Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
4ce3121f NS |
2 | * Copyright (c) 2000-2003 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
4ce3121f NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
4ce3121f NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
4ce3121f NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 LT |
18 | #include "xfs.h" |
19 | #include "xfs_fs.h" | |
a844f451 | 20 | #include "xfs_bit.h" |
1da177e4 | 21 | #include "xfs_log.h" |
a844f451 | 22 | #include "xfs_inum.h" |
1da177e4 LT |
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" | |
1da177e4 LT |
32 | #include "xfs_rtalloc.h" |
33 | #include "xfs_error.h" | |
34 | #include "xfs_itable.h" | |
1da177e4 LT |
35 | #include "xfs_attr.h" |
36 | #include "xfs_buf_item.h" | |
37 | #include "xfs_trans_space.h" | |
38 | #include "xfs_trans_priv.h" | |
1da177e4 | 39 | #include "xfs_qm.h" |
0b1b213f | 40 | #include "xfs_trace.h" |
1da177e4 | 41 | |
1da177e4 | 42 | /* |
bf72de31 CH |
43 | * Lock order: |
44 | * | |
45 | * ip->i_lock | |
46 | * qh->qh_lock | |
47 | * qi->qi_dqlist_lock | |
48 | * dquot->q_qlock (xfs_dqlock() and friends) | |
49 | * dquot->q_flush (xfs_dqflock() and friends) | |
50 | * xfs_Gqm->qm_dqfrlist_lock | |
51 | * | |
52 | * If two dquots need to be locked the order is user before group/project, | |
53 | * otherwise by the lowest id first, see xfs_dqlock2. | |
54 | */ | |
1da177e4 | 55 | |
1da177e4 LT |
56 | #ifdef DEBUG |
57 | xfs_buftarg_t *xfs_dqerror_target; | |
58 | int xfs_do_dqerror; | |
59 | int xfs_dqreq_num; | |
60 | int xfs_dqerror_mod = 33; | |
61 | #endif | |
62 | ||
98b8c7a0 CH |
63 | static struct lock_class_key xfs_dquot_other_class; |
64 | ||
1da177e4 LT |
65 | /* |
66 | * This is called to free all the memory associated with a dquot | |
67 | */ | |
68 | void | |
69 | xfs_qm_dqdestroy( | |
70 | xfs_dquot_t *dqp) | |
71 | { | |
3a8406f6 | 72 | ASSERT(list_empty(&dqp->q_freelist)); |
1da177e4 LT |
73 | |
74 | mutex_destroy(&dqp->q_qlock); | |
1da177e4 | 75 | kmem_zone_free(xfs_Gqm->qm_dqzone, dqp); |
0b1b213f | 76 | |
48776fd2 | 77 | XFS_STATS_DEC(xs_qm_dquot); |
1da177e4 LT |
78 | } |
79 | ||
1da177e4 LT |
80 | /* |
81 | * If default limits are in force, push them into the dquot now. | |
82 | * We overwrite the dquot limits only if they are zero and this | |
83 | * is not the root dquot. | |
84 | */ | |
85 | void | |
86 | xfs_qm_adjust_dqlimits( | |
87 | xfs_mount_t *mp, | |
88 | xfs_disk_dquot_t *d) | |
89 | { | |
90 | xfs_quotainfo_t *q = mp->m_quotainfo; | |
91 | ||
92 | ASSERT(d->d_id); | |
93 | ||
94 | if (q->qi_bsoftlimit && !d->d_blk_softlimit) | |
1149d96a | 95 | d->d_blk_softlimit = cpu_to_be64(q->qi_bsoftlimit); |
1da177e4 | 96 | if (q->qi_bhardlimit && !d->d_blk_hardlimit) |
1149d96a | 97 | d->d_blk_hardlimit = cpu_to_be64(q->qi_bhardlimit); |
1da177e4 | 98 | if (q->qi_isoftlimit && !d->d_ino_softlimit) |
1149d96a | 99 | d->d_ino_softlimit = cpu_to_be64(q->qi_isoftlimit); |
1da177e4 | 100 | if (q->qi_ihardlimit && !d->d_ino_hardlimit) |
1149d96a | 101 | d->d_ino_hardlimit = cpu_to_be64(q->qi_ihardlimit); |
1da177e4 | 102 | if (q->qi_rtbsoftlimit && !d->d_rtb_softlimit) |
1149d96a | 103 | d->d_rtb_softlimit = cpu_to_be64(q->qi_rtbsoftlimit); |
1da177e4 | 104 | if (q->qi_rtbhardlimit && !d->d_rtb_hardlimit) |
1149d96a | 105 | d->d_rtb_hardlimit = cpu_to_be64(q->qi_rtbhardlimit); |
1da177e4 LT |
106 | } |
107 | ||
108 | /* | |
109 | * Check the limits and timers of a dquot and start or reset timers | |
110 | * if necessary. | |
111 | * This gets called even when quota enforcement is OFF, which makes our | |
112 | * life a little less complicated. (We just don't reject any quota | |
113 | * reservations in that case, when enforcement is off). | |
114 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
115 | * enforcement's off. | |
116 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
117 | * 'automatically' get started when limits get exceeded. They do |
118 | * get reset to zero, however, when we find the count to be under | |
119 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
120 | */ |
121 | void | |
122 | xfs_qm_adjust_dqtimers( | |
123 | xfs_mount_t *mp, | |
124 | xfs_disk_dquot_t *d) | |
125 | { | |
126 | ASSERT(d->d_id); | |
127 | ||
ea15ab3c | 128 | #ifdef DEBUG |
1149d96a CH |
129 | if (d->d_blk_hardlimit) |
130 | ASSERT(be64_to_cpu(d->d_blk_softlimit) <= | |
131 | be64_to_cpu(d->d_blk_hardlimit)); | |
132 | if (d->d_ino_hardlimit) | |
133 | ASSERT(be64_to_cpu(d->d_ino_softlimit) <= | |
134 | be64_to_cpu(d->d_ino_hardlimit)); | |
135 | if (d->d_rtb_hardlimit) | |
136 | ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= | |
137 | be64_to_cpu(d->d_rtb_hardlimit)); | |
1da177e4 | 138 | #endif |
ea15ab3c | 139 | |
1da177e4 | 140 | if (!d->d_btimer) { |
1149d96a | 141 | if ((d->d_blk_softlimit && |
d0a3fe67 | 142 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
143 | be64_to_cpu(d->d_blk_softlimit))) || |
144 | (d->d_blk_hardlimit && | |
d0a3fe67 | 145 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
146 | be64_to_cpu(d->d_blk_hardlimit)))) { |
147 | d->d_btimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 148 | mp->m_quotainfo->qi_btimelimit); |
754002b4 NS |
149 | } else { |
150 | d->d_bwarns = 0; | |
1da177e4 LT |
151 | } |
152 | } else { | |
153 | if ((!d->d_blk_softlimit || | |
d0a3fe67 | 154 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 155 | be64_to_cpu(d->d_blk_softlimit))) && |
1da177e4 | 156 | (!d->d_blk_hardlimit || |
d0a3fe67 | 157 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 158 | be64_to_cpu(d->d_blk_hardlimit)))) { |
1da177e4 LT |
159 | d->d_btimer = 0; |
160 | } | |
161 | } | |
162 | ||
163 | if (!d->d_itimer) { | |
1149d96a | 164 | if ((d->d_ino_softlimit && |
d0a3fe67 | 165 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
166 | be64_to_cpu(d->d_ino_softlimit))) || |
167 | (d->d_ino_hardlimit && | |
d0a3fe67 | 168 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
169 | be64_to_cpu(d->d_ino_hardlimit)))) { |
170 | d->d_itimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 171 | mp->m_quotainfo->qi_itimelimit); |
754002b4 NS |
172 | } else { |
173 | d->d_iwarns = 0; | |
1da177e4 LT |
174 | } |
175 | } else { | |
176 | if ((!d->d_ino_softlimit || | |
d0a3fe67 | 177 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 178 | be64_to_cpu(d->d_ino_softlimit))) && |
1da177e4 | 179 | (!d->d_ino_hardlimit || |
d0a3fe67 | 180 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 181 | be64_to_cpu(d->d_ino_hardlimit)))) { |
1da177e4 LT |
182 | d->d_itimer = 0; |
183 | } | |
184 | } | |
185 | ||
186 | if (!d->d_rtbtimer) { | |
1149d96a | 187 | if ((d->d_rtb_softlimit && |
d0a3fe67 | 188 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
189 | be64_to_cpu(d->d_rtb_softlimit))) || |
190 | (d->d_rtb_hardlimit && | |
d0a3fe67 | 191 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
192 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
193 | d->d_rtbtimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 194 | mp->m_quotainfo->qi_rtbtimelimit); |
754002b4 NS |
195 | } else { |
196 | d->d_rtbwarns = 0; | |
1da177e4 LT |
197 | } |
198 | } else { | |
199 | if ((!d->d_rtb_softlimit || | |
d0a3fe67 | 200 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 201 | be64_to_cpu(d->d_rtb_softlimit))) && |
1da177e4 | 202 | (!d->d_rtb_hardlimit || |
d0a3fe67 | 203 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 204 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
1da177e4 LT |
205 | d->d_rtbtimer = 0; |
206 | } | |
207 | } | |
208 | } | |
209 | ||
1da177e4 LT |
210 | /* |
211 | * initialize a buffer full of dquots and log the whole thing | |
212 | */ | |
213 | STATIC void | |
214 | xfs_qm_init_dquot_blk( | |
215 | xfs_trans_t *tp, | |
216 | xfs_mount_t *mp, | |
217 | xfs_dqid_t id, | |
218 | uint type, | |
219 | xfs_buf_t *bp) | |
220 | { | |
8a7b8a89 | 221 | struct xfs_quotainfo *q = mp->m_quotainfo; |
1da177e4 LT |
222 | xfs_dqblk_t *d; |
223 | int curid, i; | |
224 | ||
225 | ASSERT(tp); | |
0c842ad4 | 226 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 227 | |
62926044 | 228 | d = bp->b_addr; |
1da177e4 LT |
229 | |
230 | /* | |
231 | * ID of the first dquot in the block - id's are zero based. | |
232 | */ | |
8a7b8a89 | 233 | curid = id - (id % q->qi_dqperchunk); |
1da177e4 | 234 | ASSERT(curid >= 0); |
8a7b8a89 | 235 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
49d35a5c CH |
236 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) { |
237 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
238 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
239 | d->dd_diskdq.d_id = cpu_to_be32(curid); | |
240 | d->dd_diskdq.d_flags = type; | |
241 | } | |
242 | ||
1da177e4 | 243 | xfs_trans_dquot_buf(tp, bp, |
c1155410 DC |
244 | (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : |
245 | ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : | |
246 | XFS_BLF_GDQUOT_BUF))); | |
8a7b8a89 | 247 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); |
1da177e4 LT |
248 | } |
249 | ||
250 | ||
251 | ||
252 | /* | |
253 | * Allocate a block and fill it with dquots. | |
254 | * This is called when the bmapi finds a hole. | |
255 | */ | |
256 | STATIC int | |
257 | xfs_qm_dqalloc( | |
efa092f3 | 258 | xfs_trans_t **tpp, |
1da177e4 LT |
259 | xfs_mount_t *mp, |
260 | xfs_dquot_t *dqp, | |
261 | xfs_inode_t *quotip, | |
262 | xfs_fileoff_t offset_fsb, | |
263 | xfs_buf_t **O_bpp) | |
264 | { | |
265 | xfs_fsblock_t firstblock; | |
266 | xfs_bmap_free_t flist; | |
267 | xfs_bmbt_irec_t map; | |
268 | int nmaps, error, committed; | |
269 | xfs_buf_t *bp; | |
efa092f3 | 270 | xfs_trans_t *tp = *tpp; |
1da177e4 LT |
271 | |
272 | ASSERT(tp != NULL); | |
0b1b213f CH |
273 | |
274 | trace_xfs_dqalloc(dqp); | |
1da177e4 LT |
275 | |
276 | /* | |
277 | * Initialize the bmap freelist prior to calling bmapi code. | |
278 | */ | |
9d87c319 | 279 | xfs_bmap_init(&flist, &firstblock); |
1da177e4 LT |
280 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
281 | /* | |
282 | * Return if this type of quotas is turned off while we didn't | |
283 | * have an inode lock | |
284 | */ | |
6967b964 | 285 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 LT |
286 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
287 | return (ESRCH); | |
288 | } | |
289 | ||
ddc3415a | 290 | xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); |
1da177e4 | 291 | nmaps = 1; |
c0dc7828 DC |
292 | error = xfs_bmapi_write(tp, quotip, offset_fsb, |
293 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
294 | &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), | |
295 | &map, &nmaps, &flist); | |
296 | if (error) | |
1da177e4 | 297 | goto error0; |
1da177e4 LT |
298 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
299 | ASSERT(nmaps == 1); | |
300 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
301 | (map.br_startblock != HOLESTARTBLOCK)); | |
302 | ||
303 | /* | |
304 | * Keep track of the blkno to save a lookup later | |
305 | */ | |
306 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
307 | ||
308 | /* now we can just get the buffer (there's nothing to read yet) */ | |
309 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
310 | dqp->q_blkno, | |
8a7b8a89 | 311 | mp->m_quotainfo->qi_dqchunklen, |
1da177e4 | 312 | 0); |
2a30f36d CS |
313 | |
314 | error = xfs_buf_geterror(bp); | |
315 | if (error) | |
1da177e4 | 316 | goto error1; |
2a30f36d | 317 | |
1da177e4 LT |
318 | /* |
319 | * Make a chunk of dquots out of this buffer and log | |
320 | * the entire thing. | |
321 | */ | |
1149d96a | 322 | xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 323 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
1da177e4 | 324 | |
efa092f3 TS |
325 | /* |
326 | * xfs_bmap_finish() may commit the current transaction and | |
327 | * start a second transaction if the freelist is not empty. | |
328 | * | |
329 | * Since we still want to modify this buffer, we need to | |
330 | * ensure that the buffer is not released on commit of | |
331 | * the first transaction and ensure the buffer is added to the | |
332 | * second transaction. | |
333 | * | |
334 | * If there is only one transaction then don't stop the buffer | |
335 | * from being released when it commits later on. | |
336 | */ | |
337 | ||
338 | xfs_trans_bhold(tp, bp); | |
339 | ||
f7c99b6f | 340 | if ((error = xfs_bmap_finish(tpp, &flist, &committed))) { |
1da177e4 LT |
341 | goto error1; |
342 | } | |
343 | ||
efa092f3 TS |
344 | if (committed) { |
345 | tp = *tpp; | |
346 | xfs_trans_bjoin(tp, bp); | |
347 | } else { | |
348 | xfs_trans_bhold_release(tp, bp); | |
349 | } | |
350 | ||
1da177e4 LT |
351 | *O_bpp = bp; |
352 | return 0; | |
353 | ||
354 | error1: | |
355 | xfs_bmap_cancel(&flist); | |
356 | error0: | |
357 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); | |
358 | ||
359 | return (error); | |
360 | } | |
361 | ||
362 | /* | |
363 | * Maps a dquot to the buffer containing its on-disk version. | |
364 | * This returns a ptr to the buffer containing the on-disk dquot | |
365 | * in the bpp param, and a ptr to the on-disk dquot within that buffer | |
366 | */ | |
367 | STATIC int | |
368 | xfs_qm_dqtobp( | |
efa092f3 | 369 | xfs_trans_t **tpp, |
1da177e4 LT |
370 | xfs_dquot_t *dqp, |
371 | xfs_disk_dquot_t **O_ddpp, | |
372 | xfs_buf_t **O_bpp, | |
373 | uint flags) | |
374 | { | |
375 | xfs_bmbt_irec_t map; | |
acecf1b5 | 376 | int nmaps = 1, error; |
1da177e4 | 377 | xfs_buf_t *bp; |
acecf1b5 CH |
378 | xfs_inode_t *quotip = XFS_DQ_TO_QIP(dqp); |
379 | xfs_mount_t *mp = dqp->q_mount; | |
1da177e4 | 380 | xfs_disk_dquot_t *ddq; |
acecf1b5 | 381 | xfs_dqid_t id = be32_to_cpu(dqp->q_core.d_id); |
efa092f3 | 382 | xfs_trans_t *tp = (tpp ? *tpp : NULL); |
1da177e4 | 383 | |
acecf1b5 | 384 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
1da177e4 | 385 | |
acecf1b5 | 386 | xfs_ilock(quotip, XFS_ILOCK_SHARED); |
6967b964 | 387 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
1da177e4 | 388 | /* |
acecf1b5 CH |
389 | * Return if this type of quotas is turned off while we |
390 | * didn't have the quota inode lock. | |
1da177e4 | 391 | */ |
acecf1b5 CH |
392 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); |
393 | return ESRCH; | |
394 | } | |
395 | ||
396 | /* | |
397 | * Find the block map; no allocations yet | |
398 | */ | |
5c8ed202 DC |
399 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
400 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); | |
acecf1b5 CH |
401 | |
402 | xfs_iunlock(quotip, XFS_ILOCK_SHARED); | |
403 | if (error) | |
404 | return error; | |
405 | ||
406 | ASSERT(nmaps == 1); | |
407 | ASSERT(map.br_blockcount == 1); | |
408 | ||
409 | /* | |
410 | * Offset of dquot in the (fixed sized) dquot chunk. | |
411 | */ | |
412 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
413 | sizeof(xfs_dqblk_t); | |
414 | ||
415 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
416 | if (map.br_startblock == HOLESTARTBLOCK) { | |
1da177e4 | 417 | /* |
acecf1b5 | 418 | * We don't allocate unless we're asked to |
1da177e4 | 419 | */ |
acecf1b5 CH |
420 | if (!(flags & XFS_QMOPT_DQALLOC)) |
421 | return ENOENT; | |
1da177e4 | 422 | |
acecf1b5 CH |
423 | ASSERT(tp); |
424 | error = xfs_qm_dqalloc(tpp, mp, dqp, quotip, | |
425 | dqp->q_fileoffset, &bp); | |
1da177e4 | 426 | if (error) |
acecf1b5 CH |
427 | return error; |
428 | tp = *tpp; | |
429 | } else { | |
430 | trace_xfs_dqtobp_read(dqp); | |
1da177e4 LT |
431 | |
432 | /* | |
acecf1b5 CH |
433 | * store the blkno etc so that we don't have to do the |
434 | * mapping all the time | |
1da177e4 | 435 | */ |
acecf1b5 | 436 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
0b1b213f | 437 | |
8a7b8a89 CH |
438 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
439 | dqp->q_blkno, | |
440 | mp->m_quotainfo->qi_dqchunklen, | |
441 | 0, &bp); | |
1da177e4 LT |
442 | if (error || !bp) |
443 | return XFS_ERROR(error); | |
444 | } | |
acecf1b5 | 445 | |
0c842ad4 | 446 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
447 | |
448 | /* | |
449 | * calculate the location of the dquot inside the buffer. | |
450 | */ | |
62926044 | 451 | ddq = bp->b_addr + dqp->q_bufoffset; |
1da177e4 LT |
452 | |
453 | /* | |
454 | * A simple sanity check in case we got a corrupted dquot... | |
455 | */ | |
a0fa2b67 | 456 | error = xfs_qm_dqcheck(mp, ddq, id, dqp->dq_flags & XFS_DQ_ALLTYPES, |
1da177e4 | 457 | flags & (XFS_QMOPT_DQREPAIR|XFS_QMOPT_DOWARN), |
a0fa2b67 DC |
458 | "dqtobp"); |
459 | if (error) { | |
1da177e4 LT |
460 | if (!(flags & XFS_QMOPT_DQREPAIR)) { |
461 | xfs_trans_brelse(tp, bp); | |
462 | return XFS_ERROR(EIO); | |
463 | } | |
1da177e4 LT |
464 | } |
465 | ||
466 | *O_bpp = bp; | |
467 | *O_ddpp = ddq; | |
468 | ||
469 | return (0); | |
470 | } | |
471 | ||
472 | ||
473 | /* | |
474 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
475 | * and release the buffer immediately. | |
476 | * | |
97e7ade5 | 477 | * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed. |
1da177e4 | 478 | */ |
7ae44407 | 479 | int |
1da177e4 | 480 | xfs_qm_dqread( |
97e7ade5 CH |
481 | struct xfs_mount *mp, |
482 | xfs_dqid_t id, | |
483 | uint type, | |
484 | uint flags, | |
485 | struct xfs_dquot **O_dqpp) | |
1da177e4 | 486 | { |
97e7ade5 CH |
487 | struct xfs_dquot *dqp; |
488 | struct xfs_disk_dquot *ddqp; | |
489 | struct xfs_buf *bp; | |
490 | struct xfs_trans *tp = NULL; | |
491 | int error; | |
492 | int cancelflags = 0; | |
efa092f3 | 493 | |
92b2e5b3 CH |
494 | |
495 | dqp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP); | |
496 | ||
497 | dqp->dq_flags = type; | |
498 | dqp->q_core.d_id = cpu_to_be32(id); | |
499 | dqp->q_mount = mp; | |
500 | INIT_LIST_HEAD(&dqp->q_freelist); | |
501 | mutex_init(&dqp->q_qlock); | |
502 | init_waitqueue_head(&dqp->q_pinwait); | |
503 | ||
504 | /* | |
505 | * Because we want to use a counting completion, complete | |
506 | * the flush completion once to allow a single access to | |
507 | * the flush completion without blocking. | |
508 | */ | |
509 | init_completion(&dqp->q_flush); | |
510 | complete(&dqp->q_flush); | |
511 | ||
512 | /* | |
513 | * Make sure group quotas have a different lock class than user | |
514 | * quotas. | |
515 | */ | |
516 | if (!(type & XFS_DQ_USER)) | |
517 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class); | |
518 | ||
48776fd2 | 519 | XFS_STATS_INC(xs_qm_dquot); |
1da177e4 | 520 | |
0b1b213f CH |
521 | trace_xfs_dqread(dqp); |
522 | ||
97e7ade5 CH |
523 | if (flags & XFS_QMOPT_DQALLOC) { |
524 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_DQALLOC); | |
525 | error = xfs_trans_reserve(tp, XFS_QM_DQALLOC_SPACE_RES(mp), | |
526 | XFS_WRITE_LOG_RES(mp) + | |
527 | /* | |
528 | * Round the chunklen up to the next multiple | |
529 | * of 128 (buf log item chunk size)). | |
530 | */ | |
531 | BBTOB(mp->m_quotainfo->qi_dqchunklen) - 1 + 128, | |
532 | 0, | |
533 | XFS_TRANS_PERM_LOG_RES, | |
534 | XFS_WRITE_LOG_COUNT); | |
535 | if (error) | |
536 | goto error1; | |
537 | cancelflags = XFS_TRANS_RELEASE_LOG_RES; | |
538 | } | |
539 | ||
1da177e4 LT |
540 | /* |
541 | * get a pointer to the on-disk dquot and the buffer containing it | |
542 | * dqp already knows its own type (GROUP/USER). | |
543 | */ | |
97e7ade5 CH |
544 | error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags); |
545 | if (error) { | |
546 | /* | |
547 | * This can happen if quotas got turned off (ESRCH), | |
548 | * or if the dquot didn't exist on disk and we ask to | |
549 | * allocate (ENOENT). | |
550 | */ | |
551 | trace_xfs_dqread_fail(dqp); | |
552 | cancelflags |= XFS_TRANS_ABORT; | |
553 | goto error1; | |
1da177e4 LT |
554 | } |
555 | ||
556 | /* copy everything from disk dquot to the incore dquot */ | |
557 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
1da177e4 LT |
558 | xfs_qm_dquot_logitem_init(dqp); |
559 | ||
560 | /* | |
561 | * Reservation counters are defined as reservation plus current usage | |
25985edc | 562 | * to avoid having to add every time. |
1da177e4 | 563 | */ |
1149d96a CH |
564 | dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); |
565 | dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); | |
566 | dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); | |
1da177e4 LT |
567 | |
568 | /* Mark the buf so that this will stay incore a little longer */ | |
38f23232 | 569 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 LT |
570 | |
571 | /* | |
572 | * We got the buffer with a xfs_trans_read_buf() (in dqtobp()) | |
573 | * So we need to release with xfs_trans_brelse(). | |
574 | * The strategy here is identical to that of inodes; we lock | |
575 | * the dquot in xfs_qm_dqget() before making it accessible to | |
576 | * others. This is because dquots, like inodes, need a good level of | |
577 | * concurrency, and we don't want to take locks on the entire buffers | |
578 | * for dquot accesses. | |
579 | * Note also that the dquot buffer may even be dirty at this point, if | |
580 | * this particular dquot was repaired. We still aren't afraid to | |
581 | * brelse it because we have the changes incore. | |
582 | */ | |
0c842ad4 | 583 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
584 | xfs_trans_brelse(tp, bp); |
585 | ||
1da177e4 | 586 | if (tp) { |
97e7ade5 CH |
587 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
588 | if (error) | |
589 | goto error0; | |
1da177e4 LT |
590 | } |
591 | ||
592 | *O_dqpp = dqp; | |
97e7ade5 | 593 | return error; |
1da177e4 | 594 | |
97e7ade5 | 595 | error1: |
1da177e4 LT |
596 | if (tp) |
597 | xfs_trans_cancel(tp, cancelflags); | |
97e7ade5 | 598 | error0: |
1da177e4 LT |
599 | xfs_qm_dqdestroy(dqp); |
600 | *O_dqpp = NULL; | |
97e7ade5 | 601 | return error; |
1da177e4 LT |
602 | } |
603 | ||
604 | /* | |
605 | * Lookup a dquot in the incore dquot hashtable. We keep two separate | |
606 | * hashtables for user and group dquots; and, these are global tables | |
607 | * inside the XQM, not per-filesystem tables. | |
608 | * The hash chain must be locked by caller, and it is left locked | |
609 | * on return. Returning dquot is locked. | |
610 | */ | |
611 | STATIC int | |
612 | xfs_qm_dqlookup( | |
613 | xfs_mount_t *mp, | |
614 | xfs_dqid_t id, | |
615 | xfs_dqhash_t *qh, | |
616 | xfs_dquot_t **O_dqpp) | |
617 | { | |
618 | xfs_dquot_t *dqp; | |
1da177e4 | 619 | |
c9a192dc | 620 | ASSERT(mutex_is_locked(&qh->qh_lock)); |
1da177e4 | 621 | |
1da177e4 LT |
622 | /* |
623 | * Traverse the hashchain looking for a match | |
624 | */ | |
e6a81f13 | 625 | list_for_each_entry(dqp, &qh->qh_list, q_hashlist) { |
1da177e4 LT |
626 | /* |
627 | * We already have the hashlock. We don't need the | |
628 | * dqlock to look at the id field of the dquot, since the | |
629 | * id can't be modified without the hashlock anyway. | |
630 | */ | |
be7ffc38 CH |
631 | if (be32_to_cpu(dqp->q_core.d_id) != id || dqp->q_mount != mp) |
632 | continue; | |
0b1b213f | 633 | |
be7ffc38 | 634 | trace_xfs_dqlookup_found(dqp); |
1da177e4 | 635 | |
be7ffc38 | 636 | xfs_dqlock(dqp); |
92678554 CH |
637 | if (dqp->dq_flags & XFS_DQ_FREEING) { |
638 | *O_dqpp = NULL; | |
639 | xfs_dqunlock(dqp); | |
640 | return -1; | |
641 | } | |
642 | ||
78e55892 | 643 | dqp->q_nrefs++; |
be7ffc38 CH |
644 | |
645 | /* | |
646 | * move the dquot to the front of the hashchain | |
647 | */ | |
648 | list_move(&dqp->q_hashlist, &qh->qh_list); | |
649 | trace_xfs_dqlookup_done(dqp); | |
650 | *O_dqpp = dqp; | |
651 | return 0; | |
1da177e4 LT |
652 | } |
653 | ||
654 | *O_dqpp = NULL; | |
be7ffc38 | 655 | return 1; |
1da177e4 LT |
656 | } |
657 | ||
658 | /* | |
659 | * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a | |
660 | * a locked dquot, doing an allocation (if requested) as needed. | |
661 | * When both an inode and an id are given, the inode's id takes precedence. | |
662 | * That is, if the id changes while we don't hold the ilock inside this | |
663 | * function, the new dquot is returned, not necessarily the one requested | |
664 | * in the id argument. | |
665 | */ | |
666 | int | |
667 | xfs_qm_dqget( | |
668 | xfs_mount_t *mp, | |
669 | xfs_inode_t *ip, /* locked inode (optional) */ | |
c8ad20ff NS |
670 | xfs_dqid_t id, /* uid/projid/gid depending on type */ |
671 | uint type, /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */ | |
1da177e4 LT |
672 | uint flags, /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */ |
673 | xfs_dquot_t **O_dqpp) /* OUT : locked incore dquot */ | |
674 | { | |
36731410 | 675 | xfs_dquot_t *dqp, *dqp1; |
1da177e4 LT |
676 | xfs_dqhash_t *h; |
677 | uint version; | |
678 | int error; | |
679 | ||
680 | ASSERT(XFS_IS_QUOTA_RUNNING(mp)); | |
681 | if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) || | |
c8ad20ff | 682 | (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) || |
1da177e4 LT |
683 | (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) { |
684 | return (ESRCH); | |
685 | } | |
686 | h = XFS_DQ_HASH(mp, id, type); | |
687 | ||
688 | #ifdef DEBUG | |
689 | if (xfs_do_dqerror) { | |
690 | if ((xfs_dqerror_target == mp->m_ddev_targp) && | |
691 | (xfs_dqreq_num++ % xfs_dqerror_mod) == 0) { | |
0b932ccc | 692 | xfs_debug(mp, "Returning error in dqget"); |
1da177e4 LT |
693 | return (EIO); |
694 | } | |
695 | } | |
1da177e4 | 696 | |
c8ad20ff NS |
697 | ASSERT(type == XFS_DQ_USER || |
698 | type == XFS_DQ_PROJ || | |
699 | type == XFS_DQ_GROUP); | |
1da177e4 | 700 | if (ip) { |
579aa9ca | 701 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
36731410 | 702 | ASSERT(xfs_inode_dquot(ip, type) == NULL); |
1da177e4 LT |
703 | } |
704 | #endif | |
92678554 CH |
705 | |
706 | restart: | |
c9a192dc | 707 | mutex_lock(&h->qh_lock); |
1da177e4 LT |
708 | |
709 | /* | |
710 | * Look in the cache (hashtable). | |
711 | * The chain is kept locked during lookup. | |
712 | */ | |
92678554 CH |
713 | switch (xfs_qm_dqlookup(mp, id, h, O_dqpp)) { |
714 | case -1: | |
48776fd2 | 715 | XFS_STATS_INC(xs_qm_dquot_dups); |
92678554 CH |
716 | mutex_unlock(&h->qh_lock); |
717 | delay(1); | |
718 | goto restart; | |
719 | case 0: | |
48776fd2 | 720 | XFS_STATS_INC(xs_qm_dqcachehits); |
1da177e4 LT |
721 | /* |
722 | * The dquot was found, moved to the front of the chain, | |
723 | * taken off the freelist if it was on it, and locked | |
724 | * at this point. Just unlock the hashchain and return. | |
725 | */ | |
726 | ASSERT(*O_dqpp); | |
727 | ASSERT(XFS_DQ_IS_LOCKED(*O_dqpp)); | |
c9a192dc | 728 | mutex_unlock(&h->qh_lock); |
0b1b213f | 729 | trace_xfs_dqget_hit(*O_dqpp); |
92678554 CH |
730 | return 0; /* success */ |
731 | default: | |
48776fd2 | 732 | XFS_STATS_INC(xs_qm_dqcachemisses); |
92678554 | 733 | break; |
1da177e4 | 734 | } |
1da177e4 LT |
735 | |
736 | /* | |
737 | * Dquot cache miss. We don't want to keep the inode lock across | |
738 | * a (potential) disk read. Also we don't want to deal with the lock | |
739 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
740 | * lock here means dealing with a chown that can happen before | |
741 | * we re-acquire the lock. | |
742 | */ | |
743 | if (ip) | |
744 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
745 | /* | |
746 | * Save the hashchain version stamp, and unlock the chain, so that | |
747 | * we don't keep the lock across a disk read | |
748 | */ | |
749 | version = h->qh_version; | |
c9a192dc | 750 | mutex_unlock(&h->qh_lock); |
1da177e4 | 751 | |
97e7ade5 | 752 | error = xfs_qm_dqread(mp, id, type, flags, &dqp); |
1da177e4 | 753 | |
7ae44407 CH |
754 | if (ip) |
755 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
756 | ||
757 | if (error) | |
758 | return error; | |
1da177e4 LT |
759 | |
760 | /* | |
761 | * Dquot lock comes after hashlock in the lock ordering | |
762 | */ | |
763 | if (ip) { | |
1da177e4 LT |
764 | /* |
765 | * A dquot could be attached to this inode by now, since | |
766 | * we had dropped the ilock. | |
767 | */ | |
36731410 CS |
768 | if (xfs_this_quota_on(mp, type)) { |
769 | dqp1 = xfs_inode_dquot(ip, type); | |
770 | if (dqp1) { | |
1da177e4 | 771 | xfs_qm_dqdestroy(dqp); |
36731410 | 772 | dqp = dqp1; |
1da177e4 LT |
773 | xfs_dqlock(dqp); |
774 | goto dqret; | |
775 | } | |
776 | } else { | |
36731410 CS |
777 | /* inode stays locked on return */ |
778 | xfs_qm_dqdestroy(dqp); | |
779 | return XFS_ERROR(ESRCH); | |
1da177e4 LT |
780 | } |
781 | } | |
782 | ||
783 | /* | |
784 | * Hashlock comes after ilock in lock order | |
785 | */ | |
c9a192dc | 786 | mutex_lock(&h->qh_lock); |
1da177e4 LT |
787 | if (version != h->qh_version) { |
788 | xfs_dquot_t *tmpdqp; | |
789 | /* | |
790 | * Now, see if somebody else put the dquot in the | |
791 | * hashtable before us. This can happen because we didn't | |
792 | * keep the hashchain lock. We don't have to worry about | |
793 | * lock order between the two dquots here since dqp isn't | |
794 | * on any findable lists yet. | |
795 | */ | |
92678554 CH |
796 | switch (xfs_qm_dqlookup(mp, id, h, &tmpdqp)) { |
797 | case 0: | |
798 | case -1: | |
1da177e4 | 799 | /* |
92678554 CH |
800 | * Duplicate found, either in cache or on its way out. |
801 | * Just throw away the new dquot and start over. | |
1da177e4 | 802 | */ |
92678554 CH |
803 | if (tmpdqp) |
804 | xfs_qm_dqput(tmpdqp); | |
c9a192dc | 805 | mutex_unlock(&h->qh_lock); |
1da177e4 | 806 | xfs_qm_dqdestroy(dqp); |
48776fd2 | 807 | XFS_STATS_INC(xs_qm_dquot_dups); |
92678554 CH |
808 | goto restart; |
809 | default: | |
810 | break; | |
1da177e4 LT |
811 | } |
812 | } | |
813 | ||
814 | /* | |
815 | * Put the dquot at the beginning of the hash-chain and mp's list | |
816 | * LOCK ORDER: hashlock, freelistlock, mplistlock, udqlock, gdqlock .. | |
817 | */ | |
c9a192dc | 818 | ASSERT(mutex_is_locked(&h->qh_lock)); |
1da177e4 | 819 | dqp->q_hash = h; |
e6a81f13 DC |
820 | list_add(&dqp->q_hashlist, &h->qh_list); |
821 | h->qh_version++; | |
1da177e4 LT |
822 | |
823 | /* | |
824 | * Attach this dquot to this filesystem's list of all dquots, | |
825 | * kept inside the mount structure in m_quotainfo field | |
826 | */ | |
3a25404b | 827 | mutex_lock(&mp->m_quotainfo->qi_dqlist_lock); |
1da177e4 LT |
828 | |
829 | /* | |
830 | * We return a locked dquot to the caller, with a reference taken | |
831 | */ | |
832 | xfs_dqlock(dqp); | |
833 | dqp->q_nrefs = 1; | |
834 | ||
3a25404b DC |
835 | list_add(&dqp->q_mplist, &mp->m_quotainfo->qi_dqlist); |
836 | mp->m_quotainfo->qi_dquots++; | |
837 | mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock); | |
c9a192dc | 838 | mutex_unlock(&h->qh_lock); |
1da177e4 | 839 | dqret: |
579aa9ca | 840 | ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b1b213f | 841 | trace_xfs_dqget_miss(dqp); |
1da177e4 LT |
842 | *O_dqpp = dqp; |
843 | return (0); | |
844 | } | |
845 | ||
846 | ||
847 | /* | |
848 | * Release a reference to the dquot (decrement ref-count) | |
849 | * and unlock it. If there is a group quota attached to this | |
850 | * dquot, carefully release that too without tripping over | |
851 | * deadlocks'n'stuff. | |
852 | */ | |
853 | void | |
854 | xfs_qm_dqput( | |
bf72de31 | 855 | struct xfs_dquot *dqp) |
1da177e4 | 856 | { |
bf72de31 | 857 | struct xfs_dquot *gdqp; |
1da177e4 LT |
858 | |
859 | ASSERT(dqp->q_nrefs > 0); | |
860 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
0b1b213f CH |
861 | |
862 | trace_xfs_dqput(dqp); | |
1da177e4 | 863 | |
bf72de31 CH |
864 | recurse: |
865 | if (--dqp->q_nrefs > 0) { | |
1da177e4 LT |
866 | xfs_dqunlock(dqp); |
867 | return; | |
868 | } | |
869 | ||
bf72de31 CH |
870 | trace_xfs_dqput_free(dqp); |
871 | ||
872 | mutex_lock(&xfs_Gqm->qm_dqfrlist_lock); | |
873 | if (list_empty(&dqp->q_freelist)) { | |
874 | list_add_tail(&dqp->q_freelist, &xfs_Gqm->qm_dqfrlist); | |
875 | xfs_Gqm->qm_dqfrlist_cnt++; | |
48776fd2 | 876 | XFS_STATS_INC(xs_qm_dquot_unused); |
bf72de31 CH |
877 | } |
878 | mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); | |
879 | ||
1da177e4 | 880 | /* |
bf72de31 CH |
881 | * If we just added a udquot to the freelist, then we want to release |
882 | * the gdquot reference that it (probably) has. Otherwise it'll keep | |
883 | * the gdquot from getting reclaimed. | |
1da177e4 | 884 | */ |
bf72de31 CH |
885 | gdqp = dqp->q_gdquot; |
886 | if (gdqp) { | |
887 | xfs_dqlock(gdqp); | |
888 | dqp->q_gdquot = NULL; | |
1da177e4 | 889 | } |
bf72de31 | 890 | xfs_dqunlock(dqp); |
1da177e4 | 891 | |
bf72de31 CH |
892 | /* |
893 | * If we had a group quota hint, release it now. | |
894 | */ | |
895 | if (gdqp) { | |
1da177e4 | 896 | dqp = gdqp; |
bf72de31 | 897 | goto recurse; |
1da177e4 | 898 | } |
1da177e4 LT |
899 | } |
900 | ||
901 | /* | |
902 | * Release a dquot. Flush it if dirty, then dqput() it. | |
903 | * dquot must not be locked. | |
904 | */ | |
905 | void | |
906 | xfs_qm_dqrele( | |
907 | xfs_dquot_t *dqp) | |
908 | { | |
7d095257 CH |
909 | if (!dqp) |
910 | return; | |
911 | ||
0b1b213f | 912 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
913 | |
914 | xfs_dqlock(dqp); | |
915 | /* | |
916 | * We don't care to flush it if the dquot is dirty here. | |
917 | * That will create stutters that we want to avoid. | |
918 | * Instead we do a delayed write when we try to reclaim | |
919 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
920 | */ | |
921 | xfs_qm_dqput(dqp); | |
922 | } | |
923 | ||
ca30b2a7 CH |
924 | /* |
925 | * This is the dquot flushing I/O completion routine. It is called | |
926 | * from interrupt level when the buffer containing the dquot is | |
927 | * flushed to disk. It is responsible for removing the dquot logitem | |
928 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
929 | * flush lock. This behavior is very similar to that of inodes.. | |
930 | */ | |
931 | STATIC void | |
932 | xfs_qm_dqflush_done( | |
933 | struct xfs_buf *bp, | |
934 | struct xfs_log_item *lip) | |
935 | { | |
936 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
937 | xfs_dquot_t *dqp = qip->qli_dquot; | |
938 | struct xfs_ail *ailp = lip->li_ailp; | |
939 | ||
940 | /* | |
941 | * We only want to pull the item from the AIL if its | |
942 | * location in the log has not changed since we started the flush. | |
943 | * Thus, we only bother if the dquot's lsn has | |
944 | * not changed. First we check the lsn outside the lock | |
945 | * since it's cheaper, and then we recheck while | |
946 | * holding the lock before removing the dquot from the AIL. | |
947 | */ | |
948 | if ((lip->li_flags & XFS_LI_IN_AIL) && | |
949 | lip->li_lsn == qip->qli_flush_lsn) { | |
950 | ||
951 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
952 | spin_lock(&ailp->xa_lock); | |
953 | if (lip->li_lsn == qip->qli_flush_lsn) | |
954 | xfs_trans_ail_delete(ailp, lip); | |
955 | else | |
956 | spin_unlock(&ailp->xa_lock); | |
957 | } | |
958 | ||
959 | /* | |
960 | * Release the dq's flush lock since we're done with it. | |
961 | */ | |
962 | xfs_dqfunlock(dqp); | |
963 | } | |
1da177e4 LT |
964 | |
965 | /* | |
966 | * Write a modified dquot to disk. | |
967 | * The dquot must be locked and the flush lock too taken by caller. | |
968 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
969 | * but the dquot is free to be unlocked and modified by the caller | |
970 | * in the interim. Dquot is still locked on return. This behavior is | |
971 | * identical to that of inodes. | |
972 | */ | |
973 | int | |
974 | xfs_qm_dqflush( | |
975 | xfs_dquot_t *dqp, | |
976 | uint flags) | |
977 | { | |
acecf1b5 CH |
978 | struct xfs_mount *mp = dqp->q_mount; |
979 | struct xfs_buf *bp; | |
980 | struct xfs_disk_dquot *ddqp; | |
1da177e4 | 981 | int error; |
1da177e4 LT |
982 | |
983 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 984 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 985 | |
0b1b213f | 986 | trace_xfs_dqflush(dqp); |
1da177e4 LT |
987 | |
988 | /* | |
acecf1b5 | 989 | * If not dirty, or it's pinned and we are not supposed to block, nada. |
1da177e4 | 990 | */ |
2f8a3ce1 | 991 | if (!XFS_DQ_IS_DIRTY(dqp) || |
fdedf28b | 992 | ((flags & SYNC_TRYLOCK) && atomic_read(&dqp->q_pincount) > 0)) { |
1da177e4 | 993 | xfs_dqfunlock(dqp); |
2f8a3ce1 | 994 | return 0; |
1da177e4 | 995 | } |
1da177e4 LT |
996 | xfs_qm_dqunpin_wait(dqp); |
997 | ||
998 | /* | |
999 | * This may have been unpinned because the filesystem is shutting | |
1000 | * down forcibly. If that's the case we must not write this dquot | |
1001 | * to disk, because the log record didn't make it to disk! | |
1002 | */ | |
acecf1b5 CH |
1003 | if (XFS_FORCED_SHUTDOWN(mp)) { |
1004 | dqp->dq_flags &= ~XFS_DQ_DIRTY; | |
1da177e4 LT |
1005 | xfs_dqfunlock(dqp); |
1006 | return XFS_ERROR(EIO); | |
1007 | } | |
1008 | ||
1009 | /* | |
1010 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1011 | */ |
acecf1b5 CH |
1012 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
1013 | mp->m_quotainfo->qi_dqchunklen, 0, &bp); | |
1014 | if (error) { | |
1da177e4 | 1015 | ASSERT(error != ENOENT); |
1da177e4 | 1016 | xfs_dqfunlock(dqp); |
acecf1b5 | 1017 | return error; |
1da177e4 LT |
1018 | } |
1019 | ||
acecf1b5 CH |
1020 | /* |
1021 | * Calculate the location of the dquot inside the buffer. | |
1022 | */ | |
62926044 | 1023 | ddqp = bp->b_addr + dqp->q_bufoffset; |
acecf1b5 CH |
1024 | |
1025 | /* | |
1026 | * A simple sanity check in case we got a corrupted dquot.. | |
1027 | */ | |
a0fa2b67 DC |
1028 | error = xfs_qm_dqcheck(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, |
1029 | XFS_QMOPT_DOWARN, "dqflush (incore copy)"); | |
1030 | if (error) { | |
acecf1b5 CH |
1031 | xfs_buf_relse(bp); |
1032 | xfs_dqfunlock(dqp); | |
1033 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
1da177e4 LT |
1034 | return XFS_ERROR(EIO); |
1035 | } | |
1036 | ||
1037 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1038 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1039 | |
1040 | /* | |
1041 | * Clear the dirty field and remember the flush lsn for later use. | |
1042 | */ | |
acecf1b5 | 1043 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1044 | |
7b2e2a31 DC |
1045 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1046 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 LT |
1047 | |
1048 | /* | |
1049 | * Attach an iodone routine so that we can remove this dquot from the | |
1050 | * AIL and release the flush lock once the dquot is synced to disk. | |
1051 | */ | |
ca30b2a7 CH |
1052 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1053 | &dqp->q_logitem.qli_item); | |
1054 | ||
1da177e4 LT |
1055 | /* |
1056 | * If the buffer is pinned then push on the log so we won't | |
1057 | * get stuck waiting in the write for too long. | |
1058 | */ | |
811e64c7 | 1059 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1060 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1061 | xfs_log_force(mp, 0); |
1da177e4 LT |
1062 | } |
1063 | ||
20026d92 | 1064 | if (flags & SYNC_WAIT) |
c2b006c1 CH |
1065 | error = xfs_bwrite(bp); |
1066 | else | |
61551f1e | 1067 | xfs_buf_delwri_queue(bp); |
c2b006c1 CH |
1068 | |
1069 | xfs_buf_relse(bp); | |
0b1b213f CH |
1070 | |
1071 | trace_xfs_dqflush_done(dqp); | |
1072 | ||
1da177e4 LT |
1073 | /* |
1074 | * dqp is still locked, but caller is free to unlock it now. | |
1075 | */ | |
20026d92 | 1076 | return error; |
1da177e4 LT |
1077 | |
1078 | } | |
1079 | ||
5bb87a33 CH |
1080 | /* |
1081 | * Lock two xfs_dquot structures. | |
1082 | * | |
1083 | * To avoid deadlocks we always lock the quota structure with | |
1084 | * the lowerd id first. | |
1085 | */ | |
1da177e4 LT |
1086 | void |
1087 | xfs_dqlock2( | |
1088 | xfs_dquot_t *d1, | |
1089 | xfs_dquot_t *d2) | |
1090 | { | |
1091 | if (d1 && d2) { | |
1092 | ASSERT(d1 != d2); | |
1149d96a CH |
1093 | if (be32_to_cpu(d1->q_core.d_id) > |
1094 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1095 | mutex_lock(&d2->q_qlock); |
1096 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1097 | } else { |
5bb87a33 CH |
1098 | mutex_lock(&d1->q_qlock); |
1099 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1100 | } |
5bb87a33 CH |
1101 | } else if (d1) { |
1102 | mutex_lock(&d1->q_qlock); | |
1103 | } else if (d2) { | |
1104 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1105 | } |
1106 | } | |
1107 | ||
1da177e4 | 1108 | /* |
92678554 CH |
1109 | * Take a dquot out of the mount's dqlist as well as the hashlist. This is |
1110 | * called via unmount as well as quotaoff, and the purge will always succeed. | |
1da177e4 | 1111 | */ |
92678554 | 1112 | void |
1da177e4 | 1113 | xfs_qm_dqpurge( |
92678554 | 1114 | struct xfs_dquot *dqp) |
1da177e4 | 1115 | { |
92678554 CH |
1116 | struct xfs_mount *mp = dqp->q_mount; |
1117 | struct xfs_dqhash *qh = dqp->q_hash; | |
80a376bf | 1118 | |
1da177e4 | 1119 | xfs_dqlock(dqp); |
1da177e4 LT |
1120 | |
1121 | /* | |
1122 | * If we're turning off quotas, we have to make sure that, for | |
1123 | * example, we don't delete quota disk blocks while dquots are | |
1124 | * in the process of getting written to those disk blocks. | |
1125 | * This dquot might well be on AIL, and we can't leave it there | |
1126 | * if we're turning off quotas. Basically, we need this flush | |
1127 | * lock, and are willing to block on it. | |
1128 | */ | |
e1f49cf2 | 1129 | if (!xfs_dqflock_nowait(dqp)) { |
1da177e4 LT |
1130 | /* |
1131 | * Block on the flush lock after nudging dquot buffer, | |
1132 | * if it is incore. | |
1133 | */ | |
800b484e | 1134 | xfs_dqflock_pushbuf_wait(dqp); |
1da177e4 LT |
1135 | } |
1136 | ||
1137 | /* | |
92678554 | 1138 | * If we are turning this type of quotas off, we don't care |
1da177e4 LT |
1139 | * about the dirty metadata sitting in this dquot. OTOH, if |
1140 | * we're unmounting, we do care, so we flush it and wait. | |
1141 | */ | |
1142 | if (XFS_DQ_IS_DIRTY(dqp)) { | |
3c56836f | 1143 | int error; |
0b1b213f | 1144 | |
1da177e4 | 1145 | /* |
1da177e4 LT |
1146 | * We don't care about getting disk errors here. We need |
1147 | * to purge this dquot anyway, so we go ahead regardless. | |
1148 | */ | |
20026d92 | 1149 | error = xfs_qm_dqflush(dqp, SYNC_WAIT); |
3c56836f | 1150 | if (error) |
53487786 DC |
1151 | xfs_warn(mp, "%s: dquot %p flush failed", |
1152 | __func__, dqp); | |
1da177e4 LT |
1153 | xfs_dqflock(dqp); |
1154 | } | |
92678554 | 1155 | |
bc3048e3 | 1156 | ASSERT(atomic_read(&dqp->q_pincount) == 0); |
1da177e4 LT |
1157 | ASSERT(XFS_FORCED_SHUTDOWN(mp) || |
1158 | !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL)); | |
1159 | ||
92678554 CH |
1160 | xfs_dqfunlock(dqp); |
1161 | xfs_dqunlock(dqp); | |
1162 | ||
1163 | mutex_lock(&qh->qh_lock); | |
e6a81f13 DC |
1164 | list_del_init(&dqp->q_hashlist); |
1165 | qh->qh_version++; | |
92678554 | 1166 | mutex_unlock(&qh->qh_lock); |
80a376bf | 1167 | |
92678554 | 1168 | mutex_lock(&mp->m_quotainfo->qi_dqlist_lock); |
3a25404b DC |
1169 | list_del_init(&dqp->q_mplist); |
1170 | mp->m_quotainfo->qi_dqreclaims++; | |
1171 | mp->m_quotainfo->qi_dquots--; | |
92678554 | 1172 | mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock); |
1da177e4 | 1173 | |
92678554 CH |
1174 | /* |
1175 | * We move dquots to the freelist as soon as their reference count | |
1176 | * hits zero, so it really should be on the freelist here. | |
1177 | */ | |
1178 | mutex_lock(&xfs_Gqm->qm_dqfrlist_lock); | |
1179 | ASSERT(!list_empty(&dqp->q_freelist)); | |
80a376bf CH |
1180 | list_del_init(&dqp->q_freelist); |
1181 | xfs_Gqm->qm_dqfrlist_cnt--; | |
48776fd2 | 1182 | XFS_STATS_DEC(xs_qm_dquot_unused); |
80a376bf | 1183 | mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); |
1da177e4 | 1184 | |
80a376bf | 1185 | xfs_qm_dqdestroy(dqp); |
80a376bf | 1186 | } |
1da177e4 | 1187 | |
1da177e4 LT |
1188 | /* |
1189 | * Give the buffer a little push if it is incore and | |
1190 | * wait on the flush lock. | |
1191 | */ | |
1192 | void | |
800b484e | 1193 | xfs_dqflock_pushbuf_wait( |
1da177e4 LT |
1194 | xfs_dquot_t *dqp) |
1195 | { | |
8a7b8a89 | 1196 | xfs_mount_t *mp = dqp->q_mount; |
1da177e4 LT |
1197 | xfs_buf_t *bp; |
1198 | ||
1199 | /* | |
1200 | * Check to see if the dquot has been flushed delayed | |
1201 | * write. If so, grab its buffer and send it | |
1202 | * out immediately. We'll be able to acquire | |
1203 | * the flush lock when the I/O completes. | |
1204 | */ | |
8a7b8a89 CH |
1205 | bp = xfs_incore(mp->m_ddev_targp, dqp->q_blkno, |
1206 | mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK); | |
7d6a7bde DC |
1207 | if (!bp) |
1208 | goto out_lock; | |
1209 | ||
1210 | if (XFS_BUF_ISDELAYWRITE(bp)) { | |
811e64c7 | 1211 | if (xfs_buf_ispinned(bp)) |
8a7b8a89 | 1212 | xfs_log_force(mp, 0); |
7d6a7bde DC |
1213 | xfs_buf_delwri_promote(bp); |
1214 | wake_up_process(bp->b_target->bt_task); | |
1da177e4 | 1215 | } |
7d6a7bde DC |
1216 | xfs_buf_relse(bp); |
1217 | out_lock: | |
1da177e4 LT |
1218 | xfs_dqflock(dqp); |
1219 | } |