blob: 02c01bbbc7899410e9d23f423c51086d23afa940 [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 Chinner70a98832013-10-23 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_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_inode.h"
27#include "xfs_bmap.h"
Dave Chinner68988112013-08-12 20:49:42 +100028#include "xfs_bmap_util.h"
Dave Chinner239880e2013-10-23 10:50:10 +110029#include "xfs_alloc.h"
30#include "xfs_quota.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_error.h"
Dave Chinner239880e2013-10-23 10:50:10 +110032#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_buf_item.h"
34#include "xfs_trans_space.h"
35#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_qm.h"
Christoph Hellwig3fe58f32013-04-03 16:11:16 +110037#include "xfs_cksum.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000038#include "xfs_trace.h"
Dave Chinner239880e2013-10-23 10:50:10 +110039#include "xfs_log.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110040#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/*
Christoph Hellwigbf72de32011-12-06 21:58:19 +000043 * Lock order:
44 *
45 * ip->i_lock
Christoph Hellwig9f920f12012-03-13 08:52:35 +000046 * qi->qi_tree_lock
Christoph Hellwigb84a3a92012-03-14 11:53:34 -050047 * dquot->q_qlock (xfs_dqlock() and friends)
48 * dquot->q_flush (xfs_dqflock() and friends)
49 * qi->qi_lru_lock
Christoph Hellwigbf72de32011-12-06 21:58:19 +000050 *
51 * If two dquots need to be locked the order is user before group/project,
52 * otherwise by the lowest id first, see xfs_dqlock2.
53 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#ifdef DEBUG
56xfs_buftarg_t *xfs_dqerror_target;
57int xfs_do_dqerror;
58int xfs_dqreq_num;
59int xfs_dqerror_mod = 33;
60#endif
61
Christoph Hellwiga05931c2012-03-13 08:52:37 +000062struct kmem_zone *xfs_qm_dqtrxzone;
63static struct kmem_zone *xfs_qm_dqzone;
64
Dave Chinnerf112a042013-09-30 09:37:03 +100065static struct lock_class_key xfs_dquot_group_class;
66static struct lock_class_key xfs_dquot_project_class;
Christoph Hellwig98b8c7a2009-01-19 02:03:25 +010067
Linus Torvalds1da177e2005-04-16 15:20:36 -070068/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 * This is called to free all the memory associated with a dquot
70 */
71void
72xfs_qm_dqdestroy(
73 xfs_dquot_t *dqp)
74{
Christoph Hellwigf8739c32012-03-13 08:52:34 +000075 ASSERT(list_empty(&dqp->q_lru));
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
77 mutex_destroy(&dqp->q_qlock);
Christoph Hellwiga05931c2012-03-13 08:52:37 +000078 kmem_zone_free(xfs_qm_dqzone, dqp);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000079
Christoph Hellwig48776fd2012-03-13 08:52:33 +000080 XFS_STATS_DEC(xs_qm_dquot);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081}
82
83/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 * If default limits are in force, push them into the dquot now.
85 * We overwrite the dquot limits only if they are zero and this
86 * is not the root dquot.
87 */
88void
89xfs_qm_adjust_dqlimits(
Brian Foster4b6eae2e2013-03-18 10:51:45 -040090 struct xfs_mount *mp,
91 struct xfs_dquot *dq)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092{
Brian Foster4b6eae2e2013-03-18 10:51:45 -040093 struct xfs_quotainfo *q = mp->m_quotainfo;
94 struct xfs_disk_dquot *d = &dq->q_core;
Brian Fosterb1366452013-03-18 10:51:46 -040095 int prealloc = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
97 ASSERT(d->d_id);
98
Brian Fosterb1366452013-03-18 10:51:46 -040099 if (q->qi_bsoftlimit && !d->d_blk_softlimit) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100100 d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit);
Brian Fosterb1366452013-03-18 10:51:46 -0400101 prealloc = 1;
102 }
103 if (q->qi_bhardlimit && !d->d_blk_hardlimit) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100104 d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit);
Brian Fosterb1366452013-03-18 10:51:46 -0400105 prealloc = 1;
106 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 if (q->qi_isoftlimit && !d->d_ino_softlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100108 d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 if (q->qi_ihardlimit && !d->d_ino_hardlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100110 d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100112 d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit)
Christoph Hellwig1149d962005-11-02 15:01:12 +1100114 d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit);
Brian Fosterb1366452013-03-18 10:51:46 -0400115
116 if (prealloc)
117 xfs_dquot_set_prealloc_limits(dq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118}
119
120/*
121 * Check the limits and timers of a dquot and start or reset timers
122 * if necessary.
123 * This gets called even when quota enforcement is OFF, which makes our
124 * life a little less complicated. (We just don't reject any quota
125 * reservations in that case, when enforcement is off).
126 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
127 * enforcement's off.
128 * In contrast, warnings are a little different in that they don't
Nathan Scott754002b2005-06-21 15:49:06 +1000129 * 'automatically' get started when limits get exceeded. They do
130 * get reset to zero, however, when we find the count to be under
131 * the soft limit (they are only ever set non-zero via userspace).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 */
133void
134xfs_qm_adjust_dqtimers(
135 xfs_mount_t *mp,
136 xfs_disk_dquot_t *d)
137{
138 ASSERT(d->d_id);
139
Christoph Hellwigea15ab32011-07-13 13:43:50 +0200140#ifdef DEBUG
Christoph Hellwig1149d962005-11-02 15:01:12 +1100141 if (d->d_blk_hardlimit)
142 ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
143 be64_to_cpu(d->d_blk_hardlimit));
144 if (d->d_ino_hardlimit)
145 ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
146 be64_to_cpu(d->d_ino_hardlimit));
147 if (d->d_rtb_hardlimit)
148 ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
149 be64_to_cpu(d->d_rtb_hardlimit));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150#endif
Christoph Hellwigea15ab32011-07-13 13:43:50 +0200151
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 if (!d->d_btimer) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100153 if ((d->d_blk_softlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000154 (be64_to_cpu(d->d_bcount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +1100155 be64_to_cpu(d->d_blk_softlimit))) ||
156 (d->d_blk_hardlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000157 (be64_to_cpu(d->d_bcount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +1100158 be64_to_cpu(d->d_blk_hardlimit)))) {
159 d->d_btimer = cpu_to_be32(get_seconds() +
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000160 mp->m_quotainfo->qi_btimelimit);
Nathan Scott754002b2005-06-21 15:49:06 +1000161 } else {
162 d->d_bwarns = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 }
164 } else {
165 if ((!d->d_blk_softlimit ||
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000166 (be64_to_cpu(d->d_bcount) <=
Christoph Hellwig1149d962005-11-02 15:01:12 +1100167 be64_to_cpu(d->d_blk_softlimit))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 (!d->d_blk_hardlimit ||
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000169 (be64_to_cpu(d->d_bcount) <=
Christoph Hellwig1149d962005-11-02 15:01:12 +1100170 be64_to_cpu(d->d_blk_hardlimit)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 d->d_btimer = 0;
172 }
173 }
174
175 if (!d->d_itimer) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100176 if ((d->d_ino_softlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000177 (be64_to_cpu(d->d_icount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +1100178 be64_to_cpu(d->d_ino_softlimit))) ||
179 (d->d_ino_hardlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000180 (be64_to_cpu(d->d_icount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +1100181 be64_to_cpu(d->d_ino_hardlimit)))) {
182 d->d_itimer = cpu_to_be32(get_seconds() +
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000183 mp->m_quotainfo->qi_itimelimit);
Nathan Scott754002b2005-06-21 15:49:06 +1000184 } else {
185 d->d_iwarns = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 }
187 } else {
188 if ((!d->d_ino_softlimit ||
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000189 (be64_to_cpu(d->d_icount) <=
Christoph Hellwig1149d962005-11-02 15:01:12 +1100190 be64_to_cpu(d->d_ino_softlimit))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 (!d->d_ino_hardlimit ||
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000192 (be64_to_cpu(d->d_icount) <=
Christoph Hellwig1149d962005-11-02 15:01:12 +1100193 be64_to_cpu(d->d_ino_hardlimit)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 d->d_itimer = 0;
195 }
196 }
197
198 if (!d->d_rtbtimer) {
Christoph Hellwig1149d962005-11-02 15:01:12 +1100199 if ((d->d_rtb_softlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000200 (be64_to_cpu(d->d_rtbcount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +1100201 be64_to_cpu(d->d_rtb_softlimit))) ||
202 (d->d_rtb_hardlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000203 (be64_to_cpu(d->d_rtbcount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +1100204 be64_to_cpu(d->d_rtb_hardlimit)))) {
205 d->d_rtbtimer = cpu_to_be32(get_seconds() +
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000206 mp->m_quotainfo->qi_rtbtimelimit);
Nathan Scott754002b2005-06-21 15:49:06 +1000207 } else {
208 d->d_rtbwarns = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 }
210 } else {
211 if ((!d->d_rtb_softlimit ||
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000212 (be64_to_cpu(d->d_rtbcount) <=
Christoph Hellwig1149d962005-11-02 15:01:12 +1100213 be64_to_cpu(d->d_rtb_softlimit))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 (!d->d_rtb_hardlimit ||
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +0000215 (be64_to_cpu(d->d_rtbcount) <=
Christoph Hellwig1149d962005-11-02 15:01:12 +1100216 be64_to_cpu(d->d_rtb_hardlimit)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 d->d_rtbtimer = 0;
218 }
219 }
220}
221
222/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 * initialize a buffer full of dquots and log the whole thing
224 */
225STATIC void
226xfs_qm_init_dquot_blk(
227 xfs_trans_t *tp,
228 xfs_mount_t *mp,
229 xfs_dqid_t id,
230 uint type,
231 xfs_buf_t *bp)
232{
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000233 struct xfs_quotainfo *q = mp->m_quotainfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 xfs_dqblk_t *d;
235 int curid, i;
236
237 ASSERT(tp);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200238 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Chandra Seetharaman62926042011-07-22 23:40:15 +0000240 d = bp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
242 /*
243 * ID of the first dquot in the block - id's are zero based.
244 */
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000245 curid = id - (id % q->qi_dqperchunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 ASSERT(curid >= 0);
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000247 memset(d, 0, BBTOB(q->qi_dqchunklen));
Christoph Hellwig49d35a52011-12-06 21:58:23 +0000248 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
249 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
250 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
251 d->dd_diskdq.d_id = cpu_to_be32(curid);
252 d->dd_diskdq.d_flags = type;
Dave Chinner6fcdc592013-06-03 15:28:46 +1000253 if (xfs_sb_version_hascrc(&mp->m_sb)) {
Christoph Hellwig3fe58f32013-04-03 16:11:16 +1100254 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid);
Dave Chinner6fcdc592013-06-03 15:28:46 +1000255 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
256 XFS_DQUOT_CRC_OFF);
257 }
Christoph Hellwig49d35a52011-12-06 21:58:23 +0000258 }
259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 xfs_trans_dquot_buf(tp, bp,
Dave Chinnerc1155412010-05-07 11:05:19 +1000261 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
262 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
263 XFS_BLF_GDQUOT_BUF)));
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000264 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
266
Brian Fosterb1366452013-03-18 10:51:46 -0400267/*
268 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
269 * watermarks correspond to the soft and hard limits by default. If a soft limit
270 * is not specified, we use 95% of the hard limit.
271 */
272void
273xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
274{
275 __uint64_t space;
276
277 dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit);
278 dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit);
279 if (!dqp->q_prealloc_lo_wmark) {
280 dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
281 do_div(dqp->q_prealloc_lo_wmark, 100);
282 dqp->q_prealloc_lo_wmark *= 95;
283 }
284
285 space = dqp->q_prealloc_hi_wmark;
286
287 do_div(space, 100);
288 dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
289 dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
290 dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
291}
292
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293/*
294 * Allocate a block and fill it with dquots.
295 * This is called when the bmapi finds a hole.
296 */
297STATIC int
298xfs_qm_dqalloc(
Tim Shimminefa092f2005-09-05 08:29:01 +1000299 xfs_trans_t **tpp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 xfs_mount_t *mp,
301 xfs_dquot_t *dqp,
302 xfs_inode_t *quotip,
303 xfs_fileoff_t offset_fsb,
304 xfs_buf_t **O_bpp)
305{
306 xfs_fsblock_t firstblock;
307 xfs_bmap_free_t flist;
308 xfs_bmbt_irec_t map;
309 int nmaps, error, committed;
310 xfs_buf_t *bp;
Tim Shimminefa092f2005-09-05 08:29:01 +1000311 xfs_trans_t *tp = *tpp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
313 ASSERT(tp != NULL);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000314
315 trace_xfs_dqalloc(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317 /*
318 * Initialize the bmap freelist prior to calling bmapi code.
319 */
Eric Sandeen9d87c312009-01-14 23:22:07 -0600320 xfs_bmap_init(&flist, &firstblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 xfs_ilock(quotip, XFS_ILOCK_EXCL);
322 /*
323 * Return if this type of quotas is turned off while we didn't
324 * have an inode lock
325 */
Chandra Seetharaman6967b962012-01-23 17:31:25 +0000326 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
Dave Chinner24513372014-06-25 14:58:08 +1000328 return -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 }
330
Christoph Hellwigddc34152011-09-19 15:00:54 +0000331 xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 nmaps = 1;
Dave Chinnerc0dc7822011-09-18 20:40:52 +0000333 error = xfs_bmapi_write(tp, quotip, offset_fsb,
334 XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
335 &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp),
336 &map, &nmaps, &flist);
337 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 goto error0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
340 ASSERT(nmaps == 1);
341 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
342 (map.br_startblock != HOLESTARTBLOCK));
343
344 /*
345 * Keep track of the blkno to save a lookup later
346 */
347 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
348
349 /* now we can just get the buffer (there's nothing to read yet) */
350 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
351 dqp->q_blkno,
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000352 mp->m_quotainfo->qi_dqchunklen,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 0);
Dave Chinner36de9552014-06-06 16:02:12 +1000354 if (!bp) {
Dave Chinner24513372014-06-25 14:58:08 +1000355 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 goto error1;
Dave Chinner36de9552014-06-06 16:02:12 +1000357 }
Dave Chinner1813dd62012-11-14 17:54:40 +1100358 bp->b_ops = &xfs_dquot_buf_ops;
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +0000359
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 /*
361 * Make a chunk of dquots out of this buffer and log
362 * the entire thing.
363 */
Christoph Hellwig1149d962005-11-02 15:01:12 +1100364 xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000365 dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
Tim Shimminefa092f2005-09-05 08:29:01 +1000367 /*
368 * xfs_bmap_finish() may commit the current transaction and
369 * start a second transaction if the freelist is not empty.
370 *
371 * Since we still want to modify this buffer, we need to
372 * ensure that the buffer is not released on commit of
373 * the first transaction and ensure the buffer is added to the
374 * second transaction.
375 *
376 * If there is only one transaction then don't stop the buffer
377 * from being released when it commits later on.
378 */
379
380 xfs_trans_bhold(tp, bp);
381
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100382 if ((error = xfs_bmap_finish(tpp, &flist, &committed))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 goto error1;
384 }
385
Tim Shimminefa092f2005-09-05 08:29:01 +1000386 if (committed) {
387 tp = *tpp;
388 xfs_trans_bjoin(tp, bp);
389 } else {
390 xfs_trans_bhold_release(tp, bp);
391 }
392
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 *O_bpp = bp;
394 return 0;
395
396 error1:
397 xfs_bmap_cancel(&flist);
398 error0:
399 xfs_iunlock(quotip, XFS_ILOCK_EXCL);
400
Eric Sandeend99831f2014-06-22 15:03:54 +1000401 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
Dave Chinner9aede1d2013-10-15 09:17:52 +1100403
Dave Chinnerc6319192012-11-14 17:50:13 +1100404STATIC int
405xfs_qm_dqrepair(
406 struct xfs_mount *mp,
407 struct xfs_trans *tp,
408 struct xfs_dquot *dqp,
409 xfs_dqid_t firstid,
410 struct xfs_buf **bpp)
411{
412 int error;
413 struct xfs_disk_dquot *ddq;
414 struct xfs_dqblk *d;
415 int i;
416
417 /*
418 * Read the buffer without verification so we get the corrupted
Dave Chinner612cfbf2012-11-14 17:52:32 +1100419 * buffer returned to us. make sure we verify it on write, though.
Dave Chinnerc6319192012-11-14 17:50:13 +1100420 */
421 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno,
422 mp->m_quotainfo->qi_dqchunklen,
423 0, bpp, NULL);
424
425 if (error) {
426 ASSERT(*bpp == NULL);
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000427 return error;
Dave Chinnerc6319192012-11-14 17:50:13 +1100428 }
Dave Chinner1813dd62012-11-14 17:54:40 +1100429 (*bpp)->b_ops = &xfs_dquot_buf_ops;
Dave Chinnerc6319192012-11-14 17:50:13 +1100430
431 ASSERT(xfs_buf_islocked(*bpp));
432 d = (struct xfs_dqblk *)(*bpp)->b_addr;
433
434 /* Do the actual repair of dquots in this buffer */
435 for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) {
436 ddq = &d[i].dd_diskdq;
Dave Chinner9aede1d2013-10-15 09:17:52 +1100437 error = xfs_dqcheck(mp, ddq, firstid + i,
Dave Chinnerc6319192012-11-14 17:50:13 +1100438 dqp->dq_flags & XFS_DQ_ALLTYPES,
439 XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair");
440 if (error) {
441 /* repair failed, we're screwed */
442 xfs_trans_brelse(tp, *bpp);
Dave Chinner24513372014-06-25 14:58:08 +1000443 return -EIO;
Dave Chinnerc6319192012-11-14 17:50:13 +1100444 }
445 }
446
447 return 0;
448}
449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450/*
451 * Maps a dquot to the buffer containing its on-disk version.
452 * This returns a ptr to the buffer containing the on-disk dquot
453 * in the bpp param, and a ptr to the on-disk dquot within that buffer
454 */
455STATIC int
456xfs_qm_dqtobp(
Tim Shimminefa092f2005-09-05 08:29:01 +1000457 xfs_trans_t **tpp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 xfs_dquot_t *dqp,
459 xfs_disk_dquot_t **O_ddpp,
460 xfs_buf_t **O_bpp,
461 uint flags)
462{
Chandra Seetharaman113a5682013-06-27 17:25:07 -0500463 struct xfs_bmbt_irec map;
464 int nmaps = 1, error;
465 struct xfs_buf *bp;
Chandra Seetharaman995961c2013-06-27 17:25:06 -0500466 struct xfs_inode *quotip = xfs_dq_to_quota_inode(dqp);
Chandra Seetharaman113a5682013-06-27 17:25:07 -0500467 struct xfs_mount *mp = dqp->q_mount;
468 xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id);
469 struct xfs_trans *tp = (tpp ? *tpp : NULL);
Christoph Hellwigf4df8ad2013-12-06 12:30:13 -0800470 uint lock_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000472 dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Christoph Hellwigf4df8ad2013-12-06 12:30:13 -0800474 lock_mode = xfs_ilock_data_map_shared(quotip);
Chandra Seetharaman6967b962012-01-23 17:31:25 +0000475 if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 /*
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000477 * Return if this type of quotas is turned off while we
478 * didn't have the quota inode lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 */
Christoph Hellwigf4df8ad2013-12-06 12:30:13 -0800480 xfs_iunlock(quotip, lock_mode);
Dave Chinner24513372014-06-25 14:58:08 +1000481 return -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
484 /*
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000485 * Find the block map; no allocations yet
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 */
Dave Chinner5c8ed202011-09-18 20:40:45 +0000487 error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
488 XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000489
Christoph Hellwigf4df8ad2013-12-06 12:30:13 -0800490 xfs_iunlock(quotip, lock_mode);
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000491 if (error)
492 return error;
493
494 ASSERT(nmaps == 1);
495 ASSERT(map.br_blockcount == 1);
496
497 /*
498 * Offset of dquot in the (fixed sized) dquot chunk.
499 */
500 dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
501 sizeof(xfs_dqblk_t);
502
503 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
504 if (map.br_startblock == HOLESTARTBLOCK) {
505 /*
506 * We don't allocate unless we're asked to
507 */
508 if (!(flags & XFS_QMOPT_DQALLOC))
Dave Chinner24513372014-06-25 14:58:08 +1000509 return -ENOENT;
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000510
511 ASSERT(tp);
512 error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
513 dqp->q_fileoffset, &bp);
514 if (error)
515 return error;
516 tp = *tpp;
517 } else {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000518 trace_xfs_dqtobp_read(dqp);
519
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000520 /*
521 * store the blkno etc so that we don't have to do the
522 * mapping all the time
523 */
524 dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
525
Christoph Hellwig8a7b8a82010-04-20 17:01:30 +1000526 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
527 dqp->q_blkno,
528 mp->m_quotainfo->qi_dqchunklen,
Dave Chinner1813dd62012-11-14 17:54:40 +1100529 0, &bp, &xfs_dquot_buf_ops);
Dave Chinnerc6319192012-11-14 17:50:13 +1100530
Dave Chinner24513372014-06-25 14:58:08 +1000531 if (error == -EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) {
Dave Chinnerc6319192012-11-14 17:50:13 +1100532 xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff *
533 mp->m_quotainfo->qi_dqperchunk;
534 ASSERT(bp == NULL);
535 error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp);
536 }
537
538 if (error) {
539 ASSERT(bp == NULL);
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000540 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
543
Dave Chinnerc6319192012-11-14 17:50:13 +1100544 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 *O_bpp = bp;
Dave Chinnerc6319192012-11-14 17:50:13 +1100546 *O_ddpp = bp->b_addr + dqp->q_bufoffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
Eric Sandeend99831f2014-06-22 15:03:54 +1000548 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549}
550
551
552/*
553 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
554 * and release the buffer immediately.
555 *
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000556 * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 */
Christoph Hellwig7ae44402011-12-06 21:58:25 +0000558int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559xfs_qm_dqread(
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000560 struct xfs_mount *mp,
561 xfs_dqid_t id,
562 uint type,
563 uint flags,
564 struct xfs_dquot **O_dqpp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000566 struct xfs_dquot *dqp;
567 struct xfs_disk_dquot *ddqp;
568 struct xfs_buf *bp;
569 struct xfs_trans *tp = NULL;
570 int error;
571 int cancelflags = 0;
Tim Shimminefa092f2005-09-05 08:29:01 +1000572
Christoph Hellwig92b2e5b2012-02-01 13:57:20 +0000573
Christoph Hellwiga05931c2012-03-13 08:52:37 +0000574 dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP);
Christoph Hellwig92b2e5b2012-02-01 13:57:20 +0000575
576 dqp->dq_flags = type;
577 dqp->q_core.d_id = cpu_to_be32(id);
578 dqp->q_mount = mp;
Christoph Hellwigf8739c32012-03-13 08:52:34 +0000579 INIT_LIST_HEAD(&dqp->q_lru);
Christoph Hellwig92b2e5b2012-02-01 13:57:20 +0000580 mutex_init(&dqp->q_qlock);
581 init_waitqueue_head(&dqp->q_pinwait);
582
583 /*
584 * Because we want to use a counting completion, complete
585 * the flush completion once to allow a single access to
586 * the flush completion without blocking.
587 */
588 init_completion(&dqp->q_flush);
589 complete(&dqp->q_flush);
590
591 /*
592 * Make sure group quotas have a different lock class than user
593 * quotas.
594 */
Dave Chinnerf112a042013-09-30 09:37:03 +1000595 switch (type) {
596 case XFS_DQ_USER:
597 /* uses the default lock class */
598 break;
599 case XFS_DQ_GROUP:
600 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
601 break;
602 case XFS_DQ_PROJ:
603 lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
604 break;
605 default:
606 ASSERT(0);
607 break;
608 }
Christoph Hellwig92b2e5b2012-02-01 13:57:20 +0000609
Christoph Hellwig48776fd2012-03-13 08:52:33 +0000610 XFS_STATS_INC(xs_qm_dquot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000612 trace_xfs_dqread(dqp);
613
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000614 if (flags & XFS_QMOPT_DQALLOC) {
615 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC);
Brian Foster410b11a2014-02-07 14:55:54 +1100616 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_dqalloc,
Jie Liu3d3c8b52013-08-12 20:49:59 +1000617 XFS_QM_DQALLOC_SPACE_RES(mp), 0);
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000618 if (error)
619 goto error1;
620 cancelflags = XFS_TRANS_RELEASE_LOG_RES;
621 }
622
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 /*
624 * get a pointer to the on-disk dquot and the buffer containing it
625 * dqp already knows its own type (GROUP/USER).
626 */
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000627 error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags);
628 if (error) {
629 /*
630 * This can happen if quotas got turned off (ESRCH),
631 * or if the dquot didn't exist on disk and we ask to
632 * allocate (ENOENT).
633 */
634 trace_xfs_dqread_fail(dqp);
635 cancelflags |= XFS_TRANS_ABORT;
636 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 }
638
639 /* copy everything from disk dquot to the incore dquot */
640 memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 xfs_qm_dquot_logitem_init(dqp);
642
643 /*
644 * Reservation counters are defined as reservation plus current usage
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300645 * to avoid having to add every time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 */
Christoph Hellwig1149d962005-11-02 15:01:12 +1100647 dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
648 dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
649 dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Brian Fosterb1366452013-03-18 10:51:46 -0400651 /* initialize the dquot speculative prealloc thresholds */
652 xfs_dquot_set_prealloc_limits(dqp);
653
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 /* Mark the buf so that this will stay incore a little longer */
Christoph Hellwig38f23232011-10-10 16:52:45 +0000655 xfs_buf_set_ref(bp, XFS_DQUOT_REF);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
657 /*
658 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
659 * So we need to release with xfs_trans_brelse().
660 * The strategy here is identical to that of inodes; we lock
661 * the dquot in xfs_qm_dqget() before making it accessible to
662 * others. This is because dquots, like inodes, need a good level of
663 * concurrency, and we don't want to take locks on the entire buffers
664 * for dquot accesses.
665 * Note also that the dquot buffer may even be dirty at this point, if
666 * this particular dquot was repaired. We still aren't afraid to
667 * brelse it because we have the changes incore.
668 */
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200669 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 xfs_trans_brelse(tp, bp);
671
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 if (tp) {
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000673 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
674 if (error)
675 goto error0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 }
677
678 *O_dqpp = dqp;
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000679 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000681error1:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 if (tp)
683 xfs_trans_cancel(tp, cancelflags);
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000684error0:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 xfs_qm_dqdestroy(dqp);
686 *O_dqpp = NULL;
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000687 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688}
689
690/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
692 * a locked dquot, doing an allocation (if requested) as needed.
693 * When both an inode and an id are given, the inode's id takes precedence.
694 * That is, if the id changes while we don't hold the ilock inside this
695 * function, the new dquot is returned, not necessarily the one requested
696 * in the id argument.
697 */
698int
699xfs_qm_dqget(
700 xfs_mount_t *mp,
701 xfs_inode_t *ip, /* locked inode (optional) */
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000702 xfs_dqid_t id, /* uid/projid/gid depending on type */
703 uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
705 xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */
706{
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000707 struct xfs_quotainfo *qi = mp->m_quotainfo;
Chandra Seetharaman329e0872013-06-27 17:25:05 -0500708 struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000709 struct xfs_dquot *dqp;
710 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
712 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
713 if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000714 (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
Dave Chinner24513372014-06-25 14:58:08 +1000716 return -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
719#ifdef DEBUG
720 if (xfs_do_dqerror) {
721 if ((xfs_dqerror_target == mp->m_ddev_targp) &&
722 (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100723 xfs_debug(mp, "Returning error in dqget");
Dave Chinner24513372014-06-25 14:58:08 +1000724 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 }
726 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000728 ASSERT(type == XFS_DQ_USER ||
729 type == XFS_DQ_PROJ ||
730 type == XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 if (ip) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000732 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Chandra Seetharaman36731412012-01-23 17:31:30 +0000733 ASSERT(xfs_inode_dquot(ip, type) == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 }
735#endif
Christoph Hellwig92678552011-12-06 21:58:18 +0000736
737restart:
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000738 mutex_lock(&qi->qi_tree_lock);
739 dqp = radix_tree_lookup(tree, id);
740 if (dqp) {
741 xfs_dqlock(dqp);
742 if (dqp->dq_flags & XFS_DQ_FREEING) {
743 xfs_dqunlock(dqp);
744 mutex_unlock(&qi->qi_tree_lock);
745 trace_xfs_dqget_freeing(dqp);
746 delay(1);
747 goto restart;
748 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000750 dqp->q_nrefs++;
751 mutex_unlock(&qi->qi_tree_lock);
752
753 trace_xfs_dqget_hit(dqp);
Christoph Hellwig48776fd2012-03-13 08:52:33 +0000754 XFS_STATS_INC(xs_qm_dqcachehits);
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000755 *O_dqpp = dqp;
756 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 }
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000758 mutex_unlock(&qi->qi_tree_lock);
759 XFS_STATS_INC(xs_qm_dqcachemisses);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
761 /*
762 * Dquot cache miss. We don't want to keep the inode lock across
763 * a (potential) disk read. Also we don't want to deal with the lock
764 * ordering between quotainode and this inode. OTOH, dropping the inode
765 * lock here means dealing with a chown that can happen before
766 * we re-acquire the lock.
767 */
768 if (ip)
769 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Christoph Hellwig97e7ade2011-12-06 21:58:24 +0000771 error = xfs_qm_dqread(mp, id, type, flags, &dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
Christoph Hellwig7ae44402011-12-06 21:58:25 +0000773 if (ip)
774 xfs_ilock(ip, XFS_ILOCK_EXCL);
775
776 if (error)
777 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 if (ip) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 /*
781 * A dquot could be attached to this inode by now, since
782 * we had dropped the ilock.
783 */
Chandra Seetharaman36731412012-01-23 17:31:30 +0000784 if (xfs_this_quota_on(mp, type)) {
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000785 struct xfs_dquot *dqp1;
786
Chandra Seetharaman36731412012-01-23 17:31:30 +0000787 dqp1 = xfs_inode_dquot(ip, type);
788 if (dqp1) {
Christoph Hellwig191f8482010-04-20 17:01:53 +1000789 xfs_qm_dqdestroy(dqp);
Chandra Seetharaman36731412012-01-23 17:31:30 +0000790 dqp = dqp1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 xfs_dqlock(dqp);
792 goto dqret;
793 }
794 } else {
Chandra Seetharaman36731412012-01-23 17:31:30 +0000795 /* inode stays locked on return */
796 xfs_qm_dqdestroy(dqp);
Dave Chinner24513372014-06-25 14:58:08 +1000797 return -ESRCH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 }
799 }
800
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000801 mutex_lock(&qi->qi_tree_lock);
Dave Chinner24513372014-06-25 14:58:08 +1000802 error = radix_tree_insert(tree, id, dqp);
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000803 if (unlikely(error)) {
Dave Chinner24513372014-06-25 14:58:08 +1000804 WARN_ON(error != -EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000806 /*
807 * Duplicate found. Just throw away the new dquot and start
808 * over.
809 */
810 mutex_unlock(&qi->qi_tree_lock);
811 trace_xfs_dqget_dup(dqp);
812 xfs_qm_dqdestroy(dqp);
813 XFS_STATS_INC(xs_qm_dquot_dups);
814 goto restart;
815 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
817 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 * We return a locked dquot to the caller, with a reference taken
819 */
820 xfs_dqlock(dqp);
821 dqp->q_nrefs = 1;
822
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000823 qi->qi_dquots++;
Christoph Hellwig9f920f12012-03-13 08:52:35 +0000824 mutex_unlock(&qi->qi_tree_lock);
825
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 dqret:
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000827 ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000828 trace_xfs_dqget_miss(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 *O_dqpp = dqp;
Eric Sandeend99831f2014-06-22 15:03:54 +1000830 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831}
832
Christoph Hellwigf8739c32012-03-13 08:52:34 +0000833/*
834 * Release a reference to the dquot (decrement ref-count) and unlock it.
835 *
836 * If there is a group quota attached to this dquot, carefully release that
837 * too without tripping over deadlocks'n'stuff.
838 */
839void
840xfs_qm_dqput(
841 struct xfs_dquot *dqp)
842{
843 ASSERT(dqp->q_nrefs > 0);
844 ASSERT(XFS_DQ_IS_LOCKED(dqp));
845
846 trace_xfs_dqput(dqp);
847
Dave Chinner3c35337572014-05-05 17:30:15 +1000848 if (--dqp->q_nrefs == 0) {
849 struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo;
850 trace_xfs_dqput_free(dqp);
851
852 if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
853 XFS_STATS_INC(xs_qm_dquot_unused);
854 }
855 xfs_dqunlock(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856}
857
858/*
859 * Release a dquot. Flush it if dirty, then dqput() it.
860 * dquot must not be locked.
861 */
862void
863xfs_qm_dqrele(
864 xfs_dquot_t *dqp)
865{
Christoph Hellwig7d095252009-06-08 15:33:32 +0200866 if (!dqp)
867 return;
868
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000869 trace_xfs_dqrele(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
871 xfs_dqlock(dqp);
872 /*
873 * We don't care to flush it if the dquot is dirty here.
874 * That will create stutters that we want to avoid.
875 * Instead we do a delayed write when we try to reclaim
876 * a dirty dquot. Also xfs_sync will take part of the burden...
877 */
878 xfs_qm_dqput(dqp);
879}
880
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000881/*
882 * This is the dquot flushing I/O completion routine. It is called
883 * from interrupt level when the buffer containing the dquot is
884 * flushed to disk. It is responsible for removing the dquot logitem
885 * from the AIL if it has not been re-logged, and unlocking the dquot's
886 * flush lock. This behavior is very similar to that of inodes..
887 */
888STATIC void
889xfs_qm_dqflush_done(
890 struct xfs_buf *bp,
891 struct xfs_log_item *lip)
892{
893 xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip;
894 xfs_dquot_t *dqp = qip->qli_dquot;
895 struct xfs_ail *ailp = lip->li_ailp;
896
897 /*
898 * We only want to pull the item from the AIL if its
899 * location in the log has not changed since we started the flush.
900 * Thus, we only bother if the dquot's lsn has
901 * not changed. First we check the lsn outside the lock
902 * since it's cheaper, and then we recheck while
903 * holding the lock before removing the dquot from the AIL.
904 */
905 if ((lip->li_flags & XFS_LI_IN_AIL) &&
906 lip->li_lsn == qip->qli_flush_lsn) {
907
908 /* xfs_trans_ail_delete() drops the AIL lock. */
909 spin_lock(&ailp->xa_lock);
910 if (lip->li_lsn == qip->qli_flush_lsn)
Dave Chinner04913fd2012-04-23 15:58:41 +1000911 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000912 else
913 spin_unlock(&ailp->xa_lock);
914 }
915
916 /*
917 * Release the dq's flush lock since we're done with it.
918 */
919 xfs_dqfunlock(dqp);
920}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
922/*
923 * Write a modified dquot to disk.
924 * The dquot must be locked and the flush lock too taken by caller.
925 * The flush lock will not be unlocked until the dquot reaches the disk,
926 * but the dquot is free to be unlocked and modified by the caller
927 * in the interim. Dquot is still locked on return. This behavior is
928 * identical to that of inodes.
929 */
930int
931xfs_qm_dqflush(
Christoph Hellwigfe7257f2012-04-23 15:58:37 +1000932 struct xfs_dquot *dqp,
933 struct xfs_buf **bpp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934{
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000935 struct xfs_mount *mp = dqp->q_mount;
936 struct xfs_buf *bp;
937 struct xfs_disk_dquot *ddqp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
940 ASSERT(XFS_DQ_IS_LOCKED(dqp));
David Chinnere1f49cf2008-08-13 16:41:43 +1000941 ASSERT(!completion_done(&dqp->q_flush));
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000942
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000943 trace_xfs_dqflush(dqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
Christoph Hellwigfe7257f2012-04-23 15:58:37 +1000945 *bpp = NULL;
946
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 xfs_qm_dqunpin_wait(dqp);
948
949 /*
950 * This may have been unpinned because the filesystem is shutting
951 * down forcibly. If that's the case we must not write this dquot
Christoph Hellwigdea96092012-04-23 15:58:31 +1000952 * to disk, because the log record didn't make it to disk.
953 *
954 * We also have to remove the log item from the AIL in this case,
955 * as we wait for an emptry AIL as part of the unmount process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 */
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000957 if (XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwigdea96092012-04-23 15:58:31 +1000958 struct xfs_log_item *lip = &dqp->q_logitem.qli_item;
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000959 dqp->dq_flags &= ~XFS_DQ_DIRTY;
Christoph Hellwigdea96092012-04-23 15:58:31 +1000960
961 spin_lock(&mp->m_ail->xa_lock);
962 if (lip->li_flags & XFS_LI_IN_AIL)
Dave Chinner04913fd2012-04-23 15:58:41 +1000963 xfs_trans_ail_delete(mp->m_ail, lip,
964 SHUTDOWN_CORRUPT_INCORE);
Christoph Hellwigdea96092012-04-23 15:58:31 +1000965 else
966 spin_unlock(&mp->m_ail->xa_lock);
Dave Chinner24513372014-06-25 14:58:08 +1000967 error = -EIO;
Christoph Hellwigfe7257f2012-04-23 15:58:37 +1000968 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 }
970
971 /*
972 * Get the buffer containing the on-disk dquot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 */
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000974 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
Dave Chinner5fd364f2014-08-04 12:43:26 +1000975 mp->m_quotainfo->qi_dqchunklen, 0, &bp,
976 &xfs_dquot_buf_ops);
Christoph Hellwigfe7257f2012-04-23 15:58:37 +1000977 if (error)
978 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000980 /*
981 * Calculate the location of the dquot inside the buffer.
982 */
Chandra Seetharaman62926042011-07-22 23:40:15 +0000983 ddqp = bp->b_addr + dqp->q_bufoffset;
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000984
985 /*
986 * A simple sanity check in case we got a corrupted dquot..
987 */
Dave Chinner9aede1d2013-10-15 09:17:52 +1100988 error = xfs_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0,
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100989 XFS_QMOPT_DOWARN, "dqflush (incore copy)");
990 if (error) {
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000991 xfs_buf_relse(bp);
992 xfs_dqfunlock(dqp);
993 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Dave Chinner24513372014-06-25 14:58:08 +1000994 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 }
996
997 /* This is the only portion of data that needs to persist */
Christoph Hellwigacecf1b2010-09-06 01:44:45 +0000998 memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
1000 /*
1001 * Clear the dirty field and remember the flush lsn for later use.
1002 */
Christoph Hellwigacecf1b2010-09-06 01:44:45 +00001003 dqp->dq_flags &= ~XFS_DQ_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004
David Chinner7b2e2a32008-10-30 17:39:12 +11001005 xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1006 &dqp->q_logitem.qli_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
1008 /*
Christoph Hellwig3fe58f32013-04-03 16:11:16 +11001009 * copy the lsn into the on-disk dquot now while we have the in memory
1010 * dquot here. This can't be done later in the write verifier as we
1011 * can't get access to the log item at that point in time.
Dave Chinner6fcdc592013-06-03 15:28:46 +10001012 *
1013 * We also calculate the CRC here so that the on-disk dquot in the
1014 * buffer always has a valid CRC. This ensures there is no possibility
1015 * of a dquot without an up-to-date CRC getting to disk.
Christoph Hellwig3fe58f32013-04-03 16:11:16 +11001016 */
1017 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1018 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp;
1019
1020 dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
Dave Chinner6fcdc592013-06-03 15:28:46 +10001021 xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk),
1022 XFS_DQUOT_CRC_OFF);
Christoph Hellwig3fe58f32013-04-03 16:11:16 +11001023 }
1024
1025 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 * Attach an iodone routine so that we can remove this dquot from the
1027 * AIL and release the flush lock once the dquot is synced to disk.
1028 */
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001029 xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1030 &dqp->q_logitem.qli_item);
1031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 /*
1033 * If the buffer is pinned then push on the log so we won't
1034 * get stuck waiting in the write for too long.
1035 */
Chandra Seetharaman811e64c2011-07-22 23:40:27 +00001036 if (xfs_buf_ispinned(bp)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001037 trace_xfs_dqflush_force(dqp);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001038 xfs_log_force(mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 }
1040
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001041 trace_xfs_dqflush_done(dqp);
Christoph Hellwigfe7257f2012-04-23 15:58:37 +10001042 *bpp = bp;
1043 return 0;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001044
Christoph Hellwigfe7257f2012-04-23 15:58:37 +10001045out_unlock:
1046 xfs_dqfunlock(dqp);
Dave Chinner24513372014-06-25 14:58:08 +10001047 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048}
1049
Christoph Hellwig5bb87a32009-01-19 02:03:19 +01001050/*
1051 * Lock two xfs_dquot structures.
1052 *
1053 * To avoid deadlocks we always lock the quota structure with
1054 * the lowerd id first.
1055 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056void
1057xfs_dqlock2(
1058 xfs_dquot_t *d1,
1059 xfs_dquot_t *d2)
1060{
1061 if (d1 && d2) {
1062 ASSERT(d1 != d2);
Christoph Hellwig1149d962005-11-02 15:01:12 +11001063 if (be32_to_cpu(d1->q_core.d_id) >
1064 be32_to_cpu(d2->q_core.d_id)) {
Christoph Hellwig5bb87a32009-01-19 02:03:19 +01001065 mutex_lock(&d2->q_qlock);
1066 mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 } else {
Christoph Hellwig5bb87a32009-01-19 02:03:19 +01001068 mutex_lock(&d1->q_qlock);
1069 mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 }
Christoph Hellwig5bb87a32009-01-19 02:03:19 +01001071 } else if (d1) {
1072 mutex_lock(&d1->q_qlock);
1073 } else if (d2) {
1074 mutex_lock(&d2->q_qlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 }
1076}
1077
Christoph Hellwiga05931c2012-03-13 08:52:37 +00001078int __init
1079xfs_qm_init(void)
1080{
1081 xfs_qm_dqzone =
1082 kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
1083 if (!xfs_qm_dqzone)
1084 goto out;
1085
1086 xfs_qm_dqtrxzone =
1087 kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
1088 if (!xfs_qm_dqtrxzone)
1089 goto out_free_dqzone;
1090
1091 return 0;
1092
1093out_free_dqzone:
1094 kmem_zone_destroy(xfs_qm_dqzone);
1095out:
1096 return -ENOMEM;
1097}
1098
Gerard Snitselaar1c2ccc62012-03-16 18:36:18 +00001099void
Christoph Hellwiga05931c2012-03-13 08:52:37 +00001100xfs_qm_exit(void)
1101{
1102 kmem_zone_destroy(xfs_qm_dqtrxzone);
1103 kmem_zone_destroy(xfs_qm_dqzone);
1104}