zvol.c revision 241297
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 * Copyright (c) 2006-2010 Pawel Jakub Dawidek <pjd@FreeBSD.org> 25 * All rights reserved. 26 */ 27 28/* Portions Copyright 2010 Robert Milkowski */ 29/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 30 31/* 32 * ZFS volume emulation driver. 33 * 34 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 35 * Volumes are accessed through the symbolic links named: 36 * 37 * /dev/zvol/dsk/<pool_name>/<dataset_name> 38 * /dev/zvol/rdsk/<pool_name>/<dataset_name> 39 * 40 * These links are created by the /dev filesystem (sdev_zvolops.c). 41 * Volumes are persistent through reboot. No user command needs to be 42 * run before opening and using a device. 43 * 44 * FreeBSD notes. 45 * On FreeBSD ZVOLs are simply GEOM providers like any other storage device 46 * in the system. 47 */ 48 49#include <sys/types.h> 50#include <sys/param.h> 51#include <sys/kernel.h> 52#include <sys/errno.h> 53#include <sys/uio.h> 54#include <sys/bio.h> 55#include <sys/buf.h> 56#include <sys/kmem.h> 57#include <sys/conf.h> 58#include <sys/cmn_err.h> 59#include <sys/stat.h> 60#include <sys/zap.h> 61#include <sys/spa.h> 62#include <sys/zio.h> 63#include <sys/dmu_traverse.h> 64#include <sys/dnode.h> 65#include <sys/dsl_dataset.h> 66#include <sys/dsl_prop.h> 67#include <sys/dkio.h> 68#include <sys/byteorder.h> 69#include <sys/sunddi.h> 70#include <sys/dirent.h> 71#include <sys/policy.h> 72#include <sys/fs/zfs.h> 73#include <sys/zfs_ioctl.h> 74#include <sys/zil.h> 75#include <sys/refcount.h> 76#include <sys/zfs_znode.h> 77#include <sys/zfs_rlock.h> 78#include <sys/vdev_impl.h> 79#include <sys/zvol.h> 80#include <sys/zil_impl.h> 81#include <geom/geom.h> 82 83#include "zfs_namecheck.h" 84 85struct g_class zfs_zvol_class = { 86 .name = "ZFS::ZVOL", 87 .version = G_VERSION, 88}; 89 90DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol); 91 92void *zfsdev_state; 93static char *zvol_tag = "zvol_tag"; 94 95#define ZVOL_DUMPSIZE "dumpsize" 96 97/* 98 * The spa_namespace_lock protects the zfsdev_state structure from being 99 * modified while it's being used, e.g. an open that comes in before a 100 * create finishes. It also protects temporary opens of the dataset so that, 101 * e.g., an open doesn't get a spurious EBUSY. 102 */ 103static uint32_t zvol_minors; 104 105typedef struct zvol_extent { 106 list_node_t ze_node; 107 dva_t ze_dva; /* dva associated with this extent */ 108 uint64_t ze_nblks; /* number of blocks in extent */ 109} zvol_extent_t; 110 111/* 112 * The in-core state of each volume. 113 */ 114typedef struct zvol_state { 115 char zv_name[MAXPATHLEN]; /* pool/dd name */ 116 uint64_t zv_volsize; /* amount of space we advertise */ 117 uint64_t zv_volblocksize; /* volume block size */ 118 struct g_provider *zv_provider; /* GEOM provider */ 119 uint8_t zv_min_bs; /* minimum addressable block shift */ 120 uint8_t zv_flags; /* readonly, dumpified, etc. */ 121 objset_t *zv_objset; /* objset handle */ 122 uint32_t zv_total_opens; /* total open count */ 123 zilog_t *zv_zilog; /* ZIL handle */ 124 list_t zv_extents; /* List of extents for dump */ 125 znode_t zv_znode; /* for range locking */ 126 dmu_buf_t *zv_dbuf; /* bonus handle */ 127 int zv_state; 128 struct bio_queue_head zv_queue; 129 struct mtx zv_queue_mtx; /* zv_queue mutex */ 130} zvol_state_t; 131 132/* 133 * zvol specific flags 134 */ 135#define ZVOL_RDONLY 0x1 136#define ZVOL_DUMPIFIED 0x2 137#define ZVOL_EXCL 0x4 138#define ZVOL_WCE 0x8 139 140/* 141 * zvol maximum transfer in one DMU tx. 142 */ 143int zvol_maxphys = DMU_MAX_ACCESS/2; 144 145extern int zfs_set_prop_nvlist(const char *, zprop_source_t, 146 nvlist_t *, nvlist_t **); 147static int zvol_remove_zv(zvol_state_t *); 148static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio); 149static int zvol_dumpify(zvol_state_t *zv); 150static int zvol_dump_fini(zvol_state_t *zv); 151static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); 152 153static zvol_state_t *zvol_geom_create(const char *name); 154static void zvol_geom_run(zvol_state_t *zv); 155static void zvol_geom_destroy(zvol_state_t *zv); 156static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace); 157static void zvol_geom_start(struct bio *bp); 158static void zvol_geom_worker(void *arg); 159 160static void 161zvol_size_changed(zvol_state_t *zv) 162{ 163#ifdef sun 164 dev_t dev = makedevice(maj, min); 165 166 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 167 "Size", volsize) == DDI_SUCCESS); 168 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 169 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS); 170 171 /* Notify specfs to invalidate the cached size */ 172 spec_size_invalidate(dev, VBLK); 173 spec_size_invalidate(dev, VCHR); 174#else /* !sun */ 175 struct g_provider *pp; 176 177 pp = zv->zv_provider; 178 if (pp == NULL) 179 return; 180 g_topology_lock(); 181 g_resize_provider(pp, zv->zv_volsize); 182 g_topology_unlock(); 183#endif /* !sun */ 184} 185 186int 187zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 188{ 189 if (volsize == 0) 190 return (EINVAL); 191 192 if (volsize % blocksize != 0) 193 return (EINVAL); 194 195#ifdef _ILP32 196 if (volsize - 1 > SPEC_MAXOFFSET_T) 197 return (EOVERFLOW); 198#endif 199 return (0); 200} 201 202int 203zvol_check_volblocksize(uint64_t volblocksize) 204{ 205 if (volblocksize < SPA_MINBLOCKSIZE || 206 volblocksize > SPA_MAXBLOCKSIZE || 207 !ISP2(volblocksize)) 208 return (EDOM); 209 210 return (0); 211} 212 213int 214zvol_get_stats(objset_t *os, nvlist_t *nv) 215{ 216 int error; 217 dmu_object_info_t doi; 218 uint64_t val; 219 220 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 221 if (error) 222 return (error); 223 224 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 225 226 error = dmu_object_info(os, ZVOL_OBJ, &doi); 227 228 if (error == 0) { 229 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 230 doi.doi_data_block_size); 231 } 232 233 return (error); 234} 235 236static zvol_state_t * 237zvol_minor_lookup(const char *name) 238{ 239 struct g_provider *pp; 240 struct g_geom *gp; 241 zvol_state_t *zv = NULL; 242 243 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 244 245 g_topology_lock(); 246 LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) { 247 pp = LIST_FIRST(&gp->provider); 248 if (pp == NULL) 249 continue; 250 zv = pp->private; 251 if (zv == NULL) 252 continue; 253 if (strcmp(zv->zv_name, name) == 0) 254 break; 255 } 256 g_topology_unlock(); 257 258 return (gp != NULL ? zv : NULL); 259} 260 261/* extent mapping arg */ 262struct maparg { 263 zvol_state_t *ma_zv; 264 uint64_t ma_blks; 265}; 266 267/*ARGSUSED*/ 268static int 269zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf, 270 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 271{ 272 struct maparg *ma = arg; 273 zvol_extent_t *ze; 274 int bs = ma->ma_zv->zv_volblocksize; 275 276 if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 277 return (0); 278 279 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 280 ma->ma_blks++; 281 282 /* Abort immediately if we have encountered gang blocks */ 283 if (BP_IS_GANG(bp)) 284 return (EFRAGS); 285 286 /* 287 * See if the block is at the end of the previous extent. 288 */ 289 ze = list_tail(&ma->ma_zv->zv_extents); 290 if (ze && 291 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 292 DVA_GET_OFFSET(BP_IDENTITY(bp)) == 293 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 294 ze->ze_nblks++; 295 return (0); 296 } 297 298 dprintf_bp(bp, "%s", "next blkptr:"); 299 300 /* start a new extent */ 301 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 302 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 303 ze->ze_nblks = 1; 304 list_insert_tail(&ma->ma_zv->zv_extents, ze); 305 return (0); 306} 307 308static void 309zvol_free_extents(zvol_state_t *zv) 310{ 311 zvol_extent_t *ze; 312 313 while (ze = list_head(&zv->zv_extents)) { 314 list_remove(&zv->zv_extents, ze); 315 kmem_free(ze, sizeof (zvol_extent_t)); 316 } 317} 318 319static int 320zvol_get_lbas(zvol_state_t *zv) 321{ 322 objset_t *os = zv->zv_objset; 323 struct maparg ma; 324 int err; 325 326 ma.ma_zv = zv; 327 ma.ma_blks = 0; 328 zvol_free_extents(zv); 329 330 /* commit any in-flight changes before traversing the dataset */ 331 txg_wait_synced(dmu_objset_pool(os), 0); 332 err = traverse_dataset(dmu_objset_ds(os), 0, 333 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 334 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 335 zvol_free_extents(zv); 336 return (err ? err : EIO); 337 } 338 339 return (0); 340} 341 342/* ARGSUSED */ 343void 344zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 345{ 346 zfs_creat_t *zct = arg; 347 nvlist_t *nvprops = zct->zct_props; 348 int error; 349 uint64_t volblocksize, volsize; 350 351 VERIFY(nvlist_lookup_uint64(nvprops, 352 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 353 if (nvlist_lookup_uint64(nvprops, 354 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 355 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 356 357 /* 358 * These properties must be removed from the list so the generic 359 * property setting step won't apply to them. 360 */ 361 VERIFY(nvlist_remove_all(nvprops, 362 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 363 (void) nvlist_remove_all(nvprops, 364 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 365 366 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 367 DMU_OT_NONE, 0, tx); 368 ASSERT(error == 0); 369 370 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 371 DMU_OT_NONE, 0, tx); 372 ASSERT(error == 0); 373 374 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 375 ASSERT(error == 0); 376} 377 378/* 379 * Replay a TX_WRITE ZIL transaction that didn't get committed 380 * after a system failure 381 */ 382static int 383zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 384{ 385 objset_t *os = zv->zv_objset; 386 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 387 uint64_t offset, length; 388 dmu_tx_t *tx; 389 int error; 390 391 if (byteswap) 392 byteswap_uint64_array(lr, sizeof (*lr)); 393 394 offset = lr->lr_offset; 395 length = lr->lr_length; 396 397 /* If it's a dmu_sync() block, write the whole block */ 398 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 399 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 400 if (length < blocksize) { 401 offset -= offset % blocksize; 402 length = blocksize; 403 } 404 } 405 406 tx = dmu_tx_create(os); 407 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 408 error = dmu_tx_assign(tx, TXG_WAIT); 409 if (error) { 410 dmu_tx_abort(tx); 411 } else { 412 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 413 dmu_tx_commit(tx); 414 } 415 416 return (error); 417} 418 419/* ARGSUSED */ 420static int 421zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 422{ 423 return (ENOTSUP); 424} 425 426/* 427 * Callback vectors for replaying records. 428 * Only TX_WRITE is needed for zvol. 429 */ 430zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 431 zvol_replay_err, /* 0 no such transaction type */ 432 zvol_replay_err, /* TX_CREATE */ 433 zvol_replay_err, /* TX_MKDIR */ 434 zvol_replay_err, /* TX_MKXATTR */ 435 zvol_replay_err, /* TX_SYMLINK */ 436 zvol_replay_err, /* TX_REMOVE */ 437 zvol_replay_err, /* TX_RMDIR */ 438 zvol_replay_err, /* TX_LINK */ 439 zvol_replay_err, /* TX_RENAME */ 440 zvol_replay_write, /* TX_WRITE */ 441 zvol_replay_err, /* TX_TRUNCATE */ 442 zvol_replay_err, /* TX_SETATTR */ 443 zvol_replay_err, /* TX_ACL */ 444 zvol_replay_err, /* TX_CREATE_ACL */ 445 zvol_replay_err, /* TX_CREATE_ATTR */ 446 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 447 zvol_replay_err, /* TX_MKDIR_ACL */ 448 zvol_replay_err, /* TX_MKDIR_ATTR */ 449 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 450 zvol_replay_err, /* TX_WRITE2 */ 451}; 452 453#ifdef sun 454int 455zvol_name2minor(const char *name, minor_t *minor) 456{ 457 zvol_state_t *zv; 458 459 mutex_enter(&spa_namespace_lock); 460 zv = zvol_minor_lookup(name); 461 if (minor && zv) 462 *minor = zv->zv_minor; 463 mutex_exit(&spa_namespace_lock); 464 return (zv ? 0 : -1); 465} 466#endif /* sun */ 467 468/* 469 * Create a minor node (plus a whole lot more) for the specified volume. 470 */ 471int 472zvol_create_minor(const char *name) 473{ 474 zfs_soft_state_t *zs; 475 zvol_state_t *zv; 476 objset_t *os; 477 dmu_object_info_t doi; 478 uint64_t volsize; 479 int error; 480 481 ZFS_LOG(1, "Creating ZVOL %s...", name); 482 483 mutex_enter(&spa_namespace_lock); 484 485 if (zvol_minor_lookup(name) != NULL) { 486 mutex_exit(&spa_namespace_lock); 487 return (EEXIST); 488 } 489 490 /* lie and say we're read-only */ 491 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os); 492 493 if (error) { 494 mutex_exit(&spa_namespace_lock); 495 return (error); 496 } 497 498#ifdef sun 499 if ((minor = zfsdev_minor_alloc()) == 0) { 500 dmu_objset_disown(os, FTAG); 501 mutex_exit(&spa_namespace_lock); 502 return (ENXIO); 503 } 504 505 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 506 dmu_objset_disown(os, FTAG); 507 mutex_exit(&spa_namespace_lock); 508 return (EAGAIN); 509 } 510 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 511 (char *)name); 512 513 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 514 515 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 516 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 517 ddi_soft_state_free(zfsdev_state, minor); 518 dmu_objset_disown(os, FTAG); 519 mutex_exit(&spa_namespace_lock); 520 return (EAGAIN); 521 } 522 523 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 524 525 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 526 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 527 ddi_remove_minor_node(zfs_dip, chrbuf); 528 ddi_soft_state_free(zfsdev_state, minor); 529 dmu_objset_disown(os, FTAG); 530 mutex_exit(&spa_namespace_lock); 531 return (EAGAIN); 532 } 533 534 zs = ddi_get_soft_state(zfsdev_state, minor); 535 zs->zss_type = ZSST_ZVOL; 536 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 537#else /* !sun */ 538 539 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 540 if (error) { 541 ASSERT(error == 0); 542 dmu_objset_disown(os, zvol_tag); 543 mutex_exit(&spa_namespace_lock); 544 return (error); 545 } 546 547 DROP_GIANT(); 548 g_topology_lock(); 549 zv = zvol_geom_create(name); 550 zv->zv_volsize = volsize; 551 zv->zv_provider->mediasize = zv->zv_volsize; 552 553#endif /* !sun */ 554 555 (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 556 zv->zv_min_bs = DEV_BSHIFT; 557 zv->zv_objset = os; 558 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) 559 zv->zv_flags |= ZVOL_RDONLY; 560 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 561 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 562 sizeof (rl_t), offsetof(rl_t, r_node)); 563 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 564 offsetof(zvol_extent_t, ze_node)); 565 /* get and cache the blocksize */ 566 error = dmu_object_info(os, ZVOL_OBJ, &doi); 567 ASSERT(error == 0); 568 zv->zv_volblocksize = doi.doi_data_block_size; 569 570 if (spa_writeable(dmu_objset_spa(os))) { 571 if (zil_replay_disable) 572 zil_destroy(dmu_objset_zil(os), B_FALSE); 573 else 574 zil_replay(os, zv, zvol_replay_vector); 575 } 576 dmu_objset_disown(os, FTAG); 577 zv->zv_objset = NULL; 578 579 zvol_minors++; 580 581 mutex_exit(&spa_namespace_lock); 582 583 zvol_geom_run(zv); 584 585 g_topology_unlock(); 586 PICKUP_GIANT(); 587 588 ZFS_LOG(1, "ZVOL %s created.", name); 589 590 return (0); 591} 592 593/* 594 * Remove minor node for the specified volume. 595 */ 596static int 597zvol_remove_zv(zvol_state_t *zv) 598{ 599#ifdef sun 600 minor_t minor = zv->zv_minor; 601#endif 602 603 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 604 if (zv->zv_total_opens != 0) 605 return (EBUSY); 606 607 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); 608 609#ifdef sun 610 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 611 ddi_remove_minor_node(zfs_dip, nmbuf); 612#endif /* sun */ 613 614 avl_destroy(&zv->zv_znode.z_range_avl); 615 mutex_destroy(&zv->zv_znode.z_range_lock); 616 617 zvol_geom_destroy(zv); 618 619 zvol_minors--; 620 return (0); 621} 622 623int 624zvol_remove_minor(const char *name) 625{ 626 zvol_state_t *zv; 627 int rc; 628 629 mutex_enter(&spa_namespace_lock); 630 if ((zv = zvol_minor_lookup(name)) == NULL) { 631 mutex_exit(&spa_namespace_lock); 632 return (ENXIO); 633 } 634 g_topology_lock(); 635 rc = zvol_remove_zv(zv); 636 g_topology_unlock(); 637 mutex_exit(&spa_namespace_lock); 638 return (rc); 639} 640 641int 642zvol_first_open(zvol_state_t *zv) 643{ 644 objset_t *os; 645 uint64_t volsize; 646 int error; 647 uint64_t readonly; 648 649 /* lie and say we're read-only */ 650 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 651 zvol_tag, &os); 652 if (error) 653 return (error); 654 655 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 656 if (error) { 657 ASSERT(error == 0); 658 dmu_objset_disown(os, zvol_tag); 659 return (error); 660 } 661 zv->zv_objset = os; 662 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 663 if (error) { 664 dmu_objset_disown(os, zvol_tag); 665 return (error); 666 } 667 zv->zv_volsize = volsize; 668 zv->zv_zilog = zil_open(os, zvol_get_data); 669 zvol_size_changed(zv); 670 671 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 672 NULL) == 0); 673 if (readonly || dmu_objset_is_snapshot(os) || 674 !spa_writeable(dmu_objset_spa(os))) 675 zv->zv_flags |= ZVOL_RDONLY; 676 else 677 zv->zv_flags &= ~ZVOL_RDONLY; 678 return (error); 679} 680 681void 682zvol_last_close(zvol_state_t *zv) 683{ 684 zil_close(zv->zv_zilog); 685 zv->zv_zilog = NULL; 686 687 dmu_buf_rele(zv->zv_dbuf, zvol_tag); 688 zv->zv_dbuf = NULL; 689 690 /* 691 * Evict cached data 692 */ 693 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && 694 !(zv->zv_flags & ZVOL_RDONLY)) 695 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 696 (void) dmu_objset_evict_dbufs(zv->zv_objset); 697 698 dmu_objset_disown(zv->zv_objset, zvol_tag); 699 zv->zv_objset = NULL; 700} 701 702#ifdef sun 703int 704zvol_prealloc(zvol_state_t *zv) 705{ 706 objset_t *os = zv->zv_objset; 707 dmu_tx_t *tx; 708 uint64_t refd, avail, usedobjs, availobjs; 709 uint64_t resid = zv->zv_volsize; 710 uint64_t off = 0; 711 712 /* Check the space usage before attempting to allocate the space */ 713 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 714 if (avail < zv->zv_volsize) 715 return (ENOSPC); 716 717 /* Free old extents if they exist */ 718 zvol_free_extents(zv); 719 720 while (resid != 0) { 721 int error; 722 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); 723 724 tx = dmu_tx_create(os); 725 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 726 error = dmu_tx_assign(tx, TXG_WAIT); 727 if (error) { 728 dmu_tx_abort(tx); 729 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 730 return (error); 731 } 732 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 733 dmu_tx_commit(tx); 734 off += bytes; 735 resid -= bytes; 736 } 737 txg_wait_synced(dmu_objset_pool(os), 0); 738 739 return (0); 740} 741#endif /* sun */ 742 743int 744zvol_update_volsize(objset_t *os, uint64_t volsize) 745{ 746 dmu_tx_t *tx; 747 int error; 748 749 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 750 751 tx = dmu_tx_create(os); 752 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 753 error = dmu_tx_assign(tx, TXG_WAIT); 754 if (error) { 755 dmu_tx_abort(tx); 756 return (error); 757 } 758 759 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 760 &volsize, tx); 761 dmu_tx_commit(tx); 762 763 if (error == 0) 764 error = dmu_free_long_range(os, 765 ZVOL_OBJ, volsize, DMU_OBJECT_END); 766 return (error); 767} 768 769void 770zvol_remove_minors(const char *name) 771{ 772 struct g_geom *gp, *gptmp; 773 struct g_provider *pp; 774 zvol_state_t *zv; 775 size_t namelen; 776 777 namelen = strlen(name); 778 779 DROP_GIANT(); 780 mutex_enter(&spa_namespace_lock); 781 g_topology_lock(); 782 783 LIST_FOREACH_SAFE(gp, &zfs_zvol_class.geom, geom, gptmp) { 784 pp = LIST_FIRST(&gp->provider); 785 if (pp == NULL) 786 continue; 787 zv = pp->private; 788 if (zv == NULL) 789 continue; 790 if (strcmp(zv->zv_name, name) == 0 || 791 (strncmp(zv->zv_name, name, namelen) == 0 && 792 zv->zv_name[namelen] == '/')) { 793 (void) zvol_remove_zv(zv); 794 } 795 } 796 797 g_topology_unlock(); 798 mutex_exit(&spa_namespace_lock); 799 PICKUP_GIANT(); 800} 801 802int 803zvol_set_volsize(const char *name, major_t maj, uint64_t volsize) 804{ 805 zvol_state_t *zv = NULL; 806 objset_t *os; 807 int error; 808 dmu_object_info_t doi; 809 uint64_t old_volsize = 0ULL; 810 uint64_t readonly; 811 812 mutex_enter(&spa_namespace_lock); 813 zv = zvol_minor_lookup(name); 814 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 815 mutex_exit(&spa_namespace_lock); 816 return (error); 817 } 818 819 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 820 (error = zvol_check_volsize(volsize, 821 doi.doi_data_block_size)) != 0) 822 goto out; 823 824 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, 825 NULL) == 0); 826 if (readonly) { 827 error = EROFS; 828 goto out; 829 } 830 831 error = zvol_update_volsize(os, volsize); 832 /* 833 * Reinitialize the dump area to the new size. If we 834 * failed to resize the dump area then restore it back to 835 * its original size. 836 */ 837 if (zv && error == 0) { 838#ifdef ZVOL_DUMP 839 if (zv->zv_flags & ZVOL_DUMPIFIED) { 840 old_volsize = zv->zv_volsize; 841 zv->zv_volsize = volsize; 842 if ((error = zvol_dumpify(zv)) != 0 || 843 (error = dumpvp_resize()) != 0) { 844 (void) zvol_update_volsize(os, old_volsize); 845 zv->zv_volsize = old_volsize; 846 error = zvol_dumpify(zv); 847 } 848 } 849#endif /* ZVOL_DUMP */ 850 if (error == 0) { 851 zv->zv_volsize = volsize; 852 zvol_size_changed(zv); 853 } 854 } 855 856#ifdef sun 857 /* 858 * Generate a LUN expansion event. 859 */ 860 if (zv && error == 0) { 861 sysevent_id_t eid; 862 nvlist_t *attr; 863 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 864 865 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 866 zv->zv_minor); 867 868 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 869 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 870 871 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 872 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 873 874 nvlist_free(attr); 875 kmem_free(physpath, MAXPATHLEN); 876 } 877#endif /* sun */ 878 879out: 880 dmu_objset_rele(os, FTAG); 881 882 mutex_exit(&spa_namespace_lock); 883 884 return (error); 885} 886 887/*ARGSUSED*/ 888static int 889zvol_open(struct g_provider *pp, int flag, int count) 890{ 891 zvol_state_t *zv; 892 int err = 0; 893 boolean_t locked = B_FALSE; 894 895 /* 896 * Protect against recursively entering spa_namespace_lock 897 * when spa_open() is used for a pool on a (local) ZVOL(s). 898 * This is needed since we replaced upstream zfsdev_state_lock 899 * with spa_namespace_lock in the ZVOL code. 900 * We are using the same trick as spa_open(). 901 * Note that calls in zvol_first_open which need to resolve 902 * pool name to a spa object will enter spa_open() 903 * recursively, but that function already has all the 904 * necessary protection. 905 */ 906 if (!MUTEX_HELD(&spa_namespace_lock)) { 907 mutex_enter(&spa_namespace_lock); 908 locked = B_TRUE; 909 } 910 911 zv = pp->private; 912 if (zv == NULL) { 913 if (locked) 914 mutex_exit(&spa_namespace_lock); 915 return (ENXIO); 916 } 917 918 if (zv->zv_total_opens == 0) 919 err = zvol_first_open(zv); 920 if (err) { 921 if (locked) 922 mutex_exit(&spa_namespace_lock); 923 return (err); 924 } 925 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 926 err = EROFS; 927 goto out; 928 } 929 if (zv->zv_flags & ZVOL_EXCL) { 930 err = EBUSY; 931 goto out; 932 } 933#ifdef FEXCL 934 if (flag & FEXCL) { 935 if (zv->zv_total_opens != 0) { 936 err = EBUSY; 937 goto out; 938 } 939 zv->zv_flags |= ZVOL_EXCL; 940 } 941#endif 942 943 zv->zv_total_opens += count; 944 if (locked) 945 mutex_exit(&spa_namespace_lock); 946 947 return (err); 948out: 949 if (zv->zv_total_opens == 0) 950 zvol_last_close(zv); 951 if (locked) 952 mutex_exit(&spa_namespace_lock); 953 return (err); 954} 955 956/*ARGSUSED*/ 957static int 958zvol_close(struct g_provider *pp, int flag, int count) 959{ 960 zvol_state_t *zv; 961 int error = 0; 962 boolean_t locked = B_FALSE; 963 964 /* See comment in zvol_open(). */ 965 if (!MUTEX_HELD(&spa_namespace_lock)) { 966 mutex_enter(&spa_namespace_lock); 967 locked = B_TRUE; 968 } 969 970 zv = pp->private; 971 if (zv == NULL) { 972 if (locked) 973 mutex_exit(&spa_namespace_lock); 974 return (ENXIO); 975 } 976 977 if (zv->zv_flags & ZVOL_EXCL) { 978 ASSERT(zv->zv_total_opens == 1); 979 zv->zv_flags &= ~ZVOL_EXCL; 980 } 981 982 /* 983 * If the open count is zero, this is a spurious close. 984 * That indicates a bug in the kernel / DDI framework. 985 */ 986 ASSERT(zv->zv_total_opens != 0); 987 988 /* 989 * You may get multiple opens, but only one close. 990 */ 991 zv->zv_total_opens -= count; 992 993 if (zv->zv_total_opens == 0) 994 zvol_last_close(zv); 995 996 if (locked) 997 mutex_exit(&spa_namespace_lock); 998 return (error); 999} 1000 1001static void 1002zvol_get_done(zgd_t *zgd, int error) 1003{ 1004 if (zgd->zgd_db) 1005 dmu_buf_rele(zgd->zgd_db, zgd); 1006 1007 zfs_range_unlock(zgd->zgd_rl); 1008 1009 if (error == 0 && zgd->zgd_bp) 1010 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1011 1012 kmem_free(zgd, sizeof (zgd_t)); 1013} 1014 1015/* 1016 * Get data to generate a TX_WRITE intent log record. 1017 */ 1018static int 1019zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1020{ 1021 zvol_state_t *zv = arg; 1022 objset_t *os = zv->zv_objset; 1023 uint64_t object = ZVOL_OBJ; 1024 uint64_t offset = lr->lr_offset; 1025 uint64_t size = lr->lr_length; /* length of user data */ 1026 blkptr_t *bp = &lr->lr_blkptr; 1027 dmu_buf_t *db; 1028 zgd_t *zgd; 1029 int error; 1030 1031 ASSERT(zio != NULL); 1032 ASSERT(size != 0); 1033 1034 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1035 zgd->zgd_zilog = zv->zv_zilog; 1036 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 1037 1038 /* 1039 * Write records come in two flavors: immediate and indirect. 1040 * For small writes it's cheaper to store the data with the 1041 * log record (immediate); for large writes it's cheaper to 1042 * sync the data and get a pointer to it (indirect) so that 1043 * we don't have to write the data twice. 1044 */ 1045 if (buf != NULL) { /* immediate write */ 1046 error = dmu_read(os, object, offset, size, buf, 1047 DMU_READ_NO_PREFETCH); 1048 } else { 1049 size = zv->zv_volblocksize; 1050 offset = P2ALIGN(offset, size); 1051 error = dmu_buf_hold(os, object, offset, zgd, &db, 1052 DMU_READ_NO_PREFETCH); 1053 if (error == 0) { 1054 zgd->zgd_db = db; 1055 zgd->zgd_bp = bp; 1056 1057 ASSERT(db->db_offset == offset); 1058 ASSERT(db->db_size == size); 1059 1060 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1061 zvol_get_done, zgd); 1062 1063 if (error == 0) 1064 return (0); 1065 } 1066 } 1067 1068 zvol_get_done(zgd, error); 1069 1070 return (error); 1071} 1072 1073/* 1074 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 1075 * 1076 * We store data in the log buffers if it's small enough. 1077 * Otherwise we will later flush the data out via dmu_sync(). 1078 */ 1079ssize_t zvol_immediate_write_sz = 32768; 1080 1081static void 1082zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1083 boolean_t sync) 1084{ 1085 uint32_t blocksize = zv->zv_volblocksize; 1086 zilog_t *zilog = zv->zv_zilog; 1087 boolean_t slogging; 1088 ssize_t immediate_write_sz; 1089 1090 if (zil_replaying(zilog, tx)) 1091 return; 1092 1093 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1094 ? 0 : zvol_immediate_write_sz; 1095 1096 slogging = spa_has_slogs(zilog->zl_spa) && 1097 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 1098 1099 while (resid) { 1100 itx_t *itx; 1101 lr_write_t *lr; 1102 ssize_t len; 1103 itx_wr_state_t write_state; 1104 1105 /* 1106 * Unlike zfs_log_write() we can be called with 1107 * upto DMU_MAX_ACCESS/2 (5MB) writes. 1108 */ 1109 if (blocksize > immediate_write_sz && !slogging && 1110 resid >= blocksize && off % blocksize == 0) { 1111 write_state = WR_INDIRECT; /* uses dmu_sync */ 1112 len = blocksize; 1113 } else if (sync) { 1114 write_state = WR_COPIED; 1115 len = MIN(ZIL_MAX_LOG_DATA, resid); 1116 } else { 1117 write_state = WR_NEED_COPY; 1118 len = MIN(ZIL_MAX_LOG_DATA, resid); 1119 } 1120 1121 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1122 (write_state == WR_COPIED ? len : 0)); 1123 lr = (lr_write_t *)&itx->itx_lr; 1124 if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 1125 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1126 zil_itx_destroy(itx); 1127 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1128 lr = (lr_write_t *)&itx->itx_lr; 1129 write_state = WR_NEED_COPY; 1130 } 1131 1132 itx->itx_wr_state = write_state; 1133 if (write_state == WR_NEED_COPY) 1134 itx->itx_sod += len; 1135 lr->lr_foid = ZVOL_OBJ; 1136 lr->lr_offset = off; 1137 lr->lr_length = len; 1138 lr->lr_blkoff = 0; 1139 BP_ZERO(&lr->lr_blkptr); 1140 1141 itx->itx_private = zv; 1142 itx->itx_sync = sync; 1143 1144 zil_itx_assign(zilog, itx, tx); 1145 1146 off += len; 1147 resid -= len; 1148 } 1149} 1150 1151#ifdef sun 1152static int 1153zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t size, 1154 boolean_t doread, boolean_t isdump) 1155{ 1156 vdev_disk_t *dvd; 1157 int c; 1158 int numerrors = 0; 1159 1160 for (c = 0; c < vd->vdev_children; c++) { 1161 ASSERT(vd->vdev_ops == &vdev_mirror_ops || 1162 vd->vdev_ops == &vdev_replacing_ops || 1163 vd->vdev_ops == &vdev_spare_ops); 1164 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1165 addr, offset, size, doread, isdump); 1166 if (err != 0) { 1167 numerrors++; 1168 } else if (doread) { 1169 break; 1170 } 1171 } 1172 1173 if (!vd->vdev_ops->vdev_op_leaf) 1174 return (numerrors < vd->vdev_children ? 0 : EIO); 1175 1176 if (doread && !vdev_readable(vd)) 1177 return (EIO); 1178 else if (!doread && !vdev_writeable(vd)) 1179 return (EIO); 1180 1181 dvd = vd->vdev_tsd; 1182 ASSERT3P(dvd, !=, NULL); 1183 offset += VDEV_LABEL_START_SIZE; 1184 1185 if (ddi_in_panic() || isdump) { 1186 ASSERT(!doread); 1187 if (doread) 1188 return (EIO); 1189 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1190 lbtodb(size))); 1191 } else { 1192 return (vdev_disk_physio(dvd->vd_lh, addr, size, offset, 1193 doread ? B_READ : B_WRITE)); 1194 } 1195} 1196 1197static int 1198zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1199 boolean_t doread, boolean_t isdump) 1200{ 1201 vdev_t *vd; 1202 int error; 1203 zvol_extent_t *ze; 1204 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1205 1206 /* Must be sector aligned, and not stradle a block boundary. */ 1207 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1208 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1209 return (EINVAL); 1210 } 1211 ASSERT(size <= zv->zv_volblocksize); 1212 1213 /* Locate the extent this belongs to */ 1214 ze = list_head(&zv->zv_extents); 1215 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1216 offset -= ze->ze_nblks * zv->zv_volblocksize; 1217 ze = list_next(&zv->zv_extents, ze); 1218 } 1219 1220 if (!ddi_in_panic()) 1221 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1222 1223 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1224 offset += DVA_GET_OFFSET(&ze->ze_dva); 1225 error = zvol_dumpio_vdev(vd, addr, offset, size, doread, isdump); 1226 1227 if (!ddi_in_panic()) 1228 spa_config_exit(spa, SCL_STATE, FTAG); 1229 1230 return (error); 1231} 1232#endif /* sun */ 1233 1234int 1235zvol_strategy(struct bio *bp) 1236{ 1237 zvol_state_t *zv = bp->bio_to->private; 1238 uint64_t off, volsize; 1239 size_t resid; 1240 char *addr; 1241 objset_t *os; 1242 rl_t *rl; 1243 int error = 0; 1244 boolean_t doread = (bp->bio_cmd == BIO_READ); 1245 boolean_t sync; 1246 1247 if (zv == NULL) { 1248 g_io_deliver(bp, ENXIO); 1249 return (0); 1250 } 1251 1252 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) { 1253 g_io_deliver(bp, EROFS); 1254 return (0); 1255 } 1256 1257 off = bp->bio_offset; 1258 volsize = zv->zv_volsize; 1259 1260 os = zv->zv_objset; 1261 ASSERT(os != NULL); 1262 1263 addr = bp->bio_data; 1264 resid = bp->bio_length; 1265 1266 if (resid > 0 && (off < 0 || off >= volsize)) { 1267 g_io_deliver(bp, EIO); 1268 return (0); 1269 } 1270 1271 sync = !doread && zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; 1272 1273 /* 1274 * There must be no buffer changes when doing a dmu_sync() because 1275 * we can't change the data whilst calculating the checksum. 1276 */ 1277 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1278 doread ? RL_READER : RL_WRITER); 1279 1280 while (resid != 0 && off < volsize) { 1281 size_t size = MIN(resid, zvol_maxphys); 1282 if (doread) { 1283 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1284 DMU_READ_PREFETCH); 1285 } else { 1286 dmu_tx_t *tx = dmu_tx_create(os); 1287 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1288 error = dmu_tx_assign(tx, TXG_WAIT); 1289 if (error) { 1290 dmu_tx_abort(tx); 1291 } else { 1292 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1293 zvol_log_write(zv, tx, off, size, sync); 1294 dmu_tx_commit(tx); 1295 } 1296 } 1297 if (error) { 1298 /* convert checksum errors into IO errors */ 1299 if (error == ECKSUM) 1300 error = EIO; 1301 break; 1302 } 1303 off += size; 1304 addr += size; 1305 resid -= size; 1306 } 1307 zfs_range_unlock(rl); 1308 1309 bp->bio_completed = bp->bio_length - resid; 1310 if (bp->bio_completed < bp->bio_length) 1311 bp->bio_error = (off > volsize ? EINVAL : error); 1312 1313 if (sync) 1314 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1315 g_io_deliver(bp, 0); 1316 1317 return (0); 1318} 1319 1320#ifdef sun 1321/* 1322 * Set the buffer count to the zvol maximum transfer. 1323 * Using our own routine instead of the default minphys() 1324 * means that for larger writes we write bigger buffers on X86 1325 * (128K instead of 56K) and flush the disk write cache less often 1326 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1327 * 56K on X86 and 128K on sparc). 1328 */ 1329void 1330zvol_minphys(struct buf *bp) 1331{ 1332 if (bp->b_bcount > zvol_maxphys) 1333 bp->b_bcount = zvol_maxphys; 1334} 1335 1336int 1337zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1338{ 1339 minor_t minor = getminor(dev); 1340 zvol_state_t *zv; 1341 int error = 0; 1342 uint64_t size; 1343 uint64_t boff; 1344 uint64_t resid; 1345 1346 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1347 if (zv == NULL) 1348 return (ENXIO); 1349 1350 boff = ldbtob(blkno); 1351 resid = ldbtob(nblocks); 1352 1353 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1354 1355 while (resid) { 1356 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1357 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1358 if (error) 1359 break; 1360 boff += size; 1361 addr += size; 1362 resid -= size; 1363 } 1364 1365 return (error); 1366} 1367 1368/*ARGSUSED*/ 1369int 1370zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1371{ 1372 minor_t minor = getminor(dev); 1373 zvol_state_t *zv; 1374 uint64_t volsize; 1375 rl_t *rl; 1376 int error = 0; 1377 1378 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1379 if (zv == NULL) 1380 return (ENXIO); 1381 1382 volsize = zv->zv_volsize; 1383 if (uio->uio_resid > 0 && 1384 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1385 return (EIO); 1386 1387 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1388 error = physio(zvol_strategy, NULL, dev, B_READ, 1389 zvol_minphys, uio); 1390 return (error); 1391 } 1392 1393 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1394 RL_READER); 1395 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1396 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1397 1398 /* don't read past the end */ 1399 if (bytes > volsize - uio->uio_loffset) 1400 bytes = volsize - uio->uio_loffset; 1401 1402 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1403 if (error) { 1404 /* convert checksum errors into IO errors */ 1405 if (error == ECKSUM) 1406 error = EIO; 1407 break; 1408 } 1409 } 1410 zfs_range_unlock(rl); 1411 return (error); 1412} 1413 1414/*ARGSUSED*/ 1415int 1416zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1417{ 1418 minor_t minor = getminor(dev); 1419 zvol_state_t *zv; 1420 uint64_t volsize; 1421 rl_t *rl; 1422 int error = 0; 1423 boolean_t sync; 1424 1425 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1426 if (zv == NULL) 1427 return (ENXIO); 1428 1429 volsize = zv->zv_volsize; 1430 if (uio->uio_resid > 0 && 1431 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1432 return (EIO); 1433 1434 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1435 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1436 zvol_minphys, uio); 1437 return (error); 1438 } 1439 1440 sync = !(zv->zv_flags & ZVOL_WCE) || 1441 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1442 1443 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1444 RL_WRITER); 1445 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1446 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1447 uint64_t off = uio->uio_loffset; 1448 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1449 1450 if (bytes > volsize - off) /* don't write past the end */ 1451 bytes = volsize - off; 1452 1453 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1454 error = dmu_tx_assign(tx, TXG_WAIT); 1455 if (error) { 1456 dmu_tx_abort(tx); 1457 break; 1458 } 1459 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1460 if (error == 0) 1461 zvol_log_write(zv, tx, off, bytes, sync); 1462 dmu_tx_commit(tx); 1463 1464 if (error) 1465 break; 1466 } 1467 zfs_range_unlock(rl); 1468 if (sync) 1469 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1470 return (error); 1471} 1472 1473int 1474zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1475{ 1476 struct uuid uuid = EFI_RESERVED; 1477 efi_gpe_t gpe = { 0 }; 1478 uint32_t crc; 1479 dk_efi_t efi; 1480 int length; 1481 char *ptr; 1482 1483 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1484 return (EFAULT); 1485 ptr = (char *)(uintptr_t)efi.dki_data_64; 1486 length = efi.dki_length; 1487 /* 1488 * Some clients may attempt to request a PMBR for the 1489 * zvol. Currently this interface will return EINVAL to 1490 * such requests. These requests could be supported by 1491 * adding a check for lba == 0 and consing up an appropriate 1492 * PMBR. 1493 */ 1494 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1495 return (EINVAL); 1496 1497 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1498 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1499 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1500 1501 if (efi.dki_lba == 1) { 1502 efi_gpt_t gpt = { 0 }; 1503 1504 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1505 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1506 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1507 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1508 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1509 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1510 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1511 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1512 gpt.efi_gpt_SizeOfPartitionEntry = 1513 LE_32(sizeof (efi_gpe_t)); 1514 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1515 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1516 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1517 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1518 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1519 flag)) 1520 return (EFAULT); 1521 ptr += sizeof (gpt); 1522 length -= sizeof (gpt); 1523 } 1524 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1525 length), flag)) 1526 return (EFAULT); 1527 return (0); 1528} 1529 1530/* 1531 * BEGIN entry points to allow external callers access to the volume. 1532 */ 1533/* 1534 * Return the volume parameters needed for access from an external caller. 1535 * These values are invariant as long as the volume is held open. 1536 */ 1537int 1538zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1539 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1540 void **rl_hdl, void **bonus_hdl) 1541{ 1542 zvol_state_t *zv; 1543 1544 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1545 if (zv == NULL) 1546 return (ENXIO); 1547 if (zv->zv_flags & ZVOL_DUMPIFIED) 1548 return (ENXIO); 1549 1550 ASSERT(blksize && max_xfer_len && minor_hdl && 1551 objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 1552 1553 *blksize = zv->zv_volblocksize; 1554 *max_xfer_len = (uint64_t)zvol_maxphys; 1555 *minor_hdl = zv; 1556 *objset_hdl = zv->zv_objset; 1557 *zil_hdl = zv->zv_zilog; 1558 *rl_hdl = &zv->zv_znode; 1559 *bonus_hdl = zv->zv_dbuf; 1560 return (0); 1561} 1562 1563/* 1564 * Return the current volume size to an external caller. 1565 * The size can change while the volume is open. 1566 */ 1567uint64_t 1568zvol_get_volume_size(void *minor_hdl) 1569{ 1570 zvol_state_t *zv = minor_hdl; 1571 1572 return (zv->zv_volsize); 1573} 1574 1575/* 1576 * Return the current WCE setting to an external caller. 1577 * The WCE setting can change while the volume is open. 1578 */ 1579int 1580zvol_get_volume_wce(void *minor_hdl) 1581{ 1582 zvol_state_t *zv = minor_hdl; 1583 1584 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 1585} 1586 1587/* 1588 * Entry point for external callers to zvol_log_write 1589 */ 1590void 1591zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 1592 boolean_t sync) 1593{ 1594 zvol_state_t *zv = minor_hdl; 1595 1596 zvol_log_write(zv, tx, off, resid, sync); 1597} 1598/* 1599 * END entry points to allow external callers access to the volume. 1600 */ 1601 1602/* 1603 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1604 */ 1605/*ARGSUSED*/ 1606int 1607zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1608{ 1609 zvol_state_t *zv; 1610 struct dk_cinfo dki; 1611 struct dk_minfo dkm; 1612 struct dk_callback *dkc; 1613 int error = 0; 1614 rl_t *rl; 1615 1616 mutex_enter(&spa_namespace_lock); 1617 1618 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 1619 1620 if (zv == NULL) { 1621 mutex_exit(&spa_namespace_lock); 1622 return (ENXIO); 1623 } 1624 ASSERT(zv->zv_total_opens > 0); 1625 1626 switch (cmd) { 1627 1628 case DKIOCINFO: 1629 bzero(&dki, sizeof (dki)); 1630 (void) strcpy(dki.dki_cname, "zvol"); 1631 (void) strcpy(dki.dki_dname, "zvol"); 1632 dki.dki_ctype = DKC_UNKNOWN; 1633 dki.dki_unit = getminor(dev); 1634 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); 1635 mutex_exit(&spa_namespace_lock); 1636 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1637 error = EFAULT; 1638 return (error); 1639 1640 case DKIOCGMEDIAINFO: 1641 bzero(&dkm, sizeof (dkm)); 1642 dkm.dki_lbsize = 1U << zv->zv_min_bs; 1643 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1644 dkm.dki_media_type = DK_UNKNOWN; 1645 mutex_exit(&spa_namespace_lock); 1646 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1647 error = EFAULT; 1648 return (error); 1649 1650 case DKIOCGETEFI: 1651 { 1652 uint64_t vs = zv->zv_volsize; 1653 uint8_t bs = zv->zv_min_bs; 1654 1655 mutex_exit(&spa_namespace_lock); 1656 error = zvol_getefi((void *)arg, flag, vs, bs); 1657 return (error); 1658 } 1659 1660 case DKIOCFLUSHWRITECACHE: 1661 dkc = (struct dk_callback *)arg; 1662 mutex_exit(&spa_namespace_lock); 1663 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1664 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1665 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1666 error = 0; 1667 } 1668 return (error); 1669 1670 case DKIOCGETWCE: 1671 { 1672 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1673 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1674 flag)) 1675 error = EFAULT; 1676 break; 1677 } 1678 case DKIOCSETWCE: 1679 { 1680 int wce; 1681 if (ddi_copyin((void *)arg, &wce, sizeof (int), 1682 flag)) { 1683 error = EFAULT; 1684 break; 1685 } 1686 if (wce) { 1687 zv->zv_flags |= ZVOL_WCE; 1688 mutex_exit(&spa_namespace_lock); 1689 } else { 1690 zv->zv_flags &= ~ZVOL_WCE; 1691 mutex_exit(&spa_namespace_lock); 1692 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1693 } 1694 return (0); 1695 } 1696 1697 case DKIOCGGEOM: 1698 case DKIOCGVTOC: 1699 /* 1700 * commands using these (like prtvtoc) expect ENOTSUP 1701 * since we're emulating an EFI label 1702 */ 1703 error = ENOTSUP; 1704 break; 1705 1706 case DKIOCDUMPINIT: 1707 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1708 RL_WRITER); 1709 error = zvol_dumpify(zv); 1710 zfs_range_unlock(rl); 1711 break; 1712 1713 case DKIOCDUMPFINI: 1714 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 1715 break; 1716 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1717 RL_WRITER); 1718 error = zvol_dump_fini(zv); 1719 zfs_range_unlock(rl); 1720 break; 1721 1722 default: 1723 error = ENOTTY; 1724 break; 1725 1726 } 1727 mutex_exit(&spa_namespace_lock); 1728 return (error); 1729} 1730#endif /* sun */ 1731 1732int 1733zvol_busy(void) 1734{ 1735 return (zvol_minors != 0); 1736} 1737 1738void 1739zvol_init(void) 1740{ 1741 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 1742 1) == 0); 1743 ZFS_LOG(1, "ZVOL Initialized."); 1744} 1745 1746void 1747zvol_fini(void) 1748{ 1749 ddi_soft_state_fini(&zfsdev_state); 1750 ZFS_LOG(1, "ZVOL Deinitialized."); 1751} 1752 1753#ifdef sun 1754static int 1755zvol_dump_init(zvol_state_t *zv, boolean_t resize) 1756{ 1757 dmu_tx_t *tx; 1758 int error = 0; 1759 objset_t *os = zv->zv_objset; 1760 nvlist_t *nv = NULL; 1761 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 1762 1763 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1764 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 1765 DMU_OBJECT_END); 1766 /* wait for dmu_free_long_range to actually free the blocks */ 1767 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1768 1769 tx = dmu_tx_create(os); 1770 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1771 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1772 error = dmu_tx_assign(tx, TXG_WAIT); 1773 if (error) { 1774 dmu_tx_abort(tx); 1775 return (error); 1776 } 1777 1778 /* 1779 * If we are resizing the dump device then we only need to 1780 * update the refreservation to match the newly updated 1781 * zvolsize. Otherwise, we save off the original state of the 1782 * zvol so that we can restore them if the zvol is ever undumpified. 1783 */ 1784 if (resize) { 1785 error = zap_update(os, ZVOL_ZAP_OBJ, 1786 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1787 &zv->zv_volsize, tx); 1788 } else { 1789 uint64_t checksum, compress, refresrv, vbs, dedup; 1790 1791 error = dsl_prop_get_integer(zv->zv_name, 1792 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 1793 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1794 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL); 1795 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1796 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL); 1797 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1798 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL); 1799 if (version >= SPA_VERSION_DEDUP) { 1800 error = error ? error : 1801 dsl_prop_get_integer(zv->zv_name, 1802 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 1803 } 1804 1805 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1806 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 1807 &compress, tx); 1808 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1809 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx); 1810 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1811 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1812 &refresrv, tx); 1813 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1814 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 1815 &vbs, tx); 1816 error = error ? error : dmu_object_set_blocksize( 1817 os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx); 1818 if (version >= SPA_VERSION_DEDUP) { 1819 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1820 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 1821 &dedup, tx); 1822 } 1823 if (error == 0) 1824 zv->zv_volblocksize = SPA_MAXBLOCKSIZE; 1825 } 1826 dmu_tx_commit(tx); 1827 1828 /* 1829 * We only need update the zvol's property if we are initializing 1830 * the dump area for the first time. 1831 */ 1832 if (!resize) { 1833 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1834 VERIFY(nvlist_add_uint64(nv, 1835 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 1836 VERIFY(nvlist_add_uint64(nv, 1837 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 1838 ZIO_COMPRESS_OFF) == 0); 1839 VERIFY(nvlist_add_uint64(nv, 1840 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 1841 ZIO_CHECKSUM_OFF) == 0); 1842 if (version >= SPA_VERSION_DEDUP) { 1843 VERIFY(nvlist_add_uint64(nv, 1844 zfs_prop_to_name(ZFS_PROP_DEDUP), 1845 ZIO_CHECKSUM_OFF) == 0); 1846 } 1847 1848 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 1849 nv, NULL); 1850 nvlist_free(nv); 1851 1852 if (error) 1853 return (error); 1854 } 1855 1856 /* Allocate the space for the dump */ 1857 error = zvol_prealloc(zv); 1858 return (error); 1859} 1860 1861static int 1862zvol_dumpify(zvol_state_t *zv) 1863{ 1864 int error = 0; 1865 uint64_t dumpsize = 0; 1866 dmu_tx_t *tx; 1867 objset_t *os = zv->zv_objset; 1868 1869 if (zv->zv_flags & ZVOL_RDONLY) 1870 return (EROFS); 1871 1872 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 1873 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 1874 boolean_t resize = (dumpsize > 0) ? B_TRUE : B_FALSE; 1875 1876 if ((error = zvol_dump_init(zv, resize)) != 0) { 1877 (void) zvol_dump_fini(zv); 1878 return (error); 1879 } 1880 } 1881 1882 /* 1883 * Build up our lba mapping. 1884 */ 1885 error = zvol_get_lbas(zv); 1886 if (error) { 1887 (void) zvol_dump_fini(zv); 1888 return (error); 1889 } 1890 1891 tx = dmu_tx_create(os); 1892 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1893 error = dmu_tx_assign(tx, TXG_WAIT); 1894 if (error) { 1895 dmu_tx_abort(tx); 1896 (void) zvol_dump_fini(zv); 1897 return (error); 1898 } 1899 1900 zv->zv_flags |= ZVOL_DUMPIFIED; 1901 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 1902 &zv->zv_volsize, tx); 1903 dmu_tx_commit(tx); 1904 1905 if (error) { 1906 (void) zvol_dump_fini(zv); 1907 return (error); 1908 } 1909 1910 txg_wait_synced(dmu_objset_pool(os), 0); 1911 return (0); 1912} 1913 1914static int 1915zvol_dump_fini(zvol_state_t *zv) 1916{ 1917 dmu_tx_t *tx; 1918 objset_t *os = zv->zv_objset; 1919 nvlist_t *nv; 1920 int error = 0; 1921 uint64_t checksum, compress, refresrv, vbs, dedup; 1922 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 1923 1924 /* 1925 * Attempt to restore the zvol back to its pre-dumpified state. 1926 * This is a best-effort attempt as it's possible that not all 1927 * of these properties were initialized during the dumpify process 1928 * (i.e. error during zvol_dump_init). 1929 */ 1930 1931 tx = dmu_tx_create(os); 1932 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1933 error = dmu_tx_assign(tx, TXG_WAIT); 1934 if (error) { 1935 dmu_tx_abort(tx); 1936 return (error); 1937 } 1938 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 1939 dmu_tx_commit(tx); 1940 1941 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1942 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 1943 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1944 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 1945 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1946 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 1947 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1948 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 1949 1950 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1951 (void) nvlist_add_uint64(nv, 1952 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 1953 (void) nvlist_add_uint64(nv, 1954 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 1955 (void) nvlist_add_uint64(nv, 1956 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 1957 if (version >= SPA_VERSION_DEDUP && 1958 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1959 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 1960 (void) nvlist_add_uint64(nv, 1961 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 1962 } 1963 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 1964 nv, NULL); 1965 nvlist_free(nv); 1966 1967 zvol_free_extents(zv); 1968 zv->zv_flags &= ~ZVOL_DUMPIFIED; 1969 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 1970 /* wait for dmu_free_long_range to actually free the blocks */ 1971 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1972 tx = dmu_tx_create(os); 1973 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1974 error = dmu_tx_assign(tx, TXG_WAIT); 1975 if (error) { 1976 dmu_tx_abort(tx); 1977 return (error); 1978 } 1979 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 1980 zv->zv_volblocksize = vbs; 1981 dmu_tx_commit(tx); 1982 1983 return (0); 1984} 1985#endif /* sun */ 1986 1987static zvol_state_t * 1988zvol_geom_create(const char *name) 1989{ 1990 struct g_provider *pp; 1991 struct g_geom *gp; 1992 zvol_state_t *zv; 1993 1994 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name); 1995 gp->start = zvol_geom_start; 1996 gp->access = zvol_geom_access; 1997 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name); 1998 pp->sectorsize = DEV_BSIZE; 1999 2000 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP); 2001 zv->zv_provider = pp; 2002 zv->zv_state = 0; 2003 bioq_init(&zv->zv_queue); 2004 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF); 2005 2006 pp->private = zv; 2007 2008 return (zv); 2009} 2010 2011static void 2012zvol_geom_run(zvol_state_t *zv) 2013{ 2014 struct g_provider *pp; 2015 2016 pp = zv->zv_provider; 2017 g_error_provider(pp, 0); 2018 2019 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0, 2020 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER)); 2021} 2022 2023static void 2024zvol_geom_destroy(zvol_state_t *zv) 2025{ 2026 struct g_provider *pp; 2027 2028 g_topology_assert(); 2029 2030 mtx_lock(&zv->zv_queue_mtx); 2031 zv->zv_state = 1; 2032 wakeup_one(&zv->zv_queue); 2033 while (zv->zv_state != 2) 2034 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0); 2035 mtx_destroy(&zv->zv_queue_mtx); 2036 2037 pp = zv->zv_provider; 2038 zv->zv_provider = NULL; 2039 pp->private = NULL; 2040 g_wither_geom(pp->geom, ENXIO); 2041 2042 kmem_free(zv, sizeof(*zv)); 2043} 2044 2045static int 2046zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) 2047{ 2048 int count, error, flags; 2049 2050 g_topology_assert(); 2051 2052 /* 2053 * To make it easier we expect either open or close, but not both 2054 * at the same time. 2055 */ 2056 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) || 2057 (acr <= 0 && acw <= 0 && ace <= 0), 2058 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", 2059 pp->name, acr, acw, ace)); 2060 2061 if (pp->private == NULL) { 2062 if (acr <= 0 && acw <= 0 && ace <= 0) 2063 return (0); 2064 return (pp->error); 2065 } 2066 2067 /* 2068 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0, 2069 * because GEOM already handles that and handles it a bit differently. 2070 * GEOM allows for multiple read/exclusive consumers and ZFS allows 2071 * only one exclusive consumer, no matter if it is reader or writer. 2072 * I like better the way GEOM works so I'll leave it for GEOM to 2073 * decide what to do. 2074 */ 2075 2076 count = acr + acw + ace; 2077 if (count == 0) 2078 return (0); 2079 2080 flags = 0; 2081 if (acr != 0 || ace != 0) 2082 flags |= FREAD; 2083 if (acw != 0) 2084 flags |= FWRITE; 2085 2086 g_topology_unlock(); 2087 if (count > 0) 2088 error = zvol_open(pp, flags, count); 2089 else 2090 error = zvol_close(pp, flags, -count); 2091 g_topology_lock(); 2092 return (error); 2093} 2094 2095static void 2096zvol_geom_start(struct bio *bp) 2097{ 2098 zvol_state_t *zv; 2099 boolean_t first; 2100 2101 switch (bp->bio_cmd) { 2102 case BIO_READ: 2103 case BIO_WRITE: 2104 case BIO_FLUSH: 2105 zv = bp->bio_to->private; 2106 ASSERT(zv != NULL); 2107 mtx_lock(&zv->zv_queue_mtx); 2108 first = (bioq_first(&zv->zv_queue) == NULL); 2109 bioq_insert_tail(&zv->zv_queue, bp); 2110 mtx_unlock(&zv->zv_queue_mtx); 2111 if (first) 2112 wakeup_one(&zv->zv_queue); 2113 break; 2114 case BIO_GETATTR: 2115 case BIO_DELETE: 2116 default: 2117 g_io_deliver(bp, EOPNOTSUPP); 2118 break; 2119 } 2120} 2121 2122static void 2123zvol_geom_worker(void *arg) 2124{ 2125 zvol_state_t *zv; 2126 struct bio *bp; 2127 2128 thread_lock(curthread); 2129 sched_prio(curthread, PRIBIO); 2130 thread_unlock(curthread); 2131 2132 zv = arg; 2133 for (;;) { 2134 mtx_lock(&zv->zv_queue_mtx); 2135 bp = bioq_takefirst(&zv->zv_queue); 2136 if (bp == NULL) { 2137 if (zv->zv_state == 1) { 2138 zv->zv_state = 2; 2139 wakeup(&zv->zv_state); 2140 mtx_unlock(&zv->zv_queue_mtx); 2141 kthread_exit(); 2142 } 2143 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP, 2144 "zvol:io", 0); 2145 continue; 2146 } 2147 mtx_unlock(&zv->zv_queue_mtx); 2148 switch (bp->bio_cmd) { 2149 case BIO_FLUSH: 2150 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2151 g_io_deliver(bp, 0); 2152 break; 2153 case BIO_READ: 2154 case BIO_WRITE: 2155 zvol_strategy(bp); 2156 break; 2157 } 2158 } 2159} 2160 2161extern boolean_t dataset_name_hidden(const char *name); 2162 2163static int 2164zvol_create_snapshots(objset_t *os, const char *name) 2165{ 2166 uint64_t cookie, obj; 2167 char *sname; 2168 int error, len; 2169 2170 cookie = obj = 0; 2171 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2172 2173 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL, 2174 DS_FIND_SNAPSHOTS); 2175 2176 for (;;) { 2177 len = snprintf(sname, MAXPATHLEN, "%s@", name); 2178 if (len >= MAXPATHLEN) { 2179 dmu_objset_rele(os, FTAG); 2180 error = ENAMETOOLONG; 2181 break; 2182 } 2183 2184 error = dmu_snapshot_list_next(os, MAXPATHLEN - len, 2185 sname + len, &obj, &cookie, NULL); 2186 if (error != 0) { 2187 if (error == ENOENT) 2188 error = 0; 2189 break; 2190 } 2191 2192 if ((error = zvol_create_minor(sname)) != 0) { 2193 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2194 sname, error); 2195 break; 2196 } 2197 } 2198 2199 kmem_free(sname, MAXPATHLEN); 2200 return (error); 2201} 2202 2203int 2204zvol_create_minors(const char *name) 2205{ 2206 uint64_t cookie; 2207 objset_t *os; 2208 char *osname, *p; 2209 int error, len; 2210 2211 if (dataset_name_hidden(name)) 2212 return (0); 2213 2214 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2215 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2216 name, error); 2217 return (error); 2218 } 2219 if (dmu_objset_type(os) == DMU_OST_ZVOL) { 2220 if ((error = zvol_create_minor(name)) == 0) 2221 error = zvol_create_snapshots(os, name); 2222 else { 2223 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2224 name, error); 2225 } 2226 dmu_objset_rele(os, FTAG); 2227 return (error); 2228 } 2229 if (dmu_objset_type(os) != DMU_OST_ZFS) { 2230 dmu_objset_rele(os, FTAG); 2231 return (0); 2232 } 2233 2234 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2235 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) { 2236 dmu_objset_rele(os, FTAG); 2237 kmem_free(osname, MAXPATHLEN); 2238 return (ENOENT); 2239 } 2240 p = osname + strlen(osname); 2241 len = MAXPATHLEN - (p - osname); 2242 2243 /* Prefetch the datasets. */ 2244 cookie = 0; 2245 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) { 2246 if (!dataset_name_hidden(osname)) 2247 (void) dmu_objset_prefetch(osname, NULL); 2248 } 2249 2250 cookie = 0; 2251 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL, 2252 &cookie) == 0) { 2253 dmu_objset_rele(os, FTAG); 2254 (void)zvol_create_minors(osname); 2255 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2256 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2257 name, error); 2258 return (error); 2259 } 2260 } 2261 2262 dmu_objset_rele(os, FTAG); 2263 kmem_free(osname, MAXPATHLEN); 2264 return (0); 2265} 2266 2267static void 2268zvol_rename_minor(struct g_geom *gp, const char *newname) 2269{ 2270 struct g_provider *pp; 2271 zvol_state_t *zv; 2272 2273 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2274 g_topology_assert(); 2275 2276 pp = LIST_FIRST(&gp->provider); 2277 ASSERT(pp != NULL); 2278 zv = pp->private; 2279 ASSERT(zv != NULL); 2280 2281 zv->zv_provider = NULL; 2282 g_wither_provider(pp, ENXIO); 2283 2284 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname); 2285 pp->sectorsize = DEV_BSIZE; 2286 pp->mediasize = zv->zv_volsize; 2287 pp->private = zv; 2288 zv->zv_provider = pp; 2289 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name)); 2290 g_error_provider(pp, 0); 2291} 2292 2293void 2294zvol_rename_minors(const char *oldname, const char *newname) 2295{ 2296 char name[MAXPATHLEN]; 2297 struct g_provider *pp; 2298 struct g_geom *gp; 2299 size_t oldnamelen, newnamelen; 2300 zvol_state_t *zv; 2301 char *namebuf; 2302 2303 oldnamelen = strlen(oldname); 2304 newnamelen = strlen(newname); 2305 2306 DROP_GIANT(); 2307 mutex_enter(&spa_namespace_lock); 2308 g_topology_lock(); 2309 2310 LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) { 2311 pp = LIST_FIRST(&gp->provider); 2312 if (pp == NULL) 2313 continue; 2314 zv = pp->private; 2315 if (zv == NULL) 2316 continue; 2317 if (strcmp(zv->zv_name, oldname) == 0) { 2318 zvol_rename_minor(gp, newname); 2319 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && 2320 (zv->zv_name[oldnamelen] == '/' || 2321 zv->zv_name[oldnamelen] == '@')) { 2322 snprintf(name, sizeof(name), "%s%c%s", newname, 2323 zv->zv_name[oldnamelen], 2324 zv->zv_name + oldnamelen + 1); 2325 zvol_rename_minor(gp, name); 2326 } 2327 } 2328 2329 g_topology_unlock(); 2330 mutex_exit(&spa_namespace_lock); 2331 PICKUP_GIANT(); 2332} 2333