]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - fs/xfs/xfs_qm.c
Merge tag 'zynqmp-dt-for-v5.8' of https://github.com/Xilinx/linux-xlnx into arm/dt
[mirror_ubuntu-hirsute-kernel.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
27 /*
28 * The global quota manager. There is only one of these for the entire
29 * system, _not_ one per file system. XQM keeps track of the overall
30 * quota functionality, including maintaining the freelist and hash
31 * tables of dquots.
32 */
33 STATIC int xfs_qm_init_quotainos(struct xfs_mount *mp);
34 STATIC int xfs_qm_init_quotainfo(struct xfs_mount *mp);
35
36 STATIC void xfs_qm_destroy_quotainos(struct xfs_quotainfo *qi);
37 STATIC void xfs_qm_dqfree_one(struct xfs_dquot *dqp);
38 /*
39 * We use the batch lookup interface to iterate over the dquots as it
40 * currently is the only interface into the radix tree code that allows
41 * fuzzy lookups instead of exact matches. Holding the lock over multiple
42 * operations is fine as all callers are used either during mount/umount
43 * or quotaoff.
44 */
45 #define XFS_DQ_LOOKUP_BATCH 32
46
47 STATIC int
48 xfs_qm_dquot_walk(
49 struct xfs_mount *mp,
50 int type,
51 int (*execute)(struct xfs_dquot *dqp, void *data),
52 void *data)
53 {
54 struct xfs_quotainfo *qi = mp->m_quotainfo;
55 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
56 uint32_t next_index;
57 int last_error = 0;
58 int skipped;
59 int nr_found;
60
61 restart:
62 skipped = 0;
63 next_index = 0;
64 nr_found = 0;
65
66 while (1) {
67 struct xfs_dquot *batch[XFS_DQ_LOOKUP_BATCH];
68 int error = 0;
69 int i;
70
71 mutex_lock(&qi->qi_tree_lock);
72 nr_found = radix_tree_gang_lookup(tree, (void **)batch,
73 next_index, XFS_DQ_LOOKUP_BATCH);
74 if (!nr_found) {
75 mutex_unlock(&qi->qi_tree_lock);
76 break;
77 }
78
79 for (i = 0; i < nr_found; i++) {
80 struct xfs_dquot *dqp = batch[i];
81
82 next_index = be32_to_cpu(dqp->q_core.d_id) + 1;
83
84 error = execute(batch[i], data);
85 if (error == -EAGAIN) {
86 skipped++;
87 continue;
88 }
89 if (error && last_error != -EFSCORRUPTED)
90 last_error = error;
91 }
92
93 mutex_unlock(&qi->qi_tree_lock);
94
95 /* bail out if the filesystem is corrupted. */
96 if (last_error == -EFSCORRUPTED) {
97 skipped = 0;
98 break;
99 }
100 /* we're done if id overflows back to zero */
101 if (!next_index)
102 break;
103 }
104
105 if (skipped) {
106 delay(1);
107 goto restart;
108 }
109
110 return last_error;
111 }
112
113
114 /*
115 * Purge a dquot from all tracking data structures and free it.
116 */
117 STATIC int
118 xfs_qm_dqpurge(
119 struct xfs_dquot *dqp,
120 void *data)
121 {
122 struct xfs_mount *mp = dqp->q_mount;
123 struct xfs_quotainfo *qi = mp->m_quotainfo;
124 int error = -EAGAIN;
125
126 xfs_dqlock(dqp);
127 if ((dqp->dq_flags & XFS_DQ_FREEING) || dqp->q_nrefs != 0)
128 goto out_unlock;
129
130 dqp->dq_flags |= XFS_DQ_FREEING;
131
132 xfs_dqflock(dqp);
133
134 /*
135 * If we are turning this type of quotas off, we don't care
136 * about the dirty metadata sitting in this dquot. OTOH, if
137 * we're unmounting, we do care, so we flush it and wait.
138 */
139 if (XFS_DQ_IS_DIRTY(dqp)) {
140 struct xfs_buf *bp = NULL;
141
142 /*
143 * We don't care about getting disk errors here. We need
144 * to purge this dquot anyway, so we go ahead regardless.
145 */
146 error = xfs_qm_dqflush(dqp, &bp);
147 if (!error) {
148 error = xfs_bwrite(bp);
149 xfs_buf_relse(bp);
150 } else if (error == -EAGAIN) {
151 goto out_unlock;
152 }
153 xfs_dqflock(dqp);
154 }
155
156 ASSERT(atomic_read(&dqp->q_pincount) == 0);
157 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
158 !test_bit(XFS_LI_IN_AIL, &dqp->q_logitem.qli_item.li_flags));
159
160 xfs_dqfunlock(dqp);
161 xfs_dqunlock(dqp);
162
163 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
164 be32_to_cpu(dqp->q_core.d_id));
165 qi->qi_dquots--;
166
167 /*
168 * We move dquots to the freelist as soon as their reference count
169 * hits zero, so it really should be on the freelist here.
170 */
171 ASSERT(!list_empty(&dqp->q_lru));
172 list_lru_del(&qi->qi_lru, &dqp->q_lru);
173 XFS_STATS_DEC(mp, xs_qm_dquot_unused);
174
175 xfs_qm_dqdestroy(dqp);
176 return 0;
177
178 out_unlock:
179 xfs_dqunlock(dqp);
180 return error;
181 }
182
183 /*
184 * Purge the dquot cache.
185 */
186 void
187 xfs_qm_dqpurge_all(
188 struct xfs_mount *mp,
189 uint flags)
190 {
191 if (flags & XFS_QMOPT_UQUOTA)
192 xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_dqpurge, NULL);
193 if (flags & XFS_QMOPT_GQUOTA)
194 xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_dqpurge, NULL);
195 if (flags & XFS_QMOPT_PQUOTA)
196 xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_dqpurge, NULL);
197 }
198
199 /*
200 * Just destroy the quotainfo structure.
201 */
202 void
203 xfs_qm_unmount(
204 struct xfs_mount *mp)
205 {
206 if (mp->m_quotainfo) {
207 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
208 xfs_qm_destroy_quotainfo(mp);
209 }
210 }
211
212 /*
213 * Called from the vfsops layer.
214 */
215 void
216 xfs_qm_unmount_quotas(
217 xfs_mount_t *mp)
218 {
219 /*
220 * Release the dquots that root inode, et al might be holding,
221 * before we flush quotas and blow away the quotainfo structure.
222 */
223 ASSERT(mp->m_rootip);
224 xfs_qm_dqdetach(mp->m_rootip);
225 if (mp->m_rbmip)
226 xfs_qm_dqdetach(mp->m_rbmip);
227 if (mp->m_rsumip)
228 xfs_qm_dqdetach(mp->m_rsumip);
229
230 /*
231 * Release the quota inodes.
232 */
233 if (mp->m_quotainfo) {
234 if (mp->m_quotainfo->qi_uquotaip) {
235 xfs_irele(mp->m_quotainfo->qi_uquotaip);
236 mp->m_quotainfo->qi_uquotaip = NULL;
237 }
238 if (mp->m_quotainfo->qi_gquotaip) {
239 xfs_irele(mp->m_quotainfo->qi_gquotaip);
240 mp->m_quotainfo->qi_gquotaip = NULL;
241 }
242 if (mp->m_quotainfo->qi_pquotaip) {
243 xfs_irele(mp->m_quotainfo->qi_pquotaip);
244 mp->m_quotainfo->qi_pquotaip = NULL;
245 }
246 }
247 }
248
249 STATIC int
250 xfs_qm_dqattach_one(
251 struct xfs_inode *ip,
252 xfs_dqid_t id,
253 uint type,
254 bool doalloc,
255 struct xfs_dquot **IO_idqpp)
256 {
257 struct xfs_dquot *dqp;
258 int error;
259
260 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
261 error = 0;
262
263 /*
264 * See if we already have it in the inode itself. IO_idqpp is &i_udquot
265 * or &i_gdquot. This made the code look weird, but made the logic a lot
266 * simpler.
267 */
268 dqp = *IO_idqpp;
269 if (dqp) {
270 trace_xfs_dqattach_found(dqp);
271 return 0;
272 }
273
274 /*
275 * Find the dquot from somewhere. This bumps the reference count of
276 * dquot and returns it locked. This can return ENOENT if dquot didn't
277 * exist on disk and we didn't ask it to allocate; ESRCH if quotas got
278 * turned off suddenly.
279 */
280 error = xfs_qm_dqget_inode(ip, type, doalloc, &dqp);
281 if (error)
282 return error;
283
284 trace_xfs_dqattach_get(dqp);
285
286 /*
287 * dqget may have dropped and re-acquired the ilock, but it guarantees
288 * that the dquot returned is the one that should go in the inode.
289 */
290 *IO_idqpp = dqp;
291 xfs_dqunlock(dqp);
292 return 0;
293 }
294
295 static bool
296 xfs_qm_need_dqattach(
297 struct xfs_inode *ip)
298 {
299 struct xfs_mount *mp = ip->i_mount;
300
301 if (!XFS_IS_QUOTA_RUNNING(mp))
302 return false;
303 if (!XFS_IS_QUOTA_ON(mp))
304 return false;
305 if (!XFS_NOT_DQATTACHED(mp, ip))
306 return false;
307 if (xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
308 return false;
309 return true;
310 }
311
312 /*
313 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
314 * into account.
315 * If @doalloc is true, the dquot(s) will be allocated if needed.
316 * Inode may get unlocked and relocked in here, and the caller must deal with
317 * the consequences.
318 */
319 int
320 xfs_qm_dqattach_locked(
321 xfs_inode_t *ip,
322 bool doalloc)
323 {
324 xfs_mount_t *mp = ip->i_mount;
325 int error = 0;
326
327 if (!xfs_qm_need_dqattach(ip))
328 return 0;
329
330 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
331
332 if (XFS_IS_UQUOTA_ON(mp) && !ip->i_udquot) {
333 error = xfs_qm_dqattach_one(ip, i_uid_read(VFS_I(ip)),
334 XFS_DQ_USER, doalloc, &ip->i_udquot);
335 if (error)
336 goto done;
337 ASSERT(ip->i_udquot);
338 }
339
340 if (XFS_IS_GQUOTA_ON(mp) && !ip->i_gdquot) {
341 error = xfs_qm_dqattach_one(ip, i_gid_read(VFS_I(ip)),
342 XFS_DQ_GROUP, doalloc, &ip->i_gdquot);
343 if (error)
344 goto done;
345 ASSERT(ip->i_gdquot);
346 }
347
348 if (XFS_IS_PQUOTA_ON(mp) && !ip->i_pdquot) {
349 error = xfs_qm_dqattach_one(ip, ip->i_d.di_projid, XFS_DQ_PROJ,
350 doalloc, &ip->i_pdquot);
351 if (error)
352 goto done;
353 ASSERT(ip->i_pdquot);
354 }
355
356 done:
357 /*
358 * Don't worry about the dquots that we may have attached before any
359 * error - they'll get detached later if it has not already been done.
360 */
361 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
362 return error;
363 }
364
365 int
366 xfs_qm_dqattach(
367 struct xfs_inode *ip)
368 {
369 int error;
370
371 if (!xfs_qm_need_dqattach(ip))
372 return 0;
373
374 xfs_ilock(ip, XFS_ILOCK_EXCL);
375 error = xfs_qm_dqattach_locked(ip, false);
376 xfs_iunlock(ip, XFS_ILOCK_EXCL);
377
378 return error;
379 }
380
381 /*
382 * Release dquots (and their references) if any.
383 * The inode should be locked EXCL except when this's called by
384 * xfs_ireclaim.
385 */
386 void
387 xfs_qm_dqdetach(
388 xfs_inode_t *ip)
389 {
390 if (!(ip->i_udquot || ip->i_gdquot || ip->i_pdquot))
391 return;
392
393 trace_xfs_dquot_dqdetach(ip);
394
395 ASSERT(!xfs_is_quota_inode(&ip->i_mount->m_sb, ip->i_ino));
396 if (ip->i_udquot) {
397 xfs_qm_dqrele(ip->i_udquot);
398 ip->i_udquot = NULL;
399 }
400 if (ip->i_gdquot) {
401 xfs_qm_dqrele(ip->i_gdquot);
402 ip->i_gdquot = NULL;
403 }
404 if (ip->i_pdquot) {
405 xfs_qm_dqrele(ip->i_pdquot);
406 ip->i_pdquot = NULL;
407 }
408 }
409
410 struct xfs_qm_isolate {
411 struct list_head buffers;
412 struct list_head dispose;
413 };
414
415 static enum lru_status
416 xfs_qm_dquot_isolate(
417 struct list_head *item,
418 struct list_lru_one *lru,
419 spinlock_t *lru_lock,
420 void *arg)
421 __releases(lru_lock) __acquires(lru_lock)
422 {
423 struct xfs_dquot *dqp = container_of(item,
424 struct xfs_dquot, q_lru);
425 struct xfs_qm_isolate *isol = arg;
426
427 if (!xfs_dqlock_nowait(dqp))
428 goto out_miss_busy;
429
430 /*
431 * This dquot has acquired a reference in the meantime remove it from
432 * the freelist and try again.
433 */
434 if (dqp->q_nrefs) {
435 xfs_dqunlock(dqp);
436 XFS_STATS_INC(dqp->q_mount, xs_qm_dqwants);
437
438 trace_xfs_dqreclaim_want(dqp);
439 list_lru_isolate(lru, &dqp->q_lru);
440 XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot_unused);
441 return LRU_REMOVED;
442 }
443
444 /*
445 * If the dquot is dirty, flush it. If it's already being flushed, just
446 * skip it so there is time for the IO to complete before we try to
447 * reclaim it again on the next LRU pass.
448 */
449 if (!xfs_dqflock_nowait(dqp)) {
450 xfs_dqunlock(dqp);
451 goto out_miss_busy;
452 }
453
454 if (XFS_DQ_IS_DIRTY(dqp)) {
455 struct xfs_buf *bp = NULL;
456 int error;
457
458 trace_xfs_dqreclaim_dirty(dqp);
459
460 /* we have to drop the LRU lock to flush the dquot */
461 spin_unlock(lru_lock);
462
463 error = xfs_qm_dqflush(dqp, &bp);
464 if (error)
465 goto out_unlock_dirty;
466
467 xfs_buf_delwri_queue(bp, &isol->buffers);
468 xfs_buf_relse(bp);
469 goto out_unlock_dirty;
470 }
471 xfs_dqfunlock(dqp);
472
473 /*
474 * Prevent lookups now that we are past the point of no return.
475 */
476 dqp->dq_flags |= XFS_DQ_FREEING;
477 xfs_dqunlock(dqp);
478
479 ASSERT(dqp->q_nrefs == 0);
480 list_lru_isolate_move(lru, &dqp->q_lru, &isol->dispose);
481 XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot_unused);
482 trace_xfs_dqreclaim_done(dqp);
483 XFS_STATS_INC(dqp->q_mount, xs_qm_dqreclaims);
484 return LRU_REMOVED;
485
486 out_miss_busy:
487 trace_xfs_dqreclaim_busy(dqp);
488 XFS_STATS_INC(dqp->q_mount, xs_qm_dqreclaim_misses);
489 return LRU_SKIP;
490
491 out_unlock_dirty:
492 trace_xfs_dqreclaim_busy(dqp);
493 XFS_STATS_INC(dqp->q_mount, xs_qm_dqreclaim_misses);
494 xfs_dqunlock(dqp);
495 spin_lock(lru_lock);
496 return LRU_RETRY;
497 }
498
499 static unsigned long
500 xfs_qm_shrink_scan(
501 struct shrinker *shrink,
502 struct shrink_control *sc)
503 {
504 struct xfs_quotainfo *qi = container_of(shrink,
505 struct xfs_quotainfo, qi_shrinker);
506 struct xfs_qm_isolate isol;
507 unsigned long freed;
508 int error;
509
510 if ((sc->gfp_mask & (__GFP_FS|__GFP_DIRECT_RECLAIM)) != (__GFP_FS|__GFP_DIRECT_RECLAIM))
511 return 0;
512
513 INIT_LIST_HEAD(&isol.buffers);
514 INIT_LIST_HEAD(&isol.dispose);
515
516 freed = list_lru_shrink_walk(&qi->qi_lru, sc,
517 xfs_qm_dquot_isolate, &isol);
518
519 error = xfs_buf_delwri_submit(&isol.buffers);
520 if (error)
521 xfs_warn(NULL, "%s: dquot reclaim failed", __func__);
522
523 while (!list_empty(&isol.dispose)) {
524 struct xfs_dquot *dqp;
525
526 dqp = list_first_entry(&isol.dispose, struct xfs_dquot, q_lru);
527 list_del_init(&dqp->q_lru);
528 xfs_qm_dqfree_one(dqp);
529 }
530
531 return freed;
532 }
533
534 static unsigned long
535 xfs_qm_shrink_count(
536 struct shrinker *shrink,
537 struct shrink_control *sc)
538 {
539 struct xfs_quotainfo *qi = container_of(shrink,
540 struct xfs_quotainfo, qi_shrinker);
541
542 return list_lru_shrink_count(&qi->qi_lru, sc);
543 }
544
545 STATIC void
546 xfs_qm_set_defquota(
547 struct xfs_mount *mp,
548 uint type,
549 struct xfs_quotainfo *qinf)
550 {
551 struct xfs_dquot *dqp;
552 struct xfs_def_quota *defq;
553 struct xfs_disk_dquot *ddqp;
554 int error;
555
556 error = xfs_qm_dqget_uncached(mp, 0, type, &dqp);
557 if (error)
558 return;
559
560 ddqp = &dqp->q_core;
561 defq = xfs_get_defquota(dqp, qinf);
562
563 /*
564 * Timers and warnings have been already set, let's just set the
565 * default limits for this quota type
566 */
567 defq->bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
568 defq->bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
569 defq->ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
570 defq->isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
571 defq->rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
572 defq->rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
573 xfs_qm_dqdestroy(dqp);
574 }
575
576 /* Initialize quota time limits from the root dquot. */
577 static void
578 xfs_qm_init_timelimits(
579 struct xfs_mount *mp,
580 struct xfs_quotainfo *qinf)
581 {
582 struct xfs_disk_dquot *ddqp;
583 struct xfs_dquot *dqp;
584 uint type;
585 int error;
586
587 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
588 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
589 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
590 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
591 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
592 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
593
594 /*
595 * We try to get the limits from the superuser's limits fields.
596 * This is quite hacky, but it is standard quota practice.
597 *
598 * Since we may not have done a quotacheck by this point, just read
599 * the dquot without attaching it to any hashtables or lists.
600 *
601 * Timers and warnings are globally set by the first timer found in
602 * user/group/proj quota types, otherwise a default value is used.
603 * This should be split into different fields per quota type.
604 */
605 if (XFS_IS_UQUOTA_RUNNING(mp))
606 type = XFS_DQ_USER;
607 else if (XFS_IS_GQUOTA_RUNNING(mp))
608 type = XFS_DQ_GROUP;
609 else
610 type = XFS_DQ_PROJ;
611 error = xfs_qm_dqget_uncached(mp, 0, type, &dqp);
612 if (error)
613 return;
614
615 ddqp = &dqp->q_core;
616 /*
617 * The warnings and timers set the grace period given to
618 * a user or group before he or she can not perform any
619 * more writing. If it is zero, a default is used.
620 */
621 if (ddqp->d_btimer)
622 qinf->qi_btimelimit = be32_to_cpu(ddqp->d_btimer);
623 if (ddqp->d_itimer)
624 qinf->qi_itimelimit = be32_to_cpu(ddqp->d_itimer);
625 if (ddqp->d_rtbtimer)
626 qinf->qi_rtbtimelimit = be32_to_cpu(ddqp->d_rtbtimer);
627 if (ddqp->d_bwarns)
628 qinf->qi_bwarnlimit = be16_to_cpu(ddqp->d_bwarns);
629 if (ddqp->d_iwarns)
630 qinf->qi_iwarnlimit = be16_to_cpu(ddqp->d_iwarns);
631 if (ddqp->d_rtbwarns)
632 qinf->qi_rtbwarnlimit = be16_to_cpu(ddqp->d_rtbwarns);
633
634 xfs_qm_dqdestroy(dqp);
635 }
636
637 /*
638 * This initializes all the quota information that's kept in the
639 * mount structure
640 */
641 STATIC int
642 xfs_qm_init_quotainfo(
643 struct xfs_mount *mp)
644 {
645 struct xfs_quotainfo *qinf;
646 int error;
647
648 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
649
650 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(struct xfs_quotainfo), 0);
651
652 error = list_lru_init(&qinf->qi_lru);
653 if (error)
654 goto out_free_qinf;
655
656 /*
657 * See if quotainodes are setup, and if not, allocate them,
658 * and change the superblock accordingly.
659 */
660 error = xfs_qm_init_quotainos(mp);
661 if (error)
662 goto out_free_lru;
663
664 INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
665 INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
666 INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS);
667 mutex_init(&qinf->qi_tree_lock);
668
669 /* mutex used to serialize quotaoffs */
670 mutex_init(&qinf->qi_quotaofflock);
671
672 /* Precalc some constants */
673 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
674 qinf->qi_dqperchunk = xfs_calc_dquots_per_chunk(qinf->qi_dqchunklen);
675
676 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
677
678 xfs_qm_init_timelimits(mp, qinf);
679
680 if (XFS_IS_UQUOTA_RUNNING(mp))
681 xfs_qm_set_defquota(mp, XFS_DQ_USER, qinf);
682 if (XFS_IS_GQUOTA_RUNNING(mp))
683 xfs_qm_set_defquota(mp, XFS_DQ_GROUP, qinf);
684 if (XFS_IS_PQUOTA_RUNNING(mp))
685 xfs_qm_set_defquota(mp, XFS_DQ_PROJ, qinf);
686
687 qinf->qi_shrinker.count_objects = xfs_qm_shrink_count;
688 qinf->qi_shrinker.scan_objects = xfs_qm_shrink_scan;
689 qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
690 qinf->qi_shrinker.flags = SHRINKER_NUMA_AWARE;
691
692 error = register_shrinker(&qinf->qi_shrinker);
693 if (error)
694 goto out_free_inos;
695
696 return 0;
697
698 out_free_inos:
699 mutex_destroy(&qinf->qi_quotaofflock);
700 mutex_destroy(&qinf->qi_tree_lock);
701 xfs_qm_destroy_quotainos(qinf);
702 out_free_lru:
703 list_lru_destroy(&qinf->qi_lru);
704 out_free_qinf:
705 kmem_free(qinf);
706 mp->m_quotainfo = NULL;
707 return error;
708 }
709
710 /*
711 * Gets called when unmounting a filesystem or when all quotas get
712 * turned off.
713 * This purges the quota inodes, destroys locks and frees itself.
714 */
715 void
716 xfs_qm_destroy_quotainfo(
717 struct xfs_mount *mp)
718 {
719 struct xfs_quotainfo *qi;
720
721 qi = mp->m_quotainfo;
722 ASSERT(qi != NULL);
723
724 unregister_shrinker(&qi->qi_shrinker);
725 list_lru_destroy(&qi->qi_lru);
726 xfs_qm_destroy_quotainos(qi);
727 mutex_destroy(&qi->qi_tree_lock);
728 mutex_destroy(&qi->qi_quotaofflock);
729 kmem_free(qi);
730 mp->m_quotainfo = NULL;
731 }
732
733 /*
734 * Create an inode and return with a reference already taken, but unlocked
735 * This is how we create quota inodes
736 */
737 STATIC int
738 xfs_qm_qino_alloc(
739 xfs_mount_t *mp,
740 xfs_inode_t **ip,
741 uint flags)
742 {
743 xfs_trans_t *tp;
744 int error;
745 bool need_alloc = true;
746
747 *ip = NULL;
748 /*
749 * With superblock that doesn't have separate pquotino, we
750 * share an inode between gquota and pquota. If the on-disk
751 * superblock has GQUOTA and the filesystem is now mounted
752 * with PQUOTA, just use sb_gquotino for sb_pquotino and
753 * vice-versa.
754 */
755 if (!xfs_sb_version_has_pquotino(&mp->m_sb) &&
756 (flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) {
757 xfs_ino_t ino = NULLFSINO;
758
759 if ((flags & XFS_QMOPT_PQUOTA) &&
760 (mp->m_sb.sb_gquotino != NULLFSINO)) {
761 ino = mp->m_sb.sb_gquotino;
762 if (XFS_IS_CORRUPT(mp,
763 mp->m_sb.sb_pquotino != NULLFSINO))
764 return -EFSCORRUPTED;
765 } else if ((flags & XFS_QMOPT_GQUOTA) &&
766 (mp->m_sb.sb_pquotino != NULLFSINO)) {
767 ino = mp->m_sb.sb_pquotino;
768 if (XFS_IS_CORRUPT(mp,
769 mp->m_sb.sb_gquotino != NULLFSINO))
770 return -EFSCORRUPTED;
771 }
772 if (ino != NULLFSINO) {
773 error = xfs_iget(mp, NULL, ino, 0, 0, ip);
774 if (error)
775 return error;
776 mp->m_sb.sb_gquotino = NULLFSINO;
777 mp->m_sb.sb_pquotino = NULLFSINO;
778 need_alloc = false;
779 }
780 }
781
782 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_create,
783 XFS_QM_QINOCREATE_SPACE_RES(mp), 0, 0, &tp);
784 if (error)
785 return error;
786
787 if (need_alloc) {
788 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, ip);
789 if (error) {
790 xfs_trans_cancel(tp);
791 return error;
792 }
793 }
794
795 /*
796 * Make the changes in the superblock, and log those too.
797 * sbfields arg may contain fields other than *QUOTINO;
798 * VERSIONNUM for example.
799 */
800 spin_lock(&mp->m_sb_lock);
801 if (flags & XFS_QMOPT_SBVERSION) {
802 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
803
804 xfs_sb_version_addquota(&mp->m_sb);
805 mp->m_sb.sb_uquotino = NULLFSINO;
806 mp->m_sb.sb_gquotino = NULLFSINO;
807 mp->m_sb.sb_pquotino = NULLFSINO;
808
809 /* qflags will get updated fully _after_ quotacheck */
810 mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT;
811 }
812 if (flags & XFS_QMOPT_UQUOTA)
813 mp->m_sb.sb_uquotino = (*ip)->i_ino;
814 else if (flags & XFS_QMOPT_GQUOTA)
815 mp->m_sb.sb_gquotino = (*ip)->i_ino;
816 else
817 mp->m_sb.sb_pquotino = (*ip)->i_ino;
818 spin_unlock(&mp->m_sb_lock);
819 xfs_log_sb(tp);
820
821 error = xfs_trans_commit(tp);
822 if (error) {
823 ASSERT(XFS_FORCED_SHUTDOWN(mp));
824 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
825 }
826 if (need_alloc)
827 xfs_finish_inode_setup(*ip);
828 return error;
829 }
830
831
832 STATIC void
833 xfs_qm_reset_dqcounts(
834 xfs_mount_t *mp,
835 xfs_buf_t *bp,
836 xfs_dqid_t id,
837 uint type)
838 {
839 struct xfs_dqblk *dqb;
840 int j;
841 xfs_failaddr_t fa;
842
843 trace_xfs_reset_dqcounts(bp, _RET_IP_);
844
845 /*
846 * Reset all counters and timers. They'll be
847 * started afresh by xfs_qm_quotacheck.
848 */
849 #ifdef DEBUG
850 j = (int)XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB) /
851 sizeof(xfs_dqblk_t);
852 ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
853 #endif
854 dqb = bp->b_addr;
855 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
856 struct xfs_disk_dquot *ddq;
857
858 ddq = (struct xfs_disk_dquot *)&dqb[j];
859
860 /*
861 * Do a sanity check, and if needed, repair the dqblk. Don't
862 * output any warnings because it's perfectly possible to
863 * find uninitialised dquot blks. See comment in
864 * xfs_dquot_verify.
865 */
866 fa = xfs_dqblk_verify(mp, &dqb[j], id + j, type);
867 if (fa)
868 xfs_dqblk_repair(mp, &dqb[j], id + j, type);
869
870 /*
871 * Reset type in case we are reusing group quota file for
872 * project quotas or vice versa
873 */
874 ddq->d_flags = type;
875 ddq->d_bcount = 0;
876 ddq->d_icount = 0;
877 ddq->d_rtbcount = 0;
878
879 /*
880 * dquot id 0 stores the default grace period and the maximum
881 * warning limit that were set by the administrator, so we
882 * should not reset them.
883 */
884 if (ddq->d_id != 0) {
885 ddq->d_btimer = 0;
886 ddq->d_itimer = 0;
887 ddq->d_rtbtimer = 0;
888 ddq->d_bwarns = 0;
889 ddq->d_iwarns = 0;
890 ddq->d_rtbwarns = 0;
891 }
892
893 if (xfs_sb_version_hascrc(&mp->m_sb)) {
894 xfs_update_cksum((char *)&dqb[j],
895 sizeof(struct xfs_dqblk),
896 XFS_DQUOT_CRC_OFF);
897 }
898 }
899 }
900
901 STATIC int
902 xfs_qm_reset_dqcounts_all(
903 struct xfs_mount *mp,
904 xfs_dqid_t firstid,
905 xfs_fsblock_t bno,
906 xfs_filblks_t blkcnt,
907 uint flags,
908 struct list_head *buffer_list)
909 {
910 struct xfs_buf *bp;
911 int error;
912 int type;
913
914 ASSERT(blkcnt > 0);
915 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
916 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
917 error = 0;
918
919 /*
920 * Blkcnt arg can be a very big number, and might even be
921 * larger than the log itself. So, we have to break it up into
922 * manageable-sized transactions.
923 * Note that we don't start a permanent transaction here; we might
924 * not be able to get a log reservation for the whole thing up front,
925 * and we don't really care to either, because we just discard
926 * everything if we were to crash in the middle of this loop.
927 */
928 while (blkcnt--) {
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 &xfs_dquot_buf_ops);
933
934 /*
935 * CRC and validation errors will return a EFSCORRUPTED here. If
936 * this occurs, re-read without CRC validation so that we can
937 * repair the damage via xfs_qm_reset_dqcounts(). This process
938 * will leave a trace in the log indicating corruption has
939 * been detected.
940 */
941 if (error == -EFSCORRUPTED) {
942 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
943 XFS_FSB_TO_DADDR(mp, bno),
944 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
945 NULL);
946 }
947
948 if (error)
949 break;
950
951 /*
952 * A corrupt buffer might not have a verifier attached, so
953 * make sure we have the correct one attached before writeback
954 * occurs.
955 */
956 bp->b_ops = &xfs_dquot_buf_ops;
957 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
958 xfs_buf_delwri_queue(bp, buffer_list);
959 xfs_buf_relse(bp);
960
961 /* goto the next block. */
962 bno++;
963 firstid += mp->m_quotainfo->qi_dqperchunk;
964 }
965
966 return error;
967 }
968
969 /*
970 * Iterate over all allocated dquot blocks in this quota inode, zeroing all
971 * counters for every chunk of dquots that we find.
972 */
973 STATIC int
974 xfs_qm_reset_dqcounts_buf(
975 struct xfs_mount *mp,
976 struct xfs_inode *qip,
977 uint flags,
978 struct list_head *buffer_list)
979 {
980 struct xfs_bmbt_irec *map;
981 int i, nmaps; /* number of map entries */
982 int error; /* return value */
983 xfs_fileoff_t lblkno;
984 xfs_filblks_t maxlblkcnt;
985 xfs_dqid_t firstid;
986 xfs_fsblock_t rablkno;
987 xfs_filblks_t rablkcnt;
988
989 error = 0;
990 /*
991 * This looks racy, but we can't keep an inode lock across a
992 * trans_reserve. But, this gets called during quotacheck, and that
993 * happens only at mount time which is single threaded.
994 */
995 if (qip->i_d.di_nblocks == 0)
996 return 0;
997
998 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), 0);
999
1000 lblkno = 0;
1001 maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1002 do {
1003 uint lock_mode;
1004
1005 nmaps = XFS_DQITER_MAP_SIZE;
1006 /*
1007 * We aren't changing the inode itself. Just changing
1008 * some of its data. No new blocks are added here, and
1009 * the inode is never added to the transaction.
1010 */
1011 lock_mode = xfs_ilock_data_map_shared(qip);
1012 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
1013 map, &nmaps, 0);
1014 xfs_iunlock(qip, lock_mode);
1015 if (error)
1016 break;
1017
1018 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1019 for (i = 0; i < nmaps; i++) {
1020 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1021 ASSERT(map[i].br_blockcount);
1022
1023
1024 lblkno += map[i].br_blockcount;
1025
1026 if (map[i].br_startblock == HOLESTARTBLOCK)
1027 continue;
1028
1029 firstid = (xfs_dqid_t) map[i].br_startoff *
1030 mp->m_quotainfo->qi_dqperchunk;
1031 /*
1032 * Do a read-ahead on the next extent.
1033 */
1034 if ((i+1 < nmaps) &&
1035 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1036 rablkcnt = map[i+1].br_blockcount;
1037 rablkno = map[i+1].br_startblock;
1038 while (rablkcnt--) {
1039 xfs_buf_readahead(mp->m_ddev_targp,
1040 XFS_FSB_TO_DADDR(mp, rablkno),
1041 mp->m_quotainfo->qi_dqchunklen,
1042 &xfs_dquot_buf_ops);
1043 rablkno++;
1044 }
1045 }
1046 /*
1047 * Iterate thru all the blks in the extent and
1048 * reset the counters of all the dquots inside them.
1049 */
1050 error = xfs_qm_reset_dqcounts_all(mp, firstid,
1051 map[i].br_startblock,
1052 map[i].br_blockcount,
1053 flags, buffer_list);
1054 if (error)
1055 goto out;
1056 }
1057 } while (nmaps > 0);
1058
1059 out:
1060 kmem_free(map);
1061 return error;
1062 }
1063
1064 /*
1065 * Called by dqusage_adjust in doing a quotacheck.
1066 *
1067 * Given the inode, and a dquot id this updates both the incore dqout as well
1068 * as the buffer copy. This is so that once the quotacheck is done, we can
1069 * just log all the buffers, as opposed to logging numerous updates to
1070 * individual dquots.
1071 */
1072 STATIC int
1073 xfs_qm_quotacheck_dqadjust(
1074 struct xfs_inode *ip,
1075 uint type,
1076 xfs_qcnt_t nblks,
1077 xfs_qcnt_t rtblks)
1078 {
1079 struct xfs_mount *mp = ip->i_mount;
1080 struct xfs_dquot *dqp;
1081 xfs_dqid_t id;
1082 int error;
1083
1084 id = xfs_qm_id_for_quotatype(ip, type);
1085 error = xfs_qm_dqget(mp, id, type, true, &dqp);
1086 if (error) {
1087 /*
1088 * Shouldn't be able to turn off quotas here.
1089 */
1090 ASSERT(error != -ESRCH);
1091 ASSERT(error != -ENOENT);
1092 return error;
1093 }
1094
1095 trace_xfs_dqadjust(dqp);
1096
1097 /*
1098 * Adjust the inode count and the block count to reflect this inode's
1099 * resource usage.
1100 */
1101 be64_add_cpu(&dqp->q_core.d_icount, 1);
1102 dqp->q_res_icount++;
1103 if (nblks) {
1104 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1105 dqp->q_res_bcount += nblks;
1106 }
1107 if (rtblks) {
1108 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1109 dqp->q_res_rtbcount += rtblks;
1110 }
1111
1112 /*
1113 * Set default limits, adjust timers (since we changed usages)
1114 *
1115 * There are no timers for the default values set in the root dquot.
1116 */
1117 if (dqp->q_core.d_id) {
1118 xfs_qm_adjust_dqlimits(mp, dqp);
1119 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1120 }
1121
1122 dqp->dq_flags |= XFS_DQ_DIRTY;
1123 xfs_qm_dqput(dqp);
1124 return 0;
1125 }
1126
1127 /*
1128 * callback routine supplied to bulkstat(). Given an inumber, find its
1129 * dquots and update them to account for resources taken by that inode.
1130 */
1131 /* ARGSUSED */
1132 STATIC int
1133 xfs_qm_dqusage_adjust(
1134 struct xfs_mount *mp,
1135 struct xfs_trans *tp,
1136 xfs_ino_t ino,
1137 void *data)
1138 {
1139 struct xfs_inode *ip;
1140 xfs_qcnt_t nblks;
1141 xfs_filblks_t rtblks = 0; /* total rt blks */
1142 int error;
1143
1144 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1145
1146 /*
1147 * rootino must have its resources accounted for, not so with the quota
1148 * inodes.
1149 */
1150 if (xfs_is_quota_inode(&mp->m_sb, ino))
1151 return 0;
1152
1153 /*
1154 * We don't _need_ to take the ilock EXCL here because quotacheck runs
1155 * at mount time and therefore nobody will be racing chown/chproj.
1156 */
1157 error = xfs_iget(mp, tp, ino, XFS_IGET_DONTCACHE, 0, &ip);
1158 if (error == -EINVAL || error == -ENOENT)
1159 return 0;
1160 if (error)
1161 return error;
1162
1163 ASSERT(ip->i_delayed_blks == 0);
1164
1165 if (XFS_IS_REALTIME_INODE(ip)) {
1166 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1167
1168 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1169 error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
1170 if (error)
1171 goto error0;
1172 }
1173
1174 xfs_bmap_count_leaves(ifp, &rtblks);
1175 }
1176
1177 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1178
1179 /*
1180 * Add the (disk blocks and inode) resources occupied by this
1181 * inode to its dquots. We do this adjustment in the incore dquot,
1182 * and also copy the changes to its buffer.
1183 * We don't care about putting these changes in a transaction
1184 * envelope because if we crash in the middle of a 'quotacheck'
1185 * we have to start from the beginning anyway.
1186 * Once we're done, we'll log all the dquot bufs.
1187 *
1188 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1189 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1190 */
1191 if (XFS_IS_UQUOTA_ON(mp)) {
1192 error = xfs_qm_quotacheck_dqadjust(ip, XFS_DQ_USER, nblks,
1193 rtblks);
1194 if (error)
1195 goto error0;
1196 }
1197
1198 if (XFS_IS_GQUOTA_ON(mp)) {
1199 error = xfs_qm_quotacheck_dqadjust(ip, XFS_DQ_GROUP, nblks,
1200 rtblks);
1201 if (error)
1202 goto error0;
1203 }
1204
1205 if (XFS_IS_PQUOTA_ON(mp)) {
1206 error = xfs_qm_quotacheck_dqadjust(ip, XFS_DQ_PROJ, nblks,
1207 rtblks);
1208 if (error)
1209 goto error0;
1210 }
1211
1212 error0:
1213 xfs_irele(ip);
1214 return error;
1215 }
1216
1217 STATIC int
1218 xfs_qm_flush_one(
1219 struct xfs_dquot *dqp,
1220 void *data)
1221 {
1222 struct xfs_mount *mp = dqp->q_mount;
1223 struct list_head *buffer_list = data;
1224 struct xfs_buf *bp = NULL;
1225 int error = 0;
1226
1227 xfs_dqlock(dqp);
1228 if (dqp->dq_flags & XFS_DQ_FREEING)
1229 goto out_unlock;
1230 if (!XFS_DQ_IS_DIRTY(dqp))
1231 goto out_unlock;
1232
1233 /*
1234 * The only way the dquot is already flush locked by the time quotacheck
1235 * gets here is if reclaim flushed it before the dqadjust walk dirtied
1236 * it for the final time. Quotacheck collects all dquot bufs in the
1237 * local delwri queue before dquots are dirtied, so reclaim can't have
1238 * possibly queued it for I/O. The only way out is to push the buffer to
1239 * cycle the flush lock.
1240 */
1241 if (!xfs_dqflock_nowait(dqp)) {
1242 /* buf is pinned in-core by delwri list */
1243 bp = xfs_buf_incore(mp->m_ddev_targp, dqp->q_blkno,
1244 mp->m_quotainfo->qi_dqchunklen, 0);
1245 if (!bp) {
1246 error = -EINVAL;
1247 goto out_unlock;
1248 }
1249 xfs_buf_unlock(bp);
1250
1251 xfs_buf_delwri_pushbuf(bp, buffer_list);
1252 xfs_buf_rele(bp);
1253
1254 error = -EAGAIN;
1255 goto out_unlock;
1256 }
1257
1258 error = xfs_qm_dqflush(dqp, &bp);
1259 if (error)
1260 goto out_unlock;
1261
1262 xfs_buf_delwri_queue(bp, buffer_list);
1263 xfs_buf_relse(bp);
1264 out_unlock:
1265 xfs_dqunlock(dqp);
1266 return error;
1267 }
1268
1269 /*
1270 * Walk thru all the filesystem inodes and construct a consistent view
1271 * of the disk quota world. If the quotacheck fails, disable quotas.
1272 */
1273 STATIC int
1274 xfs_qm_quotacheck(
1275 xfs_mount_t *mp)
1276 {
1277 int error, error2;
1278 uint flags;
1279 LIST_HEAD (buffer_list);
1280 struct xfs_inode *uip = mp->m_quotainfo->qi_uquotaip;
1281 struct xfs_inode *gip = mp->m_quotainfo->qi_gquotaip;
1282 struct xfs_inode *pip = mp->m_quotainfo->qi_pquotaip;
1283
1284 flags = 0;
1285
1286 ASSERT(uip || gip || pip);
1287 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1288
1289 xfs_notice(mp, "Quotacheck needed: Please wait.");
1290
1291 /*
1292 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1293 * their counters to zero. We need a clean slate.
1294 * We don't log our changes till later.
1295 */
1296 if (uip) {
1297 error = xfs_qm_reset_dqcounts_buf(mp, uip, XFS_QMOPT_UQUOTA,
1298 &buffer_list);
1299 if (error)
1300 goto error_return;
1301 flags |= XFS_UQUOTA_CHKD;
1302 }
1303
1304 if (gip) {
1305 error = xfs_qm_reset_dqcounts_buf(mp, gip, XFS_QMOPT_GQUOTA,
1306 &buffer_list);
1307 if (error)
1308 goto error_return;
1309 flags |= XFS_GQUOTA_CHKD;
1310 }
1311
1312 if (pip) {
1313 error = xfs_qm_reset_dqcounts_buf(mp, pip, XFS_QMOPT_PQUOTA,
1314 &buffer_list);
1315 if (error)
1316 goto error_return;
1317 flags |= XFS_PQUOTA_CHKD;
1318 }
1319
1320 error = xfs_iwalk_threaded(mp, 0, 0, xfs_qm_dqusage_adjust, 0, true,
1321 NULL);
1322 if (error)
1323 goto error_return;
1324
1325 /*
1326 * We've made all the changes that we need to make incore. Flush them
1327 * down to disk buffers if everything was updated successfully.
1328 */
1329 if (XFS_IS_UQUOTA_ON(mp)) {
1330 error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one,
1331 &buffer_list);
1332 }
1333 if (XFS_IS_GQUOTA_ON(mp)) {
1334 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one,
1335 &buffer_list);
1336 if (!error)
1337 error = error2;
1338 }
1339 if (XFS_IS_PQUOTA_ON(mp)) {
1340 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one,
1341 &buffer_list);
1342 if (!error)
1343 error = error2;
1344 }
1345
1346 error2 = xfs_buf_delwri_submit(&buffer_list);
1347 if (!error)
1348 error = error2;
1349
1350 /*
1351 * We can get this error if we couldn't do a dquot allocation inside
1352 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1353 * dirty dquots that might be cached, we just want to get rid of them
1354 * and turn quotaoff. The dquots won't be attached to any of the inodes
1355 * at this point (because we intentionally didn't in dqget_noattach).
1356 */
1357 if (error) {
1358 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1359 goto error_return;
1360 }
1361
1362 /*
1363 * If one type of quotas is off, then it will lose its
1364 * quotachecked status, since we won't be doing accounting for
1365 * that type anymore.
1366 */
1367 mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1368 mp->m_qflags |= flags;
1369
1370 error_return:
1371 xfs_buf_delwri_cancel(&buffer_list);
1372
1373 if (error) {
1374 xfs_warn(mp,
1375 "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1376 error);
1377 /*
1378 * We must turn off quotas.
1379 */
1380 ASSERT(mp->m_quotainfo != NULL);
1381 xfs_qm_destroy_quotainfo(mp);
1382 if (xfs_mount_reset_sbqflags(mp)) {
1383 xfs_warn(mp,
1384 "Quotacheck: Failed to reset quota flags.");
1385 }
1386 } else
1387 xfs_notice(mp, "Quotacheck: Done.");
1388 return error;
1389 }
1390
1391 /*
1392 * This is called from xfs_mountfs to start quotas and initialize all
1393 * necessary data structures like quotainfo. This is also responsible for
1394 * running a quotacheck as necessary. We are guaranteed that the superblock
1395 * is consistently read in at this point.
1396 *
1397 * If we fail here, the mount will continue with quota turned off. We don't
1398 * need to inidicate success or failure at all.
1399 */
1400 void
1401 xfs_qm_mount_quotas(
1402 struct xfs_mount *mp)
1403 {
1404 int error = 0;
1405 uint sbf;
1406
1407 /*
1408 * If quotas on realtime volumes is not supported, we disable
1409 * quotas immediately.
1410 */
1411 if (mp->m_sb.sb_rextents) {
1412 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
1413 mp->m_qflags = 0;
1414 goto write_changes;
1415 }
1416
1417 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1418
1419 /*
1420 * Allocate the quotainfo structure inside the mount struct, and
1421 * create quotainode(s), and change/rev superblock if necessary.
1422 */
1423 error = xfs_qm_init_quotainfo(mp);
1424 if (error) {
1425 /*
1426 * We must turn off quotas.
1427 */
1428 ASSERT(mp->m_quotainfo == NULL);
1429 mp->m_qflags = 0;
1430 goto write_changes;
1431 }
1432 /*
1433 * If any of the quotas are not consistent, do a quotacheck.
1434 */
1435 if (XFS_QM_NEED_QUOTACHECK(mp)) {
1436 error = xfs_qm_quotacheck(mp);
1437 if (error) {
1438 /* Quotacheck failed and disabled quotas. */
1439 return;
1440 }
1441 }
1442 /*
1443 * If one type of quotas is off, then it will lose its
1444 * quotachecked status, since we won't be doing accounting for
1445 * that type anymore.
1446 */
1447 if (!XFS_IS_UQUOTA_ON(mp))
1448 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
1449 if (!XFS_IS_GQUOTA_ON(mp))
1450 mp->m_qflags &= ~XFS_GQUOTA_CHKD;
1451 if (!XFS_IS_PQUOTA_ON(mp))
1452 mp->m_qflags &= ~XFS_PQUOTA_CHKD;
1453
1454 write_changes:
1455 /*
1456 * We actually don't have to acquire the m_sb_lock at all.
1457 * This can only be called from mount, and that's single threaded. XXX
1458 */
1459 spin_lock(&mp->m_sb_lock);
1460 sbf = mp->m_sb.sb_qflags;
1461 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
1462 spin_unlock(&mp->m_sb_lock);
1463
1464 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
1465 if (xfs_sync_sb(mp, false)) {
1466 /*
1467 * We could only have been turning quotas off.
1468 * We aren't in very good shape actually because
1469 * the incore structures are convinced that quotas are
1470 * off, but the on disk superblock doesn't know that !
1471 */
1472 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
1473 xfs_alert(mp, "%s: Superblock update failed!",
1474 __func__);
1475 }
1476 }
1477
1478 if (error) {
1479 xfs_warn(mp, "Failed to initialize disk quotas.");
1480 return;
1481 }
1482 }
1483
1484 /*
1485 * This is called after the superblock has been read in and we're ready to
1486 * iget the quota inodes.
1487 */
1488 STATIC int
1489 xfs_qm_init_quotainos(
1490 xfs_mount_t *mp)
1491 {
1492 struct xfs_inode *uip = NULL;
1493 struct xfs_inode *gip = NULL;
1494 struct xfs_inode *pip = NULL;
1495 int error;
1496 uint flags = 0;
1497
1498 ASSERT(mp->m_quotainfo);
1499
1500 /*
1501 * Get the uquota and gquota inodes
1502 */
1503 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1504 if (XFS_IS_UQUOTA_ON(mp) &&
1505 mp->m_sb.sb_uquotino != NULLFSINO) {
1506 ASSERT(mp->m_sb.sb_uquotino > 0);
1507 error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1508 0, 0, &uip);
1509 if (error)
1510 return error;
1511 }
1512 if (XFS_IS_GQUOTA_ON(mp) &&
1513 mp->m_sb.sb_gquotino != NULLFSINO) {
1514 ASSERT(mp->m_sb.sb_gquotino > 0);
1515 error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1516 0, 0, &gip);
1517 if (error)
1518 goto error_rele;
1519 }
1520 if (XFS_IS_PQUOTA_ON(mp) &&
1521 mp->m_sb.sb_pquotino != NULLFSINO) {
1522 ASSERT(mp->m_sb.sb_pquotino > 0);
1523 error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino,
1524 0, 0, &pip);
1525 if (error)
1526 goto error_rele;
1527 }
1528 } else {
1529 flags |= XFS_QMOPT_SBVERSION;
1530 }
1531
1532 /*
1533 * Create the three inodes, if they don't exist already. The changes
1534 * made above will get added to a transaction and logged in one of
1535 * the qino_alloc calls below. If the device is readonly,
1536 * temporarily switch to read-write to do this.
1537 */
1538 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1539 error = xfs_qm_qino_alloc(mp, &uip,
1540 flags | XFS_QMOPT_UQUOTA);
1541 if (error)
1542 goto error_rele;
1543
1544 flags &= ~XFS_QMOPT_SBVERSION;
1545 }
1546 if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
1547 error = xfs_qm_qino_alloc(mp, &gip,
1548 flags | XFS_QMOPT_GQUOTA);
1549 if (error)
1550 goto error_rele;
1551
1552 flags &= ~XFS_QMOPT_SBVERSION;
1553 }
1554 if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) {
1555 error = xfs_qm_qino_alloc(mp, &pip,
1556 flags | XFS_QMOPT_PQUOTA);
1557 if (error)
1558 goto error_rele;
1559 }
1560
1561 mp->m_quotainfo->qi_uquotaip = uip;
1562 mp->m_quotainfo->qi_gquotaip = gip;
1563 mp->m_quotainfo->qi_pquotaip = pip;
1564
1565 return 0;
1566
1567 error_rele:
1568 if (uip)
1569 xfs_irele(uip);
1570 if (gip)
1571 xfs_irele(gip);
1572 if (pip)
1573 xfs_irele(pip);
1574 return error;
1575 }
1576
1577 STATIC void
1578 xfs_qm_destroy_quotainos(
1579 struct xfs_quotainfo *qi)
1580 {
1581 if (qi->qi_uquotaip) {
1582 xfs_irele(qi->qi_uquotaip);
1583 qi->qi_uquotaip = NULL; /* paranoia */
1584 }
1585 if (qi->qi_gquotaip) {
1586 xfs_irele(qi->qi_gquotaip);
1587 qi->qi_gquotaip = NULL;
1588 }
1589 if (qi->qi_pquotaip) {
1590 xfs_irele(qi->qi_pquotaip);
1591 qi->qi_pquotaip = NULL;
1592 }
1593 }
1594
1595 STATIC void
1596 xfs_qm_dqfree_one(
1597 struct xfs_dquot *dqp)
1598 {
1599 struct xfs_mount *mp = dqp->q_mount;
1600 struct xfs_quotainfo *qi = mp->m_quotainfo;
1601
1602 mutex_lock(&qi->qi_tree_lock);
1603 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
1604 be32_to_cpu(dqp->q_core.d_id));
1605
1606 qi->qi_dquots--;
1607 mutex_unlock(&qi->qi_tree_lock);
1608
1609 xfs_qm_dqdestroy(dqp);
1610 }
1611
1612 /* --------------- utility functions for vnodeops ---------------- */
1613
1614
1615 /*
1616 * Given an inode, a uid, gid and prid make sure that we have
1617 * allocated relevant dquot(s) on disk, and that we won't exceed inode
1618 * quotas by creating this file.
1619 * This also attaches dquot(s) to the given inode after locking it,
1620 * and returns the dquots corresponding to the uid and/or gid.
1621 *
1622 * in : inode (unlocked)
1623 * out : udquot, gdquot with references taken and unlocked
1624 */
1625 int
1626 xfs_qm_vop_dqalloc(
1627 struct xfs_inode *ip,
1628 kuid_t uid,
1629 kgid_t gid,
1630 prid_t prid,
1631 uint flags,
1632 struct xfs_dquot **O_udqpp,
1633 struct xfs_dquot **O_gdqpp,
1634 struct xfs_dquot **O_pdqpp)
1635 {
1636 struct xfs_mount *mp = ip->i_mount;
1637 struct inode *inode = VFS_I(ip);
1638 struct user_namespace *user_ns = inode->i_sb->s_user_ns;
1639 struct xfs_dquot *uq = NULL;
1640 struct xfs_dquot *gq = NULL;
1641 struct xfs_dquot *pq = NULL;
1642 int error;
1643 uint lockflags;
1644
1645 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1646 return 0;
1647
1648 lockflags = XFS_ILOCK_EXCL;
1649 xfs_ilock(ip, lockflags);
1650
1651 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1652 gid = inode->i_gid;
1653
1654 /*
1655 * Attach the dquot(s) to this inode, doing a dquot allocation
1656 * if necessary. The dquot(s) will not be locked.
1657 */
1658 if (XFS_NOT_DQATTACHED(mp, ip)) {
1659 error = xfs_qm_dqattach_locked(ip, true);
1660 if (error) {
1661 xfs_iunlock(ip, lockflags);
1662 return error;
1663 }
1664 }
1665
1666 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1667 if (!uid_eq(inode->i_uid, uid)) {
1668 /*
1669 * What we need is the dquot that has this uid, and
1670 * if we send the inode to dqget, the uid of the inode
1671 * takes priority over what's sent in the uid argument.
1672 * We must unlock inode here before calling dqget if
1673 * we're not sending the inode, because otherwise
1674 * we'll deadlock by doing trans_reserve while
1675 * holding ilock.
1676 */
1677 xfs_iunlock(ip, lockflags);
1678 error = xfs_qm_dqget(mp, from_kuid(user_ns, uid),
1679 XFS_DQ_USER, true, &uq);
1680 if (error) {
1681 ASSERT(error != -ENOENT);
1682 return error;
1683 }
1684 /*
1685 * Get the ilock in the right order.
1686 */
1687 xfs_dqunlock(uq);
1688 lockflags = XFS_ILOCK_SHARED;
1689 xfs_ilock(ip, lockflags);
1690 } else {
1691 /*
1692 * Take an extra reference, because we'll return
1693 * this to caller
1694 */
1695 ASSERT(ip->i_udquot);
1696 uq = xfs_qm_dqhold(ip->i_udquot);
1697 }
1698 }
1699 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1700 if (!gid_eq(inode->i_gid, gid)) {
1701 xfs_iunlock(ip, lockflags);
1702 error = xfs_qm_dqget(mp, from_kgid(user_ns, gid),
1703 XFS_DQ_GROUP, true, &gq);
1704 if (error) {
1705 ASSERT(error != -ENOENT);
1706 goto error_rele;
1707 }
1708 xfs_dqunlock(gq);
1709 lockflags = XFS_ILOCK_SHARED;
1710 xfs_ilock(ip, lockflags);
1711 } else {
1712 ASSERT(ip->i_gdquot);
1713 gq = xfs_qm_dqhold(ip->i_gdquot);
1714 }
1715 }
1716 if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1717 if (ip->i_d.di_projid != prid) {
1718 xfs_iunlock(ip, lockflags);
1719 error = xfs_qm_dqget(mp, (xfs_dqid_t)prid, XFS_DQ_PROJ,
1720 true, &pq);
1721 if (error) {
1722 ASSERT(error != -ENOENT);
1723 goto error_rele;
1724 }
1725 xfs_dqunlock(pq);
1726 lockflags = XFS_ILOCK_SHARED;
1727 xfs_ilock(ip, lockflags);
1728 } else {
1729 ASSERT(ip->i_pdquot);
1730 pq = xfs_qm_dqhold(ip->i_pdquot);
1731 }
1732 }
1733 if (uq)
1734 trace_xfs_dquot_dqalloc(ip);
1735
1736 xfs_iunlock(ip, lockflags);
1737 if (O_udqpp)
1738 *O_udqpp = uq;
1739 else
1740 xfs_qm_dqrele(uq);
1741 if (O_gdqpp)
1742 *O_gdqpp = gq;
1743 else
1744 xfs_qm_dqrele(gq);
1745 if (O_pdqpp)
1746 *O_pdqpp = pq;
1747 else
1748 xfs_qm_dqrele(pq);
1749 return 0;
1750
1751 error_rele:
1752 xfs_qm_dqrele(gq);
1753 xfs_qm_dqrele(uq);
1754 return error;
1755 }
1756
1757 /*
1758 * Actually transfer ownership, and do dquot modifications.
1759 * These were already reserved.
1760 */
1761 struct xfs_dquot *
1762 xfs_qm_vop_chown(
1763 struct xfs_trans *tp,
1764 struct xfs_inode *ip,
1765 struct xfs_dquot **IO_olddq,
1766 struct xfs_dquot *newdq)
1767 {
1768 struct xfs_dquot *prevdq;
1769 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
1770 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1771
1772
1773 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1774 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1775
1776 /* old dquot */
1777 prevdq = *IO_olddq;
1778 ASSERT(prevdq);
1779 ASSERT(prevdq != newdq);
1780
1781 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
1782 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1783
1784 /* the sparkling new dquot */
1785 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
1786 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1787
1788 /*
1789 * Take an extra reference, because the inode is going to keep
1790 * this dquot pointer even after the trans_commit.
1791 */
1792 *IO_olddq = xfs_qm_dqhold(newdq);
1793
1794 return prevdq;
1795 }
1796
1797 /*
1798 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1799 */
1800 int
1801 xfs_qm_vop_chown_reserve(
1802 struct xfs_trans *tp,
1803 struct xfs_inode *ip,
1804 struct xfs_dquot *udqp,
1805 struct xfs_dquot *gdqp,
1806 struct xfs_dquot *pdqp,
1807 uint flags)
1808 {
1809 struct xfs_mount *mp = ip->i_mount;
1810 uint64_t delblks;
1811 unsigned int blkflags, prjflags = 0;
1812 struct xfs_dquot *udq_unres = NULL;
1813 struct xfs_dquot *gdq_unres = NULL;
1814 struct xfs_dquot *pdq_unres = NULL;
1815 struct xfs_dquot *udq_delblks = NULL;
1816 struct xfs_dquot *gdq_delblks = NULL;
1817 struct xfs_dquot *pdq_delblks = NULL;
1818 int error;
1819
1820
1821 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1822 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1823
1824 delblks = ip->i_delayed_blks;
1825 blkflags = XFS_IS_REALTIME_INODE(ip) ?
1826 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
1827
1828 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
1829 i_uid_read(VFS_I(ip)) != be32_to_cpu(udqp->q_core.d_id)) {
1830 udq_delblks = udqp;
1831 /*
1832 * If there are delayed allocation blocks, then we have to
1833 * unreserve those from the old dquot, and add them to the
1834 * new dquot.
1835 */
1836 if (delblks) {
1837 ASSERT(ip->i_udquot);
1838 udq_unres = ip->i_udquot;
1839 }
1840 }
1841 if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp &&
1842 i_gid_read(VFS_I(ip)) != be32_to_cpu(gdqp->q_core.d_id)) {
1843 gdq_delblks = gdqp;
1844 if (delblks) {
1845 ASSERT(ip->i_gdquot);
1846 gdq_unres = ip->i_gdquot;
1847 }
1848 }
1849
1850 if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
1851 ip->i_d.di_projid != be32_to_cpu(pdqp->q_core.d_id)) {
1852 prjflags = XFS_QMOPT_ENOSPC;
1853 pdq_delblks = pdqp;
1854 if (delblks) {
1855 ASSERT(ip->i_pdquot);
1856 pdq_unres = ip->i_pdquot;
1857 }
1858 }
1859
1860 error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
1861 udq_delblks, gdq_delblks, pdq_delblks,
1862 ip->i_d.di_nblocks, 1,
1863 flags | blkflags | prjflags);
1864 if (error)
1865 return error;
1866
1867 /*
1868 * Do the delayed blks reservations/unreservations now. Since, these
1869 * are done without the help of a transaction, if a reservation fails
1870 * its previous reservations won't be automatically undone by trans
1871 * code. So, we have to do it manually here.
1872 */
1873 if (delblks) {
1874 /*
1875 * Do the reservations first. Unreservation can't fail.
1876 */
1877 ASSERT(udq_delblks || gdq_delblks || pdq_delblks);
1878 ASSERT(udq_unres || gdq_unres || pdq_unres);
1879 error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
1880 udq_delblks, gdq_delblks, pdq_delblks,
1881 (xfs_qcnt_t)delblks, 0,
1882 flags | blkflags | prjflags);
1883 if (error)
1884 return error;
1885 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
1886 udq_unres, gdq_unres, pdq_unres,
1887 -((xfs_qcnt_t)delblks), 0, blkflags);
1888 }
1889
1890 return 0;
1891 }
1892
1893 int
1894 xfs_qm_vop_rename_dqattach(
1895 struct xfs_inode **i_tab)
1896 {
1897 struct xfs_mount *mp = i_tab[0]->i_mount;
1898 int i;
1899
1900 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1901 return 0;
1902
1903 for (i = 0; (i < 4 && i_tab[i]); i++) {
1904 struct xfs_inode *ip = i_tab[i];
1905 int error;
1906
1907 /*
1908 * Watch out for duplicate entries in the table.
1909 */
1910 if (i == 0 || ip != i_tab[i-1]) {
1911 if (XFS_NOT_DQATTACHED(mp, ip)) {
1912 error = xfs_qm_dqattach(ip);
1913 if (error)
1914 return error;
1915 }
1916 }
1917 }
1918 return 0;
1919 }
1920
1921 void
1922 xfs_qm_vop_create_dqattach(
1923 struct xfs_trans *tp,
1924 struct xfs_inode *ip,
1925 struct xfs_dquot *udqp,
1926 struct xfs_dquot *gdqp,
1927 struct xfs_dquot *pdqp)
1928 {
1929 struct xfs_mount *mp = tp->t_mountp;
1930
1931 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1932 return;
1933
1934 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1935 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1936
1937 if (udqp && XFS_IS_UQUOTA_ON(mp)) {
1938 ASSERT(ip->i_udquot == NULL);
1939 ASSERT(i_uid_read(VFS_I(ip)) == be32_to_cpu(udqp->q_core.d_id));
1940
1941 ip->i_udquot = xfs_qm_dqhold(udqp);
1942 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
1943 }
1944 if (gdqp && XFS_IS_GQUOTA_ON(mp)) {
1945 ASSERT(ip->i_gdquot == NULL);
1946 ASSERT(i_gid_read(VFS_I(ip)) == be32_to_cpu(gdqp->q_core.d_id));
1947
1948 ip->i_gdquot = xfs_qm_dqhold(gdqp);
1949 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
1950 }
1951 if (pdqp && XFS_IS_PQUOTA_ON(mp)) {
1952 ASSERT(ip->i_pdquot == NULL);
1953 ASSERT(ip->i_d.di_projid == be32_to_cpu(pdqp->q_core.d_id));
1954
1955 ip->i_pdquot = xfs_qm_dqhold(pdqp);
1956 xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1);
1957 }
1958 }
1959