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 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* 27 * ZFS volume emulation driver. 28 * 29 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 30 * Volumes are accessed through the symbolic links named: 31 * 32 * /dev/zvol/dsk/<pool_name>/<dataset_name> 33 * /dev/zvol/rdsk/<pool_name>/<dataset_name> 34 * 35 * These links are created by the ZFS-specific devfsadm link generator. 36 * Volumes are persistent through reboot. No user command needs to be 37 * run before opening and using a device. 38 */ 39 40 #include <sys/types.h> 41 #include <sys/param.h> 42 #include <sys/errno.h> 43 #include <sys/uio.h> 44 #include <sys/buf.h> 45 #include <sys/modctl.h> 46 #include <sys/open.h> 47 #include <sys/kmem.h> 48 #include <sys/conf.h> 49 #include <sys/cmn_err.h> 50 #include <sys/stat.h> 51 #include <sys/zap.h> 52 #include <sys/spa.h> 53 #include <sys/zio.h> 54 #include <sys/dmu_traverse.h> 55 #include <sys/dnode.h> 56 #include <sys/dsl_dataset.h> 57 #include <sys/dsl_prop.h> 58 #include <sys/dkio.h> 59 #include <sys/efi_partition.h> 60 #include <sys/byteorder.h> 61 #include <sys/pathname.h> 62 #include <sys/ddi.h> 63 #include <sys/sunddi.h> 64 #include <sys/crc32.h> 65 #include <sys/dirent.h> 66 #include <sys/policy.h> 67 #include <sys/fs/zfs.h> 68 #include <sys/zfs_ioctl.h> 69 #include <sys/mkdev.h> 70 #include <sys/zil.h> 71 #include <sys/refcount.h> 72 #include <sys/zfs_znode.h> 73 #include <sys/zfs_rlock.h> 74 #include <sys/vdev_disk.h> 75 #include <sys/vdev_impl.h> 76 #include <sys/zvol.h> 77 #include <sys/dumphdr.h> 78 #include <sys/zil_impl.h> 79 80 #include "zfs_namecheck.h" 81 82 static void *zvol_state; 83 84 #define ZVOL_DUMPSIZE "dumpsize" 85 86 /* 87 * This lock protects the zvol_state structure from being modified 88 * while it's being used, e.g. an open that comes in before a create 89 * finishes. It also protects temporary opens of the dataset so that, 90 * e.g., an open doesn't get a spurious EBUSY. 91 */ 92 static kmutex_t zvol_state_lock; 93 static uint32_t zvol_minors; 94 95 typedef struct zvol_extent { 96 list_node_t ze_node; 97 dva_t ze_dva; /* dva associated with this extent */ 98 uint64_t ze_nblks; /* number of blocks in extent */ 99 } zvol_extent_t; 100 101 /* 102 * The in-core state of each volume. 103 */ 104 typedef struct zvol_state { 105 char zv_name[MAXPATHLEN]; /* pool/dd name */ 106 uint64_t zv_volsize; /* amount of space we advertise */ 107 uint64_t zv_volblocksize; /* volume block size */ 108 minor_t zv_minor; /* minor number */ 109 uint8_t zv_min_bs; /* minimum addressable block shift */ 110 uint8_t zv_flags; /* readonly, dumpified, etc. */ 111 objset_t *zv_objset; /* objset handle */ 112 uint32_t zv_mode; /* DS_MODE_* flags at open time */ 113 uint32_t zv_open_count[OTYPCNT]; /* open counts */ 114 uint32_t zv_total_opens; /* total open count */ 115 zilog_t *zv_zilog; /* ZIL handle */ 116 list_t zv_extents; /* List of extents for dump */ 117 znode_t zv_znode; /* for range locking */ 118 } zvol_state_t; 119 120 /* 121 * zvol specific flags 122 */ 123 #define ZVOL_RDONLY 0x1 124 #define ZVOL_DUMPIFIED 0x2 125 #define ZVOL_EXCL 0x4 126 #define ZVOL_WCE 0x8 127 128 /* 129 * zvol maximum transfer in one DMU tx. 130 */ 131 int zvol_maxphys = DMU_MAX_ACCESS/2; 132 133 extern int zfs_set_prop_nvlist(const char *, nvlist_t *); 134 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio); 135 static int zvol_dumpify(zvol_state_t *zv); 136 static int zvol_dump_fini(zvol_state_t *zv); 137 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); 138 139 static void 140 zvol_size_changed(zvol_state_t *zv, major_t maj) 141 { 142 dev_t dev = makedevice(maj, zv->zv_minor); 143 144 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 145 "Size", zv->zv_volsize) == DDI_SUCCESS); 146 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 147 "Nblocks", lbtodb(zv->zv_volsize)) == DDI_SUCCESS); 148 149 /* Notify specfs to invalidate the cached size */ 150 spec_size_invalidate(dev, VBLK); 151 spec_size_invalidate(dev, VCHR); 152 } 153 154 int 155 zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 156 { 157 if (volsize == 0) 158 return (EINVAL); 159 160 if (volsize % blocksize != 0) 161 return (EINVAL); 162 163 #ifdef _ILP32 164 if (volsize - 1 > SPEC_MAXOFFSET_T) 165 return (EOVERFLOW); 166 #endif 167 return (0); 168 } 169 170 int 171 zvol_check_volblocksize(uint64_t volblocksize) 172 { 173 if (volblocksize < SPA_MINBLOCKSIZE || 174 volblocksize > SPA_MAXBLOCKSIZE || 175 !ISP2(volblocksize)) 176 return (EDOM); 177 178 return (0); 179 } 180 181 static void 182 zvol_readonly_changed_cb(void *arg, uint64_t newval) 183 { 184 zvol_state_t *zv = arg; 185 186 if (newval) 187 zv->zv_flags |= ZVOL_RDONLY; 188 else 189 zv->zv_flags &= ~ZVOL_RDONLY; 190 } 191 192 int 193 zvol_get_stats(objset_t *os, nvlist_t *nv) 194 { 195 int error; 196 dmu_object_info_t doi; 197 uint64_t val; 198 199 200 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 201 if (error) 202 return (error); 203 204 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 205 206 error = dmu_object_info(os, ZVOL_OBJ, &doi); 207 208 if (error == 0) { 209 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 210 doi.doi_data_block_size); 211 } 212 213 return (error); 214 } 215 216 /* 217 * Find a free minor number. 218 */ 219 static minor_t 220 zvol_minor_alloc(void) 221 { 222 minor_t minor; 223 224 ASSERT(MUTEX_HELD(&zvol_state_lock)); 225 226 for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++) 227 if (ddi_get_soft_state(zvol_state, minor) == NULL) 228 return (minor); 229 230 return (0); 231 } 232 233 static zvol_state_t * 234 zvol_minor_lookup(const char *name) 235 { 236 minor_t minor; 237 zvol_state_t *zv; 238 239 ASSERT(MUTEX_HELD(&zvol_state_lock)); 240 241 for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++) { 242 zv = ddi_get_soft_state(zvol_state, minor); 243 if (zv == NULL) 244 continue; 245 if (strcmp(zv->zv_name, name) == 0) 246 break; 247 } 248 249 return (zv); 250 } 251 252 /* extent mapping arg */ 253 struct maparg { 254 zvol_state_t *ma_zv; 255 uint64_t ma_blks; 256 }; 257 258 /*ARGSUSED*/ 259 static int 260 zvol_map_block(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb, 261 const dnode_phys_t *dnp, void *arg) 262 { 263 struct maparg *ma = arg; 264 zvol_extent_t *ze; 265 int bs = ma->ma_zv->zv_volblocksize; 266 267 if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 268 return (0); 269 270 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 271 ma->ma_blks++; 272 273 /* Abort immediately if we have encountered gang blocks */ 274 if (BP_IS_GANG(bp)) 275 return (EFRAGS); 276 277 /* 278 * See if the block is at the end of the previous extent. 279 */ 280 ze = list_tail(&ma->ma_zv->zv_extents); 281 if (ze && 282 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 283 DVA_GET_OFFSET(BP_IDENTITY(bp)) == 284 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 285 ze->ze_nblks++; 286 return (0); 287 } 288 289 dprintf_bp(bp, "%s", "next blkptr:"); 290 291 /* start a new extent */ 292 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 293 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 294 ze->ze_nblks = 1; 295 list_insert_tail(&ma->ma_zv->zv_extents, ze); 296 return (0); 297 } 298 299 static void 300 zvol_free_extents(zvol_state_t *zv) 301 { 302 zvol_extent_t *ze; 303 304 while (ze = list_head(&zv->zv_extents)) { 305 list_remove(&zv->zv_extents, ze); 306 kmem_free(ze, sizeof (zvol_extent_t)); 307 } 308 } 309 310 static int 311 zvol_get_lbas(zvol_state_t *zv) 312 { 313 struct maparg ma; 314 int err; 315 316 ma.ma_zv = zv; 317 ma.ma_blks = 0; 318 zvol_free_extents(zv); 319 320 err = traverse_dataset(dmu_objset_ds(zv->zv_objset), 0, 321 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 322 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 323 zvol_free_extents(zv); 324 return (err ? err : EIO); 325 } 326 327 return (0); 328 } 329 330 /* ARGSUSED */ 331 void 332 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 333 { 334 zfs_creat_t *zct = arg; 335 nvlist_t *nvprops = zct->zct_props; 336 int error; 337 uint64_t volblocksize, volsize; 338 339 VERIFY(nvlist_lookup_uint64(nvprops, 340 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 341 if (nvlist_lookup_uint64(nvprops, 342 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 343 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 344 345 /* 346 * These properties must be removed from the list so the generic 347 * property setting step won't apply to them. 348 */ 349 VERIFY(nvlist_remove_all(nvprops, 350 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 351 (void) nvlist_remove_all(nvprops, 352 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 353 354 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 355 DMU_OT_NONE, 0, tx); 356 ASSERT(error == 0); 357 358 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 359 DMU_OT_NONE, 0, tx); 360 ASSERT(error == 0); 361 362 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 363 ASSERT(error == 0); 364 } 365 366 /* 367 * Replay a TX_WRITE ZIL transaction that didn't get committed 368 * after a system failure 369 */ 370 static int 371 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 372 { 373 objset_t *os = zv->zv_objset; 374 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 375 uint64_t off = lr->lr_offset; 376 uint64_t len = lr->lr_length; 377 dmu_tx_t *tx; 378 int error; 379 380 if (byteswap) 381 byteswap_uint64_array(lr, sizeof (*lr)); 382 383 tx = dmu_tx_create(os); 384 dmu_tx_hold_write(tx, ZVOL_OBJ, off, len); 385 error = dmu_tx_assign(tx, TXG_WAIT); 386 if (error) { 387 dmu_tx_abort(tx); 388 } else { 389 dmu_write(os, ZVOL_OBJ, off, len, data, tx); 390 dmu_tx_commit(tx); 391 } 392 393 return (error); 394 } 395 396 /* ARGSUSED */ 397 static int 398 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 399 { 400 return (ENOTSUP); 401 } 402 403 /* 404 * Callback vectors for replaying records. 405 * Only TX_WRITE is needed for zvol. 406 */ 407 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 408 zvol_replay_err, /* 0 no such transaction type */ 409 zvol_replay_err, /* TX_CREATE */ 410 zvol_replay_err, /* TX_MKDIR */ 411 zvol_replay_err, /* TX_MKXATTR */ 412 zvol_replay_err, /* TX_SYMLINK */ 413 zvol_replay_err, /* TX_REMOVE */ 414 zvol_replay_err, /* TX_RMDIR */ 415 zvol_replay_err, /* TX_LINK */ 416 zvol_replay_err, /* TX_RENAME */ 417 zvol_replay_write, /* TX_WRITE */ 418 zvol_replay_err, /* TX_TRUNCATE */ 419 zvol_replay_err, /* TX_SETATTR */ 420 zvol_replay_err, /* TX_ACL */ 421 }; 422 423 /* 424 * Create a minor node (plus a whole lot more) for the specified volume. 425 */ 426 int 427 zvol_create_minor(const char *name, major_t maj) 428 { 429 zvol_state_t *zv; 430 objset_t *os; 431 dmu_object_info_t doi; 432 uint64_t volsize; 433 minor_t minor = 0; 434 struct pathname linkpath; 435 int ds_mode = DS_MODE_OWNER; 436 vnode_t *vp = NULL; 437 char *devpath; 438 size_t devpathlen = strlen(ZVOL_FULL_DEV_DIR) + strlen(name) + 1; 439 char chrbuf[30], blkbuf[30]; 440 int error; 441 442 mutex_enter(&zvol_state_lock); 443 444 if ((zv = zvol_minor_lookup(name)) != NULL) { 445 mutex_exit(&zvol_state_lock); 446 return (EEXIST); 447 } 448 449 if (strchr(name, '@') != 0) 450 ds_mode |= DS_MODE_READONLY; 451 452 error = dmu_objset_open(name, DMU_OST_ZVOL, ds_mode, &os); 453 454 if (error) { 455 mutex_exit(&zvol_state_lock); 456 return (error); 457 } 458 459 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 460 461 if (error) { 462 dmu_objset_close(os); 463 mutex_exit(&zvol_state_lock); 464 return (error); 465 } 466 467 /* 468 * If there's an existing /dev/zvol symlink, try to use the 469 * same minor number we used last time. 470 */ 471 devpath = kmem_alloc(devpathlen, KM_SLEEP); 472 473 (void) sprintf(devpath, "%s%s", ZVOL_FULL_DEV_DIR, name); 474 475 error = lookupname(devpath, UIO_SYSSPACE, NO_FOLLOW, NULL, &vp); 476 477 kmem_free(devpath, devpathlen); 478 479 if (error == 0 && vp->v_type != VLNK) 480 error = EINVAL; 481 482 if (error == 0) { 483 pn_alloc(&linkpath); 484 error = pn_getsymlink(vp, &linkpath, kcred); 485 if (error == 0) { 486 char *ms = strstr(linkpath.pn_path, ZVOL_PSEUDO_DEV); 487 if (ms != NULL) { 488 ms += strlen(ZVOL_PSEUDO_DEV); 489 minor = stoi(&ms); 490 } 491 } 492 pn_free(&linkpath); 493 } 494 495 if (vp != NULL) 496 VN_RELE(vp); 497 498 /* 499 * If we found a minor but it's already in use, we must pick a new one. 500 */ 501 if (minor != 0 && ddi_get_soft_state(zvol_state, minor) != NULL) 502 minor = 0; 503 504 if (minor == 0) 505 minor = zvol_minor_alloc(); 506 507 if (minor == 0) { 508 dmu_objset_close(os); 509 mutex_exit(&zvol_state_lock); 510 return (ENXIO); 511 } 512 513 if (ddi_soft_state_zalloc(zvol_state, minor) != DDI_SUCCESS) { 514 dmu_objset_close(os); 515 mutex_exit(&zvol_state_lock); 516 return (EAGAIN); 517 } 518 519 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 520 (char *)name); 521 522 (void) sprintf(chrbuf, "%uc,raw", minor); 523 524 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 525 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 526 ddi_soft_state_free(zvol_state, minor); 527 dmu_objset_close(os); 528 mutex_exit(&zvol_state_lock); 529 return (EAGAIN); 530 } 531 532 (void) sprintf(blkbuf, "%uc", minor); 533 534 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 535 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 536 ddi_remove_minor_node(zfs_dip, chrbuf); 537 ddi_soft_state_free(zvol_state, minor); 538 dmu_objset_close(os); 539 mutex_exit(&zvol_state_lock); 540 return (EAGAIN); 541 } 542 543 zv = ddi_get_soft_state(zvol_state, minor); 544 545 (void) strcpy(zv->zv_name, name); 546 zv->zv_min_bs = DEV_BSHIFT; 547 zv->zv_minor = minor; 548 zv->zv_volsize = volsize; 549 zv->zv_objset = os; 550 zv->zv_mode = ds_mode; 551 zv->zv_zilog = zil_open(os, zvol_get_data); 552 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 553 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 554 sizeof (rl_t), offsetof(rl_t, r_node)); 555 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 556 offsetof(zvol_extent_t, ze_node)); 557 /* get and cache the blocksize */ 558 error = dmu_object_info(os, ZVOL_OBJ, &doi); 559 ASSERT(error == 0); 560 zv->zv_volblocksize = doi.doi_data_block_size; 561 562 zil_replay(os, zv, zvol_replay_vector); 563 zvol_size_changed(zv, maj); 564 565 /* XXX this should handle the possible i/o error */ 566 VERIFY(dsl_prop_register(dmu_objset_ds(zv->zv_objset), 567 "readonly", zvol_readonly_changed_cb, zv) == 0); 568 569 zvol_minors++; 570 571 mutex_exit(&zvol_state_lock); 572 573 return (0); 574 } 575 576 /* 577 * Remove minor node for the specified volume. 578 */ 579 int 580 zvol_remove_minor(const char *name) 581 { 582 zvol_state_t *zv; 583 char namebuf[30]; 584 585 mutex_enter(&zvol_state_lock); 586 587 if ((zv = zvol_minor_lookup(name)) == NULL) { 588 mutex_exit(&zvol_state_lock); 589 return (ENXIO); 590 } 591 592 if (zv->zv_total_opens != 0) { 593 mutex_exit(&zvol_state_lock); 594 return (EBUSY); 595 } 596 597 (void) sprintf(namebuf, "%uc,raw", zv->zv_minor); 598 ddi_remove_minor_node(zfs_dip, namebuf); 599 600 (void) sprintf(namebuf, "%uc", zv->zv_minor); 601 ddi_remove_minor_node(zfs_dip, namebuf); 602 603 VERIFY(dsl_prop_unregister(dmu_objset_ds(zv->zv_objset), 604 "readonly", zvol_readonly_changed_cb, zv) == 0); 605 606 zil_close(zv->zv_zilog); 607 zv->zv_zilog = NULL; 608 dmu_objset_close(zv->zv_objset); 609 zv->zv_objset = NULL; 610 avl_destroy(&zv->zv_znode.z_range_avl); 611 mutex_destroy(&zv->zv_znode.z_range_lock); 612 613 ddi_soft_state_free(zvol_state, zv->zv_minor); 614 615 zvol_minors--; 616 617 mutex_exit(&zvol_state_lock); 618 619 return (0); 620 } 621 622 int 623 zvol_prealloc(zvol_state_t *zv) 624 { 625 objset_t *os = zv->zv_objset; 626 dmu_tx_t *tx; 627 uint64_t refd, avail, usedobjs, availobjs; 628 uint64_t resid = zv->zv_volsize; 629 uint64_t off = 0; 630 631 /* Check the space usage before attempting to allocate the space */ 632 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 633 if (avail < zv->zv_volsize) 634 return (ENOSPC); 635 636 /* Free old extents if they exist */ 637 zvol_free_extents(zv); 638 639 while (resid != 0) { 640 int error; 641 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); 642 643 tx = dmu_tx_create(os); 644 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 645 error = dmu_tx_assign(tx, TXG_WAIT); 646 if (error) { 647 dmu_tx_abort(tx); 648 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 649 return (error); 650 } 651 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 652 dmu_tx_commit(tx); 653 off += bytes; 654 resid -= bytes; 655 } 656 txg_wait_synced(dmu_objset_pool(os), 0); 657 658 return (0); 659 } 660 661 int 662 zvol_update_volsize(zvol_state_t *zv, major_t maj, uint64_t volsize) 663 { 664 dmu_tx_t *tx; 665 int error; 666 667 ASSERT(MUTEX_HELD(&zvol_state_lock)); 668 669 tx = dmu_tx_create(zv->zv_objset); 670 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 671 error = dmu_tx_assign(tx, TXG_WAIT); 672 if (error) { 673 dmu_tx_abort(tx); 674 return (error); 675 } 676 677 error = zap_update(zv->zv_objset, ZVOL_ZAP_OBJ, "size", 8, 1, 678 &volsize, tx); 679 dmu_tx_commit(tx); 680 681 if (error == 0) 682 error = dmu_free_long_range(zv->zv_objset, 683 ZVOL_OBJ, volsize, DMU_OBJECT_END); 684 685 /* 686 * If we are using a faked-up state (zv_minor == 0) then don't 687 * try to update the in-core zvol state. 688 */ 689 if (error == 0 && zv->zv_minor) { 690 zv->zv_volsize = volsize; 691 zvol_size_changed(zv, maj); 692 } 693 return (error); 694 } 695 696 int 697 zvol_set_volsize(const char *name, major_t maj, uint64_t volsize) 698 { 699 zvol_state_t *zv; 700 int error; 701 dmu_object_info_t doi; 702 uint64_t old_volsize = 0ULL; 703 zvol_state_t state = { 0 }; 704 705 mutex_enter(&zvol_state_lock); 706 707 if ((zv = zvol_minor_lookup(name)) == NULL) { 708 /* 709 * If we are doing a "zfs clone -o volsize=", then the 710 * minor node won't exist yet. 711 */ 712 error = dmu_objset_open(name, DMU_OST_ZVOL, DS_MODE_OWNER, 713 &state.zv_objset); 714 if (error != 0) 715 goto out; 716 zv = &state; 717 } 718 old_volsize = zv->zv_volsize; 719 720 if ((error = dmu_object_info(zv->zv_objset, ZVOL_OBJ, &doi)) != 0 || 721 (error = zvol_check_volsize(volsize, 722 doi.doi_data_block_size)) != 0) 723 goto out; 724 725 if (zv->zv_flags & ZVOL_RDONLY || (zv->zv_mode & DS_MODE_READONLY)) { 726 error = EROFS; 727 goto out; 728 } 729 730 error = zvol_update_volsize(zv, maj, volsize); 731 732 /* 733 * Reinitialize the dump area to the new size. If we 734 * failed to resize the dump area then restore the it back to 735 * it's original size. 736 */ 737 if (error == 0 && zv->zv_flags & ZVOL_DUMPIFIED) { 738 if ((error = zvol_dumpify(zv)) != 0 || 739 (error = dumpvp_resize()) != 0) { 740 (void) zvol_update_volsize(zv, maj, old_volsize); 741 error = zvol_dumpify(zv); 742 } 743 } 744 745 /* 746 * Generate a LUN expansion event. 747 */ 748 if (error == 0) { 749 sysevent_id_t eid; 750 nvlist_t *attr; 751 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 752 753 (void) snprintf(physpath, MAXPATHLEN, "%s%uc", ZVOL_PSEUDO_DEV, 754 zv->zv_minor); 755 756 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 757 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 758 759 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 760 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 761 762 nvlist_free(attr); 763 kmem_free(physpath, MAXPATHLEN); 764 } 765 766 out: 767 if (state.zv_objset) 768 dmu_objset_close(state.zv_objset); 769 770 mutex_exit(&zvol_state_lock); 771 772 return (error); 773 } 774 775 int 776 zvol_set_volblocksize(const char *name, uint64_t volblocksize) 777 { 778 zvol_state_t *zv; 779 dmu_tx_t *tx; 780 int error; 781 boolean_t needlock; 782 783 /* 784 * The lock may already be held if we are being called from 785 * zvol_dump_init(). 786 */ 787 needlock = !MUTEX_HELD(&zvol_state_lock); 788 if (needlock) 789 mutex_enter(&zvol_state_lock); 790 791 if ((zv = zvol_minor_lookup(name)) == NULL) { 792 if (needlock) 793 mutex_exit(&zvol_state_lock); 794 return (ENXIO); 795 } 796 if (zv->zv_flags & ZVOL_RDONLY || (zv->zv_mode & DS_MODE_READONLY)) { 797 if (needlock) 798 mutex_exit(&zvol_state_lock); 799 return (EROFS); 800 } 801 802 tx = dmu_tx_create(zv->zv_objset); 803 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 804 error = dmu_tx_assign(tx, TXG_WAIT); 805 if (error) { 806 dmu_tx_abort(tx); 807 } else { 808 error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ, 809 volblocksize, 0, tx); 810 if (error == ENOTSUP) 811 error = EBUSY; 812 dmu_tx_commit(tx); 813 if (error == 0) 814 zv->zv_volblocksize = volblocksize; 815 } 816 817 if (needlock) 818 mutex_exit(&zvol_state_lock); 819 820 return (error); 821 } 822 823 /*ARGSUSED*/ 824 int 825 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr) 826 { 827 minor_t minor = getminor(*devp); 828 zvol_state_t *zv; 829 830 if (minor == 0) /* This is the control device */ 831 return (0); 832 833 mutex_enter(&zvol_state_lock); 834 835 zv = ddi_get_soft_state(zvol_state, minor); 836 if (zv == NULL) { 837 mutex_exit(&zvol_state_lock); 838 return (ENXIO); 839 } 840 841 ASSERT(zv->zv_objset != NULL); 842 843 if ((flag & FWRITE) && 844 (zv->zv_flags & ZVOL_RDONLY || (zv->zv_mode & DS_MODE_READONLY))) { 845 mutex_exit(&zvol_state_lock); 846 return (EROFS); 847 } 848 if (zv->zv_flags & ZVOL_EXCL) { 849 mutex_exit(&zvol_state_lock); 850 return (EBUSY); 851 } 852 if (flag & FEXCL) { 853 if (zv->zv_total_opens != 0) { 854 mutex_exit(&zvol_state_lock); 855 return (EBUSY); 856 } 857 zv->zv_flags |= ZVOL_EXCL; 858 } 859 860 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) { 861 zv->zv_open_count[otyp]++; 862 zv->zv_total_opens++; 863 } 864 865 mutex_exit(&zvol_state_lock); 866 867 return (0); 868 } 869 870 /*ARGSUSED*/ 871 int 872 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr) 873 { 874 minor_t minor = getminor(dev); 875 zvol_state_t *zv; 876 877 if (minor == 0) /* This is the control device */ 878 return (0); 879 880 mutex_enter(&zvol_state_lock); 881 882 zv = ddi_get_soft_state(zvol_state, minor); 883 if (zv == NULL) { 884 mutex_exit(&zvol_state_lock); 885 return (ENXIO); 886 } 887 888 if (zv->zv_flags & ZVOL_EXCL) { 889 ASSERT(zv->zv_total_opens == 1); 890 zv->zv_flags &= ~ZVOL_EXCL; 891 } 892 893 /* 894 * If the open count is zero, this is a spurious close. 895 * That indicates a bug in the kernel / DDI framework. 896 */ 897 ASSERT(zv->zv_open_count[otyp] != 0); 898 ASSERT(zv->zv_total_opens != 0); 899 900 /* 901 * You may get multiple opens, but only one close. 902 */ 903 zv->zv_open_count[otyp]--; 904 zv->zv_total_opens--; 905 906 mutex_exit(&zvol_state_lock); 907 908 return (0); 909 } 910 911 static void 912 zvol_get_done(dmu_buf_t *db, void *vzgd) 913 { 914 zgd_t *zgd = (zgd_t *)vzgd; 915 rl_t *rl = zgd->zgd_rl; 916 917 dmu_buf_rele(db, vzgd); 918 zfs_range_unlock(rl); 919 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 920 kmem_free(zgd, sizeof (zgd_t)); 921 } 922 923 /* 924 * Get data to generate a TX_WRITE intent log record. 925 */ 926 static int 927 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 928 { 929 zvol_state_t *zv = arg; 930 objset_t *os = zv->zv_objset; 931 dmu_buf_t *db; 932 rl_t *rl; 933 zgd_t *zgd; 934 uint64_t boff; /* block starting offset */ 935 int dlen = lr->lr_length; /* length of user data */ 936 int error; 937 938 ASSERT(zio); 939 ASSERT(dlen != 0); 940 941 /* 942 * Write records come in two flavors: immediate and indirect. 943 * For small writes it's cheaper to store the data with the 944 * log record (immediate); for large writes it's cheaper to 945 * sync the data and get a pointer to it (indirect) so that 946 * we don't have to write the data twice. 947 */ 948 if (buf != NULL) /* immediate write */ 949 return (dmu_read(os, ZVOL_OBJ, lr->lr_offset, dlen, buf, 950 DMU_READ_NO_PREFETCH)); 951 952 zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP); 953 zgd->zgd_zilog = zv->zv_zilog; 954 zgd->zgd_bp = &lr->lr_blkptr; 955 956 /* 957 * Lock the range of the block to ensure that when the data is 958 * written out and its checksum is being calculated that no other 959 * thread can change the block. 960 */ 961 boff = P2ALIGN_TYPED(lr->lr_offset, zv->zv_volblocksize, uint64_t); 962 rl = zfs_range_lock(&zv->zv_znode, boff, zv->zv_volblocksize, 963 RL_READER); 964 zgd->zgd_rl = rl; 965 966 VERIFY(0 == dmu_buf_hold(os, ZVOL_OBJ, lr->lr_offset, zgd, &db)); 967 error = dmu_sync(zio, db, &lr->lr_blkptr, 968 lr->lr_common.lrc_txg, zvol_get_done, zgd); 969 if (error == 0) 970 zil_add_block(zv->zv_zilog, &lr->lr_blkptr); 971 /* 972 * If we get EINPROGRESS, then we need to wait for a 973 * write IO initiated by dmu_sync() to complete before 974 * we can release this dbuf. We will finish everything 975 * up in the zvol_get_done() callback. 976 */ 977 if (error == EINPROGRESS) 978 return (0); 979 dmu_buf_rele(db, zgd); 980 zfs_range_unlock(rl); 981 kmem_free(zgd, sizeof (zgd_t)); 982 return (error); 983 } 984 985 /* 986 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 987 * 988 * We store data in the log buffers if it's small enough. 989 * Otherwise we will later flush the data out via dmu_sync(). 990 */ 991 ssize_t zvol_immediate_write_sz = 32768; 992 993 static void 994 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 995 boolean_t sync) 996 { 997 uint32_t blocksize = zv->zv_volblocksize; 998 zilog_t *zilog = zv->zv_zilog; 999 boolean_t slogging; 1000 1001 if (zil_disable) 1002 return; 1003 1004 if (zilog->zl_replay) { 1005 dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); 1006 zilog->zl_replayed_seq[dmu_tx_get_txg(tx) & TXG_MASK] = 1007 zilog->zl_replaying_seq; 1008 return; 1009 } 1010 1011 slogging = spa_has_slogs(zilog->zl_spa); 1012 1013 while (resid) { 1014 itx_t *itx; 1015 lr_write_t *lr; 1016 ssize_t len; 1017 itx_wr_state_t write_state; 1018 1019 /* 1020 * Unlike zfs_log_write() we can be called with 1021 * upto DMU_MAX_ACCESS/2 (5MB) writes. 1022 */ 1023 if (blocksize > zvol_immediate_write_sz && !slogging && 1024 resid >= blocksize && off % blocksize == 0) { 1025 write_state = WR_INDIRECT; /* uses dmu_sync */ 1026 len = blocksize; 1027 } else if (sync) { 1028 write_state = WR_COPIED; 1029 len = MIN(ZIL_MAX_LOG_DATA, resid); 1030 } else { 1031 write_state = WR_NEED_COPY; 1032 len = MIN(ZIL_MAX_LOG_DATA, resid); 1033 } 1034 1035 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1036 (write_state == WR_COPIED ? len : 0)); 1037 lr = (lr_write_t *)&itx->itx_lr; 1038 if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 1039 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1040 kmem_free(itx, offsetof(itx_t, itx_lr) + 1041 itx->itx_lr.lrc_reclen); 1042 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1043 lr = (lr_write_t *)&itx->itx_lr; 1044 write_state = WR_NEED_COPY; 1045 } 1046 1047 itx->itx_wr_state = write_state; 1048 if (write_state == WR_NEED_COPY) 1049 itx->itx_sod += len; 1050 lr->lr_foid = ZVOL_OBJ; 1051 lr->lr_offset = off; 1052 lr->lr_length = len; 1053 lr->lr_blkoff = off - P2ALIGN_TYPED(off, blocksize, uint64_t); 1054 BP_ZERO(&lr->lr_blkptr); 1055 1056 itx->itx_private = zv; 1057 itx->itx_sync = sync; 1058 1059 (void) zil_itx_assign(zilog, itx, tx); 1060 1061 off += len; 1062 resid -= len; 1063 } 1064 } 1065 1066 static int 1067 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t size, 1068 boolean_t doread, boolean_t isdump) 1069 { 1070 vdev_disk_t *dvd; 1071 int c; 1072 int numerrors = 0; 1073 1074 for (c = 0; c < vd->vdev_children; c++) { 1075 ASSERT(vd->vdev_ops == &vdev_mirror_ops || 1076 vd->vdev_ops == &vdev_replacing_ops || 1077 vd->vdev_ops == &vdev_spare_ops); 1078 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1079 addr, offset, size, doread, isdump); 1080 if (err != 0) { 1081 numerrors++; 1082 } else if (doread) { 1083 break; 1084 } 1085 } 1086 1087 if (!vd->vdev_ops->vdev_op_leaf) 1088 return (numerrors < vd->vdev_children ? 0 : EIO); 1089 1090 if (doread && !vdev_readable(vd)) 1091 return (EIO); 1092 else if (!doread && !vdev_writeable(vd)) 1093 return (EIO); 1094 1095 dvd = vd->vdev_tsd; 1096 ASSERT3P(dvd, !=, NULL); 1097 offset += VDEV_LABEL_START_SIZE; 1098 1099 if (ddi_in_panic() || isdump) { 1100 ASSERT(!doread); 1101 if (doread) 1102 return (EIO); 1103 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1104 lbtodb(size))); 1105 } else { 1106 return (vdev_disk_physio(dvd->vd_lh, addr, size, offset, 1107 doread ? B_READ : B_WRITE)); 1108 } 1109 } 1110 1111 static int 1112 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1113 boolean_t doread, boolean_t isdump) 1114 { 1115 vdev_t *vd; 1116 int error; 1117 zvol_extent_t *ze; 1118 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1119 1120 /* Must be sector aligned, and not stradle a block boundary. */ 1121 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1122 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1123 return (EINVAL); 1124 } 1125 ASSERT(size <= zv->zv_volblocksize); 1126 1127 /* Locate the extent this belongs to */ 1128 ze = list_head(&zv->zv_extents); 1129 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1130 offset -= ze->ze_nblks * zv->zv_volblocksize; 1131 ze = list_next(&zv->zv_extents, ze); 1132 } 1133 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1134 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1135 offset += DVA_GET_OFFSET(&ze->ze_dva); 1136 error = zvol_dumpio_vdev(vd, addr, offset, size, doread, isdump); 1137 spa_config_exit(spa, SCL_STATE, FTAG); 1138 return (error); 1139 } 1140 1141 int 1142 zvol_strategy(buf_t *bp) 1143 { 1144 zvol_state_t *zv = ddi_get_soft_state(zvol_state, getminor(bp->b_edev)); 1145 uint64_t off, volsize; 1146 size_t resid; 1147 char *addr; 1148 objset_t *os; 1149 rl_t *rl; 1150 int error = 0; 1151 boolean_t doread = bp->b_flags & B_READ; 1152 boolean_t is_dump = zv->zv_flags & ZVOL_DUMPIFIED; 1153 boolean_t sync; 1154 1155 if (zv == NULL) { 1156 bioerror(bp, ENXIO); 1157 biodone(bp); 1158 return (0); 1159 } 1160 1161 if (getminor(bp->b_edev) == 0) { 1162 bioerror(bp, EINVAL); 1163 biodone(bp); 1164 return (0); 1165 } 1166 1167 if (!(bp->b_flags & B_READ) && 1168 (zv->zv_flags & ZVOL_RDONLY || 1169 zv->zv_mode & DS_MODE_READONLY)) { 1170 bioerror(bp, EROFS); 1171 biodone(bp); 1172 return (0); 1173 } 1174 1175 off = ldbtob(bp->b_blkno); 1176 volsize = zv->zv_volsize; 1177 1178 os = zv->zv_objset; 1179 ASSERT(os != NULL); 1180 1181 bp_mapin(bp); 1182 addr = bp->b_un.b_addr; 1183 resid = bp->b_bcount; 1184 1185 if (resid > 0 && (off < 0 || off >= volsize)) { 1186 bioerror(bp, EIO); 1187 biodone(bp); 1188 return (0); 1189 } 1190 1191 sync = !(bp->b_flags & B_ASYNC) && !doread && !is_dump && 1192 !(zv->zv_flags & ZVOL_WCE) && !zil_disable; 1193 1194 /* 1195 * There must be no buffer changes when doing a dmu_sync() because 1196 * we can't change the data whilst calculating the checksum. 1197 */ 1198 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1199 doread ? RL_READER : RL_WRITER); 1200 1201 while (resid != 0 && off < volsize) { 1202 size_t size = MIN(resid, zvol_maxphys); 1203 if (is_dump) { 1204 size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 1205 error = zvol_dumpio(zv, addr, off, size, 1206 doread, B_FALSE); 1207 } else if (doread) { 1208 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1209 DMU_READ_PREFETCH); 1210 } else { 1211 dmu_tx_t *tx = dmu_tx_create(os); 1212 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1213 error = dmu_tx_assign(tx, TXG_WAIT); 1214 if (error) { 1215 dmu_tx_abort(tx); 1216 } else { 1217 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1218 zvol_log_write(zv, tx, off, size, sync); 1219 dmu_tx_commit(tx); 1220 } 1221 } 1222 if (error) { 1223 /* convert checksum errors into IO errors */ 1224 if (error == ECKSUM) 1225 error = EIO; 1226 break; 1227 } 1228 off += size; 1229 addr += size; 1230 resid -= size; 1231 } 1232 zfs_range_unlock(rl); 1233 1234 if ((bp->b_resid = resid) == bp->b_bcount) 1235 bioerror(bp, off > volsize ? EINVAL : error); 1236 1237 if (sync) 1238 zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ); 1239 biodone(bp); 1240 1241 return (0); 1242 } 1243 1244 /* 1245 * Set the buffer count to the zvol maximum transfer. 1246 * Using our own routine instead of the default minphys() 1247 * means that for larger writes we write bigger buffers on X86 1248 * (128K instead of 56K) and flush the disk write cache less often 1249 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1250 * 56K on X86 and 128K on sparc). 1251 */ 1252 void 1253 zvol_minphys(struct buf *bp) 1254 { 1255 if (bp->b_bcount > zvol_maxphys) 1256 bp->b_bcount = zvol_maxphys; 1257 } 1258 1259 int 1260 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1261 { 1262 minor_t minor = getminor(dev); 1263 zvol_state_t *zv; 1264 int error = 0; 1265 uint64_t size; 1266 uint64_t boff; 1267 uint64_t resid; 1268 1269 if (minor == 0) /* This is the control device */ 1270 return (ENXIO); 1271 1272 zv = ddi_get_soft_state(zvol_state, minor); 1273 if (zv == NULL) 1274 return (ENXIO); 1275 1276 boff = ldbtob(blkno); 1277 resid = ldbtob(nblocks); 1278 1279 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1280 1281 while (resid) { 1282 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1283 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1284 if (error) 1285 break; 1286 boff += size; 1287 addr += size; 1288 resid -= size; 1289 } 1290 1291 return (error); 1292 } 1293 1294 /*ARGSUSED*/ 1295 int 1296 zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1297 { 1298 minor_t minor = getminor(dev); 1299 zvol_state_t *zv; 1300 uint64_t volsize; 1301 rl_t *rl; 1302 int error = 0; 1303 1304 if (minor == 0) /* This is the control device */ 1305 return (ENXIO); 1306 1307 zv = ddi_get_soft_state(zvol_state, minor); 1308 if (zv == NULL) 1309 return (ENXIO); 1310 1311 volsize = zv->zv_volsize; 1312 if (uio->uio_resid > 0 && 1313 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1314 return (EIO); 1315 1316 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1317 error = physio(zvol_strategy, NULL, dev, B_READ, 1318 zvol_minphys, uio); 1319 return (error); 1320 } 1321 1322 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1323 RL_READER); 1324 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1325 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1326 1327 /* don't read past the end */ 1328 if (bytes > volsize - uio->uio_loffset) 1329 bytes = volsize - uio->uio_loffset; 1330 1331 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1332 if (error) { 1333 /* convert checksum errors into IO errors */ 1334 if (error == ECKSUM) 1335 error = EIO; 1336 break; 1337 } 1338 } 1339 zfs_range_unlock(rl); 1340 return (error); 1341 } 1342 1343 /*ARGSUSED*/ 1344 int 1345 zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1346 { 1347 minor_t minor = getminor(dev); 1348 zvol_state_t *zv; 1349 uint64_t volsize; 1350 rl_t *rl; 1351 int error = 0; 1352 boolean_t sync; 1353 1354 if (minor == 0) /* This is the control device */ 1355 return (ENXIO); 1356 1357 zv = ddi_get_soft_state(zvol_state, minor); 1358 if (zv == NULL) 1359 return (ENXIO); 1360 1361 volsize = zv->zv_volsize; 1362 if (uio->uio_resid > 0 && 1363 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1364 return (EIO); 1365 1366 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1367 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1368 zvol_minphys, uio); 1369 return (error); 1370 } 1371 1372 sync = !(zv->zv_flags & ZVOL_WCE) && !zil_disable; 1373 1374 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1375 RL_WRITER); 1376 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1377 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1378 uint64_t off = uio->uio_loffset; 1379 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1380 1381 if (bytes > volsize - off) /* don't write past the end */ 1382 bytes = volsize - off; 1383 1384 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1385 error = dmu_tx_assign(tx, TXG_WAIT); 1386 if (error) { 1387 dmu_tx_abort(tx); 1388 break; 1389 } 1390 error = dmu_write_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes, tx); 1391 if (error == 0) 1392 zvol_log_write(zv, tx, off, bytes, sync); 1393 dmu_tx_commit(tx); 1394 1395 if (error) 1396 break; 1397 } 1398 zfs_range_unlock(rl); 1399 if (sync) 1400 zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ); 1401 return (error); 1402 } 1403 1404 int 1405 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1406 { 1407 struct uuid uuid = EFI_RESERVED; 1408 efi_gpe_t gpe = { 0 }; 1409 uint32_t crc; 1410 dk_efi_t efi; 1411 int length; 1412 char *ptr; 1413 1414 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1415 return (EFAULT); 1416 ptr = (char *)(uintptr_t)efi.dki_data_64; 1417 length = efi.dki_length; 1418 /* 1419 * Some clients may attempt to request a PMBR for the 1420 * zvol. Currently this interface will return EINVAL to 1421 * such requests. These requests could be supported by 1422 * adding a check for lba == 0 and consing up an appropriate 1423 * PMBR. 1424 */ 1425 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1426 return (EINVAL); 1427 1428 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1429 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1430 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1431 1432 if (efi.dki_lba == 1) { 1433 efi_gpt_t gpt = { 0 }; 1434 1435 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1436 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1437 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1438 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1439 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1440 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1441 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1442 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1443 gpt.efi_gpt_SizeOfPartitionEntry = 1444 LE_32(sizeof (efi_gpe_t)); 1445 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1446 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1447 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1448 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1449 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1450 flag)) 1451 return (EFAULT); 1452 ptr += sizeof (gpt); 1453 length -= sizeof (gpt); 1454 } 1455 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1456 length), flag)) 1457 return (EFAULT); 1458 return (0); 1459 } 1460 1461 /* 1462 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1463 */ 1464 /*ARGSUSED*/ 1465 int 1466 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1467 { 1468 zvol_state_t *zv; 1469 struct dk_cinfo dki; 1470 struct dk_minfo dkm; 1471 struct dk_callback *dkc; 1472 int error = 0; 1473 rl_t *rl; 1474 1475 mutex_enter(&zvol_state_lock); 1476 1477 zv = ddi_get_soft_state(zvol_state, getminor(dev)); 1478 1479 if (zv == NULL) { 1480 mutex_exit(&zvol_state_lock); 1481 return (ENXIO); 1482 } 1483 ASSERT(zv->zv_total_opens > 0); 1484 1485 switch (cmd) { 1486 1487 case DKIOCINFO: 1488 bzero(&dki, sizeof (dki)); 1489 (void) strcpy(dki.dki_cname, "zvol"); 1490 (void) strcpy(dki.dki_dname, "zvol"); 1491 dki.dki_ctype = DKC_UNKNOWN; 1492 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); 1493 mutex_exit(&zvol_state_lock); 1494 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1495 error = EFAULT; 1496 return (error); 1497 1498 case DKIOCGMEDIAINFO: 1499 bzero(&dkm, sizeof (dkm)); 1500 dkm.dki_lbsize = 1U << zv->zv_min_bs; 1501 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1502 dkm.dki_media_type = DK_UNKNOWN; 1503 mutex_exit(&zvol_state_lock); 1504 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1505 error = EFAULT; 1506 return (error); 1507 1508 case DKIOCGETEFI: 1509 { 1510 uint64_t vs = zv->zv_volsize; 1511 uint8_t bs = zv->zv_min_bs; 1512 1513 mutex_exit(&zvol_state_lock); 1514 error = zvol_getefi((void *)arg, flag, vs, bs); 1515 return (error); 1516 } 1517 1518 case DKIOCFLUSHWRITECACHE: 1519 dkc = (struct dk_callback *)arg; 1520 mutex_exit(&zvol_state_lock); 1521 zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ); 1522 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1523 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1524 error = 0; 1525 } 1526 return (error); 1527 1528 case DKIOCGETWCE: 1529 { 1530 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1531 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1532 flag)) 1533 error = EFAULT; 1534 break; 1535 } 1536 case DKIOCSETWCE: 1537 { 1538 int wce; 1539 if (ddi_copyin((void *)arg, &wce, sizeof (int), 1540 flag)) { 1541 error = EFAULT; 1542 break; 1543 } 1544 if (wce) { 1545 zv->zv_flags |= ZVOL_WCE; 1546 mutex_exit(&zvol_state_lock); 1547 } else { 1548 zv->zv_flags &= ~ZVOL_WCE; 1549 mutex_exit(&zvol_state_lock); 1550 zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ); 1551 } 1552 return (0); 1553 } 1554 1555 case DKIOCGGEOM: 1556 case DKIOCGVTOC: 1557 /* 1558 * commands using these (like prtvtoc) expect ENOTSUP 1559 * since we're emulating an EFI label 1560 */ 1561 error = ENOTSUP; 1562 break; 1563 1564 case DKIOCDUMPINIT: 1565 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1566 RL_WRITER); 1567 error = zvol_dumpify(zv); 1568 zfs_range_unlock(rl); 1569 break; 1570 1571 case DKIOCDUMPFINI: 1572 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 1573 break; 1574 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1575 RL_WRITER); 1576 error = zvol_dump_fini(zv); 1577 zfs_range_unlock(rl); 1578 break; 1579 1580 default: 1581 error = ENOTTY; 1582 break; 1583 1584 } 1585 mutex_exit(&zvol_state_lock); 1586 return (error); 1587 } 1588 1589 int 1590 zvol_busy(void) 1591 { 1592 return (zvol_minors != 0); 1593 } 1594 1595 void 1596 zvol_init(void) 1597 { 1598 VERIFY(ddi_soft_state_init(&zvol_state, sizeof (zvol_state_t), 1) == 0); 1599 mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL); 1600 } 1601 1602 void 1603 zvol_fini(void) 1604 { 1605 mutex_destroy(&zvol_state_lock); 1606 ddi_soft_state_fini(&zvol_state); 1607 } 1608 1609 static boolean_t 1610 zvol_is_swap(zvol_state_t *zv) 1611 { 1612 vnode_t *vp; 1613 boolean_t ret = B_FALSE; 1614 char *devpath; 1615 size_t devpathlen; 1616 int error; 1617 1618 devpathlen = strlen(ZVOL_FULL_DEV_DIR) + strlen(zv->zv_name) + 1; 1619 devpath = kmem_alloc(devpathlen, KM_SLEEP); 1620 (void) sprintf(devpath, "%s%s", ZVOL_FULL_DEV_DIR, zv->zv_name); 1621 error = lookupname(devpath, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp); 1622 kmem_free(devpath, devpathlen); 1623 1624 ret = !error && IS_SWAPVP(common_specvp(vp)); 1625 1626 if (vp != NULL) 1627 VN_RELE(vp); 1628 1629 return (ret); 1630 } 1631 1632 static int 1633 zvol_dump_init(zvol_state_t *zv, boolean_t resize) 1634 { 1635 dmu_tx_t *tx; 1636 int error = 0; 1637 objset_t *os = zv->zv_objset; 1638 nvlist_t *nv = NULL; 1639 1640 ASSERT(MUTEX_HELD(&zvol_state_lock)); 1641 1642 tx = dmu_tx_create(os); 1643 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1644 error = dmu_tx_assign(tx, TXG_WAIT); 1645 if (error) { 1646 dmu_tx_abort(tx); 1647 return (error); 1648 } 1649 1650 /* 1651 * If we are resizing the dump device then we only need to 1652 * update the refreservation to match the newly updated 1653 * zvolsize. Otherwise, we save off the original state of the 1654 * zvol so that we can restore them if the zvol is ever undumpified. 1655 */ 1656 if (resize) { 1657 error = zap_update(os, ZVOL_ZAP_OBJ, 1658 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1659 &zv->zv_volsize, tx); 1660 } else { 1661 uint64_t checksum, compress, refresrv, vbs; 1662 1663 error = dsl_prop_get_integer(zv->zv_name, 1664 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 1665 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1666 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL); 1667 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1668 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL); 1669 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1670 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL); 1671 1672 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1673 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 1674 &compress, tx); 1675 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1676 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx); 1677 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1678 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1679 &refresrv, tx); 1680 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1681 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 1682 &vbs, tx); 1683 } 1684 dmu_tx_commit(tx); 1685 1686 /* Truncate the file */ 1687 if (!error) 1688 error = dmu_free_long_range(zv->zv_objset, 1689 ZVOL_OBJ, 0, DMU_OBJECT_END); 1690 1691 if (error) 1692 return (error); 1693 1694 /* 1695 * We only need update the zvol's property if we are initializing 1696 * the dump area for the first time. 1697 */ 1698 if (!resize) { 1699 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1700 VERIFY(nvlist_add_uint64(nv, 1701 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 1702 VERIFY(nvlist_add_uint64(nv, 1703 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 1704 ZIO_COMPRESS_OFF) == 0); 1705 VERIFY(nvlist_add_uint64(nv, 1706 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 1707 ZIO_CHECKSUM_OFF) == 0); 1708 VERIFY(nvlist_add_uint64(nv, 1709 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 1710 SPA_MAXBLOCKSIZE) == 0); 1711 1712 error = zfs_set_prop_nvlist(zv->zv_name, nv); 1713 nvlist_free(nv); 1714 1715 if (error) 1716 return (error); 1717 } 1718 1719 /* Allocate the space for the dump */ 1720 error = zvol_prealloc(zv); 1721 return (error); 1722 } 1723 1724 static int 1725 zvol_dumpify(zvol_state_t *zv) 1726 { 1727 int error = 0; 1728 uint64_t dumpsize = 0; 1729 dmu_tx_t *tx; 1730 objset_t *os = zv->zv_objset; 1731 1732 if (zv->zv_flags & ZVOL_RDONLY || (zv->zv_mode & DS_MODE_READONLY)) 1733 return (EROFS); 1734 1735 /* 1736 * We do not support swap devices acting as dump devices. 1737 */ 1738 if (zvol_is_swap(zv)) 1739 return (ENOTSUP); 1740 1741 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 1742 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 1743 boolean_t resize = (dumpsize > 0) ? B_TRUE : B_FALSE; 1744 1745 if ((error = zvol_dump_init(zv, resize)) != 0) { 1746 (void) zvol_dump_fini(zv); 1747 return (error); 1748 } 1749 } 1750 1751 /* 1752 * Build up our lba mapping. 1753 */ 1754 error = zvol_get_lbas(zv); 1755 if (error) { 1756 (void) zvol_dump_fini(zv); 1757 return (error); 1758 } 1759 1760 tx = dmu_tx_create(os); 1761 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1762 error = dmu_tx_assign(tx, TXG_WAIT); 1763 if (error) { 1764 dmu_tx_abort(tx); 1765 (void) zvol_dump_fini(zv); 1766 return (error); 1767 } 1768 1769 zv->zv_flags |= ZVOL_DUMPIFIED; 1770 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 1771 &zv->zv_volsize, tx); 1772 dmu_tx_commit(tx); 1773 1774 if (error) { 1775 (void) zvol_dump_fini(zv); 1776 return (error); 1777 } 1778 1779 txg_wait_synced(dmu_objset_pool(os), 0); 1780 return (0); 1781 } 1782 1783 static int 1784 zvol_dump_fini(zvol_state_t *zv) 1785 { 1786 dmu_tx_t *tx; 1787 objset_t *os = zv->zv_objset; 1788 nvlist_t *nv; 1789 int error = 0; 1790 uint64_t checksum, compress, refresrv, vbs; 1791 1792 /* 1793 * Attempt to restore the zvol back to its pre-dumpified state. 1794 * This is a best-effort attempt as it's possible that not all 1795 * of these properties were initialized during the dumpify process 1796 * (i.e. error during zvol_dump_init). 1797 */ 1798 1799 tx = dmu_tx_create(os); 1800 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1801 error = dmu_tx_assign(tx, TXG_WAIT); 1802 if (error) { 1803 dmu_tx_abort(tx); 1804 return (error); 1805 } 1806 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 1807 dmu_tx_commit(tx); 1808 1809 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1810 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 1811 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1812 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 1813 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1814 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 1815 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1816 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 1817 1818 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1819 (void) nvlist_add_uint64(nv, 1820 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 1821 (void) nvlist_add_uint64(nv, 1822 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 1823 (void) nvlist_add_uint64(nv, 1824 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 1825 (void) nvlist_add_uint64(nv, 1826 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), vbs); 1827 (void) zfs_set_prop_nvlist(zv->zv_name, nv); 1828 nvlist_free(nv); 1829 1830 zvol_free_extents(zv); 1831 zv->zv_flags &= ~ZVOL_DUMPIFIED; 1832 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 1833 1834 return (0); 1835 } 1836