]>
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" | |
239880ef | 17 | #include "xfs_quota.h" |
239880ef | 18 | #include "xfs_trans.h" |
1da177e4 LT |
19 | #include "xfs_buf_item.h" |
20 | #include "xfs_trans_space.h" | |
21 | #include "xfs_trans_priv.h" | |
1da177e4 | 22 | #include "xfs_qm.h" |
0b1b213f | 23 | #include "xfs_trace.h" |
239880ef | 24 | #include "xfs_log.h" |
a4fbe6ab | 25 | #include "xfs_bmap_btree.h" |
afeda600 | 26 | #include "xfs_error.h" |
1da177e4 | 27 | |
1da177e4 | 28 | /* |
bf72de31 CH |
29 | * Lock order: |
30 | * | |
31 | * ip->i_lock | |
9f920f11 | 32 | * qi->qi_tree_lock |
b84a3a96 CH |
33 | * dquot->q_qlock (xfs_dqlock() and friends) |
34 | * dquot->q_flush (xfs_dqflock() and friends) | |
35 | * qi->qi_lru_lock | |
bf72de31 CH |
36 | * |
37 | * If two dquots need to be locked the order is user before group/project, | |
38 | * otherwise by the lowest id first, see xfs_dqlock2. | |
39 | */ | |
1da177e4 | 40 | |
a05931ce CH |
41 | struct kmem_zone *xfs_qm_dqtrxzone; |
42 | static struct kmem_zone *xfs_qm_dqzone; | |
43 | ||
f112a049 DC |
44 | static struct lock_class_key xfs_dquot_group_class; |
45 | static struct lock_class_key xfs_dquot_project_class; | |
98b8c7a0 | 46 | |
1da177e4 LT |
47 | /* |
48 | * This is called to free all the memory associated with a dquot | |
49 | */ | |
50 | void | |
51 | xfs_qm_dqdestroy( | |
aefe69a4 | 52 | struct xfs_dquot *dqp) |
1da177e4 | 53 | { |
f8739c3c | 54 | ASSERT(list_empty(&dqp->q_lru)); |
1da177e4 | 55 | |
b1c5ebb2 | 56 | kmem_free(dqp->q_logitem.qli_item.li_lv_shadow); |
1da177e4 | 57 | mutex_destroy(&dqp->q_qlock); |
0b1b213f | 58 | |
ff6d6af2 | 59 | XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot); |
377bcd5f | 60 | kmem_cache_free(xfs_qm_dqzone, dqp); |
1da177e4 LT |
61 | } |
62 | ||
1da177e4 LT |
63 | /* |
64 | * If default limits are in force, push them into the dquot now. | |
65 | * We overwrite the dquot limits only if they are zero and this | |
66 | * is not the root dquot. | |
67 | */ | |
68 | void | |
69 | xfs_qm_adjust_dqlimits( | |
4b6eae2e BF |
70 | struct xfs_mount *mp, |
71 | struct xfs_dquot *dq) | |
1da177e4 | 72 | { |
4b6eae2e | 73 | struct xfs_quotainfo *q = mp->m_quotainfo; |
be607946 | 74 | struct xfs_def_quota *defq; |
b1366451 | 75 | int prealloc = 0; |
1da177e4 | 76 | |
c51df733 | 77 | ASSERT(dq->q_id); |
ce6e7e79 | 78 | defq = xfs_get_defquota(q, xfs_dquot_type(dq)); |
1da177e4 | 79 | |
d3537cf9 DW |
80 | if (defq->bsoftlimit && !dq->q_blk.softlimit) { |
81 | dq->q_blk.softlimit = defq->bsoftlimit; | |
b1366451 BF |
82 | prealloc = 1; |
83 | } | |
d3537cf9 DW |
84 | if (defq->bhardlimit && !dq->q_blk.hardlimit) { |
85 | dq->q_blk.hardlimit = defq->bhardlimit; | |
b1366451 BF |
86 | prealloc = 1; |
87 | } | |
d3537cf9 DW |
88 | if (defq->isoftlimit && !dq->q_ino.softlimit) |
89 | dq->q_ino.softlimit = defq->isoftlimit; | |
90 | if (defq->ihardlimit && !dq->q_ino.hardlimit) | |
91 | dq->q_ino.hardlimit = defq->ihardlimit; | |
92 | if (defq->rtbsoftlimit && !dq->q_rtb.softlimit) | |
93 | dq->q_rtb.softlimit = defq->rtbsoftlimit; | |
94 | if (defq->rtbhardlimit && !dq->q_rtb.hardlimit) | |
95 | dq->q_rtb.hardlimit = defq->rtbhardlimit; | |
b1366451 BF |
96 | |
97 | if (prealloc) | |
98 | xfs_dquot_set_prealloc_limits(dq); | |
1da177e4 LT |
99 | } |
100 | ||
101 | /* | |
102 | * Check the limits and timers of a dquot and start or reset timers | |
103 | * if necessary. | |
104 | * This gets called even when quota enforcement is OFF, which makes our | |
105 | * life a little less complicated. (We just don't reject any quota | |
106 | * reservations in that case, when enforcement is off). | |
107 | * We also return 0 as the values of the timers in Q_GETQUOTA calls, when | |
108 | * enforcement's off. | |
109 | * In contrast, warnings are a little different in that they don't | |
754002b4 NS |
110 | * 'automatically' get started when limits get exceeded. They do |
111 | * get reset to zero, however, when we find the count to be under | |
112 | * the soft limit (they are only ever set non-zero via userspace). | |
1da177e4 LT |
113 | */ |
114 | void | |
115 | xfs_qm_adjust_dqtimers( | |
aefe69a4 | 116 | struct xfs_mount *mp, |
3dbb9aa3 | 117 | struct xfs_dquot *dq) |
1da177e4 | 118 | { |
e850301f | 119 | struct xfs_quotainfo *qi = mp->m_quotainfo; |
3dbb9aa3 | 120 | struct xfs_disk_dquot *d = &dq->q_core; |
e850301f ES |
121 | struct xfs_def_quota *defq; |
122 | ||
c51df733 | 123 | ASSERT(dq->q_id); |
e850301f | 124 | defq = xfs_get_defquota(qi, xfs_dquot_type(dq)); |
1da177e4 | 125 | |
ea15ab3c | 126 | #ifdef DEBUG |
d3537cf9 DW |
127 | if (dq->q_blk.hardlimit) |
128 | ASSERT(dq->q_blk.softlimit <= dq->q_blk.hardlimit); | |
129 | if (dq->q_ino.hardlimit) | |
130 | ASSERT(dq->q_ino.softlimit <= dq->q_ino.hardlimit); | |
131 | if (dq->q_rtb.hardlimit) | |
132 | ASSERT(dq->q_rtb.softlimit <= dq->q_rtb.hardlimit); | |
1da177e4 | 133 | #endif |
ea15ab3c | 134 | |
1da177e4 | 135 | if (!d->d_btimer) { |
d3537cf9 | 136 | if ((dq->q_blk.softlimit && |
be37d40c | 137 | (dq->q_blk.count > dq->q_blk.softlimit)) || |
d3537cf9 | 138 | (dq->q_blk.hardlimit && |
be37d40c | 139 | (dq->q_blk.count > dq->q_blk.hardlimit))) { |
b8a0880a | 140 | d->d_btimer = cpu_to_be32(ktime_get_real_seconds() + |
e850301f | 141 | defq->btimelimit); |
754002b4 | 142 | } else { |
c8c45fb2 | 143 | dq->q_blk.warnings = 0; |
1da177e4 LT |
144 | } |
145 | } else { | |
d3537cf9 | 146 | if ((!dq->q_blk.softlimit || |
be37d40c | 147 | (dq->q_blk.count <= dq->q_blk.softlimit)) && |
d3537cf9 | 148 | (!dq->q_blk.hardlimit || |
be37d40c | 149 | (dq->q_blk.count <= dq->q_blk.hardlimit))) { |
1da177e4 LT |
150 | d->d_btimer = 0; |
151 | } | |
152 | } | |
153 | ||
154 | if (!d->d_itimer) { | |
d3537cf9 | 155 | if ((dq->q_ino.softlimit && |
be37d40c | 156 | (dq->q_ino.count > dq->q_ino.softlimit)) || |
d3537cf9 | 157 | (dq->q_ino.hardlimit && |
be37d40c | 158 | (dq->q_ino.count > dq->q_ino.hardlimit))) { |
b8a0880a | 159 | d->d_itimer = cpu_to_be32(ktime_get_real_seconds() + |
e850301f | 160 | defq->itimelimit); |
754002b4 | 161 | } else { |
c8c45fb2 | 162 | dq->q_ino.warnings = 0; |
1da177e4 LT |
163 | } |
164 | } else { | |
d3537cf9 | 165 | if ((!dq->q_ino.softlimit || |
be37d40c | 166 | (dq->q_ino.count <= dq->q_ino.softlimit)) && |
d3537cf9 | 167 | (!dq->q_ino.hardlimit || |
be37d40c | 168 | (dq->q_ino.count <= dq->q_ino.hardlimit))) { |
1da177e4 LT |
169 | d->d_itimer = 0; |
170 | } | |
171 | } | |
172 | ||
173 | if (!d->d_rtbtimer) { | |
d3537cf9 | 174 | if ((dq->q_rtb.softlimit && |
be37d40c | 175 | (dq->q_rtb.count > dq->q_rtb.softlimit)) || |
d3537cf9 | 176 | (dq->q_rtb.hardlimit && |
be37d40c | 177 | (dq->q_rtb.count > dq->q_rtb.hardlimit))) { |
b8a0880a | 178 | d->d_rtbtimer = cpu_to_be32(ktime_get_real_seconds() + |
e850301f | 179 | defq->rtbtimelimit); |
754002b4 | 180 | } else { |
c8c45fb2 | 181 | dq->q_rtb.warnings = 0; |
1da177e4 LT |
182 | } |
183 | } else { | |
d3537cf9 | 184 | if ((!dq->q_rtb.softlimit || |
be37d40c | 185 | (dq->q_rtb.count <= dq->q_rtb.softlimit)) && |
d3537cf9 | 186 | (!dq->q_rtb.hardlimit || |
be37d40c | 187 | (dq->q_rtb.count <= dq->q_rtb.hardlimit))) { |
1da177e4 LT |
188 | d->d_rtbtimer = 0; |
189 | } | |
190 | } | |
191 | } | |
192 | ||
1da177e4 LT |
193 | /* |
194 | * initialize a buffer full of dquots and log the whole thing | |
195 | */ | |
196 | STATIC void | |
197 | xfs_qm_init_dquot_blk( | |
78bba5c8 DW |
198 | struct xfs_trans *tp, |
199 | struct xfs_mount *mp, | |
200 | xfs_dqid_t id, | |
201 | uint type, | |
202 | struct xfs_buf *bp) | |
1da177e4 | 203 | { |
8a7b8a89 | 204 | struct xfs_quotainfo *q = mp->m_quotainfo; |
78bba5c8 DW |
205 | struct xfs_dqblk *d; |
206 | xfs_dqid_t curid; | |
207 | unsigned int qflag; | |
208 | unsigned int blftype; | |
209 | int i; | |
1da177e4 LT |
210 | |
211 | ASSERT(tp); | |
0c842ad4 | 212 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 213 | |
62926044 | 214 | d = bp->b_addr; |
1da177e4 LT |
215 | |
216 | /* | |
217 | * ID of the first dquot in the block - id's are zero based. | |
218 | */ | |
8a7b8a89 | 219 | curid = id - (id % q->qi_dqperchunk); |
8a7b8a89 | 220 | memset(d, 0, BBTOB(q->qi_dqchunklen)); |
49d35a5c CH |
221 | for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) { |
222 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); | |
223 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; | |
224 | d->dd_diskdq.d_id = cpu_to_be32(curid); | |
225 | d->dd_diskdq.d_flags = type; | |
6fcdc59d | 226 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
92863451 | 227 | uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid); |
6fcdc59d DC |
228 | xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk), |
229 | XFS_DQUOT_CRC_OFF); | |
230 | } | |
49d35a5c CH |
231 | } |
232 | ||
78bba5c8 DW |
233 | if (type & XFS_DQ_USER) { |
234 | qflag = XFS_UQUOTA_CHKD; | |
235 | blftype = XFS_BLF_UDQUOT_BUF; | |
236 | } else if (type & XFS_DQ_PROJ) { | |
237 | qflag = XFS_PQUOTA_CHKD; | |
238 | blftype = XFS_BLF_PDQUOT_BUF; | |
239 | } else { | |
240 | qflag = XFS_GQUOTA_CHKD; | |
241 | blftype = XFS_BLF_GDQUOT_BUF; | |
242 | } | |
243 | ||
244 | xfs_trans_dquot_buf(tp, bp, blftype); | |
245 | ||
246 | /* | |
247 | * quotacheck uses delayed writes to update all the dquots on disk in an | |
248 | * efficient manner instead of logging the individual dquot changes as | |
249 | * they are made. However if we log the buffer allocated here and crash | |
250 | * after quotacheck while the logged initialisation is still in the | |
251 | * active region of the log, log recovery can replay the dquot buffer | |
252 | * initialisation over the top of the checked dquots and corrupt quota | |
253 | * accounting. | |
254 | * | |
255 | * To avoid this problem, quotacheck cannot log the initialised buffer. | |
256 | * We must still dirty the buffer and write it back before the | |
257 | * allocation transaction clears the log. Therefore, mark the buffer as | |
258 | * ordered instead of logging it directly. This is safe for quotacheck | |
259 | * because it detects and repairs allocated but initialized dquot blocks | |
260 | * in the quota inodes. | |
261 | */ | |
262 | if (!(mp->m_qflags & qflag)) | |
263 | xfs_trans_ordered_buf(tp, bp); | |
264 | else | |
265 | xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); | |
1da177e4 LT |
266 | } |
267 | ||
b1366451 BF |
268 | /* |
269 | * Initialize the dynamic speculative preallocation thresholds. The lo/hi | |
270 | * watermarks correspond to the soft and hard limits by default. If a soft limit | |
271 | * is not specified, we use 95% of the hard limit. | |
272 | */ | |
273 | void | |
274 | xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp) | |
275 | { | |
c8ce540d | 276 | uint64_t space; |
b1366451 | 277 | |
d3537cf9 DW |
278 | dqp->q_prealloc_hi_wmark = dqp->q_blk.hardlimit; |
279 | dqp->q_prealloc_lo_wmark = dqp->q_blk.softlimit; | |
b1366451 BF |
280 | if (!dqp->q_prealloc_lo_wmark) { |
281 | dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark; | |
282 | do_div(dqp->q_prealloc_lo_wmark, 100); | |
283 | dqp->q_prealloc_lo_wmark *= 95; | |
284 | } | |
285 | ||
286 | space = dqp->q_prealloc_hi_wmark; | |
287 | ||
288 | do_div(space, 100); | |
289 | dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space; | |
290 | dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3; | |
291 | dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5; | |
292 | } | |
293 | ||
1da177e4 | 294 | /* |
d63192c8 | 295 | * Ensure that the given in-core dquot has a buffer on disk backing it, and |
710d707d DW |
296 | * return the buffer locked and held. This is called when the bmapi finds a |
297 | * hole. | |
1da177e4 LT |
298 | */ |
299 | STATIC int | |
d63192c8 DW |
300 | xfs_dquot_disk_alloc( |
301 | struct xfs_trans **tpp, | |
302 | struct xfs_dquot *dqp, | |
303 | struct xfs_buf **bpp) | |
1da177e4 | 304 | { |
d63192c8 | 305 | struct xfs_bmbt_irec map; |
2ba13721 BF |
306 | struct xfs_trans *tp = *tpp; |
307 | struct xfs_mount *mp = tp->t_mountp; | |
d63192c8 DW |
308 | struct xfs_buf *bp; |
309 | struct xfs_inode *quotip = xfs_quota_inode(mp, dqp->dq_flags); | |
d63192c8 DW |
310 | int nmaps = 1; |
311 | int error; | |
0b1b213f CH |
312 | |
313 | trace_xfs_dqalloc(dqp); | |
1da177e4 | 314 | |
1da177e4 | 315 | xfs_ilock(quotip, XFS_ILOCK_EXCL); |
6967b964 | 316 | if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) { |
d63192c8 DW |
317 | /* |
318 | * Return if this type of quotas is turned off while we didn't | |
319 | * have an inode lock | |
320 | */ | |
1da177e4 | 321 | xfs_iunlock(quotip, XFS_ILOCK_EXCL); |
2451337d | 322 | return -ESRCH; |
1da177e4 LT |
323 | } |
324 | ||
d63192c8 | 325 | /* Create the block mapping. */ |
2ba13721 BF |
326 | xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL); |
327 | error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset, | |
da781e64 BF |
328 | XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0, &map, |
329 | &nmaps); | |
c0dc7828 | 330 | if (error) |
73971b17 | 331 | return error; |
1da177e4 LT |
332 | ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB); |
333 | ASSERT(nmaps == 1); | |
334 | ASSERT((map.br_startblock != DELAYSTARTBLOCK) && | |
335 | (map.br_startblock != HOLESTARTBLOCK)); | |
336 | ||
337 | /* | |
338 | * Keep track of the blkno to save a lookup later | |
339 | */ | |
340 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); | |
341 | ||
342 | /* now we can just get the buffer (there's nothing to read yet) */ | |
ce92464c DW |
343 | error = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno, |
344 | mp->m_quotainfo->qi_dqchunklen, 0, &bp); | |
345 | if (error) | |
346 | return error; | |
1813dd64 | 347 | bp->b_ops = &xfs_dquot_buf_ops; |
2a30f36d | 348 | |
1da177e4 LT |
349 | /* |
350 | * Make a chunk of dquots out of this buffer and log | |
351 | * the entire thing. | |
352 | */ | |
c51df733 | 353 | xfs_qm_init_dquot_blk(tp, mp, dqp->q_id, |
c8ad20ff | 354 | dqp->dq_flags & XFS_DQ_ALLTYPES, bp); |
d63192c8 | 355 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
1da177e4 | 356 | |
efa092f3 | 357 | /* |
7b6b50f5 DW |
358 | * Hold the buffer and join it to the dfops so that we'll still own |
359 | * the buffer when we return to the caller. The buffer disposal on | |
360 | * error must be paid attention to very carefully, as it has been | |
361 | * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota | |
362 | * code when allocating a new dquot record" in 2005, and the later | |
363 | * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep | |
364 | * the buffer locked across the _defer_finish call. We can now do | |
365 | * this correctly with xfs_defer_bjoin. | |
efa092f3 | 366 | * |
73971b17 BF |
367 | * Above, we allocated a disk block for the dquot information and used |
368 | * get_buf to initialize the dquot. If the _defer_finish fails, the old | |
7b6b50f5 DW |
369 | * transaction is gone but the new buffer is not joined or held to any |
370 | * transaction, so we must _buf_relse it. | |
efa092f3 | 371 | * |
7b6b50f5 | 372 | * If everything succeeds, the caller of this function is returned a |
d63192c8 | 373 | * buffer that is locked and held to the transaction. The caller |
7b6b50f5 | 374 | * is responsible for unlocking any buffer passed back, either |
710d707d DW |
375 | * manually or by committing the transaction. On error, the buffer is |
376 | * released and not passed back. | |
efa092f3 | 377 | */ |
2ba13721 | 378 | xfs_trans_bhold(tp, bp); |
9e28a242 | 379 | error = xfs_defer_finish(tpp); |
7b6b50f5 | 380 | if (error) { |
710d707d DW |
381 | xfs_trans_bhold_release(*tpp, bp); |
382 | xfs_trans_brelse(*tpp, bp); | |
73971b17 | 383 | return error; |
efa092f3 | 384 | } |
d63192c8 | 385 | *bpp = bp; |
1da177e4 | 386 | return 0; |
1da177e4 | 387 | } |
9aede1d8 | 388 | |
1da177e4 | 389 | /* |
d63192c8 DW |
390 | * Read in the in-core dquot's on-disk metadata and return the buffer. |
391 | * Returns ENOENT to signal a hole. | |
1da177e4 LT |
392 | */ |
393 | STATIC int | |
d63192c8 DW |
394 | xfs_dquot_disk_read( |
395 | struct xfs_mount *mp, | |
396 | struct xfs_dquot *dqp, | |
397 | struct xfs_buf **bpp) | |
1da177e4 | 398 | { |
113a5683 | 399 | struct xfs_bmbt_irec map; |
113a5683 | 400 | struct xfs_buf *bp; |
d63192c8 | 401 | struct xfs_inode *quotip = xfs_quota_inode(mp, dqp->dq_flags); |
0891f997 | 402 | uint lock_mode; |
d63192c8 DW |
403 | int nmaps = 1; |
404 | int error; | |
1da177e4 | 405 | |
0891f997 | 406 | lock_mode = xfs_ilock_data_map_shared(quotip); |
d63192c8 | 407 | if (!xfs_this_quota_on(mp, dqp->dq_flags)) { |
1da177e4 | 408 | /* |
acecf1b5 CH |
409 | * Return if this type of quotas is turned off while we |
410 | * didn't have the quota inode lock. | |
1da177e4 | 411 | */ |
0891f997 | 412 | xfs_iunlock(quotip, lock_mode); |
2451337d | 413 | return -ESRCH; |
acecf1b5 CH |
414 | } |
415 | ||
416 | /* | |
417 | * Find the block map; no allocations yet | |
418 | */ | |
5c8ed202 | 419 | error = xfs_bmapi_read(quotip, dqp->q_fileoffset, |
d63192c8 | 420 | XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0); |
0891f997 | 421 | xfs_iunlock(quotip, lock_mode); |
acecf1b5 CH |
422 | if (error) |
423 | return error; | |
424 | ||
425 | ASSERT(nmaps == 1); | |
d63192c8 DW |
426 | ASSERT(map.br_blockcount >= 1); |
427 | ASSERT(map.br_startblock != DELAYSTARTBLOCK); | |
428 | if (map.br_startblock == HOLESTARTBLOCK) | |
429 | return -ENOENT; | |
430 | ||
431 | trace_xfs_dqtobp_read(dqp); | |
acecf1b5 CH |
432 | |
433 | /* | |
d63192c8 DW |
434 | * store the blkno etc so that we don't have to do the |
435 | * mapping all the time | |
acecf1b5 | 436 | */ |
d63192c8 | 437 | dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock); |
1da177e4 | 438 | |
d63192c8 DW |
439 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
440 | mp->m_quotainfo->qi_dqchunklen, 0, &bp, | |
441 | &xfs_dquot_buf_ops); | |
442 | if (error) { | |
443 | ASSERT(bp == NULL); | |
444 | return error; | |
1da177e4 LT |
445 | } |
446 | ||
c6319198 | 447 | ASSERT(xfs_buf_islocked(bp)); |
d63192c8 DW |
448 | xfs_buf_set_ref(bp, XFS_DQUOT_REF); |
449 | *bpp = bp; | |
1da177e4 | 450 | |
d99831ff | 451 | return 0; |
1da177e4 LT |
452 | } |
453 | ||
617cd5c1 DW |
454 | /* Allocate and initialize everything we need for an incore dquot. */ |
455 | STATIC struct xfs_dquot * | |
456 | xfs_dquot_alloc( | |
97e7ade5 CH |
457 | struct xfs_mount *mp, |
458 | xfs_dqid_t id, | |
617cd5c1 | 459 | uint type) |
1da177e4 | 460 | { |
97e7ade5 | 461 | struct xfs_dquot *dqp; |
92b2e5b3 | 462 | |
707e0dda | 463 | dqp = kmem_zone_zalloc(xfs_qm_dqzone, 0); |
92b2e5b3 CH |
464 | |
465 | dqp->dq_flags = type; | |
c51df733 | 466 | dqp->q_id = id; |
92b2e5b3 | 467 | dqp->q_mount = mp; |
f8739c3c | 468 | INIT_LIST_HEAD(&dqp->q_lru); |
92b2e5b3 CH |
469 | mutex_init(&dqp->q_qlock); |
470 | init_waitqueue_head(&dqp->q_pinwait); | |
d63192c8 DW |
471 | dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk; |
472 | /* | |
473 | * Offset of dquot in the (fixed sized) dquot chunk. | |
474 | */ | |
475 | dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) * | |
476 | sizeof(xfs_dqblk_t); | |
92b2e5b3 CH |
477 | |
478 | /* | |
479 | * Because we want to use a counting completion, complete | |
480 | * the flush completion once to allow a single access to | |
481 | * the flush completion without blocking. | |
482 | */ | |
483 | init_completion(&dqp->q_flush); | |
484 | complete(&dqp->q_flush); | |
485 | ||
486 | /* | |
487 | * Make sure group quotas have a different lock class than user | |
488 | * quotas. | |
489 | */ | |
f112a049 DC |
490 | switch (type) { |
491 | case XFS_DQ_USER: | |
492 | /* uses the default lock class */ | |
493 | break; | |
494 | case XFS_DQ_GROUP: | |
495 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class); | |
496 | break; | |
497 | case XFS_DQ_PROJ: | |
498 | lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class); | |
499 | break; | |
500 | default: | |
501 | ASSERT(0); | |
502 | break; | |
503 | } | |
92b2e5b3 | 504 | |
617cd5c1 DW |
505 | xfs_qm_dquot_logitem_init(dqp); |
506 | ||
ff6d6af2 | 507 | XFS_STATS_INC(mp, xs_qm_dquot); |
617cd5c1 DW |
508 | return dqp; |
509 | } | |
510 | ||
511 | /* Copy the in-core quota fields in from the on-disk buffer. */ | |
afeda600 | 512 | STATIC int |
617cd5c1 DW |
513 | xfs_dquot_from_disk( |
514 | struct xfs_dquot *dqp, | |
d63192c8 | 515 | struct xfs_buf *bp) |
617cd5c1 | 516 | { |
d63192c8 DW |
517 | struct xfs_disk_dquot *ddqp = bp->b_addr + dqp->q_bufoffset; |
518 | ||
afeda600 DW |
519 | /* |
520 | * Ensure that we got the type and ID we were looking for. | |
521 | * Everything else was checked by the dquot buffer verifier. | |
522 | */ | |
523 | if ((ddqp->d_flags & XFS_DQ_ALLTYPES) != dqp->dq_flags || | |
c51df733 | 524 | be32_to_cpu(ddqp->d_id) != dqp->q_id) { |
afeda600 DW |
525 | xfs_alert_tag(bp->b_mount, XFS_PTAG_VERIFIER_ERROR, |
526 | "Metadata corruption detected at %pS, quota %u", | |
c51df733 | 527 | __this_address, dqp->q_id); |
afeda600 DW |
528 | xfs_alert(bp->b_mount, "Unmount and run xfs_repair"); |
529 | return -EFSCORRUPTED; | |
530 | } | |
531 | ||
617cd5c1 | 532 | /* copy everything from disk dquot to the incore dquot */ |
aefe69a4 | 533 | memcpy(&dqp->q_core, ddqp, sizeof(struct xfs_disk_dquot)); |
d3537cf9 DW |
534 | dqp->q_blk.hardlimit = be64_to_cpu(ddqp->d_blk_hardlimit); |
535 | dqp->q_blk.softlimit = be64_to_cpu(ddqp->d_blk_softlimit); | |
536 | dqp->q_ino.hardlimit = be64_to_cpu(ddqp->d_ino_hardlimit); | |
537 | dqp->q_ino.softlimit = be64_to_cpu(ddqp->d_ino_softlimit); | |
538 | dqp->q_rtb.hardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit); | |
539 | dqp->q_rtb.softlimit = be64_to_cpu(ddqp->d_rtb_softlimit); | |
617cd5c1 | 540 | |
be37d40c DW |
541 | dqp->q_blk.count = be64_to_cpu(ddqp->d_bcount); |
542 | dqp->q_ino.count = be64_to_cpu(ddqp->d_icount); | |
543 | dqp->q_rtb.count = be64_to_cpu(ddqp->d_rtbcount); | |
544 | ||
c8c45fb2 DW |
545 | dqp->q_blk.warnings = be16_to_cpu(ddqp->d_bwarns); |
546 | dqp->q_ino.warnings = be16_to_cpu(ddqp->d_iwarns); | |
547 | dqp->q_rtb.warnings = be16_to_cpu(ddqp->d_rtbwarns); | |
548 | ||
617cd5c1 DW |
549 | /* |
550 | * Reservation counters are defined as reservation plus current usage | |
551 | * to avoid having to add every time. | |
552 | */ | |
be37d40c DW |
553 | dqp->q_blk.reserved = dqp->q_blk.count; |
554 | dqp->q_ino.reserved = dqp->q_ino.count; | |
555 | dqp->q_rtb.reserved = dqp->q_rtb.count; | |
617cd5c1 DW |
556 | |
557 | /* initialize the dquot speculative prealloc thresholds */ | |
558 | xfs_dquot_set_prealloc_limits(dqp); | |
afeda600 | 559 | return 0; |
617cd5c1 | 560 | } |
1da177e4 | 561 | |
0b0fa1d1 DW |
562 | /* Copy the in-core quota fields into the on-disk buffer. */ |
563 | void | |
564 | xfs_dquot_to_disk( | |
565 | struct xfs_disk_dquot *ddqp, | |
566 | struct xfs_dquot *dqp) | |
567 | { | |
568 | memcpy(ddqp, &dqp->q_core, sizeof(struct xfs_disk_dquot)); | |
d3537cf9 DW |
569 | ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit); |
570 | ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit); | |
571 | ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit); | |
572 | ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit); | |
573 | ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit); | |
574 | ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit); | |
be37d40c DW |
575 | |
576 | ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count); | |
577 | ddqp->d_icount = cpu_to_be64(dqp->q_ino.count); | |
578 | ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count); | |
c8c45fb2 DW |
579 | |
580 | ddqp->d_bwarns = cpu_to_be16(dqp->q_blk.warnings); | |
581 | ddqp->d_iwarns = cpu_to_be16(dqp->q_ino.warnings); | |
582 | ddqp->d_rtbwarns = cpu_to_be16(dqp->q_rtb.warnings); | |
0b0fa1d1 DW |
583 | } |
584 | ||
d63192c8 DW |
585 | /* Allocate and initialize the dquot buffer for this in-core dquot. */ |
586 | static int | |
587 | xfs_qm_dqread_alloc( | |
588 | struct xfs_mount *mp, | |
589 | struct xfs_dquot *dqp, | |
590 | struct xfs_buf **bpp) | |
591 | { | |
592 | struct xfs_trans *tp; | |
d63192c8 DW |
593 | int error; |
594 | ||
595 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc, | |
596 | XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp); | |
597 | if (error) | |
598 | goto err; | |
599 | ||
710d707d | 600 | error = xfs_dquot_disk_alloc(&tp, dqp, bpp); |
d63192c8 DW |
601 | if (error) |
602 | goto err_cancel; | |
603 | ||
604 | error = xfs_trans_commit(tp); | |
605 | if (error) { | |
606 | /* | |
607 | * Buffer was held to the transaction, so we have to unlock it | |
608 | * manually here because we're not passing it back. | |
609 | */ | |
710d707d DW |
610 | xfs_buf_relse(*bpp); |
611 | *bpp = NULL; | |
d63192c8 DW |
612 | goto err; |
613 | } | |
d63192c8 DW |
614 | return 0; |
615 | ||
616 | err_cancel: | |
617 | xfs_trans_cancel(tp); | |
618 | err: | |
619 | return error; | |
620 | } | |
621 | ||
617cd5c1 DW |
622 | /* |
623 | * Read in the ondisk dquot using dqtobp() then copy it to an incore version, | |
30ab2dcf DW |
624 | * and release the buffer immediately. If @can_alloc is true, fill any |
625 | * holes in the on-disk metadata. | |
617cd5c1 | 626 | */ |
114e73cc | 627 | static int |
617cd5c1 DW |
628 | xfs_qm_dqread( |
629 | struct xfs_mount *mp, | |
630 | xfs_dqid_t id, | |
631 | uint type, | |
30ab2dcf | 632 | bool can_alloc, |
d63192c8 | 633 | struct xfs_dquot **dqpp) |
617cd5c1 DW |
634 | { |
635 | struct xfs_dquot *dqp; | |
617cd5c1 | 636 | struct xfs_buf *bp; |
617cd5c1 DW |
637 | int error; |
638 | ||
639 | dqp = xfs_dquot_alloc(mp, id, type); | |
0b1b213f CH |
640 | trace_xfs_dqread(dqp); |
641 | ||
d63192c8 DW |
642 | /* Try to read the buffer, allocating if necessary. */ |
643 | error = xfs_dquot_disk_read(mp, dqp, &bp); | |
30ab2dcf | 644 | if (error == -ENOENT && can_alloc) |
d63192c8 DW |
645 | error = xfs_qm_dqread_alloc(mp, dqp, &bp); |
646 | if (error) | |
647 | goto err; | |
1da177e4 LT |
648 | |
649 | /* | |
d63192c8 DW |
650 | * At this point we should have a clean locked buffer. Copy the data |
651 | * to the incore dquot and release the buffer since the incore dquot | |
652 | * has its own locking protocol so we needn't tie up the buffer any | |
653 | * further. | |
1da177e4 | 654 | */ |
0c842ad4 | 655 | ASSERT(xfs_buf_islocked(bp)); |
afeda600 | 656 | error = xfs_dquot_from_disk(dqp, bp); |
d63192c8 | 657 | xfs_buf_relse(bp); |
afeda600 DW |
658 | if (error) |
659 | goto err; | |
660 | ||
d63192c8 | 661 | *dqpp = dqp; |
97e7ade5 | 662 | return error; |
1da177e4 | 663 | |
d63192c8 DW |
664 | err: |
665 | trace_xfs_dqread_fail(dqp); | |
1da177e4 | 666 | xfs_qm_dqdestroy(dqp); |
d63192c8 | 667 | *dqpp = NULL; |
97e7ade5 | 668 | return error; |
1da177e4 LT |
669 | } |
670 | ||
296c24e2 ES |
671 | /* |
672 | * Advance to the next id in the current chunk, or if at the | |
673 | * end of the chunk, skip ahead to first id in next allocated chunk | |
674 | * using the SEEK_DATA interface. | |
675 | */ | |
6e3e6d55 | 676 | static int |
296c24e2 | 677 | xfs_dq_get_next_id( |
bda250db | 678 | struct xfs_mount *mp, |
296c24e2 | 679 | uint type, |
bda250db | 680 | xfs_dqid_t *id) |
296c24e2 | 681 | { |
bda250db CH |
682 | struct xfs_inode *quotip = xfs_quota_inode(mp, type); |
683 | xfs_dqid_t next_id = *id + 1; /* simple advance */ | |
684 | uint lock_flags; | |
685 | struct xfs_bmbt_irec got; | |
b2b1712a | 686 | struct xfs_iext_cursor cur; |
296c24e2 | 687 | xfs_fsblock_t start; |
296c24e2 ES |
688 | int error = 0; |
689 | ||
657bdfb7 ES |
690 | /* If we'd wrap past the max ID, stop */ |
691 | if (next_id < *id) | |
692 | return -ENOENT; | |
693 | ||
296c24e2 ES |
694 | /* If new ID is within the current chunk, advancing it sufficed */ |
695 | if (next_id % mp->m_quotainfo->qi_dqperchunk) { | |
696 | *id = next_id; | |
697 | return 0; | |
698 | } | |
699 | ||
700 | /* Nope, next_id is now past the current chunk, so find the next one */ | |
701 | start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk; | |
702 | ||
bda250db CH |
703 | lock_flags = xfs_ilock_data_map_shared(quotip); |
704 | if (!(quotip->i_df.if_flags & XFS_IFEXTENTS)) { | |
705 | error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK); | |
706 | if (error) | |
707 | return error; | |
708 | } | |
296c24e2 | 709 | |
b2b1712a | 710 | if (xfs_iext_lookup_extent(quotip, "ip->i_df, start, &cur, &got)) { |
2192b0ba BF |
711 | /* contiguous chunk, bump startoff for the id calculation */ |
712 | if (got.br_startoff < start) | |
713 | got.br_startoff = start; | |
bda250db | 714 | *id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk; |
2192b0ba | 715 | } else { |
bda250db | 716 | error = -ENOENT; |
2192b0ba BF |
717 | } |
718 | ||
bda250db | 719 | xfs_iunlock(quotip, lock_flags); |
296c24e2 | 720 | |
bda250db | 721 | return error; |
296c24e2 ES |
722 | } |
723 | ||
cc2047c4 DW |
724 | /* |
725 | * Look up the dquot in the in-core cache. If found, the dquot is returned | |
726 | * locked and ready to go. | |
727 | */ | |
728 | static struct xfs_dquot * | |
729 | xfs_qm_dqget_cache_lookup( | |
730 | struct xfs_mount *mp, | |
731 | struct xfs_quotainfo *qi, | |
732 | struct radix_tree_root *tree, | |
733 | xfs_dqid_t id) | |
734 | { | |
735 | struct xfs_dquot *dqp; | |
736 | ||
737 | restart: | |
738 | mutex_lock(&qi->qi_tree_lock); | |
739 | dqp = radix_tree_lookup(tree, id); | |
740 | if (!dqp) { | |
741 | mutex_unlock(&qi->qi_tree_lock); | |
742 | XFS_STATS_INC(mp, xs_qm_dqcachemisses); | |
743 | return NULL; | |
744 | } | |
745 | ||
746 | xfs_dqlock(dqp); | |
985a78fd | 747 | if (dqp->q_flags & XFS_DQFLAG_FREEING) { |
cc2047c4 DW |
748 | xfs_dqunlock(dqp); |
749 | mutex_unlock(&qi->qi_tree_lock); | |
750 | trace_xfs_dqget_freeing(dqp); | |
751 | delay(1); | |
752 | goto restart; | |
753 | } | |
754 | ||
755 | dqp->q_nrefs++; | |
756 | mutex_unlock(&qi->qi_tree_lock); | |
757 | ||
758 | trace_xfs_dqget_hit(dqp); | |
759 | XFS_STATS_INC(mp, xs_qm_dqcachehits); | |
760 | return dqp; | |
761 | } | |
762 | ||
763 | /* | |
764 | * Try to insert a new dquot into the in-core cache. If an error occurs the | |
765 | * caller should throw away the dquot and start over. Otherwise, the dquot | |
766 | * is returned locked (and held by the cache) as if there had been a cache | |
767 | * hit. | |
768 | */ | |
769 | static int | |
770 | xfs_qm_dqget_cache_insert( | |
771 | struct xfs_mount *mp, | |
772 | struct xfs_quotainfo *qi, | |
773 | struct radix_tree_root *tree, | |
774 | xfs_dqid_t id, | |
775 | struct xfs_dquot *dqp) | |
776 | { | |
777 | int error; | |
778 | ||
779 | mutex_lock(&qi->qi_tree_lock); | |
780 | error = radix_tree_insert(tree, id, dqp); | |
781 | if (unlikely(error)) { | |
782 | /* Duplicate found! Caller must try again. */ | |
783 | WARN_ON(error != -EEXIST); | |
784 | mutex_unlock(&qi->qi_tree_lock); | |
785 | trace_xfs_dqget_dup(dqp); | |
786 | return error; | |
787 | } | |
788 | ||
789 | /* Return a locked dquot to the caller, with a reference taken. */ | |
790 | xfs_dqlock(dqp); | |
791 | dqp->q_nrefs = 1; | |
792 | ||
793 | qi->qi_dquots++; | |
794 | mutex_unlock(&qi->qi_tree_lock); | |
795 | ||
796 | return 0; | |
797 | } | |
798 | ||
d7103eeb DW |
799 | /* Check our input parameters. */ |
800 | static int | |
801 | xfs_qm_dqget_checks( | |
802 | struct xfs_mount *mp, | |
803 | uint type) | |
804 | { | |
805 | if (WARN_ON_ONCE(!XFS_IS_QUOTA_RUNNING(mp))) | |
806 | return -ESRCH; | |
807 | ||
808 | switch (type) { | |
809 | case XFS_DQ_USER: | |
810 | if (!XFS_IS_UQUOTA_ON(mp)) | |
811 | return -ESRCH; | |
812 | return 0; | |
813 | case XFS_DQ_GROUP: | |
814 | if (!XFS_IS_GQUOTA_ON(mp)) | |
815 | return -ESRCH; | |
816 | return 0; | |
817 | case XFS_DQ_PROJ: | |
818 | if (!XFS_IS_PQUOTA_ON(mp)) | |
819 | return -ESRCH; | |
820 | return 0; | |
821 | default: | |
822 | WARN_ON_ONCE(0); | |
823 | return -EINVAL; | |
824 | } | |
825 | } | |
826 | ||
1da177e4 | 827 | /* |
4882c19d DW |
828 | * Given the file system, id, and type (UDQUOT/GDQUOT), return a a locked |
829 | * dquot, doing an allocation (if requested) as needed. | |
1da177e4 LT |
830 | */ |
831 | int | |
832 | xfs_qm_dqget( | |
4882c19d DW |
833 | struct xfs_mount *mp, |
834 | xfs_dqid_t id, | |
835 | uint type, | |
30ab2dcf | 836 | bool can_alloc, |
4882c19d | 837 | struct xfs_dquot **O_dqpp) |
1da177e4 | 838 | { |
9f920f11 | 839 | struct xfs_quotainfo *qi = mp->m_quotainfo; |
4882c19d | 840 | struct radix_tree_root *tree = xfs_dquot_tree(qi, type); |
9f920f11 CH |
841 | struct xfs_dquot *dqp; |
842 | int error; | |
1da177e4 | 843 | |
d7103eeb DW |
844 | error = xfs_qm_dqget_checks(mp, type); |
845 | if (error) | |
846 | return error; | |
1da177e4 | 847 | |
4882c19d DW |
848 | restart: |
849 | dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id); | |
850 | if (dqp) { | |
851 | *O_dqpp = dqp; | |
852 | return 0; | |
853 | } | |
854 | ||
30ab2dcf | 855 | error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp); |
4882c19d DW |
856 | if (error) |
857 | return error; | |
858 | ||
859 | error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp); | |
860 | if (error) { | |
861 | /* | |
862 | * Duplicate found. Just throw away the new dquot and start | |
863 | * over. | |
864 | */ | |
865 | xfs_qm_dqdestroy(dqp); | |
866 | XFS_STATS_INC(mp, xs_qm_dquot_dups); | |
867 | goto restart; | |
868 | } | |
869 | ||
870 | trace_xfs_dqget_miss(dqp); | |
871 | *O_dqpp = dqp; | |
872 | return 0; | |
873 | } | |
874 | ||
114e73cc DW |
875 | /* |
876 | * Given a dquot id and type, read and initialize a dquot from the on-disk | |
877 | * metadata. This function is only for use during quota initialization so | |
878 | * it ignores the dquot cache assuming that the dquot shrinker isn't set up. | |
879 | * The caller is responsible for _qm_dqdestroy'ing the returned dquot. | |
880 | */ | |
881 | int | |
882 | xfs_qm_dqget_uncached( | |
883 | struct xfs_mount *mp, | |
884 | xfs_dqid_t id, | |
885 | uint type, | |
886 | struct xfs_dquot **dqpp) | |
887 | { | |
888 | int error; | |
889 | ||
890 | error = xfs_qm_dqget_checks(mp, type); | |
891 | if (error) | |
892 | return error; | |
893 | ||
894 | return xfs_qm_dqread(mp, id, type, 0, dqpp); | |
895 | } | |
896 | ||
4882c19d DW |
897 | /* Return the quota id for a given inode and type. */ |
898 | xfs_dqid_t | |
899 | xfs_qm_id_for_quotatype( | |
900 | struct xfs_inode *ip, | |
901 | uint type) | |
902 | { | |
903 | switch (type) { | |
904 | case XFS_DQ_USER: | |
ba8adad5 | 905 | return i_uid_read(VFS_I(ip)); |
4882c19d | 906 | case XFS_DQ_GROUP: |
ba8adad5 | 907 | return i_gid_read(VFS_I(ip)); |
4882c19d | 908 | case XFS_DQ_PROJ: |
de7a866f | 909 | return ip->i_d.di_projid; |
1da177e4 | 910 | } |
4882c19d DW |
911 | ASSERT(0); |
912 | return 0; | |
913 | } | |
914 | ||
915 | /* | |
916 | * Return the dquot for a given inode and type. If @can_alloc is true, then | |
917 | * allocate blocks if needed. The inode's ILOCK must be held and it must not | |
918 | * have already had an inode attached. | |
919 | */ | |
920 | int | |
921 | xfs_qm_dqget_inode( | |
922 | struct xfs_inode *ip, | |
923 | uint type, | |
924 | bool can_alloc, | |
925 | struct xfs_dquot **O_dqpp) | |
926 | { | |
927 | struct xfs_mount *mp = ip->i_mount; | |
928 | struct xfs_quotainfo *qi = mp->m_quotainfo; | |
929 | struct radix_tree_root *tree = xfs_dquot_tree(qi, type); | |
930 | struct xfs_dquot *dqp; | |
931 | xfs_dqid_t id; | |
4882c19d DW |
932 | int error; |
933 | ||
934 | error = xfs_qm_dqget_checks(mp, type); | |
935 | if (error) | |
936 | return error; | |
937 | ||
4882c19d DW |
938 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
939 | ASSERT(xfs_inode_dquot(ip, type) == NULL); | |
940 | ||
941 | id = xfs_qm_id_for_quotatype(ip, type); | |
92678554 CH |
942 | |
943 | restart: | |
cc2047c4 | 944 | dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id); |
9f920f11 | 945 | if (dqp) { |
9f920f11 CH |
946 | *O_dqpp = dqp; |
947 | return 0; | |
1da177e4 | 948 | } |
1da177e4 LT |
949 | |
950 | /* | |
951 | * Dquot cache miss. We don't want to keep the inode lock across | |
952 | * a (potential) disk read. Also we don't want to deal with the lock | |
953 | * ordering between quotainode and this inode. OTOH, dropping the inode | |
954 | * lock here means dealing with a chown that can happen before | |
955 | * we re-acquire the lock. | |
956 | */ | |
4882c19d | 957 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
30ab2dcf | 958 | error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp); |
4882c19d | 959 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
7ae44407 CH |
960 | if (error) |
961 | return error; | |
1da177e4 | 962 | |
4882c19d DW |
963 | /* |
964 | * A dquot could be attached to this inode by now, since we had | |
965 | * dropped the ilock. | |
966 | */ | |
967 | if (xfs_this_quota_on(mp, type)) { | |
968 | struct xfs_dquot *dqp1; | |
969 | ||
970 | dqp1 = xfs_inode_dquot(ip, type); | |
971 | if (dqp1) { | |
36731410 | 972 | xfs_qm_dqdestroy(dqp); |
4882c19d DW |
973 | dqp = dqp1; |
974 | xfs_dqlock(dqp); | |
975 | goto dqret; | |
1da177e4 | 976 | } |
4882c19d DW |
977 | } else { |
978 | /* inode stays locked on return */ | |
979 | xfs_qm_dqdestroy(dqp); | |
980 | return -ESRCH; | |
1da177e4 LT |
981 | } |
982 | ||
cc2047c4 DW |
983 | error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp); |
984 | if (error) { | |
1da177e4 | 985 | /* |
9f920f11 CH |
986 | * Duplicate found. Just throw away the new dquot and start |
987 | * over. | |
1da177e4 | 988 | */ |
9f920f11 | 989 | xfs_qm_dqdestroy(dqp); |
ff6d6af2 | 990 | XFS_STATS_INC(mp, xs_qm_dquot_dups); |
9f920f11 | 991 | goto restart; |
1da177e4 LT |
992 | } |
993 | ||
4882c19d DW |
994 | dqret: |
995 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
0b1b213f | 996 | trace_xfs_dqget_miss(dqp); |
1da177e4 | 997 | *O_dqpp = dqp; |
d99831ff | 998 | return 0; |
1da177e4 LT |
999 | } |
1000 | ||
2e330e76 DW |
1001 | /* |
1002 | * Starting at @id and progressing upwards, look for an initialized incore | |
1003 | * dquot, lock it, and return it. | |
1004 | */ | |
1005 | int | |
1006 | xfs_qm_dqget_next( | |
1007 | struct xfs_mount *mp, | |
1008 | xfs_dqid_t id, | |
1009 | uint type, | |
1010 | struct xfs_dquot **dqpp) | |
1011 | { | |
1012 | struct xfs_dquot *dqp; | |
1013 | int error = 0; | |
1014 | ||
1015 | *dqpp = NULL; | |
1016 | for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) { | |
30ab2dcf | 1017 | error = xfs_qm_dqget(mp, id, type, false, &dqp); |
2e330e76 DW |
1018 | if (error == -ENOENT) |
1019 | continue; | |
1020 | else if (error != 0) | |
1021 | break; | |
1022 | ||
1023 | if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) { | |
1024 | *dqpp = dqp; | |
1025 | return 0; | |
1026 | } | |
1027 | ||
1028 | xfs_qm_dqput(dqp); | |
1029 | } | |
1030 | ||
1031 | return error; | |
1032 | } | |
1033 | ||
f8739c3c CH |
1034 | /* |
1035 | * Release a reference to the dquot (decrement ref-count) and unlock it. | |
1036 | * | |
1037 | * If there is a group quota attached to this dquot, carefully release that | |
1038 | * too without tripping over deadlocks'n'stuff. | |
1039 | */ | |
1040 | void | |
1041 | xfs_qm_dqput( | |
1042 | struct xfs_dquot *dqp) | |
1043 | { | |
1044 | ASSERT(dqp->q_nrefs > 0); | |
1045 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
1046 | ||
1047 | trace_xfs_dqput(dqp); | |
1048 | ||
3c353375 DC |
1049 | if (--dqp->q_nrefs == 0) { |
1050 | struct xfs_quotainfo *qi = dqp->q_mount->m_quotainfo; | |
1051 | trace_xfs_dqput_free(dqp); | |
1052 | ||
1053 | if (list_lru_add(&qi->qi_lru, &dqp->q_lru)) | |
ff6d6af2 | 1054 | XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused); |
3c353375 DC |
1055 | } |
1056 | xfs_dqunlock(dqp); | |
1da177e4 LT |
1057 | } |
1058 | ||
1059 | /* | |
1060 | * Release a dquot. Flush it if dirty, then dqput() it. | |
1061 | * dquot must not be locked. | |
1062 | */ | |
1063 | void | |
1064 | xfs_qm_dqrele( | |
aefe69a4 | 1065 | struct xfs_dquot *dqp) |
1da177e4 | 1066 | { |
7d095257 CH |
1067 | if (!dqp) |
1068 | return; | |
1069 | ||
0b1b213f | 1070 | trace_xfs_dqrele(dqp); |
1da177e4 LT |
1071 | |
1072 | xfs_dqlock(dqp); | |
1073 | /* | |
1074 | * We don't care to flush it if the dquot is dirty here. | |
1075 | * That will create stutters that we want to avoid. | |
1076 | * Instead we do a delayed write when we try to reclaim | |
1077 | * a dirty dquot. Also xfs_sync will take part of the burden... | |
1078 | */ | |
1079 | xfs_qm_dqput(dqp); | |
1080 | } | |
1081 | ||
ca30b2a7 CH |
1082 | /* |
1083 | * This is the dquot flushing I/O completion routine. It is called | |
1084 | * from interrupt level when the buffer containing the dquot is | |
1085 | * flushed to disk. It is responsible for removing the dquot logitem | |
1086 | * from the AIL if it has not been re-logged, and unlocking the dquot's | |
1087 | * flush lock. This behavior is very similar to that of inodes.. | |
1088 | */ | |
6f5de180 | 1089 | static void |
ca30b2a7 | 1090 | xfs_qm_dqflush_done( |
ca30b2a7 CH |
1091 | struct xfs_log_item *lip) |
1092 | { | |
fd8b81db | 1093 | struct xfs_dq_logitem *qip = (struct xfs_dq_logitem *)lip; |
aefe69a4 | 1094 | struct xfs_dquot *dqp = qip->qli_dquot; |
ca30b2a7 | 1095 | struct xfs_ail *ailp = lip->li_ailp; |
849274c1 | 1096 | xfs_lsn_t tail_lsn; |
ca30b2a7 CH |
1097 | |
1098 | /* | |
1099 | * We only want to pull the item from the AIL if its | |
1100 | * location in the log has not changed since we started the flush. | |
1101 | * Thus, we only bother if the dquot's lsn has | |
1102 | * not changed. First we check the lsn outside the lock | |
1103 | * since it's cheaper, and then we recheck while | |
1104 | * holding the lock before removing the dquot from the AIL. | |
1105 | */ | |
22525c17 | 1106 | if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) && |
373b0589 | 1107 | ((lip->li_lsn == qip->qli_flush_lsn) || |
22525c17 | 1108 | test_bit(XFS_LI_FAILED, &lip->li_flags))) { |
ca30b2a7 | 1109 | |
57e80956 | 1110 | spin_lock(&ailp->ail_lock); |
e98084b8 | 1111 | xfs_clear_li_failed(lip); |
373b0589 | 1112 | if (lip->li_lsn == qip->qli_flush_lsn) { |
849274c1 BF |
1113 | /* xfs_ail_update_finish() drops the AIL lock */ |
1114 | tail_lsn = xfs_ail_delete_one(ailp, lip); | |
1115 | xfs_ail_update_finish(ailp, tail_lsn); | |
373b0589 | 1116 | } else { |
57e80956 | 1117 | spin_unlock(&ailp->ail_lock); |
373b0589 | 1118 | } |
ca30b2a7 CH |
1119 | } |
1120 | ||
1121 | /* | |
1122 | * Release the dq's flush lock since we're done with it. | |
1123 | */ | |
1124 | xfs_dqfunlock(dqp); | |
1125 | } | |
1da177e4 | 1126 | |
6f5de180 DC |
1127 | void |
1128 | xfs_dquot_done( | |
1129 | struct xfs_buf *bp) | |
1130 | { | |
1131 | struct xfs_log_item *lip, *n; | |
1132 | ||
1133 | list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) { | |
1134 | list_del_init(&lip->li_bio_list); | |
1135 | xfs_qm_dqflush_done(lip); | |
1136 | } | |
1137 | } | |
1138 | ||
0b0fa1d1 DW |
1139 | /* Check incore dquot for errors before we flush. */ |
1140 | static xfs_failaddr_t | |
1141 | xfs_qm_dqflush_check( | |
1142 | struct xfs_dquot *dqp) | |
1143 | { | |
d3537cf9 | 1144 | struct xfs_disk_dquot *ddq = &dqp->q_core; |
0b0fa1d1 DW |
1145 | __u8 type = dqp->dq_flags & XFS_DQ_ALLTYPES; |
1146 | ||
1147 | if (type != XFS_DQ_USER && | |
1148 | type != XFS_DQ_GROUP && | |
1149 | type != XFS_DQ_PROJ) | |
1150 | return __this_address; | |
1151 | ||
d3537cf9 DW |
1152 | if (dqp->q_id == 0) |
1153 | return NULL; | |
1154 | ||
be37d40c | 1155 | if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit && |
d3537cf9 DW |
1156 | !ddq->d_btimer) |
1157 | return __this_address; | |
1158 | ||
be37d40c | 1159 | if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit && |
d3537cf9 DW |
1160 | !ddq->d_itimer) |
1161 | return __this_address; | |
1162 | ||
be37d40c | 1163 | if (dqp->q_rtb.softlimit && dqp->q_rtb.count > dqp->q_rtb.softlimit && |
d3537cf9 DW |
1164 | !ddq->d_rtbtimer) |
1165 | return __this_address; | |
1166 | ||
0b0fa1d1 DW |
1167 | return NULL; |
1168 | } | |
1169 | ||
1da177e4 LT |
1170 | /* |
1171 | * Write a modified dquot to disk. | |
1172 | * The dquot must be locked and the flush lock too taken by caller. | |
1173 | * The flush lock will not be unlocked until the dquot reaches the disk, | |
1174 | * but the dquot is free to be unlocked and modified by the caller | |
1175 | * in the interim. Dquot is still locked on return. This behavior is | |
1176 | * identical to that of inodes. | |
1177 | */ | |
1178 | int | |
1179 | xfs_qm_dqflush( | |
fe7257fd CH |
1180 | struct xfs_dquot *dqp, |
1181 | struct xfs_buf **bpp) | |
1da177e4 | 1182 | { |
acecf1b5 | 1183 | struct xfs_mount *mp = dqp->q_mount; |
b707fffd | 1184 | struct xfs_log_item *lip = &dqp->q_logitem.qli_item; |
acecf1b5 | 1185 | struct xfs_buf *bp; |
7224fa48 | 1186 | struct xfs_dqblk *dqb; |
acecf1b5 | 1187 | struct xfs_disk_dquot *ddqp; |
eebf3cab | 1188 | xfs_failaddr_t fa; |
1da177e4 | 1189 | int error; |
1da177e4 LT |
1190 | |
1191 | ASSERT(XFS_DQ_IS_LOCKED(dqp)); | |
e1f49cf2 | 1192 | ASSERT(!completion_done(&dqp->q_flush)); |
acecf1b5 | 1193 | |
0b1b213f | 1194 | trace_xfs_dqflush(dqp); |
1da177e4 | 1195 | |
fe7257fd CH |
1196 | *bpp = NULL; |
1197 | ||
1da177e4 LT |
1198 | xfs_qm_dqunpin_wait(dqp); |
1199 | ||
1da177e4 LT |
1200 | /* |
1201 | * Get the buffer containing the on-disk dquot | |
1da177e4 | 1202 | */ |
acecf1b5 | 1203 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno, |
8d3d7e2b BF |
1204 | mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK, |
1205 | &bp, &xfs_dquot_buf_ops); | |
b707fffd | 1206 | if (error == -EAGAIN) |
fe7257fd | 1207 | goto out_unlock; |
b707fffd BF |
1208 | if (error) |
1209 | goto out_abort; | |
1da177e4 | 1210 | |
acecf1b5 CH |
1211 | /* |
1212 | * Calculate the location of the dquot inside the buffer. | |
1213 | */ | |
7224fa48 ES |
1214 | dqb = bp->b_addr + dqp->q_bufoffset; |
1215 | ddqp = &dqb->dd_diskdq; | |
acecf1b5 | 1216 | |
629dcb38 | 1217 | /* sanity check the in-core structure before we flush */ |
c51df733 | 1218 | fa = xfs_dquot_verify(mp, &dqp->q_core, dqp->q_id, 0); |
eebf3cab DW |
1219 | if (fa) { |
1220 | xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS", | |
c51df733 | 1221 | dqp->q_id, fa); |
acecf1b5 | 1222 | xfs_buf_relse(bp); |
b707fffd BF |
1223 | error = -EFSCORRUPTED; |
1224 | goto out_abort; | |
1da177e4 LT |
1225 | } |
1226 | ||
0b0fa1d1 DW |
1227 | fa = xfs_qm_dqflush_check(dqp); |
1228 | if (fa) { | |
1229 | xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS", | |
c51df733 | 1230 | dqp->q_id, fa); |
0b0fa1d1 DW |
1231 | xfs_buf_relse(bp); |
1232 | error = -EFSCORRUPTED; | |
1233 | goto out_abort; | |
1234 | } | |
1235 | ||
1236 | xfs_dquot_to_disk(ddqp, dqp); | |
1da177e4 LT |
1237 | |
1238 | /* | |
1239 | * Clear the dirty field and remember the flush lsn for later use. | |
1240 | */ | |
985a78fd | 1241 | dqp->q_flags &= ~XFS_DQFLAG_DIRTY; |
1da177e4 | 1242 | |
7b2e2a31 DC |
1243 | xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn, |
1244 | &dqp->q_logitem.qli_item.li_lsn); | |
1da177e4 | 1245 | |
3fe58f30 CH |
1246 | /* |
1247 | * copy the lsn into the on-disk dquot now while we have the in memory | |
1248 | * dquot here. This can't be done later in the write verifier as we | |
1249 | * can't get access to the log item at that point in time. | |
6fcdc59d DC |
1250 | * |
1251 | * We also calculate the CRC here so that the on-disk dquot in the | |
1252 | * buffer always has a valid CRC. This ensures there is no possibility | |
1253 | * of a dquot without an up-to-date CRC getting to disk. | |
3fe58f30 CH |
1254 | */ |
1255 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
3fe58f30 | 1256 | dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn); |
6fcdc59d DC |
1257 | xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk), |
1258 | XFS_DQUOT_CRC_OFF); | |
3fe58f30 CH |
1259 | } |
1260 | ||
1da177e4 | 1261 | /* |
2ef3f7f5 DC |
1262 | * Attach the dquot to the buffer so that we can remove this dquot from |
1263 | * the AIL and release the flush lock once the dquot is synced to disk. | |
1da177e4 | 1264 | */ |
0c7e5afb | 1265 | bp->b_flags |= _XBF_DQUOTS; |
2ef3f7f5 | 1266 | list_add_tail(&dqp->q_logitem.qli_item.li_bio_list, &bp->b_li_list); |
ca30b2a7 | 1267 | |
1da177e4 LT |
1268 | /* |
1269 | * If the buffer is pinned then push on the log so we won't | |
1270 | * get stuck waiting in the write for too long. | |
1271 | */ | |
811e64c7 | 1272 | if (xfs_buf_ispinned(bp)) { |
0b1b213f | 1273 | trace_xfs_dqflush_force(dqp); |
a14a348b | 1274 | xfs_log_force(mp, 0); |
1da177e4 LT |
1275 | } |
1276 | ||
0b1b213f | 1277 | trace_xfs_dqflush_done(dqp); |
fe7257fd CH |
1278 | *bpp = bp; |
1279 | return 0; | |
0b1b213f | 1280 | |
b707fffd | 1281 | out_abort: |
985a78fd | 1282 | dqp->q_flags &= ~XFS_DQFLAG_DIRTY; |
2b3cf093 | 1283 | xfs_trans_ail_delete(lip, 0); |
b707fffd | 1284 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
fe7257fd CH |
1285 | out_unlock: |
1286 | xfs_dqfunlock(dqp); | |
8d3d7e2b | 1287 | return error; |
1da177e4 LT |
1288 | } |
1289 | ||
5bb87a33 CH |
1290 | /* |
1291 | * Lock two xfs_dquot structures. | |
1292 | * | |
1293 | * To avoid deadlocks we always lock the quota structure with | |
1294 | * the lowerd id first. | |
1295 | */ | |
1da177e4 LT |
1296 | void |
1297 | xfs_dqlock2( | |
aefe69a4 PR |
1298 | struct xfs_dquot *d1, |
1299 | struct xfs_dquot *d2) | |
1da177e4 LT |
1300 | { |
1301 | if (d1 && d2) { | |
1302 | ASSERT(d1 != d2); | |
c51df733 | 1303 | if (d1->q_id > d2->q_id) { |
5bb87a33 CH |
1304 | mutex_lock(&d2->q_qlock); |
1305 | mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1306 | } else { |
5bb87a33 CH |
1307 | mutex_lock(&d1->q_qlock); |
1308 | mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED); | |
1da177e4 | 1309 | } |
5bb87a33 CH |
1310 | } else if (d1) { |
1311 | mutex_lock(&d1->q_qlock); | |
1312 | } else if (d2) { | |
1313 | mutex_lock(&d2->q_qlock); | |
1da177e4 LT |
1314 | } |
1315 | } | |
1316 | ||
a05931ce CH |
1317 | int __init |
1318 | xfs_qm_init(void) | |
1319 | { | |
b1231760 CM |
1320 | xfs_qm_dqzone = kmem_cache_create("xfs_dquot", |
1321 | sizeof(struct xfs_dquot), | |
1322 | 0, 0, NULL); | |
a05931ce CH |
1323 | if (!xfs_qm_dqzone) |
1324 | goto out; | |
1325 | ||
b1231760 CM |
1326 | xfs_qm_dqtrxzone = kmem_cache_create("xfs_dqtrx", |
1327 | sizeof(struct xfs_dquot_acct), | |
1328 | 0, 0, NULL); | |
a05931ce CH |
1329 | if (!xfs_qm_dqtrxzone) |
1330 | goto out_free_dqzone; | |
1331 | ||
1332 | return 0; | |
1333 | ||
1334 | out_free_dqzone: | |
aaf54eb8 | 1335 | kmem_cache_destroy(xfs_qm_dqzone); |
a05931ce CH |
1336 | out: |
1337 | return -ENOMEM; | |
1338 | } | |
1339 | ||
1c2ccc66 | 1340 | void |
a05931ce CH |
1341 | xfs_qm_exit(void) |
1342 | { | |
aaf54eb8 CM |
1343 | kmem_cache_destroy(xfs_qm_dqtrxzone); |
1344 | kmem_cache_destroy(xfs_qm_dqzone); | |
a05931ce | 1345 | } |
554ba965 DW |
1346 | |
1347 | /* | |
1348 | * Iterate every dquot of a particular type. The caller must ensure that the | |
1349 | * particular quota type is active. iter_fn can return negative error codes, | |
e7ee96df | 1350 | * or -ECANCELED to indicate that it wants to stop iterating. |
554ba965 DW |
1351 | */ |
1352 | int | |
1353 | xfs_qm_dqiterate( | |
1354 | struct xfs_mount *mp, | |
1355 | uint dqtype, | |
1356 | xfs_qm_dqiterate_fn iter_fn, | |
1357 | void *priv) | |
1358 | { | |
1359 | struct xfs_dquot *dq; | |
1360 | xfs_dqid_t id = 0; | |
1361 | int error; | |
1362 | ||
1363 | do { | |
1364 | error = xfs_qm_dqget_next(mp, id, dqtype, &dq); | |
1365 | if (error == -ENOENT) | |
1366 | return 0; | |
1367 | if (error) | |
1368 | return error; | |
1369 | ||
1370 | error = iter_fn(dq, dqtype, priv); | |
c51df733 | 1371 | id = dq->q_id; |
554ba965 | 1372 | xfs_qm_dqput(dq); |
554ba965 DW |
1373 | } while (error == 0 && id != 0); |
1374 | ||
1375 | return error; | |
1376 | } |