]>
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 | 41 | |
1da177e4 | 42 | /* |
bf72de31 CH |
43 | * Lock order: |
44 | * | |
45 | * ip->i_lock | |
46 | * qh->qh_lock | |
47 | * qi->qi_dqlist_lock | |
48 | * dquot->q_qlock (xfs_dqlock() and friends) | |
49 | * dquot->q_flush (xfs_dqflock() and friends) | |
50 | * xfs_Gqm->qm_dqfrlist_lock | |
51 | * | |
52 | * If two dquots need to be locked the order is user before group/project, | |
53 | * otherwise by the lowest id first, see xfs_dqlock2. | |
54 | */ | |
1da177e4 | 55 | |
1da177e4 LT |
56 | #ifdef DEBUG |
57 | xfs_buftarg_t *xfs_dqerror_target; | |
58 | int xfs_do_dqerror; | |
59 | int xfs_dqreq_num; | |
60 | int xfs_dqerror_mod = 33; | |
61 | #endif | |
62 | ||
98b8c7a0 CH |
63 | static struct lock_class_key xfs_dquot_other_class; |
64 | ||
1da177e4 LT |
65 | /* |
66 | * Allocate and initialize a dquot. We don't always allocate fresh memory; | |
67 | * we try to reclaim a free dquot if the number of incore dquots are above | |
68 | * a threshold. | |
69 | * The only field inside the core that gets initialized at this point | |
70 | * is the d_id field. The idea is to fill in the entire q_core | |
71 | * when we read in the on disk dquot. | |
72 | */ | |
ba0f32d4 | 73 | STATIC xfs_dquot_t * |
1da177e4 LT |
74 | xfs_qm_dqinit( |
75 | xfs_mount_t *mp, | |
76 | xfs_dqid_t id, | |
77 | uint type) | |
78 | { | |
79 | xfs_dquot_t *dqp; | |
80 | boolean_t brandnewdquot; | |
81 | ||
82 | brandnewdquot = xfs_qm_dqalloc_incore(&dqp); | |
83 | dqp->dq_flags = type; | |
1149d96a | 84 | dqp->q_core.d_id = cpu_to_be32(id); |
1da177e4 LT |
85 | dqp->q_mount = mp; |
86 | ||
87 | /* | |
88 | * No need to re-initialize these if this is a reclaimed dquot. | |
89 | */ | |
90 | if (brandnewdquot) { | |
3a8406f6 | 91 | INIT_LIST_HEAD(&dqp->q_freelist); |
794ee1ba | 92 | mutex_init(&dqp->q_qlock); |
bc3048e3 | 93 | init_waitqueue_head(&dqp->q_pinwait); |
1da177e4 | 94 | |
e1f49cf2 DC |
95 | /* |
96 | * Because we want to use a counting completion, complete | |
97 | * the flush completion once to allow a single access to | |
98 | * the flush completion without blocking. | |
99 | */ | |
100 | init_completion(&dqp->q_flush); | |
101 | complete(&dqp->q_flush); | |
102 | ||
0b1b213f | 103 | trace_xfs_dqinit(dqp); |
1da177e4 LT |
104 | } else { |
105 | /* | |
106 | * Only the q_core portion was zeroed in dqreclaim_one(). | |
107 | * So, we need to reset others. | |
108 | */ | |
3a8406f6 DC |
109 | dqp->q_nrefs = 0; |
110 | dqp->q_blkno = 0; | |
111 | INIT_LIST_HEAD(&dqp->q_mplist); | |
112 | INIT_LIST_HEAD(&dqp->q_hashlist); | |
113 | dqp->q_bufoffset = 0; | |
114 | dqp->q_fileoffset = 0; | |
115 | dqp->q_transp = NULL; | |
116 | dqp->q_gdquot = NULL; | |
117 | dqp->q_res_bcount = 0; | |
118 | dqp->q_res_icount = 0; | |
119 | dqp->q_res_rtbcount = 0; | |
120 | atomic_set(&dqp->q_pincount, 0); | |
121 | dqp->q_hash = NULL; | |
122 | ASSERT(list_empty(&dqp->q_freelist)); | |
1da177e4 | 123 | |
0b1b213f | 124 | trace_xfs_dqreuse(dqp); |
98b8c7a0 CH |
125 | } |
126 | ||
127 | /* | |
128 | * In either case we need to make sure group quotas have a different | |
129 | * lock class than user quotas, to make sure lockdep knows we can | |
130 | * locks of one of each at the same time. | |
131 | */ | |
132 | if (!(type & XFS_DQ_USER)) | |
133 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class); | |
1da177e4 LT |
134 | |
135 | /* | |
136 | * log item gets initialized later | |
137 | */ | |
138 | return (dqp); | |
139 | } | |
140 | ||
141 | /* | |
142 | * This is called to free all the memory associated with a dquot | |
143 | */ | |
144 | void | |
145 | xfs_qm_dqdestroy( | |
146 | xfs_dquot_t *dqp) | |
147 | { | |
3a8406f6 | 148 | ASSERT(list_empty(&dqp->q_freelist)); |
1da177e4 LT |
149 | |
150 | mutex_destroy(&dqp->q_qlock); | |
1da177e4 | 151 | kmem_zone_free(xfs_Gqm->qm_dqzone, dqp); |
0b1b213f | 152 | |
1da177e4 LT |
153 | atomic_dec(&xfs_Gqm->qm_totaldquots); |
154 | } | |
155 | ||
156 | /* | |
157 | * This is what a 'fresh' dquot inside a dquot chunk looks like on disk. | |
158 | */ | |
159 | STATIC void | |
160 | xfs_qm_dqinit_core( | |
161 | xfs_dqid_t id, | |
162 | uint type, | |
163 | xfs_dqblk_t *d) | |
164 | { | |
165 | /* | |
166 | * Caller has zero'd the entire dquot 'chunk' already. | |
167 | */ | |
1149d96a CH |
168 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); |
169 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
170 | d->dd_diskdq.d_id = cpu_to_be32(id); | |
171 | d->dd_diskdq.d_flags = type; | |
1da177e4 LT |
172 | } |
173 | ||
1da177e4 LT |
174 | /* |
175 | * If default limits are in force, push them into the dquot now. | |
176 | * We overwrite the dquot limits only if they are zero and this | |
177 | * is not the root dquot. | |
178 | */ | |
179 | void | |
180 | xfs_qm_adjust_dqlimits( | |
181 | xfs_mount_t *mp, | |
182 | xfs_disk_dquot_t *d) | |
183 | { | |
184 | xfs_quotainfo_t *q = mp->m_quotainfo; | |
185 | ||
186 | ASSERT(d->d_id); | |
187 | ||
188 | if (q->qi_bsoftlimit && !d->d_blk_softlimit) | |
1149d96a | 189 | d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit); |
1da177e4 | 190 | if (q->qi_bhardlimit && !d->d_blk_hardlimit) |
1149d96a | 191 | d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit); |
1da177e4 | 192 | if (q->qi_isoftlimit && !d->d_ino_softlimit) |
1149d96a | 193 | d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit); |
1da177e4 | 194 | if (q->qi_ihardlimit && !d->d_ino_hardlimit) |
1149d96a | 195 | d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit); |
1da177e4 | 196 | if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit) |
1149d96a | 197 | d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit); |
1da177e4 | 198 | if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit) |
1149d96a | 199 | d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit); |
1da177e4 LT |
200 | } |
201 | ||
202 | /* | |
203 | * Check the limits and timers of a dquot and start or reset timers | |
204 | * if necessary. | |
205 | * This gets called even when quota enforcement is OFF, which makes our | |
206 | * life a little less complicated. (We just don't reject any quota | |
207 | * reservations in that case, when enforcement is off). | |
208 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
209 | * enforcement's off. | |
210 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
211 | * 'automatically' get started when limits get exceeded. They do |
212 | * get reset to zero, however, when we find the count to be under | |
213 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
214 | */ |
215 | void | |
216 | xfs_qm_adjust_dqtimers( | |
217 | xfs_mount_t *mp, | |
218 | xfs_disk_dquot_t *d) | |
219 | { | |
220 | ASSERT(d->d_id); | |
221 | ||
ea15ab3c | 222 | #ifdef DEBUG |
1149d96a CH |
223 | if (d->d_blk_hardlimit) |
224 | ASSERT(be64_to_cpu(d->d_blk_softlimit) <= | |
225 | be64_to_cpu(d->d_blk_hardlimit)); | |
226 | if (d->d_ino_hardlimit) | |
227 | ASSERT(be64_to_cpu(d->d_ino_softlimit) <= | |
228 | be64_to_cpu(d->d_ino_hardlimit)); | |
229 | if (d->d_rtb_hardlimit) | |
230 | ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= | |
231 | be64_to_cpu(d->d_rtb_hardlimit)); | |
1da177e4 | 232 | #endif |
ea15ab3c | 233 | |
1da177e4 | 234 | if (!d->d_btimer) { |
1149d96a CH |
235 | if ((d->d_blk_softlimit && |
236 | (be64_to_cpu(d->d_bcount) >= | |
237 | be64_to_cpu(d->d_blk_softlimit))) || | |
238 | (d->d_blk_hardlimit && | |
239 | (be64_to_cpu(d->d_bcount) >= | |
240 | be64_to_cpu(d->d_blk_hardlimit)))) { | |
241 | d->d_btimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 242 | mp->m_quotainfo->qi_btimelimit); |
754002b4 NS |
243 | } else { |
244 | d->d_bwarns = 0; | |
1da177e4 LT |
245 | } |
246 | } else { | |
247 | if ((!d->d_blk_softlimit || | |
1149d96a CH |
248 | (be64_to_cpu(d->d_bcount) < |
249 | be64_to_cpu(d->d_blk_softlimit))) && | |
1da177e4 | 250 | (!d->d_blk_hardlimit || |
1149d96a CH |
251 | (be64_to_cpu(d->d_bcount) < |
252 | be64_to_cpu(d->d_blk_hardlimit)))) { | |
1da177e4 LT |
253 | d->d_btimer = 0; |
254 | } | |
255 | } | |
256 | ||
257 | if (!d->d_itimer) { | |
1149d96a CH |
258 | if ((d->d_ino_softlimit && |
259 | (be64_to_cpu(d->d_icount) >= | |
260 | be64_to_cpu(d->d_ino_softlimit))) || | |
261 | (d->d_ino_hardlimit && | |
262 | (be64_to_cpu(d->d_icount) >= | |
263 | be64_to_cpu(d->d_ino_hardlimit)))) { | |
264 | d->d_itimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 265 | mp->m_quotainfo->qi_itimelimit); |
754002b4 NS |
266 | } else { |
267 | d->d_iwarns = 0; | |
1da177e4 LT |
268 | } |
269 | } else { | |
270 | if ((!d->d_ino_softlimit || | |
1149d96a CH |
271 | (be64_to_cpu(d->d_icount) < |
272 | be64_to_cpu(d->d_ino_softlimit))) && | |
1da177e4 | 273 | (!d->d_ino_hardlimit || |
1149d96a CH |
274 | (be64_to_cpu(d->d_icount) < |
275 | be64_to_cpu(d->d_ino_hardlimit)))) { | |
1da177e4 LT |
276 | d->d_itimer = 0; |
277 | } | |
278 | } | |
279 | ||
280 | if (!d->d_rtbtimer) { | |
1149d96a CH |
281 | if ((d->d_rtb_softlimit && |
282 | (be64_to_cpu(d->d_rtbcount) >= | |
283 | be64_to_cpu(d->d_rtb_softlimit))) || | |
284 | (d->d_rtb_hardlimit && | |
285 | (be64_to_cpu(d->d_rtbcount) >= | |
286 | be64_to_cpu(d->d_rtb_hardlimit)))) { | |
287 | d->d_rtbtimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 288 | mp->m_quotainfo->qi_rtbtimelimit); |
754002b4 NS |
289 | } else { |
290 | d->d_rtbwarns = 0; | |
1da177e4 LT |
291 | } |
292 | } else { | |
293 | if ((!d->d_rtb_softlimit || | |
1149d96a CH |
294 | (be64_to_cpu(d->d_rtbcount) < |
295 | be64_to_cpu(d->d_rtb_softlimit))) && | |
1da177e4 | 296 | (!d->d_rtb_hardlimit || |
1149d96a CH |
297 | (be64_to_cpu(d->d_rtbcount) < |
298 | be64_to_cpu(d->d_rtb_hardlimit)))) { | |
1da177e4 LT |
299 | d->d_rtbtimer = 0; |
300 | } | |
301 | } | |
302 | } | |
303 | ||
1da177e4 LT |
304 | /* |
305 | * initialize a buffer full of dquots and log the whole thing | |
306 | */ | |
307 | STATIC void | |
308 | xfs_qm_init_dquot_blk( | |
309 | xfs_trans_t *tp, | |
310 | xfs_mount_t *mp, | |
311 | xfs_dqid_t id, | |
312 | uint type, | |
313 | xfs_buf_t *bp) | |
314 | { | |
8a7b8a89 | 315 | struct xfs_quotainfo *q = mp->m_quotainfo; |
1da177e4 LT |
316 | xfs_dqblk_t *d; |
317 | int curid, i; | |
318 | ||
319 | ASSERT(tp); | |
0c842ad4 | 320 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 321 | |
62926044 | 322 | d = bp->b_addr; |
1da177e4 LT |
323 | |
324 | /* | |
325 | * ID of the first dquot in the block - id's are zero based. | |
326 | */ | |
8a7b8a89 | 327 | curid = id - (id % q->qi_dqperchunk); |
1da177e4 | 328 | ASSERT(curid >= 0); |
8a7b8a89 CH |
329 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
330 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) | |
1da177e4 LT |
331 | xfs_qm_dqinit_core(curid, type, d); |
332 | xfs_trans_dquot_buf(tp, bp, | |
c1155410 DC |
333 | (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : |
334 | ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : | |
335 | XFS_BLF_GDQUOT_BUF))); | |
8a7b8a89 | 336 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); |
1da177e4 LT |
337 | } |
338 | ||
339 | ||
340 | ||
341 | /* | |
342 | * Allocate a block and fill it with dquots. | |
343 | * This is called when the bmapi finds a hole. | |
344 | */ | |
345 | STATIC int | |
346 | xfs_qm_dqalloc( | |
efa092f3 | 347 | xfs_trans_t **tpp, |
1da177e4 LT |
348 | xfs_mount_t *mp, |
349 | xfs_dquot_t *dqp, | |
350 | xfs_inode_t *quotip, | |
351 | xfs_fileoff_t offset_fsb, | |
352 | xfs_buf_t **O_bpp) | |
353 | { | |
354 | xfs_fsblock_t firstblock; | |
355 | xfs_bmap_free_t flist; | |
356 | xfs_bmbt_irec_t map; | |
357 | int nmaps, error, committed; | |
358 | xfs_buf_t *bp; | |
efa092f3 | 359 | xfs_trans_t *tp = *tpp; |
1da177e4 LT |
360 | |
361 | ASSERT(tp != NULL); | |
0b1b213f CH |
362 | |
363 | trace_xfs_dqalloc(dqp); | |
1da177e4 LT |
364 | |
365 | /* | |
366 | * Initialize the bmap freelist prior to calling bmapi code. | |
367 | */ | |
9d87c319 | 368 | xfs_bmap_init(&flist, &firstblock); |
1da177e4 LT |
369 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
370 | /* | |
371 | * Return if this type of quotas is turned off while we didn't | |
372 | * have an inode lock | |
373 | */ | |
374 | if (XFS_IS_THIS_QUOTA_OFF(dqp)) { | |
375 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
376 | return (ESRCH); | |
377 | } | |
378 | ||
ddc3415a | 379 | xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); |
1da177e4 | 380 | nmaps = 1; |
c0dc7828 DC |
381 | error = xfs_bmapi_write(tp, quotip, offset_fsb, |
382 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
383 | &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), | |
384 | &map, &nmaps, &flist); | |
385 | if (error) | |
1da177e4 | 386 | goto error0; |
1da177e4 LT |
387 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
388 | ASSERT(nmaps == 1); | |
389 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
390 | (map.br_startblock != HOLESTARTBLOCK)); | |
391 | ||
392 | /* | |
393 | * Keep track of the blkno to save a lookup later | |
394 | */ | |
395 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
396 | ||
397 | /* now we can just get the buffer (there's nothing to read yet) */ | |
398 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
399 | dqp->q_blkno, | |
8a7b8a89 | 400 | mp->m_quotainfo->qi_dqchunklen, |
1da177e4 | 401 | 0); |
2a30f36d CS |
402 | |
403 | error = xfs_buf_geterror(bp); | |
404 | if (error) | |
1da177e4 | 405 | goto error1; |
2a30f36d | 406 | |
1da177e4 LT |
407 | /* |
408 | * Make a chunk of dquots out of this buffer and log | |
409 | * the entire thing. | |
410 | */ | |
1149d96a | 411 | xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 412 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
1da177e4 | 413 | |
efa092f3 TS |
414 | /* |
415 | * xfs_bmap_finish() may commit the current transaction and | |
416 | * start a second transaction if the freelist is not empty. | |
417 | * | |
418 | * Since we still want to modify this buffer, we need to | |
419 | * ensure that the buffer is not released on commit of | |
420 | * the first transaction and ensure the buffer is added to the | |
421 | * second transaction. | |
422 | * | |
423 | * If there is only one transaction then don't stop the buffer | |
424 | * from being released when it commits later on. | |
425 | */ | |
426 | ||
427 | xfs_trans_bhold(tp, bp); | |
428 | ||
f7c99b6f | 429 | if ((error = xfs_bmap_finish(tpp, &flist, &committed))) { |
1da177e4 LT |
430 | goto error1; |
431 | } | |
432 | ||
efa092f3 TS |
433 | if (committed) { |
434 | tp = *tpp; | |
435 | xfs_trans_bjoin(tp, bp); | |
436 | } else { | |
437 | xfs_trans_bhold_release(tp, bp); | |
438 | } | |
439 | ||
1da177e4 LT |
440 | *O_bpp = bp; |
441 | return 0; | |
442 | ||
443 | error1: | |
444 | xfs_bmap_cancel(&flist); | |
445 | error0: | |
446 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
447 | ||
448 | return (error); | |
449 | } | |
450 | ||
451 | /* | |
452 | * Maps a dquot to the buffer containing its on-disk version. | |
453 | * This returns a ptr to the buffer containing the on-disk dquot | |
454 | * in the bpp param, and a ptr to the on-disk dquot within that buffer | |
455 | */ | |
456 | STATIC int | |
457 | xfs_qm_dqtobp( | |
efa092f3 | 458 | xfs_trans_t **tpp, |
1da177e4 LT |
459 | xfs_dquot_t *dqp, |
460 | xfs_disk_dquot_t **O_ddpp, | |
461 | xfs_buf_t **O_bpp, | |
462 | uint flags) | |
463 | { | |
464 | xfs_bmbt_irec_t map; | |
acecf1b5 | 465 | int nmaps = 1, error; |
1da177e4 | 466 | xfs_buf_t *bp; |
acecf1b5 CH |
467 | xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp); |
468 | xfs_mount_t *mp = dqp->q_mount; | |
1da177e4 | 469 | xfs_disk_dquot_t *ddq; |
acecf1b5 | 470 | xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); |
efa092f3 | 471 | xfs_trans_t *tp = (tpp ? *tpp : NULL); |
1da177e4 | 472 | |
acecf1b5 | 473 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
1da177e4 | 474 | |
acecf1b5 CH |
475 | xfs_ilock(quotip, XFS_ILOCK_SHARED); |
476 | if (XFS_IS_THIS_QUOTA_OFF(dqp)) { | |
1da177e4 | 477 | /* |
acecf1b5 CH |
478 | * Return if this type of quotas is turned off while we |
479 | * didn't have the quota inode lock. | |
1da177e4 | 480 | */ |
acecf1b5 CH |
481 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); |
482 | return ESRCH; | |
483 | } | |
484 | ||
485 | /* | |
486 | * Find the block map; no allocations yet | |
487 | */ | |
5c8ed202 DC |
488 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
489 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); | |
acecf1b5 CH |
490 | |
491 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); | |
492 | if (error) | |
493 | return error; | |
494 | ||
495 | ASSERT(nmaps == 1); | |
496 | ASSERT(map.br_blockcount == 1); | |
497 | ||
498 | /* | |
499 | * Offset of dquot in the (fixed sized) dquot chunk. | |
500 | */ | |
501 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
502 | sizeof(xfs_dqblk_t); | |
503 | ||
504 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
505 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 506 | /* |
acecf1b5 | 507 | * We don't allocate unless we're asked to |
1da177e4 | 508 | */ |
acecf1b5 CH |
509 | if (!(flags & XFS_QMOPT_DQALLOC)) |
510 | return ENOENT; | |
1da177e4 | 511 | |
acecf1b5 CH |
512 | ASSERT(tp); |
513 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
514 | dqp->q_fileoffset, &bp); | |
1da177e4 | 515 | if (error) |
acecf1b5 CH |
516 | return error; |
517 | tp = *tpp; | |
518 | } else { | |
519 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
520 | |
521 | /* | |
acecf1b5 CH |
522 | * store the blkno etc so that we don't have to do the |
523 | * mapping all the time | |
1da177e4 | 524 | */ |
acecf1b5 | 525 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 526 | |
8a7b8a89 CH |
527 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
528 | dqp->q_blkno, | |
529 | mp->m_quotainfo->qi_dqchunklen, | |
530 | 0, &bp); | |
1da177e4 LT |
531 | if (error || !bp) |
532 | return XFS_ERROR(error); | |
533 | } | |
acecf1b5 | 534 | |
0c842ad4 | 535 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
536 | |
537 | /* | |
538 | * calculate the location of the dquot inside the buffer. | |
539 | */ | |
62926044 | 540 | ddq = bp->b_addr + dqp->q_bufoffset; |
1da177e4 LT |
541 | |
542 | /* | |
543 | * A simple sanity check in case we got a corrupted dquot... | |
544 | */ | |
a0fa2b67 | 545 | error = xfs_qm_dqcheck(mp, ddq, id, dqp->dq_flags & XFS_DQ_ALLTYPES, |
1da177e4 | 546 | flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN), |
a0fa2b67 DC |
547 | "dqtobp"); |
548 | if (error) { | |
1da177e4 LT |
549 | if (!(flags & XFS_QMOPT_DQREPAIR)) { |
550 | xfs_trans_brelse(tp, bp); | |
551 | return XFS_ERROR(EIO); | |
552 | } | |
1da177e4 LT |
553 | } |
554 | ||
555 | *O_bpp = bp; | |
556 | *O_ddpp = ddq; | |
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 | * | |
566 | */ | |
567 | /* ARGSUSED */ | |
568 | STATIC int | |
569 | xfs_qm_dqread( | |
efa092f3 | 570 | xfs_trans_t **tpp, |
1da177e4 LT |
571 | xfs_dqid_t id, |
572 | xfs_dquot_t *dqp, /* dquot to get filled in */ | |
573 | uint flags) | |
574 | { | |
575 | xfs_disk_dquot_t *ddqp; | |
576 | xfs_buf_t *bp; | |
577 | int error; | |
efa092f3 TS |
578 | xfs_trans_t *tp; |
579 | ||
580 | ASSERT(tpp); | |
1da177e4 | 581 | |
0b1b213f CH |
582 | trace_xfs_dqread(dqp); |
583 | ||
1da177e4 LT |
584 | /* |
585 | * get a pointer to the on-disk dquot and the buffer containing it | |
586 | * dqp already knows its own type (GROUP/USER). | |
587 | */ | |
efa092f3 | 588 | if ((error = xfs_qm_dqtobp(tpp, dqp, &ddqp, &bp, flags))) { |
1da177e4 LT |
589 | return (error); |
590 | } | |
efa092f3 | 591 | tp = *tpp; |
1da177e4 LT |
592 | |
593 | /* copy everything from disk dquot to the incore dquot */ | |
594 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
1149d96a | 595 | ASSERT(be32_to_cpu(dqp->q_core.d_id) == id); |
1da177e4 LT |
596 | xfs_qm_dquot_logitem_init(dqp); |
597 | ||
598 | /* | |
599 | * Reservation counters are defined as reservation plus current usage | |
25985edc | 600 | * to avoid having to add every time. |
1da177e4 | 601 | */ |
1149d96a CH |
602 | dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); |
603 | dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); | |
604 | dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); | |
1da177e4 LT |
605 | |
606 | /* Mark the buf so that this will stay incore a little longer */ | |
38f23232 | 607 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 LT |
608 | |
609 | /* | |
610 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
611 | * So we need to release with xfs_trans_brelse(). | |
612 | * The strategy here is identical to that of inodes; we lock | |
613 | * the dquot in xfs_qm_dqget() before making it accessible to | |
614 | * others. This is because dquots, like inodes, need a good level of | |
615 | * concurrency, and we don't want to take locks on the entire buffers | |
616 | * for dquot accesses. | |
617 | * Note also that the dquot buffer may even be dirty at this point, if | |
618 | * this particular dquot was repaired. We still aren't afraid to | |
619 | * brelse it because we have the changes incore. | |
620 | */ | |
0c842ad4 | 621 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
622 | xfs_trans_brelse(tp, bp); |
623 | ||
624 | return (error); | |
625 | } | |
626 | ||
627 | ||
628 | /* | |
629 | * allocate an incore dquot from the kernel heap, | |
630 | * and fill its core with quota information kept on disk. | |
631 | * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk | |
632 | * if it wasn't already allocated. | |
633 | */ | |
634 | STATIC int | |
635 | xfs_qm_idtodq( | |
636 | xfs_mount_t *mp, | |
637 | xfs_dqid_t id, /* gid or uid, depending on type */ | |
638 | uint type, /* UDQUOT or GDQUOT */ | |
639 | uint flags, /* DQALLOC, DQREPAIR */ | |
640 | xfs_dquot_t **O_dqpp)/* OUT : incore dquot, not locked */ | |
641 | { | |
642 | xfs_dquot_t *dqp; | |
643 | int error; | |
644 | xfs_trans_t *tp; | |
645 | int cancelflags=0; | |
646 | ||
647 | dqp = xfs_qm_dqinit(mp, id, type); | |
648 | tp = NULL; | |
649 | if (flags & XFS_QMOPT_DQALLOC) { | |
650 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC); | |
8a7b8a89 CH |
651 | error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp), |
652 | XFS_WRITE_LOG_RES(mp) + | |
653 | BBTOB(mp->m_quotainfo->qi_dqchunklen) - 1 + | |
654 | 128, | |
655 | 0, | |
656 | XFS_TRANS_PERM_LOG_RES, | |
657 | XFS_WRITE_LOG_COUNT); | |
658 | if (error) { | |
1da177e4 LT |
659 | cancelflags = 0; |
660 | goto error0; | |
661 | } | |
662 | cancelflags = XFS_TRANS_RELEASE_LOG_RES; | |
663 | } | |
664 | ||
665 | /* | |
666 | * Read it from disk; xfs_dqread() takes care of | |
667 | * all the necessary initialization of dquot's fields (locks, etc) | |
668 | */ | |
efa092f3 | 669 | if ((error = xfs_qm_dqread(&tp, id, dqp, flags))) { |
1da177e4 LT |
670 | /* |
671 | * This can happen if quotas got turned off (ESRCH), | |
672 | * or if the dquot didn't exist on disk and we ask to | |
673 | * allocate (ENOENT). | |
674 | */ | |
0b1b213f | 675 | trace_xfs_dqread_fail(dqp); |
1da177e4 LT |
676 | cancelflags |= XFS_TRANS_ABORT; |
677 | goto error0; | |
678 | } | |
679 | if (tp) { | |
1c72bf90 | 680 | if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) |
1da177e4 LT |
681 | goto error1; |
682 | } | |
683 | ||
684 | *O_dqpp = dqp; | |
685 | return (0); | |
686 | ||
687 | error0: | |
688 | ASSERT(error); | |
689 | if (tp) | |
690 | xfs_trans_cancel(tp, cancelflags); | |
691 | error1: | |
692 | xfs_qm_dqdestroy(dqp); | |
693 | *O_dqpp = NULL; | |
694 | return (error); | |
695 | } | |
696 | ||
697 | /* | |
698 | * Lookup a dquot in the incore dquot hashtable. We keep two separate | |
699 | * hashtables for user and group dquots; and, these are global tables | |
700 | * inside the XQM, not per-filesystem tables. | |
701 | * The hash chain must be locked by caller, and it is left locked | |
702 | * on return. Returning dquot is locked. | |
703 | */ | |
704 | STATIC int | |
705 | xfs_qm_dqlookup( | |
706 | xfs_mount_t *mp, | |
707 | xfs_dqid_t id, | |
708 | xfs_dqhash_t *qh, | |
709 | xfs_dquot_t **O_dqpp) | |
710 | { | |
711 | xfs_dquot_t *dqp; | |
1da177e4 | 712 | |
c9a192dc | 713 | ASSERT(mutex_is_locked(&qh->qh_lock)); |
1da177e4 | 714 | |
1da177e4 LT |
715 | /* |
716 | * Traverse the hashchain looking for a match | |
717 | */ | |
e6a81f13 | 718 | list_for_each_entry(dqp, &qh->qh_list, q_hashlist) { |
1da177e4 LT |
719 | /* |
720 | * We already have the hashlock. We don't need the | |
721 | * dqlock to look at the id field of the dquot, since the | |
722 | * id can't be modified without the hashlock anyway. | |
723 | */ | |
be7ffc38 CH |
724 | if (be32_to_cpu(dqp->q_core.d_id) != id || dqp->q_mount != mp) |
725 | continue; | |
0b1b213f | 726 | |
be7ffc38 | 727 | trace_xfs_dqlookup_found(dqp); |
1da177e4 | 728 | |
be7ffc38 | 729 | xfs_dqlock(dqp); |
92678554 CH |
730 | if (dqp->dq_flags & XFS_DQ_FREEING) { |
731 | *O_dqpp = NULL; | |
732 | xfs_dqunlock(dqp); | |
733 | return -1; | |
734 | } | |
735 | ||
78e55892 | 736 | dqp->q_nrefs++; |
be7ffc38 CH |
737 | |
738 | /* | |
739 | * move the dquot to the front of the hashchain | |
740 | */ | |
741 | list_move(&dqp->q_hashlist, &qh->qh_list); | |
742 | trace_xfs_dqlookup_done(dqp); | |
743 | *O_dqpp = dqp; | |
744 | return 0; | |
1da177e4 LT |
745 | } |
746 | ||
747 | *O_dqpp = NULL; | |
be7ffc38 | 748 | return 1; |
1da177e4 LT |
749 | } |
750 | ||
751 | /* | |
752 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
753 | * a locked dquot, doing an allocation (if requested) as needed. | |
754 | * When both an inode and an id are given, the inode's id takes precedence. | |
755 | * That is, if the id changes while we don't hold the ilock inside this | |
756 | * function, the new dquot is returned, not necessarily the one requested | |
757 | * in the id argument. | |
758 | */ | |
759 | int | |
760 | xfs_qm_dqget( | |
761 | xfs_mount_t *mp, | |
762 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
763 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
764 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
765 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
766 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
767 | { | |
768 | xfs_dquot_t *dqp; | |
769 | xfs_dqhash_t *h; | |
770 | uint version; | |
771 | int error; | |
772 | ||
773 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
774 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 775 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 LT |
776 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
777 | return (ESRCH); | |
778 | } | |
779 | h = XFS_DQ_HASH(mp, id, type); | |
780 | ||
781 | #ifdef DEBUG | |
782 | if (xfs_do_dqerror) { | |
783 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
784 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 785 | xfs_debug(mp, "Returning error in dqget"); |
1da177e4 LT |
786 | return (EIO); |
787 | } | |
788 | } | |
1da177e4 | 789 | |
c8ad20ff NS |
790 | ASSERT(type == XFS_DQ_USER || |
791 | type == XFS_DQ_PROJ || | |
792 | type == XFS_DQ_GROUP); | |
1da177e4 | 793 | if (ip) { |
579aa9ca | 794 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
795 | if (type == XFS_DQ_USER) |
796 | ASSERT(ip->i_udquot == NULL); | |
797 | else | |
798 | ASSERT(ip->i_gdquot == NULL); | |
799 | } | |
800 | #endif | |
92678554 CH |
801 | |
802 | restart: | |
c9a192dc | 803 | mutex_lock(&h->qh_lock); |
1da177e4 LT |
804 | |
805 | /* | |
806 | * Look in the cache (hashtable). | |
807 | * The chain is kept locked during lookup. | |
808 | */ | |
92678554 CH |
809 | switch (xfs_qm_dqlookup(mp, id, h, O_dqpp)) { |
810 | case -1: | |
811 | XQM_STATS_INC(xqmstats.xs_qm_dquot_dups); | |
812 | mutex_unlock(&h->qh_lock); | |
813 | delay(1); | |
814 | goto restart; | |
815 | case 0: | |
1da177e4 LT |
816 | XQM_STATS_INC(xqmstats.xs_qm_dqcachehits); |
817 | /* | |
818 | * The dquot was found, moved to the front of the chain, | |
819 | * taken off the freelist if it was on it, and locked | |
820 | * at this point. Just unlock the hashchain and return. | |
821 | */ | |
822 | ASSERT(*O_dqpp); | |
823 | ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp)); | |
c9a192dc | 824 | mutex_unlock(&h->qh_lock); |
0b1b213f | 825 | trace_xfs_dqget_hit(*O_dqpp); |
92678554 CH |
826 | return 0; /* success */ |
827 | default: | |
828 | XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses); | |
829 | break; | |
1da177e4 | 830 | } |
1da177e4 LT |
831 | |
832 | /* | |
833 | * Dquot cache miss. We don't want to keep the inode lock across | |
834 | * a (potential) disk read. Also we don't want to deal with the lock | |
835 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
836 | * lock here means dealing with a chown that can happen before | |
837 | * we re-acquire the lock. | |
838 | */ | |
839 | if (ip) | |
840 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
841 | /* | |
842 | * Save the hashchain version stamp, and unlock the chain, so that | |
843 | * we don't keep the lock across a disk read | |
844 | */ | |
845 | version = h->qh_version; | |
c9a192dc | 846 | mutex_unlock(&h->qh_lock); |
1da177e4 LT |
847 | |
848 | /* | |
849 | * Allocate the dquot on the kernel heap, and read the ondisk | |
850 | * portion off the disk. Also, do all the necessary initialization | |
851 | * This can return ENOENT if dquot didn't exist on disk and we didn't | |
852 | * ask it to allocate; ESRCH if quotas got turned off suddenly. | |
853 | */ | |
854 | if ((error = xfs_qm_idtodq(mp, id, type, | |
855 | flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR| | |
856 | XFS_QMOPT_DOWARN), | |
857 | &dqp))) { | |
858 | if (ip) | |
859 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
860 | return (error); | |
861 | } | |
862 | ||
863 | /* | |
864 | * See if this is mount code calling to look at the overall quota limits | |
865 | * which are stored in the id == 0 user or group's dquot. | |
866 | * Since we may not have done a quotacheck by this point, just return | |
867 | * the dquot without attaching it to any hashtables, lists, etc, or even | |
868 | * taking a reference. | |
869 | * The caller must dqdestroy this once done. | |
870 | */ | |
871 | if (flags & XFS_QMOPT_DQSUSER) { | |
872 | ASSERT(id == 0); | |
873 | ASSERT(! ip); | |
874 | goto dqret; | |
875 | } | |
876 | ||
877 | /* | |
878 | * Dquot lock comes after hashlock in the lock ordering | |
879 | */ | |
880 | if (ip) { | |
881 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
191f8488 | 882 | |
1da177e4 LT |
883 | /* |
884 | * A dquot could be attached to this inode by now, since | |
885 | * we had dropped the ilock. | |
886 | */ | |
887 | if (type == XFS_DQ_USER) { | |
191f8488 CH |
888 | if (!XFS_IS_UQUOTA_ON(mp)) { |
889 | /* inode stays locked on return */ | |
890 | xfs_qm_dqdestroy(dqp); | |
891 | return XFS_ERROR(ESRCH); | |
892 | } | |
1da177e4 LT |
893 | if (ip->i_udquot) { |
894 | xfs_qm_dqdestroy(dqp); | |
895 | dqp = ip->i_udquot; | |
896 | xfs_dqlock(dqp); | |
897 | goto dqret; | |
898 | } | |
899 | } else { | |
191f8488 CH |
900 | if (!XFS_IS_OQUOTA_ON(mp)) { |
901 | /* inode stays locked on return */ | |
902 | xfs_qm_dqdestroy(dqp); | |
903 | return XFS_ERROR(ESRCH); | |
904 | } | |
1da177e4 LT |
905 | if (ip->i_gdquot) { |
906 | xfs_qm_dqdestroy(dqp); | |
907 | dqp = ip->i_gdquot; | |
908 | xfs_dqlock(dqp); | |
909 | goto dqret; | |
910 | } | |
911 | } | |
912 | } | |
913 | ||
914 | /* | |
915 | * Hashlock comes after ilock in lock order | |
916 | */ | |
c9a192dc | 917 | mutex_lock(&h->qh_lock); |
1da177e4 LT |
918 | if (version != h->qh_version) { |
919 | xfs_dquot_t *tmpdqp; | |
920 | /* | |
921 | * Now, see if somebody else put the dquot in the | |
922 | * hashtable before us. This can happen because we didn't | |
923 | * keep the hashchain lock. We don't have to worry about | |
924 | * lock order between the two dquots here since dqp isn't | |
925 | * on any findable lists yet. | |
926 | */ | |
92678554 CH |
927 | switch (xfs_qm_dqlookup(mp, id, h, &tmpdqp)) { |
928 | case 0: | |
929 | case -1: | |
1da177e4 | 930 | /* |
92678554 CH |
931 | * Duplicate found, either in cache or on its way out. |
932 | * Just throw away the new dquot and start over. | |
1da177e4 | 933 | */ |
92678554 CH |
934 | if (tmpdqp) |
935 | xfs_qm_dqput(tmpdqp); | |
c9a192dc | 936 | mutex_unlock(&h->qh_lock); |
1da177e4 LT |
937 | xfs_qm_dqdestroy(dqp); |
938 | XQM_STATS_INC(xqmstats.xs_qm_dquot_dups); | |
92678554 CH |
939 | goto restart; |
940 | default: | |
941 | break; | |
1da177e4 LT |
942 | } |
943 | } | |
944 | ||
945 | /* | |
946 | * Put the dquot at the beginning of the hash-chain and mp's list | |
947 | * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock .. | |
948 | */ | |
c9a192dc | 949 | ASSERT(mutex_is_locked(&h->qh_lock)); |
1da177e4 | 950 | dqp->q_hash = h; |
e6a81f13 DC |
951 | list_add(&dqp->q_hashlist, &h->qh_list); |
952 | h->qh_version++; | |
1da177e4 LT |
953 | |
954 | /* | |
955 | * Attach this dquot to this filesystem's list of all dquots, | |
956 | * kept inside the mount structure in m_quotainfo field | |
957 | */ | |
3a25404b | 958 | mutex_lock(&mp->m_quotainfo->qi_dqlist_lock); |
1da177e4 LT |
959 | |
960 | /* | |
961 | * We return a locked dquot to the caller, with a reference taken | |
962 | */ | |
963 | xfs_dqlock(dqp); | |
964 | dqp->q_nrefs = 1; | |
965 | ||
3a25404b DC |
966 | list_add(&dqp->q_mplist, &mp->m_quotainfo->qi_dqlist); |
967 | mp->m_quotainfo->qi_dquots++; | |
968 | mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock); | |
c9a192dc | 969 | mutex_unlock(&h->qh_lock); |
1da177e4 | 970 | dqret: |
579aa9ca | 971 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 972 | trace_xfs_dqget_miss(dqp); |
1da177e4 LT |
973 | *O_dqpp = dqp; |
974 | return (0); | |
975 | } | |
976 | ||
977 | ||
978 | /* | |
979 | * Release a reference to the dquot (decrement ref-count) | |
980 | * and unlock it. If there is a group quota attached to this | |
981 | * dquot, carefully release that too without tripping over | |
982 | * deadlocks'n'stuff. | |
983 | */ | |
984 | void | |
985 | xfs_qm_dqput( | |
bf72de31 | 986 | struct xfs_dquot *dqp) |
1da177e4 | 987 | { |
bf72de31 | 988 | struct xfs_dquot *gdqp; |
1da177e4 LT |
989 | |
990 | ASSERT(dqp->q_nrefs > 0); | |
991 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
0b1b213f CH |
992 | |
993 | trace_xfs_dqput(dqp); | |
1da177e4 | 994 | |
bf72de31 CH |
995 | recurse: |
996 | if (--dqp->q_nrefs > 0) { | |
1da177e4 LT |
997 | xfs_dqunlock(dqp); |
998 | return; | |
999 | } | |
1000 | ||
bf72de31 CH |
1001 | trace_xfs_dqput_free(dqp); |
1002 | ||
1003 | mutex_lock(&xfs_Gqm->qm_dqfrlist_lock); | |
1004 | if (list_empty(&dqp->q_freelist)) { | |
1005 | list_add_tail(&dqp->q_freelist, &xfs_Gqm->qm_dqfrlist); | |
1006 | xfs_Gqm->qm_dqfrlist_cnt++; | |
1007 | } | |
1008 | mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); | |
1009 | ||
1da177e4 | 1010 | /* |
bf72de31 CH |
1011 | * If we just added a udquot to the freelist, then we want to release |
1012 | * the gdquot reference that it (probably) has. Otherwise it'll keep | |
1013 | * the gdquot from getting reclaimed. | |
1da177e4 | 1014 | */ |
bf72de31 CH |
1015 | gdqp = dqp->q_gdquot; |
1016 | if (gdqp) { | |
1017 | xfs_dqlock(gdqp); | |
1018 | dqp->q_gdquot = NULL; | |
1da177e4 | 1019 | } |
bf72de31 | 1020 | xfs_dqunlock(dqp); |
1da177e4 | 1021 | |
bf72de31 CH |
1022 | /* |
1023 | * If we had a group quota hint, release it now. | |
1024 | */ | |
1025 | if (gdqp) { | |
1da177e4 | 1026 | dqp = gdqp; |
bf72de31 | 1027 | goto recurse; |
1da177e4 | 1028 | } |
1da177e4 LT |
1029 | } |
1030 | ||
1031 | /* | |
1032 | * Release a dquot. Flush it if dirty, then dqput() it. | |
1033 | * dquot must not be locked. | |
1034 | */ | |
1035 | void | |
1036 | xfs_qm_dqrele( | |
1037 | xfs_dquot_t *dqp) | |
1038 | { | |
7d095257 CH |
1039 | if (!dqp) |
1040 | return; | |
1041 | ||
0b1b213f | 1042 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
1043 | |
1044 | xfs_dqlock(dqp); | |
1045 | /* | |
1046 | * We don't care to flush it if the dquot is dirty here. | |
1047 | * That will create stutters that we want to avoid. | |
1048 | * Instead we do a delayed write when we try to reclaim | |
1049 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
1050 | */ | |
1051 | xfs_qm_dqput(dqp); | |
1052 | } | |
1053 | ||
ca30b2a7 CH |
1054 | /* |
1055 | * This is the dquot flushing I/O completion routine. It is called | |
1056 | * from interrupt level when the buffer containing the dquot is | |
1057 | * flushed to disk. It is responsible for removing the dquot logitem | |
1058 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
1059 | * flush lock. This behavior is very similar to that of inodes.. | |
1060 | */ | |
1061 | STATIC void | |
1062 | xfs_qm_dqflush_done( | |
1063 | struct xfs_buf *bp, | |
1064 | struct xfs_log_item *lip) | |
1065 | { | |
1066 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
1067 | xfs_dquot_t *dqp = qip->qli_dquot; | |
1068 | struct xfs_ail *ailp = lip->li_ailp; | |
1069 | ||
1070 | /* | |
1071 | * We only want to pull the item from the AIL if its | |
1072 | * location in the log has not changed since we started the flush. | |
1073 | * Thus, we only bother if the dquot's lsn has | |
1074 | * not changed. First we check the lsn outside the lock | |
1075 | * since it's cheaper, and then we recheck while | |
1076 | * holding the lock before removing the dquot from the AIL. | |
1077 | */ | |
1078 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
1079 | lip->li_lsn == qip->qli_flush_lsn) { | |
1080 | ||
1081 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
1082 | spin_lock(&ailp->xa_lock); | |
1083 | if (lip->li_lsn == qip->qli_flush_lsn) | |
1084 | xfs_trans_ail_delete(ailp, lip); | |
1085 | else | |
1086 | spin_unlock(&ailp->xa_lock); | |
1087 | } | |
1088 | ||
1089 | /* | |
1090 | * Release the dq's flush lock since we're done with it. | |
1091 | */ | |
1092 | xfs_dqfunlock(dqp); | |
1093 | } | |
1da177e4 LT |
1094 | |
1095 | /* | |
1096 | * Write a modified dquot to disk. | |
1097 | * The dquot must be locked and the flush lock too taken by caller. | |
1098 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
1099 | * but the dquot is free to be unlocked and modified by the caller | |
1100 | * in the interim. Dquot is still locked on return. This behavior is | |
1101 | * identical to that of inodes. | |
1102 | */ | |
1103 | int | |
1104 | xfs_qm_dqflush( | |
1105 | xfs_dquot_t *dqp, | |
1106 | uint flags) | |
1107 | { | |
acecf1b5 CH |
1108 | struct xfs_mount *mp = dqp->q_mount; |
1109 | struct xfs_buf *bp; | |
1110 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 1111 | int error; |
1da177e4 LT |
1112 | |
1113 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 1114 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 1115 | |
0b1b213f | 1116 | trace_xfs_dqflush(dqp); |
1da177e4 LT |
1117 | |
1118 | /* | |
acecf1b5 | 1119 | * If not dirty, or it's pinned and we are not supposed to block, nada. |
1da177e4 | 1120 | */ |
2f8a3ce1 | 1121 | if (!XFS_DQ_IS_DIRTY(dqp) || |
fdedf28b | 1122 | ((flags & SYNC_TRYLOCK) && atomic_read(&dqp->q_pincount) > 0)) { |
1da177e4 | 1123 | xfs_dqfunlock(dqp); |
2f8a3ce1 | 1124 | return 0; |
1da177e4 | 1125 | } |
1da177e4 LT |
1126 | xfs_qm_dqunpin_wait(dqp); |
1127 | ||
1128 | /* | |
1129 | * This may have been unpinned because the filesystem is shutting | |
1130 | * down forcibly. If that's the case we must not write this dquot | |
1131 | * to disk, because the log record didn't make it to disk! | |
1132 | */ | |
acecf1b5 CH |
1133 | if (XFS_FORCED_SHUTDOWN(mp)) { |
1134 | dqp->dq_flags &= ~XFS_DQ_DIRTY; | |
1da177e4 LT |
1135 | xfs_dqfunlock(dqp); |
1136 | return XFS_ERROR(EIO); | |
1137 | } | |
1138 | ||
1139 | /* | |
1140 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1141 | */ |
acecf1b5 CH |
1142 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
1143 | mp->m_quotainfo->qi_dqchunklen, 0, &bp); | |
1144 | if (error) { | |
1da177e4 | 1145 | ASSERT(error != ENOENT); |
1da177e4 | 1146 | xfs_dqfunlock(dqp); |
acecf1b5 | 1147 | return error; |
1da177e4 LT |
1148 | } |
1149 | ||
acecf1b5 CH |
1150 | /* |
1151 | * Calculate the location of the dquot inside the buffer. | |
1152 | */ | |
62926044 | 1153 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1154 | |
1155 | /* | |
1156 | * A simple sanity check in case we got a corrupted dquot.. | |
1157 | */ | |
a0fa2b67 DC |
1158 | error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
1159 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); | |
1160 | if (error) { | |
acecf1b5 CH |
1161 | xfs_buf_relse(bp); |
1162 | xfs_dqfunlock(dqp); | |
1163 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
1da177e4 LT |
1164 | return XFS_ERROR(EIO); |
1165 | } | |
1166 | ||
1167 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1168 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1169 | |
1170 | /* | |
1171 | * Clear the dirty field and remember the flush lsn for later use. | |
1172 | */ | |
acecf1b5 | 1173 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1174 | |
7b2e2a31 DC |
1175 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1176 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 LT |
1177 | |
1178 | /* | |
1179 | * Attach an iodone routine so that we can remove this dquot from the | |
1180 | * AIL and release the flush lock once the dquot is synced to disk. | |
1181 | */ | |
ca30b2a7 CH |
1182 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1183 | &dqp->q_logitem.qli_item); | |
1184 | ||
1da177e4 LT |
1185 | /* |
1186 | * If the buffer is pinned then push on the log so we won't | |
1187 | * get stuck waiting in the write for too long. | |
1188 | */ | |
811e64c7 | 1189 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1190 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1191 | xfs_log_force(mp, 0); |
1da177e4 LT |
1192 | } |
1193 | ||
20026d92 | 1194 | if (flags & SYNC_WAIT) |
c2b006c1 CH |
1195 | error = xfs_bwrite(bp); |
1196 | else | |
61551f1e | 1197 | xfs_buf_delwri_queue(bp); |
c2b006c1 CH |
1198 | |
1199 | xfs_buf_relse(bp); | |
0b1b213f CH |
1200 | |
1201 | trace_xfs_dqflush_done(dqp); | |
1202 | ||
1da177e4 LT |
1203 | /* |
1204 | * dqp is still locked, but caller is free to unlock it now. | |
1205 | */ | |
20026d92 | 1206 | return error; |
1da177e4 LT |
1207 | |
1208 | } | |
1209 | ||
1da177e4 LT |
1210 | void |
1211 | xfs_dqunlock( | |
1212 | xfs_dquot_t *dqp) | |
1213 | { | |
800b484e | 1214 | xfs_dqunlock_nonotify(dqp); |
1da177e4 | 1215 | if (dqp->q_logitem.qli_dquot == dqp) { |
783a2f65 | 1216 | xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_ailp, |
800b484e | 1217 | &dqp->q_logitem.qli_item); |
1da177e4 LT |
1218 | } |
1219 | } | |
1220 | ||
5bb87a33 CH |
1221 | /* |
1222 | * Lock two xfs_dquot structures. | |
1223 | * | |
1224 | * To avoid deadlocks we always lock the quota structure with | |
1225 | * the lowerd id first. | |
1226 | */ | |
1da177e4 LT |
1227 | void |
1228 | xfs_dqlock2( | |
1229 | xfs_dquot_t *d1, | |
1230 | xfs_dquot_t *d2) | |
1231 | { | |
1232 | if (d1 && d2) { | |
1233 | ASSERT(d1 != d2); | |
1149d96a CH |
1234 | if (be32_to_cpu(d1->q_core.d_id) > |
1235 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1236 | mutex_lock(&d2->q_qlock); |
1237 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1238 | } else { |
5bb87a33 CH |
1239 | mutex_lock(&d1->q_qlock); |
1240 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1241 | } |
5bb87a33 CH |
1242 | } else if (d1) { |
1243 | mutex_lock(&d1->q_qlock); | |
1244 | } else if (d2) { | |
1245 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1246 | } |
1247 | } | |
1248 | ||
1da177e4 | 1249 | /* |
92678554 CH |
1250 | * Take a dquot out of the mount's dqlist as well as the hashlist. This is |
1251 | * called via unmount as well as quotaoff, and the purge will always succeed. | |
1da177e4 | 1252 | */ |
92678554 | 1253 | void |
1da177e4 | 1254 | xfs_qm_dqpurge( |
92678554 | 1255 | struct xfs_dquot *dqp) |
1da177e4 | 1256 | { |
92678554 CH |
1257 | struct xfs_mount *mp = dqp->q_mount; |
1258 | struct xfs_dqhash *qh = dqp->q_hash; | |
80a376bf | 1259 | |
1da177e4 | 1260 | xfs_dqlock(dqp); |
1da177e4 LT |
1261 | |
1262 | /* | |
1263 | * If we're turning off quotas, we have to make sure that, for | |
1264 | * example, we don't delete quota disk blocks while dquots are | |
1265 | * in the process of getting written to those disk blocks. | |
1266 | * This dquot might well be on AIL, and we can't leave it there | |
1267 | * if we're turning off quotas. Basically, we need this flush | |
1268 | * lock, and are willing to block on it. | |
1269 | */ | |
e1f49cf2 | 1270 | if (!xfs_dqflock_nowait(dqp)) { |
1da177e4 LT |
1271 | /* |
1272 | * Block on the flush lock after nudging dquot buffer, | |
1273 | * if it is incore. | |
1274 | */ | |
800b484e | 1275 | xfs_dqflock_pushbuf_wait(dqp); |
1da177e4 LT |
1276 | } |
1277 | ||
1278 | /* | |
92678554 | 1279 | * If we are turning this type of quotas off, we don't care |
1da177e4 LT |
1280 | * about the dirty metadata sitting in this dquot. OTOH, if |
1281 | * we're unmounting, we do care, so we flush it and wait. | |
1282 | */ | |
1283 | if (XFS_DQ_IS_DIRTY(dqp)) { | |
3c56836f | 1284 | int error; |
0b1b213f | 1285 | |
1da177e4 | 1286 | /* |
1da177e4 LT |
1287 | * We don't care about getting disk errors here. We need |
1288 | * to purge this dquot anyway, so we go ahead regardless. | |
1289 | */ | |
20026d92 | 1290 | error = xfs_qm_dqflush(dqp, SYNC_WAIT); |
3c56836f | 1291 | if (error) |
53487786 DC |
1292 | xfs_warn(mp, "%s: dquot %p flush failed", |
1293 | __func__, dqp); | |
1da177e4 LT |
1294 | xfs_dqflock(dqp); |
1295 | } | |
92678554 | 1296 | |
bc3048e3 | 1297 | ASSERT(atomic_read(&dqp->q_pincount) == 0); |
1da177e4 LT |
1298 | ASSERT(XFS_FORCED_SHUTDOWN(mp) || |
1299 | !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL)); | |
1300 | ||
92678554 CH |
1301 | xfs_dqfunlock(dqp); |
1302 | xfs_dqunlock(dqp); | |
1303 | ||
1304 | mutex_lock(&qh->qh_lock); | |
e6a81f13 DC |
1305 | list_del_init(&dqp->q_hashlist); |
1306 | qh->qh_version++; | |
92678554 | 1307 | mutex_unlock(&qh->qh_lock); |
80a376bf | 1308 | |
92678554 | 1309 | mutex_lock(&mp->m_quotainfo->qi_dqlist_lock); |
3a25404b DC |
1310 | list_del_init(&dqp->q_mplist); |
1311 | mp->m_quotainfo->qi_dqreclaims++; | |
1312 | mp->m_quotainfo->qi_dquots--; | |
92678554 | 1313 | mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock); |
1da177e4 | 1314 | |
92678554 CH |
1315 | /* |
1316 | * We move dquots to the freelist as soon as their reference count | |
1317 | * hits zero, so it really should be on the freelist here. | |
1318 | */ | |
1319 | mutex_lock(&xfs_Gqm->qm_dqfrlist_lock); | |
1320 | ASSERT(!list_empty(&dqp->q_freelist)); | |
80a376bf CH |
1321 | list_del_init(&dqp->q_freelist); |
1322 | xfs_Gqm->qm_dqfrlist_cnt--; | |
80a376bf | 1323 | mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); |
1da177e4 | 1324 | |
80a376bf | 1325 | xfs_qm_dqdestroy(dqp); |
80a376bf | 1326 | } |
1da177e4 | 1327 | |
1da177e4 LT |
1328 | /* |
1329 | * Give the buffer a little push if it is incore and | |
1330 | * wait on the flush lock. | |
1331 | */ | |
1332 | void | |
800b484e | 1333 | xfs_dqflock_pushbuf_wait( |
1da177e4 LT |
1334 | xfs_dquot_t *dqp) |
1335 | { | |
8a7b8a89 | 1336 | xfs_mount_t *mp = dqp->q_mount; |
1da177e4 LT |
1337 | xfs_buf_t *bp; |
1338 | ||
1339 | /* | |
1340 | * Check to see if the dquot has been flushed delayed | |
1341 | * write. If so, grab its buffer and send it | |
1342 | * out immediately. We'll be able to acquire | |
1343 | * the flush lock when the I/O completes. | |
1344 | */ | |
8a7b8a89 CH |
1345 | bp = xfs_incore(mp->m_ddev_targp, dqp->q_blkno, |
1346 | mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK); | |
7d6a7bde DC |
1347 | if (!bp) |
1348 | goto out_lock; | |
1349 | ||
1350 | if (XFS_BUF_ISDELAYWRITE(bp)) { | |
811e64c7 | 1351 | if (xfs_buf_ispinned(bp)) |
8a7b8a89 | 1352 | xfs_log_force(mp, 0); |
7d6a7bde DC |
1353 | xfs_buf_delwri_promote(bp); |
1354 | wake_up_process(bp->b_target->bt_task); | |
1da177e4 | 1355 | } |
7d6a7bde DC |
1356 | xfs_buf_relse(bp); |
1357 | out_lock: | |
1da177e4 LT |
1358 | xfs_dqflock(dqp); |
1359 | } |