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" | |
a844f451 | 20 | #include "xfs_bit.h" |
1da177e4 LT |
21 | #include "xfs_log.h" |
22 | #include "xfs_trans.h" | |
23 | #include "xfs_sb.h" | |
24 | #include "xfs_ag.h" | |
1da177e4 | 25 | #include "xfs_alloc.h" |
1da177e4 LT |
26 | #include "xfs_quota.h" |
27 | #include "xfs_mount.h" | |
1da177e4 | 28 | #include "xfs_bmap_btree.h" |
1da177e4 LT |
29 | #include "xfs_inode.h" |
30 | #include "xfs_bmap.h" | |
1da177e4 LT |
31 | #include "xfs_rtalloc.h" |
32 | #include "xfs_error.h" | |
33 | #include "xfs_itable.h" | |
1da177e4 LT |
34 | #include "xfs_attr.h" |
35 | #include "xfs_buf_item.h" | |
36 | #include "xfs_trans_space.h" | |
37 | #include "xfs_trans_priv.h" | |
1da177e4 | 38 | #include "xfs_qm.h" |
0b1b213f | 39 | #include "xfs_trace.h" |
1da177e4 | 40 | |
1da177e4 | 41 | /* |
bf72de31 CH |
42 | * Lock order: |
43 | * | |
44 | * ip->i_lock | |
9f920f11 | 45 | * qi->qi_tree_lock |
b84a3a96 CH |
46 | * dquot->q_qlock (xfs_dqlock() and friends) |
47 | * dquot->q_flush (xfs_dqflock() and friends) | |
48 | * qi->qi_lru_lock | |
bf72de31 CH |
49 | * |
50 | * If two dquots need to be locked the order is user before group/project, | |
51 | * otherwise by the lowest id first, see xfs_dqlock2. | |
52 | */ | |
1da177e4 | 53 | |
1da177e4 LT |
54 | #ifdef DEBUG |
55 | xfs_buftarg_t *xfs_dqerror_target; | |
56 | int xfs_do_dqerror; | |
57 | int xfs_dqreq_num; | |
58 | int xfs_dqerror_mod = 33; | |
59 | #endif | |
60 | ||
a05931ce CH |
61 | struct kmem_zone *xfs_qm_dqtrxzone; |
62 | static struct kmem_zone *xfs_qm_dqzone; | |
63 | ||
98b8c7a0 CH |
64 | static struct lock_class_key xfs_dquot_other_class; |
65 | ||
1da177e4 LT |
66 | /* |
67 | * This is called to free all the memory associated with a dquot | |
68 | */ | |
69 | void | |
70 | xfs_qm_dqdestroy( | |
71 | xfs_dquot_t *dqp) | |
72 | { | |
f8739c3c | 73 | ASSERT(list_empty(&dqp->q_lru)); |
1da177e4 LT |
74 | |
75 | mutex_destroy(&dqp->q_qlock); | |
a05931ce | 76 | kmem_zone_free(xfs_qm_dqzone, dqp); |
0b1b213f | 77 | |
48776fd2 | 78 | XFS_STATS_DEC(xs_qm_dquot); |
1da177e4 LT |
79 | } |
80 | ||
1da177e4 LT |
81 | /* |
82 | * If default limits are in force, push them into the dquot now. | |
83 | * We overwrite the dquot limits only if they are zero and this | |
84 | * is not the root dquot. | |
85 | */ | |
86 | void | |
87 | xfs_qm_adjust_dqlimits( | |
88 | xfs_mount_t *mp, | |
89 | xfs_disk_dquot_t *d) | |
90 | { | |
91 | xfs_quotainfo_t *q = mp->m_quotainfo; | |
92 | ||
93 | ASSERT(d->d_id); | |
94 | ||
95 | if (q->qi_bsoftlimit && !d->d_blk_softlimit) | |
1149d96a | 96 | d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit); |
1da177e4 | 97 | if (q->qi_bhardlimit && !d->d_blk_hardlimit) |
1149d96a | 98 | d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit); |
1da177e4 | 99 | if (q->qi_isoftlimit && !d->d_ino_softlimit) |
1149d96a | 100 | d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit); |
1da177e4 | 101 | if (q->qi_ihardlimit && !d->d_ino_hardlimit) |
1149d96a | 102 | d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit); |
1da177e4 | 103 | if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit) |
1149d96a | 104 | d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit); |
1da177e4 | 105 | if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit) |
1149d96a | 106 | d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit); |
1da177e4 LT |
107 | } |
108 | ||
109 | /* | |
110 | * Check the limits and timers of a dquot and start or reset timers | |
111 | * if necessary. | |
112 | * This gets called even when quota enforcement is OFF, which makes our | |
113 | * life a little less complicated. (We just don't reject any quota | |
114 | * reservations in that case, when enforcement is off). | |
115 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
116 | * enforcement's off. | |
117 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
118 | * 'automatically' get started when limits get exceeded. They do |
119 | * get reset to zero, however, when we find the count to be under | |
120 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
121 | */ |
122 | void | |
123 | xfs_qm_adjust_dqtimers( | |
124 | xfs_mount_t *mp, | |
125 | xfs_disk_dquot_t *d) | |
126 | { | |
127 | ASSERT(d->d_id); | |
128 | ||
ea15ab3c | 129 | #ifdef DEBUG |
1149d96a CH |
130 | if (d->d_blk_hardlimit) |
131 | ASSERT(be64_to_cpu(d->d_blk_softlimit) <= | |
132 | be64_to_cpu(d->d_blk_hardlimit)); | |
133 | if (d->d_ino_hardlimit) | |
134 | ASSERT(be64_to_cpu(d->d_ino_softlimit) <= | |
135 | be64_to_cpu(d->d_ino_hardlimit)); | |
136 | if (d->d_rtb_hardlimit) | |
137 | ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= | |
138 | be64_to_cpu(d->d_rtb_hardlimit)); | |
1da177e4 | 139 | #endif |
ea15ab3c | 140 | |
1da177e4 | 141 | if (!d->d_btimer) { |
1149d96a | 142 | if ((d->d_blk_softlimit && |
d0a3fe67 | 143 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
144 | be64_to_cpu(d->d_blk_softlimit))) || |
145 | (d->d_blk_hardlimit && | |
d0a3fe67 | 146 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
147 | be64_to_cpu(d->d_blk_hardlimit)))) { |
148 | d->d_btimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 149 | mp->m_quotainfo->qi_btimelimit); |
754002b4 NS |
150 | } else { |
151 | d->d_bwarns = 0; | |
1da177e4 LT |
152 | } |
153 | } else { | |
154 | if ((!d->d_blk_softlimit || | |
d0a3fe67 | 155 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 156 | be64_to_cpu(d->d_blk_softlimit))) && |
1da177e4 | 157 | (!d->d_blk_hardlimit || |
d0a3fe67 | 158 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 159 | be64_to_cpu(d->d_blk_hardlimit)))) { |
1da177e4 LT |
160 | d->d_btimer = 0; |
161 | } | |
162 | } | |
163 | ||
164 | if (!d->d_itimer) { | |
1149d96a | 165 | if ((d->d_ino_softlimit && |
d0a3fe67 | 166 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
167 | be64_to_cpu(d->d_ino_softlimit))) || |
168 | (d->d_ino_hardlimit && | |
d0a3fe67 | 169 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
170 | be64_to_cpu(d->d_ino_hardlimit)))) { |
171 | d->d_itimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 172 | mp->m_quotainfo->qi_itimelimit); |
754002b4 NS |
173 | } else { |
174 | d->d_iwarns = 0; | |
1da177e4 LT |
175 | } |
176 | } else { | |
177 | if ((!d->d_ino_softlimit || | |
d0a3fe67 | 178 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 179 | be64_to_cpu(d->d_ino_softlimit))) && |
1da177e4 | 180 | (!d->d_ino_hardlimit || |
d0a3fe67 | 181 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 182 | be64_to_cpu(d->d_ino_hardlimit)))) { |
1da177e4 LT |
183 | d->d_itimer = 0; |
184 | } | |
185 | } | |
186 | ||
187 | if (!d->d_rtbtimer) { | |
1149d96a | 188 | if ((d->d_rtb_softlimit && |
d0a3fe67 | 189 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
190 | be64_to_cpu(d->d_rtb_softlimit))) || |
191 | (d->d_rtb_hardlimit && | |
d0a3fe67 | 192 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
193 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
194 | d->d_rtbtimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 195 | mp->m_quotainfo->qi_rtbtimelimit); |
754002b4 NS |
196 | } else { |
197 | d->d_rtbwarns = 0; | |
1da177e4 LT |
198 | } |
199 | } else { | |
200 | if ((!d->d_rtb_softlimit || | |
d0a3fe67 | 201 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 202 | be64_to_cpu(d->d_rtb_softlimit))) && |
1da177e4 | 203 | (!d->d_rtb_hardlimit || |
d0a3fe67 | 204 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 205 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
1da177e4 LT |
206 | d->d_rtbtimer = 0; |
207 | } | |
208 | } | |
209 | } | |
210 | ||
1da177e4 LT |
211 | /* |
212 | * initialize a buffer full of dquots and log the whole thing | |
213 | */ | |
214 | STATIC void | |
215 | xfs_qm_init_dquot_blk( | |
216 | xfs_trans_t *tp, | |
217 | xfs_mount_t *mp, | |
218 | xfs_dqid_t id, | |
219 | uint type, | |
220 | xfs_buf_t *bp) | |
221 | { | |
8a7b8a89 | 222 | struct xfs_quotainfo *q = mp->m_quotainfo; |
1da177e4 LT |
223 | xfs_dqblk_t *d; |
224 | int curid, i; | |
225 | ||
226 | ASSERT(tp); | |
0c842ad4 | 227 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 228 | |
62926044 | 229 | d = bp->b_addr; |
1da177e4 LT |
230 | |
231 | /* | |
232 | * ID of the first dquot in the block - id's are zero based. | |
233 | */ | |
8a7b8a89 | 234 | curid = id - (id % q->qi_dqperchunk); |
1da177e4 | 235 | ASSERT(curid >= 0); |
8a7b8a89 | 236 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
49d35a5c CH |
237 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) { |
238 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
239 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
240 | d->dd_diskdq.d_id = cpu_to_be32(curid); | |
241 | d->dd_diskdq.d_flags = type; | |
242 | } | |
243 | ||
1da177e4 | 244 | xfs_trans_dquot_buf(tp, bp, |
c1155410 DC |
245 | (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : |
246 | ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : | |
247 | XFS_BLF_GDQUOT_BUF))); | |
8a7b8a89 | 248 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); |
1da177e4 LT |
249 | } |
250 | ||
b0f539de DC |
251 | static void |
252 | xfs_dquot_buf_verify( | |
253 | struct xfs_buf *bp) | |
254 | { | |
255 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
256 | struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr; | |
257 | struct xfs_disk_dquot *ddq; | |
258 | xfs_dqid_t id = 0; | |
259 | int i; | |
260 | ||
261 | /* | |
262 | * On the first read of the buffer, verify that each dquot is valid. | |
263 | * We don't know what the id of the dquot is supposed to be, just that | |
264 | * they should be increasing monotonically within the buffer. If the | |
265 | * first id is corrupt, then it will fail on the second dquot in the | |
266 | * buffer so corruptions could point to the wrong dquot in this case. | |
267 | */ | |
268 | for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) { | |
269 | int error; | |
270 | ||
271 | ddq = &d[i].dd_diskdq; | |
272 | ||
273 | if (i == 0) | |
274 | id = be32_to_cpu(ddq->d_id); | |
275 | ||
276 | error = xfs_qm_dqcheck(mp, ddq, id + i, 0, XFS_QMOPT_DOWARN, | |
277 | "xfs_dquot_read_verify"); | |
278 | if (error) { | |
279 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, d); | |
280 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
281 | break; | |
282 | } | |
283 | } | |
284 | } | |
285 | ||
286 | static void | |
1813dd64 | 287 | xfs_dquot_buf_read_verify( |
b0f539de DC |
288 | struct xfs_buf *bp) |
289 | { | |
290 | xfs_dquot_buf_verify(bp); | |
291 | } | |
1da177e4 | 292 | |
b0f539de | 293 | void |
1813dd64 | 294 | xfs_dquot_buf_write_verify( |
b0f539de DC |
295 | struct xfs_buf *bp) |
296 | { | |
297 | xfs_dquot_buf_verify(bp); | |
b0f539de | 298 | } |
1da177e4 | 299 | |
1813dd64 DC |
300 | const struct xfs_buf_ops xfs_dquot_buf_ops = { |
301 | .verify_read = xfs_dquot_buf_read_verify, | |
302 | .verify_write = xfs_dquot_buf_write_verify, | |
303 | }; | |
304 | ||
1da177e4 LT |
305 | /* |
306 | * Allocate a block and fill it with dquots. | |
307 | * This is called when the bmapi finds a hole. | |
308 | */ | |
309 | STATIC int | |
310 | xfs_qm_dqalloc( | |
efa092f3 | 311 | xfs_trans_t **tpp, |
1da177e4 LT |
312 | xfs_mount_t *mp, |
313 | xfs_dquot_t *dqp, | |
314 | xfs_inode_t *quotip, | |
315 | xfs_fileoff_t offset_fsb, | |
316 | xfs_buf_t **O_bpp) | |
317 | { | |
318 | xfs_fsblock_t firstblock; | |
319 | xfs_bmap_free_t flist; | |
320 | xfs_bmbt_irec_t map; | |
321 | int nmaps, error, committed; | |
322 | xfs_buf_t *bp; | |
efa092f3 | 323 | xfs_trans_t *tp = *tpp; |
1da177e4 LT |
324 | |
325 | ASSERT(tp != NULL); | |
0b1b213f CH |
326 | |
327 | trace_xfs_dqalloc(dqp); | |
1da177e4 LT |
328 | |
329 | /* | |
330 | * Initialize the bmap freelist prior to calling bmapi code. | |
331 | */ | |
9d87c319 | 332 | xfs_bmap_init(&flist, &firstblock); |
1da177e4 LT |
333 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
334 | /* | |
335 | * Return if this type of quotas is turned off while we didn't | |
336 | * have an inode lock | |
337 | */ | |
6967b964 | 338 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 LT |
339 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
340 | return (ESRCH); | |
341 | } | |
342 | ||
ddc3415a | 343 | xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); |
1da177e4 | 344 | nmaps = 1; |
c0dc7828 DC |
345 | error = xfs_bmapi_write(tp, quotip, offset_fsb, |
346 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
347 | &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), | |
348 | &map, &nmaps, &flist); | |
349 | if (error) | |
1da177e4 | 350 | goto error0; |
1da177e4 LT |
351 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
352 | ASSERT(nmaps == 1); | |
353 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
354 | (map.br_startblock != HOLESTARTBLOCK)); | |
355 | ||
356 | /* | |
357 | * Keep track of the blkno to save a lookup later | |
358 | */ | |
359 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
360 | ||
361 | /* now we can just get the buffer (there's nothing to read yet) */ | |
362 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
363 | dqp->q_blkno, | |
8a7b8a89 | 364 | mp->m_quotainfo->qi_dqchunklen, |
1da177e4 | 365 | 0); |
2a30f36d CS |
366 | |
367 | error = xfs_buf_geterror(bp); | |
368 | if (error) | |
1da177e4 | 369 | goto error1; |
1813dd64 | 370 | bp->b_ops = &xfs_dquot_buf_ops; |
2a30f36d | 371 | |
1da177e4 LT |
372 | /* |
373 | * Make a chunk of dquots out of this buffer and log | |
374 | * the entire thing. | |
375 | */ | |
1149d96a | 376 | xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 377 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
1da177e4 | 378 | |
efa092f3 TS |
379 | /* |
380 | * xfs_bmap_finish() may commit the current transaction and | |
381 | * start a second transaction if the freelist is not empty. | |
382 | * | |
383 | * Since we still want to modify this buffer, we need to | |
384 | * ensure that the buffer is not released on commit of | |
385 | * the first transaction and ensure the buffer is added to the | |
386 | * second transaction. | |
387 | * | |
388 | * If there is only one transaction then don't stop the buffer | |
389 | * from being released when it commits later on. | |
390 | */ | |
391 | ||
392 | xfs_trans_bhold(tp, bp); | |
393 | ||
f7c99b6f | 394 | if ((error = xfs_bmap_finish(tpp, &flist, &committed))) { |
1da177e4 LT |
395 | goto error1; |
396 | } | |
397 | ||
efa092f3 TS |
398 | if (committed) { |
399 | tp = *tpp; | |
400 | xfs_trans_bjoin(tp, bp); | |
401 | } else { | |
402 | xfs_trans_bhold_release(tp, bp); | |
403 | } | |
404 | ||
1da177e4 LT |
405 | *O_bpp = bp; |
406 | return 0; | |
407 | ||
408 | error1: | |
409 | xfs_bmap_cancel(&flist); | |
410 | error0: | |
411 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
412 | ||
413 | return (error); | |
414 | } | |
c6319198 DC |
415 | STATIC int |
416 | xfs_qm_dqrepair( | |
417 | struct xfs_mount *mp, | |
418 | struct xfs_trans *tp, | |
419 | struct xfs_dquot *dqp, | |
420 | xfs_dqid_t firstid, | |
421 | struct xfs_buf **bpp) | |
422 | { | |
423 | int error; | |
424 | struct xfs_disk_dquot *ddq; | |
425 | struct xfs_dqblk *d; | |
426 | int i; | |
427 | ||
428 | /* | |
429 | * Read the buffer without verification so we get the corrupted | |
612cfbfe | 430 | * buffer returned to us. make sure we verify it on write, though. |
c6319198 DC |
431 | */ |
432 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno, | |
433 | mp->m_quotainfo->qi_dqchunklen, | |
434 | 0, bpp, NULL); | |
435 | ||
436 | if (error) { | |
437 | ASSERT(*bpp == NULL); | |
438 | return XFS_ERROR(error); | |
439 | } | |
1813dd64 | 440 | (*bpp)->b_ops = &xfs_dquot_buf_ops; |
c6319198 DC |
441 | |
442 | ASSERT(xfs_buf_islocked(*bpp)); | |
443 | d = (struct xfs_dqblk *)(*bpp)->b_addr; | |
444 | ||
445 | /* Do the actual repair of dquots in this buffer */ | |
446 | for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) { | |
447 | ddq = &d[i].dd_diskdq; | |
448 | error = xfs_qm_dqcheck(mp, ddq, firstid + i, | |
449 | dqp->dq_flags & XFS_DQ_ALLTYPES, | |
450 | XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair"); | |
451 | if (error) { | |
452 | /* repair failed, we're screwed */ | |
453 | xfs_trans_brelse(tp, *bpp); | |
454 | return XFS_ERROR(EIO); | |
455 | } | |
456 | } | |
457 | ||
458 | return 0; | |
459 | } | |
460 | ||
1da177e4 LT |
461 | /* |
462 | * Maps a dquot to the buffer containing its on-disk version. | |
463 | * This returns a ptr to the buffer containing the on-disk dquot | |
464 | * in the bpp param, and a ptr to the on-disk dquot within that buffer | |
465 | */ | |
466 | STATIC int | |
467 | xfs_qm_dqtobp( | |
efa092f3 | 468 | xfs_trans_t **tpp, |
1da177e4 LT |
469 | xfs_dquot_t *dqp, |
470 | xfs_disk_dquot_t **O_ddpp, | |
471 | xfs_buf_t **O_bpp, | |
472 | uint flags) | |
473 | { | |
474 | xfs_bmbt_irec_t map; | |
acecf1b5 | 475 | int nmaps = 1, error; |
1da177e4 | 476 | xfs_buf_t *bp; |
acecf1b5 CH |
477 | xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp); |
478 | xfs_mount_t *mp = dqp->q_mount; | |
acecf1b5 | 479 | xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); |
efa092f3 | 480 | xfs_trans_t *tp = (tpp ? *tpp : NULL); |
1da177e4 | 481 | |
acecf1b5 | 482 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
1da177e4 | 483 | |
acecf1b5 | 484 | xfs_ilock(quotip, XFS_ILOCK_SHARED); |
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 | */ |
acecf1b5 CH |
490 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); |
491 | return ESRCH; | |
492 | } | |
493 | ||
494 | /* | |
495 | * Find the block map; no allocations yet | |
496 | */ | |
5c8ed202 DC |
497 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
498 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); | |
acecf1b5 CH |
499 | |
500 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); | |
501 | if (error) | |
502 | return error; | |
503 | ||
504 | ASSERT(nmaps == 1); | |
505 | ASSERT(map.br_blockcount == 1); | |
506 | ||
507 | /* | |
508 | * Offset of dquot in the (fixed sized) dquot chunk. | |
509 | */ | |
510 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
511 | sizeof(xfs_dqblk_t); | |
512 | ||
513 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
514 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 515 | /* |
acecf1b5 | 516 | * We don't allocate unless we're asked to |
1da177e4 | 517 | */ |
acecf1b5 CH |
518 | if (!(flags & XFS_QMOPT_DQALLOC)) |
519 | return ENOENT; | |
1da177e4 | 520 | |
acecf1b5 CH |
521 | ASSERT(tp); |
522 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
523 | dqp->q_fileoffset, &bp); | |
1da177e4 | 524 | if (error) |
acecf1b5 CH |
525 | return error; |
526 | tp = *tpp; | |
527 | } else { | |
528 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
529 | |
530 | /* | |
acecf1b5 CH |
531 | * store the blkno etc so that we don't have to do the |
532 | * mapping all the time | |
1da177e4 | 533 | */ |
acecf1b5 | 534 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 535 | |
8a7b8a89 CH |
536 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
537 | dqp->q_blkno, | |
538 | mp->m_quotainfo->qi_dqchunklen, | |
1813dd64 | 539 | 0, &bp, &xfs_dquot_buf_ops); |
acecf1b5 | 540 | |
c6319198 DC |
541 | if (error == EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) { |
542 | xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff * | |
543 | mp->m_quotainfo->qi_dqperchunk; | |
544 | ASSERT(bp == NULL); | |
545 | error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp); | |
546 | } | |
1da177e4 | 547 | |
c6319198 DC |
548 | if (error) { |
549 | ASSERT(bp == NULL); | |
550 | return XFS_ERROR(error); | |
1da177e4 | 551 | } |
1da177e4 LT |
552 | } |
553 | ||
c6319198 | 554 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 555 | *O_bpp = bp; |
c6319198 | 556 | *O_ddpp = bp->b_addr + dqp->q_bufoffset; |
1da177e4 LT |
557 | |
558 | return (0); | |
559 | } | |
560 | ||
561 | ||
562 | /* | |
563 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
564 | * and release the buffer immediately. | |
565 | * | |
97e7ade5 | 566 | * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed. |
1da177e4 | 567 | */ |
7ae44407 | 568 | int |
1da177e4 | 569 | xfs_qm_dqread( |
97e7ade5 CH |
570 | struct xfs_mount *mp, |
571 | xfs_dqid_t id, | |
572 | uint type, | |
573 | uint flags, | |
574 | struct xfs_dquot **O_dqpp) | |
1da177e4 | 575 | { |
97e7ade5 CH |
576 | struct xfs_dquot *dqp; |
577 | struct xfs_disk_dquot *ddqp; | |
578 | struct xfs_buf *bp; | |
579 | struct xfs_trans *tp = NULL; | |
580 | int error; | |
581 | int cancelflags = 0; | |
efa092f3 | 582 | |
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 | */ | |
605 | if (!(type & XFS_DQ_USER)) | |
606 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class); | |
607 | ||
48776fd2 | 608 | XFS_STATS_INC(xs_qm_dquot); |
1da177e4 | 609 | |
0b1b213f CH |
610 | trace_xfs_dqread(dqp); |
611 | ||
97e7ade5 CH |
612 | if (flags & XFS_QMOPT_DQALLOC) { |
613 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC); | |
614 | error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp), | |
48001044 JL |
615 | XFS_QM_DQALLOC_LOG_RES(mp), 0, |
616 | XFS_TRANS_PERM_LOG_RES, | |
617 | XFS_WRITE_LOG_COUNT); | |
97e7ade5 CH |
618 | if (error) |
619 | goto error1; | |
620 | cancelflags = XFS_TRANS_RELEASE_LOG_RES; | |
621 | } | |
622 | ||
1da177e4 LT |
623 | /* |
624 | * get a pointer to the on-disk dquot and the buffer containing it | |
625 | * dqp already knows its own type (GROUP/USER). | |
626 | */ | |
97e7ade5 CH |
627 | 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; | |
1da177e4 LT |
637 | } |
638 | ||
639 | /* copy everything from disk dquot to the incore dquot */ | |
640 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
1da177e4 LT |
641 | xfs_qm_dquot_logitem_init(dqp); |
642 | ||
643 | /* | |
644 | * Reservation counters are defined as reservation plus current usage | |
25985edc | 645 | * to avoid having to add every time. |
1da177e4 | 646 | */ |
1149d96a CH |
647 | 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); | |
1da177e4 LT |
650 | |
651 | /* Mark the buf so that this will stay incore a little longer */ | |
38f23232 | 652 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 LT |
653 | |
654 | /* | |
655 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
656 | * So we need to release with xfs_trans_brelse(). | |
657 | * The strategy here is identical to that of inodes; we lock | |
658 | * the dquot in xfs_qm_dqget() before making it accessible to | |
659 | * others. This is because dquots, like inodes, need a good level of | |
660 | * concurrency, and we don't want to take locks on the entire buffers | |
661 | * for dquot accesses. | |
662 | * Note also that the dquot buffer may even be dirty at this point, if | |
663 | * this particular dquot was repaired. We still aren't afraid to | |
664 | * brelse it because we have the changes incore. | |
665 | */ | |
0c842ad4 | 666 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
667 | xfs_trans_brelse(tp, bp); |
668 | ||
1da177e4 | 669 | if (tp) { |
97e7ade5 CH |
670 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
671 | if (error) | |
672 | goto error0; | |
1da177e4 LT |
673 | } |
674 | ||
675 | *O_dqpp = dqp; | |
97e7ade5 | 676 | return error; |
1da177e4 | 677 | |
97e7ade5 | 678 | error1: |
1da177e4 LT |
679 | if (tp) |
680 | xfs_trans_cancel(tp, cancelflags); | |
97e7ade5 | 681 | error0: |
1da177e4 LT |
682 | xfs_qm_dqdestroy(dqp); |
683 | *O_dqpp = NULL; | |
97e7ade5 | 684 | return error; |
1da177e4 LT |
685 | } |
686 | ||
1da177e4 LT |
687 | /* |
688 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
689 | * a locked dquot, doing an allocation (if requested) as needed. | |
690 | * When both an inode and an id are given, the inode's id takes precedence. | |
691 | * That is, if the id changes while we don't hold the ilock inside this | |
692 | * function, the new dquot is returned, not necessarily the one requested | |
693 | * in the id argument. | |
694 | */ | |
695 | int | |
696 | xfs_qm_dqget( | |
697 | xfs_mount_t *mp, | |
698 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
699 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
700 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
701 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
702 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
703 | { | |
9f920f11 CH |
704 | struct xfs_quotainfo *qi = mp->m_quotainfo; |
705 | struct radix_tree_root *tree = XFS_DQUOT_TREE(qi, type); | |
706 | struct xfs_dquot *dqp; | |
707 | int error; | |
1da177e4 LT |
708 | |
709 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
710 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 711 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 LT |
712 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
713 | return (ESRCH); | |
714 | } | |
1da177e4 LT |
715 | |
716 | #ifdef DEBUG | |
717 | if (xfs_do_dqerror) { | |
718 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
719 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 720 | xfs_debug(mp, "Returning error in dqget"); |
1da177e4 LT |
721 | return (EIO); |
722 | } | |
723 | } | |
1da177e4 | 724 | |
c8ad20ff NS |
725 | ASSERT(type == XFS_DQ_USER || |
726 | type == XFS_DQ_PROJ || | |
727 | type == XFS_DQ_GROUP); | |
1da177e4 | 728 | if (ip) { |
579aa9ca | 729 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
36731410 | 730 | ASSERT(xfs_inode_dquot(ip, type) == NULL); |
1da177e4 LT |
731 | } |
732 | #endif | |
92678554 CH |
733 | |
734 | restart: | |
9f920f11 CH |
735 | mutex_lock(&qi->qi_tree_lock); |
736 | dqp = radix_tree_lookup(tree, id); | |
737 | if (dqp) { | |
738 | xfs_dqlock(dqp); | |
739 | if (dqp->dq_flags & XFS_DQ_FREEING) { | |
740 | xfs_dqunlock(dqp); | |
741 | mutex_unlock(&qi->qi_tree_lock); | |
742 | trace_xfs_dqget_freeing(dqp); | |
743 | delay(1); | |
744 | goto restart; | |
745 | } | |
1da177e4 | 746 | |
9f920f11 CH |
747 | dqp->q_nrefs++; |
748 | mutex_unlock(&qi->qi_tree_lock); | |
749 | ||
750 | trace_xfs_dqget_hit(dqp); | |
48776fd2 | 751 | XFS_STATS_INC(xs_qm_dqcachehits); |
9f920f11 CH |
752 | *O_dqpp = dqp; |
753 | return 0; | |
1da177e4 | 754 | } |
9f920f11 CH |
755 | mutex_unlock(&qi->qi_tree_lock); |
756 | XFS_STATS_INC(xs_qm_dqcachemisses); | |
1da177e4 LT |
757 | |
758 | /* | |
759 | * Dquot cache miss. We don't want to keep the inode lock across | |
760 | * a (potential) disk read. Also we don't want to deal with the lock | |
761 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
762 | * lock here means dealing with a chown that can happen before | |
763 | * we re-acquire the lock. | |
764 | */ | |
765 | if (ip) | |
766 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 | 767 | |
97e7ade5 | 768 | error = xfs_qm_dqread(mp, id, type, flags, &dqp); |
1da177e4 | 769 | |
7ae44407 CH |
770 | if (ip) |
771 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
772 | ||
773 | if (error) | |
774 | return error; | |
1da177e4 | 775 | |
1da177e4 | 776 | if (ip) { |
1da177e4 LT |
777 | /* |
778 | * A dquot could be attached to this inode by now, since | |
779 | * we had dropped the ilock. | |
780 | */ | |
36731410 | 781 | if (xfs_this_quota_on(mp, type)) { |
9f920f11 CH |
782 | struct xfs_dquot *dqp1; |
783 | ||
36731410 CS |
784 | dqp1 = xfs_inode_dquot(ip, type); |
785 | if (dqp1) { | |
1da177e4 | 786 | xfs_qm_dqdestroy(dqp); |
36731410 | 787 | dqp = dqp1; |
1da177e4 LT |
788 | xfs_dqlock(dqp); |
789 | goto dqret; | |
790 | } | |
791 | } else { | |
36731410 CS |
792 | /* inode stays locked on return */ |
793 | xfs_qm_dqdestroy(dqp); | |
794 | return XFS_ERROR(ESRCH); | |
1da177e4 LT |
795 | } |
796 | } | |
797 | ||
9f920f11 CH |
798 | mutex_lock(&qi->qi_tree_lock); |
799 | error = -radix_tree_insert(tree, id, dqp); | |
800 | if (unlikely(error)) { | |
801 | WARN_ON(error != EEXIST); | |
802 | ||
1da177e4 | 803 | /* |
9f920f11 CH |
804 | * Duplicate found. Just throw away the new dquot and start |
805 | * over. | |
1da177e4 | 806 | */ |
9f920f11 CH |
807 | mutex_unlock(&qi->qi_tree_lock); |
808 | trace_xfs_dqget_dup(dqp); | |
809 | xfs_qm_dqdestroy(dqp); | |
810 | XFS_STATS_INC(xs_qm_dquot_dups); | |
811 | goto restart; | |
1da177e4 LT |
812 | } |
813 | ||
1da177e4 LT |
814 | /* |
815 | * We return a locked dquot to the caller, with a reference taken | |
816 | */ | |
817 | xfs_dqlock(dqp); | |
818 | dqp->q_nrefs = 1; | |
819 | ||
9f920f11 | 820 | qi->qi_dquots++; |
9f920f11 CH |
821 | mutex_unlock(&qi->qi_tree_lock); |
822 | ||
1da177e4 | 823 | dqret: |
579aa9ca | 824 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 825 | trace_xfs_dqget_miss(dqp); |
1da177e4 LT |
826 | *O_dqpp = dqp; |
827 | return (0); | |
828 | } | |
829 | ||
830 | ||
f8739c3c CH |
831 | STATIC void |
832 | xfs_qm_dqput_final( | |
bf72de31 | 833 | struct xfs_dquot *dqp) |
1da177e4 | 834 | { |
f8739c3c | 835 | struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo; |
bf72de31 | 836 | struct xfs_dquot *gdqp; |
1da177e4 | 837 | |
bf72de31 CH |
838 | trace_xfs_dqput_free(dqp); |
839 | ||
f8739c3c CH |
840 | mutex_lock(&qi->qi_lru_lock); |
841 | if (list_empty(&dqp->q_lru)) { | |
842 | list_add_tail(&dqp->q_lru, &qi->qi_lru_list); | |
843 | qi->qi_lru_count++; | |
48776fd2 | 844 | XFS_STATS_INC(xs_qm_dquot_unused); |
bf72de31 | 845 | } |
f8739c3c | 846 | mutex_unlock(&qi->qi_lru_lock); |
bf72de31 | 847 | |
1da177e4 | 848 | /* |
bf72de31 CH |
849 | * If we just added a udquot to the freelist, then we want to release |
850 | * the gdquot reference that it (probably) has. Otherwise it'll keep | |
851 | * the gdquot from getting reclaimed. | |
1da177e4 | 852 | */ |
bf72de31 CH |
853 | gdqp = dqp->q_gdquot; |
854 | if (gdqp) { | |
855 | xfs_dqlock(gdqp); | |
856 | dqp->q_gdquot = NULL; | |
1da177e4 | 857 | } |
bf72de31 | 858 | xfs_dqunlock(dqp); |
1da177e4 | 859 | |
bf72de31 CH |
860 | /* |
861 | * If we had a group quota hint, release it now. | |
862 | */ | |
f8739c3c CH |
863 | if (gdqp) |
864 | xfs_qm_dqput(gdqp); | |
865 | } | |
866 | ||
867 | /* | |
868 | * Release a reference to the dquot (decrement ref-count) and unlock it. | |
869 | * | |
870 | * If there is a group quota attached to this dquot, carefully release that | |
871 | * too without tripping over deadlocks'n'stuff. | |
872 | */ | |
873 | void | |
874 | xfs_qm_dqput( | |
875 | struct xfs_dquot *dqp) | |
876 | { | |
877 | ASSERT(dqp->q_nrefs > 0); | |
878 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
879 | ||
880 | trace_xfs_dqput(dqp); | |
881 | ||
882 | if (--dqp->q_nrefs > 0) | |
883 | xfs_dqunlock(dqp); | |
884 | else | |
885 | xfs_qm_dqput_final(dqp); | |
1da177e4 LT |
886 | } |
887 | ||
888 | /* | |
889 | * Release a dquot. Flush it if dirty, then dqput() it. | |
890 | * dquot must not be locked. | |
891 | */ | |
892 | void | |
893 | xfs_qm_dqrele( | |
894 | xfs_dquot_t *dqp) | |
895 | { | |
7d095257 CH |
896 | if (!dqp) |
897 | return; | |
898 | ||
0b1b213f | 899 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
900 | |
901 | xfs_dqlock(dqp); | |
902 | /* | |
903 | * We don't care to flush it if the dquot is dirty here. | |
904 | * That will create stutters that we want to avoid. | |
905 | * Instead we do a delayed write when we try to reclaim | |
906 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
907 | */ | |
908 | xfs_qm_dqput(dqp); | |
909 | } | |
910 | ||
ca30b2a7 CH |
911 | /* |
912 | * This is the dquot flushing I/O completion routine. It is called | |
913 | * from interrupt level when the buffer containing the dquot is | |
914 | * flushed to disk. It is responsible for removing the dquot logitem | |
915 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
916 | * flush lock. This behavior is very similar to that of inodes.. | |
917 | */ | |
918 | STATIC void | |
919 | xfs_qm_dqflush_done( | |
920 | struct xfs_buf *bp, | |
921 | struct xfs_log_item *lip) | |
922 | { | |
923 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
924 | xfs_dquot_t *dqp = qip->qli_dquot; | |
925 | struct xfs_ail *ailp = lip->li_ailp; | |
926 | ||
927 | /* | |
928 | * We only want to pull the item from the AIL if its | |
929 | * location in the log has not changed since we started the flush. | |
930 | * Thus, we only bother if the dquot's lsn has | |
931 | * not changed. First we check the lsn outside the lock | |
932 | * since it's cheaper, and then we recheck while | |
933 | * holding the lock before removing the dquot from the AIL. | |
934 | */ | |
935 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
936 | lip->li_lsn == qip->qli_flush_lsn) { | |
937 | ||
938 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
939 | spin_lock(&ailp->xa_lock); | |
940 | if (lip->li_lsn == qip->qli_flush_lsn) | |
04913fdd | 941 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 CH |
942 | else |
943 | spin_unlock(&ailp->xa_lock); | |
944 | } | |
945 | ||
946 | /* | |
947 | * Release the dq's flush lock since we're done with it. | |
948 | */ | |
949 | xfs_dqfunlock(dqp); | |
950 | } | |
1da177e4 LT |
951 | |
952 | /* | |
953 | * Write a modified dquot to disk. | |
954 | * The dquot must be locked and the flush lock too taken by caller. | |
955 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
956 | * but the dquot is free to be unlocked and modified by the caller | |
957 | * in the interim. Dquot is still locked on return. This behavior is | |
958 | * identical to that of inodes. | |
959 | */ | |
960 | int | |
961 | xfs_qm_dqflush( | |
fe7257fd CH |
962 | struct xfs_dquot *dqp, |
963 | struct xfs_buf **bpp) | |
1da177e4 | 964 | { |
acecf1b5 CH |
965 | struct xfs_mount *mp = dqp->q_mount; |
966 | struct xfs_buf *bp; | |
967 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 968 | int error; |
1da177e4 LT |
969 | |
970 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 971 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 972 | |
0b1b213f | 973 | trace_xfs_dqflush(dqp); |
1da177e4 | 974 | |
fe7257fd CH |
975 | *bpp = NULL; |
976 | ||
1da177e4 LT |
977 | xfs_qm_dqunpin_wait(dqp); |
978 | ||
979 | /* | |
980 | * This may have been unpinned because the filesystem is shutting | |
981 | * down forcibly. If that's the case we must not write this dquot | |
dea96095 CH |
982 | * to disk, because the log record didn't make it to disk. |
983 | * | |
984 | * We also have to remove the log item from the AIL in this case, | |
985 | * as we wait for an emptry AIL as part of the unmount process. | |
1da177e4 | 986 | */ |
acecf1b5 | 987 | if (XFS_FORCED_SHUTDOWN(mp)) { |
dea96095 | 988 | struct xfs_log_item *lip = &dqp->q_logitem.qli_item; |
acecf1b5 | 989 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
dea96095 CH |
990 | |
991 | spin_lock(&mp->m_ail->xa_lock); | |
992 | if (lip->li_flags & XFS_LI_IN_AIL) | |
04913fdd DC |
993 | xfs_trans_ail_delete(mp->m_ail, lip, |
994 | SHUTDOWN_CORRUPT_INCORE); | |
dea96095 CH |
995 | else |
996 | spin_unlock(&mp->m_ail->xa_lock); | |
fe7257fd CH |
997 | error = XFS_ERROR(EIO); |
998 | goto out_unlock; | |
1da177e4 LT |
999 | } |
1000 | ||
1001 | /* | |
1002 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1003 | */ |
acecf1b5 | 1004 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
c3f8fc73 | 1005 | mp->m_quotainfo->qi_dqchunklen, 0, &bp, NULL); |
fe7257fd CH |
1006 | if (error) |
1007 | goto out_unlock; | |
1da177e4 | 1008 | |
acecf1b5 CH |
1009 | /* |
1010 | * Calculate the location of the dquot inside the buffer. | |
1011 | */ | |
62926044 | 1012 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1013 | |
1014 | /* | |
1015 | * A simple sanity check in case we got a corrupted dquot.. | |
1016 | */ | |
a0fa2b67 DC |
1017 | error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
1018 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); | |
1019 | if (error) { | |
acecf1b5 CH |
1020 | xfs_buf_relse(bp); |
1021 | xfs_dqfunlock(dqp); | |
1022 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
1da177e4 LT |
1023 | return XFS_ERROR(EIO); |
1024 | } | |
1025 | ||
1026 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1027 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1028 | |
1029 | /* | |
1030 | * Clear the dirty field and remember the flush lsn for later use. | |
1031 | */ | |
acecf1b5 | 1032 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1033 | |
7b2e2a31 DC |
1034 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1035 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 LT |
1036 | |
1037 | /* | |
1038 | * Attach an iodone routine so that we can remove this dquot from the | |
1039 | * AIL and release the flush lock once the dquot is synced to disk. | |
1040 | */ | |
ca30b2a7 CH |
1041 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1042 | &dqp->q_logitem.qli_item); | |
1043 | ||
1da177e4 LT |
1044 | /* |
1045 | * If the buffer is pinned then push on the log so we won't | |
1046 | * get stuck waiting in the write for too long. | |
1047 | */ | |
811e64c7 | 1048 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1049 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1050 | xfs_log_force(mp, 0); |
1da177e4 LT |
1051 | } |
1052 | ||
0b1b213f | 1053 | trace_xfs_dqflush_done(dqp); |
fe7257fd CH |
1054 | *bpp = bp; |
1055 | return 0; | |
0b1b213f | 1056 | |
fe7257fd CH |
1057 | out_unlock: |
1058 | xfs_dqfunlock(dqp); | |
1059 | return XFS_ERROR(EIO); | |
1da177e4 LT |
1060 | } |
1061 | ||
5bb87a33 CH |
1062 | /* |
1063 | * Lock two xfs_dquot structures. | |
1064 | * | |
1065 | * To avoid deadlocks we always lock the quota structure with | |
1066 | * the lowerd id first. | |
1067 | */ | |
1da177e4 LT |
1068 | void |
1069 | xfs_dqlock2( | |
1070 | xfs_dquot_t *d1, | |
1071 | xfs_dquot_t *d2) | |
1072 | { | |
1073 | if (d1 && d2) { | |
1074 | ASSERT(d1 != d2); | |
1149d96a CH |
1075 | if (be32_to_cpu(d1->q_core.d_id) > |
1076 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1077 | mutex_lock(&d2->q_qlock); |
1078 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1079 | } else { |
5bb87a33 CH |
1080 | mutex_lock(&d1->q_qlock); |
1081 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1082 | } |
5bb87a33 CH |
1083 | } else if (d1) { |
1084 | mutex_lock(&d1->q_qlock); | |
1085 | } else if (d2) { | |
1086 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1087 | } |
1088 | } | |
1089 | ||
a05931ce CH |
1090 | int __init |
1091 | xfs_qm_init(void) | |
1092 | { | |
1093 | xfs_qm_dqzone = | |
1094 | kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot"); | |
1095 | if (!xfs_qm_dqzone) | |
1096 | goto out; | |
1097 | ||
1098 | xfs_qm_dqtrxzone = | |
1099 | kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx"); | |
1100 | if (!xfs_qm_dqtrxzone) | |
1101 | goto out_free_dqzone; | |
1102 | ||
1103 | return 0; | |
1104 | ||
1105 | out_free_dqzone: | |
1106 | kmem_zone_destroy(xfs_qm_dqzone); | |
1107 | out: | |
1108 | return -ENOMEM; | |
1109 | } | |
1110 | ||
1c2ccc66 | 1111 | void |
a05931ce CH |
1112 | xfs_qm_exit(void) |
1113 | { | |
1114 | kmem_zone_destroy(xfs_qm_dqtrxzone); | |
1115 | kmem_zone_destroy(xfs_qm_dqzone); | |
1116 | } |