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