zvol.c revision 297112
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 cdev *dev; 591 struct g_provider *pp; 592 struct g_geom *gp; 593 uint64_t volsize, mode; 594#endif 595 int error; 596 597#ifndef illumos 598 ZFS_LOG(1, "Creating ZVOL %s...", name); 599#endif 600 601 mutex_enter(&zfsdev_state_lock); 602 603 if (zvol_minor_lookup(name) != NULL) { 604 mutex_exit(&zfsdev_state_lock); 605 return (SET_ERROR(EEXIST)); 606 } 607 608 /* lie and say we're read-only */ 609 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os); 610 611 if (error) { 612 mutex_exit(&zfsdev_state_lock); 613 return (error); 614 } 615 616#ifdef illumos 617 if ((minor = zfsdev_minor_alloc()) == 0) { 618 dmu_objset_disown(os, FTAG); 619 mutex_exit(&zfsdev_state_lock); 620 return (SET_ERROR(ENXIO)); 621 } 622 623 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 624 dmu_objset_disown(os, FTAG); 625 mutex_exit(&zfsdev_state_lock); 626 return (SET_ERROR(EAGAIN)); 627 } 628 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 629 (char *)name); 630 631 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 632 633 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 634 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 635 ddi_soft_state_free(zfsdev_state, minor); 636 dmu_objset_disown(os, FTAG); 637 mutex_exit(&zfsdev_state_lock); 638 return (SET_ERROR(EAGAIN)); 639 } 640 641 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 642 643 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 644 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 645 ddi_remove_minor_node(zfs_dip, chrbuf); 646 ddi_soft_state_free(zfsdev_state, minor); 647 dmu_objset_disown(os, FTAG); 648 mutex_exit(&zfsdev_state_lock); 649 return (SET_ERROR(EAGAIN)); 650 } 651 652 zs = ddi_get_soft_state(zfsdev_state, minor); 653 zs->zss_type = ZSST_ZVOL; 654 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 655#else /* !illumos */ 656 657 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP); 658 zv->zv_state = 0; 659 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 660 if (error) { 661 kmem_free(zv, sizeof(*zv)); 662 dmu_objset_disown(os, zvol_tag); 663 mutex_exit(&zfsdev_state_lock); 664 return (error); 665 } 666 error = dsl_prop_get_integer(name, 667 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL); 668 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT) 669 mode = volmode; 670 671 DROP_GIANT(); 672 zv->zv_volsize = volsize; 673 zv->zv_volmode = mode; 674 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 675 g_topology_lock(); 676 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name); 677 gp->start = zvol_geom_start; 678 gp->access = zvol_geom_access; 679 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name); 680 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 681 pp->sectorsize = DEV_BSIZE; 682 pp->mediasize = zv->zv_volsize; 683 pp->private = zv; 684 685 zv->zv_provider = pp; 686 bioq_init(&zv->zv_queue); 687 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF); 688 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { 689 if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, 690 &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 691 0640, "%s/%s", ZVOL_DRIVER, name) != 0) { 692 kmem_free(zv, sizeof(*zv)); 693 dmu_objset_disown(os, FTAG); 694 mutex_exit(&zfsdev_state_lock); 695 return (SET_ERROR(ENXIO)); 696 } 697 zv->zv_dev = dev; 698 dev->si_iosize_max = MAXPHYS; 699 dev->si_drv2 = zv; 700 } 701 LIST_INSERT_HEAD(&all_zvols, zv, zv_links); 702#endif /* illumos */ 703 704 (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 705 zv->zv_min_bs = DEV_BSHIFT; 706#ifdef illumos 707 zv->zv_minor = minor; 708#endif 709 zv->zv_objset = os; 710 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) 711 zv->zv_flags |= ZVOL_RDONLY; 712 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 713 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 714 sizeof (rl_t), offsetof(rl_t, r_node)); 715 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 716 offsetof(zvol_extent_t, ze_node)); 717 /* get and cache the blocksize */ 718 error = dmu_object_info(os, ZVOL_OBJ, &doi); 719 ASSERT(error == 0); 720 zv->zv_volblocksize = doi.doi_data_block_size; 721 722 if (spa_writeable(dmu_objset_spa(os))) { 723 if (zil_replay_disable) 724 zil_destroy(dmu_objset_zil(os), B_FALSE); 725 else 726 zil_replay(os, zv, zvol_replay_vector); 727 } 728 dmu_objset_disown(os, FTAG); 729 zv->zv_objset = NULL; 730 731 zvol_minors++; 732 733 mutex_exit(&zfsdev_state_lock); 734#ifndef illumos 735 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 736 zvol_geom_run(zv); 737 g_topology_unlock(); 738 } 739 PICKUP_GIANT(); 740 741 ZFS_LOG(1, "ZVOL %s created.", name); 742#endif 743 744 return (0); 745} 746 747/* 748 * Remove minor node for the specified volume. 749 */ 750static int 751zvol_remove_zv(zvol_state_t *zv) 752{ 753#ifdef illumos 754 char nmbuf[20]; 755 minor_t minor = zv->zv_minor; 756#endif 757 758 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 759 if (zv->zv_total_opens != 0) 760 return (SET_ERROR(EBUSY)); 761 762#ifdef illumos 763 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 764 ddi_remove_minor_node(zfs_dip, nmbuf); 765 766 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor); 767 ddi_remove_minor_node(zfs_dip, nmbuf); 768#else 769 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); 770 771 LIST_REMOVE(zv, zv_links); 772 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 773 g_topology_lock(); 774 zvol_geom_destroy(zv); 775 g_topology_unlock(); 776 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) 777 destroy_dev(zv->zv_dev); 778#endif 779 780 avl_destroy(&zv->zv_znode.z_range_avl); 781 mutex_destroy(&zv->zv_znode.z_range_lock); 782 783 kmem_free(zv, sizeof (zvol_state_t)); 784#ifdef illumos 785 ddi_soft_state_free(zfsdev_state, minor); 786#endif 787 zvol_minors--; 788 return (0); 789} 790 791int 792zvol_remove_minor(const char *name) 793{ 794 zvol_state_t *zv; 795 int rc; 796 797 mutex_enter(&zfsdev_state_lock); 798 if ((zv = zvol_minor_lookup(name)) == NULL) { 799 mutex_exit(&zfsdev_state_lock); 800 return (SET_ERROR(ENXIO)); 801 } 802 rc = zvol_remove_zv(zv); 803 mutex_exit(&zfsdev_state_lock); 804 return (rc); 805} 806 807int 808zvol_first_open(zvol_state_t *zv) 809{ 810 objset_t *os; 811 uint64_t volsize; 812 int error; 813 uint64_t readonly; 814 815 /* lie and say we're read-only */ 816 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 817 zvol_tag, &os); 818 if (error) 819 return (error); 820 821 zv->zv_objset = os; 822 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 823 if (error) { 824 ASSERT(error == 0); 825 dmu_objset_disown(os, zvol_tag); 826 return (error); 827 } 828 829 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 830 if (error) { 831 dmu_objset_disown(os, zvol_tag); 832 return (error); 833 } 834 835 zvol_size_changed(zv, volsize); 836 zv->zv_zilog = zil_open(os, zvol_get_data); 837 838 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 839 NULL) == 0); 840 if (readonly || dmu_objset_is_snapshot(os) || 841 !spa_writeable(dmu_objset_spa(os))) 842 zv->zv_flags |= ZVOL_RDONLY; 843 else 844 zv->zv_flags &= ~ZVOL_RDONLY; 845 return (error); 846} 847 848void 849zvol_last_close(zvol_state_t *zv) 850{ 851 zil_close(zv->zv_zilog); 852 zv->zv_zilog = NULL; 853 854 dmu_buf_rele(zv->zv_dbuf, zvol_tag); 855 zv->zv_dbuf = NULL; 856 857 /* 858 * Evict cached data 859 */ 860 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && 861 !(zv->zv_flags & ZVOL_RDONLY)) 862 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 863 dmu_objset_evict_dbufs(zv->zv_objset); 864 865 dmu_objset_disown(zv->zv_objset, zvol_tag); 866 zv->zv_objset = NULL; 867} 868 869#ifdef illumos 870int 871zvol_prealloc(zvol_state_t *zv) 872{ 873 objset_t *os = zv->zv_objset; 874 dmu_tx_t *tx; 875 uint64_t refd, avail, usedobjs, availobjs; 876 uint64_t resid = zv->zv_volsize; 877 uint64_t off = 0; 878 879 /* Check the space usage before attempting to allocate the space */ 880 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 881 if (avail < zv->zv_volsize) 882 return (SET_ERROR(ENOSPC)); 883 884 /* Free old extents if they exist */ 885 zvol_free_extents(zv); 886 887 while (resid != 0) { 888 int error; 889 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE); 890 891 tx = dmu_tx_create(os); 892 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 893 error = dmu_tx_assign(tx, TXG_WAIT); 894 if (error) { 895 dmu_tx_abort(tx); 896 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 897 return (error); 898 } 899 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 900 dmu_tx_commit(tx); 901 off += bytes; 902 resid -= bytes; 903 } 904 txg_wait_synced(dmu_objset_pool(os), 0); 905 906 return (0); 907} 908#endif /* illumos */ 909 910static int 911zvol_update_volsize(objset_t *os, uint64_t volsize) 912{ 913 dmu_tx_t *tx; 914 int error; 915 916 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 917 918 tx = dmu_tx_create(os); 919 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 920 dmu_tx_mark_netfree(tx); 921 error = dmu_tx_assign(tx, TXG_WAIT); 922 if (error) { 923 dmu_tx_abort(tx); 924 return (error); 925 } 926 927 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 928 &volsize, tx); 929 dmu_tx_commit(tx); 930 931 if (error == 0) 932 error = dmu_free_long_range(os, 933 ZVOL_OBJ, volsize, DMU_OBJECT_END); 934 return (error); 935} 936 937void 938zvol_remove_minors(const char *name) 939{ 940#ifdef illumos 941 zvol_state_t *zv; 942 char *namebuf; 943 minor_t minor; 944 945 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP); 946 (void) strncpy(namebuf, name, strlen(name)); 947 (void) strcat(namebuf, "/"); 948 mutex_enter(&zfsdev_state_lock); 949 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) { 950 951 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 952 if (zv == NULL) 953 continue; 954 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0) 955 (void) zvol_remove_zv(zv); 956 } 957 kmem_free(namebuf, strlen(name) + 2); 958 959 mutex_exit(&zfsdev_state_lock); 960#else /* !illumos */ 961 zvol_state_t *zv, *tzv; 962 size_t namelen; 963 964 namelen = strlen(name); 965 966 DROP_GIANT(); 967 mutex_enter(&zfsdev_state_lock); 968 969 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) { 970 if (strcmp(zv->zv_name, name) == 0 || 971 (strncmp(zv->zv_name, name, namelen) == 0 && 972 strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' || 973 zv->zv_name[namelen] == '@'))) { 974 (void) zvol_remove_zv(zv); 975 } 976 } 977 978 mutex_exit(&zfsdev_state_lock); 979 PICKUP_GIANT(); 980#endif /* illumos */ 981} 982 983static int 984zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize) 985{ 986 uint64_t old_volsize = 0ULL; 987 int error = 0; 988 989 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 990 991 /* 992 * Reinitialize the dump area to the new size. If we 993 * failed to resize the dump area then restore it back to 994 * its original size. We must set the new volsize prior 995 * to calling dumpvp_resize() to ensure that the devices' 996 * size(9P) is not visible by the dump subsystem. 997 */ 998 old_volsize = zv->zv_volsize; 999 zvol_size_changed(zv, volsize); 1000 1001#ifdef ZVOL_DUMP 1002 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1003 if ((error = zvol_dumpify(zv)) != 0 || 1004 (error = dumpvp_resize()) != 0) { 1005 int dumpify_error; 1006 1007 (void) zvol_update_volsize(zv->zv_objset, old_volsize); 1008 zvol_size_changed(zv, old_volsize); 1009 dumpify_error = zvol_dumpify(zv); 1010 error = dumpify_error ? dumpify_error : error; 1011 } 1012 } 1013#endif /* ZVOL_DUMP */ 1014 1015#ifdef illumos 1016 /* 1017 * Generate a LUN expansion event. 1018 */ 1019 if (error == 0) { 1020 sysevent_id_t eid; 1021 nvlist_t *attr; 1022 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 1023 1024 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 1025 zv->zv_minor); 1026 1027 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1028 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 1029 1030 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 1031 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 1032 1033 nvlist_free(attr); 1034 kmem_free(physpath, MAXPATHLEN); 1035 } 1036#endif /* illumos */ 1037 return (error); 1038} 1039 1040int 1041zvol_set_volsize(const char *name, uint64_t volsize) 1042{ 1043 zvol_state_t *zv = NULL; 1044 objset_t *os; 1045 int error; 1046 dmu_object_info_t doi; 1047 uint64_t readonly; 1048 boolean_t owned = B_FALSE; 1049 1050 error = dsl_prop_get_integer(name, 1051 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL); 1052 if (error != 0) 1053 return (error); 1054 if (readonly) 1055 return (SET_ERROR(EROFS)); 1056 1057 mutex_enter(&zfsdev_state_lock); 1058 zv = zvol_minor_lookup(name); 1059 1060 if (zv == NULL || zv->zv_objset == NULL) { 1061 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, 1062 FTAG, &os)) != 0) { 1063 mutex_exit(&zfsdev_state_lock); 1064 return (error); 1065 } 1066 owned = B_TRUE; 1067 if (zv != NULL) 1068 zv->zv_objset = os; 1069 } else { 1070 os = zv->zv_objset; 1071 } 1072 1073 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 1074 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0) 1075 goto out; 1076 1077 error = zvol_update_volsize(os, volsize); 1078 1079 if (error == 0 && zv != NULL) 1080 error = zvol_update_live_volsize(zv, volsize); 1081out: 1082 if (owned) { 1083 dmu_objset_disown(os, FTAG); 1084 if (zv != NULL) 1085 zv->zv_objset = NULL; 1086 } 1087 mutex_exit(&zfsdev_state_lock); 1088 return (error); 1089} 1090 1091/*ARGSUSED*/ 1092#ifdef illumos 1093int 1094zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr) 1095#else 1096static int 1097zvol_open(struct g_provider *pp, int flag, int count) 1098#endif 1099{ 1100 zvol_state_t *zv; 1101 int err = 0; 1102#ifdef illumos 1103 1104 mutex_enter(&zfsdev_state_lock); 1105 1106 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL); 1107 if (zv == NULL) { 1108 mutex_exit(&zfsdev_state_lock); 1109 return (SET_ERROR(ENXIO)); 1110 } 1111 1112 if (zv->zv_total_opens == 0) 1113 err = zvol_first_open(zv); 1114 if (err) { 1115 mutex_exit(&zfsdev_state_lock); 1116 return (err); 1117 } 1118#else /* !illumos */ 1119 boolean_t locked = B_FALSE; 1120 1121 /* 1122 * Protect against recursively entering spa_namespace_lock 1123 * when spa_open() is used for a pool on a (local) ZVOL(s). 1124 * This is needed since we replaced upstream zfsdev_state_lock 1125 * with spa_namespace_lock in the ZVOL code. 1126 * We are using the same trick as spa_open(). 1127 * Note that calls in zvol_first_open which need to resolve 1128 * pool name to a spa object will enter spa_open() 1129 * recursively, but that function already has all the 1130 * necessary protection. 1131 */ 1132 if (!MUTEX_HELD(&zfsdev_state_lock)) { 1133 mutex_enter(&zfsdev_state_lock); 1134 locked = B_TRUE; 1135 } 1136 1137 zv = pp->private; 1138 if (zv == NULL) { 1139 if (locked) 1140 mutex_exit(&zfsdev_state_lock); 1141 return (SET_ERROR(ENXIO)); 1142 } 1143 1144 if (zv->zv_total_opens == 0) { 1145 err = zvol_first_open(zv); 1146 if (err) { 1147 if (locked) 1148 mutex_exit(&zfsdev_state_lock); 1149 return (err); 1150 } 1151 pp->mediasize = zv->zv_volsize; 1152 pp->stripeoffset = 0; 1153 pp->stripesize = zv->zv_volblocksize; 1154 } 1155#endif /* illumos */ 1156 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 1157 err = SET_ERROR(EROFS); 1158 goto out; 1159 } 1160 if (zv->zv_flags & ZVOL_EXCL) { 1161 err = SET_ERROR(EBUSY); 1162 goto out; 1163 } 1164#ifdef FEXCL 1165 if (flag & FEXCL) { 1166 if (zv->zv_total_opens != 0) { 1167 err = SET_ERROR(EBUSY); 1168 goto out; 1169 } 1170 zv->zv_flags |= ZVOL_EXCL; 1171 } 1172#endif 1173 1174#ifdef illumos 1175 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) { 1176 zv->zv_open_count[otyp]++; 1177 zv->zv_total_opens++; 1178 } 1179 mutex_exit(&zfsdev_state_lock); 1180#else 1181 zv->zv_total_opens += count; 1182 if (locked) 1183 mutex_exit(&zfsdev_state_lock); 1184#endif 1185 1186 return (err); 1187out: 1188 if (zv->zv_total_opens == 0) 1189 zvol_last_close(zv); 1190#ifdef illumos 1191 mutex_exit(&zfsdev_state_lock); 1192#else 1193 if (locked) 1194 mutex_exit(&zfsdev_state_lock); 1195#endif 1196 return (err); 1197} 1198 1199/*ARGSUSED*/ 1200#ifdef illumos 1201int 1202zvol_close(dev_t dev, int flag, int otyp, cred_t *cr) 1203{ 1204 minor_t minor = getminor(dev); 1205 zvol_state_t *zv; 1206 int error = 0; 1207 1208 mutex_enter(&zfsdev_state_lock); 1209 1210 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1211 if (zv == NULL) { 1212 mutex_exit(&zfsdev_state_lock); 1213#else /* !illumos */ 1214static int 1215zvol_close(struct g_provider *pp, int flag, int count) 1216{ 1217 zvol_state_t *zv; 1218 int error = 0; 1219 boolean_t locked = B_FALSE; 1220 1221 /* See comment in zvol_open(). */ 1222 if (!MUTEX_HELD(&zfsdev_state_lock)) { 1223 mutex_enter(&zfsdev_state_lock); 1224 locked = B_TRUE; 1225 } 1226 1227 zv = pp->private; 1228 if (zv == NULL) { 1229 if (locked) 1230 mutex_exit(&zfsdev_state_lock); 1231#endif /* illumos */ 1232 return (SET_ERROR(ENXIO)); 1233 } 1234 1235 if (zv->zv_flags & ZVOL_EXCL) { 1236 ASSERT(zv->zv_total_opens == 1); 1237 zv->zv_flags &= ~ZVOL_EXCL; 1238 } 1239 1240 /* 1241 * If the open count is zero, this is a spurious close. 1242 * That indicates a bug in the kernel / DDI framework. 1243 */ 1244#ifdef illumos 1245 ASSERT(zv->zv_open_count[otyp] != 0); 1246#endif 1247 ASSERT(zv->zv_total_opens != 0); 1248 1249 /* 1250 * You may get multiple opens, but only one close. 1251 */ 1252#ifdef illumos 1253 zv->zv_open_count[otyp]--; 1254 zv->zv_total_opens--; 1255#else 1256 zv->zv_total_opens -= count; 1257#endif 1258 1259 if (zv->zv_total_opens == 0) 1260 zvol_last_close(zv); 1261 1262#ifdef illumos 1263 mutex_exit(&zfsdev_state_lock); 1264#else 1265 if (locked) 1266 mutex_exit(&zfsdev_state_lock); 1267#endif 1268 return (error); 1269} 1270 1271static void 1272zvol_get_done(zgd_t *zgd, int error) 1273{ 1274 if (zgd->zgd_db) 1275 dmu_buf_rele(zgd->zgd_db, zgd); 1276 1277 zfs_range_unlock(zgd->zgd_rl); 1278 1279 if (error == 0 && zgd->zgd_bp) 1280 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1281 1282 kmem_free(zgd, sizeof (zgd_t)); 1283} 1284 1285/* 1286 * Get data to generate a TX_WRITE intent log record. 1287 */ 1288static int 1289zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1290{ 1291 zvol_state_t *zv = arg; 1292 objset_t *os = zv->zv_objset; 1293 uint64_t object = ZVOL_OBJ; 1294 uint64_t offset = lr->lr_offset; 1295 uint64_t size = lr->lr_length; /* length of user data */ 1296 blkptr_t *bp = &lr->lr_blkptr; 1297 dmu_buf_t *db; 1298 zgd_t *zgd; 1299 int error; 1300 1301 ASSERT(zio != NULL); 1302 ASSERT(size != 0); 1303 1304 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1305 zgd->zgd_zilog = zv->zv_zilog; 1306 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 1307 1308 /* 1309 * Write records come in two flavors: immediate and indirect. 1310 * For small writes it's cheaper to store the data with the 1311 * log record (immediate); for large writes it's cheaper to 1312 * sync the data and get a pointer to it (indirect) so that 1313 * we don't have to write the data twice. 1314 */ 1315 if (buf != NULL) { /* immediate write */ 1316 error = dmu_read(os, object, offset, size, buf, 1317 DMU_READ_NO_PREFETCH); 1318 } else { 1319 size = zv->zv_volblocksize; 1320 offset = P2ALIGN(offset, size); 1321 error = dmu_buf_hold(os, object, offset, zgd, &db, 1322 DMU_READ_NO_PREFETCH); 1323 if (error == 0) { 1324 blkptr_t *obp = dmu_buf_get_blkptr(db); 1325 if (obp) { 1326 ASSERT(BP_IS_HOLE(bp)); 1327 *bp = *obp; 1328 } 1329 1330 zgd->zgd_db = db; 1331 zgd->zgd_bp = bp; 1332 1333 ASSERT(db->db_offset == offset); 1334 ASSERT(db->db_size == size); 1335 1336 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1337 zvol_get_done, zgd); 1338 1339 if (error == 0) 1340 return (0); 1341 } 1342 } 1343 1344 zvol_get_done(zgd, error); 1345 1346 return (error); 1347} 1348 1349/* 1350 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 1351 * 1352 * We store data in the log buffers if it's small enough. 1353 * Otherwise we will later flush the data out via dmu_sync(). 1354 */ 1355ssize_t zvol_immediate_write_sz = 32768; 1356 1357static void 1358zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1359 boolean_t sync) 1360{ 1361 uint32_t blocksize = zv->zv_volblocksize; 1362 zilog_t *zilog = zv->zv_zilog; 1363 boolean_t slogging; 1364 ssize_t immediate_write_sz; 1365 1366 if (zil_replaying(zilog, tx)) 1367 return; 1368 1369 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1370 ? 0 : zvol_immediate_write_sz; 1371 1372 slogging = spa_has_slogs(zilog->zl_spa) && 1373 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 1374 1375 while (resid) { 1376 itx_t *itx; 1377 lr_write_t *lr; 1378 ssize_t len; 1379 itx_wr_state_t write_state; 1380 1381 /* 1382 * Unlike zfs_log_write() we can be called with 1383 * upto DMU_MAX_ACCESS/2 (5MB) writes. 1384 */ 1385 if (blocksize > immediate_write_sz && !slogging && 1386 resid >= blocksize && off % blocksize == 0) { 1387 write_state = WR_INDIRECT; /* uses dmu_sync */ 1388 len = blocksize; 1389 } else if (sync) { 1390 write_state = WR_COPIED; 1391 len = MIN(ZIL_MAX_LOG_DATA, resid); 1392 } else { 1393 write_state = WR_NEED_COPY; 1394 len = MIN(ZIL_MAX_LOG_DATA, resid); 1395 } 1396 1397 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1398 (write_state == WR_COPIED ? len : 0)); 1399 lr = (lr_write_t *)&itx->itx_lr; 1400 if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 1401 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1402 zil_itx_destroy(itx); 1403 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1404 lr = (lr_write_t *)&itx->itx_lr; 1405 write_state = WR_NEED_COPY; 1406 } 1407 1408 itx->itx_wr_state = write_state; 1409 if (write_state == WR_NEED_COPY) 1410 itx->itx_sod += len; 1411 lr->lr_foid = ZVOL_OBJ; 1412 lr->lr_offset = off; 1413 lr->lr_length = len; 1414 lr->lr_blkoff = 0; 1415 BP_ZERO(&lr->lr_blkptr); 1416 1417 itx->itx_private = zv; 1418 itx->itx_sync = sync; 1419 1420 zil_itx_assign(zilog, itx, tx); 1421 1422 off += len; 1423 resid -= len; 1424 } 1425} 1426 1427#ifdef illumos 1428static int 1429zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset, 1430 uint64_t size, boolean_t doread, boolean_t isdump) 1431{ 1432 vdev_disk_t *dvd; 1433 int c; 1434 int numerrors = 0; 1435 1436 if (vd->vdev_ops == &vdev_mirror_ops || 1437 vd->vdev_ops == &vdev_replacing_ops || 1438 vd->vdev_ops == &vdev_spare_ops) { 1439 for (c = 0; c < vd->vdev_children; c++) { 1440 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1441 addr, offset, origoffset, size, doread, isdump); 1442 if (err != 0) { 1443 numerrors++; 1444 } else if (doread) { 1445 break; 1446 } 1447 } 1448 } 1449 1450 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops) 1451 return (numerrors < vd->vdev_children ? 0 : EIO); 1452 1453 if (doread && !vdev_readable(vd)) 1454 return (SET_ERROR(EIO)); 1455 else if (!doread && !vdev_writeable(vd)) 1456 return (SET_ERROR(EIO)); 1457 1458 if (vd->vdev_ops == &vdev_raidz_ops) { 1459 return (vdev_raidz_physio(vd, 1460 addr, size, offset, origoffset, doread, isdump)); 1461 } 1462 1463 offset += VDEV_LABEL_START_SIZE; 1464 1465 if (ddi_in_panic() || isdump) { 1466 ASSERT(!doread); 1467 if (doread) 1468 return (SET_ERROR(EIO)); 1469 dvd = vd->vdev_tsd; 1470 ASSERT3P(dvd, !=, NULL); 1471 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1472 lbtodb(size))); 1473 } else { 1474 dvd = vd->vdev_tsd; 1475 ASSERT3P(dvd, !=, NULL); 1476 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size, 1477 offset, doread ? B_READ : B_WRITE)); 1478 } 1479} 1480 1481static int 1482zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1483 boolean_t doread, boolean_t isdump) 1484{ 1485 vdev_t *vd; 1486 int error; 1487 zvol_extent_t *ze; 1488 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1489 1490 /* Must be sector aligned, and not stradle a block boundary. */ 1491 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1492 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1493 return (SET_ERROR(EINVAL)); 1494 } 1495 ASSERT(size <= zv->zv_volblocksize); 1496 1497 /* Locate the extent this belongs to */ 1498 ze = list_head(&zv->zv_extents); 1499 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1500 offset -= ze->ze_nblks * zv->zv_volblocksize; 1501 ze = list_next(&zv->zv_extents, ze); 1502 } 1503 1504 if (ze == NULL) 1505 return (SET_ERROR(EINVAL)); 1506 1507 if (!ddi_in_panic()) 1508 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1509 1510 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1511 offset += DVA_GET_OFFSET(&ze->ze_dva); 1512 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva), 1513 size, doread, isdump); 1514 1515 if (!ddi_in_panic()) 1516 spa_config_exit(spa, SCL_STATE, FTAG); 1517 1518 return (error); 1519} 1520 1521int 1522zvol_strategy(buf_t *bp) 1523{ 1524 zfs_soft_state_t *zs = NULL; 1525#else /* !illumos */ 1526void 1527zvol_strategy(struct bio *bp) 1528{ 1529#endif /* illumos */ 1530 zvol_state_t *zv; 1531 uint64_t off, volsize; 1532 size_t resid; 1533 char *addr; 1534 objset_t *os; 1535 rl_t *rl; 1536 int error = 0; 1537#ifdef illumos 1538 boolean_t doread = bp->b_flags & B_READ; 1539#else 1540 boolean_t doread = 0; 1541#endif 1542 boolean_t is_dumpified; 1543 boolean_t sync; 1544 1545#ifdef illumos 1546 if (getminor(bp->b_edev) == 0) { 1547 error = SET_ERROR(EINVAL); 1548 } else { 1549 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev)); 1550 if (zs == NULL) 1551 error = SET_ERROR(ENXIO); 1552 else if (zs->zss_type != ZSST_ZVOL) 1553 error = SET_ERROR(EINVAL); 1554 } 1555 1556 if (error) { 1557 bioerror(bp, error); 1558 biodone(bp); 1559 return (0); 1560 } 1561 1562 zv = zs->zss_data; 1563 1564 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) { 1565 bioerror(bp, EROFS); 1566 biodone(bp); 1567 return (0); 1568 } 1569 1570 off = ldbtob(bp->b_blkno); 1571#else /* !illumos */ 1572 if (bp->bio_to) 1573 zv = bp->bio_to->private; 1574 else 1575 zv = bp->bio_dev->si_drv2; 1576 1577 if (zv == NULL) { 1578 error = SET_ERROR(ENXIO); 1579 goto out; 1580 } 1581 1582 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) { 1583 error = SET_ERROR(EROFS); 1584 goto out; 1585 } 1586 1587 switch (bp->bio_cmd) { 1588 case BIO_FLUSH: 1589 goto sync; 1590 case BIO_READ: 1591 doread = 1; 1592 case BIO_WRITE: 1593 case BIO_DELETE: 1594 break; 1595 default: 1596 error = EOPNOTSUPP; 1597 goto out; 1598 } 1599 1600 off = bp->bio_offset; 1601#endif /* illumos */ 1602 volsize = zv->zv_volsize; 1603 1604 os = zv->zv_objset; 1605 ASSERT(os != NULL); 1606 1607#ifdef illumos 1608 bp_mapin(bp); 1609 addr = bp->b_un.b_addr; 1610 resid = bp->b_bcount; 1611 1612 if (resid > 0 && (off < 0 || off >= volsize)) { 1613 bioerror(bp, EIO); 1614 biodone(bp); 1615 return (0); 1616 } 1617 1618 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED; 1619 sync = ((!(bp->b_flags & B_ASYNC) && 1620 !(zv->zv_flags & ZVOL_WCE)) || 1621 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) && 1622 !doread && !is_dumpified; 1623#else /* !illumos */ 1624 addr = bp->bio_data; 1625 resid = bp->bio_length; 1626 1627 if (resid > 0 && (off < 0 || off >= volsize)) { 1628 error = SET_ERROR(EIO); 1629 goto out; 1630 } 1631 1632 is_dumpified = B_FALSE; 1633 sync = !doread && !is_dumpified && 1634 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; 1635#endif /* illumos */ 1636 1637 /* 1638 * There must be no buffer changes when doing a dmu_sync() because 1639 * we can't change the data whilst calculating the checksum. 1640 */ 1641 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1642 doread ? RL_READER : RL_WRITER); 1643 1644#ifndef illumos 1645 if (bp->bio_cmd == BIO_DELETE) { 1646 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1647 error = dmu_tx_assign(tx, TXG_WAIT); 1648 if (error != 0) { 1649 dmu_tx_abort(tx); 1650 } else { 1651 zvol_log_truncate(zv, tx, off, resid, B_TRUE); 1652 dmu_tx_commit(tx); 1653 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1654 off, resid); 1655 resid = 0; 1656 } 1657 goto unlock; 1658 } 1659#endif 1660 while (resid != 0 && off < volsize) { 1661 size_t size = MIN(resid, zvol_maxphys); 1662#ifdef illumos 1663 if (is_dumpified) { 1664 size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 1665 error = zvol_dumpio(zv, addr, off, size, 1666 doread, B_FALSE); 1667 } else if (doread) { 1668#else 1669 if (doread) { 1670#endif 1671 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1672 DMU_READ_PREFETCH); 1673 } else { 1674 dmu_tx_t *tx = dmu_tx_create(os); 1675 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1676 error = dmu_tx_assign(tx, TXG_WAIT); 1677 if (error) { 1678 dmu_tx_abort(tx); 1679 } else { 1680 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1681 zvol_log_write(zv, tx, off, size, sync); 1682 dmu_tx_commit(tx); 1683 } 1684 } 1685 if (error) { 1686 /* convert checksum errors into IO errors */ 1687 if (error == ECKSUM) 1688 error = SET_ERROR(EIO); 1689 break; 1690 } 1691 off += size; 1692 addr += size; 1693 resid -= size; 1694 } 1695#ifndef illumos 1696unlock: 1697#endif 1698 zfs_range_unlock(rl); 1699 1700#ifdef illumos 1701 if ((bp->b_resid = resid) == bp->b_bcount) 1702 bioerror(bp, off > volsize ? EINVAL : error); 1703 1704 if (sync) 1705 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1706 biodone(bp); 1707 1708 return (0); 1709#else /* !illumos */ 1710 bp->bio_completed = bp->bio_length - resid; 1711 if (bp->bio_completed < bp->bio_length && off > volsize) 1712 error = EINVAL; 1713 1714 if (sync) { 1715sync: 1716 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1717 } 1718out: 1719 if (bp->bio_to) 1720 g_io_deliver(bp, error); 1721 else 1722 biofinish(bp, NULL, error); 1723#endif /* illumos */ 1724} 1725 1726#ifdef illumos 1727/* 1728 * Set the buffer count to the zvol maximum transfer. 1729 * Using our own routine instead of the default minphys() 1730 * means that for larger writes we write bigger buffers on X86 1731 * (128K instead of 56K) and flush the disk write cache less often 1732 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1733 * 56K on X86 and 128K on sparc). 1734 */ 1735void 1736zvol_minphys(struct buf *bp) 1737{ 1738 if (bp->b_bcount > zvol_maxphys) 1739 bp->b_bcount = zvol_maxphys; 1740} 1741 1742int 1743zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1744{ 1745 minor_t minor = getminor(dev); 1746 zvol_state_t *zv; 1747 int error = 0; 1748 uint64_t size; 1749 uint64_t boff; 1750 uint64_t resid; 1751 1752 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1753 if (zv == NULL) 1754 return (SET_ERROR(ENXIO)); 1755 1756 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) 1757 return (SET_ERROR(EINVAL)); 1758 1759 boff = ldbtob(blkno); 1760 resid = ldbtob(nblocks); 1761 1762 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1763 1764 while (resid) { 1765 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1766 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1767 if (error) 1768 break; 1769 boff += size; 1770 addr += size; 1771 resid -= size; 1772 } 1773 1774 return (error); 1775} 1776 1777/*ARGSUSED*/ 1778int 1779zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1780{ 1781 minor_t minor = getminor(dev); 1782#else /* !illumos */ 1783int 1784zvol_read(struct cdev *dev, struct uio *uio, int ioflag) 1785{ 1786#endif /* illumos */ 1787 zvol_state_t *zv; 1788 uint64_t volsize; 1789 rl_t *rl; 1790 int error = 0; 1791 1792#ifdef illumos 1793 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1794 if (zv == NULL) 1795 return (SET_ERROR(ENXIO)); 1796#else 1797 zv = dev->si_drv2; 1798#endif 1799 1800 volsize = zv->zv_volsize; 1801 /* uio_loffset == volsize isn't an error as its required for EOF processing. */ 1802 if (uio->uio_resid > 0 && 1803 (uio->uio_loffset < 0 || uio->uio_loffset > volsize)) 1804 return (SET_ERROR(EIO)); 1805 1806#ifdef illumos 1807 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1808 error = physio(zvol_strategy, NULL, dev, B_READ, 1809 zvol_minphys, uio); 1810 return (error); 1811 } 1812#endif 1813 1814 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1815 RL_READER); 1816 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1817 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1818 1819 /* don't read past the end */ 1820 if (bytes > volsize - uio->uio_loffset) 1821 bytes = volsize - uio->uio_loffset; 1822 1823 error = dmu_read_uio_dbuf(zv->zv_dbuf, uio, bytes); 1824 if (error) { 1825 /* convert checksum errors into IO errors */ 1826 if (error == ECKSUM) 1827 error = SET_ERROR(EIO); 1828 break; 1829 } 1830 } 1831 zfs_range_unlock(rl); 1832 return (error); 1833} 1834 1835#ifdef illumos 1836/*ARGSUSED*/ 1837int 1838zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1839{ 1840 minor_t minor = getminor(dev); 1841#else /* !illumos */ 1842int 1843zvol_write(struct cdev *dev, struct uio *uio, int ioflag) 1844{ 1845#endif /* illumos */ 1846 zvol_state_t *zv; 1847 uint64_t volsize; 1848 rl_t *rl; 1849 int error = 0; 1850 boolean_t sync; 1851 1852#ifdef illumos 1853 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1854 if (zv == NULL) 1855 return (SET_ERROR(ENXIO)); 1856#else 1857 zv = dev->si_drv2; 1858#endif 1859 1860 volsize = zv->zv_volsize; 1861 /* uio_loffset == volsize isn't an error as its required for EOF processing. */ 1862 if (uio->uio_resid > 0 && 1863 (uio->uio_loffset < 0 || uio->uio_loffset > volsize)) 1864 return (SET_ERROR(EIO)); 1865 1866#ifdef illumos 1867 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1868 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1869 zvol_minphys, uio); 1870 return (error); 1871 } 1872 1873 sync = !(zv->zv_flags & ZVOL_WCE) || 1874#else 1875 sync = (ioflag & IO_SYNC) || 1876#endif 1877 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1878 1879 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1880 RL_WRITER); 1881 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1882 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1883 uint64_t off = uio->uio_loffset; 1884 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1885 1886 if (bytes > volsize - off) /* don't write past the end */ 1887 bytes = volsize - off; 1888 1889 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1890 error = dmu_tx_assign(tx, TXG_WAIT); 1891 if (error) { 1892 dmu_tx_abort(tx); 1893 break; 1894 } 1895 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1896 if (error == 0) 1897 zvol_log_write(zv, tx, off, bytes, sync); 1898 dmu_tx_commit(tx); 1899 1900 if (error) 1901 break; 1902 } 1903 zfs_range_unlock(rl); 1904 if (sync) 1905 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1906 return (error); 1907} 1908 1909#ifdef illumos 1910int 1911zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1912{ 1913 struct uuid uuid = EFI_RESERVED; 1914 efi_gpe_t gpe = { 0 }; 1915 uint32_t crc; 1916 dk_efi_t efi; 1917 int length; 1918 char *ptr; 1919 1920 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1921 return (SET_ERROR(EFAULT)); 1922 ptr = (char *)(uintptr_t)efi.dki_data_64; 1923 length = efi.dki_length; 1924 /* 1925 * Some clients may attempt to request a PMBR for the 1926 * zvol. Currently this interface will return EINVAL to 1927 * such requests. These requests could be supported by 1928 * adding a check for lba == 0 and consing up an appropriate 1929 * PMBR. 1930 */ 1931 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1932 return (SET_ERROR(EINVAL)); 1933 1934 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1935 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1936 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1937 1938 if (efi.dki_lba == 1) { 1939 efi_gpt_t gpt = { 0 }; 1940 1941 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1942 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1943 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1944 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1945 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1946 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1947 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1948 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1949 gpt.efi_gpt_SizeOfPartitionEntry = 1950 LE_32(sizeof (efi_gpe_t)); 1951 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1952 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1953 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1954 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1955 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1956 flag)) 1957 return (SET_ERROR(EFAULT)); 1958 ptr += sizeof (gpt); 1959 length -= sizeof (gpt); 1960 } 1961 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1962 length), flag)) 1963 return (SET_ERROR(EFAULT)); 1964 return (0); 1965} 1966 1967/* 1968 * BEGIN entry points to allow external callers access to the volume. 1969 */ 1970/* 1971 * Return the volume parameters needed for access from an external caller. 1972 * These values are invariant as long as the volume is held open. 1973 */ 1974int 1975zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1976 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1977 void **rl_hdl, void **bonus_hdl) 1978{ 1979 zvol_state_t *zv; 1980 1981 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1982 if (zv == NULL) 1983 return (SET_ERROR(ENXIO)); 1984 if (zv->zv_flags & ZVOL_DUMPIFIED) 1985 return (SET_ERROR(ENXIO)); 1986 1987 ASSERT(blksize && max_xfer_len && minor_hdl && 1988 objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 1989 1990 *blksize = zv->zv_volblocksize; 1991 *max_xfer_len = (uint64_t)zvol_maxphys; 1992 *minor_hdl = zv; 1993 *objset_hdl = zv->zv_objset; 1994 *zil_hdl = zv->zv_zilog; 1995 *rl_hdl = &zv->zv_znode; 1996 *bonus_hdl = zv->zv_dbuf; 1997 return (0); 1998} 1999 2000/* 2001 * Return the current volume size to an external caller. 2002 * The size can change while the volume is open. 2003 */ 2004uint64_t 2005zvol_get_volume_size(void *minor_hdl) 2006{ 2007 zvol_state_t *zv = minor_hdl; 2008 2009 return (zv->zv_volsize); 2010} 2011 2012/* 2013 * Return the current WCE setting to an external caller. 2014 * The WCE setting can change while the volume is open. 2015 */ 2016int 2017zvol_get_volume_wce(void *minor_hdl) 2018{ 2019 zvol_state_t *zv = minor_hdl; 2020 2021 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 2022} 2023 2024/* 2025 * Entry point for external callers to zvol_log_write 2026 */ 2027void 2028zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 2029 boolean_t sync) 2030{ 2031 zvol_state_t *zv = minor_hdl; 2032 2033 zvol_log_write(zv, tx, off, resid, sync); 2034} 2035/* 2036 * END entry points to allow external callers access to the volume. 2037 */ 2038#endif /* illumos */ 2039 2040/* 2041 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. 2042 */ 2043static void 2044zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, 2045 boolean_t sync) 2046{ 2047 itx_t *itx; 2048 lr_truncate_t *lr; 2049 zilog_t *zilog = zv->zv_zilog; 2050 2051 if (zil_replaying(zilog, tx)) 2052 return; 2053 2054 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 2055 lr = (lr_truncate_t *)&itx->itx_lr; 2056 lr->lr_foid = ZVOL_OBJ; 2057 lr->lr_offset = off; 2058 lr->lr_length = len; 2059 2060 itx->itx_sync = sync; 2061 zil_itx_assign(zilog, itx, tx); 2062} 2063 2064#ifdef illumos 2065/* 2066 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 2067 * Also a dirtbag dkio ioctl for unmap/free-block functionality. 2068 */ 2069/*ARGSUSED*/ 2070int 2071zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 2072{ 2073 zvol_state_t *zv; 2074 struct dk_callback *dkc; 2075 int error = 0; 2076 rl_t *rl; 2077 2078 mutex_enter(&zfsdev_state_lock); 2079 2080 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 2081 2082 if (zv == NULL) { 2083 mutex_exit(&zfsdev_state_lock); 2084 return (SET_ERROR(ENXIO)); 2085 } 2086 ASSERT(zv->zv_total_opens > 0); 2087 2088 switch (cmd) { 2089 2090 case DKIOCINFO: 2091 { 2092 struct dk_cinfo dki; 2093 2094 bzero(&dki, sizeof (dki)); 2095 (void) strcpy(dki.dki_cname, "zvol"); 2096 (void) strcpy(dki.dki_dname, "zvol"); 2097 dki.dki_ctype = DKC_UNKNOWN; 2098 dki.dki_unit = getminor(dev); 2099 dki.dki_maxtransfer = 2100 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs); 2101 mutex_exit(&zfsdev_state_lock); 2102 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 2103 error = SET_ERROR(EFAULT); 2104 return (error); 2105 } 2106 2107 case DKIOCGMEDIAINFO: 2108 { 2109 struct dk_minfo dkm; 2110 2111 bzero(&dkm, sizeof (dkm)); 2112 dkm.dki_lbsize = 1U << zv->zv_min_bs; 2113 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 2114 dkm.dki_media_type = DK_UNKNOWN; 2115 mutex_exit(&zfsdev_state_lock); 2116 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 2117 error = SET_ERROR(EFAULT); 2118 return (error); 2119 } 2120 2121 case DKIOCGMEDIAINFOEXT: 2122 { 2123 struct dk_minfo_ext dkmext; 2124 2125 bzero(&dkmext, sizeof (dkmext)); 2126 dkmext.dki_lbsize = 1U << zv->zv_min_bs; 2127 dkmext.dki_pbsize = zv->zv_volblocksize; 2128 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 2129 dkmext.dki_media_type = DK_UNKNOWN; 2130 mutex_exit(&zfsdev_state_lock); 2131 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag)) 2132 error = SET_ERROR(EFAULT); 2133 return (error); 2134 } 2135 2136 case DKIOCGETEFI: 2137 { 2138 uint64_t vs = zv->zv_volsize; 2139 uint8_t bs = zv->zv_min_bs; 2140 2141 mutex_exit(&zfsdev_state_lock); 2142 error = zvol_getefi((void *)arg, flag, vs, bs); 2143 return (error); 2144 } 2145 2146 case DKIOCFLUSHWRITECACHE: 2147 dkc = (struct dk_callback *)arg; 2148 mutex_exit(&zfsdev_state_lock); 2149 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2150 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 2151 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 2152 error = 0; 2153 } 2154 return (error); 2155 2156 case DKIOCGETWCE: 2157 { 2158 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 2159 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 2160 flag)) 2161 error = SET_ERROR(EFAULT); 2162 break; 2163 } 2164 case DKIOCSETWCE: 2165 { 2166 int wce; 2167 if (ddi_copyin((void *)arg, &wce, sizeof (int), 2168 flag)) { 2169 error = SET_ERROR(EFAULT); 2170 break; 2171 } 2172 if (wce) { 2173 zv->zv_flags |= ZVOL_WCE; 2174 mutex_exit(&zfsdev_state_lock); 2175 } else { 2176 zv->zv_flags &= ~ZVOL_WCE; 2177 mutex_exit(&zfsdev_state_lock); 2178 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2179 } 2180 return (0); 2181 } 2182 2183 case DKIOCGGEOM: 2184 case DKIOCGVTOC: 2185 /* 2186 * commands using these (like prtvtoc) expect ENOTSUP 2187 * since we're emulating an EFI label 2188 */ 2189 error = SET_ERROR(ENOTSUP); 2190 break; 2191 2192 case DKIOCDUMPINIT: 2193 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 2194 RL_WRITER); 2195 error = zvol_dumpify(zv); 2196 zfs_range_unlock(rl); 2197 break; 2198 2199 case DKIOCDUMPFINI: 2200 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 2201 break; 2202 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 2203 RL_WRITER); 2204 error = zvol_dump_fini(zv); 2205 zfs_range_unlock(rl); 2206 break; 2207 2208 case DKIOCFREE: 2209 { 2210 dkioc_free_t df; 2211 dmu_tx_t *tx; 2212 2213 if (!zvol_unmap_enabled) 2214 break; 2215 2216 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) { 2217 error = SET_ERROR(EFAULT); 2218 break; 2219 } 2220 2221 /* 2222 * Apply Postel's Law to length-checking. If they overshoot, 2223 * just blank out until the end, if there's a need to blank 2224 * out anything. 2225 */ 2226 if (df.df_start >= zv->zv_volsize) 2227 break; /* No need to do anything... */ 2228 2229 mutex_exit(&zfsdev_state_lock); 2230 2231 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length, 2232 RL_WRITER); 2233 tx = dmu_tx_create(zv->zv_objset); 2234 dmu_tx_mark_netfree(tx); 2235 error = dmu_tx_assign(tx, TXG_WAIT); 2236 if (error != 0) { 2237 dmu_tx_abort(tx); 2238 } else { 2239 zvol_log_truncate(zv, tx, df.df_start, 2240 df.df_length, B_TRUE); 2241 dmu_tx_commit(tx); 2242 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 2243 df.df_start, df.df_length); 2244 } 2245 2246 zfs_range_unlock(rl); 2247 2248 if (error == 0) { 2249 /* 2250 * If the write-cache is disabled or 'sync' property 2251 * is set to 'always' then treat this as a synchronous 2252 * operation (i.e. commit to zil). 2253 */ 2254 if (!(zv->zv_flags & ZVOL_WCE) || 2255 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) 2256 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2257 2258 /* 2259 * If the caller really wants synchronous writes, and 2260 * can't wait for them, don't return until the write 2261 * is done. 2262 */ 2263 if (df.df_flags & DF_WAIT_SYNC) { 2264 txg_wait_synced( 2265 dmu_objset_pool(zv->zv_objset), 0); 2266 } 2267 } 2268 return (error); 2269 } 2270 2271 default: 2272 error = SET_ERROR(ENOTTY); 2273 break; 2274 2275 } 2276 mutex_exit(&zfsdev_state_lock); 2277 return (error); 2278} 2279#endif /* illumos */ 2280 2281int 2282zvol_busy(void) 2283{ 2284 return (zvol_minors != 0); 2285} 2286 2287void 2288zvol_init(void) 2289{ 2290 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 2291 1) == 0); 2292#ifdef illumos 2293 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL); 2294#else 2295 ZFS_LOG(1, "ZVOL Initialized."); 2296#endif 2297} 2298 2299void 2300zvol_fini(void) 2301{ 2302#ifdef illumos 2303 mutex_destroy(&zfsdev_state_lock); 2304#endif 2305 ddi_soft_state_fini(&zfsdev_state); 2306 ZFS_LOG(1, "ZVOL Deinitialized."); 2307} 2308 2309#ifdef illumos 2310/*ARGSUSED*/ 2311static int 2312zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx) 2313{ 2314 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 2315 2316 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 2317 return (1); 2318 return (0); 2319} 2320 2321/*ARGSUSED*/ 2322static void 2323zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx) 2324{ 2325 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 2326 2327 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx); 2328} 2329 2330static int 2331zvol_dump_init(zvol_state_t *zv, boolean_t resize) 2332{ 2333 dmu_tx_t *tx; 2334 int error; 2335 objset_t *os = zv->zv_objset; 2336 spa_t *spa = dmu_objset_spa(os); 2337 vdev_t *vd = spa->spa_root_vdev; 2338 nvlist_t *nv = NULL; 2339 uint64_t version = spa_version(spa); 2340 uint64_t checksum, compress, refresrv, vbs, dedup; 2341 2342 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 2343 ASSERT(vd->vdev_ops == &vdev_root_ops); 2344 2345 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 2346 DMU_OBJECT_END); 2347 if (error != 0) 2348 return (error); 2349 /* wait for dmu_free_long_range to actually free the blocks */ 2350 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2351 2352 /* 2353 * If the pool on which the dump device is being initialized has more 2354 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is 2355 * enabled. If so, bump that feature's counter to indicate that the 2356 * feature is active. We also check the vdev type to handle the 2357 * following case: 2358 * # zpool create test raidz disk1 disk2 disk3 2359 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev), 2360 * the raidz vdev itself has 3 children. 2361 */ 2362 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) { 2363 if (!spa_feature_is_enabled(spa, 2364 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 2365 return (SET_ERROR(ENOTSUP)); 2366 (void) dsl_sync_task(spa_name(spa), 2367 zfs_mvdev_dump_feature_check, 2368 zfs_mvdev_dump_activate_feature_sync, NULL, 2369 2, ZFS_SPACE_CHECK_RESERVED); 2370 } 2371 2372 if (!resize) { 2373 error = dsl_prop_get_integer(zv->zv_name, 2374 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 2375 if (error == 0) { 2376 error = dsl_prop_get_integer(zv->zv_name, 2377 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, 2378 NULL); 2379 } 2380 if (error == 0) { 2381 error = dsl_prop_get_integer(zv->zv_name, 2382 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 2383 &refresrv, NULL); 2384 } 2385 if (error == 0) { 2386 error = dsl_prop_get_integer(zv->zv_name, 2387 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, 2388 NULL); 2389 } 2390 if (version >= SPA_VERSION_DEDUP && error == 0) { 2391 error = dsl_prop_get_integer(zv->zv_name, 2392 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 2393 } 2394 } 2395 if (error != 0) 2396 return (error); 2397 2398 tx = dmu_tx_create(os); 2399 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2400 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2401 error = dmu_tx_assign(tx, TXG_WAIT); 2402 if (error != 0) { 2403 dmu_tx_abort(tx); 2404 return (error); 2405 } 2406 2407 /* 2408 * If we are resizing the dump device then we only need to 2409 * update the refreservation to match the newly updated 2410 * zvolsize. Otherwise, we save off the original state of the 2411 * zvol so that we can restore them if the zvol is ever undumpified. 2412 */ 2413 if (resize) { 2414 error = zap_update(os, ZVOL_ZAP_OBJ, 2415 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2416 &zv->zv_volsize, tx); 2417 } else { 2418 error = zap_update(os, ZVOL_ZAP_OBJ, 2419 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 2420 &compress, tx); 2421 if (error == 0) { 2422 error = zap_update(os, ZVOL_ZAP_OBJ, 2423 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, 2424 &checksum, tx); 2425 } 2426 if (error == 0) { 2427 error = zap_update(os, ZVOL_ZAP_OBJ, 2428 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2429 &refresrv, tx); 2430 } 2431 if (error == 0) { 2432 error = zap_update(os, ZVOL_ZAP_OBJ, 2433 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 2434 &vbs, tx); 2435 } 2436 if (error == 0) { 2437 error = dmu_object_set_blocksize( 2438 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx); 2439 } 2440 if (version >= SPA_VERSION_DEDUP && error == 0) { 2441 error = zap_update(os, ZVOL_ZAP_OBJ, 2442 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 2443 &dedup, tx); 2444 } 2445 if (error == 0) 2446 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE; 2447 } 2448 dmu_tx_commit(tx); 2449 2450 /* 2451 * We only need update the zvol's property if we are initializing 2452 * the dump area for the first time. 2453 */ 2454 if (error == 0 && !resize) { 2455 /* 2456 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum 2457 * function. Otherwise, use the old default -- OFF. 2458 */ 2459 checksum = spa_feature_is_active(spa, 2460 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY : 2461 ZIO_CHECKSUM_OFF; 2462 2463 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2464 VERIFY(nvlist_add_uint64(nv, 2465 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 2466 VERIFY(nvlist_add_uint64(nv, 2467 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 2468 ZIO_COMPRESS_OFF) == 0); 2469 VERIFY(nvlist_add_uint64(nv, 2470 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 2471 checksum) == 0); 2472 if (version >= SPA_VERSION_DEDUP) { 2473 VERIFY(nvlist_add_uint64(nv, 2474 zfs_prop_to_name(ZFS_PROP_DEDUP), 2475 ZIO_CHECKSUM_OFF) == 0); 2476 } 2477 2478 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2479 nv, NULL); 2480 nvlist_free(nv); 2481 } 2482 2483 /* Allocate the space for the dump */ 2484 if (error == 0) 2485 error = zvol_prealloc(zv); 2486 return (error); 2487} 2488 2489static int 2490zvol_dumpify(zvol_state_t *zv) 2491{ 2492 int error = 0; 2493 uint64_t dumpsize = 0; 2494 dmu_tx_t *tx; 2495 objset_t *os = zv->zv_objset; 2496 2497 if (zv->zv_flags & ZVOL_RDONLY) 2498 return (SET_ERROR(EROFS)); 2499 2500 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 2501 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 2502 boolean_t resize = (dumpsize > 0); 2503 2504 if ((error = zvol_dump_init(zv, resize)) != 0) { 2505 (void) zvol_dump_fini(zv); 2506 return (error); 2507 } 2508 } 2509 2510 /* 2511 * Build up our lba mapping. 2512 */ 2513 error = zvol_get_lbas(zv); 2514 if (error) { 2515 (void) zvol_dump_fini(zv); 2516 return (error); 2517 } 2518 2519 tx = dmu_tx_create(os); 2520 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2521 error = dmu_tx_assign(tx, TXG_WAIT); 2522 if (error) { 2523 dmu_tx_abort(tx); 2524 (void) zvol_dump_fini(zv); 2525 return (error); 2526 } 2527 2528 zv->zv_flags |= ZVOL_DUMPIFIED; 2529 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 2530 &zv->zv_volsize, tx); 2531 dmu_tx_commit(tx); 2532 2533 if (error) { 2534 (void) zvol_dump_fini(zv); 2535 return (error); 2536 } 2537 2538 txg_wait_synced(dmu_objset_pool(os), 0); 2539 return (0); 2540} 2541 2542static int 2543zvol_dump_fini(zvol_state_t *zv) 2544{ 2545 dmu_tx_t *tx; 2546 objset_t *os = zv->zv_objset; 2547 nvlist_t *nv; 2548 int error = 0; 2549 uint64_t checksum, compress, refresrv, vbs, dedup; 2550 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 2551 2552 /* 2553 * Attempt to restore the zvol back to its pre-dumpified state. 2554 * This is a best-effort attempt as it's possible that not all 2555 * of these properties were initialized during the dumpify process 2556 * (i.e. error during zvol_dump_init). 2557 */ 2558 2559 tx = dmu_tx_create(os); 2560 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2561 error = dmu_tx_assign(tx, TXG_WAIT); 2562 if (error) { 2563 dmu_tx_abort(tx); 2564 return (error); 2565 } 2566 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 2567 dmu_tx_commit(tx); 2568 2569 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2570 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 2571 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2572 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 2573 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2574 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 2575 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2576 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 2577 2578 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2579 (void) nvlist_add_uint64(nv, 2580 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 2581 (void) nvlist_add_uint64(nv, 2582 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 2583 (void) nvlist_add_uint64(nv, 2584 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 2585 if (version >= SPA_VERSION_DEDUP && 2586 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2587 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 2588 (void) nvlist_add_uint64(nv, 2589 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 2590 } 2591 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2592 nv, NULL); 2593 nvlist_free(nv); 2594 2595 zvol_free_extents(zv); 2596 zv->zv_flags &= ~ZVOL_DUMPIFIED; 2597 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 2598 /* wait for dmu_free_long_range to actually free the blocks */ 2599 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2600 tx = dmu_tx_create(os); 2601 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2602 error = dmu_tx_assign(tx, TXG_WAIT); 2603 if (error) { 2604 dmu_tx_abort(tx); 2605 return (error); 2606 } 2607 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 2608 zv->zv_volblocksize = vbs; 2609 dmu_tx_commit(tx); 2610 2611 return (0); 2612} 2613#else /* !illumos */ 2614 2615static void 2616zvol_geom_run(zvol_state_t *zv) 2617{ 2618 struct g_provider *pp; 2619 2620 pp = zv->zv_provider; 2621 g_error_provider(pp, 0); 2622 2623 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0, 2624 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER)); 2625} 2626 2627static void 2628zvol_geom_destroy(zvol_state_t *zv) 2629{ 2630 struct g_provider *pp; 2631 2632 g_topology_assert(); 2633 2634 mtx_lock(&zv->zv_queue_mtx); 2635 zv->zv_state = 1; 2636 wakeup_one(&zv->zv_queue); 2637 while (zv->zv_state != 2) 2638 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0); 2639 mtx_destroy(&zv->zv_queue_mtx); 2640 2641 pp = zv->zv_provider; 2642 zv->zv_provider = NULL; 2643 pp->private = NULL; 2644 g_wither_geom(pp->geom, ENXIO); 2645} 2646 2647static int 2648zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) 2649{ 2650 int count, error, flags; 2651 2652 g_topology_assert(); 2653 2654 /* 2655 * To make it easier we expect either open or close, but not both 2656 * at the same time. 2657 */ 2658 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) || 2659 (acr <= 0 && acw <= 0 && ace <= 0), 2660 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", 2661 pp->name, acr, acw, ace)); 2662 2663 if (pp->private == NULL) { 2664 if (acr <= 0 && acw <= 0 && ace <= 0) 2665 return (0); 2666 return (pp->error); 2667 } 2668 2669 /* 2670 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0, 2671 * because GEOM already handles that and handles it a bit differently. 2672 * GEOM allows for multiple read/exclusive consumers and ZFS allows 2673 * only one exclusive consumer, no matter if it is reader or writer. 2674 * I like better the way GEOM works so I'll leave it for GEOM to 2675 * decide what to do. 2676 */ 2677 2678 count = acr + acw + ace; 2679 if (count == 0) 2680 return (0); 2681 2682 flags = 0; 2683 if (acr != 0 || ace != 0) 2684 flags |= FREAD; 2685 if (acw != 0) 2686 flags |= FWRITE; 2687 2688 g_topology_unlock(); 2689 if (count > 0) 2690 error = zvol_open(pp, flags, count); 2691 else 2692 error = zvol_close(pp, flags, -count); 2693 g_topology_lock(); 2694 return (error); 2695} 2696 2697static void 2698zvol_geom_start(struct bio *bp) 2699{ 2700 zvol_state_t *zv; 2701 boolean_t first; 2702 2703 zv = bp->bio_to->private; 2704 ASSERT(zv != NULL); 2705 switch (bp->bio_cmd) { 2706 case BIO_FLUSH: 2707 if (!THREAD_CAN_SLEEP()) 2708 goto enqueue; 2709 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2710 g_io_deliver(bp, 0); 2711 break; 2712 case BIO_READ: 2713 case BIO_WRITE: 2714 case BIO_DELETE: 2715 if (!THREAD_CAN_SLEEP()) 2716 goto enqueue; 2717 zvol_strategy(bp); 2718 break; 2719 case BIO_GETATTR: { 2720 spa_t *spa = dmu_objset_spa(zv->zv_objset); 2721 uint64_t refd, avail, usedobjs, availobjs, val; 2722 2723 if (g_handleattr_int(bp, "GEOM::candelete", 1)) 2724 return; 2725 if (strcmp(bp->bio_attribute, "blocksavail") == 0) { 2726 dmu_objset_space(zv->zv_objset, &refd, &avail, 2727 &usedobjs, &availobjs); 2728 if (g_handleattr_off_t(bp, "blocksavail", 2729 avail / DEV_BSIZE)) 2730 return; 2731 } else if (strcmp(bp->bio_attribute, "blocksused") == 0) { 2732 dmu_objset_space(zv->zv_objset, &refd, &avail, 2733 &usedobjs, &availobjs); 2734 if (g_handleattr_off_t(bp, "blocksused", 2735 refd / DEV_BSIZE)) 2736 return; 2737 } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) { 2738 avail = metaslab_class_get_space(spa_normal_class(spa)); 2739 avail -= metaslab_class_get_alloc(spa_normal_class(spa)); 2740 if (g_handleattr_off_t(bp, "poolblocksavail", 2741 avail / DEV_BSIZE)) 2742 return; 2743 } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) { 2744 refd = metaslab_class_get_alloc(spa_normal_class(spa)); 2745 if (g_handleattr_off_t(bp, "poolblocksused", 2746 refd / DEV_BSIZE)) 2747 return; 2748 } 2749 /* FALLTHROUGH */ 2750 } 2751 default: 2752 g_io_deliver(bp, EOPNOTSUPP); 2753 break; 2754 } 2755 return; 2756 2757enqueue: 2758 mtx_lock(&zv->zv_queue_mtx); 2759 first = (bioq_first(&zv->zv_queue) == NULL); 2760 bioq_insert_tail(&zv->zv_queue, bp); 2761 mtx_unlock(&zv->zv_queue_mtx); 2762 if (first) 2763 wakeup_one(&zv->zv_queue); 2764} 2765 2766static void 2767zvol_geom_worker(void *arg) 2768{ 2769 zvol_state_t *zv; 2770 struct bio *bp; 2771 2772 thread_lock(curthread); 2773 sched_prio(curthread, PRIBIO); 2774 thread_unlock(curthread); 2775 2776 zv = arg; 2777 for (;;) { 2778 mtx_lock(&zv->zv_queue_mtx); 2779 bp = bioq_takefirst(&zv->zv_queue); 2780 if (bp == NULL) { 2781 if (zv->zv_state == 1) { 2782 zv->zv_state = 2; 2783 wakeup(&zv->zv_state); 2784 mtx_unlock(&zv->zv_queue_mtx); 2785 kthread_exit(); 2786 } 2787 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP, 2788 "zvol:io", 0); 2789 continue; 2790 } 2791 mtx_unlock(&zv->zv_queue_mtx); 2792 switch (bp->bio_cmd) { 2793 case BIO_FLUSH: 2794 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2795 g_io_deliver(bp, 0); 2796 break; 2797 case BIO_READ: 2798 case BIO_WRITE: 2799 case BIO_DELETE: 2800 zvol_strategy(bp); 2801 break; 2802 default: 2803 g_io_deliver(bp, EOPNOTSUPP); 2804 break; 2805 } 2806 } 2807} 2808 2809extern boolean_t dataset_name_hidden(const char *name); 2810 2811static int 2812zvol_create_snapshots(objset_t *os, const char *name) 2813{ 2814 uint64_t cookie, obj; 2815 char *sname; 2816 int error, len; 2817 2818 cookie = obj = 0; 2819 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2820 2821#if 0 2822 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL, 2823 DS_FIND_SNAPSHOTS); 2824#endif 2825 2826 for (;;) { 2827 len = snprintf(sname, MAXPATHLEN, "%s@", name); 2828 if (len >= MAXPATHLEN) { 2829 dmu_objset_rele(os, FTAG); 2830 error = ENAMETOOLONG; 2831 break; 2832 } 2833 2834 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 2835 error = dmu_snapshot_list_next(os, MAXPATHLEN - len, 2836 sname + len, &obj, &cookie, NULL); 2837 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 2838 if (error != 0) { 2839 if (error == ENOENT) 2840 error = 0; 2841 break; 2842 } 2843 2844 if ((error = zvol_create_minor(sname)) != 0) { 2845 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2846 sname, error); 2847 break; 2848 } 2849 } 2850 2851 kmem_free(sname, MAXPATHLEN); 2852 return (error); 2853} 2854 2855int 2856zvol_create_minors(const char *name) 2857{ 2858 uint64_t cookie; 2859 objset_t *os; 2860 char *osname, *p; 2861 int error, len; 2862 2863 if (dataset_name_hidden(name)) 2864 return (0); 2865 2866 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2867 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2868 name, error); 2869 return (error); 2870 } 2871 if (dmu_objset_type(os) == DMU_OST_ZVOL) { 2872 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG); 2873 dsl_pool_rele(dmu_objset_pool(os), FTAG); 2874 error = zvol_create_minor(name); 2875 if (error == 0 || error == EEXIST) { 2876 error = zvol_create_snapshots(os, name); 2877 } else { 2878 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2879 name, error); 2880 } 2881 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG); 2882 dsl_dataset_rele(os->os_dsl_dataset, FTAG); 2883 return (error); 2884 } 2885 if (dmu_objset_type(os) != DMU_OST_ZFS) { 2886 dmu_objset_rele(os, FTAG); 2887 return (0); 2888 } 2889 2890 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2891 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) { 2892 dmu_objset_rele(os, FTAG); 2893 kmem_free(osname, MAXPATHLEN); 2894 return (ENOENT); 2895 } 2896 p = osname + strlen(osname); 2897 len = MAXPATHLEN - (p - osname); 2898 2899#if 0 2900 /* Prefetch the datasets. */ 2901 cookie = 0; 2902 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) { 2903 if (!dataset_name_hidden(osname)) 2904 (void) dmu_objset_prefetch(osname, NULL); 2905 } 2906#endif 2907 2908 cookie = 0; 2909 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL, 2910 &cookie) == 0) { 2911 dmu_objset_rele(os, FTAG); 2912 (void)zvol_create_minors(osname); 2913 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2914 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2915 name, error); 2916 return (error); 2917 } 2918 } 2919 2920 dmu_objset_rele(os, FTAG); 2921 kmem_free(osname, MAXPATHLEN); 2922 return (0); 2923} 2924 2925static void 2926zvol_rename_minor(zvol_state_t *zv, const char *newname) 2927{ 2928 struct g_geom *gp; 2929 struct g_provider *pp; 2930 struct cdev *dev; 2931 2932 ASSERT(MUTEX_HELD(&zfsdev_state_lock)); 2933 2934 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 2935 g_topology_lock(); 2936 pp = zv->zv_provider; 2937 ASSERT(pp != NULL); 2938 gp = pp->geom; 2939 ASSERT(gp != NULL); 2940 2941 zv->zv_provider = NULL; 2942 g_wither_provider(pp, ENXIO); 2943 2944 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname); 2945 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 2946 pp->sectorsize = DEV_BSIZE; 2947 pp->mediasize = zv->zv_volsize; 2948 pp->private = zv; 2949 zv->zv_provider = pp; 2950 g_error_provider(pp, 0); 2951 g_topology_unlock(); 2952 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { 2953 dev = zv->zv_dev; 2954 ASSERT(dev != NULL); 2955 zv->zv_dev = NULL; 2956 destroy_dev(dev); 2957 2958 if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, 2959 &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 2960 0640, "%s/%s", ZVOL_DRIVER, newname) == 0) { 2961 zv->zv_dev = dev; 2962 dev->si_iosize_max = MAXPHYS; 2963 dev->si_drv2 = zv; 2964 } 2965 } 2966 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name)); 2967} 2968 2969void 2970zvol_rename_minors(const char *oldname, const char *newname) 2971{ 2972 char name[MAXPATHLEN]; 2973 struct g_provider *pp; 2974 struct g_geom *gp; 2975 size_t oldnamelen, newnamelen; 2976 zvol_state_t *zv; 2977 char *namebuf; 2978 boolean_t locked = B_FALSE; 2979 2980 oldnamelen = strlen(oldname); 2981 newnamelen = strlen(newname); 2982 2983 DROP_GIANT(); 2984 /* See comment in zvol_open(). */ 2985 if (!MUTEX_HELD(&zfsdev_state_lock)) { 2986 mutex_enter(&zfsdev_state_lock); 2987 locked = B_TRUE; 2988 } 2989 2990 LIST_FOREACH(zv, &all_zvols, zv_links) { 2991 if (strcmp(zv->zv_name, oldname) == 0) { 2992 zvol_rename_minor(zv, newname); 2993 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && 2994 (zv->zv_name[oldnamelen] == '/' || 2995 zv->zv_name[oldnamelen] == '@')) { 2996 snprintf(name, sizeof(name), "%s%c%s", newname, 2997 zv->zv_name[oldnamelen], 2998 zv->zv_name + oldnamelen + 1); 2999 zvol_rename_minor(zv, name); 3000 } 3001 } 3002 3003 if (locked) 3004 mutex_exit(&zfsdev_state_lock); 3005 PICKUP_GIANT(); 3006} 3007 3008static int 3009zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td) 3010{ 3011 zvol_state_t *zv; 3012 int err = 0; 3013 3014 mutex_enter(&zfsdev_state_lock); 3015 zv = dev->si_drv2; 3016 if (zv == NULL) { 3017 mutex_exit(&zfsdev_state_lock); 3018 return(ENXIO); /* zvol_create_minor() not done yet */ 3019 } 3020 3021 if (zv->zv_total_opens == 0) 3022 err = zvol_first_open(zv); 3023 if (err) { 3024 mutex_exit(&zfsdev_state_lock); 3025 return (err); 3026 } 3027 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 3028 err = SET_ERROR(EROFS); 3029 goto out; 3030 } 3031 if (zv->zv_flags & ZVOL_EXCL) { 3032 err = SET_ERROR(EBUSY); 3033 goto out; 3034 } 3035#ifdef FEXCL 3036 if (flags & FEXCL) { 3037 if (zv->zv_total_opens != 0) { 3038 err = SET_ERROR(EBUSY); 3039 goto out; 3040 } 3041 zv->zv_flags |= ZVOL_EXCL; 3042 } 3043#endif 3044 3045 zv->zv_total_opens++; 3046 mutex_exit(&zfsdev_state_lock); 3047 return (err); 3048out: 3049 if (zv->zv_total_opens == 0) 3050 zvol_last_close(zv); 3051 mutex_exit(&zfsdev_state_lock); 3052 return (err); 3053} 3054 3055static int 3056zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td) 3057{ 3058 zvol_state_t *zv; 3059 int err = 0; 3060 3061 mutex_enter(&zfsdev_state_lock); 3062 zv = dev->si_drv2; 3063 if (zv == NULL) { 3064 mutex_exit(&zfsdev_state_lock); 3065 return(ENXIO); 3066 } 3067 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