Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
4ce3121f NS |
3 | * Copyright (c) 2000-2003 Silicon Graphics, Inc. |
4 | * All Rights Reserved. | |
1da177e4 | 5 | */ |
1da177e4 LT |
6 | #include "xfs.h" |
7 | #include "xfs_fs.h" | |
6ca1c906 | 8 | #include "xfs_format.h" |
239880ef | 9 | #include "xfs_log_format.h" |
70a9883c | 10 | #include "xfs_shared.h" |
239880ef | 11 | #include "xfs_trans_resv.h" |
a844f451 | 12 | #include "xfs_bit.h" |
1da177e4 | 13 | #include "xfs_mount.h" |
3ab78df2 | 14 | #include "xfs_defer.h" |
1da177e4 LT |
15 | #include "xfs_inode.h" |
16 | #include "xfs_bmap.h" | |
68988114 | 17 | #include "xfs_bmap_util.h" |
239880ef DC |
18 | #include "xfs_alloc.h" |
19 | #include "xfs_quota.h" | |
1da177e4 | 20 | #include "xfs_error.h" |
239880ef | 21 | #include "xfs_trans.h" |
1da177e4 LT |
22 | #include "xfs_buf_item.h" |
23 | #include "xfs_trans_space.h" | |
24 | #include "xfs_trans_priv.h" | |
1da177e4 | 25 | #include "xfs_qm.h" |
3fe58f30 | 26 | #include "xfs_cksum.h" |
0b1b213f | 27 | #include "xfs_trace.h" |
239880ef | 28 | #include "xfs_log.h" |
a4fbe6ab | 29 | #include "xfs_bmap_btree.h" |
1da177e4 | 30 | |
1da177e4 | 31 | /* |
bf72de31 CH |
32 | * Lock order: |
33 | * | |
34 | * ip->i_lock | |
9f920f11 | 35 | * qi->qi_tree_lock |
b84a3a96 CH |
36 | * dquot->q_qlock (xfs_dqlock() and friends) |
37 | * dquot->q_flush (xfs_dqflock() and friends) | |
38 | * qi->qi_lru_lock | |
bf72de31 CH |
39 | * |
40 | * If two dquots need to be locked the order is user before group/project, | |
41 | * otherwise by the lowest id first, see xfs_dqlock2. | |
42 | */ | |
1da177e4 | 43 | |
a05931ce CH |
44 | struct kmem_zone *xfs_qm_dqtrxzone; |
45 | static struct kmem_zone *xfs_qm_dqzone; | |
46 | ||
f112a049 DC |
47 | static struct lock_class_key xfs_dquot_group_class; |
48 | static struct lock_class_key xfs_dquot_project_class; | |
98b8c7a0 | 49 | |
1da177e4 LT |
50 | /* |
51 | * This is called to free all the memory associated with a dquot | |
52 | */ | |
53 | void | |
54 | xfs_qm_dqdestroy( | |
55 | xfs_dquot_t *dqp) | |
56 | { | |
f8739c3c | 57 | ASSERT(list_empty(&dqp->q_lru)); |
1da177e4 | 58 | |
b1c5ebb2 | 59 | kmem_free(dqp->q_logitem.qli_item.li_lv_shadow); |
1da177e4 | 60 | mutex_destroy(&dqp->q_qlock); |
0b1b213f | 61 | |
ff6d6af2 BD |
62 | XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot); |
63 | kmem_zone_free(xfs_qm_dqzone, dqp); | |
1da177e4 LT |
64 | } |
65 | ||
1da177e4 LT |
66 | /* |
67 | * If default limits are in force, push them into the dquot now. | |
68 | * We overwrite the dquot limits only if they are zero and this | |
69 | * is not the root dquot. | |
70 | */ | |
71 | void | |
72 | xfs_qm_adjust_dqlimits( | |
4b6eae2e BF |
73 | struct xfs_mount *mp, |
74 | struct xfs_dquot *dq) | |
1da177e4 | 75 | { |
4b6eae2e BF |
76 | struct xfs_quotainfo *q = mp->m_quotainfo; |
77 | struct xfs_disk_dquot *d = &dq->q_core; | |
be607946 | 78 | struct xfs_def_quota *defq; |
b1366451 | 79 | int prealloc = 0; |
1da177e4 LT |
80 | |
81 | ASSERT(d->d_id); | |
be607946 | 82 | defq = xfs_get_defquota(dq, q); |
1da177e4 | 83 | |
be607946 CM |
84 | if (defq->bsoftlimit && !d->d_blk_softlimit) { |
85 | d->d_blk_softlimit = cpu_to_be64(defq->bsoftlimit); | |
b1366451 BF |
86 | prealloc = 1; |
87 | } | |
be607946 CM |
88 | if (defq->bhardlimit && !d->d_blk_hardlimit) { |
89 | d->d_blk_hardlimit = cpu_to_be64(defq->bhardlimit); | |
b1366451 BF |
90 | prealloc = 1; |
91 | } | |
be607946 CM |
92 | if (defq->isoftlimit && !d->d_ino_softlimit) |
93 | d->d_ino_softlimit = cpu_to_be64(defq->isoftlimit); | |
94 | if (defq->ihardlimit && !d->d_ino_hardlimit) | |
95 | d->d_ino_hardlimit = cpu_to_be64(defq->ihardlimit); | |
96 | if (defq->rtbsoftlimit && !d->d_rtb_softlimit) | |
97 | d->d_rtb_softlimit = cpu_to_be64(defq->rtbsoftlimit); | |
98 | if (defq->rtbhardlimit && !d->d_rtb_hardlimit) | |
99 | d->d_rtb_hardlimit = cpu_to_be64(defq->rtbhardlimit); | |
b1366451 BF |
100 | |
101 | if (prealloc) | |
102 | xfs_dquot_set_prealloc_limits(dq); | |
1da177e4 LT |
103 | } |
104 | ||
105 | /* | |
106 | * Check the limits and timers of a dquot and start or reset timers | |
107 | * if necessary. | |
108 | * This gets called even when quota enforcement is OFF, which makes our | |
109 | * life a little less complicated. (We just don't reject any quota | |
110 | * reservations in that case, when enforcement is off). | |
111 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
112 | * enforcement's off. | |
113 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
114 | * 'automatically' get started when limits get exceeded. They do |
115 | * get reset to zero, however, when we find the count to be under | |
116 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
117 | */ |
118 | void | |
119 | xfs_qm_adjust_dqtimers( | |
120 | xfs_mount_t *mp, | |
121 | xfs_disk_dquot_t *d) | |
122 | { | |
123 | ASSERT(d->d_id); | |
124 | ||
ea15ab3c | 125 | #ifdef DEBUG |
1149d96a CH |
126 | if (d->d_blk_hardlimit) |
127 | ASSERT(be64_to_cpu(d->d_blk_softlimit) <= | |
128 | be64_to_cpu(d->d_blk_hardlimit)); | |
129 | if (d->d_ino_hardlimit) | |
130 | ASSERT(be64_to_cpu(d->d_ino_softlimit) <= | |
131 | be64_to_cpu(d->d_ino_hardlimit)); | |
132 | if (d->d_rtb_hardlimit) | |
133 | ASSERT(be64_to_cpu(d->d_rtb_softlimit) <= | |
134 | be64_to_cpu(d->d_rtb_hardlimit)); | |
1da177e4 | 135 | #endif |
ea15ab3c | 136 | |
1da177e4 | 137 | if (!d->d_btimer) { |
1149d96a | 138 | if ((d->d_blk_softlimit && |
d0a3fe67 | 139 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
140 | be64_to_cpu(d->d_blk_softlimit))) || |
141 | (d->d_blk_hardlimit && | |
d0a3fe67 | 142 | (be64_to_cpu(d->d_bcount) > |
1149d96a CH |
143 | be64_to_cpu(d->d_blk_hardlimit)))) { |
144 | d->d_btimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 145 | mp->m_quotainfo->qi_btimelimit); |
754002b4 NS |
146 | } else { |
147 | d->d_bwarns = 0; | |
1da177e4 LT |
148 | } |
149 | } else { | |
150 | if ((!d->d_blk_softlimit || | |
d0a3fe67 | 151 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 152 | be64_to_cpu(d->d_blk_softlimit))) && |
1da177e4 | 153 | (!d->d_blk_hardlimit || |
d0a3fe67 | 154 | (be64_to_cpu(d->d_bcount) <= |
1149d96a | 155 | be64_to_cpu(d->d_blk_hardlimit)))) { |
1da177e4 LT |
156 | d->d_btimer = 0; |
157 | } | |
158 | } | |
159 | ||
160 | if (!d->d_itimer) { | |
1149d96a | 161 | if ((d->d_ino_softlimit && |
d0a3fe67 | 162 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
163 | be64_to_cpu(d->d_ino_softlimit))) || |
164 | (d->d_ino_hardlimit && | |
d0a3fe67 | 165 | (be64_to_cpu(d->d_icount) > |
1149d96a CH |
166 | be64_to_cpu(d->d_ino_hardlimit)))) { |
167 | d->d_itimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 168 | mp->m_quotainfo->qi_itimelimit); |
754002b4 NS |
169 | } else { |
170 | d->d_iwarns = 0; | |
1da177e4 LT |
171 | } |
172 | } else { | |
173 | if ((!d->d_ino_softlimit || | |
d0a3fe67 | 174 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 175 | be64_to_cpu(d->d_ino_softlimit))) && |
1da177e4 | 176 | (!d->d_ino_hardlimit || |
d0a3fe67 | 177 | (be64_to_cpu(d->d_icount) <= |
1149d96a | 178 | be64_to_cpu(d->d_ino_hardlimit)))) { |
1da177e4 LT |
179 | d->d_itimer = 0; |
180 | } | |
181 | } | |
182 | ||
183 | if (!d->d_rtbtimer) { | |
1149d96a | 184 | if ((d->d_rtb_softlimit && |
d0a3fe67 | 185 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
186 | be64_to_cpu(d->d_rtb_softlimit))) || |
187 | (d->d_rtb_hardlimit && | |
d0a3fe67 | 188 | (be64_to_cpu(d->d_rtbcount) > |
1149d96a CH |
189 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
190 | d->d_rtbtimer = cpu_to_be32(get_seconds() + | |
8a7b8a89 | 191 | mp->m_quotainfo->qi_rtbtimelimit); |
754002b4 NS |
192 | } else { |
193 | d->d_rtbwarns = 0; | |
1da177e4 LT |
194 | } |
195 | } else { | |
196 | if ((!d->d_rtb_softlimit || | |
d0a3fe67 | 197 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 198 | be64_to_cpu(d->d_rtb_softlimit))) && |
1da177e4 | 199 | (!d->d_rtb_hardlimit || |
d0a3fe67 | 200 | (be64_to_cpu(d->d_rtbcount) <= |
1149d96a | 201 | be64_to_cpu(d->d_rtb_hardlimit)))) { |
1da177e4 LT |
202 | d->d_rtbtimer = 0; |
203 | } | |
204 | } | |
205 | } | |
206 | ||
1da177e4 LT |
207 | /* |
208 | * initialize a buffer full of dquots and log the whole thing | |
209 | */ | |
210 | STATIC void | |
211 | xfs_qm_init_dquot_blk( | |
212 | xfs_trans_t *tp, | |
213 | xfs_mount_t *mp, | |
214 | xfs_dqid_t id, | |
215 | uint type, | |
216 | xfs_buf_t *bp) | |
217 | { | |
8a7b8a89 | 218 | struct xfs_quotainfo *q = mp->m_quotainfo; |
1da177e4 | 219 | xfs_dqblk_t *d; |
a484bcdd ES |
220 | xfs_dqid_t curid; |
221 | int i; | |
1da177e4 LT |
222 | |
223 | ASSERT(tp); | |
0c842ad4 | 224 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 225 | |
62926044 | 226 | d = bp->b_addr; |
1da177e4 LT |
227 | |
228 | /* | |
229 | * ID of the first dquot in the block - id's are zero based. | |
230 | */ | |
8a7b8a89 | 231 | curid = id - (id % q->qi_dqperchunk); |
8a7b8a89 | 232 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
49d35a5c CH |
233 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) { |
234 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
235 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
236 | d->dd_diskdq.d_id = cpu_to_be32(curid); | |
237 | d->dd_diskdq.d_flags = type; | |
6fcdc59d | 238 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
92863451 | 239 | uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid); |
6fcdc59d DC |
240 | xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk), |
241 | XFS_DQUOT_CRC_OFF); | |
242 | } | |
49d35a5c CH |
243 | } |
244 | ||
1da177e4 | 245 | xfs_trans_dquot_buf(tp, bp, |
c1155410 DC |
246 | (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF : |
247 | ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF : | |
248 | XFS_BLF_GDQUOT_BUF))); | |
8a7b8a89 | 249 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); |
1da177e4 LT |
250 | } |
251 | ||
b1366451 BF |
252 | /* |
253 | * Initialize the dynamic speculative preallocation thresholds. The lo/hi | |
254 | * watermarks correspond to the soft and hard limits by default. If a soft limit | |
255 | * is not specified, we use 95% of the hard limit. | |
256 | */ | |
257 | void | |
258 | xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp) | |
259 | { | |
c8ce540d | 260 | uint64_t space; |
b1366451 BF |
261 | |
262 | dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit); | |
263 | dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit); | |
264 | if (!dqp->q_prealloc_lo_wmark) { | |
265 | dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark; | |
266 | do_div(dqp->q_prealloc_lo_wmark, 100); | |
267 | dqp->q_prealloc_lo_wmark *= 95; | |
268 | } | |
269 | ||
270 | space = dqp->q_prealloc_hi_wmark; | |
271 | ||
272 | do_div(space, 100); | |
273 | dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space; | |
274 | dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3; | |
275 | dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5; | |
276 | } | |
277 | ||
1da177e4 | 278 | /* |
d63192c8 DW |
279 | * Ensure that the given in-core dquot has a buffer on disk backing it, and |
280 | * return the buffer. This is called when the bmapi finds a hole. | |
1da177e4 LT |
281 | */ |
282 | STATIC int | |
d63192c8 DW |
283 | xfs_dquot_disk_alloc( |
284 | struct xfs_trans **tpp, | |
285 | struct xfs_dquot *dqp, | |
286 | struct xfs_buf **bpp) | |
1da177e4 | 287 | { |
d63192c8 DW |
288 | struct xfs_bmbt_irec map; |
289 | struct xfs_defer_ops dfops; | |
290 | struct xfs_mount *mp = (*tpp)->t_mountp; | |
291 | struct xfs_buf *bp; | |
292 | struct xfs_inode *quotip = xfs_quota_inode(mp, dqp->dq_flags); | |
293 | xfs_fsblock_t firstblock; | |
294 | int nmaps = 1; | |
295 | int error; | |
0b1b213f CH |
296 | |
297 | trace_xfs_dqalloc(dqp); | |
1da177e4 | 298 | |
2c3234d1 | 299 | xfs_defer_init(&dfops, &firstblock); |
1da177e4 | 300 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
6967b964 | 301 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
d63192c8 DW |
302 | /* |
303 | * Return if this type of quotas is turned off while we didn't | |
304 | * have an inode lock | |
305 | */ | |
1da177e4 | 306 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
2451337d | 307 | return -ESRCH; |
1da177e4 LT |
308 | } |
309 | ||
d63192c8 DW |
310 | /* Create the block mapping. */ |
311 | xfs_trans_ijoin(*tpp, quotip, XFS_ILOCK_EXCL); | |
312 | error = xfs_bmapi_write(*tpp, quotip, dqp->q_fileoffset, | |
313 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, | |
314 | &firstblock, XFS_QM_DQALLOC_SPACE_RES(mp), | |
315 | &map, &nmaps, &dfops); | |
c0dc7828 | 316 | if (error) |
1da177e4 | 317 | goto error0; |
1da177e4 LT |
318 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
319 | ASSERT(nmaps == 1); | |
320 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
321 | (map.br_startblock != HOLESTARTBLOCK)); | |
322 | ||
323 | /* | |
324 | * Keep track of the blkno to save a lookup later | |
325 | */ | |
326 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
327 | ||
328 | /* now we can just get the buffer (there's nothing to read yet) */ | |
d63192c8 DW |
329 | bp = xfs_trans_get_buf(*tpp, mp->m_ddev_targp, dqp->q_blkno, |
330 | mp->m_quotainfo->qi_dqchunklen, 0); | |
36de9556 | 331 | if (!bp) { |
2451337d | 332 | error = -ENOMEM; |
1da177e4 | 333 | goto error1; |
36de9556 | 334 | } |
1813dd64 | 335 | bp->b_ops = &xfs_dquot_buf_ops; |
2a30f36d | 336 | |
1da177e4 LT |
337 | /* |
338 | * Make a chunk of dquots out of this buffer and log | |
339 | * the entire thing. | |
340 | */ | |
d63192c8 | 341 | xfs_qm_init_dquot_blk(*tpp, mp, be32_to_cpu(dqp->q_core.d_id), |
c8ad20ff | 342 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
d63192c8 | 343 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 | 344 | |
efa092f3 | 345 | /* |
7b6b50f5 DW |
346 | * Hold the buffer and join it to the dfops so that we'll still own |
347 | * the buffer when we return to the caller. The buffer disposal on | |
348 | * error must be paid attention to very carefully, as it has been | |
349 | * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota | |
350 | * code when allocating a new dquot record" in 2005, and the later | |
351 | * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep | |
352 | * the buffer locked across the _defer_finish call. We can now do | |
353 | * this correctly with xfs_defer_bjoin. | |
efa092f3 | 354 | * |
7b6b50f5 DW |
355 | * Above, we allocated a disk block for the dquot information and |
356 | * used get_buf to initialize the dquot. If the _defer_bjoin fails, | |
357 | * the buffer is still locked to *tpp, so we must _bhold_release and | |
358 | * then _trans_brelse the buffer. If the _defer_finish fails, the old | |
359 | * transaction is gone but the new buffer is not joined or held to any | |
360 | * transaction, so we must _buf_relse it. | |
efa092f3 | 361 | * |
7b6b50f5 | 362 | * If everything succeeds, the caller of this function is returned a |
d63192c8 | 363 | * buffer that is locked and held to the transaction. The caller |
7b6b50f5 DW |
364 | * is responsible for unlocking any buffer passed back, either |
365 | * manually or by committing the transaction. | |
efa092f3 | 366 | */ |
7b6b50f5 DW |
367 | xfs_trans_bhold(*tpp, bp); |
368 | error = xfs_defer_bjoin(&dfops, bp); | |
369 | if (error) { | |
370 | xfs_trans_bhold_release(*tpp, bp); | |
371 | xfs_trans_brelse(*tpp, bp); | |
372 | goto error1; | |
373 | } | |
8ad7c629 | 374 | error = xfs_defer_finish(tpp, &dfops); |
7b6b50f5 DW |
375 | if (error) { |
376 | xfs_buf_relse(bp); | |
1da177e4 | 377 | goto error1; |
efa092f3 | 378 | } |
d63192c8 | 379 | *bpp = bp; |
1da177e4 LT |
380 | return 0; |
381 | ||
f6106efa | 382 | error1: |
2c3234d1 | 383 | xfs_defer_cancel(&dfops); |
f6106efa | 384 | error0: |
d99831ff | 385 | return error; |
1da177e4 | 386 | } |
9aede1d8 | 387 | |
1da177e4 | 388 | /* |
d63192c8 DW |
389 | * Read in the in-core dquot's on-disk metadata and return the buffer. |
390 | * Returns ENOENT to signal a hole. | |
1da177e4 LT |
391 | */ |
392 | STATIC int | |
d63192c8 DW |
393 | xfs_dquot_disk_read( |
394 | struct xfs_mount *mp, | |
395 | struct xfs_dquot *dqp, | |
396 | struct xfs_buf **bpp) | |
1da177e4 | 397 | { |
113a5683 | 398 | struct xfs_bmbt_irec map; |
113a5683 | 399 | struct xfs_buf *bp; |
d63192c8 | 400 | struct xfs_inode *quotip = xfs_quota_inode(mp, dqp->dq_flags); |
0891f997 | 401 | uint lock_mode; |
d63192c8 DW |
402 | int nmaps = 1; |
403 | int error; | |
1da177e4 | 404 | |
0891f997 | 405 | lock_mode = xfs_ilock_data_map_shared(quotip); |
d63192c8 | 406 | if (!xfs_this_quota_on(mp, dqp->dq_flags)) { |
1da177e4 | 407 | /* |
acecf1b5 CH |
408 | * Return if this type of quotas is turned off while we |
409 | * didn't have the quota inode lock. | |
1da177e4 | 410 | */ |
0891f997 | 411 | xfs_iunlock(quotip, lock_mode); |
2451337d | 412 | return -ESRCH; |
acecf1b5 CH |
413 | } |
414 | ||
415 | /* | |
416 | * Find the block map; no allocations yet | |
417 | */ | |
5c8ed202 | 418 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
d63192c8 | 419 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); |
0891f997 | 420 | xfs_iunlock(quotip, lock_mode); |
acecf1b5 CH |
421 | if (error) |
422 | return error; | |
423 | ||
424 | ASSERT(nmaps == 1); | |
d63192c8 DW |
425 | ASSERT(map.br_blockcount >= 1); |
426 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
427 | if (map.br_startblock == HOLESTARTBLOCK) | |
428 | return -ENOENT; | |
429 | ||
430 | trace_xfs_dqtobp_read(dqp); | |
acecf1b5 CH |
431 | |
432 | /* | |
d63192c8 DW |
433 | * store the blkno etc so that we don't have to do the |
434 | * mapping all the time | |
acecf1b5 | 435 | */ |
d63192c8 | 436 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
1da177e4 | 437 | |
d63192c8 DW |
438 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
439 | mp->m_quotainfo->qi_dqchunklen, 0, &bp, | |
440 | &xfs_dquot_buf_ops); | |
441 | if (error) { | |
442 | ASSERT(bp == NULL); | |
443 | return error; | |
1da177e4 LT |
444 | } |
445 | ||
c6319198 | 446 | ASSERT(xfs_buf_islocked(bp)); |
d63192c8 DW |
447 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
448 | *bpp = bp; | |
1da177e4 | 449 | |
d99831ff | 450 | return 0; |
1da177e4 LT |
451 | } |
452 | ||
617cd5c1 DW |
453 | /* Allocate and initialize everything we need for an incore dquot. */ |
454 | STATIC struct xfs_dquot * | |
455 | xfs_dquot_alloc( | |
97e7ade5 CH |
456 | struct xfs_mount *mp, |
457 | xfs_dqid_t id, | |
617cd5c1 | 458 | uint type) |
1da177e4 | 459 | { |
97e7ade5 | 460 | struct xfs_dquot *dqp; |
92b2e5b3 | 461 | |
a05931ce | 462 | dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP); |
92b2e5b3 CH |
463 | |
464 | dqp->dq_flags = type; | |
465 | dqp->q_core.d_id = cpu_to_be32(id); | |
466 | dqp->q_mount = mp; | |
f8739c3c | 467 | INIT_LIST_HEAD(&dqp->q_lru); |
92b2e5b3 CH |
468 | mutex_init(&dqp->q_qlock); |
469 | init_waitqueue_head(&dqp->q_pinwait); | |
d63192c8 DW |
470 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
471 | /* | |
472 | * Offset of dquot in the (fixed sized) dquot chunk. | |
473 | */ | |
474 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
475 | sizeof(xfs_dqblk_t); | |
92b2e5b3 CH |
476 | |
477 | /* | |
478 | * Because we want to use a counting completion, complete | |
479 | * the flush completion once to allow a single access to | |
480 | * the flush completion without blocking. | |
481 | */ | |
482 | init_completion(&dqp->q_flush); | |
483 | complete(&dqp->q_flush); | |
484 | ||
485 | /* | |
486 | * Make sure group quotas have a different lock class than user | |
487 | * quotas. | |
488 | */ | |
f112a049 DC |
489 | switch (type) { |
490 | case XFS_DQ_USER: | |
491 | /* uses the default lock class */ | |
492 | break; | |
493 | case XFS_DQ_GROUP: | |
494 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class); | |
495 | break; | |
496 | case XFS_DQ_PROJ: | |
497 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class); | |
498 | break; | |
499 | default: | |
500 | ASSERT(0); | |
501 | break; | |
502 | } | |
92b2e5b3 | 503 | |
617cd5c1 DW |
504 | xfs_qm_dquot_logitem_init(dqp); |
505 | ||
ff6d6af2 | 506 | XFS_STATS_INC(mp, xs_qm_dquot); |
617cd5c1 DW |
507 | return dqp; |
508 | } | |
509 | ||
510 | /* Copy the in-core quota fields in from the on-disk buffer. */ | |
511 | STATIC void | |
512 | xfs_dquot_from_disk( | |
513 | struct xfs_dquot *dqp, | |
d63192c8 | 514 | struct xfs_buf *bp) |
617cd5c1 | 515 | { |
d63192c8 DW |
516 | struct xfs_disk_dquot *ddqp = bp->b_addr + dqp->q_bufoffset; |
517 | ||
617cd5c1 DW |
518 | /* copy everything from disk dquot to the incore dquot */ |
519 | memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t)); | |
520 | ||
521 | /* | |
522 | * Reservation counters are defined as reservation plus current usage | |
523 | * to avoid having to add every time. | |
524 | */ | |
525 | dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount); | |
526 | dqp->q_res_icount = be64_to_cpu(ddqp->d_icount); | |
527 | dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount); | |
528 | ||
529 | /* initialize the dquot speculative prealloc thresholds */ | |
530 | xfs_dquot_set_prealloc_limits(dqp); | |
531 | } | |
1da177e4 | 532 | |
d63192c8 DW |
533 | /* Allocate and initialize the dquot buffer for this in-core dquot. */ |
534 | static int | |
535 | xfs_qm_dqread_alloc( | |
536 | struct xfs_mount *mp, | |
537 | struct xfs_dquot *dqp, | |
538 | struct xfs_buf **bpp) | |
539 | { | |
540 | struct xfs_trans *tp; | |
541 | struct xfs_buf *bp; | |
542 | int error; | |
543 | ||
544 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc, | |
545 | XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp); | |
546 | if (error) | |
547 | goto err; | |
548 | ||
549 | error = xfs_dquot_disk_alloc(&tp, dqp, &bp); | |
550 | if (error) | |
551 | goto err_cancel; | |
552 | ||
553 | error = xfs_trans_commit(tp); | |
554 | if (error) { | |
555 | /* | |
556 | * Buffer was held to the transaction, so we have to unlock it | |
557 | * manually here because we're not passing it back. | |
558 | */ | |
559 | xfs_buf_relse(bp); | |
560 | goto err; | |
561 | } | |
562 | *bpp = bp; | |
563 | return 0; | |
564 | ||
565 | err_cancel: | |
566 | xfs_trans_cancel(tp); | |
567 | err: | |
568 | return error; | |
569 | } | |
570 | ||
617cd5c1 DW |
571 | /* |
572 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
30ab2dcf DW |
573 | * and release the buffer immediately. If @can_alloc is true, fill any |
574 | * holes in the on-disk metadata. | |
617cd5c1 | 575 | */ |
114e73cc | 576 | static int |
617cd5c1 DW |
577 | xfs_qm_dqread( |
578 | struct xfs_mount *mp, | |
579 | xfs_dqid_t id, | |
580 | uint type, | |
30ab2dcf | 581 | bool can_alloc, |
d63192c8 | 582 | struct xfs_dquot **dqpp) |
617cd5c1 DW |
583 | { |
584 | struct xfs_dquot *dqp; | |
617cd5c1 | 585 | struct xfs_buf *bp; |
617cd5c1 DW |
586 | int error; |
587 | ||
588 | dqp = xfs_dquot_alloc(mp, id, type); | |
0b1b213f CH |
589 | trace_xfs_dqread(dqp); |
590 | ||
d63192c8 DW |
591 | /* Try to read the buffer, allocating if necessary. */ |
592 | error = xfs_dquot_disk_read(mp, dqp, &bp); | |
30ab2dcf | 593 | if (error == -ENOENT && can_alloc) |
d63192c8 DW |
594 | error = xfs_qm_dqread_alloc(mp, dqp, &bp); |
595 | if (error) | |
596 | goto err; | |
1da177e4 LT |
597 | |
598 | /* | |
d63192c8 DW |
599 | * At this point we should have a clean locked buffer. Copy the data |
600 | * to the incore dquot and release the buffer since the incore dquot | |
601 | * has its own locking protocol so we needn't tie up the buffer any | |
602 | * further. | |
1da177e4 | 603 | */ |
0c842ad4 | 604 | ASSERT(xfs_buf_islocked(bp)); |
d63192c8 | 605 | xfs_dquot_from_disk(dqp, bp); |
1da177e4 | 606 | |
d63192c8 DW |
607 | xfs_buf_relse(bp); |
608 | *dqpp = dqp; | |
97e7ade5 | 609 | return error; |
1da177e4 | 610 | |
d63192c8 DW |
611 | err: |
612 | trace_xfs_dqread_fail(dqp); | |
1da177e4 | 613 | xfs_qm_dqdestroy(dqp); |
d63192c8 | 614 | *dqpp = NULL; |
97e7ade5 | 615 | return error; |
1da177e4 LT |
616 | } |
617 | ||
296c24e2 ES |
618 | /* |
619 | * Advance to the next id in the current chunk, or if at the | |
620 | * end of the chunk, skip ahead to first id in next allocated chunk | |
621 | * using the SEEK_DATA interface. | |
622 | */ | |
6e3e6d55 | 623 | static int |
296c24e2 | 624 | xfs_dq_get_next_id( |
bda250db | 625 | struct xfs_mount *mp, |
296c24e2 | 626 | uint type, |
bda250db | 627 | xfs_dqid_t *id) |
296c24e2 | 628 | { |
bda250db CH |
629 | struct xfs_inode *quotip = xfs_quota_inode(mp, type); |
630 | xfs_dqid_t next_id = *id + 1; /* simple advance */ | |
631 | uint lock_flags; | |
632 | struct xfs_bmbt_irec got; | |
b2b1712a | 633 | struct xfs_iext_cursor cur; |
296c24e2 | 634 | xfs_fsblock_t start; |
296c24e2 ES |
635 | int error = 0; |
636 | ||
657bdfb7 ES |
637 | /* If we'd wrap past the max ID, stop */ |
638 | if (next_id < *id) | |
639 | return -ENOENT; | |
640 | ||
296c24e2 ES |
641 | /* If new ID is within the current chunk, advancing it sufficed */ |
642 | if (next_id % mp->m_quotainfo->qi_dqperchunk) { | |
643 | *id = next_id; | |
644 | return 0; | |
645 | } | |
646 | ||
647 | /* Nope, next_id is now past the current chunk, so find the next one */ | |
648 | start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk; | |
649 | ||
bda250db CH |
650 | lock_flags = xfs_ilock_data_map_shared(quotip); |
651 | if (!(quotip->i_df.if_flags & XFS_IFEXTENTS)) { | |
652 | error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK); | |
653 | if (error) | |
654 | return error; | |
655 | } | |
296c24e2 | 656 | |
b2b1712a | 657 | if (xfs_iext_lookup_extent(quotip, "ip->i_df, start, &cur, &got)) { |
2192b0ba BF |
658 | /* contiguous chunk, bump startoff for the id calculation */ |
659 | if (got.br_startoff < start) | |
660 | got.br_startoff = start; | |
bda250db | 661 | *id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk; |
2192b0ba | 662 | } else { |
bda250db | 663 | error = -ENOENT; |
2192b0ba BF |
664 | } |
665 | ||
bda250db | 666 | xfs_iunlock(quotip, lock_flags); |
296c24e2 | 667 | |
bda250db | 668 | return error; |
296c24e2 ES |
669 | } |
670 | ||
cc2047c4 DW |
671 | /* |
672 | * Look up the dquot in the in-core cache. If found, the dquot is returned | |
673 | * locked and ready to go. | |
674 | */ | |
675 | static struct xfs_dquot * | |
676 | xfs_qm_dqget_cache_lookup( | |
677 | struct xfs_mount *mp, | |
678 | struct xfs_quotainfo *qi, | |
679 | struct radix_tree_root *tree, | |
680 | xfs_dqid_t id) | |
681 | { | |
682 | struct xfs_dquot *dqp; | |
683 | ||
684 | restart: | |
685 | mutex_lock(&qi->qi_tree_lock); | |
686 | dqp = radix_tree_lookup(tree, id); | |
687 | if (!dqp) { | |
688 | mutex_unlock(&qi->qi_tree_lock); | |
689 | XFS_STATS_INC(mp, xs_qm_dqcachemisses); | |
690 | return NULL; | |
691 | } | |
692 | ||
693 | xfs_dqlock(dqp); | |
694 | if (dqp->dq_flags & XFS_DQ_FREEING) { | |
695 | xfs_dqunlock(dqp); | |
696 | mutex_unlock(&qi->qi_tree_lock); | |
697 | trace_xfs_dqget_freeing(dqp); | |
698 | delay(1); | |
699 | goto restart; | |
700 | } | |
701 | ||
702 | dqp->q_nrefs++; | |
703 | mutex_unlock(&qi->qi_tree_lock); | |
704 | ||
705 | trace_xfs_dqget_hit(dqp); | |
706 | XFS_STATS_INC(mp, xs_qm_dqcachehits); | |
707 | return dqp; | |
708 | } | |
709 | ||
710 | /* | |
711 | * Try to insert a new dquot into the in-core cache. If an error occurs the | |
712 | * caller should throw away the dquot and start over. Otherwise, the dquot | |
713 | * is returned locked (and held by the cache) as if there had been a cache | |
714 | * hit. | |
715 | */ | |
716 | static int | |
717 | xfs_qm_dqget_cache_insert( | |
718 | struct xfs_mount *mp, | |
719 | struct xfs_quotainfo *qi, | |
720 | struct radix_tree_root *tree, | |
721 | xfs_dqid_t id, | |
722 | struct xfs_dquot *dqp) | |
723 | { | |
724 | int error; | |
725 | ||
726 | mutex_lock(&qi->qi_tree_lock); | |
727 | error = radix_tree_insert(tree, id, dqp); | |
728 | if (unlikely(error)) { | |
729 | /* Duplicate found! Caller must try again. */ | |
730 | WARN_ON(error != -EEXIST); | |
731 | mutex_unlock(&qi->qi_tree_lock); | |
732 | trace_xfs_dqget_dup(dqp); | |
733 | return error; | |
734 | } | |
735 | ||
736 | /* Return a locked dquot to the caller, with a reference taken. */ | |
737 | xfs_dqlock(dqp); | |
738 | dqp->q_nrefs = 1; | |
739 | ||
740 | qi->qi_dquots++; | |
741 | mutex_unlock(&qi->qi_tree_lock); | |
742 | ||
743 | return 0; | |
744 | } | |
745 | ||
d7103eeb DW |
746 | /* Check our input parameters. */ |
747 | static int | |
748 | xfs_qm_dqget_checks( | |
749 | struct xfs_mount *mp, | |
750 | uint type) | |
751 | { | |
752 | if (WARN_ON_ONCE(!XFS_IS_QUOTA_RUNNING(mp))) | |
753 | return -ESRCH; | |
754 | ||
755 | switch (type) { | |
756 | case XFS_DQ_USER: | |
757 | if (!XFS_IS_UQUOTA_ON(mp)) | |
758 | return -ESRCH; | |
759 | return 0; | |
760 | case XFS_DQ_GROUP: | |
761 | if (!XFS_IS_GQUOTA_ON(mp)) | |
762 | return -ESRCH; | |
763 | return 0; | |
764 | case XFS_DQ_PROJ: | |
765 | if (!XFS_IS_PQUOTA_ON(mp)) | |
766 | return -ESRCH; | |
767 | return 0; | |
768 | default: | |
769 | WARN_ON_ONCE(0); | |
770 | return -EINVAL; | |
771 | } | |
772 | } | |
773 | ||
1da177e4 | 774 | /* |
4882c19d DW |
775 | * Given the file system, id, and type (UDQUOT/GDQUOT), return a a locked |
776 | * dquot, doing an allocation (if requested) as needed. | |
1da177e4 LT |
777 | */ |
778 | int | |
779 | xfs_qm_dqget( | |
4882c19d DW |
780 | struct xfs_mount *mp, |
781 | xfs_dqid_t id, | |
782 | uint type, | |
30ab2dcf | 783 | bool can_alloc, |
4882c19d | 784 | struct xfs_dquot **O_dqpp) |
1da177e4 | 785 | { |
9f920f11 | 786 | struct xfs_quotainfo *qi = mp->m_quotainfo; |
4882c19d | 787 | struct radix_tree_root *tree = xfs_dquot_tree(qi, type); |
9f920f11 CH |
788 | struct xfs_dquot *dqp; |
789 | int error; | |
1da177e4 | 790 | |
d7103eeb DW |
791 | error = xfs_qm_dqget_checks(mp, type); |
792 | if (error) | |
793 | return error; | |
1da177e4 | 794 | |
4882c19d DW |
795 | restart: |
796 | dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id); | |
797 | if (dqp) { | |
798 | *O_dqpp = dqp; | |
799 | return 0; | |
800 | } | |
801 | ||
30ab2dcf | 802 | error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp); |
4882c19d DW |
803 | if (error) |
804 | return error; | |
805 | ||
806 | error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp); | |
807 | if (error) { | |
808 | /* | |
809 | * Duplicate found. Just throw away the new dquot and start | |
810 | * over. | |
811 | */ | |
812 | xfs_qm_dqdestroy(dqp); | |
813 | XFS_STATS_INC(mp, xs_qm_dquot_dups); | |
814 | goto restart; | |
815 | } | |
816 | ||
817 | trace_xfs_dqget_miss(dqp); | |
818 | *O_dqpp = dqp; | |
819 | return 0; | |
820 | } | |
821 | ||
114e73cc DW |
822 | /* |
823 | * Given a dquot id and type, read and initialize a dquot from the on-disk | |
824 | * metadata. This function is only for use during quota initialization so | |
825 | * it ignores the dquot cache assuming that the dquot shrinker isn't set up. | |
826 | * The caller is responsible for _qm_dqdestroy'ing the returned dquot. | |
827 | */ | |
828 | int | |
829 | xfs_qm_dqget_uncached( | |
830 | struct xfs_mount *mp, | |
831 | xfs_dqid_t id, | |
832 | uint type, | |
833 | struct xfs_dquot **dqpp) | |
834 | { | |
835 | int error; | |
836 | ||
837 | error = xfs_qm_dqget_checks(mp, type); | |
838 | if (error) | |
839 | return error; | |
840 | ||
841 | return xfs_qm_dqread(mp, id, type, 0, dqpp); | |
842 | } | |
843 | ||
4882c19d DW |
844 | /* Return the quota id for a given inode and type. */ |
845 | xfs_dqid_t | |
846 | xfs_qm_id_for_quotatype( | |
847 | struct xfs_inode *ip, | |
848 | uint type) | |
849 | { | |
850 | switch (type) { | |
851 | case XFS_DQ_USER: | |
852 | return ip->i_d.di_uid; | |
853 | case XFS_DQ_GROUP: | |
854 | return ip->i_d.di_gid; | |
855 | case XFS_DQ_PROJ: | |
856 | return xfs_get_projid(ip); | |
1da177e4 | 857 | } |
4882c19d DW |
858 | ASSERT(0); |
859 | return 0; | |
860 | } | |
861 | ||
862 | /* | |
863 | * Return the dquot for a given inode and type. If @can_alloc is true, then | |
864 | * allocate blocks if needed. The inode's ILOCK must be held and it must not | |
865 | * have already had an inode attached. | |
866 | */ | |
867 | int | |
868 | xfs_qm_dqget_inode( | |
869 | struct xfs_inode *ip, | |
870 | uint type, | |
871 | bool can_alloc, | |
872 | struct xfs_dquot **O_dqpp) | |
873 | { | |
874 | struct xfs_mount *mp = ip->i_mount; | |
875 | struct xfs_quotainfo *qi = mp->m_quotainfo; | |
876 | struct radix_tree_root *tree = xfs_dquot_tree(qi, type); | |
877 | struct xfs_dquot *dqp; | |
878 | xfs_dqid_t id; | |
4882c19d DW |
879 | int error; |
880 | ||
881 | error = xfs_qm_dqget_checks(mp, type); | |
882 | if (error) | |
883 | return error; | |
884 | ||
4882c19d DW |
885 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
886 | ASSERT(xfs_inode_dquot(ip, type) == NULL); | |
887 | ||
888 | id = xfs_qm_id_for_quotatype(ip, type); | |
92678554 CH |
889 | |
890 | restart: | |
cc2047c4 | 891 | dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id); |
9f920f11 | 892 | if (dqp) { |
9f920f11 CH |
893 | *O_dqpp = dqp; |
894 | return 0; | |
1da177e4 | 895 | } |
1da177e4 LT |
896 | |
897 | /* | |
898 | * Dquot cache miss. We don't want to keep the inode lock across | |
899 | * a (potential) disk read. Also we don't want to deal with the lock | |
900 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
901 | * lock here means dealing with a chown that can happen before | |
902 | * we re-acquire the lock. | |
903 | */ | |
4882c19d | 904 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
30ab2dcf | 905 | error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp); |
4882c19d | 906 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
7ae44407 CH |
907 | if (error) |
908 | return error; | |
1da177e4 | 909 | |
4882c19d DW |
910 | /* |
911 | * A dquot could be attached to this inode by now, since we had | |
912 | * dropped the ilock. | |
913 | */ | |
914 | if (xfs_this_quota_on(mp, type)) { | |
915 | struct xfs_dquot *dqp1; | |
916 | ||
917 | dqp1 = xfs_inode_dquot(ip, type); | |
918 | if (dqp1) { | |
36731410 | 919 | xfs_qm_dqdestroy(dqp); |
4882c19d DW |
920 | dqp = dqp1; |
921 | xfs_dqlock(dqp); | |
922 | goto dqret; | |
1da177e4 | 923 | } |
4882c19d DW |
924 | } else { |
925 | /* inode stays locked on return */ | |
926 | xfs_qm_dqdestroy(dqp); | |
927 | return -ESRCH; | |
1da177e4 LT |
928 | } |
929 | ||
cc2047c4 DW |
930 | error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp); |
931 | if (error) { | |
1da177e4 | 932 | /* |
9f920f11 CH |
933 | * Duplicate found. Just throw away the new dquot and start |
934 | * over. | |
1da177e4 | 935 | */ |
9f920f11 | 936 | xfs_qm_dqdestroy(dqp); |
ff6d6af2 | 937 | XFS_STATS_INC(mp, xs_qm_dquot_dups); |
9f920f11 | 938 | goto restart; |
1da177e4 LT |
939 | } |
940 | ||
4882c19d DW |
941 | dqret: |
942 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
0b1b213f | 943 | trace_xfs_dqget_miss(dqp); |
1da177e4 | 944 | *O_dqpp = dqp; |
d99831ff | 945 | return 0; |
1da177e4 LT |
946 | } |
947 | ||
2e330e76 DW |
948 | /* |
949 | * Starting at @id and progressing upwards, look for an initialized incore | |
950 | * dquot, lock it, and return it. | |
951 | */ | |
952 | int | |
953 | xfs_qm_dqget_next( | |
954 | struct xfs_mount *mp, | |
955 | xfs_dqid_t id, | |
956 | uint type, | |
957 | struct xfs_dquot **dqpp) | |
958 | { | |
959 | struct xfs_dquot *dqp; | |
960 | int error = 0; | |
961 | ||
962 | *dqpp = NULL; | |
963 | for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) { | |
30ab2dcf | 964 | error = xfs_qm_dqget(mp, id, type, false, &dqp); |
2e330e76 DW |
965 | if (error == -ENOENT) |
966 | continue; | |
967 | else if (error != 0) | |
968 | break; | |
969 | ||
970 | if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) { | |
971 | *dqpp = dqp; | |
972 | return 0; | |
973 | } | |
974 | ||
975 | xfs_qm_dqput(dqp); | |
976 | } | |
977 | ||
978 | return error; | |
979 | } | |
980 | ||
f8739c3c CH |
981 | /* |
982 | * Release a reference to the dquot (decrement ref-count) and unlock it. | |
983 | * | |
984 | * If there is a group quota attached to this dquot, carefully release that | |
985 | * too without tripping over deadlocks'n'stuff. | |
986 | */ | |
987 | void | |
988 | xfs_qm_dqput( | |
989 | struct xfs_dquot *dqp) | |
990 | { | |
991 | ASSERT(dqp->q_nrefs > 0); | |
992 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
993 | ||
994 | trace_xfs_dqput(dqp); | |
995 | ||
3c353375 DC |
996 | if (--dqp->q_nrefs == 0) { |
997 | struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo; | |
998 | trace_xfs_dqput_free(dqp); | |
999 | ||
1000 | if (list_lru_add(&qi->qi_lru, &dqp->q_lru)) | |
ff6d6af2 | 1001 | XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused); |
3c353375 DC |
1002 | } |
1003 | xfs_dqunlock(dqp); | |
1da177e4 LT |
1004 | } |
1005 | ||
1006 | /* | |
1007 | * Release a dquot. Flush it if dirty, then dqput() it. | |
1008 | * dquot must not be locked. | |
1009 | */ | |
1010 | void | |
1011 | xfs_qm_dqrele( | |
1012 | xfs_dquot_t *dqp) | |
1013 | { | |
7d095257 CH |
1014 | if (!dqp) |
1015 | return; | |
1016 | ||
0b1b213f | 1017 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
1018 | |
1019 | xfs_dqlock(dqp); | |
1020 | /* | |
1021 | * We don't care to flush it if the dquot is dirty here. | |
1022 | * That will create stutters that we want to avoid. | |
1023 | * Instead we do a delayed write when we try to reclaim | |
1024 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
1025 | */ | |
1026 | xfs_qm_dqput(dqp); | |
1027 | } | |
1028 | ||
ca30b2a7 CH |
1029 | /* |
1030 | * This is the dquot flushing I/O completion routine. It is called | |
1031 | * from interrupt level when the buffer containing the dquot is | |
1032 | * flushed to disk. It is responsible for removing the dquot logitem | |
1033 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
1034 | * flush lock. This behavior is very similar to that of inodes.. | |
1035 | */ | |
1036 | STATIC void | |
1037 | xfs_qm_dqflush_done( | |
1038 | struct xfs_buf *bp, | |
1039 | struct xfs_log_item *lip) | |
1040 | { | |
1041 | xfs_dq_logitem_t *qip = (struct xfs_dq_logitem *)lip; | |
1042 | xfs_dquot_t *dqp = qip->qli_dquot; | |
1043 | struct xfs_ail *ailp = lip->li_ailp; | |
1044 | ||
1045 | /* | |
1046 | * We only want to pull the item from the AIL if its | |
1047 | * location in the log has not changed since we started the flush. | |
1048 | * Thus, we only bother if the dquot's lsn has | |
1049 | * not changed. First we check the lsn outside the lock | |
1050 | * since it's cheaper, and then we recheck while | |
1051 | * holding the lock before removing the dquot from the AIL. | |
1052 | */ | |
22525c17 | 1053 | if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) && |
373b0589 | 1054 | ((lip->li_lsn == qip->qli_flush_lsn) || |
22525c17 | 1055 | test_bit(XFS_LI_FAILED, &lip->li_flags))) { |
ca30b2a7 CH |
1056 | |
1057 | /* xfs_trans_ail_delete() drops the AIL lock. */ | |
57e80956 | 1058 | spin_lock(&ailp->ail_lock); |
373b0589 | 1059 | if (lip->li_lsn == qip->qli_flush_lsn) { |
04913fdd | 1060 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
373b0589 CM |
1061 | } else { |
1062 | /* | |
1063 | * Clear the failed state since we are about to drop the | |
1064 | * flush lock | |
1065 | */ | |
22525c17 | 1066 | xfs_clear_li_failed(lip); |
57e80956 | 1067 | spin_unlock(&ailp->ail_lock); |
373b0589 | 1068 | } |
ca30b2a7 CH |
1069 | } |
1070 | ||
1071 | /* | |
1072 | * Release the dq's flush lock since we're done with it. | |
1073 | */ | |
1074 | xfs_dqfunlock(dqp); | |
1075 | } | |
1da177e4 LT |
1076 | |
1077 | /* | |
1078 | * Write a modified dquot to disk. | |
1079 | * The dquot must be locked and the flush lock too taken by caller. | |
1080 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
1081 | * but the dquot is free to be unlocked and modified by the caller | |
1082 | * in the interim. Dquot is still locked on return. This behavior is | |
1083 | * identical to that of inodes. | |
1084 | */ | |
1085 | int | |
1086 | xfs_qm_dqflush( | |
fe7257fd CH |
1087 | struct xfs_dquot *dqp, |
1088 | struct xfs_buf **bpp) | |
1da177e4 | 1089 | { |
acecf1b5 CH |
1090 | struct xfs_mount *mp = dqp->q_mount; |
1091 | struct xfs_buf *bp; | |
7224fa48 | 1092 | struct xfs_dqblk *dqb; |
acecf1b5 | 1093 | struct xfs_disk_dquot *ddqp; |
eebf3cab | 1094 | xfs_failaddr_t fa; |
1da177e4 | 1095 | int error; |
1da177e4 LT |
1096 | |
1097 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 1098 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 1099 | |
0b1b213f | 1100 | trace_xfs_dqflush(dqp); |
1da177e4 | 1101 | |
fe7257fd CH |
1102 | *bpp = NULL; |
1103 | ||
1da177e4 LT |
1104 | xfs_qm_dqunpin_wait(dqp); |
1105 | ||
1106 | /* | |
1107 | * This may have been unpinned because the filesystem is shutting | |
1108 | * down forcibly. If that's the case we must not write this dquot | |
dea96095 CH |
1109 | * to disk, because the log record didn't make it to disk. |
1110 | * | |
1111 | * We also have to remove the log item from the AIL in this case, | |
1112 | * as we wait for an emptry AIL as part of the unmount process. | |
1da177e4 | 1113 | */ |
acecf1b5 | 1114 | if (XFS_FORCED_SHUTDOWN(mp)) { |
dea96095 | 1115 | struct xfs_log_item *lip = &dqp->q_logitem.qli_item; |
acecf1b5 | 1116 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
dea96095 | 1117 | |
146e54b7 BF |
1118 | xfs_trans_ail_remove(lip, SHUTDOWN_CORRUPT_INCORE); |
1119 | ||
2451337d | 1120 | error = -EIO; |
fe7257fd | 1121 | goto out_unlock; |
1da177e4 LT |
1122 | } |
1123 | ||
1124 | /* | |
1125 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1126 | */ |
acecf1b5 | 1127 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
5fd364fe DC |
1128 | mp->m_quotainfo->qi_dqchunklen, 0, &bp, |
1129 | &xfs_dquot_buf_ops); | |
fe7257fd CH |
1130 | if (error) |
1131 | goto out_unlock; | |
1da177e4 | 1132 | |
acecf1b5 CH |
1133 | /* |
1134 | * Calculate the location of the dquot inside the buffer. | |
1135 | */ | |
7224fa48 ES |
1136 | dqb = bp->b_addr + dqp->q_bufoffset; |
1137 | ddqp = &dqb->dd_diskdq; | |
acecf1b5 CH |
1138 | |
1139 | /* | |
7224fa48 | 1140 | * A simple sanity check in case we got a corrupted dquot. |
acecf1b5 | 1141 | */ |
7224fa48 | 1142 | fa = xfs_dqblk_verify(mp, dqb, be32_to_cpu(ddqp->d_id), 0); |
eebf3cab DW |
1143 | if (fa) { |
1144 | xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS", | |
1145 | be32_to_cpu(ddqp->d_id), fa); | |
acecf1b5 CH |
1146 | xfs_buf_relse(bp); |
1147 | xfs_dqfunlock(dqp); | |
1148 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
2451337d | 1149 | return -EIO; |
1da177e4 LT |
1150 | } |
1151 | ||
1152 | /* This is the only portion of data that needs to persist */ | |
acecf1b5 | 1153 | memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t)); |
1da177e4 LT |
1154 | |
1155 | /* | |
1156 | * Clear the dirty field and remember the flush lsn for later use. | |
1157 | */ | |
acecf1b5 | 1158 | dqp->dq_flags &= ~XFS_DQ_DIRTY; |
1da177e4 | 1159 | |
7b2e2a31 DC |
1160 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1161 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 | 1162 | |
3fe58f30 CH |
1163 | /* |
1164 | * copy the lsn into the on-disk dquot now while we have the in memory | |
1165 | * dquot here. This can't be done later in the write verifier as we | |
1166 | * can't get access to the log item at that point in time. | |
6fcdc59d DC |
1167 | * |
1168 | * We also calculate the CRC here so that the on-disk dquot in the | |
1169 | * buffer always has a valid CRC. This ensures there is no possibility | |
1170 | * of a dquot without an up-to-date CRC getting to disk. | |
3fe58f30 CH |
1171 | */ |
1172 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
3fe58f30 | 1173 | dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn); |
6fcdc59d DC |
1174 | xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk), |
1175 | XFS_DQUOT_CRC_OFF); | |
3fe58f30 CH |
1176 | } |
1177 | ||
1da177e4 LT |
1178 | /* |
1179 | * Attach an iodone routine so that we can remove this dquot from the | |
1180 | * AIL and release the flush lock once the dquot is synced to disk. | |
1181 | */ | |
ca30b2a7 CH |
1182 | xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done, |
1183 | &dqp->q_logitem.qli_item); | |
1184 | ||
1da177e4 LT |
1185 | /* |
1186 | * If the buffer is pinned then push on the log so we won't | |
1187 | * get stuck waiting in the write for too long. | |
1188 | */ | |
811e64c7 | 1189 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1190 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1191 | xfs_log_force(mp, 0); |
1da177e4 LT |
1192 | } |
1193 | ||
0b1b213f | 1194 | trace_xfs_dqflush_done(dqp); |
fe7257fd CH |
1195 | *bpp = bp; |
1196 | return 0; | |
0b1b213f | 1197 | |
fe7257fd CH |
1198 | out_unlock: |
1199 | xfs_dqfunlock(dqp); | |
2451337d | 1200 | return -EIO; |
1da177e4 LT |
1201 | } |
1202 | ||
5bb87a33 CH |
1203 | /* |
1204 | * Lock two xfs_dquot structures. | |
1205 | * | |
1206 | * To avoid deadlocks we always lock the quota structure with | |
1207 | * the lowerd id first. | |
1208 | */ | |
1da177e4 LT |
1209 | void |
1210 | xfs_dqlock2( | |
1211 | xfs_dquot_t *d1, | |
1212 | xfs_dquot_t *d2) | |
1213 | { | |
1214 | if (d1 && d2) { | |
1215 | ASSERT(d1 != d2); | |
1149d96a CH |
1216 | if (be32_to_cpu(d1->q_core.d_id) > |
1217 | be32_to_cpu(d2->q_core.d_id)) { | |
5bb87a33 CH |
1218 | mutex_lock(&d2->q_qlock); |
1219 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1220 | } else { |
5bb87a33 CH |
1221 | mutex_lock(&d1->q_qlock); |
1222 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1223 | } |
5bb87a33 CH |
1224 | } else if (d1) { |
1225 | mutex_lock(&d1->q_qlock); | |
1226 | } else if (d2) { | |
1227 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1228 | } |
1229 | } | |
1230 | ||
a05931ce CH |
1231 | int __init |
1232 | xfs_qm_init(void) | |
1233 | { | |
1234 | xfs_qm_dqzone = | |
1235 | kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot"); | |
1236 | if (!xfs_qm_dqzone) | |
1237 | goto out; | |
1238 | ||
1239 | xfs_qm_dqtrxzone = | |
1240 | kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx"); | |
1241 | if (!xfs_qm_dqtrxzone) | |
1242 | goto out_free_dqzone; | |
1243 | ||
1244 | return 0; | |
1245 | ||
1246 | out_free_dqzone: | |
1247 | kmem_zone_destroy(xfs_qm_dqzone); | |
1248 | out: | |
1249 | return -ENOMEM; | |
1250 | } | |
1251 | ||
1c2ccc66 | 1252 | void |
a05931ce CH |
1253 | xfs_qm_exit(void) |
1254 | { | |
1255 | kmem_zone_destroy(xfs_qm_dqtrxzone); | |
1256 | kmem_zone_destroy(xfs_qm_dqzone); | |
1257 | } | |
554ba965 DW |
1258 | |
1259 | /* | |
1260 | * Iterate every dquot of a particular type. The caller must ensure that the | |
1261 | * particular quota type is active. iter_fn can return negative error codes, | |
1262 | * or XFS_BTREE_QUERY_RANGE_ABORT to indicate that it wants to stop iterating. | |
1263 | */ | |
1264 | int | |
1265 | xfs_qm_dqiterate( | |
1266 | struct xfs_mount *mp, | |
1267 | uint dqtype, | |
1268 | xfs_qm_dqiterate_fn iter_fn, | |
1269 | void *priv) | |
1270 | { | |
1271 | struct xfs_dquot *dq; | |
1272 | xfs_dqid_t id = 0; | |
1273 | int error; | |
1274 | ||
1275 | do { | |
1276 | error = xfs_qm_dqget_next(mp, id, dqtype, &dq); | |
1277 | if (error == -ENOENT) | |
1278 | return 0; | |
1279 | if (error) | |
1280 | return error; | |
1281 | ||
1282 | error = iter_fn(dq, dqtype, priv); | |
1283 | id = be32_to_cpu(dq->q_core.d_id); | |
1284 | xfs_qm_dqput(dq); | |
1285 | id++; | |
1286 | } while (error == 0 && id != 0); | |
1287 | ||
1288 | return error; | |
1289 | } |