]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/xfs/xfs_dquot.c
xfs: unlock the inode before log force in xfs_change_file_space
[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 381 nmaps = 1;
c0dc7828
DC
382 error = xfs_bmapi_write(tp, quotip, offset_fsb,
383 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
384 &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp),
385 &map, &nmaps, &flist);
386 if (error)
1da177e4 387 goto error0;
1da177e4
LT
388 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
389 ASSERT(nmaps == 1);
390 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
391 (map.br_startblock != HOLESTARTBLOCK));
392
393 /*
394 * Keep track of the blkno to save a lookup later
395 */
396 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
397
398 /* now we can just get the buffer (there's nothing to read yet) */
399 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
400 dqp->q_blkno,
8a7b8a89 401 mp->m_quotainfo->qi_dqchunklen,
1da177e4 402 0);
2a30f36d
CS
403
404 error = xfs_buf_geterror(bp);
405 if (error)
1da177e4 406 goto error1;
2a30f36d 407
1da177e4
LT
408 /*
409 * Make a chunk of dquots out of this buffer and log
410 * the entire thing.
411 */
1149d96a 412 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
c8ad20ff 413 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
1da177e4 414
efa092f3
TS
415 /*
416 * xfs_bmap_finish() may commit the current transaction and
417 * start a second transaction if the freelist is not empty.
418 *
419 * Since we still want to modify this buffer, we need to
420 * ensure that the buffer is not released on commit of
421 * the first transaction and ensure the buffer is added to the
422 * second transaction.
423 *
424 * If there is only one transaction then don't stop the buffer
425 * from being released when it commits later on.
426 */
427
428 xfs_trans_bhold(tp, bp);
429
f7c99b6f 430 if ((error = xfs_bmap_finish(tpp, &flist, &committed))) {
1da177e4
LT
431 goto error1;
432 }
433
efa092f3
TS
434 if (committed) {
435 tp = *tpp;
436 xfs_trans_bjoin(tp, bp);
437 } else {
438 xfs_trans_bhold_release(tp, bp);
439 }
440
1da177e4
LT
441 *O_bpp = bp;
442 return 0;
443
444 error1:
445 xfs_bmap_cancel(&flist);
446 error0:
447 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
448
449 return (error);
450}
451
452/*
453 * Maps a dquot to the buffer containing its on-disk version.
454 * This returns a ptr to the buffer containing the on-disk dquot
455 * in the bpp param, and a ptr to the on-disk dquot within that buffer
456 */
457STATIC int
458xfs_qm_dqtobp(
efa092f3 459 xfs_trans_t **tpp,
1da177e4
LT
460 xfs_dquot_t *dqp,
461 xfs_disk_dquot_t **O_ddpp,
462 xfs_buf_t **O_bpp,
463 uint flags)
464{
465 xfs_bmbt_irec_t map;
acecf1b5 466 int nmaps = 1, error;
1da177e4 467 xfs_buf_t *bp;
acecf1b5
CH
468 xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp);
469 xfs_mount_t *mp = dqp->q_mount;
1da177e4 470 xfs_disk_dquot_t *ddq;
acecf1b5 471 xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id);
efa092f3 472 xfs_trans_t *tp = (tpp ? *tpp : NULL);
1da177e4 473
acecf1b5 474 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
1da177e4 475
acecf1b5
CH
476 xfs_ilock(quotip, XFS_ILOCK_SHARED);
477 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
1da177e4 478 /*
acecf1b5
CH
479 * Return if this type of quotas is turned off while we
480 * didn't have the quota inode lock.
1da177e4 481 */
acecf1b5
CH
482 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
483 return ESRCH;
484 }
485
486 /*
487 * Find the block map; no allocations yet
488 */
5c8ed202
DC
489 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
490 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
acecf1b5
CH
491
492 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
493 if (error)
494 return error;
495
496 ASSERT(nmaps == 1);
497 ASSERT(map.br_blockcount == 1);
498
499 /*
500 * Offset of dquot in the (fixed sized) dquot chunk.
501 */
502 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
503 sizeof(xfs_dqblk_t);
504
505 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
506 if (map.br_startblock == HOLESTARTBLOCK) {
1da177e4 507 /*
acecf1b5 508 * We don't allocate unless we're asked to
1da177e4 509 */
acecf1b5
CH
510 if (!(flags & XFS_QMOPT_DQALLOC))
511 return ENOENT;
1da177e4 512
acecf1b5
CH
513 ASSERT(tp);
514 error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
515 dqp->q_fileoffset, &bp);
1da177e4 516 if (error)
acecf1b5
CH
517 return error;
518 tp = *tpp;
519 } else {
520 trace_xfs_dqtobp_read(dqp);
1da177e4
LT
521
522 /*
acecf1b5
CH
523 * store the blkno etc so that we don't have to do the
524 * mapping all the time
1da177e4 525 */
acecf1b5 526 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
0b1b213f 527
8a7b8a89
CH
528 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
529 dqp->q_blkno,
530 mp->m_quotainfo->qi_dqchunklen,
531 0, &bp);
1da177e4
LT
532 if (error || !bp)
533 return XFS_ERROR(error);
534 }
acecf1b5 535
0c842ad4 536 ASSERT(xfs_buf_islocked(bp));
1da177e4
LT
537
538 /*
539 * calculate the location of the dquot inside the buffer.
540 */
62926044 541 ddq = bp->b_addr + dqp->q_bufoffset;
1da177e4
LT
542
543 /*
544 * A simple sanity check in case we got a corrupted dquot...
545 */
a0fa2b67 546 error = xfs_qm_dqcheck(mp, ddq, id, dqp->dq_flags & XFS_DQ_ALLTYPES,
1da177e4 547 flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN),
a0fa2b67
DC
548 "dqtobp");
549 if (error) {
1da177e4
LT
550 if (!(flags & XFS_QMOPT_DQREPAIR)) {
551 xfs_trans_brelse(tp, bp);
552 return XFS_ERROR(EIO);
553 }
1da177e4
LT
554 }
555
556 *O_bpp = bp;
557 *O_ddpp = ddq;
558
559 return (0);
560}
561
562
563/*
564 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
565 * and release the buffer immediately.
566 *
567 */
568/* ARGSUSED */
569STATIC int
570xfs_qm_dqread(
efa092f3 571 xfs_trans_t **tpp,
1da177e4
LT
572 xfs_dqid_t id,
573 xfs_dquot_t *dqp, /* dquot to get filled in */
574 uint flags)
575{
576 xfs_disk_dquot_t *ddqp;
577 xfs_buf_t *bp;
578 int error;
efa092f3
TS
579 xfs_trans_t *tp;
580
581 ASSERT(tpp);
1da177e4 582
0b1b213f
CH
583 trace_xfs_dqread(dqp);
584
1da177e4
LT
585 /*
586 * get a pointer to the on-disk dquot and the buffer containing it
587 * dqp already knows its own type (GROUP/USER).
588 */
efa092f3 589 if ((error = xfs_qm_dqtobp(tpp, dqp, &ddqp, &bp, flags))) {
1da177e4
LT
590 return (error);
591 }
efa092f3 592 tp = *tpp;
1da177e4
LT
593
594 /* copy everything from disk dquot to the incore dquot */
595 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
1149d96a 596 ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);
1da177e4
LT
597 xfs_qm_dquot_logitem_init(dqp);
598
599 /*
600 * Reservation counters are defined as reservation plus current usage
25985edc 601 * to avoid having to add every time.
1da177e4 602 */
1149d96a
CH
603 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
604 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
605 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
1da177e4
LT
606
607 /* Mark the buf so that this will stay incore a little longer */
608 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DQUOT, XFS_DQUOT_REF);
609
610 /*
611 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
612 * So we need to release with xfs_trans_brelse().
613 * The strategy here is identical to that of inodes; we lock
614 * the dquot in xfs_qm_dqget() before making it accessible to
615 * others. This is because dquots, like inodes, need a good level of
616 * concurrency, and we don't want to take locks on the entire buffers
617 * for dquot accesses.
618 * Note also that the dquot buffer may even be dirty at this point, if
619 * this particular dquot was repaired. We still aren't afraid to
620 * brelse it because we have the changes incore.
621 */
0c842ad4 622 ASSERT(xfs_buf_islocked(bp));
1da177e4
LT
623 xfs_trans_brelse(tp, bp);
624
625 return (error);
626}
627
628
629/*
630 * allocate an incore dquot from the kernel heap,
631 * and fill its core with quota information kept on disk.
632 * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk
633 * if it wasn't already allocated.
634 */
635STATIC int
636xfs_qm_idtodq(
637 xfs_mount_t *mp,
638 xfs_dqid_t id, /* gid or uid, depending on type */
639 uint type, /* UDQUOT or GDQUOT */
640 uint flags, /* DQALLOC, DQREPAIR */
641 xfs_dquot_t **O_dqpp)/* OUT : incore dquot, not locked */
642{
643 xfs_dquot_t *dqp;
644 int error;
645 xfs_trans_t *tp;
646 int cancelflags=0;
647
648 dqp = xfs_qm_dqinit(mp, id, type);
649 tp = NULL;
650 if (flags & XFS_QMOPT_DQALLOC) {
651 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
8a7b8a89
CH
652 error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp),
653 XFS_WRITE_LOG_RES(mp) +
654 BBTOB(mp->m_quotainfo->qi_dqchunklen) - 1 +
655 128,
656 0,
657 XFS_TRANS_PERM_LOG_RES,
658 XFS_WRITE_LOG_COUNT);
659 if (error) {
1da177e4
LT
660 cancelflags = 0;
661 goto error0;
662 }
663 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
664 }
665
666 /*
667 * Read it from disk; xfs_dqread() takes care of
668 * all the necessary initialization of dquot's fields (locks, etc)
669 */
efa092f3 670 if ((error = xfs_qm_dqread(&tp, id, dqp, flags))) {
1da177e4
LT
671 /*
672 * This can happen if quotas got turned off (ESRCH),
673 * or if the dquot didn't exist on disk and we ask to
674 * allocate (ENOENT).
675 */
0b1b213f 676 trace_xfs_dqread_fail(dqp);
1da177e4
LT
677 cancelflags |= XFS_TRANS_ABORT;
678 goto error0;
679 }
680 if (tp) {
1c72bf90 681 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES)))
1da177e4
LT
682 goto error1;
683 }
684
685 *O_dqpp = dqp;
686 return (0);
687
688 error0:
689 ASSERT(error);
690 if (tp)
691 xfs_trans_cancel(tp, cancelflags);
692 error1:
693 xfs_qm_dqdestroy(dqp);
694 *O_dqpp = NULL;
695 return (error);
696}
697
698/*
699 * Lookup a dquot in the incore dquot hashtable. We keep two separate
700 * hashtables for user and group dquots; and, these are global tables
701 * inside the XQM, not per-filesystem tables.
702 * The hash chain must be locked by caller, and it is left locked
703 * on return. Returning dquot is locked.
704 */
705STATIC int
706xfs_qm_dqlookup(
707 xfs_mount_t *mp,
708 xfs_dqid_t id,
709 xfs_dqhash_t *qh,
710 xfs_dquot_t **O_dqpp)
711{
712 xfs_dquot_t *dqp;
713 uint flist_locked;
1da177e4 714
c9a192dc 715 ASSERT(mutex_is_locked(&qh->qh_lock));
1da177e4
LT
716
717 flist_locked = B_FALSE;
718
719 /*
720 * Traverse the hashchain looking for a match
721 */
e6a81f13 722 list_for_each_entry(dqp, &qh->qh_list, q_hashlist) {
1da177e4
LT
723 /*
724 * We already have the hashlock. We don't need the
725 * dqlock to look at the id field of the dquot, since the
726 * id can't be modified without the hashlock anyway.
727 */
1149d96a 728 if (be32_to_cpu(dqp->q_core.d_id) == id && dqp->q_mount == mp) {
0b1b213f
CH
729 trace_xfs_dqlookup_found(dqp);
730
1da177e4
LT
731 /*
732 * All in core dquots must be on the dqlist of mp
733 */
3a25404b 734 ASSERT(!list_empty(&dqp->q_mplist));
1da177e4
LT
735
736 xfs_dqlock(dqp);
737 if (dqp->q_nrefs == 0) {
3a8406f6
DC
738 ASSERT(!list_empty(&dqp->q_freelist));
739 if (!mutex_trylock(&xfs_Gqm->qm_dqfrlist_lock)) {
0b1b213f 740 trace_xfs_dqlookup_want(dqp);
1da177e4
LT
741
742 /*
743 * We may have raced with dqreclaim_one()
744 * (and lost). So, flag that we don't
745 * want the dquot to be reclaimed.
746 */
747 dqp->dq_flags |= XFS_DQ_WANT;
748 xfs_dqunlock(dqp);
3a8406f6 749 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
750 xfs_dqlock(dqp);
751 dqp->dq_flags &= ~(XFS_DQ_WANT);
752 }
753 flist_locked = B_TRUE;
754 }
755
756 /*
757 * id couldn't have changed; we had the hashlock all
758 * along
759 */
1149d96a 760 ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);
1da177e4
LT
761
762 if (flist_locked) {
763 if (dqp->q_nrefs != 0) {
3a8406f6 764 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
765 flist_locked = B_FALSE;
766 } else {
3a8406f6 767 /* take it off the freelist */
0b1b213f 768 trace_xfs_dqlookup_freelist(dqp);
3a8406f6
DC
769 list_del_init(&dqp->q_freelist);
770 xfs_Gqm->qm_dqfrlist_cnt--;
1da177e4
LT
771 }
772 }
773
1da177e4
LT
774 XFS_DQHOLD(dqp);
775
776 if (flist_locked)
3a8406f6 777 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
778 /*
779 * move the dquot to the front of the hashchain
780 */
c9a192dc 781 ASSERT(mutex_is_locked(&qh->qh_lock));
e6a81f13 782 list_move(&dqp->q_hashlist, &qh->qh_list);
0b1b213f 783 trace_xfs_dqlookup_done(dqp);
1da177e4 784 *O_dqpp = dqp;
e6a81f13 785 return 0;
1da177e4
LT
786 }
787 }
788
789 *O_dqpp = NULL;
c9a192dc 790 ASSERT(mutex_is_locked(&qh->qh_lock));
1da177e4
LT
791 return (1);
792}
793
794/*
795 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
796 * a locked dquot, doing an allocation (if requested) as needed.
797 * When both an inode and an id are given, the inode's id takes precedence.
798 * That is, if the id changes while we don't hold the ilock inside this
799 * function, the new dquot is returned, not necessarily the one requested
800 * in the id argument.
801 */
802int
803xfs_qm_dqget(
804 xfs_mount_t *mp,
805 xfs_inode_t *ip, /* locked inode (optional) */
c8ad20ff
NS
806 xfs_dqid_t id, /* uid/projid/gid depending on type */
807 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
1da177e4
LT
808 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
809 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
810{
811 xfs_dquot_t *dqp;
812 xfs_dqhash_t *h;
813 uint version;
814 int error;
815
816 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
817 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
c8ad20ff 818 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
1da177e4
LT
819 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
820 return (ESRCH);
821 }
822 h = XFS_DQ_HASH(mp, id, type);
823
824#ifdef DEBUG
825 if (xfs_do_dqerror) {
826 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
827 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
0b932ccc 828 xfs_debug(mp, "Returning error in dqget");
1da177e4
LT
829 return (EIO);
830 }
831 }
832#endif
833
834 again:
835
836#ifdef DEBUG
c8ad20ff
NS
837 ASSERT(type == XFS_DQ_USER ||
838 type == XFS_DQ_PROJ ||
839 type == XFS_DQ_GROUP);
1da177e4 840 if (ip) {
579aa9ca 841 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4
LT
842 if (type == XFS_DQ_USER)
843 ASSERT(ip->i_udquot == NULL);
844 else
845 ASSERT(ip->i_gdquot == NULL);
846 }
847#endif
c9a192dc 848 mutex_lock(&h->qh_lock);
1da177e4
LT
849
850 /*
851 * Look in the cache (hashtable).
852 * The chain is kept locked during lookup.
853 */
854 if (xfs_qm_dqlookup(mp, id, h, O_dqpp) == 0) {
855 XQM_STATS_INC(xqmstats.xs_qm_dqcachehits);
856 /*
857 * The dquot was found, moved to the front of the chain,
858 * taken off the freelist if it was on it, and locked
859 * at this point. Just unlock the hashchain and return.
860 */
861 ASSERT(*O_dqpp);
862 ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp));
c9a192dc 863 mutex_unlock(&h->qh_lock);
0b1b213f 864 trace_xfs_dqget_hit(*O_dqpp);
1da177e4
LT
865 return (0); /* success */
866 }
867 XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses);
868
869 /*
870 * Dquot cache miss. We don't want to keep the inode lock across
871 * a (potential) disk read. Also we don't want to deal with the lock
872 * ordering between quotainode and this inode. OTOH, dropping the inode
873 * lock here means dealing with a chown that can happen before
874 * we re-acquire the lock.
875 */
876 if (ip)
877 xfs_iunlock(ip, XFS_ILOCK_EXCL);
878 /*
879 * Save the hashchain version stamp, and unlock the chain, so that
880 * we don't keep the lock across a disk read
881 */
882 version = h->qh_version;
c9a192dc 883 mutex_unlock(&h->qh_lock);
1da177e4
LT
884
885 /*
886 * Allocate the dquot on the kernel heap, and read the ondisk
887 * portion off the disk. Also, do all the necessary initialization
888 * This can return ENOENT if dquot didn't exist on disk and we didn't
889 * ask it to allocate; ESRCH if quotas got turned off suddenly.
890 */
891 if ((error = xfs_qm_idtodq(mp, id, type,
892 flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR|
893 XFS_QMOPT_DOWARN),
894 &dqp))) {
895 if (ip)
896 xfs_ilock(ip, XFS_ILOCK_EXCL);
897 return (error);
898 }
899
900 /*
901 * See if this is mount code calling to look at the overall quota limits
902 * which are stored in the id == 0 user or group's dquot.
903 * Since we may not have done a quotacheck by this point, just return
904 * the dquot without attaching it to any hashtables, lists, etc, or even
905 * taking a reference.
906 * The caller must dqdestroy this once done.
907 */
908 if (flags & XFS_QMOPT_DQSUSER) {
909 ASSERT(id == 0);
910 ASSERT(! ip);
911 goto dqret;
912 }
913
914 /*
915 * Dquot lock comes after hashlock in the lock ordering
916 */
917 if (ip) {
918 xfs_ilock(ip, XFS_ILOCK_EXCL);
191f8488 919
1da177e4
LT
920 /*
921 * A dquot could be attached to this inode by now, since
922 * we had dropped the ilock.
923 */
924 if (type == XFS_DQ_USER) {
191f8488
CH
925 if (!XFS_IS_UQUOTA_ON(mp)) {
926 /* inode stays locked on return */
927 xfs_qm_dqdestroy(dqp);
928 return XFS_ERROR(ESRCH);
929 }
1da177e4
LT
930 if (ip->i_udquot) {
931 xfs_qm_dqdestroy(dqp);
932 dqp = ip->i_udquot;
933 xfs_dqlock(dqp);
934 goto dqret;
935 }
936 } else {
191f8488
CH
937 if (!XFS_IS_OQUOTA_ON(mp)) {
938 /* inode stays locked on return */
939 xfs_qm_dqdestroy(dqp);
940 return XFS_ERROR(ESRCH);
941 }
1da177e4
LT
942 if (ip->i_gdquot) {
943 xfs_qm_dqdestroy(dqp);
944 dqp = ip->i_gdquot;
945 xfs_dqlock(dqp);
946 goto dqret;
947 }
948 }
949 }
950
951 /*
952 * Hashlock comes after ilock in lock order
953 */
c9a192dc 954 mutex_lock(&h->qh_lock);
1da177e4
LT
955 if (version != h->qh_version) {
956 xfs_dquot_t *tmpdqp;
957 /*
958 * Now, see if somebody else put the dquot in the
959 * hashtable before us. This can happen because we didn't
960 * keep the hashchain lock. We don't have to worry about
961 * lock order between the two dquots here since dqp isn't
962 * on any findable lists yet.
963 */
964 if (xfs_qm_dqlookup(mp, id, h, &tmpdqp) == 0) {
965 /*
966 * Duplicate found. Just throw away the new dquot
967 * and start over.
968 */
969 xfs_qm_dqput(tmpdqp);
c9a192dc 970 mutex_unlock(&h->qh_lock);
1da177e4
LT
971 xfs_qm_dqdestroy(dqp);
972 XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
973 goto again;
974 }
975 }
976
977 /*
978 * Put the dquot at the beginning of the hash-chain and mp's list
979 * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock ..
980 */
c9a192dc 981 ASSERT(mutex_is_locked(&h->qh_lock));
1da177e4 982 dqp->q_hash = h;
e6a81f13
DC
983 list_add(&dqp->q_hashlist, &h->qh_list);
984 h->qh_version++;
1da177e4
LT
985
986 /*
987 * Attach this dquot to this filesystem's list of all dquots,
988 * kept inside the mount structure in m_quotainfo field
989 */
3a25404b 990 mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
1da177e4
LT
991
992 /*
993 * We return a locked dquot to the caller, with a reference taken
994 */
995 xfs_dqlock(dqp);
996 dqp->q_nrefs = 1;
997
3a25404b
DC
998 list_add(&dqp->q_mplist, &mp->m_quotainfo->qi_dqlist);
999 mp->m_quotainfo->qi_dquots++;
1000 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
c9a192dc 1001 mutex_unlock(&h->qh_lock);
1da177e4 1002 dqret:
579aa9ca 1003 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
0b1b213f 1004 trace_xfs_dqget_miss(dqp);
1da177e4
LT
1005 *O_dqpp = dqp;
1006 return (0);
1007}
1008
1009
1010/*
1011 * Release a reference to the dquot (decrement ref-count)
1012 * and unlock it. If there is a group quota attached to this
1013 * dquot, carefully release that too without tripping over
1014 * deadlocks'n'stuff.
1015 */
1016void
1017xfs_qm_dqput(
1018 xfs_dquot_t *dqp)
1019{
1020 xfs_dquot_t *gdqp;
1021
1022 ASSERT(dqp->q_nrefs > 0);
1023 ASSERT(XFS_DQ_IS_LOCKED(dqp));
0b1b213f
CH
1024
1025 trace_xfs_dqput(dqp);
1da177e4
LT
1026
1027 if (dqp->q_nrefs != 1) {
1028 dqp->q_nrefs--;
1029 xfs_dqunlock(dqp);
1030 return;
1031 }
1032
1033 /*
1034 * drop the dqlock and acquire the freelist and dqlock
1035 * in the right order; but try to get it out-of-order first
1036 */
3a8406f6 1037 if (!mutex_trylock(&xfs_Gqm->qm_dqfrlist_lock)) {
0b1b213f 1038 trace_xfs_dqput_wait(dqp);
1da177e4 1039 xfs_dqunlock(dqp);
3a8406f6 1040 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
1041 xfs_dqlock(dqp);
1042 }
1043
1044 while (1) {
1045 gdqp = NULL;
1046
1047 /* We can't depend on nrefs being == 1 here */
1048 if (--dqp->q_nrefs == 0) {
0b1b213f
CH
1049 trace_xfs_dqput_free(dqp);
1050
3a8406f6
DC
1051 list_add_tail(&dqp->q_freelist, &xfs_Gqm->qm_dqfrlist);
1052 xfs_Gqm->qm_dqfrlist_cnt++;
1da177e4
LT
1053
1054 /*
1055 * If we just added a udquot to the freelist, then
1056 * we want to release the gdquot reference that
1057 * it (probably) has. Otherwise it'll keep the
1058 * gdquot from getting reclaimed.
1059 */
1060 if ((gdqp = dqp->q_gdquot)) {
1061 /*
1062 * Avoid a recursive dqput call
1063 */
1064 xfs_dqlock(gdqp);
1065 dqp->q_gdquot = NULL;
1066 }
1da177e4
LT
1067 }
1068 xfs_dqunlock(dqp);
1069
1070 /*
1071 * If we had a group quota inside the user quota as a hint,
1072 * release it now.
1073 */
1074 if (! gdqp)
1075 break;
1076 dqp = gdqp;
1077 }
3a8406f6 1078 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
1079}
1080
1081/*
1082 * Release a dquot. Flush it if dirty, then dqput() it.
1083 * dquot must not be locked.
1084 */
1085void
1086xfs_qm_dqrele(
1087 xfs_dquot_t *dqp)
1088{
7d095257
CH
1089 if (!dqp)
1090 return;
1091
0b1b213f 1092 trace_xfs_dqrele(dqp);
1da177e4
LT
1093
1094 xfs_dqlock(dqp);
1095 /*
1096 * We don't care to flush it if the dquot is dirty here.
1097 * That will create stutters that we want to avoid.
1098 * Instead we do a delayed write when we try to reclaim
1099 * a dirty dquot. Also xfs_sync will take part of the burden...
1100 */
1101 xfs_qm_dqput(dqp);
1102}
1103
ca30b2a7
CH
1104/*
1105 * This is the dquot flushing I/O completion routine. It is called
1106 * from interrupt level when the buffer containing the dquot is
1107 * flushed to disk. It is responsible for removing the dquot logitem
1108 * from the AIL if it has not been re-logged, and unlocking the dquot's
1109 * flush lock. This behavior is very similar to that of inodes..
1110 */
1111STATIC void
1112xfs_qm_dqflush_done(
1113 struct xfs_buf *bp,
1114 struct xfs_log_item *lip)
1115{
1116 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
1117 xfs_dquot_t *dqp = qip->qli_dquot;
1118 struct xfs_ail *ailp = lip->li_ailp;
1119
1120 /*
1121 * We only want to pull the item from the AIL if its
1122 * location in the log has not changed since we started the flush.
1123 * Thus, we only bother if the dquot's lsn has
1124 * not changed. First we check the lsn outside the lock
1125 * since it's cheaper, and then we recheck while
1126 * holding the lock before removing the dquot from the AIL.
1127 */
1128 if ((lip->li_flags & XFS_LI_IN_AIL) &&
1129 lip->li_lsn == qip->qli_flush_lsn) {
1130
1131 /* xfs_trans_ail_delete() drops the AIL lock. */
1132 spin_lock(&ailp->xa_lock);
1133 if (lip->li_lsn == qip->qli_flush_lsn)
1134 xfs_trans_ail_delete(ailp, lip);
1135 else
1136 spin_unlock(&ailp->xa_lock);
1137 }
1138
1139 /*
1140 * Release the dq's flush lock since we're done with it.
1141 */
1142 xfs_dqfunlock(dqp);
1143}
1da177e4
LT
1144
1145/*
1146 * Write a modified dquot to disk.
1147 * The dquot must be locked and the flush lock too taken by caller.
1148 * The flush lock will not be unlocked until the dquot reaches the disk,
1149 * but the dquot is free to be unlocked and modified by the caller
1150 * in the interim. Dquot is still locked on return. This behavior is
1151 * identical to that of inodes.
1152 */
1153int
1154xfs_qm_dqflush(
1155 xfs_dquot_t *dqp,
1156 uint flags)
1157{
acecf1b5
CH
1158 struct xfs_mount *mp = dqp->q_mount;
1159 struct xfs_buf *bp;
1160 struct xfs_disk_dquot *ddqp;
1da177e4 1161 int error;
1da177e4
LT
1162
1163 ASSERT(XFS_DQ_IS_LOCKED(dqp));
e1f49cf2 1164 ASSERT(!completion_done(&dqp->q_flush));
acecf1b5 1165
0b1b213f 1166 trace_xfs_dqflush(dqp);
1da177e4
LT
1167
1168 /*
acecf1b5 1169 * If not dirty, or it's pinned and we are not supposed to block, nada.
1da177e4 1170 */
2f8a3ce1 1171 if (!XFS_DQ_IS_DIRTY(dqp) ||
20026d92 1172 (!(flags & SYNC_WAIT) && atomic_read(&dqp->q_pincount) > 0)) {
1da177e4 1173 xfs_dqfunlock(dqp);
2f8a3ce1 1174 return 0;
1da177e4 1175 }
1da177e4
LT
1176 xfs_qm_dqunpin_wait(dqp);
1177
1178 /*
1179 * This may have been unpinned because the filesystem is shutting
1180 * down forcibly. If that's the case we must not write this dquot
1181 * to disk, because the log record didn't make it to disk!
1182 */
acecf1b5
CH
1183 if (XFS_FORCED_SHUTDOWN(mp)) {
1184 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1da177e4
LT
1185 xfs_dqfunlock(dqp);
1186 return XFS_ERROR(EIO);
1187 }
1188
1189 /*
1190 * Get the buffer containing the on-disk dquot
1da177e4 1191 */
acecf1b5
CH
1192 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1193 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
1194 if (error) {
1da177e4 1195 ASSERT(error != ENOENT);
1da177e4 1196 xfs_dqfunlock(dqp);
acecf1b5 1197 return error;
1da177e4
LT
1198 }
1199
acecf1b5
CH
1200 /*
1201 * Calculate the location of the dquot inside the buffer.
1202 */
62926044 1203 ddqp = bp->b_addr + dqp->q_bufoffset;
acecf1b5
CH
1204
1205 /*
1206 * A simple sanity check in case we got a corrupted dquot..
1207 */
a0fa2b67
DC
1208 error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
1209 XFS_QMOPT_DOWARN, "dqflush (incore copy)");
1210 if (error) {
acecf1b5
CH
1211 xfs_buf_relse(bp);
1212 xfs_dqfunlock(dqp);
1213 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1da177e4
LT
1214 return XFS_ERROR(EIO);
1215 }
1216
1217 /* This is the only portion of data that needs to persist */
acecf1b5 1218 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
1da177e4
LT
1219
1220 /*
1221 * Clear the dirty field and remember the flush lsn for later use.
1222 */
acecf1b5 1223 dqp->dq_flags &= ~XFS_DQ_DIRTY;
1da177e4 1224
7b2e2a31
DC
1225 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1226 &dqp->q_logitem.qli_item.li_lsn);
1da177e4
LT
1227
1228 /*
1229 * Attach an iodone routine so that we can remove this dquot from the
1230 * AIL and release the flush lock once the dquot is synced to disk.
1231 */
ca30b2a7
CH
1232 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1233 &dqp->q_logitem.qli_item);
1234
1da177e4
LT
1235 /*
1236 * If the buffer is pinned then push on the log so we won't
1237 * get stuck waiting in the write for too long.
1238 */
811e64c7 1239 if (xfs_buf_ispinned(bp)) {
0b1b213f 1240 trace_xfs_dqflush_force(dqp);
a14a348b 1241 xfs_log_force(mp, 0);
1da177e4
LT
1242 }
1243
20026d92 1244 if (flags & SYNC_WAIT)
c2b006c1
CH
1245 error = xfs_bwrite(bp);
1246 else
61551f1e 1247 xfs_buf_delwri_queue(bp);
c2b006c1
CH
1248
1249 xfs_buf_relse(bp);
0b1b213f
CH
1250
1251 trace_xfs_dqflush_done(dqp);
1252
1da177e4
LT
1253 /*
1254 * dqp is still locked, but caller is free to unlock it now.
1255 */
20026d92 1256 return error;
1da177e4
LT
1257
1258}
1259
1da177e4
LT
1260int
1261xfs_qm_dqlock_nowait(
1262 xfs_dquot_t *dqp)
1263{
e1f49cf2 1264 return mutex_trylock(&dqp->q_qlock);
1da177e4
LT
1265}
1266
1267void
1268xfs_dqlock(
1269 xfs_dquot_t *dqp)
1270{
e1f49cf2 1271 mutex_lock(&dqp->q_qlock);
1da177e4
LT
1272}
1273
1274void
1275xfs_dqunlock(
1276 xfs_dquot_t *dqp)
1277{
1278 mutex_unlock(&(dqp->q_qlock));
1279 if (dqp->q_logitem.qli_dquot == dqp) {
1280 /* Once was dqp->q_mount, but might just have been cleared */
783a2f65 1281 xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_ailp,
1da177e4
LT
1282 (xfs_log_item_t*)&(dqp->q_logitem));
1283 }
1284}
1285
1286
1287void
1288xfs_dqunlock_nonotify(
1289 xfs_dquot_t *dqp)
1290{
1291 mutex_unlock(&(dqp->q_qlock));
1292}
1293
5bb87a33
CH
1294/*
1295 * Lock two xfs_dquot structures.
1296 *
1297 * To avoid deadlocks we always lock the quota structure with
1298 * the lowerd id first.
1299 */
1da177e4
LT
1300void
1301xfs_dqlock2(
1302 xfs_dquot_t *d1,
1303 xfs_dquot_t *d2)
1304{
1305 if (d1 && d2) {
1306 ASSERT(d1 != d2);
1149d96a
CH
1307 if (be32_to_cpu(d1->q_core.d_id) >
1308 be32_to_cpu(d2->q_core.d_id)) {
5bb87a33
CH
1309 mutex_lock(&d2->q_qlock);
1310 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1da177e4 1311 } else {
5bb87a33
CH
1312 mutex_lock(&d1->q_qlock);
1313 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1da177e4 1314 }
5bb87a33
CH
1315 } else if (d1) {
1316 mutex_lock(&d1->q_qlock);
1317 } else if (d2) {
1318 mutex_lock(&d2->q_qlock);
1da177e4
LT
1319 }
1320}
1321
1322
1323/*
1324 * Take a dquot out of the mount's dqlist as well as the hashlist.
1325 * This is called via unmount as well as quotaoff, and the purge
1326 * will always succeed unless there are soft (temp) references
1327 * outstanding.
1328 *
1329 * This returns 0 if it was purged, 1 if it wasn't. It's not an error code
1330 * that we're returning! XXXsup - not cool.
1331 */
1332/* ARGSUSED */
1333int
1334xfs_qm_dqpurge(
4f0e8a98 1335 xfs_dquot_t *dqp)
1da177e4 1336{
e6a81f13 1337 xfs_dqhash_t *qh = dqp->q_hash;
3c56836f 1338 xfs_mount_t *mp = dqp->q_mount;
1da177e4 1339
3a25404b 1340 ASSERT(mutex_is_locked(&mp->m_quotainfo->qi_dqlist_lock));
c9a192dc 1341 ASSERT(mutex_is_locked(&dqp->q_hash->qh_lock));
1da177e4
LT
1342
1343 xfs_dqlock(dqp);
1344 /*
1345 * We really can't afford to purge a dquot that is
1346 * referenced, because these are hard refs.
1347 * It shouldn't happen in general because we went thru _all_ inodes in
1348 * dqrele_all_inodes before calling this and didn't let the mountlock go.
1349 * However it is possible that we have dquots with temporary
1350 * references that are not attached to an inode. e.g. see xfs_setattr().
1351 */
1352 if (dqp->q_nrefs != 0) {
1353 xfs_dqunlock(dqp);
c9a192dc 1354 mutex_unlock(&dqp->q_hash->qh_lock);
1da177e4
LT
1355 return (1);
1356 }
1357
3a8406f6 1358 ASSERT(!list_empty(&dqp->q_freelist));
1da177e4
LT
1359
1360 /*
1361 * If we're turning off quotas, we have to make sure that, for
1362 * example, we don't delete quota disk blocks while dquots are
1363 * in the process of getting written to those disk blocks.
1364 * This dquot might well be on AIL, and we can't leave it there
1365 * if we're turning off quotas. Basically, we need this flush
1366 * lock, and are willing to block on it.
1367 */
e1f49cf2 1368 if (!xfs_dqflock_nowait(dqp)) {
1da177e4
LT
1369 /*
1370 * Block on the flush lock after nudging dquot buffer,
1371 * if it is incore.
1372 */
1373 xfs_qm_dqflock_pushbuf_wait(dqp);
1374 }
1375
1376 /*
1377 * XXXIf we're turning this type of quotas off, we don't care
1378 * about the dirty metadata sitting in this dquot. OTOH, if
1379 * we're unmounting, we do care, so we flush it and wait.
1380 */
1381 if (XFS_DQ_IS_DIRTY(dqp)) {
3c56836f 1382 int error;
0b1b213f 1383
1da177e4
LT
1384 /* dqflush unlocks dqflock */
1385 /*
1386 * Given that dqpurge is a very rare occurrence, it is OK
1387 * that we're holding the hashlist and mplist locks
1388 * across the disk write. But, ... XXXsup
1389 *
1390 * We don't care about getting disk errors here. We need
1391 * to purge this dquot anyway, so we go ahead regardless.
1392 */
20026d92 1393 error = xfs_qm_dqflush(dqp, SYNC_WAIT);
3c56836f 1394 if (error)
53487786
DC
1395 xfs_warn(mp, "%s: dquot %p flush failed",
1396 __func__, dqp);
1da177e4
LT
1397 xfs_dqflock(dqp);
1398 }
bc3048e3 1399 ASSERT(atomic_read(&dqp->q_pincount) == 0);
1da177e4
LT
1400 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1401 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
1402
e6a81f13
DC
1403 list_del_init(&dqp->q_hashlist);
1404 qh->qh_version++;
3a25404b
DC
1405 list_del_init(&dqp->q_mplist);
1406 mp->m_quotainfo->qi_dqreclaims++;
1407 mp->m_quotainfo->qi_dquots--;
1da177e4
LT
1408 /*
1409 * XXX Move this to the front of the freelist, if we can get the
1410 * freelist lock.
1411 */
3a8406f6 1412 ASSERT(!list_empty(&dqp->q_freelist));
1da177e4
LT
1413
1414 dqp->q_mount = NULL;
1415 dqp->q_hash = NULL;
1416 dqp->dq_flags = XFS_DQ_INACTIVE;
1417 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
1418 xfs_dqfunlock(dqp);
1419 xfs_dqunlock(dqp);
e6a81f13 1420 mutex_unlock(&qh->qh_lock);
1da177e4
LT
1421 return (0);
1422}
1423
1424
1da177e4
LT
1425/*
1426 * Give the buffer a little push if it is incore and
1427 * wait on the flush lock.
1428 */
1429void
1430xfs_qm_dqflock_pushbuf_wait(
1431 xfs_dquot_t *dqp)
1432{
8a7b8a89 1433 xfs_mount_t *mp = dqp->q_mount;
1da177e4
LT
1434 xfs_buf_t *bp;
1435
1436 /*
1437 * Check to see if the dquot has been flushed delayed
1438 * write. If so, grab its buffer and send it
1439 * out immediately. We'll be able to acquire
1440 * the flush lock when the I/O completes.
1441 */
8a7b8a89
CH
1442 bp = xfs_incore(mp->m_ddev_targp, dqp->q_blkno,
1443 mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK);
7d6a7bde
DC
1444 if (!bp)
1445 goto out_lock;
1446
1447 if (XFS_BUF_ISDELAYWRITE(bp)) {
811e64c7 1448 if (xfs_buf_ispinned(bp))
8a7b8a89 1449 xfs_log_force(mp, 0);
7d6a7bde
DC
1450 xfs_buf_delwri_promote(bp);
1451 wake_up_process(bp->b_target->bt_task);
1da177e4 1452 }
7d6a7bde
DC
1453 xfs_buf_relse(bp);
1454out_lock:
1da177e4
LT
1455 xfs_dqflock(dqp);
1456}