]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/xfs/xfs_dquot.c
xfs: factor unwritten extent map manipulations out of xfs_bmapi
[mirror_ubuntu-zesty-kernel.git] / fs / xfs / xfs_dquot.c
CommitLineData
1da177e4 1/*
4ce3121f
NS
2 * Copyright (c) 2000-2003 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 29#include "xfs_bmap_btree.h"
1da177e4
LT
30#include "xfs_inode.h"
31#include "xfs_bmap.h"
1da177e4
LT
32#include "xfs_rtalloc.h"
33#include "xfs_error.h"
34#include "xfs_itable.h"
1da177e4
LT
35#include "xfs_attr.h"
36#include "xfs_buf_item.h"
37#include "xfs_trans_space.h"
38#include "xfs_trans_priv.h"
1da177e4 39#include "xfs_qm.h"
0b1b213f 40#include "xfs_trace.h"
1da177e4
LT
41
42
43/*
44 LOCK ORDER
45
46 inode lock (ilock)
47 dquot hash-chain lock (hashlock)
48 xqm dquot freelist lock (freelistlock
49 mount's dquot list lock (mplistlock)
50 user dquot lock - lock ordering among dquots is based on the uid or gid
51 group dquot lock - similar to udquots. Between the two dquots, the udquot
52 has to be locked first.
53 pin lock - the dquot lock must be held to take this lock.
54 flush lock - ditto.
55*/
56
1da177e4
LT
57#ifdef DEBUG
58xfs_buftarg_t *xfs_dqerror_target;
59int xfs_do_dqerror;
60int xfs_dqreq_num;
61int xfs_dqerror_mod = 33;
62#endif
63
98b8c7a0
CH
64static struct lock_class_key xfs_dquot_other_class;
65
1da177e4
LT
66/*
67 * Allocate and initialize a dquot. We don't always allocate fresh memory;
68 * we try to reclaim a free dquot if the number of incore dquots are above
69 * a threshold.
70 * The only field inside the core that gets initialized at this point
71 * is the d_id field. The idea is to fill in the entire q_core
72 * when we read in the on disk dquot.
73 */
ba0f32d4 74STATIC xfs_dquot_t *
1da177e4
LT
75xfs_qm_dqinit(
76 xfs_mount_t *mp,
77 xfs_dqid_t id,
78 uint type)
79{
80 xfs_dquot_t *dqp;
81 boolean_t brandnewdquot;
82
83 brandnewdquot = xfs_qm_dqalloc_incore(&dqp);
84 dqp->dq_flags = type;
1149d96a 85 dqp->q_core.d_id = cpu_to_be32(id);
1da177e4
LT
86 dqp->q_mount = mp;
87
88 /*
89 * No need to re-initialize these if this is a reclaimed dquot.
90 */
91 if (brandnewdquot) {
3a8406f6 92 INIT_LIST_HEAD(&dqp->q_freelist);
794ee1ba 93 mutex_init(&dqp->q_qlock);
bc3048e3 94 init_waitqueue_head(&dqp->q_pinwait);
1da177e4 95
e1f49cf2
DC
96 /*
97 * Because we want to use a counting completion, complete
98 * the flush completion once to allow a single access to
99 * the flush completion without blocking.
100 */
101 init_completion(&dqp->q_flush);
102 complete(&dqp->q_flush);
103
0b1b213f 104 trace_xfs_dqinit(dqp);
1da177e4
LT
105 } else {
106 /*
107 * Only the q_core portion was zeroed in dqreclaim_one().
108 * So, we need to reset others.
109 */
3a8406f6
DC
110 dqp->q_nrefs = 0;
111 dqp->q_blkno = 0;
112 INIT_LIST_HEAD(&dqp->q_mplist);
113 INIT_LIST_HEAD(&dqp->q_hashlist);
114 dqp->q_bufoffset = 0;
115 dqp->q_fileoffset = 0;
116 dqp->q_transp = NULL;
117 dqp->q_gdquot = NULL;
118 dqp->q_res_bcount = 0;
119 dqp->q_res_icount = 0;
120 dqp->q_res_rtbcount = 0;
121 atomic_set(&dqp->q_pincount, 0);
122 dqp->q_hash = NULL;
123 ASSERT(list_empty(&dqp->q_freelist));
1da177e4 124
0b1b213f 125 trace_xfs_dqreuse(dqp);
98b8c7a0
CH
126 }
127
128 /*
129 * In either case we need to make sure group quotas have a different
130 * lock class than user quotas, to make sure lockdep knows we can
131 * locks of one of each at the same time.
132 */
133 if (!(type & XFS_DQ_USER))
134 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class);
1da177e4
LT
135
136 /*
137 * log item gets initialized later
138 */
139 return (dqp);
140}
141
142/*
143 * This is called to free all the memory associated with a dquot
144 */
145void
146xfs_qm_dqdestroy(
147 xfs_dquot_t *dqp)
148{
3a8406f6 149 ASSERT(list_empty(&dqp->q_freelist));
1da177e4
LT
150
151 mutex_destroy(&dqp->q_qlock);
1da177e4 152 kmem_zone_free(xfs_Gqm->qm_dqzone, dqp);
0b1b213f 153
1da177e4
LT
154 atomic_dec(&xfs_Gqm->qm_totaldquots);
155}
156
157/*
158 * This is what a 'fresh' dquot inside a dquot chunk looks like on disk.
159 */
160STATIC void
161xfs_qm_dqinit_core(
162 xfs_dqid_t id,
163 uint type,
164 xfs_dqblk_t *d)
165{
166 /*
167 * Caller has zero'd the entire dquot 'chunk' already.
168 */
1149d96a
CH
169 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
170 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
171 d->dd_diskdq.d_id = cpu_to_be32(id);
172 d->dd_diskdq.d_flags = type;
1da177e4
LT
173}
174
1da177e4
LT
175/*
176 * If default limits are in force, push them into the dquot now.
177 * We overwrite the dquot limits only if they are zero and this
178 * is not the root dquot.
179 */
180void
181xfs_qm_adjust_dqlimits(
182 xfs_mount_t *mp,
183 xfs_disk_dquot_t *d)
184{
185 xfs_quotainfo_t *q = mp->m_quotainfo;
186
187 ASSERT(d->d_id);
188
189 if (q->qi_bsoftlimit && !d->d_blk_softlimit)
1149d96a 190 d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit);
1da177e4 191 if (q->qi_bhardlimit && !d->d_blk_hardlimit)
1149d96a 192 d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit);
1da177e4 193 if (q->qi_isoftlimit && !d->d_ino_softlimit)
1149d96a 194 d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit);
1da177e4 195 if (q->qi_ihardlimit && !d->d_ino_hardlimit)
1149d96a 196 d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit);
1da177e4 197 if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit)
1149d96a 198 d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit);
1da177e4 199 if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit)
1149d96a 200 d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit);
1da177e4
LT
201}
202
203/*
204 * Check the limits and timers of a dquot and start or reset timers
205 * if necessary.
206 * This gets called even when quota enforcement is OFF, which makes our
207 * life a little less complicated. (We just don't reject any quota
208 * reservations in that case, when enforcement is off).
209 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
210 * enforcement's off.
211 * In contrast, warnings are a little different in that they don't
754002b4
NS
212 * 'automatically' get started when limits get exceeded. They do
213 * get reset to zero, however, when we find the count to be under
214 * the soft limit (they are only ever set non-zero via userspace).
1da177e4
LT
215 */
216void
217xfs_qm_adjust_dqtimers(
218 xfs_mount_t *mp,
219 xfs_disk_dquot_t *d)
220{
221 ASSERT(d->d_id);
222
ea15ab3c 223#ifdef DEBUG
1149d96a
CH
224 if (d->d_blk_hardlimit)
225 ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
226 be64_to_cpu(d->d_blk_hardlimit));
227 if (d->d_ino_hardlimit)
228 ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
229 be64_to_cpu(d->d_ino_hardlimit));
230 if (d->d_rtb_hardlimit)
231 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
232 be64_to_cpu(d->d_rtb_hardlimit));
1da177e4 233#endif
ea15ab3c 234
1da177e4 235 if (!d->d_btimer) {
1149d96a
CH
236 if ((d->d_blk_softlimit &&
237 (be64_to_cpu(d->d_bcount) >=
238 be64_to_cpu(d->d_blk_softlimit))) ||
239 (d->d_blk_hardlimit &&
240 (be64_to_cpu(d->d_bcount) >=
241 be64_to_cpu(d->d_blk_hardlimit)))) {
242 d->d_btimer = cpu_to_be32(get_seconds() +
8a7b8a89 243 mp->m_quotainfo->qi_btimelimit);
754002b4
NS
244 } else {
245 d->d_bwarns = 0;
1da177e4
LT
246 }
247 } else {
248 if ((!d->d_blk_softlimit ||
1149d96a
CH
249 (be64_to_cpu(d->d_bcount) <
250 be64_to_cpu(d->d_blk_softlimit))) &&
1da177e4 251 (!d->d_blk_hardlimit ||
1149d96a
CH
252 (be64_to_cpu(d->d_bcount) <
253 be64_to_cpu(d->d_blk_hardlimit)))) {
1da177e4
LT
254 d->d_btimer = 0;
255 }
256 }
257
258 if (!d->d_itimer) {
1149d96a
CH
259 if ((d->d_ino_softlimit &&
260 (be64_to_cpu(d->d_icount) >=
261 be64_to_cpu(d->d_ino_softlimit))) ||
262 (d->d_ino_hardlimit &&
263 (be64_to_cpu(d->d_icount) >=
264 be64_to_cpu(d->d_ino_hardlimit)))) {
265 d->d_itimer = cpu_to_be32(get_seconds() +
8a7b8a89 266 mp->m_quotainfo->qi_itimelimit);
754002b4
NS
267 } else {
268 d->d_iwarns = 0;
1da177e4
LT
269 }
270 } else {
271 if ((!d->d_ino_softlimit ||
1149d96a
CH
272 (be64_to_cpu(d->d_icount) <
273 be64_to_cpu(d->d_ino_softlimit))) &&
1da177e4 274 (!d->d_ino_hardlimit ||
1149d96a
CH
275 (be64_to_cpu(d->d_icount) <
276 be64_to_cpu(d->d_ino_hardlimit)))) {
1da177e4
LT
277 d->d_itimer = 0;
278 }
279 }
280
281 if (!d->d_rtbtimer) {
1149d96a
CH
282 if ((d->d_rtb_softlimit &&
283 (be64_to_cpu(d->d_rtbcount) >=
284 be64_to_cpu(d->d_rtb_softlimit))) ||
285 (d->d_rtb_hardlimit &&
286 (be64_to_cpu(d->d_rtbcount) >=
287 be64_to_cpu(d->d_rtb_hardlimit)))) {
288 d->d_rtbtimer = cpu_to_be32(get_seconds() +
8a7b8a89 289 mp->m_quotainfo->qi_rtbtimelimit);
754002b4
NS
290 } else {
291 d->d_rtbwarns = 0;
1da177e4
LT
292 }
293 } else {
294 if ((!d->d_rtb_softlimit ||
1149d96a
CH
295 (be64_to_cpu(d->d_rtbcount) <
296 be64_to_cpu(d->d_rtb_softlimit))) &&
1da177e4 297 (!d->d_rtb_hardlimit ||
1149d96a
CH
298 (be64_to_cpu(d->d_rtbcount) <
299 be64_to_cpu(d->d_rtb_hardlimit)))) {
1da177e4
LT
300 d->d_rtbtimer = 0;
301 }
302 }
303}
304
1da177e4
LT
305/*
306 * initialize a buffer full of dquots and log the whole thing
307 */
308STATIC void
309xfs_qm_init_dquot_blk(
310 xfs_trans_t *tp,
311 xfs_mount_t *mp,
312 xfs_dqid_t id,
313 uint type,
314 xfs_buf_t *bp)
315{
8a7b8a89 316 struct xfs_quotainfo *q = mp->m_quotainfo;
1da177e4
LT
317 xfs_dqblk_t *d;
318 int curid, i;
319
320 ASSERT(tp);
0c842ad4 321 ASSERT(xfs_buf_islocked(bp));
1da177e4 322
62926044 323 d = bp->b_addr;
1da177e4
LT
324
325 /*
326 * ID of the first dquot in the block - id's are zero based.
327 */
8a7b8a89 328 curid = id - (id % q->qi_dqperchunk);
1da177e4 329 ASSERT(curid >= 0);
8a7b8a89
CH
330 memset(d, 0, BBTOB(q->qi_dqchunklen));
331 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++)
1da177e4
LT
332 xfs_qm_dqinit_core(curid, type, d);
333 xfs_trans_dquot_buf(tp, bp,
c1155410
DC
334 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
335 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
336 XFS_BLF_GDQUOT_BUF)));
8a7b8a89 337 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
1da177e4
LT
338}
339
340
341
342/*
343 * Allocate a block and fill it with dquots.
344 * This is called when the bmapi finds a hole.
345 */
346STATIC int
347xfs_qm_dqalloc(
efa092f3 348 xfs_trans_t **tpp,
1da177e4
LT
349 xfs_mount_t *mp,
350 xfs_dquot_t *dqp,
351 xfs_inode_t *quotip,
352 xfs_fileoff_t offset_fsb,
353 xfs_buf_t **O_bpp)
354{
355 xfs_fsblock_t firstblock;
356 xfs_bmap_free_t flist;
357 xfs_bmbt_irec_t map;
358 int nmaps, error, committed;
359 xfs_buf_t *bp;
efa092f3 360 xfs_trans_t *tp = *tpp;
1da177e4
LT
361
362 ASSERT(tp != NULL);
0b1b213f
CH
363
364 trace_xfs_dqalloc(dqp);
1da177e4
LT
365
366 /*
367 * Initialize the bmap freelist prior to calling bmapi code.
368 */
9d87c319 369 xfs_bmap_init(&flist, &firstblock);
1da177e4
LT
370 xfs_ilock(quotip, XFS_ILOCK_EXCL);
371 /*
372 * Return if this type of quotas is turned off while we didn't
373 * have an inode lock
374 */
375 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
376 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
377 return (ESRCH);
378 }
379
898621d5 380 xfs_trans_ijoin_ref(tp, quotip, XFS_ILOCK_EXCL);
1da177e4
LT
381 nmaps = 1;
382 if ((error = xfs_bmapi(tp, quotip,
383 offset_fsb, XFS_DQUOT_CLUSTER_SIZE_FSB,
384 XFS_BMAPI_METADATA | XFS_BMAPI_WRITE,
385 &firstblock,
386 XFS_QM_DQALLOC_SPACE_RES(mp),
b4e9181e 387 &map, &nmaps, &flist))) {
1da177e4
LT
388 goto error0;
389 }
390 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
391 ASSERT(nmaps == 1);
392 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
393 (map.br_startblock != HOLESTARTBLOCK));
394
395 /*
396 * Keep track of the blkno to save a lookup later
397 */
398 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
399
400 /* now we can just get the buffer (there's nothing to read yet) */
401 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
402 dqp->q_blkno,
8a7b8a89 403 mp->m_quotainfo->qi_dqchunklen,
1da177e4 404 0);
2a30f36d
CS
405
406 error = xfs_buf_geterror(bp);
407 if (error)
1da177e4 408 goto error1;
2a30f36d 409
1da177e4
LT
410 /*
411 * Make a chunk of dquots out of this buffer and log
412 * the entire thing.
413 */
1149d96a 414 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
c8ad20ff 415 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
1da177e4 416
efa092f3
TS
417 /*
418 * xfs_bmap_finish() may commit the current transaction and
419 * start a second transaction if the freelist is not empty.
420 *
421 * Since we still want to modify this buffer, we need to
422 * ensure that the buffer is not released on commit of
423 * the first transaction and ensure the buffer is added to the
424 * second transaction.
425 *
426 * If there is only one transaction then don't stop the buffer
427 * from being released when it commits later on.
428 */
429
430 xfs_trans_bhold(tp, bp);
431
f7c99b6f 432 if ((error = xfs_bmap_finish(tpp, &flist, &committed))) {
1da177e4
LT
433 goto error1;
434 }
435
efa092f3
TS
436 if (committed) {
437 tp = *tpp;
438 xfs_trans_bjoin(tp, bp);
439 } else {
440 xfs_trans_bhold_release(tp, bp);
441 }
442
1da177e4
LT
443 *O_bpp = bp;
444 return 0;
445
446 error1:
447 xfs_bmap_cancel(&flist);
448 error0:
449 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
450
451 return (error);
452}
453
454/*
455 * Maps a dquot to the buffer containing its on-disk version.
456 * This returns a ptr to the buffer containing the on-disk dquot
457 * in the bpp param, and a ptr to the on-disk dquot within that buffer
458 */
459STATIC int
460xfs_qm_dqtobp(
efa092f3 461 xfs_trans_t **tpp,
1da177e4
LT
462 xfs_dquot_t *dqp,
463 xfs_disk_dquot_t **O_ddpp,
464 xfs_buf_t **O_bpp,
465 uint flags)
466{
467 xfs_bmbt_irec_t map;
acecf1b5 468 int nmaps = 1, error;
1da177e4 469 xfs_buf_t *bp;
acecf1b5
CH
470 xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp);
471 xfs_mount_t *mp = dqp->q_mount;
1da177e4 472 xfs_disk_dquot_t *ddq;
acecf1b5 473 xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id);
efa092f3 474 xfs_trans_t *tp = (tpp ? *tpp : NULL);
1da177e4 475
acecf1b5 476 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
1da177e4 477
acecf1b5
CH
478 xfs_ilock(quotip, XFS_ILOCK_SHARED);
479 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
1da177e4 480 /*
acecf1b5
CH
481 * Return if this type of quotas is turned off while we
482 * didn't have the quota inode lock.
1da177e4 483 */
acecf1b5
CH
484 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
485 return ESRCH;
486 }
487
488 /*
489 * Find the block map; no allocations yet
490 */
5c8ed202
DC
491 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
492 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
acecf1b5
CH
493
494 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
495 if (error)
496 return error;
497
498 ASSERT(nmaps == 1);
499 ASSERT(map.br_blockcount == 1);
500
501 /*
502 * Offset of dquot in the (fixed sized) dquot chunk.
503 */
504 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
505 sizeof(xfs_dqblk_t);
506
507 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
508 if (map.br_startblock == HOLESTARTBLOCK) {
1da177e4 509 /*
acecf1b5 510 * We don't allocate unless we're asked to
1da177e4 511 */
acecf1b5
CH
512 if (!(flags & XFS_QMOPT_DQALLOC))
513 return ENOENT;
1da177e4 514
acecf1b5
CH
515 ASSERT(tp);
516 error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
517 dqp->q_fileoffset, &bp);
1da177e4 518 if (error)
acecf1b5
CH
519 return error;
520 tp = *tpp;
521 } else {
522 trace_xfs_dqtobp_read(dqp);
1da177e4
LT
523
524 /*
acecf1b5
CH
525 * store the blkno etc so that we don't have to do the
526 * mapping all the time
1da177e4 527 */
acecf1b5 528 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
0b1b213f 529
8a7b8a89
CH
530 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
531 dqp->q_blkno,
532 mp->m_quotainfo->qi_dqchunklen,
533 0, &bp);
1da177e4
LT
534 if (error || !bp)
535 return XFS_ERROR(error);
536 }
acecf1b5 537
0c842ad4 538 ASSERT(xfs_buf_islocked(bp));
1da177e4
LT
539
540 /*
541 * calculate the location of the dquot inside the buffer.
542 */
62926044 543 ddq = bp->b_addr + dqp->q_bufoffset;
1da177e4
LT
544
545 /*
546 * A simple sanity check in case we got a corrupted dquot...
547 */
a0fa2b67 548 error = xfs_qm_dqcheck(mp, ddq, id, dqp->dq_flags & XFS_DQ_ALLTYPES,
1da177e4 549 flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN),
a0fa2b67
DC
550 "dqtobp");
551 if (error) {
1da177e4
LT
552 if (!(flags & XFS_QMOPT_DQREPAIR)) {
553 xfs_trans_brelse(tp, bp);
554 return XFS_ERROR(EIO);
555 }
1da177e4
LT
556 }
557
558 *O_bpp = bp;
559 *O_ddpp = ddq;
560
561 return (0);
562}
563
564
565/*
566 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
567 * and release the buffer immediately.
568 *
569 */
570/* ARGSUSED */
571STATIC int
572xfs_qm_dqread(
efa092f3 573 xfs_trans_t **tpp,
1da177e4
LT
574 xfs_dqid_t id,
575 xfs_dquot_t *dqp, /* dquot to get filled in */
576 uint flags)
577{
578 xfs_disk_dquot_t *ddqp;
579 xfs_buf_t *bp;
580 int error;
efa092f3
TS
581 xfs_trans_t *tp;
582
583 ASSERT(tpp);
1da177e4 584
0b1b213f
CH
585 trace_xfs_dqread(dqp);
586
1da177e4
LT
587 /*
588 * get a pointer to the on-disk dquot and the buffer containing it
589 * dqp already knows its own type (GROUP/USER).
590 */
efa092f3 591 if ((error = xfs_qm_dqtobp(tpp, dqp, &ddqp, &bp, flags))) {
1da177e4
LT
592 return (error);
593 }
efa092f3 594 tp = *tpp;
1da177e4
LT
595
596 /* copy everything from disk dquot to the incore dquot */
597 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
1149d96a 598 ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);
1da177e4
LT
599 xfs_qm_dquot_logitem_init(dqp);
600
601 /*
602 * Reservation counters are defined as reservation plus current usage
25985edc 603 * to avoid having to add every time.
1da177e4 604 */
1149d96a
CH
605 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
606 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
607 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
1da177e4
LT
608
609 /* Mark the buf so that this will stay incore a little longer */
610 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DQUOT, XFS_DQUOT_REF);
611
612 /*
613 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
614 * So we need to release with xfs_trans_brelse().
615 * The strategy here is identical to that of inodes; we lock
616 * the dquot in xfs_qm_dqget() before making it accessible to
617 * others. This is because dquots, like inodes, need a good level of
618 * concurrency, and we don't want to take locks on the entire buffers
619 * for dquot accesses.
620 * Note also that the dquot buffer may even be dirty at this point, if
621 * this particular dquot was repaired. We still aren't afraid to
622 * brelse it because we have the changes incore.
623 */
0c842ad4 624 ASSERT(xfs_buf_islocked(bp));
1da177e4
LT
625 xfs_trans_brelse(tp, bp);
626
627 return (error);
628}
629
630
631/*
632 * allocate an incore dquot from the kernel heap,
633 * and fill its core with quota information kept on disk.
634 * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk
635 * if it wasn't already allocated.
636 */
637STATIC int
638xfs_qm_idtodq(
639 xfs_mount_t *mp,
640 xfs_dqid_t id, /* gid or uid, depending on type */
641 uint type, /* UDQUOT or GDQUOT */
642 uint flags, /* DQALLOC, DQREPAIR */
643 xfs_dquot_t **O_dqpp)/* OUT : incore dquot, not locked */
644{
645 xfs_dquot_t *dqp;
646 int error;
647 xfs_trans_t *tp;
648 int cancelflags=0;
649
650 dqp = xfs_qm_dqinit(mp, id, type);
651 tp = NULL;
652 if (flags & XFS_QMOPT_DQALLOC) {
653 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
8a7b8a89
CH
654 error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp),
655 XFS_WRITE_LOG_RES(mp) +
656 BBTOB(mp->m_quotainfo->qi_dqchunklen) - 1 +
657 128,
658 0,
659 XFS_TRANS_PERM_LOG_RES,
660 XFS_WRITE_LOG_COUNT);
661 if (error) {
1da177e4
LT
662 cancelflags = 0;
663 goto error0;
664 }
665 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
666 }
667
668 /*
669 * Read it from disk; xfs_dqread() takes care of
670 * all the necessary initialization of dquot's fields (locks, etc)
671 */
efa092f3 672 if ((error = xfs_qm_dqread(&tp, id, dqp, flags))) {
1da177e4
LT
673 /*
674 * This can happen if quotas got turned off (ESRCH),
675 * or if the dquot didn't exist on disk and we ask to
676 * allocate (ENOENT).
677 */
0b1b213f 678 trace_xfs_dqread_fail(dqp);
1da177e4
LT
679 cancelflags |= XFS_TRANS_ABORT;
680 goto error0;
681 }
682 if (tp) {
1c72bf90 683 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES)))
1da177e4
LT
684 goto error1;
685 }
686
687 *O_dqpp = dqp;
688 return (0);
689
690 error0:
691 ASSERT(error);
692 if (tp)
693 xfs_trans_cancel(tp, cancelflags);
694 error1:
695 xfs_qm_dqdestroy(dqp);
696 *O_dqpp = NULL;
697 return (error);
698}
699
700/*
701 * Lookup a dquot in the incore dquot hashtable. We keep two separate
702 * hashtables for user and group dquots; and, these are global tables
703 * inside the XQM, not per-filesystem tables.
704 * The hash chain must be locked by caller, and it is left locked
705 * on return. Returning dquot is locked.
706 */
707STATIC int
708xfs_qm_dqlookup(
709 xfs_mount_t *mp,
710 xfs_dqid_t id,
711 xfs_dqhash_t *qh,
712 xfs_dquot_t **O_dqpp)
713{
714 xfs_dquot_t *dqp;
715 uint flist_locked;
1da177e4 716
c9a192dc 717 ASSERT(mutex_is_locked(&qh->qh_lock));
1da177e4
LT
718
719 flist_locked = B_FALSE;
720
721 /*
722 * Traverse the hashchain looking for a match
723 */
e6a81f13 724 list_for_each_entry(dqp, &qh->qh_list, q_hashlist) {
1da177e4
LT
725 /*
726 * We already have the hashlock. We don't need the
727 * dqlock to look at the id field of the dquot, since the
728 * id can't be modified without the hashlock anyway.
729 */
1149d96a 730 if (be32_to_cpu(dqp->q_core.d_id) == id && dqp->q_mount == mp) {
0b1b213f
CH
731 trace_xfs_dqlookup_found(dqp);
732
1da177e4
LT
733 /*
734 * All in core dquots must be on the dqlist of mp
735 */
3a25404b 736 ASSERT(!list_empty(&dqp->q_mplist));
1da177e4
LT
737
738 xfs_dqlock(dqp);
739 if (dqp->q_nrefs == 0) {
3a8406f6
DC
740 ASSERT(!list_empty(&dqp->q_freelist));
741 if (!mutex_trylock(&xfs_Gqm->qm_dqfrlist_lock)) {
0b1b213f 742 trace_xfs_dqlookup_want(dqp);
1da177e4
LT
743
744 /*
745 * We may have raced with dqreclaim_one()
746 * (and lost). So, flag that we don't
747 * want the dquot to be reclaimed.
748 */
749 dqp->dq_flags |= XFS_DQ_WANT;
750 xfs_dqunlock(dqp);
3a8406f6 751 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
752 xfs_dqlock(dqp);
753 dqp->dq_flags &= ~(XFS_DQ_WANT);
754 }
755 flist_locked = B_TRUE;
756 }
757
758 /*
759 * id couldn't have changed; we had the hashlock all
760 * along
761 */
1149d96a 762 ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);
1da177e4
LT
763
764 if (flist_locked) {
765 if (dqp->q_nrefs != 0) {
3a8406f6 766 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
767 flist_locked = B_FALSE;
768 } else {
3a8406f6 769 /* take it off the freelist */
0b1b213f 770 trace_xfs_dqlookup_freelist(dqp);
3a8406f6
DC
771 list_del_init(&dqp->q_freelist);
772 xfs_Gqm->qm_dqfrlist_cnt--;
1da177e4
LT
773 }
774 }
775
1da177e4
LT
776 XFS_DQHOLD(dqp);
777
778 if (flist_locked)
3a8406f6 779 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
780 /*
781 * move the dquot to the front of the hashchain
782 */
c9a192dc 783 ASSERT(mutex_is_locked(&qh->qh_lock));
e6a81f13 784 list_move(&dqp->q_hashlist, &qh->qh_list);
0b1b213f 785 trace_xfs_dqlookup_done(dqp);
1da177e4 786 *O_dqpp = dqp;
e6a81f13 787 return 0;
1da177e4
LT
788 }
789 }
790
791 *O_dqpp = NULL;
c9a192dc 792 ASSERT(mutex_is_locked(&qh->qh_lock));
1da177e4
LT
793 return (1);
794}
795
796/*
797 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
798 * a locked dquot, doing an allocation (if requested) as needed.
799 * When both an inode and an id are given, the inode's id takes precedence.
800 * That is, if the id changes while we don't hold the ilock inside this
801 * function, the new dquot is returned, not necessarily the one requested
802 * in the id argument.
803 */
804int
805xfs_qm_dqget(
806 xfs_mount_t *mp,
807 xfs_inode_t *ip, /* locked inode (optional) */
c8ad20ff
NS
808 xfs_dqid_t id, /* uid/projid/gid depending on type */
809 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
1da177e4
LT
810 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
811 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
812{
813 xfs_dquot_t *dqp;
814 xfs_dqhash_t *h;
815 uint version;
816 int error;
817
818 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
819 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
c8ad20ff 820 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
1da177e4
LT
821 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
822 return (ESRCH);
823 }
824 h = XFS_DQ_HASH(mp, id, type);
825
826#ifdef DEBUG
827 if (xfs_do_dqerror) {
828 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
829 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
0b932ccc 830 xfs_debug(mp, "Returning error in dqget");
1da177e4
LT
831 return (EIO);
832 }
833 }
834#endif
835
836 again:
837
838#ifdef DEBUG
c8ad20ff
NS
839 ASSERT(type == XFS_DQ_USER ||
840 type == XFS_DQ_PROJ ||
841 type == XFS_DQ_GROUP);
1da177e4 842 if (ip) {
579aa9ca 843 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4
LT
844 if (type == XFS_DQ_USER)
845 ASSERT(ip->i_udquot == NULL);
846 else
847 ASSERT(ip->i_gdquot == NULL);
848 }
849#endif
c9a192dc 850 mutex_lock(&h->qh_lock);
1da177e4
LT
851
852 /*
853 * Look in the cache (hashtable).
854 * The chain is kept locked during lookup.
855 */
856 if (xfs_qm_dqlookup(mp, id, h, O_dqpp) == 0) {
857 XQM_STATS_INC(xqmstats.xs_qm_dqcachehits);
858 /*
859 * The dquot was found, moved to the front of the chain,
860 * taken off the freelist if it was on it, and locked
861 * at this point. Just unlock the hashchain and return.
862 */
863 ASSERT(*O_dqpp);
864 ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp));
c9a192dc 865 mutex_unlock(&h->qh_lock);
0b1b213f 866 trace_xfs_dqget_hit(*O_dqpp);
1da177e4
LT
867 return (0); /* success */
868 }
869 XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses);
870
871 /*
872 * Dquot cache miss. We don't want to keep the inode lock across
873 * a (potential) disk read. Also we don't want to deal with the lock
874 * ordering between quotainode and this inode. OTOH, dropping the inode
875 * lock here means dealing with a chown that can happen before
876 * we re-acquire the lock.
877 */
878 if (ip)
879 xfs_iunlock(ip, XFS_ILOCK_EXCL);
880 /*
881 * Save the hashchain version stamp, and unlock the chain, so that
882 * we don't keep the lock across a disk read
883 */
884 version = h->qh_version;
c9a192dc 885 mutex_unlock(&h->qh_lock);
1da177e4
LT
886
887 /*
888 * Allocate the dquot on the kernel heap, and read the ondisk
889 * portion off the disk. Also, do all the necessary initialization
890 * This can return ENOENT if dquot didn't exist on disk and we didn't
891 * ask it to allocate; ESRCH if quotas got turned off suddenly.
892 */
893 if ((error = xfs_qm_idtodq(mp, id, type,
894 flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR|
895 XFS_QMOPT_DOWARN),
896 &dqp))) {
897 if (ip)
898 xfs_ilock(ip, XFS_ILOCK_EXCL);
899 return (error);
900 }
901
902 /*
903 * See if this is mount code calling to look at the overall quota limits
904 * which are stored in the id == 0 user or group's dquot.
905 * Since we may not have done a quotacheck by this point, just return
906 * the dquot without attaching it to any hashtables, lists, etc, or even
907 * taking a reference.
908 * The caller must dqdestroy this once done.
909 */
910 if (flags & XFS_QMOPT_DQSUSER) {
911 ASSERT(id == 0);
912 ASSERT(! ip);
913 goto dqret;
914 }
915
916 /*
917 * Dquot lock comes after hashlock in the lock ordering
918 */
919 if (ip) {
920 xfs_ilock(ip, XFS_ILOCK_EXCL);
191f8488 921
1da177e4
LT
922 /*
923 * A dquot could be attached to this inode by now, since
924 * we had dropped the ilock.
925 */
926 if (type == XFS_DQ_USER) {
191f8488
CH
927 if (!XFS_IS_UQUOTA_ON(mp)) {
928 /* inode stays locked on return */
929 xfs_qm_dqdestroy(dqp);
930 return XFS_ERROR(ESRCH);
931 }
1da177e4
LT
932 if (ip->i_udquot) {
933 xfs_qm_dqdestroy(dqp);
934 dqp = ip->i_udquot;
935 xfs_dqlock(dqp);
936 goto dqret;
937 }
938 } else {
191f8488
CH
939 if (!XFS_IS_OQUOTA_ON(mp)) {
940 /* inode stays locked on return */
941 xfs_qm_dqdestroy(dqp);
942 return XFS_ERROR(ESRCH);
943 }
1da177e4
LT
944 if (ip->i_gdquot) {
945 xfs_qm_dqdestroy(dqp);
946 dqp = ip->i_gdquot;
947 xfs_dqlock(dqp);
948 goto dqret;
949 }
950 }
951 }
952
953 /*
954 * Hashlock comes after ilock in lock order
955 */
c9a192dc 956 mutex_lock(&h->qh_lock);
1da177e4
LT
957 if (version != h->qh_version) {
958 xfs_dquot_t *tmpdqp;
959 /*
960 * Now, see if somebody else put the dquot in the
961 * hashtable before us. This can happen because we didn't
962 * keep the hashchain lock. We don't have to worry about
963 * lock order between the two dquots here since dqp isn't
964 * on any findable lists yet.
965 */
966 if (xfs_qm_dqlookup(mp, id, h, &tmpdqp) == 0) {
967 /*
968 * Duplicate found. Just throw away the new dquot
969 * and start over.
970 */
971 xfs_qm_dqput(tmpdqp);
c9a192dc 972 mutex_unlock(&h->qh_lock);
1da177e4
LT
973 xfs_qm_dqdestroy(dqp);
974 XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
975 goto again;
976 }
977 }
978
979 /*
980 * Put the dquot at the beginning of the hash-chain and mp's list
981 * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock ..
982 */
c9a192dc 983 ASSERT(mutex_is_locked(&h->qh_lock));
1da177e4 984 dqp->q_hash = h;
e6a81f13
DC
985 list_add(&dqp->q_hashlist, &h->qh_list);
986 h->qh_version++;
1da177e4
LT
987
988 /*
989 * Attach this dquot to this filesystem's list of all dquots,
990 * kept inside the mount structure in m_quotainfo field
991 */
3a25404b 992 mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
1da177e4
LT
993
994 /*
995 * We return a locked dquot to the caller, with a reference taken
996 */
997 xfs_dqlock(dqp);
998 dqp->q_nrefs = 1;
999
3a25404b
DC
1000 list_add(&dqp->q_mplist, &mp->m_quotainfo->qi_dqlist);
1001 mp->m_quotainfo->qi_dquots++;
1002 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
c9a192dc 1003 mutex_unlock(&h->qh_lock);
1da177e4 1004 dqret:
579aa9ca 1005 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
0b1b213f 1006 trace_xfs_dqget_miss(dqp);
1da177e4
LT
1007 *O_dqpp = dqp;
1008 return (0);
1009}
1010
1011
1012/*
1013 * Release a reference to the dquot (decrement ref-count)
1014 * and unlock it. If there is a group quota attached to this
1015 * dquot, carefully release that too without tripping over
1016 * deadlocks'n'stuff.
1017 */
1018void
1019xfs_qm_dqput(
1020 xfs_dquot_t *dqp)
1021{
1022 xfs_dquot_t *gdqp;
1023
1024 ASSERT(dqp->q_nrefs > 0);
1025 ASSERT(XFS_DQ_IS_LOCKED(dqp));
0b1b213f
CH
1026
1027 trace_xfs_dqput(dqp);
1da177e4
LT
1028
1029 if (dqp->q_nrefs != 1) {
1030 dqp->q_nrefs--;
1031 xfs_dqunlock(dqp);
1032 return;
1033 }
1034
1035 /*
1036 * drop the dqlock and acquire the freelist and dqlock
1037 * in the right order; but try to get it out-of-order first
1038 */
3a8406f6 1039 if (!mutex_trylock(&xfs_Gqm->qm_dqfrlist_lock)) {
0b1b213f 1040 trace_xfs_dqput_wait(dqp);
1da177e4 1041 xfs_dqunlock(dqp);
3a8406f6 1042 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
1043 xfs_dqlock(dqp);
1044 }
1045
1046 while (1) {
1047 gdqp = NULL;
1048
1049 /* We can't depend on nrefs being == 1 here */
1050 if (--dqp->q_nrefs == 0) {
0b1b213f
CH
1051 trace_xfs_dqput_free(dqp);
1052
3a8406f6
DC
1053 list_add_tail(&dqp->q_freelist, &xfs_Gqm->qm_dqfrlist);
1054 xfs_Gqm->qm_dqfrlist_cnt++;
1da177e4
LT
1055
1056 /*
1057 * If we just added a udquot to the freelist, then
1058 * we want to release the gdquot reference that
1059 * it (probably) has. Otherwise it'll keep the
1060 * gdquot from getting reclaimed.
1061 */
1062 if ((gdqp = dqp->q_gdquot)) {
1063 /*
1064 * Avoid a recursive dqput call
1065 */
1066 xfs_dqlock(gdqp);
1067 dqp->q_gdquot = NULL;
1068 }
1da177e4
LT
1069 }
1070 xfs_dqunlock(dqp);
1071
1072 /*
1073 * If we had a group quota inside the user quota as a hint,
1074 * release it now.
1075 */
1076 if (! gdqp)
1077 break;
1078 dqp = gdqp;
1079 }
3a8406f6 1080 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
1081}
1082
1083/*
1084 * Release a dquot. Flush it if dirty, then dqput() it.
1085 * dquot must not be locked.
1086 */
1087void
1088xfs_qm_dqrele(
1089 xfs_dquot_t *dqp)
1090{
7d095257
CH
1091 if (!dqp)
1092 return;
1093
0b1b213f 1094 trace_xfs_dqrele(dqp);
1da177e4
LT
1095
1096 xfs_dqlock(dqp);
1097 /*
1098 * We don't care to flush it if the dquot is dirty here.
1099 * That will create stutters that we want to avoid.
1100 * Instead we do a delayed write when we try to reclaim
1101 * a dirty dquot. Also xfs_sync will take part of the burden...
1102 */
1103 xfs_qm_dqput(dqp);
1104}
1105
ca30b2a7
CH
1106/*
1107 * This is the dquot flushing I/O completion routine. It is called
1108 * from interrupt level when the buffer containing the dquot is
1109 * flushed to disk. It is responsible for removing the dquot logitem
1110 * from the AIL if it has not been re-logged, and unlocking the dquot's
1111 * flush lock. This behavior is very similar to that of inodes..
1112 */
1113STATIC void
1114xfs_qm_dqflush_done(
1115 struct xfs_buf *bp,
1116 struct xfs_log_item *lip)
1117{
1118 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
1119 xfs_dquot_t *dqp = qip->qli_dquot;
1120 struct xfs_ail *ailp = lip->li_ailp;
1121
1122 /*
1123 * We only want to pull the item from the AIL if its
1124 * location in the log has not changed since we started the flush.
1125 * Thus, we only bother if the dquot's lsn has
1126 * not changed. First we check the lsn outside the lock
1127 * since it's cheaper, and then we recheck while
1128 * holding the lock before removing the dquot from the AIL.
1129 */
1130 if ((lip->li_flags & XFS_LI_IN_AIL) &&
1131 lip->li_lsn == qip->qli_flush_lsn) {
1132
1133 /* xfs_trans_ail_delete() drops the AIL lock. */
1134 spin_lock(&ailp->xa_lock);
1135 if (lip->li_lsn == qip->qli_flush_lsn)
1136 xfs_trans_ail_delete(ailp, lip);
1137 else
1138 spin_unlock(&ailp->xa_lock);
1139 }
1140
1141 /*
1142 * Release the dq's flush lock since we're done with it.
1143 */
1144 xfs_dqfunlock(dqp);
1145}
1da177e4
LT
1146
1147/*
1148 * Write a modified dquot to disk.
1149 * The dquot must be locked and the flush lock too taken by caller.
1150 * The flush lock will not be unlocked until the dquot reaches the disk,
1151 * but the dquot is free to be unlocked and modified by the caller
1152 * in the interim. Dquot is still locked on return. This behavior is
1153 * identical to that of inodes.
1154 */
1155int
1156xfs_qm_dqflush(
1157 xfs_dquot_t *dqp,
1158 uint flags)
1159{
acecf1b5
CH
1160 struct xfs_mount *mp = dqp->q_mount;
1161 struct xfs_buf *bp;
1162 struct xfs_disk_dquot *ddqp;
1da177e4 1163 int error;
1da177e4
LT
1164
1165 ASSERT(XFS_DQ_IS_LOCKED(dqp));
e1f49cf2 1166 ASSERT(!completion_done(&dqp->q_flush));
acecf1b5 1167
0b1b213f 1168 trace_xfs_dqflush(dqp);
1da177e4
LT
1169
1170 /*
acecf1b5 1171 * If not dirty, or it's pinned and we are not supposed to block, nada.
1da177e4 1172 */
2f8a3ce1 1173 if (!XFS_DQ_IS_DIRTY(dqp) ||
20026d92 1174 (!(flags & SYNC_WAIT) && atomic_read(&dqp->q_pincount) > 0)) {
1da177e4 1175 xfs_dqfunlock(dqp);
2f8a3ce1 1176 return 0;
1da177e4 1177 }
1da177e4
LT
1178 xfs_qm_dqunpin_wait(dqp);
1179
1180 /*
1181 * This may have been unpinned because the filesystem is shutting
1182 * down forcibly. If that's the case we must not write this dquot
1183 * to disk, because the log record didn't make it to disk!
1184 */
acecf1b5
CH
1185 if (XFS_FORCED_SHUTDOWN(mp)) {
1186 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1da177e4
LT
1187 xfs_dqfunlock(dqp);
1188 return XFS_ERROR(EIO);
1189 }
1190
1191 /*
1192 * Get the buffer containing the on-disk dquot
1da177e4 1193 */
acecf1b5
CH
1194 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1195 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
1196 if (error) {
1da177e4 1197 ASSERT(error != ENOENT);
1da177e4 1198 xfs_dqfunlock(dqp);
acecf1b5 1199 return error;
1da177e4
LT
1200 }
1201
acecf1b5
CH
1202 /*
1203 * Calculate the location of the dquot inside the buffer.
1204 */
62926044 1205 ddqp = bp->b_addr + dqp->q_bufoffset;
acecf1b5
CH
1206
1207 /*
1208 * A simple sanity check in case we got a corrupted dquot..
1209 */
a0fa2b67
DC
1210 error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
1211 XFS_QMOPT_DOWARN, "dqflush (incore copy)");
1212 if (error) {
acecf1b5
CH
1213 xfs_buf_relse(bp);
1214 xfs_dqfunlock(dqp);
1215 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1da177e4
LT
1216 return XFS_ERROR(EIO);
1217 }
1218
1219 /* This is the only portion of data that needs to persist */
acecf1b5 1220 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
1da177e4
LT
1221
1222 /*
1223 * Clear the dirty field and remember the flush lsn for later use.
1224 */
acecf1b5 1225 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1da177e4 1226
7b2e2a31
DC
1227 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1228 &dqp->q_logitem.qli_item.li_lsn);
1da177e4
LT
1229
1230 /*
1231 * Attach an iodone routine so that we can remove this dquot from the
1232 * AIL and release the flush lock once the dquot is synced to disk.
1233 */
ca30b2a7
CH
1234 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1235 &dqp->q_logitem.qli_item);
1236
1da177e4
LT
1237 /*
1238 * If the buffer is pinned then push on the log so we won't
1239 * get stuck waiting in the write for too long.
1240 */
811e64c7 1241 if (xfs_buf_ispinned(bp)) {
0b1b213f 1242 trace_xfs_dqflush_force(dqp);
a14a348b 1243 xfs_log_force(mp, 0);
1da177e4
LT
1244 }
1245
20026d92 1246 if (flags & SYNC_WAIT)
c2b006c1
CH
1247 error = xfs_bwrite(bp);
1248 else
61551f1e 1249 xfs_buf_delwri_queue(bp);
c2b006c1
CH
1250
1251 xfs_buf_relse(bp);
0b1b213f
CH
1252
1253 trace_xfs_dqflush_done(dqp);
1254
1da177e4
LT
1255 /*
1256 * dqp is still locked, but caller is free to unlock it now.
1257 */
20026d92 1258 return error;
1da177e4
LT
1259
1260}
1261
1da177e4
LT
1262int
1263xfs_qm_dqlock_nowait(
1264 xfs_dquot_t *dqp)
1265{
e1f49cf2 1266 return mutex_trylock(&dqp->q_qlock);
1da177e4
LT
1267}
1268
1269void
1270xfs_dqlock(
1271 xfs_dquot_t *dqp)
1272{
e1f49cf2 1273 mutex_lock(&dqp->q_qlock);
1da177e4
LT
1274}
1275
1276void
1277xfs_dqunlock(
1278 xfs_dquot_t *dqp)
1279{
1280 mutex_unlock(&(dqp->q_qlock));
1281 if (dqp->q_logitem.qli_dquot == dqp) {
1282 /* Once was dqp->q_mount, but might just have been cleared */
783a2f65 1283 xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_ailp,
1da177e4
LT
1284 (xfs_log_item_t*)&(dqp->q_logitem));
1285 }
1286}
1287
1288
1289void
1290xfs_dqunlock_nonotify(
1291 xfs_dquot_t *dqp)
1292{
1293 mutex_unlock(&(dqp->q_qlock));
1294}
1295
5bb87a33
CH
1296/*
1297 * Lock two xfs_dquot structures.
1298 *
1299 * To avoid deadlocks we always lock the quota structure with
1300 * the lowerd id first.
1301 */
1da177e4
LT
1302void
1303xfs_dqlock2(
1304 xfs_dquot_t *d1,
1305 xfs_dquot_t *d2)
1306{
1307 if (d1 && d2) {
1308 ASSERT(d1 != d2);
1149d96a
CH
1309 if (be32_to_cpu(d1->q_core.d_id) >
1310 be32_to_cpu(d2->q_core.d_id)) {
5bb87a33
CH
1311 mutex_lock(&d2->q_qlock);
1312 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1da177e4 1313 } else {
5bb87a33
CH
1314 mutex_lock(&d1->q_qlock);
1315 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1da177e4 1316 }
5bb87a33
CH
1317 } else if (d1) {
1318 mutex_lock(&d1->q_qlock);
1319 } else if (d2) {
1320 mutex_lock(&d2->q_qlock);
1da177e4
LT
1321 }
1322}
1323
1324
1325/*
1326 * Take a dquot out of the mount's dqlist as well as the hashlist.
1327 * This is called via unmount as well as quotaoff, and the purge
1328 * will always succeed unless there are soft (temp) references
1329 * outstanding.
1330 *
1331 * This returns 0 if it was purged, 1 if it wasn't. It's not an error code
1332 * that we're returning! XXXsup - not cool.
1333 */
1334/* ARGSUSED */
1335int
1336xfs_qm_dqpurge(
4f0e8a98 1337 xfs_dquot_t *dqp)
1da177e4 1338{
e6a81f13 1339 xfs_dqhash_t *qh = dqp->q_hash;
3c56836f 1340 xfs_mount_t *mp = dqp->q_mount;
1da177e4 1341
3a25404b 1342 ASSERT(mutex_is_locked(&mp->m_quotainfo->qi_dqlist_lock));
c9a192dc 1343 ASSERT(mutex_is_locked(&dqp->q_hash->qh_lock));
1da177e4
LT
1344
1345 xfs_dqlock(dqp);
1346 /*
1347 * We really can't afford to purge a dquot that is
1348 * referenced, because these are hard refs.
1349 * It shouldn't happen in general because we went thru _all_ inodes in
1350 * dqrele_all_inodes before calling this and didn't let the mountlock go.
1351 * However it is possible that we have dquots with temporary
1352 * references that are not attached to an inode. e.g. see xfs_setattr().
1353 */
1354 if (dqp->q_nrefs != 0) {
1355 xfs_dqunlock(dqp);
c9a192dc 1356 mutex_unlock(&dqp->q_hash->qh_lock);
1da177e4
LT
1357 return (1);
1358 }
1359
3a8406f6 1360 ASSERT(!list_empty(&dqp->q_freelist));
1da177e4
LT
1361
1362 /*
1363 * If we're turning off quotas, we have to make sure that, for
1364 * example, we don't delete quota disk blocks while dquots are
1365 * in the process of getting written to those disk blocks.
1366 * This dquot might well be on AIL, and we can't leave it there
1367 * if we're turning off quotas. Basically, we need this flush
1368 * lock, and are willing to block on it.
1369 */
e1f49cf2 1370 if (!xfs_dqflock_nowait(dqp)) {
1da177e4
LT
1371 /*
1372 * Block on the flush lock after nudging dquot buffer,
1373 * if it is incore.
1374 */
1375 xfs_qm_dqflock_pushbuf_wait(dqp);
1376 }
1377
1378 /*
1379 * XXXIf we're turning this type of quotas off, we don't care
1380 * about the dirty metadata sitting in this dquot. OTOH, if
1381 * we're unmounting, we do care, so we flush it and wait.
1382 */
1383 if (XFS_DQ_IS_DIRTY(dqp)) {
3c56836f 1384 int error;
0b1b213f 1385
1da177e4
LT
1386 /* dqflush unlocks dqflock */
1387 /*
1388 * Given that dqpurge is a very rare occurrence, it is OK
1389 * that we're holding the hashlist and mplist locks
1390 * across the disk write. But, ... XXXsup
1391 *
1392 * We don't care about getting disk errors here. We need
1393 * to purge this dquot anyway, so we go ahead regardless.
1394 */
20026d92 1395 error = xfs_qm_dqflush(dqp, SYNC_WAIT);
3c56836f 1396 if (error)
53487786
DC
1397 xfs_warn(mp, "%s: dquot %p flush failed",
1398 __func__, dqp);
1da177e4
LT
1399 xfs_dqflock(dqp);
1400 }
bc3048e3 1401 ASSERT(atomic_read(&dqp->q_pincount) == 0);
1da177e4
LT
1402 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1403 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
1404
e6a81f13
DC
1405 list_del_init(&dqp->q_hashlist);
1406 qh->qh_version++;
3a25404b
DC
1407 list_del_init(&dqp->q_mplist);
1408 mp->m_quotainfo->qi_dqreclaims++;
1409 mp->m_quotainfo->qi_dquots--;
1da177e4
LT
1410 /*
1411 * XXX Move this to the front of the freelist, if we can get the
1412 * freelist lock.
1413 */
3a8406f6 1414 ASSERT(!list_empty(&dqp->q_freelist));
1da177e4
LT
1415
1416 dqp->q_mount = NULL;
1417 dqp->q_hash = NULL;
1418 dqp->dq_flags = XFS_DQ_INACTIVE;
1419 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
1420 xfs_dqfunlock(dqp);
1421 xfs_dqunlock(dqp);
e6a81f13 1422 mutex_unlock(&qh->qh_lock);
1da177e4
LT
1423 return (0);
1424}
1425
1426
1da177e4
LT
1427/*
1428 * Give the buffer a little push if it is incore and
1429 * wait on the flush lock.
1430 */
1431void
1432xfs_qm_dqflock_pushbuf_wait(
1433 xfs_dquot_t *dqp)
1434{
8a7b8a89 1435 xfs_mount_t *mp = dqp->q_mount;
1da177e4
LT
1436 xfs_buf_t *bp;
1437
1438 /*
1439 * Check to see if the dquot has been flushed delayed
1440 * write. If so, grab its buffer and send it
1441 * out immediately. We'll be able to acquire
1442 * the flush lock when the I/O completes.
1443 */
8a7b8a89
CH
1444 bp = xfs_incore(mp->m_ddev_targp, dqp->q_blkno,
1445 mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK);
7d6a7bde
DC
1446 if (!bp)
1447 goto out_lock;
1448
1449 if (XFS_BUF_ISDELAYWRITE(bp)) {
811e64c7 1450 if (xfs_buf_ispinned(bp))
8a7b8a89 1451 xfs_log_force(mp, 0);
7d6a7bde
DC
1452 xfs_buf_delwri_promote(bp);
1453 wake_up_process(bp->b_target->bt_task);
1da177e4 1454 }
7d6a7bde
DC
1455 xfs_buf_relse(bp);
1456out_lock:
1da177e4
LT
1457 xfs_dqflock(dqp);
1458}