]>
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" | |
6ca1c906 | 20 | #include "xfs_format.h" |
239880ef | 21 | #include "xfs_log_format.h" |
70a9883c | 22 | #include "xfs_shared.h" |
239880ef | 23 | #include "xfs_trans_resv.h" |
a844f451 | 24 | #include "xfs_bit.h" |
1da177e4 LT |
25 | #include "xfs_sb.h" |
26 | #include "xfs_ag.h" | |
1da177e4 | 27 | #include "xfs_mount.h" |
1da177e4 | 28 | #include "xfs_bmap_btree.h" |
1da177e4 LT |
29 | #include "xfs_inode.h" |
30 | #include "xfs_bmap.h" | |
68988114 | 31 | #include "xfs_bmap_util.h" |
239880ef DC |
32 | #include "xfs_alloc.h" |
33 | #include "xfs_quota.h" | |
1da177e4 LT |
34 | #include "xfs_rtalloc.h" |
35 | #include "xfs_error.h" | |
36 | #include "xfs_itable.h" | |
1da177e4 | 37 | #include "xfs_attr.h" |
239880ef | 38 | #include "xfs_trans.h" |
1da177e4 LT |
39 | #include "xfs_buf_item.h" |
40 | #include "xfs_trans_space.h" | |
41 | #include "xfs_trans_priv.h" | |
1da177e4 | 42 | #include "xfs_qm.h" |
3fe58f30 | 43 | #include "xfs_cksum.h" |
0b1b213f | 44 | #include "xfs_trace.h" |
239880ef | 45 | #include "xfs_log.h" |
1da177e4 | 46 | |
1da177e4 | 47 | /* |
bf72de31 CH |
48 | * Lock order: |
49 | * | |
50 | * ip->i_lock | |
9f920f11 | 51 | * qi->qi_tree_lock |
b84a3a96 CH |
52 | * dquot->q_qlock (xfs_dqlock() and friends) |
53 | * dquot->q_flush (xfs_dqflock() and friends) | |
54 | * qi->qi_lru_lock | |
bf72de31 CH |
55 | * |
56 | * If two dquots need to be locked the order is user before group/project, | |
57 | * otherwise by the lowest id first, see xfs_dqlock2. | |
58 | */ | |
1da177e4 | 59 | |
1da177e4 LT |
60 | #ifdef DEBUG |
61 | xfs_buftarg_t *xfs_dqerror_target; | |
62 | int xfs_do_dqerror; | |
63 | int xfs_dqreq_num; | |
64 | int xfs_dqerror_mod = 33; | |
65 | #endif | |
66 | ||
a05931ce CH |
67 | struct kmem_zone *xfs_qm_dqtrxzone; |
68 | static struct kmem_zone *xfs_qm_dqzone; | |
69 | ||
f112a049 DC |
70 | static struct lock_class_key xfs_dquot_group_class; |
71 | static struct lock_class_key xfs_dquot_project_class; | |
98b8c7a0 | 72 | |
1da177e4 LT |
73 | /* |
74 | * This is called to free all the memory associated with a dquot | |
75 | */ | |
76 | void | |
77 | xfs_qm_dqdestroy( | |
78 | xfs_dquot_t *dqp) | |
79 | { | |
f8739c3c | 80 | ASSERT(list_empty(&dqp->q_lru)); |
1da177e4 LT |
81 | |
82 | mutex_destroy(&dqp->q_qlock); | |
a05931ce | 83 | kmem_zone_free(xfs_qm_dqzone, dqp); |
0b1b213f | 84 | |
48776fd2 | 85 | XFS_STATS_DEC(xs_qm_dquot); |
1da177e4 LT |
86 | } |
87 | ||
1da177e4 LT |
88 | /* |
89 | * If default limits are in force, push them into the dquot now. | |
90 | * We overwrite the dquot limits only if they are zero and this | |
91 | * is not the root dquot. | |
92 | */ | |
93 | void | |
94 | xfs_qm_adjust_dqlimits( | |
4b6eae2e BF |
95 | struct xfs_mount *mp, |
96 | struct xfs_dquot *dq) | |
1da177e4 | 97 | { |
4b6eae2e BF |
98 | struct xfs_quotainfo *q = mp->m_quotainfo; |
99 | struct xfs_disk_dquot *d = &dq->q_core; | |
b1366451 | 100 | int prealloc = 0; |
1da177e4 LT |
101 | |
102 | ASSERT(d->d_id); | |
103 | ||
b1366451 | 104 | if (q->qi_bsoftlimit && !d->d_blk_softlimit) { |
1149d96a | 105 | d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit); |
b1366451 BF |
106 | prealloc = 1; |
107 | } | |
108 | if (q->qi_bhardlimit && !d->d_blk_hardlimit) { | |
1149d96a | 109 | d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit); |
b1366451 BF |
110 | prealloc = 1; |
111 | } | |
1da177e4 | 112 | if (q->qi_isoftlimit && !d->d_ino_softlimit) |
1149d96a | 113 | d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit); |
1da177e4 | 114 | if (q->qi_ihardlimit && !d->d_ino_hardlimit) |
1149d96a | 115 | d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit); |
1da177e4 | 116 | if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit) |
1149d96a | 117 | d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit); |
1da177e4 | 118 | if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit) |
1149d96a | 119 | d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit); |
b1366451 BF |
120 | |
121 | if (prealloc) | |
122 | xfs_dquot_set_prealloc_limits(dq); | |
1da177e4 LT |
123 | } |
124 | ||
125 | /* | |
126 | * Check the limits and timers of a dquot and start or reset timers | |
127 | * if necessary. | |
128 | * This gets called even when quota enforcement is OFF, which makes our | |
129 | * life a little less complicated. (We just don't reject any quota | |
130 | * reservations in that case, when enforcement is off). | |
131 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
132 | * enforcement's off. | |
133 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
134 | * 'automatically' get started when limits get exceeded. They do |
135 | * get reset to zero, however, when we find the count to be under | |
136 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
137 | */ |
138 | void | |
139 | xfs_qm_adjust_dqtimers( | |
140 | xfs_mount_t *mp, | |
141 | xfs_disk_dquot_t *d) | |
142 | { | |
143 | ASSERT(d->d_id); | |
144 | ||
ea15ab3c | 145 | #ifdef DEBUG |
1149d96a CH |
146 | if (d->d_blk_hardlimit) |
147 | ASSERT(be64_to_cpu(d->d_blk_softlimit) <= | |
148 | be64_to_cpu(d->d_blk_hardlimit)); | |
149 | if (d->d_ino_hardlimit) | |
150 | ASSERT(be64_to_cpu(d->d_ino_softlimit) <= | |
151 | be64_to_cpu(d->d_ino_hardlimit)); | |
152 | if (d->d_rtb_hardlimit) | |
153 | ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= | |
154 | be64_to_cpu(d->d_rtb_hardlimit)); | |
1da177e4 | 155 | #endif |
ea15ab3c | 156 | |
1da177e4 | 157 | if (!d->d_btimer) { |
1149d96a | 158 | if ((d->d_blk_softlimit && |
d0a3fe67 | 159 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
160 | be64_to_cpu(d->d_blk_softlimit))) || |
161 | (d->d_blk_hardlimit && | |
d0a3fe67 | 162 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
163 | be64_to_cpu(d->d_blk_hardlimit)))) { |
164 | d->d_btimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 165 | mp->m_quotainfo->qi_btimelimit); |
754002b4 NS |
166 | } else { |
167 | d->d_bwarns = 0; | |
1da177e4 LT |
168 | } |
169 | } else { | |
170 | if ((!d->d_blk_softlimit || | |
d0a3fe67 | 171 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 172 | be64_to_cpu(d->d_blk_softlimit))) && |
1da177e4 | 173 | (!d->d_blk_hardlimit || |
d0a3fe67 | 174 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 175 | be64_to_cpu(d->d_blk_hardlimit)))) { |
1da177e4 LT |
176 | d->d_btimer = 0; |
177 | } | |
178 | } | |
179 | ||
180 | if (!d->d_itimer) { | |
1149d96a | 181 | if ((d->d_ino_softlimit && |
d0a3fe67 | 182 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
183 | be64_to_cpu(d->d_ino_softlimit))) || |
184 | (d->d_ino_hardlimit && | |
d0a3fe67 | 185 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
186 | be64_to_cpu(d->d_ino_hardlimit)))) { |
187 | d->d_itimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 188 | mp->m_quotainfo->qi_itimelimit); |
754002b4 NS |
189 | } else { |
190 | d->d_iwarns = 0; | |
1da177e4 LT |
191 | } |
192 | } else { | |
193 | if ((!d->d_ino_softlimit || | |
d0a3fe67 | 194 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 195 | be64_to_cpu(d->d_ino_softlimit))) && |
1da177e4 | 196 | (!d->d_ino_hardlimit || |
d0a3fe67 | 197 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 198 | be64_to_cpu(d->d_ino_hardlimit)))) { |
1da177e4 LT |
199 | d->d_itimer = 0; |
200 | } | |
201 | } | |
202 | ||
203 | if (!d->d_rtbtimer) { | |
1149d96a | 204 | if ((d->d_rtb_softlimit && |
d0a3fe67 | 205 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
206 | be64_to_cpu(d->d_rtb_softlimit))) || |
207 | (d->d_rtb_hardlimit && | |
d0a3fe67 | 208 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
209 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
210 | d->d_rtbtimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 211 | mp->m_quotainfo->qi_rtbtimelimit); |
754002b4 NS |
212 | } else { |
213 | d->d_rtbwarns = 0; | |
1da177e4 LT |
214 | } |
215 | } else { | |
216 | if ((!d->d_rtb_softlimit || | |
d0a3fe67 | 217 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 218 | be64_to_cpu(d->d_rtb_softlimit))) && |
1da177e4 | 219 | (!d->d_rtb_hardlimit || |
d0a3fe67 | 220 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 221 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
1da177e4 LT |
222 | d->d_rtbtimer = 0; |
223 | } | |
224 | } | |
225 | } | |
226 | ||
1da177e4 LT |
227 | /* |
228 | * initialize a buffer full of dquots and log the whole thing | |
229 | */ | |
230 | STATIC void | |
231 | xfs_qm_init_dquot_blk( | |
232 | xfs_trans_t *tp, | |
233 | xfs_mount_t *mp, | |
234 | xfs_dqid_t id, | |
235 | uint type, | |
236 | xfs_buf_t *bp) | |
237 | { | |
8a7b8a89 | 238 | struct xfs_quotainfo *q = mp->m_quotainfo; |
1da177e4 LT |
239 | xfs_dqblk_t *d; |
240 | int curid, i; | |
241 | ||
242 | ASSERT(tp); | |
0c842ad4 | 243 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 244 | |
62926044 | 245 | d = bp->b_addr; |
1da177e4 LT |
246 | |
247 | /* | |
248 | * ID of the first dquot in the block - id's are zero based. | |
249 | */ | |
8a7b8a89 | 250 | curid = id - (id % q->qi_dqperchunk); |
1da177e4 | 251 | ASSERT(curid >= 0); |
8a7b8a89 | 252 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
49d35a5c CH |
253 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) { |
254 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
255 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
256 | d->dd_diskdq.d_id = cpu_to_be32(curid); | |
257 | d->dd_diskdq.d_flags = type; | |
6fcdc59d | 258 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
3fe58f30 | 259 | uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid); |
6fcdc59d DC |
260 | xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk), |
261 | XFS_DQUOT_CRC_OFF); | |
262 | } | |
49d35a5c CH |
263 | } |
264 | ||
1da177e4 | 265 | xfs_trans_dquot_buf(tp, bp, |
c1155410 DC |
266 | (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : |
267 | ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : | |
268 | XFS_BLF_GDQUOT_BUF))); | |
8a7b8a89 | 269 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); |
1da177e4 LT |
270 | } |
271 | ||
b1366451 BF |
272 | /* |
273 | * Initialize the dynamic speculative preallocation thresholds. The lo/hi | |
274 | * watermarks correspond to the soft and hard limits by default. If a soft limit | |
275 | * is not specified, we use 95% of the hard limit. | |
276 | */ | |
277 | void | |
278 | xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp) | |
279 | { | |
280 | __uint64_t space; | |
281 | ||
282 | dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit); | |
283 | dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit); | |
284 | if (!dqp->q_prealloc_lo_wmark) { | |
285 | dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark; | |
286 | do_div(dqp->q_prealloc_lo_wmark, 100); | |
287 | dqp->q_prealloc_lo_wmark *= 95; | |
288 | } | |
289 | ||
290 | space = dqp->q_prealloc_hi_wmark; | |
291 | ||
292 | do_div(space, 100); | |
293 | dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space; | |
294 | dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3; | |
295 | dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5; | |
296 | } | |
297 | ||
1da177e4 LT |
298 | /* |
299 | * Allocate a block and fill it with dquots. | |
300 | * This is called when the bmapi finds a hole. | |
301 | */ | |
302 | STATIC int | |
303 | xfs_qm_dqalloc( | |
efa092f3 | 304 | xfs_trans_t **tpp, |
1da177e4 LT |
305 | xfs_mount_t *mp, |
306 | xfs_dquot_t *dqp, | |
307 | xfs_inode_t *quotip, | |
308 | xfs_fileoff_t offset_fsb, | |
309 | xfs_buf_t **O_bpp) | |
310 | { | |
311 | xfs_fsblock_t firstblock; | |
312 | xfs_bmap_free_t flist; | |
313 | xfs_bmbt_irec_t map; | |
314 | int nmaps, error, committed; | |
315 | xfs_buf_t *bp; | |
efa092f3 | 316 | xfs_trans_t *tp = *tpp; |
1da177e4 LT |
317 | |
318 | ASSERT(tp != NULL); | |
0b1b213f CH |
319 | |
320 | trace_xfs_dqalloc(dqp); | |
1da177e4 LT |
321 | |
322 | /* | |
323 | * Initialize the bmap freelist prior to calling bmapi code. | |
324 | */ | |
9d87c319 | 325 | xfs_bmap_init(&flist, &firstblock); |
1da177e4 LT |
326 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
327 | /* | |
328 | * Return if this type of quotas is turned off while we didn't | |
329 | * have an inode lock | |
330 | */ | |
6967b964 | 331 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 LT |
332 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
333 | return (ESRCH); | |
334 | } | |
335 | ||
ddc3415a | 336 | xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); |
1da177e4 | 337 | nmaps = 1; |
c0dc7828 DC |
338 | error = xfs_bmapi_write(tp, quotip, offset_fsb, |
339 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
340 | &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), | |
341 | &map, &nmaps, &flist); | |
342 | if (error) | |
1da177e4 | 343 | goto error0; |
1da177e4 LT |
344 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
345 | ASSERT(nmaps == 1); | |
346 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
347 | (map.br_startblock != HOLESTARTBLOCK)); | |
348 | ||
349 | /* | |
350 | * Keep track of the blkno to save a lookup later | |
351 | */ | |
352 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
353 | ||
354 | /* now we can just get the buffer (there's nothing to read yet) */ | |
355 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
356 | dqp->q_blkno, | |
8a7b8a89 | 357 | mp->m_quotainfo->qi_dqchunklen, |
1da177e4 | 358 | 0); |
2a30f36d CS |
359 | |
360 | error = xfs_buf_geterror(bp); | |
361 | if (error) | |
1da177e4 | 362 | goto error1; |
1813dd64 | 363 | bp->b_ops = &xfs_dquot_buf_ops; |
2a30f36d | 364 | |
1da177e4 LT |
365 | /* |
366 | * Make a chunk of dquots out of this buffer and log | |
367 | * the entire thing. | |
368 | */ | |
1149d96a | 369 | xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 370 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
1da177e4 | 371 | |
efa092f3 TS |
372 | /* |
373 | * xfs_bmap_finish() may commit the current transaction and | |
374 | * start a second transaction if the freelist is not empty. | |
375 | * | |
376 | * Since we still want to modify this buffer, we need to | |
377 | * ensure that the buffer is not released on commit of | |
378 | * the first transaction and ensure the buffer is added to the | |
379 | * second transaction. | |
380 | * | |
381 | * If there is only one transaction then don't stop the buffer | |
382 | * from being released when it commits later on. | |
383 | */ | |
384 | ||
385 | xfs_trans_bhold(tp, bp); | |
386 | ||
f7c99b6f | 387 | if ((error = xfs_bmap_finish(tpp, &flist, &committed))) { |
1da177e4 LT |
388 | goto error1; |
389 | } | |
390 | ||
efa092f3 TS |
391 | if (committed) { |
392 | tp = *tpp; | |
393 | xfs_trans_bjoin(tp, bp); | |
394 | } else { | |
395 | xfs_trans_bhold_release(tp, bp); | |
396 | } | |
397 | ||
1da177e4 LT |
398 | *O_bpp = bp; |
399 | return 0; | |
400 | ||
401 | error1: | |
402 | xfs_bmap_cancel(&flist); | |
403 | error0: | |
404 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
405 | ||
406 | return (error); | |
407 | } | |
9aede1d8 | 408 | |
c6319198 DC |
409 | STATIC int |
410 | xfs_qm_dqrepair( | |
411 | struct xfs_mount *mp, | |
412 | struct xfs_trans *tp, | |
413 | struct xfs_dquot *dqp, | |
414 | xfs_dqid_t firstid, | |
415 | struct xfs_buf **bpp) | |
416 | { | |
417 | int error; | |
418 | struct xfs_disk_dquot *ddq; | |
419 | struct xfs_dqblk *d; | |
420 | int i; | |
421 | ||
422 | /* | |
423 | * Read the buffer without verification so we get the corrupted | |
612cfbfe | 424 | * buffer returned to us. make sure we verify it on write, though. |
c6319198 DC |
425 | */ |
426 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno, | |
427 | mp->m_quotainfo->qi_dqchunklen, | |
428 | 0, bpp, NULL); | |
429 | ||
430 | if (error) { | |
431 | ASSERT(*bpp == NULL); | |
432 | return XFS_ERROR(error); | |
433 | } | |
1813dd64 | 434 | (*bpp)->b_ops = &xfs_dquot_buf_ops; |
c6319198 DC |
435 | |
436 | ASSERT(xfs_buf_islocked(*bpp)); | |
437 | d = (struct xfs_dqblk *)(*bpp)->b_addr; | |
438 | ||
439 | /* Do the actual repair of dquots in this buffer */ | |
440 | for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) { | |
441 | ddq = &d[i].dd_diskdq; | |
9aede1d8 | 442 | error = xfs_dqcheck(mp, ddq, firstid + i, |
c6319198 DC |
443 | dqp->dq_flags & XFS_DQ_ALLTYPES, |
444 | XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair"); | |
445 | if (error) { | |
446 | /* repair failed, we're screwed */ | |
447 | xfs_trans_brelse(tp, *bpp); | |
448 | return XFS_ERROR(EIO); | |
449 | } | |
450 | } | |
451 | ||
452 | return 0; | |
453 | } | |
454 | ||
1da177e4 LT |
455 | /* |
456 | * Maps a dquot to the buffer containing its on-disk version. | |
457 | * This returns a ptr to the buffer containing the on-disk dquot | |
458 | * in the bpp param, and a ptr to the on-disk dquot within that buffer | |
459 | */ | |
460 | STATIC int | |
461 | xfs_qm_dqtobp( | |
efa092f3 | 462 | xfs_trans_t **tpp, |
1da177e4 LT |
463 | xfs_dquot_t *dqp, |
464 | xfs_disk_dquot_t **O_ddpp, | |
465 | xfs_buf_t **O_bpp, | |
466 | uint flags) | |
467 | { | |
113a5683 CS |
468 | struct xfs_bmbt_irec map; |
469 | int nmaps = 1, error; | |
470 | struct xfs_buf *bp; | |
995961c4 | 471 | struct xfs_inode *quotip = xfs_dq_to_quota_inode(dqp); |
113a5683 CS |
472 | struct xfs_mount *mp = dqp->q_mount; |
473 | xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); | |
474 | struct xfs_trans *tp = (tpp ? *tpp : NULL); | |
1da177e4 | 475 | |
acecf1b5 | 476 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
1da177e4 | 477 | |
acecf1b5 | 478 | xfs_ilock(quotip, XFS_ILOCK_SHARED); |
6967b964 | 479 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 | 480 | /* |
acecf1b5 CH |
481 | * Return if this type of quotas is turned off while we |
482 | * didn't have the quota inode lock. | |
1da177e4 | 483 | */ |
acecf1b5 CH |
484 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); |
485 | return ESRCH; | |
486 | } | |
487 | ||
488 | /* | |
489 | * Find the block map; no allocations yet | |
490 | */ | |
5c8ed202 DC |
491 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
492 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); | |
acecf1b5 CH |
493 | |
494 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); | |
495 | if (error) | |
496 | return error; | |
497 | ||
498 | ASSERT(nmaps == 1); | |
499 | ASSERT(map.br_blockcount == 1); | |
500 | ||
501 | /* | |
502 | * Offset of dquot in the (fixed sized) dquot chunk. | |
503 | */ | |
504 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
505 | sizeof(xfs_dqblk_t); | |
506 | ||
507 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
508 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 509 | /* |
acecf1b5 | 510 | * We don't allocate unless we're asked to |
1da177e4 | 511 | */ |
acecf1b5 CH |
512 | if (!(flags & XFS_QMOPT_DQALLOC)) |
513 | return ENOENT; | |
1da177e4 | 514 | |
acecf1b5 CH |
515 | ASSERT(tp); |
516 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
517 | dqp->q_fileoffset, &bp); | |
1da177e4 | 518 | if (error) |
acecf1b5 CH |
519 | return error; |
520 | tp = *tpp; | |
521 | } else { | |
522 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
523 | |
524 | /* | |
acecf1b5 CH |
525 | * store the blkno etc so that we don't have to do the |
526 | * mapping all the time | |
1da177e4 | 527 | */ |
acecf1b5 | 528 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 529 | |
8a7b8a89 CH |
530 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
531 | dqp->q_blkno, | |
532 | mp->m_quotainfo->qi_dqchunklen, | |
1813dd64 | 533 | 0, &bp, &xfs_dquot_buf_ops); |
acecf1b5 | 534 | |
c6319198 DC |
535 | if (error == EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) { |
536 | xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff * | |
537 | mp->m_quotainfo->qi_dqperchunk; | |
538 | ASSERT(bp == NULL); | |
539 | error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp); | |
540 | } | |
1da177e4 | 541 | |
c6319198 DC |
542 | if (error) { |
543 | ASSERT(bp == NULL); | |
544 | return XFS_ERROR(error); | |
1da177e4 | 545 | } |
1da177e4 LT |
546 | } |
547 | ||
c6319198 | 548 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 549 | *O_bpp = bp; |
c6319198 | 550 | *O_ddpp = bp->b_addr + dqp->q_bufoffset; |
1da177e4 LT |
551 | |
552 | return (0); | |
553 | } | |
554 | ||
555 | ||
556 | /* | |
557 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
558 | * and release the buffer immediately. | |
559 | * | |
97e7ade5 | 560 | * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed. |
1da177e4 | 561 | */ |
7ae44407 | 562 | int |
1da177e4 | 563 | xfs_qm_dqread( |
97e7ade5 CH |
564 | struct xfs_mount *mp, |
565 | xfs_dqid_t id, | |
566 | uint type, | |
567 | uint flags, | |
568 | struct xfs_dquot **O_dqpp) | |
1da177e4 | 569 | { |
97e7ade5 CH |
570 | struct xfs_dquot *dqp; |
571 | struct xfs_disk_dquot *ddqp; | |
572 | struct xfs_buf *bp; | |
573 | struct xfs_trans *tp = NULL; | |
574 | int error; | |
575 | int cancelflags = 0; | |
efa092f3 | 576 | |
92b2e5b3 | 577 | |
a05931ce | 578 | dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP); |
92b2e5b3 CH |
579 | |
580 | dqp->dq_flags = type; | |
581 | dqp->q_core.d_id = cpu_to_be32(id); | |
582 | dqp->q_mount = mp; | |
f8739c3c | 583 | INIT_LIST_HEAD(&dqp->q_lru); |
92b2e5b3 CH |
584 | mutex_init(&dqp->q_qlock); |
585 | init_waitqueue_head(&dqp->q_pinwait); | |
586 | ||
587 | /* | |
588 | * Because we want to use a counting completion, complete | |
589 | * the flush completion once to allow a single access to | |
590 | * the flush completion without blocking. | |
591 | */ | |
592 | init_completion(&dqp->q_flush); | |
593 | complete(&dqp->q_flush); | |
594 | ||
595 | /* | |
596 | * Make sure group quotas have a different lock class than user | |
597 | * quotas. | |
598 | */ | |
f112a049 DC |
599 | switch (type) { |
600 | case XFS_DQ_USER: | |
601 | /* uses the default lock class */ | |
602 | break; | |
603 | case XFS_DQ_GROUP: | |
604 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class); | |
605 | break; | |
606 | case XFS_DQ_PROJ: | |
607 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class); | |
608 | break; | |
609 | default: | |
610 | ASSERT(0); | |
611 | break; | |
612 | } | |
92b2e5b3 | 613 | |
48776fd2 | 614 | XFS_STATS_INC(xs_qm_dquot); |
1da177e4 | 615 | |
0b1b213f CH |
616 | trace_xfs_dqread(dqp); |
617 | ||
97e7ade5 CH |
618 | if (flags & XFS_QMOPT_DQALLOC) { |
619 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC); | |
3d3c8b52 JL |
620 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_attrsetm, |
621 | XFS_QM_DQALLOC_SPACE_RES(mp), 0); | |
97e7ade5 CH |
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 | 654 | |
b1366451 BF |
655 | /* initialize the dquot speculative prealloc thresholds */ |
656 | xfs_dquot_set_prealloc_limits(dqp); | |
657 | ||
1da177e4 | 658 | /* Mark the buf so that this will stay incore a little longer */ |
38f23232 | 659 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 LT |
660 | |
661 | /* | |
662 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
663 | * So we need to release with xfs_trans_brelse(). | |
664 | * The strategy here is identical to that of inodes; we lock | |
665 | * the dquot in xfs_qm_dqget() before making it accessible to | |
666 | * others. This is because dquots, like inodes, need a good level of | |
667 | * concurrency, and we don't want to take locks on the entire buffers | |
668 | * for dquot accesses. | |
669 | * Note also that the dquot buffer may even be dirty at this point, if | |
670 | * this particular dquot was repaired. We still aren't afraid to | |
671 | * brelse it because we have the changes incore. | |
672 | */ | |
0c842ad4 | 673 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
674 | xfs_trans_brelse(tp, bp); |
675 | ||
1da177e4 | 676 | if (tp) { |
97e7ade5 CH |
677 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
678 | if (error) | |
679 | goto error0; | |
1da177e4 LT |
680 | } |
681 | ||
682 | *O_dqpp = dqp; | |
97e7ade5 | 683 | return error; |
1da177e4 | 684 | |
97e7ade5 | 685 | error1: |
1da177e4 LT |
686 | if (tp) |
687 | xfs_trans_cancel(tp, cancelflags); | |
97e7ade5 | 688 | error0: |
1da177e4 LT |
689 | xfs_qm_dqdestroy(dqp); |
690 | *O_dqpp = NULL; | |
97e7ade5 | 691 | return error; |
1da177e4 LT |
692 | } |
693 | ||
1da177e4 LT |
694 | /* |
695 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
696 | * a locked dquot, doing an allocation (if requested) as needed. | |
697 | * When both an inode and an id are given, the inode's id takes precedence. | |
698 | * That is, if the id changes while we don't hold the ilock inside this | |
699 | * function, the new dquot is returned, not necessarily the one requested | |
700 | * in the id argument. | |
701 | */ | |
702 | int | |
703 | xfs_qm_dqget( | |
704 | xfs_mount_t *mp, | |
705 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
706 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
707 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
708 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
709 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
710 | { | |
9f920f11 | 711 | struct xfs_quotainfo *qi = mp->m_quotainfo; |
329e0875 | 712 | struct radix_tree_root *tree = xfs_dquot_tree(qi, type); |
9f920f11 CH |
713 | struct xfs_dquot *dqp; |
714 | int error; | |
1da177e4 LT |
715 | |
716 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
717 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 718 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 LT |
719 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
720 | return (ESRCH); | |
721 | } | |
1da177e4 LT |
722 | |
723 | #ifdef DEBUG | |
724 | if (xfs_do_dqerror) { | |
725 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
726 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 727 | xfs_debug(mp, "Returning error in dqget"); |
1da177e4 LT |
728 | return (EIO); |
729 | } | |
730 | } | |
1da177e4 | 731 | |
c8ad20ff NS |
732 | ASSERT(type == XFS_DQ_USER || |
733 | type == XFS_DQ_PROJ || | |
734 | type == XFS_DQ_GROUP); | |
1da177e4 | 735 | if (ip) { |
579aa9ca | 736 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
36731410 | 737 | ASSERT(xfs_inode_dquot(ip, type) == NULL); |
1da177e4 LT |
738 | } |
739 | #endif | |
92678554 CH |
740 | |
741 | restart: | |
9f920f11 CH |
742 | mutex_lock(&qi->qi_tree_lock); |
743 | dqp = radix_tree_lookup(tree, id); | |
744 | if (dqp) { | |
745 | xfs_dqlock(dqp); | |
746 | if (dqp->dq_flags & XFS_DQ_FREEING) { | |
747 | xfs_dqunlock(dqp); | |
748 | mutex_unlock(&qi->qi_tree_lock); | |
749 | trace_xfs_dqget_freeing(dqp); | |
750 | delay(1); | |
751 | goto restart; | |
752 | } | |
1da177e4 | 753 | |
9f920f11 CH |
754 | dqp->q_nrefs++; |
755 | mutex_unlock(&qi->qi_tree_lock); | |
756 | ||
757 | trace_xfs_dqget_hit(dqp); | |
48776fd2 | 758 | XFS_STATS_INC(xs_qm_dqcachehits); |
9f920f11 CH |
759 | *O_dqpp = dqp; |
760 | return 0; | |
1da177e4 | 761 | } |
9f920f11 CH |
762 | mutex_unlock(&qi->qi_tree_lock); |
763 | XFS_STATS_INC(xs_qm_dqcachemisses); | |
1da177e4 LT |
764 | |
765 | /* | |
766 | * Dquot cache miss. We don't want to keep the inode lock across | |
767 | * a (potential) disk read. Also we don't want to deal with the lock | |
768 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
769 | * lock here means dealing with a chown that can happen before | |
770 | * we re-acquire the lock. | |
771 | */ | |
772 | if (ip) | |
773 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 | 774 | |
97e7ade5 | 775 | error = xfs_qm_dqread(mp, id, type, flags, &dqp); |
1da177e4 | 776 | |
7ae44407 CH |
777 | if (ip) |
778 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
779 | ||
780 | if (error) | |
781 | return error; | |
1da177e4 | 782 | |
1da177e4 | 783 | if (ip) { |
1da177e4 LT |
784 | /* |
785 | * A dquot could be attached to this inode by now, since | |
786 | * we had dropped the ilock. | |
787 | */ | |
36731410 | 788 | if (xfs_this_quota_on(mp, type)) { |
9f920f11 CH |
789 | struct xfs_dquot *dqp1; |
790 | ||
36731410 CS |
791 | dqp1 = xfs_inode_dquot(ip, type); |
792 | if (dqp1) { | |
1da177e4 | 793 | xfs_qm_dqdestroy(dqp); |
36731410 | 794 | dqp = dqp1; |
1da177e4 LT |
795 | xfs_dqlock(dqp); |
796 | goto dqret; | |
797 | } | |
798 | } else { | |
36731410 CS |
799 | /* inode stays locked on return */ |
800 | xfs_qm_dqdestroy(dqp); | |
801 | return XFS_ERROR(ESRCH); | |
1da177e4 LT |
802 | } |
803 | } | |
804 | ||
9f920f11 CH |
805 | mutex_lock(&qi->qi_tree_lock); |
806 | error = -radix_tree_insert(tree, id, dqp); | |
807 | if (unlikely(error)) { | |
808 | WARN_ON(error != EEXIST); | |
809 | ||
1da177e4 | 810 | /* |
9f920f11 CH |
811 | * Duplicate found. Just throw away the new dquot and start |
812 | * over. | |
1da177e4 | 813 | */ |
9f920f11 CH |
814 | mutex_unlock(&qi->qi_tree_lock); |
815 | trace_xfs_dqget_dup(dqp); | |
816 | xfs_qm_dqdestroy(dqp); | |
817 | XFS_STATS_INC(xs_qm_dquot_dups); | |
818 | goto restart; | |
1da177e4 LT |
819 | } |
820 | ||
1da177e4 LT |
821 | /* |
822 | * We return a locked dquot to the caller, with a reference taken | |
823 | */ | |
824 | xfs_dqlock(dqp); | |
825 | dqp->q_nrefs = 1; | |
826 | ||
9f920f11 | 827 | qi->qi_dquots++; |
9f920f11 CH |
828 | mutex_unlock(&qi->qi_tree_lock); |
829 | ||
1da177e4 | 830 | dqret: |
579aa9ca | 831 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 832 | trace_xfs_dqget_miss(dqp); |
1da177e4 LT |
833 | *O_dqpp = dqp; |
834 | return (0); | |
835 | } | |
836 | ||
837 | ||
f8739c3c CH |
838 | STATIC void |
839 | xfs_qm_dqput_final( | |
bf72de31 | 840 | struct xfs_dquot *dqp) |
1da177e4 | 841 | { |
f8739c3c | 842 | struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo; |
bf72de31 | 843 | struct xfs_dquot *gdqp; |
92f8ff73 | 844 | struct xfs_dquot *pdqp; |
1da177e4 | 845 | |
bf72de31 CH |
846 | trace_xfs_dqput_free(dqp); |
847 | ||
cd56a39a | 848 | if (list_lru_add(&qi->qi_lru, &dqp->q_lru)) |
48776fd2 | 849 | XFS_STATS_INC(xs_qm_dquot_unused); |
bf72de31 | 850 | |
1da177e4 | 851 | /* |
bf72de31 | 852 | * If we just added a udquot to the freelist, then we want to release |
92f8ff73 CS |
853 | * the gdquot/pdquot reference that it (probably) has. Otherwise it'll |
854 | * keep the gdquot/pdquot from getting reclaimed. | |
1da177e4 | 855 | */ |
bf72de31 CH |
856 | gdqp = dqp->q_gdquot; |
857 | if (gdqp) { | |
858 | xfs_dqlock(gdqp); | |
859 | dqp->q_gdquot = NULL; | |
1da177e4 | 860 | } |
92f8ff73 CS |
861 | |
862 | pdqp = dqp->q_pdquot; | |
863 | if (pdqp) { | |
864 | xfs_dqlock(pdqp); | |
865 | dqp->q_pdquot = NULL; | |
866 | } | |
bf72de31 | 867 | xfs_dqunlock(dqp); |
1da177e4 | 868 | |
bf72de31 | 869 | /* |
92f8ff73 | 870 | * If we had a group/project quota hint, release it now. |
bf72de31 | 871 | */ |
f8739c3c CH |
872 | if (gdqp) |
873 | xfs_qm_dqput(gdqp); | |
92f8ff73 CS |
874 | if (pdqp) |
875 | xfs_qm_dqput(pdqp); | |
f8739c3c CH |
876 | } |
877 | ||
878 | /* | |
879 | * Release a reference to the dquot (decrement ref-count) and unlock it. | |
880 | * | |
881 | * If there is a group quota attached to this dquot, carefully release that | |
882 | * too without tripping over deadlocks'n'stuff. | |
883 | */ | |
884 | void | |
885 | xfs_qm_dqput( | |
886 | struct xfs_dquot *dqp) | |
887 | { | |
888 | ASSERT(dqp->q_nrefs > 0); | |
889 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
890 | ||
891 | trace_xfs_dqput(dqp); | |
892 | ||
893 | if (--dqp->q_nrefs > 0) | |
894 | xfs_dqunlock(dqp); | |
895 | else | |
896 | xfs_qm_dqput_final(dqp); | |
1da177e4 LT |
897 | } |
898 | ||
899 | /* | |
900 | * Release a dquot. Flush it if dirty, then dqput() it. | |
901 | * dquot must not be locked. | |
902 | */ | |
903 | void | |
904 | xfs_qm_dqrele( | |
905 | xfs_dquot_t *dqp) | |
906 | { | |
7d095257 CH |
907 | if (!dqp) |
908 | return; | |
909 | ||
0b1b213f | 910 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
911 | |
912 | xfs_dqlock(dqp); | |
913 | /* | |
914 | * We don't care to flush it if the dquot is dirty here. | |
915 | * That will create stutters that we want to avoid. | |
916 | * Instead we do a delayed write when we try to reclaim | |
917 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
918 | */ | |
919 | xfs_qm_dqput(dqp); | |
920 | } | |
921 | ||
ca30b2a7 CH |
922 | /* |
923 | * This is the dquot flushing I/O completion routine. It is called | |
924 | * from interrupt level when the buffer containing the dquot is | |
925 | * flushed to disk. It is responsible for removing the dquot logitem | |
926 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
927 | * flush lock. This behavior is very similar to that of inodes.. | |
928 | */ | |
929 | STATIC void | |
930 | xfs_qm_dqflush_done( | |
931 | struct xfs_buf *bp, | |
932 | struct xfs_log_item *lip) | |
933 | { | |
934 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
935 | xfs_dquot_t *dqp = qip->qli_dquot; | |
936 | struct xfs_ail *ailp = lip->li_ailp; | |
937 | ||
938 | /* | |
939 | * We only want to pull the item from the AIL if its | |
940 | * location in the log has not changed since we started the flush. | |
941 | * Thus, we only bother if the dquot's lsn has | |
942 | * not changed. First we check the lsn outside the lock | |
943 | * since it's cheaper, and then we recheck while | |
944 | * holding the lock before removing the dquot from the AIL. | |
945 | */ | |
946 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
947 | lip->li_lsn == qip->qli_flush_lsn) { | |
948 | ||
949 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
950 | spin_lock(&ailp->xa_lock); | |
951 | if (lip->li_lsn == qip->qli_flush_lsn) | |
04913fdd | 952 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 CH |
953 | else |
954 | spin_unlock(&ailp->xa_lock); | |
955 | } | |
956 | ||
957 | /* | |
958 | * Release the dq's flush lock since we're done with it. | |
959 | */ | |
960 | xfs_dqfunlock(dqp); | |
961 | } | |
1da177e4 LT |
962 | |
963 | /* | |
964 | * Write a modified dquot to disk. | |
965 | * The dquot must be locked and the flush lock too taken by caller. | |
966 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
967 | * but the dquot is free to be unlocked and modified by the caller | |
968 | * in the interim. Dquot is still locked on return. This behavior is | |
969 | * identical to that of inodes. | |
970 | */ | |
971 | int | |
972 | xfs_qm_dqflush( | |
fe7257fd CH |
973 | struct xfs_dquot *dqp, |
974 | struct xfs_buf **bpp) | |
1da177e4 | 975 | { |
acecf1b5 CH |
976 | struct xfs_mount *mp = dqp->q_mount; |
977 | struct xfs_buf *bp; | |
978 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 979 | int error; |
1da177e4 LT |
980 | |
981 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 982 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 983 | |
0b1b213f | 984 | trace_xfs_dqflush(dqp); |
1da177e4 | 985 | |
fe7257fd CH |
986 | *bpp = NULL; |
987 | ||
1da177e4 LT |
988 | xfs_qm_dqunpin_wait(dqp); |
989 | ||
990 | /* | |
991 | * This may have been unpinned because the filesystem is shutting | |
992 | * down forcibly. If that's the case we must not write this dquot | |
dea96095 CH |
993 | * to disk, because the log record didn't make it to disk. |
994 | * | |
995 | * We also have to remove the log item from the AIL in this case, | |
996 | * as we wait for an emptry AIL as part of the unmount process. | |
1da177e4 | 997 | */ |
acecf1b5 | 998 | if (XFS_FORCED_SHUTDOWN(mp)) { |
dea96095 | 999 | struct xfs_log_item *lip = &dqp->q_logitem.qli_item; |
acecf1b5 | 1000 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
dea96095 CH |
1001 | |
1002 | spin_lock(&mp->m_ail->xa_lock); | |
1003 | if (lip->li_flags & XFS_LI_IN_AIL) | |
04913fdd DC |
1004 | xfs_trans_ail_delete(mp->m_ail, lip, |
1005 | SHUTDOWN_CORRUPT_INCORE); | |
dea96095 CH |
1006 | else |
1007 | spin_unlock(&mp->m_ail->xa_lock); | |
fe7257fd CH |
1008 | error = XFS_ERROR(EIO); |
1009 | goto out_unlock; | |
1da177e4 LT |
1010 | } |
1011 | ||
1012 | /* | |
1013 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1014 | */ |
acecf1b5 | 1015 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
c3f8fc73 | 1016 | mp->m_quotainfo->qi_dqchunklen, 0, &bp, NULL); |
fe7257fd CH |
1017 | if (error) |
1018 | goto out_unlock; | |
1da177e4 | 1019 | |
acecf1b5 CH |
1020 | /* |
1021 | * Calculate the location of the dquot inside the buffer. | |
1022 | */ | |
62926044 | 1023 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1024 | |
1025 | /* | |
1026 | * A simple sanity check in case we got a corrupted dquot.. | |
1027 | */ | |
9aede1d8 | 1028 | error = xfs_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
a0fa2b67 DC |
1029 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); |
1030 | if (error) { | |
acecf1b5 CH |
1031 | xfs_buf_relse(bp); |
1032 | xfs_dqfunlock(dqp); | |
1033 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
1da177e4 LT |
1034 | return XFS_ERROR(EIO); |
1035 | } | |
1036 | ||
1037 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1038 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1039 | |
1040 | /* | |
1041 | * Clear the dirty field and remember the flush lsn for later use. | |
1042 | */ | |
acecf1b5 | 1043 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1044 | |
7b2e2a31 DC |
1045 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1046 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 | 1047 | |
3fe58f30 CH |
1048 | /* |
1049 | * copy the lsn into the on-disk dquot now while we have the in memory | |
1050 | * dquot here. This can't be done later in the write verifier as we | |
1051 | * can't get access to the log item at that point in time. | |
6fcdc59d DC |
1052 | * |
1053 | * We also calculate the CRC here so that the on-disk dquot in the | |
1054 | * buffer always has a valid CRC. This ensures there is no possibility | |
1055 | * of a dquot without an up-to-date CRC getting to disk. | |
3fe58f30 CH |
1056 | */ |
1057 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
1058 | struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp; | |
1059 | ||
1060 | dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn); | |
6fcdc59d DC |
1061 | xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk), |
1062 | XFS_DQUOT_CRC_OFF); | |
3fe58f30 CH |
1063 | } |
1064 | ||
1da177e4 LT |
1065 | /* |
1066 | * Attach an iodone routine so that we can remove this dquot from the | |
1067 | * AIL and release the flush lock once the dquot is synced to disk. | |
1068 | */ | |
ca30b2a7 CH |
1069 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1070 | &dqp->q_logitem.qli_item); | |
1071 | ||
1da177e4 LT |
1072 | /* |
1073 | * If the buffer is pinned then push on the log so we won't | |
1074 | * get stuck waiting in the write for too long. | |
1075 | */ | |
811e64c7 | 1076 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1077 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1078 | xfs_log_force(mp, 0); |
1da177e4 LT |
1079 | } |
1080 | ||
0b1b213f | 1081 | trace_xfs_dqflush_done(dqp); |
fe7257fd CH |
1082 | *bpp = bp; |
1083 | return 0; | |
0b1b213f | 1084 | |
fe7257fd CH |
1085 | out_unlock: |
1086 | xfs_dqfunlock(dqp); | |
1087 | return XFS_ERROR(EIO); | |
1da177e4 LT |
1088 | } |
1089 | ||
5bb87a33 CH |
1090 | /* |
1091 | * Lock two xfs_dquot structures. | |
1092 | * | |
1093 | * To avoid deadlocks we always lock the quota structure with | |
1094 | * the lowerd id first. | |
1095 | */ | |
1da177e4 LT |
1096 | void |
1097 | xfs_dqlock2( | |
1098 | xfs_dquot_t *d1, | |
1099 | xfs_dquot_t *d2) | |
1100 | { | |
1101 | if (d1 && d2) { | |
1102 | ASSERT(d1 != d2); | |
1149d96a CH |
1103 | if (be32_to_cpu(d1->q_core.d_id) > |
1104 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1105 | mutex_lock(&d2->q_qlock); |
1106 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1107 | } else { |
5bb87a33 CH |
1108 | mutex_lock(&d1->q_qlock); |
1109 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1110 | } |
5bb87a33 CH |
1111 | } else if (d1) { |
1112 | mutex_lock(&d1->q_qlock); | |
1113 | } else if (d2) { | |
1114 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1115 | } |
1116 | } | |
1117 | ||
a05931ce CH |
1118 | int __init |
1119 | xfs_qm_init(void) | |
1120 | { | |
1121 | xfs_qm_dqzone = | |
1122 | kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot"); | |
1123 | if (!xfs_qm_dqzone) | |
1124 | goto out; | |
1125 | ||
1126 | xfs_qm_dqtrxzone = | |
1127 | kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx"); | |
1128 | if (!xfs_qm_dqtrxzone) | |
1129 | goto out_free_dqzone; | |
1130 | ||
1131 | return 0; | |
1132 | ||
1133 | out_free_dqzone: | |
1134 | kmem_zone_destroy(xfs_qm_dqzone); | |
1135 | out: | |
1136 | return -ENOMEM; | |
1137 | } | |
1138 | ||
1c2ccc66 | 1139 | void |
a05931ce CH |
1140 | xfs_qm_exit(void) |
1141 | { | |
1142 | kmem_zone_destroy(xfs_qm_dqtrxzone); | |
1143 | kmem_zone_destroy(xfs_qm_dqzone); | |
1144 | } |