Merge tag 'regulator-fix-v5.14-rc2' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6-microblaze.git] / fs / xfs / xfs_qm.c
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_iwalk.h"
17 #include "xfs_quota.h"
18 #include "xfs_bmap.h"
19 #include "xfs_bmap_util.h"
20 #include "xfs_trans.h"
21 #include "xfs_trans_space.h"
22 #include "xfs_qm.h"
23 #include "xfs_trace.h"
24 #include "xfs_icache.h"
25 #include "xfs_error.h"
26 #include "xfs_ag.h"
27 #include "xfs_ialloc.h"
28
29 /*
30  * The global quota manager. There is only one of these for the entire
31  * system, _not_ one per file system. XQM keeps track of the overall
32  * quota functionality, including maintaining the freelist and hash
33  * tables of dquots.
34  */
35 STATIC int      xfs_qm_init_quotainos(struct xfs_mount *mp);
36 STATIC int      xfs_qm_init_quotainfo(struct xfs_mount *mp);
37
38 STATIC void     xfs_qm_destroy_quotainos(struct xfs_quotainfo *qi);
39 STATIC void     xfs_qm_dqfree_one(struct xfs_dquot *dqp);
40 /*
41  * We use the batch lookup interface to iterate over the dquots as it
42  * currently is the only interface into the radix tree code that allows
43  * fuzzy lookups instead of exact matches.  Holding the lock over multiple
44  * operations is fine as all callers are used either during mount/umount
45  * or quotaoff.
46  */
47 #define XFS_DQ_LOOKUP_BATCH     32
48
49 STATIC int
50 xfs_qm_dquot_walk(
51         struct xfs_mount        *mp,
52         xfs_dqtype_t            type,
53         int                     (*execute)(struct xfs_dquot *dqp, void *data),
54         void                    *data)
55 {
56         struct xfs_quotainfo    *qi = mp->m_quotainfo;
57         struct radix_tree_root  *tree = xfs_dquot_tree(qi, type);
58         uint32_t                next_index;
59         int                     last_error = 0;
60         int                     skipped;
61         int                     nr_found;
62
63 restart:
64         skipped = 0;
65         next_index = 0;
66         nr_found = 0;
67
68         while (1) {
69                 struct xfs_dquot *batch[XFS_DQ_LOOKUP_BATCH];
70                 int             error = 0;
71                 int             i;
72
73                 mutex_lock(&qi->qi_tree_lock);
74                 nr_found = radix_tree_gang_lookup(tree, (void **)batch,
75                                         next_index, XFS_DQ_LOOKUP_BATCH);
76                 if (!nr_found) {
77                         mutex_unlock(&qi->qi_tree_lock);
78                         break;
79                 }
80
81                 for (i = 0; i < nr_found; i++) {
82                         struct xfs_dquot *dqp = batch[i];
83
84                         next_index = dqp->q_id + 1;
85
86                         error = execute(batch[i], data);
87                         if (error == -EAGAIN) {
88                                 skipped++;
89                                 continue;
90                         }
91                         if (error && last_error != -EFSCORRUPTED)
92                                 last_error = error;
93                 }
94
95                 mutex_unlock(&qi->qi_tree_lock);
96
97                 /* bail out if the filesystem is corrupted.  */
98                 if (last_error == -EFSCORRUPTED) {
99                         skipped = 0;
100                         break;
101                 }
102                 /* we're done if id overflows back to zero */
103                 if (!next_index)
104                         break;
105         }
106
107         if (skipped) {
108                 delay(1);
109                 goto restart;
110         }
111
112         return last_error;
113 }
114
115
116 /*
117  * Purge a dquot from all tracking data structures and free it.
118  */
119 STATIC int
120 xfs_qm_dqpurge(
121         struct xfs_dquot        *dqp,
122         void                    *data)
123 {
124         struct xfs_mount        *mp = dqp->q_mount;
125         struct xfs_quotainfo    *qi = mp->m_quotainfo;
126         int                     error = -EAGAIN;
127
128         xfs_dqlock(dqp);
129         if ((dqp->q_flags & XFS_DQFLAG_FREEING) || dqp->q_nrefs != 0)
130                 goto out_unlock;
131
132         dqp->q_flags |= XFS_DQFLAG_FREEING;
133
134         xfs_dqflock(dqp);
135
136         /*
137          * If we are turning this type of quotas off, we don't care
138          * about the dirty metadata sitting in this dquot. OTOH, if
139          * we're unmounting, we do care, so we flush it and wait.
140          */
141         if (XFS_DQ_IS_DIRTY(dqp)) {
142                 struct xfs_buf  *bp = NULL;
143
144                 /*
145                  * We don't care about getting disk errors here. We need
146                  * to purge this dquot anyway, so we go ahead regardless.
147                  */
148                 error = xfs_qm_dqflush(dqp, &bp);
149                 if (!error) {
150                         error = xfs_bwrite(bp);
151                         xfs_buf_relse(bp);
152                 } else if (error == -EAGAIN) {
153                         dqp->q_flags &= ~XFS_DQFLAG_FREEING;
154                         goto out_unlock;
155                 }
156                 xfs_dqflock(dqp);
157         }
158
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));
162
163         xfs_dqfunlock(dqp);
164         xfs_dqunlock(dqp);
165
166         radix_tree_delete(xfs_dquot_tree(qi, xfs_dquot_type(dqp)), dqp->q_id);
167         qi->qi_dquots--;
168
169         /*
170          * We move dquots to the freelist as soon as their reference count
171          * hits zero, so it really should be on the freelist here.
172          */
173         ASSERT(!list_empty(&dqp->q_lru));
174         list_lru_del(&qi->qi_lru, &dqp->q_lru);
175         XFS_STATS_DEC(mp, xs_qm_dquot_unused);
176
177         xfs_qm_dqdestroy(dqp);
178         return 0;
179
180 out_unlock:
181         xfs_dqunlock(dqp);
182         return error;
183 }
184
185 /*
186  * Purge the dquot cache.
187  */
188 void
189 xfs_qm_dqpurge_all(
190         struct xfs_mount        *mp,
191         uint                    flags)
192 {
193         if (flags & XFS_QMOPT_UQUOTA)
194                 xfs_qm_dquot_walk(mp, XFS_DQTYPE_USER, xfs_qm_dqpurge, NULL);
195         if (flags & XFS_QMOPT_GQUOTA)
196                 xfs_qm_dquot_walk(mp, XFS_DQTYPE_GROUP, xfs_qm_dqpurge, NULL);
197         if (flags & XFS_QMOPT_PQUOTA)
198                 xfs_qm_dquot_walk(mp, XFS_DQTYPE_PROJ, xfs_qm_dqpurge, NULL);
199 }
200
201 /*
202  * Just destroy the quotainfo structure.
203  */
204 void
205 xfs_qm_unmount(
206         struct xfs_mount        *mp)
207 {
208         if (mp->m_quotainfo) {
209                 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
210                 xfs_qm_destroy_quotainfo(mp);
211         }
212 }
213
214 /*
215  * Called from the vfsops layer.
216  */
217 void
218 xfs_qm_unmount_quotas(
219         xfs_mount_t     *mp)
220 {
221         /*
222          * Release the dquots that root inode, et al might be holding,
223          * before we flush quotas and blow away the quotainfo structure.
224          */
225         ASSERT(mp->m_rootip);
226         xfs_qm_dqdetach(mp->m_rootip);
227         if (mp->m_rbmip)
228                 xfs_qm_dqdetach(mp->m_rbmip);
229         if (mp->m_rsumip)
230                 xfs_qm_dqdetach(mp->m_rsumip);
231
232         /*
233          * Release the quota inodes.
234          */
235         if (mp->m_quotainfo) {
236                 if (mp->m_quotainfo->qi_uquotaip) {
237                         xfs_irele(mp->m_quotainfo->qi_uquotaip);
238                         mp->m_quotainfo->qi_uquotaip = NULL;
239                 }
240                 if (mp->m_quotainfo->qi_gquotaip) {
241                         xfs_irele(mp->m_quotainfo->qi_gquotaip);
242                         mp->m_quotainfo->qi_gquotaip = NULL;
243                 }
244                 if (mp->m_quotainfo->qi_pquotaip) {
245                         xfs_irele(mp->m_quotainfo->qi_pquotaip);
246                         mp->m_quotainfo->qi_pquotaip = NULL;
247                 }
248         }
249 }
250
251 STATIC int
252 xfs_qm_dqattach_one(
253         struct xfs_inode        *ip,
254         xfs_dqtype_t            type,
255         bool                    doalloc,
256         struct xfs_dquot        **IO_idqpp)
257 {
258         struct xfs_dquot        *dqp;
259         int                     error;
260
261         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
262         error = 0;
263
264         /*
265          * See if we already have it in the inode itself. IO_idqpp is &i_udquot
266          * or &i_gdquot. This made the code look weird, but made the logic a lot
267          * simpler.
268          */
269         dqp = *IO_idqpp;
270         if (dqp) {
271                 trace_xfs_dqattach_found(dqp);
272                 return 0;
273         }
274
275         /*
276          * Find the dquot from somewhere. This bumps the reference count of
277          * dquot and returns it locked.  This can return ENOENT if dquot didn't
278          * exist on disk and we didn't ask it to allocate; ESRCH if quotas got
279          * turned off suddenly.
280          */
281         error = xfs_qm_dqget_inode(ip, type, doalloc, &dqp);
282         if (error)
283                 return error;
284
285         trace_xfs_dqattach_get(dqp);
286
287         /*
288          * dqget may have dropped and re-acquired the ilock, but it guarantees
289          * that the dquot returned is the one that should go in the inode.
290          */
291         *IO_idqpp = dqp;
292         xfs_dqunlock(dqp);
293         return 0;
294 }
295
296 static bool
297 xfs_qm_need_dqattach(
298         struct xfs_inode        *ip)
299 {
300         struct xfs_mount        *mp = ip->i_mount;
301
302         if (!XFS_IS_QUOTA_RUNNING(mp))
303                 return false;
304         if (!XFS_IS_QUOTA_ON(mp))
305                 return false;
306         if (!XFS_NOT_DQATTACHED(mp, ip))
307                 return false;
308         if (xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
309                 return false;
310         return true;
311 }
312
313 /*
314  * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
315  * into account.
316  * If @doalloc is true, the dquot(s) will be allocated if needed.
317  * Inode may get unlocked and relocked in here, and the caller must deal with
318  * the consequences.
319  */
320 int
321 xfs_qm_dqattach_locked(
322         xfs_inode_t     *ip,
323         bool            doalloc)
324 {
325         xfs_mount_t     *mp = ip->i_mount;
326         int             error = 0;
327
328         if (!xfs_qm_need_dqattach(ip))
329                 return 0;
330
331         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
332
333         if (XFS_IS_UQUOTA_ON(mp) && !ip->i_udquot) {
334                 error = xfs_qm_dqattach_one(ip, XFS_DQTYPE_USER,
335                                 doalloc, &ip->i_udquot);
336                 if (error)
337                         goto done;
338                 ASSERT(ip->i_udquot);
339         }
340
341         if (XFS_IS_GQUOTA_ON(mp) && !ip->i_gdquot) {
342                 error = xfs_qm_dqattach_one(ip, XFS_DQTYPE_GROUP,
343                                 doalloc, &ip->i_gdquot);
344                 if (error)
345                         goto done;
346                 ASSERT(ip->i_gdquot);
347         }
348
349         if (XFS_IS_PQUOTA_ON(mp) && !ip->i_pdquot) {
350                 error = xfs_qm_dqattach_one(ip, XFS_DQTYPE_PROJ,
351                                 doalloc, &ip->i_pdquot);
352                 if (error)
353                         goto done;
354                 ASSERT(ip->i_pdquot);
355         }
356
357 done:
358         /*
359          * Don't worry about the dquots that we may have attached before any
360          * error - they'll get detached later if it has not already been done.
361          */
362         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
363         return error;
364 }
365
366 int
367 xfs_qm_dqattach(
368         struct xfs_inode        *ip)
369 {
370         int                     error;
371
372         if (!xfs_qm_need_dqattach(ip))
373                 return 0;
374
375         xfs_ilock(ip, XFS_ILOCK_EXCL);
376         error = xfs_qm_dqattach_locked(ip, false);
377         xfs_iunlock(ip, XFS_ILOCK_EXCL);
378
379         return error;
380 }
381
382 /*
383  * Release dquots (and their references) if any.
384  * The inode should be locked EXCL except when this's called by
385  * xfs_ireclaim.
386  */
387 void
388 xfs_qm_dqdetach(
389         xfs_inode_t     *ip)
390 {
391         if (!(ip->i_udquot || ip->i_gdquot || ip->i_pdquot))
392                 return;
393
394         trace_xfs_dquot_dqdetach(ip);
395
396         ASSERT(!xfs_is_quota_inode(&ip->i_mount->m_sb, ip->i_ino));
397         if (ip->i_udquot) {
398                 xfs_qm_dqrele(ip->i_udquot);
399                 ip->i_udquot = NULL;
400         }
401         if (ip->i_gdquot) {
402                 xfs_qm_dqrele(ip->i_gdquot);
403                 ip->i_gdquot = NULL;
404         }
405         if (ip->i_pdquot) {
406                 xfs_qm_dqrele(ip->i_pdquot);
407                 ip->i_pdquot = NULL;
408         }
409 }
410
411 struct xfs_qm_isolate {
412         struct list_head        buffers;
413         struct list_head        dispose;
414 };
415
416 static enum lru_status
417 xfs_qm_dquot_isolate(
418         struct list_head        *item,
419         struct list_lru_one     *lru,
420         spinlock_t              *lru_lock,
421         void                    *arg)
422                 __releases(lru_lock) __acquires(lru_lock)
423 {
424         struct xfs_dquot        *dqp = container_of(item,
425                                                 struct xfs_dquot, q_lru);
426         struct xfs_qm_isolate   *isol = arg;
427
428         if (!xfs_dqlock_nowait(dqp))
429                 goto out_miss_busy;
430
431         /*
432          * This dquot has acquired a reference in the meantime remove it from
433          * the freelist and try again.
434          */
435         if (dqp->q_nrefs) {
436                 xfs_dqunlock(dqp);
437                 XFS_STATS_INC(dqp->q_mount, xs_qm_dqwants);
438
439                 trace_xfs_dqreclaim_want(dqp);
440                 list_lru_isolate(lru, &dqp->q_lru);
441                 XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot_unused);
442                 return LRU_REMOVED;
443         }
444
445         /*
446          * If the dquot is dirty, flush it. If it's already being flushed, just
447          * skip it so there is time for the IO to complete before we try to
448          * reclaim it again on the next LRU pass.
449          */
450         if (!xfs_dqflock_nowait(dqp)) {
451                 xfs_dqunlock(dqp);
452                 goto out_miss_busy;
453         }
454
455         if (XFS_DQ_IS_DIRTY(dqp)) {
456                 struct xfs_buf  *bp = NULL;
457                 int             error;
458
459                 trace_xfs_dqreclaim_dirty(dqp);
460
461                 /* we have to drop the LRU lock to flush the dquot */
462                 spin_unlock(lru_lock);
463
464                 error = xfs_qm_dqflush(dqp, &bp);
465                 if (error)
466                         goto out_unlock_dirty;
467
468                 xfs_buf_delwri_queue(bp, &isol->buffers);
469                 xfs_buf_relse(bp);
470                 goto out_unlock_dirty;
471         }
472         xfs_dqfunlock(dqp);
473
474         /*
475          * Prevent lookups now that we are past the point of no return.
476          */
477         dqp->q_flags |= XFS_DQFLAG_FREEING;
478         xfs_dqunlock(dqp);
479
480         ASSERT(dqp->q_nrefs == 0);
481         list_lru_isolate_move(lru, &dqp->q_lru, &isol->dispose);
482         XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot_unused);
483         trace_xfs_dqreclaim_done(dqp);
484         XFS_STATS_INC(dqp->q_mount, xs_qm_dqreclaims);
485         return LRU_REMOVED;
486
487 out_miss_busy:
488         trace_xfs_dqreclaim_busy(dqp);
489         XFS_STATS_INC(dqp->q_mount, xs_qm_dqreclaim_misses);
490         return LRU_SKIP;
491
492 out_unlock_dirty:
493         trace_xfs_dqreclaim_busy(dqp);
494         XFS_STATS_INC(dqp->q_mount, xs_qm_dqreclaim_misses);
495         xfs_dqunlock(dqp);
496         spin_lock(lru_lock);
497         return LRU_RETRY;
498 }
499
500 static unsigned long
501 xfs_qm_shrink_scan(
502         struct shrinker         *shrink,
503         struct shrink_control   *sc)
504 {
505         struct xfs_quotainfo    *qi = container_of(shrink,
506                                         struct xfs_quotainfo, qi_shrinker);
507         struct xfs_qm_isolate   isol;
508         unsigned long           freed;
509         int                     error;
510
511         if ((sc->gfp_mask & (__GFP_FS|__GFP_DIRECT_RECLAIM)) != (__GFP_FS|__GFP_DIRECT_RECLAIM))
512                 return 0;
513
514         INIT_LIST_HEAD(&isol.buffers);
515         INIT_LIST_HEAD(&isol.dispose);
516
517         freed = list_lru_shrink_walk(&qi->qi_lru, sc,
518                                      xfs_qm_dquot_isolate, &isol);
519
520         error = xfs_buf_delwri_submit(&isol.buffers);
521         if (error)
522                 xfs_warn(NULL, "%s: dquot reclaim failed", __func__);
523
524         while (!list_empty(&isol.dispose)) {
525                 struct xfs_dquot        *dqp;
526
527                 dqp = list_first_entry(&isol.dispose, struct xfs_dquot, q_lru);
528                 list_del_init(&dqp->q_lru);
529                 xfs_qm_dqfree_one(dqp);
530         }
531
532         return freed;
533 }
534
535 static unsigned long
536 xfs_qm_shrink_count(
537         struct shrinker         *shrink,
538         struct shrink_control   *sc)
539 {
540         struct xfs_quotainfo    *qi = container_of(shrink,
541                                         struct xfs_quotainfo, qi_shrinker);
542
543         return list_lru_shrink_count(&qi->qi_lru, sc);
544 }
545
546 STATIC void
547 xfs_qm_set_defquota(
548         struct xfs_mount        *mp,
549         xfs_dqtype_t            type,
550         struct xfs_quotainfo    *qinf)
551 {
552         struct xfs_dquot        *dqp;
553         struct xfs_def_quota    *defq;
554         int                     error;
555
556         error = xfs_qm_dqget_uncached(mp, 0, type, &dqp);
557         if (error)
558                 return;
559
560         defq = xfs_get_defquota(qinf, xfs_dquot_type(dqp));
561
562         /*
563          * Timers and warnings have been already set, let's just set the
564          * default limits for this quota type
565          */
566         defq->blk.hard = dqp->q_blk.hardlimit;
567         defq->blk.soft = dqp->q_blk.softlimit;
568         defq->ino.hard = dqp->q_ino.hardlimit;
569         defq->ino.soft = dqp->q_ino.softlimit;
570         defq->rtb.hard = dqp->q_rtb.hardlimit;
571         defq->rtb.soft = dqp->q_rtb.softlimit;
572         xfs_qm_dqdestroy(dqp);
573 }
574
575 /* Initialize quota time limits from the root dquot. */
576 static void
577 xfs_qm_init_timelimits(
578         struct xfs_mount        *mp,
579         xfs_dqtype_t            type)
580 {
581         struct xfs_quotainfo    *qinf = mp->m_quotainfo;
582         struct xfs_def_quota    *defq;
583         struct xfs_dquot        *dqp;
584         int                     error;
585
586         defq = xfs_get_defquota(qinf, type);
587
588         defq->blk.time = XFS_QM_BTIMELIMIT;
589         defq->ino.time = XFS_QM_ITIMELIMIT;
590         defq->rtb.time = XFS_QM_RTBTIMELIMIT;
591         defq->blk.warn = XFS_QM_BWARNLIMIT;
592         defq->ino.warn = XFS_QM_IWARNLIMIT;
593         defq->rtb.warn = XFS_QM_RTBWARNLIMIT;
594
595         /*
596          * We try to get the limits from the superuser's limits fields.
597          * This is quite hacky, but it is standard quota practice.
598          *
599          * Since we may not have done a quotacheck by this point, just read
600          * the dquot without attaching it to any hashtables or lists.
601          */
602         error = xfs_qm_dqget_uncached(mp, 0, type, &dqp);
603         if (error)
604                 return;
605
606         /*
607          * The warnings and timers set the grace period given to
608          * a user or group before he or she can not perform any
609          * more writing. If it is zero, a default is used.
610          */
611         if (dqp->q_blk.timer)
612                 defq->blk.time = dqp->q_blk.timer;
613         if (dqp->q_ino.timer)
614                 defq->ino.time = dqp->q_ino.timer;
615         if (dqp->q_rtb.timer)
616                 defq->rtb.time = dqp->q_rtb.timer;
617         if (dqp->q_blk.warnings)
618                 defq->blk.warn = dqp->q_blk.warnings;
619         if (dqp->q_ino.warnings)
620                 defq->ino.warn = dqp->q_ino.warnings;
621         if (dqp->q_rtb.warnings)
622                 defq->rtb.warn = dqp->q_rtb.warnings;
623
624         xfs_qm_dqdestroy(dqp);
625 }
626
627 /*
628  * This initializes all the quota information that's kept in the
629  * mount structure
630  */
631 STATIC int
632 xfs_qm_init_quotainfo(
633         struct xfs_mount        *mp)
634 {
635         struct xfs_quotainfo    *qinf;
636         int                     error;
637
638         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
639
640         qinf = mp->m_quotainfo = kmem_zalloc(sizeof(struct xfs_quotainfo), 0);
641
642         error = list_lru_init(&qinf->qi_lru);
643         if (error)
644                 goto out_free_qinf;
645
646         /*
647          * See if quotainodes are setup, and if not, allocate them,
648          * and change the superblock accordingly.
649          */
650         error = xfs_qm_init_quotainos(mp);
651         if (error)
652                 goto out_free_lru;
653
654         INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
655         INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
656         INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS);
657         mutex_init(&qinf->qi_tree_lock);
658
659         /* mutex used to serialize quotaoffs */
660         mutex_init(&qinf->qi_quotaofflock);
661
662         /* Precalc some constants */
663         qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
664         qinf->qi_dqperchunk = xfs_calc_dquots_per_chunk(qinf->qi_dqchunklen);
665         if (xfs_sb_version_hasbigtime(&mp->m_sb)) {
666                 qinf->qi_expiry_min =
667                         xfs_dq_bigtime_to_unix(XFS_DQ_BIGTIME_EXPIRY_MIN);
668                 qinf->qi_expiry_max =
669                         xfs_dq_bigtime_to_unix(XFS_DQ_BIGTIME_EXPIRY_MAX);
670         } else {
671                 qinf->qi_expiry_min = XFS_DQ_LEGACY_EXPIRY_MIN;
672                 qinf->qi_expiry_max = XFS_DQ_LEGACY_EXPIRY_MAX;
673         }
674         trace_xfs_quota_expiry_range(mp, qinf->qi_expiry_min,
675                         qinf->qi_expiry_max);
676
677         mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
678
679         xfs_qm_init_timelimits(mp, XFS_DQTYPE_USER);
680         xfs_qm_init_timelimits(mp, XFS_DQTYPE_GROUP);
681         xfs_qm_init_timelimits(mp, XFS_DQTYPE_PROJ);
682
683         if (XFS_IS_UQUOTA_RUNNING(mp))
684                 xfs_qm_set_defquota(mp, XFS_DQTYPE_USER, qinf);
685         if (XFS_IS_GQUOTA_RUNNING(mp))
686                 xfs_qm_set_defquota(mp, XFS_DQTYPE_GROUP, qinf);
687         if (XFS_IS_PQUOTA_RUNNING(mp))
688                 xfs_qm_set_defquota(mp, XFS_DQTYPE_PROJ, qinf);
689
690         qinf->qi_shrinker.count_objects = xfs_qm_shrink_count;
691         qinf->qi_shrinker.scan_objects = xfs_qm_shrink_scan;
692         qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
693         qinf->qi_shrinker.flags = SHRINKER_NUMA_AWARE;
694
695         error = register_shrinker(&qinf->qi_shrinker);
696         if (error)
697                 goto out_free_inos;
698
699         return 0;
700
701 out_free_inos:
702         mutex_destroy(&qinf->qi_quotaofflock);
703         mutex_destroy(&qinf->qi_tree_lock);
704         xfs_qm_destroy_quotainos(qinf);
705 out_free_lru:
706         list_lru_destroy(&qinf->qi_lru);
707 out_free_qinf:
708         kmem_free(qinf);
709         mp->m_quotainfo = NULL;
710         return error;
711 }
712
713 /*
714  * Gets called when unmounting a filesystem or when all quotas get
715  * turned off.
716  * This purges the quota inodes, destroys locks and frees itself.
717  */
718 void
719 xfs_qm_destroy_quotainfo(
720         struct xfs_mount        *mp)
721 {
722         struct xfs_quotainfo    *qi;
723
724         qi = mp->m_quotainfo;
725         ASSERT(qi != NULL);
726
727         unregister_shrinker(&qi->qi_shrinker);
728         list_lru_destroy(&qi->qi_lru);
729         xfs_qm_destroy_quotainos(qi);
730         mutex_destroy(&qi->qi_tree_lock);
731         mutex_destroy(&qi->qi_quotaofflock);
732         kmem_free(qi);
733         mp->m_quotainfo = NULL;
734 }
735
736 /*
737  * Create an inode and return with a reference already taken, but unlocked
738  * This is how we create quota inodes
739  */
740 STATIC int
741 xfs_qm_qino_alloc(
742         struct xfs_mount        *mp,
743         struct xfs_inode        **ipp,
744         unsigned int            flags)
745 {
746         struct xfs_trans        *tp;
747         int                     error;
748         bool                    need_alloc = true;
749
750         *ipp = NULL;
751         /*
752          * With superblock that doesn't have separate pquotino, we
753          * share an inode between gquota and pquota. If the on-disk
754          * superblock has GQUOTA and the filesystem is now mounted
755          * with PQUOTA, just use sb_gquotino for sb_pquotino and
756          * vice-versa.
757          */
758         if (!xfs_sb_version_has_pquotino(&mp->m_sb) &&
759                         (flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) {
760                 xfs_ino_t ino = NULLFSINO;
761
762                 if ((flags & XFS_QMOPT_PQUOTA) &&
763                              (mp->m_sb.sb_gquotino != NULLFSINO)) {
764                         ino = mp->m_sb.sb_gquotino;
765                         if (XFS_IS_CORRUPT(mp,
766                                            mp->m_sb.sb_pquotino != NULLFSINO))
767                                 return -EFSCORRUPTED;
768                 } else if ((flags & XFS_QMOPT_GQUOTA) &&
769                              (mp->m_sb.sb_pquotino != NULLFSINO)) {
770                         ino = mp->m_sb.sb_pquotino;
771                         if (XFS_IS_CORRUPT(mp,
772                                            mp->m_sb.sb_gquotino != NULLFSINO))
773                                 return -EFSCORRUPTED;
774                 }
775                 if (ino != NULLFSINO) {
776                         error = xfs_iget(mp, NULL, ino, 0, 0, ipp);
777                         if (error)
778                                 return error;
779                         mp->m_sb.sb_gquotino = NULLFSINO;
780                         mp->m_sb.sb_pquotino = NULLFSINO;
781                         need_alloc = false;
782                 }
783         }
784
785         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_create,
786                         need_alloc ? XFS_QM_QINOCREATE_SPACE_RES(mp) : 0,
787                         0, 0, &tp);
788         if (error)
789                 return error;
790
791         if (need_alloc) {
792                 xfs_ino_t       ino;
793
794                 error = xfs_dialloc(&tp, 0, S_IFREG, &ino);
795                 if (!error)
796                         error = xfs_init_new_inode(&init_user_ns, tp, NULL, ino,
797                                         S_IFREG, 1, 0, 0, false, ipp);
798                 if (error) {
799                         xfs_trans_cancel(tp);
800                         return error;
801                 }
802         }
803
804         /*
805          * Make the changes in the superblock, and log those too.
806          * sbfields arg may contain fields other than *QUOTINO;
807          * VERSIONNUM for example.
808          */
809         spin_lock(&mp->m_sb_lock);
810         if (flags & XFS_QMOPT_SBVERSION) {
811                 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
812
813                 xfs_sb_version_addquota(&mp->m_sb);
814                 mp->m_sb.sb_uquotino = NULLFSINO;
815                 mp->m_sb.sb_gquotino = NULLFSINO;
816                 mp->m_sb.sb_pquotino = NULLFSINO;
817
818                 /* qflags will get updated fully _after_ quotacheck */
819                 mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT;
820         }
821         if (flags & XFS_QMOPT_UQUOTA)
822                 mp->m_sb.sb_uquotino = (*ipp)->i_ino;
823         else if (flags & XFS_QMOPT_GQUOTA)
824                 mp->m_sb.sb_gquotino = (*ipp)->i_ino;
825         else
826                 mp->m_sb.sb_pquotino = (*ipp)->i_ino;
827         spin_unlock(&mp->m_sb_lock);
828         xfs_log_sb(tp);
829
830         error = xfs_trans_commit(tp);
831         if (error) {
832                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
833                 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
834         }
835         if (need_alloc)
836                 xfs_finish_inode_setup(*ipp);
837         return error;
838 }
839
840
841 STATIC void
842 xfs_qm_reset_dqcounts(
843         struct xfs_mount        *mp,
844         struct xfs_buf          *bp,
845         xfs_dqid_t              id,
846         xfs_dqtype_t            type)
847 {
848         struct xfs_dqblk        *dqb;
849         int                     j;
850
851         trace_xfs_reset_dqcounts(bp, _RET_IP_);
852
853         /*
854          * Reset all counters and timers. They'll be
855          * started afresh by xfs_qm_quotacheck.
856          */
857 #ifdef DEBUG
858         j = (int)XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB) /
859                 sizeof(xfs_dqblk_t);
860         ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
861 #endif
862         dqb = bp->b_addr;
863         for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
864                 struct xfs_disk_dquot   *ddq;
865
866                 ddq = (struct xfs_disk_dquot *)&dqb[j];
867
868                 /*
869                  * Do a sanity check, and if needed, repair the dqblk. Don't
870                  * output any warnings because it's perfectly possible to
871                  * find uninitialised dquot blks. See comment in
872                  * xfs_dquot_verify.
873                  */
874                 if (xfs_dqblk_verify(mp, &dqb[j], id + j) ||
875                     (dqb[j].dd_diskdq.d_type & XFS_DQTYPE_REC_MASK) != type)
876                         xfs_dqblk_repair(mp, &dqb[j], id + j, type);
877
878                 /*
879                  * Reset type in case we are reusing group quota file for
880                  * project quotas or vice versa
881                  */
882                 ddq->d_type = type;
883                 ddq->d_bcount = 0;
884                 ddq->d_icount = 0;
885                 ddq->d_rtbcount = 0;
886
887                 /*
888                  * dquot id 0 stores the default grace period and the maximum
889                  * warning limit that were set by the administrator, so we
890                  * should not reset them.
891                  */
892                 if (ddq->d_id != 0) {
893                         ddq->d_btimer = 0;
894                         ddq->d_itimer = 0;
895                         ddq->d_rtbtimer = 0;
896                         ddq->d_bwarns = 0;
897                         ddq->d_iwarns = 0;
898                         ddq->d_rtbwarns = 0;
899                         if (xfs_sb_version_hasbigtime(&mp->m_sb))
900                                 ddq->d_type |= XFS_DQTYPE_BIGTIME;
901                 }
902
903                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
904                         xfs_update_cksum((char *)&dqb[j],
905                                          sizeof(struct xfs_dqblk),
906                                          XFS_DQUOT_CRC_OFF);
907                 }
908         }
909 }
910
911 STATIC int
912 xfs_qm_reset_dqcounts_all(
913         struct xfs_mount        *mp,
914         xfs_dqid_t              firstid,
915         xfs_fsblock_t           bno,
916         xfs_filblks_t           blkcnt,
917         xfs_dqtype_t            type,
918         struct list_head        *buffer_list)
919 {
920         struct xfs_buf          *bp;
921         int                     error = 0;
922
923         ASSERT(blkcnt > 0);
924
925         /*
926          * Blkcnt arg can be a very big number, and might even be
927          * larger than the log itself. So, we have to break it up into
928          * manageable-sized transactions.
929          * Note that we don't start a permanent transaction here; we might
930          * not be able to get a log reservation for the whole thing up front,
931          * and we don't really care to either, because we just discard
932          * everything if we were to crash in the middle of this loop.
933          */
934         while (blkcnt--) {
935                 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
936                               XFS_FSB_TO_DADDR(mp, bno),
937                               mp->m_quotainfo->qi_dqchunklen, 0, &bp,
938                               &xfs_dquot_buf_ops);
939
940                 /*
941                  * CRC and validation errors will return a EFSCORRUPTED here. If
942                  * this occurs, re-read without CRC validation so that we can
943                  * repair the damage via xfs_qm_reset_dqcounts(). This process
944                  * will leave a trace in the log indicating corruption has
945                  * been detected.
946                  */
947                 if (error == -EFSCORRUPTED) {
948                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
949                                       XFS_FSB_TO_DADDR(mp, bno),
950                                       mp->m_quotainfo->qi_dqchunklen, 0, &bp,
951                                       NULL);
952                 }
953
954                 if (error)
955                         break;
956
957                 /*
958                  * A corrupt buffer might not have a verifier attached, so
959                  * make sure we have the correct one attached before writeback
960                  * occurs.
961                  */
962                 bp->b_ops = &xfs_dquot_buf_ops;
963                 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
964                 xfs_buf_delwri_queue(bp, buffer_list);
965                 xfs_buf_relse(bp);
966
967                 /* goto the next block. */
968                 bno++;
969                 firstid += mp->m_quotainfo->qi_dqperchunk;
970         }
971
972         return error;
973 }
974
975 /*
976  * Iterate over all allocated dquot blocks in this quota inode, zeroing all
977  * counters for every chunk of dquots that we find.
978  */
979 STATIC int
980 xfs_qm_reset_dqcounts_buf(
981         struct xfs_mount        *mp,
982         struct xfs_inode        *qip,
983         xfs_dqtype_t            type,
984         struct list_head        *buffer_list)
985 {
986         struct xfs_bmbt_irec    *map;
987         int                     i, nmaps;       /* number of map entries */
988         int                     error;          /* return value */
989         xfs_fileoff_t           lblkno;
990         xfs_filblks_t           maxlblkcnt;
991         xfs_dqid_t              firstid;
992         xfs_fsblock_t           rablkno;
993         xfs_filblks_t           rablkcnt;
994
995         error = 0;
996         /*
997          * This looks racy, but we can't keep an inode lock across a
998          * trans_reserve. But, this gets called during quotacheck, and that
999          * happens only at mount time which is single threaded.
1000          */
1001         if (qip->i_nblocks == 0)
1002                 return 0;
1003
1004         map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), 0);
1005
1006         lblkno = 0;
1007         maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1008         do {
1009                 uint            lock_mode;
1010
1011                 nmaps = XFS_DQITER_MAP_SIZE;
1012                 /*
1013                  * We aren't changing the inode itself. Just changing
1014                  * some of its data. No new blocks are added here, and
1015                  * the inode is never added to the transaction.
1016                  */
1017                 lock_mode = xfs_ilock_data_map_shared(qip);
1018                 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
1019                                        map, &nmaps, 0);
1020                 xfs_iunlock(qip, lock_mode);
1021                 if (error)
1022                         break;
1023
1024                 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1025                 for (i = 0; i < nmaps; i++) {
1026                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1027                         ASSERT(map[i].br_blockcount);
1028
1029
1030                         lblkno += map[i].br_blockcount;
1031
1032                         if (map[i].br_startblock == HOLESTARTBLOCK)
1033                                 continue;
1034
1035                         firstid = (xfs_dqid_t) map[i].br_startoff *
1036                                 mp->m_quotainfo->qi_dqperchunk;
1037                         /*
1038                          * Do a read-ahead on the next extent.
1039                          */
1040                         if ((i+1 < nmaps) &&
1041                             (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1042                                 rablkcnt =  map[i+1].br_blockcount;
1043                                 rablkno = map[i+1].br_startblock;
1044                                 while (rablkcnt--) {
1045                                         xfs_buf_readahead(mp->m_ddev_targp,
1046                                                XFS_FSB_TO_DADDR(mp, rablkno),
1047                                                mp->m_quotainfo->qi_dqchunklen,
1048                                                &xfs_dquot_buf_ops);
1049                                         rablkno++;
1050                                 }
1051                         }
1052                         /*
1053                          * Iterate thru all the blks in the extent and
1054                          * reset the counters of all the dquots inside them.
1055                          */
1056                         error = xfs_qm_reset_dqcounts_all(mp, firstid,
1057                                                    map[i].br_startblock,
1058                                                    map[i].br_blockcount,
1059                                                    type, buffer_list);
1060                         if (error)
1061                                 goto out;
1062                 }
1063         } while (nmaps > 0);
1064
1065 out:
1066         kmem_free(map);
1067         return error;
1068 }
1069
1070 /*
1071  * Called by dqusage_adjust in doing a quotacheck.
1072  *
1073  * Given the inode, and a dquot id this updates both the incore dqout as well
1074  * as the buffer copy. This is so that once the quotacheck is done, we can
1075  * just log all the buffers, as opposed to logging numerous updates to
1076  * individual dquots.
1077  */
1078 STATIC int
1079 xfs_qm_quotacheck_dqadjust(
1080         struct xfs_inode        *ip,
1081         xfs_dqtype_t            type,
1082         xfs_qcnt_t              nblks,
1083         xfs_qcnt_t              rtblks)
1084 {
1085         struct xfs_mount        *mp = ip->i_mount;
1086         struct xfs_dquot        *dqp;
1087         xfs_dqid_t              id;
1088         int                     error;
1089
1090         id = xfs_qm_id_for_quotatype(ip, type);
1091         error = xfs_qm_dqget(mp, id, type, true, &dqp);
1092         if (error) {
1093                 /*
1094                  * Shouldn't be able to turn off quotas here.
1095                  */
1096                 ASSERT(error != -ESRCH);
1097                 ASSERT(error != -ENOENT);
1098                 return error;
1099         }
1100
1101         trace_xfs_dqadjust(dqp);
1102
1103         /*
1104          * Adjust the inode count and the block count to reflect this inode's
1105          * resource usage.
1106          */
1107         dqp->q_ino.count++;
1108         dqp->q_ino.reserved++;
1109         if (nblks) {
1110                 dqp->q_blk.count += nblks;
1111                 dqp->q_blk.reserved += nblks;
1112         }
1113         if (rtblks) {
1114                 dqp->q_rtb.count += rtblks;
1115                 dqp->q_rtb.reserved += rtblks;
1116         }
1117
1118         /*
1119          * Set default limits, adjust timers (since we changed usages)
1120          *
1121          * There are no timers for the default values set in the root dquot.
1122          */
1123         if (dqp->q_id) {
1124                 xfs_qm_adjust_dqlimits(dqp);
1125                 xfs_qm_adjust_dqtimers(dqp);
1126         }
1127
1128         dqp->q_flags |= XFS_DQFLAG_DIRTY;
1129         xfs_qm_dqput(dqp);
1130         return 0;
1131 }
1132
1133 /*
1134  * callback routine supplied to bulkstat(). Given an inumber, find its
1135  * dquots and update them to account for resources taken by that inode.
1136  */
1137 /* ARGSUSED */
1138 STATIC int
1139 xfs_qm_dqusage_adjust(
1140         struct xfs_mount        *mp,
1141         struct xfs_trans        *tp,
1142         xfs_ino_t               ino,
1143         void                    *data)
1144 {
1145         struct xfs_inode        *ip;
1146         xfs_qcnt_t              nblks;
1147         xfs_filblks_t           rtblks = 0;     /* total rt blks */
1148         int                     error;
1149
1150         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1151
1152         /*
1153          * rootino must have its resources accounted for, not so with the quota
1154          * inodes.
1155          */
1156         if (xfs_is_quota_inode(&mp->m_sb, ino))
1157                 return 0;
1158
1159         /*
1160          * We don't _need_ to take the ilock EXCL here because quotacheck runs
1161          * at mount time and therefore nobody will be racing chown/chproj.
1162          */
1163         error = xfs_iget(mp, tp, ino, XFS_IGET_DONTCACHE, 0, &ip);
1164         if (error == -EINVAL || error == -ENOENT)
1165                 return 0;
1166         if (error)
1167                 return error;
1168
1169         ASSERT(ip->i_delayed_blks == 0);
1170
1171         if (XFS_IS_REALTIME_INODE(ip)) {
1172                 struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1173
1174                 error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
1175                 if (error)
1176                         goto error0;
1177
1178                 xfs_bmap_count_leaves(ifp, &rtblks);
1179         }
1180
1181         nblks = (xfs_qcnt_t)ip->i_nblocks - rtblks;
1182
1183         /*
1184          * Add the (disk blocks and inode) resources occupied by this
1185          * inode to its dquots. We do this adjustment in the incore dquot,
1186          * and also copy the changes to its buffer.
1187          * We don't care about putting these changes in a transaction
1188          * envelope because if we crash in the middle of a 'quotacheck'
1189          * we have to start from the beginning anyway.
1190          * Once we're done, we'll log all the dquot bufs.
1191          *
1192          * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1193          * and quotaoffs don't race. (Quotachecks happen at mount time only).
1194          */
1195         if (XFS_IS_UQUOTA_ON(mp)) {
1196                 error = xfs_qm_quotacheck_dqadjust(ip, XFS_DQTYPE_USER, nblks,
1197                                 rtblks);
1198                 if (error)
1199                         goto error0;
1200         }
1201
1202         if (XFS_IS_GQUOTA_ON(mp)) {
1203                 error = xfs_qm_quotacheck_dqadjust(ip, XFS_DQTYPE_GROUP, nblks,
1204                                 rtblks);
1205                 if (error)
1206                         goto error0;
1207         }
1208
1209         if (XFS_IS_PQUOTA_ON(mp)) {
1210                 error = xfs_qm_quotacheck_dqadjust(ip, XFS_DQTYPE_PROJ, nblks,
1211                                 rtblks);
1212                 if (error)
1213                         goto error0;
1214         }
1215
1216 error0:
1217         xfs_irele(ip);
1218         return error;
1219 }
1220
1221 STATIC int
1222 xfs_qm_flush_one(
1223         struct xfs_dquot        *dqp,
1224         void                    *data)
1225 {
1226         struct xfs_mount        *mp = dqp->q_mount;
1227         struct list_head        *buffer_list = data;
1228         struct xfs_buf          *bp = NULL;
1229         int                     error = 0;
1230
1231         xfs_dqlock(dqp);
1232         if (dqp->q_flags & XFS_DQFLAG_FREEING)
1233                 goto out_unlock;
1234         if (!XFS_DQ_IS_DIRTY(dqp))
1235                 goto out_unlock;
1236
1237         /*
1238          * The only way the dquot is already flush locked by the time quotacheck
1239          * gets here is if reclaim flushed it before the dqadjust walk dirtied
1240          * it for the final time. Quotacheck collects all dquot bufs in the
1241          * local delwri queue before dquots are dirtied, so reclaim can't have
1242          * possibly queued it for I/O. The only way out is to push the buffer to
1243          * cycle the flush lock.
1244          */
1245         if (!xfs_dqflock_nowait(dqp)) {
1246                 /* buf is pinned in-core by delwri list */
1247                 bp = xfs_buf_incore(mp->m_ddev_targp, dqp->q_blkno,
1248                                 mp->m_quotainfo->qi_dqchunklen, 0);
1249                 if (!bp) {
1250                         error = -EINVAL;
1251                         goto out_unlock;
1252                 }
1253                 xfs_buf_unlock(bp);
1254
1255                 xfs_buf_delwri_pushbuf(bp, buffer_list);
1256                 xfs_buf_rele(bp);
1257
1258                 error = -EAGAIN;
1259                 goto out_unlock;
1260         }
1261
1262         error = xfs_qm_dqflush(dqp, &bp);
1263         if (error)
1264                 goto out_unlock;
1265
1266         xfs_buf_delwri_queue(bp, buffer_list);
1267         xfs_buf_relse(bp);
1268 out_unlock:
1269         xfs_dqunlock(dqp);
1270         return error;
1271 }
1272
1273 /*
1274  * Walk thru all the filesystem inodes and construct a consistent view
1275  * of the disk quota world. If the quotacheck fails, disable quotas.
1276  */
1277 STATIC int
1278 xfs_qm_quotacheck(
1279         xfs_mount_t     *mp)
1280 {
1281         int                     error, error2;
1282         uint                    flags;
1283         LIST_HEAD               (buffer_list);
1284         struct xfs_inode        *uip = mp->m_quotainfo->qi_uquotaip;
1285         struct xfs_inode        *gip = mp->m_quotainfo->qi_gquotaip;
1286         struct xfs_inode        *pip = mp->m_quotainfo->qi_pquotaip;
1287
1288         flags = 0;
1289
1290         ASSERT(uip || gip || pip);
1291         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1292
1293         xfs_notice(mp, "Quotacheck needed: Please wait.");
1294
1295         /*
1296          * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1297          * their counters to zero. We need a clean slate.
1298          * We don't log our changes till later.
1299          */
1300         if (uip) {
1301                 error = xfs_qm_reset_dqcounts_buf(mp, uip, XFS_DQTYPE_USER,
1302                                          &buffer_list);
1303                 if (error)
1304                         goto error_return;
1305                 flags |= XFS_UQUOTA_CHKD;
1306         }
1307
1308         if (gip) {
1309                 error = xfs_qm_reset_dqcounts_buf(mp, gip, XFS_DQTYPE_GROUP,
1310                                          &buffer_list);
1311                 if (error)
1312                         goto error_return;
1313                 flags |= XFS_GQUOTA_CHKD;
1314         }
1315
1316         if (pip) {
1317                 error = xfs_qm_reset_dqcounts_buf(mp, pip, XFS_DQTYPE_PROJ,
1318                                          &buffer_list);
1319                 if (error)
1320                         goto error_return;
1321                 flags |= XFS_PQUOTA_CHKD;
1322         }
1323
1324         error = xfs_iwalk_threaded(mp, 0, 0, xfs_qm_dqusage_adjust, 0, true,
1325                         NULL);
1326         if (error)
1327                 goto error_return;
1328
1329         /*
1330          * We've made all the changes that we need to make incore.  Flush them
1331          * down to disk buffers if everything was updated successfully.
1332          */
1333         if (XFS_IS_UQUOTA_ON(mp)) {
1334                 error = xfs_qm_dquot_walk(mp, XFS_DQTYPE_USER, xfs_qm_flush_one,
1335                                           &buffer_list);
1336         }
1337         if (XFS_IS_GQUOTA_ON(mp)) {
1338                 error2 = xfs_qm_dquot_walk(mp, XFS_DQTYPE_GROUP, xfs_qm_flush_one,
1339                                            &buffer_list);
1340                 if (!error)
1341                         error = error2;
1342         }
1343         if (XFS_IS_PQUOTA_ON(mp)) {
1344                 error2 = xfs_qm_dquot_walk(mp, XFS_DQTYPE_PROJ, xfs_qm_flush_one,
1345                                            &buffer_list);
1346                 if (!error)
1347                         error = error2;
1348         }
1349
1350         error2 = xfs_buf_delwri_submit(&buffer_list);
1351         if (!error)
1352                 error = error2;
1353
1354         /*
1355          * We can get this error if we couldn't do a dquot allocation inside
1356          * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1357          * dirty dquots that might be cached, we just want to get rid of them
1358          * and turn quotaoff. The dquots won't be attached to any of the inodes
1359          * at this point (because we intentionally didn't in dqget_noattach).
1360          */
1361         if (error) {
1362                 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1363                 goto error_return;
1364         }
1365
1366         /*
1367          * If one type of quotas is off, then it will lose its
1368          * quotachecked status, since we won't be doing accounting for
1369          * that type anymore.
1370          */
1371         mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1372         mp->m_qflags |= flags;
1373
1374  error_return:
1375         xfs_buf_delwri_cancel(&buffer_list);
1376
1377         if (error) {
1378                 xfs_warn(mp,
1379         "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1380                         error);
1381                 /*
1382                  * We must turn off quotas.
1383                  */
1384                 ASSERT(mp->m_quotainfo != NULL);
1385                 xfs_qm_destroy_quotainfo(mp);
1386                 if (xfs_mount_reset_sbqflags(mp)) {
1387                         xfs_warn(mp,
1388                                 "Quotacheck: Failed to reset quota flags.");
1389                 }
1390         } else
1391                 xfs_notice(mp, "Quotacheck: Done.");
1392         return error;
1393 }
1394
1395 /*
1396  * This is called from xfs_mountfs to start quotas and initialize all
1397  * necessary data structures like quotainfo.  This is also responsible for
1398  * running a quotacheck as necessary.  We are guaranteed that the superblock
1399  * is consistently read in at this point.
1400  *
1401  * If we fail here, the mount will continue with quota turned off. We don't
1402  * need to inidicate success or failure at all.
1403  */
1404 void
1405 xfs_qm_mount_quotas(
1406         struct xfs_mount        *mp)
1407 {
1408         int                     error = 0;
1409         uint                    sbf;
1410
1411         /*
1412          * If quotas on realtime volumes is not supported, we disable
1413          * quotas immediately.
1414          */
1415         if (mp->m_sb.sb_rextents) {
1416                 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
1417                 mp->m_qflags = 0;
1418                 goto write_changes;
1419         }
1420
1421         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1422
1423         /*
1424          * Allocate the quotainfo structure inside the mount struct, and
1425          * create quotainode(s), and change/rev superblock if necessary.
1426          */
1427         error = xfs_qm_init_quotainfo(mp);
1428         if (error) {
1429                 /*
1430                  * We must turn off quotas.
1431                  */
1432                 ASSERT(mp->m_quotainfo == NULL);
1433                 mp->m_qflags = 0;
1434                 goto write_changes;
1435         }
1436         /*
1437          * If any of the quotas are not consistent, do a quotacheck.
1438          */
1439         if (XFS_QM_NEED_QUOTACHECK(mp)) {
1440                 error = xfs_qm_quotacheck(mp);
1441                 if (error) {
1442                         /* Quotacheck failed and disabled quotas. */
1443                         return;
1444                 }
1445         }
1446         /*
1447          * If one type of quotas is off, then it will lose its
1448          * quotachecked status, since we won't be doing accounting for
1449          * that type anymore.
1450          */
1451         if (!XFS_IS_UQUOTA_ON(mp))
1452                 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
1453         if (!XFS_IS_GQUOTA_ON(mp))
1454                 mp->m_qflags &= ~XFS_GQUOTA_CHKD;
1455         if (!XFS_IS_PQUOTA_ON(mp))
1456                 mp->m_qflags &= ~XFS_PQUOTA_CHKD;
1457
1458  write_changes:
1459         /*
1460          * We actually don't have to acquire the m_sb_lock at all.
1461          * This can only be called from mount, and that's single threaded. XXX
1462          */
1463         spin_lock(&mp->m_sb_lock);
1464         sbf = mp->m_sb.sb_qflags;
1465         mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
1466         spin_unlock(&mp->m_sb_lock);
1467
1468         if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
1469                 if (xfs_sync_sb(mp, false)) {
1470                         /*
1471                          * We could only have been turning quotas off.
1472                          * We aren't in very good shape actually because
1473                          * the incore structures are convinced that quotas are
1474                          * off, but the on disk superblock doesn't know that !
1475                          */
1476                         ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
1477                         xfs_alert(mp, "%s: Superblock update failed!",
1478                                 __func__);
1479                 }
1480         }
1481
1482         if (error) {
1483                 xfs_warn(mp, "Failed to initialize disk quotas.");
1484                 return;
1485         }
1486 }
1487
1488 /*
1489  * This is called after the superblock has been read in and we're ready to
1490  * iget the quota inodes.
1491  */
1492 STATIC int
1493 xfs_qm_init_quotainos(
1494         xfs_mount_t     *mp)
1495 {
1496         struct xfs_inode        *uip = NULL;
1497         struct xfs_inode        *gip = NULL;
1498         struct xfs_inode        *pip = NULL;
1499         int                     error;
1500         uint                    flags = 0;
1501
1502         ASSERT(mp->m_quotainfo);
1503
1504         /*
1505          * Get the uquota and gquota inodes
1506          */
1507         if (xfs_sb_version_hasquota(&mp->m_sb)) {
1508                 if (XFS_IS_UQUOTA_ON(mp) &&
1509                     mp->m_sb.sb_uquotino != NULLFSINO) {
1510                         ASSERT(mp->m_sb.sb_uquotino > 0);
1511                         error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1512                                              0, 0, &uip);
1513                         if (error)
1514                                 return error;
1515                 }
1516                 if (XFS_IS_GQUOTA_ON(mp) &&
1517                     mp->m_sb.sb_gquotino != NULLFSINO) {
1518                         ASSERT(mp->m_sb.sb_gquotino > 0);
1519                         error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1520                                              0, 0, &gip);
1521                         if (error)
1522                                 goto error_rele;
1523                 }
1524                 if (XFS_IS_PQUOTA_ON(mp) &&
1525                     mp->m_sb.sb_pquotino != NULLFSINO) {
1526                         ASSERT(mp->m_sb.sb_pquotino > 0);
1527                         error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino,
1528                                              0, 0, &pip);
1529                         if (error)
1530                                 goto error_rele;
1531                 }
1532         } else {
1533                 flags |= XFS_QMOPT_SBVERSION;
1534         }
1535
1536         /*
1537          * Create the three inodes, if they don't exist already. The changes
1538          * made above will get added to a transaction and logged in one of
1539          * the qino_alloc calls below.  If the device is readonly,
1540          * temporarily switch to read-write to do this.
1541          */
1542         if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1543                 error = xfs_qm_qino_alloc(mp, &uip,
1544                                               flags | XFS_QMOPT_UQUOTA);
1545                 if (error)
1546                         goto error_rele;
1547
1548                 flags &= ~XFS_QMOPT_SBVERSION;
1549         }
1550         if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
1551                 error = xfs_qm_qino_alloc(mp, &gip,
1552                                           flags | XFS_QMOPT_GQUOTA);
1553                 if (error)
1554                         goto error_rele;
1555
1556                 flags &= ~XFS_QMOPT_SBVERSION;
1557         }
1558         if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) {
1559                 error = xfs_qm_qino_alloc(mp, &pip,
1560                                           flags | XFS_QMOPT_PQUOTA);
1561                 if (error)
1562                         goto error_rele;
1563         }
1564
1565         mp->m_quotainfo->qi_uquotaip = uip;
1566         mp->m_quotainfo->qi_gquotaip = gip;
1567         mp->m_quotainfo->qi_pquotaip = pip;
1568
1569         return 0;
1570
1571 error_rele:
1572         if (uip)
1573                 xfs_irele(uip);
1574         if (gip)
1575                 xfs_irele(gip);
1576         if (pip)
1577                 xfs_irele(pip);
1578         return error;
1579 }
1580
1581 STATIC void
1582 xfs_qm_destroy_quotainos(
1583         struct xfs_quotainfo    *qi)
1584 {
1585         if (qi->qi_uquotaip) {
1586                 xfs_irele(qi->qi_uquotaip);
1587                 qi->qi_uquotaip = NULL; /* paranoia */
1588         }
1589         if (qi->qi_gquotaip) {
1590                 xfs_irele(qi->qi_gquotaip);
1591                 qi->qi_gquotaip = NULL;
1592         }
1593         if (qi->qi_pquotaip) {
1594                 xfs_irele(qi->qi_pquotaip);
1595                 qi->qi_pquotaip = NULL;
1596         }
1597 }
1598
1599 STATIC void
1600 xfs_qm_dqfree_one(
1601         struct xfs_dquot        *dqp)
1602 {
1603         struct xfs_mount        *mp = dqp->q_mount;
1604         struct xfs_quotainfo    *qi = mp->m_quotainfo;
1605
1606         mutex_lock(&qi->qi_tree_lock);
1607         radix_tree_delete(xfs_dquot_tree(qi, xfs_dquot_type(dqp)), dqp->q_id);
1608
1609         qi->qi_dquots--;
1610         mutex_unlock(&qi->qi_tree_lock);
1611
1612         xfs_qm_dqdestroy(dqp);
1613 }
1614
1615 /* --------------- utility functions for vnodeops ---------------- */
1616
1617
1618 /*
1619  * Given an inode, a uid, gid and prid make sure that we have
1620  * allocated relevant dquot(s) on disk, and that we won't exceed inode
1621  * quotas by creating this file.
1622  * This also attaches dquot(s) to the given inode after locking it,
1623  * and returns the dquots corresponding to the uid and/or gid.
1624  *
1625  * in   : inode (unlocked)
1626  * out  : udquot, gdquot with references taken and unlocked
1627  */
1628 int
1629 xfs_qm_vop_dqalloc(
1630         struct xfs_inode        *ip,
1631         kuid_t                  uid,
1632         kgid_t                  gid,
1633         prid_t                  prid,
1634         uint                    flags,
1635         struct xfs_dquot        **O_udqpp,
1636         struct xfs_dquot        **O_gdqpp,
1637         struct xfs_dquot        **O_pdqpp)
1638 {
1639         struct xfs_mount        *mp = ip->i_mount;
1640         struct inode            *inode = VFS_I(ip);
1641         struct user_namespace   *user_ns = inode->i_sb->s_user_ns;
1642         struct xfs_dquot        *uq = NULL;
1643         struct xfs_dquot        *gq = NULL;
1644         struct xfs_dquot        *pq = NULL;
1645         int                     error;
1646         uint                    lockflags;
1647
1648         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1649                 return 0;
1650
1651         lockflags = XFS_ILOCK_EXCL;
1652         xfs_ilock(ip, lockflags);
1653
1654         if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1655                 gid = inode->i_gid;
1656
1657         /*
1658          * Attach the dquot(s) to this inode, doing a dquot allocation
1659          * if necessary. The dquot(s) will not be locked.
1660          */
1661         if (XFS_NOT_DQATTACHED(mp, ip)) {
1662                 error = xfs_qm_dqattach_locked(ip, true);
1663                 if (error) {
1664                         xfs_iunlock(ip, lockflags);
1665                         return error;
1666                 }
1667         }
1668
1669         if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1670                 ASSERT(O_udqpp);
1671                 if (!uid_eq(inode->i_uid, uid)) {
1672                         /*
1673                          * What we need is the dquot that has this uid, and
1674                          * if we send the inode to dqget, the uid of the inode
1675                          * takes priority over what's sent in the uid argument.
1676                          * We must unlock inode here before calling dqget if
1677                          * we're not sending the inode, because otherwise
1678                          * we'll deadlock by doing trans_reserve while
1679                          * holding ilock.
1680                          */
1681                         xfs_iunlock(ip, lockflags);
1682                         error = xfs_qm_dqget(mp, from_kuid(user_ns, uid),
1683                                         XFS_DQTYPE_USER, true, &uq);
1684                         if (error) {
1685                                 ASSERT(error != -ENOENT);
1686                                 return error;
1687                         }
1688                         /*
1689                          * Get the ilock in the right order.
1690                          */
1691                         xfs_dqunlock(uq);
1692                         lockflags = XFS_ILOCK_SHARED;
1693                         xfs_ilock(ip, lockflags);
1694                 } else {
1695                         /*
1696                          * Take an extra reference, because we'll return
1697                          * this to caller
1698                          */
1699                         ASSERT(ip->i_udquot);
1700                         uq = xfs_qm_dqhold(ip->i_udquot);
1701                 }
1702         }
1703         if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1704                 ASSERT(O_gdqpp);
1705                 if (!gid_eq(inode->i_gid, gid)) {
1706                         xfs_iunlock(ip, lockflags);
1707                         error = xfs_qm_dqget(mp, from_kgid(user_ns, gid),
1708                                         XFS_DQTYPE_GROUP, true, &gq);
1709                         if (error) {
1710                                 ASSERT(error != -ENOENT);
1711                                 goto error_rele;
1712                         }
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);
1719                 }
1720         }
1721         if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1722                 ASSERT(O_pdqpp);
1723                 if (ip->i_projid != prid) {
1724                         xfs_iunlock(ip, lockflags);
1725                         error = xfs_qm_dqget(mp, prid,
1726                                         XFS_DQTYPE_PROJ, true, &pq);
1727                         if (error) {
1728                                 ASSERT(error != -ENOENT);
1729                                 goto error_rele;
1730                         }
1731                         xfs_dqunlock(pq);
1732                         lockflags = XFS_ILOCK_SHARED;
1733                         xfs_ilock(ip, lockflags);
1734                 } else {
1735                         ASSERT(ip->i_pdquot);
1736                         pq = xfs_qm_dqhold(ip->i_pdquot);
1737                 }
1738         }
1739         trace_xfs_dquot_dqalloc(ip);
1740
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;
1755
1756 error_rele:
1757         xfs_qm_dqrele(gq);
1758         xfs_qm_dqrele(uq);
1759         return error;
1760 }
1761
1762 /*
1763  * Actually transfer ownership, and do dquot modifications.
1764  * These were already reserved.
1765  */
1766 struct xfs_dquot *
1767 xfs_qm_vop_chown(
1768         struct xfs_trans        *tp,
1769         struct xfs_inode        *ip,
1770         struct xfs_dquot        **IO_olddq,
1771         struct xfs_dquot        *newdq)
1772 {
1773         struct xfs_dquot        *prevdq;
1774         uint            bfield = XFS_IS_REALTIME_INODE(ip) ?
1775                                  XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1776
1777
1778         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1779         ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1780
1781         /* old dquot */
1782         prevdq = *IO_olddq;
1783         ASSERT(prevdq);
1784         ASSERT(prevdq != newdq);
1785
1786         xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_nblocks));
1787         xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1788
1789         /* the sparkling new dquot */
1790         xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_nblocks);
1791         xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1792
1793         /*
1794          * Back when we made quota reservations for the chown, we reserved the
1795          * ondisk blocks + delalloc blocks with the new dquot.  Now that we've
1796          * switched the dquots, decrease the new dquot's block reservation
1797          * (having already bumped up the real counter) so that we don't have
1798          * any reservation to give back when we commit.
1799          */
1800         xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_RES_BLKS,
1801                         -ip->i_delayed_blks);
1802
1803         /*
1804          * Give the incore reservation for delalloc blocks back to the old
1805          * dquot.  We don't normally handle delalloc quota reservations
1806          * transactionally, so just lock the dquot and subtract from the
1807          * reservation.  Dirty the transaction because it's too late to turn
1808          * back now.
1809          */
1810         tp->t_flags |= XFS_TRANS_DIRTY;
1811         xfs_dqlock(prevdq);
1812         ASSERT(prevdq->q_blk.reserved >= ip->i_delayed_blks);
1813         prevdq->q_blk.reserved -= ip->i_delayed_blks;
1814         xfs_dqunlock(prevdq);
1815
1816         /*
1817          * Take an extra reference, because the inode is going to keep
1818          * this dquot pointer even after the trans_commit.
1819          */
1820         *IO_olddq = xfs_qm_dqhold(newdq);
1821
1822         return prevdq;
1823 }
1824
1825 int
1826 xfs_qm_vop_rename_dqattach(
1827         struct xfs_inode        **i_tab)
1828 {
1829         struct xfs_mount        *mp = i_tab[0]->i_mount;
1830         int                     i;
1831
1832         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1833                 return 0;
1834
1835         for (i = 0; (i < 4 && i_tab[i]); i++) {
1836                 struct xfs_inode        *ip = i_tab[i];
1837                 int                     error;
1838
1839                 /*
1840                  * Watch out for duplicate entries in the table.
1841                  */
1842                 if (i == 0 || ip != i_tab[i-1]) {
1843                         if (XFS_NOT_DQATTACHED(mp, ip)) {
1844                                 error = xfs_qm_dqattach(ip);
1845                                 if (error)
1846                                         return error;
1847                         }
1848                 }
1849         }
1850         return 0;
1851 }
1852
1853 void
1854 xfs_qm_vop_create_dqattach(
1855         struct xfs_trans        *tp,
1856         struct xfs_inode        *ip,
1857         struct xfs_dquot        *udqp,
1858         struct xfs_dquot        *gdqp,
1859         struct xfs_dquot        *pdqp)
1860 {
1861         struct xfs_mount        *mp = tp->t_mountp;
1862
1863         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1864                 return;
1865
1866         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1867
1868         if (udqp && XFS_IS_UQUOTA_ON(mp)) {
1869                 ASSERT(ip->i_udquot == NULL);
1870                 ASSERT(i_uid_read(VFS_I(ip)) == udqp->q_id);
1871
1872                 ip->i_udquot = xfs_qm_dqhold(udqp);
1873                 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
1874         }
1875         if (gdqp && XFS_IS_GQUOTA_ON(mp)) {
1876                 ASSERT(ip->i_gdquot == NULL);
1877                 ASSERT(i_gid_read(VFS_I(ip)) == gdqp->q_id);
1878
1879                 ip->i_gdquot = xfs_qm_dqhold(gdqp);
1880                 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
1881         }
1882         if (pdqp && XFS_IS_PQUOTA_ON(mp)) {
1883                 ASSERT(ip->i_pdquot == NULL);
1884                 ASSERT(ip->i_projid == pdqp->q_id);
1885
1886                 ip->i_pdquot = xfs_qm_dqhold(pdqp);
1887                 xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1);
1888         }
1889 }
1890