]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/xfs/xfs_qm.c
xfs: simplify xfs_qm_dqattach_grouphint
[mirror_ubuntu-zesty-kernel.git] / fs / xfs / xfs_qm.c
CommitLineData
1da177e4 1/*
4ce3121f
NS
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
1da177e4 4 *
4ce3121f
NS
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
1da177e4
LT
7 * published by the Free Software Foundation.
8 *
4ce3121f
NS
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.
1da177e4 13 *
4ce3121f
NS
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
1da177e4 17 */
1da177e4
LT
18#include "xfs.h"
19#include "xfs_fs.h"
a844f451 20#include "xfs_bit.h"
1da177e4 21#include "xfs_log.h"
a844f451 22#include "xfs_inum.h"
1da177e4
LT
23#include "xfs_trans.h"
24#include "xfs_sb.h"
25#include "xfs_ag.h"
1da177e4 26#include "xfs_alloc.h"
1da177e4
LT
27#include "xfs_quota.h"
28#include "xfs_mount.h"
1da177e4
LT
29#include "xfs_bmap_btree.h"
30#include "xfs_ialloc_btree.h"
1da177e4
LT
31#include "xfs_dinode.h"
32#include "xfs_inode.h"
a844f451
NS
33#include "xfs_ialloc.h"
34#include "xfs_itable.h"
1da177e4
LT
35#include "xfs_rtalloc.h"
36#include "xfs_error.h"
a844f451 37#include "xfs_bmap.h"
1da177e4
LT
38#include "xfs_attr.h"
39#include "xfs_buf_item.h"
40#include "xfs_trans_space.h"
41#include "xfs_utils.h"
1da177e4 42#include "xfs_qm.h"
0b1b213f 43#include "xfs_trace.h"
1da177e4
LT
44
45/*
46 * The global quota manager. There is only one of these for the entire
47 * system, _not_ one per file system. XQM keeps track of the overall
48 * quota functionality, including maintaining the freelist and hash
49 * tables of dquots.
50 */
a0b0b8a5 51struct mutex xfs_Gqm_lock;
1da177e4 52struct xfs_qm *xfs_Gqm;
6b3f6b5b 53uint ndquot;
1da177e4
LT
54
55kmem_zone_t *qm_dqzone;
56kmem_zone_t *qm_dqtrxzone;
1da177e4
LT
57
58STATIC void xfs_qm_list_init(xfs_dqlist_t *, char *, int);
59STATIC void xfs_qm_list_destroy(xfs_dqlist_t *);
60
61STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
ba0f32d4 62STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
1495f230 63STATIC int xfs_qm_shake(struct shrinker *, struct shrink_control *);
1da177e4 64
8e1f936b
RR
65static struct shrinker xfs_qm_shaker = {
66 .shrink = xfs_qm_shake,
67 .seeks = DEFAULT_SEEKS,
68};
69
1da177e4
LT
70/*
71 * Initialize the XQM structure.
72 * Note that there is not one quota manager per file system.
73 */
74STATIC struct xfs_qm *
75xfs_Gqm_init(void)
76{
6b3f6b5b
NS
77 xfs_dqhash_t *udqhash, *gdqhash;
78 xfs_qm_t *xqm;
215101c3
NS
79 size_t hsize;
80 uint i;
1da177e4
LT
81
82 /*
83 * Initialize the dquot hash tables.
84 */
77e4635a 85 udqhash = kmem_zalloc_greedy(&hsize,
5995cb7d 86 XFS_QM_HASHSIZE_LOW * sizeof(xfs_dqhash_t),
bdfb0430
CH
87 XFS_QM_HASHSIZE_HIGH * sizeof(xfs_dqhash_t));
88 if (!udqhash)
89 goto out;
90
91 gdqhash = kmem_zalloc_large(hsize);
d67b1b03 92 if (!gdqhash)
bdfb0430
CH
93 goto out_free_udqhash;
94
77e4635a 95 hsize /= sizeof(xfs_dqhash_t);
6b3f6b5b 96 ndquot = hsize << 8;
1da177e4 97
6b3f6b5b
NS
98 xqm = kmem_zalloc(sizeof(xfs_qm_t), KM_SLEEP);
99 xqm->qm_dqhashmask = hsize - 1;
100 xqm->qm_usr_dqhtable = udqhash;
101 xqm->qm_grp_dqhtable = gdqhash;
1da177e4
LT
102 ASSERT(xqm->qm_usr_dqhtable != NULL);
103 ASSERT(xqm->qm_grp_dqhtable != NULL);
104
105 for (i = 0; i < hsize; i++) {
106 xfs_qm_list_init(&(xqm->qm_usr_dqhtable[i]), "uxdqh", i);
107 xfs_qm_list_init(&(xqm->qm_grp_dqhtable[i]), "gxdqh", i);
108 }
109
110 /*
111 * Freelist of all dquots of all file systems
112 */
3a8406f6
DC
113 INIT_LIST_HEAD(&xqm->qm_dqfrlist);
114 xqm->qm_dqfrlist_cnt = 0;
115 mutex_init(&xqm->qm_dqfrlist_lock);
1da177e4
LT
116
117 /*
118 * dquot zone. we register our own low-memory callback.
119 */
120 if (!qm_dqzone) {
121 xqm->qm_dqzone = kmem_zone_init(sizeof(xfs_dquot_t),
122 "xfs_dquots");
123 qm_dqzone = xqm->qm_dqzone;
124 } else
125 xqm->qm_dqzone = qm_dqzone;
126
8e1f936b 127 register_shrinker(&xfs_qm_shaker);
1da177e4
LT
128
129 /*
130 * The t_dqinfo portion of transactions.
131 */
132 if (!qm_dqtrxzone) {
133 xqm->qm_dqtrxzone = kmem_zone_init(sizeof(xfs_dquot_acct_t),
134 "xfs_dqtrx");
135 qm_dqtrxzone = xqm->qm_dqtrxzone;
136 } else
137 xqm->qm_dqtrxzone = qm_dqtrxzone;
138
139 atomic_set(&xqm->qm_totaldquots, 0);
140 xqm->qm_dqfree_ratio = XFS_QM_DQFREE_RATIO;
141 xqm->qm_nrefs = 0;
1da177e4 142 return xqm;
bdfb0430
CH
143
144 out_free_udqhash:
145 kmem_free_large(udqhash);
146 out:
147 return NULL;
1da177e4
LT
148}
149
150/*
151 * Destroy the global quota manager when its reference count goes to zero.
152 */
ba0f32d4 153STATIC void
1da177e4
LT
154xfs_qm_destroy(
155 struct xfs_qm *xqm)
156{
157 int hsize, i;
158
159 ASSERT(xqm != NULL);
160 ASSERT(xqm->qm_nrefs == 0);
80a376bf 161
8e1f936b 162 unregister_shrinker(&xfs_qm_shaker);
80a376bf
CH
163
164 mutex_lock(&xqm->qm_dqfrlist_lock);
165 ASSERT(list_empty(&xqm->qm_dqfrlist));
166 mutex_unlock(&xqm->qm_dqfrlist_lock);
167
1da177e4
LT
168 hsize = xqm->qm_dqhashmask + 1;
169 for (i = 0; i < hsize; i++) {
170 xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
171 xfs_qm_list_destroy(&(xqm->qm_grp_dqhtable[i]));
172 }
bdfb0430
CH
173 kmem_free_large(xqm->qm_usr_dqhtable);
174 kmem_free_large(xqm->qm_grp_dqhtable);
1da177e4
LT
175 xqm->qm_usr_dqhtable = NULL;
176 xqm->qm_grp_dqhtable = NULL;
177 xqm->qm_dqhashmask = 0;
3a8406f6 178
f0e2d93c 179 kmem_free(xqm);
1da177e4
LT
180}
181
182/*
183 * Called at mount time to let XQM know that another file system is
184 * starting quotas. This isn't crucial information as the individual mount
185 * structures are pretty independent, but it helps the XQM keep a
186 * global view of what's going on.
187 */
188/* ARGSUSED */
189STATIC int
190xfs_qm_hold_quotafs_ref(
191 struct xfs_mount *mp)
192{
193 /*
194 * Need to lock the xfs_Gqm structure for things like this. For example,
195 * the structure could disappear between the entry to this routine and
196 * a HOLD operation if not locked.
197 */
e2494582 198 mutex_lock(&xfs_Gqm_lock);
1da177e4 199
bdfb0430 200 if (!xfs_Gqm) {
1da177e4 201 xfs_Gqm = xfs_Gqm_init();
38e712ab
JL
202 if (!xfs_Gqm) {
203 mutex_unlock(&xfs_Gqm_lock);
bdfb0430 204 return ENOMEM;
38e712ab 205 }
bdfb0430
CH
206 }
207
1da177e4
LT
208 /*
209 * We can keep a list of all filesystems with quotas mounted for
210 * debugging and statistical purposes, but ...
211 * Just take a reference and get out.
212 */
e2494582
CH
213 xfs_Gqm->qm_nrefs++;
214 mutex_unlock(&xfs_Gqm_lock);
1da177e4
LT
215
216 return 0;
217}
218
219
220/*
221 * Release the reference that a filesystem took at mount time,
222 * so that we know when we need to destroy the entire quota manager.
223 */
224/* ARGSUSED */
225STATIC void
226xfs_qm_rele_quotafs_ref(
227 struct xfs_mount *mp)
228{
1da177e4
LT
229 ASSERT(xfs_Gqm);
230 ASSERT(xfs_Gqm->qm_nrefs > 0);
231
1da177e4
LT
232 /*
233 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
234 * be restarted.
235 */
e2494582
CH
236 mutex_lock(&xfs_Gqm_lock);
237 if (--xfs_Gqm->qm_nrefs == 0) {
1da177e4
LT
238 xfs_qm_destroy(xfs_Gqm);
239 xfs_Gqm = NULL;
240 }
e2494582 241 mutex_unlock(&xfs_Gqm_lock);
1da177e4
LT
242}
243
1da177e4
LT
244/*
245 * Just destroy the quotainfo structure.
246 */
247void
7d095257
CH
248xfs_qm_unmount(
249 struct xfs_mount *mp)
1da177e4 250{
7d095257 251 if (mp->m_quotainfo) {
8112e9dc 252 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1da177e4 253 xfs_qm_destroy_quotainfo(mp);
7d095257 254 }
1da177e4
LT
255}
256
257
258/*
259 * This is called from xfs_mountfs to start quotas and initialize all
260 * necessary data structures like quotainfo. This is also responsible for
261 * running a quotacheck as necessary. We are guaranteed that the superblock
262 * is consistently read in at this point.
53aa7915
DC
263 *
264 * If we fail here, the mount will continue with quota turned off. We don't
265 * need to inidicate success or failure at all.
1da177e4 266 */
53aa7915 267void
1da177e4 268xfs_qm_mount_quotas(
4249023a 269 xfs_mount_t *mp)
1da177e4 270{
1da177e4
LT
271 int error = 0;
272 uint sbf;
273
1da177e4
LT
274 /*
275 * If quotas on realtime volumes is not supported, we disable
276 * quotas immediately.
277 */
278 if (mp->m_sb.sb_rextents) {
0b932ccc 279 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
1da177e4
LT
280 mp->m_qflags = 0;
281 goto write_changes;
282 }
283
1da177e4 284 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
155ffd07 285
1da177e4
LT
286 /*
287 * Allocate the quotainfo structure inside the mount struct, and
288 * create quotainode(s), and change/rev superblock if necessary.
289 */
53aa7915
DC
290 error = xfs_qm_init_quotainfo(mp);
291 if (error) {
1da177e4
LT
292 /*
293 * We must turn off quotas.
294 */
295 ASSERT(mp->m_quotainfo == NULL);
296 mp->m_qflags = 0;
297 goto write_changes;
298 }
299 /*
300 * If any of the quotas are not consistent, do a quotacheck.
301 */
4249023a 302 if (XFS_QM_NEED_QUOTACHECK(mp)) {
53aa7915
DC
303 error = xfs_qm_quotacheck(mp);
304 if (error) {
305 /* Quotacheck failed and disabled quotas. */
306 return;
1da177e4 307 }
1da177e4 308 }
646d5bda
DD
309 /*
310 * If one type of quotas is off, then it will lose its
311 * quotachecked status, since we won't be doing accounting for
312 * that type anymore.
313 */
53aa7915 314 if (!XFS_IS_UQUOTA_ON(mp))
646d5bda 315 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
53aa7915 316 if (!(XFS_IS_GQUOTA_ON(mp) || XFS_IS_PQUOTA_ON(mp)))
646d5bda 317 mp->m_qflags &= ~XFS_OQUOTA_CHKD;
155ffd07 318
1da177e4
LT
319 write_changes:
320 /*
3685c2a1 321 * We actually don't have to acquire the m_sb_lock at all.
1da177e4
LT
322 * This can only be called from mount, and that's single threaded. XXX
323 */
3685c2a1 324 spin_lock(&mp->m_sb_lock);
1da177e4
LT
325 sbf = mp->m_sb.sb_qflags;
326 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
3685c2a1 327 spin_unlock(&mp->m_sb_lock);
1da177e4
LT
328
329 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
330 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
331 /*
332 * We could only have been turning quotas off.
333 * We aren't in very good shape actually because
334 * the incore structures are convinced that quotas are
335 * off, but the on disk superblock doesn't know that !
336 */
337 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
53487786
DC
338 xfs_alert(mp, "%s: Superblock update failed!",
339 __func__);
1da177e4
LT
340 }
341 }
342
343 if (error) {
53487786 344 xfs_warn(mp, "Failed to initialize disk quotas.");
7d095257 345 return;
1da177e4 346 }
1da177e4
LT
347}
348
349/*
350 * Called from the vfsops layer.
351 */
e57481dc 352void
1da177e4
LT
353xfs_qm_unmount_quotas(
354 xfs_mount_t *mp)
355{
1da177e4
LT
356 /*
357 * Release the dquots that root inode, et al might be holding,
358 * before we flush quotas and blow away the quotainfo structure.
359 */
360 ASSERT(mp->m_rootip);
361 xfs_qm_dqdetach(mp->m_rootip);
362 if (mp->m_rbmip)
363 xfs_qm_dqdetach(mp->m_rbmip);
364 if (mp->m_rsumip)
365 xfs_qm_dqdetach(mp->m_rsumip);
366
367 /*
e57481dc 368 * Release the quota inodes.
1da177e4 369 */
1da177e4 370 if (mp->m_quotainfo) {
e57481dc
CH
371 if (mp->m_quotainfo->qi_uquotaip) {
372 IRELE(mp->m_quotainfo->qi_uquotaip);
373 mp->m_quotainfo->qi_uquotaip = NULL;
1da177e4 374 }
e57481dc
CH
375 if (mp->m_quotainfo->qi_gquotaip) {
376 IRELE(mp->m_quotainfo->qi_gquotaip);
377 mp->m_quotainfo->qi_gquotaip = NULL;
1da177e4
LT
378 }
379 }
1da177e4
LT
380}
381
382/*
383 * Flush all dquots of the given file system to disk. The dquots are
384 * _not_ purged from memory here, just their data written to disk.
385 */
ba0f32d4 386STATIC int
1da177e4 387xfs_qm_dqflush_all(
a7ef9bd7 388 struct xfs_mount *mp)
1da177e4 389{
8a7b8a89
CH
390 struct xfs_quotainfo *q = mp->m_quotainfo;
391 int recl;
392 struct xfs_dquot *dqp;
8a7b8a89 393 int error;
1da177e4 394
8a7b8a89 395 if (!q)
014c2544 396 return 0;
1da177e4 397again:
8a7b8a89
CH
398 mutex_lock(&q->qi_dqlist_lock);
399 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
1da177e4 400 xfs_dqlock(dqp);
92678554
CH
401 if ((dqp->dq_flags & XFS_DQ_FREEING) ||
402 !XFS_DQ_IS_DIRTY(dqp)) {
1da177e4
LT
403 xfs_dqunlock(dqp);
404 continue;
405 }
0b1b213f 406
1da177e4 407 /* XXX a sentinel would be better */
8a7b8a89 408 recl = q->qi_dqreclaims;
e1f49cf2 409 if (!xfs_dqflock_nowait(dqp)) {
1da177e4
LT
410 /*
411 * If we can't grab the flush lock then check
412 * to see if the dquot has been flushed delayed
413 * write. If so, grab its buffer and send it
414 * out immediately. We'll be able to acquire
415 * the flush lock when the I/O completes.
416 */
800b484e 417 xfs_dqflock_pushbuf_wait(dqp);
1da177e4
LT
418 }
419 /*
420 * Let go of the mplist lock. We don't want to hold it
421 * across a disk write.
422 */
8a7b8a89 423 mutex_unlock(&q->qi_dqlist_lock);
a7ef9bd7 424 error = xfs_qm_dqflush(dqp, 0);
1da177e4
LT
425 xfs_dqunlock(dqp);
426 if (error)
014c2544 427 return error;
1da177e4 428
8a7b8a89
CH
429 mutex_lock(&q->qi_dqlist_lock);
430 if (recl != q->qi_dqreclaims) {
431 mutex_unlock(&q->qi_dqlist_lock);
1da177e4
LT
432 /* XXX restart limit */
433 goto again;
434 }
435 }
436
8a7b8a89 437 mutex_unlock(&q->qi_dqlist_lock);
1da177e4 438 /* return ! busy */
014c2544 439 return 0;
1da177e4 440}
92678554 441
1da177e4
LT
442/*
443 * Release the group dquot pointers the user dquots may be
444 * carrying around as a hint. mplist is locked on entry and exit.
445 */
446STATIC void
447xfs_qm_detach_gdquots(
8a7b8a89 448 struct xfs_mount *mp)
1da177e4 449{
8a7b8a89
CH
450 struct xfs_quotainfo *q = mp->m_quotainfo;
451 struct xfs_dquot *dqp, *gdqp;
452 int nrecl;
1da177e4
LT
453
454 again:
8a7b8a89
CH
455 ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
456 list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
1da177e4 457 xfs_dqlock(dqp);
92678554
CH
458 if (dqp->dq_flags & XFS_DQ_FREEING) {
459 xfs_dqunlock(dqp);
460 mutex_unlock(&q->qi_dqlist_lock);
461 delay(1);
462 mutex_lock(&q->qi_dqlist_lock);
463 goto again;
464 }
1da177e4
LT
465 if ((gdqp = dqp->q_gdquot)) {
466 xfs_dqlock(gdqp);
467 dqp->q_gdquot = NULL;
468 }
469 xfs_dqunlock(dqp);
470
471 if (gdqp) {
472 /*
473 * Can't hold the mplist lock across a dqput.
474 * XXXmust convert to marker based iterations here.
475 */
8a7b8a89
CH
476 nrecl = q->qi_dqreclaims;
477 mutex_unlock(&q->qi_dqlist_lock);
1da177e4
LT
478 xfs_qm_dqput(gdqp);
479
8a7b8a89
CH
480 mutex_lock(&q->qi_dqlist_lock);
481 if (nrecl != q->qi_dqreclaims)
1da177e4
LT
482 goto again;
483 }
1da177e4
LT
484 }
485}
486
487/*
488 * Go through all the incore dquots of this file system and take them
489 * off the mplist and hashlist, if the dquot type matches the dqtype
490 * parameter. This is used when turning off quota accounting for
491 * users and/or groups, as well as when the filesystem is unmounting.
492 */
493STATIC int
494xfs_qm_dqpurge_int(
8a7b8a89
CH
495 struct xfs_mount *mp,
496 uint flags)
1da177e4 497{
8a7b8a89
CH
498 struct xfs_quotainfo *q = mp->m_quotainfo;
499 struct xfs_dquot *dqp, *n;
500 uint dqtype;
92678554
CH
501 int nmisses = 0;
502 LIST_HEAD (dispose_list);
1da177e4 503
8a7b8a89 504 if (!q)
014c2544 505 return 0;
1da177e4
LT
506
507 dqtype = (flags & XFS_QMOPT_UQUOTA) ? XFS_DQ_USER : 0;
c8ad20ff 508 dqtype |= (flags & XFS_QMOPT_PQUOTA) ? XFS_DQ_PROJ : 0;
1da177e4
LT
509 dqtype |= (flags & XFS_QMOPT_GQUOTA) ? XFS_DQ_GROUP : 0;
510
8a7b8a89 511 mutex_lock(&q->qi_dqlist_lock);
1da177e4
LT
512
513 /*
514 * In the first pass through all incore dquots of this filesystem,
515 * we release the group dquot pointers the user dquots may be
516 * carrying around as a hint. We need to do this irrespective of
517 * what's being turned off.
518 */
519 xfs_qm_detach_gdquots(mp);
520
1da177e4 521 /*
92678554 522 * Try to get rid of all of the unwanted dquots.
1da177e4 523 */
8a7b8a89 524 list_for_each_entry_safe(dqp, n, &q->qi_dqlist, q_mplist) {
be7ffc38 525 xfs_dqlock(dqp);
92678554
CH
526 if ((dqp->dq_flags & dqtype) != 0 &&
527 !(dqp->dq_flags & XFS_DQ_FREEING)) {
528 if (dqp->q_nrefs == 0) {
529 dqp->dq_flags |= XFS_DQ_FREEING;
530 list_move_tail(&dqp->q_mplist, &dispose_list);
531 } else
532 nmisses++;
be7ffc38
CH
533 }
534 xfs_dqunlock(dqp);
1da177e4 535 }
8a7b8a89 536 mutex_unlock(&q->qi_dqlist_lock);
92678554
CH
537
538 list_for_each_entry_safe(dqp, n, &dispose_list, q_mplist)
539 xfs_qm_dqpurge(dqp);
540
1da177e4
LT
541 return nmisses;
542}
543
544int
545xfs_qm_dqpurge_all(
546 xfs_mount_t *mp,
547 uint flags)
548{
549 int ndquots;
550
551 /*
552 * Purge the dquot cache.
553 * None of the dquots should really be busy at this point.
554 */
555 if (mp->m_quotainfo) {
556 while ((ndquots = xfs_qm_dqpurge_int(mp, flags))) {
557 delay(ndquots * 10);
558 }
559 }
560 return 0;
561}
562
563STATIC int
564xfs_qm_dqattach_one(
565 xfs_inode_t *ip,
566 xfs_dqid_t id,
567 uint type,
568 uint doalloc,
1da177e4
LT
569 xfs_dquot_t *udqhint, /* hint */
570 xfs_dquot_t **IO_idqpp)
571{
572 xfs_dquot_t *dqp;
573 int error;
574
579aa9ca 575 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4 576 error = 0;
8e9b6e7f 577
1da177e4
LT
578 /*
579 * See if we already have it in the inode itself. IO_idqpp is
580 * &i_udquot or &i_gdquot. This made the code look weird, but
581 * made the logic a lot simpler.
582 */
8e9b6e7f
CH
583 dqp = *IO_idqpp;
584 if (dqp) {
0b1b213f 585 trace_xfs_dqattach_found(dqp);
8e9b6e7f 586 return 0;
1da177e4
LT
587 }
588
589 /*
590 * udqhint is the i_udquot field in inode, and is non-NULL only
c8ad20ff 591 * when the type arg is group/project. Its purpose is to save a
1da177e4
LT
592 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
593 * the user dquot.
594 */
8e9b6e7f
CH
595 if (udqhint) {
596 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
1da177e4
LT
597 xfs_dqlock(udqhint);
598
8e9b6e7f
CH
599 /*
600 * No need to take dqlock to look at the id.
601 *
602 * The ID can't change until it gets reclaimed, and it won't
603 * be reclaimed as long as we have a ref from inode and we
604 * hold the ilock.
605 */
606 dqp = udqhint->q_gdquot;
607 if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
608 xfs_dqlock(dqp);
609 XFS_DQHOLD(dqp);
610 ASSERT(*IO_idqpp == NULL);
611 *IO_idqpp = dqp;
612
1da177e4
LT
613 xfs_dqunlock(dqp);
614 xfs_dqunlock(udqhint);
8e9b6e7f 615 return 0;
1da177e4 616 }
8e9b6e7f
CH
617
618 /*
619 * We can't hold a dquot lock when we call the dqget code.
620 * We'll deadlock in no time, because of (not conforming to)
621 * lock ordering - the inodelock comes before any dquot lock,
622 * and we may drop and reacquire the ilock in xfs_qm_dqget().
623 */
1da177e4 624 xfs_dqunlock(udqhint);
8e9b6e7f
CH
625 }
626
1da177e4
LT
627 /*
628 * Find the dquot from somewhere. This bumps the
629 * reference count of dquot and returns it locked.
630 * This can return ENOENT if dquot didn't exist on
631 * disk and we didn't ask it to allocate;
632 * ESRCH if quotas got turned off suddenly.
633 */
db3e74b5
MH
634 error = xfs_qm_dqget(ip->i_mount, ip, id, type,
635 doalloc | XFS_QMOPT_DOWARN, &dqp);
8e9b6e7f
CH
636 if (error)
637 return error;
1da177e4 638
0b1b213f 639 trace_xfs_dqattach_get(dqp);
8e9b6e7f 640
1da177e4
LT
641 /*
642 * dqget may have dropped and re-acquired the ilock, but it guarantees
643 * that the dquot returned is the one that should go in the inode.
644 */
645 *IO_idqpp = dqp;
8e9b6e7f
CH
646 xfs_dqunlock(dqp);
647 return 0;
1da177e4
LT
648}
649
650
651/*
652 * Given a udquot and gdquot, attach a ptr to the group dquot in the
ab680bb7 653 * udquot as a hint for future lookups.
1da177e4
LT
654 */
655STATIC void
656xfs_qm_dqattach_grouphint(
657 xfs_dquot_t *udq,
8e9b6e7f 658 xfs_dquot_t *gdq)
1da177e4
LT
659{
660 xfs_dquot_t *tmp;
661
8e9b6e7f 662 xfs_dqlock(udq);
1da177e4 663
ab680bb7
CH
664 tmp = udq->q_gdquot;
665 if (tmp) {
666 if (tmp == gdq)
667 goto done;
1da177e4
LT
668
669 udq->q_gdquot = NULL;
1da177e4 670 xfs_qm_dqrele(tmp);
1da177e4 671 }
8e9b6e7f 672
ab680bb7
CH
673 xfs_dqlock(gdq);
674 XFS_DQHOLD(gdq);
8e9b6e7f 675 xfs_dqunlock(gdq);
ab680bb7
CH
676
677 udq->q_gdquot = gdq;
678done:
8e9b6e7f 679 xfs_dqunlock(udq);
1da177e4
LT
680}
681
682
683/*
c8ad20ff
NS
684 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
685 * into account.
1da177e4 686 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
1da177e4
LT
687 * Inode may get unlocked and relocked in here, and the caller must deal with
688 * the consequences.
689 */
690int
7d095257 691xfs_qm_dqattach_locked(
1da177e4
LT
692 xfs_inode_t *ip,
693 uint flags)
694{
695 xfs_mount_t *mp = ip->i_mount;
696 uint nquotas = 0;
697 int error = 0;
698
7d095257
CH
699 if (!XFS_IS_QUOTA_RUNNING(mp) ||
700 !XFS_IS_QUOTA_ON(mp) ||
701 !XFS_NOT_DQATTACHED(mp, ip) ||
702 ip->i_ino == mp->m_sb.sb_uquotino ||
703 ip->i_ino == mp->m_sb.sb_gquotino)
014c2544 704 return 0;
1da177e4 705
7d095257 706 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4
LT
707
708 if (XFS_IS_UQUOTA_ON(mp)) {
709 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
710 flags & XFS_QMOPT_DQALLOC,
1da177e4
LT
711 NULL, &ip->i_udquot);
712 if (error)
713 goto done;
714 nquotas++;
715 }
579aa9ca
CH
716
717 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
c8ad20ff
NS
718 if (XFS_IS_OQUOTA_ON(mp)) {
719 error = XFS_IS_GQUOTA_ON(mp) ?
720 xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
721 flags & XFS_QMOPT_DQALLOC,
c8ad20ff 722 ip->i_udquot, &ip->i_gdquot) :
6743099c 723 xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
1da177e4 724 flags & XFS_QMOPT_DQALLOC,
1da177e4
LT
725 ip->i_udquot, &ip->i_gdquot);
726 /*
727 * Don't worry about the udquot that we may have
728 * attached above. It'll get detached, if not already.
729 */
730 if (error)
731 goto done;
732 nquotas++;
733 }
734
735 /*
736 * Attach this group quota to the user quota as a hint.
737 * This WON'T, in general, result in a thrash.
738 */
739 if (nquotas == 2) {
579aa9ca 740 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4
LT
741 ASSERT(ip->i_udquot);
742 ASSERT(ip->i_gdquot);
743
744 /*
ab680bb7
CH
745 * We do not have i_udquot locked at this point, but this check
746 * is OK since we don't depend on the i_gdquot to be accurate
747 * 100% all the time. It is just a hint, and this will
748 * succeed in general.
1da177e4 749 */
ab680bb7
CH
750 if (ip->i_udquot->q_gdquot != ip->i_gdquot)
751 xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot);
1da177e4
LT
752 }
753
7d095257 754 done:
ea15ab3c
CH
755#ifdef DEBUG
756 if (!error) {
1da177e4
LT
757 if (XFS_IS_UQUOTA_ON(mp))
758 ASSERT(ip->i_udquot);
c8ad20ff 759 if (XFS_IS_OQUOTA_ON(mp))
1da177e4
LT
760 ASSERT(ip->i_gdquot);
761 }
7d095257 762 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4 763#endif
7d095257
CH
764 return error;
765}
1da177e4 766
7d095257
CH
767int
768xfs_qm_dqattach(
769 struct xfs_inode *ip,
770 uint flags)
771{
772 int error;
773
774 xfs_ilock(ip, XFS_ILOCK_EXCL);
775 error = xfs_qm_dqattach_locked(ip, flags);
776 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1da177e4 777
014c2544 778 return error;
1da177e4
LT
779}
780
781/*
782 * Release dquots (and their references) if any.
783 * The inode should be locked EXCL except when this's called by
784 * xfs_ireclaim.
785 */
786void
787xfs_qm_dqdetach(
788 xfs_inode_t *ip)
789{
790 if (!(ip->i_udquot || ip->i_gdquot))
791 return;
792
0b1b213f
CH
793 trace_xfs_dquot_dqdetach(ip);
794
1da177e4
LT
795 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
796 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
1da177e4
LT
797 if (ip->i_udquot) {
798 xfs_qm_dqrele(ip->i_udquot);
799 ip->i_udquot = NULL;
800 }
801 if (ip->i_gdquot) {
802 xfs_qm_dqrele(ip->i_gdquot);
803 ip->i_gdquot = NULL;
804 }
805}
806
a4edd1da
CH
807/*
808 * The hash chains and the mplist use the same xfs_dqhash structure as
809 * their list head, but we can take the mplist qh_lock and one of the
810 * hash qh_locks at the same time without any problem as they aren't
811 * related.
812 */
813static struct lock_class_key xfs_quota_mplist_class;
1da177e4
LT
814
815/*
816 * This initializes all the quota information that's kept in the
817 * mount structure
818 */
ba0f32d4 819STATIC int
1da177e4
LT
820xfs_qm_init_quotainfo(
821 xfs_mount_t *mp)
822{
823 xfs_quotainfo_t *qinf;
824 int error;
825 xfs_dquot_t *dqp;
826
827 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
828
829 /*
830 * Tell XQM that we exist as soon as possible.
831 */
832 if ((error = xfs_qm_hold_quotafs_ref(mp))) {
014c2544 833 return error;
1da177e4
LT
834 }
835
836 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
837
838 /*
839 * See if quotainodes are setup, and if not, allocate them,
840 * and change the superblock accordingly.
841 */
842 if ((error = xfs_qm_init_quotainos(mp))) {
f0e2d93c 843 kmem_free(qinf);
1da177e4 844 mp->m_quotainfo = NULL;
014c2544 845 return error;
1da177e4
LT
846 }
847
3a25404b
DC
848 INIT_LIST_HEAD(&qinf->qi_dqlist);
849 mutex_init(&qinf->qi_dqlist_lock);
850 lockdep_set_class(&qinf->qi_dqlist_lock, &xfs_quota_mplist_class);
a4edd1da 851
1da177e4
LT
852 qinf->qi_dqreclaims = 0;
853
854 /* mutex used to serialize quotaoffs */
794ee1ba 855 mutex_init(&qinf->qi_quotaofflock);
1da177e4
LT
856
857 /* Precalc some constants */
858 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
859 ASSERT(qinf->qi_dqchunklen);
860 qinf->qi_dqperchunk = BBTOB(qinf->qi_dqchunklen);
861 do_div(qinf->qi_dqperchunk, sizeof(xfs_dqblk_t));
862
863 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
864
865 /*
866 * We try to get the limits from the superuser's limits fields.
867 * This is quite hacky, but it is standard quota practice.
868 * We look at the USR dquot with id == 0 first, but if user quotas
869 * are not enabled we goto the GRP dquot with id == 0.
870 * We don't really care to keep separate default limits for user
871 * and group quotas, at least not at this point.
872 */
873 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)0,
c8ad20ff
NS
874 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
875 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
876 XFS_DQ_PROJ),
1da177e4
LT
877 XFS_QMOPT_DQSUSER|XFS_QMOPT_DOWARN,
878 &dqp);
879 if (! error) {
880 xfs_disk_dquot_t *ddqp = &dqp->q_core;
881
882 /*
883 * The warnings and timers set the grace period given to
884 * a user or group before he or she can not perform any
885 * more writing. If it is zero, a default is used.
886 */
1149d96a
CH
887 qinf->qi_btimelimit = ddqp->d_btimer ?
888 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
889 qinf->qi_itimelimit = ddqp->d_itimer ?
890 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
891 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
892 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
893 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
894 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
895 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
896 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
897 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
898 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
899 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
900 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
901 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
902 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
903 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
904 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
1da177e4
LT
905
906 /*
907 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
908 * we don't want this dquot cached. We haven't done a
909 * quotacheck yet, and quotacheck doesn't like incore dquots.
910 */
911 xfs_qm_dqdestroy(dqp);
912 } else {
913 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
914 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
915 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
916 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
917 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
06d10dd9 918 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
1da177e4
LT
919 }
920
014c2544 921 return 0;
1da177e4
LT
922}
923
924
925/*
926 * Gets called when unmounting a filesystem or when all quotas get
927 * turned off.
928 * This purges the quota inodes, destroys locks and frees itself.
929 */
930void
931xfs_qm_destroy_quotainfo(
932 xfs_mount_t *mp)
933{
934 xfs_quotainfo_t *qi;
935
936 qi = mp->m_quotainfo;
937 ASSERT(qi != NULL);
938 ASSERT(xfs_Gqm != NULL);
939
940 /*
941 * Release the reference that XQM kept, so that we know
942 * when the XQM structure should be freed. We cannot assume
943 * that xfs_Gqm is non-null after this point.
944 */
945 xfs_qm_rele_quotafs_ref(mp);
946
3a25404b
DC
947 ASSERT(list_empty(&qi->qi_dqlist));
948 mutex_destroy(&qi->qi_dqlist_lock);
1da177e4
LT
949
950 if (qi->qi_uquotaip) {
26cc0021 951 IRELE(qi->qi_uquotaip);
1da177e4
LT
952 qi->qi_uquotaip = NULL; /* paranoia */
953 }
954 if (qi->qi_gquotaip) {
26cc0021 955 IRELE(qi->qi_gquotaip);
1da177e4
LT
956 qi->qi_gquotaip = NULL;
957 }
958 mutex_destroy(&qi->qi_quotaofflock);
f0e2d93c 959 kmem_free(qi);
1da177e4
LT
960 mp->m_quotainfo = NULL;
961}
962
963
964
965/* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
966
967/* ARGSUSED */
968STATIC void
969xfs_qm_list_init(
970 xfs_dqlist_t *list,
971 char *str,
972 int n)
973{
794ee1ba 974 mutex_init(&list->qh_lock);
e6a81f13 975 INIT_LIST_HEAD(&list->qh_list);
1da177e4
LT
976 list->qh_version = 0;
977 list->qh_nelems = 0;
978}
979
980STATIC void
981xfs_qm_list_destroy(
982 xfs_dqlist_t *list)
983{
984 mutex_destroy(&(list->qh_lock));
985}
986
1da177e4
LT
987/*
988 * Create an inode and return with a reference already taken, but unlocked
989 * This is how we create quota inodes
990 */
991STATIC int
992xfs_qm_qino_alloc(
993 xfs_mount_t *mp,
994 xfs_inode_t **ip,
995 __int64_t sbfields,
996 uint flags)
997{
998 xfs_trans_t *tp;
999 int error;
1da177e4
LT
1000 int committed;
1001
061f7209 1002 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
1da177e4
LT
1003 if ((error = xfs_trans_reserve(tp,
1004 XFS_QM_QINOCREATE_SPACE_RES(mp),
1005 XFS_CREATE_LOG_RES(mp), 0,
1006 XFS_TRANS_PERM_LOG_RES,
1007 XFS_CREATE_LOG_COUNT))) {
1008 xfs_trans_cancel(tp, 0);
014c2544 1009 return error;
1da177e4 1010 }
1da177e4 1011
6c77b0ea
CH
1012 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip, &committed);
1013 if (error) {
1da177e4
LT
1014 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1015 XFS_TRANS_ABORT);
014c2544 1016 return error;
1da177e4
LT
1017 }
1018
1da177e4
LT
1019 /*
1020 * Make the changes in the superblock, and log those too.
1021 * sbfields arg may contain fields other than *QUOTINO;
1022 * VERSIONNUM for example.
1023 */
3685c2a1 1024 spin_lock(&mp->m_sb_lock);
1da177e4 1025 if (flags & XFS_QMOPT_SBVERSION) {
62118709 1026 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1da177e4
LT
1027 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1028 XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
1029 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1030 XFS_SB_GQUOTINO | XFS_SB_QFLAGS));
1031
62118709 1032 xfs_sb_version_addquota(&mp->m_sb);
1da177e4
LT
1033 mp->m_sb.sb_uquotino = NULLFSINO;
1034 mp->m_sb.sb_gquotino = NULLFSINO;
1035
1036 /* qflags will get updated _after_ quotacheck */
1037 mp->m_sb.sb_qflags = 0;
1da177e4
LT
1038 }
1039 if (flags & XFS_QMOPT_UQUOTA)
1040 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1041 else
1042 mp->m_sb.sb_gquotino = (*ip)->i_ino;
3685c2a1 1043 spin_unlock(&mp->m_sb_lock);
1da177e4
LT
1044 xfs_mod_sb(tp, sbfields);
1045
1c72bf90 1046 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
53487786 1047 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
014c2544 1048 return error;
1da177e4 1049 }
014c2544 1050 return 0;
1da177e4
LT
1051}
1052
1053
5b139738 1054STATIC void
1da177e4
LT
1055xfs_qm_reset_dqcounts(
1056 xfs_mount_t *mp,
1057 xfs_buf_t *bp,
1058 xfs_dqid_t id,
1059 uint type)
1060{
1061 xfs_disk_dquot_t *ddq;
1062 int j;
1063
0b1b213f
CH
1064 trace_xfs_reset_dqcounts(bp, _RET_IP_);
1065
1da177e4
LT
1066 /*
1067 * Reset all counters and timers. They'll be
1068 * started afresh by xfs_qm_quotacheck.
1069 */
1070#ifdef DEBUG
1071 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1072 do_div(j, sizeof(xfs_dqblk_t));
8a7b8a89 1073 ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
1da177e4 1074#endif
62926044 1075 ddq = bp->b_addr;
8a7b8a89 1076 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
1da177e4
LT
1077 /*
1078 * Do a sanity check, and if needed, repair the dqblk. Don't
1079 * output any warnings because it's perfectly possible to
c41564b5 1080 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1da177e4 1081 */
a0fa2b67 1082 (void) xfs_qm_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1da177e4 1083 "xfs_quotacheck");
1149d96a
CH
1084 ddq->d_bcount = 0;
1085 ddq->d_icount = 0;
1086 ddq->d_rtbcount = 0;
1087 ddq->d_btimer = 0;
1088 ddq->d_itimer = 0;
1089 ddq->d_rtbtimer = 0;
1090 ddq->d_bwarns = 0;
1091 ddq->d_iwarns = 0;
1092 ddq->d_rtbwarns = 0;
1da177e4
LT
1093 ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
1094 }
1da177e4
LT
1095}
1096
1097STATIC int
1098xfs_qm_dqiter_bufs(
1099 xfs_mount_t *mp,
1100 xfs_dqid_t firstid,
1101 xfs_fsblock_t bno,
1102 xfs_filblks_t blkcnt,
1103 uint flags)
1104{
1105 xfs_buf_t *bp;
1106 int error;
c8ad20ff 1107 int type;
1da177e4
LT
1108
1109 ASSERT(blkcnt > 0);
c8ad20ff
NS
1110 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1111 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1da177e4
LT
1112 error = 0;
1113
1114 /*
1115 * Blkcnt arg can be a very big number, and might even be
1116 * larger than the log itself. So, we have to break it up into
1117 * manageable-sized transactions.
1118 * Note that we don't start a permanent transaction here; we might
1119 * not be able to get a log reservation for the whole thing up front,
1120 * and we don't really care to either, because we just discard
1121 * everything if we were to crash in the middle of this loop.
1122 */
1123 while (blkcnt--) {
1124 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1125 XFS_FSB_TO_DADDR(mp, bno),
8a7b8a89 1126 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
1da177e4
LT
1127 if (error)
1128 break;
1129
5b139738 1130 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
61551f1e
CH
1131 xfs_buf_delwri_queue(bp);
1132 xfs_buf_relse(bp);
1da177e4
LT
1133 /*
1134 * goto the next block.
1135 */
1136 bno++;
8a7b8a89 1137 firstid += mp->m_quotainfo->qi_dqperchunk;
1da177e4 1138 }
014c2544 1139 return error;
1da177e4
LT
1140}
1141
1142/*
c8ad20ff 1143 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1da177e4
LT
1144 * caller supplied function for every chunk of dquots that we find.
1145 */
1146STATIC int
1147xfs_qm_dqiterate(
1148 xfs_mount_t *mp,
1149 xfs_inode_t *qip,
1150 uint flags)
1151{
1152 xfs_bmbt_irec_t *map;
1153 int i, nmaps; /* number of map entries */
1154 int error; /* return value */
1155 xfs_fileoff_t lblkno;
1156 xfs_filblks_t maxlblkcnt;
1157 xfs_dqid_t firstid;
1158 xfs_fsblock_t rablkno;
1159 xfs_filblks_t rablkcnt;
1160
1161 error = 0;
1162 /*
c41564b5 1163 * This looks racy, but we can't keep an inode lock across a
1da177e4
LT
1164 * trans_reserve. But, this gets called during quotacheck, and that
1165 * happens only at mount time which is single threaded.
1166 */
1167 if (qip->i_d.di_nblocks == 0)
014c2544 1168 return 0;
1da177e4
LT
1169
1170 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1171
1172 lblkno = 0;
1173 maxlblkcnt = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1174 do {
1175 nmaps = XFS_DQITER_MAP_SIZE;
1176 /*
1177 * We aren't changing the inode itself. Just changing
1178 * some of its data. No new blocks are added here, and
1179 * the inode is never added to the transaction.
1180 */
1181 xfs_ilock(qip, XFS_ILOCK_SHARED);
5c8ed202
DC
1182 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
1183 map, &nmaps, 0);
1da177e4
LT
1184 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1185 if (error)
1186 break;
1187
1188 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1189 for (i = 0; i < nmaps; i++) {
1190 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1191 ASSERT(map[i].br_blockcount);
1192
1193
1194 lblkno += map[i].br_blockcount;
1195
1196 if (map[i].br_startblock == HOLESTARTBLOCK)
1197 continue;
1198
1199 firstid = (xfs_dqid_t) map[i].br_startoff *
8a7b8a89 1200 mp->m_quotainfo->qi_dqperchunk;
1da177e4
LT
1201 /*
1202 * Do a read-ahead on the next extent.
1203 */
1204 if ((i+1 < nmaps) &&
1205 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1206 rablkcnt = map[i+1].br_blockcount;
1207 rablkno = map[i+1].br_startblock;
1208 while (rablkcnt--) {
1a1a3e97 1209 xfs_buf_readahead(mp->m_ddev_targp,
1da177e4 1210 XFS_FSB_TO_DADDR(mp, rablkno),
8a7b8a89 1211 mp->m_quotainfo->qi_dqchunklen);
1da177e4
LT
1212 rablkno++;
1213 }
1214 }
1215 /*
1216 * Iterate thru all the blks in the extent and
1217 * reset the counters of all the dquots inside them.
1218 */
1219 if ((error = xfs_qm_dqiter_bufs(mp,
1220 firstid,
1221 map[i].br_startblock,
1222 map[i].br_blockcount,
1223 flags))) {
1224 break;
1225 }
1226 }
1227
1228 if (error)
1229 break;
1230 } while (nmaps > 0);
1231
f0e2d93c 1232 kmem_free(map);
1da177e4 1233
014c2544 1234 return error;
1da177e4
LT
1235}
1236
1237/*
1238 * Called by dqusage_adjust in doing a quotacheck.
52fda114
CH
1239 *
1240 * Given the inode, and a dquot id this updates both the incore dqout as well
1241 * as the buffer copy. This is so that once the quotacheck is done, we can
1242 * just log all the buffers, as opposed to logging numerous updates to
1243 * individual dquots.
1da177e4 1244 */
52fda114 1245STATIC int
1da177e4 1246xfs_qm_quotacheck_dqadjust(
52fda114
CH
1247 struct xfs_inode *ip,
1248 xfs_dqid_t id,
1249 uint type,
1da177e4
LT
1250 xfs_qcnt_t nblks,
1251 xfs_qcnt_t rtblks)
1252{
52fda114
CH
1253 struct xfs_mount *mp = ip->i_mount;
1254 struct xfs_dquot *dqp;
1255 int error;
1256
1257 error = xfs_qm_dqget(mp, ip, id, type,
1258 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1259 if (error) {
1260 /*
1261 * Shouldn't be able to turn off quotas here.
1262 */
1263 ASSERT(error != ESRCH);
1264 ASSERT(error != ENOENT);
1265 return error;
1266 }
0b1b213f
CH
1267
1268 trace_xfs_dqadjust(dqp);
1269
1da177e4
LT
1270 /*
1271 * Adjust the inode count and the block count to reflect this inode's
1272 * resource usage.
1273 */
413d57c9 1274 be64_add_cpu(&dqp->q_core.d_icount, 1);
1da177e4
LT
1275 dqp->q_res_icount++;
1276 if (nblks) {
413d57c9 1277 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1da177e4
LT
1278 dqp->q_res_bcount += nblks;
1279 }
1280 if (rtblks) {
413d57c9 1281 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1da177e4
LT
1282 dqp->q_res_rtbcount += rtblks;
1283 }
1284
1285 /*
1286 * Set default limits, adjust timers (since we changed usages)
191f8488
CH
1287 *
1288 * There are no timers for the default values set in the root dquot.
1da177e4 1289 */
191f8488 1290 if (dqp->q_core.d_id) {
52fda114
CH
1291 xfs_qm_adjust_dqlimits(mp, &dqp->q_core);
1292 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1da177e4
LT
1293 }
1294
1295 dqp->dq_flags |= XFS_DQ_DIRTY;
52fda114
CH
1296 xfs_qm_dqput(dqp);
1297 return 0;
1da177e4
LT
1298}
1299
1300STATIC int
1301xfs_qm_get_rtblks(
1302 xfs_inode_t *ip,
1303 xfs_qcnt_t *O_rtblks)
1304{
1305 xfs_filblks_t rtblks; /* total rt blks */
4eea22f0 1306 xfs_extnum_t idx; /* extent record index */
1da177e4
LT
1307 xfs_ifork_t *ifp; /* inode fork pointer */
1308 xfs_extnum_t nextents; /* number of extent entries */
1da177e4
LT
1309 int error;
1310
1311 ASSERT(XFS_IS_REALTIME_INODE(ip));
1312 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1313 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1314 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
014c2544 1315 return error;
1da177e4
LT
1316 }
1317 rtblks = 0;
4eea22f0 1318 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
a6f64d4a
CH
1319 for (idx = 0; idx < nextents; idx++)
1320 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1da177e4 1321 *O_rtblks = (xfs_qcnt_t)rtblks;
014c2544 1322 return 0;
1da177e4
LT
1323}
1324
1325/*
1326 * callback routine supplied to bulkstat(). Given an inumber, find its
1327 * dquots and update them to account for resources taken by that inode.
1328 */
1329/* ARGSUSED */
1330STATIC int
1331xfs_qm_dqusage_adjust(
1332 xfs_mount_t *mp, /* mount point for filesystem */
1333 xfs_ino_t ino, /* inode number to get data for */
1334 void __user *buffer, /* not used */
1335 int ubsize, /* not used */
1da177e4 1336 int *ubused, /* not used */
1da177e4
LT
1337 int *res) /* result code value */
1338{
1339 xfs_inode_t *ip;
52fda114 1340 xfs_qcnt_t nblks, rtblks = 0;
1da177e4
LT
1341 int error;
1342
1343 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1344
1345 /*
1346 * rootino must have its resources accounted for, not so with the quota
1347 * inodes.
1348 */
1349 if (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino) {
1350 *res = BULKSTAT_RV_NOTHING;
1351 return XFS_ERROR(EINVAL);
1352 }
1353
1354 /*
1355 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1356 * interface expects the inode to be exclusively locked because that's
1357 * the case in all other instances. It's OK that we do this because
1358 * quotacheck is done only at mount time.
1359 */
52fda114
CH
1360 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1361 if (error) {
1da177e4 1362 *res = BULKSTAT_RV_NOTHING;
014c2544 1363 return error;
1da177e4
LT
1364 }
1365
52fda114 1366 ASSERT(ip->i_delayed_blks == 0);
1da177e4 1367
52fda114 1368 if (XFS_IS_REALTIME_INODE(ip)) {
1da177e4
LT
1369 /*
1370 * Walk thru the extent list and count the realtime blocks.
1371 */
52fda114
CH
1372 error = xfs_qm_get_rtblks(ip, &rtblks);
1373 if (error)
1374 goto error0;
1da177e4 1375 }
1da177e4 1376
52fda114 1377 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1da177e4
LT
1378
1379 /*
1380 * Add the (disk blocks and inode) resources occupied by this
1381 * inode to its dquots. We do this adjustment in the incore dquot,
1382 * and also copy the changes to its buffer.
1383 * We don't care about putting these changes in a transaction
1384 * envelope because if we crash in the middle of a 'quotacheck'
1385 * we have to start from the beginning anyway.
1386 * Once we're done, we'll log all the dquot bufs.
1387 *
c41564b5 1388 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1da177e4
LT
1389 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1390 */
1391 if (XFS_IS_UQUOTA_ON(mp)) {
52fda114
CH
1392 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1393 XFS_DQ_USER, nblks, rtblks);
1394 if (error)
1395 goto error0;
1da177e4 1396 }
52fda114
CH
1397
1398 if (XFS_IS_GQUOTA_ON(mp)) {
1399 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1400 XFS_DQ_GROUP, nblks, rtblks);
1401 if (error)
1402 goto error0;
1da177e4 1403 }
1da177e4 1404
52fda114 1405 if (XFS_IS_PQUOTA_ON(mp)) {
6743099c 1406 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
52fda114
CH
1407 XFS_DQ_PROJ, nblks, rtblks);
1408 if (error)
1409 goto error0;
1410 }
1411
1412 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1413 IRELE(ip);
1da177e4 1414 *res = BULKSTAT_RV_DIDONE;
014c2544 1415 return 0;
52fda114
CH
1416
1417error0:
1418 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1419 IRELE(ip);
1420 *res = BULKSTAT_RV_GIVEUP;
1421 return error;
1da177e4
LT
1422}
1423
1424/*
1425 * Walk thru all the filesystem inodes and construct a consistent view
1426 * of the disk quota world. If the quotacheck fails, disable quotas.
1427 */
1428int
1429xfs_qm_quotacheck(
1430 xfs_mount_t *mp)
1431{
1432 int done, count, error;
1433 xfs_ino_t lastino;
1434 size_t structsz;
1435 xfs_inode_t *uip, *gip;
1436 uint flags;
1437
1438 count = INT_MAX;
1439 structsz = 1;
1440 lastino = 0;
1441 flags = 0;
1442
8a7b8a89 1443 ASSERT(mp->m_quotainfo->qi_uquotaip || mp->m_quotainfo->qi_gquotaip);
1da177e4
LT
1444 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1445
1446 /*
1447 * There should be no cached dquots. The (simplistic) quotacheck
1448 * algorithm doesn't like that.
1449 */
3a25404b 1450 ASSERT(list_empty(&mp->m_quotainfo->qi_dqlist));
1da177e4 1451
0b932ccc 1452 xfs_notice(mp, "Quotacheck needed: Please wait.");
1da177e4
LT
1453
1454 /*
c8ad20ff 1455 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1da177e4
LT
1456 * their counters to zero. We need a clean slate.
1457 * We don't log our changes till later.
1458 */
8a7b8a89
CH
1459 uip = mp->m_quotainfo->qi_uquotaip;
1460 if (uip) {
1461 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA);
1462 if (error)
1da177e4
LT
1463 goto error_return;
1464 flags |= XFS_UQUOTA_CHKD;
1465 }
1466
8a7b8a89
CH
1467 gip = mp->m_quotainfo->qi_gquotaip;
1468 if (gip) {
1469 error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
1470 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1471 if (error)
1da177e4 1472 goto error_return;
c8ad20ff 1473 flags |= XFS_OQUOTA_CHKD;
1da177e4
LT
1474 }
1475
1476 do {
1477 /*
1478 * Iterate thru all the inodes in the file system,
1479 * adjusting the corresponding dquot counters in core.
1480 */
7dce11db
CH
1481 error = xfs_bulkstat(mp, &lastino, &count,
1482 xfs_qm_dqusage_adjust,
1483 structsz, NULL, &done);
1484 if (error)
1da177e4
LT
1485 break;
1486
7dce11db 1487 } while (!done);
1da177e4 1488
4b8879df
DC
1489 /*
1490 * We've made all the changes that we need to make incore.
1491 * Flush them down to disk buffers if everything was updated
1492 * successfully.
1493 */
1494 if (!error)
a7ef9bd7 1495 error = xfs_qm_dqflush_all(mp);
4b8879df 1496
1da177e4
LT
1497 /*
1498 * We can get this error if we couldn't do a dquot allocation inside
1499 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1500 * dirty dquots that might be cached, we just want to get rid of them
1501 * and turn quotaoff. The dquots won't be attached to any of the inodes
1502 * at this point (because we intentionally didn't in dqget_noattach).
1503 */
1504 if (error) {
8112e9dc 1505 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1da177e4
LT
1506 goto error_return;
1507 }
1da177e4
LT
1508
1509 /*
1510 * We didn't log anything, because if we crashed, we'll have to
1511 * start the quotacheck from scratch anyway. However, we must make
1512 * sure that our dquot changes are secure before we put the
1513 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1514 * flush.
1515 */
a9add83e 1516 xfs_flush_buftarg(mp->m_ddev_targp, 1);
1da177e4
LT
1517
1518 /*
1519 * If one type of quotas is off, then it will lose its
1520 * quotachecked status, since we won't be doing accounting for
1521 * that type anymore.
1522 */
c8ad20ff 1523 mp->m_qflags &= ~(XFS_OQUOTA_CHKD | XFS_UQUOTA_CHKD);
1da177e4
LT
1524 mp->m_qflags |= flags;
1525
1da177e4
LT
1526 error_return:
1527 if (error) {
0b932ccc
DC
1528 xfs_warn(mp,
1529 "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1530 error);
1da177e4
LT
1531 /*
1532 * We must turn off quotas.
1533 */
1534 ASSERT(mp->m_quotainfo != NULL);
1535 ASSERT(xfs_Gqm != NULL);
1536 xfs_qm_destroy_quotainfo(mp);
31d5577b 1537 if (xfs_mount_reset_sbqflags(mp)) {
0b932ccc
DC
1538 xfs_warn(mp,
1539 "Quotacheck: Failed to reset quota flags.");
31d5577b 1540 }
0b932ccc
DC
1541 } else
1542 xfs_notice(mp, "Quotacheck: Done.");
1da177e4
LT
1543 return (error);
1544}
1545
1546/*
1547 * This is called after the superblock has been read in and we're ready to
1548 * iget the quota inodes.
1549 */
1550STATIC int
1551xfs_qm_init_quotainos(
1552 xfs_mount_t *mp)
1553{
1554 xfs_inode_t *uip, *gip;
1555 int error;
1556 __int64_t sbflags;
1557 uint flags;
1558
1559 ASSERT(mp->m_quotainfo);
1560 uip = gip = NULL;
1561 sbflags = 0;
1562 flags = 0;
1563
1564 /*
1565 * Get the uquota and gquota inodes
1566 */
62118709 1567 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1da177e4
LT
1568 if (XFS_IS_UQUOTA_ON(mp) &&
1569 mp->m_sb.sb_uquotino != NULLFSINO) {
1570 ASSERT(mp->m_sb.sb_uquotino > 0);
1571 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
7b6259e7 1572 0, 0, &uip)))
1da177e4
LT
1573 return XFS_ERROR(error);
1574 }
c8ad20ff 1575 if (XFS_IS_OQUOTA_ON(mp) &&
1da177e4
LT
1576 mp->m_sb.sb_gquotino != NULLFSINO) {
1577 ASSERT(mp->m_sb.sb_gquotino > 0);
1578 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
7b6259e7 1579 0, 0, &gip))) {
1da177e4 1580 if (uip)
43355099 1581 IRELE(uip);
1da177e4
LT
1582 return XFS_ERROR(error);
1583 }
1584 }
1585 } else {
1586 flags |= XFS_QMOPT_SBVERSION;
1587 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1588 XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
1589 }
1590
1591 /*
1592 * Create the two inodes, if they don't exist already. The changes
1593 * made above will get added to a transaction and logged in one of
1594 * the qino_alloc calls below. If the device is readonly,
1595 * temporarily switch to read-write to do this.
1596 */
1597 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1598 if ((error = xfs_qm_qino_alloc(mp, &uip,
1599 sbflags | XFS_SB_UQUOTINO,
1600 flags | XFS_QMOPT_UQUOTA)))
1601 return XFS_ERROR(error);
1602
1603 flags &= ~XFS_QMOPT_SBVERSION;
1604 }
c8ad20ff
NS
1605 if (XFS_IS_OQUOTA_ON(mp) && gip == NULL) {
1606 flags |= (XFS_IS_GQUOTA_ON(mp) ?
1607 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1608 error = xfs_qm_qino_alloc(mp, &gip,
1609 sbflags | XFS_SB_GQUOTINO, flags);
1610 if (error) {
1da177e4 1611 if (uip)
43355099 1612 IRELE(uip);
1da177e4
LT
1613
1614 return XFS_ERROR(error);
1615 }
1616 }
1617
8a7b8a89
CH
1618 mp->m_quotainfo->qi_uquotaip = uip;
1619 mp->m_quotainfo->qi_gquotaip = gip;
1da177e4 1620
014c2544 1621 return 0;
1da177e4
LT
1622}
1623
1624
368e1361 1625
1da177e4 1626/*
92678554 1627 * Pop the least recently used dquot off the freelist and recycle it.
1da177e4 1628 */
92678554 1629STATIC struct xfs_dquot *
368e1361 1630xfs_qm_dqreclaim_one(void)
1da177e4 1631{
92678554
CH
1632 struct xfs_dquot *dqp;
1633 int restarts = 0;
1da177e4 1634
3a8406f6 1635 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
92678554 1636restart:
3a8406f6 1637 list_for_each_entry(dqp, &xfs_Gqm->qm_dqfrlist, q_freelist) {
3a25404b 1638 struct xfs_mount *mp = dqp->q_mount;
bf72de31
CH
1639
1640 if (!xfs_dqlock_nowait(dqp))
1641 continue;
1da177e4
LT
1642
1643 /*
be7ffc38
CH
1644 * This dquot has already been grabbed by dqlookup.
1645 * Remove it from the freelist and try again.
1da177e4 1646 */
be7ffc38 1647 if (dqp->q_nrefs) {
368e1361 1648 trace_xfs_dqreclaim_want(dqp);
1da177e4 1649 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
be7ffc38
CH
1650
1651 list_del_init(&dqp->q_freelist);
1652 xfs_Gqm->qm_dqfrlist_cnt--;
0fbca4d1 1653 restarts++;
0fbca4d1 1654 goto dqunlock;
1da177e4
LT
1655 }
1656
368e1361 1657 ASSERT(dqp->q_hash);
3a25404b 1658 ASSERT(!list_empty(&dqp->q_mplist));
368e1361 1659
1da177e4 1660 /*
0fbca4d1
DC
1661 * Try to grab the flush lock. If this dquot is in the process
1662 * of getting flushed to disk, we don't want to reclaim it.
1da177e4 1663 */
0fbca4d1
DC
1664 if (!xfs_dqflock_nowait(dqp))
1665 goto dqunlock;
1da177e4
LT
1666
1667 /*
1668 * We have the flush lock so we know that this is not in the
1669 * process of being flushed. So, if this is dirty, flush it
1670 * DELWRI so that we don't get a freelist infested with
1671 * dirty dquots.
1672 */
1673 if (XFS_DQ_IS_DIRTY(dqp)) {
3c56836f 1674 int error;
0b1b213f 1675
368e1361 1676 trace_xfs_dqreclaim_dirty(dqp);
0b1b213f 1677
1da177e4
LT
1678 /*
1679 * We flush it delayed write, so don't bother
368e1361 1680 * releasing the freelist lock.
1da177e4 1681 */
fdedf28b 1682 error = xfs_qm_dqflush(dqp, SYNC_TRYLOCK);
3c56836f 1683 if (error) {
53487786
DC
1684 xfs_warn(mp, "%s: dquot %p flush failed",
1685 __func__, dqp);
3c56836f 1686 }
0fbca4d1 1687 goto dqunlock;
1da177e4 1688 }
92678554 1689 xfs_dqfunlock(dqp);
368e1361 1690
1da177e4 1691 /*
92678554
CH
1692 * Prevent lookup now that we are going to reclaim the dquot.
1693 * Once XFS_DQ_FREEING is set lookup won't touch the dquot,
1694 * thus we can drop the lock now.
1da177e4 1695 */
92678554
CH
1696 dqp->dq_flags |= XFS_DQ_FREEING;
1697 xfs_dqunlock(dqp);
368e1361 1698
92678554
CH
1699 mutex_lock(&dqp->q_hash->qh_lock);
1700 list_del_init(&dqp->q_hashlist);
1701 dqp->q_hash->qh_version++;
1702 mutex_unlock(&dqp->q_hash->qh_lock);
0b1b213f 1703
92678554 1704 mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
3a25404b
DC
1705 list_del_init(&dqp->q_mplist);
1706 mp->m_quotainfo->qi_dquots--;
1707 mp->m_quotainfo->qi_dqreclaims++;
92678554
CH
1708 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
1709
1710 ASSERT(dqp->q_nrefs == 0);
3a8406f6
DC
1711 list_del_init(&dqp->q_freelist);
1712 xfs_Gqm->qm_dqfrlist_cnt--;
92678554
CH
1713
1714 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1715 return dqp;
0fbca4d1 1716dqunlock:
1da177e4 1717 xfs_dqunlock(dqp);
368e1361 1718 if (restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
0fbca4d1 1719 break;
92678554 1720 goto restart;
1da177e4 1721 }
92678554 1722
3a8406f6 1723 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
92678554 1724 return NULL;
1da177e4
LT
1725}
1726
368e1361
DC
1727/*
1728 * Traverse the freelist of dquots and attempt to reclaim a maximum of
1729 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
1730 * favor the lookup function ...
1731 */
1732STATIC int
1733xfs_qm_shake_freelist(
1734 int howmany)
1735{
1736 int nreclaimed = 0;
1737 xfs_dquot_t *dqp;
1738
1739 if (howmany <= 0)
1740 return 0;
1741
1742 while (nreclaimed < howmany) {
1743 dqp = xfs_qm_dqreclaim_one();
1744 if (!dqp)
1745 return nreclaimed;
1746 xfs_qm_dqdestroy(dqp);
1747 nreclaimed++;
1748 }
1749 return nreclaimed;
1750}
1da177e4
LT
1751
1752/*
1753 * The kmem_shake interface is invoked when memory is running low.
1754 */
1755/* ARGSUSED */
1756STATIC int
7f8275d0
DC
1757xfs_qm_shake(
1758 struct shrinker *shrink,
1495f230 1759 struct shrink_control *sc)
1da177e4
LT
1760{
1761 int ndqused, nfree, n;
1495f230 1762 gfp_t gfp_mask = sc->gfp_mask;
1da177e4
LT
1763
1764 if (!kmem_shake_allow(gfp_mask))
014c2544 1765 return 0;
1da177e4 1766 if (!xfs_Gqm)
014c2544 1767 return 0;
1da177e4 1768
3a8406f6 1769 nfree = xfs_Gqm->qm_dqfrlist_cnt; /* free dquots */
1da177e4
LT
1770 /* incore dquots in all f/s's */
1771 ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;
1772
1773 ASSERT(ndqused >= 0);
1774
1775 if (nfree <= ndqused && nfree < ndquot)
014c2544 1776 return 0;
1da177e4
LT
1777
1778 ndqused *= xfs_Gqm->qm_dqfree_ratio; /* target # of free dquots */
1779 n = nfree - ndqused - ndquot; /* # over target */
1780
1781 return xfs_qm_shake_freelist(MAX(nfree, n));
1782}
1783
1784
1da177e4
LT
1785/*------------------------------------------------------------------*/
1786
1787/*
1788 * Return a new incore dquot. Depending on the number of
1789 * dquots in the system, we either allocate a new one on the kernel heap,
1790 * or reclaim a free one.
1791 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
1792 * to reclaim an existing one from the freelist.
1793 */
1794boolean_t
1795xfs_qm_dqalloc_incore(
1796 xfs_dquot_t **O_dqpp)
1797{
1798 xfs_dquot_t *dqp;
1799
1800 /*
1801 * Check against high water mark to see if we want to pop
1802 * a nincompoop dquot off the freelist.
1803 */
1804 if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
1805 /*
1806 * Try to recycle a dquot from the freelist.
1807 */
1808 if ((dqp = xfs_qm_dqreclaim_one())) {
1809 XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
1810 /*
1811 * Just zero the core here. The rest will get
1812 * reinitialized by caller. XXX we shouldn't even
1813 * do this zero ...
1814 */
1815 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
1816 *O_dqpp = dqp;
014c2544 1817 return B_FALSE;
1da177e4
LT
1818 }
1819 XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
1820 }
1821
1822 /*
1823 * Allocate a brand new dquot on the kernel heap and return it
1824 * to the caller to initialize.
1825 */
1826 ASSERT(xfs_Gqm->qm_dqzone != NULL);
1827 *O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
1828 atomic_inc(&xfs_Gqm->qm_totaldquots);
1829
014c2544 1830 return B_TRUE;
1da177e4
LT
1831}
1832
1833
1834/*
1835 * Start a transaction and write the incore superblock changes to
1836 * disk. flags parameter indicates which fields have changed.
1837 */
1838int
1839xfs_qm_write_sb_changes(
1840 xfs_mount_t *mp,
1841 __int64_t flags)
1842{
1843 xfs_trans_t *tp;
1844 int error;
1845
1da177e4
LT
1846 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1847 if ((error = xfs_trans_reserve(tp, 0,
1848 mp->m_sb.sb_sectsize + 128, 0,
1849 0,
1850 XFS_DEFAULT_LOG_COUNT))) {
1851 xfs_trans_cancel(tp, 0);
014c2544 1852 return error;
1da177e4
LT
1853 }
1854
1855 xfs_mod_sb(tp, flags);
e5720eec 1856 error = xfs_trans_commit(tp, 0);
1da177e4 1857
e5720eec 1858 return error;
1da177e4
LT
1859}
1860
1861
1862/* --------------- utility functions for vnodeops ---------------- */
1863
1864
1865/*
6c77b0ea 1866 * Given an inode, a uid, gid and prid make sure that we have
1da177e4
LT
1867 * allocated relevant dquot(s) on disk, and that we won't exceed inode
1868 * quotas by creating this file.
1869 * This also attaches dquot(s) to the given inode after locking it,
1870 * and returns the dquots corresponding to the uid and/or gid.
1871 *
1872 * in : inode (unlocked)
1873 * out : udquot, gdquot with references taken and unlocked
1874 */
1875int
1876xfs_qm_vop_dqalloc(
7d095257
CH
1877 struct xfs_inode *ip,
1878 uid_t uid,
1879 gid_t gid,
1880 prid_t prid,
1881 uint flags,
1882 struct xfs_dquot **O_udqpp,
1883 struct xfs_dquot **O_gdqpp)
1da177e4 1884{
7d095257
CH
1885 struct xfs_mount *mp = ip->i_mount;
1886 struct xfs_dquot *uq, *gq;
1887 int error;
1888 uint lockflags;
1da177e4 1889
7d095257 1890 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1da177e4
LT
1891 return 0;
1892
1893 lockflags = XFS_ILOCK_EXCL;
1894 xfs_ilock(ip, lockflags);
1895
bd186aa9 1896 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1da177e4
LT
1897 gid = ip->i_d.di_gid;
1898
1899 /*
1900 * Attach the dquot(s) to this inode, doing a dquot allocation
1901 * if necessary. The dquot(s) will not be locked.
1902 */
1903 if (XFS_NOT_DQATTACHED(mp, ip)) {
7d095257
CH
1904 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
1905 if (error) {
1da177e4 1906 xfs_iunlock(ip, lockflags);
014c2544 1907 return error;
1da177e4
LT
1908 }
1909 }
1910
1911 uq = gq = NULL;
c8ad20ff 1912 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1da177e4
LT
1913 if (ip->i_d.di_uid != uid) {
1914 /*
1915 * What we need is the dquot that has this uid, and
1916 * if we send the inode to dqget, the uid of the inode
1917 * takes priority over what's sent in the uid argument.
1918 * We must unlock inode here before calling dqget if
1919 * we're not sending the inode, because otherwise
1920 * we'll deadlock by doing trans_reserve while
1921 * holding ilock.
1922 */
1923 xfs_iunlock(ip, lockflags);
1924 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
1925 XFS_DQ_USER,
1926 XFS_QMOPT_DQALLOC |
1927 XFS_QMOPT_DOWARN,
1928 &uq))) {
1929 ASSERT(error != ENOENT);
014c2544 1930 return error;
1da177e4
LT
1931 }
1932 /*
1933 * Get the ilock in the right order.
1934 */
1935 xfs_dqunlock(uq);
1936 lockflags = XFS_ILOCK_SHARED;
1937 xfs_ilock(ip, lockflags);
1938 } else {
1939 /*
1940 * Take an extra reference, because we'll return
1941 * this to caller
1942 */
1943 ASSERT(ip->i_udquot);
1944 uq = ip->i_udquot;
1945 xfs_dqlock(uq);
1946 XFS_DQHOLD(uq);
1947 xfs_dqunlock(uq);
1948 }
1949 }
c8ad20ff 1950 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1da177e4
LT
1951 if (ip->i_d.di_gid != gid) {
1952 xfs_iunlock(ip, lockflags);
1953 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
1954 XFS_DQ_GROUP,
1955 XFS_QMOPT_DQALLOC |
1956 XFS_QMOPT_DOWARN,
1957 &gq))) {
1958 if (uq)
1959 xfs_qm_dqrele(uq);
1960 ASSERT(error != ENOENT);
014c2544 1961 return error;
1da177e4
LT
1962 }
1963 xfs_dqunlock(gq);
1964 lockflags = XFS_ILOCK_SHARED;
1965 xfs_ilock(ip, lockflags);
1966 } else {
1967 ASSERT(ip->i_gdquot);
1968 gq = ip->i_gdquot;
1969 xfs_dqlock(gq);
1970 XFS_DQHOLD(gq);
1971 xfs_dqunlock(gq);
1972 }
c8ad20ff 1973 } else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
6743099c 1974 if (xfs_get_projid(ip) != prid) {
c8ad20ff
NS
1975 xfs_iunlock(ip, lockflags);
1976 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
1977 XFS_DQ_PROJ,
1978 XFS_QMOPT_DQALLOC |
1979 XFS_QMOPT_DOWARN,
1980 &gq))) {
1981 if (uq)
1982 xfs_qm_dqrele(uq);
1983 ASSERT(error != ENOENT);
1984 return (error);
1985 }
1986 xfs_dqunlock(gq);
1987 lockflags = XFS_ILOCK_SHARED;
1988 xfs_ilock(ip, lockflags);
1989 } else {
1990 ASSERT(ip->i_gdquot);
1991 gq = ip->i_gdquot;
1992 xfs_dqlock(gq);
1993 XFS_DQHOLD(gq);
1994 xfs_dqunlock(gq);
1995 }
1da177e4
LT
1996 }
1997 if (uq)
0b1b213f 1998 trace_xfs_dquot_dqalloc(ip);
1da177e4
LT
1999
2000 xfs_iunlock(ip, lockflags);
2001 if (O_udqpp)
2002 *O_udqpp = uq;
2003 else if (uq)
2004 xfs_qm_dqrele(uq);
2005 if (O_gdqpp)
2006 *O_gdqpp = gq;
2007 else if (gq)
2008 xfs_qm_dqrele(gq);
014c2544 2009 return 0;
1da177e4
LT
2010}
2011
2012/*
2013 * Actually transfer ownership, and do dquot modifications.
2014 * These were already reserved.
2015 */
2016xfs_dquot_t *
2017xfs_qm_vop_chown(
2018 xfs_trans_t *tp,
2019 xfs_inode_t *ip,
2020 xfs_dquot_t **IO_olddq,
2021 xfs_dquot_t *newdq)
2022{
2023 xfs_dquot_t *prevdq;
06d10dd9
NS
2024 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
2025 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
2026
7d095257 2027
579aa9ca 2028 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4
LT
2029 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
2030
2031 /* old dquot */
2032 prevdq = *IO_olddq;
2033 ASSERT(prevdq);
2034 ASSERT(prevdq != newdq);
2035
06d10dd9
NS
2036 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
2037 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1da177e4
LT
2038
2039 /* the sparkling new dquot */
06d10dd9
NS
2040 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
2041 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1da177e4
LT
2042
2043 /*
2044 * Take an extra reference, because the inode
2045 * is going to keep this dquot pointer even
2046 * after the trans_commit.
2047 */
2048 xfs_dqlock(newdq);
2049 XFS_DQHOLD(newdq);
2050 xfs_dqunlock(newdq);
2051 *IO_olddq = newdq;
2052
014c2544 2053 return prevdq;
1da177e4
LT
2054}
2055
2056/*
c8ad20ff 2057 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1da177e4
LT
2058 */
2059int
2060xfs_qm_vop_chown_reserve(
2061 xfs_trans_t *tp,
2062 xfs_inode_t *ip,
2063 xfs_dquot_t *udqp,
2064 xfs_dquot_t *gdqp,
2065 uint flags)
2066{
7d095257 2067 xfs_mount_t *mp = ip->i_mount;
9a2a7de2 2068 uint delblks, blkflags, prjflags = 0;
1da177e4 2069 xfs_dquot_t *unresudq, *unresgdq, *delblksudq, *delblksgdq;
7d095257
CH
2070 int error;
2071
1da177e4 2072
579aa9ca 2073 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1da177e4
LT
2074 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2075
2076 delblks = ip->i_delayed_blks;
2077 delblksudq = delblksgdq = unresudq = unresgdq = NULL;
06d10dd9
NS
2078 blkflags = XFS_IS_REALTIME_INODE(ip) ?
2079 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
1da177e4
LT
2080
2081 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
1149d96a 2082 ip->i_d.di_uid != (uid_t)be32_to_cpu(udqp->q_core.d_id)) {
1da177e4
LT
2083 delblksudq = udqp;
2084 /*
2085 * If there are delayed allocation blocks, then we have to
2086 * unreserve those from the old dquot, and add them to the
2087 * new dquot.
2088 */
2089 if (delblks) {
2090 ASSERT(ip->i_udquot);
2091 unresudq = ip->i_udquot;
2092 }
2093 }
c8ad20ff 2094 if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
9a2a7de2 2095 if (XFS_IS_PQUOTA_ON(ip->i_mount) &&
6743099c 2096 xfs_get_projid(ip) != be32_to_cpu(gdqp->q_core.d_id))
9a2a7de2
NS
2097 prjflags = XFS_QMOPT_ENOSPC;
2098
2099 if (prjflags ||
2100 (XFS_IS_GQUOTA_ON(ip->i_mount) &&
2101 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id))) {
c8ad20ff
NS
2102 delblksgdq = gdqp;
2103 if (delblks) {
2104 ASSERT(ip->i_gdquot);
2105 unresgdq = ip->i_gdquot;
2106 }
1da177e4
LT
2107 }
2108 }
2109
2110 if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2111 delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
9a2a7de2 2112 flags | blkflags | prjflags)))
1da177e4
LT
2113 return (error);
2114
2115 /*
2116 * Do the delayed blks reservations/unreservations now. Since, these
2117 * are done without the help of a transaction, if a reservation fails
2118 * its previous reservations won't be automatically undone by trans
2119 * code. So, we have to do it manually here.
2120 */
2121 if (delblks) {
2122 /*
2123 * Do the reservations first. Unreservation can't fail.
2124 */
2125 ASSERT(delblksudq || delblksgdq);
2126 ASSERT(unresudq || unresgdq);
2127 if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2128 delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
9a2a7de2 2129 flags | blkflags | prjflags)))
1da177e4
LT
2130 return (error);
2131 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2132 unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
06d10dd9 2133 blkflags);
1da177e4
LT
2134 }
2135
2136 return (0);
2137}
2138
2139int
2140xfs_qm_vop_rename_dqattach(
7d095257 2141 struct xfs_inode **i_tab)
1da177e4 2142{
7d095257
CH
2143 struct xfs_mount *mp = i_tab[0]->i_mount;
2144 int i;
1da177e4 2145
7d095257 2146 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
014c2544 2147 return 0;
1da177e4 2148
7d095257
CH
2149 for (i = 0; (i < 4 && i_tab[i]); i++) {
2150 struct xfs_inode *ip = i_tab[i];
2151 int error;
2152
1da177e4
LT
2153 /*
2154 * Watch out for duplicate entries in the table.
2155 */
7d095257
CH
2156 if (i == 0 || ip != i_tab[i-1]) {
2157 if (XFS_NOT_DQATTACHED(mp, ip)) {
1da177e4
LT
2158 error = xfs_qm_dqattach(ip, 0);
2159 if (error)
014c2544 2160 return error;
1da177e4
LT
2161 }
2162 }
2163 }
014c2544 2164 return 0;
1da177e4
LT
2165}
2166
2167void
7d095257
CH
2168xfs_qm_vop_create_dqattach(
2169 struct xfs_trans *tp,
2170 struct xfs_inode *ip,
2171 struct xfs_dquot *udqp,
2172 struct xfs_dquot *gdqp)
1da177e4 2173{
7d095257
CH
2174 struct xfs_mount *mp = tp->t_mountp;
2175
2176 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1da177e4
LT
2177 return;
2178
579aa9ca 2179 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
7d095257 2180 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1da177e4
LT
2181
2182 if (udqp) {
2183 xfs_dqlock(udqp);
2184 XFS_DQHOLD(udqp);
2185 xfs_dqunlock(udqp);
2186 ASSERT(ip->i_udquot == NULL);
2187 ip->i_udquot = udqp;
7d095257 2188 ASSERT(XFS_IS_UQUOTA_ON(mp));
1149d96a 2189 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
1da177e4
LT
2190 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2191 }
2192 if (gdqp) {
2193 xfs_dqlock(gdqp);
2194 XFS_DQHOLD(gdqp);
2195 xfs_dqunlock(gdqp);
2196 ASSERT(ip->i_gdquot == NULL);
2197 ip->i_gdquot = gdqp;
7d095257
CH
2198 ASSERT(XFS_IS_OQUOTA_ON(mp));
2199 ASSERT((XFS_IS_GQUOTA_ON(mp) ?
6743099c 2200 ip->i_d.di_gid : xfs_get_projid(ip)) ==
ee2a4f7c 2201 be32_to_cpu(gdqp->q_core.d_id));
1da177e4
LT
2202 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
2203 }
2204}
2205