqdestroy(dqp);
1614 /* --------------- utility functions for vnodeops ---------------- */
1618 * Given an inode, a uid, gid and prid make sure that we have
1619 * allocated relevant dquot(s) on disk, and that we won't exceed inode
1620 * quotas by creating this file.
1621 * This also attaches dquot(s) to the given inode after locking it,
1622 * and returns the dquots corresponding to the uid and/or gid.
1624 * in : inode (unlocked)
1625 * out : udquot, gdquot with references taken and unlocked
1628 xfs_qm_vop_dqalloc(
1629 struct xfs_inode *ip,
1630 xfs_dqid_t uid,
1631 xfs_dqid_t gid,
1632 prid_t prid,
1633 uint flags,
1634 struct xfs_dquot **O_udqpp,
1635 struct xfs_dquot **O_gdqpp,
1636 struct xfs_dquot **O_pdqpp)
1638 struct xfs_mount *mp = ip->i_mount;
1639 struct xfs_dquot *uq = NULL;
1640 struct xfs_dquot *gq = NULL;
1641 struct xfs_dquot *pq = NULL;
1642 int error;
1643 uint lockflags;
1645 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1646 return 0;
1648 lockflags = XFS_ILOCK_EXCL;
1649 xfs_ilock(ip, lockflags);
1651 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1652 gid = ip->i_d.di_gid;
1655 * Attach the dquot(s) to this inode, doing a dquot allocation
1656 * if necessary. The dquot(s) will not be locked.
1658 if (XFS_NOT_DQATTACHED(mp, ip)) {
1659 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
1660 if (error) {
1661 xfs_iunlock(ip, lockflags);
1662 return error;
1666 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1667 if (ip->i_d.di_uid != uid) {
1669 * What we need is the dquot that has this uid, and
1670 * if we send the inode to dqget, the uid of the inode
1671 * takes priority over what's sent in the uid argument.
1672 * We must unlock inode here before calling dqget if
1673 * we're not sending the inode, because otherwise
1674 * we'll deadlock by doing trans_reserve while
1675 * holding ilock.
1677 xfs_iunlock(ip, lockflags);
1678 error = xfs_qm_dqget(mp, NULL, uid,
1679 XFS_DQ_USER,
1680 XFS_QMOPT_DQALLOC |
1681 XFS_QMOPT_DOWARN,
1682 &uq);
1683 if (error) {
1684 ASSERT(error != -ENOENT);
1685 return error;
1688 * Get the ilock in the right order.
1690 xfs_dqunlock(uq);
1691 lockflags = XFS_ILOCK_SHARED;
1692 xfs_ilock(ip, lockflags);
1693 } else {
1695 * Take an extra reference, because we'll return
1696 * this to caller
1698 ASSERT(ip->i_udquot);
1699 uq = xfs_qm_dqhold(ip->i_udquot);
1702 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1703 if (ip->i_d.di_gid != gid) {
1704 xfs_iunlock(ip, lockflags);
1705 error = xfs_qm_dqget(mp, NULL, gid,
1706 XFS_DQ_GROUP,
1707 XFS_QMOPT_DQALLOC |
1708 XFS_QMOPT_DOWARN,
1709 &gq);
1710 if (error) {
1711 ASSERT(error != -ENOENT);
1712 goto error_rele;
1714 xfs_dqunlock(gq);
1715 lockflags = XFS_ILOCK_SHARED;
1716 xfs_ilock(ip, lockflags);
1717 } else {
1718 ASSERT(ip->i_gdquot);
1719 gq = xfs_qm_dqhold(ip->i_gdquot);
1722 if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1723 if (xfs_get_projid(ip) != prid) {
1724 xfs_iunlock(ip, lockflags);
1725 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
1726 XFS_DQ_PROJ,
1727 XFS_QMOPT_DQALLOC |
1728 XFS_QMOPT_DOWARN,
1729 &pq);
1730 if (error) {
1731 ASSERT(error != -ENOENT);
1732 goto error_rele;
1734 xfs_dqunlock(pq);
1735 lockflags = XFS_ILOCK_SHARED;
1736 xfs_ilock(ip, lockflags);
1737 } else {
1738 ASSERT(ip->i_pdquot);
1739 pq = xfs_qm_dqhold(ip->i_pdquot);
1742 if (uq)
1743 trace_xfs_dquot_dqalloc(ip);
1745 xfs_iunlock(ip, lockflags);
1746 if (O_udqpp)
1747 *O_udqpp = uq;
1748 else
1749 xfs_qm_dqrele(uq);
1750 if (O_gdqpp)
1751 *O_gdqpp = gq;
1752 else
1753 xfs_qm_dqrele(gq);
1754 if (O_pdqpp)
1755 *O_pdqpp = pq;
1756 else
1757 xfs_qm_dqrele(pq);
1758 return 0;
1760 error_rele:
1761 xfs_qm_dqrele(gq);
1762 xfs_qm_dqrele(uq);
1763 return error;
1767 * Actually transfer ownership, and do dquot modifications.
1768 * These were already reserved.
1770 xfs_dquot_t *
1771 xfs_qm_vop_chown(
1772 xfs_trans_t *tp,
1773 xfs_inode_t *ip,
1774 xfs_dquot_t **IO_olddq,
1775 xfs_dquot_t *newdq)
1777 xfs_dquot_t *prevdq;
1778 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
1779 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1782 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1783 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1785 /* old dquot */
1786 prevdq = *IO_olddq;
1787 ASSERT(prevdq);
1788 ASSERT(prevdq != newdq);
1790 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
1791 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1793 /* the sparkling new dquot */
1794 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
1795 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1798 * Take an extra reference, because the inode is going to keep
1799 * this dquot pointer even after the trans_commit.
1801 *IO_olddq = xfs_qm_dqhold(newdq);
1803 return prevdq;
1807 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1810 xfs_qm_vop_chown_reserve(
1811 struct xfs_trans *tp,
1812 struct xfs_inode *ip,
1813 struct xfs_dquot *udqp,
1814 struct xfs_dquot *gdqp,
1815 struct xfs_dquot *pdqp,
1816 uint flags)
1818 struct xfs_mount *mp = ip->i_mount;
1819 uint delblks, blkflags, prjflags = 0;
1820 struct xfs_dquot *udq_unres = NULL;
1821 struct xfs_dquot *gdq_unres = NULL;
1822 struct xfs_dquot *pdq_unres = NULL;
1823 struct xfs_dquot *udq_delblks = NULL;
1824 struct xfs_dquot *gdq_delblks = NULL;
1825 struct xfs_dquot *pdq_delblks = NULL;
1826 int error;
1829 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1830 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1832 delblks = ip->i_delayed_blks;
1833 blkflags = XFS_IS_REALTIME_INODE(ip) ?
1834 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
1836 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
1837 ip->i_d.di_uid != be32_to_cpu(udqp->q_core.d_id)) {
1838 udq_delblks = udqp;
1840 * If there are delayed allocation blocks, then we have to
1841 * unreserve those from the old dquot, and add them to the
1842 * new dquot.
1844 if (delblks) {
1845 ASSERT(ip->i_udquot);
1846 udq_unres = ip->i_udquot;
1849 if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp &&
1850 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id)) {
1851 gdq_delblks = gdqp;
1852 if (delblks) {
1853 ASSERT(ip->i_gdquot);
1854 gdq_unres = ip->i_gdquot;
1858 if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
1859 xfs_get_projid(ip) != be32_to_cpu(pdqp->q_core.d_id)) {
1860 prjflags = XFS_QMOPT_ENOSPC;
1861 pdq_delblks = pdqp;
1862 if (delblks) {
1863 ASSERT(ip->i_pdquot);
1864 pdq_unres = ip->i_pdquot;
1868 error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
1869 udq_delblks, gdq_delblks, pdq_delblks,
1870 ip->i_d.di_nblocks, 1,
1871 flags | blkflags | prjflags);
1872 if (error)
1873 return error;
1876 * Do the delayed blks reservations/unreservations now. Since, these
1877 * are done without the help of a transaction, if a reservation fails
1878 * its previous reservations won't be automatically undone by trans
1879 * code. So, we have to do it manually here.
1881 if (delblks) {
1883 * Do the reservations first. Unreservation can't fail.
1885 ASSERT(udq_delblks || gdq_delblks || pdq_delblks);
1886 ASSERT(udq_unres || gdq_unres || pdq_unres);
1887 error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
1888 udq_delblks, gdq_delblks, pdq_delblks,
1889 (xfs_qcnt_t)delblks, 0,
1890 flags | blkflags | prjflags);
1891 if (error)
1892 return error;
1893 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
1894 udq_unres, gdq_unres, pdq_unres,
1895 -((xfs_qcnt_t)delblks), 0, blkflags);
1898 return 0;
1902 xfs_qm_vop_rename_dqattach(
1903 struct xfs_inode **i_tab)
1905 struct xfs_mount *mp = i_tab[0]->i_mount;
1906 int i;
1908 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1909 return 0;
1911 for (i = 0; (i < 4 && i_tab[i]); i++) {
1912 struct xfs_inode *ip = i_tab[i];
1913 int error;
1916 * Watch out for duplicate entries in the table.
1918 if (i == 0 || ip != i_tab[i-1]) {
1919 if (XFS_NOT_DQATTACHED(mp, ip)) {
1920 error = xfs_qm_dqattach(ip, 0);
1921 if (error)
1922 return error;
1926 return 0;
1929 void
1930 xfs_qm_vop_create_dqattach(
1931 struct xfs_trans *tp,
1932 struct xfs_inode *ip,
1933 struct xfs_dquot *udqp,
1934 struct xfs_dquot *gdqp,
1935 struct xfs_dquot *pdqp)
1937 struct xfs_mount *mp = tp->t_mountp;
1939 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1940 return;
1942 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1943 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1945 if (udqp && XFS_IS_UQUOTA_ON(mp)) {
1946 ASSERT(ip->i_udquot == NULL);
1947 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
1949 ip->i_udquot = xfs_qm_dqhold(udqp);
1950 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
1952 if (gdqp && XFS_IS_GQUOTA_ON(mp)) {
1953 ASSERT(ip->i_gdquot == NULL);
1954 ASSERT(ip->i_d.di_gid == be32_to_cpu(gdqp->q_core.d_id));
1955 ip->i_gdquot = xfs_qm_dqhold(gdqp);
1956 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
1958 if (pdqp && XFS_IS_PQUOTA_ON(mp)) {
1959 ASSERT(ip->i_pdquot == NULL);
1960 ASSERT(xfs_get_projid(ip) == be32_to_cpu(pdqp->q_core.d_id));
1962 ip->i_pdquot = xfs_qm_dqhold(pdqp);
1963 xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1);