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