blob: 740d20d331874cee3bec1c0d2257f08629747acb [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (c) 2000-2003 Silicon Graphics, Inc. All Rights Reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11 *
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
18 *
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22 *
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
25 *
26 * http://www.sgi.com
27 *
28 * For further information regarding this notice, see:
29 *
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31 */
32
33#include "xfs.h"
34#include "xfs_fs.h"
35#include "xfs_inum.h"
36#include "xfs_log.h"
37#include "xfs_trans.h"
38#include "xfs_sb.h"
39#include "xfs_ag.h"
40#include "xfs_dir.h"
41#include "xfs_dir2.h"
42#include "xfs_alloc.h"
43#include "xfs_dmapi.h"
44#include "xfs_quota.h"
45#include "xfs_mount.h"
46#include "xfs_alloc_btree.h"
47#include "xfs_bmap_btree.h"
48#include "xfs_ialloc_btree.h"
49#include "xfs_btree.h"
50#include "xfs_ialloc.h"
51#include "xfs_attr_sf.h"
52#include "xfs_dir_sf.h"
53#include "xfs_dir2_sf.h"
54#include "xfs_dinode.h"
55#include "xfs_inode.h"
56#include "xfs_bmap.h"
57#include "xfs_bit.h"
58#include "xfs_rtalloc.h"
59#include "xfs_error.h"
60#include "xfs_itable.h"
61#include "xfs_rw.h"
62#include "xfs_acl.h"
63#include "xfs_cap.h"
64#include "xfs_mac.h"
65#include "xfs_attr.h"
66#include "xfs_buf_item.h"
67#include "xfs_trans_space.h"
68#include "xfs_trans_priv.h"
69
70#include "xfs_qm.h"
71
72
73/*
74 LOCK ORDER
75
76 inode lock (ilock)
77 dquot hash-chain lock (hashlock)
78 xqm dquot freelist lock (freelistlock
79 mount's dquot list lock (mplistlock)
80 user dquot lock - lock ordering among dquots is based on the uid or gid
81 group dquot lock - similar to udquots. Between the two dquots, the udquot
82 has to be locked first.
83 pin lock - the dquot lock must be held to take this lock.
84 flush lock - ditto.
85*/
86
87STATIC void xfs_qm_dqflush_done(xfs_buf_t *, xfs_dq_logitem_t *);
88
89#ifdef DEBUG
90xfs_buftarg_t *xfs_dqerror_target;
91int xfs_do_dqerror;
92int xfs_dqreq_num;
93int xfs_dqerror_mod = 33;
94#endif
95
96/*
97 * Allocate and initialize a dquot. We don't always allocate fresh memory;
98 * we try to reclaim a free dquot if the number of incore dquots are above
99 * a threshold.
100 * The only field inside the core that gets initialized at this point
101 * is the d_id field. The idea is to fill in the entire q_core
102 * when we read in the on disk dquot.
103 */
104xfs_dquot_t *
105xfs_qm_dqinit(
106 xfs_mount_t *mp,
107 xfs_dqid_t id,
108 uint type)
109{
110 xfs_dquot_t *dqp;
111 boolean_t brandnewdquot;
112
113 brandnewdquot = xfs_qm_dqalloc_incore(&dqp);
114 dqp->dq_flags = type;
115 INT_SET(dqp->q_core.d_id, ARCH_CONVERT, id);
116 dqp->q_mount = mp;
117
118 /*
119 * No need to re-initialize these if this is a reclaimed dquot.
120 */
121 if (brandnewdquot) {
122 dqp->dq_flnext = dqp->dq_flprev = dqp;
123 mutex_init(&dqp->q_qlock, MUTEX_DEFAULT, "xdq");
124 initnsema(&dqp->q_flock, 1, "fdq");
125 sv_init(&dqp->q_pinwait, SV_DEFAULT, "pdq");
126
127#ifdef XFS_DQUOT_TRACE
128 dqp->q_trace = ktrace_alloc(DQUOT_TRACE_SIZE, KM_SLEEP);
129 xfs_dqtrace_entry(dqp, "DQINIT");
130#endif
131 } else {
132 /*
133 * Only the q_core portion was zeroed in dqreclaim_one().
134 * So, we need to reset others.
135 */
136 dqp->q_nrefs = 0;
137 dqp->q_blkno = 0;
138 dqp->MPL_NEXT = dqp->HL_NEXT = NULL;
139 dqp->HL_PREVP = dqp->MPL_PREVP = NULL;
140 dqp->q_bufoffset = 0;
141 dqp->q_fileoffset = 0;
142 dqp->q_transp = NULL;
143 dqp->q_gdquot = NULL;
144 dqp->q_res_bcount = 0;
145 dqp->q_res_icount = 0;
146 dqp->q_res_rtbcount = 0;
147 dqp->q_pincount = 0;
148 dqp->q_hash = NULL;
149 ASSERT(dqp->dq_flnext == dqp->dq_flprev);
150
151#ifdef XFS_DQUOT_TRACE
152 ASSERT(dqp->q_trace);
153 xfs_dqtrace_entry(dqp, "DQRECLAIMED_INIT");
154#endif
155 }
156
157 /*
158 * log item gets initialized later
159 */
160 return (dqp);
161}
162
163/*
164 * This is called to free all the memory associated with a dquot
165 */
166void
167xfs_qm_dqdestroy(
168 xfs_dquot_t *dqp)
169{
170 ASSERT(! XFS_DQ_IS_ON_FREELIST(dqp));
171
172 mutex_destroy(&dqp->q_qlock);
173 freesema(&dqp->q_flock);
174 sv_destroy(&dqp->q_pinwait);
175
176#ifdef XFS_DQUOT_TRACE
177 if (dqp->q_trace)
178 ktrace_free(dqp->q_trace);
179 dqp->q_trace = NULL;
180#endif
181 kmem_zone_free(xfs_Gqm->qm_dqzone, dqp);
182 atomic_dec(&xfs_Gqm->qm_totaldquots);
183}
184
185/*
186 * This is what a 'fresh' dquot inside a dquot chunk looks like on disk.
187 */
188STATIC void
189xfs_qm_dqinit_core(
190 xfs_dqid_t id,
191 uint type,
192 xfs_dqblk_t *d)
193{
194 /*
195 * Caller has zero'd the entire dquot 'chunk' already.
196 */
197 INT_SET(d->dd_diskdq.d_magic, ARCH_CONVERT, XFS_DQUOT_MAGIC);
198 INT_SET(d->dd_diskdq.d_version, ARCH_CONVERT, XFS_DQUOT_VERSION);
199 INT_SET(d->dd_diskdq.d_id, ARCH_CONVERT, id);
200 INT_SET(d->dd_diskdq.d_flags, ARCH_CONVERT, type);
201}
202
203
204#ifdef XFS_DQUOT_TRACE
205/*
206 * Dquot tracing for debugging.
207 */
208/* ARGSUSED */
209void
210__xfs_dqtrace_entry(
211 xfs_dquot_t *dqp,
212 char *func,
213 void *retaddr,
214 xfs_inode_t *ip)
215{
216 xfs_dquot_t *udqp = NULL;
217 xfs_ino_t ino = 0;
218
219 ASSERT(dqp->q_trace);
220 if (ip) {
221 ino = ip->i_ino;
222 udqp = ip->i_udquot;
223 }
224 ktrace_enter(dqp->q_trace,
225 (void *)(__psint_t)DQUOT_KTRACE_ENTRY,
226 (void *)func,
227 (void *)(__psint_t)dqp->q_nrefs,
228 (void *)(__psint_t)dqp->dq_flags,
229 (void *)(__psint_t)dqp->q_res_bcount,
230 (void *)(__psint_t)INT_GET(dqp->q_core.d_bcount,
231 ARCH_CONVERT),
232 (void *)(__psint_t)INT_GET(dqp->q_core.d_icount,
233 ARCH_CONVERT),
234 (void *)(__psint_t)INT_GET(dqp->q_core.d_blk_hardlimit,
235 ARCH_CONVERT),
236 (void *)(__psint_t)INT_GET(dqp->q_core.d_blk_softlimit,
237 ARCH_CONVERT),
238 (void *)(__psint_t)INT_GET(dqp->q_core.d_ino_hardlimit,
239 ARCH_CONVERT),
240 (void *)(__psint_t)INT_GET(dqp->q_core.d_ino_softlimit,
241 ARCH_CONVERT),
242 (void *)(__psint_t)INT_GET(dqp->q_core.d_id, ARCH_CONVERT),
243 (void *)(__psint_t)current_pid(),
244 (void *)(__psint_t)ino,
245 (void *)(__psint_t)retaddr,
246 (void *)(__psint_t)udqp);
247 return;
248}
249#endif
250
251
252/*
253 * If default limits are in force, push them into the dquot now.
254 * We overwrite the dquot limits only if they are zero and this
255 * is not the root dquot.
256 */
257void
258xfs_qm_adjust_dqlimits(
259 xfs_mount_t *mp,
260 xfs_disk_dquot_t *d)
261{
262 xfs_quotainfo_t *q = mp->m_quotainfo;
263
264 ASSERT(d->d_id);
265
266 if (q->qi_bsoftlimit && !d->d_blk_softlimit)
267 INT_SET(d->d_blk_softlimit, ARCH_CONVERT, q->qi_bsoftlimit);
268 if (q->qi_bhardlimit && !d->d_blk_hardlimit)
269 INT_SET(d->d_blk_hardlimit, ARCH_CONVERT, q->qi_bhardlimit);
270 if (q->qi_isoftlimit && !d->d_ino_softlimit)
271 INT_SET(d->d_ino_softlimit, ARCH_CONVERT, q->qi_isoftlimit);
272 if (q->qi_ihardlimit && !d->d_ino_hardlimit)
273 INT_SET(d->d_ino_hardlimit, ARCH_CONVERT, q->qi_ihardlimit);
274 if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit)
275 INT_SET(d->d_rtb_softlimit, ARCH_CONVERT, q->qi_rtbsoftlimit);
276 if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit)
277 INT_SET(d->d_rtb_hardlimit, ARCH_CONVERT, q->qi_rtbhardlimit);
278}
279
280/*
281 * Check the limits and timers of a dquot and start or reset timers
282 * if necessary.
283 * This gets called even when quota enforcement is OFF, which makes our
284 * life a little less complicated. (We just don't reject any quota
285 * reservations in that case, when enforcement is off).
286 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
287 * enforcement's off.
288 * In contrast, warnings are a little different in that they don't
289 * 'automatically' get started when limits get exceeded.
290 */
291void
292xfs_qm_adjust_dqtimers(
293 xfs_mount_t *mp,
294 xfs_disk_dquot_t *d)
295{
296 ASSERT(d->d_id);
297
298#ifdef QUOTADEBUG
299 if (INT_GET(d->d_blk_hardlimit, ARCH_CONVERT))
300 ASSERT(INT_GET(d->d_blk_softlimit, ARCH_CONVERT) <=
301 INT_GET(d->d_blk_hardlimit, ARCH_CONVERT));
302 if (INT_GET(d->d_ino_hardlimit, ARCH_CONVERT))
303 ASSERT(INT_GET(d->d_ino_softlimit, ARCH_CONVERT) <=
304 INT_GET(d->d_ino_hardlimit, ARCH_CONVERT));
305 if (INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT))
306 ASSERT(INT_GET(d->d_rtb_softlimit, ARCH_CONVERT) <=
307 INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT));
308#endif
309 if (!d->d_btimer) {
310 if ((INT_GET(d->d_blk_softlimit, ARCH_CONVERT) &&
311 (INT_GET(d->d_bcount, ARCH_CONVERT) >=
312 INT_GET(d->d_blk_softlimit, ARCH_CONVERT))) ||
313 (INT_GET(d->d_blk_hardlimit, ARCH_CONVERT) &&
314 (INT_GET(d->d_bcount, ARCH_CONVERT) >=
315 INT_GET(d->d_blk_hardlimit, ARCH_CONVERT)))) {
316 INT_SET(d->d_btimer, ARCH_CONVERT,
317 get_seconds() + XFS_QI_BTIMELIMIT(mp));
318 }
319 } else {
320 if ((!d->d_blk_softlimit ||
321 (INT_GET(d->d_bcount, ARCH_CONVERT) <
322 INT_GET(d->d_blk_softlimit, ARCH_CONVERT))) &&
323 (!d->d_blk_hardlimit ||
324 (INT_GET(d->d_bcount, ARCH_CONVERT) <
325 INT_GET(d->d_blk_hardlimit, ARCH_CONVERT)))) {
326 d->d_btimer = 0;
327 }
328 }
329
330 if (!d->d_itimer) {
331 if ((INT_GET(d->d_ino_softlimit, ARCH_CONVERT) &&
332 (INT_GET(d->d_icount, ARCH_CONVERT) >=
333 INT_GET(d->d_ino_softlimit, ARCH_CONVERT))) ||
334 (INT_GET(d->d_ino_hardlimit, ARCH_CONVERT) &&
335 (INT_GET(d->d_icount, ARCH_CONVERT) >=
336 INT_GET(d->d_ino_hardlimit, ARCH_CONVERT)))) {
337 INT_SET(d->d_itimer, ARCH_CONVERT,
338 get_seconds() + XFS_QI_ITIMELIMIT(mp));
339 }
340 } else {
341 if ((!d->d_ino_softlimit ||
342 (INT_GET(d->d_icount, ARCH_CONVERT) <
343 INT_GET(d->d_ino_softlimit, ARCH_CONVERT))) &&
344 (!d->d_ino_hardlimit ||
345 (INT_GET(d->d_icount, ARCH_CONVERT) <
346 INT_GET(d->d_ino_hardlimit, ARCH_CONVERT)))) {
347 d->d_itimer = 0;
348 }
349 }
350
351 if (!d->d_rtbtimer) {
352 if ((INT_GET(d->d_rtb_softlimit, ARCH_CONVERT) &&
353 (INT_GET(d->d_rtbcount, ARCH_CONVERT) >=
354 INT_GET(d->d_rtb_softlimit, ARCH_CONVERT))) ||
355 (INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT) &&
356 (INT_GET(d->d_rtbcount, ARCH_CONVERT) >=
357 INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT)))) {
358 INT_SET(d->d_rtbtimer, ARCH_CONVERT,
359 get_seconds() + XFS_QI_RTBTIMELIMIT(mp));
360 }
361 } else {
362 if ((!d->d_rtb_softlimit ||
363 (INT_GET(d->d_rtbcount, ARCH_CONVERT) <
364 INT_GET(d->d_rtb_softlimit, ARCH_CONVERT))) &&
365 (!d->d_rtb_hardlimit ||
366 (INT_GET(d->d_rtbcount, ARCH_CONVERT) <
367 INT_GET(d->d_rtb_hardlimit, ARCH_CONVERT)))) {
368 d->d_rtbtimer = 0;
369 }
370 }
371}
372
373/*
374 * Increment or reset warnings of a given dquot.
375 */
376int
377xfs_qm_dqwarn(
378 xfs_disk_dquot_t *d,
379 uint flags)
380{
381 int warned;
382
383 /*
384 * root's limits are not real limits.
385 */
386 if (!d->d_id)
387 return (0);
388
389 warned = 0;
390 if (INT_GET(d->d_blk_softlimit, ARCH_CONVERT) &&
391 (INT_GET(d->d_bcount, ARCH_CONVERT) >=
392 INT_GET(d->d_blk_softlimit, ARCH_CONVERT))) {
393 if (flags & XFS_QMOPT_DOWARN) {
394 INT_MOD(d->d_bwarns, ARCH_CONVERT, +1);
395 warned++;
396 }
397 } else {
398 if (!d->d_blk_softlimit ||
399 (INT_GET(d->d_bcount, ARCH_CONVERT) <
400 INT_GET(d->d_blk_softlimit, ARCH_CONVERT))) {
401 d->d_bwarns = 0;
402 }
403 }
404
405 if (INT_GET(d->d_ino_softlimit, ARCH_CONVERT) > 0 &&
406 (INT_GET(d->d_icount, ARCH_CONVERT) >=
407 INT_GET(d->d_ino_softlimit, ARCH_CONVERT))) {
408 if (flags & XFS_QMOPT_DOWARN) {
409 INT_MOD(d->d_iwarns, ARCH_CONVERT, +1);
410 warned++;
411 }
412 } else {
413 if (!d->d_ino_softlimit ||
414 (INT_GET(d->d_icount, ARCH_CONVERT) <
415 INT_GET(d->d_ino_softlimit, ARCH_CONVERT))) {
416 d->d_iwarns = 0;
417 }
418 }
419#ifdef QUOTADEBUG
420 if (INT_GET(d->d_iwarns, ARCH_CONVERT))
421 cmn_err(CE_DEBUG,
422 "--------@@Inode warnings running : %Lu >= %Lu",
423 INT_GET(d->d_icount, ARCH_CONVERT),
424 INT_GET(d->d_ino_softlimit, ARCH_CONVERT));
425 if (INT_GET(d->d_bwarns, ARCH_CONVERT))
426 cmn_err(CE_DEBUG,
427 "--------@@Blks warnings running : %Lu >= %Lu",
428 INT_GET(d->d_bcount, ARCH_CONVERT),
429 INT_GET(d->d_blk_softlimit, ARCH_CONVERT));
430#endif
431 return (warned);
432}
433
434
435/*
436 * initialize a buffer full of dquots and log the whole thing
437 */
438STATIC void
439xfs_qm_init_dquot_blk(
440 xfs_trans_t *tp,
441 xfs_mount_t *mp,
442 xfs_dqid_t id,
443 uint type,
444 xfs_buf_t *bp)
445{
446 xfs_dqblk_t *d;
447 int curid, i;
448
449 ASSERT(tp);
450 ASSERT(XFS_BUF_ISBUSY(bp));
451 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
452
453 d = (xfs_dqblk_t *)XFS_BUF_PTR(bp);
454
455 /*
456 * ID of the first dquot in the block - id's are zero based.
457 */
458 curid = id - (id % XFS_QM_DQPERBLK(mp));
459 ASSERT(curid >= 0);
460 memset(d, 0, BBTOB(XFS_QI_DQCHUNKLEN(mp)));
461 for (i = 0; i < XFS_QM_DQPERBLK(mp); i++, d++, curid++)
462 xfs_qm_dqinit_core(curid, type, d);
463 xfs_trans_dquot_buf(tp, bp,
464 type & XFS_DQ_USER ?
465 XFS_BLI_UDQUOT_BUF :
466 XFS_BLI_GDQUOT_BUF);
467 xfs_trans_log_buf(tp, bp, 0, BBTOB(XFS_QI_DQCHUNKLEN(mp)) - 1);
468}
469
470
471
472/*
473 * Allocate a block and fill it with dquots.
474 * This is called when the bmapi finds a hole.
475 */
476STATIC int
477xfs_qm_dqalloc(
478 xfs_trans_t *tp,
479 xfs_mount_t *mp,
480 xfs_dquot_t *dqp,
481 xfs_inode_t *quotip,
482 xfs_fileoff_t offset_fsb,
483 xfs_buf_t **O_bpp)
484{
485 xfs_fsblock_t firstblock;
486 xfs_bmap_free_t flist;
487 xfs_bmbt_irec_t map;
488 int nmaps, error, committed;
489 xfs_buf_t *bp;
490
491 ASSERT(tp != NULL);
492 xfs_dqtrace_entry(dqp, "DQALLOC");
493
494 /*
495 * Initialize the bmap freelist prior to calling bmapi code.
496 */
497 XFS_BMAP_INIT(&flist, &firstblock);
498 xfs_ilock(quotip, XFS_ILOCK_EXCL);
499 /*
500 * Return if this type of quotas is turned off while we didn't
501 * have an inode lock
502 */
503 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
504 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
505 return (ESRCH);
506 }
507
508 /*
509 * xfs_trans_commit normally decrements the vnode ref count
510 * when it unlocks the inode. Since we want to keep the quota
511 * inode around, we bump the vnode ref count now.
512 */
513 VN_HOLD(XFS_ITOV(quotip));
514
515 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
516 nmaps = 1;
517 if ((error = xfs_bmapi(tp, quotip,
518 offset_fsb, XFS_DQUOT_CLUSTER_SIZE_FSB,
519 XFS_BMAPI_METADATA | XFS_BMAPI_WRITE,
520 &firstblock,
521 XFS_QM_DQALLOC_SPACE_RES(mp),
522 &map, &nmaps, &flist))) {
523 goto error0;
524 }
525 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
526 ASSERT(nmaps == 1);
527 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
528 (map.br_startblock != HOLESTARTBLOCK));
529
530 /*
531 * Keep track of the blkno to save a lookup later
532 */
533 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
534
535 /* now we can just get the buffer (there's nothing to read yet) */
536 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
537 dqp->q_blkno,
538 XFS_QI_DQCHUNKLEN(mp),
539 0);
540 if (!bp || (error = XFS_BUF_GETERROR(bp)))
541 goto error1;
542 /*
543 * Make a chunk of dquots out of this buffer and log
544 * the entire thing.
545 */
546 xfs_qm_init_dquot_blk(tp, mp, INT_GET(dqp->q_core.d_id, ARCH_CONVERT),
547 dqp->dq_flags & (XFS_DQ_USER|XFS_DQ_GROUP),
548 bp);
549
550 if ((error = xfs_bmap_finish(&tp, &flist, firstblock, &committed))) {
551 goto error1;
552 }
553
554 *O_bpp = bp;
555 return 0;
556
557 error1:
558 xfs_bmap_cancel(&flist);
559 error0:
560 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
561
562 return (error);
563}
564
565/*
566 * Maps a dquot to the buffer containing its on-disk version.
567 * This returns a ptr to the buffer containing the on-disk dquot
568 * in the bpp param, and a ptr to the on-disk dquot within that buffer
569 */
570STATIC int
571xfs_qm_dqtobp(
572 xfs_trans_t *tp,
573 xfs_dquot_t *dqp,
574 xfs_disk_dquot_t **O_ddpp,
575 xfs_buf_t **O_bpp,
576 uint flags)
577{
578 xfs_bmbt_irec_t map;
579 int nmaps, error;
580 xfs_buf_t *bp;
581 xfs_inode_t *quotip;
582 xfs_mount_t *mp;
583 xfs_disk_dquot_t *ddq;
584 xfs_dqid_t id;
585 boolean_t newdquot;
586
587 mp = dqp->q_mount;
588 id = INT_GET(dqp->q_core.d_id, ARCH_CONVERT);
589 nmaps = 1;
590 newdquot = B_FALSE;
591
592 /*
593 * If we don't know where the dquot lives, find out.
594 */
595 if (dqp->q_blkno == (xfs_daddr_t) 0) {
596 /* We use the id as an index */
597 dqp->q_fileoffset = (xfs_fileoff_t) ((uint)id /
598 XFS_QM_DQPERBLK(mp));
599 nmaps = 1;
600 quotip = XFS_DQ_TO_QIP(dqp);
601 xfs_ilock(quotip, XFS_ILOCK_SHARED);
602 /*
603 * Return if this type of quotas is turned off while we didn't
604 * have an inode lock
605 */
606 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
607 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
608 return (ESRCH);
609 }
610 /*
611 * Find the block map; no allocations yet
612 */
613 error = xfs_bmapi(NULL, quotip, dqp->q_fileoffset,
614 XFS_DQUOT_CLUSTER_SIZE_FSB,
615 XFS_BMAPI_METADATA,
616 NULL, 0, &map, &nmaps, NULL);
617
618 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
619 if (error)
620 return (error);
621 ASSERT(nmaps == 1);
622 ASSERT(map.br_blockcount == 1);
623
624 /*
625 * offset of dquot in the (fixed sized) dquot chunk.
626 */
627 dqp->q_bufoffset = (id % XFS_QM_DQPERBLK(mp)) *
628 sizeof(xfs_dqblk_t);
629 if (map.br_startblock == HOLESTARTBLOCK) {
630 /*
631 * We don't allocate unless we're asked to
632 */
633 if (!(flags & XFS_QMOPT_DQALLOC))
634 return (ENOENT);
635
636 ASSERT(tp);
637 if ((error = xfs_qm_dqalloc(tp, mp, dqp, quotip,
638 dqp->q_fileoffset, &bp)))
639 return (error);
640 newdquot = B_TRUE;
641 } else {
642 /*
643 * store the blkno etc so that we don't have to do the
644 * mapping all the time
645 */
646 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
647 }
648 }
649 ASSERT(dqp->q_blkno != DELAYSTARTBLOCK);
650 ASSERT(dqp->q_blkno != HOLESTARTBLOCK);
651
652 /*
653 * Read in the buffer, unless we've just done the allocation
654 * (in which case we already have the buf).
655 */
656 if (! newdquot) {
657 xfs_dqtrace_entry(dqp, "DQTOBP READBUF");
658 if ((error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
659 dqp->q_blkno,
660 XFS_QI_DQCHUNKLEN(mp),
661 0, &bp))) {
662 return (error);
663 }
664 if (error || !bp)
665 return XFS_ERROR(error);
666 }
667 ASSERT(XFS_BUF_ISBUSY(bp));
668 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
669
670 /*
671 * calculate the location of the dquot inside the buffer.
672 */
673 ddq = (xfs_disk_dquot_t *)((char *)XFS_BUF_PTR(bp) + dqp->q_bufoffset);
674
675 /*
676 * A simple sanity check in case we got a corrupted dquot...
677 */
678 if (xfs_qm_dqcheck(ddq, id,
679 dqp->dq_flags & (XFS_DQ_USER|XFS_DQ_GROUP),
680 flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN),
681 "dqtobp")) {
682 if (!(flags & XFS_QMOPT_DQREPAIR)) {
683 xfs_trans_brelse(tp, bp);
684 return XFS_ERROR(EIO);
685 }
686 XFS_BUF_BUSY(bp); /* We dirtied this */
687 }
688
689 *O_bpp = bp;
690 *O_ddpp = ddq;
691
692 return (0);
693}
694
695
696/*
697 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
698 * and release the buffer immediately.
699 *
700 */
701/* ARGSUSED */
702STATIC int
703xfs_qm_dqread(
704 xfs_trans_t *tp,
705 xfs_dqid_t id,
706 xfs_dquot_t *dqp, /* dquot to get filled in */
707 uint flags)
708{
709 xfs_disk_dquot_t *ddqp;
710 xfs_buf_t *bp;
711 int error;
712
713 /*
714 * get a pointer to the on-disk dquot and the buffer containing it
715 * dqp already knows its own type (GROUP/USER).
716 */
717 xfs_dqtrace_entry(dqp, "DQREAD");
718 if ((error = xfs_qm_dqtobp(tp, dqp, &ddqp, &bp, flags))) {
719 return (error);
720 }
721
722 /* copy everything from disk dquot to the incore dquot */
723 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
724 ASSERT(INT_GET(dqp->q_core.d_id, ARCH_CONVERT) == id);
725 xfs_qm_dquot_logitem_init(dqp);
726
727 /*
728 * Reservation counters are defined as reservation plus current usage
729 * to avoid having to add everytime.
730 */
731 dqp->q_res_bcount = INT_GET(ddqp->d_bcount, ARCH_CONVERT);
732 dqp->q_res_icount = INT_GET(ddqp->d_icount, ARCH_CONVERT);
733 dqp->q_res_rtbcount = INT_GET(ddqp->d_rtbcount, ARCH_CONVERT);
734
735 /* Mark the buf so that this will stay incore a little longer */
736 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DQUOT, XFS_DQUOT_REF);
737
738 /*
739 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
740 * So we need to release with xfs_trans_brelse().
741 * The strategy here is identical to that of inodes; we lock
742 * the dquot in xfs_qm_dqget() before making it accessible to
743 * others. This is because dquots, like inodes, need a good level of
744 * concurrency, and we don't want to take locks on the entire buffers
745 * for dquot accesses.
746 * Note also that the dquot buffer may even be dirty at this point, if
747 * this particular dquot was repaired. We still aren't afraid to
748 * brelse it because we have the changes incore.
749 */
750 ASSERT(XFS_BUF_ISBUSY(bp));
751 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
752 xfs_trans_brelse(tp, bp);
753
754 return (error);
755}
756
757
758/*
759 * allocate an incore dquot from the kernel heap,
760 * and fill its core with quota information kept on disk.
761 * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk
762 * if it wasn't already allocated.
763 */
764STATIC int
765xfs_qm_idtodq(
766 xfs_mount_t *mp,
767 xfs_dqid_t id, /* gid or uid, depending on type */
768 uint type, /* UDQUOT or GDQUOT */
769 uint flags, /* DQALLOC, DQREPAIR */
770 xfs_dquot_t **O_dqpp)/* OUT : incore dquot, not locked */
771{
772 xfs_dquot_t *dqp;
773 int error;
774 xfs_trans_t *tp;
775 int cancelflags=0;
776
777 dqp = xfs_qm_dqinit(mp, id, type);
778 tp = NULL;
779 if (flags & XFS_QMOPT_DQALLOC) {
780 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
781 if ((error = xfs_trans_reserve(tp,
782 XFS_QM_DQALLOC_SPACE_RES(mp),
783 XFS_WRITE_LOG_RES(mp) +
784 BBTOB(XFS_QI_DQCHUNKLEN(mp)) - 1 +
785 128,
786 0,
787 XFS_TRANS_PERM_LOG_RES,
788 XFS_WRITE_LOG_COUNT))) {
789 cancelflags = 0;
790 goto error0;
791 }
792 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
793 }
794
795 /*
796 * Read it from disk; xfs_dqread() takes care of
797 * all the necessary initialization of dquot's fields (locks, etc)
798 */
799 if ((error = xfs_qm_dqread(tp, id, dqp, flags))) {
800 /*
801 * This can happen if quotas got turned off (ESRCH),
802 * or if the dquot didn't exist on disk and we ask to
803 * allocate (ENOENT).
804 */
805 xfs_dqtrace_entry(dqp, "DQREAD FAIL");
806 cancelflags |= XFS_TRANS_ABORT;
807 goto error0;
808 }
809 if (tp) {
810 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES,
811 NULL)))
812 goto error1;
813 }
814
815 *O_dqpp = dqp;
816 return (0);
817
818 error0:
819 ASSERT(error);
820 if (tp)
821 xfs_trans_cancel(tp, cancelflags);
822 error1:
823 xfs_qm_dqdestroy(dqp);
824 *O_dqpp = NULL;
825 return (error);
826}
827
828/*
829 * Lookup a dquot in the incore dquot hashtable. We keep two separate
830 * hashtables for user and group dquots; and, these are global tables
831 * inside the XQM, not per-filesystem tables.
832 * The hash chain must be locked by caller, and it is left locked
833 * on return. Returning dquot is locked.
834 */
835STATIC int
836xfs_qm_dqlookup(
837 xfs_mount_t *mp,
838 xfs_dqid_t id,
839 xfs_dqhash_t *qh,
840 xfs_dquot_t **O_dqpp)
841{
842 xfs_dquot_t *dqp;
843 uint flist_locked;
844 xfs_dquot_t *d;
845
846 ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
847
848 flist_locked = B_FALSE;
849
850 /*
851 * Traverse the hashchain looking for a match
852 */
853 for (dqp = qh->qh_next; dqp != NULL; dqp = dqp->HL_NEXT) {
854 /*
855 * We already have the hashlock. We don't need the
856 * dqlock to look at the id field of the dquot, since the
857 * id can't be modified without the hashlock anyway.
858 */
859 if (INT_GET(dqp->q_core.d_id, ARCH_CONVERT) == id && dqp->q_mount == mp) {
860 xfs_dqtrace_entry(dqp, "DQFOUND BY LOOKUP");
861 /*
862 * All in core dquots must be on the dqlist of mp
863 */
864 ASSERT(dqp->MPL_PREVP != NULL);
865
866 xfs_dqlock(dqp);
867 if (dqp->q_nrefs == 0) {
868 ASSERT (XFS_DQ_IS_ON_FREELIST(dqp));
869 if (! xfs_qm_freelist_lock_nowait(xfs_Gqm)) {
870 xfs_dqtrace_entry(dqp, "DQLOOKUP: WANT");
871
872 /*
873 * We may have raced with dqreclaim_one()
874 * (and lost). So, flag that we don't
875 * want the dquot to be reclaimed.
876 */
877 dqp->dq_flags |= XFS_DQ_WANT;
878 xfs_dqunlock(dqp);
879 xfs_qm_freelist_lock(xfs_Gqm);
880 xfs_dqlock(dqp);
881 dqp->dq_flags &= ~(XFS_DQ_WANT);
882 }
883 flist_locked = B_TRUE;
884 }
885
886 /*
887 * id couldn't have changed; we had the hashlock all
888 * along
889 */
890 ASSERT(INT_GET(dqp->q_core.d_id, ARCH_CONVERT) == id);
891
892 if (flist_locked) {
893 if (dqp->q_nrefs != 0) {
894 xfs_qm_freelist_unlock(xfs_Gqm);
895 flist_locked = B_FALSE;
896 } else {
897 /*
898 * take it off the freelist
899 */
900 xfs_dqtrace_entry(dqp,
901 "DQLOOKUP: TAKEOFF FL");
902 XQM_FREELIST_REMOVE(dqp);
903 /* xfs_qm_freelist_print(&(xfs_Gqm->
904 qm_dqfreelist),
905 "after removal"); */
906 }
907 }
908
909 /*
910 * grab a reference
911 */
912 XFS_DQHOLD(dqp);
913
914 if (flist_locked)
915 xfs_qm_freelist_unlock(xfs_Gqm);
916 /*
917 * move the dquot to the front of the hashchain
918 */
919 ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
920 if (dqp->HL_PREVP != &qh->qh_next) {
921 xfs_dqtrace_entry(dqp,
922 "DQLOOKUP: HASH MOVETOFRONT");
923 if ((d = dqp->HL_NEXT))
924 d->HL_PREVP = dqp->HL_PREVP;
925 *(dqp->HL_PREVP) = d;
926 d = qh->qh_next;
927 d->HL_PREVP = &dqp->HL_NEXT;
928 dqp->HL_NEXT = d;
929 dqp->HL_PREVP = &qh->qh_next;
930 qh->qh_next = dqp;
931 }
932 xfs_dqtrace_entry(dqp, "LOOKUP END");
933 *O_dqpp = dqp;
934 ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
935 return (0);
936 }
937 }
938
939 *O_dqpp = NULL;
940 ASSERT(XFS_DQ_IS_HASH_LOCKED(qh));
941 return (1);
942}
943
944/*
945 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
946 * a locked dquot, doing an allocation (if requested) as needed.
947 * When both an inode and an id are given, the inode's id takes precedence.
948 * That is, if the id changes while we don't hold the ilock inside this
949 * function, the new dquot is returned, not necessarily the one requested
950 * in the id argument.
951 */
952int
953xfs_qm_dqget(
954 xfs_mount_t *mp,
955 xfs_inode_t *ip, /* locked inode (optional) */
956 xfs_dqid_t id, /* gid or uid, depending on type */
957 uint type, /* UDQUOT or GDQUOT */
958 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
959 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
960{
961 xfs_dquot_t *dqp;
962 xfs_dqhash_t *h;
963 uint version;
964 int error;
965
966 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
967 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
968 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
969 return (ESRCH);
970 }
971 h = XFS_DQ_HASH(mp, id, type);
972
973#ifdef DEBUG
974 if (xfs_do_dqerror) {
975 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
976 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
977 cmn_err(CE_DEBUG, "Returning error in dqget");
978 return (EIO);
979 }
980 }
981#endif
982
983 again:
984
985#ifdef DEBUG
986 ASSERT(type == XFS_DQ_USER || type == XFS_DQ_GROUP);
987 if (ip) {
988 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
989 if (type == XFS_DQ_USER)
990 ASSERT(ip->i_udquot == NULL);
991 else
992 ASSERT(ip->i_gdquot == NULL);
993 }
994#endif
995 XFS_DQ_HASH_LOCK(h);
996
997 /*
998 * Look in the cache (hashtable).
999 * The chain is kept locked during lookup.
1000 */
1001 if (xfs_qm_dqlookup(mp, id, h, O_dqpp) == 0) {
1002 XQM_STATS_INC(xqmstats.xs_qm_dqcachehits);
1003 /*
1004 * The dquot was found, moved to the front of the chain,
1005 * taken off the freelist if it was on it, and locked
1006 * at this point. Just unlock the hashchain and return.
1007 */
1008 ASSERT(*O_dqpp);
1009 ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp));
1010 XFS_DQ_HASH_UNLOCK(h);
1011 xfs_dqtrace_entry(*O_dqpp, "DQGET DONE (FROM CACHE)");
1012 return (0); /* success */
1013 }
1014 XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses);
1015
1016 /*
1017 * Dquot cache miss. We don't want to keep the inode lock across
1018 * a (potential) disk read. Also we don't want to deal with the lock
1019 * ordering between quotainode and this inode. OTOH, dropping the inode
1020 * lock here means dealing with a chown that can happen before
1021 * we re-acquire the lock.
1022 */
1023 if (ip)
1024 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1025 /*
1026 * Save the hashchain version stamp, and unlock the chain, so that
1027 * we don't keep the lock across a disk read
1028 */
1029 version = h->qh_version;
1030 XFS_DQ_HASH_UNLOCK(h);
1031
1032 /*
1033 * Allocate the dquot on the kernel heap, and read the ondisk
1034 * portion off the disk. Also, do all the necessary initialization
1035 * This can return ENOENT if dquot didn't exist on disk and we didn't
1036 * ask it to allocate; ESRCH if quotas got turned off suddenly.
1037 */
1038 if ((error = xfs_qm_idtodq(mp, id, type,
1039 flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR|
1040 XFS_QMOPT_DOWARN),
1041 &dqp))) {
1042 if (ip)
1043 xfs_ilock(ip, XFS_ILOCK_EXCL);
1044 return (error);
1045 }
1046
1047 /*
1048 * See if this is mount code calling to look at the overall quota limits
1049 * which are stored in the id == 0 user or group's dquot.
1050 * Since we may not have done a quotacheck by this point, just return
1051 * the dquot without attaching it to any hashtables, lists, etc, or even
1052 * taking a reference.
1053 * The caller must dqdestroy this once done.
1054 */
1055 if (flags & XFS_QMOPT_DQSUSER) {
1056 ASSERT(id == 0);
1057 ASSERT(! ip);
1058 goto dqret;
1059 }
1060
1061 /*
1062 * Dquot lock comes after hashlock in the lock ordering
1063 */
1064 if (ip) {
1065 xfs_ilock(ip, XFS_ILOCK_EXCL);
1066 if (! XFS_IS_DQTYPE_ON(mp, type)) {
1067 /* inode stays locked on return */
1068 xfs_qm_dqdestroy(dqp);
1069 return XFS_ERROR(ESRCH);
1070 }
1071 /*
1072 * A dquot could be attached to this inode by now, since
1073 * we had dropped the ilock.
1074 */
1075 if (type == XFS_DQ_USER) {
1076 if (ip->i_udquot) {
1077 xfs_qm_dqdestroy(dqp);
1078 dqp = ip->i_udquot;
1079 xfs_dqlock(dqp);
1080 goto dqret;
1081 }
1082 } else {
1083 if (ip->i_gdquot) {
1084 xfs_qm_dqdestroy(dqp);
1085 dqp = ip->i_gdquot;
1086 xfs_dqlock(dqp);
1087 goto dqret;
1088 }
1089 }
1090 }
1091
1092 /*
1093 * Hashlock comes after ilock in lock order
1094 */
1095 XFS_DQ_HASH_LOCK(h);
1096 if (version != h->qh_version) {
1097 xfs_dquot_t *tmpdqp;
1098 /*
1099 * Now, see if somebody else put the dquot in the
1100 * hashtable before us. This can happen because we didn't
1101 * keep the hashchain lock. We don't have to worry about
1102 * lock order between the two dquots here since dqp isn't
1103 * on any findable lists yet.
1104 */
1105 if (xfs_qm_dqlookup(mp, id, h, &tmpdqp) == 0) {
1106 /*
1107 * Duplicate found. Just throw away the new dquot
1108 * and start over.
1109 */
1110 xfs_qm_dqput(tmpdqp);
1111 XFS_DQ_HASH_UNLOCK(h);
1112 xfs_qm_dqdestroy(dqp);
1113 XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
1114 goto again;
1115 }
1116 }
1117
1118 /*
1119 * Put the dquot at the beginning of the hash-chain and mp's list
1120 * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock ..
1121 */
1122 ASSERT(XFS_DQ_IS_HASH_LOCKED(h));
1123 dqp->q_hash = h;
1124 XQM_HASHLIST_INSERT(h, dqp);
1125
1126 /*
1127 * Attach this dquot to this filesystem's list of all dquots,
1128 * kept inside the mount structure in m_quotainfo field
1129 */
1130 xfs_qm_mplist_lock(mp);
1131
1132 /*
1133 * We return a locked dquot to the caller, with a reference taken
1134 */
1135 xfs_dqlock(dqp);
1136 dqp->q_nrefs = 1;
1137
1138 XQM_MPLIST_INSERT(&(XFS_QI_MPL_LIST(mp)), dqp);
1139
1140 xfs_qm_mplist_unlock(mp);
1141 XFS_DQ_HASH_UNLOCK(h);
1142 dqret:
1143 ASSERT((ip == NULL) || XFS_ISLOCKED_INODE_EXCL(ip));
1144 xfs_dqtrace_entry(dqp, "DQGET DONE");
1145 *O_dqpp = dqp;
1146 return (0);
1147}
1148
1149
1150/*
1151 * Release a reference to the dquot (decrement ref-count)
1152 * and unlock it. If there is a group quota attached to this
1153 * dquot, carefully release that too without tripping over
1154 * deadlocks'n'stuff.
1155 */
1156void
1157xfs_qm_dqput(
1158 xfs_dquot_t *dqp)
1159{
1160 xfs_dquot_t *gdqp;
1161
1162 ASSERT(dqp->q_nrefs > 0);
1163 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1164 xfs_dqtrace_entry(dqp, "DQPUT");
1165
1166 if (dqp->q_nrefs != 1) {
1167 dqp->q_nrefs--;
1168 xfs_dqunlock(dqp);
1169 return;
1170 }
1171
1172 /*
1173 * drop the dqlock and acquire the freelist and dqlock
1174 * in the right order; but try to get it out-of-order first
1175 */
1176 if (! xfs_qm_freelist_lock_nowait(xfs_Gqm)) {
1177 xfs_dqtrace_entry(dqp, "DQPUT: FLLOCK-WAIT");
1178 xfs_dqunlock(dqp);
1179 xfs_qm_freelist_lock(xfs_Gqm);
1180 xfs_dqlock(dqp);
1181 }
1182
1183 while (1) {
1184 gdqp = NULL;
1185
1186 /* We can't depend on nrefs being == 1 here */
1187 if (--dqp->q_nrefs == 0) {
1188 xfs_dqtrace_entry(dqp, "DQPUT: ON FREELIST");
1189 /*
1190 * insert at end of the freelist.
1191 */
1192 XQM_FREELIST_INSERT(&(xfs_Gqm->qm_dqfreelist), dqp);
1193
1194 /*
1195 * If we just added a udquot to the freelist, then
1196 * we want to release the gdquot reference that
1197 * it (probably) has. Otherwise it'll keep the
1198 * gdquot from getting reclaimed.
1199 */
1200 if ((gdqp = dqp->q_gdquot)) {
1201 /*
1202 * Avoid a recursive dqput call
1203 */
1204 xfs_dqlock(gdqp);
1205 dqp->q_gdquot = NULL;
1206 }
1207
1208 /* xfs_qm_freelist_print(&(xfs_Gqm->qm_dqfreelist),
1209 "@@@@@++ Free list (after append) @@@@@+");
1210 */
1211 }
1212 xfs_dqunlock(dqp);
1213
1214 /*
1215 * If we had a group quota inside the user quota as a hint,
1216 * release it now.
1217 */
1218 if (! gdqp)
1219 break;
1220 dqp = gdqp;
1221 }
1222 xfs_qm_freelist_unlock(xfs_Gqm);
1223}
1224
1225/*
1226 * Release a dquot. Flush it if dirty, then dqput() it.
1227 * dquot must not be locked.
1228 */
1229void
1230xfs_qm_dqrele(
1231 xfs_dquot_t *dqp)
1232{
1233 ASSERT(dqp);
1234 xfs_dqtrace_entry(dqp, "DQRELE");
1235
1236 xfs_dqlock(dqp);
1237 /*
1238 * We don't care to flush it if the dquot is dirty here.
1239 * That will create stutters that we want to avoid.
1240 * Instead we do a delayed write when we try to reclaim
1241 * a dirty dquot. Also xfs_sync will take part of the burden...
1242 */
1243 xfs_qm_dqput(dqp);
1244}
1245
1246
1247/*
1248 * Write a modified dquot to disk.
1249 * The dquot must be locked and the flush lock too taken by caller.
1250 * The flush lock will not be unlocked until the dquot reaches the disk,
1251 * but the dquot is free to be unlocked and modified by the caller
1252 * in the interim. Dquot is still locked on return. This behavior is
1253 * identical to that of inodes.
1254 */
1255int
1256xfs_qm_dqflush(
1257 xfs_dquot_t *dqp,
1258 uint flags)
1259{
1260 xfs_mount_t *mp;
1261 xfs_buf_t *bp;
1262 xfs_disk_dquot_t *ddqp;
1263 int error;
1264 SPLDECL(s);
1265
1266 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1267 ASSERT(XFS_DQ_IS_FLUSH_LOCKED(dqp));
1268 xfs_dqtrace_entry(dqp, "DQFLUSH");
1269
1270 /*
1271 * If not dirty, nada.
1272 */
1273 if (!XFS_DQ_IS_DIRTY(dqp)) {
1274 xfs_dqfunlock(dqp);
1275 return (0);
1276 }
1277
1278 /*
1279 * Cant flush a pinned dquot. Wait for it.
1280 */
1281 xfs_qm_dqunpin_wait(dqp);
1282
1283 /*
1284 * This may have been unpinned because the filesystem is shutting
1285 * down forcibly. If that's the case we must not write this dquot
1286 * to disk, because the log record didn't make it to disk!
1287 */
1288 if (XFS_FORCED_SHUTDOWN(dqp->q_mount)) {
1289 dqp->dq_flags &= ~(XFS_DQ_DIRTY);
1290 xfs_dqfunlock(dqp);
1291 return XFS_ERROR(EIO);
1292 }
1293
1294 /*
1295 * Get the buffer containing the on-disk dquot
1296 * We don't need a transaction envelope because we know that the
1297 * the ondisk-dquot has already been allocated for.
1298 */
1299 if ((error = xfs_qm_dqtobp(NULL, dqp, &ddqp, &bp, XFS_QMOPT_DOWARN))) {
1300 xfs_dqtrace_entry(dqp, "DQTOBP FAIL");
1301 ASSERT(error != ENOENT);
1302 /*
1303 * Quotas could have gotten turned off (ESRCH)
1304 */
1305 xfs_dqfunlock(dqp);
1306 return (error);
1307 }
1308
1309 if (xfs_qm_dqcheck(&dqp->q_core, INT_GET(ddqp->d_id, ARCH_CONVERT), 0, XFS_QMOPT_DOWARN,
1310 "dqflush (incore copy)")) {
1311 xfs_force_shutdown(dqp->q_mount, XFS_CORRUPT_INCORE);
1312 return XFS_ERROR(EIO);
1313 }
1314
1315 /* This is the only portion of data that needs to persist */
1316 memcpy(ddqp, &(dqp->q_core), sizeof(xfs_disk_dquot_t));
1317
1318 /*
1319 * Clear the dirty field and remember the flush lsn for later use.
1320 */
1321 dqp->dq_flags &= ~(XFS_DQ_DIRTY);
1322 mp = dqp->q_mount;
1323
1324 /* lsn is 64 bits */
1325 AIL_LOCK(mp, s);
1326 dqp->q_logitem.qli_flush_lsn = dqp->q_logitem.qli_item.li_lsn;
1327 AIL_UNLOCK(mp, s);
1328
1329 /*
1330 * Attach an iodone routine so that we can remove this dquot from the
1331 * AIL and release the flush lock once the dquot is synced to disk.
1332 */
1333 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t *, xfs_log_item_t *))
1334 xfs_qm_dqflush_done, &(dqp->q_logitem.qli_item));
1335 /*
1336 * If the buffer is pinned then push on the log so we won't
1337 * get stuck waiting in the write for too long.
1338 */
1339 if (XFS_BUF_ISPINNED(bp)) {
1340 xfs_dqtrace_entry(dqp, "DQFLUSH LOG FORCE");
1341 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1342 }
1343
1344 if (flags & XFS_QMOPT_DELWRI) {
1345 xfs_bdwrite(mp, bp);
1346 } else if (flags & XFS_QMOPT_ASYNC) {
1347 xfs_bawrite(mp, bp);
1348 } else {
1349 error = xfs_bwrite(mp, bp);
1350 }
1351 xfs_dqtrace_entry(dqp, "DQFLUSH END");
1352 /*
1353 * dqp is still locked, but caller is free to unlock it now.
1354 */
1355 return (error);
1356
1357}
1358
1359/*
1360 * This is the dquot flushing I/O completion routine. It is called
1361 * from interrupt level when the buffer containing the dquot is
1362 * flushed to disk. It is responsible for removing the dquot logitem
1363 * from the AIL if it has not been re-logged, and unlocking the dquot's
1364 * flush lock. This behavior is very similar to that of inodes..
1365 */
1366/*ARGSUSED*/
1367STATIC void
1368xfs_qm_dqflush_done(
1369 xfs_buf_t *bp,
1370 xfs_dq_logitem_t *qip)
1371{
1372 xfs_dquot_t *dqp;
1373 SPLDECL(s);
1374
1375 dqp = qip->qli_dquot;
1376
1377 /*
1378 * We only want to pull the item from the AIL if its
1379 * location in the log has not changed since we started the flush.
1380 * Thus, we only bother if the dquot's lsn has
1381 * not changed. First we check the lsn outside the lock
1382 * since it's cheaper, and then we recheck while
1383 * holding the lock before removing the dquot from the AIL.
1384 */
1385 if ((qip->qli_item.li_flags & XFS_LI_IN_AIL) &&
1386 qip->qli_item.li_lsn == qip->qli_flush_lsn) {
1387
1388 AIL_LOCK(dqp->q_mount, s);
1389 /*
1390 * xfs_trans_delete_ail() drops the AIL lock.
1391 */
1392 if (qip->qli_item.li_lsn == qip->qli_flush_lsn)
1393 xfs_trans_delete_ail(dqp->q_mount,
1394 (xfs_log_item_t*)qip, s);
1395 else
1396 AIL_UNLOCK(dqp->q_mount, s);
1397 }
1398
1399 /*
1400 * Release the dq's flush lock since we're done with it.
1401 */
1402 xfs_dqfunlock(dqp);
1403}
1404
1405
1406int
1407xfs_qm_dqflock_nowait(
1408 xfs_dquot_t *dqp)
1409{
1410 int locked;
1411
1412 locked = cpsema(&((dqp)->q_flock));
1413
1414 /* XXX ifdef these out */
1415 if (locked)
1416 (dqp)->dq_flags |= XFS_DQ_FLOCKED;
1417 return (locked);
1418}
1419
1420
1421int
1422xfs_qm_dqlock_nowait(
1423 xfs_dquot_t *dqp)
1424{
1425 return (mutex_trylock(&((dqp)->q_qlock)));
1426}
1427
1428void
1429xfs_dqlock(
1430 xfs_dquot_t *dqp)
1431{
1432 mutex_lock(&(dqp->q_qlock), PINOD);
1433}
1434
1435void
1436xfs_dqunlock(
1437 xfs_dquot_t *dqp)
1438{
1439 mutex_unlock(&(dqp->q_qlock));
1440 if (dqp->q_logitem.qli_dquot == dqp) {
1441 /* Once was dqp->q_mount, but might just have been cleared */
1442 xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_mountp,
1443 (xfs_log_item_t*)&(dqp->q_logitem));
1444 }
1445}
1446
1447
1448void
1449xfs_dqunlock_nonotify(
1450 xfs_dquot_t *dqp)
1451{
1452 mutex_unlock(&(dqp->q_qlock));
1453}
1454
1455void
1456xfs_dqlock2(
1457 xfs_dquot_t *d1,
1458 xfs_dquot_t *d2)
1459{
1460 if (d1 && d2) {
1461 ASSERT(d1 != d2);
1462 if (INT_GET(d1->q_core.d_id, ARCH_CONVERT) > INT_GET(d2->q_core.d_id, ARCH_CONVERT)) {
1463 xfs_dqlock(d2);
1464 xfs_dqlock(d1);
1465 } else {
1466 xfs_dqlock(d1);
1467 xfs_dqlock(d2);
1468 }
1469 } else {
1470 if (d1) {
1471 xfs_dqlock(d1);
1472 } else if (d2) {
1473 xfs_dqlock(d2);
1474 }
1475 }
1476}
1477
1478
1479/*
1480 * Take a dquot out of the mount's dqlist as well as the hashlist.
1481 * This is called via unmount as well as quotaoff, and the purge
1482 * will always succeed unless there are soft (temp) references
1483 * outstanding.
1484 *
1485 * This returns 0 if it was purged, 1 if it wasn't. It's not an error code
1486 * that we're returning! XXXsup - not cool.
1487 */
1488/* ARGSUSED */
1489int
1490xfs_qm_dqpurge(
1491 xfs_dquot_t *dqp,
1492 uint flags)
1493{
1494 xfs_dqhash_t *thishash;
1495 xfs_mount_t *mp;
1496
1497 mp = dqp->q_mount;
1498
1499 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp));
1500 ASSERT(XFS_DQ_IS_HASH_LOCKED(dqp->q_hash));
1501
1502 xfs_dqlock(dqp);
1503 /*
1504 * We really can't afford to purge a dquot that is
1505 * referenced, because these are hard refs.
1506 * It shouldn't happen in general because we went thru _all_ inodes in
1507 * dqrele_all_inodes before calling this and didn't let the mountlock go.
1508 * However it is possible that we have dquots with temporary
1509 * references that are not attached to an inode. e.g. see xfs_setattr().
1510 */
1511 if (dqp->q_nrefs != 0) {
1512 xfs_dqunlock(dqp);
1513 XFS_DQ_HASH_UNLOCK(dqp->q_hash);
1514 return (1);
1515 }
1516
1517 ASSERT(XFS_DQ_IS_ON_FREELIST(dqp));
1518
1519 /*
1520 * If we're turning off quotas, we have to make sure that, for
1521 * example, we don't delete quota disk blocks while dquots are
1522 * in the process of getting written to those disk blocks.
1523 * This dquot might well be on AIL, and we can't leave it there
1524 * if we're turning off quotas. Basically, we need this flush
1525 * lock, and are willing to block on it.
1526 */
1527 if (! xfs_qm_dqflock_nowait(dqp)) {
1528 /*
1529 * Block on the flush lock after nudging dquot buffer,
1530 * if it is incore.
1531 */
1532 xfs_qm_dqflock_pushbuf_wait(dqp);
1533 }
1534
1535 /*
1536 * XXXIf we're turning this type of quotas off, we don't care
1537 * about the dirty metadata sitting in this dquot. OTOH, if
1538 * we're unmounting, we do care, so we flush it and wait.
1539 */
1540 if (XFS_DQ_IS_DIRTY(dqp)) {
1541 xfs_dqtrace_entry(dqp, "DQPURGE ->DQFLUSH: DQDIRTY");
1542 /* dqflush unlocks dqflock */
1543 /*
1544 * Given that dqpurge is a very rare occurrence, it is OK
1545 * that we're holding the hashlist and mplist locks
1546 * across the disk write. But, ... XXXsup
1547 *
1548 * We don't care about getting disk errors here. We need
1549 * to purge this dquot anyway, so we go ahead regardless.
1550 */
1551 (void) xfs_qm_dqflush(dqp, XFS_QMOPT_SYNC);
1552 xfs_dqflock(dqp);
1553 }
1554 ASSERT(dqp->q_pincount == 0);
1555 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1556 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
1557
1558 thishash = dqp->q_hash;
1559 XQM_HASHLIST_REMOVE(thishash, dqp);
1560 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(mp)), dqp);
1561 /*
1562 * XXX Move this to the front of the freelist, if we can get the
1563 * freelist lock.
1564 */
1565 ASSERT(XFS_DQ_IS_ON_FREELIST(dqp));
1566
1567 dqp->q_mount = NULL;
1568 dqp->q_hash = NULL;
1569 dqp->dq_flags = XFS_DQ_INACTIVE;
1570 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
1571 xfs_dqfunlock(dqp);
1572 xfs_dqunlock(dqp);
1573 XFS_DQ_HASH_UNLOCK(thishash);
1574 return (0);
1575}
1576
1577
1578#ifdef QUOTADEBUG
1579void
1580xfs_qm_dqprint(xfs_dquot_t *dqp)
1581{
1582 cmn_err(CE_DEBUG, "-----------KERNEL DQUOT----------------");
1583 cmn_err(CE_DEBUG, "---- dquotID = %d",
1584 (int)INT_GET(dqp->q_core.d_id, ARCH_CONVERT));
1585 cmn_err(CE_DEBUG, "---- type = %s",
1586 XFS_QM_ISUDQ(dqp) ? "USR" : "GRP");
1587 cmn_err(CE_DEBUG, "---- fs = 0x%p", dqp->q_mount);
1588 cmn_err(CE_DEBUG, "---- blkno = 0x%x", (int) dqp->q_blkno);
1589 cmn_err(CE_DEBUG, "---- boffset = 0x%x", (int) dqp->q_bufoffset);
1590 cmn_err(CE_DEBUG, "---- blkhlimit = %Lu (0x%x)",
1591 INT_GET(dqp->q_core.d_blk_hardlimit, ARCH_CONVERT),
1592 (int) INT_GET(dqp->q_core.d_blk_hardlimit, ARCH_CONVERT));
1593 cmn_err(CE_DEBUG, "---- blkslimit = %Lu (0x%x)",
1594 INT_GET(dqp->q_core.d_blk_softlimit, ARCH_CONVERT),
1595 (int)INT_GET(dqp->q_core.d_blk_softlimit, ARCH_CONVERT));
1596 cmn_err(CE_DEBUG, "---- inohlimit = %Lu (0x%x)",
1597 INT_GET(dqp->q_core.d_ino_hardlimit, ARCH_CONVERT),
1598 (int)INT_GET(dqp->q_core.d_ino_hardlimit, ARCH_CONVERT));
1599 cmn_err(CE_DEBUG, "---- inoslimit = %Lu (0x%x)",
1600 INT_GET(dqp->q_core.d_ino_softlimit, ARCH_CONVERT),
1601 (int)INT_GET(dqp->q_core.d_ino_softlimit, ARCH_CONVERT));
1602 cmn_err(CE_DEBUG, "---- bcount = %Lu (0x%x)",
1603 INT_GET(dqp->q_core.d_bcount, ARCH_CONVERT),
1604 (int)INT_GET(dqp->q_core.d_bcount, ARCH_CONVERT));
1605 cmn_err(CE_DEBUG, "---- icount = %Lu (0x%x)",
1606 INT_GET(dqp->q_core.d_icount, ARCH_CONVERT),
1607 (int)INT_GET(dqp->q_core.d_icount, ARCH_CONVERT));
1608 cmn_err(CE_DEBUG, "---- btimer = %d",
1609 (int)INT_GET(dqp->q_core.d_btimer, ARCH_CONVERT));
1610 cmn_err(CE_DEBUG, "---- itimer = %d",
1611 (int)INT_GET(dqp->q_core.d_itimer, ARCH_CONVERT));
1612 cmn_err(CE_DEBUG, "---------------------------");
1613}
1614#endif
1615
1616/*
1617 * Give the buffer a little push if it is incore and
1618 * wait on the flush lock.
1619 */
1620void
1621xfs_qm_dqflock_pushbuf_wait(
1622 xfs_dquot_t *dqp)
1623{
1624 xfs_buf_t *bp;
1625
1626 /*
1627 * Check to see if the dquot has been flushed delayed
1628 * write. If so, grab its buffer and send it
1629 * out immediately. We'll be able to acquire
1630 * the flush lock when the I/O completes.
1631 */
1632 bp = xfs_incore(dqp->q_mount->m_ddev_targp, dqp->q_blkno,
1633 XFS_QI_DQCHUNKLEN(dqp->q_mount),
1634 XFS_INCORE_TRYLOCK);
1635 if (bp != NULL) {
1636 if (XFS_BUF_ISDELAYWRITE(bp)) {
1637 if (XFS_BUF_ISPINNED(bp)) {
1638 xfs_log_force(dqp->q_mount,
1639 (xfs_lsn_t)0,
1640 XFS_LOG_FORCE);
1641 }
1642 xfs_bawrite(dqp->q_mount, bp);
1643 } else {
1644 xfs_buf_relse(bp);
1645 }
1646 }
1647 xfs_dqflock(dqp);
1648}