blob: 63c2de49f61de1782e066156dbec263addcfbbec [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"
Dave Chinner6ca1c902013-08-12 20:49:26 +100020#include "xfs_format.h"
Dave Chinner239880e2013-10-23 10:50:10 +110021#include "xfs_log_format.h"
Dave Chinner70a9883c2013-10-23 10:36:05 +110022#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110023#include "xfs_trans_resv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_inode.h"
29#include "xfs_bmap.h"
Dave Chinner68988112013-08-12 20:49:42 +100030#include "xfs_bmap_util.h"
Dave Chinner239880e2013-10-23 10:50:10 +110031#include "xfs_alloc.h"
32#include "xfs_quota.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_error.h"
Dave Chinner239880e2013-10-23 10:50:10 +110034#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_buf_item.h"
36#include "xfs_trans_space.h"
37#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include "xfs_qm.h"
Christoph Hellwig3fe58f32013-04-03 16:11:16 +110039#include "xfs_cksum.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000040#include "xfs_trace.h"
Dave Chinner239880e2013-10-23 10:50:10 +110041#include "xfs_log.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110042#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/*
Christoph Hellwigbf72de32011-12-06 21:58:19 +000045 * Lock order:
46 *
47 * ip->i_lock
Christoph Hellwig9f920f12012-03-13 08:52:35 +000048 * qi->qi_tree_lock
Christoph Hellwigb84a3a92012-03-14 11:53:34 -050049 * dquot->q_qlock (xfs_dqlock() and friends)
50 * dquot->q_flush (xfs_dqflock() and friends)
51 * qi->qi_lru_lock
Christoph Hellwigbf72de32011-12-06 21:58:19 +000052 *
53 * If two dquots need to be locked the order is user before group/project,
54 * otherwise by the lowest id first, see xfs_dqlock2.
55 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
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 Hellwiga05931c2012-03-13 08:52:37 +000064struct kmem_zone *xfs_qm_dqtrxzone;
65static struct kmem_zone *xfs_qm_dqzone;
66
Dave Chinnerf112a042013-09-30 09:37:03 +100067static struct lock_class_key xfs_dquot_group_class;
68static struct lock_class_key xfs_dquot_project_class;
Christoph Hellwig98b8c7a2009-01-19 02:03:25 +010069
Linus Torvalds1da177e2005-04-16 15:20:36 -070070/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 * This is called to free all the memory associated with a dquot
72 */
73void
74xfs_qm_dqdestroy(
75 xfs_dquot_t *dqp)
76{
Christoph Hellwigf8739c32012-03-13 08:52:34 +000077 ASSERT(list_empty(&dqp->q_lru));
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
79 mutex_destroy(&dqp->q_qlock);
Christoph Hellwiga05931c2012-03-13 08:52:37 +000080 kmem_zone_free(xfs_qm_dqzone, dqp);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000081
Christoph Hellwig48776fd2012-03-13 08:52:33 +000082 XFS_STATS_DEC(xs_qm_dquot);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083}
84
85/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 * If default limits are in force, push them into the dquot now.
87 * We overwrite the dquot limits only if they are zero and this
88 * is not the root dquot.
89 */
90void
91xfs_qm_adjust_dqlimits(
Brian Foster4b6eae2e2013-03-18 10:51:45 -040092 struct xfs_mount *mp,
93 struct xfs_dquot *dq)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094{
Brian Foster4b6eae2e2013-03-18 10:51:45 -040095 struct xfs_quotainfo *q = mp->m_quotainfo;
96 struct xfs_disk_dquot *d = &dq->q_core;
Brian Fosterb1366452013-03-18 10:51:46 -040097 int prealloc = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
99 ASSERT(d->d_id);
100
Brian Fosterb1366452013-03-18 10:51:46 -0400101 if (q->qi_bsoftlimit && !d->d_blk_softlimit) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100102 d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit);
Brian Fosterb1366452013-03-18 10:51:46 -0400103 prealloc = 1;
104 }
105 if (q->qi_bhardlimit && !d->d_blk_hardlimit) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100106 d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit);
Brian Fosterb1366452013-03-18 10:51:46 -0400107 prealloc = 1;
108 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 if (q->qi_isoftlimit && !d->d_ino_softlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100110 d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 if (q->qi_ihardlimit && !d->d_ino_hardlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100112 d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100114 d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100116 d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit);
Brian Fosterb1366452013-03-18 10:51:46 -0400117
118 if (prealloc)
119 xfs_dquot_set_prealloc_limits(dq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120}
121
122/*
123 * Check the limits and timers of a dquot and start or reset timers
124 * if necessary.
125 * This gets called even when quota enforcement is OFF, which makes our
126 * life a little less complicated. (We just don't reject any quota
127 * reservations in that case, when enforcement is off).
128 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
129 * enforcement's off.
130 * In contrast, warnings are a little different in that they don't
Nathan Scott754002b2005-06-21 15:49:06 +1000131 * 'automatically' get started when limits get exceeded. They do
132 * get reset to zero, however, when we find the count to be under
133 * the soft limit (they are only ever set non-zero via userspace).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 */
135void
136xfs_qm_adjust_dqtimers(
137 xfs_mount_t *mp,
138 xfs_disk_dquot_t *d)
139{
140 ASSERT(d->d_id);
141
Christoph Hellwigea15ab32011-07-13 13:43:50 +0200142#ifdef DEBUG
Christoph Hellwig1149d962005-11-02 15:01:12 +1100143 if (d->d_blk_hardlimit)
144 ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
145 be64_to_cpu(d->d_blk_hardlimit));
146 if (d->d_ino_hardlimit)
147 ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
148 be64_to_cpu(d->d_ino_hardlimit));
149 if (d->d_rtb_hardlimit)
150 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
151 be64_to_cpu(d->d_rtb_hardlimit));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152#endif
Christoph Hellwigea15ab32011-07-13 13:43:50 +0200153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 if (!d->d_btimer) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100155 if ((d->d_blk_softlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000156 (be64_to_cpu(d->d_bcount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +1100157 be64_to_cpu(d->d_blk_softlimit))) ||
158 (d->d_blk_hardlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000159 (be64_to_cpu(d->d_bcount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +1100160 be64_to_cpu(d->d_blk_hardlimit)))) {
161 d->d_btimer = cpu_to_be32(get_seconds() +
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000162 mp->m_quotainfo->qi_btimelimit);
Nathan Scott754002b2005-06-21 15:49:06 +1000163 } else {
164 d->d_bwarns = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 }
166 } else {
167 if ((!d->d_blk_softlimit ||
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000168 (be64_to_cpu(d->d_bcount) <=
Christoph Hellwig1149d962005-11-02 15:01:12 +1100169 be64_to_cpu(d->d_blk_softlimit))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 (!d->d_blk_hardlimit ||
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000171 (be64_to_cpu(d->d_bcount) <=
Christoph Hellwig1149d962005-11-02 15:01:12 +1100172 be64_to_cpu(d->d_blk_hardlimit)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 d->d_btimer = 0;
174 }
175 }
176
177 if (!d->d_itimer) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100178 if ((d->d_ino_softlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000179 (be64_to_cpu(d->d_icount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +1100180 be64_to_cpu(d->d_ino_softlimit))) ||
181 (d->d_ino_hardlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000182 (be64_to_cpu(d->d_icount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +1100183 be64_to_cpu(d->d_ino_hardlimit)))) {
184 d->d_itimer = cpu_to_be32(get_seconds() +
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000185 mp->m_quotainfo->qi_itimelimit);
Nathan Scott754002b2005-06-21 15:49:06 +1000186 } else {
187 d->d_iwarns = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 }
189 } else {
190 if ((!d->d_ino_softlimit ||
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000191 (be64_to_cpu(d->d_icount) <=
Christoph Hellwig1149d962005-11-02 15:01:12 +1100192 be64_to_cpu(d->d_ino_softlimit))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 (!d->d_ino_hardlimit ||
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000194 (be64_to_cpu(d->d_icount) <=
Christoph Hellwig1149d962005-11-02 15:01:12 +1100195 be64_to_cpu(d->d_ino_hardlimit)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 d->d_itimer = 0;
197 }
198 }
199
200 if (!d->d_rtbtimer) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100201 if ((d->d_rtb_softlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000202 (be64_to_cpu(d->d_rtbcount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +1100203 be64_to_cpu(d->d_rtb_softlimit))) ||
204 (d->d_rtb_hardlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000205 (be64_to_cpu(d->d_rtbcount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +1100206 be64_to_cpu(d->d_rtb_hardlimit)))) {
207 d->d_rtbtimer = cpu_to_be32(get_seconds() +
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000208 mp->m_quotainfo->qi_rtbtimelimit);
Nathan Scott754002b2005-06-21 15:49:06 +1000209 } else {
210 d->d_rtbwarns = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 }
212 } else {
213 if ((!d->d_rtb_softlimit ||
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000214 (be64_to_cpu(d->d_rtbcount) <=
Christoph Hellwig1149d962005-11-02 15:01:12 +1100215 be64_to_cpu(d->d_rtb_softlimit))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 (!d->d_rtb_hardlimit ||
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000217 (be64_to_cpu(d->d_rtbcount) <=
Christoph Hellwig1149d962005-11-02 15:01:12 +1100218 be64_to_cpu(d->d_rtb_hardlimit)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 d->d_rtbtimer = 0;
220 }
221 }
222}
223
224/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 * initialize a buffer full of dquots and log the whole thing
226 */
227STATIC void
228xfs_qm_init_dquot_blk(
229 xfs_trans_t *tp,
230 xfs_mount_t *mp,
231 xfs_dqid_t id,
232 uint type,
233 xfs_buf_t *bp)
234{
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000235 struct xfs_quotainfo *q = mp->m_quotainfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 xfs_dqblk_t *d;
237 int curid, i;
238
239 ASSERT(tp);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200240 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
Chandra Seetharaman62926042011-07-22 23:40:15 +0000242 d = bp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
244 /*
245 * ID of the first dquot in the block - id's are zero based.
246 */
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000247 curid = id - (id % q->qi_dqperchunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 ASSERT(curid >= 0);
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000249 memset(d, 0, BBTOB(q->qi_dqchunklen));
Christoph Hellwig49d35a52011-12-06 21:58:23 +0000250 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
251 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
252 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
253 d->dd_diskdq.d_id = cpu_to_be32(curid);
254 d->dd_diskdq.d_flags = type;
Dave Chinner6fcdc592013-06-03 15:28:46 +1000255 if (xfs_sb_version_hascrc(&mp->m_sb)) {
Christoph Hellwig3fe58f32013-04-03 16:11:16 +1100256 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid);
Dave Chinner6fcdc592013-06-03 15:28:46 +1000257 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
258 XFS_DQUOT_CRC_OFF);
259 }
Christoph Hellwig49d35a52011-12-06 21:58:23 +0000260 }
261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 xfs_trans_dquot_buf(tp, bp,
Dave Chinnerc1155412010-05-07 11:05:19 +1000263 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
264 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
265 XFS_BLF_GDQUOT_BUF)));
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000266 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267}
268
Brian Fosterb1366452013-03-18 10:51:46 -0400269/*
270 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
271 * watermarks correspond to the soft and hard limits by default. If a soft limit
272 * is not specified, we use 95% of the hard limit.
273 */
274void
275xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
276{
277 __uint64_t space;
278
279 dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit);
280 dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit);
281 if (!dqp->q_prealloc_lo_wmark) {
282 dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
283 do_div(dqp->q_prealloc_lo_wmark, 100);
284 dqp->q_prealloc_lo_wmark *= 95;
285 }
286
287 space = dqp->q_prealloc_hi_wmark;
288
289 do_div(space, 100);
290 dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
291 dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
292 dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
293}
294
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295/*
296 * Allocate a block and fill it with dquots.
297 * This is called when the bmapi finds a hole.
298 */
299STATIC int
300xfs_qm_dqalloc(
Tim Shimminefa092f2005-09-05 08:29:01 +1000301 xfs_trans_t **tpp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 xfs_mount_t *mp,
303 xfs_dquot_t *dqp,
304 xfs_inode_t *quotip,
305 xfs_fileoff_t offset_fsb,
306 xfs_buf_t **O_bpp)
307{
308 xfs_fsblock_t firstblock;
309 xfs_bmap_free_t flist;
310 xfs_bmbt_irec_t map;
311 int nmaps, error, committed;
312 xfs_buf_t *bp;
Tim Shimminefa092f2005-09-05 08:29:01 +1000313 xfs_trans_t *tp = *tpp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
315 ASSERT(tp != NULL);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000316
317 trace_xfs_dqalloc(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
319 /*
320 * Initialize the bmap freelist prior to calling bmapi code.
321 */
Eric Sandeen9d87c312009-01-14 23:22:07 -0600322 xfs_bmap_init(&flist, &firstblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 xfs_ilock(quotip, XFS_ILOCK_EXCL);
324 /*
325 * Return if this type of quotas is turned off while we didn't
326 * have an inode lock
327 */
Chandra Seetharaman6967b962012-01-23 17:31:25 +0000328 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
Dave Chinner24513372014-06-25 14:58:08 +1000330 return -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 }
332
Christoph Hellwigddc34152011-09-19 15:00:54 +0000333 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 nmaps = 1;
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000335 error = xfs_bmapi_write(tp, quotip, offset_fsb,
336 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
337 &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp),
338 &map, &nmaps, &flist);
339 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 goto error0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
342 ASSERT(nmaps == 1);
343 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
344 (map.br_startblock != HOLESTARTBLOCK));
345
346 /*
347 * Keep track of the blkno to save a lookup later
348 */
349 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
350
351 /* now we can just get the buffer (there's nothing to read yet) */
352 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
353 dqp->q_blkno,
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000354 mp->m_quotainfo->qi_dqchunklen,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 0);
Dave Chinner36de9552014-06-06 16:02:12 +1000356 if (!bp) {
Dave Chinner24513372014-06-25 14:58:08 +1000357 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 goto error1;
Dave Chinner36de9552014-06-06 16:02:12 +1000359 }
Dave Chinner1813dd62012-11-14 17:54:40 +1100360 bp->b_ops = &xfs_dquot_buf_ops;
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +0000361
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 /*
363 * Make a chunk of dquots out of this buffer and log
364 * the entire thing.
365 */
Christoph Hellwig1149d962005-11-02 15:01:12 +1100366 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000367 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
Tim Shimminefa092f2005-09-05 08:29:01 +1000369 /*
370 * xfs_bmap_finish() may commit the current transaction and
371 * start a second transaction if the freelist is not empty.
372 *
373 * Since we still want to modify this buffer, we need to
374 * ensure that the buffer is not released on commit of
375 * the first transaction and ensure the buffer is added to the
376 * second transaction.
377 *
378 * If there is only one transaction then don't stop the buffer
379 * from being released when it commits later on.
380 */
381
382 xfs_trans_bhold(tp, bp);
383
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100384 if ((error = xfs_bmap_finish(tpp, &flist, &committed))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 goto error1;
386 }
387
Tim Shimminefa092f2005-09-05 08:29:01 +1000388 if (committed) {
389 tp = *tpp;
390 xfs_trans_bjoin(tp, bp);
391 } else {
392 xfs_trans_bhold_release(tp, bp);
393 }
394
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 *O_bpp = bp;
396 return 0;
397
398 error1:
399 xfs_bmap_cancel(&flist);
400 error0:
401 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
402
Eric Sandeend99831f2014-06-22 15:03:54 +1000403 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404}
Dave Chinner9aede1d2013-10-15 09:17:52 +1100405
Dave Chinnerc6319192012-11-14 17:50:13 +1100406STATIC int
407xfs_qm_dqrepair(
408 struct xfs_mount *mp,
409 struct xfs_trans *tp,
410 struct xfs_dquot *dqp,
411 xfs_dqid_t firstid,
412 struct xfs_buf **bpp)
413{
414 int error;
415 struct xfs_disk_dquot *ddq;
416 struct xfs_dqblk *d;
417 int i;
418
419 /*
420 * Read the buffer without verification so we get the corrupted
Dave Chinner612cfbf2012-11-14 17:52:32 +1100421 * buffer returned to us. make sure we verify it on write, though.
Dave Chinnerc6319192012-11-14 17:50:13 +1100422 */
423 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno,
424 mp->m_quotainfo->qi_dqchunklen,
425 0, bpp, NULL);
426
427 if (error) {
428 ASSERT(*bpp == NULL);
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000429 return error;
Dave Chinnerc6319192012-11-14 17:50:13 +1100430 }
Dave Chinner1813dd62012-11-14 17:54:40 +1100431 (*bpp)->b_ops = &xfs_dquot_buf_ops;
Dave Chinnerc6319192012-11-14 17:50:13 +1100432
433 ASSERT(xfs_buf_islocked(*bpp));
434 d = (struct xfs_dqblk *)(*bpp)->b_addr;
435
436 /* Do the actual repair of dquots in this buffer */
437 for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) {
438 ddq = &d[i].dd_diskdq;
Dave Chinner9aede1d2013-10-15 09:17:52 +1100439 error = xfs_dqcheck(mp, ddq, firstid + i,
Dave Chinnerc6319192012-11-14 17:50:13 +1100440 dqp->dq_flags & XFS_DQ_ALLTYPES,
441 XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair");
442 if (error) {
443 /* repair failed, we're screwed */
444 xfs_trans_brelse(tp, *bpp);
Dave Chinner24513372014-06-25 14:58:08 +1000445 return -EIO;
Dave Chinnerc6319192012-11-14 17:50:13 +1100446 }
447 }
448
449 return 0;
450}
451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452/*
453 * Maps a dquot to the buffer containing its on-disk version.
454 * This returns a ptr to the buffer containing the on-disk dquot
455 * in the bpp param, and a ptr to the on-disk dquot within that buffer
456 */
457STATIC int
458xfs_qm_dqtobp(
Tim Shimminefa092f2005-09-05 08:29:01 +1000459 xfs_trans_t **tpp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 xfs_dquot_t *dqp,
461 xfs_disk_dquot_t **O_ddpp,
462 xfs_buf_t **O_bpp,
463 uint flags)
464{
Chandra Seetharaman113a5682013-06-27 17:25:07 -0500465 struct xfs_bmbt_irec map;
466 int nmaps = 1, error;
467 struct xfs_buf *bp;
Chandra Seetharaman995961c2013-06-27 17:25:06 -0500468 struct xfs_inode *quotip = xfs_dq_to_quota_inode(dqp);
Chandra Seetharaman113a5682013-06-27 17:25:07 -0500469 struct xfs_mount *mp = dqp->q_mount;
470 xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id);
471 struct xfs_trans *tp = (tpp ? *tpp : NULL);
Christoph Hellwigf4df8ad2013-12-06 12:30:13 -0800472 uint lock_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000474 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
Christoph Hellwigf4df8ad2013-12-06 12:30:13 -0800476 lock_mode = xfs_ilock_data_map_shared(quotip);
Chandra Seetharaman6967b962012-01-23 17:31:25 +0000477 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 /*
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000479 * Return if this type of quotas is turned off while we
480 * didn't have the quota inode lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 */
Christoph Hellwigf4df8ad2013-12-06 12:30:13 -0800482 xfs_iunlock(quotip, lock_mode);
Dave Chinner24513372014-06-25 14:58:08 +1000483 return -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
486 /*
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000487 * Find the block map; no allocations yet
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 */
Dave Chinner5c8ed202011-09-18 20:40:45 +0000489 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
490 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000491
Christoph Hellwigf4df8ad2013-12-06 12:30:13 -0800492 xfs_iunlock(quotip, lock_mode);
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000493 if (error)
494 return error;
495
496 ASSERT(nmaps == 1);
497 ASSERT(map.br_blockcount == 1);
498
499 /*
500 * Offset of dquot in the (fixed sized) dquot chunk.
501 */
502 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
503 sizeof(xfs_dqblk_t);
504
505 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
506 if (map.br_startblock == HOLESTARTBLOCK) {
507 /*
508 * We don't allocate unless we're asked to
509 */
510 if (!(flags & XFS_QMOPT_DQALLOC))
Dave Chinner24513372014-06-25 14:58:08 +1000511 return -ENOENT;
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000512
513 ASSERT(tp);
514 error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
515 dqp->q_fileoffset, &bp);
516 if (error)
517 return error;
518 tp = *tpp;
519 } else {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000520 trace_xfs_dqtobp_read(dqp);
521
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000522 /*
523 * store the blkno etc so that we don't have to do the
524 * mapping all the time
525 */
526 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
527
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000528 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
529 dqp->q_blkno,
530 mp->m_quotainfo->qi_dqchunklen,
Dave Chinner1813dd62012-11-14 17:54:40 +1100531 0, &bp, &xfs_dquot_buf_ops);
Dave Chinnerc6319192012-11-14 17:50:13 +1100532
Dave Chinner24513372014-06-25 14:58:08 +1000533 if (error == -EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) {
Dave Chinnerc6319192012-11-14 17:50:13 +1100534 xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff *
535 mp->m_quotainfo->qi_dqperchunk;
536 ASSERT(bp == NULL);
537 error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp);
538 }
539
540 if (error) {
541 ASSERT(bp == NULL);
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000542 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 }
545
Dave Chinnerc6319192012-11-14 17:50:13 +1100546 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 *O_bpp = bp;
Dave Chinnerc6319192012-11-14 17:50:13 +1100548 *O_ddpp = bp->b_addr + dqp->q_bufoffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
Eric Sandeend99831f2014-06-22 15:03:54 +1000550 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551}
552
553
554/*
555 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
556 * and release the buffer immediately.
557 *
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000558 * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 */
Christoph Hellwig7ae44402011-12-06 21:58:25 +0000560int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561xfs_qm_dqread(
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000562 struct xfs_mount *mp,
563 xfs_dqid_t id,
564 uint type,
565 uint flags,
566 struct xfs_dquot **O_dqpp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000568 struct xfs_dquot *dqp;
569 struct xfs_disk_dquot *ddqp;
570 struct xfs_buf *bp;
571 struct xfs_trans *tp = NULL;
572 int error;
573 int cancelflags = 0;
Tim Shimminefa092f2005-09-05 08:29:01 +1000574
Christoph Hellwig92b2e5b32012-02-01 13:57:20 +0000575
Christoph Hellwiga05931c2012-03-13 08:52:37 +0000576 dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP);
Christoph Hellwig92b2e5b32012-02-01 13:57:20 +0000577
578 dqp->dq_flags = type;
579 dqp->q_core.d_id = cpu_to_be32(id);
580 dqp->q_mount = mp;
Christoph Hellwigf8739c32012-03-13 08:52:34 +0000581 INIT_LIST_HEAD(&dqp->q_lru);
Christoph Hellwig92b2e5b32012-02-01 13:57:20 +0000582 mutex_init(&dqp->q_qlock);
583 init_waitqueue_head(&dqp->q_pinwait);
584
585 /*
586 * Because we want to use a counting completion, complete
587 * the flush completion once to allow a single access to
588 * the flush completion without blocking.
589 */
590 init_completion(&dqp->q_flush);
591 complete(&dqp->q_flush);
592
593 /*
594 * Make sure group quotas have a different lock class than user
595 * quotas.
596 */
Dave Chinnerf112a042013-09-30 09:37:03 +1000597 switch (type) {
598 case XFS_DQ_USER:
599 /* uses the default lock class */
600 break;
601 case XFS_DQ_GROUP:
602 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
603 break;
604 case XFS_DQ_PROJ:
605 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
606 break;
607 default:
608 ASSERT(0);
609 break;
610 }
Christoph Hellwig92b2e5b32012-02-01 13:57:20 +0000611
Christoph Hellwig48776fd2012-03-13 08:52:33 +0000612 XFS_STATS_INC(xs_qm_dquot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000614 trace_xfs_dqread(dqp);
615
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000616 if (flags & XFS_QMOPT_DQALLOC) {
617 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
Brian Foster410b11a2014-02-07 14:55:54 +1100618 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_dqalloc,
Jie Liu3d3c8b52013-08-12 20:49:59 +1000619 XFS_QM_DQALLOC_SPACE_RES(mp), 0);
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000620 if (error)
621 goto error1;
622 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
623 }
624
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 /*
626 * get a pointer to the on-disk dquot and the buffer containing it
627 * dqp already knows its own type (GROUP/USER).
628 */
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000629 error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags);
630 if (error) {
631 /*
632 * This can happen if quotas got turned off (ESRCH),
633 * or if the dquot didn't exist on disk and we ask to
634 * allocate (ENOENT).
635 */
636 trace_xfs_dqread_fail(dqp);
637 cancelflags |= XFS_TRANS_ABORT;
638 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 }
640
641 /* copy everything from disk dquot to the incore dquot */
642 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 xfs_qm_dquot_logitem_init(dqp);
644
645 /*
646 * Reservation counters are defined as reservation plus current usage
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300647 * to avoid having to add every time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 */
Christoph Hellwig1149d962005-11-02 15:01:12 +1100649 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
650 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
651 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Brian Fosterb1366452013-03-18 10:51:46 -0400653 /* initialize the dquot speculative prealloc thresholds */
654 xfs_dquot_set_prealloc_limits(dqp);
655
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 /* Mark the buf so that this will stay incore a little longer */
Christoph Hellwig38f23232011-10-10 16:52:45 +0000657 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
659 /*
660 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
661 * So we need to release with xfs_trans_brelse().
662 * The strategy here is identical to that of inodes; we lock
663 * the dquot in xfs_qm_dqget() before making it accessible to
664 * others. This is because dquots, like inodes, need a good level of
665 * concurrency, and we don't want to take locks on the entire buffers
666 * for dquot accesses.
667 * Note also that the dquot buffer may even be dirty at this point, if
668 * this particular dquot was repaired. We still aren't afraid to
669 * brelse it because we have the changes incore.
670 */
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200671 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 xfs_trans_brelse(tp, bp);
673
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 if (tp) {
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000675 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
676 if (error)
677 goto error0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
679
680 *O_dqpp = dqp;
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000681 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000683error1:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 if (tp)
685 xfs_trans_cancel(tp, cancelflags);
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000686error0:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 xfs_qm_dqdestroy(dqp);
688 *O_dqpp = NULL;
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000689 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690}
691
692/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
694 * a locked dquot, doing an allocation (if requested) as needed.
695 * When both an inode and an id are given, the inode's id takes precedence.
696 * That is, if the id changes while we don't hold the ilock inside this
697 * function, the new dquot is returned, not necessarily the one requested
698 * in the id argument.
699 */
700int
701xfs_qm_dqget(
702 xfs_mount_t *mp,
703 xfs_inode_t *ip, /* locked inode (optional) */
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000704 xfs_dqid_t id, /* uid/projid/gid depending on type */
705 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
707 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
708{
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000709 struct xfs_quotainfo *qi = mp->m_quotainfo;
Chandra Seetharaman329e0872013-06-27 17:25:05 -0500710 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000711 struct xfs_dquot *dqp;
712 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
714 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
715 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000716 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
Dave Chinner24513372014-06-25 14:58:08 +1000718 return -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
721#ifdef DEBUG
722 if (xfs_do_dqerror) {
723 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
724 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100725 xfs_debug(mp, "Returning error in dqget");
Dave Chinner24513372014-06-25 14:58:08 +1000726 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 }
728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000730 ASSERT(type == XFS_DQ_USER ||
731 type == XFS_DQ_PROJ ||
732 type == XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 if (ip) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000734 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Chandra Seetharaman36731412012-01-23 17:31:30 +0000735 ASSERT(xfs_inode_dquot(ip, type) == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 }
737#endif
Christoph Hellwig92678552011-12-06 21:58:18 +0000738
739restart:
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000740 mutex_lock(&qi->qi_tree_lock);
741 dqp = radix_tree_lookup(tree, id);
742 if (dqp) {
743 xfs_dqlock(dqp);
744 if (dqp->dq_flags & XFS_DQ_FREEING) {
745 xfs_dqunlock(dqp);
746 mutex_unlock(&qi->qi_tree_lock);
747 trace_xfs_dqget_freeing(dqp);
748 delay(1);
749 goto restart;
750 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000752 dqp->q_nrefs++;
753 mutex_unlock(&qi->qi_tree_lock);
754
755 trace_xfs_dqget_hit(dqp);
Christoph Hellwig48776fd2012-03-13 08:52:33 +0000756 XFS_STATS_INC(xs_qm_dqcachehits);
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000757 *O_dqpp = dqp;
758 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 }
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000760 mutex_unlock(&qi->qi_tree_lock);
761 XFS_STATS_INC(xs_qm_dqcachemisses);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
763 /*
764 * Dquot cache miss. We don't want to keep the inode lock across
765 * a (potential) disk read. Also we don't want to deal with the lock
766 * ordering between quotainode and this inode. OTOH, dropping the inode
767 * lock here means dealing with a chown that can happen before
768 * we re-acquire the lock.
769 */
770 if (ip)
771 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000773 error = xfs_qm_dqread(mp, id, type, flags, &dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Christoph Hellwig7ae44402011-12-06 21:58:25 +0000775 if (ip)
776 xfs_ilock(ip, XFS_ILOCK_EXCL);
777
778 if (error)
779 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 if (ip) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 /*
783 * A dquot could be attached to this inode by now, since
784 * we had dropped the ilock.
785 */
Chandra Seetharaman36731412012-01-23 17:31:30 +0000786 if (xfs_this_quota_on(mp, type)) {
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000787 struct xfs_dquot *dqp1;
788
Chandra Seetharaman36731412012-01-23 17:31:30 +0000789 dqp1 = xfs_inode_dquot(ip, type);
790 if (dqp1) {
Christoph Hellwig191f8482010-04-20 17:01:53 +1000791 xfs_qm_dqdestroy(dqp);
Chandra Seetharaman36731412012-01-23 17:31:30 +0000792 dqp = dqp1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 xfs_dqlock(dqp);
794 goto dqret;
795 }
796 } else {
Chandra Seetharaman36731412012-01-23 17:31:30 +0000797 /* inode stays locked on return */
798 xfs_qm_dqdestroy(dqp);
Dave Chinner24513372014-06-25 14:58:08 +1000799 return -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 }
801 }
802
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000803 mutex_lock(&qi->qi_tree_lock);
Dave Chinner24513372014-06-25 14:58:08 +1000804 error = radix_tree_insert(tree, id, dqp);
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000805 if (unlikely(error)) {
Dave Chinner24513372014-06-25 14:58:08 +1000806 WARN_ON(error != -EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000808 /*
809 * Duplicate found. Just throw away the new dquot and start
810 * over.
811 */
812 mutex_unlock(&qi->qi_tree_lock);
813 trace_xfs_dqget_dup(dqp);
814 xfs_qm_dqdestroy(dqp);
815 XFS_STATS_INC(xs_qm_dquot_dups);
816 goto restart;
817 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
819 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 * We return a locked dquot to the caller, with a reference taken
821 */
822 xfs_dqlock(dqp);
823 dqp->q_nrefs = 1;
824
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000825 qi->qi_dquots++;
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000826 mutex_unlock(&qi->qi_tree_lock);
827
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 dqret:
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000829 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000830 trace_xfs_dqget_miss(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 *O_dqpp = dqp;
Eric Sandeend99831f2014-06-22 15:03:54 +1000832 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
834
Christoph Hellwigf8739c32012-03-13 08:52:34 +0000835/*
836 * Release a reference to the dquot (decrement ref-count) and unlock it.
837 *
838 * If there is a group quota attached to this dquot, carefully release that
839 * too without tripping over deadlocks'n'stuff.
840 */
841void
842xfs_qm_dqput(
843 struct xfs_dquot *dqp)
844{
845 ASSERT(dqp->q_nrefs > 0);
846 ASSERT(XFS_DQ_IS_LOCKED(dqp));
847
848 trace_xfs_dqput(dqp);
849
Dave Chinner3c35337572014-05-05 17:30:15 +1000850 if (--dqp->q_nrefs == 0) {
851 struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
852 trace_xfs_dqput_free(dqp);
853
854 if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
855 XFS_STATS_INC(xs_qm_dquot_unused);
856 }
857 xfs_dqunlock(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858}
859
860/*
861 * Release a dquot. Flush it if dirty, then dqput() it.
862 * dquot must not be locked.
863 */
864void
865xfs_qm_dqrele(
866 xfs_dquot_t *dqp)
867{
Christoph Hellwig7d095252009-06-08 15:33:32 +0200868 if (!dqp)
869 return;
870
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000871 trace_xfs_dqrele(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
873 xfs_dqlock(dqp);
874 /*
875 * We don't care to flush it if the dquot is dirty here.
876 * That will create stutters that we want to avoid.
877 * Instead we do a delayed write when we try to reclaim
878 * a dirty dquot. Also xfs_sync will take part of the burden...
879 */
880 xfs_qm_dqput(dqp);
881}
882
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000883/*
884 * This is the dquot flushing I/O completion routine. It is called
885 * from interrupt level when the buffer containing the dquot is
886 * flushed to disk. It is responsible for removing the dquot logitem
887 * from the AIL if it has not been re-logged, and unlocking the dquot's
888 * flush lock. This behavior is very similar to that of inodes..
889 */
890STATIC void
891xfs_qm_dqflush_done(
892 struct xfs_buf *bp,
893 struct xfs_log_item *lip)
894{
895 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
896 xfs_dquot_t *dqp = qip->qli_dquot;
897 struct xfs_ail *ailp = lip->li_ailp;
898
899 /*
900 * We only want to pull the item from the AIL if its
901 * location in the log has not changed since we started the flush.
902 * Thus, we only bother if the dquot's lsn has
903 * not changed. First we check the lsn outside the lock
904 * since it's cheaper, and then we recheck while
905 * holding the lock before removing the dquot from the AIL.
906 */
907 if ((lip->li_flags & XFS_LI_IN_AIL) &&
908 lip->li_lsn == qip->qli_flush_lsn) {
909
910 /* xfs_trans_ail_delete() drops the AIL lock. */
911 spin_lock(&ailp->xa_lock);
912 if (lip->li_lsn == qip->qli_flush_lsn)
Dave Chinner04913fd2012-04-23 15:58:41 +1000913 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000914 else
915 spin_unlock(&ailp->xa_lock);
916 }
917
918 /*
919 * Release the dq's flush lock since we're done with it.
920 */
921 xfs_dqfunlock(dqp);
922}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
924/*
925 * Write a modified dquot to disk.
926 * The dquot must be locked and the flush lock too taken by caller.
927 * The flush lock will not be unlocked until the dquot reaches the disk,
928 * but the dquot is free to be unlocked and modified by the caller
929 * in the interim. Dquot is still locked on return. This behavior is
930 * identical to that of inodes.
931 */
932int
933xfs_qm_dqflush(
Christoph Hellwigfe7257f2012-04-23 15:58:37 +1000934 struct xfs_dquot *dqp,
935 struct xfs_buf **bpp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936{
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000937 struct xfs_mount *mp = dqp->q_mount;
938 struct xfs_buf *bp;
939 struct xfs_disk_dquot *ddqp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
942 ASSERT(XFS_DQ_IS_LOCKED(dqp));
David Chinnere1f49cf2008-08-13 16:41:43 +1000943 ASSERT(!completion_done(&dqp->q_flush));
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000944
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000945 trace_xfs_dqflush(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
Christoph Hellwigfe7257f2012-04-23 15:58:37 +1000947 *bpp = NULL;
948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 xfs_qm_dqunpin_wait(dqp);
950
951 /*
952 * This may have been unpinned because the filesystem is shutting
953 * down forcibly. If that's the case we must not write this dquot
Christoph Hellwigdea96092012-04-23 15:58:31 +1000954 * to disk, because the log record didn't make it to disk.
955 *
956 * We also have to remove the log item from the AIL in this case,
957 * as we wait for an emptry AIL as part of the unmount process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 */
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000959 if (XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwigdea96092012-04-23 15:58:31 +1000960 struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000961 dqp->dq_flags &= ~XFS_DQ_DIRTY;
Christoph Hellwigdea96092012-04-23 15:58:31 +1000962
963 spin_lock(&mp->m_ail->xa_lock);
964 if (lip->li_flags & XFS_LI_IN_AIL)
Dave Chinner04913fd2012-04-23 15:58:41 +1000965 xfs_trans_ail_delete(mp->m_ail, lip,
966 SHUTDOWN_CORRUPT_INCORE);
Christoph Hellwigdea96092012-04-23 15:58:31 +1000967 else
968 spin_unlock(&mp->m_ail->xa_lock);
Dave Chinner24513372014-06-25 14:58:08 +1000969 error = -EIO;
Christoph Hellwigfe7257f2012-04-23 15:58:37 +1000970 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 }
972
973 /*
974 * Get the buffer containing the on-disk dquot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 */
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000976 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
Dave Chinner5fd364f2014-08-04 12:43:26 +1000977 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
978 &xfs_dquot_buf_ops);
Christoph Hellwigfe7257f2012-04-23 15:58:37 +1000979 if (error)
980 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000982 /*
983 * Calculate the location of the dquot inside the buffer.
984 */
Chandra Seetharaman62926042011-07-22 23:40:15 +0000985 ddqp = bp->b_addr + dqp->q_bufoffset;
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000986
987 /*
988 * A simple sanity check in case we got a corrupted dquot..
989 */
Dave Chinner9aede1d2013-10-15 09:17:52 +1100990 error = xfs_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100991 XFS_QMOPT_DOWARN, "dqflush (incore copy)");
992 if (error) {
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000993 xfs_buf_relse(bp);
994 xfs_dqfunlock(dqp);
995 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Dave Chinner24513372014-06-25 14:58:08 +1000996 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 }
998
999 /* This is the only portion of data that needs to persist */
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001000 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001
1002 /*
1003 * Clear the dirty field and remember the flush lsn for later use.
1004 */
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001005 dqp->dq_flags &= ~XFS_DQ_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
David Chinner7b2e2a32008-10-30 17:39:12 +11001007 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1008 &dqp->q_logitem.qli_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
1010 /*
Christoph Hellwig3fe58f32013-04-03 16:11:16 +11001011 * copy the lsn into the on-disk dquot now while we have the in memory
1012 * dquot here. This can't be done later in the write verifier as we
1013 * can't get access to the log item at that point in time.
Dave Chinner6fcdc592013-06-03 15:28:46 +10001014 *
1015 * We also calculate the CRC here so that the on-disk dquot in the
1016 * buffer always has a valid CRC. This ensures there is no possibility
1017 * of a dquot without an up-to-date CRC getting to disk.
Christoph Hellwig3fe58f32013-04-03 16:11:16 +11001018 */
1019 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1020 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp;
1021
1022 dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
Dave Chinner6fcdc592013-06-03 15:28:46 +10001023 xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk),
1024 XFS_DQUOT_CRC_OFF);
Christoph Hellwig3fe58f32013-04-03 16:11:16 +11001025 }
1026
1027 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 * Attach an iodone routine so that we can remove this dquot from the
1029 * AIL and release the flush lock once the dquot is synced to disk.
1030 */
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001031 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1032 &dqp->q_logitem.qli_item);
1033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 /*
1035 * If the buffer is pinned then push on the log so we won't
1036 * get stuck waiting in the write for too long.
1037 */
Chandra Seetharaman811e64c2011-07-22 23:40:27 +00001038 if (xfs_buf_ispinned(bp)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001039 trace_xfs_dqflush_force(dqp);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001040 xfs_log_force(mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 }
1042
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001043 trace_xfs_dqflush_done(dqp);
Christoph Hellwigfe7257f2012-04-23 15:58:37 +10001044 *bpp = bp;
1045 return 0;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001046
Christoph Hellwigfe7257f2012-04-23 15:58:37 +10001047out_unlock:
1048 xfs_dqfunlock(dqp);
Dave Chinner24513372014-06-25 14:58:08 +10001049 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050}
1051
Christoph Hellwig5bb87a32009-01-19 02:03:19 +01001052/*
1053 * Lock two xfs_dquot structures.
1054 *
1055 * To avoid deadlocks we always lock the quota structure with
1056 * the lowerd id first.
1057 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058void
1059xfs_dqlock2(
1060 xfs_dquot_t *d1,
1061 xfs_dquot_t *d2)
1062{
1063 if (d1 && d2) {
1064 ASSERT(d1 != d2);
Christoph Hellwig1149d962005-11-02 15:01:12 +11001065 if (be32_to_cpu(d1->q_core.d_id) >
1066 be32_to_cpu(d2->q_core.d_id)) {
Christoph Hellwig5bb87a32009-01-19 02:03:19 +01001067 mutex_lock(&d2->q_qlock);
1068 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 } else {
Christoph Hellwig5bb87a32009-01-19 02:03:19 +01001070 mutex_lock(&d1->q_qlock);
1071 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
Christoph Hellwig5bb87a32009-01-19 02:03:19 +01001073 } else if (d1) {
1074 mutex_lock(&d1->q_qlock);
1075 } else if (d2) {
1076 mutex_lock(&d2->q_qlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 }
1078}
1079
Christoph Hellwiga05931c2012-03-13 08:52:37 +00001080int __init
1081xfs_qm_init(void)
1082{
1083 xfs_qm_dqzone =
1084 kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
1085 if (!xfs_qm_dqzone)
1086 goto out;
1087
1088 xfs_qm_dqtrxzone =
1089 kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
1090 if (!xfs_qm_dqtrxzone)
1091 goto out_free_dqzone;
1092
1093 return 0;
1094
1095out_free_dqzone:
1096 kmem_zone_destroy(xfs_qm_dqzone);
1097out:
1098 return -ENOMEM;
1099}
1100
Gerard Snitselaar1c2ccc62012-03-16 18:36:18 +00001101void
Christoph Hellwiga05931c2012-03-13 08:52:37 +00001102xfs_qm_exit(void)
1103{
1104 kmem_zone_destroy(xfs_qm_dqtrxzone);
1105 kmem_zone_destroy(xfs_qm_dqzone);
1106}