]>
Commit | Line | Data |
---|---|---|
5df7e9d8 MM |
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 | */ | |
0929c4de MA |
21 | /* |
22 | * Copyright (c) 2012, 2020 by Delphix. All rights reserved. | |
23 | */ | |
5df7e9d8 MM |
24 | |
25 | #include <sys/dataset_kstats.h> | |
26 | #include <sys/dbuf.h> | |
27 | #include <sys/dmu_traverse.h> | |
28 | #include <sys/dsl_dataset.h> | |
29 | #include <sys/dsl_prop.h> | |
30 | #include <sys/dsl_dir.h> | |
31 | #include <sys/zap.h> | |
32 | #include <sys/zfeature.h> | |
33 | #include <sys/zil_impl.h> | |
34 | #include <sys/dmu_tx.h> | |
35 | #include <sys/zio.h> | |
36 | #include <sys/zfs_rlock.h> | |
37 | #include <sys/spa_impl.h> | |
38 | #include <sys/zvol.h> | |
39 | #include <sys/zvol_impl.h> | |
40 | ||
41 | #include <linux/blkdev_compat.h> | |
42 | #include <linux/task_io_accounting_ops.h> | |
43 | ||
44 | unsigned int zvol_major = ZVOL_MAJOR; | |
45 | unsigned int zvol_request_sync = 0; | |
46 | unsigned int zvol_prefetch_bytes = (128 * 1024); | |
47 | unsigned long zvol_max_discard_blocks = 16384; | |
48 | unsigned int zvol_threads = 32; | |
49 | ||
50 | struct zvol_state_os { | |
51 | struct gendisk *zvo_disk; /* generic disk */ | |
52 | struct request_queue *zvo_queue; /* request queue */ | |
5df7e9d8 MM |
53 | dev_t zvo_dev; /* device id */ |
54 | }; | |
55 | ||
56 | taskq_t *zvol_taskq; | |
57 | static struct ida zvol_ida; | |
58 | ||
59 | typedef struct zv_request { | |
60 | zvol_state_t *zv; | |
61 | struct bio *bio; | |
0929c4de | 62 | taskq_ent_t ent; |
5df7e9d8 MM |
63 | } zv_request_t; |
64 | ||
65 | /* | |
66 | * Given a path, return TRUE if path is a ZVOL. | |
67 | */ | |
68 | static boolean_t | |
b7281c88 | 69 | zvol_is_zvol_impl(const char *path) |
5df7e9d8 | 70 | { |
b7281c88 | 71 | dev_t dev = 0; |
5df7e9d8 | 72 | |
b7281c88 | 73 | if (vdev_lookup_bdev(path, &dev) != 0) |
5df7e9d8 MM |
74 | return (B_FALSE); |
75 | ||
b7281c88 | 76 | if (MAJOR(dev) == zvol_major) |
5df7e9d8 MM |
77 | return (B_TRUE); |
78 | ||
79 | return (B_FALSE); | |
80 | } | |
81 | ||
5df7e9d8 MM |
82 | static void |
83 | zvol_write(void *arg) | |
84 | { | |
5df7e9d8 MM |
85 | zv_request_t *zvr = arg; |
86 | struct bio *bio = zvr->bio; | |
1c2358c1 BB |
87 | int error = 0; |
88 | uio_t uio; | |
89 | ||
90 | uio_bvec_init(&uio, bio); | |
5df7e9d8 MM |
91 | |
92 | zvol_state_t *zv = zvr->zv; | |
0b32d817 RM |
93 | ASSERT3P(zv, !=, NULL); |
94 | ASSERT3U(zv->zv_open_count, >, 0); | |
95 | ASSERT3P(zv->zv_zilog, !=, NULL); | |
5df7e9d8 | 96 | |
0929c4de MA |
97 | /* bio marked as FLUSH need to flush before write */ |
98 | if (bio_is_flush(bio)) | |
99 | zil_commit(zv->zv_zilog, ZVOL_OBJ); | |
100 | ||
101 | /* Some requests are just for flush and nothing else. */ | |
102 | if (uio.uio_resid == 0) { | |
103 | rw_exit(&zv->zv_suspend_lock); | |
104 | BIO_END_IO(bio, 0); | |
105 | kmem_free(zvr, sizeof (zv_request_t)); | |
106 | return; | |
107 | } | |
108 | ||
5df7e9d8 MM |
109 | ssize_t start_resid = uio.uio_resid; |
110 | unsigned long start_jif = jiffies; | |
111 | blk_generic_start_io_acct(zv->zv_zso->zvo_queue, WRITE, | |
112 | bio_sectors(bio), &zv->zv_zso->zvo_disk->part0); | |
113 | ||
114 | boolean_t sync = | |
115 | bio_is_fua(bio) || zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; | |
116 | ||
0929c4de MA |
117 | zfs_locked_range_t *lr = zfs_rangelock_enter(&zv->zv_rangelock, |
118 | uio.uio_loffset, uio.uio_resid, RL_WRITER); | |
119 | ||
5df7e9d8 MM |
120 | uint64_t volsize = zv->zv_volsize; |
121 | while (uio.uio_resid > 0 && uio.uio_loffset < volsize) { | |
122 | uint64_t bytes = MIN(uio.uio_resid, DMU_MAX_ACCESS >> 1); | |
123 | uint64_t off = uio.uio_loffset; | |
124 | dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); | |
125 | ||
126 | if (bytes > volsize - off) /* don't write past the end */ | |
127 | bytes = volsize - off; | |
128 | ||
20f28785 | 129 | dmu_tx_hold_write_by_dnode(tx, zv->zv_dn, off, bytes); |
5df7e9d8 MM |
130 | |
131 | /* This will only fail for ENOSPC */ | |
132 | error = dmu_tx_assign(tx, TXG_WAIT); | |
133 | if (error) { | |
134 | dmu_tx_abort(tx); | |
135 | break; | |
136 | } | |
137 | error = dmu_write_uio_dnode(zv->zv_dn, &uio, bytes, tx); | |
138 | if (error == 0) { | |
139 | zvol_log_write(zv, tx, off, bytes, sync); | |
140 | } | |
141 | dmu_tx_commit(tx); | |
142 | ||
143 | if (error) | |
144 | break; | |
145 | } | |
0929c4de | 146 | zfs_rangelock_exit(lr); |
5df7e9d8 MM |
147 | |
148 | int64_t nwritten = start_resid - uio.uio_resid; | |
4547fc4e | 149 | dataset_kstats_update_write_kstats(&zv->zv_kstat, nwritten); |
5df7e9d8 MM |
150 | task_io_account_write(nwritten); |
151 | ||
152 | if (sync) | |
153 | zil_commit(zv->zv_zilog, ZVOL_OBJ); | |
154 | ||
155 | rw_exit(&zv->zv_suspend_lock); | |
156 | blk_generic_end_io_acct(zv->zv_zso->zvo_queue, | |
157 | WRITE, &zv->zv_zso->zvo_disk->part0, start_jif); | |
158 | BIO_END_IO(bio, -error); | |
159 | kmem_free(zvr, sizeof (zv_request_t)); | |
160 | } | |
161 | ||
162 | static void | |
163 | zvol_discard(void *arg) | |
164 | { | |
165 | zv_request_t *zvr = arg; | |
166 | struct bio *bio = zvr->bio; | |
167 | zvol_state_t *zv = zvr->zv; | |
168 | uint64_t start = BIO_BI_SECTOR(bio) << 9; | |
169 | uint64_t size = BIO_BI_SIZE(bio); | |
170 | uint64_t end = start + size; | |
171 | boolean_t sync; | |
172 | int error = 0; | |
173 | dmu_tx_t *tx; | |
174 | unsigned long start_jif; | |
175 | ||
0b32d817 RM |
176 | ASSERT3P(zv, !=, NULL); |
177 | ASSERT3U(zv->zv_open_count, >, 0); | |
178 | ASSERT3P(zv->zv_zilog, !=, NULL); | |
5df7e9d8 MM |
179 | |
180 | start_jif = jiffies; | |
181 | blk_generic_start_io_acct(zv->zv_zso->zvo_queue, WRITE, | |
182 | bio_sectors(bio), &zv->zv_zso->zvo_disk->part0); | |
183 | ||
184 | sync = bio_is_fua(bio) || zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; | |
185 | ||
186 | if (end > zv->zv_volsize) { | |
187 | error = SET_ERROR(EIO); | |
188 | goto unlock; | |
189 | } | |
190 | ||
191 | /* | |
192 | * Align the request to volume block boundaries when a secure erase is | |
193 | * not required. This will prevent dnode_free_range() from zeroing out | |
194 | * the unaligned parts which is slow (read-modify-write) and useless | |
195 | * since we are not freeing any space by doing so. | |
196 | */ | |
197 | if (!bio_is_secure_erase(bio)) { | |
198 | start = P2ROUNDUP(start, zv->zv_volblocksize); | |
199 | end = P2ALIGN(end, zv->zv_volblocksize); | |
200 | size = end - start; | |
201 | } | |
202 | ||
203 | if (start >= end) | |
204 | goto unlock; | |
205 | ||
0929c4de MA |
206 | zfs_locked_range_t *lr = zfs_rangelock_enter(&zv->zv_rangelock, |
207 | start, size, RL_WRITER); | |
208 | ||
5df7e9d8 MM |
209 | tx = dmu_tx_create(zv->zv_objset); |
210 | dmu_tx_mark_netfree(tx); | |
211 | error = dmu_tx_assign(tx, TXG_WAIT); | |
212 | if (error != 0) { | |
213 | dmu_tx_abort(tx); | |
214 | } else { | |
215 | zvol_log_truncate(zv, tx, start, size, B_TRUE); | |
216 | dmu_tx_commit(tx); | |
217 | error = dmu_free_long_range(zv->zv_objset, | |
218 | ZVOL_OBJ, start, size); | |
219 | } | |
0929c4de | 220 | zfs_rangelock_exit(lr); |
5df7e9d8 MM |
221 | |
222 | if (error == 0 && sync) | |
223 | zil_commit(zv->zv_zilog, ZVOL_OBJ); | |
224 | ||
0929c4de | 225 | unlock: |
5df7e9d8 MM |
226 | rw_exit(&zv->zv_suspend_lock); |
227 | blk_generic_end_io_acct(zv->zv_zso->zvo_queue, WRITE, | |
228 | &zv->zv_zso->zvo_disk->part0, start_jif); | |
229 | BIO_END_IO(bio, -error); | |
230 | kmem_free(zvr, sizeof (zv_request_t)); | |
231 | } | |
232 | ||
233 | static void | |
234 | zvol_read(void *arg) | |
235 | { | |
5df7e9d8 MM |
236 | zv_request_t *zvr = arg; |
237 | struct bio *bio = zvr->bio; | |
1c2358c1 BB |
238 | int error = 0; |
239 | uio_t uio; | |
240 | ||
241 | uio_bvec_init(&uio, bio); | |
5df7e9d8 MM |
242 | |
243 | zvol_state_t *zv = zvr->zv; | |
0b32d817 RM |
244 | ASSERT3P(zv, !=, NULL); |
245 | ASSERT3U(zv->zv_open_count, >, 0); | |
5df7e9d8 MM |
246 | |
247 | ssize_t start_resid = uio.uio_resid; | |
248 | unsigned long start_jif = jiffies; | |
249 | blk_generic_start_io_acct(zv->zv_zso->zvo_queue, READ, bio_sectors(bio), | |
250 | &zv->zv_zso->zvo_disk->part0); | |
251 | ||
0929c4de MA |
252 | zfs_locked_range_t *lr = zfs_rangelock_enter(&zv->zv_rangelock, |
253 | uio.uio_loffset, uio.uio_resid, RL_READER); | |
254 | ||
5df7e9d8 MM |
255 | uint64_t volsize = zv->zv_volsize; |
256 | while (uio.uio_resid > 0 && uio.uio_loffset < volsize) { | |
257 | uint64_t bytes = MIN(uio.uio_resid, DMU_MAX_ACCESS >> 1); | |
258 | ||
259 | /* don't read past the end */ | |
260 | if (bytes > volsize - uio.uio_loffset) | |
261 | bytes = volsize - uio.uio_loffset; | |
262 | ||
263 | error = dmu_read_uio_dnode(zv->zv_dn, &uio, bytes); | |
264 | if (error) { | |
265 | /* convert checksum errors into IO errors */ | |
266 | if (error == ECKSUM) | |
267 | error = SET_ERROR(EIO); | |
268 | break; | |
269 | } | |
270 | } | |
0929c4de | 271 | zfs_rangelock_exit(lr); |
5df7e9d8 MM |
272 | |
273 | int64_t nread = start_resid - uio.uio_resid; | |
4547fc4e | 274 | dataset_kstats_update_read_kstats(&zv->zv_kstat, nread); |
5df7e9d8 MM |
275 | task_io_account_read(nread); |
276 | ||
277 | rw_exit(&zv->zv_suspend_lock); | |
278 | blk_generic_end_io_acct(zv->zv_zso->zvo_queue, READ, | |
279 | &zv->zv_zso->zvo_disk->part0, start_jif); | |
280 | BIO_END_IO(bio, -error); | |
281 | kmem_free(zvr, sizeof (zv_request_t)); | |
282 | } | |
283 | ||
d817c171 CK |
284 | #ifdef HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS |
285 | static blk_qc_t | |
286 | zvol_submit_bio(struct bio *bio) | |
287 | #else | |
5df7e9d8 MM |
288 | static MAKE_REQUEST_FN_RET |
289 | zvol_request(struct request_queue *q, struct bio *bio) | |
d817c171 | 290 | #endif |
5df7e9d8 | 291 | { |
d817c171 CK |
292 | #ifdef HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS |
293 | struct request_queue *q = bio->bi_disk->queue; | |
294 | #endif | |
5df7e9d8 MM |
295 | zvol_state_t *zv = q->queuedata; |
296 | fstrans_cookie_t cookie = spl_fstrans_mark(); | |
297 | uint64_t offset = BIO_BI_SECTOR(bio) << 9; | |
298 | uint64_t size = BIO_BI_SIZE(bio); | |
299 | int rw = bio_data_dir(bio); | |
300 | zv_request_t *zvr; | |
301 | ||
302 | if (bio_has_data(bio) && offset + size > zv->zv_volsize) { | |
303 | printk(KERN_INFO | |
304 | "%s: bad access: offset=%llu, size=%lu\n", | |
305 | zv->zv_zso->zvo_disk->disk_name, | |
306 | (long long unsigned)offset, | |
307 | (long unsigned)size); | |
308 | ||
309 | BIO_END_IO(bio, -SET_ERROR(EIO)); | |
310 | goto out; | |
311 | } | |
312 | ||
313 | if (rw == WRITE) { | |
5df7e9d8 MM |
314 | if (unlikely(zv->zv_flags & ZVOL_RDONLY)) { |
315 | BIO_END_IO(bio, -SET_ERROR(EROFS)); | |
316 | goto out; | |
317 | } | |
318 | ||
319 | /* | |
0929c4de MA |
320 | * Prevents the zvol from being suspended, or the ZIL being |
321 | * concurrently opened. Will be released after the i/o | |
322 | * completes. | |
5df7e9d8 MM |
323 | */ |
324 | rw_enter(&zv->zv_suspend_lock, RW_READER); | |
325 | ||
326 | /* | |
327 | * Open a ZIL if this is the first time we have written to this | |
328 | * zvol. We protect zv->zv_zilog with zv_suspend_lock rather | |
329 | * than zv_state_lock so that we don't need to acquire an | |
330 | * additional lock in this path. | |
331 | */ | |
332 | if (zv->zv_zilog == NULL) { | |
333 | rw_exit(&zv->zv_suspend_lock); | |
334 | rw_enter(&zv->zv_suspend_lock, RW_WRITER); | |
335 | if (zv->zv_zilog == NULL) { | |
336 | zv->zv_zilog = zil_open(zv->zv_objset, | |
337 | zvol_get_data); | |
338 | zv->zv_flags |= ZVOL_WRITTEN_TO; | |
339 | } | |
340 | rw_downgrade(&zv->zv_suspend_lock); | |
341 | } | |
342 | ||
5df7e9d8 MM |
343 | zvr = kmem_alloc(sizeof (zv_request_t), KM_SLEEP); |
344 | zvr->zv = zv; | |
345 | zvr->bio = bio; | |
0929c4de | 346 | taskq_init_ent(&zvr->ent); |
5df7e9d8 MM |
347 | |
348 | /* | |
0929c4de MA |
349 | * We don't want this thread to be blocked waiting for i/o to |
350 | * complete, so we instead wait from a taskq callback. The | |
351 | * i/o may be a ZIL write (via zil_commit()), or a read of an | |
352 | * indirect block, or a read of a data block (if this is a | |
353 | * partial-block write). We will indicate that the i/o is | |
354 | * complete by calling BIO_END_IO() from the taskq callback. | |
355 | * | |
356 | * This design allows the calling thread to continue and | |
357 | * initiate more concurrent operations by calling | |
358 | * zvol_request() again. There are typically only a small | |
359 | * number of threads available to call zvol_request() (e.g. | |
360 | * one per iSCSI target), so keeping the latency of | |
361 | * zvol_request() low is important for performance. | |
362 | * | |
363 | * The zvol_request_sync module parameter allows this | |
364 | * behavior to be altered, for performance evaluation | |
365 | * purposes. If the callback blocks, setting | |
366 | * zvol_request_sync=1 will result in much worse performance. | |
367 | * | |
368 | * We can have up to zvol_threads concurrent i/o's being | |
369 | * processed for all zvols on the system. This is typically | |
370 | * a vast improvement over the zvol_request_sync=1 behavior | |
371 | * of one i/o at a time per zvol. However, an even better | |
372 | * design would be for zvol_request() to initiate the zio | |
373 | * directly, and then be notified by the zio_done callback, | |
374 | * which would call BIO_END_IO(). Unfortunately, the DMU/ZIL | |
375 | * interfaces lack this functionality (they block waiting for | |
376 | * the i/o to complete). | |
5df7e9d8 | 377 | */ |
5df7e9d8 | 378 | if (bio_is_discard(bio) || bio_is_secure_erase(bio)) { |
0929c4de | 379 | if (zvol_request_sync) { |
5df7e9d8 | 380 | zvol_discard(zvr); |
0929c4de MA |
381 | } else { |
382 | taskq_dispatch_ent(zvol_taskq, | |
383 | zvol_discard, zvr, 0, &zvr->ent); | |
384 | } | |
5df7e9d8 | 385 | } else { |
0929c4de | 386 | if (zvol_request_sync) { |
5df7e9d8 | 387 | zvol_write(zvr); |
0929c4de MA |
388 | } else { |
389 | taskq_dispatch_ent(zvol_taskq, | |
390 | zvol_write, zvr, 0, &zvr->ent); | |
391 | } | |
5df7e9d8 MM |
392 | } |
393 | } else { | |
394 | /* | |
395 | * The SCST driver, and possibly others, may issue READ I/Os | |
396 | * with a length of zero bytes. These empty I/Os contain no | |
397 | * data and require no additional handling. | |
398 | */ | |
399 | if (size == 0) { | |
400 | BIO_END_IO(bio, 0); | |
401 | goto out; | |
402 | } | |
403 | ||
404 | zvr = kmem_alloc(sizeof (zv_request_t), KM_SLEEP); | |
405 | zvr->zv = zv; | |
406 | zvr->bio = bio; | |
0929c4de | 407 | taskq_init_ent(&zvr->ent); |
5df7e9d8 MM |
408 | |
409 | rw_enter(&zv->zv_suspend_lock, RW_READER); | |
410 | ||
0929c4de MA |
411 | /* See comment in WRITE case above. */ |
412 | if (zvol_request_sync) { | |
5df7e9d8 | 413 | zvol_read(zvr); |
0929c4de MA |
414 | } else { |
415 | taskq_dispatch_ent(zvol_taskq, | |
416 | zvol_read, zvr, 0, &zvr->ent); | |
417 | } | |
5df7e9d8 MM |
418 | } |
419 | ||
420 | out: | |
421 | spl_fstrans_unmark(cookie); | |
d817c171 CK |
422 | #if defined(HAVE_MAKE_REQUEST_FN_RET_QC) || \ |
423 | defined(HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS) | |
5df7e9d8 MM |
424 | return (BLK_QC_T_NONE); |
425 | #endif | |
426 | } | |
427 | ||
428 | static int | |
429 | zvol_open(struct block_device *bdev, fmode_t flag) | |
430 | { | |
431 | zvol_state_t *zv; | |
432 | int error = 0; | |
433 | boolean_t drop_suspend = B_TRUE; | |
434 | ||
435 | rw_enter(&zvol_state_lock, RW_READER); | |
436 | /* | |
437 | * Obtain a copy of private_data under the zvol_state_lock to make | |
438 | * sure that either the result of zvol free code path setting | |
439 | * bdev->bd_disk->private_data to NULL is observed, or zvol_free() | |
440 | * is not called on this zv because of the positive zv_open_count. | |
441 | */ | |
442 | zv = bdev->bd_disk->private_data; | |
443 | if (zv == NULL) { | |
444 | rw_exit(&zvol_state_lock); | |
445 | return (SET_ERROR(-ENXIO)); | |
446 | } | |
447 | ||
448 | mutex_enter(&zv->zv_state_lock); | |
449 | /* | |
450 | * make sure zvol is not suspended during first open | |
451 | * (hold zv_suspend_lock) and respect proper lock acquisition | |
452 | * ordering - zv_suspend_lock before zv_state_lock | |
453 | */ | |
454 | if (zv->zv_open_count == 0) { | |
455 | if (!rw_tryenter(&zv->zv_suspend_lock, RW_READER)) { | |
456 | mutex_exit(&zv->zv_state_lock); | |
457 | rw_enter(&zv->zv_suspend_lock, RW_READER); | |
458 | mutex_enter(&zv->zv_state_lock); | |
459 | /* check to see if zv_suspend_lock is needed */ | |
460 | if (zv->zv_open_count != 0) { | |
461 | rw_exit(&zv->zv_suspend_lock); | |
462 | drop_suspend = B_FALSE; | |
463 | } | |
464 | } | |
465 | } else { | |
466 | drop_suspend = B_FALSE; | |
467 | } | |
468 | rw_exit(&zvol_state_lock); | |
469 | ||
470 | ASSERT(MUTEX_HELD(&zv->zv_state_lock)); | |
5df7e9d8 MM |
471 | |
472 | if (zv->zv_open_count == 0) { | |
0b32d817 | 473 | ASSERT(RW_READ_HELD(&zv->zv_suspend_lock)); |
5df7e9d8 MM |
474 | error = -zvol_first_open(zv, !(flag & FMODE_WRITE)); |
475 | if (error) | |
476 | goto out_mutex; | |
477 | } | |
478 | ||
479 | if ((flag & FMODE_WRITE) && (zv->zv_flags & ZVOL_RDONLY)) { | |
480 | error = -EROFS; | |
481 | goto out_open_count; | |
482 | } | |
483 | ||
484 | zv->zv_open_count++; | |
485 | ||
486 | mutex_exit(&zv->zv_state_lock); | |
487 | if (drop_suspend) | |
488 | rw_exit(&zv->zv_suspend_lock); | |
489 | ||
ae15f1c1 | 490 | zfs_check_media_change(bdev); |
5df7e9d8 MM |
491 | |
492 | return (0); | |
493 | ||
494 | out_open_count: | |
495 | if (zv->zv_open_count == 0) | |
496 | zvol_last_close(zv); | |
497 | ||
498 | out_mutex: | |
499 | mutex_exit(&zv->zv_state_lock); | |
500 | if (drop_suspend) | |
501 | rw_exit(&zv->zv_suspend_lock); | |
502 | if (error == -EINTR) { | |
503 | error = -ERESTARTSYS; | |
504 | schedule(); | |
505 | } | |
506 | return (SET_ERROR(error)); | |
507 | } | |
508 | ||
5df7e9d8 | 509 | static void |
5df7e9d8 MM |
510 | zvol_release(struct gendisk *disk, fmode_t mode) |
511 | { | |
512 | zvol_state_t *zv; | |
513 | boolean_t drop_suspend = B_TRUE; | |
514 | ||
515 | rw_enter(&zvol_state_lock, RW_READER); | |
516 | zv = disk->private_data; | |
517 | ||
518 | mutex_enter(&zv->zv_state_lock); | |
0b32d817 | 519 | ASSERT3U(zv->zv_open_count, >, 0); |
5df7e9d8 MM |
520 | /* |
521 | * make sure zvol is not suspended during last close | |
522 | * (hold zv_suspend_lock) and respect proper lock acquisition | |
523 | * ordering - zv_suspend_lock before zv_state_lock | |
524 | */ | |
525 | if (zv->zv_open_count == 1) { | |
526 | if (!rw_tryenter(&zv->zv_suspend_lock, RW_READER)) { | |
527 | mutex_exit(&zv->zv_state_lock); | |
528 | rw_enter(&zv->zv_suspend_lock, RW_READER); | |
529 | mutex_enter(&zv->zv_state_lock); | |
530 | /* check to see if zv_suspend_lock is needed */ | |
531 | if (zv->zv_open_count != 1) { | |
532 | rw_exit(&zv->zv_suspend_lock); | |
533 | drop_suspend = B_FALSE; | |
534 | } | |
535 | } | |
536 | } else { | |
537 | drop_suspend = B_FALSE; | |
538 | } | |
539 | rw_exit(&zvol_state_lock); | |
540 | ||
541 | ASSERT(MUTEX_HELD(&zv->zv_state_lock)); | |
5df7e9d8 MM |
542 | |
543 | zv->zv_open_count--; | |
0b32d817 RM |
544 | if (zv->zv_open_count == 0) { |
545 | ASSERT(RW_READ_HELD(&zv->zv_suspend_lock)); | |
5df7e9d8 | 546 | zvol_last_close(zv); |
0b32d817 | 547 | } |
5df7e9d8 MM |
548 | |
549 | mutex_exit(&zv->zv_state_lock); | |
550 | ||
551 | if (drop_suspend) | |
552 | rw_exit(&zv->zv_suspend_lock); | |
5df7e9d8 MM |
553 | } |
554 | ||
555 | static int | |
556 | zvol_ioctl(struct block_device *bdev, fmode_t mode, | |
557 | unsigned int cmd, unsigned long arg) | |
558 | { | |
559 | zvol_state_t *zv = bdev->bd_disk->private_data; | |
560 | int error = 0; | |
561 | ||
562 | ASSERT3U(zv->zv_open_count, >, 0); | |
563 | ||
564 | switch (cmd) { | |
565 | case BLKFLSBUF: | |
566 | fsync_bdev(bdev); | |
567 | invalidate_bdev(bdev); | |
568 | rw_enter(&zv->zv_suspend_lock, RW_READER); | |
569 | ||
570 | if (!(zv->zv_flags & ZVOL_RDONLY)) | |
571 | txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); | |
572 | ||
573 | rw_exit(&zv->zv_suspend_lock); | |
574 | break; | |
575 | ||
576 | case BLKZNAME: | |
577 | mutex_enter(&zv->zv_state_lock); | |
578 | error = copy_to_user((void *)arg, zv->zv_name, MAXNAMELEN); | |
579 | mutex_exit(&zv->zv_state_lock); | |
580 | break; | |
581 | ||
582 | default: | |
583 | error = -ENOTTY; | |
584 | break; | |
585 | } | |
586 | ||
587 | return (SET_ERROR(error)); | |
588 | } | |
589 | ||
590 | #ifdef CONFIG_COMPAT | |
591 | static int | |
592 | zvol_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
593 | unsigned cmd, unsigned long arg) | |
594 | { | |
595 | return (zvol_ioctl(bdev, mode, cmd, arg)); | |
596 | } | |
597 | #else | |
598 | #define zvol_compat_ioctl NULL | |
599 | #endif | |
600 | ||
5df7e9d8 MM |
601 | static unsigned int |
602 | zvol_check_events(struct gendisk *disk, unsigned int clearing) | |
603 | { | |
604 | unsigned int mask = 0; | |
605 | ||
606 | rw_enter(&zvol_state_lock, RW_READER); | |
607 | ||
608 | zvol_state_t *zv = disk->private_data; | |
609 | if (zv != NULL) { | |
610 | mutex_enter(&zv->zv_state_lock); | |
611 | mask = zv->zv_changed ? DISK_EVENT_MEDIA_CHANGE : 0; | |
612 | zv->zv_changed = 0; | |
613 | mutex_exit(&zv->zv_state_lock); | |
614 | } | |
615 | ||
616 | rw_exit(&zvol_state_lock); | |
617 | ||
618 | return (mask); | |
619 | } | |
5df7e9d8 MM |
620 | |
621 | static int | |
622 | zvol_revalidate_disk(struct gendisk *disk) | |
623 | { | |
624 | rw_enter(&zvol_state_lock, RW_READER); | |
625 | ||
626 | zvol_state_t *zv = disk->private_data; | |
627 | if (zv != NULL) { | |
628 | mutex_enter(&zv->zv_state_lock); | |
629 | set_capacity(zv->zv_zso->zvo_disk, | |
630 | zv->zv_volsize >> SECTOR_BITS); | |
631 | mutex_exit(&zv->zv_state_lock); | |
632 | } | |
633 | ||
634 | rw_exit(&zvol_state_lock); | |
635 | ||
636 | return (0); | |
637 | } | |
638 | ||
65c7cc49 | 639 | static int |
5df7e9d8 MM |
640 | zvol_update_volsize(zvol_state_t *zv, uint64_t volsize) |
641 | { | |
1c0bbd52 | 642 | struct gendisk *disk = zv->zv_zso->zvo_disk; |
5df7e9d8 | 643 | |
59b68723 | 644 | #ifdef HAVE_REVALIDATE_DISK_SIZE |
1c0bbd52 | 645 | revalidate_disk_size(disk, zvol_revalidate_disk(disk) == 0); |
59b68723 | 646 | #else |
1c0bbd52 | 647 | revalidate_disk(disk); |
59b68723 | 648 | #endif |
5df7e9d8 MM |
649 | return (0); |
650 | } | |
651 | ||
652 | static void | |
653 | zvol_clear_private(zvol_state_t *zv) | |
654 | { | |
655 | /* | |
656 | * Cleared while holding zvol_state_lock as a writer | |
657 | * which will prevent zvol_open() from opening it. | |
658 | */ | |
659 | zv->zv_zso->zvo_disk->private_data = NULL; | |
660 | } | |
661 | ||
662 | /* | |
663 | * Provide a simple virtual geometry for legacy compatibility. For devices | |
664 | * smaller than 1 MiB a small head and sector count is used to allow very | |
665 | * tiny devices. For devices over 1 Mib a standard head and sector count | |
666 | * is used to keep the cylinders count reasonable. | |
667 | */ | |
668 | static int | |
669 | zvol_getgeo(struct block_device *bdev, struct hd_geometry *geo) | |
670 | { | |
671 | zvol_state_t *zv = bdev->bd_disk->private_data; | |
672 | sector_t sectors; | |
673 | ||
674 | ASSERT3U(zv->zv_open_count, >, 0); | |
675 | ||
676 | sectors = get_capacity(zv->zv_zso->zvo_disk); | |
677 | ||
678 | if (sectors > 2048) { | |
679 | geo->heads = 16; | |
680 | geo->sectors = 63; | |
681 | } else { | |
682 | geo->heads = 2; | |
683 | geo->sectors = 4; | |
684 | } | |
685 | ||
686 | geo->start = 0; | |
687 | geo->cylinders = sectors / (geo->heads * geo->sectors); | |
688 | ||
689 | return (0); | |
690 | } | |
691 | ||
692 | /* | |
693 | * Find a zvol_state_t given the full major+minor dev_t. If found, | |
694 | * return with zv_state_lock taken, otherwise, return (NULL) without | |
695 | * taking zv_state_lock. | |
696 | */ | |
697 | static zvol_state_t * | |
698 | zvol_find_by_dev(dev_t dev) | |
699 | { | |
700 | zvol_state_t *zv; | |
701 | ||
702 | rw_enter(&zvol_state_lock, RW_READER); | |
703 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
704 | zv = list_next(&zvol_state_list, zv)) { | |
705 | mutex_enter(&zv->zv_state_lock); | |
706 | if (zv->zv_zso->zvo_dev == dev) { | |
707 | rw_exit(&zvol_state_lock); | |
708 | return (zv); | |
709 | } | |
710 | mutex_exit(&zv->zv_state_lock); | |
711 | } | |
712 | rw_exit(&zvol_state_lock); | |
713 | ||
714 | return (NULL); | |
715 | } | |
716 | ||
5df7e9d8 MM |
717 | static struct kobject * |
718 | zvol_probe(dev_t dev, int *part, void *arg) | |
719 | { | |
720 | zvol_state_t *zv; | |
721 | struct kobject *kobj; | |
722 | ||
723 | zv = zvol_find_by_dev(dev); | |
724 | kobj = zv ? get_disk_and_module(zv->zv_zso->zvo_disk) : NULL; | |
725 | ASSERT(zv == NULL || MUTEX_HELD(&zv->zv_state_lock)); | |
726 | if (zv) | |
727 | mutex_exit(&zv->zv_state_lock); | |
728 | ||
729 | return (kobj); | |
730 | } | |
731 | ||
732 | static struct block_device_operations zvol_ops = { | |
733 | .open = zvol_open, | |
734 | .release = zvol_release, | |
735 | .ioctl = zvol_ioctl, | |
736 | .compat_ioctl = zvol_compat_ioctl, | |
5df7e9d8 | 737 | .check_events = zvol_check_events, |
5df7e9d8 MM |
738 | .revalidate_disk = zvol_revalidate_disk, |
739 | .getgeo = zvol_getgeo, | |
740 | .owner = THIS_MODULE, | |
d817c171 CK |
741 | #ifdef HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS |
742 | .submit_bio = zvol_submit_bio, | |
743 | #endif | |
5df7e9d8 MM |
744 | }; |
745 | ||
746 | /* | |
747 | * Allocate memory for a new zvol_state_t and setup the required | |
748 | * request queue and generic disk structures for the block device. | |
749 | */ | |
750 | static zvol_state_t * | |
751 | zvol_alloc(dev_t dev, const char *name) | |
752 | { | |
753 | zvol_state_t *zv; | |
68dde63d | 754 | struct zvol_state_os *zso; |
5df7e9d8 MM |
755 | uint64_t volmode; |
756 | ||
757 | if (dsl_prop_get_integer(name, "volmode", &volmode, NULL) != 0) | |
758 | return (NULL); | |
759 | ||
760 | if (volmode == ZFS_VOLMODE_DEFAULT) | |
761 | volmode = zvol_volmode; | |
762 | ||
763 | if (volmode == ZFS_VOLMODE_NONE) | |
764 | return (NULL); | |
765 | ||
766 | zv = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); | |
68dde63d BB |
767 | zso = kmem_zalloc(sizeof (struct zvol_state_os), KM_SLEEP); |
768 | zv->zv_zso = zso; | |
0ca45cb3 | 769 | zv->zv_volmode = volmode; |
5df7e9d8 MM |
770 | |
771 | list_link_init(&zv->zv_next); | |
5df7e9d8 MM |
772 | mutex_init(&zv->zv_state_lock, NULL, MUTEX_DEFAULT, NULL); |
773 | ||
d817c171 CK |
774 | #ifdef HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS |
775 | zso->zvo_queue = blk_alloc_queue(NUMA_NO_NODE); | |
776 | #else | |
68dde63d | 777 | zso->zvo_queue = blk_generic_alloc_queue(zvol_request, NUMA_NO_NODE); |
d817c171 | 778 | #endif |
68dde63d | 779 | if (zso->zvo_queue == NULL) |
5df7e9d8 MM |
780 | goto out_kmem; |
781 | ||
68dde63d | 782 | blk_queue_set_write_cache(zso->zvo_queue, B_TRUE, B_TRUE); |
5df7e9d8 MM |
783 | |
784 | /* Limit read-ahead to a single page to prevent over-prefetching. */ | |
68dde63d | 785 | blk_queue_set_read_ahead(zso->zvo_queue, 1); |
5df7e9d8 MM |
786 | |
787 | /* Disable write merging in favor of the ZIO pipeline. */ | |
68dde63d | 788 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, zso->zvo_queue); |
5df7e9d8 | 789 | |
68dde63d BB |
790 | zso->zvo_disk = alloc_disk(ZVOL_MINORS); |
791 | if (zso->zvo_disk == NULL) | |
5df7e9d8 MM |
792 | goto out_queue; |
793 | ||
68dde63d BB |
794 | zso->zvo_queue->queuedata = zv; |
795 | zso->zvo_dev = dev; | |
5df7e9d8 MM |
796 | zv->zv_open_count = 0; |
797 | strlcpy(zv->zv_name, name, MAXNAMELEN); | |
798 | ||
2cc479d0 | 799 | zfs_rangelock_init(&zv->zv_rangelock, NULL, NULL); |
5df7e9d8 MM |
800 | rw_init(&zv->zv_suspend_lock, NULL, RW_DEFAULT, NULL); |
801 | ||
68dde63d BB |
802 | zso->zvo_disk->major = zvol_major; |
803 | zso->zvo_disk->events = DISK_EVENT_MEDIA_CHANGE; | |
5df7e9d8 MM |
804 | |
805 | if (volmode == ZFS_VOLMODE_DEV) { | |
806 | /* | |
807 | * ZFS_VOLMODE_DEV disable partitioning on ZVOL devices: set | |
808 | * gendisk->minors = 1 as noted in include/linux/genhd.h. | |
809 | * Also disable extended partition numbers (GENHD_FL_EXT_DEVT) | |
810 | * and suppresses partition scanning (GENHD_FL_NO_PART_SCAN) | |
811 | * setting gendisk->flags accordingly. | |
812 | */ | |
68dde63d | 813 | zso->zvo_disk->minors = 1; |
5df7e9d8 | 814 | #if defined(GENHD_FL_EXT_DEVT) |
68dde63d | 815 | zso->zvo_disk->flags &= ~GENHD_FL_EXT_DEVT; |
5df7e9d8 MM |
816 | #endif |
817 | #if defined(GENHD_FL_NO_PART_SCAN) | |
68dde63d | 818 | zso->zvo_disk->flags |= GENHD_FL_NO_PART_SCAN; |
5df7e9d8 MM |
819 | #endif |
820 | } | |
68dde63d BB |
821 | zso->zvo_disk->first_minor = (dev & MINORMASK); |
822 | zso->zvo_disk->fops = &zvol_ops; | |
823 | zso->zvo_disk->private_data = zv; | |
824 | zso->zvo_disk->queue = zso->zvo_queue; | |
825 | snprintf(zso->zvo_disk->disk_name, DISK_NAME_LEN, "%s%d", | |
5df7e9d8 MM |
826 | ZVOL_DEV_NAME, (dev & MINORMASK)); |
827 | ||
828 | return (zv); | |
829 | ||
830 | out_queue: | |
68dde63d | 831 | blk_cleanup_queue(zso->zvo_queue); |
5df7e9d8 | 832 | out_kmem: |
68dde63d | 833 | kmem_free(zso, sizeof (struct zvol_state_os)); |
5df7e9d8 MM |
834 | kmem_free(zv, sizeof (zvol_state_t)); |
835 | return (NULL); | |
836 | } | |
837 | ||
838 | /* | |
839 | * Cleanup then free a zvol_state_t which was created by zvol_alloc(). | |
840 | * At this time, the structure is not opened by anyone, is taken off | |
841 | * the zvol_state_list, and has its private data set to NULL. | |
842 | * The zvol_state_lock is dropped. | |
99573cc0 PS |
843 | * |
844 | * This function may take many milliseconds to complete (e.g. we've seen | |
845 | * it take over 256ms), due to the calls to "blk_cleanup_queue" and | |
846 | * "del_gendisk". Thus, consumers need to be careful to account for this | |
847 | * latency when calling this function. | |
5df7e9d8 MM |
848 | */ |
849 | static void | |
850 | zvol_free(zvol_state_t *zv) | |
851 | { | |
852 | ||
853 | ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock)); | |
854 | ASSERT(!MUTEX_HELD(&zv->zv_state_lock)); | |
0b32d817 RM |
855 | ASSERT0(zv->zv_open_count); |
856 | ASSERT3P(zv->zv_zso->zvo_disk->private_data, ==, NULL); | |
5df7e9d8 MM |
857 | |
858 | rw_destroy(&zv->zv_suspend_lock); | |
2cc479d0 | 859 | zfs_rangelock_fini(&zv->zv_rangelock); |
5df7e9d8 MM |
860 | |
861 | del_gendisk(zv->zv_zso->zvo_disk); | |
862 | blk_cleanup_queue(zv->zv_zso->zvo_queue); | |
863 | put_disk(zv->zv_zso->zvo_disk); | |
864 | ||
865 | ida_simple_remove(&zvol_ida, | |
866 | MINOR(zv->zv_zso->zvo_dev) >> ZVOL_MINOR_BITS); | |
867 | ||
868 | mutex_destroy(&zv->zv_state_lock); | |
4547fc4e | 869 | dataset_kstats_destroy(&zv->zv_kstat); |
5df7e9d8 MM |
870 | |
871 | kmem_free(zv->zv_zso, sizeof (struct zvol_state_os)); | |
872 | kmem_free(zv, sizeof (zvol_state_t)); | |
873 | } | |
874 | ||
0ca45cb3 MM |
875 | void |
876 | zvol_wait_close(zvol_state_t *zv) | |
877 | { | |
878 | } | |
879 | ||
5df7e9d8 MM |
880 | /* |
881 | * Create a block device minor node and setup the linkage between it | |
882 | * and the specified volume. Once this function returns the block | |
883 | * device is live and ready for use. | |
884 | */ | |
885 | static int | |
ec213971 | 886 | zvol_os_create_minor(const char *name) |
5df7e9d8 MM |
887 | { |
888 | zvol_state_t *zv; | |
889 | objset_t *os; | |
890 | dmu_object_info_t *doi; | |
891 | uint64_t volsize; | |
892 | uint64_t len; | |
893 | unsigned minor = 0; | |
894 | int error = 0; | |
895 | int idx; | |
896 | uint64_t hash = zvol_name_hash(name); | |
897 | ||
898 | if (zvol_inhibit_dev) | |
899 | return (0); | |
900 | ||
901 | idx = ida_simple_get(&zvol_ida, 0, 0, kmem_flags_convert(KM_SLEEP)); | |
902 | if (idx < 0) | |
903 | return (SET_ERROR(-idx)); | |
904 | minor = idx << ZVOL_MINOR_BITS; | |
905 | ||
906 | zv = zvol_find_by_name_hash(name, hash, RW_NONE); | |
907 | if (zv) { | |
908 | ASSERT(MUTEX_HELD(&zv->zv_state_lock)); | |
909 | mutex_exit(&zv->zv_state_lock); | |
910 | ida_simple_remove(&zvol_ida, idx); | |
911 | return (SET_ERROR(EEXIST)); | |
912 | } | |
913 | ||
914 | doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP); | |
915 | ||
916 | error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, B_TRUE, FTAG, &os); | |
917 | if (error) | |
918 | goto out_doi; | |
919 | ||
920 | error = dmu_object_info(os, ZVOL_OBJ, doi); | |
921 | if (error) | |
922 | goto out_dmu_objset_disown; | |
923 | ||
924 | error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); | |
925 | if (error) | |
926 | goto out_dmu_objset_disown; | |
927 | ||
928 | zv = zvol_alloc(MKDEV(zvol_major, minor), name); | |
929 | if (zv == NULL) { | |
930 | error = SET_ERROR(EAGAIN); | |
931 | goto out_dmu_objset_disown; | |
932 | } | |
933 | zv->zv_hash = hash; | |
934 | ||
935 | if (dmu_objset_is_snapshot(os)) | |
936 | zv->zv_flags |= ZVOL_RDONLY; | |
937 | ||
938 | zv->zv_volblocksize = doi->doi_data_block_size; | |
939 | zv->zv_volsize = volsize; | |
940 | zv->zv_objset = os; | |
941 | ||
942 | set_capacity(zv->zv_zso->zvo_disk, zv->zv_volsize >> 9); | |
943 | ||
944 | blk_queue_max_hw_sectors(zv->zv_zso->zvo_queue, | |
945 | (DMU_MAX_ACCESS / 4) >> 9); | |
946 | blk_queue_max_segments(zv->zv_zso->zvo_queue, UINT16_MAX); | |
947 | blk_queue_max_segment_size(zv->zv_zso->zvo_queue, UINT_MAX); | |
948 | blk_queue_physical_block_size(zv->zv_zso->zvo_queue, | |
949 | zv->zv_volblocksize); | |
950 | blk_queue_io_opt(zv->zv_zso->zvo_queue, zv->zv_volblocksize); | |
951 | blk_queue_max_discard_sectors(zv->zv_zso->zvo_queue, | |
952 | (zvol_max_discard_blocks * zv->zv_volblocksize) >> 9); | |
953 | blk_queue_discard_granularity(zv->zv_zso->zvo_queue, | |
954 | zv->zv_volblocksize); | |
955 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, zv->zv_zso->zvo_queue); | |
956 | #ifdef QUEUE_FLAG_NONROT | |
957 | blk_queue_flag_set(QUEUE_FLAG_NONROT, zv->zv_zso->zvo_queue); | |
958 | #endif | |
959 | #ifdef QUEUE_FLAG_ADD_RANDOM | |
960 | blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zv->zv_zso->zvo_queue); | |
961 | #endif | |
962 | /* This flag was introduced in kernel version 4.12. */ | |
963 | #ifdef QUEUE_FLAG_SCSI_PASSTHROUGH | |
964 | blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, zv->zv_zso->zvo_queue); | |
965 | #endif | |
966 | ||
967 | if (spa_writeable(dmu_objset_spa(os))) { | |
968 | if (zil_replay_disable) | |
969 | zil_destroy(dmu_objset_zil(os), B_FALSE); | |
970 | else | |
971 | zil_replay(os, zv, zvol_replay_vector); | |
972 | } | |
4547fc4e AJ |
973 | ASSERT3P(zv->zv_kstat.dk_kstats, ==, NULL); |
974 | dataset_kstats_create(&zv->zv_kstat, zv->zv_objset); | |
5df7e9d8 MM |
975 | |
976 | /* | |
977 | * When udev detects the addition of the device it will immediately | |
978 | * invoke blkid(8) to determine the type of content on the device. | |
979 | * Prefetching the blocks commonly scanned by blkid(8) will speed | |
980 | * up this process. | |
981 | */ | |
982 | len = MIN(MAX(zvol_prefetch_bytes, 0), SPA_MAXBLOCKSIZE); | |
983 | if (len > 0) { | |
984 | dmu_prefetch(os, ZVOL_OBJ, 0, 0, len, ZIO_PRIORITY_SYNC_READ); | |
985 | dmu_prefetch(os, ZVOL_OBJ, 0, volsize - len, len, | |
986 | ZIO_PRIORITY_SYNC_READ); | |
987 | } | |
988 | ||
989 | zv->zv_objset = NULL; | |
990 | out_dmu_objset_disown: | |
991 | dmu_objset_disown(os, B_TRUE, FTAG); | |
992 | out_doi: | |
993 | kmem_free(doi, sizeof (dmu_object_info_t)); | |
994 | ||
995 | /* | |
996 | * Keep in mind that once add_disk() is called, the zvol is | |
997 | * announced to the world, and zvol_open()/zvol_release() can | |
998 | * be called at any time. Incidentally, add_disk() itself calls | |
999 | * zvol_open()->zvol_first_open() and zvol_release()->zvol_last_close() | |
1000 | * directly as well. | |
1001 | */ | |
1002 | if (error == 0) { | |
1003 | rw_enter(&zvol_state_lock, RW_WRITER); | |
1004 | zvol_insert(zv); | |
1005 | rw_exit(&zvol_state_lock); | |
1006 | add_disk(zv->zv_zso->zvo_disk); | |
1007 | } else { | |
1008 | ida_simple_remove(&zvol_ida, idx); | |
1009 | } | |
1010 | ||
ec213971 | 1011 | return (error); |
5df7e9d8 MM |
1012 | } |
1013 | ||
1014 | static void | |
1015 | zvol_rename_minor(zvol_state_t *zv, const char *newname) | |
1016 | { | |
1017 | int readonly = get_disk_ro(zv->zv_zso->zvo_disk); | |
1018 | ||
1019 | ASSERT(RW_LOCK_HELD(&zvol_state_lock)); | |
1020 | ASSERT(MUTEX_HELD(&zv->zv_state_lock)); | |
1021 | ||
1022 | strlcpy(zv->zv_name, newname, sizeof (zv->zv_name)); | |
1023 | ||
1024 | /* move to new hashtable entry */ | |
1025 | zv->zv_hash = zvol_name_hash(zv->zv_name); | |
1026 | hlist_del(&zv->zv_hlink); | |
1027 | hlist_add_head(&zv->zv_hlink, ZVOL_HT_HEAD(zv->zv_hash)); | |
1028 | ||
1029 | /* | |
1030 | * The block device's read-only state is briefly changed causing | |
1031 | * a KOBJ_CHANGE uevent to be issued. This ensures udev detects | |
1032 | * the name change and fixes the symlinks. This does not change | |
1033 | * ZVOL_RDONLY in zv->zv_flags so the actual read-only state never | |
1034 | * changes. This would normally be done using kobject_uevent() but | |
1035 | * that is a GPL-only symbol which is why we need this workaround. | |
1036 | */ | |
1037 | set_disk_ro(zv->zv_zso->zvo_disk, !readonly); | |
1038 | set_disk_ro(zv->zv_zso->zvo_disk, readonly); | |
1039 | } | |
1040 | ||
1041 | static void | |
1042 | zvol_set_disk_ro_impl(zvol_state_t *zv, int flags) | |
1043 | { | |
1044 | ||
1045 | set_disk_ro(zv->zv_zso->zvo_disk, flags); | |
1046 | } | |
1047 | ||
1048 | static void | |
1049 | zvol_set_capacity_impl(zvol_state_t *zv, uint64_t capacity) | |
1050 | { | |
1051 | ||
1052 | set_capacity(zv->zv_zso->zvo_disk, capacity); | |
1053 | } | |
1054 | ||
1055 | const static zvol_platform_ops_t zvol_linux_ops = { | |
1056 | .zv_free = zvol_free, | |
1057 | .zv_rename_minor = zvol_rename_minor, | |
ec213971 | 1058 | .zv_create_minor = zvol_os_create_minor, |
5df7e9d8 MM |
1059 | .zv_update_volsize = zvol_update_volsize, |
1060 | .zv_clear_private = zvol_clear_private, | |
1061 | .zv_is_zvol = zvol_is_zvol_impl, | |
1062 | .zv_set_disk_ro = zvol_set_disk_ro_impl, | |
1063 | .zv_set_capacity = zvol_set_capacity_impl, | |
1064 | }; | |
1065 | ||
1066 | int | |
1067 | zvol_init(void) | |
1068 | { | |
1069 | int error; | |
1070 | int threads = MIN(MAX(zvol_threads, 1), 1024); | |
1071 | ||
1072 | error = register_blkdev(zvol_major, ZVOL_DRIVER); | |
1073 | if (error) { | |
1074 | printk(KERN_INFO "ZFS: register_blkdev() failed %d\n", error); | |
1075 | return (error); | |
1076 | } | |
1077 | zvol_taskq = taskq_create(ZVOL_DRIVER, threads, maxclsyspri, | |
1078 | threads * 2, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC); | |
1079 | if (zvol_taskq == NULL) { | |
1080 | unregister_blkdev(zvol_major, ZVOL_DRIVER); | |
1081 | return (-ENOMEM); | |
1082 | } | |
1083 | zvol_init_impl(); | |
1084 | blk_register_region(MKDEV(zvol_major, 0), 1UL << MINORBITS, | |
1085 | THIS_MODULE, zvol_probe, NULL, NULL); | |
1086 | ||
1087 | ida_init(&zvol_ida); | |
1088 | zvol_register_ops(&zvol_linux_ops); | |
1089 | return (0); | |
1090 | } | |
1091 | ||
1092 | void | |
1093 | zvol_fini(void) | |
1094 | { | |
5df7e9d8 MM |
1095 | zvol_fini_impl(); |
1096 | blk_unregister_region(MKDEV(zvol_major, 0), 1UL << MINORBITS); | |
1097 | unregister_blkdev(zvol_major, ZVOL_DRIVER); | |
1098 | taskq_destroy(zvol_taskq); | |
1099 | ida_destroy(&zvol_ida); | |
1100 | } | |
1101 | ||
1102 | /* BEGIN CSTYLED */ | |
1103 | module_param(zvol_inhibit_dev, uint, 0644); | |
1104 | MODULE_PARM_DESC(zvol_inhibit_dev, "Do not create zvol device nodes"); | |
1105 | ||
1106 | module_param(zvol_major, uint, 0444); | |
1107 | MODULE_PARM_DESC(zvol_major, "Major number for zvol device"); | |
1108 | ||
1109 | module_param(zvol_threads, uint, 0444); | |
1110 | MODULE_PARM_DESC(zvol_threads, "Max number of threads to handle I/O requests"); | |
1111 | ||
1112 | module_param(zvol_request_sync, uint, 0644); | |
1113 | MODULE_PARM_DESC(zvol_request_sync, "Synchronously handle bio requests"); | |
1114 | ||
1115 | module_param(zvol_max_discard_blocks, ulong, 0444); | |
1116 | MODULE_PARM_DESC(zvol_max_discard_blocks, "Max number of blocks to discard"); | |
1117 | ||
1118 | module_param(zvol_prefetch_bytes, uint, 0644); | |
1119 | MODULE_PARM_DESC(zvol_prefetch_bytes, "Prefetch N bytes at zvol start+end"); | |
1120 | ||
1121 | module_param(zvol_volmode, uint, 0644); | |
1122 | MODULE_PARM_DESC(zvol_volmode, "Default volmode property value"); | |
1123 | /* END CSTYLED */ |