]>
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 | ||
38 | #include <sys/dmu_traverse.h> | |
39 | #include <sys/dsl_dataset.h> | |
40 | #include <sys/dsl_prop.h> | |
41 | #include <sys/zap.h> | |
42 | #include <sys/zil_impl.h> | |
43 | #include <sys/zio.h> | |
44 | #include <sys/zfs_rlock.h> | |
45 | #include <sys/zfs_znode.h> | |
46 | #include <sys/zvol.h> | |
47 | ||
48 | unsigned int zvol_major = ZVOL_MAJOR; | |
49 | unsigned int zvol_threads = 0; | |
50 | ||
51 | static taskq_t *zvol_taskq; | |
52 | static kmutex_t zvol_state_lock; | |
53 | static list_t zvol_state_list; | |
54 | static char *zvol_tag = "zvol_tag"; | |
55 | ||
56 | /* | |
57 | * The in-core state of each volume. | |
58 | */ | |
59 | typedef struct zvol_state { | |
60 | char zv_name[DISK_NAME_LEN]; /* name */ | |
61 | uint64_t zv_volsize; /* advertised space */ | |
62 | uint64_t zv_volblocksize;/* volume block size */ | |
63 | objset_t *zv_objset; /* objset handle */ | |
64 | uint32_t zv_flags; /* ZVOL_* flags */ | |
65 | uint32_t zv_open_count; /* open counts */ | |
66 | uint32_t zv_changed; /* disk changed */ | |
67 | zilog_t *zv_zilog; /* ZIL handle */ | |
68 | znode_t zv_znode; /* for range locking */ | |
69 | dmu_buf_t *zv_dbuf; /* bonus handle */ | |
70 | dev_t zv_dev; /* device id */ | |
71 | struct gendisk *zv_disk; /* generic disk */ | |
72 | struct request_queue *zv_queue; /* request queue */ | |
73 | spinlock_t zv_lock; /* request queue lock */ | |
74 | list_node_t zv_next; /* next zvol_state_t linkage */ | |
75 | } zvol_state_t; | |
76 | ||
77 | #define ZVOL_RDONLY 0x1 | |
78 | ||
79 | /* | |
80 | * Find the next available range of ZVOL_MINORS minor numbers. The | |
81 | * zvol_state_list is kept in ascending minor order so we simply need | |
82 | * to scan the list for the first gap in the sequence. This allows us | |
83 | * to recycle minor number as devices are created and removed. | |
84 | */ | |
85 | static int | |
86 | zvol_find_minor(unsigned *minor) | |
87 | { | |
88 | zvol_state_t *zv; | |
89 | ||
90 | *minor = 0; | |
91 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
92 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
93 | zv = list_next(&zvol_state_list, zv), *minor += ZVOL_MINORS) { | |
94 | if (MINOR(zv->zv_dev) != MINOR(*minor)) | |
95 | break; | |
96 | } | |
97 | ||
98 | /* All minors are in use */ | |
99 | if (*minor >= (1 << MINORBITS)) | |
100 | return ENXIO; | |
101 | ||
102 | return 0; | |
103 | } | |
104 | ||
105 | /* | |
106 | * Find a zvol_state_t given the full major+minor dev_t. | |
107 | */ | |
108 | static zvol_state_t * | |
109 | zvol_find_by_dev(dev_t dev) | |
110 | { | |
111 | zvol_state_t *zv; | |
112 | ||
113 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
114 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
115 | zv = list_next(&zvol_state_list, zv)) { | |
116 | if (zv->zv_dev == dev) | |
117 | return zv; | |
118 | } | |
119 | ||
120 | return NULL; | |
121 | } | |
122 | ||
123 | /* | |
124 | * Find a zvol_state_t given the name provided at zvol_alloc() time. | |
125 | */ | |
126 | static zvol_state_t * | |
127 | zvol_find_by_name(const char *name) | |
128 | { | |
129 | zvol_state_t *zv; | |
130 | ||
131 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
132 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
133 | zv = list_next(&zvol_state_list, zv)) { | |
134 | if (!strncmp(zv->zv_name, name, DISK_NAME_LEN)) | |
135 | return zv; | |
136 | } | |
137 | ||
138 | return NULL; | |
139 | } | |
140 | ||
141 | /* | |
142 | * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation. | |
143 | */ | |
144 | void | |
145 | zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) | |
146 | { | |
147 | zfs_creat_t *zct = arg; | |
148 | nvlist_t *nvprops = zct->zct_props; | |
149 | int error; | |
150 | uint64_t volblocksize, volsize; | |
151 | ||
152 | VERIFY(nvlist_lookup_uint64(nvprops, | |
153 | zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); | |
154 | if (nvlist_lookup_uint64(nvprops, | |
155 | zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) | |
156 | volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); | |
157 | ||
158 | /* | |
159 | * These properties must be removed from the list so the generic | |
160 | * property setting step won't apply to them. | |
161 | */ | |
162 | VERIFY(nvlist_remove_all(nvprops, | |
163 | zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); | |
164 | (void) nvlist_remove_all(nvprops, | |
165 | zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); | |
166 | ||
167 | error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, | |
168 | DMU_OT_NONE, 0, tx); | |
169 | ASSERT(error == 0); | |
170 | ||
171 | error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, | |
172 | DMU_OT_NONE, 0, tx); | |
173 | ASSERT(error == 0); | |
174 | ||
175 | error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); | |
176 | ASSERT(error == 0); | |
177 | } | |
178 | ||
179 | /* | |
180 | * ZFS_IOC_OBJSET_STATS entry point. | |
181 | */ | |
182 | int | |
183 | zvol_get_stats(objset_t *os, nvlist_t *nv) | |
184 | { | |
185 | int error; | |
186 | dmu_object_info_t *doi; | |
187 | uint64_t val; | |
188 | ||
189 | error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); | |
190 | if (error) | |
191 | return (error); | |
192 | ||
193 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); | |
194 | doi = kmem_alloc(sizeof(dmu_object_info_t), KM_SLEEP); | |
195 | error = dmu_object_info(os, ZVOL_OBJ, doi); | |
196 | ||
197 | if (error == 0) { | |
198 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, | |
199 | doi->doi_data_block_size); | |
200 | } | |
201 | ||
202 | kmem_free(doi, sizeof(dmu_object_info_t)); | |
203 | ||
204 | return (error); | |
205 | } | |
206 | ||
207 | /* | |
208 | * Sanity check volume size. | |
209 | */ | |
210 | int | |
211 | zvol_check_volsize(uint64_t volsize, uint64_t blocksize) | |
212 | { | |
213 | if (volsize == 0) | |
214 | return (EINVAL); | |
215 | ||
216 | if (volsize % blocksize != 0) | |
217 | return (EINVAL); | |
218 | ||
219 | #ifdef _ILP32 | |
220 | if (volsize - 1 > MAXOFFSET_T) | |
221 | return (EOVERFLOW); | |
222 | #endif | |
223 | return (0); | |
224 | } | |
225 | ||
226 | /* | |
227 | * Ensure the zap is flushed then inform the VFS of the capacity change. | |
228 | */ | |
229 | static int | |
230 | zvol_update_volsize(zvol_state_t *zv, uint64_t volsize) | |
231 | { | |
232 | struct block_device *bdev; | |
233 | dmu_tx_t *tx; | |
234 | int error; | |
235 | ||
236 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
237 | ||
238 | tx = dmu_tx_create(zv->zv_objset); | |
239 | dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); | |
240 | error = dmu_tx_assign(tx, TXG_WAIT); | |
241 | if (error) { | |
242 | dmu_tx_abort(tx); | |
243 | return (error); | |
244 | } | |
245 | ||
246 | error = zap_update(zv->zv_objset, ZVOL_ZAP_OBJ, "size", 8, 1, | |
247 | &volsize, tx); | |
248 | dmu_tx_commit(tx); | |
249 | ||
250 | if (error) | |
251 | return (error); | |
252 | ||
253 | error = dmu_free_long_range(zv->zv_objset, | |
254 | ZVOL_OBJ, volsize, DMU_OBJECT_END); | |
255 | if (error) | |
256 | return (error); | |
257 | ||
258 | zv->zv_volsize = volsize; | |
259 | zv->zv_changed = 1; | |
260 | ||
261 | bdev = bdget_disk(zv->zv_disk, 0); | |
262 | if (!bdev) | |
263 | return EIO; | |
264 | ||
265 | error = check_disk_change(bdev); | |
266 | ASSERT3U(error, !=, 0); | |
267 | bdput(bdev); | |
268 | ||
269 | return (0); | |
270 | } | |
271 | ||
272 | /* | |
273 | * Set ZFS_PROP_VOLSIZE set entry point. | |
274 | */ | |
275 | int | |
276 | zvol_set_volsize(const char *name, uint64_t volsize) | |
277 | { | |
278 | zvol_state_t *zv; | |
279 | dmu_object_info_t *doi; | |
280 | objset_t *os = NULL; | |
281 | uint64_t readonly; | |
282 | int error; | |
283 | ||
284 | mutex_enter(&zvol_state_lock); | |
285 | ||
286 | zv = zvol_find_by_name(name); | |
287 | if (zv == NULL) { | |
288 | error = ENXIO; | |
289 | goto out; | |
290 | } | |
291 | ||
292 | doi = kmem_alloc(sizeof(dmu_object_info_t), KM_SLEEP); | |
293 | ||
294 | error = dmu_objset_hold(name, FTAG, &os); | |
295 | if (error) | |
296 | goto out_doi; | |
297 | ||
298 | if ((error = dmu_object_info(os, ZVOL_OBJ, doi)) != 0 || | |
299 | (error = zvol_check_volsize(volsize,doi->doi_data_block_size)) != 0) | |
300 | goto out_doi; | |
301 | ||
302 | VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, NULL) == 0); | |
303 | if (readonly) { | |
304 | error = EROFS; | |
305 | goto out_doi; | |
306 | } | |
307 | ||
308 | if (get_disk_ro(zv->zv_disk) || (zv->zv_flags & ZVOL_RDONLY)) { | |
309 | error = EROFS; | |
310 | goto out_doi; | |
311 | } | |
312 | ||
313 | error = zvol_update_volsize(zv, volsize); | |
314 | out_doi: | |
315 | kmem_free(doi, sizeof(dmu_object_info_t)); | |
316 | out: | |
317 | if (os) | |
318 | dmu_objset_rele(os, FTAG); | |
319 | ||
320 | mutex_exit(&zvol_state_lock); | |
321 | ||
322 | return (error); | |
323 | } | |
324 | ||
325 | /* | |
326 | * Sanity check volume block size. | |
327 | */ | |
328 | int | |
329 | zvol_check_volblocksize(uint64_t volblocksize) | |
330 | { | |
331 | if (volblocksize < SPA_MINBLOCKSIZE || | |
332 | volblocksize > SPA_MAXBLOCKSIZE || | |
333 | !ISP2(volblocksize)) | |
334 | return (EDOM); | |
335 | ||
336 | return (0); | |
337 | } | |
338 | ||
339 | /* | |
340 | * Set ZFS_PROP_VOLBLOCKSIZE set entry point. | |
341 | */ | |
342 | int | |
343 | zvol_set_volblocksize(const char *name, uint64_t volblocksize) | |
344 | { | |
345 | zvol_state_t *zv; | |
346 | dmu_tx_t *tx; | |
347 | int error; | |
348 | ||
349 | mutex_enter(&zvol_state_lock); | |
350 | ||
351 | zv = zvol_find_by_name(name); | |
352 | if (zv == NULL) { | |
353 | error = ENXIO; | |
354 | goto out; | |
355 | } | |
356 | ||
357 | if (get_disk_ro(zv->zv_disk) || (zv->zv_flags & ZVOL_RDONLY)) { | |
358 | error = EROFS; | |
359 | goto out; | |
360 | } | |
361 | ||
362 | tx = dmu_tx_create(zv->zv_objset); | |
363 | dmu_tx_hold_bonus(tx, ZVOL_OBJ); | |
364 | error = dmu_tx_assign(tx, TXG_WAIT); | |
365 | if (error) { | |
366 | dmu_tx_abort(tx); | |
367 | } else { | |
368 | error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ, | |
369 | volblocksize, 0, tx); | |
370 | if (error == ENOTSUP) | |
371 | error = EBUSY; | |
372 | dmu_tx_commit(tx); | |
373 | if (error == 0) | |
374 | zv->zv_volblocksize = volblocksize; | |
375 | } | |
376 | out: | |
377 | mutex_exit(&zvol_state_lock); | |
378 | ||
379 | return (error); | |
380 | } | |
381 | ||
382 | /* | |
383 | * Replay a TX_WRITE ZIL transaction that didn't get committed | |
384 | * after a system failure | |
385 | */ | |
386 | static int | |
387 | zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) | |
388 | { | |
389 | objset_t *os = zv->zv_objset; | |
390 | char *data = (char *)(lr + 1); /* data follows lr_write_t */ | |
391 | uint64_t off = lr->lr_offset; | |
392 | uint64_t len = lr->lr_length; | |
393 | dmu_tx_t *tx; | |
394 | int error; | |
395 | ||
396 | if (byteswap) | |
397 | byteswap_uint64_array(lr, sizeof (*lr)); | |
398 | ||
399 | tx = dmu_tx_create(os); | |
400 | dmu_tx_hold_write(tx, ZVOL_OBJ, off, len); | |
401 | error = dmu_tx_assign(tx, TXG_WAIT); | |
402 | if (error) { | |
403 | dmu_tx_abort(tx); | |
404 | } else { | |
405 | dmu_write(os, ZVOL_OBJ, off, len, data, tx); | |
406 | dmu_tx_commit(tx); | |
407 | } | |
408 | ||
409 | return (error); | |
410 | } | |
411 | ||
412 | static int | |
413 | zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) | |
414 | { | |
415 | return (ENOTSUP); | |
416 | } | |
417 | ||
418 | /* | |
419 | * Callback vectors for replaying records. | |
420 | * Only TX_WRITE is needed for zvol. | |
421 | */ | |
422 | zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { | |
423 | (zil_replay_func_t *)zvol_replay_err, /* no such transaction type */ | |
424 | (zil_replay_func_t *)zvol_replay_err, /* TX_CREATE */ | |
425 | (zil_replay_func_t *)zvol_replay_err, /* TX_MKDIR */ | |
426 | (zil_replay_func_t *)zvol_replay_err, /* TX_MKXATTR */ | |
427 | (zil_replay_func_t *)zvol_replay_err, /* TX_SYMLINK */ | |
428 | (zil_replay_func_t *)zvol_replay_err, /* TX_REMOVE */ | |
429 | (zil_replay_func_t *)zvol_replay_err, /* TX_RMDIR */ | |
430 | (zil_replay_func_t *)zvol_replay_err, /* TX_LINK */ | |
431 | (zil_replay_func_t *)zvol_replay_err, /* TX_RENAME */ | |
432 | (zil_replay_func_t *)zvol_replay_write, /* TX_WRITE */ | |
433 | (zil_replay_func_t *)zvol_replay_err, /* TX_TRUNCATE */ | |
434 | (zil_replay_func_t *)zvol_replay_err, /* TX_SETATTR */ | |
435 | (zil_replay_func_t *)zvol_replay_err, /* TX_ACL */ | |
436 | }; | |
437 | ||
438 | /* | |
439 | * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. | |
440 | * | |
441 | * We store data in the log buffers if it's small enough. | |
442 | * Otherwise we will later flush the data out via dmu_sync(). | |
443 | */ | |
444 | ssize_t zvol_immediate_write_sz = 32768; | |
445 | ||
446 | static void | |
447 | zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, | |
448 | uint64_t offset, uint64_t size, int sync) | |
449 | { | |
450 | uint32_t blocksize = zv->zv_volblocksize; | |
451 | zilog_t *zilog = zv->zv_zilog; | |
452 | boolean_t slogging; | |
453 | ||
454 | if (zil_replaying(zilog, tx)) | |
455 | return; | |
456 | ||
457 | slogging = spa_has_slogs(zilog->zl_spa); | |
458 | ||
459 | while (size) { | |
460 | itx_t *itx; | |
461 | lr_write_t *lr; | |
462 | ssize_t len; | |
463 | itx_wr_state_t write_state; | |
464 | ||
465 | /* | |
466 | * Unlike zfs_log_write() we can be called with | |
467 | * up to DMU_MAX_ACCESS/2 (5MB) writes. | |
468 | */ | |
469 | if (blocksize > zvol_immediate_write_sz && !slogging && | |
470 | size >= blocksize && offset % blocksize == 0) { | |
471 | write_state = WR_INDIRECT; /* uses dmu_sync */ | |
472 | len = blocksize; | |
473 | } else if (sync) { | |
474 | write_state = WR_COPIED; | |
475 | len = MIN(ZIL_MAX_LOG_DATA, size); | |
476 | } else { | |
477 | write_state = WR_NEED_COPY; | |
478 | len = MIN(ZIL_MAX_LOG_DATA, size); | |
479 | } | |
480 | ||
481 | itx = zil_itx_create(TX_WRITE, sizeof (*lr) + | |
482 | (write_state == WR_COPIED ? len : 0)); | |
483 | lr = (lr_write_t *)&itx->itx_lr; | |
484 | if (write_state == WR_COPIED && dmu_read(zv->zv_objset, | |
485 | ZVOL_OBJ, offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) { | |
486 | zil_itx_destroy(itx); | |
487 | itx = zil_itx_create(TX_WRITE, sizeof (*lr)); | |
488 | lr = (lr_write_t *)&itx->itx_lr; | |
489 | write_state = WR_NEED_COPY; | |
490 | } | |
491 | ||
492 | itx->itx_wr_state = write_state; | |
493 | if (write_state == WR_NEED_COPY) | |
494 | itx->itx_sod += len; | |
495 | lr->lr_foid = ZVOL_OBJ; | |
496 | lr->lr_offset = offset; | |
497 | lr->lr_length = len; | |
498 | lr->lr_blkoff = 0; | |
499 | BP_ZERO(&lr->lr_blkptr); | |
500 | ||
501 | itx->itx_private = zv; | |
502 | itx->itx_sync = sync; | |
503 | ||
504 | (void) zil_itx_assign(zilog, itx, tx); | |
505 | ||
506 | offset += len; | |
507 | size -= len; | |
508 | } | |
509 | } | |
510 | ||
511 | /* | |
512 | * Common write path running under the zvol taskq context. This function | |
513 | * is responsible for copying the request structure data in to the DMU and | |
514 | * signaling the request queue with the result of the copy. | |
515 | */ | |
516 | static void | |
517 | zvol_write(void *arg) | |
518 | { | |
519 | struct request *req = (struct request *)arg; | |
520 | struct request_queue *q = req->q; | |
521 | zvol_state_t *zv = q->queuedata; | |
522 | uint64_t offset = blk_rq_pos(req) << 9; | |
523 | uint64_t size = blk_rq_bytes(req); | |
524 | int error = 0; | |
525 | dmu_tx_t *tx; | |
526 | rl_t *rl; | |
527 | ||
528 | rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_WRITER); | |
529 | ||
530 | tx = dmu_tx_create(zv->zv_objset); | |
531 | dmu_tx_hold_write(tx, ZVOL_OBJ, offset, size); | |
532 | ||
533 | /* This will only fail for ENOSPC */ | |
534 | error = dmu_tx_assign(tx, TXG_WAIT); | |
535 | if (error) { | |
536 | dmu_tx_abort(tx); | |
537 | zfs_range_unlock(rl); | |
538 | blk_end_request(req, -error, size); | |
539 | return; | |
540 | } | |
541 | ||
542 | error = dmu_write_req(zv->zv_objset, ZVOL_OBJ, req, tx); | |
543 | if (error == 0) | |
544 | zvol_log_write(zv, tx, offset, size, rq_is_sync(req)); | |
545 | ||
546 | dmu_tx_commit(tx); | |
547 | zfs_range_unlock(rl); | |
548 | ||
549 | if (rq_is_sync(req)) | |
550 | zil_commit(zv->zv_zilog, ZVOL_OBJ); | |
551 | ||
552 | blk_end_request(req, -error, size); | |
553 | } | |
554 | ||
555 | /* | |
556 | * Common read path running under the zvol taskq context. This function | |
557 | * is responsible for copying the requested data out of the DMU and in to | |
558 | * a linux request structure. It then must signal the request queue with | |
559 | * an error code describing the result of the copy. | |
560 | */ | |
561 | static void | |
562 | zvol_read(void *arg) | |
563 | { | |
564 | struct request *req = (struct request *)arg; | |
565 | struct request_queue *q = req->q; | |
566 | zvol_state_t *zv = q->queuedata; | |
567 | uint64_t offset = blk_rq_pos(req) << 9; | |
568 | uint64_t size = blk_rq_bytes(req); | |
569 | int error; | |
570 | rl_t *rl; | |
571 | ||
572 | rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); | |
573 | ||
574 | error = dmu_read_req(zv->zv_objset, ZVOL_OBJ, req); | |
575 | ||
576 | zfs_range_unlock(rl); | |
577 | ||
578 | /* convert checksum errors into IO errors */ | |
579 | if (error == ECKSUM) | |
580 | error = EIO; | |
581 | ||
582 | blk_end_request(req, -error, size); | |
583 | } | |
584 | ||
585 | /* | |
586 | * Request will be added back to the request queue and retried if | |
587 | * it cannot be immediately dispatched to the taskq for handling | |
588 | */ | |
589 | static inline void | |
590 | zvol_dispatch(task_func_t func, struct request *req) | |
591 | { | |
592 | if (!taskq_dispatch(zvol_taskq, func, (void *)req, TQ_NOSLEEP)) | |
593 | blk_requeue_request(req->q, req); | |
594 | } | |
595 | ||
596 | /* | |
597 | * Common request path. Rather than registering a custom make_request() | |
598 | * function we use the generic Linux version. This is done because it allows | |
599 | * us to easily merge read requests which would otherwise we performed | |
600 | * synchronously by the DMU. This is less critical in write case where the | |
601 | * DMU will perform the correct merging within a transaction group. Using | |
602 | * the generic make_request() also let's use leverage the fact that the | |
603 | * elevator with ensure correct ordering in regards to barrior IOs. On | |
604 | * the downside it means that in the write case we end up doing request | |
605 | * merging twice once in the elevator and once in the DMU. | |
606 | * | |
607 | * The request handler is called under a spin lock so all the real work | |
608 | * is handed off to be done in the context of the zvol taskq. This function | |
609 | * simply performs basic request sanity checking and hands off the request. | |
610 | */ | |
611 | static void | |
612 | zvol_request(struct request_queue *q) | |
613 | { | |
614 | zvol_state_t *zv = q->queuedata; | |
615 | struct request *req; | |
616 | unsigned int size; | |
617 | ||
618 | while ((req = blk_fetch_request(q)) != NULL) { | |
619 | size = blk_rq_bytes(req); | |
620 | ||
621 | if (blk_rq_pos(req) + blk_rq_sectors(req) > | |
622 | get_capacity(zv->zv_disk)) { | |
623 | printk(KERN_INFO | |
624 | "%s: bad access: block=%llu, count=%lu\n", | |
625 | req->rq_disk->disk_name, | |
626 | (long long unsigned)blk_rq_pos(req), | |
627 | (long unsigned)blk_rq_sectors(req)); | |
628 | __blk_end_request(req, -EIO, size); | |
629 | continue; | |
630 | } | |
631 | ||
632 | if (!blk_fs_request(req)) { | |
633 | printk(KERN_INFO "%s: non-fs cmd\n", | |
634 | req->rq_disk->disk_name); | |
635 | __blk_end_request(req, -EIO, size); | |
636 | continue; | |
637 | } | |
638 | ||
639 | switch (rq_data_dir(req)) { | |
640 | case READ: | |
641 | zvol_dispatch(zvol_read, req); | |
642 | break; | |
643 | case WRITE: | |
644 | if (unlikely(get_disk_ro(zv->zv_disk)) || | |
645 | unlikely(zv->zv_flags & ZVOL_RDONLY)) { | |
646 | __blk_end_request(req, -EROFS, size); | |
647 | break; | |
648 | } | |
649 | ||
650 | zvol_dispatch(zvol_write, req); | |
651 | break; | |
652 | default: | |
653 | printk(KERN_INFO "%s: unknown cmd: %d\n", | |
654 | req->rq_disk->disk_name, (int)rq_data_dir(req)); | |
655 | __blk_end_request(req, -EIO, size); | |
656 | break; | |
657 | } | |
658 | } | |
659 | } | |
660 | ||
661 | static void | |
662 | zvol_get_done(zgd_t *zgd, int error) | |
663 | { | |
664 | if (zgd->zgd_db) | |
665 | dmu_buf_rele(zgd->zgd_db, zgd); | |
666 | ||
667 | zfs_range_unlock(zgd->zgd_rl); | |
668 | ||
669 | if (error == 0 && zgd->zgd_bp) | |
670 | zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); | |
671 | ||
672 | kmem_free(zgd, sizeof (zgd_t)); | |
673 | } | |
674 | ||
675 | /* | |
676 | * Get data to generate a TX_WRITE intent log record. | |
677 | */ | |
678 | static int | |
679 | zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) | |
680 | { | |
681 | zvol_state_t *zv = arg; | |
682 | objset_t *os = zv->zv_objset; | |
683 | uint64_t offset = lr->lr_offset; | |
684 | uint64_t size = lr->lr_length; | |
685 | dmu_buf_t *db; | |
686 | zgd_t *zgd; | |
687 | int error; | |
688 | ||
689 | ASSERT(zio != NULL); | |
690 | ASSERT(size != 0); | |
691 | ||
692 | zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); | |
693 | zgd->zgd_zilog = zv->zv_zilog; | |
694 | zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); | |
695 | ||
696 | /* | |
697 | * Write records come in two flavors: immediate and indirect. | |
698 | * For small writes it's cheaper to store the data with the | |
699 | * log record (immediate); for large writes it's cheaper to | |
700 | * sync the data and get a pointer to it (indirect) so that | |
701 | * we don't have to write the data twice. | |
702 | */ | |
703 | if (buf != NULL) { /* immediate write */ | |
704 | error = dmu_read(os, ZVOL_OBJ, offset, size, buf, | |
705 | DMU_READ_NO_PREFETCH); | |
706 | } else { | |
707 | size = zv->zv_volblocksize; | |
708 | offset = P2ALIGN_TYPED(offset, size, uint64_t); | |
709 | error = dmu_buf_hold(os, ZVOL_OBJ, offset, zgd, &db, | |
710 | DMU_READ_NO_PREFETCH); | |
711 | if (error == 0) { | |
712 | zgd->zgd_db = db; | |
713 | zgd->zgd_bp = &lr->lr_blkptr; | |
714 | ||
715 | ASSERT(db != NULL); | |
716 | ASSERT(db->db_offset == offset); | |
717 | ASSERT(db->db_size == size); | |
718 | ||
719 | error = dmu_sync(zio, lr->lr_common.lrc_txg, | |
720 | zvol_get_done, zgd); | |
721 | ||
722 | if (error == 0) | |
723 | return (0); | |
724 | } | |
725 | } | |
726 | ||
727 | zvol_get_done(zgd, error); | |
728 | ||
729 | return (error); | |
730 | } | |
731 | ||
732 | /* | |
733 | * The zvol_state_t's are inserted in increasing MINOR(dev_t) order. | |
734 | */ | |
735 | static void | |
736 | zvol_insert(zvol_state_t *zv_insert) | |
737 | { | |
738 | zvol_state_t *zv = NULL; | |
739 | ||
740 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
741 | ASSERT3U(MINOR(zv_insert->zv_dev) & ZVOL_MINOR_MASK, ==, 0); | |
742 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
743 | zv = list_next(&zvol_state_list, zv)) { | |
744 | if (MINOR(zv->zv_dev) > MINOR(zv_insert->zv_dev)) | |
745 | break; | |
746 | } | |
747 | ||
748 | list_insert_before(&zvol_state_list, zv, zv_insert); | |
749 | } | |
750 | ||
751 | /* | |
752 | * Simply remove the zvol from to list of zvols. | |
753 | */ | |
754 | static void | |
755 | zvol_remove(zvol_state_t *zv_remove) | |
756 | { | |
757 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
758 | list_remove(&zvol_state_list, zv_remove); | |
759 | } | |
760 | ||
761 | static int | |
762 | zvol_first_open(zvol_state_t *zv) | |
763 | { | |
764 | objset_t *os; | |
765 | uint64_t volsize; | |
766 | int error; | |
767 | uint64_t ro; | |
768 | ||
769 | /* lie and say we're read-only */ | |
770 | error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, 1, zvol_tag, &os); | |
771 | if (error) | |
772 | return (-error); | |
773 | ||
774 | error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); | |
775 | if (error) { | |
776 | dmu_objset_disown(os, zvol_tag); | |
777 | return (-error); | |
778 | } | |
779 | ||
780 | zv->zv_objset = os; | |
781 | error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); | |
782 | if (error) { | |
783 | dmu_objset_disown(os, zvol_tag); | |
784 | return (-error); | |
785 | } | |
786 | ||
787 | set_capacity(zv->zv_disk, volsize >> 9); | |
788 | zv->zv_volsize = volsize; | |
789 | zv->zv_zilog = zil_open(os, zvol_get_data); | |
790 | ||
791 | VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL) == 0); | |
792 | if (ro || dmu_objset_is_snapshot(os)) { | |
793 | set_disk_ro(zv->zv_disk, 1); | |
794 | zv->zv_flags |= ZVOL_RDONLY; | |
795 | } else { | |
796 | set_disk_ro(zv->zv_disk, 0); | |
797 | zv->zv_flags &= ~ZVOL_RDONLY; | |
798 | } | |
799 | ||
800 | return (-error); | |
801 | } | |
802 | ||
803 | static void | |
804 | zvol_last_close(zvol_state_t *zv) | |
805 | { | |
806 | zil_close(zv->zv_zilog); | |
807 | zv->zv_zilog = NULL; | |
808 | dmu_buf_rele(zv->zv_dbuf, zvol_tag); | |
809 | zv->zv_dbuf = NULL; | |
810 | dmu_objset_disown(zv->zv_objset, zvol_tag); | |
811 | zv->zv_objset = NULL; | |
812 | } | |
813 | ||
814 | static int | |
815 | zvol_open(struct block_device *bdev, fmode_t flag) | |
816 | { | |
817 | zvol_state_t *zv = bdev->bd_disk->private_data; | |
818 | int error = 0, drop_mutex = 0; | |
819 | ||
820 | /* | |
821 | * If the caller is already holding the mutex do not take it | |
822 | * again, this will happen as part of zvol_create_minor(). | |
823 | * Once add_disk() is called the device is live and the kernel | |
824 | * will attempt to open it to read the partition information. | |
825 | */ | |
826 | if (!mutex_owned(&zvol_state_lock)) { | |
827 | mutex_enter(&zvol_state_lock); | |
828 | drop_mutex = 1; | |
829 | } | |
830 | ||
831 | ASSERT3P(zv, !=, NULL); | |
832 | ||
833 | if (zv->zv_open_count == 0) { | |
834 | error = zvol_first_open(zv); | |
835 | if (error) | |
836 | goto out_mutex; | |
837 | } | |
838 | ||
839 | if ((flag & FMODE_WRITE) && | |
840 | (get_disk_ro(zv->zv_disk) || (zv->zv_flags & ZVOL_RDONLY))) { | |
841 | error = -EROFS; | |
842 | goto out_open_count; | |
843 | } | |
844 | ||
845 | zv->zv_open_count++; | |
846 | ||
847 | out_open_count: | |
848 | if (zv->zv_open_count == 0) | |
849 | zvol_last_close(zv); | |
850 | ||
851 | out_mutex: | |
852 | if (drop_mutex) | |
853 | mutex_exit(&zvol_state_lock); | |
854 | ||
855 | check_disk_change(bdev); | |
856 | ||
857 | return (error); | |
858 | } | |
859 | ||
860 | static int | |
861 | zvol_release(struct gendisk *disk, fmode_t mode) | |
862 | { | |
863 | zvol_state_t *zv = disk->private_data; | |
864 | int drop_mutex = 0; | |
865 | ||
866 | if (!mutex_owned(&zvol_state_lock)) { | |
867 | mutex_enter(&zvol_state_lock); | |
868 | drop_mutex = 1; | |
869 | } | |
870 | ||
871 | ASSERT3P(zv, !=, NULL); | |
872 | ASSERT3U(zv->zv_open_count, >, 0); | |
873 | zv->zv_open_count--; | |
874 | if (zv->zv_open_count == 0) | |
875 | zvol_last_close(zv); | |
876 | ||
877 | if (drop_mutex) | |
878 | mutex_exit(&zvol_state_lock); | |
879 | ||
880 | return (0); | |
881 | } | |
882 | ||
883 | static int | |
884 | zvol_ioctl(struct block_device *bdev, fmode_t mode, | |
885 | unsigned int cmd, unsigned long arg) | |
886 | { | |
887 | zvol_state_t *zv = bdev->bd_disk->private_data; | |
888 | int error = 0; | |
889 | ||
890 | if (zv == NULL) | |
891 | return (-ENXIO); | |
892 | ||
893 | switch (cmd) { | |
894 | case BLKFLSBUF: | |
895 | zil_commit(zv->zv_zilog, ZVOL_OBJ); | |
896 | break; | |
897 | ||
898 | default: | |
899 | error = -ENOTTY; | |
900 | break; | |
901 | ||
902 | } | |
903 | ||
904 | return (error); | |
905 | } | |
906 | ||
907 | #ifdef CONFIG_COMPAT | |
908 | static int | |
909 | zvol_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
910 | unsigned cmd, unsigned long arg) | |
911 | { | |
912 | return zvol_ioctl(bdev, mode, cmd, arg); | |
913 | } | |
914 | #else | |
915 | #define zvol_compat_ioctl NULL | |
916 | #endif | |
917 | ||
918 | static int zvol_media_changed(struct gendisk *disk) | |
919 | { | |
920 | zvol_state_t *zv = disk->private_data; | |
921 | ||
922 | return zv->zv_changed; | |
923 | } | |
924 | ||
925 | static int zvol_revalidate_disk(struct gendisk *disk) | |
926 | { | |
927 | zvol_state_t *zv = disk->private_data; | |
928 | ||
929 | zv->zv_changed = 0; | |
930 | set_capacity(zv->zv_disk, zv->zv_volsize >> 9); | |
931 | ||
932 | return 0; | |
933 | } | |
934 | ||
935 | /* | |
936 | * Provide a simple virtual geometry for legacy compatibility. For devices | |
937 | * smaller than 1 MiB a small head and sector count is used to allow very | |
938 | * tiny devices. For devices over 1 Mib a standard head and sector count | |
939 | * is used to keep the cylinders count reasonable. | |
940 | */ | |
941 | static int | |
942 | zvol_getgeo(struct block_device *bdev, struct hd_geometry *geo) | |
943 | { | |
944 | zvol_state_t *zv = bdev->bd_disk->private_data; | |
945 | sector_t sectors = get_capacity(zv->zv_disk); | |
946 | ||
947 | if (sectors > 2048) { | |
948 | geo->heads = 16; | |
949 | geo->sectors = 63; | |
950 | } else { | |
951 | geo->heads = 2; | |
952 | geo->sectors = 4; | |
953 | } | |
954 | ||
955 | geo->start = 0; | |
956 | geo->cylinders = sectors / (geo->heads * geo->sectors); | |
957 | ||
958 | return 0; | |
959 | } | |
960 | ||
961 | static struct kobject * | |
962 | zvol_probe(dev_t dev, int *part, void *arg) | |
963 | { | |
964 | zvol_state_t *zv; | |
965 | struct kobject *kobj; | |
966 | ||
967 | mutex_enter(&zvol_state_lock); | |
968 | zv = zvol_find_by_dev(dev); | |
969 | kobj = zv ? get_disk(zv->zv_disk) : ERR_PTR(-ENOENT); | |
970 | mutex_exit(&zvol_state_lock); | |
971 | ||
972 | return kobj; | |
973 | } | |
974 | ||
975 | #ifdef HAVE_BDEV_BLOCK_DEVICE_OPERATIONS | |
976 | static struct block_device_operations zvol_ops = { | |
977 | .open = zvol_open, | |
978 | .release = zvol_release, | |
979 | .ioctl = zvol_ioctl, | |
980 | .compat_ioctl = zvol_compat_ioctl, | |
981 | .media_changed = zvol_media_changed, | |
982 | .revalidate_disk = zvol_revalidate_disk, | |
983 | .getgeo = zvol_getgeo, | |
984 | .owner = THIS_MODULE, | |
985 | }; | |
986 | ||
987 | #else /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */ | |
988 | ||
989 | static int | |
990 | zvol_open_by_inode(struct inode *inode, struct file *file) | |
991 | { | |
992 | return zvol_open(inode->i_bdev, file->f_mode); | |
993 | } | |
994 | ||
995 | static int | |
996 | zvol_release_by_inode(struct inode *inode, struct file *file) | |
997 | { | |
998 | return zvol_release(inode->i_bdev->bd_disk, file->f_mode); | |
999 | } | |
1000 | ||
1001 | static int | |
1002 | zvol_ioctl_by_inode(struct inode *inode, struct file *file, | |
1003 | unsigned int cmd, unsigned long arg) | |
1004 | { | |
b1c58213 NB |
1005 | if (file == NULL || inode == NULL) |
1006 | return -EINVAL; | |
60101509 BB |
1007 | return zvol_ioctl(inode->i_bdev, file->f_mode, cmd, arg); |
1008 | } | |
1009 | ||
1010 | # ifdef CONFIG_COMPAT | |
1011 | static long | |
1012 | zvol_compat_ioctl_by_inode(struct file *file, | |
1013 | unsigned int cmd, unsigned long arg) | |
1014 | { | |
b1c58213 NB |
1015 | if (file == NULL) |
1016 | return -EINVAL; | |
60101509 BB |
1017 | return zvol_compat_ioctl(file->f_dentry->d_inode->i_bdev, |
1018 | file->f_mode, cmd, arg); | |
1019 | } | |
1020 | # else | |
1021 | # define zvol_compat_ioctl_by_inode NULL | |
1022 | # endif | |
1023 | ||
1024 | static struct block_device_operations zvol_ops = { | |
1025 | .open = zvol_open_by_inode, | |
1026 | .release = zvol_release_by_inode, | |
1027 | .ioctl = zvol_ioctl_by_inode, | |
1028 | .compat_ioctl = zvol_compat_ioctl_by_inode, | |
1029 | .media_changed = zvol_media_changed, | |
1030 | .revalidate_disk = zvol_revalidate_disk, | |
1031 | .getgeo = zvol_getgeo, | |
1032 | .owner = THIS_MODULE, | |
1033 | }; | |
1034 | #endif /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */ | |
1035 | ||
1036 | /* | |
1037 | * Allocate memory for a new zvol_state_t and setup the required | |
1038 | * request queue and generic disk structures for the block device. | |
1039 | */ | |
1040 | static zvol_state_t * | |
1041 | zvol_alloc(dev_t dev, const char *name) | |
1042 | { | |
1043 | zvol_state_t *zv; | |
1044 | ||
1045 | zv = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); | |
1046 | if (zv == NULL) | |
1047 | goto out; | |
1048 | ||
1049 | zv->zv_queue = blk_init_queue(zvol_request, &zv->zv_lock); | |
1050 | if (zv->zv_queue == NULL) | |
1051 | goto out_kmem; | |
1052 | ||
1053 | zv->zv_disk = alloc_disk(ZVOL_MINORS); | |
1054 | if (zv->zv_disk == NULL) | |
1055 | goto out_queue; | |
1056 | ||
1057 | zv->zv_queue->queuedata = zv; | |
1058 | zv->zv_dev = dev; | |
1059 | zv->zv_open_count = 0; | |
1060 | strlcpy(zv->zv_name, name, DISK_NAME_LEN); | |
1061 | ||
1062 | mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); | |
1063 | avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, | |
1064 | sizeof (rl_t), offsetof(rl_t, r_node)); | |
3c4988c8 BB |
1065 | zv->zv_znode.z_is_zvol = TRUE; |
1066 | ||
60101509 BB |
1067 | spin_lock_init(&zv->zv_lock); |
1068 | list_link_init(&zv->zv_next); | |
1069 | ||
1070 | zv->zv_disk->major = zvol_major; | |
1071 | zv->zv_disk->first_minor = (dev & MINORMASK); | |
1072 | zv->zv_disk->fops = &zvol_ops; | |
1073 | zv->zv_disk->private_data = zv; | |
1074 | zv->zv_disk->queue = zv->zv_queue; | |
1075 | snprintf(zv->zv_disk->disk_name, DISK_NAME_LEN, "%s", name); | |
1076 | ||
1077 | return zv; | |
1078 | ||
1079 | out_queue: | |
1080 | blk_cleanup_queue(zv->zv_queue); | |
1081 | out_kmem: | |
1082 | kmem_free(zv, sizeof (zvol_state_t)); | |
1083 | out: | |
1084 | return NULL; | |
1085 | } | |
1086 | ||
1087 | /* | |
1088 | * Cleanup then free a zvol_state_t which was created by zvol_alloc(). | |
1089 | */ | |
1090 | static void | |
1091 | zvol_free(zvol_state_t *zv) | |
1092 | { | |
1093 | avl_destroy(&zv->zv_znode.z_range_avl); | |
1094 | mutex_destroy(&zv->zv_znode.z_range_lock); | |
1095 | ||
1096 | del_gendisk(zv->zv_disk); | |
1097 | blk_cleanup_queue(zv->zv_queue); | |
1098 | put_disk(zv->zv_disk); | |
1099 | ||
1100 | kmem_free(zv, sizeof (zvol_state_t)); | |
1101 | } | |
1102 | ||
1103 | static int | |
1104 | __zvol_create_minor(const char *name) | |
1105 | { | |
1106 | zvol_state_t *zv; | |
1107 | objset_t *os; | |
1108 | dmu_object_info_t *doi; | |
1109 | uint64_t volsize; | |
1110 | unsigned minor = 0; | |
1111 | int error = 0; | |
1112 | ||
1113 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
1114 | ||
1115 | zv = zvol_find_by_name(name); | |
1116 | if (zv) { | |
1117 | error = EEXIST; | |
1118 | goto out; | |
1119 | } | |
1120 | ||
1121 | doi = kmem_alloc(sizeof(dmu_object_info_t), KM_SLEEP); | |
1122 | ||
1123 | error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, zvol_tag, &os); | |
1124 | if (error) | |
1125 | goto out_doi; | |
1126 | ||
1127 | error = dmu_object_info(os, ZVOL_OBJ, doi); | |
1128 | if (error) | |
1129 | goto out_dmu_objset_disown; | |
1130 | ||
1131 | error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); | |
1132 | if (error) | |
1133 | goto out_dmu_objset_disown; | |
1134 | ||
1135 | error = zvol_find_minor(&minor); | |
1136 | if (error) | |
1137 | goto out_dmu_objset_disown; | |
1138 | ||
1139 | zv = zvol_alloc(MKDEV(zvol_major, minor), name); | |
1140 | if (zv == NULL) { | |
1141 | error = EAGAIN; | |
1142 | goto out_dmu_objset_disown; | |
1143 | } | |
1144 | ||
1145 | if (dmu_objset_is_snapshot(os)) | |
1146 | zv->zv_flags |= ZVOL_RDONLY; | |
1147 | ||
1148 | zv->zv_volblocksize = doi->doi_data_block_size; | |
1149 | zv->zv_volsize = volsize; | |
1150 | zv->zv_objset = os; | |
1151 | ||
1152 | set_capacity(zv->zv_disk, zv->zv_volsize >> 9); | |
1153 | ||
1154 | if (zil_replay_disable) | |
1155 | zil_destroy(dmu_objset_zil(os), B_FALSE); | |
1156 | else | |
1157 | zil_replay(os, zv, zvol_replay_vector); | |
1158 | ||
1159 | out_dmu_objset_disown: | |
1160 | dmu_objset_disown(os, zvol_tag); | |
1161 | zv->zv_objset = NULL; | |
1162 | out_doi: | |
1163 | kmem_free(doi, sizeof(dmu_object_info_t)); | |
1164 | out: | |
1165 | ||
1166 | if (error == 0) { | |
1167 | zvol_insert(zv); | |
1168 | add_disk(zv->zv_disk); | |
1169 | } | |
1170 | ||
1171 | return (error); | |
1172 | } | |
1173 | ||
1174 | /* | |
1175 | * Create a block device minor node and setup the linkage between it | |
1176 | * and the specified volume. Once this function returns the block | |
1177 | * device is live and ready for use. | |
1178 | */ | |
1179 | int | |
1180 | zvol_create_minor(const char *name) | |
1181 | { | |
1182 | int error; | |
1183 | ||
1184 | mutex_enter(&zvol_state_lock); | |
1185 | error = __zvol_create_minor(name); | |
1186 | mutex_exit(&zvol_state_lock); | |
1187 | ||
1188 | return (error); | |
1189 | } | |
1190 | ||
1191 | static int | |
1192 | __zvol_remove_minor(const char *name) | |
1193 | { | |
1194 | zvol_state_t *zv; | |
1195 | ||
1196 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
1197 | ||
1198 | zv = zvol_find_by_name(name); | |
1199 | if (zv == NULL) | |
1200 | return (ENXIO); | |
1201 | ||
1202 | if (zv->zv_open_count > 0) | |
1203 | return (EBUSY); | |
1204 | ||
1205 | zvol_remove(zv); | |
1206 | zvol_free(zv); | |
1207 | ||
1208 | return (0); | |
1209 | } | |
1210 | ||
1211 | /* | |
1212 | * Remove a block device minor node for the specified volume. | |
1213 | */ | |
1214 | int | |
1215 | zvol_remove_minor(const char *name) | |
1216 | { | |
1217 | int error; | |
1218 | ||
1219 | mutex_enter(&zvol_state_lock); | |
1220 | error = __zvol_remove_minor(name); | |
1221 | mutex_exit(&zvol_state_lock); | |
1222 | ||
1223 | return (error); | |
1224 | } | |
1225 | ||
1226 | static int | |
1227 | zvol_create_minors_cb(spa_t *spa, uint64_t dsobj, | |
1228 | const char *dsname, void *arg) | |
1229 | { | |
1230 | if (strchr(dsname, '/') == NULL) | |
1231 | return 0; | |
1232 | ||
1233 | return __zvol_create_minor(dsname); | |
1234 | } | |
1235 | ||
1236 | /* | |
1237 | * Create minors for specified pool, if pool is NULL create minors | |
1238 | * for all available pools. | |
1239 | */ | |
1240 | int | |
1241 | zvol_create_minors(const char *pool) | |
1242 | { | |
1243 | spa_t *spa = NULL; | |
1244 | int error = 0; | |
1245 | ||
1246 | mutex_enter(&zvol_state_lock); | |
1247 | if (pool) { | |
1248 | error = dmu_objset_find_spa(NULL, pool, zvol_create_minors_cb, | |
1249 | NULL, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); | |
1250 | } else { | |
1251 | mutex_enter(&spa_namespace_lock); | |
1252 | while ((spa = spa_next(spa)) != NULL) { | |
1253 | error = dmu_objset_find_spa(NULL, | |
1254 | spa_name(spa), zvol_create_minors_cb, NULL, | |
1255 | DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); | |
1256 | if (error) | |
1257 | break; | |
1258 | } | |
1259 | mutex_exit(&spa_namespace_lock); | |
1260 | } | |
1261 | mutex_exit(&zvol_state_lock); | |
1262 | ||
1263 | return error; | |
1264 | } | |
1265 | ||
1266 | /* | |
1267 | * Remove minors for specified pool, if pool is NULL remove all minors. | |
1268 | */ | |
1269 | void | |
1270 | zvol_remove_minors(const char *pool) | |
1271 | { | |
1272 | zvol_state_t *zv, *zv_next; | |
1273 | char *str; | |
1274 | ||
1275 | str = kmem_zalloc(DISK_NAME_LEN, KM_SLEEP); | |
1276 | if (pool) { | |
1277 | (void) strncpy(str, pool, strlen(pool)); | |
1278 | (void) strcat(str, "/"); | |
1279 | } | |
1280 | ||
1281 | mutex_enter(&zvol_state_lock); | |
1282 | for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { | |
1283 | zv_next = list_next(&zvol_state_list, zv); | |
1284 | ||
1285 | if (pool == NULL || !strncmp(str, zv->zv_name, strlen(str))) { | |
1286 | zvol_remove(zv); | |
1287 | zvol_free(zv); | |
1288 | } | |
1289 | } | |
1290 | mutex_exit(&zvol_state_lock); | |
1291 | kmem_free(str, DISK_NAME_LEN); | |
1292 | } | |
1293 | ||
1294 | int | |
1295 | zvol_init(void) | |
1296 | { | |
1297 | int error; | |
1298 | ||
1299 | if (!zvol_threads) | |
1300 | zvol_threads = num_online_cpus(); | |
1301 | ||
1302 | zvol_taskq = taskq_create(ZVOL_DRIVER, zvol_threads, maxclsyspri, | |
1303 | zvol_threads, INT_MAX, TASKQ_PREPOPULATE); | |
1304 | if (zvol_taskq == NULL) { | |
1305 | printk(KERN_INFO "ZFS: taskq_create() failed\n"); | |
1306 | return (-ENOMEM); | |
1307 | } | |
1308 | ||
1309 | error = register_blkdev(zvol_major, ZVOL_DRIVER); | |
1310 | if (error) { | |
1311 | printk(KERN_INFO "ZFS: register_blkdev() failed %d\n", error); | |
1312 | taskq_destroy(zvol_taskq); | |
1313 | return (error); | |
1314 | } | |
1315 | ||
1316 | blk_register_region(MKDEV(zvol_major, 0), 1UL << MINORBITS, | |
1317 | THIS_MODULE, zvol_probe, NULL, NULL); | |
1318 | ||
1319 | mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL); | |
1320 | list_create(&zvol_state_list, sizeof (zvol_state_t), | |
1321 | offsetof(zvol_state_t, zv_next)); | |
1322 | ||
1323 | (void) zvol_create_minors(NULL); | |
1324 | ||
1325 | return (0); | |
1326 | } | |
1327 | ||
1328 | void | |
1329 | zvol_fini(void) | |
1330 | { | |
1331 | zvol_remove_minors(NULL); | |
1332 | blk_unregister_region(MKDEV(zvol_major, 0), 1UL << MINORBITS); | |
1333 | unregister_blkdev(zvol_major, ZVOL_DRIVER); | |
1334 | taskq_destroy(zvol_taskq); | |
1335 | mutex_destroy(&zvol_state_lock); | |
1336 | list_destroy(&zvol_state_list); | |
1337 | } | |
1338 | ||
1339 | module_param(zvol_major, uint, 0); | |
1340 | MODULE_PARM_DESC(zvol_major, "Major number for zvol device"); | |
1341 | ||
1342 | module_param(zvol_threads, uint, 0); | |
1343 | MODULE_PARM_DESC(zvol_threads, "Number of threads for zvol device"); |