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