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 | ||
251 | ||
252 | ||
253 | /* | |
254 | * Allocate a block and fill it with dquots. | |
255 | * This is called when the bmapi finds a hole. | |
256 | */ | |
257 | STATIC int | |
258 | xfs_qm_dqalloc( | |
efa092f3 | 259 | xfs_trans_t **tpp, |
1da177e4 LT |
260 | xfs_mount_t *mp, |
261 | xfs_dquot_t *dqp, | |
262 | xfs_inode_t *quotip, | |
263 | xfs_fileoff_t offset_fsb, | |
264 | xfs_buf_t **O_bpp) | |
265 | { | |
266 | xfs_fsblock_t firstblock; | |
267 | xfs_bmap_free_t flist; | |
268 | xfs_bmbt_irec_t map; | |
269 | int nmaps, error, committed; | |
270 | xfs_buf_t *bp; | |
efa092f3 | 271 | xfs_trans_t *tp = *tpp; |
1da177e4 LT |
272 | |
273 | ASSERT(tp != NULL); | |
0b1b213f CH |
274 | |
275 | trace_xfs_dqalloc(dqp); | |
1da177e4 LT |
276 | |
277 | /* | |
278 | * Initialize the bmap freelist prior to calling bmapi code. | |
279 | */ | |
9d87c319 | 280 | xfs_bmap_init(&flist, &firstblock); |
1da177e4 LT |
281 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
282 | /* | |
283 | * Return if this type of quotas is turned off while we didn't | |
284 | * have an inode lock | |
285 | */ | |
6967b964 | 286 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 LT |
287 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
288 | return (ESRCH); | |
289 | } | |
290 | ||
ddc3415a | 291 | xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); |
1da177e4 | 292 | nmaps = 1; |
c0dc7828 DC |
293 | error = xfs_bmapi_write(tp, quotip, offset_fsb, |
294 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
295 | &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), | |
296 | &map, &nmaps, &flist); | |
297 | if (error) | |
1da177e4 | 298 | goto error0; |
1da177e4 LT |
299 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
300 | ASSERT(nmaps == 1); | |
301 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
302 | (map.br_startblock != HOLESTARTBLOCK)); | |
303 | ||
304 | /* | |
305 | * Keep track of the blkno to save a lookup later | |
306 | */ | |
307 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
308 | ||
309 | /* now we can just get the buffer (there's nothing to read yet) */ | |
310 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
311 | dqp->q_blkno, | |
8a7b8a89 | 312 | mp->m_quotainfo->qi_dqchunklen, |
1da177e4 | 313 | 0); |
2a30f36d CS |
314 | |
315 | error = xfs_buf_geterror(bp); | |
316 | if (error) | |
1da177e4 | 317 | goto error1; |
2a30f36d | 318 | |
1da177e4 LT |
319 | /* |
320 | * Make a chunk of dquots out of this buffer and log | |
321 | * the entire thing. | |
322 | */ | |
1149d96a | 323 | xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 324 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
1da177e4 | 325 | |
efa092f3 TS |
326 | /* |
327 | * xfs_bmap_finish() may commit the current transaction and | |
328 | * start a second transaction if the freelist is not empty. | |
329 | * | |
330 | * Since we still want to modify this buffer, we need to | |
331 | * ensure that the buffer is not released on commit of | |
332 | * the first transaction and ensure the buffer is added to the | |
333 | * second transaction. | |
334 | * | |
335 | * If there is only one transaction then don't stop the buffer | |
336 | * from being released when it commits later on. | |
337 | */ | |
338 | ||
339 | xfs_trans_bhold(tp, bp); | |
340 | ||
f7c99b6f | 341 | if ((error = xfs_bmap_finish(tpp, &flist, &committed))) { |
1da177e4 LT |
342 | goto error1; |
343 | } | |
344 | ||
efa092f3 TS |
345 | if (committed) { |
346 | tp = *tpp; | |
347 | xfs_trans_bjoin(tp, bp); | |
348 | } else { | |
349 | xfs_trans_bhold_release(tp, bp); | |
350 | } | |
351 | ||
1da177e4 LT |
352 | *O_bpp = bp; |
353 | return 0; | |
354 | ||
355 | error1: | |
356 | xfs_bmap_cancel(&flist); | |
357 | error0: | |
358 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
359 | ||
360 | return (error); | |
361 | } | |
362 | ||
c6319198 DC |
363 | void |
364 | xfs_dquot_read_verify( | |
365 | struct xfs_buf *bp) | |
366 | { | |
367 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
368 | struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr; | |
369 | struct xfs_disk_dquot *ddq; | |
370 | xfs_dqid_t id = 0; | |
371 | int i; | |
372 | ||
373 | /* | |
374 | * On the first read of the buffer, verify that each dquot is valid. | |
375 | * We don't know what the id of the dquot is supposed to be, just that | |
376 | * they should be increasing monotonically within the buffer. If the | |
377 | * first id is corrupt, then it will fail on the second dquot in the | |
378 | * buffer so corruptions could point to the wrong dquot in this case. | |
379 | */ | |
380 | for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) { | |
381 | int error; | |
382 | ||
383 | ddq = &d[i].dd_diskdq; | |
384 | ||
385 | if (i == 0) | |
386 | id = be32_to_cpu(ddq->d_id); | |
387 | ||
388 | error = xfs_qm_dqcheck(mp, ddq, id + i, 0, XFS_QMOPT_DOWARN, | |
389 | "xfs_dquot_read_verify"); | |
390 | if (error) { | |
391 | XFS_CORRUPTION_ERROR("xfs_dquot_read_verify", | |
392 | XFS_ERRLEVEL_LOW, mp, d); | |
393 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
394 | break; | |
395 | } | |
396 | } | |
397 | bp->b_iodone = NULL; | |
398 | xfs_buf_ioend(bp, 0); | |
399 | } | |
400 | ||
401 | STATIC int | |
402 | xfs_qm_dqrepair( | |
403 | struct xfs_mount *mp, | |
404 | struct xfs_trans *tp, | |
405 | struct xfs_dquot *dqp, | |
406 | xfs_dqid_t firstid, | |
407 | struct xfs_buf **bpp) | |
408 | { | |
409 | int error; | |
410 | struct xfs_disk_dquot *ddq; | |
411 | struct xfs_dqblk *d; | |
412 | int i; | |
413 | ||
414 | /* | |
415 | * Read the buffer without verification so we get the corrupted | |
416 | * buffer returned to us. | |
417 | */ | |
418 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno, | |
419 | mp->m_quotainfo->qi_dqchunklen, | |
420 | 0, bpp, NULL); | |
421 | ||
422 | if (error) { | |
423 | ASSERT(*bpp == NULL); | |
424 | return XFS_ERROR(error); | |
425 | } | |
426 | ||
427 | ASSERT(xfs_buf_islocked(*bpp)); | |
428 | d = (struct xfs_dqblk *)(*bpp)->b_addr; | |
429 | ||
430 | /* Do the actual repair of dquots in this buffer */ | |
431 | for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) { | |
432 | ddq = &d[i].dd_diskdq; | |
433 | error = xfs_qm_dqcheck(mp, ddq, firstid + i, | |
434 | dqp->dq_flags & XFS_DQ_ALLTYPES, | |
435 | XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair"); | |
436 | if (error) { | |
437 | /* repair failed, we're screwed */ | |
438 | xfs_trans_brelse(tp, *bpp); | |
439 | return XFS_ERROR(EIO); | |
440 | } | |
441 | } | |
442 | ||
443 | return 0; | |
444 | } | |
445 | ||
1da177e4 LT |
446 | /* |
447 | * Maps a dquot to the buffer containing its on-disk version. | |
448 | * This returns a ptr to the buffer containing the on-disk dquot | |
449 | * in the bpp param, and a ptr to the on-disk dquot within that buffer | |
450 | */ | |
451 | STATIC int | |
452 | xfs_qm_dqtobp( | |
efa092f3 | 453 | xfs_trans_t **tpp, |
1da177e4 LT |
454 | xfs_dquot_t *dqp, |
455 | xfs_disk_dquot_t **O_ddpp, | |
456 | xfs_buf_t **O_bpp, | |
457 | uint flags) | |
458 | { | |
459 | xfs_bmbt_irec_t map; | |
acecf1b5 | 460 | int nmaps = 1, error; |
1da177e4 | 461 | xfs_buf_t *bp; |
acecf1b5 CH |
462 | xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp); |
463 | xfs_mount_t *mp = dqp->q_mount; | |
acecf1b5 | 464 | xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); |
efa092f3 | 465 | xfs_trans_t *tp = (tpp ? *tpp : NULL); |
1da177e4 | 466 | |
acecf1b5 | 467 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
1da177e4 | 468 | |
acecf1b5 | 469 | xfs_ilock(quotip, XFS_ILOCK_SHARED); |
6967b964 | 470 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 | 471 | /* |
acecf1b5 CH |
472 | * Return if this type of quotas is turned off while we |
473 | * didn't have the quota inode lock. | |
1da177e4 | 474 | */ |
acecf1b5 CH |
475 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); |
476 | return ESRCH; | |
477 | } | |
478 | ||
479 | /* | |
480 | * Find the block map; no allocations yet | |
481 | */ | |
5c8ed202 DC |
482 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
483 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); | |
acecf1b5 CH |
484 | |
485 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); | |
486 | if (error) | |
487 | return error; | |
488 | ||
489 | ASSERT(nmaps == 1); | |
490 | ASSERT(map.br_blockcount == 1); | |
491 | ||
492 | /* | |
493 | * Offset of dquot in the (fixed sized) dquot chunk. | |
494 | */ | |
495 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
496 | sizeof(xfs_dqblk_t); | |
497 | ||
498 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
499 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 500 | /* |
acecf1b5 | 501 | * We don't allocate unless we're asked to |
1da177e4 | 502 | */ |
acecf1b5 CH |
503 | if (!(flags & XFS_QMOPT_DQALLOC)) |
504 | return ENOENT; | |
1da177e4 | 505 | |
acecf1b5 CH |
506 | ASSERT(tp); |
507 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
508 | dqp->q_fileoffset, &bp); | |
1da177e4 | 509 | if (error) |
acecf1b5 CH |
510 | return error; |
511 | tp = *tpp; | |
512 | } else { | |
513 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
514 | |
515 | /* | |
acecf1b5 CH |
516 | * store the blkno etc so that we don't have to do the |
517 | * mapping all the time | |
1da177e4 | 518 | */ |
acecf1b5 | 519 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 520 | |
8a7b8a89 CH |
521 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
522 | dqp->q_blkno, | |
523 | mp->m_quotainfo->qi_dqchunklen, | |
c6319198 | 524 | 0, &bp, xfs_dquot_read_verify); |
acecf1b5 | 525 | |
c6319198 DC |
526 | if (error == EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) { |
527 | xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff * | |
528 | mp->m_quotainfo->qi_dqperchunk; | |
529 | ASSERT(bp == NULL); | |
530 | error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp); | |
531 | } | |
1da177e4 | 532 | |
c6319198 DC |
533 | if (error) { |
534 | ASSERT(bp == NULL); | |
535 | return XFS_ERROR(error); | |
1da177e4 | 536 | } |
1da177e4 LT |
537 | } |
538 | ||
c6319198 | 539 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 540 | *O_bpp = bp; |
c6319198 | 541 | *O_ddpp = bp->b_addr + dqp->q_bufoffset; |
1da177e4 LT |
542 | |
543 | return (0); | |
544 | } | |
545 | ||
546 | ||
547 | /* | |
548 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
549 | * and release the buffer immediately. | |
550 | * | |
97e7ade5 | 551 | * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed. |
1da177e4 | 552 | */ |
7ae44407 | 553 | int |
1da177e4 | 554 | xfs_qm_dqread( |
97e7ade5 CH |
555 | struct xfs_mount *mp, |
556 | xfs_dqid_t id, | |
557 | uint type, | |
558 | uint flags, | |
559 | struct xfs_dquot **O_dqpp) | |
1da177e4 | 560 | { |
97e7ade5 CH |
561 | struct xfs_dquot *dqp; |
562 | struct xfs_disk_dquot *ddqp; | |
563 | struct xfs_buf *bp; | |
564 | struct xfs_trans *tp = NULL; | |
565 | int error; | |
566 | int cancelflags = 0; | |
efa092f3 | 567 | |
92b2e5b3 | 568 | |
a05931ce | 569 | dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP); |
92b2e5b3 CH |
570 | |
571 | dqp->dq_flags = type; | |
572 | dqp->q_core.d_id = cpu_to_be32(id); | |
573 | dqp->q_mount = mp; | |
f8739c3c | 574 | INIT_LIST_HEAD(&dqp->q_lru); |
92b2e5b3 CH |
575 | mutex_init(&dqp->q_qlock); |
576 | init_waitqueue_head(&dqp->q_pinwait); | |
577 | ||
578 | /* | |
579 | * Because we want to use a counting completion, complete | |
580 | * the flush completion once to allow a single access to | |
581 | * the flush completion without blocking. | |
582 | */ | |
583 | init_completion(&dqp->q_flush); | |
584 | complete(&dqp->q_flush); | |
585 | ||
586 | /* | |
587 | * Make sure group quotas have a different lock class than user | |
588 | * quotas. | |
589 | */ | |
590 | if (!(type & XFS_DQ_USER)) | |
591 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class); | |
592 | ||
48776fd2 | 593 | XFS_STATS_INC(xs_qm_dquot); |
1da177e4 | 594 | |
0b1b213f CH |
595 | trace_xfs_dqread(dqp); |
596 | ||
97e7ade5 CH |
597 | if (flags & XFS_QMOPT_DQALLOC) { |
598 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC); | |
599 | error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp), | |
600 | XFS_WRITE_LOG_RES(mp) + | |
601 | /* | |
602 | * Round the chunklen up to the next multiple | |
603 | * of 128 (buf log item chunk size)). | |
604 | */ | |
605 | BBTOB(mp->m_quotainfo->qi_dqchunklen) - 1 + 128, | |
606 | 0, | |
607 | XFS_TRANS_PERM_LOG_RES, | |
608 | XFS_WRITE_LOG_COUNT); | |
609 | if (error) | |
610 | goto error1; | |
611 | cancelflags = XFS_TRANS_RELEASE_LOG_RES; | |
612 | } | |
613 | ||
1da177e4 LT |
614 | /* |
615 | * get a pointer to the on-disk dquot and the buffer containing it | |
616 | * dqp already knows its own type (GROUP/USER). | |
617 | */ | |
97e7ade5 CH |
618 | error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags); |
619 | if (error) { | |
620 | /* | |
621 | * This can happen if quotas got turned off (ESRCH), | |
622 | * or if the dquot didn't exist on disk and we ask to | |
623 | * allocate (ENOENT). | |
624 | */ | |
625 | trace_xfs_dqread_fail(dqp); | |
626 | cancelflags |= XFS_TRANS_ABORT; | |
627 | goto error1; | |
1da177e4 LT |
628 | } |
629 | ||
630 | /* copy everything from disk dquot to the incore dquot */ | |
631 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
1da177e4 LT |
632 | xfs_qm_dquot_logitem_init(dqp); |
633 | ||
634 | /* | |
635 | * Reservation counters are defined as reservation plus current usage | |
25985edc | 636 | * to avoid having to add every time. |
1da177e4 | 637 | */ |
1149d96a CH |
638 | dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); |
639 | dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); | |
640 | dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); | |
1da177e4 LT |
641 | |
642 | /* Mark the buf so that this will stay incore a little longer */ | |
38f23232 | 643 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 LT |
644 | |
645 | /* | |
646 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
647 | * So we need to release with xfs_trans_brelse(). | |
648 | * The strategy here is identical to that of inodes; we lock | |
649 | * the dquot in xfs_qm_dqget() before making it accessible to | |
650 | * others. This is because dquots, like inodes, need a good level of | |
651 | * concurrency, and we don't want to take locks on the entire buffers | |
652 | * for dquot accesses. | |
653 | * Note also that the dquot buffer may even be dirty at this point, if | |
654 | * this particular dquot was repaired. We still aren't afraid to | |
655 | * brelse it because we have the changes incore. | |
656 | */ | |
0c842ad4 | 657 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
658 | xfs_trans_brelse(tp, bp); |
659 | ||
1da177e4 | 660 | if (tp) { |
97e7ade5 CH |
661 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
662 | if (error) | |
663 | goto error0; | |
1da177e4 LT |
664 | } |
665 | ||
666 | *O_dqpp = dqp; | |
97e7ade5 | 667 | return error; |
1da177e4 | 668 | |
97e7ade5 | 669 | error1: |
1da177e4 LT |
670 | if (tp) |
671 | xfs_trans_cancel(tp, cancelflags); | |
97e7ade5 | 672 | error0: |
1da177e4 LT |
673 | xfs_qm_dqdestroy(dqp); |
674 | *O_dqpp = NULL; | |
97e7ade5 | 675 | return error; |
1da177e4 LT |
676 | } |
677 | ||
1da177e4 LT |
678 | /* |
679 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
680 | * a locked dquot, doing an allocation (if requested) as needed. | |
681 | * When both an inode and an id are given, the inode's id takes precedence. | |
682 | * That is, if the id changes while we don't hold the ilock inside this | |
683 | * function, the new dquot is returned, not necessarily the one requested | |
684 | * in the id argument. | |
685 | */ | |
686 | int | |
687 | xfs_qm_dqget( | |
688 | xfs_mount_t *mp, | |
689 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
690 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
691 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
692 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
693 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
694 | { | |
9f920f11 CH |
695 | struct xfs_quotainfo *qi = mp->m_quotainfo; |
696 | struct radix_tree_root *tree = XFS_DQUOT_TREE(qi, type); | |
697 | struct xfs_dquot *dqp; | |
698 | int error; | |
1da177e4 LT |
699 | |
700 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
701 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 702 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 LT |
703 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
704 | return (ESRCH); | |
705 | } | |
1da177e4 LT |
706 | |
707 | #ifdef DEBUG | |
708 | if (xfs_do_dqerror) { | |
709 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
710 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 711 | xfs_debug(mp, "Returning error in dqget"); |
1da177e4 LT |
712 | return (EIO); |
713 | } | |
714 | } | |
1da177e4 | 715 | |
c8ad20ff NS |
716 | ASSERT(type == XFS_DQ_USER || |
717 | type == XFS_DQ_PROJ || | |
718 | type == XFS_DQ_GROUP); | |
1da177e4 | 719 | if (ip) { |
579aa9ca | 720 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
36731410 | 721 | ASSERT(xfs_inode_dquot(ip, type) == NULL); |
1da177e4 LT |
722 | } |
723 | #endif | |
92678554 CH |
724 | |
725 | restart: | |
9f920f11 CH |
726 | mutex_lock(&qi->qi_tree_lock); |
727 | dqp = radix_tree_lookup(tree, id); | |
728 | if (dqp) { | |
729 | xfs_dqlock(dqp); | |
730 | if (dqp->dq_flags & XFS_DQ_FREEING) { | |
731 | xfs_dqunlock(dqp); | |
732 | mutex_unlock(&qi->qi_tree_lock); | |
733 | trace_xfs_dqget_freeing(dqp); | |
734 | delay(1); | |
735 | goto restart; | |
736 | } | |
1da177e4 | 737 | |
9f920f11 CH |
738 | dqp->q_nrefs++; |
739 | mutex_unlock(&qi->qi_tree_lock); | |
740 | ||
741 | trace_xfs_dqget_hit(dqp); | |
48776fd2 | 742 | XFS_STATS_INC(xs_qm_dqcachehits); |
9f920f11 CH |
743 | *O_dqpp = dqp; |
744 | return 0; | |
1da177e4 | 745 | } |
9f920f11 CH |
746 | mutex_unlock(&qi->qi_tree_lock); |
747 | XFS_STATS_INC(xs_qm_dqcachemisses); | |
1da177e4 LT |
748 | |
749 | /* | |
750 | * Dquot cache miss. We don't want to keep the inode lock across | |
751 | * a (potential) disk read. Also we don't want to deal with the lock | |
752 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
753 | * lock here means dealing with a chown that can happen before | |
754 | * we re-acquire the lock. | |
755 | */ | |
756 | if (ip) | |
757 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 | 758 | |
97e7ade5 | 759 | error = xfs_qm_dqread(mp, id, type, flags, &dqp); |
1da177e4 | 760 | |
7ae44407 CH |
761 | if (ip) |
762 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
763 | ||
764 | if (error) | |
765 | return error; | |
1da177e4 | 766 | |
1da177e4 | 767 | if (ip) { |
1da177e4 LT |
768 | /* |
769 | * A dquot could be attached to this inode by now, since | |
770 | * we had dropped the ilock. | |
771 | */ | |
36731410 | 772 | if (xfs_this_quota_on(mp, type)) { |
9f920f11 CH |
773 | struct xfs_dquot *dqp1; |
774 | ||
36731410 CS |
775 | dqp1 = xfs_inode_dquot(ip, type); |
776 | if (dqp1) { | |
1da177e4 | 777 | xfs_qm_dqdestroy(dqp); |
36731410 | 778 | dqp = dqp1; |
1da177e4 LT |
779 | xfs_dqlock(dqp); |
780 | goto dqret; | |
781 | } | |
782 | } else { | |
36731410 CS |
783 | /* inode stays locked on return */ |
784 | xfs_qm_dqdestroy(dqp); | |
785 | return XFS_ERROR(ESRCH); | |
1da177e4 LT |
786 | } |
787 | } | |
788 | ||
9f920f11 CH |
789 | mutex_lock(&qi->qi_tree_lock); |
790 | error = -radix_tree_insert(tree, id, dqp); | |
791 | if (unlikely(error)) { | |
792 | WARN_ON(error != EEXIST); | |
793 | ||
1da177e4 | 794 | /* |
9f920f11 CH |
795 | * Duplicate found. Just throw away the new dquot and start |
796 | * over. | |
1da177e4 | 797 | */ |
9f920f11 CH |
798 | mutex_unlock(&qi->qi_tree_lock); |
799 | trace_xfs_dqget_dup(dqp); | |
800 | xfs_qm_dqdestroy(dqp); | |
801 | XFS_STATS_INC(xs_qm_dquot_dups); | |
802 | goto restart; | |
1da177e4 LT |
803 | } |
804 | ||
1da177e4 LT |
805 | /* |
806 | * We return a locked dquot to the caller, with a reference taken | |
807 | */ | |
808 | xfs_dqlock(dqp); | |
809 | dqp->q_nrefs = 1; | |
810 | ||
9f920f11 | 811 | qi->qi_dquots++; |
9f920f11 CH |
812 | mutex_unlock(&qi->qi_tree_lock); |
813 | ||
1da177e4 | 814 | dqret: |
579aa9ca | 815 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 816 | trace_xfs_dqget_miss(dqp); |
1da177e4 LT |
817 | *O_dqpp = dqp; |
818 | return (0); | |
819 | } | |
820 | ||
821 | ||
f8739c3c CH |
822 | STATIC void |
823 | xfs_qm_dqput_final( | |
bf72de31 | 824 | struct xfs_dquot *dqp) |
1da177e4 | 825 | { |
f8739c3c | 826 | struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo; |
bf72de31 | 827 | struct xfs_dquot *gdqp; |
1da177e4 | 828 | |
bf72de31 CH |
829 | trace_xfs_dqput_free(dqp); |
830 | ||
f8739c3c CH |
831 | mutex_lock(&qi->qi_lru_lock); |
832 | if (list_empty(&dqp->q_lru)) { | |
833 | list_add_tail(&dqp->q_lru, &qi->qi_lru_list); | |
834 | qi->qi_lru_count++; | |
48776fd2 | 835 | XFS_STATS_INC(xs_qm_dquot_unused); |
bf72de31 | 836 | } |
f8739c3c | 837 | mutex_unlock(&qi->qi_lru_lock); |
bf72de31 | 838 | |
1da177e4 | 839 | /* |
bf72de31 CH |
840 | * If we just added a udquot to the freelist, then we want to release |
841 | * the gdquot reference that it (probably) has. Otherwise it'll keep | |
842 | * the gdquot from getting reclaimed. | |
1da177e4 | 843 | */ |
bf72de31 CH |
844 | gdqp = dqp->q_gdquot; |
845 | if (gdqp) { | |
846 | xfs_dqlock(gdqp); | |
847 | dqp->q_gdquot = NULL; | |
1da177e4 | 848 | } |
bf72de31 | 849 | xfs_dqunlock(dqp); |
1da177e4 | 850 | |
bf72de31 CH |
851 | /* |
852 | * If we had a group quota hint, release it now. | |
853 | */ | |
f8739c3c CH |
854 | if (gdqp) |
855 | xfs_qm_dqput(gdqp); | |
856 | } | |
857 | ||
858 | /* | |
859 | * Release a reference to the dquot (decrement ref-count) and unlock it. | |
860 | * | |
861 | * If there is a group quota attached to this dquot, carefully release that | |
862 | * too without tripping over deadlocks'n'stuff. | |
863 | */ | |
864 | void | |
865 | xfs_qm_dqput( | |
866 | struct xfs_dquot *dqp) | |
867 | { | |
868 | ASSERT(dqp->q_nrefs > 0); | |
869 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
870 | ||
871 | trace_xfs_dqput(dqp); | |
872 | ||
873 | if (--dqp->q_nrefs > 0) | |
874 | xfs_dqunlock(dqp); | |
875 | else | |
876 | xfs_qm_dqput_final(dqp); | |
1da177e4 LT |
877 | } |
878 | ||
879 | /* | |
880 | * Release a dquot. Flush it if dirty, then dqput() it. | |
881 | * dquot must not be locked. | |
882 | */ | |
883 | void | |
884 | xfs_qm_dqrele( | |
885 | xfs_dquot_t *dqp) | |
886 | { | |
7d095257 CH |
887 | if (!dqp) |
888 | return; | |
889 | ||
0b1b213f | 890 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
891 | |
892 | xfs_dqlock(dqp); | |
893 | /* | |
894 | * We don't care to flush it if the dquot is dirty here. | |
895 | * That will create stutters that we want to avoid. | |
896 | * Instead we do a delayed write when we try to reclaim | |
897 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
898 | */ | |
899 | xfs_qm_dqput(dqp); | |
900 | } | |
901 | ||
ca30b2a7 CH |
902 | /* |
903 | * This is the dquot flushing I/O completion routine. It is called | |
904 | * from interrupt level when the buffer containing the dquot is | |
905 | * flushed to disk. It is responsible for removing the dquot logitem | |
906 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
907 | * flush lock. This behavior is very similar to that of inodes.. | |
908 | */ | |
909 | STATIC void | |
910 | xfs_qm_dqflush_done( | |
911 | struct xfs_buf *bp, | |
912 | struct xfs_log_item *lip) | |
913 | { | |
914 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
915 | xfs_dquot_t *dqp = qip->qli_dquot; | |
916 | struct xfs_ail *ailp = lip->li_ailp; | |
917 | ||
918 | /* | |
919 | * We only want to pull the item from the AIL if its | |
920 | * location in the log has not changed since we started the flush. | |
921 | * Thus, we only bother if the dquot's lsn has | |
922 | * not changed. First we check the lsn outside the lock | |
923 | * since it's cheaper, and then we recheck while | |
924 | * holding the lock before removing the dquot from the AIL. | |
925 | */ | |
926 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
927 | lip->li_lsn == qip->qli_flush_lsn) { | |
928 | ||
929 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
930 | spin_lock(&ailp->xa_lock); | |
931 | if (lip->li_lsn == qip->qli_flush_lsn) | |
04913fdd | 932 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 CH |
933 | else |
934 | spin_unlock(&ailp->xa_lock); | |
935 | } | |
936 | ||
937 | /* | |
938 | * Release the dq's flush lock since we're done with it. | |
939 | */ | |
940 | xfs_dqfunlock(dqp); | |
941 | } | |
1da177e4 LT |
942 | |
943 | /* | |
944 | * Write a modified dquot to disk. | |
945 | * The dquot must be locked and the flush lock too taken by caller. | |
946 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
947 | * but the dquot is free to be unlocked and modified by the caller | |
948 | * in the interim. Dquot is still locked on return. This behavior is | |
949 | * identical to that of inodes. | |
950 | */ | |
951 | int | |
952 | xfs_qm_dqflush( | |
fe7257fd CH |
953 | struct xfs_dquot *dqp, |
954 | struct xfs_buf **bpp) | |
1da177e4 | 955 | { |
acecf1b5 CH |
956 | struct xfs_mount *mp = dqp->q_mount; |
957 | struct xfs_buf *bp; | |
958 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 959 | int error; |
1da177e4 LT |
960 | |
961 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 962 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 963 | |
0b1b213f | 964 | trace_xfs_dqflush(dqp); |
1da177e4 | 965 | |
fe7257fd CH |
966 | *bpp = NULL; |
967 | ||
1da177e4 LT |
968 | xfs_qm_dqunpin_wait(dqp); |
969 | ||
970 | /* | |
971 | * This may have been unpinned because the filesystem is shutting | |
972 | * down forcibly. If that's the case we must not write this dquot | |
dea96095 CH |
973 | * to disk, because the log record didn't make it to disk. |
974 | * | |
975 | * We also have to remove the log item from the AIL in this case, | |
976 | * as we wait for an emptry AIL as part of the unmount process. | |
1da177e4 | 977 | */ |
acecf1b5 | 978 | if (XFS_FORCED_SHUTDOWN(mp)) { |
dea96095 | 979 | struct xfs_log_item *lip = &dqp->q_logitem.qli_item; |
acecf1b5 | 980 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
dea96095 CH |
981 | |
982 | spin_lock(&mp->m_ail->xa_lock); | |
983 | if (lip->li_flags & XFS_LI_IN_AIL) | |
04913fdd DC |
984 | xfs_trans_ail_delete(mp->m_ail, lip, |
985 | SHUTDOWN_CORRUPT_INCORE); | |
dea96095 CH |
986 | else |
987 | spin_unlock(&mp->m_ail->xa_lock); | |
fe7257fd CH |
988 | error = XFS_ERROR(EIO); |
989 | goto out_unlock; | |
1da177e4 LT |
990 | } |
991 | ||
992 | /* | |
993 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 994 | */ |
acecf1b5 | 995 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
c3f8fc73 | 996 | mp->m_quotainfo->qi_dqchunklen, 0, &bp, NULL); |
fe7257fd CH |
997 | if (error) |
998 | goto out_unlock; | |
1da177e4 | 999 | |
acecf1b5 CH |
1000 | /* |
1001 | * Calculate the location of the dquot inside the buffer. | |
1002 | */ | |
62926044 | 1003 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1004 | |
1005 | /* | |
1006 | * A simple sanity check in case we got a corrupted dquot.. | |
1007 | */ | |
a0fa2b67 DC |
1008 | error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
1009 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); | |
1010 | if (error) { | |
acecf1b5 CH |
1011 | xfs_buf_relse(bp); |
1012 | xfs_dqfunlock(dqp); | |
1013 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
1da177e4 LT |
1014 | return XFS_ERROR(EIO); |
1015 | } | |
1016 | ||
1017 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1018 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1019 | |
1020 | /* | |
1021 | * Clear the dirty field and remember the flush lsn for later use. | |
1022 | */ | |
acecf1b5 | 1023 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1024 | |
7b2e2a31 DC |
1025 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1026 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 LT |
1027 | |
1028 | /* | |
1029 | * Attach an iodone routine so that we can remove this dquot from the | |
1030 | * AIL and release the flush lock once the dquot is synced to disk. | |
1031 | */ | |
ca30b2a7 CH |
1032 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1033 | &dqp->q_logitem.qli_item); | |
1034 | ||
1da177e4 LT |
1035 | /* |
1036 | * If the buffer is pinned then push on the log so we won't | |
1037 | * get stuck waiting in the write for too long. | |
1038 | */ | |
811e64c7 | 1039 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1040 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1041 | xfs_log_force(mp, 0); |
1da177e4 LT |
1042 | } |
1043 | ||
0b1b213f | 1044 | trace_xfs_dqflush_done(dqp); |
fe7257fd CH |
1045 | *bpp = bp; |
1046 | return 0; | |
0b1b213f | 1047 | |
fe7257fd CH |
1048 | out_unlock: |
1049 | xfs_dqfunlock(dqp); | |
1050 | return XFS_ERROR(EIO); | |
1da177e4 LT |
1051 | } |
1052 | ||
5bb87a33 CH |
1053 | /* |
1054 | * Lock two xfs_dquot structures. | |
1055 | * | |
1056 | * To avoid deadlocks we always lock the quota structure with | |
1057 | * the lowerd id first. | |
1058 | */ | |
1da177e4 LT |
1059 | void |
1060 | xfs_dqlock2( | |
1061 | xfs_dquot_t *d1, | |
1062 | xfs_dquot_t *d2) | |
1063 | { | |
1064 | if (d1 && d2) { | |
1065 | ASSERT(d1 != d2); | |
1149d96a CH |
1066 | if (be32_to_cpu(d1->q_core.d_id) > |
1067 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1068 | mutex_lock(&d2->q_qlock); |
1069 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1070 | } else { |
5bb87a33 CH |
1071 | mutex_lock(&d1->q_qlock); |
1072 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1073 | } |
5bb87a33 CH |
1074 | } else if (d1) { |
1075 | mutex_lock(&d1->q_qlock); | |
1076 | } else if (d2) { | |
1077 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1078 | } |
1079 | } | |
1080 | ||
a05931ce CH |
1081 | int __init |
1082 | xfs_qm_init(void) | |
1083 | { | |
1084 | xfs_qm_dqzone = | |
1085 | kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot"); | |
1086 | if (!xfs_qm_dqzone) | |
1087 | goto out; | |
1088 | ||
1089 | xfs_qm_dqtrxzone = | |
1090 | kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx"); | |
1091 | if (!xfs_qm_dqtrxzone) | |
1092 | goto out_free_dqzone; | |
1093 | ||
1094 | return 0; | |
1095 | ||
1096 | out_free_dqzone: | |
1097 | kmem_zone_destroy(xfs_qm_dqzone); | |
1098 | out: | |
1099 | return -ENOMEM; | |
1100 | } | |
1101 | ||
1c2ccc66 | 1102 | void |
a05931ce CH |
1103 | xfs_qm_exit(void) |
1104 | { | |
1105 | kmem_zone_destroy(xfs_qm_dqtrxzone); | |
1106 | kmem_zone_destroy(xfs_qm_dqzone); | |
1107 | } |