1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * 24 * Copyright (c) 2006-2010 Pawel Jakub Dawidek <pjd@FreeBSD.org> 25 * All rights reserved. 26 */ 27 28/* Portions Copyright 2010 Robert Milkowski */ 29/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 30 31/* 32 * ZFS volume emulation driver. 33 * 34 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 35 * Volumes are accessed through the symbolic links named: 36 * 37 * /dev/zvol/dsk/<pool_name>/<dataset_name> 38 * /dev/zvol/rdsk/<pool_name>/<dataset_name> 39 * 40 * These links are created by the /dev filesystem (sdev_zvolops.c). 41 * Volumes are persistent through reboot. No user command needs to be 42 * run before opening and using a device. 43 * 44 * FreeBSD notes. 45 * On FreeBSD ZVOLs are simply GEOM providers like any other storage device 46 * in the system. 47 */ 48 49#include <sys/types.h> 50#include <sys/param.h> 51#include <sys/kernel.h> 52#include <sys/errno.h> 53#include <sys/uio.h> 54#include <sys/bio.h> 55#include <sys/buf.h> 56#include <sys/kmem.h> 57#include <sys/conf.h> 58#include <sys/cmn_err.h> 59#include <sys/stat.h> 60#include <sys/zap.h> 61#include <sys/spa.h> 62#include <sys/zio.h> 63#include <sys/dmu_traverse.h> 64#include <sys/dnode.h> 65#include <sys/dsl_dataset.h> 66#include <sys/dsl_prop.h> 67#include <sys/dkio.h> 68#include <sys/byteorder.h> 69#include <sys/sunddi.h> 70#include <sys/dirent.h> 71#include <sys/policy.h> 72#include <sys/fs/zfs.h> 73#include <sys/zfs_ioctl.h> 74#include <sys/zil.h> 75#include <sys/refcount.h> 76#include <sys/zfs_znode.h> 77#include <sys/zfs_rlock.h> 78#include <sys/vdev_impl.h> 79#include <sys/zvol.h> 80#include <sys/zil_impl.h> 81#include <geom/geom.h> 82 83#include "zfs_namecheck.h" 84 85struct g_class zfs_zvol_class = { 86 .name = "ZFS::ZVOL", 87 .version = G_VERSION, 88}; 89 90DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol); 91 92void *zfsdev_state; 93static char *zvol_tag = "zvol_tag"; 94 95#define ZVOL_DUMPSIZE "dumpsize" 96 97/* 98 * The spa_namespace_lock protects the zfsdev_state structure from being 99 * modified while it's being used, e.g. an open that comes in before a 100 * create finishes. It also protects temporary opens of the dataset so that, 101 * e.g., an open doesn't get a spurious EBUSY. 102 */ 103static uint32_t zvol_minors; 104 105typedef struct zvol_extent { 106 list_node_t ze_node; 107 dva_t ze_dva; /* dva associated with this extent */ 108 uint64_t ze_nblks; /* number of blocks in extent */ 109} zvol_extent_t; 110 111/* 112 * The in-core state of each volume. 113 */ 114typedef struct zvol_state { 115 char zv_name[MAXPATHLEN]; /* pool/dd name */ 116 uint64_t zv_volsize; /* amount of space we advertise */ 117 uint64_t zv_volblocksize; /* volume block size */ 118 struct g_provider *zv_provider; /* GEOM provider */ 119 uint8_t zv_min_bs; /* minimum addressable block shift */ 120 uint8_t zv_flags; /* readonly, dumpified, etc. */ 121 objset_t *zv_objset; /* objset handle */ 122 uint32_t zv_total_opens; /* total open count */ 123 zilog_t *zv_zilog; /* ZIL handle */ 124 list_t zv_extents; /* List of extents for dump */ 125 znode_t zv_znode; /* for range locking */ 126 dmu_buf_t *zv_dbuf; /* bonus handle */ 127 int zv_state; 128 struct bio_queue_head zv_queue; 129 struct mtx zv_queue_mtx; /* zv_queue mutex */ 130} zvol_state_t; 131 132/* 133 * zvol specific flags 134 */ 135#define ZVOL_RDONLY 0x1 136#define ZVOL_DUMPIFIED 0x2 137#define ZVOL_EXCL 0x4 138#define ZVOL_WCE 0x8 139 140/* 141 * zvol maximum transfer in one DMU tx. 142 */ 143int zvol_maxphys = DMU_MAX_ACCESS/2; 144 145extern int zfs_set_prop_nvlist(const char *, zprop_source_t, 146 nvlist_t *, nvlist_t **); 147static int zvol_remove_zv(zvol_state_t *); 148static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio); 149static int zvol_dumpify(zvol_state_t *zv); 150static int zvol_dump_fini(zvol_state_t *zv); 151static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); 152 153static zvol_state_t *zvol_geom_create(const char *name); 154static void zvol_geom_run(zvol_state_t *zv); 155static void zvol_geom_destroy(zvol_state_t *zv); 156static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace); 157static void zvol_geom_start(struct bio *bp); 158static void zvol_geom_worker(void *arg); 159 160static void 161zvol_size_changed(zvol_state_t *zv) 162{ 163#ifdef sun 164 dev_t dev = makedevice(maj, min); 165 166 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 167 "Size", volsize) == DDI_SUCCESS); 168 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 169 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS); 170 171 /* Notify specfs to invalidate the cached size */ 172 spec_size_invalidate(dev, VBLK); 173 spec_size_invalidate(dev, VCHR); 174#else /* !sun */ 175 struct g_provider *pp; 176 177 pp = zv->zv_provider; 178 if (pp == NULL) 179 return; 180 if (zv->zv_volsize == pp->mediasize) 181 return; 182 /* 183 * Changing provider size is not really supported by GEOM, but it 184 * should be safe when provider is closed. 185 */ 186 if (zv->zv_total_opens > 0) 187 return; 188 pp->mediasize = zv->zv_volsize; 189#endif /* !sun */ 190} 191 192int 193zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 194{ 195 if (volsize == 0) 196 return (EINVAL); 197 198 if (volsize % blocksize != 0) 199 return (EINVAL); 200 201#ifdef _ILP32 202 if (volsize - 1 > SPEC_MAXOFFSET_T) 203 return (EOVERFLOW); 204#endif 205 return (0); 206} 207 208int 209zvol_check_volblocksize(uint64_t volblocksize) 210{ 211 if (volblocksize < SPA_MINBLOCKSIZE || 212 volblocksize > SPA_MAXBLOCKSIZE || 213 !ISP2(volblocksize)) 214 return (EDOM); 215 216 return (0); 217} 218 219int 220zvol_get_stats(objset_t *os, nvlist_t *nv) 221{ 222 int error; 223 dmu_object_info_t doi; 224 uint64_t val; 225 226 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 227 if (error) 228 return (error); 229 230 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 231 232 error = dmu_object_info(os, ZVOL_OBJ, &doi); 233 234 if (error == 0) { 235 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 236 doi.doi_data_block_size); 237 } 238 239 return (error); 240} 241 242static zvol_state_t * 243zvol_minor_lookup(const char *name) 244{ 245 struct g_provider *pp; 246 struct g_geom *gp; 247 zvol_state_t *zv = NULL; 248 249 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 250 251 g_topology_lock(); 252 LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) { 253 pp = LIST_FIRST(&gp->provider); 254 if (pp == NULL) 255 continue; 256 zv = pp->private; 257 if (zv == NULL) 258 continue; 259 if (strcmp(zv->zv_name, name) == 0) 260 break; 261 } 262 g_topology_unlock(); 263 264 return (gp != NULL ? zv : NULL); 265} 266 267/* extent mapping arg */ 268struct maparg { 269 zvol_state_t *ma_zv; 270 uint64_t ma_blks; 271}; 272 273/*ARGSUSED*/ 274static int 275zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf, 276 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 277{ 278 struct maparg *ma = arg; 279 zvol_extent_t *ze; 280 int bs = ma->ma_zv->zv_volblocksize; 281 282 if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 283 return (0); 284 285 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 286 ma->ma_blks++; 287 288 /* Abort immediately if we have encountered gang blocks */ 289 if (BP_IS_GANG(bp)) 290 return (EFRAGS); 291 292 /* 293 * See if the block is at the end of the previous extent. 294 */ 295 ze = list_tail(&ma->ma_zv->zv_extents); 296 if (ze && 297 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 298 DVA_GET_OFFSET(BP_IDENTITY(bp)) == 299 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 300 ze->ze_nblks++; 301 return (0); 302 } 303 304 dprintf_bp(bp, "%s", "next blkptr:"); 305 306 /* start a new extent */ 307 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 308 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 309 ze->ze_nblks = 1; 310 list_insert_tail(&ma->ma_zv->zv_extents, ze); 311 return (0); 312} 313 314static void 315zvol_free_extents(zvol_state_t *zv) 316{ 317 zvol_extent_t *ze; 318 319 while (ze = list_head(&zv->zv_extents)) { 320 list_remove(&zv->zv_extents, ze); 321 kmem_free(ze, sizeof (zvol_extent_t)); 322 } 323} 324 325static int 326zvol_get_lbas(zvol_state_t *zv) 327{ 328 objset_t *os = zv->zv_objset; 329 struct maparg ma; 330 int err; 331 332 ma.ma_zv = zv; 333 ma.ma_blks = 0; 334 zvol_free_extents(zv); 335 336 /* commit any in-flight changes before traversing the dataset */ 337 txg_wait_synced(dmu_objset_pool(os), 0); 338 err = traverse_dataset(dmu_objset_ds(os), 0, 339 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 340 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 341 zvol_free_extents(zv); 342 return (err ? err : EIO); 343 } 344 345 return (0); 346} 347 348/* ARGSUSED */ 349void 350zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 351{ 352 zfs_creat_t *zct = arg; 353 nvlist_t *nvprops = zct->zct_props; 354 int error; 355 uint64_t volblocksize, volsize; 356 357 VERIFY(nvlist_lookup_uint64(nvprops, 358 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 359 if (nvlist_lookup_uint64(nvprops, 360 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 361 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 362 363 /* 364 * These properties must be removed from the list so the generic 365 * property setting step won't apply to them. 366 */ 367 VERIFY(nvlist_remove_all(nvprops, 368 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 369 (void) nvlist_remove_all(nvprops, 370 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 371 372 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 373 DMU_OT_NONE, 0, tx); 374 ASSERT(error == 0); 375 376 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 377 DMU_OT_NONE, 0, tx); 378 ASSERT(error == 0); 379 380 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 381 ASSERT(error == 0); 382} 383 384/* 385 * Replay a TX_WRITE ZIL transaction that didn't get committed 386 * after a system failure 387 */ 388static int 389zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 390{ 391 objset_t *os = zv->zv_objset; 392 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 393 uint64_t offset, length; 394 dmu_tx_t *tx; 395 int error; 396 397 if (byteswap) 398 byteswap_uint64_array(lr, sizeof (*lr)); 399 400 offset = lr->lr_offset; 401 length = lr->lr_length; 402 403 /* If it's a dmu_sync() block, write the whole block */ 404 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 405 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 406 if (length < blocksize) { 407 offset -= offset % blocksize; 408 length = blocksize; 409 } 410 } 411 412 tx = dmu_tx_create(os); 413 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 414 error = dmu_tx_assign(tx, TXG_WAIT); 415 if (error) { 416 dmu_tx_abort(tx); 417 } else { 418 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 419 dmu_tx_commit(tx); 420 } 421 422 return (error); 423} 424 425/* ARGSUSED */ 426static int 427zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 428{ 429 return (ENOTSUP); 430} 431 432/* 433 * Callback vectors for replaying records. 434 * Only TX_WRITE is needed for zvol. 435 */ 436zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 437 zvol_replay_err, /* 0 no such transaction type */ 438 zvol_replay_err, /* TX_CREATE */ 439 zvol_replay_err, /* TX_MKDIR */ 440 zvol_replay_err, /* TX_MKXATTR */ 441 zvol_replay_err, /* TX_SYMLINK */ 442 zvol_replay_err, /* TX_REMOVE */ 443 zvol_replay_err, /* TX_RMDIR */ 444 zvol_replay_err, /* TX_LINK */ 445 zvol_replay_err, /* TX_RENAME */ 446 zvol_replay_write, /* TX_WRITE */ 447 zvol_replay_err, /* TX_TRUNCATE */ 448 zvol_replay_err, /* TX_SETATTR */ 449 zvol_replay_err, /* TX_ACL */ 450 zvol_replay_err, /* TX_CREATE_ACL */ 451 zvol_replay_err, /* TX_CREATE_ATTR */ 452 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 453 zvol_replay_err, /* TX_MKDIR_ACL */ 454 zvol_replay_err, /* TX_MKDIR_ATTR */ 455 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 456 zvol_replay_err, /* TX_WRITE2 */ 457}; 458 459#ifdef sun 460int 461zvol_name2minor(const char *name, minor_t *minor) 462{ 463 zvol_state_t *zv; 464 465 mutex_enter(&spa_namespace_lock); 466 zv = zvol_minor_lookup(name); 467 if (minor && zv) 468 *minor = zv->zv_minor; 469 mutex_exit(&spa_namespace_lock); 470 return (zv ? 0 : -1); 471} 472#endif /* sun */ 473 474/* 475 * Create a minor node (plus a whole lot more) for the specified volume. 476 */ 477int 478zvol_create_minor(const char *name) 479{ 480 zfs_soft_state_t *zs; 481 zvol_state_t *zv; 482 objset_t *os; 483 dmu_object_info_t doi; 484 uint64_t volsize; 485 int error; 486 487 ZFS_LOG(1, "Creating ZVOL %s...", name); 488 489 mutex_enter(&spa_namespace_lock); 490 491 if (zvol_minor_lookup(name) != NULL) { 492 mutex_exit(&spa_namespace_lock); 493 return (EEXIST); 494 } 495 496 /* lie and say we're read-only */ 497 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os); 498 499 if (error) { 500 mutex_exit(&spa_namespace_lock); 501 return (error); 502 } 503 504#ifdef sun 505 if ((minor = zfsdev_minor_alloc()) == 0) { 506 dmu_objset_disown(os, FTAG); 507 mutex_exit(&spa_namespace_lock); 508 return (ENXIO); 509 } 510 511 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 512 dmu_objset_disown(os, FTAG); 513 mutex_exit(&spa_namespace_lock); 514 return (EAGAIN); 515 } 516 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 517 (char *)name); 518 519 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 520 521 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 522 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 523 ddi_soft_state_free(zfsdev_state, minor); 524 dmu_objset_disown(os, FTAG); 525 mutex_exit(&spa_namespace_lock); 526 return (EAGAIN); 527 } 528 529 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 530 531 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 532 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 533 ddi_remove_minor_node(zfs_dip, chrbuf); 534 ddi_soft_state_free(zfsdev_state, minor); 535 dmu_objset_disown(os, FTAG); 536 mutex_exit(&spa_namespace_lock); 537 return (EAGAIN); 538 } 539 540 zs = ddi_get_soft_state(zfsdev_state, minor); 541 zs->zss_type = ZSST_ZVOL; 542 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 543#else /* !sun */ 544 545 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 546 if (error) { 547 ASSERT(error == 0); 548 dmu_objset_disown(os, zvol_tag); 549 mutex_exit(&spa_namespace_lock); 550 return (error); 551 } 552 553 DROP_GIANT(); 554 g_topology_lock(); 555 zv = zvol_geom_create(name); 556 zv->zv_volsize = volsize; 557 zv->zv_provider->mediasize = zv->zv_volsize; 558 559#endif /* !sun */ 560 561 (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 562 zv->zv_min_bs = DEV_BSHIFT; 563 zv->zv_objset = os; 564 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) 565 zv->zv_flags |= ZVOL_RDONLY; 566 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 567 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 568 sizeof (rl_t), offsetof(rl_t, r_node)); 569 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 570 offsetof(zvol_extent_t, ze_node)); 571 /* get and cache the blocksize */ 572 error = dmu_object_info(os, ZVOL_OBJ, &doi); 573 ASSERT(error == 0); 574 zv->zv_volblocksize = doi.doi_data_block_size; 575 576 if (spa_writeable(dmu_objset_spa(os))) { 577 if (zil_replay_disable) 578 zil_destroy(dmu_objset_zil(os), B_FALSE); 579 else 580 zil_replay(os, zv, zvol_replay_vector); 581 } 582 dmu_objset_disown(os, FTAG); 583 zv->zv_objset = NULL; 584 585 zvol_minors++; 586 587 mutex_exit(&spa_namespace_lock); 588 589 zvol_geom_run(zv); 590 591 g_topology_unlock(); 592 PICKUP_GIANT(); 593 594 ZFS_LOG(1, "ZVOL %s created.", name); 595 596 return (0); 597} 598 599/* 600 * Remove minor node for the specified volume. 601 */ 602static int 603zvol_remove_zv(zvol_state_t *zv) 604{ 605#ifdef sun 606 minor_t minor = zv->zv_minor; 607#endif 608 609 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 610 if (zv->zv_total_opens != 0) 611 return (EBUSY); 612 613 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); 614 615#ifdef sun 616 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 617 ddi_remove_minor_node(zfs_dip, nmbuf); 618#endif /* sun */ 619 620 avl_destroy(&zv->zv_znode.z_range_avl); 621 mutex_destroy(&zv->zv_znode.z_range_lock); 622 623 zvol_geom_destroy(zv); 624 625 zvol_minors--; 626 return (0); 627} 628 629int 630zvol_remove_minor(const char *name) 631{ 632 zvol_state_t *zv; 633 int rc; 634 635 mutex_enter(&spa_namespace_lock); 636 if ((zv = zvol_minor_lookup(name)) == NULL) { 637 mutex_exit(&spa_namespace_lock); 638 return (ENXIO); 639 } 640 g_topology_lock(); 641 rc = zvol_remove_zv(zv); 642 g_topology_unlock(); 643 mutex_exit(&spa_namespace_lock); 644 return (rc); 645} 646 647int 648zvol_first_open(zvol_state_t *zv) 649{ 650 objset_t *os; 651 uint64_t volsize; 652 int error; 653 uint64_t readonly; 654 655 /* lie and say we're read-only */ 656 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 657 zvol_tag, &os); 658 if (error) 659 return (error); 660 661 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 662 if (error) { 663 ASSERT(error == 0); 664 dmu_objset_disown(os, zvol_tag); 665 return (error); 666 } 667 zv->zv_objset = os; 668 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 669 if (error) { 670 dmu_objset_disown(os, zvol_tag); 671 return (error); 672 } 673 zv->zv_volsize = volsize; 674 zv->zv_zilog = zil_open(os, zvol_get_data); 675 zvol_size_changed(zv); 676 677 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 678 NULL) == 0); 679 if (readonly || dmu_objset_is_snapshot(os) || 680 !spa_writeable(dmu_objset_spa(os))) 681 zv->zv_flags |= ZVOL_RDONLY; 682 else 683 zv->zv_flags &= ~ZVOL_RDONLY; 684 return (error); 685} 686 687void 688zvol_last_close(zvol_state_t *zv) 689{ 690 zil_close(zv->zv_zilog); 691 zv->zv_zilog = NULL;
|
704 dmu_objset_disown(zv->zv_objset, zvol_tag); 705 zv->zv_objset = NULL; 706} 707 708#ifdef sun 709int 710zvol_prealloc(zvol_state_t *zv) 711{ 712 objset_t *os = zv->zv_objset; 713 dmu_tx_t *tx; 714 uint64_t refd, avail, usedobjs, availobjs; 715 uint64_t resid = zv->zv_volsize; 716 uint64_t off = 0; 717 718 /* Check the space usage before attempting to allocate the space */ 719 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 720 if (avail < zv->zv_volsize) 721 return (ENOSPC); 722 723 /* Free old extents if they exist */ 724 zvol_free_extents(zv); 725 726 while (resid != 0) { 727 int error; 728 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); 729 730 tx = dmu_tx_create(os); 731 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 732 error = dmu_tx_assign(tx, TXG_WAIT); 733 if (error) { 734 dmu_tx_abort(tx); 735 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 736 return (error); 737 } 738 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 739 dmu_tx_commit(tx); 740 off += bytes; 741 resid -= bytes; 742 } 743 txg_wait_synced(dmu_objset_pool(os), 0); 744 745 return (0); 746} 747#endif /* sun */ 748 749int 750zvol_update_volsize(objset_t *os, uint64_t volsize) 751{ 752 dmu_tx_t *tx; 753 int error; 754 755 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 756 757 tx = dmu_tx_create(os); 758 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 759 error = dmu_tx_assign(tx, TXG_WAIT); 760 if (error) { 761 dmu_tx_abort(tx); 762 return (error); 763 } 764 765 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 766 &volsize, tx); 767 dmu_tx_commit(tx); 768 769 if (error == 0) 770 error = dmu_free_long_range(os, 771 ZVOL_OBJ, volsize, DMU_OBJECT_END); 772 return (error); 773} 774 775void 776zvol_remove_minors(const char *name) 777{ 778 struct g_geom *gp, *gptmp; 779 struct g_provider *pp; 780 zvol_state_t *zv; 781 size_t namelen; 782 783 namelen = strlen(name); 784 785 DROP_GIANT(); 786 mutex_enter(&spa_namespace_lock); 787 g_topology_lock(); 788 789 LIST_FOREACH_SAFE(gp, &zfs_zvol_class.geom, geom, gptmp) { 790 pp = LIST_FIRST(&gp->provider); 791 if (pp == NULL) 792 continue; 793 zv = pp->private; 794 if (zv == NULL) 795 continue; 796 if (strcmp(zv->zv_name, name) == 0 || 797 (strncmp(zv->zv_name, name, namelen) == 0 && 798 zv->zv_name[namelen] == '/')) { 799 (void) zvol_remove_zv(zv); 800 } 801 } 802 803 g_topology_unlock(); 804 mutex_exit(&spa_namespace_lock); 805 PICKUP_GIANT(); 806} 807 808int 809zvol_set_volsize(const char *name, major_t maj, uint64_t volsize) 810{ 811 zvol_state_t *zv = NULL; 812 objset_t *os; 813 int error; 814 dmu_object_info_t doi; 815 uint64_t old_volsize = 0ULL; 816 uint64_t readonly; 817 818 mutex_enter(&spa_namespace_lock); 819 zv = zvol_minor_lookup(name); 820 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 821 mutex_exit(&spa_namespace_lock); 822 return (error); 823 } 824 825 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 826 (error = zvol_check_volsize(volsize, 827 doi.doi_data_block_size)) != 0) 828 goto out; 829 830 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, 831 NULL) == 0); 832 if (readonly) { 833 error = EROFS; 834 goto out; 835 } 836 837 error = zvol_update_volsize(os, volsize); 838 /* 839 * Reinitialize the dump area to the new size. If we 840 * failed to resize the dump area then restore it back to 841 * its original size. 842 */ 843 if (zv && error == 0) { 844#ifdef ZVOL_DUMP 845 if (zv->zv_flags & ZVOL_DUMPIFIED) { 846 old_volsize = zv->zv_volsize; 847 zv->zv_volsize = volsize; 848 if ((error = zvol_dumpify(zv)) != 0 || 849 (error = dumpvp_resize()) != 0) { 850 (void) zvol_update_volsize(os, old_volsize); 851 zv->zv_volsize = old_volsize; 852 error = zvol_dumpify(zv); 853 } 854 } 855#endif /* ZVOL_DUMP */ 856 if (error == 0) { 857 zv->zv_volsize = volsize; 858 zvol_size_changed(zv); 859 } 860 } 861 862#ifdef sun 863 /* 864 * Generate a LUN expansion event. 865 */ 866 if (zv && error == 0) { 867 sysevent_id_t eid; 868 nvlist_t *attr; 869 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 870 871 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 872 zv->zv_minor); 873 874 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 875 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 876 877 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 878 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 879 880 nvlist_free(attr); 881 kmem_free(physpath, MAXPATHLEN); 882 } 883#endif /* sun */ 884 885out: 886 dmu_objset_rele(os, FTAG); 887 888 mutex_exit(&spa_namespace_lock); 889 890 return (error); 891} 892 893/*ARGSUSED*/ 894static int 895zvol_open(struct g_provider *pp, int flag, int count) 896{ 897 zvol_state_t *zv; 898 int err = 0; 899 boolean_t locked = B_FALSE; 900 901 /* 902 * Protect against recursively entering spa_namespace_lock 903 * when spa_open() is used for a pool on a (local) ZVOL(s). 904 * This is needed since we replaced upstream zfsdev_state_lock 905 * with spa_namespace_lock in the ZVOL code. 906 * We are using the same trick as spa_open(). 907 * Note that calls in zvol_first_open which need to resolve 908 * pool name to a spa object will enter spa_open() 909 * recursively, but that function already has all the 910 * necessary protection. 911 */ 912 if (!MUTEX_HELD(&spa_namespace_lock)) { 913 mutex_enter(&spa_namespace_lock); 914 locked = B_TRUE; 915 } 916 917 zv = pp->private; 918 if (zv == NULL) { 919 if (locked) 920 mutex_exit(&spa_namespace_lock); 921 return (ENXIO); 922 } 923 924 if (zv->zv_total_opens == 0) 925 err = zvol_first_open(zv); 926 if (err) { 927 if (locked) 928 mutex_exit(&spa_namespace_lock); 929 return (err); 930 } 931 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 932 err = EROFS; 933 goto out; 934 } 935 if (zv->zv_flags & ZVOL_EXCL) { 936 err = EBUSY; 937 goto out; 938 } 939#ifdef FEXCL 940 if (flag & FEXCL) { 941 if (zv->zv_total_opens != 0) { 942 err = EBUSY; 943 goto out; 944 } 945 zv->zv_flags |= ZVOL_EXCL; 946 } 947#endif 948 949 zv->zv_total_opens += count; 950 if (locked) 951 mutex_exit(&spa_namespace_lock); 952 953 return (err); 954out: 955 if (zv->zv_total_opens == 0) 956 zvol_last_close(zv); 957 if (locked) 958 mutex_exit(&spa_namespace_lock); 959 return (err); 960} 961 962/*ARGSUSED*/ 963static int 964zvol_close(struct g_provider *pp, int flag, int count) 965{ 966 zvol_state_t *zv; 967 int error = 0; 968 boolean_t locked = B_FALSE; 969 970 /* See comment in zvol_open(). */ 971 if (!MUTEX_HELD(&spa_namespace_lock)) { 972 mutex_enter(&spa_namespace_lock); 973 locked = B_TRUE; 974 } 975 976 zv = pp->private; 977 if (zv == NULL) { 978 if (locked) 979 mutex_exit(&spa_namespace_lock); 980 return (ENXIO); 981 } 982 983 if (zv->zv_flags & ZVOL_EXCL) { 984 ASSERT(zv->zv_total_opens == 1); 985 zv->zv_flags &= ~ZVOL_EXCL; 986 } 987 988 /* 989 * If the open count is zero, this is a spurious close. 990 * That indicates a bug in the kernel / DDI framework. 991 */ 992 ASSERT(zv->zv_total_opens != 0); 993 994 /* 995 * You may get multiple opens, but only one close. 996 */ 997 zv->zv_total_opens -= count; 998 999 if (zv->zv_total_opens == 0) 1000 zvol_last_close(zv); 1001 1002 if (locked) 1003 mutex_exit(&spa_namespace_lock); 1004 return (error); 1005} 1006 1007static void 1008zvol_get_done(zgd_t *zgd, int error) 1009{ 1010 if (zgd->zgd_db) 1011 dmu_buf_rele(zgd->zgd_db, zgd); 1012 1013 zfs_range_unlock(zgd->zgd_rl); 1014 1015 if (error == 0 && zgd->zgd_bp) 1016 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1017 1018 kmem_free(zgd, sizeof (zgd_t)); 1019} 1020 1021/* 1022 * Get data to generate a TX_WRITE intent log record. 1023 */ 1024static int 1025zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1026{ 1027 zvol_state_t *zv = arg; 1028 objset_t *os = zv->zv_objset; 1029 uint64_t object = ZVOL_OBJ; 1030 uint64_t offset = lr->lr_offset; 1031 uint64_t size = lr->lr_length; /* length of user data */ 1032 blkptr_t *bp = &lr->lr_blkptr; 1033 dmu_buf_t *db; 1034 zgd_t *zgd; 1035 int error; 1036 1037 ASSERT(zio != NULL); 1038 ASSERT(size != 0); 1039 1040 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1041 zgd->zgd_zilog = zv->zv_zilog; 1042 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 1043 1044 /* 1045 * Write records come in two flavors: immediate and indirect. 1046 * For small writes it's cheaper to store the data with the 1047 * log record (immediate); for large writes it's cheaper to 1048 * sync the data and get a pointer to it (indirect) so that 1049 * we don't have to write the data twice. 1050 */ 1051 if (buf != NULL) { /* immediate write */ 1052 error = dmu_read(os, object, offset, size, buf, 1053 DMU_READ_NO_PREFETCH); 1054 } else { 1055 size = zv->zv_volblocksize; 1056 offset = P2ALIGN(offset, size); 1057 error = dmu_buf_hold(os, object, offset, zgd, &db, 1058 DMU_READ_NO_PREFETCH); 1059 if (error == 0) { 1060 zgd->zgd_db = db; 1061 zgd->zgd_bp = bp; 1062 1063 ASSERT(db->db_offset == offset); 1064 ASSERT(db->db_size == size); 1065 1066 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1067 zvol_get_done, zgd); 1068 1069 if (error == 0) 1070 return (0); 1071 } 1072 } 1073 1074 zvol_get_done(zgd, error); 1075 1076 return (error); 1077} 1078 1079/* 1080 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 1081 * 1082 * We store data in the log buffers if it's small enough. 1083 * Otherwise we will later flush the data out via dmu_sync(). 1084 */ 1085ssize_t zvol_immediate_write_sz = 32768; 1086 1087static void 1088zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1089 boolean_t sync) 1090{ 1091 uint32_t blocksize = zv->zv_volblocksize; 1092 zilog_t *zilog = zv->zv_zilog; 1093 boolean_t slogging; 1094 ssize_t immediate_write_sz; 1095 1096 if (zil_replaying(zilog, tx)) 1097 return; 1098 1099 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1100 ? 0 : zvol_immediate_write_sz; 1101 1102 slogging = spa_has_slogs(zilog->zl_spa) && 1103 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 1104 1105 while (resid) { 1106 itx_t *itx; 1107 lr_write_t *lr; 1108 ssize_t len; 1109 itx_wr_state_t write_state; 1110 1111 /* 1112 * Unlike zfs_log_write() we can be called with 1113 * upto DMU_MAX_ACCESS/2 (5MB) writes. 1114 */ 1115 if (blocksize > immediate_write_sz && !slogging && 1116 resid >= blocksize && off % blocksize == 0) { 1117 write_state = WR_INDIRECT; /* uses dmu_sync */ 1118 len = blocksize; 1119 } else if (sync) { 1120 write_state = WR_COPIED; 1121 len = MIN(ZIL_MAX_LOG_DATA, resid); 1122 } else { 1123 write_state = WR_NEED_COPY; 1124 len = MIN(ZIL_MAX_LOG_DATA, resid); 1125 } 1126 1127 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1128 (write_state == WR_COPIED ? len : 0)); 1129 lr = (lr_write_t *)&itx->itx_lr; 1130 if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 1131 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1132 zil_itx_destroy(itx); 1133 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1134 lr = (lr_write_t *)&itx->itx_lr; 1135 write_state = WR_NEED_COPY; 1136 } 1137 1138 itx->itx_wr_state = write_state; 1139 if (write_state == WR_NEED_COPY) 1140 itx->itx_sod += len; 1141 lr->lr_foid = ZVOL_OBJ; 1142 lr->lr_offset = off; 1143 lr->lr_length = len; 1144 lr->lr_blkoff = 0; 1145 BP_ZERO(&lr->lr_blkptr); 1146 1147 itx->itx_private = zv; 1148 itx->itx_sync = sync; 1149 1150 zil_itx_assign(zilog, itx, tx); 1151 1152 off += len; 1153 resid -= len; 1154 } 1155} 1156 1157#ifdef sun 1158static int 1159zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t size, 1160 boolean_t doread, boolean_t isdump) 1161{ 1162 vdev_disk_t *dvd; 1163 int c; 1164 int numerrors = 0; 1165 1166 for (c = 0; c < vd->vdev_children; c++) { 1167 ASSERT(vd->vdev_ops == &vdev_mirror_ops || 1168 vd->vdev_ops == &vdev_replacing_ops || 1169 vd->vdev_ops == &vdev_spare_ops); 1170 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1171 addr, offset, size, doread, isdump); 1172 if (err != 0) { 1173 numerrors++; 1174 } else if (doread) { 1175 break; 1176 } 1177 } 1178 1179 if (!vd->vdev_ops->vdev_op_leaf) 1180 return (numerrors < vd->vdev_children ? 0 : EIO); 1181 1182 if (doread && !vdev_readable(vd)) 1183 return (EIO); 1184 else if (!doread && !vdev_writeable(vd)) 1185 return (EIO); 1186 1187 dvd = vd->vdev_tsd; 1188 ASSERT3P(dvd, !=, NULL); 1189 offset += VDEV_LABEL_START_SIZE; 1190 1191 if (ddi_in_panic() || isdump) { 1192 ASSERT(!doread); 1193 if (doread) 1194 return (EIO); 1195 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1196 lbtodb(size))); 1197 } else { 1198 return (vdev_disk_physio(dvd->vd_lh, addr, size, offset, 1199 doread ? B_READ : B_WRITE)); 1200 } 1201} 1202 1203static int 1204zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1205 boolean_t doread, boolean_t isdump) 1206{ 1207 vdev_t *vd; 1208 int error; 1209 zvol_extent_t *ze; 1210 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1211 1212 /* Must be sector aligned, and not stradle a block boundary. */ 1213 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1214 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1215 return (EINVAL); 1216 } 1217 ASSERT(size <= zv->zv_volblocksize); 1218 1219 /* Locate the extent this belongs to */ 1220 ze = list_head(&zv->zv_extents); 1221 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1222 offset -= ze->ze_nblks * zv->zv_volblocksize; 1223 ze = list_next(&zv->zv_extents, ze); 1224 } 1225 1226 if (!ddi_in_panic()) 1227 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1228 1229 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1230 offset += DVA_GET_OFFSET(&ze->ze_dva); 1231 error = zvol_dumpio_vdev(vd, addr, offset, size, doread, isdump); 1232 1233 if (!ddi_in_panic()) 1234 spa_config_exit(spa, SCL_STATE, FTAG); 1235 1236 return (error); 1237} 1238#endif /* sun */ 1239 1240int 1241zvol_strategy(struct bio *bp) 1242{ 1243 zvol_state_t *zv = bp->bio_to->private; 1244 uint64_t off, volsize; 1245 size_t resid; 1246 char *addr; 1247 objset_t *os; 1248 rl_t *rl; 1249 int error = 0; 1250 boolean_t doread = (bp->bio_cmd == BIO_READ); 1251 boolean_t sync; 1252 1253 if (zv == NULL) { 1254 g_io_deliver(bp, ENXIO); 1255 return (0); 1256 } 1257 1258 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) { 1259 g_io_deliver(bp, EROFS); 1260 return (0); 1261 } 1262 1263 off = bp->bio_offset; 1264 volsize = zv->zv_volsize; 1265 1266 os = zv->zv_objset; 1267 ASSERT(os != NULL); 1268 1269 addr = bp->bio_data; 1270 resid = bp->bio_length; 1271 1272 if (resid > 0 && (off < 0 || off >= volsize)) { 1273 g_io_deliver(bp, EIO); 1274 return (0); 1275 } 1276 1277 sync = !doread && zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; 1278 1279 /* 1280 * There must be no buffer changes when doing a dmu_sync() because 1281 * we can't change the data whilst calculating the checksum. 1282 */ 1283 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1284 doread ? RL_READER : RL_WRITER); 1285 1286 while (resid != 0 && off < volsize) { 1287 size_t size = MIN(resid, zvol_maxphys); 1288 if (doread) { 1289 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1290 DMU_READ_PREFETCH); 1291 } else { 1292 dmu_tx_t *tx = dmu_tx_create(os); 1293 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1294 error = dmu_tx_assign(tx, TXG_WAIT); 1295 if (error) { 1296 dmu_tx_abort(tx); 1297 } else { 1298 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1299 zvol_log_write(zv, tx, off, size, sync); 1300 dmu_tx_commit(tx); 1301 } 1302 } 1303 if (error) { 1304 /* convert checksum errors into IO errors */ 1305 if (error == ECKSUM) 1306 error = EIO; 1307 break; 1308 } 1309 off += size; 1310 addr += size; 1311 resid -= size; 1312 } 1313 zfs_range_unlock(rl); 1314 1315 bp->bio_completed = bp->bio_length - resid; 1316 if (bp->bio_completed < bp->bio_length) 1317 bp->bio_error = (off > volsize ? EINVAL : error); 1318 1319 if (sync) 1320 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1321 g_io_deliver(bp, 0); 1322 1323 return (0); 1324} 1325 1326#ifdef sun 1327/* 1328 * Set the buffer count to the zvol maximum transfer. 1329 * Using our own routine instead of the default minphys() 1330 * means that for larger writes we write bigger buffers on X86 1331 * (128K instead of 56K) and flush the disk write cache less often 1332 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1333 * 56K on X86 and 128K on sparc). 1334 */ 1335void 1336zvol_minphys(struct buf *bp) 1337{ 1338 if (bp->b_bcount > zvol_maxphys) 1339 bp->b_bcount = zvol_maxphys; 1340} 1341 1342int 1343zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1344{ 1345 minor_t minor = getminor(dev); 1346 zvol_state_t *zv; 1347 int error = 0; 1348 uint64_t size; 1349 uint64_t boff; 1350 uint64_t resid; 1351 1352 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1353 if (zv == NULL) 1354 return (ENXIO); 1355 1356 boff = ldbtob(blkno); 1357 resid = ldbtob(nblocks); 1358 1359 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1360 1361 while (resid) { 1362 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1363 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1364 if (error) 1365 break; 1366 boff += size; 1367 addr += size; 1368 resid -= size; 1369 } 1370 1371 return (error); 1372} 1373 1374/*ARGSUSED*/ 1375int 1376zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1377{ 1378 minor_t minor = getminor(dev); 1379 zvol_state_t *zv; 1380 uint64_t volsize; 1381 rl_t *rl; 1382 int error = 0; 1383 1384 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1385 if (zv == NULL) 1386 return (ENXIO); 1387 1388 volsize = zv->zv_volsize; 1389 if (uio->uio_resid > 0 && 1390 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1391 return (EIO); 1392 1393 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1394 error = physio(zvol_strategy, NULL, dev, B_READ, 1395 zvol_minphys, uio); 1396 return (error); 1397 } 1398 1399 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1400 RL_READER); 1401 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1402 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1403 1404 /* don't read past the end */ 1405 if (bytes > volsize - uio->uio_loffset) 1406 bytes = volsize - uio->uio_loffset; 1407 1408 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1409 if (error) { 1410 /* convert checksum errors into IO errors */ 1411 if (error == ECKSUM) 1412 error = EIO; 1413 break; 1414 } 1415 } 1416 zfs_range_unlock(rl); 1417 return (error); 1418} 1419 1420/*ARGSUSED*/ 1421int 1422zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1423{ 1424 minor_t minor = getminor(dev); 1425 zvol_state_t *zv; 1426 uint64_t volsize; 1427 rl_t *rl; 1428 int error = 0; 1429 boolean_t sync; 1430 1431 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1432 if (zv == NULL) 1433 return (ENXIO); 1434 1435 volsize = zv->zv_volsize; 1436 if (uio->uio_resid > 0 && 1437 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1438 return (EIO); 1439 1440 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1441 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1442 zvol_minphys, uio); 1443 return (error); 1444 } 1445 1446 sync = !(zv->zv_flags & ZVOL_WCE) || 1447 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1448 1449 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1450 RL_WRITER); 1451 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1452 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1453 uint64_t off = uio->uio_loffset; 1454 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1455 1456 if (bytes > volsize - off) /* don't write past the end */ 1457 bytes = volsize - off; 1458 1459 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1460 error = dmu_tx_assign(tx, TXG_WAIT); 1461 if (error) { 1462 dmu_tx_abort(tx); 1463 break; 1464 } 1465 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1466 if (error == 0) 1467 zvol_log_write(zv, tx, off, bytes, sync); 1468 dmu_tx_commit(tx); 1469 1470 if (error) 1471 break; 1472 } 1473 zfs_range_unlock(rl); 1474 if (sync) 1475 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1476 return (error); 1477} 1478 1479int 1480zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1481{ 1482 struct uuid uuid = EFI_RESERVED; 1483 efi_gpe_t gpe = { 0 }; 1484 uint32_t crc; 1485 dk_efi_t efi; 1486 int length; 1487 char *ptr; 1488 1489 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1490 return (EFAULT); 1491 ptr = (char *)(uintptr_t)efi.dki_data_64; 1492 length = efi.dki_length; 1493 /* 1494 * Some clients may attempt to request a PMBR for the 1495 * zvol. Currently this interface will return EINVAL to 1496 * such requests. These requests could be supported by 1497 * adding a check for lba == 0 and consing up an appropriate 1498 * PMBR. 1499 */ 1500 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1501 return (EINVAL); 1502 1503 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1504 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1505 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1506 1507 if (efi.dki_lba == 1) { 1508 efi_gpt_t gpt = { 0 }; 1509 1510 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1511 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1512 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1513 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1514 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1515 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1516 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1517 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1518 gpt.efi_gpt_SizeOfPartitionEntry = 1519 LE_32(sizeof (efi_gpe_t)); 1520 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1521 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1522 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1523 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1524 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1525 flag)) 1526 return (EFAULT); 1527 ptr += sizeof (gpt); 1528 length -= sizeof (gpt); 1529 } 1530 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1531 length), flag)) 1532 return (EFAULT); 1533 return (0); 1534} 1535 1536/* 1537 * BEGIN entry points to allow external callers access to the volume. 1538 */ 1539/* 1540 * Return the volume parameters needed for access from an external caller. 1541 * These values are invariant as long as the volume is held open. 1542 */ 1543int 1544zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1545 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1546 void **rl_hdl, void **bonus_hdl) 1547{ 1548 zvol_state_t *zv; 1549 1550 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1551 if (zv == NULL) 1552 return (ENXIO); 1553 if (zv->zv_flags & ZVOL_DUMPIFIED) 1554 return (ENXIO); 1555 1556 ASSERT(blksize && max_xfer_len && minor_hdl && 1557 objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 1558 1559 *blksize = zv->zv_volblocksize; 1560 *max_xfer_len = (uint64_t)zvol_maxphys; 1561 *minor_hdl = zv; 1562 *objset_hdl = zv->zv_objset; 1563 *zil_hdl = zv->zv_zilog; 1564 *rl_hdl = &zv->zv_znode; 1565 *bonus_hdl = zv->zv_dbuf; 1566 return (0); 1567} 1568 1569/* 1570 * Return the current volume size to an external caller. 1571 * The size can change while the volume is open. 1572 */ 1573uint64_t 1574zvol_get_volume_size(void *minor_hdl) 1575{ 1576 zvol_state_t *zv = minor_hdl; 1577 1578 return (zv->zv_volsize); 1579} 1580 1581/* 1582 * Return the current WCE setting to an external caller. 1583 * The WCE setting can change while the volume is open. 1584 */ 1585int 1586zvol_get_volume_wce(void *minor_hdl) 1587{ 1588 zvol_state_t *zv = minor_hdl; 1589 1590 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 1591} 1592 1593/* 1594 * Entry point for external callers to zvol_log_write 1595 */ 1596void 1597zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 1598 boolean_t sync) 1599{ 1600 zvol_state_t *zv = minor_hdl; 1601 1602 zvol_log_write(zv, tx, off, resid, sync); 1603} 1604/* 1605 * END entry points to allow external callers access to the volume. 1606 */ 1607 1608/* 1609 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1610 */ 1611/*ARGSUSED*/ 1612int 1613zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1614{ 1615 zvol_state_t *zv; 1616 struct dk_cinfo dki; 1617 struct dk_minfo dkm; 1618 struct dk_callback *dkc; 1619 int error = 0; 1620 rl_t *rl; 1621 1622 mutex_enter(&spa_namespace_lock); 1623 1624 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 1625 1626 if (zv == NULL) { 1627 mutex_exit(&spa_namespace_lock); 1628 return (ENXIO); 1629 } 1630 ASSERT(zv->zv_total_opens > 0); 1631 1632 switch (cmd) { 1633 1634 case DKIOCINFO: 1635 bzero(&dki, sizeof (dki)); 1636 (void) strcpy(dki.dki_cname, "zvol"); 1637 (void) strcpy(dki.dki_dname, "zvol"); 1638 dki.dki_ctype = DKC_UNKNOWN; 1639 dki.dki_unit = getminor(dev); 1640 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); 1641 mutex_exit(&spa_namespace_lock); 1642 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1643 error = EFAULT; 1644 return (error); 1645 1646 case DKIOCGMEDIAINFO: 1647 bzero(&dkm, sizeof (dkm)); 1648 dkm.dki_lbsize = 1U << zv->zv_min_bs; 1649 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1650 dkm.dki_media_type = DK_UNKNOWN; 1651 mutex_exit(&spa_namespace_lock); 1652 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1653 error = EFAULT; 1654 return (error); 1655 1656 case DKIOCGETEFI: 1657 { 1658 uint64_t vs = zv->zv_volsize; 1659 uint8_t bs = zv->zv_min_bs; 1660 1661 mutex_exit(&spa_namespace_lock); 1662 error = zvol_getefi((void *)arg, flag, vs, bs); 1663 return (error); 1664 } 1665 1666 case DKIOCFLUSHWRITECACHE: 1667 dkc = (struct dk_callback *)arg; 1668 mutex_exit(&spa_namespace_lock); 1669 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1670 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1671 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1672 error = 0; 1673 } 1674 return (error); 1675 1676 case DKIOCGETWCE: 1677 { 1678 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1679 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1680 flag)) 1681 error = EFAULT; 1682 break; 1683 } 1684 case DKIOCSETWCE: 1685 { 1686 int wce; 1687 if (ddi_copyin((void *)arg, &wce, sizeof (int), 1688 flag)) { 1689 error = EFAULT; 1690 break; 1691 } 1692 if (wce) { 1693 zv->zv_flags |= ZVOL_WCE; 1694 mutex_exit(&spa_namespace_lock); 1695 } else { 1696 zv->zv_flags &= ~ZVOL_WCE; 1697 mutex_exit(&spa_namespace_lock); 1698 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1699 } 1700 return (0); 1701 } 1702 1703 case DKIOCGGEOM: 1704 case DKIOCGVTOC: 1705 /* 1706 * commands using these (like prtvtoc) expect ENOTSUP 1707 * since we're emulating an EFI label 1708 */ 1709 error = ENOTSUP; 1710 break; 1711 1712 case DKIOCDUMPINIT: 1713 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1714 RL_WRITER); 1715 error = zvol_dumpify(zv); 1716 zfs_range_unlock(rl); 1717 break; 1718 1719 case DKIOCDUMPFINI: 1720 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 1721 break; 1722 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1723 RL_WRITER); 1724 error = zvol_dump_fini(zv); 1725 zfs_range_unlock(rl); 1726 break; 1727 1728 default: 1729 error = ENOTTY; 1730 break; 1731 1732 } 1733 mutex_exit(&spa_namespace_lock); 1734 return (error); 1735} 1736#endif /* sun */ 1737 1738int 1739zvol_busy(void) 1740{ 1741 return (zvol_minors != 0); 1742} 1743 1744void 1745zvol_init(void) 1746{ 1747 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 1748 1) == 0); 1749 ZFS_LOG(1, "ZVOL Initialized."); 1750} 1751 1752void 1753zvol_fini(void) 1754{ 1755 ddi_soft_state_fini(&zfsdev_state); 1756 ZFS_LOG(1, "ZVOL Deinitialized."); 1757} 1758 1759#ifdef sun 1760static int 1761zvol_dump_init(zvol_state_t *zv, boolean_t resize) 1762{ 1763 dmu_tx_t *tx; 1764 int error = 0; 1765 objset_t *os = zv->zv_objset; 1766 nvlist_t *nv = NULL; 1767 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 1768 1769 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1770 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 1771 DMU_OBJECT_END); 1772 /* wait for dmu_free_long_range to actually free the blocks */ 1773 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1774 1775 tx = dmu_tx_create(os); 1776 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1777 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1778 error = dmu_tx_assign(tx, TXG_WAIT); 1779 if (error) { 1780 dmu_tx_abort(tx); 1781 return (error); 1782 } 1783 1784 /* 1785 * If we are resizing the dump device then we only need to 1786 * update the refreservation to match the newly updated 1787 * zvolsize. Otherwise, we save off the original state of the 1788 * zvol so that we can restore them if the zvol is ever undumpified. 1789 */ 1790 if (resize) { 1791 error = zap_update(os, ZVOL_ZAP_OBJ, 1792 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1793 &zv->zv_volsize, tx); 1794 } else { 1795 uint64_t checksum, compress, refresrv, vbs, dedup; 1796 1797 error = dsl_prop_get_integer(zv->zv_name, 1798 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 1799 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1800 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL); 1801 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1802 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL); 1803 error = error ? error : dsl_prop_get_integer(zv->zv_name, 1804 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL); 1805 if (version >= SPA_VERSION_DEDUP) { 1806 error = error ? error : 1807 dsl_prop_get_integer(zv->zv_name, 1808 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 1809 } 1810 1811 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1812 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 1813 &compress, tx); 1814 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1815 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx); 1816 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1817 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 1818 &refresrv, tx); 1819 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1820 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 1821 &vbs, tx); 1822 error = error ? error : dmu_object_set_blocksize( 1823 os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx); 1824 if (version >= SPA_VERSION_DEDUP) { 1825 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 1826 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 1827 &dedup, tx); 1828 } 1829 if (error == 0) 1830 zv->zv_volblocksize = SPA_MAXBLOCKSIZE; 1831 } 1832 dmu_tx_commit(tx); 1833 1834 /* 1835 * We only need update the zvol's property if we are initializing 1836 * the dump area for the first time. 1837 */ 1838 if (!resize) { 1839 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1840 VERIFY(nvlist_add_uint64(nv, 1841 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 1842 VERIFY(nvlist_add_uint64(nv, 1843 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 1844 ZIO_COMPRESS_OFF) == 0); 1845 VERIFY(nvlist_add_uint64(nv, 1846 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 1847 ZIO_CHECKSUM_OFF) == 0); 1848 if (version >= SPA_VERSION_DEDUP) { 1849 VERIFY(nvlist_add_uint64(nv, 1850 zfs_prop_to_name(ZFS_PROP_DEDUP), 1851 ZIO_CHECKSUM_OFF) == 0); 1852 } 1853 1854 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 1855 nv, NULL); 1856 nvlist_free(nv); 1857 1858 if (error) 1859 return (error); 1860 } 1861 1862 /* Allocate the space for the dump */ 1863 error = zvol_prealloc(zv); 1864 return (error); 1865} 1866 1867static int 1868zvol_dumpify(zvol_state_t *zv) 1869{ 1870 int error = 0; 1871 uint64_t dumpsize = 0; 1872 dmu_tx_t *tx; 1873 objset_t *os = zv->zv_objset; 1874 1875 if (zv->zv_flags & ZVOL_RDONLY) 1876 return (EROFS); 1877 1878 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 1879 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 1880 boolean_t resize = (dumpsize > 0) ? B_TRUE : B_FALSE; 1881 1882 if ((error = zvol_dump_init(zv, resize)) != 0) { 1883 (void) zvol_dump_fini(zv); 1884 return (error); 1885 } 1886 } 1887 1888 /* 1889 * Build up our lba mapping. 1890 */ 1891 error = zvol_get_lbas(zv); 1892 if (error) { 1893 (void) zvol_dump_fini(zv); 1894 return (error); 1895 } 1896 1897 tx = dmu_tx_create(os); 1898 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1899 error = dmu_tx_assign(tx, TXG_WAIT); 1900 if (error) { 1901 dmu_tx_abort(tx); 1902 (void) zvol_dump_fini(zv); 1903 return (error); 1904 } 1905 1906 zv->zv_flags |= ZVOL_DUMPIFIED; 1907 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 1908 &zv->zv_volsize, tx); 1909 dmu_tx_commit(tx); 1910 1911 if (error) { 1912 (void) zvol_dump_fini(zv); 1913 return (error); 1914 } 1915 1916 txg_wait_synced(dmu_objset_pool(os), 0); 1917 return (0); 1918} 1919 1920static int 1921zvol_dump_fini(zvol_state_t *zv) 1922{ 1923 dmu_tx_t *tx; 1924 objset_t *os = zv->zv_objset; 1925 nvlist_t *nv; 1926 int error = 0; 1927 uint64_t checksum, compress, refresrv, vbs, dedup; 1928 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 1929 1930 /* 1931 * Attempt to restore the zvol back to its pre-dumpified state. 1932 * This is a best-effort attempt as it's possible that not all 1933 * of these properties were initialized during the dumpify process 1934 * (i.e. error during zvol_dump_init). 1935 */ 1936 1937 tx = dmu_tx_create(os); 1938 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1939 error = dmu_tx_assign(tx, TXG_WAIT); 1940 if (error) { 1941 dmu_tx_abort(tx); 1942 return (error); 1943 } 1944 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 1945 dmu_tx_commit(tx); 1946 1947 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1948 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 1949 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1950 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 1951 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1952 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 1953 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1954 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 1955 1956 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1957 (void) nvlist_add_uint64(nv, 1958 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 1959 (void) nvlist_add_uint64(nv, 1960 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 1961 (void) nvlist_add_uint64(nv, 1962 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 1963 if (version >= SPA_VERSION_DEDUP && 1964 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 1965 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 1966 (void) nvlist_add_uint64(nv, 1967 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 1968 } 1969 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 1970 nv, NULL); 1971 nvlist_free(nv); 1972 1973 zvol_free_extents(zv); 1974 zv->zv_flags &= ~ZVOL_DUMPIFIED; 1975 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 1976 /* wait for dmu_free_long_range to actually free the blocks */ 1977 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1978 tx = dmu_tx_create(os); 1979 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1980 error = dmu_tx_assign(tx, TXG_WAIT); 1981 if (error) { 1982 dmu_tx_abort(tx); 1983 return (error); 1984 } 1985 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 1986 zv->zv_volblocksize = vbs; 1987 dmu_tx_commit(tx); 1988 1989 return (0); 1990} 1991#endif /* sun */ 1992 1993static zvol_state_t * 1994zvol_geom_create(const char *name) 1995{ 1996 struct g_provider *pp; 1997 struct g_geom *gp; 1998 zvol_state_t *zv; 1999 2000 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name); 2001 gp->start = zvol_geom_start; 2002 gp->access = zvol_geom_access; 2003 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name); 2004 pp->sectorsize = DEV_BSIZE; 2005 2006 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP); 2007 zv->zv_provider = pp; 2008 zv->zv_state = 0; 2009 bioq_init(&zv->zv_queue); 2010 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF); 2011 2012 pp->private = zv; 2013 2014 return (zv); 2015} 2016 2017static void 2018zvol_geom_run(zvol_state_t *zv) 2019{ 2020 struct g_provider *pp; 2021 2022 pp = zv->zv_provider; 2023 g_error_provider(pp, 0); 2024 2025 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0, 2026 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER)); 2027} 2028 2029static void 2030zvol_geom_destroy(zvol_state_t *zv) 2031{ 2032 struct g_provider *pp; 2033 2034 g_topology_assert(); 2035 2036 mtx_lock(&zv->zv_queue_mtx); 2037 zv->zv_state = 1; 2038 wakeup_one(&zv->zv_queue); 2039 while (zv->zv_state != 2) 2040 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0); 2041 mtx_destroy(&zv->zv_queue_mtx); 2042 2043 pp = zv->zv_provider; 2044 zv->zv_provider = NULL; 2045 pp->private = NULL; 2046 g_wither_geom(pp->geom, ENXIO); 2047 2048 kmem_free(zv, sizeof(*zv)); 2049} 2050 2051static int 2052zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) 2053{ 2054 int count, error, flags; 2055 2056 g_topology_assert(); 2057 2058 /* 2059 * To make it easier we expect either open or close, but not both 2060 * at the same time. 2061 */ 2062 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) || 2063 (acr <= 0 && acw <= 0 && ace <= 0), 2064 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", 2065 pp->name, acr, acw, ace)); 2066 2067 if (pp->private == NULL) { 2068 if (acr <= 0 && acw <= 0 && ace <= 0) 2069 return (0); 2070 return (pp->error); 2071 } 2072 2073 /* 2074 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0, 2075 * because GEOM already handles that and handles it a bit differently. 2076 * GEOM allows for multiple read/exclusive consumers and ZFS allows 2077 * only one exclusive consumer, no matter if it is reader or writer. 2078 * I like better the way GEOM works so I'll leave it for GEOM to 2079 * decide what to do. 2080 */ 2081 2082 count = acr + acw + ace; 2083 if (count == 0) 2084 return (0); 2085 2086 flags = 0; 2087 if (acr != 0 || ace != 0) 2088 flags |= FREAD; 2089 if (acw != 0) 2090 flags |= FWRITE; 2091 2092 g_topology_unlock(); 2093 if (count > 0) 2094 error = zvol_open(pp, flags, count); 2095 else 2096 error = zvol_close(pp, flags, -count); 2097 g_topology_lock(); 2098 return (error); 2099} 2100 2101static void 2102zvol_geom_start(struct bio *bp) 2103{ 2104 zvol_state_t *zv; 2105 boolean_t first; 2106 2107 switch (bp->bio_cmd) { 2108 case BIO_READ: 2109 case BIO_WRITE: 2110 case BIO_FLUSH: 2111 zv = bp->bio_to->private; 2112 ASSERT(zv != NULL); 2113 mtx_lock(&zv->zv_queue_mtx); 2114 first = (bioq_first(&zv->zv_queue) == NULL); 2115 bioq_insert_tail(&zv->zv_queue, bp); 2116 mtx_unlock(&zv->zv_queue_mtx); 2117 if (first) 2118 wakeup_one(&zv->zv_queue); 2119 break; 2120 case BIO_GETATTR: 2121 case BIO_DELETE: 2122 default: 2123 g_io_deliver(bp, EOPNOTSUPP); 2124 break; 2125 } 2126} 2127 2128static void 2129zvol_geom_worker(void *arg) 2130{ 2131 zvol_state_t *zv; 2132 struct bio *bp; 2133 2134 thread_lock(curthread); 2135 sched_prio(curthread, PRIBIO); 2136 thread_unlock(curthread); 2137 2138 zv = arg; 2139 for (;;) { 2140 mtx_lock(&zv->zv_queue_mtx); 2141 bp = bioq_takefirst(&zv->zv_queue); 2142 if (bp == NULL) { 2143 if (zv->zv_state == 1) { 2144 zv->zv_state = 2; 2145 wakeup(&zv->zv_state); 2146 mtx_unlock(&zv->zv_queue_mtx); 2147 kthread_exit(); 2148 } 2149 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP, 2150 "zvol:io", 0); 2151 continue; 2152 } 2153 mtx_unlock(&zv->zv_queue_mtx); 2154 switch (bp->bio_cmd) { 2155 case BIO_FLUSH: 2156 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2157 g_io_deliver(bp, 0); 2158 break; 2159 case BIO_READ: 2160 case BIO_WRITE: 2161 zvol_strategy(bp); 2162 break; 2163 } 2164 } 2165} 2166 2167extern boolean_t dataset_name_hidden(const char *name); 2168 2169static int 2170zvol_create_snapshots(objset_t *os, const char *name) 2171{ 2172 uint64_t cookie, obj; 2173 char *sname; 2174 int error, len; 2175 2176 cookie = obj = 0; 2177 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2178 2179 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL, 2180 DS_FIND_SNAPSHOTS); 2181 2182 for (;;) { 2183 len = snprintf(sname, MAXPATHLEN, "%s@", name); 2184 if (len >= MAXPATHLEN) { 2185 dmu_objset_rele(os, FTAG); 2186 error = ENAMETOOLONG; 2187 break; 2188 } 2189 2190 error = dmu_snapshot_list_next(os, MAXPATHLEN - len, 2191 sname + len, &obj, &cookie, NULL); 2192 if (error != 0) { 2193 if (error == ENOENT) 2194 error = 0; 2195 break; 2196 } 2197 2198 if ((error = zvol_create_minor(sname)) != 0) { 2199 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2200 sname, error); 2201 break; 2202 } 2203 } 2204 2205 kmem_free(sname, MAXPATHLEN); 2206 return (error); 2207} 2208 2209int 2210zvol_create_minors(const char *name) 2211{ 2212 uint64_t cookie; 2213 objset_t *os; 2214 char *osname, *p; 2215 int error, len; 2216 2217 if (dataset_name_hidden(name)) 2218 return (0); 2219 2220 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2221 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2222 name, error); 2223 return (error); 2224 } 2225 if (dmu_objset_type(os) == DMU_OST_ZVOL) { 2226 if ((error = zvol_create_minor(name)) == 0) 2227 error = zvol_create_snapshots(os, name); 2228 else { 2229 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2230 name, error); 2231 } 2232 dmu_objset_rele(os, FTAG); 2233 return (error); 2234 } 2235 if (dmu_objset_type(os) != DMU_OST_ZFS) { 2236 dmu_objset_rele(os, FTAG); 2237 return (0); 2238 } 2239 2240 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2241 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) { 2242 dmu_objset_rele(os, FTAG); 2243 kmem_free(osname, MAXPATHLEN); 2244 return (ENOENT); 2245 } 2246 p = osname + strlen(osname); 2247 len = MAXPATHLEN - (p - osname); 2248 2249 /* Prefetch the datasets. */ 2250 cookie = 0; 2251 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) { 2252 if (!dataset_name_hidden(osname)) 2253 (void) dmu_objset_prefetch(osname, NULL); 2254 } 2255 2256 cookie = 0; 2257 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL, 2258 &cookie) == 0) { 2259 dmu_objset_rele(os, FTAG); 2260 (void)zvol_create_minors(osname); 2261 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2262 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2263 name, error); 2264 return (error); 2265 } 2266 } 2267 2268 dmu_objset_rele(os, FTAG); 2269 kmem_free(osname, MAXPATHLEN); 2270 return (0); 2271} 2272 2273static void 2274zvol_rename_minor(struct g_geom *gp, const char *newname) 2275{ 2276 struct g_provider *pp; 2277 zvol_state_t *zv; 2278 2279 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2280 g_topology_assert(); 2281 2282 pp = LIST_FIRST(&gp->provider); 2283 ASSERT(pp != NULL); 2284 zv = pp->private; 2285 ASSERT(zv != NULL); 2286 2287 zv->zv_provider = NULL; 2288 g_wither_provider(pp, ENXIO); 2289 2290 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname); 2291 pp->sectorsize = DEV_BSIZE; 2292 pp->mediasize = zv->zv_volsize; 2293 pp->private = zv; 2294 zv->zv_provider = pp; 2295 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name)); 2296 g_error_provider(pp, 0); 2297} 2298 2299void 2300zvol_rename_minors(const char *oldname, const char *newname) 2301{ 2302 char name[MAXPATHLEN]; 2303 struct g_provider *pp; 2304 struct g_geom *gp; 2305 size_t oldnamelen, newnamelen; 2306 zvol_state_t *zv; 2307 char *namebuf; 2308 2309 oldnamelen = strlen(oldname); 2310 newnamelen = strlen(newname); 2311 2312 DROP_GIANT(); 2313 mutex_enter(&spa_namespace_lock); 2314 g_topology_lock(); 2315 2316 LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) { 2317 pp = LIST_FIRST(&gp->provider); 2318 if (pp == NULL) 2319 continue; 2320 zv = pp->private; 2321 if (zv == NULL) 2322 continue; 2323 if (strcmp(zv->zv_name, oldname) == 0) { 2324 zvol_rename_minor(gp, newname); 2325 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && 2326 (zv->zv_name[oldnamelen] == '/' || 2327 zv->zv_name[oldnamelen] == '@')) { 2328 snprintf(name, sizeof(name), "%s%c%s", newname, 2329 zv->zv_name[oldnamelen], 2330 zv->zv_name + oldnamelen + 1); 2331 zvol_rename_minor(gp, name); 2332 } 2333 } 2334 2335 g_topology_unlock(); 2336 mutex_exit(&spa_namespace_lock); 2337 PICKUP_GIANT(); 2338}
|