zvol.c revision 268649
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * 24 * Copyright (c) 2006-2010 Pawel Jakub Dawidek <pjd@FreeBSD.org> 25 * All rights reserved. 26 * Copyright (c) 2013 by Delphix. All rights reserved. 27 * Copyright (c) 2013, Joyent, Inc. All rights reserved. 28 * 29 * Portions Copyright 2010 Robert Milkowski 30 * 31 * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 32 */ 33 34/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 35 36/* 37 * ZFS volume emulation driver. 38 * 39 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 40 * Volumes are accessed through the symbolic links named: 41 * 42 * /dev/zvol/dsk/<pool_name>/<dataset_name> 43 * /dev/zvol/rdsk/<pool_name>/<dataset_name> 44 * 45 * These links are created by the /dev filesystem (sdev_zvolops.c). 46 * Volumes are persistent through reboot. No user command needs to be 47 * run before opening and using a device. 48 * 49 * FreeBSD notes. 50 * On FreeBSD ZVOLs are simply GEOM providers like any other storage device 51 * in the system. 52 */ 53 54#include <sys/types.h> 55#include <sys/param.h> 56#include <sys/kernel.h> 57#include <sys/errno.h> 58#include <sys/uio.h> 59#include <sys/bio.h> 60#include <sys/buf.h> 61#include <sys/kmem.h> 62#include <sys/conf.h> 63#include <sys/cmn_err.h> 64#include <sys/stat.h> 65#include <sys/zap.h> 66#include <sys/spa.h> 67#include <sys/spa_impl.h> 68#include <sys/zio.h> 69#include <sys/disk.h> 70#include <sys/dmu_traverse.h> 71#include <sys/dnode.h> 72#include <sys/dsl_dataset.h> 73#include <sys/dsl_prop.h> 74#include <sys/dkio.h> 75#include <sys/byteorder.h> 76#include <sys/sunddi.h> 77#include <sys/dirent.h> 78#include <sys/policy.h> 79#include <sys/queue.h> 80#include <sys/fs/zfs.h> 81#include <sys/zfs_ioctl.h> 82#include <sys/zil.h> 83#include <sys/refcount.h> 84#include <sys/zfs_znode.h> 85#include <sys/zfs_rlock.h> 86#include <sys/vdev_impl.h> 87#include <sys/vdev_raidz.h> 88#include <sys/zvol.h> 89#include <sys/zil_impl.h> 90#include <sys/dbuf.h> 91#include <sys/dmu_tx.h> 92#include <sys/zfeature.h> 93#include <sys/zio_checksum.h> 94 95#include <geom/geom.h> 96 97#include "zfs_namecheck.h" 98 99struct g_class zfs_zvol_class = { 100 .name = "ZFS::ZVOL", 101 .version = G_VERSION, 102}; 103 104DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol); 105 106void *zfsdev_state; 107static char *zvol_tag = "zvol_tag"; 108 109#define ZVOL_DUMPSIZE "dumpsize" 110 111/* 112 * The spa_namespace_lock protects the zfsdev_state structure from being 113 * modified while it's being used, e.g. an open that comes in before a 114 * create finishes. It also protects temporary opens of the dataset so that, 115 * e.g., an open doesn't get a spurious EBUSY. 116 */ 117static uint32_t zvol_minors; 118 119SYSCTL_DECL(_vfs_zfs); 120SYSCTL_NODE(_vfs_zfs, OID_AUTO, vol, CTLFLAG_RW, 0, "ZFS VOLUME"); 121static int volmode = ZFS_VOLMODE_GEOM; 122TUNABLE_INT("vfs.zfs.vol.mode", &volmode); 123SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &volmode, 0, 124 "Expose as GEOM providers (1), device files (2) or neither"); 125 126typedef struct zvol_extent { 127 list_node_t ze_node; 128 dva_t ze_dva; /* dva associated with this extent */ 129 uint64_t ze_nblks; /* number of blocks in extent */ 130} zvol_extent_t; 131 132/* 133 * The in-core state of each volume. 134 */ 135typedef struct zvol_state { 136 LIST_ENTRY(zvol_state) zv_links; 137 char zv_name[MAXPATHLEN]; /* pool/dd name */ 138 uint64_t zv_volsize; /* amount of space we advertise */ 139 uint64_t zv_volblocksize; /* volume block size */ 140 struct cdev *zv_dev; /* non-GEOM device */ 141 struct g_provider *zv_provider; /* GEOM provider */ 142 uint8_t zv_min_bs; /* minimum addressable block shift */ 143 uint8_t zv_flags; /* readonly, dumpified, etc. */ 144 objset_t *zv_objset; /* objset handle */ 145 uint32_t zv_total_opens; /* total open count */ 146 zilog_t *zv_zilog; /* ZIL handle */ 147 list_t zv_extents; /* List of extents for dump */ 148 znode_t zv_znode; /* for range locking */ 149 dmu_buf_t *zv_dbuf; /* bonus handle */ 150 int zv_state; 151 int zv_volmode; /* Provide GEOM or cdev */ 152 struct bio_queue_head zv_queue; 153 struct mtx zv_queue_mtx; /* zv_queue mutex */ 154} zvol_state_t; 155 156static LIST_HEAD(, zvol_state) all_zvols; 157 158/* 159 * zvol specific flags 160 */ 161#define ZVOL_RDONLY 0x1 162#define ZVOL_DUMPIFIED 0x2 163#define ZVOL_EXCL 0x4 164#define ZVOL_WCE 0x8 165 166/* 167 * zvol maximum transfer in one DMU tx. 168 */ 169int zvol_maxphys = DMU_MAX_ACCESS/2; 170 171static d_open_t zvol_d_open; 172static d_close_t zvol_d_close; 173static d_read_t zvol_read; 174static d_write_t zvol_write; 175static d_ioctl_t zvol_d_ioctl; 176static d_strategy_t zvol_strategy; 177 178static struct cdevsw zvol_cdevsw = { 179 .d_version = D_VERSION, 180 .d_open = zvol_d_open, 181 .d_close = zvol_d_close, 182 .d_read = zvol_read, 183 .d_write = zvol_write, 184 .d_ioctl = zvol_d_ioctl, 185 .d_strategy = zvol_strategy, 186 .d_name = "zvol", 187 .d_flags = D_DISK | D_TRACKCLOSE, 188}; 189 190extern int zfs_set_prop_nvlist(const char *, zprop_source_t, 191 nvlist_t *, nvlist_t *); 192static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, 193 uint64_t len, boolean_t sync); 194static int zvol_remove_zv(zvol_state_t *); 195static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio); 196static int zvol_dumpify(zvol_state_t *zv); 197static int zvol_dump_fini(zvol_state_t *zv); 198static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); 199 200static void zvol_geom_run(zvol_state_t *zv); 201static void zvol_geom_destroy(zvol_state_t *zv); 202static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace); 203static void zvol_geom_start(struct bio *bp); 204static void zvol_geom_worker(void *arg); 205 206static void 207zvol_size_changed(zvol_state_t *zv) 208{ 209#ifdef sun 210 dev_t dev = makedevice(maj, min); 211 212 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 213 "Size", volsize) == DDI_SUCCESS); 214 VERIFY(ddi_prop_update_int64(dev, zfs_dip, 215 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS); 216 217 /* Notify specfs to invalidate the cached size */ 218 spec_size_invalidate(dev, VBLK); 219 spec_size_invalidate(dev, VCHR); 220#else /* !sun */ 221 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 222 struct g_provider *pp; 223 224 pp = zv->zv_provider; 225 if (pp == NULL) 226 return; 227 g_topology_lock(); 228 g_resize_provider(pp, zv->zv_volsize); 229 g_topology_unlock(); 230 } 231#endif /* !sun */ 232} 233 234int 235zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 236{ 237 if (volsize == 0) 238 return (SET_ERROR(EINVAL)); 239 240 if (volsize % blocksize != 0) 241 return (SET_ERROR(EINVAL)); 242 243#ifdef _ILP32 244 if (volsize - 1 > SPEC_MAXOFFSET_T) 245 return (SET_ERROR(EOVERFLOW)); 246#endif 247 return (0); 248} 249 250int 251zvol_check_volblocksize(uint64_t volblocksize) 252{ 253 if (volblocksize < SPA_MINBLOCKSIZE || 254 volblocksize > SPA_MAXBLOCKSIZE || 255 !ISP2(volblocksize)) 256 return (SET_ERROR(EDOM)); 257 258 return (0); 259} 260 261int 262zvol_get_stats(objset_t *os, nvlist_t *nv) 263{ 264 int error; 265 dmu_object_info_t doi; 266 uint64_t val; 267 268 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 269 if (error) 270 return (error); 271 272 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 273 274 error = dmu_object_info(os, ZVOL_OBJ, &doi); 275 276 if (error == 0) { 277 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 278 doi.doi_data_block_size); 279 } 280 281 return (error); 282} 283 284static zvol_state_t * 285zvol_minor_lookup(const char *name) 286{ 287 zvol_state_t *zv; 288 289 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 290 291 LIST_FOREACH(zv, &all_zvols, zv_links) { 292 if (strcmp(zv->zv_name, name) == 0) 293 break; 294 } 295 296 return (zv); 297} 298 299/* extent mapping arg */ 300struct maparg { 301 zvol_state_t *ma_zv; 302 uint64_t ma_blks; 303}; 304 305/*ARGSUSED*/ 306static int 307zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 308 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 309{ 310 struct maparg *ma = arg; 311 zvol_extent_t *ze; 312 int bs = ma->ma_zv->zv_volblocksize; 313 314 if (BP_IS_HOLE(bp) || 315 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) 316 return (0); 317 318 VERIFY(!BP_IS_EMBEDDED(bp)); 319 320 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 321 ma->ma_blks++; 322 323 /* Abort immediately if we have encountered gang blocks */ 324 if (BP_IS_GANG(bp)) 325 return (SET_ERROR(EFRAGS)); 326 327 /* 328 * See if the block is at the end of the previous extent. 329 */ 330 ze = list_tail(&ma->ma_zv->zv_extents); 331 if (ze && 332 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 333 DVA_GET_OFFSET(BP_IDENTITY(bp)) == 334 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 335 ze->ze_nblks++; 336 return (0); 337 } 338 339 dprintf_bp(bp, "%s", "next blkptr:"); 340 341 /* start a new extent */ 342 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 343 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 344 ze->ze_nblks = 1; 345 list_insert_tail(&ma->ma_zv->zv_extents, ze); 346 return (0); 347} 348 349static void 350zvol_free_extents(zvol_state_t *zv) 351{ 352 zvol_extent_t *ze; 353 354 while (ze = list_head(&zv->zv_extents)) { 355 list_remove(&zv->zv_extents, ze); 356 kmem_free(ze, sizeof (zvol_extent_t)); 357 } 358} 359 360static int 361zvol_get_lbas(zvol_state_t *zv) 362{ 363 objset_t *os = zv->zv_objset; 364 struct maparg ma; 365 int err; 366 367 ma.ma_zv = zv; 368 ma.ma_blks = 0; 369 zvol_free_extents(zv); 370 371 /* commit any in-flight changes before traversing the dataset */ 372 txg_wait_synced(dmu_objset_pool(os), 0); 373 err = traverse_dataset(dmu_objset_ds(os), 0, 374 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 375 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 376 zvol_free_extents(zv); 377 return (err ? err : EIO); 378 } 379 380 return (0); 381} 382 383/* ARGSUSED */ 384void 385zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 386{ 387 zfs_creat_t *zct = arg; 388 nvlist_t *nvprops = zct->zct_props; 389 int error; 390 uint64_t volblocksize, volsize; 391 392 VERIFY(nvlist_lookup_uint64(nvprops, 393 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 394 if (nvlist_lookup_uint64(nvprops, 395 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 396 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 397 398 /* 399 * These properties must be removed from the list so the generic 400 * property setting step won't apply to them. 401 */ 402 VERIFY(nvlist_remove_all(nvprops, 403 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 404 (void) nvlist_remove_all(nvprops, 405 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 406 407 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 408 DMU_OT_NONE, 0, tx); 409 ASSERT(error == 0); 410 411 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 412 DMU_OT_NONE, 0, tx); 413 ASSERT(error == 0); 414 415 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 416 ASSERT(error == 0); 417} 418 419/* 420 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we 421 * implement DKIOCFREE/free-long-range. 422 */ 423static int 424zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap) 425{ 426 uint64_t offset, length; 427 428 if (byteswap) 429 byteswap_uint64_array(lr, sizeof (*lr)); 430 431 offset = lr->lr_offset; 432 length = lr->lr_length; 433 434 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length)); 435} 436 437/* 438 * Replay a TX_WRITE ZIL transaction that didn't get committed 439 * after a system failure 440 */ 441static int 442zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 443{ 444 objset_t *os = zv->zv_objset; 445 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 446 uint64_t offset, length; 447 dmu_tx_t *tx; 448 int error; 449 450 if (byteswap) 451 byteswap_uint64_array(lr, sizeof (*lr)); 452 453 offset = lr->lr_offset; 454 length = lr->lr_length; 455 456 /* If it's a dmu_sync() block, write the whole block */ 457 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 458 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 459 if (length < blocksize) { 460 offset -= offset % blocksize; 461 length = blocksize; 462 } 463 } 464 465 tx = dmu_tx_create(os); 466 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 467 error = dmu_tx_assign(tx, TXG_WAIT); 468 if (error) { 469 dmu_tx_abort(tx); 470 } else { 471 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 472 dmu_tx_commit(tx); 473 } 474 475 return (error); 476} 477 478/* ARGSUSED */ 479static int 480zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 481{ 482 return (SET_ERROR(ENOTSUP)); 483} 484 485/* 486 * Callback vectors for replaying records. 487 * Only TX_WRITE and TX_TRUNCATE are needed for zvol. 488 */ 489zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 490 zvol_replay_err, /* 0 no such transaction type */ 491 zvol_replay_err, /* TX_CREATE */ 492 zvol_replay_err, /* TX_MKDIR */ 493 zvol_replay_err, /* TX_MKXATTR */ 494 zvol_replay_err, /* TX_SYMLINK */ 495 zvol_replay_err, /* TX_REMOVE */ 496 zvol_replay_err, /* TX_RMDIR */ 497 zvol_replay_err, /* TX_LINK */ 498 zvol_replay_err, /* TX_RENAME */ 499 zvol_replay_write, /* TX_WRITE */ 500 zvol_replay_truncate, /* TX_TRUNCATE */ 501 zvol_replay_err, /* TX_SETATTR */ 502 zvol_replay_err, /* TX_ACL */ 503 zvol_replay_err, /* TX_CREATE_ACL */ 504 zvol_replay_err, /* TX_CREATE_ATTR */ 505 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 506 zvol_replay_err, /* TX_MKDIR_ACL */ 507 zvol_replay_err, /* TX_MKDIR_ATTR */ 508 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 509 zvol_replay_err, /* TX_WRITE2 */ 510}; 511 512#ifdef sun 513int 514zvol_name2minor(const char *name, minor_t *minor) 515{ 516 zvol_state_t *zv; 517 518 mutex_enter(&spa_namespace_lock); 519 zv = zvol_minor_lookup(name); 520 if (minor && zv) 521 *minor = zv->zv_minor; 522 mutex_exit(&spa_namespace_lock); 523 return (zv ? 0 : -1); 524} 525#endif /* sun */ 526 527/* 528 * Create a minor node (plus a whole lot more) for the specified volume. 529 */ 530int 531zvol_create_minor(const char *name) 532{ 533 zfs_soft_state_t *zs; 534 zvol_state_t *zv; 535 objset_t *os; 536 struct cdev *dev; 537 struct g_provider *pp; 538 struct g_geom *gp; 539 dmu_object_info_t doi; 540 uint64_t volsize, mode; 541 int error; 542 543 ZFS_LOG(1, "Creating ZVOL %s...", name); 544 545 mutex_enter(&spa_namespace_lock); 546 547 if (zvol_minor_lookup(name) != NULL) { 548 mutex_exit(&spa_namespace_lock); 549 return (SET_ERROR(EEXIST)); 550 } 551 552 /* lie and say we're read-only */ 553 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os); 554 555 if (error) { 556 mutex_exit(&spa_namespace_lock); 557 return (error); 558 } 559 560#ifdef sun 561 if ((minor = zfsdev_minor_alloc()) == 0) { 562 dmu_objset_disown(os, FTAG); 563 mutex_exit(&spa_namespace_lock); 564 return (SET_ERROR(ENXIO)); 565 } 566 567 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 568 dmu_objset_disown(os, FTAG); 569 mutex_exit(&spa_namespace_lock); 570 return (SET_ERROR(EAGAIN)); 571 } 572 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 573 (char *)name); 574 575 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 576 577 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 578 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 579 ddi_soft_state_free(zfsdev_state, minor); 580 dmu_objset_disown(os, FTAG); 581 mutex_exit(&spa_namespace_lock); 582 return (SET_ERROR(EAGAIN)); 583 } 584 585 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 586 587 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 588 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 589 ddi_remove_minor_node(zfs_dip, chrbuf); 590 ddi_soft_state_free(zfsdev_state, minor); 591 dmu_objset_disown(os, FTAG); 592 mutex_exit(&spa_namespace_lock); 593 return (SET_ERROR(EAGAIN)); 594 } 595 596 zs = ddi_get_soft_state(zfsdev_state, minor); 597 zs->zss_type = ZSST_ZVOL; 598 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 599#else /* !sun */ 600 601 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP); 602 zv->zv_state = 0; 603 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 604 if (error) { 605 kmem_free(zv, sizeof(*zv)); 606 dmu_objset_disown(os, zvol_tag); 607 mutex_exit(&spa_namespace_lock); 608 return (error); 609 } 610 error = dsl_prop_get_integer(name, 611 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL); 612 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT) 613 mode = volmode; 614 615 DROP_GIANT(); 616 zv->zv_volsize = volsize; 617 zv->zv_volmode = mode; 618 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 619 g_topology_lock(); 620 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name); 621 gp->start = zvol_geom_start; 622 gp->access = zvol_geom_access; 623 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name); 624 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 625 pp->sectorsize = DEV_BSIZE; 626 pp->mediasize = zv->zv_volsize; 627 pp->private = zv; 628 629 zv->zv_provider = pp; 630 bioq_init(&zv->zv_queue); 631 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF); 632 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { 633 if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, 634 &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 635 0640, "%s/%s", ZVOL_DRIVER, name) != 0) { 636 kmem_free(zv, sizeof(*zv)); 637 dmu_objset_disown(os, FTAG); 638 mutex_exit(&spa_namespace_lock); 639 return (SET_ERROR(ENXIO)); 640 } 641 zv->zv_dev = dev; 642 dev->si_iosize_max = MAXPHYS; 643 dev->si_drv2 = zv; 644 } 645 LIST_INSERT_HEAD(&all_zvols, zv, zv_links); 646#endif /* !sun */ 647 648 (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 649 zv->zv_min_bs = DEV_BSHIFT; 650 zv->zv_objset = os; 651 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) 652 zv->zv_flags |= ZVOL_RDONLY; 653 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 654 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 655 sizeof (rl_t), offsetof(rl_t, r_node)); 656 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 657 offsetof(zvol_extent_t, ze_node)); 658 /* get and cache the blocksize */ 659 error = dmu_object_info(os, ZVOL_OBJ, &doi); 660 ASSERT(error == 0); 661 zv->zv_volblocksize = doi.doi_data_block_size; 662 663 if (spa_writeable(dmu_objset_spa(os))) { 664 if (zil_replay_disable) 665 zil_destroy(dmu_objset_zil(os), B_FALSE); 666 else 667 zil_replay(os, zv, zvol_replay_vector); 668 } 669 dmu_objset_disown(os, FTAG); 670 zv->zv_objset = NULL; 671 672 zvol_minors++; 673 674 mutex_exit(&spa_namespace_lock); 675 676#ifndef sun 677 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 678 zvol_geom_run(zv); 679 g_topology_unlock(); 680 } 681 PICKUP_GIANT(); 682#endif 683 684 ZFS_LOG(1, "ZVOL %s created.", name); 685 686 return (0); 687} 688 689/* 690 * Remove minor node for the specified volume. 691 */ 692static int 693zvol_remove_zv(zvol_state_t *zv) 694{ 695#ifdef sun 696 minor_t minor = zv->zv_minor; 697#endif 698 699 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 700 if (zv->zv_total_opens != 0) 701 return (SET_ERROR(EBUSY)); 702 703 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); 704 705#ifdef sun 706 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 707 ddi_remove_minor_node(zfs_dip, nmbuf); 708#else 709 LIST_REMOVE(zv, zv_links); 710 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 711 g_topology_lock(); 712 zvol_geom_destroy(zv); 713 g_topology_unlock(); 714 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) 715 destroy_dev(zv->zv_dev); 716#endif /* sun */ 717 718 avl_destroy(&zv->zv_znode.z_range_avl); 719 mutex_destroy(&zv->zv_znode.z_range_lock); 720 721 kmem_free(zv, sizeof(*zv)); 722 723 zvol_minors--; 724 return (0); 725} 726 727int 728zvol_remove_minor(const char *name) 729{ 730 zvol_state_t *zv; 731 int rc; 732 733 mutex_enter(&spa_namespace_lock); 734 if ((zv = zvol_minor_lookup(name)) == NULL) { 735 mutex_exit(&spa_namespace_lock); 736 return (SET_ERROR(ENXIO)); 737 } 738 rc = zvol_remove_zv(zv); 739 mutex_exit(&spa_namespace_lock); 740 return (rc); 741} 742 743int 744zvol_first_open(zvol_state_t *zv) 745{ 746 objset_t *os; 747 uint64_t volsize; 748 int error; 749 uint64_t readonly; 750 751 /* lie and say we're read-only */ 752 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 753 zvol_tag, &os); 754 if (error) 755 return (error); 756 757 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 758 if (error) { 759 ASSERT(error == 0); 760 dmu_objset_disown(os, zvol_tag); 761 return (error); 762 } 763 zv->zv_objset = os; 764 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 765 if (error) { 766 dmu_objset_disown(os, zvol_tag); 767 return (error); 768 } 769 zv->zv_volsize = volsize; 770 zv->zv_zilog = zil_open(os, zvol_get_data); 771 zvol_size_changed(zv); 772 773 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 774 NULL) == 0); 775 if (readonly || dmu_objset_is_snapshot(os) || 776 !spa_writeable(dmu_objset_spa(os))) 777 zv->zv_flags |= ZVOL_RDONLY; 778 else 779 zv->zv_flags &= ~ZVOL_RDONLY; 780 return (error); 781} 782 783void 784zvol_last_close(zvol_state_t *zv) 785{ 786 zil_close(zv->zv_zilog); 787 zv->zv_zilog = NULL; 788 789 dmu_buf_rele(zv->zv_dbuf, zvol_tag); 790 zv->zv_dbuf = NULL; 791 792 /* 793 * Evict cached data 794 */ 795 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && 796 !(zv->zv_flags & ZVOL_RDONLY)) 797 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 798 dmu_objset_evict_dbufs(zv->zv_objset); 799 800 dmu_objset_disown(zv->zv_objset, zvol_tag); 801 zv->zv_objset = NULL; 802} 803 804#ifdef sun 805int 806zvol_prealloc(zvol_state_t *zv) 807{ 808 objset_t *os = zv->zv_objset; 809 dmu_tx_t *tx; 810 uint64_t refd, avail, usedobjs, availobjs; 811 uint64_t resid = zv->zv_volsize; 812 uint64_t off = 0; 813 814 /* Check the space usage before attempting to allocate the space */ 815 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 816 if (avail < zv->zv_volsize) 817 return (SET_ERROR(ENOSPC)); 818 819 /* Free old extents if they exist */ 820 zvol_free_extents(zv); 821 822 while (resid != 0) { 823 int error; 824 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); 825 826 tx = dmu_tx_create(os); 827 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 828 error = dmu_tx_assign(tx, TXG_WAIT); 829 if (error) { 830 dmu_tx_abort(tx); 831 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 832 return (error); 833 } 834 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 835 dmu_tx_commit(tx); 836 off += bytes; 837 resid -= bytes; 838 } 839 txg_wait_synced(dmu_objset_pool(os), 0); 840 841 return (0); 842} 843#endif /* sun */ 844 845static int 846zvol_update_volsize(objset_t *os, uint64_t volsize) 847{ 848 dmu_tx_t *tx; 849 int error; 850 851 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 852 853 tx = dmu_tx_create(os); 854 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 855 error = dmu_tx_assign(tx, TXG_WAIT); 856 if (error) { 857 dmu_tx_abort(tx); 858 return (error); 859 } 860 861 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 862 &volsize, tx); 863 dmu_tx_commit(tx); 864 865 if (error == 0) 866 error = dmu_free_long_range(os, 867 ZVOL_OBJ, volsize, DMU_OBJECT_END); 868 return (error); 869} 870 871void 872zvol_remove_minors(const char *name) 873{ 874 zvol_state_t *zv, *tzv; 875 size_t namelen; 876 877 namelen = strlen(name); 878 879 DROP_GIANT(); 880 mutex_enter(&spa_namespace_lock); 881 882 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) { 883 if (strcmp(zv->zv_name, name) == 0 || 884 (strncmp(zv->zv_name, name, namelen) == 0 && 885 zv->zv_name[namelen] == '/')) { 886 (void) zvol_remove_zv(zv); 887 } 888 } 889 890 mutex_exit(&spa_namespace_lock); 891 PICKUP_GIANT(); 892} 893 894int 895zvol_set_volsize(const char *name, major_t maj, uint64_t volsize) 896{ 897 zvol_state_t *zv = NULL; 898 objset_t *os; 899 int error; 900 dmu_object_info_t doi; 901 uint64_t old_volsize = 0ULL; 902 uint64_t readonly; 903 904 mutex_enter(&spa_namespace_lock); 905 zv = zvol_minor_lookup(name); 906 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 907 mutex_exit(&spa_namespace_lock); 908 return (error); 909 } 910 911 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 912 (error = zvol_check_volsize(volsize, 913 doi.doi_data_block_size)) != 0) 914 goto out; 915 916 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, 917 NULL) == 0); 918 if (readonly) { 919 error = EROFS; 920 goto out; 921 } 922 923 error = zvol_update_volsize(os, volsize); 924 /* 925 * Reinitialize the dump area to the new size. If we 926 * failed to resize the dump area then restore it back to 927 * its original size. 928 */ 929 if (zv && error == 0) { 930#ifdef ZVOL_DUMP 931 if (zv->zv_flags & ZVOL_DUMPIFIED) { 932 old_volsize = zv->zv_volsize; 933 zv->zv_volsize = volsize; 934 if ((error = zvol_dumpify(zv)) != 0 || 935 (error = dumpvp_resize()) != 0) { 936 (void) zvol_update_volsize(os, old_volsize); 937 zv->zv_volsize = old_volsize; 938 error = zvol_dumpify(zv); 939 } 940 } 941#endif /* ZVOL_DUMP */ 942 if (error == 0) { 943 zv->zv_volsize = volsize; 944 zvol_size_changed(zv); 945 } 946 } 947 948#ifdef sun 949 /* 950 * Generate a LUN expansion event. 951 */ 952 if (zv && error == 0) { 953 sysevent_id_t eid; 954 nvlist_t *attr; 955 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 956 957 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 958 zv->zv_minor); 959 960 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 961 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 962 963 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 964 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 965 966 nvlist_free(attr); 967 kmem_free(physpath, MAXPATHLEN); 968 } 969#endif /* sun */ 970 971out: 972 dmu_objset_rele(os, FTAG); 973 974 mutex_exit(&spa_namespace_lock); 975 976 return (error); 977} 978 979/*ARGSUSED*/ 980static int 981zvol_open(struct g_provider *pp, int flag, int count) 982{ 983 zvol_state_t *zv; 984 int err = 0; 985 boolean_t locked = B_FALSE; 986 987 /* 988 * Protect against recursively entering spa_namespace_lock 989 * when spa_open() is used for a pool on a (local) ZVOL(s). 990 * This is needed since we replaced upstream zfsdev_state_lock 991 * with spa_namespace_lock in the ZVOL code. 992 * We are using the same trick as spa_open(). 993 * Note that calls in zvol_first_open which need to resolve 994 * pool name to a spa object will enter spa_open() 995 * recursively, but that function already has all the 996 * necessary protection. 997 */ 998 if (!MUTEX_HELD(&spa_namespace_lock)) { 999 mutex_enter(&spa_namespace_lock); 1000 locked = B_TRUE; 1001 } 1002 1003 zv = pp->private; 1004 if (zv == NULL) { 1005 if (locked) 1006 mutex_exit(&spa_namespace_lock); 1007 return (SET_ERROR(ENXIO)); 1008 } 1009 1010 if (zv->zv_total_opens == 0) { 1011 err = zvol_first_open(zv); 1012 if (err) { 1013 if (locked) 1014 mutex_exit(&spa_namespace_lock); 1015 return (err); 1016 } 1017 pp->mediasize = zv->zv_volsize; 1018 pp->stripeoffset = 0; 1019 pp->stripesize = zv->zv_volblocksize; 1020 } 1021 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 1022 err = SET_ERROR(EROFS); 1023 goto out; 1024 } 1025 if (zv->zv_flags & ZVOL_EXCL) { 1026 err = SET_ERROR(EBUSY); 1027 goto out; 1028 } 1029#ifdef FEXCL 1030 if (flag & FEXCL) { 1031 if (zv->zv_total_opens != 0) { 1032 err = SET_ERROR(EBUSY); 1033 goto out; 1034 } 1035 zv->zv_flags |= ZVOL_EXCL; 1036 } 1037#endif 1038 1039 zv->zv_total_opens += count; 1040 if (locked) 1041 mutex_exit(&spa_namespace_lock); 1042 1043 return (err); 1044out: 1045 if (zv->zv_total_opens == 0) 1046 zvol_last_close(zv); 1047 if (locked) 1048 mutex_exit(&spa_namespace_lock); 1049 return (err); 1050} 1051 1052/*ARGSUSED*/ 1053static int 1054zvol_close(struct g_provider *pp, int flag, int count) 1055{ 1056 zvol_state_t *zv; 1057 int error = 0; 1058 boolean_t locked = B_FALSE; 1059 1060 /* See comment in zvol_open(). */ 1061 if (!MUTEX_HELD(&spa_namespace_lock)) { 1062 mutex_enter(&spa_namespace_lock); 1063 locked = B_TRUE; 1064 } 1065 1066 zv = pp->private; 1067 if (zv == NULL) { 1068 if (locked) 1069 mutex_exit(&spa_namespace_lock); 1070 return (SET_ERROR(ENXIO)); 1071 } 1072 1073 if (zv->zv_flags & ZVOL_EXCL) { 1074 ASSERT(zv->zv_total_opens == 1); 1075 zv->zv_flags &= ~ZVOL_EXCL; 1076 } 1077 1078 /* 1079 * If the open count is zero, this is a spurious close. 1080 * That indicates a bug in the kernel / DDI framework. 1081 */ 1082 ASSERT(zv->zv_total_opens != 0); 1083 1084 /* 1085 * You may get multiple opens, but only one close. 1086 */ 1087 zv->zv_total_opens -= count; 1088 1089 if (zv->zv_total_opens == 0) 1090 zvol_last_close(zv); 1091 1092 if (locked) 1093 mutex_exit(&spa_namespace_lock); 1094 return (error); 1095} 1096 1097static void 1098zvol_get_done(zgd_t *zgd, int error) 1099{ 1100 if (zgd->zgd_db) 1101 dmu_buf_rele(zgd->zgd_db, zgd); 1102 1103 zfs_range_unlock(zgd->zgd_rl); 1104 1105 if (error == 0 && zgd->zgd_bp) 1106 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1107 1108 kmem_free(zgd, sizeof (zgd_t)); 1109} 1110 1111/* 1112 * Get data to generate a TX_WRITE intent log record. 1113 */ 1114static int 1115zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1116{ 1117 zvol_state_t *zv = arg; 1118 objset_t *os = zv->zv_objset; 1119 uint64_t object = ZVOL_OBJ; 1120 uint64_t offset = lr->lr_offset; 1121 uint64_t size = lr->lr_length; /* length of user data */ 1122 blkptr_t *bp = &lr->lr_blkptr; 1123 dmu_buf_t *db; 1124 zgd_t *zgd; 1125 int error; 1126 1127 ASSERT(zio != NULL); 1128 ASSERT(size != 0); 1129 1130 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1131 zgd->zgd_zilog = zv->zv_zilog; 1132 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 1133 1134 /* 1135 * Write records come in two flavors: immediate and indirect. 1136 * For small writes it's cheaper to store the data with the 1137 * log record (immediate); for large writes it's cheaper to 1138 * sync the data and get a pointer to it (indirect) so that 1139 * we don't have to write the data twice. 1140 */ 1141 if (buf != NULL) { /* immediate write */ 1142 error = dmu_read(os, object, offset, size, buf, 1143 DMU_READ_NO_PREFETCH); 1144 } else { 1145 size = zv->zv_volblocksize; 1146 offset = P2ALIGN(offset, size); 1147 error = dmu_buf_hold(os, object, offset, zgd, &db, 1148 DMU_READ_NO_PREFETCH); 1149 if (error == 0) { 1150 blkptr_t *obp = dmu_buf_get_blkptr(db); 1151 if (obp) { 1152 ASSERT(BP_IS_HOLE(bp)); 1153 *bp = *obp; 1154 } 1155 1156 zgd->zgd_db = db; 1157 zgd->zgd_bp = bp; 1158 1159 ASSERT(db->db_offset == offset); 1160 ASSERT(db->db_size == size); 1161 1162 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1163 zvol_get_done, zgd); 1164 1165 if (error == 0) 1166 return (0); 1167 } 1168 } 1169 1170 zvol_get_done(zgd, error); 1171 1172 return (error); 1173} 1174 1175/* 1176 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 1177 * 1178 * We store data in the log buffers if it's small enough. 1179 * Otherwise we will later flush the data out via dmu_sync(). 1180 */ 1181ssize_t zvol_immediate_write_sz = 32768; 1182 1183static void 1184zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1185 boolean_t sync) 1186{ 1187 uint32_t blocksize = zv->zv_volblocksize; 1188 zilog_t *zilog = zv->zv_zilog; 1189 boolean_t slogging; 1190 ssize_t immediate_write_sz; 1191 1192 if (zil_replaying(zilog, tx)) 1193 return; 1194 1195 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1196 ? 0 : zvol_immediate_write_sz; 1197 1198 slogging = spa_has_slogs(zilog->zl_spa) && 1199 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 1200 1201 while (resid) { 1202 itx_t *itx; 1203 lr_write_t *lr; 1204 ssize_t len; 1205 itx_wr_state_t write_state; 1206 1207 /* 1208 * Unlike zfs_log_write() we can be called with 1209 * upto DMU_MAX_ACCESS/2 (5MB) writes. 1210 */ 1211 if (blocksize > immediate_write_sz && !slogging && 1212 resid >= blocksize && off % blocksize == 0) { 1213 write_state = WR_INDIRECT; /* uses dmu_sync */ 1214 len = blocksize; 1215 } else if (sync) { 1216 write_state = WR_COPIED; 1217 len = MIN(ZIL_MAX_LOG_DATA, resid); 1218 } else { 1219 write_state = WR_NEED_COPY; 1220 len = MIN(ZIL_MAX_LOG_DATA, resid); 1221 } 1222 1223 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1224 (write_state == WR_COPIED ? len : 0)); 1225 lr = (lr_write_t *)&itx->itx_lr; 1226 if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 1227 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1228 zil_itx_destroy(itx); 1229 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1230 lr = (lr_write_t *)&itx->itx_lr; 1231 write_state = WR_NEED_COPY; 1232 } 1233 1234 itx->itx_wr_state = write_state; 1235 if (write_state == WR_NEED_COPY) 1236 itx->itx_sod += len; 1237 lr->lr_foid = ZVOL_OBJ; 1238 lr->lr_offset = off; 1239 lr->lr_length = len; 1240 lr->lr_blkoff = 0; 1241 BP_ZERO(&lr->lr_blkptr); 1242 1243 itx->itx_private = zv; 1244 itx->itx_sync = sync; 1245 1246 zil_itx_assign(zilog, itx, tx); 1247 1248 off += len; 1249 resid -= len; 1250 } 1251} 1252 1253#ifdef sun 1254static int 1255zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset, 1256 uint64_t size, boolean_t doread, boolean_t isdump) 1257{ 1258 vdev_disk_t *dvd; 1259 int c; 1260 int numerrors = 0; 1261 1262 if (vd->vdev_ops == &vdev_mirror_ops || 1263 vd->vdev_ops == &vdev_replacing_ops || 1264 vd->vdev_ops == &vdev_spare_ops) { 1265 for (c = 0; c < vd->vdev_children; c++) { 1266 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1267 addr, offset, origoffset, size, doread, isdump); 1268 if (err != 0) { 1269 numerrors++; 1270 } else if (doread) { 1271 break; 1272 } 1273 } 1274 } 1275 1276 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops) 1277 return (numerrors < vd->vdev_children ? 0 : EIO); 1278 1279 if (doread && !vdev_readable(vd)) 1280 return (SET_ERROR(EIO)); 1281 else if (!doread && !vdev_writeable(vd)) 1282 return (SET_ERROR(EIO)); 1283 1284 if (vd->vdev_ops == &vdev_raidz_ops) { 1285 return (vdev_raidz_physio(vd, 1286 addr, size, offset, origoffset, doread, isdump)); 1287 } 1288 1289 offset += VDEV_LABEL_START_SIZE; 1290 1291 if (ddi_in_panic() || isdump) { 1292 ASSERT(!doread); 1293 if (doread) 1294 return (SET_ERROR(EIO)); 1295 dvd = vd->vdev_tsd; 1296 ASSERT3P(dvd, !=, NULL); 1297 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1298 lbtodb(size))); 1299 } else { 1300 dvd = vd->vdev_tsd; 1301 ASSERT3P(dvd, !=, NULL); 1302 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size, 1303 offset, doread ? B_READ : B_WRITE)); 1304 } 1305} 1306 1307static int 1308zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1309 boolean_t doread, boolean_t isdump) 1310{ 1311 vdev_t *vd; 1312 int error; 1313 zvol_extent_t *ze; 1314 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1315 1316 /* Must be sector aligned, and not stradle a block boundary. */ 1317 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1318 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1319 return (SET_ERROR(EINVAL)); 1320 } 1321 ASSERT(size <= zv->zv_volblocksize); 1322 1323 /* Locate the extent this belongs to */ 1324 ze = list_head(&zv->zv_extents); 1325 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1326 offset -= ze->ze_nblks * zv->zv_volblocksize; 1327 ze = list_next(&zv->zv_extents, ze); 1328 } 1329 1330 if (ze == NULL) 1331 return (SET_ERROR(EINVAL)); 1332 1333 if (!ddi_in_panic()) 1334 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1335 1336 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1337 offset += DVA_GET_OFFSET(&ze->ze_dva); 1338 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva), 1339 size, doread, isdump); 1340 1341 if (!ddi_in_panic()) 1342 spa_config_exit(spa, SCL_STATE, FTAG); 1343 1344 return (error); 1345} 1346#endif /* sun */ 1347 1348void 1349zvol_strategy(struct bio *bp) 1350{ 1351 zvol_state_t *zv; 1352 uint64_t off, volsize; 1353 size_t resid; 1354 char *addr; 1355 objset_t *os; 1356 rl_t *rl; 1357 int error = 0; 1358 boolean_t doread = 0; 1359 boolean_t is_dumpified; 1360 boolean_t sync; 1361 1362 if (bp->bio_to) 1363 zv = bp->bio_to->private; 1364 else 1365 zv = bp->bio_dev->si_drv2; 1366 1367 if (zv == NULL) { 1368 error = ENXIO; 1369 goto out; 1370 } 1371 1372 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) { 1373 error = EROFS; 1374 goto out; 1375 } 1376 1377 switch (bp->bio_cmd) { 1378 case BIO_FLUSH: 1379 goto sync; 1380 case BIO_READ: 1381 doread = 1; 1382 case BIO_WRITE: 1383 case BIO_DELETE: 1384 break; 1385 default: 1386 error = EOPNOTSUPP; 1387 goto out; 1388 } 1389 1390 off = bp->bio_offset; 1391 volsize = zv->zv_volsize; 1392 1393 os = zv->zv_objset; 1394 ASSERT(os != NULL); 1395 1396 addr = bp->bio_data; 1397 resid = bp->bio_length; 1398 1399 if (resid > 0 && (off < 0 || off >= volsize)) { 1400 error = EIO; 1401 goto out; 1402 } 1403 1404#ifdef illumos 1405 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED; 1406#else 1407 is_dumpified = B_FALSE; 1408#endif 1409 sync = !doread && !is_dumpified && 1410 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; 1411 1412 /* 1413 * There must be no buffer changes when doing a dmu_sync() because 1414 * we can't change the data whilst calculating the checksum. 1415 */ 1416 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1417 doread ? RL_READER : RL_WRITER); 1418 1419 if (bp->bio_cmd == BIO_DELETE) { 1420 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1421 error = dmu_tx_assign(tx, TXG_WAIT); 1422 if (error != 0) { 1423 dmu_tx_abort(tx); 1424 } else { 1425 zvol_log_truncate(zv, tx, off, resid, B_TRUE); 1426 dmu_tx_commit(tx); 1427 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1428 off, resid); 1429 resid = 0; 1430 } 1431 goto unlock; 1432 } 1433 1434 while (resid != 0 && off < volsize) { 1435 size_t size = MIN(resid, zvol_maxphys); 1436#ifdef illumos 1437 if (is_dumpified) { 1438 size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 1439 error = zvol_dumpio(zv, addr, off, size, 1440 doread, B_FALSE); 1441 } else if (doread) { 1442#else 1443 if (doread) { 1444#endif 1445 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1446 DMU_READ_PREFETCH); 1447 } else { 1448 dmu_tx_t *tx = dmu_tx_create(os); 1449 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1450 error = dmu_tx_assign(tx, TXG_WAIT); 1451 if (error) { 1452 dmu_tx_abort(tx); 1453 } else { 1454 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1455 zvol_log_write(zv, tx, off, size, sync); 1456 dmu_tx_commit(tx); 1457 } 1458 } 1459 if (error) { 1460 /* convert checksum errors into IO errors */ 1461 if (error == ECKSUM) 1462 error = SET_ERROR(EIO); 1463 break; 1464 } 1465 off += size; 1466 addr += size; 1467 resid -= size; 1468 } 1469unlock: 1470 zfs_range_unlock(rl); 1471 1472 bp->bio_completed = bp->bio_length - resid; 1473 if (bp->bio_completed < bp->bio_length && off > volsize) 1474 error = EINVAL; 1475 1476 if (sync) { 1477sync: 1478 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1479 } 1480out: 1481 if (bp->bio_to) 1482 g_io_deliver(bp, error); 1483 else 1484 biofinish(bp, NULL, error); 1485} 1486 1487#ifdef sun 1488/* 1489 * Set the buffer count to the zvol maximum transfer. 1490 * Using our own routine instead of the default minphys() 1491 * means that for larger writes we write bigger buffers on X86 1492 * (128K instead of 56K) and flush the disk write cache less often 1493 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1494 * 56K on X86 and 128K on sparc). 1495 */ 1496void 1497zvol_minphys(struct buf *bp) 1498{ 1499 if (bp->b_bcount > zvol_maxphys) 1500 bp->b_bcount = zvol_maxphys; 1501} 1502 1503int 1504zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1505{ 1506 minor_t minor = getminor(dev); 1507 zvol_state_t *zv; 1508 int error = 0; 1509 uint64_t size; 1510 uint64_t boff; 1511 uint64_t resid; 1512 1513 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1514 if (zv == NULL) 1515 return (SET_ERROR(ENXIO)); 1516 1517 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) 1518 return (SET_ERROR(EINVAL)); 1519 1520 boff = ldbtob(blkno); 1521 resid = ldbtob(nblocks); 1522 1523 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1524 1525 while (resid) { 1526 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1527 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1528 if (error) 1529 break; 1530 boff += size; 1531 addr += size; 1532 resid -= size; 1533 } 1534 1535 return (error); 1536} 1537 1538/*ARGSUSED*/ 1539int 1540zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1541{ 1542 minor_t minor = getminor(dev); 1543#else 1544int 1545zvol_read(struct cdev *dev, struct uio *uio, int ioflag) 1546{ 1547#endif 1548 zvol_state_t *zv; 1549 uint64_t volsize; 1550 rl_t *rl; 1551 int error = 0; 1552 1553#ifdef sun 1554 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1555 if (zv == NULL) 1556 return (SET_ERROR(ENXIO)); 1557#else 1558 zv = dev->si_drv2; 1559#endif 1560 1561 volsize = zv->zv_volsize; 1562 if (uio->uio_resid > 0 && 1563 (uio->uio_loffset < 0 || uio->uio_loffset > volsize)) 1564 return (SET_ERROR(EIO)); 1565 1566#ifdef illumos 1567 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1568 error = physio(zvol_strategy, NULL, dev, B_READ, 1569 zvol_minphys, uio); 1570 return (error); 1571 } 1572#endif 1573 1574 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1575 RL_READER); 1576 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1577 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1578 1579 /* don't read past the end */ 1580 if (bytes > volsize - uio->uio_loffset) 1581 bytes = volsize - uio->uio_loffset; 1582 1583 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1584 if (error) { 1585 /* convert checksum errors into IO errors */ 1586 if (error == ECKSUM) 1587 error = SET_ERROR(EIO); 1588 break; 1589 } 1590 } 1591 zfs_range_unlock(rl); 1592 return (error); 1593} 1594 1595#ifdef sun 1596/*ARGSUSED*/ 1597int 1598zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1599{ 1600 minor_t minor = getminor(dev); 1601#else 1602int 1603zvol_write(struct cdev *dev, struct uio *uio, int ioflag) 1604{ 1605#endif 1606 zvol_state_t *zv; 1607 uint64_t volsize; 1608 rl_t *rl; 1609 int error = 0; 1610 boolean_t sync; 1611 1612#ifdef sun 1613 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1614 if (zv == NULL) 1615 return (SET_ERROR(ENXIO)); 1616#else 1617 zv = dev->si_drv2; 1618#endif 1619 1620 volsize = zv->zv_volsize; 1621 if (uio->uio_resid > 0 && 1622 (uio->uio_loffset < 0 || uio->uio_loffset > volsize)) 1623 return (SET_ERROR(EIO)); 1624 1625#ifdef illumos 1626 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1627 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1628 zvol_minphys, uio); 1629 return (error); 1630 } 1631#endif 1632 1633#ifdef sun 1634 sync = !(zv->zv_flags & ZVOL_WCE) || 1635#else 1636 sync = 1637#endif 1638 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1639 1640 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1641 RL_WRITER); 1642 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1643 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1644 uint64_t off = uio->uio_loffset; 1645 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1646 1647 if (bytes > volsize - off) /* don't write past the end */ 1648 bytes = volsize - off; 1649 1650 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1651 error = dmu_tx_assign(tx, TXG_WAIT); 1652 if (error) { 1653 dmu_tx_abort(tx); 1654 break; 1655 } 1656 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1657 if (error == 0) 1658 zvol_log_write(zv, tx, off, bytes, sync); 1659 dmu_tx_commit(tx); 1660 1661 if (error) 1662 break; 1663 } 1664 zfs_range_unlock(rl); 1665 if (sync) 1666 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1667 return (error); 1668} 1669 1670#ifdef sun 1671int 1672zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1673{ 1674 struct uuid uuid = EFI_RESERVED; 1675 efi_gpe_t gpe = { 0 }; 1676 uint32_t crc; 1677 dk_efi_t efi; 1678 int length; 1679 char *ptr; 1680 1681 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1682 return (SET_ERROR(EFAULT)); 1683 ptr = (char *)(uintptr_t)efi.dki_data_64; 1684 length = efi.dki_length; 1685 /* 1686 * Some clients may attempt to request a PMBR for the 1687 * zvol. Currently this interface will return EINVAL to 1688 * such requests. These requests could be supported by 1689 * adding a check for lba == 0 and consing up an appropriate 1690 * PMBR. 1691 */ 1692 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1693 return (SET_ERROR(EINVAL)); 1694 1695 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1696 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1697 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1698 1699 if (efi.dki_lba == 1) { 1700 efi_gpt_t gpt = { 0 }; 1701 1702 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1703 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1704 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1705 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1706 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1707 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1708 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1709 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1710 gpt.efi_gpt_SizeOfPartitionEntry = 1711 LE_32(sizeof (efi_gpe_t)); 1712 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1713 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1714 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1715 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1716 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1717 flag)) 1718 return (SET_ERROR(EFAULT)); 1719 ptr += sizeof (gpt); 1720 length -= sizeof (gpt); 1721 } 1722 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1723 length), flag)) 1724 return (SET_ERROR(EFAULT)); 1725 return (0); 1726} 1727 1728/* 1729 * BEGIN entry points to allow external callers access to the volume. 1730 */ 1731/* 1732 * Return the volume parameters needed for access from an external caller. 1733 * These values are invariant as long as the volume is held open. 1734 */ 1735int 1736zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1737 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1738 void **rl_hdl, void **bonus_hdl) 1739{ 1740 zvol_state_t *zv; 1741 1742 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1743 if (zv == NULL) 1744 return (SET_ERROR(ENXIO)); 1745 if (zv->zv_flags & ZVOL_DUMPIFIED) 1746 return (SET_ERROR(ENXIO)); 1747 1748 ASSERT(blksize && max_xfer_len && minor_hdl && 1749 objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 1750 1751 *blksize = zv->zv_volblocksize; 1752 *max_xfer_len = (uint64_t)zvol_maxphys; 1753 *minor_hdl = zv; 1754 *objset_hdl = zv->zv_objset; 1755 *zil_hdl = zv->zv_zilog; 1756 *rl_hdl = &zv->zv_znode; 1757 *bonus_hdl = zv->zv_dbuf; 1758 return (0); 1759} 1760 1761/* 1762 * Return the current volume size to an external caller. 1763 * The size can change while the volume is open. 1764 */ 1765uint64_t 1766zvol_get_volume_size(void *minor_hdl) 1767{ 1768 zvol_state_t *zv = minor_hdl; 1769 1770 return (zv->zv_volsize); 1771} 1772 1773/* 1774 * Return the current WCE setting to an external caller. 1775 * The WCE setting can change while the volume is open. 1776 */ 1777int 1778zvol_get_volume_wce(void *minor_hdl) 1779{ 1780 zvol_state_t *zv = minor_hdl; 1781 1782 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 1783} 1784 1785/* 1786 * Entry point for external callers to zvol_log_write 1787 */ 1788void 1789zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 1790 boolean_t sync) 1791{ 1792 zvol_state_t *zv = minor_hdl; 1793 1794 zvol_log_write(zv, tx, off, resid, sync); 1795} 1796/* 1797 * END entry points to allow external callers access to the volume. 1798 */ 1799#endif /* sun */ 1800 1801/* 1802 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. 1803 */ 1804static void 1805zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, 1806 boolean_t sync) 1807{ 1808 itx_t *itx; 1809 lr_truncate_t *lr; 1810 zilog_t *zilog = zv->zv_zilog; 1811 1812 if (zil_replaying(zilog, tx)) 1813 return; 1814 1815 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 1816 lr = (lr_truncate_t *)&itx->itx_lr; 1817 lr->lr_foid = ZVOL_OBJ; 1818 lr->lr_offset = off; 1819 lr->lr_length = len; 1820 1821 itx->itx_sync = sync; 1822 zil_itx_assign(zilog, itx, tx); 1823} 1824 1825#ifdef sun 1826/* 1827 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1828 * Also a dirtbag dkio ioctl for unmap/free-block functionality. 1829 */ 1830/*ARGSUSED*/ 1831int 1832zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1833{ 1834 zvol_state_t *zv; 1835 struct dk_callback *dkc; 1836 int error = 0; 1837 rl_t *rl; 1838 1839 mutex_enter(&spa_namespace_lock); 1840 1841 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 1842 1843 if (zv == NULL) { 1844 mutex_exit(&spa_namespace_lock); 1845 return (SET_ERROR(ENXIO)); 1846 } 1847 ASSERT(zv->zv_total_opens > 0); 1848 1849 switch (cmd) { 1850 1851 case DKIOCINFO: 1852 { 1853 struct dk_cinfo dki; 1854 1855 bzero(&dki, sizeof (dki)); 1856 (void) strcpy(dki.dki_cname, "zvol"); 1857 (void) strcpy(dki.dki_dname, "zvol"); 1858 dki.dki_ctype = DKC_UNKNOWN; 1859 dki.dki_unit = getminor(dev); 1860 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); 1861 mutex_exit(&spa_namespace_lock); 1862 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1863 error = SET_ERROR(EFAULT); 1864 return (error); 1865 } 1866 1867 case DKIOCGMEDIAINFO: 1868 { 1869 struct dk_minfo dkm; 1870 1871 bzero(&dkm, sizeof (dkm)); 1872 dkm.dki_lbsize = 1U << zv->zv_min_bs; 1873 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1874 dkm.dki_media_type = DK_UNKNOWN; 1875 mutex_exit(&spa_namespace_lock); 1876 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1877 error = SET_ERROR(EFAULT); 1878 return (error); 1879 } 1880 1881 case DKIOCGMEDIAINFOEXT: 1882 { 1883 struct dk_minfo_ext dkmext; 1884 1885 bzero(&dkmext, sizeof (dkmext)); 1886 dkmext.dki_lbsize = 1U << zv->zv_min_bs; 1887 dkmext.dki_pbsize = zv->zv_volblocksize; 1888 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1889 dkmext.dki_media_type = DK_UNKNOWN; 1890 mutex_exit(&spa_namespace_lock); 1891 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag)) 1892 error = SET_ERROR(EFAULT); 1893 return (error); 1894 } 1895 1896 case DKIOCGETEFI: 1897 { 1898 uint64_t vs = zv->zv_volsize; 1899 uint8_t bs = zv->zv_min_bs; 1900 1901 mutex_exit(&spa_namespace_lock); 1902 error = zvol_getefi((void *)arg, flag, vs, bs); 1903 return (error); 1904 } 1905 1906 case DKIOCFLUSHWRITECACHE: 1907 dkc = (struct dk_callback *)arg; 1908 mutex_exit(&spa_namespace_lock); 1909 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1910 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1911 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1912 error = 0; 1913 } 1914 return (error); 1915 1916 case DKIOCGETWCE: 1917 { 1918 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1919 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1920 flag)) 1921 error = SET_ERROR(EFAULT); 1922 break; 1923 } 1924 case DKIOCSETWCE: 1925 { 1926 int wce; 1927 if (ddi_copyin((void *)arg, &wce, sizeof (int), 1928 flag)) { 1929 error = SET_ERROR(EFAULT); 1930 break; 1931 } 1932 if (wce) { 1933 zv->zv_flags |= ZVOL_WCE; 1934 mutex_exit(&spa_namespace_lock); 1935 } else { 1936 zv->zv_flags &= ~ZVOL_WCE; 1937 mutex_exit(&spa_namespace_lock); 1938 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1939 } 1940 return (0); 1941 } 1942 1943 case DKIOCGGEOM: 1944 case DKIOCGVTOC: 1945 /* 1946 * commands using these (like prtvtoc) expect ENOTSUP 1947 * since we're emulating an EFI label 1948 */ 1949 error = SET_ERROR(ENOTSUP); 1950 break; 1951 1952 case DKIOCDUMPINIT: 1953 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1954 RL_WRITER); 1955 error = zvol_dumpify(zv); 1956 zfs_range_unlock(rl); 1957 break; 1958 1959 case DKIOCDUMPFINI: 1960 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 1961 break; 1962 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1963 RL_WRITER); 1964 error = zvol_dump_fini(zv); 1965 zfs_range_unlock(rl); 1966 break; 1967 1968 case DKIOCFREE: 1969 { 1970 dkioc_free_t df; 1971 dmu_tx_t *tx; 1972 1973 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) { 1974 error = SET_ERROR(EFAULT); 1975 break; 1976 } 1977 1978 /* 1979 * Apply Postel's Law to length-checking. If they overshoot, 1980 * just blank out until the end, if there's a need to blank 1981 * out anything. 1982 */ 1983 if (df.df_start >= zv->zv_volsize) 1984 break; /* No need to do anything... */ 1985 if (df.df_start + df.df_length > zv->zv_volsize) 1986 df.df_length = DMU_OBJECT_END; 1987 1988 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length, 1989 RL_WRITER); 1990 tx = dmu_tx_create(zv->zv_objset); 1991 error = dmu_tx_assign(tx, TXG_WAIT); 1992 if (error != 0) { 1993 dmu_tx_abort(tx); 1994 } else { 1995 zvol_log_truncate(zv, tx, df.df_start, 1996 df.df_length, B_TRUE); 1997 dmu_tx_commit(tx); 1998 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1999 df.df_start, df.df_length); 2000 } 2001 2002 zfs_range_unlock(rl); 2003 2004 if (error == 0) { 2005 /* 2006 * If the write-cache is disabled or 'sync' property 2007 * is set to 'always' then treat this as a synchronous 2008 * operation (i.e. commit to zil). 2009 */ 2010 if (!(zv->zv_flags & ZVOL_WCE) || 2011 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) 2012 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2013 2014 /* 2015 * If the caller really wants synchronous writes, and 2016 * can't wait for them, don't return until the write 2017 * is done. 2018 */ 2019 if (df.df_flags & DF_WAIT_SYNC) { 2020 txg_wait_synced( 2021 dmu_objset_pool(zv->zv_objset), 0); 2022 } 2023 } 2024 break; 2025 } 2026 2027 default: 2028 error = SET_ERROR(ENOTTY); 2029 break; 2030 2031 } 2032 mutex_exit(&spa_namespace_lock); 2033 return (error); 2034} 2035#endif /* sun */ 2036 2037int 2038zvol_busy(void) 2039{ 2040 return (zvol_minors != 0); 2041} 2042 2043void 2044zvol_init(void) 2045{ 2046 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 2047 1) == 0); 2048 ZFS_LOG(1, "ZVOL Initialized."); 2049} 2050 2051void 2052zvol_fini(void) 2053{ 2054 ddi_soft_state_fini(&zfsdev_state); 2055 ZFS_LOG(1, "ZVOL Deinitialized."); 2056} 2057 2058#ifdef sun 2059/*ARGSUSED*/ 2060static int 2061zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx) 2062{ 2063 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 2064 2065 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 2066 return (1); 2067 return (0); 2068} 2069 2070/*ARGSUSED*/ 2071static void 2072zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx) 2073{ 2074 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 2075 2076 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx); 2077} 2078 2079static int 2080zvol_dump_init(zvol_state_t *zv, boolean_t resize) 2081{ 2082 dmu_tx_t *tx; 2083 int error; 2084 objset_t *os = zv->zv_objset; 2085 spa_t *spa = dmu_objset_spa(os); 2086 vdev_t *vd = spa->spa_root_vdev; 2087 nvlist_t *nv = NULL; 2088 uint64_t version = spa_version(spa); 2089 enum zio_checksum checksum; 2090 2091 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2092 ASSERT(vd->vdev_ops == &vdev_root_ops); 2093 2094 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 2095 DMU_OBJECT_END); 2096 /* wait for dmu_free_long_range to actually free the blocks */ 2097 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2098 2099 /* 2100 * If the pool on which the dump device is being initialized has more 2101 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is 2102 * enabled. If so, bump that feature's counter to indicate that the 2103 * feature is active. We also check the vdev type to handle the 2104 * following case: 2105 * # zpool create test raidz disk1 disk2 disk3 2106 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev), 2107 * the raidz vdev itself has 3 children. 2108 */ 2109 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) { 2110 if (!spa_feature_is_enabled(spa, 2111 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 2112 return (SET_ERROR(ENOTSUP)); 2113 (void) dsl_sync_task(spa_name(spa), 2114 zfs_mvdev_dump_feature_check, 2115 zfs_mvdev_dump_activate_feature_sync, NULL, 2); 2116 } 2117 2118 tx = dmu_tx_create(os); 2119 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2120 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2121 error = dmu_tx_assign(tx, TXG_WAIT); 2122 if (error) { 2123 dmu_tx_abort(tx); 2124 return (error); 2125 } 2126 2127 /* 2128 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum 2129 * function. Otherwise, use the old default -- OFF. 2130 */ 2131 checksum = spa_feature_is_active(spa, 2132 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY : 2133 ZIO_CHECKSUM_OFF; 2134 2135 /* 2136 * If we are resizing the dump device then we only need to 2137 * update the refreservation to match the newly updated 2138 * zvolsize. Otherwise, we save off the original state of the 2139 * zvol so that we can restore them if the zvol is ever undumpified. 2140 */ 2141 if (resize) { 2142 error = zap_update(os, ZVOL_ZAP_OBJ, 2143 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2144 &zv->zv_volsize, tx); 2145 } else { 2146 uint64_t checksum, compress, refresrv, vbs, dedup; 2147 2148 error = dsl_prop_get_integer(zv->zv_name, 2149 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 2150 error = error ? error : dsl_prop_get_integer(zv->zv_name, 2151 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL); 2152 error = error ? error : dsl_prop_get_integer(zv->zv_name, 2153 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL); 2154 error = error ? error : dsl_prop_get_integer(zv->zv_name, 2155 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL); 2156 if (version >= SPA_VERSION_DEDUP) { 2157 error = error ? error : 2158 dsl_prop_get_integer(zv->zv_name, 2159 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 2160 } 2161 2162 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2163 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 2164 &compress, tx); 2165 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2166 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx); 2167 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2168 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2169 &refresrv, tx); 2170 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2171 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 2172 &vbs, tx); 2173 error = error ? error : dmu_object_set_blocksize( 2174 os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx); 2175 if (version >= SPA_VERSION_DEDUP) { 2176 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2177 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 2178 &dedup, tx); 2179 } 2180 if (error == 0) 2181 zv->zv_volblocksize = SPA_MAXBLOCKSIZE; 2182 } 2183 dmu_tx_commit(tx); 2184 2185 /* 2186 * We only need update the zvol's property if we are initializing 2187 * the dump area for the first time. 2188 */ 2189 if (!resize) { 2190 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2191 VERIFY(nvlist_add_uint64(nv, 2192 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 2193 VERIFY(nvlist_add_uint64(nv, 2194 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 2195 ZIO_COMPRESS_OFF) == 0); 2196 VERIFY(nvlist_add_uint64(nv, 2197 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 2198 checksum) == 0); 2199 if (version >= SPA_VERSION_DEDUP) { 2200 VERIFY(nvlist_add_uint64(nv, 2201 zfs_prop_to_name(ZFS_PROP_DEDUP), 2202 ZIO_CHECKSUM_OFF) == 0); 2203 } 2204 2205 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2206 nv, NULL); 2207 nvlist_free(nv); 2208 2209 if (error) 2210 return (error); 2211 } 2212 2213 /* Allocate the space for the dump */ 2214 error = zvol_prealloc(zv); 2215 return (error); 2216} 2217 2218static int 2219zvol_dumpify(zvol_state_t *zv) 2220{ 2221 int error = 0; 2222 uint64_t dumpsize = 0; 2223 dmu_tx_t *tx; 2224 objset_t *os = zv->zv_objset; 2225 2226 if (zv->zv_flags & ZVOL_RDONLY) 2227 return (SET_ERROR(EROFS)); 2228 2229 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 2230 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 2231 boolean_t resize = (dumpsize > 0); 2232 2233 if ((error = zvol_dump_init(zv, resize)) != 0) { 2234 (void) zvol_dump_fini(zv); 2235 return (error); 2236 } 2237 } 2238 2239 /* 2240 * Build up our lba mapping. 2241 */ 2242 error = zvol_get_lbas(zv); 2243 if (error) { 2244 (void) zvol_dump_fini(zv); 2245 return (error); 2246 } 2247 2248 tx = dmu_tx_create(os); 2249 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2250 error = dmu_tx_assign(tx, TXG_WAIT); 2251 if (error) { 2252 dmu_tx_abort(tx); 2253 (void) zvol_dump_fini(zv); 2254 return (error); 2255 } 2256 2257 zv->zv_flags |= ZVOL_DUMPIFIED; 2258 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 2259 &zv->zv_volsize, tx); 2260 dmu_tx_commit(tx); 2261 2262 if (error) { 2263 (void) zvol_dump_fini(zv); 2264 return (error); 2265 } 2266 2267 txg_wait_synced(dmu_objset_pool(os), 0); 2268 return (0); 2269} 2270 2271static int 2272zvol_dump_fini(zvol_state_t *zv) 2273{ 2274 dmu_tx_t *tx; 2275 objset_t *os = zv->zv_objset; 2276 nvlist_t *nv; 2277 int error = 0; 2278 uint64_t checksum, compress, refresrv, vbs, dedup; 2279 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 2280 2281 /* 2282 * Attempt to restore the zvol back to its pre-dumpified state. 2283 * This is a best-effort attempt as it's possible that not all 2284 * of these properties were initialized during the dumpify process 2285 * (i.e. error during zvol_dump_init). 2286 */ 2287 2288 tx = dmu_tx_create(os); 2289 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2290 error = dmu_tx_assign(tx, TXG_WAIT); 2291 if (error) { 2292 dmu_tx_abort(tx); 2293 return (error); 2294 } 2295 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 2296 dmu_tx_commit(tx); 2297 2298 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2299 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 2300 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2301 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 2302 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2303 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 2304 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2305 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 2306 2307 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2308 (void) nvlist_add_uint64(nv, 2309 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 2310 (void) nvlist_add_uint64(nv, 2311 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 2312 (void) nvlist_add_uint64(nv, 2313 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 2314 if (version >= SPA_VERSION_DEDUP && 2315 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2316 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 2317 (void) nvlist_add_uint64(nv, 2318 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 2319 } 2320 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2321 nv, NULL); 2322 nvlist_free(nv); 2323 2324 zvol_free_extents(zv); 2325 zv->zv_flags &= ~ZVOL_DUMPIFIED; 2326 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 2327 /* wait for dmu_free_long_range to actually free the blocks */ 2328 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2329 tx = dmu_tx_create(os); 2330 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2331 error = dmu_tx_assign(tx, TXG_WAIT); 2332 if (error) { 2333 dmu_tx_abort(tx); 2334 return (error); 2335 } 2336 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 2337 zv->zv_volblocksize = vbs; 2338 dmu_tx_commit(tx); 2339 2340 return (0); 2341} 2342#endif /* sun */ 2343 2344static void 2345zvol_geom_run(zvol_state_t *zv) 2346{ 2347 struct g_provider *pp; 2348 2349 pp = zv->zv_provider; 2350 g_error_provider(pp, 0); 2351 2352 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0, 2353 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER)); 2354} 2355 2356static void 2357zvol_geom_destroy(zvol_state_t *zv) 2358{ 2359 struct g_provider *pp; 2360 2361 g_topology_assert(); 2362 2363 mtx_lock(&zv->zv_queue_mtx); 2364 zv->zv_state = 1; 2365 wakeup_one(&zv->zv_queue); 2366 while (zv->zv_state != 2) 2367 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0); 2368 mtx_destroy(&zv->zv_queue_mtx); 2369 2370 pp = zv->zv_provider; 2371 zv->zv_provider = NULL; 2372 pp->private = NULL; 2373 g_wither_geom(pp->geom, ENXIO); 2374} 2375 2376static int 2377zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) 2378{ 2379 int count, error, flags; 2380 2381 g_topology_assert(); 2382 2383 /* 2384 * To make it easier we expect either open or close, but not both 2385 * at the same time. 2386 */ 2387 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) || 2388 (acr <= 0 && acw <= 0 && ace <= 0), 2389 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", 2390 pp->name, acr, acw, ace)); 2391 2392 if (pp->private == NULL) { 2393 if (acr <= 0 && acw <= 0 && ace <= 0) 2394 return (0); 2395 return (pp->error); 2396 } 2397 2398 /* 2399 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0, 2400 * because GEOM already handles that and handles it a bit differently. 2401 * GEOM allows for multiple read/exclusive consumers and ZFS allows 2402 * only one exclusive consumer, no matter if it is reader or writer. 2403 * I like better the way GEOM works so I'll leave it for GEOM to 2404 * decide what to do. 2405 */ 2406 2407 count = acr + acw + ace; 2408 if (count == 0) 2409 return (0); 2410 2411 flags = 0; 2412 if (acr != 0 || ace != 0) 2413 flags |= FREAD; 2414 if (acw != 0) 2415 flags |= FWRITE; 2416 2417 g_topology_unlock(); 2418 if (count > 0) 2419 error = zvol_open(pp, flags, count); 2420 else 2421 error = zvol_close(pp, flags, -count); 2422 g_topology_lock(); 2423 return (error); 2424} 2425 2426static void 2427zvol_geom_start(struct bio *bp) 2428{ 2429 zvol_state_t *zv; 2430 boolean_t first; 2431 2432 zv = bp->bio_to->private; 2433 ASSERT(zv != NULL); 2434 switch (bp->bio_cmd) { 2435 case BIO_FLUSH: 2436 if (!THREAD_CAN_SLEEP()) 2437 goto enqueue; 2438 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2439 g_io_deliver(bp, 0); 2440 break; 2441 case BIO_READ: 2442 case BIO_WRITE: 2443 case BIO_DELETE: 2444 if (!THREAD_CAN_SLEEP()) 2445 goto enqueue; 2446 zvol_strategy(bp); 2447 break; 2448 case BIO_GETATTR: 2449 if (g_handleattr_int(bp, "GEOM::candelete", 1)) 2450 return; 2451 /* FALLTHROUGH */ 2452 default: 2453 g_io_deliver(bp, EOPNOTSUPP); 2454 break; 2455 } 2456 return; 2457 2458enqueue: 2459 mtx_lock(&zv->zv_queue_mtx); 2460 first = (bioq_first(&zv->zv_queue) == NULL); 2461 bioq_insert_tail(&zv->zv_queue, bp); 2462 mtx_unlock(&zv->zv_queue_mtx); 2463 if (first) 2464 wakeup_one(&zv->zv_queue); 2465} 2466 2467static void 2468zvol_geom_worker(void *arg) 2469{ 2470 zvol_state_t *zv; 2471 struct bio *bp; 2472 2473 thread_lock(curthread); 2474 sched_prio(curthread, PRIBIO); 2475 thread_unlock(curthread); 2476 2477 zv = arg; 2478 for (;;) { 2479 mtx_lock(&zv->zv_queue_mtx); 2480 bp = bioq_takefirst(&zv->zv_queue); 2481 if (bp == NULL) { 2482 if (zv->zv_state == 1) { 2483 zv->zv_state = 2; 2484 wakeup(&zv->zv_state); 2485 mtx_unlock(&zv->zv_queue_mtx); 2486 kthread_exit(); 2487 } 2488 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP, 2489 "zvol:io", 0); 2490 continue; 2491 } 2492 mtx_unlock(&zv->zv_queue_mtx); 2493 switch (bp->bio_cmd) { 2494 case BIO_FLUSH: 2495 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2496 g_io_deliver(bp, 0); 2497 break; 2498 case BIO_READ: 2499 case BIO_WRITE: 2500 zvol_strategy(bp); 2501 break; 2502 } 2503 } 2504} 2505 2506extern boolean_t dataset_name_hidden(const char *name); 2507 2508static int 2509zvol_create_snapshots(objset_t *os, const char *name) 2510{ 2511 uint64_t cookie, obj; 2512 char *sname; 2513 int error, len; 2514 2515 cookie = obj = 0; 2516 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2517 2518#if 0 2519 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL, 2520 DS_FIND_SNAPSHOTS); 2521#endif 2522 2523 for (;;) { 2524 len = snprintf(sname, MAXPATHLEN, "%s@", name); 2525 if (len >= MAXPATHLEN) { 2526 dmu_objset_rele(os, FTAG); 2527 error = ENAMETOOLONG; 2528 break; 2529 } 2530 2531 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 2532 error = dmu_snapshot_list_next(os, MAXPATHLEN - len, 2533 sname + len, &obj, &cookie, NULL); 2534 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 2535 if (error != 0) { 2536 if (error == ENOENT) 2537 error = 0; 2538 break; 2539 } 2540 2541 if ((error = zvol_create_minor(sname)) != 0) { 2542 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2543 sname, error); 2544 break; 2545 } 2546 } 2547 2548 kmem_free(sname, MAXPATHLEN); 2549 return (error); 2550} 2551 2552int 2553zvol_create_minors(const char *name) 2554{ 2555 uint64_t cookie; 2556 objset_t *os; 2557 char *osname, *p; 2558 int error, len; 2559 2560 if (dataset_name_hidden(name)) 2561 return (0); 2562 2563 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2564 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2565 name, error); 2566 return (error); 2567 } 2568 if (dmu_objset_type(os) == DMU_OST_ZVOL) { 2569 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG); 2570 dsl_pool_rele(dmu_objset_pool(os), FTAG); 2571 if ((error = zvol_create_minor(name)) == 0) 2572 error = zvol_create_snapshots(os, name); 2573 else { 2574 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2575 name, error); 2576 } 2577 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG); 2578 dsl_dataset_rele(os->os_dsl_dataset, FTAG); 2579 return (error); 2580 } 2581 if (dmu_objset_type(os) != DMU_OST_ZFS) { 2582 dmu_objset_rele(os, FTAG); 2583 return (0); 2584 } 2585 2586 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2587 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) { 2588 dmu_objset_rele(os, FTAG); 2589 kmem_free(osname, MAXPATHLEN); 2590 return (ENOENT); 2591 } 2592 p = osname + strlen(osname); 2593 len = MAXPATHLEN - (p - osname); 2594 2595#if 0 2596 /* Prefetch the datasets. */ 2597 cookie = 0; 2598 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) { 2599 if (!dataset_name_hidden(osname)) 2600 (void) dmu_objset_prefetch(osname, NULL); 2601 } 2602#endif 2603 2604 cookie = 0; 2605 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL, 2606 &cookie) == 0) { 2607 dmu_objset_rele(os, FTAG); 2608 (void)zvol_create_minors(osname); 2609 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2610 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2611 name, error); 2612 return (error); 2613 } 2614 } 2615 2616 dmu_objset_rele(os, FTAG); 2617 kmem_free(osname, MAXPATHLEN); 2618 return (0); 2619} 2620 2621static void 2622zvol_rename_minor(zvol_state_t *zv, const char *newname) 2623{ 2624 struct g_geom *gp; 2625 struct g_provider *pp; 2626 struct cdev *dev; 2627 2628 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2629 2630 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 2631 g_topology_lock(); 2632 pp = zv->zv_provider; 2633 ASSERT(pp != NULL); 2634 gp = pp->geom; 2635 ASSERT(gp != NULL); 2636 2637 zv->zv_provider = NULL; 2638 g_wither_provider(pp, ENXIO); 2639 2640 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname); 2641 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 2642 pp->sectorsize = DEV_BSIZE; 2643 pp->mediasize = zv->zv_volsize; 2644 pp->private = zv; 2645 zv->zv_provider = pp; 2646 g_error_provider(pp, 0); 2647 g_topology_unlock(); 2648 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { 2649 dev = zv->zv_dev; 2650 ASSERT(dev != NULL); 2651 zv->zv_dev = NULL; 2652 destroy_dev(dev); 2653 2654 if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, 2655 &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 2656 0640, "%s/%s", ZVOL_DRIVER, newname) == 0) { 2657 zv->zv_dev = dev; 2658 dev->si_iosize_max = MAXPHYS; 2659 dev->si_drv2 = zv; 2660 } 2661 } 2662 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name)); 2663} 2664 2665void 2666zvol_rename_minors(const char *oldname, const char *newname) 2667{ 2668 char name[MAXPATHLEN]; 2669 struct g_provider *pp; 2670 struct g_geom *gp; 2671 size_t oldnamelen, newnamelen; 2672 zvol_state_t *zv; 2673 char *namebuf; 2674 2675 oldnamelen = strlen(oldname); 2676 newnamelen = strlen(newname); 2677 2678 DROP_GIANT(); 2679 mutex_enter(&spa_namespace_lock); 2680 2681 LIST_FOREACH(zv, &all_zvols, zv_links) { 2682 if (strcmp(zv->zv_name, oldname) == 0) { 2683 zvol_rename_minor(zv, newname); 2684 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && 2685 (zv->zv_name[oldnamelen] == '/' || 2686 zv->zv_name[oldnamelen] == '@')) { 2687 snprintf(name, sizeof(name), "%s%c%s", newname, 2688 zv->zv_name[oldnamelen], 2689 zv->zv_name + oldnamelen + 1); 2690 zvol_rename_minor(zv, name); 2691 } 2692 } 2693 2694 mutex_exit(&spa_namespace_lock); 2695 PICKUP_GIANT(); 2696} 2697 2698static int 2699zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td) 2700{ 2701 zvol_state_t *zv; 2702 int err = 0; 2703 2704 mutex_enter(&spa_namespace_lock); 2705 zv = dev->si_drv2; 2706 if (zv == NULL) { 2707 mutex_exit(&spa_namespace_lock); 2708 return(ENXIO); /* zvol_create_minor() not done yet */ 2709 } 2710 2711 if (zv->zv_total_opens == 0) 2712 err = zvol_first_open(zv); 2713 if (err) { 2714 mutex_exit(&spa_namespace_lock); 2715 return (err); 2716 } 2717 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 2718 err = SET_ERROR(EROFS); 2719 goto out; 2720 } 2721 if (zv->zv_flags & ZVOL_EXCL) { 2722 err = SET_ERROR(EBUSY); 2723 goto out; 2724 } 2725#ifdef FEXCL 2726 if (flags & FEXCL) { 2727 if (zv->zv_total_opens != 0) { 2728 err = SET_ERROR(EBUSY); 2729 goto out; 2730 } 2731 zv->zv_flags |= ZVOL_EXCL; 2732 } 2733#endif 2734 2735 zv->zv_total_opens++; 2736 mutex_exit(&spa_namespace_lock); 2737 return (err); 2738out: 2739 if (zv->zv_total_opens == 0) 2740 zvol_last_close(zv); 2741 mutex_exit(&spa_namespace_lock); 2742 return (err); 2743} 2744 2745static int 2746zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td) 2747{ 2748 zvol_state_t *zv; 2749 int err = 0; 2750 2751 mutex_enter(&spa_namespace_lock); 2752 zv = dev->si_drv2; 2753 if (zv == NULL) { 2754 mutex_exit(&spa_namespace_lock); 2755 return(ENXIO); 2756 } 2757 2758 if (zv->zv_flags & ZVOL_EXCL) { 2759 ASSERT(zv->zv_total_opens == 1); 2760 zv->zv_flags &= ~ZVOL_EXCL; 2761 } 2762 2763 /* 2764 * If the open count is zero, this is a spurious close. 2765 * That indicates a bug in the kernel / DDI framework. 2766 */ 2767 ASSERT(zv->zv_total_opens != 0); 2768 2769 /* 2770 * You may get multiple opens, but only one close. 2771 */ 2772 zv->zv_total_opens--; 2773 2774 if (zv->zv_total_opens == 0) 2775 zvol_last_close(zv); 2776 2777 mutex_exit(&spa_namespace_lock); 2778 return (0); 2779} 2780 2781static int 2782zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 2783{ 2784 zvol_state_t *zv; 2785 rl_t *rl; 2786 off_t offset, length, chunk; 2787 int i, error; 2788 u_int u; 2789 2790 zv = dev->si_drv2; 2791 2792 error = 0; 2793 KASSERT(zv->zv_total_opens > 0, 2794 ("Device with zero access count in zvol_d_ioctl")); 2795 2796 i = IOCPARM_LEN(cmd); 2797 switch (cmd) { 2798 case DIOCGSECTORSIZE: 2799 *(u_int *)data = DEV_BSIZE; 2800 break; 2801 case DIOCGMEDIASIZE: 2802 *(off_t *)data = zv->zv_volsize; 2803 break; 2804 case DIOCGFLUSH: 2805 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2806 break; 2807 case DIOCGDELETE: 2808 offset = ((off_t *)data)[0]; 2809 length = ((off_t *)data)[1]; 2810 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 || 2811 offset < 0 || offset >= zv->zv_volsize || 2812 length <= 0) { 2813 printf("%s: offset=%jd length=%jd\n", __func__, offset, 2814 length); 2815 error = EINVAL; 2816 break; 2817 } 2818 2819 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER); 2820 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 2821 error = dmu_tx_assign(tx, TXG_WAIT); 2822 if (error != 0) { 2823 dmu_tx_abort(tx); 2824 } else { 2825 zvol_log_truncate(zv, tx, offset, length, B_TRUE); 2826 dmu_tx_commit(tx); 2827 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 2828 offset, length); 2829 } 2830 zfs_range_unlock(rl); 2831 if (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) 2832 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2833 break; 2834 case DIOCGSTRIPESIZE: 2835 *(off_t *)data = zv->zv_volblocksize; 2836 break; 2837 case DIOCGSTRIPEOFFSET: 2838 *(off_t *)data = 0; 2839 break; 2840 default: 2841 error = ENOIOCTL; 2842 } 2843 2844 return (error); 2845} 2846