blk-mq: always free hctx after request queue is freed
[linux/fpc-iii.git] / fs / xfs / xfs_qm.c
blob52ed7904df10b75ad5182e9391589b8e351cf9f4
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_inode.h"
16 #include "xfs_ialloc.h"
17 #include "xfs_itable.h"
18 #include "xfs_quota.h"
19 #include "xfs_error.h"
20 #include "xfs_bmap.h"
21 #include "xfs_bmap_btree.h"
22 #include "xfs_bmap_util.h"
23 #include "xfs_trans.h"
24 #include "xfs_trans_space.h"
25 #include "xfs_qm.h"
26 #include "xfs_trace.h"
27 #include "xfs_icache.h"
28 #include "xfs_cksum.h"
31 * The global quota manager. There is only one of these for the entire
32 * system, _not_ one per file system. XQM keeps track of the overall
33 * quota functionality, including maintaining the freelist and hash
34 * tables of dquots.
36 STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
37 STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
39 STATIC void xfs_qm_destroy_quotainos(xfs_quotainfo_t *qi);
40 STATIC void xfs_qm_dqfree_one(struct xfs_dquot *dqp);
42 * We use the batch lookup interface to iterate over the dquots as it
43 * currently is the only interface into the radix tree code that allows
44 * fuzzy lookups instead of exact matches. Holding the lock over multiple
45 * operations is fine as all callers are used either during mount/umount
46 * or quotaoff.
48 #define XFS_DQ_LOOKUP_BATCH 32
50 STATIC int
51 xfs_qm_dquot_walk(
52 struct xfs_mount *mp,
53 int type,
54 int (*execute)(struct xfs_dquot *dqp, void *data),
55 void *data)
57 struct xfs_quotainfo *qi = mp->m_quotainfo;
58 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
59 uint32_t next_index;
60 int last_error = 0;
61 int skipped;
62 int nr_found;
64 restart:
65 skipped = 0;
66 next_index = 0;
67 nr_found = 0;
69 while (1) {
70 struct xfs_dquot *batch[XFS_DQ_LOOKUP_BATCH];
71 int error = 0;
72 int i;
74 mutex_lock(&qi->qi_tree_lock);
75 nr_found = radix_tree_gang_lookup(tree, (void **)batch,
76 next_index, XFS_DQ_LOOKUP_BATCH);
77 if (!nr_found) {
78 mutex_unlock(&qi->qi_tree_lock);
79 break;
82 for (i = 0; i < nr_found; i++) {
83 struct xfs_dquot *dqp = batch[i];
85 next_index = be32_to_cpu(dqp->q_core.d_id) + 1;
87 error = execute(batch[i], data);
88 if (error == -EAGAIN) {
89 skipped++;
90 continue;
92 if (error && last_error != -EFSCORRUPTED)
93 last_error = error;
96 mutex_unlock(&qi->qi_tree_lock);
98 /* bail out if the filesystem is corrupted. */
99 if (last_error == -EFSCORRUPTED) {
100 skipped = 0;
101 break;
103 /* we're done if id overflows back to zero */
104 if (!next_index)
105 break;
108 if (skipped) {
109 delay(1);
110 goto restart;
113 return last_error;
118 * Purge a dquot from all tracking data structures and free it.
120 STATIC int
121 xfs_qm_dqpurge(
122 struct xfs_dquot *dqp,
123 void *data)
125 struct xfs_mount *mp = dqp->q_mount;
126 struct xfs_quotainfo *qi = mp->m_quotainfo;
128 xfs_dqlock(dqp);
129 if ((dqp->dq_flags & XFS_DQ_FREEING) || dqp->q_nrefs != 0) {
130 xfs_dqunlock(dqp);
131 return -EAGAIN;
134 dqp->dq_flags |= XFS_DQ_FREEING;
136 xfs_dqflock(dqp);
139 * If we are turning this type of quotas off, we don't care
140 * about the dirty metadata sitting in this dquot. OTOH, if
141 * we're unmounting, we do care, so we flush it and wait.
143 if (XFS_DQ_IS_DIRTY(dqp)) {
144 struct xfs_buf *bp = NULL;
145 int error;
148 * We don't care about getting disk errors here. We need
149 * to purge this dquot anyway, so we go ahead regardless.
151 error = xfs_qm_dqflush(dqp, &bp);
152 if (!error) {
153 error = xfs_bwrite(bp);
154 xfs_buf_relse(bp);
156 xfs_dqflock(dqp);
159 ASSERT(atomic_read(&dqp->q_pincount) == 0);
160 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
161 !test_bit(XFS_LI_IN_AIL, &dqp->q_logitem.qli_item.li_flags));
163 xfs_dqfunlock(dqp);
164 xfs_dqunlock(dqp);
166 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
167 be32_to_cpu(dqp->q_core.d_id));
168 qi->qi_dquots--;
171 * We move dquots to the freelist as soon as their reference count
172 * hits zero, so it really should be on the freelist here.
174 ASSERT(!list_empty(&dqp->q_lru));
175 list_lru_del(&qi->qi_lru, &dqp->q_lru);
176 XFS_STATS_DEC(mp, xs_qm_dquot_unused);
178 xfs_qm_dqdestroy(dqp);
179 return 0;
183 * Purge the dquot cache.
185 void
186 xfs_qm_dqpurge_all(
187 struct xfs_mount *mp,
188 uint flags)
190 if (flags & XFS_QMOPT_UQUOTA)
191 xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_dqpurge, NULL);
192 if (flags & XFS_QMOPT_GQUOTA)
193 xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_dqpurge, NULL);
194 if (flags & XFS_QMOPT_PQUOTA)
195 xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_dqpurge, NULL);
199 * Just destroy the quotainfo structure.
201 void
202 xfs_qm_unmount(
203 struct xfs_mount *mp)
205 if (mp->m_quotainfo) {
206 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
207 xfs_qm_destroy_quotainfo(mp);
212 * Called from the vfsops layer.
214 void
215 xfs_qm_unmount_quotas(
216 xfs_mount_t *mp)
219 * Release the dquots that root inode, et al might be holding,
220 * before we flush quotas and blow away the quotainfo structure.
222 ASSERT(mp->m_rootip);
223 xfs_qm_dqdetach(mp->m_rootip);
224 if (mp->m_rbmip)
225 xfs_qm_dqdetach(mp->m_rbmip);
226 if (mp->m_rsumip)
227 xfs_qm_dqdetach(mp->m_rsumip);
230 * Release the quota inodes.
232 if (mp->m_quotainfo) {
233 if (mp->m_quotainfo->qi_uquotaip) {
234 xfs_irele(mp->m_quotainfo->qi_uquotaip);
235 mp->m_quotainfo->qi_uquotaip = NULL;
237 if (mp->m_quotainfo->qi_gquotaip) {
238 xfs_irele(mp->m_quotainfo->qi_gquotaip);
239 mp->m_quotainfo->qi_gquotaip = NULL;
241 if (mp->m_quotainfo->qi_pquotaip) {
242 xfs_irele(mp->m_quotainfo->qi_pquotaip);
243 mp->m_quotainfo->qi_pquotaip = NULL;
248 STATIC int
249 xfs_qm_dqattach_one(
250 xfs_inode_t *ip,
251 xfs_dqid_t id,
252 uint type,
253 bool doalloc,
254 xfs_dquot_t **IO_idqpp)
256 xfs_dquot_t *dqp;
257 int error;
259 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
260 error = 0;
263 * See if we already have it in the inode itself. IO_idqpp is &i_udquot
264 * or &i_gdquot. This made the code look weird, but made the logic a lot
265 * simpler.
267 dqp = *IO_idqpp;
268 if (dqp) {
269 trace_xfs_dqattach_found(dqp);
270 return 0;
274 * Find the dquot from somewhere. This bumps the reference count of
275 * dquot and returns it locked. This can return ENOENT if dquot didn't
276 * exist on disk and we didn't ask it to allocate; ESRCH if quotas got
277 * turned off suddenly.
279 error = xfs_qm_dqget_inode(ip, type, doalloc, &dqp);
280 if (error)
281 return error;
283 trace_xfs_dqattach_get(dqp);
286 * dqget may have dropped and re-acquired the ilock, but it guarantees
287 * that the dquot returned is the one that should go in the inode.
289 *IO_idqpp = dqp;
290 xfs_dqunlock(dqp);
291 return 0;
294 static bool
295 xfs_qm_need_dqattach(
296 struct xfs_inode *ip)
298 struct xfs_mount *mp = ip->i_mount;
300 if (!XFS_IS_QUOTA_RUNNING(mp))
301 return false;
302 if (!XFS_IS_QUOTA_ON(mp))
303 return false;
304 if (!XFS_NOT_DQATTACHED(mp, ip))
305 return false;
306 if (xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
307 return false;
308 return true;
312 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
313 * into account.
314 * If @doalloc is true, the dquot(s) will be allocated if needed.
315 * Inode may get unlocked and relocked in here, and the caller must deal with
316 * the consequences.
319 xfs_qm_dqattach_locked(
320 xfs_inode_t *ip,
321 bool doalloc)
323 xfs_mount_t *mp = ip->i_mount;
324 int error = 0;
326 if (!xfs_qm_need_dqattach(ip))
327 return 0;
329 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
331 if (XFS_IS_UQUOTA_ON(mp) && !ip->i_udquot) {
332 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
333 doalloc, &ip->i_udquot);
334 if (error)
335 goto done;
336 ASSERT(ip->i_udquot);
339 if (XFS_IS_GQUOTA_ON(mp) && !ip->i_gdquot) {
340 error = xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
341 doalloc, &ip->i_gdquot);
342 if (error)
343 goto done;
344 ASSERT(ip->i_gdquot);
347 if (XFS_IS_PQUOTA_ON(mp) && !ip->i_pdquot) {
348 error = xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
349 doalloc, &ip->i_pdquot);
350 if (error)
351 goto done;
352 ASSERT(ip->i_pdquot);
355 done:
357 * Don't worry about the dquots that we may have attached before any
358 * error - they'll get detached later if it has not already been done.
360 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
361 return error;
365 xfs_qm_dqattach(
366 struct xfs_inode *ip)
368 int error;
370 if (!xfs_qm_need_dqattach(ip))
371 return 0;
373 xfs_ilock(ip, XFS_ILOCK_EXCL);
374 error = xfs_qm_dqattach_locked(ip, false);
375 xfs_iunlock(ip, XFS_ILOCK_EXCL);
377 return error;
381 * Release dquots (and their references) if any.
382 * The inode should be locked EXCL except when this's called by
383 * xfs_ireclaim.
385 void
386 xfs_qm_dqdetach(
387 xfs_inode_t *ip)
389 if (!(ip->i_udquot || ip->i_gdquot || ip->i_pdquot))
390 return;
392 trace_xfs_dquot_dqdetach(ip);
394 ASSERT(!xfs_is_quota_inode(&ip->i_mount->m_sb, ip->i_ino));
395 if (ip->i_udquot) {
396 xfs_qm_dqrele(ip->i_udquot);
397 ip->i_udquot = NULL;
399 if (ip->i_gdquot) {
400 xfs_qm_dqrele(ip->i_gdquot);
401 ip->i_gdquot = NULL;
403 if (ip->i_pdquot) {
404 xfs_qm_dqrele(ip->i_pdquot);
405 ip->i_pdquot = NULL;
409 struct xfs_qm_isolate {
410 struct list_head buffers;
411 struct list_head dispose;
414 static enum lru_status
415 xfs_qm_dquot_isolate(
416 struct list_head *item,
417 struct list_lru_one *lru,
418 spinlock_t *lru_lock,
419 void *arg)
420 __releases(lru_lock) __acquires(lru_lock)
422 struct xfs_dquot *dqp = container_of(item,
423 struct xfs_dquot, q_lru);
424 struct xfs_qm_isolate *isol = arg;
426 if (!xfs_dqlock_nowait(dqp))
427 goto out_miss_busy;
430 * This dquot has acquired a reference in the meantime remove it from
431 * the freelist and try again.
433 if (dqp->q_nrefs) {
434 xfs_dqunlock(dqp);
435 XFS_STATS_INC(dqp->q_mount, xs_qm_dqwants);
437 trace_xfs_dqreclaim_want(dqp);
438 list_lru_isolate(lru, &dqp->q_lru);
439 XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot_unused);
440 return LRU_REMOVED;
444 * If the dquot is dirty, flush it. If it's already being flushed, just
445 * skip it so there is time for the IO to complete before we try to
446 * reclaim it again on the next LRU pass.
448 if (!xfs_dqflock_nowait(dqp)) {
449 xfs_dqunlock(dqp);
450 goto out_miss_busy;
453 if (XFS_DQ_IS_DIRTY(dqp)) {
454 struct xfs_buf *bp = NULL;
455 int error;
457 trace_xfs_dqreclaim_dirty(dqp);
459 /* we have to drop the LRU lock to flush the dquot */
460 spin_unlock(lru_lock);
462 error = xfs_qm_dqflush(dqp, &bp);
463 if (error)
464 goto out_unlock_dirty;
466 xfs_buf_delwri_queue(bp, &isol->buffers);
467 xfs_buf_relse(bp);
468 goto out_unlock_dirty;
470 xfs_dqfunlock(dqp);
473 * Prevent lookups now that we are past the point of no return.
475 dqp->dq_flags |= XFS_DQ_FREEING;
476 xfs_dqunlock(dqp);
478 ASSERT(dqp->q_nrefs == 0);
479 list_lru_isolate_move(lru, &dqp->q_lru, &isol->dispose);
480 XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot_unused);
481 trace_xfs_dqreclaim_done(dqp);
482 XFS_STATS_INC(dqp->q_mount, xs_qm_dqreclaims);
483 return LRU_REMOVED;
485 out_miss_busy:
486 trace_xfs_dqreclaim_busy(dqp);
487 XFS_STATS_INC(dqp->q_mount, xs_qm_dqreclaim_misses);
488 return LRU_SKIP;
490 out_unlock_dirty:
491 trace_xfs_dqreclaim_busy(dqp);
492 XFS_STATS_INC(dqp->q_mount, xs_qm_dqreclaim_misses);
493 xfs_dqunlock(dqp);
494 spin_lock(lru_lock);
495 return LRU_RETRY;
498 static unsigned long
499 xfs_qm_shrink_scan(
500 struct shrinker *shrink,
501 struct shrink_control *sc)
503 struct xfs_quotainfo *qi = container_of(shrink,
504 struct xfs_quotainfo, qi_shrinker);
505 struct xfs_qm_isolate isol;
506 unsigned long freed;
507 int error;
509 if ((sc->gfp_mask & (__GFP_FS|__GFP_DIRECT_RECLAIM)) != (__GFP_FS|__GFP_DIRECT_RECLAIM))
510 return 0;
512 INIT_LIST_HEAD(&isol.buffers);
513 INIT_LIST_HEAD(&isol.dispose);
515 freed = list_lru_shrink_walk(&qi->qi_lru, sc,
516 xfs_qm_dquot_isolate, &isol);
518 error = xfs_buf_delwri_submit(&isol.buffers);
519 if (error)
520 xfs_warn(NULL, "%s: dquot reclaim failed", __func__);
522 while (!list_empty(&isol.dispose)) {
523 struct xfs_dquot *dqp;
525 dqp = list_first_entry(&isol.dispose, struct xfs_dquot, q_lru);
526 list_del_init(&dqp->q_lru);
527 xfs_qm_dqfree_one(dqp);
530 return freed;
533 static unsigned long
534 xfs_qm_shrink_count(
535 struct shrinker *shrink,
536 struct shrink_control *sc)
538 struct xfs_quotainfo *qi = container_of(shrink,
539 struct xfs_quotainfo, qi_shrinker);
541 return list_lru_shrink_count(&qi->qi_lru, sc);
544 STATIC void
545 xfs_qm_set_defquota(
546 xfs_mount_t *mp,
547 uint type,
548 xfs_quotainfo_t *qinf)
550 xfs_dquot_t *dqp;
551 struct xfs_def_quota *defq;
552 struct xfs_disk_dquot *ddqp;
553 int error;
555 error = xfs_qm_dqget_uncached(mp, 0, type, &dqp);
556 if (error)
557 return;
559 ddqp = &dqp->q_core;
560 defq = xfs_get_defquota(dqp, qinf);
563 * Timers and warnings have been already set, let's just set the
564 * default limits for this quota type
566 defq->bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
567 defq->bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
568 defq->ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
569 defq->isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
570 defq->rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
571 defq->rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
572 xfs_qm_dqdestroy(dqp);
575 /* Initialize quota time limits from the root dquot. */
576 static void
577 xfs_qm_init_timelimits(
578 struct xfs_mount *mp,
579 struct xfs_quotainfo *qinf)
581 struct xfs_disk_dquot *ddqp;
582 struct xfs_dquot *dqp;
583 uint type;
584 int error;
586 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
587 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
588 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
589 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
590 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
591 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
594 * We try to get the limits from the superuser's limits fields.
595 * This is quite hacky, but it is standard quota practice.
597 * Since we may not have done a quotacheck by this point, just read
598 * the dquot without attaching it to any hashtables or lists.
600 * Timers and warnings are globally set by the first timer found in
601 * user/group/proj quota types, otherwise a default value is used.
602 * This should be split into different fields per quota type.
604 if (XFS_IS_UQUOTA_RUNNING(mp))
605 type = XFS_DQ_USER;
606 else if (XFS_IS_GQUOTA_RUNNING(mp))
607 type = XFS_DQ_GROUP;
608 else
609 type = XFS_DQ_PROJ;
610 error = xfs_qm_dqget_uncached(mp, 0, type, &dqp);
611 if (error)
612 return;
614 ddqp = &dqp->q_core;
616 * The warnings and timers set the grace period given to
617 * a user or group before he or she can not perform any
618 * more writing. If it is zero, a default is used.
620 if (ddqp->d_btimer)
621 qinf->qi_btimelimit = be32_to_cpu(ddqp->d_btimer);
622 if (ddqp->d_itimer)
623 qinf->qi_itimelimit = be32_to_cpu(ddqp->d_itimer);
624 if (ddqp->d_rtbtimer)
625 qinf->qi_rtbtimelimit = be32_to_cpu(ddqp->d_rtbtimer);
626 if (ddqp->d_bwarns)
627 qinf->qi_bwarnlimit = be16_to_cpu(ddqp->d_bwarns);
628 if (ddqp->d_iwarns)
629 qinf->qi_iwarnlimit = be16_to_cpu(ddqp->d_iwarns);
630 if (ddqp->d_rtbwarns)
631 qinf->qi_rtbwarnlimit = be16_to_cpu(ddqp->d_rtbwarns);
633 xfs_qm_dqdestroy(dqp);
637 * This initializes all the quota information that's kept in the
638 * mount structure
640 STATIC int
641 xfs_qm_init_quotainfo(
642 struct xfs_mount *mp)
644 struct xfs_quotainfo *qinf;
645 int error;
647 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
649 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
651 error = list_lru_init(&qinf->qi_lru);
652 if (error)
653 goto out_free_qinf;
656 * See if quotainodes are setup, and if not, allocate them,
657 * and change the superblock accordingly.
659 error = xfs_qm_init_quotainos(mp);
660 if (error)
661 goto out_free_lru;
663 INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
664 INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
665 INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS);
666 mutex_init(&qinf->qi_tree_lock);
668 /* mutex used to serialize quotaoffs */
669 mutex_init(&qinf->qi_quotaofflock);
671 /* Precalc some constants */
672 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
673 qinf->qi_dqperchunk = xfs_calc_dquots_per_chunk(qinf->qi_dqchunklen);
675 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
677 xfs_qm_init_timelimits(mp, qinf);
679 if (XFS_IS_UQUOTA_RUNNING(mp))
680 xfs_qm_set_defquota(mp, XFS_DQ_USER, qinf);
681 if (XFS_IS_GQUOTA_RUNNING(mp))
682 xfs_qm_set_defquota(mp, XFS_DQ_GROUP, qinf);
683 if (XFS_IS_PQUOTA_RUNNING(mp))
684 xfs_qm_set_defquota(mp, XFS_DQ_PROJ, qinf);
686 qinf->qi_shrinker.count_objects = xfs_qm_shrink_count;
687 qinf->qi_shrinker.scan_objects = xfs_qm_shrink_scan;
688 qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
689 qinf->qi_shrinker.flags = SHRINKER_NUMA_AWARE;
691 error = register_shrinker(&qinf->qi_shrinker);
692 if (error)
693 goto out_free_inos;
695 return 0;
697 out_free_inos:
698 mutex_destroy(&qinf->qi_quotaofflock);
699 mutex_destroy(&qinf->qi_tree_lock);
700 xfs_qm_destroy_quotainos(qinf);
701 out_free_lru:
702 list_lru_destroy(&qinf->qi_lru);
703 out_free_qinf:
704 kmem_free(qinf);
705 mp->m_quotainfo = NULL;
706 return error;
710 * Gets called when unmounting a filesystem or when all quotas get
711 * turned off.
712 * This purges the quota inodes, destroys locks and frees itself.
714 void
715 xfs_qm_destroy_quotainfo(
716 xfs_mount_t *mp)
718 xfs_quotainfo_t *qi;
720 qi = mp->m_quotainfo;
721 ASSERT(qi != NULL);
723 unregister_shrinker(&qi->qi_shrinker);
724 list_lru_destroy(&qi->qi_lru);
725 xfs_qm_destroy_quotainos(qi);
726 mutex_destroy(&qi->qi_tree_lock);
727 mutex_destroy(&qi->qi_quotaofflock);
728 kmem_free(qi);
729 mp->m_quotainfo = NULL;
733 * Create an inode and return with a reference already taken, but unlocked
734 * This is how we create quota inodes
736 STATIC int
737 xfs_qm_qino_alloc(
738 xfs_mount_t *mp,
739 xfs_inode_t **ip,
740 uint flags)
742 xfs_trans_t *tp;
743 int error;
744 bool need_alloc = true;
746 *ip = NULL;
748 * With superblock that doesn't have separate pquotino, we
749 * share an inode between gquota and pquota. If the on-disk
750 * superblock has GQUOTA and the filesystem is now mounted
751 * with PQUOTA, just use sb_gquotino for sb_pquotino and
752 * vice-versa.
754 if (!xfs_sb_version_has_pquotino(&mp->m_sb) &&
755 (flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) {
756 xfs_ino_t ino = NULLFSINO;
758 if ((flags & XFS_QMOPT_PQUOTA) &&
759 (mp->m_sb.sb_gquotino != NULLFSINO)) {
760 ino = mp->m_sb.sb_gquotino;
761 ASSERT(mp->m_sb.sb_pquotino == NULLFSINO);
762 } else if ((flags & XFS_QMOPT_GQUOTA) &&
763 (mp->m_sb.sb_pquotino != NULLFSINO)) {
764 ino = mp->m_sb.sb_pquotino;
765 ASSERT(mp->m_sb.sb_gquotino == NULLFSINO);
767 if (ino != NULLFSINO) {
768 error = xfs_iget(mp, NULL, ino, 0, 0, ip);
769 if (error)
770 return error;
771 mp->m_sb.sb_gquotino = NULLFSINO;
772 mp->m_sb.sb_pquotino = NULLFSINO;
773 need_alloc = false;
777 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_create,
778 XFS_QM_QINOCREATE_SPACE_RES(mp), 0, 0, &tp);
779 if (error)
780 return error;
782 if (need_alloc) {
783 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, ip);
784 if (error) {
785 xfs_trans_cancel(tp);
786 return error;
791 * Make the changes in the superblock, and log those too.
792 * sbfields arg may contain fields other than *QUOTINO;
793 * VERSIONNUM for example.
795 spin_lock(&mp->m_sb_lock);
796 if (flags & XFS_QMOPT_SBVERSION) {
797 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
799 xfs_sb_version_addquota(&mp->m_sb);
800 mp->m_sb.sb_uquotino = NULLFSINO;
801 mp->m_sb.sb_gquotino = NULLFSINO;
802 mp->m_sb.sb_pquotino = NULLFSINO;
804 /* qflags will get updated fully _after_ quotacheck */
805 mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT;
807 if (flags & XFS_QMOPT_UQUOTA)
808 mp->m_sb.sb_uquotino = (*ip)->i_ino;
809 else if (flags & XFS_QMOPT_GQUOTA)
810 mp->m_sb.sb_gquotino = (*ip)->i_ino;
811 else
812 mp->m_sb.sb_pquotino = (*ip)->i_ino;
813 spin_unlock(&mp->m_sb_lock);
814 xfs_log_sb(tp);
816 error = xfs_trans_commit(tp);
817 if (error) {
818 ASSERT(XFS_FORCED_SHUTDOWN(mp));
819 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
821 if (need_alloc)
822 xfs_finish_inode_setup(*ip);
823 return error;
827 STATIC void
828 xfs_qm_reset_dqcounts(
829 xfs_mount_t *mp,
830 xfs_buf_t *bp,
831 xfs_dqid_t id,
832 uint type)
834 struct xfs_dqblk *dqb;
835 int j;
836 xfs_failaddr_t fa;
838 trace_xfs_reset_dqcounts(bp, _RET_IP_);
841 * Reset all counters and timers. They'll be
842 * started afresh by xfs_qm_quotacheck.
844 #ifdef DEBUG
845 j = (int)XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB) /
846 sizeof(xfs_dqblk_t);
847 ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
848 #endif
849 dqb = bp->b_addr;
850 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
851 struct xfs_disk_dquot *ddq;
853 ddq = (struct xfs_disk_dquot *)&dqb[j];
856 * Do a sanity check, and if needed, repair the dqblk. Don't
857 * output any warnings because it's perfectly possible to
858 * find uninitialised dquot blks. See comment in
859 * xfs_dquot_verify.
861 fa = xfs_dqblk_verify(mp, &dqb[j], id + j, type);
862 if (fa)
863 xfs_dqblk_repair(mp, &dqb[j], id + j, type);
866 * Reset type in case we are reusing group quota file for
867 * project quotas or vice versa
869 ddq->d_flags = type;
870 ddq->d_bcount = 0;
871 ddq->d_icount = 0;
872 ddq->d_rtbcount = 0;
873 ddq->d_btimer = 0;
874 ddq->d_itimer = 0;
875 ddq->d_rtbtimer = 0;
876 ddq->d_bwarns = 0;
877 ddq->d_iwarns = 0;
878 ddq->d_rtbwarns = 0;
880 if (xfs_sb_version_hascrc(&mp->m_sb)) {
881 xfs_update_cksum((char *)&dqb[j],
882 sizeof(struct xfs_dqblk),
883 XFS_DQUOT_CRC_OFF);
888 STATIC int
889 xfs_qm_reset_dqcounts_all(
890 struct xfs_mount *mp,
891 xfs_dqid_t firstid,
892 xfs_fsblock_t bno,
893 xfs_filblks_t blkcnt,
894 uint flags,
895 struct list_head *buffer_list)
897 struct xfs_buf *bp;
898 int error;
899 int type;
901 ASSERT(blkcnt > 0);
902 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
903 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
904 error = 0;
907 * Blkcnt arg can be a very big number, and might even be
908 * larger than the log itself. So, we have to break it up into
909 * manageable-sized transactions.
910 * Note that we don't start a permanent transaction here; we might
911 * not be able to get a log reservation for the whole thing up front,
912 * and we don't really care to either, because we just discard
913 * everything if we were to crash in the middle of this loop.
915 while (blkcnt--) {
916 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
917 XFS_FSB_TO_DADDR(mp, bno),
918 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
919 &xfs_dquot_buf_ops);
922 * CRC and validation errors will return a EFSCORRUPTED here. If
923 * this occurs, re-read without CRC validation so that we can
924 * repair the damage via xfs_qm_reset_dqcounts(). This process
925 * will leave a trace in the log indicating corruption has
926 * been detected.
928 if (error == -EFSCORRUPTED) {
929 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
930 XFS_FSB_TO_DADDR(mp, bno),
931 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
932 NULL);
935 if (error)
936 break;
939 * A corrupt buffer might not have a verifier attached, so
940 * make sure we have the correct one attached before writeback
941 * occurs.
943 bp->b_ops = &xfs_dquot_buf_ops;
944 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
945 xfs_buf_delwri_queue(bp, buffer_list);
946 xfs_buf_relse(bp);
948 /* goto the next block. */
949 bno++;
950 firstid += mp->m_quotainfo->qi_dqperchunk;
953 return error;
957 * Iterate over all allocated dquot blocks in this quota inode, zeroing all
958 * counters for every chunk of dquots that we find.
960 STATIC int
961 xfs_qm_reset_dqcounts_buf(
962 struct xfs_mount *mp,
963 struct xfs_inode *qip,
964 uint flags,
965 struct list_head *buffer_list)
967 struct xfs_bmbt_irec *map;
968 int i, nmaps; /* number of map entries */
969 int error; /* return value */
970 xfs_fileoff_t lblkno;
971 xfs_filblks_t maxlblkcnt;
972 xfs_dqid_t firstid;
973 xfs_fsblock_t rablkno;
974 xfs_filblks_t rablkcnt;
976 error = 0;
978 * This looks racy, but we can't keep an inode lock across a
979 * trans_reserve. But, this gets called during quotacheck, and that
980 * happens only at mount time which is single threaded.
982 if (qip->i_d.di_nblocks == 0)
983 return 0;
985 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
987 lblkno = 0;
988 maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
989 do {
990 uint lock_mode;
992 nmaps = XFS_DQITER_MAP_SIZE;
994 * We aren't changing the inode itself. Just changing
995 * some of its data. No new blocks are added here, and
996 * the inode is never added to the transaction.
998 lock_mode = xfs_ilock_data_map_shared(qip);
999 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
1000 map, &nmaps, 0);
1001 xfs_iunlock(qip, lock_mode);
1002 if (error)
1003 break;
1005 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1006 for (i = 0; i < nmaps; i++) {
1007 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1008 ASSERT(map[i].br_blockcount);
1011 lblkno += map[i].br_blockcount;
1013 if (map[i].br_startblock == HOLESTARTBLOCK)
1014 continue;
1016 firstid = (xfs_dqid_t) map[i].br_startoff *
1017 mp->m_quotainfo->qi_dqperchunk;
1019 * Do a read-ahead on the next extent.
1021 if ((i+1 < nmaps) &&
1022 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1023 rablkcnt = map[i+1].br_blockcount;
1024 rablkno = map[i+1].br_startblock;
1025 while (rablkcnt--) {
1026 xfs_buf_readahead(mp->m_ddev_targp,
1027 XFS_FSB_TO_DADDR(mp, rablkno),
1028 mp->m_quotainfo->qi_dqchunklen,
1029 &xfs_dquot_buf_ops);
1030 rablkno++;
1034 * Iterate thru all the blks in the extent and
1035 * reset the counters of all the dquots inside them.
1037 error = xfs_qm_reset_dqcounts_all(mp, firstid,
1038 map[i].br_startblock,
1039 map[i].br_blockcount,
1040 flags, buffer_list);
1041 if (error)
1042 goto out;
1044 } while (nmaps > 0);
1046 out:
1047 kmem_free(map);
1048 return error;
1052 * Called by dqusage_adjust in doing a quotacheck.
1054 * Given the inode, and a dquot id this updates both the incore dqout as well
1055 * as the buffer copy. This is so that once the quotacheck is done, we can
1056 * just log all the buffers, as opposed to logging numerous updates to
1057 * individual dquots.
1059 STATIC int
1060 xfs_qm_quotacheck_dqadjust(
1061 struct xfs_inode *ip,
1062 uint type,
1063 xfs_qcnt_t nblks,
1064 xfs_qcnt_t rtblks)
1066 struct xfs_mount *mp = ip->i_mount;
1067 struct xfs_dquot *dqp;
1068 xfs_dqid_t id;
1069 int error;
1071 id = xfs_qm_id_for_quotatype(ip, type);
1072 error = xfs_qm_dqget(mp, id, type, true, &dqp);
1073 if (error) {
1075 * Shouldn't be able to turn off quotas here.
1077 ASSERT(error != -ESRCH);
1078 ASSERT(error != -ENOENT);
1079 return error;
1082 trace_xfs_dqadjust(dqp);
1085 * Adjust the inode count and the block count to reflect this inode's
1086 * resource usage.
1088 be64_add_cpu(&dqp->q_core.d_icount, 1);
1089 dqp->q_res_icount++;
1090 if (nblks) {
1091 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1092 dqp->q_res_bcount += nblks;
1094 if (rtblks) {
1095 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1096 dqp->q_res_rtbcount += rtblks;
1100 * Set default limits, adjust timers (since we changed usages)
1102 * There are no timers for the default values set in the root dquot.
1104 if (dqp->q_core.d_id) {
1105 xfs_qm_adjust_dqlimits(mp, dqp);
1106 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1109 dqp->dq_flags |= XFS_DQ_DIRTY;
1110 xfs_qm_dqput(dqp);
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;
1130 xfs_filblks_t rtblks = 0; /* total rt blks */
1131 int error;
1133 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1136 * rootino must have its resources accounted for, not so with the quota
1137 * inodes.
1139 if (xfs_is_quota_inode(&mp->m_sb, ino)) {
1140 *res = BULKSTAT_RV_NOTHING;
1141 return -EINVAL;
1145 * We don't _need_ to take the ilock EXCL here because quotacheck runs
1146 * at mount time and therefore nobody will be racing chown/chproj.
1148 error = xfs_iget(mp, NULL, ino, XFS_IGET_DONTCACHE, 0, &ip);
1149 if (error) {
1150 *res = BULKSTAT_RV_NOTHING;
1151 return error;
1154 ASSERT(ip->i_delayed_blks == 0);
1156 if (XFS_IS_REALTIME_INODE(ip)) {
1157 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1159 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1160 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1161 if (error)
1162 goto error0;
1165 xfs_bmap_count_leaves(ifp, &rtblks);
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, XFS_DQ_USER, nblks,
1184 rtblks);
1185 if (error)
1186 goto error0;
1189 if (XFS_IS_GQUOTA_ON(mp)) {
1190 error = xfs_qm_quotacheck_dqadjust(ip, XFS_DQ_GROUP, nblks,
1191 rtblks);
1192 if (error)
1193 goto error0;
1196 if (XFS_IS_PQUOTA_ON(mp)) {
1197 error = xfs_qm_quotacheck_dqadjust(ip, XFS_DQ_PROJ, nblks,
1198 rtblks);
1199 if (error)
1200 goto error0;
1203 xfs_irele(ip);
1204 *res = BULKSTAT_RV_DIDONE;
1205 return 0;
1207 error0:
1208 xfs_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 xfs_mount *mp = dqp->q_mount;
1219 struct list_head *buffer_list = data;
1220 struct xfs_buf *bp = NULL;
1221 int error = 0;
1223 xfs_dqlock(dqp);
1224 if (dqp->dq_flags & XFS_DQ_FREEING)
1225 goto out_unlock;
1226 if (!XFS_DQ_IS_DIRTY(dqp))
1227 goto out_unlock;
1230 * The only way the dquot is already flush locked by the time quotacheck
1231 * gets here is if reclaim flushed it before the dqadjust walk dirtied
1232 * it for the final time. Quotacheck collects all dquot bufs in the
1233 * local delwri queue before dquots are dirtied, so reclaim can't have
1234 * possibly queued it for I/O. The only way out is to push the buffer to
1235 * cycle the flush lock.
1237 if (!xfs_dqflock_nowait(dqp)) {
1238 /* buf is pinned in-core by delwri list */
1239 bp = xfs_buf_incore(mp->m_ddev_targp, dqp->q_blkno,
1240 mp->m_quotainfo->qi_dqchunklen, 0);
1241 if (!bp) {
1242 error = -EINVAL;
1243 goto out_unlock;
1245 xfs_buf_unlock(bp);
1247 xfs_buf_delwri_pushbuf(bp, buffer_list);
1248 xfs_buf_rele(bp);
1250 error = -EAGAIN;
1251 goto out_unlock;
1254 error = xfs_qm_dqflush(dqp, &bp);
1255 if (error)
1256 goto out_unlock;
1258 xfs_buf_delwri_queue(bp, buffer_list);
1259 xfs_buf_relse(bp);
1260 out_unlock:
1261 xfs_dqunlock(dqp);
1262 return error;
1266 * Walk thru all the filesystem inodes and construct a consistent view
1267 * of the disk quota world. If the quotacheck fails, disable quotas.
1269 STATIC int
1270 xfs_qm_quotacheck(
1271 xfs_mount_t *mp)
1273 int done, count, error, error2;
1274 xfs_ino_t lastino;
1275 size_t structsz;
1276 uint flags;
1277 LIST_HEAD (buffer_list);
1278 struct xfs_inode *uip = mp->m_quotainfo->qi_uquotaip;
1279 struct xfs_inode *gip = mp->m_quotainfo->qi_gquotaip;
1280 struct xfs_inode *pip = mp->m_quotainfo->qi_pquotaip;
1282 count = INT_MAX;
1283 structsz = 1;
1284 lastino = 0;
1285 flags = 0;
1287 ASSERT(uip || gip || pip);
1288 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1290 xfs_notice(mp, "Quotacheck needed: Please wait.");
1293 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1294 * their counters to zero. We need a clean slate.
1295 * We don't log our changes till later.
1297 if (uip) {
1298 error = xfs_qm_reset_dqcounts_buf(mp, uip, XFS_QMOPT_UQUOTA,
1299 &buffer_list);
1300 if (error)
1301 goto error_return;
1302 flags |= XFS_UQUOTA_CHKD;
1305 if (gip) {
1306 error = xfs_qm_reset_dqcounts_buf(mp, gip, XFS_QMOPT_GQUOTA,
1307 &buffer_list);
1308 if (error)
1309 goto error_return;
1310 flags |= XFS_GQUOTA_CHKD;
1313 if (pip) {
1314 error = xfs_qm_reset_dqcounts_buf(mp, pip, XFS_QMOPT_PQUOTA,
1315 &buffer_list);
1316 if (error)
1317 goto error_return;
1318 flags |= XFS_PQUOTA_CHKD;
1321 do {
1323 * Iterate thru all the inodes in the file system,
1324 * adjusting the corresponding dquot counters in core.
1326 error = xfs_bulkstat(mp, &lastino, &count,
1327 xfs_qm_dqusage_adjust,
1328 structsz, NULL, &done);
1329 if (error)
1330 break;
1332 } while (!done);
1335 * We've made all the changes that we need to make incore. Flush them
1336 * down to disk buffers if everything was updated successfully.
1338 if (XFS_IS_UQUOTA_ON(mp)) {
1339 error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one,
1340 &buffer_list);
1342 if (XFS_IS_GQUOTA_ON(mp)) {
1343 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one,
1344 &buffer_list);
1345 if (!error)
1346 error = error2;
1348 if (XFS_IS_PQUOTA_ON(mp)) {
1349 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one,
1350 &buffer_list);
1351 if (!error)
1352 error = error2;
1355 error2 = xfs_buf_delwri_submit(&buffer_list);
1356 if (!error)
1357 error = error2;
1360 * We can get this error if we couldn't do a dquot allocation inside
1361 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1362 * dirty dquots that might be cached, we just want to get rid of them
1363 * and turn quotaoff. The dquots won't be attached to any of the inodes
1364 * at this point (because we intentionally didn't in dqget_noattach).
1366 if (error) {
1367 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1368 goto error_return;
1372 * If one type of quotas is off, then it will lose its
1373 * quotachecked status, since we won't be doing accounting for
1374 * that type anymore.
1376 mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1377 mp->m_qflags |= flags;
1379 error_return:
1380 xfs_buf_delwri_cancel(&buffer_list);
1382 if (error) {
1383 xfs_warn(mp,
1384 "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1385 error);
1387 * We must turn off quotas.
1389 ASSERT(mp->m_quotainfo != NULL);
1390 xfs_qm_destroy_quotainfo(mp);
1391 if (xfs_mount_reset_sbqflags(mp)) {
1392 xfs_warn(mp,
1393 "Quotacheck: Failed to reset quota flags.");
1395 } else
1396 xfs_notice(mp, "Quotacheck: Done.");
1397 return error;
1401 * This is called from xfs_mountfs to start quotas and initialize all
1402 * necessary data structures like quotainfo. This is also responsible for
1403 * running a quotacheck as necessary. We are guaranteed that the superblock
1404 * is consistently read in at this point.
1406 * If we fail here, the mount will continue with quota turned off. We don't
1407 * need to inidicate success or failure at all.
1409 void
1410 xfs_qm_mount_quotas(
1411 struct xfs_mount *mp)
1413 int error = 0;
1414 uint sbf;
1417 * If quotas on realtime volumes is not supported, we disable
1418 * quotas immediately.
1420 if (mp->m_sb.sb_rextents) {
1421 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
1422 mp->m_qflags = 0;
1423 goto write_changes;
1426 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1429 * Allocate the quotainfo structure inside the mount struct, and
1430 * create quotainode(s), and change/rev superblock if necessary.
1432 error = xfs_qm_init_quotainfo(mp);
1433 if (error) {
1435 * We must turn off quotas.
1437 ASSERT(mp->m_quotainfo == NULL);
1438 mp->m_qflags = 0;
1439 goto write_changes;
1442 * If any of the quotas are not consistent, do a quotacheck.
1444 if (XFS_QM_NEED_QUOTACHECK(mp)) {
1445 error = xfs_qm_quotacheck(mp);
1446 if (error) {
1447 /* Quotacheck failed and disabled quotas. */
1448 return;
1452 * If one type of quotas is off, then it will lose its
1453 * quotachecked status, since we won't be doing accounting for
1454 * that type anymore.
1456 if (!XFS_IS_UQUOTA_ON(mp))
1457 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
1458 if (!XFS_IS_GQUOTA_ON(mp))
1459 mp->m_qflags &= ~XFS_GQUOTA_CHKD;
1460 if (!XFS_IS_PQUOTA_ON(mp))
1461 mp->m_qflags &= ~XFS_PQUOTA_CHKD;
1463 write_changes:
1465 * We actually don't have to acquire the m_sb_lock at all.
1466 * This can only be called from mount, and that's single threaded. XXX
1468 spin_lock(&mp->m_sb_lock);
1469 sbf = mp->m_sb.sb_qflags;
1470 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
1471 spin_unlock(&mp->m_sb_lock);
1473 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
1474 if (xfs_sync_sb(mp, false)) {
1476 * We could only have been turning quotas off.
1477 * We aren't in very good shape actually because
1478 * the incore structures are convinced that quotas are
1479 * off, but the on disk superblock doesn't know that !
1481 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
1482 xfs_alert(mp, "%s: Superblock update failed!",
1483 __func__);
1487 if (error) {
1488 xfs_warn(mp, "Failed to initialize disk quotas.");
1489 return;
1494 * This is called after the superblock has been read in and we're ready to
1495 * iget the quota inodes.
1497 STATIC int
1498 xfs_qm_init_quotainos(
1499 xfs_mount_t *mp)
1501 struct xfs_inode *uip = NULL;
1502 struct xfs_inode *gip = NULL;
1503 struct xfs_inode *pip = NULL;
1504 int error;
1505 uint flags = 0;
1507 ASSERT(mp->m_quotainfo);
1510 * Get the uquota and gquota inodes
1512 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1513 if (XFS_IS_UQUOTA_ON(mp) &&
1514 mp->m_sb.sb_uquotino != NULLFSINO) {
1515 ASSERT(mp->m_sb.sb_uquotino > 0);
1516 error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1517 0, 0, &uip);
1518 if (error)
1519 return error;
1521 if (XFS_IS_GQUOTA_ON(mp) &&
1522 mp->m_sb.sb_gquotino != NULLFSINO) {
1523 ASSERT(mp->m_sb.sb_gquotino > 0);
1524 error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1525 0, 0, &gip);
1526 if (error)
1527 goto error_rele;
1529 if (XFS_IS_PQUOTA_ON(mp) &&
1530 mp->m_sb.sb_pquotino != NULLFSINO) {
1531 ASSERT(mp->m_sb.sb_pquotino > 0);
1532 error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino,
1533 0, 0, &pip);
1534 if (error)
1535 goto error_rele;
1537 } else {
1538 flags |= XFS_QMOPT_SBVERSION;
1542 * Create the three inodes, if they don't exist already. The changes
1543 * made above will get added to a transaction and logged in one of
1544 * the qino_alloc calls below. If the device is readonly,
1545 * temporarily switch to read-write to do this.
1547 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1548 error = xfs_qm_qino_alloc(mp, &uip,
1549 flags | XFS_QMOPT_UQUOTA);
1550 if (error)
1551 goto error_rele;
1553 flags &= ~XFS_QMOPT_SBVERSION;
1555 if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
1556 error = xfs_qm_qino_alloc(mp, &gip,
1557 flags | XFS_QMOPT_GQUOTA);
1558 if (error)
1559 goto error_rele;
1561 flags &= ~XFS_QMOPT_SBVERSION;
1563 if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) {
1564 error = xfs_qm_qino_alloc(mp, &pip,
1565 flags | XFS_QMOPT_PQUOTA);
1566 if (error)
1567 goto error_rele;
1570 mp->m_quotainfo->qi_uquotaip = uip;
1571 mp->m_quotainfo->qi_gquotaip = gip;
1572 mp->m_quotainfo->qi_pquotaip = pip;
1574 return 0;
1576 error_rele:
1577 if (uip)
1578 xfs_irele(uip);
1579 if (gip)
1580 xfs_irele(gip);
1581 if (pip)
1582 xfs_irele(pip);
1583 return error;
1586 STATIC void
1587 xfs_qm_destroy_quotainos(
1588 xfs_quotainfo_t *qi)
1590 if (qi->qi_uquotaip) {
1591 xfs_irele(qi->qi_uquotaip);
1592 qi->qi_uquotaip = NULL; /* paranoia */
1594 if (qi->qi_gquotaip) {
1595 xfs_irele(qi->qi_gquotaip);
1596 qi->qi_gquotaip = NULL;
1598 if (qi->qi_pquotaip) {
1599 xfs_irele(qi->qi_pquotaip);
1600 qi->qi_pquotaip = NULL;
1604 STATIC void
1605 xfs_qm_dqfree_one(
1606 struct xfs_dquot *dqp)
1608 struct xfs_mount *mp = dqp->q_mount;
1609 struct xfs_quotainfo *qi = mp->m_quotainfo;
1611 mutex_lock(&qi->qi_tree_lock);
1612 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
1613 be32_to_cpu(dqp->q_core.d_id));
1615 qi->qi_dquots--;
1616 mutex_unlock(&qi->qi_tree_lock);
1618 xfs_qm_dqdestroy(dqp);
1621 /* --------------- utility functions for vnodeops ---------------- */
1625 * Given an inode, a uid, gid and prid make sure that we have
1626 * allocated relevant dquot(s) on disk, and that we won't exceed inode
1627 * quotas by creating this file.
1628 * This also attaches dquot(s) to the given inode after locking it,
1629 * and returns the dquots corresponding to the uid and/or gid.
1631 * in : inode (unlocked)
1632 * out : udquot, gdquot with references taken and unlocked
1635 xfs_qm_vop_dqalloc(
1636 struct xfs_inode *ip,
1637 xfs_dqid_t uid,
1638 xfs_dqid_t gid,
1639 prid_t prid,
1640 uint flags,
1641 struct xfs_dquot **O_udqpp,
1642 struct xfs_dquot **O_gdqpp,
1643 struct xfs_dquot **O_pdqpp)
1645 struct xfs_mount *mp = ip->i_mount;
1646 struct xfs_dquot *uq = NULL;
1647 struct xfs_dquot *gq = NULL;
1648 struct xfs_dquot *pq = NULL;
1649 int error;
1650 uint lockflags;
1652 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1653 return 0;
1655 lockflags = XFS_ILOCK_EXCL;
1656 xfs_ilock(ip, lockflags);
1658 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1659 gid = ip->i_d.di_gid;
1662 * Attach the dquot(s) to this inode, doing a dquot allocation
1663 * if necessary. The dquot(s) will not be locked.
1665 if (XFS_NOT_DQATTACHED(mp, ip)) {
1666 error = xfs_qm_dqattach_locked(ip, true);
1667 if (error) {
1668 xfs_iunlock(ip, lockflags);
1669 return error;
1673 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1674 if (ip->i_d.di_uid != uid) {
1676 * What we need is the dquot that has this uid, and
1677 * if we send the inode to dqget, the uid of the inode
1678 * takes priority over what's sent in the uid argument.
1679 * We must unlock inode here before calling dqget if
1680 * we're not sending the inode, because otherwise
1681 * we'll deadlock by doing trans_reserve while
1682 * holding ilock.
1684 xfs_iunlock(ip, lockflags);
1685 error = xfs_qm_dqget(mp, uid, XFS_DQ_USER, true, &uq);
1686 if (error) {
1687 ASSERT(error != -ENOENT);
1688 return error;
1691 * Get the ilock in the right order.
1693 xfs_dqunlock(uq);
1694 lockflags = XFS_ILOCK_SHARED;
1695 xfs_ilock(ip, lockflags);
1696 } else {
1698 * Take an extra reference, because we'll return
1699 * this to caller
1701 ASSERT(ip->i_udquot);
1702 uq = xfs_qm_dqhold(ip->i_udquot);
1705 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1706 if (ip->i_d.di_gid != gid) {
1707 xfs_iunlock(ip, lockflags);
1708 error = xfs_qm_dqget(mp, gid, XFS_DQ_GROUP, true, &gq);
1709 if (error) {
1710 ASSERT(error != -ENOENT);
1711 goto error_rele;
1713 xfs_dqunlock(gq);
1714 lockflags = XFS_ILOCK_SHARED;
1715 xfs_ilock(ip, lockflags);
1716 } else {
1717 ASSERT(ip->i_gdquot);
1718 gq = xfs_qm_dqhold(ip->i_gdquot);
1721 if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1722 if (xfs_get_projid(ip) != prid) {
1723 xfs_iunlock(ip, lockflags);
1724 error = xfs_qm_dqget(mp, (xfs_dqid_t)prid, XFS_DQ_PROJ,
1725 true, &pq);
1726 if (error) {
1727 ASSERT(error != -ENOENT);
1728 goto error_rele;
1730 xfs_dqunlock(pq);
1731 lockflags = XFS_ILOCK_SHARED;
1732 xfs_ilock(ip, lockflags);
1733 } else {
1734 ASSERT(ip->i_pdquot);
1735 pq = xfs_qm_dqhold(ip->i_pdquot);
1738 if (uq)
1739 trace_xfs_dquot_dqalloc(ip);
1741 xfs_iunlock(ip, lockflags);
1742 if (O_udqpp)
1743 *O_udqpp = uq;
1744 else
1745 xfs_qm_dqrele(uq);
1746 if (O_gdqpp)
1747 *O_gdqpp = gq;
1748 else
1749 xfs_qm_dqrele(gq);
1750 if (O_pdqpp)
1751 *O_pdqpp = pq;
1752 else
1753 xfs_qm_dqrele(pq);
1754 return 0;
1756 error_rele:
1757 xfs_qm_dqrele(gq);
1758 xfs_qm_dqrele(uq);
1759 return error;
1763 * Actually transfer ownership, and do dquot modifications.
1764 * These were already reserved.
1766 xfs_dquot_t *
1767 xfs_qm_vop_chown(
1768 xfs_trans_t *tp,
1769 xfs_inode_t *ip,
1770 xfs_dquot_t **IO_olddq,
1771 xfs_dquot_t *newdq)
1773 xfs_dquot_t *prevdq;
1774 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
1775 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1778 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1779 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1781 /* old dquot */
1782 prevdq = *IO_olddq;
1783 ASSERT(prevdq);
1784 ASSERT(prevdq != newdq);
1786 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
1787 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1789 /* the sparkling new dquot */
1790 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
1791 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1794 * Take an extra reference, because the inode is going to keep
1795 * this dquot pointer even after the trans_commit.
1797 *IO_olddq = xfs_qm_dqhold(newdq);
1799 return prevdq;
1803 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1806 xfs_qm_vop_chown_reserve(
1807 struct xfs_trans *tp,
1808 struct xfs_inode *ip,
1809 struct xfs_dquot *udqp,
1810 struct xfs_dquot *gdqp,
1811 struct xfs_dquot *pdqp,
1812 uint flags)
1814 struct xfs_mount *mp = ip->i_mount;
1815 uint delblks, blkflags, prjflags = 0;
1816 struct xfs_dquot *udq_unres = NULL;
1817 struct xfs_dquot *gdq_unres = NULL;
1818 struct xfs_dquot *pdq_unres = NULL;
1819 struct xfs_dquot *udq_delblks = NULL;
1820 struct xfs_dquot *gdq_delblks = NULL;
1821 struct xfs_dquot *pdq_delblks = NULL;
1822 int error;
1825 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1826 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1828 delblks = ip->i_delayed_blks;
1829 blkflags = XFS_IS_REALTIME_INODE(ip) ?
1830 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
1832 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
1833 ip->i_d.di_uid != be32_to_cpu(udqp->q_core.d_id)) {
1834 udq_delblks = udqp;
1836 * If there are delayed allocation blocks, then we have to
1837 * unreserve those from the old dquot, and add them to the
1838 * new dquot.
1840 if (delblks) {
1841 ASSERT(ip->i_udquot);
1842 udq_unres = ip->i_udquot;
1845 if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp &&
1846 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id)) {
1847 gdq_delblks = gdqp;
1848 if (delblks) {
1849 ASSERT(ip->i_gdquot);
1850 gdq_unres = ip->i_gdquot;
1854 if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
1855 xfs_get_projid(ip) != be32_to_cpu(pdqp->q_core.d_id)) {
1856 prjflags = XFS_QMOPT_ENOSPC;
1857 pdq_delblks = pdqp;
1858 if (delblks) {
1859 ASSERT(ip->i_pdquot);
1860 pdq_unres = ip->i_pdquot;
1864 error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
1865 udq_delblks, gdq_delblks, pdq_delblks,
1866 ip->i_d.di_nblocks, 1,
1867 flags | blkflags | prjflags);
1868 if (error)
1869 return error;
1872 * Do the delayed blks reservations/unreservations now. Since, these
1873 * are done without the help of a transaction, if a reservation fails
1874 * its previous reservations won't be automatically undone by trans
1875 * code. So, we have to do it manually here.
1877 if (delblks) {
1879 * Do the reservations first. Unreservation can't fail.
1881 ASSERT(udq_delblks || gdq_delblks || pdq_delblks);
1882 ASSERT(udq_unres || gdq_unres || pdq_unres);
1883 error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
1884 udq_delblks, gdq_delblks, pdq_delblks,
1885 (xfs_qcnt_t)delblks, 0,
1886 flags | blkflags | prjflags);
1887 if (error)
1888 return error;
1889 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
1890 udq_unres, gdq_unres, pdq_unres,
1891 -((xfs_qcnt_t)delblks), 0, blkflags);
1894 return 0;
1898 xfs_qm_vop_rename_dqattach(
1899 struct xfs_inode **i_tab)
1901 struct xfs_mount *mp = i_tab[0]->i_mount;
1902 int i;
1904 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1905 return 0;
1907 for (i = 0; (i < 4 && i_tab[i]); i++) {
1908 struct xfs_inode *ip = i_tab[i];
1909 int error;
1912 * Watch out for duplicate entries in the table.
1914 if (i == 0 || ip != i_tab[i-1]) {
1915 if (XFS_NOT_DQATTACHED(mp, ip)) {
1916 error = xfs_qm_dqattach(ip);
1917 if (error)
1918 return error;
1922 return 0;
1925 void
1926 xfs_qm_vop_create_dqattach(
1927 struct xfs_trans *tp,
1928 struct xfs_inode *ip,
1929 struct xfs_dquot *udqp,
1930 struct xfs_dquot *gdqp,
1931 struct xfs_dquot *pdqp)
1933 struct xfs_mount *mp = tp->t_mountp;
1935 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1936 return;
1938 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1939 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1941 if (udqp && XFS_IS_UQUOTA_ON(mp)) {
1942 ASSERT(ip->i_udquot == NULL);
1943 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
1945 ip->i_udquot = xfs_qm_dqhold(udqp);
1946 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
1948 if (gdqp && XFS_IS_GQUOTA_ON(mp)) {
1949 ASSERT(ip->i_gdquot == NULL);
1950 ASSERT(ip->i_d.di_gid == be32_to_cpu(gdqp->q_core.d_id));
1951 ip->i_gdquot = xfs_qm_dqhold(gdqp);
1952 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
1954 if (pdqp && XFS_IS_PQUOTA_ON(mp)) {
1955 ASSERT(ip->i_pdquot == NULL);
1956 ASSERT(xfs_get_projid(ip) == be32_to_cpu(pdqp->q_core.d_id));
1958 ip->i_pdquot = xfs_qm_dqhold(pdqp);
1959 xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1);