]>
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 | { | |
c8ce540d | 279 | uint64_t space; |
b1366451 BF |
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); | |
50e0bdbe | 475 | uint lock_mode = 0; |
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 | |
50e0bdbe DW |
480 | ASSERT(!(flags & XFS_QMOPT_NOLOCK) || |
481 | xfs_isilocked(quotip, XFS_ILOCK_SHARED) || | |
482 | xfs_isilocked(quotip, XFS_ILOCK_EXCL)); | |
483 | if (!(flags & XFS_QMOPT_NOLOCK)) | |
484 | lock_mode = xfs_ilock_data_map_shared(quotip); | |
6967b964 | 485 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 | 486 | /* |
acecf1b5 CH |
487 | * Return if this type of quotas is turned off while we |
488 | * didn't have the quota inode lock. | |
1da177e4 | 489 | */ |
50e0bdbe DW |
490 | if (lock_mode) |
491 | xfs_iunlock(quotip, lock_mode); | |
2451337d | 492 | return -ESRCH; |
acecf1b5 CH |
493 | } |
494 | ||
495 | /* | |
496 | * Find the block map; no allocations yet | |
497 | */ | |
5c8ed202 DC |
498 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
499 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); | |
acecf1b5 | 500 | |
50e0bdbe DW |
501 | if (lock_mode) |
502 | xfs_iunlock(quotip, lock_mode); | |
acecf1b5 CH |
503 | if (error) |
504 | return error; | |
505 | ||
506 | ASSERT(nmaps == 1); | |
507 | ASSERT(map.br_blockcount == 1); | |
508 | ||
509 | /* | |
510 | * Offset of dquot in the (fixed sized) dquot chunk. | |
511 | */ | |
512 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
513 | sizeof(xfs_dqblk_t); | |
514 | ||
515 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
516 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 517 | /* |
acecf1b5 | 518 | * We don't allocate unless we're asked to |
1da177e4 | 519 | */ |
acecf1b5 | 520 | if (!(flags & XFS_QMOPT_DQALLOC)) |
2451337d | 521 | return -ENOENT; |
1da177e4 | 522 | |
acecf1b5 CH |
523 | ASSERT(tp); |
524 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
525 | dqp->q_fileoffset, &bp); | |
1da177e4 | 526 | if (error) |
acecf1b5 CH |
527 | return error; |
528 | tp = *tpp; | |
529 | } else { | |
530 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
531 | |
532 | /* | |
acecf1b5 CH |
533 | * store the blkno etc so that we don't have to do the |
534 | * mapping all the time | |
1da177e4 | 535 | */ |
acecf1b5 | 536 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 537 | |
8a7b8a89 CH |
538 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
539 | dqp->q_blkno, | |
540 | mp->m_quotainfo->qi_dqchunklen, | |
1813dd64 | 541 | 0, &bp, &xfs_dquot_buf_ops); |
acecf1b5 | 542 | |
2451337d | 543 | if (error == -EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) { |
c6319198 DC |
544 | xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff * |
545 | mp->m_quotainfo->qi_dqperchunk; | |
546 | ASSERT(bp == NULL); | |
547 | error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp); | |
548 | } | |
1da177e4 | 549 | |
c6319198 DC |
550 | if (error) { |
551 | ASSERT(bp == NULL); | |
b474c7ae | 552 | return error; |
1da177e4 | 553 | } |
1da177e4 LT |
554 | } |
555 | ||
c6319198 | 556 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 557 | *O_bpp = bp; |
c6319198 | 558 | *O_ddpp = bp->b_addr + dqp->q_bufoffset; |
1da177e4 | 559 | |
d99831ff | 560 | return 0; |
1da177e4 LT |
561 | } |
562 | ||
563 | ||
564 | /* | |
565 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
566 | * and release the buffer immediately. | |
567 | * | |
97e7ade5 | 568 | * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed. |
1da177e4 | 569 | */ |
7ae44407 | 570 | int |
1da177e4 | 571 | xfs_qm_dqread( |
97e7ade5 CH |
572 | struct xfs_mount *mp, |
573 | xfs_dqid_t id, | |
574 | uint type, | |
575 | uint flags, | |
576 | struct xfs_dquot **O_dqpp) | |
1da177e4 | 577 | { |
97e7ade5 CH |
578 | struct xfs_dquot *dqp; |
579 | struct xfs_disk_dquot *ddqp; | |
580 | struct xfs_buf *bp; | |
581 | struct xfs_trans *tp = NULL; | |
582 | int error; | |
92b2e5b3 | 583 | |
a05931ce | 584 | dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP); |
92b2e5b3 CH |
585 | |
586 | dqp->dq_flags = type; | |
587 | dqp->q_core.d_id = cpu_to_be32(id); | |
588 | dqp->q_mount = mp; | |
f8739c3c | 589 | INIT_LIST_HEAD(&dqp->q_lru); |
92b2e5b3 CH |
590 | mutex_init(&dqp->q_qlock); |
591 | init_waitqueue_head(&dqp->q_pinwait); | |
592 | ||
593 | /* | |
594 | * Because we want to use a counting completion, complete | |
595 | * the flush completion once to allow a single access to | |
596 | * the flush completion without blocking. | |
597 | */ | |
598 | init_completion(&dqp->q_flush); | |
599 | complete(&dqp->q_flush); | |
600 | ||
601 | /* | |
602 | * Make sure group quotas have a different lock class than user | |
603 | * quotas. | |
604 | */ | |
f112a049 DC |
605 | switch (type) { |
606 | case XFS_DQ_USER: | |
607 | /* uses the default lock class */ | |
608 | break; | |
609 | case XFS_DQ_GROUP: | |
610 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class); | |
611 | break; | |
612 | case XFS_DQ_PROJ: | |
613 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class); | |
614 | break; | |
615 | default: | |
616 | ASSERT(0); | |
617 | break; | |
618 | } | |
92b2e5b3 | 619 | |
ff6d6af2 | 620 | XFS_STATS_INC(mp, xs_qm_dquot); |
1da177e4 | 621 | |
0b1b213f CH |
622 | trace_xfs_dqread(dqp); |
623 | ||
97e7ade5 | 624 | if (flags & XFS_QMOPT_DQALLOC) { |
253f4911 CH |
625 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc, |
626 | XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp); | |
97e7ade5 | 627 | if (error) |
253f4911 | 628 | goto error0; |
97e7ade5 CH |
629 | } |
630 | ||
1da177e4 LT |
631 | /* |
632 | * get a pointer to the on-disk dquot and the buffer containing it | |
633 | * dqp already knows its own type (GROUP/USER). | |
634 | */ | |
97e7ade5 CH |
635 | error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags); |
636 | if (error) { | |
637 | /* | |
638 | * This can happen if quotas got turned off (ESRCH), | |
639 | * or if the dquot didn't exist on disk and we ask to | |
640 | * allocate (ENOENT). | |
641 | */ | |
642 | trace_xfs_dqread_fail(dqp); | |
97e7ade5 | 643 | goto error1; |
1da177e4 LT |
644 | } |
645 | ||
646 | /* copy everything from disk dquot to the incore dquot */ | |
647 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
1da177e4 LT |
648 | xfs_qm_dquot_logitem_init(dqp); |
649 | ||
650 | /* | |
651 | * Reservation counters are defined as reservation plus current usage | |
25985edc | 652 | * to avoid having to add every time. |
1da177e4 | 653 | */ |
1149d96a CH |
654 | dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); |
655 | dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); | |
656 | dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); | |
1da177e4 | 657 | |
b1366451 BF |
658 | /* initialize the dquot speculative prealloc thresholds */ |
659 | xfs_dquot_set_prealloc_limits(dqp); | |
660 | ||
1da177e4 | 661 | /* Mark the buf so that this will stay incore a little longer */ |
38f23232 | 662 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 LT |
663 | |
664 | /* | |
665 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
666 | * So we need to release with xfs_trans_brelse(). | |
667 | * The strategy here is identical to that of inodes; we lock | |
668 | * the dquot in xfs_qm_dqget() before making it accessible to | |
669 | * others. This is because dquots, like inodes, need a good level of | |
670 | * concurrency, and we don't want to take locks on the entire buffers | |
671 | * for dquot accesses. | |
672 | * Note also that the dquot buffer may even be dirty at this point, if | |
673 | * this particular dquot was repaired. We still aren't afraid to | |
674 | * brelse it because we have the changes incore. | |
675 | */ | |
0c842ad4 | 676 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
677 | xfs_trans_brelse(tp, bp); |
678 | ||
1da177e4 | 679 | if (tp) { |
70393313 | 680 | error = xfs_trans_commit(tp); |
97e7ade5 CH |
681 | if (error) |
682 | goto error0; | |
1da177e4 LT |
683 | } |
684 | ||
685 | *O_dqpp = dqp; | |
97e7ade5 | 686 | return error; |
1da177e4 | 687 | |
97e7ade5 | 688 | error1: |
1da177e4 | 689 | if (tp) |
4906e215 | 690 | xfs_trans_cancel(tp); |
97e7ade5 | 691 | error0: |
1da177e4 LT |
692 | xfs_qm_dqdestroy(dqp); |
693 | *O_dqpp = NULL; | |
97e7ade5 | 694 | return error; |
1da177e4 LT |
695 | } |
696 | ||
296c24e2 ES |
697 | /* |
698 | * Advance to the next id in the current chunk, or if at the | |
699 | * end of the chunk, skip ahead to first id in next allocated chunk | |
700 | * using the SEEK_DATA interface. | |
701 | */ | |
6e3e6d55 | 702 | static int |
296c24e2 | 703 | xfs_dq_get_next_id( |
bda250db | 704 | struct xfs_mount *mp, |
296c24e2 | 705 | uint type, |
bda250db | 706 | xfs_dqid_t *id) |
296c24e2 | 707 | { |
bda250db CH |
708 | struct xfs_inode *quotip = xfs_quota_inode(mp, type); |
709 | xfs_dqid_t next_id = *id + 1; /* simple advance */ | |
710 | uint lock_flags; | |
711 | struct xfs_bmbt_irec got; | |
712 | xfs_extnum_t idx; | |
296c24e2 | 713 | xfs_fsblock_t start; |
296c24e2 ES |
714 | int error = 0; |
715 | ||
657bdfb7 ES |
716 | /* If we'd wrap past the max ID, stop */ |
717 | if (next_id < *id) | |
718 | return -ENOENT; | |
719 | ||
296c24e2 ES |
720 | /* If new ID is within the current chunk, advancing it sufficed */ |
721 | if (next_id % mp->m_quotainfo->qi_dqperchunk) { | |
722 | *id = next_id; | |
723 | return 0; | |
724 | } | |
725 | ||
726 | /* Nope, next_id is now past the current chunk, so find the next one */ | |
727 | start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk; | |
728 | ||
bda250db CH |
729 | lock_flags = xfs_ilock_data_map_shared(quotip); |
730 | if (!(quotip->i_df.if_flags & XFS_IFEXTENTS)) { | |
731 | error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK); | |
732 | if (error) | |
733 | return error; | |
734 | } | |
296c24e2 | 735 | |
2192b0ba BF |
736 | if (xfs_iext_lookup_extent(quotip, "ip->i_df, start, &idx, &got)) { |
737 | /* contiguous chunk, bump startoff for the id calculation */ | |
738 | if (got.br_startoff < start) | |
739 | got.br_startoff = start; | |
bda250db | 740 | *id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk; |
2192b0ba | 741 | } else { |
bda250db | 742 | error = -ENOENT; |
2192b0ba BF |
743 | } |
744 | ||
bda250db | 745 | xfs_iunlock(quotip, lock_flags); |
296c24e2 | 746 | |
bda250db | 747 | return error; |
296c24e2 ES |
748 | } |
749 | ||
1da177e4 LT |
750 | /* |
751 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
752 | * a locked dquot, doing an allocation (if requested) as needed. | |
753 | * When both an inode and an id are given, the inode's id takes precedence. | |
754 | * That is, if the id changes while we don't hold the ilock inside this | |
755 | * function, the new dquot is returned, not necessarily the one requested | |
756 | * in the id argument. | |
757 | */ | |
758 | int | |
759 | xfs_qm_dqget( | |
760 | xfs_mount_t *mp, | |
761 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
762 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
763 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
764 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
765 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
766 | { | |
9f920f11 | 767 | struct xfs_quotainfo *qi = mp->m_quotainfo; |
329e0875 | 768 | struct radix_tree_root *tree = xfs_dquot_tree(qi, type); |
9f920f11 CH |
769 | struct xfs_dquot *dqp; |
770 | int error; | |
1da177e4 LT |
771 | |
772 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
773 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 774 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 | 775 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
2451337d | 776 | return -ESRCH; |
1da177e4 | 777 | } |
1da177e4 LT |
778 | |
779 | #ifdef DEBUG | |
780 | if (xfs_do_dqerror) { | |
781 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
782 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 783 | xfs_debug(mp, "Returning error in dqget"); |
2451337d | 784 | return -EIO; |
1da177e4 LT |
785 | } |
786 | } | |
1da177e4 | 787 | |
c8ad20ff NS |
788 | ASSERT(type == XFS_DQ_USER || |
789 | type == XFS_DQ_PROJ || | |
790 | type == XFS_DQ_GROUP); | |
1da177e4 | 791 | if (ip) { |
579aa9ca | 792 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
36731410 | 793 | ASSERT(xfs_inode_dquot(ip, type) == NULL); |
1da177e4 LT |
794 | } |
795 | #endif | |
92678554 CH |
796 | |
797 | restart: | |
9f920f11 CH |
798 | mutex_lock(&qi->qi_tree_lock); |
799 | dqp = radix_tree_lookup(tree, id); | |
800 | if (dqp) { | |
801 | xfs_dqlock(dqp); | |
802 | if (dqp->dq_flags & XFS_DQ_FREEING) { | |
803 | xfs_dqunlock(dqp); | |
804 | mutex_unlock(&qi->qi_tree_lock); | |
805 | trace_xfs_dqget_freeing(dqp); | |
806 | delay(1); | |
807 | goto restart; | |
808 | } | |
1da177e4 | 809 | |
296c24e2 ES |
810 | /* uninit / unused quota found in radix tree, keep looking */ |
811 | if (flags & XFS_QMOPT_DQNEXT) { | |
812 | if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) { | |
813 | xfs_dqunlock(dqp); | |
814 | mutex_unlock(&qi->qi_tree_lock); | |
bda250db | 815 | error = xfs_dq_get_next_id(mp, type, &id); |
296c24e2 ES |
816 | if (error) |
817 | return error; | |
818 | goto restart; | |
819 | } | |
820 | } | |
821 | ||
9f920f11 CH |
822 | dqp->q_nrefs++; |
823 | mutex_unlock(&qi->qi_tree_lock); | |
824 | ||
825 | trace_xfs_dqget_hit(dqp); | |
ff6d6af2 | 826 | XFS_STATS_INC(mp, xs_qm_dqcachehits); |
9f920f11 CH |
827 | *O_dqpp = dqp; |
828 | return 0; | |
1da177e4 | 829 | } |
9f920f11 | 830 | mutex_unlock(&qi->qi_tree_lock); |
ff6d6af2 | 831 | XFS_STATS_INC(mp, xs_qm_dqcachemisses); |
1da177e4 LT |
832 | |
833 | /* | |
834 | * Dquot cache miss. We don't want to keep the inode lock across | |
835 | * a (potential) disk read. Also we don't want to deal with the lock | |
836 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
837 | * lock here means dealing with a chown that can happen before | |
838 | * we re-acquire the lock. | |
839 | */ | |
840 | if (ip) | |
841 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 | 842 | |
97e7ade5 | 843 | error = xfs_qm_dqread(mp, id, type, flags, &dqp); |
1da177e4 | 844 | |
7ae44407 CH |
845 | if (ip) |
846 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
847 | ||
296c24e2 ES |
848 | /* If we are asked to find next active id, keep looking */ |
849 | if (error == -ENOENT && (flags & XFS_QMOPT_DQNEXT)) { | |
bda250db | 850 | error = xfs_dq_get_next_id(mp, type, &id); |
296c24e2 ES |
851 | if (!error) |
852 | goto restart; | |
853 | } | |
854 | ||
7ae44407 CH |
855 | if (error) |
856 | return error; | |
1da177e4 | 857 | |
1da177e4 | 858 | if (ip) { |
1da177e4 LT |
859 | /* |
860 | * A dquot could be attached to this inode by now, since | |
861 | * we had dropped the ilock. | |
862 | */ | |
36731410 | 863 | if (xfs_this_quota_on(mp, type)) { |
9f920f11 CH |
864 | struct xfs_dquot *dqp1; |
865 | ||
36731410 CS |
866 | dqp1 = xfs_inode_dquot(ip, type); |
867 | if (dqp1) { | |
1da177e4 | 868 | xfs_qm_dqdestroy(dqp); |
36731410 | 869 | dqp = dqp1; |
1da177e4 LT |
870 | xfs_dqlock(dqp); |
871 | goto dqret; | |
872 | } | |
873 | } else { | |
36731410 CS |
874 | /* inode stays locked on return */ |
875 | xfs_qm_dqdestroy(dqp); | |
2451337d | 876 | return -ESRCH; |
1da177e4 LT |
877 | } |
878 | } | |
879 | ||
9f920f11 | 880 | mutex_lock(&qi->qi_tree_lock); |
2451337d | 881 | error = radix_tree_insert(tree, id, dqp); |
9f920f11 | 882 | if (unlikely(error)) { |
2451337d | 883 | WARN_ON(error != -EEXIST); |
9f920f11 | 884 | |
1da177e4 | 885 | /* |
9f920f11 CH |
886 | * Duplicate found. Just throw away the new dquot and start |
887 | * over. | |
1da177e4 | 888 | */ |
9f920f11 CH |
889 | mutex_unlock(&qi->qi_tree_lock); |
890 | trace_xfs_dqget_dup(dqp); | |
891 | xfs_qm_dqdestroy(dqp); | |
ff6d6af2 | 892 | XFS_STATS_INC(mp, xs_qm_dquot_dups); |
9f920f11 | 893 | goto restart; |
1da177e4 LT |
894 | } |
895 | ||
1da177e4 LT |
896 | /* |
897 | * We return a locked dquot to the caller, with a reference taken | |
898 | */ | |
899 | xfs_dqlock(dqp); | |
900 | dqp->q_nrefs = 1; | |
901 | ||
9f920f11 | 902 | qi->qi_dquots++; |
9f920f11 CH |
903 | mutex_unlock(&qi->qi_tree_lock); |
904 | ||
296c24e2 ES |
905 | /* If we are asked to find next active id, keep looking */ |
906 | if (flags & XFS_QMOPT_DQNEXT) { | |
907 | if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) { | |
908 | xfs_qm_dqput(dqp); | |
bda250db | 909 | error = xfs_dq_get_next_id(mp, type, &id); |
296c24e2 ES |
910 | if (error) |
911 | return error; | |
912 | goto restart; | |
913 | } | |
914 | } | |
915 | ||
1da177e4 | 916 | dqret: |
579aa9ca | 917 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 918 | trace_xfs_dqget_miss(dqp); |
1da177e4 | 919 | *O_dqpp = dqp; |
d99831ff | 920 | return 0; |
1da177e4 LT |
921 | } |
922 | ||
f8739c3c CH |
923 | /* |
924 | * Release a reference to the dquot (decrement ref-count) and unlock it. | |
925 | * | |
926 | * If there is a group quota attached to this dquot, carefully release that | |
927 | * too without tripping over deadlocks'n'stuff. | |
928 | */ | |
929 | void | |
930 | xfs_qm_dqput( | |
931 | struct xfs_dquot *dqp) | |
932 | { | |
933 | ASSERT(dqp->q_nrefs > 0); | |
934 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
935 | ||
936 | trace_xfs_dqput(dqp); | |
937 | ||
3c353375 DC |
938 | if (--dqp->q_nrefs == 0) { |
939 | struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo; | |
940 | trace_xfs_dqput_free(dqp); | |
941 | ||
942 | if (list_lru_add(&qi->qi_lru, &dqp->q_lru)) | |
ff6d6af2 | 943 | XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused); |
3c353375 DC |
944 | } |
945 | xfs_dqunlock(dqp); | |
1da177e4 LT |
946 | } |
947 | ||
948 | /* | |
949 | * Release a dquot. Flush it if dirty, then dqput() it. | |
950 | * dquot must not be locked. | |
951 | */ | |
952 | void | |
953 | xfs_qm_dqrele( | |
954 | xfs_dquot_t *dqp) | |
955 | { | |
7d095257 CH |
956 | if (!dqp) |
957 | return; | |
958 | ||
0b1b213f | 959 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
960 | |
961 | xfs_dqlock(dqp); | |
962 | /* | |
963 | * We don't care to flush it if the dquot is dirty here. | |
964 | * That will create stutters that we want to avoid. | |
965 | * Instead we do a delayed write when we try to reclaim | |
966 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
967 | */ | |
968 | xfs_qm_dqput(dqp); | |
969 | } | |
970 | ||
ca30b2a7 CH |
971 | /* |
972 | * This is the dquot flushing I/O completion routine. It is called | |
973 | * from interrupt level when the buffer containing the dquot is | |
974 | * flushed to disk. It is responsible for removing the dquot logitem | |
975 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
976 | * flush lock. This behavior is very similar to that of inodes.. | |
977 | */ | |
978 | STATIC void | |
979 | xfs_qm_dqflush_done( | |
980 | struct xfs_buf *bp, | |
981 | struct xfs_log_item *lip) | |
982 | { | |
983 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
984 | xfs_dquot_t *dqp = qip->qli_dquot; | |
985 | struct xfs_ail *ailp = lip->li_ailp; | |
986 | ||
987 | /* | |
988 | * We only want to pull the item from the AIL if its | |
989 | * location in the log has not changed since we started the flush. | |
990 | * Thus, we only bother if the dquot's lsn has | |
991 | * not changed. First we check the lsn outside the lock | |
992 | * since it's cheaper, and then we recheck while | |
993 | * holding the lock before removing the dquot from the AIL. | |
994 | */ | |
995 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
996 | lip->li_lsn == qip->qli_flush_lsn) { | |
997 | ||
998 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
999 | spin_lock(&ailp->xa_lock); | |
1000 | if (lip->li_lsn == qip->qli_flush_lsn) | |
04913fdd | 1001 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 CH |
1002 | else |
1003 | spin_unlock(&ailp->xa_lock); | |
1004 | } | |
1005 | ||
1006 | /* | |
1007 | * Release the dq's flush lock since we're done with it. | |
1008 | */ | |
1009 | xfs_dqfunlock(dqp); | |
1010 | } | |
1da177e4 LT |
1011 | |
1012 | /* | |
1013 | * Write a modified dquot to disk. | |
1014 | * The dquot must be locked and the flush lock too taken by caller. | |
1015 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
1016 | * but the dquot is free to be unlocked and modified by the caller | |
1017 | * in the interim. Dquot is still locked on return. This behavior is | |
1018 | * identical to that of inodes. | |
1019 | */ | |
1020 | int | |
1021 | xfs_qm_dqflush( | |
fe7257fd CH |
1022 | struct xfs_dquot *dqp, |
1023 | struct xfs_buf **bpp) | |
1da177e4 | 1024 | { |
acecf1b5 CH |
1025 | struct xfs_mount *mp = dqp->q_mount; |
1026 | struct xfs_buf *bp; | |
1027 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 1028 | int error; |
1da177e4 LT |
1029 | |
1030 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 1031 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 1032 | |
0b1b213f | 1033 | trace_xfs_dqflush(dqp); |
1da177e4 | 1034 | |
fe7257fd CH |
1035 | *bpp = NULL; |
1036 | ||
1da177e4 LT |
1037 | xfs_qm_dqunpin_wait(dqp); |
1038 | ||
1039 | /* | |
1040 | * This may have been unpinned because the filesystem is shutting | |
1041 | * down forcibly. If that's the case we must not write this dquot | |
dea96095 CH |
1042 | * to disk, because the log record didn't make it to disk. |
1043 | * | |
1044 | * We also have to remove the log item from the AIL in this case, | |
1045 | * as we wait for an emptry AIL as part of the unmount process. | |
1da177e4 | 1046 | */ |
acecf1b5 | 1047 | if (XFS_FORCED_SHUTDOWN(mp)) { |
dea96095 | 1048 | struct xfs_log_item *lip = &dqp->q_logitem.qli_item; |
acecf1b5 | 1049 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
dea96095 | 1050 | |
146e54b7 BF |
1051 | xfs_trans_ail_remove(lip, SHUTDOWN_CORRUPT_INCORE); |
1052 | ||
2451337d | 1053 | error = -EIO; |
fe7257fd | 1054 | goto out_unlock; |
1da177e4 LT |
1055 | } |
1056 | ||
1057 | /* | |
1058 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1059 | */ |
acecf1b5 | 1060 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
5fd364fe DC |
1061 | mp->m_quotainfo->qi_dqchunklen, 0, &bp, |
1062 | &xfs_dquot_buf_ops); | |
fe7257fd CH |
1063 | if (error) |
1064 | goto out_unlock; | |
1da177e4 | 1065 | |
acecf1b5 CH |
1066 | /* |
1067 | * Calculate the location of the dquot inside the buffer. | |
1068 | */ | |
62926044 | 1069 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1070 | |
1071 | /* | |
1072 | * A simple sanity check in case we got a corrupted dquot.. | |
1073 | */ | |
9aede1d8 | 1074 | error = xfs_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
a0fa2b67 DC |
1075 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); |
1076 | if (error) { | |
acecf1b5 CH |
1077 | xfs_buf_relse(bp); |
1078 | xfs_dqfunlock(dqp); | |
1079 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
2451337d | 1080 | return -EIO; |
1da177e4 LT |
1081 | } |
1082 | ||
1083 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1084 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1085 | |
1086 | /* | |
1087 | * Clear the dirty field and remember the flush lsn for later use. | |
1088 | */ | |
acecf1b5 | 1089 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1090 | |
7b2e2a31 DC |
1091 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1092 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 | 1093 | |
3fe58f30 CH |
1094 | /* |
1095 | * copy the lsn into the on-disk dquot now while we have the in memory | |
1096 | * dquot here. This can't be done later in the write verifier as we | |
1097 | * can't get access to the log item at that point in time. | |
6fcdc59d DC |
1098 | * |
1099 | * We also calculate the CRC here so that the on-disk dquot in the | |
1100 | * buffer always has a valid CRC. This ensures there is no possibility | |
1101 | * of a dquot without an up-to-date CRC getting to disk. | |
3fe58f30 CH |
1102 | */ |
1103 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
1104 | struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp; | |
1105 | ||
1106 | dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn); | |
6fcdc59d DC |
1107 | xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk), |
1108 | XFS_DQUOT_CRC_OFF); | |
3fe58f30 CH |
1109 | } |
1110 | ||
1da177e4 LT |
1111 | /* |
1112 | * Attach an iodone routine so that we can remove this dquot from the | |
1113 | * AIL and release the flush lock once the dquot is synced to disk. | |
1114 | */ | |
ca30b2a7 CH |
1115 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1116 | &dqp->q_logitem.qli_item); | |
1117 | ||
1da177e4 LT |
1118 | /* |
1119 | * If the buffer is pinned then push on the log so we won't | |
1120 | * get stuck waiting in the write for too long. | |
1121 | */ | |
811e64c7 | 1122 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1123 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1124 | xfs_log_force(mp, 0); |
1da177e4 LT |
1125 | } |
1126 | ||
0b1b213f | 1127 | trace_xfs_dqflush_done(dqp); |
fe7257fd CH |
1128 | *bpp = bp; |
1129 | return 0; | |
0b1b213f | 1130 | |
fe7257fd CH |
1131 | out_unlock: |
1132 | xfs_dqfunlock(dqp); | |
2451337d | 1133 | return -EIO; |
1da177e4 LT |
1134 | } |
1135 | ||
5bb87a33 CH |
1136 | /* |
1137 | * Lock two xfs_dquot structures. | |
1138 | * | |
1139 | * To avoid deadlocks we always lock the quota structure with | |
1140 | * the lowerd id first. | |
1141 | */ | |
1da177e4 LT |
1142 | void |
1143 | xfs_dqlock2( | |
1144 | xfs_dquot_t *d1, | |
1145 | xfs_dquot_t *d2) | |
1146 | { | |
1147 | if (d1 && d2) { | |
1148 | ASSERT(d1 != d2); | |
1149d96a CH |
1149 | if (be32_to_cpu(d1->q_core.d_id) > |
1150 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1151 | mutex_lock(&d2->q_qlock); |
1152 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1153 | } else { |
5bb87a33 CH |
1154 | mutex_lock(&d1->q_qlock); |
1155 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1156 | } |
5bb87a33 CH |
1157 | } else if (d1) { |
1158 | mutex_lock(&d1->q_qlock); | |
1159 | } else if (d2) { | |
1160 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1161 | } |
1162 | } | |
1163 | ||
a05931ce CH |
1164 | int __init |
1165 | xfs_qm_init(void) | |
1166 | { | |
1167 | xfs_qm_dqzone = | |
1168 | kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot"); | |
1169 | if (!xfs_qm_dqzone) | |
1170 | goto out; | |
1171 | ||
1172 | xfs_qm_dqtrxzone = | |
1173 | kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx"); | |
1174 | if (!xfs_qm_dqtrxzone) | |
1175 | goto out_free_dqzone; | |
1176 | ||
1177 | return 0; | |
1178 | ||
1179 | out_free_dqzone: | |
1180 | kmem_zone_destroy(xfs_qm_dqzone); | |
1181 | out: | |
1182 | return -ENOMEM; | |
1183 | } | |
1184 | ||
1c2ccc66 | 1185 | void |
a05931ce CH |
1186 | xfs_qm_exit(void) |
1187 | { | |
1188 | kmem_zone_destroy(xfs_qm_dqtrxzone); | |
1189 | kmem_zone_destroy(xfs_qm_dqzone); | |
1190 | } |