zvol.c revision 265678
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 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); 319 ma->ma_blks++; 320 321 /* Abort immediately if we have encountered gang blocks */ 322 if (BP_IS_GANG(bp)) 323 return (SET_ERROR(EFRAGS)); 324 325 /* 326 * See if the block is at the end of the previous extent. 327 */ 328 ze = list_tail(&ma->ma_zv->zv_extents); 329 if (ze && 330 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && 331 DVA_GET_OFFSET(BP_IDENTITY(bp)) == 332 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { 333 ze->ze_nblks++; 334 return (0); 335 } 336 337 dprintf_bp(bp, "%s", "next blkptr:"); 338 339 /* start a new extent */ 340 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); 341 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ 342 ze->ze_nblks = 1; 343 list_insert_tail(&ma->ma_zv->zv_extents, ze); 344 return (0); 345} 346 347static void 348zvol_free_extents(zvol_state_t *zv) 349{ 350 zvol_extent_t *ze; 351 352 while (ze = list_head(&zv->zv_extents)) { 353 list_remove(&zv->zv_extents, ze); 354 kmem_free(ze, sizeof (zvol_extent_t)); 355 } 356} 357 358static int 359zvol_get_lbas(zvol_state_t *zv) 360{ 361 objset_t *os = zv->zv_objset; 362 struct maparg ma; 363 int err; 364 365 ma.ma_zv = zv; 366 ma.ma_blks = 0; 367 zvol_free_extents(zv); 368 369 /* commit any in-flight changes before traversing the dataset */ 370 txg_wait_synced(dmu_objset_pool(os), 0); 371 err = traverse_dataset(dmu_objset_ds(os), 0, 372 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); 373 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { 374 zvol_free_extents(zv); 375 return (err ? err : EIO); 376 } 377 378 return (0); 379} 380 381/* ARGSUSED */ 382void 383zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 384{ 385 zfs_creat_t *zct = arg; 386 nvlist_t *nvprops = zct->zct_props; 387 int error; 388 uint64_t volblocksize, volsize; 389 390 VERIFY(nvlist_lookup_uint64(nvprops, 391 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 392 if (nvlist_lookup_uint64(nvprops, 393 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 394 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 395 396 /* 397 * These properties must be removed from the list so the generic 398 * property setting step won't apply to them. 399 */ 400 VERIFY(nvlist_remove_all(nvprops, 401 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 402 (void) nvlist_remove_all(nvprops, 403 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 404 405 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 406 DMU_OT_NONE, 0, tx); 407 ASSERT(error == 0); 408 409 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 410 DMU_OT_NONE, 0, tx); 411 ASSERT(error == 0); 412 413 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 414 ASSERT(error == 0); 415} 416 417/* 418 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we 419 * implement DKIOCFREE/free-long-range. 420 */ 421static int 422zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap) 423{ 424 uint64_t offset, length; 425 426 if (byteswap) 427 byteswap_uint64_array(lr, sizeof (*lr)); 428 429 offset = lr->lr_offset; 430 length = lr->lr_length; 431 432 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length)); 433} 434 435/* 436 * Replay a TX_WRITE ZIL transaction that didn't get committed 437 * after a system failure 438 */ 439static int 440zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) 441{ 442 objset_t *os = zv->zv_objset; 443 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 444 uint64_t offset, length; 445 dmu_tx_t *tx; 446 int error; 447 448 if (byteswap) 449 byteswap_uint64_array(lr, sizeof (*lr)); 450 451 offset = lr->lr_offset; 452 length = lr->lr_length; 453 454 /* If it's a dmu_sync() block, write the whole block */ 455 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 456 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 457 if (length < blocksize) { 458 offset -= offset % blocksize; 459 length = blocksize; 460 } 461 } 462 463 tx = dmu_tx_create(os); 464 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 465 error = dmu_tx_assign(tx, TXG_WAIT); 466 if (error) { 467 dmu_tx_abort(tx); 468 } else { 469 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 470 dmu_tx_commit(tx); 471 } 472 473 return (error); 474} 475 476/* ARGSUSED */ 477static int 478zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) 479{ 480 return (SET_ERROR(ENOTSUP)); 481} 482 483/* 484 * Callback vectors for replaying records. 485 * Only TX_WRITE and TX_TRUNCATE are needed for zvol. 486 */ 487zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 488 zvol_replay_err, /* 0 no such transaction type */ 489 zvol_replay_err, /* TX_CREATE */ 490 zvol_replay_err, /* TX_MKDIR */ 491 zvol_replay_err, /* TX_MKXATTR */ 492 zvol_replay_err, /* TX_SYMLINK */ 493 zvol_replay_err, /* TX_REMOVE */ 494 zvol_replay_err, /* TX_RMDIR */ 495 zvol_replay_err, /* TX_LINK */ 496 zvol_replay_err, /* TX_RENAME */ 497 zvol_replay_write, /* TX_WRITE */ 498 zvol_replay_truncate, /* TX_TRUNCATE */ 499 zvol_replay_err, /* TX_SETATTR */ 500 zvol_replay_err, /* TX_ACL */ 501 zvol_replay_err, /* TX_CREATE_ACL */ 502 zvol_replay_err, /* TX_CREATE_ATTR */ 503 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 504 zvol_replay_err, /* TX_MKDIR_ACL */ 505 zvol_replay_err, /* TX_MKDIR_ATTR */ 506 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 507 zvol_replay_err, /* TX_WRITE2 */ 508}; 509 510#ifdef sun 511int 512zvol_name2minor(const char *name, minor_t *minor) 513{ 514 zvol_state_t *zv; 515 516 mutex_enter(&spa_namespace_lock); 517 zv = zvol_minor_lookup(name); 518 if (minor && zv) 519 *minor = zv->zv_minor; 520 mutex_exit(&spa_namespace_lock); 521 return (zv ? 0 : -1); 522} 523#endif /* sun */ 524 525/* 526 * Create a minor node (plus a whole lot more) for the specified volume. 527 */ 528int 529zvol_create_minor(const char *name) 530{ 531 zfs_soft_state_t *zs; 532 zvol_state_t *zv; 533 objset_t *os; 534 struct cdev *dev; 535 struct g_provider *pp; 536 struct g_geom *gp; 537 dmu_object_info_t doi; 538 uint64_t volsize, mode; 539 int error; 540 541 ZFS_LOG(1, "Creating ZVOL %s...", name); 542 543 mutex_enter(&spa_namespace_lock); 544 545 if (zvol_minor_lookup(name) != NULL) { 546 mutex_exit(&spa_namespace_lock); 547 return (SET_ERROR(EEXIST)); 548 } 549 550 /* lie and say we're read-only */ 551 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os); 552 553 if (error) { 554 mutex_exit(&spa_namespace_lock); 555 return (error); 556 } 557 558#ifdef sun 559 if ((minor = zfsdev_minor_alloc()) == 0) { 560 dmu_objset_disown(os, FTAG); 561 mutex_exit(&spa_namespace_lock); 562 return (SET_ERROR(ENXIO)); 563 } 564 565 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { 566 dmu_objset_disown(os, FTAG); 567 mutex_exit(&spa_namespace_lock); 568 return (SET_ERROR(EAGAIN)); 569 } 570 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, 571 (char *)name); 572 573 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); 574 575 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, 576 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 577 ddi_soft_state_free(zfsdev_state, minor); 578 dmu_objset_disown(os, FTAG); 579 mutex_exit(&spa_namespace_lock); 580 return (SET_ERROR(EAGAIN)); 581 } 582 583 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); 584 585 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, 586 minor, DDI_PSEUDO, 0) == DDI_FAILURE) { 587 ddi_remove_minor_node(zfs_dip, chrbuf); 588 ddi_soft_state_free(zfsdev_state, minor); 589 dmu_objset_disown(os, FTAG); 590 mutex_exit(&spa_namespace_lock); 591 return (SET_ERROR(EAGAIN)); 592 } 593 594 zs = ddi_get_soft_state(zfsdev_state, minor); 595 zs->zss_type = ZSST_ZVOL; 596 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); 597#else /* !sun */ 598 599 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP); 600 zv->zv_state = 0; 601 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 602 if (error) { 603 kmem_free(zv, sizeof(*zv)); 604 dmu_objset_disown(os, zvol_tag); 605 mutex_exit(&spa_namespace_lock); 606 return (error); 607 } 608 error = dsl_prop_get_integer(name, 609 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL); 610 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT) 611 mode = volmode; 612 613 DROP_GIANT(); 614 zv->zv_volsize = volsize; 615 zv->zv_volmode = mode; 616 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 617 g_topology_lock(); 618 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name); 619 gp->start = zvol_geom_start; 620 gp->access = zvol_geom_access; 621 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name); 622 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 623 pp->sectorsize = DEV_BSIZE; 624 pp->mediasize = zv->zv_volsize; 625 pp->private = zv; 626 627 zv->zv_provider = pp; 628 bioq_init(&zv->zv_queue); 629 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF); 630 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { 631 if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, 632 &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 633 0640, "%s/%s", ZVOL_DRIVER, name) != 0) { 634 kmem_free(zv, sizeof(*zv)); 635 dmu_objset_disown(os, FTAG); 636 mutex_exit(&spa_namespace_lock); 637 return (SET_ERROR(ENXIO)); 638 } 639 zv->zv_dev = dev; 640 dev->si_iosize_max = MAXPHYS; 641 dev->si_drv2 = zv; 642 } 643 LIST_INSERT_HEAD(&all_zvols, zv, zv_links); 644#endif /* !sun */ 645 646 (void) strlcpy(zv->zv_name, name, MAXPATHLEN); 647 zv->zv_min_bs = DEV_BSHIFT; 648 zv->zv_objset = os; 649 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) 650 zv->zv_flags |= ZVOL_RDONLY; 651 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); 652 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, 653 sizeof (rl_t), offsetof(rl_t, r_node)); 654 list_create(&zv->zv_extents, sizeof (zvol_extent_t), 655 offsetof(zvol_extent_t, ze_node)); 656 /* get and cache the blocksize */ 657 error = dmu_object_info(os, ZVOL_OBJ, &doi); 658 ASSERT(error == 0); 659 zv->zv_volblocksize = doi.doi_data_block_size; 660 661 if (spa_writeable(dmu_objset_spa(os))) { 662 if (zil_replay_disable) 663 zil_destroy(dmu_objset_zil(os), B_FALSE); 664 else 665 zil_replay(os, zv, zvol_replay_vector); 666 } 667 dmu_objset_disown(os, FTAG); 668 zv->zv_objset = NULL; 669 670 zvol_minors++; 671 672 mutex_exit(&spa_namespace_lock); 673 674#ifndef sun 675 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 676 zvol_geom_run(zv); 677 g_topology_unlock(); 678 } 679 PICKUP_GIANT(); 680#endif 681 682 ZFS_LOG(1, "ZVOL %s created.", name); 683 684 return (0); 685} 686 687/* 688 * Remove minor node for the specified volume. 689 */ 690static int 691zvol_remove_zv(zvol_state_t *zv) 692{ 693#ifdef sun 694 minor_t minor = zv->zv_minor; 695#endif 696 697 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 698 if (zv->zv_total_opens != 0) 699 return (SET_ERROR(EBUSY)); 700 701 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); 702 703#ifdef sun 704 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); 705 ddi_remove_minor_node(zfs_dip, nmbuf); 706#else 707 LIST_REMOVE(zv, zv_links); 708 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 709 g_topology_lock(); 710 zvol_geom_destroy(zv); 711 g_topology_unlock(); 712 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) 713 destroy_dev(zv->zv_dev); 714#endif /* sun */ 715 716 avl_destroy(&zv->zv_znode.z_range_avl); 717 mutex_destroy(&zv->zv_znode.z_range_lock); 718 719 kmem_free(zv, sizeof(*zv)); 720 721 zvol_minors--; 722 return (0); 723} 724 725int 726zvol_remove_minor(const char *name) 727{ 728 zvol_state_t *zv; 729 int rc; 730 731 mutex_enter(&spa_namespace_lock); 732 if ((zv = zvol_minor_lookup(name)) == NULL) { 733 mutex_exit(&spa_namespace_lock); 734 return (SET_ERROR(ENXIO)); 735 } 736 rc = zvol_remove_zv(zv); 737 mutex_exit(&spa_namespace_lock); 738 return (rc); 739} 740 741int 742zvol_first_open(zvol_state_t *zv) 743{ 744 objset_t *os; 745 uint64_t volsize; 746 int error; 747 uint64_t readonly; 748 749 /* lie and say we're read-only */ 750 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, 751 zvol_tag, &os); 752 if (error) 753 return (error); 754 755 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 756 if (error) { 757 ASSERT(error == 0); 758 dmu_objset_disown(os, zvol_tag); 759 return (error); 760 } 761 zv->zv_objset = os; 762 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); 763 if (error) { 764 dmu_objset_disown(os, zvol_tag); 765 return (error); 766 } 767 zv->zv_volsize = volsize; 768 zv->zv_zilog = zil_open(os, zvol_get_data); 769 zvol_size_changed(zv); 770 771 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, 772 NULL) == 0); 773 if (readonly || dmu_objset_is_snapshot(os) || 774 !spa_writeable(dmu_objset_spa(os))) 775 zv->zv_flags |= ZVOL_RDONLY; 776 else 777 zv->zv_flags &= ~ZVOL_RDONLY; 778 return (error); 779} 780 781void 782zvol_last_close(zvol_state_t *zv) 783{ 784 zil_close(zv->zv_zilog); 785 zv->zv_zilog = NULL; 786 787 dmu_buf_rele(zv->zv_dbuf, zvol_tag); 788 zv->zv_dbuf = NULL; 789 790 /* 791 * Evict cached data 792 */ 793 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && 794 !(zv->zv_flags & ZVOL_RDONLY)) 795 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 796 dmu_objset_evict_dbufs(zv->zv_objset); 797 798 dmu_objset_disown(zv->zv_objset, zvol_tag); 799 zv->zv_objset = NULL; 800} 801 802#ifdef sun 803int 804zvol_prealloc(zvol_state_t *zv) 805{ 806 objset_t *os = zv->zv_objset; 807 dmu_tx_t *tx; 808 uint64_t refd, avail, usedobjs, availobjs; 809 uint64_t resid = zv->zv_volsize; 810 uint64_t off = 0; 811 812 /* Check the space usage before attempting to allocate the space */ 813 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); 814 if (avail < zv->zv_volsize) 815 return (SET_ERROR(ENOSPC)); 816 817 /* Free old extents if they exist */ 818 zvol_free_extents(zv); 819 820 while (resid != 0) { 821 int error; 822 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); 823 824 tx = dmu_tx_create(os); 825 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 826 error = dmu_tx_assign(tx, TXG_WAIT); 827 if (error) { 828 dmu_tx_abort(tx); 829 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); 830 return (error); 831 } 832 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); 833 dmu_tx_commit(tx); 834 off += bytes; 835 resid -= bytes; 836 } 837 txg_wait_synced(dmu_objset_pool(os), 0); 838 839 return (0); 840} 841#endif /* sun */ 842 843static int 844zvol_update_volsize(objset_t *os, uint64_t volsize) 845{ 846 dmu_tx_t *tx; 847 int error; 848 849 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 850 851 tx = dmu_tx_create(os); 852 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 853 error = dmu_tx_assign(tx, TXG_WAIT); 854 if (error) { 855 dmu_tx_abort(tx); 856 return (error); 857 } 858 859 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 860 &volsize, tx); 861 dmu_tx_commit(tx); 862 863 if (error == 0) 864 error = dmu_free_long_range(os, 865 ZVOL_OBJ, volsize, DMU_OBJECT_END); 866 return (error); 867} 868 869void 870zvol_remove_minors(const char *name) 871{ 872 zvol_state_t *zv, *tzv; 873 size_t namelen; 874 875 namelen = strlen(name); 876 877 DROP_GIANT(); 878 mutex_enter(&spa_namespace_lock); 879 880 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) { 881 if (strcmp(zv->zv_name, name) == 0 || 882 (strncmp(zv->zv_name, name, namelen) == 0 && 883 zv->zv_name[namelen] == '/')) { 884 (void) zvol_remove_zv(zv); 885 } 886 } 887 888 mutex_exit(&spa_namespace_lock); 889 PICKUP_GIANT(); 890} 891 892int 893zvol_set_volsize(const char *name, major_t maj, uint64_t volsize) 894{ 895 zvol_state_t *zv = NULL; 896 objset_t *os; 897 int error; 898 dmu_object_info_t doi; 899 uint64_t old_volsize = 0ULL; 900 uint64_t readonly; 901 902 mutex_enter(&spa_namespace_lock); 903 zv = zvol_minor_lookup(name); 904 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 905 mutex_exit(&spa_namespace_lock); 906 return (error); 907 } 908 909 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || 910 (error = zvol_check_volsize(volsize, 911 doi.doi_data_block_size)) != 0) 912 goto out; 913 914 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, 915 NULL) == 0); 916 if (readonly) { 917 error = EROFS; 918 goto out; 919 } 920 921 error = zvol_update_volsize(os, volsize); 922 /* 923 * Reinitialize the dump area to the new size. If we 924 * failed to resize the dump area then restore it back to 925 * its original size. 926 */ 927 if (zv && error == 0) { 928#ifdef ZVOL_DUMP 929 if (zv->zv_flags & ZVOL_DUMPIFIED) { 930 old_volsize = zv->zv_volsize; 931 zv->zv_volsize = volsize; 932 if ((error = zvol_dumpify(zv)) != 0 || 933 (error = dumpvp_resize()) != 0) { 934 (void) zvol_update_volsize(os, old_volsize); 935 zv->zv_volsize = old_volsize; 936 error = zvol_dumpify(zv); 937 } 938 } 939#endif /* ZVOL_DUMP */ 940 if (error == 0) { 941 zv->zv_volsize = volsize; 942 zvol_size_changed(zv); 943 } 944 } 945 946#ifdef sun 947 /* 948 * Generate a LUN expansion event. 949 */ 950 if (zv && error == 0) { 951 sysevent_id_t eid; 952 nvlist_t *attr; 953 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 954 955 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, 956 zv->zv_minor); 957 958 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 959 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 960 961 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 962 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 963 964 nvlist_free(attr); 965 kmem_free(physpath, MAXPATHLEN); 966 } 967#endif /* sun */ 968 969out: 970 dmu_objset_rele(os, FTAG); 971 972 mutex_exit(&spa_namespace_lock); 973 974 return (error); 975} 976 977/*ARGSUSED*/ 978static int 979zvol_open(struct g_provider *pp, int flag, int count) 980{ 981 zvol_state_t *zv; 982 int err = 0; 983 boolean_t locked = B_FALSE; 984 985 /* 986 * Protect against recursively entering spa_namespace_lock 987 * when spa_open() is used for a pool on a (local) ZVOL(s). 988 * This is needed since we replaced upstream zfsdev_state_lock 989 * with spa_namespace_lock in the ZVOL code. 990 * We are using the same trick as spa_open(). 991 * Note that calls in zvol_first_open which need to resolve 992 * pool name to a spa object will enter spa_open() 993 * recursively, but that function already has all the 994 * necessary protection. 995 */ 996 if (!MUTEX_HELD(&spa_namespace_lock)) { 997 mutex_enter(&spa_namespace_lock); 998 locked = B_TRUE; 999 } 1000 1001 zv = pp->private; 1002 if (zv == NULL) { 1003 if (locked) 1004 mutex_exit(&spa_namespace_lock); 1005 return (SET_ERROR(ENXIO)); 1006 } 1007 1008 if (zv->zv_total_opens == 0) { 1009 err = zvol_first_open(zv); 1010 if (err) { 1011 if (locked) 1012 mutex_exit(&spa_namespace_lock); 1013 return (err); 1014 } 1015 pp->mediasize = zv->zv_volsize; 1016 pp->stripeoffset = 0; 1017 pp->stripesize = zv->zv_volblocksize; 1018 } 1019 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 1020 err = SET_ERROR(EROFS); 1021 goto out; 1022 } 1023 if (zv->zv_flags & ZVOL_EXCL) { 1024 err = SET_ERROR(EBUSY); 1025 goto out; 1026 } 1027#ifdef FEXCL 1028 if (flag & FEXCL) { 1029 if (zv->zv_total_opens != 0) { 1030 err = SET_ERROR(EBUSY); 1031 goto out; 1032 } 1033 zv->zv_flags |= ZVOL_EXCL; 1034 } 1035#endif 1036 1037 zv->zv_total_opens += count; 1038 if (locked) 1039 mutex_exit(&spa_namespace_lock); 1040 1041 return (err); 1042out: 1043 if (zv->zv_total_opens == 0) 1044 zvol_last_close(zv); 1045 if (locked) 1046 mutex_exit(&spa_namespace_lock); 1047 return (err); 1048} 1049 1050/*ARGSUSED*/ 1051static int 1052zvol_close(struct g_provider *pp, int flag, int count) 1053{ 1054 zvol_state_t *zv; 1055 int error = 0; 1056 boolean_t locked = B_FALSE; 1057 1058 /* See comment in zvol_open(). */ 1059 if (!MUTEX_HELD(&spa_namespace_lock)) { 1060 mutex_enter(&spa_namespace_lock); 1061 locked = B_TRUE; 1062 } 1063 1064 zv = pp->private; 1065 if (zv == NULL) { 1066 if (locked) 1067 mutex_exit(&spa_namespace_lock); 1068 return (SET_ERROR(ENXIO)); 1069 } 1070 1071 if (zv->zv_flags & ZVOL_EXCL) { 1072 ASSERT(zv->zv_total_opens == 1); 1073 zv->zv_flags &= ~ZVOL_EXCL; 1074 } 1075 1076 /* 1077 * If the open count is zero, this is a spurious close. 1078 * That indicates a bug in the kernel / DDI framework. 1079 */ 1080 ASSERT(zv->zv_total_opens != 0); 1081 1082 /* 1083 * You may get multiple opens, but only one close. 1084 */ 1085 zv->zv_total_opens -= count; 1086 1087 if (zv->zv_total_opens == 0) 1088 zvol_last_close(zv); 1089 1090 if (locked) 1091 mutex_exit(&spa_namespace_lock); 1092 return (error); 1093} 1094 1095static void 1096zvol_get_done(zgd_t *zgd, int error) 1097{ 1098 if (zgd->zgd_db) 1099 dmu_buf_rele(zgd->zgd_db, zgd); 1100 1101 zfs_range_unlock(zgd->zgd_rl); 1102 1103 if (error == 0 && zgd->zgd_bp) 1104 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1105 1106 kmem_free(zgd, sizeof (zgd_t)); 1107} 1108 1109/* 1110 * Get data to generate a TX_WRITE intent log record. 1111 */ 1112static int 1113zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1114{ 1115 zvol_state_t *zv = arg; 1116 objset_t *os = zv->zv_objset; 1117 uint64_t object = ZVOL_OBJ; 1118 uint64_t offset = lr->lr_offset; 1119 uint64_t size = lr->lr_length; /* length of user data */ 1120 blkptr_t *bp = &lr->lr_blkptr; 1121 dmu_buf_t *db; 1122 zgd_t *zgd; 1123 int error; 1124 1125 ASSERT(zio != NULL); 1126 ASSERT(size != 0); 1127 1128 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1129 zgd->zgd_zilog = zv->zv_zilog; 1130 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); 1131 1132 /* 1133 * Write records come in two flavors: immediate and indirect. 1134 * For small writes it's cheaper to store the data with the 1135 * log record (immediate); for large writes it's cheaper to 1136 * sync the data and get a pointer to it (indirect) so that 1137 * we don't have to write the data twice. 1138 */ 1139 if (buf != NULL) { /* immediate write */ 1140 error = dmu_read(os, object, offset, size, buf, 1141 DMU_READ_NO_PREFETCH); 1142 } else { 1143 size = zv->zv_volblocksize; 1144 offset = P2ALIGN(offset, size); 1145 error = dmu_buf_hold(os, object, offset, zgd, &db, 1146 DMU_READ_NO_PREFETCH); 1147 if (error == 0) { 1148 blkptr_t *obp = dmu_buf_get_blkptr(db); 1149 if (obp) { 1150 ASSERT(BP_IS_HOLE(bp)); 1151 *bp = *obp; 1152 } 1153 1154 zgd->zgd_db = db; 1155 zgd->zgd_bp = bp; 1156 1157 ASSERT(db->db_offset == offset); 1158 ASSERT(db->db_size == size); 1159 1160 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1161 zvol_get_done, zgd); 1162 1163 if (error == 0) 1164 return (0); 1165 } 1166 } 1167 1168 zvol_get_done(zgd, error); 1169 1170 return (error); 1171} 1172 1173/* 1174 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 1175 * 1176 * We store data in the log buffers if it's small enough. 1177 * Otherwise we will later flush the data out via dmu_sync(). 1178 */ 1179ssize_t zvol_immediate_write_sz = 32768; 1180 1181static void 1182zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, 1183 boolean_t sync) 1184{ 1185 uint32_t blocksize = zv->zv_volblocksize; 1186 zilog_t *zilog = zv->zv_zilog; 1187 boolean_t slogging; 1188 ssize_t immediate_write_sz; 1189 1190 if (zil_replaying(zilog, tx)) 1191 return; 1192 1193 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 1194 ? 0 : zvol_immediate_write_sz; 1195 1196 slogging = spa_has_slogs(zilog->zl_spa) && 1197 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); 1198 1199 while (resid) { 1200 itx_t *itx; 1201 lr_write_t *lr; 1202 ssize_t len; 1203 itx_wr_state_t write_state; 1204 1205 /* 1206 * Unlike zfs_log_write() we can be called with 1207 * upto DMU_MAX_ACCESS/2 (5MB) writes. 1208 */ 1209 if (blocksize > immediate_write_sz && !slogging && 1210 resid >= blocksize && off % blocksize == 0) { 1211 write_state = WR_INDIRECT; /* uses dmu_sync */ 1212 len = blocksize; 1213 } else if (sync) { 1214 write_state = WR_COPIED; 1215 len = MIN(ZIL_MAX_LOG_DATA, resid); 1216 } else { 1217 write_state = WR_NEED_COPY; 1218 len = MIN(ZIL_MAX_LOG_DATA, resid); 1219 } 1220 1221 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 1222 (write_state == WR_COPIED ? len : 0)); 1223 lr = (lr_write_t *)&itx->itx_lr; 1224 if (write_state == WR_COPIED && dmu_read(zv->zv_objset, 1225 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { 1226 zil_itx_destroy(itx); 1227 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 1228 lr = (lr_write_t *)&itx->itx_lr; 1229 write_state = WR_NEED_COPY; 1230 } 1231 1232 itx->itx_wr_state = write_state; 1233 if (write_state == WR_NEED_COPY) 1234 itx->itx_sod += len; 1235 lr->lr_foid = ZVOL_OBJ; 1236 lr->lr_offset = off; 1237 lr->lr_length = len; 1238 lr->lr_blkoff = 0; 1239 BP_ZERO(&lr->lr_blkptr); 1240 1241 itx->itx_private = zv; 1242 itx->itx_sync = sync; 1243 1244 zil_itx_assign(zilog, itx, tx); 1245 1246 off += len; 1247 resid -= len; 1248 } 1249} 1250 1251#ifdef sun 1252static int 1253zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset, 1254 uint64_t size, boolean_t doread, boolean_t isdump) 1255{ 1256 vdev_disk_t *dvd; 1257 int c; 1258 int numerrors = 0; 1259 1260 if (vd->vdev_ops == &vdev_mirror_ops || 1261 vd->vdev_ops == &vdev_replacing_ops || 1262 vd->vdev_ops == &vdev_spare_ops) { 1263 for (c = 0; c < vd->vdev_children; c++) { 1264 int err = zvol_dumpio_vdev(vd->vdev_child[c], 1265 addr, offset, origoffset, size, doread, isdump); 1266 if (err != 0) { 1267 numerrors++; 1268 } else if (doread) { 1269 break; 1270 } 1271 } 1272 } 1273 1274 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops) 1275 return (numerrors < vd->vdev_children ? 0 : EIO); 1276 1277 if (doread && !vdev_readable(vd)) 1278 return (SET_ERROR(EIO)); 1279 else if (!doread && !vdev_writeable(vd)) 1280 return (SET_ERROR(EIO)); 1281 1282 if (vd->vdev_ops == &vdev_raidz_ops) { 1283 return (vdev_raidz_physio(vd, 1284 addr, size, offset, origoffset, doread, isdump)); 1285 } 1286 1287 offset += VDEV_LABEL_START_SIZE; 1288 1289 if (ddi_in_panic() || isdump) { 1290 ASSERT(!doread); 1291 if (doread) 1292 return (SET_ERROR(EIO)); 1293 dvd = vd->vdev_tsd; 1294 ASSERT3P(dvd, !=, NULL); 1295 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), 1296 lbtodb(size))); 1297 } else { 1298 dvd = vd->vdev_tsd; 1299 ASSERT3P(dvd, !=, NULL); 1300 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size, 1301 offset, doread ? B_READ : B_WRITE)); 1302 } 1303} 1304 1305static int 1306zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, 1307 boolean_t doread, boolean_t isdump) 1308{ 1309 vdev_t *vd; 1310 int error; 1311 zvol_extent_t *ze; 1312 spa_t *spa = dmu_objset_spa(zv->zv_objset); 1313 1314 /* Must be sector aligned, and not stradle a block boundary. */ 1315 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || 1316 P2BOUNDARY(offset, size, zv->zv_volblocksize)) { 1317 return (SET_ERROR(EINVAL)); 1318 } 1319 ASSERT(size <= zv->zv_volblocksize); 1320 1321 /* Locate the extent this belongs to */ 1322 ze = list_head(&zv->zv_extents); 1323 while (offset >= ze->ze_nblks * zv->zv_volblocksize) { 1324 offset -= ze->ze_nblks * zv->zv_volblocksize; 1325 ze = list_next(&zv->zv_extents, ze); 1326 } 1327 1328 if (ze == NULL) 1329 return (SET_ERROR(EINVAL)); 1330 1331 if (!ddi_in_panic()) 1332 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 1333 1334 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); 1335 offset += DVA_GET_OFFSET(&ze->ze_dva); 1336 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva), 1337 size, doread, isdump); 1338 1339 if (!ddi_in_panic()) 1340 spa_config_exit(spa, SCL_STATE, FTAG); 1341 1342 return (error); 1343} 1344#endif /* sun */ 1345 1346void 1347zvol_strategy(struct bio *bp) 1348{ 1349 zvol_state_t *zv; 1350 uint64_t off, volsize; 1351 size_t resid; 1352 char *addr; 1353 objset_t *os; 1354 rl_t *rl; 1355 int error = 0; 1356 boolean_t doread = 0; 1357 boolean_t is_dumpified; 1358 boolean_t sync; 1359 1360 if (bp->bio_to) 1361 zv = bp->bio_to->private; 1362 else 1363 zv = bp->bio_dev->si_drv2; 1364 1365 if (zv == NULL) { 1366 error = ENXIO; 1367 goto out; 1368 } 1369 1370 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) { 1371 error = EROFS; 1372 goto out; 1373 } 1374 1375 switch (bp->bio_cmd) { 1376 case BIO_FLUSH: 1377 goto sync; 1378 case BIO_READ: 1379 doread = 1; 1380 case BIO_WRITE: 1381 case BIO_DELETE: 1382 break; 1383 default: 1384 error = EOPNOTSUPP; 1385 goto out; 1386 } 1387 1388 off = bp->bio_offset; 1389 volsize = zv->zv_volsize; 1390 1391 os = zv->zv_objset; 1392 ASSERT(os != NULL); 1393 1394 addr = bp->bio_data; 1395 resid = bp->bio_length; 1396 1397 if (resid > 0 && (off < 0 || off >= volsize)) { 1398 error = EIO; 1399 goto out; 1400 } 1401 1402#ifdef illumos 1403 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED; 1404#else 1405 is_dumpified = B_FALSE; 1406#endif 1407 sync = !doread && !is_dumpified && 1408 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; 1409 1410 /* 1411 * There must be no buffer changes when doing a dmu_sync() because 1412 * we can't change the data whilst calculating the checksum. 1413 */ 1414 rl = zfs_range_lock(&zv->zv_znode, off, resid, 1415 doread ? RL_READER : RL_WRITER); 1416 1417 if (bp->bio_cmd == BIO_DELETE) { 1418 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1419 error = dmu_tx_assign(tx, TXG_WAIT); 1420 if (error != 0) { 1421 dmu_tx_abort(tx); 1422 } else { 1423 zvol_log_truncate(zv, tx, off, resid, B_TRUE); 1424 dmu_tx_commit(tx); 1425 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1426 off, resid); 1427 resid = 0; 1428 } 1429 goto unlock; 1430 } 1431 1432 while (resid != 0 && off < volsize) { 1433 size_t size = MIN(resid, zvol_maxphys); 1434#ifdef illumos 1435 if (is_dumpified) { 1436 size = MIN(size, P2END(off, zv->zv_volblocksize) - off); 1437 error = zvol_dumpio(zv, addr, off, size, 1438 doread, B_FALSE); 1439 } else if (doread) { 1440#else 1441 if (doread) { 1442#endif 1443 error = dmu_read(os, ZVOL_OBJ, off, size, addr, 1444 DMU_READ_PREFETCH); 1445 } else { 1446 dmu_tx_t *tx = dmu_tx_create(os); 1447 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); 1448 error = dmu_tx_assign(tx, TXG_WAIT); 1449 if (error) { 1450 dmu_tx_abort(tx); 1451 } else { 1452 dmu_write(os, ZVOL_OBJ, off, size, addr, tx); 1453 zvol_log_write(zv, tx, off, size, sync); 1454 dmu_tx_commit(tx); 1455 } 1456 } 1457 if (error) { 1458 /* convert checksum errors into IO errors */ 1459 if (error == ECKSUM) 1460 error = SET_ERROR(EIO); 1461 break; 1462 } 1463 off += size; 1464 addr += size; 1465 resid -= size; 1466 } 1467unlock: 1468 zfs_range_unlock(rl); 1469 1470 bp->bio_completed = bp->bio_length - resid; 1471 if (bp->bio_completed < bp->bio_length && off > volsize) 1472 error = EINVAL; 1473 1474 if (sync) { 1475sync: 1476 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1477 } 1478out: 1479 if (bp->bio_to) 1480 g_io_deliver(bp, error); 1481 else 1482 biofinish(bp, NULL, error); 1483} 1484 1485#ifdef sun 1486/* 1487 * Set the buffer count to the zvol maximum transfer. 1488 * Using our own routine instead of the default minphys() 1489 * means that for larger writes we write bigger buffers on X86 1490 * (128K instead of 56K) and flush the disk write cache less often 1491 * (every zvol_maxphys - currently 1MB) instead of minphys (currently 1492 * 56K on X86 and 128K on sparc). 1493 */ 1494void 1495zvol_minphys(struct buf *bp) 1496{ 1497 if (bp->b_bcount > zvol_maxphys) 1498 bp->b_bcount = zvol_maxphys; 1499} 1500 1501int 1502zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) 1503{ 1504 minor_t minor = getminor(dev); 1505 zvol_state_t *zv; 1506 int error = 0; 1507 uint64_t size; 1508 uint64_t boff; 1509 uint64_t resid; 1510 1511 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1512 if (zv == NULL) 1513 return (SET_ERROR(ENXIO)); 1514 1515 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) 1516 return (SET_ERROR(EINVAL)); 1517 1518 boff = ldbtob(blkno); 1519 resid = ldbtob(nblocks); 1520 1521 VERIFY3U(boff + resid, <=, zv->zv_volsize); 1522 1523 while (resid) { 1524 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); 1525 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); 1526 if (error) 1527 break; 1528 boff += size; 1529 addr += size; 1530 resid -= size; 1531 } 1532 1533 return (error); 1534} 1535 1536/*ARGSUSED*/ 1537int 1538zvol_read(dev_t dev, uio_t *uio, cred_t *cr) 1539{ 1540 minor_t minor = getminor(dev); 1541#else 1542int 1543zvol_read(struct cdev *dev, struct uio *uio, int ioflag) 1544{ 1545#endif 1546 zvol_state_t *zv; 1547 uint64_t volsize; 1548 rl_t *rl; 1549 int error = 0; 1550 1551#ifdef sun 1552 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1553 if (zv == NULL) 1554 return (SET_ERROR(ENXIO)); 1555#else 1556 zv = dev->si_drv2; 1557#endif 1558 1559 volsize = zv->zv_volsize; 1560 if (uio->uio_resid > 0 && 1561 (uio->uio_loffset < 0 || uio->uio_loffset > volsize)) 1562 return (SET_ERROR(EIO)); 1563 1564#ifdef illumos 1565 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1566 error = physio(zvol_strategy, NULL, dev, B_READ, 1567 zvol_minphys, uio); 1568 return (error); 1569 } 1570#endif 1571 1572 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1573 RL_READER); 1574 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1575 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1576 1577 /* don't read past the end */ 1578 if (bytes > volsize - uio->uio_loffset) 1579 bytes = volsize - uio->uio_loffset; 1580 1581 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); 1582 if (error) { 1583 /* convert checksum errors into IO errors */ 1584 if (error == ECKSUM) 1585 error = SET_ERROR(EIO); 1586 break; 1587 } 1588 } 1589 zfs_range_unlock(rl); 1590 return (error); 1591} 1592 1593#ifdef sun 1594/*ARGSUSED*/ 1595int 1596zvol_write(dev_t dev, uio_t *uio, cred_t *cr) 1597{ 1598 minor_t minor = getminor(dev); 1599#else 1600int 1601zvol_write(struct cdev *dev, struct uio *uio, int ioflag) 1602{ 1603#endif 1604 zvol_state_t *zv; 1605 uint64_t volsize; 1606 rl_t *rl; 1607 int error = 0; 1608 boolean_t sync; 1609 1610#ifdef sun 1611 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1612 if (zv == NULL) 1613 return (SET_ERROR(ENXIO)); 1614#else 1615 zv = dev->si_drv2; 1616#endif 1617 1618 volsize = zv->zv_volsize; 1619 if (uio->uio_resid > 0 && 1620 (uio->uio_loffset < 0 || uio->uio_loffset > volsize)) 1621 return (SET_ERROR(EIO)); 1622 1623#ifdef illumos 1624 if (zv->zv_flags & ZVOL_DUMPIFIED) { 1625 error = physio(zvol_strategy, NULL, dev, B_WRITE, 1626 zvol_minphys, uio); 1627 return (error); 1628 } 1629#endif 1630 1631 sync = !(zv->zv_flags & ZVOL_WCE) || 1632 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); 1633 1634 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, 1635 RL_WRITER); 1636 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { 1637 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); 1638 uint64_t off = uio->uio_loffset; 1639 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 1640 1641 if (bytes > volsize - off) /* don't write past the end */ 1642 bytes = volsize - off; 1643 1644 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); 1645 error = dmu_tx_assign(tx, TXG_WAIT); 1646 if (error) { 1647 dmu_tx_abort(tx); 1648 break; 1649 } 1650 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); 1651 if (error == 0) 1652 zvol_log_write(zv, tx, off, bytes, sync); 1653 dmu_tx_commit(tx); 1654 1655 if (error) 1656 break; 1657 } 1658 zfs_range_unlock(rl); 1659 if (sync) 1660 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1661 return (error); 1662} 1663 1664#ifdef sun 1665int 1666zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) 1667{ 1668 struct uuid uuid = EFI_RESERVED; 1669 efi_gpe_t gpe = { 0 }; 1670 uint32_t crc; 1671 dk_efi_t efi; 1672 int length; 1673 char *ptr; 1674 1675 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) 1676 return (SET_ERROR(EFAULT)); 1677 ptr = (char *)(uintptr_t)efi.dki_data_64; 1678 length = efi.dki_length; 1679 /* 1680 * Some clients may attempt to request a PMBR for the 1681 * zvol. Currently this interface will return EINVAL to 1682 * such requests. These requests could be supported by 1683 * adding a check for lba == 0 and consing up an appropriate 1684 * PMBR. 1685 */ 1686 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) 1687 return (SET_ERROR(EINVAL)); 1688 1689 gpe.efi_gpe_StartingLBA = LE_64(34ULL); 1690 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); 1691 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); 1692 1693 if (efi.dki_lba == 1) { 1694 efi_gpt_t gpt = { 0 }; 1695 1696 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1697 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); 1698 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); 1699 gpt.efi_gpt_MyLBA = LE_64(1ULL); 1700 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); 1701 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); 1702 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1703 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); 1704 gpt.efi_gpt_SizeOfPartitionEntry = 1705 LE_32(sizeof (efi_gpe_t)); 1706 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); 1707 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); 1708 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); 1709 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); 1710 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), 1711 flag)) 1712 return (SET_ERROR(EFAULT)); 1713 ptr += sizeof (gpt); 1714 length -= sizeof (gpt); 1715 } 1716 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), 1717 length), flag)) 1718 return (SET_ERROR(EFAULT)); 1719 return (0); 1720} 1721 1722/* 1723 * BEGIN entry points to allow external callers access to the volume. 1724 */ 1725/* 1726 * Return the volume parameters needed for access from an external caller. 1727 * These values are invariant as long as the volume is held open. 1728 */ 1729int 1730zvol_get_volume_params(minor_t minor, uint64_t *blksize, 1731 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, 1732 void **rl_hdl, void **bonus_hdl) 1733{ 1734 zvol_state_t *zv; 1735 1736 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); 1737 if (zv == NULL) 1738 return (SET_ERROR(ENXIO)); 1739 if (zv->zv_flags & ZVOL_DUMPIFIED) 1740 return (SET_ERROR(ENXIO)); 1741 1742 ASSERT(blksize && max_xfer_len && minor_hdl && 1743 objset_hdl && zil_hdl && rl_hdl && bonus_hdl); 1744 1745 *blksize = zv->zv_volblocksize; 1746 *max_xfer_len = (uint64_t)zvol_maxphys; 1747 *minor_hdl = zv; 1748 *objset_hdl = zv->zv_objset; 1749 *zil_hdl = zv->zv_zilog; 1750 *rl_hdl = &zv->zv_znode; 1751 *bonus_hdl = zv->zv_dbuf; 1752 return (0); 1753} 1754 1755/* 1756 * Return the current volume size to an external caller. 1757 * The size can change while the volume is open. 1758 */ 1759uint64_t 1760zvol_get_volume_size(void *minor_hdl) 1761{ 1762 zvol_state_t *zv = minor_hdl; 1763 1764 return (zv->zv_volsize); 1765} 1766 1767/* 1768 * Return the current WCE setting to an external caller. 1769 * The WCE setting can change while the volume is open. 1770 */ 1771int 1772zvol_get_volume_wce(void *minor_hdl) 1773{ 1774 zvol_state_t *zv = minor_hdl; 1775 1776 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); 1777} 1778 1779/* 1780 * Entry point for external callers to zvol_log_write 1781 */ 1782void 1783zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, 1784 boolean_t sync) 1785{ 1786 zvol_state_t *zv = minor_hdl; 1787 1788 zvol_log_write(zv, tx, off, resid, sync); 1789} 1790/* 1791 * END entry points to allow external callers access to the volume. 1792 */ 1793#endif /* sun */ 1794 1795/* 1796 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. 1797 */ 1798static void 1799zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, 1800 boolean_t sync) 1801{ 1802 itx_t *itx; 1803 lr_truncate_t *lr; 1804 zilog_t *zilog = zv->zv_zilog; 1805 1806 if (zil_replaying(zilog, tx)) 1807 return; 1808 1809 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 1810 lr = (lr_truncate_t *)&itx->itx_lr; 1811 lr->lr_foid = ZVOL_OBJ; 1812 lr->lr_offset = off; 1813 lr->lr_length = len; 1814 1815 itx->itx_sync = sync; 1816 zil_itx_assign(zilog, itx, tx); 1817} 1818 1819#ifdef sun 1820/* 1821 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). 1822 * Also a dirtbag dkio ioctl for unmap/free-block functionality. 1823 */ 1824/*ARGSUSED*/ 1825int 1826zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) 1827{ 1828 zvol_state_t *zv; 1829 struct dk_callback *dkc; 1830 int error = 0; 1831 rl_t *rl; 1832 1833 mutex_enter(&spa_namespace_lock); 1834 1835 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); 1836 1837 if (zv == NULL) { 1838 mutex_exit(&spa_namespace_lock); 1839 return (SET_ERROR(ENXIO)); 1840 } 1841 ASSERT(zv->zv_total_opens > 0); 1842 1843 switch (cmd) { 1844 1845 case DKIOCINFO: 1846 { 1847 struct dk_cinfo dki; 1848 1849 bzero(&dki, sizeof (dki)); 1850 (void) strcpy(dki.dki_cname, "zvol"); 1851 (void) strcpy(dki.dki_dname, "zvol"); 1852 dki.dki_ctype = DKC_UNKNOWN; 1853 dki.dki_unit = getminor(dev); 1854 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); 1855 mutex_exit(&spa_namespace_lock); 1856 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) 1857 error = SET_ERROR(EFAULT); 1858 return (error); 1859 } 1860 1861 case DKIOCGMEDIAINFO: 1862 { 1863 struct dk_minfo dkm; 1864 1865 bzero(&dkm, sizeof (dkm)); 1866 dkm.dki_lbsize = 1U << zv->zv_min_bs; 1867 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1868 dkm.dki_media_type = DK_UNKNOWN; 1869 mutex_exit(&spa_namespace_lock); 1870 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) 1871 error = SET_ERROR(EFAULT); 1872 return (error); 1873 } 1874 1875 case DKIOCGMEDIAINFOEXT: 1876 { 1877 struct dk_minfo_ext dkmext; 1878 1879 bzero(&dkmext, sizeof (dkmext)); 1880 dkmext.dki_lbsize = 1U << zv->zv_min_bs; 1881 dkmext.dki_pbsize = zv->zv_volblocksize; 1882 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; 1883 dkmext.dki_media_type = DK_UNKNOWN; 1884 mutex_exit(&spa_namespace_lock); 1885 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag)) 1886 error = SET_ERROR(EFAULT); 1887 return (error); 1888 } 1889 1890 case DKIOCGETEFI: 1891 { 1892 uint64_t vs = zv->zv_volsize; 1893 uint8_t bs = zv->zv_min_bs; 1894 1895 mutex_exit(&spa_namespace_lock); 1896 error = zvol_getefi((void *)arg, flag, vs, bs); 1897 return (error); 1898 } 1899 1900 case DKIOCFLUSHWRITECACHE: 1901 dkc = (struct dk_callback *)arg; 1902 mutex_exit(&spa_namespace_lock); 1903 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1904 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { 1905 (*dkc->dkc_callback)(dkc->dkc_cookie, error); 1906 error = 0; 1907 } 1908 return (error); 1909 1910 case DKIOCGETWCE: 1911 { 1912 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; 1913 if (ddi_copyout(&wce, (void *)arg, sizeof (int), 1914 flag)) 1915 error = SET_ERROR(EFAULT); 1916 break; 1917 } 1918 case DKIOCSETWCE: 1919 { 1920 int wce; 1921 if (ddi_copyin((void *)arg, &wce, sizeof (int), 1922 flag)) { 1923 error = SET_ERROR(EFAULT); 1924 break; 1925 } 1926 if (wce) { 1927 zv->zv_flags |= ZVOL_WCE; 1928 mutex_exit(&spa_namespace_lock); 1929 } else { 1930 zv->zv_flags &= ~ZVOL_WCE; 1931 mutex_exit(&spa_namespace_lock); 1932 zil_commit(zv->zv_zilog, ZVOL_OBJ); 1933 } 1934 return (0); 1935 } 1936 1937 case DKIOCGGEOM: 1938 case DKIOCGVTOC: 1939 /* 1940 * commands using these (like prtvtoc) expect ENOTSUP 1941 * since we're emulating an EFI label 1942 */ 1943 error = SET_ERROR(ENOTSUP); 1944 break; 1945 1946 case DKIOCDUMPINIT: 1947 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1948 RL_WRITER); 1949 error = zvol_dumpify(zv); 1950 zfs_range_unlock(rl); 1951 break; 1952 1953 case DKIOCDUMPFINI: 1954 if (!(zv->zv_flags & ZVOL_DUMPIFIED)) 1955 break; 1956 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, 1957 RL_WRITER); 1958 error = zvol_dump_fini(zv); 1959 zfs_range_unlock(rl); 1960 break; 1961 1962 case DKIOCFREE: 1963 { 1964 dkioc_free_t df; 1965 dmu_tx_t *tx; 1966 1967 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) { 1968 error = SET_ERROR(EFAULT); 1969 break; 1970 } 1971 1972 /* 1973 * Apply Postel's Law to length-checking. If they overshoot, 1974 * just blank out until the end, if there's a need to blank 1975 * out anything. 1976 */ 1977 if (df.df_start >= zv->zv_volsize) 1978 break; /* No need to do anything... */ 1979 if (df.df_start + df.df_length > zv->zv_volsize) 1980 df.df_length = DMU_OBJECT_END; 1981 1982 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length, 1983 RL_WRITER); 1984 tx = dmu_tx_create(zv->zv_objset); 1985 error = dmu_tx_assign(tx, TXG_WAIT); 1986 if (error != 0) { 1987 dmu_tx_abort(tx); 1988 } else { 1989 zvol_log_truncate(zv, tx, df.df_start, 1990 df.df_length, B_TRUE); 1991 dmu_tx_commit(tx); 1992 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 1993 df.df_start, df.df_length); 1994 } 1995 1996 zfs_range_unlock(rl); 1997 1998 if (error == 0) { 1999 /* 2000 * If the write-cache is disabled or 'sync' property 2001 * is set to 'always' then treat this as a synchronous 2002 * operation (i.e. commit to zil). 2003 */ 2004 if (!(zv->zv_flags & ZVOL_WCE) || 2005 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) 2006 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2007 2008 /* 2009 * If the caller really wants synchronous writes, and 2010 * can't wait for them, don't return until the write 2011 * is done. 2012 */ 2013 if (df.df_flags & DF_WAIT_SYNC) { 2014 txg_wait_synced( 2015 dmu_objset_pool(zv->zv_objset), 0); 2016 } 2017 } 2018 break; 2019 } 2020 2021 default: 2022 error = SET_ERROR(ENOTTY); 2023 break; 2024 2025 } 2026 mutex_exit(&spa_namespace_lock); 2027 return (error); 2028} 2029#endif /* sun */ 2030 2031int 2032zvol_busy(void) 2033{ 2034 return (zvol_minors != 0); 2035} 2036 2037void 2038zvol_init(void) 2039{ 2040 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 2041 1) == 0); 2042 ZFS_LOG(1, "ZVOL Initialized."); 2043} 2044 2045void 2046zvol_fini(void) 2047{ 2048 ddi_soft_state_fini(&zfsdev_state); 2049 ZFS_LOG(1, "ZVOL Deinitialized."); 2050} 2051 2052#ifdef sun 2053/*ARGSUSED*/ 2054static int 2055zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx) 2056{ 2057 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 2058 2059 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 2060 return (1); 2061 return (0); 2062} 2063 2064/*ARGSUSED*/ 2065static void 2066zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx) 2067{ 2068 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 2069 2070 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx); 2071} 2072 2073static int 2074zvol_dump_init(zvol_state_t *zv, boolean_t resize) 2075{ 2076 dmu_tx_t *tx; 2077 int error; 2078 objset_t *os = zv->zv_objset; 2079 spa_t *spa = dmu_objset_spa(os); 2080 vdev_t *vd = spa->spa_root_vdev; 2081 nvlist_t *nv = NULL; 2082 uint64_t version = spa_version(spa); 2083 enum zio_checksum checksum; 2084 2085 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2086 ASSERT(vd->vdev_ops == &vdev_root_ops); 2087 2088 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, 2089 DMU_OBJECT_END); 2090 /* wait for dmu_free_long_range to actually free the blocks */ 2091 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2092 2093 /* 2094 * If the pool on which the dump device is being initialized has more 2095 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is 2096 * enabled. If so, bump that feature's counter to indicate that the 2097 * feature is active. We also check the vdev type to handle the 2098 * following case: 2099 * # zpool create test raidz disk1 disk2 disk3 2100 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev), 2101 * the raidz vdev itself has 3 children. 2102 */ 2103 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) { 2104 if (!spa_feature_is_enabled(spa, 2105 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) 2106 return (SET_ERROR(ENOTSUP)); 2107 (void) dsl_sync_task(spa_name(spa), 2108 zfs_mvdev_dump_feature_check, 2109 zfs_mvdev_dump_activate_feature_sync, NULL, 2); 2110 } 2111 2112 tx = dmu_tx_create(os); 2113 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2114 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2115 error = dmu_tx_assign(tx, TXG_WAIT); 2116 if (error) { 2117 dmu_tx_abort(tx); 2118 return (error); 2119 } 2120 2121 /* 2122 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum 2123 * function. Otherwise, use the old default -- OFF. 2124 */ 2125 checksum = spa_feature_is_active(spa, 2126 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY : 2127 ZIO_CHECKSUM_OFF; 2128 2129 /* 2130 * If we are resizing the dump device then we only need to 2131 * update the refreservation to match the newly updated 2132 * zvolsize. Otherwise, we save off the original state of the 2133 * zvol so that we can restore them if the zvol is ever undumpified. 2134 */ 2135 if (resize) { 2136 error = zap_update(os, ZVOL_ZAP_OBJ, 2137 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2138 &zv->zv_volsize, tx); 2139 } else { 2140 uint64_t checksum, compress, refresrv, vbs, dedup; 2141 2142 error = dsl_prop_get_integer(zv->zv_name, 2143 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); 2144 error = error ? error : dsl_prop_get_integer(zv->zv_name, 2145 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL); 2146 error = error ? error : dsl_prop_get_integer(zv->zv_name, 2147 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL); 2148 error = error ? error : dsl_prop_get_integer(zv->zv_name, 2149 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL); 2150 if (version >= SPA_VERSION_DEDUP) { 2151 error = error ? error : 2152 dsl_prop_get_integer(zv->zv_name, 2153 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); 2154 } 2155 2156 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2157 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, 2158 &compress, tx); 2159 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2160 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx); 2161 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2162 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, 2163 &refresrv, tx); 2164 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2165 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, 2166 &vbs, tx); 2167 error = error ? error : dmu_object_set_blocksize( 2168 os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx); 2169 if (version >= SPA_VERSION_DEDUP) { 2170 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, 2171 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, 2172 &dedup, tx); 2173 } 2174 if (error == 0) 2175 zv->zv_volblocksize = SPA_MAXBLOCKSIZE; 2176 } 2177 dmu_tx_commit(tx); 2178 2179 /* 2180 * We only need update the zvol's property if we are initializing 2181 * the dump area for the first time. 2182 */ 2183 if (!resize) { 2184 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2185 VERIFY(nvlist_add_uint64(nv, 2186 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); 2187 VERIFY(nvlist_add_uint64(nv, 2188 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 2189 ZIO_COMPRESS_OFF) == 0); 2190 VERIFY(nvlist_add_uint64(nv, 2191 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 2192 checksum) == 0); 2193 if (version >= SPA_VERSION_DEDUP) { 2194 VERIFY(nvlist_add_uint64(nv, 2195 zfs_prop_to_name(ZFS_PROP_DEDUP), 2196 ZIO_CHECKSUM_OFF) == 0); 2197 } 2198 2199 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2200 nv, NULL); 2201 nvlist_free(nv); 2202 2203 if (error) 2204 return (error); 2205 } 2206 2207 /* Allocate the space for the dump */ 2208 error = zvol_prealloc(zv); 2209 return (error); 2210} 2211 2212static int 2213zvol_dumpify(zvol_state_t *zv) 2214{ 2215 int error = 0; 2216 uint64_t dumpsize = 0; 2217 dmu_tx_t *tx; 2218 objset_t *os = zv->zv_objset; 2219 2220 if (zv->zv_flags & ZVOL_RDONLY) 2221 return (SET_ERROR(EROFS)); 2222 2223 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 2224 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { 2225 boolean_t resize = (dumpsize > 0); 2226 2227 if ((error = zvol_dump_init(zv, resize)) != 0) { 2228 (void) zvol_dump_fini(zv); 2229 return (error); 2230 } 2231 } 2232 2233 /* 2234 * Build up our lba mapping. 2235 */ 2236 error = zvol_get_lbas(zv); 2237 if (error) { 2238 (void) zvol_dump_fini(zv); 2239 return (error); 2240 } 2241 2242 tx = dmu_tx_create(os); 2243 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2244 error = dmu_tx_assign(tx, TXG_WAIT); 2245 if (error) { 2246 dmu_tx_abort(tx); 2247 (void) zvol_dump_fini(zv); 2248 return (error); 2249 } 2250 2251 zv->zv_flags |= ZVOL_DUMPIFIED; 2252 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, 2253 &zv->zv_volsize, tx); 2254 dmu_tx_commit(tx); 2255 2256 if (error) { 2257 (void) zvol_dump_fini(zv); 2258 return (error); 2259 } 2260 2261 txg_wait_synced(dmu_objset_pool(os), 0); 2262 return (0); 2263} 2264 2265static int 2266zvol_dump_fini(zvol_state_t *zv) 2267{ 2268 dmu_tx_t *tx; 2269 objset_t *os = zv->zv_objset; 2270 nvlist_t *nv; 2271 int error = 0; 2272 uint64_t checksum, compress, refresrv, vbs, dedup; 2273 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); 2274 2275 /* 2276 * Attempt to restore the zvol back to its pre-dumpified state. 2277 * This is a best-effort attempt as it's possible that not all 2278 * of these properties were initialized during the dumpify process 2279 * (i.e. error during zvol_dump_init). 2280 */ 2281 2282 tx = dmu_tx_create(os); 2283 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 2284 error = dmu_tx_assign(tx, TXG_WAIT); 2285 if (error) { 2286 dmu_tx_abort(tx); 2287 return (error); 2288 } 2289 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); 2290 dmu_tx_commit(tx); 2291 2292 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2293 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); 2294 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2295 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); 2296 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2297 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); 2298 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2299 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); 2300 2301 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2302 (void) nvlist_add_uint64(nv, 2303 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); 2304 (void) nvlist_add_uint64(nv, 2305 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); 2306 (void) nvlist_add_uint64(nv, 2307 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); 2308 if (version >= SPA_VERSION_DEDUP && 2309 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, 2310 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { 2311 (void) nvlist_add_uint64(nv, 2312 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); 2313 } 2314 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, 2315 nv, NULL); 2316 nvlist_free(nv); 2317 2318 zvol_free_extents(zv); 2319 zv->zv_flags &= ~ZVOL_DUMPIFIED; 2320 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); 2321 /* wait for dmu_free_long_range to actually free the blocks */ 2322 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 2323 tx = dmu_tx_create(os); 2324 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 2325 error = dmu_tx_assign(tx, TXG_WAIT); 2326 if (error) { 2327 dmu_tx_abort(tx); 2328 return (error); 2329 } 2330 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) 2331 zv->zv_volblocksize = vbs; 2332 dmu_tx_commit(tx); 2333 2334 return (0); 2335} 2336#endif /* sun */ 2337 2338static void 2339zvol_geom_run(zvol_state_t *zv) 2340{ 2341 struct g_provider *pp; 2342 2343 pp = zv->zv_provider; 2344 g_error_provider(pp, 0); 2345 2346 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0, 2347 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER)); 2348} 2349 2350static void 2351zvol_geom_destroy(zvol_state_t *zv) 2352{ 2353 struct g_provider *pp; 2354 2355 g_topology_assert(); 2356 2357 mtx_lock(&zv->zv_queue_mtx); 2358 zv->zv_state = 1; 2359 wakeup_one(&zv->zv_queue); 2360 while (zv->zv_state != 2) 2361 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0); 2362 mtx_destroy(&zv->zv_queue_mtx); 2363 2364 pp = zv->zv_provider; 2365 zv->zv_provider = NULL; 2366 pp->private = NULL; 2367 g_wither_geom(pp->geom, ENXIO); 2368} 2369 2370static int 2371zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) 2372{ 2373 int count, error, flags; 2374 2375 g_topology_assert(); 2376 2377 /* 2378 * To make it easier we expect either open or close, but not both 2379 * at the same time. 2380 */ 2381 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) || 2382 (acr <= 0 && acw <= 0 && ace <= 0), 2383 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", 2384 pp->name, acr, acw, ace)); 2385 2386 if (pp->private == NULL) { 2387 if (acr <= 0 && acw <= 0 && ace <= 0) 2388 return (0); 2389 return (pp->error); 2390 } 2391 2392 /* 2393 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0, 2394 * because GEOM already handles that and handles it a bit differently. 2395 * GEOM allows for multiple read/exclusive consumers and ZFS allows 2396 * only one exclusive consumer, no matter if it is reader or writer. 2397 * I like better the way GEOM works so I'll leave it for GEOM to 2398 * decide what to do. 2399 */ 2400 2401 count = acr + acw + ace; 2402 if (count == 0) 2403 return (0); 2404 2405 flags = 0; 2406 if (acr != 0 || ace != 0) 2407 flags |= FREAD; 2408 if (acw != 0) 2409 flags |= FWRITE; 2410 2411 g_topology_unlock(); 2412 if (count > 0) 2413 error = zvol_open(pp, flags, count); 2414 else 2415 error = zvol_close(pp, flags, -count); 2416 g_topology_lock(); 2417 return (error); 2418} 2419 2420static void 2421zvol_geom_start(struct bio *bp) 2422{ 2423 zvol_state_t *zv; 2424 boolean_t first; 2425 2426 zv = bp->bio_to->private; 2427 ASSERT(zv != NULL); 2428 switch (bp->bio_cmd) { 2429 case BIO_FLUSH: 2430 if (!THREAD_CAN_SLEEP()) 2431 goto enqueue; 2432 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2433 g_io_deliver(bp, 0); 2434 break; 2435 case BIO_READ: 2436 case BIO_WRITE: 2437 case BIO_DELETE: 2438 if (!THREAD_CAN_SLEEP()) 2439 goto enqueue; 2440 zvol_strategy(bp); 2441 break; 2442 case BIO_GETATTR: 2443 if (g_handleattr_int(bp, "GEOM::candelete", 1)) 2444 return; 2445 /* FALLTHROUGH */ 2446 default: 2447 g_io_deliver(bp, EOPNOTSUPP); 2448 break; 2449 } 2450 return; 2451 2452enqueue: 2453 mtx_lock(&zv->zv_queue_mtx); 2454 first = (bioq_first(&zv->zv_queue) == NULL); 2455 bioq_insert_tail(&zv->zv_queue, bp); 2456 mtx_unlock(&zv->zv_queue_mtx); 2457 if (first) 2458 wakeup_one(&zv->zv_queue); 2459} 2460 2461static void 2462zvol_geom_worker(void *arg) 2463{ 2464 zvol_state_t *zv; 2465 struct bio *bp; 2466 2467 thread_lock(curthread); 2468 sched_prio(curthread, PRIBIO); 2469 thread_unlock(curthread); 2470 2471 zv = arg; 2472 for (;;) { 2473 mtx_lock(&zv->zv_queue_mtx); 2474 bp = bioq_takefirst(&zv->zv_queue); 2475 if (bp == NULL) { 2476 if (zv->zv_state == 1) { 2477 zv->zv_state = 2; 2478 wakeup(&zv->zv_state); 2479 mtx_unlock(&zv->zv_queue_mtx); 2480 kthread_exit(); 2481 } 2482 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP, 2483 "zvol:io", 0); 2484 continue; 2485 } 2486 mtx_unlock(&zv->zv_queue_mtx); 2487 switch (bp->bio_cmd) { 2488 case BIO_FLUSH: 2489 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2490 g_io_deliver(bp, 0); 2491 break; 2492 case BIO_READ: 2493 case BIO_WRITE: 2494 zvol_strategy(bp); 2495 break; 2496 } 2497 } 2498} 2499 2500extern boolean_t dataset_name_hidden(const char *name); 2501 2502static int 2503zvol_create_snapshots(objset_t *os, const char *name) 2504{ 2505 uint64_t cookie, obj; 2506 char *sname; 2507 int error, len; 2508 2509 cookie = obj = 0; 2510 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2511 2512#if 0 2513 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL, 2514 DS_FIND_SNAPSHOTS); 2515#endif 2516 2517 for (;;) { 2518 len = snprintf(sname, MAXPATHLEN, "%s@", name); 2519 if (len >= MAXPATHLEN) { 2520 dmu_objset_rele(os, FTAG); 2521 error = ENAMETOOLONG; 2522 break; 2523 } 2524 2525 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 2526 error = dmu_snapshot_list_next(os, MAXPATHLEN - len, 2527 sname + len, &obj, &cookie, NULL); 2528 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 2529 if (error != 0) { 2530 if (error == ENOENT) 2531 error = 0; 2532 break; 2533 } 2534 2535 if ((error = zvol_create_minor(sname)) != 0) { 2536 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2537 sname, error); 2538 break; 2539 } 2540 } 2541 2542 kmem_free(sname, MAXPATHLEN); 2543 return (error); 2544} 2545 2546int 2547zvol_create_minors(const char *name) 2548{ 2549 uint64_t cookie; 2550 objset_t *os; 2551 char *osname, *p; 2552 int error, len; 2553 2554 if (dataset_name_hidden(name)) 2555 return (0); 2556 2557 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2558 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2559 name, error); 2560 return (error); 2561 } 2562 if (dmu_objset_type(os) == DMU_OST_ZVOL) { 2563 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG); 2564 dsl_pool_rele(dmu_objset_pool(os), FTAG); 2565 if ((error = zvol_create_minor(name)) == 0) 2566 error = zvol_create_snapshots(os, name); 2567 else { 2568 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", 2569 name, error); 2570 } 2571 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG); 2572 dsl_dataset_rele(os->os_dsl_dataset, FTAG); 2573 return (error); 2574 } 2575 if (dmu_objset_type(os) != DMU_OST_ZFS) { 2576 dmu_objset_rele(os, FTAG); 2577 return (0); 2578 } 2579 2580 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2581 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) { 2582 dmu_objset_rele(os, FTAG); 2583 kmem_free(osname, MAXPATHLEN); 2584 return (ENOENT); 2585 } 2586 p = osname + strlen(osname); 2587 len = MAXPATHLEN - (p - osname); 2588 2589#if 0 2590 /* Prefetch the datasets. */ 2591 cookie = 0; 2592 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) { 2593 if (!dataset_name_hidden(osname)) 2594 (void) dmu_objset_prefetch(osname, NULL); 2595 } 2596#endif 2597 2598 cookie = 0; 2599 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL, 2600 &cookie) == 0) { 2601 dmu_objset_rele(os, FTAG); 2602 (void)zvol_create_minors(osname); 2603 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { 2604 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", 2605 name, error); 2606 return (error); 2607 } 2608 } 2609 2610 dmu_objset_rele(os, FTAG); 2611 kmem_free(osname, MAXPATHLEN); 2612 return (0); 2613} 2614 2615static void 2616zvol_rename_minor(zvol_state_t *zv, const char *newname) 2617{ 2618 struct g_geom *gp; 2619 struct g_provider *pp; 2620 struct cdev *dev; 2621 2622 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2623 2624 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { 2625 g_topology_lock(); 2626 pp = zv->zv_provider; 2627 ASSERT(pp != NULL); 2628 gp = pp->geom; 2629 ASSERT(gp != NULL); 2630 2631 zv->zv_provider = NULL; 2632 g_wither_provider(pp, ENXIO); 2633 2634 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname); 2635 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; 2636 pp->sectorsize = DEV_BSIZE; 2637 pp->mediasize = zv->zv_volsize; 2638 pp->private = zv; 2639 zv->zv_provider = pp; 2640 g_error_provider(pp, 0); 2641 g_topology_unlock(); 2642 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { 2643 dev = zv->zv_dev; 2644 ASSERT(dev != NULL); 2645 zv->zv_dev = NULL; 2646 destroy_dev(dev); 2647 2648 if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, 2649 &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 2650 0640, "%s/%s", ZVOL_DRIVER, newname) == 0) { 2651 zv->zv_dev = dev; 2652 dev->si_iosize_max = MAXPHYS; 2653 dev->si_drv2 = zv; 2654 } 2655 } 2656 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name)); 2657} 2658 2659void 2660zvol_rename_minors(const char *oldname, const char *newname) 2661{ 2662 char name[MAXPATHLEN]; 2663 struct g_provider *pp; 2664 struct g_geom *gp; 2665 size_t oldnamelen, newnamelen; 2666 zvol_state_t *zv; 2667 char *namebuf; 2668 2669 oldnamelen = strlen(oldname); 2670 newnamelen = strlen(newname); 2671 2672 DROP_GIANT(); 2673 mutex_enter(&spa_namespace_lock); 2674 2675 LIST_FOREACH(zv, &all_zvols, zv_links) { 2676 if (strcmp(zv->zv_name, oldname) == 0) { 2677 zvol_rename_minor(zv, newname); 2678 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && 2679 (zv->zv_name[oldnamelen] == '/' || 2680 zv->zv_name[oldnamelen] == '@')) { 2681 snprintf(name, sizeof(name), "%s%c%s", newname, 2682 zv->zv_name[oldnamelen], 2683 zv->zv_name + oldnamelen + 1); 2684 zvol_rename_minor(zv, name); 2685 } 2686 } 2687 2688 mutex_exit(&spa_namespace_lock); 2689 PICKUP_GIANT(); 2690} 2691 2692static int 2693zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td) 2694{ 2695 zvol_state_t *zv; 2696 int err = 0; 2697 2698 mutex_enter(&spa_namespace_lock); 2699 zv = dev->si_drv2; 2700 if (zv == NULL) { 2701 mutex_exit(&spa_namespace_lock); 2702 return(ENXIO); /* zvol_create_minor() not done yet */ 2703 } 2704 2705 if (zv->zv_total_opens == 0) 2706 err = zvol_first_open(zv); 2707 if (err) { 2708 mutex_exit(&spa_namespace_lock); 2709 return (err); 2710 } 2711 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { 2712 err = SET_ERROR(EROFS); 2713 goto out; 2714 } 2715 if (zv->zv_flags & ZVOL_EXCL) { 2716 err = SET_ERROR(EBUSY); 2717 goto out; 2718 } 2719#ifdef FEXCL 2720 if (flags & FEXCL) { 2721 if (zv->zv_total_opens != 0) { 2722 err = SET_ERROR(EBUSY); 2723 goto out; 2724 } 2725 zv->zv_flags |= ZVOL_EXCL; 2726 } 2727#endif 2728 2729 zv->zv_total_opens++; 2730 mutex_exit(&spa_namespace_lock); 2731 return (err); 2732out: 2733 if (zv->zv_total_opens == 0) 2734 zvol_last_close(zv); 2735 mutex_exit(&spa_namespace_lock); 2736 return (err); 2737} 2738 2739static int 2740zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td) 2741{ 2742 zvol_state_t *zv; 2743 int err = 0; 2744 2745 mutex_enter(&spa_namespace_lock); 2746 zv = dev->si_drv2; 2747 if (zv == NULL) { 2748 mutex_exit(&spa_namespace_lock); 2749 return(ENXIO); 2750 } 2751 2752 if (zv->zv_flags & ZVOL_EXCL) { 2753 ASSERT(zv->zv_total_opens == 1); 2754 zv->zv_flags &= ~ZVOL_EXCL; 2755 } 2756 2757 /* 2758 * If the open count is zero, this is a spurious close. 2759 * That indicates a bug in the kernel / DDI framework. 2760 */ 2761 ASSERT(zv->zv_total_opens != 0); 2762 2763 /* 2764 * You may get multiple opens, but only one close. 2765 */ 2766 zv->zv_total_opens--; 2767 2768 if (zv->zv_total_opens == 0) 2769 zvol_last_close(zv); 2770 2771 mutex_exit(&spa_namespace_lock); 2772 return (0); 2773} 2774 2775static int 2776zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 2777{ 2778 zvol_state_t *zv; 2779 rl_t *rl; 2780 off_t offset, length, chunk; 2781 int i, error; 2782 u_int u; 2783 2784 zv = dev->si_drv2; 2785 2786 error = 0; 2787 KASSERT(zv->zv_total_opens > 0, 2788 ("Device with zero access count in zvol_d_ioctl")); 2789 2790 i = IOCPARM_LEN(cmd); 2791 switch (cmd) { 2792 case DIOCGSECTORSIZE: 2793 *(u_int *)data = DEV_BSIZE; 2794 break; 2795 case DIOCGMEDIASIZE: 2796 *(off_t *)data = zv->zv_volsize; 2797 break; 2798 case DIOCGFLUSH: 2799 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2800 break; 2801 case DIOCGDELETE: 2802 offset = ((off_t *)data)[0]; 2803 length = ((off_t *)data)[1]; 2804 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 || 2805 offset < 0 || offset >= zv->zv_volsize || 2806 length <= 0) { 2807 printf("%s: offset=%jd length=%jd\n", __func__, offset, 2808 length); 2809 error = EINVAL; 2810 break; 2811 } 2812 2813 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER); 2814 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 2815 error = dmu_tx_assign(tx, TXG_WAIT); 2816 if (error != 0) { 2817 dmu_tx_abort(tx); 2818 } else { 2819 zvol_log_truncate(zv, tx, offset, length, B_TRUE); 2820 dmu_tx_commit(tx); 2821 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 2822 offset, length); 2823 } 2824 zfs_range_unlock(rl); 2825 if (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) 2826 zil_commit(zv->zv_zilog, ZVOL_OBJ); 2827 break; 2828 case DIOCGSTRIPESIZE: 2829 *(off_t *)data = zv->zv_volblocksize; 2830 break; 2831 case DIOCGSTRIPEOFFSET: 2832 *(off_t *)data = 0; 2833 break; 2834 default: 2835 error = ENOIOCTL; 2836 } 2837 2838 return (error); 2839} 2840