]>
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" |
70a9883c | 21 | #include "xfs_shared.h" |
a844f451 | 22 | #include "xfs_bit.h" |
1da177e4 LT |
23 | #include "xfs_log.h" |
24 | #include "xfs_trans.h" | |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_ag.h" | |
1da177e4 | 27 | #include "xfs_alloc.h" |
1da177e4 LT |
28 | #include "xfs_quota.h" |
29 | #include "xfs_mount.h" | |
1da177e4 | 30 | #include "xfs_bmap_btree.h" |
1da177e4 LT |
31 | #include "xfs_inode.h" |
32 | #include "xfs_bmap.h" | |
68988114 | 33 | #include "xfs_bmap_util.h" |
1da177e4 LT |
34 | #include "xfs_rtalloc.h" |
35 | #include "xfs_error.h" | |
36 | #include "xfs_itable.h" | |
1da177e4 LT |
37 | #include "xfs_attr.h" |
38 | #include "xfs_buf_item.h" | |
39 | #include "xfs_trans_space.h" | |
40 | #include "xfs_trans_priv.h" | |
1da177e4 | 41 | #include "xfs_qm.h" |
3fe58f30 | 42 | #include "xfs_cksum.h" |
0b1b213f | 43 | #include "xfs_trace.h" |
1da177e4 | 44 | |
1da177e4 | 45 | /* |
bf72de31 CH |
46 | * Lock order: |
47 | * | |
48 | * ip->i_lock | |
9f920f11 | 49 | * qi->qi_tree_lock |
b84a3a96 CH |
50 | * dquot->q_qlock (xfs_dqlock() and friends) |
51 | * dquot->q_flush (xfs_dqflock() and friends) | |
52 | * qi->qi_lru_lock | |
bf72de31 CH |
53 | * |
54 | * If two dquots need to be locked the order is user before group/project, | |
55 | * otherwise by the lowest id first, see xfs_dqlock2. | |
56 | */ | |
1da177e4 | 57 | |
1da177e4 LT |
58 | #ifdef DEBUG |
59 | xfs_buftarg_t *xfs_dqerror_target; | |
60 | int xfs_do_dqerror; | |
61 | int xfs_dqreq_num; | |
62 | int xfs_dqerror_mod = 33; | |
63 | #endif | |
64 | ||
a05931ce CH |
65 | struct kmem_zone *xfs_qm_dqtrxzone; |
66 | static struct kmem_zone *xfs_qm_dqzone; | |
67 | ||
f112a049 DC |
68 | static struct lock_class_key xfs_dquot_group_class; |
69 | static struct lock_class_key xfs_dquot_project_class; | |
98b8c7a0 | 70 | |
1da177e4 LT |
71 | /* |
72 | * This is called to free all the memory associated with a dquot | |
73 | */ | |
74 | void | |
75 | xfs_qm_dqdestroy( | |
76 | xfs_dquot_t *dqp) | |
77 | { | |
f8739c3c | 78 | ASSERT(list_empty(&dqp->q_lru)); |
1da177e4 LT |
79 | |
80 | mutex_destroy(&dqp->q_qlock); | |
a05931ce | 81 | kmem_zone_free(xfs_qm_dqzone, dqp); |
0b1b213f | 82 | |
48776fd2 | 83 | XFS_STATS_DEC(xs_qm_dquot); |
1da177e4 LT |
84 | } |
85 | ||
1da177e4 LT |
86 | /* |
87 | * If default limits are in force, push them into the dquot now. | |
88 | * We overwrite the dquot limits only if they are zero and this | |
89 | * is not the root dquot. | |
90 | */ | |
91 | void | |
92 | xfs_qm_adjust_dqlimits( | |
4b6eae2e BF |
93 | struct xfs_mount *mp, |
94 | struct xfs_dquot *dq) | |
1da177e4 | 95 | { |
4b6eae2e BF |
96 | struct xfs_quotainfo *q = mp->m_quotainfo; |
97 | struct xfs_disk_dquot *d = &dq->q_core; | |
b1366451 | 98 | int prealloc = 0; |
1da177e4 LT |
99 | |
100 | ASSERT(d->d_id); | |
101 | ||
b1366451 | 102 | if (q->qi_bsoftlimit && !d->d_blk_softlimit) { |
1149d96a | 103 | d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit); |
b1366451 BF |
104 | prealloc = 1; |
105 | } | |
106 | if (q->qi_bhardlimit && !d->d_blk_hardlimit) { | |
1149d96a | 107 | d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit); |
b1366451 BF |
108 | prealloc = 1; |
109 | } | |
1da177e4 | 110 | if (q->qi_isoftlimit && !d->d_ino_softlimit) |
1149d96a | 111 | d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit); |
1da177e4 | 112 | if (q->qi_ihardlimit && !d->d_ino_hardlimit) |
1149d96a | 113 | d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit); |
1da177e4 | 114 | if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit) |
1149d96a | 115 | d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit); |
1da177e4 | 116 | if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit) |
1149d96a | 117 | d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit); |
b1366451 BF |
118 | |
119 | if (prealloc) | |
120 | xfs_dquot_set_prealloc_limits(dq); | |
1da177e4 LT |
121 | } |
122 | ||
123 | /* | |
124 | * Check the limits and timers of a dquot and start or reset timers | |
125 | * if necessary. | |
126 | * This gets called even when quota enforcement is OFF, which makes our | |
127 | * life a little less complicated. (We just don't reject any quota | |
128 | * reservations in that case, when enforcement is off). | |
129 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
130 | * enforcement's off. | |
131 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
132 | * 'automatically' get started when limits get exceeded. They do |
133 | * get reset to zero, however, when we find the count to be under | |
134 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
135 | */ |
136 | void | |
137 | xfs_qm_adjust_dqtimers( | |
138 | xfs_mount_t *mp, | |
139 | xfs_disk_dquot_t *d) | |
140 | { | |
141 | ASSERT(d->d_id); | |
142 | ||
ea15ab3c | 143 | #ifdef DEBUG |
1149d96a CH |
144 | if (d->d_blk_hardlimit) |
145 | ASSERT(be64_to_cpu(d->d_blk_softlimit) <= | |
146 | be64_to_cpu(d->d_blk_hardlimit)); | |
147 | if (d->d_ino_hardlimit) | |
148 | ASSERT(be64_to_cpu(d->d_ino_softlimit) <= | |
149 | be64_to_cpu(d->d_ino_hardlimit)); | |
150 | if (d->d_rtb_hardlimit) | |
151 | ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= | |
152 | be64_to_cpu(d->d_rtb_hardlimit)); | |
1da177e4 | 153 | #endif |
ea15ab3c | 154 | |
1da177e4 | 155 | if (!d->d_btimer) { |
1149d96a | 156 | if ((d->d_blk_softlimit && |
d0a3fe67 | 157 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
158 | be64_to_cpu(d->d_blk_softlimit))) || |
159 | (d->d_blk_hardlimit && | |
d0a3fe67 | 160 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
161 | be64_to_cpu(d->d_blk_hardlimit)))) { |
162 | d->d_btimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 163 | mp->m_quotainfo->qi_btimelimit); |
754002b4 NS |
164 | } else { |
165 | d->d_bwarns = 0; | |
1da177e4 LT |
166 | } |
167 | } else { | |
168 | if ((!d->d_blk_softlimit || | |
d0a3fe67 | 169 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 170 | be64_to_cpu(d->d_blk_softlimit))) && |
1da177e4 | 171 | (!d->d_blk_hardlimit || |
d0a3fe67 | 172 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 173 | be64_to_cpu(d->d_blk_hardlimit)))) { |
1da177e4 LT |
174 | d->d_btimer = 0; |
175 | } | |
176 | } | |
177 | ||
178 | if (!d->d_itimer) { | |
1149d96a | 179 | if ((d->d_ino_softlimit && |
d0a3fe67 | 180 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
181 | be64_to_cpu(d->d_ino_softlimit))) || |
182 | (d->d_ino_hardlimit && | |
d0a3fe67 | 183 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
184 | be64_to_cpu(d->d_ino_hardlimit)))) { |
185 | d->d_itimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 186 | mp->m_quotainfo->qi_itimelimit); |
754002b4 NS |
187 | } else { |
188 | d->d_iwarns = 0; | |
1da177e4 LT |
189 | } |
190 | } else { | |
191 | if ((!d->d_ino_softlimit || | |
d0a3fe67 | 192 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 193 | be64_to_cpu(d->d_ino_softlimit))) && |
1da177e4 | 194 | (!d->d_ino_hardlimit || |
d0a3fe67 | 195 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 196 | be64_to_cpu(d->d_ino_hardlimit)))) { |
1da177e4 LT |
197 | d->d_itimer = 0; |
198 | } | |
199 | } | |
200 | ||
201 | if (!d->d_rtbtimer) { | |
1149d96a | 202 | if ((d->d_rtb_softlimit && |
d0a3fe67 | 203 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
204 | be64_to_cpu(d->d_rtb_softlimit))) || |
205 | (d->d_rtb_hardlimit && | |
d0a3fe67 | 206 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
207 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
208 | d->d_rtbtimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 209 | mp->m_quotainfo->qi_rtbtimelimit); |
754002b4 NS |
210 | } else { |
211 | d->d_rtbwarns = 0; | |
1da177e4 LT |
212 | } |
213 | } else { | |
214 | if ((!d->d_rtb_softlimit || | |
d0a3fe67 | 215 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 216 | be64_to_cpu(d->d_rtb_softlimit))) && |
1da177e4 | 217 | (!d->d_rtb_hardlimit || |
d0a3fe67 | 218 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 219 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
1da177e4 LT |
220 | d->d_rtbtimer = 0; |
221 | } | |
222 | } | |
223 | } | |
224 | ||
1da177e4 LT |
225 | /* |
226 | * initialize a buffer full of dquots and log the whole thing | |
227 | */ | |
228 | STATIC void | |
229 | xfs_qm_init_dquot_blk( | |
230 | xfs_trans_t *tp, | |
231 | xfs_mount_t *mp, | |
232 | xfs_dqid_t id, | |
233 | uint type, | |
234 | xfs_buf_t *bp) | |
235 | { | |
8a7b8a89 | 236 | struct xfs_quotainfo *q = mp->m_quotainfo; |
1da177e4 LT |
237 | xfs_dqblk_t *d; |
238 | int curid, i; | |
239 | ||
240 | ASSERT(tp); | |
0c842ad4 | 241 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 242 | |
62926044 | 243 | d = bp->b_addr; |
1da177e4 LT |
244 | |
245 | /* | |
246 | * ID of the first dquot in the block - id's are zero based. | |
247 | */ | |
8a7b8a89 | 248 | curid = id - (id % q->qi_dqperchunk); |
1da177e4 | 249 | ASSERT(curid >= 0); |
8a7b8a89 | 250 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
49d35a5c CH |
251 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) { |
252 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
253 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
254 | d->dd_diskdq.d_id = cpu_to_be32(curid); | |
255 | d->dd_diskdq.d_flags = type; | |
6fcdc59d | 256 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
3fe58f30 | 257 | uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid); |
6fcdc59d DC |
258 | xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk), |
259 | XFS_DQUOT_CRC_OFF); | |
260 | } | |
49d35a5c CH |
261 | } |
262 | ||
1da177e4 | 263 | xfs_trans_dquot_buf(tp, bp, |
c1155410 DC |
264 | (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : |
265 | ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : | |
266 | XFS_BLF_GDQUOT_BUF))); | |
8a7b8a89 | 267 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); |
1da177e4 LT |
268 | } |
269 | ||
b1366451 BF |
270 | /* |
271 | * Initialize the dynamic speculative preallocation thresholds. The lo/hi | |
272 | * watermarks correspond to the soft and hard limits by default. If a soft limit | |
273 | * is not specified, we use 95% of the hard limit. | |
274 | */ | |
275 | void | |
276 | xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp) | |
277 | { | |
278 | __uint64_t space; | |
279 | ||
280 | dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit); | |
281 | dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit); | |
282 | if (!dqp->q_prealloc_lo_wmark) { | |
283 | dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark; | |
284 | do_div(dqp->q_prealloc_lo_wmark, 100); | |
285 | dqp->q_prealloc_lo_wmark *= 95; | |
286 | } | |
287 | ||
288 | space = dqp->q_prealloc_hi_wmark; | |
289 | ||
290 | do_div(space, 100); | |
291 | dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space; | |
292 | dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3; | |
293 | dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5; | |
294 | } | |
295 | ||
3fe58f30 CH |
296 | STATIC bool |
297 | xfs_dquot_buf_verify_crc( | |
298 | struct xfs_mount *mp, | |
299 | struct xfs_buf *bp) | |
300 | { | |
301 | struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr; | |
302 | int ndquots; | |
303 | int i; | |
304 | ||
305 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
306 | return true; | |
307 | ||
308 | /* | |
309 | * if we are in log recovery, the quota subsystem has not been | |
310 | * initialised so we have no quotainfo structure. In that case, we need | |
311 | * to manually calculate the number of dquots in the buffer. | |
312 | */ | |
313 | if (mp->m_quotainfo) | |
314 | ndquots = mp->m_quotainfo->qi_dqperchunk; | |
315 | else | |
316 | ndquots = xfs_qm_calc_dquots_per_chunk(mp, | |
317 | XFS_BB_TO_FSB(mp, bp->b_length)); | |
318 | ||
319 | for (i = 0; i < ndquots; i++, d++) { | |
320 | if (!xfs_verify_cksum((char *)d, sizeof(struct xfs_dqblk), | |
6fcdc59d | 321 | XFS_DQUOT_CRC_OFF)) |
3fe58f30 CH |
322 | return false; |
323 | if (!uuid_equal(&d->dd_uuid, &mp->m_sb.sb_uuid)) | |
324 | return false; | |
325 | } | |
3fe58f30 CH |
326 | return true; |
327 | } | |
328 | ||
329 | STATIC bool | |
b0f539de | 330 | xfs_dquot_buf_verify( |
3fe58f30 | 331 | struct xfs_mount *mp, |
b0f539de DC |
332 | struct xfs_buf *bp) |
333 | { | |
b0f539de | 334 | struct xfs_dqblk *d = (struct xfs_dqblk *)bp->b_addr; |
b0f539de | 335 | xfs_dqid_t id = 0; |
3fe58f30 | 336 | int ndquots; |
b0f539de DC |
337 | int i; |
338 | ||
3fe58f30 CH |
339 | /* |
340 | * if we are in log recovery, the quota subsystem has not been | |
341 | * initialised so we have no quotainfo structure. In that case, we need | |
342 | * to manually calculate the number of dquots in the buffer. | |
343 | */ | |
344 | if (mp->m_quotainfo) | |
345 | ndquots = mp->m_quotainfo->qi_dqperchunk; | |
346 | else | |
347 | ndquots = xfs_qm_calc_dquots_per_chunk(mp, bp->b_length); | |
348 | ||
b0f539de DC |
349 | /* |
350 | * On the first read of the buffer, verify that each dquot is valid. | |
351 | * We don't know what the id of the dquot is supposed to be, just that | |
352 | * they should be increasing monotonically within the buffer. If the | |
353 | * first id is corrupt, then it will fail on the second dquot in the | |
354 | * buffer so corruptions could point to the wrong dquot in this case. | |
355 | */ | |
3fe58f30 CH |
356 | for (i = 0; i < ndquots; i++) { |
357 | struct xfs_disk_dquot *ddq; | |
358 | int error; | |
b0f539de DC |
359 | |
360 | ddq = &d[i].dd_diskdq; | |
361 | ||
362 | if (i == 0) | |
363 | id = be32_to_cpu(ddq->d_id); | |
364 | ||
365 | error = xfs_qm_dqcheck(mp, ddq, id + i, 0, XFS_QMOPT_DOWARN, | |
3fe58f30 CH |
366 | "xfs_dquot_buf_verify"); |
367 | if (error) | |
368 | return false; | |
b0f539de | 369 | } |
3fe58f30 | 370 | return true; |
b0f539de DC |
371 | } |
372 | ||
373 | static void | |
1813dd64 | 374 | xfs_dquot_buf_read_verify( |
b0f539de DC |
375 | struct xfs_buf *bp) |
376 | { | |
3fe58f30 CH |
377 | struct xfs_mount *mp = bp->b_target->bt_mount; |
378 | ||
379 | if (!xfs_dquot_buf_verify_crc(mp, bp) || !xfs_dquot_buf_verify(mp, bp)) { | |
380 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr); | |
381 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
382 | } | |
b0f539de | 383 | } |
1da177e4 | 384 | |
6fcdc59d DC |
385 | /* |
386 | * we don't calculate the CRC here as that is done when the dquot is flushed to | |
387 | * the buffer after the update is done. This ensures that the dquot in the | |
388 | * buffer always has an up-to-date CRC value. | |
389 | */ | |
b0f539de | 390 | void |
1813dd64 | 391 | xfs_dquot_buf_write_verify( |
b0f539de DC |
392 | struct xfs_buf *bp) |
393 | { | |
3fe58f30 CH |
394 | struct xfs_mount *mp = bp->b_target->bt_mount; |
395 | ||
396 | if (!xfs_dquot_buf_verify(mp, bp)) { | |
397 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr); | |
398 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
399 | return; | |
400 | } | |
b0f539de | 401 | } |
1da177e4 | 402 | |
1813dd64 DC |
403 | const struct xfs_buf_ops xfs_dquot_buf_ops = { |
404 | .verify_read = xfs_dquot_buf_read_verify, | |
405 | .verify_write = xfs_dquot_buf_write_verify, | |
406 | }; | |
407 | ||
1da177e4 LT |
408 | /* |
409 | * Allocate a block and fill it with dquots. | |
410 | * This is called when the bmapi finds a hole. | |
411 | */ | |
412 | STATIC int | |
413 | xfs_qm_dqalloc( | |
efa092f3 | 414 | xfs_trans_t **tpp, |
1da177e4 LT |
415 | xfs_mount_t *mp, |
416 | xfs_dquot_t *dqp, | |
417 | xfs_inode_t *quotip, | |
418 | xfs_fileoff_t offset_fsb, | |
419 | xfs_buf_t **O_bpp) | |
420 | { | |
421 | xfs_fsblock_t firstblock; | |
422 | xfs_bmap_free_t flist; | |
423 | xfs_bmbt_irec_t map; | |
424 | int nmaps, error, committed; | |
425 | xfs_buf_t *bp; | |
efa092f3 | 426 | xfs_trans_t *tp = *tpp; |
1da177e4 LT |
427 | |
428 | ASSERT(tp != NULL); | |
0b1b213f CH |
429 | |
430 | trace_xfs_dqalloc(dqp); | |
1da177e4 LT |
431 | |
432 | /* | |
433 | * Initialize the bmap freelist prior to calling bmapi code. | |
434 | */ | |
9d87c319 | 435 | xfs_bmap_init(&flist, &firstblock); |
1da177e4 LT |
436 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
437 | /* | |
438 | * Return if this type of quotas is turned off while we didn't | |
439 | * have an inode lock | |
440 | */ | |
6967b964 | 441 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 LT |
442 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
443 | return (ESRCH); | |
444 | } | |
445 | ||
ddc3415a | 446 | xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); |
1da177e4 | 447 | nmaps = 1; |
c0dc7828 DC |
448 | error = xfs_bmapi_write(tp, quotip, offset_fsb, |
449 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
450 | &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), | |
451 | &map, &nmaps, &flist); | |
452 | if (error) | |
1da177e4 | 453 | goto error0; |
1da177e4 LT |
454 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
455 | ASSERT(nmaps == 1); | |
456 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
457 | (map.br_startblock != HOLESTARTBLOCK)); | |
458 | ||
459 | /* | |
460 | * Keep track of the blkno to save a lookup later | |
461 | */ | |
462 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
463 | ||
464 | /* now we can just get the buffer (there's nothing to read yet) */ | |
465 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
466 | dqp->q_blkno, | |
8a7b8a89 | 467 | mp->m_quotainfo->qi_dqchunklen, |
1da177e4 | 468 | 0); |
2a30f36d CS |
469 | |
470 | error = xfs_buf_geterror(bp); | |
471 | if (error) | |
1da177e4 | 472 | goto error1; |
1813dd64 | 473 | bp->b_ops = &xfs_dquot_buf_ops; |
2a30f36d | 474 | |
1da177e4 LT |
475 | /* |
476 | * Make a chunk of dquots out of this buffer and log | |
477 | * the entire thing. | |
478 | */ | |
1149d96a | 479 | xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 480 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
1da177e4 | 481 | |
efa092f3 TS |
482 | /* |
483 | * xfs_bmap_finish() may commit the current transaction and | |
484 | * start a second transaction if the freelist is not empty. | |
485 | * | |
486 | * Since we still want to modify this buffer, we need to | |
487 | * ensure that the buffer is not released on commit of | |
488 | * the first transaction and ensure the buffer is added to the | |
489 | * second transaction. | |
490 | * | |
491 | * If there is only one transaction then don't stop the buffer | |
492 | * from being released when it commits later on. | |
493 | */ | |
494 | ||
495 | xfs_trans_bhold(tp, bp); | |
496 | ||
f7c99b6f | 497 | if ((error = xfs_bmap_finish(tpp, &flist, &committed))) { |
1da177e4 LT |
498 | goto error1; |
499 | } | |
500 | ||
efa092f3 TS |
501 | if (committed) { |
502 | tp = *tpp; | |
503 | xfs_trans_bjoin(tp, bp); | |
504 | } else { | |
505 | xfs_trans_bhold_release(tp, bp); | |
506 | } | |
507 | ||
1da177e4 LT |
508 | *O_bpp = bp; |
509 | return 0; | |
510 | ||
511 | error1: | |
512 | xfs_bmap_cancel(&flist); | |
513 | error0: | |
514 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
515 | ||
516 | return (error); | |
517 | } | |
c6319198 DC |
518 | STATIC int |
519 | xfs_qm_dqrepair( | |
520 | struct xfs_mount *mp, | |
521 | struct xfs_trans *tp, | |
522 | struct xfs_dquot *dqp, | |
523 | xfs_dqid_t firstid, | |
524 | struct xfs_buf **bpp) | |
525 | { | |
526 | int error; | |
527 | struct xfs_disk_dquot *ddq; | |
528 | struct xfs_dqblk *d; | |
529 | int i; | |
530 | ||
531 | /* | |
532 | * Read the buffer without verification so we get the corrupted | |
612cfbfe | 533 | * buffer returned to us. make sure we verify it on write, though. |
c6319198 DC |
534 | */ |
535 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno, | |
536 | mp->m_quotainfo->qi_dqchunklen, | |
537 | 0, bpp, NULL); | |
538 | ||
539 | if (error) { | |
540 | ASSERT(*bpp == NULL); | |
541 | return XFS_ERROR(error); | |
542 | } | |
1813dd64 | 543 | (*bpp)->b_ops = &xfs_dquot_buf_ops; |
c6319198 DC |
544 | |
545 | ASSERT(xfs_buf_islocked(*bpp)); | |
546 | d = (struct xfs_dqblk *)(*bpp)->b_addr; | |
547 | ||
548 | /* Do the actual repair of dquots in this buffer */ | |
549 | for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++) { | |
550 | ddq = &d[i].dd_diskdq; | |
551 | error = xfs_qm_dqcheck(mp, ddq, firstid + i, | |
552 | dqp->dq_flags & XFS_DQ_ALLTYPES, | |
553 | XFS_QMOPT_DQREPAIR, "xfs_qm_dqrepair"); | |
554 | if (error) { | |
555 | /* repair failed, we're screwed */ | |
556 | xfs_trans_brelse(tp, *bpp); | |
557 | return XFS_ERROR(EIO); | |
558 | } | |
559 | } | |
560 | ||
561 | return 0; | |
562 | } | |
563 | ||
1da177e4 LT |
564 | /* |
565 | * Maps a dquot to the buffer containing its on-disk version. | |
566 | * This returns a ptr to the buffer containing the on-disk dquot | |
567 | * in the bpp param, and a ptr to the on-disk dquot within that buffer | |
568 | */ | |
569 | STATIC int | |
570 | xfs_qm_dqtobp( | |
efa092f3 | 571 | xfs_trans_t **tpp, |
1da177e4 LT |
572 | xfs_dquot_t *dqp, |
573 | xfs_disk_dquot_t **O_ddpp, | |
574 | xfs_buf_t **O_bpp, | |
575 | uint flags) | |
576 | { | |
113a5683 CS |
577 | struct xfs_bmbt_irec map; |
578 | int nmaps = 1, error; | |
579 | struct xfs_buf *bp; | |
995961c4 | 580 | struct xfs_inode *quotip = xfs_dq_to_quota_inode(dqp); |
113a5683 CS |
581 | struct xfs_mount *mp = dqp->q_mount; |
582 | xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); | |
583 | struct xfs_trans *tp = (tpp ? *tpp : NULL); | |
1da177e4 | 584 | |
acecf1b5 | 585 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
1da177e4 | 586 | |
acecf1b5 | 587 | xfs_ilock(quotip, XFS_ILOCK_SHARED); |
6967b964 | 588 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 | 589 | /* |
acecf1b5 CH |
590 | * Return if this type of quotas is turned off while we |
591 | * didn't have the quota inode lock. | |
1da177e4 | 592 | */ |
acecf1b5 CH |
593 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); |
594 | return ESRCH; | |
595 | } | |
596 | ||
597 | /* | |
598 | * Find the block map; no allocations yet | |
599 | */ | |
5c8ed202 DC |
600 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
601 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); | |
acecf1b5 CH |
602 | |
603 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); | |
604 | if (error) | |
605 | return error; | |
606 | ||
607 | ASSERT(nmaps == 1); | |
608 | ASSERT(map.br_blockcount == 1); | |
609 | ||
610 | /* | |
611 | * Offset of dquot in the (fixed sized) dquot chunk. | |
612 | */ | |
613 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
614 | sizeof(xfs_dqblk_t); | |
615 | ||
616 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
617 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 618 | /* |
acecf1b5 | 619 | * We don't allocate unless we're asked to |
1da177e4 | 620 | */ |
acecf1b5 CH |
621 | if (!(flags & XFS_QMOPT_DQALLOC)) |
622 | return ENOENT; | |
1da177e4 | 623 | |
acecf1b5 CH |
624 | ASSERT(tp); |
625 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
626 | dqp->q_fileoffset, &bp); | |
1da177e4 | 627 | if (error) |
acecf1b5 CH |
628 | return error; |
629 | tp = *tpp; | |
630 | } else { | |
631 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
632 | |
633 | /* | |
acecf1b5 CH |
634 | * store the blkno etc so that we don't have to do the |
635 | * mapping all the time | |
1da177e4 | 636 | */ |
acecf1b5 | 637 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 638 | |
8a7b8a89 CH |
639 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
640 | dqp->q_blkno, | |
641 | mp->m_quotainfo->qi_dqchunklen, | |
1813dd64 | 642 | 0, &bp, &xfs_dquot_buf_ops); |
acecf1b5 | 643 | |
c6319198 DC |
644 | if (error == EFSCORRUPTED && (flags & XFS_QMOPT_DQREPAIR)) { |
645 | xfs_dqid_t firstid = (xfs_dqid_t)map.br_startoff * | |
646 | mp->m_quotainfo->qi_dqperchunk; | |
647 | ASSERT(bp == NULL); | |
648 | error = xfs_qm_dqrepair(mp, tp, dqp, firstid, &bp); | |
649 | } | |
1da177e4 | 650 | |
c6319198 DC |
651 | if (error) { |
652 | ASSERT(bp == NULL); | |
653 | return XFS_ERROR(error); | |
1da177e4 | 654 | } |
1da177e4 LT |
655 | } |
656 | ||
c6319198 | 657 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 658 | *O_bpp = bp; |
c6319198 | 659 | *O_ddpp = bp->b_addr + dqp->q_bufoffset; |
1da177e4 LT |
660 | |
661 | return (0); | |
662 | } | |
663 | ||
664 | ||
665 | /* | |
666 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
667 | * and release the buffer immediately. | |
668 | * | |
97e7ade5 | 669 | * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed. |
1da177e4 | 670 | */ |
7ae44407 | 671 | int |
1da177e4 | 672 | xfs_qm_dqread( |
97e7ade5 CH |
673 | struct xfs_mount *mp, |
674 | xfs_dqid_t id, | |
675 | uint type, | |
676 | uint flags, | |
677 | struct xfs_dquot **O_dqpp) | |
1da177e4 | 678 | { |
97e7ade5 CH |
679 | struct xfs_dquot *dqp; |
680 | struct xfs_disk_dquot *ddqp; | |
681 | struct xfs_buf *bp; | |
682 | struct xfs_trans *tp = NULL; | |
683 | int error; | |
684 | int cancelflags = 0; | |
efa092f3 | 685 | |
92b2e5b3 | 686 | |
a05931ce | 687 | dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP); |
92b2e5b3 CH |
688 | |
689 | dqp->dq_flags = type; | |
690 | dqp->q_core.d_id = cpu_to_be32(id); | |
691 | dqp->q_mount = mp; | |
f8739c3c | 692 | INIT_LIST_HEAD(&dqp->q_lru); |
92b2e5b3 CH |
693 | mutex_init(&dqp->q_qlock); |
694 | init_waitqueue_head(&dqp->q_pinwait); | |
695 | ||
696 | /* | |
697 | * Because we want to use a counting completion, complete | |
698 | * the flush completion once to allow a single access to | |
699 | * the flush completion without blocking. | |
700 | */ | |
701 | init_completion(&dqp->q_flush); | |
702 | complete(&dqp->q_flush); | |
703 | ||
704 | /* | |
705 | * Make sure group quotas have a different lock class than user | |
706 | * quotas. | |
707 | */ | |
f112a049 DC |
708 | switch (type) { |
709 | case XFS_DQ_USER: | |
710 | /* uses the default lock class */ | |
711 | break; | |
712 | case XFS_DQ_GROUP: | |
713 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class); | |
714 | break; | |
715 | case XFS_DQ_PROJ: | |
716 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class); | |
717 | break; | |
718 | default: | |
719 | ASSERT(0); | |
720 | break; | |
721 | } | |
92b2e5b3 | 722 | |
48776fd2 | 723 | XFS_STATS_INC(xs_qm_dquot); |
1da177e4 | 724 | |
0b1b213f CH |
725 | trace_xfs_dqread(dqp); |
726 | ||
97e7ade5 CH |
727 | if (flags & XFS_QMOPT_DQALLOC) { |
728 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC); | |
3d3c8b52 JL |
729 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_attrsetm, |
730 | XFS_QM_DQALLOC_SPACE_RES(mp), 0); | |
97e7ade5 CH |
731 | if (error) |
732 | goto error1; | |
733 | cancelflags = XFS_TRANS_RELEASE_LOG_RES; | |
734 | } | |
735 | ||
1da177e4 LT |
736 | /* |
737 | * get a pointer to the on-disk dquot and the buffer containing it | |
738 | * dqp already knows its own type (GROUP/USER). | |
739 | */ | |
97e7ade5 CH |
740 | error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags); |
741 | if (error) { | |
742 | /* | |
743 | * This can happen if quotas got turned off (ESRCH), | |
744 | * or if the dquot didn't exist on disk and we ask to | |
745 | * allocate (ENOENT). | |
746 | */ | |
747 | trace_xfs_dqread_fail(dqp); | |
748 | cancelflags |= XFS_TRANS_ABORT; | |
749 | goto error1; | |
1da177e4 LT |
750 | } |
751 | ||
752 | /* copy everything from disk dquot to the incore dquot */ | |
753 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
1da177e4 LT |
754 | xfs_qm_dquot_logitem_init(dqp); |
755 | ||
756 | /* | |
757 | * Reservation counters are defined as reservation plus current usage | |
25985edc | 758 | * to avoid having to add every time. |
1da177e4 | 759 | */ |
1149d96a CH |
760 | dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); |
761 | dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); | |
762 | dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); | |
1da177e4 | 763 | |
b1366451 BF |
764 | /* initialize the dquot speculative prealloc thresholds */ |
765 | xfs_dquot_set_prealloc_limits(dqp); | |
766 | ||
1da177e4 | 767 | /* Mark the buf so that this will stay incore a little longer */ |
38f23232 | 768 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 LT |
769 | |
770 | /* | |
771 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
772 | * So we need to release with xfs_trans_brelse(). | |
773 | * The strategy here is identical to that of inodes; we lock | |
774 | * the dquot in xfs_qm_dqget() before making it accessible to | |
775 | * others. This is because dquots, like inodes, need a good level of | |
776 | * concurrency, and we don't want to take locks on the entire buffers | |
777 | * for dquot accesses. | |
778 | * Note also that the dquot buffer may even be dirty at this point, if | |
779 | * this particular dquot was repaired. We still aren't afraid to | |
780 | * brelse it because we have the changes incore. | |
781 | */ | |
0c842ad4 | 782 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
783 | xfs_trans_brelse(tp, bp); |
784 | ||
1da177e4 | 785 | if (tp) { |
97e7ade5 CH |
786 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
787 | if (error) | |
788 | goto error0; | |
1da177e4 LT |
789 | } |
790 | ||
791 | *O_dqpp = dqp; | |
97e7ade5 | 792 | return error; |
1da177e4 | 793 | |
97e7ade5 | 794 | error1: |
1da177e4 LT |
795 | if (tp) |
796 | xfs_trans_cancel(tp, cancelflags); | |
97e7ade5 | 797 | error0: |
1da177e4 LT |
798 | xfs_qm_dqdestroy(dqp); |
799 | *O_dqpp = NULL; | |
97e7ade5 | 800 | return error; |
1da177e4 LT |
801 | } |
802 | ||
1da177e4 LT |
803 | /* |
804 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
805 | * a locked dquot, doing an allocation (if requested) as needed. | |
806 | * When both an inode and an id are given, the inode's id takes precedence. | |
807 | * That is, if the id changes while we don't hold the ilock inside this | |
808 | * function, the new dquot is returned, not necessarily the one requested | |
809 | * in the id argument. | |
810 | */ | |
811 | int | |
812 | xfs_qm_dqget( | |
813 | xfs_mount_t *mp, | |
814 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
815 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
816 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
817 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
818 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
819 | { | |
9f920f11 | 820 | struct xfs_quotainfo *qi = mp->m_quotainfo; |
329e0875 | 821 | struct radix_tree_root *tree = xfs_dquot_tree(qi, type); |
9f920f11 CH |
822 | struct xfs_dquot *dqp; |
823 | int error; | |
1da177e4 LT |
824 | |
825 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
826 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 827 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 LT |
828 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
829 | return (ESRCH); | |
830 | } | |
1da177e4 LT |
831 | |
832 | #ifdef DEBUG | |
833 | if (xfs_do_dqerror) { | |
834 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
835 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 836 | xfs_debug(mp, "Returning error in dqget"); |
1da177e4 LT |
837 | return (EIO); |
838 | } | |
839 | } | |
1da177e4 | 840 | |
c8ad20ff NS |
841 | ASSERT(type == XFS_DQ_USER || |
842 | type == XFS_DQ_PROJ || | |
843 | type == XFS_DQ_GROUP); | |
1da177e4 | 844 | if (ip) { |
579aa9ca | 845 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
36731410 | 846 | ASSERT(xfs_inode_dquot(ip, type) == NULL); |
1da177e4 LT |
847 | } |
848 | #endif | |
92678554 CH |
849 | |
850 | restart: | |
9f920f11 CH |
851 | mutex_lock(&qi->qi_tree_lock); |
852 | dqp = radix_tree_lookup(tree, id); | |
853 | if (dqp) { | |
854 | xfs_dqlock(dqp); | |
855 | if (dqp->dq_flags & XFS_DQ_FREEING) { | |
856 | xfs_dqunlock(dqp); | |
857 | mutex_unlock(&qi->qi_tree_lock); | |
858 | trace_xfs_dqget_freeing(dqp); | |
859 | delay(1); | |
860 | goto restart; | |
861 | } | |
1da177e4 | 862 | |
9f920f11 CH |
863 | dqp->q_nrefs++; |
864 | mutex_unlock(&qi->qi_tree_lock); | |
865 | ||
866 | trace_xfs_dqget_hit(dqp); | |
48776fd2 | 867 | XFS_STATS_INC(xs_qm_dqcachehits); |
9f920f11 CH |
868 | *O_dqpp = dqp; |
869 | return 0; | |
1da177e4 | 870 | } |
9f920f11 CH |
871 | mutex_unlock(&qi->qi_tree_lock); |
872 | XFS_STATS_INC(xs_qm_dqcachemisses); | |
1da177e4 LT |
873 | |
874 | /* | |
875 | * Dquot cache miss. We don't want to keep the inode lock across | |
876 | * a (potential) disk read. Also we don't want to deal with the lock | |
877 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
878 | * lock here means dealing with a chown that can happen before | |
879 | * we re-acquire the lock. | |
880 | */ | |
881 | if (ip) | |
882 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 | 883 | |
97e7ade5 | 884 | error = xfs_qm_dqread(mp, id, type, flags, &dqp); |
1da177e4 | 885 | |
7ae44407 CH |
886 | if (ip) |
887 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
888 | ||
889 | if (error) | |
890 | return error; | |
1da177e4 | 891 | |
1da177e4 | 892 | if (ip) { |
1da177e4 LT |
893 | /* |
894 | * A dquot could be attached to this inode by now, since | |
895 | * we had dropped the ilock. | |
896 | */ | |
36731410 | 897 | if (xfs_this_quota_on(mp, type)) { |
9f920f11 CH |
898 | struct xfs_dquot *dqp1; |
899 | ||
36731410 CS |
900 | dqp1 = xfs_inode_dquot(ip, type); |
901 | if (dqp1) { | |
1da177e4 | 902 | xfs_qm_dqdestroy(dqp); |
36731410 | 903 | dqp = dqp1; |
1da177e4 LT |
904 | xfs_dqlock(dqp); |
905 | goto dqret; | |
906 | } | |
907 | } else { | |
36731410 CS |
908 | /* inode stays locked on return */ |
909 | xfs_qm_dqdestroy(dqp); | |
910 | return XFS_ERROR(ESRCH); | |
1da177e4 LT |
911 | } |
912 | } | |
913 | ||
9f920f11 CH |
914 | mutex_lock(&qi->qi_tree_lock); |
915 | error = -radix_tree_insert(tree, id, dqp); | |
916 | if (unlikely(error)) { | |
917 | WARN_ON(error != EEXIST); | |
918 | ||
1da177e4 | 919 | /* |
9f920f11 CH |
920 | * Duplicate found. Just throw away the new dquot and start |
921 | * over. | |
1da177e4 | 922 | */ |
9f920f11 CH |
923 | mutex_unlock(&qi->qi_tree_lock); |
924 | trace_xfs_dqget_dup(dqp); | |
925 | xfs_qm_dqdestroy(dqp); | |
926 | XFS_STATS_INC(xs_qm_dquot_dups); | |
927 | goto restart; | |
1da177e4 LT |
928 | } |
929 | ||
1da177e4 LT |
930 | /* |
931 | * We return a locked dquot to the caller, with a reference taken | |
932 | */ | |
933 | xfs_dqlock(dqp); | |
934 | dqp->q_nrefs = 1; | |
935 | ||
9f920f11 | 936 | qi->qi_dquots++; |
9f920f11 CH |
937 | mutex_unlock(&qi->qi_tree_lock); |
938 | ||
1da177e4 | 939 | dqret: |
579aa9ca | 940 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 941 | trace_xfs_dqget_miss(dqp); |
1da177e4 LT |
942 | *O_dqpp = dqp; |
943 | return (0); | |
944 | } | |
945 | ||
946 | ||
f8739c3c CH |
947 | STATIC void |
948 | xfs_qm_dqput_final( | |
bf72de31 | 949 | struct xfs_dquot *dqp) |
1da177e4 | 950 | { |
f8739c3c | 951 | struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo; |
bf72de31 | 952 | struct xfs_dquot *gdqp; |
92f8ff73 | 953 | struct xfs_dquot *pdqp; |
1da177e4 | 954 | |
bf72de31 CH |
955 | trace_xfs_dqput_free(dqp); |
956 | ||
cd56a39a | 957 | if (list_lru_add(&qi->qi_lru, &dqp->q_lru)) |
48776fd2 | 958 | XFS_STATS_INC(xs_qm_dquot_unused); |
bf72de31 | 959 | |
1da177e4 | 960 | /* |
bf72de31 | 961 | * If we just added a udquot to the freelist, then we want to release |
92f8ff73 CS |
962 | * the gdquot/pdquot reference that it (probably) has. Otherwise it'll |
963 | * keep the gdquot/pdquot from getting reclaimed. | |
1da177e4 | 964 | */ |
bf72de31 CH |
965 | gdqp = dqp->q_gdquot; |
966 | if (gdqp) { | |
967 | xfs_dqlock(gdqp); | |
968 | dqp->q_gdquot = NULL; | |
1da177e4 | 969 | } |
92f8ff73 CS |
970 | |
971 | pdqp = dqp->q_pdquot; | |
972 | if (pdqp) { | |
973 | xfs_dqlock(pdqp); | |
974 | dqp->q_pdquot = NULL; | |
975 | } | |
bf72de31 | 976 | xfs_dqunlock(dqp); |
1da177e4 | 977 | |
bf72de31 | 978 | /* |
92f8ff73 | 979 | * If we had a group/project quota hint, release it now. |
bf72de31 | 980 | */ |
f8739c3c CH |
981 | if (gdqp) |
982 | xfs_qm_dqput(gdqp); | |
92f8ff73 CS |
983 | if (pdqp) |
984 | xfs_qm_dqput(pdqp); | |
f8739c3c CH |
985 | } |
986 | ||
987 | /* | |
988 | * Release a reference to the dquot (decrement ref-count) and unlock it. | |
989 | * | |
990 | * If there is a group quota attached to this dquot, carefully release that | |
991 | * too without tripping over deadlocks'n'stuff. | |
992 | */ | |
993 | void | |
994 | xfs_qm_dqput( | |
995 | struct xfs_dquot *dqp) | |
996 | { | |
997 | ASSERT(dqp->q_nrefs > 0); | |
998 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
999 | ||
1000 | trace_xfs_dqput(dqp); | |
1001 | ||
1002 | if (--dqp->q_nrefs > 0) | |
1003 | xfs_dqunlock(dqp); | |
1004 | else | |
1005 | xfs_qm_dqput_final(dqp); | |
1da177e4 LT |
1006 | } |
1007 | ||
1008 | /* | |
1009 | * Release a dquot. Flush it if dirty, then dqput() it. | |
1010 | * dquot must not be locked. | |
1011 | */ | |
1012 | void | |
1013 | xfs_qm_dqrele( | |
1014 | xfs_dquot_t *dqp) | |
1015 | { | |
7d095257 CH |
1016 | if (!dqp) |
1017 | return; | |
1018 | ||
0b1b213f | 1019 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
1020 | |
1021 | xfs_dqlock(dqp); | |
1022 | /* | |
1023 | * We don't care to flush it if the dquot is dirty here. | |
1024 | * That will create stutters that we want to avoid. | |
1025 | * Instead we do a delayed write when we try to reclaim | |
1026 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
1027 | */ | |
1028 | xfs_qm_dqput(dqp); | |
1029 | } | |
1030 | ||
ca30b2a7 CH |
1031 | /* |
1032 | * This is the dquot flushing I/O completion routine. It is called | |
1033 | * from interrupt level when the buffer containing the dquot is | |
1034 | * flushed to disk. It is responsible for removing the dquot logitem | |
1035 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
1036 | * flush lock. This behavior is very similar to that of inodes.. | |
1037 | */ | |
1038 | STATIC void | |
1039 | xfs_qm_dqflush_done( | |
1040 | struct xfs_buf *bp, | |
1041 | struct xfs_log_item *lip) | |
1042 | { | |
1043 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
1044 | xfs_dquot_t *dqp = qip->qli_dquot; | |
1045 | struct xfs_ail *ailp = lip->li_ailp; | |
1046 | ||
1047 | /* | |
1048 | * We only want to pull the item from the AIL if its | |
1049 | * location in the log has not changed since we started the flush. | |
1050 | * Thus, we only bother if the dquot's lsn has | |
1051 | * not changed. First we check the lsn outside the lock | |
1052 | * since it's cheaper, and then we recheck while | |
1053 | * holding the lock before removing the dquot from the AIL. | |
1054 | */ | |
1055 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
1056 | lip->li_lsn == qip->qli_flush_lsn) { | |
1057 | ||
1058 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
1059 | spin_lock(&ailp->xa_lock); | |
1060 | if (lip->li_lsn == qip->qli_flush_lsn) | |
04913fdd | 1061 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 CH |
1062 | else |
1063 | spin_unlock(&ailp->xa_lock); | |
1064 | } | |
1065 | ||
1066 | /* | |
1067 | * Release the dq's flush lock since we're done with it. | |
1068 | */ | |
1069 | xfs_dqfunlock(dqp); | |
1070 | } | |
1da177e4 LT |
1071 | |
1072 | /* | |
1073 | * Write a modified dquot to disk. | |
1074 | * The dquot must be locked and the flush lock too taken by caller. | |
1075 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
1076 | * but the dquot is free to be unlocked and modified by the caller | |
1077 | * in the interim. Dquot is still locked on return. This behavior is | |
1078 | * identical to that of inodes. | |
1079 | */ | |
1080 | int | |
1081 | xfs_qm_dqflush( | |
fe7257fd CH |
1082 | struct xfs_dquot *dqp, |
1083 | struct xfs_buf **bpp) | |
1da177e4 | 1084 | { |
acecf1b5 CH |
1085 | struct xfs_mount *mp = dqp->q_mount; |
1086 | struct xfs_buf *bp; | |
1087 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 1088 | int error; |
1da177e4 LT |
1089 | |
1090 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 1091 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 1092 | |
0b1b213f | 1093 | trace_xfs_dqflush(dqp); |
1da177e4 | 1094 | |
fe7257fd CH |
1095 | *bpp = NULL; |
1096 | ||
1da177e4 LT |
1097 | xfs_qm_dqunpin_wait(dqp); |
1098 | ||
1099 | /* | |
1100 | * This may have been unpinned because the filesystem is shutting | |
1101 | * down forcibly. If that's the case we must not write this dquot | |
dea96095 CH |
1102 | * to disk, because the log record didn't make it to disk. |
1103 | * | |
1104 | * We also have to remove the log item from the AIL in this case, | |
1105 | * as we wait for an emptry AIL as part of the unmount process. | |
1da177e4 | 1106 | */ |
acecf1b5 | 1107 | if (XFS_FORCED_SHUTDOWN(mp)) { |
dea96095 | 1108 | struct xfs_log_item *lip = &dqp->q_logitem.qli_item; |
acecf1b5 | 1109 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
dea96095 CH |
1110 | |
1111 | spin_lock(&mp->m_ail->xa_lock); | |
1112 | if (lip->li_flags & XFS_LI_IN_AIL) | |
04913fdd DC |
1113 | xfs_trans_ail_delete(mp->m_ail, lip, |
1114 | SHUTDOWN_CORRUPT_INCORE); | |
dea96095 CH |
1115 | else |
1116 | spin_unlock(&mp->m_ail->xa_lock); | |
fe7257fd CH |
1117 | error = XFS_ERROR(EIO); |
1118 | goto out_unlock; | |
1da177e4 LT |
1119 | } |
1120 | ||
1121 | /* | |
1122 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1123 | */ |
acecf1b5 | 1124 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
c3f8fc73 | 1125 | mp->m_quotainfo->qi_dqchunklen, 0, &bp, NULL); |
fe7257fd CH |
1126 | if (error) |
1127 | goto out_unlock; | |
1da177e4 | 1128 | |
acecf1b5 CH |
1129 | /* |
1130 | * Calculate the location of the dquot inside the buffer. | |
1131 | */ | |
62926044 | 1132 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1133 | |
1134 | /* | |
1135 | * A simple sanity check in case we got a corrupted dquot.. | |
1136 | */ | |
a0fa2b67 DC |
1137 | error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
1138 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); | |
1139 | if (error) { | |
acecf1b5 CH |
1140 | xfs_buf_relse(bp); |
1141 | xfs_dqfunlock(dqp); | |
1142 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
1da177e4 LT |
1143 | return XFS_ERROR(EIO); |
1144 | } | |
1145 | ||
1146 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1147 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1148 | |
1149 | /* | |
1150 | * Clear the dirty field and remember the flush lsn for later use. | |
1151 | */ | |
acecf1b5 | 1152 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1153 | |
7b2e2a31 DC |
1154 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1155 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 | 1156 | |
3fe58f30 CH |
1157 | /* |
1158 | * copy the lsn into the on-disk dquot now while we have the in memory | |
1159 | * dquot here. This can't be done later in the write verifier as we | |
1160 | * can't get access to the log item at that point in time. | |
6fcdc59d DC |
1161 | * |
1162 | * We also calculate the CRC here so that the on-disk dquot in the | |
1163 | * buffer always has a valid CRC. This ensures there is no possibility | |
1164 | * of a dquot without an up-to-date CRC getting to disk. | |
3fe58f30 CH |
1165 | */ |
1166 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
1167 | struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp; | |
1168 | ||
1169 | dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn); | |
6fcdc59d DC |
1170 | xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk), |
1171 | XFS_DQUOT_CRC_OFF); | |
3fe58f30 CH |
1172 | } |
1173 | ||
1da177e4 LT |
1174 | /* |
1175 | * Attach an iodone routine so that we can remove this dquot from the | |
1176 | * AIL and release the flush lock once the dquot is synced to disk. | |
1177 | */ | |
ca30b2a7 CH |
1178 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1179 | &dqp->q_logitem.qli_item); | |
1180 | ||
1da177e4 LT |
1181 | /* |
1182 | * If the buffer is pinned then push on the log so we won't | |
1183 | * get stuck waiting in the write for too long. | |
1184 | */ | |
811e64c7 | 1185 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1186 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1187 | xfs_log_force(mp, 0); |
1da177e4 LT |
1188 | } |
1189 | ||
0b1b213f | 1190 | trace_xfs_dqflush_done(dqp); |
fe7257fd CH |
1191 | *bpp = bp; |
1192 | return 0; | |
0b1b213f | 1193 | |
fe7257fd CH |
1194 | out_unlock: |
1195 | xfs_dqfunlock(dqp); | |
1196 | return XFS_ERROR(EIO); | |
1da177e4 LT |
1197 | } |
1198 | ||
5bb87a33 CH |
1199 | /* |
1200 | * Lock two xfs_dquot structures. | |
1201 | * | |
1202 | * To avoid deadlocks we always lock the quota structure with | |
1203 | * the lowerd id first. | |
1204 | */ | |
1da177e4 LT |
1205 | void |
1206 | xfs_dqlock2( | |
1207 | xfs_dquot_t *d1, | |
1208 | xfs_dquot_t *d2) | |
1209 | { | |
1210 | if (d1 && d2) { | |
1211 | ASSERT(d1 != d2); | |
1149d96a CH |
1212 | if (be32_to_cpu(d1->q_core.d_id) > |
1213 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1214 | mutex_lock(&d2->q_qlock); |
1215 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1216 | } else { |
5bb87a33 CH |
1217 | mutex_lock(&d1->q_qlock); |
1218 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1219 | } |
5bb87a33 CH |
1220 | } else if (d1) { |
1221 | mutex_lock(&d1->q_qlock); | |
1222 | } else if (d2) { | |
1223 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1224 | } |
1225 | } | |
1226 | ||
a05931ce CH |
1227 | int __init |
1228 | xfs_qm_init(void) | |
1229 | { | |
1230 | xfs_qm_dqzone = | |
1231 | kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot"); | |
1232 | if (!xfs_qm_dqzone) | |
1233 | goto out; | |
1234 | ||
1235 | xfs_qm_dqtrxzone = | |
1236 | kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx"); | |
1237 | if (!xfs_qm_dqtrxzone) | |
1238 | goto out_free_dqzone; | |
1239 | ||
1240 | return 0; | |
1241 | ||
1242 | out_free_dqzone: | |
1243 | kmem_zone_destroy(xfs_qm_dqzone); | |
1244 | out: | |
1245 | return -ENOMEM; | |
1246 | } | |
1247 | ||
1c2ccc66 | 1248 | void |
a05931ce CH |
1249 | xfs_qm_exit(void) |
1250 | { | |
1251 | kmem_zone_destroy(xfs_qm_dqtrxzone); | |
1252 | kmem_zone_destroy(xfs_qm_dqzone); | |
1253 | } |