]>
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 | ||
18168da7 AZ |
44 | static unsigned int zvol_major = ZVOL_MAJOR; |
45 | static unsigned int zvol_request_sync = 0; | |
46 | static unsigned int zvol_prefetch_bytes = (128 * 1024); | |
47 | static unsigned long zvol_max_discard_blocks = 16384; | |
48 | static unsigned int zvol_threads = 32; | |
abdcef47 PH |
49 | |
50 | #ifndef HAVE_BLKDEV_GET_ERESTARTSYS | |
18168da7 | 51 | static const unsigned int zvol_open_timeout_ms = 1000; |
abdcef47 | 52 | #endif |
5df7e9d8 MM |
53 | |
54 | struct zvol_state_os { | |
55 | struct gendisk *zvo_disk; /* generic disk */ | |
56 | struct request_queue *zvo_queue; /* request queue */ | |
5df7e9d8 MM |
57 | dev_t zvo_dev; /* device id */ |
58 | }; | |
59 | ||
60 | taskq_t *zvol_taskq; | |
61 | static struct ida zvol_ida; | |
62 | ||
e439ee83 | 63 | typedef struct zv_request_stack { |
5df7e9d8 MM |
64 | zvol_state_t *zv; |
65 | struct bio *bio; | |
5df7e9d8 MM |
66 | } zv_request_t; |
67 | ||
e439ee83 CS |
68 | typedef struct zv_request_task { |
69 | zv_request_t zvr; | |
70 | taskq_ent_t ent; | |
71 | } zv_request_task_t; | |
72 | ||
73 | static zv_request_task_t * | |
74 | zv_request_task_create(zv_request_t zvr) | |
75 | { | |
76 | zv_request_task_t *task; | |
77 | task = kmem_alloc(sizeof (zv_request_task_t), KM_SLEEP); | |
78 | taskq_init_ent(&task->ent); | |
79 | task->zvr = zvr; | |
80 | return (task); | |
81 | } | |
82 | ||
83 | static void | |
84 | zv_request_task_free(zv_request_task_t *task) | |
85 | { | |
86 | kmem_free(task, sizeof (*task)); | |
87 | } | |
88 | ||
5df7e9d8 MM |
89 | /* |
90 | * Given a path, return TRUE if path is a ZVOL. | |
91 | */ | |
1dccfd7a CS |
92 | boolean_t |
93 | zvol_os_is_zvol(const char *path) | |
5df7e9d8 | 94 | { |
b7281c88 | 95 | dev_t dev = 0; |
5df7e9d8 | 96 | |
b7281c88 | 97 | if (vdev_lookup_bdev(path, &dev) != 0) |
5df7e9d8 MM |
98 | return (B_FALSE); |
99 | ||
b7281c88 | 100 | if (MAJOR(dev) == zvol_major) |
5df7e9d8 MM |
101 | return (B_TRUE); |
102 | ||
103 | return (B_FALSE); | |
104 | } | |
105 | ||
5df7e9d8 | 106 | static void |
e439ee83 | 107 | zvol_write(zv_request_t *zvr) |
5df7e9d8 | 108 | { |
5df7e9d8 | 109 | struct bio *bio = zvr->bio; |
1c2358c1 | 110 | int error = 0; |
d0cd9a5c | 111 | zfs_uio_t uio; |
1c2358c1 | 112 | |
d0cd9a5c | 113 | zfs_uio_bvec_init(&uio, bio); |
5df7e9d8 MM |
114 | |
115 | zvol_state_t *zv = zvr->zv; | |
0b32d817 RM |
116 | ASSERT3P(zv, !=, NULL); |
117 | ASSERT3U(zv->zv_open_count, >, 0); | |
118 | ASSERT3P(zv->zv_zilog, !=, NULL); | |
5df7e9d8 | 119 | |
0929c4de MA |
120 | /* bio marked as FLUSH need to flush before write */ |
121 | if (bio_is_flush(bio)) | |
122 | zil_commit(zv->zv_zilog, ZVOL_OBJ); | |
123 | ||
124 | /* Some requests are just for flush and nothing else. */ | |
125 | if (uio.uio_resid == 0) { | |
126 | rw_exit(&zv->zv_suspend_lock); | |
127 | BIO_END_IO(bio, 0); | |
0929c4de MA |
128 | return; |
129 | } | |
130 | ||
a970f059 BB |
131 | struct request_queue *q = zv->zv_zso->zvo_queue; |
132 | struct gendisk *disk = zv->zv_zso->zvo_disk; | |
5df7e9d8 | 133 | ssize_t start_resid = uio.uio_resid; |
a970f059 BB |
134 | unsigned long start_time; |
135 | ||
136 | boolean_t acct = blk_queue_io_stat(q); | |
137 | if (acct) | |
138 | start_time = blk_generic_start_io_acct(q, disk, WRITE, bio); | |
5df7e9d8 MM |
139 | |
140 | boolean_t sync = | |
141 | bio_is_fua(bio) || zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; | |
142 | ||
0929c4de MA |
143 | zfs_locked_range_t *lr = zfs_rangelock_enter(&zv->zv_rangelock, |
144 | uio.uio_loffset, uio.uio_resid, RL_WRITER); | |
145 | ||
5df7e9d8 MM |
146 | uint64_t volsize = zv->zv_volsize; |
147 | while (uio.uio_resid > 0 && uio.uio_loffset < volsize) { | |
148 | uint64_t bytes = MIN(uio.uio_resid, DMU_MAX_ACCESS >> 1); | |
149 | uint64_t off = uio.uio_loffset; | |
150 | dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); | |
151 | ||
152 | if (bytes > volsize - off) /* don't write past the end */ | |
153 | bytes = volsize - off; | |
154 | ||
20f28785 | 155 | dmu_tx_hold_write_by_dnode(tx, zv->zv_dn, off, bytes); |
5df7e9d8 MM |
156 | |
157 | /* This will only fail for ENOSPC */ | |
158 | error = dmu_tx_assign(tx, TXG_WAIT); | |
159 | if (error) { | |
160 | dmu_tx_abort(tx); | |
161 | break; | |
162 | } | |
163 | error = dmu_write_uio_dnode(zv->zv_dn, &uio, bytes, tx); | |
164 | if (error == 0) { | |
165 | zvol_log_write(zv, tx, off, bytes, sync); | |
166 | } | |
167 | dmu_tx_commit(tx); | |
168 | ||
169 | if (error) | |
170 | break; | |
171 | } | |
0929c4de | 172 | zfs_rangelock_exit(lr); |
5df7e9d8 MM |
173 | |
174 | int64_t nwritten = start_resid - uio.uio_resid; | |
4547fc4e | 175 | dataset_kstats_update_write_kstats(&zv->zv_kstat, nwritten); |
5df7e9d8 MM |
176 | task_io_account_write(nwritten); |
177 | ||
178 | if (sync) | |
179 | zil_commit(zv->zv_zilog, ZVOL_OBJ); | |
180 | ||
181 | rw_exit(&zv->zv_suspend_lock); | |
a970f059 BB |
182 | |
183 | if (acct) | |
184 | blk_generic_end_io_acct(q, disk, WRITE, bio, start_time); | |
185 | ||
5df7e9d8 | 186 | BIO_END_IO(bio, -error); |
5df7e9d8 MM |
187 | } |
188 | ||
189 | static void | |
e439ee83 CS |
190 | zvol_write_task(void *arg) |
191 | { | |
192 | zv_request_task_t *task = arg; | |
193 | zvol_write(&task->zvr); | |
194 | zv_request_task_free(task); | |
195 | } | |
196 | ||
197 | static void | |
198 | zvol_discard(zv_request_t *zvr) | |
5df7e9d8 | 199 | { |
5df7e9d8 MM |
200 | struct bio *bio = zvr->bio; |
201 | zvol_state_t *zv = zvr->zv; | |
202 | uint64_t start = BIO_BI_SECTOR(bio) << 9; | |
203 | uint64_t size = BIO_BI_SIZE(bio); | |
204 | uint64_t end = start + size; | |
205 | boolean_t sync; | |
206 | int error = 0; | |
207 | dmu_tx_t *tx; | |
5df7e9d8 | 208 | |
0b32d817 RM |
209 | ASSERT3P(zv, !=, NULL); |
210 | ASSERT3U(zv->zv_open_count, >, 0); | |
211 | ASSERT3P(zv->zv_zilog, !=, NULL); | |
5df7e9d8 | 212 | |
a970f059 BB |
213 | struct request_queue *q = zv->zv_zso->zvo_queue; |
214 | struct gendisk *disk = zv->zv_zso->zvo_disk; | |
215 | unsigned long start_time; | |
216 | ||
217 | boolean_t acct = blk_queue_io_stat(q); | |
218 | if (acct) | |
219 | start_time = blk_generic_start_io_acct(q, disk, WRITE, bio); | |
5df7e9d8 MM |
220 | |
221 | sync = bio_is_fua(bio) || zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; | |
222 | ||
223 | if (end > zv->zv_volsize) { | |
224 | error = SET_ERROR(EIO); | |
225 | goto unlock; | |
226 | } | |
227 | ||
228 | /* | |
229 | * Align the request to volume block boundaries when a secure erase is | |
230 | * not required. This will prevent dnode_free_range() from zeroing out | |
231 | * the unaligned parts which is slow (read-modify-write) and useless | |
232 | * since we are not freeing any space by doing so. | |
233 | */ | |
234 | if (!bio_is_secure_erase(bio)) { | |
235 | start = P2ROUNDUP(start, zv->zv_volblocksize); | |
236 | end = P2ALIGN(end, zv->zv_volblocksize); | |
237 | size = end - start; | |
238 | } | |
239 | ||
240 | if (start >= end) | |
241 | goto unlock; | |
242 | ||
0929c4de MA |
243 | zfs_locked_range_t *lr = zfs_rangelock_enter(&zv->zv_rangelock, |
244 | start, size, RL_WRITER); | |
245 | ||
5df7e9d8 MM |
246 | tx = dmu_tx_create(zv->zv_objset); |
247 | dmu_tx_mark_netfree(tx); | |
248 | error = dmu_tx_assign(tx, TXG_WAIT); | |
249 | if (error != 0) { | |
250 | dmu_tx_abort(tx); | |
251 | } else { | |
252 | zvol_log_truncate(zv, tx, start, size, B_TRUE); | |
253 | dmu_tx_commit(tx); | |
254 | error = dmu_free_long_range(zv->zv_objset, | |
255 | ZVOL_OBJ, start, size); | |
256 | } | |
0929c4de | 257 | zfs_rangelock_exit(lr); |
5df7e9d8 MM |
258 | |
259 | if (error == 0 && sync) | |
260 | zil_commit(zv->zv_zilog, ZVOL_OBJ); | |
261 | ||
0929c4de | 262 | unlock: |
5df7e9d8 | 263 | rw_exit(&zv->zv_suspend_lock); |
a970f059 BB |
264 | |
265 | if (acct) | |
266 | blk_generic_end_io_acct(q, disk, WRITE, bio, start_time); | |
267 | ||
5df7e9d8 | 268 | BIO_END_IO(bio, -error); |
5df7e9d8 MM |
269 | } |
270 | ||
271 | static void | |
e439ee83 CS |
272 | zvol_discard_task(void *arg) |
273 | { | |
274 | zv_request_task_t *task = arg; | |
275 | zvol_discard(&task->zvr); | |
276 | zv_request_task_free(task); | |
277 | } | |
278 | ||
279 | static void | |
280 | zvol_read(zv_request_t *zvr) | |
5df7e9d8 | 281 | { |
5df7e9d8 | 282 | struct bio *bio = zvr->bio; |
1c2358c1 | 283 | int error = 0; |
d0cd9a5c | 284 | zfs_uio_t uio; |
1c2358c1 | 285 | |
d0cd9a5c | 286 | zfs_uio_bvec_init(&uio, bio); |
5df7e9d8 MM |
287 | |
288 | zvol_state_t *zv = zvr->zv; | |
0b32d817 RM |
289 | ASSERT3P(zv, !=, NULL); |
290 | ASSERT3U(zv->zv_open_count, >, 0); | |
5df7e9d8 | 291 | |
a970f059 BB |
292 | struct request_queue *q = zv->zv_zso->zvo_queue; |
293 | struct gendisk *disk = zv->zv_zso->zvo_disk; | |
5df7e9d8 | 294 | ssize_t start_resid = uio.uio_resid; |
a970f059 BB |
295 | unsigned long start_time; |
296 | ||
297 | boolean_t acct = blk_queue_io_stat(q); | |
298 | if (acct) | |
299 | start_time = blk_generic_start_io_acct(q, disk, READ, bio); | |
5df7e9d8 | 300 | |
0929c4de MA |
301 | zfs_locked_range_t *lr = zfs_rangelock_enter(&zv->zv_rangelock, |
302 | uio.uio_loffset, uio.uio_resid, RL_READER); | |
303 | ||
5df7e9d8 MM |
304 | uint64_t volsize = zv->zv_volsize; |
305 | while (uio.uio_resid > 0 && uio.uio_loffset < volsize) { | |
306 | uint64_t bytes = MIN(uio.uio_resid, DMU_MAX_ACCESS >> 1); | |
307 | ||
308 | /* don't read past the end */ | |
309 | if (bytes > volsize - uio.uio_loffset) | |
310 | bytes = volsize - uio.uio_loffset; | |
311 | ||
312 | error = dmu_read_uio_dnode(zv->zv_dn, &uio, bytes); | |
313 | if (error) { | |
314 | /* convert checksum errors into IO errors */ | |
315 | if (error == ECKSUM) | |
316 | error = SET_ERROR(EIO); | |
317 | break; | |
318 | } | |
319 | } | |
0929c4de | 320 | zfs_rangelock_exit(lr); |
5df7e9d8 MM |
321 | |
322 | int64_t nread = start_resid - uio.uio_resid; | |
4547fc4e | 323 | dataset_kstats_update_read_kstats(&zv->zv_kstat, nread); |
5df7e9d8 MM |
324 | task_io_account_read(nread); |
325 | ||
326 | rw_exit(&zv->zv_suspend_lock); | |
a970f059 BB |
327 | |
328 | if (acct) | |
329 | blk_generic_end_io_acct(q, disk, READ, bio, start_time); | |
330 | ||
5df7e9d8 | 331 | BIO_END_IO(bio, -error); |
e439ee83 CS |
332 | } |
333 | ||
334 | static void | |
335 | zvol_read_task(void *arg) | |
336 | { | |
337 | zv_request_task_t *task = arg; | |
338 | zvol_read(&task->zvr); | |
339 | zv_request_task_free(task); | |
5df7e9d8 MM |
340 | } |
341 | ||
d817c171 | 342 | #ifdef HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS |
435a451e CK |
343 | #ifdef HAVE_BDEV_SUBMIT_BIO_RETURNS_VOID |
344 | static void | |
345 | zvol_submit_bio(struct bio *bio) | |
346 | #else | |
d817c171 CK |
347 | static blk_qc_t |
348 | zvol_submit_bio(struct bio *bio) | |
435a451e | 349 | #endif |
d817c171 | 350 | #else |
5df7e9d8 MM |
351 | static MAKE_REQUEST_FN_RET |
352 | zvol_request(struct request_queue *q, struct bio *bio) | |
d817c171 | 353 | #endif |
5df7e9d8 | 354 | { |
d817c171 | 355 | #ifdef HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS |
d939930f CK |
356 | #if defined(HAVE_BIO_BDEV_DISK) |
357 | struct request_queue *q = bio->bi_bdev->bd_disk->queue; | |
358 | #else | |
d817c171 | 359 | struct request_queue *q = bio->bi_disk->queue; |
d939930f | 360 | #endif |
d817c171 | 361 | #endif |
5df7e9d8 MM |
362 | zvol_state_t *zv = q->queuedata; |
363 | fstrans_cookie_t cookie = spl_fstrans_mark(); | |
364 | uint64_t offset = BIO_BI_SECTOR(bio) << 9; | |
365 | uint64_t size = BIO_BI_SIZE(bio); | |
366 | int rw = bio_data_dir(bio); | |
5df7e9d8 MM |
367 | |
368 | if (bio_has_data(bio) && offset + size > zv->zv_volsize) { | |
369 | printk(KERN_INFO | |
370 | "%s: bad access: offset=%llu, size=%lu\n", | |
371 | zv->zv_zso->zvo_disk->disk_name, | |
372 | (long long unsigned)offset, | |
373 | (long unsigned)size); | |
374 | ||
375 | BIO_END_IO(bio, -SET_ERROR(EIO)); | |
376 | goto out; | |
377 | } | |
378 | ||
e439ee83 CS |
379 | zv_request_t zvr = { |
380 | .zv = zv, | |
381 | .bio = bio, | |
382 | }; | |
383 | zv_request_task_t *task; | |
384 | ||
5df7e9d8 | 385 | if (rw == WRITE) { |
5df7e9d8 MM |
386 | if (unlikely(zv->zv_flags & ZVOL_RDONLY)) { |
387 | BIO_END_IO(bio, -SET_ERROR(EROFS)); | |
388 | goto out; | |
389 | } | |
390 | ||
391 | /* | |
0929c4de MA |
392 | * Prevents the zvol from being suspended, or the ZIL being |
393 | * concurrently opened. Will be released after the i/o | |
394 | * completes. | |
5df7e9d8 MM |
395 | */ |
396 | rw_enter(&zv->zv_suspend_lock, RW_READER); | |
397 | ||
398 | /* | |
399 | * Open a ZIL if this is the first time we have written to this | |
400 | * zvol. We protect zv->zv_zilog with zv_suspend_lock rather | |
401 | * than zv_state_lock so that we don't need to acquire an | |
402 | * additional lock in this path. | |
403 | */ | |
404 | if (zv->zv_zilog == NULL) { | |
405 | rw_exit(&zv->zv_suspend_lock); | |
406 | rw_enter(&zv->zv_suspend_lock, RW_WRITER); | |
407 | if (zv->zv_zilog == NULL) { | |
408 | zv->zv_zilog = zil_open(zv->zv_objset, | |
409 | zvol_get_data); | |
410 | zv->zv_flags |= ZVOL_WRITTEN_TO; | |
93e36580 CS |
411 | /* replay / destroy done in zvol_create_minor */ |
412 | VERIFY0((zv->zv_zilog->zl_header->zh_flags & | |
413 | ZIL_REPLAY_NEEDED)); | |
5df7e9d8 MM |
414 | } |
415 | rw_downgrade(&zv->zv_suspend_lock); | |
416 | } | |
417 | ||
5df7e9d8 | 418 | /* |
0929c4de MA |
419 | * We don't want this thread to be blocked waiting for i/o to |
420 | * complete, so we instead wait from a taskq callback. The | |
421 | * i/o may be a ZIL write (via zil_commit()), or a read of an | |
422 | * indirect block, or a read of a data block (if this is a | |
423 | * partial-block write). We will indicate that the i/o is | |
424 | * complete by calling BIO_END_IO() from the taskq callback. | |
425 | * | |
426 | * This design allows the calling thread to continue and | |
427 | * initiate more concurrent operations by calling | |
428 | * zvol_request() again. There are typically only a small | |
429 | * number of threads available to call zvol_request() (e.g. | |
430 | * one per iSCSI target), so keeping the latency of | |
431 | * zvol_request() low is important for performance. | |
432 | * | |
433 | * The zvol_request_sync module parameter allows this | |
434 | * behavior to be altered, for performance evaluation | |
435 | * purposes. If the callback blocks, setting | |
436 | * zvol_request_sync=1 will result in much worse performance. | |
437 | * | |
438 | * We can have up to zvol_threads concurrent i/o's being | |
439 | * processed for all zvols on the system. This is typically | |
440 | * a vast improvement over the zvol_request_sync=1 behavior | |
441 | * of one i/o at a time per zvol. However, an even better | |
442 | * design would be for zvol_request() to initiate the zio | |
443 | * directly, and then be notified by the zio_done callback, | |
444 | * which would call BIO_END_IO(). Unfortunately, the DMU/ZIL | |
445 | * interfaces lack this functionality (they block waiting for | |
446 | * the i/o to complete). | |
5df7e9d8 | 447 | */ |
5df7e9d8 | 448 | if (bio_is_discard(bio) || bio_is_secure_erase(bio)) { |
0929c4de | 449 | if (zvol_request_sync) { |
e439ee83 | 450 | zvol_discard(&zvr); |
0929c4de | 451 | } else { |
e439ee83 | 452 | task = zv_request_task_create(zvr); |
0929c4de | 453 | taskq_dispatch_ent(zvol_taskq, |
e439ee83 | 454 | zvol_discard_task, task, 0, &task->ent); |
0929c4de | 455 | } |
5df7e9d8 | 456 | } else { |
0929c4de | 457 | if (zvol_request_sync) { |
e439ee83 | 458 | zvol_write(&zvr); |
0929c4de | 459 | } else { |
e439ee83 | 460 | task = zv_request_task_create(zvr); |
0929c4de | 461 | taskq_dispatch_ent(zvol_taskq, |
e439ee83 | 462 | zvol_write_task, task, 0, &task->ent); |
0929c4de | 463 | } |
5df7e9d8 MM |
464 | } |
465 | } else { | |
466 | /* | |
467 | * The SCST driver, and possibly others, may issue READ I/Os | |
468 | * with a length of zero bytes. These empty I/Os contain no | |
469 | * data and require no additional handling. | |
470 | */ | |
471 | if (size == 0) { | |
472 | BIO_END_IO(bio, 0); | |
473 | goto out; | |
474 | } | |
475 | ||
5df7e9d8 MM |
476 | rw_enter(&zv->zv_suspend_lock, RW_READER); |
477 | ||
0929c4de MA |
478 | /* See comment in WRITE case above. */ |
479 | if (zvol_request_sync) { | |
e439ee83 | 480 | zvol_read(&zvr); |
0929c4de | 481 | } else { |
e439ee83 | 482 | task = zv_request_task_create(zvr); |
0929c4de | 483 | taskq_dispatch_ent(zvol_taskq, |
e439ee83 | 484 | zvol_read_task, task, 0, &task->ent); |
0929c4de | 485 | } |
5df7e9d8 MM |
486 | } |
487 | ||
488 | out: | |
489 | spl_fstrans_unmark(cookie); | |
435a451e CK |
490 | #if (defined(HAVE_MAKE_REQUEST_FN_RET_QC) || \ |
491 | defined(HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS)) && \ | |
492 | !defined(HAVE_BDEV_SUBMIT_BIO_RETURNS_VOID) | |
5df7e9d8 MM |
493 | return (BLK_QC_T_NONE); |
494 | #endif | |
495 | } | |
496 | ||
497 | static int | |
498 | zvol_open(struct block_device *bdev, fmode_t flag) | |
499 | { | |
500 | zvol_state_t *zv; | |
501 | int error = 0; | |
8a02d01e | 502 | boolean_t drop_suspend = B_FALSE; |
77e2756d BB |
503 | #ifndef HAVE_BLKDEV_GET_ERESTARTSYS |
504 | hrtime_t timeout = MSEC2NSEC(zvol_open_timeout_ms); | |
505 | hrtime_t start = gethrtime(); | |
5df7e9d8 | 506 | |
77e2756d BB |
507 | retry: |
508 | #endif | |
5df7e9d8 MM |
509 | rw_enter(&zvol_state_lock, RW_READER); |
510 | /* | |
511 | * Obtain a copy of private_data under the zvol_state_lock to make | |
512 | * sure that either the result of zvol free code path setting | |
1dccfd7a | 513 | * bdev->bd_disk->private_data to NULL is observed, or zvol_os_free() |
5df7e9d8 MM |
514 | * is not called on this zv because of the positive zv_open_count. |
515 | */ | |
516 | zv = bdev->bd_disk->private_data; | |
517 | if (zv == NULL) { | |
518 | rw_exit(&zvol_state_lock); | |
519 | return (SET_ERROR(-ENXIO)); | |
520 | } | |
521 | ||
8a02d01e BB |
522 | mutex_enter(&zv->zv_state_lock); |
523 | /* | |
524 | * Make sure zvol is not suspended during first open | |
525 | * (hold zv_suspend_lock) and respect proper lock acquisition | |
526 | * ordering - zv_suspend_lock before zv_state_lock | |
527 | */ | |
528 | if (zv->zv_open_count == 0) { | |
529 | if (!rw_tryenter(&zv->zv_suspend_lock, RW_READER)) { | |
530 | mutex_exit(&zv->zv_state_lock); | |
531 | rw_enter(&zv->zv_suspend_lock, RW_READER); | |
532 | mutex_enter(&zv->zv_state_lock); | |
533 | /* check to see if zv_suspend_lock is needed */ | |
534 | if (zv->zv_open_count != 0) { | |
535 | rw_exit(&zv->zv_suspend_lock); | |
536 | } else { | |
537 | drop_suspend = B_TRUE; | |
538 | } | |
539 | } else { | |
540 | drop_suspend = B_TRUE; | |
541 | } | |
542 | } | |
543 | rw_exit(&zvol_state_lock); | |
544 | ||
545 | ASSERT(MUTEX_HELD(&zv->zv_state_lock)); | |
546 | ||
547 | if (zv->zv_open_count == 0) { | |
548 | boolean_t drop_namespace = B_FALSE; | |
549 | ||
550 | ASSERT(RW_READ_HELD(&zv->zv_suspend_lock)); | |
551 | ||
77e2756d BB |
552 | /* |
553 | * In all other call paths the spa_namespace_lock is taken | |
554 | * before the bdev->bd_mutex lock. However, on open(2) | |
555 | * the __blkdev_get() function calls fops->open() with the | |
556 | * bdev->bd_mutex lock held. This can result in a deadlock | |
557 | * when zvols from one pool are used as vdevs in another. | |
558 | * | |
559 | * To prevent a lock inversion deadlock we preemptively | |
560 | * take the spa_namespace_lock. Normally the lock will not | |
561 | * be contended and this is safe because spa_open_common() | |
562 | * handles the case where the caller already holds the | |
563 | * spa_namespace_lock. | |
564 | * | |
565 | * When the lock cannot be aquired after multiple retries | |
566 | * this must be the vdev on zvol deadlock case and we have | |
567 | * no choice but to return an error. For 5.12 and older | |
568 | * kernels returning -ERESTARTSYS will result in the | |
569 | * bdev->bd_mutex being dropped, then reacquired, and | |
570 | * fops->open() being called again. This process can be | |
571 | * repeated safely until both locks are acquired. For 5.13 | |
572 | * and newer the -ERESTARTSYS retry logic was removed from | |
573 | * the kernel so the only option is to return the error for | |
574 | * the caller to handle it. | |
575 | */ | |
8a02d01e BB |
576 | if (!mutex_owned(&spa_namespace_lock)) { |
577 | if (!mutex_tryenter(&spa_namespace_lock)) { | |
578 | mutex_exit(&zv->zv_state_lock); | |
579 | rw_exit(&zv->zv_suspend_lock); | |
77e2756d BB |
580 | |
581 | #ifdef HAVE_BLKDEV_GET_ERESTARTSYS | |
8a02d01e | 582 | schedule(); |
77e2756d | 583 | return (SET_ERROR(-ERESTARTSYS)); |
8a02d01e BB |
584 | #else |
585 | if ((gethrtime() - start) > timeout) | |
586 | return (SET_ERROR(-ERESTARTSYS)); | |
77e2756d | 587 | |
8a02d01e BB |
588 | schedule_timeout(MSEC_TO_TICK(10)); |
589 | goto retry; | |
77e2756d | 590 | #endif |
8a02d01e BB |
591 | } else { |
592 | drop_namespace = B_TRUE; | |
5df7e9d8 MM |
593 | } |
594 | } | |
5df7e9d8 | 595 | |
5df7e9d8 | 596 | error = -zvol_first_open(zv, !(flag & FMODE_WRITE)); |
5df7e9d8 | 597 | |
8a02d01e BB |
598 | if (drop_namespace) |
599 | mutex_exit(&spa_namespace_lock); | |
5df7e9d8 MM |
600 | } |
601 | ||
8a02d01e BB |
602 | if (error == 0) { |
603 | if ((flag & FMODE_WRITE) && (zv->zv_flags & ZVOL_RDONLY)) { | |
604 | if (zv->zv_open_count == 0) | |
605 | zvol_last_close(zv); | |
5df7e9d8 | 606 | |
8a02d01e BB |
607 | error = SET_ERROR(-EROFS); |
608 | } else { | |
609 | zv->zv_open_count++; | |
610 | } | |
611 | } | |
5df7e9d8 | 612 | |
5df7e9d8 MM |
613 | mutex_exit(&zv->zv_state_lock); |
614 | if (drop_suspend) | |
615 | rw_exit(&zv->zv_suspend_lock); | |
77e2756d | 616 | |
8a02d01e BB |
617 | if (error == 0) |
618 | zfs_check_media_change(bdev); | |
619 | ||
620 | return (error); | |
5df7e9d8 MM |
621 | } |
622 | ||
5df7e9d8 | 623 | static void |
5df7e9d8 MM |
624 | zvol_release(struct gendisk *disk, fmode_t mode) |
625 | { | |
626 | zvol_state_t *zv; | |
627 | boolean_t drop_suspend = B_TRUE; | |
628 | ||
629 | rw_enter(&zvol_state_lock, RW_READER); | |
630 | zv = disk->private_data; | |
631 | ||
632 | mutex_enter(&zv->zv_state_lock); | |
0b32d817 | 633 | ASSERT3U(zv->zv_open_count, >, 0); |
5df7e9d8 MM |
634 | /* |
635 | * make sure zvol is not suspended during last close | |
636 | * (hold zv_suspend_lock) and respect proper lock acquisition | |
637 | * ordering - zv_suspend_lock before zv_state_lock | |
638 | */ | |
639 | if (zv->zv_open_count == 1) { | |
640 | if (!rw_tryenter(&zv->zv_suspend_lock, RW_READER)) { | |
641 | mutex_exit(&zv->zv_state_lock); | |
642 | rw_enter(&zv->zv_suspend_lock, RW_READER); | |
643 | mutex_enter(&zv->zv_state_lock); | |
644 | /* check to see if zv_suspend_lock is needed */ | |
645 | if (zv->zv_open_count != 1) { | |
646 | rw_exit(&zv->zv_suspend_lock); | |
647 | drop_suspend = B_FALSE; | |
648 | } | |
649 | } | |
650 | } else { | |
651 | drop_suspend = B_FALSE; | |
652 | } | |
653 | rw_exit(&zvol_state_lock); | |
654 | ||
655 | ASSERT(MUTEX_HELD(&zv->zv_state_lock)); | |
5df7e9d8 MM |
656 | |
657 | zv->zv_open_count--; | |
0b32d817 RM |
658 | if (zv->zv_open_count == 0) { |
659 | ASSERT(RW_READ_HELD(&zv->zv_suspend_lock)); | |
5df7e9d8 | 660 | zvol_last_close(zv); |
0b32d817 | 661 | } |
5df7e9d8 MM |
662 | |
663 | mutex_exit(&zv->zv_state_lock); | |
664 | ||
665 | if (drop_suspend) | |
666 | rw_exit(&zv->zv_suspend_lock); | |
5df7e9d8 MM |
667 | } |
668 | ||
669 | static int | |
670 | zvol_ioctl(struct block_device *bdev, fmode_t mode, | |
671 | unsigned int cmd, unsigned long arg) | |
672 | { | |
673 | zvol_state_t *zv = bdev->bd_disk->private_data; | |
674 | int error = 0; | |
675 | ||
676 | ASSERT3U(zv->zv_open_count, >, 0); | |
677 | ||
678 | switch (cmd) { | |
679 | case BLKFLSBUF: | |
680 | fsync_bdev(bdev); | |
681 | invalidate_bdev(bdev); | |
682 | rw_enter(&zv->zv_suspend_lock, RW_READER); | |
683 | ||
684 | if (!(zv->zv_flags & ZVOL_RDONLY)) | |
685 | txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); | |
686 | ||
687 | rw_exit(&zv->zv_suspend_lock); | |
688 | break; | |
689 | ||
690 | case BLKZNAME: | |
691 | mutex_enter(&zv->zv_state_lock); | |
692 | error = copy_to_user((void *)arg, zv->zv_name, MAXNAMELEN); | |
693 | mutex_exit(&zv->zv_state_lock); | |
694 | break; | |
695 | ||
696 | default: | |
697 | error = -ENOTTY; | |
698 | break; | |
699 | } | |
700 | ||
701 | return (SET_ERROR(error)); | |
702 | } | |
703 | ||
704 | #ifdef CONFIG_COMPAT | |
705 | static int | |
706 | zvol_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
707 | unsigned cmd, unsigned long arg) | |
708 | { | |
709 | return (zvol_ioctl(bdev, mode, cmd, arg)); | |
710 | } | |
711 | #else | |
712 | #define zvol_compat_ioctl NULL | |
713 | #endif | |
714 | ||
5df7e9d8 MM |
715 | static unsigned int |
716 | zvol_check_events(struct gendisk *disk, unsigned int clearing) | |
717 | { | |
718 | unsigned int mask = 0; | |
719 | ||
720 | rw_enter(&zvol_state_lock, RW_READER); | |
721 | ||
722 | zvol_state_t *zv = disk->private_data; | |
723 | if (zv != NULL) { | |
724 | mutex_enter(&zv->zv_state_lock); | |
725 | mask = zv->zv_changed ? DISK_EVENT_MEDIA_CHANGE : 0; | |
726 | zv->zv_changed = 0; | |
727 | mutex_exit(&zv->zv_state_lock); | |
728 | } | |
729 | ||
730 | rw_exit(&zvol_state_lock); | |
731 | ||
732 | return (mask); | |
733 | } | |
5df7e9d8 MM |
734 | |
735 | static int | |
736 | zvol_revalidate_disk(struct gendisk *disk) | |
737 | { | |
738 | rw_enter(&zvol_state_lock, RW_READER); | |
739 | ||
740 | zvol_state_t *zv = disk->private_data; | |
741 | if (zv != NULL) { | |
742 | mutex_enter(&zv->zv_state_lock); | |
743 | set_capacity(zv->zv_zso->zvo_disk, | |
744 | zv->zv_volsize >> SECTOR_BITS); | |
745 | mutex_exit(&zv->zv_state_lock); | |
746 | } | |
747 | ||
748 | rw_exit(&zvol_state_lock); | |
749 | ||
750 | return (0); | |
751 | } | |
752 | ||
1dccfd7a CS |
753 | int |
754 | zvol_os_update_volsize(zvol_state_t *zv, uint64_t volsize) | |
5df7e9d8 | 755 | { |
1c0bbd52 | 756 | struct gendisk *disk = zv->zv_zso->zvo_disk; |
5df7e9d8 | 757 | |
19697e45 | 758 | #if defined(HAVE_REVALIDATE_DISK_SIZE) |
1c0bbd52 | 759 | revalidate_disk_size(disk, zvol_revalidate_disk(disk) == 0); |
19697e45 | 760 | #elif defined(HAVE_REVALIDATE_DISK) |
1c0bbd52 | 761 | revalidate_disk(disk); |
19697e45 BB |
762 | #else |
763 | zvol_revalidate_disk(disk); | |
59b68723 | 764 | #endif |
5df7e9d8 MM |
765 | return (0); |
766 | } | |
767 | ||
1dccfd7a CS |
768 | void |
769 | zvol_os_clear_private(zvol_state_t *zv) | |
5df7e9d8 MM |
770 | { |
771 | /* | |
772 | * Cleared while holding zvol_state_lock as a writer | |
773 | * which will prevent zvol_open() from opening it. | |
774 | */ | |
775 | zv->zv_zso->zvo_disk->private_data = NULL; | |
776 | } | |
777 | ||
778 | /* | |
779 | * Provide a simple virtual geometry for legacy compatibility. For devices | |
780 | * smaller than 1 MiB a small head and sector count is used to allow very | |
781 | * tiny devices. For devices over 1 Mib a standard head and sector count | |
782 | * is used to keep the cylinders count reasonable. | |
783 | */ | |
784 | static int | |
785 | zvol_getgeo(struct block_device *bdev, struct hd_geometry *geo) | |
786 | { | |
787 | zvol_state_t *zv = bdev->bd_disk->private_data; | |
788 | sector_t sectors; | |
789 | ||
790 | ASSERT3U(zv->zv_open_count, >, 0); | |
791 | ||
792 | sectors = get_capacity(zv->zv_zso->zvo_disk); | |
793 | ||
794 | if (sectors > 2048) { | |
795 | geo->heads = 16; | |
796 | geo->sectors = 63; | |
797 | } else { | |
798 | geo->heads = 2; | |
799 | geo->sectors = 4; | |
800 | } | |
801 | ||
802 | geo->start = 0; | |
803 | geo->cylinders = sectors / (geo->heads * geo->sectors); | |
804 | ||
805 | return (0); | |
806 | } | |
807 | ||
18168da7 | 808 | static const struct block_device_operations zvol_ops = { |
5df7e9d8 MM |
809 | .open = zvol_open, |
810 | .release = zvol_release, | |
811 | .ioctl = zvol_ioctl, | |
812 | .compat_ioctl = zvol_compat_ioctl, | |
5df7e9d8 | 813 | .check_events = zvol_check_events, |
48c7b0e4 | 814 | #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_REVALIDATE_DISK |
5df7e9d8 | 815 | .revalidate_disk = zvol_revalidate_disk, |
48c7b0e4 | 816 | #endif |
5df7e9d8 MM |
817 | .getgeo = zvol_getgeo, |
818 | .owner = THIS_MODULE, | |
d817c171 | 819 | #ifdef HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS |
1b06b03a | 820 | .submit_bio = zvol_submit_bio, |
d817c171 | 821 | #endif |
5df7e9d8 MM |
822 | }; |
823 | ||
824 | /* | |
825 | * Allocate memory for a new zvol_state_t and setup the required | |
826 | * request queue and generic disk structures for the block device. | |
827 | */ | |
828 | static zvol_state_t * | |
829 | zvol_alloc(dev_t dev, const char *name) | |
830 | { | |
831 | zvol_state_t *zv; | |
68dde63d | 832 | struct zvol_state_os *zso; |
5df7e9d8 MM |
833 | uint64_t volmode; |
834 | ||
835 | if (dsl_prop_get_integer(name, "volmode", &volmode, NULL) != 0) | |
836 | return (NULL); | |
837 | ||
838 | if (volmode == ZFS_VOLMODE_DEFAULT) | |
839 | volmode = zvol_volmode; | |
840 | ||
841 | if (volmode == ZFS_VOLMODE_NONE) | |
842 | return (NULL); | |
843 | ||
844 | zv = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); | |
68dde63d BB |
845 | zso = kmem_zalloc(sizeof (struct zvol_state_os), KM_SLEEP); |
846 | zv->zv_zso = zso; | |
0ca45cb3 | 847 | zv->zv_volmode = volmode; |
5df7e9d8 MM |
848 | |
849 | list_link_init(&zv->zv_next); | |
5df7e9d8 MM |
850 | mutex_init(&zv->zv_state_lock, NULL, MUTEX_DEFAULT, NULL); |
851 | ||
d817c171 | 852 | #ifdef HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS |
1b06b03a BB |
853 | #ifdef HAVE_BLK_ALLOC_DISK |
854 | zso->zvo_disk = blk_alloc_disk(NUMA_NO_NODE); | |
855 | if (zso->zvo_disk == NULL) | |
856 | goto out_kmem; | |
857 | ||
858 | zso->zvo_disk->minors = ZVOL_MINORS; | |
859 | zso->zvo_queue = zso->zvo_disk->queue; | |
860 | #else | |
d817c171 | 861 | zso->zvo_queue = blk_alloc_queue(NUMA_NO_NODE); |
1b06b03a BB |
862 | if (zso->zvo_queue == NULL) |
863 | goto out_kmem; | |
864 | ||
865 | zso->zvo_disk = alloc_disk(ZVOL_MINORS); | |
866 | if (zso->zvo_disk == NULL) { | |
867 | blk_cleanup_queue(zso->zvo_queue); | |
868 | goto out_kmem; | |
869 | } | |
870 | ||
871 | zso->zvo_disk->queue = zso->zvo_queue; | |
872 | #endif /* HAVE_BLK_ALLOC_DISK */ | |
d817c171 | 873 | #else |
68dde63d BB |
874 | zso->zvo_queue = blk_generic_alloc_queue(zvol_request, NUMA_NO_NODE); |
875 | if (zso->zvo_queue == NULL) | |
5df7e9d8 MM |
876 | goto out_kmem; |
877 | ||
1b06b03a BB |
878 | zso->zvo_disk = alloc_disk(ZVOL_MINORS); |
879 | if (zso->zvo_disk == NULL) { | |
880 | blk_cleanup_queue(zso->zvo_queue); | |
881 | goto out_kmem; | |
882 | } | |
883 | ||
884 | zso->zvo_disk->queue = zso->zvo_queue; | |
885 | #endif /* HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS */ | |
886 | ||
68dde63d | 887 | blk_queue_set_write_cache(zso->zvo_queue, B_TRUE, B_TRUE); |
5df7e9d8 MM |
888 | |
889 | /* Limit read-ahead to a single page to prevent over-prefetching. */ | |
68dde63d | 890 | blk_queue_set_read_ahead(zso->zvo_queue, 1); |
5df7e9d8 MM |
891 | |
892 | /* Disable write merging in favor of the ZIO pipeline. */ | |
68dde63d | 893 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, zso->zvo_queue); |
5df7e9d8 | 894 | |
ae1e40b3 BB |
895 | /* Enable /proc/diskstats */ |
896 | blk_queue_flag_set(QUEUE_FLAG_IO_STAT, zso->zvo_queue); | |
897 | ||
68dde63d BB |
898 | zso->zvo_queue->queuedata = zv; |
899 | zso->zvo_dev = dev; | |
5df7e9d8 MM |
900 | zv->zv_open_count = 0; |
901 | strlcpy(zv->zv_name, name, MAXNAMELEN); | |
902 | ||
2cc479d0 | 903 | zfs_rangelock_init(&zv->zv_rangelock, NULL, NULL); |
5df7e9d8 MM |
904 | rw_init(&zv->zv_suspend_lock, NULL, RW_DEFAULT, NULL); |
905 | ||
68dde63d BB |
906 | zso->zvo_disk->major = zvol_major; |
907 | zso->zvo_disk->events = DISK_EVENT_MEDIA_CHANGE; | |
5df7e9d8 MM |
908 | |
909 | if (volmode == ZFS_VOLMODE_DEV) { | |
910 | /* | |
911 | * ZFS_VOLMODE_DEV disable partitioning on ZVOL devices: set | |
86690775 | 912 | * gendisk->minors = 1 as noted in include/linux/blkdev.h. |
5df7e9d8 MM |
913 | * Also disable extended partition numbers (GENHD_FL_EXT_DEVT) |
914 | * and suppresses partition scanning (GENHD_FL_NO_PART_SCAN) | |
915 | * setting gendisk->flags accordingly. | |
916 | */ | |
68dde63d | 917 | zso->zvo_disk->minors = 1; |
5df7e9d8 | 918 | #if defined(GENHD_FL_EXT_DEVT) |
68dde63d | 919 | zso->zvo_disk->flags &= ~GENHD_FL_EXT_DEVT; |
5df7e9d8 MM |
920 | #endif |
921 | #if defined(GENHD_FL_NO_PART_SCAN) | |
68dde63d | 922 | zso->zvo_disk->flags |= GENHD_FL_NO_PART_SCAN; |
5df7e9d8 MM |
923 | #endif |
924 | } | |
68dde63d BB |
925 | zso->zvo_disk->first_minor = (dev & MINORMASK); |
926 | zso->zvo_disk->fops = &zvol_ops; | |
927 | zso->zvo_disk->private_data = zv; | |
68dde63d | 928 | snprintf(zso->zvo_disk->disk_name, DISK_NAME_LEN, "%s%d", |
5df7e9d8 MM |
929 | ZVOL_DEV_NAME, (dev & MINORMASK)); |
930 | ||
931 | return (zv); | |
932 | ||
5df7e9d8 | 933 | out_kmem: |
68dde63d | 934 | kmem_free(zso, sizeof (struct zvol_state_os)); |
5df7e9d8 MM |
935 | kmem_free(zv, sizeof (zvol_state_t)); |
936 | return (NULL); | |
937 | } | |
938 | ||
939 | /* | |
940 | * Cleanup then free a zvol_state_t which was created by zvol_alloc(). | |
941 | * At this time, the structure is not opened by anyone, is taken off | |
942 | * the zvol_state_list, and has its private data set to NULL. | |
943 | * The zvol_state_lock is dropped. | |
99573cc0 PS |
944 | * |
945 | * This function may take many milliseconds to complete (e.g. we've seen | |
946 | * it take over 256ms), due to the calls to "blk_cleanup_queue" and | |
947 | * "del_gendisk". Thus, consumers need to be careful to account for this | |
948 | * latency when calling this function. | |
5df7e9d8 | 949 | */ |
1dccfd7a CS |
950 | void |
951 | zvol_os_free(zvol_state_t *zv) | |
5df7e9d8 MM |
952 | { |
953 | ||
954 | ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock)); | |
955 | ASSERT(!MUTEX_HELD(&zv->zv_state_lock)); | |
0b32d817 RM |
956 | ASSERT0(zv->zv_open_count); |
957 | ASSERT3P(zv->zv_zso->zvo_disk->private_data, ==, NULL); | |
5df7e9d8 MM |
958 | |
959 | rw_destroy(&zv->zv_suspend_lock); | |
2cc479d0 | 960 | zfs_rangelock_fini(&zv->zv_rangelock); |
5df7e9d8 MM |
961 | |
962 | del_gendisk(zv->zv_zso->zvo_disk); | |
1b06b03a BB |
963 | #if defined(HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS) && \ |
964 | defined(HAVE_BLK_ALLOC_DISK) | |
965 | blk_cleanup_disk(zv->zv_zso->zvo_disk); | |
966 | #else | |
5df7e9d8 MM |
967 | blk_cleanup_queue(zv->zv_zso->zvo_queue); |
968 | put_disk(zv->zv_zso->zvo_disk); | |
1b06b03a | 969 | #endif |
5df7e9d8 MM |
970 | |
971 | ida_simple_remove(&zvol_ida, | |
972 | MINOR(zv->zv_zso->zvo_dev) >> ZVOL_MINOR_BITS); | |
973 | ||
974 | mutex_destroy(&zv->zv_state_lock); | |
4547fc4e | 975 | dataset_kstats_destroy(&zv->zv_kstat); |
5df7e9d8 MM |
976 | |
977 | kmem_free(zv->zv_zso, sizeof (struct zvol_state_os)); | |
978 | kmem_free(zv, sizeof (zvol_state_t)); | |
979 | } | |
980 | ||
0ca45cb3 MM |
981 | void |
982 | zvol_wait_close(zvol_state_t *zv) | |
983 | { | |
984 | } | |
985 | ||
5df7e9d8 MM |
986 | /* |
987 | * Create a block device minor node and setup the linkage between it | |
988 | * and the specified volume. Once this function returns the block | |
989 | * device is live and ready for use. | |
990 | */ | |
1dccfd7a | 991 | int |
ec213971 | 992 | zvol_os_create_minor(const char *name) |
5df7e9d8 MM |
993 | { |
994 | zvol_state_t *zv; | |
995 | objset_t *os; | |
996 | dmu_object_info_t *doi; | |
997 | uint64_t volsize; | |
998 | uint64_t len; | |
999 | unsigned minor = 0; | |
1000 | int error = 0; | |
1001 | int idx; | |
1002 | uint64_t hash = zvol_name_hash(name); | |
1003 | ||
1004 | if (zvol_inhibit_dev) | |
1005 | return (0); | |
1006 | ||
1007 | idx = ida_simple_get(&zvol_ida, 0, 0, kmem_flags_convert(KM_SLEEP)); | |
1008 | if (idx < 0) | |
1009 | return (SET_ERROR(-idx)); | |
1010 | minor = idx << ZVOL_MINOR_BITS; | |
1011 | ||
1012 | zv = zvol_find_by_name_hash(name, hash, RW_NONE); | |
1013 | if (zv) { | |
1014 | ASSERT(MUTEX_HELD(&zv->zv_state_lock)); | |
1015 | mutex_exit(&zv->zv_state_lock); | |
1016 | ida_simple_remove(&zvol_ida, idx); | |
1017 | return (SET_ERROR(EEXIST)); | |
1018 | } | |
1019 | ||
1020 | doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP); | |
1021 | ||
1022 | error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, B_TRUE, FTAG, &os); | |
1023 | if (error) | |
1024 | goto out_doi; | |
1025 | ||
1026 | error = dmu_object_info(os, ZVOL_OBJ, doi); | |
1027 | if (error) | |
1028 | goto out_dmu_objset_disown; | |
1029 | ||
1030 | error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); | |
1031 | if (error) | |
1032 | goto out_dmu_objset_disown; | |
1033 | ||
1034 | zv = zvol_alloc(MKDEV(zvol_major, minor), name); | |
1035 | if (zv == NULL) { | |
1036 | error = SET_ERROR(EAGAIN); | |
1037 | goto out_dmu_objset_disown; | |
1038 | } | |
1039 | zv->zv_hash = hash; | |
1040 | ||
1041 | if (dmu_objset_is_snapshot(os)) | |
1042 | zv->zv_flags |= ZVOL_RDONLY; | |
1043 | ||
1044 | zv->zv_volblocksize = doi->doi_data_block_size; | |
1045 | zv->zv_volsize = volsize; | |
1046 | zv->zv_objset = os; | |
1047 | ||
1048 | set_capacity(zv->zv_zso->zvo_disk, zv->zv_volsize >> 9); | |
1049 | ||
1050 | blk_queue_max_hw_sectors(zv->zv_zso->zvo_queue, | |
1051 | (DMU_MAX_ACCESS / 4) >> 9); | |
1052 | blk_queue_max_segments(zv->zv_zso->zvo_queue, UINT16_MAX); | |
1053 | blk_queue_max_segment_size(zv->zv_zso->zvo_queue, UINT_MAX); | |
1054 | blk_queue_physical_block_size(zv->zv_zso->zvo_queue, | |
1055 | zv->zv_volblocksize); | |
1056 | blk_queue_io_opt(zv->zv_zso->zvo_queue, zv->zv_volblocksize); | |
1057 | blk_queue_max_discard_sectors(zv->zv_zso->zvo_queue, | |
1058 | (zvol_max_discard_blocks * zv->zv_volblocksize) >> 9); | |
1059 | blk_queue_discard_granularity(zv->zv_zso->zvo_queue, | |
1060 | zv->zv_volblocksize); | |
1061 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, zv->zv_zso->zvo_queue); | |
1062 | #ifdef QUEUE_FLAG_NONROT | |
1063 | blk_queue_flag_set(QUEUE_FLAG_NONROT, zv->zv_zso->zvo_queue); | |
1064 | #endif | |
1065 | #ifdef QUEUE_FLAG_ADD_RANDOM | |
1066 | blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zv->zv_zso->zvo_queue); | |
1067 | #endif | |
1068 | /* This flag was introduced in kernel version 4.12. */ | |
1069 | #ifdef QUEUE_FLAG_SCSI_PASSTHROUGH | |
1070 | blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, zv->zv_zso->zvo_queue); | |
1071 | #endif | |
1072 | ||
93e36580 CS |
1073 | ASSERT3P(zv->zv_zilog, ==, NULL); |
1074 | zv->zv_zilog = zil_open(os, zvol_get_data); | |
5df7e9d8 MM |
1075 | if (spa_writeable(dmu_objset_spa(os))) { |
1076 | if (zil_replay_disable) | |
93e36580 | 1077 | zil_destroy(zv->zv_zilog, B_FALSE); |
5df7e9d8 MM |
1078 | else |
1079 | zil_replay(os, zv, zvol_replay_vector); | |
1080 | } | |
93e36580 CS |
1081 | zil_close(zv->zv_zilog); |
1082 | zv->zv_zilog = NULL; | |
4547fc4e AJ |
1083 | ASSERT3P(zv->zv_kstat.dk_kstats, ==, NULL); |
1084 | dataset_kstats_create(&zv->zv_kstat, zv->zv_objset); | |
5df7e9d8 MM |
1085 | |
1086 | /* | |
1087 | * When udev detects the addition of the device it will immediately | |
1088 | * invoke blkid(8) to determine the type of content on the device. | |
1089 | * Prefetching the blocks commonly scanned by blkid(8) will speed | |
1090 | * up this process. | |
1091 | */ | |
1092 | len = MIN(MAX(zvol_prefetch_bytes, 0), SPA_MAXBLOCKSIZE); | |
1093 | if (len > 0) { | |
1094 | dmu_prefetch(os, ZVOL_OBJ, 0, 0, len, ZIO_PRIORITY_SYNC_READ); | |
1095 | dmu_prefetch(os, ZVOL_OBJ, 0, volsize - len, len, | |
1096 | ZIO_PRIORITY_SYNC_READ); | |
1097 | } | |
1098 | ||
1099 | zv->zv_objset = NULL; | |
1100 | out_dmu_objset_disown: | |
1101 | dmu_objset_disown(os, B_TRUE, FTAG); | |
1102 | out_doi: | |
1103 | kmem_free(doi, sizeof (dmu_object_info_t)); | |
1104 | ||
1105 | /* | |
1106 | * Keep in mind that once add_disk() is called, the zvol is | |
1107 | * announced to the world, and zvol_open()/zvol_release() can | |
1108 | * be called at any time. Incidentally, add_disk() itself calls | |
1109 | * zvol_open()->zvol_first_open() and zvol_release()->zvol_last_close() | |
1110 | * directly as well. | |
1111 | */ | |
1112 | if (error == 0) { | |
1113 | rw_enter(&zvol_state_lock, RW_WRITER); | |
1114 | zvol_insert(zv); | |
1115 | rw_exit(&zvol_state_lock); | |
12fa250d RE |
1116 | #ifdef HAVE_ADD_DISK_RET |
1117 | error = add_disk(zv->zv_zso->zvo_disk); | |
1118 | #else | |
5df7e9d8 | 1119 | add_disk(zv->zv_zso->zvo_disk); |
12fa250d | 1120 | #endif |
5df7e9d8 MM |
1121 | } else { |
1122 | ida_simple_remove(&zvol_ida, idx); | |
1123 | } | |
1124 | ||
ec213971 | 1125 | return (error); |
5df7e9d8 MM |
1126 | } |
1127 | ||
1dccfd7a CS |
1128 | void |
1129 | zvol_os_rename_minor(zvol_state_t *zv, const char *newname) | |
5df7e9d8 MM |
1130 | { |
1131 | int readonly = get_disk_ro(zv->zv_zso->zvo_disk); | |
1132 | ||
1133 | ASSERT(RW_LOCK_HELD(&zvol_state_lock)); | |
1134 | ASSERT(MUTEX_HELD(&zv->zv_state_lock)); | |
1135 | ||
1136 | strlcpy(zv->zv_name, newname, sizeof (zv->zv_name)); | |
1137 | ||
1138 | /* move to new hashtable entry */ | |
1139 | zv->zv_hash = zvol_name_hash(zv->zv_name); | |
1140 | hlist_del(&zv->zv_hlink); | |
1141 | hlist_add_head(&zv->zv_hlink, ZVOL_HT_HEAD(zv->zv_hash)); | |
1142 | ||
1143 | /* | |
1144 | * The block device's read-only state is briefly changed causing | |
1145 | * a KOBJ_CHANGE uevent to be issued. This ensures udev detects | |
1146 | * the name change and fixes the symlinks. This does not change | |
1147 | * ZVOL_RDONLY in zv->zv_flags so the actual read-only state never | |
1148 | * changes. This would normally be done using kobject_uevent() but | |
1149 | * that is a GPL-only symbol which is why we need this workaround. | |
1150 | */ | |
1151 | set_disk_ro(zv->zv_zso->zvo_disk, !readonly); | |
1152 | set_disk_ro(zv->zv_zso->zvo_disk, readonly); | |
1153 | } | |
1154 | ||
1dccfd7a CS |
1155 | void |
1156 | zvol_os_set_disk_ro(zvol_state_t *zv, int flags) | |
5df7e9d8 MM |
1157 | { |
1158 | ||
1159 | set_disk_ro(zv->zv_zso->zvo_disk, flags); | |
1160 | } | |
1161 | ||
1dccfd7a CS |
1162 | void |
1163 | zvol_os_set_capacity(zvol_state_t *zv, uint64_t capacity) | |
5df7e9d8 MM |
1164 | { |
1165 | ||
1166 | set_capacity(zv->zv_zso->zvo_disk, capacity); | |
1167 | } | |
1168 | ||
5df7e9d8 MM |
1169 | int |
1170 | zvol_init(void) | |
1171 | { | |
1172 | int error; | |
1173 | int threads = MIN(MAX(zvol_threads, 1), 1024); | |
1174 | ||
1175 | error = register_blkdev(zvol_major, ZVOL_DRIVER); | |
1176 | if (error) { | |
1177 | printk(KERN_INFO "ZFS: register_blkdev() failed %d\n", error); | |
1178 | return (error); | |
1179 | } | |
1180 | zvol_taskq = taskq_create(ZVOL_DRIVER, threads, maxclsyspri, | |
1181 | threads * 2, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC); | |
1182 | if (zvol_taskq == NULL) { | |
1183 | unregister_blkdev(zvol_major, ZVOL_DRIVER); | |
1184 | return (-ENOMEM); | |
1185 | } | |
1186 | zvol_init_impl(); | |
5df7e9d8 | 1187 | ida_init(&zvol_ida); |
5df7e9d8 MM |
1188 | return (0); |
1189 | } | |
1190 | ||
1191 | void | |
1192 | zvol_fini(void) | |
1193 | { | |
5df7e9d8 | 1194 | zvol_fini_impl(); |
5df7e9d8 MM |
1195 | unregister_blkdev(zvol_major, ZVOL_DRIVER); |
1196 | taskq_destroy(zvol_taskq); | |
1197 | ida_destroy(&zvol_ida); | |
1198 | } | |
1199 | ||
1200 | /* BEGIN CSTYLED */ | |
1201 | module_param(zvol_inhibit_dev, uint, 0644); | |
1202 | MODULE_PARM_DESC(zvol_inhibit_dev, "Do not create zvol device nodes"); | |
1203 | ||
1204 | module_param(zvol_major, uint, 0444); | |
1205 | MODULE_PARM_DESC(zvol_major, "Major number for zvol device"); | |
1206 | ||
1207 | module_param(zvol_threads, uint, 0444); | |
1208 | MODULE_PARM_DESC(zvol_threads, "Max number of threads to handle I/O requests"); | |
1209 | ||
1210 | module_param(zvol_request_sync, uint, 0644); | |
1211 | MODULE_PARM_DESC(zvol_request_sync, "Synchronously handle bio requests"); | |
1212 | ||
1213 | module_param(zvol_max_discard_blocks, ulong, 0444); | |
1214 | MODULE_PARM_DESC(zvol_max_discard_blocks, "Max number of blocks to discard"); | |
1215 | ||
1216 | module_param(zvol_prefetch_bytes, uint, 0644); | |
1217 | MODULE_PARM_DESC(zvol_prefetch_bytes, "Prefetch N bytes at zvol start+end"); | |
1218 | ||
1219 | module_param(zvol_volmode, uint, 0644); | |
1220 | MODULE_PARM_DESC(zvol_volmode, "Default volmode property value"); | |
1221 | /* END CSTYLED */ |