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 | ||
1da177e4 LT |
156 | /* |
157 | * If default limits are in force, push them into the dquot now. | |
158 | * We overwrite the dquot limits only if they are zero and this | |
159 | * is not the root dquot. | |
160 | */ | |
161 | void | |
162 | xfs_qm_adjust_dqlimits( | |
163 | xfs_mount_t *mp, | |
164 | xfs_disk_dquot_t *d) | |
165 | { | |
166 | xfs_quotainfo_t *q = mp->m_quotainfo; | |
167 | ||
168 | ASSERT(d->d_id); | |
169 | ||
170 | if (q->qi_bsoftlimit && !d->d_blk_softlimit) | |
1149d96a | 171 | d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit); |
1da177e4 | 172 | if (q->qi_bhardlimit && !d->d_blk_hardlimit) |
1149d96a | 173 | d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit); |
1da177e4 | 174 | if (q->qi_isoftlimit && !d->d_ino_softlimit) |
1149d96a | 175 | d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit); |
1da177e4 | 176 | if (q->qi_ihardlimit && !d->d_ino_hardlimit) |
1149d96a | 177 | d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit); |
1da177e4 | 178 | if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit) |
1149d96a | 179 | d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit); |
1da177e4 | 180 | if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit) |
1149d96a | 181 | d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit); |
1da177e4 LT |
182 | } |
183 | ||
184 | /* | |
185 | * Check the limits and timers of a dquot and start or reset timers | |
186 | * if necessary. | |
187 | * This gets called even when quota enforcement is OFF, which makes our | |
188 | * life a little less complicated. (We just don't reject any quota | |
189 | * reservations in that case, when enforcement is off). | |
190 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
191 | * enforcement's off. | |
192 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
193 | * 'automatically' get started when limits get exceeded. They do |
194 | * get reset to zero, however, when we find the count to be under | |
195 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
196 | */ |
197 | void | |
198 | xfs_qm_adjust_dqtimers( | |
199 | xfs_mount_t *mp, | |
200 | xfs_disk_dquot_t *d) | |
201 | { | |
202 | ASSERT(d->d_id); | |
203 | ||
ea15ab3c | 204 | #ifdef DEBUG |
1149d96a CH |
205 | if (d->d_blk_hardlimit) |
206 | ASSERT(be64_to_cpu(d->d_blk_softlimit) <= | |
207 | be64_to_cpu(d->d_blk_hardlimit)); | |
208 | if (d->d_ino_hardlimit) | |
209 | ASSERT(be64_to_cpu(d->d_ino_softlimit) <= | |
210 | be64_to_cpu(d->d_ino_hardlimit)); | |
211 | if (d->d_rtb_hardlimit) | |
212 | ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= | |
213 | be64_to_cpu(d->d_rtb_hardlimit)); | |
1da177e4 | 214 | #endif |
ea15ab3c | 215 | |
1da177e4 | 216 | if (!d->d_btimer) { |
1149d96a CH |
217 | if ((d->d_blk_softlimit && |
218 | (be64_to_cpu(d->d_bcount) >= | |
219 | be64_to_cpu(d->d_blk_softlimit))) || | |
220 | (d->d_blk_hardlimit && | |
221 | (be64_to_cpu(d->d_bcount) >= | |
222 | be64_to_cpu(d->d_blk_hardlimit)))) { | |
223 | d->d_btimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 224 | mp->m_quotainfo->qi_btimelimit); |
754002b4 NS |
225 | } else { |
226 | d->d_bwarns = 0; | |
1da177e4 LT |
227 | } |
228 | } else { | |
229 | if ((!d->d_blk_softlimit || | |
1149d96a CH |
230 | (be64_to_cpu(d->d_bcount) < |
231 | be64_to_cpu(d->d_blk_softlimit))) && | |
1da177e4 | 232 | (!d->d_blk_hardlimit || |
1149d96a CH |
233 | (be64_to_cpu(d->d_bcount) < |
234 | be64_to_cpu(d->d_blk_hardlimit)))) { | |
1da177e4 LT |
235 | d->d_btimer = 0; |
236 | } | |
237 | } | |
238 | ||
239 | if (!d->d_itimer) { | |
1149d96a CH |
240 | if ((d->d_ino_softlimit && |
241 | (be64_to_cpu(d->d_icount) >= | |
242 | be64_to_cpu(d->d_ino_softlimit))) || | |
243 | (d->d_ino_hardlimit && | |
244 | (be64_to_cpu(d->d_icount) >= | |
245 | be64_to_cpu(d->d_ino_hardlimit)))) { | |
246 | d->d_itimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 247 | mp->m_quotainfo->qi_itimelimit); |
754002b4 NS |
248 | } else { |
249 | d->d_iwarns = 0; | |
1da177e4 LT |
250 | } |
251 | } else { | |
252 | if ((!d->d_ino_softlimit || | |
1149d96a CH |
253 | (be64_to_cpu(d->d_icount) < |
254 | be64_to_cpu(d->d_ino_softlimit))) && | |
1da177e4 | 255 | (!d->d_ino_hardlimit || |
1149d96a CH |
256 | (be64_to_cpu(d->d_icount) < |
257 | be64_to_cpu(d->d_ino_hardlimit)))) { | |
1da177e4 LT |
258 | d->d_itimer = 0; |
259 | } | |
260 | } | |
261 | ||
262 | if (!d->d_rtbtimer) { | |
1149d96a CH |
263 | if ((d->d_rtb_softlimit && |
264 | (be64_to_cpu(d->d_rtbcount) >= | |
265 | be64_to_cpu(d->d_rtb_softlimit))) || | |
266 | (d->d_rtb_hardlimit && | |
267 | (be64_to_cpu(d->d_rtbcount) >= | |
268 | be64_to_cpu(d->d_rtb_hardlimit)))) { | |
269 | d->d_rtbtimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 270 | mp->m_quotainfo->qi_rtbtimelimit); |
754002b4 NS |
271 | } else { |
272 | d->d_rtbwarns = 0; | |
1da177e4 LT |
273 | } |
274 | } else { | |
275 | if ((!d->d_rtb_softlimit || | |
1149d96a CH |
276 | (be64_to_cpu(d->d_rtbcount) < |
277 | be64_to_cpu(d->d_rtb_softlimit))) && | |
1da177e4 | 278 | (!d->d_rtb_hardlimit || |
1149d96a CH |
279 | (be64_to_cpu(d->d_rtbcount) < |
280 | be64_to_cpu(d->d_rtb_hardlimit)))) { | |
1da177e4 LT |
281 | d->d_rtbtimer = 0; |
282 | } | |
283 | } | |
284 | } | |
285 | ||
1da177e4 LT |
286 | /* |
287 | * initialize a buffer full of dquots and log the whole thing | |
288 | */ | |
289 | STATIC void | |
290 | xfs_qm_init_dquot_blk( | |
291 | xfs_trans_t *tp, | |
292 | xfs_mount_t *mp, | |
293 | xfs_dqid_t id, | |
294 | uint type, | |
295 | xfs_buf_t *bp) | |
296 | { | |
8a7b8a89 | 297 | struct xfs_quotainfo *q = mp->m_quotainfo; |
1da177e4 LT |
298 | xfs_dqblk_t *d; |
299 | int curid, i; | |
300 | ||
301 | ASSERT(tp); | |
0c842ad4 | 302 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 303 | |
62926044 | 304 | d = bp->b_addr; |
1da177e4 LT |
305 | |
306 | /* | |
307 | * ID of the first dquot in the block - id's are zero based. | |
308 | */ | |
8a7b8a89 | 309 | curid = id - (id % q->qi_dqperchunk); |
1da177e4 | 310 | ASSERT(curid >= 0); |
8a7b8a89 | 311 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
49d35a5c CH |
312 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) { |
313 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
314 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
315 | d->dd_diskdq.d_id = cpu_to_be32(curid); | |
316 | d->dd_diskdq.d_flags = type; | |
317 | } | |
318 | ||
1da177e4 | 319 | xfs_trans_dquot_buf(tp, bp, |
c1155410 DC |
320 | (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : |
321 | ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : | |
322 | XFS_BLF_GDQUOT_BUF))); | |
8a7b8a89 | 323 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); |
1da177e4 LT |
324 | } |
325 | ||
326 | ||
327 | ||
328 | /* | |
329 | * Allocate a block and fill it with dquots. | |
330 | * This is called when the bmapi finds a hole. | |
331 | */ | |
332 | STATIC int | |
333 | xfs_qm_dqalloc( | |
efa092f3 | 334 | xfs_trans_t **tpp, |
1da177e4 LT |
335 | xfs_mount_t *mp, |
336 | xfs_dquot_t *dqp, | |
337 | xfs_inode_t *quotip, | |
338 | xfs_fileoff_t offset_fsb, | |
339 | xfs_buf_t **O_bpp) | |
340 | { | |
341 | xfs_fsblock_t firstblock; | |
342 | xfs_bmap_free_t flist; | |
343 | xfs_bmbt_irec_t map; | |
344 | int nmaps, error, committed; | |
345 | xfs_buf_t *bp; | |
efa092f3 | 346 | xfs_trans_t *tp = *tpp; |
1da177e4 LT |
347 | |
348 | ASSERT(tp != NULL); | |
0b1b213f CH |
349 | |
350 | trace_xfs_dqalloc(dqp); | |
1da177e4 LT |
351 | |
352 | /* | |
353 | * Initialize the bmap freelist prior to calling bmapi code. | |
354 | */ | |
9d87c319 | 355 | xfs_bmap_init(&flist, &firstblock); |
1da177e4 LT |
356 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
357 | /* | |
358 | * Return if this type of quotas is turned off while we didn't | |
359 | * have an inode lock | |
360 | */ | |
361 | if (XFS_IS_THIS_QUOTA_OFF(dqp)) { | |
362 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
363 | return (ESRCH); | |
364 | } | |
365 | ||
ddc3415a | 366 | xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); |
1da177e4 | 367 | nmaps = 1; |
c0dc7828 DC |
368 | error = xfs_bmapi_write(tp, quotip, offset_fsb, |
369 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
370 | &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), | |
371 | &map, &nmaps, &flist); | |
372 | if (error) | |
1da177e4 | 373 | goto error0; |
1da177e4 LT |
374 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
375 | ASSERT(nmaps == 1); | |
376 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
377 | (map.br_startblock != HOLESTARTBLOCK)); | |
378 | ||
379 | /* | |
380 | * Keep track of the blkno to save a lookup later | |
381 | */ | |
382 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
383 | ||
384 | /* now we can just get the buffer (there's nothing to read yet) */ | |
385 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
386 | dqp->q_blkno, | |
8a7b8a89 | 387 | mp->m_quotainfo->qi_dqchunklen, |
1da177e4 | 388 | 0); |
2a30f36d CS |
389 | |
390 | error = xfs_buf_geterror(bp); | |
391 | if (error) | |
1da177e4 | 392 | goto error1; |
2a30f36d | 393 | |
1da177e4 LT |
394 | /* |
395 | * Make a chunk of dquots out of this buffer and log | |
396 | * the entire thing. | |
397 | */ | |
1149d96a | 398 | xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 399 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
1da177e4 | 400 | |
efa092f3 TS |
401 | /* |
402 | * xfs_bmap_finish() may commit the current transaction and | |
403 | * start a second transaction if the freelist is not empty. | |
404 | * | |
405 | * Since we still want to modify this buffer, we need to | |
406 | * ensure that the buffer is not released on commit of | |
407 | * the first transaction and ensure the buffer is added to the | |
408 | * second transaction. | |
409 | * | |
410 | * If there is only one transaction then don't stop the buffer | |
411 | * from being released when it commits later on. | |
412 | */ | |
413 | ||
414 | xfs_trans_bhold(tp, bp); | |
415 | ||
f7c99b6f | 416 | if ((error = xfs_bmap_finish(tpp, &flist, &committed))) { |
1da177e4 LT |
417 | goto error1; |
418 | } | |
419 | ||
efa092f3 TS |
420 | if (committed) { |
421 | tp = *tpp; | |
422 | xfs_trans_bjoin(tp, bp); | |
423 | } else { | |
424 | xfs_trans_bhold_release(tp, bp); | |
425 | } | |
426 | ||
1da177e4 LT |
427 | *O_bpp = bp; |
428 | return 0; | |
429 | ||
430 | error1: | |
431 | xfs_bmap_cancel(&flist); | |
432 | error0: | |
433 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
434 | ||
435 | return (error); | |
436 | } | |
437 | ||
438 | /* | |
439 | * Maps a dquot to the buffer containing its on-disk version. | |
440 | * This returns a ptr to the buffer containing the on-disk dquot | |
441 | * in the bpp param, and a ptr to the on-disk dquot within that buffer | |
442 | */ | |
443 | STATIC int | |
444 | xfs_qm_dqtobp( | |
efa092f3 | 445 | xfs_trans_t **tpp, |
1da177e4 LT |
446 | xfs_dquot_t *dqp, |
447 | xfs_disk_dquot_t **O_ddpp, | |
448 | xfs_buf_t **O_bpp, | |
449 | uint flags) | |
450 | { | |
451 | xfs_bmbt_irec_t map; | |
acecf1b5 | 452 | int nmaps = 1, error; |
1da177e4 | 453 | xfs_buf_t *bp; |
acecf1b5 CH |
454 | xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp); |
455 | xfs_mount_t *mp = dqp->q_mount; | |
1da177e4 | 456 | xfs_disk_dquot_t *ddq; |
acecf1b5 | 457 | xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); |
efa092f3 | 458 | xfs_trans_t *tp = (tpp ? *tpp : NULL); |
1da177e4 | 459 | |
acecf1b5 | 460 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
1da177e4 | 461 | |
acecf1b5 CH |
462 | xfs_ilock(quotip, XFS_ILOCK_SHARED); |
463 | if (XFS_IS_THIS_QUOTA_OFF(dqp)) { | |
1da177e4 | 464 | /* |
acecf1b5 CH |
465 | * Return if this type of quotas is turned off while we |
466 | * didn't have the quota inode lock. | |
1da177e4 | 467 | */ |
acecf1b5 CH |
468 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); |
469 | return ESRCH; | |
470 | } | |
471 | ||
472 | /* | |
473 | * Find the block map; no allocations yet | |
474 | */ | |
5c8ed202 DC |
475 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
476 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); | |
acecf1b5 CH |
477 | |
478 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); | |
479 | if (error) | |
480 | return error; | |
481 | ||
482 | ASSERT(nmaps == 1); | |
483 | ASSERT(map.br_blockcount == 1); | |
484 | ||
485 | /* | |
486 | * Offset of dquot in the (fixed sized) dquot chunk. | |
487 | */ | |
488 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
489 | sizeof(xfs_dqblk_t); | |
490 | ||
491 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
492 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 493 | /* |
acecf1b5 | 494 | * We don't allocate unless we're asked to |
1da177e4 | 495 | */ |
acecf1b5 CH |
496 | if (!(flags & XFS_QMOPT_DQALLOC)) |
497 | return ENOENT; | |
1da177e4 | 498 | |
acecf1b5 CH |
499 | ASSERT(tp); |
500 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
501 | dqp->q_fileoffset, &bp); | |
1da177e4 | 502 | if (error) |
acecf1b5 CH |
503 | return error; |
504 | tp = *tpp; | |
505 | } else { | |
506 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
507 | |
508 | /* | |
acecf1b5 CH |
509 | * store the blkno etc so that we don't have to do the |
510 | * mapping all the time | |
1da177e4 | 511 | */ |
acecf1b5 | 512 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 513 | |
8a7b8a89 CH |
514 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
515 | dqp->q_blkno, | |
516 | mp->m_quotainfo->qi_dqchunklen, | |
517 | 0, &bp); | |
1da177e4 LT |
518 | if (error || !bp) |
519 | return XFS_ERROR(error); | |
520 | } | |
acecf1b5 | 521 | |
0c842ad4 | 522 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
523 | |
524 | /* | |
525 | * calculate the location of the dquot inside the buffer. | |
526 | */ | |
62926044 | 527 | ddq = bp->b_addr + dqp->q_bufoffset; |
1da177e4 LT |
528 | |
529 | /* | |
530 | * A simple sanity check in case we got a corrupted dquot... | |
531 | */ | |
a0fa2b67 | 532 | error = xfs_qm_dqcheck(mp, ddq, id, dqp->dq_flags & XFS_DQ_ALLTYPES, |
1da177e4 | 533 | flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN), |
a0fa2b67 DC |
534 | "dqtobp"); |
535 | if (error) { | |
1da177e4 LT |
536 | if (!(flags & XFS_QMOPT_DQREPAIR)) { |
537 | xfs_trans_brelse(tp, bp); | |
538 | return XFS_ERROR(EIO); | |
539 | } | |
1da177e4 LT |
540 | } |
541 | ||
542 | *O_bpp = bp; | |
543 | *O_ddpp = ddq; | |
544 | ||
545 | return (0); | |
546 | } | |
547 | ||
548 | ||
549 | /* | |
550 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
551 | * and release the buffer immediately. | |
552 | * | |
553 | */ | |
554 | /* ARGSUSED */ | |
555 | STATIC int | |
556 | xfs_qm_dqread( | |
efa092f3 | 557 | xfs_trans_t **tpp, |
1da177e4 LT |
558 | xfs_dqid_t id, |
559 | xfs_dquot_t *dqp, /* dquot to get filled in */ | |
560 | uint flags) | |
561 | { | |
562 | xfs_disk_dquot_t *ddqp; | |
563 | xfs_buf_t *bp; | |
564 | int error; | |
efa092f3 TS |
565 | xfs_trans_t *tp; |
566 | ||
567 | ASSERT(tpp); | |
1da177e4 | 568 | |
0b1b213f CH |
569 | trace_xfs_dqread(dqp); |
570 | ||
1da177e4 LT |
571 | /* |
572 | * get a pointer to the on-disk dquot and the buffer containing it | |
573 | * dqp already knows its own type (GROUP/USER). | |
574 | */ | |
efa092f3 | 575 | if ((error = xfs_qm_dqtobp(tpp, dqp, &ddqp, &bp, flags))) { |
1da177e4 LT |
576 | return (error); |
577 | } | |
efa092f3 | 578 | tp = *tpp; |
1da177e4 LT |
579 | |
580 | /* copy everything from disk dquot to the incore dquot */ | |
581 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
1149d96a | 582 | ASSERT(be32_to_cpu(dqp->q_core.d_id) == id); |
1da177e4 LT |
583 | xfs_qm_dquot_logitem_init(dqp); |
584 | ||
585 | /* | |
586 | * Reservation counters are defined as reservation plus current usage | |
25985edc | 587 | * to avoid having to add every time. |
1da177e4 | 588 | */ |
1149d96a CH |
589 | dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); |
590 | dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); | |
591 | dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); | |
1da177e4 LT |
592 | |
593 | /* Mark the buf so that this will stay incore a little longer */ | |
38f23232 | 594 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 LT |
595 | |
596 | /* | |
597 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
598 | * So we need to release with xfs_trans_brelse(). | |
599 | * The strategy here is identical to that of inodes; we lock | |
600 | * the dquot in xfs_qm_dqget() before making it accessible to | |
601 | * others. This is because dquots, like inodes, need a good level of | |
602 | * concurrency, and we don't want to take locks on the entire buffers | |
603 | * for dquot accesses. | |
604 | * Note also that the dquot buffer may even be dirty at this point, if | |
605 | * this particular dquot was repaired. We still aren't afraid to | |
606 | * brelse it because we have the changes incore. | |
607 | */ | |
0c842ad4 | 608 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
609 | xfs_trans_brelse(tp, bp); |
610 | ||
611 | return (error); | |
612 | } | |
613 | ||
614 | ||
615 | /* | |
616 | * allocate an incore dquot from the kernel heap, | |
617 | * and fill its core with quota information kept on disk. | |
618 | * If XFS_QMOPT_DQALLOC is set, it'll allocate a dquot on disk | |
619 | * if it wasn't already allocated. | |
620 | */ | |
621 | STATIC int | |
622 | xfs_qm_idtodq( | |
623 | xfs_mount_t *mp, | |
624 | xfs_dqid_t id, /* gid or uid, depending on type */ | |
625 | uint type, /* UDQUOT or GDQUOT */ | |
626 | uint flags, /* DQALLOC, DQREPAIR */ | |
627 | xfs_dquot_t **O_dqpp)/* OUT : incore dquot, not locked */ | |
628 | { | |
629 | xfs_dquot_t *dqp; | |
630 | int error; | |
631 | xfs_trans_t *tp; | |
632 | int cancelflags=0; | |
633 | ||
634 | dqp = xfs_qm_dqinit(mp, id, type); | |
635 | tp = NULL; | |
636 | if (flags & XFS_QMOPT_DQALLOC) { | |
637 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC); | |
8a7b8a89 CH |
638 | error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp), |
639 | XFS_WRITE_LOG_RES(mp) + | |
640 | BBTOB(mp->m_quotainfo->qi_dqchunklen) - 1 + | |
641 | 128, | |
642 | 0, | |
643 | XFS_TRANS_PERM_LOG_RES, | |
644 | XFS_WRITE_LOG_COUNT); | |
645 | if (error) { | |
1da177e4 LT |
646 | cancelflags = 0; |
647 | goto error0; | |
648 | } | |
649 | cancelflags = XFS_TRANS_RELEASE_LOG_RES; | |
650 | } | |
651 | ||
652 | /* | |
653 | * Read it from disk; xfs_dqread() takes care of | |
654 | * all the necessary initialization of dquot's fields (locks, etc) | |
655 | */ | |
efa092f3 | 656 | if ((error = xfs_qm_dqread(&tp, id, dqp, flags))) { |
1da177e4 LT |
657 | /* |
658 | * This can happen if quotas got turned off (ESRCH), | |
659 | * or if the dquot didn't exist on disk and we ask to | |
660 | * allocate (ENOENT). | |
661 | */ | |
0b1b213f | 662 | trace_xfs_dqread_fail(dqp); |
1da177e4 LT |
663 | cancelflags |= XFS_TRANS_ABORT; |
664 | goto error0; | |
665 | } | |
666 | if (tp) { | |
1c72bf90 | 667 | if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) |
1da177e4 LT |
668 | goto error1; |
669 | } | |
670 | ||
671 | *O_dqpp = dqp; | |
672 | return (0); | |
673 | ||
674 | error0: | |
675 | ASSERT(error); | |
676 | if (tp) | |
677 | xfs_trans_cancel(tp, cancelflags); | |
678 | error1: | |
679 | xfs_qm_dqdestroy(dqp); | |
680 | *O_dqpp = NULL; | |
681 | return (error); | |
682 | } | |
683 | ||
684 | /* | |
685 | * Lookup a dquot in the incore dquot hashtable. We keep two separate | |
686 | * hashtables for user and group dquots; and, these are global tables | |
687 | * inside the XQM, not per-filesystem tables. | |
688 | * The hash chain must be locked by caller, and it is left locked | |
689 | * on return. Returning dquot is locked. | |
690 | */ | |
691 | STATIC int | |
692 | xfs_qm_dqlookup( | |
693 | xfs_mount_t *mp, | |
694 | xfs_dqid_t id, | |
695 | xfs_dqhash_t *qh, | |
696 | xfs_dquot_t **O_dqpp) | |
697 | { | |
698 | xfs_dquot_t *dqp; | |
1da177e4 | 699 | |
c9a192dc | 700 | ASSERT(mutex_is_locked(&qh->qh_lock)); |
1da177e4 | 701 | |
1da177e4 LT |
702 | /* |
703 | * Traverse the hashchain looking for a match | |
704 | */ | |
e6a81f13 | 705 | list_for_each_entry(dqp, &qh->qh_list, q_hashlist) { |
1da177e4 LT |
706 | /* |
707 | * We already have the hashlock. We don't need the | |
708 | * dqlock to look at the id field of the dquot, since the | |
709 | * id can't be modified without the hashlock anyway. | |
710 | */ | |
be7ffc38 CH |
711 | if (be32_to_cpu(dqp->q_core.d_id) != id || dqp->q_mount != mp) |
712 | continue; | |
0b1b213f | 713 | |
be7ffc38 | 714 | trace_xfs_dqlookup_found(dqp); |
1da177e4 | 715 | |
be7ffc38 | 716 | xfs_dqlock(dqp); |
92678554 CH |
717 | if (dqp->dq_flags & XFS_DQ_FREEING) { |
718 | *O_dqpp = NULL; | |
719 | xfs_dqunlock(dqp); | |
720 | return -1; | |
721 | } | |
722 | ||
78e55892 | 723 | dqp->q_nrefs++; |
be7ffc38 CH |
724 | |
725 | /* | |
726 | * move the dquot to the front of the hashchain | |
727 | */ | |
728 | list_move(&dqp->q_hashlist, &qh->qh_list); | |
729 | trace_xfs_dqlookup_done(dqp); | |
730 | *O_dqpp = dqp; | |
731 | return 0; | |
1da177e4 LT |
732 | } |
733 | ||
734 | *O_dqpp = NULL; | |
be7ffc38 | 735 | return 1; |
1da177e4 LT |
736 | } |
737 | ||
738 | /* | |
739 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
740 | * a locked dquot, doing an allocation (if requested) as needed. | |
741 | * When both an inode and an id are given, the inode's id takes precedence. | |
742 | * That is, if the id changes while we don't hold the ilock inside this | |
743 | * function, the new dquot is returned, not necessarily the one requested | |
744 | * in the id argument. | |
745 | */ | |
746 | int | |
747 | xfs_qm_dqget( | |
748 | xfs_mount_t *mp, | |
749 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
750 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
751 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
752 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
753 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
754 | { | |
755 | xfs_dquot_t *dqp; | |
756 | xfs_dqhash_t *h; | |
757 | uint version; | |
758 | int error; | |
759 | ||
760 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
761 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 762 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 LT |
763 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
764 | return (ESRCH); | |
765 | } | |
766 | h = XFS_DQ_HASH(mp, id, type); | |
767 | ||
768 | #ifdef DEBUG | |
769 | if (xfs_do_dqerror) { | |
770 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
771 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 772 | xfs_debug(mp, "Returning error in dqget"); |
1da177e4 LT |
773 | return (EIO); |
774 | } | |
775 | } | |
1da177e4 | 776 | |
c8ad20ff NS |
777 | ASSERT(type == XFS_DQ_USER || |
778 | type == XFS_DQ_PROJ || | |
779 | type == XFS_DQ_GROUP); | |
1da177e4 | 780 | if (ip) { |
579aa9ca | 781 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
782 | if (type == XFS_DQ_USER) |
783 | ASSERT(ip->i_udquot == NULL); | |
784 | else | |
785 | ASSERT(ip->i_gdquot == NULL); | |
786 | } | |
787 | #endif | |
92678554 CH |
788 | |
789 | restart: | |
c9a192dc | 790 | mutex_lock(&h->qh_lock); |
1da177e4 LT |
791 | |
792 | /* | |
793 | * Look in the cache (hashtable). | |
794 | * The chain is kept locked during lookup. | |
795 | */ | |
92678554 CH |
796 | switch (xfs_qm_dqlookup(mp, id, h, O_dqpp)) { |
797 | case -1: | |
798 | XQM_STATS_INC(xqmstats.xs_qm_dquot_dups); | |
799 | mutex_unlock(&h->qh_lock); | |
800 | delay(1); | |
801 | goto restart; | |
802 | case 0: | |
1da177e4 LT |
803 | XQM_STATS_INC(xqmstats.xs_qm_dqcachehits); |
804 | /* | |
805 | * The dquot was found, moved to the front of the chain, | |
806 | * taken off the freelist if it was on it, and locked | |
807 | * at this point. Just unlock the hashchain and return. | |
808 | */ | |
809 | ASSERT(*O_dqpp); | |
810 | ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp)); | |
c9a192dc | 811 | mutex_unlock(&h->qh_lock); |
0b1b213f | 812 | trace_xfs_dqget_hit(*O_dqpp); |
92678554 CH |
813 | return 0; /* success */ |
814 | default: | |
815 | XQM_STATS_INC(xqmstats.xs_qm_dqcachemisses); | |
816 | break; | |
1da177e4 | 817 | } |
1da177e4 LT |
818 | |
819 | /* | |
820 | * Dquot cache miss. We don't want to keep the inode lock across | |
821 | * a (potential) disk read. Also we don't want to deal with the lock | |
822 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
823 | * lock here means dealing with a chown that can happen before | |
824 | * we re-acquire the lock. | |
825 | */ | |
826 | if (ip) | |
827 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
828 | /* | |
829 | * Save the hashchain version stamp, and unlock the chain, so that | |
830 | * we don't keep the lock across a disk read | |
831 | */ | |
832 | version = h->qh_version; | |
c9a192dc | 833 | mutex_unlock(&h->qh_lock); |
1da177e4 LT |
834 | |
835 | /* | |
836 | * Allocate the dquot on the kernel heap, and read the ondisk | |
837 | * portion off the disk. Also, do all the necessary initialization | |
838 | * This can return ENOENT if dquot didn't exist on disk and we didn't | |
839 | * ask it to allocate; ESRCH if quotas got turned off suddenly. | |
840 | */ | |
841 | if ((error = xfs_qm_idtodq(mp, id, type, | |
842 | flags & (XFS_QMOPT_DQALLOC|XFS_QMOPT_DQREPAIR| | |
843 | XFS_QMOPT_DOWARN), | |
844 | &dqp))) { | |
845 | if (ip) | |
846 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
847 | return (error); | |
848 | } | |
849 | ||
850 | /* | |
851 | * See if this is mount code calling to look at the overall quota limits | |
852 | * which are stored in the id == 0 user or group's dquot. | |
853 | * Since we may not have done a quotacheck by this point, just return | |
854 | * the dquot without attaching it to any hashtables, lists, etc, or even | |
855 | * taking a reference. | |
856 | * The caller must dqdestroy this once done. | |
857 | */ | |
858 | if (flags & XFS_QMOPT_DQSUSER) { | |
859 | ASSERT(id == 0); | |
860 | ASSERT(! ip); | |
861 | goto dqret; | |
862 | } | |
863 | ||
864 | /* | |
865 | * Dquot lock comes after hashlock in the lock ordering | |
866 | */ | |
867 | if (ip) { | |
868 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
191f8488 | 869 | |
1da177e4 LT |
870 | /* |
871 | * A dquot could be attached to this inode by now, since | |
872 | * we had dropped the ilock. | |
873 | */ | |
874 | if (type == XFS_DQ_USER) { | |
191f8488 CH |
875 | if (!XFS_IS_UQUOTA_ON(mp)) { |
876 | /* inode stays locked on return */ | |
877 | xfs_qm_dqdestroy(dqp); | |
878 | return XFS_ERROR(ESRCH); | |
879 | } | |
1da177e4 LT |
880 | if (ip->i_udquot) { |
881 | xfs_qm_dqdestroy(dqp); | |
882 | dqp = ip->i_udquot; | |
883 | xfs_dqlock(dqp); | |
884 | goto dqret; | |
885 | } | |
886 | } else { | |
191f8488 CH |
887 | if (!XFS_IS_OQUOTA_ON(mp)) { |
888 | /* inode stays locked on return */ | |
889 | xfs_qm_dqdestroy(dqp); | |
890 | return XFS_ERROR(ESRCH); | |
891 | } | |
1da177e4 LT |
892 | if (ip->i_gdquot) { |
893 | xfs_qm_dqdestroy(dqp); | |
894 | dqp = ip->i_gdquot; | |
895 | xfs_dqlock(dqp); | |
896 | goto dqret; | |
897 | } | |
898 | } | |
899 | } | |
900 | ||
901 | /* | |
902 | * Hashlock comes after ilock in lock order | |
903 | */ | |
c9a192dc | 904 | mutex_lock(&h->qh_lock); |
1da177e4 LT |
905 | if (version != h->qh_version) { |
906 | xfs_dquot_t *tmpdqp; | |
907 | /* | |
908 | * Now, see if somebody else put the dquot in the | |
909 | * hashtable before us. This can happen because we didn't | |
910 | * keep the hashchain lock. We don't have to worry about | |
911 | * lock order between the two dquots here since dqp isn't | |
912 | * on any findable lists yet. | |
913 | */ | |
92678554 CH |
914 | switch (xfs_qm_dqlookup(mp, id, h, &tmpdqp)) { |
915 | case 0: | |
916 | case -1: | |
1da177e4 | 917 | /* |
92678554 CH |
918 | * Duplicate found, either in cache or on its way out. |
919 | * Just throw away the new dquot and start over. | |
1da177e4 | 920 | */ |
92678554 CH |
921 | if (tmpdqp) |
922 | xfs_qm_dqput(tmpdqp); | |
c9a192dc | 923 | mutex_unlock(&h->qh_lock); |
1da177e4 LT |
924 | xfs_qm_dqdestroy(dqp); |
925 | XQM_STATS_INC(xqmstats.xs_qm_dquot_dups); | |
92678554 CH |
926 | goto restart; |
927 | default: | |
928 | break; | |
1da177e4 LT |
929 | } |
930 | } | |
931 | ||
932 | /* | |
933 | * Put the dquot at the beginning of the hash-chain and mp's list | |
934 | * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock .. | |
935 | */ | |
c9a192dc | 936 | ASSERT(mutex_is_locked(&h->qh_lock)); |
1da177e4 | 937 | dqp->q_hash = h; |
e6a81f13 DC |
938 | list_add(&dqp->q_hashlist, &h->qh_list); |
939 | h->qh_version++; | |
1da177e4 LT |
940 | |
941 | /* | |
942 | * Attach this dquot to this filesystem's list of all dquots, | |
943 | * kept inside the mount structure in m_quotainfo field | |
944 | */ | |
3a25404b | 945 | mutex_lock(&mp->m_quotainfo->qi_dqlist_lock); |
1da177e4 LT |
946 | |
947 | /* | |
948 | * We return a locked dquot to the caller, with a reference taken | |
949 | */ | |
950 | xfs_dqlock(dqp); | |
951 | dqp->q_nrefs = 1; | |
952 | ||
3a25404b DC |
953 | list_add(&dqp->q_mplist, &mp->m_quotainfo->qi_dqlist); |
954 | mp->m_quotainfo->qi_dquots++; | |
955 | mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock); | |
c9a192dc | 956 | mutex_unlock(&h->qh_lock); |
1da177e4 | 957 | dqret: |
579aa9ca | 958 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 959 | trace_xfs_dqget_miss(dqp); |
1da177e4 LT |
960 | *O_dqpp = dqp; |
961 | return (0); | |
962 | } | |
963 | ||
964 | ||
965 | /* | |
966 | * Release a reference to the dquot (decrement ref-count) | |
967 | * and unlock it. If there is a group quota attached to this | |
968 | * dquot, carefully release that too without tripping over | |
969 | * deadlocks'n'stuff. | |
970 | */ | |
971 | void | |
972 | xfs_qm_dqput( | |
bf72de31 | 973 | struct xfs_dquot *dqp) |
1da177e4 | 974 | { |
bf72de31 | 975 | struct xfs_dquot *gdqp; |
1da177e4 LT |
976 | |
977 | ASSERT(dqp->q_nrefs > 0); | |
978 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
0b1b213f CH |
979 | |
980 | trace_xfs_dqput(dqp); | |
1da177e4 | 981 | |
bf72de31 CH |
982 | recurse: |
983 | if (--dqp->q_nrefs > 0) { | |
1da177e4 LT |
984 | xfs_dqunlock(dqp); |
985 | return; | |
986 | } | |
987 | ||
bf72de31 CH |
988 | trace_xfs_dqput_free(dqp); |
989 | ||
990 | mutex_lock(&xfs_Gqm->qm_dqfrlist_lock); | |
991 | if (list_empty(&dqp->q_freelist)) { | |
992 | list_add_tail(&dqp->q_freelist, &xfs_Gqm->qm_dqfrlist); | |
993 | xfs_Gqm->qm_dqfrlist_cnt++; | |
994 | } | |
995 | mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); | |
996 | ||
1da177e4 | 997 | /* |
bf72de31 CH |
998 | * If we just added a udquot to the freelist, then we want to release |
999 | * the gdquot reference that it (probably) has. Otherwise it'll keep | |
1000 | * the gdquot from getting reclaimed. | |
1da177e4 | 1001 | */ |
bf72de31 CH |
1002 | gdqp = dqp->q_gdquot; |
1003 | if (gdqp) { | |
1004 | xfs_dqlock(gdqp); | |
1005 | dqp->q_gdquot = NULL; | |
1da177e4 | 1006 | } |
bf72de31 | 1007 | xfs_dqunlock(dqp); |
1da177e4 | 1008 | |
bf72de31 CH |
1009 | /* |
1010 | * If we had a group quota hint, release it now. | |
1011 | */ | |
1012 | if (gdqp) { | |
1da177e4 | 1013 | dqp = gdqp; |
bf72de31 | 1014 | goto recurse; |
1da177e4 | 1015 | } |
1da177e4 LT |
1016 | } |
1017 | ||
1018 | /* | |
1019 | * Release a dquot. Flush it if dirty, then dqput() it. | |
1020 | * dquot must not be locked. | |
1021 | */ | |
1022 | void | |
1023 | xfs_qm_dqrele( | |
1024 | xfs_dquot_t *dqp) | |
1025 | { | |
7d095257 CH |
1026 | if (!dqp) |
1027 | return; | |
1028 | ||
0b1b213f | 1029 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
1030 | |
1031 | xfs_dqlock(dqp); | |
1032 | /* | |
1033 | * We don't care to flush it if the dquot is dirty here. | |
1034 | * That will create stutters that we want to avoid. | |
1035 | * Instead we do a delayed write when we try to reclaim | |
1036 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
1037 | */ | |
1038 | xfs_qm_dqput(dqp); | |
1039 | } | |
1040 | ||
ca30b2a7 CH |
1041 | /* |
1042 | * This is the dquot flushing I/O completion routine. It is called | |
1043 | * from interrupt level when the buffer containing the dquot is | |
1044 | * flushed to disk. It is responsible for removing the dquot logitem | |
1045 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
1046 | * flush lock. This behavior is very similar to that of inodes.. | |
1047 | */ | |
1048 | STATIC void | |
1049 | xfs_qm_dqflush_done( | |
1050 | struct xfs_buf *bp, | |
1051 | struct xfs_log_item *lip) | |
1052 | { | |
1053 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
1054 | xfs_dquot_t *dqp = qip->qli_dquot; | |
1055 | struct xfs_ail *ailp = lip->li_ailp; | |
1056 | ||
1057 | /* | |
1058 | * We only want to pull the item from the AIL if its | |
1059 | * location in the log has not changed since we started the flush. | |
1060 | * Thus, we only bother if the dquot's lsn has | |
1061 | * not changed. First we check the lsn outside the lock | |
1062 | * since it's cheaper, and then we recheck while | |
1063 | * holding the lock before removing the dquot from the AIL. | |
1064 | */ | |
1065 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
1066 | lip->li_lsn == qip->qli_flush_lsn) { | |
1067 | ||
1068 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
1069 | spin_lock(&ailp->xa_lock); | |
1070 | if (lip->li_lsn == qip->qli_flush_lsn) | |
1071 | xfs_trans_ail_delete(ailp, lip); | |
1072 | else | |
1073 | spin_unlock(&ailp->xa_lock); | |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * Release the dq's flush lock since we're done with it. | |
1078 | */ | |
1079 | xfs_dqfunlock(dqp); | |
1080 | } | |
1da177e4 LT |
1081 | |
1082 | /* | |
1083 | * Write a modified dquot to disk. | |
1084 | * The dquot must be locked and the flush lock too taken by caller. | |
1085 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
1086 | * but the dquot is free to be unlocked and modified by the caller | |
1087 | * in the interim. Dquot is still locked on return. This behavior is | |
1088 | * identical to that of inodes. | |
1089 | */ | |
1090 | int | |
1091 | xfs_qm_dqflush( | |
1092 | xfs_dquot_t *dqp, | |
1093 | uint flags) | |
1094 | { | |
acecf1b5 CH |
1095 | struct xfs_mount *mp = dqp->q_mount; |
1096 | struct xfs_buf *bp; | |
1097 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 1098 | int error; |
1da177e4 LT |
1099 | |
1100 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 1101 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 1102 | |
0b1b213f | 1103 | trace_xfs_dqflush(dqp); |
1da177e4 LT |
1104 | |
1105 | /* | |
acecf1b5 | 1106 | * If not dirty, or it's pinned and we are not supposed to block, nada. |
1da177e4 | 1107 | */ |
2f8a3ce1 | 1108 | if (!XFS_DQ_IS_DIRTY(dqp) || |
fdedf28b | 1109 | ((flags & SYNC_TRYLOCK) && atomic_read(&dqp->q_pincount) > 0)) { |
1da177e4 | 1110 | xfs_dqfunlock(dqp); |
2f8a3ce1 | 1111 | return 0; |
1da177e4 | 1112 | } |
1da177e4 LT |
1113 | xfs_qm_dqunpin_wait(dqp); |
1114 | ||
1115 | /* | |
1116 | * This may have been unpinned because the filesystem is shutting | |
1117 | * down forcibly. If that's the case we must not write this dquot | |
1118 | * to disk, because the log record didn't make it to disk! | |
1119 | */ | |
acecf1b5 CH |
1120 | if (XFS_FORCED_SHUTDOWN(mp)) { |
1121 | dqp->dq_flags &= ~XFS_DQ_DIRTY; | |
1da177e4 LT |
1122 | xfs_dqfunlock(dqp); |
1123 | return XFS_ERROR(EIO); | |
1124 | } | |
1125 | ||
1126 | /* | |
1127 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1128 | */ |
acecf1b5 CH |
1129 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
1130 | mp->m_quotainfo->qi_dqchunklen, 0, &bp); | |
1131 | if (error) { | |
1da177e4 | 1132 | ASSERT(error != ENOENT); |
1da177e4 | 1133 | xfs_dqfunlock(dqp); |
acecf1b5 | 1134 | return error; |
1da177e4 LT |
1135 | } |
1136 | ||
acecf1b5 CH |
1137 | /* |
1138 | * Calculate the location of the dquot inside the buffer. | |
1139 | */ | |
62926044 | 1140 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1141 | |
1142 | /* | |
1143 | * A simple sanity check in case we got a corrupted dquot.. | |
1144 | */ | |
a0fa2b67 DC |
1145 | error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
1146 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); | |
1147 | if (error) { | |
acecf1b5 CH |
1148 | xfs_buf_relse(bp); |
1149 | xfs_dqfunlock(dqp); | |
1150 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
1da177e4 LT |
1151 | return XFS_ERROR(EIO); |
1152 | } | |
1153 | ||
1154 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1155 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1156 | |
1157 | /* | |
1158 | * Clear the dirty field and remember the flush lsn for later use. | |
1159 | */ | |
acecf1b5 | 1160 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1161 | |
7b2e2a31 DC |
1162 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1163 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 LT |
1164 | |
1165 | /* | |
1166 | * Attach an iodone routine so that we can remove this dquot from the | |
1167 | * AIL and release the flush lock once the dquot is synced to disk. | |
1168 | */ | |
ca30b2a7 CH |
1169 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1170 | &dqp->q_logitem.qli_item); | |
1171 | ||
1da177e4 LT |
1172 | /* |
1173 | * If the buffer is pinned then push on the log so we won't | |
1174 | * get stuck waiting in the write for too long. | |
1175 | */ | |
811e64c7 | 1176 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1177 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1178 | xfs_log_force(mp, 0); |
1da177e4 LT |
1179 | } |
1180 | ||
20026d92 | 1181 | if (flags & SYNC_WAIT) |
c2b006c1 CH |
1182 | error = xfs_bwrite(bp); |
1183 | else | |
61551f1e | 1184 | xfs_buf_delwri_queue(bp); |
c2b006c1 CH |
1185 | |
1186 | xfs_buf_relse(bp); | |
0b1b213f CH |
1187 | |
1188 | trace_xfs_dqflush_done(dqp); | |
1189 | ||
1da177e4 LT |
1190 | /* |
1191 | * dqp is still locked, but caller is free to unlock it now. | |
1192 | */ | |
20026d92 | 1193 | return error; |
1da177e4 LT |
1194 | |
1195 | } | |
1196 | ||
1da177e4 LT |
1197 | void |
1198 | xfs_dqunlock( | |
1199 | xfs_dquot_t *dqp) | |
1200 | { | |
800b484e | 1201 | xfs_dqunlock_nonotify(dqp); |
1da177e4 | 1202 | if (dqp->q_logitem.qli_dquot == dqp) { |
783a2f65 | 1203 | xfs_trans_unlocked_item(dqp->q_logitem.qli_item.li_ailp, |
800b484e | 1204 | &dqp->q_logitem.qli_item); |
1da177e4 LT |
1205 | } |
1206 | } | |
1207 | ||
5bb87a33 CH |
1208 | /* |
1209 | * Lock two xfs_dquot structures. | |
1210 | * | |
1211 | * To avoid deadlocks we always lock the quota structure with | |
1212 | * the lowerd id first. | |
1213 | */ | |
1da177e4 LT |
1214 | void |
1215 | xfs_dqlock2( | |
1216 | xfs_dquot_t *d1, | |
1217 | xfs_dquot_t *d2) | |
1218 | { | |
1219 | if (d1 && d2) { | |
1220 | ASSERT(d1 != d2); | |
1149d96a CH |
1221 | if (be32_to_cpu(d1->q_core.d_id) > |
1222 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1223 | mutex_lock(&d2->q_qlock); |
1224 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1225 | } else { |
5bb87a33 CH |
1226 | mutex_lock(&d1->q_qlock); |
1227 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1228 | } |
5bb87a33 CH |
1229 | } else if (d1) { |
1230 | mutex_lock(&d1->q_qlock); | |
1231 | } else if (d2) { | |
1232 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1233 | } |
1234 | } | |
1235 | ||
1da177e4 | 1236 | /* |
92678554 CH |
1237 | * Take a dquot out of the mount's dqlist as well as the hashlist. This is |
1238 | * called via unmount as well as quotaoff, and the purge will always succeed. | |
1da177e4 | 1239 | */ |
92678554 | 1240 | void |
1da177e4 | 1241 | xfs_qm_dqpurge( |
92678554 | 1242 | struct xfs_dquot *dqp) |
1da177e4 | 1243 | { |
92678554 CH |
1244 | struct xfs_mount *mp = dqp->q_mount; |
1245 | struct xfs_dqhash *qh = dqp->q_hash; | |
80a376bf | 1246 | |
1da177e4 | 1247 | xfs_dqlock(dqp); |
1da177e4 LT |
1248 | |
1249 | /* | |
1250 | * If we're turning off quotas, we have to make sure that, for | |
1251 | * example, we don't delete quota disk blocks while dquots are | |
1252 | * in the process of getting written to those disk blocks. | |
1253 | * This dquot might well be on AIL, and we can't leave it there | |
1254 | * if we're turning off quotas. Basically, we need this flush | |
1255 | * lock, and are willing to block on it. | |
1256 | */ | |
e1f49cf2 | 1257 | if (!xfs_dqflock_nowait(dqp)) { |
1da177e4 LT |
1258 | /* |
1259 | * Block on the flush lock after nudging dquot buffer, | |
1260 | * if it is incore. | |
1261 | */ | |
800b484e | 1262 | xfs_dqflock_pushbuf_wait(dqp); |
1da177e4 LT |
1263 | } |
1264 | ||
1265 | /* | |
92678554 | 1266 | * If we are turning this type of quotas off, we don't care |
1da177e4 LT |
1267 | * about the dirty metadata sitting in this dquot. OTOH, if |
1268 | * we're unmounting, we do care, so we flush it and wait. | |
1269 | */ | |
1270 | if (XFS_DQ_IS_DIRTY(dqp)) { | |
3c56836f | 1271 | int error; |
0b1b213f | 1272 | |
1da177e4 | 1273 | /* |
1da177e4 LT |
1274 | * We don't care about getting disk errors here. We need |
1275 | * to purge this dquot anyway, so we go ahead regardless. | |
1276 | */ | |
20026d92 | 1277 | error = xfs_qm_dqflush(dqp, SYNC_WAIT); |
3c56836f | 1278 | if (error) |
53487786 DC |
1279 | xfs_warn(mp, "%s: dquot %p flush failed", |
1280 | __func__, dqp); | |
1da177e4 LT |
1281 | xfs_dqflock(dqp); |
1282 | } | |
92678554 | 1283 | |
bc3048e3 | 1284 | ASSERT(atomic_read(&dqp->q_pincount) == 0); |
1da177e4 LT |
1285 | ASSERT(XFS_FORCED_SHUTDOWN(mp) || |
1286 | !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL)); | |
1287 | ||
92678554 CH |
1288 | xfs_dqfunlock(dqp); |
1289 | xfs_dqunlock(dqp); | |
1290 | ||
1291 | mutex_lock(&qh->qh_lock); | |
e6a81f13 DC |
1292 | list_del_init(&dqp->q_hashlist); |
1293 | qh->qh_version++; | |
92678554 | 1294 | mutex_unlock(&qh->qh_lock); |
80a376bf | 1295 | |
92678554 | 1296 | mutex_lock(&mp->m_quotainfo->qi_dqlist_lock); |
3a25404b DC |
1297 | list_del_init(&dqp->q_mplist); |
1298 | mp->m_quotainfo->qi_dqreclaims++; | |
1299 | mp->m_quotainfo->qi_dquots--; | |
92678554 | 1300 | mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock); |
1da177e4 | 1301 | |
92678554 CH |
1302 | /* |
1303 | * We move dquots to the freelist as soon as their reference count | |
1304 | * hits zero, so it really should be on the freelist here. | |
1305 | */ | |
1306 | mutex_lock(&xfs_Gqm->qm_dqfrlist_lock); | |
1307 | ASSERT(!list_empty(&dqp->q_freelist)); | |
80a376bf CH |
1308 | list_del_init(&dqp->q_freelist); |
1309 | xfs_Gqm->qm_dqfrlist_cnt--; | |
80a376bf | 1310 | mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); |
1da177e4 | 1311 | |
80a376bf | 1312 | xfs_qm_dqdestroy(dqp); |
80a376bf | 1313 | } |
1da177e4 | 1314 | |
1da177e4 LT |
1315 | /* |
1316 | * Give the buffer a little push if it is incore and | |
1317 | * wait on the flush lock. | |
1318 | */ | |
1319 | void | |
800b484e | 1320 | xfs_dqflock_pushbuf_wait( |
1da177e4 LT |
1321 | xfs_dquot_t *dqp) |
1322 | { | |
8a7b8a89 | 1323 | xfs_mount_t *mp = dqp->q_mount; |
1da177e4 LT |
1324 | xfs_buf_t *bp; |
1325 | ||
1326 | /* | |
1327 | * Check to see if the dquot has been flushed delayed | |
1328 | * write. If so, grab its buffer and send it | |
1329 | * out immediately. We'll be able to acquire | |
1330 | * the flush lock when the I/O completes. | |
1331 | */ | |
8a7b8a89 CH |
1332 | bp = xfs_incore(mp->m_ddev_targp, dqp->q_blkno, |
1333 | mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK); | |
7d6a7bde DC |
1334 | if (!bp) |
1335 | goto out_lock; | |
1336 | ||
1337 | if (XFS_BUF_ISDELAYWRITE(bp)) { | |
811e64c7 | 1338 | if (xfs_buf_ispinned(bp)) |
8a7b8a89 | 1339 | xfs_log_force(mp, 0); |
7d6a7bde DC |
1340 | xfs_buf_delwri_promote(bp); |
1341 | wake_up_process(bp->b_target->bt_task); | |
1da177e4 | 1342 | } |
7d6a7bde DC |
1343 | xfs_buf_relse(bp); |
1344 | out_lock: | |
1da177e4 LT |
1345 | xfs_dqflock(dqp); |
1346 | } |