]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/xfs/quota/xfs_dquot.c
xfs: give li_cb callbacks the correct prototype
[net-next-2.6.git] / fs / xfs / quota / 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 sv_destroy(&dqp->q_pinwait);
1da177e4 153 kmem_zone_free(xfs_Gqm->qm_dqzone, dqp);
0b1b213f 154
1da177e4
LT
155 atomic_dec(&xfs_Gqm->qm_totaldquots);
156}
157
158/*
159 * This is what a 'fresh' dquot inside a dquot chunk looks like on disk.
160 */
161STATIC void
162xfs_qm_dqinit_core(
163 xfs_dqid_t id,
164 uint type,
165 xfs_dqblk_t *d)
166{
167 /*
168 * Caller has zero'd the entire dquot 'chunk' already.
169 */
1149d96a
CH
170 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
171 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
172 d->dd_diskdq.d_id = cpu_to_be32(id);
173 d->dd_diskdq.d_flags = type;
1da177e4
LT
174}
175
1da177e4
LT
176/*
177 * If default limits are in force, push them into the dquot now.
178 * We overwrite the dquot limits only if they are zero and this
179 * is not the root dquot.
180 */
181void
182xfs_qm_adjust_dqlimits(
183 xfs_mount_t *mp,
184 xfs_disk_dquot_t *d)
185{
186 xfs_quotainfo_t *q = mp->m_quotainfo;
187
188 ASSERT(d->d_id);
189
190 if (q->qi_bsoftlimit && !d->d_blk_softlimit)
1149d96a 191 d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit);
1da177e4 192 if (q->qi_bhardlimit && !d->d_blk_hardlimit)
1149d96a 193 d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit);
1da177e4 194 if (q->qi_isoftlimit && !d->d_ino_softlimit)
1149d96a 195 d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit);
1da177e4 196 if (q->qi_ihardlimit && !d->d_ino_hardlimit)
1149d96a 197 d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit);
1da177e4 198 if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit)
1149d96a 199 d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit);
1da177e4 200 if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit)
1149d96a 201 d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit);
1da177e4
LT
202}
203
204/*
205 * Check the limits and timers of a dquot and start or reset timers
206 * if necessary.
207 * This gets called even when quota enforcement is OFF, which makes our
208 * life a little less complicated. (We just don't reject any quota
209 * reservations in that case, when enforcement is off).
210 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
211 * enforcement's off.
212 * In contrast, warnings are a little different in that they don't
754002b4
NS
213 * 'automatically' get started when limits get exceeded. They do
214 * get reset to zero, however, when we find the count to be under
215 * the soft limit (they are only ever set non-zero via userspace).
1da177e4
LT
216 */
217void
218xfs_qm_adjust_dqtimers(
219 xfs_mount_t *mp,
220 xfs_disk_dquot_t *d)
221{
222 ASSERT(d->d_id);
223
224#ifdef QUOTADEBUG
1149d96a
CH
225 if (d->d_blk_hardlimit)
226 ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
227 be64_to_cpu(d->d_blk_hardlimit));
228 if (d->d_ino_hardlimit)
229 ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
230 be64_to_cpu(d->d_ino_hardlimit));
231 if (d->d_rtb_hardlimit)
232 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
233 be64_to_cpu(d->d_rtb_hardlimit));
1da177e4
LT
234#endif
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);
321 ASSERT(XFS_BUF_ISBUSY(bp));
322 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
323
324 d = (xfs_dqblk_t *)XFS_BUF_PTR(bp);
325
326 /*
327 * ID of the first dquot in the block - id's are zero based.
328 */
8a7b8a89 329 curid = id - (id % q->qi_dqperchunk);
1da177e4 330 ASSERT(curid >= 0);
8a7b8a89
CH
331 memset(d, 0, BBTOB(q->qi_dqchunklen));
332 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++)
1da177e4
LT
333 xfs_qm_dqinit_core(curid, type, d);
334 xfs_trans_dquot_buf(tp, bp,
c1155410
DC
335 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
336 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
337 XFS_BLF_GDQUOT_BUF)));
8a7b8a89 338 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
1da177e4
LT
339}
340
341
342
343/*
344 * Allocate a block and fill it with dquots.
345 * This is called when the bmapi finds a hole.
346 */
347STATIC int
348xfs_qm_dqalloc(
efa092f3 349 xfs_trans_t **tpp,
1da177e4
LT
350 xfs_mount_t *mp,
351 xfs_dquot_t *dqp,
352 xfs_inode_t *quotip,
353 xfs_fileoff_t offset_fsb,
354 xfs_buf_t **O_bpp)
355{
356 xfs_fsblock_t firstblock;
357 xfs_bmap_free_t flist;
358 xfs_bmbt_irec_t map;
359 int nmaps, error, committed;
360 xfs_buf_t *bp;
efa092f3 361 xfs_trans_t *tp = *tpp;
1da177e4
LT
362
363 ASSERT(tp != NULL);
0b1b213f
CH
364
365 trace_xfs_dqalloc(dqp);
1da177e4
LT
366
367 /*
368 * Initialize the bmap freelist prior to calling bmapi code.
369 */
9d87c319 370 xfs_bmap_init(&flist, &firstblock);
1da177e4
LT
371 xfs_ilock(quotip, XFS_ILOCK_EXCL);
372 /*
373 * Return if this type of quotas is turned off while we didn't
374 * have an inode lock
375 */
376 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
377 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
378 return (ESRCH);
379 }
380
381 /*
382 * xfs_trans_commit normally decrements the vnode ref count
383 * when it unlocks the inode. Since we want to keep the quota
384 * inode around, we bump the vnode ref count now.
385 */
0b1f9177 386 IHOLD(quotip);
1da177e4
LT
387
388 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
389 nmaps = 1;
390 if ((error = xfs_bmapi(tp, quotip,
391 offset_fsb, XFS_DQUOT_CLUSTER_SIZE_FSB,
392 XFS_BMAPI_METADATA | XFS_BMAPI_WRITE,
393 &firstblock,
394 XFS_QM_DQALLOC_SPACE_RES(mp),
3e57ecf6 395 &map, &nmaps, &flist, NULL))) {
1da177e4
LT
396 goto error0;
397 }
398 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
399 ASSERT(nmaps == 1);
400 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
401 (map.br_startblock != HOLESTARTBLOCK));
402
403 /*
404 * Keep track of the blkno to save a lookup later
405 */
406 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
407
408 /* now we can just get the buffer (there's nothing to read yet) */
409 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
410 dqp->q_blkno,
8a7b8a89 411 mp->m_quotainfo->qi_dqchunklen,
1da177e4
LT
412 0);
413 if (!bp || (error = XFS_BUF_GETERROR(bp)))
414 goto error1;
415 /*
416 * Make a chunk of dquots out of this buffer and log
417 * the entire thing.
418 */
1149d96a 419 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
c8ad20ff 420 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
1da177e4 421
efa092f3
TS
422 /*
423 * xfs_bmap_finish() may commit the current transaction and
424 * start a second transaction if the freelist is not empty.
425 *
426 * Since we still want to modify this buffer, we need to
427 * ensure that the buffer is not released on commit of
428 * the first transaction and ensure the buffer is added to the
429 * second transaction.
430 *
431 * If there is only one transaction then don't stop the buffer
432 * from being released when it commits later on.
433 */
434
435 xfs_trans_bhold(tp, bp);
436
f7c99b6f 437 if ((error = xfs_bmap_finish(tpp, &flist, &committed))) {
1da177e4
LT
438 goto error1;
439 }
440
efa092f3
TS
441 if (committed) {
442 tp = *tpp;
443 xfs_trans_bjoin(tp, bp);
444 } else {
445 xfs_trans_bhold_release(tp, bp);
446 }
447
1da177e4
LT
448 *O_bpp = bp;
449 return 0;
450
451 error1:
452 xfs_bmap_cancel(&flist);
453 error0:
454 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
455
456 return (error);
457}
458
459/*
460 * Maps a dquot to the buffer containing its on-disk version.
461 * This returns a ptr to the buffer containing the on-disk dquot
462 * in the bpp param, and a ptr to the on-disk dquot within that buffer
463 */
464STATIC int
465xfs_qm_dqtobp(
efa092f3 466 xfs_trans_t **tpp,
1da177e4
LT
467 xfs_dquot_t *dqp,
468 xfs_disk_dquot_t **O_ddpp,
469 xfs_buf_t **O_bpp,
470 uint flags)
471{
472 xfs_bmbt_irec_t map;
473 int nmaps, error;
474 xfs_buf_t *bp;
475 xfs_inode_t *quotip;
476 xfs_mount_t *mp;
477 xfs_disk_dquot_t *ddq;
478 xfs_dqid_t id;
479 boolean_t newdquot;
efa092f3 480 xfs_trans_t *tp = (tpp ? *tpp : NULL);
1da177e4
LT
481
482 mp = dqp->q_mount;
1149d96a 483 id = be32_to_cpu(dqp->q_core.d_id);
1da177e4
LT
484 nmaps = 1;
485 newdquot = B_FALSE;
486
487 /*
488 * If we don't know where the dquot lives, find out.
489 */
490 if (dqp->q_blkno == (xfs_daddr_t) 0) {
491 /* We use the id as an index */
8a7b8a89
CH
492 dqp->q_fileoffset = (xfs_fileoff_t)id /
493 mp->m_quotainfo->qi_dqperchunk;
1da177e4
LT
494 nmaps = 1;
495 quotip = XFS_DQ_TO_QIP(dqp);
496 xfs_ilock(quotip, XFS_ILOCK_SHARED);
497 /*
498 * Return if this type of quotas is turned off while we didn't
499 * have an inode lock
500 */
501 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
502 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
503 return (ESRCH);
504 }
505 /*
506 * Find the block map; no allocations yet
507 */
508 error = xfs_bmapi(NULL, quotip, dqp->q_fileoffset,
509 XFS_DQUOT_CLUSTER_SIZE_FSB,
510 XFS_BMAPI_METADATA,
3e57ecf6 511 NULL, 0, &map, &nmaps, NULL, NULL);
1da177e4
LT
512
513 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
514 if (error)
515 return (error);
516 ASSERT(nmaps == 1);
517 ASSERT(map.br_blockcount == 1);
518
519 /*
520 * offset of dquot in the (fixed sized) dquot chunk.
521 */
8a7b8a89 522 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
1da177e4
LT
523 sizeof(xfs_dqblk_t);
524 if (map.br_startblock == HOLESTARTBLOCK) {
525 /*
526 * We don't allocate unless we're asked to
527 */
528 if (!(flags & XFS_QMOPT_DQALLOC))
529 return (ENOENT);
530
531 ASSERT(tp);
efa092f3 532 if ((error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
1da177e4
LT
533 dqp->q_fileoffset, &bp)))
534 return (error);
efa092f3 535 tp = *tpp;
1da177e4
LT
536 newdquot = B_TRUE;
537 } else {
538 /*
539 * store the blkno etc so that we don't have to do the
540 * mapping all the time
541 */
542 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
543 }
544 }
545 ASSERT(dqp->q_blkno != DELAYSTARTBLOCK);
546 ASSERT(dqp->q_blkno != HOLESTARTBLOCK);
547
548 /*
549 * Read in the buffer, unless we've just done the allocation
550 * (in which case we already have the buf).
551 */
8a7b8a89 552 if (!newdquot) {
0b1b213f
CH
553 trace_xfs_dqtobp_read(dqp);
554
8a7b8a89
CH
555 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
556 dqp->q_blkno,
557 mp->m_quotainfo->qi_dqchunklen,
558 0, &bp);
1da177e4
LT
559 if (error || !bp)
560 return XFS_ERROR(error);
561 }
562 ASSERT(XFS_BUF_ISBUSY(bp));
563 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
564
565 /*
566 * calculate the location of the dquot inside the buffer.
567 */
568 ddq = (xfs_disk_dquot_t *)((char *)XFS_BUF_PTR(bp) + dqp->q_bufoffset);
569
570 /*
571 * A simple sanity check in case we got a corrupted dquot...
572 */
c8ad20ff 573 if (xfs_qm_dqcheck(ddq, id, dqp->dq_flags & XFS_DQ_ALLTYPES,
1da177e4
LT
574 flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN),
575 "dqtobp")) {
576 if (!(flags & XFS_QMOPT_DQREPAIR)) {
577 xfs_trans_brelse(tp, bp);
578 return XFS_ERROR(EIO);
579 }
580 XFS_BUF_BUSY(bp); /* We dirtied this */
581 }
582
583 *O_bpp = bp;
584 *O_ddpp = ddq;
585
586 return (0);
587}
588
589
590/*
591 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
592 * and release the buffer immediately.
593 *
594 */
595/* ARGSUSED */
596STATIC int
597xfs_qm_dqread(
efa092f3 598 xfs_trans_t **tpp,
1da177e4
LT
599 xfs_dqid_t id,
600 xfs_dquot_t *dqp, /* dquot to get filled in */
601 uint flags)
602{
603 xfs_disk_dquot_t *ddqp;
604 xfs_buf_t *bp;
605 int error;
efa092f3
TS
606 xfs_trans_t *tp;
607
608 ASSERT(tpp);
1da177e4 609
0b1b213f
CH
610 trace_xfs_dqread(dqp);
611
1da177e4
LT
612 /*
613 * get a pointer to the on-disk dquot and the buffer containing it
614 * dqp already knows its own type (GROUP/USER).
615 */
efa092f3 616 if ((error = xfs_qm_dqtobp(tpp, dqp, &ddqp, &bp, flags))) {
1da177e4
LT
617 return (error);
618 }
efa092f3 619 tp = *tpp;
1da177e4
LT
620
621 /* copy everything from disk dquot to the incore dquot */
622 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
1149d96a 623 ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);
1da177e4
LT
624 xfs_qm_dquot_logitem_init(dqp);
625
626 /*
627 * Reservation counters are defined as reservation plus current usage
628 * to avoid having to add everytime.
629 */
1149d96a
CH
630 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
631 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
632 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
1da177e4
LT
633
634 /* Mark the buf so that this will stay incore a little longer */
635 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DQUOT, XFS_DQUOT_REF);
636
637 /*
638 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
639 * So we need to release with xfs_trans_brelse().
640 * The strategy here is identical to that of inodes; we lock
641 * the dquot in xfs_qm_dqget() before making it accessible to
642 * others. This is because dquots, like inodes, need a good level of
643 * concurrency, and we don't want to take locks on the entire buffers
644 * for dquot accesses.
645 * Note also that the dquot buffer may even be dirty at this point, if
646 * this particular dquot was repaired. We still aren't afraid to
647 * brelse it because we have the changes incore.
648 */
649 ASSERT(XFS_BUF_ISBUSY(bp));
650 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
651 xfs_trans_brelse(tp, bp);
652
653 return (error);
654}
655
656
657/*
658 * allocate an incore dquot from the kernel heap,
659 * and fill its core with quota information kept on disk.
660 * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk
661 * if it wasn't already allocated.
662 */
663STATIC int
664xfs_qm_idtodq(
665 xfs_mount_t *mp,
666 xfs_dqid_t id, /* gid or uid, depending on type */
667 uint type, /* UDQUOT or GDQUOT */
668 uint flags, /* DQALLOC, DQREPAIR */
669 xfs_dquot_t **O_dqpp)/* OUT : incore dquot, not locked */
670{
671 xfs_dquot_t *dqp;
672 int error;
673 xfs_trans_t *tp;
674 int cancelflags=0;
675
676 dqp = xfs_qm_dqinit(mp, id, type);
677 tp = NULL;
678 if (flags & XFS_QMOPT_DQALLOC) {
679 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
8a7b8a89
CH
680 error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp),
681 XFS_WRITE_LOG_RES(mp) +
682 BBTOB(mp->m_quotainfo->qi_dqchunklen) - 1 +
683 128,
684 0,
685 XFS_TRANS_PERM_LOG_RES,
686 XFS_WRITE_LOG_COUNT);
687 if (error) {
1da177e4
LT
688 cancelflags = 0;
689 goto error0;
690 }
691 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
692 }
693
694 /*
695 * Read it from disk; xfs_dqread() takes care of
696 * all the necessary initialization of dquot's fields (locks, etc)
697 */
efa092f3 698 if ((error = xfs_qm_dqread(&tp, id, dqp, flags))) {
1da177e4
LT
699 /*
700 * This can happen if quotas got turned off (ESRCH),
701 * or if the dquot didn't exist on disk and we ask to
702 * allocate (ENOENT).
703 */
0b1b213f 704 trace_xfs_dqread_fail(dqp);
1da177e4
LT
705 cancelflags |= XFS_TRANS_ABORT;
706 goto error0;
707 }
708 if (tp) {
1c72bf90 709 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES)))
1da177e4
LT
710 goto error1;
711 }
712
713 *O_dqpp = dqp;
714 return (0);
715
716 error0:
717 ASSERT(error);
718 if (tp)
719 xfs_trans_cancel(tp, cancelflags);
720 error1:
721 xfs_qm_dqdestroy(dqp);
722 *O_dqpp = NULL;
723 return (error);
724}
725
726/*
727 * Lookup a dquot in the incore dquot hashtable. We keep two separate
728 * hashtables for user and group dquots; and, these are global tables
729 * inside the XQM, not per-filesystem tables.
730 * The hash chain must be locked by caller, and it is left locked
731 * on return. Returning dquot is locked.
732 */
733STATIC int
734xfs_qm_dqlookup(
735 xfs_mount_t *mp,
736 xfs_dqid_t id,
737 xfs_dqhash_t *qh,
738 xfs_dquot_t **O_dqpp)
739{
740 xfs_dquot_t *dqp;
741 uint flist_locked;
1da177e4 742
c9a192dc 743 ASSERT(mutex_is_locked(&qh->qh_lock));
1da177e4
LT
744
745 flist_locked = B_FALSE;
746
747 /*
748 * Traverse the hashchain looking for a match
749 */
e6a81f13 750 list_for_each_entry(dqp, &qh->qh_list, q_hashlist) {
1da177e4
LT
751 /*
752 * We already have the hashlock. We don't need the
753 * dqlock to look at the id field of the dquot, since the
754 * id can't be modified without the hashlock anyway.
755 */
1149d96a 756 if (be32_to_cpu(dqp->q_core.d_id) == id && dqp->q_mount == mp) {
0b1b213f
CH
757 trace_xfs_dqlookup_found(dqp);
758
1da177e4
LT
759 /*
760 * All in core dquots must be on the dqlist of mp
761 */
3a25404b 762 ASSERT(!list_empty(&dqp->q_mplist));
1da177e4
LT
763
764 xfs_dqlock(dqp);
765 if (dqp->q_nrefs == 0) {
3a8406f6
DC
766 ASSERT(!list_empty(&dqp->q_freelist));
767 if (!mutex_trylock(&xfs_Gqm->qm_dqfrlist_lock)) {
0b1b213f 768 trace_xfs_dqlookup_want(dqp);
1da177e4
LT
769
770 /*
771 * We may have raced with dqreclaim_one()
772 * (and lost). So, flag that we don't
773 * want the dquot to be reclaimed.
774 */
775 dqp->dq_flags |= XFS_DQ_WANT;
776 xfs_dqunlock(dqp);
3a8406f6 777 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
778 xfs_dqlock(dqp);
779 dqp->dq_flags &= ~(XFS_DQ_WANT);
780 }
781 flist_locked = B_TRUE;
782 }
783
784 /*
785 * id couldn't have changed; we had the hashlock all
786 * along
787 */
1149d96a 788 ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);
1da177e4
LT
789
790 if (flist_locked) {
791 if (dqp->q_nrefs != 0) {
3a8406f6 792 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
793 flist_locked = B_FALSE;
794 } else {
3a8406f6 795 /* take it off the freelist */
0b1b213f 796 trace_xfs_dqlookup_freelist(dqp);
3a8406f6
DC
797 list_del_init(&dqp->q_freelist);
798 xfs_Gqm->qm_dqfrlist_cnt--;
1da177e4
LT
799 }
800 }
801
1da177e4
LT
802 XFS_DQHOLD(dqp);
803
804 if (flist_locked)
3a8406f6 805 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
806 /*
807 * move the dquot to the front of the hashchain
808 */
c9a192dc 809 ASSERT(mutex_is_locked(&qh->qh_lock));
e6a81f13 810 list_move(&dqp->q_hashlist, &qh->qh_list);
0b1b213f 811 trace_xfs_dqlookup_done(dqp);
1da177e4 812 *O_dqpp = dqp;
e6a81f13 813 return 0;
1da177e4
LT
814 }
815 }
816
817 *O_dqpp = NULL;
c9a192dc 818 ASSERT(mutex_is_locked(&qh->qh_lock));
1da177e4
LT
819 return (1);
820}
821
822/*
823 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
824 * a locked dquot, doing an allocation (if requested) as needed.
825 * When both an inode and an id are given, the inode's id takes precedence.
826 * That is, if the id changes while we don't hold the ilock inside this
827 * function, the new dquot is returned, not necessarily the one requested
828 * in the id argument.
829 */
830int
831xfs_qm_dqget(
832 xfs_mount_t *mp,
833 xfs_inode_t *ip, /* locked inode (optional) */
c8ad20ff
NS
834 xfs_dqid_t id, /* uid/projid/gid depending on type */
835 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
1da177e4
LT
836 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
837 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
838{
839 xfs_dquot_t *dqp;
840 xfs_dqhash_t *h;
841 uint version;
842 int error;
843
844 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
845 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
c8ad20ff 846 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
1da177e4
LT
847 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
848 return (ESRCH);
849 }
850 h = XFS_DQ_HASH(mp, id, type);
851
852#ifdef DEBUG
853 if (xfs_do_dqerror) {
854 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
855 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
856 cmn_err(CE_DEBUG, "Returning error in dqget");
857 return (EIO);
858 }
859 }
860#endif
861
862 again:
863
864#ifdef DEBUG
c8ad20ff
NS
865 ASSERT(type == XFS_DQ_USER ||
866 type == XFS_DQ_PROJ ||
867 type == XFS_DQ_GROUP);
1da177e4 868 if (ip) {
579aa9ca 869 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4
LT
870 if (type == XFS_DQ_USER)
871 ASSERT(ip->i_udquot == NULL);
872 else
873 ASSERT(ip->i_gdquot == NULL);
874 }
875#endif
c9a192dc 876 mutex_lock(&h->qh_lock);
1da177e4
LT
877
878 /*
879 * Look in the cache (hashtable).
880 * The chain is kept locked during lookup.
881 */
882 if (xfs_qm_dqlookup(mp, id, h, O_dqpp) == 0) {
883 XQM_STATS_INC(xqmstats.xs_qm_dqcachehits);
884 /*
885 * The dquot was found, moved to the front of the chain,
886 * taken off the freelist if it was on it, and locked
887 * at this point. Just unlock the hashchain and return.
888 */
889 ASSERT(*O_dqpp);
890 ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp));
c9a192dc 891 mutex_unlock(&h->qh_lock);
0b1b213f 892 trace_xfs_dqget_hit(*O_dqpp);
1da177e4
LT
893 return (0); /* success */
894 }
895 XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses);
896
897 /*
898 * Dquot cache miss. We don't want to keep the inode lock across
899 * a (potential) disk read. Also we don't want to deal with the lock
900 * ordering between quotainode and this inode. OTOH, dropping the inode
901 * lock here means dealing with a chown that can happen before
902 * we re-acquire the lock.
903 */
904 if (ip)
905 xfs_iunlock(ip, XFS_ILOCK_EXCL);
906 /*
907 * Save the hashchain version stamp, and unlock the chain, so that
908 * we don't keep the lock across a disk read
909 */
910 version = h->qh_version;
c9a192dc 911 mutex_unlock(&h->qh_lock);
1da177e4
LT
912
913 /*
914 * Allocate the dquot on the kernel heap, and read the ondisk
915 * portion off the disk. Also, do all the necessary initialization
916 * This can return ENOENT if dquot didn't exist on disk and we didn't
917 * ask it to allocate; ESRCH if quotas got turned off suddenly.
918 */
919 if ((error = xfs_qm_idtodq(mp, id, type,
920 flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR|
921 XFS_QMOPT_DOWARN),
922 &dqp))) {
923 if (ip)
924 xfs_ilock(ip, XFS_ILOCK_EXCL);
925 return (error);
926 }
927
928 /*
929 * See if this is mount code calling to look at the overall quota limits
930 * which are stored in the id == 0 user or group's dquot.
931 * Since we may not have done a quotacheck by this point, just return
932 * the dquot without attaching it to any hashtables, lists, etc, or even
933 * taking a reference.
934 * The caller must dqdestroy this once done.
935 */
936 if (flags & XFS_QMOPT_DQSUSER) {
937 ASSERT(id == 0);
938 ASSERT(! ip);
939 goto dqret;
940 }
941
942 /*
943 * Dquot lock comes after hashlock in the lock ordering
944 */
945 if (ip) {
946 xfs_ilock(ip, XFS_ILOCK_EXCL);
191f8488 947
1da177e4
LT
948 /*
949 * A dquot could be attached to this inode by now, since
950 * we had dropped the ilock.
951 */
952 if (type == XFS_DQ_USER) {
191f8488
CH
953 if (!XFS_IS_UQUOTA_ON(mp)) {
954 /* inode stays locked on return */
955 xfs_qm_dqdestroy(dqp);
956 return XFS_ERROR(ESRCH);
957 }
1da177e4
LT
958 if (ip->i_udquot) {
959 xfs_qm_dqdestroy(dqp);
960 dqp = ip->i_udquot;
961 xfs_dqlock(dqp);
962 goto dqret;
963 }
964 } else {
191f8488
CH
965 if (!XFS_IS_OQUOTA_ON(mp)) {
966 /* inode stays locked on return */
967 xfs_qm_dqdestroy(dqp);
968 return XFS_ERROR(ESRCH);
969 }
1da177e4
LT
970 if (ip->i_gdquot) {
971 xfs_qm_dqdestroy(dqp);
972 dqp = ip->i_gdquot;
973 xfs_dqlock(dqp);
974 goto dqret;
975 }
976 }
977 }
978
979 /*
980 * Hashlock comes after ilock in lock order
981 */
c9a192dc 982 mutex_lock(&h->qh_lock);
1da177e4
LT
983 if (version != h->qh_version) {
984 xfs_dquot_t *tmpdqp;
985 /*
986 * Now, see if somebody else put the dquot in the
987 * hashtable before us. This can happen because we didn't
988 * keep the hashchain lock. We don't have to worry about
989 * lock order between the two dquots here since dqp isn't
990 * on any findable lists yet.
991 */
992 if (xfs_qm_dqlookup(mp, id, h, &tmpdqp) == 0) {
993 /*
994 * Duplicate found. Just throw away the new dquot
995 * and start over.
996 */
997 xfs_qm_dqput(tmpdqp);
c9a192dc 998 mutex_unlock(&h->qh_lock);
1da177e4
LT
999 xfs_qm_dqdestroy(dqp);
1000 XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
1001 goto again;
1002 }
1003 }
1004
1005 /*
1006 * Put the dquot at the beginning of the hash-chain and mp's list
1007 * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock ..
1008 */
c9a192dc 1009 ASSERT(mutex_is_locked(&h->qh_lock));
1da177e4 1010 dqp->q_hash = h;
e6a81f13
DC
1011 list_add(&dqp->q_hashlist, &h->qh_list);
1012 h->qh_version++;
1da177e4
LT
1013
1014 /*
1015 * Attach this dquot to this filesystem's list of all dquots,
1016 * kept inside the mount structure in m_quotainfo field
1017 */
3a25404b 1018 mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
1da177e4
LT
1019
1020 /*
1021 * We return a locked dquot to the caller, with a reference taken
1022 */
1023 xfs_dqlock(dqp);
1024 dqp->q_nrefs = 1;
1025
3a25404b
DC
1026 list_add(&dqp->q_mplist, &mp->m_quotainfo->qi_dqlist);
1027 mp->m_quotainfo->qi_dquots++;
1028 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
c9a192dc 1029 mutex_unlock(&h->qh_lock);
1da177e4 1030 dqret:
579aa9ca 1031 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
0b1b213f 1032 trace_xfs_dqget_miss(dqp);
1da177e4
LT
1033 *O_dqpp = dqp;
1034 return (0);
1035}
1036
1037
1038/*
1039 * Release a reference to the dquot (decrement ref-count)
1040 * and unlock it. If there is a group quota attached to this
1041 * dquot, carefully release that too without tripping over
1042 * deadlocks'n'stuff.
1043 */
1044void
1045xfs_qm_dqput(
1046 xfs_dquot_t *dqp)
1047{
1048 xfs_dquot_t *gdqp;
1049
1050 ASSERT(dqp->q_nrefs > 0);
1051 ASSERT(XFS_DQ_IS_LOCKED(dqp));
0b1b213f
CH
1052
1053 trace_xfs_dqput(dqp);
1da177e4
LT
1054
1055 if (dqp->q_nrefs != 1) {
1056 dqp->q_nrefs--;
1057 xfs_dqunlock(dqp);
1058 return;
1059 }
1060
1061 /*
1062 * drop the dqlock and acquire the freelist and dqlock
1063 * in the right order; but try to get it out-of-order first
1064 */
3a8406f6 1065 if (!mutex_trylock(&xfs_Gqm->qm_dqfrlist_lock)) {
0b1b213f 1066 trace_xfs_dqput_wait(dqp);
1da177e4 1067 xfs_dqunlock(dqp);
3a8406f6 1068 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
1069 xfs_dqlock(dqp);
1070 }
1071
1072 while (1) {
1073 gdqp = NULL;
1074
1075 /* We can't depend on nrefs being == 1 here */
1076 if (--dqp->q_nrefs == 0) {
0b1b213f
CH
1077 trace_xfs_dqput_free(dqp);
1078
3a8406f6
DC
1079 list_add_tail(&dqp->q_freelist, &xfs_Gqm->qm_dqfrlist);
1080 xfs_Gqm->qm_dqfrlist_cnt++;
1da177e4
LT
1081
1082 /*
1083 * If we just added a udquot to the freelist, then
1084 * we want to release the gdquot reference that
1085 * it (probably) has. Otherwise it'll keep the
1086 * gdquot from getting reclaimed.
1087 */
1088 if ((gdqp = dqp->q_gdquot)) {
1089 /*
1090 * Avoid a recursive dqput call
1091 */
1092 xfs_dqlock(gdqp);
1093 dqp->q_gdquot = NULL;
1094 }
1da177e4
LT
1095 }
1096 xfs_dqunlock(dqp);
1097
1098 /*
1099 * If we had a group quota inside the user quota as a hint,
1100 * release it now.
1101 */
1102 if (! gdqp)
1103 break;
1104 dqp = gdqp;
1105 }
3a8406f6 1106 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1da177e4
LT
1107}
1108
1109/*
1110 * Release a dquot. Flush it if dirty, then dqput() it.
1111 * dquot must not be locked.
1112 */
1113void
1114xfs_qm_dqrele(
1115 xfs_dquot_t *dqp)
1116{
7d095257
CH
1117 if (!dqp)
1118 return;
1119
0b1b213f 1120 trace_xfs_dqrele(dqp);
1da177e4
LT
1121
1122 xfs_dqlock(dqp);
1123 /*
1124 * We don't care to flush it if the dquot is dirty here.
1125 * That will create stutters that we want to avoid.
1126 * Instead we do a delayed write when we try to reclaim
1127 * a dirty dquot. Also xfs_sync will take part of the burden...
1128 */
1129 xfs_qm_dqput(dqp);
1130}
1131
ca30b2a7
CH
1132/*
1133 * This is the dquot flushing I/O completion routine. It is called
1134 * from interrupt level when the buffer containing the dquot is
1135 * flushed to disk. It is responsible for removing the dquot logitem
1136 * from the AIL if it has not been re-logged, and unlocking the dquot's
1137 * flush lock. This behavior is very similar to that of inodes..
1138 */
1139STATIC void
1140xfs_qm_dqflush_done(
1141 struct xfs_buf *bp,
1142 struct xfs_log_item *lip)
1143{
1144 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
1145 xfs_dquot_t *dqp = qip->qli_dquot;
1146 struct xfs_ail *ailp = lip->li_ailp;
1147
1148 /*
1149 * We only want to pull the item from the AIL if its
1150 * location in the log has not changed since we started the flush.
1151 * Thus, we only bother if the dquot's lsn has
1152 * not changed. First we check the lsn outside the lock
1153 * since it's cheaper, and then we recheck while
1154 * holding the lock before removing the dquot from the AIL.
1155 */
1156 if ((lip->li_flags & XFS_LI_IN_AIL) &&
1157 lip->li_lsn == qip->qli_flush_lsn) {
1158
1159 /* xfs_trans_ail_delete() drops the AIL lock. */
1160 spin_lock(&ailp->xa_lock);
1161 if (lip->li_lsn == qip->qli_flush_lsn)
1162 xfs_trans_ail_delete(ailp, lip);
1163 else
1164 spin_unlock(&ailp->xa_lock);
1165 }
1166
1167 /*
1168 * Release the dq's flush lock since we're done with it.
1169 */
1170 xfs_dqfunlock(dqp);
1171}
1da177e4
LT
1172
1173/*
1174 * Write a modified dquot to disk.
1175 * The dquot must be locked and the flush lock too taken by caller.
1176 * The flush lock will not be unlocked until the dquot reaches the disk,
1177 * but the dquot is free to be unlocked and modified by the caller
1178 * in the interim. Dquot is still locked on return. This behavior is
1179 * identical to that of inodes.
1180 */
1181int
1182xfs_qm_dqflush(
1183 xfs_dquot_t *dqp,
1184 uint flags)
1185{
1186 xfs_mount_t *mp;
1187 xfs_buf_t *bp;
1188 xfs_disk_dquot_t *ddqp;
1189 int error;
1da177e4
LT
1190
1191 ASSERT(XFS_DQ_IS_LOCKED(dqp));
e1f49cf2 1192 ASSERT(!completion_done(&dqp->q_flush));
0b1b213f 1193 trace_xfs_dqflush(dqp);
1da177e4
LT
1194
1195 /*
2f8a3ce1
DC
1196 * If not dirty, or it's pinned and we are not supposed to
1197 * block, nada.
1da177e4 1198 */
2f8a3ce1 1199 if (!XFS_DQ_IS_DIRTY(dqp) ||
20026d92 1200 (!(flags & SYNC_WAIT) && atomic_read(&dqp->q_pincount) > 0)) {
1da177e4 1201 xfs_dqfunlock(dqp);
2f8a3ce1 1202 return 0;
1da177e4 1203 }
1da177e4
LT
1204 xfs_qm_dqunpin_wait(dqp);
1205
1206 /*
1207 * This may have been unpinned because the filesystem is shutting
1208 * down forcibly. If that's the case we must not write this dquot
1209 * to disk, because the log record didn't make it to disk!
1210 */
1211 if (XFS_FORCED_SHUTDOWN(dqp->q_mount)) {
1212 dqp->dq_flags &= ~(XFS_DQ_DIRTY);
1213 xfs_dqfunlock(dqp);
1214 return XFS_ERROR(EIO);
1215 }
1216
1217 /*
1218 * Get the buffer containing the on-disk dquot
1219 * We don't need a transaction envelope because we know that the
1220 * the ondisk-dquot has already been allocated for.
1221 */
1222 if ((error = xfs_qm_dqtobp(NULL, dqp, &ddqp, &bp, XFS_QMOPT_DOWARN))) {
1da177e4
LT
1223 ASSERT(error != ENOENT);
1224 /*
1225 * Quotas could have gotten turned off (ESRCH)
1226 */
1227 xfs_dqfunlock(dqp);
1228 return (error);
1229 }
1230
1149d96a 1231 if (xfs_qm_dqcheck(&dqp->q_core, be32_to_cpu(ddqp->d_id),
c8ad20ff 1232 0, XFS_QMOPT_DOWARN, "dqflush (incore copy)")) {
7d04a335 1233 xfs_force_shutdown(dqp->q_mount, SHUTDOWN_CORRUPT_INCORE);
1da177e4
LT
1234 return XFS_ERROR(EIO);
1235 }
1236
1237 /* This is the only portion of data that needs to persist */
1238 memcpy(ddqp, &(dqp->q_core), sizeof(xfs_disk_dquot_t));
1239
1240 /*
1241 * Clear the dirty field and remember the flush lsn for later use.
1242 */
1243 dqp->dq_flags &= ~(XFS_DQ_DIRTY);
1244 mp = dqp->q_mount;
1245
7b2e2a31
DC
1246 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1247 &dqp->q_logitem.qli_item.li_lsn);
1da177e4
LT
1248
1249 /*
1250 * Attach an iodone routine so that we can remove this dquot from the
1251 * AIL and release the flush lock once the dquot is synced to disk.
1252 */
ca30b2a7
CH
1253 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1254 &dqp->q_logitem.qli_item);
1255
1da177e4
LT
1256 /*
1257 * If the buffer is pinned then push on the log so we won't
1258 * get stuck waiting in the write for too long.
1259 */
1260 if (XFS_BUF_ISPINNED(bp)) {
0b1b213f 1261 trace_xfs_dqflush_force(dqp);
a14a348b 1262 xfs_log_force(mp, 0);
1da177e4
LT
1263 }
1264
20026d92 1265 if (flags & SYNC_WAIT)
1da177e4 1266 error = xfs_bwrite(mp, bp);
20026d92
DC
1267 else
1268 xfs_bdwrite(mp, bp);
0b1b213f
CH
1269
1270 trace_xfs_dqflush_done(dqp);
1271
1da177e4
LT
1272 /*
1273 * dqp is still locked, but caller is free to unlock it now.
1274 */
20026d92 1275 return error;
1da177e4
LT
1276
1277}
1278
1da177e4
LT
1279int
1280xfs_qm_dqlock_nowait(
1281 xfs_dquot_t *dqp)
1282{
e1f49cf2 1283 return mutex_trylock(&dqp->q_qlock);
1da177e4
LT
1284}
1285
1286void
1287xfs_dqlock(
1288 xfs_dquot_t *dqp)
1289{
e1f49cf2 1290 mutex_lock(&dqp->q_qlock);
1da177e4
LT
1291}
1292
1293void
1294xfs_dqunlock(
1295 xfs_dquot_t *dqp)
1296{
1297 mutex_unlock(&(dqp->q_qlock));
1298 if (dqp->q_logitem.qli_dquot == dqp) {
1299 /* Once was dqp->q_mount, but might just have been cleared */
783a2f65 1300 xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_ailp,
1da177e4
LT
1301 (xfs_log_item_t*)&(dqp->q_logitem));
1302 }
1303}
1304
1305
1306void
1307xfs_dqunlock_nonotify(
1308 xfs_dquot_t *dqp)
1309{
1310 mutex_unlock(&(dqp->q_qlock));
1311}
1312
5bb87a33
CH
1313/*
1314 * Lock two xfs_dquot structures.
1315 *
1316 * To avoid deadlocks we always lock the quota structure with
1317 * the lowerd id first.
1318 */
1da177e4
LT
1319void
1320xfs_dqlock2(
1321 xfs_dquot_t *d1,
1322 xfs_dquot_t *d2)
1323{
1324 if (d1 && d2) {
1325 ASSERT(d1 != d2);
1149d96a
CH
1326 if (be32_to_cpu(d1->q_core.d_id) >
1327 be32_to_cpu(d2->q_core.d_id)) {
5bb87a33
CH
1328 mutex_lock(&d2->q_qlock);
1329 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1da177e4 1330 } else {
5bb87a33
CH
1331 mutex_lock(&d1->q_qlock);
1332 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1da177e4 1333 }
5bb87a33
CH
1334 } else if (d1) {
1335 mutex_lock(&d1->q_qlock);
1336 } else if (d2) {
1337 mutex_lock(&d2->q_qlock);
1da177e4
LT
1338 }
1339}
1340
1341
1342/*
1343 * Take a dquot out of the mount's dqlist as well as the hashlist.
1344 * This is called via unmount as well as quotaoff, and the purge
1345 * will always succeed unless there are soft (temp) references
1346 * outstanding.
1347 *
1348 * This returns 0 if it was purged, 1 if it wasn't. It's not an error code
1349 * that we're returning! XXXsup - not cool.
1350 */
1351/* ARGSUSED */
1352int
1353xfs_qm_dqpurge(
4f0e8a98 1354 xfs_dquot_t *dqp)
1da177e4 1355{
e6a81f13 1356 xfs_dqhash_t *qh = dqp->q_hash;
3c56836f 1357 xfs_mount_t *mp = dqp->q_mount;
1da177e4 1358
3a25404b 1359 ASSERT(mutex_is_locked(&mp->m_quotainfo->qi_dqlist_lock));
c9a192dc 1360 ASSERT(mutex_is_locked(&dqp->q_hash->qh_lock));
1da177e4
LT
1361
1362 xfs_dqlock(dqp);
1363 /*
1364 * We really can't afford to purge a dquot that is
1365 * referenced, because these are hard refs.
1366 * It shouldn't happen in general because we went thru _all_ inodes in
1367 * dqrele_all_inodes before calling this and didn't let the mountlock go.
1368 * However it is possible that we have dquots with temporary
1369 * references that are not attached to an inode. e.g. see xfs_setattr().
1370 */
1371 if (dqp->q_nrefs != 0) {
1372 xfs_dqunlock(dqp);
c9a192dc 1373 mutex_unlock(&dqp->q_hash->qh_lock);
1da177e4
LT
1374 return (1);
1375 }
1376
3a8406f6 1377 ASSERT(!list_empty(&dqp->q_freelist));
1da177e4
LT
1378
1379 /*
1380 * If we're turning off quotas, we have to make sure that, for
1381 * example, we don't delete quota disk blocks while dquots are
1382 * in the process of getting written to those disk blocks.
1383 * This dquot might well be on AIL, and we can't leave it there
1384 * if we're turning off quotas. Basically, we need this flush
1385 * lock, and are willing to block on it.
1386 */
e1f49cf2 1387 if (!xfs_dqflock_nowait(dqp)) {
1da177e4
LT
1388 /*
1389 * Block on the flush lock after nudging dquot buffer,
1390 * if it is incore.
1391 */
1392 xfs_qm_dqflock_pushbuf_wait(dqp);
1393 }
1394
1395 /*
1396 * XXXIf we're turning this type of quotas off, we don't care
1397 * about the dirty metadata sitting in this dquot. OTOH, if
1398 * we're unmounting, we do care, so we flush it and wait.
1399 */
1400 if (XFS_DQ_IS_DIRTY(dqp)) {
3c56836f 1401 int error;
0b1b213f 1402
1da177e4
LT
1403 /* dqflush unlocks dqflock */
1404 /*
1405 * Given that dqpurge is a very rare occurrence, it is OK
1406 * that we're holding the hashlist and mplist locks
1407 * across the disk write. But, ... XXXsup
1408 *
1409 * We don't care about getting disk errors here. We need
1410 * to purge this dquot anyway, so we go ahead regardless.
1411 */
20026d92 1412 error = xfs_qm_dqflush(dqp, SYNC_WAIT);
3c56836f
DC
1413 if (error)
1414 xfs_fs_cmn_err(CE_WARN, mp,
1415 "xfs_qm_dqpurge: dquot %p flush failed", dqp);
1da177e4
LT
1416 xfs_dqflock(dqp);
1417 }
bc3048e3 1418 ASSERT(atomic_read(&dqp->q_pincount) == 0);
1da177e4
LT
1419 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1420 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
1421
e6a81f13
DC
1422 list_del_init(&dqp->q_hashlist);
1423 qh->qh_version++;
3a25404b
DC
1424 list_del_init(&dqp->q_mplist);
1425 mp->m_quotainfo->qi_dqreclaims++;
1426 mp->m_quotainfo->qi_dquots--;
1da177e4
LT
1427 /*
1428 * XXX Move this to the front of the freelist, if we can get the
1429 * freelist lock.
1430 */
3a8406f6 1431 ASSERT(!list_empty(&dqp->q_freelist));
1da177e4
LT
1432
1433 dqp->q_mount = NULL;
1434 dqp->q_hash = NULL;
1435 dqp->dq_flags = XFS_DQ_INACTIVE;
1436 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
1437 xfs_dqfunlock(dqp);
1438 xfs_dqunlock(dqp);
e6a81f13 1439 mutex_unlock(&qh->qh_lock);
1da177e4
LT
1440 return (0);
1441}
1442
1443
1444#ifdef QUOTADEBUG
1445void
1446xfs_qm_dqprint(xfs_dquot_t *dqp)
1447{
1448 cmn_err(CE_DEBUG, "-----------KERNEL DQUOT----------------");
1449 cmn_err(CE_DEBUG, "---- dquotID = %d",
1149d96a 1450 (int)be32_to_cpu(dqp->q_core.d_id));
c8ad20ff 1451 cmn_err(CE_DEBUG, "---- type = %s", DQFLAGTO_TYPESTR(dqp));
1da177e4
LT
1452 cmn_err(CE_DEBUG, "---- fs = 0x%p", dqp->q_mount);
1453 cmn_err(CE_DEBUG, "---- blkno = 0x%x", (int) dqp->q_blkno);
1454 cmn_err(CE_DEBUG, "---- boffset = 0x%x", (int) dqp->q_bufoffset);
1455 cmn_err(CE_DEBUG, "---- blkhlimit = %Lu (0x%x)",
1149d96a
CH
1456 be64_to_cpu(dqp->q_core.d_blk_hardlimit),
1457 (int)be64_to_cpu(dqp->q_core.d_blk_hardlimit));
1da177e4 1458 cmn_err(CE_DEBUG, "---- blkslimit = %Lu (0x%x)",
1149d96a
CH
1459 be64_to_cpu(dqp->q_core.d_blk_softlimit),
1460 (int)be64_to_cpu(dqp->q_core.d_blk_softlimit));
1da177e4 1461 cmn_err(CE_DEBUG, "---- inohlimit = %Lu (0x%x)",
1149d96a
CH
1462 be64_to_cpu(dqp->q_core.d_ino_hardlimit),
1463 (int)be64_to_cpu(dqp->q_core.d_ino_hardlimit));
1da177e4 1464 cmn_err(CE_DEBUG, "---- inoslimit = %Lu (0x%x)",
1149d96a
CH
1465 be64_to_cpu(dqp->q_core.d_ino_softlimit),
1466 (int)be64_to_cpu(dqp->q_core.d_ino_softlimit));
1da177e4 1467 cmn_err(CE_DEBUG, "---- bcount = %Lu (0x%x)",
1149d96a
CH
1468 be64_to_cpu(dqp->q_core.d_bcount),
1469 (int)be64_to_cpu(dqp->q_core.d_bcount));
1da177e4 1470 cmn_err(CE_DEBUG, "---- icount = %Lu (0x%x)",
1149d96a
CH
1471 be64_to_cpu(dqp->q_core.d_icount),
1472 (int)be64_to_cpu(dqp->q_core.d_icount));
1da177e4 1473 cmn_err(CE_DEBUG, "---- btimer = %d",
1149d96a 1474 (int)be32_to_cpu(dqp->q_core.d_btimer));
1da177e4 1475 cmn_err(CE_DEBUG, "---- itimer = %d",
1149d96a 1476 (int)be32_to_cpu(dqp->q_core.d_itimer));
1da177e4
LT
1477 cmn_err(CE_DEBUG, "---------------------------");
1478}
1479#endif
1480
1481/*
1482 * Give the buffer a little push if it is incore and
1483 * wait on the flush lock.
1484 */
1485void
1486xfs_qm_dqflock_pushbuf_wait(
1487 xfs_dquot_t *dqp)
1488{
8a7b8a89 1489 xfs_mount_t *mp = dqp->q_mount;
1da177e4
LT
1490 xfs_buf_t *bp;
1491
1492 /*
1493 * Check to see if the dquot has been flushed delayed
1494 * write. If so, grab its buffer and send it
1495 * out immediately. We'll be able to acquire
1496 * the flush lock when the I/O completes.
1497 */
8a7b8a89
CH
1498 bp = xfs_incore(mp->m_ddev_targp, dqp->q_blkno,
1499 mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK);
7d6a7bde
DC
1500 if (!bp)
1501 goto out_lock;
1502
1503 if (XFS_BUF_ISDELAYWRITE(bp)) {
1504 if (XFS_BUF_ISPINNED(bp))
8a7b8a89 1505 xfs_log_force(mp, 0);
7d6a7bde
DC
1506 xfs_buf_delwri_promote(bp);
1507 wake_up_process(bp->b_target->bt_task);
1da177e4 1508 }
7d6a7bde
DC
1509 xfs_buf_relse(bp);
1510out_lock:
1da177e4
LT
1511 xfs_dqflock(dqp);
1512}