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 | 25 | #include "xfs_mount.h" |
1da177e4 LT |
26 | #include "xfs_inode.h" |
27 | #include "xfs_bmap.h" | |
68988114 | 28 | #include "xfs_bmap_util.h" |
239880ef DC |
29 | #include "xfs_alloc.h" |
30 | #include "xfs_quota.h" | |
1da177e4 | 31 | #include "xfs_error.h" |
239880ef | 32 | #include "xfs_trans.h" |
1da177e4 LT |
33 | #include "xfs_buf_item.h" |
34 | #include "xfs_trans_space.h" | |
35 | #include "xfs_trans_priv.h" | |
1da177e4 | 36 | #include "xfs_qm.h" |
3fe58f30 | 37 | #include "xfs_cksum.h" |
0b1b213f | 38 | #include "xfs_trace.h" |
239880ef | 39 | #include "xfs_log.h" |
a4fbe6ab | 40 | #include "xfs_bmap_btree.h" |
1da177e4 | 41 | |
1da177e4 | 42 | /* |
bf72de31 CH |
43 | * Lock order: |
44 | * | |
45 | * ip->i_lock | |
9f920f11 | 46 | * qi->qi_tree_lock |
b84a3a96 CH |
47 | * dquot->q_qlock (xfs_dqlock() and friends) |
48 | * dquot->q_flush (xfs_dqflock() and friends) | |
49 | * qi->qi_lru_lock | |
bf72de31 CH |
50 | * |
51 | * If two dquots need to be locked the order is user before group/project, | |
52 | * otherwise by the lowest id first, see xfs_dqlock2. | |
53 | */ | |
1da177e4 | 54 | |
1da177e4 LT |
55 | #ifdef DEBUG |
56 | xfs_buftarg_t *xfs_dqerror_target; | |
57 | int xfs_do_dqerror; | |
58 | int xfs_dqreq_num; | |
59 | int xfs_dqerror_mod = 33; | |
60 | #endif | |
61 | ||
a05931ce CH |
62 | struct kmem_zone *xfs_qm_dqtrxzone; |
63 | static struct kmem_zone *xfs_qm_dqzone; | |
64 | ||
f112a049 DC |
65 | static struct lock_class_key xfs_dquot_group_class; |
66 | static struct lock_class_key xfs_dquot_project_class; | |
98b8c7a0 | 67 | |
1da177e4 LT |
68 | /* |
69 | * This is called to free all the memory associated with a dquot | |
70 | */ | |
71 | void | |
72 | xfs_qm_dqdestroy( | |
73 | xfs_dquot_t *dqp) | |
74 | { | |
f8739c3c | 75 | ASSERT(list_empty(&dqp->q_lru)); |
1da177e4 LT |
76 | |
77 | mutex_destroy(&dqp->q_qlock); | |
0b1b213f | 78 | |
ff6d6af2 BD |
79 | XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot); |
80 | kmem_zone_free(xfs_qm_dqzone, dqp); | |
1da177e4 LT |
81 | } |
82 | ||
1da177e4 LT |
83 | /* |
84 | * If default limits are in force, push them into the dquot now. | |
85 | * We overwrite the dquot limits only if they are zero and this | |
86 | * is not the root dquot. | |
87 | */ | |
88 | void | |
89 | xfs_qm_adjust_dqlimits( | |
4b6eae2e BF |
90 | struct xfs_mount *mp, |
91 | struct xfs_dquot *dq) | |
1da177e4 | 92 | { |
4b6eae2e BF |
93 | struct xfs_quotainfo *q = mp->m_quotainfo; |
94 | struct xfs_disk_dquot *d = &dq->q_core; | |
b1366451 | 95 | int prealloc = 0; |
1da177e4 LT |
96 | |
97 | ASSERT(d->d_id); | |
98 | ||
b1366451 | 99 | if (q->qi_bsoftlimit && !d->d_blk_softlimit) { |
1149d96a | 100 | d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit); |
b1366451 BF |
101 | prealloc = 1; |
102 | } | |
103 | if (q->qi_bhardlimit && !d->d_blk_hardlimit) { | |
1149d96a | 104 | d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit); |
b1366451 BF |
105 | prealloc = 1; |
106 | } | |
1da177e4 | 107 | if (q->qi_isoftlimit && !d->d_ino_softlimit) |
1149d96a | 108 | d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit); |
1da177e4 | 109 | if (q->qi_ihardlimit && !d->d_ino_hardlimit) |
1149d96a | 110 | d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit); |
1da177e4 | 111 | if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit) |
1149d96a | 112 | d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit); |
1da177e4 | 113 | if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit) |
1149d96a | 114 | d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit); |
b1366451 BF |
115 | |
116 | if (prealloc) | |
117 | xfs_dquot_set_prealloc_limits(dq); | |
1da177e4 LT |
118 | } |
119 | ||
120 | /* | |
121 | * Check the limits and timers of a dquot and start or reset timers | |
122 | * if necessary. | |
123 | * This gets called even when quota enforcement is OFF, which makes our | |
124 | * life a little less complicated. (We just don't reject any quota | |
125 | * reservations in that case, when enforcement is off). | |
126 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
127 | * enforcement's off. | |
128 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
129 | * 'automatically' get started when limits get exceeded. They do |
130 | * get reset to zero, however, when we find the count to be under | |
131 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
132 | */ |
133 | void | |
134 | xfs_qm_adjust_dqtimers( | |
135 | xfs_mount_t *mp, | |
136 | xfs_disk_dquot_t *d) | |
137 | { | |
138 | ASSERT(d->d_id); | |
139 | ||
ea15ab3c | 140 | #ifdef DEBUG |
1149d96a CH |
141 | if (d->d_blk_hardlimit) |
142 | ASSERT(be64_to_cpu(d->d_blk_softlimit) <= | |
143 | be64_to_cpu(d->d_blk_hardlimit)); | |
144 | if (d->d_ino_hardlimit) | |
145 | ASSERT(be64_to_cpu(d->d_ino_softlimit) <= | |
146 | be64_to_cpu(d->d_ino_hardlimit)); | |
147 | if (d->d_rtb_hardlimit) | |
148 | ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= | |
149 | be64_to_cpu(d->d_rtb_hardlimit)); | |
1da177e4 | 150 | #endif |
ea15ab3c | 151 | |
1da177e4 | 152 | if (!d->d_btimer) { |
1149d96a | 153 | if ((d->d_blk_softlimit && |
d0a3fe67 | 154 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
155 | be64_to_cpu(d->d_blk_softlimit))) || |
156 | (d->d_blk_hardlimit && | |
d0a3fe67 | 157 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
158 | be64_to_cpu(d->d_blk_hardlimit)))) { |
159 | d->d_btimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 160 | mp->m_quotainfo->qi_btimelimit); |
754002b4 NS |
161 | } else { |
162 | d->d_bwarns = 0; | |
1da177e4 LT |
163 | } |
164 | } else { | |
165 | if ((!d->d_blk_softlimit || | |
d0a3fe67 | 166 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 167 | be64_to_cpu(d->d_blk_softlimit))) && |
1da177e4 | 168 | (!d->d_blk_hardlimit || |
d0a3fe67 | 169 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 170 | be64_to_cpu(d->d_blk_hardlimit)))) { |
1da177e4 LT |
171 | d->d_btimer = 0; |
172 | } | |
173 | } | |
174 | ||
175 | if (!d->d_itimer) { | |
1149d96a | 176 | if ((d->d_ino_softlimit && |
d0a3fe67 | 177 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
178 | be64_to_cpu(d->d_ino_softlimit))) || |
179 | (d->d_ino_hardlimit && | |
d0a3fe67 | 180 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
181 | be64_to_cpu(d->d_ino_hardlimit)))) { |
182 | d->d_itimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 183 | mp->m_quotainfo->qi_itimelimit); |
754002b4 NS |
184 | } else { |
185 | d->d_iwarns = 0; | |
1da177e4 LT |
186 | } |
187 | } else { | |
188 | if ((!d->d_ino_softlimit || | |
d0a3fe67 | 189 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 190 | be64_to_cpu(d->d_ino_softlimit))) && |
1da177e4 | 191 | (!d->d_ino_hardlimit || |
d0a3fe67 | 192 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 193 | be64_to_cpu(d->d_ino_hardlimit)))) { |
1da177e4 LT |
194 | d->d_itimer = 0; |
195 | } | |
196 | } | |
197 | ||
198 | if (!d->d_rtbtimer) { | |
1149d96a | 199 | if ((d->d_rtb_softlimit && |
d0a3fe67 | 200 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
201 | be64_to_cpu(d->d_rtb_softlimit))) || |
202 | (d->d_rtb_hardlimit && | |
d0a3fe67 | 203 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
204 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
205 | d->d_rtbtimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 206 | mp->m_quotainfo->qi_rtbtimelimit); |
754002b4 NS |
207 | } else { |
208 | d->d_rtbwarns = 0; | |
1da177e4 LT |
209 | } |
210 | } else { | |
211 | if ((!d->d_rtb_softlimit || | |
d0a3fe67 | 212 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 213 | be64_to_cpu(d->d_rtb_softlimit))) && |
1da177e4 | 214 | (!d->d_rtb_hardlimit || |
d0a3fe67 | 215 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 216 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
1da177e4 LT |
217 | d->d_rtbtimer = 0; |
218 | } | |
219 | } | |
220 | } | |
221 | ||
1da177e4 LT |
222 | /* |
223 | * initialize a buffer full of dquots and log the whole thing | |
224 | */ | |
225 | STATIC void | |
226 | xfs_qm_init_dquot_blk( | |
227 | xfs_trans_t *tp, | |
228 | xfs_mount_t *mp, | |
229 | xfs_dqid_t id, | |
230 | uint type, | |
231 | xfs_buf_t *bp) | |
232 | { | |
8a7b8a89 | 233 | struct xfs_quotainfo *q = mp->m_quotainfo; |
1da177e4 | 234 | xfs_dqblk_t *d; |
a484bcdd ES |
235 | xfs_dqid_t curid; |
236 | int i; | |
1da177e4 LT |
237 | |
238 | ASSERT(tp); | |
0c842ad4 | 239 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 240 | |
62926044 | 241 | d = bp->b_addr; |
1da177e4 LT |
242 | |
243 | /* | |
244 | * ID of the first dquot in the block - id's are zero based. | |
245 | */ | |
8a7b8a89 | 246 | curid = id - (id % q->qi_dqperchunk); |
8a7b8a89 | 247 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
49d35a5c CH |
248 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) { |
249 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
250 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
251 | d->dd_diskdq.d_id = cpu_to_be32(curid); | |
252 | d->dd_diskdq.d_flags = type; | |
6fcdc59d | 253 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
92863451 | 254 | uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid); |
6fcdc59d DC |
255 | xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk), |
256 | XFS_DQUOT_CRC_OFF); | |
257 | } | |
49d35a5c CH |
258 | } |
259 | ||
1da177e4 | 260 | xfs_trans_dquot_buf(tp, bp, |
c1155410 DC |
261 | (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : |
262 | ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : | |
263 | XFS_BLF_GDQUOT_BUF))); | |
8a7b8a89 | 264 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); |
1da177e4 LT |
265 | } |
266 | ||
b1366451 BF |
267 | /* |
268 | * Initialize the dynamic speculative preallocation thresholds. The lo/hi | |
269 | * watermarks correspond to the soft and hard limits by default. If a soft limit | |
270 | * is not specified, we use 95% of the hard limit. | |
271 | */ | |
272 | void | |
273 | xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp) | |
274 | { | |
275 | __uint64_t space; | |
276 | ||
277 | dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit); | |
278 | dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit); | |
279 | if (!dqp->q_prealloc_lo_wmark) { | |
280 | dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark; | |
281 | do_div(dqp->q_prealloc_lo_wmark, 100); | |
282 | dqp->q_prealloc_lo_wmark *= 95; | |
283 | } | |
284 | ||
285 | space = dqp->q_prealloc_hi_wmark; | |
286 | ||
287 | do_div(space, 100); | |
288 | dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space; | |
289 | dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3; | |
290 | dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5; | |
291 | } | |
292 | ||
1da177e4 LT |
293 | /* |
294 | * Allocate a block and fill it with dquots. | |
295 | * This is called when the bmapi finds a hole. | |
296 | */ | |
297 | STATIC int | |
298 | xfs_qm_dqalloc( | |
efa092f3 | 299 | xfs_trans_t **tpp, |
1da177e4 LT |
300 | xfs_mount_t *mp, |
301 | xfs_dquot_t *dqp, | |
302 | xfs_inode_t *quotip, | |
303 | xfs_fileoff_t offset_fsb, | |
304 | xfs_buf_t **O_bpp) | |
305 | { | |
306 | xfs_fsblock_t firstblock; | |
307 | xfs_bmap_free_t flist; | |
308 | xfs_bmbt_irec_t map; | |
f6106efa | 309 | int nmaps, error; |
1da177e4 | 310 | xfs_buf_t *bp; |
efa092f3 | 311 | xfs_trans_t *tp = *tpp; |
1da177e4 LT |
312 | |
313 | ASSERT(tp != NULL); | |
0b1b213f CH |
314 | |
315 | trace_xfs_dqalloc(dqp); | |
1da177e4 LT |
316 | |
317 | /* | |
318 | * Initialize the bmap freelist prior to calling bmapi code. | |
319 | */ | |
9d87c319 | 320 | xfs_bmap_init(&flist, &firstblock); |
1da177e4 LT |
321 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
322 | /* | |
323 | * Return if this type of quotas is turned off while we didn't | |
324 | * have an inode lock | |
325 | */ | |
6967b964 | 326 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 | 327 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
2451337d | 328 | return -ESRCH; |
1da177e4 LT |
329 | } |
330 | ||
ddc3415a | 331 | xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); |
1da177e4 | 332 | nmaps = 1; |
c0dc7828 DC |
333 | error = xfs_bmapi_write(tp, quotip, offset_fsb, |
334 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
335 | &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), | |
336 | &map, &nmaps, &flist); | |
337 | if (error) | |
1da177e4 | 338 | goto error0; |
1da177e4 LT |
339 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
340 | ASSERT(nmaps == 1); | |
341 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
342 | (map.br_startblock != HOLESTARTBLOCK)); | |
343 | ||
344 | /* | |
345 | * Keep track of the blkno to save a lookup later | |
346 | */ | |
347 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
348 | ||
349 | /* now we can just get the buffer (there's nothing to read yet) */ | |
350 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
351 | dqp->q_blkno, | |
8a7b8a89 | 352 | mp->m_quotainfo->qi_dqchunklen, |
1da177e4 | 353 | 0); |
36de9556 | 354 | if (!bp) { |
2451337d | 355 | error = -ENOMEM; |
1da177e4 | 356 | goto error1; |
36de9556 | 357 | } |
1813dd64 | 358 | bp->b_ops = &xfs_dquot_buf_ops; |
2a30f36d | 359 | |
1da177e4 LT |
360 | /* |
361 | * Make a chunk of dquots out of this buffer and log | |
362 | * the entire thing. | |
363 | */ | |
1149d96a | 364 | xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 365 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
1da177e4 | 366 | |
efa092f3 TS |
367 | /* |
368 | * xfs_bmap_finish() may commit the current transaction and | |
369 | * start a second transaction if the freelist is not empty. | |
370 | * | |
371 | * Since we still want to modify this buffer, we need to | |
372 | * ensure that the buffer is not released on commit of | |
373 | * the first transaction and ensure the buffer is added to the | |
374 | * second transaction. | |
375 | * | |
376 | * If there is only one transaction then don't stop the buffer | |
377 | * from being released when it commits later on. | |
378 | */ | |
379 | ||
380 | xfs_trans_bhold(tp, bp); | |
381 | ||
f6106efa ES |
382 | error = xfs_bmap_finish(tpp, &flist, NULL); |
383 | if (error) | |
1da177e4 | 384 | goto error1; |
1da177e4 | 385 | |
f6106efa ES |
386 | /* Transaction was committed? */ |
387 | if (*tpp != tp) { | |
efa092f3 TS |
388 | tp = *tpp; |
389 | xfs_trans_bjoin(tp, bp); | |
390 | } else { | |
391 | xfs_trans_bhold_release(tp, bp); | |
392 | } | |
393 | ||
1da177e4 LT |
394 | *O_bpp = bp; |
395 | return 0; | |
396 | ||
f6106efa | 397 | error1: |
1da177e4 | 398 | xfs_bmap_cancel(&flist); |
f6106efa | 399 | error0: |
1da177e4 LT |
400 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
401 | ||
d99831ff | 402 | return error; |
1da177e4 | 403 | } |
9aede1d8 | 404 | |
c6319198 DC |
405 | STATIC int |
406 | xfs_qm_dqrepair( | |
407 | struct xfs_mount *mp, | |
408 | struct xfs_trans *tp, | |
409 | struct xfs_dquot *dqp, | |
410 | xfs_dqid_t firstid, | |
411 | struct xfs_buf **bpp) | |
412 | { | |
413 | int error; | |
414 | struct xfs_disk_dquot *ddq; | |
415 | struct xfs_dqblk *d; | |
416 | int i; | |
417 | ||
418 | /* | |
419 | * Read the buffer without verification so we get the corrupted | |
612cfbfe | 420 | * buffer returned to us. make sure we verify it on write, though. |
c6319198 DC |
421 | */ |
422 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno, | |
423 | mp->m_quotainfo->qi_dqchunklen, | |
424 | 0, bpp, NULL); | |
425 | ||
426 | if (error) { | |
427 | ASSERT(*bpp == NULL); | |
b474c7ae | 428 | return error; |
c6319198 | 429 | } |
1813dd64 | 430 | (*bpp)->b_ops = &xfs_dquot_buf_ops; |
c6319198 DC |
431 | |
432 | ASSERT(xfs_buf_islocked(*bpp)); | |
433 | d = (struct xfs_dqblk *)(*bpp)->b_addr; | |
434 | ||
435 | /* Do the actual repair of dquots in this buffer */ | |
436 | for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) { | |
437 | ddq = &d[i].dd_diskdq; | |
9aede1d8 | 438 | error = xfs_dqcheck(mp, ddq, firstid + i, |
c6319198 DC |
439 | dqp->dq_flags & XFS_DQ_ALLTYPES, |
440 | XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair"); | |
441 | if (error) { | |
442 | /* repair failed, we're screwed */ | |
443 | xfs_trans_brelse(tp, *bpp); | |
2451337d | 444 | return -EIO; |
c6319198 DC |
445 | } |
446 | } | |
447 | ||
448 | return 0; | |
449 | } | |
450 | ||
1da177e4 LT |
451 | /* |
452 | * Maps a dquot to the buffer containing its on-disk version. | |
453 | * This returns a ptr to the buffer containing the on-disk dquot | |
454 | * in the bpp param, and a ptr to the on-disk dquot within that buffer | |
455 | */ | |
456 | STATIC int | |
457 | xfs_qm_dqtobp( | |
efa092f3 | 458 | xfs_trans_t **tpp, |
1da177e4 LT |
459 | xfs_dquot_t *dqp, |
460 | xfs_disk_dquot_t **O_ddpp, | |
461 | xfs_buf_t **O_bpp, | |
462 | uint flags) | |
463 | { | |
113a5683 CS |
464 | struct xfs_bmbt_irec map; |
465 | int nmaps = 1, error; | |
466 | struct xfs_buf *bp; | |
4d4d9523 | 467 | struct xfs_inode *quotip; |
113a5683 CS |
468 | struct xfs_mount *mp = dqp->q_mount; |
469 | xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); | |
470 | struct xfs_trans *tp = (tpp ? *tpp : NULL); | |
f4df8adc | 471 | uint lock_mode; |
1da177e4 | 472 | |
4d4d9523 | 473 | quotip = xfs_quota_inode(dqp->q_mount, dqp->dq_flags); |
acecf1b5 | 474 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
1da177e4 | 475 | |
f4df8adc | 476 | lock_mode = xfs_ilock_data_map_shared(quotip); |
6967b964 | 477 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 | 478 | /* |
acecf1b5 CH |
479 | * Return if this type of quotas is turned off while we |
480 | * didn't have the quota inode lock. | |
1da177e4 | 481 | */ |
f4df8adc | 482 | xfs_iunlock(quotip, lock_mode); |
2451337d | 483 | return -ESRCH; |
acecf1b5 CH |
484 | } |
485 | ||
486 | /* | |
487 | * Find the block map; no allocations yet | |
488 | */ | |
5c8ed202 DC |
489 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
490 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); | |
acecf1b5 | 491 | |
f4df8adc | 492 | xfs_iunlock(quotip, lock_mode); |
acecf1b5 CH |
493 | if (error) |
494 | return error; | |
495 | ||
496 | ASSERT(nmaps == 1); | |
497 | ASSERT(map.br_blockcount == 1); | |
498 | ||
499 | /* | |
500 | * Offset of dquot in the (fixed sized) dquot chunk. | |
501 | */ | |
502 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
503 | sizeof(xfs_dqblk_t); | |
504 | ||
505 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
506 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 507 | /* |
acecf1b5 | 508 | * We don't allocate unless we're asked to |
1da177e4 | 509 | */ |
acecf1b5 | 510 | if (!(flags & XFS_QMOPT_DQALLOC)) |
2451337d | 511 | return -ENOENT; |
1da177e4 | 512 | |
acecf1b5 CH |
513 | ASSERT(tp); |
514 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
515 | dqp->q_fileoffset, &bp); | |
1da177e4 | 516 | if (error) |
acecf1b5 CH |
517 | return error; |
518 | tp = *tpp; | |
519 | } else { | |
520 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
521 | |
522 | /* | |
acecf1b5 CH |
523 | * store the blkno etc so that we don't have to do the |
524 | * mapping all the time | |
1da177e4 | 525 | */ |
acecf1b5 | 526 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 527 | |
8a7b8a89 CH |
528 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
529 | dqp->q_blkno, | |
530 | mp->m_quotainfo->qi_dqchunklen, | |
1813dd64 | 531 | 0, &bp, &xfs_dquot_buf_ops); |
acecf1b5 | 532 | |
2451337d | 533 | if (error == -EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) { |
c6319198 DC |
534 | xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff * |
535 | mp->m_quotainfo->qi_dqperchunk; | |
536 | ASSERT(bp == NULL); | |
537 | error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp); | |
538 | } | |
1da177e4 | 539 | |
c6319198 DC |
540 | if (error) { |
541 | ASSERT(bp == NULL); | |
b474c7ae | 542 | return error; |
1da177e4 | 543 | } |
1da177e4 LT |
544 | } |
545 | ||
c6319198 | 546 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 547 | *O_bpp = bp; |
c6319198 | 548 | *O_ddpp = bp->b_addr + dqp->q_bufoffset; |
1da177e4 | 549 | |
d99831ff | 550 | return 0; |
1da177e4 LT |
551 | } |
552 | ||
553 | ||
554 | /* | |
555 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
556 | * and release the buffer immediately. | |
557 | * | |
97e7ade5 | 558 | * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed. |
1da177e4 | 559 | */ |
7ae44407 | 560 | int |
1da177e4 | 561 | xfs_qm_dqread( |
97e7ade5 CH |
562 | struct xfs_mount *mp, |
563 | xfs_dqid_t id, | |
564 | uint type, | |
565 | uint flags, | |
566 | struct xfs_dquot **O_dqpp) | |
1da177e4 | 567 | { |
97e7ade5 CH |
568 | struct xfs_dquot *dqp; |
569 | struct xfs_disk_dquot *ddqp; | |
570 | struct xfs_buf *bp; | |
571 | struct xfs_trans *tp = NULL; | |
572 | int error; | |
92b2e5b3 | 573 | |
a05931ce | 574 | dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP); |
92b2e5b3 CH |
575 | |
576 | dqp->dq_flags = type; | |
577 | dqp->q_core.d_id = cpu_to_be32(id); | |
578 | dqp->q_mount = mp; | |
f8739c3c | 579 | INIT_LIST_HEAD(&dqp->q_lru); |
92b2e5b3 CH |
580 | mutex_init(&dqp->q_qlock); |
581 | init_waitqueue_head(&dqp->q_pinwait); | |
582 | ||
583 | /* | |
584 | * Because we want to use a counting completion, complete | |
585 | * the flush completion once to allow a single access to | |
586 | * the flush completion without blocking. | |
587 | */ | |
588 | init_completion(&dqp->q_flush); | |
589 | complete(&dqp->q_flush); | |
590 | ||
591 | /* | |
592 | * Make sure group quotas have a different lock class than user | |
593 | * quotas. | |
594 | */ | |
f112a049 DC |
595 | switch (type) { |
596 | case XFS_DQ_USER: | |
597 | /* uses the default lock class */ | |
598 | break; | |
599 | case XFS_DQ_GROUP: | |
600 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class); | |
601 | break; | |
602 | case XFS_DQ_PROJ: | |
603 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class); | |
604 | break; | |
605 | default: | |
606 | ASSERT(0); | |
607 | break; | |
608 | } | |
92b2e5b3 | 609 | |
ff6d6af2 | 610 | XFS_STATS_INC(mp, xs_qm_dquot); |
1da177e4 | 611 | |
0b1b213f CH |
612 | trace_xfs_dqread(dqp); |
613 | ||
97e7ade5 CH |
614 | if (flags & XFS_QMOPT_DQALLOC) { |
615 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC); | |
410b11a6 | 616 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_dqalloc, |
3d3c8b52 | 617 | XFS_QM_DQALLOC_SPACE_RES(mp), 0); |
97e7ade5 CH |
618 | if (error) |
619 | goto error1; | |
97e7ade5 CH |
620 | } |
621 | ||
1da177e4 LT |
622 | /* |
623 | * get a pointer to the on-disk dquot and the buffer containing it | |
624 | * dqp already knows its own type (GROUP/USER). | |
625 | */ | |
97e7ade5 CH |
626 | error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags); |
627 | if (error) { | |
628 | /* | |
629 | * This can happen if quotas got turned off (ESRCH), | |
630 | * or if the dquot didn't exist on disk and we ask to | |
631 | * allocate (ENOENT). | |
632 | */ | |
633 | trace_xfs_dqread_fail(dqp); | |
97e7ade5 | 634 | goto error1; |
1da177e4 LT |
635 | } |
636 | ||
637 | /* copy everything from disk dquot to the incore dquot */ | |
638 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
1da177e4 LT |
639 | xfs_qm_dquot_logitem_init(dqp); |
640 | ||
641 | /* | |
642 | * Reservation counters are defined as reservation plus current usage | |
25985edc | 643 | * to avoid having to add every time. |
1da177e4 | 644 | */ |
1149d96a CH |
645 | dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); |
646 | dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); | |
647 | dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); | |
1da177e4 | 648 | |
b1366451 BF |
649 | /* initialize the dquot speculative prealloc thresholds */ |
650 | xfs_dquot_set_prealloc_limits(dqp); | |
651 | ||
1da177e4 | 652 | /* Mark the buf so that this will stay incore a little longer */ |
38f23232 | 653 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 LT |
654 | |
655 | /* | |
656 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
657 | * So we need to release with xfs_trans_brelse(). | |
658 | * The strategy here is identical to that of inodes; we lock | |
659 | * the dquot in xfs_qm_dqget() before making it accessible to | |
660 | * others. This is because dquots, like inodes, need a good level of | |
661 | * concurrency, and we don't want to take locks on the entire buffers | |
662 | * for dquot accesses. | |
663 | * Note also that the dquot buffer may even be dirty at this point, if | |
664 | * this particular dquot was repaired. We still aren't afraid to | |
665 | * brelse it because we have the changes incore. | |
666 | */ | |
0c842ad4 | 667 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
668 | xfs_trans_brelse(tp, bp); |
669 | ||
1da177e4 | 670 | if (tp) { |
70393313 | 671 | error = xfs_trans_commit(tp); |
97e7ade5 CH |
672 | if (error) |
673 | goto error0; | |
1da177e4 LT |
674 | } |
675 | ||
676 | *O_dqpp = dqp; | |
97e7ade5 | 677 | return error; |
1da177e4 | 678 | |
97e7ade5 | 679 | error1: |
1da177e4 | 680 | if (tp) |
4906e215 | 681 | xfs_trans_cancel(tp); |
97e7ade5 | 682 | error0: |
1da177e4 LT |
683 | xfs_qm_dqdestroy(dqp); |
684 | *O_dqpp = NULL; | |
97e7ade5 | 685 | return error; |
1da177e4 LT |
686 | } |
687 | ||
296c24e2 ES |
688 | /* |
689 | * Advance to the next id in the current chunk, or if at the | |
690 | * end of the chunk, skip ahead to first id in next allocated chunk | |
691 | * using the SEEK_DATA interface. | |
692 | */ | |
693 | int | |
694 | xfs_dq_get_next_id( | |
695 | xfs_mount_t *mp, | |
696 | uint type, | |
697 | xfs_dqid_t *id, | |
698 | loff_t eof) | |
699 | { | |
700 | struct xfs_inode *quotip; | |
701 | xfs_fsblock_t start; | |
702 | loff_t offset; | |
703 | uint lock; | |
704 | xfs_dqid_t next_id; | |
705 | int error = 0; | |
706 | ||
707 | /* Simple advance */ | |
708 | next_id = *id + 1; | |
709 | ||
710 | /* If new ID is within the current chunk, advancing it sufficed */ | |
711 | if (next_id % mp->m_quotainfo->qi_dqperchunk) { | |
712 | *id = next_id; | |
713 | return 0; | |
714 | } | |
715 | ||
716 | /* Nope, next_id is now past the current chunk, so find the next one */ | |
717 | start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk; | |
718 | ||
719 | quotip = xfs_quota_inode(mp, type); | |
720 | lock = xfs_ilock_data_map_shared(quotip); | |
721 | ||
722 | offset = __xfs_seek_hole_data(VFS_I(quotip), XFS_FSB_TO_B(mp, start), | |
723 | eof, SEEK_DATA); | |
724 | if (offset < 0) | |
725 | error = offset; | |
726 | ||
727 | xfs_iunlock(quotip, lock); | |
728 | ||
729 | /* -ENXIO is essentially "no more data" */ | |
730 | if (error) | |
731 | return (error == -ENXIO ? -ENOENT: error); | |
732 | ||
733 | /* Convert next data offset back to a quota id */ | |
734 | *id = XFS_B_TO_FSB(mp, offset) * mp->m_quotainfo->qi_dqperchunk; | |
735 | return 0; | |
736 | } | |
737 | ||
1da177e4 LT |
738 | /* |
739 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
740 | * a locked dquot, doing an allocation (if requested) as needed. | |
741 | * When both an inode and an id are given, the inode's id takes precedence. | |
742 | * That is, if the id changes while we don't hold the ilock inside this | |
743 | * function, the new dquot is returned, not necessarily the one requested | |
744 | * in the id argument. | |
745 | */ | |
746 | int | |
747 | xfs_qm_dqget( | |
748 | xfs_mount_t *mp, | |
749 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
750 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
751 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
752 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
753 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
754 | { | |
9f920f11 | 755 | struct xfs_quotainfo *qi = mp->m_quotainfo; |
329e0875 | 756 | struct radix_tree_root *tree = xfs_dquot_tree(qi, type); |
9f920f11 | 757 | struct xfs_dquot *dqp; |
296c24e2 | 758 | loff_t eof = 0; |
9f920f11 | 759 | int error; |
1da177e4 LT |
760 | |
761 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
762 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 763 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 | 764 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
2451337d | 765 | return -ESRCH; |
1da177e4 | 766 | } |
1da177e4 LT |
767 | |
768 | #ifdef DEBUG | |
769 | if (xfs_do_dqerror) { | |
770 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
771 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 772 | xfs_debug(mp, "Returning error in dqget"); |
2451337d | 773 | return -EIO; |
1da177e4 LT |
774 | } |
775 | } | |
1da177e4 | 776 | |
c8ad20ff NS |
777 | ASSERT(type == XFS_DQ_USER || |
778 | type == XFS_DQ_PROJ || | |
779 | type == XFS_DQ_GROUP); | |
1da177e4 | 780 | if (ip) { |
579aa9ca | 781 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
36731410 | 782 | ASSERT(xfs_inode_dquot(ip, type) == NULL); |
1da177e4 LT |
783 | } |
784 | #endif | |
92678554 | 785 | |
296c24e2 ES |
786 | /* Get the end of the quota file if we need it */ |
787 | if (flags & XFS_QMOPT_DQNEXT) { | |
788 | struct xfs_inode *quotip; | |
789 | xfs_fileoff_t last; | |
790 | uint lock_mode; | |
791 | ||
792 | quotip = xfs_quota_inode(mp, type); | |
793 | lock_mode = xfs_ilock_data_map_shared(quotip); | |
794 | error = xfs_bmap_last_offset(quotip, &last, XFS_DATA_FORK); | |
795 | xfs_iunlock(quotip, lock_mode); | |
796 | if (error) | |
797 | return error; | |
798 | eof = XFS_FSB_TO_B(mp, last); | |
799 | } | |
800 | ||
92678554 | 801 | restart: |
9f920f11 CH |
802 | mutex_lock(&qi->qi_tree_lock); |
803 | dqp = radix_tree_lookup(tree, id); | |
804 | if (dqp) { | |
805 | xfs_dqlock(dqp); | |
806 | if (dqp->dq_flags & XFS_DQ_FREEING) { | |
807 | xfs_dqunlock(dqp); | |
808 | mutex_unlock(&qi->qi_tree_lock); | |
809 | trace_xfs_dqget_freeing(dqp); | |
810 | delay(1); | |
811 | goto restart; | |
812 | } | |
1da177e4 | 813 | |
296c24e2 ES |
814 | /* uninit / unused quota found in radix tree, keep looking */ |
815 | if (flags & XFS_QMOPT_DQNEXT) { | |
816 | if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) { | |
817 | xfs_dqunlock(dqp); | |
818 | mutex_unlock(&qi->qi_tree_lock); | |
819 | error = xfs_dq_get_next_id(mp, type, &id, eof); | |
820 | if (error) | |
821 | return error; | |
822 | goto restart; | |
823 | } | |
824 | } | |
825 | ||
9f920f11 CH |
826 | dqp->q_nrefs++; |
827 | mutex_unlock(&qi->qi_tree_lock); | |
828 | ||
829 | trace_xfs_dqget_hit(dqp); | |
ff6d6af2 | 830 | XFS_STATS_INC(mp, xs_qm_dqcachehits); |
9f920f11 CH |
831 | *O_dqpp = dqp; |
832 | return 0; | |
1da177e4 | 833 | } |
9f920f11 | 834 | mutex_unlock(&qi->qi_tree_lock); |
ff6d6af2 | 835 | XFS_STATS_INC(mp, xs_qm_dqcachemisses); |
1da177e4 LT |
836 | |
837 | /* | |
838 | * Dquot cache miss. We don't want to keep the inode lock across | |
839 | * a (potential) disk read. Also we don't want to deal with the lock | |
840 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
841 | * lock here means dealing with a chown that can happen before | |
842 | * we re-acquire the lock. | |
843 | */ | |
844 | if (ip) | |
845 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 | 846 | |
97e7ade5 | 847 | error = xfs_qm_dqread(mp, id, type, flags, &dqp); |
1da177e4 | 848 | |
7ae44407 CH |
849 | if (ip) |
850 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
851 | ||
296c24e2 ES |
852 | /* If we are asked to find next active id, keep looking */ |
853 | if (error == -ENOENT && (flags & XFS_QMOPT_DQNEXT)) { | |
854 | error = xfs_dq_get_next_id(mp, type, &id, eof); | |
855 | if (!error) | |
856 | goto restart; | |
857 | } | |
858 | ||
7ae44407 CH |
859 | if (error) |
860 | return error; | |
1da177e4 | 861 | |
1da177e4 | 862 | if (ip) { |
1da177e4 LT |
863 | /* |
864 | * A dquot could be attached to this inode by now, since | |
865 | * we had dropped the ilock. | |
866 | */ | |
36731410 | 867 | if (xfs_this_quota_on(mp, type)) { |
9f920f11 CH |
868 | struct xfs_dquot *dqp1; |
869 | ||
36731410 CS |
870 | dqp1 = xfs_inode_dquot(ip, type); |
871 | if (dqp1) { | |
1da177e4 | 872 | xfs_qm_dqdestroy(dqp); |
36731410 | 873 | dqp = dqp1; |
1da177e4 LT |
874 | xfs_dqlock(dqp); |
875 | goto dqret; | |
876 | } | |
877 | } else { | |
36731410 CS |
878 | /* inode stays locked on return */ |
879 | xfs_qm_dqdestroy(dqp); | |
2451337d | 880 | return -ESRCH; |
1da177e4 LT |
881 | } |
882 | } | |
883 | ||
9f920f11 | 884 | mutex_lock(&qi->qi_tree_lock); |
2451337d | 885 | error = radix_tree_insert(tree, id, dqp); |
9f920f11 | 886 | if (unlikely(error)) { |
2451337d | 887 | WARN_ON(error != -EEXIST); |
9f920f11 | 888 | |
1da177e4 | 889 | /* |
9f920f11 CH |
890 | * Duplicate found. Just throw away the new dquot and start |
891 | * over. | |
1da177e4 | 892 | */ |
9f920f11 CH |
893 | mutex_unlock(&qi->qi_tree_lock); |
894 | trace_xfs_dqget_dup(dqp); | |
895 | xfs_qm_dqdestroy(dqp); | |
ff6d6af2 | 896 | XFS_STATS_INC(mp, xs_qm_dquot_dups); |
9f920f11 | 897 | goto restart; |
1da177e4 LT |
898 | } |
899 | ||
1da177e4 LT |
900 | /* |
901 | * We return a locked dquot to the caller, with a reference taken | |
902 | */ | |
903 | xfs_dqlock(dqp); | |
904 | dqp->q_nrefs = 1; | |
905 | ||
9f920f11 | 906 | qi->qi_dquots++; |
9f920f11 CH |
907 | mutex_unlock(&qi->qi_tree_lock); |
908 | ||
296c24e2 ES |
909 | /* If we are asked to find next active id, keep looking */ |
910 | if (flags & XFS_QMOPT_DQNEXT) { | |
911 | if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) { | |
912 | xfs_qm_dqput(dqp); | |
913 | error = xfs_dq_get_next_id(mp, type, &id, eof); | |
914 | if (error) | |
915 | return error; | |
916 | goto restart; | |
917 | } | |
918 | } | |
919 | ||
1da177e4 | 920 | dqret: |
579aa9ca | 921 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 922 | trace_xfs_dqget_miss(dqp); |
1da177e4 | 923 | *O_dqpp = dqp; |
d99831ff | 924 | return 0; |
1da177e4 LT |
925 | } |
926 | ||
f8739c3c CH |
927 | /* |
928 | * Release a reference to the dquot (decrement ref-count) and unlock it. | |
929 | * | |
930 | * If there is a group quota attached to this dquot, carefully release that | |
931 | * too without tripping over deadlocks'n'stuff. | |
932 | */ | |
933 | void | |
934 | xfs_qm_dqput( | |
935 | struct xfs_dquot *dqp) | |
936 | { | |
937 | ASSERT(dqp->q_nrefs > 0); | |
938 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
939 | ||
940 | trace_xfs_dqput(dqp); | |
941 | ||
3c353375 DC |
942 | if (--dqp->q_nrefs == 0) { |
943 | struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo; | |
944 | trace_xfs_dqput_free(dqp); | |
945 | ||
946 | if (list_lru_add(&qi->qi_lru, &dqp->q_lru)) | |
ff6d6af2 | 947 | XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused); |
3c353375 DC |
948 | } |
949 | xfs_dqunlock(dqp); | |
1da177e4 LT |
950 | } |
951 | ||
952 | /* | |
953 | * Release a dquot. Flush it if dirty, then dqput() it. | |
954 | * dquot must not be locked. | |
955 | */ | |
956 | void | |
957 | xfs_qm_dqrele( | |
958 | xfs_dquot_t *dqp) | |
959 | { | |
7d095257 CH |
960 | if (!dqp) |
961 | return; | |
962 | ||
0b1b213f | 963 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
964 | |
965 | xfs_dqlock(dqp); | |
966 | /* | |
967 | * We don't care to flush it if the dquot is dirty here. | |
968 | * That will create stutters that we want to avoid. | |
969 | * Instead we do a delayed write when we try to reclaim | |
970 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
971 | */ | |
972 | xfs_qm_dqput(dqp); | |
973 | } | |
974 | ||
ca30b2a7 CH |
975 | /* |
976 | * This is the dquot flushing I/O completion routine. It is called | |
977 | * from interrupt level when the buffer containing the dquot is | |
978 | * flushed to disk. It is responsible for removing the dquot logitem | |
979 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
980 | * flush lock. This behavior is very similar to that of inodes.. | |
981 | */ | |
982 | STATIC void | |
983 | xfs_qm_dqflush_done( | |
984 | struct xfs_buf *bp, | |
985 | struct xfs_log_item *lip) | |
986 | { | |
987 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
988 | xfs_dquot_t *dqp = qip->qli_dquot; | |
989 | struct xfs_ail *ailp = lip->li_ailp; | |
990 | ||
991 | /* | |
992 | * We only want to pull the item from the AIL if its | |
993 | * location in the log has not changed since we started the flush. | |
994 | * Thus, we only bother if the dquot's lsn has | |
995 | * not changed. First we check the lsn outside the lock | |
996 | * since it's cheaper, and then we recheck while | |
997 | * holding the lock before removing the dquot from the AIL. | |
998 | */ | |
999 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
1000 | lip->li_lsn == qip->qli_flush_lsn) { | |
1001 | ||
1002 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
1003 | spin_lock(&ailp->xa_lock); | |
1004 | if (lip->li_lsn == qip->qli_flush_lsn) | |
04913fdd | 1005 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 CH |
1006 | else |
1007 | spin_unlock(&ailp->xa_lock); | |
1008 | } | |
1009 | ||
1010 | /* | |
1011 | * Release the dq's flush lock since we're done with it. | |
1012 | */ | |
1013 | xfs_dqfunlock(dqp); | |
1014 | } | |
1da177e4 LT |
1015 | |
1016 | /* | |
1017 | * Write a modified dquot to disk. | |
1018 | * The dquot must be locked and the flush lock too taken by caller. | |
1019 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
1020 | * but the dquot is free to be unlocked and modified by the caller | |
1021 | * in the interim. Dquot is still locked on return. This behavior is | |
1022 | * identical to that of inodes. | |
1023 | */ | |
1024 | int | |
1025 | xfs_qm_dqflush( | |
fe7257fd CH |
1026 | struct xfs_dquot *dqp, |
1027 | struct xfs_buf **bpp) | |
1da177e4 | 1028 | { |
acecf1b5 CH |
1029 | struct xfs_mount *mp = dqp->q_mount; |
1030 | struct xfs_buf *bp; | |
1031 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 1032 | int error; |
1da177e4 LT |
1033 | |
1034 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 1035 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 1036 | |
0b1b213f | 1037 | trace_xfs_dqflush(dqp); |
1da177e4 | 1038 | |
fe7257fd CH |
1039 | *bpp = NULL; |
1040 | ||
1da177e4 LT |
1041 | xfs_qm_dqunpin_wait(dqp); |
1042 | ||
1043 | /* | |
1044 | * This may have been unpinned because the filesystem is shutting | |
1045 | * down forcibly. If that's the case we must not write this dquot | |
dea96095 CH |
1046 | * to disk, because the log record didn't make it to disk. |
1047 | * | |
1048 | * We also have to remove the log item from the AIL in this case, | |
1049 | * as we wait for an emptry AIL as part of the unmount process. | |
1da177e4 | 1050 | */ |
acecf1b5 | 1051 | if (XFS_FORCED_SHUTDOWN(mp)) { |
dea96095 | 1052 | struct xfs_log_item *lip = &dqp->q_logitem.qli_item; |
acecf1b5 | 1053 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
dea96095 | 1054 | |
146e54b7 BF |
1055 | xfs_trans_ail_remove(lip, SHUTDOWN_CORRUPT_INCORE); |
1056 | ||
2451337d | 1057 | error = -EIO; |
fe7257fd | 1058 | goto out_unlock; |
1da177e4 LT |
1059 | } |
1060 | ||
1061 | /* | |
1062 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1063 | */ |
acecf1b5 | 1064 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
5fd364fe DC |
1065 | mp->m_quotainfo->qi_dqchunklen, 0, &bp, |
1066 | &xfs_dquot_buf_ops); | |
fe7257fd CH |
1067 | if (error) |
1068 | goto out_unlock; | |
1da177e4 | 1069 | |
acecf1b5 CH |
1070 | /* |
1071 | * Calculate the location of the dquot inside the buffer. | |
1072 | */ | |
62926044 | 1073 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1074 | |
1075 | /* | |
1076 | * A simple sanity check in case we got a corrupted dquot.. | |
1077 | */ | |
9aede1d8 | 1078 | error = xfs_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
a0fa2b67 DC |
1079 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); |
1080 | if (error) { | |
acecf1b5 CH |
1081 | xfs_buf_relse(bp); |
1082 | xfs_dqfunlock(dqp); | |
1083 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
2451337d | 1084 | return -EIO; |
1da177e4 LT |
1085 | } |
1086 | ||
1087 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1088 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1089 | |
1090 | /* | |
1091 | * Clear the dirty field and remember the flush lsn for later use. | |
1092 | */ | |
acecf1b5 | 1093 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1094 | |
7b2e2a31 DC |
1095 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1096 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 | 1097 | |
3fe58f30 CH |
1098 | /* |
1099 | * copy the lsn into the on-disk dquot now while we have the in memory | |
1100 | * dquot here. This can't be done later in the write verifier as we | |
1101 | * can't get access to the log item at that point in time. | |
6fcdc59d DC |
1102 | * |
1103 | * We also calculate the CRC here so that the on-disk dquot in the | |
1104 | * buffer always has a valid CRC. This ensures there is no possibility | |
1105 | * of a dquot without an up-to-date CRC getting to disk. | |
3fe58f30 CH |
1106 | */ |
1107 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
1108 | struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp; | |
1109 | ||
1110 | dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn); | |
6fcdc59d DC |
1111 | xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk), |
1112 | XFS_DQUOT_CRC_OFF); | |
3fe58f30 CH |
1113 | } |
1114 | ||
1da177e4 LT |
1115 | /* |
1116 | * Attach an iodone routine so that we can remove this dquot from the | |
1117 | * AIL and release the flush lock once the dquot is synced to disk. | |
1118 | */ | |
ca30b2a7 CH |
1119 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1120 | &dqp->q_logitem.qli_item); | |
1121 | ||
1da177e4 LT |
1122 | /* |
1123 | * If the buffer is pinned then push on the log so we won't | |
1124 | * get stuck waiting in the write for too long. | |
1125 | */ | |
811e64c7 | 1126 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1127 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1128 | xfs_log_force(mp, 0); |
1da177e4 LT |
1129 | } |
1130 | ||
0b1b213f | 1131 | trace_xfs_dqflush_done(dqp); |
fe7257fd CH |
1132 | *bpp = bp; |
1133 | return 0; | |
0b1b213f | 1134 | |
fe7257fd CH |
1135 | out_unlock: |
1136 | xfs_dqfunlock(dqp); | |
2451337d | 1137 | return -EIO; |
1da177e4 LT |
1138 | } |
1139 | ||
5bb87a33 CH |
1140 | /* |
1141 | * Lock two xfs_dquot structures. | |
1142 | * | |
1143 | * To avoid deadlocks we always lock the quota structure with | |
1144 | * the lowerd id first. | |
1145 | */ | |
1da177e4 LT |
1146 | void |
1147 | xfs_dqlock2( | |
1148 | xfs_dquot_t *d1, | |
1149 | xfs_dquot_t *d2) | |
1150 | { | |
1151 | if (d1 && d2) { | |
1152 | ASSERT(d1 != d2); | |
1149d96a CH |
1153 | if (be32_to_cpu(d1->q_core.d_id) > |
1154 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1155 | mutex_lock(&d2->q_qlock); |
1156 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1157 | } else { |
5bb87a33 CH |
1158 | mutex_lock(&d1->q_qlock); |
1159 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1160 | } |
5bb87a33 CH |
1161 | } else if (d1) { |
1162 | mutex_lock(&d1->q_qlock); | |
1163 | } else if (d2) { | |
1164 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1165 | } |
1166 | } | |
1167 | ||
a05931ce CH |
1168 | int __init |
1169 | xfs_qm_init(void) | |
1170 | { | |
1171 | xfs_qm_dqzone = | |
1172 | kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot"); | |
1173 | if (!xfs_qm_dqzone) | |
1174 | goto out; | |
1175 | ||
1176 | xfs_qm_dqtrxzone = | |
1177 | kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx"); | |
1178 | if (!xfs_qm_dqtrxzone) | |
1179 | goto out_free_dqzone; | |
1180 | ||
1181 | return 0; | |
1182 | ||
1183 | out_free_dqzone: | |
1184 | kmem_zone_destroy(xfs_qm_dqzone); | |
1185 | out: | |
1186 | return -ENOMEM; | |
1187 | } | |
1188 | ||
1c2ccc66 | 1189 | void |
a05931ce CH |
1190 | xfs_qm_exit(void) |
1191 | { | |
1192 | kmem_zone_destroy(xfs_qm_dqtrxzone); | |
1193 | kmem_zone_destroy(xfs_qm_dqzone); | |
1194 | } |