zvol.c revision 263390
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 * Copyright (c) 2013 by Delphix. All rights reserved. 27 * Copyright (c) 2013, Joyent, Inc. All rights reserved. 28 */ 29 30/* Portions Copyright 2010 Robert Milkowski */ 31/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 32 33/* 34 * ZFS volume emulation driver. 35 * 36 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 37 * Volumes are accessed through the symbolic links named: 38 * 39 * /dev/zvol/dsk/<pool_name>/<dataset_name> 40 * /dev/zvol/rdsk/<pool_name>/<dataset_name> 41 * 42 * These links are created by the /dev filesystem (sdev_zvolops.c). 43 * Volumes are persistent through reboot. No user command needs to be 44 * run before opening and using a device. 45 * 46 * FreeBSD notes. 47 * On FreeBSD ZVOLs are simply GEOM providers like any other storage device 48 * in the system. 49 */ 50 51#include <sys/types.h> 52#include <sys/param.h> 53#include <sys/kernel.h> 54#include <sys/errno.h> 55#include <sys/uio.h> 56#include <sys/bio.h> 57#include <sys/buf.h> 58#include <sys/kmem.h> 59#include <sys/conf.h> 60#include <sys/cmn_err.h> 61#include <sys/stat.h> 62#include <sys/zap.h> 63#include <sys/spa.h> 64#include <sys/spa_impl.h> 65#include <sys/zio.h> 66#include <sys/dmu_traverse.h> 67#include <sys/dnode.h> 68#include <sys/dsl_dataset.h> 69#include <sys/dsl_prop.h> 70#include <sys/dkio.h> 71#include <sys/byteorder.h> 72#include <sys/sunddi.h> 73#include <sys/dirent.h> 74#include <sys/policy.h> 75#include <sys/fs/zfs.h> 76#include <sys/zfs_ioctl.h> 77#include <sys/zil.h> 78#include <sys/refcount.h> 79#include <sys/zfs_znode.h> 80#include <sys/zfs_rlock.h> 81#include <sys/vdev_impl.h> 82#include <sys/vdev_raidz.h> 83#include <sys/zvol.h> 84#include <sys/zil_impl.h> 85#include <sys/dbuf.h> 86#include <sys/dmu_tx.h> 87#include <sys/zfeature.h> 88#include <sys/zio_checksum.h> 89 90#include <geom/geom.h> 91 92#include "zfs_namecheck.h" 93 94struct g_class zfs_zvol_class = { 95 .name = "ZFS::ZVOL", 96 .version = G_VERSION, 97}; 98 99DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol); 100 101void *zfsdev_state; 102static char *zvol_tag = "zvol_tag"; 103 104#define ZVOL_DUMPSIZE "dumpsize" 105 106/* 107 * The spa_namespace_lock protects the zfsdev_state structure from being 108 * modified while it's being used, e.g. an open that comes in before a 109 * create finishes. It also protects temporary opens of the dataset so that, 110 * e.g., an open doesn't get a spurious EBUSY. 111 */ 112static uint32_t zvol_minors; 113 114typedef struct zvol_extent { 115 list_node_t ze_node; 116 dva_t ze_dva; /* dva associated with this extent */ 117 uint64_t ze_nblks; /* number of blocks in extent */ 118} zvol_extent_t; 119 120/* 121 * The in-core state of each volume. 122 */ 123typedef struct zvol_state { 124 char zv_name[MAXPATHLEN]; /* pool/dd name */ 125 uint64_t zv_volsize; /* amount of space we advertise */ 126 uint64_t zv_volblocksize; /* volume block size */ 127 struct g_provider *zv_provider; /* GEOM provider */ 128 uint8_t zv_min_bs; /* minimum addressable block shift */ 129 uint8_t zv_flags; /* readonly, dumpified, etc. */ 130 objset_t *zv_objset; /* objset handle */ 131 uint32_t zv_total_opens; /* total open count */ 132 zilog_t *zv_zilog; /* ZIL handle */ 133 list_t zv_extents; /* List of extents for dump */ 134 znode_t zv_znode; /* for range locking */ 135 dmu_buf_t *zv_dbuf; /* bonus handle */ 136 int zv_state; 137 struct bio_queue_head zv_queue; 138 struct mtx zv_queue_mtx; /* zv_queue mutex */ 139} zvol_state_t; 140 141/* 142 * zvol specific flags 143 */ 144#define ZVOL_RDONLY 0x1 145#define ZVOL_DUMPIFIED 0x2 146#define ZVOL_EXCL 0x4 147#define ZVOL_WCE 0x8 148 149/* 150 * zvol maximum transfer in one DMU tx. 151 */ 152int zvol_maxphys = DMU_MAX_ACCESS/2; 153 154extern int zfs_set_prop_nvlist(const char *, zprop_source_t, 155 nvlist_t *, nvlist_t *); 156static int zvol_remove_zv(zvol_state_t *); 157static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio); 158static int zvol_dumpify(zvol_state_t *zv); 159static int zvol_dump_fini(zvol_state_t *zv); 160static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); 161 162static zvol_state_t *zvol_geom_create(const char *name); 163static void zvol_geom_run(zvol_state_t *zv); 164static void zvol_geom_destroy(zvol_state_t *zv); 165static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace); 166static void zvol_geom_start(struct bio *bp); 167static void zvol_geom_worker(void *arg); 168 169static void 170zvol_size_changed(zvol_state_t *zv) 171{ 172#ifdef sun 173 dev_t dev = makedevice(maj, min); 174 175 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 176 "Size", volsize) == DDI_SUCCESS); 177 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 178 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS); 179 180 /* Notify specfs to invalidate the cached size */ 181 spec_size_invalidate(dev, VBLK); 182 spec_size_invalidate(dev, VCHR); 183#else /* !sun */ 184 struct g_provider *pp; 185 186 pp = zv->zv_provider; 187 if (pp == NULL) 188 return; 189 g_topology_lock(); 190 g_resize_provider(pp, zv->zv_volsize); 191 g_topology_unlock(); 192#endif /* !sun */ 193} 194 195int 196zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 197{ 198 if (volsize == 0) 199 return (SET_ERROR(EINVAL)); 200 201 if (volsize % blocksize != 0) 202 return (SET_ERROR(EINVAL)); 203 204#ifdef _ILP32 205 if (volsize - 1 > SPEC_MAXOFFSET_T) 206 return (SET_ERROR(EOVERFLOW)); 207#endif 208 return (0); 209} 210 211int 212zvol_check_volblocksize(uint64_t volblocksize) 213{ 214 if (volblocksize < SPA_MINBLOCKSIZE || 215 volblocksize > SPA_MAXBLOCKSIZE || 216 !ISP2(volblocksize)) 217 return (SET_ERROR(EDOM)); 218 219 return (0); 220} 221 222int 223zvol_get_stats(objset_t *os, nvlist_t *nv) 224{ 225 int error; 226 dmu_object_info_t doi; 227 uint64_t val; 228 229 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 230 if (error) 231 return (error); 232 233 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 234 235 error = dmu_object_info(os, ZVOL_OBJ, &doi); 236 237 if (error == 0) { 238 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 239 doi.doi_data_block_size); 240 } 241 242 return (error); 243} 244 245static zvol_state_t * 246zvol_minor_lookup(const char *name) 247{ 248 struct g_provider *pp; 249 struct g_geom *gp; 250 zvol_state_t *zv = NULL; 251 252 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 253 254 g_topology_lock(); 255 LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) { 256 pp = LIST_FIRST(&gp->provider); 257 if (pp == NULL) 258 continue; 259 zv = pp->private; 260 if (zv == NULL) 261 continue; 262 if (strcmp(zv->zv_name, name) == 0) 263 break; 264 } 265 g_topology_unlock(); 266 267 return (gp != NULL ? zv : NULL); 268} 269 270/* extent mapping arg */ 271struct maparg { 272 zvol_state_t *ma_zv; 273 uint64_t ma_blks; 274}; 275 276/*ARGSUSED*/ 277static int 278zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 279 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 280{ 281 struct maparg *ma = arg; 282 zvol_extent_t *ze; 283 int bs = ma->ma_zv->zv_volblocksize; 284 285 if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 286 return (0); 287 288 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 289 ma->ma_blks++; 290 291 /* Abort immediately if we have encountered gang blocks */ 292 if (BP_IS_GANG(bp)) 293 return (SET_ERROR(EFRAGS)); 294 295 /* 296 * See if the block is at the end of the previous extent. 297 */ 298 ze = list_tail(&ma->ma_zv->zv_extents); 299 if (ze && 300 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 301 DVA_GET_OFFSET(BP_IDENTITY(bp)) == 302 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 303 ze->ze_nblks++; 304 return (0); 305 } 306 307 dprintf_bp(bp, "%s", "next blkptr:"); 308 309 /* start a new extent */ 310 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 311 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 312 ze->ze_nblks = 1; 313 list_insert_tail(&ma->ma_zv->zv_extents, ze); 314 return (0); 315} 316 317static void 318zvol_free_extents(zvol_state_t *zv) 319{ 320 zvol_extent_t *ze; 321 322 while (ze = list_head(&zv->zv_extents)) { 323 list_remove(&zv->zv_extents, ze); 324 kmem_free(ze, sizeof (zvol_extent_t)); 325 } 326} 327 328static int 329zvol_get_lbas(zvol_state_t *zv) 330{ 331 objset_t *os = zv->zv_objset; 332 struct maparg ma; 333 int err; 334 335 ma.ma_zv = zv; 336 ma.ma_blks = 0; 337 zvol_free_extents(zv); 338 339 /* commit any in-flight changes before traversing the dataset */ 340 txg_wait_synced(dmu_objset_pool(os), 0); 341 err = traverse_dataset(dmu_objset_ds(os), 0, 342 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 343 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 344 zvol_free_extents(zv); 345 return (err ? err : EIO); 346 } 347 348 return (0); 349} 350 351/* ARGSUSED */ 352void 353zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 354{ 355 zfs_creat_t *zct = arg; 356 nvlist_t *nvprops = zct->zct_props; 357 int error; 358 uint64_t volblocksize, volsize; 359 360 VERIFY(nvlist_lookup_uint64(nvprops, 361 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 362 if (nvlist_lookup_uint64(nvprops, 363 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 364 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 365 366 /* 367 * These properties must be removed from the list so the generic 368 * property setting step won't apply to them. 369 */ 370 VERIFY(nvlist_remove_all(nvprops, 371 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 372 (void) nvlist_remove_all(nvprops, 373 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 374 375 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 376 DMU_OT_NONE, 0, tx); 377 ASSERT(error == 0); 378 379 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 380 DMU_OT_NONE, 0, tx); 381 ASSERT(error == 0); 382 383 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 384 ASSERT(error == 0); 385} 386 387/* 388 * Replay a TX_WRITE ZIL transaction that didn't get committed 389 * after a system failure 390 */ 391static int 392zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 393{ 394 objset_t *os = zv->zv_objset; 395 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 396 uint64_t offset, length; 397 dmu_tx_t *tx; 398 int error; 399 400 if (byteswap) 401 byteswap_uint64_array(lr, sizeof (*lr)); 402 403 offset = lr->lr_offset; 404 length = lr->lr_length; 405 406 /* If it's a dmu_sync() block, write the whole block */ 407 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 408 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 409 if (length < blocksize) { 410 offset -= offset % blocksize; 411 length = blocksize; 412 } 413 } 414 415 tx = dmu_tx_create(os); 416 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 417 error = dmu_tx_assign(tx, TXG_WAIT); 418 if (error) { 419 dmu_tx_abort(tx); 420 } else { 421 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 422 dmu_tx_commit(tx); 423 } 424 425 return (error); 426} 427 428/* ARGSUSED */ 429static int 430zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 431{ 432 return (SET_ERROR(ENOTSUP)); 433} 434 435/* 436 * Callback vectors for replaying records. 437 * Only TX_WRITE is needed for zvol. 438 */ 439zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 440 zvol_replay_err, /* 0 no such transaction type */ 441 zvol_replay_err, /* TX_CREATE */ 442 zvol_replay_err, /* TX_MKDIR */ 443 zvol_replay_err, /* TX_MKXATTR */ 444 zvol_replay_err, /* TX_SYMLINK */ 445 zvol_replay_err, /* TX_REMOVE */ 446 zvol_replay_err, /* TX_RMDIR */ 447 zvol_replay_err, /* TX_LINK */ 448 zvol_replay_err, /* TX_RENAME */ 449 zvol_replay_write, /* TX_WRITE */ 450 zvol_replay_err, /* TX_TRUNCATE */ 451 zvol_replay_err, /* TX_SETATTR */ 452 zvol_replay_err, /* TX_ACL */ 453 zvol_replay_err, /* TX_CREATE_ACL */ 454 zvol_replay_err, /* TX_CREATE_ATTR */ 455 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 456 zvol_replay_err, /* TX_MKDIR_ACL */ 457 zvol_replay_err, /* TX_MKDIR_ATTR */ 458 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 459 zvol_replay_err, /* TX_WRITE2 */ 460}; 461 462#ifdef sun 463int 464zvol_name2minor(const char *name, minor_t *minor) 465{ 466 zvol_state_t *zv; 467 468 mutex_enter(&spa_namespace_lock); 469 zv = zvol_minor_lookup(name); 470 if (minor && zv) 471 *minor = zv->zv_minor; 472 mutex_exit(&spa_namespace_lock); 473 return (zv ? 0 : -1); 474} 475#endif /* sun */ 476 477/* 478 * Create a minor node (plus a whole lot more) for the specified volume. 479 */ 480int 481zvol_create_minor(const char *name) 482{ 483 zfs_soft_state_t *zs; 484 zvol_state_t *zv; 485 objset_t *os; 486 dmu_object_info_t doi; 487 uint64_t volsize; 488 int error; 489 490 ZFS_LOG(1, "Creating ZVOL %s...", name); 491 492 mutex_enter(&spa_namespace_lock); 493 494 if (zvol_minor_lookup(name) != NULL) { 495 mutex_exit(&spa_namespace_lock); 496 return (SET_ERROR(EEXIST)); 497 } 498 499 /* lie and say we're read-only */ 500 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os); 501 502 if (error) { 503 mutex_exit(&spa_namespace_lock); 504 return (error); 505 } 506 507#ifdef sun 508 if ((minor = zfsdev_minor_alloc()) == 0) { 509 dmu_objset_disown(os, FTAG); 510 mutex_exit(&spa_namespace_lock); 511 return (SET_ERROR(ENXIO)); 512 } 513 514 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 515 dmu_objset_disown(os, FTAG); 516 mutex_exit(&spa_namespace_lock); 517 return (SET_ERROR(EAGAIN)); 518 } 519 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 520 (char *)name); 521 522 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 523 524 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 525 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 526 ddi_soft_state_free(zfsdev_state, minor); 527 dmu_objset_disown(os, FTAG); 528 mutex_exit(&spa_namespace_lock); 529 return (SET_ERROR(EAGAIN)); 530 } 531 532 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 533 534 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 535 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 536 ddi_remove_minor_node(zfs_dip, chrbuf); 537 ddi_soft_state_free(zfsdev_state, minor); 538 dmu_objset_disown(os, FTAG); 539 mutex_exit(&spa_namespace_lock); 540 return (SET_ERROR(EAGAIN)); 541 } 542 543 zs = ddi_get_soft_state(zfsdev_state, minor); 544 zs->zss_type = ZSST_ZVOL; 545 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 546#else /* !sun */ 547 548 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 549 if (error) { 550 ASSERT(error == 0); 551 dmu_objset_disown(os, zvol_tag); 552 mutex_exit(&spa_namespace_lock); 553 return (error); 554 } 555 556 DROP_GIANT(); 557 g_topology_lock(); 558 zv = zvol_geom_create(name); 559 zv->zv_volsize = volsize; 560 zv->zv_provider->mediasize = zv->zv_volsize; 561 562#endif /* !sun */ 563 564 (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 565 zv->zv_min_bs = DEV_BSHIFT; 566 zv->zv_objset = os; 567 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) 568 zv->zv_flags |= ZVOL_RDONLY; 569 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 570 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 571 sizeof (rl_t), offsetof(rl_t, r_node)); 572 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 573 offsetof(zvol_extent_t, ze_node)); 574 /* get and cache the blocksize */ 575 error = dmu_object_info(os, ZVOL_OBJ, &doi); 576 ASSERT(error == 0); 577 zv->zv_volblocksize = doi.doi_data_block_size; 578 579 if (spa_writeable(dmu_objset_spa(os))) { 580 if (zil_replay_disable) 581 zil_destroy(dmu_objset_zil(os), B_FALSE); 582 else 583 zil_replay(os, zv, zvol_replay_vector); 584 } 585 dmu_objset_disown(os, FTAG); 586 zv->zv_objset = NULL; 587 588 zvol_minors++; 589 590 mutex_exit(&spa_namespace_lock); 591 592 zvol_geom_run(zv); 593 594 g_topology_unlock(); 595 PICKUP_GIANT(); 596 597 ZFS_LOG(1, "ZVOL %s created.", name); 598 599 return (0); 600} 601 602/* 603 * Remove minor node for the specified volume. 604 */ 605static int 606zvol_remove_zv(zvol_state_t *zv) 607{ 608#ifdef sun 609 minor_t minor = zv->zv_minor; 610#endif 611 612 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 613 if (zv->zv_total_opens != 0) 614 return (SET_ERROR(EBUSY)); 615 616 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); 617 618#ifdef sun 619 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 620 ddi_remove_minor_node(zfs_dip, nmbuf); 621#endif /* sun */ 622 623 avl_destroy(&zv->zv_znode.z_range_avl); 624 mutex_destroy(&zv->zv_znode.z_range_lock); 625 626 zvol_geom_destroy(zv); 627 628 zvol_minors--; 629 return (0); 630} 631 632int 633zvol_remove_minor(const char *name) 634{ 635 zvol_state_t *zv; 636 int rc; 637 638 mutex_enter(&spa_namespace_lock); 639 if ((zv = zvol_minor_lookup(name)) == NULL) { 640 mutex_exit(&spa_namespace_lock); 641 return (SET_ERROR(ENXIO)); 642 } 643 g_topology_lock(); 644 rc = zvol_remove_zv(zv); 645 g_topology_unlock(); 646 mutex_exit(&spa_namespace_lock); 647 return (rc); 648} 649 650int 651zvol_first_open(zvol_state_t *zv) 652{ 653 objset_t *os; 654 uint64_t volsize; 655 int error; 656 uint64_t readonly; 657 658 /* lie and say we're read-only */ 659 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 660 zvol_tag, &os); 661 if (error) 662 return (error); 663 664 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 665 if (error) { 666 ASSERT(error == 0); 667 dmu_objset_disown(os, zvol_tag); 668 return (error); 669 } 670 zv->zv_objset = os; 671 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 672 if (error) { 673 dmu_objset_disown(os, zvol_tag); 674 return (error); 675 } 676 zv->zv_volsize = volsize; 677 zv->zv_zilog = zil_open(os, zvol_get_data); 678 zvol_size_changed(zv); 679 680 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 681 NULL) == 0); 682 if (readonly || dmu_objset_is_snapshot(os) || 683 !spa_writeable(dmu_objset_spa(os))) 684 zv->zv_flags |= ZVOL_RDONLY; 685 else 686 zv->zv_flags &= ~ZVOL_RDONLY; 687 return (error); 688} 689 690void 691zvol_last_close(zvol_state_t *zv) 692{ 693 zil_close(zv->zv_zilog); 694 zv->zv_zilog = NULL; 695 696 dmu_buf_rele(zv->zv_dbuf, zvol_tag); 697 zv->zv_dbuf = NULL; 698 699 /* 700 * Evict cached data 701 */ 702 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && 703 !(zv->zv_flags & ZVOL_RDONLY)) 704 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 705 dmu_objset_evict_dbufs(zv->zv_objset); 706 707 dmu_objset_disown(zv->zv_objset, zvol_tag); 708 zv->zv_objset = NULL; 709} 710 711#ifdef sun 712int 713zvol_prealloc(zvol_state_t *zv) 714{ 715 objset_t *os = zv->zv_objset; 716 dmu_tx_t *tx; 717 uint64_t refd, avail, usedobjs, availobjs; 718 uint64_t resid = zv->zv_volsize; 719 uint64_t off = 0; 720 721 /* Check the space usage before attempting to allocate the space */ 722 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 723 if (avail < zv->zv_volsize) 724 return (SET_ERROR(ENOSPC)); 725 726 /* Free old extents if they exist */ 727 zvol_free_extents(zv); 728 729 while (resid != 0) { 730 int error; 731 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); 732 733 tx = dmu_tx_create(os); 734 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 735 error = dmu_tx_assign(tx, TXG_WAIT); 736 if (error) { 737 dmu_tx_abort(tx); 738 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 739 return (error); 740 } 741 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 742 dmu_tx_commit(tx); 743 off += bytes; 744 resid -= bytes; 745 } 746 txg_wait_synced(dmu_objset_pool(os), 0); 747 748 return (0); 749} 750#endif /* sun */ 751 752static int 753zvol_update_volsize(objset_t *os, uint64_t volsize) 754{ 755 dmu_tx_t *tx; 756 int error; 757 758 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 759 760 tx = dmu_tx_create(os); 761 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 762 error = dmu_tx_assign(tx, TXG_WAIT); 763 if (error) { 764 dmu_tx_abort(tx); 765 return (error); 766 } 767 768 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 769 &volsize, tx); 770 dmu_tx_commit(tx); 771 772 if (error == 0) 773 error = dmu_free_long_range(os, 774 ZVOL_OBJ, volsize, DMU_OBJECT_END); 775 return (error); 776} 777 778void 779zvol_remove_minors(const char *name) 780{ 781 struct g_geom *gp, *gptmp; 782 struct g_provider *pp; 783 zvol_state_t *zv; 784 size_t namelen; 785 786 namelen = strlen(name); 787 788 DROP_GIANT(); 789 mutex_enter(&spa_namespace_lock); 790 g_topology_lock(); 791 792 LIST_FOREACH_SAFE(gp, &zfs_zvol_class.geom, geom, gptmp) { 793 pp = LIST_FIRST(&gp->provider); 794 if (pp == NULL) 795 continue; 796 zv = pp->private; 797 if (zv == NULL) 798 continue; 799 if (strcmp(zv->zv_name, name) == 0 || 800 (strncmp(zv->zv_name, name, namelen) == 0 && 801 zv->zv_name[namelen] == '/')) { 802 (void) zvol_remove_zv(zv); 803 } 804 } 805 806 g_topology_unlock(); 807 mutex_exit(&spa_namespace_lock); 808 PICKUP_GIANT(); 809} 810 811int 812zvol_set_volsize(const char *name, major_t maj, uint64_t volsize) 813{ 814 zvol_state_t *zv = NULL; 815 objset_t *os; 816 int error; 817 dmu_object_info_t doi; 818 uint64_t old_volsize = 0ULL; 819 uint64_t readonly; 820 821 mutex_enter(&spa_namespace_lock); 822 zv = zvol_minor_lookup(name); 823 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 824 mutex_exit(&spa_namespace_lock); 825 return (error); 826 } 827 828 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 829 (error = zvol_check_volsize(volsize, 830 doi.doi_data_block_size)) != 0) 831 goto out; 832 833 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, 834 NULL) == 0); 835 if (readonly) { 836 error = EROFS; 837 goto out; 838 } 839 840 error = zvol_update_volsize(os, volsize); 841 /* 842 * Reinitialize the dump area to the new size. If we 843 * failed to resize the dump area then restore it back to 844 * its original size. 845 */ 846 if (zv && error == 0) { 847#ifdef ZVOL_DUMP 848 if (zv->zv_flags & ZVOL_DUMPIFIED) { 849 old_volsize = zv->zv_volsize; 850 zv->zv_volsize = volsize; 851 if ((error = zvol_dumpify(zv)) != 0 || 852 (error = dumpvp_resize()) != 0) { 853 (void) zvol_update_volsize(os, old_volsize); 854 zv->zv_volsize = old_volsize; 855 error = zvol_dumpify(zv); 856 } 857 } 858#endif /* ZVOL_DUMP */ 859 if (error == 0) { 860 zv->zv_volsize = volsize; 861 zvol_size_changed(zv); 862 } 863 } 864 865#ifdef sun 866 /* 867 * Generate a LUN expansion event. 868 */ 869 if (zv && error == 0) { 870 sysevent_id_t eid; 871 nvlist_t *attr; 872 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 873 874 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 875 zv->zv_minor); 876 877 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 878 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 879 880 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 881 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 882 883 nvlist_free(attr); 884 kmem_free(physpath, MAXPATHLEN); 885 } 886#endif /* sun */ 887 888out: 889 dmu_objset_rele(os, FTAG); 890 891 mutex_exit(&spa_namespace_lock); 892 893 return (error); 894} 895 896/*ARGSUSED*/ 897static int 898zvol_open(struct g_provider *pp, int flag, int count) 899{ 900 zvol_state_t *zv; 901 int err = 0; 902 boolean_t locked = B_FALSE; 903 904 /* 905 * Protect against recursively entering spa_namespace_lock 906 * when spa_open() is used for a pool on a (local) ZVOL(s). 907 * This is needed since we replaced upstream zfsdev_state_lock 908 * with spa_namespace_lock in the ZVOL code. 909 * We are using the same trick as spa_open(). 910 * Note that calls in zvol_first_open which need to resolve 911 * pool name to a spa object will enter spa_open() 912 * recursively, but that function already has all the 913 * necessary protection. 914 */ 915 if (!MUTEX_HELD(&spa_namespace_lock)) { 916 mutex_enter(&spa_namespace_lock); 917 locked = B_TRUE; 918 } 919 920 zv = pp->private; 921 if (zv == NULL) { 922 if (locked) 923 mutex_exit(&spa_namespace_lock); 924 return (SET_ERROR(ENXIO)); 925 } 926 927 if (zv->zv_total_opens == 0) 928 err = zvol_first_open(zv); 929 if (err) { 930 if (locked) 931 mutex_exit(&spa_namespace_lock); 932 return (err); 933 } 934 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 935 err = SET_ERROR(EROFS); 936 goto out; 937 } 938 if (zv->zv_flags & ZVOL_EXCL) { 939 err = SET_ERROR(EBUSY); 940 goto out; 941 } 942#ifdef FEXCL 943 if (flag & FEXCL) { 944 if (zv->zv_total_opens != 0) { 945 err = SET_ERROR(EBUSY); 946 goto out; 947 } 948 zv->zv_flags |= ZVOL_EXCL; 949 } 950#endif 951 952 zv->zv_total_opens += count; 953 if (locked) 954 mutex_exit(&spa_namespace_lock); 955 956 return (err); 957out: 958 if (zv->zv_total_opens == 0) 959 zvol_last_close(zv); 960 if (locked) 961 mutex_exit(&spa_namespace_lock); 962 return (err); 963} 964 965/*ARGSUSED*/ 966static int 967zvol_close(struct g_provider *pp, int flag, int count) 968{ 969 zvol_state_t *zv; 970 int error = 0; 971 boolean_t locked = B_FALSE; 972 973 /* See comment in zvol_open(). */ 974 if (!MUTEX_HELD(&spa_namespace_lock)) { 975 mutex_enter(&spa_namespace_lock); 976 locked = B_TRUE; 977 } 978 979 zv = pp->private; 980 if (zv == NULL) { 981 if (locked) 982 mutex_exit(&spa_namespace_lock); 983 return (SET_ERROR(ENXIO)); 984 } 985 986 if (zv->zv_flags & ZVOL_EXCL) { 987 ASSERT(zv->zv_total_opens == 1); 988 zv->zv_flags &= ~ZVOL_EXCL; 989 } 990 991 /* 992 * If the open count is zero, this is a spurious close. 993 * That indicates a bug in the kernel / DDI framework. 994 */ 995 ASSERT(zv->zv_total_opens != 0); 996 997 /* 998 * You may get multiple opens, but only one close. 999 */ 1000 zv->zv_total_opens -= count; 1001 1002 if (zv->zv_total_opens == 0) 1003 zvol_last_close(zv); 1004 1005 if (locked) 1006 mutex_exit(&spa_namespace_lock); 1007 return (error); 1008} 1009 1010static void 1011zvol_get_done(zgd_t *zgd, int error) 1012{ 1013 if (zgd->zgd_db) 1014 dmu_buf_rele(zgd->zgd_db, zgd); 1015 1016 zfs_range_unlock(zgd->zgd_rl); 1017 1018 if (error == 0 && zgd->zgd_bp) 1019 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1020 1021 kmem_free(zgd, sizeof (zgd_t)); 1022} 1023 1024/* 1025 * Get data to generate a TX_WRITE intent log record. 1026 */ 1027static int 1028zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1029{ 1030 zvol_state_t *zv = arg; 1031 objset_t *os = zv->zv_objset; 1032 uint64_t object = ZVOL_OBJ; 1033 uint64_t offset = lr->lr_offset; 1034 uint64_t size = lr->lr_length; /* length of user data */ 1035 blkptr_t *bp = &lr->lr_blkptr; 1036 dmu_buf_t *db; 1037 zgd_t *zgd; 1038 int error; 1039 1040 ASSERT(zio != NULL); 1041 ASSERT(size != 0); 1042 1043 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1044 zgd->zgd_zilog = zv->zv_zilog; 1045 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 1046 1047 /* 1048 * Write records come in two flavors: immediate and indirect. 1049 * For small writes it's cheaper to store the data with the 1050 * log record (immediate); for large writes it's cheaper to 1051 * sync the data and get a pointer to it (indirect) so that 1052 * we don't have to write the data twice. 1053 */ 1054 if (buf != NULL) { /* immediate write */ 1055 error = dmu_read(os, object, offset, size, buf, 1056 DMU_READ_NO_PREFETCH); 1057 } else { 1058 size = zv->zv_volblocksize; 1059 offset = P2ALIGN(offset, size); 1060 error = dmu_buf_hold(os, object, offset, zgd, &db, 1061 DMU_READ_NO_PREFETCH); 1062 if (error == 0) { 1063 blkptr_t *obp = dmu_buf_get_blkptr(db); 1064 if (obp) { 1065 ASSERT(BP_IS_HOLE(bp)); 1066 *bp = *obp; 1067 } 1068 1069 zgd->zgd_db = db; 1070 zgd->zgd_bp = bp; 1071 1072 ASSERT(db->db_offset == offset); 1073 ASSERT(db->db_size == size); 1074 1075 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1076 zvol_get_done, zgd); 1077 1078 if (error == 0) 1079 return (0); 1080 } 1081 } 1082 1083 zvol_get_done(zgd, error); 1084 1085 return (error); 1086} 1087 1088/* 1089 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 1090 * 1091 * We store data in the log buffers if it's small enough. 1092 * Otherwise we will later flush the data out via dmu_sync(). 1093 */ 1094ssize_t zvol_immediate_write_sz = 32768; 1095 1096static void 1097zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1098 boolean_t sync) 1099{ 1100 uint32_t blocksize = zv->zv_volblocksize; 1101 zilog_t *zilog = zv->zv_zilog; 1102 boolean_t slogging; 1103 ssize_t immediate_write_sz; 1104 1105 if (zil_replaying(zilog, tx)) 1106 return; 1107 1108 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1109 ? 0 : zvol_immediate_write_sz; 1110 1111 slogging = spa_has_slogs(zilog->zl_spa) && 1112 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 1113 1114 while (resid) { 1115 itx_t *itx; 1116 lr_write_t *lr; 1117 ssize_t len; 1118 itx_wr_state_t write_state; 1119 1120 /* 1121 * Unlike zfs_log_write() we can be called with 1122 * upto DMU_MAX_ACCESS/2 (5MB) writes. 1123 */ 1124 if (blocksize > immediate_write_sz && !slogging && 1125 resid >= blocksize && off % blocksize == 0) { 1126 write_state = WR_INDIRECT; /* uses dmu_sync */ 1127 len = blocksize; 1128 } else if (sync) { 1129 write_state = WR_COPIED; 1130 len = MIN(ZIL_MAX_LOG_DATA, resid); 1131 } else { 1132 write_state = WR_NEED_COPY; 1133 len = MIN(ZIL_MAX_LOG_DATA, resid); 1134 } 1135 1136 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1137 (write_state == WR_COPIED ? len : 0)); 1138 lr = (lr_write_t *)&itx->itx_lr; 1139 if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 1140 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1141 zil_itx_destroy(itx); 1142 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1143 lr = (lr_write_t *)&itx->itx_lr; 1144 write_state = WR_NEED_COPY; 1145 } 1146 1147 itx->itx_wr_state = write_state; 1148 if (write_state == WR_NEED_COPY) 1149 itx->itx_sod += len; 1150 lr->lr_foid = ZVOL_OBJ; 1151 lr->lr_offset = off; 1152 lr->lr_length = len; 1153 lr->lr_blkoff = 0; 1154 BP_ZERO(&lr->lr_blkptr); 1155 1156 itx->itx_private = zv; 1157 itx->itx_sync = sync; 1158 1159 zil_itx_assign(zilog, itx, tx); 1160 1161 off += len; 1162 resid -= len; 1163 } 1164} 1165 1166#ifdef sun 1167static int 1168zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset, 1169 uint64_t size, boolean_t doread, boolean_t isdump) 1170{ 1171 vdev_disk_t *dvd; 1172 int c; 1173 int numerrors = 0; 1174 1175 if (vd->vdev_ops == &vdev_mirror_ops || 1176 vd->vdev_ops == &vdev_replacing_ops || 1177 vd->vdev_ops == &vdev_spare_ops) { 1178 for (c = 0; c < vd->vdev_children; c++) { 1179 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1180 addr, offset, origoffset, size, doread, isdump); 1181 if (err != 0) { 1182 numerrors++; 1183 } else if (doread) { 1184 break; 1185 } 1186 } 1187 } 1188 1189 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops) 1190 return (numerrors < vd->vdev_children ? 0 : EIO); 1191 1192 if (doread && !vdev_readable(vd)) 1193 return (SET_ERROR(EIO)); 1194 else if (!doread && !vdev_writeable(vd)) 1195 return (SET_ERROR(EIO)); 1196 1197 if (vd->vdev_ops == &vdev_raidz_ops) { 1198 return (vdev_raidz_physio(vd, 1199 addr, size, offset, origoffset, doread, isdump)); 1200 } 1201 1202 offset += VDEV_LABEL_START_SIZE; 1203 1204 if (ddi_in_panic() || isdump) { 1205 ASSERT(!doread); 1206 if (doread) 1207 return (SET_ERROR(EIO)); 1208 dvd = vd->vdev_tsd; 1209 ASSERT3P(dvd, !=, NULL); 1210 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1211 lbtodb(size))); 1212 } else { 1213 dvd = vd->vdev_tsd; 1214 ASSERT3P(dvd, !=, NULL); 1215 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size, 1216 offset, doread ? B_READ : B_WRITE)); 1217 } 1218} 1219 1220static int 1221zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1222 boolean_t doread, boolean_t isdump) 1223{ 1224 vdev_t *vd; 1225 int error; 1226 zvol_extent_t *ze; 1227 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1228 1229 /* Must be sector aligned, and not stradle a block boundary. */ 1230 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1231 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1232 return (SET_ERROR(EINVAL)); 1233 } 1234 ASSERT(size <= zv->zv_volblocksize); 1235 1236 /* Locate the extent this belongs to */ 1237 ze = list_head(&zv->zv_extents); 1238 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1239 offset -= ze->ze_nblks * zv->zv_volblocksize; 1240 ze = list_next(&zv->zv_extents, ze); 1241 } 1242 1243 if (ze == NULL) 1244 return (SET_ERROR(EINVAL)); 1245 1246 if (!ddi_in_panic()) 1247 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1248 1249 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1250 offset += DVA_GET_OFFSET(&ze->ze_dva); 1251 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva), 1252 size, doread, isdump); 1253 1254 if (!ddi_in_panic()) 1255 spa_config_exit(spa, SCL_STATE, FTAG); 1256 1257 return (error); 1258} 1259#endif /* sun */ 1260 1261int 1262zvol_strategy(struct bio *bp) 1263{ 1264 zvol_state_t *zv = bp->bio_to->private; 1265 uint64_t off, volsize; 1266 size_t resid; 1267 char *addr; 1268 objset_t *os; 1269 rl_t *rl; 1270 int error = 0; 1271 boolean_t doread = (bp->bio_cmd == BIO_READ); 1272 boolean_t is_dumpified; 1273 boolean_t sync; 1274 1275 if (zv == NULL) { 1276 g_io_deliver(bp, ENXIO); 1277 return (0); 1278 } 1279 1280 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) { 1281 g_io_deliver(bp, EROFS); 1282 return (0); 1283 } 1284 1285 off = bp->bio_offset; 1286 volsize = zv->zv_volsize; 1287 1288 os = zv->zv_objset; 1289 ASSERT(os != NULL); 1290 1291 addr = bp->bio_data; 1292 resid = bp->bio_length; 1293 1294 if (resid > 0 && (off < 0 || off >= volsize)) { 1295 g_io_deliver(bp, EIO); 1296 return (0); 1297 } 1298 1299#ifdef illumos 1300 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED; 1301#else 1302 is_dumpified = B_FALSE; 1303#endif 1304 sync = !doread && !is_dumpified && 1305 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; 1306 1307 /* 1308 * There must be no buffer changes when doing a dmu_sync() because 1309 * we can't change the data whilst calculating the checksum. 1310 */ 1311 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1312 doread ? RL_READER : RL_WRITER); 1313 1314 while (resid != 0 && off < volsize) { 1315 size_t size = MIN(resid, zvol_maxphys); 1316#ifdef illumos 1317 if (is_dumpified) { 1318 size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 1319 error = zvol_dumpio(zv, addr, off, size, 1320 doread, B_FALSE); 1321 } else if (doread) { 1322#else 1323 if (doread) { 1324#endif 1325 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1326 DMU_READ_PREFETCH); 1327 } else { 1328 dmu_tx_t *tx = dmu_tx_create(os); 1329 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1330 error = dmu_tx_assign(tx, TXG_WAIT); 1331 if (error) { 1332 dmu_tx_abort(tx); 1333 } else { 1334 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1335 zvol_log_write(zv, tx, off, size, sync); 1336 dmu_tx_commit(tx); 1337 } 1338 } 1339 if (error) { 1340 /* convert checksum errors into IO errors */ 1341 if (error == ECKSUM) 1342 error = SET_ERROR(EIO); 1343 break; 1344 } 1345 off += size; 1346 addr += size; 1347 resid -= size; 1348 } 1349 zfs_range_unlock(rl); 1350 1351 bp->bio_completed = bp->bio_length - resid; 1352 if (bp->bio_completed < bp->bio_length) 1353 bp->bio_error = (off > volsize ? EINVAL : error); 1354 1355 if (sync) 1356 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1357 g_io_deliver(bp, 0); 1358 1359 return (0); 1360} 1361 1362#ifdef sun 1363/* 1364 * Set the buffer count to the zvol maximum transfer. 1365 * Using our own routine instead of the default minphys() 1366 * means that for larger writes we write bigger buffers on X86 1367 * (128K instead of 56K) and flush the disk write cache less often 1368 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1369 * 56K on X86 and 128K on sparc). 1370 */ 1371void 1372zvol_minphys(struct buf *bp) 1373{ 1374 if (bp->b_bcount > zvol_maxphys) 1375 bp->b_bcount = zvol_maxphys; 1376} 1377 1378int 1379zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1380{ 1381 minor_t minor = getminor(dev); 1382 zvol_state_t *zv; 1383 int error = 0; 1384 uint64_t size; 1385 uint64_t boff; 1386 uint64_t resid; 1387 1388 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1389 if (zv == NULL) 1390 return (SET_ERROR(ENXIO)); 1391 1392 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) 1393 return (SET_ERROR(EINVAL)); 1394 1395 boff = ldbtob(blkno); 1396 resid = ldbtob(nblocks); 1397 1398 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1399 1400 while (resid) { 1401 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1402 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1403 if (error) 1404 break; 1405 boff += size; 1406 addr += size; 1407 resid -= size; 1408 } 1409 1410 return (error); 1411} 1412 1413/*ARGSUSED*/ 1414int 1415zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1416{ 1417 minor_t minor = getminor(dev); 1418 zvol_state_t *zv; 1419 uint64_t volsize; 1420 rl_t *rl; 1421 int error = 0; 1422 1423 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1424 if (zv == NULL) 1425 return (SET_ERROR(ENXIO)); 1426 1427 volsize = zv->zv_volsize; 1428 if (uio->uio_resid > 0 && 1429 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1430 return (SET_ERROR(EIO)); 1431 1432 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1433 error = physio(zvol_strategy, NULL, dev, B_READ, 1434 zvol_minphys, uio); 1435 return (error); 1436 } 1437 1438 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1439 RL_READER); 1440 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1441 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1442 1443 /* don't read past the end */ 1444 if (bytes > volsize - uio->uio_loffset) 1445 bytes = volsize - uio->uio_loffset; 1446 1447 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1448 if (error) { 1449 /* convert checksum errors into IO errors */ 1450 if (error == ECKSUM) 1451 error = SET_ERROR(EIO); 1452 break; 1453 } 1454 } 1455 zfs_range_unlock(rl); 1456 return (error); 1457} 1458 1459/*ARGSUSED*/ 1460int 1461zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1462{ 1463 minor_t minor = getminor(dev); 1464 zvol_state_t *zv; 1465 uint64_t volsize; 1466 rl_t *rl; 1467 int error = 0; 1468 boolean_t sync; 1469 1470 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1471 if (zv == NULL) 1472 return (SET_ERROR(ENXIO)); 1473 1474 volsize = zv->zv_volsize; 1475 if (uio->uio_resid > 0 && 1476 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1477 return (SET_ERROR(EIO)); 1478 1479 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1480 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1481 zvol_minphys, uio); 1482 return (error); 1483 } 1484 1485 sync = !(zv->zv_flags & ZVOL_WCE) || 1486 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1487 1488 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1489 RL_WRITER); 1490 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1491 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1492 uint64_t off = uio->uio_loffset; 1493 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1494 1495 if (bytes > volsize - off) /* don't write past the end */ 1496 bytes = volsize - off; 1497 1498 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1499 error = dmu_tx_assign(tx, TXG_WAIT); 1500 if (error) { 1501 dmu_tx_abort(tx); 1502 break; 1503 } 1504 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1505 if (error == 0) 1506 zvol_log_write(zv, tx, off, bytes, sync); 1507 dmu_tx_commit(tx); 1508 1509 if (error) 1510 break; 1511 } 1512 zfs_range_unlock(rl); 1513 if (sync) 1514 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1515 return (error); 1516} 1517 1518int 1519zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1520{ 1521 struct uuid uuid = EFI_RESERVED; 1522 efi_gpe_t gpe = { 0 }; 1523 uint32_t crc; 1524 dk_efi_t efi; 1525 int length; 1526 char *ptr; 1527 1528 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1529 return (SET_ERROR(EFAULT)); 1530 ptr = (char *)(uintptr_t)efi.dki_data_64; 1531 length = efi.dki_length; 1532 /* 1533 * Some clients may attempt to request a PMBR for the 1534 * zvol. Currently this interface will return EINVAL to 1535 * such requests. These requests could be supported by 1536 * adding a check for lba == 0 and consing up an appropriate 1537 * PMBR. 1538 */ 1539 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1540 return (SET_ERROR(EINVAL)); 1541 1542 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1543 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1544 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1545 1546 if (efi.dki_lba == 1) { 1547 efi_gpt_t gpt = { 0 }; 1548 1549 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1550 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1551 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1552 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1553 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1554 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1555 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1556 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1557 gpt.efi_gpt_SizeOfPartitionEntry = 1558 LE_32(sizeof (efi_gpe_t)); 1559 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1560 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1561 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1562 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1563 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1564 flag)) 1565 return (SET_ERROR(EFAULT)); 1566 ptr += sizeof (gpt); 1567 length -= sizeof (gpt); 1568 } 1569 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1570 length), flag)) 1571 return (SET_ERROR(EFAULT)); 1572 return (0); 1573} 1574 1575/* 1576 * BEGIN entry points to allow external callers access to the volume. 1577 */ 1578/* 1579 * Return the volume parameters needed for access from an external caller. 1580 * These values are invariant as long as the volume is held open. 1581 */ 1582int 1583zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1584 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1585 void **rl_hdl, void **bonus_hdl) 1586{ 1587 zvol_state_t *zv; 1588 1589 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1590 if (zv == NULL) 1591 return (SET_ERROR(ENXIO)); 1592 if (zv->zv_flags & ZVOL_DUMPIFIED) 1593 return (SET_ERROR(ENXIO)); 1594 1595 ASSERT(blksize && max_xfer_len && minor_hdl && 1596 objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 1597 1598 *blksize = zv->zv_volblocksize; 1599 *max_xfer_len = (uint64_t)zvol_maxphys; 1600 *minor_hdl = zv; 1601 *objset_hdl = zv->zv_objset; 1602 *zil_hdl = zv->zv_zilog; 1603 *rl_hdl = &zv->zv_znode; 1604 *bonus_hdl = zv->zv_dbuf; 1605 return (0); 1606} 1607 1608/* 1609 * Return the current volume size to an external caller. 1610 * The size can change while the volume is open. 1611 */ 1612uint64_t 1613zvol_get_volume_size(void *minor_hdl) 1614{ 1615 zvol_state_t *zv = minor_hdl; 1616 1617 return (zv->zv_volsize); 1618} 1619 1620/* 1621 * Return the current WCE setting to an external caller. 1622 * The WCE setting can change while the volume is open. 1623 */ 1624int 1625zvol_get_volume_wce(void *minor_hdl) 1626{ 1627 zvol_state_t *zv = minor_hdl; 1628 1629 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 1630} 1631 1632/* 1633 * Entry point for external callers to zvol_log_write 1634 */ 1635void 1636zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 1637 boolean_t sync) 1638{ 1639 zvol_state_t *zv = minor_hdl; 1640 1641 zvol_log_write(zv, tx, off, resid, sync); 1642} 1643/* 1644 * END entry points to allow external callers access to the volume. 1645 */ 1646 1647/* 1648 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1649 */ 1650/*ARGSUSED*/ 1651int 1652zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1653{ 1654 zvol_state_t *zv; 1655 struct dk_cinfo dki; 1656 struct dk_minfo dkm; 1657 struct dk_callback *dkc; 1658 int error = 0; 1659 rl_t *rl; 1660 1661 mutex_enter(&spa_namespace_lock); 1662 1663 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 1664 1665 if (zv == NULL) { 1666 mutex_exit(&spa_namespace_lock); 1667 return (SET_ERROR(ENXIO)); 1668 } 1669 ASSERT(zv->zv_total_opens > 0); 1670 1671 switch (cmd) { 1672 1673 case DKIOCINFO: 1674 bzero(&dki, sizeof (dki)); 1675 (void) strcpy(dki.dki_cname, "zvol"); 1676 (void) strcpy(dki.dki_dname, "zvol"); 1677 dki.dki_ctype = DKC_UNKNOWN; 1678 dki.dki_unit = getminor(dev); 1679 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); 1680 mutex_exit(&spa_namespace_lock); 1681 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1682 error = SET_ERROR(EFAULT); 1683 return (error); 1684 1685 case DKIOCGMEDIAINFO: 1686 bzero(&dkm, sizeof (dkm)); 1687 dkm.dki_lbsize = 1U << zv->zv_min_bs; 1688 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1689 dkm.dki_media_type = DK_UNKNOWN; 1690 mutex_exit(&spa_namespace_lock); 1691 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1692 error = SET_ERROR(EFAULT); 1693 return (error); 1694 1695 case DKIOCGETEFI: 1696 { 1697 uint64_t vs = zv->zv_volsize; 1698 uint8_t bs = zv->zv_min_bs; 1699 1700 mutex_exit(&spa_namespace_lock); 1701 error = zvol_getefi((void *)arg, flag, vs, bs); 1702 return (error); 1703 } 1704 1705 case DKIOCFLUSHWRITECACHE: 1706 dkc = (struct dk_callback *)arg; 1707 mutex_exit(&spa_namespace_lock); 1708 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1709 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1710 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1711 error = 0; 1712 } 1713 return (error); 1714 1715 case DKIOCGETWCE: 1716 { 1717 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1718 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1719 flag)) 1720 error = SET_ERROR(EFAULT); 1721 break; 1722 } 1723 case DKIOCSETWCE: 1724 { 1725 int wce; 1726 if (ddi_copyin((void *)arg, &wce, sizeof (int), 1727 flag)) { 1728 error = SET_ERROR(EFAULT); 1729 break; 1730 } 1731 if (wce) { 1732 zv->zv_flags |= ZVOL_WCE; 1733 mutex_exit(&spa_namespace_lock); 1734 } else { 1735 zv->zv_flags &= ~ZVOL_WCE; 1736 mutex_exit(&spa_namespace_lock); 1737 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1738 } 1739 return (0); 1740 } 1741 1742 case DKIOCGGEOM: 1743 case DKIOCGVTOC: 1744 /* 1745 * commands using these (like prtvtoc) expect ENOTSUP 1746 * since we're emulating an EFI label 1747 */ 1748 error = SET_ERROR(ENOTSUP); 1749 break; 1750 1751 case DKIOCDUMPINIT: 1752 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1753 RL_WRITER); 1754 error = zvol_dumpify(zv); 1755 zfs_range_unlock(rl); 1756 break; 1757 1758 case DKIOCDUMPFINI: 1759 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 1760 break; 1761 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1762 RL_WRITER); 1763 error = zvol_dump_fini(zv); 1764 zfs_range_unlock(rl); 1765 break; 1766 1767 case DKIOCFREE: 1768 { 1769 dkioc_free_t df; 1770 dmu_tx_t *tx; 1771 1772 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) { 1773 error = SET_ERROR(EFAULT); 1774 break; 1775 } 1776 1777 /* 1778 * Apply Postel's Law to length-checking. If they overshoot, 1779 * just blank out until the end, if there's a need to blank 1780 * out anything. 1781 */ 1782 if (df.df_start >= zv->zv_volsize) 1783 break; /* No need to do anything... */ 1784 if (df.df_start + df.df_length > zv->zv_volsize) 1785 df.df_length = DMU_OBJECT_END; 1786 1787 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length, 1788 RL_WRITER); 1789 tx = dmu_tx_create(zv->zv_objset); 1790 error = dmu_tx_assign(tx, TXG_WAIT); 1791 if (error != 0) { 1792 dmu_tx_abort(tx); 1793 } else { 1794 zvol_log_truncate(zv, tx, df.df_start, 1795 df.df_length, B_TRUE); 1796 dmu_tx_commit(tx); 1797 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1798 df.df_start, df.df_length); 1799 } 1800 1801 zfs_range_unlock(rl); 1802 1803 if (error == 0) { 1804 /* 1805 * If the write-cache is disabled or 'sync' property 1806 * is set to 'always' then treat this as a synchronous 1807 * operation (i.e. commit to zil). 1808 */ 1809 if (!(zv->zv_flags & ZVOL_WCE) || 1810 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) 1811 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1812 1813 /* 1814 * If the caller really wants synchronous writes, and 1815 * can't wait for them, don't return until the write 1816 * is done. 1817 */ 1818 if (df.df_flags & DF_WAIT_SYNC) { 1819 txg_wait_synced( 1820 dmu_objset_pool(zv->zv_objset), 0); 1821 } 1822 } 1823 break; 1824 } 1825 1826 default: 1827 error = SET_ERROR(ENOTTY); 1828 break; 1829 1830 } 1831 mutex_exit(&spa_namespace_lock); 1832 return (error); 1833} 1834#endif /* sun */ 1835 1836int 1837zvol_busy(void) 1838{ 1839 return (zvol_minors != 0); 1840} 1841 1842void 1843zvol_init(void) 1844{ 1845 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 1846 1) == 0); 1847 ZFS_LOG(1, "ZVOL Initialized."); 1848} 1849 1850void 1851zvol_fini(void) 1852{ 1853 ddi_soft_state_fini(&zfsdev_state); 1854 ZFS_LOG(1, "ZVOL Deinitialized."); 1855} 1856 1857#ifdef sun 1858/*ARGSUSED*/ 1859static int 1860zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx) 1861{ 1862 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1863 1864 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 1865 return (1); 1866 return (0); 1867} 1868 1869/*ARGSUSED*/ 1870static void 1871zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx) 1872{ 1873 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1874 1875 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx); 1876} 1877 1878static int 1879zvol_dump_init(zvol_state_t *zv, boolean_t resize) 1880{ 1881 dmu_tx_t *tx; 1882 int error; 1883 objset_t *os = zv->zv_objset; 1884 spa_t *spa = dmu_objset_spa(os); 1885 vdev_t *vd = spa->spa_root_vdev; 1886 nvlist_t *nv = NULL; 1887 uint64_t version = spa_version(spa); 1888 enum zio_checksum checksum; 1889 1890 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1891 ASSERT(vd->vdev_ops == &vdev_root_ops); 1892 1893 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 1894 DMU_OBJECT_END); 1895 /* wait for dmu_free_long_range to actually free the blocks */ 1896 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1897 1898 /* 1899 * If the pool on which the dump device is being initialized has more 1900 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is 1901 * enabled. If so, bump that feature's counter to indicate that the 1902 * feature is active. We also check the vdev type to handle the 1903 * following case: 1904 * # zpool create test raidz disk1 disk2 disk3 1905 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev), 1906 * the raidz vdev itself has 3 children. 1907 */ 1908 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) { 1909 if (!spa_feature_is_enabled(spa, 1910 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 1911 return (SET_ERROR(ENOTSUP)); 1912 (void) dsl_sync_task(spa_name(spa), 1913 zfs_mvdev_dump_feature_check, 1914 zfs_mvdev_dump_activate_feature_sync, NULL, 2); 1915 } 1916 1917 tx = dmu_tx_create(os); 1918 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1919 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1920 error = dmu_tx_assign(tx, TXG_WAIT); 1921 if (error) { 1922 dmu_tx_abort(tx); 1923 return (error); 1924 } 1925 1926 /* 1927 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum 1928 * function. Otherwise, use the old default -- OFF. 1929 */ 1930 checksum = spa_feature_is_active(spa, 1931 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY : 1932 ZIO_CHECKSUM_OFF; 1933 1934 /* 1935 * If we are resizing the dump device then we only need to 1936 * update the refreservation to match the newly updated 1937 * zvolsize. Otherwise, we save off the original state of the 1938 * zvol so that we can restore them if the zvol is ever undumpified. 1939 */ 1940 if (resize) { 1941 error = zap_update(os, ZVOL_ZAP_OBJ, 1942 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1943 &zv->zv_volsize, tx); 1944 } else { 1945 uint64_t checksum, compress, refresrv, vbs, dedup; 1946 1947 error = dsl_prop_get_integer(zv->zv_name, 1948 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 1949 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1950 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL); 1951 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1952 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL); 1953 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1954 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL); 1955 if (version >= SPA_VERSION_DEDUP) { 1956 error = error ? error : 1957 dsl_prop_get_integer(zv->zv_name, 1958 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 1959 } 1960 1961 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1962 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 1963 &compress, tx); 1964 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1965 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx); 1966 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1967 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1968 &refresrv, tx); 1969 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1970 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 1971 &vbs, tx); 1972 error = error ? error : dmu_object_set_blocksize( 1973 os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx); 1974 if (version >= SPA_VERSION_DEDUP) { 1975 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1976 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 1977 &dedup, tx); 1978 } 1979 if (error == 0) 1980 zv->zv_volblocksize = SPA_MAXBLOCKSIZE; 1981 } 1982 dmu_tx_commit(tx); 1983 1984 /* 1985 * We only need update the zvol's property if we are initializing 1986 * the dump area for the first time. 1987 */ 1988 if (!resize) { 1989 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1990 VERIFY(nvlist_add_uint64(nv, 1991 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 1992 VERIFY(nvlist_add_uint64(nv, 1993 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 1994 ZIO_COMPRESS_OFF) == 0); 1995 VERIFY(nvlist_add_uint64(nv, 1996 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 1997 checksum) == 0); 1998 if (version >= SPA_VERSION_DEDUP) { 1999 VERIFY(nvlist_add_uint64(nv, 2000 zfs_prop_to_name(ZFS_PROP_DEDUP), 2001 ZIO_CHECKSUM_OFF) == 0); 2002 } 2003 2004 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2005 nv, NULL); 2006 nvlist_free(nv); 2007 2008 if (error) 2009 return (error); 2010 } 2011 2012 /* Allocate the space for the dump */ 2013 error = zvol_prealloc(zv); 2014 return (error); 2015} 2016 2017static int 2018zvol_dumpify(zvol_state_t *zv) 2019{ 2020 int error = 0; 2021 uint64_t dumpsize = 0; 2022 dmu_tx_t *tx; 2023 objset_t *os = zv->zv_objset; 2024 2025 if (zv->zv_flags & ZVOL_RDONLY) 2026 return (SET_ERROR(EROFS)); 2027 2028 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 2029 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 2030 boolean_t resize = (dumpsize > 0); 2031 2032 if ((error = zvol_dump_init(zv, resize)) != 0) { 2033 (void) zvol_dump_fini(zv); 2034 return (error); 2035 } 2036 } 2037 2038 /* 2039 * Build up our lba mapping. 2040 */ 2041 error = zvol_get_lbas(zv); 2042 if (error) { 2043 (void) zvol_dump_fini(zv); 2044 return (error); 2045 } 2046 2047 tx = dmu_tx_create(os); 2048 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2049 error = dmu_tx_assign(tx, TXG_WAIT); 2050 if (error) { 2051 dmu_tx_abort(tx); 2052 (void) zvol_dump_fini(zv); 2053 return (error); 2054 } 2055 2056 zv->zv_flags |= ZVOL_DUMPIFIED; 2057 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 2058 &zv->zv_volsize, tx); 2059 dmu_tx_commit(tx); 2060 2061 if (error) { 2062 (void) zvol_dump_fini(zv); 2063 return (error); 2064 } 2065 2066 txg_wait_synced(dmu_objset_pool(os), 0); 2067 return (0); 2068} 2069 2070static int 2071zvol_dump_fini(zvol_state_t *zv) 2072{ 2073 dmu_tx_t *tx; 2074 objset_t *os = zv->zv_objset; 2075 nvlist_t *nv; 2076 int error = 0; 2077 uint64_t checksum, compress, refresrv, vbs, dedup; 2078 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 2079 2080 /* 2081 * Attempt to restore the zvol back to its pre-dumpified state. 2082 * This is a best-effort attempt as it's possible that not all 2083 * of these properties were initialized during the dumpify process 2084 * (i.e. error during zvol_dump_init). 2085 */ 2086 2087 tx = dmu_tx_create(os); 2088 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2089 error = dmu_tx_assign(tx, TXG_WAIT); 2090 if (error) { 2091 dmu_tx_abort(tx); 2092 return (error); 2093 } 2094 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 2095 dmu_tx_commit(tx); 2096 2097 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2098 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 2099 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2100 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 2101 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2102 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 2103 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2104 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 2105 2106 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2107 (void) nvlist_add_uint64(nv, 2108 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 2109 (void) nvlist_add_uint64(nv, 2110 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 2111 (void) nvlist_add_uint64(nv, 2112 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 2113 if (version >= SPA_VERSION_DEDUP && 2114 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2115 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 2116 (void) nvlist_add_uint64(nv, 2117 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 2118 } 2119 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2120 nv, NULL); 2121 nvlist_free(nv); 2122 2123 zvol_free_extents(zv); 2124 zv->zv_flags &= ~ZVOL_DUMPIFIED; 2125 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 2126 /* wait for dmu_free_long_range to actually free the blocks */ 2127 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2128 tx = dmu_tx_create(os); 2129 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2130 error = dmu_tx_assign(tx, TXG_WAIT); 2131 if (error) { 2132 dmu_tx_abort(tx); 2133 return (error); 2134 } 2135 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 2136 zv->zv_volblocksize = vbs; 2137 dmu_tx_commit(tx); 2138 2139 return (0); 2140} 2141#endif /* sun */ 2142 2143static zvol_state_t * 2144zvol_geom_create(const char *name) 2145{ 2146 struct g_provider *pp; 2147 struct g_geom *gp; 2148 zvol_state_t *zv; 2149 2150 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name); 2151 gp->start = zvol_geom_start; 2152 gp->access = zvol_geom_access; 2153 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name); 2154 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 2155 pp->sectorsize = DEV_BSIZE; 2156 2157 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP); 2158 zv->zv_provider = pp; 2159 zv->zv_state = 0; 2160 bioq_init(&zv->zv_queue); 2161 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF); 2162 2163 pp->private = zv; 2164 2165 return (zv); 2166} 2167 2168static void 2169zvol_geom_run(zvol_state_t *zv) 2170{ 2171 struct g_provider *pp; 2172 2173 pp = zv->zv_provider; 2174 g_error_provider(pp, 0); 2175 2176 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0, 2177 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER)); 2178} 2179 2180static void 2181zvol_geom_destroy(zvol_state_t *zv) 2182{ 2183 struct g_provider *pp; 2184 2185 g_topology_assert(); 2186 2187 mtx_lock(&zv->zv_queue_mtx); 2188 zv->zv_state = 1; 2189 wakeup_one(&zv->zv_queue); 2190 while (zv->zv_state != 2) 2191 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0); 2192 mtx_destroy(&zv->zv_queue_mtx); 2193 2194 pp = zv->zv_provider; 2195 zv->zv_provider = NULL; 2196 pp->private = NULL; 2197 g_wither_geom(pp->geom, ENXIO); 2198 2199 kmem_free(zv, sizeof(*zv)); 2200} 2201 2202static int 2203zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) 2204{ 2205 int count, error, flags; 2206 2207 g_topology_assert(); 2208 2209 /* 2210 * To make it easier we expect either open or close, but not both 2211 * at the same time. 2212 */ 2213 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) || 2214 (acr <= 0 && acw <= 0 && ace <= 0), 2215 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", 2216 pp->name, acr, acw, ace)); 2217 2218 if (pp->private == NULL) { 2219 if (acr <= 0 && acw <= 0 && ace <= 0) 2220 return (0); 2221 return (pp->error); 2222 } 2223 2224 /* 2225 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0, 2226 * because GEOM already handles that and handles it a bit differently. 2227 * GEOM allows for multiple read/exclusive consumers and ZFS allows 2228 * only one exclusive consumer, no matter if it is reader or writer. 2229 * I like better the way GEOM works so I'll leave it for GEOM to 2230 * decide what to do. 2231 */ 2232 2233 count = acr + acw + ace; 2234 if (count == 0) 2235 return (0); 2236 2237 flags = 0; 2238 if (acr != 0 || ace != 0) 2239 flags |= FREAD; 2240 if (acw != 0) 2241 flags |= FWRITE; 2242 2243 g_topology_unlock(); 2244 if (count > 0) 2245 error = zvol_open(pp, flags, count); 2246 else 2247 error = zvol_close(pp, flags, -count); 2248 g_topology_lock(); 2249 return (error); 2250} 2251 2252static void 2253zvol_geom_start(struct bio *bp) 2254{ 2255 zvol_state_t *zv; 2256 boolean_t first; 2257 2258 zv = bp->bio_to->private; 2259 ASSERT(zv != NULL); 2260 switch (bp->bio_cmd) { 2261 case BIO_FLUSH: 2262 if (!THREAD_CAN_SLEEP()) 2263 goto enqueue; 2264 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2265 g_io_deliver(bp, 0); 2266 break; 2267 case BIO_READ: 2268 case BIO_WRITE: 2269 if (!THREAD_CAN_SLEEP()) 2270 goto enqueue; 2271 zvol_strategy(bp); 2272 break; 2273 case BIO_GETATTR: 2274 case BIO_DELETE: 2275 default: 2276 g_io_deliver(bp, EOPNOTSUPP); 2277 break; 2278 } 2279 return; 2280 2281enqueue: 2282 mtx_lock(&zv->zv_queue_mtx); 2283 first = (bioq_first(&zv->zv_queue) == NULL); 2284 bioq_insert_tail(&zv->zv_queue, bp); 2285 mtx_unlock(&zv->zv_queue_mtx); 2286 if (first) 2287 wakeup_one(&zv->zv_queue); 2288} 2289 2290static void 2291zvol_geom_worker(void *arg) 2292{ 2293 zvol_state_t *zv; 2294 struct bio *bp; 2295 2296 thread_lock(curthread); 2297 sched_prio(curthread, PRIBIO); 2298 thread_unlock(curthread); 2299 2300 zv = arg; 2301 for (;;) { 2302 mtx_lock(&zv->zv_queue_mtx); 2303 bp = bioq_takefirst(&zv->zv_queue); 2304 if (bp == NULL) { 2305 if (zv->zv_state == 1) { 2306 zv->zv_state = 2; 2307 wakeup(&zv->zv_state); 2308 mtx_unlock(&zv->zv_queue_mtx); 2309 kthread_exit(); 2310 } 2311 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP, 2312 "zvol:io", 0); 2313 continue; 2314 } 2315 mtx_unlock(&zv->zv_queue_mtx); 2316 switch (bp->bio_cmd) { 2317 case BIO_FLUSH: 2318 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2319 g_io_deliver(bp, 0); 2320 break; 2321 case BIO_READ: 2322 case BIO_WRITE: 2323 zvol_strategy(bp); 2324 break; 2325 } 2326 } 2327} 2328 2329extern boolean_t dataset_name_hidden(const char *name); 2330 2331static int 2332zvol_create_snapshots(objset_t *os, const char *name) 2333{ 2334 uint64_t cookie, obj; 2335 char *sname; 2336 int error, len; 2337 2338 cookie = obj = 0; 2339 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2340 2341#if 0 2342 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL, 2343 DS_FIND_SNAPSHOTS); 2344#endif 2345 2346 for (;;) { 2347 len = snprintf(sname, MAXPATHLEN, "%s@", name); 2348 if (len >= MAXPATHLEN) { 2349 dmu_objset_rele(os, FTAG); 2350 error = ENAMETOOLONG; 2351 break; 2352 } 2353 2354 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 2355 error = dmu_snapshot_list_next(os, MAXPATHLEN - len, 2356 sname + len, &obj, &cookie, NULL); 2357 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 2358 if (error != 0) { 2359 if (error == ENOENT) 2360 error = 0; 2361 break; 2362 } 2363 2364 if ((error = zvol_create_minor(sname)) != 0) { 2365 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2366 sname, error); 2367 break; 2368 } 2369 } 2370 2371 kmem_free(sname, MAXPATHLEN); 2372 return (error); 2373} 2374 2375int 2376zvol_create_minors(const char *name) 2377{ 2378 uint64_t cookie; 2379 objset_t *os; 2380 char *osname, *p; 2381 int error, len; 2382 2383 if (dataset_name_hidden(name)) 2384 return (0); 2385 2386 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2387 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2388 name, error); 2389 return (error); 2390 } 2391 if (dmu_objset_type(os) == DMU_OST_ZVOL) { 2392 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG); 2393 dsl_pool_rele(dmu_objset_pool(os), FTAG); 2394 if ((error = zvol_create_minor(name)) == 0) 2395 error = zvol_create_snapshots(os, name); 2396 else { 2397 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2398 name, error); 2399 } 2400 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG); 2401 dsl_dataset_rele(os->os_dsl_dataset, FTAG); 2402 return (error); 2403 } 2404 if (dmu_objset_type(os) != DMU_OST_ZFS) { 2405 dmu_objset_rele(os, FTAG); 2406 return (0); 2407 } 2408 2409 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2410 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) { 2411 dmu_objset_rele(os, FTAG); 2412 kmem_free(osname, MAXPATHLEN); 2413 return (ENOENT); 2414 } 2415 p = osname + strlen(osname); 2416 len = MAXPATHLEN - (p - osname); 2417 2418#if 0 2419 /* Prefetch the datasets. */ 2420 cookie = 0; 2421 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) { 2422 if (!dataset_name_hidden(osname)) 2423 (void) dmu_objset_prefetch(osname, NULL); 2424 } 2425#endif 2426 2427 cookie = 0; 2428 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL, 2429 &cookie) == 0) { 2430 dmu_objset_rele(os, FTAG); 2431 (void)zvol_create_minors(osname); 2432 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2433 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2434 name, error); 2435 return (error); 2436 } 2437 } 2438 2439 dmu_objset_rele(os, FTAG); 2440 kmem_free(osname, MAXPATHLEN); 2441 return (0); 2442} 2443 2444static void 2445zvol_rename_minor(struct g_geom *gp, const char *newname) 2446{ 2447 struct g_provider *pp; 2448 zvol_state_t *zv; 2449 2450 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2451 g_topology_assert(); 2452 2453 pp = LIST_FIRST(&gp->provider); 2454 ASSERT(pp != NULL); 2455 zv = pp->private; 2456 ASSERT(zv != NULL); 2457 2458 zv->zv_provider = NULL; 2459 g_wither_provider(pp, ENXIO); 2460 2461 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname); 2462 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 2463 pp->sectorsize = DEV_BSIZE; 2464 pp->mediasize = zv->zv_volsize; 2465 pp->private = zv; 2466 zv->zv_provider = pp; 2467 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name)); 2468 g_error_provider(pp, 0); 2469} 2470 2471void 2472zvol_rename_minors(const char *oldname, const char *newname) 2473{ 2474 char name[MAXPATHLEN]; 2475 struct g_provider *pp; 2476 struct g_geom *gp; 2477 size_t oldnamelen, newnamelen; 2478 zvol_state_t *zv; 2479 char *namebuf; 2480 2481 oldnamelen = strlen(oldname); 2482 newnamelen = strlen(newname); 2483 2484 DROP_GIANT(); 2485 mutex_enter(&spa_namespace_lock); 2486 g_topology_lock(); 2487 2488 LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) { 2489 pp = LIST_FIRST(&gp->provider); 2490 if (pp == NULL) 2491 continue; 2492 zv = pp->private; 2493 if (zv == NULL) 2494 continue; 2495 if (strcmp(zv->zv_name, oldname) == 0) { 2496 zvol_rename_minor(gp, newname); 2497 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && 2498 (zv->zv_name[oldnamelen] == '/' || 2499 zv->zv_name[oldnamelen] == '@')) { 2500 snprintf(name, sizeof(name), "%s%c%s", newname, 2501 zv->zv_name[oldnamelen], 2502 zv->zv_name + oldnamelen + 1); 2503 zvol_rename_minor(gp, name); 2504 } 2505 } 2506 2507 g_topology_unlock(); 2508 mutex_exit(&spa_namespace_lock); 2509 PICKUP_GIANT(); 2510} 2511