zvol.c revision 308448
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * 24 * Copyright (c) 2006-2010 Pawel Jakub Dawidek <pjd@FreeBSD.org> 25 * All rights reserved. 26 * 27 * Portions Copyright 2010 Robert Milkowski 28 * 29 * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 30 * Copyright (c) 2012, 2014 by Delphix. All rights reserved. 31 * Copyright (c) 2013, Joyent, Inc. All rights reserved. 32 * Copyright (c) 2014 Integros [integros.com] 33 */ 34 35/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 36 37/* 38 * ZFS volume emulation driver. 39 * 40 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 41 * Volumes are accessed through the symbolic links named: 42 * 43 * /dev/zvol/dsk/<pool_name>/<dataset_name> 44 * /dev/zvol/rdsk/<pool_name>/<dataset_name> 45 * 46 * These links are created by the /dev filesystem (sdev_zvolops.c). 47 * Volumes are persistent through reboot. No user command needs to be 48 * run before opening and using a device. 49 * 50 * FreeBSD notes. 51 * On FreeBSD ZVOLs are simply GEOM providers like any other storage device 52 * in the system. 53 */ 54 55#include <sys/types.h> 56#include <sys/param.h> 57#include <sys/kernel.h> 58#include <sys/errno.h> 59#include <sys/uio.h> 60#include <sys/bio.h> 61#include <sys/buf.h> 62#include <sys/kmem.h> 63#include <sys/conf.h> 64#include <sys/cmn_err.h> 65#include <sys/stat.h> 66#include <sys/zap.h> 67#include <sys/spa.h> 68#include <sys/spa_impl.h> 69#include <sys/zio.h> 70#include <sys/disk.h> 71#include <sys/dmu_traverse.h> 72#include <sys/dnode.h> 73#include <sys/dsl_dataset.h> 74#include <sys/dsl_prop.h> 75#include <sys/dkio.h> 76#include <sys/byteorder.h> 77#include <sys/sunddi.h> 78#include <sys/dirent.h> 79#include <sys/policy.h> 80#include <sys/queue.h> 81#include <sys/fs/zfs.h> 82#include <sys/zfs_ioctl.h> 83#include <sys/zil.h> 84#include <sys/refcount.h> 85#include <sys/zfs_znode.h> 86#include <sys/zfs_rlock.h> 87#include <sys/vdev_impl.h> 88#include <sys/vdev_raidz.h> 89#include <sys/zvol.h> 90#include <sys/zil_impl.h> 91#include <sys/dbuf.h> 92#include <sys/dmu_tx.h> 93#include <sys/zfeature.h> 94#include <sys/zio_checksum.h> 95#include <sys/filio.h> 96 97#include <geom/geom.h> 98 99#include "zfs_namecheck.h" 100 101#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 if (zv->zv_dev != NULL) 785 destroy_dev(zv->zv_dev); 786 } 787#endif 788 789 avl_destroy(&zv->zv_znode.z_range_avl); 790 mutex_destroy(&zv->zv_znode.z_range_lock); 791 792 kmem_free(zv, sizeof (zvol_state_t)); 793#ifdef illumos 794 ddi_soft_state_free(zfsdev_state, minor); 795#endif 796 zvol_minors--; 797 return (0); 798} 799 800int 801zvol_remove_minor(const char *name) 802{ 803 zvol_state_t *zv; 804 int rc; 805 806 mutex_enter(&zfsdev_state_lock); 807 if ((zv = zvol_minor_lookup(name)) == NULL) { 808 mutex_exit(&zfsdev_state_lock); 809 return (SET_ERROR(ENXIO)); 810 } 811 rc = zvol_remove_zv(zv); 812 mutex_exit(&zfsdev_state_lock); 813 return (rc); 814} 815 816int 817zvol_first_open(zvol_state_t *zv) 818{ 819 objset_t *os; 820 uint64_t volsize; 821 int error; 822 uint64_t readonly; 823 824 /* lie and say we're read-only */ 825 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 826 zvol_tag, &os); 827 if (error) 828 return (error); 829 830 zv->zv_objset = os; 831 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 832 if (error) { 833 ASSERT(error == 0); 834 dmu_objset_disown(os, zvol_tag); 835 return (error); 836 } 837 838 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 839 if (error) { 840 dmu_objset_disown(os, zvol_tag); 841 return (error); 842 } 843 844 zvol_size_changed(zv, volsize); 845 zv->zv_zilog = zil_open(os, zvol_get_data); 846 847 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 848 NULL) == 0); 849 if (readonly || dmu_objset_is_snapshot(os) || 850 !spa_writeable(dmu_objset_spa(os))) 851 zv->zv_flags |= ZVOL_RDONLY; 852 else 853 zv->zv_flags &= ~ZVOL_RDONLY; 854 return (error); 855} 856 857void 858zvol_last_close(zvol_state_t *zv) 859{ 860 zil_close(zv->zv_zilog); 861 zv->zv_zilog = NULL; 862 863 dmu_buf_rele(zv->zv_dbuf, zvol_tag); 864 zv->zv_dbuf = NULL; 865 866 /* 867 * Evict cached data 868 */ 869 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && 870 !(zv->zv_flags & ZVOL_RDONLY)) 871 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 872 dmu_objset_evict_dbufs(zv->zv_objset); 873 874 dmu_objset_disown(zv->zv_objset, zvol_tag); 875 zv->zv_objset = NULL; 876} 877 878#ifdef illumos 879int 880zvol_prealloc(zvol_state_t *zv) 881{ 882 objset_t *os = zv->zv_objset; 883 dmu_tx_t *tx; 884 uint64_t refd, avail, usedobjs, availobjs; 885 uint64_t resid = zv->zv_volsize; 886 uint64_t off = 0; 887 888 /* Check the space usage before attempting to allocate the space */ 889 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 890 if (avail < zv->zv_volsize) 891 return (SET_ERROR(ENOSPC)); 892 893 /* Free old extents if they exist */ 894 zvol_free_extents(zv); 895 896 while (resid != 0) { 897 int error; 898 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE); 899 900 tx = dmu_tx_create(os); 901 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 902 error = dmu_tx_assign(tx, TXG_WAIT); 903 if (error) { 904 dmu_tx_abort(tx); 905 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 906 return (error); 907 } 908 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 909 dmu_tx_commit(tx); 910 off += bytes; 911 resid -= bytes; 912 } 913 txg_wait_synced(dmu_objset_pool(os), 0); 914 915 return (0); 916} 917#endif /* illumos */ 918 919static int 920zvol_update_volsize(objset_t *os, uint64_t volsize) 921{ 922 dmu_tx_t *tx; 923 int error; 924 925 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 926 927 tx = dmu_tx_create(os); 928 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 929 dmu_tx_mark_netfree(tx); 930 error = dmu_tx_assign(tx, TXG_WAIT); 931 if (error) { 932 dmu_tx_abort(tx); 933 return (error); 934 } 935 936 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 937 &volsize, tx); 938 dmu_tx_commit(tx); 939 940 if (error == 0) 941 error = dmu_free_long_range(os, 942 ZVOL_OBJ, volsize, DMU_OBJECT_END); 943 return (error); 944} 945 946void 947zvol_remove_minors(const char *name) 948{ 949#ifdef illumos 950 zvol_state_t *zv; 951 char *namebuf; 952 minor_t minor; 953 954 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP); 955 (void) strncpy(namebuf, name, strlen(name)); 956 (void) strcat(namebuf, "/"); 957 mutex_enter(&zfsdev_state_lock); 958 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) { 959 960 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 961 if (zv == NULL) 962 continue; 963 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0) 964 (void) zvol_remove_zv(zv); 965 } 966 kmem_free(namebuf, strlen(name) + 2); 967 968 mutex_exit(&zfsdev_state_lock); 969#else /* !illumos */ 970 zvol_state_t *zv, *tzv; 971 size_t namelen; 972 973 namelen = strlen(name); 974 975 DROP_GIANT(); 976 mutex_enter(&zfsdev_state_lock); 977 978 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) { 979 if (strcmp(zv->zv_name, name) == 0 || 980 (strncmp(zv->zv_name, name, namelen) == 0 && 981 strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' || 982 zv->zv_name[namelen] == '@'))) { 983 (void) zvol_remove_zv(zv); 984 } 985 } 986 987 mutex_exit(&zfsdev_state_lock); 988 PICKUP_GIANT(); 989#endif /* illumos */ 990} 991 992static int 993zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize) 994{ 995 uint64_t old_volsize = 0ULL; 996 int error = 0; 997 998 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 999 1000 /* 1001 * Reinitialize the dump area to the new size. If we 1002 * failed to resize the dump area then restore it back to 1003 * its original size. We must set the new volsize prior 1004 * to calling dumpvp_resize() to ensure that the devices' 1005 * size(9P) is not visible by the dump subsystem. 1006 */ 1007 old_volsize = zv->zv_volsize; 1008 zvol_size_changed(zv, volsize); 1009 1010#ifdef ZVOL_DUMP 1011 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1012 if ((error = zvol_dumpify(zv)) != 0 || 1013 (error = dumpvp_resize()) != 0) { 1014 int dumpify_error; 1015 1016 (void) zvol_update_volsize(zv->zv_objset, old_volsize); 1017 zvol_size_changed(zv, old_volsize); 1018 dumpify_error = zvol_dumpify(zv); 1019 error = dumpify_error ? dumpify_error : error; 1020 } 1021 } 1022#endif /* ZVOL_DUMP */ 1023 1024#ifdef illumos 1025 /* 1026 * Generate a LUN expansion event. 1027 */ 1028 if (error == 0) { 1029 sysevent_id_t eid; 1030 nvlist_t *attr; 1031 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 1032 1033 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 1034 zv->zv_minor); 1035 1036 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1037 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 1038 1039 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 1040 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 1041 1042 nvlist_free(attr); 1043 kmem_free(physpath, MAXPATHLEN); 1044 } 1045#endif /* illumos */ 1046 return (error); 1047} 1048 1049int 1050zvol_set_volsize(const char *name, uint64_t volsize) 1051{ 1052 zvol_state_t *zv = NULL; 1053 objset_t *os; 1054 int error; 1055 dmu_object_info_t doi; 1056 uint64_t readonly; 1057 boolean_t owned = B_FALSE; 1058 1059 error = dsl_prop_get_integer(name, 1060 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL); 1061 if (error != 0) 1062 return (error); 1063 if (readonly) 1064 return (SET_ERROR(EROFS)); 1065 1066 mutex_enter(&zfsdev_state_lock); 1067 zv = zvol_minor_lookup(name); 1068 1069 if (zv == NULL || zv->zv_objset == NULL) { 1070 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, 1071 FTAG, &os)) != 0) { 1072 mutex_exit(&zfsdev_state_lock); 1073 return (error); 1074 } 1075 owned = B_TRUE; 1076 if (zv != NULL) 1077 zv->zv_objset = os; 1078 } else { 1079 os = zv->zv_objset; 1080 } 1081 1082 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 1083 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0) 1084 goto out; 1085 1086 error = zvol_update_volsize(os, volsize); 1087 1088 if (error == 0 && zv != NULL) 1089 error = zvol_update_live_volsize(zv, volsize); 1090out: 1091 if (owned) { 1092 dmu_objset_disown(os, FTAG); 1093 if (zv != NULL) 1094 zv->zv_objset = NULL; 1095 } 1096 mutex_exit(&zfsdev_state_lock); 1097 return (error); 1098} 1099 1100/*ARGSUSED*/ 1101#ifdef illumos 1102int 1103zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr) 1104#else 1105static int 1106zvol_open(struct g_provider *pp, int flag, int count) 1107#endif 1108{ 1109 zvol_state_t *zv; 1110 int err = 0; 1111#ifdef illumos 1112 1113 mutex_enter(&zfsdev_state_lock); 1114 1115 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL); 1116 if (zv == NULL) { 1117 mutex_exit(&zfsdev_state_lock); 1118 return (SET_ERROR(ENXIO)); 1119 } 1120 1121 if (zv->zv_total_opens == 0) 1122 err = zvol_first_open(zv); 1123 if (err) { 1124 mutex_exit(&zfsdev_state_lock); 1125 return (err); 1126 } 1127#else /* !illumos */ 1128 if (tsd_get(zfs_geom_probe_vdev_key) != NULL) { 1129 /* 1130 * if zfs_geom_probe_vdev_key is set, that means that zfs is 1131 * attempting to probe geom providers while looking for a 1132 * replacement for a missing VDEV. In this case, the 1133 * spa_namespace_lock will not be held, but it is still illegal 1134 * to use a zvol as a vdev. Deadlocks can result if another 1135 * thread has spa_namespace_lock 1136 */ 1137 return (EOPNOTSUPP); 1138 } 1139 1140 mutex_enter(&zfsdev_state_lock); 1141 1142 zv = pp->private; 1143 if (zv == NULL) { 1144 mutex_exit(&zfsdev_state_lock); 1145 return (SET_ERROR(ENXIO)); 1146 } 1147 1148 if (zv->zv_total_opens == 0) { 1149 err = zvol_first_open(zv); 1150 if (err) { 1151 mutex_exit(&zfsdev_state_lock); 1152 return (err); 1153 } 1154 pp->mediasize = zv->zv_volsize; 1155 pp->stripeoffset = 0; 1156 pp->stripesize = zv->zv_volblocksize; 1157 } 1158#endif /* illumos */ 1159 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 1160 err = SET_ERROR(EROFS); 1161 goto out; 1162 } 1163 if (zv->zv_flags & ZVOL_EXCL) { 1164 err = SET_ERROR(EBUSY); 1165 goto out; 1166 } 1167#ifdef FEXCL 1168 if (flag & FEXCL) { 1169 if (zv->zv_total_opens != 0) { 1170 err = SET_ERROR(EBUSY); 1171 goto out; 1172 } 1173 zv->zv_flags |= ZVOL_EXCL; 1174 } 1175#endif 1176 1177#ifdef illumos 1178 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) { 1179 zv->zv_open_count[otyp]++; 1180 zv->zv_total_opens++; 1181 } 1182 mutex_exit(&zfsdev_state_lock); 1183#else 1184 zv->zv_total_opens += count; 1185 mutex_exit(&zfsdev_state_lock); 1186#endif 1187 1188 return (err); 1189out: 1190 if (zv->zv_total_opens == 0) 1191 zvol_last_close(zv); 1192#ifdef illumos 1193 mutex_exit(&zfsdev_state_lock); 1194#else 1195 mutex_exit(&zfsdev_state_lock); 1196#endif 1197 return (err); 1198} 1199 1200/*ARGSUSED*/ 1201#ifdef illumos 1202int 1203zvol_close(dev_t dev, int flag, int otyp, cred_t *cr) 1204{ 1205 minor_t minor = getminor(dev); 1206 zvol_state_t *zv; 1207 int error = 0; 1208 1209 mutex_enter(&zfsdev_state_lock); 1210 1211 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1212 if (zv == NULL) { 1213 mutex_exit(&zfsdev_state_lock); 1214#else /* !illumos */ 1215static int 1216zvol_close(struct g_provider *pp, int flag, int count) 1217{ 1218 zvol_state_t *zv; 1219 int error = 0; 1220 boolean_t locked = B_FALSE; 1221 1222 /* See comment in zvol_open(). */ 1223 if (!MUTEX_HELD(&zfsdev_state_lock)) { 1224 mutex_enter(&zfsdev_state_lock); 1225 locked = B_TRUE; 1226 } 1227 1228 zv = pp->private; 1229 if (zv == NULL) { 1230 if (locked) 1231 mutex_exit(&zfsdev_state_lock); 1232#endif /* illumos */ 1233 return (SET_ERROR(ENXIO)); 1234 } 1235 1236 if (zv->zv_flags & ZVOL_EXCL) { 1237 ASSERT(zv->zv_total_opens == 1); 1238 zv->zv_flags &= ~ZVOL_EXCL; 1239 } 1240 1241 /* 1242 * If the open count is zero, this is a spurious close. 1243 * That indicates a bug in the kernel / DDI framework. 1244 */ 1245#ifdef illumos 1246 ASSERT(zv->zv_open_count[otyp] != 0); 1247#endif 1248 ASSERT(zv->zv_total_opens != 0); 1249 1250 /* 1251 * You may get multiple opens, but only one close. 1252 */ 1253#ifdef illumos 1254 zv->zv_open_count[otyp]--; 1255 zv->zv_total_opens--; 1256#else 1257 zv->zv_total_opens -= count; 1258#endif 1259 1260 if (zv->zv_total_opens == 0) 1261 zvol_last_close(zv); 1262 1263#ifdef illumos 1264 mutex_exit(&zfsdev_state_lock); 1265#else 1266 if (locked) 1267 mutex_exit(&zfsdev_state_lock); 1268#endif 1269 return (error); 1270} 1271 1272static void 1273zvol_get_done(zgd_t *zgd, int error) 1274{ 1275 if (zgd->zgd_db) 1276 dmu_buf_rele(zgd->zgd_db, zgd); 1277 1278 zfs_range_unlock(zgd->zgd_rl); 1279 1280 if (error == 0 && zgd->zgd_bp) 1281 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1282 1283 kmem_free(zgd, sizeof (zgd_t)); 1284} 1285 1286/* 1287 * Get data to generate a TX_WRITE intent log record. 1288 */ 1289static int 1290zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1291{ 1292 zvol_state_t *zv = arg; 1293 objset_t *os = zv->zv_objset; 1294 uint64_t object = ZVOL_OBJ; 1295 uint64_t offset = lr->lr_offset; 1296 uint64_t size = lr->lr_length; /* length of user data */ 1297 blkptr_t *bp = &lr->lr_blkptr; 1298 dmu_buf_t *db; 1299 zgd_t *zgd; 1300 int error; 1301 1302 ASSERT(zio != NULL); 1303 ASSERT(size != 0); 1304 1305 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1306 zgd->zgd_zilog = zv->zv_zilog; 1307 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 1308 1309 /* 1310 * Write records come in two flavors: immediate and indirect. 1311 * For small writes it's cheaper to store the data with the 1312 * log record (immediate); for large writes it's cheaper to 1313 * sync the data and get a pointer to it (indirect) so that 1314 * we don't have to write the data twice. 1315 */ 1316 if (buf != NULL) { /* immediate write */ 1317 error = dmu_read(os, object, offset, size, buf, 1318 DMU_READ_NO_PREFETCH); 1319 } else { 1320 size = zv->zv_volblocksize; 1321 offset = P2ALIGN(offset, size); 1322 error = dmu_buf_hold(os, object, offset, zgd, &db, 1323 DMU_READ_NO_PREFETCH); 1324 if (error == 0) { 1325 blkptr_t *obp = dmu_buf_get_blkptr(db); 1326 if (obp) { 1327 ASSERT(BP_IS_HOLE(bp)); 1328 *bp = *obp; 1329 } 1330 1331 zgd->zgd_db = db; 1332 zgd->zgd_bp = bp; 1333 1334 ASSERT(db->db_offset == offset); 1335 ASSERT(db->db_size == size); 1336 1337 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1338 zvol_get_done, zgd); 1339 1340 if (error == 0) 1341 return (0); 1342 } 1343 } 1344 1345 zvol_get_done(zgd, error); 1346 1347 return (error); 1348} 1349 1350/* 1351 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 1352 * 1353 * We store data in the log buffers if it's small enough. 1354 * Otherwise we will later flush the data out via dmu_sync(). 1355 */ 1356ssize_t zvol_immediate_write_sz = 32768; 1357 1358static void 1359zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1360 boolean_t sync) 1361{ 1362 uint32_t blocksize = zv->zv_volblocksize; 1363 zilog_t *zilog = zv->zv_zilog; 1364 boolean_t slogging; 1365 ssize_t immediate_write_sz; 1366 1367 if (zil_replaying(zilog, tx)) 1368 return; 1369 1370 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1371 ? 0 : zvol_immediate_write_sz; 1372 1373 slogging = spa_has_slogs(zilog->zl_spa) && 1374 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 1375 1376 while (resid) { 1377 itx_t *itx; 1378 lr_write_t *lr; 1379 ssize_t len; 1380 itx_wr_state_t write_state; 1381 1382 /* 1383 * Unlike zfs_log_write() we can be called with 1384 * upto DMU_MAX_ACCESS/2 (5MB) writes. 1385 */ 1386 if (blocksize > immediate_write_sz && !slogging && 1387 resid >= blocksize && off % blocksize == 0) { 1388 write_state = WR_INDIRECT; /* uses dmu_sync */ 1389 len = blocksize; 1390 } else if (sync) { 1391 write_state = WR_COPIED; 1392 len = MIN(ZIL_MAX_LOG_DATA, resid); 1393 } else { 1394 write_state = WR_NEED_COPY; 1395 len = MIN(ZIL_MAX_LOG_DATA, resid); 1396 } 1397 1398 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1399 (write_state == WR_COPIED ? len : 0)); 1400 lr = (lr_write_t *)&itx->itx_lr; 1401 if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 1402 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1403 zil_itx_destroy(itx); 1404 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1405 lr = (lr_write_t *)&itx->itx_lr; 1406 write_state = WR_NEED_COPY; 1407 } 1408 1409 itx->itx_wr_state = write_state; 1410 if (write_state == WR_NEED_COPY) 1411 itx->itx_sod += len; 1412 lr->lr_foid = ZVOL_OBJ; 1413 lr->lr_offset = off; 1414 lr->lr_length = len; 1415 lr->lr_blkoff = 0; 1416 BP_ZERO(&lr->lr_blkptr); 1417 1418 itx->itx_private = zv; 1419 itx->itx_sync = sync; 1420 1421 zil_itx_assign(zilog, itx, tx); 1422 1423 off += len; 1424 resid -= len; 1425 } 1426} 1427 1428#ifdef illumos 1429static int 1430zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset, 1431 uint64_t size, boolean_t doread, boolean_t isdump) 1432{ 1433 vdev_disk_t *dvd; 1434 int c; 1435 int numerrors = 0; 1436 1437 if (vd->vdev_ops == &vdev_mirror_ops || 1438 vd->vdev_ops == &vdev_replacing_ops || 1439 vd->vdev_ops == &vdev_spare_ops) { 1440 for (c = 0; c < vd->vdev_children; c++) { 1441 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1442 addr, offset, origoffset, size, doread, isdump); 1443 if (err != 0) { 1444 numerrors++; 1445 } else if (doread) { 1446 break; 1447 } 1448 } 1449 } 1450 1451 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops) 1452 return (numerrors < vd->vdev_children ? 0 : EIO); 1453 1454 if (doread && !vdev_readable(vd)) 1455 return (SET_ERROR(EIO)); 1456 else if (!doread && !vdev_writeable(vd)) 1457 return (SET_ERROR(EIO)); 1458 1459 if (vd->vdev_ops == &vdev_raidz_ops) { 1460 return (vdev_raidz_physio(vd, 1461 addr, size, offset, origoffset, doread, isdump)); 1462 } 1463 1464 offset += VDEV_LABEL_START_SIZE; 1465 1466 if (ddi_in_panic() || isdump) { 1467 ASSERT(!doread); 1468 if (doread) 1469 return (SET_ERROR(EIO)); 1470 dvd = vd->vdev_tsd; 1471 ASSERT3P(dvd, !=, NULL); 1472 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1473 lbtodb(size))); 1474 } else { 1475 dvd = vd->vdev_tsd; 1476 ASSERT3P(dvd, !=, NULL); 1477 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size, 1478 offset, doread ? B_READ : B_WRITE)); 1479 } 1480} 1481 1482static int 1483zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1484 boolean_t doread, boolean_t isdump) 1485{ 1486 vdev_t *vd; 1487 int error; 1488 zvol_extent_t *ze; 1489 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1490 1491 /* Must be sector aligned, and not stradle a block boundary. */ 1492 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1493 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1494 return (SET_ERROR(EINVAL)); 1495 } 1496 ASSERT(size <= zv->zv_volblocksize); 1497 1498 /* Locate the extent this belongs to */ 1499 ze = list_head(&zv->zv_extents); 1500 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1501 offset -= ze->ze_nblks * zv->zv_volblocksize; 1502 ze = list_next(&zv->zv_extents, ze); 1503 } 1504 1505 if (ze == NULL) 1506 return (SET_ERROR(EINVAL)); 1507 1508 if (!ddi_in_panic()) 1509 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1510 1511 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1512 offset += DVA_GET_OFFSET(&ze->ze_dva); 1513 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva), 1514 size, doread, isdump); 1515 1516 if (!ddi_in_panic()) 1517 spa_config_exit(spa, SCL_STATE, FTAG); 1518 1519 return (error); 1520} 1521 1522int 1523zvol_strategy(buf_t *bp) 1524{ 1525 zfs_soft_state_t *zs = NULL; 1526#else /* !illumos */ 1527void 1528zvol_strategy(struct bio *bp) 1529{ 1530#endif /* illumos */ 1531 zvol_state_t *zv; 1532 uint64_t off, volsize; 1533 size_t resid; 1534 char *addr; 1535 objset_t *os; 1536 rl_t *rl; 1537 int error = 0; 1538#ifdef illumos 1539 boolean_t doread = bp->b_flags & B_READ; 1540#else 1541 boolean_t doread = 0; 1542#endif 1543 boolean_t is_dumpified; 1544 boolean_t sync; 1545 1546#ifdef illumos 1547 if (getminor(bp->b_edev) == 0) { 1548 error = SET_ERROR(EINVAL); 1549 } else { 1550 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev)); 1551 if (zs == NULL) 1552 error = SET_ERROR(ENXIO); 1553 else if (zs->zss_type != ZSST_ZVOL) 1554 error = SET_ERROR(EINVAL); 1555 } 1556 1557 if (error) { 1558 bioerror(bp, error); 1559 biodone(bp); 1560 return (0); 1561 } 1562 1563 zv = zs->zss_data; 1564 1565 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) { 1566 bioerror(bp, EROFS); 1567 biodone(bp); 1568 return (0); 1569 } 1570 1571 off = ldbtob(bp->b_blkno); 1572#else /* !illumos */ 1573 if (bp->bio_to) 1574 zv = bp->bio_to->private; 1575 else 1576 zv = bp->bio_dev->si_drv2; 1577 1578 if (zv == NULL) { 1579 error = SET_ERROR(ENXIO); 1580 goto out; 1581 } 1582 1583 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) { 1584 error = SET_ERROR(EROFS); 1585 goto out; 1586 } 1587 1588 switch (bp->bio_cmd) { 1589 case BIO_FLUSH: 1590 goto sync; 1591 case BIO_READ: 1592 doread = 1; 1593 case BIO_WRITE: 1594 case BIO_DELETE: 1595 break; 1596 default: 1597 error = EOPNOTSUPP; 1598 goto out; 1599 } 1600 1601 off = bp->bio_offset; 1602#endif /* illumos */ 1603 volsize = zv->zv_volsize; 1604 1605 os = zv->zv_objset; 1606 ASSERT(os != NULL); 1607 1608#ifdef illumos 1609 bp_mapin(bp); 1610 addr = bp->b_un.b_addr; 1611 resid = bp->b_bcount; 1612 1613 if (resid > 0 && (off < 0 || off >= volsize)) { 1614 bioerror(bp, EIO); 1615 biodone(bp); 1616 return (0); 1617 } 1618 1619 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED; 1620 sync = ((!(bp->b_flags & B_ASYNC) && 1621 !(zv->zv_flags & ZVOL_WCE)) || 1622 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) && 1623 !doread && !is_dumpified; 1624#else /* !illumos */ 1625 addr = bp->bio_data; 1626 resid = bp->bio_length; 1627 1628 if (resid > 0 && (off < 0 || off >= volsize)) { 1629 error = SET_ERROR(EIO); 1630 goto out; 1631 } 1632 1633 is_dumpified = B_FALSE; 1634 sync = !doread && !is_dumpified && 1635 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; 1636#endif /* illumos */ 1637 1638 /* 1639 * There must be no buffer changes when doing a dmu_sync() because 1640 * we can't change the data whilst calculating the checksum. 1641 */ 1642 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1643 doread ? RL_READER : RL_WRITER); 1644 1645#ifndef illumos 1646 if (bp->bio_cmd == BIO_DELETE) { 1647 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1648 error = dmu_tx_assign(tx, TXG_WAIT); 1649 if (error != 0) { 1650 dmu_tx_abort(tx); 1651 } else { 1652 zvol_log_truncate(zv, tx, off, resid, B_TRUE); 1653 dmu_tx_commit(tx); 1654 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1655 off, resid); 1656 resid = 0; 1657 } 1658 goto unlock; 1659 } 1660#endif 1661 while (resid != 0 && off < volsize) { 1662 size_t size = MIN(resid, zvol_maxphys); 1663#ifdef illumos 1664 if (is_dumpified) { 1665 size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 1666 error = zvol_dumpio(zv, addr, off, size, 1667 doread, B_FALSE); 1668 } else if (doread) { 1669#else 1670 if (doread) { 1671#endif 1672 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1673 DMU_READ_PREFETCH); 1674 } else { 1675 dmu_tx_t *tx = dmu_tx_create(os); 1676 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1677 error = dmu_tx_assign(tx, TXG_WAIT); 1678 if (error) { 1679 dmu_tx_abort(tx); 1680 } else { 1681 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1682 zvol_log_write(zv, tx, off, size, sync); 1683 dmu_tx_commit(tx); 1684 } 1685 } 1686 if (error) { 1687 /* convert checksum errors into IO errors */ 1688 if (error == ECKSUM) 1689 error = SET_ERROR(EIO); 1690 break; 1691 } 1692 off += size; 1693 addr += size; 1694 resid -= size; 1695 } 1696#ifndef illumos 1697unlock: 1698#endif 1699 zfs_range_unlock(rl); 1700 1701#ifdef illumos 1702 if ((bp->b_resid = resid) == bp->b_bcount) 1703 bioerror(bp, off > volsize ? EINVAL : error); 1704 1705 if (sync) 1706 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1707 biodone(bp); 1708 1709 return (0); 1710#else /* !illumos */ 1711 bp->bio_completed = bp->bio_length - resid; 1712 if (bp->bio_completed < bp->bio_length && off > volsize) 1713 error = EINVAL; 1714 1715 if (sync) { 1716sync: 1717 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1718 } 1719out: 1720 if (bp->bio_to) 1721 g_io_deliver(bp, error); 1722 else 1723 biofinish(bp, NULL, error); 1724#endif /* illumos */ 1725} 1726 1727#ifdef illumos 1728/* 1729 * Set the buffer count to the zvol maximum transfer. 1730 * Using our own routine instead of the default minphys() 1731 * means that for larger writes we write bigger buffers on X86 1732 * (128K instead of 56K) and flush the disk write cache less often 1733 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1734 * 56K on X86 and 128K on sparc). 1735 */ 1736void 1737zvol_minphys(struct buf *bp) 1738{ 1739 if (bp->b_bcount > zvol_maxphys) 1740 bp->b_bcount = zvol_maxphys; 1741} 1742 1743int 1744zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1745{ 1746 minor_t minor = getminor(dev); 1747 zvol_state_t *zv; 1748 int error = 0; 1749 uint64_t size; 1750 uint64_t boff; 1751 uint64_t resid; 1752 1753 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1754 if (zv == NULL) 1755 return (SET_ERROR(ENXIO)); 1756 1757 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) 1758 return (SET_ERROR(EINVAL)); 1759 1760 boff = ldbtob(blkno); 1761 resid = ldbtob(nblocks); 1762 1763 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1764 1765 while (resid) { 1766 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1767 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1768 if (error) 1769 break; 1770 boff += size; 1771 addr += size; 1772 resid -= size; 1773 } 1774 1775 return (error); 1776} 1777 1778/*ARGSUSED*/ 1779int 1780zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1781{ 1782 minor_t minor = getminor(dev); 1783#else /* !illumos */ 1784int 1785zvol_read(struct cdev *dev, struct uio *uio, int ioflag) 1786{ 1787#endif /* illumos */ 1788 zvol_state_t *zv; 1789 uint64_t volsize; 1790 rl_t *rl; 1791 int error = 0; 1792 1793#ifdef illumos 1794 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1795 if (zv == NULL) 1796 return (SET_ERROR(ENXIO)); 1797#else 1798 zv = dev->si_drv2; 1799#endif 1800 1801 volsize = zv->zv_volsize; 1802 /* uio_loffset == volsize isn't an error as its required for EOF processing. */ 1803 if (uio->uio_resid > 0 && 1804 (uio->uio_loffset < 0 || uio->uio_loffset > volsize)) 1805 return (SET_ERROR(EIO)); 1806 1807#ifdef illumos 1808 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1809 error = physio(zvol_strategy, NULL, dev, B_READ, 1810 zvol_minphys, uio); 1811 return (error); 1812 } 1813#endif 1814 1815 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1816 RL_READER); 1817 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1818 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1819 1820 /* don't read past the end */ 1821 if (bytes > volsize - uio->uio_loffset) 1822 bytes = volsize - uio->uio_loffset; 1823 1824 error = dmu_read_uio_dbuf(zv->zv_dbuf, uio, bytes); 1825 if (error) { 1826 /* convert checksum errors into IO errors */ 1827 if (error == ECKSUM) 1828 error = SET_ERROR(EIO); 1829 break; 1830 } 1831 } 1832 zfs_range_unlock(rl); 1833 return (error); 1834} 1835 1836#ifdef illumos 1837/*ARGSUSED*/ 1838int 1839zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1840{ 1841 minor_t minor = getminor(dev); 1842#else /* !illumos */ 1843int 1844zvol_write(struct cdev *dev, struct uio *uio, int ioflag) 1845{ 1846#endif /* illumos */ 1847 zvol_state_t *zv; 1848 uint64_t volsize; 1849 rl_t *rl; 1850 int error = 0; 1851 boolean_t sync; 1852 1853#ifdef illumos 1854 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1855 if (zv == NULL) 1856 return (SET_ERROR(ENXIO)); 1857#else 1858 zv = dev->si_drv2; 1859#endif 1860 1861 volsize = zv->zv_volsize; 1862 /* uio_loffset == volsize isn't an error as its required for EOF processing. */ 1863 if (uio->uio_resid > 0 && 1864 (uio->uio_loffset < 0 || uio->uio_loffset > volsize)) 1865 return (SET_ERROR(EIO)); 1866 1867#ifdef illumos 1868 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1869 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1870 zvol_minphys, uio); 1871 return (error); 1872 } 1873 1874 sync = !(zv->zv_flags & ZVOL_WCE) || 1875#else 1876 sync = (ioflag & IO_SYNC) || 1877#endif 1878 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1879 1880 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1881 RL_WRITER); 1882 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1883 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1884 uint64_t off = uio->uio_loffset; 1885 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1886 1887 if (bytes > volsize - off) /* don't write past the end */ 1888 bytes = volsize - off; 1889 1890 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1891 error = dmu_tx_assign(tx, TXG_WAIT); 1892 if (error) { 1893 dmu_tx_abort(tx); 1894 break; 1895 } 1896 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1897 if (error == 0) 1898 zvol_log_write(zv, tx, off, bytes, sync); 1899 dmu_tx_commit(tx); 1900 1901 if (error) 1902 break; 1903 } 1904 zfs_range_unlock(rl); 1905 if (sync) 1906 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1907 return (error); 1908} 1909 1910#ifdef illumos 1911int 1912zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1913{ 1914 struct uuid uuid = EFI_RESERVED; 1915 efi_gpe_t gpe = { 0 }; 1916 uint32_t crc; 1917 dk_efi_t efi; 1918 int length; 1919 char *ptr; 1920 1921 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1922 return (SET_ERROR(EFAULT)); 1923 ptr = (char *)(uintptr_t)efi.dki_data_64; 1924 length = efi.dki_length; 1925 /* 1926 * Some clients may attempt to request a PMBR for the 1927 * zvol. Currently this interface will return EINVAL to 1928 * such requests. These requests could be supported by 1929 * adding a check for lba == 0 and consing up an appropriate 1930 * PMBR. 1931 */ 1932 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1933 return (SET_ERROR(EINVAL)); 1934 1935 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1936 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1937 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1938 1939 if (efi.dki_lba == 1) { 1940 efi_gpt_t gpt = { 0 }; 1941 1942 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1943 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1944 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1945 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1946 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1947 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1948 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1949 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1950 gpt.efi_gpt_SizeOfPartitionEntry = 1951 LE_32(sizeof (efi_gpe_t)); 1952 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1953 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1954 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1955 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1956 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1957 flag)) 1958 return (SET_ERROR(EFAULT)); 1959 ptr += sizeof (gpt); 1960 length -= sizeof (gpt); 1961 } 1962 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1963 length), flag)) 1964 return (SET_ERROR(EFAULT)); 1965 return (0); 1966} 1967 1968/* 1969 * BEGIN entry points to allow external callers access to the volume. 1970 */ 1971/* 1972 * Return the volume parameters needed for access from an external caller. 1973 * These values are invariant as long as the volume is held open. 1974 */ 1975int 1976zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1977 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1978 void **rl_hdl, void **bonus_hdl) 1979{ 1980 zvol_state_t *zv; 1981 1982 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1983 if (zv == NULL) 1984 return (SET_ERROR(ENXIO)); 1985 if (zv->zv_flags & ZVOL_DUMPIFIED) 1986 return (SET_ERROR(ENXIO)); 1987 1988 ASSERT(blksize && max_xfer_len && minor_hdl && 1989 objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 1990 1991 *blksize = zv->zv_volblocksize; 1992 *max_xfer_len = (uint64_t)zvol_maxphys; 1993 *minor_hdl = zv; 1994 *objset_hdl = zv->zv_objset; 1995 *zil_hdl = zv->zv_zilog; 1996 *rl_hdl = &zv->zv_znode; 1997 *bonus_hdl = zv->zv_dbuf; 1998 return (0); 1999} 2000 2001/* 2002 * Return the current volume size to an external caller. 2003 * The size can change while the volume is open. 2004 */ 2005uint64_t 2006zvol_get_volume_size(void *minor_hdl) 2007{ 2008 zvol_state_t *zv = minor_hdl; 2009 2010 return (zv->zv_volsize); 2011} 2012 2013/* 2014 * Return the current WCE setting to an external caller. 2015 * The WCE setting can change while the volume is open. 2016 */ 2017int 2018zvol_get_volume_wce(void *minor_hdl) 2019{ 2020 zvol_state_t *zv = minor_hdl; 2021 2022 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 2023} 2024 2025/* 2026 * Entry point for external callers to zvol_log_write 2027 */ 2028void 2029zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 2030 boolean_t sync) 2031{ 2032 zvol_state_t *zv = minor_hdl; 2033 2034 zvol_log_write(zv, tx, off, resid, sync); 2035} 2036/* 2037 * END entry points to allow external callers access to the volume. 2038 */ 2039#endif /* illumos */ 2040 2041/* 2042 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. 2043 */ 2044static void 2045zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, 2046 boolean_t sync) 2047{ 2048 itx_t *itx; 2049 lr_truncate_t *lr; 2050 zilog_t *zilog = zv->zv_zilog; 2051 2052 if (zil_replaying(zilog, tx)) 2053 return; 2054 2055 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 2056 lr = (lr_truncate_t *)&itx->itx_lr; 2057 lr->lr_foid = ZVOL_OBJ; 2058 lr->lr_offset = off; 2059 lr->lr_length = len; 2060 2061 itx->itx_sync = sync; 2062 zil_itx_assign(zilog, itx, tx); 2063} 2064 2065#ifdef illumos 2066/* 2067 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 2068 * Also a dirtbag dkio ioctl for unmap/free-block functionality. 2069 */ 2070/*ARGSUSED*/ 2071int 2072zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 2073{ 2074 zvol_state_t *zv; 2075 struct dk_callback *dkc; 2076 int error = 0; 2077 rl_t *rl; 2078 2079 mutex_enter(&zfsdev_state_lock); 2080 2081 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 2082 2083 if (zv == NULL) { 2084 mutex_exit(&zfsdev_state_lock); 2085 return (SET_ERROR(ENXIO)); 2086 } 2087 ASSERT(zv->zv_total_opens > 0); 2088 2089 switch (cmd) { 2090 2091 case DKIOCINFO: 2092 { 2093 struct dk_cinfo dki; 2094 2095 bzero(&dki, sizeof (dki)); 2096 (void) strcpy(dki.dki_cname, "zvol"); 2097 (void) strcpy(dki.dki_dname, "zvol"); 2098 dki.dki_ctype = DKC_UNKNOWN; 2099 dki.dki_unit = getminor(dev); 2100 dki.dki_maxtransfer = 2101 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs); 2102 mutex_exit(&zfsdev_state_lock); 2103 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 2104 error = SET_ERROR(EFAULT); 2105 return (error); 2106 } 2107 2108 case DKIOCGMEDIAINFO: 2109 { 2110 struct dk_minfo dkm; 2111 2112 bzero(&dkm, sizeof (dkm)); 2113 dkm.dki_lbsize = 1U << zv->zv_min_bs; 2114 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 2115 dkm.dki_media_type = DK_UNKNOWN; 2116 mutex_exit(&zfsdev_state_lock); 2117 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 2118 error = SET_ERROR(EFAULT); 2119 return (error); 2120 } 2121 2122 case DKIOCGMEDIAINFOEXT: 2123 { 2124 struct dk_minfo_ext dkmext; 2125 2126 bzero(&dkmext, sizeof (dkmext)); 2127 dkmext.dki_lbsize = 1U << zv->zv_min_bs; 2128 dkmext.dki_pbsize = zv->zv_volblocksize; 2129 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 2130 dkmext.dki_media_type = DK_UNKNOWN; 2131 mutex_exit(&zfsdev_state_lock); 2132 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag)) 2133 error = SET_ERROR(EFAULT); 2134 return (error); 2135 } 2136 2137 case DKIOCGETEFI: 2138 { 2139 uint64_t vs = zv->zv_volsize; 2140 uint8_t bs = zv->zv_min_bs; 2141 2142 mutex_exit(&zfsdev_state_lock); 2143 error = zvol_getefi((void *)arg, flag, vs, bs); 2144 return (error); 2145 } 2146 2147 case DKIOCFLUSHWRITECACHE: 2148 dkc = (struct dk_callback *)arg; 2149 mutex_exit(&zfsdev_state_lock); 2150 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2151 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 2152 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 2153 error = 0; 2154 } 2155 return (error); 2156 2157 case DKIOCGETWCE: 2158 { 2159 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 2160 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 2161 flag)) 2162 error = SET_ERROR(EFAULT); 2163 break; 2164 } 2165 case DKIOCSETWCE: 2166 { 2167 int wce; 2168 if (ddi_copyin((void *)arg, &wce, sizeof (int), 2169 flag)) { 2170 error = SET_ERROR(EFAULT); 2171 break; 2172 } 2173 if (wce) { 2174 zv->zv_flags |= ZVOL_WCE; 2175 mutex_exit(&zfsdev_state_lock); 2176 } else { 2177 zv->zv_flags &= ~ZVOL_WCE; 2178 mutex_exit(&zfsdev_state_lock); 2179 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2180 } 2181 return (0); 2182 } 2183 2184 case DKIOCGGEOM: 2185 case DKIOCGVTOC: 2186 /* 2187 * commands using these (like prtvtoc) expect ENOTSUP 2188 * since we're emulating an EFI label 2189 */ 2190 error = SET_ERROR(ENOTSUP); 2191 break; 2192 2193 case DKIOCDUMPINIT: 2194 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 2195 RL_WRITER); 2196 error = zvol_dumpify(zv); 2197 zfs_range_unlock(rl); 2198 break; 2199 2200 case DKIOCDUMPFINI: 2201 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 2202 break; 2203 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 2204 RL_WRITER); 2205 error = zvol_dump_fini(zv); 2206 zfs_range_unlock(rl); 2207 break; 2208 2209 case DKIOCFREE: 2210 { 2211 dkioc_free_t df; 2212 dmu_tx_t *tx; 2213 2214 if (!zvol_unmap_enabled) 2215 break; 2216 2217 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) { 2218 error = SET_ERROR(EFAULT); 2219 break; 2220 } 2221 2222 /* 2223 * Apply Postel's Law to length-checking. If they overshoot, 2224 * just blank out until the end, if there's a need to blank 2225 * out anything. 2226 */ 2227 if (df.df_start >= zv->zv_volsize) 2228 break; /* No need to do anything... */ 2229 2230 mutex_exit(&zfsdev_state_lock); 2231 2232 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length, 2233 RL_WRITER); 2234 tx = dmu_tx_create(zv->zv_objset); 2235 dmu_tx_mark_netfree(tx); 2236 error = dmu_tx_assign(tx, TXG_WAIT); 2237 if (error != 0) { 2238 dmu_tx_abort(tx); 2239 } else { 2240 zvol_log_truncate(zv, tx, df.df_start, 2241 df.df_length, B_TRUE); 2242 dmu_tx_commit(tx); 2243 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 2244 df.df_start, df.df_length); 2245 } 2246 2247 zfs_range_unlock(rl); 2248 2249 if (error == 0) { 2250 /* 2251 * If the write-cache is disabled or 'sync' property 2252 * is set to 'always' then treat this as a synchronous 2253 * operation (i.e. commit to zil). 2254 */ 2255 if (!(zv->zv_flags & ZVOL_WCE) || 2256 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) 2257 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2258 2259 /* 2260 * If the caller really wants synchronous writes, and 2261 * can't wait for them, don't return until the write 2262 * is done. 2263 */ 2264 if (df.df_flags & DF_WAIT_SYNC) { 2265 txg_wait_synced( 2266 dmu_objset_pool(zv->zv_objset), 0); 2267 } 2268 } 2269 return (error); 2270 } 2271 2272 default: 2273 error = SET_ERROR(ENOTTY); 2274 break; 2275 2276 } 2277 mutex_exit(&zfsdev_state_lock); 2278 return (error); 2279} 2280#endif /* illumos */ 2281 2282int 2283zvol_busy(void) 2284{ 2285 return (zvol_minors != 0); 2286} 2287 2288void 2289zvol_init(void) 2290{ 2291 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 2292 1) == 0); 2293#ifdef illumos 2294 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL); 2295#else 2296 ZFS_LOG(1, "ZVOL Initialized."); 2297#endif 2298} 2299 2300void 2301zvol_fini(void) 2302{ 2303#ifdef illumos 2304 mutex_destroy(&zfsdev_state_lock); 2305#endif 2306 ddi_soft_state_fini(&zfsdev_state); 2307 ZFS_LOG(1, "ZVOL Deinitialized."); 2308} 2309 2310#ifdef illumos 2311/*ARGSUSED*/ 2312static int 2313zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx) 2314{ 2315 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 2316 2317 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 2318 return (1); 2319 return (0); 2320} 2321 2322/*ARGSUSED*/ 2323static void 2324zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx) 2325{ 2326 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 2327 2328 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx); 2329} 2330 2331static int 2332zvol_dump_init(zvol_state_t *zv, boolean_t resize) 2333{ 2334 dmu_tx_t *tx; 2335 int error; 2336 objset_t *os = zv->zv_objset; 2337 spa_t *spa = dmu_objset_spa(os); 2338 vdev_t *vd = spa->spa_root_vdev; 2339 nvlist_t *nv = NULL; 2340 uint64_t version = spa_version(spa); 2341 uint64_t checksum, compress, refresrv, vbs, dedup; 2342 2343 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 2344 ASSERT(vd->vdev_ops == &vdev_root_ops); 2345 2346 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 2347 DMU_OBJECT_END); 2348 if (error != 0) 2349 return (error); 2350 /* wait for dmu_free_long_range to actually free the blocks */ 2351 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2352 2353 /* 2354 * If the pool on which the dump device is being initialized has more 2355 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is 2356 * enabled. If so, bump that feature's counter to indicate that the 2357 * feature is active. We also check the vdev type to handle the 2358 * following case: 2359 * # zpool create test raidz disk1 disk2 disk3 2360 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev), 2361 * the raidz vdev itself has 3 children. 2362 */ 2363 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) { 2364 if (!spa_feature_is_enabled(spa, 2365 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 2366 return (SET_ERROR(ENOTSUP)); 2367 (void) dsl_sync_task(spa_name(spa), 2368 zfs_mvdev_dump_feature_check, 2369 zfs_mvdev_dump_activate_feature_sync, NULL, 2370 2, ZFS_SPACE_CHECK_RESERVED); 2371 } 2372 2373 if (!resize) { 2374 error = dsl_prop_get_integer(zv->zv_name, 2375 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 2376 if (error == 0) { 2377 error = dsl_prop_get_integer(zv->zv_name, 2378 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, 2379 NULL); 2380 } 2381 if (error == 0) { 2382 error = dsl_prop_get_integer(zv->zv_name, 2383 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 2384 &refresrv, NULL); 2385 } 2386 if (error == 0) { 2387 error = dsl_prop_get_integer(zv->zv_name, 2388 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, 2389 NULL); 2390 } 2391 if (version >= SPA_VERSION_DEDUP && error == 0) { 2392 error = dsl_prop_get_integer(zv->zv_name, 2393 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 2394 } 2395 } 2396 if (error != 0) 2397 return (error); 2398 2399 tx = dmu_tx_create(os); 2400 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2401 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2402 error = dmu_tx_assign(tx, TXG_WAIT); 2403 if (error != 0) { 2404 dmu_tx_abort(tx); 2405 return (error); 2406 } 2407 2408 /* 2409 * If we are resizing the dump device then we only need to 2410 * update the refreservation to match the newly updated 2411 * zvolsize. Otherwise, we save off the original state of the 2412 * zvol so that we can restore them if the zvol is ever undumpified. 2413 */ 2414 if (resize) { 2415 error = zap_update(os, ZVOL_ZAP_OBJ, 2416 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2417 &zv->zv_volsize, tx); 2418 } else { 2419 error = zap_update(os, ZVOL_ZAP_OBJ, 2420 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 2421 &compress, tx); 2422 if (error == 0) { 2423 error = zap_update(os, ZVOL_ZAP_OBJ, 2424 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, 2425 &checksum, tx); 2426 } 2427 if (error == 0) { 2428 error = zap_update(os, ZVOL_ZAP_OBJ, 2429 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2430 &refresrv, tx); 2431 } 2432 if (error == 0) { 2433 error = zap_update(os, ZVOL_ZAP_OBJ, 2434 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 2435 &vbs, tx); 2436 } 2437 if (error == 0) { 2438 error = dmu_object_set_blocksize( 2439 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx); 2440 } 2441 if (version >= SPA_VERSION_DEDUP && error == 0) { 2442 error = zap_update(os, ZVOL_ZAP_OBJ, 2443 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 2444 &dedup, tx); 2445 } 2446 if (error == 0) 2447 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE; 2448 } 2449 dmu_tx_commit(tx); 2450 2451 /* 2452 * We only need update the zvol's property if we are initializing 2453 * the dump area for the first time. 2454 */ 2455 if (error == 0 && !resize) { 2456 /* 2457 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum 2458 * function. Otherwise, use the old default -- OFF. 2459 */ 2460 checksum = spa_feature_is_active(spa, 2461 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY : 2462 ZIO_CHECKSUM_OFF; 2463 2464 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2465 VERIFY(nvlist_add_uint64(nv, 2466 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 2467 VERIFY(nvlist_add_uint64(nv, 2468 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 2469 ZIO_COMPRESS_OFF) == 0); 2470 VERIFY(nvlist_add_uint64(nv, 2471 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 2472 checksum) == 0); 2473 if (version >= SPA_VERSION_DEDUP) { 2474 VERIFY(nvlist_add_uint64(nv, 2475 zfs_prop_to_name(ZFS_PROP_DEDUP), 2476 ZIO_CHECKSUM_OFF) == 0); 2477 } 2478 2479 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2480 nv, NULL); 2481 nvlist_free(nv); 2482 } 2483 2484 /* Allocate the space for the dump */ 2485 if (error == 0) 2486 error = zvol_prealloc(zv); 2487 return (error); 2488} 2489 2490static int 2491zvol_dumpify(zvol_state_t *zv) 2492{ 2493 int error = 0; 2494 uint64_t dumpsize = 0; 2495 dmu_tx_t *tx; 2496 objset_t *os = zv->zv_objset; 2497 2498 if (zv->zv_flags & ZVOL_RDONLY) 2499 return (SET_ERROR(EROFS)); 2500 2501 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 2502 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 2503 boolean_t resize = (dumpsize > 0); 2504 2505 if ((error = zvol_dump_init(zv, resize)) != 0) { 2506 (void) zvol_dump_fini(zv); 2507 return (error); 2508 } 2509 } 2510 2511 /* 2512 * Build up our lba mapping. 2513 */ 2514 error = zvol_get_lbas(zv); 2515 if (error) { 2516 (void) zvol_dump_fini(zv); 2517 return (error); 2518 } 2519 2520 tx = dmu_tx_create(os); 2521 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2522 error = dmu_tx_assign(tx, TXG_WAIT); 2523 if (error) { 2524 dmu_tx_abort(tx); 2525 (void) zvol_dump_fini(zv); 2526 return (error); 2527 } 2528 2529 zv->zv_flags |= ZVOL_DUMPIFIED; 2530 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 2531 &zv->zv_volsize, tx); 2532 dmu_tx_commit(tx); 2533 2534 if (error) { 2535 (void) zvol_dump_fini(zv); 2536 return (error); 2537 } 2538 2539 txg_wait_synced(dmu_objset_pool(os), 0); 2540 return (0); 2541} 2542 2543static int 2544zvol_dump_fini(zvol_state_t *zv) 2545{ 2546 dmu_tx_t *tx; 2547 objset_t *os = zv->zv_objset; 2548 nvlist_t *nv; 2549 int error = 0; 2550 uint64_t checksum, compress, refresrv, vbs, dedup; 2551 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 2552 2553 /* 2554 * Attempt to restore the zvol back to its pre-dumpified state. 2555 * This is a best-effort attempt as it's possible that not all 2556 * of these properties were initialized during the dumpify process 2557 * (i.e. error during zvol_dump_init). 2558 */ 2559 2560 tx = dmu_tx_create(os); 2561 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2562 error = dmu_tx_assign(tx, TXG_WAIT); 2563 if (error) { 2564 dmu_tx_abort(tx); 2565 return (error); 2566 } 2567 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 2568 dmu_tx_commit(tx); 2569 2570 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2571 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 2572 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2573 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 2574 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2575 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 2576 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2577 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 2578 2579 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2580 (void) nvlist_add_uint64(nv, 2581 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 2582 (void) nvlist_add_uint64(nv, 2583 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 2584 (void) nvlist_add_uint64(nv, 2585 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 2586 if (version >= SPA_VERSION_DEDUP && 2587 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2588 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 2589 (void) nvlist_add_uint64(nv, 2590 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 2591 } 2592 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2593 nv, NULL); 2594 nvlist_free(nv); 2595 2596 zvol_free_extents(zv); 2597 zv->zv_flags &= ~ZVOL_DUMPIFIED; 2598 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 2599 /* wait for dmu_free_long_range to actually free the blocks */ 2600 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2601 tx = dmu_tx_create(os); 2602 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2603 error = dmu_tx_assign(tx, TXG_WAIT); 2604 if (error) { 2605 dmu_tx_abort(tx); 2606 return (error); 2607 } 2608 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 2609 zv->zv_volblocksize = vbs; 2610 dmu_tx_commit(tx); 2611 2612 return (0); 2613} 2614#else /* !illumos */ 2615 2616static void 2617zvol_geom_run(zvol_state_t *zv) 2618{ 2619 struct g_provider *pp; 2620 2621 pp = zv->zv_provider; 2622 g_error_provider(pp, 0); 2623 2624 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0, 2625 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER)); 2626} 2627 2628static void 2629zvol_geom_destroy(zvol_state_t *zv) 2630{ 2631 struct g_provider *pp; 2632 2633 g_topology_assert(); 2634 2635 mtx_lock(&zv->zv_queue_mtx); 2636 zv->zv_state = 1; 2637 wakeup_one(&zv->zv_queue); 2638 while (zv->zv_state != 2) 2639 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0); 2640 mtx_destroy(&zv->zv_queue_mtx); 2641 2642 pp = zv->zv_provider; 2643 zv->zv_provider = NULL; 2644 pp->private = NULL; 2645 g_wither_geom(pp->geom, ENXIO); 2646} 2647 2648static int 2649zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) 2650{ 2651 int count, error, flags; 2652 2653 g_topology_assert(); 2654 2655 /* 2656 * To make it easier we expect either open or close, but not both 2657 * at the same time. 2658 */ 2659 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) || 2660 (acr <= 0 && acw <= 0 && ace <= 0), 2661 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", 2662 pp->name, acr, acw, ace)); 2663 2664 if (pp->private == NULL) { 2665 if (acr <= 0 && acw <= 0 && ace <= 0) 2666 return (0); 2667 return (pp->error); 2668 } 2669 2670 /* 2671 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0, 2672 * because GEOM already handles that and handles it a bit differently. 2673 * GEOM allows for multiple read/exclusive consumers and ZFS allows 2674 * only one exclusive consumer, no matter if it is reader or writer. 2675 * I like better the way GEOM works so I'll leave it for GEOM to 2676 * decide what to do. 2677 */ 2678 2679 count = acr + acw + ace; 2680 if (count == 0) 2681 return (0); 2682 2683 flags = 0; 2684 if (acr != 0 || ace != 0) 2685 flags |= FREAD; 2686 if (acw != 0) 2687 flags |= FWRITE; 2688 2689 g_topology_unlock(); 2690 if (count > 0) 2691 error = zvol_open(pp, flags, count); 2692 else 2693 error = zvol_close(pp, flags, -count); 2694 g_topology_lock(); 2695 return (error); 2696} 2697 2698static void 2699zvol_geom_start(struct bio *bp) 2700{ 2701 zvol_state_t *zv; 2702 boolean_t first; 2703 2704 zv = bp->bio_to->private; 2705 ASSERT(zv != NULL); 2706 switch (bp->bio_cmd) { 2707 case BIO_FLUSH: 2708 if (!THREAD_CAN_SLEEP()) 2709 goto enqueue; 2710 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2711 g_io_deliver(bp, 0); 2712 break; 2713 case BIO_READ: 2714 case BIO_WRITE: 2715 case BIO_DELETE: 2716 if (!THREAD_CAN_SLEEP()) 2717 goto enqueue; 2718 zvol_strategy(bp); 2719 break; 2720 case BIO_GETATTR: { 2721 spa_t *spa = dmu_objset_spa(zv->zv_objset); 2722 uint64_t refd, avail, usedobjs, availobjs, val; 2723 2724 if (g_handleattr_int(bp, "GEOM::candelete", 1)) 2725 return; 2726 if (strcmp(bp->bio_attribute, "blocksavail") == 0) { 2727 dmu_objset_space(zv->zv_objset, &refd, &avail, 2728 &usedobjs, &availobjs); 2729 if (g_handleattr_off_t(bp, "blocksavail", 2730 avail / DEV_BSIZE)) 2731 return; 2732 } else if (strcmp(bp->bio_attribute, "blocksused") == 0) { 2733 dmu_objset_space(zv->zv_objset, &refd, &avail, 2734 &usedobjs, &availobjs); 2735 if (g_handleattr_off_t(bp, "blocksused", 2736 refd / DEV_BSIZE)) 2737 return; 2738 } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) { 2739 avail = metaslab_class_get_space(spa_normal_class(spa)); 2740 avail -= metaslab_class_get_alloc(spa_normal_class(spa)); 2741 if (g_handleattr_off_t(bp, "poolblocksavail", 2742 avail / DEV_BSIZE)) 2743 return; 2744 } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) { 2745 refd = metaslab_class_get_alloc(spa_normal_class(spa)); 2746 if (g_handleattr_off_t(bp, "poolblocksused", 2747 refd / DEV_BSIZE)) 2748 return; 2749 } 2750 /* FALLTHROUGH */ 2751 } 2752 default: 2753 g_io_deliver(bp, EOPNOTSUPP); 2754 break; 2755 } 2756 return; 2757 2758enqueue: 2759 mtx_lock(&zv->zv_queue_mtx); 2760 first = (bioq_first(&zv->zv_queue) == NULL); 2761 bioq_insert_tail(&zv->zv_queue, bp); 2762 mtx_unlock(&zv->zv_queue_mtx); 2763 if (first) 2764 wakeup_one(&zv->zv_queue); 2765} 2766 2767static void 2768zvol_geom_worker(void *arg) 2769{ 2770 zvol_state_t *zv; 2771 struct bio *bp; 2772 2773 thread_lock(curthread); 2774 sched_prio(curthread, PRIBIO); 2775 thread_unlock(curthread); 2776 2777 zv = arg; 2778 for (;;) { 2779 mtx_lock(&zv->zv_queue_mtx); 2780 bp = bioq_takefirst(&zv->zv_queue); 2781 if (bp == NULL) { 2782 if (zv->zv_state == 1) { 2783 zv->zv_state = 2; 2784 wakeup(&zv->zv_state); 2785 mtx_unlock(&zv->zv_queue_mtx); 2786 kthread_exit(); 2787 } 2788 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP, 2789 "zvol:io", 0); 2790 continue; 2791 } 2792 mtx_unlock(&zv->zv_queue_mtx); 2793 switch (bp->bio_cmd) { 2794 case BIO_FLUSH: 2795 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2796 g_io_deliver(bp, 0); 2797 break; 2798 case BIO_READ: 2799 case BIO_WRITE: 2800 case BIO_DELETE: 2801 zvol_strategy(bp); 2802 break; 2803 default: 2804 g_io_deliver(bp, EOPNOTSUPP); 2805 break; 2806 } 2807 } 2808} 2809 2810extern boolean_t dataset_name_hidden(const char *name); 2811 2812static int 2813zvol_create_snapshots(objset_t *os, const char *name) 2814{ 2815 uint64_t cookie, obj; 2816 char *sname; 2817 int error, len; 2818 2819 cookie = obj = 0; 2820 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2821 2822#if 0 2823 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL, 2824 DS_FIND_SNAPSHOTS); 2825#endif 2826 2827 for (;;) { 2828 len = snprintf(sname, MAXPATHLEN, "%s@", name); 2829 if (len >= MAXPATHLEN) { 2830 dmu_objset_rele(os, FTAG); 2831 error = ENAMETOOLONG; 2832 break; 2833 } 2834 2835 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 2836 error = dmu_snapshot_list_next(os, MAXPATHLEN - len, 2837 sname + len, &obj, &cookie, NULL); 2838 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 2839 if (error != 0) { 2840 if (error == ENOENT) 2841 error = 0; 2842 break; 2843 } 2844 2845 error = zvol_create_minor(sname); 2846 if (error != 0 && error != EEXIST) { 2847 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2848 sname, error); 2849 break; 2850 } 2851 } 2852 2853 kmem_free(sname, MAXPATHLEN); 2854 return (error); 2855} 2856 2857int 2858zvol_create_minors(const char *name) 2859{ 2860 uint64_t cookie; 2861 objset_t *os; 2862 char *osname, *p; 2863 int error, len; 2864 2865 if (dataset_name_hidden(name)) 2866 return (0); 2867 2868 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2869 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2870 name, error); 2871 return (error); 2872 } 2873 if (dmu_objset_type(os) == DMU_OST_ZVOL) { 2874 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG); 2875 dsl_pool_rele(dmu_objset_pool(os), FTAG); 2876 error = zvol_create_minor(name); 2877 if (error == 0 || error == EEXIST) { 2878 error = zvol_create_snapshots(os, name); 2879 } else { 2880 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2881 name, error); 2882 } 2883 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG); 2884 dsl_dataset_rele(os->os_dsl_dataset, FTAG); 2885 return (error); 2886 } 2887 if (dmu_objset_type(os) != DMU_OST_ZFS) { 2888 dmu_objset_rele(os, FTAG); 2889 return (0); 2890 } 2891 2892 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2893 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) { 2894 dmu_objset_rele(os, FTAG); 2895 kmem_free(osname, MAXPATHLEN); 2896 return (ENOENT); 2897 } 2898 p = osname + strlen(osname); 2899 len = MAXPATHLEN - (p - osname); 2900 2901#if 0 2902 /* Prefetch the datasets. */ 2903 cookie = 0; 2904 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) { 2905 if (!dataset_name_hidden(osname)) 2906 (void) dmu_objset_prefetch(osname, NULL); 2907 } 2908#endif 2909 2910 cookie = 0; 2911 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL, 2912 &cookie) == 0) { 2913 dmu_objset_rele(os, FTAG); 2914 (void)zvol_create_minors(osname); 2915 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2916 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2917 name, error); 2918 return (error); 2919 } 2920 } 2921 2922 dmu_objset_rele(os, FTAG); 2923 kmem_free(osname, MAXPATHLEN); 2924 return (0); 2925} 2926 2927static void 2928zvol_rename_minor(zvol_state_t *zv, const char *newname) 2929{ 2930 struct g_geom *gp; 2931 struct g_provider *pp; 2932 struct cdev *dev; 2933 2934 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 2935 2936 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 2937 g_topology_lock(); 2938 pp = zv->zv_provider; 2939 ASSERT(pp != NULL); 2940 gp = pp->geom; 2941 ASSERT(gp != NULL); 2942 2943 zv->zv_provider = NULL; 2944 g_wither_provider(pp, ENXIO); 2945 2946 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname); 2947 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 2948 pp->sectorsize = DEV_BSIZE; 2949 pp->mediasize = zv->zv_volsize; 2950 pp->private = zv; 2951 zv->zv_provider = pp; 2952 g_error_provider(pp, 0); 2953 g_topology_unlock(); 2954 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { 2955 struct make_dev_args args; 2956 2957 if ((dev = zv->zv_dev) != NULL) { 2958 zv->zv_dev = NULL; 2959 destroy_dev(dev); 2960 if (zv->zv_total_opens > 0) { 2961 zv->zv_flags &= ~ZVOL_EXCL; 2962 zv->zv_total_opens = 0; 2963 zvol_last_close(zv); 2964 } 2965 } 2966 2967 make_dev_args_init(&args); 2968 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK; 2969 args.mda_devsw = &zvol_cdevsw; 2970 args.mda_cr = NULL; 2971 args.mda_uid = UID_ROOT; 2972 args.mda_gid = GID_OPERATOR; 2973 args.mda_mode = 0640; 2974 args.mda_si_drv2 = zv; 2975 if (make_dev_s(&args, &zv->zv_dev, 2976 "%s/%s", ZVOL_DRIVER, newname) == 0) 2977 zv->zv_dev->si_iosize_max = MAXPHYS; 2978 } 2979 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name)); 2980} 2981 2982void 2983zvol_rename_minors(const char *oldname, const char *newname) 2984{ 2985 char name[MAXPATHLEN]; 2986 struct g_provider *pp; 2987 struct g_geom *gp; 2988 size_t oldnamelen, newnamelen; 2989 zvol_state_t *zv; 2990 char *namebuf; 2991 boolean_t locked = B_FALSE; 2992 2993 oldnamelen = strlen(oldname); 2994 newnamelen = strlen(newname); 2995 2996 DROP_GIANT(); 2997 /* See comment in zvol_open(). */ 2998 if (!MUTEX_HELD(&zfsdev_state_lock)) { 2999 mutex_enter(&zfsdev_state_lock); 3000 locked = B_TRUE; 3001 } 3002 3003 LIST_FOREACH(zv, &all_zvols, zv_links) { 3004 if (strcmp(zv->zv_name, oldname) == 0) { 3005 zvol_rename_minor(zv, newname); 3006 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && 3007 (zv->zv_name[oldnamelen] == '/' || 3008 zv->zv_name[oldnamelen] == '@')) { 3009 snprintf(name, sizeof(name), "%s%c%s", newname, 3010 zv->zv_name[oldnamelen], 3011 zv->zv_name + oldnamelen + 1); 3012 zvol_rename_minor(zv, name); 3013 } 3014 } 3015 3016 if (locked) 3017 mutex_exit(&zfsdev_state_lock); 3018 PICKUP_GIANT(); 3019} 3020 3021static int 3022zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td) 3023{ 3024 zvol_state_t *zv = dev->si_drv2; 3025 int err = 0; 3026 3027 mutex_enter(&zfsdev_state_lock); 3028 if (zv->zv_total_opens == 0) 3029 err = zvol_first_open(zv); 3030 if (err) { 3031 mutex_exit(&zfsdev_state_lock); 3032 return (err); 3033 } 3034 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 3035 err = SET_ERROR(EROFS); 3036 goto out; 3037 } 3038 if (zv->zv_flags & ZVOL_EXCL) { 3039 err = SET_ERROR(EBUSY); 3040 goto out; 3041 } 3042#ifdef FEXCL 3043 if (flags & FEXCL) { 3044 if (zv->zv_total_opens != 0) { 3045 err = SET_ERROR(EBUSY); 3046 goto out; 3047 } 3048 zv->zv_flags |= ZVOL_EXCL; 3049 } 3050#endif 3051 3052 zv->zv_total_opens++; 3053 mutex_exit(&zfsdev_state_lock); 3054 return (err); 3055out: 3056 if (zv->zv_total_opens == 0) 3057 zvol_last_close(zv); 3058 mutex_exit(&zfsdev_state_lock); 3059 return (err); 3060} 3061 3062static int 3063zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td) 3064{ 3065 zvol_state_t *zv = dev->si_drv2; 3066 3067 mutex_enter(&zfsdev_state_lock); 3068 if (zv->zv_flags & ZVOL_EXCL) { 3069 ASSERT(zv->zv_total_opens == 1); 3070 zv->zv_flags &= ~ZVOL_EXCL; 3071 } 3072 3073 /* 3074 * If the open count is zero, this is a spurious close. 3075 * That indicates a bug in the kernel / DDI framework. 3076 */ 3077 ASSERT(zv->zv_total_opens != 0); 3078 3079 /* 3080 * You may get multiple opens, but only one close. 3081 */ 3082 zv->zv_total_opens--; 3083 3084 if (zv->zv_total_opens == 0) 3085 zvol_last_close(zv); 3086 3087 mutex_exit(&zfsdev_state_lock); 3088 return (0); 3089} 3090 3091static int 3092zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 3093{ 3094 zvol_state_t *zv; 3095 rl_t *rl; 3096 off_t offset, length, chunk; 3097 int i, error; 3098 u_int u; 3099 3100 zv = dev->si_drv2; 3101 3102 error = 0; 3103 KASSERT(zv->zv_total_opens > 0, 3104 ("Device with zero access count in zvol_d_ioctl")); 3105 3106 i = IOCPARM_LEN(cmd); 3107 switch (cmd) { 3108 case DIOCGSECTORSIZE: 3109 *(u_int *)data = DEV_BSIZE; 3110 break; 3111 case DIOCGMEDIASIZE: 3112 *(off_t *)data = zv->zv_volsize; 3113 break; 3114 case DIOCGFLUSH: 3115 zil_commit(zv->zv_zilog, ZVOL_OBJ); 3116 break; 3117 case DIOCGDELETE: 3118 if (!zvol_unmap_enabled) 3119 break; 3120 3121 offset = ((off_t *)data)[0]; 3122 length = ((off_t *)data)[1]; 3123 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 || 3124 offset < 0 || offset >= zv->zv_volsize || 3125 length <= 0) { 3126 printf("%s: offset=%jd length=%jd\n", __func__, offset, 3127 length); 3128 error = EINVAL; 3129 break; 3130 } 3131 3132 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER); 3133 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 3134 error = dmu_tx_assign(tx, TXG_WAIT); 3135 if (error != 0) { 3136 dmu_tx_abort(tx); 3137 } else { 3138 zvol_log_truncate(zv, tx, offset, length, B_TRUE); 3139 dmu_tx_commit(tx); 3140 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 3141 offset, length); 3142 } 3143 zfs_range_unlock(rl); 3144 if (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) 3145 zil_commit(zv->zv_zilog, ZVOL_OBJ); 3146 break; 3147 case DIOCGSTRIPESIZE: 3148 *(off_t *)data = zv->zv_volblocksize; 3149 break; 3150 case DIOCGSTRIPEOFFSET: 3151 *(off_t *)data = 0; 3152 break; 3153 case DIOCGATTR: { 3154 spa_t *spa = dmu_objset_spa(zv->zv_objset); 3155 struct diocgattr_arg *arg = (struct diocgattr_arg *)data; 3156 uint64_t refd, avail, usedobjs, availobjs; 3157 3158 if (strcmp(arg->name, "GEOM::candelete") == 0) 3159 arg->value.i = 1; 3160 else if (strcmp(arg->name, "blocksavail") == 0) { 3161 dmu_objset_space(zv->zv_objset, &refd, &avail, 3162 &usedobjs, &availobjs); 3163 arg->value.off = avail / DEV_BSIZE; 3164 } else if (strcmp(arg->name, "blocksused") == 0) { 3165 dmu_objset_space(zv->zv_objset, &refd, &avail, 3166 &usedobjs, &availobjs); 3167 arg->value.off = refd / DEV_BSIZE; 3168 } else if (strcmp(arg->name, "poolblocksavail") == 0) { 3169 avail = metaslab_class_get_space(spa_normal_class(spa)); 3170 avail -= metaslab_class_get_alloc(spa_normal_class(spa)); 3171 arg->value.off = avail / DEV_BSIZE; 3172 } else if (strcmp(arg->name, "poolblocksused") == 0) { 3173 refd = metaslab_class_get_alloc(spa_normal_class(spa)); 3174 arg->value.off = refd / DEV_BSIZE; 3175 } else 3176 error = ENOIOCTL; 3177 break; 3178 } 3179 case FIOSEEKHOLE: 3180 case FIOSEEKDATA: { 3181 off_t *off = (off_t *)data; 3182 uint64_t noff; 3183 boolean_t hole; 3184 3185 hole = (cmd == FIOSEEKHOLE); 3186 noff = *off; 3187 error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff); 3188 *off = noff; 3189 break; 3190 } 3191 default: 3192 error = ENOIOCTL; 3193 } 3194 3195 return (error); 3196} 3197#endif /* illumos */ 3198