Print this page
NEX-19083 backport OS-7314 zil_commit should omit cache thrash
9962 zil_commit should omit cache thrash
Reviewed by: Matt Ahrens <matt@delphix.com>
Reviewed by: Brad Lewis <brad.lewis@delphix.com>
Reviewed by: Patrick Mooney <patrick.mooney@joyent.com>
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Approved by: Joshua M. Clulow <josh@sysmgr.org>
NEX-5367 special vdev: sync-write options (NEW)
Reviewed by: Roman Strashkin <roman.strashkin@nexenta.com>
Reviewed by: Alek Pinchuk <alek.pinchuk@nexenta.com>
6250 zvol_dump_init() can hold txg open
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Prakash Surya <prakash.surya@delphix.com>
Reviewed by: Albert Lee <trisk@omniti.com>
Reviewed by: Xin Li <delphij@freebsd.org>
Approved by: Garrett D'Amore <garrett@damore.org>
NEX-4582 update wrc test cases for allow to use write back cache per tree of datasets
Reviewed by: Steve Peng <steve.peng@nexenta.com>
Reviewed by: Alex Aizman <alex.aizman@nexenta.com>
5960 zfs recv should prefetch indirect blocks
5925 zfs receive -o origin=
Reviewed by: Prakash Surya <prakash.surya@delphix.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
NEX-3508 CLONE - Port NEX-2946 Add UNMAP/TRIM functionality to ZFS and illumos
Reviewed by: Josef Sipek <josef.sipek@nexenta.com>
Reviewed by: Alek Pinchuk <alek.pinchuk@nexenta.com>
Conflicts:
usr/src/uts/common/io/scsi/targets/sd.c
usr/src/uts/common/sys/scsi/targets/sddef.h
NEX-1142 move rwlock to vdev to protect vdev_tsd
not just ldi handle.
This way we serialize open/close, yet allow parallel I/O.
4370 avoid transmitting holes during zfs send
4371 DMU code clean up
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>
Approved by: Garrett D'Amore <garrett@damore.org>
NEX-1065 Added serialization to avoid race
between ldi notification and I/O path.
Also fixes OS-124, NEX-1051, NEX-1062.
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/zvol.c
+++ new/usr/src/uts/common/fs/zfs/zvol.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
|
↓ open down ↓ |
18 lines elided |
↑ open up ↑ |
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 *
24 24 * Portions Copyright 2010 Robert Milkowski
25 25 *
26 26 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
27 27 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
28 28 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
29 + * Copyright 2017 Nexenta Systems, Inc. All rights reserved.
29 30 * Copyright (c) 2014 Integros [integros.com]
30 31 */
31 32
32 33 /*
33 34 * ZFS volume emulation driver.
34 35 *
35 36 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
36 37 * Volumes are accessed through the symbolic links named:
37 38 *
38 39 * /dev/zvol/dsk/<pool_name>/<dataset_name>
39 40 * /dev/zvol/rdsk/<pool_name>/<dataset_name>
40 41 *
41 42 * These links are created by the /dev filesystem (sdev_zvolops.c).
42 43 * Volumes are persistent through reboot. No user command needs to be
43 44 * run before opening and using a device.
44 45 */
45 46
46 47 #include <sys/types.h>
47 48 #include <sys/param.h>
48 49 #include <sys/errno.h>
49 50 #include <sys/uio.h>
50 51 #include <sys/buf.h>
51 52 #include <sys/modctl.h>
52 53 #include <sys/open.h>
53 54 #include <sys/kmem.h>
54 55 #include <sys/conf.h>
55 56 #include <sys/cmn_err.h>
56 57 #include <sys/stat.h>
57 58 #include <sys/zap.h>
58 59 #include <sys/spa.h>
59 60 #include <sys/spa_impl.h>
60 61 #include <sys/zio.h>
61 62 #include <sys/dmu_traverse.h>
62 63 #include <sys/dnode.h>
63 64 #include <sys/dsl_dataset.h>
64 65 #include <sys/dsl_prop.h>
65 66 #include <sys/dkio.h>
66 67 #include <sys/efi_partition.h>
67 68 #include <sys/byteorder.h>
68 69 #include <sys/pathname.h>
69 70 #include <sys/ddi.h>
70 71 #include <sys/sunddi.h>
71 72 #include <sys/crc32.h>
72 73 #include <sys/dirent.h>
73 74 #include <sys/policy.h>
74 75 #include <sys/fs/zfs.h>
75 76 #include <sys/zfs_ioctl.h>
76 77 #include <sys/mkdev.h>
77 78 #include <sys/zil.h>
78 79 #include <sys/refcount.h>
79 80 #include <sys/zfs_znode.h>
80 81 #include <sys/zfs_rlock.h>
|
↓ open down ↓ |
42 lines elided |
↑ open up ↑ |
81 82 #include <sys/vdev_disk.h>
82 83 #include <sys/vdev_impl.h>
83 84 #include <sys/vdev_raidz.h>
84 85 #include <sys/zvol.h>
85 86 #include <sys/dumphdr.h>
86 87 #include <sys/zil_impl.h>
87 88 #include <sys/dbuf.h>
88 89 #include <sys/dmu_tx.h>
89 90 #include <sys/zfeature.h>
90 91 #include <sys/zio_checksum.h>
92 +#include <sys/dkioc_free_util.h>
91 93 #include <sys/zil_impl.h>
92 94
93 95 #include "zfs_namecheck.h"
94 96
95 97 void *zfsdev_state;
96 98 static char *zvol_tag = "zvol_tag";
97 99
98 100 #define ZVOL_DUMPSIZE "dumpsize"
99 101
100 102 /*
101 103 * This lock protects the zfsdev_state structure from being modified
102 104 * while it's being used, e.g. an open that comes in before a create
103 105 * finishes. It also protects temporary opens of the dataset so that,
104 106 * e.g., an open doesn't get a spurious EBUSY.
105 107 */
106 108 kmutex_t zfsdev_state_lock;
107 109 static uint32_t zvol_minors;
108 110
109 111 typedef struct zvol_extent {
110 112 list_node_t ze_node;
111 113 dva_t ze_dva; /* dva associated with this extent */
112 114 uint64_t ze_nblks; /* number of blocks in extent */
113 115 } zvol_extent_t;
114 116
115 117 /*
116 118 * The in-core state of each volume.
117 119 */
118 120 typedef struct zvol_state {
119 121 char zv_name[MAXPATHLEN]; /* pool/dd name */
120 122 uint64_t zv_volsize; /* amount of space we advertise */
121 123 uint64_t zv_volblocksize; /* volume block size */
122 124 minor_t zv_minor; /* minor number */
123 125 uint8_t zv_min_bs; /* minimum addressable block shift */
124 126 uint8_t zv_flags; /* readonly, dumpified, etc. */
125 127 objset_t *zv_objset; /* objset handle */
126 128 uint32_t zv_open_count[OTYPCNT]; /* open counts */
127 129 uint32_t zv_total_opens; /* total open count */
128 130 zilog_t *zv_zilog; /* ZIL handle */
129 131 list_t zv_extents; /* List of extents for dump */
130 132 znode_t zv_znode; /* for range locking */
131 133 dmu_buf_t *zv_dbuf; /* bonus handle */
132 134 } zvol_state_t;
133 135
134 136 /*
135 137 * zvol specific flags
136 138 */
137 139 #define ZVOL_RDONLY 0x1
138 140 #define ZVOL_DUMPIFIED 0x2
139 141 #define ZVOL_EXCL 0x4
140 142 #define ZVOL_WCE 0x8
141 143
142 144 /*
143 145 * zvol maximum transfer in one DMU tx.
144 146 */
145 147 int zvol_maxphys = DMU_MAX_ACCESS/2;
146 148
147 149 /*
148 150 * Toggle unmap functionality.
149 151 */
150 152 boolean_t zvol_unmap_enabled = B_TRUE;
151 153
152 154 /*
153 155 * If true, unmaps requested as synchronous are executed synchronously,
154 156 * otherwise all unmaps are asynchronous.
155 157 */
156 158 boolean_t zvol_unmap_sync_enabled = B_FALSE;
157 159
158 160 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
159 161 nvlist_t *, nvlist_t *);
160 162 static int zvol_remove_zv(zvol_state_t *);
161 163 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf,
162 164 struct lwb *lwb, zio_t *zio);
163 165 static int zvol_dumpify(zvol_state_t *zv);
164 166 static int zvol_dump_fini(zvol_state_t *zv);
165 167 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
166 168
167 169 static void
168 170 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
169 171 {
170 172 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
171 173
172 174 zv->zv_volsize = volsize;
173 175 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
174 176 "Size", volsize) == DDI_SUCCESS);
175 177 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
176 178 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
177 179
178 180 /* Notify specfs to invalidate the cached size */
179 181 spec_size_invalidate(dev, VBLK);
180 182 spec_size_invalidate(dev, VCHR);
181 183 }
182 184
183 185 int
184 186 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
185 187 {
186 188 if (volsize == 0)
187 189 return (SET_ERROR(EINVAL));
188 190
189 191 if (volsize % blocksize != 0)
190 192 return (SET_ERROR(EINVAL));
191 193
192 194 #ifdef _ILP32
193 195 if (volsize - 1 > SPEC_MAXOFFSET_T)
194 196 return (SET_ERROR(EOVERFLOW));
195 197 #endif
196 198 return (0);
197 199 }
198 200
199 201 int
200 202 zvol_check_volblocksize(uint64_t volblocksize)
201 203 {
202 204 if (volblocksize < SPA_MINBLOCKSIZE ||
203 205 volblocksize > SPA_OLD_MAXBLOCKSIZE ||
204 206 !ISP2(volblocksize))
205 207 return (SET_ERROR(EDOM));
206 208
207 209 return (0);
208 210 }
209 211
210 212 int
211 213 zvol_get_stats(objset_t *os, nvlist_t *nv)
212 214 {
213 215 int error;
214 216 dmu_object_info_t doi;
215 217 uint64_t val;
216 218
217 219 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
218 220 if (error)
219 221 return (error);
220 222
221 223 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
222 224
223 225 error = dmu_object_info(os, ZVOL_OBJ, &doi);
224 226
225 227 if (error == 0) {
226 228 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
227 229 doi.doi_data_block_size);
228 230 }
229 231
230 232 return (error);
231 233 }
232 234
233 235 static zvol_state_t *
234 236 zvol_minor_lookup(const char *name)
235 237 {
236 238 minor_t minor;
237 239 zvol_state_t *zv;
238 240
239 241 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
240 242
241 243 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
242 244 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
243 245 if (zv == NULL)
244 246 continue;
245 247 if (strcmp(zv->zv_name, name) == 0)
246 248 return (zv);
247 249 }
248 250
249 251 return (NULL);
250 252 }
251 253
252 254 /* extent mapping arg */
253 255 struct maparg {
254 256 zvol_state_t *ma_zv;
255 257 uint64_t ma_blks;
256 258 };
257 259
258 260 /*ARGSUSED*/
259 261 static int
260 262 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
261 263 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
262 264 {
263 265 struct maparg *ma = arg;
264 266 zvol_extent_t *ze;
265 267 int bs = ma->ma_zv->zv_volblocksize;
266 268
267 269 if (bp == NULL || BP_IS_HOLE(bp) ||
268 270 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
269 271 return (0);
270 272
271 273 VERIFY(!BP_IS_EMBEDDED(bp));
272 274
273 275 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
274 276 ma->ma_blks++;
275 277
276 278 /* Abort immediately if we have encountered gang blocks */
277 279 if (BP_IS_GANG(bp))
278 280 return (SET_ERROR(EFRAGS));
279 281
280 282 /*
281 283 * See if the block is at the end of the previous extent.
282 284 */
283 285 ze = list_tail(&ma->ma_zv->zv_extents);
284 286 if (ze &&
285 287 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
286 288 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
287 289 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
288 290 ze->ze_nblks++;
289 291 return (0);
290 292 }
291 293
292 294 dprintf_bp(bp, "%s", "next blkptr:");
293 295
294 296 /* start a new extent */
295 297 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
296 298 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
297 299 ze->ze_nblks = 1;
298 300 list_insert_tail(&ma->ma_zv->zv_extents, ze);
299 301 return (0);
300 302 }
301 303
302 304 static void
303 305 zvol_free_extents(zvol_state_t *zv)
304 306 {
305 307 zvol_extent_t *ze;
306 308
307 309 while (ze = list_head(&zv->zv_extents)) {
308 310 list_remove(&zv->zv_extents, ze);
309 311 kmem_free(ze, sizeof (zvol_extent_t));
310 312 }
311 313 }
312 314
313 315 static int
314 316 zvol_get_lbas(zvol_state_t *zv)
315 317 {
316 318 objset_t *os = zv->zv_objset;
317 319 struct maparg ma;
318 320 int err;
319 321
320 322 ma.ma_zv = zv;
321 323 ma.ma_blks = 0;
322 324 zvol_free_extents(zv);
323 325
324 326 /* commit any in-flight changes before traversing the dataset */
325 327 txg_wait_synced(dmu_objset_pool(os), 0);
326 328 err = traverse_dataset(dmu_objset_ds(os), 0,
327 329 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
328 330 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
329 331 zvol_free_extents(zv);
330 332 return (err ? err : EIO);
331 333 }
332 334
333 335 return (0);
334 336 }
335 337
336 338 /* ARGSUSED */
337 339 void
338 340 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
339 341 {
340 342 zfs_creat_t *zct = arg;
341 343 nvlist_t *nvprops = zct->zct_props;
342 344 int error;
343 345 uint64_t volblocksize, volsize;
344 346
345 347 VERIFY(nvlist_lookup_uint64(nvprops,
346 348 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
347 349 if (nvlist_lookup_uint64(nvprops,
348 350 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
349 351 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
350 352
351 353 /*
352 354 * These properties must be removed from the list so the generic
353 355 * property setting step won't apply to them.
354 356 */
355 357 VERIFY(nvlist_remove_all(nvprops,
356 358 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
357 359 (void) nvlist_remove_all(nvprops,
358 360 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
359 361
360 362 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
361 363 DMU_OT_NONE, 0, tx);
362 364 ASSERT(error == 0);
363 365
364 366 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
365 367 DMU_OT_NONE, 0, tx);
366 368 ASSERT(error == 0);
367 369
368 370 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
369 371 ASSERT(error == 0);
370 372 }
371 373
372 374 /*
373 375 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
374 376 * implement DKIOCFREE/free-long-range.
375 377 */
376 378 static int
377 379 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
378 380 {
379 381 zvol_state_t *zv = arg1;
380 382 lr_truncate_t *lr = arg2;
381 383 uint64_t offset, length;
382 384
383 385 if (byteswap)
384 386 byteswap_uint64_array(lr, sizeof (*lr));
385 387
386 388 offset = lr->lr_offset;
387 389 length = lr->lr_length;
388 390
389 391 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
390 392 }
391 393
392 394 /*
393 395 * Replay a TX_WRITE ZIL transaction that didn't get committed
394 396 * after a system failure
395 397 */
396 398 static int
397 399 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
398 400 {
399 401 zvol_state_t *zv = arg1;
400 402 lr_write_t *lr = arg2;
401 403 objset_t *os = zv->zv_objset;
402 404 char *data = (char *)(lr + 1); /* data follows lr_write_t */
403 405 uint64_t offset, length;
404 406 dmu_tx_t *tx;
405 407 int error;
406 408
407 409 if (byteswap)
408 410 byteswap_uint64_array(lr, sizeof (*lr));
409 411
410 412 offset = lr->lr_offset;
411 413 length = lr->lr_length;
412 414
413 415 /* If it's a dmu_sync() block, write the whole block */
414 416 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
415 417 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
416 418 if (length < blocksize) {
417 419 offset -= offset % blocksize;
418 420 length = blocksize;
419 421 }
420 422 }
421 423
422 424 tx = dmu_tx_create(os);
423 425 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
424 426 error = dmu_tx_assign(tx, TXG_WAIT);
425 427 if (error) {
426 428 dmu_tx_abort(tx);
427 429 } else {
428 430 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
429 431 dmu_tx_commit(tx);
430 432 }
431 433
432 434 return (error);
433 435 }
434 436
435 437 /* ARGSUSED */
436 438 static int
437 439 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
438 440 {
439 441 return (SET_ERROR(ENOTSUP));
440 442 }
441 443
442 444 /*
443 445 * Callback vectors for replaying records.
444 446 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
445 447 */
446 448 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
447 449 zvol_replay_err, /* 0 no such transaction type */
448 450 zvol_replay_err, /* TX_CREATE */
449 451 zvol_replay_err, /* TX_MKDIR */
450 452 zvol_replay_err, /* TX_MKXATTR */
451 453 zvol_replay_err, /* TX_SYMLINK */
452 454 zvol_replay_err, /* TX_REMOVE */
453 455 zvol_replay_err, /* TX_RMDIR */
454 456 zvol_replay_err, /* TX_LINK */
455 457 zvol_replay_err, /* TX_RENAME */
456 458 zvol_replay_write, /* TX_WRITE */
457 459 zvol_replay_truncate, /* TX_TRUNCATE */
458 460 zvol_replay_err, /* TX_SETATTR */
459 461 zvol_replay_err, /* TX_ACL */
460 462 zvol_replay_err, /* TX_CREATE_ACL */
461 463 zvol_replay_err, /* TX_CREATE_ATTR */
462 464 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
463 465 zvol_replay_err, /* TX_MKDIR_ACL */
464 466 zvol_replay_err, /* TX_MKDIR_ATTR */
465 467 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
466 468 zvol_replay_err, /* TX_WRITE2 */
467 469 };
468 470
469 471 int
470 472 zvol_name2minor(const char *name, minor_t *minor)
471 473 {
472 474 zvol_state_t *zv;
473 475
474 476 mutex_enter(&zfsdev_state_lock);
475 477 zv = zvol_minor_lookup(name);
476 478 if (minor && zv)
477 479 *minor = zv->zv_minor;
478 480 mutex_exit(&zfsdev_state_lock);
479 481 return (zv ? 0 : -1);
480 482 }
481 483
482 484 /*
483 485 * Create a minor node (plus a whole lot more) for the specified volume.
484 486 */
485 487 int
486 488 zvol_create_minor(const char *name)
487 489 {
488 490 zfs_soft_state_t *zs;
489 491 zvol_state_t *zv;
490 492 objset_t *os;
491 493 dmu_object_info_t doi;
492 494 minor_t minor = 0;
493 495 char chrbuf[30], blkbuf[30];
494 496 int error;
495 497
496 498 mutex_enter(&zfsdev_state_lock);
497 499
498 500 if (zvol_minor_lookup(name) != NULL) {
499 501 mutex_exit(&zfsdev_state_lock);
500 502 return (SET_ERROR(EEXIST));
501 503 }
502 504
503 505 /* lie and say we're read-only */
504 506 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
505 507
506 508 if (error) {
507 509 mutex_exit(&zfsdev_state_lock);
508 510 return (error);
509 511 }
510 512
511 513 if ((minor = zfsdev_minor_alloc()) == 0) {
512 514 dmu_objset_disown(os, FTAG);
513 515 mutex_exit(&zfsdev_state_lock);
514 516 return (SET_ERROR(ENXIO));
515 517 }
516 518
517 519 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
518 520 dmu_objset_disown(os, FTAG);
519 521 mutex_exit(&zfsdev_state_lock);
520 522 return (SET_ERROR(EAGAIN));
521 523 }
522 524 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
523 525 (char *)name);
524 526
525 527 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
526 528
527 529 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
528 530 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
529 531 ddi_soft_state_free(zfsdev_state, minor);
530 532 dmu_objset_disown(os, FTAG);
531 533 mutex_exit(&zfsdev_state_lock);
532 534 return (SET_ERROR(EAGAIN));
533 535 }
534 536
535 537 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
536 538
537 539 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
538 540 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
539 541 ddi_remove_minor_node(zfs_dip, chrbuf);
540 542 ddi_soft_state_free(zfsdev_state, minor);
541 543 dmu_objset_disown(os, FTAG);
542 544 mutex_exit(&zfsdev_state_lock);
543 545 return (SET_ERROR(EAGAIN));
544 546 }
545 547
546 548 zs = ddi_get_soft_state(zfsdev_state, minor);
547 549 zs->zss_type = ZSST_ZVOL;
548 550 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
549 551 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
550 552 zv->zv_min_bs = DEV_BSHIFT;
551 553 zv->zv_minor = minor;
552 554 zv->zv_objset = os;
553 555 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
554 556 zv->zv_flags |= ZVOL_RDONLY;
555 557 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
556 558 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
557 559 sizeof (rl_t), offsetof(rl_t, r_node));
558 560 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
559 561 offsetof(zvol_extent_t, ze_node));
560 562 /* get and cache the blocksize */
561 563 error = dmu_object_info(os, ZVOL_OBJ, &doi);
562 564 ASSERT(error == 0);
563 565 zv->zv_volblocksize = doi.doi_data_block_size;
564 566
565 567 if (spa_writeable(dmu_objset_spa(os))) {
566 568 if (zil_replay_disable)
567 569 zil_destroy(dmu_objset_zil(os), B_FALSE);
568 570 else
569 571 zil_replay(os, zv, zvol_replay_vector);
570 572 }
571 573 dmu_objset_disown(os, FTAG);
572 574 zv->zv_objset = NULL;
573 575
574 576 zvol_minors++;
575 577
576 578 mutex_exit(&zfsdev_state_lock);
577 579
578 580 return (0);
579 581 }
580 582
581 583 /*
582 584 * Remove minor node for the specified volume.
583 585 */
584 586 static int
585 587 zvol_remove_zv(zvol_state_t *zv)
586 588 {
587 589 char nmbuf[20];
588 590 minor_t minor = zv->zv_minor;
589 591
590 592 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
591 593 if (zv->zv_total_opens != 0)
592 594 return (SET_ERROR(EBUSY));
593 595
594 596 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
595 597 ddi_remove_minor_node(zfs_dip, nmbuf);
596 598
597 599 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
598 600 ddi_remove_minor_node(zfs_dip, nmbuf);
599 601
600 602 avl_destroy(&zv->zv_znode.z_range_avl);
601 603 mutex_destroy(&zv->zv_znode.z_range_lock);
602 604
603 605 kmem_free(zv, sizeof (zvol_state_t));
604 606
605 607 ddi_soft_state_free(zfsdev_state, minor);
606 608
607 609 zvol_minors--;
608 610 return (0);
609 611 }
610 612
611 613 int
612 614 zvol_remove_minor(const char *name)
613 615 {
614 616 zvol_state_t *zv;
615 617 int rc;
616 618
617 619 mutex_enter(&zfsdev_state_lock);
618 620 if ((zv = zvol_minor_lookup(name)) == NULL) {
619 621 mutex_exit(&zfsdev_state_lock);
620 622 return (SET_ERROR(ENXIO));
621 623 }
622 624 rc = zvol_remove_zv(zv);
623 625 mutex_exit(&zfsdev_state_lock);
624 626 return (rc);
625 627 }
626 628
627 629 int
628 630 zvol_first_open(zvol_state_t *zv)
629 631 {
630 632 objset_t *os;
631 633 uint64_t volsize;
632 634 int error;
633 635 uint64_t readonly;
634 636
635 637 /* lie and say we're read-only */
636 638 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
637 639 zvol_tag, &os);
638 640 if (error)
639 641 return (error);
640 642
641 643 zv->zv_objset = os;
642 644 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
643 645 if (error) {
644 646 ASSERT(error == 0);
645 647 dmu_objset_disown(os, zvol_tag);
646 648 return (error);
647 649 }
648 650
649 651 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
650 652 if (error) {
651 653 dmu_objset_disown(os, zvol_tag);
652 654 return (error);
653 655 }
654 656
655 657 zvol_size_changed(zv, volsize);
656 658 zv->zv_zilog = zil_open(os, zvol_get_data);
657 659
658 660 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
659 661 NULL) == 0);
660 662 if (readonly || dmu_objset_is_snapshot(os) ||
661 663 !spa_writeable(dmu_objset_spa(os)))
662 664 zv->zv_flags |= ZVOL_RDONLY;
663 665 else
664 666 zv->zv_flags &= ~ZVOL_RDONLY;
665 667 return (error);
666 668 }
667 669
668 670 void
669 671 zvol_last_close(zvol_state_t *zv)
670 672 {
671 673 zil_close(zv->zv_zilog);
672 674 zv->zv_zilog = NULL;
673 675
674 676 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
675 677 zv->zv_dbuf = NULL;
676 678
677 679 /*
678 680 * Evict cached data
679 681 */
680 682 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
681 683 !(zv->zv_flags & ZVOL_RDONLY))
682 684 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
683 685 dmu_objset_evict_dbufs(zv->zv_objset);
684 686
685 687 dmu_objset_disown(zv->zv_objset, zvol_tag);
686 688 zv->zv_objset = NULL;
687 689 }
688 690
689 691 int
690 692 zvol_prealloc(zvol_state_t *zv)
691 693 {
692 694 objset_t *os = zv->zv_objset;
693 695 dmu_tx_t *tx;
694 696 uint64_t refd, avail, usedobjs, availobjs;
695 697 uint64_t resid = zv->zv_volsize;
696 698 uint64_t off = 0;
697 699
698 700 /* Check the space usage before attempting to allocate the space */
699 701 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
700 702 if (avail < zv->zv_volsize)
701 703 return (SET_ERROR(ENOSPC));
702 704
703 705 /* Free old extents if they exist */
704 706 zvol_free_extents(zv);
705 707
706 708 while (resid != 0) {
707 709 int error;
708 710 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
709 711
710 712 tx = dmu_tx_create(os);
711 713 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
712 714 error = dmu_tx_assign(tx, TXG_WAIT);
713 715 if (error) {
714 716 dmu_tx_abort(tx);
715 717 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
716 718 return (error);
717 719 }
718 720 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
719 721 dmu_tx_commit(tx);
720 722 off += bytes;
721 723 resid -= bytes;
722 724 }
723 725 txg_wait_synced(dmu_objset_pool(os), 0);
724 726
725 727 return (0);
726 728 }
727 729
728 730 static int
729 731 zvol_update_volsize(objset_t *os, uint64_t volsize)
730 732 {
731 733 dmu_tx_t *tx;
732 734 int error;
733 735
734 736 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
735 737
736 738 tx = dmu_tx_create(os);
737 739 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
738 740 dmu_tx_mark_netfree(tx);
739 741 error = dmu_tx_assign(tx, TXG_WAIT);
740 742 if (error) {
741 743 dmu_tx_abort(tx);
742 744 return (error);
743 745 }
744 746
745 747 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
746 748 &volsize, tx);
747 749 dmu_tx_commit(tx);
748 750
749 751 if (error == 0)
750 752 error = dmu_free_long_range(os,
751 753 ZVOL_OBJ, volsize, DMU_OBJECT_END);
752 754 return (error);
753 755 }
754 756
755 757 void
756 758 zvol_remove_minors(const char *name)
757 759 {
758 760 zvol_state_t *zv;
759 761 char *namebuf;
760 762 minor_t minor;
761 763
762 764 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
763 765 (void) strncpy(namebuf, name, strlen(name));
764 766 (void) strcat(namebuf, "/");
765 767 mutex_enter(&zfsdev_state_lock);
766 768 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
767 769
768 770 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
769 771 if (zv == NULL)
770 772 continue;
771 773 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
772 774 (void) zvol_remove_zv(zv);
773 775 }
774 776 kmem_free(namebuf, strlen(name) + 2);
775 777
776 778 mutex_exit(&zfsdev_state_lock);
777 779 }
778 780
779 781 static int
780 782 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
781 783 {
782 784 uint64_t old_volsize = 0ULL;
783 785 int error = 0;
784 786
785 787 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
786 788
787 789 /*
788 790 * Reinitialize the dump area to the new size. If we
789 791 * failed to resize the dump area then restore it back to
790 792 * its original size. We must set the new volsize prior
791 793 * to calling dumpvp_resize() to ensure that the devices'
792 794 * size(9P) is not visible by the dump subsystem.
793 795 */
794 796 old_volsize = zv->zv_volsize;
795 797 zvol_size_changed(zv, volsize);
796 798
797 799 if (zv->zv_flags & ZVOL_DUMPIFIED) {
798 800 if ((error = zvol_dumpify(zv)) != 0 ||
799 801 (error = dumpvp_resize()) != 0) {
800 802 int dumpify_error;
801 803
802 804 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
803 805 zvol_size_changed(zv, old_volsize);
804 806 dumpify_error = zvol_dumpify(zv);
805 807 error = dumpify_error ? dumpify_error : error;
806 808 }
807 809 }
808 810
809 811 /*
810 812 * Generate a LUN expansion event.
811 813 */
812 814 if (error == 0) {
813 815 sysevent_id_t eid;
814 816 nvlist_t *attr;
815 817 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
816 818
817 819 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
818 820 zv->zv_minor);
819 821
820 822 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
821 823 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
822 824
823 825 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
824 826 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
825 827
826 828 nvlist_free(attr);
827 829 kmem_free(physpath, MAXPATHLEN);
828 830 }
829 831 return (error);
830 832 }
831 833
832 834 int
833 835 zvol_set_volsize(const char *name, uint64_t volsize)
834 836 {
835 837 zvol_state_t *zv = NULL;
836 838 objset_t *os;
837 839 int error;
838 840 dmu_object_info_t doi;
839 841 uint64_t readonly;
840 842 boolean_t owned = B_FALSE;
841 843
842 844 error = dsl_prop_get_integer(name,
843 845 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
844 846 if (error != 0)
845 847 return (error);
846 848 if (readonly)
847 849 return (SET_ERROR(EROFS));
848 850
849 851 mutex_enter(&zfsdev_state_lock);
850 852 zv = zvol_minor_lookup(name);
851 853
852 854 if (zv == NULL || zv->zv_objset == NULL) {
853 855 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
854 856 FTAG, &os)) != 0) {
855 857 mutex_exit(&zfsdev_state_lock);
856 858 return (error);
857 859 }
858 860 owned = B_TRUE;
859 861 if (zv != NULL)
860 862 zv->zv_objset = os;
861 863 } else {
862 864 os = zv->zv_objset;
863 865 }
864 866
865 867 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
866 868 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
867 869 goto out;
868 870
869 871 error = zvol_update_volsize(os, volsize);
870 872
871 873 if (error == 0 && zv != NULL)
872 874 error = zvol_update_live_volsize(zv, volsize);
873 875 out:
874 876 if (owned) {
875 877 dmu_objset_disown(os, FTAG);
876 878 if (zv != NULL)
877 879 zv->zv_objset = NULL;
878 880 }
879 881 mutex_exit(&zfsdev_state_lock);
880 882 return (error);
881 883 }
882 884
883 885 /*ARGSUSED*/
884 886 int
885 887 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
886 888 {
887 889 zvol_state_t *zv;
888 890 int err = 0;
889 891
890 892 mutex_enter(&zfsdev_state_lock);
891 893
892 894 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
893 895 if (zv == NULL) {
894 896 mutex_exit(&zfsdev_state_lock);
895 897 return (SET_ERROR(ENXIO));
896 898 }
897 899
898 900 if (zv->zv_total_opens == 0)
899 901 err = zvol_first_open(zv);
900 902 if (err) {
901 903 mutex_exit(&zfsdev_state_lock);
902 904 return (err);
903 905 }
904 906 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
905 907 err = SET_ERROR(EROFS);
906 908 goto out;
907 909 }
908 910 if (zv->zv_flags & ZVOL_EXCL) {
909 911 err = SET_ERROR(EBUSY);
910 912 goto out;
911 913 }
912 914 if (flag & FEXCL) {
913 915 if (zv->zv_total_opens != 0) {
914 916 err = SET_ERROR(EBUSY);
915 917 goto out;
916 918 }
917 919 zv->zv_flags |= ZVOL_EXCL;
918 920 }
919 921
920 922 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
921 923 zv->zv_open_count[otyp]++;
922 924 zv->zv_total_opens++;
923 925 }
924 926 mutex_exit(&zfsdev_state_lock);
925 927
926 928 return (err);
927 929 out:
928 930 if (zv->zv_total_opens == 0)
929 931 zvol_last_close(zv);
930 932 mutex_exit(&zfsdev_state_lock);
931 933 return (err);
932 934 }
933 935
934 936 /*ARGSUSED*/
935 937 int
936 938 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
937 939 {
938 940 minor_t minor = getminor(dev);
939 941 zvol_state_t *zv;
940 942 int error = 0;
941 943
942 944 mutex_enter(&zfsdev_state_lock);
943 945
944 946 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
945 947 if (zv == NULL) {
946 948 mutex_exit(&zfsdev_state_lock);
947 949 return (SET_ERROR(ENXIO));
948 950 }
949 951
950 952 if (zv->zv_flags & ZVOL_EXCL) {
951 953 ASSERT(zv->zv_total_opens == 1);
952 954 zv->zv_flags &= ~ZVOL_EXCL;
953 955 }
954 956
955 957 /*
956 958 * If the open count is zero, this is a spurious close.
957 959 * That indicates a bug in the kernel / DDI framework.
958 960 */
959 961 ASSERT(zv->zv_open_count[otyp] != 0);
960 962 ASSERT(zv->zv_total_opens != 0);
961 963
962 964 /*
963 965 * You may get multiple opens, but only one close.
964 966 */
|
↓ open down ↓ |
864 lines elided |
↑ open up ↑ |
965 967 zv->zv_open_count[otyp]--;
966 968 zv->zv_total_opens--;
967 969
968 970 if (zv->zv_total_opens == 0)
969 971 zvol_last_close(zv);
970 972
971 973 mutex_exit(&zfsdev_state_lock);
972 974 return (error);
973 975 }
974 976
977 +/* ARGSUSED */
975 978 static void
976 979 zvol_get_done(zgd_t *zgd, int error)
977 980 {
978 981 if (zgd->zgd_db)
979 982 dmu_buf_rele(zgd->zgd_db, zgd);
980 983
981 984 zfs_range_unlock(zgd->zgd_rl);
982 985
983 - if (error == 0 && zgd->zgd_bp)
984 - zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
985 -
986 986 kmem_free(zgd, sizeof (zgd_t));
987 987 }
988 988
989 989 /*
990 990 * Get data to generate a TX_WRITE intent log record.
991 991 */
992 992 static int
993 993 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
994 994 {
995 995 zvol_state_t *zv = arg;
996 996 objset_t *os = zv->zv_objset;
997 997 uint64_t object = ZVOL_OBJ;
998 998 uint64_t offset = lr->lr_offset;
999 999 uint64_t size = lr->lr_length; /* length of user data */
1000 1000 dmu_buf_t *db;
1001 1001 zgd_t *zgd;
1002 1002 int error;
1003 1003
1004 1004 ASSERT3P(lwb, !=, NULL);
1005 1005 ASSERT3P(zio, !=, NULL);
1006 1006 ASSERT3U(size, !=, 0);
1007 1007
1008 1008 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1009 1009 zgd->zgd_lwb = lwb;
1010 1010
1011 1011 /*
1012 1012 * Write records come in two flavors: immediate and indirect.
1013 1013 * For small writes it's cheaper to store the data with the
1014 1014 * log record (immediate); for large writes it's cheaper to
1015 1015 * sync the data and get a pointer to it (indirect) so that
1016 1016 * we don't have to write the data twice.
1017 1017 */
1018 1018 if (buf != NULL) { /* immediate write */
1019 1019 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size,
1020 1020 RL_READER);
1021 1021 error = dmu_read(os, object, offset, size, buf,
1022 1022 DMU_READ_NO_PREFETCH);
1023 1023 } else { /* indirect write */
1024 1024 /*
1025 1025 * Have to lock the whole block to ensure when it's written out
1026 1026 * and its checksum is being calculated that no one can change
1027 1027 * the data. Contrarily to zfs_get_data we need not re-check
1028 1028 * blocksize after we get the lock because it cannot be changed.
1029 1029 */
1030 1030 size = zv->zv_volblocksize;
1031 1031 offset = P2ALIGN(offset, size);
1032 1032 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size,
1033 1033 RL_READER);
1034 1034 error = dmu_buf_hold(os, object, offset, zgd, &db,
1035 1035 DMU_READ_NO_PREFETCH);
1036 1036 if (error == 0) {
1037 1037 blkptr_t *bp = &lr->lr_blkptr;
1038 1038
1039 1039 zgd->zgd_db = db;
1040 1040 zgd->zgd_bp = bp;
1041 1041
1042 1042 ASSERT(db->db_offset == offset);
1043 1043 ASSERT(db->db_size == size);
1044 1044
1045 1045 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1046 1046 zvol_get_done, zgd);
1047 1047
1048 1048 if (error == 0)
1049 1049 return (0);
1050 1050 }
1051 1051 }
1052 1052
1053 1053 zvol_get_done(zgd, error);
1054 1054
1055 1055 return (error);
1056 1056 }
1057 1057
1058 1058 /*
1059 1059 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1060 1060 *
1061 1061 * We store data in the log buffers if it's small enough.
|
↓ open down ↓ |
66 lines elided |
↑ open up ↑ |
1062 1062 * Otherwise we will later flush the data out via dmu_sync().
1063 1063 */
1064 1064 ssize_t zvol_immediate_write_sz = 32768;
1065 1065
1066 1066 static void
1067 1067 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1068 1068 boolean_t sync)
1069 1069 {
1070 1070 uint32_t blocksize = zv->zv_volblocksize;
1071 1071 zilog_t *zilog = zv->zv_zilog;
1072 + spa_t *spa = zilog->zl_spa;
1073 + spa_meta_placement_t *mp = &spa->spa_meta_policy;
1074 + boolean_t slogging, zil_to_special, write_to_special;
1075 + ssize_t immediate_write_sz;
1072 1076 itx_wr_state_t write_state;
1073 1077
1074 1078 if (zil_replaying(zilog, tx))
1075 1079 return;
1076 1080
1077 - if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1081 + /*
1082 + * See comments in zfs_log_write()
1083 + */
1084 +
1085 + immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1086 + ? 0 : zvol_immediate_write_sz;
1087 +
1088 + zil_to_special = !spa_has_slogs(spa) &&
1089 + spa_can_special_be_used(spa) &&
1090 + mp->spa_sync_to_special != SYNC_TO_SPECIAL_DISABLED;
1091 +
1092 + write_to_special = !spa_has_slogs(spa) &&
1093 + spa_write_data_to_special(spa, zilog->zl_os) &&
1094 + (mp->spa_sync_to_special == SYNC_TO_SPECIAL_ALWAYS ||
1095 + (mp->spa_sync_to_special == SYNC_TO_SPECIAL_BALANCED &&
1096 + spa->spa_avg_stat_rotor % 100 < spa->spa_special_to_normal_ratio));
1097 +
1098 + slogging = (spa_has_slogs(spa) || zil_to_special) &&
1099 + (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1100 +
1101 + if (blocksize > immediate_write_sz && !slogging &&
1102 + resid >= blocksize && off % blocksize == 0)
1078 1103 write_state = WR_INDIRECT;
1104 + else if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1105 + write_state = WR_INDIRECT;
1079 1106 else if (!spa_has_slogs(zilog->zl_spa) &&
1080 1107 resid >= blocksize && blocksize > zvol_immediate_write_sz)
1081 1108 write_state = WR_INDIRECT;
1109 + else if (write_to_special)
1110 + write_state = WR_INDIRECT;
1082 1111 else if (sync)
1083 1112 write_state = WR_COPIED;
1084 1113 else
1085 1114 write_state = WR_NEED_COPY;
1086 1115
1087 1116 while (resid) {
1088 1117 itx_t *itx;
1089 1118 lr_write_t *lr;
1090 1119 itx_wr_state_t wr_state = write_state;
1091 1120 ssize_t len = resid;
1092 1121
1093 1122 if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA)
1094 1123 wr_state = WR_NEED_COPY;
1095 1124 else if (wr_state == WR_INDIRECT)
1096 1125 len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1097 1126
1098 1127 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1099 1128 (wr_state == WR_COPIED ? len : 0));
1100 1129 lr = (lr_write_t *)&itx->itx_lr;
1101 1130 if (wr_state == WR_COPIED && dmu_read(zv->zv_objset,
1102 1131 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1103 1132 zil_itx_destroy(itx);
1104 1133 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1105 1134 lr = (lr_write_t *)&itx->itx_lr;
1106 1135 wr_state = WR_NEED_COPY;
1107 1136 }
1108 1137
1109 1138 itx->itx_wr_state = wr_state;
1110 1139 lr->lr_foid = ZVOL_OBJ;
1111 1140 lr->lr_offset = off;
1112 1141 lr->lr_length = len;
1113 1142 lr->lr_blkoff = 0;
1114 1143 BP_ZERO(&lr->lr_blkptr);
1115 1144
1116 1145 itx->itx_private = zv;
1117 1146 itx->itx_sync = sync;
1118 1147
1119 1148 zil_itx_assign(zilog, itx, tx);
1120 1149
|
↓ open down ↓ |
29 lines elided |
↑ open up ↑ |
1121 1150 off += len;
1122 1151 resid -= len;
1123 1152 }
1124 1153 }
1125 1154
1126 1155 static int
1127 1156 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1128 1157 uint64_t size, boolean_t doread, boolean_t isdump)
1129 1158 {
1130 1159 vdev_disk_t *dvd;
1131 - int c;
1160 + int c, rc;
1132 1161 int numerrors = 0;
1133 1162
1134 1163 if (vd->vdev_ops == &vdev_mirror_ops ||
1135 1164 vd->vdev_ops == &vdev_replacing_ops ||
1136 1165 vd->vdev_ops == &vdev_spare_ops) {
1137 1166 for (c = 0; c < vd->vdev_children; c++) {
1138 1167 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1139 1168 addr, offset, origoffset, size, doread, isdump);
1140 1169 if (err != 0) {
1141 1170 numerrors++;
1142 1171 } else if (doread) {
1143 1172 break;
1144 1173 }
1145 1174 }
1146 1175 }
1147 1176
1148 1177 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1149 1178 return (numerrors < vd->vdev_children ? 0 : EIO);
1150 1179
1151 1180 if (doread && !vdev_readable(vd))
1152 1181 return (SET_ERROR(EIO));
|
↓ open down ↓ |
11 lines elided |
↑ open up ↑ |
1153 1182 else if (!doread && !vdev_writeable(vd))
1154 1183 return (SET_ERROR(EIO));
1155 1184
1156 1185 if (vd->vdev_ops == &vdev_raidz_ops) {
1157 1186 return (vdev_raidz_physio(vd,
1158 1187 addr, size, offset, origoffset, doread, isdump));
1159 1188 }
1160 1189
1161 1190 offset += VDEV_LABEL_START_SIZE;
1162 1191
1192 + rw_enter(&vd->vdev_tsd_lock, RW_READER);
1193 + dvd = vd->vdev_tsd;
1163 1194 if (ddi_in_panic() || isdump) {
1164 1195 ASSERT(!doread);
1165 - if (doread)
1196 + if (doread) {
1197 + rw_exit(&vd->vdev_tsd_lock);
1166 1198 return (SET_ERROR(EIO));
1167 - dvd = vd->vdev_tsd;
1199 + }
1200 + /* We assume here dvd is not NULL */
1168 1201 ASSERT3P(dvd, !=, NULL);
1169 - return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1170 - lbtodb(size)));
1202 +
1203 + /* If our assumption is wrong, we do not want to crash */
1204 + if (dvd != NULL && dvd->vd_lh != NULL) {
1205 + rc = ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1206 + lbtodb(size));
1207 + } else {
1208 + rc = SET_ERROR(ENXIO);
1209 + }
1171 1210 } else {
1172 - dvd = vd->vdev_tsd;
1211 + /* We assume here dvd is not NULL */
1173 1212 ASSERT3P(dvd, !=, NULL);
1174 - return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1175 - offset, doread ? B_READ : B_WRITE));
1213 +
1214 + /* If our assumption is wrong, we do not want to crash */
1215 + if (dvd != NULL && dvd->vd_lh != NULL) {
1216 + rc = vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1217 + offset, doread ? B_READ : B_WRITE);
1218 + } else {
1219 + rc = SET_ERROR(ENXIO);
1220 + }
1176 1221 }
1222 + rw_exit(&vd->vdev_tsd_lock);
1223 + return (rc);
1177 1224 }
1178 1225
1179 1226 static int
1180 1227 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1181 1228 boolean_t doread, boolean_t isdump)
1182 1229 {
1183 1230 vdev_t *vd;
1184 1231 int error;
1185 1232 zvol_extent_t *ze;
1186 1233 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1187 1234
1188 1235 /* Must be sector aligned, and not stradle a block boundary. */
1189 1236 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1190 1237 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1191 1238 return (SET_ERROR(EINVAL));
1192 1239 }
1193 1240 ASSERT(size <= zv->zv_volblocksize);
1194 1241
1195 1242 /* Locate the extent this belongs to */
1196 1243 ze = list_head(&zv->zv_extents);
1197 1244 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1198 1245 offset -= ze->ze_nblks * zv->zv_volblocksize;
1199 1246 ze = list_next(&zv->zv_extents, ze);
1200 1247 }
1201 1248
1202 1249 if (ze == NULL)
1203 1250 return (SET_ERROR(EINVAL));
1204 1251
1205 1252 if (!ddi_in_panic())
1206 1253 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1207 1254
1208 1255 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1209 1256 offset += DVA_GET_OFFSET(&ze->ze_dva);
1210 1257 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1211 1258 size, doread, isdump);
1212 1259
1213 1260 if (!ddi_in_panic())
1214 1261 spa_config_exit(spa, SCL_STATE, FTAG);
1215 1262
1216 1263 return (error);
1217 1264 }
1218 1265
1219 1266 int
1220 1267 zvol_strategy(buf_t *bp)
1221 1268 {
1222 1269 zfs_soft_state_t *zs = NULL;
1223 1270 zvol_state_t *zv;
1224 1271 uint64_t off, volsize;
1225 1272 size_t resid;
1226 1273 char *addr;
1227 1274 objset_t *os;
1228 1275 rl_t *rl;
1229 1276 int error = 0;
1230 1277 boolean_t doread = bp->b_flags & B_READ;
1231 1278 boolean_t is_dumpified;
1232 1279 boolean_t sync;
1233 1280
1234 1281 if (getminor(bp->b_edev) == 0) {
1235 1282 error = SET_ERROR(EINVAL);
1236 1283 } else {
1237 1284 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1238 1285 if (zs == NULL)
1239 1286 error = SET_ERROR(ENXIO);
1240 1287 else if (zs->zss_type != ZSST_ZVOL)
1241 1288 error = SET_ERROR(EINVAL);
1242 1289 }
1243 1290
1244 1291 if (error) {
1245 1292 bioerror(bp, error);
1246 1293 biodone(bp);
1247 1294 return (0);
1248 1295 }
1249 1296
1250 1297 zv = zs->zss_data;
1251 1298
1252 1299 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1253 1300 bioerror(bp, EROFS);
1254 1301 biodone(bp);
1255 1302 return (0);
1256 1303 }
1257 1304
1258 1305 off = ldbtob(bp->b_blkno);
1259 1306 volsize = zv->zv_volsize;
1260 1307
1261 1308 os = zv->zv_objset;
1262 1309 ASSERT(os != NULL);
1263 1310
1264 1311 bp_mapin(bp);
1265 1312 addr = bp->b_un.b_addr;
1266 1313 resid = bp->b_bcount;
1267 1314
1268 1315 if (resid > 0 && (off < 0 || off >= volsize)) {
1269 1316 bioerror(bp, EIO);
1270 1317 biodone(bp);
1271 1318 return (0);
1272 1319 }
1273 1320
1274 1321 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1275 1322 sync = ((!(bp->b_flags & B_ASYNC) &&
1276 1323 !(zv->zv_flags & ZVOL_WCE)) ||
1277 1324 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1278 1325 !doread && !is_dumpified;
1279 1326
1280 1327 /*
1281 1328 * There must be no buffer changes when doing a dmu_sync() because
1282 1329 * we can't change the data whilst calculating the checksum.
1283 1330 */
1284 1331 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1285 1332 doread ? RL_READER : RL_WRITER);
1286 1333
1287 1334 while (resid != 0 && off < volsize) {
1288 1335 size_t size = MIN(resid, zvol_maxphys);
1289 1336 if (is_dumpified) {
1290 1337 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1291 1338 error = zvol_dumpio(zv, addr, off, size,
1292 1339 doread, B_FALSE);
1293 1340 } else if (doread) {
1294 1341 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1295 1342 DMU_READ_PREFETCH);
1296 1343 } else {
1297 1344 dmu_tx_t *tx = dmu_tx_create(os);
1298 1345 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1299 1346 error = dmu_tx_assign(tx, TXG_WAIT);
1300 1347 if (error) {
1301 1348 dmu_tx_abort(tx);
1302 1349 } else {
1303 1350 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1304 1351 zvol_log_write(zv, tx, off, size, sync);
1305 1352 dmu_tx_commit(tx);
1306 1353 }
1307 1354 }
1308 1355 if (error) {
1309 1356 /* convert checksum errors into IO errors */
1310 1357 if (error == ECKSUM)
1311 1358 error = SET_ERROR(EIO);
1312 1359 break;
1313 1360 }
1314 1361 off += size;
1315 1362 addr += size;
1316 1363 resid -= size;
1317 1364 }
1318 1365 zfs_range_unlock(rl);
1319 1366
1320 1367 if ((bp->b_resid = resid) == bp->b_bcount)
1321 1368 bioerror(bp, off > volsize ? EINVAL : error);
1322 1369
1323 1370 if (sync)
1324 1371 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1325 1372 biodone(bp);
1326 1373
1327 1374 return (0);
1328 1375 }
1329 1376
1330 1377 /*
1331 1378 * Set the buffer count to the zvol maximum transfer.
1332 1379 * Using our own routine instead of the default minphys()
1333 1380 * means that for larger writes we write bigger buffers on X86
1334 1381 * (128K instead of 56K) and flush the disk write cache less often
1335 1382 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1336 1383 * 56K on X86 and 128K on sparc).
1337 1384 */
1338 1385 void
1339 1386 zvol_minphys(struct buf *bp)
1340 1387 {
1341 1388 if (bp->b_bcount > zvol_maxphys)
1342 1389 bp->b_bcount = zvol_maxphys;
1343 1390 }
1344 1391
1345 1392 int
1346 1393 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1347 1394 {
1348 1395 minor_t minor = getminor(dev);
1349 1396 zvol_state_t *zv;
1350 1397 int error = 0;
1351 1398 uint64_t size;
1352 1399 uint64_t boff;
1353 1400 uint64_t resid;
1354 1401
1355 1402 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1356 1403 if (zv == NULL)
1357 1404 return (SET_ERROR(ENXIO));
1358 1405
1359 1406 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1360 1407 return (SET_ERROR(EINVAL));
1361 1408
1362 1409 boff = ldbtob(blkno);
1363 1410 resid = ldbtob(nblocks);
1364 1411
1365 1412 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1366 1413
1367 1414 while (resid) {
1368 1415 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1369 1416 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1370 1417 if (error)
1371 1418 break;
1372 1419 boff += size;
1373 1420 addr += size;
1374 1421 resid -= size;
1375 1422 }
1376 1423
1377 1424 return (error);
1378 1425 }
1379 1426
1380 1427 /*ARGSUSED*/
1381 1428 int
1382 1429 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1383 1430 {
1384 1431 minor_t minor = getminor(dev);
1385 1432 zvol_state_t *zv;
1386 1433 uint64_t volsize;
1387 1434 rl_t *rl;
1388 1435 int error = 0;
1389 1436
1390 1437 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1391 1438 if (zv == NULL)
1392 1439 return (SET_ERROR(ENXIO));
1393 1440
1394 1441 volsize = zv->zv_volsize;
1395 1442 if (uio->uio_resid > 0 &&
1396 1443 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1397 1444 return (SET_ERROR(EIO));
1398 1445
1399 1446 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1400 1447 error = physio(zvol_strategy, NULL, dev, B_READ,
1401 1448 zvol_minphys, uio);
1402 1449 return (error);
1403 1450 }
1404 1451
1405 1452 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1406 1453 RL_READER);
1407 1454 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1408 1455 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1409 1456
1410 1457 /* don't read past the end */
1411 1458 if (bytes > volsize - uio->uio_loffset)
1412 1459 bytes = volsize - uio->uio_loffset;
1413 1460
1414 1461 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1415 1462 if (error) {
1416 1463 /* convert checksum errors into IO errors */
1417 1464 if (error == ECKSUM)
1418 1465 error = SET_ERROR(EIO);
1419 1466 break;
1420 1467 }
1421 1468 }
1422 1469 zfs_range_unlock(rl);
1423 1470 return (error);
1424 1471 }
1425 1472
1426 1473 /*ARGSUSED*/
1427 1474 int
1428 1475 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1429 1476 {
1430 1477 minor_t minor = getminor(dev);
1431 1478 zvol_state_t *zv;
1432 1479 uint64_t volsize;
1433 1480 rl_t *rl;
1434 1481 int error = 0;
1435 1482 boolean_t sync;
1436 1483
1437 1484 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1438 1485 if (zv == NULL)
1439 1486 return (SET_ERROR(ENXIO));
1440 1487
1441 1488 volsize = zv->zv_volsize;
1442 1489 if (uio->uio_resid > 0 &&
1443 1490 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1444 1491 return (SET_ERROR(EIO));
1445 1492
1446 1493 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1447 1494 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1448 1495 zvol_minphys, uio);
1449 1496 return (error);
1450 1497 }
1451 1498
1452 1499 sync = !(zv->zv_flags & ZVOL_WCE) ||
1453 1500 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1454 1501
1455 1502 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1456 1503 RL_WRITER);
1457 1504 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1458 1505 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1459 1506 uint64_t off = uio->uio_loffset;
1460 1507 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1461 1508
1462 1509 if (bytes > volsize - off) /* don't write past the end */
1463 1510 bytes = volsize - off;
1464 1511
1465 1512 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1466 1513 error = dmu_tx_assign(tx, TXG_WAIT);
1467 1514 if (error) {
1468 1515 dmu_tx_abort(tx);
1469 1516 break;
1470 1517 }
1471 1518 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1472 1519 if (error == 0)
1473 1520 zvol_log_write(zv, tx, off, bytes, sync);
1474 1521 dmu_tx_commit(tx);
1475 1522
1476 1523 if (error)
1477 1524 break;
1478 1525 }
1479 1526 zfs_range_unlock(rl);
1480 1527 if (sync)
1481 1528 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1482 1529 return (error);
1483 1530 }
1484 1531
1485 1532 int
1486 1533 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1487 1534 {
1488 1535 struct uuid uuid = EFI_RESERVED;
1489 1536 efi_gpe_t gpe = { 0 };
1490 1537 uint32_t crc;
1491 1538 dk_efi_t efi;
1492 1539 int length;
1493 1540 char *ptr;
1494 1541
1495 1542 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1496 1543 return (SET_ERROR(EFAULT));
1497 1544 ptr = (char *)(uintptr_t)efi.dki_data_64;
1498 1545 length = efi.dki_length;
1499 1546 /*
1500 1547 * Some clients may attempt to request a PMBR for the
1501 1548 * zvol. Currently this interface will return EINVAL to
1502 1549 * such requests. These requests could be supported by
1503 1550 * adding a check for lba == 0 and consing up an appropriate
1504 1551 * PMBR.
1505 1552 */
1506 1553 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1507 1554 return (SET_ERROR(EINVAL));
1508 1555
1509 1556 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1510 1557 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1511 1558 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1512 1559
1513 1560 if (efi.dki_lba == 1) {
1514 1561 efi_gpt_t gpt = { 0 };
1515 1562
1516 1563 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1517 1564 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1518 1565 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1519 1566 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1520 1567 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1521 1568 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1522 1569 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1523 1570 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1524 1571 gpt.efi_gpt_SizeOfPartitionEntry =
1525 1572 LE_32(sizeof (efi_gpe_t));
1526 1573 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1527 1574 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1528 1575 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1529 1576 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1530 1577 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1531 1578 flag))
1532 1579 return (SET_ERROR(EFAULT));
1533 1580 ptr += sizeof (gpt);
1534 1581 length -= sizeof (gpt);
1535 1582 }
1536 1583 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1537 1584 length), flag))
1538 1585 return (SET_ERROR(EFAULT));
1539 1586 return (0);
1540 1587 }
1541 1588
1542 1589 /*
1543 1590 * BEGIN entry points to allow external callers access to the volume.
1544 1591 */
1545 1592 /*
1546 1593 * Return the volume parameters needed for access from an external caller.
1547 1594 * These values are invariant as long as the volume is held open.
1548 1595 */
1549 1596 int
1550 1597 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1551 1598 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1552 1599 void **rl_hdl, void **bonus_hdl)
1553 1600 {
1554 1601 zvol_state_t *zv;
1555 1602
1556 1603 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1557 1604 if (zv == NULL)
1558 1605 return (SET_ERROR(ENXIO));
1559 1606 if (zv->zv_flags & ZVOL_DUMPIFIED)
1560 1607 return (SET_ERROR(ENXIO));
1561 1608
1562 1609 ASSERT(blksize && max_xfer_len && minor_hdl &&
1563 1610 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1564 1611
1565 1612 *blksize = zv->zv_volblocksize;
1566 1613 *max_xfer_len = (uint64_t)zvol_maxphys;
1567 1614 *minor_hdl = zv;
1568 1615 *objset_hdl = zv->zv_objset;
1569 1616 *zil_hdl = zv->zv_zilog;
1570 1617 *rl_hdl = &zv->zv_znode;
1571 1618 *bonus_hdl = zv->zv_dbuf;
1572 1619 return (0);
1573 1620 }
1574 1621
1575 1622 /*
1576 1623 * Return the current volume size to an external caller.
1577 1624 * The size can change while the volume is open.
1578 1625 */
1579 1626 uint64_t
1580 1627 zvol_get_volume_size(void *minor_hdl)
1581 1628 {
1582 1629 zvol_state_t *zv = minor_hdl;
1583 1630
1584 1631 return (zv->zv_volsize);
1585 1632 }
1586 1633
1587 1634 /*
1588 1635 * Return the current WCE setting to an external caller.
1589 1636 * The WCE setting can change while the volume is open.
1590 1637 */
1591 1638 int
1592 1639 zvol_get_volume_wce(void *minor_hdl)
1593 1640 {
1594 1641 zvol_state_t *zv = minor_hdl;
1595 1642
1596 1643 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1597 1644 }
1598 1645
1599 1646 /*
1600 1647 * Entry point for external callers to zvol_log_write
1601 1648 */
1602 1649 void
1603 1650 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1604 1651 boolean_t sync)
1605 1652 {
1606 1653 zvol_state_t *zv = minor_hdl;
1607 1654
1608 1655 zvol_log_write(zv, tx, off, resid, sync);
1609 1656 }
1610 1657 /*
1611 1658 * END entry points to allow external callers access to the volume.
1612 1659 */
1613 1660
1614 1661 /*
1615 1662 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1616 1663 */
1617 1664 static void
1618 1665 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1619 1666 boolean_t sync)
1620 1667 {
1621 1668 itx_t *itx;
1622 1669 lr_truncate_t *lr;
1623 1670 zilog_t *zilog = zv->zv_zilog;
1624 1671
1625 1672 if (zil_replaying(zilog, tx))
1626 1673 return;
1627 1674
1628 1675 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1629 1676 lr = (lr_truncate_t *)&itx->itx_lr;
1630 1677 lr->lr_foid = ZVOL_OBJ;
1631 1678 lr->lr_offset = off;
1632 1679 lr->lr_length = len;
1633 1680
1634 1681 itx->itx_sync = sync;
1635 1682 zil_itx_assign(zilog, itx, tx);
1636 1683 }
1637 1684
1638 1685 /*
1639 1686 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
1640 1687 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1641 1688 */
1642 1689 /*ARGSUSED*/
1643 1690 int
1644 1691 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1645 1692 {
1646 1693 zvol_state_t *zv;
1647 1694 struct dk_callback *dkc;
1648 1695 int error = 0;
1649 1696 rl_t *rl;
1650 1697
1651 1698 mutex_enter(&zfsdev_state_lock);
1652 1699
1653 1700 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1654 1701
1655 1702 if (zv == NULL) {
1656 1703 mutex_exit(&zfsdev_state_lock);
1657 1704 return (SET_ERROR(ENXIO));
1658 1705 }
1659 1706 ASSERT(zv->zv_total_opens > 0);
1660 1707
1661 1708 switch (cmd) {
1662 1709
1663 1710 case DKIOCINFO:
1664 1711 {
1665 1712 struct dk_cinfo dki;
1666 1713
1667 1714 bzero(&dki, sizeof (dki));
1668 1715 (void) strcpy(dki.dki_cname, "zvol");
1669 1716 (void) strcpy(dki.dki_dname, "zvol");
1670 1717 dki.dki_ctype = DKC_UNKNOWN;
1671 1718 dki.dki_unit = getminor(dev);
1672 1719 dki.dki_maxtransfer =
1673 1720 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
1674 1721 mutex_exit(&zfsdev_state_lock);
1675 1722 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1676 1723 error = SET_ERROR(EFAULT);
1677 1724 return (error);
1678 1725 }
1679 1726
1680 1727 case DKIOCGMEDIAINFO:
1681 1728 {
1682 1729 struct dk_minfo dkm;
1683 1730
1684 1731 bzero(&dkm, sizeof (dkm));
1685 1732 dkm.dki_lbsize = 1U << zv->zv_min_bs;
1686 1733 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1687 1734 dkm.dki_media_type = DK_UNKNOWN;
1688 1735 mutex_exit(&zfsdev_state_lock);
1689 1736 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1690 1737 error = SET_ERROR(EFAULT);
1691 1738 return (error);
1692 1739 }
1693 1740
1694 1741 case DKIOCGMEDIAINFOEXT:
1695 1742 {
1696 1743 struct dk_minfo_ext dkmext;
1697 1744
1698 1745 bzero(&dkmext, sizeof (dkmext));
1699 1746 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
1700 1747 dkmext.dki_pbsize = zv->zv_volblocksize;
1701 1748 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1702 1749 dkmext.dki_media_type = DK_UNKNOWN;
1703 1750 mutex_exit(&zfsdev_state_lock);
1704 1751 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
1705 1752 error = SET_ERROR(EFAULT);
1706 1753 return (error);
1707 1754 }
1708 1755
1709 1756 case DKIOCGETEFI:
1710 1757 {
1711 1758 uint64_t vs = zv->zv_volsize;
1712 1759 uint8_t bs = zv->zv_min_bs;
1713 1760
1714 1761 mutex_exit(&zfsdev_state_lock);
1715 1762 error = zvol_getefi((void *)arg, flag, vs, bs);
1716 1763 return (error);
1717 1764 }
1718 1765
1719 1766 case DKIOCFLUSHWRITECACHE:
1720 1767 dkc = (struct dk_callback *)arg;
1721 1768 mutex_exit(&zfsdev_state_lock);
1722 1769 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1723 1770 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1724 1771 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
1725 1772 error = 0;
1726 1773 }
1727 1774 return (error);
1728 1775
1729 1776 case DKIOCGETWCE:
1730 1777 {
1731 1778 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1732 1779 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1733 1780 flag))
1734 1781 error = SET_ERROR(EFAULT);
1735 1782 break;
1736 1783 }
1737 1784 case DKIOCSETWCE:
1738 1785 {
1739 1786 int wce;
1740 1787 if (ddi_copyin((void *)arg, &wce, sizeof (int),
1741 1788 flag)) {
1742 1789 error = SET_ERROR(EFAULT);
1743 1790 break;
1744 1791 }
1745 1792 if (wce) {
1746 1793 zv->zv_flags |= ZVOL_WCE;
1747 1794 mutex_exit(&zfsdev_state_lock);
1748 1795 } else {
1749 1796 zv->zv_flags &= ~ZVOL_WCE;
1750 1797 mutex_exit(&zfsdev_state_lock);
1751 1798 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1752 1799 }
1753 1800 return (0);
1754 1801 }
1755 1802
1756 1803 case DKIOCGGEOM:
1757 1804 case DKIOCGVTOC:
1758 1805 /*
1759 1806 * commands using these (like prtvtoc) expect ENOTSUP
1760 1807 * since we're emulating an EFI label
1761 1808 */
1762 1809 error = SET_ERROR(ENOTSUP);
1763 1810 break;
1764 1811
1765 1812 case DKIOCDUMPINIT:
1766 1813 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1767 1814 RL_WRITER);
1768 1815 error = zvol_dumpify(zv);
1769 1816 zfs_range_unlock(rl);
1770 1817 break;
1771 1818
1772 1819 case DKIOCDUMPFINI:
|
↓ open down ↓ |
586 lines elided |
↑ open up ↑ |
1773 1820 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1774 1821 break;
1775 1822 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1776 1823 RL_WRITER);
1777 1824 error = zvol_dump_fini(zv);
1778 1825 zfs_range_unlock(rl);
1779 1826 break;
1780 1827
1781 1828 case DKIOCFREE:
1782 1829 {
1783 - dkioc_free_t df;
1830 + dkioc_free_list_t *dfl;
1784 1831 dmu_tx_t *tx;
1785 1832
1833 + mutex_exit(&zfsdev_state_lock);
1834 +
1786 1835 if (!zvol_unmap_enabled)
1787 1836 break;
1788 1837
1789 - if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
1790 - error = SET_ERROR(EFAULT);
1791 - break;
1838 + if (!(flag & FKIOCTL)) {
1839 + dfl = dfl_copyin((void *)arg, flag, KM_SLEEP);
1840 + if (dfl == NULL) {
1841 + error = SET_ERROR(EFAULT);
1842 + break;
1843 + }
1844 + } else {
1845 + dfl = (dkioc_free_list_t *)arg;
1792 1846 }
1793 1847
1794 - /*
1795 - * Apply Postel's Law to length-checking. If they overshoot,
1796 - * just blank out until the end, if there's a need to blank
1797 - * out anything.
1798 - */
1799 - if (df.df_start >= zv->zv_volsize)
1800 - break; /* No need to do anything... */
1848 + for (int i = 0; i < dfl->dfl_num_exts; i++) {
1849 + uint64_t start = dfl->dfl_exts[i].dfle_start,
1850 + length = dfl->dfl_exts[i].dfle_length,
1851 + end = start + length;
1801 1852
1802 - mutex_exit(&zfsdev_state_lock);
1853 + /*
1854 + * Apply Postel's Law to length-checking. If they
1855 + * overshoot, just blank out until the end, if there's
1856 + * a need to blank out anything.
1857 + */
1858 + if (start >= zv->zv_volsize)
1859 + continue; /* No need to do anything... */
1860 + if (end > zv->zv_volsize) {
1861 + end = DMU_OBJECT_END;
1862 + length = end - start;
1863 + }
1803 1864
1804 - rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
1805 - RL_WRITER);
1806 - tx = dmu_tx_create(zv->zv_objset);
1807 - dmu_tx_mark_netfree(tx);
1808 - error = dmu_tx_assign(tx, TXG_WAIT);
1809 - if (error != 0) {
1810 - dmu_tx_abort(tx);
1811 - } else {
1812 - zvol_log_truncate(zv, tx, df.df_start,
1813 - df.df_length, B_TRUE);
1814 - dmu_tx_commit(tx);
1815 - error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1816 - df.df_start, df.df_length);
1817 - }
1865 + rl = zfs_range_lock(&zv->zv_znode, start, length,
1866 + RL_WRITER);
1867 + tx = dmu_tx_create(zv->zv_objset);
1868 + error = dmu_tx_assign(tx, TXG_WAIT);
1869 + if (error != 0) {
1870 + dmu_tx_abort(tx);
1871 + } else {
1872 + zvol_log_truncate(zv, tx, start, length,
1873 + B_TRUE);
1874 + dmu_tx_commit(tx);
1875 + error = dmu_free_long_range(zv->zv_objset,
1876 + ZVOL_OBJ, start, length);
1877 + }
1818 1878
1819 - zfs_range_unlock(rl);
1879 + zfs_range_unlock(rl);
1820 1880
1881 + if (error != 0)
1882 + break;
1883 + }
1884 +
1821 1885 /*
1822 1886 * If the write-cache is disabled, 'sync' property
1823 1887 * is set to 'always', or if the caller is asking for
1824 1888 * a synchronous free, commit this operation to the zil.
1825 1889 * This will sync any previous uncommitted writes to the
1826 1890 * zvol object.
1827 1891 * Can be overridden by the zvol_unmap_sync_enabled tunable.
1828 1892 */
1829 1893 if ((error == 0) && zvol_unmap_sync_enabled &&
1830 1894 (!(zv->zv_flags & ZVOL_WCE) ||
1831 1895 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) ||
1832 - (df.df_flags & DF_WAIT_SYNC))) {
1896 + (dfl->dfl_flags & DF_WAIT_SYNC))) {
1833 1897 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1834 1898 }
1835 1899
1900 + if (!(flag & FKIOCTL))
1901 + dfl_free(dfl);
1902 +
1836 1903 return (error);
1837 1904 }
1838 1905
1839 1906 default:
1840 1907 error = SET_ERROR(ENOTTY);
1841 1908 break;
1842 1909
1843 1910 }
1844 1911 mutex_exit(&zfsdev_state_lock);
1845 1912 return (error);
1846 1913 }
1847 1914
1848 1915 int
1849 1916 zvol_busy(void)
1850 1917 {
1851 1918 return (zvol_minors != 0);
1852 1919 }
1853 1920
1854 1921 void
1855 1922 zvol_init(void)
1856 1923 {
1857 1924 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1858 1925 1) == 0);
1859 1926 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1860 1927 }
1861 1928
1862 1929 void
1863 1930 zvol_fini(void)
1864 1931 {
1865 1932 mutex_destroy(&zfsdev_state_lock);
1866 1933 ddi_soft_state_fini(&zfsdev_state);
1867 1934 }
1868 1935
1869 1936 /*ARGSUSED*/
1870 1937 static int
1871 1938 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
1872 1939 {
1873 1940 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1874 1941
1875 1942 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1876 1943 return (1);
1877 1944 return (0);
1878 1945 }
1879 1946
1880 1947 /*ARGSUSED*/
1881 1948 static void
1882 1949 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
1883 1950 {
1884 1951 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1885 1952
1886 1953 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
1887 1954 }
1888 1955
1889 1956 static int
1890 1957 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1891 1958 {
1892 1959 dmu_tx_t *tx;
1893 1960 int error;
1894 1961 objset_t *os = zv->zv_objset;
1895 1962 spa_t *spa = dmu_objset_spa(os);
1896 1963 vdev_t *vd = spa->spa_root_vdev;
1897 1964 nvlist_t *nv = NULL;
1898 1965 uint64_t version = spa_version(spa);
1899 1966 uint64_t checksum, compress, refresrv, vbs, dedup;
1900 1967
1901 1968 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1902 1969 ASSERT(vd->vdev_ops == &vdev_root_ops);
1903 1970
1904 1971 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1905 1972 DMU_OBJECT_END);
1906 1973 if (error != 0)
1907 1974 return (error);
1908 1975 /* wait for dmu_free_long_range to actually free the blocks */
1909 1976 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1910 1977
1911 1978 /*
1912 1979 * If the pool on which the dump device is being initialized has more
1913 1980 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
1914 1981 * enabled. If so, bump that feature's counter to indicate that the
1915 1982 * feature is active. We also check the vdev type to handle the
1916 1983 * following case:
1917 1984 * # zpool create test raidz disk1 disk2 disk3
1918 1985 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
1919 1986 * the raidz vdev itself has 3 children.
1920 1987 */
1921 1988 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
1922 1989 if (!spa_feature_is_enabled(spa,
1923 1990 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1924 1991 return (SET_ERROR(ENOTSUP));
1925 1992 (void) dsl_sync_task(spa_name(spa),
1926 1993 zfs_mvdev_dump_feature_check,
1927 1994 zfs_mvdev_dump_activate_feature_sync, NULL,
1928 1995 2, ZFS_SPACE_CHECK_RESERVED);
1929 1996 }
1930 1997
1931 1998 if (!resize) {
1932 1999 error = dsl_prop_get_integer(zv->zv_name,
1933 2000 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1934 2001 if (error == 0) {
1935 2002 error = dsl_prop_get_integer(zv->zv_name,
1936 2003 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
1937 2004 NULL);
1938 2005 }
1939 2006 if (error == 0) {
1940 2007 error = dsl_prop_get_integer(zv->zv_name,
1941 2008 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
1942 2009 &refresrv, NULL);
1943 2010 }
1944 2011 if (error == 0) {
1945 2012 error = dsl_prop_get_integer(zv->zv_name,
1946 2013 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
1947 2014 NULL);
1948 2015 }
1949 2016 if (version >= SPA_VERSION_DEDUP && error == 0) {
1950 2017 error = dsl_prop_get_integer(zv->zv_name,
1951 2018 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1952 2019 }
1953 2020 }
1954 2021 if (error != 0)
1955 2022 return (error);
1956 2023
1957 2024 tx = dmu_tx_create(os);
1958 2025 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1959 2026 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1960 2027 error = dmu_tx_assign(tx, TXG_WAIT);
1961 2028 if (error != 0) {
1962 2029 dmu_tx_abort(tx);
1963 2030 return (error);
1964 2031 }
1965 2032
1966 2033 /*
1967 2034 * If we are resizing the dump device then we only need to
1968 2035 * update the refreservation to match the newly updated
1969 2036 * zvolsize. Otherwise, we save off the original state of the
1970 2037 * zvol so that we can restore them if the zvol is ever undumpified.
1971 2038 */
1972 2039 if (resize) {
1973 2040 error = zap_update(os, ZVOL_ZAP_OBJ,
1974 2041 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1975 2042 &zv->zv_volsize, tx);
1976 2043 } else {
1977 2044 error = zap_update(os, ZVOL_ZAP_OBJ,
1978 2045 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1979 2046 &compress, tx);
1980 2047 if (error == 0) {
1981 2048 error = zap_update(os, ZVOL_ZAP_OBJ,
1982 2049 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
1983 2050 &checksum, tx);
1984 2051 }
1985 2052 if (error == 0) {
1986 2053 error = zap_update(os, ZVOL_ZAP_OBJ,
1987 2054 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1988 2055 &refresrv, tx);
1989 2056 }
1990 2057 if (error == 0) {
1991 2058 error = zap_update(os, ZVOL_ZAP_OBJ,
1992 2059 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1993 2060 &vbs, tx);
1994 2061 }
1995 2062 if (error == 0) {
1996 2063 error = dmu_object_set_blocksize(
1997 2064 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
1998 2065 }
1999 2066 if (version >= SPA_VERSION_DEDUP && error == 0) {
2000 2067 error = zap_update(os, ZVOL_ZAP_OBJ,
2001 2068 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2002 2069 &dedup, tx);
2003 2070 }
2004 2071 if (error == 0)
2005 2072 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2006 2073 }
2007 2074 dmu_tx_commit(tx);
2008 2075
2009 2076 /*
2010 2077 * We only need update the zvol's property if we are initializing
2011 2078 * the dump area for the first time.
2012 2079 */
2013 2080 if (error == 0 && !resize) {
2014 2081 /*
2015 2082 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2016 2083 * function. Otherwise, use the old default -- OFF.
2017 2084 */
2018 2085 checksum = spa_feature_is_active(spa,
2019 2086 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2020 2087 ZIO_CHECKSUM_OFF;
2021 2088
2022 2089 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2023 2090 VERIFY(nvlist_add_uint64(nv,
2024 2091 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2025 2092 VERIFY(nvlist_add_uint64(nv,
2026 2093 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2027 2094 ZIO_COMPRESS_OFF) == 0);
2028 2095 VERIFY(nvlist_add_uint64(nv,
2029 2096 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2030 2097 checksum) == 0);
2031 2098 if (version >= SPA_VERSION_DEDUP) {
2032 2099 VERIFY(nvlist_add_uint64(nv,
2033 2100 zfs_prop_to_name(ZFS_PROP_DEDUP),
2034 2101 ZIO_CHECKSUM_OFF) == 0);
2035 2102 }
2036 2103
2037 2104 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2038 2105 nv, NULL);
2039 2106 nvlist_free(nv);
2040 2107 }
2041 2108
2042 2109 /* Allocate the space for the dump */
2043 2110 if (error == 0)
2044 2111 error = zvol_prealloc(zv);
2045 2112 return (error);
2046 2113 }
2047 2114
2048 2115 static int
2049 2116 zvol_dumpify(zvol_state_t *zv)
2050 2117 {
2051 2118 int error = 0;
2052 2119 uint64_t dumpsize = 0;
2053 2120 dmu_tx_t *tx;
2054 2121 objset_t *os = zv->zv_objset;
2055 2122
2056 2123 if (zv->zv_flags & ZVOL_RDONLY)
2057 2124 return (SET_ERROR(EROFS));
2058 2125
2059 2126 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2060 2127 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2061 2128 boolean_t resize = (dumpsize > 0);
2062 2129
2063 2130 if ((error = zvol_dump_init(zv, resize)) != 0) {
2064 2131 (void) zvol_dump_fini(zv);
2065 2132 return (error);
2066 2133 }
2067 2134 }
2068 2135
2069 2136 /*
2070 2137 * Build up our lba mapping.
2071 2138 */
2072 2139 error = zvol_get_lbas(zv);
2073 2140 if (error) {
2074 2141 (void) zvol_dump_fini(zv);
2075 2142 return (error);
2076 2143 }
2077 2144
2078 2145 tx = dmu_tx_create(os);
2079 2146 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2080 2147 error = dmu_tx_assign(tx, TXG_WAIT);
2081 2148 if (error) {
2082 2149 dmu_tx_abort(tx);
2083 2150 (void) zvol_dump_fini(zv);
2084 2151 return (error);
2085 2152 }
2086 2153
2087 2154 zv->zv_flags |= ZVOL_DUMPIFIED;
2088 2155 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2089 2156 &zv->zv_volsize, tx);
2090 2157 dmu_tx_commit(tx);
2091 2158
2092 2159 if (error) {
2093 2160 (void) zvol_dump_fini(zv);
2094 2161 return (error);
2095 2162 }
2096 2163
2097 2164 txg_wait_synced(dmu_objset_pool(os), 0);
2098 2165 return (0);
2099 2166 }
2100 2167
2101 2168 static int
2102 2169 zvol_dump_fini(zvol_state_t *zv)
2103 2170 {
2104 2171 dmu_tx_t *tx;
2105 2172 objset_t *os = zv->zv_objset;
2106 2173 nvlist_t *nv;
2107 2174 int error = 0;
2108 2175 uint64_t checksum, compress, refresrv, vbs, dedup;
2109 2176 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2110 2177
2111 2178 /*
2112 2179 * Attempt to restore the zvol back to its pre-dumpified state.
2113 2180 * This is a best-effort attempt as it's possible that not all
2114 2181 * of these properties were initialized during the dumpify process
2115 2182 * (i.e. error during zvol_dump_init).
2116 2183 */
2117 2184
2118 2185 tx = dmu_tx_create(os);
2119 2186 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2120 2187 error = dmu_tx_assign(tx, TXG_WAIT);
2121 2188 if (error) {
2122 2189 dmu_tx_abort(tx);
2123 2190 return (error);
2124 2191 }
2125 2192 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2126 2193 dmu_tx_commit(tx);
2127 2194
2128 2195 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2129 2196 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2130 2197 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2131 2198 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2132 2199 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2133 2200 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2134 2201 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2135 2202 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2136 2203
2137 2204 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2138 2205 (void) nvlist_add_uint64(nv,
2139 2206 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2140 2207 (void) nvlist_add_uint64(nv,
2141 2208 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2142 2209 (void) nvlist_add_uint64(nv,
2143 2210 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2144 2211 if (version >= SPA_VERSION_DEDUP &&
2145 2212 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2146 2213 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2147 2214 (void) nvlist_add_uint64(nv,
2148 2215 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2149 2216 }
2150 2217 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2151 2218 nv, NULL);
2152 2219 nvlist_free(nv);
2153 2220
2154 2221 zvol_free_extents(zv);
2155 2222 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2156 2223 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2157 2224 /* wait for dmu_free_long_range to actually free the blocks */
2158 2225 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2159 2226 tx = dmu_tx_create(os);
2160 2227 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2161 2228 error = dmu_tx_assign(tx, TXG_WAIT);
2162 2229 if (error) {
2163 2230 dmu_tx_abort(tx);
2164 2231 return (error);
2165 2232 }
2166 2233 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2167 2234 zv->zv_volblocksize = vbs;
2168 2235 dmu_tx_commit(tx);
2169 2236
2170 2237 return (0);
2171 2238 }
|
↓ open down ↓ |
326 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX