of/overlay: Do not generate duplicate nodes
[linux-2.6/btrfs-unstable.git] / fs / xfs / xfs_qm.c
blob79fb19dd9c833de5fefc8b44a332f6e47250e13f
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(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 spinlock_t *lru_lock,
434 void *arg)
435 __releases(lru_lock) __acquires(lru_lock)
437 struct xfs_dquot *dqp = container_of(item,
438 struct xfs_dquot, q_lru);
439 struct xfs_qm_isolate *isol = arg;
441 if (!xfs_dqlock_nowait(dqp))
442 goto out_miss_busy;
445 * This dquot has acquired a reference in the meantime remove it from
446 * the freelist and try again.
448 if (dqp->q_nrefs) {
449 xfs_dqunlock(dqp);
450 XFS_STATS_INC(xs_qm_dqwants);
452 trace_xfs_dqreclaim_want(dqp);
453 list_del_init(&dqp->q_lru);
454 XFS_STATS_DEC(xs_qm_dquot_unused);
455 return LRU_REMOVED;
459 * If the dquot is dirty, flush it. If it's already being flushed, just
460 * skip it so there is time for the IO to complete before we try to
461 * reclaim it again on the next LRU pass.
463 if (!xfs_dqflock_nowait(dqp)) {
464 xfs_dqunlock(dqp);
465 goto out_miss_busy;
468 if (XFS_DQ_IS_DIRTY(dqp)) {
469 struct xfs_buf *bp = NULL;
470 int error;
472 trace_xfs_dqreclaim_dirty(dqp);
474 /* we have to drop the LRU lock to flush the dquot */
475 spin_unlock(lru_lock);
477 error = xfs_qm_dqflush(dqp, &bp);
478 if (error) {
479 xfs_warn(dqp->q_mount, "%s: dquot %p flush failed",
480 __func__, dqp);
481 goto out_unlock_dirty;
484 xfs_buf_delwri_queue(bp, &isol->buffers);
485 xfs_buf_relse(bp);
486 goto out_unlock_dirty;
488 xfs_dqfunlock(dqp);
491 * Prevent lookups now that we are past the point of no return.
493 dqp->dq_flags |= XFS_DQ_FREEING;
494 xfs_dqunlock(dqp);
496 ASSERT(dqp->q_nrefs == 0);
497 list_move_tail(&dqp->q_lru, &isol->dispose);
498 XFS_STATS_DEC(xs_qm_dquot_unused);
499 trace_xfs_dqreclaim_done(dqp);
500 XFS_STATS_INC(xs_qm_dqreclaims);
501 return LRU_REMOVED;
503 out_miss_busy:
504 trace_xfs_dqreclaim_busy(dqp);
505 XFS_STATS_INC(xs_qm_dqreclaim_misses);
506 return LRU_SKIP;
508 out_unlock_dirty:
509 trace_xfs_dqreclaim_busy(dqp);
510 XFS_STATS_INC(xs_qm_dqreclaim_misses);
511 xfs_dqunlock(dqp);
512 spin_lock(lru_lock);
513 return LRU_RETRY;
516 static unsigned long
517 xfs_qm_shrink_scan(
518 struct shrinker *shrink,
519 struct shrink_control *sc)
521 struct xfs_quotainfo *qi = container_of(shrink,
522 struct xfs_quotainfo, qi_shrinker);
523 struct xfs_qm_isolate isol;
524 unsigned long freed;
525 int error;
526 unsigned long nr_to_scan = sc->nr_to_scan;
528 if ((sc->gfp_mask & (__GFP_FS|__GFP_WAIT)) != (__GFP_FS|__GFP_WAIT))
529 return 0;
531 INIT_LIST_HEAD(&isol.buffers);
532 INIT_LIST_HEAD(&isol.dispose);
534 freed = list_lru_walk_node(&qi->qi_lru, sc->nid, xfs_qm_dquot_isolate, &isol,
535 &nr_to_scan);
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_count_node(&qi->qi_lru, sc->nid);
564 * This initializes all the quota information that's kept in the
565 * mount structure
567 STATIC int
568 xfs_qm_init_quotainfo(
569 xfs_mount_t *mp)
571 xfs_quotainfo_t *qinf;
572 int error;
573 xfs_dquot_t *dqp;
575 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
577 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
579 error = list_lru_init(&qinf->qi_lru);
580 if (error)
581 goto out_free_qinf;
584 * See if quotainodes are setup, and if not, allocate them,
585 * and change the superblock accordingly.
587 error = xfs_qm_init_quotainos(mp);
588 if (error)
589 goto out_free_lru;
591 INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
592 INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
593 INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS);
594 mutex_init(&qinf->qi_tree_lock);
596 /* mutex used to serialize quotaoffs */
597 mutex_init(&qinf->qi_quotaofflock);
599 /* Precalc some constants */
600 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
601 qinf->qi_dqperchunk = xfs_calc_dquots_per_chunk(qinf->qi_dqchunklen);
603 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
606 * We try to get the limits from the superuser's limits fields.
607 * This is quite hacky, but it is standard quota practice.
609 * We look at the USR dquot with id == 0 first, but if user quotas
610 * are not enabled we goto the GRP dquot with id == 0.
611 * We don't really care to keep separate default limits for user
612 * and group quotas, at least not at this point.
614 * Since we may not have done a quotacheck by this point, just read
615 * the dquot without attaching it to any hashtables or lists.
617 error = xfs_qm_dqread(mp, 0,
618 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
619 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
620 XFS_DQ_PROJ),
621 XFS_QMOPT_DOWARN, &dqp);
622 if (!error) {
623 xfs_disk_dquot_t *ddqp = &dqp->q_core;
626 * The warnings and timers set the grace period given to
627 * a user or group before he or she can not perform any
628 * more writing. If it is zero, a default is used.
630 qinf->qi_btimelimit = ddqp->d_btimer ?
631 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
632 qinf->qi_itimelimit = ddqp->d_itimer ?
633 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
634 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
635 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
636 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
637 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
638 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
639 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
640 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
641 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
642 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
643 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
644 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
645 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
646 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
647 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
649 xfs_qm_dqdestroy(dqp);
650 } else {
651 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
652 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
653 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
654 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
655 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
656 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
659 qinf->qi_shrinker.count_objects = xfs_qm_shrink_count;
660 qinf->qi_shrinker.scan_objects = xfs_qm_shrink_scan;
661 qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
662 qinf->qi_shrinker.flags = SHRINKER_NUMA_AWARE;
663 register_shrinker(&qinf->qi_shrinker);
664 return 0;
666 out_free_lru:
667 list_lru_destroy(&qinf->qi_lru);
668 out_free_qinf:
669 kmem_free(qinf);
670 mp->m_quotainfo = NULL;
671 return error;
676 * Gets called when unmounting a filesystem or when all quotas get
677 * turned off.
678 * This purges the quota inodes, destroys locks and frees itself.
680 void
681 xfs_qm_destroy_quotainfo(
682 xfs_mount_t *mp)
684 xfs_quotainfo_t *qi;
686 qi = mp->m_quotainfo;
687 ASSERT(qi != NULL);
689 unregister_shrinker(&qi->qi_shrinker);
690 list_lru_destroy(&qi->qi_lru);
692 if (qi->qi_uquotaip) {
693 IRELE(qi->qi_uquotaip);
694 qi->qi_uquotaip = NULL; /* paranoia */
696 if (qi->qi_gquotaip) {
697 IRELE(qi->qi_gquotaip);
698 qi->qi_gquotaip = NULL;
700 if (qi->qi_pquotaip) {
701 IRELE(qi->qi_pquotaip);
702 qi->qi_pquotaip = NULL;
704 mutex_destroy(&qi->qi_quotaofflock);
705 kmem_free(qi);
706 mp->m_quotainfo = NULL;
710 * Create an inode and return with a reference already taken, but unlocked
711 * This is how we create quota inodes
713 STATIC int
714 xfs_qm_qino_alloc(
715 xfs_mount_t *mp,
716 xfs_inode_t **ip,
717 __int64_t sbfields,
718 uint flags)
720 xfs_trans_t *tp;
721 int error;
722 int committed;
724 *ip = NULL;
726 * With superblock that doesn't have separate pquotino, we
727 * share an inode between gquota and pquota. If the on-disk
728 * superblock has GQUOTA and the filesystem is now mounted
729 * with PQUOTA, just use sb_gquotino for sb_pquotino and
730 * vice-versa.
732 if (!xfs_sb_version_has_pquotino(&mp->m_sb) &&
733 (flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) {
734 xfs_ino_t ino = NULLFSINO;
736 if ((flags & XFS_QMOPT_PQUOTA) &&
737 (mp->m_sb.sb_gquotino != NULLFSINO)) {
738 ino = mp->m_sb.sb_gquotino;
739 ASSERT(mp->m_sb.sb_pquotino == NULLFSINO);
740 } else if ((flags & XFS_QMOPT_GQUOTA) &&
741 (mp->m_sb.sb_pquotino != NULLFSINO)) {
742 ino = mp->m_sb.sb_pquotino;
743 ASSERT(mp->m_sb.sb_gquotino == NULLFSINO);
745 if (ino != NULLFSINO) {
746 error = xfs_iget(mp, NULL, ino, 0, 0, ip);
747 if (error)
748 return error;
749 mp->m_sb.sb_gquotino = NULLFSINO;
750 mp->m_sb.sb_pquotino = NULLFSINO;
754 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
755 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_create,
756 XFS_QM_QINOCREATE_SPACE_RES(mp), 0);
757 if (error) {
758 xfs_trans_cancel(tp, 0);
759 return error;
762 if (!*ip) {
763 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip,
764 &committed);
765 if (error) {
766 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
767 XFS_TRANS_ABORT);
768 return error;
773 * Make the changes in the superblock, and log those too.
774 * sbfields arg may contain fields other than *QUOTINO;
775 * VERSIONNUM for example.
777 spin_lock(&mp->m_sb_lock);
778 if (flags & XFS_QMOPT_SBVERSION) {
779 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
780 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
781 XFS_SB_GQUOTINO | XFS_SB_PQUOTINO | XFS_SB_QFLAGS)) ==
782 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
783 XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
784 XFS_SB_QFLAGS));
786 xfs_sb_version_addquota(&mp->m_sb);
787 mp->m_sb.sb_uquotino = NULLFSINO;
788 mp->m_sb.sb_gquotino = NULLFSINO;
789 mp->m_sb.sb_pquotino = NULLFSINO;
791 /* qflags will get updated fully _after_ quotacheck */
792 mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT;
794 if (flags & XFS_QMOPT_UQUOTA)
795 mp->m_sb.sb_uquotino = (*ip)->i_ino;
796 else if (flags & XFS_QMOPT_GQUOTA)
797 mp->m_sb.sb_gquotino = (*ip)->i_ino;
798 else
799 mp->m_sb.sb_pquotino = (*ip)->i_ino;
800 spin_unlock(&mp->m_sb_lock);
801 xfs_mod_sb(tp, sbfields);
803 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
804 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
805 return error;
807 return 0;
811 STATIC void
812 xfs_qm_reset_dqcounts(
813 xfs_mount_t *mp,
814 xfs_buf_t *bp,
815 xfs_dqid_t id,
816 uint type)
818 struct xfs_dqblk *dqb;
819 int j;
821 trace_xfs_reset_dqcounts(bp, _RET_IP_);
824 * Reset all counters and timers. They'll be
825 * started afresh by xfs_qm_quotacheck.
827 #ifdef DEBUG
828 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
829 do_div(j, sizeof(xfs_dqblk_t));
830 ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
831 #endif
832 dqb = bp->b_addr;
833 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
834 struct xfs_disk_dquot *ddq;
836 ddq = (struct xfs_disk_dquot *)&dqb[j];
839 * Do a sanity check, and if needed, repair the dqblk. Don't
840 * output any warnings because it's perfectly possible to
841 * find uninitialised dquot blks. See comment in xfs_dqcheck.
843 xfs_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
844 "xfs_quotacheck");
845 ddq->d_bcount = 0;
846 ddq->d_icount = 0;
847 ddq->d_rtbcount = 0;
848 ddq->d_btimer = 0;
849 ddq->d_itimer = 0;
850 ddq->d_rtbtimer = 0;
851 ddq->d_bwarns = 0;
852 ddq->d_iwarns = 0;
853 ddq->d_rtbwarns = 0;
855 if (xfs_sb_version_hascrc(&mp->m_sb)) {
856 xfs_update_cksum((char *)&dqb[j],
857 sizeof(struct xfs_dqblk),
858 XFS_DQUOT_CRC_OFF);
863 STATIC int
864 xfs_qm_dqiter_bufs(
865 struct xfs_mount *mp,
866 xfs_dqid_t firstid,
867 xfs_fsblock_t bno,
868 xfs_filblks_t blkcnt,
869 uint flags,
870 struct list_head *buffer_list)
872 struct xfs_buf *bp;
873 int error;
874 int type;
876 ASSERT(blkcnt > 0);
877 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
878 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
879 error = 0;
882 * Blkcnt arg can be a very big number, and might even be
883 * larger than the log itself. So, we have to break it up into
884 * manageable-sized transactions.
885 * Note that we don't start a permanent transaction here; we might
886 * not be able to get a log reservation for the whole thing up front,
887 * and we don't really care to either, because we just discard
888 * everything if we were to crash in the middle of this loop.
890 while (blkcnt--) {
891 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
892 XFS_FSB_TO_DADDR(mp, bno),
893 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
894 &xfs_dquot_buf_ops);
897 * CRC and validation errors will return a EFSCORRUPTED here. If
898 * this occurs, re-read without CRC validation so that we can
899 * repair the damage via xfs_qm_reset_dqcounts(). This process
900 * will leave a trace in the log indicating corruption has
901 * been detected.
903 if (error == -EFSCORRUPTED) {
904 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
905 XFS_FSB_TO_DADDR(mp, bno),
906 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
907 NULL);
910 if (error)
911 break;
914 * A corrupt buffer might not have a verifier attached, so
915 * make sure we have the correct one attached before writeback
916 * occurs.
918 bp->b_ops = &xfs_dquot_buf_ops;
919 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
920 xfs_buf_delwri_queue(bp, buffer_list);
921 xfs_buf_relse(bp);
923 /* goto the next block. */
924 bno++;
925 firstid += mp->m_quotainfo->qi_dqperchunk;
928 return error;
932 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
933 * caller supplied function for every chunk of dquots that we find.
935 STATIC int
936 xfs_qm_dqiterate(
937 struct xfs_mount *mp,
938 struct xfs_inode *qip,
939 uint flags,
940 struct list_head *buffer_list)
942 struct xfs_bmbt_irec *map;
943 int i, nmaps; /* number of map entries */
944 int error; /* return value */
945 xfs_fileoff_t lblkno;
946 xfs_filblks_t maxlblkcnt;
947 xfs_dqid_t firstid;
948 xfs_fsblock_t rablkno;
949 xfs_filblks_t rablkcnt;
951 error = 0;
953 * This looks racy, but we can't keep an inode lock across a
954 * trans_reserve. But, this gets called during quotacheck, and that
955 * happens only at mount time which is single threaded.
957 if (qip->i_d.di_nblocks == 0)
958 return 0;
960 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
962 lblkno = 0;
963 maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
964 do {
965 uint lock_mode;
967 nmaps = XFS_DQITER_MAP_SIZE;
969 * We aren't changing the inode itself. Just changing
970 * some of its data. No new blocks are added here, and
971 * the inode is never added to the transaction.
973 lock_mode = xfs_ilock_data_map_shared(qip);
974 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
975 map, &nmaps, 0);
976 xfs_iunlock(qip, lock_mode);
977 if (error)
978 break;
980 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
981 for (i = 0; i < nmaps; i++) {
982 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
983 ASSERT(map[i].br_blockcount);
986 lblkno += map[i].br_blockcount;
988 if (map[i].br_startblock == HOLESTARTBLOCK)
989 continue;
991 firstid = (xfs_dqid_t) map[i].br_startoff *
992 mp->m_quotainfo->qi_dqperchunk;
994 * Do a read-ahead on the next extent.
996 if ((i+1 < nmaps) &&
997 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
998 rablkcnt = map[i+1].br_blockcount;
999 rablkno = map[i+1].br_startblock;
1000 while (rablkcnt--) {
1001 xfs_buf_readahead(mp->m_ddev_targp,
1002 XFS_FSB_TO_DADDR(mp, rablkno),
1003 mp->m_quotainfo->qi_dqchunklen,
1004 &xfs_dquot_buf_ops);
1005 rablkno++;
1009 * Iterate thru all the blks in the extent and
1010 * reset the counters of all the dquots inside them.
1012 error = xfs_qm_dqiter_bufs(mp, firstid,
1013 map[i].br_startblock,
1014 map[i].br_blockcount,
1015 flags, buffer_list);
1016 if (error)
1017 goto out;
1019 } while (nmaps > 0);
1021 out:
1022 kmem_free(map);
1023 return error;
1027 * Called by dqusage_adjust in doing a quotacheck.
1029 * Given the inode, and a dquot id this updates both the incore dqout as well
1030 * as the buffer copy. This is so that once the quotacheck is done, we can
1031 * just log all the buffers, as opposed to logging numerous updates to
1032 * individual dquots.
1034 STATIC int
1035 xfs_qm_quotacheck_dqadjust(
1036 struct xfs_inode *ip,
1037 xfs_dqid_t id,
1038 uint type,
1039 xfs_qcnt_t nblks,
1040 xfs_qcnt_t rtblks)
1042 struct xfs_mount *mp = ip->i_mount;
1043 struct xfs_dquot *dqp;
1044 int error;
1046 error = xfs_qm_dqget(mp, ip, id, type,
1047 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1048 if (error) {
1050 * Shouldn't be able to turn off quotas here.
1052 ASSERT(error != -ESRCH);
1053 ASSERT(error != -ENOENT);
1054 return error;
1057 trace_xfs_dqadjust(dqp);
1060 * Adjust the inode count and the block count to reflect this inode's
1061 * resource usage.
1063 be64_add_cpu(&dqp->q_core.d_icount, 1);
1064 dqp->q_res_icount++;
1065 if (nblks) {
1066 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1067 dqp->q_res_bcount += nblks;
1069 if (rtblks) {
1070 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1071 dqp->q_res_rtbcount += rtblks;
1075 * Set default limits, adjust timers (since we changed usages)
1077 * There are no timers for the default values set in the root dquot.
1079 if (dqp->q_core.d_id) {
1080 xfs_qm_adjust_dqlimits(mp, dqp);
1081 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1084 dqp->dq_flags |= XFS_DQ_DIRTY;
1085 xfs_qm_dqput(dqp);
1086 return 0;
1089 STATIC int
1090 xfs_qm_get_rtblks(
1091 xfs_inode_t *ip,
1092 xfs_qcnt_t *O_rtblks)
1094 xfs_filblks_t rtblks; /* total rt blks */
1095 xfs_extnum_t idx; /* extent record index */
1096 xfs_ifork_t *ifp; /* inode fork pointer */
1097 xfs_extnum_t nextents; /* number of extent entries */
1098 int error;
1100 ASSERT(XFS_IS_REALTIME_INODE(ip));
1101 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1102 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1103 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1104 return error;
1106 rtblks = 0;
1107 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1108 for (idx = 0; idx < nextents; idx++)
1109 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1110 *O_rtblks = (xfs_qcnt_t)rtblks;
1111 return 0;
1115 * callback routine supplied to bulkstat(). Given an inumber, find its
1116 * dquots and update them to account for resources taken by that inode.
1118 /* ARGSUSED */
1119 STATIC int
1120 xfs_qm_dqusage_adjust(
1121 xfs_mount_t *mp, /* mount point for filesystem */
1122 xfs_ino_t ino, /* inode number to get data for */
1123 void __user *buffer, /* not used */
1124 int ubsize, /* not used */
1125 int *ubused, /* not used */
1126 int *res) /* result code value */
1128 xfs_inode_t *ip;
1129 xfs_qcnt_t nblks, rtblks = 0;
1130 int error;
1132 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1135 * rootino must have its resources accounted for, not so with the quota
1136 * inodes.
1138 if (xfs_is_quota_inode(&mp->m_sb, ino)) {
1139 *res = BULKSTAT_RV_NOTHING;
1140 return -EINVAL;
1144 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1145 * interface expects the inode to be exclusively locked because that's
1146 * the case in all other instances. It's OK that we do this because
1147 * quotacheck is done only at mount time.
1149 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1150 if (error) {
1151 *res = BULKSTAT_RV_NOTHING;
1152 return error;
1155 ASSERT(ip->i_delayed_blks == 0);
1157 if (XFS_IS_REALTIME_INODE(ip)) {
1159 * Walk thru the extent list and count the realtime blocks.
1161 error = xfs_qm_get_rtblks(ip, &rtblks);
1162 if (error)
1163 goto error0;
1166 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1169 * Add the (disk blocks and inode) resources occupied by this
1170 * inode to its dquots. We do this adjustment in the incore dquot,
1171 * and also copy the changes to its buffer.
1172 * We don't care about putting these changes in a transaction
1173 * envelope because if we crash in the middle of a 'quotacheck'
1174 * we have to start from the beginning anyway.
1175 * Once we're done, we'll log all the dquot bufs.
1177 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1178 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1180 if (XFS_IS_UQUOTA_ON(mp)) {
1181 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1182 XFS_DQ_USER, nblks, rtblks);
1183 if (error)
1184 goto error0;
1187 if (XFS_IS_GQUOTA_ON(mp)) {
1188 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1189 XFS_DQ_GROUP, nblks, rtblks);
1190 if (error)
1191 goto error0;
1194 if (XFS_IS_PQUOTA_ON(mp)) {
1195 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1196 XFS_DQ_PROJ, nblks, rtblks);
1197 if (error)
1198 goto error0;
1201 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1202 IRELE(ip);
1203 *res = BULKSTAT_RV_DIDONE;
1204 return 0;
1206 error0:
1207 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1208 IRELE(ip);
1209 *res = BULKSTAT_RV_GIVEUP;
1210 return error;
1213 STATIC int
1214 xfs_qm_flush_one(
1215 struct xfs_dquot *dqp,
1216 void *data)
1218 struct list_head *buffer_list = data;
1219 struct xfs_buf *bp = NULL;
1220 int error = 0;
1222 xfs_dqlock(dqp);
1223 if (dqp->dq_flags & XFS_DQ_FREEING)
1224 goto out_unlock;
1225 if (!XFS_DQ_IS_DIRTY(dqp))
1226 goto out_unlock;
1228 xfs_dqflock(dqp);
1229 error = xfs_qm_dqflush(dqp, &bp);
1230 if (error)
1231 goto out_unlock;
1233 xfs_buf_delwri_queue(bp, buffer_list);
1234 xfs_buf_relse(bp);
1235 out_unlock:
1236 xfs_dqunlock(dqp);
1237 return error;
1241 * Walk thru all the filesystem inodes and construct a consistent view
1242 * of the disk quota world. If the quotacheck fails, disable quotas.
1244 STATIC int
1245 xfs_qm_quotacheck(
1246 xfs_mount_t *mp)
1248 int done, count, error, error2;
1249 xfs_ino_t lastino;
1250 size_t structsz;
1251 uint flags;
1252 LIST_HEAD (buffer_list);
1253 struct xfs_inode *uip = mp->m_quotainfo->qi_uquotaip;
1254 struct xfs_inode *gip = mp->m_quotainfo->qi_gquotaip;
1255 struct xfs_inode *pip = mp->m_quotainfo->qi_pquotaip;
1257 count = INT_MAX;
1258 structsz = 1;
1259 lastino = 0;
1260 flags = 0;
1262 ASSERT(uip || gip || pip);
1263 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1265 xfs_notice(mp, "Quotacheck needed: Please wait.");
1268 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1269 * their counters to zero. We need a clean slate.
1270 * We don't log our changes till later.
1272 if (uip) {
1273 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA,
1274 &buffer_list);
1275 if (error)
1276 goto error_return;
1277 flags |= XFS_UQUOTA_CHKD;
1280 if (gip) {
1281 error = xfs_qm_dqiterate(mp, gip, XFS_QMOPT_GQUOTA,
1282 &buffer_list);
1283 if (error)
1284 goto error_return;
1285 flags |= XFS_GQUOTA_CHKD;
1288 if (pip) {
1289 error = xfs_qm_dqiterate(mp, pip, XFS_QMOPT_PQUOTA,
1290 &buffer_list);
1291 if (error)
1292 goto error_return;
1293 flags |= XFS_PQUOTA_CHKD;
1296 do {
1298 * Iterate thru all the inodes in the file system,
1299 * adjusting the corresponding dquot counters in core.
1301 error = xfs_bulkstat(mp, &lastino, &count,
1302 xfs_qm_dqusage_adjust,
1303 structsz, NULL, &done);
1304 if (error)
1305 break;
1307 } while (!done);
1310 * We've made all the changes that we need to make incore. Flush them
1311 * down to disk buffers if everything was updated successfully.
1313 if (XFS_IS_UQUOTA_ON(mp)) {
1314 error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one,
1315 &buffer_list);
1317 if (XFS_IS_GQUOTA_ON(mp)) {
1318 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one,
1319 &buffer_list);
1320 if (!error)
1321 error = error2;
1323 if (XFS_IS_PQUOTA_ON(mp)) {
1324 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one,
1325 &buffer_list);
1326 if (!error)
1327 error = error2;
1330 error2 = xfs_buf_delwri_submit(&buffer_list);
1331 if (!error)
1332 error = error2;
1335 * We can get this error if we couldn't do a dquot allocation inside
1336 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1337 * dirty dquots that might be cached, we just want to get rid of them
1338 * and turn quotaoff. The dquots won't be attached to any of the inodes
1339 * at this point (because we intentionally didn't in dqget_noattach).
1341 if (error) {
1342 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1343 goto error_return;
1347 * If one type of quotas is off, then it will lose its
1348 * quotachecked status, since we won't be doing accounting for
1349 * that type anymore.
1351 mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1352 mp->m_qflags |= flags;
1354 error_return:
1355 while (!list_empty(&buffer_list)) {
1356 struct xfs_buf *bp =
1357 list_first_entry(&buffer_list, struct xfs_buf, b_list);
1358 list_del_init(&bp->b_list);
1359 xfs_buf_relse(bp);
1362 if (error) {
1363 xfs_warn(mp,
1364 "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1365 error);
1367 * We must turn off quotas.
1369 ASSERT(mp->m_quotainfo != NULL);
1370 xfs_qm_destroy_quotainfo(mp);
1371 if (xfs_mount_reset_sbqflags(mp)) {
1372 xfs_warn(mp,
1373 "Quotacheck: Failed to reset quota flags.");
1375 } else
1376 xfs_notice(mp, "Quotacheck: Done.");
1377 return error;
1381 * This is called from xfs_mountfs to start quotas and initialize all
1382 * necessary data structures like quotainfo. This is also responsible for
1383 * running a quotacheck as necessary. We are guaranteed that the superblock
1384 * is consistently read in at this point.
1386 * If we fail here, the mount will continue with quota turned off. We don't
1387 * need to inidicate success or failure at all.
1389 void
1390 xfs_qm_mount_quotas(
1391 struct xfs_mount *mp)
1393 int error = 0;
1394 uint sbf;
1397 * If quotas on realtime volumes is not supported, we disable
1398 * quotas immediately.
1400 if (mp->m_sb.sb_rextents) {
1401 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
1402 mp->m_qflags = 0;
1403 goto write_changes;
1406 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1409 * Allocate the quotainfo structure inside the mount struct, and
1410 * create quotainode(s), and change/rev superblock if necessary.
1412 error = xfs_qm_init_quotainfo(mp);
1413 if (error) {
1415 * We must turn off quotas.
1417 ASSERT(mp->m_quotainfo == NULL);
1418 mp->m_qflags = 0;
1419 goto write_changes;
1422 * If any of the quotas are not consistent, do a quotacheck.
1424 if (XFS_QM_NEED_QUOTACHECK(mp)) {
1425 error = xfs_qm_quotacheck(mp);
1426 if (error) {
1427 /* Quotacheck failed and disabled quotas. */
1428 return;
1432 * If one type of quotas is off, then it will lose its
1433 * quotachecked status, since we won't be doing accounting for
1434 * that type anymore.
1436 if (!XFS_IS_UQUOTA_ON(mp))
1437 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
1438 if (!XFS_IS_GQUOTA_ON(mp))
1439 mp->m_qflags &= ~XFS_GQUOTA_CHKD;
1440 if (!XFS_IS_PQUOTA_ON(mp))
1441 mp->m_qflags &= ~XFS_PQUOTA_CHKD;
1443 write_changes:
1445 * We actually don't have to acquire the m_sb_lock at all.
1446 * This can only be called from mount, and that's single threaded. XXX
1448 spin_lock(&mp->m_sb_lock);
1449 sbf = mp->m_sb.sb_qflags;
1450 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
1451 spin_unlock(&mp->m_sb_lock);
1453 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
1454 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
1456 * We could only have been turning quotas off.
1457 * We aren't in very good shape actually because
1458 * the incore structures are convinced that quotas are
1459 * off, but the on disk superblock doesn't know that !
1461 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
1462 xfs_alert(mp, "%s: Superblock update failed!",
1463 __func__);
1467 if (error) {
1468 xfs_warn(mp, "Failed to initialize disk quotas.");
1469 return;
1474 * This is called after the superblock has been read in and we're ready to
1475 * iget the quota inodes.
1477 STATIC int
1478 xfs_qm_init_quotainos(
1479 xfs_mount_t *mp)
1481 struct xfs_inode *uip = NULL;
1482 struct xfs_inode *gip = NULL;
1483 struct xfs_inode *pip = NULL;
1484 int error;
1485 __int64_t sbflags = 0;
1486 uint flags = 0;
1488 ASSERT(mp->m_quotainfo);
1491 * Get the uquota and gquota inodes
1493 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1494 if (XFS_IS_UQUOTA_ON(mp) &&
1495 mp->m_sb.sb_uquotino != NULLFSINO) {
1496 ASSERT(mp->m_sb.sb_uquotino > 0);
1497 error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1498 0, 0, &uip);
1499 if (error)
1500 return error;
1502 if (XFS_IS_GQUOTA_ON(mp) &&
1503 mp->m_sb.sb_gquotino != NULLFSINO) {
1504 ASSERT(mp->m_sb.sb_gquotino > 0);
1505 error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1506 0, 0, &gip);
1507 if (error)
1508 goto error_rele;
1510 if (XFS_IS_PQUOTA_ON(mp) &&
1511 mp->m_sb.sb_pquotino != NULLFSINO) {
1512 ASSERT(mp->m_sb.sb_pquotino > 0);
1513 error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino,
1514 0, 0, &pip);
1515 if (error)
1516 goto error_rele;
1518 } else {
1519 flags |= XFS_QMOPT_SBVERSION;
1520 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1521 XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1522 XFS_SB_QFLAGS);
1526 * Create the three inodes, if they don't exist already. The changes
1527 * made above will get added to a transaction and logged in one of
1528 * the qino_alloc calls below. If the device is readonly,
1529 * temporarily switch to read-write to do this.
1531 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1532 error = xfs_qm_qino_alloc(mp, &uip,
1533 sbflags | XFS_SB_UQUOTINO,
1534 flags | XFS_QMOPT_UQUOTA);
1535 if (error)
1536 goto error_rele;
1538 flags &= ~XFS_QMOPT_SBVERSION;
1540 if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
1541 error = xfs_qm_qino_alloc(mp, &gip,
1542 sbflags | XFS_SB_GQUOTINO,
1543 flags | XFS_QMOPT_GQUOTA);
1544 if (error)
1545 goto error_rele;
1547 flags &= ~XFS_QMOPT_SBVERSION;
1549 if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) {
1550 error = xfs_qm_qino_alloc(mp, &pip,
1551 sbflags | XFS_SB_PQUOTINO,
1552 flags | XFS_QMOPT_PQUOTA);
1553 if (error)
1554 goto error_rele;
1557 mp->m_quotainfo->qi_uquotaip = uip;
1558 mp->m_quotainfo->qi_gquotaip = gip;
1559 mp->m_quotainfo->qi_pquotaip = pip;
1561 return 0;
1563 error_rele:
1564 if (uip)
1565 IRELE(uip);
1566 if (gip)
1567 IRELE(gip);
1568 if (pip)
1569 IRELE(pip);
1570 return error;
1573 STATIC void
1574 xfs_qm_dqfree_one(
1575 struct xfs_dquot *dqp)
1577 struct xfs_mount *mp = dqp->q_mount;
1578 struct xfs_quotainfo *qi = mp->m_quotainfo;
1580 mutex_lock(&qi->qi_tree_lock);
1581 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
1582 be32_to_cpu(dqp->q_core.d_id));
1584 qi->qi_dquots--;
1585 mutex_unlock(&qi->qi_tree_lock);
1587 xfs_qm_dqdestroy(dqp);
1591 * Start a transaction and write the incore superblock changes to
1592 * disk. flags parameter indicates which fields have changed.
1595 xfs_qm_write_sb_changes(
1596 xfs_mount_t *mp,
1597 __int64_t flags)
1599 xfs_trans_t *tp;
1600 int error;
1602 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1603 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
1604 if (error) {
1605 xfs_trans_cancel(tp, 0);
1606 return error;
1609 xfs_mod_sb(tp, flags);
1610 error = xfs_trans_commit(tp, 0);
1612 return error;
1616 /* --------------- utility functions for vnodeops ---------------- */
1620 * Given an inode, a uid, gid and prid make sure that we have
1621 * allocated relevant dquot(s) on disk, and that we won't exceed inode
1622 * quotas by creating this file.
1623 * This also attaches dquot(s) to the given inode after locking it,
1624 * and returns the dquots corresponding to the uid and/or gid.
1626 * in : inode (unlocked)
1627 * out : udquot, gdquot with references taken and unlocked
1630 xfs_qm_vop_dqalloc(
1631 struct xfs_inode *ip,
1632 xfs_dqid_t uid,
1633 xfs_dqid_t gid,
1634 prid_t prid,
1635 uint flags,
1636 struct xfs_dquot **O_udqpp,
1637 struct xfs_dquot **O_gdqpp,
1638 struct xfs_dquot **O_pdqpp)
1640 struct xfs_mount *mp = ip->i_mount;
1641 struct xfs_dquot *uq = NULL;
1642 struct xfs_dquot *gq = NULL;
1643 struct xfs_dquot *pq = NULL;
1644 int error;
1645 uint lockflags;
1647 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1648 return 0;
1650 lockflags = XFS_ILOCK_EXCL;
1651 xfs_ilock(ip, lockflags);
1653 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1654 gid = ip->i_d.di_gid;
1657 * Attach the dquot(s) to this inode, doing a dquot allocation
1658 * if necessary. The dquot(s) will not be locked.
1660 if (XFS_NOT_DQATTACHED(mp, ip)) {
1661 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
1662 if (error) {
1663 xfs_iunlock(ip, lockflags);
1664 return error;
1668 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1669 if (ip->i_d.di_uid != uid) {
1671 * What we need is the dquot that has this uid, and
1672 * if we send the inode to dqget, the uid of the inode
1673 * takes priority over what's sent in the uid argument.
1674 * We must unlock inode here before calling dqget if
1675 * we're not sending the inode, because otherwise
1676 * we'll deadlock by doing trans_reserve while
1677 * holding ilock.
1679 xfs_iunlock(ip, lockflags);
1680 error = xfs_qm_dqget(mp, NULL, uid,
1681 XFS_DQ_USER,
1682 XFS_QMOPT_DQALLOC |
1683 XFS_QMOPT_DOWARN,
1684 &uq);
1685 if (error) {
1686 ASSERT(error != -ENOENT);
1687 return error;
1690 * Get the ilock in the right order.
1692 xfs_dqunlock(uq);
1693 lockflags = XFS_ILOCK_SHARED;
1694 xfs_ilock(ip, lockflags);
1695 } else {
1697 * Take an extra reference, because we'll return
1698 * this to caller
1700 ASSERT(ip->i_udquot);
1701 uq = xfs_qm_dqhold(ip->i_udquot);
1704 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1705 if (ip->i_d.di_gid != gid) {
1706 xfs_iunlock(ip, lockflags);
1707 error = xfs_qm_dqget(mp, NULL, gid,
1708 XFS_DQ_GROUP,
1709 XFS_QMOPT_DQALLOC |
1710 XFS_QMOPT_DOWARN,
1711 &gq);
1712 if (error) {
1713 ASSERT(error != -ENOENT);
1714 goto error_rele;
1716 xfs_dqunlock(gq);
1717 lockflags = XFS_ILOCK_SHARED;
1718 xfs_ilock(ip, lockflags);
1719 } else {
1720 ASSERT(ip->i_gdquot);
1721 gq = xfs_qm_dqhold(ip->i_gdquot);
1724 if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1725 if (xfs_get_projid(ip) != prid) {
1726 xfs_iunlock(ip, lockflags);
1727 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
1728 XFS_DQ_PROJ,
1729 XFS_QMOPT_DQALLOC |
1730 XFS_QMOPT_DOWARN,
1731 &pq);
1732 if (error) {
1733 ASSERT(error != -ENOENT);
1734 goto error_rele;
1736 xfs_dqunlock(pq);
1737 lockflags = XFS_ILOCK_SHARED;
1738 xfs_ilock(ip, lockflags);
1739 } else {
1740 ASSERT(ip->i_pdquot);
1741 pq = xfs_qm_dqhold(ip->i_pdquot);
1744 if (uq)
1745 trace_xfs_dquot_dqalloc(ip);
1747 xfs_iunlock(ip, lockflags);
1748 if (O_udqpp)
1749 *O_udqpp = uq;
1750 else
1751 xfs_qm_dqrele(uq);
1752 if (O_gdqpp)
1753 *O_gdqpp = gq;
1754 else
1755 xfs_qm_dqrele(gq);
1756 if (O_pdqpp)
1757 *O_pdqpp = pq;
1758 else
1759 xfs_qm_dqrele(pq);
1760 return 0;
1762 error_rele:
1763 xfs_qm_dqrele(gq);
1764 xfs_qm_dqrele(uq);
1765 return error;
1769 * Actually transfer ownership, and do dquot modifications.
1770 * These were already reserved.
1772 xfs_dquot_t *
1773 xfs_qm_vop_chown(
1774 xfs_trans_t *tp,
1775 xfs_inode_t *ip,
1776 xfs_dquot_t **IO_olddq,
1777 xfs_dquot_t *newdq)
1779 xfs_dquot_t *prevdq;
1780 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
1781 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1784 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1785 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1787 /* old dquot */
1788 prevdq = *IO_olddq;
1789 ASSERT(prevdq);
1790 ASSERT(prevdq != newdq);
1792 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
1793 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1795 /* the sparkling new dquot */
1796 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
1797 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1800 * Take an extra reference, because the inode is going to keep
1801 * this dquot pointer even after the trans_commit.
1803 *IO_olddq = xfs_qm_dqhold(newdq);
1805 return prevdq;
1809 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1812 xfs_qm_vop_chown_reserve(
1813 struct xfs_trans *tp,
1814 struct xfs_inode *ip,
1815 struct xfs_dquot *udqp,
1816 struct xfs_dquot *gdqp,
1817 struct xfs_dquot *pdqp,
1818 uint flags)
1820 struct xfs_mount *mp = ip->i_mount;
1821 uint delblks, blkflags, prjflags = 0;
1822 struct xfs_dquot *udq_unres = NULL;
1823 struct xfs_dquot *gdq_unres = NULL;
1824 struct xfs_dquot *pdq_unres = NULL;
1825 struct xfs_dquot *udq_delblks = NULL;
1826 struct xfs_dquot *gdq_delblks = NULL;
1827 struct xfs_dquot *pdq_delblks = NULL;
1828 int error;
1831 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1832 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1834 delblks = ip->i_delayed_blks;
1835 blkflags = XFS_IS_REALTIME_INODE(ip) ?
1836 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
1838 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
1839 ip->i_d.di_uid != be32_to_cpu(udqp->q_core.d_id)) {
1840 udq_delblks = udqp;
1842 * If there are delayed allocation blocks, then we have to
1843 * unreserve those from the old dquot, and add them to the
1844 * new dquot.
1846 if (delblks) {
1847 ASSERT(ip->i_udquot);
1848 udq_unres = ip->i_udquot;
1851 if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp &&
1852 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id)) {
1853 gdq_delblks = gdqp;
1854 if (delblks) {
1855 ASSERT(ip->i_gdquot);
1856 gdq_unres = ip->i_gdquot;
1860 if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
1861 xfs_get_projid(ip) != be32_to_cpu(pdqp->q_core.d_id)) {
1862 prjflags = XFS_QMOPT_ENOSPC;
1863 pdq_delblks = pdqp;
1864 if (delblks) {
1865 ASSERT(ip->i_pdquot);
1866 pdq_unres = ip->i_pdquot;
1870 error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
1871 udq_delblks, gdq_delblks, pdq_delblks,
1872 ip->i_d.di_nblocks, 1,
1873 flags | blkflags | prjflags);
1874 if (error)
1875 return error;
1878 * Do the delayed blks reservations/unreservations now. Since, these
1879 * are done without the help of a transaction, if a reservation fails
1880 * its previous reservations won't be automatically undone by trans
1881 * code. So, we have to do it manually here.
1883 if (delblks) {
1885 * Do the reservations first. Unreservation can't fail.
1887 ASSERT(udq_delblks || gdq_delblks || pdq_delblks);
1888 ASSERT(udq_unres || gdq_unres || pdq_unres);
1889 error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
1890 udq_delblks, gdq_delblks, pdq_delblks,
1891 (xfs_qcnt_t)delblks, 0,
1892 flags | blkflags | prjflags);
1893 if (error)
1894 return error;
1895 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
1896 udq_unres, gdq_unres, pdq_unres,
1897 -((xfs_qcnt_t)delblks), 0, blkflags);
1900 return 0;
1904 xfs_qm_vop_rename_dqattach(
1905 struct xfs_inode **i_tab)
1907 struct xfs_mount *mp = i_tab[0]->i_mount;
1908 int i;
1910 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1911 return 0;
1913 for (i = 0; (i < 4 && i_tab[i]); i++) {
1914 struct xfs_inode *ip = i_tab[i];
1915 int error;
1918 * Watch out for duplicate entries in the table.
1920 if (i == 0 || ip != i_tab[i-1]) {
1921 if (XFS_NOT_DQATTACHED(mp, ip)) {
1922 error = xfs_qm_dqattach(ip, 0);
1923 if (error)
1924 return error;
1928 return 0;
1931 void
1932 xfs_qm_vop_create_dqattach(
1933 struct xfs_trans *tp,
1934 struct xfs_inode *ip,
1935 struct xfs_dquot *udqp,
1936 struct xfs_dquot *gdqp,
1937 struct xfs_dquot *pdqp)
1939 struct xfs_mount *mp = tp->t_mountp;
1941 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1942 return;
1944 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1945 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1947 if (udqp && XFS_IS_UQUOTA_ON(mp)) {
1948 ASSERT(ip->i_udquot == NULL);
1949 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
1951 ip->i_udquot = xfs_qm_dqhold(udqp);
1952 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
1954 if (gdqp && XFS_IS_GQUOTA_ON(mp)) {
1955 ASSERT(ip->i_gdquot == NULL);
1956 ASSERT(ip->i_d.di_gid == be32_to_cpu(gdqp->q_core.d_id));
1957 ip->i_gdquot = xfs_qm_dqhold(gdqp);
1958 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
1960 if (pdqp && XFS_IS_PQUOTA_ON(mp)) {
1961 ASSERT(ip->i_pdquot == NULL);
1962 ASSERT(xfs_get_projid(ip) == be32_to_cpu(pdqp->q_core.d_id));
1964 ip->i_pdquot = xfs_qm_dqhold(pdqp);
1965 xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1);