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 | 21 | #include "xfs_log.h" |
a844f451 | 22 | #include "xfs_inum.h" |
1da177e4 LT |
23 | #include "xfs_trans.h" |
24 | #include "xfs_sb.h" | |
25 | #include "xfs_ag.h" | |
1da177e4 | 26 | #include "xfs_alloc.h" |
1da177e4 LT |
27 | #include "xfs_quota.h" |
28 | #include "xfs_mount.h" | |
1da177e4 | 29 | #include "xfs_bmap_btree.h" |
1da177e4 LT |
30 | #include "xfs_inode.h" |
31 | #include "xfs_bmap.h" | |
1da177e4 LT |
32 | #include "xfs_rtalloc.h" |
33 | #include "xfs_error.h" | |
34 | #include "xfs_itable.h" | |
1da177e4 LT |
35 | #include "xfs_attr.h" |
36 | #include "xfs_buf_item.h" | |
37 | #include "xfs_trans_space.h" | |
38 | #include "xfs_trans_priv.h" | |
1da177e4 | 39 | #include "xfs_qm.h" |
0b1b213f | 40 | #include "xfs_trace.h" |
1da177e4 LT |
41 | |
42 | ||
43 | /* | |
44 | LOCK ORDER | |
45 | ||
46 | inode lock (ilock) | |
47 | dquot hash-chain lock (hashlock) | |
48 | xqm dquot freelist lock (freelistlock | |
49 | mount's dquot list lock (mplistlock) | |
50 | user dquot lock - lock ordering among dquots is based on the uid or gid | |
51 | group dquot lock - similar to udquots. Between the two dquots, the udquot | |
52 | has to be locked first. | |
53 | pin lock - the dquot lock must be held to take this lock. | |
54 | flush lock - ditto. | |
55 | */ | |
56 | ||
1da177e4 LT |
57 | #ifdef DEBUG |
58 | xfs_buftarg_t *xfs_dqerror_target; | |
59 | int xfs_do_dqerror; | |
60 | int xfs_dqreq_num; | |
61 | int xfs_dqerror_mod = 33; | |
62 | #endif | |
63 | ||
98b8c7a0 CH |
64 | static struct lock_class_key xfs_dquot_other_class; |
65 | ||
1da177e4 LT |
66 | /* |
67 | * Allocate and initialize a dquot. We don't always allocate fresh memory; | |
68 | * we try to reclaim a free dquot if the number of incore dquots are above | |
69 | * a threshold. | |
70 | * The only field inside the core that gets initialized at this point | |
71 | * is the d_id field. The idea is to fill in the entire q_core | |
72 | * when we read in the on disk dquot. | |
73 | */ | |
ba0f32d4 | 74 | STATIC xfs_dquot_t * |
1da177e4 LT |
75 | xfs_qm_dqinit( |
76 | xfs_mount_t *mp, | |
77 | xfs_dqid_t id, | |
78 | uint type) | |
79 | { | |
80 | xfs_dquot_t *dqp; | |
81 | boolean_t brandnewdquot; | |
82 | ||
83 | brandnewdquot = xfs_qm_dqalloc_incore(&dqp); | |
84 | dqp->dq_flags = type; | |
1149d96a | 85 | dqp->q_core.d_id = cpu_to_be32(id); |
1da177e4 LT |
86 | dqp->q_mount = mp; |
87 | ||
88 | /* | |
89 | * No need to re-initialize these if this is a reclaimed dquot. | |
90 | */ | |
91 | if (brandnewdquot) { | |
3a8406f6 | 92 | INIT_LIST_HEAD(&dqp->q_freelist); |
794ee1ba | 93 | mutex_init(&dqp->q_qlock); |
bc3048e3 | 94 | init_waitqueue_head(&dqp->q_pinwait); |
1da177e4 | 95 | |
e1f49cf2 DC |
96 | /* |
97 | * Because we want to use a counting completion, complete | |
98 | * the flush completion once to allow a single access to | |
99 | * the flush completion without blocking. | |
100 | */ | |
101 | init_completion(&dqp->q_flush); | |
102 | complete(&dqp->q_flush); | |
103 | ||
0b1b213f | 104 | trace_xfs_dqinit(dqp); |
1da177e4 LT |
105 | } else { |
106 | /* | |
107 | * Only the q_core portion was zeroed in dqreclaim_one(). | |
108 | * So, we need to reset others. | |
109 | */ | |
3a8406f6 DC |
110 | dqp->q_nrefs = 0; |
111 | dqp->q_blkno = 0; | |
112 | INIT_LIST_HEAD(&dqp->q_mplist); | |
113 | INIT_LIST_HEAD(&dqp->q_hashlist); | |
114 | dqp->q_bufoffset = 0; | |
115 | dqp->q_fileoffset = 0; | |
116 | dqp->q_transp = NULL; | |
117 | dqp->q_gdquot = NULL; | |
118 | dqp->q_res_bcount = 0; | |
119 | dqp->q_res_icount = 0; | |
120 | dqp->q_res_rtbcount = 0; | |
121 | atomic_set(&dqp->q_pincount, 0); | |
122 | dqp->q_hash = NULL; | |
123 | ASSERT(list_empty(&dqp->q_freelist)); | |
1da177e4 | 124 | |
0b1b213f | 125 | trace_xfs_dqreuse(dqp); |
98b8c7a0 CH |
126 | } |
127 | ||
128 | /* | |
129 | * In either case we need to make sure group quotas have a different | |
130 | * lock class than user quotas, to make sure lockdep knows we can | |
131 | * locks of one of each at the same time. | |
132 | */ | |
133 | if (!(type & XFS_DQ_USER)) | |
134 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class); | |
1da177e4 LT |
135 | |
136 | /* | |
137 | * log item gets initialized later | |
138 | */ | |
139 | return (dqp); | |
140 | } | |
141 | ||
142 | /* | |
143 | * This is called to free all the memory associated with a dquot | |
144 | */ | |
145 | void | |
146 | xfs_qm_dqdestroy( | |
147 | xfs_dquot_t *dqp) | |
148 | { | |
3a8406f6 | 149 | ASSERT(list_empty(&dqp->q_freelist)); |
1da177e4 LT |
150 | |
151 | mutex_destroy(&dqp->q_qlock); | |
1da177e4 | 152 | kmem_zone_free(xfs_Gqm->qm_dqzone, dqp); |
0b1b213f | 153 | |
1da177e4 LT |
154 | atomic_dec(&xfs_Gqm->qm_totaldquots); |
155 | } | |
156 | ||
157 | /* | |
158 | * This is what a 'fresh' dquot inside a dquot chunk looks like on disk. | |
159 | */ | |
160 | STATIC void | |
161 | xfs_qm_dqinit_core( | |
162 | xfs_dqid_t id, | |
163 | uint type, | |
164 | xfs_dqblk_t *d) | |
165 | { | |
166 | /* | |
167 | * Caller has zero'd the entire dquot 'chunk' already. | |
168 | */ | |
1149d96a CH |
169 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); |
170 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
171 | d->dd_diskdq.d_id = cpu_to_be32(id); | |
172 | d->dd_diskdq.d_flags = type; | |
1da177e4 LT |
173 | } |
174 | ||
1da177e4 LT |
175 | /* |
176 | * If default limits are in force, push them into the dquot now. | |
177 | * We overwrite the dquot limits only if they are zero and this | |
178 | * is not the root dquot. | |
179 | */ | |
180 | void | |
181 | xfs_qm_adjust_dqlimits( | |
182 | xfs_mount_t *mp, | |
183 | xfs_disk_dquot_t *d) | |
184 | { | |
185 | xfs_quotainfo_t *q = mp->m_quotainfo; | |
186 | ||
187 | ASSERT(d->d_id); | |
188 | ||
189 | if (q->qi_bsoftlimit && !d->d_blk_softlimit) | |
1149d96a | 190 | d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit); |
1da177e4 | 191 | if (q->qi_bhardlimit && !d->d_blk_hardlimit) |
1149d96a | 192 | d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit); |
1da177e4 | 193 | if (q->qi_isoftlimit && !d->d_ino_softlimit) |
1149d96a | 194 | d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit); |
1da177e4 | 195 | if (q->qi_ihardlimit && !d->d_ino_hardlimit) |
1149d96a | 196 | d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit); |
1da177e4 | 197 | if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit) |
1149d96a | 198 | d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit); |
1da177e4 | 199 | if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit) |
1149d96a | 200 | d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit); |
1da177e4 LT |
201 | } |
202 | ||
203 | /* | |
204 | * Check the limits and timers of a dquot and start or reset timers | |
205 | * if necessary. | |
206 | * This gets called even when quota enforcement is OFF, which makes our | |
207 | * life a little less complicated. (We just don't reject any quota | |
208 | * reservations in that case, when enforcement is off). | |
209 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
210 | * enforcement's off. | |
211 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
212 | * 'automatically' get started when limits get exceeded. They do |
213 | * get reset to zero, however, when we find the count to be under | |
214 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
215 | */ |
216 | void | |
217 | xfs_qm_adjust_dqtimers( | |
218 | xfs_mount_t *mp, | |
219 | xfs_disk_dquot_t *d) | |
220 | { | |
221 | ASSERT(d->d_id); | |
222 | ||
ea15ab3c | 223 | #ifdef DEBUG |
1149d96a CH |
224 | if (d->d_blk_hardlimit) |
225 | ASSERT(be64_to_cpu(d->d_blk_softlimit) <= | |
226 | be64_to_cpu(d->d_blk_hardlimit)); | |
227 | if (d->d_ino_hardlimit) | |
228 | ASSERT(be64_to_cpu(d->d_ino_softlimit) <= | |
229 | be64_to_cpu(d->d_ino_hardlimit)); | |
230 | if (d->d_rtb_hardlimit) | |
231 | ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= | |
232 | be64_to_cpu(d->d_rtb_hardlimit)); | |
1da177e4 | 233 | #endif |
ea15ab3c | 234 | |
1da177e4 | 235 | if (!d->d_btimer) { |
1149d96a CH |
236 | if ((d->d_blk_softlimit && |
237 | (be64_to_cpu(d->d_bcount) >= | |
238 | be64_to_cpu(d->d_blk_softlimit))) || | |
239 | (d->d_blk_hardlimit && | |
240 | (be64_to_cpu(d->d_bcount) >= | |
241 | be64_to_cpu(d->d_blk_hardlimit)))) { | |
242 | d->d_btimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 243 | mp->m_quotainfo->qi_btimelimit); |
754002b4 NS |
244 | } else { |
245 | d->d_bwarns = 0; | |
1da177e4 LT |
246 | } |
247 | } else { | |
248 | if ((!d->d_blk_softlimit || | |
1149d96a CH |
249 | (be64_to_cpu(d->d_bcount) < |
250 | be64_to_cpu(d->d_blk_softlimit))) && | |
1da177e4 | 251 | (!d->d_blk_hardlimit || |
1149d96a CH |
252 | (be64_to_cpu(d->d_bcount) < |
253 | be64_to_cpu(d->d_blk_hardlimit)))) { | |
1da177e4 LT |
254 | d->d_btimer = 0; |
255 | } | |
256 | } | |
257 | ||
258 | if (!d->d_itimer) { | |
1149d96a CH |
259 | if ((d->d_ino_softlimit && |
260 | (be64_to_cpu(d->d_icount) >= | |
261 | be64_to_cpu(d->d_ino_softlimit))) || | |
262 | (d->d_ino_hardlimit && | |
263 | (be64_to_cpu(d->d_icount) >= | |
264 | be64_to_cpu(d->d_ino_hardlimit)))) { | |
265 | d->d_itimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 266 | mp->m_quotainfo->qi_itimelimit); |
754002b4 NS |
267 | } else { |
268 | d->d_iwarns = 0; | |
1da177e4 LT |
269 | } |
270 | } else { | |
271 | if ((!d->d_ino_softlimit || | |
1149d96a CH |
272 | (be64_to_cpu(d->d_icount) < |
273 | be64_to_cpu(d->d_ino_softlimit))) && | |
1da177e4 | 274 | (!d->d_ino_hardlimit || |
1149d96a CH |
275 | (be64_to_cpu(d->d_icount) < |
276 | be64_to_cpu(d->d_ino_hardlimit)))) { | |
1da177e4 LT |
277 | d->d_itimer = 0; |
278 | } | |
279 | } | |
280 | ||
281 | if (!d->d_rtbtimer) { | |
1149d96a CH |
282 | if ((d->d_rtb_softlimit && |
283 | (be64_to_cpu(d->d_rtbcount) >= | |
284 | be64_to_cpu(d->d_rtb_softlimit))) || | |
285 | (d->d_rtb_hardlimit && | |
286 | (be64_to_cpu(d->d_rtbcount) >= | |
287 | be64_to_cpu(d->d_rtb_hardlimit)))) { | |
288 | d->d_rtbtimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 289 | mp->m_quotainfo->qi_rtbtimelimit); |
754002b4 NS |
290 | } else { |
291 | d->d_rtbwarns = 0; | |
1da177e4 LT |
292 | } |
293 | } else { | |
294 | if ((!d->d_rtb_softlimit || | |
1149d96a CH |
295 | (be64_to_cpu(d->d_rtbcount) < |
296 | be64_to_cpu(d->d_rtb_softlimit))) && | |
1da177e4 | 297 | (!d->d_rtb_hardlimit || |
1149d96a CH |
298 | (be64_to_cpu(d->d_rtbcount) < |
299 | be64_to_cpu(d->d_rtb_hardlimit)))) { | |
1da177e4 LT |
300 | d->d_rtbtimer = 0; |
301 | } | |
302 | } | |
303 | } | |
304 | ||
1da177e4 LT |
305 | /* |
306 | * initialize a buffer full of dquots and log the whole thing | |
307 | */ | |
308 | STATIC void | |
309 | xfs_qm_init_dquot_blk( | |
310 | xfs_trans_t *tp, | |
311 | xfs_mount_t *mp, | |
312 | xfs_dqid_t id, | |
313 | uint type, | |
314 | xfs_buf_t *bp) | |
315 | { | |
8a7b8a89 | 316 | struct xfs_quotainfo *q = mp->m_quotainfo; |
1da177e4 LT |
317 | xfs_dqblk_t *d; |
318 | int curid, i; | |
319 | ||
320 | ASSERT(tp); | |
0c842ad4 | 321 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 322 | |
62926044 | 323 | d = bp->b_addr; |
1da177e4 LT |
324 | |
325 | /* | |
326 | * ID of the first dquot in the block - id's are zero based. | |
327 | */ | |
8a7b8a89 | 328 | curid = id - (id % q->qi_dqperchunk); |
1da177e4 | 329 | ASSERT(curid >= 0); |
8a7b8a89 CH |
330 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
331 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) | |
1da177e4 LT |
332 | xfs_qm_dqinit_core(curid, type, d); |
333 | xfs_trans_dquot_buf(tp, bp, | |
c1155410 DC |
334 | (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : |
335 | ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : | |
336 | XFS_BLF_GDQUOT_BUF))); | |
8a7b8a89 | 337 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); |
1da177e4 LT |
338 | } |
339 | ||
340 | ||
341 | ||
342 | /* | |
343 | * Allocate a block and fill it with dquots. | |
344 | * This is called when the bmapi finds a hole. | |
345 | */ | |
346 | STATIC int | |
347 | xfs_qm_dqalloc( | |
efa092f3 | 348 | xfs_trans_t **tpp, |
1da177e4 LT |
349 | xfs_mount_t *mp, |
350 | xfs_dquot_t *dqp, | |
351 | xfs_inode_t *quotip, | |
352 | xfs_fileoff_t offset_fsb, | |
353 | xfs_buf_t **O_bpp) | |
354 | { | |
355 | xfs_fsblock_t firstblock; | |
356 | xfs_bmap_free_t flist; | |
357 | xfs_bmbt_irec_t map; | |
358 | int nmaps, error, committed; | |
359 | xfs_buf_t *bp; | |
efa092f3 | 360 | xfs_trans_t *tp = *tpp; |
1da177e4 LT |
361 | |
362 | ASSERT(tp != NULL); | |
0b1b213f CH |
363 | |
364 | trace_xfs_dqalloc(dqp); | |
1da177e4 LT |
365 | |
366 | /* | |
367 | * Initialize the bmap freelist prior to calling bmapi code. | |
368 | */ | |
9d87c319 | 369 | xfs_bmap_init(&flist, &firstblock); |
1da177e4 LT |
370 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
371 | /* | |
372 | * Return if this type of quotas is turned off while we didn't | |
373 | * have an inode lock | |
374 | */ | |
375 | if (XFS_IS_THIS_QUOTA_OFF(dqp)) { | |
376 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
377 | return (ESRCH); | |
378 | } | |
379 | ||
898621d5 | 380 | xfs_trans_ijoin_ref(tp, quotip, XFS_ILOCK_EXCL); |
1da177e4 LT |
381 | nmaps = 1; |
382 | if ((error = xfs_bmapi(tp, quotip, | |
383 | offset_fsb, XFS_DQUOT_CLUSTER_SIZE_FSB, | |
384 | XFS_BMAPI_METADATA | XFS_BMAPI_WRITE, | |
385 | &firstblock, | |
386 | XFS_QM_DQALLOC_SPACE_RES(mp), | |
b4e9181e | 387 | &map, &nmaps, &flist))) { |
1da177e4 LT |
388 | goto error0; |
389 | } | |
390 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); | |
391 | ASSERT(nmaps == 1); | |
392 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
393 | (map.br_startblock != HOLESTARTBLOCK)); | |
394 | ||
395 | /* | |
396 | * Keep track of the blkno to save a lookup later | |
397 | */ | |
398 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
399 | ||
400 | /* now we can just get the buffer (there's nothing to read yet) */ | |
401 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
402 | dqp->q_blkno, | |
8a7b8a89 | 403 | mp->m_quotainfo->qi_dqchunklen, |
1da177e4 | 404 | 0); |
2a30f36d CS |
405 | |
406 | error = xfs_buf_geterror(bp); | |
407 | if (error) | |
1da177e4 | 408 | goto error1; |
2a30f36d | 409 | |
1da177e4 LT |
410 | /* |
411 | * Make a chunk of dquots out of this buffer and log | |
412 | * the entire thing. | |
413 | */ | |
1149d96a | 414 | xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 415 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
1da177e4 | 416 | |
efa092f3 TS |
417 | /* |
418 | * xfs_bmap_finish() may commit the current transaction and | |
419 | * start a second transaction if the freelist is not empty. | |
420 | * | |
421 | * Since we still want to modify this buffer, we need to | |
422 | * ensure that the buffer is not released on commit of | |
423 | * the first transaction and ensure the buffer is added to the | |
424 | * second transaction. | |
425 | * | |
426 | * If there is only one transaction then don't stop the buffer | |
427 | * from being released when it commits later on. | |
428 | */ | |
429 | ||
430 | xfs_trans_bhold(tp, bp); | |
431 | ||
f7c99b6f | 432 | if ((error = xfs_bmap_finish(tpp, &flist, &committed))) { |
1da177e4 LT |
433 | goto error1; |
434 | } | |
435 | ||
efa092f3 TS |
436 | if (committed) { |
437 | tp = *tpp; | |
438 | xfs_trans_bjoin(tp, bp); | |
439 | } else { | |
440 | xfs_trans_bhold_release(tp, bp); | |
441 | } | |
442 | ||
1da177e4 LT |
443 | *O_bpp = bp; |
444 | return 0; | |
445 | ||
446 | error1: | |
447 | xfs_bmap_cancel(&flist); | |
448 | error0: | |
449 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
450 | ||
451 | return (error); | |
452 | } | |
453 | ||
454 | /* | |
455 | * Maps a dquot to the buffer containing its on-disk version. | |
456 | * This returns a ptr to the buffer containing the on-disk dquot | |
457 | * in the bpp param, and a ptr to the on-disk dquot within that buffer | |
458 | */ | |
459 | STATIC int | |
460 | xfs_qm_dqtobp( | |
efa092f3 | 461 | xfs_trans_t **tpp, |
1da177e4 LT |
462 | xfs_dquot_t *dqp, |
463 | xfs_disk_dquot_t **O_ddpp, | |
464 | xfs_buf_t **O_bpp, | |
465 | uint flags) | |
466 | { | |
467 | xfs_bmbt_irec_t map; | |
acecf1b5 | 468 | int nmaps = 1, error; |
1da177e4 | 469 | xfs_buf_t *bp; |
acecf1b5 CH |
470 | xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp); |
471 | xfs_mount_t *mp = dqp->q_mount; | |
1da177e4 | 472 | xfs_disk_dquot_t *ddq; |
acecf1b5 | 473 | xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); |
efa092f3 | 474 | xfs_trans_t *tp = (tpp ? *tpp : NULL); |
1da177e4 | 475 | |
acecf1b5 | 476 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
1da177e4 | 477 | |
acecf1b5 CH |
478 | xfs_ilock(quotip, XFS_ILOCK_SHARED); |
479 | if (XFS_IS_THIS_QUOTA_OFF(dqp)) { | |
1da177e4 | 480 | /* |
acecf1b5 CH |
481 | * Return if this type of quotas is turned off while we |
482 | * didn't have the quota inode lock. | |
1da177e4 | 483 | */ |
acecf1b5 CH |
484 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); |
485 | return ESRCH; | |
486 | } | |
487 | ||
488 | /* | |
489 | * Find the block map; no allocations yet | |
490 | */ | |
491 | error = xfs_bmapi(NULL, quotip, dqp->q_fileoffset, | |
492 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
493 | NULL, 0, &map, &nmaps, NULL); | |
494 | ||
495 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); | |
496 | if (error) | |
497 | return error; | |
498 | ||
499 | ASSERT(nmaps == 1); | |
500 | ASSERT(map.br_blockcount == 1); | |
501 | ||
502 | /* | |
503 | * Offset of dquot in the (fixed sized) dquot chunk. | |
504 | */ | |
505 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
506 | sizeof(xfs_dqblk_t); | |
507 | ||
508 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
509 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 510 | /* |
acecf1b5 | 511 | * We don't allocate unless we're asked to |
1da177e4 | 512 | */ |
acecf1b5 CH |
513 | if (!(flags & XFS_QMOPT_DQALLOC)) |
514 | return ENOENT; | |
1da177e4 | 515 | |
acecf1b5 CH |
516 | ASSERT(tp); |
517 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
518 | dqp->q_fileoffset, &bp); | |
1da177e4 | 519 | if (error) |
acecf1b5 CH |
520 | return error; |
521 | tp = *tpp; | |
522 | } else { | |
523 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
524 | |
525 | /* | |
acecf1b5 CH |
526 | * store the blkno etc so that we don't have to do the |
527 | * mapping all the time | |
1da177e4 | 528 | */ |
acecf1b5 | 529 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 530 | |
8a7b8a89 CH |
531 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
532 | dqp->q_blkno, | |
533 | mp->m_quotainfo->qi_dqchunklen, | |
534 | 0, &bp); | |
1da177e4 LT |
535 | if (error || !bp) |
536 | return XFS_ERROR(error); | |
537 | } | |
acecf1b5 | 538 | |
0c842ad4 | 539 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
540 | |
541 | /* | |
542 | * calculate the location of the dquot inside the buffer. | |
543 | */ | |
62926044 | 544 | ddq = bp->b_addr + dqp->q_bufoffset; |
1da177e4 LT |
545 | |
546 | /* | |
547 | * A simple sanity check in case we got a corrupted dquot... | |
548 | */ | |
a0fa2b67 | 549 | error = xfs_qm_dqcheck(mp, ddq, id, dqp->dq_flags & XFS_DQ_ALLTYPES, |
1da177e4 | 550 | flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN), |
a0fa2b67 DC |
551 | "dqtobp"); |
552 | if (error) { | |
1da177e4 LT |
553 | if (!(flags & XFS_QMOPT_DQREPAIR)) { |
554 | xfs_trans_brelse(tp, bp); | |
555 | return XFS_ERROR(EIO); | |
556 | } | |
1da177e4 LT |
557 | } |
558 | ||
559 | *O_bpp = bp; | |
560 | *O_ddpp = ddq; | |
561 | ||
562 | return (0); | |
563 | } | |
564 | ||
565 | ||
566 | /* | |
567 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
568 | * and release the buffer immediately. | |
569 | * | |
570 | */ | |
571 | /* ARGSUSED */ | |
572 | STATIC int | |
573 | xfs_qm_dqread( | |
efa092f3 | 574 | xfs_trans_t **tpp, |
1da177e4 LT |
575 | xfs_dqid_t id, |
576 | xfs_dquot_t *dqp, /* dquot to get filled in */ | |
577 | uint flags) | |
578 | { | |
579 | xfs_disk_dquot_t *ddqp; | |
580 | xfs_buf_t *bp; | |
581 | int error; | |
efa092f3 TS |
582 | xfs_trans_t *tp; |
583 | ||
584 | ASSERT(tpp); | |
1da177e4 | 585 | |
0b1b213f CH |
586 | trace_xfs_dqread(dqp); |
587 | ||
1da177e4 LT |
588 | /* |
589 | * get a pointer to the on-disk dquot and the buffer containing it | |
590 | * dqp already knows its own type (GROUP/USER). | |
591 | */ | |
efa092f3 | 592 | if ((error = xfs_qm_dqtobp(tpp, dqp, &ddqp, &bp, flags))) { |
1da177e4 LT |
593 | return (error); |
594 | } | |
efa092f3 | 595 | tp = *tpp; |
1da177e4 LT |
596 | |
597 | /* copy everything from disk dquot to the incore dquot */ | |
598 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
1149d96a | 599 | ASSERT(be32_to_cpu(dqp->q_core.d_id) == id); |
1da177e4 LT |
600 | xfs_qm_dquot_logitem_init(dqp); |
601 | ||
602 | /* | |
603 | * Reservation counters are defined as reservation plus current usage | |
25985edc | 604 | * to avoid having to add every time. |
1da177e4 | 605 | */ |
1149d96a CH |
606 | dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); |
607 | dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); | |
608 | dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); | |
1da177e4 LT |
609 | |
610 | /* Mark the buf so that this will stay incore a little longer */ | |
611 | XFS_BUF_SET_VTYPE_REF(bp, B_FS_DQUOT, XFS_DQUOT_REF); | |
612 | ||
613 | /* | |
614 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
615 | * So we need to release with xfs_trans_brelse(). | |
616 | * The strategy here is identical to that of inodes; we lock | |
617 | * the dquot in xfs_qm_dqget() before making it accessible to | |
618 | * others. This is because dquots, like inodes, need a good level of | |
619 | * concurrency, and we don't want to take locks on the entire buffers | |
620 | * for dquot accesses. | |
621 | * Note also that the dquot buffer may even be dirty at this point, if | |
622 | * this particular dquot was repaired. We still aren't afraid to | |
623 | * brelse it because we have the changes incore. | |
624 | */ | |
0c842ad4 | 625 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
626 | xfs_trans_brelse(tp, bp); |
627 | ||
628 | return (error); | |
629 | } | |
630 | ||
631 | ||
632 | /* | |
633 | * allocate an incore dquot from the kernel heap, | |
634 | * and fill its core with quota information kept on disk. | |
635 | * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk | |
636 | * if it wasn't already allocated. | |
637 | */ | |
638 | STATIC int | |
639 | xfs_qm_idtodq( | |
640 | xfs_mount_t *mp, | |
641 | xfs_dqid_t id, /* gid or uid, depending on type */ | |
642 | uint type, /* UDQUOT or GDQUOT */ | |
643 | uint flags, /* DQALLOC, DQREPAIR */ | |
644 | xfs_dquot_t **O_dqpp)/* OUT : incore dquot, not locked */ | |
645 | { | |
646 | xfs_dquot_t *dqp; | |
647 | int error; | |
648 | xfs_trans_t *tp; | |
649 | int cancelflags=0; | |
650 | ||
651 | dqp = xfs_qm_dqinit(mp, id, type); | |
652 | tp = NULL; | |
653 | if (flags & XFS_QMOPT_DQALLOC) { | |
654 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC); | |
8a7b8a89 CH |
655 | error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp), |
656 | XFS_WRITE_LOG_RES(mp) + | |
657 | BBTOB(mp->m_quotainfo->qi_dqchunklen) - 1 + | |
658 | 128, | |
659 | 0, | |
660 | XFS_TRANS_PERM_LOG_RES, | |
661 | XFS_WRITE_LOG_COUNT); | |
662 | if (error) { | |
1da177e4 LT |
663 | cancelflags = 0; |
664 | goto error0; | |
665 | } | |
666 | cancelflags = XFS_TRANS_RELEASE_LOG_RES; | |
667 | } | |
668 | ||
669 | /* | |
670 | * Read it from disk; xfs_dqread() takes care of | |
671 | * all the necessary initialization of dquot's fields (locks, etc) | |
672 | */ | |
efa092f3 | 673 | if ((error = xfs_qm_dqread(&tp, id, dqp, flags))) { |
1da177e4 LT |
674 | /* |
675 | * This can happen if quotas got turned off (ESRCH), | |
676 | * or if the dquot didn't exist on disk and we ask to | |
677 | * allocate (ENOENT). | |
678 | */ | |
0b1b213f | 679 | trace_xfs_dqread_fail(dqp); |
1da177e4 LT |
680 | cancelflags |= XFS_TRANS_ABORT; |
681 | goto error0; | |
682 | } | |
683 | if (tp) { | |
1c72bf90 | 684 | if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) |
1da177e4 LT |
685 | goto error1; |
686 | } | |
687 | ||
688 | *O_dqpp = dqp; | |
689 | return (0); | |
690 | ||
691 | error0: | |
692 | ASSERT(error); | |
693 | if (tp) | |
694 | xfs_trans_cancel(tp, cancelflags); | |
695 | error1: | |
696 | xfs_qm_dqdestroy(dqp); | |
697 | *O_dqpp = NULL; | |
698 | return (error); | |
699 | } | |
700 | ||
701 | /* | |
702 | * Lookup a dquot in the incore dquot hashtable. We keep two separate | |
703 | * hashtables for user and group dquots; and, these are global tables | |
704 | * inside the XQM, not per-filesystem tables. | |
705 | * The hash chain must be locked by caller, and it is left locked | |
706 | * on return. Returning dquot is locked. | |
707 | */ | |
708 | STATIC int | |
709 | xfs_qm_dqlookup( | |
710 | xfs_mount_t *mp, | |
711 | xfs_dqid_t id, | |
712 | xfs_dqhash_t *qh, | |
713 | xfs_dquot_t **O_dqpp) | |
714 | { | |
715 | xfs_dquot_t *dqp; | |
716 | uint flist_locked; | |
1da177e4 | 717 | |
c9a192dc | 718 | ASSERT(mutex_is_locked(&qh->qh_lock)); |
1da177e4 LT |
719 | |
720 | flist_locked = B_FALSE; | |
721 | ||
722 | /* | |
723 | * Traverse the hashchain looking for a match | |
724 | */ | |
e6a81f13 | 725 | list_for_each_entry(dqp, &qh->qh_list, q_hashlist) { |
1da177e4 LT |
726 | /* |
727 | * We already have the hashlock. We don't need the | |
728 | * dqlock to look at the id field of the dquot, since the | |
729 | * id can't be modified without the hashlock anyway. | |
730 | */ | |
1149d96a | 731 | if (be32_to_cpu(dqp->q_core.d_id) == id && dqp->q_mount == mp) { |
0b1b213f CH |
732 | trace_xfs_dqlookup_found(dqp); |
733 | ||
1da177e4 LT |
734 | /* |
735 | * All in core dquots must be on the dqlist of mp | |
736 | */ | |
3a25404b | 737 | ASSERT(!list_empty(&dqp->q_mplist)); |
1da177e4 LT |
738 | |
739 | xfs_dqlock(dqp); | |
740 | if (dqp->q_nrefs == 0) { | |
3a8406f6 DC |
741 | ASSERT(!list_empty(&dqp->q_freelist)); |
742 | if (!mutex_trylock(&xfs_Gqm->qm_dqfrlist_lock)) { | |
0b1b213f | 743 | trace_xfs_dqlookup_want(dqp); |
1da177e4 LT |
744 | |
745 | /* | |
746 | * We may have raced with dqreclaim_one() | |
747 | * (and lost). So, flag that we don't | |
748 | * want the dquot to be reclaimed. | |
749 | */ | |
750 | dqp->dq_flags |= XFS_DQ_WANT; | |
751 | xfs_dqunlock(dqp); | |
3a8406f6 | 752 | mutex_lock(&xfs_Gqm->qm_dqfrlist_lock); |
1da177e4 LT |
753 | xfs_dqlock(dqp); |
754 | dqp->dq_flags &= ~(XFS_DQ_WANT); | |
755 | } | |
756 | flist_locked = B_TRUE; | |
757 | } | |
758 | ||
759 | /* | |
760 | * id couldn't have changed; we had the hashlock all | |
761 | * along | |
762 | */ | |
1149d96a | 763 | ASSERT(be32_to_cpu(dqp->q_core.d_id) == id); |
1da177e4 LT |
764 | |
765 | if (flist_locked) { | |
766 | if (dqp->q_nrefs != 0) { | |
3a8406f6 | 767 | mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); |
1da177e4 LT |
768 | flist_locked = B_FALSE; |
769 | } else { | |
3a8406f6 | 770 | /* take it off the freelist */ |
0b1b213f | 771 | trace_xfs_dqlookup_freelist(dqp); |
3a8406f6 DC |
772 | list_del_init(&dqp->q_freelist); |
773 | xfs_Gqm->qm_dqfrlist_cnt--; | |
1da177e4 LT |
774 | } |
775 | } | |
776 | ||
1da177e4 LT |
777 | XFS_DQHOLD(dqp); |
778 | ||
779 | if (flist_locked) | |
3a8406f6 | 780 | mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); |
1da177e4 LT |
781 | /* |
782 | * move the dquot to the front of the hashchain | |
783 | */ | |
c9a192dc | 784 | ASSERT(mutex_is_locked(&qh->qh_lock)); |
e6a81f13 | 785 | list_move(&dqp->q_hashlist, &qh->qh_list); |
0b1b213f | 786 | trace_xfs_dqlookup_done(dqp); |
1da177e4 | 787 | *O_dqpp = dqp; |
e6a81f13 | 788 | return 0; |
1da177e4 LT |
789 | } |
790 | } | |
791 | ||
792 | *O_dqpp = NULL; | |
c9a192dc | 793 | ASSERT(mutex_is_locked(&qh->qh_lock)); |
1da177e4 LT |
794 | return (1); |
795 | } | |
796 | ||
797 | /* | |
798 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
799 | * a locked dquot, doing an allocation (if requested) as needed. | |
800 | * When both an inode and an id are given, the inode's id takes precedence. | |
801 | * That is, if the id changes while we don't hold the ilock inside this | |
802 | * function, the new dquot is returned, not necessarily the one requested | |
803 | * in the id argument. | |
804 | */ | |
805 | int | |
806 | xfs_qm_dqget( | |
807 | xfs_mount_t *mp, | |
808 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
809 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
810 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
811 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
812 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
813 | { | |
814 | xfs_dquot_t *dqp; | |
815 | xfs_dqhash_t *h; | |
816 | uint version; | |
817 | int error; | |
818 | ||
819 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
820 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 821 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 LT |
822 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
823 | return (ESRCH); | |
824 | } | |
825 | h = XFS_DQ_HASH(mp, id, type); | |
826 | ||
827 | #ifdef DEBUG | |
828 | if (xfs_do_dqerror) { | |
829 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
830 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 831 | xfs_debug(mp, "Returning error in dqget"); |
1da177e4 LT |
832 | return (EIO); |
833 | } | |
834 | } | |
835 | #endif | |
836 | ||
837 | again: | |
838 | ||
839 | #ifdef DEBUG | |
c8ad20ff NS |
840 | ASSERT(type == XFS_DQ_USER || |
841 | type == XFS_DQ_PROJ || | |
842 | type == XFS_DQ_GROUP); | |
1da177e4 | 843 | if (ip) { |
579aa9ca | 844 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
845 | if (type == XFS_DQ_USER) |
846 | ASSERT(ip->i_udquot == NULL); | |
847 | else | |
848 | ASSERT(ip->i_gdquot == NULL); | |
849 | } | |
850 | #endif | |
c9a192dc | 851 | mutex_lock(&h->qh_lock); |
1da177e4 LT |
852 | |
853 | /* | |
854 | * Look in the cache (hashtable). | |
855 | * The chain is kept locked during lookup. | |
856 | */ | |
857 | if (xfs_qm_dqlookup(mp, id, h, O_dqpp) == 0) { | |
858 | XQM_STATS_INC(xqmstats.xs_qm_dqcachehits); | |
859 | /* | |
860 | * The dquot was found, moved to the front of the chain, | |
861 | * taken off the freelist if it was on it, and locked | |
862 | * at this point. Just unlock the hashchain and return. | |
863 | */ | |
864 | ASSERT(*O_dqpp); | |
865 | ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp)); | |
c9a192dc | 866 | mutex_unlock(&h->qh_lock); |
0b1b213f | 867 | trace_xfs_dqget_hit(*O_dqpp); |
1da177e4 LT |
868 | return (0); /* success */ |
869 | } | |
870 | XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses); | |
871 | ||
872 | /* | |
873 | * Dquot cache miss. We don't want to keep the inode lock across | |
874 | * a (potential) disk read. Also we don't want to deal with the lock | |
875 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
876 | * lock here means dealing with a chown that can happen before | |
877 | * we re-acquire the lock. | |
878 | */ | |
879 | if (ip) | |
880 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
881 | /* | |
882 | * Save the hashchain version stamp, and unlock the chain, so that | |
883 | * we don't keep the lock across a disk read | |
884 | */ | |
885 | version = h->qh_version; | |
c9a192dc | 886 | mutex_unlock(&h->qh_lock); |
1da177e4 LT |
887 | |
888 | /* | |
889 | * Allocate the dquot on the kernel heap, and read the ondisk | |
890 | * portion off the disk. Also, do all the necessary initialization | |
891 | * This can return ENOENT if dquot didn't exist on disk and we didn't | |
892 | * ask it to allocate; ESRCH if quotas got turned off suddenly. | |
893 | */ | |
894 | if ((error = xfs_qm_idtodq(mp, id, type, | |
895 | flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR| | |
896 | XFS_QMOPT_DOWARN), | |
897 | &dqp))) { | |
898 | if (ip) | |
899 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
900 | return (error); | |
901 | } | |
902 | ||
903 | /* | |
904 | * See if this is mount code calling to look at the overall quota limits | |
905 | * which are stored in the id == 0 user or group's dquot. | |
906 | * Since we may not have done a quotacheck by this point, just return | |
907 | * the dquot without attaching it to any hashtables, lists, etc, or even | |
908 | * taking a reference. | |
909 | * The caller must dqdestroy this once done. | |
910 | */ | |
911 | if (flags & XFS_QMOPT_DQSUSER) { | |
912 | ASSERT(id == 0); | |
913 | ASSERT(! ip); | |
914 | goto dqret; | |
915 | } | |
916 | ||
917 | /* | |
918 | * Dquot lock comes after hashlock in the lock ordering | |
919 | */ | |
920 | if (ip) { | |
921 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
191f8488 | 922 | |
1da177e4 LT |
923 | /* |
924 | * A dquot could be attached to this inode by now, since | |
925 | * we had dropped the ilock. | |
926 | */ | |
927 | if (type == XFS_DQ_USER) { | |
191f8488 CH |
928 | if (!XFS_IS_UQUOTA_ON(mp)) { |
929 | /* inode stays locked on return */ | |
930 | xfs_qm_dqdestroy(dqp); | |
931 | return XFS_ERROR(ESRCH); | |
932 | } | |
1da177e4 LT |
933 | if (ip->i_udquot) { |
934 | xfs_qm_dqdestroy(dqp); | |
935 | dqp = ip->i_udquot; | |
936 | xfs_dqlock(dqp); | |
937 | goto dqret; | |
938 | } | |
939 | } else { | |
191f8488 CH |
940 | if (!XFS_IS_OQUOTA_ON(mp)) { |
941 | /* inode stays locked on return */ | |
942 | xfs_qm_dqdestroy(dqp); | |
943 | return XFS_ERROR(ESRCH); | |
944 | } | |
1da177e4 LT |
945 | if (ip->i_gdquot) { |
946 | xfs_qm_dqdestroy(dqp); | |
947 | dqp = ip->i_gdquot; | |
948 | xfs_dqlock(dqp); | |
949 | goto dqret; | |
950 | } | |
951 | } | |
952 | } | |
953 | ||
954 | /* | |
955 | * Hashlock comes after ilock in lock order | |
956 | */ | |
c9a192dc | 957 | mutex_lock(&h->qh_lock); |
1da177e4 LT |
958 | if (version != h->qh_version) { |
959 | xfs_dquot_t *tmpdqp; | |
960 | /* | |
961 | * Now, see if somebody else put the dquot in the | |
962 | * hashtable before us. This can happen because we didn't | |
963 | * keep the hashchain lock. We don't have to worry about | |
964 | * lock order between the two dquots here since dqp isn't | |
965 | * on any findable lists yet. | |
966 | */ | |
967 | if (xfs_qm_dqlookup(mp, id, h, &tmpdqp) == 0) { | |
968 | /* | |
969 | * Duplicate found. Just throw away the new dquot | |
970 | * and start over. | |
971 | */ | |
972 | xfs_qm_dqput(tmpdqp); | |
c9a192dc | 973 | mutex_unlock(&h->qh_lock); |
1da177e4 LT |
974 | xfs_qm_dqdestroy(dqp); |
975 | XQM_STATS_INC(xqmstats.xs_qm_dquot_dups); | |
976 | goto again; | |
977 | } | |
978 | } | |
979 | ||
980 | /* | |
981 | * Put the dquot at the beginning of the hash-chain and mp's list | |
982 | * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock .. | |
983 | */ | |
c9a192dc | 984 | ASSERT(mutex_is_locked(&h->qh_lock)); |
1da177e4 | 985 | dqp->q_hash = h; |
e6a81f13 DC |
986 | list_add(&dqp->q_hashlist, &h->qh_list); |
987 | h->qh_version++; | |
1da177e4 LT |
988 | |
989 | /* | |
990 | * Attach this dquot to this filesystem's list of all dquots, | |
991 | * kept inside the mount structure in m_quotainfo field | |
992 | */ | |
3a25404b | 993 | mutex_lock(&mp->m_quotainfo->qi_dqlist_lock); |
1da177e4 LT |
994 | |
995 | /* | |
996 | * We return a locked dquot to the caller, with a reference taken | |
997 | */ | |
998 | xfs_dqlock(dqp); | |
999 | dqp->q_nrefs = 1; | |
1000 | ||
3a25404b DC |
1001 | list_add(&dqp->q_mplist, &mp->m_quotainfo->qi_dqlist); |
1002 | mp->m_quotainfo->qi_dquots++; | |
1003 | mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock); | |
c9a192dc | 1004 | mutex_unlock(&h->qh_lock); |
1da177e4 | 1005 | dqret: |
579aa9ca | 1006 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 1007 | trace_xfs_dqget_miss(dqp); |
1da177e4 LT |
1008 | *O_dqpp = dqp; |
1009 | return (0); | |
1010 | } | |
1011 | ||
1012 | ||
1013 | /* | |
1014 | * Release a reference to the dquot (decrement ref-count) | |
1015 | * and unlock it. If there is a group quota attached to this | |
1016 | * dquot, carefully release that too without tripping over | |
1017 | * deadlocks'n'stuff. | |
1018 | */ | |
1019 | void | |
1020 | xfs_qm_dqput( | |
1021 | xfs_dquot_t *dqp) | |
1022 | { | |
1023 | xfs_dquot_t *gdqp; | |
1024 | ||
1025 | ASSERT(dqp->q_nrefs > 0); | |
1026 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
0b1b213f CH |
1027 | |
1028 | trace_xfs_dqput(dqp); | |
1da177e4 LT |
1029 | |
1030 | if (dqp->q_nrefs != 1) { | |
1031 | dqp->q_nrefs--; | |
1032 | xfs_dqunlock(dqp); | |
1033 | return; | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * drop the dqlock and acquire the freelist and dqlock | |
1038 | * in the right order; but try to get it out-of-order first | |
1039 | */ | |
3a8406f6 | 1040 | if (!mutex_trylock(&xfs_Gqm->qm_dqfrlist_lock)) { |
0b1b213f | 1041 | trace_xfs_dqput_wait(dqp); |
1da177e4 | 1042 | xfs_dqunlock(dqp); |
3a8406f6 | 1043 | mutex_lock(&xfs_Gqm->qm_dqfrlist_lock); |
1da177e4 LT |
1044 | xfs_dqlock(dqp); |
1045 | } | |
1046 | ||
1047 | while (1) { | |
1048 | gdqp = NULL; | |
1049 | ||
1050 | /* We can't depend on nrefs being == 1 here */ | |
1051 | if (--dqp->q_nrefs == 0) { | |
0b1b213f CH |
1052 | trace_xfs_dqput_free(dqp); |
1053 | ||
3a8406f6 DC |
1054 | list_add_tail(&dqp->q_freelist, &xfs_Gqm->qm_dqfrlist); |
1055 | xfs_Gqm->qm_dqfrlist_cnt++; | |
1da177e4 LT |
1056 | |
1057 | /* | |
1058 | * If we just added a udquot to the freelist, then | |
1059 | * we want to release the gdquot reference that | |
1060 | * it (probably) has. Otherwise it'll keep the | |
1061 | * gdquot from getting reclaimed. | |
1062 | */ | |
1063 | if ((gdqp = dqp->q_gdquot)) { | |
1064 | /* | |
1065 | * Avoid a recursive dqput call | |
1066 | */ | |
1067 | xfs_dqlock(gdqp); | |
1068 | dqp->q_gdquot = NULL; | |
1069 | } | |
1da177e4 LT |
1070 | } |
1071 | xfs_dqunlock(dqp); | |
1072 | ||
1073 | /* | |
1074 | * If we had a group quota inside the user quota as a hint, | |
1075 | * release it now. | |
1076 | */ | |
1077 | if (! gdqp) | |
1078 | break; | |
1079 | dqp = gdqp; | |
1080 | } | |
3a8406f6 | 1081 | mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); |
1da177e4 LT |
1082 | } |
1083 | ||
1084 | /* | |
1085 | * Release a dquot. Flush it if dirty, then dqput() it. | |
1086 | * dquot must not be locked. | |
1087 | */ | |
1088 | void | |
1089 | xfs_qm_dqrele( | |
1090 | xfs_dquot_t *dqp) | |
1091 | { | |
7d095257 CH |
1092 | if (!dqp) |
1093 | return; | |
1094 | ||
0b1b213f | 1095 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
1096 | |
1097 | xfs_dqlock(dqp); | |
1098 | /* | |
1099 | * We don't care to flush it if the dquot is dirty here. | |
1100 | * That will create stutters that we want to avoid. | |
1101 | * Instead we do a delayed write when we try to reclaim | |
1102 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
1103 | */ | |
1104 | xfs_qm_dqput(dqp); | |
1105 | } | |
1106 | ||
ca30b2a7 CH |
1107 | /* |
1108 | * This is the dquot flushing I/O completion routine. It is called | |
1109 | * from interrupt level when the buffer containing the dquot is | |
1110 | * flushed to disk. It is responsible for removing the dquot logitem | |
1111 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
1112 | * flush lock. This behavior is very similar to that of inodes.. | |
1113 | */ | |
1114 | STATIC void | |
1115 | xfs_qm_dqflush_done( | |
1116 | struct xfs_buf *bp, | |
1117 | struct xfs_log_item *lip) | |
1118 | { | |
1119 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
1120 | xfs_dquot_t *dqp = qip->qli_dquot; | |
1121 | struct xfs_ail *ailp = lip->li_ailp; | |
1122 | ||
1123 | /* | |
1124 | * We only want to pull the item from the AIL if its | |
1125 | * location in the log has not changed since we started the flush. | |
1126 | * Thus, we only bother if the dquot's lsn has | |
1127 | * not changed. First we check the lsn outside the lock | |
1128 | * since it's cheaper, and then we recheck while | |
1129 | * holding the lock before removing the dquot from the AIL. | |
1130 | */ | |
1131 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
1132 | lip->li_lsn == qip->qli_flush_lsn) { | |
1133 | ||
1134 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
1135 | spin_lock(&ailp->xa_lock); | |
1136 | if (lip->li_lsn == qip->qli_flush_lsn) | |
1137 | xfs_trans_ail_delete(ailp, lip); | |
1138 | else | |
1139 | spin_unlock(&ailp->xa_lock); | |
1140 | } | |
1141 | ||
1142 | /* | |
1143 | * Release the dq's flush lock since we're done with it. | |
1144 | */ | |
1145 | xfs_dqfunlock(dqp); | |
1146 | } | |
1da177e4 LT |
1147 | |
1148 | /* | |
1149 | * Write a modified dquot to disk. | |
1150 | * The dquot must be locked and the flush lock too taken by caller. | |
1151 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
1152 | * but the dquot is free to be unlocked and modified by the caller | |
1153 | * in the interim. Dquot is still locked on return. This behavior is | |
1154 | * identical to that of inodes. | |
1155 | */ | |
1156 | int | |
1157 | xfs_qm_dqflush( | |
1158 | xfs_dquot_t *dqp, | |
1159 | uint flags) | |
1160 | { | |
acecf1b5 CH |
1161 | struct xfs_mount *mp = dqp->q_mount; |
1162 | struct xfs_buf *bp; | |
1163 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 1164 | int error; |
1da177e4 LT |
1165 | |
1166 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 1167 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 1168 | |
0b1b213f | 1169 | trace_xfs_dqflush(dqp); |
1da177e4 LT |
1170 | |
1171 | /* | |
acecf1b5 | 1172 | * If not dirty, or it's pinned and we are not supposed to block, nada. |
1da177e4 | 1173 | */ |
2f8a3ce1 | 1174 | if (!XFS_DQ_IS_DIRTY(dqp) || |
20026d92 | 1175 | (!(flags & SYNC_WAIT) && atomic_read(&dqp->q_pincount) > 0)) { |
1da177e4 | 1176 | xfs_dqfunlock(dqp); |
2f8a3ce1 | 1177 | return 0; |
1da177e4 | 1178 | } |
1da177e4 LT |
1179 | xfs_qm_dqunpin_wait(dqp); |
1180 | ||
1181 | /* | |
1182 | * This may have been unpinned because the filesystem is shutting | |
1183 | * down forcibly. If that's the case we must not write this dquot | |
1184 | * to disk, because the log record didn't make it to disk! | |
1185 | */ | |
acecf1b5 CH |
1186 | if (XFS_FORCED_SHUTDOWN(mp)) { |
1187 | dqp->dq_flags &= ~XFS_DQ_DIRTY; | |
1da177e4 LT |
1188 | xfs_dqfunlock(dqp); |
1189 | return XFS_ERROR(EIO); | |
1190 | } | |
1191 | ||
1192 | /* | |
1193 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1194 | */ |
acecf1b5 CH |
1195 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
1196 | mp->m_quotainfo->qi_dqchunklen, 0, &bp); | |
1197 | if (error) { | |
1da177e4 | 1198 | ASSERT(error != ENOENT); |
1da177e4 | 1199 | xfs_dqfunlock(dqp); |
acecf1b5 | 1200 | return error; |
1da177e4 LT |
1201 | } |
1202 | ||
acecf1b5 CH |
1203 | /* |
1204 | * Calculate the location of the dquot inside the buffer. | |
1205 | */ | |
62926044 | 1206 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1207 | |
1208 | /* | |
1209 | * A simple sanity check in case we got a corrupted dquot.. | |
1210 | */ | |
a0fa2b67 DC |
1211 | error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
1212 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); | |
1213 | if (error) { | |
acecf1b5 CH |
1214 | xfs_buf_relse(bp); |
1215 | xfs_dqfunlock(dqp); | |
1216 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
1da177e4 LT |
1217 | return XFS_ERROR(EIO); |
1218 | } | |
1219 | ||
1220 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1221 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1222 | |
1223 | /* | |
1224 | * Clear the dirty field and remember the flush lsn for later use. | |
1225 | */ | |
acecf1b5 | 1226 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1227 | |
7b2e2a31 DC |
1228 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1229 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 LT |
1230 | |
1231 | /* | |
1232 | * Attach an iodone routine so that we can remove this dquot from the | |
1233 | * AIL and release the flush lock once the dquot is synced to disk. | |
1234 | */ | |
ca30b2a7 CH |
1235 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1236 | &dqp->q_logitem.qli_item); | |
1237 | ||
1da177e4 LT |
1238 | /* |
1239 | * If the buffer is pinned then push on the log so we won't | |
1240 | * get stuck waiting in the write for too long. | |
1241 | */ | |
811e64c7 | 1242 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1243 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1244 | xfs_log_force(mp, 0); |
1da177e4 LT |
1245 | } |
1246 | ||
20026d92 | 1247 | if (flags & SYNC_WAIT) |
c2b006c1 CH |
1248 | error = xfs_bwrite(bp); |
1249 | else | |
61551f1e | 1250 | xfs_buf_delwri_queue(bp); |
c2b006c1 CH |
1251 | |
1252 | xfs_buf_relse(bp); | |
0b1b213f CH |
1253 | |
1254 | trace_xfs_dqflush_done(dqp); | |
1255 | ||
1da177e4 LT |
1256 | /* |
1257 | * dqp is still locked, but caller is free to unlock it now. | |
1258 | */ | |
20026d92 | 1259 | return error; |
1da177e4 LT |
1260 | |
1261 | } | |
1262 | ||
1da177e4 LT |
1263 | int |
1264 | xfs_qm_dqlock_nowait( | |
1265 | xfs_dquot_t *dqp) | |
1266 | { | |
e1f49cf2 | 1267 | return mutex_trylock(&dqp->q_qlock); |
1da177e4 LT |
1268 | } |
1269 | ||
1270 | void | |
1271 | xfs_dqlock( | |
1272 | xfs_dquot_t *dqp) | |
1273 | { | |
e1f49cf2 | 1274 | mutex_lock(&dqp->q_qlock); |
1da177e4 LT |
1275 | } |
1276 | ||
1277 | void | |
1278 | xfs_dqunlock( | |
1279 | xfs_dquot_t *dqp) | |
1280 | { | |
1281 | mutex_unlock(&(dqp->q_qlock)); | |
1282 | if (dqp->q_logitem.qli_dquot == dqp) { | |
1283 | /* Once was dqp->q_mount, but might just have been cleared */ | |
783a2f65 | 1284 | xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_ailp, |
1da177e4 LT |
1285 | (xfs_log_item_t*)&(dqp->q_logitem)); |
1286 | } | |
1287 | } | |
1288 | ||
1289 | ||
1290 | void | |
1291 | xfs_dqunlock_nonotify( | |
1292 | xfs_dquot_t *dqp) | |
1293 | { | |
1294 | mutex_unlock(&(dqp->q_qlock)); | |
1295 | } | |
1296 | ||
5bb87a33 CH |
1297 | /* |
1298 | * Lock two xfs_dquot structures. | |
1299 | * | |
1300 | * To avoid deadlocks we always lock the quota structure with | |
1301 | * the lowerd id first. | |
1302 | */ | |
1da177e4 LT |
1303 | void |
1304 | xfs_dqlock2( | |
1305 | xfs_dquot_t *d1, | |
1306 | xfs_dquot_t *d2) | |
1307 | { | |
1308 | if (d1 && d2) { | |
1309 | ASSERT(d1 != d2); | |
1149d96a CH |
1310 | if (be32_to_cpu(d1->q_core.d_id) > |
1311 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1312 | mutex_lock(&d2->q_qlock); |
1313 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1314 | } else { |
5bb87a33 CH |
1315 | mutex_lock(&d1->q_qlock); |
1316 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1317 | } |
5bb87a33 CH |
1318 | } else if (d1) { |
1319 | mutex_lock(&d1->q_qlock); | |
1320 | } else if (d2) { | |
1321 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1322 | } |
1323 | } | |
1324 | ||
1325 | ||
1326 | /* | |
1327 | * Take a dquot out of the mount's dqlist as well as the hashlist. | |
1328 | * This is called via unmount as well as quotaoff, and the purge | |
1329 | * will always succeed unless there are soft (temp) references | |
1330 | * outstanding. | |
1331 | * | |
1332 | * This returns 0 if it was purged, 1 if it wasn't. It's not an error code | |
1333 | * that we're returning! XXXsup - not cool. | |
1334 | */ | |
1335 | /* ARGSUSED */ | |
1336 | int | |
1337 | xfs_qm_dqpurge( | |
4f0e8a98 | 1338 | xfs_dquot_t *dqp) |
1da177e4 | 1339 | { |
e6a81f13 | 1340 | xfs_dqhash_t *qh = dqp->q_hash; |
3c56836f | 1341 | xfs_mount_t *mp = dqp->q_mount; |
1da177e4 | 1342 | |
3a25404b | 1343 | ASSERT(mutex_is_locked(&mp->m_quotainfo->qi_dqlist_lock)); |
c9a192dc | 1344 | ASSERT(mutex_is_locked(&dqp->q_hash->qh_lock)); |
1da177e4 LT |
1345 | |
1346 | xfs_dqlock(dqp); | |
1347 | /* | |
1348 | * We really can't afford to purge a dquot that is | |
1349 | * referenced, because these are hard refs. | |
1350 | * It shouldn't happen in general because we went thru _all_ inodes in | |
1351 | * dqrele_all_inodes before calling this and didn't let the mountlock go. | |
1352 | * However it is possible that we have dquots with temporary | |
1353 | * references that are not attached to an inode. e.g. see xfs_setattr(). | |
1354 | */ | |
1355 | if (dqp->q_nrefs != 0) { | |
1356 | xfs_dqunlock(dqp); | |
c9a192dc | 1357 | mutex_unlock(&dqp->q_hash->qh_lock); |
1da177e4 LT |
1358 | return (1); |
1359 | } | |
1360 | ||
3a8406f6 | 1361 | ASSERT(!list_empty(&dqp->q_freelist)); |
1da177e4 LT |
1362 | |
1363 | /* | |
1364 | * If we're turning off quotas, we have to make sure that, for | |
1365 | * example, we don't delete quota disk blocks while dquots are | |
1366 | * in the process of getting written to those disk blocks. | |
1367 | * This dquot might well be on AIL, and we can't leave it there | |
1368 | * if we're turning off quotas. Basically, we need this flush | |
1369 | * lock, and are willing to block on it. | |
1370 | */ | |
e1f49cf2 | 1371 | if (!xfs_dqflock_nowait(dqp)) { |
1da177e4 LT |
1372 | /* |
1373 | * Block on the flush lock after nudging dquot buffer, | |
1374 | * if it is incore. | |
1375 | */ | |
1376 | xfs_qm_dqflock_pushbuf_wait(dqp); | |
1377 | } | |
1378 | ||
1379 | /* | |
1380 | * XXXIf we're turning this type of quotas off, we don't care | |
1381 | * about the dirty metadata sitting in this dquot. OTOH, if | |
1382 | * we're unmounting, we do care, so we flush it and wait. | |
1383 | */ | |
1384 | if (XFS_DQ_IS_DIRTY(dqp)) { | |
3c56836f | 1385 | int error; |
0b1b213f | 1386 | |
1da177e4 LT |
1387 | /* dqflush unlocks dqflock */ |
1388 | /* | |
1389 | * Given that dqpurge is a very rare occurrence, it is OK | |
1390 | * that we're holding the hashlist and mplist locks | |
1391 | * across the disk write. But, ... XXXsup | |
1392 | * | |
1393 | * We don't care about getting disk errors here. We need | |
1394 | * to purge this dquot anyway, so we go ahead regardless. | |
1395 | */ | |
20026d92 | 1396 | error = xfs_qm_dqflush(dqp, SYNC_WAIT); |
3c56836f | 1397 | if (error) |
53487786 DC |
1398 | xfs_warn(mp, "%s: dquot %p flush failed", |
1399 | __func__, dqp); | |
1da177e4 LT |
1400 | xfs_dqflock(dqp); |
1401 | } | |
bc3048e3 | 1402 | ASSERT(atomic_read(&dqp->q_pincount) == 0); |
1da177e4 LT |
1403 | ASSERT(XFS_FORCED_SHUTDOWN(mp) || |
1404 | !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL)); | |
1405 | ||
e6a81f13 DC |
1406 | list_del_init(&dqp->q_hashlist); |
1407 | qh->qh_version++; | |
3a25404b DC |
1408 | list_del_init(&dqp->q_mplist); |
1409 | mp->m_quotainfo->qi_dqreclaims++; | |
1410 | mp->m_quotainfo->qi_dquots--; | |
1da177e4 LT |
1411 | /* |
1412 | * XXX Move this to the front of the freelist, if we can get the | |
1413 | * freelist lock. | |
1414 | */ | |
3a8406f6 | 1415 | ASSERT(!list_empty(&dqp->q_freelist)); |
1da177e4 LT |
1416 | |
1417 | dqp->q_mount = NULL; | |
1418 | dqp->q_hash = NULL; | |
1419 | dqp->dq_flags = XFS_DQ_INACTIVE; | |
1420 | memset(&dqp->q_core, 0, sizeof(dqp->q_core)); | |
1421 | xfs_dqfunlock(dqp); | |
1422 | xfs_dqunlock(dqp); | |
e6a81f13 | 1423 | mutex_unlock(&qh->qh_lock); |
1da177e4 LT |
1424 | return (0); |
1425 | } | |
1426 | ||
1427 | ||
1da177e4 LT |
1428 | /* |
1429 | * Give the buffer a little push if it is incore and | |
1430 | * wait on the flush lock. | |
1431 | */ | |
1432 | void | |
1433 | xfs_qm_dqflock_pushbuf_wait( | |
1434 | xfs_dquot_t *dqp) | |
1435 | { | |
8a7b8a89 | 1436 | xfs_mount_t *mp = dqp->q_mount; |
1da177e4 LT |
1437 | xfs_buf_t *bp; |
1438 | ||
1439 | /* | |
1440 | * Check to see if the dquot has been flushed delayed | |
1441 | * write. If so, grab its buffer and send it | |
1442 | * out immediately. We'll be able to acquire | |
1443 | * the flush lock when the I/O completes. | |
1444 | */ | |
8a7b8a89 CH |
1445 | bp = xfs_incore(mp->m_ddev_targp, dqp->q_blkno, |
1446 | mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK); | |
7d6a7bde DC |
1447 | if (!bp) |
1448 | goto out_lock; | |
1449 | ||
1450 | if (XFS_BUF_ISDELAYWRITE(bp)) { | |
811e64c7 | 1451 | if (xfs_buf_ispinned(bp)) |
8a7b8a89 | 1452 | xfs_log_force(mp, 0); |
7d6a7bde DC |
1453 | xfs_buf_delwri_promote(bp); |
1454 | wake_up_process(bp->b_target->bt_task); | |
1da177e4 | 1455 | } |
7d6a7bde DC |
1456 | xfs_buf_relse(bp); |
1457 | out_lock: | |
1da177e4 LT |
1458 | xfs_dqflock(dqp); |
1459 | } |