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