]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
4ce3121f NS |
2 | * Copyright (c) 2000-2003 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
4ce3121f NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
4ce3121f NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
4ce3121f NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 LT |
18 | #include "xfs.h" |
19 | #include "xfs_fs.h" | |
a844f451 | 20 | #include "xfs_bit.h" |
1da177e4 LT |
21 | #include "xfs_log.h" |
22 | #include "xfs_trans.h" | |
23 | #include "xfs_sb.h" | |
24 | #include "xfs_ag.h" | |
1da177e4 | 25 | #include "xfs_alloc.h" |
1da177e4 LT |
26 | #include "xfs_quota.h" |
27 | #include "xfs_mount.h" | |
1da177e4 | 28 | #include "xfs_bmap_btree.h" |
1da177e4 LT |
29 | #include "xfs_inode.h" |
30 | #include "xfs_bmap.h" | |
1da177e4 LT |
31 | #include "xfs_rtalloc.h" |
32 | #include "xfs_error.h" | |
33 | #include "xfs_itable.h" | |
1da177e4 LT |
34 | #include "xfs_attr.h" |
35 | #include "xfs_buf_item.h" | |
36 | #include "xfs_trans_space.h" | |
37 | #include "xfs_trans_priv.h" | |
1da177e4 | 38 | #include "xfs_qm.h" |
0b1b213f | 39 | #include "xfs_trace.h" |
1da177e4 | 40 | |
1da177e4 | 41 | /* |
bf72de31 CH |
42 | * Lock order: |
43 | * | |
44 | * ip->i_lock | |
9f920f11 | 45 | * qi->qi_tree_lock |
b84a3a96 CH |
46 | * dquot->q_qlock (xfs_dqlock() and friends) |
47 | * dquot->q_flush (xfs_dqflock() and friends) | |
48 | * qi->qi_lru_lock | |
bf72de31 CH |
49 | * |
50 | * If two dquots need to be locked the order is user before group/project, | |
51 | * otherwise by the lowest id first, see xfs_dqlock2. | |
52 | */ | |
1da177e4 | 53 | |
1da177e4 LT |
54 | #ifdef DEBUG |
55 | xfs_buftarg_t *xfs_dqerror_target; | |
56 | int xfs_do_dqerror; | |
57 | int xfs_dqreq_num; | |
58 | int xfs_dqerror_mod = 33; | |
59 | #endif | |
60 | ||
a05931ce CH |
61 | struct kmem_zone *xfs_qm_dqtrxzone; |
62 | static struct kmem_zone *xfs_qm_dqzone; | |
63 | ||
98b8c7a0 CH |
64 | static struct lock_class_key xfs_dquot_other_class; |
65 | ||
1da177e4 LT |
66 | /* |
67 | * This is called to free all the memory associated with a dquot | |
68 | */ | |
69 | void | |
70 | xfs_qm_dqdestroy( | |
71 | xfs_dquot_t *dqp) | |
72 | { | |
f8739c3c | 73 | ASSERT(list_empty(&dqp->q_lru)); |
1da177e4 LT |
74 | |
75 | mutex_destroy(&dqp->q_qlock); | |
a05931ce | 76 | kmem_zone_free(xfs_qm_dqzone, dqp); |
0b1b213f | 77 | |
48776fd2 | 78 | XFS_STATS_DEC(xs_qm_dquot); |
1da177e4 LT |
79 | } |
80 | ||
1da177e4 LT |
81 | /* |
82 | * If default limits are in force, push them into the dquot now. | |
83 | * We overwrite the dquot limits only if they are zero and this | |
84 | * is not the root dquot. | |
85 | */ | |
86 | void | |
87 | xfs_qm_adjust_dqlimits( | |
88 | xfs_mount_t *mp, | |
89 | xfs_disk_dquot_t *d) | |
90 | { | |
91 | xfs_quotainfo_t *q = mp->m_quotainfo; | |
92 | ||
93 | ASSERT(d->d_id); | |
94 | ||
95 | if (q->qi_bsoftlimit && !d->d_blk_softlimit) | |
1149d96a | 96 | d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit); |
1da177e4 | 97 | if (q->qi_bhardlimit && !d->d_blk_hardlimit) |
1149d96a | 98 | d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit); |
1da177e4 | 99 | if (q->qi_isoftlimit && !d->d_ino_softlimit) |
1149d96a | 100 | d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit); |
1da177e4 | 101 | if (q->qi_ihardlimit && !d->d_ino_hardlimit) |
1149d96a | 102 | d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit); |
1da177e4 | 103 | if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit) |
1149d96a | 104 | d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit); |
1da177e4 | 105 | if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit) |
1149d96a | 106 | d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit); |
1da177e4 LT |
107 | } |
108 | ||
109 | /* | |
110 | * Check the limits and timers of a dquot and start or reset timers | |
111 | * if necessary. | |
112 | * This gets called even when quota enforcement is OFF, which makes our | |
113 | * life a little less complicated. (We just don't reject any quota | |
114 | * reservations in that case, when enforcement is off). | |
115 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
116 | * enforcement's off. | |
117 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
118 | * 'automatically' get started when limits get exceeded. They do |
119 | * get reset to zero, however, when we find the count to be under | |
120 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
121 | */ |
122 | void | |
123 | xfs_qm_adjust_dqtimers( | |
124 | xfs_mount_t *mp, | |
125 | xfs_disk_dquot_t *d) | |
126 | { | |
127 | ASSERT(d->d_id); | |
128 | ||
ea15ab3c | 129 | #ifdef DEBUG |
1149d96a CH |
130 | if (d->d_blk_hardlimit) |
131 | ASSERT(be64_to_cpu(d->d_blk_softlimit) <= | |
132 | be64_to_cpu(d->d_blk_hardlimit)); | |
133 | if (d->d_ino_hardlimit) | |
134 | ASSERT(be64_to_cpu(d->d_ino_softlimit) <= | |
135 | be64_to_cpu(d->d_ino_hardlimit)); | |
136 | if (d->d_rtb_hardlimit) | |
137 | ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= | |
138 | be64_to_cpu(d->d_rtb_hardlimit)); | |
1da177e4 | 139 | #endif |
ea15ab3c | 140 | |
1da177e4 | 141 | if (!d->d_btimer) { |
1149d96a | 142 | if ((d->d_blk_softlimit && |
d0a3fe67 | 143 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
144 | be64_to_cpu(d->d_blk_softlimit))) || |
145 | (d->d_blk_hardlimit && | |
d0a3fe67 | 146 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
147 | be64_to_cpu(d->d_blk_hardlimit)))) { |
148 | d->d_btimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 149 | mp->m_quotainfo->qi_btimelimit); |
754002b4 NS |
150 | } else { |
151 | d->d_bwarns = 0; | |
1da177e4 LT |
152 | } |
153 | } else { | |
154 | if ((!d->d_blk_softlimit || | |
d0a3fe67 | 155 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 156 | be64_to_cpu(d->d_blk_softlimit))) && |
1da177e4 | 157 | (!d->d_blk_hardlimit || |
d0a3fe67 | 158 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 159 | be64_to_cpu(d->d_blk_hardlimit)))) { |
1da177e4 LT |
160 | d->d_btimer = 0; |
161 | } | |
162 | } | |
163 | ||
164 | if (!d->d_itimer) { | |
1149d96a | 165 | if ((d->d_ino_softlimit && |
d0a3fe67 | 166 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
167 | be64_to_cpu(d->d_ino_softlimit))) || |
168 | (d->d_ino_hardlimit && | |
d0a3fe67 | 169 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
170 | be64_to_cpu(d->d_ino_hardlimit)))) { |
171 | d->d_itimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 172 | mp->m_quotainfo->qi_itimelimit); |
754002b4 NS |
173 | } else { |
174 | d->d_iwarns = 0; | |
1da177e4 LT |
175 | } |
176 | } else { | |
177 | if ((!d->d_ino_softlimit || | |
d0a3fe67 | 178 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 179 | be64_to_cpu(d->d_ino_softlimit))) && |
1da177e4 | 180 | (!d->d_ino_hardlimit || |
d0a3fe67 | 181 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 182 | be64_to_cpu(d->d_ino_hardlimit)))) { |
1da177e4 LT |
183 | d->d_itimer = 0; |
184 | } | |
185 | } | |
186 | ||
187 | if (!d->d_rtbtimer) { | |
1149d96a | 188 | if ((d->d_rtb_softlimit && |
d0a3fe67 | 189 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
190 | be64_to_cpu(d->d_rtb_softlimit))) || |
191 | (d->d_rtb_hardlimit && | |
d0a3fe67 | 192 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
193 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
194 | d->d_rtbtimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 195 | mp->m_quotainfo->qi_rtbtimelimit); |
754002b4 NS |
196 | } else { |
197 | d->d_rtbwarns = 0; | |
1da177e4 LT |
198 | } |
199 | } else { | |
200 | if ((!d->d_rtb_softlimit || | |
d0a3fe67 | 201 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 202 | be64_to_cpu(d->d_rtb_softlimit))) && |
1da177e4 | 203 | (!d->d_rtb_hardlimit || |
d0a3fe67 | 204 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 205 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
1da177e4 LT |
206 | d->d_rtbtimer = 0; |
207 | } | |
208 | } | |
209 | } | |
210 | ||
1da177e4 LT |
211 | /* |
212 | * initialize a buffer full of dquots and log the whole thing | |
213 | */ | |
214 | STATIC void | |
215 | xfs_qm_init_dquot_blk( | |
216 | xfs_trans_t *tp, | |
217 | xfs_mount_t *mp, | |
218 | xfs_dqid_t id, | |
219 | uint type, | |
220 | xfs_buf_t *bp) | |
221 | { | |
8a7b8a89 | 222 | struct xfs_quotainfo *q = mp->m_quotainfo; |
1da177e4 LT |
223 | xfs_dqblk_t *d; |
224 | int curid, i; | |
225 | ||
226 | ASSERT(tp); | |
0c842ad4 | 227 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 228 | |
62926044 | 229 | d = bp->b_addr; |
1da177e4 LT |
230 | |
231 | /* | |
232 | * ID of the first dquot in the block - id's are zero based. | |
233 | */ | |
8a7b8a89 | 234 | curid = id - (id % q->qi_dqperchunk); |
1da177e4 | 235 | ASSERT(curid >= 0); |
8a7b8a89 | 236 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
49d35a5c CH |
237 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) { |
238 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
239 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
240 | d->dd_diskdq.d_id = cpu_to_be32(curid); | |
241 | d->dd_diskdq.d_flags = type; | |
242 | } | |
243 | ||
1da177e4 | 244 | xfs_trans_dquot_buf(tp, bp, |
c1155410 DC |
245 | (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : |
246 | ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : | |
247 | XFS_BLF_GDQUOT_BUF))); | |
8a7b8a89 | 248 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); |
1da177e4 LT |
249 | } |
250 | ||
b0f539de DC |
251 | static void |
252 | xfs_dquot_buf_verify( | |
253 | struct xfs_buf *bp) | |
254 | { | |
255 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
256 | struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr; | |
257 | struct xfs_disk_dquot *ddq; | |
258 | xfs_dqid_t id = 0; | |
259 | int i; | |
260 | ||
261 | /* | |
262 | * On the first read of the buffer, verify that each dquot is valid. | |
263 | * We don't know what the id of the dquot is supposed to be, just that | |
264 | * they should be increasing monotonically within the buffer. If the | |
265 | * first id is corrupt, then it will fail on the second dquot in the | |
266 | * buffer so corruptions could point to the wrong dquot in this case. | |
267 | */ | |
268 | for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) { | |
269 | int error; | |
270 | ||
271 | ddq = &d[i].dd_diskdq; | |
272 | ||
273 | if (i == 0) | |
274 | id = be32_to_cpu(ddq->d_id); | |
275 | ||
276 | error = xfs_qm_dqcheck(mp, ddq, id + i, 0, XFS_QMOPT_DOWARN, | |
277 | "xfs_dquot_read_verify"); | |
278 | if (error) { | |
279 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, d); | |
280 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
281 | break; | |
282 | } | |
283 | } | |
284 | } | |
285 | ||
286 | static void | |
287 | xfs_dquot_buf_write_verify( | |
288 | struct xfs_buf *bp) | |
289 | { | |
290 | xfs_dquot_buf_verify(bp); | |
291 | } | |
1da177e4 | 292 | |
b0f539de DC |
293 | void |
294 | xfs_dquot_buf_read_verify( | |
295 | struct xfs_buf *bp) | |
296 | { | |
297 | xfs_dquot_buf_verify(bp); | |
298 | bp->b_pre_io = xfs_dquot_buf_write_verify; | |
299 | bp->b_iodone = NULL; | |
300 | xfs_buf_ioend(bp, 0); | |
301 | } | |
1da177e4 LT |
302 | |
303 | /* | |
304 | * Allocate a block and fill it with dquots. | |
305 | * This is called when the bmapi finds a hole. | |
306 | */ | |
307 | STATIC int | |
308 | xfs_qm_dqalloc( | |
efa092f3 | 309 | xfs_trans_t **tpp, |
1da177e4 LT |
310 | xfs_mount_t *mp, |
311 | xfs_dquot_t *dqp, | |
312 | xfs_inode_t *quotip, | |
313 | xfs_fileoff_t offset_fsb, | |
314 | xfs_buf_t **O_bpp) | |
315 | { | |
316 | xfs_fsblock_t firstblock; | |
317 | xfs_bmap_free_t flist; | |
318 | xfs_bmbt_irec_t map; | |
319 | int nmaps, error, committed; | |
320 | xfs_buf_t *bp; | |
efa092f3 | 321 | xfs_trans_t *tp = *tpp; |
1da177e4 LT |
322 | |
323 | ASSERT(tp != NULL); | |
0b1b213f CH |
324 | |
325 | trace_xfs_dqalloc(dqp); | |
1da177e4 LT |
326 | |
327 | /* | |
328 | * Initialize the bmap freelist prior to calling bmapi code. | |
329 | */ | |
9d87c319 | 330 | xfs_bmap_init(&flist, &firstblock); |
1da177e4 LT |
331 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
332 | /* | |
333 | * Return if this type of quotas is turned off while we didn't | |
334 | * have an inode lock | |
335 | */ | |
6967b964 | 336 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 LT |
337 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
338 | return (ESRCH); | |
339 | } | |
340 | ||
ddc3415a | 341 | xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); |
1da177e4 | 342 | nmaps = 1; |
c0dc7828 DC |
343 | error = xfs_bmapi_write(tp, quotip, offset_fsb, |
344 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
345 | &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), | |
346 | &map, &nmaps, &flist); | |
347 | if (error) | |
1da177e4 | 348 | goto error0; |
1da177e4 LT |
349 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
350 | ASSERT(nmaps == 1); | |
351 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
352 | (map.br_startblock != HOLESTARTBLOCK)); | |
353 | ||
354 | /* | |
355 | * Keep track of the blkno to save a lookup later | |
356 | */ | |
357 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
358 | ||
359 | /* now we can just get the buffer (there's nothing to read yet) */ | |
360 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
361 | dqp->q_blkno, | |
8a7b8a89 | 362 | mp->m_quotainfo->qi_dqchunklen, |
1da177e4 | 363 | 0); |
2a30f36d CS |
364 | |
365 | error = xfs_buf_geterror(bp); | |
366 | if (error) | |
1da177e4 | 367 | goto error1; |
b0f539de | 368 | bp->b_pre_io = xfs_dquot_buf_write_verify; |
2a30f36d | 369 | |
1da177e4 LT |
370 | /* |
371 | * Make a chunk of dquots out of this buffer and log | |
372 | * the entire thing. | |
373 | */ | |
1149d96a | 374 | xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 375 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
1da177e4 | 376 | |
efa092f3 TS |
377 | /* |
378 | * xfs_bmap_finish() may commit the current transaction and | |
379 | * start a second transaction if the freelist is not empty. | |
380 | * | |
381 | * Since we still want to modify this buffer, we need to | |
382 | * ensure that the buffer is not released on commit of | |
383 | * the first transaction and ensure the buffer is added to the | |
384 | * second transaction. | |
385 | * | |
386 | * If there is only one transaction then don't stop the buffer | |
387 | * from being released when it commits later on. | |
388 | */ | |
389 | ||
390 | xfs_trans_bhold(tp, bp); | |
391 | ||
f7c99b6f | 392 | if ((error = xfs_bmap_finish(tpp, &flist, &committed))) { |
1da177e4 LT |
393 | goto error1; |
394 | } | |
395 | ||
efa092f3 TS |
396 | if (committed) { |
397 | tp = *tpp; | |
398 | xfs_trans_bjoin(tp, bp); | |
399 | } else { | |
400 | xfs_trans_bhold_release(tp, bp); | |
401 | } | |
402 | ||
1da177e4 LT |
403 | *O_bpp = bp; |
404 | return 0; | |
405 | ||
406 | error1: | |
407 | xfs_bmap_cancel(&flist); | |
408 | error0: | |
409 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
410 | ||
411 | return (error); | |
412 | } | |
c6319198 DC |
413 | STATIC int |
414 | xfs_qm_dqrepair( | |
415 | struct xfs_mount *mp, | |
416 | struct xfs_trans *tp, | |
417 | struct xfs_dquot *dqp, | |
418 | xfs_dqid_t firstid, | |
419 | struct xfs_buf **bpp) | |
420 | { | |
421 | int error; | |
422 | struct xfs_disk_dquot *ddq; | |
423 | struct xfs_dqblk *d; | |
424 | int i; | |
425 | ||
426 | /* | |
427 | * Read the buffer without verification so we get the corrupted | |
612cfbfe | 428 | * buffer returned to us. make sure we verify it on write, though. |
c6319198 DC |
429 | */ |
430 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno, | |
431 | mp->m_quotainfo->qi_dqchunklen, | |
432 | 0, bpp, NULL); | |
433 | ||
434 | if (error) { | |
435 | ASSERT(*bpp == NULL); | |
436 | return XFS_ERROR(error); | |
437 | } | |
612cfbfe | 438 | (*bpp)->b_pre_io = xfs_dquot_buf_write_verify; |
c6319198 DC |
439 | |
440 | ASSERT(xfs_buf_islocked(*bpp)); | |
441 | d = (struct xfs_dqblk *)(*bpp)->b_addr; | |
442 | ||
443 | /* Do the actual repair of dquots in this buffer */ | |
444 | for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) { | |
445 | ddq = &d[i].dd_diskdq; | |
446 | error = xfs_qm_dqcheck(mp, ddq, firstid + i, | |
447 | dqp->dq_flags & XFS_DQ_ALLTYPES, | |
448 | XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair"); | |
449 | if (error) { | |
450 | /* repair failed, we're screwed */ | |
451 | xfs_trans_brelse(tp, *bpp); | |
452 | return XFS_ERROR(EIO); | |
453 | } | |
454 | } | |
455 | ||
456 | return 0; | |
457 | } | |
458 | ||
1da177e4 LT |
459 | /* |
460 | * Maps a dquot to the buffer containing its on-disk version. | |
461 | * This returns a ptr to the buffer containing the on-disk dquot | |
462 | * in the bpp param, and a ptr to the on-disk dquot within that buffer | |
463 | */ | |
464 | STATIC int | |
465 | xfs_qm_dqtobp( | |
efa092f3 | 466 | xfs_trans_t **tpp, |
1da177e4 LT |
467 | xfs_dquot_t *dqp, |
468 | xfs_disk_dquot_t **O_ddpp, | |
469 | xfs_buf_t **O_bpp, | |
470 | uint flags) | |
471 | { | |
472 | xfs_bmbt_irec_t map; | |
acecf1b5 | 473 | int nmaps = 1, error; |
1da177e4 | 474 | xfs_buf_t *bp; |
acecf1b5 CH |
475 | xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp); |
476 | xfs_mount_t *mp = dqp->q_mount; | |
acecf1b5 | 477 | xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); |
efa092f3 | 478 | xfs_trans_t *tp = (tpp ? *tpp : NULL); |
1da177e4 | 479 | |
acecf1b5 | 480 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
1da177e4 | 481 | |
acecf1b5 | 482 | xfs_ilock(quotip, XFS_ILOCK_SHARED); |
6967b964 | 483 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 | 484 | /* |
acecf1b5 CH |
485 | * Return if this type of quotas is turned off while we |
486 | * didn't have the quota inode lock. | |
1da177e4 | 487 | */ |
acecf1b5 CH |
488 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); |
489 | return ESRCH; | |
490 | } | |
491 | ||
492 | /* | |
493 | * Find the block map; no allocations yet | |
494 | */ | |
5c8ed202 DC |
495 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
496 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); | |
acecf1b5 CH |
497 | |
498 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); | |
499 | if (error) | |
500 | return error; | |
501 | ||
502 | ASSERT(nmaps == 1); | |
503 | ASSERT(map.br_blockcount == 1); | |
504 | ||
505 | /* | |
506 | * Offset of dquot in the (fixed sized) dquot chunk. | |
507 | */ | |
508 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
509 | sizeof(xfs_dqblk_t); | |
510 | ||
511 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
512 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 513 | /* |
acecf1b5 | 514 | * We don't allocate unless we're asked to |
1da177e4 | 515 | */ |
acecf1b5 CH |
516 | if (!(flags & XFS_QMOPT_DQALLOC)) |
517 | return ENOENT; | |
1da177e4 | 518 | |
acecf1b5 CH |
519 | ASSERT(tp); |
520 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
521 | dqp->q_fileoffset, &bp); | |
1da177e4 | 522 | if (error) |
acecf1b5 CH |
523 | return error; |
524 | tp = *tpp; | |
525 | } else { | |
526 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
527 | |
528 | /* | |
acecf1b5 CH |
529 | * store the blkno etc so that we don't have to do the |
530 | * mapping all the time | |
1da177e4 | 531 | */ |
acecf1b5 | 532 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 533 | |
8a7b8a89 CH |
534 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
535 | dqp->q_blkno, | |
536 | mp->m_quotainfo->qi_dqchunklen, | |
612cfbfe | 537 | 0, &bp, xfs_dquot_buf_read_verify); |
acecf1b5 | 538 | |
c6319198 DC |
539 | if (error == EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) { |
540 | xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff * | |
541 | mp->m_quotainfo->qi_dqperchunk; | |
542 | ASSERT(bp == NULL); | |
543 | error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp); | |
544 | } | |
1da177e4 | 545 | |
c6319198 DC |
546 | if (error) { |
547 | ASSERT(bp == NULL); | |
548 | return XFS_ERROR(error); | |
1da177e4 | 549 | } |
1da177e4 LT |
550 | } |
551 | ||
c6319198 | 552 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 553 | *O_bpp = bp; |
c6319198 | 554 | *O_ddpp = bp->b_addr + dqp->q_bufoffset; |
1da177e4 LT |
555 | |
556 | return (0); | |
557 | } | |
558 | ||
559 | ||
560 | /* | |
561 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
562 | * and release the buffer immediately. | |
563 | * | |
97e7ade5 | 564 | * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed. |
1da177e4 | 565 | */ |
7ae44407 | 566 | int |
1da177e4 | 567 | xfs_qm_dqread( |
97e7ade5 CH |
568 | struct xfs_mount *mp, |
569 | xfs_dqid_t id, | |
570 | uint type, | |
571 | uint flags, | |
572 | struct xfs_dquot **O_dqpp) | |
1da177e4 | 573 | { |
97e7ade5 CH |
574 | struct xfs_dquot *dqp; |
575 | struct xfs_disk_dquot *ddqp; | |
576 | struct xfs_buf *bp; | |
577 | struct xfs_trans *tp = NULL; | |
578 | int error; | |
579 | int cancelflags = 0; | |
efa092f3 | 580 | |
92b2e5b3 | 581 | |
a05931ce | 582 | dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP); |
92b2e5b3 CH |
583 | |
584 | dqp->dq_flags = type; | |
585 | dqp->q_core.d_id = cpu_to_be32(id); | |
586 | dqp->q_mount = mp; | |
f8739c3c | 587 | INIT_LIST_HEAD(&dqp->q_lru); |
92b2e5b3 CH |
588 | mutex_init(&dqp->q_qlock); |
589 | init_waitqueue_head(&dqp->q_pinwait); | |
590 | ||
591 | /* | |
592 | * Because we want to use a counting completion, complete | |
593 | * the flush completion once to allow a single access to | |
594 | * the flush completion without blocking. | |
595 | */ | |
596 | init_completion(&dqp->q_flush); | |
597 | complete(&dqp->q_flush); | |
598 | ||
599 | /* | |
600 | * Make sure group quotas have a different lock class than user | |
601 | * quotas. | |
602 | */ | |
603 | if (!(type & XFS_DQ_USER)) | |
604 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class); | |
605 | ||
48776fd2 | 606 | XFS_STATS_INC(xs_qm_dquot); |
1da177e4 | 607 | |
0b1b213f CH |
608 | trace_xfs_dqread(dqp); |
609 | ||
97e7ade5 CH |
610 | if (flags & XFS_QMOPT_DQALLOC) { |
611 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC); | |
612 | error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp), | |
613 | XFS_WRITE_LOG_RES(mp) + | |
614 | /* | |
615 | * Round the chunklen up to the next multiple | |
616 | * of 128 (buf log item chunk size)). | |
617 | */ | |
618 | BBTOB(mp->m_quotainfo->qi_dqchunklen) - 1 + 128, | |
619 | 0, | |
620 | XFS_TRANS_PERM_LOG_RES, | |
621 | XFS_WRITE_LOG_COUNT); | |
622 | if (error) | |
623 | goto error1; | |
624 | cancelflags = XFS_TRANS_RELEASE_LOG_RES; | |
625 | } | |
626 | ||
1da177e4 LT |
627 | /* |
628 | * get a pointer to the on-disk dquot and the buffer containing it | |
629 | * dqp already knows its own type (GROUP/USER). | |
630 | */ | |
97e7ade5 CH |
631 | error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags); |
632 | if (error) { | |
633 | /* | |
634 | * This can happen if quotas got turned off (ESRCH), | |
635 | * or if the dquot didn't exist on disk and we ask to | |
636 | * allocate (ENOENT). | |
637 | */ | |
638 | trace_xfs_dqread_fail(dqp); | |
639 | cancelflags |= XFS_TRANS_ABORT; | |
640 | goto error1; | |
1da177e4 LT |
641 | } |
642 | ||
643 | /* copy everything from disk dquot to the incore dquot */ | |
644 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
1da177e4 LT |
645 | xfs_qm_dquot_logitem_init(dqp); |
646 | ||
647 | /* | |
648 | * Reservation counters are defined as reservation plus current usage | |
25985edc | 649 | * to avoid having to add every time. |
1da177e4 | 650 | */ |
1149d96a CH |
651 | dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); |
652 | dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); | |
653 | dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); | |
1da177e4 LT |
654 | |
655 | /* Mark the buf so that this will stay incore a little longer */ | |
38f23232 | 656 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 LT |
657 | |
658 | /* | |
659 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
660 | * So we need to release with xfs_trans_brelse(). | |
661 | * The strategy here is identical to that of inodes; we lock | |
662 | * the dquot in xfs_qm_dqget() before making it accessible to | |
663 | * others. This is because dquots, like inodes, need a good level of | |
664 | * concurrency, and we don't want to take locks on the entire buffers | |
665 | * for dquot accesses. | |
666 | * Note also that the dquot buffer may even be dirty at this point, if | |
667 | * this particular dquot was repaired. We still aren't afraid to | |
668 | * brelse it because we have the changes incore. | |
669 | */ | |
0c842ad4 | 670 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
671 | xfs_trans_brelse(tp, bp); |
672 | ||
1da177e4 | 673 | if (tp) { |
97e7ade5 CH |
674 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
675 | if (error) | |
676 | goto error0; | |
1da177e4 LT |
677 | } |
678 | ||
679 | *O_dqpp = dqp; | |
97e7ade5 | 680 | return error; |
1da177e4 | 681 | |
97e7ade5 | 682 | error1: |
1da177e4 LT |
683 | if (tp) |
684 | xfs_trans_cancel(tp, cancelflags); | |
97e7ade5 | 685 | error0: |
1da177e4 LT |
686 | xfs_qm_dqdestroy(dqp); |
687 | *O_dqpp = NULL; | |
97e7ade5 | 688 | return error; |
1da177e4 LT |
689 | } |
690 | ||
1da177e4 LT |
691 | /* |
692 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
693 | * a locked dquot, doing an allocation (if requested) as needed. | |
694 | * When both an inode and an id are given, the inode's id takes precedence. | |
695 | * That is, if the id changes while we don't hold the ilock inside this | |
696 | * function, the new dquot is returned, not necessarily the one requested | |
697 | * in the id argument. | |
698 | */ | |
699 | int | |
700 | xfs_qm_dqget( | |
701 | xfs_mount_t *mp, | |
702 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
703 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
704 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
705 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
706 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
707 | { | |
9f920f11 CH |
708 | struct xfs_quotainfo *qi = mp->m_quotainfo; |
709 | struct radix_tree_root *tree = XFS_DQUOT_TREE(qi, type); | |
710 | struct xfs_dquot *dqp; | |
711 | int error; | |
1da177e4 LT |
712 | |
713 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
714 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 715 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 LT |
716 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
717 | return (ESRCH); | |
718 | } | |
1da177e4 LT |
719 | |
720 | #ifdef DEBUG | |
721 | if (xfs_do_dqerror) { | |
722 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
723 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 724 | xfs_debug(mp, "Returning error in dqget"); |
1da177e4 LT |
725 | return (EIO); |
726 | } | |
727 | } | |
1da177e4 | 728 | |
c8ad20ff NS |
729 | ASSERT(type == XFS_DQ_USER || |
730 | type == XFS_DQ_PROJ || | |
731 | type == XFS_DQ_GROUP); | |
1da177e4 | 732 | if (ip) { |
579aa9ca | 733 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
36731410 | 734 | ASSERT(xfs_inode_dquot(ip, type) == NULL); |
1da177e4 LT |
735 | } |
736 | #endif | |
92678554 CH |
737 | |
738 | restart: | |
9f920f11 CH |
739 | mutex_lock(&qi->qi_tree_lock); |
740 | dqp = radix_tree_lookup(tree, id); | |
741 | if (dqp) { | |
742 | xfs_dqlock(dqp); | |
743 | if (dqp->dq_flags & XFS_DQ_FREEING) { | |
744 | xfs_dqunlock(dqp); | |
745 | mutex_unlock(&qi->qi_tree_lock); | |
746 | trace_xfs_dqget_freeing(dqp); | |
747 | delay(1); | |
748 | goto restart; | |
749 | } | |
1da177e4 | 750 | |
9f920f11 CH |
751 | dqp->q_nrefs++; |
752 | mutex_unlock(&qi->qi_tree_lock); | |
753 | ||
754 | trace_xfs_dqget_hit(dqp); | |
48776fd2 | 755 | XFS_STATS_INC(xs_qm_dqcachehits); |
9f920f11 CH |
756 | *O_dqpp = dqp; |
757 | return 0; | |
1da177e4 | 758 | } |
9f920f11 CH |
759 | mutex_unlock(&qi->qi_tree_lock); |
760 | XFS_STATS_INC(xs_qm_dqcachemisses); | |
1da177e4 LT |
761 | |
762 | /* | |
763 | * Dquot cache miss. We don't want to keep the inode lock across | |
764 | * a (potential) disk read. Also we don't want to deal with the lock | |
765 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
766 | * lock here means dealing with a chown that can happen before | |
767 | * we re-acquire the lock. | |
768 | */ | |
769 | if (ip) | |
770 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 | 771 | |
97e7ade5 | 772 | error = xfs_qm_dqread(mp, id, type, flags, &dqp); |
1da177e4 | 773 | |
7ae44407 CH |
774 | if (ip) |
775 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
776 | ||
777 | if (error) | |
778 | return error; | |
1da177e4 | 779 | |
1da177e4 | 780 | if (ip) { |
1da177e4 LT |
781 | /* |
782 | * A dquot could be attached to this inode by now, since | |
783 | * we had dropped the ilock. | |
784 | */ | |
36731410 | 785 | if (xfs_this_quota_on(mp, type)) { |
9f920f11 CH |
786 | struct xfs_dquot *dqp1; |
787 | ||
36731410 CS |
788 | dqp1 = xfs_inode_dquot(ip, type); |
789 | if (dqp1) { | |
1da177e4 | 790 | xfs_qm_dqdestroy(dqp); |
36731410 | 791 | dqp = dqp1; |
1da177e4 LT |
792 | xfs_dqlock(dqp); |
793 | goto dqret; | |
794 | } | |
795 | } else { | |
36731410 CS |
796 | /* inode stays locked on return */ |
797 | xfs_qm_dqdestroy(dqp); | |
798 | return XFS_ERROR(ESRCH); | |
1da177e4 LT |
799 | } |
800 | } | |
801 | ||
9f920f11 CH |
802 | mutex_lock(&qi->qi_tree_lock); |
803 | error = -radix_tree_insert(tree, id, dqp); | |
804 | if (unlikely(error)) { | |
805 | WARN_ON(error != EEXIST); | |
806 | ||
1da177e4 | 807 | /* |
9f920f11 CH |
808 | * Duplicate found. Just throw away the new dquot and start |
809 | * over. | |
1da177e4 | 810 | */ |
9f920f11 CH |
811 | mutex_unlock(&qi->qi_tree_lock); |
812 | trace_xfs_dqget_dup(dqp); | |
813 | xfs_qm_dqdestroy(dqp); | |
814 | XFS_STATS_INC(xs_qm_dquot_dups); | |
815 | goto restart; | |
1da177e4 LT |
816 | } |
817 | ||
1da177e4 LT |
818 | /* |
819 | * We return a locked dquot to the caller, with a reference taken | |
820 | */ | |
821 | xfs_dqlock(dqp); | |
822 | dqp->q_nrefs = 1; | |
823 | ||
9f920f11 | 824 | qi->qi_dquots++; |
9f920f11 CH |
825 | mutex_unlock(&qi->qi_tree_lock); |
826 | ||
1da177e4 | 827 | dqret: |
579aa9ca | 828 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 829 | trace_xfs_dqget_miss(dqp); |
1da177e4 LT |
830 | *O_dqpp = dqp; |
831 | return (0); | |
832 | } | |
833 | ||
834 | ||
f8739c3c CH |
835 | STATIC void |
836 | xfs_qm_dqput_final( | |
bf72de31 | 837 | struct xfs_dquot *dqp) |
1da177e4 | 838 | { |
f8739c3c | 839 | struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo; |
bf72de31 | 840 | struct xfs_dquot *gdqp; |
1da177e4 | 841 | |
bf72de31 CH |
842 | trace_xfs_dqput_free(dqp); |
843 | ||
f8739c3c CH |
844 | mutex_lock(&qi->qi_lru_lock); |
845 | if (list_empty(&dqp->q_lru)) { | |
846 | list_add_tail(&dqp->q_lru, &qi->qi_lru_list); | |
847 | qi->qi_lru_count++; | |
48776fd2 | 848 | XFS_STATS_INC(xs_qm_dquot_unused); |
bf72de31 | 849 | } |
f8739c3c | 850 | mutex_unlock(&qi->qi_lru_lock); |
bf72de31 | 851 | |
1da177e4 | 852 | /* |
bf72de31 CH |
853 | * If we just added a udquot to the freelist, then we want to release |
854 | * the gdquot reference that it (probably) has. Otherwise it'll keep | |
855 | * the gdquot from getting reclaimed. | |
1da177e4 | 856 | */ |
bf72de31 CH |
857 | gdqp = dqp->q_gdquot; |
858 | if (gdqp) { | |
859 | xfs_dqlock(gdqp); | |
860 | dqp->q_gdquot = NULL; | |
1da177e4 | 861 | } |
bf72de31 | 862 | xfs_dqunlock(dqp); |
1da177e4 | 863 | |
bf72de31 CH |
864 | /* |
865 | * If we had a group quota hint, release it now. | |
866 | */ | |
f8739c3c CH |
867 | if (gdqp) |
868 | xfs_qm_dqput(gdqp); | |
869 | } | |
870 | ||
871 | /* | |
872 | * Release a reference to the dquot (decrement ref-count) and unlock it. | |
873 | * | |
874 | * If there is a group quota attached to this dquot, carefully release that | |
875 | * too without tripping over deadlocks'n'stuff. | |
876 | */ | |
877 | void | |
878 | xfs_qm_dqput( | |
879 | struct xfs_dquot *dqp) | |
880 | { | |
881 | ASSERT(dqp->q_nrefs > 0); | |
882 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
883 | ||
884 | trace_xfs_dqput(dqp); | |
885 | ||
886 | if (--dqp->q_nrefs > 0) | |
887 | xfs_dqunlock(dqp); | |
888 | else | |
889 | xfs_qm_dqput_final(dqp); | |
1da177e4 LT |
890 | } |
891 | ||
892 | /* | |
893 | * Release a dquot. Flush it if dirty, then dqput() it. | |
894 | * dquot must not be locked. | |
895 | */ | |
896 | void | |
897 | xfs_qm_dqrele( | |
898 | xfs_dquot_t *dqp) | |
899 | { | |
7d095257 CH |
900 | if (!dqp) |
901 | return; | |
902 | ||
0b1b213f | 903 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
904 | |
905 | xfs_dqlock(dqp); | |
906 | /* | |
907 | * We don't care to flush it if the dquot is dirty here. | |
908 | * That will create stutters that we want to avoid. | |
909 | * Instead we do a delayed write when we try to reclaim | |
910 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
911 | */ | |
912 | xfs_qm_dqput(dqp); | |
913 | } | |
914 | ||
ca30b2a7 CH |
915 | /* |
916 | * This is the dquot flushing I/O completion routine. It is called | |
917 | * from interrupt level when the buffer containing the dquot is | |
918 | * flushed to disk. It is responsible for removing the dquot logitem | |
919 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
920 | * flush lock. This behavior is very similar to that of inodes.. | |
921 | */ | |
922 | STATIC void | |
923 | xfs_qm_dqflush_done( | |
924 | struct xfs_buf *bp, | |
925 | struct xfs_log_item *lip) | |
926 | { | |
927 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
928 | xfs_dquot_t *dqp = qip->qli_dquot; | |
929 | struct xfs_ail *ailp = lip->li_ailp; | |
930 | ||
931 | /* | |
932 | * We only want to pull the item from the AIL if its | |
933 | * location in the log has not changed since we started the flush. | |
934 | * Thus, we only bother if the dquot's lsn has | |
935 | * not changed. First we check the lsn outside the lock | |
936 | * since it's cheaper, and then we recheck while | |
937 | * holding the lock before removing the dquot from the AIL. | |
938 | */ | |
939 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
940 | lip->li_lsn == qip->qli_flush_lsn) { | |
941 | ||
942 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
943 | spin_lock(&ailp->xa_lock); | |
944 | if (lip->li_lsn == qip->qli_flush_lsn) | |
04913fdd | 945 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 CH |
946 | else |
947 | spin_unlock(&ailp->xa_lock); | |
948 | } | |
949 | ||
950 | /* | |
951 | * Release the dq's flush lock since we're done with it. | |
952 | */ | |
953 | xfs_dqfunlock(dqp); | |
954 | } | |
1da177e4 LT |
955 | |
956 | /* | |
957 | * Write a modified dquot to disk. | |
958 | * The dquot must be locked and the flush lock too taken by caller. | |
959 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
960 | * but the dquot is free to be unlocked and modified by the caller | |
961 | * in the interim. Dquot is still locked on return. This behavior is | |
962 | * identical to that of inodes. | |
963 | */ | |
964 | int | |
965 | xfs_qm_dqflush( | |
fe7257fd CH |
966 | struct xfs_dquot *dqp, |
967 | struct xfs_buf **bpp) | |
1da177e4 | 968 | { |
acecf1b5 CH |
969 | struct xfs_mount *mp = dqp->q_mount; |
970 | struct xfs_buf *bp; | |
971 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 972 | int error; |
1da177e4 LT |
973 | |
974 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 975 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 976 | |
0b1b213f | 977 | trace_xfs_dqflush(dqp); |
1da177e4 | 978 | |
fe7257fd CH |
979 | *bpp = NULL; |
980 | ||
1da177e4 LT |
981 | xfs_qm_dqunpin_wait(dqp); |
982 | ||
983 | /* | |
984 | * This may have been unpinned because the filesystem is shutting | |
985 | * down forcibly. If that's the case we must not write this dquot | |
dea96095 CH |
986 | * to disk, because the log record didn't make it to disk. |
987 | * | |
988 | * We also have to remove the log item from the AIL in this case, | |
989 | * as we wait for an emptry AIL as part of the unmount process. | |
1da177e4 | 990 | */ |
acecf1b5 | 991 | if (XFS_FORCED_SHUTDOWN(mp)) { |
dea96095 | 992 | struct xfs_log_item *lip = &dqp->q_logitem.qli_item; |
acecf1b5 | 993 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
dea96095 CH |
994 | |
995 | spin_lock(&mp->m_ail->xa_lock); | |
996 | if (lip->li_flags & XFS_LI_IN_AIL) | |
04913fdd DC |
997 | xfs_trans_ail_delete(mp->m_ail, lip, |
998 | SHUTDOWN_CORRUPT_INCORE); | |
dea96095 CH |
999 | else |
1000 | spin_unlock(&mp->m_ail->xa_lock); | |
fe7257fd CH |
1001 | error = XFS_ERROR(EIO); |
1002 | goto out_unlock; | |
1da177e4 LT |
1003 | } |
1004 | ||
1005 | /* | |
1006 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1007 | */ |
acecf1b5 | 1008 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
c3f8fc73 | 1009 | mp->m_quotainfo->qi_dqchunklen, 0, &bp, NULL); |
fe7257fd CH |
1010 | if (error) |
1011 | goto out_unlock; | |
1da177e4 | 1012 | |
acecf1b5 CH |
1013 | /* |
1014 | * Calculate the location of the dquot inside the buffer. | |
1015 | */ | |
62926044 | 1016 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1017 | |
1018 | /* | |
1019 | * A simple sanity check in case we got a corrupted dquot.. | |
1020 | */ | |
a0fa2b67 DC |
1021 | error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
1022 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); | |
1023 | if (error) { | |
acecf1b5 CH |
1024 | xfs_buf_relse(bp); |
1025 | xfs_dqfunlock(dqp); | |
1026 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
1da177e4 LT |
1027 | return XFS_ERROR(EIO); |
1028 | } | |
1029 | ||
1030 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1031 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1032 | |
1033 | /* | |
1034 | * Clear the dirty field and remember the flush lsn for later use. | |
1035 | */ | |
acecf1b5 | 1036 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1037 | |
7b2e2a31 DC |
1038 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1039 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 LT |
1040 | |
1041 | /* | |
1042 | * Attach an iodone routine so that we can remove this dquot from the | |
1043 | * AIL and release the flush lock once the dquot is synced to disk. | |
1044 | */ | |
ca30b2a7 CH |
1045 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1046 | &dqp->q_logitem.qli_item); | |
1047 | ||
1da177e4 LT |
1048 | /* |
1049 | * If the buffer is pinned then push on the log so we won't | |
1050 | * get stuck waiting in the write for too long. | |
1051 | */ | |
811e64c7 | 1052 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1053 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1054 | xfs_log_force(mp, 0); |
1da177e4 LT |
1055 | } |
1056 | ||
0b1b213f | 1057 | trace_xfs_dqflush_done(dqp); |
fe7257fd CH |
1058 | *bpp = bp; |
1059 | return 0; | |
0b1b213f | 1060 | |
fe7257fd CH |
1061 | out_unlock: |
1062 | xfs_dqfunlock(dqp); | |
1063 | return XFS_ERROR(EIO); | |
1da177e4 LT |
1064 | } |
1065 | ||
5bb87a33 CH |
1066 | /* |
1067 | * Lock two xfs_dquot structures. | |
1068 | * | |
1069 | * To avoid deadlocks we always lock the quota structure with | |
1070 | * the lowerd id first. | |
1071 | */ | |
1da177e4 LT |
1072 | void |
1073 | xfs_dqlock2( | |
1074 | xfs_dquot_t *d1, | |
1075 | xfs_dquot_t *d2) | |
1076 | { | |
1077 | if (d1 && d2) { | |
1078 | ASSERT(d1 != d2); | |
1149d96a CH |
1079 | if (be32_to_cpu(d1->q_core.d_id) > |
1080 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1081 | mutex_lock(&d2->q_qlock); |
1082 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1083 | } else { |
5bb87a33 CH |
1084 | mutex_lock(&d1->q_qlock); |
1085 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1086 | } |
5bb87a33 CH |
1087 | } else if (d1) { |
1088 | mutex_lock(&d1->q_qlock); | |
1089 | } else if (d2) { | |
1090 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1091 | } |
1092 | } | |
1093 | ||
a05931ce CH |
1094 | int __init |
1095 | xfs_qm_init(void) | |
1096 | { | |
1097 | xfs_qm_dqzone = | |
1098 | kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot"); | |
1099 | if (!xfs_qm_dqzone) | |
1100 | goto out; | |
1101 | ||
1102 | xfs_qm_dqtrxzone = | |
1103 | kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx"); | |
1104 | if (!xfs_qm_dqtrxzone) | |
1105 | goto out_free_dqzone; | |
1106 | ||
1107 | return 0; | |
1108 | ||
1109 | out_free_dqzone: | |
1110 | kmem_zone_destroy(xfs_qm_dqzone); | |
1111 | out: | |
1112 | return -ENOMEM; | |
1113 | } | |
1114 | ||
1c2ccc66 | 1115 | void |
a05931ce CH |
1116 | xfs_qm_exit(void) |
1117 | { | |
1118 | kmem_zone_destroy(xfs_qm_dqtrxzone); | |
1119 | kmem_zone_destroy(xfs_qm_dqzone); | |
1120 | } |