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 25 /* Portions Copyright 2010 Robert Milkowski */ 26 27 /* 28 * ZFS volume emulation driver. 29 * 30 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 31 * Volumes are accessed through the symbolic links named: 32 * 33 * /dev/zvol/dsk/<pool_name>/<dataset_name> 34 * /dev/zvol/rdsk/<pool_name>/<dataset_name> 35 * 36 * These links are created by the /dev filesystem (sdev_zvolops.c). 37 * Volumes are persistent through reboot. No user command needs to be 38 * run before opening and using a device. 39 */ 40 41 #include <sys/types.h> 42 #include <sys/param.h> 43 #include <sys/errno.h> 44 #include <sys/uio.h> 45 #include <sys/buf.h> 46 #include <sys/modctl.h> 47 #include <sys/open.h> 48 #include <sys/kmem.h> 49 #include <sys/conf.h> 50 #include <sys/cmn_err.h> 51 #include <sys/stat.h> 52 #include <sys/zap.h> 53 #include <sys/spa.h> 54 #include <sys/zio.h> 55 #include <sys/dmu_traverse.h> 56 #include <sys/dnode.h> 57 #include <sys/dsl_dataset.h> 58 #include <sys/dsl_prop.h> 59 #include <sys/dkio.h> 60 #include <sys/efi_partition.h> 61 #include <sys/byteorder.h> 62 #include <sys/pathname.h> 63 #include <sys/ddi.h> 64 #include <sys/sunddi.h> 65 #include <sys/crc32.h> 66 #include <sys/dirent.h> 67 #include <sys/policy.h> 68 #include <sys/fs/zfs.h> 69 #include <sys/zfs_ioctl.h> 70 #include <sys/mkdev.h> 71 #include <sys/zil.h> 72 #include <sys/refcount.h> 73 #include <sys/zfs_znode.h> 74 #include <sys/zfs_rlock.h> 75 #include <sys/vdev_disk.h> 76 #include <sys/vdev_impl.h> 77 #include <sys/zvol.h> 78 #include <sys/dumphdr.h> 79 #include <sys/zil_impl.h> 80 81 #include "zfs_namecheck.h" 82 83 void *zfsdev_state; 84 static char *zvol_tag = "zvol_tag"; 85 86 #define ZVOL_DUMPSIZE "dumpsize" 87 88 /* 89 * This lock protects the zfsdev_state structure from being modified 90 * while it's being used, e.g. an open that comes in before a create 91 * finishes. It also protects temporary opens of the dataset so that, 92 * e.g., an open doesn't get a spurious EBUSY. 93 */ 94 kmutex_t zfsdev_state_lock; 95 static uint32_t zvol_minors; 96 97 typedef struct zvol_extent { 98 list_node_t ze_node; 99 dva_t ze_dva; /* dva associated with this extent */ 100 uint64_t ze_nblks; /* number of blocks in extent */ 101 } zvol_extent_t; 102 103 /* 104 * The in-core state of each volume. 105 */ 106 typedef struct zvol_state { 107 char zv_name[MAXPATHLEN]; /* pool/dd name */ 108 uint64_t zv_volsize; /* amount of space we advertise */ 109 uint64_t zv_volblocksize; /* volume block size */ 110 minor_t zv_minor; /* minor number */ 111 uint8_t zv_min_bs; /* minimum addressable block shift */ 112 uint8_t zv_flags; /* readonly, dumpified, etc. */ 113 objset_t *zv_objset; /* objset handle */ 114 uint32_t zv_open_count[OTYPCNT]; /* open counts */ 115 uint32_t zv_total_opens; /* total open count */ 116 zilog_t *zv_zilog; /* ZIL handle */ 117 list_t zv_extents; /* List of extents for dump */ 118 znode_t zv_znode; /* for range locking */ 119 dmu_buf_t *zv_dbuf; /* bonus handle */ 120 } zvol_state_t; 121 122 /* 123 * zvol specific flags 124 */ 125 #define ZVOL_RDONLY 0x1 126 #define ZVOL_DUMPIFIED 0x2 127 #define ZVOL_EXCL 0x4 128 #define ZVOL_WCE 0x8 129 130 /* 131 * zvol maximum transfer in one DMU tx. 132 */ 133 int zvol_maxphys = DMU_MAX_ACCESS/2; 134 135 extern int zfs_set_prop_nvlist(const char *, zprop_source_t, 136 nvlist_t *, nvlist_t **); 137 static int zvol_remove_zv(zvol_state_t *); 138 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio); 139 static int zvol_dumpify(zvol_state_t *zv); 140 static int zvol_dump_fini(zvol_state_t *zv); 141 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); 142 143 static void 144 zvol_size_changed(uint64_t volsize, major_t maj, minor_t min) 145 { 146 dev_t dev = makedevice(maj, min); 147 148 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 149 "Size", volsize) == DDI_SUCCESS); 150 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 151 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS); 152 153 /* Notify specfs to invalidate the cached size */ 154 spec_size_invalidate(dev, VBLK); 155 spec_size_invalidate(dev, VCHR); 156 } 157 158 int 159 zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 160 { 161 if (volsize == 0) 162 return (EINVAL); 163 164 if (volsize % blocksize != 0) 165 return (EINVAL); 166 167 #ifdef _ILP32 168 if (volsize - 1 > SPEC_MAXOFFSET_T) 169 return (EOVERFLOW); 170 #endif 171 return (0); 172 } 173 174 int 175 zvol_check_volblocksize(uint64_t volblocksize) 176 { 177 if (volblocksize < SPA_MINBLOCKSIZE || 178 volblocksize > SPA_MAXBLOCKSIZE || 179 !ISP2(volblocksize)) 180 return (EDOM); 181 182 return (0); 183 } 184 185 int 186 zvol_get_stats(objset_t *os, nvlist_t *nv) 187 { 188 int error; 189 dmu_object_info_t doi; 190 uint64_t val; 191 192 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 193 if (error) 194 return (error); 195 196 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 197 198 error = dmu_object_info(os, ZVOL_OBJ, &doi); 199 200 if (error == 0) { 201 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 202 doi.doi_data_block_size); 203 } 204 205 return (error); 206 } 207 208 static zvol_state_t * 209 zvol_minor_lookup(const char *name) 210 { 211 minor_t minor; 212 zvol_state_t *zv; 213 214 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 215 216 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) { 217 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 218 if (zv == NULL) 219 continue; 220 if (strcmp(zv->zv_name, name) == 0) 221 return (zv); 222 } 223 224 return (NULL); 225 } 226 227 /* extent mapping arg */ 228 struct maparg { 229 zvol_state_t *ma_zv; 230 uint64_t ma_blks; 231 }; 232 233 /*ARGSUSED*/ 234 static int 235 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf, 236 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 237 { 238 struct maparg *ma = arg; 239 zvol_extent_t *ze; 240 int bs = ma->ma_zv->zv_volblocksize; 241 242 if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 243 return (0); 244 245 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 246 ma->ma_blks++; 247 248 /* Abort immediately if we have encountered gang blocks */ 249 if (BP_IS_GANG(bp)) 250 return (EFRAGS); 251 252 /* 253 * See if the block is at the end of the previous extent. 254 */ 255 ze = list_tail(&ma->ma_zv->zv_extents); 256 if (ze && 257 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 258 DVA_GET_OFFSET(BP_IDENTITY(bp)) == 259 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 260 ze->ze_nblks++; 261 return (0); 262 } 263 264 dprintf_bp(bp, "%s", "next blkptr:"); 265 266 /* start a new extent */ 267 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 268 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 269 ze->ze_nblks = 1; 270 list_insert_tail(&ma->ma_zv->zv_extents, ze); 271 return (0); 272 } 273 274 static void 275 zvol_free_extents(zvol_state_t *zv) 276 { 277 zvol_extent_t *ze; 278 279 while (ze = list_head(&zv->zv_extents)) { 280 list_remove(&zv->zv_extents, ze); 281 kmem_free(ze, sizeof (zvol_extent_t)); 282 } 283 } 284 285 static int 286 zvol_get_lbas(zvol_state_t *zv) 287 { 288 objset_t *os = zv->zv_objset; 289 struct maparg ma; 290 int err; 291 292 ma.ma_zv = zv; 293 ma.ma_blks = 0; 294 zvol_free_extents(zv); 295 296 /* commit any in-flight changes before traversing the dataset */ 297 txg_wait_synced(dmu_objset_pool(os), 0); 298 err = traverse_dataset(dmu_objset_ds(os), 0, 299 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 300 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 301 zvol_free_extents(zv); 302 return (err ? err : EIO); 303 } 304 305 return (0); 306 } 307 308 /* ARGSUSED */ 309 void 310 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 311 { 312 zfs_creat_t *zct = arg; 313 nvlist_t *nvprops = zct->zct_props; 314 int error; 315 uint64_t volblocksize, volsize; 316 317 VERIFY(nvlist_lookup_uint64(nvprops, 318 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 319 if (nvlist_lookup_uint64(nvprops, 320 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 321 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 322 323 /* 324 * These properties must be removed from the list so the generic 325 * property setting step won't apply to them. 326 */ 327 VERIFY(nvlist_remove_all(nvprops, 328 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 329 (void) nvlist_remove_all(nvprops, 330 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 331 332 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 333 DMU_OT_NONE, 0, tx); 334 ASSERT(error == 0); 335 336 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 337 DMU_OT_NONE, 0, tx); 338 ASSERT(error == 0); 339 340 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 341 ASSERT(error == 0); 342 } 343 344 /* 345 * Replay a TX_WRITE ZIL transaction that didn't get committed 346 * after a system failure 347 */ 348 static int 349 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 350 { 351 objset_t *os = zv->zv_objset; 352 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 353 uint64_t offset, length; 354 dmu_tx_t *tx; 355 int error; 356 357 if (byteswap) 358 byteswap_uint64_array(lr, sizeof (*lr)); 359 360 offset = lr->lr_offset; 361 length = lr->lr_length; 362 363 /* If it's a dmu_sync() block, write the whole block */ 364 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 365 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 366 if (length < blocksize) { 367 offset -= offset % blocksize; 368 length = blocksize; 369 } 370 } 371 372 tx = dmu_tx_create(os); 373 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 374 error = dmu_tx_assign(tx, TXG_WAIT); 375 if (error) { 376 dmu_tx_abort(tx); 377 } else { 378 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 379 dmu_tx_commit(tx); 380 } 381 382 return (error); 383 } 384 385 /* ARGSUSED */ 386 static int 387 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 388 { 389 return (ENOTSUP); 390 } 391 392 /* 393 * Callback vectors for replaying records. 394 * Only TX_WRITE is needed for zvol. 395 */ 396 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 397 zvol_replay_err, /* 0 no such transaction type */ 398 zvol_replay_err, /* TX_CREATE */ 399 zvol_replay_err, /* TX_MKDIR */ 400 zvol_replay_err, /* TX_MKXATTR */ 401 zvol_replay_err, /* TX_SYMLINK */ 402 zvol_replay_err, /* TX_REMOVE */ 403 zvol_replay_err, /* TX_RMDIR */ 404 zvol_replay_err, /* TX_LINK */ 405 zvol_replay_err, /* TX_RENAME */ 406 zvol_replay_write, /* TX_WRITE */ 407 zvol_replay_err, /* TX_TRUNCATE */ 408 zvol_replay_err, /* TX_SETATTR */ 409 zvol_replay_err, /* TX_ACL */ 410 zvol_replay_err, /* TX_CREATE_ACL */ 411 zvol_replay_err, /* TX_CREATE_ATTR */ 412 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 413 zvol_replay_err, /* TX_MKDIR_ACL */ 414 zvol_replay_err, /* TX_MKDIR_ATTR */ 415 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 416 zvol_replay_err, /* TX_WRITE2 */ 417 }; 418 419 int 420 zvol_name2minor(const char *name, minor_t *minor) 421 { 422 zvol_state_t *zv; 423 424 mutex_enter(&zfsdev_state_lock); 425 zv = zvol_minor_lookup(name); 426 if (minor && zv) 427 *minor = zv->zv_minor; 428 mutex_exit(&zfsdev_state_lock); 429 return (zv ? 0 : -1); 430 } 431 432 /* 433 * Create a minor node (plus a whole lot more) for the specified volume. 434 */ 435 int 436 zvol_create_minor(const char *name) 437 { 438 zfs_soft_state_t *zs; 439 zvol_state_t *zv; 440 objset_t *os; 441 dmu_object_info_t doi; 442 minor_t minor = 0; 443 char chrbuf[30], blkbuf[30]; 444 int error; 445 446 mutex_enter(&zfsdev_state_lock); 447 448 if (zvol_minor_lookup(name) != NULL) { 449 mutex_exit(&zfsdev_state_lock); 450 return (EEXIST); 451 } 452 453 /* lie and say we're read-only */ 454 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, zvol_tag, &os); 455 456 if (error) { 457 mutex_exit(&zfsdev_state_lock); 458 return (error); 459 } 460 461 if ((minor = zfsdev_minor_alloc()) == 0) { 462 dmu_objset_disown(os, zvol_tag); 463 mutex_exit(&zfsdev_state_lock); 464 return (ENXIO); 465 } 466 467 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 468 dmu_objset_disown(os, zvol_tag); 469 mutex_exit(&zfsdev_state_lock); 470 return (EAGAIN); 471 } 472 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 473 (char *)name); 474 475 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 476 477 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 478 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 479 ddi_soft_state_free(zfsdev_state, minor); 480 dmu_objset_disown(os, zvol_tag); 481 mutex_exit(&zfsdev_state_lock); 482 return (EAGAIN); 483 } 484 485 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 486 487 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 488 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 489 ddi_remove_minor_node(zfs_dip, chrbuf); 490 ddi_soft_state_free(zfsdev_state, minor); 491 dmu_objset_disown(os, zvol_tag); 492 mutex_exit(&zfsdev_state_lock); 493 return (EAGAIN); 494 } 495 496 zs = ddi_get_soft_state(zfsdev_state, minor); 497 zs->zss_type = ZSST_ZVOL; 498 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 499 (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 500 zv->zv_min_bs = DEV_BSHIFT; 501 zv->zv_minor = minor; 502 zv->zv_objset = os; 503 if (dmu_objset_is_snapshot(os)) 504 zv->zv_flags |= ZVOL_RDONLY; 505 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 506 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 507 sizeof (rl_t), offsetof(rl_t, r_node)); 508 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 509 offsetof(zvol_extent_t, ze_node)); 510 /* get and cache the blocksize */ 511 error = dmu_object_info(os, ZVOL_OBJ, &doi); 512 ASSERT(error == 0); 513 zv->zv_volblocksize = doi.doi_data_block_size; 514 515 if (zil_replay_disable) 516 zil_destroy(dmu_objset_zil(os), B_FALSE); 517 else 518 zil_replay(os, zv, zvol_replay_vector); 519 dmu_objset_disown(os, zvol_tag); 520 zv->zv_objset = NULL; 521 522 zvol_minors++; 523 524 mutex_exit(&zfsdev_state_lock); 525 526 return (0); 527 } 528 529 /* 530 * Remove minor node for the specified volume. 531 */ 532 static int 533 zvol_remove_zv(zvol_state_t *zv) 534 { 535 char nmbuf[20]; 536 minor_t minor = zv->zv_minor; 537 538 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 539 if (zv->zv_total_opens != 0) 540 return (EBUSY); 541 542 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 543 ddi_remove_minor_node(zfs_dip, nmbuf); 544 545 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor); 546 ddi_remove_minor_node(zfs_dip, nmbuf); 547 548 avl_destroy(&zv->zv_znode.z_range_avl); 549 mutex_destroy(&zv->zv_znode.z_range_lock); 550 551 kmem_free(zv, sizeof (zvol_state_t)); 552 553 ddi_soft_state_free(zfsdev_state, minor); 554 555 zvol_minors--; 556 return (0); 557 } 558 559 int 560 zvol_remove_minor(const char *name) 561 { 562 zvol_state_t *zv; 563 int rc; 564 565 mutex_enter(&zfsdev_state_lock); 566 if ((zv = zvol_minor_lookup(name)) == NULL) { 567 mutex_exit(&zfsdev_state_lock); 568 return (ENXIO); 569 } 570 rc = zvol_remove_zv(zv); 571 mutex_exit(&zfsdev_state_lock); 572 return (rc); 573 } 574 575 int 576 zvol_first_open(zvol_state_t *zv) 577 { 578 objset_t *os; 579 uint64_t volsize; 580 int error; 581 uint64_t readonly; 582 583 /* lie and say we're read-only */ 584 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 585 zvol_tag, &os); 586 if (error) 587 return (error); 588 589 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 590 if (error) { 591 ASSERT(error == 0); 592 dmu_objset_disown(os, zvol_tag); 593 return (error); 594 } 595 zv->zv_objset = os; 596 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 597 if (error) { 598 dmu_objset_disown(os, zvol_tag); 599 return (error); 600 } 601 zv->zv_volsize = volsize; 602 zv->zv_zilog = zil_open(os, zvol_get_data); 603 zvol_size_changed(zv->zv_volsize, ddi_driver_major(zfs_dip), 604 zv->zv_minor); 605 606 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 607 NULL) == 0); 608 if (readonly || dmu_objset_is_snapshot(os)) 609 zv->zv_flags |= ZVOL_RDONLY; 610 else 611 zv->zv_flags &= ~ZVOL_RDONLY; 612 return (error); 613 } 614 615 void 616 zvol_last_close(zvol_state_t *zv) 617 { 618 zil_close(zv->zv_zilog); 619 zv->zv_zilog = NULL; 620 dmu_buf_rele(zv->zv_dbuf, zvol_tag); 621 zv->zv_dbuf = NULL; 622 dmu_objset_disown(zv->zv_objset, zvol_tag); 623 zv->zv_objset = NULL; 624 } 625 626 int 627 zvol_prealloc(zvol_state_t *zv) 628 { 629 objset_t *os = zv->zv_objset; 630 dmu_tx_t *tx; 631 uint64_t refd, avail, usedobjs, availobjs; 632 uint64_t resid = zv->zv_volsize; 633 uint64_t off = 0; 634 635 /* Check the space usage before attempting to allocate the space */ 636 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 637 if (avail < zv->zv_volsize) 638 return (ENOSPC); 639 640 /* Free old extents if they exist */ 641 zvol_free_extents(zv); 642 643 while (resid != 0) { 644 int error; 645 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); 646 647 tx = dmu_tx_create(os); 648 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 649 error = dmu_tx_assign(tx, TXG_WAIT); 650 if (error) { 651 dmu_tx_abort(tx); 652 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 653 return (error); 654 } 655 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 656 dmu_tx_commit(tx); 657 off += bytes; 658 resid -= bytes; 659 } 660 txg_wait_synced(dmu_objset_pool(os), 0); 661 662 return (0); 663 } 664 665 int 666 zvol_update_volsize(objset_t *os, uint64_t volsize) 667 { 668 dmu_tx_t *tx; 669 int error; 670 671 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 672 673 tx = dmu_tx_create(os); 674 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 675 error = dmu_tx_assign(tx, TXG_WAIT); 676 if (error) { 677 dmu_tx_abort(tx); 678 return (error); 679 } 680 681 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 682 &volsize, tx); 683 dmu_tx_commit(tx); 684 685 if (error == 0) 686 error = dmu_free_long_range(os, 687 ZVOL_OBJ, volsize, DMU_OBJECT_END); 688 return (error); 689 } 690 691 void 692 zvol_remove_minors(const char *name) 693 { 694 zvol_state_t *zv; 695 char *namebuf; 696 minor_t minor; 697 698 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP); 699 (void) strncpy(namebuf, name, strlen(name)); 700 (void) strcat(namebuf, "/"); 701 mutex_enter(&zfsdev_state_lock); 702 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) { 703 704 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 705 if (zv == NULL) 706 continue; 707 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0) 708 (void) zvol_remove_zv(zv); 709 } 710 kmem_free(namebuf, strlen(name) + 2); 711 712 mutex_exit(&zfsdev_state_lock); 713 } 714 715 int 716 zvol_set_volsize(const char *name, major_t maj, uint64_t volsize) 717 { 718 zvol_state_t *zv = NULL; 719 objset_t *os; 720 int error; 721 dmu_object_info_t doi; 722 uint64_t old_volsize = 0ULL; 723 uint64_t readonly; 724 725 mutex_enter(&zfsdev_state_lock); 726 zv = zvol_minor_lookup(name); 727 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 728 mutex_exit(&zfsdev_state_lock); 729 return (error); 730 } 731 732 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 733 (error = zvol_check_volsize(volsize, 734 doi.doi_data_block_size)) != 0) 735 goto out; 736 737 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, 738 NULL) == 0); 739 if (readonly) { 740 error = EROFS; 741 goto out; 742 } 743 744 error = zvol_update_volsize(os, volsize); 745 /* 746 * Reinitialize the dump area to the new size. If we 747 * failed to resize the dump area then restore it back to 748 * its original size. 749 */ 750 if (zv && error == 0) { 751 if (zv->zv_flags & ZVOL_DUMPIFIED) { 752 old_volsize = zv->zv_volsize; 753 zv->zv_volsize = volsize; 754 if ((error = zvol_dumpify(zv)) != 0 || 755 (error = dumpvp_resize()) != 0) { 756 (void) zvol_update_volsize(os, old_volsize); 757 zv->zv_volsize = old_volsize; 758 error = zvol_dumpify(zv); 759 } 760 } 761 if (error == 0) { 762 zv->zv_volsize = volsize; 763 zvol_size_changed(volsize, maj, zv->zv_minor); 764 } 765 } 766 767 /* 768 * Generate a LUN expansion event. 769 */ 770 if (zv && error == 0) { 771 sysevent_id_t eid; 772 nvlist_t *attr; 773 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 774 775 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 776 zv->zv_minor); 777 778 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 779 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 780 781 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 782 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 783 784 nvlist_free(attr); 785 kmem_free(physpath, MAXPATHLEN); 786 } 787 788 out: 789 dmu_objset_rele(os, FTAG); 790 791 mutex_exit(&zfsdev_state_lock); 792 793 return (error); 794 } 795 796 /*ARGSUSED*/ 797 int 798 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr) 799 { 800 zvol_state_t *zv; 801 int err = 0; 802 803 mutex_enter(&zfsdev_state_lock); 804 805 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL); 806 if (zv == NULL) { 807 mutex_exit(&zfsdev_state_lock); 808 return (ENXIO); 809 } 810 811 if (zv->zv_total_opens == 0) 812 err = zvol_first_open(zv); 813 if (err) { 814 mutex_exit(&zfsdev_state_lock); 815 return (err); 816 } 817 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 818 err = EROFS; 819 goto out; 820 } 821 if (zv->zv_flags & ZVOL_EXCL) { 822 err = EBUSY; 823 goto out; 824 } 825 if (flag & FEXCL) { 826 if (zv->zv_total_opens != 0) { 827 err = EBUSY; 828 goto out; 829 } 830 zv->zv_flags |= ZVOL_EXCL; 831 } 832 833 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) { 834 zv->zv_open_count[otyp]++; 835 zv->zv_total_opens++; 836 } 837 mutex_exit(&zfsdev_state_lock); 838 839 return (err); 840 out: 841 if (zv->zv_total_opens == 0) 842 zvol_last_close(zv); 843 mutex_exit(&zfsdev_state_lock); 844 return (err); 845 } 846 847 /*ARGSUSED*/ 848 int 849 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr) 850 { 851 minor_t minor = getminor(dev); 852 zvol_state_t *zv; 853 int error = 0; 854 855 mutex_enter(&zfsdev_state_lock); 856 857 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 858 if (zv == NULL) { 859 mutex_exit(&zfsdev_state_lock); 860 return (ENXIO); 861 } 862 863 if (zv->zv_flags & ZVOL_EXCL) { 864 ASSERT(zv->zv_total_opens == 1); 865 zv->zv_flags &= ~ZVOL_EXCL; 866 } 867 868 /* 869 * If the open count is zero, this is a spurious close. 870 * That indicates a bug in the kernel / DDI framework. 871 */ 872 ASSERT(zv->zv_open_count[otyp] != 0); 873 ASSERT(zv->zv_total_opens != 0); 874 875 /* 876 * You may get multiple opens, but only one close. 877 */ 878 zv->zv_open_count[otyp]--; 879 zv->zv_total_opens--; 880 881 if (zv->zv_total_opens == 0) 882 zvol_last_close(zv); 883 884 mutex_exit(&zfsdev_state_lock); 885 return (error); 886 } 887 888 static void 889 zvol_get_done(zgd_t *zgd, int error) 890 { 891 if (zgd->zgd_db) 892 dmu_buf_rele(zgd->zgd_db, zgd); 893 894 zfs_range_unlock(zgd->zgd_rl); 895 896 if (error == 0 && zgd->zgd_bp) 897 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 898 899 kmem_free(zgd, sizeof (zgd_t)); 900 } 901 902 /* 903 * Get data to generate a TX_WRITE intent log record. 904 */ 905 static int 906 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 907 { 908 zvol_state_t *zv = arg; 909 objset_t *os = zv->zv_objset; 910 uint64_t object = ZVOL_OBJ; 911 uint64_t offset = lr->lr_offset; 912 uint64_t size = lr->lr_length; /* length of user data */ 913 blkptr_t *bp = &lr->lr_blkptr; 914 dmu_buf_t *db; 915 zgd_t *zgd; 916 int error; 917 918 ASSERT(zio != NULL); 919 ASSERT(size != 0); 920 921 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 922 zgd->zgd_zilog = zv->zv_zilog; 923 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 924 925 /* 926 * Write records come in two flavors: immediate and indirect. 927 * For small writes it's cheaper to store the data with the 928 * log record (immediate); for large writes it's cheaper to 929 * sync the data and get a pointer to it (indirect) so that 930 * we don't have to write the data twice. 931 */ 932 if (buf != NULL) { /* immediate write */ 933 error = dmu_read(os, object, offset, size, buf, 934 DMU_READ_NO_PREFETCH); 935 } else { 936 size = zv->zv_volblocksize; 937 offset = P2ALIGN(offset, size); 938 error = dmu_buf_hold(os, object, offset, zgd, &db, 939 DMU_READ_NO_PREFETCH); 940 if (error == 0) { 941 zgd->zgd_db = db; 942 zgd->zgd_bp = bp; 943 944 ASSERT(db->db_offset == offset); 945 ASSERT(db->db_size == size); 946 947 error = dmu_sync(zio, lr->lr_common.lrc_txg, 948 zvol_get_done, zgd); 949 950 if (error == 0) 951 return (0); 952 } 953 } 954 955 zvol_get_done(zgd, error); 956 957 return (error); 958 } 959 960 /* 961 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 962 * 963 * We store data in the log buffers if it's small enough. 964 * Otherwise we will later flush the data out via dmu_sync(). 965 */ 966 ssize_t zvol_immediate_write_sz = 32768; 967 968 static void 969 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 970 boolean_t sync) 971 { 972 uint32_t blocksize = zv->zv_volblocksize; 973 zilog_t *zilog = zv->zv_zilog; 974 boolean_t slogging; 975 ssize_t immediate_write_sz; 976 977 if (zil_replaying(zilog, tx)) 978 return; 979 980 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 981 ? 0 : zvol_immediate_write_sz; 982 983 slogging = spa_has_slogs(zilog->zl_spa) && 984 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 985 986 while (resid) { 987 itx_t *itx; 988 lr_write_t *lr; 989 ssize_t len; 990 itx_wr_state_t write_state; 991 992 /* 993 * Unlike zfs_log_write() we can be called with 994 * upto DMU_MAX_ACCESS/2 (5MB) writes. 995 */ 996 if (blocksize > immediate_write_sz && !slogging && 997 resid >= blocksize && off % blocksize == 0) { 998 write_state = WR_INDIRECT; /* uses dmu_sync */ 999 len = blocksize; 1000 } else if (sync) { 1001 write_state = WR_COPIED; 1002 len = MIN(ZIL_MAX_LOG_DATA, resid); 1003 } else { 1004 write_state = WR_NEED_COPY; 1005 len = MIN(ZIL_MAX_LOG_DATA, resid); 1006 } 1007 1008 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1009 (write_state == WR_COPIED ? len : 0)); 1010 lr = (lr_write_t *)&itx->itx_lr; 1011 if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 1012 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1013 zil_itx_destroy(itx); 1014 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1015 lr = (lr_write_t *)&itx->itx_lr; 1016 write_state = WR_NEED_COPY; 1017 } 1018 1019 itx->itx_wr_state = write_state; 1020 if (write_state == WR_NEED_COPY) 1021 itx->itx_sod += len; 1022 lr->lr_foid = ZVOL_OBJ; 1023 lr->lr_offset = off; 1024 lr->lr_length = len; 1025 lr->lr_blkoff = 0; 1026 BP_ZERO(&lr->lr_blkptr); 1027 1028 itx->itx_private = zv; 1029 itx->itx_sync = sync; 1030 1031 (void) zil_itx_assign(zilog, itx, tx); 1032 1033 off += len; 1034 resid -= len; 1035 } 1036 } 1037 1038 static int 1039 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t size, 1040 boolean_t doread, boolean_t isdump) 1041 { 1042 vdev_disk_t *dvd; 1043 int c; 1044 int numerrors = 0; 1045 1046 for (c = 0; c < vd->vdev_children; c++) { 1047 ASSERT(vd->vdev_ops == &vdev_mirror_ops || 1048 vd->vdev_ops == &vdev_replacing_ops || 1049 vd->vdev_ops == &vdev_spare_ops); 1050 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1051 addr, offset, size, doread, isdump); 1052 if (err != 0) { 1053 numerrors++; 1054 } else if (doread) { 1055 break; 1056 } 1057 } 1058 1059 if (!vd->vdev_ops->vdev_op_leaf) 1060 return (numerrors < vd->vdev_children ? 0 : EIO); 1061 1062 if (doread && !vdev_readable(vd)) 1063 return (EIO); 1064 else if (!doread && !vdev_writeable(vd)) 1065 return (EIO); 1066 1067 dvd = vd->vdev_tsd; 1068 ASSERT3P(dvd, !=, NULL); 1069 offset += VDEV_LABEL_START_SIZE; 1070 1071 if (ddi_in_panic() || isdump) { 1072 ASSERT(!doread); 1073 if (doread) 1074 return (EIO); 1075 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1076 lbtodb(size))); 1077 } else { 1078 return (vdev_disk_physio(dvd->vd_lh, addr, size, offset, 1079 doread ? B_READ : B_WRITE)); 1080 } 1081 } 1082 1083 static int 1084 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1085 boolean_t doread, boolean_t isdump) 1086 { 1087 vdev_t *vd; 1088 int error; 1089 zvol_extent_t *ze; 1090 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1091 1092 /* Must be sector aligned, and not stradle a block boundary. */ 1093 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1094 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1095 return (EINVAL); 1096 } 1097 ASSERT(size <= zv->zv_volblocksize); 1098 1099 /* Locate the extent this belongs to */ 1100 ze = list_head(&zv->zv_extents); 1101 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1102 offset -= ze->ze_nblks * zv->zv_volblocksize; 1103 ze = list_next(&zv->zv_extents, ze); 1104 } 1105 1106 if (!ddi_in_panic()) 1107 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1108 1109 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1110 offset += DVA_GET_OFFSET(&ze->ze_dva); 1111 error = zvol_dumpio_vdev(vd, addr, offset, size, doread, isdump); 1112 1113 if (!ddi_in_panic()) 1114 spa_config_exit(spa, SCL_STATE, FTAG); 1115 1116 return (error); 1117 } 1118 1119 int 1120 zvol_strategy(buf_t *bp) 1121 { 1122 zfs_soft_state_t *zs = NULL; 1123 zvol_state_t *zv; 1124 uint64_t off, volsize; 1125 size_t resid; 1126 char *addr; 1127 objset_t *os; 1128 rl_t *rl; 1129 int error = 0; 1130 boolean_t doread = bp->b_flags & B_READ; 1131 boolean_t is_dump; 1132 boolean_t sync; 1133 1134 if (getminor(bp->b_edev) == 0) { 1135 error = EINVAL; 1136 } else { 1137 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev)); 1138 if (zs == NULL) 1139 error = ENXIO; 1140 else if (zs->zss_type != ZSST_ZVOL) 1141 error = EINVAL; 1142 } 1143 1144 if (error) { 1145 bioerror(bp, error); 1146 biodone(bp); 1147 return (0); 1148 } 1149 1150 zv = zs->zss_data; 1151 1152 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) { 1153 bioerror(bp, EROFS); 1154 biodone(bp); 1155 return (0); 1156 } 1157 1158 off = ldbtob(bp->b_blkno); 1159 volsize = zv->zv_volsize; 1160 1161 os = zv->zv_objset; 1162 ASSERT(os != NULL); 1163 1164 bp_mapin(bp); 1165 addr = bp->b_un.b_addr; 1166 resid = bp->b_bcount; 1167 1168 if (resid > 0 && (off < 0 || off >= volsize)) { 1169 bioerror(bp, EIO); 1170 biodone(bp); 1171 return (0); 1172 } 1173 1174 is_dump = zv->zv_flags & ZVOL_DUMPIFIED; 1175 sync = ((!(bp->b_flags & B_ASYNC) && 1176 !(zv->zv_flags & ZVOL_WCE)) || 1177 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) && 1178 !doread && !is_dump; 1179 1180 /* 1181 * There must be no buffer changes when doing a dmu_sync() because 1182 * we can't change the data whilst calculating the checksum. 1183 */ 1184 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1185 doread ? RL_READER : RL_WRITER); 1186 1187 while (resid != 0 && off < volsize) { 1188 size_t size = MIN(resid, zvol_maxphys); 1189 if (is_dump) { 1190 size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 1191 error = zvol_dumpio(zv, addr, off, size, 1192 doread, B_FALSE); 1193 } else if (doread) { 1194 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1195 DMU_READ_PREFETCH); 1196 } else { 1197 dmu_tx_t *tx = dmu_tx_create(os); 1198 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1199 error = dmu_tx_assign(tx, TXG_WAIT); 1200 if (error) { 1201 dmu_tx_abort(tx); 1202 } else { 1203 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1204 zvol_log_write(zv, tx, off, size, sync); 1205 dmu_tx_commit(tx); 1206 } 1207 } 1208 if (error) { 1209 /* convert checksum errors into IO errors */ 1210 if (error == ECKSUM) 1211 error = EIO; 1212 break; 1213 } 1214 off += size; 1215 addr += size; 1216 resid -= size; 1217 } 1218 zfs_range_unlock(rl); 1219 1220 if ((bp->b_resid = resid) == bp->b_bcount) 1221 bioerror(bp, off > volsize ? EINVAL : error); 1222 1223 if (sync) 1224 zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ); 1225 biodone(bp); 1226 1227 return (0); 1228 } 1229 1230 /* 1231 * Set the buffer count to the zvol maximum transfer. 1232 * Using our own routine instead of the default minphys() 1233 * means that for larger writes we write bigger buffers on X86 1234 * (128K instead of 56K) and flush the disk write cache less often 1235 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1236 * 56K on X86 and 128K on sparc). 1237 */ 1238 void 1239 zvol_minphys(struct buf *bp) 1240 { 1241 if (bp->b_bcount > zvol_maxphys) 1242 bp->b_bcount = zvol_maxphys; 1243 } 1244 1245 int 1246 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1247 { 1248 minor_t minor = getminor(dev); 1249 zvol_state_t *zv; 1250 int error = 0; 1251 uint64_t size; 1252 uint64_t boff; 1253 uint64_t resid; 1254 1255 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1256 if (zv == NULL) 1257 return (ENXIO); 1258 1259 boff = ldbtob(blkno); 1260 resid = ldbtob(nblocks); 1261 1262 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1263 1264 while (resid) { 1265 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1266 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1267 if (error) 1268 break; 1269 boff += size; 1270 addr += size; 1271 resid -= size; 1272 } 1273 1274 return (error); 1275 } 1276 1277 /*ARGSUSED*/ 1278 int 1279 zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1280 { 1281 minor_t minor = getminor(dev); 1282 zvol_state_t *zv; 1283 uint64_t volsize; 1284 rl_t *rl; 1285 int error = 0; 1286 1287 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1288 if (zv == NULL) 1289 return (ENXIO); 1290 1291 volsize = zv->zv_volsize; 1292 if (uio->uio_resid > 0 && 1293 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1294 return (EIO); 1295 1296 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1297 error = physio(zvol_strategy, NULL, dev, B_READ, 1298 zvol_minphys, uio); 1299 return (error); 1300 } 1301 1302 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1303 RL_READER); 1304 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1305 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1306 1307 /* don't read past the end */ 1308 if (bytes > volsize - uio->uio_loffset) 1309 bytes = volsize - uio->uio_loffset; 1310 1311 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1312 if (error) { 1313 /* convert checksum errors into IO errors */ 1314 if (error == ECKSUM) 1315 error = EIO; 1316 break; 1317 } 1318 } 1319 zfs_range_unlock(rl); 1320 return (error); 1321 } 1322 1323 /*ARGSUSED*/ 1324 int 1325 zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1326 { 1327 minor_t minor = getminor(dev); 1328 zvol_state_t *zv; 1329 uint64_t volsize; 1330 rl_t *rl; 1331 int error = 0; 1332 boolean_t sync; 1333 1334 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1335 if (zv == NULL) 1336 return (ENXIO); 1337 1338 volsize = zv->zv_volsize; 1339 if (uio->uio_resid > 0 && 1340 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1341 return (EIO); 1342 1343 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1344 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1345 zvol_minphys, uio); 1346 return (error); 1347 } 1348 1349 sync = !(zv->zv_flags & ZVOL_WCE) || 1350 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1351 1352 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1353 RL_WRITER); 1354 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1355 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1356 uint64_t off = uio->uio_loffset; 1357 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1358 1359 if (bytes > volsize - off) /* don't write past the end */ 1360 bytes = volsize - off; 1361 1362 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1363 error = dmu_tx_assign(tx, TXG_WAIT); 1364 if (error) { 1365 dmu_tx_abort(tx); 1366 break; 1367 } 1368 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1369 if (error == 0) 1370 zvol_log_write(zv, tx, off, bytes, sync); 1371 dmu_tx_commit(tx); 1372 1373 if (error) 1374 break; 1375 } 1376 zfs_range_unlock(rl); 1377 if (sync) 1378 zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ); 1379 return (error); 1380 } 1381 1382 int 1383 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1384 { 1385 struct uuid uuid = EFI_RESERVED; 1386 efi_gpe_t gpe = { 0 }; 1387 uint32_t crc; 1388 dk_efi_t efi; 1389 int length; 1390 char *ptr; 1391 1392 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1393 return (EFAULT); 1394 ptr = (char *)(uintptr_t)efi.dki_data_64; 1395 length = efi.dki_length; 1396 /* 1397 * Some clients may attempt to request a PMBR for the 1398 * zvol. Currently this interface will return EINVAL to 1399 * such requests. These requests could be supported by 1400 * adding a check for lba == 0 and consing up an appropriate 1401 * PMBR. 1402 */ 1403 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1404 return (EINVAL); 1405 1406 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1407 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1408 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1409 1410 if (efi.dki_lba == 1) { 1411 efi_gpt_t gpt = { 0 }; 1412 1413 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1414 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1415 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1416 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1417 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1418 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1419 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1420 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1421 gpt.efi_gpt_SizeOfPartitionEntry = 1422 LE_32(sizeof (efi_gpe_t)); 1423 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1424 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1425 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1426 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1427 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1428 flag)) 1429 return (EFAULT); 1430 ptr += sizeof (gpt); 1431 length -= sizeof (gpt); 1432 } 1433 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1434 length), flag)) 1435 return (EFAULT); 1436 return (0); 1437 } 1438 1439 /* 1440 * BEGIN entry points to allow external callers access to the volume. 1441 */ 1442 /* 1443 * Return the volume parameters needed for access from an external caller. 1444 * These values are invariant as long as the volume is held open. 1445 */ 1446 int 1447 zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1448 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1449 void **rl_hdl, void **bonus_hdl) 1450 { 1451 zvol_state_t *zv; 1452 1453 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1454 if (zv == NULL) 1455 return (ENXIO); 1456 if (zv->zv_flags & ZVOL_DUMPIFIED) 1457 return (ENXIO); 1458 1459 ASSERT(blksize && max_xfer_len && minor_hdl && 1460 objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 1461 1462 *blksize = zv->zv_volblocksize; 1463 *max_xfer_len = (uint64_t)zvol_maxphys; 1464 *minor_hdl = zv; 1465 *objset_hdl = zv->zv_objset; 1466 *zil_hdl = zv->zv_zilog; 1467 *rl_hdl = &zv->zv_znode; 1468 *bonus_hdl = zv->zv_dbuf; 1469 return (0); 1470 } 1471 1472 /* 1473 * Return the current volume size to an external caller. 1474 * The size can change while the volume is open. 1475 */ 1476 uint64_t 1477 zvol_get_volume_size(void *minor_hdl) 1478 { 1479 zvol_state_t *zv = minor_hdl; 1480 1481 return (zv->zv_volsize); 1482 } 1483 1484 /* 1485 * Return the current WCE setting to an external caller. 1486 * The WCE setting can change while the volume is open. 1487 */ 1488 int 1489 zvol_get_volume_wce(void *minor_hdl) 1490 { 1491 zvol_state_t *zv = minor_hdl; 1492 1493 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 1494 } 1495 1496 /* 1497 * Entry point for external callers to zvol_log_write 1498 */ 1499 void 1500 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 1501 boolean_t sync) 1502 { 1503 zvol_state_t *zv = minor_hdl; 1504 1505 zvol_log_write(zv, tx, off, resid, sync); 1506 } 1507 /* 1508 * END entry points to allow external callers access to the volume. 1509 */ 1510 1511 /* 1512 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1513 */ 1514 /*ARGSUSED*/ 1515 int 1516 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1517 { 1518 zvol_state_t *zv; 1519 struct dk_cinfo dki; 1520 struct dk_minfo dkm; 1521 struct dk_callback *dkc; 1522 int error = 0; 1523 rl_t *rl; 1524 1525 mutex_enter(&zfsdev_state_lock); 1526 1527 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 1528 1529 if (zv == NULL) { 1530 mutex_exit(&zfsdev_state_lock); 1531 return (ENXIO); 1532 } 1533 ASSERT(zv->zv_total_opens > 0); 1534 1535 switch (cmd) { 1536 1537 case DKIOCINFO: 1538 bzero(&dki, sizeof (dki)); 1539 (void) strcpy(dki.dki_cname, "zvol"); 1540 (void) strcpy(dki.dki_dname, "zvol"); 1541 dki.dki_ctype = DKC_UNKNOWN; 1542 dki.dki_unit = getminor(dev); 1543 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); 1544 mutex_exit(&zfsdev_state_lock); 1545 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1546 error = EFAULT; 1547 return (error); 1548 1549 case DKIOCGMEDIAINFO: 1550 bzero(&dkm, sizeof (dkm)); 1551 dkm.dki_lbsize = 1U << zv->zv_min_bs; 1552 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1553 dkm.dki_media_type = DK_UNKNOWN; 1554 mutex_exit(&zfsdev_state_lock); 1555 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1556 error = EFAULT; 1557 return (error); 1558 1559 case DKIOCGETEFI: 1560 { 1561 uint64_t vs = zv->zv_volsize; 1562 uint8_t bs = zv->zv_min_bs; 1563 1564 mutex_exit(&zfsdev_state_lock); 1565 error = zvol_getefi((void *)arg, flag, vs, bs); 1566 return (error); 1567 } 1568 1569 case DKIOCFLUSHWRITECACHE: 1570 dkc = (struct dk_callback *)arg; 1571 mutex_exit(&zfsdev_state_lock); 1572 zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ); 1573 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1574 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1575 error = 0; 1576 } 1577 return (error); 1578 1579 case DKIOCGETWCE: 1580 { 1581 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1582 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1583 flag)) 1584 error = EFAULT; 1585 break; 1586 } 1587 case DKIOCSETWCE: 1588 { 1589 int wce; 1590 if (ddi_copyin((void *)arg, &wce, sizeof (int), 1591 flag)) { 1592 error = EFAULT; 1593 break; 1594 } 1595 if (wce) { 1596 zv->zv_flags |= ZVOL_WCE; 1597 mutex_exit(&zfsdev_state_lock); 1598 } else { 1599 zv->zv_flags &= ~ZVOL_WCE; 1600 mutex_exit(&zfsdev_state_lock); 1601 zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ); 1602 } 1603 return (0); 1604 } 1605 1606 case DKIOCGGEOM: 1607 case DKIOCGVTOC: 1608 /* 1609 * commands using these (like prtvtoc) expect ENOTSUP 1610 * since we're emulating an EFI label 1611 */ 1612 error = ENOTSUP; 1613 break; 1614 1615 case DKIOCDUMPINIT: 1616 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1617 RL_WRITER); 1618 error = zvol_dumpify(zv); 1619 zfs_range_unlock(rl); 1620 break; 1621 1622 case DKIOCDUMPFINI: 1623 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 1624 break; 1625 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1626 RL_WRITER); 1627 error = zvol_dump_fini(zv); 1628 zfs_range_unlock(rl); 1629 break; 1630 1631 default: 1632 error = ENOTTY; 1633 break; 1634 1635 } 1636 mutex_exit(&zfsdev_state_lock); 1637 return (error); 1638 } 1639 1640 int 1641 zvol_busy(void) 1642 { 1643 return (zvol_minors != 0); 1644 } 1645 1646 void 1647 zvol_init(void) 1648 { 1649 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 1650 1) == 0); 1651 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL); 1652 } 1653 1654 void 1655 zvol_fini(void) 1656 { 1657 mutex_destroy(&zfsdev_state_lock); 1658 ddi_soft_state_fini(&zfsdev_state); 1659 } 1660 1661 static int 1662 zvol_dump_init(zvol_state_t *zv, boolean_t resize) 1663 { 1664 dmu_tx_t *tx; 1665 int error = 0; 1666 objset_t *os = zv->zv_objset; 1667 nvlist_t *nv = NULL; 1668 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 1669 1670 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 1671 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 1672 DMU_OBJECT_END); 1673 /* wait for dmu_free_long_range to actually free the blocks */ 1674 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1675 1676 tx = dmu_tx_create(os); 1677 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1678 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1679 error = dmu_tx_assign(tx, TXG_WAIT); 1680 if (error) { 1681 dmu_tx_abort(tx); 1682 return (error); 1683 } 1684 1685 /* 1686 * If we are resizing the dump device then we only need to 1687 * update the refreservation to match the newly updated 1688 * zvolsize. Otherwise, we save off the original state of the 1689 * zvol so that we can restore them if the zvol is ever undumpified. 1690 */ 1691 if (resize) { 1692 error = zap_update(os, ZVOL_ZAP_OBJ, 1693 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1694 &zv->zv_volsize, tx); 1695 } else { 1696 uint64_t checksum, compress, refresrv, vbs, dedup; 1697 1698 error = dsl_prop_get_integer(zv->zv_name, 1699 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 1700 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1701 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL); 1702 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1703 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL); 1704 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1705 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL); 1706 if (version >= SPA_VERSION_DEDUP) { 1707 error = error ? error : 1708 dsl_prop_get_integer(zv->zv_name, 1709 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 1710 } 1711 1712 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1713 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 1714 &compress, tx); 1715 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1716 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx); 1717 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1718 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1719 &refresrv, tx); 1720 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1721 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 1722 &vbs, tx); 1723 error = error ? error : dmu_object_set_blocksize( 1724 os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx); 1725 if (version >= SPA_VERSION_DEDUP) { 1726 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1727 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 1728 &dedup, tx); 1729 } 1730 if (error == 0) 1731 zv->zv_volblocksize = SPA_MAXBLOCKSIZE; 1732 } 1733 dmu_tx_commit(tx); 1734 1735 /* 1736 * We only need update the zvol's property if we are initializing 1737 * the dump area for the first time. 1738 */ 1739 if (!resize) { 1740 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1741 VERIFY(nvlist_add_uint64(nv, 1742 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 1743 VERIFY(nvlist_add_uint64(nv, 1744 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 1745 ZIO_COMPRESS_OFF) == 0); 1746 VERIFY(nvlist_add_uint64(nv, 1747 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 1748 ZIO_CHECKSUM_OFF) == 0); 1749 if (version >= SPA_VERSION_DEDUP) { 1750 VERIFY(nvlist_add_uint64(nv, 1751 zfs_prop_to_name(ZFS_PROP_DEDUP), 1752 ZIO_CHECKSUM_OFF) == 0); 1753 } 1754 1755 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 1756 nv, NULL); 1757 nvlist_free(nv); 1758 1759 if (error) 1760 return (error); 1761 } 1762 1763 /* Allocate the space for the dump */ 1764 error = zvol_prealloc(zv); 1765 return (error); 1766 } 1767 1768 static int 1769 zvol_dumpify(zvol_state_t *zv) 1770 { 1771 int error = 0; 1772 uint64_t dumpsize = 0; 1773 dmu_tx_t *tx; 1774 objset_t *os = zv->zv_objset; 1775 1776 if (zv->zv_flags & ZVOL_RDONLY) 1777 return (EROFS); 1778 1779 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 1780 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 1781 boolean_t resize = (dumpsize > 0) ? B_TRUE : B_FALSE; 1782 1783 if ((error = zvol_dump_init(zv, resize)) != 0) { 1784 (void) zvol_dump_fini(zv); 1785 return (error); 1786 } 1787 } 1788 1789 /* 1790 * Build up our lba mapping. 1791 */ 1792 error = zvol_get_lbas(zv); 1793 if (error) { 1794 (void) zvol_dump_fini(zv); 1795 return (error); 1796 } 1797 1798 tx = dmu_tx_create(os); 1799 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1800 error = dmu_tx_assign(tx, TXG_WAIT); 1801 if (error) { 1802 dmu_tx_abort(tx); 1803 (void) zvol_dump_fini(zv); 1804 return (error); 1805 } 1806 1807 zv->zv_flags |= ZVOL_DUMPIFIED; 1808 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 1809 &zv->zv_volsize, tx); 1810 dmu_tx_commit(tx); 1811 1812 if (error) { 1813 (void) zvol_dump_fini(zv); 1814 return (error); 1815 } 1816 1817 txg_wait_synced(dmu_objset_pool(os), 0); 1818 return (0); 1819 } 1820 1821 static int 1822 zvol_dump_fini(zvol_state_t *zv) 1823 { 1824 dmu_tx_t *tx; 1825 objset_t *os = zv->zv_objset; 1826 nvlist_t *nv; 1827 int error = 0; 1828 uint64_t checksum, compress, refresrv, vbs, dedup; 1829 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 1830 1831 /* 1832 * Attempt to restore the zvol back to its pre-dumpified state. 1833 * This is a best-effort attempt as it's possible that not all 1834 * of these properties were initialized during the dumpify process 1835 * (i.e. error during zvol_dump_init). 1836 */ 1837 1838 tx = dmu_tx_create(os); 1839 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1840 error = dmu_tx_assign(tx, TXG_WAIT); 1841 if (error) { 1842 dmu_tx_abort(tx); 1843 return (error); 1844 } 1845 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 1846 dmu_tx_commit(tx); 1847 1848 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1849 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 1850 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1851 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 1852 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1853 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 1854 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1855 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 1856 1857 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1858 (void) nvlist_add_uint64(nv, 1859 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 1860 (void) nvlist_add_uint64(nv, 1861 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 1862 (void) nvlist_add_uint64(nv, 1863 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 1864 if (version >= SPA_VERSION_DEDUP && 1865 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1866 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 1867 (void) nvlist_add_uint64(nv, 1868 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 1869 } 1870 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 1871 nv, NULL); 1872 nvlist_free(nv); 1873 1874 zvol_free_extents(zv); 1875 zv->zv_flags &= ~ZVOL_DUMPIFIED; 1876 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 1877 /* wait for dmu_free_long_range to actually free the blocks */ 1878 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1879 tx = dmu_tx_create(os); 1880 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1881 error = dmu_tx_assign(tx, TXG_WAIT); 1882 if (error) { 1883 dmu_tx_abort(tx); 1884 return (error); 1885 } 1886 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 1887 zv->zv_volblocksize = vbs; 1888 dmu_tx_commit(tx); 1889 1890 return (0); 1891 } 1892