Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
4ce3121f NS |
2 | * Copyright (c) 2000-2003 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
4ce3121f NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
4ce3121f NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
4ce3121f NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 LT |
18 | #include "xfs.h" |
19 | #include "xfs_fs.h" | |
6ca1c906 | 20 | #include "xfs_format.h" |
239880ef | 21 | #include "xfs_log_format.h" |
70a9883c | 22 | #include "xfs_shared.h" |
239880ef | 23 | #include "xfs_trans_resv.h" |
a844f451 | 24 | #include "xfs_bit.h" |
1da177e4 | 25 | #include "xfs_mount.h" |
3ab78df2 | 26 | #include "xfs_defer.h" |
1da177e4 LT |
27 | #include "xfs_inode.h" |
28 | #include "xfs_bmap.h" | |
68988114 | 29 | #include "xfs_bmap_util.h" |
239880ef DC |
30 | #include "xfs_alloc.h" |
31 | #include "xfs_quota.h" | |
1da177e4 | 32 | #include "xfs_error.h" |
239880ef | 33 | #include "xfs_trans.h" |
1da177e4 LT |
34 | #include "xfs_buf_item.h" |
35 | #include "xfs_trans_space.h" | |
36 | #include "xfs_trans_priv.h" | |
1da177e4 | 37 | #include "xfs_qm.h" |
3fe58f30 | 38 | #include "xfs_cksum.h" |
0b1b213f | 39 | #include "xfs_trace.h" |
239880ef | 40 | #include "xfs_log.h" |
a4fbe6ab | 41 | #include "xfs_bmap_btree.h" |
1da177e4 | 42 | |
1da177e4 | 43 | /* |
bf72de31 CH |
44 | * Lock order: |
45 | * | |
46 | * ip->i_lock | |
9f920f11 | 47 | * qi->qi_tree_lock |
b84a3a96 CH |
48 | * dquot->q_qlock (xfs_dqlock() and friends) |
49 | * dquot->q_flush (xfs_dqflock() and friends) | |
50 | * qi->qi_lru_lock | |
bf72de31 CH |
51 | * |
52 | * If two dquots need to be locked the order is user before group/project, | |
53 | * otherwise by the lowest id first, see xfs_dqlock2. | |
54 | */ | |
1da177e4 | 55 | |
1da177e4 LT |
56 | #ifdef DEBUG |
57 | xfs_buftarg_t *xfs_dqerror_target; | |
58 | int xfs_do_dqerror; | |
59 | int xfs_dqreq_num; | |
60 | int xfs_dqerror_mod = 33; | |
61 | #endif | |
62 | ||
a05931ce CH |
63 | struct kmem_zone *xfs_qm_dqtrxzone; |
64 | static struct kmem_zone *xfs_qm_dqzone; | |
65 | ||
f112a049 DC |
66 | static struct lock_class_key xfs_dquot_group_class; |
67 | static struct lock_class_key xfs_dquot_project_class; | |
98b8c7a0 | 68 | |
1da177e4 LT |
69 | /* |
70 | * This is called to free all the memory associated with a dquot | |
71 | */ | |
72 | void | |
73 | xfs_qm_dqdestroy( | |
74 | xfs_dquot_t *dqp) | |
75 | { | |
f8739c3c | 76 | ASSERT(list_empty(&dqp->q_lru)); |
1da177e4 | 77 | |
b1c5ebb2 | 78 | kmem_free(dqp->q_logitem.qli_item.li_lv_shadow); |
1da177e4 | 79 | mutex_destroy(&dqp->q_qlock); |
0b1b213f | 80 | |
ff6d6af2 BD |
81 | XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot); |
82 | kmem_zone_free(xfs_qm_dqzone, dqp); | |
1da177e4 LT |
83 | } |
84 | ||
1da177e4 LT |
85 | /* |
86 | * 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 | */ | |
90 | void | |
91 | xfs_qm_adjust_dqlimits( | |
4b6eae2e BF |
92 | struct xfs_mount *mp, |
93 | struct xfs_dquot *dq) | |
1da177e4 | 94 | { |
4b6eae2e BF |
95 | struct xfs_quotainfo *q = mp->m_quotainfo; |
96 | struct xfs_disk_dquot *d = &dq->q_core; | |
be607946 | 97 | struct xfs_def_quota *defq; |
b1366451 | 98 | int prealloc = 0; |
1da177e4 LT |
99 | |
100 | ASSERT(d->d_id); | |
be607946 | 101 | defq = xfs_get_defquota(dq, q); |
1da177e4 | 102 | |
be607946 CM |
103 | if (defq->bsoftlimit && !d->d_blk_softlimit) { |
104 | d->d_blk_softlimit = cpu_to_be64(defq->bsoftlimit); | |
b1366451 BF |
105 | prealloc = 1; |
106 | } | |
be607946 CM |
107 | if (defq->bhardlimit && !d->d_blk_hardlimit) { |
108 | d->d_blk_hardlimit = cpu_to_be64(defq->bhardlimit); | |
b1366451 BF |
109 | prealloc = 1; |
110 | } | |
be607946 CM |
111 | if (defq->isoftlimit && !d->d_ino_softlimit) |
112 | d->d_ino_softlimit = cpu_to_be64(defq->isoftlimit); | |
113 | if (defq->ihardlimit && !d->d_ino_hardlimit) | |
114 | d->d_ino_hardlimit = cpu_to_be64(defq->ihardlimit); | |
115 | if (defq->rtbsoftlimit && !d->d_rtb_softlimit) | |
116 | d->d_rtb_softlimit = cpu_to_be64(defq->rtbsoftlimit); | |
117 | if (defq->rtbhardlimit && !d->d_rtb_hardlimit) | |
118 | d->d_rtb_hardlimit = cpu_to_be64(defq->rtbhardlimit); | |
b1366451 BF |
119 | |
120 | if (prealloc) | |
121 | xfs_dquot_set_prealloc_limits(dq); | |
1da177e4 LT |
122 | } |
123 | ||
124 | /* | |
125 | * Check the limits and timers of a dquot and start or reset timers | |
126 | * if necessary. | |
127 | * This gets called even when quota enforcement is OFF, which makes our | |
128 | * life a little less complicated. (We just don't reject any quota | |
129 | * reservations in that case, when enforcement is off). | |
130 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
131 | * enforcement's off. | |
132 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
133 | * 'automatically' get started when limits get exceeded. They do |
134 | * get reset to zero, however, when we find the count to be under | |
135 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
136 | */ |
137 | void | |
138 | xfs_qm_adjust_dqtimers( | |
139 | xfs_mount_t *mp, | |
140 | xfs_disk_dquot_t *d) | |
141 | { | |
142 | ASSERT(d->d_id); | |
143 | ||
ea15ab3c | 144 | #ifdef DEBUG |
1149d96a CH |
145 | if (d->d_blk_hardlimit) |
146 | ASSERT(be64_to_cpu(d->d_blk_softlimit) <= | |
147 | be64_to_cpu(d->d_blk_hardlimit)); | |
148 | if (d->d_ino_hardlimit) | |
149 | ASSERT(be64_to_cpu(d->d_ino_softlimit) <= | |
150 | be64_to_cpu(d->d_ino_hardlimit)); | |
151 | if (d->d_rtb_hardlimit) | |
152 | ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= | |
153 | be64_to_cpu(d->d_rtb_hardlimit)); | |
1da177e4 | 154 | #endif |
ea15ab3c | 155 | |
1da177e4 | 156 | if (!d->d_btimer) { |
1149d96a | 157 | if ((d->d_blk_softlimit && |
d0a3fe67 | 158 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
159 | be64_to_cpu(d->d_blk_softlimit))) || |
160 | (d->d_blk_hardlimit && | |
d0a3fe67 | 161 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
162 | be64_to_cpu(d->d_blk_hardlimit)))) { |
163 | d->d_btimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 164 | mp->m_quotainfo->qi_btimelimit); |
754002b4 NS |
165 | } else { |
166 | d->d_bwarns = 0; | |
1da177e4 LT |
167 | } |
168 | } else { | |
169 | if ((!d->d_blk_softlimit || | |
d0a3fe67 | 170 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 171 | be64_to_cpu(d->d_blk_softlimit))) && |
1da177e4 | 172 | (!d->d_blk_hardlimit || |
d0a3fe67 | 173 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 174 | be64_to_cpu(d->d_blk_hardlimit)))) { |
1da177e4 LT |
175 | d->d_btimer = 0; |
176 | } | |
177 | } | |
178 | ||
179 | if (!d->d_itimer) { | |
1149d96a | 180 | if ((d->d_ino_softlimit && |
d0a3fe67 | 181 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
182 | be64_to_cpu(d->d_ino_softlimit))) || |
183 | (d->d_ino_hardlimit && | |
d0a3fe67 | 184 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
185 | be64_to_cpu(d->d_ino_hardlimit)))) { |
186 | d->d_itimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 187 | mp->m_quotainfo->qi_itimelimit); |
754002b4 NS |
188 | } else { |
189 | d->d_iwarns = 0; | |
1da177e4 LT |
190 | } |
191 | } else { | |
192 | if ((!d->d_ino_softlimit || | |
d0a3fe67 | 193 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 194 | be64_to_cpu(d->d_ino_softlimit))) && |
1da177e4 | 195 | (!d->d_ino_hardlimit || |
d0a3fe67 | 196 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 197 | be64_to_cpu(d->d_ino_hardlimit)))) { |
1da177e4 LT |
198 | d->d_itimer = 0; |
199 | } | |
200 | } | |
201 | ||
202 | if (!d->d_rtbtimer) { | |
1149d96a | 203 | if ((d->d_rtb_softlimit && |
d0a3fe67 | 204 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
205 | be64_to_cpu(d->d_rtb_softlimit))) || |
206 | (d->d_rtb_hardlimit && | |
d0a3fe67 | 207 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
208 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
209 | d->d_rtbtimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 210 | mp->m_quotainfo->qi_rtbtimelimit); |
754002b4 NS |
211 | } else { |
212 | d->d_rtbwarns = 0; | |
1da177e4 LT |
213 | } |
214 | } else { | |
215 | if ((!d->d_rtb_softlimit || | |
d0a3fe67 | 216 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 217 | be64_to_cpu(d->d_rtb_softlimit))) && |
1da177e4 | 218 | (!d->d_rtb_hardlimit || |
d0a3fe67 | 219 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 220 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
1da177e4 LT |
221 | d->d_rtbtimer = 0; |
222 | } | |
223 | } | |
224 | } | |
225 | ||
1da177e4 LT |
226 | /* |
227 | * initialize a buffer full of dquots and log the whole thing | |
228 | */ | |
229 | STATIC void | |
230 | xfs_qm_init_dquot_blk( | |
231 | xfs_trans_t *tp, | |
232 | xfs_mount_t *mp, | |
233 | xfs_dqid_t id, | |
234 | uint type, | |
235 | xfs_buf_t *bp) | |
236 | { | |
8a7b8a89 | 237 | struct xfs_quotainfo *q = mp->m_quotainfo; |
1da177e4 | 238 | xfs_dqblk_t *d; |
a484bcdd ES |
239 | xfs_dqid_t curid; |
240 | int i; | |
1da177e4 LT |
241 | |
242 | ASSERT(tp); | |
0c842ad4 | 243 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 244 | |
62926044 | 245 | d = bp->b_addr; |
1da177e4 LT |
246 | |
247 | /* | |
248 | * ID of the first dquot in the block - id's are zero based. | |
249 | */ | |
8a7b8a89 | 250 | curid = id - (id % q->qi_dqperchunk); |
8a7b8a89 | 251 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
49d35a5c CH |
252 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) { |
253 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
254 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
255 | d->dd_diskdq.d_id = cpu_to_be32(curid); | |
256 | d->dd_diskdq.d_flags = type; | |
6fcdc59d | 257 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
92863451 | 258 | uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid); |
6fcdc59d DC |
259 | xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk), |
260 | XFS_DQUOT_CRC_OFF); | |
261 | } | |
49d35a5c CH |
262 | } |
263 | ||
1da177e4 | 264 | xfs_trans_dquot_buf(tp, bp, |
c1155410 DC |
265 | (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : |
266 | ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : | |
267 | XFS_BLF_GDQUOT_BUF))); | |
8a7b8a89 | 268 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); |
1da177e4 LT |
269 | } |
270 | ||
b1366451 BF |
271 | /* |
272 | * Initialize the dynamic speculative preallocation thresholds. The lo/hi | |
273 | * watermarks correspond to the soft and hard limits by default. If a soft limit | |
274 | * is not specified, we use 95% of the hard limit. | |
275 | */ | |
276 | void | |
277 | xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp) | |
278 | { | |
279 | __uint64_t space; | |
280 | ||
281 | dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit); | |
282 | dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit); | |
283 | if (!dqp->q_prealloc_lo_wmark) { | |
284 | dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark; | |
285 | do_div(dqp->q_prealloc_lo_wmark, 100); | |
286 | dqp->q_prealloc_lo_wmark *= 95; | |
287 | } | |
288 | ||
289 | space = dqp->q_prealloc_hi_wmark; | |
290 | ||
291 | do_div(space, 100); | |
292 | dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space; | |
293 | dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3; | |
294 | dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5; | |
295 | } | |
296 | ||
1da177e4 LT |
297 | /* |
298 | * Allocate a block and fill it with dquots. | |
299 | * This is called when the bmapi finds a hole. | |
300 | */ | |
301 | STATIC int | |
302 | xfs_qm_dqalloc( | |
efa092f3 | 303 | xfs_trans_t **tpp, |
1da177e4 LT |
304 | xfs_mount_t *mp, |
305 | xfs_dquot_t *dqp, | |
306 | xfs_inode_t *quotip, | |
307 | xfs_fileoff_t offset_fsb, | |
308 | xfs_buf_t **O_bpp) | |
309 | { | |
310 | xfs_fsblock_t firstblock; | |
2c3234d1 | 311 | struct xfs_defer_ops dfops; |
1da177e4 | 312 | xfs_bmbt_irec_t map; |
f6106efa | 313 | int nmaps, error; |
1da177e4 | 314 | xfs_buf_t *bp; |
efa092f3 | 315 | xfs_trans_t *tp = *tpp; |
1da177e4 LT |
316 | |
317 | ASSERT(tp != NULL); | |
0b1b213f CH |
318 | |
319 | trace_xfs_dqalloc(dqp); | |
1da177e4 LT |
320 | |
321 | /* | |
322 | * Initialize the bmap freelist prior to calling bmapi code. | |
323 | */ | |
2c3234d1 | 324 | xfs_defer_init(&dfops, &firstblock); |
1da177e4 LT |
325 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
326 | /* | |
327 | * Return if this type of quotas is turned off while we didn't | |
328 | * have an inode lock | |
329 | */ | |
6967b964 | 330 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 | 331 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
2451337d | 332 | return -ESRCH; |
1da177e4 LT |
333 | } |
334 | ||
ddc3415a | 335 | xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); |
1da177e4 | 336 | nmaps = 1; |
c0dc7828 DC |
337 | error = xfs_bmapi_write(tp, quotip, offset_fsb, |
338 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
339 | &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), | |
2c3234d1 | 340 | &map, &nmaps, &dfops); |
c0dc7828 | 341 | if (error) |
1da177e4 | 342 | goto error0; |
1da177e4 LT |
343 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
344 | ASSERT(nmaps == 1); | |
345 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
346 | (map.br_startblock != HOLESTARTBLOCK)); | |
347 | ||
348 | /* | |
349 | * Keep track of the blkno to save a lookup later | |
350 | */ | |
351 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
352 | ||
353 | /* now we can just get the buffer (there's nothing to read yet) */ | |
354 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
355 | dqp->q_blkno, | |
8a7b8a89 | 356 | mp->m_quotainfo->qi_dqchunklen, |
1da177e4 | 357 | 0); |
36de9556 | 358 | if (!bp) { |
2451337d | 359 | error = -ENOMEM; |
1da177e4 | 360 | goto error1; |
36de9556 | 361 | } |
1813dd64 | 362 | bp->b_ops = &xfs_dquot_buf_ops; |
2a30f36d | 363 | |
1da177e4 LT |
364 | /* |
365 | * Make a chunk of dquots out of this buffer and log | |
366 | * the entire thing. | |
367 | */ | |
1149d96a | 368 | xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 369 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
1da177e4 | 370 | |
efa092f3 | 371 | /* |
310a75a3 | 372 | * xfs_defer_finish() may commit the current transaction and |
efa092f3 TS |
373 | * start a second transaction if the freelist is not empty. |
374 | * | |
375 | * Since we still want to modify this buffer, we need to | |
376 | * ensure that the buffer is not released on commit of | |
377 | * the first transaction and ensure the buffer is added to the | |
378 | * second transaction. | |
379 | * | |
380 | * If there is only one transaction then don't stop the buffer | |
381 | * from being released when it commits later on. | |
382 | */ | |
383 | ||
384 | xfs_trans_bhold(tp, bp); | |
385 | ||
2c3234d1 | 386 | error = xfs_defer_finish(tpp, &dfops, NULL); |
f6106efa | 387 | if (error) |
1da177e4 | 388 | goto error1; |
1da177e4 | 389 | |
f6106efa ES |
390 | /* Transaction was committed? */ |
391 | if (*tpp != tp) { | |
efa092f3 TS |
392 | tp = *tpp; |
393 | xfs_trans_bjoin(tp, bp); | |
394 | } else { | |
395 | xfs_trans_bhold_release(tp, bp); | |
396 | } | |
397 | ||
1da177e4 LT |
398 | *O_bpp = bp; |
399 | return 0; | |
400 | ||
f6106efa | 401 | error1: |
2c3234d1 | 402 | xfs_defer_cancel(&dfops); |
f6106efa | 403 | error0: |
1da177e4 LT |
404 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
405 | ||
d99831ff | 406 | return error; |
1da177e4 | 407 | } |
9aede1d8 | 408 | |
c6319198 DC |
409 | STATIC int |
410 | xfs_qm_dqrepair( | |
411 | struct xfs_mount *mp, | |
412 | struct xfs_trans *tp, | |
413 | struct xfs_dquot *dqp, | |
414 | xfs_dqid_t firstid, | |
415 | struct xfs_buf **bpp) | |
416 | { | |
417 | int error; | |
418 | struct xfs_disk_dquot *ddq; | |
419 | struct xfs_dqblk *d; | |
420 | int i; | |
421 | ||
422 | /* | |
423 | * Read the buffer without verification so we get the corrupted | |
612cfbfe | 424 | * buffer returned to us. make sure we verify it on write, though. |
c6319198 DC |
425 | */ |
426 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno, | |
427 | mp->m_quotainfo->qi_dqchunklen, | |
428 | 0, bpp, NULL); | |
429 | ||
430 | if (error) { | |
431 | ASSERT(*bpp == NULL); | |
b474c7ae | 432 | return error; |
c6319198 | 433 | } |
1813dd64 | 434 | (*bpp)->b_ops = &xfs_dquot_buf_ops; |
c6319198 DC |
435 | |
436 | ASSERT(xfs_buf_islocked(*bpp)); | |
437 | d = (struct xfs_dqblk *)(*bpp)->b_addr; | |
438 | ||
439 | /* Do the actual repair of dquots in this buffer */ | |
440 | for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) { | |
441 | ddq = &d[i].dd_diskdq; | |
9aede1d8 | 442 | error = xfs_dqcheck(mp, ddq, firstid + i, |
c6319198 DC |
443 | dqp->dq_flags & XFS_DQ_ALLTYPES, |
444 | XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair"); | |
445 | if (error) { | |
446 | /* repair failed, we're screwed */ | |
447 | xfs_trans_brelse(tp, *bpp); | |
2451337d | 448 | return -EIO; |
c6319198 DC |
449 | } |
450 | } | |
451 | ||
452 | return 0; | |
453 | } | |
454 | ||
1da177e4 LT |
455 | /* |
456 | * Maps a dquot to the buffer containing its on-disk version. | |
457 | * This returns a ptr to the buffer containing the on-disk dquot | |
458 | * in the bpp param, and a ptr to the on-disk dquot within that buffer | |
459 | */ | |
460 | STATIC int | |
461 | xfs_qm_dqtobp( | |
efa092f3 | 462 | xfs_trans_t **tpp, |
1da177e4 LT |
463 | xfs_dquot_t *dqp, |
464 | xfs_disk_dquot_t **O_ddpp, | |
465 | xfs_buf_t **O_bpp, | |
466 | uint flags) | |
467 | { | |
113a5683 CS |
468 | struct xfs_bmbt_irec map; |
469 | int nmaps = 1, error; | |
470 | struct xfs_buf *bp; | |
4d4d9523 | 471 | struct xfs_inode *quotip; |
113a5683 CS |
472 | struct xfs_mount *mp = dqp->q_mount; |
473 | xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); | |
474 | struct xfs_trans *tp = (tpp ? *tpp : NULL); | |
f4df8adc | 475 | uint lock_mode; |
1da177e4 | 476 | |
4d4d9523 | 477 | quotip = xfs_quota_inode(dqp->q_mount, dqp->dq_flags); |
acecf1b5 | 478 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
1da177e4 | 479 | |
f4df8adc | 480 | lock_mode = xfs_ilock_data_map_shared(quotip); |
6967b964 | 481 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 | 482 | /* |
acecf1b5 CH |
483 | * Return if this type of quotas is turned off while we |
484 | * didn't have the quota inode lock. | |
1da177e4 | 485 | */ |
f4df8adc | 486 | xfs_iunlock(quotip, lock_mode); |
2451337d | 487 | return -ESRCH; |
acecf1b5 CH |
488 | } |
489 | ||
490 | /* | |
491 | * Find the block map; no allocations yet | |
492 | */ | |
5c8ed202 DC |
493 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
494 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); | |
acecf1b5 | 495 | |
f4df8adc | 496 | xfs_iunlock(quotip, lock_mode); |
acecf1b5 CH |
497 | if (error) |
498 | return error; | |
499 | ||
500 | ASSERT(nmaps == 1); | |
501 | ASSERT(map.br_blockcount == 1); | |
502 | ||
503 | /* | |
504 | * Offset of dquot in the (fixed sized) dquot chunk. | |
505 | */ | |
506 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
507 | sizeof(xfs_dqblk_t); | |
508 | ||
509 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
510 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 511 | /* |
acecf1b5 | 512 | * We don't allocate unless we're asked to |
1da177e4 | 513 | */ |
acecf1b5 | 514 | if (!(flags & XFS_QMOPT_DQALLOC)) |
2451337d | 515 | return -ENOENT; |
1da177e4 | 516 | |
acecf1b5 CH |
517 | ASSERT(tp); |
518 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
519 | dqp->q_fileoffset, &bp); | |
1da177e4 | 520 | if (error) |
acecf1b5 CH |
521 | return error; |
522 | tp = *tpp; | |
523 | } else { | |
524 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
525 | |
526 | /* | |
acecf1b5 CH |
527 | * store the blkno etc so that we don't have to do the |
528 | * mapping all the time | |
1da177e4 | 529 | */ |
acecf1b5 | 530 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 531 | |
8a7b8a89 CH |
532 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
533 | dqp->q_blkno, | |
534 | mp->m_quotainfo->qi_dqchunklen, | |
1813dd64 | 535 | 0, &bp, &xfs_dquot_buf_ops); |
acecf1b5 | 536 | |
2451337d | 537 | if (error == -EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) { |
c6319198 DC |
538 | xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff * |
539 | mp->m_quotainfo->qi_dqperchunk; | |
540 | ASSERT(bp == NULL); | |
541 | error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp); | |
542 | } | |
1da177e4 | 543 | |
c6319198 DC |
544 | if (error) { |
545 | ASSERT(bp == NULL); | |
b474c7ae | 546 | return error; |
1da177e4 | 547 | } |
1da177e4 LT |
548 | } |
549 | ||
c6319198 | 550 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 551 | *O_bpp = bp; |
c6319198 | 552 | *O_ddpp = bp->b_addr + dqp->q_bufoffset; |
1da177e4 | 553 | |
d99831ff | 554 | return 0; |
1da177e4 LT |
555 | } |
556 | ||
557 | ||
558 | /* | |
559 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
560 | * and release the buffer immediately. | |
561 | * | |
97e7ade5 | 562 | * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed. |
1da177e4 | 563 | */ |
7ae44407 | 564 | int |
1da177e4 | 565 | xfs_qm_dqread( |
97e7ade5 CH |
566 | struct xfs_mount *mp, |
567 | xfs_dqid_t id, | |
568 | uint type, | |
569 | uint flags, | |
570 | struct xfs_dquot **O_dqpp) | |
1da177e4 | 571 | { |
97e7ade5 CH |
572 | struct xfs_dquot *dqp; |
573 | struct xfs_disk_dquot *ddqp; | |
574 | struct xfs_buf *bp; | |
575 | struct xfs_trans *tp = NULL; | |
576 | int error; | |
92b2e5b3 | 577 | |
a05931ce | 578 | dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP); |
92b2e5b3 CH |
579 | |
580 | dqp->dq_flags = type; | |
581 | dqp->q_core.d_id = cpu_to_be32(id); | |
582 | dqp->q_mount = mp; | |
f8739c3c | 583 | INIT_LIST_HEAD(&dqp->q_lru); |
92b2e5b3 CH |
584 | mutex_init(&dqp->q_qlock); |
585 | init_waitqueue_head(&dqp->q_pinwait); | |
586 | ||
587 | /* | |
588 | * Because we want to use a counting completion, complete | |
589 | * the flush completion once to allow a single access to | |
590 | * the flush completion without blocking. | |
591 | */ | |
592 | init_completion(&dqp->q_flush); | |
593 | complete(&dqp->q_flush); | |
594 | ||
595 | /* | |
596 | * Make sure group quotas have a different lock class than user | |
597 | * quotas. | |
598 | */ | |
f112a049 DC |
599 | switch (type) { |
600 | case XFS_DQ_USER: | |
601 | /* uses the default lock class */ | |
602 | break; | |
603 | case XFS_DQ_GROUP: | |
604 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class); | |
605 | break; | |
606 | case XFS_DQ_PROJ: | |
607 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class); | |
608 | break; | |
609 | default: | |
610 | ASSERT(0); | |
611 | break; | |
612 | } | |
92b2e5b3 | 613 | |
ff6d6af2 | 614 | XFS_STATS_INC(mp, xs_qm_dquot); |
1da177e4 | 615 | |
0b1b213f CH |
616 | trace_xfs_dqread(dqp); |
617 | ||
97e7ade5 | 618 | if (flags & XFS_QMOPT_DQALLOC) { |
253f4911 CH |
619 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc, |
620 | XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp); | |
97e7ade5 | 621 | if (error) |
253f4911 | 622 | goto error0; |
97e7ade5 CH |
623 | } |
624 | ||
1da177e4 LT |
625 | /* |
626 | * get a pointer to the on-disk dquot and the buffer containing it | |
627 | * dqp already knows its own type (GROUP/USER). | |
628 | */ | |
97e7ade5 CH |
629 | 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); | |
97e7ade5 | 637 | goto error1; |
1da177e4 LT |
638 | } |
639 | ||
640 | /* copy everything from disk dquot to the incore dquot */ | |
641 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
1da177e4 LT |
642 | xfs_qm_dquot_logitem_init(dqp); |
643 | ||
644 | /* | |
645 | * Reservation counters are defined as reservation plus current usage | |
25985edc | 646 | * to avoid having to add every time. |
1da177e4 | 647 | */ |
1149d96a CH |
648 | dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); |
649 | dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); | |
650 | dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); | |
1da177e4 | 651 | |
b1366451 BF |
652 | /* initialize the dquot speculative prealloc thresholds */ |
653 | xfs_dquot_set_prealloc_limits(dqp); | |
654 | ||
1da177e4 | 655 | /* Mark the buf so that this will stay incore a little longer */ |
38f23232 | 656 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 LT |
657 | |
658 | /* | |
659 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
660 | * So we need to release with xfs_trans_brelse(). | |
661 | * The strategy here is identical to that of inodes; we lock | |
662 | * the dquot in xfs_qm_dqget() before making it accessible to | |
663 | * others. This is because dquots, like inodes, need a good level of | |
664 | * concurrency, and we don't want to take locks on the entire buffers | |
665 | * for dquot accesses. | |
666 | * Note also that the dquot buffer may even be dirty at this point, if | |
667 | * this particular dquot was repaired. We still aren't afraid to | |
668 | * brelse it because we have the changes incore. | |
669 | */ | |
0c842ad4 | 670 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
671 | xfs_trans_brelse(tp, bp); |
672 | ||
1da177e4 | 673 | if (tp) { |
70393313 | 674 | error = xfs_trans_commit(tp); |
97e7ade5 CH |
675 | if (error) |
676 | goto error0; | |
1da177e4 LT |
677 | } |
678 | ||
679 | *O_dqpp = dqp; | |
97e7ade5 | 680 | return error; |
1da177e4 | 681 | |
97e7ade5 | 682 | error1: |
1da177e4 | 683 | if (tp) |
4906e215 | 684 | xfs_trans_cancel(tp); |
97e7ade5 | 685 | error0: |
1da177e4 LT |
686 | xfs_qm_dqdestroy(dqp); |
687 | *O_dqpp = NULL; | |
97e7ade5 | 688 | return error; |
1da177e4 LT |
689 | } |
690 | ||
296c24e2 ES |
691 | /* |
692 | * Advance to the next id in the current chunk, or if at the | |
693 | * end of the chunk, skip ahead to first id in next allocated chunk | |
694 | * using the SEEK_DATA interface. | |
695 | */ | |
6e3e6d55 | 696 | static int |
296c24e2 ES |
697 | xfs_dq_get_next_id( |
698 | xfs_mount_t *mp, | |
699 | uint type, | |
700 | xfs_dqid_t *id, | |
701 | loff_t eof) | |
702 | { | |
703 | struct xfs_inode *quotip; | |
704 | xfs_fsblock_t start; | |
705 | loff_t offset; | |
706 | uint lock; | |
707 | xfs_dqid_t next_id; | |
708 | int error = 0; | |
709 | ||
710 | /* Simple advance */ | |
711 | next_id = *id + 1; | |
712 | ||
713 | /* If new ID is within the current chunk, advancing it sufficed */ | |
714 | if (next_id % mp->m_quotainfo->qi_dqperchunk) { | |
715 | *id = next_id; | |
716 | return 0; | |
717 | } | |
718 | ||
719 | /* Nope, next_id is now past the current chunk, so find the next one */ | |
720 | start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk; | |
721 | ||
722 | quotip = xfs_quota_inode(mp, type); | |
723 | lock = xfs_ilock_data_map_shared(quotip); | |
724 | ||
725 | offset = __xfs_seek_hole_data(VFS_I(quotip), XFS_FSB_TO_B(mp, start), | |
726 | eof, SEEK_DATA); | |
727 | if (offset < 0) | |
728 | error = offset; | |
729 | ||
730 | xfs_iunlock(quotip, lock); | |
731 | ||
732 | /* -ENXIO is essentially "no more data" */ | |
733 | if (error) | |
734 | return (error == -ENXIO ? -ENOENT: error); | |
735 | ||
736 | /* Convert next data offset back to a quota id */ | |
737 | *id = XFS_B_TO_FSB(mp, offset) * mp->m_quotainfo->qi_dqperchunk; | |
738 | return 0; | |
739 | } | |
740 | ||
1da177e4 LT |
741 | /* |
742 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
743 | * a locked dquot, doing an allocation (if requested) as needed. | |
744 | * When both an inode and an id are given, the inode's id takes precedence. | |
745 | * That is, if the id changes while we don't hold the ilock inside this | |
746 | * function, the new dquot is returned, not necessarily the one requested | |
747 | * in the id argument. | |
748 | */ | |
749 | int | |
750 | xfs_qm_dqget( | |
751 | xfs_mount_t *mp, | |
752 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
753 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
754 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
755 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
756 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
757 | { | |
9f920f11 | 758 | struct xfs_quotainfo *qi = mp->m_quotainfo; |
329e0875 | 759 | struct radix_tree_root *tree = xfs_dquot_tree(qi, type); |
9f920f11 | 760 | struct xfs_dquot *dqp; |
296c24e2 | 761 | loff_t eof = 0; |
9f920f11 | 762 | int error; |
1da177e4 LT |
763 | |
764 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
765 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 766 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 | 767 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
2451337d | 768 | return -ESRCH; |
1da177e4 | 769 | } |
1da177e4 LT |
770 | |
771 | #ifdef DEBUG | |
772 | if (xfs_do_dqerror) { | |
773 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
774 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 775 | xfs_debug(mp, "Returning error in dqget"); |
2451337d | 776 | return -EIO; |
1da177e4 LT |
777 | } |
778 | } | |
1da177e4 | 779 | |
c8ad20ff NS |
780 | ASSERT(type == XFS_DQ_USER || |
781 | type == XFS_DQ_PROJ || | |
782 | type == XFS_DQ_GROUP); | |
1da177e4 | 783 | if (ip) { |
579aa9ca | 784 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
36731410 | 785 | ASSERT(xfs_inode_dquot(ip, type) == NULL); |
1da177e4 LT |
786 | } |
787 | #endif | |
92678554 | 788 | |
296c24e2 ES |
789 | /* Get the end of the quota file if we need it */ |
790 | if (flags & XFS_QMOPT_DQNEXT) { | |
791 | struct xfs_inode *quotip; | |
792 | xfs_fileoff_t last; | |
793 | uint lock_mode; | |
794 | ||
795 | quotip = xfs_quota_inode(mp, type); | |
796 | lock_mode = xfs_ilock_data_map_shared(quotip); | |
797 | error = xfs_bmap_last_offset(quotip, &last, XFS_DATA_FORK); | |
798 | xfs_iunlock(quotip, lock_mode); | |
799 | if (error) | |
800 | return error; | |
801 | eof = XFS_FSB_TO_B(mp, last); | |
802 | } | |
803 | ||
92678554 | 804 | restart: |
9f920f11 CH |
805 | mutex_lock(&qi->qi_tree_lock); |
806 | dqp = radix_tree_lookup(tree, id); | |
807 | if (dqp) { | |
808 | xfs_dqlock(dqp); | |
809 | if (dqp->dq_flags & XFS_DQ_FREEING) { | |
810 | xfs_dqunlock(dqp); | |
811 | mutex_unlock(&qi->qi_tree_lock); | |
812 | trace_xfs_dqget_freeing(dqp); | |
813 | delay(1); | |
814 | goto restart; | |
815 | } | |
1da177e4 | 816 | |
296c24e2 ES |
817 | /* uninit / unused quota found in radix tree, keep looking */ |
818 | if (flags & XFS_QMOPT_DQNEXT) { | |
819 | if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) { | |
820 | xfs_dqunlock(dqp); | |
821 | mutex_unlock(&qi->qi_tree_lock); | |
822 | error = xfs_dq_get_next_id(mp, type, &id, eof); | |
823 | if (error) | |
824 | return error; | |
825 | goto restart; | |
826 | } | |
827 | } | |
828 | ||
9f920f11 CH |
829 | dqp->q_nrefs++; |
830 | mutex_unlock(&qi->qi_tree_lock); | |
831 | ||
832 | trace_xfs_dqget_hit(dqp); | |
ff6d6af2 | 833 | XFS_STATS_INC(mp, xs_qm_dqcachehits); |
9f920f11 CH |
834 | *O_dqpp = dqp; |
835 | return 0; | |
1da177e4 | 836 | } |
9f920f11 | 837 | mutex_unlock(&qi->qi_tree_lock); |
ff6d6af2 | 838 | XFS_STATS_INC(mp, xs_qm_dqcachemisses); |
1da177e4 LT |
839 | |
840 | /* | |
841 | * Dquot cache miss. We don't want to keep the inode lock across | |
842 | * a (potential) disk read. Also we don't want to deal with the lock | |
843 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
844 | * lock here means dealing with a chown that can happen before | |
845 | * we re-acquire the lock. | |
846 | */ | |
847 | if (ip) | |
848 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 | 849 | |
97e7ade5 | 850 | error = xfs_qm_dqread(mp, id, type, flags, &dqp); |
1da177e4 | 851 | |
7ae44407 CH |
852 | if (ip) |
853 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
854 | ||
296c24e2 ES |
855 | /* If we are asked to find next active id, keep looking */ |
856 | if (error == -ENOENT && (flags & XFS_QMOPT_DQNEXT)) { | |
857 | error = xfs_dq_get_next_id(mp, type, &id, eof); | |
858 | if (!error) | |
859 | goto restart; | |
860 | } | |
861 | ||
7ae44407 CH |
862 | if (error) |
863 | return error; | |
1da177e4 | 864 | |
1da177e4 | 865 | if (ip) { |
1da177e4 LT |
866 | /* |
867 | * A dquot could be attached to this inode by now, since | |
868 | * we had dropped the ilock. | |
869 | */ | |
36731410 | 870 | if (xfs_this_quota_on(mp, type)) { |
9f920f11 CH |
871 | struct xfs_dquot *dqp1; |
872 | ||
36731410 CS |
873 | dqp1 = xfs_inode_dquot(ip, type); |
874 | if (dqp1) { | |
1da177e4 | 875 | xfs_qm_dqdestroy(dqp); |
36731410 | 876 | dqp = dqp1; |
1da177e4 LT |
877 | xfs_dqlock(dqp); |
878 | goto dqret; | |
879 | } | |
880 | } else { | |
36731410 CS |
881 | /* inode stays locked on return */ |
882 | xfs_qm_dqdestroy(dqp); | |
2451337d | 883 | return -ESRCH; |
1da177e4 LT |
884 | } |
885 | } | |
886 | ||
9f920f11 | 887 | mutex_lock(&qi->qi_tree_lock); |
2451337d | 888 | error = radix_tree_insert(tree, id, dqp); |
9f920f11 | 889 | if (unlikely(error)) { |
2451337d | 890 | WARN_ON(error != -EEXIST); |
9f920f11 | 891 | |
1da177e4 | 892 | /* |
9f920f11 CH |
893 | * Duplicate found. Just throw away the new dquot and start |
894 | * over. | |
1da177e4 | 895 | */ |
9f920f11 CH |
896 | mutex_unlock(&qi->qi_tree_lock); |
897 | trace_xfs_dqget_dup(dqp); | |
898 | xfs_qm_dqdestroy(dqp); | |
ff6d6af2 | 899 | XFS_STATS_INC(mp, xs_qm_dquot_dups); |
9f920f11 | 900 | goto restart; |
1da177e4 LT |
901 | } |
902 | ||
1da177e4 LT |
903 | /* |
904 | * We return a locked dquot to the caller, with a reference taken | |
905 | */ | |
906 | xfs_dqlock(dqp); | |
907 | dqp->q_nrefs = 1; | |
908 | ||
9f920f11 | 909 | qi->qi_dquots++; |
9f920f11 CH |
910 | mutex_unlock(&qi->qi_tree_lock); |
911 | ||
296c24e2 ES |
912 | /* If we are asked to find next active id, keep looking */ |
913 | if (flags & XFS_QMOPT_DQNEXT) { | |
914 | if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) { | |
915 | xfs_qm_dqput(dqp); | |
916 | error = xfs_dq_get_next_id(mp, type, &id, eof); | |
917 | if (error) | |
918 | return error; | |
919 | goto restart; | |
920 | } | |
921 | } | |
922 | ||
1da177e4 | 923 | dqret: |
579aa9ca | 924 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 925 | trace_xfs_dqget_miss(dqp); |
1da177e4 | 926 | *O_dqpp = dqp; |
d99831ff | 927 | return 0; |
1da177e4 LT |
928 | } |
929 | ||
f8739c3c CH |
930 | /* |
931 | * Release a reference to the dquot (decrement ref-count) and unlock it. | |
932 | * | |
933 | * If there is a group quota attached to this dquot, carefully release that | |
934 | * too without tripping over deadlocks'n'stuff. | |
935 | */ | |
936 | void | |
937 | xfs_qm_dqput( | |
938 | struct xfs_dquot *dqp) | |
939 | { | |
940 | ASSERT(dqp->q_nrefs > 0); | |
941 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
942 | ||
943 | trace_xfs_dqput(dqp); | |
944 | ||
3c353375 DC |
945 | if (--dqp->q_nrefs == 0) { |
946 | struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo; | |
947 | trace_xfs_dqput_free(dqp); | |
948 | ||
949 | if (list_lru_add(&qi->qi_lru, &dqp->q_lru)) | |
ff6d6af2 | 950 | XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused); |
3c353375 DC |
951 | } |
952 | xfs_dqunlock(dqp); | |
1da177e4 LT |
953 | } |
954 | ||
955 | /* | |
956 | * Release a dquot. Flush it if dirty, then dqput() it. | |
957 | * dquot must not be locked. | |
958 | */ | |
959 | void | |
960 | xfs_qm_dqrele( | |
961 | xfs_dquot_t *dqp) | |
962 | { | |
7d095257 CH |
963 | if (!dqp) |
964 | return; | |
965 | ||
0b1b213f | 966 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
967 | |
968 | xfs_dqlock(dqp); | |
969 | /* | |
970 | * We don't care to flush it if the dquot is dirty here. | |
971 | * That will create stutters that we want to avoid. | |
972 | * Instead we do a delayed write when we try to reclaim | |
973 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
974 | */ | |
975 | xfs_qm_dqput(dqp); | |
976 | } | |
977 | ||
ca30b2a7 CH |
978 | /* |
979 | * This is the dquot flushing I/O completion routine. It is called | |
980 | * from interrupt level when the buffer containing the dquot is | |
981 | * flushed to disk. It is responsible for removing the dquot logitem | |
982 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
983 | * flush lock. This behavior is very similar to that of inodes.. | |
984 | */ | |
985 | STATIC void | |
986 | xfs_qm_dqflush_done( | |
987 | struct xfs_buf *bp, | |
988 | struct xfs_log_item *lip) | |
989 | { | |
990 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
991 | xfs_dquot_t *dqp = qip->qli_dquot; | |
992 | struct xfs_ail *ailp = lip->li_ailp; | |
993 | ||
994 | /* | |
995 | * We only want to pull the item from the AIL if its | |
996 | * location in the log has not changed since we started the flush. | |
997 | * Thus, we only bother if the dquot's lsn has | |
998 | * not changed. First we check the lsn outside the lock | |
999 | * since it's cheaper, and then we recheck while | |
1000 | * holding the lock before removing the dquot from the AIL. | |
1001 | */ | |
1002 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
1003 | lip->li_lsn == qip->qli_flush_lsn) { | |
1004 | ||
1005 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
1006 | spin_lock(&ailp->xa_lock); | |
1007 | if (lip->li_lsn == qip->qli_flush_lsn) | |
04913fdd | 1008 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 CH |
1009 | else |
1010 | spin_unlock(&ailp->xa_lock); | |
1011 | } | |
1012 | ||
1013 | /* | |
1014 | * Release the dq's flush lock since we're done with it. | |
1015 | */ | |
1016 | xfs_dqfunlock(dqp); | |
1017 | } | |
1da177e4 LT |
1018 | |
1019 | /* | |
1020 | * Write a modified dquot to disk. | |
1021 | * The dquot must be locked and the flush lock too taken by caller. | |
1022 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
1023 | * but the dquot is free to be unlocked and modified by the caller | |
1024 | * in the interim. Dquot is still locked on return. This behavior is | |
1025 | * identical to that of inodes. | |
1026 | */ | |
1027 | int | |
1028 | xfs_qm_dqflush( | |
fe7257fd CH |
1029 | struct xfs_dquot *dqp, |
1030 | struct xfs_buf **bpp) | |
1da177e4 | 1031 | { |
acecf1b5 CH |
1032 | struct xfs_mount *mp = dqp->q_mount; |
1033 | struct xfs_buf *bp; | |
1034 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 1035 | int error; |
1da177e4 LT |
1036 | |
1037 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 1038 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 1039 | |
0b1b213f | 1040 | trace_xfs_dqflush(dqp); |
1da177e4 | 1041 | |
fe7257fd CH |
1042 | *bpp = NULL; |
1043 | ||
1da177e4 LT |
1044 | xfs_qm_dqunpin_wait(dqp); |
1045 | ||
1046 | /* | |
1047 | * This may have been unpinned because the filesystem is shutting | |
1048 | * down forcibly. If that's the case we must not write this dquot | |
dea96095 CH |
1049 | * to disk, because the log record didn't make it to disk. |
1050 | * | |
1051 | * We also have to remove the log item from the AIL in this case, | |
1052 | * as we wait for an emptry AIL as part of the unmount process. | |
1da177e4 | 1053 | */ |
acecf1b5 | 1054 | if (XFS_FORCED_SHUTDOWN(mp)) { |
dea96095 | 1055 | struct xfs_log_item *lip = &dqp->q_logitem.qli_item; |
acecf1b5 | 1056 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
dea96095 | 1057 | |
146e54b7 BF |
1058 | xfs_trans_ail_remove(lip, SHUTDOWN_CORRUPT_INCORE); |
1059 | ||
2451337d | 1060 | error = -EIO; |
fe7257fd | 1061 | goto out_unlock; |
1da177e4 LT |
1062 | } |
1063 | ||
1064 | /* | |
1065 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1066 | */ |
acecf1b5 | 1067 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
5fd364fe DC |
1068 | mp->m_quotainfo->qi_dqchunklen, 0, &bp, |
1069 | &xfs_dquot_buf_ops); | |
fe7257fd CH |
1070 | if (error) |
1071 | goto out_unlock; | |
1da177e4 | 1072 | |
acecf1b5 CH |
1073 | /* |
1074 | * Calculate the location of the dquot inside the buffer. | |
1075 | */ | |
62926044 | 1076 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1077 | |
1078 | /* | |
1079 | * A simple sanity check in case we got a corrupted dquot.. | |
1080 | */ | |
9aede1d8 | 1081 | error = xfs_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
a0fa2b67 DC |
1082 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); |
1083 | if (error) { | |
acecf1b5 CH |
1084 | xfs_buf_relse(bp); |
1085 | xfs_dqfunlock(dqp); | |
1086 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
2451337d | 1087 | return -EIO; |
1da177e4 LT |
1088 | } |
1089 | ||
1090 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1091 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1092 | |
1093 | /* | |
1094 | * Clear the dirty field and remember the flush lsn for later use. | |
1095 | */ | |
acecf1b5 | 1096 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1097 | |
7b2e2a31 DC |
1098 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1099 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 | 1100 | |
3fe58f30 CH |
1101 | /* |
1102 | * copy the lsn into the on-disk dquot now while we have the in memory | |
1103 | * dquot here. This can't be done later in the write verifier as we | |
1104 | * can't get access to the log item at that point in time. | |
6fcdc59d DC |
1105 | * |
1106 | * We also calculate the CRC here so that the on-disk dquot in the | |
1107 | * buffer always has a valid CRC. This ensures there is no possibility | |
1108 | * of a dquot without an up-to-date CRC getting to disk. | |
3fe58f30 CH |
1109 | */ |
1110 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
1111 | struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp; | |
1112 | ||
1113 | dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn); | |
6fcdc59d DC |
1114 | xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk), |
1115 | XFS_DQUOT_CRC_OFF); | |
3fe58f30 CH |
1116 | } |
1117 | ||
1da177e4 LT |
1118 | /* |
1119 | * Attach an iodone routine so that we can remove this dquot from the | |
1120 | * AIL and release the flush lock once the dquot is synced to disk. | |
1121 | */ | |
ca30b2a7 CH |
1122 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1123 | &dqp->q_logitem.qli_item); | |
1124 | ||
1da177e4 LT |
1125 | /* |
1126 | * If the buffer is pinned then push on the log so we won't | |
1127 | * get stuck waiting in the write for too long. | |
1128 | */ | |
811e64c7 | 1129 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1130 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1131 | xfs_log_force(mp, 0); |
1da177e4 LT |
1132 | } |
1133 | ||
0b1b213f | 1134 | trace_xfs_dqflush_done(dqp); |
fe7257fd CH |
1135 | *bpp = bp; |
1136 | return 0; | |
0b1b213f | 1137 | |
fe7257fd CH |
1138 | out_unlock: |
1139 | xfs_dqfunlock(dqp); | |
2451337d | 1140 | return -EIO; |
1da177e4 LT |
1141 | } |
1142 | ||
5bb87a33 CH |
1143 | /* |
1144 | * Lock two xfs_dquot structures. | |
1145 | * | |
1146 | * To avoid deadlocks we always lock the quota structure with | |
1147 | * the lowerd id first. | |
1148 | */ | |
1da177e4 LT |
1149 | void |
1150 | xfs_dqlock2( | |
1151 | xfs_dquot_t *d1, | |
1152 | xfs_dquot_t *d2) | |
1153 | { | |
1154 | if (d1 && d2) { | |
1155 | ASSERT(d1 != d2); | |
1149d96a CH |
1156 | if (be32_to_cpu(d1->q_core.d_id) > |
1157 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1158 | mutex_lock(&d2->q_qlock); |
1159 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1160 | } else { |
5bb87a33 CH |
1161 | mutex_lock(&d1->q_qlock); |
1162 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1163 | } |
5bb87a33 CH |
1164 | } else if (d1) { |
1165 | mutex_lock(&d1->q_qlock); | |
1166 | } else if (d2) { | |
1167 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1168 | } |
1169 | } | |
1170 | ||
a05931ce CH |
1171 | int __init |
1172 | xfs_qm_init(void) | |
1173 | { | |
1174 | xfs_qm_dqzone = | |
1175 | kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot"); | |
1176 | if (!xfs_qm_dqzone) | |
1177 | goto out; | |
1178 | ||
1179 | xfs_qm_dqtrxzone = | |
1180 | kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx"); | |
1181 | if (!xfs_qm_dqtrxzone) | |
1182 | goto out_free_dqzone; | |
1183 | ||
1184 | return 0; | |
1185 | ||
1186 | out_free_dqzone: | |
1187 | kmem_zone_destroy(xfs_qm_dqzone); | |
1188 | out: | |
1189 | return -ENOMEM; | |
1190 | } | |
1191 | ||
1c2ccc66 | 1192 | void |
a05931ce CH |
1193 | xfs_qm_exit(void) |
1194 | { | |
1195 | kmem_zone_destroy(xfs_qm_dqtrxzone); | |
1196 | kmem_zone_destroy(xfs_qm_dqzone); | |
1197 | } |