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( | |
4b6eae2e BF |
88 | struct xfs_mount *mp, |
89 | struct xfs_dquot *dq) | |
1da177e4 | 90 | { |
4b6eae2e BF |
91 | struct xfs_quotainfo *q = mp->m_quotainfo; |
92 | struct xfs_disk_dquot *d = &dq->q_core; | |
b1366451 | 93 | int prealloc = 0; |
1da177e4 LT |
94 | |
95 | ASSERT(d->d_id); | |
96 | ||
b1366451 | 97 | if (q->qi_bsoftlimit && !d->d_blk_softlimit) { |
1149d96a | 98 | d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit); |
b1366451 BF |
99 | prealloc = 1; |
100 | } | |
101 | if (q->qi_bhardlimit && !d->d_blk_hardlimit) { | |
1149d96a | 102 | d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit); |
b1366451 BF |
103 | prealloc = 1; |
104 | } | |
1da177e4 | 105 | if (q->qi_isoftlimit && !d->d_ino_softlimit) |
1149d96a | 106 | d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit); |
1da177e4 | 107 | if (q->qi_ihardlimit && !d->d_ino_hardlimit) |
1149d96a | 108 | d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit); |
1da177e4 | 109 | if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit) |
1149d96a | 110 | d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit); |
1da177e4 | 111 | if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit) |
1149d96a | 112 | d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit); |
b1366451 BF |
113 | |
114 | if (prealloc) | |
115 | xfs_dquot_set_prealloc_limits(dq); | |
1da177e4 LT |
116 | } |
117 | ||
118 | /* | |
119 | * Check the limits and timers of a dquot and start or reset timers | |
120 | * if necessary. | |
121 | * This gets called even when quota enforcement is OFF, which makes our | |
122 | * life a little less complicated. (We just don't reject any quota | |
123 | * reservations in that case, when enforcement is off). | |
124 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
125 | * enforcement's off. | |
126 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
127 | * 'automatically' get started when limits get exceeded. They do |
128 | * get reset to zero, however, when we find the count to be under | |
129 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
130 | */ |
131 | void | |
132 | xfs_qm_adjust_dqtimers( | |
133 | xfs_mount_t *mp, | |
134 | xfs_disk_dquot_t *d) | |
135 | { | |
136 | ASSERT(d->d_id); | |
137 | ||
ea15ab3c | 138 | #ifdef DEBUG |
1149d96a CH |
139 | if (d->d_blk_hardlimit) |
140 | ASSERT(be64_to_cpu(d->d_blk_softlimit) <= | |
141 | be64_to_cpu(d->d_blk_hardlimit)); | |
142 | if (d->d_ino_hardlimit) | |
143 | ASSERT(be64_to_cpu(d->d_ino_softlimit) <= | |
144 | be64_to_cpu(d->d_ino_hardlimit)); | |
145 | if (d->d_rtb_hardlimit) | |
146 | ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= | |
147 | be64_to_cpu(d->d_rtb_hardlimit)); | |
1da177e4 | 148 | #endif |
ea15ab3c | 149 | |
1da177e4 | 150 | if (!d->d_btimer) { |
1149d96a | 151 | if ((d->d_blk_softlimit && |
d0a3fe67 | 152 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
153 | be64_to_cpu(d->d_blk_softlimit))) || |
154 | (d->d_blk_hardlimit && | |
d0a3fe67 | 155 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
156 | be64_to_cpu(d->d_blk_hardlimit)))) { |
157 | d->d_btimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 158 | mp->m_quotainfo->qi_btimelimit); |
754002b4 NS |
159 | } else { |
160 | d->d_bwarns = 0; | |
1da177e4 LT |
161 | } |
162 | } else { | |
163 | if ((!d->d_blk_softlimit || | |
d0a3fe67 | 164 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 165 | be64_to_cpu(d->d_blk_softlimit))) && |
1da177e4 | 166 | (!d->d_blk_hardlimit || |
d0a3fe67 | 167 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 168 | be64_to_cpu(d->d_blk_hardlimit)))) { |
1da177e4 LT |
169 | d->d_btimer = 0; |
170 | } | |
171 | } | |
172 | ||
173 | if (!d->d_itimer) { | |
1149d96a | 174 | if ((d->d_ino_softlimit && |
d0a3fe67 | 175 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
176 | be64_to_cpu(d->d_ino_softlimit))) || |
177 | (d->d_ino_hardlimit && | |
d0a3fe67 | 178 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
179 | be64_to_cpu(d->d_ino_hardlimit)))) { |
180 | d->d_itimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 181 | mp->m_quotainfo->qi_itimelimit); |
754002b4 NS |
182 | } else { |
183 | d->d_iwarns = 0; | |
1da177e4 LT |
184 | } |
185 | } else { | |
186 | if ((!d->d_ino_softlimit || | |
d0a3fe67 | 187 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 188 | be64_to_cpu(d->d_ino_softlimit))) && |
1da177e4 | 189 | (!d->d_ino_hardlimit || |
d0a3fe67 | 190 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 191 | be64_to_cpu(d->d_ino_hardlimit)))) { |
1da177e4 LT |
192 | d->d_itimer = 0; |
193 | } | |
194 | } | |
195 | ||
196 | if (!d->d_rtbtimer) { | |
1149d96a | 197 | if ((d->d_rtb_softlimit && |
d0a3fe67 | 198 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
199 | be64_to_cpu(d->d_rtb_softlimit))) || |
200 | (d->d_rtb_hardlimit && | |
d0a3fe67 | 201 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
202 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
203 | d->d_rtbtimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 204 | mp->m_quotainfo->qi_rtbtimelimit); |
754002b4 NS |
205 | } else { |
206 | d->d_rtbwarns = 0; | |
1da177e4 LT |
207 | } |
208 | } else { | |
209 | if ((!d->d_rtb_softlimit || | |
d0a3fe67 | 210 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 211 | be64_to_cpu(d->d_rtb_softlimit))) && |
1da177e4 | 212 | (!d->d_rtb_hardlimit || |
d0a3fe67 | 213 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 214 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
1da177e4 LT |
215 | d->d_rtbtimer = 0; |
216 | } | |
217 | } | |
218 | } | |
219 | ||
1da177e4 LT |
220 | /* |
221 | * initialize a buffer full of dquots and log the whole thing | |
222 | */ | |
223 | STATIC void | |
224 | xfs_qm_init_dquot_blk( | |
225 | xfs_trans_t *tp, | |
226 | xfs_mount_t *mp, | |
227 | xfs_dqid_t id, | |
228 | uint type, | |
229 | xfs_buf_t *bp) | |
230 | { | |
8a7b8a89 | 231 | struct xfs_quotainfo *q = mp->m_quotainfo; |
1da177e4 LT |
232 | xfs_dqblk_t *d; |
233 | int curid, i; | |
234 | ||
235 | ASSERT(tp); | |
0c842ad4 | 236 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 237 | |
62926044 | 238 | d = bp->b_addr; |
1da177e4 LT |
239 | |
240 | /* | |
241 | * ID of the first dquot in the block - id's are zero based. | |
242 | */ | |
8a7b8a89 | 243 | curid = id - (id % q->qi_dqperchunk); |
1da177e4 | 244 | ASSERT(curid >= 0); |
8a7b8a89 | 245 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
49d35a5c CH |
246 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) { |
247 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
248 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
249 | d->dd_diskdq.d_id = cpu_to_be32(curid); | |
250 | d->dd_diskdq.d_flags = type; | |
251 | } | |
252 | ||
1da177e4 | 253 | xfs_trans_dquot_buf(tp, bp, |
c1155410 DC |
254 | (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : |
255 | ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : | |
256 | XFS_BLF_GDQUOT_BUF))); | |
8a7b8a89 | 257 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); |
1da177e4 LT |
258 | } |
259 | ||
b1366451 BF |
260 | /* |
261 | * Initialize the dynamic speculative preallocation thresholds. The lo/hi | |
262 | * watermarks correspond to the soft and hard limits by default. If a soft limit | |
263 | * is not specified, we use 95% of the hard limit. | |
264 | */ | |
265 | void | |
266 | xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp) | |
267 | { | |
268 | __uint64_t space; | |
269 | ||
270 | dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit); | |
271 | dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit); | |
272 | if (!dqp->q_prealloc_lo_wmark) { | |
273 | dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark; | |
274 | do_div(dqp->q_prealloc_lo_wmark, 100); | |
275 | dqp->q_prealloc_lo_wmark *= 95; | |
276 | } | |
277 | ||
278 | space = dqp->q_prealloc_hi_wmark; | |
279 | ||
280 | do_div(space, 100); | |
281 | dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space; | |
282 | dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3; | |
283 | dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5; | |
284 | } | |
285 | ||
b0f539de DC |
286 | static void |
287 | xfs_dquot_buf_verify( | |
288 | struct xfs_buf *bp) | |
289 | { | |
290 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
291 | struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr; | |
292 | struct xfs_disk_dquot *ddq; | |
293 | xfs_dqid_t id = 0; | |
294 | int i; | |
295 | ||
296 | /* | |
297 | * On the first read of the buffer, verify that each dquot is valid. | |
298 | * We don't know what the id of the dquot is supposed to be, just that | |
299 | * they should be increasing monotonically within the buffer. If the | |
300 | * first id is corrupt, then it will fail on the second dquot in the | |
301 | * buffer so corruptions could point to the wrong dquot in this case. | |
302 | */ | |
303 | for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) { | |
304 | int error; | |
305 | ||
306 | ddq = &d[i].dd_diskdq; | |
307 | ||
308 | if (i == 0) | |
309 | id = be32_to_cpu(ddq->d_id); | |
310 | ||
311 | error = xfs_qm_dqcheck(mp, ddq, id + i, 0, XFS_QMOPT_DOWARN, | |
312 | "xfs_dquot_read_verify"); | |
313 | if (error) { | |
314 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, d); | |
315 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
316 | break; | |
317 | } | |
318 | } | |
319 | } | |
320 | ||
321 | static void | |
1813dd64 | 322 | xfs_dquot_buf_read_verify( |
b0f539de DC |
323 | struct xfs_buf *bp) |
324 | { | |
325 | xfs_dquot_buf_verify(bp); | |
326 | } | |
1da177e4 | 327 | |
b0f539de | 328 | void |
1813dd64 | 329 | xfs_dquot_buf_write_verify( |
b0f539de DC |
330 | struct xfs_buf *bp) |
331 | { | |
332 | xfs_dquot_buf_verify(bp); | |
b0f539de | 333 | } |
1da177e4 | 334 | |
1813dd64 DC |
335 | const struct xfs_buf_ops xfs_dquot_buf_ops = { |
336 | .verify_read = xfs_dquot_buf_read_verify, | |
337 | .verify_write = xfs_dquot_buf_write_verify, | |
338 | }; | |
339 | ||
1da177e4 LT |
340 | /* |
341 | * Allocate a block and fill it with dquots. | |
342 | * This is called when the bmapi finds a hole. | |
343 | */ | |
344 | STATIC int | |
345 | xfs_qm_dqalloc( | |
efa092f3 | 346 | xfs_trans_t **tpp, |
1da177e4 LT |
347 | xfs_mount_t *mp, |
348 | xfs_dquot_t *dqp, | |
349 | xfs_inode_t *quotip, | |
350 | xfs_fileoff_t offset_fsb, | |
351 | xfs_buf_t **O_bpp) | |
352 | { | |
353 | xfs_fsblock_t firstblock; | |
354 | xfs_bmap_free_t flist; | |
355 | xfs_bmbt_irec_t map; | |
356 | int nmaps, error, committed; | |
357 | xfs_buf_t *bp; | |
efa092f3 | 358 | xfs_trans_t *tp = *tpp; |
1da177e4 LT |
359 | |
360 | ASSERT(tp != NULL); | |
0b1b213f CH |
361 | |
362 | trace_xfs_dqalloc(dqp); | |
1da177e4 LT |
363 | |
364 | /* | |
365 | * Initialize the bmap freelist prior to calling bmapi code. | |
366 | */ | |
9d87c319 | 367 | xfs_bmap_init(&flist, &firstblock); |
1da177e4 LT |
368 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
369 | /* | |
370 | * Return if this type of quotas is turned off while we didn't | |
371 | * have an inode lock | |
372 | */ | |
6967b964 | 373 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 LT |
374 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
375 | return (ESRCH); | |
376 | } | |
377 | ||
ddc3415a | 378 | xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); |
1da177e4 | 379 | nmaps = 1; |
c0dc7828 DC |
380 | error = xfs_bmapi_write(tp, quotip, offset_fsb, |
381 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
382 | &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), | |
383 | &map, &nmaps, &flist); | |
384 | if (error) | |
1da177e4 | 385 | goto error0; |
1da177e4 LT |
386 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
387 | ASSERT(nmaps == 1); | |
388 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
389 | (map.br_startblock != HOLESTARTBLOCK)); | |
390 | ||
391 | /* | |
392 | * Keep track of the blkno to save a lookup later | |
393 | */ | |
394 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
395 | ||
396 | /* now we can just get the buffer (there's nothing to read yet) */ | |
397 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
398 | dqp->q_blkno, | |
8a7b8a89 | 399 | mp->m_quotainfo->qi_dqchunklen, |
1da177e4 | 400 | 0); |
2a30f36d CS |
401 | |
402 | error = xfs_buf_geterror(bp); | |
403 | if (error) | |
1da177e4 | 404 | goto error1; |
1813dd64 | 405 | bp->b_ops = &xfs_dquot_buf_ops; |
2a30f36d | 406 | |
1da177e4 LT |
407 | /* |
408 | * Make a chunk of dquots out of this buffer and log | |
409 | * the entire thing. | |
410 | */ | |
1149d96a | 411 | xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 412 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
1da177e4 | 413 | |
efa092f3 TS |
414 | /* |
415 | * xfs_bmap_finish() may commit the current transaction and | |
416 | * start a second transaction if the freelist is not empty. | |
417 | * | |
418 | * Since we still want to modify this buffer, we need to | |
419 | * ensure that the buffer is not released on commit of | |
420 | * the first transaction and ensure the buffer is added to the | |
421 | * second transaction. | |
422 | * | |
423 | * If there is only one transaction then don't stop the buffer | |
424 | * from being released when it commits later on. | |
425 | */ | |
426 | ||
427 | xfs_trans_bhold(tp, bp); | |
428 | ||
f7c99b6f | 429 | if ((error = xfs_bmap_finish(tpp, &flist, &committed))) { |
1da177e4 LT |
430 | goto error1; |
431 | } | |
432 | ||
efa092f3 TS |
433 | if (committed) { |
434 | tp = *tpp; | |
435 | xfs_trans_bjoin(tp, bp); | |
436 | } else { | |
437 | xfs_trans_bhold_release(tp, bp); | |
438 | } | |
439 | ||
1da177e4 LT |
440 | *O_bpp = bp; |
441 | return 0; | |
442 | ||
443 | error1: | |
444 | xfs_bmap_cancel(&flist); | |
445 | error0: | |
446 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
447 | ||
448 | return (error); | |
449 | } | |
c6319198 DC |
450 | STATIC int |
451 | xfs_qm_dqrepair( | |
452 | struct xfs_mount *mp, | |
453 | struct xfs_trans *tp, | |
454 | struct xfs_dquot *dqp, | |
455 | xfs_dqid_t firstid, | |
456 | struct xfs_buf **bpp) | |
457 | { | |
458 | int error; | |
459 | struct xfs_disk_dquot *ddq; | |
460 | struct xfs_dqblk *d; | |
461 | int i; | |
462 | ||
463 | /* | |
464 | * Read the buffer without verification so we get the corrupted | |
612cfbfe | 465 | * buffer returned to us. make sure we verify it on write, though. |
c6319198 DC |
466 | */ |
467 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno, | |
468 | mp->m_quotainfo->qi_dqchunklen, | |
469 | 0, bpp, NULL); | |
470 | ||
471 | if (error) { | |
472 | ASSERT(*bpp == NULL); | |
473 | return XFS_ERROR(error); | |
474 | } | |
1813dd64 | 475 | (*bpp)->b_ops = &xfs_dquot_buf_ops; |
c6319198 DC |
476 | |
477 | ASSERT(xfs_buf_islocked(*bpp)); | |
478 | d = (struct xfs_dqblk *)(*bpp)->b_addr; | |
479 | ||
480 | /* Do the actual repair of dquots in this buffer */ | |
481 | for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) { | |
482 | ddq = &d[i].dd_diskdq; | |
483 | error = xfs_qm_dqcheck(mp, ddq, firstid + i, | |
484 | dqp->dq_flags & XFS_DQ_ALLTYPES, | |
485 | XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair"); | |
486 | if (error) { | |
487 | /* repair failed, we're screwed */ | |
488 | xfs_trans_brelse(tp, *bpp); | |
489 | return XFS_ERROR(EIO); | |
490 | } | |
491 | } | |
492 | ||
493 | return 0; | |
494 | } | |
495 | ||
1da177e4 LT |
496 | /* |
497 | * Maps a dquot to the buffer containing its on-disk version. | |
498 | * This returns a ptr to the buffer containing the on-disk dquot | |
499 | * in the bpp param, and a ptr to the on-disk dquot within that buffer | |
500 | */ | |
501 | STATIC int | |
502 | xfs_qm_dqtobp( | |
efa092f3 | 503 | xfs_trans_t **tpp, |
1da177e4 LT |
504 | xfs_dquot_t *dqp, |
505 | xfs_disk_dquot_t **O_ddpp, | |
506 | xfs_buf_t **O_bpp, | |
507 | uint flags) | |
508 | { | |
509 | xfs_bmbt_irec_t map; | |
acecf1b5 | 510 | int nmaps = 1, error; |
1da177e4 | 511 | xfs_buf_t *bp; |
acecf1b5 CH |
512 | xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp); |
513 | xfs_mount_t *mp = dqp->q_mount; | |
acecf1b5 | 514 | xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); |
efa092f3 | 515 | xfs_trans_t *tp = (tpp ? *tpp : NULL); |
1da177e4 | 516 | |
acecf1b5 | 517 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
1da177e4 | 518 | |
acecf1b5 | 519 | xfs_ilock(quotip, XFS_ILOCK_SHARED); |
6967b964 | 520 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 | 521 | /* |
acecf1b5 CH |
522 | * Return if this type of quotas is turned off while we |
523 | * didn't have the quota inode lock. | |
1da177e4 | 524 | */ |
acecf1b5 CH |
525 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); |
526 | return ESRCH; | |
527 | } | |
528 | ||
529 | /* | |
530 | * Find the block map; no allocations yet | |
531 | */ | |
5c8ed202 DC |
532 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
533 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); | |
acecf1b5 CH |
534 | |
535 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); | |
536 | if (error) | |
537 | return error; | |
538 | ||
539 | ASSERT(nmaps == 1); | |
540 | ASSERT(map.br_blockcount == 1); | |
541 | ||
542 | /* | |
543 | * Offset of dquot in the (fixed sized) dquot chunk. | |
544 | */ | |
545 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
546 | sizeof(xfs_dqblk_t); | |
547 | ||
548 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
549 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 550 | /* |
acecf1b5 | 551 | * We don't allocate unless we're asked to |
1da177e4 | 552 | */ |
acecf1b5 CH |
553 | if (!(flags & XFS_QMOPT_DQALLOC)) |
554 | return ENOENT; | |
1da177e4 | 555 | |
acecf1b5 CH |
556 | ASSERT(tp); |
557 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
558 | dqp->q_fileoffset, &bp); | |
1da177e4 | 559 | if (error) |
acecf1b5 CH |
560 | return error; |
561 | tp = *tpp; | |
562 | } else { | |
563 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
564 | |
565 | /* | |
acecf1b5 CH |
566 | * store the blkno etc so that we don't have to do the |
567 | * mapping all the time | |
1da177e4 | 568 | */ |
acecf1b5 | 569 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 570 | |
8a7b8a89 CH |
571 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
572 | dqp->q_blkno, | |
573 | mp->m_quotainfo->qi_dqchunklen, | |
1813dd64 | 574 | 0, &bp, &xfs_dquot_buf_ops); |
acecf1b5 | 575 | |
c6319198 DC |
576 | if (error == EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) { |
577 | xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff * | |
578 | mp->m_quotainfo->qi_dqperchunk; | |
579 | ASSERT(bp == NULL); | |
580 | error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp); | |
581 | } | |
1da177e4 | 582 | |
c6319198 DC |
583 | if (error) { |
584 | ASSERT(bp == NULL); | |
585 | return XFS_ERROR(error); | |
1da177e4 | 586 | } |
1da177e4 LT |
587 | } |
588 | ||
c6319198 | 589 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 590 | *O_bpp = bp; |
c6319198 | 591 | *O_ddpp = bp->b_addr + dqp->q_bufoffset; |
1da177e4 LT |
592 | |
593 | return (0); | |
594 | } | |
595 | ||
596 | ||
597 | /* | |
598 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
599 | * and release the buffer immediately. | |
600 | * | |
97e7ade5 | 601 | * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed. |
1da177e4 | 602 | */ |
7ae44407 | 603 | int |
1da177e4 | 604 | xfs_qm_dqread( |
97e7ade5 CH |
605 | struct xfs_mount *mp, |
606 | xfs_dqid_t id, | |
607 | uint type, | |
608 | uint flags, | |
609 | struct xfs_dquot **O_dqpp) | |
1da177e4 | 610 | { |
97e7ade5 CH |
611 | struct xfs_dquot *dqp; |
612 | struct xfs_disk_dquot *ddqp; | |
613 | struct xfs_buf *bp; | |
614 | struct xfs_trans *tp = NULL; | |
615 | int error; | |
616 | int cancelflags = 0; | |
efa092f3 | 617 | |
92b2e5b3 | 618 | |
a05931ce | 619 | dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP); |
92b2e5b3 CH |
620 | |
621 | dqp->dq_flags = type; | |
622 | dqp->q_core.d_id = cpu_to_be32(id); | |
623 | dqp->q_mount = mp; | |
f8739c3c | 624 | INIT_LIST_HEAD(&dqp->q_lru); |
92b2e5b3 CH |
625 | mutex_init(&dqp->q_qlock); |
626 | init_waitqueue_head(&dqp->q_pinwait); | |
627 | ||
628 | /* | |
629 | * Because we want to use a counting completion, complete | |
630 | * the flush completion once to allow a single access to | |
631 | * the flush completion without blocking. | |
632 | */ | |
633 | init_completion(&dqp->q_flush); | |
634 | complete(&dqp->q_flush); | |
635 | ||
636 | /* | |
637 | * Make sure group quotas have a different lock class than user | |
638 | * quotas. | |
639 | */ | |
640 | if (!(type & XFS_DQ_USER)) | |
641 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class); | |
642 | ||
48776fd2 | 643 | XFS_STATS_INC(xs_qm_dquot); |
1da177e4 | 644 | |
0b1b213f CH |
645 | trace_xfs_dqread(dqp); |
646 | ||
97e7ade5 CH |
647 | if (flags & XFS_QMOPT_DQALLOC) { |
648 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC); | |
649 | error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp), | |
48001044 JL |
650 | XFS_QM_DQALLOC_LOG_RES(mp), 0, |
651 | XFS_TRANS_PERM_LOG_RES, | |
652 | XFS_WRITE_LOG_COUNT); | |
97e7ade5 CH |
653 | if (error) |
654 | goto error1; | |
655 | cancelflags = XFS_TRANS_RELEASE_LOG_RES; | |
656 | } | |
657 | ||
1da177e4 LT |
658 | /* |
659 | * get a pointer to the on-disk dquot and the buffer containing it | |
660 | * dqp already knows its own type (GROUP/USER). | |
661 | */ | |
97e7ade5 CH |
662 | error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags); |
663 | if (error) { | |
664 | /* | |
665 | * This can happen if quotas got turned off (ESRCH), | |
666 | * or if the dquot didn't exist on disk and we ask to | |
667 | * allocate (ENOENT). | |
668 | */ | |
669 | trace_xfs_dqread_fail(dqp); | |
670 | cancelflags |= XFS_TRANS_ABORT; | |
671 | goto error1; | |
1da177e4 LT |
672 | } |
673 | ||
674 | /* copy everything from disk dquot to the incore dquot */ | |
675 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
1da177e4 LT |
676 | xfs_qm_dquot_logitem_init(dqp); |
677 | ||
678 | /* | |
679 | * Reservation counters are defined as reservation plus current usage | |
25985edc | 680 | * to avoid having to add every time. |
1da177e4 | 681 | */ |
1149d96a CH |
682 | dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); |
683 | dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); | |
684 | dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); | |
1da177e4 | 685 | |
b1366451 BF |
686 | /* initialize the dquot speculative prealloc thresholds */ |
687 | xfs_dquot_set_prealloc_limits(dqp); | |
688 | ||
1da177e4 | 689 | /* Mark the buf so that this will stay incore a little longer */ |
38f23232 | 690 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 LT |
691 | |
692 | /* | |
693 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
694 | * So we need to release with xfs_trans_brelse(). | |
695 | * The strategy here is identical to that of inodes; we lock | |
696 | * the dquot in xfs_qm_dqget() before making it accessible to | |
697 | * others. This is because dquots, like inodes, need a good level of | |
698 | * concurrency, and we don't want to take locks on the entire buffers | |
699 | * for dquot accesses. | |
700 | * Note also that the dquot buffer may even be dirty at this point, if | |
701 | * this particular dquot was repaired. We still aren't afraid to | |
702 | * brelse it because we have the changes incore. | |
703 | */ | |
0c842ad4 | 704 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
705 | xfs_trans_brelse(tp, bp); |
706 | ||
1da177e4 | 707 | if (tp) { |
97e7ade5 CH |
708 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
709 | if (error) | |
710 | goto error0; | |
1da177e4 LT |
711 | } |
712 | ||
713 | *O_dqpp = dqp; | |
97e7ade5 | 714 | return error; |
1da177e4 | 715 | |
97e7ade5 | 716 | error1: |
1da177e4 LT |
717 | if (tp) |
718 | xfs_trans_cancel(tp, cancelflags); | |
97e7ade5 | 719 | error0: |
1da177e4 LT |
720 | xfs_qm_dqdestroy(dqp); |
721 | *O_dqpp = NULL; | |
97e7ade5 | 722 | return error; |
1da177e4 LT |
723 | } |
724 | ||
1da177e4 LT |
725 | /* |
726 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
727 | * a locked dquot, doing an allocation (if requested) as needed. | |
728 | * When both an inode and an id are given, the inode's id takes precedence. | |
729 | * That is, if the id changes while we don't hold the ilock inside this | |
730 | * function, the new dquot is returned, not necessarily the one requested | |
731 | * in the id argument. | |
732 | */ | |
733 | int | |
734 | xfs_qm_dqget( | |
735 | xfs_mount_t *mp, | |
736 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
737 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
738 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
739 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
740 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
741 | { | |
9f920f11 CH |
742 | struct xfs_quotainfo *qi = mp->m_quotainfo; |
743 | struct radix_tree_root *tree = XFS_DQUOT_TREE(qi, type); | |
744 | struct xfs_dquot *dqp; | |
745 | int error; | |
1da177e4 LT |
746 | |
747 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
748 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 749 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 LT |
750 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
751 | return (ESRCH); | |
752 | } | |
1da177e4 LT |
753 | |
754 | #ifdef DEBUG | |
755 | if (xfs_do_dqerror) { | |
756 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
757 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 758 | xfs_debug(mp, "Returning error in dqget"); |
1da177e4 LT |
759 | return (EIO); |
760 | } | |
761 | } | |
1da177e4 | 762 | |
c8ad20ff NS |
763 | ASSERT(type == XFS_DQ_USER || |
764 | type == XFS_DQ_PROJ || | |
765 | type == XFS_DQ_GROUP); | |
1da177e4 | 766 | if (ip) { |
579aa9ca | 767 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
36731410 | 768 | ASSERT(xfs_inode_dquot(ip, type) == NULL); |
1da177e4 LT |
769 | } |
770 | #endif | |
92678554 CH |
771 | |
772 | restart: | |
9f920f11 CH |
773 | mutex_lock(&qi->qi_tree_lock); |
774 | dqp = radix_tree_lookup(tree, id); | |
775 | if (dqp) { | |
776 | xfs_dqlock(dqp); | |
777 | if (dqp->dq_flags & XFS_DQ_FREEING) { | |
778 | xfs_dqunlock(dqp); | |
779 | mutex_unlock(&qi->qi_tree_lock); | |
780 | trace_xfs_dqget_freeing(dqp); | |
781 | delay(1); | |
782 | goto restart; | |
783 | } | |
1da177e4 | 784 | |
9f920f11 CH |
785 | dqp->q_nrefs++; |
786 | mutex_unlock(&qi->qi_tree_lock); | |
787 | ||
788 | trace_xfs_dqget_hit(dqp); | |
48776fd2 | 789 | XFS_STATS_INC(xs_qm_dqcachehits); |
9f920f11 CH |
790 | *O_dqpp = dqp; |
791 | return 0; | |
1da177e4 | 792 | } |
9f920f11 CH |
793 | mutex_unlock(&qi->qi_tree_lock); |
794 | XFS_STATS_INC(xs_qm_dqcachemisses); | |
1da177e4 LT |
795 | |
796 | /* | |
797 | * Dquot cache miss. We don't want to keep the inode lock across | |
798 | * a (potential) disk read. Also we don't want to deal with the lock | |
799 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
800 | * lock here means dealing with a chown that can happen before | |
801 | * we re-acquire the lock. | |
802 | */ | |
803 | if (ip) | |
804 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 | 805 | |
97e7ade5 | 806 | error = xfs_qm_dqread(mp, id, type, flags, &dqp); |
1da177e4 | 807 | |
7ae44407 CH |
808 | if (ip) |
809 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
810 | ||
811 | if (error) | |
812 | return error; | |
1da177e4 | 813 | |
1da177e4 | 814 | if (ip) { |
1da177e4 LT |
815 | /* |
816 | * A dquot could be attached to this inode by now, since | |
817 | * we had dropped the ilock. | |
818 | */ | |
36731410 | 819 | if (xfs_this_quota_on(mp, type)) { |
9f920f11 CH |
820 | struct xfs_dquot *dqp1; |
821 | ||
36731410 CS |
822 | dqp1 = xfs_inode_dquot(ip, type); |
823 | if (dqp1) { | |
1da177e4 | 824 | xfs_qm_dqdestroy(dqp); |
36731410 | 825 | dqp = dqp1; |
1da177e4 LT |
826 | xfs_dqlock(dqp); |
827 | goto dqret; | |
828 | } | |
829 | } else { | |
36731410 CS |
830 | /* inode stays locked on return */ |
831 | xfs_qm_dqdestroy(dqp); | |
832 | return XFS_ERROR(ESRCH); | |
1da177e4 LT |
833 | } |
834 | } | |
835 | ||
9f920f11 CH |
836 | mutex_lock(&qi->qi_tree_lock); |
837 | error = -radix_tree_insert(tree, id, dqp); | |
838 | if (unlikely(error)) { | |
839 | WARN_ON(error != EEXIST); | |
840 | ||
1da177e4 | 841 | /* |
9f920f11 CH |
842 | * Duplicate found. Just throw away the new dquot and start |
843 | * over. | |
1da177e4 | 844 | */ |
9f920f11 CH |
845 | mutex_unlock(&qi->qi_tree_lock); |
846 | trace_xfs_dqget_dup(dqp); | |
847 | xfs_qm_dqdestroy(dqp); | |
848 | XFS_STATS_INC(xs_qm_dquot_dups); | |
849 | goto restart; | |
1da177e4 LT |
850 | } |
851 | ||
1da177e4 LT |
852 | /* |
853 | * We return a locked dquot to the caller, with a reference taken | |
854 | */ | |
855 | xfs_dqlock(dqp); | |
856 | dqp->q_nrefs = 1; | |
857 | ||
9f920f11 | 858 | qi->qi_dquots++; |
9f920f11 CH |
859 | mutex_unlock(&qi->qi_tree_lock); |
860 | ||
1da177e4 | 861 | dqret: |
579aa9ca | 862 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 863 | trace_xfs_dqget_miss(dqp); |
1da177e4 LT |
864 | *O_dqpp = dqp; |
865 | return (0); | |
866 | } | |
867 | ||
868 | ||
f8739c3c CH |
869 | STATIC void |
870 | xfs_qm_dqput_final( | |
bf72de31 | 871 | struct xfs_dquot *dqp) |
1da177e4 | 872 | { |
f8739c3c | 873 | struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo; |
bf72de31 | 874 | struct xfs_dquot *gdqp; |
1da177e4 | 875 | |
bf72de31 CH |
876 | trace_xfs_dqput_free(dqp); |
877 | ||
f8739c3c CH |
878 | mutex_lock(&qi->qi_lru_lock); |
879 | if (list_empty(&dqp->q_lru)) { | |
880 | list_add_tail(&dqp->q_lru, &qi->qi_lru_list); | |
881 | qi->qi_lru_count++; | |
48776fd2 | 882 | XFS_STATS_INC(xs_qm_dquot_unused); |
bf72de31 | 883 | } |
f8739c3c | 884 | mutex_unlock(&qi->qi_lru_lock); |
bf72de31 | 885 | |
1da177e4 | 886 | /* |
bf72de31 CH |
887 | * If we just added a udquot to the freelist, then we want to release |
888 | * the gdquot reference that it (probably) has. Otherwise it'll keep | |
889 | * the gdquot from getting reclaimed. | |
1da177e4 | 890 | */ |
bf72de31 CH |
891 | gdqp = dqp->q_gdquot; |
892 | if (gdqp) { | |
893 | xfs_dqlock(gdqp); | |
894 | dqp->q_gdquot = NULL; | |
1da177e4 | 895 | } |
bf72de31 | 896 | xfs_dqunlock(dqp); |
1da177e4 | 897 | |
bf72de31 CH |
898 | /* |
899 | * If we had a group quota hint, release it now. | |
900 | */ | |
f8739c3c CH |
901 | if (gdqp) |
902 | xfs_qm_dqput(gdqp); | |
903 | } | |
904 | ||
905 | /* | |
906 | * Release a reference to the dquot (decrement ref-count) and unlock it. | |
907 | * | |
908 | * If there is a group quota attached to this dquot, carefully release that | |
909 | * too without tripping over deadlocks'n'stuff. | |
910 | */ | |
911 | void | |
912 | xfs_qm_dqput( | |
913 | struct xfs_dquot *dqp) | |
914 | { | |
915 | ASSERT(dqp->q_nrefs > 0); | |
916 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
917 | ||
918 | trace_xfs_dqput(dqp); | |
919 | ||
920 | if (--dqp->q_nrefs > 0) | |
921 | xfs_dqunlock(dqp); | |
922 | else | |
923 | xfs_qm_dqput_final(dqp); | |
1da177e4 LT |
924 | } |
925 | ||
926 | /* | |
927 | * Release a dquot. Flush it if dirty, then dqput() it. | |
928 | * dquot must not be locked. | |
929 | */ | |
930 | void | |
931 | xfs_qm_dqrele( | |
932 | xfs_dquot_t *dqp) | |
933 | { | |
7d095257 CH |
934 | if (!dqp) |
935 | return; | |
936 | ||
0b1b213f | 937 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
938 | |
939 | xfs_dqlock(dqp); | |
940 | /* | |
941 | * We don't care to flush it if the dquot is dirty here. | |
942 | * That will create stutters that we want to avoid. | |
943 | * Instead we do a delayed write when we try to reclaim | |
944 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
945 | */ | |
946 | xfs_qm_dqput(dqp); | |
947 | } | |
948 | ||
ca30b2a7 CH |
949 | /* |
950 | * This is the dquot flushing I/O completion routine. It is called | |
951 | * from interrupt level when the buffer containing the dquot is | |
952 | * flushed to disk. It is responsible for removing the dquot logitem | |
953 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
954 | * flush lock. This behavior is very similar to that of inodes.. | |
955 | */ | |
956 | STATIC void | |
957 | xfs_qm_dqflush_done( | |
958 | struct xfs_buf *bp, | |
959 | struct xfs_log_item *lip) | |
960 | { | |
961 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
962 | xfs_dquot_t *dqp = qip->qli_dquot; | |
963 | struct xfs_ail *ailp = lip->li_ailp; | |
964 | ||
965 | /* | |
966 | * We only want to pull the item from the AIL if its | |
967 | * location in the log has not changed since we started the flush. | |
968 | * Thus, we only bother if the dquot's lsn has | |
969 | * not changed. First we check the lsn outside the lock | |
970 | * since it's cheaper, and then we recheck while | |
971 | * holding the lock before removing the dquot from the AIL. | |
972 | */ | |
973 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
974 | lip->li_lsn == qip->qli_flush_lsn) { | |
975 | ||
976 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
977 | spin_lock(&ailp->xa_lock); | |
978 | if (lip->li_lsn == qip->qli_flush_lsn) | |
04913fdd | 979 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 CH |
980 | else |
981 | spin_unlock(&ailp->xa_lock); | |
982 | } | |
983 | ||
984 | /* | |
985 | * Release the dq's flush lock since we're done with it. | |
986 | */ | |
987 | xfs_dqfunlock(dqp); | |
988 | } | |
1da177e4 LT |
989 | |
990 | /* | |
991 | * Write a modified dquot to disk. | |
992 | * The dquot must be locked and the flush lock too taken by caller. | |
993 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
994 | * but the dquot is free to be unlocked and modified by the caller | |
995 | * in the interim. Dquot is still locked on return. This behavior is | |
996 | * identical to that of inodes. | |
997 | */ | |
998 | int | |
999 | xfs_qm_dqflush( | |
fe7257fd CH |
1000 | struct xfs_dquot *dqp, |
1001 | struct xfs_buf **bpp) | |
1da177e4 | 1002 | { |
acecf1b5 CH |
1003 | struct xfs_mount *mp = dqp->q_mount; |
1004 | struct xfs_buf *bp; | |
1005 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 1006 | int error; |
1da177e4 LT |
1007 | |
1008 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 1009 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 1010 | |
0b1b213f | 1011 | trace_xfs_dqflush(dqp); |
1da177e4 | 1012 | |
fe7257fd CH |
1013 | *bpp = NULL; |
1014 | ||
1da177e4 LT |
1015 | xfs_qm_dqunpin_wait(dqp); |
1016 | ||
1017 | /* | |
1018 | * This may have been unpinned because the filesystem is shutting | |
1019 | * down forcibly. If that's the case we must not write this dquot | |
dea96095 CH |
1020 | * to disk, because the log record didn't make it to disk. |
1021 | * | |
1022 | * We also have to remove the log item from the AIL in this case, | |
1023 | * as we wait for an emptry AIL as part of the unmount process. | |
1da177e4 | 1024 | */ |
acecf1b5 | 1025 | if (XFS_FORCED_SHUTDOWN(mp)) { |
dea96095 | 1026 | struct xfs_log_item *lip = &dqp->q_logitem.qli_item; |
acecf1b5 | 1027 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
dea96095 CH |
1028 | |
1029 | spin_lock(&mp->m_ail->xa_lock); | |
1030 | if (lip->li_flags & XFS_LI_IN_AIL) | |
04913fdd DC |
1031 | xfs_trans_ail_delete(mp->m_ail, lip, |
1032 | SHUTDOWN_CORRUPT_INCORE); | |
dea96095 CH |
1033 | else |
1034 | spin_unlock(&mp->m_ail->xa_lock); | |
fe7257fd CH |
1035 | error = XFS_ERROR(EIO); |
1036 | goto out_unlock; | |
1da177e4 LT |
1037 | } |
1038 | ||
1039 | /* | |
1040 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1041 | */ |
acecf1b5 | 1042 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
c3f8fc73 | 1043 | mp->m_quotainfo->qi_dqchunklen, 0, &bp, NULL); |
fe7257fd CH |
1044 | if (error) |
1045 | goto out_unlock; | |
1da177e4 | 1046 | |
acecf1b5 CH |
1047 | /* |
1048 | * Calculate the location of the dquot inside the buffer. | |
1049 | */ | |
62926044 | 1050 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1051 | |
1052 | /* | |
1053 | * A simple sanity check in case we got a corrupted dquot.. | |
1054 | */ | |
a0fa2b67 DC |
1055 | error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
1056 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); | |
1057 | if (error) { | |
acecf1b5 CH |
1058 | xfs_buf_relse(bp); |
1059 | xfs_dqfunlock(dqp); | |
1060 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
1da177e4 LT |
1061 | return XFS_ERROR(EIO); |
1062 | } | |
1063 | ||
1064 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1065 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1066 | |
1067 | /* | |
1068 | * Clear the dirty field and remember the flush lsn for later use. | |
1069 | */ | |
acecf1b5 | 1070 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1071 | |
7b2e2a31 DC |
1072 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1073 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 LT |
1074 | |
1075 | /* | |
1076 | * Attach an iodone routine so that we can remove this dquot from the | |
1077 | * AIL and release the flush lock once the dquot is synced to disk. | |
1078 | */ | |
ca30b2a7 CH |
1079 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1080 | &dqp->q_logitem.qli_item); | |
1081 | ||
1da177e4 LT |
1082 | /* |
1083 | * If the buffer is pinned then push on the log so we won't | |
1084 | * get stuck waiting in the write for too long. | |
1085 | */ | |
811e64c7 | 1086 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1087 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1088 | xfs_log_force(mp, 0); |
1da177e4 LT |
1089 | } |
1090 | ||
0b1b213f | 1091 | trace_xfs_dqflush_done(dqp); |
fe7257fd CH |
1092 | *bpp = bp; |
1093 | return 0; | |
0b1b213f | 1094 | |
fe7257fd CH |
1095 | out_unlock: |
1096 | xfs_dqfunlock(dqp); | |
1097 | return XFS_ERROR(EIO); | |
1da177e4 LT |
1098 | } |
1099 | ||
5bb87a33 CH |
1100 | /* |
1101 | * Lock two xfs_dquot structures. | |
1102 | * | |
1103 | * To avoid deadlocks we always lock the quota structure with | |
1104 | * the lowerd id first. | |
1105 | */ | |
1da177e4 LT |
1106 | void |
1107 | xfs_dqlock2( | |
1108 | xfs_dquot_t *d1, | |
1109 | xfs_dquot_t *d2) | |
1110 | { | |
1111 | if (d1 && d2) { | |
1112 | ASSERT(d1 != d2); | |
1149d96a CH |
1113 | if (be32_to_cpu(d1->q_core.d_id) > |
1114 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1115 | mutex_lock(&d2->q_qlock); |
1116 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1117 | } else { |
5bb87a33 CH |
1118 | mutex_lock(&d1->q_qlock); |
1119 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1120 | } |
5bb87a33 CH |
1121 | } else if (d1) { |
1122 | mutex_lock(&d1->q_qlock); | |
1123 | } else if (d2) { | |
1124 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1125 | } |
1126 | } | |
1127 | ||
a05931ce CH |
1128 | int __init |
1129 | xfs_qm_init(void) | |
1130 | { | |
1131 | xfs_qm_dqzone = | |
1132 | kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot"); | |
1133 | if (!xfs_qm_dqzone) | |
1134 | goto out; | |
1135 | ||
1136 | xfs_qm_dqtrxzone = | |
1137 | kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx"); | |
1138 | if (!xfs_qm_dqtrxzone) | |
1139 | goto out_free_dqzone; | |
1140 | ||
1141 | return 0; | |
1142 | ||
1143 | out_free_dqzone: | |
1144 | kmem_zone_destroy(xfs_qm_dqzone); | |
1145 | out: | |
1146 | return -ENOMEM; | |
1147 | } | |
1148 | ||
1c2ccc66 | 1149 | void |
a05931ce CH |
1150 | xfs_qm_exit(void) |
1151 | { | |
1152 | kmem_zone_destroy(xfs_qm_dqtrxzone); | |
1153 | kmem_zone_destroy(xfs_qm_dqzone); | |
1154 | } |