]>
Commit | Line | Data |
---|---|---|
60101509 BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
22 | * Copyright (C) 2008-2010 Lawrence Livermore National Security, LLC. | |
23 | * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER). | |
24 | * Rewritten for Linux by Brian Behlendorf <behlendorf1@llnl.gov>. | |
25 | * LLNL-CODE-403049. | |
26 | * | |
27 | * ZFS volume emulation driver. | |
28 | * | |
29 | * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. | |
30 | * Volumes are accessed through the symbolic links named: | |
31 | * | |
32 | * /dev/<pool_name>/<dataset_name> | |
33 | * | |
34 | * Volumes are persistent through reboot and module load. No user command | |
35 | * needs to be run before opening and using a device. | |
36 | */ | |
37 | ||
03c6040b | 38 | #include <sys/dbuf.h> |
60101509 BB |
39 | #include <sys/dmu_traverse.h> |
40 | #include <sys/dsl_dataset.h> | |
41 | #include <sys/dsl_prop.h> | |
42 | #include <sys/zap.h> | |
43 | #include <sys/zil_impl.h> | |
44 | #include <sys/zio.h> | |
45 | #include <sys/zfs_rlock.h> | |
46 | #include <sys/zfs_znode.h> | |
47 | #include <sys/zvol.h> | |
61e90960 | 48 | #include <linux/blkdev_compat.h> |
60101509 | 49 | |
74497b7a | 50 | unsigned int zvol_inhibit_dev = 0; |
60101509 | 51 | unsigned int zvol_major = ZVOL_MAJOR; |
dde9380a | 52 | unsigned int zvol_threads = 32; |
7c0e5708 | 53 | unsigned long zvol_max_discard_blocks = 16384; |
60101509 BB |
54 | |
55 | static taskq_t *zvol_taskq; | |
56 | static kmutex_t zvol_state_lock; | |
57 | static list_t zvol_state_list; | |
58 | static char *zvol_tag = "zvol_tag"; | |
59 | ||
60 | /* | |
61 | * The in-core state of each volume. | |
62 | */ | |
63 | typedef struct zvol_state { | |
4c0d8e50 | 64 | char zv_name[MAXNAMELEN]; /* name */ |
60101509 BB |
65 | uint64_t zv_volsize; /* advertised space */ |
66 | uint64_t zv_volblocksize;/* volume block size */ | |
67 | objset_t *zv_objset; /* objset handle */ | |
68 | uint32_t zv_flags; /* ZVOL_* flags */ | |
69 | uint32_t zv_open_count; /* open counts */ | |
70 | uint32_t zv_changed; /* disk changed */ | |
71 | zilog_t *zv_zilog; /* ZIL handle */ | |
72 | znode_t zv_znode; /* for range locking */ | |
73 | dmu_buf_t *zv_dbuf; /* bonus handle */ | |
74 | dev_t zv_dev; /* device id */ | |
75 | struct gendisk *zv_disk; /* generic disk */ | |
76 | struct request_queue *zv_queue; /* request queue */ | |
77 | spinlock_t zv_lock; /* request queue lock */ | |
78 | list_node_t zv_next; /* next zvol_state_t linkage */ | |
79 | } zvol_state_t; | |
80 | ||
81 | #define ZVOL_RDONLY 0x1 | |
82 | ||
83 | /* | |
84 | * Find the next available range of ZVOL_MINORS minor numbers. The | |
85 | * zvol_state_list is kept in ascending minor order so we simply need | |
86 | * to scan the list for the first gap in the sequence. This allows us | |
87 | * to recycle minor number as devices are created and removed. | |
88 | */ | |
89 | static int | |
90 | zvol_find_minor(unsigned *minor) | |
91 | { | |
92 | zvol_state_t *zv; | |
93 | ||
94 | *minor = 0; | |
95 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
96 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
97 | zv = list_next(&zvol_state_list, zv), *minor += ZVOL_MINORS) { | |
98 | if (MINOR(zv->zv_dev) != MINOR(*minor)) | |
99 | break; | |
100 | } | |
101 | ||
102 | /* All minors are in use */ | |
103 | if (*minor >= (1 << MINORBITS)) | |
104 | return ENXIO; | |
105 | ||
106 | return 0; | |
107 | } | |
108 | ||
109 | /* | |
110 | * Find a zvol_state_t given the full major+minor dev_t. | |
111 | */ | |
112 | static zvol_state_t * | |
113 | zvol_find_by_dev(dev_t dev) | |
114 | { | |
115 | zvol_state_t *zv; | |
116 | ||
117 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
118 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
119 | zv = list_next(&zvol_state_list, zv)) { | |
120 | if (zv->zv_dev == dev) | |
121 | return zv; | |
122 | } | |
123 | ||
124 | return NULL; | |
125 | } | |
126 | ||
127 | /* | |
128 | * Find a zvol_state_t given the name provided at zvol_alloc() time. | |
129 | */ | |
130 | static zvol_state_t * | |
131 | zvol_find_by_name(const char *name) | |
132 | { | |
133 | zvol_state_t *zv; | |
134 | ||
135 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
136 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
137 | zv = list_next(&zvol_state_list, zv)) { | |
4c0d8e50 | 138 | if (!strncmp(zv->zv_name, name, MAXNAMELEN)) |
60101509 BB |
139 | return zv; |
140 | } | |
141 | ||
142 | return NULL; | |
143 | } | |
144 | ||
6c285672 JL |
145 | |
146 | /* | |
147 | * Given a path, return TRUE if path is a ZVOL. | |
148 | */ | |
149 | boolean_t | |
150 | zvol_is_zvol(const char *device) | |
151 | { | |
152 | struct block_device *bdev; | |
153 | unsigned int major; | |
154 | ||
155 | bdev = lookup_bdev(device); | |
156 | if (IS_ERR(bdev)) | |
157 | return (B_FALSE); | |
158 | ||
159 | major = MAJOR(bdev->bd_dev); | |
160 | bdput(bdev); | |
161 | ||
162 | if (major == zvol_major) | |
163 | return (B_TRUE); | |
164 | ||
165 | return (B_FALSE); | |
166 | } | |
167 | ||
60101509 BB |
168 | /* |
169 | * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation. | |
170 | */ | |
171 | void | |
172 | zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) | |
173 | { | |
174 | zfs_creat_t *zct = arg; | |
175 | nvlist_t *nvprops = zct->zct_props; | |
176 | int error; | |
177 | uint64_t volblocksize, volsize; | |
178 | ||
179 | VERIFY(nvlist_lookup_uint64(nvprops, | |
180 | zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); | |
181 | if (nvlist_lookup_uint64(nvprops, | |
182 | zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) | |
183 | volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); | |
184 | ||
185 | /* | |
186 | * These properties must be removed from the list so the generic | |
187 | * property setting step won't apply to them. | |
188 | */ | |
189 | VERIFY(nvlist_remove_all(nvprops, | |
190 | zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); | |
191 | (void) nvlist_remove_all(nvprops, | |
192 | zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); | |
193 | ||
194 | error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, | |
195 | DMU_OT_NONE, 0, tx); | |
196 | ASSERT(error == 0); | |
197 | ||
198 | error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, | |
199 | DMU_OT_NONE, 0, tx); | |
200 | ASSERT(error == 0); | |
201 | ||
202 | error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); | |
203 | ASSERT(error == 0); | |
204 | } | |
205 | ||
206 | /* | |
207 | * ZFS_IOC_OBJSET_STATS entry point. | |
208 | */ | |
209 | int | |
210 | zvol_get_stats(objset_t *os, nvlist_t *nv) | |
211 | { | |
212 | int error; | |
213 | dmu_object_info_t *doi; | |
214 | uint64_t val; | |
215 | ||
216 | error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); | |
217 | if (error) | |
218 | return (error); | |
219 | ||
220 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); | |
221 | doi = kmem_alloc(sizeof(dmu_object_info_t), KM_SLEEP); | |
222 | error = dmu_object_info(os, ZVOL_OBJ, doi); | |
223 | ||
224 | if (error == 0) { | |
225 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, | |
226 | doi->doi_data_block_size); | |
227 | } | |
228 | ||
229 | kmem_free(doi, sizeof(dmu_object_info_t)); | |
230 | ||
231 | return (error); | |
232 | } | |
233 | ||
234 | /* | |
235 | * Sanity check volume size. | |
236 | */ | |
237 | int | |
238 | zvol_check_volsize(uint64_t volsize, uint64_t blocksize) | |
239 | { | |
240 | if (volsize == 0) | |
2e528b49 | 241 | return (SET_ERROR(EINVAL)); |
60101509 BB |
242 | |
243 | if (volsize % blocksize != 0) | |
2e528b49 | 244 | return (SET_ERROR(EINVAL)); |
60101509 BB |
245 | |
246 | #ifdef _ILP32 | |
247 | if (volsize - 1 > MAXOFFSET_T) | |
2e528b49 | 248 | return (SET_ERROR(EOVERFLOW)); |
60101509 BB |
249 | #endif |
250 | return (0); | |
251 | } | |
252 | ||
253 | /* | |
254 | * Ensure the zap is flushed then inform the VFS of the capacity change. | |
255 | */ | |
256 | static int | |
df554c14 | 257 | zvol_update_volsize(zvol_state_t *zv, uint64_t volsize, objset_t *os) |
60101509 BB |
258 | { |
259 | struct block_device *bdev; | |
260 | dmu_tx_t *tx; | |
261 | int error; | |
262 | ||
263 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
264 | ||
df554c14 | 265 | tx = dmu_tx_create(os); |
60101509 BB |
266 | dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); |
267 | error = dmu_tx_assign(tx, TXG_WAIT); | |
268 | if (error) { | |
269 | dmu_tx_abort(tx); | |
270 | return (error); | |
271 | } | |
272 | ||
df554c14 | 273 | error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, |
60101509 BB |
274 | &volsize, tx); |
275 | dmu_tx_commit(tx); | |
276 | ||
277 | if (error) | |
278 | return (error); | |
279 | ||
df554c14 | 280 | error = dmu_free_long_range(os, |
60101509 BB |
281 | ZVOL_OBJ, volsize, DMU_OBJECT_END); |
282 | if (error) | |
283 | return (error); | |
284 | ||
60101509 BB |
285 | bdev = bdget_disk(zv->zv_disk, 0); |
286 | if (!bdev) | |
2e528b49 | 287 | return (SET_ERROR(EIO)); |
df554c14 BB |
288 | /* |
289 | * 2.6.28 API change | |
290 | * Added check_disk_size_change() helper function. | |
291 | */ | |
292 | #ifdef HAVE_CHECK_DISK_SIZE_CHANGE | |
293 | set_capacity(zv->zv_disk, volsize >> 9); | |
294 | zv->zv_volsize = volsize; | |
295 | check_disk_size_change(zv->zv_disk, bdev); | |
296 | #else | |
297 | zv->zv_volsize = volsize; | |
298 | zv->zv_changed = 1; | |
299 | (void) check_disk_change(bdev); | |
300 | #endif /* HAVE_CHECK_DISK_SIZE_CHANGE */ | |
60101509 | 301 | |
60101509 BB |
302 | bdput(bdev); |
303 | ||
304 | return (0); | |
305 | } | |
306 | ||
307 | /* | |
308 | * Set ZFS_PROP_VOLSIZE set entry point. | |
309 | */ | |
310 | int | |
311 | zvol_set_volsize(const char *name, uint64_t volsize) | |
312 | { | |
313 | zvol_state_t *zv; | |
314 | dmu_object_info_t *doi; | |
315 | objset_t *os = NULL; | |
316 | uint64_t readonly; | |
317 | int error; | |
318 | ||
13fe0198 MA |
319 | error = dsl_prop_get_integer(name, |
320 | zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL); | |
321 | if (error != 0) | |
322 | return (error); | |
323 | if (readonly) | |
2e528b49 | 324 | return (SET_ERROR(EROFS)); |
13fe0198 | 325 | |
60101509 BB |
326 | mutex_enter(&zvol_state_lock); |
327 | ||
328 | zv = zvol_find_by_name(name); | |
329 | if (zv == NULL) { | |
2e528b49 | 330 | error = SET_ERROR(ENXIO); |
60101509 BB |
331 | goto out; |
332 | } | |
333 | ||
334 | doi = kmem_alloc(sizeof(dmu_object_info_t), KM_SLEEP); | |
335 | ||
336 | error = dmu_objset_hold(name, FTAG, &os); | |
337 | if (error) | |
338 | goto out_doi; | |
339 | ||
340 | if ((error = dmu_object_info(os, ZVOL_OBJ, doi)) != 0 || | |
341 | (error = zvol_check_volsize(volsize,doi->doi_data_block_size)) != 0) | |
342 | goto out_doi; | |
343 | ||
344 | VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, NULL) == 0); | |
345 | if (readonly) { | |
2e528b49 | 346 | error = SET_ERROR(EROFS); |
60101509 BB |
347 | goto out_doi; |
348 | } | |
349 | ||
ba6a2402 | 350 | if (zv->zv_flags & ZVOL_RDONLY) { |
2e528b49 | 351 | error = SET_ERROR(EROFS); |
60101509 BB |
352 | goto out_doi; |
353 | } | |
354 | ||
df554c14 | 355 | error = zvol_update_volsize(zv, volsize, os); |
60101509 BB |
356 | out_doi: |
357 | kmem_free(doi, sizeof(dmu_object_info_t)); | |
358 | out: | |
359 | if (os) | |
360 | dmu_objset_rele(os, FTAG); | |
361 | ||
362 | mutex_exit(&zvol_state_lock); | |
363 | ||
364 | return (error); | |
365 | } | |
366 | ||
367 | /* | |
368 | * Sanity check volume block size. | |
369 | */ | |
370 | int | |
371 | zvol_check_volblocksize(uint64_t volblocksize) | |
372 | { | |
373 | if (volblocksize < SPA_MINBLOCKSIZE || | |
374 | volblocksize > SPA_MAXBLOCKSIZE || | |
375 | !ISP2(volblocksize)) | |
2e528b49 | 376 | return (SET_ERROR(EDOM)); |
60101509 BB |
377 | |
378 | return (0); | |
379 | } | |
380 | ||
381 | /* | |
382 | * Set ZFS_PROP_VOLBLOCKSIZE set entry point. | |
383 | */ | |
384 | int | |
385 | zvol_set_volblocksize(const char *name, uint64_t volblocksize) | |
386 | { | |
387 | zvol_state_t *zv; | |
388 | dmu_tx_t *tx; | |
389 | int error; | |
390 | ||
391 | mutex_enter(&zvol_state_lock); | |
392 | ||
393 | zv = zvol_find_by_name(name); | |
394 | if (zv == NULL) { | |
2e528b49 | 395 | error = SET_ERROR(ENXIO); |
60101509 BB |
396 | goto out; |
397 | } | |
398 | ||
ba6a2402 | 399 | if (zv->zv_flags & ZVOL_RDONLY) { |
2e528b49 | 400 | error = SET_ERROR(EROFS); |
60101509 BB |
401 | goto out; |
402 | } | |
403 | ||
404 | tx = dmu_tx_create(zv->zv_objset); | |
405 | dmu_tx_hold_bonus(tx, ZVOL_OBJ); | |
406 | error = dmu_tx_assign(tx, TXG_WAIT); | |
407 | if (error) { | |
408 | dmu_tx_abort(tx); | |
409 | } else { | |
410 | error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ, | |
411 | volblocksize, 0, tx); | |
412 | if (error == ENOTSUP) | |
2e528b49 | 413 | error = SET_ERROR(EBUSY); |
60101509 BB |
414 | dmu_tx_commit(tx); |
415 | if (error == 0) | |
416 | zv->zv_volblocksize = volblocksize; | |
417 | } | |
418 | out: | |
419 | mutex_exit(&zvol_state_lock); | |
420 | ||
421 | return (error); | |
422 | } | |
423 | ||
424 | /* | |
425 | * Replay a TX_WRITE ZIL transaction that didn't get committed | |
426 | * after a system failure | |
427 | */ | |
428 | static int | |
429 | zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) | |
430 | { | |
431 | objset_t *os = zv->zv_objset; | |
432 | char *data = (char *)(lr + 1); /* data follows lr_write_t */ | |
433 | uint64_t off = lr->lr_offset; | |
434 | uint64_t len = lr->lr_length; | |
435 | dmu_tx_t *tx; | |
436 | int error; | |
437 | ||
438 | if (byteswap) | |
439 | byteswap_uint64_array(lr, sizeof (*lr)); | |
440 | ||
441 | tx = dmu_tx_create(os); | |
442 | dmu_tx_hold_write(tx, ZVOL_OBJ, off, len); | |
443 | error = dmu_tx_assign(tx, TXG_WAIT); | |
444 | if (error) { | |
445 | dmu_tx_abort(tx); | |
446 | } else { | |
447 | dmu_write(os, ZVOL_OBJ, off, len, data, tx); | |
448 | dmu_tx_commit(tx); | |
449 | } | |
450 | ||
451 | return (error); | |
452 | } | |
453 | ||
454 | static int | |
455 | zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) | |
456 | { | |
2e528b49 | 457 | return (SET_ERROR(ENOTSUP)); |
60101509 BB |
458 | } |
459 | ||
460 | /* | |
461 | * Callback vectors for replaying records. | |
462 | * Only TX_WRITE is needed for zvol. | |
463 | */ | |
b01615d5 RY |
464 | zil_replay_func_t zvol_replay_vector[TX_MAX_TYPE] = { |
465 | (zil_replay_func_t)zvol_replay_err, /* no such transaction type */ | |
466 | (zil_replay_func_t)zvol_replay_err, /* TX_CREATE */ | |
467 | (zil_replay_func_t)zvol_replay_err, /* TX_MKDIR */ | |
468 | (zil_replay_func_t)zvol_replay_err, /* TX_MKXATTR */ | |
469 | (zil_replay_func_t)zvol_replay_err, /* TX_SYMLINK */ | |
470 | (zil_replay_func_t)zvol_replay_err, /* TX_REMOVE */ | |
471 | (zil_replay_func_t)zvol_replay_err, /* TX_RMDIR */ | |
472 | (zil_replay_func_t)zvol_replay_err, /* TX_LINK */ | |
473 | (zil_replay_func_t)zvol_replay_err, /* TX_RENAME */ | |
474 | (zil_replay_func_t)zvol_replay_write, /* TX_WRITE */ | |
475 | (zil_replay_func_t)zvol_replay_err, /* TX_TRUNCATE */ | |
476 | (zil_replay_func_t)zvol_replay_err, /* TX_SETATTR */ | |
477 | (zil_replay_func_t)zvol_replay_err, /* TX_ACL */ | |
60101509 BB |
478 | }; |
479 | ||
480 | /* | |
481 | * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. | |
482 | * | |
483 | * We store data in the log buffers if it's small enough. | |
484 | * Otherwise we will later flush the data out via dmu_sync(). | |
485 | */ | |
486 | ssize_t zvol_immediate_write_sz = 32768; | |
487 | ||
488 | static void | |
489 | zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, | |
490 | uint64_t offset, uint64_t size, int sync) | |
491 | { | |
492 | uint32_t blocksize = zv->zv_volblocksize; | |
493 | zilog_t *zilog = zv->zv_zilog; | |
494 | boolean_t slogging; | |
ab85f845 | 495 | ssize_t immediate_write_sz; |
60101509 BB |
496 | |
497 | if (zil_replaying(zilog, tx)) | |
498 | return; | |
499 | ||
ab85f845 ED |
500 | immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) |
501 | ? 0 : zvol_immediate_write_sz; | |
502 | slogging = spa_has_slogs(zilog->zl_spa) && | |
503 | (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); | |
60101509 BB |
504 | |
505 | while (size) { | |
506 | itx_t *itx; | |
507 | lr_write_t *lr; | |
508 | ssize_t len; | |
509 | itx_wr_state_t write_state; | |
510 | ||
511 | /* | |
512 | * Unlike zfs_log_write() we can be called with | |
513 | * up to DMU_MAX_ACCESS/2 (5MB) writes. | |
514 | */ | |
ab85f845 | 515 | if (blocksize > immediate_write_sz && !slogging && |
60101509 BB |
516 | size >= blocksize && offset % blocksize == 0) { |
517 | write_state = WR_INDIRECT; /* uses dmu_sync */ | |
518 | len = blocksize; | |
519 | } else if (sync) { | |
520 | write_state = WR_COPIED; | |
521 | len = MIN(ZIL_MAX_LOG_DATA, size); | |
522 | } else { | |
523 | write_state = WR_NEED_COPY; | |
524 | len = MIN(ZIL_MAX_LOG_DATA, size); | |
525 | } | |
526 | ||
527 | itx = zil_itx_create(TX_WRITE, sizeof (*lr) + | |
528 | (write_state == WR_COPIED ? len : 0)); | |
529 | lr = (lr_write_t *)&itx->itx_lr; | |
530 | if (write_state == WR_COPIED && dmu_read(zv->zv_objset, | |
531 | ZVOL_OBJ, offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) { | |
532 | zil_itx_destroy(itx); | |
533 | itx = zil_itx_create(TX_WRITE, sizeof (*lr)); | |
534 | lr = (lr_write_t *)&itx->itx_lr; | |
535 | write_state = WR_NEED_COPY; | |
536 | } | |
537 | ||
538 | itx->itx_wr_state = write_state; | |
539 | if (write_state == WR_NEED_COPY) | |
540 | itx->itx_sod += len; | |
541 | lr->lr_foid = ZVOL_OBJ; | |
542 | lr->lr_offset = offset; | |
543 | lr->lr_length = len; | |
544 | lr->lr_blkoff = 0; | |
545 | BP_ZERO(&lr->lr_blkptr); | |
546 | ||
547 | itx->itx_private = zv; | |
548 | itx->itx_sync = sync; | |
549 | ||
550 | (void) zil_itx_assign(zilog, itx, tx); | |
551 | ||
552 | offset += len; | |
553 | size -= len; | |
554 | } | |
555 | } | |
556 | ||
557 | /* | |
558 | * Common write path running under the zvol taskq context. This function | |
559 | * is responsible for copying the request structure data in to the DMU and | |
560 | * signaling the request queue with the result of the copy. | |
561 | */ | |
562 | static void | |
563 | zvol_write(void *arg) | |
564 | { | |
565 | struct request *req = (struct request *)arg; | |
566 | struct request_queue *q = req->q; | |
567 | zvol_state_t *zv = q->queuedata; | |
568 | uint64_t offset = blk_rq_pos(req) << 9; | |
569 | uint64_t size = blk_rq_bytes(req); | |
570 | int error = 0; | |
571 | dmu_tx_t *tx; | |
572 | rl_t *rl; | |
573 | ||
8630650a BB |
574 | /* |
575 | * Annotate this call path with a flag that indicates that it is | |
576 | * unsafe to use KM_SLEEP during memory allocations due to the | |
577 | * potential for a deadlock. KM_PUSHPAGE should be used instead. | |
578 | */ | |
579 | ASSERT(!(current->flags & PF_NOFS)); | |
580 | current->flags |= PF_NOFS; | |
581 | ||
b18019d2 ED |
582 | if (req->cmd_flags & VDEV_REQ_FLUSH) |
583 | zil_commit(zv->zv_zilog, ZVOL_OBJ); | |
584 | ||
585 | /* | |
586 | * Some requests are just for flush and nothing else. | |
587 | */ | |
588 | if (size == 0) { | |
589 | blk_end_request(req, 0, size); | |
8630650a | 590 | goto out; |
b18019d2 ED |
591 | } |
592 | ||
60101509 BB |
593 | rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_WRITER); |
594 | ||
595 | tx = dmu_tx_create(zv->zv_objset); | |
596 | dmu_tx_hold_write(tx, ZVOL_OBJ, offset, size); | |
597 | ||
598 | /* This will only fail for ENOSPC */ | |
599 | error = dmu_tx_assign(tx, TXG_WAIT); | |
600 | if (error) { | |
601 | dmu_tx_abort(tx); | |
602 | zfs_range_unlock(rl); | |
603 | blk_end_request(req, -error, size); | |
8630650a | 604 | goto out; |
60101509 BB |
605 | } |
606 | ||
607 | error = dmu_write_req(zv->zv_objset, ZVOL_OBJ, req, tx); | |
608 | if (error == 0) | |
b18019d2 ED |
609 | zvol_log_write(zv, tx, offset, size, |
610 | req->cmd_flags & VDEV_REQ_FUA); | |
60101509 BB |
611 | |
612 | dmu_tx_commit(tx); | |
613 | zfs_range_unlock(rl); | |
614 | ||
b18019d2 ED |
615 | if ((req->cmd_flags & VDEV_REQ_FUA) || |
616 | zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) | |
60101509 BB |
617 | zil_commit(zv->zv_zilog, ZVOL_OBJ); |
618 | ||
619 | blk_end_request(req, -error, size); | |
8630650a BB |
620 | out: |
621 | current->flags &= ~PF_NOFS; | |
60101509 BB |
622 | } |
623 | ||
30930fba ED |
624 | #ifdef HAVE_BLK_QUEUE_DISCARD |
625 | static void | |
626 | zvol_discard(void *arg) | |
627 | { | |
628 | struct request *req = (struct request *)arg; | |
629 | struct request_queue *q = req->q; | |
630 | zvol_state_t *zv = q->queuedata; | |
089fa91b ED |
631 | uint64_t start = blk_rq_pos(req) << 9; |
632 | uint64_t end = start + blk_rq_bytes(req); | |
30930fba ED |
633 | int error; |
634 | rl_t *rl; | |
635 | ||
8630650a BB |
636 | /* |
637 | * Annotate this call path with a flag that indicates that it is | |
638 | * unsafe to use KM_SLEEP during memory allocations due to the | |
639 | * potential for a deadlock. KM_PUSHPAGE should be used instead. | |
640 | */ | |
641 | ASSERT(!(current->flags & PF_NOFS)); | |
642 | current->flags |= PF_NOFS; | |
643 | ||
089fa91b ED |
644 | if (end > zv->zv_volsize) { |
645 | blk_end_request(req, -EIO, blk_rq_bytes(req)); | |
8630650a | 646 | goto out; |
30930fba ED |
647 | } |
648 | ||
089fa91b ED |
649 | /* |
650 | * Align the request to volume block boundaries. If we don't, | |
651 | * then this will force dnode_free_range() to zero out the | |
652 | * unaligned parts, which is slow (read-modify-write) and | |
653 | * useless since we are not freeing any space by doing so. | |
654 | */ | |
655 | start = P2ROUNDUP(start, zv->zv_volblocksize); | |
656 | end = P2ALIGN(end, zv->zv_volblocksize); | |
657 | ||
658 | if (start >= end) { | |
659 | blk_end_request(req, 0, blk_rq_bytes(req)); | |
8630650a | 660 | goto out; |
30930fba ED |
661 | } |
662 | ||
089fa91b | 663 | rl = zfs_range_lock(&zv->zv_znode, start, end - start, RL_WRITER); |
30930fba | 664 | |
089fa91b | 665 | error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, start, end - start); |
30930fba ED |
666 | |
667 | /* | |
668 | * TODO: maybe we should add the operation to the log. | |
669 | */ | |
670 | ||
671 | zfs_range_unlock(rl); | |
672 | ||
089fa91b | 673 | blk_end_request(req, -error, blk_rq_bytes(req)); |
8630650a BB |
674 | out: |
675 | current->flags &= ~PF_NOFS; | |
30930fba ED |
676 | } |
677 | #endif /* HAVE_BLK_QUEUE_DISCARD */ | |
678 | ||
60101509 BB |
679 | /* |
680 | * Common read path running under the zvol taskq context. This function | |
681 | * is responsible for copying the requested data out of the DMU and in to | |
682 | * a linux request structure. It then must signal the request queue with | |
683 | * an error code describing the result of the copy. | |
684 | */ | |
685 | static void | |
686 | zvol_read(void *arg) | |
687 | { | |
688 | struct request *req = (struct request *)arg; | |
689 | struct request_queue *q = req->q; | |
690 | zvol_state_t *zv = q->queuedata; | |
691 | uint64_t offset = blk_rq_pos(req) << 9; | |
692 | uint64_t size = blk_rq_bytes(req); | |
693 | int error; | |
694 | rl_t *rl; | |
695 | ||
b18019d2 ED |
696 | if (size == 0) { |
697 | blk_end_request(req, 0, size); | |
698 | return; | |
699 | } | |
700 | ||
60101509 BB |
701 | rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); |
702 | ||
703 | error = dmu_read_req(zv->zv_objset, ZVOL_OBJ, req); | |
704 | ||
705 | zfs_range_unlock(rl); | |
706 | ||
707 | /* convert checksum errors into IO errors */ | |
708 | if (error == ECKSUM) | |
2e528b49 | 709 | error = SET_ERROR(EIO); |
60101509 BB |
710 | |
711 | blk_end_request(req, -error, size); | |
712 | } | |
713 | ||
714 | /* | |
715 | * Request will be added back to the request queue and retried if | |
716 | * it cannot be immediately dispatched to the taskq for handling | |
717 | */ | |
718 | static inline void | |
719 | zvol_dispatch(task_func_t func, struct request *req) | |
720 | { | |
721 | if (!taskq_dispatch(zvol_taskq, func, (void *)req, TQ_NOSLEEP)) | |
722 | blk_requeue_request(req->q, req); | |
723 | } | |
724 | ||
725 | /* | |
726 | * Common request path. Rather than registering a custom make_request() | |
727 | * function we use the generic Linux version. This is done because it allows | |
728 | * us to easily merge read requests which would otherwise we performed | |
729 | * synchronously by the DMU. This is less critical in write case where the | |
730 | * DMU will perform the correct merging within a transaction group. Using | |
731 | * the generic make_request() also let's use leverage the fact that the | |
732 | * elevator with ensure correct ordering in regards to barrior IOs. On | |
733 | * the downside it means that in the write case we end up doing request | |
734 | * merging twice once in the elevator and once in the DMU. | |
735 | * | |
736 | * The request handler is called under a spin lock so all the real work | |
737 | * is handed off to be done in the context of the zvol taskq. This function | |
738 | * simply performs basic request sanity checking and hands off the request. | |
739 | */ | |
740 | static void | |
741 | zvol_request(struct request_queue *q) | |
742 | { | |
743 | zvol_state_t *zv = q->queuedata; | |
744 | struct request *req; | |
745 | unsigned int size; | |
746 | ||
747 | while ((req = blk_fetch_request(q)) != NULL) { | |
748 | size = blk_rq_bytes(req); | |
749 | ||
b18019d2 | 750 | if (size != 0 && blk_rq_pos(req) + blk_rq_sectors(req) > |
60101509 BB |
751 | get_capacity(zv->zv_disk)) { |
752 | printk(KERN_INFO | |
753 | "%s: bad access: block=%llu, count=%lu\n", | |
754 | req->rq_disk->disk_name, | |
755 | (long long unsigned)blk_rq_pos(req), | |
756 | (long unsigned)blk_rq_sectors(req)); | |
757 | __blk_end_request(req, -EIO, size); | |
758 | continue; | |
759 | } | |
760 | ||
761 | if (!blk_fs_request(req)) { | |
762 | printk(KERN_INFO "%s: non-fs cmd\n", | |
763 | req->rq_disk->disk_name); | |
764 | __blk_end_request(req, -EIO, size); | |
765 | continue; | |
766 | } | |
767 | ||
768 | switch (rq_data_dir(req)) { | |
769 | case READ: | |
770 | zvol_dispatch(zvol_read, req); | |
771 | break; | |
772 | case WRITE: | |
ba6a2402 | 773 | if (unlikely(zv->zv_flags & ZVOL_RDONLY)) { |
60101509 BB |
774 | __blk_end_request(req, -EROFS, size); |
775 | break; | |
776 | } | |
777 | ||
30930fba ED |
778 | #ifdef HAVE_BLK_QUEUE_DISCARD |
779 | if (req->cmd_flags & VDEV_REQ_DISCARD) { | |
780 | zvol_dispatch(zvol_discard, req); | |
781 | break; | |
782 | } | |
783 | #endif /* HAVE_BLK_QUEUE_DISCARD */ | |
784 | ||
60101509 BB |
785 | zvol_dispatch(zvol_write, req); |
786 | break; | |
787 | default: | |
788 | printk(KERN_INFO "%s: unknown cmd: %d\n", | |
789 | req->rq_disk->disk_name, (int)rq_data_dir(req)); | |
790 | __blk_end_request(req, -EIO, size); | |
791 | break; | |
792 | } | |
793 | } | |
794 | } | |
795 | ||
796 | static void | |
797 | zvol_get_done(zgd_t *zgd, int error) | |
798 | { | |
799 | if (zgd->zgd_db) | |
800 | dmu_buf_rele(zgd->zgd_db, zgd); | |
801 | ||
802 | zfs_range_unlock(zgd->zgd_rl); | |
803 | ||
804 | if (error == 0 && zgd->zgd_bp) | |
805 | zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); | |
806 | ||
807 | kmem_free(zgd, sizeof (zgd_t)); | |
808 | } | |
809 | ||
810 | /* | |
811 | * Get data to generate a TX_WRITE intent log record. | |
812 | */ | |
813 | static int | |
814 | zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) | |
815 | { | |
816 | zvol_state_t *zv = arg; | |
817 | objset_t *os = zv->zv_objset; | |
03c6040b | 818 | uint64_t object = ZVOL_OBJ; |
60101509 BB |
819 | uint64_t offset = lr->lr_offset; |
820 | uint64_t size = lr->lr_length; | |
03c6040b | 821 | blkptr_t *bp = &lr->lr_blkptr; |
60101509 BB |
822 | dmu_buf_t *db; |
823 | zgd_t *zgd; | |
824 | int error; | |
825 | ||
826 | ASSERT(zio != NULL); | |
827 | ASSERT(size != 0); | |
828 | ||
b8d06fca | 829 | zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_PUSHPAGE); |
60101509 BB |
830 | zgd->zgd_zilog = zv->zv_zilog; |
831 | zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); | |
832 | ||
833 | /* | |
834 | * Write records come in two flavors: immediate and indirect. | |
835 | * For small writes it's cheaper to store the data with the | |
836 | * log record (immediate); for large writes it's cheaper to | |
837 | * sync the data and get a pointer to it (indirect) so that | |
838 | * we don't have to write the data twice. | |
839 | */ | |
840 | if (buf != NULL) { /* immediate write */ | |
03c6040b | 841 | error = dmu_read(os, object, offset, size, buf, |
60101509 BB |
842 | DMU_READ_NO_PREFETCH); |
843 | } else { | |
844 | size = zv->zv_volblocksize; | |
845 | offset = P2ALIGN_TYPED(offset, size, uint64_t); | |
03c6040b | 846 | error = dmu_buf_hold(os, object, offset, zgd, &db, |
60101509 BB |
847 | DMU_READ_NO_PREFETCH); |
848 | if (error == 0) { | |
03c6040b GW |
849 | blkptr_t *obp = dmu_buf_get_blkptr(db); |
850 | if (obp) { | |
851 | ASSERT(BP_IS_HOLE(bp)); | |
852 | *bp = *obp; | |
853 | } | |
854 | ||
60101509 BB |
855 | zgd->zgd_db = db; |
856 | zgd->zgd_bp = &lr->lr_blkptr; | |
857 | ||
858 | ASSERT(db != NULL); | |
859 | ASSERT(db->db_offset == offset); | |
860 | ASSERT(db->db_size == size); | |
861 | ||
862 | error = dmu_sync(zio, lr->lr_common.lrc_txg, | |
863 | zvol_get_done, zgd); | |
864 | ||
865 | if (error == 0) | |
866 | return (0); | |
867 | } | |
868 | } | |
869 | ||
870 | zvol_get_done(zgd, error); | |
871 | ||
872 | return (error); | |
873 | } | |
874 | ||
875 | /* | |
876 | * The zvol_state_t's are inserted in increasing MINOR(dev_t) order. | |
877 | */ | |
878 | static void | |
879 | zvol_insert(zvol_state_t *zv_insert) | |
880 | { | |
881 | zvol_state_t *zv = NULL; | |
882 | ||
883 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
884 | ASSERT3U(MINOR(zv_insert->zv_dev) & ZVOL_MINOR_MASK, ==, 0); | |
885 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
886 | zv = list_next(&zvol_state_list, zv)) { | |
887 | if (MINOR(zv->zv_dev) > MINOR(zv_insert->zv_dev)) | |
888 | break; | |
889 | } | |
890 | ||
891 | list_insert_before(&zvol_state_list, zv, zv_insert); | |
892 | } | |
893 | ||
894 | /* | |
895 | * Simply remove the zvol from to list of zvols. | |
896 | */ | |
897 | static void | |
898 | zvol_remove(zvol_state_t *zv_remove) | |
899 | { | |
900 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
901 | list_remove(&zvol_state_list, zv_remove); | |
902 | } | |
903 | ||
904 | static int | |
905 | zvol_first_open(zvol_state_t *zv) | |
906 | { | |
907 | objset_t *os; | |
908 | uint64_t volsize; | |
65d56083 | 909 | int locked = 0; |
60101509 BB |
910 | int error; |
911 | uint64_t ro; | |
912 | ||
65d56083 BB |
913 | /* |
914 | * In all other cases the spa_namespace_lock is taken before the | |
915 | * bdev->bd_mutex lock. But in this case the Linux __blkdev_get() | |
916 | * function calls fops->open() with the bdev->bd_mutex lock held. | |
917 | * | |
918 | * To avoid a potential lock inversion deadlock we preemptively | |
919 | * try to take the spa_namespace_lock(). Normally it will not | |
920 | * be contended and this is safe because spa_open_common() handles | |
921 | * the case where the caller already holds the spa_namespace_lock. | |
922 | * | |
923 | * When it is contended we risk a lock inversion if we were to | |
924 | * block waiting for the lock. Luckily, the __blkdev_get() | |
925 | * function allows us to return -ERESTARTSYS which will result in | |
926 | * bdev->bd_mutex being dropped, reacquired, and fops->open() being | |
927 | * called again. This process can be repeated safely until both | |
928 | * locks are acquired. | |
929 | */ | |
930 | if (!mutex_owned(&spa_namespace_lock)) { | |
931 | locked = mutex_tryenter(&spa_namespace_lock); | |
932 | if (!locked) | |
2e528b49 | 933 | return (-SET_ERROR(ERESTARTSYS)); |
65d56083 BB |
934 | } |
935 | ||
60101509 BB |
936 | /* lie and say we're read-only */ |
937 | error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, 1, zvol_tag, &os); | |
938 | if (error) | |
babf3f9b | 939 | goto out_mutex; |
60101509 BB |
940 | |
941 | error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); | |
942 | if (error) { | |
babf3f9b MM |
943 | dmu_objset_disown(os, zvol_tag); |
944 | goto out_mutex; | |
60101509 BB |
945 | } |
946 | ||
947 | zv->zv_objset = os; | |
948 | error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); | |
949 | if (error) { | |
babf3f9b MM |
950 | dmu_objset_disown(os, zvol_tag); |
951 | goto out_mutex; | |
60101509 BB |
952 | } |
953 | ||
954 | set_capacity(zv->zv_disk, volsize >> 9); | |
955 | zv->zv_volsize = volsize; | |
956 | zv->zv_zilog = zil_open(os, zvol_get_data); | |
957 | ||
958 | VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL) == 0); | |
a4430fce GW |
959 | if (ro || dmu_objset_is_snapshot(os) || |
960 | !spa_writeable(dmu_objset_spa(os))) { | |
babf3f9b MM |
961 | set_disk_ro(zv->zv_disk, 1); |
962 | zv->zv_flags |= ZVOL_RDONLY; | |
60101509 | 963 | } else { |
babf3f9b MM |
964 | set_disk_ro(zv->zv_disk, 0); |
965 | zv->zv_flags &= ~ZVOL_RDONLY; | |
60101509 BB |
966 | } |
967 | ||
babf3f9b MM |
968 | out_mutex: |
969 | if (locked) | |
970 | mutex_exit(&spa_namespace_lock); | |
971 | ||
60101509 BB |
972 | return (-error); |
973 | } | |
974 | ||
975 | static void | |
976 | zvol_last_close(zvol_state_t *zv) | |
977 | { | |
978 | zil_close(zv->zv_zilog); | |
979 | zv->zv_zilog = NULL; | |
04434775 | 980 | |
60101509 BB |
981 | dmu_buf_rele(zv->zv_dbuf, zvol_tag); |
982 | zv->zv_dbuf = NULL; | |
04434775 MA |
983 | |
984 | /* | |
985 | * Evict cached data | |
986 | */ | |
987 | if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && | |
988 | !(zv->zv_flags & ZVOL_RDONLY)) | |
989 | txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); | |
990 | (void) dmu_objset_evict_dbufs(zv->zv_objset); | |
991 | ||
60101509 BB |
992 | dmu_objset_disown(zv->zv_objset, zvol_tag); |
993 | zv->zv_objset = NULL; | |
994 | } | |
995 | ||
996 | static int | |
997 | zvol_open(struct block_device *bdev, fmode_t flag) | |
998 | { | |
999 | zvol_state_t *zv = bdev->bd_disk->private_data; | |
1000 | int error = 0, drop_mutex = 0; | |
1001 | ||
1002 | /* | |
1003 | * If the caller is already holding the mutex do not take it | |
1004 | * again, this will happen as part of zvol_create_minor(). | |
1005 | * Once add_disk() is called the device is live and the kernel | |
1006 | * will attempt to open it to read the partition information. | |
1007 | */ | |
1008 | if (!mutex_owned(&zvol_state_lock)) { | |
1009 | mutex_enter(&zvol_state_lock); | |
1010 | drop_mutex = 1; | |
1011 | } | |
1012 | ||
1013 | ASSERT3P(zv, !=, NULL); | |
1014 | ||
1015 | if (zv->zv_open_count == 0) { | |
1016 | error = zvol_first_open(zv); | |
1017 | if (error) | |
1018 | goto out_mutex; | |
1019 | } | |
1020 | ||
ba6a2402 | 1021 | if ((flag & FMODE_WRITE) && (zv->zv_flags & ZVOL_RDONLY)) { |
60101509 BB |
1022 | error = -EROFS; |
1023 | goto out_open_count; | |
1024 | } | |
1025 | ||
1026 | zv->zv_open_count++; | |
1027 | ||
1028 | out_open_count: | |
1029 | if (zv->zv_open_count == 0) | |
1030 | zvol_last_close(zv); | |
1031 | ||
1032 | out_mutex: | |
1033 | if (drop_mutex) | |
1034 | mutex_exit(&zvol_state_lock); | |
1035 | ||
1036 | check_disk_change(bdev); | |
1037 | ||
1038 | return (error); | |
1039 | } | |
1040 | ||
a1d9543a CD |
1041 | #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID |
1042 | static void | |
1043 | #else | |
60101509 | 1044 | static int |
a1d9543a | 1045 | #endif |
60101509 BB |
1046 | zvol_release(struct gendisk *disk, fmode_t mode) |
1047 | { | |
1048 | zvol_state_t *zv = disk->private_data; | |
1049 | int drop_mutex = 0; | |
1050 | ||
1051 | if (!mutex_owned(&zvol_state_lock)) { | |
1052 | mutex_enter(&zvol_state_lock); | |
1053 | drop_mutex = 1; | |
1054 | } | |
1055 | ||
1056 | ASSERT3P(zv, !=, NULL); | |
1057 | ASSERT3U(zv->zv_open_count, >, 0); | |
1058 | zv->zv_open_count--; | |
1059 | if (zv->zv_open_count == 0) | |
1060 | zvol_last_close(zv); | |
1061 | ||
1062 | if (drop_mutex) | |
1063 | mutex_exit(&zvol_state_lock); | |
1064 | ||
a1d9543a | 1065 | #ifndef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID |
60101509 | 1066 | return (0); |
a1d9543a | 1067 | #endif |
60101509 BB |
1068 | } |
1069 | ||
1070 | static int | |
1071 | zvol_ioctl(struct block_device *bdev, fmode_t mode, | |
1072 | unsigned int cmd, unsigned long arg) | |
1073 | { | |
1074 | zvol_state_t *zv = bdev->bd_disk->private_data; | |
1075 | int error = 0; | |
1076 | ||
1077 | if (zv == NULL) | |
2e528b49 | 1078 | return (-SET_ERROR(ENXIO)); |
60101509 BB |
1079 | |
1080 | switch (cmd) { | |
1081 | case BLKFLSBUF: | |
1082 | zil_commit(zv->zv_zilog, ZVOL_OBJ); | |
1083 | break; | |
4c0d8e50 FN |
1084 | case BLKZNAME: |
1085 | error = copy_to_user((void *)arg, zv->zv_name, MAXNAMELEN); | |
1086 | break; | |
60101509 BB |
1087 | |
1088 | default: | |
1089 | error = -ENOTTY; | |
1090 | break; | |
1091 | ||
1092 | } | |
1093 | ||
1094 | return (error); | |
1095 | } | |
1096 | ||
1097 | #ifdef CONFIG_COMPAT | |
1098 | static int | |
1099 | zvol_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
1100 | unsigned cmd, unsigned long arg) | |
1101 | { | |
1102 | return zvol_ioctl(bdev, mode, cmd, arg); | |
1103 | } | |
1104 | #else | |
1105 | #define zvol_compat_ioctl NULL | |
1106 | #endif | |
1107 | ||
1108 | static int zvol_media_changed(struct gendisk *disk) | |
1109 | { | |
1110 | zvol_state_t *zv = disk->private_data; | |
1111 | ||
1112 | return zv->zv_changed; | |
1113 | } | |
1114 | ||
1115 | static int zvol_revalidate_disk(struct gendisk *disk) | |
1116 | { | |
1117 | zvol_state_t *zv = disk->private_data; | |
1118 | ||
1119 | zv->zv_changed = 0; | |
1120 | set_capacity(zv->zv_disk, zv->zv_volsize >> 9); | |
1121 | ||
1122 | return 0; | |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * Provide a simple virtual geometry for legacy compatibility. For devices | |
1127 | * smaller than 1 MiB a small head and sector count is used to allow very | |
1128 | * tiny devices. For devices over 1 Mib a standard head and sector count | |
1129 | * is used to keep the cylinders count reasonable. | |
1130 | */ | |
1131 | static int | |
1132 | zvol_getgeo(struct block_device *bdev, struct hd_geometry *geo) | |
1133 | { | |
1134 | zvol_state_t *zv = bdev->bd_disk->private_data; | |
1135 | sector_t sectors = get_capacity(zv->zv_disk); | |
1136 | ||
1137 | if (sectors > 2048) { | |
1138 | geo->heads = 16; | |
1139 | geo->sectors = 63; | |
1140 | } else { | |
1141 | geo->heads = 2; | |
1142 | geo->sectors = 4; | |
1143 | } | |
1144 | ||
1145 | geo->start = 0; | |
1146 | geo->cylinders = sectors / (geo->heads * geo->sectors); | |
1147 | ||
1148 | return 0; | |
1149 | } | |
1150 | ||
1151 | static struct kobject * | |
1152 | zvol_probe(dev_t dev, int *part, void *arg) | |
1153 | { | |
1154 | zvol_state_t *zv; | |
1155 | struct kobject *kobj; | |
1156 | ||
1157 | mutex_enter(&zvol_state_lock); | |
1158 | zv = zvol_find_by_dev(dev); | |
23a61ccc | 1159 | kobj = zv ? get_disk(zv->zv_disk) : NULL; |
60101509 BB |
1160 | mutex_exit(&zvol_state_lock); |
1161 | ||
1162 | return kobj; | |
1163 | } | |
1164 | ||
1165 | #ifdef HAVE_BDEV_BLOCK_DEVICE_OPERATIONS | |
1166 | static struct block_device_operations zvol_ops = { | |
1167 | .open = zvol_open, | |
1168 | .release = zvol_release, | |
1169 | .ioctl = zvol_ioctl, | |
1170 | .compat_ioctl = zvol_compat_ioctl, | |
1171 | .media_changed = zvol_media_changed, | |
1172 | .revalidate_disk = zvol_revalidate_disk, | |
1173 | .getgeo = zvol_getgeo, | |
1174 | .owner = THIS_MODULE, | |
1175 | }; | |
1176 | ||
1177 | #else /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */ | |
1178 | ||
1179 | static int | |
1180 | zvol_open_by_inode(struct inode *inode, struct file *file) | |
1181 | { | |
1182 | return zvol_open(inode->i_bdev, file->f_mode); | |
1183 | } | |
1184 | ||
1185 | static int | |
1186 | zvol_release_by_inode(struct inode *inode, struct file *file) | |
1187 | { | |
1188 | return zvol_release(inode->i_bdev->bd_disk, file->f_mode); | |
1189 | } | |
1190 | ||
1191 | static int | |
1192 | zvol_ioctl_by_inode(struct inode *inode, struct file *file, | |
1193 | unsigned int cmd, unsigned long arg) | |
1194 | { | |
b1c58213 NB |
1195 | if (file == NULL || inode == NULL) |
1196 | return -EINVAL; | |
60101509 BB |
1197 | return zvol_ioctl(inode->i_bdev, file->f_mode, cmd, arg); |
1198 | } | |
1199 | ||
1200 | # ifdef CONFIG_COMPAT | |
1201 | static long | |
1202 | zvol_compat_ioctl_by_inode(struct file *file, | |
1203 | unsigned int cmd, unsigned long arg) | |
1204 | { | |
b1c58213 NB |
1205 | if (file == NULL) |
1206 | return -EINVAL; | |
60101509 BB |
1207 | return zvol_compat_ioctl(file->f_dentry->d_inode->i_bdev, |
1208 | file->f_mode, cmd, arg); | |
1209 | } | |
1210 | # else | |
1211 | # define zvol_compat_ioctl_by_inode NULL | |
1212 | # endif | |
1213 | ||
1214 | static struct block_device_operations zvol_ops = { | |
1215 | .open = zvol_open_by_inode, | |
1216 | .release = zvol_release_by_inode, | |
1217 | .ioctl = zvol_ioctl_by_inode, | |
1218 | .compat_ioctl = zvol_compat_ioctl_by_inode, | |
1219 | .media_changed = zvol_media_changed, | |
1220 | .revalidate_disk = zvol_revalidate_disk, | |
1221 | .getgeo = zvol_getgeo, | |
1222 | .owner = THIS_MODULE, | |
1223 | }; | |
1224 | #endif /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */ | |
1225 | ||
1226 | /* | |
1227 | * Allocate memory for a new zvol_state_t and setup the required | |
1228 | * request queue and generic disk structures for the block device. | |
1229 | */ | |
1230 | static zvol_state_t * | |
1231 | zvol_alloc(dev_t dev, const char *name) | |
1232 | { | |
1233 | zvol_state_t *zv; | |
7bd04f2d | 1234 | int error = 0; |
60101509 | 1235 | |
ba6a2402 | 1236 | zv = kmem_zalloc(sizeof (zvol_state_t), KM_PUSHPAGE); |
60101509 | 1237 | |
2a3871d4 RY |
1238 | spin_lock_init(&zv->zv_lock); |
1239 | list_link_init(&zv->zv_next); | |
1240 | ||
60101509 BB |
1241 | zv->zv_queue = blk_init_queue(zvol_request, &zv->zv_lock); |
1242 | if (zv->zv_queue == NULL) | |
1243 | goto out_kmem; | |
1244 | ||
7bd04f2d BB |
1245 | #ifdef HAVE_ELEVATOR_CHANGE |
1246 | error = elevator_change(zv->zv_queue, "noop"); | |
1247 | #endif /* HAVE_ELEVATOR_CHANGE */ | |
1248 | if (error) { | |
1249 | printk("ZFS: Unable to set \"%s\" scheduler for zvol %s: %d\n", | |
1250 | "noop", name, error); | |
1251 | goto out_queue; | |
1252 | } | |
1253 | ||
b18019d2 ED |
1254 | #ifdef HAVE_BLK_QUEUE_FLUSH |
1255 | blk_queue_flush(zv->zv_queue, VDEV_REQ_FLUSH | VDEV_REQ_FUA); | |
1256 | #else | |
1257 | blk_queue_ordered(zv->zv_queue, QUEUE_ORDERED_DRAIN, NULL); | |
1258 | #endif /* HAVE_BLK_QUEUE_FLUSH */ | |
1259 | ||
60101509 BB |
1260 | zv->zv_disk = alloc_disk(ZVOL_MINORS); |
1261 | if (zv->zv_disk == NULL) | |
1262 | goto out_queue; | |
1263 | ||
1264 | zv->zv_queue->queuedata = zv; | |
1265 | zv->zv_dev = dev; | |
1266 | zv->zv_open_count = 0; | |
4c0d8e50 | 1267 | strlcpy(zv->zv_name, name, MAXNAMELEN); |
60101509 BB |
1268 | |
1269 | mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); | |
1270 | avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, | |
1271 | sizeof (rl_t), offsetof(rl_t, r_node)); | |
3c4988c8 BB |
1272 | zv->zv_znode.z_is_zvol = TRUE; |
1273 | ||
60101509 BB |
1274 | zv->zv_disk->major = zvol_major; |
1275 | zv->zv_disk->first_minor = (dev & MINORMASK); | |
1276 | zv->zv_disk->fops = &zvol_ops; | |
1277 | zv->zv_disk->private_data = zv; | |
1278 | zv->zv_disk->queue = zv->zv_queue; | |
4c0d8e50 FN |
1279 | snprintf(zv->zv_disk->disk_name, DISK_NAME_LEN, "%s%d", |
1280 | ZVOL_DEV_NAME, (dev & MINORMASK)); | |
60101509 BB |
1281 | |
1282 | return zv; | |
1283 | ||
1284 | out_queue: | |
1285 | blk_cleanup_queue(zv->zv_queue); | |
1286 | out_kmem: | |
1287 | kmem_free(zv, sizeof (zvol_state_t)); | |
0a6bef26 | 1288 | |
60101509 BB |
1289 | return NULL; |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * Cleanup then free a zvol_state_t which was created by zvol_alloc(). | |
1294 | */ | |
1295 | static void | |
1296 | zvol_free(zvol_state_t *zv) | |
1297 | { | |
1298 | avl_destroy(&zv->zv_znode.z_range_avl); | |
1299 | mutex_destroy(&zv->zv_znode.z_range_lock); | |
1300 | ||
1301 | del_gendisk(zv->zv_disk); | |
1302 | blk_cleanup_queue(zv->zv_queue); | |
1303 | put_disk(zv->zv_disk); | |
1304 | ||
1305 | kmem_free(zv, sizeof (zvol_state_t)); | |
1306 | } | |
1307 | ||
1308 | static int | |
0b4d1b58 ED |
1309 | __zvol_snapdev_hidden(const char *name) |
1310 | { | |
1311 | uint64_t snapdev; | |
1312 | char *parent; | |
1313 | char *atp; | |
1314 | int error = 0; | |
1315 | ||
ba6a2402 | 1316 | parent = kmem_alloc(MAXPATHLEN, KM_PUSHPAGE); |
0b4d1b58 ED |
1317 | (void) strlcpy(parent, name, MAXPATHLEN); |
1318 | ||
1319 | if ((atp = strrchr(parent, '@')) != NULL) { | |
1320 | *atp = '\0'; | |
1321 | error = dsl_prop_get_integer(parent, "snapdev", &snapdev, NULL); | |
1322 | if ((error == 0) && (snapdev == ZFS_SNAPDEV_HIDDEN)) | |
2e528b49 | 1323 | error = SET_ERROR(ENODEV); |
0b4d1b58 ED |
1324 | } |
1325 | kmem_free(parent, MAXPATHLEN); | |
1326 | return (error); | |
1327 | } | |
1328 | ||
1329 | static int | |
1330 | __zvol_create_minor(const char *name, boolean_t ignore_snapdev) | |
60101509 BB |
1331 | { |
1332 | zvol_state_t *zv; | |
1333 | objset_t *os; | |
1334 | dmu_object_info_t *doi; | |
1335 | uint64_t volsize; | |
1336 | unsigned minor = 0; | |
1337 | int error = 0; | |
1338 | ||
1339 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
1340 | ||
1341 | zv = zvol_find_by_name(name); | |
1342 | if (zv) { | |
2e528b49 | 1343 | error = SET_ERROR(EEXIST); |
60101509 BB |
1344 | goto out; |
1345 | } | |
1346 | ||
0b4d1b58 ED |
1347 | if (ignore_snapdev == B_FALSE) { |
1348 | error = __zvol_snapdev_hidden(name); | |
1349 | if (error) | |
1350 | goto out; | |
1351 | } | |
1352 | ||
ba6a2402 | 1353 | doi = kmem_alloc(sizeof(dmu_object_info_t), KM_PUSHPAGE); |
60101509 BB |
1354 | |
1355 | error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, zvol_tag, &os); | |
1356 | if (error) | |
1357 | goto out_doi; | |
1358 | ||
1359 | error = dmu_object_info(os, ZVOL_OBJ, doi); | |
1360 | if (error) | |
1361 | goto out_dmu_objset_disown; | |
1362 | ||
1363 | error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); | |
1364 | if (error) | |
1365 | goto out_dmu_objset_disown; | |
1366 | ||
1367 | error = zvol_find_minor(&minor); | |
1368 | if (error) | |
1369 | goto out_dmu_objset_disown; | |
1370 | ||
1371 | zv = zvol_alloc(MKDEV(zvol_major, minor), name); | |
1372 | if (zv == NULL) { | |
2e528b49 | 1373 | error = SET_ERROR(EAGAIN); |
60101509 BB |
1374 | goto out_dmu_objset_disown; |
1375 | } | |
1376 | ||
1377 | if (dmu_objset_is_snapshot(os)) | |
1378 | zv->zv_flags |= ZVOL_RDONLY; | |
1379 | ||
1380 | zv->zv_volblocksize = doi->doi_data_block_size; | |
1381 | zv->zv_volsize = volsize; | |
1382 | zv->zv_objset = os; | |
1383 | ||
1384 | set_capacity(zv->zv_disk, zv->zv_volsize >> 9); | |
1385 | ||
34037afe ED |
1386 | blk_queue_max_hw_sectors(zv->zv_queue, UINT_MAX); |
1387 | blk_queue_max_segments(zv->zv_queue, UINT16_MAX); | |
1388 | blk_queue_max_segment_size(zv->zv_queue, UINT_MAX); | |
1389 | blk_queue_physical_block_size(zv->zv_queue, zv->zv_volblocksize); | |
1390 | blk_queue_io_opt(zv->zv_queue, zv->zv_volblocksize); | |
30930fba | 1391 | #ifdef HAVE_BLK_QUEUE_DISCARD |
7c0e5708 ED |
1392 | blk_queue_max_discard_sectors(zv->zv_queue, |
1393 | (zvol_max_discard_blocks * zv->zv_volblocksize) >> 9); | |
ee5fd0bb | 1394 | blk_queue_discard_granularity(zv->zv_queue, zv->zv_volblocksize); |
30930fba ED |
1395 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zv->zv_queue); |
1396 | #endif | |
34037afe ED |
1397 | #ifdef HAVE_BLK_QUEUE_NONROT |
1398 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zv->zv_queue); | |
1399 | #endif | |
1400 | ||
a4430fce GW |
1401 | if (spa_writeable(dmu_objset_spa(os))) { |
1402 | if (zil_replay_disable) | |
1403 | zil_destroy(dmu_objset_zil(os), B_FALSE); | |
1404 | else | |
1405 | zil_replay(os, zv, zvol_replay_vector); | |
1406 | } | |
60101509 | 1407 | |
f74a147c | 1408 | zv->zv_objset = NULL; |
60101509 BB |
1409 | out_dmu_objset_disown: |
1410 | dmu_objset_disown(os, zvol_tag); | |
60101509 BB |
1411 | out_doi: |
1412 | kmem_free(doi, sizeof(dmu_object_info_t)); | |
1413 | out: | |
1414 | ||
1415 | if (error == 0) { | |
1416 | zvol_insert(zv); | |
1417 | add_disk(zv->zv_disk); | |
1418 | } | |
1419 | ||
1420 | return (error); | |
1421 | } | |
1422 | ||
1423 | /* | |
1424 | * Create a block device minor node and setup the linkage between it | |
1425 | * and the specified volume. Once this function returns the block | |
1426 | * device is live and ready for use. | |
1427 | */ | |
1428 | int | |
1429 | zvol_create_minor(const char *name) | |
1430 | { | |
1431 | int error; | |
1432 | ||
1433 | mutex_enter(&zvol_state_lock); | |
0b4d1b58 | 1434 | error = __zvol_create_minor(name, B_FALSE); |
60101509 BB |
1435 | mutex_exit(&zvol_state_lock); |
1436 | ||
1437 | return (error); | |
1438 | } | |
1439 | ||
1440 | static int | |
1441 | __zvol_remove_minor(const char *name) | |
1442 | { | |
1443 | zvol_state_t *zv; | |
1444 | ||
1445 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
1446 | ||
1447 | zv = zvol_find_by_name(name); | |
1448 | if (zv == NULL) | |
2e528b49 | 1449 | return (SET_ERROR(ENXIO)); |
60101509 BB |
1450 | |
1451 | if (zv->zv_open_count > 0) | |
2e528b49 | 1452 | return (SET_ERROR(EBUSY)); |
60101509 BB |
1453 | |
1454 | zvol_remove(zv); | |
1455 | zvol_free(zv); | |
1456 | ||
1457 | return (0); | |
1458 | } | |
1459 | ||
1460 | /* | |
1461 | * Remove a block device minor node for the specified volume. | |
1462 | */ | |
1463 | int | |
1464 | zvol_remove_minor(const char *name) | |
1465 | { | |
1466 | int error; | |
1467 | ||
1468 | mutex_enter(&zvol_state_lock); | |
1469 | error = __zvol_remove_minor(name); | |
1470 | mutex_exit(&zvol_state_lock); | |
1471 | ||
1472 | return (error); | |
1473 | } | |
1474 | ||
ba6a2402 BB |
1475 | /* |
1476 | * Rename a block device minor mode for the specified volume. | |
1477 | */ | |
1478 | static void | |
1479 | __zvol_rename_minor(zvol_state_t *zv, const char *newname) | |
1480 | { | |
1481 | int readonly = get_disk_ro(zv->zv_disk); | |
1482 | ||
1483 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
1484 | ||
1485 | strlcpy(zv->zv_name, newname, sizeof (zv->zv_name)); | |
1486 | ||
1487 | /* | |
1488 | * The block device's read-only state is briefly changed causing | |
1489 | * a KOBJ_CHANGE uevent to be issued. This ensures udev detects | |
1490 | * the name change and fixes the symlinks. This does not change | |
1491 | * ZVOL_RDONLY in zv->zv_flags so the actual read-only state never | |
1492 | * changes. This would normally be done using kobject_uevent() but | |
1493 | * that is a GPL-only symbol which is why we need this workaround. | |
1494 | */ | |
1495 | set_disk_ro(zv->zv_disk, !readonly); | |
1496 | set_disk_ro(zv->zv_disk, readonly); | |
1497 | } | |
1498 | ||
60101509 | 1499 | static int |
13fe0198 | 1500 | zvol_create_minors_cb(const char *dsname, void *arg) |
60101509 | 1501 | { |
ba6a2402 | 1502 | (void) zvol_create_minor(dsname); |
60101509 | 1503 | |
d5674448 | 1504 | return (0); |
60101509 BB |
1505 | } |
1506 | ||
1507 | /* | |
ba6a2402 | 1508 | * Create minors for specified dataset including children and snapshots. |
60101509 BB |
1509 | */ |
1510 | int | |
ba6a2402 | 1511 | zvol_create_minors(const char *name) |
60101509 | 1512 | { |
60101509 BB |
1513 | int error = 0; |
1514 | ||
ba6a2402 BB |
1515 | if (!zvol_inhibit_dev) |
1516 | error = dmu_objset_find((char *)name, zvol_create_minors_cb, | |
1517 | NULL, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); | |
1518 | ||
1519 | return (SET_ERROR(error)); | |
1520 | } | |
1521 | ||
1522 | /* | |
1523 | * Remove minors for specified dataset including children and snapshots. | |
1524 | */ | |
1525 | void | |
1526 | zvol_remove_minors(const char *name) | |
1527 | { | |
1528 | zvol_state_t *zv, *zv_next; | |
1529 | int namelen = ((name) ? strlen(name) : 0); | |
1530 | ||
74497b7a | 1531 | if (zvol_inhibit_dev) |
ba6a2402 | 1532 | return; |
74497b7a | 1533 | |
60101509 | 1534 | mutex_enter(&zvol_state_lock); |
ba6a2402 BB |
1535 | |
1536 | for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { | |
1537 | zv_next = list_next(&zvol_state_list, zv); | |
1538 | ||
1539 | if (name == NULL || strcmp(zv->zv_name, name) == 0 || | |
1540 | (strncmp(zv->zv_name, name, namelen) == 0 && | |
1541 | zv->zv_name[namelen] == '/')) { | |
1542 | zvol_remove(zv); | |
1543 | zvol_free(zv); | |
60101509 | 1544 | } |
60101509 | 1545 | } |
60101509 | 1546 | |
ba6a2402 | 1547 | mutex_exit(&zvol_state_lock); |
60101509 BB |
1548 | } |
1549 | ||
1550 | /* | |
ba6a2402 | 1551 | * Rename minors for specified dataset including children and snapshots. |
60101509 BB |
1552 | */ |
1553 | void | |
ba6a2402 | 1554 | zvol_rename_minors(const char *oldname, const char *newname) |
60101509 BB |
1555 | { |
1556 | zvol_state_t *zv, *zv_next; | |
ba6a2402 BB |
1557 | int oldnamelen, newnamelen; |
1558 | char *name; | |
60101509 | 1559 | |
74497b7a DH |
1560 | if (zvol_inhibit_dev) |
1561 | return; | |
1562 | ||
ba6a2402 BB |
1563 | oldnamelen = strlen(oldname); |
1564 | newnamelen = strlen(newname); | |
1565 | name = kmem_alloc(MAXNAMELEN, KM_PUSHPAGE); | |
60101509 BB |
1566 | |
1567 | mutex_enter(&zvol_state_lock); | |
ba6a2402 | 1568 | |
60101509 BB |
1569 | for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { |
1570 | zv_next = list_next(&zvol_state_list, zv); | |
1571 | ||
ba6a2402 BB |
1572 | if (strcmp(zv->zv_name, oldname) == 0) { |
1573 | __zvol_rename_minor(zv, newname); | |
1574 | } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && | |
1575 | (zv->zv_name[oldnamelen] == '/' || | |
1576 | zv->zv_name[oldnamelen] == '@')) { | |
1577 | snprintf(name, MAXNAMELEN, "%s%c%s", newname, | |
1578 | zv->zv_name[oldnamelen], | |
1579 | zv->zv_name + oldnamelen + 1); | |
1580 | __zvol_rename_minor(zv, name); | |
60101509 BB |
1581 | } |
1582 | } | |
ba6a2402 | 1583 | |
60101509 | 1584 | mutex_exit(&zvol_state_lock); |
ba6a2402 BB |
1585 | |
1586 | kmem_free(name, MAXNAMELEN); | |
60101509 BB |
1587 | } |
1588 | ||
0b4d1b58 ED |
1589 | static int |
1590 | snapdev_snapshot_changed_cb(const char *dsname, void *arg) { | |
1591 | uint64_t snapdev = *(uint64_t *) arg; | |
1592 | ||
1593 | if (strchr(dsname, '@') == NULL) | |
ba6a2402 | 1594 | return (0); |
0b4d1b58 ED |
1595 | |
1596 | switch (snapdev) { | |
1597 | case ZFS_SNAPDEV_VISIBLE: | |
1598 | mutex_enter(&zvol_state_lock); | |
1599 | (void) __zvol_create_minor(dsname, B_TRUE); | |
1600 | mutex_exit(&zvol_state_lock); | |
1601 | break; | |
1602 | case ZFS_SNAPDEV_HIDDEN: | |
1603 | (void) zvol_remove_minor(dsname); | |
1604 | break; | |
1605 | } | |
ba6a2402 BB |
1606 | |
1607 | return (0); | |
0b4d1b58 ED |
1608 | } |
1609 | ||
1610 | int | |
1611 | zvol_set_snapdev(const char *dsname, uint64_t snapdev) { | |
1612 | (void) dmu_objset_find((char *) dsname, snapdev_snapshot_changed_cb, | |
1613 | &snapdev, DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); | |
1614 | /* caller should continue to modify snapdev property */ | |
1615 | return (-1); | |
1616 | } | |
1617 | ||
1618 | ||
60101509 BB |
1619 | int |
1620 | zvol_init(void) | |
1621 | { | |
1622 | int error; | |
1623 | ||
2a3871d4 RY |
1624 | list_create(&zvol_state_list, sizeof (zvol_state_t), |
1625 | offsetof(zvol_state_t, zv_next)); | |
1626 | mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL); | |
1627 | ||
60101509 | 1628 | zvol_taskq = taskq_create(ZVOL_DRIVER, zvol_threads, maxclsyspri, |
71011408 | 1629 | zvol_threads, INT_MAX, TASKQ_PREPOPULATE); |
60101509 BB |
1630 | if (zvol_taskq == NULL) { |
1631 | printk(KERN_INFO "ZFS: taskq_create() failed\n"); | |
2a3871d4 RY |
1632 | error = -ENOMEM; |
1633 | goto out1; | |
60101509 BB |
1634 | } |
1635 | ||
1636 | error = register_blkdev(zvol_major, ZVOL_DRIVER); | |
1637 | if (error) { | |
1638 | printk(KERN_INFO "ZFS: register_blkdev() failed %d\n", error); | |
2a3871d4 | 1639 | goto out2; |
60101509 BB |
1640 | } |
1641 | ||
1642 | blk_register_region(MKDEV(zvol_major, 0), 1UL << MINORBITS, | |
1643 | THIS_MODULE, zvol_probe, NULL, NULL); | |
1644 | ||
60101509 | 1645 | return (0); |
2a3871d4 RY |
1646 | |
1647 | out2: | |
1648 | taskq_destroy(zvol_taskq); | |
1649 | out1: | |
1650 | mutex_destroy(&zvol_state_lock); | |
1651 | list_destroy(&zvol_state_list); | |
1652 | ||
1653 | return (error); | |
60101509 BB |
1654 | } |
1655 | ||
1656 | void | |
1657 | zvol_fini(void) | |
1658 | { | |
1659 | zvol_remove_minors(NULL); | |
1660 | blk_unregister_region(MKDEV(zvol_major, 0), 1UL << MINORBITS); | |
1661 | unregister_blkdev(zvol_major, ZVOL_DRIVER); | |
1662 | taskq_destroy(zvol_taskq); | |
1663 | mutex_destroy(&zvol_state_lock); | |
1664 | list_destroy(&zvol_state_list); | |
1665 | } | |
1666 | ||
74497b7a DH |
1667 | module_param(zvol_inhibit_dev, uint, 0644); |
1668 | MODULE_PARM_DESC(zvol_inhibit_dev, "Do not create zvol device nodes"); | |
1669 | ||
30a9524e | 1670 | module_param(zvol_major, uint, 0444); |
60101509 BB |
1671 | MODULE_PARM_DESC(zvol_major, "Major number for zvol device"); |
1672 | ||
30a9524e | 1673 | module_param(zvol_threads, uint, 0444); |
60101509 | 1674 | MODULE_PARM_DESC(zvol_threads, "Number of threads for zvol device"); |
7c0e5708 ED |
1675 | |
1676 | module_param(zvol_max_discard_blocks, ulong, 0444); | |
1677 | MODULE_PARM_DESC(zvol_max_discard_blocks, "Max number of blocks to discard at once"); |