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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 *
24 * Copyright (c) 2006-2010 Pawel Jakub Dawidek <pjd@FreeBSD.org>
25 * All rights reserved.
26 *
27 * Portions Copyright 2010 Robert Milkowski
28 *
29 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
30 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
31 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
32 */
33
34/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */
35
36/*
37 * ZFS volume emulation driver.
38 *
39 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
40 * Volumes are accessed through the symbolic links named:
41 *
42 * /dev/zvol/dsk/<pool_name>/<dataset_name>
43 * /dev/zvol/rdsk/<pool_name>/<dataset_name>
44 *
45 * These links are created by the /dev filesystem (sdev_zvolops.c).
46 * Volumes are persistent through reboot. No user command needs to be
47 * run before opening and using a device.
48 *
49 * FreeBSD notes.
50 * On FreeBSD ZVOLs are simply GEOM providers like any other storage device
51 * in the system.
52 */
53
54#include <sys/types.h>
55#include <sys/param.h>
56#include <sys/kernel.h>
57#include <sys/errno.h>
58#include <sys/uio.h>
59#include <sys/bio.h>
60#include <sys/buf.h>
61#include <sys/kmem.h>
62#include <sys/conf.h>
63#include <sys/cmn_err.h>
64#include <sys/stat.h>
65#include <sys/zap.h>
66#include <sys/spa.h>
67#include <sys/spa_impl.h>
68#include <sys/zio.h>
69#include <sys/disk.h>
70#include <sys/dmu_traverse.h>
71#include <sys/dnode.h>
72#include <sys/dsl_dataset.h>
73#include <sys/dsl_prop.h>
74#include <sys/dkio.h>
75#include <sys/byteorder.h>
76#include <sys/sunddi.h>
77#include <sys/dirent.h>
78#include <sys/policy.h>
79#include <sys/queue.h>
80#include <sys/fs/zfs.h>
81#include <sys/zfs_ioctl.h>
82#include <sys/zil.h>
83#include <sys/refcount.h>
84#include <sys/zfs_znode.h>
85#include <sys/zfs_rlock.h>
86#include <sys/vdev_impl.h>
87#include <sys/vdev_raidz.h>
88#include <sys/zvol.h>
89#include <sys/zil_impl.h>
90#include <sys/dbuf.h>
91#include <sys/dmu_tx.h>
92#include <sys/zfeature.h>
93#include <sys/zio_checksum.h>
94
95#include <geom/geom.h>
96
97#include "zfs_namecheck.h"
98
99struct g_class zfs_zvol_class = {
100 .name = "ZFS::ZVOL",
101 .version = G_VERSION,
102};
103
104DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol);
105
106void *zfsdev_state;
107static char *zvol_tag = "zvol_tag";
108
109#define ZVOL_DUMPSIZE "dumpsize"
110
111/*
112 * The spa_namespace_lock protects the zfsdev_state structure from being
113 * modified while it's being used, e.g. an open that comes in before a
114 * create finishes. It also protects temporary opens of the dataset so that,
115 * e.g., an open doesn't get a spurious EBUSY.
116 */
117static uint32_t zvol_minors;
118
119SYSCTL_DECL(_vfs_zfs);
120SYSCTL_NODE(_vfs_zfs, OID_AUTO, vol, CTLFLAG_RW, 0, "ZFS VOLUME");
121static int volmode = ZFS_VOLMODE_GEOM;
122TUNABLE_INT("vfs.zfs.vol.mode", &volmode);
123SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &volmode, 0,
124 "Expose as GEOM providers (1), device files (2) or neither");
125
126typedef struct zvol_extent {
127 list_node_t ze_node;
128 dva_t ze_dva; /* dva associated with this extent */
129 uint64_t ze_nblks; /* number of blocks in extent */
130} zvol_extent_t;
131
132/*
133 * The in-core state of each volume.
134 */
135typedef struct zvol_state {
136 LIST_ENTRY(zvol_state) zv_links;
137 char zv_name[MAXPATHLEN]; /* pool/dd name */
138 uint64_t zv_volsize; /* amount of space we advertise */
139 uint64_t zv_volblocksize; /* volume block size */
140 struct cdev *zv_dev; /* non-GEOM device */
141 struct g_provider *zv_provider; /* GEOM provider */
142 uint8_t zv_min_bs; /* minimum addressable block shift */
143 uint8_t zv_flags; /* readonly, dumpified, etc. */
144 objset_t *zv_objset; /* objset handle */
145 uint32_t zv_total_opens; /* total open count */
146 zilog_t *zv_zilog; /* ZIL handle */
147 list_t zv_extents; /* List of extents for dump */
148 znode_t zv_znode; /* for range locking */
149 dmu_buf_t *zv_dbuf; /* bonus handle */
150 int zv_state;
151 int zv_volmode; /* Provide GEOM or cdev */
152 struct bio_queue_head zv_queue;
153 struct mtx zv_queue_mtx; /* zv_queue mutex */
154} zvol_state_t;
155
156static LIST_HEAD(, zvol_state) all_zvols;
157
158/*
159 * zvol specific flags
160 */
161#define ZVOL_RDONLY 0x1
162#define ZVOL_DUMPIFIED 0x2
163#define ZVOL_EXCL 0x4
164#define ZVOL_WCE 0x8
165
166/*
167 * zvol maximum transfer in one DMU tx.
168 */
169int zvol_maxphys = DMU_MAX_ACCESS/2;
170
171static d_open_t zvol_d_open;
172static d_close_t zvol_d_close;
173static d_read_t zvol_read;
174static d_write_t zvol_write;
175static d_ioctl_t zvol_d_ioctl;
176static d_strategy_t zvol_strategy;
177
178static struct cdevsw zvol_cdevsw = {
179 .d_version = D_VERSION,
180 .d_open = zvol_d_open,
181 .d_close = zvol_d_close,
182 .d_read = zvol_read,
183 .d_write = zvol_write,
184 .d_ioctl = zvol_d_ioctl,
185 .d_strategy = zvol_strategy,
186 .d_name = "zvol",
187 .d_flags = D_DISK | D_TRACKCLOSE,
188};
189
190extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
191 nvlist_t *, nvlist_t *);
192static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off,
193 uint64_t len, boolean_t sync);
194static int zvol_remove_zv(zvol_state_t *);
195static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
196static int zvol_dumpify(zvol_state_t *zv);
197static int zvol_dump_fini(zvol_state_t *zv);
198static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
199
200static void zvol_geom_run(zvol_state_t *zv);
201static void zvol_geom_destroy(zvol_state_t *zv);
202static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
203static void zvol_geom_start(struct bio *bp);
204static void zvol_geom_worker(void *arg);
205
206static void
207zvol_size_changed(zvol_state_t *zv)
208{
209#ifdef sun
210 dev_t dev = makedevice(maj, min);
211
212 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
213 "Size", volsize) == DDI_SUCCESS);
214 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
215 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
216
217 /* Notify specfs to invalidate the cached size */
218 spec_size_invalidate(dev, VBLK);
219 spec_size_invalidate(dev, VCHR);
220#else /* !sun */
221 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
222 struct g_provider *pp;
223
224 pp = zv->zv_provider;
225 if (pp == NULL)
226 return;
227 g_topology_lock();
228 g_resize_provider(pp, zv->zv_volsize);
229 g_topology_unlock();
230 }
231#endif /* !sun */
232}
233
234int
235zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
236{
237 if (volsize == 0)
238 return (SET_ERROR(EINVAL));
239
240 if (volsize % blocksize != 0)
241 return (SET_ERROR(EINVAL));
242
243#ifdef _ILP32
244 if (volsize - 1 > SPEC_MAXOFFSET_T)
245 return (SET_ERROR(EOVERFLOW));
246#endif
247 return (0);
248}
249
250int
251zvol_check_volblocksize(uint64_t volblocksize)
252{
253 if (volblocksize < SPA_MINBLOCKSIZE ||
254 volblocksize > SPA_MAXBLOCKSIZE ||
255 !ISP2(volblocksize))
256 return (SET_ERROR(EDOM));
257
258 return (0);
259}
260
261int
262zvol_get_stats(objset_t *os, nvlist_t *nv)
263{
264 int error;
265 dmu_object_info_t doi;
266 uint64_t val;
267
268 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
269 if (error)
270 return (error);
271
272 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
273
274 error = dmu_object_info(os, ZVOL_OBJ, &doi);
275
276 if (error == 0) {
277 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
278 doi.doi_data_block_size);
279 }
280
281 return (error);
282}
283
284static zvol_state_t *
285zvol_minor_lookup(const char *name)
286{
287 zvol_state_t *zv;
288
289 ASSERT(MUTEX_HELD(&spa_namespace_lock));
290
291 LIST_FOREACH(zv, &all_zvols, zv_links) {
292 if (strcmp(zv->zv_name, name) == 0)
293 break;
294 }
295
296 return (zv);
297}
298
299/* extent mapping arg */
300struct maparg {
301 zvol_state_t *ma_zv;
302 uint64_t ma_blks;
303};
304
305/*ARGSUSED*/
306static int
307zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
308 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
309{
310 struct maparg *ma = arg;
311 zvol_extent_t *ze;
312 int bs = ma->ma_zv->zv_volblocksize;
313
314 if (BP_IS_HOLE(bp) ||
315 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
316 return (0);
317
318 VERIFY(!BP_IS_EMBEDDED(bp));
319
320 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
321 ma->ma_blks++;
322
323 /* Abort immediately if we have encountered gang blocks */
324 if (BP_IS_GANG(bp))
325 return (SET_ERROR(EFRAGS));
326
327 /*
328 * See if the block is at the end of the previous extent.
329 */
330 ze = list_tail(&ma->ma_zv->zv_extents);
331 if (ze &&
332 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
333 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
334 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
335 ze->ze_nblks++;
336 return (0);
337 }
338
339 dprintf_bp(bp, "%s", "next blkptr:");
340
341 /* start a new extent */
342 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
343 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
344 ze->ze_nblks = 1;
345 list_insert_tail(&ma->ma_zv->zv_extents, ze);
346 return (0);
347}
348
349static void
350zvol_free_extents(zvol_state_t *zv)
351{
352 zvol_extent_t *ze;
353
354 while (ze = list_head(&zv->zv_extents)) {
355 list_remove(&zv->zv_extents, ze);
356 kmem_free(ze, sizeof (zvol_extent_t));
357 }
358}
359
360static int
361zvol_get_lbas(zvol_state_t *zv)
362{
363 objset_t *os = zv->zv_objset;
364 struct maparg ma;
365 int err;
366
367 ma.ma_zv = zv;
368 ma.ma_blks = 0;
369 zvol_free_extents(zv);
370
371 /* commit any in-flight changes before traversing the dataset */
372 txg_wait_synced(dmu_objset_pool(os), 0);
373 err = traverse_dataset(dmu_objset_ds(os), 0,
374 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
375 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
376 zvol_free_extents(zv);
377 return (err ? err : EIO);
378 }
379
380 return (0);
381}
382
383/* ARGSUSED */
384void
385zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
386{
387 zfs_creat_t *zct = arg;
388 nvlist_t *nvprops = zct->zct_props;
389 int error;
390 uint64_t volblocksize, volsize;
391
392 VERIFY(nvlist_lookup_uint64(nvprops,
393 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
394 if (nvlist_lookup_uint64(nvprops,
395 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
396 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
397
398 /*
399 * These properties must be removed from the list so the generic
400 * property setting step won't apply to them.
401 */
402 VERIFY(nvlist_remove_all(nvprops,
403 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
404 (void) nvlist_remove_all(nvprops,
405 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
406
407 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
408 DMU_OT_NONE, 0, tx);
409 ASSERT(error == 0);
410
411 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
412 DMU_OT_NONE, 0, tx);
413 ASSERT(error == 0);
414
415 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
416 ASSERT(error == 0);
417}
418
419/*
420 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
421 * implement DKIOCFREE/free-long-range.
422 */
423static int
424zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
425{
426 uint64_t offset, length;
427
428 if (byteswap)
429 byteswap_uint64_array(lr, sizeof (*lr));
430
431 offset = lr->lr_offset;
432 length = lr->lr_length;
433
434 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
435}
436
437/*
438 * Replay a TX_WRITE ZIL transaction that didn't get committed
439 * after a system failure
440 */
441static int
442zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
443{
444 objset_t *os = zv->zv_objset;
445 char *data = (char *)(lr + 1); /* data follows lr_write_t */
446 uint64_t offset, length;
447 dmu_tx_t *tx;
448 int error;
449
450 if (byteswap)
451 byteswap_uint64_array(lr, sizeof (*lr));
452
453 offset = lr->lr_offset;
454 length = lr->lr_length;
455
456 /* If it's a dmu_sync() block, write the whole block */
457 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
458 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
459 if (length < blocksize) {
460 offset -= offset % blocksize;
461 length = blocksize;
462 }
463 }
464
465 tx = dmu_tx_create(os);
466 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
467 error = dmu_tx_assign(tx, TXG_WAIT);
468 if (error) {
469 dmu_tx_abort(tx);
470 } else {
471 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
472 dmu_tx_commit(tx);
473 }
474
475 return (error);
476}
477
478/* ARGSUSED */
479static int
480zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
481{
482 return (SET_ERROR(ENOTSUP));
483}
484
485/*
486 * Callback vectors for replaying records.
487 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
488 */
489zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
490 zvol_replay_err, /* 0 no such transaction type */
491 zvol_replay_err, /* TX_CREATE */
492 zvol_replay_err, /* TX_MKDIR */
493 zvol_replay_err, /* TX_MKXATTR */
494 zvol_replay_err, /* TX_SYMLINK */
495 zvol_replay_err, /* TX_REMOVE */
496 zvol_replay_err, /* TX_RMDIR */
497 zvol_replay_err, /* TX_LINK */
498 zvol_replay_err, /* TX_RENAME */
499 zvol_replay_write, /* TX_WRITE */
500 zvol_replay_truncate, /* TX_TRUNCATE */
501 zvol_replay_err, /* TX_SETATTR */
502 zvol_replay_err, /* TX_ACL */
503 zvol_replay_err, /* TX_CREATE_ACL */
504 zvol_replay_err, /* TX_CREATE_ATTR */
505 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
506 zvol_replay_err, /* TX_MKDIR_ACL */
507 zvol_replay_err, /* TX_MKDIR_ATTR */
508 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
509 zvol_replay_err, /* TX_WRITE2 */
510};
511
512#ifdef sun
513int
514zvol_name2minor(const char *name, minor_t *minor)
515{
516 zvol_state_t *zv;
517
518 mutex_enter(&spa_namespace_lock);
519 zv = zvol_minor_lookup(name);
520 if (minor && zv)
521 *minor = zv->zv_minor;
522 mutex_exit(&spa_namespace_lock);
523 return (zv ? 0 : -1);
524}
525#endif /* sun */
526
527/*
528 * Create a minor node (plus a whole lot more) for the specified volume.
529 */
530int
531zvol_create_minor(const char *name)
532{
533 zfs_soft_state_t *zs;
534 zvol_state_t *zv;
535 objset_t *os;
536 struct cdev *dev;
537 struct g_provider *pp;
538 struct g_geom *gp;
539 dmu_object_info_t doi;
540 uint64_t volsize, mode;
541 int error;
542
543 ZFS_LOG(1, "Creating ZVOL %s...", name);
544
545 mutex_enter(&spa_namespace_lock);
546
547 if (zvol_minor_lookup(name) != NULL) {
548 mutex_exit(&spa_namespace_lock);
549 return (SET_ERROR(EEXIST));
550 }
551
552 /* lie and say we're read-only */
553 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
554
555 if (error) {
556 mutex_exit(&spa_namespace_lock);
557 return (error);
558 }
559
560#ifdef sun
561 if ((minor = zfsdev_minor_alloc()) == 0) {
562 dmu_objset_disown(os, FTAG);
563 mutex_exit(&spa_namespace_lock);
564 return (SET_ERROR(ENXIO));
565 }
566
567 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
568 dmu_objset_disown(os, FTAG);
569 mutex_exit(&spa_namespace_lock);
570 return (SET_ERROR(EAGAIN));
571 }
572 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
573 (char *)name);
574
575 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
576
577 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
578 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
579 ddi_soft_state_free(zfsdev_state, minor);
580 dmu_objset_disown(os, FTAG);
581 mutex_exit(&spa_namespace_lock);
582 return (SET_ERROR(EAGAIN));
583 }
584
585 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
586
587 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
588 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
589 ddi_remove_minor_node(zfs_dip, chrbuf);
590 ddi_soft_state_free(zfsdev_state, minor);
591 dmu_objset_disown(os, FTAG);
592 mutex_exit(&spa_namespace_lock);
593 return (SET_ERROR(EAGAIN));
594 }
595
596 zs = ddi_get_soft_state(zfsdev_state, minor);
597 zs->zss_type = ZSST_ZVOL;
598 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
599#else /* !sun */
600
601 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
602 zv->zv_state = 0;
603 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
604 if (error) {
605 kmem_free(zv, sizeof(*zv));
606 dmu_objset_disown(os, zvol_tag);
607 mutex_exit(&spa_namespace_lock);
608 return (error);
609 }
610 error = dsl_prop_get_integer(name,
611 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
612 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
613 mode = volmode;
614
615 DROP_GIANT();
616 zv->zv_volsize = volsize;
617 zv->zv_volmode = mode;
618 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
619 g_topology_lock();
620 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
621 gp->start = zvol_geom_start;
622 gp->access = zvol_geom_access;
623 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
624 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
625 pp->sectorsize = DEV_BSIZE;
626 pp->mediasize = zv->zv_volsize;
627 pp->private = zv;
628
629 zv->zv_provider = pp;
630 bioq_init(&zv->zv_queue);
631 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
632 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
633 if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
634 &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR,
635 0640, "%s/%s", ZVOL_DRIVER, name) != 0) {
636 kmem_free(zv, sizeof(*zv));
637 dmu_objset_disown(os, FTAG);
638 mutex_exit(&spa_namespace_lock);
639 return (SET_ERROR(ENXIO));
640 }
641 zv->zv_dev = dev;
642 dev->si_iosize_max = MAXPHYS;
643 dev->si_drv2 = zv;
644 }
645 LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
646#endif /* !sun */
647
648 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
649 zv->zv_min_bs = DEV_BSHIFT;
650 zv->zv_objset = os;
651 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
652 zv->zv_flags |= ZVOL_RDONLY;
653 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
654 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
655 sizeof (rl_t), offsetof(rl_t, r_node));
656 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
657 offsetof(zvol_extent_t, ze_node));
658 /* get and cache the blocksize */
659 error = dmu_object_info(os, ZVOL_OBJ, &doi);
660 ASSERT(error == 0);
661 zv->zv_volblocksize = doi.doi_data_block_size;
662
663 if (spa_writeable(dmu_objset_spa(os))) {
664 if (zil_replay_disable)
665 zil_destroy(dmu_objset_zil(os), B_FALSE);
666 else
667 zil_replay(os, zv, zvol_replay_vector);
668 }
669 dmu_objset_disown(os, FTAG);
670 zv->zv_objset = NULL;
671
672 zvol_minors++;
673
674 mutex_exit(&spa_namespace_lock);
675
676#ifndef sun
677 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
678 zvol_geom_run(zv);
679 g_topology_unlock();
680 }
681 PICKUP_GIANT();
682#endif
683
684 ZFS_LOG(1, "ZVOL %s created.", name);
685
686 return (0);
687}
688
689/*
690 * Remove minor node for the specified volume.
691 */
692static int
693zvol_remove_zv(zvol_state_t *zv)
694{
695#ifdef sun
696 minor_t minor = zv->zv_minor;
697#endif
698
699 ASSERT(MUTEX_HELD(&spa_namespace_lock));
700 if (zv->zv_total_opens != 0)
701 return (SET_ERROR(EBUSY));
702
703 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
704
705#ifdef sun
706 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
707 ddi_remove_minor_node(zfs_dip, nmbuf);
708#else
709 LIST_REMOVE(zv, zv_links);
710 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
711 g_topology_lock();
712 zvol_geom_destroy(zv);
713 g_topology_unlock();
714 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV)
715 destroy_dev(zv->zv_dev);
716#endif /* sun */
717
718 avl_destroy(&zv->zv_znode.z_range_avl);
719 mutex_destroy(&zv->zv_znode.z_range_lock);
720
721 kmem_free(zv, sizeof(*zv));
722
723 zvol_minors--;
724 return (0);
725}
726
727int
728zvol_remove_minor(const char *name)
729{
730 zvol_state_t *zv;
731 int rc;
732
733 mutex_enter(&spa_namespace_lock);
734 if ((zv = zvol_minor_lookup(name)) == NULL) {
735 mutex_exit(&spa_namespace_lock);
736 return (SET_ERROR(ENXIO));
737 }
738 rc = zvol_remove_zv(zv);
739 mutex_exit(&spa_namespace_lock);
740 return (rc);
741}
742
743int
744zvol_first_open(zvol_state_t *zv)
745{
746 objset_t *os;
747 uint64_t volsize;
748 int error;
749 uint64_t readonly;
750
751 /* lie and say we're read-only */
752 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
753 zvol_tag, &os);
754 if (error)
755 return (error);
756
757 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
758 if (error) {
759 ASSERT(error == 0);
760 dmu_objset_disown(os, zvol_tag);
761 return (error);
762 }
763 zv->zv_objset = os;
764 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
765 if (error) {
766 dmu_objset_disown(os, zvol_tag);
767 return (error);
768 }
769 zv->zv_volsize = volsize;
770 zv->zv_zilog = zil_open(os, zvol_get_data);
771 zvol_size_changed(zv);
772
773 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
774 NULL) == 0);
775 if (readonly || dmu_objset_is_snapshot(os) ||
776 !spa_writeable(dmu_objset_spa(os)))
777 zv->zv_flags |= ZVOL_RDONLY;
778 else
779 zv->zv_flags &= ~ZVOL_RDONLY;
780 return (error);
781}
782
783void
784zvol_last_close(zvol_state_t *zv)
785{
786 zil_close(zv->zv_zilog);
787 zv->zv_zilog = NULL;
788
789 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
790 zv->zv_dbuf = NULL;
791
792 /*
793 * Evict cached data
794 */
795 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
796 !(zv->zv_flags & ZVOL_RDONLY))
797 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
798 dmu_objset_evict_dbufs(zv->zv_objset);
799
800 dmu_objset_disown(zv->zv_objset, zvol_tag);
801 zv->zv_objset = NULL;
802}
803
804#ifdef sun
805int
806zvol_prealloc(zvol_state_t *zv)
807{
808 objset_t *os = zv->zv_objset;
809 dmu_tx_t *tx;
810 uint64_t refd, avail, usedobjs, availobjs;
811 uint64_t resid = zv->zv_volsize;
812 uint64_t off = 0;
813
814 /* Check the space usage before attempting to allocate the space */
815 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
816 if (avail < zv->zv_volsize)
817 return (SET_ERROR(ENOSPC));
818
819 /* Free old extents if they exist */
820 zvol_free_extents(zv);
821
822 while (resid != 0) {
823 int error;
824 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE);
825
826 tx = dmu_tx_create(os);
827 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
828 error = dmu_tx_assign(tx, TXG_WAIT);
829 if (error) {
830 dmu_tx_abort(tx);
831 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
832 return (error);
833 }
834 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
835 dmu_tx_commit(tx);
836 off += bytes;
837 resid -= bytes;
838 }
839 txg_wait_synced(dmu_objset_pool(os), 0);
840
841 return (0);
842}
843#endif /* sun */
844
845static int
846zvol_update_volsize(objset_t *os, uint64_t volsize)
847{
848 dmu_tx_t *tx;
849 int error;
850
851 ASSERT(MUTEX_HELD(&spa_namespace_lock));
852
853 tx = dmu_tx_create(os);
854 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
855 dmu_tx_mark_netfree(tx);
856 error = dmu_tx_assign(tx, TXG_WAIT);
857 if (error) {
858 dmu_tx_abort(tx);
859 return (error);
860 }
861
862 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
863 &volsize, tx);
864 dmu_tx_commit(tx);
865
866 if (error == 0)
867 error = dmu_free_long_range(os,
868 ZVOL_OBJ, volsize, DMU_OBJECT_END);
869 return (error);
870}
871
872void
873zvol_remove_minors(const char *name)
874{
875 zvol_state_t *zv, *tzv;
876 size_t namelen;
877
878 namelen = strlen(name);
879
880 DROP_GIANT();
881 mutex_enter(&spa_namespace_lock);
882
883 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) {
884 if (strcmp(zv->zv_name, name) == 0 ||
885 (strncmp(zv->zv_name, name, namelen) == 0 &&
886 zv->zv_name[namelen] == '/')) {
887 (void) zvol_remove_zv(zv);
888 }
889 }
890
891 mutex_exit(&spa_namespace_lock);
892 PICKUP_GIANT();
893}
894
895int
896zvol_set_volsize(const char *name, major_t maj, uint64_t volsize)
897{
898 zvol_state_t *zv = NULL;
899 objset_t *os;
900 int error;
901 dmu_object_info_t doi;
902 uint64_t old_volsize = 0ULL;
903 uint64_t readonly;
904
905 mutex_enter(&spa_namespace_lock);
906 zv = zvol_minor_lookup(name);
907 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
908 mutex_exit(&spa_namespace_lock);
909 return (error);
910 }
911
912 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
913 (error = zvol_check_volsize(volsize,
914 doi.doi_data_block_size)) != 0)
915 goto out;
916
917 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly,
918 NULL) == 0);
919 if (readonly) {
920 error = EROFS;
921 goto out;
922 }
923
924 error = zvol_update_volsize(os, volsize);
925 /*
926 * Reinitialize the dump area to the new size. If we
927 * failed to resize the dump area then restore it back to
928 * its original size.
929 */
930 if (zv && error == 0) {
931#ifdef ZVOL_DUMP
932 if (zv->zv_flags & ZVOL_DUMPIFIED) {
933 old_volsize = zv->zv_volsize;
934 zv->zv_volsize = volsize;
935 if ((error = zvol_dumpify(zv)) != 0 ||
936 (error = dumpvp_resize()) != 0) {
937 (void) zvol_update_volsize(os, old_volsize);
938 zv->zv_volsize = old_volsize;
939 error = zvol_dumpify(zv);
940 }
941 }
942#endif /* ZVOL_DUMP */
943 if (error == 0) {
944 zv->zv_volsize = volsize;
945 zvol_size_changed(zv);
946 }
947 }
948
949#ifdef sun
950 /*
951 * Generate a LUN expansion event.
952 */
953 if (zv && error == 0) {
954 sysevent_id_t eid;
955 nvlist_t *attr;
956 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
957
958 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
959 zv->zv_minor);
960
961 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
962 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
963
964 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
965 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
966
967 nvlist_free(attr);
968 kmem_free(physpath, MAXPATHLEN);
969 }
970#endif /* sun */
971
972out:
973 dmu_objset_rele(os, FTAG);
974
975 mutex_exit(&spa_namespace_lock);
976
977 return (error);
978}
979
980/*ARGSUSED*/
981static int
982zvol_open(struct g_provider *pp, int flag, int count)
983{
984 zvol_state_t *zv;
985 int err = 0;
986 boolean_t locked = B_FALSE;
987
988 /*
989 * Protect against recursively entering spa_namespace_lock
990 * when spa_open() is used for a pool on a (local) ZVOL(s).
991 * This is needed since we replaced upstream zfsdev_state_lock
992 * with spa_namespace_lock in the ZVOL code.
993 * We are using the same trick as spa_open().
994 * Note that calls in zvol_first_open which need to resolve
995 * pool name to a spa object will enter spa_open()
996 * recursively, but that function already has all the
997 * necessary protection.
998 */
999 if (!MUTEX_HELD(&spa_namespace_lock)) {
1000 mutex_enter(&spa_namespace_lock);
1001 locked = B_TRUE;
1002 }
1003
1004 zv = pp->private;
1005 if (zv == NULL) {
1006 if (locked)
1007 mutex_exit(&spa_namespace_lock);
1008 return (SET_ERROR(ENXIO));
1009 }
1010
1011 if (zv->zv_total_opens == 0) {
1012 err = zvol_first_open(zv);
1013 if (err) {
1014 if (locked)
1015 mutex_exit(&spa_namespace_lock);
1016 return (err);
1017 }
1018 pp->mediasize = zv->zv_volsize;
1019 pp->stripeoffset = 0;
1020 pp->stripesize = zv->zv_volblocksize;
1021 }
1022 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1023 err = SET_ERROR(EROFS);
1024 goto out;
1025 }
1026 if (zv->zv_flags & ZVOL_EXCL) {
1027 err = SET_ERROR(EBUSY);
1028 goto out;
1029 }
1030#ifdef FEXCL
1031 if (flag & FEXCL) {
1032 if (zv->zv_total_opens != 0) {
1033 err = SET_ERROR(EBUSY);
1034 goto out;
1035 }
1036 zv->zv_flags |= ZVOL_EXCL;
1037 }
1038#endif
1039
1040 zv->zv_total_opens += count;
1041 if (locked)
1042 mutex_exit(&spa_namespace_lock);
1043
1044 return (err);
1045out:
1046 if (zv->zv_total_opens == 0)
1047 zvol_last_close(zv);
1048 if (locked)
1049 mutex_exit(&spa_namespace_lock);
1050 return (err);
1051}
1052
1053/*ARGSUSED*/
1054static int
1055zvol_close(struct g_provider *pp, int flag, int count)
1056{
1057 zvol_state_t *zv;
1058 int error = 0;
1059 boolean_t locked = B_FALSE;
1060
1061 /* See comment in zvol_open(). */
1062 if (!MUTEX_HELD(&spa_namespace_lock)) {
1063 mutex_enter(&spa_namespace_lock);
1064 locked = B_TRUE;
1065 }
1066
1067 zv = pp->private;
1068 if (zv == NULL) {
1069 if (locked)
1070 mutex_exit(&spa_namespace_lock);
1071 return (SET_ERROR(ENXIO));
1072 }
1073
1074 if (zv->zv_flags & ZVOL_EXCL) {
1075 ASSERT(zv->zv_total_opens == 1);
1076 zv->zv_flags &= ~ZVOL_EXCL;
1077 }
1078
1079 /*
1080 * If the open count is zero, this is a spurious close.
1081 * That indicates a bug in the kernel / DDI framework.
1082 */
1083 ASSERT(zv->zv_total_opens != 0);
1084
1085 /*
1086 * You may get multiple opens, but only one close.
1087 */
1088 zv->zv_total_opens -= count;
1089
1090 if (zv->zv_total_opens == 0)
1091 zvol_last_close(zv);
1092
1093 if (locked)
1094 mutex_exit(&spa_namespace_lock);
1095 return (error);
1096}
1097
1098static void
1099zvol_get_done(zgd_t *zgd, int error)
1100{
1101 if (zgd->zgd_db)
1102 dmu_buf_rele(zgd->zgd_db, zgd);
1103
1104 zfs_range_unlock(zgd->zgd_rl);
1105
1106 if (error == 0 && zgd->zgd_bp)
1107 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1108
1109 kmem_free(zgd, sizeof (zgd_t));
1110}
1111
1112/*
1113 * Get data to generate a TX_WRITE intent log record.
1114 */
1115static int
1116zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1117{
1118 zvol_state_t *zv = arg;
1119 objset_t *os = zv->zv_objset;
1120 uint64_t object = ZVOL_OBJ;
1121 uint64_t offset = lr->lr_offset;
1122 uint64_t size = lr->lr_length; /* length of user data */
1123 blkptr_t *bp = &lr->lr_blkptr;
1124 dmu_buf_t *db;
1125 zgd_t *zgd;
1126 int error;
1127
1128 ASSERT(zio != NULL);
1129 ASSERT(size != 0);
1130
1131 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1132 zgd->zgd_zilog = zv->zv_zilog;
1133 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
1134
1135 /*
1136 * Write records come in two flavors: immediate and indirect.
1137 * For small writes it's cheaper to store the data with the
1138 * log record (immediate); for large writes it's cheaper to
1139 * sync the data and get a pointer to it (indirect) so that
1140 * we don't have to write the data twice.
1141 */
1142 if (buf != NULL) { /* immediate write */
1143 error = dmu_read(os, object, offset, size, buf,
1144 DMU_READ_NO_PREFETCH);
1145 } else {
1146 size = zv->zv_volblocksize;
1147 offset = P2ALIGN(offset, size);
1148 error = dmu_buf_hold(os, object, offset, zgd, &db,
1149 DMU_READ_NO_PREFETCH);
1150 if (error == 0) {
1151 blkptr_t *obp = dmu_buf_get_blkptr(db);
1152 if (obp) {
1153 ASSERT(BP_IS_HOLE(bp));
1154 *bp = *obp;
1155 }
1156
1157 zgd->zgd_db = db;
1158 zgd->zgd_bp = bp;
1159
1160 ASSERT(db->db_offset == offset);
1161 ASSERT(db->db_size == size);
1162
1163 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1164 zvol_get_done, zgd);
1165
1166 if (error == 0)
1167 return (0);
1168 }
1169 }
1170
1171 zvol_get_done(zgd, error);
1172
1173 return (error);
1174}
1175
1176/*
1177 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1178 *
1179 * We store data in the log buffers if it's small enough.
1180 * Otherwise we will later flush the data out via dmu_sync().
1181 */
1182ssize_t zvol_immediate_write_sz = 32768;
1183
1184static void
1185zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1186 boolean_t sync)
1187{
1188 uint32_t blocksize = zv->zv_volblocksize;
1189 zilog_t *zilog = zv->zv_zilog;
1190 boolean_t slogging;
1191 ssize_t immediate_write_sz;
1192
1193 if (zil_replaying(zilog, tx))
1194 return;
1195
1196 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1197 ? 0 : zvol_immediate_write_sz;
1198
1199 slogging = spa_has_slogs(zilog->zl_spa) &&
1200 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1201
1202 while (resid) {
1203 itx_t *itx;
1204 lr_write_t *lr;
1205 ssize_t len;
1206 itx_wr_state_t write_state;
1207
1208 /*
1209 * Unlike zfs_log_write() we can be called with
1210 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1211 */
1212 if (blocksize > immediate_write_sz && !slogging &&
1213 resid >= blocksize && off % blocksize == 0) {
1214 write_state = WR_INDIRECT; /* uses dmu_sync */
1215 len = blocksize;
1216 } else if (sync) {
1217 write_state = WR_COPIED;
1218 len = MIN(ZIL_MAX_LOG_DATA, resid);
1219 } else {
1220 write_state = WR_NEED_COPY;
1221 len = MIN(ZIL_MAX_LOG_DATA, resid);
1222 }
1223
1224 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1225 (write_state == WR_COPIED ? len : 0));
1226 lr = (lr_write_t *)&itx->itx_lr;
1227 if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1228 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1229 zil_itx_destroy(itx);
1230 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1231 lr = (lr_write_t *)&itx->itx_lr;
1232 write_state = WR_NEED_COPY;
1233 }
1234
1235 itx->itx_wr_state = write_state;
1236 if (write_state == WR_NEED_COPY)
1237 itx->itx_sod += len;
1238 lr->lr_foid = ZVOL_OBJ;
1239 lr->lr_offset = off;
1240 lr->lr_length = len;
1241 lr->lr_blkoff = 0;
1242 BP_ZERO(&lr->lr_blkptr);
1243
1244 itx->itx_private = zv;
1245 itx->itx_sync = sync;
1246
1247 zil_itx_assign(zilog, itx, tx);
1248
1249 off += len;
1250 resid -= len;
1251 }
1252}
1253
1254#ifdef sun
1255static int
1256zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1257 uint64_t size, boolean_t doread, boolean_t isdump)
1258{
1259 vdev_disk_t *dvd;
1260 int c;
1261 int numerrors = 0;
1262
1263 if (vd->vdev_ops == &vdev_mirror_ops ||
1264 vd->vdev_ops == &vdev_replacing_ops ||
1265 vd->vdev_ops == &vdev_spare_ops) {
1266 for (c = 0; c < vd->vdev_children; c++) {
1267 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1268 addr, offset, origoffset, size, doread, isdump);
1269 if (err != 0) {
1270 numerrors++;
1271 } else if (doread) {
1272 break;
1273 }
1274 }
1275 }
1276
1277 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1278 return (numerrors < vd->vdev_children ? 0 : EIO);
1279
1280 if (doread && !vdev_readable(vd))
1281 return (SET_ERROR(EIO));
1282 else if (!doread && !vdev_writeable(vd))
1283 return (SET_ERROR(EIO));
1284
1285 if (vd->vdev_ops == &vdev_raidz_ops) {
1286 return (vdev_raidz_physio(vd,
1287 addr, size, offset, origoffset, doread, isdump));
1288 }
1289
1290 offset += VDEV_LABEL_START_SIZE;
1291
1292 if (ddi_in_panic() || isdump) {
1293 ASSERT(!doread);
1294 if (doread)
1295 return (SET_ERROR(EIO));
1296 dvd = vd->vdev_tsd;
1297 ASSERT3P(dvd, !=, NULL);
1298 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1299 lbtodb(size)));
1300 } else {
1301 dvd = vd->vdev_tsd;
1302 ASSERT3P(dvd, !=, NULL);
1303 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1304 offset, doread ? B_READ : B_WRITE));
1305 }
1306}
1307
1308static int
1309zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1310 boolean_t doread, boolean_t isdump)
1311{
1312 vdev_t *vd;
1313 int error;
1314 zvol_extent_t *ze;
1315 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1316
1317 /* Must be sector aligned, and not stradle a block boundary. */
1318 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1319 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1320 return (SET_ERROR(EINVAL));
1321 }
1322 ASSERT(size <= zv->zv_volblocksize);
1323
1324 /* Locate the extent this belongs to */
1325 ze = list_head(&zv->zv_extents);
1326 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1327 offset -= ze->ze_nblks * zv->zv_volblocksize;
1328 ze = list_next(&zv->zv_extents, ze);
1329 }
1330
1331 if (ze == NULL)
1332 return (SET_ERROR(EINVAL));
1333
1334 if (!ddi_in_panic())
1335 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1336
1337 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1338 offset += DVA_GET_OFFSET(&ze->ze_dva);
1339 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1340 size, doread, isdump);
1341
1342 if (!ddi_in_panic())
1343 spa_config_exit(spa, SCL_STATE, FTAG);
1344
1345 return (error);
1346}
1347#endif /* sun */
1348
1349void
1350zvol_strategy(struct bio *bp)
1351{
1352 zvol_state_t *zv;
1353 uint64_t off, volsize;
1354 size_t resid;
1355 char *addr;
1356 objset_t *os;
1357 rl_t *rl;
1358 int error = 0;
1359 boolean_t doread = 0;
1360 boolean_t is_dumpified;
1361 boolean_t sync;
1362
1363 if (bp->bio_to)
1364 zv = bp->bio_to->private;
1365 else
1366 zv = bp->bio_dev->si_drv2;
1367
1368 if (zv == NULL) {
1369 error = ENXIO;
1370 goto out;
1371 }
1372
1373 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1374 error = EROFS;
1375 goto out;
1376 }
1377
1378 switch (bp->bio_cmd) {
1379 case BIO_FLUSH:
1380 goto sync;
1381 case BIO_READ:
1382 doread = 1;
1383 case BIO_WRITE:
1384 case BIO_DELETE:
1385 break;
1386 default:
1387 error = EOPNOTSUPP;
1388 goto out;
1389 }
1390
1391 off = bp->bio_offset;
1392 volsize = zv->zv_volsize;
1393
1394 os = zv->zv_objset;
1395 ASSERT(os != NULL);
1396
1397 addr = bp->bio_data;
1398 resid = bp->bio_length;
1399
1400 if (resid > 0 && (off < 0 || off >= volsize)) {
1401 error = EIO;
1402 goto out;
1403 }
1404
1405#ifdef illumos
1406 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1407#else
1408 is_dumpified = B_FALSE;
1409#endif
1410 sync = !doread && !is_dumpified &&
1411 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1412
1413 /*
1414 * There must be no buffer changes when doing a dmu_sync() because
1415 * we can't change the data whilst calculating the checksum.
1416 */
1417 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1418 doread ? RL_READER : RL_WRITER);
1419
1420 if (bp->bio_cmd == BIO_DELETE) {
1421 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1422 error = dmu_tx_assign(tx, TXG_WAIT);
1423 if (error != 0) {
1424 dmu_tx_abort(tx);
1425 } else {
1426 zvol_log_truncate(zv, tx, off, resid, B_TRUE);
1427 dmu_tx_commit(tx);
1428 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1429 off, resid);
1430 resid = 0;
1431 }
1432 goto unlock;
1433 }
1434
1435 while (resid != 0 && off < volsize) {
1436 size_t size = MIN(resid, zvol_maxphys);
1437#ifdef illumos
1438 if (is_dumpified) {
1439 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1440 error = zvol_dumpio(zv, addr, off, size,
1441 doread, B_FALSE);
1442 } else if (doread) {
1443#else
1444 if (doread) {
1445#endif
1446 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1447 DMU_READ_PREFETCH);
1448 } else {
1449 dmu_tx_t *tx = dmu_tx_create(os);
1450 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1451 error = dmu_tx_assign(tx, TXG_WAIT);
1452 if (error) {
1453 dmu_tx_abort(tx);
1454 } else {
1455 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1456 zvol_log_write(zv, tx, off, size, sync);
1457 dmu_tx_commit(tx);
1458 }
1459 }
1460 if (error) {
1461 /* convert checksum errors into IO errors */
1462 if (error == ECKSUM)
1463 error = SET_ERROR(EIO);
1464 break;
1465 }
1466 off += size;
1467 addr += size;
1468 resid -= size;
1469 }
1470unlock:
1471 zfs_range_unlock(rl);
1472
1473 bp->bio_completed = bp->bio_length - resid;
1474 if (bp->bio_completed < bp->bio_length && off > volsize)
1475 error = EINVAL;
1476
1477 if (sync) {
1478sync:
1479 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1480 }
1481out:
1482 if (bp->bio_to)
1483 g_io_deliver(bp, error);
1484 else
1485 biofinish(bp, NULL, error);
1486}
1487
1488#ifdef sun
1489/*
1490 * Set the buffer count to the zvol maximum transfer.
1491 * Using our own routine instead of the default minphys()
1492 * means that for larger writes we write bigger buffers on X86
1493 * (128K instead of 56K) and flush the disk write cache less often
1494 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1495 * 56K on X86 and 128K on sparc).
1496 */
1497void
1498zvol_minphys(struct buf *bp)
1499{
1500 if (bp->b_bcount > zvol_maxphys)
1501 bp->b_bcount = zvol_maxphys;
1502}
1503
1504int
1505zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1506{
1507 minor_t minor = getminor(dev);
1508 zvol_state_t *zv;
1509 int error = 0;
1510 uint64_t size;
1511 uint64_t boff;
1512 uint64_t resid;
1513
1514 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1515 if (zv == NULL)
1516 return (SET_ERROR(ENXIO));
1517
1518 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1519 return (SET_ERROR(EINVAL));
1520
1521 boff = ldbtob(blkno);
1522 resid = ldbtob(nblocks);
1523
1524 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1525
1526 while (resid) {
1527 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1528 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1529 if (error)
1530 break;
1531 boff += size;
1532 addr += size;
1533 resid -= size;
1534 }
1535
1536 return (error);
1537}
1538
1539/*ARGSUSED*/
1540int
1541zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1542{
1543 minor_t minor = getminor(dev);
1544#else
1545int
1546zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1547{
1548#endif
1549 zvol_state_t *zv;
1550 uint64_t volsize;
1551 rl_t *rl;
1552 int error = 0;
1553
1554#ifdef sun
1555 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1556 if (zv == NULL)
1557 return (SET_ERROR(ENXIO));
1558#else
1559 zv = dev->si_drv2;
1560#endif
1561
1562 volsize = zv->zv_volsize;
1563 if (uio->uio_resid > 0 &&
1564 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1565 return (SET_ERROR(EIO));
1566
1567#ifdef illumos
1568 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1569 error = physio(zvol_strategy, NULL, dev, B_READ,
1570 zvol_minphys, uio);
1571 return (error);
1572 }
1573#endif
1574
1575 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1576 RL_READER);
1577 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1578 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1579
1580 /* don't read past the end */
1581 if (bytes > volsize - uio->uio_loffset)
1582 bytes = volsize - uio->uio_loffset;
1583
1584 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1585 if (error) {
1586 /* convert checksum errors into IO errors */
1587 if (error == ECKSUM)
1588 error = SET_ERROR(EIO);
1589 break;
1590 }
1591 }
1592 zfs_range_unlock(rl);
1593 return (error);
1594}
1595
1596#ifdef sun
1597/*ARGSUSED*/
1598int
1599zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1600{
1601 minor_t minor = getminor(dev);
1602#else
1603int
1604zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1605{
1606#endif
1607 zvol_state_t *zv;
1608 uint64_t volsize;
1609 rl_t *rl;
1610 int error = 0;
1611 boolean_t sync;
1612
1613#ifdef sun
1614 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1615 if (zv == NULL)
1616 return (SET_ERROR(ENXIO));
1617#else
1618 zv = dev->si_drv2;
1619#endif
1620
1621 volsize = zv->zv_volsize;
1622 if (uio->uio_resid > 0 &&
1623 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1624 return (SET_ERROR(EIO));
1625
1626#ifdef illumos
1627 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1628 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1629 zvol_minphys, uio);
1630 return (error);
1631 }
1632#endif
1633
1634#ifdef sun
1635 sync = !(zv->zv_flags & ZVOL_WCE) ||
1636#else
1637 sync =
1638#endif
1639 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1640
1641 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1642 RL_WRITER);
1643 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1644 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1645 uint64_t off = uio->uio_loffset;
1646 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1647
1648 if (bytes > volsize - off) /* don't write past the end */
1649 bytes = volsize - off;
1650
1651 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1652 error = dmu_tx_assign(tx, TXG_WAIT);
1653 if (error) {
1654 dmu_tx_abort(tx);
1655 break;
1656 }
1657 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1658 if (error == 0)
1659 zvol_log_write(zv, tx, off, bytes, sync);
1660 dmu_tx_commit(tx);
1661
1662 if (error)
1663 break;
1664 }
1665 zfs_range_unlock(rl);
1666 if (sync)
1667 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1668 return (error);
1669}
1670
1671#ifdef sun
1672int
1673zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1674{
1675 struct uuid uuid = EFI_RESERVED;
1676 efi_gpe_t gpe = { 0 };
1677 uint32_t crc;
1678 dk_efi_t efi;
1679 int length;
1680 char *ptr;
1681
1682 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1683 return (SET_ERROR(EFAULT));
1684 ptr = (char *)(uintptr_t)efi.dki_data_64;
1685 length = efi.dki_length;
1686 /*
1687 * Some clients may attempt to request a PMBR for the
1688 * zvol. Currently this interface will return EINVAL to
1689 * such requests. These requests could be supported by
1690 * adding a check for lba == 0 and consing up an appropriate
1691 * PMBR.
1692 */
1693 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1694 return (SET_ERROR(EINVAL));
1695
1696 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1697 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1698 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1699
1700 if (efi.dki_lba == 1) {
1701 efi_gpt_t gpt = { 0 };
1702
1703 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1704 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1705 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1706 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1707 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1708 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1709 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1710 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1711 gpt.efi_gpt_SizeOfPartitionEntry =
1712 LE_32(sizeof (efi_gpe_t));
1713 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1714 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1715 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1716 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1717 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1718 flag))
1719 return (SET_ERROR(EFAULT));
1720 ptr += sizeof (gpt);
1721 length -= sizeof (gpt);
1722 }
1723 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1724 length), flag))
1725 return (SET_ERROR(EFAULT));
1726 return (0);
1727}
1728
1729/*
1730 * BEGIN entry points to allow external callers access to the volume.
1731 */
1732/*
1733 * Return the volume parameters needed for access from an external caller.
1734 * These values are invariant as long as the volume is held open.
1735 */
1736int
1737zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1738 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1739 void **rl_hdl, void **bonus_hdl)
1740{
1741 zvol_state_t *zv;
1742
1743 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1744 if (zv == NULL)
1745 return (SET_ERROR(ENXIO));
1746 if (zv->zv_flags & ZVOL_DUMPIFIED)
1747 return (SET_ERROR(ENXIO));
1748
1749 ASSERT(blksize && max_xfer_len && minor_hdl &&
1750 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1751
1752 *blksize = zv->zv_volblocksize;
1753 *max_xfer_len = (uint64_t)zvol_maxphys;
1754 *minor_hdl = zv;
1755 *objset_hdl = zv->zv_objset;
1756 *zil_hdl = zv->zv_zilog;
1757 *rl_hdl = &zv->zv_znode;
1758 *bonus_hdl = zv->zv_dbuf;
1759 return (0);
1760}
1761
1762/*
1763 * Return the current volume size to an external caller.
1764 * The size can change while the volume is open.
1765 */
1766uint64_t
1767zvol_get_volume_size(void *minor_hdl)
1768{
1769 zvol_state_t *zv = minor_hdl;
1770
1771 return (zv->zv_volsize);
1772}
1773
1774/*
1775 * Return the current WCE setting to an external caller.
1776 * The WCE setting can change while the volume is open.
1777 */
1778int
1779zvol_get_volume_wce(void *minor_hdl)
1780{
1781 zvol_state_t *zv = minor_hdl;
1782
1783 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1784}
1785
1786/*
1787 * Entry point for external callers to zvol_log_write
1788 */
1789void
1790zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1791 boolean_t sync)
1792{
1793 zvol_state_t *zv = minor_hdl;
1794
1795 zvol_log_write(zv, tx, off, resid, sync);
1796}
1797/*
1798 * END entry points to allow external callers access to the volume.
1799 */
1800#endif /* sun */
1801
1802/*
1803 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1804 */
1805static void
1806zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1807 boolean_t sync)
1808{
1809 itx_t *itx;
1810 lr_truncate_t *lr;
1811 zilog_t *zilog = zv->zv_zilog;
1812
1813 if (zil_replaying(zilog, tx))
1814 return;
1815
1816 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1817 lr = (lr_truncate_t *)&itx->itx_lr;
1818 lr->lr_foid = ZVOL_OBJ;
1819 lr->lr_offset = off;
1820 lr->lr_length = len;
1821
1822 itx->itx_sync = sync;
1823 zil_itx_assign(zilog, itx, tx);
1824}
1825
1826#ifdef sun
1827/*
1828 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
1829 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1830 */
1831/*ARGSUSED*/
1832int
1833zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1834{
1835 zvol_state_t *zv;
1836 struct dk_callback *dkc;
1837 int error = 0;
1838 rl_t *rl;
1839
1840 mutex_enter(&spa_namespace_lock);
1841
1842 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1843
1844 if (zv == NULL) {
1845 mutex_exit(&spa_namespace_lock);
1846 return (SET_ERROR(ENXIO));
1847 }
1848 ASSERT(zv->zv_total_opens > 0);
1849
1850 switch (cmd) {
1851
1852 case DKIOCINFO:
1853 {
1854 struct dk_cinfo dki;
1855
1856 bzero(&dki, sizeof (dki));
1857 (void) strcpy(dki.dki_cname, "zvol");
1858 (void) strcpy(dki.dki_dname, "zvol");
1859 dki.dki_ctype = DKC_UNKNOWN;
1860 dki.dki_unit = getminor(dev);
1861 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
1862 mutex_exit(&spa_namespace_lock);
1863 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1864 error = SET_ERROR(EFAULT);
1865 return (error);
1866 }
1867
1868 case DKIOCGMEDIAINFO:
1869 {
1870 struct dk_minfo dkm;
1871
1872 bzero(&dkm, sizeof (dkm));
1873 dkm.dki_lbsize = 1U << zv->zv_min_bs;
1874 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1875 dkm.dki_media_type = DK_UNKNOWN;
1876 mutex_exit(&spa_namespace_lock);
1877 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1878 error = SET_ERROR(EFAULT);
1879 return (error);
1880 }
1881
1882 case DKIOCGMEDIAINFOEXT:
1883 {
1884 struct dk_minfo_ext dkmext;
1885
1886 bzero(&dkmext, sizeof (dkmext));
1887 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
1888 dkmext.dki_pbsize = zv->zv_volblocksize;
1889 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1890 dkmext.dki_media_type = DK_UNKNOWN;
1891 mutex_exit(&spa_namespace_lock);
1892 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
1893 error = SET_ERROR(EFAULT);
1894 return (error);
1895 }
1896
1897 case DKIOCGETEFI:
1898 {
1899 uint64_t vs = zv->zv_volsize;
1900 uint8_t bs = zv->zv_min_bs;
1901
1902 mutex_exit(&spa_namespace_lock);
1903 error = zvol_getefi((void *)arg, flag, vs, bs);
1904 return (error);
1905 }
1906
1907 case DKIOCFLUSHWRITECACHE:
1908 dkc = (struct dk_callback *)arg;
1909 mutex_exit(&spa_namespace_lock);
1910 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1911 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1912 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
1913 error = 0;
1914 }
1915 return (error);
1916
1917 case DKIOCGETWCE:
1918 {
1919 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1920 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1921 flag))
1922 error = SET_ERROR(EFAULT);
1923 break;
1924 }
1925 case DKIOCSETWCE:
1926 {
1927 int wce;
1928 if (ddi_copyin((void *)arg, &wce, sizeof (int),
1929 flag)) {
1930 error = SET_ERROR(EFAULT);
1931 break;
1932 }
1933 if (wce) {
1934 zv->zv_flags |= ZVOL_WCE;
1935 mutex_exit(&spa_namespace_lock);
1936 } else {
1937 zv->zv_flags &= ~ZVOL_WCE;
1938 mutex_exit(&spa_namespace_lock);
1939 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1940 }
1941 return (0);
1942 }
1943
1944 case DKIOCGGEOM:
1945 case DKIOCGVTOC:
1946 /*
1947 * commands using these (like prtvtoc) expect ENOTSUP
1948 * since we're emulating an EFI label
1949 */
1950 error = SET_ERROR(ENOTSUP);
1951 break;
1952
1953 case DKIOCDUMPINIT:
1954 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1955 RL_WRITER);
1956 error = zvol_dumpify(zv);
1957 zfs_range_unlock(rl);
1958 break;
1959
1960 case DKIOCDUMPFINI:
1961 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1962 break;
1963 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1964 RL_WRITER);
1965 error = zvol_dump_fini(zv);
1966 zfs_range_unlock(rl);
1967 break;
1968
1969 case DKIOCFREE:
1970 {
1971 dkioc_free_t df;
1972 dmu_tx_t *tx;
1973
1974 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
1975 error = SET_ERROR(EFAULT);
1976 break;
1977 }
1978
1979 /*
1980 * Apply Postel's Law to length-checking. If they overshoot,
1981 * just blank out until the end, if there's a need to blank
1982 * out anything.
1983 */
1984 if (df.df_start >= zv->zv_volsize)
1985 break; /* No need to do anything... */
1986 if (df.df_start + df.df_length > zv->zv_volsize)
1987 df.df_length = DMU_OBJECT_END;
1988
1989 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
1990 RL_WRITER);
1991 tx = dmu_tx_create(zv->zv_objset);
1992 dmu_tx_mark_netfree(tx);
1993 error = dmu_tx_assign(tx, TXG_WAIT);
1994 if (error != 0) {
1995 dmu_tx_abort(tx);
1996 } else {
1997 zvol_log_truncate(zv, tx, df.df_start,
1998 df.df_length, B_TRUE);
1999 dmu_tx_commit(tx);
2000 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
2001 df.df_start, df.df_length);
2002 }
2003
2004 zfs_range_unlock(rl);
2005
2006 if (error == 0) {
2007 /*
2008 * If the write-cache is disabled or 'sync' property
2009 * is set to 'always' then treat this as a synchronous
2010 * operation (i.e. commit to zil).
2011 */
2012 if (!(zv->zv_flags & ZVOL_WCE) ||
2013 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
2014 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2015
2016 /*
2017 * If the caller really wants synchronous writes, and
2018 * can't wait for them, don't return until the write
2019 * is done.
2020 */
2021 if (df.df_flags & DF_WAIT_SYNC) {
2022 txg_wait_synced(
2023 dmu_objset_pool(zv->zv_objset), 0);
2024 }
2025 }
2026 break;
2027 }
2028
2029 default:
2030 error = SET_ERROR(ENOTTY);
2031 break;
2032
2033 }
2034 mutex_exit(&spa_namespace_lock);
2035 return (error);
2036}
2037#endif /* sun */
2038
2039int
2040zvol_busy(void)
2041{
2042 return (zvol_minors != 0);
2043}
2044
2045void
2046zvol_init(void)
2047{
2048 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2049 1) == 0);
2050 ZFS_LOG(1, "ZVOL Initialized.");
2051}
2052
2053void
2054zvol_fini(void)
2055{
2056 ddi_soft_state_fini(&zfsdev_state);
2057 ZFS_LOG(1, "ZVOL Deinitialized.");
2058}
2059
2060#ifdef sun
2061/*ARGSUSED*/
2062static int
2063zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2064{
2065 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2066
2067 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2068 return (1);
2069 return (0);
2070}
2071
2072/*ARGSUSED*/
2073static void
2074zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2075{
2076 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2077
2078 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2079}
2080
2081static int
2082zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2083{
2084 dmu_tx_t *tx;
2085 int error;
2086 objset_t *os = zv->zv_objset;
2087 spa_t *spa = dmu_objset_spa(os);
2088 vdev_t *vd = spa->spa_root_vdev;
2089 nvlist_t *nv = NULL;
2090 uint64_t version = spa_version(spa);
2091 enum zio_checksum checksum;
2092
2093 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2094 ASSERT(vd->vdev_ops == &vdev_root_ops);
2095
2096 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2097 DMU_OBJECT_END);
2098 /* wait for dmu_free_long_range to actually free the blocks */
2099 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2100
2101 /*
2102 * If the pool on which the dump device is being initialized has more
2103 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2104 * enabled. If so, bump that feature's counter to indicate that the
2105 * feature is active. We also check the vdev type to handle the
2106 * following case:
2107 * # zpool create test raidz disk1 disk2 disk3
2108 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2109 * the raidz vdev itself has 3 children.
2110 */
2111 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2112 if (!spa_feature_is_enabled(spa,
2113 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2114 return (SET_ERROR(ENOTSUP));
2115 (void) dsl_sync_task(spa_name(spa),
2116 zfs_mvdev_dump_feature_check,
| 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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 *
24 * Copyright (c) 2006-2010 Pawel Jakub Dawidek <pjd@FreeBSD.org>
25 * All rights reserved.
26 *
27 * Portions Copyright 2010 Robert Milkowski
28 *
29 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
30 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
31 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
32 */
33
34/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */
35
36/*
37 * ZFS volume emulation driver.
38 *
39 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
40 * Volumes are accessed through the symbolic links named:
41 *
42 * /dev/zvol/dsk/<pool_name>/<dataset_name>
43 * /dev/zvol/rdsk/<pool_name>/<dataset_name>
44 *
45 * These links are created by the /dev filesystem (sdev_zvolops.c).
46 * Volumes are persistent through reboot. No user command needs to be
47 * run before opening and using a device.
48 *
49 * FreeBSD notes.
50 * On FreeBSD ZVOLs are simply GEOM providers like any other storage device
51 * in the system.
52 */
53
54#include <sys/types.h>
55#include <sys/param.h>
56#include <sys/kernel.h>
57#include <sys/errno.h>
58#include <sys/uio.h>
59#include <sys/bio.h>
60#include <sys/buf.h>
61#include <sys/kmem.h>
62#include <sys/conf.h>
63#include <sys/cmn_err.h>
64#include <sys/stat.h>
65#include <sys/zap.h>
66#include <sys/spa.h>
67#include <sys/spa_impl.h>
68#include <sys/zio.h>
69#include <sys/disk.h>
70#include <sys/dmu_traverse.h>
71#include <sys/dnode.h>
72#include <sys/dsl_dataset.h>
73#include <sys/dsl_prop.h>
74#include <sys/dkio.h>
75#include <sys/byteorder.h>
76#include <sys/sunddi.h>
77#include <sys/dirent.h>
78#include <sys/policy.h>
79#include <sys/queue.h>
80#include <sys/fs/zfs.h>
81#include <sys/zfs_ioctl.h>
82#include <sys/zil.h>
83#include <sys/refcount.h>
84#include <sys/zfs_znode.h>
85#include <sys/zfs_rlock.h>
86#include <sys/vdev_impl.h>
87#include <sys/vdev_raidz.h>
88#include <sys/zvol.h>
89#include <sys/zil_impl.h>
90#include <sys/dbuf.h>
91#include <sys/dmu_tx.h>
92#include <sys/zfeature.h>
93#include <sys/zio_checksum.h>
94
95#include <geom/geom.h>
96
97#include "zfs_namecheck.h"
98
99struct g_class zfs_zvol_class = {
100 .name = "ZFS::ZVOL",
101 .version = G_VERSION,
102};
103
104DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol);
105
106void *zfsdev_state;
107static char *zvol_tag = "zvol_tag";
108
109#define ZVOL_DUMPSIZE "dumpsize"
110
111/*
112 * The spa_namespace_lock protects the zfsdev_state structure from being
113 * modified while it's being used, e.g. an open that comes in before a
114 * create finishes. It also protects temporary opens of the dataset so that,
115 * e.g., an open doesn't get a spurious EBUSY.
116 */
117static uint32_t zvol_minors;
118
119SYSCTL_DECL(_vfs_zfs);
120SYSCTL_NODE(_vfs_zfs, OID_AUTO, vol, CTLFLAG_RW, 0, "ZFS VOLUME");
121static int volmode = ZFS_VOLMODE_GEOM;
122TUNABLE_INT("vfs.zfs.vol.mode", &volmode);
123SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &volmode, 0,
124 "Expose as GEOM providers (1), device files (2) or neither");
125
126typedef struct zvol_extent {
127 list_node_t ze_node;
128 dva_t ze_dva; /* dva associated with this extent */
129 uint64_t ze_nblks; /* number of blocks in extent */
130} zvol_extent_t;
131
132/*
133 * The in-core state of each volume.
134 */
135typedef struct zvol_state {
136 LIST_ENTRY(zvol_state) zv_links;
137 char zv_name[MAXPATHLEN]; /* pool/dd name */
138 uint64_t zv_volsize; /* amount of space we advertise */
139 uint64_t zv_volblocksize; /* volume block size */
140 struct cdev *zv_dev; /* non-GEOM device */
141 struct g_provider *zv_provider; /* GEOM provider */
142 uint8_t zv_min_bs; /* minimum addressable block shift */
143 uint8_t zv_flags; /* readonly, dumpified, etc. */
144 objset_t *zv_objset; /* objset handle */
145 uint32_t zv_total_opens; /* total open count */
146 zilog_t *zv_zilog; /* ZIL handle */
147 list_t zv_extents; /* List of extents for dump */
148 znode_t zv_znode; /* for range locking */
149 dmu_buf_t *zv_dbuf; /* bonus handle */
150 int zv_state;
151 int zv_volmode; /* Provide GEOM or cdev */
152 struct bio_queue_head zv_queue;
153 struct mtx zv_queue_mtx; /* zv_queue mutex */
154} zvol_state_t;
155
156static LIST_HEAD(, zvol_state) all_zvols;
157
158/*
159 * zvol specific flags
160 */
161#define ZVOL_RDONLY 0x1
162#define ZVOL_DUMPIFIED 0x2
163#define ZVOL_EXCL 0x4
164#define ZVOL_WCE 0x8
165
166/*
167 * zvol maximum transfer in one DMU tx.
168 */
169int zvol_maxphys = DMU_MAX_ACCESS/2;
170
171static d_open_t zvol_d_open;
172static d_close_t zvol_d_close;
173static d_read_t zvol_read;
174static d_write_t zvol_write;
175static d_ioctl_t zvol_d_ioctl;
176static d_strategy_t zvol_strategy;
177
178static struct cdevsw zvol_cdevsw = {
179 .d_version = D_VERSION,
180 .d_open = zvol_d_open,
181 .d_close = zvol_d_close,
182 .d_read = zvol_read,
183 .d_write = zvol_write,
184 .d_ioctl = zvol_d_ioctl,
185 .d_strategy = zvol_strategy,
186 .d_name = "zvol",
187 .d_flags = D_DISK | D_TRACKCLOSE,
188};
189
190extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
191 nvlist_t *, nvlist_t *);
192static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off,
193 uint64_t len, boolean_t sync);
194static int zvol_remove_zv(zvol_state_t *);
195static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
196static int zvol_dumpify(zvol_state_t *zv);
197static int zvol_dump_fini(zvol_state_t *zv);
198static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
199
200static void zvol_geom_run(zvol_state_t *zv);
201static void zvol_geom_destroy(zvol_state_t *zv);
202static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
203static void zvol_geom_start(struct bio *bp);
204static void zvol_geom_worker(void *arg);
205
206static void
207zvol_size_changed(zvol_state_t *zv)
208{
209#ifdef sun
210 dev_t dev = makedevice(maj, min);
211
212 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
213 "Size", volsize) == DDI_SUCCESS);
214 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
215 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
216
217 /* Notify specfs to invalidate the cached size */
218 spec_size_invalidate(dev, VBLK);
219 spec_size_invalidate(dev, VCHR);
220#else /* !sun */
221 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
222 struct g_provider *pp;
223
224 pp = zv->zv_provider;
225 if (pp == NULL)
226 return;
227 g_topology_lock();
228 g_resize_provider(pp, zv->zv_volsize);
229 g_topology_unlock();
230 }
231#endif /* !sun */
232}
233
234int
235zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
236{
237 if (volsize == 0)
238 return (SET_ERROR(EINVAL));
239
240 if (volsize % blocksize != 0)
241 return (SET_ERROR(EINVAL));
242
243#ifdef _ILP32
244 if (volsize - 1 > SPEC_MAXOFFSET_T)
245 return (SET_ERROR(EOVERFLOW));
246#endif
247 return (0);
248}
249
250int
251zvol_check_volblocksize(uint64_t volblocksize)
252{
253 if (volblocksize < SPA_MINBLOCKSIZE ||
254 volblocksize > SPA_MAXBLOCKSIZE ||
255 !ISP2(volblocksize))
256 return (SET_ERROR(EDOM));
257
258 return (0);
259}
260
261int
262zvol_get_stats(objset_t *os, nvlist_t *nv)
263{
264 int error;
265 dmu_object_info_t doi;
266 uint64_t val;
267
268 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
269 if (error)
270 return (error);
271
272 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
273
274 error = dmu_object_info(os, ZVOL_OBJ, &doi);
275
276 if (error == 0) {
277 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
278 doi.doi_data_block_size);
279 }
280
281 return (error);
282}
283
284static zvol_state_t *
285zvol_minor_lookup(const char *name)
286{
287 zvol_state_t *zv;
288
289 ASSERT(MUTEX_HELD(&spa_namespace_lock));
290
291 LIST_FOREACH(zv, &all_zvols, zv_links) {
292 if (strcmp(zv->zv_name, name) == 0)
293 break;
294 }
295
296 return (zv);
297}
298
299/* extent mapping arg */
300struct maparg {
301 zvol_state_t *ma_zv;
302 uint64_t ma_blks;
303};
304
305/*ARGSUSED*/
306static int
307zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
308 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
309{
310 struct maparg *ma = arg;
311 zvol_extent_t *ze;
312 int bs = ma->ma_zv->zv_volblocksize;
313
314 if (BP_IS_HOLE(bp) ||
315 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
316 return (0);
317
318 VERIFY(!BP_IS_EMBEDDED(bp));
319
320 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
321 ma->ma_blks++;
322
323 /* Abort immediately if we have encountered gang blocks */
324 if (BP_IS_GANG(bp))
325 return (SET_ERROR(EFRAGS));
326
327 /*
328 * See if the block is at the end of the previous extent.
329 */
330 ze = list_tail(&ma->ma_zv->zv_extents);
331 if (ze &&
332 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
333 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
334 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
335 ze->ze_nblks++;
336 return (0);
337 }
338
339 dprintf_bp(bp, "%s", "next blkptr:");
340
341 /* start a new extent */
342 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
343 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
344 ze->ze_nblks = 1;
345 list_insert_tail(&ma->ma_zv->zv_extents, ze);
346 return (0);
347}
348
349static void
350zvol_free_extents(zvol_state_t *zv)
351{
352 zvol_extent_t *ze;
353
354 while (ze = list_head(&zv->zv_extents)) {
355 list_remove(&zv->zv_extents, ze);
356 kmem_free(ze, sizeof (zvol_extent_t));
357 }
358}
359
360static int
361zvol_get_lbas(zvol_state_t *zv)
362{
363 objset_t *os = zv->zv_objset;
364 struct maparg ma;
365 int err;
366
367 ma.ma_zv = zv;
368 ma.ma_blks = 0;
369 zvol_free_extents(zv);
370
371 /* commit any in-flight changes before traversing the dataset */
372 txg_wait_synced(dmu_objset_pool(os), 0);
373 err = traverse_dataset(dmu_objset_ds(os), 0,
374 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
375 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
376 zvol_free_extents(zv);
377 return (err ? err : EIO);
378 }
379
380 return (0);
381}
382
383/* ARGSUSED */
384void
385zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
386{
387 zfs_creat_t *zct = arg;
388 nvlist_t *nvprops = zct->zct_props;
389 int error;
390 uint64_t volblocksize, volsize;
391
392 VERIFY(nvlist_lookup_uint64(nvprops,
393 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
394 if (nvlist_lookup_uint64(nvprops,
395 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
396 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
397
398 /*
399 * These properties must be removed from the list so the generic
400 * property setting step won't apply to them.
401 */
402 VERIFY(nvlist_remove_all(nvprops,
403 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
404 (void) nvlist_remove_all(nvprops,
405 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
406
407 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
408 DMU_OT_NONE, 0, tx);
409 ASSERT(error == 0);
410
411 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
412 DMU_OT_NONE, 0, tx);
413 ASSERT(error == 0);
414
415 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
416 ASSERT(error == 0);
417}
418
419/*
420 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
421 * implement DKIOCFREE/free-long-range.
422 */
423static int
424zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
425{
426 uint64_t offset, length;
427
428 if (byteswap)
429 byteswap_uint64_array(lr, sizeof (*lr));
430
431 offset = lr->lr_offset;
432 length = lr->lr_length;
433
434 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
435}
436
437/*
438 * Replay a TX_WRITE ZIL transaction that didn't get committed
439 * after a system failure
440 */
441static int
442zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
443{
444 objset_t *os = zv->zv_objset;
445 char *data = (char *)(lr + 1); /* data follows lr_write_t */
446 uint64_t offset, length;
447 dmu_tx_t *tx;
448 int error;
449
450 if (byteswap)
451 byteswap_uint64_array(lr, sizeof (*lr));
452
453 offset = lr->lr_offset;
454 length = lr->lr_length;
455
456 /* If it's a dmu_sync() block, write the whole block */
457 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
458 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
459 if (length < blocksize) {
460 offset -= offset % blocksize;
461 length = blocksize;
462 }
463 }
464
465 tx = dmu_tx_create(os);
466 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
467 error = dmu_tx_assign(tx, TXG_WAIT);
468 if (error) {
469 dmu_tx_abort(tx);
470 } else {
471 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
472 dmu_tx_commit(tx);
473 }
474
475 return (error);
476}
477
478/* ARGSUSED */
479static int
480zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
481{
482 return (SET_ERROR(ENOTSUP));
483}
484
485/*
486 * Callback vectors for replaying records.
487 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
488 */
489zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
490 zvol_replay_err, /* 0 no such transaction type */
491 zvol_replay_err, /* TX_CREATE */
492 zvol_replay_err, /* TX_MKDIR */
493 zvol_replay_err, /* TX_MKXATTR */
494 zvol_replay_err, /* TX_SYMLINK */
495 zvol_replay_err, /* TX_REMOVE */
496 zvol_replay_err, /* TX_RMDIR */
497 zvol_replay_err, /* TX_LINK */
498 zvol_replay_err, /* TX_RENAME */
499 zvol_replay_write, /* TX_WRITE */
500 zvol_replay_truncate, /* TX_TRUNCATE */
501 zvol_replay_err, /* TX_SETATTR */
502 zvol_replay_err, /* TX_ACL */
503 zvol_replay_err, /* TX_CREATE_ACL */
504 zvol_replay_err, /* TX_CREATE_ATTR */
505 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
506 zvol_replay_err, /* TX_MKDIR_ACL */
507 zvol_replay_err, /* TX_MKDIR_ATTR */
508 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
509 zvol_replay_err, /* TX_WRITE2 */
510};
511
512#ifdef sun
513int
514zvol_name2minor(const char *name, minor_t *minor)
515{
516 zvol_state_t *zv;
517
518 mutex_enter(&spa_namespace_lock);
519 zv = zvol_minor_lookup(name);
520 if (minor && zv)
521 *minor = zv->zv_minor;
522 mutex_exit(&spa_namespace_lock);
523 return (zv ? 0 : -1);
524}
525#endif /* sun */
526
527/*
528 * Create a minor node (plus a whole lot more) for the specified volume.
529 */
530int
531zvol_create_minor(const char *name)
532{
533 zfs_soft_state_t *zs;
534 zvol_state_t *zv;
535 objset_t *os;
536 struct cdev *dev;
537 struct g_provider *pp;
538 struct g_geom *gp;
539 dmu_object_info_t doi;
540 uint64_t volsize, mode;
541 int error;
542
543 ZFS_LOG(1, "Creating ZVOL %s...", name);
544
545 mutex_enter(&spa_namespace_lock);
546
547 if (zvol_minor_lookup(name) != NULL) {
548 mutex_exit(&spa_namespace_lock);
549 return (SET_ERROR(EEXIST));
550 }
551
552 /* lie and say we're read-only */
553 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
554
555 if (error) {
556 mutex_exit(&spa_namespace_lock);
557 return (error);
558 }
559
560#ifdef sun
561 if ((minor = zfsdev_minor_alloc()) == 0) {
562 dmu_objset_disown(os, FTAG);
563 mutex_exit(&spa_namespace_lock);
564 return (SET_ERROR(ENXIO));
565 }
566
567 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
568 dmu_objset_disown(os, FTAG);
569 mutex_exit(&spa_namespace_lock);
570 return (SET_ERROR(EAGAIN));
571 }
572 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
573 (char *)name);
574
575 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
576
577 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
578 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
579 ddi_soft_state_free(zfsdev_state, minor);
580 dmu_objset_disown(os, FTAG);
581 mutex_exit(&spa_namespace_lock);
582 return (SET_ERROR(EAGAIN));
583 }
584
585 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
586
587 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
588 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
589 ddi_remove_minor_node(zfs_dip, chrbuf);
590 ddi_soft_state_free(zfsdev_state, minor);
591 dmu_objset_disown(os, FTAG);
592 mutex_exit(&spa_namespace_lock);
593 return (SET_ERROR(EAGAIN));
594 }
595
596 zs = ddi_get_soft_state(zfsdev_state, minor);
597 zs->zss_type = ZSST_ZVOL;
598 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
599#else /* !sun */
600
601 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
602 zv->zv_state = 0;
603 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
604 if (error) {
605 kmem_free(zv, sizeof(*zv));
606 dmu_objset_disown(os, zvol_tag);
607 mutex_exit(&spa_namespace_lock);
608 return (error);
609 }
610 error = dsl_prop_get_integer(name,
611 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
612 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
613 mode = volmode;
614
615 DROP_GIANT();
616 zv->zv_volsize = volsize;
617 zv->zv_volmode = mode;
618 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
619 g_topology_lock();
620 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
621 gp->start = zvol_geom_start;
622 gp->access = zvol_geom_access;
623 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
624 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
625 pp->sectorsize = DEV_BSIZE;
626 pp->mediasize = zv->zv_volsize;
627 pp->private = zv;
628
629 zv->zv_provider = pp;
630 bioq_init(&zv->zv_queue);
631 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
632 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
633 if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
634 &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR,
635 0640, "%s/%s", ZVOL_DRIVER, name) != 0) {
636 kmem_free(zv, sizeof(*zv));
637 dmu_objset_disown(os, FTAG);
638 mutex_exit(&spa_namespace_lock);
639 return (SET_ERROR(ENXIO));
640 }
641 zv->zv_dev = dev;
642 dev->si_iosize_max = MAXPHYS;
643 dev->si_drv2 = zv;
644 }
645 LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
646#endif /* !sun */
647
648 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
649 zv->zv_min_bs = DEV_BSHIFT;
650 zv->zv_objset = os;
651 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
652 zv->zv_flags |= ZVOL_RDONLY;
653 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
654 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
655 sizeof (rl_t), offsetof(rl_t, r_node));
656 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
657 offsetof(zvol_extent_t, ze_node));
658 /* get and cache the blocksize */
659 error = dmu_object_info(os, ZVOL_OBJ, &doi);
660 ASSERT(error == 0);
661 zv->zv_volblocksize = doi.doi_data_block_size;
662
663 if (spa_writeable(dmu_objset_spa(os))) {
664 if (zil_replay_disable)
665 zil_destroy(dmu_objset_zil(os), B_FALSE);
666 else
667 zil_replay(os, zv, zvol_replay_vector);
668 }
669 dmu_objset_disown(os, FTAG);
670 zv->zv_objset = NULL;
671
672 zvol_minors++;
673
674 mutex_exit(&spa_namespace_lock);
675
676#ifndef sun
677 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
678 zvol_geom_run(zv);
679 g_topology_unlock();
680 }
681 PICKUP_GIANT();
682#endif
683
684 ZFS_LOG(1, "ZVOL %s created.", name);
685
686 return (0);
687}
688
689/*
690 * Remove minor node for the specified volume.
691 */
692static int
693zvol_remove_zv(zvol_state_t *zv)
694{
695#ifdef sun
696 minor_t minor = zv->zv_minor;
697#endif
698
699 ASSERT(MUTEX_HELD(&spa_namespace_lock));
700 if (zv->zv_total_opens != 0)
701 return (SET_ERROR(EBUSY));
702
703 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
704
705#ifdef sun
706 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
707 ddi_remove_minor_node(zfs_dip, nmbuf);
708#else
709 LIST_REMOVE(zv, zv_links);
710 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
711 g_topology_lock();
712 zvol_geom_destroy(zv);
713 g_topology_unlock();
714 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV)
715 destroy_dev(zv->zv_dev);
716#endif /* sun */
717
718 avl_destroy(&zv->zv_znode.z_range_avl);
719 mutex_destroy(&zv->zv_znode.z_range_lock);
720
721 kmem_free(zv, sizeof(*zv));
722
723 zvol_minors--;
724 return (0);
725}
726
727int
728zvol_remove_minor(const char *name)
729{
730 zvol_state_t *zv;
731 int rc;
732
733 mutex_enter(&spa_namespace_lock);
734 if ((zv = zvol_minor_lookup(name)) == NULL) {
735 mutex_exit(&spa_namespace_lock);
736 return (SET_ERROR(ENXIO));
737 }
738 rc = zvol_remove_zv(zv);
739 mutex_exit(&spa_namespace_lock);
740 return (rc);
741}
742
743int
744zvol_first_open(zvol_state_t *zv)
745{
746 objset_t *os;
747 uint64_t volsize;
748 int error;
749 uint64_t readonly;
750
751 /* lie and say we're read-only */
752 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
753 zvol_tag, &os);
754 if (error)
755 return (error);
756
757 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
758 if (error) {
759 ASSERT(error == 0);
760 dmu_objset_disown(os, zvol_tag);
761 return (error);
762 }
763 zv->zv_objset = os;
764 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
765 if (error) {
766 dmu_objset_disown(os, zvol_tag);
767 return (error);
768 }
769 zv->zv_volsize = volsize;
770 zv->zv_zilog = zil_open(os, zvol_get_data);
771 zvol_size_changed(zv);
772
773 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
774 NULL) == 0);
775 if (readonly || dmu_objset_is_snapshot(os) ||
776 !spa_writeable(dmu_objset_spa(os)))
777 zv->zv_flags |= ZVOL_RDONLY;
778 else
779 zv->zv_flags &= ~ZVOL_RDONLY;
780 return (error);
781}
782
783void
784zvol_last_close(zvol_state_t *zv)
785{
786 zil_close(zv->zv_zilog);
787 zv->zv_zilog = NULL;
788
789 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
790 zv->zv_dbuf = NULL;
791
792 /*
793 * Evict cached data
794 */
795 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
796 !(zv->zv_flags & ZVOL_RDONLY))
797 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
798 dmu_objset_evict_dbufs(zv->zv_objset);
799
800 dmu_objset_disown(zv->zv_objset, zvol_tag);
801 zv->zv_objset = NULL;
802}
803
804#ifdef sun
805int
806zvol_prealloc(zvol_state_t *zv)
807{
808 objset_t *os = zv->zv_objset;
809 dmu_tx_t *tx;
810 uint64_t refd, avail, usedobjs, availobjs;
811 uint64_t resid = zv->zv_volsize;
812 uint64_t off = 0;
813
814 /* Check the space usage before attempting to allocate the space */
815 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
816 if (avail < zv->zv_volsize)
817 return (SET_ERROR(ENOSPC));
818
819 /* Free old extents if they exist */
820 zvol_free_extents(zv);
821
822 while (resid != 0) {
823 int error;
824 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE);
825
826 tx = dmu_tx_create(os);
827 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
828 error = dmu_tx_assign(tx, TXG_WAIT);
829 if (error) {
830 dmu_tx_abort(tx);
831 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
832 return (error);
833 }
834 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
835 dmu_tx_commit(tx);
836 off += bytes;
837 resid -= bytes;
838 }
839 txg_wait_synced(dmu_objset_pool(os), 0);
840
841 return (0);
842}
843#endif /* sun */
844
845static int
846zvol_update_volsize(objset_t *os, uint64_t volsize)
847{
848 dmu_tx_t *tx;
849 int error;
850
851 ASSERT(MUTEX_HELD(&spa_namespace_lock));
852
853 tx = dmu_tx_create(os);
854 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
855 dmu_tx_mark_netfree(tx);
856 error = dmu_tx_assign(tx, TXG_WAIT);
857 if (error) {
858 dmu_tx_abort(tx);
859 return (error);
860 }
861
862 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
863 &volsize, tx);
864 dmu_tx_commit(tx);
865
866 if (error == 0)
867 error = dmu_free_long_range(os,
868 ZVOL_OBJ, volsize, DMU_OBJECT_END);
869 return (error);
870}
871
872void
873zvol_remove_minors(const char *name)
874{
875 zvol_state_t *zv, *tzv;
876 size_t namelen;
877
878 namelen = strlen(name);
879
880 DROP_GIANT();
881 mutex_enter(&spa_namespace_lock);
882
883 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) {
884 if (strcmp(zv->zv_name, name) == 0 ||
885 (strncmp(zv->zv_name, name, namelen) == 0 &&
886 zv->zv_name[namelen] == '/')) {
887 (void) zvol_remove_zv(zv);
888 }
889 }
890
891 mutex_exit(&spa_namespace_lock);
892 PICKUP_GIANT();
893}
894
895int
896zvol_set_volsize(const char *name, major_t maj, uint64_t volsize)
897{
898 zvol_state_t *zv = NULL;
899 objset_t *os;
900 int error;
901 dmu_object_info_t doi;
902 uint64_t old_volsize = 0ULL;
903 uint64_t readonly;
904
905 mutex_enter(&spa_namespace_lock);
906 zv = zvol_minor_lookup(name);
907 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
908 mutex_exit(&spa_namespace_lock);
909 return (error);
910 }
911
912 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
913 (error = zvol_check_volsize(volsize,
914 doi.doi_data_block_size)) != 0)
915 goto out;
916
917 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly,
918 NULL) == 0);
919 if (readonly) {
920 error = EROFS;
921 goto out;
922 }
923
924 error = zvol_update_volsize(os, volsize);
925 /*
926 * Reinitialize the dump area to the new size. If we
927 * failed to resize the dump area then restore it back to
928 * its original size.
929 */
930 if (zv && error == 0) {
931#ifdef ZVOL_DUMP
932 if (zv->zv_flags & ZVOL_DUMPIFIED) {
933 old_volsize = zv->zv_volsize;
934 zv->zv_volsize = volsize;
935 if ((error = zvol_dumpify(zv)) != 0 ||
936 (error = dumpvp_resize()) != 0) {
937 (void) zvol_update_volsize(os, old_volsize);
938 zv->zv_volsize = old_volsize;
939 error = zvol_dumpify(zv);
940 }
941 }
942#endif /* ZVOL_DUMP */
943 if (error == 0) {
944 zv->zv_volsize = volsize;
945 zvol_size_changed(zv);
946 }
947 }
948
949#ifdef sun
950 /*
951 * Generate a LUN expansion event.
952 */
953 if (zv && error == 0) {
954 sysevent_id_t eid;
955 nvlist_t *attr;
956 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
957
958 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
959 zv->zv_minor);
960
961 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
962 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
963
964 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
965 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
966
967 nvlist_free(attr);
968 kmem_free(physpath, MAXPATHLEN);
969 }
970#endif /* sun */
971
972out:
973 dmu_objset_rele(os, FTAG);
974
975 mutex_exit(&spa_namespace_lock);
976
977 return (error);
978}
979
980/*ARGSUSED*/
981static int
982zvol_open(struct g_provider *pp, int flag, int count)
983{
984 zvol_state_t *zv;
985 int err = 0;
986 boolean_t locked = B_FALSE;
987
988 /*
989 * Protect against recursively entering spa_namespace_lock
990 * when spa_open() is used for a pool on a (local) ZVOL(s).
991 * This is needed since we replaced upstream zfsdev_state_lock
992 * with spa_namespace_lock in the ZVOL code.
993 * We are using the same trick as spa_open().
994 * Note that calls in zvol_first_open which need to resolve
995 * pool name to a spa object will enter spa_open()
996 * recursively, but that function already has all the
997 * necessary protection.
998 */
999 if (!MUTEX_HELD(&spa_namespace_lock)) {
1000 mutex_enter(&spa_namespace_lock);
1001 locked = B_TRUE;
1002 }
1003
1004 zv = pp->private;
1005 if (zv == NULL) {
1006 if (locked)
1007 mutex_exit(&spa_namespace_lock);
1008 return (SET_ERROR(ENXIO));
1009 }
1010
1011 if (zv->zv_total_opens == 0) {
1012 err = zvol_first_open(zv);
1013 if (err) {
1014 if (locked)
1015 mutex_exit(&spa_namespace_lock);
1016 return (err);
1017 }
1018 pp->mediasize = zv->zv_volsize;
1019 pp->stripeoffset = 0;
1020 pp->stripesize = zv->zv_volblocksize;
1021 }
1022 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1023 err = SET_ERROR(EROFS);
1024 goto out;
1025 }
1026 if (zv->zv_flags & ZVOL_EXCL) {
1027 err = SET_ERROR(EBUSY);
1028 goto out;
1029 }
1030#ifdef FEXCL
1031 if (flag & FEXCL) {
1032 if (zv->zv_total_opens != 0) {
1033 err = SET_ERROR(EBUSY);
1034 goto out;
1035 }
1036 zv->zv_flags |= ZVOL_EXCL;
1037 }
1038#endif
1039
1040 zv->zv_total_opens += count;
1041 if (locked)
1042 mutex_exit(&spa_namespace_lock);
1043
1044 return (err);
1045out:
1046 if (zv->zv_total_opens == 0)
1047 zvol_last_close(zv);
1048 if (locked)
1049 mutex_exit(&spa_namespace_lock);
1050 return (err);
1051}
1052
1053/*ARGSUSED*/
1054static int
1055zvol_close(struct g_provider *pp, int flag, int count)
1056{
1057 zvol_state_t *zv;
1058 int error = 0;
1059 boolean_t locked = B_FALSE;
1060
1061 /* See comment in zvol_open(). */
1062 if (!MUTEX_HELD(&spa_namespace_lock)) {
1063 mutex_enter(&spa_namespace_lock);
1064 locked = B_TRUE;
1065 }
1066
1067 zv = pp->private;
1068 if (zv == NULL) {
1069 if (locked)
1070 mutex_exit(&spa_namespace_lock);
1071 return (SET_ERROR(ENXIO));
1072 }
1073
1074 if (zv->zv_flags & ZVOL_EXCL) {
1075 ASSERT(zv->zv_total_opens == 1);
1076 zv->zv_flags &= ~ZVOL_EXCL;
1077 }
1078
1079 /*
1080 * If the open count is zero, this is a spurious close.
1081 * That indicates a bug in the kernel / DDI framework.
1082 */
1083 ASSERT(zv->zv_total_opens != 0);
1084
1085 /*
1086 * You may get multiple opens, but only one close.
1087 */
1088 zv->zv_total_opens -= count;
1089
1090 if (zv->zv_total_opens == 0)
1091 zvol_last_close(zv);
1092
1093 if (locked)
1094 mutex_exit(&spa_namespace_lock);
1095 return (error);
1096}
1097
1098static void
1099zvol_get_done(zgd_t *zgd, int error)
1100{
1101 if (zgd->zgd_db)
1102 dmu_buf_rele(zgd->zgd_db, zgd);
1103
1104 zfs_range_unlock(zgd->zgd_rl);
1105
1106 if (error == 0 && zgd->zgd_bp)
1107 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1108
1109 kmem_free(zgd, sizeof (zgd_t));
1110}
1111
1112/*
1113 * Get data to generate a TX_WRITE intent log record.
1114 */
1115static int
1116zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1117{
1118 zvol_state_t *zv = arg;
1119 objset_t *os = zv->zv_objset;
1120 uint64_t object = ZVOL_OBJ;
1121 uint64_t offset = lr->lr_offset;
1122 uint64_t size = lr->lr_length; /* length of user data */
1123 blkptr_t *bp = &lr->lr_blkptr;
1124 dmu_buf_t *db;
1125 zgd_t *zgd;
1126 int error;
1127
1128 ASSERT(zio != NULL);
1129 ASSERT(size != 0);
1130
1131 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1132 zgd->zgd_zilog = zv->zv_zilog;
1133 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
1134
1135 /*
1136 * Write records come in two flavors: immediate and indirect.
1137 * For small writes it's cheaper to store the data with the
1138 * log record (immediate); for large writes it's cheaper to
1139 * sync the data and get a pointer to it (indirect) so that
1140 * we don't have to write the data twice.
1141 */
1142 if (buf != NULL) { /* immediate write */
1143 error = dmu_read(os, object, offset, size, buf,
1144 DMU_READ_NO_PREFETCH);
1145 } else {
1146 size = zv->zv_volblocksize;
1147 offset = P2ALIGN(offset, size);
1148 error = dmu_buf_hold(os, object, offset, zgd, &db,
1149 DMU_READ_NO_PREFETCH);
1150 if (error == 0) {
1151 blkptr_t *obp = dmu_buf_get_blkptr(db);
1152 if (obp) {
1153 ASSERT(BP_IS_HOLE(bp));
1154 *bp = *obp;
1155 }
1156
1157 zgd->zgd_db = db;
1158 zgd->zgd_bp = bp;
1159
1160 ASSERT(db->db_offset == offset);
1161 ASSERT(db->db_size == size);
1162
1163 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1164 zvol_get_done, zgd);
1165
1166 if (error == 0)
1167 return (0);
1168 }
1169 }
1170
1171 zvol_get_done(zgd, error);
1172
1173 return (error);
1174}
1175
1176/*
1177 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1178 *
1179 * We store data in the log buffers if it's small enough.
1180 * Otherwise we will later flush the data out via dmu_sync().
1181 */
1182ssize_t zvol_immediate_write_sz = 32768;
1183
1184static void
1185zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1186 boolean_t sync)
1187{
1188 uint32_t blocksize = zv->zv_volblocksize;
1189 zilog_t *zilog = zv->zv_zilog;
1190 boolean_t slogging;
1191 ssize_t immediate_write_sz;
1192
1193 if (zil_replaying(zilog, tx))
1194 return;
1195
1196 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1197 ? 0 : zvol_immediate_write_sz;
1198
1199 slogging = spa_has_slogs(zilog->zl_spa) &&
1200 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1201
1202 while (resid) {
1203 itx_t *itx;
1204 lr_write_t *lr;
1205 ssize_t len;
1206 itx_wr_state_t write_state;
1207
1208 /*
1209 * Unlike zfs_log_write() we can be called with
1210 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1211 */
1212 if (blocksize > immediate_write_sz && !slogging &&
1213 resid >= blocksize && off % blocksize == 0) {
1214 write_state = WR_INDIRECT; /* uses dmu_sync */
1215 len = blocksize;
1216 } else if (sync) {
1217 write_state = WR_COPIED;
1218 len = MIN(ZIL_MAX_LOG_DATA, resid);
1219 } else {
1220 write_state = WR_NEED_COPY;
1221 len = MIN(ZIL_MAX_LOG_DATA, resid);
1222 }
1223
1224 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1225 (write_state == WR_COPIED ? len : 0));
1226 lr = (lr_write_t *)&itx->itx_lr;
1227 if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1228 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1229 zil_itx_destroy(itx);
1230 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1231 lr = (lr_write_t *)&itx->itx_lr;
1232 write_state = WR_NEED_COPY;
1233 }
1234
1235 itx->itx_wr_state = write_state;
1236 if (write_state == WR_NEED_COPY)
1237 itx->itx_sod += len;
1238 lr->lr_foid = ZVOL_OBJ;
1239 lr->lr_offset = off;
1240 lr->lr_length = len;
1241 lr->lr_blkoff = 0;
1242 BP_ZERO(&lr->lr_blkptr);
1243
1244 itx->itx_private = zv;
1245 itx->itx_sync = sync;
1246
1247 zil_itx_assign(zilog, itx, tx);
1248
1249 off += len;
1250 resid -= len;
1251 }
1252}
1253
1254#ifdef sun
1255static int
1256zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1257 uint64_t size, boolean_t doread, boolean_t isdump)
1258{
1259 vdev_disk_t *dvd;
1260 int c;
1261 int numerrors = 0;
1262
1263 if (vd->vdev_ops == &vdev_mirror_ops ||
1264 vd->vdev_ops == &vdev_replacing_ops ||
1265 vd->vdev_ops == &vdev_spare_ops) {
1266 for (c = 0; c < vd->vdev_children; c++) {
1267 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1268 addr, offset, origoffset, size, doread, isdump);
1269 if (err != 0) {
1270 numerrors++;
1271 } else if (doread) {
1272 break;
1273 }
1274 }
1275 }
1276
1277 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1278 return (numerrors < vd->vdev_children ? 0 : EIO);
1279
1280 if (doread && !vdev_readable(vd))
1281 return (SET_ERROR(EIO));
1282 else if (!doread && !vdev_writeable(vd))
1283 return (SET_ERROR(EIO));
1284
1285 if (vd->vdev_ops == &vdev_raidz_ops) {
1286 return (vdev_raidz_physio(vd,
1287 addr, size, offset, origoffset, doread, isdump));
1288 }
1289
1290 offset += VDEV_LABEL_START_SIZE;
1291
1292 if (ddi_in_panic() || isdump) {
1293 ASSERT(!doread);
1294 if (doread)
1295 return (SET_ERROR(EIO));
1296 dvd = vd->vdev_tsd;
1297 ASSERT3P(dvd, !=, NULL);
1298 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1299 lbtodb(size)));
1300 } else {
1301 dvd = vd->vdev_tsd;
1302 ASSERT3P(dvd, !=, NULL);
1303 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1304 offset, doread ? B_READ : B_WRITE));
1305 }
1306}
1307
1308static int
1309zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1310 boolean_t doread, boolean_t isdump)
1311{
1312 vdev_t *vd;
1313 int error;
1314 zvol_extent_t *ze;
1315 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1316
1317 /* Must be sector aligned, and not stradle a block boundary. */
1318 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1319 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1320 return (SET_ERROR(EINVAL));
1321 }
1322 ASSERT(size <= zv->zv_volblocksize);
1323
1324 /* Locate the extent this belongs to */
1325 ze = list_head(&zv->zv_extents);
1326 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1327 offset -= ze->ze_nblks * zv->zv_volblocksize;
1328 ze = list_next(&zv->zv_extents, ze);
1329 }
1330
1331 if (ze == NULL)
1332 return (SET_ERROR(EINVAL));
1333
1334 if (!ddi_in_panic())
1335 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1336
1337 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1338 offset += DVA_GET_OFFSET(&ze->ze_dva);
1339 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1340 size, doread, isdump);
1341
1342 if (!ddi_in_panic())
1343 spa_config_exit(spa, SCL_STATE, FTAG);
1344
1345 return (error);
1346}
1347#endif /* sun */
1348
1349void
1350zvol_strategy(struct bio *bp)
1351{
1352 zvol_state_t *zv;
1353 uint64_t off, volsize;
1354 size_t resid;
1355 char *addr;
1356 objset_t *os;
1357 rl_t *rl;
1358 int error = 0;
1359 boolean_t doread = 0;
1360 boolean_t is_dumpified;
1361 boolean_t sync;
1362
1363 if (bp->bio_to)
1364 zv = bp->bio_to->private;
1365 else
1366 zv = bp->bio_dev->si_drv2;
1367
1368 if (zv == NULL) {
1369 error = ENXIO;
1370 goto out;
1371 }
1372
1373 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1374 error = EROFS;
1375 goto out;
1376 }
1377
1378 switch (bp->bio_cmd) {
1379 case BIO_FLUSH:
1380 goto sync;
1381 case BIO_READ:
1382 doread = 1;
1383 case BIO_WRITE:
1384 case BIO_DELETE:
1385 break;
1386 default:
1387 error = EOPNOTSUPP;
1388 goto out;
1389 }
1390
1391 off = bp->bio_offset;
1392 volsize = zv->zv_volsize;
1393
1394 os = zv->zv_objset;
1395 ASSERT(os != NULL);
1396
1397 addr = bp->bio_data;
1398 resid = bp->bio_length;
1399
1400 if (resid > 0 && (off < 0 || off >= volsize)) {
1401 error = EIO;
1402 goto out;
1403 }
1404
1405#ifdef illumos
1406 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1407#else
1408 is_dumpified = B_FALSE;
1409#endif
1410 sync = !doread && !is_dumpified &&
1411 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1412
1413 /*
1414 * There must be no buffer changes when doing a dmu_sync() because
1415 * we can't change the data whilst calculating the checksum.
1416 */
1417 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1418 doread ? RL_READER : RL_WRITER);
1419
1420 if (bp->bio_cmd == BIO_DELETE) {
1421 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1422 error = dmu_tx_assign(tx, TXG_WAIT);
1423 if (error != 0) {
1424 dmu_tx_abort(tx);
1425 } else {
1426 zvol_log_truncate(zv, tx, off, resid, B_TRUE);
1427 dmu_tx_commit(tx);
1428 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1429 off, resid);
1430 resid = 0;
1431 }
1432 goto unlock;
1433 }
1434
1435 while (resid != 0 && off < volsize) {
1436 size_t size = MIN(resid, zvol_maxphys);
1437#ifdef illumos
1438 if (is_dumpified) {
1439 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1440 error = zvol_dumpio(zv, addr, off, size,
1441 doread, B_FALSE);
1442 } else if (doread) {
1443#else
1444 if (doread) {
1445#endif
1446 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1447 DMU_READ_PREFETCH);
1448 } else {
1449 dmu_tx_t *tx = dmu_tx_create(os);
1450 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1451 error = dmu_tx_assign(tx, TXG_WAIT);
1452 if (error) {
1453 dmu_tx_abort(tx);
1454 } else {
1455 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1456 zvol_log_write(zv, tx, off, size, sync);
1457 dmu_tx_commit(tx);
1458 }
1459 }
1460 if (error) {
1461 /* convert checksum errors into IO errors */
1462 if (error == ECKSUM)
1463 error = SET_ERROR(EIO);
1464 break;
1465 }
1466 off += size;
1467 addr += size;
1468 resid -= size;
1469 }
1470unlock:
1471 zfs_range_unlock(rl);
1472
1473 bp->bio_completed = bp->bio_length - resid;
1474 if (bp->bio_completed < bp->bio_length && off > volsize)
1475 error = EINVAL;
1476
1477 if (sync) {
1478sync:
1479 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1480 }
1481out:
1482 if (bp->bio_to)
1483 g_io_deliver(bp, error);
1484 else
1485 biofinish(bp, NULL, error);
1486}
1487
1488#ifdef sun
1489/*
1490 * Set the buffer count to the zvol maximum transfer.
1491 * Using our own routine instead of the default minphys()
1492 * means that for larger writes we write bigger buffers on X86
1493 * (128K instead of 56K) and flush the disk write cache less often
1494 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1495 * 56K on X86 and 128K on sparc).
1496 */
1497void
1498zvol_minphys(struct buf *bp)
1499{
1500 if (bp->b_bcount > zvol_maxphys)
1501 bp->b_bcount = zvol_maxphys;
1502}
1503
1504int
1505zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1506{
1507 minor_t minor = getminor(dev);
1508 zvol_state_t *zv;
1509 int error = 0;
1510 uint64_t size;
1511 uint64_t boff;
1512 uint64_t resid;
1513
1514 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1515 if (zv == NULL)
1516 return (SET_ERROR(ENXIO));
1517
1518 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1519 return (SET_ERROR(EINVAL));
1520
1521 boff = ldbtob(blkno);
1522 resid = ldbtob(nblocks);
1523
1524 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1525
1526 while (resid) {
1527 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1528 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1529 if (error)
1530 break;
1531 boff += size;
1532 addr += size;
1533 resid -= size;
1534 }
1535
1536 return (error);
1537}
1538
1539/*ARGSUSED*/
1540int
1541zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1542{
1543 minor_t minor = getminor(dev);
1544#else
1545int
1546zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1547{
1548#endif
1549 zvol_state_t *zv;
1550 uint64_t volsize;
1551 rl_t *rl;
1552 int error = 0;
1553
1554#ifdef sun
1555 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1556 if (zv == NULL)
1557 return (SET_ERROR(ENXIO));
1558#else
1559 zv = dev->si_drv2;
1560#endif
1561
1562 volsize = zv->zv_volsize;
1563 if (uio->uio_resid > 0 &&
1564 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1565 return (SET_ERROR(EIO));
1566
1567#ifdef illumos
1568 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1569 error = physio(zvol_strategy, NULL, dev, B_READ,
1570 zvol_minphys, uio);
1571 return (error);
1572 }
1573#endif
1574
1575 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1576 RL_READER);
1577 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1578 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1579
1580 /* don't read past the end */
1581 if (bytes > volsize - uio->uio_loffset)
1582 bytes = volsize - uio->uio_loffset;
1583
1584 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1585 if (error) {
1586 /* convert checksum errors into IO errors */
1587 if (error == ECKSUM)
1588 error = SET_ERROR(EIO);
1589 break;
1590 }
1591 }
1592 zfs_range_unlock(rl);
1593 return (error);
1594}
1595
1596#ifdef sun
1597/*ARGSUSED*/
1598int
1599zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1600{
1601 minor_t minor = getminor(dev);
1602#else
1603int
1604zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1605{
1606#endif
1607 zvol_state_t *zv;
1608 uint64_t volsize;
1609 rl_t *rl;
1610 int error = 0;
1611 boolean_t sync;
1612
1613#ifdef sun
1614 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1615 if (zv == NULL)
1616 return (SET_ERROR(ENXIO));
1617#else
1618 zv = dev->si_drv2;
1619#endif
1620
1621 volsize = zv->zv_volsize;
1622 if (uio->uio_resid > 0 &&
1623 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1624 return (SET_ERROR(EIO));
1625
1626#ifdef illumos
1627 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1628 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1629 zvol_minphys, uio);
1630 return (error);
1631 }
1632#endif
1633
1634#ifdef sun
1635 sync = !(zv->zv_flags & ZVOL_WCE) ||
1636#else
1637 sync =
1638#endif
1639 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1640
1641 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1642 RL_WRITER);
1643 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1644 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1645 uint64_t off = uio->uio_loffset;
1646 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1647
1648 if (bytes > volsize - off) /* don't write past the end */
1649 bytes = volsize - off;
1650
1651 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1652 error = dmu_tx_assign(tx, TXG_WAIT);
1653 if (error) {
1654 dmu_tx_abort(tx);
1655 break;
1656 }
1657 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1658 if (error == 0)
1659 zvol_log_write(zv, tx, off, bytes, sync);
1660 dmu_tx_commit(tx);
1661
1662 if (error)
1663 break;
1664 }
1665 zfs_range_unlock(rl);
1666 if (sync)
1667 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1668 return (error);
1669}
1670
1671#ifdef sun
1672int
1673zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1674{
1675 struct uuid uuid = EFI_RESERVED;
1676 efi_gpe_t gpe = { 0 };
1677 uint32_t crc;
1678 dk_efi_t efi;
1679 int length;
1680 char *ptr;
1681
1682 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1683 return (SET_ERROR(EFAULT));
1684 ptr = (char *)(uintptr_t)efi.dki_data_64;
1685 length = efi.dki_length;
1686 /*
1687 * Some clients may attempt to request a PMBR for the
1688 * zvol. Currently this interface will return EINVAL to
1689 * such requests. These requests could be supported by
1690 * adding a check for lba == 0 and consing up an appropriate
1691 * PMBR.
1692 */
1693 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1694 return (SET_ERROR(EINVAL));
1695
1696 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1697 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1698 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1699
1700 if (efi.dki_lba == 1) {
1701 efi_gpt_t gpt = { 0 };
1702
1703 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1704 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1705 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1706 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1707 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1708 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1709 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1710 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1711 gpt.efi_gpt_SizeOfPartitionEntry =
1712 LE_32(sizeof (efi_gpe_t));
1713 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1714 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1715 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1716 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1717 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1718 flag))
1719 return (SET_ERROR(EFAULT));
1720 ptr += sizeof (gpt);
1721 length -= sizeof (gpt);
1722 }
1723 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1724 length), flag))
1725 return (SET_ERROR(EFAULT));
1726 return (0);
1727}
1728
1729/*
1730 * BEGIN entry points to allow external callers access to the volume.
1731 */
1732/*
1733 * Return the volume parameters needed for access from an external caller.
1734 * These values are invariant as long as the volume is held open.
1735 */
1736int
1737zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1738 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1739 void **rl_hdl, void **bonus_hdl)
1740{
1741 zvol_state_t *zv;
1742
1743 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1744 if (zv == NULL)
1745 return (SET_ERROR(ENXIO));
1746 if (zv->zv_flags & ZVOL_DUMPIFIED)
1747 return (SET_ERROR(ENXIO));
1748
1749 ASSERT(blksize && max_xfer_len && minor_hdl &&
1750 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1751
1752 *blksize = zv->zv_volblocksize;
1753 *max_xfer_len = (uint64_t)zvol_maxphys;
1754 *minor_hdl = zv;
1755 *objset_hdl = zv->zv_objset;
1756 *zil_hdl = zv->zv_zilog;
1757 *rl_hdl = &zv->zv_znode;
1758 *bonus_hdl = zv->zv_dbuf;
1759 return (0);
1760}
1761
1762/*
1763 * Return the current volume size to an external caller.
1764 * The size can change while the volume is open.
1765 */
1766uint64_t
1767zvol_get_volume_size(void *minor_hdl)
1768{
1769 zvol_state_t *zv = minor_hdl;
1770
1771 return (zv->zv_volsize);
1772}
1773
1774/*
1775 * Return the current WCE setting to an external caller.
1776 * The WCE setting can change while the volume is open.
1777 */
1778int
1779zvol_get_volume_wce(void *minor_hdl)
1780{
1781 zvol_state_t *zv = minor_hdl;
1782
1783 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1784}
1785
1786/*
1787 * Entry point for external callers to zvol_log_write
1788 */
1789void
1790zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1791 boolean_t sync)
1792{
1793 zvol_state_t *zv = minor_hdl;
1794
1795 zvol_log_write(zv, tx, off, resid, sync);
1796}
1797/*
1798 * END entry points to allow external callers access to the volume.
1799 */
1800#endif /* sun */
1801
1802/*
1803 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1804 */
1805static void
1806zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1807 boolean_t sync)
1808{
1809 itx_t *itx;
1810 lr_truncate_t *lr;
1811 zilog_t *zilog = zv->zv_zilog;
1812
1813 if (zil_replaying(zilog, tx))
1814 return;
1815
1816 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1817 lr = (lr_truncate_t *)&itx->itx_lr;
1818 lr->lr_foid = ZVOL_OBJ;
1819 lr->lr_offset = off;
1820 lr->lr_length = len;
1821
1822 itx->itx_sync = sync;
1823 zil_itx_assign(zilog, itx, tx);
1824}
1825
1826#ifdef sun
1827/*
1828 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
1829 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1830 */
1831/*ARGSUSED*/
1832int
1833zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1834{
1835 zvol_state_t *zv;
1836 struct dk_callback *dkc;
1837 int error = 0;
1838 rl_t *rl;
1839
1840 mutex_enter(&spa_namespace_lock);
1841
1842 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1843
1844 if (zv == NULL) {
1845 mutex_exit(&spa_namespace_lock);
1846 return (SET_ERROR(ENXIO));
1847 }
1848 ASSERT(zv->zv_total_opens > 0);
1849
1850 switch (cmd) {
1851
1852 case DKIOCINFO:
1853 {
1854 struct dk_cinfo dki;
1855
1856 bzero(&dki, sizeof (dki));
1857 (void) strcpy(dki.dki_cname, "zvol");
1858 (void) strcpy(dki.dki_dname, "zvol");
1859 dki.dki_ctype = DKC_UNKNOWN;
1860 dki.dki_unit = getminor(dev);
1861 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
1862 mutex_exit(&spa_namespace_lock);
1863 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1864 error = SET_ERROR(EFAULT);
1865 return (error);
1866 }
1867
1868 case DKIOCGMEDIAINFO:
1869 {
1870 struct dk_minfo dkm;
1871
1872 bzero(&dkm, sizeof (dkm));
1873 dkm.dki_lbsize = 1U << zv->zv_min_bs;
1874 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1875 dkm.dki_media_type = DK_UNKNOWN;
1876 mutex_exit(&spa_namespace_lock);
1877 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1878 error = SET_ERROR(EFAULT);
1879 return (error);
1880 }
1881
1882 case DKIOCGMEDIAINFOEXT:
1883 {
1884 struct dk_minfo_ext dkmext;
1885
1886 bzero(&dkmext, sizeof (dkmext));
1887 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
1888 dkmext.dki_pbsize = zv->zv_volblocksize;
1889 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1890 dkmext.dki_media_type = DK_UNKNOWN;
1891 mutex_exit(&spa_namespace_lock);
1892 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
1893 error = SET_ERROR(EFAULT);
1894 return (error);
1895 }
1896
1897 case DKIOCGETEFI:
1898 {
1899 uint64_t vs = zv->zv_volsize;
1900 uint8_t bs = zv->zv_min_bs;
1901
1902 mutex_exit(&spa_namespace_lock);
1903 error = zvol_getefi((void *)arg, flag, vs, bs);
1904 return (error);
1905 }
1906
1907 case DKIOCFLUSHWRITECACHE:
1908 dkc = (struct dk_callback *)arg;
1909 mutex_exit(&spa_namespace_lock);
1910 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1911 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1912 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
1913 error = 0;
1914 }
1915 return (error);
1916
1917 case DKIOCGETWCE:
1918 {
1919 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1920 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1921 flag))
1922 error = SET_ERROR(EFAULT);
1923 break;
1924 }
1925 case DKIOCSETWCE:
1926 {
1927 int wce;
1928 if (ddi_copyin((void *)arg, &wce, sizeof (int),
1929 flag)) {
1930 error = SET_ERROR(EFAULT);
1931 break;
1932 }
1933 if (wce) {
1934 zv->zv_flags |= ZVOL_WCE;
1935 mutex_exit(&spa_namespace_lock);
1936 } else {
1937 zv->zv_flags &= ~ZVOL_WCE;
1938 mutex_exit(&spa_namespace_lock);
1939 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1940 }
1941 return (0);
1942 }
1943
1944 case DKIOCGGEOM:
1945 case DKIOCGVTOC:
1946 /*
1947 * commands using these (like prtvtoc) expect ENOTSUP
1948 * since we're emulating an EFI label
1949 */
1950 error = SET_ERROR(ENOTSUP);
1951 break;
1952
1953 case DKIOCDUMPINIT:
1954 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1955 RL_WRITER);
1956 error = zvol_dumpify(zv);
1957 zfs_range_unlock(rl);
1958 break;
1959
1960 case DKIOCDUMPFINI:
1961 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1962 break;
1963 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1964 RL_WRITER);
1965 error = zvol_dump_fini(zv);
1966 zfs_range_unlock(rl);
1967 break;
1968
1969 case DKIOCFREE:
1970 {
1971 dkioc_free_t df;
1972 dmu_tx_t *tx;
1973
1974 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
1975 error = SET_ERROR(EFAULT);
1976 break;
1977 }
1978
1979 /*
1980 * Apply Postel's Law to length-checking. If they overshoot,
1981 * just blank out until the end, if there's a need to blank
1982 * out anything.
1983 */
1984 if (df.df_start >= zv->zv_volsize)
1985 break; /* No need to do anything... */
1986 if (df.df_start + df.df_length > zv->zv_volsize)
1987 df.df_length = DMU_OBJECT_END;
1988
1989 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
1990 RL_WRITER);
1991 tx = dmu_tx_create(zv->zv_objset);
1992 dmu_tx_mark_netfree(tx);
1993 error = dmu_tx_assign(tx, TXG_WAIT);
1994 if (error != 0) {
1995 dmu_tx_abort(tx);
1996 } else {
1997 zvol_log_truncate(zv, tx, df.df_start,
1998 df.df_length, B_TRUE);
1999 dmu_tx_commit(tx);
2000 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
2001 df.df_start, df.df_length);
2002 }
2003
2004 zfs_range_unlock(rl);
2005
2006 if (error == 0) {
2007 /*
2008 * If the write-cache is disabled or 'sync' property
2009 * is set to 'always' then treat this as a synchronous
2010 * operation (i.e. commit to zil).
2011 */
2012 if (!(zv->zv_flags & ZVOL_WCE) ||
2013 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
2014 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2015
2016 /*
2017 * If the caller really wants synchronous writes, and
2018 * can't wait for them, don't return until the write
2019 * is done.
2020 */
2021 if (df.df_flags & DF_WAIT_SYNC) {
2022 txg_wait_synced(
2023 dmu_objset_pool(zv->zv_objset), 0);
2024 }
2025 }
2026 break;
2027 }
2028
2029 default:
2030 error = SET_ERROR(ENOTTY);
2031 break;
2032
2033 }
2034 mutex_exit(&spa_namespace_lock);
2035 return (error);
2036}
2037#endif /* sun */
2038
2039int
2040zvol_busy(void)
2041{
2042 return (zvol_minors != 0);
2043}
2044
2045void
2046zvol_init(void)
2047{
2048 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2049 1) == 0);
2050 ZFS_LOG(1, "ZVOL Initialized.");
2051}
2052
2053void
2054zvol_fini(void)
2055{
2056 ddi_soft_state_fini(&zfsdev_state);
2057 ZFS_LOG(1, "ZVOL Deinitialized.");
2058}
2059
2060#ifdef sun
2061/*ARGSUSED*/
2062static int
2063zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2064{
2065 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2066
2067 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2068 return (1);
2069 return (0);
2070}
2071
2072/*ARGSUSED*/
2073static void
2074zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2075{
2076 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2077
2078 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2079}
2080
2081static int
2082zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2083{
2084 dmu_tx_t *tx;
2085 int error;
2086 objset_t *os = zv->zv_objset;
2087 spa_t *spa = dmu_objset_spa(os);
2088 vdev_t *vd = spa->spa_root_vdev;
2089 nvlist_t *nv = NULL;
2090 uint64_t version = spa_version(spa);
2091 enum zio_checksum checksum;
2092
2093 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2094 ASSERT(vd->vdev_ops == &vdev_root_ops);
2095
2096 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2097 DMU_OBJECT_END);
2098 /* wait for dmu_free_long_range to actually free the blocks */
2099 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2100
2101 /*
2102 * If the pool on which the dump device is being initialized has more
2103 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2104 * enabled. If so, bump that feature's counter to indicate that the
2105 * feature is active. We also check the vdev type to handle the
2106 * following case:
2107 * # zpool create test raidz disk1 disk2 disk3
2108 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2109 * the raidz vdev itself has 3 children.
2110 */
2111 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2112 if (!spa_feature_is_enabled(spa,
2113 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2114 return (SET_ERROR(ENOTSUP));
2115 (void) dsl_sync_task(spa_name(spa),
2116 zfs_mvdev_dump_feature_check,
|
2118 }
2119
2120 tx = dmu_tx_create(os);
2121 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2122 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2123 error = dmu_tx_assign(tx, TXG_WAIT);
2124 if (error) {
2125 dmu_tx_abort(tx);
2126 return (error);
2127 }
2128
2129 /*
2130 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2131 * function. Otherwise, use the old default -- OFF.
2132 */
2133 checksum = spa_feature_is_active(spa,
2134 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2135 ZIO_CHECKSUM_OFF;
2136
2137 /*
2138 * If we are resizing the dump device then we only need to
2139 * update the refreservation to match the newly updated
2140 * zvolsize. Otherwise, we save off the original state of the
2141 * zvol so that we can restore them if the zvol is ever undumpified.
2142 */
2143 if (resize) {
2144 error = zap_update(os, ZVOL_ZAP_OBJ,
2145 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2146 &zv->zv_volsize, tx);
2147 } else {
2148 uint64_t checksum, compress, refresrv, vbs, dedup;
2149
2150 error = dsl_prop_get_integer(zv->zv_name,
2151 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2152 error = error ? error : dsl_prop_get_integer(zv->zv_name,
2153 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
2154 error = error ? error : dsl_prop_get_integer(zv->zv_name,
2155 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
2156 error = error ? error : dsl_prop_get_integer(zv->zv_name,
2157 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
2158 if (version >= SPA_VERSION_DEDUP) {
2159 error = error ? error :
2160 dsl_prop_get_integer(zv->zv_name,
2161 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2162 }
2163
2164 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
2165 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2166 &compress, tx);
2167 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
2168 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
2169 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
2170 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2171 &refresrv, tx);
2172 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
2173 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2174 &vbs, tx);
2175 error = error ? error : dmu_object_set_blocksize(
2176 os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx);
2177 if (version >= SPA_VERSION_DEDUP) {
2178 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
2179 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2180 &dedup, tx);
2181 }
2182 if (error == 0)
2183 zv->zv_volblocksize = SPA_MAXBLOCKSIZE;
2184 }
2185 dmu_tx_commit(tx);
2186
2187 /*
2188 * We only need update the zvol's property if we are initializing
2189 * the dump area for the first time.
2190 */
2191 if (!resize) {
2192 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2193 VERIFY(nvlist_add_uint64(nv,
2194 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2195 VERIFY(nvlist_add_uint64(nv,
2196 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2197 ZIO_COMPRESS_OFF) == 0);
2198 VERIFY(nvlist_add_uint64(nv,
2199 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2200 checksum) == 0);
2201 if (version >= SPA_VERSION_DEDUP) {
2202 VERIFY(nvlist_add_uint64(nv,
2203 zfs_prop_to_name(ZFS_PROP_DEDUP),
2204 ZIO_CHECKSUM_OFF) == 0);
2205 }
2206
2207 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2208 nv, NULL);
2209 nvlist_free(nv);
2210
2211 if (error)
2212 return (error);
2213 }
2214
2215 /* Allocate the space for the dump */
2216 error = zvol_prealloc(zv);
2217 return (error);
2218}
2219
2220static int
2221zvol_dumpify(zvol_state_t *zv)
2222{
2223 int error = 0;
2224 uint64_t dumpsize = 0;
2225 dmu_tx_t *tx;
2226 objset_t *os = zv->zv_objset;
2227
2228 if (zv->zv_flags & ZVOL_RDONLY)
2229 return (SET_ERROR(EROFS));
2230
2231 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2232 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2233 boolean_t resize = (dumpsize > 0);
2234
2235 if ((error = zvol_dump_init(zv, resize)) != 0) {
2236 (void) zvol_dump_fini(zv);
2237 return (error);
2238 }
2239 }
2240
2241 /*
2242 * Build up our lba mapping.
2243 */
2244 error = zvol_get_lbas(zv);
2245 if (error) {
2246 (void) zvol_dump_fini(zv);
2247 return (error);
2248 }
2249
2250 tx = dmu_tx_create(os);
2251 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2252 error = dmu_tx_assign(tx, TXG_WAIT);
2253 if (error) {
2254 dmu_tx_abort(tx);
2255 (void) zvol_dump_fini(zv);
2256 return (error);
2257 }
2258
2259 zv->zv_flags |= ZVOL_DUMPIFIED;
2260 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2261 &zv->zv_volsize, tx);
2262 dmu_tx_commit(tx);
2263
2264 if (error) {
2265 (void) zvol_dump_fini(zv);
2266 return (error);
2267 }
2268
2269 txg_wait_synced(dmu_objset_pool(os), 0);
2270 return (0);
2271}
2272
2273static int
2274zvol_dump_fini(zvol_state_t *zv)
2275{
2276 dmu_tx_t *tx;
2277 objset_t *os = zv->zv_objset;
2278 nvlist_t *nv;
2279 int error = 0;
2280 uint64_t checksum, compress, refresrv, vbs, dedup;
2281 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2282
2283 /*
2284 * Attempt to restore the zvol back to its pre-dumpified state.
2285 * This is a best-effort attempt as it's possible that not all
2286 * of these properties were initialized during the dumpify process
2287 * (i.e. error during zvol_dump_init).
2288 */
2289
2290 tx = dmu_tx_create(os);
2291 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2292 error = dmu_tx_assign(tx, TXG_WAIT);
2293 if (error) {
2294 dmu_tx_abort(tx);
2295 return (error);
2296 }
2297 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2298 dmu_tx_commit(tx);
2299
2300 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2301 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2302 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2303 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2304 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2305 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2306 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2307 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2308
2309 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2310 (void) nvlist_add_uint64(nv,
2311 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2312 (void) nvlist_add_uint64(nv,
2313 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2314 (void) nvlist_add_uint64(nv,
2315 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2316 if (version >= SPA_VERSION_DEDUP &&
2317 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2318 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2319 (void) nvlist_add_uint64(nv,
2320 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2321 }
2322 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2323 nv, NULL);
2324 nvlist_free(nv);
2325
2326 zvol_free_extents(zv);
2327 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2328 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2329 /* wait for dmu_free_long_range to actually free the blocks */
2330 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2331 tx = dmu_tx_create(os);
2332 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2333 error = dmu_tx_assign(tx, TXG_WAIT);
2334 if (error) {
2335 dmu_tx_abort(tx);
2336 return (error);
2337 }
2338 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2339 zv->zv_volblocksize = vbs;
2340 dmu_tx_commit(tx);
2341
2342 return (0);
2343}
2344#endif /* sun */
2345
2346static void
2347zvol_geom_run(zvol_state_t *zv)
2348{
2349 struct g_provider *pp;
2350
2351 pp = zv->zv_provider;
2352 g_error_provider(pp, 0);
2353
2354 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
2355 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2356}
2357
2358static void
2359zvol_geom_destroy(zvol_state_t *zv)
2360{
2361 struct g_provider *pp;
2362
2363 g_topology_assert();
2364
2365 mtx_lock(&zv->zv_queue_mtx);
2366 zv->zv_state = 1;
2367 wakeup_one(&zv->zv_queue);
2368 while (zv->zv_state != 2)
2369 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2370 mtx_destroy(&zv->zv_queue_mtx);
2371
2372 pp = zv->zv_provider;
2373 zv->zv_provider = NULL;
2374 pp->private = NULL;
2375 g_wither_geom(pp->geom, ENXIO);
2376}
2377
2378static int
2379zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2380{
2381 int count, error, flags;
2382
2383 g_topology_assert();
2384
2385 /*
2386 * To make it easier we expect either open or close, but not both
2387 * at the same time.
2388 */
2389 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2390 (acr <= 0 && acw <= 0 && ace <= 0),
2391 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2392 pp->name, acr, acw, ace));
2393
2394 if (pp->private == NULL) {
2395 if (acr <= 0 && acw <= 0 && ace <= 0)
2396 return (0);
2397 return (pp->error);
2398 }
2399
2400 /*
2401 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2402 * because GEOM already handles that and handles it a bit differently.
2403 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2404 * only one exclusive consumer, no matter if it is reader or writer.
2405 * I like better the way GEOM works so I'll leave it for GEOM to
2406 * decide what to do.
2407 */
2408
2409 count = acr + acw + ace;
2410 if (count == 0)
2411 return (0);
2412
2413 flags = 0;
2414 if (acr != 0 || ace != 0)
2415 flags |= FREAD;
2416 if (acw != 0)
2417 flags |= FWRITE;
2418
2419 g_topology_unlock();
2420 if (count > 0)
2421 error = zvol_open(pp, flags, count);
2422 else
2423 error = zvol_close(pp, flags, -count);
2424 g_topology_lock();
2425 return (error);
2426}
2427
2428static void
2429zvol_geom_start(struct bio *bp)
2430{
2431 zvol_state_t *zv;
2432 boolean_t first;
2433
2434 zv = bp->bio_to->private;
2435 ASSERT(zv != NULL);
2436 switch (bp->bio_cmd) {
2437 case BIO_FLUSH:
2438 if (!THREAD_CAN_SLEEP())
2439 goto enqueue;
2440 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2441 g_io_deliver(bp, 0);
2442 break;
2443 case BIO_READ:
2444 case BIO_WRITE:
2445 case BIO_DELETE:
2446 if (!THREAD_CAN_SLEEP())
2447 goto enqueue;
2448 zvol_strategy(bp);
2449 break;
2450 case BIO_GETATTR:
2451 if (g_handleattr_int(bp, "GEOM::candelete", 1))
2452 return;
2453 /* FALLTHROUGH */
2454 default:
2455 g_io_deliver(bp, EOPNOTSUPP);
2456 break;
2457 }
2458 return;
2459
2460enqueue:
2461 mtx_lock(&zv->zv_queue_mtx);
2462 first = (bioq_first(&zv->zv_queue) == NULL);
2463 bioq_insert_tail(&zv->zv_queue, bp);
2464 mtx_unlock(&zv->zv_queue_mtx);
2465 if (first)
2466 wakeup_one(&zv->zv_queue);
2467}
2468
2469static void
2470zvol_geom_worker(void *arg)
2471{
2472 zvol_state_t *zv;
2473 struct bio *bp;
2474
2475 thread_lock(curthread);
2476 sched_prio(curthread, PRIBIO);
2477 thread_unlock(curthread);
2478
2479 zv = arg;
2480 for (;;) {
2481 mtx_lock(&zv->zv_queue_mtx);
2482 bp = bioq_takefirst(&zv->zv_queue);
2483 if (bp == NULL) {
2484 if (zv->zv_state == 1) {
2485 zv->zv_state = 2;
2486 wakeup(&zv->zv_state);
2487 mtx_unlock(&zv->zv_queue_mtx);
2488 kthread_exit();
2489 }
2490 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2491 "zvol:io", 0);
2492 continue;
2493 }
2494 mtx_unlock(&zv->zv_queue_mtx);
2495 switch (bp->bio_cmd) {
2496 case BIO_FLUSH:
2497 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2498 g_io_deliver(bp, 0);
2499 break;
2500 case BIO_READ:
2501 case BIO_WRITE:
2502 zvol_strategy(bp);
2503 break;
2504 }
2505 }
2506}
2507
2508extern boolean_t dataset_name_hidden(const char *name);
2509
2510static int
2511zvol_create_snapshots(objset_t *os, const char *name)
2512{
2513 uint64_t cookie, obj;
2514 char *sname;
2515 int error, len;
2516
2517 cookie = obj = 0;
2518 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2519
2520#if 0
2521 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2522 DS_FIND_SNAPSHOTS);
2523#endif
2524
2525 for (;;) {
2526 len = snprintf(sname, MAXPATHLEN, "%s@", name);
2527 if (len >= MAXPATHLEN) {
2528 dmu_objset_rele(os, FTAG);
2529 error = ENAMETOOLONG;
2530 break;
2531 }
2532
2533 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2534 error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2535 sname + len, &obj, &cookie, NULL);
2536 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2537 if (error != 0) {
2538 if (error == ENOENT)
2539 error = 0;
2540 break;
2541 }
2542
2543 if ((error = zvol_create_minor(sname)) != 0) {
2544 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2545 sname, error);
2546 break;
2547 }
2548 }
2549
2550 kmem_free(sname, MAXPATHLEN);
2551 return (error);
2552}
2553
2554int
2555zvol_create_minors(const char *name)
2556{
2557 uint64_t cookie;
2558 objset_t *os;
2559 char *osname, *p;
2560 int error, len;
2561
2562 if (dataset_name_hidden(name))
2563 return (0);
2564
2565 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2566 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2567 name, error);
2568 return (error);
2569 }
2570 if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2571 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2572 dsl_pool_rele(dmu_objset_pool(os), FTAG);
2573 if ((error = zvol_create_minor(name)) == 0)
2574 error = zvol_create_snapshots(os, name);
2575 else {
2576 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2577 name, error);
2578 }
2579 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2580 dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2581 return (error);
2582 }
2583 if (dmu_objset_type(os) != DMU_OST_ZFS) {
2584 dmu_objset_rele(os, FTAG);
2585 return (0);
2586 }
2587
2588 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2589 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2590 dmu_objset_rele(os, FTAG);
2591 kmem_free(osname, MAXPATHLEN);
2592 return (ENOENT);
2593 }
2594 p = osname + strlen(osname);
2595 len = MAXPATHLEN - (p - osname);
2596
2597#if 0
2598 /* Prefetch the datasets. */
2599 cookie = 0;
2600 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2601 if (!dataset_name_hidden(osname))
2602 (void) dmu_objset_prefetch(osname, NULL);
2603 }
2604#endif
2605
2606 cookie = 0;
2607 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2608 &cookie) == 0) {
2609 dmu_objset_rele(os, FTAG);
2610 (void)zvol_create_minors(osname);
2611 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2612 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2613 name, error);
2614 return (error);
2615 }
2616 }
2617
2618 dmu_objset_rele(os, FTAG);
2619 kmem_free(osname, MAXPATHLEN);
2620 return (0);
2621}
2622
2623static void
2624zvol_rename_minor(zvol_state_t *zv, const char *newname)
2625{
2626 struct g_geom *gp;
2627 struct g_provider *pp;
2628 struct cdev *dev;
2629
2630 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2631
2632 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
2633 g_topology_lock();
2634 pp = zv->zv_provider;
2635 ASSERT(pp != NULL);
2636 gp = pp->geom;
2637 ASSERT(gp != NULL);
2638
2639 zv->zv_provider = NULL;
2640 g_wither_provider(pp, ENXIO);
2641
2642 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
2643 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
2644 pp->sectorsize = DEV_BSIZE;
2645 pp->mediasize = zv->zv_volsize;
2646 pp->private = zv;
2647 zv->zv_provider = pp;
2648 g_error_provider(pp, 0);
2649 g_topology_unlock();
2650 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
2651 dev = zv->zv_dev;
2652 ASSERT(dev != NULL);
2653 zv->zv_dev = NULL;
2654 destroy_dev(dev);
2655
2656 if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
2657 &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR,
2658 0640, "%s/%s", ZVOL_DRIVER, newname) == 0) {
2659 zv->zv_dev = dev;
2660 dev->si_iosize_max = MAXPHYS;
2661 dev->si_drv2 = zv;
2662 }
2663 }
2664 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
2665}
2666
2667void
2668zvol_rename_minors(const char *oldname, const char *newname)
2669{
2670 char name[MAXPATHLEN];
2671 struct g_provider *pp;
2672 struct g_geom *gp;
2673 size_t oldnamelen, newnamelen;
2674 zvol_state_t *zv;
2675 char *namebuf;
2676
2677 oldnamelen = strlen(oldname);
2678 newnamelen = strlen(newname);
2679
2680 DROP_GIANT();
2681 mutex_enter(&spa_namespace_lock);
2682
2683 LIST_FOREACH(zv, &all_zvols, zv_links) {
2684 if (strcmp(zv->zv_name, oldname) == 0) {
2685 zvol_rename_minor(zv, newname);
2686 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
2687 (zv->zv_name[oldnamelen] == '/' ||
2688 zv->zv_name[oldnamelen] == '@')) {
2689 snprintf(name, sizeof(name), "%s%c%s", newname,
2690 zv->zv_name[oldnamelen],
2691 zv->zv_name + oldnamelen + 1);
2692 zvol_rename_minor(zv, name);
2693 }
2694 }
2695
2696 mutex_exit(&spa_namespace_lock);
2697 PICKUP_GIANT();
2698}
2699
2700static int
2701zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2702{
2703 zvol_state_t *zv;
2704 int err = 0;
2705
2706 mutex_enter(&spa_namespace_lock);
2707 zv = dev->si_drv2;
2708 if (zv == NULL) {
2709 mutex_exit(&spa_namespace_lock);
2710 return(ENXIO); /* zvol_create_minor() not done yet */
2711 }
2712
2713 if (zv->zv_total_opens == 0)
2714 err = zvol_first_open(zv);
2715 if (err) {
2716 mutex_exit(&spa_namespace_lock);
2717 return (err);
2718 }
2719 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
2720 err = SET_ERROR(EROFS);
2721 goto out;
2722 }
2723 if (zv->zv_flags & ZVOL_EXCL) {
2724 err = SET_ERROR(EBUSY);
2725 goto out;
2726 }
2727#ifdef FEXCL
2728 if (flags & FEXCL) {
2729 if (zv->zv_total_opens != 0) {
2730 err = SET_ERROR(EBUSY);
2731 goto out;
2732 }
2733 zv->zv_flags |= ZVOL_EXCL;
2734 }
2735#endif
2736
2737 zv->zv_total_opens++;
2738 mutex_exit(&spa_namespace_lock);
2739 return (err);
2740out:
2741 if (zv->zv_total_opens == 0)
2742 zvol_last_close(zv);
2743 mutex_exit(&spa_namespace_lock);
2744 return (err);
2745}
2746
2747static int
2748zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2749{
2750 zvol_state_t *zv;
2751 int err = 0;
2752
2753 mutex_enter(&spa_namespace_lock);
2754 zv = dev->si_drv2;
2755 if (zv == NULL) {
2756 mutex_exit(&spa_namespace_lock);
2757 return(ENXIO);
2758 }
2759
2760 if (zv->zv_flags & ZVOL_EXCL) {
2761 ASSERT(zv->zv_total_opens == 1);
2762 zv->zv_flags &= ~ZVOL_EXCL;
2763 }
2764
2765 /*
2766 * If the open count is zero, this is a spurious close.
2767 * That indicates a bug in the kernel / DDI framework.
2768 */
2769 ASSERT(zv->zv_total_opens != 0);
2770
2771 /*
2772 * You may get multiple opens, but only one close.
2773 */
2774 zv->zv_total_opens--;
2775
2776 if (zv->zv_total_opens == 0)
2777 zvol_last_close(zv);
2778
2779 mutex_exit(&spa_namespace_lock);
2780 return (0);
2781}
2782
2783static int
2784zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
2785{
2786 zvol_state_t *zv;
2787 rl_t *rl;
2788 off_t offset, length, chunk;
2789 int i, error;
2790 u_int u;
2791
2792 zv = dev->si_drv2;
2793
2794 error = 0;
2795 KASSERT(zv->zv_total_opens > 0,
2796 ("Device with zero access count in zvol_d_ioctl"));
2797
2798 i = IOCPARM_LEN(cmd);
2799 switch (cmd) {
2800 case DIOCGSECTORSIZE:
2801 *(u_int *)data = DEV_BSIZE;
2802 break;
2803 case DIOCGMEDIASIZE:
2804 *(off_t *)data = zv->zv_volsize;
2805 break;
2806 case DIOCGFLUSH:
2807 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2808 break;
2809 case DIOCGDELETE:
2810 offset = ((off_t *)data)[0];
2811 length = ((off_t *)data)[1];
2812 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
2813 offset < 0 || offset >= zv->zv_volsize ||
2814 length <= 0) {
2815 printf("%s: offset=%jd length=%jd\n", __func__, offset,
2816 length);
2817 error = EINVAL;
2818 break;
2819 }
2820
2821 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
2822 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
2823 error = dmu_tx_assign(tx, TXG_WAIT);
2824 if (error != 0) {
2825 dmu_tx_abort(tx);
2826 } else {
2827 zvol_log_truncate(zv, tx, offset, length, B_TRUE);
2828 dmu_tx_commit(tx);
2829 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
2830 offset, length);
2831 }
2832 zfs_range_unlock(rl);
2833 if (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)
2834 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2835 break;
2836 case DIOCGSTRIPESIZE:
2837 *(off_t *)data = zv->zv_volblocksize;
2838 break;
2839 case DIOCGSTRIPEOFFSET:
2840 *(off_t *)data = 0;
2841 break;
2842 default:
2843 error = ENOIOCTL;
2844 }
2845
2846 return (error);
2847}
| 2119 }
2120
2121 tx = dmu_tx_create(os);
2122 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2123 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2124 error = dmu_tx_assign(tx, TXG_WAIT);
2125 if (error) {
2126 dmu_tx_abort(tx);
2127 return (error);
2128 }
2129
2130 /*
2131 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2132 * function. Otherwise, use the old default -- OFF.
2133 */
2134 checksum = spa_feature_is_active(spa,
2135 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2136 ZIO_CHECKSUM_OFF;
2137
2138 /*
2139 * If we are resizing the dump device then we only need to
2140 * update the refreservation to match the newly updated
2141 * zvolsize. Otherwise, we save off the original state of the
2142 * zvol so that we can restore them if the zvol is ever undumpified.
2143 */
2144 if (resize) {
2145 error = zap_update(os, ZVOL_ZAP_OBJ,
2146 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2147 &zv->zv_volsize, tx);
2148 } else {
2149 uint64_t checksum, compress, refresrv, vbs, dedup;
2150
2151 error = dsl_prop_get_integer(zv->zv_name,
2152 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2153 error = error ? error : dsl_prop_get_integer(zv->zv_name,
2154 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
2155 error = error ? error : dsl_prop_get_integer(zv->zv_name,
2156 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
2157 error = error ? error : dsl_prop_get_integer(zv->zv_name,
2158 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
2159 if (version >= SPA_VERSION_DEDUP) {
2160 error = error ? error :
2161 dsl_prop_get_integer(zv->zv_name,
2162 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2163 }
2164
2165 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
2166 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2167 &compress, tx);
2168 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
2169 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
2170 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
2171 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2172 &refresrv, tx);
2173 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
2174 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2175 &vbs, tx);
2176 error = error ? error : dmu_object_set_blocksize(
2177 os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx);
2178 if (version >= SPA_VERSION_DEDUP) {
2179 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
2180 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2181 &dedup, tx);
2182 }
2183 if (error == 0)
2184 zv->zv_volblocksize = SPA_MAXBLOCKSIZE;
2185 }
2186 dmu_tx_commit(tx);
2187
2188 /*
2189 * We only need update the zvol's property if we are initializing
2190 * the dump area for the first time.
2191 */
2192 if (!resize) {
2193 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2194 VERIFY(nvlist_add_uint64(nv,
2195 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2196 VERIFY(nvlist_add_uint64(nv,
2197 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2198 ZIO_COMPRESS_OFF) == 0);
2199 VERIFY(nvlist_add_uint64(nv,
2200 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2201 checksum) == 0);
2202 if (version >= SPA_VERSION_DEDUP) {
2203 VERIFY(nvlist_add_uint64(nv,
2204 zfs_prop_to_name(ZFS_PROP_DEDUP),
2205 ZIO_CHECKSUM_OFF) == 0);
2206 }
2207
2208 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2209 nv, NULL);
2210 nvlist_free(nv);
2211
2212 if (error)
2213 return (error);
2214 }
2215
2216 /* Allocate the space for the dump */
2217 error = zvol_prealloc(zv);
2218 return (error);
2219}
2220
2221static int
2222zvol_dumpify(zvol_state_t *zv)
2223{
2224 int error = 0;
2225 uint64_t dumpsize = 0;
2226 dmu_tx_t *tx;
2227 objset_t *os = zv->zv_objset;
2228
2229 if (zv->zv_flags & ZVOL_RDONLY)
2230 return (SET_ERROR(EROFS));
2231
2232 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2233 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2234 boolean_t resize = (dumpsize > 0);
2235
2236 if ((error = zvol_dump_init(zv, resize)) != 0) {
2237 (void) zvol_dump_fini(zv);
2238 return (error);
2239 }
2240 }
2241
2242 /*
2243 * Build up our lba mapping.
2244 */
2245 error = zvol_get_lbas(zv);
2246 if (error) {
2247 (void) zvol_dump_fini(zv);
2248 return (error);
2249 }
2250
2251 tx = dmu_tx_create(os);
2252 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2253 error = dmu_tx_assign(tx, TXG_WAIT);
2254 if (error) {
2255 dmu_tx_abort(tx);
2256 (void) zvol_dump_fini(zv);
2257 return (error);
2258 }
2259
2260 zv->zv_flags |= ZVOL_DUMPIFIED;
2261 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2262 &zv->zv_volsize, tx);
2263 dmu_tx_commit(tx);
2264
2265 if (error) {
2266 (void) zvol_dump_fini(zv);
2267 return (error);
2268 }
2269
2270 txg_wait_synced(dmu_objset_pool(os), 0);
2271 return (0);
2272}
2273
2274static int
2275zvol_dump_fini(zvol_state_t *zv)
2276{
2277 dmu_tx_t *tx;
2278 objset_t *os = zv->zv_objset;
2279 nvlist_t *nv;
2280 int error = 0;
2281 uint64_t checksum, compress, refresrv, vbs, dedup;
2282 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2283
2284 /*
2285 * Attempt to restore the zvol back to its pre-dumpified state.
2286 * This is a best-effort attempt as it's possible that not all
2287 * of these properties were initialized during the dumpify process
2288 * (i.e. error during zvol_dump_init).
2289 */
2290
2291 tx = dmu_tx_create(os);
2292 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2293 error = dmu_tx_assign(tx, TXG_WAIT);
2294 if (error) {
2295 dmu_tx_abort(tx);
2296 return (error);
2297 }
2298 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2299 dmu_tx_commit(tx);
2300
2301 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2302 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2303 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2304 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2305 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2306 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2307 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2308 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2309
2310 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2311 (void) nvlist_add_uint64(nv,
2312 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2313 (void) nvlist_add_uint64(nv,
2314 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2315 (void) nvlist_add_uint64(nv,
2316 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2317 if (version >= SPA_VERSION_DEDUP &&
2318 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2319 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2320 (void) nvlist_add_uint64(nv,
2321 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2322 }
2323 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2324 nv, NULL);
2325 nvlist_free(nv);
2326
2327 zvol_free_extents(zv);
2328 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2329 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2330 /* wait for dmu_free_long_range to actually free the blocks */
2331 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2332 tx = dmu_tx_create(os);
2333 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2334 error = dmu_tx_assign(tx, TXG_WAIT);
2335 if (error) {
2336 dmu_tx_abort(tx);
2337 return (error);
2338 }
2339 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2340 zv->zv_volblocksize = vbs;
2341 dmu_tx_commit(tx);
2342
2343 return (0);
2344}
2345#endif /* sun */
2346
2347static void
2348zvol_geom_run(zvol_state_t *zv)
2349{
2350 struct g_provider *pp;
2351
2352 pp = zv->zv_provider;
2353 g_error_provider(pp, 0);
2354
2355 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
2356 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2357}
2358
2359static void
2360zvol_geom_destroy(zvol_state_t *zv)
2361{
2362 struct g_provider *pp;
2363
2364 g_topology_assert();
2365
2366 mtx_lock(&zv->zv_queue_mtx);
2367 zv->zv_state = 1;
2368 wakeup_one(&zv->zv_queue);
2369 while (zv->zv_state != 2)
2370 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2371 mtx_destroy(&zv->zv_queue_mtx);
2372
2373 pp = zv->zv_provider;
2374 zv->zv_provider = NULL;
2375 pp->private = NULL;
2376 g_wither_geom(pp->geom, ENXIO);
2377}
2378
2379static int
2380zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2381{
2382 int count, error, flags;
2383
2384 g_topology_assert();
2385
2386 /*
2387 * To make it easier we expect either open or close, but not both
2388 * at the same time.
2389 */
2390 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2391 (acr <= 0 && acw <= 0 && ace <= 0),
2392 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2393 pp->name, acr, acw, ace));
2394
2395 if (pp->private == NULL) {
2396 if (acr <= 0 && acw <= 0 && ace <= 0)
2397 return (0);
2398 return (pp->error);
2399 }
2400
2401 /*
2402 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2403 * because GEOM already handles that and handles it a bit differently.
2404 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2405 * only one exclusive consumer, no matter if it is reader or writer.
2406 * I like better the way GEOM works so I'll leave it for GEOM to
2407 * decide what to do.
2408 */
2409
2410 count = acr + acw + ace;
2411 if (count == 0)
2412 return (0);
2413
2414 flags = 0;
2415 if (acr != 0 || ace != 0)
2416 flags |= FREAD;
2417 if (acw != 0)
2418 flags |= FWRITE;
2419
2420 g_topology_unlock();
2421 if (count > 0)
2422 error = zvol_open(pp, flags, count);
2423 else
2424 error = zvol_close(pp, flags, -count);
2425 g_topology_lock();
2426 return (error);
2427}
2428
2429static void
2430zvol_geom_start(struct bio *bp)
2431{
2432 zvol_state_t *zv;
2433 boolean_t first;
2434
2435 zv = bp->bio_to->private;
2436 ASSERT(zv != NULL);
2437 switch (bp->bio_cmd) {
2438 case BIO_FLUSH:
2439 if (!THREAD_CAN_SLEEP())
2440 goto enqueue;
2441 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2442 g_io_deliver(bp, 0);
2443 break;
2444 case BIO_READ:
2445 case BIO_WRITE:
2446 case BIO_DELETE:
2447 if (!THREAD_CAN_SLEEP())
2448 goto enqueue;
2449 zvol_strategy(bp);
2450 break;
2451 case BIO_GETATTR:
2452 if (g_handleattr_int(bp, "GEOM::candelete", 1))
2453 return;
2454 /* FALLTHROUGH */
2455 default:
2456 g_io_deliver(bp, EOPNOTSUPP);
2457 break;
2458 }
2459 return;
2460
2461enqueue:
2462 mtx_lock(&zv->zv_queue_mtx);
2463 first = (bioq_first(&zv->zv_queue) == NULL);
2464 bioq_insert_tail(&zv->zv_queue, bp);
2465 mtx_unlock(&zv->zv_queue_mtx);
2466 if (first)
2467 wakeup_one(&zv->zv_queue);
2468}
2469
2470static void
2471zvol_geom_worker(void *arg)
2472{
2473 zvol_state_t *zv;
2474 struct bio *bp;
2475
2476 thread_lock(curthread);
2477 sched_prio(curthread, PRIBIO);
2478 thread_unlock(curthread);
2479
2480 zv = arg;
2481 for (;;) {
2482 mtx_lock(&zv->zv_queue_mtx);
2483 bp = bioq_takefirst(&zv->zv_queue);
2484 if (bp == NULL) {
2485 if (zv->zv_state == 1) {
2486 zv->zv_state = 2;
2487 wakeup(&zv->zv_state);
2488 mtx_unlock(&zv->zv_queue_mtx);
2489 kthread_exit();
2490 }
2491 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2492 "zvol:io", 0);
2493 continue;
2494 }
2495 mtx_unlock(&zv->zv_queue_mtx);
2496 switch (bp->bio_cmd) {
2497 case BIO_FLUSH:
2498 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2499 g_io_deliver(bp, 0);
2500 break;
2501 case BIO_READ:
2502 case BIO_WRITE:
2503 zvol_strategy(bp);
2504 break;
2505 }
2506 }
2507}
2508
2509extern boolean_t dataset_name_hidden(const char *name);
2510
2511static int
2512zvol_create_snapshots(objset_t *os, const char *name)
2513{
2514 uint64_t cookie, obj;
2515 char *sname;
2516 int error, len;
2517
2518 cookie = obj = 0;
2519 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2520
2521#if 0
2522 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2523 DS_FIND_SNAPSHOTS);
2524#endif
2525
2526 for (;;) {
2527 len = snprintf(sname, MAXPATHLEN, "%s@", name);
2528 if (len >= MAXPATHLEN) {
2529 dmu_objset_rele(os, FTAG);
2530 error = ENAMETOOLONG;
2531 break;
2532 }
2533
2534 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2535 error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2536 sname + len, &obj, &cookie, NULL);
2537 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2538 if (error != 0) {
2539 if (error == ENOENT)
2540 error = 0;
2541 break;
2542 }
2543
2544 if ((error = zvol_create_minor(sname)) != 0) {
2545 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2546 sname, error);
2547 break;
2548 }
2549 }
2550
2551 kmem_free(sname, MAXPATHLEN);
2552 return (error);
2553}
2554
2555int
2556zvol_create_minors(const char *name)
2557{
2558 uint64_t cookie;
2559 objset_t *os;
2560 char *osname, *p;
2561 int error, len;
2562
2563 if (dataset_name_hidden(name))
2564 return (0);
2565
2566 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2567 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2568 name, error);
2569 return (error);
2570 }
2571 if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2572 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2573 dsl_pool_rele(dmu_objset_pool(os), FTAG);
2574 if ((error = zvol_create_minor(name)) == 0)
2575 error = zvol_create_snapshots(os, name);
2576 else {
2577 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2578 name, error);
2579 }
2580 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2581 dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2582 return (error);
2583 }
2584 if (dmu_objset_type(os) != DMU_OST_ZFS) {
2585 dmu_objset_rele(os, FTAG);
2586 return (0);
2587 }
2588
2589 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2590 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2591 dmu_objset_rele(os, FTAG);
2592 kmem_free(osname, MAXPATHLEN);
2593 return (ENOENT);
2594 }
2595 p = osname + strlen(osname);
2596 len = MAXPATHLEN - (p - osname);
2597
2598#if 0
2599 /* Prefetch the datasets. */
2600 cookie = 0;
2601 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2602 if (!dataset_name_hidden(osname))
2603 (void) dmu_objset_prefetch(osname, NULL);
2604 }
2605#endif
2606
2607 cookie = 0;
2608 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2609 &cookie) == 0) {
2610 dmu_objset_rele(os, FTAG);
2611 (void)zvol_create_minors(osname);
2612 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2613 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2614 name, error);
2615 return (error);
2616 }
2617 }
2618
2619 dmu_objset_rele(os, FTAG);
2620 kmem_free(osname, MAXPATHLEN);
2621 return (0);
2622}
2623
2624static void
2625zvol_rename_minor(zvol_state_t *zv, const char *newname)
2626{
2627 struct g_geom *gp;
2628 struct g_provider *pp;
2629 struct cdev *dev;
2630
2631 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2632
2633 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
2634 g_topology_lock();
2635 pp = zv->zv_provider;
2636 ASSERT(pp != NULL);
2637 gp = pp->geom;
2638 ASSERT(gp != NULL);
2639
2640 zv->zv_provider = NULL;
2641 g_wither_provider(pp, ENXIO);
2642
2643 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
2644 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
2645 pp->sectorsize = DEV_BSIZE;
2646 pp->mediasize = zv->zv_volsize;
2647 pp->private = zv;
2648 zv->zv_provider = pp;
2649 g_error_provider(pp, 0);
2650 g_topology_unlock();
2651 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
2652 dev = zv->zv_dev;
2653 ASSERT(dev != NULL);
2654 zv->zv_dev = NULL;
2655 destroy_dev(dev);
2656
2657 if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
2658 &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR,
2659 0640, "%s/%s", ZVOL_DRIVER, newname) == 0) {
2660 zv->zv_dev = dev;
2661 dev->si_iosize_max = MAXPHYS;
2662 dev->si_drv2 = zv;
2663 }
2664 }
2665 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
2666}
2667
2668void
2669zvol_rename_minors(const char *oldname, const char *newname)
2670{
2671 char name[MAXPATHLEN];
2672 struct g_provider *pp;
2673 struct g_geom *gp;
2674 size_t oldnamelen, newnamelen;
2675 zvol_state_t *zv;
2676 char *namebuf;
2677
2678 oldnamelen = strlen(oldname);
2679 newnamelen = strlen(newname);
2680
2681 DROP_GIANT();
2682 mutex_enter(&spa_namespace_lock);
2683
2684 LIST_FOREACH(zv, &all_zvols, zv_links) {
2685 if (strcmp(zv->zv_name, oldname) == 0) {
2686 zvol_rename_minor(zv, newname);
2687 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
2688 (zv->zv_name[oldnamelen] == '/' ||
2689 zv->zv_name[oldnamelen] == '@')) {
2690 snprintf(name, sizeof(name), "%s%c%s", newname,
2691 zv->zv_name[oldnamelen],
2692 zv->zv_name + oldnamelen + 1);
2693 zvol_rename_minor(zv, name);
2694 }
2695 }
2696
2697 mutex_exit(&spa_namespace_lock);
2698 PICKUP_GIANT();
2699}
2700
2701static int
2702zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2703{
2704 zvol_state_t *zv;
2705 int err = 0;
2706
2707 mutex_enter(&spa_namespace_lock);
2708 zv = dev->si_drv2;
2709 if (zv == NULL) {
2710 mutex_exit(&spa_namespace_lock);
2711 return(ENXIO); /* zvol_create_minor() not done yet */
2712 }
2713
2714 if (zv->zv_total_opens == 0)
2715 err = zvol_first_open(zv);
2716 if (err) {
2717 mutex_exit(&spa_namespace_lock);
2718 return (err);
2719 }
2720 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
2721 err = SET_ERROR(EROFS);
2722 goto out;
2723 }
2724 if (zv->zv_flags & ZVOL_EXCL) {
2725 err = SET_ERROR(EBUSY);
2726 goto out;
2727 }
2728#ifdef FEXCL
2729 if (flags & FEXCL) {
2730 if (zv->zv_total_opens != 0) {
2731 err = SET_ERROR(EBUSY);
2732 goto out;
2733 }
2734 zv->zv_flags |= ZVOL_EXCL;
2735 }
2736#endif
2737
2738 zv->zv_total_opens++;
2739 mutex_exit(&spa_namespace_lock);
2740 return (err);
2741out:
2742 if (zv->zv_total_opens == 0)
2743 zvol_last_close(zv);
2744 mutex_exit(&spa_namespace_lock);
2745 return (err);
2746}
2747
2748static int
2749zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2750{
2751 zvol_state_t *zv;
2752 int err = 0;
2753
2754 mutex_enter(&spa_namespace_lock);
2755 zv = dev->si_drv2;
2756 if (zv == NULL) {
2757 mutex_exit(&spa_namespace_lock);
2758 return(ENXIO);
2759 }
2760
2761 if (zv->zv_flags & ZVOL_EXCL) {
2762 ASSERT(zv->zv_total_opens == 1);
2763 zv->zv_flags &= ~ZVOL_EXCL;
2764 }
2765
2766 /*
2767 * If the open count is zero, this is a spurious close.
2768 * That indicates a bug in the kernel / DDI framework.
2769 */
2770 ASSERT(zv->zv_total_opens != 0);
2771
2772 /*
2773 * You may get multiple opens, but only one close.
2774 */
2775 zv->zv_total_opens--;
2776
2777 if (zv->zv_total_opens == 0)
2778 zvol_last_close(zv);
2779
2780 mutex_exit(&spa_namespace_lock);
2781 return (0);
2782}
2783
2784static int
2785zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
2786{
2787 zvol_state_t *zv;
2788 rl_t *rl;
2789 off_t offset, length, chunk;
2790 int i, error;
2791 u_int u;
2792
2793 zv = dev->si_drv2;
2794
2795 error = 0;
2796 KASSERT(zv->zv_total_opens > 0,
2797 ("Device with zero access count in zvol_d_ioctl"));
2798
2799 i = IOCPARM_LEN(cmd);
2800 switch (cmd) {
2801 case DIOCGSECTORSIZE:
2802 *(u_int *)data = DEV_BSIZE;
2803 break;
2804 case DIOCGMEDIASIZE:
2805 *(off_t *)data = zv->zv_volsize;
2806 break;
2807 case DIOCGFLUSH:
2808 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2809 break;
2810 case DIOCGDELETE:
2811 offset = ((off_t *)data)[0];
2812 length = ((off_t *)data)[1];
2813 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
2814 offset < 0 || offset >= zv->zv_volsize ||
2815 length <= 0) {
2816 printf("%s: offset=%jd length=%jd\n", __func__, offset,
2817 length);
2818 error = EINVAL;
2819 break;
2820 }
2821
2822 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
2823 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
2824 error = dmu_tx_assign(tx, TXG_WAIT);
2825 if (error != 0) {
2826 dmu_tx_abort(tx);
2827 } else {
2828 zvol_log_truncate(zv, tx, offset, length, B_TRUE);
2829 dmu_tx_commit(tx);
2830 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
2831 offset, length);
2832 }
2833 zfs_range_unlock(rl);
2834 if (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)
2835 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2836 break;
2837 case DIOCGSTRIPESIZE:
2838 *(off_t *)data = zv->zv_volblocksize;
2839 break;
2840 case DIOCGSTRIPEOFFSET:
2841 *(off_t *)data = 0;
2842 break;
2843 default:
2844 error = ENOIOCTL;
2845 }
2846
2847 return (error);
2848}
|