zvol.c revision 264732
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 * Portions Copyright 2010 Robert Milkowski 30 * 31 * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 32 */ 33 34/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 35 36/* 37 * ZFS volume emulation driver. 38 * 39 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 40 * Volumes are accessed through the symbolic links named: 41 * 42 * /dev/zvol/dsk/<pool_name>/<dataset_name> 43 * /dev/zvol/rdsk/<pool_name>/<dataset_name> 44 * 45 * These links are created by the /dev filesystem (sdev_zvolops.c). 46 * Volumes are persistent through reboot. No user command needs to be 47 * run before opening and using a device. 48 * 49 * FreeBSD notes. 50 * On FreeBSD ZVOLs are simply GEOM providers like any other storage device 51 * in the system. 52 */ 53 54#include <sys/types.h> 55#include <sys/param.h> 56#include <sys/kernel.h> 57#include <sys/errno.h> 58#include <sys/uio.h> 59#include <sys/bio.h> 60#include <sys/buf.h> 61#include <sys/kmem.h> 62#include <sys/conf.h> 63#include <sys/cmn_err.h> 64#include <sys/stat.h> 65#include <sys/zap.h> 66#include <sys/spa.h> 67#include <sys/spa_impl.h> 68#include <sys/zio.h> 69#include <sys/dmu_traverse.h> 70#include <sys/dnode.h> 71#include <sys/dsl_dataset.h> 72#include <sys/dsl_prop.h> 73#include <sys/dkio.h> 74#include <sys/byteorder.h> 75#include <sys/sunddi.h> 76#include <sys/dirent.h> 77#include <sys/policy.h> 78#include <sys/fs/zfs.h> 79#include <sys/zfs_ioctl.h> 80#include <sys/zil.h> 81#include <sys/refcount.h> 82#include <sys/zfs_znode.h> 83#include <sys/zfs_rlock.h> 84#include <sys/vdev_impl.h> 85#include <sys/vdev_raidz.h> 86#include <sys/zvol.h> 87#include <sys/zil_impl.h> 88#include <sys/dbuf.h> 89#include <sys/dmu_tx.h> 90#include <sys/zfeature.h> 91#include <sys/zio_checksum.h> 92 93#include <geom/geom.h> 94 95#include "zfs_namecheck.h" 96 97struct g_class zfs_zvol_class = { 98 .name = "ZFS::ZVOL", 99 .version = G_VERSION, 100}; 101 102DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol); 103 104void *zfsdev_state; 105static char *zvol_tag = "zvol_tag"; 106 107#define ZVOL_DUMPSIZE "dumpsize" 108 109/* 110 * The spa_namespace_lock protects the zfsdev_state structure from being 111 * modified while it's being used, e.g. an open that comes in before a 112 * create finishes. It also protects temporary opens of the dataset so that, 113 * e.g., an open doesn't get a spurious EBUSY. 114 */ 115static uint32_t zvol_minors; 116 117typedef struct zvol_extent { 118 list_node_t ze_node; 119 dva_t ze_dva; /* dva associated with this extent */ 120 uint64_t ze_nblks; /* number of blocks in extent */ 121} zvol_extent_t; 122 123/* 124 * The in-core state of each volume. 125 */ 126typedef struct zvol_state { 127 char zv_name[MAXPATHLEN]; /* pool/dd name */ 128 uint64_t zv_volsize; /* amount of space we advertise */ 129 uint64_t zv_volblocksize; /* volume block size */ 130 struct g_provider *zv_provider; /* GEOM provider */ 131 uint8_t zv_min_bs; /* minimum addressable block shift */ 132 uint8_t zv_flags; /* readonly, dumpified, etc. */ 133 objset_t *zv_objset; /* objset handle */ 134 uint32_t zv_total_opens; /* total open count */ 135 zilog_t *zv_zilog; /* ZIL handle */ 136 list_t zv_extents; /* List of extents for dump */ 137 znode_t zv_znode; /* for range locking */ 138 dmu_buf_t *zv_dbuf; /* bonus handle */ 139 int zv_state; 140 struct bio_queue_head zv_queue; 141 struct mtx zv_queue_mtx; /* zv_queue mutex */ 142} zvol_state_t; 143 144/* 145 * zvol specific flags 146 */ 147#define ZVOL_RDONLY 0x1 148#define ZVOL_DUMPIFIED 0x2 149#define ZVOL_EXCL 0x4 150#define ZVOL_WCE 0x8 151 152/* 153 * zvol maximum transfer in one DMU tx. 154 */ 155int zvol_maxphys = DMU_MAX_ACCESS/2; 156 157extern int zfs_set_prop_nvlist(const char *, zprop_source_t, 158 nvlist_t *, nvlist_t *); 159static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, 160 uint64_t len, boolean_t sync); 161static int zvol_remove_zv(zvol_state_t *); 162static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio); 163static int zvol_dumpify(zvol_state_t *zv); 164static int zvol_dump_fini(zvol_state_t *zv); 165static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); 166 167static zvol_state_t *zvol_geom_create(const char *name); 168static void zvol_geom_run(zvol_state_t *zv); 169static void zvol_geom_destroy(zvol_state_t *zv); 170static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace); 171static void zvol_geom_start(struct bio *bp); 172static void zvol_geom_worker(void *arg); 173 174static void 175zvol_size_changed(zvol_state_t *zv) 176{ 177#ifdef sun 178 dev_t dev = makedevice(maj, min); 179 180 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 181 "Size", volsize) == DDI_SUCCESS); 182 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 183 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS); 184 185 /* Notify specfs to invalidate the cached size */ 186 spec_size_invalidate(dev, VBLK); 187 spec_size_invalidate(dev, VCHR); 188#else /* !sun */ 189 struct g_provider *pp; 190 191 pp = zv->zv_provider; 192 if (pp == NULL) 193 return; 194 g_topology_lock(); 195 g_resize_provider(pp, zv->zv_volsize); 196 g_topology_unlock(); 197#endif /* !sun */ 198} 199 200int 201zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 202{ 203 if (volsize == 0) 204 return (SET_ERROR(EINVAL)); 205 206 if (volsize % blocksize != 0) 207 return (SET_ERROR(EINVAL)); 208 209#ifdef _ILP32 210 if (volsize - 1 > SPEC_MAXOFFSET_T) 211 return (SET_ERROR(EOVERFLOW)); 212#endif 213 return (0); 214} 215 216int 217zvol_check_volblocksize(uint64_t volblocksize) 218{ 219 if (volblocksize < SPA_MINBLOCKSIZE || 220 volblocksize > SPA_MAXBLOCKSIZE || 221 !ISP2(volblocksize)) 222 return (SET_ERROR(EDOM)); 223 224 return (0); 225} 226 227int 228zvol_get_stats(objset_t *os, nvlist_t *nv) 229{ 230 int error; 231 dmu_object_info_t doi; 232 uint64_t val; 233 234 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 235 if (error) 236 return (error); 237 238 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 239 240 error = dmu_object_info(os, ZVOL_OBJ, &doi); 241 242 if (error == 0) { 243 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 244 doi.doi_data_block_size); 245 } 246 247 return (error); 248} 249 250static zvol_state_t * 251zvol_minor_lookup(const char *name) 252{ 253 struct g_provider *pp; 254 struct g_geom *gp; 255 zvol_state_t *zv = NULL; 256 257 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 258 259 g_topology_lock(); 260 LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) { 261 pp = LIST_FIRST(&gp->provider); 262 if (pp == NULL) 263 continue; 264 zv = pp->private; 265 if (zv == NULL) 266 continue; 267 if (strcmp(zv->zv_name, name) == 0) 268 break; 269 } 270 g_topology_unlock(); 271 272 return (gp != NULL ? zv : NULL); 273} 274 275/* extent mapping arg */ 276struct maparg { 277 zvol_state_t *ma_zv; 278 uint64_t ma_blks; 279}; 280 281/*ARGSUSED*/ 282static int 283zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 284 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 285{ 286 struct maparg *ma = arg; 287 zvol_extent_t *ze; 288 int bs = ma->ma_zv->zv_volblocksize; 289 290 if (BP_IS_HOLE(bp) || 291 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 292 return (0); 293 294 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 295 ma->ma_blks++; 296 297 /* Abort immediately if we have encountered gang blocks */ 298 if (BP_IS_GANG(bp)) 299 return (SET_ERROR(EFRAGS)); 300 301 /* 302 * See if the block is at the end of the previous extent. 303 */ 304 ze = list_tail(&ma->ma_zv->zv_extents); 305 if (ze && 306 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 307 DVA_GET_OFFSET(BP_IDENTITY(bp)) == 308 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 309 ze->ze_nblks++; 310 return (0); 311 } 312 313 dprintf_bp(bp, "%s", "next blkptr:"); 314 315 /* start a new extent */ 316 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 317 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 318 ze->ze_nblks = 1; 319 list_insert_tail(&ma->ma_zv->zv_extents, ze); 320 return (0); 321} 322 323static void 324zvol_free_extents(zvol_state_t *zv) 325{ 326 zvol_extent_t *ze; 327 328 while (ze = list_head(&zv->zv_extents)) { 329 list_remove(&zv->zv_extents, ze); 330 kmem_free(ze, sizeof (zvol_extent_t)); 331 } 332} 333 334static int 335zvol_get_lbas(zvol_state_t *zv) 336{ 337 objset_t *os = zv->zv_objset; 338 struct maparg ma; 339 int err; 340 341 ma.ma_zv = zv; 342 ma.ma_blks = 0; 343 zvol_free_extents(zv); 344 345 /* commit any in-flight changes before traversing the dataset */ 346 txg_wait_synced(dmu_objset_pool(os), 0); 347 err = traverse_dataset(dmu_objset_ds(os), 0, 348 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 349 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 350 zvol_free_extents(zv); 351 return (err ? err : EIO); 352 } 353 354 return (0); 355} 356 357/* ARGSUSED */ 358void 359zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 360{ 361 zfs_creat_t *zct = arg; 362 nvlist_t *nvprops = zct->zct_props; 363 int error; 364 uint64_t volblocksize, volsize; 365 366 VERIFY(nvlist_lookup_uint64(nvprops, 367 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 368 if (nvlist_lookup_uint64(nvprops, 369 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 370 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 371 372 /* 373 * These properties must be removed from the list so the generic 374 * property setting step won't apply to them. 375 */ 376 VERIFY(nvlist_remove_all(nvprops, 377 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 378 (void) nvlist_remove_all(nvprops, 379 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 380 381 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 382 DMU_OT_NONE, 0, tx); 383 ASSERT(error == 0); 384 385 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 386 DMU_OT_NONE, 0, tx); 387 ASSERT(error == 0); 388 389 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 390 ASSERT(error == 0); 391} 392 393/* 394 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we 395 * implement DKIOCFREE/free-long-range. 396 */ 397static int 398zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap) 399{ 400 uint64_t offset, length; 401 402 if (byteswap) 403 byteswap_uint64_array(lr, sizeof (*lr)); 404 405 offset = lr->lr_offset; 406 length = lr->lr_length; 407 408 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length)); 409} 410 411/* 412 * Replay a TX_WRITE ZIL transaction that didn't get committed 413 * after a system failure 414 */ 415static int 416zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 417{ 418 objset_t *os = zv->zv_objset; 419 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 420 uint64_t offset, length; 421 dmu_tx_t *tx; 422 int error; 423 424 if (byteswap) 425 byteswap_uint64_array(lr, sizeof (*lr)); 426 427 offset = lr->lr_offset; 428 length = lr->lr_length; 429 430 /* If it's a dmu_sync() block, write the whole block */ 431 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 432 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 433 if (length < blocksize) { 434 offset -= offset % blocksize; 435 length = blocksize; 436 } 437 } 438 439 tx = dmu_tx_create(os); 440 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 441 error = dmu_tx_assign(tx, TXG_WAIT); 442 if (error) { 443 dmu_tx_abort(tx); 444 } else { 445 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 446 dmu_tx_commit(tx); 447 } 448 449 return (error); 450} 451 452/* ARGSUSED */ 453static int 454zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 455{ 456 return (SET_ERROR(ENOTSUP)); 457} 458 459/* 460 * Callback vectors for replaying records. 461 * Only TX_WRITE and TX_TRUNCATE are needed for zvol. 462 */ 463zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 464 zvol_replay_err, /* 0 no such transaction type */ 465 zvol_replay_err, /* TX_CREATE */ 466 zvol_replay_err, /* TX_MKDIR */ 467 zvol_replay_err, /* TX_MKXATTR */ 468 zvol_replay_err, /* TX_SYMLINK */ 469 zvol_replay_err, /* TX_REMOVE */ 470 zvol_replay_err, /* TX_RMDIR */ 471 zvol_replay_err, /* TX_LINK */ 472 zvol_replay_err, /* TX_RENAME */ 473 zvol_replay_write, /* TX_WRITE */ 474 zvol_replay_truncate, /* TX_TRUNCATE */ 475 zvol_replay_err, /* TX_SETATTR */ 476 zvol_replay_err, /* TX_ACL */ 477 zvol_replay_err, /* TX_CREATE_ACL */ 478 zvol_replay_err, /* TX_CREATE_ATTR */ 479 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 480 zvol_replay_err, /* TX_MKDIR_ACL */ 481 zvol_replay_err, /* TX_MKDIR_ATTR */ 482 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 483 zvol_replay_err, /* TX_WRITE2 */ 484}; 485 486#ifdef sun 487int 488zvol_name2minor(const char *name, minor_t *minor) 489{ 490 zvol_state_t *zv; 491 492 mutex_enter(&spa_namespace_lock); 493 zv = zvol_minor_lookup(name); 494 if (minor && zv) 495 *minor = zv->zv_minor; 496 mutex_exit(&spa_namespace_lock); 497 return (zv ? 0 : -1); 498} 499#endif /* sun */ 500 501/* 502 * Create a minor node (plus a whole lot more) for the specified volume. 503 */ 504int 505zvol_create_minor(const char *name) 506{ 507 zfs_soft_state_t *zs; 508 zvol_state_t *zv; 509 objset_t *os; 510 dmu_object_info_t doi; 511 uint64_t volsize; 512 int error; 513 514 ZFS_LOG(1, "Creating ZVOL %s...", name); 515 516 mutex_enter(&spa_namespace_lock); 517 518 if (zvol_minor_lookup(name) != NULL) { 519 mutex_exit(&spa_namespace_lock); 520 return (SET_ERROR(EEXIST)); 521 } 522 523 /* lie and say we're read-only */ 524 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os); 525 526 if (error) { 527 mutex_exit(&spa_namespace_lock); 528 return (error); 529 } 530 531#ifdef sun 532 if ((minor = zfsdev_minor_alloc()) == 0) { 533 dmu_objset_disown(os, FTAG); 534 mutex_exit(&spa_namespace_lock); 535 return (SET_ERROR(ENXIO)); 536 } 537 538 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 539 dmu_objset_disown(os, FTAG); 540 mutex_exit(&spa_namespace_lock); 541 return (SET_ERROR(EAGAIN)); 542 } 543 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 544 (char *)name); 545 546 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 547 548 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 549 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 550 ddi_soft_state_free(zfsdev_state, minor); 551 dmu_objset_disown(os, FTAG); 552 mutex_exit(&spa_namespace_lock); 553 return (SET_ERROR(EAGAIN)); 554 } 555 556 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 557 558 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 559 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 560 ddi_remove_minor_node(zfs_dip, chrbuf); 561 ddi_soft_state_free(zfsdev_state, minor); 562 dmu_objset_disown(os, FTAG); 563 mutex_exit(&spa_namespace_lock); 564 return (SET_ERROR(EAGAIN)); 565 } 566 567 zs = ddi_get_soft_state(zfsdev_state, minor); 568 zs->zss_type = ZSST_ZVOL; 569 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 570#else /* !sun */ 571 572 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 573 if (error) { 574 ASSERT(error == 0); 575 dmu_objset_disown(os, zvol_tag); 576 mutex_exit(&spa_namespace_lock); 577 return (error); 578 } 579 580 DROP_GIANT(); 581 g_topology_lock(); 582 zv = zvol_geom_create(name); 583 zv->zv_volsize = volsize; 584 zv->zv_provider->mediasize = zv->zv_volsize; 585 586#endif /* !sun */ 587 588 (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 589 zv->zv_min_bs = DEV_BSHIFT; 590 zv->zv_objset = os; 591 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) 592 zv->zv_flags |= ZVOL_RDONLY; 593 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 594 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 595 sizeof (rl_t), offsetof(rl_t, r_node)); 596 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 597 offsetof(zvol_extent_t, ze_node)); 598 /* get and cache the blocksize */ 599 error = dmu_object_info(os, ZVOL_OBJ, &doi); 600 ASSERT(error == 0); 601 zv->zv_volblocksize = doi.doi_data_block_size; 602 603 if (spa_writeable(dmu_objset_spa(os))) { 604 if (zil_replay_disable) 605 zil_destroy(dmu_objset_zil(os), B_FALSE); 606 else 607 zil_replay(os, zv, zvol_replay_vector); 608 } 609 dmu_objset_disown(os, FTAG); 610 zv->zv_objset = NULL; 611 612 zvol_minors++; 613 614 mutex_exit(&spa_namespace_lock); 615 616 zvol_geom_run(zv); 617 618 g_topology_unlock(); 619 PICKUP_GIANT(); 620 621 ZFS_LOG(1, "ZVOL %s created.", name); 622 623 return (0); 624} 625 626/* 627 * Remove minor node for the specified volume. 628 */ 629static int 630zvol_remove_zv(zvol_state_t *zv) 631{ 632#ifdef sun 633 minor_t minor = zv->zv_minor; 634#endif 635 636 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 637 if (zv->zv_total_opens != 0) 638 return (SET_ERROR(EBUSY)); 639 640 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); 641 642#ifdef sun 643 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 644 ddi_remove_minor_node(zfs_dip, nmbuf); 645#endif /* sun */ 646 647 avl_destroy(&zv->zv_znode.z_range_avl); 648 mutex_destroy(&zv->zv_znode.z_range_lock); 649 650 zvol_geom_destroy(zv); 651 652 zvol_minors--; 653 return (0); 654} 655 656int 657zvol_remove_minor(const char *name) 658{ 659 zvol_state_t *zv; 660 int rc; 661 662 mutex_enter(&spa_namespace_lock); 663 if ((zv = zvol_minor_lookup(name)) == NULL) { 664 mutex_exit(&spa_namespace_lock); 665 return (SET_ERROR(ENXIO)); 666 } 667 g_topology_lock(); 668 rc = zvol_remove_zv(zv); 669 g_topology_unlock(); 670 mutex_exit(&spa_namespace_lock); 671 return (rc); 672} 673 674int 675zvol_first_open(zvol_state_t *zv) 676{ 677 objset_t *os; 678 uint64_t volsize; 679 int error; 680 uint64_t readonly; 681 682 /* lie and say we're read-only */ 683 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 684 zvol_tag, &os); 685 if (error) 686 return (error); 687 688 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 689 if (error) { 690 ASSERT(error == 0); 691 dmu_objset_disown(os, zvol_tag); 692 return (error); 693 } 694 zv->zv_objset = os; 695 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 696 if (error) { 697 dmu_objset_disown(os, zvol_tag); 698 return (error); 699 } 700 zv->zv_volsize = volsize; 701 zv->zv_zilog = zil_open(os, zvol_get_data); 702 zvol_size_changed(zv); 703 704 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 705 NULL) == 0); 706 if (readonly || dmu_objset_is_snapshot(os) || 707 !spa_writeable(dmu_objset_spa(os))) 708 zv->zv_flags |= ZVOL_RDONLY; 709 else 710 zv->zv_flags &= ~ZVOL_RDONLY; 711 return (error); 712} 713 714void 715zvol_last_close(zvol_state_t *zv) 716{ 717 zil_close(zv->zv_zilog); 718 zv->zv_zilog = NULL; 719 720 dmu_buf_rele(zv->zv_dbuf, zvol_tag); 721 zv->zv_dbuf = NULL; 722 723 /* 724 * Evict cached data 725 */ 726 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && 727 !(zv->zv_flags & ZVOL_RDONLY)) 728 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 729 dmu_objset_evict_dbufs(zv->zv_objset); 730 731 dmu_objset_disown(zv->zv_objset, zvol_tag); 732 zv->zv_objset = NULL; 733} 734 735#ifdef sun 736int 737zvol_prealloc(zvol_state_t *zv) 738{ 739 objset_t *os = zv->zv_objset; 740 dmu_tx_t *tx; 741 uint64_t refd, avail, usedobjs, availobjs; 742 uint64_t resid = zv->zv_volsize; 743 uint64_t off = 0; 744 745 /* Check the space usage before attempting to allocate the space */ 746 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 747 if (avail < zv->zv_volsize) 748 return (SET_ERROR(ENOSPC)); 749 750 /* Free old extents if they exist */ 751 zvol_free_extents(zv); 752 753 while (resid != 0) { 754 int error; 755 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); 756 757 tx = dmu_tx_create(os); 758 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 759 error = dmu_tx_assign(tx, TXG_WAIT); 760 if (error) { 761 dmu_tx_abort(tx); 762 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 763 return (error); 764 } 765 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 766 dmu_tx_commit(tx); 767 off += bytes; 768 resid -= bytes; 769 } 770 txg_wait_synced(dmu_objset_pool(os), 0); 771 772 return (0); 773} 774#endif /* sun */ 775 776static int 777zvol_update_volsize(objset_t *os, uint64_t volsize) 778{ 779 dmu_tx_t *tx; 780 int error; 781 782 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 783 784 tx = dmu_tx_create(os); 785 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 786 error = dmu_tx_assign(tx, TXG_WAIT); 787 if (error) { 788 dmu_tx_abort(tx); 789 return (error); 790 } 791 792 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 793 &volsize, tx); 794 dmu_tx_commit(tx); 795 796 if (error == 0) 797 error = dmu_free_long_range(os, 798 ZVOL_OBJ, volsize, DMU_OBJECT_END); 799 return (error); 800} 801 802void 803zvol_remove_minors(const char *name) 804{ 805 struct g_geom *gp, *gptmp; 806 struct g_provider *pp; 807 zvol_state_t *zv; 808 size_t namelen; 809 810 namelen = strlen(name); 811 812 DROP_GIANT(); 813 mutex_enter(&spa_namespace_lock); 814 g_topology_lock(); 815 816 LIST_FOREACH_SAFE(gp, &zfs_zvol_class.geom, geom, gptmp) { 817 pp = LIST_FIRST(&gp->provider); 818 if (pp == NULL) 819 continue; 820 zv = pp->private; 821 if (zv == NULL) 822 continue; 823 if (strcmp(zv->zv_name, name) == 0 || 824 (strncmp(zv->zv_name, name, namelen) == 0 && 825 zv->zv_name[namelen] == '/')) { 826 (void) zvol_remove_zv(zv); 827 } 828 } 829 830 g_topology_unlock(); 831 mutex_exit(&spa_namespace_lock); 832 PICKUP_GIANT(); 833} 834 835int 836zvol_set_volsize(const char *name, major_t maj, uint64_t volsize) 837{ 838 zvol_state_t *zv = NULL; 839 objset_t *os; 840 int error; 841 dmu_object_info_t doi; 842 uint64_t old_volsize = 0ULL; 843 uint64_t readonly; 844 845 mutex_enter(&spa_namespace_lock); 846 zv = zvol_minor_lookup(name); 847 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 848 mutex_exit(&spa_namespace_lock); 849 return (error); 850 } 851 852 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 853 (error = zvol_check_volsize(volsize, 854 doi.doi_data_block_size)) != 0) 855 goto out; 856 857 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, 858 NULL) == 0); 859 if (readonly) { 860 error = EROFS; 861 goto out; 862 } 863 864 error = zvol_update_volsize(os, volsize); 865 /* 866 * Reinitialize the dump area to the new size. If we 867 * failed to resize the dump area then restore it back to 868 * its original size. 869 */ 870 if (zv && error == 0) { 871#ifdef ZVOL_DUMP 872 if (zv->zv_flags & ZVOL_DUMPIFIED) { 873 old_volsize = zv->zv_volsize; 874 zv->zv_volsize = volsize; 875 if ((error = zvol_dumpify(zv)) != 0 || 876 (error = dumpvp_resize()) != 0) { 877 (void) zvol_update_volsize(os, old_volsize); 878 zv->zv_volsize = old_volsize; 879 error = zvol_dumpify(zv); 880 } 881 } 882#endif /* ZVOL_DUMP */ 883 if (error == 0) { 884 zv->zv_volsize = volsize; 885 zvol_size_changed(zv); 886 } 887 } 888 889#ifdef sun 890 /* 891 * Generate a LUN expansion event. 892 */ 893 if (zv && error == 0) { 894 sysevent_id_t eid; 895 nvlist_t *attr; 896 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 897 898 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 899 zv->zv_minor); 900 901 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 902 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 903 904 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 905 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 906 907 nvlist_free(attr); 908 kmem_free(physpath, MAXPATHLEN); 909 } 910#endif /* sun */ 911 912out: 913 dmu_objset_rele(os, FTAG); 914 915 mutex_exit(&spa_namespace_lock); 916 917 return (error); 918} 919 920/*ARGSUSED*/ 921static int 922zvol_open(struct g_provider *pp, int flag, int count) 923{ 924 zvol_state_t *zv; 925 int err = 0; 926 boolean_t locked = B_FALSE; 927 928 /* 929 * Protect against recursively entering spa_namespace_lock 930 * when spa_open() is used for a pool on a (local) ZVOL(s). 931 * This is needed since we replaced upstream zfsdev_state_lock 932 * with spa_namespace_lock in the ZVOL code. 933 * We are using the same trick as spa_open(). 934 * Note that calls in zvol_first_open which need to resolve 935 * pool name to a spa object will enter spa_open() 936 * recursively, but that function already has all the 937 * necessary protection. 938 */ 939 if (!MUTEX_HELD(&spa_namespace_lock)) { 940 mutex_enter(&spa_namespace_lock); 941 locked = B_TRUE; 942 } 943 944 zv = pp->private; 945 if (zv == NULL) { 946 if (locked) 947 mutex_exit(&spa_namespace_lock); 948 return (SET_ERROR(ENXIO)); 949 } 950 951 if (zv->zv_total_opens == 0) { 952 err = zvol_first_open(zv); 953 if (err) { 954 if (locked) 955 mutex_exit(&spa_namespace_lock); 956 return (err); 957 } 958 pp->mediasize = zv->zv_volsize; 959 pp->stripeoffset = 0; 960 pp->stripesize = zv->zv_volblocksize; 961 } 962 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 963 err = SET_ERROR(EROFS); 964 goto out; 965 } 966 if (zv->zv_flags & ZVOL_EXCL) { 967 err = SET_ERROR(EBUSY); 968 goto out; 969 } 970#ifdef FEXCL 971 if (flag & FEXCL) { 972 if (zv->zv_total_opens != 0) { 973 err = SET_ERROR(EBUSY); 974 goto out; 975 } 976 zv->zv_flags |= ZVOL_EXCL; 977 } 978#endif 979 980 zv->zv_total_opens += count; 981 if (locked) 982 mutex_exit(&spa_namespace_lock); 983 984 return (err); 985out: 986 if (zv->zv_total_opens == 0) 987 zvol_last_close(zv); 988 if (locked) 989 mutex_exit(&spa_namespace_lock); 990 return (err); 991} 992 993/*ARGSUSED*/ 994static int 995zvol_close(struct g_provider *pp, int flag, int count) 996{ 997 zvol_state_t *zv; 998 int error = 0; 999 boolean_t locked = B_FALSE; 1000 1001 /* See comment in zvol_open(). */ 1002 if (!MUTEX_HELD(&spa_namespace_lock)) { 1003 mutex_enter(&spa_namespace_lock); 1004 locked = B_TRUE; 1005 } 1006 1007 zv = pp->private; 1008 if (zv == NULL) { 1009 if (locked) 1010 mutex_exit(&spa_namespace_lock); 1011 return (SET_ERROR(ENXIO)); 1012 } 1013 1014 if (zv->zv_flags & ZVOL_EXCL) { 1015 ASSERT(zv->zv_total_opens == 1); 1016 zv->zv_flags &= ~ZVOL_EXCL; 1017 } 1018 1019 /* 1020 * If the open count is zero, this is a spurious close. 1021 * That indicates a bug in the kernel / DDI framework. 1022 */ 1023 ASSERT(zv->zv_total_opens != 0); 1024 1025 /* 1026 * You may get multiple opens, but only one close. 1027 */ 1028 zv->zv_total_opens -= count; 1029 1030 if (zv->zv_total_opens == 0) 1031 zvol_last_close(zv); 1032 1033 if (locked) 1034 mutex_exit(&spa_namespace_lock); 1035 return (error); 1036} 1037 1038static void 1039zvol_get_done(zgd_t *zgd, int error) 1040{ 1041 if (zgd->zgd_db) 1042 dmu_buf_rele(zgd->zgd_db, zgd); 1043 1044 zfs_range_unlock(zgd->zgd_rl); 1045 1046 if (error == 0 && zgd->zgd_bp) 1047 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1048 1049 kmem_free(zgd, sizeof (zgd_t)); 1050} 1051 1052/* 1053 * Get data to generate a TX_WRITE intent log record. 1054 */ 1055static int 1056zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1057{ 1058 zvol_state_t *zv = arg; 1059 objset_t *os = zv->zv_objset; 1060 uint64_t object = ZVOL_OBJ; 1061 uint64_t offset = lr->lr_offset; 1062 uint64_t size = lr->lr_length; /* length of user data */ 1063 blkptr_t *bp = &lr->lr_blkptr; 1064 dmu_buf_t *db; 1065 zgd_t *zgd; 1066 int error; 1067 1068 ASSERT(zio != NULL); 1069 ASSERT(size != 0); 1070 1071 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1072 zgd->zgd_zilog = zv->zv_zilog; 1073 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 1074 1075 /* 1076 * Write records come in two flavors: immediate and indirect. 1077 * For small writes it's cheaper to store the data with the 1078 * log record (immediate); for large writes it's cheaper to 1079 * sync the data and get a pointer to it (indirect) so that 1080 * we don't have to write the data twice. 1081 */ 1082 if (buf != NULL) { /* immediate write */ 1083 error = dmu_read(os, object, offset, size, buf, 1084 DMU_READ_NO_PREFETCH); 1085 } else { 1086 size = zv->zv_volblocksize; 1087 offset = P2ALIGN(offset, size); 1088 error = dmu_buf_hold(os, object, offset, zgd, &db, 1089 DMU_READ_NO_PREFETCH); 1090 if (error == 0) { 1091 blkptr_t *obp = dmu_buf_get_blkptr(db); 1092 if (obp) { 1093 ASSERT(BP_IS_HOLE(bp)); 1094 *bp = *obp; 1095 } 1096 1097 zgd->zgd_db = db; 1098 zgd->zgd_bp = bp; 1099 1100 ASSERT(db->db_offset == offset); 1101 ASSERT(db->db_size == size); 1102 1103 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1104 zvol_get_done, zgd); 1105 1106 if (error == 0) 1107 return (0); 1108 } 1109 } 1110 1111 zvol_get_done(zgd, error); 1112 1113 return (error); 1114} 1115 1116/* 1117 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 1118 * 1119 * We store data in the log buffers if it's small enough. 1120 * Otherwise we will later flush the data out via dmu_sync(). 1121 */ 1122ssize_t zvol_immediate_write_sz = 32768; 1123 1124static void 1125zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1126 boolean_t sync) 1127{ 1128 uint32_t blocksize = zv->zv_volblocksize; 1129 zilog_t *zilog = zv->zv_zilog; 1130 boolean_t slogging; 1131 ssize_t immediate_write_sz; 1132 1133 if (zil_replaying(zilog, tx)) 1134 return; 1135 1136 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1137 ? 0 : zvol_immediate_write_sz; 1138 1139 slogging = spa_has_slogs(zilog->zl_spa) && 1140 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 1141 1142 while (resid) { 1143 itx_t *itx; 1144 lr_write_t *lr; 1145 ssize_t len; 1146 itx_wr_state_t write_state; 1147 1148 /* 1149 * Unlike zfs_log_write() we can be called with 1150 * upto DMU_MAX_ACCESS/2 (5MB) writes. 1151 */ 1152 if (blocksize > immediate_write_sz && !slogging && 1153 resid >= blocksize && off % blocksize == 0) { 1154 write_state = WR_INDIRECT; /* uses dmu_sync */ 1155 len = blocksize; 1156 } else if (sync) { 1157 write_state = WR_COPIED; 1158 len = MIN(ZIL_MAX_LOG_DATA, resid); 1159 } else { 1160 write_state = WR_NEED_COPY; 1161 len = MIN(ZIL_MAX_LOG_DATA, resid); 1162 } 1163 1164 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1165 (write_state == WR_COPIED ? len : 0)); 1166 lr = (lr_write_t *)&itx->itx_lr; 1167 if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 1168 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1169 zil_itx_destroy(itx); 1170 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1171 lr = (lr_write_t *)&itx->itx_lr; 1172 write_state = WR_NEED_COPY; 1173 } 1174 1175 itx->itx_wr_state = write_state; 1176 if (write_state == WR_NEED_COPY) 1177 itx->itx_sod += len; 1178 lr->lr_foid = ZVOL_OBJ; 1179 lr->lr_offset = off; 1180 lr->lr_length = len; 1181 lr->lr_blkoff = 0; 1182 BP_ZERO(&lr->lr_blkptr); 1183 1184 itx->itx_private = zv; 1185 itx->itx_sync = sync; 1186 1187 zil_itx_assign(zilog, itx, tx); 1188 1189 off += len; 1190 resid -= len; 1191 } 1192} 1193 1194#ifdef sun 1195static int 1196zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset, 1197 uint64_t size, boolean_t doread, boolean_t isdump) 1198{ 1199 vdev_disk_t *dvd; 1200 int c; 1201 int numerrors = 0; 1202 1203 if (vd->vdev_ops == &vdev_mirror_ops || 1204 vd->vdev_ops == &vdev_replacing_ops || 1205 vd->vdev_ops == &vdev_spare_ops) { 1206 for (c = 0; c < vd->vdev_children; c++) { 1207 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1208 addr, offset, origoffset, size, doread, isdump); 1209 if (err != 0) { 1210 numerrors++; 1211 } else if (doread) { 1212 break; 1213 } 1214 } 1215 } 1216 1217 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops) 1218 return (numerrors < vd->vdev_children ? 0 : EIO); 1219 1220 if (doread && !vdev_readable(vd)) 1221 return (SET_ERROR(EIO)); 1222 else if (!doread && !vdev_writeable(vd)) 1223 return (SET_ERROR(EIO)); 1224 1225 if (vd->vdev_ops == &vdev_raidz_ops) { 1226 return (vdev_raidz_physio(vd, 1227 addr, size, offset, origoffset, doread, isdump)); 1228 } 1229 1230 offset += VDEV_LABEL_START_SIZE; 1231 1232 if (ddi_in_panic() || isdump) { 1233 ASSERT(!doread); 1234 if (doread) 1235 return (SET_ERROR(EIO)); 1236 dvd = vd->vdev_tsd; 1237 ASSERT3P(dvd, !=, NULL); 1238 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1239 lbtodb(size))); 1240 } else { 1241 dvd = vd->vdev_tsd; 1242 ASSERT3P(dvd, !=, NULL); 1243 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size, 1244 offset, doread ? B_READ : B_WRITE)); 1245 } 1246} 1247 1248static int 1249zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1250 boolean_t doread, boolean_t isdump) 1251{ 1252 vdev_t *vd; 1253 int error; 1254 zvol_extent_t *ze; 1255 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1256 1257 /* Must be sector aligned, and not stradle a block boundary. */ 1258 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1259 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1260 return (SET_ERROR(EINVAL)); 1261 } 1262 ASSERT(size <= zv->zv_volblocksize); 1263 1264 /* Locate the extent this belongs to */ 1265 ze = list_head(&zv->zv_extents); 1266 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1267 offset -= ze->ze_nblks * zv->zv_volblocksize; 1268 ze = list_next(&zv->zv_extents, ze); 1269 } 1270 1271 if (ze == NULL) 1272 return (SET_ERROR(EINVAL)); 1273 1274 if (!ddi_in_panic()) 1275 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1276 1277 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1278 offset += DVA_GET_OFFSET(&ze->ze_dva); 1279 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva), 1280 size, doread, isdump); 1281 1282 if (!ddi_in_panic()) 1283 spa_config_exit(spa, SCL_STATE, FTAG); 1284 1285 return (error); 1286} 1287#endif /* sun */ 1288 1289int 1290zvol_strategy(struct bio *bp) 1291{ 1292 zvol_state_t *zv = bp->bio_to->private; 1293 uint64_t off, volsize; 1294 size_t resid; 1295 char *addr; 1296 objset_t *os; 1297 rl_t *rl; 1298 int error = 0; 1299 boolean_t doread = (bp->bio_cmd == BIO_READ); 1300 boolean_t is_dumpified; 1301 boolean_t sync; 1302 1303 if (zv == NULL) { 1304 g_io_deliver(bp, ENXIO); 1305 return (0); 1306 } 1307 1308 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) { 1309 g_io_deliver(bp, EROFS); 1310 return (0); 1311 } 1312 1313 off = bp->bio_offset; 1314 volsize = zv->zv_volsize; 1315 1316 os = zv->zv_objset; 1317 ASSERT(os != NULL); 1318 1319 addr = bp->bio_data; 1320 resid = bp->bio_length; 1321 1322 if (resid > 0 && (off < 0 || off >= volsize)) { 1323 g_io_deliver(bp, EIO); 1324 return (0); 1325 } 1326 1327#ifdef illumos 1328 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED; 1329#else 1330 is_dumpified = B_FALSE; 1331#endif 1332 sync = !doread && !is_dumpified && 1333 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; 1334 1335 /* 1336 * There must be no buffer changes when doing a dmu_sync() because 1337 * we can't change the data whilst calculating the checksum. 1338 */ 1339 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1340 doread ? RL_READER : RL_WRITER); 1341 1342 if (bp->bio_cmd == BIO_DELETE) { 1343 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1344 error = dmu_tx_assign(tx, TXG_WAIT); 1345 if (error != 0) { 1346 dmu_tx_abort(tx); 1347 } else { 1348 zvol_log_truncate(zv, tx, off, resid, B_TRUE); 1349 dmu_tx_commit(tx); 1350 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1351 off, resid); 1352 resid = 0; 1353 } 1354 goto unlock; 1355 } 1356 1357 while (resid != 0 && off < volsize) { 1358 size_t size = MIN(resid, zvol_maxphys); 1359#ifdef illumos 1360 if (is_dumpified) { 1361 size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 1362 error = zvol_dumpio(zv, addr, off, size, 1363 doread, B_FALSE); 1364 } else if (doread) { 1365#else 1366 if (doread) { 1367#endif 1368 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1369 DMU_READ_PREFETCH); 1370 } else { 1371 dmu_tx_t *tx = dmu_tx_create(os); 1372 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1373 error = dmu_tx_assign(tx, TXG_WAIT); 1374 if (error) { 1375 dmu_tx_abort(tx); 1376 } else { 1377 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1378 zvol_log_write(zv, tx, off, size, sync); 1379 dmu_tx_commit(tx); 1380 } 1381 } 1382 if (error) { 1383 /* convert checksum errors into IO errors */ 1384 if (error == ECKSUM) 1385 error = SET_ERROR(EIO); 1386 break; 1387 } 1388 off += size; 1389 addr += size; 1390 resid -= size; 1391 } 1392unlock: 1393 zfs_range_unlock(rl); 1394 1395 bp->bio_completed = bp->bio_length - resid; 1396 if (bp->bio_completed < bp->bio_length) 1397 bp->bio_error = (off > volsize ? EINVAL : error); 1398 1399 if (sync) 1400 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1401 g_io_deliver(bp, 0); 1402 1403 return (0); 1404} 1405 1406#ifdef sun 1407/* 1408 * Set the buffer count to the zvol maximum transfer. 1409 * Using our own routine instead of the default minphys() 1410 * means that for larger writes we write bigger buffers on X86 1411 * (128K instead of 56K) and flush the disk write cache less often 1412 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1413 * 56K on X86 and 128K on sparc). 1414 */ 1415void 1416zvol_minphys(struct buf *bp) 1417{ 1418 if (bp->b_bcount > zvol_maxphys) 1419 bp->b_bcount = zvol_maxphys; 1420} 1421 1422int 1423zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1424{ 1425 minor_t minor = getminor(dev); 1426 zvol_state_t *zv; 1427 int error = 0; 1428 uint64_t size; 1429 uint64_t boff; 1430 uint64_t resid; 1431 1432 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1433 if (zv == NULL) 1434 return (SET_ERROR(ENXIO)); 1435 1436 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) 1437 return (SET_ERROR(EINVAL)); 1438 1439 boff = ldbtob(blkno); 1440 resid = ldbtob(nblocks); 1441 1442 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1443 1444 while (resid) { 1445 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1446 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1447 if (error) 1448 break; 1449 boff += size; 1450 addr += size; 1451 resid -= size; 1452 } 1453 1454 return (error); 1455} 1456 1457/*ARGSUSED*/ 1458int 1459zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1460{ 1461 minor_t minor = getminor(dev); 1462 zvol_state_t *zv; 1463 uint64_t volsize; 1464 rl_t *rl; 1465 int error = 0; 1466 1467 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1468 if (zv == NULL) 1469 return (SET_ERROR(ENXIO)); 1470 1471 volsize = zv->zv_volsize; 1472 if (uio->uio_resid > 0 && 1473 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1474 return (SET_ERROR(EIO)); 1475 1476 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1477 error = physio(zvol_strategy, NULL, dev, B_READ, 1478 zvol_minphys, uio); 1479 return (error); 1480 } 1481 1482 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1483 RL_READER); 1484 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1485 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1486 1487 /* don't read past the end */ 1488 if (bytes > volsize - uio->uio_loffset) 1489 bytes = volsize - uio->uio_loffset; 1490 1491 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1492 if (error) { 1493 /* convert checksum errors into IO errors */ 1494 if (error == ECKSUM) 1495 error = SET_ERROR(EIO); 1496 break; 1497 } 1498 } 1499 zfs_range_unlock(rl); 1500 return (error); 1501} 1502 1503/*ARGSUSED*/ 1504int 1505zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1506{ 1507 minor_t minor = getminor(dev); 1508 zvol_state_t *zv; 1509 uint64_t volsize; 1510 rl_t *rl; 1511 int error = 0; 1512 boolean_t sync; 1513 1514 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1515 if (zv == NULL) 1516 return (SET_ERROR(ENXIO)); 1517 1518 volsize = zv->zv_volsize; 1519 if (uio->uio_resid > 0 && 1520 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize)) 1521 return (SET_ERROR(EIO)); 1522 1523 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1524 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1525 zvol_minphys, uio); 1526 return (error); 1527 } 1528 1529 sync = !(zv->zv_flags & ZVOL_WCE) || 1530 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1531 1532 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1533 RL_WRITER); 1534 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1535 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1536 uint64_t off = uio->uio_loffset; 1537 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1538 1539 if (bytes > volsize - off) /* don't write past the end */ 1540 bytes = volsize - off; 1541 1542 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1543 error = dmu_tx_assign(tx, TXG_WAIT); 1544 if (error) { 1545 dmu_tx_abort(tx); 1546 break; 1547 } 1548 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1549 if (error == 0) 1550 zvol_log_write(zv, tx, off, bytes, sync); 1551 dmu_tx_commit(tx); 1552 1553 if (error) 1554 break; 1555 } 1556 zfs_range_unlock(rl); 1557 if (sync) 1558 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1559 return (error); 1560} 1561 1562int 1563zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1564{ 1565 struct uuid uuid = EFI_RESERVED; 1566 efi_gpe_t gpe = { 0 }; 1567 uint32_t crc; 1568 dk_efi_t efi; 1569 int length; 1570 char *ptr; 1571 1572 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1573 return (SET_ERROR(EFAULT)); 1574 ptr = (char *)(uintptr_t)efi.dki_data_64; 1575 length = efi.dki_length; 1576 /* 1577 * Some clients may attempt to request a PMBR for the 1578 * zvol. Currently this interface will return EINVAL to 1579 * such requests. These requests could be supported by 1580 * adding a check for lba == 0 and consing up an appropriate 1581 * PMBR. 1582 */ 1583 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1584 return (SET_ERROR(EINVAL)); 1585 1586 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1587 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1588 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1589 1590 if (efi.dki_lba == 1) { 1591 efi_gpt_t gpt = { 0 }; 1592 1593 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1594 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1595 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1596 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1597 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1598 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1599 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1600 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1601 gpt.efi_gpt_SizeOfPartitionEntry = 1602 LE_32(sizeof (efi_gpe_t)); 1603 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1604 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1605 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1606 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1607 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1608 flag)) 1609 return (SET_ERROR(EFAULT)); 1610 ptr += sizeof (gpt); 1611 length -= sizeof (gpt); 1612 } 1613 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1614 length), flag)) 1615 return (SET_ERROR(EFAULT)); 1616 return (0); 1617} 1618 1619/* 1620 * BEGIN entry points to allow external callers access to the volume. 1621 */ 1622/* 1623 * Return the volume parameters needed for access from an external caller. 1624 * These values are invariant as long as the volume is held open. 1625 */ 1626int 1627zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1628 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1629 void **rl_hdl, void **bonus_hdl) 1630{ 1631 zvol_state_t *zv; 1632 1633 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1634 if (zv == NULL) 1635 return (SET_ERROR(ENXIO)); 1636 if (zv->zv_flags & ZVOL_DUMPIFIED) 1637 return (SET_ERROR(ENXIO)); 1638 1639 ASSERT(blksize && max_xfer_len && minor_hdl && 1640 objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 1641 1642 *blksize = zv->zv_volblocksize; 1643 *max_xfer_len = (uint64_t)zvol_maxphys; 1644 *minor_hdl = zv; 1645 *objset_hdl = zv->zv_objset; 1646 *zil_hdl = zv->zv_zilog; 1647 *rl_hdl = &zv->zv_znode; 1648 *bonus_hdl = zv->zv_dbuf; 1649 return (0); 1650} 1651 1652/* 1653 * Return the current volume size to an external caller. 1654 * The size can change while the volume is open. 1655 */ 1656uint64_t 1657zvol_get_volume_size(void *minor_hdl) 1658{ 1659 zvol_state_t *zv = minor_hdl; 1660 1661 return (zv->zv_volsize); 1662} 1663 1664/* 1665 * Return the current WCE setting to an external caller. 1666 * The WCE setting can change while the volume is open. 1667 */ 1668int 1669zvol_get_volume_wce(void *minor_hdl) 1670{ 1671 zvol_state_t *zv = minor_hdl; 1672 1673 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 1674} 1675 1676/* 1677 * Entry point for external callers to zvol_log_write 1678 */ 1679void 1680zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 1681 boolean_t sync) 1682{ 1683 zvol_state_t *zv = minor_hdl; 1684 1685 zvol_log_write(zv, tx, off, resid, sync); 1686} 1687/* 1688 * END entry points to allow external callers access to the volume. 1689 */ 1690#endif /* sun */ 1691 1692/* 1693 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. 1694 */ 1695static void 1696zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, 1697 boolean_t sync) 1698{ 1699 itx_t *itx; 1700 lr_truncate_t *lr; 1701 zilog_t *zilog = zv->zv_zilog; 1702 1703 if (zil_replaying(zilog, tx)) 1704 return; 1705 1706 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 1707 lr = (lr_truncate_t *)&itx->itx_lr; 1708 lr->lr_foid = ZVOL_OBJ; 1709 lr->lr_offset = off; 1710 lr->lr_length = len; 1711 1712 itx->itx_sync = sync; 1713 zil_itx_assign(zilog, itx, tx); 1714} 1715 1716#ifdef sun 1717/* 1718 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1719 * Also a dirtbag dkio ioctl for unmap/free-block functionality. 1720 */ 1721/*ARGSUSED*/ 1722int 1723zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1724{ 1725 zvol_state_t *zv; 1726 struct dk_cinfo dki; 1727 struct dk_minfo dkm; 1728 struct dk_callback *dkc; 1729 int error = 0; 1730 rl_t *rl; 1731 1732 mutex_enter(&spa_namespace_lock); 1733 1734 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 1735 1736 if (zv == NULL) { 1737 mutex_exit(&spa_namespace_lock); 1738 return (SET_ERROR(ENXIO)); 1739 } 1740 ASSERT(zv->zv_total_opens > 0); 1741 1742 switch (cmd) { 1743 1744 case DKIOCINFO: 1745 bzero(&dki, sizeof (dki)); 1746 (void) strcpy(dki.dki_cname, "zvol"); 1747 (void) strcpy(dki.dki_dname, "zvol"); 1748 dki.dki_ctype = DKC_UNKNOWN; 1749 dki.dki_unit = getminor(dev); 1750 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); 1751 mutex_exit(&spa_namespace_lock); 1752 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1753 error = SET_ERROR(EFAULT); 1754 return (error); 1755 1756 case DKIOCGMEDIAINFO: 1757 bzero(&dkm, sizeof (dkm)); 1758 dkm.dki_lbsize = 1U << zv->zv_min_bs; 1759 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1760 dkm.dki_media_type = DK_UNKNOWN; 1761 mutex_exit(&spa_namespace_lock); 1762 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1763 error = SET_ERROR(EFAULT); 1764 return (error); 1765 1766 case DKIOCGETEFI: 1767 { 1768 uint64_t vs = zv->zv_volsize; 1769 uint8_t bs = zv->zv_min_bs; 1770 1771 mutex_exit(&spa_namespace_lock); 1772 error = zvol_getefi((void *)arg, flag, vs, bs); 1773 return (error); 1774 } 1775 1776 case DKIOCFLUSHWRITECACHE: 1777 dkc = (struct dk_callback *)arg; 1778 mutex_exit(&spa_namespace_lock); 1779 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1780 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1781 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1782 error = 0; 1783 } 1784 return (error); 1785 1786 case DKIOCGETWCE: 1787 { 1788 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1789 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1790 flag)) 1791 error = SET_ERROR(EFAULT); 1792 break; 1793 } 1794 case DKIOCSETWCE: 1795 { 1796 int wce; 1797 if (ddi_copyin((void *)arg, &wce, sizeof (int), 1798 flag)) { 1799 error = SET_ERROR(EFAULT); 1800 break; 1801 } 1802 if (wce) { 1803 zv->zv_flags |= ZVOL_WCE; 1804 mutex_exit(&spa_namespace_lock); 1805 } else { 1806 zv->zv_flags &= ~ZVOL_WCE; 1807 mutex_exit(&spa_namespace_lock); 1808 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1809 } 1810 return (0); 1811 } 1812 1813 case DKIOCGGEOM: 1814 case DKIOCGVTOC: 1815 /* 1816 * commands using these (like prtvtoc) expect ENOTSUP 1817 * since we're emulating an EFI label 1818 */ 1819 error = SET_ERROR(ENOTSUP); 1820 break; 1821 1822 case DKIOCDUMPINIT: 1823 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1824 RL_WRITER); 1825 error = zvol_dumpify(zv); 1826 zfs_range_unlock(rl); 1827 break; 1828 1829 case DKIOCDUMPFINI: 1830 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 1831 break; 1832 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1833 RL_WRITER); 1834 error = zvol_dump_fini(zv); 1835 zfs_range_unlock(rl); 1836 break; 1837 1838 case DKIOCFREE: 1839 { 1840 dkioc_free_t df; 1841 dmu_tx_t *tx; 1842 1843 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) { 1844 error = SET_ERROR(EFAULT); 1845 break; 1846 } 1847 1848 /* 1849 * Apply Postel's Law to length-checking. If they overshoot, 1850 * just blank out until the end, if there's a need to blank 1851 * out anything. 1852 */ 1853 if (df.df_start >= zv->zv_volsize) 1854 break; /* No need to do anything... */ 1855 if (df.df_start + df.df_length > zv->zv_volsize) 1856 df.df_length = DMU_OBJECT_END; 1857 1858 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length, 1859 RL_WRITER); 1860 tx = dmu_tx_create(zv->zv_objset); 1861 error = dmu_tx_assign(tx, TXG_WAIT); 1862 if (error != 0) { 1863 dmu_tx_abort(tx); 1864 } else { 1865 zvol_log_truncate(zv, tx, df.df_start, 1866 df.df_length, B_TRUE); 1867 dmu_tx_commit(tx); 1868 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1869 df.df_start, df.df_length); 1870 } 1871 1872 zfs_range_unlock(rl); 1873 1874 if (error == 0) { 1875 /* 1876 * If the write-cache is disabled or 'sync' property 1877 * is set to 'always' then treat this as a synchronous 1878 * operation (i.e. commit to zil). 1879 */ 1880 if (!(zv->zv_flags & ZVOL_WCE) || 1881 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) 1882 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1883 1884 /* 1885 * If the caller really wants synchronous writes, and 1886 * can't wait for them, don't return until the write 1887 * is done. 1888 */ 1889 if (df.df_flags & DF_WAIT_SYNC) { 1890 txg_wait_synced( 1891 dmu_objset_pool(zv->zv_objset), 0); 1892 } 1893 } 1894 break; 1895 } 1896 1897 default: 1898 error = SET_ERROR(ENOTTY); 1899 break; 1900 1901 } 1902 mutex_exit(&spa_namespace_lock); 1903 return (error); 1904} 1905#endif /* sun */ 1906 1907int 1908zvol_busy(void) 1909{ 1910 return (zvol_minors != 0); 1911} 1912 1913void 1914zvol_init(void) 1915{ 1916 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 1917 1) == 0); 1918 ZFS_LOG(1, "ZVOL Initialized."); 1919} 1920 1921void 1922zvol_fini(void) 1923{ 1924 ddi_soft_state_fini(&zfsdev_state); 1925 ZFS_LOG(1, "ZVOL Deinitialized."); 1926} 1927 1928#ifdef sun 1929/*ARGSUSED*/ 1930static int 1931zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx) 1932{ 1933 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1934 1935 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 1936 return (1); 1937 return (0); 1938} 1939 1940/*ARGSUSED*/ 1941static void 1942zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx) 1943{ 1944 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1945 1946 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx); 1947} 1948 1949static int 1950zvol_dump_init(zvol_state_t *zv, boolean_t resize) 1951{ 1952 dmu_tx_t *tx; 1953 int error; 1954 objset_t *os = zv->zv_objset; 1955 spa_t *spa = dmu_objset_spa(os); 1956 vdev_t *vd = spa->spa_root_vdev; 1957 nvlist_t *nv = NULL; 1958 uint64_t version = spa_version(spa); 1959 enum zio_checksum checksum; 1960 1961 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1962 ASSERT(vd->vdev_ops == &vdev_root_ops); 1963 1964 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 1965 DMU_OBJECT_END); 1966 /* wait for dmu_free_long_range to actually free the blocks */ 1967 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 1968 1969 /* 1970 * If the pool on which the dump device is being initialized has more 1971 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is 1972 * enabled. If so, bump that feature's counter to indicate that the 1973 * feature is active. We also check the vdev type to handle the 1974 * following case: 1975 * # zpool create test raidz disk1 disk2 disk3 1976 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev), 1977 * the raidz vdev itself has 3 children. 1978 */ 1979 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) { 1980 if (!spa_feature_is_enabled(spa, 1981 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 1982 return (SET_ERROR(ENOTSUP)); 1983 (void) dsl_sync_task(spa_name(spa), 1984 zfs_mvdev_dump_feature_check, 1985 zfs_mvdev_dump_activate_feature_sync, NULL, 2); 1986 } 1987 1988 tx = dmu_tx_create(os); 1989 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 1990 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 1991 error = dmu_tx_assign(tx, TXG_WAIT); 1992 if (error) { 1993 dmu_tx_abort(tx); 1994 return (error); 1995 } 1996 1997 /* 1998 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum 1999 * function. Otherwise, use the old default -- OFF. 2000 */ 2001 checksum = spa_feature_is_active(spa, 2002 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY : 2003 ZIO_CHECKSUM_OFF; 2004 2005 /* 2006 * If we are resizing the dump device then we only need to 2007 * update the refreservation to match the newly updated 2008 * zvolsize. Otherwise, we save off the original state of the 2009 * zvol so that we can restore them if the zvol is ever undumpified. 2010 */ 2011 if (resize) { 2012 error = zap_update(os, ZVOL_ZAP_OBJ, 2013 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2014 &zv->zv_volsize, tx); 2015 } else { 2016 uint64_t checksum, compress, refresrv, vbs, dedup; 2017 2018 error = dsl_prop_get_integer(zv->zv_name, 2019 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 2020 error = error ? error : dsl_prop_get_integer(zv->zv_name, 2021 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL); 2022 error = error ? error : dsl_prop_get_integer(zv->zv_name, 2023 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL); 2024 error = error ? error : dsl_prop_get_integer(zv->zv_name, 2025 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL); 2026 if (version >= SPA_VERSION_DEDUP) { 2027 error = error ? error : 2028 dsl_prop_get_integer(zv->zv_name, 2029 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 2030 } 2031 2032 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2033 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 2034 &compress, tx); 2035 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2036 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx); 2037 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2038 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2039 &refresrv, tx); 2040 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2041 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 2042 &vbs, tx); 2043 error = error ? error : dmu_object_set_blocksize( 2044 os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx); 2045 if (version >= SPA_VERSION_DEDUP) { 2046 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2047 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 2048 &dedup, tx); 2049 } 2050 if (error == 0) 2051 zv->zv_volblocksize = SPA_MAXBLOCKSIZE; 2052 } 2053 dmu_tx_commit(tx); 2054 2055 /* 2056 * We only need update the zvol's property if we are initializing 2057 * the dump area for the first time. 2058 */ 2059 if (!resize) { 2060 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2061 VERIFY(nvlist_add_uint64(nv, 2062 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 2063 VERIFY(nvlist_add_uint64(nv, 2064 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 2065 ZIO_COMPRESS_OFF) == 0); 2066 VERIFY(nvlist_add_uint64(nv, 2067 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 2068 checksum) == 0); 2069 if (version >= SPA_VERSION_DEDUP) { 2070 VERIFY(nvlist_add_uint64(nv, 2071 zfs_prop_to_name(ZFS_PROP_DEDUP), 2072 ZIO_CHECKSUM_OFF) == 0); 2073 } 2074 2075 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2076 nv, NULL); 2077 nvlist_free(nv); 2078 2079 if (error) 2080 return (error); 2081 } 2082 2083 /* Allocate the space for the dump */ 2084 error = zvol_prealloc(zv); 2085 return (error); 2086} 2087 2088static int 2089zvol_dumpify(zvol_state_t *zv) 2090{ 2091 int error = 0; 2092 uint64_t dumpsize = 0; 2093 dmu_tx_t *tx; 2094 objset_t *os = zv->zv_objset; 2095 2096 if (zv->zv_flags & ZVOL_RDONLY) 2097 return (SET_ERROR(EROFS)); 2098 2099 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 2100 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 2101 boolean_t resize = (dumpsize > 0); 2102 2103 if ((error = zvol_dump_init(zv, resize)) != 0) { 2104 (void) zvol_dump_fini(zv); 2105 return (error); 2106 } 2107 } 2108 2109 /* 2110 * Build up our lba mapping. 2111 */ 2112 error = zvol_get_lbas(zv); 2113 if (error) { 2114 (void) zvol_dump_fini(zv); 2115 return (error); 2116 } 2117 2118 tx = dmu_tx_create(os); 2119 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2120 error = dmu_tx_assign(tx, TXG_WAIT); 2121 if (error) { 2122 dmu_tx_abort(tx); 2123 (void) zvol_dump_fini(zv); 2124 return (error); 2125 } 2126 2127 zv->zv_flags |= ZVOL_DUMPIFIED; 2128 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 2129 &zv->zv_volsize, tx); 2130 dmu_tx_commit(tx); 2131 2132 if (error) { 2133 (void) zvol_dump_fini(zv); 2134 return (error); 2135 } 2136 2137 txg_wait_synced(dmu_objset_pool(os), 0); 2138 return (0); 2139} 2140 2141static int 2142zvol_dump_fini(zvol_state_t *zv) 2143{ 2144 dmu_tx_t *tx; 2145 objset_t *os = zv->zv_objset; 2146 nvlist_t *nv; 2147 int error = 0; 2148 uint64_t checksum, compress, refresrv, vbs, dedup; 2149 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 2150 2151 /* 2152 * Attempt to restore the zvol back to its pre-dumpified state. 2153 * This is a best-effort attempt as it's possible that not all 2154 * of these properties were initialized during the dumpify process 2155 * (i.e. error during zvol_dump_init). 2156 */ 2157 2158 tx = dmu_tx_create(os); 2159 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2160 error = dmu_tx_assign(tx, TXG_WAIT); 2161 if (error) { 2162 dmu_tx_abort(tx); 2163 return (error); 2164 } 2165 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 2166 dmu_tx_commit(tx); 2167 2168 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2169 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 2170 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2171 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 2172 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2173 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 2174 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2175 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 2176 2177 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2178 (void) nvlist_add_uint64(nv, 2179 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 2180 (void) nvlist_add_uint64(nv, 2181 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 2182 (void) nvlist_add_uint64(nv, 2183 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 2184 if (version >= SPA_VERSION_DEDUP && 2185 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2186 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 2187 (void) nvlist_add_uint64(nv, 2188 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 2189 } 2190 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2191 nv, NULL); 2192 nvlist_free(nv); 2193 2194 zvol_free_extents(zv); 2195 zv->zv_flags &= ~ZVOL_DUMPIFIED; 2196 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 2197 /* wait for dmu_free_long_range to actually free the blocks */ 2198 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2199 tx = dmu_tx_create(os); 2200 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2201 error = dmu_tx_assign(tx, TXG_WAIT); 2202 if (error) { 2203 dmu_tx_abort(tx); 2204 return (error); 2205 } 2206 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 2207 zv->zv_volblocksize = vbs; 2208 dmu_tx_commit(tx); 2209 2210 return (0); 2211} 2212#endif /* sun */ 2213 2214static zvol_state_t * 2215zvol_geom_create(const char *name) 2216{ 2217 struct g_provider *pp; 2218 struct g_geom *gp; 2219 zvol_state_t *zv; 2220 2221 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name); 2222 gp->start = zvol_geom_start; 2223 gp->access = zvol_geom_access; 2224 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name); 2225 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 2226 pp->sectorsize = DEV_BSIZE; 2227 2228 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP); 2229 zv->zv_provider = pp; 2230 zv->zv_state = 0; 2231 bioq_init(&zv->zv_queue); 2232 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF); 2233 2234 pp->private = zv; 2235 2236 return (zv); 2237} 2238 2239static void 2240zvol_geom_run(zvol_state_t *zv) 2241{ 2242 struct g_provider *pp; 2243 2244 pp = zv->zv_provider; 2245 g_error_provider(pp, 0); 2246 2247 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0, 2248 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER)); 2249} 2250 2251static void 2252zvol_geom_destroy(zvol_state_t *zv) 2253{ 2254 struct g_provider *pp; 2255 2256 g_topology_assert(); 2257 2258 mtx_lock(&zv->zv_queue_mtx); 2259 zv->zv_state = 1; 2260 wakeup_one(&zv->zv_queue); 2261 while (zv->zv_state != 2) 2262 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0); 2263 mtx_destroy(&zv->zv_queue_mtx); 2264 2265 pp = zv->zv_provider; 2266 zv->zv_provider = NULL; 2267 pp->private = NULL; 2268 g_wither_geom(pp->geom, ENXIO); 2269 2270 kmem_free(zv, sizeof(*zv)); 2271} 2272 2273static int 2274zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) 2275{ 2276 int count, error, flags; 2277 2278 g_topology_assert(); 2279 2280 /* 2281 * To make it easier we expect either open or close, but not both 2282 * at the same time. 2283 */ 2284 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) || 2285 (acr <= 0 && acw <= 0 && ace <= 0), 2286 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", 2287 pp->name, acr, acw, ace)); 2288 2289 if (pp->private == NULL) { 2290 if (acr <= 0 && acw <= 0 && ace <= 0) 2291 return (0); 2292 return (pp->error); 2293 } 2294 2295 /* 2296 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0, 2297 * because GEOM already handles that and handles it a bit differently. 2298 * GEOM allows for multiple read/exclusive consumers and ZFS allows 2299 * only one exclusive consumer, no matter if it is reader or writer. 2300 * I like better the way GEOM works so I'll leave it for GEOM to 2301 * decide what to do. 2302 */ 2303 2304 count = acr + acw + ace; 2305 if (count == 0) 2306 return (0); 2307 2308 flags = 0; 2309 if (acr != 0 || ace != 0) 2310 flags |= FREAD; 2311 if (acw != 0) 2312 flags |= FWRITE; 2313 2314 g_topology_unlock(); 2315 if (count > 0) 2316 error = zvol_open(pp, flags, count); 2317 else 2318 error = zvol_close(pp, flags, -count); 2319 g_topology_lock(); 2320 return (error); 2321} 2322 2323static void 2324zvol_geom_start(struct bio *bp) 2325{ 2326 zvol_state_t *zv; 2327 boolean_t first; 2328 2329 zv = bp->bio_to->private; 2330 ASSERT(zv != NULL); 2331 switch (bp->bio_cmd) { 2332 case BIO_FLUSH: 2333 if (!THREAD_CAN_SLEEP()) 2334 goto enqueue; 2335 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2336 g_io_deliver(bp, 0); 2337 break; 2338 case BIO_READ: 2339 case BIO_WRITE: 2340 case BIO_DELETE: 2341 if (!THREAD_CAN_SLEEP()) 2342 goto enqueue; 2343 zvol_strategy(bp); 2344 break; 2345 case BIO_GETATTR: 2346 default: 2347 g_io_deliver(bp, EOPNOTSUPP); 2348 break; 2349 } 2350 return; 2351 2352enqueue: 2353 mtx_lock(&zv->zv_queue_mtx); 2354 first = (bioq_first(&zv->zv_queue) == NULL); 2355 bioq_insert_tail(&zv->zv_queue, bp); 2356 mtx_unlock(&zv->zv_queue_mtx); 2357 if (first) 2358 wakeup_one(&zv->zv_queue); 2359} 2360 2361static void 2362zvol_geom_worker(void *arg) 2363{ 2364 zvol_state_t *zv; 2365 struct bio *bp; 2366 2367 thread_lock(curthread); 2368 sched_prio(curthread, PRIBIO); 2369 thread_unlock(curthread); 2370 2371 zv = arg; 2372 for (;;) { 2373 mtx_lock(&zv->zv_queue_mtx); 2374 bp = bioq_takefirst(&zv->zv_queue); 2375 if (bp == NULL) { 2376 if (zv->zv_state == 1) { 2377 zv->zv_state = 2; 2378 wakeup(&zv->zv_state); 2379 mtx_unlock(&zv->zv_queue_mtx); 2380 kthread_exit(); 2381 } 2382 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP, 2383 "zvol:io", 0); 2384 continue; 2385 } 2386 mtx_unlock(&zv->zv_queue_mtx); 2387 switch (bp->bio_cmd) { 2388 case BIO_FLUSH: 2389 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2390 g_io_deliver(bp, 0); 2391 break; 2392 case BIO_READ: 2393 case BIO_WRITE: 2394 zvol_strategy(bp); 2395 break; 2396 } 2397 } 2398} 2399 2400extern boolean_t dataset_name_hidden(const char *name); 2401 2402static int 2403zvol_create_snapshots(objset_t *os, const char *name) 2404{ 2405 uint64_t cookie, obj; 2406 char *sname; 2407 int error, len; 2408 2409 cookie = obj = 0; 2410 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2411 2412#if 0 2413 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL, 2414 DS_FIND_SNAPSHOTS); 2415#endif 2416 2417 for (;;) { 2418 len = snprintf(sname, MAXPATHLEN, "%s@", name); 2419 if (len >= MAXPATHLEN) { 2420 dmu_objset_rele(os, FTAG); 2421 error = ENAMETOOLONG; 2422 break; 2423 } 2424 2425 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 2426 error = dmu_snapshot_list_next(os, MAXPATHLEN - len, 2427 sname + len, &obj, &cookie, NULL); 2428 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 2429 if (error != 0) { 2430 if (error == ENOENT) 2431 error = 0; 2432 break; 2433 } 2434 2435 if ((error = zvol_create_minor(sname)) != 0) { 2436 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2437 sname, error); 2438 break; 2439 } 2440 } 2441 2442 kmem_free(sname, MAXPATHLEN); 2443 return (error); 2444} 2445 2446int 2447zvol_create_minors(const char *name) 2448{ 2449 uint64_t cookie; 2450 objset_t *os; 2451 char *osname, *p; 2452 int error, len; 2453 2454 if (dataset_name_hidden(name)) 2455 return (0); 2456 2457 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2458 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2459 name, error); 2460 return (error); 2461 } 2462 if (dmu_objset_type(os) == DMU_OST_ZVOL) { 2463 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG); 2464 dsl_pool_rele(dmu_objset_pool(os), FTAG); 2465 if ((error = zvol_create_minor(name)) == 0) 2466 error = zvol_create_snapshots(os, name); 2467 else { 2468 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2469 name, error); 2470 } 2471 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG); 2472 dsl_dataset_rele(os->os_dsl_dataset, FTAG); 2473 return (error); 2474 } 2475 if (dmu_objset_type(os) != DMU_OST_ZFS) { 2476 dmu_objset_rele(os, FTAG); 2477 return (0); 2478 } 2479 2480 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2481 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) { 2482 dmu_objset_rele(os, FTAG); 2483 kmem_free(osname, MAXPATHLEN); 2484 return (ENOENT); 2485 } 2486 p = osname + strlen(osname); 2487 len = MAXPATHLEN - (p - osname); 2488 2489#if 0 2490 /* Prefetch the datasets. */ 2491 cookie = 0; 2492 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) { 2493 if (!dataset_name_hidden(osname)) 2494 (void) dmu_objset_prefetch(osname, NULL); 2495 } 2496#endif 2497 2498 cookie = 0; 2499 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL, 2500 &cookie) == 0) { 2501 dmu_objset_rele(os, FTAG); 2502 (void)zvol_create_minors(osname); 2503 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2504 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2505 name, error); 2506 return (error); 2507 } 2508 } 2509 2510 dmu_objset_rele(os, FTAG); 2511 kmem_free(osname, MAXPATHLEN); 2512 return (0); 2513} 2514 2515static void 2516zvol_rename_minor(struct g_geom *gp, const char *newname) 2517{ 2518 struct g_provider *pp; 2519 zvol_state_t *zv; 2520 2521 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2522 g_topology_assert(); 2523 2524 pp = LIST_FIRST(&gp->provider); 2525 ASSERT(pp != NULL); 2526 zv = pp->private; 2527 ASSERT(zv != NULL); 2528 2529 zv->zv_provider = NULL; 2530 g_wither_provider(pp, ENXIO); 2531 2532 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname); 2533 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 2534 pp->sectorsize = DEV_BSIZE; 2535 pp->mediasize = zv->zv_volsize; 2536 pp->private = zv; 2537 zv->zv_provider = pp; 2538 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name)); 2539 g_error_provider(pp, 0); 2540} 2541 2542void 2543zvol_rename_minors(const char *oldname, const char *newname) 2544{ 2545 char name[MAXPATHLEN]; 2546 struct g_provider *pp; 2547 struct g_geom *gp; 2548 size_t oldnamelen, newnamelen; 2549 zvol_state_t *zv; 2550 char *namebuf; 2551 2552 oldnamelen = strlen(oldname); 2553 newnamelen = strlen(newname); 2554 2555 DROP_GIANT(); 2556 mutex_enter(&spa_namespace_lock); 2557 g_topology_lock(); 2558 2559 LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) { 2560 pp = LIST_FIRST(&gp->provider); 2561 if (pp == NULL) 2562 continue; 2563 zv = pp->private; 2564 if (zv == NULL) 2565 continue; 2566 if (strcmp(zv->zv_name, oldname) == 0) { 2567 zvol_rename_minor(gp, newname); 2568 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && 2569 (zv->zv_name[oldnamelen] == '/' || 2570 zv->zv_name[oldnamelen] == '@')) { 2571 snprintf(name, sizeof(name), "%s%c%s", newname, 2572 zv->zv_name[oldnamelen], 2573 zv->zv_name + oldnamelen + 1); 2574 zvol_rename_minor(gp, name); 2575 } 2576 } 2577 2578 g_topology_unlock(); 2579 mutex_exit(&spa_namespace_lock); 2580 PICKUP_GIANT(); 2581} 2582