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