?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Patrick Welche <prlw1@cam.ac.uk>
[netbsd-mini2440.git] / external / cddl / osnet / dist / uts / common / fs / zfs / zvol.c
blobf04e4476760ec061b625d090234741feb13996aa
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
26 /*
27 * ZFS volume emulation driver.
28 *
29 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
30 * Volumes are accessed through the symbolic links named:
31 *
32 * /dev/zvol/dsk/<pool_name>/<dataset_name>
33 * /dev/zvol/rdsk/<pool_name>/<dataset_name>
34 *
35 * These links are created by the ZFS-specific devfsadm link generator.
36 * Volumes are persistent through reboot. No user command needs to be
37 * run before opening and using a device.
38 */
39
40 #include <sys/types.h>
41 #include <sys/param.h>
42 #include <sys/errno.h>
43 #include <sys/uio.h>
44 #include <sys/buf.h>
45 #include <sys/modctl.h>
46 #include <sys/open.h>
47 #include <sys/kmem.h>
48 #include <sys/conf.h>
49 #include <sys/cmn_err.h>
50 #include <sys/stat.h>
51 #include <sys/zap.h>
52 #include <sys/spa.h>
53 #include <sys/zio.h>
54 #include <sys/dmu_traverse.h>
55 #include <sys/dnode.h>
56 #include <sys/dsl_dataset.h>
57 #include <sys/dsl_prop.h>
58 #include <sys/dkio.h>
59 #include <sys/efi_partition.h>
60 #include <sys/byteorder.h>
61 #include <sys/pathname.h>
62 #include <sys/ddi.h>
63 #include <sys/sunddi.h>
64 #include <sys/crc32.h>
65 #include <sys/dirent.h>
66 #include <sys/policy.h>
67 #include <sys/fs/zfs.h>
68 #include <sys/zfs_ioctl.h>
69 #include <sys/mkdev.h>
70 #include <sys/zil.h>
71 #include <sys/refcount.h>
72 #include <sys/zfs_znode.h>
73 #include <sys/zfs_rlock.h>
74 #include <sys/vdev_disk.h>
75 #include <sys/vdev_impl.h>
76 #include <sys/zvol.h>
77 #include <sys/disk.h>
78 #include <sys/dkio.h>
79 #include <sys/disklabel.h>
80
81 #ifdef __NetBSD__
82 #include <prop/proplib.h>
83 #endif
84
85 #include "zfs_namecheck.h"
86
87 static void *zvol_state;
88
89 #define ZVOL_DUMPSIZE "dumpsize"
90
91 void zvol_minphys(struct buf *);
92
93 static struct dkdriver zvol_dkdriver = { zvol_strategy, zvol_minphys };
94
95 /*
96 * This lock protects the zvol_state structure from being modified
97 * while it's being used, e.g. an open that comes in before a create
98 * finishes. It also protects temporary opens of the dataset so that,
99 * e.g., an open doesn't get a spurious EBUSY.
100 */
101 static kmutex_t zvol_state_lock;
102 static uint32_t zvol_minors;
103
104 typedef struct zvol_extent {
105 list_node_t ze_node;
106 dva_t ze_dva; /* dva associated with this extent */
107 uint64_t ze_nblks; /* number of blocks in extent */
108 } zvol_extent_t;
109
110 /*
111 * The in-core state of each volume.
112 */
113 typedef struct zvol_state {
114 char zv_name[MAXPATHLEN]; /* pool/dd name */
115 uint64_t zv_volsize; /* amount of space we advertise */
116 uint64_t zv_volblocksize; /* volume block size */
117 minor_t zv_minor; /* minor number */
118 uint8_t zv_min_bs; /* minimum addressable block shift */
119 uint8_t zv_flags; /* readonly; dumpified */
120 objset_t *zv_objset; /* objset handle */
121 uint32_t zv_mode; /* DS_MODE_* flags at open time */
122 uint32_t zv_open_count[OTYPCNT]; /* open counts */
123 uint32_t zv_total_opens; /* total open count */
124 zilog_t *zv_zilog; /* ZIL handle */
125 list_t zv_extents; /* List of extents for dump */
126 uint64_t zv_txg_assign; /* txg to assign during ZIL replay */
127 znode_t zv_znode; /* for range locking */
128 #ifdef __NetBSD__
129 struct disk zv_dk; /* disk statistics */
130 kmutex_t zv_dklock; /* disk statistics */
131 #endif
132 } zvol_state_t;
133
134 /*
135 * zvol specific flags
136 */
137 #define ZVOL_RDONLY 0x1
138 #define ZVOL_DUMPIFIED 0x2
139 #define ZVOL_EXCL 0x4
140
141 /*
142 * zvol maximum transfer in one DMU tx.
143 */
144 int zvol_maxphys = DMU_MAX_ACCESS/2;
145
146 extern int zfs_set_prop_nvlist(const char *, nvlist_t *);
147 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
148 static int zvol_dumpify(zvol_state_t *zv);
149 static int zvol_dump_fini(zvol_state_t *zv);
150 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
151
152 static void
153 zvol_size_changed(zvol_state_t *zv, major_t maj)
154 {
155 #ifdef __NetBSD__
156 prop_dictionary_t disk_info, odisk_info, geom;
157 struct disk *disk;
158
159 disk = &zv->zv_dk;
160
161 disk_info = prop_dictionary_create();
162
163 prop_dictionary_set_cstring_nocopy(disk_info, "type", "ESDI");
164
165 geom = prop_dictionary_create();
166
167 prop_dictionary_set_uint64(geom, "sectors-per-unit", zv->zv_volsize);
168
169 prop_dictionary_set_uint32(geom, "sector-size",
170 DEV_BSIZE /* XXX 512? */);
171
172 prop_dictionary_set_uint32(geom, "sectors-per-track", 32);
173
174 prop_dictionary_set_uint32(geom, "tracks-per-cylinder", 64);
175
176 prop_dictionary_set_uint32(geom, "cylinders-per-unit", zv->zv_volsize / 2048);
177
178 prop_dictionary_set(disk_info, "geometry", geom);
179 prop_object_release(geom);
180
181 odisk_info = disk->dk_info;
182
183 disk->dk_info = disk_info;
184
185 if (odisk_info != NULL)
186 prop_object_release(odisk_info);
187
188 #else
189 dev_t dev = makedevice(maj, zv->zv_minor);
190
191 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
192 "Size", zv->zv_volsize) == DDI_SUCCESS);
193 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
194 "Nblocks", lbtodb(zv->zv_volsize)) == DDI_SUCCESS);
195
196 /* Notify specfs to invalidate the cached size */
197 spec_size_invalidate(dev, VBLK);
198 spec_size_invalidate(dev, VCHR);
199 #endif
200 }
201
202 int
203 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
204 {
205 if (volsize == 0)
206 return (EINVAL);
207
208 if (volsize % blocksize != 0)
209 return (EINVAL);
210
211 #ifdef _ILP32
212 if (volsize - 1 > SPEC_MAXOFFSET_T)
213 return (EOVERFLOW);
214 #endif
215 return (0);
216 }
217
218 int
219 zvol_check_volblocksize(uint64_t volblocksize)
220 {
221 if (volblocksize < SPA_MINBLOCKSIZE ||
222 volblocksize > SPA_MAXBLOCKSIZE ||
223 !ISP2(volblocksize))
224 return (EDOM);
225
226 return (0);
227 }
228
229 static void
230 zvol_readonly_changed_cb(void *arg, uint64_t newval)
231 {
232 zvol_state_t *zv = arg;
233
234 if (newval)
235 zv->zv_flags |= ZVOL_RDONLY;
236 else
237 zv->zv_flags &= ~ZVOL_RDONLY;
238 }
239
240 int
241 zvol_get_stats(objset_t *os, nvlist_t *nv)
242 {
243 int error;
244 dmu_object_info_t doi;
245 uint64_t val;
246
247
248 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
249 if (error)
250 return (error);
251
252 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
253
254 error = dmu_object_info(os, ZVOL_OBJ, &doi);
255
256 if (error == 0) {
257 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
258 doi.doi_data_block_size);
259 }
260
261 return (error);
262 }
263
264 /*
265 * Find a free minor number.
266 */
267 static minor_t
268 zvol_minor_alloc(void)
269 {
270 minor_t minor;
271
272 ASSERT(MUTEX_HELD(&zvol_state_lock));
273
274 for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++)
275 if (ddi_get_soft_state(zvol_state, minor) == NULL)
276 return (minor);
277
278 return (0);
279 }
280
281 static zvol_state_t *
282 zvol_minor_lookup(const char *name)
283 {
284 minor_t minor;
285 zvol_state_t *zv;
286
287 ASSERT(MUTEX_HELD(&zvol_state_lock));
288
289 for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++) {
290 zv = ddi_get_soft_state(zvol_state, minor);
291 if (zv == NULL)
292 continue;
293 if (strcmp(zv->zv_name, name) == 0)
294 break;
295 }
296
297 return (zv);
298 }
299
300 /* extent mapping arg */
301 struct maparg {
302 zvol_state_t *ma_zv;
303 uint64_t ma_blks;
304 };
305
306 /*ARGSUSED*/
307 static int
308 zvol_map_block(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb,
309 const dnode_phys_t *dnp, void *arg)
310 {
311 struct maparg *ma = arg;
312 zvol_extent_t *ze;
313 int bs = ma->ma_zv->zv_volblocksize;
314
315 if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
316 return (0);
317
318 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
319 ma->ma_blks++;
320
321 /* Abort immediately if we have encountered gang blocks */
322 if (BP_IS_GANG(bp))
323 return (EFRAGS);
324
325 /*
326 * See if the block is at the end of the previous extent.
327 */
328 ze = list_tail(&ma->ma_zv->zv_extents);
329 if (ze &&
330 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
331 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
332 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
333 ze->ze_nblks++;
334 return (0);
335 }
336
337 dprintf_bp(bp, "%s", "next blkptr:");
338
339 /* start a new extent */
340 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
341 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
342 ze->ze_nblks = 1;
343 list_insert_tail(&ma->ma_zv->zv_extents, ze);
344 return (0);
345 }
346
347 static void
348 zvol_free_extents(zvol_state_t *zv)
349 {
350 zvol_extent_t *ze;
351
352 while (ze = list_head(&zv->zv_extents)) {
353 list_remove(&zv->zv_extents, ze);
354 kmem_free(ze, sizeof (zvol_extent_t));
355 }
356 }
357
358 static int
359 zvol_get_lbas(zvol_state_t *zv)
360 {
361 struct maparg ma;
362 int err;
363
364 ma.ma_zv = zv;
365 ma.ma_blks = 0;
366 zvol_free_extents(zv);
367
368 err = traverse_dataset(dmu_objset_ds(zv->zv_objset), 0,
369 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
370 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
371 zvol_free_extents(zv);
372 return (err ? err : EIO);
373 }
374
375 return (0);
376 }
377
378 /* ARGSUSED */
379 void
380 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
381 {
382 zfs_creat_t *zct = arg;
383 nvlist_t *nvprops = zct->zct_props;
384 int error;
385 uint64_t volblocksize, volsize;
386
387 VERIFY(nvlist_lookup_uint64(nvprops,
388 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
389 if (nvlist_lookup_uint64(nvprops,
390 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
391 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
392
393 /*
394 * These properties must be removed from the list so the generic
395 * property setting step won't apply to them.
396 */
397 VERIFY(nvlist_remove_all(nvprops,
398 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
399 (void) nvlist_remove_all(nvprops,
400 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
401
402 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
403 DMU_OT_NONE, 0, tx);
404 ASSERT(error == 0);
405
406 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
407 DMU_OT_NONE, 0, tx);
408 ASSERT(error == 0);
409
410 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
411 ASSERT(error == 0);
412 }
413
414 /*
415 * Replay a TX_WRITE ZIL transaction that didn't get committed
416 * after a system failure
417 */
418 static int
419 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
420 {
421 objset_t *os = zv->zv_objset;
422 char *data = (char *)(lr + 1); /* data follows lr_write_t */
423 uint64_t off = lr->lr_offset;
424 uint64_t len = lr->lr_length;
425 dmu_tx_t *tx;
426 int error;
427
428 if (byteswap)
429 byteswap_uint64_array(lr, sizeof (*lr));
430
431 tx = dmu_tx_create(os);
432 dmu_tx_hold_write(tx, ZVOL_OBJ, off, len);
433 error = dmu_tx_assign(tx, zv->zv_txg_assign);
434 if (error) {
435 dmu_tx_abort(tx);
436 } else {
437 dmu_write(os, ZVOL_OBJ, off, len, data, tx);
438 dmu_tx_commit(tx);
439 }
440
441 return (error);
442 }
443
444 /* ARGSUSED */
445 static int
446 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
447 {
448 return (ENOTSUP);
449 }
450
451 /*
452 * Callback vectors for replaying records.
453 * Only TX_WRITE is needed for zvol.
454 */
455 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
456 zvol_replay_err, /* 0 no such transaction type */
457 zvol_replay_err, /* TX_CREATE */
458 zvol_replay_err, /* TX_MKDIR */
459 zvol_replay_err, /* TX_MKXATTR */
460 zvol_replay_err, /* TX_SYMLINK */
461 zvol_replay_err, /* TX_REMOVE */
462 zvol_replay_err, /* TX_RMDIR */
463 zvol_replay_err, /* TX_LINK */
464 zvol_replay_err, /* TX_RENAME */
465 zvol_replay_write, /* TX_WRITE */
466 zvol_replay_err, /* TX_TRUNCATE */
467 zvol_replay_err, /* TX_SETATTR */
468 zvol_replay_err, /* TX_ACL */
469 };
470
471 /*
472 * Create a minor node (plus a whole lot more) for the specified volume.
473 */
474 int
475 zvol_create_minor(const char *name, major_t maj)
476 {
477 zvol_state_t *zv;
478 objset_t *os;
479 dmu_object_info_t doi;
480 uint64_t volsize;
481 minor_t minor = 0;
482 int ds_mode = DS_MODE_OWNER;
483 vnode_t *vp = NULL;
484 char *devpath;
485 size_t devpathlen = strlen(ZVOL_FULL_DEV_DIR) + strlen(name) + 1;
486 int error;
487
488 mutex_enter(&zvol_state_lock);
489
490 if ((zv = zvol_minor_lookup(name)) != NULL) {
491 mutex_exit(&zvol_state_lock);
492 return (EEXIST);
493 }
494
495 if (strchr(name, '@') != 0)
496 ds_mode |= DS_MODE_READONLY;
497
498 error = dmu_objset_open(name, DMU_OST_ZVOL, ds_mode, &os);
499
500 if (error) {
501 mutex_exit(&zvol_state_lock);
502 return (error);
503 }
504
505 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
506
507 if (error) {
508 dmu_objset_close(os);
509 mutex_exit(&zvol_state_lock);
510 return (error);
511 }
512
513 /*
514 * If there's an existing /dev/zvol symlink, try to use the
515 * same minor number we used last time.
516 */
517 devpath = kmem_alloc(devpathlen, KM_SLEEP);
518
519 /* Get full path to ZFS volume disk device */
520 (void) sprintf(devpath, "%s/%s", ZVOL_FULL_DEV_DIR, name);
521
522 error = lookupname(devpath, UIO_SYSSPACE, NULL, &vp);
523
524 if (error == 0 && vp->v_type != VBLK) {
525 error = EINVAL;
526 }
527
528 if (error == 0) {
529 struct stat sb;
530 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
531 error = vn_stat(vp, &sb);
532 VOP_UNLOCK(vp, 0);
533 if (error == 0) {
534 minor = getminor(sb.st_rdev);
535 }
536 }
537
538 if (vp != NULL)
539 VN_RELE(vp);
540
541 /*
542 * If we found a minor but it's already in use, we must pick a new one.
543 */
544 if (minor != 0 && ddi_get_soft_state(zvol_state, minor) != NULL)
545 minor = 0;
546
547 if (minor == 0)
548 minor = zvol_minor_alloc();
549
550 if (minor == 0) {
551 dmu_objset_close(os);
552 mutex_exit(&zvol_state_lock);
553 kmem_free(devpath, devpathlen);
554 return (ENXIO);
555 }
556
557 if (ddi_soft_state_zalloc(zvol_state, minor) != DDI_SUCCESS) {
558 dmu_objset_close(os);
559 mutex_exit(&zvol_state_lock);
560 kmem_free(devpath, devpathlen);
561 return (EAGAIN);
562 }
563
564 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
565 (char *)name);
566
567 if (ddi_create_minor_node(zfs_dip, (char *)name, S_IFCHR,
568 minor, DDI_PSEUDO, maj) == DDI_FAILURE) {
569 ddi_soft_state_free(zvol_state, minor);
570 dmu_objset_close(os);
571 mutex_exit(&zvol_state_lock);
572 kmem_free(devpath, devpathlen);
573 return (EAGAIN);
574 }
575
576 if (ddi_create_minor_node(zfs_dip, (char *)name, S_IFBLK,
577 minor, DDI_PSEUDO, maj) == DDI_FAILURE) {
578 ddi_remove_minor_node(zfs_dip, (char *)name);
579 ddi_soft_state_free(zvol_state, minor);
580 dmu_objset_close(os);
581 mutex_exit(&zvol_state_lock);
582 kmem_free(devpath, devpathlen);
583 return (EAGAIN);
584 }
585 zv = ddi_get_soft_state(zvol_state, minor);
586
587 (void) strcpy(zv->zv_name, name);
588 zv->zv_min_bs = DEV_BSHIFT;
589 zv->zv_minor = minor;
590 zv->zv_volsize = volsize;
591 zv->zv_objset = os;
592 zv->zv_mode = ds_mode;
593 zv->zv_zilog = zil_open(os, zvol_get_data);
594 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
595 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
596 sizeof (rl_t), offsetof(rl_t, r_node));
597 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
598 offsetof(zvol_extent_t, ze_node));
599 /* get and cache the blocksize */
600 error = dmu_object_info(os, ZVOL_OBJ, &doi);
601 ASSERT(error == 0);
602 zv->zv_volblocksize = doi.doi_data_block_size;
603
604 disk_init(&zv->zv_dk, name, &zvol_dkdriver);
605 disk_attach(&zv->zv_dk);
606 mutex_init(&zv->zv_dklock, NULL, MUTEX_DEFAULT, NULL);
607
608 zil_replay(os, zv, &zv->zv_txg_assign, zvol_replay_vector, NULL);
609 zvol_size_changed(zv, maj);
610
611 /* XXX this should handle the possible i/o error */
612 VERIFY(dsl_prop_register(dmu_objset_ds(zv->zv_objset),
613 "readonly", zvol_readonly_changed_cb, zv) == 0);
614
615 zvol_minors++;
616
617 mutex_exit(&zvol_state_lock);
618
619 // kmem_free(devpath, devpathlen);
620
621 return (0);
622 }
623
624 /*
625 * Remove minor node for the specified volume.
626 */
627 int
628 zvol_remove_minor(const char *name)
629 {
630 zvol_state_t *zv;
631 char namebuf[30], blkbuf[30];
632
633 mutex_enter(&zvol_state_lock);
634
635 if ((zv = zvol_minor_lookup(name)) == NULL) {
636 mutex_exit(&zvol_state_lock);
637 return (ENXIO);
638 }
639
640 if (zv->zv_total_opens != 0) {
641 mutex_exit(&zvol_state_lock);
642 return (EBUSY);
643 }
644
645 (void) snprintf(namebuf, sizeof(namebuf), "%s", name);
646 ddi_remove_minor_node(zfs_dip, namebuf);
647
648 (void) snprintf(blkbuf, sizeof(blkbuf), "%s", name);
649 ddi_remove_minor_node(zfs_dip, blkbuf);
650
651 VERIFY(dsl_prop_unregister(dmu_objset_ds(zv->zv_objset),
652 "readonly", zvol_readonly_changed_cb, zv) == 0);
653
654 zil_close(zv->zv_zilog);
655 zv->zv_zilog = NULL;
656 dmu_objset_close(zv->zv_objset);
657 zv->zv_objset = NULL;
658 avl_destroy(&zv->zv_znode.z_range_avl);
659 mutex_destroy(&zv->zv_znode.z_range_lock);
660
661 ddi_soft_state_free(zvol_state, zv->zv_minor);
662
663 zvol_minors--;
664
665 disk_detach(&zv->zv_dk);
666 disk_destroy(&zv->zv_dk);
667 mutex_destroy(&zv->zv_dklock);
668
669 mutex_exit(&zvol_state_lock);
670
671 return (0);
672 }
673
674 int
675 zvol_prealloc(zvol_state_t *zv)
676 {
677 objset_t *os = zv->zv_objset;
678 dmu_tx_t *tx;
679 uint64_t refd, avail, usedobjs, availobjs;
680 uint64_t resid = zv->zv_volsize;
681 uint64_t off = 0;
682
683 /* Check the space usage before attempting to allocate the space */
684 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
685 if (avail < zv->zv_volsize)
686 return (ENOSPC);
687
688 /* Free old extents if they exist */
689 zvol_free_extents(zv);
690
691 while (resid != 0) {
692 int error;
693 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE);
694
695 tx = dmu_tx_create(os);
696 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
697 error = dmu_tx_assign(tx, TXG_WAIT);
698 if (error) {
699 dmu_tx_abort(tx);
700 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
701 return (error);
702 }
703 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
704 dmu_tx_commit(tx);
705 off += bytes;
706 resid -= bytes;
707 }
708 txg_wait_synced(dmu_objset_pool(os), 0);
709
710 return (0);
711 }
712
713 int
714 zvol_update_volsize(zvol_state_t *zv, major_t maj, uint64_t volsize)
715 {
716 dmu_tx_t *tx;
717 int error;
718
719 ASSERT(MUTEX_HELD(&zvol_state_lock));
720
721 tx = dmu_tx_create(zv->zv_objset);
722 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
723 error = dmu_tx_assign(tx, TXG_WAIT);
724 if (error) {
725 dmu_tx_abort(tx);
726 return (error);
727 }
728
729 error = zap_update(zv->zv_objset, ZVOL_ZAP_OBJ, "size", 8, 1,
730 &volsize, tx);
731 dmu_tx_commit(tx);
732
733 if (error == 0)
734 error = dmu_free_long_range(zv->zv_objset,
735 ZVOL_OBJ, volsize, DMU_OBJECT_END);
736
737 /*
738 * If we are using a faked-up state (zv_minor == 0) then don't
739 * try to update the in-core zvol state.
740 */
741 if (error == 0 && zv->zv_minor) {
742 zv->zv_volsize = volsize;
743 zvol_size_changed(zv, maj);
744 }
745 return (error);
746 }
747
748 int
749 zvol_set_volsize(const char *name, major_t maj, uint64_t volsize)
750 {
751 zvol_state_t *zv;
752 int error;
753 dmu_object_info_t doi;
754 uint64_t old_volsize = 0ULL;
755 zvol_state_t state = { 0 };
756
757 mutex_enter(&zvol_state_lock);
758
759 if ((zv = zvol_minor_lookup(name)) == NULL) {
760 /*
761 * If we are doing a "zfs clone -o volsize=", then the
762 * minor node won't exist yet.
763 */
764 error = dmu_objset_open(name, DMU_OST_ZVOL, DS_MODE_OWNER,
765 &state.zv_objset);
766 if (error != 0)
767 goto out;
768 zv = &state;
769 }
770 old_volsize = zv->zv_volsize;
771
772 if ((error = dmu_object_info(zv->zv_objset, ZVOL_OBJ, &doi)) != 0 ||
773 (error = zvol_check_volsize(volsize,
774 doi.doi_data_block_size)) != 0)
775 goto out;
776
777 if (zv->zv_flags & ZVOL_RDONLY || (zv->zv_mode & DS_MODE_READONLY)) {
778 error = EROFS;
779 goto out;
780 }
781
782 error = zvol_update_volsize(zv, maj, volsize);
783
784 #ifndef __NetBSD__
785 /*
786 * Reinitialize the dump area to the new size. If we
787 * failed to resize the dump area then restore the it back to
788 * it's original size.
789 */
790 if (error == 0 && zv->zv_flags & ZVOL_DUMPIFIED) {
791 if ((error = zvol_dumpify(zv)) != 0 ||
792 (error = dumpvp_resize()) != 0) {
793 (void) zvol_update_volsize(zv, maj, old_volsize);
794 error = zvol_dumpify(zv);
795 }
796 }
797 #endif
798
799 out:
800 if (state.zv_objset)
801 dmu_objset_close(state.zv_objset);
802
803 mutex_exit(&zvol_state_lock);
804
805 return (error);
806 }
807
808 int
809 zvol_set_volblocksize(const char *name, uint64_t volblocksize)
810 {
811 zvol_state_t *zv;
812 dmu_tx_t *tx;
813 int error;
814 boolean_t needlock;
815
816 /*
817 * The lock may already be held if we are being called from
818 * zvol_dump_init().
819 */
820 needlock = !MUTEX_HELD(&zvol_state_lock);
821 if (needlock)
822 mutex_enter(&zvol_state_lock);
823
824 if ((zv = zvol_minor_lookup(name)) == NULL) {
825 if (needlock)
826 mutex_exit(&zvol_state_lock);
827 return (ENXIO);
828 }
829 if (zv->zv_flags & ZVOL_RDONLY || (zv->zv_mode & DS_MODE_READONLY)) {
830 if (needlock)
831 mutex_exit(&zvol_state_lock);
832 return (EROFS);
833 }
834
835 tx = dmu_tx_create(zv->zv_objset);
836 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
837 error = dmu_tx_assign(tx, TXG_WAIT);
838 if (error) {
839 dmu_tx_abort(tx);
840 } else {
841 error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ,
842 volblocksize, 0, tx);
843 if (error == ENOTSUP)
844 error = EBUSY;
845 dmu_tx_commit(tx);
846 if (error == 0)
847 zv->zv_volblocksize = volblocksize;
848 }
849
850 if (needlock)
851 mutex_exit(&zvol_state_lock);
852
853 return (error);
854 }
855
856 /*ARGSUSED*/
857 int
858 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
859 {
860 minor_t minor = getminor(*devp);
861 zvol_state_t *zv;
862
863 if (minor == 0) /* This is the control device */
864 return (0);
865
866 mutex_enter(&zvol_state_lock);
867
868 zv = ddi_get_soft_state(zvol_state, minor);
869 if (zv == NULL) {
870 mutex_exit(&zvol_state_lock);
871 return (ENXIO);
872 }
873
874 ASSERT(zv->zv_objset != NULL);
875
876 if ((flag & FWRITE) &&
877 (zv->zv_flags & ZVOL_RDONLY || (zv->zv_mode & DS_MODE_READONLY))) {
878 mutex_exit(&zvol_state_lock);
879 return (EROFS);
880 }
881 if (zv->zv_flags & ZVOL_EXCL) {
882 mutex_exit(&zvol_state_lock);
883 return (EBUSY);
884 }
885 if (flag & FEXCL) {
886 if (zv->zv_total_opens != 0) {
887 mutex_exit(&zvol_state_lock);
888 return (EBUSY);
889 }
890 zv->zv_flags |= ZVOL_EXCL;
891 }
892
893 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
894 zv->zv_open_count[otyp]++;
895 zv->zv_total_opens++;
896 }
897
898 mutex_exit(&zvol_state_lock);
899 return (0);
900 }
901
902 /*ARGSUSED*/
903 int
904 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
905 {
906 minor_t minor = getminor(dev);
907 zvol_state_t *zv;
908
909 if (minor == 0) /* This is the control device */
910 return (0);
911
912 mutex_enter(&zvol_state_lock);
913
914 zv = ddi_get_soft_state(zvol_state, minor);
915 if (zv == NULL) {
916 mutex_exit(&zvol_state_lock);
917 return (ENXIO);
918 }
919
920 if (zv->zv_flags & ZVOL_EXCL) {
921 ASSERT(zv->zv_total_opens == 1);
922 zv->zv_flags &= ~ZVOL_EXCL;
923 }
924
925 /*
926 * If the open count is zero, this is a spurious close.
927 * That indicates a bug in the kernel / DDI framework.
928 */
929 ASSERT(zv->zv_open_count[otyp] != 0);
930 ASSERT(zv->zv_total_opens != 0);
931
932 /*
933 * You may get multiple opens, but only one close.
934 */
935 zv->zv_open_count[otyp]--;
936 zv->zv_total_opens--;
937
938 mutex_exit(&zvol_state_lock);
939
940 return (0);
941 }
942
943 static void
944 zvol_get_done(dmu_buf_t *db, void *vzgd)
945 {
946 zgd_t *zgd = (zgd_t *)vzgd;
947 rl_t *rl = zgd->zgd_rl;
948
949 dmu_buf_rele(db, vzgd);
950 zfs_range_unlock(rl);
951 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
952 kmem_free(zgd, sizeof (zgd_t));
953 }
954
955 /*
956 * Get data to generate a TX_WRITE intent log record.
957 */
958 static int
959 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
960 {
961 zvol_state_t *zv = arg;
962 objset_t *os = zv->zv_objset;
963 dmu_buf_t *db;
964 rl_t *rl;
965 zgd_t *zgd;
966 uint64_t boff; /* block starting offset */
967 int dlen = lr->lr_length; /* length of user data */
968 int error;
969
970 ASSERT(zio);
971 ASSERT(dlen != 0);
972
973 /*
974 * Write records come in two flavors: immediate and indirect.
975 * For small writes it's cheaper to store the data with the
976 * log record (immediate); for large writes it's cheaper to
977 * sync the data and get a pointer to it (indirect) so that
978 * we don't have to write the data twice.
979 */
980 if (buf != NULL) /* immediate write */
981 return (dmu_read(os, ZVOL_OBJ, lr->lr_offset, dlen, buf));
982
983 zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP);
984 zgd->zgd_zilog = zv->zv_zilog;
985 zgd->zgd_bp = &lr->lr_blkptr;
986
987 /*
988 * Lock the range of the block to ensure that when the data is
989 * written out and its checksum is being calculated that no other
990 * thread can change the block.
991 */
992 boff = P2ALIGN_TYPED(lr->lr_offset, zv->zv_volblocksize, uint64_t);
993 rl = zfs_range_lock(&zv->zv_znode, boff, zv->zv_volblocksize,
994 RL_READER);
995 zgd->zgd_rl = rl;
996
997 VERIFY(0 == dmu_buf_hold(os, ZVOL_OBJ, lr->lr_offset, zgd, &db));
998 error = dmu_sync(zio, db, &lr->lr_blkptr,
999 lr->lr_common.lrc_txg, zvol_get_done, zgd);
1000 if (error == 0)
1001 zil_add_block(zv->zv_zilog, &lr->lr_blkptr);
1002 /*
1003 * If we get EINPROGRESS, then we need to wait for a
1004 * write IO initiated by dmu_sync() to complete before
1005 * we can release this dbuf. We will finish everything
1006 * up in the zvol_get_done() callback.
1007 */
1008 if (error == EINPROGRESS)
1009 return (0);
1010 dmu_buf_rele(db, zgd);
1011 zfs_range_unlock(rl);
1012 kmem_free(zgd, sizeof (zgd_t));
1013 return (error);
1014 }
1015
1016 /*
1017 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1018 *
1019 * We store data in the log buffers if it's small enough.
1020 * Otherwise we will later flush the data out via dmu_sync().
1021 */
1022 ssize_t zvol_immediate_write_sz = 32768;
1023
1024 static void
1025 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t len)
1026 {
1027 uint32_t blocksize = zv->zv_volblocksize;
1028 lr_write_t *lr;
1029
1030 while (len) {
1031 ssize_t nbytes = MIN(len, blocksize - P2PHASE(off, blocksize));
1032 itx_t *itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1033
1034 itx->itx_wr_state =
1035 len > zvol_immediate_write_sz ? WR_INDIRECT : WR_NEED_COPY;
1036 itx->itx_private = zv;
1037 lr = (lr_write_t *)&itx->itx_lr;
1038 lr->lr_foid = ZVOL_OBJ;
1039 lr->lr_offset = off;
1040 lr->lr_length = nbytes;
1041 lr->lr_blkoff = off - P2ALIGN_TYPED(off, blocksize, uint64_t);
1042 BP_ZERO(&lr->lr_blkptr);
1043
1044 (void) zil_itx_assign(zv->zv_zilog, itx, tx);
1045 len -= nbytes;
1046 off += nbytes;
1047 }
1048 }
1049
1050 #ifndef __NetBSD__
1051 static int
1052 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t size,
1053 boolean_t doread, boolean_t isdump)
1054 {
1055 vdev_disk_t *dvd;
1056 int c;
1057 int numerrors = 0;
1058
1059 for (c = 0; c < vd->vdev_children; c++) {
1060 ASSERT(vd->vdev_ops == &vdev_mirror_ops);
1061 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1062 addr, offset, size, doread, isdump);
1063 if (err != 0) {
1064 numerrors++;
1065 } else if (doread) {
1066 break;
1067 }
1068 }
1069
1070 if (!vd->vdev_ops->vdev_op_leaf)
1071 return (numerrors < vd->vdev_children ? 0 : EIO);
1072
1073 if (doread && !vdev_readable(vd))
1074 return (EIO);
1075 else if (!doread && !vdev_writeable(vd))
1076 return (EIO);
1077
1078 dvd = vd->vdev_tsd;
1079 ASSERT3P(dvd, !=, NULL);
1080 offset += VDEV_LABEL_START_SIZE;
1081
1082 if (ddi_in_panic() || isdump) {
1083 ASSERT(!doread);
1084 if (doread)
1085 return (EIO);
1086 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1087 lbtodb(size)));
1088 } else {
1089 return (vdev_disk_physio(dvd->vd_lh, addr, size, offset,
1090 doread ? B_READ : B_WRITE));
1091 }
1092 }
1093
1094 static int
1095 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1096 boolean_t doread, boolean_t isdump)
1097 {
1098 vdev_t *vd;
1099 int error;
1100 zvol_extent_t *ze;
1101 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1102
1103 /* Must be sector aligned, and not stradle a block boundary. */
1104 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1105 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1106 return (EINVAL);
1107 }
1108 ASSERT(size <= zv->zv_volblocksize);
1109
1110 /* Locate the extent this belongs to */
1111 ze = list_head(&zv->zv_extents);
1112 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1113 offset -= ze->ze_nblks * zv->zv_volblocksize;
1114 ze = list_next(&zv->zv_extents, ze);
1115 }
1116 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1117 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1118 offset += DVA_GET_OFFSET(&ze->ze_dva);
1119 error = zvol_dumpio_vdev(vd, addr, offset, size, doread, isdump);
1120 spa_config_exit(spa, SCL_STATE, FTAG);
1121 return (error);
1122 }
1123 #endif /* __NetBSD__ */
1124
1125 void
1126 zvol_strategy(buf_t *bp)
1127 {
1128 zvol_state_t *zv = ddi_get_soft_state(zvol_state, getminor(bp->b_edev));
1129 uint64_t off, volsize;
1130 size_t resid;
1131 char *addr;
1132 objset_t *os;
1133 rl_t *rl;
1134 int error = 0;
1135 boolean_t doread = bp->b_flags & B_READ;
1136 boolean_t is_dump = zv->zv_flags & ZVOL_DUMPIFIED;
1137
1138 if (zv == NULL) {
1139 bioerror(bp, ENXIO);
1140 biodone(bp);
1141 return;
1142 }
1143
1144 if (getminor(bp->b_edev) == 0) {
1145 bioerror(bp, EINVAL);
1146 biodone(bp);
1147 return;
1148 }
1149
1150 if (!(bp->b_flags & B_READ) &&
1151 (zv->zv_flags & ZVOL_RDONLY ||
1152 zv->zv_mode & DS_MODE_READONLY)) {
1153 bioerror(bp, EROFS);
1154 biodone(bp);
1155 return;
1156 }
1157
1158 off = (uint64_t)bp->b_blkno * DEV_BSIZE;
1159 volsize = zv->zv_volsize;
1160
1161 os = zv->zv_objset;
1162 ASSERT(os != NULL);
1163
1164 addr = bp->b_data;
1165 resid = bp->b_bcount;
1166
1167 if (resid > 0 && (off < 0 || off >= volsize)) {
1168 bioerror(bp, EIO);
1169 biodone(bp);
1170 return;
1171 }
1172
1173 /*
1174 * There must be no buffer changes when doing a dmu_sync() because
1175 * we can't change the data whilst calculating the checksum.
1176 */
1177 mutex_enter(&zv->zv_dklock);
1178 disk_busy(&zv->zv_dk);
1179 mutex_exit(&zv->zv_dklock);
1180
1181 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1182 doread ? RL_READER : RL_WRITER);
1183
1184 while (resid != 0 && off < volsize) {
1185 size_t size = MIN(resid, zvol_maxphys);
1186 if (is_dump) {
1187 #ifdef __NetBSD__
1188 printf("XXXNETBSD zvol_strategy: how?");
1189 #else
1190 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1191 error = zvol_dumpio(zv, addr, off, size,
1192 doread, B_FALSE);
1193 #endif
1194 } else if (doread) {
1195 error = dmu_read(os, ZVOL_OBJ, off, size, addr);
1196 } else {
1197 dmu_tx_t *tx = dmu_tx_create(os);
1198 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1199 error = dmu_tx_assign(tx, TXG_WAIT);
1200 if (error) {
1201 dmu_tx_abort(tx);
1202 } else {
1203 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1204 zvol_log_write(zv, tx, off, size);
1205 dmu_tx_commit(tx);
1206 }
1207 }
1208 if (error) {
1209 /* convert checksum errors into IO errors */
1210 if (error == ECKSUM)
1211 error = EIO;
1212 break;
1213 }
1214 off += size;
1215 addr += size;
1216 resid -= size;
1217 }
1218 zfs_range_unlock(rl);
1219
1220 if ((bp->b_resid = resid) == bp->b_bcount)
1221 bioerror(bp, off > volsize ? EINVAL : error);
1222
1223 if (!(bp->b_flags & B_ASYNC) && !doread && !zil_disable && !is_dump)
1224 zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ);
1225 mutex_enter(&zv->zv_dklock);
1226 disk_unbusy(&zv->zv_dk, bp->b_bcount - bp->b_resid, doread);
1227 mutex_exit(&zv->zv_dklock);
1228 biodone(bp);
1229
1230 return;
1231 }
1232
1233 /*
1234 * Set the buffer count to the zvol maximum transfer.
1235 * Using our own routine instead of the default minphys()
1236 * means that for larger writes we write bigger buffers on X86
1237 * (128K instead of 56K) and flush the disk write cache less often
1238 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1239 * 56K on X86 and 128K on sparc).
1240 */
1241 void
1242 zvol_minphys(struct buf *bp)
1243 {
1244 if (bp->b_bcount > zvol_maxphys)
1245 bp->b_bcount = zvol_maxphys;
1246 }
1247
1248 #ifndef __NetBSD__
1249 int
1250 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1251 {
1252 minor_t minor = getminor(dev);
1253 zvol_state_t *zv;
1254 int error = 0;
1255 uint64_t size;
1256 uint64_t boff;
1257 uint64_t resid;
1258
1259 if (minor == 0) /* This is the control device */
1260 return (ENXIO);
1261
1262 zv = ddi_get_soft_state(zvol_state, minor);
1263 if (zv == NULL)
1264 return (ENXIO);
1265
1266 boff = ldbtob(blkno);
1267 resid = ldbtob(nblocks);
1268
1269 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1270
1271 while (resid) {
1272 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1273 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1274 if (error)
1275 break;
1276 boff += size;
1277 addr += size;
1278 resid -= size;
1279 }
1280
1281 return (error);
1282 }
1283 #endif /* !__NetBSD__ */
1284
1285 /*ARGSUSED*/
1286 int
1287 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1288 {
1289 minor_t minor = getminor(dev);
1290 zvol_state_t *zv;
1291 uint64_t volsize;
1292 rl_t *rl;
1293 int error = 0;
1294
1295 if (minor == 0) /* This is the control device */
1296 return (ENXIO);
1297
1298 zv = ddi_get_soft_state(zvol_state, minor);
1299 if (zv == NULL)
1300 return (ENXIO);
1301
1302 volsize = zv->zv_volsize;
1303 if (uio->uio_resid > 0 &&
1304 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1305 return (EIO);
1306
1307 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1308 error = physio(zvol_strategy, NULL, dev, B_READ,
1309 zvol_minphys, uio);
1310 return (error);
1311 }
1312
1313 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1314 RL_READER);
1315 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1316 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1317
1318 /* don't read past the end */
1319 if (bytes > volsize - uio->uio_loffset)
1320 bytes = volsize - uio->uio_loffset;
1321
1322 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1323 if (error) {
1324 /* convert checksum errors into IO errors */
1325 if (error == ECKSUM)
1326 error = EIO;
1327 break;
1328 }
1329 }
1330 zfs_range_unlock(rl);
1331 return (error);
1332 }
1333
1334 /*ARGSUSED*/
1335 int
1336 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1337 {
1338 minor_t minor = getminor(dev);
1339 zvol_state_t *zv;
1340 uint64_t volsize;
1341 rl_t *rl;
1342 int error = 0;
1343
1344 if (minor == 0) /* This is the control device */
1345 return (ENXIO);
1346
1347 zv = ddi_get_soft_state(zvol_state, minor);
1348 if (zv == NULL)
1349 return (ENXIO);
1350
1351 volsize = zv->zv_volsize;
1352 if (uio->uio_resid > 0 &&
1353 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1354 return (EIO);
1355
1356 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1357 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1358 zvol_minphys, uio);
1359 return (error);
1360 }
1361
1362 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1363 RL_WRITER);
1364 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1365 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1366 uint64_t off = uio->uio_loffset;
1367 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1368
1369 if (bytes > volsize - off) /* don't write past the end */
1370 bytes = volsize - off;
1371
1372 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1373 error = dmu_tx_assign(tx, TXG_WAIT);
1374 if (error) {
1375 dmu_tx_abort(tx);
1376 break;
1377 }
1378 error = dmu_write_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes, tx);
1379 if (error == 0)
1380 zvol_log_write(zv, tx, off, bytes);
1381 dmu_tx_commit(tx);
1382
1383 if (error)
1384 break;
1385 }
1386 zfs_range_unlock(rl);
1387 return (error);
1388 }
1389
1390 #ifdef __NetBSD__
1391
1392 /*
1393 * Dirtbag ioctls to support newfs(1) for UFS filesystems.
1394 */
1395 /*ARGSUSED*/
1396 int
1397 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1398 {
1399 zvol_state_t *zv;
1400 int error = 0;
1401
1402 mutex_enter(&zvol_state_lock);
1403
1404 zv = ddi_get_soft_state(zvol_state, getminor(dev));
1405
1406 if (zv == NULL) {
1407 mutex_exit(&zvol_state_lock);
1408 return (ENXIO);
1409 }
1410
1411 switch(cmd) {
1412 case DIOCGWEDGEINFO:
1413 {
1414 struct dkwedge_info *dkw = (void *) arg;
1415
1416 strlcpy(dkw->dkw_devname, zv->zv_name, 16);
1417 strlcpy(dkw->dkw_wname, zv->zv_name, MAXPATHLEN);
1418 strlcpy(dkw->dkw_parent, zv->zv_name, 16);
1419
1420 dkw->dkw_offset = 0;
1421 /* XXX NetBSD supports only DEV_BSIZE device block
1422 size zv_volblocksize >> DEV_BSIZE*/
1423 dkw->dkw_size = (zv->zv_volsize / DEV_BSIZE);
1424 dprintf("dkw %"PRIu64" volsize %"PRIu64" volblock %"PRIu64" \n",
1425 dkw->dkw_size, zv->zv_volsize, zv->zv_volblocksize);
1426 strcpy(dkw->dkw_ptype, DKW_PTYPE_FFS);
1427
1428 break;
1429 }
1430
1431 case DIOCGDISKINFO:
1432 {
1433 struct plistref *pref = (struct plistref *) arg;
1434
1435 if (zv->zv_dk.dk_info == NULL) {
1436 mutex_exit(&zvol_state_lock);
1437 return ENOTSUP;
1438 } else
1439 prop_dictionary_copyout_ioctl(pref, cmd,
1440 zv->zv_dk.dk_info);
1441
1442 break;
1443 }
1444
1445 default:
1446 aprint_debug("unknown disk_ioctl called\n");
1447 error = ENOTTY;
1448 break;
1449 }
1450
1451 mutex_exit(&zvol_state_lock);
1452 return (error);
1453 }
1454
1455 #else /* __NetBSD__ */
1456
1457 int
1458 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1459 {
1460 struct uuid uuid = EFI_RESERVED;
1461 efi_gpe_t gpe = { 0 };
1462 uint32_t crc;
1463 dk_efi_t efi;
1464 int length;
1465 char *ptr;
1466
1467 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1468 return (EFAULT);
1469 ptr = (char *)(uintptr_t)efi.dki_data_64;
1470 length = efi.dki_length;
1471 /*
1472 * Some clients may attempt to request a PMBR for the
1473 * zvol. Currently this interface will return EINVAL to
1474 * such requests. These requests could be supported by
1475 * adding a check for lba == 0 and consing up an appropriate
1476 * PMBR.
1477 */
1478 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1479 return (EINVAL);
1480
1481 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1482 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1483 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1484
1485 if (efi.dki_lba == 1) {
1486 efi_gpt_t gpt = { 0 };
1487
1488 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1489 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1490 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1491 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1492 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1493 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1494 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1495 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1496 gpt.efi_gpt_SizeOfPartitionEntry =
1497 LE_32(sizeof (efi_gpe_t));
1498 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1499 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1500 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1501 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1502 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1503 flag))
1504 return (EFAULT);
1505 ptr += sizeof (gpt);
1506 length -= sizeof (gpt);
1507 }
1508 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1509 length), flag))
1510 return (EFAULT);
1511 return (0);
1512 }
1513
1514 /*
1515 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
1516 */
1517 /*ARGSUSED*/
1518 int
1519 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1520 {
1521 zvol_state_t *zv;
1522 struct dk_cinfo dki;
1523 struct dk_minfo dkm;
1524 struct dk_callback *dkc;
1525 int error = 0;
1526 rl_t *rl;
1527
1528 mutex_enter(&zvol_state_lock);
1529
1530 zv = ddi_get_soft_state(zvol_state, getminor(dev));
1531
1532 if (zv == NULL) {
1533 mutex_exit(&zvol_state_lock);
1534 return (ENXIO);
1535 }
1536
1537 switch (cmd) {
1538
1539 case DKIOCINFO:
1540 bzero(&dki, sizeof (dki));
1541 (void) strcpy(dki.dki_cname, "zvol");
1542 (void) strcpy(dki.dki_dname, "zvol");
1543 dki.dki_ctype = DKC_UNKNOWN;
1544 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
1545 mutex_exit(&zvol_state_lock);
1546 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1547 error = EFAULT;
1548 return (error);
1549
1550 case DKIOCGMEDIAINFO:
1551 bzero(&dkm, sizeof (dkm));
1552 dkm.dki_lbsize = 1U << zv->zv_min_bs;
1553 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1554 dkm.dki_media_type = DK_UNKNOWN;
1555 mutex_exit(&zvol_state_lock);
1556 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1557 error = EFAULT;
1558 return (error);
1559
1560 case DKIOCGETEFI:
1561 {
1562 uint64_t vs = zv->zv_volsize;
1563 uint8_t bs = zv->zv_min_bs;
1564
1565 mutex_exit(&zvol_state_lock);
1566 error = zvol_getefi((void *)arg, flag, vs, bs);
1567 return (error);
1568 }
1569
1570 case DKIOCFLUSHWRITECACHE:
1571 dkc = (struct dk_callback *)arg;
1572 zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ);
1573 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1574 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
1575 error = 0;
1576 }
1577 break;
1578
1579 case DKIOCGGEOM:
1580 case DKIOCGVTOC:
1581 /*
1582 * commands using these (like prtvtoc) expect ENOTSUP
1583 * since we're emulating an EFI label
1584 */
1585 error = ENOTSUP;
1586 break;
1587
1588 case DKIOCDUMPINIT:
1589 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1590 RL_WRITER);
1591 error = zvol_dumpify(zv);
1592 zfs_range_unlock(rl);
1593 break;
1594
1595 case DKIOCDUMPFINI:
1596 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1597 RL_WRITER);
1598 error = zvol_dump_fini(zv);
1599 zfs_range_unlock(rl);
1600 break;
1601
1602 default:
1603 error = ENOTTY;
1604 break;
1605
1606 }
1607 mutex_exit(&zvol_state_lock);
1608 return (error);
1609 }
1610
1611 #endif /* __NetBSD__ */
1612
1613 int
1614 zvol_busy(void)
1615 {
1616 return (zvol_minors != 0);
1617 }
1618
1619 void
1620 zvol_init(void)
1621 {
1622 VERIFY(ddi_soft_state_init(&zvol_state, sizeof (zvol_state_t), 1) == 0);
1623 mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL);
1624 }
1625
1626 void
1627 zvol_fini(void)
1628 {
1629 mutex_destroy(&zvol_state_lock);
1630 ddi_soft_state_fini(&zvol_state);
1631 }
1632
1633 #ifndef __NetBSD__
1634 static boolean_t
1635 zvol_is_swap(zvol_state_t *zv)
1636 {
1637 vnode_t *vp;
1638 boolean_t ret = B_FALSE;
1639 char *devpath;
1640 size_t devpathlen;
1641 int error;
1642
1643 devpathlen = strlen(ZVOL_FULL_DEV_DIR) + strlen(zv->zv_name) + 1;
1644 devpath = kmem_alloc(devpathlen, KM_SLEEP);
1645 (void) sprintf(devpath, "%s%s", ZVOL_FULL_DEV_DIR, zv->zv_name);
1646 error = lookupname(devpath, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp);
1647 kmem_free(devpath, devpathlen);
1648
1649 ret = !error && IS_SWAPVP(common_specvp(vp));
1650
1651 if (vp != NULL)
1652 VN_RELE(vp);
1653
1654 return (ret);
1655 }
1656
1657 static int
1658 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1659 {
1660 dmu_tx_t *tx;
1661 int error = 0;
1662 objset_t *os = zv->zv_objset;
1663 nvlist_t *nv = NULL;
1664
1665 ASSERT(MUTEX_HELD(&zvol_state_lock));
1666
1667 tx = dmu_tx_create(os);
1668 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1669 error = dmu_tx_assign(tx, TXG_WAIT);
1670 if (error) {
1671 dmu_tx_abort(tx);
1672 return (error);
1673 }
1674
1675 /*
1676 * If we are resizing the dump device then we only need to
1677 * update the refreservation to match the newly updated
1678 * zvolsize. Otherwise, we save off the original state of the
1679 * zvol so that we can restore them if the zvol is ever undumpified.
1680 */
1681 if (resize) {
1682 error = zap_update(os, ZVOL_ZAP_OBJ,
1683 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1684 &zv->zv_volsize, tx);
1685 } else {
1686 uint64_t checksum, compress, refresrv, vbs;
1687
1688 error = dsl_prop_get_integer(zv->zv_name,
1689 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1690 error = error ? error : dsl_prop_get_integer(zv->zv_name,
1691 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
1692 error = error ? error : dsl_prop_get_integer(zv->zv_name,
1693 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
1694 error = error ? error : dsl_prop_get_integer(zv->zv_name,
1695 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
1696
1697 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1698 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1699 &compress, tx);
1700 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1701 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
1702 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1703 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1704 &refresrv, tx);
1705 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1706 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1707 &vbs, tx);
1708 }
1709 dmu_tx_commit(tx);
1710
1711 /* Truncate the file */
1712 if (!error)
1713 error = dmu_free_long_range(zv->zv_objset,
1714 ZVOL_OBJ, 0, DMU_OBJECT_END);
1715
1716 if (error)
1717 return (error);
1718
1719 /*
1720 * We only need update the zvol's property if we are initializing
1721 * the dump area for the first time.
1722 */
1723 if (!resize) {
1724 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1725 VERIFY(nvlist_add_uint64(nv,
1726 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
1727 VERIFY(nvlist_add_uint64(nv,
1728 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
1729 ZIO_COMPRESS_OFF) == 0);
1730 VERIFY(nvlist_add_uint64(nv,
1731 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
1732 ZIO_CHECKSUM_OFF) == 0);
1733 VERIFY(nvlist_add_uint64(nv,
1734 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE),
1735 SPA_MAXBLOCKSIZE) == 0);
1736
1737 error = zfs_set_prop_nvlist(zv->zv_name, nv);
1738 nvlist_free(nv);
1739
1740 if (error)
1741 return (error);
1742 }
1743
1744 /* Allocate the space for the dump */
1745 error = zvol_prealloc(zv);
1746 return (error);
1747 }
1748
1749 static int
1750 zvol_dumpify(zvol_state_t *zv)
1751 {
1752 int error = 0;
1753 uint64_t dumpsize = 0;
1754 dmu_tx_t *tx;
1755 objset_t *os = zv->zv_objset;
1756
1757 if (zv->zv_flags & ZVOL_RDONLY || (zv->zv_mode & DS_MODE_READONLY))
1758 return (EROFS);
1759
1760 /*
1761 * We do not support swap devices acting as dump devices.
1762 */
1763 if (zvol_is_swap(zv))
1764 return (ENOTSUP);
1765
1766 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
1767 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
1768 boolean_t resize = (dumpsize > 0) ? B_TRUE : B_FALSE;
1769
1770 if ((error = zvol_dump_init(zv, resize)) != 0) {
1771 (void) zvol_dump_fini(zv);
1772 return (error);
1773 }
1774 }
1775
1776 /*
1777 * Build up our lba mapping.
1778 */
1779 error = zvol_get_lbas(zv);
1780 if (error) {
1781 (void) zvol_dump_fini(zv);
1782 return (error);
1783 }
1784
1785 tx = dmu_tx_create(os);
1786 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1787 error = dmu_tx_assign(tx, TXG_WAIT);
1788 if (error) {
1789 dmu_tx_abort(tx);
1790 (void) zvol_dump_fini(zv);
1791 return (error);
1792 }
1793
1794 zv->zv_flags |= ZVOL_DUMPIFIED;
1795 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
1796 &zv->zv_volsize, tx);
1797 dmu_tx_commit(tx);
1798
1799 if (error) {
1800 (void) zvol_dump_fini(zv);
1801 return (error);
1802 }
1803
1804 txg_wait_synced(dmu_objset_pool(os), 0);
1805 return (0);
1806 }
1807
1808 static int
1809 zvol_dump_fini(zvol_state_t *zv)
1810 {
1811 dmu_tx_t *tx;
1812 objset_t *os = zv->zv_objset;
1813 nvlist_t *nv;
1814 int error = 0;
1815 uint64_t checksum, compress, refresrv, vbs;
1816
1817 /*
1818 * Attempt to restore the zvol back to its pre-dumpified state.
1819 * This is a best-effort attempt as it's possible that not all
1820 * of these properties were initialized during the dumpify process
1821 * (i.e. error during zvol_dump_init).
1822 */
1823
1824 tx = dmu_tx_create(os);
1825 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1826 error = dmu_tx_assign(tx, TXG_WAIT);
1827 if (error) {
1828 dmu_tx_abort(tx);
1829 return (error);
1830 }
1831 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
1832 dmu_tx_commit(tx);
1833
1834 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1835 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
1836 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1837 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
1838 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1839 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
1840 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1841 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
1842
1843 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1844 (void) nvlist_add_uint64(nv,
1845 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
1846 (void) nvlist_add_uint64(nv,
1847 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
1848 (void) nvlist_add_uint64(nv,
1849 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
1850 (void) nvlist_add_uint64(nv,
1851 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), vbs);
1852 (void) zfs_set_prop_nvlist(zv->zv_name, nv);
1853 nvlist_free(nv);
1854
1855 zvol_free_extents(zv);
1856 zv->zv_flags &= ~ZVOL_DUMPIFIED;
1857 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
1858
1859 return (0);
1860 }
1861 #endif /* !__NetBSD__ */
NetBSD on the FriendlyARM MINI2440