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 2017 Nexenta Systems, Inc. All rights reserved. 27 * Copyright (c) 2012, 2017 by Delphix. All rights reserved. 28 * Copyright (c) 2014 Integros [integros.com] 29 * Copyright 2019 Joyent, Inc. 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 #include <sys/zil_impl.h> 92 #include <sys/smt.h> 93 #include <sys/dkioc_free_util.h> 94 #include <sys/zfs_rlock.h> 95 96 #include "zfs_namecheck.h" 97 98 void *zfsdev_state; 99 static char *zvol_tag = "zvol_tag"; 100 101 #define ZVOL_DUMPSIZE "dumpsize" 102 103 /* 104 * This lock protects the zfsdev_state structure from being modified 105 * while it's being used, e.g. an open that comes in before a create 106 * finishes. It also protects temporary opens of the dataset so that, 107 * e.g., an open doesn't get a spurious EBUSY. 108 */ 109 kmutex_t zfsdev_state_lock; 110 static uint32_t zvol_minors; 111 112 typedef struct zvol_extent { 113 list_node_t ze_node; 114 dva_t ze_dva; /* dva associated with this extent */ 115 uint64_t ze_nblks; /* number of blocks in extent */ 116 } zvol_extent_t; 117 118 /* 119 * The in-core state of each volume. 120 */ 121 typedef struct zvol_state { 122 char zv_name[MAXPATHLEN]; /* pool/dd name */ 123 uint64_t zv_volsize; /* amount of space we advertise */ 124 uint64_t zv_volblocksize; /* volume block size */ 125 minor_t zv_minor; /* minor number */ 126 uint8_t zv_min_bs; /* minimum addressable block shift */ 127 uint8_t zv_flags; /* readonly, dumpified, etc. */ 128 objset_t *zv_objset; /* objset handle */ 129 uint32_t zv_open_count[OTYPCNT]; /* open counts */ 130 uint32_t zv_total_opens; /* total open count */ 131 zilog_t *zv_zilog; /* ZIL handle */ 132 list_t zv_extents; /* List of extents for dump */ 133 rangelock_t zv_rangelock; 134 dnode_t *zv_dn; /* dnode hold */ 135 } zvol_state_t; 136 137 /* 138 * zvol specific flags 139 */ 140 #define ZVOL_RDONLY 0x1 141 #define ZVOL_DUMPIFIED 0x2 142 #define ZVOL_EXCL 0x4 143 #define ZVOL_WCE 0x8 144 145 /* 146 * zvol maximum transfer in one DMU tx. 147 */ 148 int zvol_maxphys = DMU_MAX_ACCESS/2; 149 150 /* 151 * Toggle unmap functionality. 152 */ 153 boolean_t zvol_unmap_enabled = B_TRUE; 154 155 /* 156 * If true, unmaps requested as synchronous are executed synchronously, 157 * otherwise all unmaps are asynchronous. 158 */ 159 boolean_t zvol_unmap_sync_enabled = B_FALSE; 160 161 extern int zfs_set_prop_nvlist(const char *, zprop_source_t, 162 nvlist_t *, nvlist_t *); 163 static int zvol_remove_zv(zvol_state_t *); 164 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, 165 struct lwb *lwb, zio_t *zio); 166 static int zvol_dumpify(zvol_state_t *zv); 167 static int zvol_dump_fini(zvol_state_t *zv); 168 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); 169 170 static void 171 zvol_size_changed(zvol_state_t *zv, uint64_t volsize) 172 { 173 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor); 174 175 zv->zv_volsize = volsize; 176 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 177 "Size", volsize) == DDI_SUCCESS); 178 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 179 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS); 180 181 /* Notify specfs to invalidate the cached size */ 182 spec_size_invalidate(dev, VBLK); 183 spec_size_invalidate(dev, VCHR); 184 } 185 186 int 187 zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 188 { 189 if (volsize == 0) 190 return (SET_ERROR(EINVAL)); 191 192 if (volsize % blocksize != 0) 193 return (SET_ERROR(EINVAL)); 194 195 #ifdef _ILP32 196 if (volsize - 1 > SPEC_MAXOFFSET_T) 197 return (SET_ERROR(EOVERFLOW)); 198 #endif 199 return (0); 200 } 201 202 int 203 zvol_check_volblocksize(uint64_t volblocksize) 204 { 205 if (volblocksize < SPA_MINBLOCKSIZE || 206 volblocksize > SPA_OLD_MAXBLOCKSIZE || 207 !ISP2(volblocksize)) 208 return (SET_ERROR(EDOM)); 209 210 return (0); 211 } 212 213 int 214 zvol_get_stats(objset_t *os, nvlist_t *nv) 215 { 216 int error; 217 dmu_object_info_t doi; 218 uint64_t val; 219 220 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 221 if (error) 222 return (error); 223 224 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 225 226 error = dmu_object_info(os, ZVOL_OBJ, &doi); 227 228 if (error == 0) { 229 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 230 doi.doi_data_block_size); 231 } 232 233 return (error); 234 } 235 236 static zvol_state_t * 237 zvol_minor_lookup(const char *name) 238 { 239 minor_t minor; 240 zvol_state_t *zv; 241 242 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 243 244 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) { 245 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 246 if (zv == NULL) 247 continue; 248 if (strcmp(zv->zv_name, name) == 0) 249 return (zv); 250 } 251 252 return (NULL); 253 } 254 255 /* extent mapping arg */ 256 struct maparg { 257 zvol_state_t *ma_zv; 258 uint64_t ma_blks; 259 }; 260 261 /*ARGSUSED*/ 262 static int 263 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 264 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 265 { 266 struct maparg *ma = arg; 267 zvol_extent_t *ze; 268 int bs = ma->ma_zv->zv_volblocksize; 269 270 if (bp == NULL || BP_IS_HOLE(bp) || 271 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 272 return (0); 273 274 VERIFY(!BP_IS_EMBEDDED(bp)); 275 276 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 277 ma->ma_blks++; 278 279 /* Abort immediately if we have encountered gang blocks */ 280 if (BP_IS_GANG(bp)) 281 return (SET_ERROR(EFRAGS)); 282 283 /* 284 * See if the block is at the end of the previous extent. 285 */ 286 ze = list_tail(&ma->ma_zv->zv_extents); 287 if (ze && 288 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 289 DVA_GET_OFFSET(BP_IDENTITY(bp)) == 290 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 291 ze->ze_nblks++; 292 return (0); 293 } 294 295 dprintf_bp(bp, "%s", "next blkptr:"); 296 297 /* start a new extent */ 298 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 299 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 300 ze->ze_nblks = 1; 301 list_insert_tail(&ma->ma_zv->zv_extents, ze); 302 return (0); 303 } 304 305 static void 306 zvol_free_extents(zvol_state_t *zv) 307 { 308 zvol_extent_t *ze; 309 310 while (ze = list_head(&zv->zv_extents)) { 311 list_remove(&zv->zv_extents, ze); 312 kmem_free(ze, sizeof (zvol_extent_t)); 313 } 314 } 315 316 static int 317 zvol_get_lbas(zvol_state_t *zv) 318 { 319 objset_t *os = zv->zv_objset; 320 struct maparg ma; 321 int err; 322 323 ma.ma_zv = zv; 324 ma.ma_blks = 0; 325 zvol_free_extents(zv); 326 327 /* commit any in-flight changes before traversing the dataset */ 328 txg_wait_synced(dmu_objset_pool(os), 0); 329 err = traverse_dataset(dmu_objset_ds(os), 0, 330 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 331 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 332 zvol_free_extents(zv); 333 return (err ? err : EIO); 334 } 335 336 return (0); 337 } 338 339 /* ARGSUSED */ 340 void 341 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 342 { 343 zfs_creat_t *zct = arg; 344 nvlist_t *nvprops = zct->zct_props; 345 int error; 346 uint64_t volblocksize, volsize; 347 348 VERIFY(nvlist_lookup_uint64(nvprops, 349 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 350 if (nvlist_lookup_uint64(nvprops, 351 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 352 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 353 354 /* 355 * These properties must be removed from the list so the generic 356 * property setting step won't apply to them. 357 */ 358 VERIFY(nvlist_remove_all(nvprops, 359 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 360 (void) nvlist_remove_all(nvprops, 361 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 362 363 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 364 DMU_OT_NONE, 0, tx); 365 ASSERT(error == 0); 366 367 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 368 DMU_OT_NONE, 0, tx); 369 ASSERT(error == 0); 370 371 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 372 ASSERT(error == 0); 373 } 374 375 /* 376 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we 377 * implement DKIOCFREE/free-long-range. 378 */ 379 static int 380 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap) 381 { 382 zvol_state_t *zv = arg1; 383 lr_truncate_t *lr = arg2; 384 uint64_t offset, length; 385 386 if (byteswap) 387 byteswap_uint64_array(lr, sizeof (*lr)); 388 389 offset = lr->lr_offset; 390 length = lr->lr_length; 391 392 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length)); 393 } 394 395 /* 396 * Replay a TX_WRITE ZIL transaction that didn't get committed 397 * after a system failure 398 */ 399 /* ARGSUSED */ 400 static int 401 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap) 402 { 403 zvol_state_t *zv = arg1; 404 lr_write_t *lr = arg2; 405 objset_t *os = zv->zv_objset; 406 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 407 uint64_t offset, length; 408 dmu_tx_t *tx; 409 int error; 410 411 if (byteswap) 412 byteswap_uint64_array(lr, sizeof (*lr)); 413 414 offset = lr->lr_offset; 415 length = lr->lr_length; 416 417 /* If it's a dmu_sync() block, write the whole block */ 418 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 419 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 420 if (length < blocksize) { 421 offset -= offset % blocksize; 422 length = blocksize; 423 } 424 } 425 426 tx = dmu_tx_create(os); 427 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 428 error = dmu_tx_assign(tx, TXG_WAIT); 429 if (error) { 430 dmu_tx_abort(tx); 431 } else { 432 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 433 dmu_tx_commit(tx); 434 } 435 436 return (error); 437 } 438 439 /* ARGSUSED */ 440 static int 441 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap) 442 { 443 return (SET_ERROR(ENOTSUP)); 444 } 445 446 /* 447 * Callback vectors for replaying records. 448 * Only TX_WRITE and TX_TRUNCATE are needed for zvol. 449 */ 450 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 451 zvol_replay_err, /* 0 no such transaction type */ 452 zvol_replay_err, /* TX_CREATE */ 453 zvol_replay_err, /* TX_MKDIR */ 454 zvol_replay_err, /* TX_MKXATTR */ 455 zvol_replay_err, /* TX_SYMLINK */ 456 zvol_replay_err, /* TX_REMOVE */ 457 zvol_replay_err, /* TX_RMDIR */ 458 zvol_replay_err, /* TX_LINK */ 459 zvol_replay_err, /* TX_RENAME */ 460 zvol_replay_write, /* TX_WRITE */ 461 zvol_replay_truncate, /* TX_TRUNCATE */ 462 zvol_replay_err, /* TX_SETATTR */ 463 zvol_replay_err, /* TX_ACL */ 464 zvol_replay_err, /* TX_CREATE_ACL */ 465 zvol_replay_err, /* TX_CREATE_ATTR */ 466 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 467 zvol_replay_err, /* TX_MKDIR_ACL */ 468 zvol_replay_err, /* TX_MKDIR_ATTR */ 469 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 470 zvol_replay_err, /* TX_WRITE2 */ 471 }; 472 473 int 474 zvol_name2minor(const char *name, minor_t *minor) 475 { 476 zvol_state_t *zv; 477 478 mutex_enter(&zfsdev_state_lock); 479 zv = zvol_minor_lookup(name); 480 if (minor && zv) 481 *minor = zv->zv_minor; 482 mutex_exit(&zfsdev_state_lock); 483 return (zv ? 0 : -1); 484 } 485 486 /* 487 * Create a minor node (plus a whole lot more) for the specified volume. 488 */ 489 int 490 zvol_create_minor(const char *name) 491 { 492 zfs_soft_state_t *zs; 493 zvol_state_t *zv; 494 objset_t *os; 495 dmu_object_info_t doi; 496 minor_t minor = 0; 497 char chrbuf[30], blkbuf[30]; 498 int error; 499 500 mutex_enter(&zfsdev_state_lock); 501 502 if (zvol_minor_lookup(name) != NULL) { 503 mutex_exit(&zfsdev_state_lock); 504 return (SET_ERROR(EEXIST)); 505 } 506 507 /* lie and say we're read-only */ 508 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, B_TRUE, FTAG, &os); 509 510 if (error) { 511 mutex_exit(&zfsdev_state_lock); 512 return (error); 513 } 514 515 if ((minor = zfsdev_minor_alloc()) == 0) { 516 dmu_objset_disown(os, 1, FTAG); 517 mutex_exit(&zfsdev_state_lock); 518 return (SET_ERROR(ENXIO)); 519 } 520 521 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 522 dmu_objset_disown(os, 1, FTAG); 523 mutex_exit(&zfsdev_state_lock); 524 return (SET_ERROR(EAGAIN)); 525 } 526 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 527 (char *)name); 528 529 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 530 531 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 532 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 533 ddi_soft_state_free(zfsdev_state, minor); 534 dmu_objset_disown(os, 1, FTAG); 535 mutex_exit(&zfsdev_state_lock); 536 return (SET_ERROR(EAGAIN)); 537 } 538 539 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 540 541 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 542 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 543 ddi_remove_minor_node(zfs_dip, chrbuf); 544 ddi_soft_state_free(zfsdev_state, minor); 545 dmu_objset_disown(os, 1, FTAG); 546 mutex_exit(&zfsdev_state_lock); 547 return (SET_ERROR(EAGAIN)); 548 } 549 550 zs = ddi_get_soft_state(zfsdev_state, minor); 551 zs->zss_type = ZSST_ZVOL; 552 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 553 (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 554 zv->zv_min_bs = DEV_BSHIFT; 555 zv->zv_minor = minor; 556 zv->zv_objset = os; 557 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) 558 zv->zv_flags |= ZVOL_RDONLY; 559 rangelock_init(&zv->zv_rangelock, NULL, NULL); 560 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 561 offsetof(zvol_extent_t, ze_node)); 562 /* get and cache the blocksize */ 563 error = dmu_object_info(os, ZVOL_OBJ, &doi); 564 ASSERT(error == 0); 565 zv->zv_volblocksize = doi.doi_data_block_size; 566 567 if (spa_writeable(dmu_objset_spa(os))) { 568 if (zil_replay_disable) 569 zil_destroy(dmu_objset_zil(os), B_FALSE); 570 else 571 zil_replay(os, zv, zvol_replay_vector); 572 } 573 dmu_objset_disown(os, 1, FTAG); 574 zv->zv_objset = NULL; 575 576 zvol_minors++; 577 578 mutex_exit(&zfsdev_state_lock); 579 580 return (0); 581 } 582 583 /* 584 * Remove minor node for the specified volume. 585 */ 586 static int 587 zvol_remove_zv(zvol_state_t *zv) 588 { 589 char nmbuf[20]; 590 minor_t minor = zv->zv_minor; 591 592 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 593 if (zv->zv_total_opens != 0) 594 return (SET_ERROR(EBUSY)); 595 596 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 597 ddi_remove_minor_node(zfs_dip, nmbuf); 598 599 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor); 600 ddi_remove_minor_node(zfs_dip, nmbuf); 601 602 rangelock_fini(&zv->zv_rangelock); 603 604 kmem_free(zv, sizeof (zvol_state_t)); 605 606 ddi_soft_state_free(zfsdev_state, minor); 607 608 zvol_minors--; 609 return (0); 610 } 611 612 int 613 zvol_remove_minor(const char *name) 614 { 615 zvol_state_t *zv; 616 int rc; 617 618 mutex_enter(&zfsdev_state_lock); 619 if ((zv = zvol_minor_lookup(name)) == NULL) { 620 mutex_exit(&zfsdev_state_lock); 621 return (SET_ERROR(ENXIO)); 622 } 623 rc = zvol_remove_zv(zv); 624 mutex_exit(&zfsdev_state_lock); 625 return (rc); 626 } 627 628 int 629 zvol_first_open(zvol_state_t *zv) 630 { 631 objset_t *os; 632 uint64_t volsize; 633 int error; 634 uint64_t readonly; 635 636 /* lie and say we're read-only */ 637 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, B_TRUE, 638 zvol_tag, &os); 639 if (error) 640 return (error); 641 642 zv->zv_objset = os; 643 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 644 if (error) { 645 ASSERT(error == 0); 646 dmu_objset_disown(os, 1, zvol_tag); 647 return (error); 648 } 649 650 error = dnode_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dn); 651 if (error) { 652 dmu_objset_disown(os, 1, zvol_tag); 653 return (error); 654 } 655 656 zvol_size_changed(zv, volsize); 657 zv->zv_zilog = zil_open(os, zvol_get_data); 658 659 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 660 NULL) == 0); 661 if (readonly || dmu_objset_is_snapshot(os) || 662 !spa_writeable(dmu_objset_spa(os))) 663 zv->zv_flags |= ZVOL_RDONLY; 664 else 665 zv->zv_flags &= ~ZVOL_RDONLY; 666 return (error); 667 } 668 669 void 670 zvol_last_close(zvol_state_t *zv) 671 { 672 zil_close(zv->zv_zilog); 673 zv->zv_zilog = NULL; 674 675 dnode_rele(zv->zv_dn, zvol_tag); 676 zv->zv_dn = NULL; 677 678 /* 679 * Evict cached data 680 */ 681 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && 682 !(zv->zv_flags & ZVOL_RDONLY)) 683 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 684 dmu_objset_evict_dbufs(zv->zv_objset); 685 686 dmu_objset_disown(zv->zv_objset, 1, zvol_tag); 687 zv->zv_objset = NULL; 688 } 689 690 int 691 zvol_prealloc(zvol_state_t *zv) 692 { 693 objset_t *os = zv->zv_objset; 694 dmu_tx_t *tx; 695 uint64_t refd, avail, usedobjs, availobjs; 696 uint64_t resid = zv->zv_volsize; 697 uint64_t off = 0; 698 699 /* Check the space usage before attempting to allocate the space */ 700 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 701 if (avail < zv->zv_volsize) 702 return (SET_ERROR(ENOSPC)); 703 704 /* Free old extents if they exist */ 705 zvol_free_extents(zv); 706 707 while (resid != 0) { 708 int error; 709 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE); 710 711 tx = dmu_tx_create(os); 712 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 713 error = dmu_tx_assign(tx, TXG_WAIT); 714 if (error) { 715 dmu_tx_abort(tx); 716 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 717 return (error); 718 } 719 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 720 dmu_tx_commit(tx); 721 off += bytes; 722 resid -= bytes; 723 } 724 txg_wait_synced(dmu_objset_pool(os), 0); 725 726 return (0); 727 } 728 729 static int 730 zvol_update_volsize(objset_t *os, uint64_t volsize) 731 { 732 dmu_tx_t *tx; 733 int error; 734 uint64_t txg; 735 736 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 737 738 tx = dmu_tx_create(os); 739 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 740 dmu_tx_mark_netfree(tx); 741 error = dmu_tx_assign(tx, TXG_WAIT); 742 if (error) { 743 dmu_tx_abort(tx); 744 return (error); 745 } 746 txg = dmu_tx_get_txg(tx); 747 748 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 749 &volsize, tx); 750 dmu_tx_commit(tx); 751 752 txg_wait_synced(dmu_objset_pool(os), txg); 753 754 if (error == 0) 755 error = dmu_free_long_range(os, 756 ZVOL_OBJ, volsize, DMU_OBJECT_END); 757 return (error); 758 } 759 760 void 761 zvol_remove_minors(const char *name) 762 { 763 zvol_state_t *zv; 764 char *namebuf; 765 minor_t minor; 766 767 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP); 768 (void) strncpy(namebuf, name, strlen(name)); 769 (void) strcat(namebuf, "/"); 770 mutex_enter(&zfsdev_state_lock); 771 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) { 772 773 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 774 if (zv == NULL) 775 continue; 776 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0) 777 (void) zvol_remove_zv(zv); 778 } 779 kmem_free(namebuf, strlen(name) + 2); 780 781 mutex_exit(&zfsdev_state_lock); 782 } 783 784 static int 785 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize) 786 { 787 uint64_t old_volsize = 0ULL; 788 int error = 0; 789 790 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 791 792 /* 793 * Reinitialize the dump area to the new size. If we 794 * failed to resize the dump area then restore it back to 795 * its original size. We must set the new volsize prior 796 * to calling dumpvp_resize() to ensure that the devices' 797 * size(9P) is not visible by the dump subsystem. 798 */ 799 old_volsize = zv->zv_volsize; 800 zvol_size_changed(zv, volsize); 801 802 if (zv->zv_flags & ZVOL_DUMPIFIED) { 803 if ((error = zvol_dumpify(zv)) != 0 || 804 (error = dumpvp_resize()) != 0) { 805 int dumpify_error; 806 807 (void) zvol_update_volsize(zv->zv_objset, old_volsize); 808 zvol_size_changed(zv, old_volsize); 809 dumpify_error = zvol_dumpify(zv); 810 error = dumpify_error ? dumpify_error : error; 811 } 812 } 813 814 /* 815 * Generate a LUN expansion event. 816 */ 817 if (error == 0) { 818 sysevent_id_t eid; 819 nvlist_t *attr; 820 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 821 822 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 823 zv->zv_minor); 824 825 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 826 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 827 828 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 829 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 830 831 nvlist_free(attr); 832 kmem_free(physpath, MAXPATHLEN); 833 } 834 return (error); 835 } 836 837 int 838 zvol_set_volsize(const char *name, uint64_t volsize) 839 { 840 zvol_state_t *zv = NULL; 841 objset_t *os; 842 int error; 843 dmu_object_info_t doi; 844 uint64_t readonly; 845 boolean_t owned = B_FALSE; 846 847 error = dsl_prop_get_integer(name, 848 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL); 849 if (error != 0) 850 return (error); 851 if (readonly) 852 return (SET_ERROR(EROFS)); 853 854 mutex_enter(&zfsdev_state_lock); 855 zv = zvol_minor_lookup(name); 856 857 if (zv == NULL || zv->zv_objset == NULL) { 858 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, B_TRUE, 859 FTAG, &os)) != 0) { 860 mutex_exit(&zfsdev_state_lock); 861 return (error); 862 } 863 owned = B_TRUE; 864 if (zv != NULL) 865 zv->zv_objset = os; 866 } else { 867 os = zv->zv_objset; 868 } 869 870 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 871 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0) 872 goto out; 873 874 error = zvol_update_volsize(os, volsize); 875 876 if (error == 0 && zv != NULL) 877 error = zvol_update_live_volsize(zv, volsize); 878 out: 879 if (owned) { 880 dmu_objset_disown(os, B_TRUE, FTAG); 881 if (zv != NULL) 882 zv->zv_objset = NULL; 883 } 884 mutex_exit(&zfsdev_state_lock); 885 return (error); 886 } 887 888 /*ARGSUSED*/ 889 int 890 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr) 891 { 892 zvol_state_t *zv; 893 int err = 0; 894 895 mutex_enter(&zfsdev_state_lock); 896 897 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL); 898 if (zv == NULL) { 899 mutex_exit(&zfsdev_state_lock); 900 return (SET_ERROR(ENXIO)); 901 } 902 903 if (zv->zv_total_opens == 0) 904 err = zvol_first_open(zv); 905 if (err) { 906 mutex_exit(&zfsdev_state_lock); 907 return (err); 908 } 909 /* 910 * Check for a bad on-disk format version now since we 911 * lied about owning the dataset readonly before. 912 */ 913 if ((flag & FWRITE) && ((zv->zv_flags & ZVOL_RDONLY) || 914 dmu_objset_incompatible_encryption_version(zv->zv_objset))) { 915 err = SET_ERROR(EROFS); 916 goto out; 917 } 918 if (zv->zv_flags & ZVOL_EXCL) { 919 err = SET_ERROR(EBUSY); 920 goto out; 921 } 922 if (flag & FEXCL) { 923 if (zv->zv_total_opens != 0) { 924 err = SET_ERROR(EBUSY); 925 goto out; 926 } 927 zv->zv_flags |= ZVOL_EXCL; 928 } 929 930 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) { 931 zv->zv_open_count[otyp]++; 932 zv->zv_total_opens++; 933 } 934 mutex_exit(&zfsdev_state_lock); 935 936 return (err); 937 out: 938 if (zv->zv_total_opens == 0) 939 zvol_last_close(zv); 940 mutex_exit(&zfsdev_state_lock); 941 return (err); 942 } 943 944 /*ARGSUSED*/ 945 int 946 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr) 947 { 948 minor_t minor = getminor(dev); 949 zvol_state_t *zv; 950 int error = 0; 951 952 mutex_enter(&zfsdev_state_lock); 953 954 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 955 if (zv == NULL) { 956 mutex_exit(&zfsdev_state_lock); 957 return (SET_ERROR(ENXIO)); 958 } 959 960 if (zv->zv_flags & ZVOL_EXCL) { 961 ASSERT(zv->zv_total_opens == 1); 962 zv->zv_flags &= ~ZVOL_EXCL; 963 } 964 965 /* 966 * If the open count is zero, this is a spurious close. 967 * That indicates a bug in the kernel / DDI framework. 968 */ 969 ASSERT(zv->zv_open_count[otyp] != 0); 970 ASSERT(zv->zv_total_opens != 0); 971 972 /* 973 * You may get multiple opens, but only one close. 974 */ 975 zv->zv_open_count[otyp]--; 976 zv->zv_total_opens--; 977 978 if (zv->zv_total_opens == 0) 979 zvol_last_close(zv); 980 981 mutex_exit(&zfsdev_state_lock); 982 return (error); 983 } 984 985 /* ARGSUSED */ 986 static void 987 zvol_get_done(zgd_t *zgd, int error) 988 { 989 if (zgd->zgd_db) 990 dmu_buf_rele(zgd->zgd_db, zgd); 991 992 rangelock_exit(zgd->zgd_lr); 993 994 kmem_free(zgd, sizeof (zgd_t)); 995 } 996 997 /* 998 * Get data to generate a TX_WRITE intent log record. 999 */ 1000 static int 1001 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio) 1002 { 1003 zvol_state_t *zv = arg; 1004 uint64_t offset = lr->lr_offset; 1005 uint64_t size = lr->lr_length; /* length of user data */ 1006 dmu_buf_t *db; 1007 zgd_t *zgd; 1008 int error; 1009 1010 ASSERT3P(lwb, !=, NULL); 1011 ASSERT3P(zio, !=, NULL); 1012 ASSERT3U(size, !=, 0); 1013 1014 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1015 zgd->zgd_lwb = lwb; 1016 1017 /* 1018 * Write records come in two flavors: immediate and indirect. 1019 * For small writes it's cheaper to store the data with the 1020 * log record (immediate); for large writes it's cheaper to 1021 * sync the data and get a pointer to it (indirect) so that 1022 * we don't have to write the data twice. 1023 */ 1024 if (buf != NULL) { /* immediate write */ 1025 zgd->zgd_lr = rangelock_enter(&zv->zv_rangelock, offset, size, 1026 RL_READER); 1027 error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf, 1028 DMU_READ_NO_PREFETCH); 1029 } else { /* indirect write */ 1030 /* 1031 * Have to lock the whole block to ensure when it's written out 1032 * and its checksum is being calculated that no one can change 1033 * the data. Contrarily to zfs_get_data we need not re-check 1034 * blocksize after we get the lock because it cannot be changed. 1035 */ 1036 size = zv->zv_volblocksize; 1037 offset = P2ALIGN(offset, size); 1038 zgd->zgd_lr = rangelock_enter(&zv->zv_rangelock, offset, size, 1039 RL_READER); 1040 error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db, 1041 DMU_READ_NO_PREFETCH); 1042 if (error == 0) { 1043 blkptr_t *bp = &lr->lr_blkptr; 1044 1045 zgd->zgd_db = db; 1046 zgd->zgd_bp = bp; 1047 1048 ASSERT(db->db_offset == offset); 1049 ASSERT(db->db_size == size); 1050 1051 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1052 zvol_get_done, zgd); 1053 1054 if (error == 0) 1055 return (0); 1056 } 1057 } 1058 1059 zvol_get_done(zgd, error); 1060 1061 return (error); 1062 } 1063 1064 /* 1065 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 1066 * 1067 * We store data in the log buffers if it's small enough. 1068 * Otherwise we will later flush the data out via dmu_sync(). 1069 */ 1070 ssize_t zvol_immediate_write_sz = 32768; 1071 1072 static void 1073 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1074 boolean_t sync) 1075 { 1076 uint32_t blocksize = zv->zv_volblocksize; 1077 zilog_t *zilog = zv->zv_zilog; 1078 itx_wr_state_t write_state; 1079 1080 if (zil_replaying(zilog, tx)) 1081 return; 1082 1083 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1084 write_state = WR_INDIRECT; 1085 else if (!spa_has_slogs(zilog->zl_spa) && 1086 resid >= blocksize && blocksize > zvol_immediate_write_sz) 1087 write_state = WR_INDIRECT; 1088 else if (sync) 1089 write_state = WR_COPIED; 1090 else 1091 write_state = WR_NEED_COPY; 1092 1093 while (resid) { 1094 itx_t *itx; 1095 lr_write_t *lr; 1096 itx_wr_state_t wr_state = write_state; 1097 ssize_t len = resid; 1098 1099 if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA) 1100 wr_state = WR_NEED_COPY; 1101 else if (wr_state == WR_INDIRECT) 1102 len = MIN(blocksize - P2PHASE(off, blocksize), resid); 1103 1104 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1105 (wr_state == WR_COPIED ? len : 0)); 1106 lr = (lr_write_t *)&itx->itx_lr; 1107 if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn, 1108 off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1109 zil_itx_destroy(itx); 1110 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1111 lr = (lr_write_t *)&itx->itx_lr; 1112 wr_state = WR_NEED_COPY; 1113 } 1114 1115 itx->itx_wr_state = wr_state; 1116 lr->lr_foid = ZVOL_OBJ; 1117 lr->lr_offset = off; 1118 lr->lr_length = len; 1119 lr->lr_blkoff = 0; 1120 BP_ZERO(&lr->lr_blkptr); 1121 1122 itx->itx_private = zv; 1123 itx->itx_sync = sync; 1124 1125 zil_itx_assign(zilog, itx, tx); 1126 1127 off += len; 1128 resid -= len; 1129 } 1130 } 1131 1132 static int 1133 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset, 1134 uint64_t size, boolean_t doread, boolean_t isdump) 1135 { 1136 vdev_disk_t *dvd; 1137 int c; 1138 int numerrors = 0; 1139 1140 if (vd->vdev_ops == &vdev_mirror_ops || 1141 vd->vdev_ops == &vdev_replacing_ops || 1142 vd->vdev_ops == &vdev_spare_ops) { 1143 for (c = 0; c < vd->vdev_children; c++) { 1144 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1145 addr, offset, origoffset, size, doread, isdump); 1146 if (err != 0) { 1147 numerrors++; 1148 } else if (doread) { 1149 break; 1150 } 1151 } 1152 } 1153 1154 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops) 1155 return (numerrors < vd->vdev_children ? 0 : EIO); 1156 1157 if (doread && !vdev_readable(vd)) 1158 return (SET_ERROR(EIO)); 1159 else if (!doread && !vdev_writeable(vd)) 1160 return (SET_ERROR(EIO)); 1161 1162 if (vd->vdev_ops == &vdev_raidz_ops) { 1163 return (vdev_raidz_physio(vd, 1164 addr, size, offset, origoffset, doread, isdump)); 1165 } 1166 1167 offset += VDEV_LABEL_START_SIZE; 1168 1169 if (ddi_in_panic() || isdump) { 1170 ASSERT(!doread); 1171 if (doread) 1172 return (SET_ERROR(EIO)); 1173 dvd = vd->vdev_tsd; 1174 ASSERT3P(dvd, !=, NULL); 1175 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1176 lbtodb(size))); 1177 } else { 1178 dvd = vd->vdev_tsd; 1179 ASSERT3P(dvd, !=, NULL); 1180 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size, 1181 offset, doread ? B_READ : B_WRITE)); 1182 } 1183 } 1184 1185 static int 1186 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1187 boolean_t doread, boolean_t isdump) 1188 { 1189 vdev_t *vd; 1190 int error; 1191 zvol_extent_t *ze; 1192 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1193 1194 /* Must be sector aligned, and not stradle a block boundary. */ 1195 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1196 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1197 return (SET_ERROR(EINVAL)); 1198 } 1199 ASSERT(size <= zv->zv_volblocksize); 1200 1201 /* Locate the extent this belongs to */ 1202 ze = list_head(&zv->zv_extents); 1203 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1204 offset -= ze->ze_nblks * zv->zv_volblocksize; 1205 ze = list_next(&zv->zv_extents, ze); 1206 } 1207 1208 if (ze == NULL) 1209 return (SET_ERROR(EINVAL)); 1210 1211 if (!ddi_in_panic()) 1212 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1213 1214 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1215 offset += DVA_GET_OFFSET(&ze->ze_dva); 1216 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva), 1217 size, doread, isdump); 1218 1219 if (!ddi_in_panic()) 1220 spa_config_exit(spa, SCL_STATE, FTAG); 1221 1222 return (error); 1223 } 1224 1225 int 1226 zvol_strategy(buf_t *bp) 1227 { 1228 zfs_soft_state_t *zs = NULL; 1229 zvol_state_t *zv; 1230 uint64_t off, volsize; 1231 size_t resid; 1232 char *addr; 1233 objset_t *os; 1234 int error = 0; 1235 boolean_t doread = bp->b_flags & B_READ; 1236 boolean_t is_dumpified; 1237 boolean_t sync; 1238 1239 if (getminor(bp->b_edev) == 0) { 1240 error = SET_ERROR(EINVAL); 1241 } else { 1242 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev)); 1243 if (zs == NULL) 1244 error = SET_ERROR(ENXIO); 1245 else if (zs->zss_type != ZSST_ZVOL) 1246 error = SET_ERROR(EINVAL); 1247 } 1248 1249 if (error) { 1250 bioerror(bp, error); 1251 biodone(bp); 1252 return (0); 1253 } 1254 1255 zv = zs->zss_data; 1256 1257 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) { 1258 bioerror(bp, EROFS); 1259 biodone(bp); 1260 return (0); 1261 } 1262 1263 off = ldbtob(bp->b_blkno); 1264 volsize = zv->zv_volsize; 1265 1266 os = zv->zv_objset; 1267 ASSERT(os != NULL); 1268 1269 bp_mapin(bp); 1270 addr = bp->b_un.b_addr; 1271 resid = bp->b_bcount; 1272 1273 if (resid > 0 && (off < 0 || off >= volsize)) { 1274 bioerror(bp, EIO); 1275 biodone(bp); 1276 return (0); 1277 } 1278 1279 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED; 1280 sync = ((!(bp->b_flags & B_ASYNC) && 1281 !(zv->zv_flags & ZVOL_WCE)) || 1282 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) && 1283 !doread && !is_dumpified; 1284 1285 smt_begin_unsafe(); 1286 1287 /* 1288 * There must be no buffer changes when doing a dmu_sync() because 1289 * we can't change the data whilst calculating the checksum. 1290 */ 1291 locked_range_t *lr = rangelock_enter(&zv->zv_rangelock, off, resid, 1292 doread ? RL_READER : RL_WRITER); 1293 1294 while (resid != 0 && off < volsize) { 1295 size_t size = MIN(resid, zvol_maxphys); 1296 if (is_dumpified) { 1297 size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 1298 error = zvol_dumpio(zv, addr, off, size, 1299 doread, B_FALSE); 1300 } else if (doread) { 1301 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1302 DMU_READ_PREFETCH); 1303 } else { 1304 dmu_tx_t *tx = dmu_tx_create(os); 1305 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1306 error = dmu_tx_assign(tx, TXG_WAIT); 1307 if (error) { 1308 dmu_tx_abort(tx); 1309 } else { 1310 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1311 zvol_log_write(zv, tx, off, size, sync); 1312 dmu_tx_commit(tx); 1313 } 1314 } 1315 if (error) { 1316 /* convert checksum errors into IO errors */ 1317 if (error == ECKSUM) 1318 error = SET_ERROR(EIO); 1319 break; 1320 } 1321 off += size; 1322 addr += size; 1323 resid -= size; 1324 } 1325 rangelock_exit(lr); 1326 1327 if ((bp->b_resid = resid) == bp->b_bcount) 1328 bioerror(bp, off > volsize ? EINVAL : error); 1329 1330 if (sync) 1331 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1332 biodone(bp); 1333 1334 smt_end_unsafe(); 1335 1336 return (0); 1337 } 1338 1339 /* 1340 * Set the buffer count to the zvol maximum transfer. 1341 * Using our own routine instead of the default minphys() 1342 * means that for larger writes we write bigger buffers on X86 1343 * (128K instead of 56K) and flush the disk write cache less often 1344 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1345 * 56K on X86 and 128K on sparc). 1346 */ 1347 void 1348 zvol_minphys(struct buf *bp) 1349 { 1350 if (bp->b_bcount > zvol_maxphys) 1351 bp->b_bcount = zvol_maxphys; 1352 } 1353 1354 int 1355 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1356 { 1357 minor_t minor = getminor(dev); 1358 zvol_state_t *zv; 1359 int error = 0; 1360 uint64_t size; 1361 uint64_t boff; 1362 uint64_t resid; 1363 1364 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1365 if (zv == NULL) 1366 return (SET_ERROR(ENXIO)); 1367 1368 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) 1369 return (SET_ERROR(EINVAL)); 1370 1371 boff = ldbtob(blkno); 1372 resid = ldbtob(nblocks); 1373 1374 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1375 1376 while (resid) { 1377 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1378 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1379 if (error) 1380 break; 1381 boff += size; 1382 addr += size; 1383 resid -= size; 1384 } 1385 1386 return (error); 1387 } 1388 1389 /*ARGSUSED*/ 1390 int 1391 zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1392 { 1393 minor_t minor = getminor(dev); 1394 zvol_state_t *zv; 1395 uint64_t volsize; 1396 int error = 0; 1397 1398 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1399 if (zv == NULL) 1400 return (SET_ERROR(ENXIO)); 1401 1402 volsize = zv->zv_volsize; 1403 if (uio->uio_resid > 0 && 1404 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1405 return (SET_ERROR(EIO)); 1406 1407 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1408 error = physio(zvol_strategy, NULL, dev, B_READ, 1409 zvol_minphys, uio); 1410 return (error); 1411 } 1412 1413 smt_begin_unsafe(); 1414 1415 locked_range_t *lr = rangelock_enter(&zv->zv_rangelock, 1416 uio->uio_loffset, uio->uio_resid, RL_READER); 1417 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1418 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1419 1420 /* don't read past the end */ 1421 if (bytes > volsize - uio->uio_loffset) 1422 bytes = volsize - uio->uio_loffset; 1423 1424 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1425 if (error) { 1426 /* convert checksum errors into IO errors */ 1427 if (error == ECKSUM) 1428 error = SET_ERROR(EIO); 1429 break; 1430 } 1431 } 1432 rangelock_exit(lr); 1433 1434 smt_end_unsafe(); 1435 1436 return (error); 1437 } 1438 1439 /*ARGSUSED*/ 1440 int 1441 zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1442 { 1443 minor_t minor = getminor(dev); 1444 zvol_state_t *zv; 1445 uint64_t volsize; 1446 int error = 0; 1447 boolean_t sync; 1448 1449 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1450 if (zv == NULL) 1451 return (SET_ERROR(ENXIO)); 1452 1453 volsize = zv->zv_volsize; 1454 if (uio->uio_resid > 0 && 1455 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1456 return (SET_ERROR(EIO)); 1457 1458 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1459 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1460 zvol_minphys, uio); 1461 return (error); 1462 } 1463 1464 smt_begin_unsafe(); 1465 1466 sync = !(zv->zv_flags & ZVOL_WCE) || 1467 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1468 1469 locked_range_t *lr = rangelock_enter(&zv->zv_rangelock, 1470 uio->uio_loffset, uio->uio_resid, RL_WRITER); 1471 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1472 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1473 uint64_t off = uio->uio_loffset; 1474 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1475 1476 if (bytes > volsize - off) /* don't write past the end */ 1477 bytes = volsize - off; 1478 1479 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1480 error = dmu_tx_assign(tx, TXG_WAIT); 1481 if (error) { 1482 dmu_tx_abort(tx); 1483 break; 1484 } 1485 error = dmu_write_uio_dnode(zv->zv_dn, uio, bytes, tx); 1486 if (error == 0) 1487 zvol_log_write(zv, tx, off, bytes, sync); 1488 dmu_tx_commit(tx); 1489 1490 if (error) 1491 break; 1492 } 1493 rangelock_exit(lr); 1494 1495 if (sync) 1496 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1497 1498 smt_end_unsafe(); 1499 1500 return (error); 1501 } 1502 1503 int 1504 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1505 { 1506 struct uuid uuid = EFI_RESERVED; 1507 efi_gpe_t gpe = { 0 }; 1508 uint32_t crc; 1509 dk_efi_t efi; 1510 int length; 1511 char *ptr; 1512 1513 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1514 return (SET_ERROR(EFAULT)); 1515 ptr = (char *)(uintptr_t)efi.dki_data_64; 1516 length = efi.dki_length; 1517 /* 1518 * Some clients may attempt to request a PMBR for the 1519 * zvol. Currently this interface will return EINVAL to 1520 * such requests. These requests could be supported by 1521 * adding a check for lba == 0 and consing up an appropriate 1522 * PMBR. 1523 */ 1524 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1525 return (SET_ERROR(EINVAL)); 1526 1527 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1528 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1529 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1530 1531 if (efi.dki_lba == 1) { 1532 efi_gpt_t gpt = { 0 }; 1533 1534 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1535 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1536 gpt.efi_gpt_HeaderSize = LE_32(EFI_HEADER_SIZE); 1537 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1538 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1539 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1540 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1541 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1542 gpt.efi_gpt_SizeOfPartitionEntry = 1543 LE_32(sizeof (efi_gpe_t)); 1544 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1545 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1546 CRC32(crc, &gpt, EFI_HEADER_SIZE, -1U, crc32_table); 1547 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1548 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1549 flag)) 1550 return (SET_ERROR(EFAULT)); 1551 ptr += sizeof (gpt); 1552 length -= sizeof (gpt); 1553 } 1554 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1555 length), flag)) 1556 return (SET_ERROR(EFAULT)); 1557 return (0); 1558 } 1559 1560 /* 1561 * BEGIN entry points to allow external callers access to the volume. 1562 */ 1563 /* 1564 * Return the volume parameters needed for access from an external caller. 1565 * These values are invariant as long as the volume is held open. 1566 */ 1567 int 1568 zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1569 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1570 void **rl_hdl, void **dnode_hdl) 1571 { 1572 zvol_state_t *zv; 1573 1574 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1575 if (zv == NULL) 1576 return (SET_ERROR(ENXIO)); 1577 if (zv->zv_flags & ZVOL_DUMPIFIED) 1578 return (SET_ERROR(ENXIO)); 1579 1580 ASSERT(blksize && max_xfer_len && minor_hdl && 1581 objset_hdl && zil_hdl && rl_hdl && dnode_hdl); 1582 1583 *blksize = zv->zv_volblocksize; 1584 *max_xfer_len = (uint64_t)zvol_maxphys; 1585 *minor_hdl = zv; 1586 *objset_hdl = zv->zv_objset; 1587 *zil_hdl = zv->zv_zilog; 1588 *rl_hdl = &zv->zv_rangelock; 1589 *dnode_hdl = zv->zv_dn; 1590 return (0); 1591 } 1592 1593 /* 1594 * Return the current volume size to an external caller. 1595 * The size can change while the volume is open. 1596 */ 1597 uint64_t 1598 zvol_get_volume_size(void *minor_hdl) 1599 { 1600 zvol_state_t *zv = minor_hdl; 1601 1602 return (zv->zv_volsize); 1603 } 1604 1605 /* 1606 * Return the current WCE setting to an external caller. 1607 * The WCE setting can change while the volume is open. 1608 */ 1609 int 1610 zvol_get_volume_wce(void *minor_hdl) 1611 { 1612 zvol_state_t *zv = minor_hdl; 1613 1614 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 1615 } 1616 1617 /* 1618 * Entry point for external callers to zvol_log_write 1619 */ 1620 void 1621 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 1622 boolean_t sync) 1623 { 1624 zvol_state_t *zv = minor_hdl; 1625 1626 zvol_log_write(zv, tx, off, resid, sync); 1627 } 1628 /* 1629 * END entry points to allow external callers access to the volume. 1630 */ 1631 1632 /* 1633 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. 1634 */ 1635 static void 1636 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, 1637 boolean_t sync) 1638 { 1639 itx_t *itx; 1640 lr_truncate_t *lr; 1641 zilog_t *zilog = zv->zv_zilog; 1642 1643 if (zil_replaying(zilog, tx)) 1644 return; 1645 1646 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 1647 lr = (lr_truncate_t *)&itx->itx_lr; 1648 lr->lr_foid = ZVOL_OBJ; 1649 lr->lr_offset = off; 1650 lr->lr_length = len; 1651 1652 itx->itx_sync = sync; 1653 zil_itx_assign(zilog, itx, tx); 1654 } 1655 1656 /* 1657 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1658 * Also a dirtbag dkio ioctl for unmap/free-block functionality. 1659 */ 1660 /*ARGSUSED*/ 1661 int 1662 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1663 { 1664 zvol_state_t *zv; 1665 struct dk_callback *dkc; 1666 int error = 0; 1667 locked_range_t *lr; 1668 1669 mutex_enter(&zfsdev_state_lock); 1670 1671 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 1672 1673 if (zv == NULL) { 1674 mutex_exit(&zfsdev_state_lock); 1675 return (SET_ERROR(ENXIO)); 1676 } 1677 ASSERT(zv->zv_total_opens > 0); 1678 1679 switch (cmd) { 1680 1681 case DKIOCINFO: 1682 { 1683 struct dk_cinfo dki; 1684 1685 bzero(&dki, sizeof (dki)); 1686 (void) strcpy(dki.dki_cname, "zvol"); 1687 (void) strcpy(dki.dki_dname, "zvol"); 1688 dki.dki_ctype = DKC_UNKNOWN; 1689 dki.dki_unit = getminor(dev); 1690 dki.dki_maxtransfer = 1691 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs); 1692 mutex_exit(&zfsdev_state_lock); 1693 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1694 error = SET_ERROR(EFAULT); 1695 return (error); 1696 } 1697 1698 case DKIOCGMEDIAINFO: 1699 { 1700 struct dk_minfo dkm; 1701 1702 bzero(&dkm, sizeof (dkm)); 1703 dkm.dki_lbsize = 1U << zv->zv_min_bs; 1704 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1705 dkm.dki_media_type = DK_UNKNOWN; 1706 mutex_exit(&zfsdev_state_lock); 1707 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1708 error = SET_ERROR(EFAULT); 1709 return (error); 1710 } 1711 1712 case DKIOCGMEDIAINFOEXT: 1713 { 1714 struct dk_minfo_ext dkmext; 1715 1716 bzero(&dkmext, sizeof (dkmext)); 1717 dkmext.dki_lbsize = 1U << zv->zv_min_bs; 1718 dkmext.dki_pbsize = zv->zv_volblocksize; 1719 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1720 dkmext.dki_media_type = DK_UNKNOWN; 1721 mutex_exit(&zfsdev_state_lock); 1722 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag)) 1723 error = SET_ERROR(EFAULT); 1724 return (error); 1725 } 1726 1727 case DKIOCGETEFI: 1728 { 1729 uint64_t vs = zv->zv_volsize; 1730 uint8_t bs = zv->zv_min_bs; 1731 1732 mutex_exit(&zfsdev_state_lock); 1733 error = zvol_getefi((void *)arg, flag, vs, bs); 1734 return (error); 1735 } 1736 1737 case DKIOCFLUSHWRITECACHE: 1738 dkc = (struct dk_callback *)arg; 1739 mutex_exit(&zfsdev_state_lock); 1740 1741 smt_begin_unsafe(); 1742 1743 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1744 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1745 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1746 error = 0; 1747 } 1748 1749 smt_end_unsafe(); 1750 1751 return (error); 1752 1753 case DKIOCGETWCE: 1754 { 1755 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1756 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1757 flag)) 1758 error = SET_ERROR(EFAULT); 1759 break; 1760 } 1761 case DKIOCSETWCE: 1762 { 1763 int wce; 1764 if (ddi_copyin((void *)arg, &wce, sizeof (int), 1765 flag)) { 1766 error = SET_ERROR(EFAULT); 1767 break; 1768 } 1769 if (wce) { 1770 zv->zv_flags |= ZVOL_WCE; 1771 mutex_exit(&zfsdev_state_lock); 1772 } else { 1773 zv->zv_flags &= ~ZVOL_WCE; 1774 mutex_exit(&zfsdev_state_lock); 1775 smt_begin_unsafe(); 1776 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1777 smt_end_unsafe(); 1778 } 1779 return (0); 1780 } 1781 1782 case DKIOCGGEOM: 1783 case DKIOCGVTOC: 1784 /* 1785 * commands using these (like prtvtoc) expect ENOTSUP 1786 * since we're emulating an EFI label 1787 */ 1788 error = SET_ERROR(ENOTSUP); 1789 break; 1790 1791 case DKIOCDUMPINIT: 1792 lr = rangelock_enter(&zv->zv_rangelock, 0, zv->zv_volsize, 1793 RL_WRITER); 1794 error = zvol_dumpify(zv); 1795 rangelock_exit(lr); 1796 break; 1797 1798 case DKIOCDUMPFINI: 1799 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 1800 break; 1801 lr = rangelock_enter(&zv->zv_rangelock, 0, zv->zv_volsize, 1802 RL_WRITER); 1803 error = zvol_dump_fini(zv); 1804 rangelock_exit(lr); 1805 break; 1806 1807 case DKIOCFREE: 1808 { 1809 dkioc_free_list_t *dfl; 1810 dmu_tx_t *tx; 1811 1812 if (!zvol_unmap_enabled) 1813 break; 1814 1815 if (!(flag & FKIOCTL)) { 1816 error = dfl_copyin((void *)arg, &dfl, flag, KM_SLEEP); 1817 if (error != 0) 1818 break; 1819 } else { 1820 dfl = (dkioc_free_list_t *)arg; 1821 ASSERT3U(dfl->dfl_num_exts, <=, DFL_COPYIN_MAX_EXTS); 1822 if (dfl->dfl_num_exts > DFL_COPYIN_MAX_EXTS) { 1823 error = SET_ERROR(EINVAL); 1824 break; 1825 } 1826 } 1827 1828 mutex_exit(&zfsdev_state_lock); 1829 1830 smt_begin_unsafe(); 1831 1832 for (int i = 0; i < dfl->dfl_num_exts; i++) { 1833 uint64_t start = dfl->dfl_exts[i].dfle_start, 1834 length = dfl->dfl_exts[i].dfle_length, 1835 end = start + length; 1836 1837 /* 1838 * Apply Postel's Law to length-checking. If they 1839 * overshoot, just blank out until the end, if there's 1840 * a need to blank out anything. 1841 */ 1842 if (start >= zv->zv_volsize) 1843 continue; /* No need to do anything... */ 1844 if (end > zv->zv_volsize) { 1845 end = DMU_OBJECT_END; 1846 length = end - start; 1847 } 1848 1849 lr = rangelock_enter(&zv->zv_rangelock, start, length, 1850 RL_WRITER); 1851 tx = dmu_tx_create(zv->zv_objset); 1852 error = dmu_tx_assign(tx, TXG_WAIT); 1853 if (error != 0) { 1854 dmu_tx_abort(tx); 1855 } else { 1856 zvol_log_truncate(zv, tx, start, length, 1857 B_TRUE); 1858 dmu_tx_commit(tx); 1859 error = dmu_free_long_range(zv->zv_objset, 1860 ZVOL_OBJ, start, length); 1861 } 1862 1863 rangelock_exit(lr); 1864 1865 if (error != 0) 1866 break; 1867 } 1868 1869 /* 1870 * If the write-cache is disabled, 'sync' property 1871 * is set to 'always', or if the caller is asking for 1872 * a synchronous free, commit this operation to the zil. 1873 * This will sync any previous uncommitted writes to the 1874 * zvol object. 1875 * Can be overridden by the zvol_unmap_sync_enabled tunable. 1876 */ 1877 if ((error == 0) && zvol_unmap_sync_enabled && 1878 (!(zv->zv_flags & ZVOL_WCE) || 1879 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) || 1880 (dfl->dfl_flags & DF_WAIT_SYNC))) { 1881 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1882 } 1883 1884 if (!(flag & FKIOCTL)) 1885 dfl_free(dfl); 1886 1887 smt_end_unsafe(); 1888 1889 return (error); 1890 } 1891 1892 default: 1893 error = SET_ERROR(ENOTTY); 1894 break; 1895 1896 } 1897 mutex_exit(&zfsdev_state_lock); 1898 return (error); 1899 } 1900 1901 int 1902 zvol_busy(void) 1903 { 1904 return (zvol_minors != 0); 1905 } 1906 1907 void 1908 zvol_init(void) 1909 { 1910 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 1911 1) == 0); 1912 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL); 1913 } 1914 1915 void 1916 zvol_fini(void) 1917 { 1918 mutex_destroy(&zfsdev_state_lock); 1919 ddi_soft_state_fini(&zfsdev_state); 1920 } 1921 1922 /*ARGSUSED*/ 1923 static int 1924 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx) 1925 { 1926 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1927 1928 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 1929 return (1); 1930 return (0); 1931 } 1932 1933 /*ARGSUSED*/ 1934 static void 1935 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx) 1936 { 1937 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1938 1939 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx); 1940 } 1941 1942 static int 1943 zvol_dump_init(zvol_state_t *zv, boolean_t resize) 1944 { 1945 dmu_tx_t *tx; 1946 int error; 1947 objset_t *os = zv->zv_objset; 1948 spa_t *spa = dmu_objset_spa(os); 1949 vdev_t *vd = spa->spa_root_vdev; 1950 nvlist_t *nv = NULL; 1951 uint64_t version = spa_version(spa); 1952 uint64_t checksum, compress, refresrv, vbs, dedup; 1953 1954 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 1955 ASSERT(vd->vdev_ops == &vdev_root_ops); 1956 1957 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 1958 DMU_OBJECT_END); 1959 if (error != 0) 1960 return (error); 1961 /* wait for dmu_free_long_range to actually free the blocks */ 1962 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1963 1964 /* 1965 * If the pool on which the dump device is being initialized has more 1966 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is 1967 * enabled. If so, bump that feature's counter to indicate that the 1968 * feature is active. We also check the vdev type to handle the 1969 * following case: 1970 * # zpool create test raidz disk1 disk2 disk3 1971 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev), 1972 * the raidz vdev itself has 3 children. 1973 */ 1974 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) { 1975 if (!spa_feature_is_enabled(spa, 1976 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 1977 return (SET_ERROR(ENOTSUP)); 1978 (void) dsl_sync_task(spa_name(spa), 1979 zfs_mvdev_dump_feature_check, 1980 zfs_mvdev_dump_activate_feature_sync, NULL, 1981 2, ZFS_SPACE_CHECK_RESERVED); 1982 } 1983 1984 if (!resize) { 1985 error = dsl_prop_get_integer(zv->zv_name, 1986 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 1987 if (error == 0) { 1988 error = dsl_prop_get_integer(zv->zv_name, 1989 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, 1990 NULL); 1991 } 1992 if (error == 0) { 1993 error = dsl_prop_get_integer(zv->zv_name, 1994 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 1995 &refresrv, NULL); 1996 } 1997 if (error == 0) { 1998 error = dsl_prop_get_integer(zv->zv_name, 1999 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, 2000 NULL); 2001 } 2002 if (version >= SPA_VERSION_DEDUP && error == 0) { 2003 error = dsl_prop_get_integer(zv->zv_name, 2004 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 2005 } 2006 } 2007 if (error != 0) 2008 return (error); 2009 2010 tx = dmu_tx_create(os); 2011 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2012 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2013 error = dmu_tx_assign(tx, TXG_WAIT); 2014 if (error != 0) { 2015 dmu_tx_abort(tx); 2016 return (error); 2017 } 2018 2019 /* 2020 * If we are resizing the dump device then we only need to 2021 * update the refreservation to match the newly updated 2022 * zvolsize. Otherwise, we save off the original state of the 2023 * zvol so that we can restore them if the zvol is ever undumpified. 2024 */ 2025 if (resize) { 2026 error = zap_update(os, ZVOL_ZAP_OBJ, 2027 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2028 &zv->zv_volsize, tx); 2029 } else { 2030 error = zap_update(os, ZVOL_ZAP_OBJ, 2031 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 2032 &compress, tx); 2033 if (error == 0) { 2034 error = zap_update(os, ZVOL_ZAP_OBJ, 2035 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, 2036 &checksum, tx); 2037 } 2038 if (error == 0) { 2039 error = zap_update(os, ZVOL_ZAP_OBJ, 2040 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2041 &refresrv, tx); 2042 } 2043 if (error == 0) { 2044 error = zap_update(os, ZVOL_ZAP_OBJ, 2045 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 2046 &vbs, tx); 2047 } 2048 if (error == 0) { 2049 error = dmu_object_set_blocksize( 2050 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx); 2051 } 2052 if (version >= SPA_VERSION_DEDUP && error == 0) { 2053 error = zap_update(os, ZVOL_ZAP_OBJ, 2054 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 2055 &dedup, tx); 2056 } 2057 if (error == 0) 2058 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE; 2059 } 2060 dmu_tx_commit(tx); 2061 2062 /* 2063 * We only need update the zvol's property if we are initializing 2064 * the dump area for the first time. 2065 */ 2066 if (error == 0 && !resize) { 2067 /* 2068 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum 2069 * function. Otherwise, use the old default -- OFF. 2070 */ 2071 checksum = spa_feature_is_active(spa, 2072 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY : 2073 ZIO_CHECKSUM_OFF; 2074 2075 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2076 VERIFY(nvlist_add_uint64(nv, 2077 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 2078 VERIFY(nvlist_add_uint64(nv, 2079 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 2080 ZIO_COMPRESS_OFF) == 0); 2081 VERIFY(nvlist_add_uint64(nv, 2082 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 2083 checksum) == 0); 2084 if (version >= SPA_VERSION_DEDUP) { 2085 VERIFY(nvlist_add_uint64(nv, 2086 zfs_prop_to_name(ZFS_PROP_DEDUP), 2087 ZIO_CHECKSUM_OFF) == 0); 2088 } 2089 2090 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2091 nv, NULL); 2092 nvlist_free(nv); 2093 } 2094 2095 /* Allocate the space for the dump */ 2096 if (error == 0) 2097 error = zvol_prealloc(zv); 2098 return (error); 2099 } 2100 2101 static int 2102 zvol_dumpify(zvol_state_t *zv) 2103 { 2104 int error = 0; 2105 uint64_t dumpsize = 0; 2106 dmu_tx_t *tx; 2107 objset_t *os = zv->zv_objset; 2108 2109 if (zv->zv_flags & ZVOL_RDONLY) 2110 return (SET_ERROR(EROFS)); 2111 2112 if (os->os_encrypted) 2113 return (SET_ERROR(ENOTSUP)); 2114 2115 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 2116 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 2117 boolean_t resize = (dumpsize > 0); 2118 2119 if ((error = zvol_dump_init(zv, resize)) != 0) { 2120 (void) zvol_dump_fini(zv); 2121 return (error); 2122 } 2123 } 2124 2125 /* 2126 * Build up our lba mapping. 2127 */ 2128 error = zvol_get_lbas(zv); 2129 if (error) { 2130 (void) zvol_dump_fini(zv); 2131 return (error); 2132 } 2133 2134 tx = dmu_tx_create(os); 2135 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2136 error = dmu_tx_assign(tx, TXG_WAIT); 2137 if (error) { 2138 dmu_tx_abort(tx); 2139 (void) zvol_dump_fini(zv); 2140 return (error); 2141 } 2142 2143 zv->zv_flags |= ZVOL_DUMPIFIED; 2144 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 2145 &zv->zv_volsize, tx); 2146 dmu_tx_commit(tx); 2147 2148 if (error) { 2149 (void) zvol_dump_fini(zv); 2150 return (error); 2151 } 2152 2153 txg_wait_synced(dmu_objset_pool(os), 0); 2154 return (0); 2155 } 2156 2157 static int 2158 zvol_dump_fini(zvol_state_t *zv) 2159 { 2160 dmu_tx_t *tx; 2161 objset_t *os = zv->zv_objset; 2162 nvlist_t *nv; 2163 int error = 0; 2164 uint64_t checksum, compress, refresrv, vbs, dedup; 2165 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 2166 2167 /* 2168 * Attempt to restore the zvol back to its pre-dumpified state. 2169 * This is a best-effort attempt as it's possible that not all 2170 * of these properties were initialized during the dumpify process 2171 * (i.e. error during zvol_dump_init). 2172 */ 2173 2174 tx = dmu_tx_create(os); 2175 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2176 error = dmu_tx_assign(tx, TXG_WAIT); 2177 if (error) { 2178 dmu_tx_abort(tx); 2179 return (error); 2180 } 2181 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 2182 dmu_tx_commit(tx); 2183 2184 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2185 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 2186 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2187 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 2188 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2189 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 2190 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2191 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 2192 2193 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2194 (void) nvlist_add_uint64(nv, 2195 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 2196 (void) nvlist_add_uint64(nv, 2197 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 2198 (void) nvlist_add_uint64(nv, 2199 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 2200 if (version >= SPA_VERSION_DEDUP && 2201 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2202 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 2203 (void) nvlist_add_uint64(nv, 2204 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 2205 } 2206 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2207 nv, NULL); 2208 nvlist_free(nv); 2209 2210 zvol_free_extents(zv); 2211 zv->zv_flags &= ~ZVOL_DUMPIFIED; 2212 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 2213 /* wait for dmu_free_long_range to actually free the blocks */ 2214 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2215 tx = dmu_tx_create(os); 2216 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2217 error = dmu_tx_assign(tx, TXG_WAIT); 2218 if (error) { 2219 dmu_tx_abort(tx); 2220 return (error); 2221 } 2222 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 2223 zv->zv_volblocksize = vbs; 2224 dmu_tx_commit(tx); 2225 2226 return (0); 2227 } 2228