]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/xfs/xfs_qm.c
Merge tag 'blackfin-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/realm...
[mirror_ubuntu-zesty-kernel.git] / fs / xfs / xfs_qm.c
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_inode.h"
29 #include "xfs_ialloc.h"
30 #include "xfs_itable.h"
31 #include "xfs_quota.h"
32 #include "xfs_error.h"
33 #include "xfs_bmap.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_trans.h"
36 #include "xfs_trans_space.h"
37 #include "xfs_qm.h"
38 #include "xfs_trace.h"
39 #include "xfs_icache.h"
40 #include "xfs_cksum.h"
41 #include "xfs_dinode.h"
42
43 /*
44 * The global quota manager. There is only one of these for the entire
45 * system, _not_ one per file system. XQM keeps track of the overall
46 * quota functionality, including maintaining the freelist and hash
47 * tables of dquots.
48 */
49 STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
50 STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
51
52
53 STATIC void xfs_qm_dqfree_one(struct xfs_dquot *dqp);
54 /*
55 * We use the batch lookup interface to iterate over the dquots as it
56 * currently is the only interface into the radix tree code that allows
57 * fuzzy lookups instead of exact matches. Holding the lock over multiple
58 * operations is fine as all callers are used either during mount/umount
59 * or quotaoff.
60 */
61 #define XFS_DQ_LOOKUP_BATCH 32
62
63 STATIC int
64 xfs_qm_dquot_walk(
65 struct xfs_mount *mp,
66 int type,
67 int (*execute)(struct xfs_dquot *dqp, void *data),
68 void *data)
69 {
70 struct xfs_quotainfo *qi = mp->m_quotainfo;
71 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
72 uint32_t next_index;
73 int last_error = 0;
74 int skipped;
75 int nr_found;
76
77 restart:
78 skipped = 0;
79 next_index = 0;
80 nr_found = 0;
81
82 while (1) {
83 struct xfs_dquot *batch[XFS_DQ_LOOKUP_BATCH];
84 int error = 0;
85 int i;
86
87 mutex_lock(&qi->qi_tree_lock);
88 nr_found = radix_tree_gang_lookup(tree, (void **)batch,
89 next_index, XFS_DQ_LOOKUP_BATCH);
90 if (!nr_found) {
91 mutex_unlock(&qi->qi_tree_lock);
92 break;
93 }
94
95 for (i = 0; i < nr_found; i++) {
96 struct xfs_dquot *dqp = batch[i];
97
98 next_index = be32_to_cpu(dqp->q_core.d_id) + 1;
99
100 error = execute(batch[i], data);
101 if (error == EAGAIN) {
102 skipped++;
103 continue;
104 }
105 if (error && last_error != EFSCORRUPTED)
106 last_error = error;
107 }
108
109 mutex_unlock(&qi->qi_tree_lock);
110
111 /* bail out if the filesystem is corrupted. */
112 if (last_error == EFSCORRUPTED) {
113 skipped = 0;
114 break;
115 }
116 }
117
118 if (skipped) {
119 delay(1);
120 goto restart;
121 }
122
123 return last_error;
124 }
125
126
127 /*
128 * Purge a dquot from all tracking data structures and free it.
129 */
130 STATIC int
131 xfs_qm_dqpurge(
132 struct xfs_dquot *dqp,
133 void *data)
134 {
135 struct xfs_mount *mp = dqp->q_mount;
136 struct xfs_quotainfo *qi = mp->m_quotainfo;
137
138 xfs_dqlock(dqp);
139 if ((dqp->dq_flags & XFS_DQ_FREEING) || dqp->q_nrefs != 0) {
140 xfs_dqunlock(dqp);
141 return EAGAIN;
142 }
143
144 dqp->dq_flags |= XFS_DQ_FREEING;
145
146 xfs_dqflock(dqp);
147
148 /*
149 * If we are turning this type of quotas off, we don't care
150 * about the dirty metadata sitting in this dquot. OTOH, if
151 * we're unmounting, we do care, so we flush it and wait.
152 */
153 if (XFS_DQ_IS_DIRTY(dqp)) {
154 struct xfs_buf *bp = NULL;
155 int error;
156
157 /*
158 * We don't care about getting disk errors here. We need
159 * to purge this dquot anyway, so we go ahead regardless.
160 */
161 error = xfs_qm_dqflush(dqp, &bp);
162 if (error) {
163 xfs_warn(mp, "%s: dquot %p flush failed",
164 __func__, dqp);
165 } else {
166 error = xfs_bwrite(bp);
167 xfs_buf_relse(bp);
168 }
169 xfs_dqflock(dqp);
170 }
171
172 ASSERT(atomic_read(&dqp->q_pincount) == 0);
173 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
174 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
175
176 xfs_dqfunlock(dqp);
177 xfs_dqunlock(dqp);
178
179 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
180 be32_to_cpu(dqp->q_core.d_id));
181 qi->qi_dquots--;
182
183 /*
184 * We move dquots to the freelist as soon as their reference count
185 * hits zero, so it really should be on the freelist here.
186 */
187 ASSERT(!list_empty(&dqp->q_lru));
188 list_lru_del(&qi->qi_lru, &dqp->q_lru);
189 XFS_STATS_DEC(xs_qm_dquot_unused);
190
191 xfs_qm_dqdestroy(dqp);
192 return 0;
193 }
194
195 /*
196 * Release the group or project dquot pointers the user dquots maybe carrying
197 * around as a hint, and proceed to purge the user dquot cache if requested.
198 */
199 STATIC int
200 xfs_qm_dqpurge_hints(
201 struct xfs_dquot *dqp,
202 void *data)
203 {
204 struct xfs_dquot *gdqp = NULL;
205 struct xfs_dquot *pdqp = NULL;
206 uint flags = *((uint *)data);
207
208 xfs_dqlock(dqp);
209 if (dqp->dq_flags & XFS_DQ_FREEING) {
210 xfs_dqunlock(dqp);
211 return EAGAIN;
212 }
213
214 /* If this quota has a hint attached, prepare for releasing it now */
215 gdqp = dqp->q_gdquot;
216 if (gdqp)
217 dqp->q_gdquot = NULL;
218
219 pdqp = dqp->q_pdquot;
220 if (pdqp)
221 dqp->q_pdquot = NULL;
222
223 xfs_dqunlock(dqp);
224
225 if (gdqp)
226 xfs_qm_dqrele(gdqp);
227 if (pdqp)
228 xfs_qm_dqrele(pdqp);
229
230 if (flags & XFS_QMOPT_UQUOTA)
231 return xfs_qm_dqpurge(dqp, NULL);
232
233 return 0;
234 }
235
236 /*
237 * Purge the dquot cache.
238 */
239 void
240 xfs_qm_dqpurge_all(
241 struct xfs_mount *mp,
242 uint flags)
243 {
244 /*
245 * We have to release group/project dquot hint(s) from the user dquot
246 * at first if they are there, otherwise we would run into an infinite
247 * loop while walking through radix tree to purge other type of dquots
248 * since their refcount is not zero if the user dquot refers to them
249 * as hint.
250 *
251 * Call the special xfs_qm_dqpurge_hints() will end up go through the
252 * general xfs_qm_dqpurge() against user dquot cache if requested.
253 */
254 xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_dqpurge_hints, &flags);
255
256 if (flags & XFS_QMOPT_GQUOTA)
257 xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_dqpurge, NULL);
258 if (flags & XFS_QMOPT_PQUOTA)
259 xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_dqpurge, NULL);
260 }
261
262 /*
263 * Just destroy the quotainfo structure.
264 */
265 void
266 xfs_qm_unmount(
267 struct xfs_mount *mp)
268 {
269 if (mp->m_quotainfo) {
270 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
271 xfs_qm_destroy_quotainfo(mp);
272 }
273 }
274
275
276 /*
277 * This is called from xfs_mountfs to start quotas and initialize all
278 * necessary data structures like quotainfo. This is also responsible for
279 * running a quotacheck as necessary. We are guaranteed that the superblock
280 * is consistently read in at this point.
281 *
282 * If we fail here, the mount will continue with quota turned off. We don't
283 * need to inidicate success or failure at all.
284 */
285 void
286 xfs_qm_mount_quotas(
287 xfs_mount_t *mp)
288 {
289 int error = 0;
290 uint sbf;
291
292 /*
293 * If quotas on realtime volumes is not supported, we disable
294 * quotas immediately.
295 */
296 if (mp->m_sb.sb_rextents) {
297 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
298 mp->m_qflags = 0;
299 goto write_changes;
300 }
301
302 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
303
304 /*
305 * Allocate the quotainfo structure inside the mount struct, and
306 * create quotainode(s), and change/rev superblock if necessary.
307 */
308 error = xfs_qm_init_quotainfo(mp);
309 if (error) {
310 /*
311 * We must turn off quotas.
312 */
313 ASSERT(mp->m_quotainfo == NULL);
314 mp->m_qflags = 0;
315 goto write_changes;
316 }
317 /*
318 * If any of the quotas are not consistent, do a quotacheck.
319 */
320 if (XFS_QM_NEED_QUOTACHECK(mp)) {
321 error = xfs_qm_quotacheck(mp);
322 if (error) {
323 /* Quotacheck failed and disabled quotas. */
324 return;
325 }
326 }
327 /*
328 * If one type of quotas is off, then it will lose its
329 * quotachecked status, since we won't be doing accounting for
330 * that type anymore.
331 */
332 if (!XFS_IS_UQUOTA_ON(mp))
333 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
334 if (!XFS_IS_GQUOTA_ON(mp))
335 mp->m_qflags &= ~XFS_GQUOTA_CHKD;
336 if (!XFS_IS_PQUOTA_ON(mp))
337 mp->m_qflags &= ~XFS_PQUOTA_CHKD;
338
339 write_changes:
340 /*
341 * We actually don't have to acquire the m_sb_lock at all.
342 * This can only be called from mount, and that's single threaded. XXX
343 */
344 spin_lock(&mp->m_sb_lock);
345 sbf = mp->m_sb.sb_qflags;
346 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
347 spin_unlock(&mp->m_sb_lock);
348
349 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
350 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
351 /*
352 * We could only have been turning quotas off.
353 * We aren't in very good shape actually because
354 * the incore structures are convinced that quotas are
355 * off, but the on disk superblock doesn't know that !
356 */
357 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
358 xfs_alert(mp, "%s: Superblock update failed!",
359 __func__);
360 }
361 }
362
363 if (error) {
364 xfs_warn(mp, "Failed to initialize disk quotas.");
365 return;
366 }
367 }
368
369 /*
370 * Called from the vfsops layer.
371 */
372 void
373 xfs_qm_unmount_quotas(
374 xfs_mount_t *mp)
375 {
376 /*
377 * Release the dquots that root inode, et al might be holding,
378 * before we flush quotas and blow away the quotainfo structure.
379 */
380 ASSERT(mp->m_rootip);
381 xfs_qm_dqdetach(mp->m_rootip);
382 if (mp->m_rbmip)
383 xfs_qm_dqdetach(mp->m_rbmip);
384 if (mp->m_rsumip)
385 xfs_qm_dqdetach(mp->m_rsumip);
386
387 /*
388 * Release the quota inodes.
389 */
390 if (mp->m_quotainfo) {
391 if (mp->m_quotainfo->qi_uquotaip) {
392 IRELE(mp->m_quotainfo->qi_uquotaip);
393 mp->m_quotainfo->qi_uquotaip = NULL;
394 }
395 if (mp->m_quotainfo->qi_gquotaip) {
396 IRELE(mp->m_quotainfo->qi_gquotaip);
397 mp->m_quotainfo->qi_gquotaip = NULL;
398 }
399 if (mp->m_quotainfo->qi_pquotaip) {
400 IRELE(mp->m_quotainfo->qi_pquotaip);
401 mp->m_quotainfo->qi_pquotaip = NULL;
402 }
403 }
404 }
405
406 STATIC int
407 xfs_qm_dqattach_one(
408 xfs_inode_t *ip,
409 xfs_dqid_t id,
410 uint type,
411 uint doalloc,
412 xfs_dquot_t *udqhint, /* hint */
413 xfs_dquot_t **IO_idqpp)
414 {
415 xfs_dquot_t *dqp;
416 int error;
417
418 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
419 error = 0;
420
421 /*
422 * See if we already have it in the inode itself. IO_idqpp is
423 * &i_udquot or &i_gdquot. This made the code look weird, but
424 * made the logic a lot simpler.
425 */
426 dqp = *IO_idqpp;
427 if (dqp) {
428 trace_xfs_dqattach_found(dqp);
429 return 0;
430 }
431
432 /*
433 * udqhint is the i_udquot field in inode, and is non-NULL only
434 * when the type arg is group/project. Its purpose is to save a
435 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
436 * the user dquot.
437 */
438 if (udqhint) {
439 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
440 xfs_dqlock(udqhint);
441
442 /*
443 * No need to take dqlock to look at the id.
444 *
445 * The ID can't change until it gets reclaimed, and it won't
446 * be reclaimed as long as we have a ref from inode and we
447 * hold the ilock.
448 */
449 if (type == XFS_DQ_GROUP)
450 dqp = udqhint->q_gdquot;
451 else
452 dqp = udqhint->q_pdquot;
453 if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
454 ASSERT(*IO_idqpp == NULL);
455
456 *IO_idqpp = xfs_qm_dqhold(dqp);
457 xfs_dqunlock(udqhint);
458 return 0;
459 }
460
461 /*
462 * We can't hold a dquot lock when we call the dqget code.
463 * We'll deadlock in no time, because of (not conforming to)
464 * lock ordering - the inodelock comes before any dquot lock,
465 * and we may drop and reacquire the ilock in xfs_qm_dqget().
466 */
467 xfs_dqunlock(udqhint);
468 }
469
470 /*
471 * Find the dquot from somewhere. This bumps the
472 * reference count of dquot and returns it locked.
473 * This can return ENOENT if dquot didn't exist on
474 * disk and we didn't ask it to allocate;
475 * ESRCH if quotas got turned off suddenly.
476 */
477 error = xfs_qm_dqget(ip->i_mount, ip, id, type,
478 doalloc | XFS_QMOPT_DOWARN, &dqp);
479 if (error)
480 return error;
481
482 trace_xfs_dqattach_get(dqp);
483
484 /*
485 * dqget may have dropped and re-acquired the ilock, but it guarantees
486 * that the dquot returned is the one that should go in the inode.
487 */
488 *IO_idqpp = dqp;
489 xfs_dqunlock(dqp);
490 return 0;
491 }
492
493
494 /*
495 * Given a udquot and group/project type, attach the group/project
496 * dquot pointer to the udquot as a hint for future lookups.
497 */
498 STATIC void
499 xfs_qm_dqattach_hint(
500 struct xfs_inode *ip,
501 int type)
502 {
503 struct xfs_dquot **dqhintp;
504 struct xfs_dquot *dqp;
505 struct xfs_dquot *udq = ip->i_udquot;
506
507 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
508
509 xfs_dqlock(udq);
510
511 if (type == XFS_DQ_GROUP) {
512 dqp = ip->i_gdquot;
513 dqhintp = &udq->q_gdquot;
514 } else {
515 dqp = ip->i_pdquot;
516 dqhintp = &udq->q_pdquot;
517 }
518
519 if (*dqhintp) {
520 struct xfs_dquot *tmp;
521
522 if (*dqhintp == dqp)
523 goto done;
524
525 tmp = *dqhintp;
526 *dqhintp = NULL;
527 xfs_qm_dqrele(tmp);
528 }
529
530 *dqhintp = xfs_qm_dqhold(dqp);
531 done:
532 xfs_dqunlock(udq);
533 }
534
535 static bool
536 xfs_qm_need_dqattach(
537 struct xfs_inode *ip)
538 {
539 struct xfs_mount *mp = ip->i_mount;
540
541 if (!XFS_IS_QUOTA_RUNNING(mp))
542 return false;
543 if (!XFS_IS_QUOTA_ON(mp))
544 return false;
545 if (!XFS_NOT_DQATTACHED(mp, ip))
546 return false;
547 if (xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
548 return false;
549 return true;
550 }
551
552 /*
553 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
554 * into account.
555 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
556 * Inode may get unlocked and relocked in here, and the caller must deal with
557 * the consequences.
558 */
559 int
560 xfs_qm_dqattach_locked(
561 xfs_inode_t *ip,
562 uint flags)
563 {
564 xfs_mount_t *mp = ip->i_mount;
565 uint nquotas = 0;
566 int error = 0;
567
568 if (!xfs_qm_need_dqattach(ip))
569 return 0;
570
571 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
572
573 if (XFS_IS_UQUOTA_ON(mp)) {
574 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
575 flags & XFS_QMOPT_DQALLOC,
576 NULL, &ip->i_udquot);
577 if (error)
578 goto done;
579 nquotas++;
580 }
581
582 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
583 if (XFS_IS_GQUOTA_ON(mp)) {
584 error = xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
585 flags & XFS_QMOPT_DQALLOC,
586 ip->i_udquot, &ip->i_gdquot);
587 /*
588 * Don't worry about the udquot that we may have
589 * attached above. It'll get detached, if not already.
590 */
591 if (error)
592 goto done;
593 nquotas++;
594 }
595
596 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
597 if (XFS_IS_PQUOTA_ON(mp)) {
598 error = xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
599 flags & XFS_QMOPT_DQALLOC,
600 ip->i_udquot, &ip->i_pdquot);
601 /*
602 * Don't worry about the udquot that we may have
603 * attached above. It'll get detached, if not already.
604 */
605 if (error)
606 goto done;
607 nquotas++;
608 }
609
610 /*
611 * Attach this group/project quota to the user quota as a hint.
612 * This WON'T, in general, result in a thrash.
613 */
614 if (nquotas > 1 && ip->i_udquot) {
615 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
616 ASSERT(ip->i_gdquot || !XFS_IS_GQUOTA_ON(mp));
617 ASSERT(ip->i_pdquot || !XFS_IS_PQUOTA_ON(mp));
618
619 /*
620 * We do not have i_udquot locked at this point, but this check
621 * is OK since we don't depend on the i_gdquot to be accurate
622 * 100% all the time. It is just a hint, and this will
623 * succeed in general.
624 */
625 if (ip->i_udquot->q_gdquot != ip->i_gdquot)
626 xfs_qm_dqattach_hint(ip, XFS_DQ_GROUP);
627
628 if (ip->i_udquot->q_pdquot != ip->i_pdquot)
629 xfs_qm_dqattach_hint(ip, XFS_DQ_PROJ);
630 }
631
632 done:
633 #ifdef DEBUG
634 if (!error) {
635 if (XFS_IS_UQUOTA_ON(mp))
636 ASSERT(ip->i_udquot);
637 if (XFS_IS_GQUOTA_ON(mp))
638 ASSERT(ip->i_gdquot);
639 if (XFS_IS_PQUOTA_ON(mp))
640 ASSERT(ip->i_pdquot);
641 }
642 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
643 #endif
644 return error;
645 }
646
647 int
648 xfs_qm_dqattach(
649 struct xfs_inode *ip,
650 uint flags)
651 {
652 int error;
653
654 if (!xfs_qm_need_dqattach(ip))
655 return 0;
656
657 xfs_ilock(ip, XFS_ILOCK_EXCL);
658 error = xfs_qm_dqattach_locked(ip, flags);
659 xfs_iunlock(ip, XFS_ILOCK_EXCL);
660
661 return error;
662 }
663
664 /*
665 * Release dquots (and their references) if any.
666 * The inode should be locked EXCL except when this's called by
667 * xfs_ireclaim.
668 */
669 void
670 xfs_qm_dqdetach(
671 xfs_inode_t *ip)
672 {
673 if (!(ip->i_udquot || ip->i_gdquot || ip->i_pdquot))
674 return;
675
676 trace_xfs_dquot_dqdetach(ip);
677
678 ASSERT(!xfs_is_quota_inode(&ip->i_mount->m_sb, ip->i_ino));
679 if (ip->i_udquot) {
680 xfs_qm_dqrele(ip->i_udquot);
681 ip->i_udquot = NULL;
682 }
683 if (ip->i_gdquot) {
684 xfs_qm_dqrele(ip->i_gdquot);
685 ip->i_gdquot = NULL;
686 }
687 if (ip->i_pdquot) {
688 xfs_qm_dqrele(ip->i_pdquot);
689 ip->i_pdquot = NULL;
690 }
691 }
692
693 struct xfs_qm_isolate {
694 struct list_head buffers;
695 struct list_head dispose;
696 };
697
698 static enum lru_status
699 xfs_qm_dquot_isolate(
700 struct list_head *item,
701 spinlock_t *lru_lock,
702 void *arg)
703 {
704 struct xfs_dquot *dqp = container_of(item,
705 struct xfs_dquot, q_lru);
706 struct xfs_qm_isolate *isol = arg;
707
708 if (!xfs_dqlock_nowait(dqp))
709 goto out_miss_busy;
710
711 /*
712 * This dquot has acquired a reference in the meantime remove it from
713 * the freelist and try again.
714 */
715 if (dqp->q_nrefs) {
716 xfs_dqunlock(dqp);
717 XFS_STATS_INC(xs_qm_dqwants);
718
719 trace_xfs_dqreclaim_want(dqp);
720 list_del_init(&dqp->q_lru);
721 XFS_STATS_DEC(xs_qm_dquot_unused);
722 return LRU_REMOVED;
723 }
724
725 /*
726 * If the dquot is dirty, flush it. If it's already being flushed, just
727 * skip it so there is time for the IO to complete before we try to
728 * reclaim it again on the next LRU pass.
729 */
730 if (!xfs_dqflock_nowait(dqp)) {
731 xfs_dqunlock(dqp);
732 goto out_miss_busy;
733 }
734
735 if (XFS_DQ_IS_DIRTY(dqp)) {
736 struct xfs_buf *bp = NULL;
737 int error;
738
739 trace_xfs_dqreclaim_dirty(dqp);
740
741 /* we have to drop the LRU lock to flush the dquot */
742 spin_unlock(lru_lock);
743
744 error = xfs_qm_dqflush(dqp, &bp);
745 if (error) {
746 xfs_warn(dqp->q_mount, "%s: dquot %p flush failed",
747 __func__, dqp);
748 goto out_unlock_dirty;
749 }
750
751 xfs_buf_delwri_queue(bp, &isol->buffers);
752 xfs_buf_relse(bp);
753 goto out_unlock_dirty;
754 }
755 xfs_dqfunlock(dqp);
756
757 /*
758 * Prevent lookups now that we are past the point of no return.
759 */
760 dqp->dq_flags |= XFS_DQ_FREEING;
761 xfs_dqunlock(dqp);
762
763 ASSERT(dqp->q_nrefs == 0);
764 list_move_tail(&dqp->q_lru, &isol->dispose);
765 XFS_STATS_DEC(xs_qm_dquot_unused);
766 trace_xfs_dqreclaim_done(dqp);
767 XFS_STATS_INC(xs_qm_dqreclaims);
768 return LRU_REMOVED;
769
770 out_miss_busy:
771 trace_xfs_dqreclaim_busy(dqp);
772 XFS_STATS_INC(xs_qm_dqreclaim_misses);
773 return LRU_SKIP;
774
775 out_unlock_dirty:
776 trace_xfs_dqreclaim_busy(dqp);
777 XFS_STATS_INC(xs_qm_dqreclaim_misses);
778 xfs_dqunlock(dqp);
779 spin_lock(lru_lock);
780 return LRU_RETRY;
781 }
782
783 static unsigned long
784 xfs_qm_shrink_scan(
785 struct shrinker *shrink,
786 struct shrink_control *sc)
787 {
788 struct xfs_quotainfo *qi = container_of(shrink,
789 struct xfs_quotainfo, qi_shrinker);
790 struct xfs_qm_isolate isol;
791 unsigned long freed;
792 int error;
793 unsigned long nr_to_scan = sc->nr_to_scan;
794
795 if ((sc->gfp_mask & (__GFP_FS|__GFP_WAIT)) != (__GFP_FS|__GFP_WAIT))
796 return 0;
797
798 INIT_LIST_HEAD(&isol.buffers);
799 INIT_LIST_HEAD(&isol.dispose);
800
801 freed = list_lru_walk_node(&qi->qi_lru, sc->nid, xfs_qm_dquot_isolate, &isol,
802 &nr_to_scan);
803
804 error = xfs_buf_delwri_submit(&isol.buffers);
805 if (error)
806 xfs_warn(NULL, "%s: dquot reclaim failed", __func__);
807
808 while (!list_empty(&isol.dispose)) {
809 struct xfs_dquot *dqp;
810
811 dqp = list_first_entry(&isol.dispose, struct xfs_dquot, q_lru);
812 list_del_init(&dqp->q_lru);
813 xfs_qm_dqfree_one(dqp);
814 }
815
816 return freed;
817 }
818
819 static unsigned long
820 xfs_qm_shrink_count(
821 struct shrinker *shrink,
822 struct shrink_control *sc)
823 {
824 struct xfs_quotainfo *qi = container_of(shrink,
825 struct xfs_quotainfo, qi_shrinker);
826
827 return list_lru_count_node(&qi->qi_lru, sc->nid);
828 }
829
830 /*
831 * This initializes all the quota information that's kept in the
832 * mount structure
833 */
834 STATIC int
835 xfs_qm_init_quotainfo(
836 xfs_mount_t *mp)
837 {
838 xfs_quotainfo_t *qinf;
839 int error;
840 xfs_dquot_t *dqp;
841
842 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
843
844 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
845
846 if ((error = list_lru_init(&qinf->qi_lru))) {
847 kmem_free(qinf);
848 mp->m_quotainfo = NULL;
849 return error;
850 }
851
852 /*
853 * See if quotainodes are setup, and if not, allocate them,
854 * and change the superblock accordingly.
855 */
856 if ((error = xfs_qm_init_quotainos(mp))) {
857 list_lru_destroy(&qinf->qi_lru);
858 kmem_free(qinf);
859 mp->m_quotainfo = NULL;
860 return error;
861 }
862
863 INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
864 INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
865 INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS);
866 mutex_init(&qinf->qi_tree_lock);
867
868 /* mutex used to serialize quotaoffs */
869 mutex_init(&qinf->qi_quotaofflock);
870
871 /* Precalc some constants */
872 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
873 qinf->qi_dqperchunk = xfs_calc_dquots_per_chunk(mp,
874 qinf->qi_dqchunklen);
875
876 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
877
878 /*
879 * We try to get the limits from the superuser's limits fields.
880 * This is quite hacky, but it is standard quota practice.
881 *
882 * We look at the USR dquot with id == 0 first, but if user quotas
883 * are not enabled we goto the GRP dquot with id == 0.
884 * We don't really care to keep separate default limits for user
885 * and group quotas, at least not at this point.
886 *
887 * Since we may not have done a quotacheck by this point, just read
888 * the dquot without attaching it to any hashtables or lists.
889 */
890 error = xfs_qm_dqread(mp, 0,
891 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
892 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
893 XFS_DQ_PROJ),
894 XFS_QMOPT_DOWARN, &dqp);
895 if (!error) {
896 xfs_disk_dquot_t *ddqp = &dqp->q_core;
897
898 /*
899 * The warnings and timers set the grace period given to
900 * a user or group before he or she can not perform any
901 * more writing. If it is zero, a default is used.
902 */
903 qinf->qi_btimelimit = ddqp->d_btimer ?
904 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
905 qinf->qi_itimelimit = ddqp->d_itimer ?
906 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
907 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
908 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
909 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
910 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
911 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
912 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
913 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
914 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
915 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
916 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
917 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
918 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
919 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
920 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
921
922 xfs_qm_dqdestroy(dqp);
923 } else {
924 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
925 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
926 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
927 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
928 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
929 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
930 }
931
932 qinf->qi_shrinker.count_objects = xfs_qm_shrink_count;
933 qinf->qi_shrinker.scan_objects = xfs_qm_shrink_scan;
934 qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
935 qinf->qi_shrinker.flags = SHRINKER_NUMA_AWARE;
936 register_shrinker(&qinf->qi_shrinker);
937 return 0;
938 }
939
940
941 /*
942 * Gets called when unmounting a filesystem or when all quotas get
943 * turned off.
944 * This purges the quota inodes, destroys locks and frees itself.
945 */
946 void
947 xfs_qm_destroy_quotainfo(
948 xfs_mount_t *mp)
949 {
950 xfs_quotainfo_t *qi;
951
952 qi = mp->m_quotainfo;
953 ASSERT(qi != NULL);
954
955 unregister_shrinker(&qi->qi_shrinker);
956 list_lru_destroy(&qi->qi_lru);
957
958 if (qi->qi_uquotaip) {
959 IRELE(qi->qi_uquotaip);
960 qi->qi_uquotaip = NULL; /* paranoia */
961 }
962 if (qi->qi_gquotaip) {
963 IRELE(qi->qi_gquotaip);
964 qi->qi_gquotaip = NULL;
965 }
966 if (qi->qi_pquotaip) {
967 IRELE(qi->qi_pquotaip);
968 qi->qi_pquotaip = NULL;
969 }
970 mutex_destroy(&qi->qi_quotaofflock);
971 kmem_free(qi);
972 mp->m_quotainfo = NULL;
973 }
974
975 /*
976 * Create an inode and return with a reference already taken, but unlocked
977 * This is how we create quota inodes
978 */
979 STATIC int
980 xfs_qm_qino_alloc(
981 xfs_mount_t *mp,
982 xfs_inode_t **ip,
983 __int64_t sbfields,
984 uint flags)
985 {
986 xfs_trans_t *tp;
987 int error;
988 int committed;
989
990 *ip = NULL;
991 /*
992 * With superblock that doesn't have separate pquotino, we
993 * share an inode between gquota and pquota. If the on-disk
994 * superblock has GQUOTA and the filesystem is now mounted
995 * with PQUOTA, just use sb_gquotino for sb_pquotino and
996 * vice-versa.
997 */
998 if (!xfs_sb_version_has_pquotino(&mp->m_sb) &&
999 (flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) {
1000 xfs_ino_t ino = NULLFSINO;
1001
1002 if ((flags & XFS_QMOPT_PQUOTA) &&
1003 (mp->m_sb.sb_gquotino != NULLFSINO)) {
1004 ino = mp->m_sb.sb_gquotino;
1005 ASSERT(mp->m_sb.sb_pquotino == NULLFSINO);
1006 } else if ((flags & XFS_QMOPT_GQUOTA) &&
1007 (mp->m_sb.sb_pquotino != NULLFSINO)) {
1008 ino = mp->m_sb.sb_pquotino;
1009 ASSERT(mp->m_sb.sb_gquotino == NULLFSINO);
1010 }
1011 if (ino != NULLFSINO) {
1012 error = xfs_iget(mp, NULL, ino, 0, 0, ip);
1013 if (error)
1014 return error;
1015 mp->m_sb.sb_gquotino = NULLFSINO;
1016 mp->m_sb.sb_pquotino = NULLFSINO;
1017 }
1018 }
1019
1020 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
1021 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_create,
1022 XFS_QM_QINOCREATE_SPACE_RES(mp), 0);
1023 if (error) {
1024 xfs_trans_cancel(tp, 0);
1025 return error;
1026 }
1027
1028 if (!*ip) {
1029 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip,
1030 &committed);
1031 if (error) {
1032 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1033 XFS_TRANS_ABORT);
1034 return error;
1035 }
1036 }
1037
1038 /*
1039 * Make the changes in the superblock, and log those too.
1040 * sbfields arg may contain fields other than *QUOTINO;
1041 * VERSIONNUM for example.
1042 */
1043 spin_lock(&mp->m_sb_lock);
1044 if (flags & XFS_QMOPT_SBVERSION) {
1045 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1046 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1047 XFS_SB_GQUOTINO | XFS_SB_PQUOTINO | XFS_SB_QFLAGS)) ==
1048 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1049 XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1050 XFS_SB_QFLAGS));
1051
1052 xfs_sb_version_addquota(&mp->m_sb);
1053 mp->m_sb.sb_uquotino = NULLFSINO;
1054 mp->m_sb.sb_gquotino = NULLFSINO;
1055 mp->m_sb.sb_pquotino = NULLFSINO;
1056
1057 /* qflags will get updated fully _after_ quotacheck */
1058 mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT;
1059 }
1060 if (flags & XFS_QMOPT_UQUOTA)
1061 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1062 else if (flags & XFS_QMOPT_GQUOTA)
1063 mp->m_sb.sb_gquotino = (*ip)->i_ino;
1064 else
1065 mp->m_sb.sb_pquotino = (*ip)->i_ino;
1066 spin_unlock(&mp->m_sb_lock);
1067 xfs_mod_sb(tp, sbfields);
1068
1069 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
1070 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
1071 return error;
1072 }
1073 return 0;
1074 }
1075
1076
1077 STATIC void
1078 xfs_qm_reset_dqcounts(
1079 xfs_mount_t *mp,
1080 xfs_buf_t *bp,
1081 xfs_dqid_t id,
1082 uint type)
1083 {
1084 struct xfs_dqblk *dqb;
1085 int j;
1086
1087 trace_xfs_reset_dqcounts(bp, _RET_IP_);
1088
1089 /*
1090 * Reset all counters and timers. They'll be
1091 * started afresh by xfs_qm_quotacheck.
1092 */
1093 #ifdef DEBUG
1094 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1095 do_div(j, sizeof(xfs_dqblk_t));
1096 ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
1097 #endif
1098 dqb = bp->b_addr;
1099 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
1100 struct xfs_disk_dquot *ddq;
1101
1102 ddq = (struct xfs_disk_dquot *)&dqb[j];
1103
1104 /*
1105 * Do a sanity check, and if needed, repair the dqblk. Don't
1106 * output any warnings because it's perfectly possible to
1107 * find uninitialised dquot blks. See comment in xfs_dqcheck.
1108 */
1109 xfs_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1110 "xfs_quotacheck");
1111 ddq->d_bcount = 0;
1112 ddq->d_icount = 0;
1113 ddq->d_rtbcount = 0;
1114 ddq->d_btimer = 0;
1115 ddq->d_itimer = 0;
1116 ddq->d_rtbtimer = 0;
1117 ddq->d_bwarns = 0;
1118 ddq->d_iwarns = 0;
1119 ddq->d_rtbwarns = 0;
1120
1121 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1122 xfs_update_cksum((char *)&dqb[j],
1123 sizeof(struct xfs_dqblk),
1124 XFS_DQUOT_CRC_OFF);
1125 }
1126 }
1127 }
1128
1129 STATIC int
1130 xfs_qm_dqiter_bufs(
1131 struct xfs_mount *mp,
1132 xfs_dqid_t firstid,
1133 xfs_fsblock_t bno,
1134 xfs_filblks_t blkcnt,
1135 uint flags,
1136 struct list_head *buffer_list)
1137 {
1138 struct xfs_buf *bp;
1139 int error;
1140 int type;
1141
1142 ASSERT(blkcnt > 0);
1143 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1144 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1145 error = 0;
1146
1147 /*
1148 * Blkcnt arg can be a very big number, and might even be
1149 * larger than the log itself. So, we have to break it up into
1150 * manageable-sized transactions.
1151 * Note that we don't start a permanent transaction here; we might
1152 * not be able to get a log reservation for the whole thing up front,
1153 * and we don't really care to either, because we just discard
1154 * everything if we were to crash in the middle of this loop.
1155 */
1156 while (blkcnt--) {
1157 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1158 XFS_FSB_TO_DADDR(mp, bno),
1159 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1160 &xfs_dquot_buf_ops);
1161
1162 /*
1163 * CRC and validation errors will return a EFSCORRUPTED here. If
1164 * this occurs, re-read without CRC validation so that we can
1165 * repair the damage via xfs_qm_reset_dqcounts(). This process
1166 * will leave a trace in the log indicating corruption has
1167 * been detected.
1168 */
1169 if (error == EFSCORRUPTED) {
1170 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1171 XFS_FSB_TO_DADDR(mp, bno),
1172 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1173 NULL);
1174 }
1175
1176 if (error)
1177 break;
1178
1179 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1180 xfs_buf_delwri_queue(bp, buffer_list);
1181 xfs_buf_relse(bp);
1182
1183 /* goto the next block. */
1184 bno++;
1185 firstid += mp->m_quotainfo->qi_dqperchunk;
1186 }
1187
1188 return error;
1189 }
1190
1191 /*
1192 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1193 * caller supplied function for every chunk of dquots that we find.
1194 */
1195 STATIC int
1196 xfs_qm_dqiterate(
1197 struct xfs_mount *mp,
1198 struct xfs_inode *qip,
1199 uint flags,
1200 struct list_head *buffer_list)
1201 {
1202 struct xfs_bmbt_irec *map;
1203 int i, nmaps; /* number of map entries */
1204 int error; /* return value */
1205 xfs_fileoff_t lblkno;
1206 xfs_filblks_t maxlblkcnt;
1207 xfs_dqid_t firstid;
1208 xfs_fsblock_t rablkno;
1209 xfs_filblks_t rablkcnt;
1210
1211 error = 0;
1212 /*
1213 * This looks racy, but we can't keep an inode lock across a
1214 * trans_reserve. But, this gets called during quotacheck, and that
1215 * happens only at mount time which is single threaded.
1216 */
1217 if (qip->i_d.di_nblocks == 0)
1218 return 0;
1219
1220 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1221
1222 lblkno = 0;
1223 maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1224 do {
1225 uint lock_mode;
1226
1227 nmaps = XFS_DQITER_MAP_SIZE;
1228 /*
1229 * We aren't changing the inode itself. Just changing
1230 * some of its data. No new blocks are added here, and
1231 * the inode is never added to the transaction.
1232 */
1233 lock_mode = xfs_ilock_data_map_shared(qip);
1234 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
1235 map, &nmaps, 0);
1236 xfs_iunlock(qip, lock_mode);
1237 if (error)
1238 break;
1239
1240 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1241 for (i = 0; i < nmaps; i++) {
1242 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1243 ASSERT(map[i].br_blockcount);
1244
1245
1246 lblkno += map[i].br_blockcount;
1247
1248 if (map[i].br_startblock == HOLESTARTBLOCK)
1249 continue;
1250
1251 firstid = (xfs_dqid_t) map[i].br_startoff *
1252 mp->m_quotainfo->qi_dqperchunk;
1253 /*
1254 * Do a read-ahead on the next extent.
1255 */
1256 if ((i+1 < nmaps) &&
1257 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1258 rablkcnt = map[i+1].br_blockcount;
1259 rablkno = map[i+1].br_startblock;
1260 while (rablkcnt--) {
1261 xfs_buf_readahead(mp->m_ddev_targp,
1262 XFS_FSB_TO_DADDR(mp, rablkno),
1263 mp->m_quotainfo->qi_dqchunklen,
1264 NULL);
1265 rablkno++;
1266 }
1267 }
1268 /*
1269 * Iterate thru all the blks in the extent and
1270 * reset the counters of all the dquots inside them.
1271 */
1272 error = xfs_qm_dqiter_bufs(mp, firstid,
1273 map[i].br_startblock,
1274 map[i].br_blockcount,
1275 flags, buffer_list);
1276 if (error)
1277 goto out;
1278 }
1279 } while (nmaps > 0);
1280
1281 out:
1282 kmem_free(map);
1283 return error;
1284 }
1285
1286 /*
1287 * Called by dqusage_adjust in doing a quotacheck.
1288 *
1289 * Given the inode, and a dquot id this updates both the incore dqout as well
1290 * as the buffer copy. This is so that once the quotacheck is done, we can
1291 * just log all the buffers, as opposed to logging numerous updates to
1292 * individual dquots.
1293 */
1294 STATIC int
1295 xfs_qm_quotacheck_dqadjust(
1296 struct xfs_inode *ip,
1297 xfs_dqid_t id,
1298 uint type,
1299 xfs_qcnt_t nblks,
1300 xfs_qcnt_t rtblks)
1301 {
1302 struct xfs_mount *mp = ip->i_mount;
1303 struct xfs_dquot *dqp;
1304 int error;
1305
1306 error = xfs_qm_dqget(mp, ip, id, type,
1307 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1308 if (error) {
1309 /*
1310 * Shouldn't be able to turn off quotas here.
1311 */
1312 ASSERT(error != ESRCH);
1313 ASSERT(error != ENOENT);
1314 return error;
1315 }
1316
1317 trace_xfs_dqadjust(dqp);
1318
1319 /*
1320 * Adjust the inode count and the block count to reflect this inode's
1321 * resource usage.
1322 */
1323 be64_add_cpu(&dqp->q_core.d_icount, 1);
1324 dqp->q_res_icount++;
1325 if (nblks) {
1326 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1327 dqp->q_res_bcount += nblks;
1328 }
1329 if (rtblks) {
1330 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1331 dqp->q_res_rtbcount += rtblks;
1332 }
1333
1334 /*
1335 * Set default limits, adjust timers (since we changed usages)
1336 *
1337 * There are no timers for the default values set in the root dquot.
1338 */
1339 if (dqp->q_core.d_id) {
1340 xfs_qm_adjust_dqlimits(mp, dqp);
1341 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1342 }
1343
1344 dqp->dq_flags |= XFS_DQ_DIRTY;
1345 xfs_qm_dqput(dqp);
1346 return 0;
1347 }
1348
1349 STATIC int
1350 xfs_qm_get_rtblks(
1351 xfs_inode_t *ip,
1352 xfs_qcnt_t *O_rtblks)
1353 {
1354 xfs_filblks_t rtblks; /* total rt blks */
1355 xfs_extnum_t idx; /* extent record index */
1356 xfs_ifork_t *ifp; /* inode fork pointer */
1357 xfs_extnum_t nextents; /* number of extent entries */
1358 int error;
1359
1360 ASSERT(XFS_IS_REALTIME_INODE(ip));
1361 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1362 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1363 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1364 return error;
1365 }
1366 rtblks = 0;
1367 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1368 for (idx = 0; idx < nextents; idx++)
1369 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1370 *O_rtblks = (xfs_qcnt_t)rtblks;
1371 return 0;
1372 }
1373
1374 /*
1375 * callback routine supplied to bulkstat(). Given an inumber, find its
1376 * dquots and update them to account for resources taken by that inode.
1377 */
1378 /* ARGSUSED */
1379 STATIC int
1380 xfs_qm_dqusage_adjust(
1381 xfs_mount_t *mp, /* mount point for filesystem */
1382 xfs_ino_t ino, /* inode number to get data for */
1383 void __user *buffer, /* not used */
1384 int ubsize, /* not used */
1385 int *ubused, /* not used */
1386 int *res) /* result code value */
1387 {
1388 xfs_inode_t *ip;
1389 xfs_qcnt_t nblks, rtblks = 0;
1390 int error;
1391
1392 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1393
1394 /*
1395 * rootino must have its resources accounted for, not so with the quota
1396 * inodes.
1397 */
1398 if (xfs_is_quota_inode(&mp->m_sb, ino)) {
1399 *res = BULKSTAT_RV_NOTHING;
1400 return XFS_ERROR(EINVAL);
1401 }
1402
1403 /*
1404 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1405 * interface expects the inode to be exclusively locked because that's
1406 * the case in all other instances. It's OK that we do this because
1407 * quotacheck is done only at mount time.
1408 */
1409 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1410 if (error) {
1411 *res = BULKSTAT_RV_NOTHING;
1412 return error;
1413 }
1414
1415 ASSERT(ip->i_delayed_blks == 0);
1416
1417 if (XFS_IS_REALTIME_INODE(ip)) {
1418 /*
1419 * Walk thru the extent list and count the realtime blocks.
1420 */
1421 error = xfs_qm_get_rtblks(ip, &rtblks);
1422 if (error)
1423 goto error0;
1424 }
1425
1426 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1427
1428 /*
1429 * Add the (disk blocks and inode) resources occupied by this
1430 * inode to its dquots. We do this adjustment in the incore dquot,
1431 * and also copy the changes to its buffer.
1432 * We don't care about putting these changes in a transaction
1433 * envelope because if we crash in the middle of a 'quotacheck'
1434 * we have to start from the beginning anyway.
1435 * Once we're done, we'll log all the dquot bufs.
1436 *
1437 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1438 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1439 */
1440 if (XFS_IS_UQUOTA_ON(mp)) {
1441 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1442 XFS_DQ_USER, nblks, rtblks);
1443 if (error)
1444 goto error0;
1445 }
1446
1447 if (XFS_IS_GQUOTA_ON(mp)) {
1448 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1449 XFS_DQ_GROUP, nblks, rtblks);
1450 if (error)
1451 goto error0;
1452 }
1453
1454 if (XFS_IS_PQUOTA_ON(mp)) {
1455 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1456 XFS_DQ_PROJ, nblks, rtblks);
1457 if (error)
1458 goto error0;
1459 }
1460
1461 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1462 IRELE(ip);
1463 *res = BULKSTAT_RV_DIDONE;
1464 return 0;
1465
1466 error0:
1467 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1468 IRELE(ip);
1469 *res = BULKSTAT_RV_GIVEUP;
1470 return error;
1471 }
1472
1473 STATIC int
1474 xfs_qm_flush_one(
1475 struct xfs_dquot *dqp,
1476 void *data)
1477 {
1478 struct list_head *buffer_list = data;
1479 struct xfs_buf *bp = NULL;
1480 int error = 0;
1481
1482 xfs_dqlock(dqp);
1483 if (dqp->dq_flags & XFS_DQ_FREEING)
1484 goto out_unlock;
1485 if (!XFS_DQ_IS_DIRTY(dqp))
1486 goto out_unlock;
1487
1488 xfs_dqflock(dqp);
1489 error = xfs_qm_dqflush(dqp, &bp);
1490 if (error)
1491 goto out_unlock;
1492
1493 xfs_buf_delwri_queue(bp, buffer_list);
1494 xfs_buf_relse(bp);
1495 out_unlock:
1496 xfs_dqunlock(dqp);
1497 return error;
1498 }
1499
1500 /*
1501 * Walk thru all the filesystem inodes and construct a consistent view
1502 * of the disk quota world. If the quotacheck fails, disable quotas.
1503 */
1504 int
1505 xfs_qm_quotacheck(
1506 xfs_mount_t *mp)
1507 {
1508 int done, count, error, error2;
1509 xfs_ino_t lastino;
1510 size_t structsz;
1511 uint flags;
1512 LIST_HEAD (buffer_list);
1513 struct xfs_inode *uip = mp->m_quotainfo->qi_uquotaip;
1514 struct xfs_inode *gip = mp->m_quotainfo->qi_gquotaip;
1515 struct xfs_inode *pip = mp->m_quotainfo->qi_pquotaip;
1516
1517 count = INT_MAX;
1518 structsz = 1;
1519 lastino = 0;
1520 flags = 0;
1521
1522 ASSERT(uip || gip || pip);
1523 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1524
1525 xfs_notice(mp, "Quotacheck needed: Please wait.");
1526
1527 /*
1528 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1529 * their counters to zero. We need a clean slate.
1530 * We don't log our changes till later.
1531 */
1532 if (uip) {
1533 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA,
1534 &buffer_list);
1535 if (error)
1536 goto error_return;
1537 flags |= XFS_UQUOTA_CHKD;
1538 }
1539
1540 if (gip) {
1541 error = xfs_qm_dqiterate(mp, gip, XFS_QMOPT_GQUOTA,
1542 &buffer_list);
1543 if (error)
1544 goto error_return;
1545 flags |= XFS_GQUOTA_CHKD;
1546 }
1547
1548 if (pip) {
1549 error = xfs_qm_dqiterate(mp, pip, XFS_QMOPT_PQUOTA,
1550 &buffer_list);
1551 if (error)
1552 goto error_return;
1553 flags |= XFS_PQUOTA_CHKD;
1554 }
1555
1556 do {
1557 /*
1558 * Iterate thru all the inodes in the file system,
1559 * adjusting the corresponding dquot counters in core.
1560 */
1561 error = xfs_bulkstat(mp, &lastino, &count,
1562 xfs_qm_dqusage_adjust,
1563 structsz, NULL, &done);
1564 if (error)
1565 break;
1566
1567 } while (!done);
1568
1569 /*
1570 * We've made all the changes that we need to make incore. Flush them
1571 * down to disk buffers if everything was updated successfully.
1572 */
1573 if (XFS_IS_UQUOTA_ON(mp)) {
1574 error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one,
1575 &buffer_list);
1576 }
1577 if (XFS_IS_GQUOTA_ON(mp)) {
1578 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one,
1579 &buffer_list);
1580 if (!error)
1581 error = error2;
1582 }
1583 if (XFS_IS_PQUOTA_ON(mp)) {
1584 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one,
1585 &buffer_list);
1586 if (!error)
1587 error = error2;
1588 }
1589
1590 error2 = xfs_buf_delwri_submit(&buffer_list);
1591 if (!error)
1592 error = error2;
1593
1594 /*
1595 * We can get this error if we couldn't do a dquot allocation inside
1596 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1597 * dirty dquots that might be cached, we just want to get rid of them
1598 * and turn quotaoff. The dquots won't be attached to any of the inodes
1599 * at this point (because we intentionally didn't in dqget_noattach).
1600 */
1601 if (error) {
1602 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1603 goto error_return;
1604 }
1605
1606 /*
1607 * If one type of quotas is off, then it will lose its
1608 * quotachecked status, since we won't be doing accounting for
1609 * that type anymore.
1610 */
1611 mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1612 mp->m_qflags |= flags;
1613
1614 error_return:
1615 while (!list_empty(&buffer_list)) {
1616 struct xfs_buf *bp =
1617 list_first_entry(&buffer_list, struct xfs_buf, b_list);
1618 list_del_init(&bp->b_list);
1619 xfs_buf_relse(bp);
1620 }
1621
1622 if (error) {
1623 xfs_warn(mp,
1624 "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1625 error);
1626 /*
1627 * We must turn off quotas.
1628 */
1629 ASSERT(mp->m_quotainfo != NULL);
1630 xfs_qm_destroy_quotainfo(mp);
1631 if (xfs_mount_reset_sbqflags(mp)) {
1632 xfs_warn(mp,
1633 "Quotacheck: Failed to reset quota flags.");
1634 }
1635 } else
1636 xfs_notice(mp, "Quotacheck: Done.");
1637 return (error);
1638 }
1639
1640 /*
1641 * This is called after the superblock has been read in and we're ready to
1642 * iget the quota inodes.
1643 */
1644 STATIC int
1645 xfs_qm_init_quotainos(
1646 xfs_mount_t *mp)
1647 {
1648 struct xfs_inode *uip = NULL;
1649 struct xfs_inode *gip = NULL;
1650 struct xfs_inode *pip = NULL;
1651 int error;
1652 __int64_t sbflags = 0;
1653 uint flags = 0;
1654
1655 ASSERT(mp->m_quotainfo);
1656
1657 /*
1658 * Get the uquota and gquota inodes
1659 */
1660 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1661 if (XFS_IS_UQUOTA_ON(mp) &&
1662 mp->m_sb.sb_uquotino != NULLFSINO) {
1663 ASSERT(mp->m_sb.sb_uquotino > 0);
1664 error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1665 0, 0, &uip);
1666 if (error)
1667 return XFS_ERROR(error);
1668 }
1669 if (XFS_IS_GQUOTA_ON(mp) &&
1670 mp->m_sb.sb_gquotino != NULLFSINO) {
1671 ASSERT(mp->m_sb.sb_gquotino > 0);
1672 error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1673 0, 0, &gip);
1674 if (error)
1675 goto error_rele;
1676 }
1677 if (XFS_IS_PQUOTA_ON(mp) &&
1678 mp->m_sb.sb_pquotino != NULLFSINO) {
1679 ASSERT(mp->m_sb.sb_pquotino > 0);
1680 error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino,
1681 0, 0, &pip);
1682 if (error)
1683 goto error_rele;
1684 }
1685 } else {
1686 flags |= XFS_QMOPT_SBVERSION;
1687 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1688 XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1689 XFS_SB_QFLAGS);
1690 }
1691
1692 /*
1693 * Create the three inodes, if they don't exist already. The changes
1694 * made above will get added to a transaction and logged in one of
1695 * the qino_alloc calls below. If the device is readonly,
1696 * temporarily switch to read-write to do this.
1697 */
1698 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1699 error = xfs_qm_qino_alloc(mp, &uip,
1700 sbflags | XFS_SB_UQUOTINO,
1701 flags | XFS_QMOPT_UQUOTA);
1702 if (error)
1703 goto error_rele;
1704
1705 flags &= ~XFS_QMOPT_SBVERSION;
1706 }
1707 if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
1708 error = xfs_qm_qino_alloc(mp, &gip,
1709 sbflags | XFS_SB_GQUOTINO,
1710 flags | XFS_QMOPT_GQUOTA);
1711 if (error)
1712 goto error_rele;
1713
1714 flags &= ~XFS_QMOPT_SBVERSION;
1715 }
1716 if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) {
1717 error = xfs_qm_qino_alloc(mp, &pip,
1718 sbflags | XFS_SB_PQUOTINO,
1719 flags | XFS_QMOPT_PQUOTA);
1720 if (error)
1721 goto error_rele;
1722 }
1723
1724 mp->m_quotainfo->qi_uquotaip = uip;
1725 mp->m_quotainfo->qi_gquotaip = gip;
1726 mp->m_quotainfo->qi_pquotaip = pip;
1727
1728 return 0;
1729
1730 error_rele:
1731 if (uip)
1732 IRELE(uip);
1733 if (gip)
1734 IRELE(gip);
1735 if (pip)
1736 IRELE(pip);
1737 return XFS_ERROR(error);
1738 }
1739
1740 STATIC void
1741 xfs_qm_dqfree_one(
1742 struct xfs_dquot *dqp)
1743 {
1744 struct xfs_mount *mp = dqp->q_mount;
1745 struct xfs_quotainfo *qi = mp->m_quotainfo;
1746
1747 mutex_lock(&qi->qi_tree_lock);
1748 radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
1749 be32_to_cpu(dqp->q_core.d_id));
1750
1751 qi->qi_dquots--;
1752 mutex_unlock(&qi->qi_tree_lock);
1753
1754 xfs_qm_dqdestroy(dqp);
1755 }
1756
1757 /*
1758 * Start a transaction and write the incore superblock changes to
1759 * disk. flags parameter indicates which fields have changed.
1760 */
1761 int
1762 xfs_qm_write_sb_changes(
1763 xfs_mount_t *mp,
1764 __int64_t flags)
1765 {
1766 xfs_trans_t *tp;
1767 int error;
1768
1769 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1770 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
1771 if (error) {
1772 xfs_trans_cancel(tp, 0);
1773 return error;
1774 }
1775
1776 xfs_mod_sb(tp, flags);
1777 error = xfs_trans_commit(tp, 0);
1778
1779 return error;
1780 }
1781
1782
1783 /* --------------- utility functions for vnodeops ---------------- */
1784
1785
1786 /*
1787 * Given an inode, a uid, gid and prid make sure that we have
1788 * allocated relevant dquot(s) on disk, and that we won't exceed inode
1789 * quotas by creating this file.
1790 * This also attaches dquot(s) to the given inode after locking it,
1791 * and returns the dquots corresponding to the uid and/or gid.
1792 *
1793 * in : inode (unlocked)
1794 * out : udquot, gdquot with references taken and unlocked
1795 */
1796 int
1797 xfs_qm_vop_dqalloc(
1798 struct xfs_inode *ip,
1799 xfs_dqid_t uid,
1800 xfs_dqid_t gid,
1801 prid_t prid,
1802 uint flags,
1803 struct xfs_dquot **O_udqpp,
1804 struct xfs_dquot **O_gdqpp,
1805 struct xfs_dquot **O_pdqpp)
1806 {
1807 struct xfs_mount *mp = ip->i_mount;
1808 struct xfs_dquot *uq = NULL;
1809 struct xfs_dquot *gq = NULL;
1810 struct xfs_dquot *pq = NULL;
1811 int error;
1812 uint lockflags;
1813
1814 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1815 return 0;
1816
1817 lockflags = XFS_ILOCK_EXCL;
1818 xfs_ilock(ip, lockflags);
1819
1820 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1821 gid = ip->i_d.di_gid;
1822
1823 /*
1824 * Attach the dquot(s) to this inode, doing a dquot allocation
1825 * if necessary. The dquot(s) will not be locked.
1826 */
1827 if (XFS_NOT_DQATTACHED(mp, ip)) {
1828 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
1829 if (error) {
1830 xfs_iunlock(ip, lockflags);
1831 return error;
1832 }
1833 }
1834
1835 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1836 if (ip->i_d.di_uid != uid) {
1837 /*
1838 * What we need is the dquot that has this uid, and
1839 * if we send the inode to dqget, the uid of the inode
1840 * takes priority over what's sent in the uid argument.
1841 * We must unlock inode here before calling dqget if
1842 * we're not sending the inode, because otherwise
1843 * we'll deadlock by doing trans_reserve while
1844 * holding ilock.
1845 */
1846 xfs_iunlock(ip, lockflags);
1847 error = xfs_qm_dqget(mp, NULL, uid,
1848 XFS_DQ_USER,
1849 XFS_QMOPT_DQALLOC |
1850 XFS_QMOPT_DOWARN,
1851 &uq);
1852 if (error) {
1853 ASSERT(error != ENOENT);
1854 return error;
1855 }
1856 /*
1857 * Get the ilock in the right order.
1858 */
1859 xfs_dqunlock(uq);
1860 lockflags = XFS_ILOCK_SHARED;
1861 xfs_ilock(ip, lockflags);
1862 } else {
1863 /*
1864 * Take an extra reference, because we'll return
1865 * this to caller
1866 */
1867 ASSERT(ip->i_udquot);
1868 uq = xfs_qm_dqhold(ip->i_udquot);
1869 }
1870 }
1871 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1872 if (ip->i_d.di_gid != gid) {
1873 xfs_iunlock(ip, lockflags);
1874 error = xfs_qm_dqget(mp, NULL, gid,
1875 XFS_DQ_GROUP,
1876 XFS_QMOPT_DQALLOC |
1877 XFS_QMOPT_DOWARN,
1878 &gq);
1879 if (error) {
1880 ASSERT(error != ENOENT);
1881 goto error_rele;
1882 }
1883 xfs_dqunlock(gq);
1884 lockflags = XFS_ILOCK_SHARED;
1885 xfs_ilock(ip, lockflags);
1886 } else {
1887 ASSERT(ip->i_gdquot);
1888 gq = xfs_qm_dqhold(ip->i_gdquot);
1889 }
1890 }
1891 if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1892 if (xfs_get_projid(ip) != prid) {
1893 xfs_iunlock(ip, lockflags);
1894 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
1895 XFS_DQ_PROJ,
1896 XFS_QMOPT_DQALLOC |
1897 XFS_QMOPT_DOWARN,
1898 &pq);
1899 if (error) {
1900 ASSERT(error != ENOENT);
1901 goto error_rele;
1902 }
1903 xfs_dqunlock(pq);
1904 lockflags = XFS_ILOCK_SHARED;
1905 xfs_ilock(ip, lockflags);
1906 } else {
1907 ASSERT(ip->i_pdquot);
1908 pq = xfs_qm_dqhold(ip->i_pdquot);
1909 }
1910 }
1911 if (uq)
1912 trace_xfs_dquot_dqalloc(ip);
1913
1914 xfs_iunlock(ip, lockflags);
1915 if (O_udqpp)
1916 *O_udqpp = uq;
1917 else if (uq)
1918 xfs_qm_dqrele(uq);
1919 if (O_gdqpp)
1920 *O_gdqpp = gq;
1921 else if (gq)
1922 xfs_qm_dqrele(gq);
1923 if (O_pdqpp)
1924 *O_pdqpp = pq;
1925 else if (pq)
1926 xfs_qm_dqrele(pq);
1927 return 0;
1928
1929 error_rele:
1930 if (gq)
1931 xfs_qm_dqrele(gq);
1932 if (uq)
1933 xfs_qm_dqrele(uq);
1934 return error;
1935 }
1936
1937 /*
1938 * Actually transfer ownership, and do dquot modifications.
1939 * These were already reserved.
1940 */
1941 xfs_dquot_t *
1942 xfs_qm_vop_chown(
1943 xfs_trans_t *tp,
1944 xfs_inode_t *ip,
1945 xfs_dquot_t **IO_olddq,
1946 xfs_dquot_t *newdq)
1947 {
1948 xfs_dquot_t *prevdq;
1949 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
1950 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1951
1952
1953 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1954 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1955
1956 /* old dquot */
1957 prevdq = *IO_olddq;
1958 ASSERT(prevdq);
1959 ASSERT(prevdq != newdq);
1960
1961 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
1962 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1963
1964 /* the sparkling new dquot */
1965 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
1966 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1967
1968 /*
1969 * Take an extra reference, because the inode is going to keep
1970 * this dquot pointer even after the trans_commit.
1971 */
1972 *IO_olddq = xfs_qm_dqhold(newdq);
1973
1974 return prevdq;
1975 }
1976
1977 /*
1978 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1979 */
1980 int
1981 xfs_qm_vop_chown_reserve(
1982 struct xfs_trans *tp,
1983 struct xfs_inode *ip,
1984 struct xfs_dquot *udqp,
1985 struct xfs_dquot *gdqp,
1986 struct xfs_dquot *pdqp,
1987 uint flags)
1988 {
1989 struct xfs_mount *mp = ip->i_mount;
1990 uint delblks, blkflags, prjflags = 0;
1991 struct xfs_dquot *udq_unres = NULL;
1992 struct xfs_dquot *gdq_unres = NULL;
1993 struct xfs_dquot *pdq_unres = NULL;
1994 struct xfs_dquot *udq_delblks = NULL;
1995 struct xfs_dquot *gdq_delblks = NULL;
1996 struct xfs_dquot *pdq_delblks = NULL;
1997 int error;
1998
1999
2000 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2001 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2002
2003 delblks = ip->i_delayed_blks;
2004 blkflags = XFS_IS_REALTIME_INODE(ip) ?
2005 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
2006
2007 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2008 ip->i_d.di_uid != be32_to_cpu(udqp->q_core.d_id)) {
2009 udq_delblks = udqp;
2010 /*
2011 * If there are delayed allocation blocks, then we have to
2012 * unreserve those from the old dquot, and add them to the
2013 * new dquot.
2014 */
2015 if (delblks) {
2016 ASSERT(ip->i_udquot);
2017 udq_unres = ip->i_udquot;
2018 }
2019 }
2020 if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp &&
2021 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id)) {
2022 gdq_delblks = gdqp;
2023 if (delblks) {
2024 ASSERT(ip->i_gdquot);
2025 gdq_unres = ip->i_gdquot;
2026 }
2027 }
2028
2029 if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
2030 xfs_get_projid(ip) != be32_to_cpu(pdqp->q_core.d_id)) {
2031 prjflags = XFS_QMOPT_ENOSPC;
2032 pdq_delblks = pdqp;
2033 if (delblks) {
2034 ASSERT(ip->i_pdquot);
2035 pdq_unres = ip->i_pdquot;
2036 }
2037 }
2038
2039 error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2040 udq_delblks, gdq_delblks, pdq_delblks,
2041 ip->i_d.di_nblocks, 1,
2042 flags | blkflags | prjflags);
2043 if (error)
2044 return error;
2045
2046 /*
2047 * Do the delayed blks reservations/unreservations now. Since, these
2048 * are done without the help of a transaction, if a reservation fails
2049 * its previous reservations won't be automatically undone by trans
2050 * code. So, we have to do it manually here.
2051 */
2052 if (delblks) {
2053 /*
2054 * Do the reservations first. Unreservation can't fail.
2055 */
2056 ASSERT(udq_delblks || gdq_delblks || pdq_delblks);
2057 ASSERT(udq_unres || gdq_unres || pdq_unres);
2058 error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2059 udq_delblks, gdq_delblks, pdq_delblks,
2060 (xfs_qcnt_t)delblks, 0,
2061 flags | blkflags | prjflags);
2062 if (error)
2063 return error;
2064 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2065 udq_unres, gdq_unres, pdq_unres,
2066 -((xfs_qcnt_t)delblks), 0, blkflags);
2067 }
2068
2069 return (0);
2070 }
2071
2072 int
2073 xfs_qm_vop_rename_dqattach(
2074 struct xfs_inode **i_tab)
2075 {
2076 struct xfs_mount *mp = i_tab[0]->i_mount;
2077 int i;
2078
2079 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2080 return 0;
2081
2082 for (i = 0; (i < 4 && i_tab[i]); i++) {
2083 struct xfs_inode *ip = i_tab[i];
2084 int error;
2085
2086 /*
2087 * Watch out for duplicate entries in the table.
2088 */
2089 if (i == 0 || ip != i_tab[i-1]) {
2090 if (XFS_NOT_DQATTACHED(mp, ip)) {
2091 error = xfs_qm_dqattach(ip, 0);
2092 if (error)
2093 return error;
2094 }
2095 }
2096 }
2097 return 0;
2098 }
2099
2100 void
2101 xfs_qm_vop_create_dqattach(
2102 struct xfs_trans *tp,
2103 struct xfs_inode *ip,
2104 struct xfs_dquot *udqp,
2105 struct xfs_dquot *gdqp,
2106 struct xfs_dquot *pdqp)
2107 {
2108 struct xfs_mount *mp = tp->t_mountp;
2109
2110 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2111 return;
2112
2113 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2114 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2115
2116 if (udqp && XFS_IS_UQUOTA_ON(mp)) {
2117 ASSERT(ip->i_udquot == NULL);
2118 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
2119
2120 ip->i_udquot = xfs_qm_dqhold(udqp);
2121 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2122 }
2123 if (gdqp && XFS_IS_GQUOTA_ON(mp)) {
2124 ASSERT(ip->i_gdquot == NULL);
2125 ASSERT(ip->i_d.di_gid == be32_to_cpu(gdqp->q_core.d_id));
2126 ip->i_gdquot = xfs_qm_dqhold(gdqp);
2127 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
2128 }
2129 if (pdqp && XFS_IS_PQUOTA_ON(mp)) {
2130 ASSERT(ip->i_pdquot == NULL);
2131 ASSERT(xfs_get_projid(ip) == be32_to_cpu(pdqp->q_core.d_id));
2132
2133 ip->i_pdquot = xfs_qm_dqhold(pdqp);
2134 xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1);
2135 }
2136 }
2137