zvol.c revision 297548
1209920Snwhitehorn/* 2209920Snwhitehorn * CDDL HEADER START 3209920Snwhitehorn * 4209920Snwhitehorn * The contents of this file are subject to the terms of the 5209920Snwhitehorn * Common Development and Distribution License (the "License"). 6209920Snwhitehorn * You may not use this file except in compliance with the License. 7209920Snwhitehorn * 8209920Snwhitehorn * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9209920Snwhitehorn * or http://www.opensolaris.org/os/licensing. 10209920Snwhitehorn * See the License for the specific language governing permissions 11209920Snwhitehorn * and limitations under the License. 12209920Snwhitehorn * 13209920Snwhitehorn * When distributing Covered Code, include this CDDL HEADER in each 14209920Snwhitehorn * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15209920Snwhitehorn * If applicable, add the following below this CDDL HEADER, with the 16209920Snwhitehorn * fields enclosed by brackets "[]" replaced with your own identifying 17209920Snwhitehorn * information: Portions Copyright [yyyy] [name of copyright owner] 18209920Snwhitehorn * 19209920Snwhitehorn * CDDL HEADER END 20209920Snwhitehorn */ 21209920Snwhitehorn/* 22209920Snwhitehorn * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23209920Snwhitehorn * 24209920Snwhitehorn * Copyright (c) 2006-2010 Pawel Jakub Dawidek <pjd@FreeBSD.org> 25209920Snwhitehorn * All rights reserved. 26209920Snwhitehorn * 27209920Snwhitehorn * Portions Copyright 2010 Robert Milkowski 28209920Snwhitehorn * 29209920Snwhitehorn * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 30209920Snwhitehorn * Copyright (c) 2012, 2014 by Delphix. All rights reserved. 31209920Snwhitehorn * Copyright (c) 2013, Joyent, Inc. All rights reserved. 32209920Snwhitehorn * Copyright (c) 2014 Integros [integros.com] 33209920Snwhitehorn */ 34209920Snwhitehorn 35209920Snwhitehorn/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 36209920Snwhitehorn 37209920Snwhitehorn/* 38209920Snwhitehorn * ZFS volume emulation driver. 39209920Snwhitehorn * 40209920Snwhitehorn * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 41209920Snwhitehorn * Volumes are accessed through the symbolic links named: 42209920Snwhitehorn * 43209920Snwhitehorn * /dev/zvol/dsk/<pool_name>/<dataset_name> 44209920Snwhitehorn * /dev/zvol/rdsk/<pool_name>/<dataset_name> 45209920Snwhitehorn * 46209920Snwhitehorn * These links are created by the /dev filesystem (sdev_zvolops.c). 47209920Snwhitehorn * Volumes are persistent through reboot. No user command needs to be 48209920Snwhitehorn * run before opening and using a device. 49209920Snwhitehorn * 50209920Snwhitehorn * FreeBSD notes. 51209920Snwhitehorn * On FreeBSD ZVOLs are simply GEOM providers like any other storage device 52209920Snwhitehorn * in the system. 53209920Snwhitehorn */ 54209920Snwhitehorn 55209920Snwhitehorn#include <sys/types.h> 56209920Snwhitehorn#include <sys/param.h> 57209920Snwhitehorn#include <sys/kernel.h> 58209920Snwhitehorn#include <sys/errno.h> 59209920Snwhitehorn#include <sys/uio.h> 60209920Snwhitehorn#include <sys/bio.h> 61209920Snwhitehorn#include <sys/buf.h> 62209920Snwhitehorn#include <sys/kmem.h> 63209920Snwhitehorn#include <sys/conf.h> 64209920Snwhitehorn#include <sys/cmn_err.h> 65209920Snwhitehorn#include <sys/stat.h> 66209920Snwhitehorn#include <sys/zap.h> 67209920Snwhitehorn#include <sys/spa.h> 68209920Snwhitehorn#include <sys/spa_impl.h> 69209920Snwhitehorn#include <sys/zio.h> 70209920Snwhitehorn#include <sys/disk.h> 71209920Snwhitehorn#include <sys/dmu_traverse.h> 72209920Snwhitehorn#include <sys/dnode.h> 73209920Snwhitehorn#include <sys/dsl_dataset.h> 74209920Snwhitehorn#include <sys/dsl_prop.h> 75209920Snwhitehorn#include <sys/dkio.h> 76209920Snwhitehorn#include <sys/byteorder.h> 77209920Snwhitehorn#include <sys/sunddi.h> 78209920Snwhitehorn#include <sys/dirent.h> 79209920Snwhitehorn#include <sys/policy.h> 80209920Snwhitehorn#include <sys/queue.h> 81209920Snwhitehorn#include <sys/fs/zfs.h> 82209920Snwhitehorn#include <sys/zfs_ioctl.h> 83209920Snwhitehorn#include <sys/zil.h> 84209920Snwhitehorn#include <sys/refcount.h> 85209920Snwhitehorn#include <sys/zfs_znode.h> 86209920Snwhitehorn#include <sys/zfs_rlock.h> 87209920Snwhitehorn#include <sys/vdev_impl.h> 88209920Snwhitehorn#include <sys/vdev_raidz.h> 89209920Snwhitehorn#include <sys/zvol.h> 90209920Snwhitehorn#include <sys/zil_impl.h> 91209920Snwhitehorn#include <sys/dbuf.h> 92209920Snwhitehorn#include <sys/dmu_tx.h> 93209920Snwhitehorn#include <sys/zfeature.h> 94209920Snwhitehorn#include <sys/zio_checksum.h> 95209920Snwhitehorn#include <sys/filio.h> 96209920Snwhitehorn 97209920Snwhitehorn#include <geom/geom.h> 98209920Snwhitehorn 99209920Snwhitehorn#include "zfs_namecheck.h" 100209920Snwhitehorn 101209920Snwhitehorn#ifndef illumos 102struct g_class zfs_zvol_class = { 103 .name = "ZFS::ZVOL", 104 .version = G_VERSION, 105}; 106 107DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol); 108 109#endif 110void *zfsdev_state; 111static char *zvol_tag = "zvol_tag"; 112 113#define ZVOL_DUMPSIZE "dumpsize" 114 115/* 116 * This lock protects the zfsdev_state structure from being modified 117 * while it's being used, e.g. an open that comes in before a create 118 * finishes. It also protects temporary opens of the dataset so that, 119 * e.g., an open doesn't get a spurious EBUSY. 120 */ 121#ifdef illumos 122kmutex_t zfsdev_state_lock; 123#else 124/* 125 * In FreeBSD we've replaced the upstream zfsdev_state_lock with the 126 * spa_namespace_lock in the ZVOL code. 127 */ 128#define zfsdev_state_lock spa_namespace_lock 129#endif 130static uint32_t zvol_minors; 131 132#ifndef illumos 133SYSCTL_DECL(_vfs_zfs); 134SYSCTL_NODE(_vfs_zfs, OID_AUTO, vol, CTLFLAG_RW, 0, "ZFS VOLUME"); 135static int volmode = ZFS_VOLMODE_GEOM; 136TUNABLE_INT("vfs.zfs.vol.mode", &volmode); 137SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &volmode, 0, 138 "Expose as GEOM providers (1), device files (2) or neither"); 139 140#endif 141typedef struct zvol_extent { 142 list_node_t ze_node; 143 dva_t ze_dva; /* dva associated with this extent */ 144 uint64_t ze_nblks; /* number of blocks in extent */ 145} zvol_extent_t; 146 147/* 148 * The in-core state of each volume. 149 */ 150typedef struct zvol_state { 151#ifndef illumos 152 LIST_ENTRY(zvol_state) zv_links; 153#endif 154 char zv_name[MAXPATHLEN]; /* pool/dd name */ 155 uint64_t zv_volsize; /* amount of space we advertise */ 156 uint64_t zv_volblocksize; /* volume block size */ 157#ifdef illumos 158 minor_t zv_minor; /* minor number */ 159#else 160 struct cdev *zv_dev; /* non-GEOM device */ 161 struct g_provider *zv_provider; /* GEOM provider */ 162#endif 163 uint8_t zv_min_bs; /* minimum addressable block shift */ 164 uint8_t zv_flags; /* readonly, dumpified, etc. */ 165 objset_t *zv_objset; /* objset handle */ 166#ifdef illumos 167 uint32_t zv_open_count[OTYPCNT]; /* open counts */ 168#endif 169 uint32_t zv_total_opens; /* total open count */ 170 zilog_t *zv_zilog; /* ZIL handle */ 171 list_t zv_extents; /* List of extents for dump */ 172 znode_t zv_znode; /* for range locking */ 173 dmu_buf_t *zv_dbuf; /* bonus handle */ 174#ifndef illumos 175 int zv_state; 176 int zv_volmode; /* Provide GEOM or cdev */ 177 struct bio_queue_head zv_queue; 178 struct mtx zv_queue_mtx; /* zv_queue mutex */ 179#endif 180} zvol_state_t; 181 182#ifndef illumos 183static LIST_HEAD(, zvol_state) all_zvols; 184#endif 185/* 186 * zvol specific flags 187 */ 188#define ZVOL_RDONLY 0x1 189#define ZVOL_DUMPIFIED 0x2 190#define ZVOL_EXCL 0x4 191#define ZVOL_WCE 0x8 192 193/* 194 * zvol maximum transfer in one DMU tx. 195 */ 196int zvol_maxphys = DMU_MAX_ACCESS/2; 197 198/* 199 * Toggle unmap functionality. 200 */ 201boolean_t zvol_unmap_enabled = B_TRUE; 202#ifndef illumos 203SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_enabled, CTLFLAG_RWTUN, 204 &zvol_unmap_enabled, 0, 205 "Enable UNMAP functionality"); 206 207static d_open_t zvol_d_open; 208static d_close_t zvol_d_close; 209static d_read_t zvol_read; 210static d_write_t zvol_write; 211static d_ioctl_t zvol_d_ioctl; 212static d_strategy_t zvol_strategy; 213 214static struct cdevsw zvol_cdevsw = { 215 .d_version = D_VERSION, 216 .d_open = zvol_d_open, 217 .d_close = zvol_d_close, 218 .d_read = zvol_read, 219 .d_write = zvol_write, 220 .d_ioctl = zvol_d_ioctl, 221 .d_strategy = zvol_strategy, 222 .d_name = "zvol", 223 .d_flags = D_DISK | D_TRACKCLOSE, 224}; 225 226static void zvol_geom_run(zvol_state_t *zv); 227static void zvol_geom_destroy(zvol_state_t *zv); 228static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace); 229static void zvol_geom_start(struct bio *bp); 230static void zvol_geom_worker(void *arg); 231static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, 232 uint64_t len, boolean_t sync); 233#endif /* !illumos */ 234 235extern int zfs_set_prop_nvlist(const char *, zprop_source_t, 236 nvlist_t *, nvlist_t *); 237static int zvol_remove_zv(zvol_state_t *); 238static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio); 239static int zvol_dumpify(zvol_state_t *zv); 240static int zvol_dump_fini(zvol_state_t *zv); 241static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); 242 243static void 244zvol_size_changed(zvol_state_t *zv, uint64_t volsize) 245{ 246#ifdef illumos 247 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor); 248 249 zv->zv_volsize = volsize; 250 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 251 "Size", volsize) == DDI_SUCCESS); 252 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 253 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS); 254 255 /* Notify specfs to invalidate the cached size */ 256 spec_size_invalidate(dev, VBLK); 257 spec_size_invalidate(dev, VCHR); 258#else /* !illumos */ 259 zv->zv_volsize = volsize; 260 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 261 struct g_provider *pp; 262 263 pp = zv->zv_provider; 264 if (pp == NULL) 265 return; 266 g_topology_lock(); 267 g_resize_provider(pp, zv->zv_volsize); 268 g_topology_unlock(); 269 } 270#endif /* illumos */ 271} 272 273int 274zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 275{ 276 if (volsize == 0) 277 return (SET_ERROR(EINVAL)); 278 279 if (volsize % blocksize != 0) 280 return (SET_ERROR(EINVAL)); 281 282#ifdef _ILP32 283 if (volsize - 1 > SPEC_MAXOFFSET_T) 284 return (SET_ERROR(EOVERFLOW)); 285#endif 286 return (0); 287} 288 289int 290zvol_check_volblocksize(uint64_t volblocksize) 291{ 292 if (volblocksize < SPA_MINBLOCKSIZE || 293 volblocksize > SPA_OLD_MAXBLOCKSIZE || 294 !ISP2(volblocksize)) 295 return (SET_ERROR(EDOM)); 296 297 return (0); 298} 299 300int 301zvol_get_stats(objset_t *os, nvlist_t *nv) 302{ 303 int error; 304 dmu_object_info_t doi; 305 uint64_t val; 306 307 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 308 if (error) 309 return (error); 310 311 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 312 313 error = dmu_object_info(os, ZVOL_OBJ, &doi); 314 315 if (error == 0) { 316 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 317 doi.doi_data_block_size); 318 } 319 320 return (error); 321} 322 323static zvol_state_t * 324zvol_minor_lookup(const char *name) 325{ 326#ifdef illumos 327 minor_t minor; 328#endif 329 zvol_state_t *zv; 330 331 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 332 333#ifdef illumos 334 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) { 335 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 336 if (zv == NULL) 337 continue; 338#else 339 LIST_FOREACH(zv, &all_zvols, zv_links) { 340#endif 341 if (strcmp(zv->zv_name, name) == 0) 342 return (zv); 343 } 344 345 return (NULL); 346} 347 348/* extent mapping arg */ 349struct maparg { 350 zvol_state_t *ma_zv; 351 uint64_t ma_blks; 352}; 353 354/*ARGSUSED*/ 355static int 356zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 357 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 358{ 359 struct maparg *ma = arg; 360 zvol_extent_t *ze; 361 int bs = ma->ma_zv->zv_volblocksize; 362 363 if (bp == NULL || BP_IS_HOLE(bp) || 364 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 365 return (0); 366 367 VERIFY(!BP_IS_EMBEDDED(bp)); 368 369 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 370 ma->ma_blks++; 371 372 /* Abort immediately if we have encountered gang blocks */ 373 if (BP_IS_GANG(bp)) 374 return (SET_ERROR(EFRAGS)); 375 376 /* 377 * See if the block is at the end of the previous extent. 378 */ 379 ze = list_tail(&ma->ma_zv->zv_extents); 380 if (ze && 381 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 382 DVA_GET_OFFSET(BP_IDENTITY(bp)) == 383 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 384 ze->ze_nblks++; 385 return (0); 386 } 387 388 dprintf_bp(bp, "%s", "next blkptr:"); 389 390 /* start a new extent */ 391 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 392 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 393 ze->ze_nblks = 1; 394 list_insert_tail(&ma->ma_zv->zv_extents, ze); 395 return (0); 396} 397 398static void 399zvol_free_extents(zvol_state_t *zv) 400{ 401 zvol_extent_t *ze; 402 403 while (ze = list_head(&zv->zv_extents)) { 404 list_remove(&zv->zv_extents, ze); 405 kmem_free(ze, sizeof (zvol_extent_t)); 406 } 407} 408 409static int 410zvol_get_lbas(zvol_state_t *zv) 411{ 412 objset_t *os = zv->zv_objset; 413 struct maparg ma; 414 int err; 415 416 ma.ma_zv = zv; 417 ma.ma_blks = 0; 418 zvol_free_extents(zv); 419 420 /* commit any in-flight changes before traversing the dataset */ 421 txg_wait_synced(dmu_objset_pool(os), 0); 422 err = traverse_dataset(dmu_objset_ds(os), 0, 423 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 424 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 425 zvol_free_extents(zv); 426 return (err ? err : EIO); 427 } 428 429 return (0); 430} 431 432/* ARGSUSED */ 433void 434zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 435{ 436 zfs_creat_t *zct = arg; 437 nvlist_t *nvprops = zct->zct_props; 438 int error; 439 uint64_t volblocksize, volsize; 440 441 VERIFY(nvlist_lookup_uint64(nvprops, 442 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 443 if (nvlist_lookup_uint64(nvprops, 444 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 445 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 446 447 /* 448 * These properties must be removed from the list so the generic 449 * property setting step won't apply to them. 450 */ 451 VERIFY(nvlist_remove_all(nvprops, 452 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 453 (void) nvlist_remove_all(nvprops, 454 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 455 456 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 457 DMU_OT_NONE, 0, tx); 458 ASSERT(error == 0); 459 460 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 461 DMU_OT_NONE, 0, tx); 462 ASSERT(error == 0); 463 464 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 465 ASSERT(error == 0); 466} 467 468/* 469 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we 470 * implement DKIOCFREE/free-long-range. 471 */ 472static int 473zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap) 474{ 475 uint64_t offset, length; 476 477 if (byteswap) 478 byteswap_uint64_array(lr, sizeof (*lr)); 479 480 offset = lr->lr_offset; 481 length = lr->lr_length; 482 483 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length)); 484} 485 486/* 487 * Replay a TX_WRITE ZIL transaction that didn't get committed 488 * after a system failure 489 */ 490static int 491zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 492{ 493 objset_t *os = zv->zv_objset; 494 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 495 uint64_t offset, length; 496 dmu_tx_t *tx; 497 int error; 498 499 if (byteswap) 500 byteswap_uint64_array(lr, sizeof (*lr)); 501 502 offset = lr->lr_offset; 503 length = lr->lr_length; 504 505 /* If it's a dmu_sync() block, write the whole block */ 506 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 507 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 508 if (length < blocksize) { 509 offset -= offset % blocksize; 510 length = blocksize; 511 } 512 } 513 514 tx = dmu_tx_create(os); 515 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 516 error = dmu_tx_assign(tx, TXG_WAIT); 517 if (error) { 518 dmu_tx_abort(tx); 519 } else { 520 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 521 dmu_tx_commit(tx); 522 } 523 524 return (error); 525} 526 527/* ARGSUSED */ 528static int 529zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 530{ 531 return (SET_ERROR(ENOTSUP)); 532} 533 534/* 535 * Callback vectors for replaying records. 536 * Only TX_WRITE and TX_TRUNCATE are needed for zvol. 537 */ 538zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 539 zvol_replay_err, /* 0 no such transaction type */ 540 zvol_replay_err, /* TX_CREATE */ 541 zvol_replay_err, /* TX_MKDIR */ 542 zvol_replay_err, /* TX_MKXATTR */ 543 zvol_replay_err, /* TX_SYMLINK */ 544 zvol_replay_err, /* TX_REMOVE */ 545 zvol_replay_err, /* TX_RMDIR */ 546 zvol_replay_err, /* TX_LINK */ 547 zvol_replay_err, /* TX_RENAME */ 548 zvol_replay_write, /* TX_WRITE */ 549 zvol_replay_truncate, /* TX_TRUNCATE */ 550 zvol_replay_err, /* TX_SETATTR */ 551 zvol_replay_err, /* TX_ACL */ 552 zvol_replay_err, /* TX_CREATE_ACL */ 553 zvol_replay_err, /* TX_CREATE_ATTR */ 554 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 555 zvol_replay_err, /* TX_MKDIR_ACL */ 556 zvol_replay_err, /* TX_MKDIR_ATTR */ 557 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 558 zvol_replay_err, /* TX_WRITE2 */ 559}; 560 561#ifdef illumos 562int 563zvol_name2minor(const char *name, minor_t *minor) 564{ 565 zvol_state_t *zv; 566 567 mutex_enter(&zfsdev_state_lock); 568 zv = zvol_minor_lookup(name); 569 if (minor && zv) 570 *minor = zv->zv_minor; 571 mutex_exit(&zfsdev_state_lock); 572 return (zv ? 0 : -1); 573} 574#endif /* illumos */ 575 576/* 577 * Create a minor node (plus a whole lot more) for the specified volume. 578 */ 579int 580zvol_create_minor(const char *name) 581{ 582 zfs_soft_state_t *zs; 583 zvol_state_t *zv; 584 objset_t *os; 585 dmu_object_info_t doi; 586#ifdef illumos 587 minor_t minor = 0; 588 char chrbuf[30], blkbuf[30]; 589#else 590 struct g_provider *pp; 591 struct g_geom *gp; 592 uint64_t volsize, mode; 593#endif 594 int error; 595 596#ifndef illumos 597 ZFS_LOG(1, "Creating ZVOL %s...", name); 598#endif 599 600 mutex_enter(&zfsdev_state_lock); 601 602 if (zvol_minor_lookup(name) != NULL) { 603 mutex_exit(&zfsdev_state_lock); 604 return (SET_ERROR(EEXIST)); 605 } 606 607 /* lie and say we're read-only */ 608 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os); 609 610 if (error) { 611 mutex_exit(&zfsdev_state_lock); 612 return (error); 613 } 614 615#ifdef illumos 616 if ((minor = zfsdev_minor_alloc()) == 0) { 617 dmu_objset_disown(os, FTAG); 618 mutex_exit(&zfsdev_state_lock); 619 return (SET_ERROR(ENXIO)); 620 } 621 622 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 623 dmu_objset_disown(os, FTAG); 624 mutex_exit(&zfsdev_state_lock); 625 return (SET_ERROR(EAGAIN)); 626 } 627 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 628 (char *)name); 629 630 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 631 632 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 633 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 634 ddi_soft_state_free(zfsdev_state, minor); 635 dmu_objset_disown(os, FTAG); 636 mutex_exit(&zfsdev_state_lock); 637 return (SET_ERROR(EAGAIN)); 638 } 639 640 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 641 642 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 643 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 644 ddi_remove_minor_node(zfs_dip, chrbuf); 645 ddi_soft_state_free(zfsdev_state, minor); 646 dmu_objset_disown(os, FTAG); 647 mutex_exit(&zfsdev_state_lock); 648 return (SET_ERROR(EAGAIN)); 649 } 650 651 zs = ddi_get_soft_state(zfsdev_state, minor); 652 zs->zss_type = ZSST_ZVOL; 653 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 654#else /* !illumos */ 655 656 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP); 657 zv->zv_state = 0; 658 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 659 if (error) { 660 kmem_free(zv, sizeof(*zv)); 661 dmu_objset_disown(os, zvol_tag); 662 mutex_exit(&zfsdev_state_lock); 663 return (error); 664 } 665 error = dsl_prop_get_integer(name, 666 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL); 667 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT) 668 mode = volmode; 669 670 DROP_GIANT(); 671 zv->zv_volsize = volsize; 672 zv->zv_volmode = mode; 673 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 674 g_topology_lock(); 675 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name); 676 gp->start = zvol_geom_start; 677 gp->access = zvol_geom_access; 678 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name); 679 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 680 pp->sectorsize = DEV_BSIZE; 681 pp->mediasize = zv->zv_volsize; 682 pp->private = zv; 683 684 zv->zv_provider = pp; 685 bioq_init(&zv->zv_queue); 686 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF); 687 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { 688 struct make_dev_args args; 689 690 make_dev_args_init(&args); 691 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK; 692 args.mda_devsw = &zvol_cdevsw; 693 args.mda_cr = NULL; 694 args.mda_uid = UID_ROOT; 695 args.mda_gid = GID_OPERATOR; 696 args.mda_mode = 0640; 697 args.mda_si_drv2 = zv; 698 error = make_dev_s(&args, &zv->zv_dev, 699 "%s/%s", ZVOL_DRIVER, name); 700 if (error != 0) { 701 kmem_free(zv, sizeof(*zv)); 702 dmu_objset_disown(os, FTAG); 703 mutex_exit(&zfsdev_state_lock); 704 return (error); 705 } 706 zv->zv_dev->si_iosize_max = MAXPHYS; 707 } 708 LIST_INSERT_HEAD(&all_zvols, zv, zv_links); 709#endif /* illumos */ 710 711 (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 712 zv->zv_min_bs = DEV_BSHIFT; 713#ifdef illumos 714 zv->zv_minor = minor; 715#endif 716 zv->zv_objset = os; 717 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) 718 zv->zv_flags |= ZVOL_RDONLY; 719 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 720 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 721 sizeof (rl_t), offsetof(rl_t, r_node)); 722 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 723 offsetof(zvol_extent_t, ze_node)); 724 /* get and cache the blocksize */ 725 error = dmu_object_info(os, ZVOL_OBJ, &doi); 726 ASSERT(error == 0); 727 zv->zv_volblocksize = doi.doi_data_block_size; 728 729 if (spa_writeable(dmu_objset_spa(os))) { 730 if (zil_replay_disable) 731 zil_destroy(dmu_objset_zil(os), B_FALSE); 732 else 733 zil_replay(os, zv, zvol_replay_vector); 734 } 735 dmu_objset_disown(os, FTAG); 736 zv->zv_objset = NULL; 737 738 zvol_minors++; 739 740 mutex_exit(&zfsdev_state_lock); 741#ifndef illumos 742 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 743 zvol_geom_run(zv); 744 g_topology_unlock(); 745 } 746 PICKUP_GIANT(); 747 748 ZFS_LOG(1, "ZVOL %s created.", name); 749#endif 750 751 return (0); 752} 753 754/* 755 * Remove minor node for the specified volume. 756 */ 757static int 758zvol_remove_zv(zvol_state_t *zv) 759{ 760#ifdef illumos 761 char nmbuf[20]; 762 minor_t minor = zv->zv_minor; 763#endif 764 765 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 766 if (zv->zv_total_opens != 0) 767 return (SET_ERROR(EBUSY)); 768 769#ifdef illumos 770 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 771 ddi_remove_minor_node(zfs_dip, nmbuf); 772 773 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor); 774 ddi_remove_minor_node(zfs_dip, nmbuf); 775#else 776 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); 777 778 LIST_REMOVE(zv, zv_links); 779 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 780 g_topology_lock(); 781 zvol_geom_destroy(zv); 782 g_topology_unlock(); 783 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) 784 destroy_dev(zv->zv_dev); 785#endif 786 787 avl_destroy(&zv->zv_znode.z_range_avl); 788 mutex_destroy(&zv->zv_znode.z_range_lock); 789 790 kmem_free(zv, sizeof (zvol_state_t)); 791#ifdef illumos 792 ddi_soft_state_free(zfsdev_state, minor); 793#endif 794 zvol_minors--; 795 return (0); 796} 797 798int 799zvol_remove_minor(const char *name) 800{ 801 zvol_state_t *zv; 802 int rc; 803 804 mutex_enter(&zfsdev_state_lock); 805 if ((zv = zvol_minor_lookup(name)) == NULL) { 806 mutex_exit(&zfsdev_state_lock); 807 return (SET_ERROR(ENXIO)); 808 } 809 rc = zvol_remove_zv(zv); 810 mutex_exit(&zfsdev_state_lock); 811 return (rc); 812} 813 814int 815zvol_first_open(zvol_state_t *zv) 816{ 817 objset_t *os; 818 uint64_t volsize; 819 int error; 820 uint64_t readonly; 821 822 /* lie and say we're read-only */ 823 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 824 zvol_tag, &os); 825 if (error) 826 return (error); 827 828 zv->zv_objset = os; 829 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 830 if (error) { 831 ASSERT(error == 0); 832 dmu_objset_disown(os, zvol_tag); 833 return (error); 834 } 835 836 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 837 if (error) { 838 dmu_objset_disown(os, zvol_tag); 839 return (error); 840 } 841 842 zvol_size_changed(zv, volsize); 843 zv->zv_zilog = zil_open(os, zvol_get_data); 844 845 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 846 NULL) == 0); 847 if (readonly || dmu_objset_is_snapshot(os) || 848 !spa_writeable(dmu_objset_spa(os))) 849 zv->zv_flags |= ZVOL_RDONLY; 850 else 851 zv->zv_flags &= ~ZVOL_RDONLY; 852 return (error); 853} 854 855void 856zvol_last_close(zvol_state_t *zv) 857{ 858 zil_close(zv->zv_zilog); 859 zv->zv_zilog = NULL; 860 861 dmu_buf_rele(zv->zv_dbuf, zvol_tag); 862 zv->zv_dbuf = NULL; 863 864 /* 865 * Evict cached data 866 */ 867 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && 868 !(zv->zv_flags & ZVOL_RDONLY)) 869 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 870 dmu_objset_evict_dbufs(zv->zv_objset); 871 872 dmu_objset_disown(zv->zv_objset, zvol_tag); 873 zv->zv_objset = NULL; 874} 875 876#ifdef illumos 877int 878zvol_prealloc(zvol_state_t *zv) 879{ 880 objset_t *os = zv->zv_objset; 881 dmu_tx_t *tx; 882 uint64_t refd, avail, usedobjs, availobjs; 883 uint64_t resid = zv->zv_volsize; 884 uint64_t off = 0; 885 886 /* Check the space usage before attempting to allocate the space */ 887 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 888 if (avail < zv->zv_volsize) 889 return (SET_ERROR(ENOSPC)); 890 891 /* Free old extents if they exist */ 892 zvol_free_extents(zv); 893 894 while (resid != 0) { 895 int error; 896 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE); 897 898 tx = dmu_tx_create(os); 899 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 900 error = dmu_tx_assign(tx, TXG_WAIT); 901 if (error) { 902 dmu_tx_abort(tx); 903 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 904 return (error); 905 } 906 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 907 dmu_tx_commit(tx); 908 off += bytes; 909 resid -= bytes; 910 } 911 txg_wait_synced(dmu_objset_pool(os), 0); 912 913 return (0); 914} 915#endif /* illumos */ 916 917static int 918zvol_update_volsize(objset_t *os, uint64_t volsize) 919{ 920 dmu_tx_t *tx; 921 int error; 922 923 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 924 925 tx = dmu_tx_create(os); 926 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 927 dmu_tx_mark_netfree(tx); 928 error = dmu_tx_assign(tx, TXG_WAIT); 929 if (error) { 930 dmu_tx_abort(tx); 931 return (error); 932 } 933 934 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 935 &volsize, tx); 936 dmu_tx_commit(tx); 937 938 if (error == 0) 939 error = dmu_free_long_range(os, 940 ZVOL_OBJ, volsize, DMU_OBJECT_END); 941 return (error); 942} 943 944void 945zvol_remove_minors(const char *name) 946{ 947#ifdef illumos 948 zvol_state_t *zv; 949 char *namebuf; 950 minor_t minor; 951 952 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP); 953 (void) strncpy(namebuf, name, strlen(name)); 954 (void) strcat(namebuf, "/"); 955 mutex_enter(&zfsdev_state_lock); 956 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) { 957 958 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 959 if (zv == NULL) 960 continue; 961 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0) 962 (void) zvol_remove_zv(zv); 963 } 964 kmem_free(namebuf, strlen(name) + 2); 965 966 mutex_exit(&zfsdev_state_lock); 967#else /* !illumos */ 968 zvol_state_t *zv, *tzv; 969 size_t namelen; 970 971 namelen = strlen(name); 972 973 DROP_GIANT(); 974 mutex_enter(&zfsdev_state_lock); 975 976 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) { 977 if (strcmp(zv->zv_name, name) == 0 || 978 (strncmp(zv->zv_name, name, namelen) == 0 && 979 strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' || 980 zv->zv_name[namelen] == '@'))) { 981 (void) zvol_remove_zv(zv); 982 } 983 } 984 985 mutex_exit(&zfsdev_state_lock); 986 PICKUP_GIANT(); 987#endif /* illumos */ 988} 989 990static int 991zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize) 992{ 993 uint64_t old_volsize = 0ULL; 994 int error = 0; 995 996 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 997 998 /* 999 * Reinitialize the dump area to the new size. If we 1000 * failed to resize the dump area then restore it back to 1001 * its original size. We must set the new volsize prior 1002 * to calling dumpvp_resize() to ensure that the devices' 1003 * size(9P) is not visible by the dump subsystem. 1004 */ 1005 old_volsize = zv->zv_volsize; 1006 zvol_size_changed(zv, volsize); 1007 1008#ifdef ZVOL_DUMP 1009 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1010 if ((error = zvol_dumpify(zv)) != 0 || 1011 (error = dumpvp_resize()) != 0) { 1012 int dumpify_error; 1013 1014 (void) zvol_update_volsize(zv->zv_objset, old_volsize); 1015 zvol_size_changed(zv, old_volsize); 1016 dumpify_error = zvol_dumpify(zv); 1017 error = dumpify_error ? dumpify_error : error; 1018 } 1019 } 1020#endif /* ZVOL_DUMP */ 1021 1022#ifdef illumos 1023 /* 1024 * Generate a LUN expansion event. 1025 */ 1026 if (error == 0) { 1027 sysevent_id_t eid; 1028 nvlist_t *attr; 1029 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 1030 1031 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 1032 zv->zv_minor); 1033 1034 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1035 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 1036 1037 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 1038 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 1039 1040 nvlist_free(attr); 1041 kmem_free(physpath, MAXPATHLEN); 1042 } 1043#endif /* illumos */ 1044 return (error); 1045} 1046 1047int 1048zvol_set_volsize(const char *name, uint64_t volsize) 1049{ 1050 zvol_state_t *zv = NULL; 1051 objset_t *os; 1052 int error; 1053 dmu_object_info_t doi; 1054 uint64_t readonly; 1055 boolean_t owned = B_FALSE; 1056 1057 error = dsl_prop_get_integer(name, 1058 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL); 1059 if (error != 0) 1060 return (error); 1061 if (readonly) 1062 return (SET_ERROR(EROFS)); 1063 1064 mutex_enter(&zfsdev_state_lock); 1065 zv = zvol_minor_lookup(name); 1066 1067 if (zv == NULL || zv->zv_objset == NULL) { 1068 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, 1069 FTAG, &os)) != 0) { 1070 mutex_exit(&zfsdev_state_lock); 1071 return (error); 1072 } 1073 owned = B_TRUE; 1074 if (zv != NULL) 1075 zv->zv_objset = os; 1076 } else { 1077 os = zv->zv_objset; 1078 } 1079 1080 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 1081 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0) 1082 goto out; 1083 1084 error = zvol_update_volsize(os, volsize); 1085 1086 if (error == 0 && zv != NULL) 1087 error = zvol_update_live_volsize(zv, volsize); 1088out: 1089 if (owned) { 1090 dmu_objset_disown(os, FTAG); 1091 if (zv != NULL) 1092 zv->zv_objset = NULL; 1093 } 1094 mutex_exit(&zfsdev_state_lock); 1095 return (error); 1096} 1097 1098/*ARGSUSED*/ 1099#ifdef illumos 1100int 1101zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr) 1102#else 1103static int 1104zvol_open(struct g_provider *pp, int flag, int count) 1105#endif 1106{ 1107 zvol_state_t *zv; 1108 int err = 0; 1109#ifdef illumos 1110 1111 mutex_enter(&zfsdev_state_lock); 1112 1113 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL); 1114 if (zv == NULL) { 1115 mutex_exit(&zfsdev_state_lock); 1116 return (SET_ERROR(ENXIO)); 1117 } 1118 1119 if (zv->zv_total_opens == 0) 1120 err = zvol_first_open(zv); 1121 if (err) { 1122 mutex_exit(&zfsdev_state_lock); 1123 return (err); 1124 } 1125#else /* !illumos */ 1126 boolean_t locked = B_FALSE; 1127 1128 /* 1129 * Protect against recursively entering spa_namespace_lock 1130 * when spa_open() is used for a pool on a (local) ZVOL(s). 1131 * This is needed since we replaced upstream zfsdev_state_lock 1132 * with spa_namespace_lock in the ZVOL code. 1133 * We are using the same trick as spa_open(). 1134 * Note that calls in zvol_first_open which need to resolve 1135 * pool name to a spa object will enter spa_open() 1136 * recursively, but that function already has all the 1137 * necessary protection. 1138 */ 1139 if (!MUTEX_HELD(&zfsdev_state_lock)) { 1140 mutex_enter(&zfsdev_state_lock); 1141 locked = B_TRUE; 1142 } 1143 1144 zv = pp->private; 1145 if (zv == NULL) { 1146 if (locked) 1147 mutex_exit(&zfsdev_state_lock); 1148 return (SET_ERROR(ENXIO)); 1149 } 1150 1151 if (zv->zv_total_opens == 0) { 1152 err = zvol_first_open(zv); 1153 if (err) { 1154 if (locked) 1155 mutex_exit(&zfsdev_state_lock); 1156 return (err); 1157 } 1158 pp->mediasize = zv->zv_volsize; 1159 pp->stripeoffset = 0; 1160 pp->stripesize = zv->zv_volblocksize; 1161 } 1162#endif /* illumos */ 1163 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 1164 err = SET_ERROR(EROFS); 1165 goto out; 1166 } 1167 if (zv->zv_flags & ZVOL_EXCL) { 1168 err = SET_ERROR(EBUSY); 1169 goto out; 1170 } 1171#ifdef FEXCL 1172 if (flag & FEXCL) { 1173 if (zv->zv_total_opens != 0) { 1174 err = SET_ERROR(EBUSY); 1175 goto out; 1176 } 1177 zv->zv_flags |= ZVOL_EXCL; 1178 } 1179#endif 1180 1181#ifdef illumos 1182 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) { 1183 zv->zv_open_count[otyp]++; 1184 zv->zv_total_opens++; 1185 } 1186 mutex_exit(&zfsdev_state_lock); 1187#else 1188 zv->zv_total_opens += count; 1189 if (locked) 1190 mutex_exit(&zfsdev_state_lock); 1191#endif 1192 1193 return (err); 1194out: 1195 if (zv->zv_total_opens == 0) 1196 zvol_last_close(zv); 1197#ifdef illumos 1198 mutex_exit(&zfsdev_state_lock); 1199#else 1200 if (locked) 1201 mutex_exit(&zfsdev_state_lock); 1202#endif 1203 return (err); 1204} 1205 1206/*ARGSUSED*/ 1207#ifdef illumos 1208int 1209zvol_close(dev_t dev, int flag, int otyp, cred_t *cr) 1210{ 1211 minor_t minor = getminor(dev); 1212 zvol_state_t *zv; 1213 int error = 0; 1214 1215 mutex_enter(&zfsdev_state_lock); 1216 1217 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1218 if (zv == NULL) { 1219 mutex_exit(&zfsdev_state_lock); 1220#else /* !illumos */ 1221static int 1222zvol_close(struct g_provider *pp, int flag, int count) 1223{ 1224 zvol_state_t *zv; 1225 int error = 0; 1226 boolean_t locked = B_FALSE; 1227 1228 /* See comment in zvol_open(). */ 1229 if (!MUTEX_HELD(&zfsdev_state_lock)) { 1230 mutex_enter(&zfsdev_state_lock); 1231 locked = B_TRUE; 1232 } 1233 1234 zv = pp->private; 1235 if (zv == NULL) { 1236 if (locked) 1237 mutex_exit(&zfsdev_state_lock); 1238#endif /* illumos */ 1239 return (SET_ERROR(ENXIO)); 1240 } 1241 1242 if (zv->zv_flags & ZVOL_EXCL) { 1243 ASSERT(zv->zv_total_opens == 1); 1244 zv->zv_flags &= ~ZVOL_EXCL; 1245 } 1246 1247 /* 1248 * If the open count is zero, this is a spurious close. 1249 * That indicates a bug in the kernel / DDI framework. 1250 */ 1251#ifdef illumos 1252 ASSERT(zv->zv_open_count[otyp] != 0); 1253#endif 1254 ASSERT(zv->zv_total_opens != 0); 1255 1256 /* 1257 * You may get multiple opens, but only one close. 1258 */ 1259#ifdef illumos 1260 zv->zv_open_count[otyp]--; 1261 zv->zv_total_opens--; 1262#else 1263 zv->zv_total_opens -= count; 1264#endif 1265 1266 if (zv->zv_total_opens == 0) 1267 zvol_last_close(zv); 1268 1269#ifdef illumos 1270 mutex_exit(&zfsdev_state_lock); 1271#else 1272 if (locked) 1273 mutex_exit(&zfsdev_state_lock); 1274#endif 1275 return (error); 1276} 1277 1278static void 1279zvol_get_done(zgd_t *zgd, int error) 1280{ 1281 if (zgd->zgd_db) 1282 dmu_buf_rele(zgd->zgd_db, zgd); 1283 1284 zfs_range_unlock(zgd->zgd_rl); 1285 1286 if (error == 0 && zgd->zgd_bp) 1287 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1288 1289 kmem_free(zgd, sizeof (zgd_t)); 1290} 1291 1292/* 1293 * Get data to generate a TX_WRITE intent log record. 1294 */ 1295static int 1296zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1297{ 1298 zvol_state_t *zv = arg; 1299 objset_t *os = zv->zv_objset; 1300 uint64_t object = ZVOL_OBJ; 1301 uint64_t offset = lr->lr_offset; 1302 uint64_t size = lr->lr_length; /* length of user data */ 1303 blkptr_t *bp = &lr->lr_blkptr; 1304 dmu_buf_t *db; 1305 zgd_t *zgd; 1306 int error; 1307 1308 ASSERT(zio != NULL); 1309 ASSERT(size != 0); 1310 1311 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1312 zgd->zgd_zilog = zv->zv_zilog; 1313 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 1314 1315 /* 1316 * Write records come in two flavors: immediate and indirect. 1317 * For small writes it's cheaper to store the data with the 1318 * log record (immediate); for large writes it's cheaper to 1319 * sync the data and get a pointer to it (indirect) so that 1320 * we don't have to write the data twice. 1321 */ 1322 if (buf != NULL) { /* immediate write */ 1323 error = dmu_read(os, object, offset, size, buf, 1324 DMU_READ_NO_PREFETCH); 1325 } else { 1326 size = zv->zv_volblocksize; 1327 offset = P2ALIGN(offset, size); 1328 error = dmu_buf_hold(os, object, offset, zgd, &db, 1329 DMU_READ_NO_PREFETCH); 1330 if (error == 0) { 1331 blkptr_t *obp = dmu_buf_get_blkptr(db); 1332 if (obp) { 1333 ASSERT(BP_IS_HOLE(bp)); 1334 *bp = *obp; 1335 } 1336 1337 zgd->zgd_db = db; 1338 zgd->zgd_bp = bp; 1339 1340 ASSERT(db->db_offset == offset); 1341 ASSERT(db->db_size == size); 1342 1343 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1344 zvol_get_done, zgd); 1345 1346 if (error == 0) 1347 return (0); 1348 } 1349 } 1350 1351 zvol_get_done(zgd, error); 1352 1353 return (error); 1354} 1355 1356/* 1357 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 1358 * 1359 * We store data in the log buffers if it's small enough. 1360 * Otherwise we will later flush the data out via dmu_sync(). 1361 */ 1362ssize_t zvol_immediate_write_sz = 32768; 1363 1364static void 1365zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1366 boolean_t sync) 1367{ 1368 uint32_t blocksize = zv->zv_volblocksize; 1369 zilog_t *zilog = zv->zv_zilog; 1370 boolean_t slogging; 1371 ssize_t immediate_write_sz; 1372 1373 if (zil_replaying(zilog, tx)) 1374 return; 1375 1376 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1377 ? 0 : zvol_immediate_write_sz; 1378 1379 slogging = spa_has_slogs(zilog->zl_spa) && 1380 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 1381 1382 while (resid) { 1383 itx_t *itx; 1384 lr_write_t *lr; 1385 ssize_t len; 1386 itx_wr_state_t write_state; 1387 1388 /* 1389 * Unlike zfs_log_write() we can be called with 1390 * upto DMU_MAX_ACCESS/2 (5MB) writes. 1391 */ 1392 if (blocksize > immediate_write_sz && !slogging && 1393 resid >= blocksize && off % blocksize == 0) { 1394 write_state = WR_INDIRECT; /* uses dmu_sync */ 1395 len = blocksize; 1396 } else if (sync) { 1397 write_state = WR_COPIED; 1398 len = MIN(ZIL_MAX_LOG_DATA, resid); 1399 } else { 1400 write_state = WR_NEED_COPY; 1401 len = MIN(ZIL_MAX_LOG_DATA, resid); 1402 } 1403 1404 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1405 (write_state == WR_COPIED ? len : 0)); 1406 lr = (lr_write_t *)&itx->itx_lr; 1407 if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 1408 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1409 zil_itx_destroy(itx); 1410 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1411 lr = (lr_write_t *)&itx->itx_lr; 1412 write_state = WR_NEED_COPY; 1413 } 1414 1415 itx->itx_wr_state = write_state; 1416 if (write_state == WR_NEED_COPY) 1417 itx->itx_sod += len; 1418 lr->lr_foid = ZVOL_OBJ; 1419 lr->lr_offset = off; 1420 lr->lr_length = len; 1421 lr->lr_blkoff = 0; 1422 BP_ZERO(&lr->lr_blkptr); 1423 1424 itx->itx_private = zv; 1425 itx->itx_sync = sync; 1426 1427 zil_itx_assign(zilog, itx, tx); 1428 1429 off += len; 1430 resid -= len; 1431 } 1432} 1433 1434#ifdef illumos 1435static int 1436zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset, 1437 uint64_t size, boolean_t doread, boolean_t isdump) 1438{ 1439 vdev_disk_t *dvd; 1440 int c; 1441 int numerrors = 0; 1442 1443 if (vd->vdev_ops == &vdev_mirror_ops || 1444 vd->vdev_ops == &vdev_replacing_ops || 1445 vd->vdev_ops == &vdev_spare_ops) { 1446 for (c = 0; c < vd->vdev_children; c++) { 1447 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1448 addr, offset, origoffset, size, doread, isdump); 1449 if (err != 0) { 1450 numerrors++; 1451 } else if (doread) { 1452 break; 1453 } 1454 } 1455 } 1456 1457 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops) 1458 return (numerrors < vd->vdev_children ? 0 : EIO); 1459 1460 if (doread && !vdev_readable(vd)) 1461 return (SET_ERROR(EIO)); 1462 else if (!doread && !vdev_writeable(vd)) 1463 return (SET_ERROR(EIO)); 1464 1465 if (vd->vdev_ops == &vdev_raidz_ops) { 1466 return (vdev_raidz_physio(vd, 1467 addr, size, offset, origoffset, doread, isdump)); 1468 } 1469 1470 offset += VDEV_LABEL_START_SIZE; 1471 1472 if (ddi_in_panic() || isdump) { 1473 ASSERT(!doread); 1474 if (doread) 1475 return (SET_ERROR(EIO)); 1476 dvd = vd->vdev_tsd; 1477 ASSERT3P(dvd, !=, NULL); 1478 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1479 lbtodb(size))); 1480 } else { 1481 dvd = vd->vdev_tsd; 1482 ASSERT3P(dvd, !=, NULL); 1483 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size, 1484 offset, doread ? B_READ : B_WRITE)); 1485 } 1486} 1487 1488static int 1489zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1490 boolean_t doread, boolean_t isdump) 1491{ 1492 vdev_t *vd; 1493 int error; 1494 zvol_extent_t *ze; 1495 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1496 1497 /* Must be sector aligned, and not stradle a block boundary. */ 1498 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1499 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1500 return (SET_ERROR(EINVAL)); 1501 } 1502 ASSERT(size <= zv->zv_volblocksize); 1503 1504 /* Locate the extent this belongs to */ 1505 ze = list_head(&zv->zv_extents); 1506 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1507 offset -= ze->ze_nblks * zv->zv_volblocksize; 1508 ze = list_next(&zv->zv_extents, ze); 1509 } 1510 1511 if (ze == NULL) 1512 return (SET_ERROR(EINVAL)); 1513 1514 if (!ddi_in_panic()) 1515 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1516 1517 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1518 offset += DVA_GET_OFFSET(&ze->ze_dva); 1519 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva), 1520 size, doread, isdump); 1521 1522 if (!ddi_in_panic()) 1523 spa_config_exit(spa, SCL_STATE, FTAG); 1524 1525 return (error); 1526} 1527 1528int 1529zvol_strategy(buf_t *bp) 1530{ 1531 zfs_soft_state_t *zs = NULL; 1532#else /* !illumos */ 1533void 1534zvol_strategy(struct bio *bp) 1535{ 1536#endif /* illumos */ 1537 zvol_state_t *zv; 1538 uint64_t off, volsize; 1539 size_t resid; 1540 char *addr; 1541 objset_t *os; 1542 rl_t *rl; 1543 int error = 0; 1544#ifdef illumos 1545 boolean_t doread = bp->b_flags & B_READ; 1546#else 1547 boolean_t doread = 0; 1548#endif 1549 boolean_t is_dumpified; 1550 boolean_t sync; 1551 1552#ifdef illumos 1553 if (getminor(bp->b_edev) == 0) { 1554 error = SET_ERROR(EINVAL); 1555 } else { 1556 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev)); 1557 if (zs == NULL) 1558 error = SET_ERROR(ENXIO); 1559 else if (zs->zss_type != ZSST_ZVOL) 1560 error = SET_ERROR(EINVAL); 1561 } 1562 1563 if (error) { 1564 bioerror(bp, error); 1565 biodone(bp); 1566 return (0); 1567 } 1568 1569 zv = zs->zss_data; 1570 1571 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) { 1572 bioerror(bp, EROFS); 1573 biodone(bp); 1574 return (0); 1575 } 1576 1577 off = ldbtob(bp->b_blkno); 1578#else /* !illumos */ 1579 if (bp->bio_to) 1580 zv = bp->bio_to->private; 1581 else 1582 zv = bp->bio_dev->si_drv2; 1583 1584 if (zv == NULL) { 1585 error = SET_ERROR(ENXIO); 1586 goto out; 1587 } 1588 1589 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) { 1590 error = SET_ERROR(EROFS); 1591 goto out; 1592 } 1593 1594 switch (bp->bio_cmd) { 1595 case BIO_FLUSH: 1596 goto sync; 1597 case BIO_READ: 1598 doread = 1; 1599 case BIO_WRITE: 1600 case BIO_DELETE: 1601 break; 1602 default: 1603 error = EOPNOTSUPP; 1604 goto out; 1605 } 1606 1607 off = bp->bio_offset; 1608#endif /* illumos */ 1609 volsize = zv->zv_volsize; 1610 1611 os = zv->zv_objset; 1612 ASSERT(os != NULL); 1613 1614#ifdef illumos 1615 bp_mapin(bp); 1616 addr = bp->b_un.b_addr; 1617 resid = bp->b_bcount; 1618 1619 if (resid > 0 && (off < 0 || off >= volsize)) { 1620 bioerror(bp, EIO); 1621 biodone(bp); 1622 return (0); 1623 } 1624 1625 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED; 1626 sync = ((!(bp->b_flags & B_ASYNC) && 1627 !(zv->zv_flags & ZVOL_WCE)) || 1628 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) && 1629 !doread && !is_dumpified; 1630#else /* !illumos */ 1631 addr = bp->bio_data; 1632 resid = bp->bio_length; 1633 1634 if (resid > 0 && (off < 0 || off >= volsize)) { 1635 error = SET_ERROR(EIO); 1636 goto out; 1637 } 1638 1639 is_dumpified = B_FALSE; 1640 sync = !doread && !is_dumpified && 1641 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; 1642#endif /* illumos */ 1643 1644 /* 1645 * There must be no buffer changes when doing a dmu_sync() because 1646 * we can't change the data whilst calculating the checksum. 1647 */ 1648 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1649 doread ? RL_READER : RL_WRITER); 1650 1651#ifndef illumos 1652 if (bp->bio_cmd == BIO_DELETE) { 1653 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1654 error = dmu_tx_assign(tx, TXG_WAIT); 1655 if (error != 0) { 1656 dmu_tx_abort(tx); 1657 } else { 1658 zvol_log_truncate(zv, tx, off, resid, B_TRUE); 1659 dmu_tx_commit(tx); 1660 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1661 off, resid); 1662 resid = 0; 1663 } 1664 goto unlock; 1665 } 1666#endif 1667 while (resid != 0 && off < volsize) { 1668 size_t size = MIN(resid, zvol_maxphys); 1669#ifdef illumos 1670 if (is_dumpified) { 1671 size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 1672 error = zvol_dumpio(zv, addr, off, size, 1673 doread, B_FALSE); 1674 } else if (doread) { 1675#else 1676 if (doread) { 1677#endif 1678 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1679 DMU_READ_PREFETCH); 1680 } else { 1681 dmu_tx_t *tx = dmu_tx_create(os); 1682 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1683 error = dmu_tx_assign(tx, TXG_WAIT); 1684 if (error) { 1685 dmu_tx_abort(tx); 1686 } else { 1687 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1688 zvol_log_write(zv, tx, off, size, sync); 1689 dmu_tx_commit(tx); 1690 } 1691 } 1692 if (error) { 1693 /* convert checksum errors into IO errors */ 1694 if (error == ECKSUM) 1695 error = SET_ERROR(EIO); 1696 break; 1697 } 1698 off += size; 1699 addr += size; 1700 resid -= size; 1701 } 1702#ifndef illumos 1703unlock: 1704#endif 1705 zfs_range_unlock(rl); 1706 1707#ifdef illumos 1708 if ((bp->b_resid = resid) == bp->b_bcount) 1709 bioerror(bp, off > volsize ? EINVAL : error); 1710 1711 if (sync) 1712 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1713 biodone(bp); 1714 1715 return (0); 1716#else /* !illumos */ 1717 bp->bio_completed = bp->bio_length - resid; 1718 if (bp->bio_completed < bp->bio_length && off > volsize) 1719 error = EINVAL; 1720 1721 if (sync) { 1722sync: 1723 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1724 } 1725out: 1726 if (bp->bio_to) 1727 g_io_deliver(bp, error); 1728 else 1729 biofinish(bp, NULL, error); 1730#endif /* illumos */ 1731} 1732 1733#ifdef illumos 1734/* 1735 * Set the buffer count to the zvol maximum transfer. 1736 * Using our own routine instead of the default minphys() 1737 * means that for larger writes we write bigger buffers on X86 1738 * (128K instead of 56K) and flush the disk write cache less often 1739 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1740 * 56K on X86 and 128K on sparc). 1741 */ 1742void 1743zvol_minphys(struct buf *bp) 1744{ 1745 if (bp->b_bcount > zvol_maxphys) 1746 bp->b_bcount = zvol_maxphys; 1747} 1748 1749int 1750zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1751{ 1752 minor_t minor = getminor(dev); 1753 zvol_state_t *zv; 1754 int error = 0; 1755 uint64_t size; 1756 uint64_t boff; 1757 uint64_t resid; 1758 1759 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1760 if (zv == NULL) 1761 return (SET_ERROR(ENXIO)); 1762 1763 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) 1764 return (SET_ERROR(EINVAL)); 1765 1766 boff = ldbtob(blkno); 1767 resid = ldbtob(nblocks); 1768 1769 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1770 1771 while (resid) { 1772 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1773 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1774 if (error) 1775 break; 1776 boff += size; 1777 addr += size; 1778 resid -= size; 1779 } 1780 1781 return (error); 1782} 1783 1784/*ARGSUSED*/ 1785int 1786zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1787{ 1788 minor_t minor = getminor(dev); 1789#else /* !illumos */ 1790int 1791zvol_read(struct cdev *dev, struct uio *uio, int ioflag) 1792{ 1793#endif /* illumos */ 1794 zvol_state_t *zv; 1795 uint64_t volsize; 1796 rl_t *rl; 1797 int error = 0; 1798 1799#ifdef illumos 1800 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1801 if (zv == NULL) 1802 return (SET_ERROR(ENXIO)); 1803#else 1804 zv = dev->si_drv2; 1805#endif 1806 1807 volsize = zv->zv_volsize; 1808 /* uio_loffset == volsize isn't an error as its required for EOF processing. */ 1809 if (uio->uio_resid > 0 && 1810 (uio->uio_loffset < 0 || uio->uio_loffset > volsize)) 1811 return (SET_ERROR(EIO)); 1812 1813#ifdef illumos 1814 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1815 error = physio(zvol_strategy, NULL, dev, B_READ, 1816 zvol_minphys, uio); 1817 return (error); 1818 } 1819#endif 1820 1821 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1822 RL_READER); 1823 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1824 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1825 1826 /* don't read past the end */ 1827 if (bytes > volsize - uio->uio_loffset) 1828 bytes = volsize - uio->uio_loffset; 1829 1830 error = dmu_read_uio_dbuf(zv->zv_dbuf, uio, bytes); 1831 if (error) { 1832 /* convert checksum errors into IO errors */ 1833 if (error == ECKSUM) 1834 error = SET_ERROR(EIO); 1835 break; 1836 } 1837 } 1838 zfs_range_unlock(rl); 1839 return (error); 1840} 1841 1842#ifdef illumos 1843/*ARGSUSED*/ 1844int 1845zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1846{ 1847 minor_t minor = getminor(dev); 1848#else /* !illumos */ 1849int 1850zvol_write(struct cdev *dev, struct uio *uio, int ioflag) 1851{ 1852#endif /* illumos */ 1853 zvol_state_t *zv; 1854 uint64_t volsize; 1855 rl_t *rl; 1856 int error = 0; 1857 boolean_t sync; 1858 1859#ifdef illumos 1860 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1861 if (zv == NULL) 1862 return (SET_ERROR(ENXIO)); 1863#else 1864 zv = dev->si_drv2; 1865#endif 1866 1867 volsize = zv->zv_volsize; 1868 /* uio_loffset == volsize isn't an error as its required for EOF processing. */ 1869 if (uio->uio_resid > 0 && 1870 (uio->uio_loffset < 0 || uio->uio_loffset > volsize)) 1871 return (SET_ERROR(EIO)); 1872 1873#ifdef illumos 1874 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1875 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1876 zvol_minphys, uio); 1877 return (error); 1878 } 1879 1880 sync = !(zv->zv_flags & ZVOL_WCE) || 1881#else 1882 sync = (ioflag & IO_SYNC) || 1883#endif 1884 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1885 1886 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1887 RL_WRITER); 1888 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1889 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1890 uint64_t off = uio->uio_loffset; 1891 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1892 1893 if (bytes > volsize - off) /* don't write past the end */ 1894 bytes = volsize - off; 1895 1896 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1897 error = dmu_tx_assign(tx, TXG_WAIT); 1898 if (error) { 1899 dmu_tx_abort(tx); 1900 break; 1901 } 1902 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1903 if (error == 0) 1904 zvol_log_write(zv, tx, off, bytes, sync); 1905 dmu_tx_commit(tx); 1906 1907 if (error) 1908 break; 1909 } 1910 zfs_range_unlock(rl); 1911 if (sync) 1912 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1913 return (error); 1914} 1915 1916#ifdef illumos 1917int 1918zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1919{ 1920 struct uuid uuid = EFI_RESERVED; 1921 efi_gpe_t gpe = { 0 }; 1922 uint32_t crc; 1923 dk_efi_t efi; 1924 int length; 1925 char *ptr; 1926 1927 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1928 return (SET_ERROR(EFAULT)); 1929 ptr = (char *)(uintptr_t)efi.dki_data_64; 1930 length = efi.dki_length; 1931 /* 1932 * Some clients may attempt to request a PMBR for the 1933 * zvol. Currently this interface will return EINVAL to 1934 * such requests. These requests could be supported by 1935 * adding a check for lba == 0 and consing up an appropriate 1936 * PMBR. 1937 */ 1938 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1939 return (SET_ERROR(EINVAL)); 1940 1941 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1942 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1943 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1944 1945 if (efi.dki_lba == 1) { 1946 efi_gpt_t gpt = { 0 }; 1947 1948 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1949 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1950 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1951 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1952 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1953 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1954 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1955 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1956 gpt.efi_gpt_SizeOfPartitionEntry = 1957 LE_32(sizeof (efi_gpe_t)); 1958 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1959 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1960 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1961 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1962 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1963 flag)) 1964 return (SET_ERROR(EFAULT)); 1965 ptr += sizeof (gpt); 1966 length -= sizeof (gpt); 1967 } 1968 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1969 length), flag)) 1970 return (SET_ERROR(EFAULT)); 1971 return (0); 1972} 1973 1974/* 1975 * BEGIN entry points to allow external callers access to the volume. 1976 */ 1977/* 1978 * Return the volume parameters needed for access from an external caller. 1979 * These values are invariant as long as the volume is held open. 1980 */ 1981int 1982zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1983 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1984 void **rl_hdl, void **bonus_hdl) 1985{ 1986 zvol_state_t *zv; 1987 1988 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1989 if (zv == NULL) 1990 return (SET_ERROR(ENXIO)); 1991 if (zv->zv_flags & ZVOL_DUMPIFIED) 1992 return (SET_ERROR(ENXIO)); 1993 1994 ASSERT(blksize && max_xfer_len && minor_hdl && 1995 objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 1996 1997 *blksize = zv->zv_volblocksize; 1998 *max_xfer_len = (uint64_t)zvol_maxphys; 1999 *minor_hdl = zv; 2000 *objset_hdl = zv->zv_objset; 2001 *zil_hdl = zv->zv_zilog; 2002 *rl_hdl = &zv->zv_znode; 2003 *bonus_hdl = zv->zv_dbuf; 2004 return (0); 2005} 2006 2007/* 2008 * Return the current volume size to an external caller. 2009 * The size can change while the volume is open. 2010 */ 2011uint64_t 2012zvol_get_volume_size(void *minor_hdl) 2013{ 2014 zvol_state_t *zv = minor_hdl; 2015 2016 return (zv->zv_volsize); 2017} 2018 2019/* 2020 * Return the current WCE setting to an external caller. 2021 * The WCE setting can change while the volume is open. 2022 */ 2023int 2024zvol_get_volume_wce(void *minor_hdl) 2025{ 2026 zvol_state_t *zv = minor_hdl; 2027 2028 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 2029} 2030 2031/* 2032 * Entry point for external callers to zvol_log_write 2033 */ 2034void 2035zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 2036 boolean_t sync) 2037{ 2038 zvol_state_t *zv = minor_hdl; 2039 2040 zvol_log_write(zv, tx, off, resid, sync); 2041} 2042/* 2043 * END entry points to allow external callers access to the volume. 2044 */ 2045#endif /* illumos */ 2046 2047/* 2048 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. 2049 */ 2050static void 2051zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, 2052 boolean_t sync) 2053{ 2054 itx_t *itx; 2055 lr_truncate_t *lr; 2056 zilog_t *zilog = zv->zv_zilog; 2057 2058 if (zil_replaying(zilog, tx)) 2059 return; 2060 2061 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 2062 lr = (lr_truncate_t *)&itx->itx_lr; 2063 lr->lr_foid = ZVOL_OBJ; 2064 lr->lr_offset = off; 2065 lr->lr_length = len; 2066 2067 itx->itx_sync = sync; 2068 zil_itx_assign(zilog, itx, tx); 2069} 2070 2071#ifdef illumos 2072/* 2073 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 2074 * Also a dirtbag dkio ioctl for unmap/free-block functionality. 2075 */ 2076/*ARGSUSED*/ 2077int 2078zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 2079{ 2080 zvol_state_t *zv; 2081 struct dk_callback *dkc; 2082 int error = 0; 2083 rl_t *rl; 2084 2085 mutex_enter(&zfsdev_state_lock); 2086 2087 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 2088 2089 if (zv == NULL) { 2090 mutex_exit(&zfsdev_state_lock); 2091 return (SET_ERROR(ENXIO)); 2092 } 2093 ASSERT(zv->zv_total_opens > 0); 2094 2095 switch (cmd) { 2096 2097 case DKIOCINFO: 2098 { 2099 struct dk_cinfo dki; 2100 2101 bzero(&dki, sizeof (dki)); 2102 (void) strcpy(dki.dki_cname, "zvol"); 2103 (void) strcpy(dki.dki_dname, "zvol"); 2104 dki.dki_ctype = DKC_UNKNOWN; 2105 dki.dki_unit = getminor(dev); 2106 dki.dki_maxtransfer = 2107 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs); 2108 mutex_exit(&zfsdev_state_lock); 2109 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 2110 error = SET_ERROR(EFAULT); 2111 return (error); 2112 } 2113 2114 case DKIOCGMEDIAINFO: 2115 { 2116 struct dk_minfo dkm; 2117 2118 bzero(&dkm, sizeof (dkm)); 2119 dkm.dki_lbsize = 1U << zv->zv_min_bs; 2120 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 2121 dkm.dki_media_type = DK_UNKNOWN; 2122 mutex_exit(&zfsdev_state_lock); 2123 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 2124 error = SET_ERROR(EFAULT); 2125 return (error); 2126 } 2127 2128 case DKIOCGMEDIAINFOEXT: 2129 { 2130 struct dk_minfo_ext dkmext; 2131 2132 bzero(&dkmext, sizeof (dkmext)); 2133 dkmext.dki_lbsize = 1U << zv->zv_min_bs; 2134 dkmext.dki_pbsize = zv->zv_volblocksize; 2135 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 2136 dkmext.dki_media_type = DK_UNKNOWN; 2137 mutex_exit(&zfsdev_state_lock); 2138 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag)) 2139 error = SET_ERROR(EFAULT); 2140 return (error); 2141 } 2142 2143 case DKIOCGETEFI: 2144 { 2145 uint64_t vs = zv->zv_volsize; 2146 uint8_t bs = zv->zv_min_bs; 2147 2148 mutex_exit(&zfsdev_state_lock); 2149 error = zvol_getefi((void *)arg, flag, vs, bs); 2150 return (error); 2151 } 2152 2153 case DKIOCFLUSHWRITECACHE: 2154 dkc = (struct dk_callback *)arg; 2155 mutex_exit(&zfsdev_state_lock); 2156 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2157 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 2158 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 2159 error = 0; 2160 } 2161 return (error); 2162 2163 case DKIOCGETWCE: 2164 { 2165 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 2166 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 2167 flag)) 2168 error = SET_ERROR(EFAULT); 2169 break; 2170 } 2171 case DKIOCSETWCE: 2172 { 2173 int wce; 2174 if (ddi_copyin((void *)arg, &wce, sizeof (int), 2175 flag)) { 2176 error = SET_ERROR(EFAULT); 2177 break; 2178 } 2179 if (wce) { 2180 zv->zv_flags |= ZVOL_WCE; 2181 mutex_exit(&zfsdev_state_lock); 2182 } else { 2183 zv->zv_flags &= ~ZVOL_WCE; 2184 mutex_exit(&zfsdev_state_lock); 2185 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2186 } 2187 return (0); 2188 } 2189 2190 case DKIOCGGEOM: 2191 case DKIOCGVTOC: 2192 /* 2193 * commands using these (like prtvtoc) expect ENOTSUP 2194 * since we're emulating an EFI label 2195 */ 2196 error = SET_ERROR(ENOTSUP); 2197 break; 2198 2199 case DKIOCDUMPINIT: 2200 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 2201 RL_WRITER); 2202 error = zvol_dumpify(zv); 2203 zfs_range_unlock(rl); 2204 break; 2205 2206 case DKIOCDUMPFINI: 2207 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 2208 break; 2209 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 2210 RL_WRITER); 2211 error = zvol_dump_fini(zv); 2212 zfs_range_unlock(rl); 2213 break; 2214 2215 case DKIOCFREE: 2216 { 2217 dkioc_free_t df; 2218 dmu_tx_t *tx; 2219 2220 if (!zvol_unmap_enabled) 2221 break; 2222 2223 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) { 2224 error = SET_ERROR(EFAULT); 2225 break; 2226 } 2227 2228 /* 2229 * Apply Postel's Law to length-checking. If they overshoot, 2230 * just blank out until the end, if there's a need to blank 2231 * out anything. 2232 */ 2233 if (df.df_start >= zv->zv_volsize) 2234 break; /* No need to do anything... */ 2235 2236 mutex_exit(&zfsdev_state_lock); 2237 2238 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length, 2239 RL_WRITER); 2240 tx = dmu_tx_create(zv->zv_objset); 2241 dmu_tx_mark_netfree(tx); 2242 error = dmu_tx_assign(tx, TXG_WAIT); 2243 if (error != 0) { 2244 dmu_tx_abort(tx); 2245 } else { 2246 zvol_log_truncate(zv, tx, df.df_start, 2247 df.df_length, B_TRUE); 2248 dmu_tx_commit(tx); 2249 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 2250 df.df_start, df.df_length); 2251 } 2252 2253 zfs_range_unlock(rl); 2254 2255 if (error == 0) { 2256 /* 2257 * If the write-cache is disabled or 'sync' property 2258 * is set to 'always' then treat this as a synchronous 2259 * operation (i.e. commit to zil). 2260 */ 2261 if (!(zv->zv_flags & ZVOL_WCE) || 2262 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) 2263 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2264 2265 /* 2266 * If the caller really wants synchronous writes, and 2267 * can't wait for them, don't return until the write 2268 * is done. 2269 */ 2270 if (df.df_flags & DF_WAIT_SYNC) { 2271 txg_wait_synced( 2272 dmu_objset_pool(zv->zv_objset), 0); 2273 } 2274 } 2275 return (error); 2276 } 2277 2278 default: 2279 error = SET_ERROR(ENOTTY); 2280 break; 2281 2282 } 2283 mutex_exit(&zfsdev_state_lock); 2284 return (error); 2285} 2286#endif /* illumos */ 2287 2288int 2289zvol_busy(void) 2290{ 2291 return (zvol_minors != 0); 2292} 2293 2294void 2295zvol_init(void) 2296{ 2297 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 2298 1) == 0); 2299#ifdef illumos 2300 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL); 2301#else 2302 ZFS_LOG(1, "ZVOL Initialized."); 2303#endif 2304} 2305 2306void 2307zvol_fini(void) 2308{ 2309#ifdef illumos 2310 mutex_destroy(&zfsdev_state_lock); 2311#endif 2312 ddi_soft_state_fini(&zfsdev_state); 2313 ZFS_LOG(1, "ZVOL Deinitialized."); 2314} 2315 2316#ifdef illumos 2317/*ARGSUSED*/ 2318static int 2319zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx) 2320{ 2321 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 2322 2323 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 2324 return (1); 2325 return (0); 2326} 2327 2328/*ARGSUSED*/ 2329static void 2330zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx) 2331{ 2332 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 2333 2334 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx); 2335} 2336 2337static int 2338zvol_dump_init(zvol_state_t *zv, boolean_t resize) 2339{ 2340 dmu_tx_t *tx; 2341 int error; 2342 objset_t *os = zv->zv_objset; 2343 spa_t *spa = dmu_objset_spa(os); 2344 vdev_t *vd = spa->spa_root_vdev; 2345 nvlist_t *nv = NULL; 2346 uint64_t version = spa_version(spa); 2347 uint64_t checksum, compress, refresrv, vbs, dedup; 2348 2349 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 2350 ASSERT(vd->vdev_ops == &vdev_root_ops); 2351 2352 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 2353 DMU_OBJECT_END); 2354 if (error != 0) 2355 return (error); 2356 /* wait for dmu_free_long_range to actually free the blocks */ 2357 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2358 2359 /* 2360 * If the pool on which the dump device is being initialized has more 2361 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is 2362 * enabled. If so, bump that feature's counter to indicate that the 2363 * feature is active. We also check the vdev type to handle the 2364 * following case: 2365 * # zpool create test raidz disk1 disk2 disk3 2366 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev), 2367 * the raidz vdev itself has 3 children. 2368 */ 2369 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) { 2370 if (!spa_feature_is_enabled(spa, 2371 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 2372 return (SET_ERROR(ENOTSUP)); 2373 (void) dsl_sync_task(spa_name(spa), 2374 zfs_mvdev_dump_feature_check, 2375 zfs_mvdev_dump_activate_feature_sync, NULL, 2376 2, ZFS_SPACE_CHECK_RESERVED); 2377 } 2378 2379 if (!resize) { 2380 error = dsl_prop_get_integer(zv->zv_name, 2381 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 2382 if (error == 0) { 2383 error = dsl_prop_get_integer(zv->zv_name, 2384 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, 2385 NULL); 2386 } 2387 if (error == 0) { 2388 error = dsl_prop_get_integer(zv->zv_name, 2389 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 2390 &refresrv, NULL); 2391 } 2392 if (error == 0) { 2393 error = dsl_prop_get_integer(zv->zv_name, 2394 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, 2395 NULL); 2396 } 2397 if (version >= SPA_VERSION_DEDUP && error == 0) { 2398 error = dsl_prop_get_integer(zv->zv_name, 2399 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 2400 } 2401 } 2402 if (error != 0) 2403 return (error); 2404 2405 tx = dmu_tx_create(os); 2406 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2407 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2408 error = dmu_tx_assign(tx, TXG_WAIT); 2409 if (error != 0) { 2410 dmu_tx_abort(tx); 2411 return (error); 2412 } 2413 2414 /* 2415 * If we are resizing the dump device then we only need to 2416 * update the refreservation to match the newly updated 2417 * zvolsize. Otherwise, we save off the original state of the 2418 * zvol so that we can restore them if the zvol is ever undumpified. 2419 */ 2420 if (resize) { 2421 error = zap_update(os, ZVOL_ZAP_OBJ, 2422 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2423 &zv->zv_volsize, tx); 2424 } else { 2425 error = zap_update(os, ZVOL_ZAP_OBJ, 2426 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 2427 &compress, tx); 2428 if (error == 0) { 2429 error = zap_update(os, ZVOL_ZAP_OBJ, 2430 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, 2431 &checksum, tx); 2432 } 2433 if (error == 0) { 2434 error = zap_update(os, ZVOL_ZAP_OBJ, 2435 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2436 &refresrv, tx); 2437 } 2438 if (error == 0) { 2439 error = zap_update(os, ZVOL_ZAP_OBJ, 2440 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 2441 &vbs, tx); 2442 } 2443 if (error == 0) { 2444 error = dmu_object_set_blocksize( 2445 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx); 2446 } 2447 if (version >= SPA_VERSION_DEDUP && error == 0) { 2448 error = zap_update(os, ZVOL_ZAP_OBJ, 2449 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 2450 &dedup, tx); 2451 } 2452 if (error == 0) 2453 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE; 2454 } 2455 dmu_tx_commit(tx); 2456 2457 /* 2458 * We only need update the zvol's property if we are initializing 2459 * the dump area for the first time. 2460 */ 2461 if (error == 0 && !resize) { 2462 /* 2463 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum 2464 * function. Otherwise, use the old default -- OFF. 2465 */ 2466 checksum = spa_feature_is_active(spa, 2467 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY : 2468 ZIO_CHECKSUM_OFF; 2469 2470 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2471 VERIFY(nvlist_add_uint64(nv, 2472 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 2473 VERIFY(nvlist_add_uint64(nv, 2474 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 2475 ZIO_COMPRESS_OFF) == 0); 2476 VERIFY(nvlist_add_uint64(nv, 2477 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 2478 checksum) == 0); 2479 if (version >= SPA_VERSION_DEDUP) { 2480 VERIFY(nvlist_add_uint64(nv, 2481 zfs_prop_to_name(ZFS_PROP_DEDUP), 2482 ZIO_CHECKSUM_OFF) == 0); 2483 } 2484 2485 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2486 nv, NULL); 2487 nvlist_free(nv); 2488 } 2489 2490 /* Allocate the space for the dump */ 2491 if (error == 0) 2492 error = zvol_prealloc(zv); 2493 return (error); 2494} 2495 2496static int 2497zvol_dumpify(zvol_state_t *zv) 2498{ 2499 int error = 0; 2500 uint64_t dumpsize = 0; 2501 dmu_tx_t *tx; 2502 objset_t *os = zv->zv_objset; 2503 2504 if (zv->zv_flags & ZVOL_RDONLY) 2505 return (SET_ERROR(EROFS)); 2506 2507 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 2508 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 2509 boolean_t resize = (dumpsize > 0); 2510 2511 if ((error = zvol_dump_init(zv, resize)) != 0) { 2512 (void) zvol_dump_fini(zv); 2513 return (error); 2514 } 2515 } 2516 2517 /* 2518 * Build up our lba mapping. 2519 */ 2520 error = zvol_get_lbas(zv); 2521 if (error) { 2522 (void) zvol_dump_fini(zv); 2523 return (error); 2524 } 2525 2526 tx = dmu_tx_create(os); 2527 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2528 error = dmu_tx_assign(tx, TXG_WAIT); 2529 if (error) { 2530 dmu_tx_abort(tx); 2531 (void) zvol_dump_fini(zv); 2532 return (error); 2533 } 2534 2535 zv->zv_flags |= ZVOL_DUMPIFIED; 2536 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 2537 &zv->zv_volsize, tx); 2538 dmu_tx_commit(tx); 2539 2540 if (error) { 2541 (void) zvol_dump_fini(zv); 2542 return (error); 2543 } 2544 2545 txg_wait_synced(dmu_objset_pool(os), 0); 2546 return (0); 2547} 2548 2549static int 2550zvol_dump_fini(zvol_state_t *zv) 2551{ 2552 dmu_tx_t *tx; 2553 objset_t *os = zv->zv_objset; 2554 nvlist_t *nv; 2555 int error = 0; 2556 uint64_t checksum, compress, refresrv, vbs, dedup; 2557 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 2558 2559 /* 2560 * Attempt to restore the zvol back to its pre-dumpified state. 2561 * This is a best-effort attempt as it's possible that not all 2562 * of these properties were initialized during the dumpify process 2563 * (i.e. error during zvol_dump_init). 2564 */ 2565 2566 tx = dmu_tx_create(os); 2567 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2568 error = dmu_tx_assign(tx, TXG_WAIT); 2569 if (error) { 2570 dmu_tx_abort(tx); 2571 return (error); 2572 } 2573 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 2574 dmu_tx_commit(tx); 2575 2576 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2577 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 2578 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2579 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 2580 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2581 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 2582 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2583 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 2584 2585 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2586 (void) nvlist_add_uint64(nv, 2587 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 2588 (void) nvlist_add_uint64(nv, 2589 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 2590 (void) nvlist_add_uint64(nv, 2591 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 2592 if (version >= SPA_VERSION_DEDUP && 2593 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2594 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 2595 (void) nvlist_add_uint64(nv, 2596 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 2597 } 2598 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2599 nv, NULL); 2600 nvlist_free(nv); 2601 2602 zvol_free_extents(zv); 2603 zv->zv_flags &= ~ZVOL_DUMPIFIED; 2604 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 2605 /* wait for dmu_free_long_range to actually free the blocks */ 2606 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2607 tx = dmu_tx_create(os); 2608 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2609 error = dmu_tx_assign(tx, TXG_WAIT); 2610 if (error) { 2611 dmu_tx_abort(tx); 2612 return (error); 2613 } 2614 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 2615 zv->zv_volblocksize = vbs; 2616 dmu_tx_commit(tx); 2617 2618 return (0); 2619} 2620#else /* !illumos */ 2621 2622static void 2623zvol_geom_run(zvol_state_t *zv) 2624{ 2625 struct g_provider *pp; 2626 2627 pp = zv->zv_provider; 2628 g_error_provider(pp, 0); 2629 2630 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0, 2631 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER)); 2632} 2633 2634static void 2635zvol_geom_destroy(zvol_state_t *zv) 2636{ 2637 struct g_provider *pp; 2638 2639 g_topology_assert(); 2640 2641 mtx_lock(&zv->zv_queue_mtx); 2642 zv->zv_state = 1; 2643 wakeup_one(&zv->zv_queue); 2644 while (zv->zv_state != 2) 2645 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0); 2646 mtx_destroy(&zv->zv_queue_mtx); 2647 2648 pp = zv->zv_provider; 2649 zv->zv_provider = NULL; 2650 pp->private = NULL; 2651 g_wither_geom(pp->geom, ENXIO); 2652} 2653 2654static int 2655zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) 2656{ 2657 int count, error, flags; 2658 2659 g_topology_assert(); 2660 2661 /* 2662 * To make it easier we expect either open or close, but not both 2663 * at the same time. 2664 */ 2665 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) || 2666 (acr <= 0 && acw <= 0 && ace <= 0), 2667 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", 2668 pp->name, acr, acw, ace)); 2669 2670 if (pp->private == NULL) { 2671 if (acr <= 0 && acw <= 0 && ace <= 0) 2672 return (0); 2673 return (pp->error); 2674 } 2675 2676 /* 2677 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0, 2678 * because GEOM already handles that and handles it a bit differently. 2679 * GEOM allows for multiple read/exclusive consumers and ZFS allows 2680 * only one exclusive consumer, no matter if it is reader or writer. 2681 * I like better the way GEOM works so I'll leave it for GEOM to 2682 * decide what to do. 2683 */ 2684 2685 count = acr + acw + ace; 2686 if (count == 0) 2687 return (0); 2688 2689 flags = 0; 2690 if (acr != 0 || ace != 0) 2691 flags |= FREAD; 2692 if (acw != 0) 2693 flags |= FWRITE; 2694 2695 g_topology_unlock(); 2696 if (count > 0) 2697 error = zvol_open(pp, flags, count); 2698 else 2699 error = zvol_close(pp, flags, -count); 2700 g_topology_lock(); 2701 return (error); 2702} 2703 2704static void 2705zvol_geom_start(struct bio *bp) 2706{ 2707 zvol_state_t *zv; 2708 boolean_t first; 2709 2710 zv = bp->bio_to->private; 2711 ASSERT(zv != NULL); 2712 switch (bp->bio_cmd) { 2713 case BIO_FLUSH: 2714 if (!THREAD_CAN_SLEEP()) 2715 goto enqueue; 2716 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2717 g_io_deliver(bp, 0); 2718 break; 2719 case BIO_READ: 2720 case BIO_WRITE: 2721 case BIO_DELETE: 2722 if (!THREAD_CAN_SLEEP()) 2723 goto enqueue; 2724 zvol_strategy(bp); 2725 break; 2726 case BIO_GETATTR: { 2727 spa_t *spa = dmu_objset_spa(zv->zv_objset); 2728 uint64_t refd, avail, usedobjs, availobjs, val; 2729 2730 if (g_handleattr_int(bp, "GEOM::candelete", 1)) 2731 return; 2732 if (strcmp(bp->bio_attribute, "blocksavail") == 0) { 2733 dmu_objset_space(zv->zv_objset, &refd, &avail, 2734 &usedobjs, &availobjs); 2735 if (g_handleattr_off_t(bp, "blocksavail", 2736 avail / DEV_BSIZE)) 2737 return; 2738 } else if (strcmp(bp->bio_attribute, "blocksused") == 0) { 2739 dmu_objset_space(zv->zv_objset, &refd, &avail, 2740 &usedobjs, &availobjs); 2741 if (g_handleattr_off_t(bp, "blocksused", 2742 refd / DEV_BSIZE)) 2743 return; 2744 } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) { 2745 avail = metaslab_class_get_space(spa_normal_class(spa)); 2746 avail -= metaslab_class_get_alloc(spa_normal_class(spa)); 2747 if (g_handleattr_off_t(bp, "poolblocksavail", 2748 avail / DEV_BSIZE)) 2749 return; 2750 } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) { 2751 refd = metaslab_class_get_alloc(spa_normal_class(spa)); 2752 if (g_handleattr_off_t(bp, "poolblocksused", 2753 refd / DEV_BSIZE)) 2754 return; 2755 } 2756 /* FALLTHROUGH */ 2757 } 2758 default: 2759 g_io_deliver(bp, EOPNOTSUPP); 2760 break; 2761 } 2762 return; 2763 2764enqueue: 2765 mtx_lock(&zv->zv_queue_mtx); 2766 first = (bioq_first(&zv->zv_queue) == NULL); 2767 bioq_insert_tail(&zv->zv_queue, bp); 2768 mtx_unlock(&zv->zv_queue_mtx); 2769 if (first) 2770 wakeup_one(&zv->zv_queue); 2771} 2772 2773static void 2774zvol_geom_worker(void *arg) 2775{ 2776 zvol_state_t *zv; 2777 struct bio *bp; 2778 2779 thread_lock(curthread); 2780 sched_prio(curthread, PRIBIO); 2781 thread_unlock(curthread); 2782 2783 zv = arg; 2784 for (;;) { 2785 mtx_lock(&zv->zv_queue_mtx); 2786 bp = bioq_takefirst(&zv->zv_queue); 2787 if (bp == NULL) { 2788 if (zv->zv_state == 1) { 2789 zv->zv_state = 2; 2790 wakeup(&zv->zv_state); 2791 mtx_unlock(&zv->zv_queue_mtx); 2792 kthread_exit(); 2793 } 2794 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP, 2795 "zvol:io", 0); 2796 continue; 2797 } 2798 mtx_unlock(&zv->zv_queue_mtx); 2799 switch (bp->bio_cmd) { 2800 case BIO_FLUSH: 2801 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2802 g_io_deliver(bp, 0); 2803 break; 2804 case BIO_READ: 2805 case BIO_WRITE: 2806 case BIO_DELETE: 2807 zvol_strategy(bp); 2808 break; 2809 default: 2810 g_io_deliver(bp, EOPNOTSUPP); 2811 break; 2812 } 2813 } 2814} 2815 2816extern boolean_t dataset_name_hidden(const char *name); 2817 2818static int 2819zvol_create_snapshots(objset_t *os, const char *name) 2820{ 2821 uint64_t cookie, obj; 2822 char *sname; 2823 int error, len; 2824 2825 cookie = obj = 0; 2826 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2827 2828#if 0 2829 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL, 2830 DS_FIND_SNAPSHOTS); 2831#endif 2832 2833 for (;;) { 2834 len = snprintf(sname, MAXPATHLEN, "%s@", name); 2835 if (len >= MAXPATHLEN) { 2836 dmu_objset_rele(os, FTAG); 2837 error = ENAMETOOLONG; 2838 break; 2839 } 2840 2841 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 2842 error = dmu_snapshot_list_next(os, MAXPATHLEN - len, 2843 sname + len, &obj, &cookie, NULL); 2844 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 2845 if (error != 0) { 2846 if (error == ENOENT) 2847 error = 0; 2848 break; 2849 } 2850 2851 error = zvol_create_minor(sname); 2852 if (error != 0 && error != EEXIST) { 2853 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2854 sname, error); 2855 break; 2856 } 2857 } 2858 2859 kmem_free(sname, MAXPATHLEN); 2860 return (error); 2861} 2862 2863int 2864zvol_create_minors(const char *name) 2865{ 2866 uint64_t cookie; 2867 objset_t *os; 2868 char *osname, *p; 2869 int error, len; 2870 2871 if (dataset_name_hidden(name)) 2872 return (0); 2873 2874 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2875 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2876 name, error); 2877 return (error); 2878 } 2879 if (dmu_objset_type(os) == DMU_OST_ZVOL) { 2880 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG); 2881 dsl_pool_rele(dmu_objset_pool(os), FTAG); 2882 error = zvol_create_minor(name); 2883 if (error == 0 || error == EEXIST) { 2884 error = zvol_create_snapshots(os, name); 2885 } else { 2886 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2887 name, error); 2888 } 2889 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG); 2890 dsl_dataset_rele(os->os_dsl_dataset, FTAG); 2891 return (error); 2892 } 2893 if (dmu_objset_type(os) != DMU_OST_ZFS) { 2894 dmu_objset_rele(os, FTAG); 2895 return (0); 2896 } 2897 2898 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2899 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) { 2900 dmu_objset_rele(os, FTAG); 2901 kmem_free(osname, MAXPATHLEN); 2902 return (ENOENT); 2903 } 2904 p = osname + strlen(osname); 2905 len = MAXPATHLEN - (p - osname); 2906 2907#if 0 2908 /* Prefetch the datasets. */ 2909 cookie = 0; 2910 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) { 2911 if (!dataset_name_hidden(osname)) 2912 (void) dmu_objset_prefetch(osname, NULL); 2913 } 2914#endif 2915 2916 cookie = 0; 2917 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL, 2918 &cookie) == 0) { 2919 dmu_objset_rele(os, FTAG); 2920 (void)zvol_create_minors(osname); 2921 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2922 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2923 name, error); 2924 return (error); 2925 } 2926 } 2927 2928 dmu_objset_rele(os, FTAG); 2929 kmem_free(osname, MAXPATHLEN); 2930 return (0); 2931} 2932 2933static void 2934zvol_rename_minor(zvol_state_t *zv, const char *newname) 2935{ 2936 struct g_geom *gp; 2937 struct g_provider *pp; 2938 struct cdev *dev; 2939 2940 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 2941 2942 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 2943 g_topology_lock(); 2944 pp = zv->zv_provider; 2945 ASSERT(pp != NULL); 2946 gp = pp->geom; 2947 ASSERT(gp != NULL); 2948 2949 zv->zv_provider = NULL; 2950 g_wither_provider(pp, ENXIO); 2951 2952 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname); 2953 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 2954 pp->sectorsize = DEV_BSIZE; 2955 pp->mediasize = zv->zv_volsize; 2956 pp->private = zv; 2957 zv->zv_provider = pp; 2958 g_error_provider(pp, 0); 2959 g_topology_unlock(); 2960 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { 2961 struct make_dev_args args; 2962 2963 dev = zv->zv_dev; 2964 ASSERT(dev != NULL); 2965 zv->zv_dev = NULL; 2966 destroy_dev(dev); 2967 2968 make_dev_args_init(&args); 2969 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK; 2970 args.mda_devsw = &zvol_cdevsw; 2971 args.mda_cr = NULL; 2972 args.mda_uid = UID_ROOT; 2973 args.mda_gid = GID_OPERATOR; 2974 args.mda_mode = 0640; 2975 args.mda_si_drv2 = zv; 2976 if (make_dev_s(&args, &zv->zv_dev, 2977 "%s/%s", ZVOL_DRIVER, newname) == 0) 2978 zv->zv_dev->si_iosize_max = MAXPHYS; 2979 } 2980 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name)); 2981} 2982 2983void 2984zvol_rename_minors(const char *oldname, const char *newname) 2985{ 2986 char name[MAXPATHLEN]; 2987 struct g_provider *pp; 2988 struct g_geom *gp; 2989 size_t oldnamelen, newnamelen; 2990 zvol_state_t *zv; 2991 char *namebuf; 2992 boolean_t locked = B_FALSE; 2993 2994 oldnamelen = strlen(oldname); 2995 newnamelen = strlen(newname); 2996 2997 DROP_GIANT(); 2998 /* See comment in zvol_open(). */ 2999 if (!MUTEX_HELD(&zfsdev_state_lock)) { 3000 mutex_enter(&zfsdev_state_lock); 3001 locked = B_TRUE; 3002 } 3003 3004 LIST_FOREACH(zv, &all_zvols, zv_links) { 3005 if (strcmp(zv->zv_name, oldname) == 0) { 3006 zvol_rename_minor(zv, newname); 3007 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && 3008 (zv->zv_name[oldnamelen] == '/' || 3009 zv->zv_name[oldnamelen] == '@')) { 3010 snprintf(name, sizeof(name), "%s%c%s", newname, 3011 zv->zv_name[oldnamelen], 3012 zv->zv_name + oldnamelen + 1); 3013 zvol_rename_minor(zv, name); 3014 } 3015 } 3016 3017 if (locked) 3018 mutex_exit(&zfsdev_state_lock); 3019 PICKUP_GIANT(); 3020} 3021 3022static int 3023zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td) 3024{ 3025 zvol_state_t *zv = dev->si_drv2; 3026 int err = 0; 3027 3028 mutex_enter(&zfsdev_state_lock); 3029 if (zv->zv_total_opens == 0) 3030 err = zvol_first_open(zv); 3031 if (err) { 3032 mutex_exit(&zfsdev_state_lock); 3033 return (err); 3034 } 3035 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 3036 err = SET_ERROR(EROFS); 3037 goto out; 3038 } 3039 if (zv->zv_flags & ZVOL_EXCL) { 3040 err = SET_ERROR(EBUSY); 3041 goto out; 3042 } 3043#ifdef FEXCL 3044 if (flags & FEXCL) { 3045 if (zv->zv_total_opens != 0) { 3046 err = SET_ERROR(EBUSY); 3047 goto out; 3048 } 3049 zv->zv_flags |= ZVOL_EXCL; 3050 } 3051#endif 3052 3053 zv->zv_total_opens++; 3054 mutex_exit(&zfsdev_state_lock); 3055 return (err); 3056out: 3057 if (zv->zv_total_opens == 0) 3058 zvol_last_close(zv); 3059 mutex_exit(&zfsdev_state_lock); 3060 return (err); 3061} 3062 3063static int 3064zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td) 3065{ 3066 zvol_state_t *zv = dev->si_drv2; 3067 3068 mutex_enter(&zfsdev_state_lock); 3069 if (zv->zv_flags & ZVOL_EXCL) { 3070 ASSERT(zv->zv_total_opens == 1); 3071 zv->zv_flags &= ~ZVOL_EXCL; 3072 } 3073 3074 /* 3075 * If the open count is zero, this is a spurious close. 3076 * That indicates a bug in the kernel / DDI framework. 3077 */ 3078 ASSERT(zv->zv_total_opens != 0); 3079 3080 /* 3081 * You may get multiple opens, but only one close. 3082 */ 3083 zv->zv_total_opens--; 3084 3085 if (zv->zv_total_opens == 0) 3086 zvol_last_close(zv); 3087 3088 mutex_exit(&zfsdev_state_lock); 3089 return (0); 3090} 3091 3092static int 3093zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 3094{ 3095 zvol_state_t *zv; 3096 rl_t *rl; 3097 off_t offset, length, chunk; 3098 int i, error; 3099 u_int u; 3100 3101 zv = dev->si_drv2; 3102 3103 error = 0; 3104 KASSERT(zv->zv_total_opens > 0, 3105 ("Device with zero access count in zvol_d_ioctl")); 3106 3107 i = IOCPARM_LEN(cmd); 3108 switch (cmd) { 3109 case DIOCGSECTORSIZE: 3110 *(u_int *)data = DEV_BSIZE; 3111 break; 3112 case DIOCGMEDIASIZE: 3113 *(off_t *)data = zv->zv_volsize; 3114 break; 3115 case DIOCGFLUSH: 3116 zil_commit(zv->zv_zilog, ZVOL_OBJ); 3117 break; 3118 case DIOCGDELETE: 3119 if (!zvol_unmap_enabled) 3120 break; 3121 3122 offset = ((off_t *)data)[0]; 3123 length = ((off_t *)data)[1]; 3124 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 || 3125 offset < 0 || offset >= zv->zv_volsize || 3126 length <= 0) { 3127 printf("%s: offset=%jd length=%jd\n", __func__, offset, 3128 length); 3129 error = EINVAL; 3130 break; 3131 } 3132 3133 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER); 3134 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 3135 error = dmu_tx_assign(tx, TXG_WAIT); 3136 if (error != 0) { 3137 dmu_tx_abort(tx); 3138 } else { 3139 zvol_log_truncate(zv, tx, offset, length, B_TRUE); 3140 dmu_tx_commit(tx); 3141 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 3142 offset, length); 3143 } 3144 zfs_range_unlock(rl); 3145 if (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) 3146 zil_commit(zv->zv_zilog, ZVOL_OBJ); 3147 break; 3148 case DIOCGSTRIPESIZE: 3149 *(off_t *)data = zv->zv_volblocksize; 3150 break; 3151 case DIOCGSTRIPEOFFSET: 3152 *(off_t *)data = 0; 3153 break; 3154 case DIOCGATTR: { 3155 spa_t *spa = dmu_objset_spa(zv->zv_objset); 3156 struct diocgattr_arg *arg = (struct diocgattr_arg *)data; 3157 uint64_t refd, avail, usedobjs, availobjs; 3158 3159 if (strcmp(arg->name, "GEOM::candelete") == 0) 3160 arg->value.i = 1; 3161 else if (strcmp(arg->name, "blocksavail") == 0) { 3162 dmu_objset_space(zv->zv_objset, &refd, &avail, 3163 &usedobjs, &availobjs); 3164 arg->value.off = avail / DEV_BSIZE; 3165 } else if (strcmp(arg->name, "blocksused") == 0) { 3166 dmu_objset_space(zv->zv_objset, &refd, &avail, 3167 &usedobjs, &availobjs); 3168 arg->value.off = refd / DEV_BSIZE; 3169 } else if (strcmp(arg->name, "poolblocksavail") == 0) { 3170 avail = metaslab_class_get_space(spa_normal_class(spa)); 3171 avail -= metaslab_class_get_alloc(spa_normal_class(spa)); 3172 arg->value.off = avail / DEV_BSIZE; 3173 } else if (strcmp(arg->name, "poolblocksused") == 0) { 3174 refd = metaslab_class_get_alloc(spa_normal_class(spa)); 3175 arg->value.off = refd / DEV_BSIZE; 3176 } else 3177 error = ENOIOCTL; 3178 break; 3179 } 3180 case FIOSEEKHOLE: 3181 case FIOSEEKDATA: { 3182 off_t *off = (off_t *)data; 3183 uint64_t noff; 3184 boolean_t hole; 3185 3186 hole = (cmd == FIOSEEKHOLE); 3187 noff = *off; 3188 error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff); 3189 *off = noff; 3190 break; 3191 } 3192 default: 3193 error = ENOIOCTL; 3194 } 3195 3196 return (error); 3197} 3198#endif /* illumos */ 3199