blob: c377961657ee24245703a1f1c74d0a193eb33846 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott4ce31212005-11-02 14:59:41 +11002 * Copyright (c) 2000-2003 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott4ce31212005-11-02 14:59:41 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott4ce31212005-11-02 14:59:41 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott4ce31212005-11-02 14:59:41 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
19#include "xfs_fs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110020#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_trans.h"
24#include "xfs_sb.h"
25#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_quota.h"
28#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_inode.h"
31#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_rtalloc.h"
33#include "xfs_error.h"
34#include "xfs_itable.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_attr.h"
36#include "xfs_buf_item.h"
37#include "xfs_trans_space.h"
38#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_qm.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000040#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#ifdef DEBUG
58xfs_buftarg_t *xfs_dqerror_target;
59int xfs_do_dqerror;
60int xfs_dqreq_num;
61int xfs_dqerror_mod = 33;
62#endif
63
Christoph Hellwig98b8c7a2009-01-19 02:03:25 +010064static struct lock_class_key xfs_dquot_other_class;
65
Linus Torvalds1da177e2005-04-16 15:20:36 -070066/*
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 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100074STATIC xfs_dquot_t *
Linus Torvalds1da177e2005-04-16 15:20:36 -070075xfs_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;
Christoph Hellwig1149d962005-11-02 15:01:12 +110085 dqp->q_core.d_id = cpu_to_be32(id);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 dqp->q_mount = mp;
87
88 /*
89 * No need to re-initialize these if this is a reclaimed dquot.
90 */
91 if (brandnewdquot) {
Dave Chinner3a8406f2010-04-13 15:06:52 +100092 INIT_LIST_HEAD(&dqp->q_freelist);
Jes Sorensen794ee1b2006-01-09 15:59:21 -080093 mutex_init(&dqp->q_qlock);
Peter Leckiebc3048e2008-10-30 17:05:04 +110094 init_waitqueue_head(&dqp->q_pinwait);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
David Chinnere1f49cf2008-08-13 16:41:43 +100096 /*
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
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000104 trace_xfs_dqinit(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 } else {
106 /*
107 * Only the q_core portion was zeroed in dqreclaim_one().
108 * So, we need to reset others.
109 */
Dave Chinner3a8406f2010-04-13 15:06:52 +1000110 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));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000125 trace_xfs_dqreuse(dqp);
Christoph Hellwig98b8c7a2009-01-19 02:03:25 +0100126 }
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135
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{
Dave Chinner3a8406f2010-04-13 15:06:52 +1000149 ASSERT(list_empty(&dqp->q_freelist));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
151 mutex_destroy(&dqp->q_qlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 kmem_zone_free(xfs_Gqm->qm_dqzone, dqp);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 atomic_dec(&xfs_Gqm->qm_totaldquots);
155}
156
157/*
158 * This is what a 'fresh' dquot inside a dquot chunk looks like on disk.
159 */
160STATIC void
161xfs_qm_dqinit_core(
162 xfs_dqid_t id,
163 uint type,
164 xfs_dqblk_t *d)
165{
166 /*
167 * Caller has zero'd the entire dquot 'chunk' already.
168 */
Christoph Hellwig1149d962005-11-02 15:01:12 +1100169 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
170 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
171 d->dd_diskdq.d_id = cpu_to_be32(id);
172 d->dd_diskdq.d_flags = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173}
174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175/*
176 * If default limits are in force, push them into the dquot now.
177 * We overwrite the dquot limits only if they are zero and this
178 * is not the root dquot.
179 */
180void
181xfs_qm_adjust_dqlimits(
182 xfs_mount_t *mp,
183 xfs_disk_dquot_t *d)
184{
185 xfs_quotainfo_t *q = mp->m_quotainfo;
186
187 ASSERT(d->d_id);
188
189 if (q->qi_bsoftlimit && !d->d_blk_softlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100190 d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 if (q->qi_bhardlimit && !d->d_blk_hardlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100192 d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 if (q->qi_isoftlimit && !d->d_ino_softlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100194 d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 if (q->qi_ihardlimit && !d->d_ino_hardlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100196 d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100198 d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100200 d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201}
202
203/*
204 * Check the limits and timers of a dquot and start or reset timers
205 * if necessary.
206 * This gets called even when quota enforcement is OFF, which makes our
207 * life a little less complicated. (We just don't reject any quota
208 * reservations in that case, when enforcement is off).
209 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
210 * enforcement's off.
211 * In contrast, warnings are a little different in that they don't
Nathan Scott754002b2005-06-21 15:49:06 +1000212 * 'automatically' get started when limits get exceeded. They do
213 * get reset to zero, however, when we find the count to be under
214 * the soft limit (they are only ever set non-zero via userspace).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 */
216void
217xfs_qm_adjust_dqtimers(
218 xfs_mount_t *mp,
219 xfs_disk_dquot_t *d)
220{
221 ASSERT(d->d_id);
222
Christoph Hellwigea15ab32011-07-13 13:43:50 +0200223#ifdef DEBUG
Christoph Hellwig1149d962005-11-02 15:01:12 +1100224 if (d->d_blk_hardlimit)
225 ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
226 be64_to_cpu(d->d_blk_hardlimit));
227 if (d->d_ino_hardlimit)
228 ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
229 be64_to_cpu(d->d_ino_hardlimit));
230 if (d->d_rtb_hardlimit)
231 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
232 be64_to_cpu(d->d_rtb_hardlimit));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233#endif
Christoph Hellwigea15ab32011-07-13 13:43:50 +0200234
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 if (!d->d_btimer) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100236 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() +
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000243 mp->m_quotainfo->qi_btimelimit);
Nathan Scott754002b2005-06-21 15:49:06 +1000244 } else {
245 d->d_bwarns = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 }
247 } else {
248 if ((!d->d_blk_softlimit ||
Christoph Hellwig1149d962005-11-02 15:01:12 +1100249 (be64_to_cpu(d->d_bcount) <
250 be64_to_cpu(d->d_blk_softlimit))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 (!d->d_blk_hardlimit ||
Christoph Hellwig1149d962005-11-02 15:01:12 +1100252 (be64_to_cpu(d->d_bcount) <
253 be64_to_cpu(d->d_blk_hardlimit)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 d->d_btimer = 0;
255 }
256 }
257
258 if (!d->d_itimer) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100259 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() +
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000266 mp->m_quotainfo->qi_itimelimit);
Nathan Scott754002b2005-06-21 15:49:06 +1000267 } else {
268 d->d_iwarns = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 }
270 } else {
271 if ((!d->d_ino_softlimit ||
Christoph Hellwig1149d962005-11-02 15:01:12 +1100272 (be64_to_cpu(d->d_icount) <
273 be64_to_cpu(d->d_ino_softlimit))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 (!d->d_ino_hardlimit ||
Christoph Hellwig1149d962005-11-02 15:01:12 +1100275 (be64_to_cpu(d->d_icount) <
276 be64_to_cpu(d->d_ino_hardlimit)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 d->d_itimer = 0;
278 }
279 }
280
281 if (!d->d_rtbtimer) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100282 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() +
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000289 mp->m_quotainfo->qi_rtbtimelimit);
Nathan Scott754002b2005-06-21 15:49:06 +1000290 } else {
291 d->d_rtbwarns = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 }
293 } else {
294 if ((!d->d_rtb_softlimit ||
Christoph Hellwig1149d962005-11-02 15:01:12 +1100295 (be64_to_cpu(d->d_rtbcount) <
296 be64_to_cpu(d->d_rtb_softlimit))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 (!d->d_rtb_hardlimit ||
Christoph Hellwig1149d962005-11-02 15:01:12 +1100298 (be64_to_cpu(d->d_rtbcount) <
299 be64_to_cpu(d->d_rtb_hardlimit)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 d->d_rtbtimer = 0;
301 }
302 }
303}
304
305/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 * 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{
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000316 struct xfs_quotainfo *q = mp->m_quotainfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 xfs_dqblk_t *d;
318 int curid, i;
319
320 ASSERT(tp);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200321 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
Chandra Seetharaman62926042011-07-22 23:40:15 +0000323 d = bp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
325 /*
326 * ID of the first dquot in the block - id's are zero based.
327 */
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000328 curid = id - (id % q->qi_dqperchunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 ASSERT(curid >= 0);
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000330 memset(d, 0, BBTOB(q->qi_dqchunklen));
331 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 xfs_qm_dqinit_core(curid, type, d);
333 xfs_trans_dquot_buf(tp, bp,
Dave Chinnerc1155412010-05-07 11:05:19 +1000334 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
335 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
336 XFS_BLF_GDQUOT_BUF)));
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000337 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338}
339
340
341
342/*
343 * Allocate a block and fill it with dquots.
344 * This is called when the bmapi finds a hole.
345 */
346STATIC int
347xfs_qm_dqalloc(
Tim Shimminefa092f2005-09-05 08:29:01 +1000348 xfs_trans_t **tpp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 xfs_mount_t *mp,
350 xfs_dquot_t *dqp,
351 xfs_inode_t *quotip,
352 xfs_fileoff_t offset_fsb,
353 xfs_buf_t **O_bpp)
354{
355 xfs_fsblock_t firstblock;
356 xfs_bmap_free_t flist;
357 xfs_bmbt_irec_t map;
358 int nmaps, error, committed;
359 xfs_buf_t *bp;
Tim Shimminefa092f2005-09-05 08:29:01 +1000360 xfs_trans_t *tp = *tpp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
362 ASSERT(tp != NULL);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000363
364 trace_xfs_dqalloc(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
366 /*
367 * Initialize the bmap freelist prior to calling bmapi code.
368 */
Eric Sandeen9d87c312009-01-14 23:22:07 -0600369 xfs_bmap_init(&flist, &firstblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 xfs_ilock(quotip, XFS_ILOCK_EXCL);
371 /*
372 * Return if this type of quotas is turned off while we didn't
373 * have an inode lock
374 */
375 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
376 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
377 return (ESRCH);
378 }
379
Christoph Hellwig898621d2010-06-24 11:36:58 +1000380 xfs_trans_ijoin_ref(tp, quotip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 nmaps = 1;
382 if ((error = xfs_bmapi(tp, quotip,
383 offset_fsb, XFS_DQUOT_CLUSTER_SIZE_FSB,
384 XFS_BMAPI_METADATA | XFS_BMAPI_WRITE,
385 &firstblock,
386 XFS_QM_DQALLOC_SPACE_RES(mp),
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000387 &map, &nmaps, &flist))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 goto error0;
389 }
390 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
391 ASSERT(nmaps == 1);
392 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
393 (map.br_startblock != HOLESTARTBLOCK));
394
395 /*
396 * Keep track of the blkno to save a lookup later
397 */
398 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
399
400 /* now we can just get the buffer (there's nothing to read yet) */
401 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
402 dqp->q_blkno,
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000403 mp->m_quotainfo->qi_dqchunklen,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 0);
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +0000405
406 error = xfs_buf_geterror(bp);
407 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 goto error1;
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +0000409
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 /*
411 * Make a chunk of dquots out of this buffer and log
412 * the entire thing.
413 */
Christoph Hellwig1149d962005-11-02 15:01:12 +1100414 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000415 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
Tim Shimminefa092f2005-09-05 08:29:01 +1000417 /*
418 * xfs_bmap_finish() may commit the current transaction and
419 * start a second transaction if the freelist is not empty.
420 *
421 * Since we still want to modify this buffer, we need to
422 * ensure that the buffer is not released on commit of
423 * the first transaction and ensure the buffer is added to the
424 * second transaction.
425 *
426 * If there is only one transaction then don't stop the buffer
427 * from being released when it commits later on.
428 */
429
430 xfs_trans_bhold(tp, bp);
431
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100432 if ((error = xfs_bmap_finish(tpp, &flist, &committed))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 goto error1;
434 }
435
Tim Shimminefa092f2005-09-05 08:29:01 +1000436 if (committed) {
437 tp = *tpp;
438 xfs_trans_bjoin(tp, bp);
439 } else {
440 xfs_trans_bhold_release(tp, bp);
441 }
442
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 *O_bpp = bp;
444 return 0;
445
446 error1:
447 xfs_bmap_cancel(&flist);
448 error0:
449 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
450
451 return (error);
452}
453
454/*
455 * Maps a dquot to the buffer containing its on-disk version.
456 * This returns a ptr to the buffer containing the on-disk dquot
457 * in the bpp param, and a ptr to the on-disk dquot within that buffer
458 */
459STATIC int
460xfs_qm_dqtobp(
Tim Shimminefa092f2005-09-05 08:29:01 +1000461 xfs_trans_t **tpp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 xfs_dquot_t *dqp,
463 xfs_disk_dquot_t **O_ddpp,
464 xfs_buf_t **O_bpp,
465 uint flags)
466{
467 xfs_bmbt_irec_t map;
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000468 int nmaps = 1, error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 xfs_buf_t *bp;
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000470 xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp);
471 xfs_mount_t *mp = dqp->q_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 xfs_disk_dquot_t *ddq;
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000473 xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id);
Tim Shimminefa092f2005-09-05 08:29:01 +1000474 xfs_trans_t *tp = (tpp ? *tpp : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000476 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000478 xfs_ilock(quotip, XFS_ILOCK_SHARED);
479 if (XFS_IS_THIS_QUOTA_OFF(dqp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 /*
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000481 * Return if this type of quotas is turned off while we
482 * didn't have the quota inode lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000485 return ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
488 /*
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000489 * Find the block map; no allocations yet
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 */
Dave Chinner5c8ed202011-09-18 20:40:45 +0000491 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
492 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000493
494 xfs_iunlock(quotip, XFS_ILOCK_SHARED);
495 if (error)
496 return error;
497
498 ASSERT(nmaps == 1);
499 ASSERT(map.br_blockcount == 1);
500
501 /*
502 * Offset of dquot in the (fixed sized) dquot chunk.
503 */
504 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
505 sizeof(xfs_dqblk_t);
506
507 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
508 if (map.br_startblock == HOLESTARTBLOCK) {
509 /*
510 * We don't allocate unless we're asked to
511 */
512 if (!(flags & XFS_QMOPT_DQALLOC))
513 return ENOENT;
514
515 ASSERT(tp);
516 error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
517 dqp->q_fileoffset, &bp);
518 if (error)
519 return error;
520 tp = *tpp;
521 } else {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000522 trace_xfs_dqtobp_read(dqp);
523
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000524 /*
525 * store the blkno etc so that we don't have to do the
526 * mapping all the time
527 */
528 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
529
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000530 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
531 dqp->q_blkno,
532 mp->m_quotainfo->qi_dqchunklen,
533 0, &bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 if (error || !bp)
535 return XFS_ERROR(error);
536 }
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000537
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200538 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
540 /*
541 * calculate the location of the dquot inside the buffer.
542 */
Chandra Seetharaman62926042011-07-22 23:40:15 +0000543 ddq = bp->b_addr + dqp->q_bufoffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
545 /*
546 * A simple sanity check in case we got a corrupted dquot...
547 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100548 error = xfs_qm_dqcheck(mp, ddq, id, dqp->dq_flags & XFS_DQ_ALLTYPES,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN),
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100550 "dqtobp");
551 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 if (!(flags & XFS_QMOPT_DQREPAIR)) {
553 xfs_trans_brelse(tp, bp);
554 return XFS_ERROR(EIO);
555 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 }
557
558 *O_bpp = bp;
559 *O_ddpp = ddq;
560
561 return (0);
562}
563
564
565/*
566 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
567 * and release the buffer immediately.
568 *
569 */
570/* ARGSUSED */
571STATIC int
572xfs_qm_dqread(
Tim Shimminefa092f2005-09-05 08:29:01 +1000573 xfs_trans_t **tpp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 xfs_dqid_t id,
575 xfs_dquot_t *dqp, /* dquot to get filled in */
576 uint flags)
577{
578 xfs_disk_dquot_t *ddqp;
579 xfs_buf_t *bp;
580 int error;
Tim Shimminefa092f2005-09-05 08:29:01 +1000581 xfs_trans_t *tp;
582
583 ASSERT(tpp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000585 trace_xfs_dqread(dqp);
586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 /*
588 * get a pointer to the on-disk dquot and the buffer containing it
589 * dqp already knows its own type (GROUP/USER).
590 */
Tim Shimminefa092f2005-09-05 08:29:01 +1000591 if ((error = xfs_qm_dqtobp(tpp, dqp, &ddqp, &bp, flags))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 return (error);
593 }
Tim Shimminefa092f2005-09-05 08:29:01 +1000594 tp = *tpp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
596 /* copy everything from disk dquot to the incore dquot */
597 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
Christoph Hellwig1149d962005-11-02 15:01:12 +1100598 ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 xfs_qm_dquot_logitem_init(dqp);
600
601 /*
602 * Reservation counters are defined as reservation plus current usage
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300603 * to avoid having to add every time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 */
Christoph Hellwig1149d962005-11-02 15:01:12 +1100605 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
606 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
607 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
609 /* Mark the buf so that this will stay incore a little longer */
610 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DQUOT, XFS_DQUOT_REF);
611
612 /*
613 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
614 * So we need to release with xfs_trans_brelse().
615 * The strategy here is identical to that of inodes; we lock
616 * the dquot in xfs_qm_dqget() before making it accessible to
617 * others. This is because dquots, like inodes, need a good level of
618 * concurrency, and we don't want to take locks on the entire buffers
619 * for dquot accesses.
620 * Note also that the dquot buffer may even be dirty at this point, if
621 * this particular dquot was repaired. We still aren't afraid to
622 * brelse it because we have the changes incore.
623 */
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200624 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 xfs_trans_brelse(tp, bp);
626
627 return (error);
628}
629
630
631/*
632 * allocate an incore dquot from the kernel heap,
633 * and fill its core with quota information kept on disk.
634 * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk
635 * if it wasn't already allocated.
636 */
637STATIC int
638xfs_qm_idtodq(
639 xfs_mount_t *mp,
640 xfs_dqid_t id, /* gid or uid, depending on type */
641 uint type, /* UDQUOT or GDQUOT */
642 uint flags, /* DQALLOC, DQREPAIR */
643 xfs_dquot_t **O_dqpp)/* OUT : incore dquot, not locked */
644{
645 xfs_dquot_t *dqp;
646 int error;
647 xfs_trans_t *tp;
648 int cancelflags=0;
649
650 dqp = xfs_qm_dqinit(mp, id, type);
651 tp = NULL;
652 if (flags & XFS_QMOPT_DQALLOC) {
653 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000654 error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp),
655 XFS_WRITE_LOG_RES(mp) +
656 BBTOB(mp->m_quotainfo->qi_dqchunklen) - 1 +
657 128,
658 0,
659 XFS_TRANS_PERM_LOG_RES,
660 XFS_WRITE_LOG_COUNT);
661 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 cancelflags = 0;
663 goto error0;
664 }
665 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
666 }
667
668 /*
669 * Read it from disk; xfs_dqread() takes care of
670 * all the necessary initialization of dquot's fields (locks, etc)
671 */
Tim Shimminefa092f2005-09-05 08:29:01 +1000672 if ((error = xfs_qm_dqread(&tp, id, dqp, flags))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 /*
674 * This can happen if quotas got turned off (ESRCH),
675 * or if the dquot didn't exist on disk and we ask to
676 * allocate (ENOENT).
677 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000678 trace_xfs_dqread_fail(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 cancelflags |= XFS_TRANS_ABORT;
680 goto error0;
681 }
682 if (tp) {
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000683 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 goto error1;
685 }
686
687 *O_dqpp = dqp;
688 return (0);
689
690 error0:
691 ASSERT(error);
692 if (tp)
693 xfs_trans_cancel(tp, cancelflags);
694 error1:
695 xfs_qm_dqdestroy(dqp);
696 *O_dqpp = NULL;
697 return (error);
698}
699
700/*
701 * Lookup a dquot in the incore dquot hashtable. We keep two separate
702 * hashtables for user and group dquots; and, these are global tables
703 * inside the XQM, not per-filesystem tables.
704 * The hash chain must be locked by caller, and it is left locked
705 * on return. Returning dquot is locked.
706 */
707STATIC int
708xfs_qm_dqlookup(
709 xfs_mount_t *mp,
710 xfs_dqid_t id,
711 xfs_dqhash_t *qh,
712 xfs_dquot_t **O_dqpp)
713{
714 xfs_dquot_t *dqp;
715 uint flist_locked;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
Christoph Hellwigc9a192d2009-02-09 08:47:22 +0100717 ASSERT(mutex_is_locked(&qh->qh_lock));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
719 flist_locked = B_FALSE;
720
721 /*
722 * Traverse the hashchain looking for a match
723 */
Dave Chinnere6a81f12010-04-13 15:06:51 +1000724 list_for_each_entry(dqp, &qh->qh_list, q_hashlist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 /*
726 * We already have the hashlock. We don't need the
727 * dqlock to look at the id field of the dquot, since the
728 * id can't be modified without the hashlock anyway.
729 */
Christoph Hellwig1149d962005-11-02 15:01:12 +1100730 if (be32_to_cpu(dqp->q_core.d_id) == id && dqp->q_mount == mp) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000731 trace_xfs_dqlookup_found(dqp);
732
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 /*
734 * All in core dquots must be on the dqlist of mp
735 */
Dave Chinner3a254042010-04-13 15:06:48 +1000736 ASSERT(!list_empty(&dqp->q_mplist));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737
738 xfs_dqlock(dqp);
739 if (dqp->q_nrefs == 0) {
Dave Chinner3a8406f2010-04-13 15:06:52 +1000740 ASSERT(!list_empty(&dqp->q_freelist));
741 if (!mutex_trylock(&xfs_Gqm->qm_dqfrlist_lock)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000742 trace_xfs_dqlookup_want(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
744 /*
745 * We may have raced with dqreclaim_one()
746 * (and lost). So, flag that we don't
747 * want the dquot to be reclaimed.
748 */
749 dqp->dq_flags |= XFS_DQ_WANT;
750 xfs_dqunlock(dqp);
Dave Chinner3a8406f2010-04-13 15:06:52 +1000751 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 xfs_dqlock(dqp);
753 dqp->dq_flags &= ~(XFS_DQ_WANT);
754 }
755 flist_locked = B_TRUE;
756 }
757
758 /*
759 * id couldn't have changed; we had the hashlock all
760 * along
761 */
Christoph Hellwig1149d962005-11-02 15:01:12 +1100762 ASSERT(be32_to_cpu(dqp->q_core.d_id) == id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
764 if (flist_locked) {
765 if (dqp->q_nrefs != 0) {
Dave Chinner3a8406f2010-04-13 15:06:52 +1000766 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 flist_locked = B_FALSE;
768 } else {
Dave Chinner3a8406f2010-04-13 15:06:52 +1000769 /* take it off the freelist */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000770 trace_xfs_dqlookup_freelist(dqp);
Dave Chinner3a8406f2010-04-13 15:06:52 +1000771 list_del_init(&dqp->q_freelist);
772 xfs_Gqm->qm_dqfrlist_cnt--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 }
774 }
775
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 XFS_DQHOLD(dqp);
777
778 if (flist_locked)
Dave Chinner3a8406f2010-04-13 15:06:52 +1000779 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 /*
781 * move the dquot to the front of the hashchain
782 */
Christoph Hellwigc9a192d2009-02-09 08:47:22 +0100783 ASSERT(mutex_is_locked(&qh->qh_lock));
Dave Chinnere6a81f12010-04-13 15:06:51 +1000784 list_move(&dqp->q_hashlist, &qh->qh_list);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000785 trace_xfs_dqlookup_done(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 *O_dqpp = dqp;
Dave Chinnere6a81f12010-04-13 15:06:51 +1000787 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
789 }
790
791 *O_dqpp = NULL;
Christoph Hellwigc9a192d2009-02-09 08:47:22 +0100792 ASSERT(mutex_is_locked(&qh->qh_lock));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 return (1);
794}
795
796/*
797 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
798 * a locked dquot, doing an allocation (if requested) as needed.
799 * When both an inode and an id are given, the inode's id takes precedence.
800 * That is, if the id changes while we don't hold the ilock inside this
801 * function, the new dquot is returned, not necessarily the one requested
802 * in the id argument.
803 */
804int
805xfs_qm_dqget(
806 xfs_mount_t *mp,
807 xfs_inode_t *ip, /* locked inode (optional) */
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000808 xfs_dqid_t id, /* uid/projid/gid depending on type */
809 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
811 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
812{
813 xfs_dquot_t *dqp;
814 xfs_dqhash_t *h;
815 uint version;
816 int error;
817
818 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
819 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000820 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
822 return (ESRCH);
823 }
824 h = XFS_DQ_HASH(mp, id, type);
825
826#ifdef DEBUG
827 if (xfs_do_dqerror) {
828 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
829 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100830 xfs_debug(mp, "Returning error in dqget");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 return (EIO);
832 }
833 }
834#endif
835
836 again:
837
838#ifdef DEBUG
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000839 ASSERT(type == XFS_DQ_USER ||
840 type == XFS_DQ_PROJ ||
841 type == XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 if (ip) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000843 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (type == XFS_DQ_USER)
845 ASSERT(ip->i_udquot == NULL);
846 else
847 ASSERT(ip->i_gdquot == NULL);
848 }
849#endif
Christoph Hellwigc9a192d2009-02-09 08:47:22 +0100850 mutex_lock(&h->qh_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
852 /*
853 * Look in the cache (hashtable).
854 * The chain is kept locked during lookup.
855 */
856 if (xfs_qm_dqlookup(mp, id, h, O_dqpp) == 0) {
857 XQM_STATS_INC(xqmstats.xs_qm_dqcachehits);
858 /*
859 * The dquot was found, moved to the front of the chain,
860 * taken off the freelist if it was on it, and locked
861 * at this point. Just unlock the hashchain and return.
862 */
863 ASSERT(*O_dqpp);
864 ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp));
Christoph Hellwigc9a192d2009-02-09 08:47:22 +0100865 mutex_unlock(&h->qh_lock);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000866 trace_xfs_dqget_hit(*O_dqpp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 return (0); /* success */
868 }
869 XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses);
870
871 /*
872 * Dquot cache miss. We don't want to keep the inode lock across
873 * a (potential) disk read. Also we don't want to deal with the lock
874 * ordering between quotainode and this inode. OTOH, dropping the inode
875 * lock here means dealing with a chown that can happen before
876 * we re-acquire the lock.
877 */
878 if (ip)
879 xfs_iunlock(ip, XFS_ILOCK_EXCL);
880 /*
881 * Save the hashchain version stamp, and unlock the chain, so that
882 * we don't keep the lock across a disk read
883 */
884 version = h->qh_version;
Christoph Hellwigc9a192d2009-02-09 08:47:22 +0100885 mutex_unlock(&h->qh_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
887 /*
888 * Allocate the dquot on the kernel heap, and read the ondisk
889 * portion off the disk. Also, do all the necessary initialization
890 * This can return ENOENT if dquot didn't exist on disk and we didn't
891 * ask it to allocate; ESRCH if quotas got turned off suddenly.
892 */
893 if ((error = xfs_qm_idtodq(mp, id, type,
894 flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR|
895 XFS_QMOPT_DOWARN),
896 &dqp))) {
897 if (ip)
898 xfs_ilock(ip, XFS_ILOCK_EXCL);
899 return (error);
900 }
901
902 /*
903 * See if this is mount code calling to look at the overall quota limits
904 * which are stored in the id == 0 user or group's dquot.
905 * Since we may not have done a quotacheck by this point, just return
906 * the dquot without attaching it to any hashtables, lists, etc, or even
907 * taking a reference.
908 * The caller must dqdestroy this once done.
909 */
910 if (flags & XFS_QMOPT_DQSUSER) {
911 ASSERT(id == 0);
912 ASSERT(! ip);
913 goto dqret;
914 }
915
916 /*
917 * Dquot lock comes after hashlock in the lock ordering
918 */
919 if (ip) {
920 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig191f8482010-04-20 17:01:53 +1000921
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 /*
923 * A dquot could be attached to this inode by now, since
924 * we had dropped the ilock.
925 */
926 if (type == XFS_DQ_USER) {
Christoph Hellwig191f8482010-04-20 17:01:53 +1000927 if (!XFS_IS_UQUOTA_ON(mp)) {
928 /* inode stays locked on return */
929 xfs_qm_dqdestroy(dqp);
930 return XFS_ERROR(ESRCH);
931 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 if (ip->i_udquot) {
933 xfs_qm_dqdestroy(dqp);
934 dqp = ip->i_udquot;
935 xfs_dqlock(dqp);
936 goto dqret;
937 }
938 } else {
Christoph Hellwig191f8482010-04-20 17:01:53 +1000939 if (!XFS_IS_OQUOTA_ON(mp)) {
940 /* inode stays locked on return */
941 xfs_qm_dqdestroy(dqp);
942 return XFS_ERROR(ESRCH);
943 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 if (ip->i_gdquot) {
945 xfs_qm_dqdestroy(dqp);
946 dqp = ip->i_gdquot;
947 xfs_dqlock(dqp);
948 goto dqret;
949 }
950 }
951 }
952
953 /*
954 * Hashlock comes after ilock in lock order
955 */
Christoph Hellwigc9a192d2009-02-09 08:47:22 +0100956 mutex_lock(&h->qh_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 if (version != h->qh_version) {
958 xfs_dquot_t *tmpdqp;
959 /*
960 * Now, see if somebody else put the dquot in the
961 * hashtable before us. This can happen because we didn't
962 * keep the hashchain lock. We don't have to worry about
963 * lock order between the two dquots here since dqp isn't
964 * on any findable lists yet.
965 */
966 if (xfs_qm_dqlookup(mp, id, h, &tmpdqp) == 0) {
967 /*
968 * Duplicate found. Just throw away the new dquot
969 * and start over.
970 */
971 xfs_qm_dqput(tmpdqp);
Christoph Hellwigc9a192d2009-02-09 08:47:22 +0100972 mutex_unlock(&h->qh_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 xfs_qm_dqdestroy(dqp);
974 XQM_STATS_INC(xqmstats.xs_qm_dquot_dups);
975 goto again;
976 }
977 }
978
979 /*
980 * Put the dquot at the beginning of the hash-chain and mp's list
981 * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock ..
982 */
Christoph Hellwigc9a192d2009-02-09 08:47:22 +0100983 ASSERT(mutex_is_locked(&h->qh_lock));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 dqp->q_hash = h;
Dave Chinnere6a81f12010-04-13 15:06:51 +1000985 list_add(&dqp->q_hashlist, &h->qh_list);
986 h->qh_version++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
988 /*
989 * Attach this dquot to this filesystem's list of all dquots,
990 * kept inside the mount structure in m_quotainfo field
991 */
Dave Chinner3a254042010-04-13 15:06:48 +1000992 mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
994 /*
995 * We return a locked dquot to the caller, with a reference taken
996 */
997 xfs_dqlock(dqp);
998 dqp->q_nrefs = 1;
999
Dave Chinner3a254042010-04-13 15:06:48 +10001000 list_add(&dqp->q_mplist, &mp->m_quotainfo->qi_dqlist);
1001 mp->m_quotainfo->qi_dquots++;
1002 mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
Christoph Hellwigc9a192d2009-02-09 08:47:22 +01001003 mutex_unlock(&h->qh_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 dqret:
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001005 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001006 trace_xfs_dqget_miss(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 *O_dqpp = dqp;
1008 return (0);
1009}
1010
1011
1012/*
1013 * Release a reference to the dquot (decrement ref-count)
1014 * and unlock it. If there is a group quota attached to this
1015 * dquot, carefully release that too without tripping over
1016 * deadlocks'n'stuff.
1017 */
1018void
1019xfs_qm_dqput(
1020 xfs_dquot_t *dqp)
1021{
1022 xfs_dquot_t *gdqp;
1023
1024 ASSERT(dqp->q_nrefs > 0);
1025 ASSERT(XFS_DQ_IS_LOCKED(dqp));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001026
1027 trace_xfs_dqput(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
1029 if (dqp->q_nrefs != 1) {
1030 dqp->q_nrefs--;
1031 xfs_dqunlock(dqp);
1032 return;
1033 }
1034
1035 /*
1036 * drop the dqlock and acquire the freelist and dqlock
1037 * in the right order; but try to get it out-of-order first
1038 */
Dave Chinner3a8406f2010-04-13 15:06:52 +10001039 if (!mutex_trylock(&xfs_Gqm->qm_dqfrlist_lock)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001040 trace_xfs_dqput_wait(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 xfs_dqunlock(dqp);
Dave Chinner3a8406f2010-04-13 15:06:52 +10001042 mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 xfs_dqlock(dqp);
1044 }
1045
1046 while (1) {
1047 gdqp = NULL;
1048
1049 /* We can't depend on nrefs being == 1 here */
1050 if (--dqp->q_nrefs == 0) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001051 trace_xfs_dqput_free(dqp);
1052
Dave Chinner3a8406f2010-04-13 15:06:52 +10001053 list_add_tail(&dqp->q_freelist, &xfs_Gqm->qm_dqfrlist);
1054 xfs_Gqm->qm_dqfrlist_cnt++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
1056 /*
1057 * If we just added a udquot to the freelist, then
1058 * we want to release the gdquot reference that
1059 * it (probably) has. Otherwise it'll keep the
1060 * gdquot from getting reclaimed.
1061 */
1062 if ((gdqp = dqp->q_gdquot)) {
1063 /*
1064 * Avoid a recursive dqput call
1065 */
1066 xfs_dqlock(gdqp);
1067 dqp->q_gdquot = NULL;
1068 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 }
1070 xfs_dqunlock(dqp);
1071
1072 /*
1073 * If we had a group quota inside the user quota as a hint,
1074 * release it now.
1075 */
1076 if (! gdqp)
1077 break;
1078 dqp = gdqp;
1079 }
Dave Chinner3a8406f2010-04-13 15:06:52 +10001080 mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081}
1082
1083/*
1084 * Release a dquot. Flush it if dirty, then dqput() it.
1085 * dquot must not be locked.
1086 */
1087void
1088xfs_qm_dqrele(
1089 xfs_dquot_t *dqp)
1090{
Christoph Hellwig7d095252009-06-08 15:33:32 +02001091 if (!dqp)
1092 return;
1093
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001094 trace_xfs_dqrele(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
1096 xfs_dqlock(dqp);
1097 /*
1098 * We don't care to flush it if the dquot is dirty here.
1099 * That will create stutters that we want to avoid.
1100 * Instead we do a delayed write when we try to reclaim
1101 * a dirty dquot. Also xfs_sync will take part of the burden...
1102 */
1103 xfs_qm_dqput(dqp);
1104}
1105
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001106/*
1107 * This is the dquot flushing I/O completion routine. It is called
1108 * from interrupt level when the buffer containing the dquot is
1109 * flushed to disk. It is responsible for removing the dquot logitem
1110 * from the AIL if it has not been re-logged, and unlocking the dquot's
1111 * flush lock. This behavior is very similar to that of inodes..
1112 */
1113STATIC void
1114xfs_qm_dqflush_done(
1115 struct xfs_buf *bp,
1116 struct xfs_log_item *lip)
1117{
1118 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
1119 xfs_dquot_t *dqp = qip->qli_dquot;
1120 struct xfs_ail *ailp = lip->li_ailp;
1121
1122 /*
1123 * We only want to pull the item from the AIL if its
1124 * location in the log has not changed since we started the flush.
1125 * Thus, we only bother if the dquot's lsn has
1126 * not changed. First we check the lsn outside the lock
1127 * since it's cheaper, and then we recheck while
1128 * holding the lock before removing the dquot from the AIL.
1129 */
1130 if ((lip->li_flags & XFS_LI_IN_AIL) &&
1131 lip->li_lsn == qip->qli_flush_lsn) {
1132
1133 /* xfs_trans_ail_delete() drops the AIL lock. */
1134 spin_lock(&ailp->xa_lock);
1135 if (lip->li_lsn == qip->qli_flush_lsn)
1136 xfs_trans_ail_delete(ailp, lip);
1137 else
1138 spin_unlock(&ailp->xa_lock);
1139 }
1140
1141 /*
1142 * Release the dq's flush lock since we're done with it.
1143 */
1144 xfs_dqfunlock(dqp);
1145}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
1147/*
1148 * Write a modified dquot to disk.
1149 * The dquot must be locked and the flush lock too taken by caller.
1150 * The flush lock will not be unlocked until the dquot reaches the disk,
1151 * but the dquot is free to be unlocked and modified by the caller
1152 * in the interim. Dquot is still locked on return. This behavior is
1153 * identical to that of inodes.
1154 */
1155int
1156xfs_qm_dqflush(
1157 xfs_dquot_t *dqp,
1158 uint flags)
1159{
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001160 struct xfs_mount *mp = dqp->q_mount;
1161 struct xfs_buf *bp;
1162 struct xfs_disk_dquot *ddqp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
1165 ASSERT(XFS_DQ_IS_LOCKED(dqp));
David Chinnere1f49cf2008-08-13 16:41:43 +10001166 ASSERT(!completion_done(&dqp->q_flush));
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001167
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001168 trace_xfs_dqflush(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
1170 /*
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001171 * If not dirty, or it's pinned and we are not supposed to block, nada.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 */
David Chinner2f8a3ce2008-10-30 17:07:20 +11001173 if (!XFS_DQ_IS_DIRTY(dqp) ||
Dave Chinner20026d92010-02-04 09:48:58 +11001174 (!(flags & SYNC_WAIT) && atomic_read(&dqp->q_pincount) > 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 xfs_dqfunlock(dqp);
David Chinner2f8a3ce2008-10-30 17:07:20 +11001176 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 xfs_qm_dqunpin_wait(dqp);
1179
1180 /*
1181 * This may have been unpinned because the filesystem is shutting
1182 * down forcibly. If that's the case we must not write this dquot
1183 * to disk, because the log record didn't make it to disk!
1184 */
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001185 if (XFS_FORCED_SHUTDOWN(mp)) {
1186 dqp->dq_flags &= ~XFS_DQ_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 xfs_dqfunlock(dqp);
1188 return XFS_ERROR(EIO);
1189 }
1190
1191 /*
1192 * Get the buffer containing the on-disk dquot
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 */
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001194 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1195 mp->m_quotainfo->qi_dqchunklen, 0, &bp);
1196 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 ASSERT(error != ENOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 xfs_dqfunlock(dqp);
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001199 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 }
1201
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001202 /*
1203 * Calculate the location of the dquot inside the buffer.
1204 */
Chandra Seetharaman62926042011-07-22 23:40:15 +00001205 ddqp = bp->b_addr + dqp->q_bufoffset;
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001206
1207 /*
1208 * A simple sanity check in case we got a corrupted dquot..
1209 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001210 error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
1211 XFS_QMOPT_DOWARN, "dqflush (incore copy)");
1212 if (error) {
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001213 xfs_buf_relse(bp);
1214 xfs_dqfunlock(dqp);
1215 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 return XFS_ERROR(EIO);
1217 }
1218
1219 /* This is the only portion of data that needs to persist */
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001220 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
1222 /*
1223 * Clear the dirty field and remember the flush lsn for later use.
1224 */
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001225 dqp->dq_flags &= ~XFS_DQ_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
David Chinner7b2e2a32008-10-30 17:39:12 +11001227 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1228 &dqp->q_logitem.qli_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
1230 /*
1231 * Attach an iodone routine so that we can remove this dquot from the
1232 * AIL and release the flush lock once the dquot is synced to disk.
1233 */
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001234 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1235 &dqp->q_logitem.qli_item);
1236
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 /*
1238 * If the buffer is pinned then push on the log so we won't
1239 * get stuck waiting in the write for too long.
1240 */
Chandra Seetharaman811e64c2011-07-22 23:40:27 +00001241 if (xfs_buf_ispinned(bp)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001242 trace_xfs_dqflush_force(dqp);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001243 xfs_log_force(mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 }
1245
Dave Chinner20026d92010-02-04 09:48:58 +11001246 if (flags & SYNC_WAIT)
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00001247 error = xfs_bwrite(bp);
1248 else
Christoph Hellwig61551f12011-08-23 08:28:06 +00001249 xfs_buf_delwri_queue(bp);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00001250
1251 xfs_buf_relse(bp);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001252
1253 trace_xfs_dqflush_done(dqp);
1254
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 /*
1256 * dqp is still locked, but caller is free to unlock it now.
1257 */
Dave Chinner20026d92010-02-04 09:48:58 +11001258 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
1260}
1261
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262int
1263xfs_qm_dqlock_nowait(
1264 xfs_dquot_t *dqp)
1265{
David Chinnere1f49cf2008-08-13 16:41:43 +10001266 return mutex_trylock(&dqp->q_qlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267}
1268
1269void
1270xfs_dqlock(
1271 xfs_dquot_t *dqp)
1272{
David Chinnere1f49cf2008-08-13 16:41:43 +10001273 mutex_lock(&dqp->q_qlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274}
1275
1276void
1277xfs_dqunlock(
1278 xfs_dquot_t *dqp)
1279{
1280 mutex_unlock(&(dqp->q_qlock));
1281 if (dqp->q_logitem.qli_dquot == dqp) {
1282 /* Once was dqp->q_mount, but might just have been cleared */
David Chinner783a2f62008-10-30 17:39:58 +11001283 xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_ailp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 (xfs_log_item_t*)&(dqp->q_logitem));
1285 }
1286}
1287
1288
1289void
1290xfs_dqunlock_nonotify(
1291 xfs_dquot_t *dqp)
1292{
1293 mutex_unlock(&(dqp->q_qlock));
1294}
1295
Christoph Hellwig5bb87a32009-01-19 02:03:19 +01001296/*
1297 * Lock two xfs_dquot structures.
1298 *
1299 * To avoid deadlocks we always lock the quota structure with
1300 * the lowerd id first.
1301 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302void
1303xfs_dqlock2(
1304 xfs_dquot_t *d1,
1305 xfs_dquot_t *d2)
1306{
1307 if (d1 && d2) {
1308 ASSERT(d1 != d2);
Christoph Hellwig1149d962005-11-02 15:01:12 +11001309 if (be32_to_cpu(d1->q_core.d_id) >
1310 be32_to_cpu(d2->q_core.d_id)) {
Christoph Hellwig5bb87a32009-01-19 02:03:19 +01001311 mutex_lock(&d2->q_qlock);
1312 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 } else {
Christoph Hellwig5bb87a32009-01-19 02:03:19 +01001314 mutex_lock(&d1->q_qlock);
1315 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 }
Christoph Hellwig5bb87a32009-01-19 02:03:19 +01001317 } else if (d1) {
1318 mutex_lock(&d1->q_qlock);
1319 } else if (d2) {
1320 mutex_lock(&d2->q_qlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 }
1322}
1323
1324
1325/*
1326 * Take a dquot out of the mount's dqlist as well as the hashlist.
1327 * This is called via unmount as well as quotaoff, and the purge
1328 * will always succeed unless there are soft (temp) references
1329 * outstanding.
1330 *
1331 * This returns 0 if it was purged, 1 if it wasn't. It's not an error code
1332 * that we're returning! XXXsup - not cool.
1333 */
1334/* ARGSUSED */
1335int
1336xfs_qm_dqpurge(
Denys Vlasenko4f0e8a92008-05-19 16:34:04 +10001337 xfs_dquot_t *dqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338{
Dave Chinnere6a81f12010-04-13 15:06:51 +10001339 xfs_dqhash_t *qh = dqp->q_hash;
David Chinner3c568362008-04-10 12:20:24 +10001340 xfs_mount_t *mp = dqp->q_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
Dave Chinner3a254042010-04-13 15:06:48 +10001342 ASSERT(mutex_is_locked(&mp->m_quotainfo->qi_dqlist_lock));
Christoph Hellwigc9a192d2009-02-09 08:47:22 +01001343 ASSERT(mutex_is_locked(&dqp->q_hash->qh_lock));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
1345 xfs_dqlock(dqp);
1346 /*
1347 * We really can't afford to purge a dquot that is
1348 * referenced, because these are hard refs.
1349 * It shouldn't happen in general because we went thru _all_ inodes in
1350 * dqrele_all_inodes before calling this and didn't let the mountlock go.
1351 * However it is possible that we have dquots with temporary
1352 * references that are not attached to an inode. e.g. see xfs_setattr().
1353 */
1354 if (dqp->q_nrefs != 0) {
1355 xfs_dqunlock(dqp);
Christoph Hellwigc9a192d2009-02-09 08:47:22 +01001356 mutex_unlock(&dqp->q_hash->qh_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 return (1);
1358 }
1359
Dave Chinner3a8406f2010-04-13 15:06:52 +10001360 ASSERT(!list_empty(&dqp->q_freelist));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
1362 /*
1363 * If we're turning off quotas, we have to make sure that, for
1364 * example, we don't delete quota disk blocks while dquots are
1365 * in the process of getting written to those disk blocks.
1366 * This dquot might well be on AIL, and we can't leave it there
1367 * if we're turning off quotas. Basically, we need this flush
1368 * lock, and are willing to block on it.
1369 */
David Chinnere1f49cf2008-08-13 16:41:43 +10001370 if (!xfs_dqflock_nowait(dqp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 /*
1372 * Block on the flush lock after nudging dquot buffer,
1373 * if it is incore.
1374 */
1375 xfs_qm_dqflock_pushbuf_wait(dqp);
1376 }
1377
1378 /*
1379 * XXXIf we're turning this type of quotas off, we don't care
1380 * about the dirty metadata sitting in this dquot. OTOH, if
1381 * we're unmounting, we do care, so we flush it and wait.
1382 */
1383 if (XFS_DQ_IS_DIRTY(dqp)) {
David Chinner3c568362008-04-10 12:20:24 +10001384 int error;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 /* dqflush unlocks dqflock */
1387 /*
1388 * Given that dqpurge is a very rare occurrence, it is OK
1389 * that we're holding the hashlist and mplist locks
1390 * across the disk write. But, ... XXXsup
1391 *
1392 * We don't care about getting disk errors here. We need
1393 * to purge this dquot anyway, so we go ahead regardless.
1394 */
Dave Chinner20026d92010-02-04 09:48:58 +11001395 error = xfs_qm_dqflush(dqp, SYNC_WAIT);
David Chinner3c568362008-04-10 12:20:24 +10001396 if (error)
Dave Chinner53487782011-03-07 10:05:35 +11001397 xfs_warn(mp, "%s: dquot %p flush failed",
1398 __func__, dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 xfs_dqflock(dqp);
1400 }
Peter Leckiebc3048e2008-10-30 17:05:04 +11001401 ASSERT(atomic_read(&dqp->q_pincount) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1403 !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
1404
Dave Chinnere6a81f12010-04-13 15:06:51 +10001405 list_del_init(&dqp->q_hashlist);
1406 qh->qh_version++;
Dave Chinner3a254042010-04-13 15:06:48 +10001407 list_del_init(&dqp->q_mplist);
1408 mp->m_quotainfo->qi_dqreclaims++;
1409 mp->m_quotainfo->qi_dquots--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 /*
1411 * XXX Move this to the front of the freelist, if we can get the
1412 * freelist lock.
1413 */
Dave Chinner3a8406f2010-04-13 15:06:52 +10001414 ASSERT(!list_empty(&dqp->q_freelist));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415
1416 dqp->q_mount = NULL;
1417 dqp->q_hash = NULL;
1418 dqp->dq_flags = XFS_DQ_INACTIVE;
1419 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
1420 xfs_dqfunlock(dqp);
1421 xfs_dqunlock(dqp);
Dave Chinnere6a81f12010-04-13 15:06:51 +10001422 mutex_unlock(&qh->qh_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 return (0);
1424}
1425
1426
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427/*
1428 * Give the buffer a little push if it is incore and
1429 * wait on the flush lock.
1430 */
1431void
1432xfs_qm_dqflock_pushbuf_wait(
1433 xfs_dquot_t *dqp)
1434{
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +10001435 xfs_mount_t *mp = dqp->q_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 xfs_buf_t *bp;
1437
1438 /*
1439 * Check to see if the dquot has been flushed delayed
1440 * write. If so, grab its buffer and send it
1441 * out immediately. We'll be able to acquire
1442 * the flush lock when the I/O completes.
1443 */
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +10001444 bp = xfs_incore(mp->m_ddev_targp, dqp->q_blkno,
1445 mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK);
Dave Chinner7d6a7bd2010-01-26 15:13:41 +11001446 if (!bp)
1447 goto out_lock;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001448
Dave Chinner7d6a7bd2010-01-26 15:13:41 +11001449 if (XFS_BUF_ISDELAYWRITE(bp)) {
Chandra Seetharaman811e64c2011-07-22 23:40:27 +00001450 if (xfs_buf_ispinned(bp))
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +10001451 xfs_log_force(mp, 0);
Dave Chinner7d6a7bd2010-01-26 15:13:41 +11001452 xfs_buf_delwri_promote(bp);
1453 wake_up_process(bp->b_target->bt_task);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 }
Dave Chinner7d6a7bd2010-01-26 15:13:41 +11001455 xfs_buf_relse(bp);
1456out_lock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 xfs_dqflock(dqp);
1458}