zvol.c revision 263397
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_IS_HOLE(bp) || 286 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 287 return (0); 288 289 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 290 ma->ma_blks++; 291 292 /* Abort immediately if we have encountered gang blocks */ 293 if (BP_IS_GANG(bp)) 294 return (SET_ERROR(EFRAGS)); 295 296 /* 297 * See if the block is at the end of the previous extent. 298 */ 299 ze = list_tail(&ma->ma_zv->zv_extents); 300 if (ze && 301 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 302 DVA_GET_OFFSET(BP_IDENTITY(bp)) == 303 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 304 ze->ze_nblks++; 305 return (0); 306 } 307 308 dprintf_bp(bp, "%s", "next blkptr:"); 309 310 /* start a new extent */ 311 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 312 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 313 ze->ze_nblks = 1; 314 list_insert_tail(&ma->ma_zv->zv_extents, ze); 315 return (0); 316} 317 318static void 319zvol_free_extents(zvol_state_t *zv) 320{ 321 zvol_extent_t *ze; 322 323 while (ze = list_head(&zv->zv_extents)) { 324 list_remove(&zv->zv_extents, ze); 325 kmem_free(ze, sizeof (zvol_extent_t)); 326 } 327} 328 329static int 330zvol_get_lbas(zvol_state_t *zv) 331{ 332 objset_t *os = zv->zv_objset; 333 struct maparg ma; 334 int err; 335 336 ma.ma_zv = zv; 337 ma.ma_blks = 0; 338 zvol_free_extents(zv); 339 340 /* commit any in-flight changes before traversing the dataset */ 341 txg_wait_synced(dmu_objset_pool(os), 0); 342 err = traverse_dataset(dmu_objset_ds(os), 0, 343 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 344 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 345 zvol_free_extents(zv); 346 return (err ? err : EIO); 347 } 348 349 return (0); 350} 351 352/* ARGSUSED */ 353void 354zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 355{ 356 zfs_creat_t *zct = arg; 357 nvlist_t *nvprops = zct->zct_props; 358 int error; 359 uint64_t volblocksize, volsize; 360 361 VERIFY(nvlist_lookup_uint64(nvprops, 362 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 363 if (nvlist_lookup_uint64(nvprops, 364 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 365 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 366 367 /* 368 * These properties must be removed from the list so the generic 369 * property setting step won't apply to them. 370 */ 371 VERIFY(nvlist_remove_all(nvprops, 372 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 373 (void) nvlist_remove_all(nvprops, 374 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 375 376 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 377 DMU_OT_NONE, 0, tx); 378 ASSERT(error == 0); 379 380 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 381 DMU_OT_NONE, 0, tx); 382 ASSERT(error == 0); 383 384 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 385 ASSERT(error == 0); 386} 387 388/* 389 * Replay a TX_WRITE ZIL transaction that didn't get committed 390 * after a system failure 391 */ 392static int 393zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 394{ 395 objset_t *os = zv->zv_objset; 396 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 397 uint64_t offset, length; 398 dmu_tx_t *tx; 399 int error; 400 401 if (byteswap) 402 byteswap_uint64_array(lr, sizeof (*lr)); 403 404 offset = lr->lr_offset; 405 length = lr->lr_length; 406 407 /* If it's a dmu_sync() block, write the whole block */ 408 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 409 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 410 if (length < blocksize) { 411 offset -= offset % blocksize; 412 length = blocksize; 413 } 414 } 415 416 tx = dmu_tx_create(os); 417 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 418 error = dmu_tx_assign(tx, TXG_WAIT); 419 if (error) { 420 dmu_tx_abort(tx); 421 } else { 422 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 423 dmu_tx_commit(tx); 424 } 425 426 return (error); 427} 428 429/* ARGSUSED */ 430static int 431zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 432{ 433 return (SET_ERROR(ENOTSUP)); 434} 435 436/* 437 * Callback vectors for replaying records. 438 * Only TX_WRITE is needed for zvol. 439 */ 440zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 441 zvol_replay_err, /* 0 no such transaction type */ 442 zvol_replay_err, /* TX_CREATE */ 443 zvol_replay_err, /* TX_MKDIR */ 444 zvol_replay_err, /* TX_MKXATTR */ 445 zvol_replay_err, /* TX_SYMLINK */ 446 zvol_replay_err, /* TX_REMOVE */ 447 zvol_replay_err, /* TX_RMDIR */ 448 zvol_replay_err, /* TX_LINK */ 449 zvol_replay_err, /* TX_RENAME */ 450 zvol_replay_write, /* TX_WRITE */ 451 zvol_replay_err, /* TX_TRUNCATE */ 452 zvol_replay_err, /* TX_SETATTR */ 453 zvol_replay_err, /* TX_ACL */ 454 zvol_replay_err, /* TX_CREATE_ACL */ 455 zvol_replay_err, /* TX_CREATE_ATTR */ 456 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 457 zvol_replay_err, /* TX_MKDIR_ACL */ 458 zvol_replay_err, /* TX_MKDIR_ATTR */ 459 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 460 zvol_replay_err, /* TX_WRITE2 */ 461}; 462 463#ifdef sun 464int 465zvol_name2minor(const char *name, minor_t *minor) 466{ 467 zvol_state_t *zv; 468 469 mutex_enter(&spa_namespace_lock); 470 zv = zvol_minor_lookup(name); 471 if (minor && zv) 472 *minor = zv->zv_minor; 473 mutex_exit(&spa_namespace_lock); 474 return (zv ? 0 : -1); 475} 476#endif /* sun */ 477 478/* 479 * Create a minor node (plus a whole lot more) for the specified volume. 480 */ 481int 482zvol_create_minor(const char *name) 483{ 484 zfs_soft_state_t *zs; 485 zvol_state_t *zv; 486 objset_t *os; 487 dmu_object_info_t doi; 488 uint64_t volsize; 489 int error; 490 491 ZFS_LOG(1, "Creating ZVOL %s...", name); 492 493 mutex_enter(&spa_namespace_lock); 494 495 if (zvol_minor_lookup(name) != NULL) { 496 mutex_exit(&spa_namespace_lock); 497 return (SET_ERROR(EEXIST)); 498 } 499 500 /* lie and say we're read-only */ 501 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os); 502 503 if (error) { 504 mutex_exit(&spa_namespace_lock); 505 return (error); 506 } 507 508#ifdef sun 509 if ((minor = zfsdev_minor_alloc()) == 0) { 510 dmu_objset_disown(os, FTAG); 511 mutex_exit(&spa_namespace_lock); 512 return (SET_ERROR(ENXIO)); 513 } 514 515 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 516 dmu_objset_disown(os, FTAG); 517 mutex_exit(&spa_namespace_lock); 518 return (SET_ERROR(EAGAIN)); 519 } 520 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 521 (char *)name); 522 523 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 524 525 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 526 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 527 ddi_soft_state_free(zfsdev_state, minor); 528 dmu_objset_disown(os, FTAG); 529 mutex_exit(&spa_namespace_lock); 530 return (SET_ERROR(EAGAIN)); 531 } 532 533 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 534 535 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 536 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 537 ddi_remove_minor_node(zfs_dip, chrbuf); 538 ddi_soft_state_free(zfsdev_state, minor); 539 dmu_objset_disown(os, FTAG); 540 mutex_exit(&spa_namespace_lock); 541 return (SET_ERROR(EAGAIN)); 542 } 543 544 zs = ddi_get_soft_state(zfsdev_state, minor); 545 zs->zss_type = ZSST_ZVOL; 546 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 547#else /* !sun */ 548 549 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 550 if (error) { 551 ASSERT(error == 0); 552 dmu_objset_disown(os, zvol_tag); 553 mutex_exit(&spa_namespace_lock); 554 return (error); 555 } 556 557 DROP_GIANT(); 558 g_topology_lock(); 559 zv = zvol_geom_create(name); 560 zv->zv_volsize = volsize; 561 zv->zv_provider->mediasize = zv->zv_volsize; 562 563#endif /* !sun */ 564 565 (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 566 zv->zv_min_bs = DEV_BSHIFT; 567 zv->zv_objset = os; 568 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) 569 zv->zv_flags |= ZVOL_RDONLY; 570 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 571 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 572 sizeof (rl_t), offsetof(rl_t, r_node)); 573 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 574 offsetof(zvol_extent_t, ze_node)); 575 /* get and cache the blocksize */ 576 error = dmu_object_info(os, ZVOL_OBJ, &doi); 577 ASSERT(error == 0); 578 zv->zv_volblocksize = doi.doi_data_block_size; 579 580 if (spa_writeable(dmu_objset_spa(os))) { 581 if (zil_replay_disable) 582 zil_destroy(dmu_objset_zil(os), B_FALSE); 583 else 584 zil_replay(os, zv, zvol_replay_vector); 585 } 586 dmu_objset_disown(os, FTAG); 587 zv->zv_objset = NULL; 588 589 zvol_minors++; 590 591 mutex_exit(&spa_namespace_lock); 592 593 zvol_geom_run(zv); 594 595 g_topology_unlock(); 596 PICKUP_GIANT(); 597 598 ZFS_LOG(1, "ZVOL %s created.", name); 599 600 return (0); 601} 602 603/* 604 * Remove minor node for the specified volume. 605 */ 606static int 607zvol_remove_zv(zvol_state_t *zv) 608{ 609#ifdef sun 610 minor_t minor = zv->zv_minor; 611#endif 612 613 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 614 if (zv->zv_total_opens != 0) 615 return (SET_ERROR(EBUSY)); 616 617 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); 618 619#ifdef sun 620 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 621 ddi_remove_minor_node(zfs_dip, nmbuf); 622#endif /* sun */ 623 624 avl_destroy(&zv->zv_znode.z_range_avl); 625 mutex_destroy(&zv->zv_znode.z_range_lock); 626 627 zvol_geom_destroy(zv); 628 629 zvol_minors--; 630 return (0); 631} 632 633int 634zvol_remove_minor(const char *name) 635{ 636 zvol_state_t *zv; 637 int rc; 638 639 mutex_enter(&spa_namespace_lock); 640 if ((zv = zvol_minor_lookup(name)) == NULL) { 641 mutex_exit(&spa_namespace_lock); 642 return (SET_ERROR(ENXIO)); 643 } 644 g_topology_lock(); 645 rc = zvol_remove_zv(zv); 646 g_topology_unlock(); 647 mutex_exit(&spa_namespace_lock); 648 return (rc); 649} 650 651int 652zvol_first_open(zvol_state_t *zv) 653{ 654 objset_t *os; 655 uint64_t volsize; 656 int error; 657 uint64_t readonly; 658 659 /* lie and say we're read-only */ 660 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 661 zvol_tag, &os); 662 if (error) 663 return (error); 664 665 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 666 if (error) { 667 ASSERT(error == 0); 668 dmu_objset_disown(os, zvol_tag); 669 return (error); 670 } 671 zv->zv_objset = os; 672 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 673 if (error) { 674 dmu_objset_disown(os, zvol_tag); 675 return (error); 676 } 677 zv->zv_volsize = volsize; 678 zv->zv_zilog = zil_open(os, zvol_get_data); 679 zvol_size_changed(zv); 680 681 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 682 NULL) == 0); 683 if (readonly || dmu_objset_is_snapshot(os) || 684 !spa_writeable(dmu_objset_spa(os))) 685 zv->zv_flags |= ZVOL_RDONLY; 686 else 687 zv->zv_flags &= ~ZVOL_RDONLY; 688 return (error); 689} 690 691void 692zvol_last_close(zvol_state_t *zv) 693{ 694 zil_close(zv->zv_zilog); 695 zv->zv_zilog = NULL; 696 697 dmu_buf_rele(zv->zv_dbuf, zvol_tag); 698 zv->zv_dbuf = NULL; 699 700 /* 701 * Evict cached data 702 */ 703 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && 704 !(zv->zv_flags & ZVOL_RDONLY)) 705 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 706 dmu_objset_evict_dbufs(zv->zv_objset); 707 708 dmu_objset_disown(zv->zv_objset, zvol_tag); 709 zv->zv_objset = NULL; 710} 711 712#ifdef sun 713int 714zvol_prealloc(zvol_state_t *zv) 715{ 716 objset_t *os = zv->zv_objset; 717 dmu_tx_t *tx; 718 uint64_t refd, avail, usedobjs, availobjs; 719 uint64_t resid = zv->zv_volsize; 720 uint64_t off = 0; 721 722 /* Check the space usage before attempting to allocate the space */ 723 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 724 if (avail < zv->zv_volsize) 725 return (SET_ERROR(ENOSPC)); 726 727 /* Free old extents if they exist */ 728 zvol_free_extents(zv); 729 730 while (resid != 0) { 731 int error; 732 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); 733 734 tx = dmu_tx_create(os); 735 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 736 error = dmu_tx_assign(tx, TXG_WAIT); 737 if (error) { 738 dmu_tx_abort(tx); 739 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 740 return (error); 741 } 742 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 743 dmu_tx_commit(tx); 744 off += bytes; 745 resid -= bytes; 746 } 747 txg_wait_synced(dmu_objset_pool(os), 0); 748 749 return (0); 750} 751#endif /* sun */ 752 753static int 754zvol_update_volsize(objset_t *os, uint64_t volsize) 755{ 756 dmu_tx_t *tx; 757 int error; 758 759 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 760 761 tx = dmu_tx_create(os); 762 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 763 error = dmu_tx_assign(tx, TXG_WAIT); 764 if (error) { 765 dmu_tx_abort(tx); 766 return (error); 767 } 768 769 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 770 &volsize, tx); 771 dmu_tx_commit(tx); 772 773 if (error == 0) 774 error = dmu_free_long_range(os, 775 ZVOL_OBJ, volsize, DMU_OBJECT_END); 776 return (error); 777} 778 779void 780zvol_remove_minors(const char *name) 781{ 782 struct g_geom *gp, *gptmp; 783 struct g_provider *pp; 784 zvol_state_t *zv; 785 size_t namelen; 786 787 namelen = strlen(name); 788 789 DROP_GIANT(); 790 mutex_enter(&spa_namespace_lock); 791 g_topology_lock(); 792 793 LIST_FOREACH_SAFE(gp, &zfs_zvol_class.geom, geom, gptmp) { 794 pp = LIST_FIRST(&gp->provider); 795 if (pp == NULL) 796 continue; 797 zv = pp->private; 798 if (zv == NULL) 799 continue; 800 if (strcmp(zv->zv_name, name) == 0 || 801 (strncmp(zv->zv_name, name, namelen) == 0 && 802 zv->zv_name[namelen] == '/')) { 803 (void) zvol_remove_zv(zv); 804 } 805 } 806 807 g_topology_unlock(); 808 mutex_exit(&spa_namespace_lock); 809 PICKUP_GIANT(); 810} 811 812int 813zvol_set_volsize(const char *name, major_t maj, uint64_t volsize) 814{ 815 zvol_state_t *zv = NULL; 816 objset_t *os; 817 int error; 818 dmu_object_info_t doi; 819 uint64_t old_volsize = 0ULL; 820 uint64_t readonly; 821 822 mutex_enter(&spa_namespace_lock); 823 zv = zvol_minor_lookup(name); 824 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 825 mutex_exit(&spa_namespace_lock); 826 return (error); 827 } 828 829 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 830 (error = zvol_check_volsize(volsize, 831 doi.doi_data_block_size)) != 0) 832 goto out; 833 834 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, 835 NULL) == 0); 836 if (readonly) { 837 error = EROFS; 838 goto out; 839 } 840 841 error = zvol_update_volsize(os, volsize); 842 /* 843 * Reinitialize the dump area to the new size. If we 844 * failed to resize the dump area then restore it back to 845 * its original size. 846 */ 847 if (zv && error == 0) { 848#ifdef ZVOL_DUMP 849 if (zv->zv_flags & ZVOL_DUMPIFIED) { 850 old_volsize = zv->zv_volsize; 851 zv->zv_volsize = volsize; 852 if ((error = zvol_dumpify(zv)) != 0 || 853 (error = dumpvp_resize()) != 0) { 854 (void) zvol_update_volsize(os, old_volsize); 855 zv->zv_volsize = old_volsize; 856 error = zvol_dumpify(zv); 857 } 858 } 859#endif /* ZVOL_DUMP */ 860 if (error == 0) { 861 zv->zv_volsize = volsize; 862 zvol_size_changed(zv); 863 } 864 } 865 866#ifdef sun 867 /* 868 * Generate a LUN expansion event. 869 */ 870 if (zv && error == 0) { 871 sysevent_id_t eid; 872 nvlist_t *attr; 873 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 874 875 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 876 zv->zv_minor); 877 878 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 879 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 880 881 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 882 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 883 884 nvlist_free(attr); 885 kmem_free(physpath, MAXPATHLEN); 886 } 887#endif /* sun */ 888 889out: 890 dmu_objset_rele(os, FTAG); 891 892 mutex_exit(&spa_namespace_lock); 893 894 return (error); 895} 896 897/*ARGSUSED*/ 898static int 899zvol_open(struct g_provider *pp, int flag, int count) 900{ 901 zvol_state_t *zv; 902 int err = 0; 903 boolean_t locked = B_FALSE; 904 905 /* 906 * Protect against recursively entering spa_namespace_lock 907 * when spa_open() is used for a pool on a (local) ZVOL(s). 908 * This is needed since we replaced upstream zfsdev_state_lock 909 * with spa_namespace_lock in the ZVOL code. 910 * We are using the same trick as spa_open(). 911 * Note that calls in zvol_first_open which need to resolve 912 * pool name to a spa object will enter spa_open() 913 * recursively, but that function already has all the 914 * necessary protection. 915 */ 916 if (!MUTEX_HELD(&spa_namespace_lock)) { 917 mutex_enter(&spa_namespace_lock); 918 locked = B_TRUE; 919 } 920 921 zv = pp->private; 922 if (zv == NULL) { 923 if (locked) 924 mutex_exit(&spa_namespace_lock); 925 return (SET_ERROR(ENXIO)); 926 } 927 928 if (zv->zv_total_opens == 0) 929 err = zvol_first_open(zv); 930 if (err) { 931 if (locked) 932 mutex_exit(&spa_namespace_lock); 933 return (err); 934 } 935 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 936 err = SET_ERROR(EROFS); 937 goto out; 938 } 939 if (zv->zv_flags & ZVOL_EXCL) { 940 err = SET_ERROR(EBUSY); 941 goto out; 942 } 943#ifdef FEXCL 944 if (flag & FEXCL) { 945 if (zv->zv_total_opens != 0) { 946 err = SET_ERROR(EBUSY); 947 goto out; 948 } 949 zv->zv_flags |= ZVOL_EXCL; 950 } 951#endif 952 953 zv->zv_total_opens += count; 954 if (locked) 955 mutex_exit(&spa_namespace_lock); 956 957 return (err); 958out: 959 if (zv->zv_total_opens == 0) 960 zvol_last_close(zv); 961 if (locked) 962 mutex_exit(&spa_namespace_lock); 963 return (err); 964} 965 966/*ARGSUSED*/ 967static int 968zvol_close(struct g_provider *pp, int flag, int count) 969{ 970 zvol_state_t *zv; 971 int error = 0; 972 boolean_t locked = B_FALSE; 973 974 /* See comment in zvol_open(). */ 975 if (!MUTEX_HELD(&spa_namespace_lock)) { 976 mutex_enter(&spa_namespace_lock); 977 locked = B_TRUE; 978 } 979 980 zv = pp->private; 981 if (zv == NULL) { 982 if (locked) 983 mutex_exit(&spa_namespace_lock); 984 return (SET_ERROR(ENXIO)); 985 } 986 987 if (zv->zv_flags & ZVOL_EXCL) { 988 ASSERT(zv->zv_total_opens == 1); 989 zv->zv_flags &= ~ZVOL_EXCL; 990 } 991 992 /* 993 * If the open count is zero, this is a spurious close. 994 * That indicates a bug in the kernel / DDI framework. 995 */ 996 ASSERT(zv->zv_total_opens != 0); 997 998 /* 999 * You may get multiple opens, but only one close. 1000 */ 1001 zv->zv_total_opens -= count; 1002 1003 if (zv->zv_total_opens == 0) 1004 zvol_last_close(zv); 1005 1006 if (locked) 1007 mutex_exit(&spa_namespace_lock); 1008 return (error); 1009} 1010 1011static void 1012zvol_get_done(zgd_t *zgd, int error) 1013{ 1014 if (zgd->zgd_db) 1015 dmu_buf_rele(zgd->zgd_db, zgd); 1016 1017 zfs_range_unlock(zgd->zgd_rl); 1018 1019 if (error == 0 && zgd->zgd_bp) 1020 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1021 1022 kmem_free(zgd, sizeof (zgd_t)); 1023} 1024 1025/* 1026 * Get data to generate a TX_WRITE intent log record. 1027 */ 1028static int 1029zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1030{ 1031 zvol_state_t *zv = arg; 1032 objset_t *os = zv->zv_objset; 1033 uint64_t object = ZVOL_OBJ; 1034 uint64_t offset = lr->lr_offset; 1035 uint64_t size = lr->lr_length; /* length of user data */ 1036 blkptr_t *bp = &lr->lr_blkptr; 1037 dmu_buf_t *db; 1038 zgd_t *zgd; 1039 int error; 1040 1041 ASSERT(zio != NULL); 1042 ASSERT(size != 0); 1043 1044 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1045 zgd->zgd_zilog = zv->zv_zilog; 1046 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 1047 1048 /* 1049 * Write records come in two flavors: immediate and indirect. 1050 * For small writes it's cheaper to store the data with the 1051 * log record (immediate); for large writes it's cheaper to 1052 * sync the data and get a pointer to it (indirect) so that 1053 * we don't have to write the data twice. 1054 */ 1055 if (buf != NULL) { /* immediate write */ 1056 error = dmu_read(os, object, offset, size, buf, 1057 DMU_READ_NO_PREFETCH); 1058 } else { 1059 size = zv->zv_volblocksize; 1060 offset = P2ALIGN(offset, size); 1061 error = dmu_buf_hold(os, object, offset, zgd, &db, 1062 DMU_READ_NO_PREFETCH); 1063 if (error == 0) { 1064 blkptr_t *obp = dmu_buf_get_blkptr(db); 1065 if (obp) { 1066 ASSERT(BP_IS_HOLE(bp)); 1067 *bp = *obp; 1068 } 1069 1070 zgd->zgd_db = db; 1071 zgd->zgd_bp = bp; 1072 1073 ASSERT(db->db_offset == offset); 1074 ASSERT(db->db_size == size); 1075 1076 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1077 zvol_get_done, zgd); 1078 1079 if (error == 0) 1080 return (0); 1081 } 1082 } 1083 1084 zvol_get_done(zgd, error); 1085 1086 return (error); 1087} 1088 1089/* 1090 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 1091 * 1092 * We store data in the log buffers if it's small enough. 1093 * Otherwise we will later flush the data out via dmu_sync(). 1094 */ 1095ssize_t zvol_immediate_write_sz = 32768; 1096 1097static void 1098zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1099 boolean_t sync) 1100{ 1101 uint32_t blocksize = zv->zv_volblocksize; 1102 zilog_t *zilog = zv->zv_zilog; 1103 boolean_t slogging; 1104 ssize_t immediate_write_sz; 1105 1106 if (zil_replaying(zilog, tx)) 1107 return; 1108 1109 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1110 ? 0 : zvol_immediate_write_sz; 1111 1112 slogging = spa_has_slogs(zilog->zl_spa) && 1113 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 1114 1115 while (resid) { 1116 itx_t *itx; 1117 lr_write_t *lr; 1118 ssize_t len; 1119 itx_wr_state_t write_state; 1120 1121 /* 1122 * Unlike zfs_log_write() we can be called with 1123 * upto DMU_MAX_ACCESS/2 (5MB) writes. 1124 */ 1125 if (blocksize > immediate_write_sz && !slogging && 1126 resid >= blocksize && off % blocksize == 0) { 1127 write_state = WR_INDIRECT; /* uses dmu_sync */ 1128 len = blocksize; 1129 } else if (sync) { 1130 write_state = WR_COPIED; 1131 len = MIN(ZIL_MAX_LOG_DATA, resid); 1132 } else { 1133 write_state = WR_NEED_COPY; 1134 len = MIN(ZIL_MAX_LOG_DATA, resid); 1135 } 1136 1137 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1138 (write_state == WR_COPIED ? len : 0)); 1139 lr = (lr_write_t *)&itx->itx_lr; 1140 if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 1141 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1142 zil_itx_destroy(itx); 1143 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1144 lr = (lr_write_t *)&itx->itx_lr; 1145 write_state = WR_NEED_COPY; 1146 } 1147 1148 itx->itx_wr_state = write_state; 1149 if (write_state == WR_NEED_COPY) 1150 itx->itx_sod += len; 1151 lr->lr_foid = ZVOL_OBJ; 1152 lr->lr_offset = off; 1153 lr->lr_length = len; 1154 lr->lr_blkoff = 0; 1155 BP_ZERO(&lr->lr_blkptr); 1156 1157 itx->itx_private = zv; 1158 itx->itx_sync = sync; 1159 1160 zil_itx_assign(zilog, itx, tx); 1161 1162 off += len; 1163 resid -= len; 1164 } 1165} 1166 1167#ifdef sun 1168static int 1169zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset, 1170 uint64_t size, boolean_t doread, boolean_t isdump) 1171{ 1172 vdev_disk_t *dvd; 1173 int c; 1174 int numerrors = 0; 1175 1176 if (vd->vdev_ops == &vdev_mirror_ops || 1177 vd->vdev_ops == &vdev_replacing_ops || 1178 vd->vdev_ops == &vdev_spare_ops) { 1179 for (c = 0; c < vd->vdev_children; c++) { 1180 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1181 addr, offset, origoffset, size, doread, isdump); 1182 if (err != 0) { 1183 numerrors++; 1184 } else if (doread) { 1185 break; 1186 } 1187 } 1188 } 1189 1190 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops) 1191 return (numerrors < vd->vdev_children ? 0 : EIO); 1192 1193 if (doread && !vdev_readable(vd)) 1194 return (SET_ERROR(EIO)); 1195 else if (!doread && !vdev_writeable(vd)) 1196 return (SET_ERROR(EIO)); 1197 1198 if (vd->vdev_ops == &vdev_raidz_ops) { 1199 return (vdev_raidz_physio(vd, 1200 addr, size, offset, origoffset, doread, isdump)); 1201 } 1202 1203 offset += VDEV_LABEL_START_SIZE; 1204 1205 if (ddi_in_panic() || isdump) { 1206 ASSERT(!doread); 1207 if (doread) 1208 return (SET_ERROR(EIO)); 1209 dvd = vd->vdev_tsd; 1210 ASSERT3P(dvd, !=, NULL); 1211 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1212 lbtodb(size))); 1213 } else { 1214 dvd = vd->vdev_tsd; 1215 ASSERT3P(dvd, !=, NULL); 1216 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size, 1217 offset, doread ? B_READ : B_WRITE)); 1218 } 1219} 1220 1221static int 1222zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1223 boolean_t doread, boolean_t isdump) 1224{ 1225 vdev_t *vd; 1226 int error; 1227 zvol_extent_t *ze; 1228 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1229 1230 /* Must be sector aligned, and not stradle a block boundary. */ 1231 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1232 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1233 return (SET_ERROR(EINVAL)); 1234 } 1235 ASSERT(size <= zv->zv_volblocksize); 1236 1237 /* Locate the extent this belongs to */ 1238 ze = list_head(&zv->zv_extents); 1239 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1240 offset -= ze->ze_nblks * zv->zv_volblocksize; 1241 ze = list_next(&zv->zv_extents, ze); 1242 } 1243 1244 if (ze == NULL) 1245 return (SET_ERROR(EINVAL)); 1246 1247 if (!ddi_in_panic()) 1248 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1249 1250 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1251 offset += DVA_GET_OFFSET(&ze->ze_dva); 1252 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva), 1253 size, doread, isdump); 1254 1255 if (!ddi_in_panic()) 1256 spa_config_exit(spa, SCL_STATE, FTAG); 1257 1258 return (error); 1259} 1260#endif /* sun */ 1261 1262int 1263zvol_strategy(struct bio *bp) 1264{ 1265 zvol_state_t *zv = bp->bio_to->private; 1266 uint64_t off, volsize; 1267 size_t resid; 1268 char *addr; 1269 objset_t *os; 1270 rl_t *rl; 1271 int error = 0; 1272 boolean_t doread = (bp->bio_cmd == BIO_READ); 1273 boolean_t is_dumpified; 1274 boolean_t sync; 1275 1276 if (zv == NULL) { 1277 g_io_deliver(bp, ENXIO); 1278 return (0); 1279 } 1280 1281 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) { 1282 g_io_deliver(bp, EROFS); 1283 return (0); 1284 } 1285 1286 off = bp->bio_offset; 1287 volsize = zv->zv_volsize; 1288 1289 os = zv->zv_objset; 1290 ASSERT(os != NULL); 1291 1292 addr = bp->bio_data; 1293 resid = bp->bio_length; 1294 1295 if (resid > 0 && (off < 0 || off >= volsize)) { 1296 g_io_deliver(bp, EIO); 1297 return (0); 1298 } 1299 1300#ifdef illumos 1301 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED; 1302#else 1303 is_dumpified = B_FALSE; 1304#endif 1305 sync = !doread && !is_dumpified && 1306 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; 1307 1308 /* 1309 * There must be no buffer changes when doing a dmu_sync() because 1310 * we can't change the data whilst calculating the checksum. 1311 */ 1312 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1313 doread ? RL_READER : RL_WRITER); 1314 1315 while (resid != 0 && off < volsize) { 1316 size_t size = MIN(resid, zvol_maxphys); 1317#ifdef illumos 1318 if (is_dumpified) { 1319 size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 1320 error = zvol_dumpio(zv, addr, off, size, 1321 doread, B_FALSE); 1322 } else if (doread) { 1323#else 1324 if (doread) { 1325#endif 1326 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1327 DMU_READ_PREFETCH); 1328 } else { 1329 dmu_tx_t *tx = dmu_tx_create(os); 1330 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1331 error = dmu_tx_assign(tx, TXG_WAIT); 1332 if (error) { 1333 dmu_tx_abort(tx); 1334 } else { 1335 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1336 zvol_log_write(zv, tx, off, size, sync); 1337 dmu_tx_commit(tx); 1338 } 1339 } 1340 if (error) { 1341 /* convert checksum errors into IO errors */ 1342 if (error == ECKSUM) 1343 error = SET_ERROR(EIO); 1344 break; 1345 } 1346 off += size; 1347 addr += size; 1348 resid -= size; 1349 } 1350 zfs_range_unlock(rl); 1351 1352 bp->bio_completed = bp->bio_length - resid; 1353 if (bp->bio_completed < bp->bio_length) 1354 bp->bio_error = (off > volsize ? EINVAL : error); 1355 1356 if (sync) 1357 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1358 g_io_deliver(bp, 0); 1359 1360 return (0); 1361} 1362 1363#ifdef sun 1364/* 1365 * Set the buffer count to the zvol maximum transfer. 1366 * Using our own routine instead of the default minphys() 1367 * means that for larger writes we write bigger buffers on X86 1368 * (128K instead of 56K) and flush the disk write cache less often 1369 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1370 * 56K on X86 and 128K on sparc). 1371 */ 1372void 1373zvol_minphys(struct buf *bp) 1374{ 1375 if (bp->b_bcount > zvol_maxphys) 1376 bp->b_bcount = zvol_maxphys; 1377} 1378 1379int 1380zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1381{ 1382 minor_t minor = getminor(dev); 1383 zvol_state_t *zv; 1384 int error = 0; 1385 uint64_t size; 1386 uint64_t boff; 1387 uint64_t resid; 1388 1389 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1390 if (zv == NULL) 1391 return (SET_ERROR(ENXIO)); 1392 1393 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) 1394 return (SET_ERROR(EINVAL)); 1395 1396 boff = ldbtob(blkno); 1397 resid = ldbtob(nblocks); 1398 1399 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1400 1401 while (resid) { 1402 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1403 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1404 if (error) 1405 break; 1406 boff += size; 1407 addr += size; 1408 resid -= size; 1409 } 1410 1411 return (error); 1412} 1413 1414/*ARGSUSED*/ 1415int 1416zvol_read(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 1424 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1425 if (zv == NULL) 1426 return (SET_ERROR(ENXIO)); 1427 1428 volsize = zv->zv_volsize; 1429 if (uio->uio_resid > 0 && 1430 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1431 return (SET_ERROR(EIO)); 1432 1433 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1434 error = physio(zvol_strategy, NULL, dev, B_READ, 1435 zvol_minphys, uio); 1436 return (error); 1437 } 1438 1439 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1440 RL_READER); 1441 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1442 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1443 1444 /* don't read past the end */ 1445 if (bytes > volsize - uio->uio_loffset) 1446 bytes = volsize - uio->uio_loffset; 1447 1448 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1449 if (error) { 1450 /* convert checksum errors into IO errors */ 1451 if (error == ECKSUM) 1452 error = SET_ERROR(EIO); 1453 break; 1454 } 1455 } 1456 zfs_range_unlock(rl); 1457 return (error); 1458} 1459 1460/*ARGSUSED*/ 1461int 1462zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1463{ 1464 minor_t minor = getminor(dev); 1465 zvol_state_t *zv; 1466 uint64_t volsize; 1467 rl_t *rl; 1468 int error = 0; 1469 boolean_t sync; 1470 1471 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1472 if (zv == NULL) 1473 return (SET_ERROR(ENXIO)); 1474 1475 volsize = zv->zv_volsize; 1476 if (uio->uio_resid > 0 && 1477 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1478 return (SET_ERROR(EIO)); 1479 1480 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1481 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1482 zvol_minphys, uio); 1483 return (error); 1484 } 1485 1486 sync = !(zv->zv_flags & ZVOL_WCE) || 1487 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1488 1489 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1490 RL_WRITER); 1491 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1492 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1493 uint64_t off = uio->uio_loffset; 1494 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1495 1496 if (bytes > volsize - off) /* don't write past the end */ 1497 bytes = volsize - off; 1498 1499 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1500 error = dmu_tx_assign(tx, TXG_WAIT); 1501 if (error) { 1502 dmu_tx_abort(tx); 1503 break; 1504 } 1505 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1506 if (error == 0) 1507 zvol_log_write(zv, tx, off, bytes, sync); 1508 dmu_tx_commit(tx); 1509 1510 if (error) 1511 break; 1512 } 1513 zfs_range_unlock(rl); 1514 if (sync) 1515 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1516 return (error); 1517} 1518 1519int 1520zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1521{ 1522 struct uuid uuid = EFI_RESERVED; 1523 efi_gpe_t gpe = { 0 }; 1524 uint32_t crc; 1525 dk_efi_t efi; 1526 int length; 1527 char *ptr; 1528 1529 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1530 return (SET_ERROR(EFAULT)); 1531 ptr = (char *)(uintptr_t)efi.dki_data_64; 1532 length = efi.dki_length; 1533 /* 1534 * Some clients may attempt to request a PMBR for the 1535 * zvol. Currently this interface will return EINVAL to 1536 * such requests. These requests could be supported by 1537 * adding a check for lba == 0 and consing up an appropriate 1538 * PMBR. 1539 */ 1540 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1541 return (SET_ERROR(EINVAL)); 1542 1543 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1544 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1545 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1546 1547 if (efi.dki_lba == 1) { 1548 efi_gpt_t gpt = { 0 }; 1549 1550 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1551 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1552 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1553 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1554 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1555 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1556 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1557 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1558 gpt.efi_gpt_SizeOfPartitionEntry = 1559 LE_32(sizeof (efi_gpe_t)); 1560 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1561 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1562 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1563 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1564 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1565 flag)) 1566 return (SET_ERROR(EFAULT)); 1567 ptr += sizeof (gpt); 1568 length -= sizeof (gpt); 1569 } 1570 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1571 length), flag)) 1572 return (SET_ERROR(EFAULT)); 1573 return (0); 1574} 1575 1576/* 1577 * BEGIN entry points to allow external callers access to the volume. 1578 */ 1579/* 1580 * Return the volume parameters needed for access from an external caller. 1581 * These values are invariant as long as the volume is held open. 1582 */ 1583int 1584zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1585 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1586 void **rl_hdl, void **bonus_hdl) 1587{ 1588 zvol_state_t *zv; 1589 1590 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1591 if (zv == NULL) 1592 return (SET_ERROR(ENXIO)); 1593 if (zv->zv_flags & ZVOL_DUMPIFIED) 1594 return (SET_ERROR(ENXIO)); 1595 1596 ASSERT(blksize && max_xfer_len && minor_hdl && 1597 objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 1598 1599 *blksize = zv->zv_volblocksize; 1600 *max_xfer_len = (uint64_t)zvol_maxphys; 1601 *minor_hdl = zv; 1602 *objset_hdl = zv->zv_objset; 1603 *zil_hdl = zv->zv_zilog; 1604 *rl_hdl = &zv->zv_znode; 1605 *bonus_hdl = zv->zv_dbuf; 1606 return (0); 1607} 1608 1609/* 1610 * Return the current volume size to an external caller. 1611 * The size can change while the volume is open. 1612 */ 1613uint64_t 1614zvol_get_volume_size(void *minor_hdl) 1615{ 1616 zvol_state_t *zv = minor_hdl; 1617 1618 return (zv->zv_volsize); 1619} 1620 1621/* 1622 * Return the current WCE setting to an external caller. 1623 * The WCE setting can change while the volume is open. 1624 */ 1625int 1626zvol_get_volume_wce(void *minor_hdl) 1627{ 1628 zvol_state_t *zv = minor_hdl; 1629 1630 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 1631} 1632 1633/* 1634 * Entry point for external callers to zvol_log_write 1635 */ 1636void 1637zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 1638 boolean_t sync) 1639{ 1640 zvol_state_t *zv = minor_hdl; 1641 1642 zvol_log_write(zv, tx, off, resid, sync); 1643} 1644/* 1645 * END entry points to allow external callers access to the volume. 1646 */ 1647 1648/* 1649 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1650 */ 1651/*ARGSUSED*/ 1652int 1653zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1654{ 1655 zvol_state_t *zv; 1656 struct dk_cinfo dki; 1657 struct dk_minfo dkm; 1658 struct dk_callback *dkc; 1659 int error = 0; 1660 rl_t *rl; 1661 1662 mutex_enter(&spa_namespace_lock); 1663 1664 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 1665 1666 if (zv == NULL) { 1667 mutex_exit(&spa_namespace_lock); 1668 return (SET_ERROR(ENXIO)); 1669 } 1670 ASSERT(zv->zv_total_opens > 0); 1671 1672 switch (cmd) { 1673 1674 case DKIOCINFO: 1675 bzero(&dki, sizeof (dki)); 1676 (void) strcpy(dki.dki_cname, "zvol"); 1677 (void) strcpy(dki.dki_dname, "zvol"); 1678 dki.dki_ctype = DKC_UNKNOWN; 1679 dki.dki_unit = getminor(dev); 1680 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); 1681 mutex_exit(&spa_namespace_lock); 1682 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1683 error = SET_ERROR(EFAULT); 1684 return (error); 1685 1686 case DKIOCGMEDIAINFO: 1687 bzero(&dkm, sizeof (dkm)); 1688 dkm.dki_lbsize = 1U << zv->zv_min_bs; 1689 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1690 dkm.dki_media_type = DK_UNKNOWN; 1691 mutex_exit(&spa_namespace_lock); 1692 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1693 error = SET_ERROR(EFAULT); 1694 return (error); 1695 1696 case DKIOCGETEFI: 1697 { 1698 uint64_t vs = zv->zv_volsize; 1699 uint8_t bs = zv->zv_min_bs; 1700 1701 mutex_exit(&spa_namespace_lock); 1702 error = zvol_getefi((void *)arg, flag, vs, bs); 1703 return (error); 1704 } 1705 1706 case DKIOCFLUSHWRITECACHE: 1707 dkc = (struct dk_callback *)arg; 1708 mutex_exit(&spa_namespace_lock); 1709 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1710 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1711 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1712 error = 0; 1713 } 1714 return (error); 1715 1716 case DKIOCGETWCE: 1717 { 1718 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1719 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1720 flag)) 1721 error = SET_ERROR(EFAULT); 1722 break; 1723 } 1724 case DKIOCSETWCE: 1725 { 1726 int wce; 1727 if (ddi_copyin((void *)arg, &wce, sizeof (int), 1728 flag)) { 1729 error = SET_ERROR(EFAULT); 1730 break; 1731 } 1732 if (wce) { 1733 zv->zv_flags |= ZVOL_WCE; 1734 mutex_exit(&spa_namespace_lock); 1735 } else { 1736 zv->zv_flags &= ~ZVOL_WCE; 1737 mutex_exit(&spa_namespace_lock); 1738 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1739 } 1740 return (0); 1741 } 1742 1743 case DKIOCGGEOM: 1744 case DKIOCGVTOC: 1745 /* 1746 * commands using these (like prtvtoc) expect ENOTSUP 1747 * since we're emulating an EFI label 1748 */ 1749 error = SET_ERROR(ENOTSUP); 1750 break; 1751 1752 case DKIOCDUMPINIT: 1753 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1754 RL_WRITER); 1755 error = zvol_dumpify(zv); 1756 zfs_range_unlock(rl); 1757 break; 1758 1759 case DKIOCDUMPFINI: 1760 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 1761 break; 1762 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1763 RL_WRITER); 1764 error = zvol_dump_fini(zv); 1765 zfs_range_unlock(rl); 1766 break; 1767 1768 case DKIOCFREE: 1769 { 1770 dkioc_free_t df; 1771 dmu_tx_t *tx; 1772 1773 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) { 1774 error = SET_ERROR(EFAULT); 1775 break; 1776 } 1777 1778 /* 1779 * Apply Postel's Law to length-checking. If they overshoot, 1780 * just blank out until the end, if there's a need to blank 1781 * out anything. 1782 */ 1783 if (df.df_start >= zv->zv_volsize) 1784 break; /* No need to do anything... */ 1785 if (df.df_start + df.df_length > zv->zv_volsize) 1786 df.df_length = DMU_OBJECT_END; 1787 1788 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length, 1789 RL_WRITER); 1790 tx = dmu_tx_create(zv->zv_objset); 1791 error = dmu_tx_assign(tx, TXG_WAIT); 1792 if (error != 0) { 1793 dmu_tx_abort(tx); 1794 } else { 1795 zvol_log_truncate(zv, tx, df.df_start, 1796 df.df_length, B_TRUE); 1797 dmu_tx_commit(tx); 1798 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1799 df.df_start, df.df_length); 1800 } 1801 1802 zfs_range_unlock(rl); 1803 1804 if (error == 0) { 1805 /* 1806 * If the write-cache is disabled or 'sync' property 1807 * is set to 'always' then treat this as a synchronous 1808 * operation (i.e. commit to zil). 1809 */ 1810 if (!(zv->zv_flags & ZVOL_WCE) || 1811 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) 1812 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1813 1814 /* 1815 * If the caller really wants synchronous writes, and 1816 * can't wait for them, don't return until the write 1817 * is done. 1818 */ 1819 if (df.df_flags & DF_WAIT_SYNC) { 1820 txg_wait_synced( 1821 dmu_objset_pool(zv->zv_objset), 0); 1822 } 1823 } 1824 break; 1825 } 1826 1827 default: 1828 error = SET_ERROR(ENOTTY); 1829 break; 1830 1831 } 1832 mutex_exit(&spa_namespace_lock); 1833 return (error); 1834} 1835#endif /* sun */ 1836 1837int 1838zvol_busy(void) 1839{ 1840 return (zvol_minors != 0); 1841} 1842 1843void 1844zvol_init(void) 1845{ 1846 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 1847 1) == 0); 1848 ZFS_LOG(1, "ZVOL Initialized."); 1849} 1850 1851void 1852zvol_fini(void) 1853{ 1854 ddi_soft_state_fini(&zfsdev_state); 1855 ZFS_LOG(1, "ZVOL Deinitialized."); 1856} 1857 1858#ifdef sun 1859/*ARGSUSED*/ 1860static int 1861zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx) 1862{ 1863 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1864 1865 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 1866 return (1); 1867 return (0); 1868} 1869 1870/*ARGSUSED*/ 1871static void 1872zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx) 1873{ 1874 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1875 1876 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx); 1877} 1878 1879static int 1880zvol_dump_init(zvol_state_t *zv, boolean_t resize) 1881{ 1882 dmu_tx_t *tx; 1883 int error; 1884 objset_t *os = zv->zv_objset; 1885 spa_t *spa = dmu_objset_spa(os); 1886 vdev_t *vd = spa->spa_root_vdev; 1887 nvlist_t *nv = NULL; 1888 uint64_t version = spa_version(spa); 1889 enum zio_checksum checksum; 1890 1891 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1892 ASSERT(vd->vdev_ops == &vdev_root_ops); 1893 1894 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 1895 DMU_OBJECT_END); 1896 /* wait for dmu_free_long_range to actually free the blocks */ 1897 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1898 1899 /* 1900 * If the pool on which the dump device is being initialized has more 1901 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is 1902 * enabled. If so, bump that feature's counter to indicate that the 1903 * feature is active. We also check the vdev type to handle the 1904 * following case: 1905 * # zpool create test raidz disk1 disk2 disk3 1906 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev), 1907 * the raidz vdev itself has 3 children. 1908 */ 1909 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) { 1910 if (!spa_feature_is_enabled(spa, 1911 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 1912 return (SET_ERROR(ENOTSUP)); 1913 (void) dsl_sync_task(spa_name(spa), 1914 zfs_mvdev_dump_feature_check, 1915 zfs_mvdev_dump_activate_feature_sync, NULL, 2); 1916 } 1917 1918 tx = dmu_tx_create(os); 1919 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1920 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1921 error = dmu_tx_assign(tx, TXG_WAIT); 1922 if (error) { 1923 dmu_tx_abort(tx); 1924 return (error); 1925 } 1926 1927 /* 1928 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum 1929 * function. Otherwise, use the old default -- OFF. 1930 */ 1931 checksum = spa_feature_is_active(spa, 1932 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY : 1933 ZIO_CHECKSUM_OFF; 1934 1935 /* 1936 * If we are resizing the dump device then we only need to 1937 * update the refreservation to match the newly updated 1938 * zvolsize. Otherwise, we save off the original state of the 1939 * zvol so that we can restore them if the zvol is ever undumpified. 1940 */ 1941 if (resize) { 1942 error = zap_update(os, ZVOL_ZAP_OBJ, 1943 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1944 &zv->zv_volsize, tx); 1945 } else { 1946 uint64_t checksum, compress, refresrv, vbs, dedup; 1947 1948 error = dsl_prop_get_integer(zv->zv_name, 1949 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 1950 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1951 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL); 1952 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1953 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL); 1954 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1955 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL); 1956 if (version >= SPA_VERSION_DEDUP) { 1957 error = error ? error : 1958 dsl_prop_get_integer(zv->zv_name, 1959 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 1960 } 1961 1962 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1963 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 1964 &compress, tx); 1965 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1966 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx); 1967 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1968 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1969 &refresrv, tx); 1970 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1971 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 1972 &vbs, tx); 1973 error = error ? error : dmu_object_set_blocksize( 1974 os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx); 1975 if (version >= SPA_VERSION_DEDUP) { 1976 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1977 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 1978 &dedup, tx); 1979 } 1980 if (error == 0) 1981 zv->zv_volblocksize = SPA_MAXBLOCKSIZE; 1982 } 1983 dmu_tx_commit(tx); 1984 1985 /* 1986 * We only need update the zvol's property if we are initializing 1987 * the dump area for the first time. 1988 */ 1989 if (!resize) { 1990 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1991 VERIFY(nvlist_add_uint64(nv, 1992 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 1993 VERIFY(nvlist_add_uint64(nv, 1994 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 1995 ZIO_COMPRESS_OFF) == 0); 1996 VERIFY(nvlist_add_uint64(nv, 1997 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 1998 checksum) == 0); 1999 if (version >= SPA_VERSION_DEDUP) { 2000 VERIFY(nvlist_add_uint64(nv, 2001 zfs_prop_to_name(ZFS_PROP_DEDUP), 2002 ZIO_CHECKSUM_OFF) == 0); 2003 } 2004 2005 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2006 nv, NULL); 2007 nvlist_free(nv); 2008 2009 if (error) 2010 return (error); 2011 } 2012 2013 /* Allocate the space for the dump */ 2014 error = zvol_prealloc(zv); 2015 return (error); 2016} 2017 2018static int 2019zvol_dumpify(zvol_state_t *zv) 2020{ 2021 int error = 0; 2022 uint64_t dumpsize = 0; 2023 dmu_tx_t *tx; 2024 objset_t *os = zv->zv_objset; 2025 2026 if (zv->zv_flags & ZVOL_RDONLY) 2027 return (SET_ERROR(EROFS)); 2028 2029 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 2030 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 2031 boolean_t resize = (dumpsize > 0); 2032 2033 if ((error = zvol_dump_init(zv, resize)) != 0) { 2034 (void) zvol_dump_fini(zv); 2035 return (error); 2036 } 2037 } 2038 2039 /* 2040 * Build up our lba mapping. 2041 */ 2042 error = zvol_get_lbas(zv); 2043 if (error) { 2044 (void) zvol_dump_fini(zv); 2045 return (error); 2046 } 2047 2048 tx = dmu_tx_create(os); 2049 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2050 error = dmu_tx_assign(tx, TXG_WAIT); 2051 if (error) { 2052 dmu_tx_abort(tx); 2053 (void) zvol_dump_fini(zv); 2054 return (error); 2055 } 2056 2057 zv->zv_flags |= ZVOL_DUMPIFIED; 2058 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 2059 &zv->zv_volsize, tx); 2060 dmu_tx_commit(tx); 2061 2062 if (error) { 2063 (void) zvol_dump_fini(zv); 2064 return (error); 2065 } 2066 2067 txg_wait_synced(dmu_objset_pool(os), 0); 2068 return (0); 2069} 2070 2071static int 2072zvol_dump_fini(zvol_state_t *zv) 2073{ 2074 dmu_tx_t *tx; 2075 objset_t *os = zv->zv_objset; 2076 nvlist_t *nv; 2077 int error = 0; 2078 uint64_t checksum, compress, refresrv, vbs, dedup; 2079 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 2080 2081 /* 2082 * Attempt to restore the zvol back to its pre-dumpified state. 2083 * This is a best-effort attempt as it's possible that not all 2084 * of these properties were initialized during the dumpify process 2085 * (i.e. error during zvol_dump_init). 2086 */ 2087 2088 tx = dmu_tx_create(os); 2089 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2090 error = dmu_tx_assign(tx, TXG_WAIT); 2091 if (error) { 2092 dmu_tx_abort(tx); 2093 return (error); 2094 } 2095 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 2096 dmu_tx_commit(tx); 2097 2098 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2099 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 2100 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2101 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 2102 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2103 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 2104 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2105 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 2106 2107 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2108 (void) nvlist_add_uint64(nv, 2109 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 2110 (void) nvlist_add_uint64(nv, 2111 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 2112 (void) nvlist_add_uint64(nv, 2113 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 2114 if (version >= SPA_VERSION_DEDUP && 2115 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2116 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 2117 (void) nvlist_add_uint64(nv, 2118 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 2119 } 2120 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2121 nv, NULL); 2122 nvlist_free(nv); 2123 2124 zvol_free_extents(zv); 2125 zv->zv_flags &= ~ZVOL_DUMPIFIED; 2126 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 2127 /* wait for dmu_free_long_range to actually free the blocks */ 2128 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2129 tx = dmu_tx_create(os); 2130 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2131 error = dmu_tx_assign(tx, TXG_WAIT); 2132 if (error) { 2133 dmu_tx_abort(tx); 2134 return (error); 2135 } 2136 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 2137 zv->zv_volblocksize = vbs; 2138 dmu_tx_commit(tx); 2139 2140 return (0); 2141} 2142#endif /* sun */ 2143 2144static zvol_state_t * 2145zvol_geom_create(const char *name) 2146{ 2147 struct g_provider *pp; 2148 struct g_geom *gp; 2149 zvol_state_t *zv; 2150 2151 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name); 2152 gp->start = zvol_geom_start; 2153 gp->access = zvol_geom_access; 2154 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name); 2155 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 2156 pp->sectorsize = DEV_BSIZE; 2157 2158 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP); 2159 zv->zv_provider = pp; 2160 zv->zv_state = 0; 2161 bioq_init(&zv->zv_queue); 2162 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF); 2163 2164 pp->private = zv; 2165 2166 return (zv); 2167} 2168 2169static void 2170zvol_geom_run(zvol_state_t *zv) 2171{ 2172 struct g_provider *pp; 2173 2174 pp = zv->zv_provider; 2175 g_error_provider(pp, 0); 2176 2177 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0, 2178 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER)); 2179} 2180 2181static void 2182zvol_geom_destroy(zvol_state_t *zv) 2183{ 2184 struct g_provider *pp; 2185 2186 g_topology_assert(); 2187 2188 mtx_lock(&zv->zv_queue_mtx); 2189 zv->zv_state = 1; 2190 wakeup_one(&zv->zv_queue); 2191 while (zv->zv_state != 2) 2192 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0); 2193 mtx_destroy(&zv->zv_queue_mtx); 2194 2195 pp = zv->zv_provider; 2196 zv->zv_provider = NULL; 2197 pp->private = NULL; 2198 g_wither_geom(pp->geom, ENXIO); 2199 2200 kmem_free(zv, sizeof(*zv)); 2201} 2202 2203static int 2204zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) 2205{ 2206 int count, error, flags; 2207 2208 g_topology_assert(); 2209 2210 /* 2211 * To make it easier we expect either open or close, but not both 2212 * at the same time. 2213 */ 2214 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) || 2215 (acr <= 0 && acw <= 0 && ace <= 0), 2216 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", 2217 pp->name, acr, acw, ace)); 2218 2219 if (pp->private == NULL) { 2220 if (acr <= 0 && acw <= 0 && ace <= 0) 2221 return (0); 2222 return (pp->error); 2223 } 2224 2225 /* 2226 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0, 2227 * because GEOM already handles that and handles it a bit differently. 2228 * GEOM allows for multiple read/exclusive consumers and ZFS allows 2229 * only one exclusive consumer, no matter if it is reader or writer. 2230 * I like better the way GEOM works so I'll leave it for GEOM to 2231 * decide what to do. 2232 */ 2233 2234 count = acr + acw + ace; 2235 if (count == 0) 2236 return (0); 2237 2238 flags = 0; 2239 if (acr != 0 || ace != 0) 2240 flags |= FREAD; 2241 if (acw != 0) 2242 flags |= FWRITE; 2243 2244 g_topology_unlock(); 2245 if (count > 0) 2246 error = zvol_open(pp, flags, count); 2247 else 2248 error = zvol_close(pp, flags, -count); 2249 g_topology_lock(); 2250 return (error); 2251} 2252 2253static void 2254zvol_geom_start(struct bio *bp) 2255{ 2256 zvol_state_t *zv; 2257 boolean_t first; 2258 2259 zv = bp->bio_to->private; 2260 ASSERT(zv != NULL); 2261 switch (bp->bio_cmd) { 2262 case BIO_FLUSH: 2263 if (!THREAD_CAN_SLEEP()) 2264 goto enqueue; 2265 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2266 g_io_deliver(bp, 0); 2267 break; 2268 case BIO_READ: 2269 case BIO_WRITE: 2270 if (!THREAD_CAN_SLEEP()) 2271 goto enqueue; 2272 zvol_strategy(bp); 2273 break; 2274 case BIO_GETATTR: 2275 case BIO_DELETE: 2276 default: 2277 g_io_deliver(bp, EOPNOTSUPP); 2278 break; 2279 } 2280 return; 2281 2282enqueue: 2283 mtx_lock(&zv->zv_queue_mtx); 2284 first = (bioq_first(&zv->zv_queue) == NULL); 2285 bioq_insert_tail(&zv->zv_queue, bp); 2286 mtx_unlock(&zv->zv_queue_mtx); 2287 if (first) 2288 wakeup_one(&zv->zv_queue); 2289} 2290 2291static void 2292zvol_geom_worker(void *arg) 2293{ 2294 zvol_state_t *zv; 2295 struct bio *bp; 2296 2297 thread_lock(curthread); 2298 sched_prio(curthread, PRIBIO); 2299 thread_unlock(curthread); 2300 2301 zv = arg; 2302 for (;;) { 2303 mtx_lock(&zv->zv_queue_mtx); 2304 bp = bioq_takefirst(&zv->zv_queue); 2305 if (bp == NULL) { 2306 if (zv->zv_state == 1) { 2307 zv->zv_state = 2; 2308 wakeup(&zv->zv_state); 2309 mtx_unlock(&zv->zv_queue_mtx); 2310 kthread_exit(); 2311 } 2312 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP, 2313 "zvol:io", 0); 2314 continue; 2315 } 2316 mtx_unlock(&zv->zv_queue_mtx); 2317 switch (bp->bio_cmd) { 2318 case BIO_FLUSH: 2319 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2320 g_io_deliver(bp, 0); 2321 break; 2322 case BIO_READ: 2323 case BIO_WRITE: 2324 zvol_strategy(bp); 2325 break; 2326 } 2327 } 2328} 2329 2330extern boolean_t dataset_name_hidden(const char *name); 2331 2332static int 2333zvol_create_snapshots(objset_t *os, const char *name) 2334{ 2335 uint64_t cookie, obj; 2336 char *sname; 2337 int error, len; 2338 2339 cookie = obj = 0; 2340 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2341 2342#if 0 2343 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL, 2344 DS_FIND_SNAPSHOTS); 2345#endif 2346 2347 for (;;) { 2348 len = snprintf(sname, MAXPATHLEN, "%s@", name); 2349 if (len >= MAXPATHLEN) { 2350 dmu_objset_rele(os, FTAG); 2351 error = ENAMETOOLONG; 2352 break; 2353 } 2354 2355 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 2356 error = dmu_snapshot_list_next(os, MAXPATHLEN - len, 2357 sname + len, &obj, &cookie, NULL); 2358 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 2359 if (error != 0) { 2360 if (error == ENOENT) 2361 error = 0; 2362 break; 2363 } 2364 2365 if ((error = zvol_create_minor(sname)) != 0) { 2366 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2367 sname, error); 2368 break; 2369 } 2370 } 2371 2372 kmem_free(sname, MAXPATHLEN); 2373 return (error); 2374} 2375 2376int 2377zvol_create_minors(const char *name) 2378{ 2379 uint64_t cookie; 2380 objset_t *os; 2381 char *osname, *p; 2382 int error, len; 2383 2384 if (dataset_name_hidden(name)) 2385 return (0); 2386 2387 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2388 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2389 name, error); 2390 return (error); 2391 } 2392 if (dmu_objset_type(os) == DMU_OST_ZVOL) { 2393 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG); 2394 dsl_pool_rele(dmu_objset_pool(os), FTAG); 2395 if ((error = zvol_create_minor(name)) == 0) 2396 error = zvol_create_snapshots(os, name); 2397 else { 2398 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2399 name, error); 2400 } 2401 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG); 2402 dsl_dataset_rele(os->os_dsl_dataset, FTAG); 2403 return (error); 2404 } 2405 if (dmu_objset_type(os) != DMU_OST_ZFS) { 2406 dmu_objset_rele(os, FTAG); 2407 return (0); 2408 } 2409 2410 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2411 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) { 2412 dmu_objset_rele(os, FTAG); 2413 kmem_free(osname, MAXPATHLEN); 2414 return (ENOENT); 2415 } 2416 p = osname + strlen(osname); 2417 len = MAXPATHLEN - (p - osname); 2418 2419#if 0 2420 /* Prefetch the datasets. */ 2421 cookie = 0; 2422 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) { 2423 if (!dataset_name_hidden(osname)) 2424 (void) dmu_objset_prefetch(osname, NULL); 2425 } 2426#endif 2427 2428 cookie = 0; 2429 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL, 2430 &cookie) == 0) { 2431 dmu_objset_rele(os, FTAG); 2432 (void)zvol_create_minors(osname); 2433 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2434 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2435 name, error); 2436 return (error); 2437 } 2438 } 2439 2440 dmu_objset_rele(os, FTAG); 2441 kmem_free(osname, MAXPATHLEN); 2442 return (0); 2443} 2444 2445static void 2446zvol_rename_minor(struct g_geom *gp, const char *newname) 2447{ 2448 struct g_provider *pp; 2449 zvol_state_t *zv; 2450 2451 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2452 g_topology_assert(); 2453 2454 pp = LIST_FIRST(&gp->provider); 2455 ASSERT(pp != NULL); 2456 zv = pp->private; 2457 ASSERT(zv != NULL); 2458 2459 zv->zv_provider = NULL; 2460 g_wither_provider(pp, ENXIO); 2461 2462 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname); 2463 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 2464 pp->sectorsize = DEV_BSIZE; 2465 pp->mediasize = zv->zv_volsize; 2466 pp->private = zv; 2467 zv->zv_provider = pp; 2468 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name)); 2469 g_error_provider(pp, 0); 2470} 2471 2472void 2473zvol_rename_minors(const char *oldname, const char *newname) 2474{ 2475 char name[MAXPATHLEN]; 2476 struct g_provider *pp; 2477 struct g_geom *gp; 2478 size_t oldnamelen, newnamelen; 2479 zvol_state_t *zv; 2480 char *namebuf; 2481 2482 oldnamelen = strlen(oldname); 2483 newnamelen = strlen(newname); 2484 2485 DROP_GIANT(); 2486 mutex_enter(&spa_namespace_lock); 2487 g_topology_lock(); 2488 2489 LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) { 2490 pp = LIST_FIRST(&gp->provider); 2491 if (pp == NULL) 2492 continue; 2493 zv = pp->private; 2494 if (zv == NULL) 2495 continue; 2496 if (strcmp(zv->zv_name, oldname) == 0) { 2497 zvol_rename_minor(gp, newname); 2498 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && 2499 (zv->zv_name[oldnamelen] == '/' || 2500 zv->zv_name[oldnamelen] == '@')) { 2501 snprintf(name, sizeof(name), "%s%c%s", newname, 2502 zv->zv_name[oldnamelen], 2503 zv->zv_name + oldnamelen + 1); 2504 zvol_rename_minor(gp, name); 2505 } 2506 } 2507 2508 g_topology_unlock(); 2509 mutex_exit(&spa_namespace_lock); 2510 PICKUP_GIANT(); 2511} 2512