hyperv: Add processing of MTU reduced by the host
[linux/fpc-iii.git] / fs / xfs / xfs_qm.c
blobd68f23021af3587cbaed9a3ad90d71343bceee7e
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_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_inode.h"
29 #include "xfs_ialloc.h"
30 #include "xfs_itable.h"
31 #include "xfs_quota.h"
32 #include "xfs_error.h"
33 #include "xfs_bmap.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_trans.h"
36 #include "xfs_trans_space.h"
37 #include "xfs_qm.h"
38 #include "xfs_trace.h"
39 #include "xfs_icache.h"
40 #include "xfs_cksum.h"
41 #include "xfs_dinode.h"
44 * The global quota manager. There is only one of these for the entire
45 * system, _not_ one per file system. XQM keeps track of the overall
46 * quota functionality, including maintaining the freelist and hash
47 * tables of dquots.
49 STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
50 STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
53 STATIC void xfs_qm_dqfree_one(struct xfs_dquot *dqp);
55 * We use the batch lookup interface to iterate over the dquots as it
56 * currently is the only interface into the radix tree code that allows
57 * fuzzy lookups instead of exact matches. Holding the lock over multiple
58 * operations is fine as all callers are used either during mount/umount
59 * or quotaoff.
61 #define XFS_DQ_LOOKUP_BATCH 32
63 STATIC int
64 xfs_qm_dquot_walk(
65 struct xfs_mount *mp,
66 int type,
67 int (*execute)(struct xfs_dquot *dqp, void *data),
68 void *data)
70 struct xfs_quotainfo *qi = mp->m_quotainfo;
71 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
72 uint32_t next_index;
73 int last_error = 0;
74 int skipped;
75 int nr_found;
77 restart:
78 skipped = 0;
79 next_index = 0;
80 nr_found = 0;
82 while (1) {
83 struct xfs_dquot *batch[XFS_DQ_LOOKUP_BATCH];
84 int error = 0;
85 int i;
87 mutex_lock(&qi->qi_tree_lock);
88 nr_found = radix_tree_gang_lookup(tree, (void **)batch,
89 next_index, XFS_DQ_LOOKUP_BATCH);
90 if (!nr_found) {
91 mutex_unlock(&qi->qi_tree_lock);
92 break;
95 for (i = 0; i < nr_found; i++) {
96 struct xfs_dquot *dqp = batch[i];
98 next_index = be32_to_cpu(dqp->q_core.d_id) + 1;
100 error = execute(batch[i], data);
101 if (error == -EAGAIN) {
102 skipped++;
103 continue;
105 if (error && last_error != -EFSCORRUPTED)
106 last_error = error;
109 mutex_unlock(&qi->qi_tree_lock);
111 /* bail out if the filesystem is corrupted. */
112 if (last_error == -EFSCORRUPTED) {
113 skipped = 0;
114 break;
118 if (skipped) {
119 delay(1);
120 goto restart;
123 return last_error;
128 * Purge a dquot from all tracking data structures and free it.
130 STATIC int
131 xfs_qm_dqpurge(
132 struct xfs_dquot *dqp,
133 void *data)
135 struct xfs_mount *mp = dqp->q_mount;
136 struct xfs_quotainfo *qi = mp->m_quotainfo;
138 xfs_dqlock(dqp);
139 if ((dqp->dq_flags & XFS_DQ_FREEING) || dqp->q_nrefs != 0) {
140 xfs_dqunlock(dqp);
141 return -EAGAIN;
144 dqp->dq_flags |= XFS_DQ_FREEING;
146 xfs_dqflock(dqp);
149 * If we are turning this type of quotas off, we don't care
150 * about the dirty metadata sitting in this dquot. OTOH, if
151 * we're unmounting, we do care, so we flush it and wait.
153 if (XFS_DQ_IS_DIRTY(dqp)) {
154 struct xfs_buf *bp = NULL;
155 int error;
158 * We don't care about getting disk errors here. We need
159 * to purge this dquot anyway, so we go ahead regardless.
161 error = xfs_qm_dqflush(dqp, &bp);
162 if (error) {
163 xfs_warn(mp, "%s: dquot %p flush failed",
164 __func__, dqp);
165 } else {
166 error = xfs_bwrite(bp);
167 xfs_buf_relse(bp);
169 xfs_dqflock(dqp);
172 ASSERT(atomic_read(&dqp->q_pincount) == 0);
173 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
174 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
176 xfs_dqfunlock(dqp);
177 xfs_dqunlock(dqp);
179 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
180 be32_to_cpu(dqp->q_core.d_id));
181 qi->qi_dquots--;
184 * We move dquots to the freelist as soon as their reference count
185 * hits zero, so it really should be on the freelist here.
187 ASSERT(!list_empty(&dqp->q_lru));
188 list_lru_del(&qi->qi_lru, &dqp->q_lru);
189 XFS_STATS_DEC(xs_qm_dquot_unused);
191 xfs_qm_dqdestroy(dqp);
192 return 0;
196 * Purge the dquot cache.
198 void
199 xfs_qm_dqpurge_all(
200 struct xfs_mount *mp,
201 uint flags)
203 if (flags & XFS_QMOPT_UQUOTA)
204 xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_dqpurge, NULL);
205 if (flags & XFS_QMOPT_GQUOTA)
206 xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_dqpurge, NULL);
207 if (flags & XFS_QMOPT_PQUOTA)
208 xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_dqpurge, NULL);
212 * Just destroy the quotainfo structure.
214 void
215 xfs_qm_unmount(
216 struct xfs_mount *mp)
218 if (mp->m_quotainfo) {
219 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
220 xfs_qm_destroy_quotainfo(mp);
225 * Called from the vfsops layer.
227 void
228 xfs_qm_unmount_quotas(
229 xfs_mount_t *mp)
232 * Release the dquots that root inode, et al might be holding,
233 * before we flush quotas and blow away the quotainfo structure.
235 ASSERT(mp->m_rootip);
236 xfs_qm_dqdetach(mp->m_rootip);
237 if (mp->m_rbmip)
238 xfs_qm_dqdetach(mp->m_rbmip);
239 if (mp->m_rsumip)
240 xfs_qm_dqdetach(mp->m_rsumip);
243 * Release the quota inodes.
245 if (mp->m_quotainfo) {
246 if (mp->m_quotainfo->qi_uquotaip) {
247 IRELE(mp->m_quotainfo->qi_uquotaip);
248 mp->m_quotainfo->qi_uquotaip = NULL;
250 if (mp->m_quotainfo->qi_gquotaip) {
251 IRELE(mp->m_quotainfo->qi_gquotaip);
252 mp->m_quotainfo->qi_gquotaip = NULL;
254 if (mp->m_quotainfo->qi_pquotaip) {
255 IRELE(mp->m_quotainfo->qi_pquotaip);
256 mp->m_quotainfo->qi_pquotaip = NULL;
261 STATIC int
262 xfs_qm_dqattach_one(
263 xfs_inode_t *ip,
264 xfs_dqid_t id,
265 uint type,
266 uint doalloc,
267 xfs_dquot_t **IO_idqpp)
269 xfs_dquot_t *dqp;
270 int error;
272 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
273 error = 0;
276 * See if we already have it in the inode itself. IO_idqpp is &i_udquot
277 * or &i_gdquot. This made the code look weird, but made the logic a lot
278 * simpler.
280 dqp = *IO_idqpp;
281 if (dqp) {
282 trace_xfs_dqattach_found(dqp);
283 return 0;
287 * Find the dquot from somewhere. This bumps the reference count of
288 * dquot and returns it locked. This can return ENOENT if dquot didn't
289 * exist on disk and we didn't ask it to allocate; ESRCH if quotas got
290 * turned off suddenly.
292 error = xfs_qm_dqget(ip->i_mount, ip, id, type,
293 doalloc | XFS_QMOPT_DOWARN, &dqp);
294 if (error)
295 return error;
297 trace_xfs_dqattach_get(dqp);
300 * dqget may have dropped and re-acquired the ilock, but it guarantees
301 * that the dquot returned is the one that should go in the inode.
303 *IO_idqpp = dqp;
304 xfs_dqunlock(dqp);
305 return 0;
308 static bool
309 xfs_qm_need_dqattach(
310 struct xfs_inode *ip)
312 struct xfs_mount *mp = ip->i_mount;
314 if (!XFS_IS_QUOTA_RUNNING(mp))
315 return false;
316 if (!XFS_IS_QUOTA_ON(mp))
317 return false;
318 if (!XFS_NOT_DQATTACHED(mp, ip))
319 return false;
320 if (xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
321 return false;
322 return true;
326 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
327 * into account.
328 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
329 * Inode may get unlocked and relocked in here, and the caller must deal with
330 * the consequences.
333 xfs_qm_dqattach_locked(
334 xfs_inode_t *ip,
335 uint flags)
337 xfs_mount_t *mp = ip->i_mount;
338 int error = 0;
340 if (!xfs_qm_need_dqattach(ip))
341 return 0;
343 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
345 if (XFS_IS_UQUOTA_ON(mp) && !ip->i_udquot) {
346 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
347 flags & XFS_QMOPT_DQALLOC,
348 &ip->i_udquot);
349 if (error)
350 goto done;
351 ASSERT(ip->i_udquot);
354 if (XFS_IS_GQUOTA_ON(mp) && !ip->i_gdquot) {
355 error = xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
356 flags & XFS_QMOPT_DQALLOC,
357 &ip->i_gdquot);
358 if (error)
359 goto done;
360 ASSERT(ip->i_gdquot);
363 if (XFS_IS_PQUOTA_ON(mp) && !ip->i_pdquot) {
364 error = xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
365 flags & XFS_QMOPT_DQALLOC,
366 &ip->i_pdquot);
367 if (error)
368 goto done;
369 ASSERT(ip->i_pdquot);
372 done:
374 * Don't worry about the dquots that we may have attached before any
375 * error - they'll get detached later if it has not already been done.
377 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
378 return error;
382 xfs_qm_dqattach(
383 struct xfs_inode *ip,
384 uint flags)
386 int error;
388 if (!xfs_qm_need_dqattach(ip))
389 return 0;
391 xfs_ilock(ip, XFS_ILOCK_EXCL);
392 error = xfs_qm_dqattach_locked(ip, flags);
393 xfs_iunlock(ip, XFS_ILOCK_EXCL);
395 return error;
399 * Release dquots (and their references) if any.
400 * The inode should be locked EXCL except when this's called by
401 * xfs_ireclaim.
403 void
404 xfs_qm_dqdetach(
405 xfs_inode_t *ip)
407 if (!(ip->i_udquot || ip->i_gdquot || ip->i_pdquot))
408 return;
410 trace_xfs_dquot_dqdetach(ip);
412 ASSERT(!xfs_is_quota_inode(&ip->i_mount->m_sb, ip->i_ino));
413 if (ip->i_udquot) {
414 xfs_qm_dqrele(ip->i_udquot);
415 ip->i_udquot = NULL;
417 if (ip->i_gdquot) {
418 xfs_qm_dqrele(ip->i_gdquot);
419 ip->i_gdquot = NULL;
421 if (ip->i_pdquot) {
422 xfs_qm_dqrele(ip->i_pdquot);
423 ip->i_pdquot = NULL;
427 struct xfs_qm_isolate {
428 struct list_head buffers;
429 struct list_head dispose;
432 static enum lru_status
433 xfs_qm_dquot_isolate(
434 struct list_head *item,
435 spinlock_t *lru_lock,
436 void *arg)
437 __releases(lru_lock) __acquires(lru_lock)
439 struct xfs_dquot *dqp = container_of(item,
440 struct xfs_dquot, q_lru);
441 struct xfs_qm_isolate *isol = arg;
443 if (!xfs_dqlock_nowait(dqp))
444 goto out_miss_busy;
447 * This dquot has acquired a reference in the meantime remove it from
448 * the freelist and try again.
450 if (dqp->q_nrefs) {
451 xfs_dqunlock(dqp);
452 XFS_STATS_INC(xs_qm_dqwants);
454 trace_xfs_dqreclaim_want(dqp);
455 list_del_init(&dqp->q_lru);
456 XFS_STATS_DEC(xs_qm_dquot_unused);
457 return LRU_REMOVED;
461 * If the dquot is dirty, flush it. If it's already being flushed, just
462 * skip it so there is time for the IO to complete before we try to
463 * reclaim it again on the next LRU pass.
465 if (!xfs_dqflock_nowait(dqp)) {
466 xfs_dqunlock(dqp);
467 goto out_miss_busy;
470 if (XFS_DQ_IS_DIRTY(dqp)) {
471 struct xfs_buf *bp = NULL;
472 int error;
474 trace_xfs_dqreclaim_dirty(dqp);
476 /* we have to drop the LRU lock to flush the dquot */
477 spin_unlock(lru_lock);
479 error = xfs_qm_dqflush(dqp, &bp);
480 if (error) {
481 xfs_warn(dqp->q_mount, "%s: dquot %p flush failed",
482 __func__, dqp);
483 goto out_unlock_dirty;
486 xfs_buf_delwri_queue(bp, &isol->buffers);
487 xfs_buf_relse(bp);
488 goto out_unlock_dirty;
490 xfs_dqfunlock(dqp);
493 * Prevent lookups now that we are past the point of no return.
495 dqp->dq_flags |= XFS_DQ_FREEING;
496 xfs_dqunlock(dqp);
498 ASSERT(dqp->q_nrefs == 0);
499 list_move_tail(&dqp->q_lru, &isol->dispose);
500 XFS_STATS_DEC(xs_qm_dquot_unused);
501 trace_xfs_dqreclaim_done(dqp);
502 XFS_STATS_INC(xs_qm_dqreclaims);
503 return LRU_REMOVED;
505 out_miss_busy:
506 trace_xfs_dqreclaim_busy(dqp);
507 XFS_STATS_INC(xs_qm_dqreclaim_misses);
508 return LRU_SKIP;
510 out_unlock_dirty:
511 trace_xfs_dqreclaim_busy(dqp);
512 XFS_STATS_INC(xs_qm_dqreclaim_misses);
513 xfs_dqunlock(dqp);
514 spin_lock(lru_lock);
515 return LRU_RETRY;
518 static unsigned long
519 xfs_qm_shrink_scan(
520 struct shrinker *shrink,
521 struct shrink_control *sc)
523 struct xfs_quotainfo *qi = container_of(shrink,
524 struct xfs_quotainfo, qi_shrinker);
525 struct xfs_qm_isolate isol;
526 unsigned long freed;
527 int error;
528 unsigned long nr_to_scan = sc->nr_to_scan;
530 if ((sc->gfp_mask & (__GFP_FS|__GFP_WAIT)) != (__GFP_FS|__GFP_WAIT))
531 return 0;
533 INIT_LIST_HEAD(&isol.buffers);
534 INIT_LIST_HEAD(&isol.dispose);
536 freed = list_lru_walk_node(&qi->qi_lru, sc->nid, xfs_qm_dquot_isolate, &isol,
537 &nr_to_scan);
539 error = xfs_buf_delwri_submit(&isol.buffers);
540 if (error)
541 xfs_warn(NULL, "%s: dquot reclaim failed", __func__);
543 while (!list_empty(&isol.dispose)) {
544 struct xfs_dquot *dqp;
546 dqp = list_first_entry(&isol.dispose, struct xfs_dquot, q_lru);
547 list_del_init(&dqp->q_lru);
548 xfs_qm_dqfree_one(dqp);
551 return freed;
554 static unsigned long
555 xfs_qm_shrink_count(
556 struct shrinker *shrink,
557 struct shrink_control *sc)
559 struct xfs_quotainfo *qi = container_of(shrink,
560 struct xfs_quotainfo, qi_shrinker);
562 return list_lru_count_node(&qi->qi_lru, sc->nid);
566 * This initializes all the quota information that's kept in the
567 * mount structure
569 STATIC int
570 xfs_qm_init_quotainfo(
571 xfs_mount_t *mp)
573 xfs_quotainfo_t *qinf;
574 int error;
575 xfs_dquot_t *dqp;
577 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
579 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
581 error = list_lru_init(&qinf->qi_lru);
582 if (error)
583 goto out_free_qinf;
586 * See if quotainodes are setup, and if not, allocate them,
587 * and change the superblock accordingly.
589 error = xfs_qm_init_quotainos(mp);
590 if (error)
591 goto out_free_lru;
593 INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
594 INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
595 INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS);
596 mutex_init(&qinf->qi_tree_lock);
598 /* mutex used to serialize quotaoffs */
599 mutex_init(&qinf->qi_quotaofflock);
601 /* Precalc some constants */
602 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
603 qinf->qi_dqperchunk = xfs_calc_dquots_per_chunk(qinf->qi_dqchunklen);
605 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
608 * We try to get the limits from the superuser's limits fields.
609 * This is quite hacky, but it is standard quota practice.
611 * We look at the USR dquot with id == 0 first, but if user quotas
612 * are not enabled we goto the GRP dquot with id == 0.
613 * We don't really care to keep separate default limits for user
614 * and group quotas, at least not at this point.
616 * Since we may not have done a quotacheck by this point, just read
617 * the dquot without attaching it to any hashtables or lists.
619 error = xfs_qm_dqread(mp, 0,
620 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
621 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
622 XFS_DQ_PROJ),
623 XFS_QMOPT_DOWARN, &dqp);
624 if (!error) {
625 xfs_disk_dquot_t *ddqp = &dqp->q_core;
628 * The warnings and timers set the grace period given to
629 * a user or group before he or she can not perform any
630 * more writing. If it is zero, a default is used.
632 qinf->qi_btimelimit = ddqp->d_btimer ?
633 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
634 qinf->qi_itimelimit = ddqp->d_itimer ?
635 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
636 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
637 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
638 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
639 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
640 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
641 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
642 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
643 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
644 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
645 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
646 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
647 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
648 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
649 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
651 xfs_qm_dqdestroy(dqp);
652 } else {
653 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
654 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
655 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
656 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
657 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
658 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
661 qinf->qi_shrinker.count_objects = xfs_qm_shrink_count;
662 qinf->qi_shrinker.scan_objects = xfs_qm_shrink_scan;
663 qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
664 qinf->qi_shrinker.flags = SHRINKER_NUMA_AWARE;
665 register_shrinker(&qinf->qi_shrinker);
666 return 0;
668 out_free_lru:
669 list_lru_destroy(&qinf->qi_lru);
670 out_free_qinf:
671 kmem_free(qinf);
672 mp->m_quotainfo = NULL;
673 return error;
678 * Gets called when unmounting a filesystem or when all quotas get
679 * turned off.
680 * This purges the quota inodes, destroys locks and frees itself.
682 void
683 xfs_qm_destroy_quotainfo(
684 xfs_mount_t *mp)
686 xfs_quotainfo_t *qi;
688 qi = mp->m_quotainfo;
689 ASSERT(qi != NULL);
691 unregister_shrinker(&qi->qi_shrinker);
692 list_lru_destroy(&qi->qi_lru);
694 if (qi->qi_uquotaip) {
695 IRELE(qi->qi_uquotaip);
696 qi->qi_uquotaip = NULL; /* paranoia */
698 if (qi->qi_gquotaip) {
699 IRELE(qi->qi_gquotaip);
700 qi->qi_gquotaip = NULL;
702 if (qi->qi_pquotaip) {
703 IRELE(qi->qi_pquotaip);
704 qi->qi_pquotaip = NULL;
706 mutex_destroy(&qi->qi_quotaofflock);
707 kmem_free(qi);
708 mp->m_quotainfo = NULL;
712 * Create an inode and return with a reference already taken, but unlocked
713 * This is how we create quota inodes
715 STATIC int
716 xfs_qm_qino_alloc(
717 xfs_mount_t *mp,
718 xfs_inode_t **ip,
719 __int64_t sbfields,
720 uint flags)
722 xfs_trans_t *tp;
723 int error;
724 int committed;
726 *ip = NULL;
728 * With superblock that doesn't have separate pquotino, we
729 * share an inode between gquota and pquota. If the on-disk
730 * superblock has GQUOTA and the filesystem is now mounted
731 * with PQUOTA, just use sb_gquotino for sb_pquotino and
732 * vice-versa.
734 if (!xfs_sb_version_has_pquotino(&mp->m_sb) &&
735 (flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) {
736 xfs_ino_t ino = NULLFSINO;
738 if ((flags & XFS_QMOPT_PQUOTA) &&
739 (mp->m_sb.sb_gquotino != NULLFSINO)) {
740 ino = mp->m_sb.sb_gquotino;
741 ASSERT(mp->m_sb.sb_pquotino == NULLFSINO);
742 } else if ((flags & XFS_QMOPT_GQUOTA) &&
743 (mp->m_sb.sb_pquotino != NULLFSINO)) {
744 ino = mp->m_sb.sb_pquotino;
745 ASSERT(mp->m_sb.sb_gquotino == NULLFSINO);
747 if (ino != NULLFSINO) {
748 error = xfs_iget(mp, NULL, ino, 0, 0, ip);
749 if (error)
750 return error;
751 mp->m_sb.sb_gquotino = NULLFSINO;
752 mp->m_sb.sb_pquotino = NULLFSINO;
756 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
757 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_create,
758 XFS_QM_QINOCREATE_SPACE_RES(mp), 0);
759 if (error) {
760 xfs_trans_cancel(tp, 0);
761 return error;
764 if (!*ip) {
765 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip,
766 &committed);
767 if (error) {
768 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
769 XFS_TRANS_ABORT);
770 return error;
775 * Make the changes in the superblock, and log those too.
776 * sbfields arg may contain fields other than *QUOTINO;
777 * VERSIONNUM for example.
779 spin_lock(&mp->m_sb_lock);
780 if (flags & XFS_QMOPT_SBVERSION) {
781 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
782 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
783 XFS_SB_GQUOTINO | XFS_SB_PQUOTINO | XFS_SB_QFLAGS)) ==
784 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
785 XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
786 XFS_SB_QFLAGS));
788 xfs_sb_version_addquota(&mp->m_sb);
789 mp->m_sb.sb_uquotino = NULLFSINO;
790 mp->m_sb.sb_gquotino = NULLFSINO;
791 mp->m_sb.sb_pquotino = NULLFSINO;
793 /* qflags will get updated fully _after_ quotacheck */
794 mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT;
796 if (flags & XFS_QMOPT_UQUOTA)
797 mp->m_sb.sb_uquotino = (*ip)->i_ino;
798 else if (flags & XFS_QMOPT_GQUOTA)
799 mp->m_sb.sb_gquotino = (*ip)->i_ino;
800 else
801 mp->m_sb.sb_pquotino = (*ip)->i_ino;
802 spin_unlock(&mp->m_sb_lock);
803 xfs_mod_sb(tp, sbfields);
805 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
806 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
807 return error;
809 return 0;
813 STATIC void
814 xfs_qm_reset_dqcounts(
815 xfs_mount_t *mp,
816 xfs_buf_t *bp,
817 xfs_dqid_t id,
818 uint type)
820 struct xfs_dqblk *dqb;
821 int j;
823 trace_xfs_reset_dqcounts(bp, _RET_IP_);
826 * Reset all counters and timers. They'll be
827 * started afresh by xfs_qm_quotacheck.
829 #ifdef DEBUG
830 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
831 do_div(j, sizeof(xfs_dqblk_t));
832 ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
833 #endif
834 dqb = bp->b_addr;
835 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
836 struct xfs_disk_dquot *ddq;
838 ddq = (struct xfs_disk_dquot *)&dqb[j];
841 * Do a sanity check, and if needed, repair the dqblk. Don't
842 * output any warnings because it's perfectly possible to
843 * find uninitialised dquot blks. See comment in xfs_dqcheck.
845 xfs_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
846 "xfs_quotacheck");
847 ddq->d_bcount = 0;
848 ddq->d_icount = 0;
849 ddq->d_rtbcount = 0;
850 ddq->d_btimer = 0;
851 ddq->d_itimer = 0;
852 ddq->d_rtbtimer = 0;
853 ddq->d_bwarns = 0;
854 ddq->d_iwarns = 0;
855 ddq->d_rtbwarns = 0;
857 if (xfs_sb_version_hascrc(&mp->m_sb)) {
858 xfs_update_cksum((char *)&dqb[j],
859 sizeof(struct xfs_dqblk),
860 XFS_DQUOT_CRC_OFF);
865 STATIC int
866 xfs_qm_dqiter_bufs(
867 struct xfs_mount *mp,
868 xfs_dqid_t firstid,
869 xfs_fsblock_t bno,
870 xfs_filblks_t blkcnt,
871 uint flags,
872 struct list_head *buffer_list)
874 struct xfs_buf *bp;
875 int error;
876 int type;
878 ASSERT(blkcnt > 0);
879 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
880 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
881 error = 0;
884 * Blkcnt arg can be a very big number, and might even be
885 * larger than the log itself. So, we have to break it up into
886 * manageable-sized transactions.
887 * Note that we don't start a permanent transaction here; we might
888 * not be able to get a log reservation for the whole thing up front,
889 * and we don't really care to either, because we just discard
890 * everything if we were to crash in the middle of this loop.
892 while (blkcnt--) {
893 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
894 XFS_FSB_TO_DADDR(mp, bno),
895 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
896 &xfs_dquot_buf_ops);
899 * CRC and validation errors will return a EFSCORRUPTED here. If
900 * this occurs, re-read without CRC validation so that we can
901 * repair the damage via xfs_qm_reset_dqcounts(). This process
902 * will leave a trace in the log indicating corruption has
903 * been detected.
905 if (error == -EFSCORRUPTED) {
906 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
907 XFS_FSB_TO_DADDR(mp, bno),
908 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
909 NULL);
912 if (error)
913 break;
916 * A corrupt buffer might not have a verifier attached, so
917 * make sure we have the correct one attached before writeback
918 * occurs.
920 bp->b_ops = &xfs_dquot_buf_ops;
921 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
922 xfs_buf_delwri_queue(bp, buffer_list);
923 xfs_buf_relse(bp);
925 /* goto the next block. */
926 bno++;
927 firstid += mp->m_quotainfo->qi_dqperchunk;
930 return error;
934 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
935 * caller supplied function for every chunk of dquots that we find.
937 STATIC int
938 xfs_qm_dqiterate(
939 struct xfs_mount *mp,
940 struct xfs_inode *qip,
941 uint flags,
942 struct list_head *buffer_list)
944 struct xfs_bmbt_irec *map;
945 int i, nmaps; /* number of map entries */
946 int error; /* return value */
947 xfs_fileoff_t lblkno;
948 xfs_filblks_t maxlblkcnt;
949 xfs_dqid_t firstid;
950 xfs_fsblock_t rablkno;
951 xfs_filblks_t rablkcnt;
953 error = 0;
955 * This looks racy, but we can't keep an inode lock across a
956 * trans_reserve. But, this gets called during quotacheck, and that
957 * happens only at mount time which is single threaded.
959 if (qip->i_d.di_nblocks == 0)
960 return 0;
962 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
964 lblkno = 0;
965 maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
966 do {
967 uint lock_mode;
969 nmaps = XFS_DQITER_MAP_SIZE;
971 * We aren't changing the inode itself. Just changing
972 * some of its data. No new blocks are added here, and
973 * the inode is never added to the transaction.
975 lock_mode = xfs_ilock_data_map_shared(qip);
976 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
977 map, &nmaps, 0);
978 xfs_iunlock(qip, lock_mode);
979 if (error)
980 break;
982 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
983 for (i = 0; i < nmaps; i++) {
984 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
985 ASSERT(map[i].br_blockcount);
988 lblkno += map[i].br_blockcount;
990 if (map[i].br_startblock == HOLESTARTBLOCK)
991 continue;
993 firstid = (xfs_dqid_t) map[i].br_startoff *
994 mp->m_quotainfo->qi_dqperchunk;
996 * Do a read-ahead on the next extent.
998 if ((i+1 < nmaps) &&
999 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1000 rablkcnt = map[i+1].br_blockcount;
1001 rablkno = map[i+1].br_startblock;
1002 while (rablkcnt--) {
1003 xfs_buf_readahead(mp->m_ddev_targp,
1004 XFS_FSB_TO_DADDR(mp, rablkno),
1005 mp->m_quotainfo->qi_dqchunklen,
1006 &xfs_dquot_buf_ops);
1007 rablkno++;
1011 * Iterate thru all the blks in the extent and
1012 * reset the counters of all the dquots inside them.
1014 error = xfs_qm_dqiter_bufs(mp, firstid,
1015 map[i].br_startblock,
1016 map[i].br_blockcount,
1017 flags, buffer_list);
1018 if (error)
1019 goto out;
1021 } while (nmaps > 0);
1023 out:
1024 kmem_free(map);
1025 return error;
1029 * Called by dqusage_adjust in doing a quotacheck.
1031 * Given the inode, and a dquot id this updates both the incore dqout as well
1032 * as the buffer copy. This is so that once the quotacheck is done, we can
1033 * just log all the buffers, as opposed to logging numerous updates to
1034 * individual dquots.
1036 STATIC int
1037 xfs_qm_quotacheck_dqadjust(
1038 struct xfs_inode *ip,
1039 xfs_dqid_t id,
1040 uint type,
1041 xfs_qcnt_t nblks,
1042 xfs_qcnt_t rtblks)
1044 struct xfs_mount *mp = ip->i_mount;
1045 struct xfs_dquot *dqp;
1046 int error;
1048 error = xfs_qm_dqget(mp, ip, id, type,
1049 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1050 if (error) {
1052 * Shouldn't be able to turn off quotas here.
1054 ASSERT(error != -ESRCH);
1055 ASSERT(error != -ENOENT);
1056 return error;
1059 trace_xfs_dqadjust(dqp);
1062 * Adjust the inode count and the block count to reflect this inode's
1063 * resource usage.
1065 be64_add_cpu(&dqp->q_core.d_icount, 1);
1066 dqp->q_res_icount++;
1067 if (nblks) {
1068 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1069 dqp->q_res_bcount += nblks;
1071 if (rtblks) {
1072 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1073 dqp->q_res_rtbcount += rtblks;
1077 * Set default limits, adjust timers (since we changed usages)
1079 * There are no timers for the default values set in the root dquot.
1081 if (dqp->q_core.d_id) {
1082 xfs_qm_adjust_dqlimits(mp, dqp);
1083 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1086 dqp->dq_flags |= XFS_DQ_DIRTY;
1087 xfs_qm_dqput(dqp);
1088 return 0;
1091 STATIC int
1092 xfs_qm_get_rtblks(
1093 xfs_inode_t *ip,
1094 xfs_qcnt_t *O_rtblks)
1096 xfs_filblks_t rtblks; /* total rt blks */
1097 xfs_extnum_t idx; /* extent record index */
1098 xfs_ifork_t *ifp; /* inode fork pointer */
1099 xfs_extnum_t nextents; /* number of extent entries */
1100 int error;
1102 ASSERT(XFS_IS_REALTIME_INODE(ip));
1103 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1104 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1105 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1106 return error;
1108 rtblks = 0;
1109 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1110 for (idx = 0; idx < nextents; idx++)
1111 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1112 *O_rtblks = (xfs_qcnt_t)rtblks;
1113 return 0;
1117 * callback routine supplied to bulkstat(). Given an inumber, find its
1118 * dquots and update them to account for resources taken by that inode.
1120 /* ARGSUSED */
1121 STATIC int
1122 xfs_qm_dqusage_adjust(
1123 xfs_mount_t *mp, /* mount point for filesystem */
1124 xfs_ino_t ino, /* inode number to get data for */
1125 void __user *buffer, /* not used */
1126 int ubsize, /* not used */
1127 int *ubused, /* not used */
1128 int *res) /* result code value */
1130 xfs_inode_t *ip;
1131 xfs_qcnt_t nblks, rtblks = 0;
1132 int error;
1134 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1137 * rootino must have its resources accounted for, not so with the quota
1138 * inodes.
1140 if (xfs_is_quota_inode(&mp->m_sb, ino)) {
1141 *res = BULKSTAT_RV_NOTHING;
1142 return -EINVAL;
1146 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1147 * interface expects the inode to be exclusively locked because that's
1148 * the case in all other instances. It's OK that we do this because
1149 * quotacheck is done only at mount time.
1151 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1152 if (error) {
1153 *res = BULKSTAT_RV_NOTHING;
1154 return error;
1157 ASSERT(ip->i_delayed_blks == 0);
1159 if (XFS_IS_REALTIME_INODE(ip)) {
1161 * Walk thru the extent list and count the realtime blocks.
1163 error = xfs_qm_get_rtblks(ip, &rtblks);
1164 if (error)
1165 goto error0;
1168 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1171 * Add the (disk blocks and inode) resources occupied by this
1172 * inode to its dquots. We do this adjustment in the incore dquot,
1173 * and also copy the changes to its buffer.
1174 * We don't care about putting these changes in a transaction
1175 * envelope because if we crash in the middle of a 'quotacheck'
1176 * we have to start from the beginning anyway.
1177 * Once we're done, we'll log all the dquot bufs.
1179 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1180 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1182 if (XFS_IS_UQUOTA_ON(mp)) {
1183 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1184 XFS_DQ_USER, nblks, rtblks);
1185 if (error)
1186 goto error0;
1189 if (XFS_IS_GQUOTA_ON(mp)) {
1190 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1191 XFS_DQ_GROUP, nblks, rtblks);
1192 if (error)
1193 goto error0;
1196 if (XFS_IS_PQUOTA_ON(mp)) {
1197 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1198 XFS_DQ_PROJ, nblks, rtblks);
1199 if (error)
1200 goto error0;
1203 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1204 IRELE(ip);
1205 *res = BULKSTAT_RV_DIDONE;
1206 return 0;
1208 error0:
1209 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1210 IRELE(ip);
1211 *res = BULKSTAT_RV_GIVEUP;
1212 return error;
1215 STATIC int
1216 xfs_qm_flush_one(
1217 struct xfs_dquot *dqp,
1218 void *data)
1220 struct list_head *buffer_list = data;
1221 struct xfs_buf *bp = NULL;
1222 int error = 0;
1224 xfs_dqlock(dqp);
1225 if (dqp->dq_flags & XFS_DQ_FREEING)
1226 goto out_unlock;
1227 if (!XFS_DQ_IS_DIRTY(dqp))
1228 goto out_unlock;
1230 xfs_dqflock(dqp);
1231 error = xfs_qm_dqflush(dqp, &bp);
1232 if (error)
1233 goto out_unlock;
1235 xfs_buf_delwri_queue(bp, buffer_list);
1236 xfs_buf_relse(bp);
1237 out_unlock:
1238 xfs_dqunlock(dqp);
1239 return error;
1243 * Walk thru all the filesystem inodes and construct a consistent view
1244 * of the disk quota world. If the quotacheck fails, disable quotas.
1246 STATIC int
1247 xfs_qm_quotacheck(
1248 xfs_mount_t *mp)
1250 int done, count, error, error2;
1251 xfs_ino_t lastino;
1252 size_t structsz;
1253 uint flags;
1254 LIST_HEAD (buffer_list);
1255 struct xfs_inode *uip = mp->m_quotainfo->qi_uquotaip;
1256 struct xfs_inode *gip = mp->m_quotainfo->qi_gquotaip;
1257 struct xfs_inode *pip = mp->m_quotainfo->qi_pquotaip;
1259 count = INT_MAX;
1260 structsz = 1;
1261 lastino = 0;
1262 flags = 0;
1264 ASSERT(uip || gip || pip);
1265 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1267 xfs_notice(mp, "Quotacheck needed: Please wait.");
1270 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1271 * their counters to zero. We need a clean slate.
1272 * We don't log our changes till later.
1274 if (uip) {
1275 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA,
1276 &buffer_list);
1277 if (error)
1278 goto error_return;
1279 flags |= XFS_UQUOTA_CHKD;
1282 if (gip) {
1283 error = xfs_qm_dqiterate(mp, gip, XFS_QMOPT_GQUOTA,
1284 &buffer_list);
1285 if (error)
1286 goto error_return;
1287 flags |= XFS_GQUOTA_CHKD;
1290 if (pip) {
1291 error = xfs_qm_dqiterate(mp, pip, XFS_QMOPT_PQUOTA,
1292 &buffer_list);
1293 if (error)
1294 goto error_return;
1295 flags |= XFS_PQUOTA_CHKD;
1298 do {
1300 * Iterate thru all the inodes in the file system,
1301 * adjusting the corresponding dquot counters in core.
1303 error = xfs_bulkstat(mp, &lastino, &count,
1304 xfs_qm_dqusage_adjust,
1305 structsz, NULL, &done);
1306 if (error)
1307 break;
1309 } while (!done);
1312 * We've made all the changes that we need to make incore. Flush them
1313 * down to disk buffers if everything was updated successfully.
1315 if (XFS_IS_UQUOTA_ON(mp)) {
1316 error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one,
1317 &buffer_list);
1319 if (XFS_IS_GQUOTA_ON(mp)) {
1320 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one,
1321 &buffer_list);
1322 if (!error)
1323 error = error2;
1325 if (XFS_IS_PQUOTA_ON(mp)) {
1326 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one,
1327 &buffer_list);
1328 if (!error)
1329 error = error2;
1332 error2 = xfs_buf_delwri_submit(&buffer_list);
1333 if (!error)
1334 error = error2;
1337 * We can get this error if we couldn't do a dquot allocation inside
1338 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1339 * dirty dquots that might be cached, we just want to get rid of them
1340 * and turn quotaoff. The dquots won't be attached to any of the inodes
1341 * at this point (because we intentionally didn't in dqget_noattach).
1343 if (error) {
1344 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1345 goto error_return;
1349 * If one type of quotas is off, then it will lose its
1350 * quotachecked status, since we won't be doing accounting for
1351 * that type anymore.
1353 mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1354 mp->m_qflags |= flags;
1356 error_return:
1357 while (!list_empty(&buffer_list)) {
1358 struct xfs_buf *bp =
1359 list_first_entry(&buffer_list, struct xfs_buf, b_list);
1360 list_del_init(&bp->b_list);
1361 xfs_buf_relse(bp);
1364 if (error) {
1365 xfs_warn(mp,
1366 "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1367 error);
1369 * We must turn off quotas.
1371 ASSERT(mp->m_quotainfo != NULL);
1372 xfs_qm_destroy_quotainfo(mp);
1373 if (xfs_mount_reset_sbqflags(mp)) {
1374 xfs_warn(mp,
1375 "Quotacheck: Failed to reset quota flags.");
1377 } else
1378 xfs_notice(mp, "Quotacheck: Done.");
1379 return error;
1383 * This is called from xfs_mountfs to start quotas and initialize all
1384 * necessary data structures like quotainfo. This is also responsible for
1385 * running a quotacheck as necessary. We are guaranteed that the superblock
1386 * is consistently read in at this point.
1388 * If we fail here, the mount will continue with quota turned off. We don't
1389 * need to inidicate success or failure at all.
1391 void
1392 xfs_qm_mount_quotas(
1393 struct xfs_mount *mp)
1395 int error = 0;
1396 uint sbf;
1399 * If quotas on realtime volumes is not supported, we disable
1400 * quotas immediately.
1402 if (mp->m_sb.sb_rextents) {
1403 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
1404 mp->m_qflags = 0;
1405 goto write_changes;
1408 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1411 * Allocate the quotainfo structure inside the mount struct, and
1412 * create quotainode(s), and change/rev superblock if necessary.
1414 error = xfs_qm_init_quotainfo(mp);
1415 if (error) {
1417 * We must turn off quotas.
1419 ASSERT(mp->m_quotainfo == NULL);
1420 mp->m_qflags = 0;
1421 goto write_changes;
1424 * If any of the quotas are not consistent, do a quotacheck.
1426 if (XFS_QM_NEED_QUOTACHECK(mp)) {
1427 error = xfs_qm_quotacheck(mp);
1428 if (error) {
1429 /* Quotacheck failed and disabled quotas. */
1430 return;
1434 * If one type of quotas is off, then it will lose its
1435 * quotachecked status, since we won't be doing accounting for
1436 * that type anymore.
1438 if (!XFS_IS_UQUOTA_ON(mp))
1439 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
1440 if (!XFS_IS_GQUOTA_ON(mp))
1441 mp->m_qflags &= ~XFS_GQUOTA_CHKD;
1442 if (!XFS_IS_PQUOTA_ON(mp))
1443 mp->m_qflags &= ~XFS_PQUOTA_CHKD;
1445 write_changes:
1447 * We actually don't have to acquire the m_sb_lock at all.
1448 * This can only be called from mount, and that's single threaded. XXX
1450 spin_lock(&mp->m_sb_lock);
1451 sbf = mp->m_sb.sb_qflags;
1452 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
1453 spin_unlock(&mp->m_sb_lock);
1455 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
1456 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
1458 * We could only have been turning quotas off.
1459 * We aren't in very good shape actually because
1460 * the incore structures are convinced that quotas are
1461 * off, but the on disk superblock doesn't know that !
1463 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
1464 xfs_alert(mp, "%s: Superblock update failed!",
1465 __func__);
1469 if (error) {
1470 xfs_warn(mp, "Failed to initialize disk quotas.");
1471 return;
1476 * This is called after the superblock has been read in and we're ready to
1477 * iget the quota inodes.
1479 STATIC int
1480 xfs_qm_init_quotainos(
1481 xfs_mount_t *mp)
1483 struct xfs_inode *uip = NULL;
1484 struct xfs_inode *gip = NULL;
1485 struct xfs_inode *pip = NULL;
1486 int error;
1487 __int64_t sbflags = 0;
1488 uint flags = 0;
1490 ASSERT(mp->m_quotainfo);
1493 * Get the uquota and gquota inodes
1495 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1496 if (XFS_IS_UQUOTA_ON(mp) &&
1497 mp->m_sb.sb_uquotino != NULLFSINO) {
1498 ASSERT(mp->m_sb.sb_uquotino > 0);
1499 error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1500 0, 0, &uip);
1501 if (error)
1502 return error;
1504 if (XFS_IS_GQUOTA_ON(mp) &&
1505 mp->m_sb.sb_gquotino != NULLFSINO) {
1506 ASSERT(mp->m_sb.sb_gquotino > 0);
1507 error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1508 0, 0, &gip);
1509 if (error)
1510 goto error_rele;
1512 if (XFS_IS_PQUOTA_ON(mp) &&
1513 mp->m_sb.sb_pquotino != NULLFSINO) {
1514 ASSERT(mp->m_sb.sb_pquotino > 0);
1515 error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino,
1516 0, 0, &pip);
1517 if (error)
1518 goto error_rele;
1520 } else {
1521 flags |= XFS_QMOPT_SBVERSION;
1522 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1523 XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1524 XFS_SB_QFLAGS);
1528 * Create the three inodes, if they don't exist already. The changes
1529 * made above will get added to a transaction and logged in one of
1530 * the qino_alloc calls below. If the device is readonly,
1531 * temporarily switch to read-write to do this.
1533 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1534 error = xfs_qm_qino_alloc(mp, &uip,
1535 sbflags | XFS_SB_UQUOTINO,
1536 flags | XFS_QMOPT_UQUOTA);
1537 if (error)
1538 goto error_rele;
1540 flags &= ~XFS_QMOPT_SBVERSION;
1542 if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
1543 error = xfs_qm_qino_alloc(mp, &gip,
1544 sbflags | XFS_SB_GQUOTINO,
1545 flags | XFS_QMOPT_GQUOTA);
1546 if (error)
1547 goto error_rele;
1549 flags &= ~XFS_QMOPT_SBVERSION;
1551 if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) {
1552 error = xfs_qm_qino_alloc(mp, &pip,
1553 sbflags | XFS_SB_PQUOTINO,
1554 flags | XFS_QMOPT_PQUOTA);
1555 if (error)
1556 goto error_rele;
1559 mp->m_quotainfo->qi_uquotaip = uip;
1560 mp->m_quotainfo->qi_gquotaip = gip;
1561 mp->m_quotainfo->qi_pquotaip = pip;
1563 return 0;
1565 error_rele:
1566 if (uip)
1567 IRELE(uip);
1568 if (gip)
1569 IRELE(gip);
1570 if (pip)
1571 IRELE(pip);
1572 return error;
1575 STATIC void
1576 xfs_qm_dqfree_one(
1577 struct xfs_dquot *dqp)
1579 struct xfs_mount *mp = dqp->q_mount;
1580 struct xfs_quotainfo *qi = mp->m_quotainfo;
1582 mutex_lock(&qi->qi_tree_lock);
1583 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
1584 be32_to_cpu(dqp->q_core.d_id));
1586 qi->qi_dquots--;
1587 mutex_unlock(&qi->qi_tree_lock);
1589 xfs_qm_dqdestroy(dqp);
1593 * Start a transaction and write the incore superblock changes to
1594 * disk. flags parameter indicates which fields have changed.
1597 xfs_qm_write_sb_changes(
1598 xfs_mount_t *mp,
1599 __int64_t flags)
1601 xfs_trans_t *tp;
1602 int error;
1604 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1605 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
1606 if (error) {
1607 xfs_trans_cancel(tp, 0);
1608 return error;
1611 xfs_mod_sb(tp, flags);
1612 error = xfs_trans_commit(tp, 0);
1614 return error;
1618 /* --------------- utility functions for vnodeops ---------------- */
1622 * Given an inode, a uid, gid and prid make sure that we have
1623 * allocated relevant dquot(s) on disk, and that we won't exceed inode
1624 * quotas by creating this file.
1625 * This also attaches dquot(s) to the given inode after locking it,
1626 * and returns the dquots corresponding to the uid and/or gid.
1628 * in : inode (unlocked)
1629 * out : udquot, gdquot with references taken and unlocked
1632 xfs_qm_vop_dqalloc(
1633 struct xfs_inode *ip,
1634 xfs_dqid_t uid,
1635 xfs_dqid_t gid,
1636 prid_t prid,
1637 uint flags,
1638 struct xfs_dquot **O_udqpp,
1639 struct xfs_dquot **O_gdqpp,
1640 struct xfs_dquot **O_pdqpp)
1642 struct xfs_mount *mp = ip->i_mount;
1643 struct xfs_dquot *uq = NULL;
1644 struct xfs_dquot *gq = NULL;
1645 struct xfs_dquot *pq = NULL;
1646 int error;
1647 uint lockflags;
1649 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1650 return 0;
1652 lockflags = XFS_ILOCK_EXCL;
1653 xfs_ilock(ip, lockflags);
1655 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1656 gid = ip->i_d.di_gid;
1659 * Attach the dquot(s) to this inode, doing a dquot allocation
1660 * if necessary. The dquot(s) will not be locked.
1662 if (XFS_NOT_DQATTACHED(mp, ip)) {
1663 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
1664 if (error) {
1665 xfs_iunlock(ip, lockflags);
1666 return error;
1670 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1671 if (ip->i_d.di_uid != uid) {
1673 * What we need is the dquot that has this uid, and
1674 * if we send the inode to dqget, the uid of the inode
1675 * takes priority over what's sent in the uid argument.
1676 * We must unlock inode here before calling dqget if
1677 * we're not sending the inode, because otherwise
1678 * we'll deadlock by doing trans_reserve while
1679 * holding ilock.
1681 xfs_iunlock(ip, lockflags);
1682 error = xfs_qm_dqget(mp, NULL, uid,
1683 XFS_DQ_USER,
1684 XFS_QMOPT_DQALLOC |
1685 XFS_QMOPT_DOWARN,
1686 &uq);
1687 if (error) {
1688 ASSERT(error != -ENOENT);
1689 return error;
1692 * Get the ilock in the right order.
1694 xfs_dqunlock(uq);
1695 lockflags = XFS_ILOCK_SHARED;
1696 xfs_ilock(ip, lockflags);
1697 } else {
1699 * Take an extra reference, because we'll return
1700 * this to caller
1702 ASSERT(ip->i_udquot);
1703 uq = xfs_qm_dqhold(ip->i_udquot);
1706 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1707 if (ip->i_d.di_gid != gid) {
1708 xfs_iunlock(ip, lockflags);
1709 error = xfs_qm_dqget(mp, NULL, gid,
1710 XFS_DQ_GROUP,
1711 XFS_QMOPT_DQALLOC |
1712 XFS_QMOPT_DOWARN,
1713 &gq);
1714 if (error) {
1715 ASSERT(error != -ENOENT);
1716 goto error_rele;
1718 xfs_dqunlock(gq);
1719 lockflags = XFS_ILOCK_SHARED;
1720 xfs_ilock(ip, lockflags);
1721 } else {
1722 ASSERT(ip->i_gdquot);
1723 gq = xfs_qm_dqhold(ip->i_gdquot);
1726 if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1727 if (xfs_get_projid(ip) != prid) {
1728 xfs_iunlock(ip, lockflags);
1729 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
1730 XFS_DQ_PROJ,
1731 XFS_QMOPT_DQALLOC |
1732 XFS_QMOPT_DOWARN,
1733 &pq);
1734 if (error) {
1735 ASSERT(error != -ENOENT);
1736 goto error_rele;
1738 xfs_dqunlock(pq);
1739 lockflags = XFS_ILOCK_SHARED;
1740 xfs_ilock(ip, lockflags);
1741 } else {
1742 ASSERT(ip->i_pdquot);
1743 pq = xfs_qm_dqhold(ip->i_pdquot);
1746 if (uq)
1747 trace_xfs_dquot_dqalloc(ip);
1749 xfs_iunlock(ip, lockflags);
1750 if (O_udqpp)
1751 *O_udqpp = uq;
1752 else if (uq)
1753 xfs_qm_dqrele(uq);
1754 if (O_gdqpp)
1755 *O_gdqpp = gq;
1756 else if (gq)
1757 xfs_qm_dqrele(gq);
1758 if (O_pdqpp)
1759 *O_pdqpp = pq;
1760 else if (pq)
1761 xfs_qm_dqrele(pq);
1762 return 0;
1764 error_rele:
1765 if (gq)
1766 xfs_qm_dqrele(gq);
1767 if (uq)
1768 xfs_qm_dqrele(uq);
1769 return error;
1773 * Actually transfer ownership, and do dquot modifications.
1774 * These were already reserved.
1776 xfs_dquot_t *
1777 xfs_qm_vop_chown(
1778 xfs_trans_t *tp,
1779 xfs_inode_t *ip,
1780 xfs_dquot_t **IO_olddq,
1781 xfs_dquot_t *newdq)
1783 xfs_dquot_t *prevdq;
1784 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
1785 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1788 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1789 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1791 /* old dquot */
1792 prevdq = *IO_olddq;
1793 ASSERT(prevdq);
1794 ASSERT(prevdq != newdq);
1796 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
1797 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1799 /* the sparkling new dquot */
1800 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
1801 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1804 * Take an extra reference, because the inode is going to keep
1805 * this dquot pointer even after the trans_commit.
1807 *IO_olddq = xfs_qm_dqhold(newdq);
1809 return prevdq;
1813 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1816 xfs_qm_vop_chown_reserve(
1817 struct xfs_trans *tp,
1818 struct xfs_inode *ip,
1819 struct xfs_dquot *udqp,
1820 struct xfs_dquot *gdqp,
1821 struct xfs_dquot *pdqp,
1822 uint flags)
1824 struct xfs_mount *mp = ip->i_mount;
1825 uint delblks, blkflags, prjflags = 0;
1826 struct xfs_dquot *udq_unres = NULL;
1827 struct xfs_dquot *gdq_unres = NULL;
1828 struct xfs_dquot *pdq_unres = NULL;
1829 struct xfs_dquot *udq_delblks = NULL;
1830 struct xfs_dquot *gdq_delblks = NULL;
1831 struct xfs_dquot *pdq_delblks = NULL;
1832 int error;
1835 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1836 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1838 delblks = ip->i_delayed_blks;
1839 blkflags = XFS_IS_REALTIME_INODE(ip) ?
1840 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
1842 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
1843 ip->i_d.di_uid != be32_to_cpu(udqp->q_core.d_id)) {
1844 udq_delblks = udqp;
1846 * If there are delayed allocation blocks, then we have to
1847 * unreserve those from the old dquot, and add them to the
1848 * new dquot.
1850 if (delblks) {
1851 ASSERT(ip->i_udquot);
1852 udq_unres = ip->i_udquot;
1855 if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp &&
1856 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id)) {
1857 gdq_delblks = gdqp;
1858 if (delblks) {
1859 ASSERT(ip->i_gdquot);
1860 gdq_unres = ip->i_gdquot;
1864 if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
1865 xfs_get_projid(ip) != be32_to_cpu(pdqp->q_core.d_id)) {
1866 prjflags = XFS_QMOPT_ENOSPC;
1867 pdq_delblks = pdqp;
1868 if (delblks) {
1869 ASSERT(ip->i_pdquot);
1870 pdq_unres = ip->i_pdquot;
1874 error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
1875 udq_delblks, gdq_delblks, pdq_delblks,
1876 ip->i_d.di_nblocks, 1,
1877 flags | blkflags | prjflags);
1878 if (error)
1879 return error;
1882 * Do the delayed blks reservations/unreservations now. Since, these
1883 * are done without the help of a transaction, if a reservation fails
1884 * its previous reservations won't be automatically undone by trans
1885 * code. So, we have to do it manually here.
1887 if (delblks) {
1889 * Do the reservations first. Unreservation can't fail.
1891 ASSERT(udq_delblks || gdq_delblks || pdq_delblks);
1892 ASSERT(udq_unres || gdq_unres || pdq_unres);
1893 error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
1894 udq_delblks, gdq_delblks, pdq_delblks,
1895 (xfs_qcnt_t)delblks, 0,
1896 flags | blkflags | prjflags);
1897 if (error)
1898 return error;
1899 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
1900 udq_unres, gdq_unres, pdq_unres,
1901 -((xfs_qcnt_t)delblks), 0, blkflags);
1904 return 0;
1908 xfs_qm_vop_rename_dqattach(
1909 struct xfs_inode **i_tab)
1911 struct xfs_mount *mp = i_tab[0]->i_mount;
1912 int i;
1914 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1915 return 0;
1917 for (i = 0; (i < 4 && i_tab[i]); i++) {
1918 struct xfs_inode *ip = i_tab[i];
1919 int error;
1922 * Watch out for duplicate entries in the table.
1924 if (i == 0 || ip != i_tab[i-1]) {
1925 if (XFS_NOT_DQATTACHED(mp, ip)) {
1926 error = xfs_qm_dqattach(ip, 0);
1927 if (error)
1928 return error;
1932 return 0;
1935 void
1936 xfs_qm_vop_create_dqattach(
1937 struct xfs_trans *tp,
1938 struct xfs_inode *ip,
1939 struct xfs_dquot *udqp,
1940 struct xfs_dquot *gdqp,
1941 struct xfs_dquot *pdqp)
1943 struct xfs_mount *mp = tp->t_mountp;
1945 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1946 return;
1948 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1949 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1951 if (udqp && XFS_IS_UQUOTA_ON(mp)) {
1952 ASSERT(ip->i_udquot == NULL);
1953 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
1955 ip->i_udquot = xfs_qm_dqhold(udqp);
1956 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
1958 if (gdqp && XFS_IS_GQUOTA_ON(mp)) {
1959 ASSERT(ip->i_gdquot == NULL);
1960 ASSERT(ip->i_d.di_gid == be32_to_cpu(gdqp->q_core.d_id));
1961 ip->i_gdquot = xfs_qm_dqhold(gdqp);
1962 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
1964 if (pdqp && XFS_IS_PQUOTA_ON(mp)) {
1965 ASSERT(ip->i_pdquot == NULL);
1966 ASSERT(xfs_get_projid(ip) == be32_to_cpu(pdqp->q_core.d_id));
1968 ip->i_pdquot = xfs_qm_dqhold(pdqp);
1969 xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1);