]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/xfs/xfs_qm.c
xfs: kill xfs_qm_idtodq
[mirror_ubuntu-artful-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) {
8e9b6e7f 608 ASSERT(*IO_idqpp == NULL);
8e9b6e7f 609
78e55892 610 *IO_idqpp = xfs_qm_dqhold(dqp);
1da177e4 611 xfs_dqunlock(udqhint);
8e9b6e7f 612 return 0;
1da177e4 613 }
8e9b6e7f
CH
614
615 /*
616 * We can't hold a dquot lock when we call the dqget code.
617 * We'll deadlock in no time, because of (not conforming to)
618 * lock ordering - the inodelock comes before any dquot lock,
619 * and we may drop and reacquire the ilock in xfs_qm_dqget().
620 */
1da177e4 621 xfs_dqunlock(udqhint);
8e9b6e7f
CH
622 }
623
1da177e4
LT
624 /*
625 * Find the dquot from somewhere. This bumps the
626 * reference count of dquot and returns it locked.
627 * This can return ENOENT if dquot didn't exist on
628 * disk and we didn't ask it to allocate;
629 * ESRCH if quotas got turned off suddenly.
630 */
db3e74b5
MH
631 error = xfs_qm_dqget(ip->i_mount, ip, id, type,
632 doalloc | XFS_QMOPT_DOWARN, &dqp);
8e9b6e7f
CH
633 if (error)
634 return error;
1da177e4 635
0b1b213f 636 trace_xfs_dqattach_get(dqp);
8e9b6e7f 637
1da177e4
LT
638 /*
639 * dqget may have dropped and re-acquired the ilock, but it guarantees
640 * that the dquot returned is the one that should go in the inode.
641 */
642 *IO_idqpp = dqp;
8e9b6e7f
CH
643 xfs_dqunlock(dqp);
644 return 0;
1da177e4
LT
645}
646
647
648/*
649 * Given a udquot and gdquot, attach a ptr to the group dquot in the
ab680bb7 650 * udquot as a hint for future lookups.
1da177e4
LT
651 */
652STATIC void
653xfs_qm_dqattach_grouphint(
654 xfs_dquot_t *udq,
8e9b6e7f 655 xfs_dquot_t *gdq)
1da177e4
LT
656{
657 xfs_dquot_t *tmp;
658
8e9b6e7f 659 xfs_dqlock(udq);
1da177e4 660
ab680bb7
CH
661 tmp = udq->q_gdquot;
662 if (tmp) {
663 if (tmp == gdq)
664 goto done;
1da177e4
LT
665
666 udq->q_gdquot = NULL;
1da177e4 667 xfs_qm_dqrele(tmp);
1da177e4 668 }
8e9b6e7f 669
78e55892 670 udq->q_gdquot = xfs_qm_dqhold(gdq);
ab680bb7 671done:
8e9b6e7f 672 xfs_dqunlock(udq);
1da177e4
LT
673}
674
675
676/*
c8ad20ff
NS
677 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
678 * into account.
1da177e4 679 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
1da177e4
LT
680 * Inode may get unlocked and relocked in here, and the caller must deal with
681 * the consequences.
682 */
683int
7d095257 684xfs_qm_dqattach_locked(
1da177e4
LT
685 xfs_inode_t *ip,
686 uint flags)
687{
688 xfs_mount_t *mp = ip->i_mount;
689 uint nquotas = 0;
690 int error = 0;
691
7d095257
CH
692 if (!XFS_IS_QUOTA_RUNNING(mp) ||
693 !XFS_IS_QUOTA_ON(mp) ||
694 !XFS_NOT_DQATTACHED(mp, ip) ||
695 ip->i_ino == mp->m_sb.sb_uquotino ||
696 ip->i_ino == mp->m_sb.sb_gquotino)
014c2544 697 return 0;
1da177e4 698
7d095257 699 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4
LT
700
701 if (XFS_IS_UQUOTA_ON(mp)) {
702 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
703 flags & XFS_QMOPT_DQALLOC,
1da177e4
LT
704 NULL, &ip->i_udquot);
705 if (error)
706 goto done;
707 nquotas++;
708 }
579aa9ca
CH
709
710 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
c8ad20ff
NS
711 if (XFS_IS_OQUOTA_ON(mp)) {
712 error = XFS_IS_GQUOTA_ON(mp) ?
713 xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
714 flags & XFS_QMOPT_DQALLOC,
c8ad20ff 715 ip->i_udquot, &ip->i_gdquot) :
6743099c 716 xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
1da177e4 717 flags & XFS_QMOPT_DQALLOC,
1da177e4
LT
718 ip->i_udquot, &ip->i_gdquot);
719 /*
720 * Don't worry about the udquot that we may have
721 * attached above. It'll get detached, if not already.
722 */
723 if (error)
724 goto done;
725 nquotas++;
726 }
727
728 /*
729 * Attach this group quota to the user quota as a hint.
730 * This WON'T, in general, result in a thrash.
731 */
732 if (nquotas == 2) {
579aa9ca 733 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4
LT
734 ASSERT(ip->i_udquot);
735 ASSERT(ip->i_gdquot);
736
737 /*
ab680bb7
CH
738 * We do not have i_udquot locked at this point, but this check
739 * is OK since we don't depend on the i_gdquot to be accurate
740 * 100% all the time. It is just a hint, and this will
741 * succeed in general.
1da177e4 742 */
ab680bb7
CH
743 if (ip->i_udquot->q_gdquot != ip->i_gdquot)
744 xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot);
1da177e4
LT
745 }
746
7d095257 747 done:
ea15ab3c
CH
748#ifdef DEBUG
749 if (!error) {
1da177e4
LT
750 if (XFS_IS_UQUOTA_ON(mp))
751 ASSERT(ip->i_udquot);
c8ad20ff 752 if (XFS_IS_OQUOTA_ON(mp))
1da177e4
LT
753 ASSERT(ip->i_gdquot);
754 }
7d095257 755 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4 756#endif
7d095257
CH
757 return error;
758}
1da177e4 759
7d095257
CH
760int
761xfs_qm_dqattach(
762 struct xfs_inode *ip,
763 uint flags)
764{
765 int error;
766
767 xfs_ilock(ip, XFS_ILOCK_EXCL);
768 error = xfs_qm_dqattach_locked(ip, flags);
769 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1da177e4 770
014c2544 771 return error;
1da177e4
LT
772}
773
774/*
775 * Release dquots (and their references) if any.
776 * The inode should be locked EXCL except when this's called by
777 * xfs_ireclaim.
778 */
779void
780xfs_qm_dqdetach(
781 xfs_inode_t *ip)
782{
783 if (!(ip->i_udquot || ip->i_gdquot))
784 return;
785
0b1b213f
CH
786 trace_xfs_dquot_dqdetach(ip);
787
1da177e4
LT
788 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
789 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
1da177e4
LT
790 if (ip->i_udquot) {
791 xfs_qm_dqrele(ip->i_udquot);
792 ip->i_udquot = NULL;
793 }
794 if (ip->i_gdquot) {
795 xfs_qm_dqrele(ip->i_gdquot);
796 ip->i_gdquot = NULL;
797 }
798}
799
a4edd1da
CH
800/*
801 * The hash chains and the mplist use the same xfs_dqhash structure as
802 * their list head, but we can take the mplist qh_lock and one of the
803 * hash qh_locks at the same time without any problem as they aren't
804 * related.
805 */
806static struct lock_class_key xfs_quota_mplist_class;
1da177e4
LT
807
808/*
809 * This initializes all the quota information that's kept in the
810 * mount structure
811 */
ba0f32d4 812STATIC int
1da177e4
LT
813xfs_qm_init_quotainfo(
814 xfs_mount_t *mp)
815{
816 xfs_quotainfo_t *qinf;
817 int error;
818 xfs_dquot_t *dqp;
819
820 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
821
822 /*
823 * Tell XQM that we exist as soon as possible.
824 */
825 if ((error = xfs_qm_hold_quotafs_ref(mp))) {
014c2544 826 return error;
1da177e4
LT
827 }
828
829 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
830
831 /*
832 * See if quotainodes are setup, and if not, allocate them,
833 * and change the superblock accordingly.
834 */
835 if ((error = xfs_qm_init_quotainos(mp))) {
f0e2d93c 836 kmem_free(qinf);
1da177e4 837 mp->m_quotainfo = NULL;
014c2544 838 return error;
1da177e4
LT
839 }
840
3a25404b
DC
841 INIT_LIST_HEAD(&qinf->qi_dqlist);
842 mutex_init(&qinf->qi_dqlist_lock);
843 lockdep_set_class(&qinf->qi_dqlist_lock, &xfs_quota_mplist_class);
a4edd1da 844
1da177e4
LT
845 qinf->qi_dqreclaims = 0;
846
847 /* mutex used to serialize quotaoffs */
794ee1ba 848 mutex_init(&qinf->qi_quotaofflock);
1da177e4
LT
849
850 /* Precalc some constants */
851 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
852 ASSERT(qinf->qi_dqchunklen);
853 qinf->qi_dqperchunk = BBTOB(qinf->qi_dqchunklen);
854 do_div(qinf->qi_dqperchunk, sizeof(xfs_dqblk_t));
855
856 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
857
858 /*
859 * We try to get the limits from the superuser's limits fields.
860 * This is quite hacky, but it is standard quota practice.
861 * We look at the USR dquot with id == 0 first, but if user quotas
862 * are not enabled we goto the GRP dquot with id == 0.
863 * We don't really care to keep separate default limits for user
864 * and group quotas, at least not at this point.
865 */
866 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)0,
c8ad20ff
NS
867 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
868 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
869 XFS_DQ_PROJ),
1da177e4
LT
870 XFS_QMOPT_DQSUSER|XFS_QMOPT_DOWARN,
871 &dqp);
872 if (! error) {
873 xfs_disk_dquot_t *ddqp = &dqp->q_core;
874
875 /*
876 * The warnings and timers set the grace period given to
877 * a user or group before he or she can not perform any
878 * more writing. If it is zero, a default is used.
879 */
1149d96a
CH
880 qinf->qi_btimelimit = ddqp->d_btimer ?
881 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
882 qinf->qi_itimelimit = ddqp->d_itimer ?
883 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
884 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
885 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
886 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
887 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
888 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
889 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
890 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
891 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
892 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
893 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
894 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
895 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
896 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
897 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
1da177e4
LT
898
899 /*
900 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
901 * we don't want this dquot cached. We haven't done a
902 * quotacheck yet, and quotacheck doesn't like incore dquots.
903 */
904 xfs_qm_dqdestroy(dqp);
905 } else {
906 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
907 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
908 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
909 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
910 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
06d10dd9 911 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
1da177e4
LT
912 }
913
014c2544 914 return 0;
1da177e4
LT
915}
916
917
918/*
919 * Gets called when unmounting a filesystem or when all quotas get
920 * turned off.
921 * This purges the quota inodes, destroys locks and frees itself.
922 */
923void
924xfs_qm_destroy_quotainfo(
925 xfs_mount_t *mp)
926{
927 xfs_quotainfo_t *qi;
928
929 qi = mp->m_quotainfo;
930 ASSERT(qi != NULL);
931 ASSERT(xfs_Gqm != NULL);
932
933 /*
934 * Release the reference that XQM kept, so that we know
935 * when the XQM structure should be freed. We cannot assume
936 * that xfs_Gqm is non-null after this point.
937 */
938 xfs_qm_rele_quotafs_ref(mp);
939
3a25404b
DC
940 ASSERT(list_empty(&qi->qi_dqlist));
941 mutex_destroy(&qi->qi_dqlist_lock);
1da177e4
LT
942
943 if (qi->qi_uquotaip) {
26cc0021 944 IRELE(qi->qi_uquotaip);
1da177e4
LT
945 qi->qi_uquotaip = NULL; /* paranoia */
946 }
947 if (qi->qi_gquotaip) {
26cc0021 948 IRELE(qi->qi_gquotaip);
1da177e4
LT
949 qi->qi_gquotaip = NULL;
950 }
951 mutex_destroy(&qi->qi_quotaofflock);
f0e2d93c 952 kmem_free(qi);
1da177e4
LT
953 mp->m_quotainfo = NULL;
954}
955
956
957
958/* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
959
960/* ARGSUSED */
961STATIC void
962xfs_qm_list_init(
963 xfs_dqlist_t *list,
964 char *str,
965 int n)
966{
794ee1ba 967 mutex_init(&list->qh_lock);
e6a81f13 968 INIT_LIST_HEAD(&list->qh_list);
1da177e4
LT
969 list->qh_version = 0;
970 list->qh_nelems = 0;
971}
972
973STATIC void
974xfs_qm_list_destroy(
975 xfs_dqlist_t *list)
976{
977 mutex_destroy(&(list->qh_lock));
978}
979
1da177e4
LT
980/*
981 * Create an inode and return with a reference already taken, but unlocked
982 * This is how we create quota inodes
983 */
984STATIC int
985xfs_qm_qino_alloc(
986 xfs_mount_t *mp,
987 xfs_inode_t **ip,
988 __int64_t sbfields,
989 uint flags)
990{
991 xfs_trans_t *tp;
992 int error;
1da177e4
LT
993 int committed;
994
061f7209 995 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
1da177e4
LT
996 if ((error = xfs_trans_reserve(tp,
997 XFS_QM_QINOCREATE_SPACE_RES(mp),
998 XFS_CREATE_LOG_RES(mp), 0,
999 XFS_TRANS_PERM_LOG_RES,
1000 XFS_CREATE_LOG_COUNT))) {
1001 xfs_trans_cancel(tp, 0);
014c2544 1002 return error;
1da177e4 1003 }
1da177e4 1004
6c77b0ea
CH
1005 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip, &committed);
1006 if (error) {
1da177e4
LT
1007 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1008 XFS_TRANS_ABORT);
014c2544 1009 return error;
1da177e4
LT
1010 }
1011
1da177e4
LT
1012 /*
1013 * Make the changes in the superblock, and log those too.
1014 * sbfields arg may contain fields other than *QUOTINO;
1015 * VERSIONNUM for example.
1016 */
3685c2a1 1017 spin_lock(&mp->m_sb_lock);
1da177e4 1018 if (flags & XFS_QMOPT_SBVERSION) {
62118709 1019 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1da177e4
LT
1020 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1021 XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
1022 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1023 XFS_SB_GQUOTINO | XFS_SB_QFLAGS));
1024
62118709 1025 xfs_sb_version_addquota(&mp->m_sb);
1da177e4
LT
1026 mp->m_sb.sb_uquotino = NULLFSINO;
1027 mp->m_sb.sb_gquotino = NULLFSINO;
1028
1029 /* qflags will get updated _after_ quotacheck */
1030 mp->m_sb.sb_qflags = 0;
1da177e4
LT
1031 }
1032 if (flags & XFS_QMOPT_UQUOTA)
1033 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1034 else
1035 mp->m_sb.sb_gquotino = (*ip)->i_ino;
3685c2a1 1036 spin_unlock(&mp->m_sb_lock);
1da177e4
LT
1037 xfs_mod_sb(tp, sbfields);
1038
1c72bf90 1039 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
53487786 1040 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
014c2544 1041 return error;
1da177e4 1042 }
014c2544 1043 return 0;
1da177e4
LT
1044}
1045
1046
5b139738 1047STATIC void
1da177e4
LT
1048xfs_qm_reset_dqcounts(
1049 xfs_mount_t *mp,
1050 xfs_buf_t *bp,
1051 xfs_dqid_t id,
1052 uint type)
1053{
1054 xfs_disk_dquot_t *ddq;
1055 int j;
1056
0b1b213f
CH
1057 trace_xfs_reset_dqcounts(bp, _RET_IP_);
1058
1da177e4
LT
1059 /*
1060 * Reset all counters and timers. They'll be
1061 * started afresh by xfs_qm_quotacheck.
1062 */
1063#ifdef DEBUG
1064 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1065 do_div(j, sizeof(xfs_dqblk_t));
8a7b8a89 1066 ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
1da177e4 1067#endif
62926044 1068 ddq = bp->b_addr;
8a7b8a89 1069 for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
1da177e4
LT
1070 /*
1071 * Do a sanity check, and if needed, repair the dqblk. Don't
1072 * output any warnings because it's perfectly possible to
c41564b5 1073 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1da177e4 1074 */
a0fa2b67 1075 (void) xfs_qm_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1da177e4 1076 "xfs_quotacheck");
1149d96a
CH
1077 ddq->d_bcount = 0;
1078 ddq->d_icount = 0;
1079 ddq->d_rtbcount = 0;
1080 ddq->d_btimer = 0;
1081 ddq->d_itimer = 0;
1082 ddq->d_rtbtimer = 0;
1083 ddq->d_bwarns = 0;
1084 ddq->d_iwarns = 0;
1085 ddq->d_rtbwarns = 0;
1da177e4
LT
1086 ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
1087 }
1da177e4
LT
1088}
1089
1090STATIC int
1091xfs_qm_dqiter_bufs(
1092 xfs_mount_t *mp,
1093 xfs_dqid_t firstid,
1094 xfs_fsblock_t bno,
1095 xfs_filblks_t blkcnt,
1096 uint flags)
1097{
1098 xfs_buf_t *bp;
1099 int error;
c8ad20ff 1100 int type;
1da177e4
LT
1101
1102 ASSERT(blkcnt > 0);
c8ad20ff
NS
1103 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1104 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1da177e4
LT
1105 error = 0;
1106
1107 /*
1108 * Blkcnt arg can be a very big number, and might even be
1109 * larger than the log itself. So, we have to break it up into
1110 * manageable-sized transactions.
1111 * Note that we don't start a permanent transaction here; we might
1112 * not be able to get a log reservation for the whole thing up front,
1113 * and we don't really care to either, because we just discard
1114 * everything if we were to crash in the middle of this loop.
1115 */
1116 while (blkcnt--) {
1117 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1118 XFS_FSB_TO_DADDR(mp, bno),
8a7b8a89 1119 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
1da177e4
LT
1120 if (error)
1121 break;
1122
5b139738 1123 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
61551f1e
CH
1124 xfs_buf_delwri_queue(bp);
1125 xfs_buf_relse(bp);
1da177e4
LT
1126 /*
1127 * goto the next block.
1128 */
1129 bno++;
8a7b8a89 1130 firstid += mp->m_quotainfo->qi_dqperchunk;
1da177e4 1131 }
014c2544 1132 return error;
1da177e4
LT
1133}
1134
1135/*
c8ad20ff 1136 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1da177e4
LT
1137 * caller supplied function for every chunk of dquots that we find.
1138 */
1139STATIC int
1140xfs_qm_dqiterate(
1141 xfs_mount_t *mp,
1142 xfs_inode_t *qip,
1143 uint flags)
1144{
1145 xfs_bmbt_irec_t *map;
1146 int i, nmaps; /* number of map entries */
1147 int error; /* return value */
1148 xfs_fileoff_t lblkno;
1149 xfs_filblks_t maxlblkcnt;
1150 xfs_dqid_t firstid;
1151 xfs_fsblock_t rablkno;
1152 xfs_filblks_t rablkcnt;
1153
1154 error = 0;
1155 /*
c41564b5 1156 * This looks racy, but we can't keep an inode lock across a
1da177e4
LT
1157 * trans_reserve. But, this gets called during quotacheck, and that
1158 * happens only at mount time which is single threaded.
1159 */
1160 if (qip->i_d.di_nblocks == 0)
014c2544 1161 return 0;
1da177e4
LT
1162
1163 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1164
1165 lblkno = 0;
1166 maxlblkcnt = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1167 do {
1168 nmaps = XFS_DQITER_MAP_SIZE;
1169 /*
1170 * We aren't changing the inode itself. Just changing
1171 * some of its data. No new blocks are added here, and
1172 * the inode is never added to the transaction.
1173 */
1174 xfs_ilock(qip, XFS_ILOCK_SHARED);
5c8ed202
DC
1175 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
1176 map, &nmaps, 0);
1da177e4
LT
1177 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1178 if (error)
1179 break;
1180
1181 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1182 for (i = 0; i < nmaps; i++) {
1183 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1184 ASSERT(map[i].br_blockcount);
1185
1186
1187 lblkno += map[i].br_blockcount;
1188
1189 if (map[i].br_startblock == HOLESTARTBLOCK)
1190 continue;
1191
1192 firstid = (xfs_dqid_t) map[i].br_startoff *
8a7b8a89 1193 mp->m_quotainfo->qi_dqperchunk;
1da177e4
LT
1194 /*
1195 * Do a read-ahead on the next extent.
1196 */
1197 if ((i+1 < nmaps) &&
1198 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1199 rablkcnt = map[i+1].br_blockcount;
1200 rablkno = map[i+1].br_startblock;
1201 while (rablkcnt--) {
1a1a3e97 1202 xfs_buf_readahead(mp->m_ddev_targp,
1da177e4 1203 XFS_FSB_TO_DADDR(mp, rablkno),
8a7b8a89 1204 mp->m_quotainfo->qi_dqchunklen);
1da177e4
LT
1205 rablkno++;
1206 }
1207 }
1208 /*
1209 * Iterate thru all the blks in the extent and
1210 * reset the counters of all the dquots inside them.
1211 */
1212 if ((error = xfs_qm_dqiter_bufs(mp,
1213 firstid,
1214 map[i].br_startblock,
1215 map[i].br_blockcount,
1216 flags))) {
1217 break;
1218 }
1219 }
1220
1221 if (error)
1222 break;
1223 } while (nmaps > 0);
1224
f0e2d93c 1225 kmem_free(map);
1da177e4 1226
014c2544 1227 return error;
1da177e4
LT
1228}
1229
1230/*
1231 * Called by dqusage_adjust in doing a quotacheck.
52fda114
CH
1232 *
1233 * Given the inode, and a dquot id this updates both the incore dqout as well
1234 * as the buffer copy. This is so that once the quotacheck is done, we can
1235 * just log all the buffers, as opposed to logging numerous updates to
1236 * individual dquots.
1da177e4 1237 */
52fda114 1238STATIC int
1da177e4 1239xfs_qm_quotacheck_dqadjust(
52fda114
CH
1240 struct xfs_inode *ip,
1241 xfs_dqid_t id,
1242 uint type,
1da177e4
LT
1243 xfs_qcnt_t nblks,
1244 xfs_qcnt_t rtblks)
1245{
52fda114
CH
1246 struct xfs_mount *mp = ip->i_mount;
1247 struct xfs_dquot *dqp;
1248 int error;
1249
1250 error = xfs_qm_dqget(mp, ip, id, type,
1251 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1252 if (error) {
1253 /*
1254 * Shouldn't be able to turn off quotas here.
1255 */
1256 ASSERT(error != ESRCH);
1257 ASSERT(error != ENOENT);
1258 return error;
1259 }
0b1b213f
CH
1260
1261 trace_xfs_dqadjust(dqp);
1262
1da177e4
LT
1263 /*
1264 * Adjust the inode count and the block count to reflect this inode's
1265 * resource usage.
1266 */
413d57c9 1267 be64_add_cpu(&dqp->q_core.d_icount, 1);
1da177e4
LT
1268 dqp->q_res_icount++;
1269 if (nblks) {
413d57c9 1270 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1da177e4
LT
1271 dqp->q_res_bcount += nblks;
1272 }
1273 if (rtblks) {
413d57c9 1274 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1da177e4
LT
1275 dqp->q_res_rtbcount += rtblks;
1276 }
1277
1278 /*
1279 * Set default limits, adjust timers (since we changed usages)
191f8488
CH
1280 *
1281 * There are no timers for the default values set in the root dquot.
1da177e4 1282 */
191f8488 1283 if (dqp->q_core.d_id) {
52fda114
CH
1284 xfs_qm_adjust_dqlimits(mp, &dqp->q_core);
1285 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1da177e4
LT
1286 }
1287
1288 dqp->dq_flags |= XFS_DQ_DIRTY;
52fda114
CH
1289 xfs_qm_dqput(dqp);
1290 return 0;
1da177e4
LT
1291}
1292
1293STATIC int
1294xfs_qm_get_rtblks(
1295 xfs_inode_t *ip,
1296 xfs_qcnt_t *O_rtblks)
1297{
1298 xfs_filblks_t rtblks; /* total rt blks */
4eea22f0 1299 xfs_extnum_t idx; /* extent record index */
1da177e4
LT
1300 xfs_ifork_t *ifp; /* inode fork pointer */
1301 xfs_extnum_t nextents; /* number of extent entries */
1da177e4
LT
1302 int error;
1303
1304 ASSERT(XFS_IS_REALTIME_INODE(ip));
1305 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1306 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1307 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
014c2544 1308 return error;
1da177e4
LT
1309 }
1310 rtblks = 0;
4eea22f0 1311 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
a6f64d4a
CH
1312 for (idx = 0; idx < nextents; idx++)
1313 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1da177e4 1314 *O_rtblks = (xfs_qcnt_t)rtblks;
014c2544 1315 return 0;
1da177e4
LT
1316}
1317
1318/*
1319 * callback routine supplied to bulkstat(). Given an inumber, find its
1320 * dquots and update them to account for resources taken by that inode.
1321 */
1322/* ARGSUSED */
1323STATIC int
1324xfs_qm_dqusage_adjust(
1325 xfs_mount_t *mp, /* mount point for filesystem */
1326 xfs_ino_t ino, /* inode number to get data for */
1327 void __user *buffer, /* not used */
1328 int ubsize, /* not used */
1da177e4 1329 int *ubused, /* not used */
1da177e4
LT
1330 int *res) /* result code value */
1331{
1332 xfs_inode_t *ip;
52fda114 1333 xfs_qcnt_t nblks, rtblks = 0;
1da177e4
LT
1334 int error;
1335
1336 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1337
1338 /*
1339 * rootino must have its resources accounted for, not so with the quota
1340 * inodes.
1341 */
1342 if (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino) {
1343 *res = BULKSTAT_RV_NOTHING;
1344 return XFS_ERROR(EINVAL);
1345 }
1346
1347 /*
1348 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1349 * interface expects the inode to be exclusively locked because that's
1350 * the case in all other instances. It's OK that we do this because
1351 * quotacheck is done only at mount time.
1352 */
52fda114
CH
1353 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1354 if (error) {
1da177e4 1355 *res = BULKSTAT_RV_NOTHING;
014c2544 1356 return error;
1da177e4
LT
1357 }
1358
52fda114 1359 ASSERT(ip->i_delayed_blks == 0);
1da177e4 1360
52fda114 1361 if (XFS_IS_REALTIME_INODE(ip)) {
1da177e4
LT
1362 /*
1363 * Walk thru the extent list and count the realtime blocks.
1364 */
52fda114
CH
1365 error = xfs_qm_get_rtblks(ip, &rtblks);
1366 if (error)
1367 goto error0;
1da177e4 1368 }
1da177e4 1369
52fda114 1370 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1da177e4
LT
1371
1372 /*
1373 * Add the (disk blocks and inode) resources occupied by this
1374 * inode to its dquots. We do this adjustment in the incore dquot,
1375 * and also copy the changes to its buffer.
1376 * We don't care about putting these changes in a transaction
1377 * envelope because if we crash in the middle of a 'quotacheck'
1378 * we have to start from the beginning anyway.
1379 * Once we're done, we'll log all the dquot bufs.
1380 *
c41564b5 1381 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1da177e4
LT
1382 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1383 */
1384 if (XFS_IS_UQUOTA_ON(mp)) {
52fda114
CH
1385 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1386 XFS_DQ_USER, nblks, rtblks);
1387 if (error)
1388 goto error0;
1da177e4 1389 }
52fda114
CH
1390
1391 if (XFS_IS_GQUOTA_ON(mp)) {
1392 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1393 XFS_DQ_GROUP, nblks, rtblks);
1394 if (error)
1395 goto error0;
1da177e4 1396 }
1da177e4 1397
52fda114 1398 if (XFS_IS_PQUOTA_ON(mp)) {
6743099c 1399 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
52fda114
CH
1400 XFS_DQ_PROJ, nblks, rtblks);
1401 if (error)
1402 goto error0;
1403 }
1404
1405 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1406 IRELE(ip);
1da177e4 1407 *res = BULKSTAT_RV_DIDONE;
014c2544 1408 return 0;
52fda114
CH
1409
1410error0:
1411 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1412 IRELE(ip);
1413 *res = BULKSTAT_RV_GIVEUP;
1414 return error;
1da177e4
LT
1415}
1416
1417/*
1418 * Walk thru all the filesystem inodes and construct a consistent view
1419 * of the disk quota world. If the quotacheck fails, disable quotas.
1420 */
1421int
1422xfs_qm_quotacheck(
1423 xfs_mount_t *mp)
1424{
1425 int done, count, error;
1426 xfs_ino_t lastino;
1427 size_t structsz;
1428 xfs_inode_t *uip, *gip;
1429 uint flags;
1430
1431 count = INT_MAX;
1432 structsz = 1;
1433 lastino = 0;
1434 flags = 0;
1435
8a7b8a89 1436 ASSERT(mp->m_quotainfo->qi_uquotaip || mp->m_quotainfo->qi_gquotaip);
1da177e4
LT
1437 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1438
1439 /*
1440 * There should be no cached dquots. The (simplistic) quotacheck
1441 * algorithm doesn't like that.
1442 */
3a25404b 1443 ASSERT(list_empty(&mp->m_quotainfo->qi_dqlist));
1da177e4 1444
0b932ccc 1445 xfs_notice(mp, "Quotacheck needed: Please wait.");
1da177e4
LT
1446
1447 /*
c8ad20ff 1448 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1da177e4
LT
1449 * their counters to zero. We need a clean slate.
1450 * We don't log our changes till later.
1451 */
8a7b8a89
CH
1452 uip = mp->m_quotainfo->qi_uquotaip;
1453 if (uip) {
1454 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA);
1455 if (error)
1da177e4
LT
1456 goto error_return;
1457 flags |= XFS_UQUOTA_CHKD;
1458 }
1459
8a7b8a89
CH
1460 gip = mp->m_quotainfo->qi_gquotaip;
1461 if (gip) {
1462 error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
1463 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1464 if (error)
1da177e4 1465 goto error_return;
c8ad20ff 1466 flags |= XFS_OQUOTA_CHKD;
1da177e4
LT
1467 }
1468
1469 do {
1470 /*
1471 * Iterate thru all the inodes in the file system,
1472 * adjusting the corresponding dquot counters in core.
1473 */
7dce11db
CH
1474 error = xfs_bulkstat(mp, &lastino, &count,
1475 xfs_qm_dqusage_adjust,
1476 structsz, NULL, &done);
1477 if (error)
1da177e4
LT
1478 break;
1479
7dce11db 1480 } while (!done);
1da177e4 1481
4b8879df
DC
1482 /*
1483 * We've made all the changes that we need to make incore.
1484 * Flush them down to disk buffers if everything was updated
1485 * successfully.
1486 */
1487 if (!error)
a7ef9bd7 1488 error = xfs_qm_dqflush_all(mp);
4b8879df 1489
1da177e4
LT
1490 /*
1491 * We can get this error if we couldn't do a dquot allocation inside
1492 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1493 * dirty dquots that might be cached, we just want to get rid of them
1494 * and turn quotaoff. The dquots won't be attached to any of the inodes
1495 * at this point (because we intentionally didn't in dqget_noattach).
1496 */
1497 if (error) {
8112e9dc 1498 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1da177e4
LT
1499 goto error_return;
1500 }
1da177e4
LT
1501
1502 /*
1503 * We didn't log anything, because if we crashed, we'll have to
1504 * start the quotacheck from scratch anyway. However, we must make
1505 * sure that our dquot changes are secure before we put the
1506 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1507 * flush.
1508 */
a9add83e 1509 xfs_flush_buftarg(mp->m_ddev_targp, 1);
1da177e4
LT
1510
1511 /*
1512 * If one type of quotas is off, then it will lose its
1513 * quotachecked status, since we won't be doing accounting for
1514 * that type anymore.
1515 */
c8ad20ff 1516 mp->m_qflags &= ~(XFS_OQUOTA_CHKD | XFS_UQUOTA_CHKD);
1da177e4
LT
1517 mp->m_qflags |= flags;
1518
1da177e4
LT
1519 error_return:
1520 if (error) {
0b932ccc
DC
1521 xfs_warn(mp,
1522 "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1523 error);
1da177e4
LT
1524 /*
1525 * We must turn off quotas.
1526 */
1527 ASSERT(mp->m_quotainfo != NULL);
1528 ASSERT(xfs_Gqm != NULL);
1529 xfs_qm_destroy_quotainfo(mp);
31d5577b 1530 if (xfs_mount_reset_sbqflags(mp)) {
0b932ccc
DC
1531 xfs_warn(mp,
1532 "Quotacheck: Failed to reset quota flags.");
31d5577b 1533 }
0b932ccc
DC
1534 } else
1535 xfs_notice(mp, "Quotacheck: Done.");
1da177e4
LT
1536 return (error);
1537}
1538
1539/*
1540 * This is called after the superblock has been read in and we're ready to
1541 * iget the quota inodes.
1542 */
1543STATIC int
1544xfs_qm_init_quotainos(
1545 xfs_mount_t *mp)
1546{
1547 xfs_inode_t *uip, *gip;
1548 int error;
1549 __int64_t sbflags;
1550 uint flags;
1551
1552 ASSERT(mp->m_quotainfo);
1553 uip = gip = NULL;
1554 sbflags = 0;
1555 flags = 0;
1556
1557 /*
1558 * Get the uquota and gquota inodes
1559 */
62118709 1560 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1da177e4
LT
1561 if (XFS_IS_UQUOTA_ON(mp) &&
1562 mp->m_sb.sb_uquotino != NULLFSINO) {
1563 ASSERT(mp->m_sb.sb_uquotino > 0);
1564 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
7b6259e7 1565 0, 0, &uip)))
1da177e4
LT
1566 return XFS_ERROR(error);
1567 }
c8ad20ff 1568 if (XFS_IS_OQUOTA_ON(mp) &&
1da177e4
LT
1569 mp->m_sb.sb_gquotino != NULLFSINO) {
1570 ASSERT(mp->m_sb.sb_gquotino > 0);
1571 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
7b6259e7 1572 0, 0, &gip))) {
1da177e4 1573 if (uip)
43355099 1574 IRELE(uip);
1da177e4
LT
1575 return XFS_ERROR(error);
1576 }
1577 }
1578 } else {
1579 flags |= XFS_QMOPT_SBVERSION;
1580 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1581 XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
1582 }
1583
1584 /*
1585 * Create the two inodes, if they don't exist already. The changes
1586 * made above will get added to a transaction and logged in one of
1587 * the qino_alloc calls below. If the device is readonly,
1588 * temporarily switch to read-write to do this.
1589 */
1590 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1591 if ((error = xfs_qm_qino_alloc(mp, &uip,
1592 sbflags | XFS_SB_UQUOTINO,
1593 flags | XFS_QMOPT_UQUOTA)))
1594 return XFS_ERROR(error);
1595
1596 flags &= ~XFS_QMOPT_SBVERSION;
1597 }
c8ad20ff
NS
1598 if (XFS_IS_OQUOTA_ON(mp) && gip == NULL) {
1599 flags |= (XFS_IS_GQUOTA_ON(mp) ?
1600 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1601 error = xfs_qm_qino_alloc(mp, &gip,
1602 sbflags | XFS_SB_GQUOTINO, flags);
1603 if (error) {
1da177e4 1604 if (uip)
43355099 1605 IRELE(uip);
1da177e4
LT
1606
1607 return XFS_ERROR(error);
1608 }
1609 }
1610
8a7b8a89
CH
1611 mp->m_quotainfo->qi_uquotaip = uip;
1612 mp->m_quotainfo->qi_gquotaip = gip;
1da177e4 1613
014c2544 1614 return 0;
1da177e4
LT
1615}
1616
1617
368e1361 1618
1da177e4 1619/*
92678554 1620 * Pop the least recently used dquot off the freelist and recycle it.
1da177e4 1621 */
92678554 1622STATIC struct xfs_dquot *
368e1361 1623xfs_qm_dqreclaim_one(void)
1da177e4 1624{
92678554
CH
1625 struct xfs_dquot *dqp;
1626 int restarts = 0;
1da177e4 1627
3a8406f6 1628 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
92678554 1629restart:
3a8406f6 1630 list_for_each_entry(dqp, &xfs_Gqm->qm_dqfrlist, q_freelist) {
3a25404b 1631 struct xfs_mount *mp = dqp->q_mount;
bf72de31
CH
1632
1633 if (!xfs_dqlock_nowait(dqp))
1634 continue;
1da177e4
LT
1635
1636 /*
be7ffc38
CH
1637 * This dquot has already been grabbed by dqlookup.
1638 * Remove it from the freelist and try again.
1da177e4 1639 */
be7ffc38 1640 if (dqp->q_nrefs) {
368e1361 1641 trace_xfs_dqreclaim_want(dqp);
1da177e4 1642 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
be7ffc38
CH
1643
1644 list_del_init(&dqp->q_freelist);
1645 xfs_Gqm->qm_dqfrlist_cnt--;
0fbca4d1 1646 restarts++;
0fbca4d1 1647 goto dqunlock;
1da177e4
LT
1648 }
1649
368e1361 1650 ASSERT(dqp->q_hash);
3a25404b 1651 ASSERT(!list_empty(&dqp->q_mplist));
368e1361 1652
1da177e4 1653 /*
0fbca4d1
DC
1654 * Try to grab the flush lock. If this dquot is in the process
1655 * of getting flushed to disk, we don't want to reclaim it.
1da177e4 1656 */
0fbca4d1
DC
1657 if (!xfs_dqflock_nowait(dqp))
1658 goto dqunlock;
1da177e4
LT
1659
1660 /*
1661 * We have the flush lock so we know that this is not in the
1662 * process of being flushed. So, if this is dirty, flush it
1663 * DELWRI so that we don't get a freelist infested with
1664 * dirty dquots.
1665 */
1666 if (XFS_DQ_IS_DIRTY(dqp)) {
3c56836f 1667 int error;
0b1b213f 1668
368e1361 1669 trace_xfs_dqreclaim_dirty(dqp);
0b1b213f 1670
1da177e4
LT
1671 /*
1672 * We flush it delayed write, so don't bother
368e1361 1673 * releasing the freelist lock.
1da177e4 1674 */
fdedf28b 1675 error = xfs_qm_dqflush(dqp, SYNC_TRYLOCK);
3c56836f 1676 if (error) {
53487786
DC
1677 xfs_warn(mp, "%s: dquot %p flush failed",
1678 __func__, dqp);
3c56836f 1679 }
0fbca4d1 1680 goto dqunlock;
1da177e4 1681 }
92678554 1682 xfs_dqfunlock(dqp);
368e1361 1683
1da177e4 1684 /*
92678554
CH
1685 * Prevent lookup now that we are going to reclaim the dquot.
1686 * Once XFS_DQ_FREEING is set lookup won't touch the dquot,
1687 * thus we can drop the lock now.
1da177e4 1688 */
92678554
CH
1689 dqp->dq_flags |= XFS_DQ_FREEING;
1690 xfs_dqunlock(dqp);
368e1361 1691
92678554
CH
1692 mutex_lock(&dqp->q_hash->qh_lock);
1693 list_del_init(&dqp->q_hashlist);
1694 dqp->q_hash->qh_version++;
1695 mutex_unlock(&dqp->q_hash->qh_lock);
0b1b213f 1696
92678554 1697 mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
3a25404b
DC
1698 list_del_init(&dqp->q_mplist);
1699 mp->m_quotainfo->qi_dquots--;
1700 mp->m_quotainfo->qi_dqreclaims++;
92678554
CH
1701 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
1702
1703 ASSERT(dqp->q_nrefs == 0);
3a8406f6
DC
1704 list_del_init(&dqp->q_freelist);
1705 xfs_Gqm->qm_dqfrlist_cnt--;
92678554
CH
1706
1707 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1708 return dqp;
0fbca4d1 1709dqunlock:
1da177e4 1710 xfs_dqunlock(dqp);
368e1361 1711 if (restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
0fbca4d1 1712 break;
92678554 1713 goto restart;
1da177e4 1714 }
92678554 1715
3a8406f6 1716 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
92678554 1717 return NULL;
1da177e4
LT
1718}
1719
368e1361
DC
1720/*
1721 * Traverse the freelist of dquots and attempt to reclaim a maximum of
1722 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
1723 * favor the lookup function ...
1724 */
1725STATIC int
1726xfs_qm_shake_freelist(
1727 int howmany)
1728{
1729 int nreclaimed = 0;
1730 xfs_dquot_t *dqp;
1731
1732 if (howmany <= 0)
1733 return 0;
1734
1735 while (nreclaimed < howmany) {
1736 dqp = xfs_qm_dqreclaim_one();
1737 if (!dqp)
1738 return nreclaimed;
1739 xfs_qm_dqdestroy(dqp);
1740 nreclaimed++;
1741 }
1742 return nreclaimed;
1743}
1da177e4
LT
1744
1745/*
1746 * The kmem_shake interface is invoked when memory is running low.
1747 */
1748/* ARGSUSED */
1749STATIC int
7f8275d0
DC
1750xfs_qm_shake(
1751 struct shrinker *shrink,
1495f230 1752 struct shrink_control *sc)
1da177e4
LT
1753{
1754 int ndqused, nfree, n;
1495f230 1755 gfp_t gfp_mask = sc->gfp_mask;
1da177e4
LT
1756
1757 if (!kmem_shake_allow(gfp_mask))
014c2544 1758 return 0;
1da177e4 1759 if (!xfs_Gqm)
014c2544 1760 return 0;
1da177e4 1761
3a8406f6 1762 nfree = xfs_Gqm->qm_dqfrlist_cnt; /* free dquots */
1da177e4
LT
1763 /* incore dquots in all f/s's */
1764 ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;
1765
1766 ASSERT(ndqused >= 0);
1767
1768 if (nfree <= ndqused && nfree < ndquot)
014c2544 1769 return 0;
1da177e4
LT
1770
1771 ndqused *= xfs_Gqm->qm_dqfree_ratio; /* target # of free dquots */
1772 n = nfree - ndqused - ndquot; /* # over target */
1773
1774 return xfs_qm_shake_freelist(MAX(nfree, n));
1775}
1776
1777
1da177e4
LT
1778/*------------------------------------------------------------------*/
1779
1780/*
1781 * Return a new incore dquot. Depending on the number of
1782 * dquots in the system, we either allocate a new one on the kernel heap,
1783 * or reclaim a free one.
1784 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
1785 * to reclaim an existing one from the freelist.
1786 */
1787boolean_t
1788xfs_qm_dqalloc_incore(
1789 xfs_dquot_t **O_dqpp)
1790{
1791 xfs_dquot_t *dqp;
1792
1793 /*
1794 * Check against high water mark to see if we want to pop
1795 * a nincompoop dquot off the freelist.
1796 */
1797 if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
1798 /*
1799 * Try to recycle a dquot from the freelist.
1800 */
1801 if ((dqp = xfs_qm_dqreclaim_one())) {
1802 XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
1803 /*
1804 * Just zero the core here. The rest will get
1805 * reinitialized by caller. XXX we shouldn't even
1806 * do this zero ...
1807 */
1808 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
1809 *O_dqpp = dqp;
014c2544 1810 return B_FALSE;
1da177e4
LT
1811 }
1812 XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
1813 }
1814
1815 /*
1816 * Allocate a brand new dquot on the kernel heap and return it
1817 * to the caller to initialize.
1818 */
1819 ASSERT(xfs_Gqm->qm_dqzone != NULL);
1820 *O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
1821 atomic_inc(&xfs_Gqm->qm_totaldquots);
1822
014c2544 1823 return B_TRUE;
1da177e4
LT
1824}
1825
1826
1827/*
1828 * Start a transaction and write the incore superblock changes to
1829 * disk. flags parameter indicates which fields have changed.
1830 */
1831int
1832xfs_qm_write_sb_changes(
1833 xfs_mount_t *mp,
1834 __int64_t flags)
1835{
1836 xfs_trans_t *tp;
1837 int error;
1838
1da177e4
LT
1839 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1840 if ((error = xfs_trans_reserve(tp, 0,
1841 mp->m_sb.sb_sectsize + 128, 0,
1842 0,
1843 XFS_DEFAULT_LOG_COUNT))) {
1844 xfs_trans_cancel(tp, 0);
014c2544 1845 return error;
1da177e4
LT
1846 }
1847
1848 xfs_mod_sb(tp, flags);
e5720eec 1849 error = xfs_trans_commit(tp, 0);
1da177e4 1850
e5720eec 1851 return error;
1da177e4
LT
1852}
1853
1854
1855/* --------------- utility functions for vnodeops ---------------- */
1856
1857
1858/*
6c77b0ea 1859 * Given an inode, a uid, gid and prid make sure that we have
1da177e4
LT
1860 * allocated relevant dquot(s) on disk, and that we won't exceed inode
1861 * quotas by creating this file.
1862 * This also attaches dquot(s) to the given inode after locking it,
1863 * and returns the dquots corresponding to the uid and/or gid.
1864 *
1865 * in : inode (unlocked)
1866 * out : udquot, gdquot with references taken and unlocked
1867 */
1868int
1869xfs_qm_vop_dqalloc(
7d095257
CH
1870 struct xfs_inode *ip,
1871 uid_t uid,
1872 gid_t gid,
1873 prid_t prid,
1874 uint flags,
1875 struct xfs_dquot **O_udqpp,
1876 struct xfs_dquot **O_gdqpp)
1da177e4 1877{
7d095257
CH
1878 struct xfs_mount *mp = ip->i_mount;
1879 struct xfs_dquot *uq, *gq;
1880 int error;
1881 uint lockflags;
1da177e4 1882
7d095257 1883 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1da177e4
LT
1884 return 0;
1885
1886 lockflags = XFS_ILOCK_EXCL;
1887 xfs_ilock(ip, lockflags);
1888
bd186aa9 1889 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1da177e4
LT
1890 gid = ip->i_d.di_gid;
1891
1892 /*
1893 * Attach the dquot(s) to this inode, doing a dquot allocation
1894 * if necessary. The dquot(s) will not be locked.
1895 */
1896 if (XFS_NOT_DQATTACHED(mp, ip)) {
7d095257
CH
1897 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
1898 if (error) {
1da177e4 1899 xfs_iunlock(ip, lockflags);
014c2544 1900 return error;
1da177e4
LT
1901 }
1902 }
1903
1904 uq = gq = NULL;
c8ad20ff 1905 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1da177e4
LT
1906 if (ip->i_d.di_uid != uid) {
1907 /*
1908 * What we need is the dquot that has this uid, and
1909 * if we send the inode to dqget, the uid of the inode
1910 * takes priority over what's sent in the uid argument.
1911 * We must unlock inode here before calling dqget if
1912 * we're not sending the inode, because otherwise
1913 * we'll deadlock by doing trans_reserve while
1914 * holding ilock.
1915 */
1916 xfs_iunlock(ip, lockflags);
1917 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
1918 XFS_DQ_USER,
1919 XFS_QMOPT_DQALLOC |
1920 XFS_QMOPT_DOWARN,
1921 &uq))) {
1922 ASSERT(error != ENOENT);
014c2544 1923 return error;
1da177e4
LT
1924 }
1925 /*
1926 * Get the ilock in the right order.
1927 */
1928 xfs_dqunlock(uq);
1929 lockflags = XFS_ILOCK_SHARED;
1930 xfs_ilock(ip, lockflags);
1931 } else {
1932 /*
1933 * Take an extra reference, because we'll return
1934 * this to caller
1935 */
1936 ASSERT(ip->i_udquot);
78e55892 1937 uq = xfs_qm_dqhold(ip->i_udquot);
1da177e4
LT
1938 }
1939 }
c8ad20ff 1940 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1da177e4
LT
1941 if (ip->i_d.di_gid != gid) {
1942 xfs_iunlock(ip, lockflags);
1943 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
1944 XFS_DQ_GROUP,
1945 XFS_QMOPT_DQALLOC |
1946 XFS_QMOPT_DOWARN,
1947 &gq))) {
1948 if (uq)
1949 xfs_qm_dqrele(uq);
1950 ASSERT(error != ENOENT);
014c2544 1951 return error;
1da177e4
LT
1952 }
1953 xfs_dqunlock(gq);
1954 lockflags = XFS_ILOCK_SHARED;
1955 xfs_ilock(ip, lockflags);
1956 } else {
1957 ASSERT(ip->i_gdquot);
78e55892 1958 gq = xfs_qm_dqhold(ip->i_gdquot);
1da177e4 1959 }
c8ad20ff 1960 } else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
6743099c 1961 if (xfs_get_projid(ip) != prid) {
c8ad20ff
NS
1962 xfs_iunlock(ip, lockflags);
1963 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
1964 XFS_DQ_PROJ,
1965 XFS_QMOPT_DQALLOC |
1966 XFS_QMOPT_DOWARN,
1967 &gq))) {
1968 if (uq)
1969 xfs_qm_dqrele(uq);
1970 ASSERT(error != ENOENT);
1971 return (error);
1972 }
1973 xfs_dqunlock(gq);
1974 lockflags = XFS_ILOCK_SHARED;
1975 xfs_ilock(ip, lockflags);
1976 } else {
1977 ASSERT(ip->i_gdquot);
78e55892 1978 gq = xfs_qm_dqhold(ip->i_gdquot);
c8ad20ff 1979 }
1da177e4
LT
1980 }
1981 if (uq)
0b1b213f 1982 trace_xfs_dquot_dqalloc(ip);
1da177e4
LT
1983
1984 xfs_iunlock(ip, lockflags);
1985 if (O_udqpp)
1986 *O_udqpp = uq;
1987 else if (uq)
1988 xfs_qm_dqrele(uq);
1989 if (O_gdqpp)
1990 *O_gdqpp = gq;
1991 else if (gq)
1992 xfs_qm_dqrele(gq);
014c2544 1993 return 0;
1da177e4
LT
1994}
1995
1996/*
1997 * Actually transfer ownership, and do dquot modifications.
1998 * These were already reserved.
1999 */
2000xfs_dquot_t *
2001xfs_qm_vop_chown(
2002 xfs_trans_t *tp,
2003 xfs_inode_t *ip,
2004 xfs_dquot_t **IO_olddq,
2005 xfs_dquot_t *newdq)
2006{
2007 xfs_dquot_t *prevdq;
06d10dd9
NS
2008 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
2009 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
2010
7d095257 2011
579aa9ca 2012 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4
LT
2013 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
2014
2015 /* old dquot */
2016 prevdq = *IO_olddq;
2017 ASSERT(prevdq);
2018 ASSERT(prevdq != newdq);
2019
06d10dd9
NS
2020 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
2021 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1da177e4
LT
2022
2023 /* the sparkling new dquot */
06d10dd9
NS
2024 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
2025 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1da177e4
LT
2026
2027 /*
78e55892
CH
2028 * Take an extra reference, because the inode is going to keep
2029 * this dquot pointer even after the trans_commit.
1da177e4 2030 */
78e55892 2031 *IO_olddq = xfs_qm_dqhold(newdq);
1da177e4 2032
014c2544 2033 return prevdq;
1da177e4
LT
2034}
2035
2036/*
c8ad20ff 2037 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1da177e4
LT
2038 */
2039int
2040xfs_qm_vop_chown_reserve(
2041 xfs_trans_t *tp,
2042 xfs_inode_t *ip,
2043 xfs_dquot_t *udqp,
2044 xfs_dquot_t *gdqp,
2045 uint flags)
2046{
7d095257 2047 xfs_mount_t *mp = ip->i_mount;
9a2a7de2 2048 uint delblks, blkflags, prjflags = 0;
1da177e4 2049 xfs_dquot_t *unresudq, *unresgdq, *delblksudq, *delblksgdq;
7d095257
CH
2050 int error;
2051
1da177e4 2052
579aa9ca 2053 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1da177e4
LT
2054 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2055
2056 delblks = ip->i_delayed_blks;
2057 delblksudq = delblksgdq = unresudq = unresgdq = NULL;
06d10dd9
NS
2058 blkflags = XFS_IS_REALTIME_INODE(ip) ?
2059 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
1da177e4
LT
2060
2061 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
1149d96a 2062 ip->i_d.di_uid != (uid_t)be32_to_cpu(udqp->q_core.d_id)) {
1da177e4
LT
2063 delblksudq = udqp;
2064 /*
2065 * If there are delayed allocation blocks, then we have to
2066 * unreserve those from the old dquot, and add them to the
2067 * new dquot.
2068 */
2069 if (delblks) {
2070 ASSERT(ip->i_udquot);
2071 unresudq = ip->i_udquot;
2072 }
2073 }
c8ad20ff 2074 if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
9a2a7de2 2075 if (XFS_IS_PQUOTA_ON(ip->i_mount) &&
6743099c 2076 xfs_get_projid(ip) != be32_to_cpu(gdqp->q_core.d_id))
9a2a7de2
NS
2077 prjflags = XFS_QMOPT_ENOSPC;
2078
2079 if (prjflags ||
2080 (XFS_IS_GQUOTA_ON(ip->i_mount) &&
2081 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id))) {
c8ad20ff
NS
2082 delblksgdq = gdqp;
2083 if (delblks) {
2084 ASSERT(ip->i_gdquot);
2085 unresgdq = ip->i_gdquot;
2086 }
1da177e4
LT
2087 }
2088 }
2089
2090 if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2091 delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
9a2a7de2 2092 flags | blkflags | prjflags)))
1da177e4
LT
2093 return (error);
2094
2095 /*
2096 * Do the delayed blks reservations/unreservations now. Since, these
2097 * are done without the help of a transaction, if a reservation fails
2098 * its previous reservations won't be automatically undone by trans
2099 * code. So, we have to do it manually here.
2100 */
2101 if (delblks) {
2102 /*
2103 * Do the reservations first. Unreservation can't fail.
2104 */
2105 ASSERT(delblksudq || delblksgdq);
2106 ASSERT(unresudq || unresgdq);
2107 if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2108 delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
9a2a7de2 2109 flags | blkflags | prjflags)))
1da177e4
LT
2110 return (error);
2111 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2112 unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
06d10dd9 2113 blkflags);
1da177e4
LT
2114 }
2115
2116 return (0);
2117}
2118
2119int
2120xfs_qm_vop_rename_dqattach(
7d095257 2121 struct xfs_inode **i_tab)
1da177e4 2122{
7d095257
CH
2123 struct xfs_mount *mp = i_tab[0]->i_mount;
2124 int i;
1da177e4 2125
7d095257 2126 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
014c2544 2127 return 0;
1da177e4 2128
7d095257
CH
2129 for (i = 0; (i < 4 && i_tab[i]); i++) {
2130 struct xfs_inode *ip = i_tab[i];
2131 int error;
2132
1da177e4
LT
2133 /*
2134 * Watch out for duplicate entries in the table.
2135 */
7d095257
CH
2136 if (i == 0 || ip != i_tab[i-1]) {
2137 if (XFS_NOT_DQATTACHED(mp, ip)) {
1da177e4
LT
2138 error = xfs_qm_dqattach(ip, 0);
2139 if (error)
014c2544 2140 return error;
1da177e4
LT
2141 }
2142 }
2143 }
014c2544 2144 return 0;
1da177e4
LT
2145}
2146
2147void
7d095257
CH
2148xfs_qm_vop_create_dqattach(
2149 struct xfs_trans *tp,
2150 struct xfs_inode *ip,
2151 struct xfs_dquot *udqp,
2152 struct xfs_dquot *gdqp)
1da177e4 2153{
7d095257
CH
2154 struct xfs_mount *mp = tp->t_mountp;
2155
2156 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1da177e4
LT
2157 return;
2158
579aa9ca 2159 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
7d095257 2160 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1da177e4
LT
2161
2162 if (udqp) {
1da177e4 2163 ASSERT(ip->i_udquot == NULL);
7d095257 2164 ASSERT(XFS_IS_UQUOTA_ON(mp));
1149d96a 2165 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
78e55892
CH
2166
2167 ip->i_udquot = xfs_qm_dqhold(udqp);
1da177e4
LT
2168 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2169 }
2170 if (gdqp) {
1da177e4 2171 ASSERT(ip->i_gdquot == NULL);
7d095257
CH
2172 ASSERT(XFS_IS_OQUOTA_ON(mp));
2173 ASSERT((XFS_IS_GQUOTA_ON(mp) ?
6743099c 2174 ip->i_d.di_gid : xfs_get_projid(ip)) ==
ee2a4f7c 2175 be32_to_cpu(gdqp->q_core.d_id));
78e55892
CH
2176
2177 ip->i_gdquot = xfs_qm_dqhold(gdqp);
1da177e4
LT
2178 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
2179 }
2180}
2181