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