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 https://opensource.org/licenses/CDDL-1.0. 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) 2008-2010 Lawrence Livermore National Security, LLC. 23 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER). 24 * Rewritten for Linux by Brian Behlendorf <behlendorf1@llnl.gov>. 25 * LLNL-CODE-403049. 26 * 27 * ZFS volume emulation driver. 28 * 29 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. 30 * Volumes are accessed through the symbolic links named: 31 * 32 * /dev/<pool_name>/<dataset_name> 33 * 34 * Volumes are persistent through reboot and module load. No user command 35 * needs to be run before opening and using a device. 36 * 37 * Copyright 2014 Nexenta Systems, Inc. All rights reserved. 38 * Copyright (c) 2016 Actifio, Inc. All rights reserved. 39 * Copyright (c) 2012, 2019 by Delphix. All rights reserved. 40 */ 41 42 /* 43 * Note on locking of zvol state structures. 44 * 45 * These structures are used to maintain internal state used to emulate block 46 * devices on top of zvols. In particular, management of device minor number 47 * operations - create, remove, rename, and set_snapdev - involves access to 48 * these structures. The zvol_state_lock is primarily used to protect the 49 * zvol_state_list. The zv->zv_state_lock is used to protect the contents 50 * of the zvol_state_t structures, as well as to make sure that when the 51 * time comes to remove the structure from the list, it is not in use, and 52 * therefore, it can be taken off zvol_state_list and freed. 53 * 54 * The zv_suspend_lock was introduced to allow for suspending I/O to a zvol, 55 * e.g. for the duration of receive and rollback operations. This lock can be 56 * held for significant periods of time. Given that it is undesirable to hold 57 * mutexes for long periods of time, the following lock ordering applies: 58 * - take zvol_state_lock if necessary, to protect zvol_state_list 59 * - take zv_suspend_lock if necessary, by the code path in question 60 * - take zv_state_lock to protect zvol_state_t 61 * 62 * The minor operations are issued to spa->spa_zvol_taskq queues, that are 63 * single-threaded (to preserve order of minor operations), and are executed 64 * through the zvol_task_cb that dispatches the specific operations. Therefore, 65 * these operations are serialized per pool. Consequently, we can be certain 66 * that for a given zvol, there is only one operation at a time in progress. 67 * That is why one can be sure that first, zvol_state_t for a given zvol is 68 * allocated and placed on zvol_state_list, and then other minor operations 69 * for this zvol are going to proceed in the order of issue. 70 * 71 */ 72 73 #include <sys/dataset_kstats.h> 74 #include <sys/dbuf.h> 75 #include <sys/dmu_traverse.h> 76 #include <sys/dsl_dataset.h> 77 #include <sys/dsl_prop.h> 78 #include <sys/dsl_dir.h> 79 #include <sys/zap.h> 80 #include <sys/zfeature.h> 81 #include <sys/zil_impl.h> 82 #include <sys/dmu_tx.h> 83 #include <sys/zio.h> 84 #include <sys/zfs_rlock.h> 85 #include <sys/spa_impl.h> 86 #include <sys/zvol.h> 87 #include <sys/zvol_impl.h> 88 89 unsigned int zvol_inhibit_dev = 0; 90 unsigned int zvol_volmode = ZFS_VOLMODE_GEOM; 91 92 struct hlist_head *zvol_htable; 93 static list_t zvol_state_list; 94 krwlock_t zvol_state_lock; 95 96 typedef enum { 97 ZVOL_ASYNC_REMOVE_MINORS, 98 ZVOL_ASYNC_RENAME_MINORS, 99 ZVOL_ASYNC_SET_SNAPDEV, 100 ZVOL_ASYNC_SET_VOLMODE, 101 ZVOL_ASYNC_MAX 102 } zvol_async_op_t; 103 104 typedef struct { 105 zvol_async_op_t op; 106 char name1[MAXNAMELEN]; 107 char name2[MAXNAMELEN]; 108 uint64_t value; 109 } zvol_task_t; 110 111 uint64_t 112 zvol_name_hash(const char *name) 113 { 114 int i; 115 uint64_t crc = -1ULL; 116 const uint8_t *p = (const uint8_t *)name; 117 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); 118 for (i = 0; i < MAXNAMELEN - 1 && *p; i++, p++) { 119 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (*p)) & 0xFF]; 120 } 121 return (crc); 122 } 123 124 /* 125 * Find a zvol_state_t given the name and hash generated by zvol_name_hash. 126 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise, 127 * return (NULL) without the taking locks. The zv_suspend_lock is always taken 128 * before zv_state_lock. The mode argument indicates the mode (including none) 129 * for zv_suspend_lock to be taken. 130 */ 131 zvol_state_t * 132 zvol_find_by_name_hash(const char *name, uint64_t hash, int mode) 133 { 134 zvol_state_t *zv; 135 struct hlist_node *p = NULL; 136 137 rw_enter(&zvol_state_lock, RW_READER); 138 hlist_for_each(p, ZVOL_HT_HEAD(hash)) { 139 zv = hlist_entry(p, zvol_state_t, zv_hlink); 140 mutex_enter(&zv->zv_state_lock); 141 if (zv->zv_hash == hash && 142 strncmp(zv->zv_name, name, MAXNAMELEN) == 0) { 143 /* 144 * this is the right zvol, take the locks in the 145 * right order 146 */ 147 if (mode != RW_NONE && 148 !rw_tryenter(&zv->zv_suspend_lock, mode)) { 149 mutex_exit(&zv->zv_state_lock); 150 rw_enter(&zv->zv_suspend_lock, mode); 151 mutex_enter(&zv->zv_state_lock); 152 /* 153 * zvol cannot be renamed as we continue 154 * to hold zvol_state_lock 155 */ 156 ASSERT(zv->zv_hash == hash && 157 strncmp(zv->zv_name, name, MAXNAMELEN) 158 == 0); 159 } 160 rw_exit(&zvol_state_lock); 161 return (zv); 162 } 163 mutex_exit(&zv->zv_state_lock); 164 } 165 rw_exit(&zvol_state_lock); 166 167 return (NULL); 168 } 169 170 /* 171 * Find a zvol_state_t given the name. 172 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise, 173 * return (NULL) without the taking locks. The zv_suspend_lock is always taken 174 * before zv_state_lock. The mode argument indicates the mode (including none) 175 * for zv_suspend_lock to be taken. 176 */ 177 static zvol_state_t * 178 zvol_find_by_name(const char *name, int mode) 179 { 180 return (zvol_find_by_name_hash(name, zvol_name_hash(name), mode)); 181 } 182 183 /* 184 * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation. 185 */ 186 void 187 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 188 { 189 zfs_creat_t *zct = arg; 190 nvlist_t *nvprops = zct->zct_props; 191 int error; 192 uint64_t volblocksize, volsize; 193 194 VERIFY(nvlist_lookup_uint64(nvprops, 195 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 196 if (nvlist_lookup_uint64(nvprops, 197 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 198 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 199 200 /* 201 * These properties must be removed from the list so the generic 202 * property setting step won't apply to them. 203 */ 204 VERIFY(nvlist_remove_all(nvprops, 205 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 206 (void) nvlist_remove_all(nvprops, 207 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 208 209 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 210 DMU_OT_NONE, 0, tx); 211 ASSERT(error == 0); 212 213 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 214 DMU_OT_NONE, 0, tx); 215 ASSERT(error == 0); 216 217 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 218 ASSERT(error == 0); 219 } 220 221 /* 222 * ZFS_IOC_OBJSET_STATS entry point. 223 */ 224 int 225 zvol_get_stats(objset_t *os, nvlist_t *nv) 226 { 227 int error; 228 dmu_object_info_t *doi; 229 uint64_t val; 230 231 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 232 if (error) 233 return (SET_ERROR(error)); 234 235 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 236 doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP); 237 error = dmu_object_info(os, ZVOL_OBJ, doi); 238 239 if (error == 0) { 240 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 241 doi->doi_data_block_size); 242 } 243 244 kmem_free(doi, sizeof (dmu_object_info_t)); 245 246 return (SET_ERROR(error)); 247 } 248 249 /* 250 * Sanity check volume size. 251 */ 252 int 253 zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 254 { 255 if (volsize == 0) 256 return (SET_ERROR(EINVAL)); 257 258 if (volsize % blocksize != 0) 259 return (SET_ERROR(EINVAL)); 260 261 #ifdef _ILP32 262 if (volsize - 1 > SPEC_MAXOFFSET_T) 263 return (SET_ERROR(EOVERFLOW)); 264 #endif 265 return (0); 266 } 267 268 /* 269 * Ensure the zap is flushed then inform the VFS of the capacity change. 270 */ 271 static int 272 zvol_update_volsize(uint64_t volsize, objset_t *os) 273 { 274 dmu_tx_t *tx; 275 int error; 276 uint64_t txg; 277 278 tx = dmu_tx_create(os); 279 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 280 dmu_tx_mark_netfree(tx); 281 error = dmu_tx_assign(tx, TXG_WAIT); 282 if (error) { 283 dmu_tx_abort(tx); 284 return (SET_ERROR(error)); 285 } 286 txg = dmu_tx_get_txg(tx); 287 288 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 289 &volsize, tx); 290 dmu_tx_commit(tx); 291 292 txg_wait_synced(dmu_objset_pool(os), txg); 293 294 if (error == 0) 295 error = dmu_free_long_range(os, 296 ZVOL_OBJ, volsize, DMU_OBJECT_END); 297 298 return (error); 299 } 300 301 /* 302 * Set ZFS_PROP_VOLSIZE set entry point. Note that modifying the volume 303 * size will result in a udev "change" event being generated. 304 */ 305 int 306 zvol_set_volsize(const char *name, uint64_t volsize) 307 { 308 objset_t *os = NULL; 309 uint64_t readonly; 310 int error; 311 boolean_t owned = B_FALSE; 312 313 error = dsl_prop_get_integer(name, 314 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL); 315 if (error != 0) 316 return (SET_ERROR(error)); 317 if (readonly) 318 return (SET_ERROR(EROFS)); 319 320 zvol_state_t *zv = zvol_find_by_name(name, RW_READER); 321 322 ASSERT(zv == NULL || (MUTEX_HELD(&zv->zv_state_lock) && 323 RW_READ_HELD(&zv->zv_suspend_lock))); 324 325 if (zv == NULL || zv->zv_objset == NULL) { 326 if (zv != NULL) 327 rw_exit(&zv->zv_suspend_lock); 328 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, B_TRUE, 329 FTAG, &os)) != 0) { 330 if (zv != NULL) 331 mutex_exit(&zv->zv_state_lock); 332 return (SET_ERROR(error)); 333 } 334 owned = B_TRUE; 335 if (zv != NULL) 336 zv->zv_objset = os; 337 } else { 338 os = zv->zv_objset; 339 } 340 341 dmu_object_info_t *doi = kmem_alloc(sizeof (*doi), KM_SLEEP); 342 343 if ((error = dmu_object_info(os, ZVOL_OBJ, doi)) || 344 (error = zvol_check_volsize(volsize, doi->doi_data_block_size))) 345 goto out; 346 347 error = zvol_update_volsize(volsize, os); 348 if (error == 0 && zv != NULL) { 349 zv->zv_volsize = volsize; 350 zv->zv_changed = 1; 351 } 352 out: 353 kmem_free(doi, sizeof (dmu_object_info_t)); 354 355 if (owned) { 356 dmu_objset_disown(os, B_TRUE, FTAG); 357 if (zv != NULL) 358 zv->zv_objset = NULL; 359 } else { 360 rw_exit(&zv->zv_suspend_lock); 361 } 362 363 if (zv != NULL) 364 mutex_exit(&zv->zv_state_lock); 365 366 if (error == 0 && zv != NULL) 367 zvol_os_update_volsize(zv, volsize); 368 369 return (SET_ERROR(error)); 370 } 371 372 /* 373 * Sanity check volume block size. 374 */ 375 int 376 zvol_check_volblocksize(const char *name, uint64_t volblocksize) 377 { 378 /* Record sizes above 128k need the feature to be enabled */ 379 if (volblocksize > SPA_OLD_MAXBLOCKSIZE) { 380 spa_t *spa; 381 int error; 382 383 if ((error = spa_open(name, &spa, FTAG)) != 0) 384 return (error); 385 386 if (!spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 387 spa_close(spa, FTAG); 388 return (SET_ERROR(ENOTSUP)); 389 } 390 391 /* 392 * We don't allow setting the property above 1MB, 393 * unless the tunable has been changed. 394 */ 395 if (volblocksize > zfs_max_recordsize) 396 return (SET_ERROR(EDOM)); 397 398 spa_close(spa, FTAG); 399 } 400 401 if (volblocksize < SPA_MINBLOCKSIZE || 402 volblocksize > SPA_MAXBLOCKSIZE || 403 !ISP2(volblocksize)) 404 return (SET_ERROR(EDOM)); 405 406 return (0); 407 } 408 409 /* 410 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we 411 * implement DKIOCFREE/free-long-range. 412 */ 413 static int 414 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap) 415 { 416 zvol_state_t *zv = arg1; 417 lr_truncate_t *lr = arg2; 418 uint64_t offset, length; 419 420 if (byteswap) 421 byteswap_uint64_array(lr, sizeof (*lr)); 422 423 offset = lr->lr_offset; 424 length = lr->lr_length; 425 426 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 427 dmu_tx_mark_netfree(tx); 428 int error = dmu_tx_assign(tx, TXG_WAIT); 429 if (error != 0) { 430 dmu_tx_abort(tx); 431 } else { 432 (void) zil_replaying(zv->zv_zilog, tx); 433 dmu_tx_commit(tx); 434 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, 435 length); 436 } 437 438 return (error); 439 } 440 441 /* 442 * Replay a TX_WRITE ZIL transaction that didn't get committed 443 * after a system failure 444 */ 445 static int 446 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap) 447 { 448 zvol_state_t *zv = arg1; 449 lr_write_t *lr = arg2; 450 objset_t *os = zv->zv_objset; 451 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 452 uint64_t offset, length; 453 dmu_tx_t *tx; 454 int error; 455 456 if (byteswap) 457 byteswap_uint64_array(lr, sizeof (*lr)); 458 459 offset = lr->lr_offset; 460 length = lr->lr_length; 461 462 /* If it's a dmu_sync() block, write the whole block */ 463 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 464 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 465 if (length < blocksize) { 466 offset -= offset % blocksize; 467 length = blocksize; 468 } 469 } 470 471 tx = dmu_tx_create(os); 472 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 473 error = dmu_tx_assign(tx, TXG_WAIT); 474 if (error) { 475 dmu_tx_abort(tx); 476 } else { 477 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 478 (void) zil_replaying(zv->zv_zilog, tx); 479 dmu_tx_commit(tx); 480 } 481 482 return (error); 483 } 484 485 /* 486 * Replay a TX_CLONE_RANGE ZIL transaction that didn't get committed 487 * after a system failure. 488 * 489 * TODO: For now we drop block cloning transations for ZVOLs as they are 490 * unsupported, but we still need to inform BRT about that as we 491 * claimed them during pool import. 492 * This situation can occur when we try to import a pool from a ZFS 493 * version supporting block cloning for ZVOLs into a system that 494 * has this ZFS version, that doesn't support block cloning for ZVOLs. 495 */ 496 static int 497 zvol_replay_clone_range(void *arg1, void *arg2, boolean_t byteswap) 498 { 499 char name[ZFS_MAX_DATASET_NAME_LEN]; 500 zvol_state_t *zv = arg1; 501 objset_t *os = zv->zv_objset; 502 lr_clone_range_t *lr = arg2; 503 blkptr_t *bp; 504 dmu_tx_t *tx; 505 spa_t *spa; 506 uint_t ii; 507 int error; 508 509 dmu_objset_name(os, name); 510 cmn_err(CE_WARN, "ZFS dropping block cloning transaction for %s.", 511 name); 512 513 if (byteswap) 514 byteswap_uint64_array(lr, sizeof (*lr)); 515 516 tx = dmu_tx_create(os); 517 error = dmu_tx_assign(tx, TXG_WAIT); 518 if (error) { 519 dmu_tx_abort(tx); 520 return (error); 521 } 522 523 spa = os->os_spa; 524 525 for (ii = 0; ii < lr->lr_nbps; ii++) { 526 bp = &lr->lr_bps[ii]; 527 528 if (!BP_IS_HOLE(bp)) { 529 zio_free(spa, dmu_tx_get_txg(tx), bp); 530 } 531 } 532 533 (void) zil_replaying(zv->zv_zilog, tx); 534 dmu_tx_commit(tx); 535 536 return (0); 537 } 538 539 static int 540 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap) 541 { 542 (void) arg1, (void) arg2, (void) byteswap; 543 return (SET_ERROR(ENOTSUP)); 544 } 545 546 /* 547 * Callback vectors for replaying records. 548 * Only TX_WRITE and TX_TRUNCATE are needed for zvol. 549 */ 550 zil_replay_func_t *const zvol_replay_vector[TX_MAX_TYPE] = { 551 zvol_replay_err, /* no such transaction type */ 552 zvol_replay_err, /* TX_CREATE */ 553 zvol_replay_err, /* TX_MKDIR */ 554 zvol_replay_err, /* TX_MKXATTR */ 555 zvol_replay_err, /* TX_SYMLINK */ 556 zvol_replay_err, /* TX_REMOVE */ 557 zvol_replay_err, /* TX_RMDIR */ 558 zvol_replay_err, /* TX_LINK */ 559 zvol_replay_err, /* TX_RENAME */ 560 zvol_replay_write, /* TX_WRITE */ 561 zvol_replay_truncate, /* TX_TRUNCATE */ 562 zvol_replay_err, /* TX_SETATTR */ 563 zvol_replay_err, /* TX_ACL */ 564 zvol_replay_err, /* TX_CREATE_ATTR */ 565 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 566 zvol_replay_err, /* TX_MKDIR_ACL */ 567 zvol_replay_err, /* TX_MKDIR_ATTR */ 568 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 569 zvol_replay_err, /* TX_WRITE2 */ 570 zvol_replay_err, /* TX_SETSAXATTR */ 571 zvol_replay_err, /* TX_RENAME_EXCHANGE */ 572 zvol_replay_err, /* TX_RENAME_WHITEOUT */ 573 zvol_replay_clone_range /* TX_CLONE_RANGE */ 574 }; 575 576 /* 577 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 578 * 579 * We store data in the log buffers if it's small enough. 580 * Otherwise we will later flush the data out via dmu_sync(). 581 */ 582 static const ssize_t zvol_immediate_write_sz = 32768; 583 584 void 585 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset, 586 uint64_t size, int sync) 587 { 588 uint32_t blocksize = zv->zv_volblocksize; 589 zilog_t *zilog = zv->zv_zilog; 590 itx_wr_state_t write_state; 591 uint64_t sz = size; 592 593 if (zil_replaying(zilog, tx)) 594 return; 595 596 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 597 write_state = WR_INDIRECT; 598 else if (!spa_has_slogs(zilog->zl_spa) && 599 size >= blocksize && blocksize > zvol_immediate_write_sz) 600 write_state = WR_INDIRECT; 601 else if (sync) 602 write_state = WR_COPIED; 603 else 604 write_state = WR_NEED_COPY; 605 606 while (size) { 607 itx_t *itx; 608 lr_write_t *lr; 609 itx_wr_state_t wr_state = write_state; 610 ssize_t len = size; 611 612 if (wr_state == WR_COPIED && size > zil_max_copied_data(zilog)) 613 wr_state = WR_NEED_COPY; 614 else if (wr_state == WR_INDIRECT) 615 len = MIN(blocksize - P2PHASE(offset, blocksize), size); 616 617 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 618 (wr_state == WR_COPIED ? len : 0)); 619 lr = (lr_write_t *)&itx->itx_lr; 620 if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn, 621 offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) { 622 zil_itx_destroy(itx); 623 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 624 lr = (lr_write_t *)&itx->itx_lr; 625 wr_state = WR_NEED_COPY; 626 } 627 628 itx->itx_wr_state = wr_state; 629 lr->lr_foid = ZVOL_OBJ; 630 lr->lr_offset = offset; 631 lr->lr_length = len; 632 lr->lr_blkoff = 0; 633 BP_ZERO(&lr->lr_blkptr); 634 635 itx->itx_private = zv; 636 itx->itx_sync = sync; 637 638 (void) zil_itx_assign(zilog, itx, tx); 639 640 offset += len; 641 size -= len; 642 } 643 644 if (write_state == WR_COPIED || write_state == WR_NEED_COPY) { 645 dsl_pool_wrlog_count(zilog->zl_dmu_pool, sz, tx->tx_txg); 646 } 647 } 648 649 /* 650 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. 651 */ 652 void 653 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, 654 boolean_t sync) 655 { 656 itx_t *itx; 657 lr_truncate_t *lr; 658 zilog_t *zilog = zv->zv_zilog; 659 660 if (zil_replaying(zilog, tx)) 661 return; 662 663 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 664 lr = (lr_truncate_t *)&itx->itx_lr; 665 lr->lr_foid = ZVOL_OBJ; 666 lr->lr_offset = off; 667 lr->lr_length = len; 668 669 itx->itx_sync = sync; 670 zil_itx_assign(zilog, itx, tx); 671 } 672 673 674 static void 675 zvol_get_done(zgd_t *zgd, int error) 676 { 677 (void) error; 678 if (zgd->zgd_db) 679 dmu_buf_rele(zgd->zgd_db, zgd); 680 681 zfs_rangelock_exit(zgd->zgd_lr); 682 683 kmem_free(zgd, sizeof (zgd_t)); 684 } 685 686 /* 687 * Get data to generate a TX_WRITE intent log record. 688 */ 689 int 690 zvol_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf, 691 struct lwb *lwb, zio_t *zio) 692 { 693 zvol_state_t *zv = arg; 694 uint64_t offset = lr->lr_offset; 695 uint64_t size = lr->lr_length; 696 dmu_buf_t *db; 697 zgd_t *zgd; 698 int error; 699 700 ASSERT3P(lwb, !=, NULL); 701 ASSERT3P(zio, !=, NULL); 702 ASSERT3U(size, !=, 0); 703 704 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 705 zgd->zgd_lwb = lwb; 706 707 /* 708 * Write records come in two flavors: immediate and indirect. 709 * For small writes it's cheaper to store the data with the 710 * log record (immediate); for large writes it's cheaper to 711 * sync the data and get a pointer to it (indirect) so that 712 * we don't have to write the data twice. 713 */ 714 if (buf != NULL) { /* immediate write */ 715 zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset, 716 size, RL_READER); 717 error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf, 718 DMU_READ_NO_PREFETCH); 719 } else { /* indirect write */ 720 /* 721 * Have to lock the whole block to ensure when it's written out 722 * and its checksum is being calculated that no one can change 723 * the data. Contrarily to zfs_get_data we need not re-check 724 * blocksize after we get the lock because it cannot be changed. 725 */ 726 size = zv->zv_volblocksize; 727 offset = P2ALIGN_TYPED(offset, size, uint64_t); 728 zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset, 729 size, RL_READER); 730 error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db, 731 DMU_READ_NO_PREFETCH); 732 if (error == 0) { 733 blkptr_t *bp = &lr->lr_blkptr; 734 735 zgd->zgd_db = db; 736 zgd->zgd_bp = bp; 737 738 ASSERT(db != NULL); 739 ASSERT(db->db_offset == offset); 740 ASSERT(db->db_size == size); 741 742 error = dmu_sync(zio, lr->lr_common.lrc_txg, 743 zvol_get_done, zgd); 744 745 if (error == 0) 746 return (0); 747 } 748 } 749 750 zvol_get_done(zgd, error); 751 752 return (SET_ERROR(error)); 753 } 754 755 /* 756 * The zvol_state_t's are inserted into zvol_state_list and zvol_htable. 757 */ 758 759 void 760 zvol_insert(zvol_state_t *zv) 761 { 762 ASSERT(RW_WRITE_HELD(&zvol_state_lock)); 763 list_insert_head(&zvol_state_list, zv); 764 hlist_add_head(&zv->zv_hlink, ZVOL_HT_HEAD(zv->zv_hash)); 765 } 766 767 /* 768 * Simply remove the zvol from to list of zvols. 769 */ 770 static void 771 zvol_remove(zvol_state_t *zv) 772 { 773 ASSERT(RW_WRITE_HELD(&zvol_state_lock)); 774 list_remove(&zvol_state_list, zv); 775 hlist_del(&zv->zv_hlink); 776 } 777 778 /* 779 * Setup zv after we just own the zv->objset 780 */ 781 static int 782 zvol_setup_zv(zvol_state_t *zv) 783 { 784 uint64_t volsize; 785 int error; 786 uint64_t ro; 787 objset_t *os = zv->zv_objset; 788 789 ASSERT(MUTEX_HELD(&zv->zv_state_lock)); 790 ASSERT(RW_LOCK_HELD(&zv->zv_suspend_lock)); 791 792 zv->zv_zilog = NULL; 793 zv->zv_flags &= ~ZVOL_WRITTEN_TO; 794 795 error = dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL); 796 if (error) 797 return (SET_ERROR(error)); 798 799 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 800 if (error) 801 return (SET_ERROR(error)); 802 803 error = dnode_hold(os, ZVOL_OBJ, zv, &zv->zv_dn); 804 if (error) 805 return (SET_ERROR(error)); 806 807 zvol_os_set_capacity(zv, volsize >> 9); 808 zv->zv_volsize = volsize; 809 810 if (ro || dmu_objset_is_snapshot(os) || 811 !spa_writeable(dmu_objset_spa(os))) { 812 zvol_os_set_disk_ro(zv, 1); 813 zv->zv_flags |= ZVOL_RDONLY; 814 } else { 815 zvol_os_set_disk_ro(zv, 0); 816 zv->zv_flags &= ~ZVOL_RDONLY; 817 } 818 return (0); 819 } 820 821 /* 822 * Shutdown every zv_objset related stuff except zv_objset itself. 823 * The is the reverse of zvol_setup_zv. 824 */ 825 static void 826 zvol_shutdown_zv(zvol_state_t *zv) 827 { 828 ASSERT(MUTEX_HELD(&zv->zv_state_lock) && 829 RW_LOCK_HELD(&zv->zv_suspend_lock)); 830 831 if (zv->zv_flags & ZVOL_WRITTEN_TO) { 832 ASSERT(zv->zv_zilog != NULL); 833 zil_close(zv->zv_zilog); 834 } 835 836 zv->zv_zilog = NULL; 837 838 dnode_rele(zv->zv_dn, zv); 839 zv->zv_dn = NULL; 840 841 /* 842 * Evict cached data. We must write out any dirty data before 843 * disowning the dataset. 844 */ 845 if (zv->zv_flags & ZVOL_WRITTEN_TO) 846 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 847 (void) dmu_objset_evict_dbufs(zv->zv_objset); 848 } 849 850 /* 851 * return the proper tag for rollback and recv 852 */ 853 void * 854 zvol_tag(zvol_state_t *zv) 855 { 856 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock)); 857 return (zv->zv_open_count > 0 ? zv : NULL); 858 } 859 860 /* 861 * Suspend the zvol for recv and rollback. 862 */ 863 zvol_state_t * 864 zvol_suspend(const char *name) 865 { 866 zvol_state_t *zv; 867 868 zv = zvol_find_by_name(name, RW_WRITER); 869 870 if (zv == NULL) 871 return (NULL); 872 873 /* block all I/O, release in zvol_resume. */ 874 ASSERT(MUTEX_HELD(&zv->zv_state_lock)); 875 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock)); 876 877 atomic_inc(&zv->zv_suspend_ref); 878 879 if (zv->zv_open_count > 0) 880 zvol_shutdown_zv(zv); 881 882 /* 883 * do not hold zv_state_lock across suspend/resume to 884 * avoid locking up zvol lookups 885 */ 886 mutex_exit(&zv->zv_state_lock); 887 888 /* zv_suspend_lock is released in zvol_resume() */ 889 return (zv); 890 } 891 892 int 893 zvol_resume(zvol_state_t *zv) 894 { 895 int error = 0; 896 897 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock)); 898 899 mutex_enter(&zv->zv_state_lock); 900 901 if (zv->zv_open_count > 0) { 902 VERIFY0(dmu_objset_hold(zv->zv_name, zv, &zv->zv_objset)); 903 VERIFY3P(zv->zv_objset->os_dsl_dataset->ds_owner, ==, zv); 904 VERIFY(dsl_dataset_long_held(zv->zv_objset->os_dsl_dataset)); 905 dmu_objset_rele(zv->zv_objset, zv); 906 907 error = zvol_setup_zv(zv); 908 } 909 910 mutex_exit(&zv->zv_state_lock); 911 912 rw_exit(&zv->zv_suspend_lock); 913 /* 914 * We need this because we don't hold zvol_state_lock while releasing 915 * zv_suspend_lock. zvol_remove_minors_impl thus cannot check 916 * zv_suspend_lock to determine it is safe to free because rwlock is 917 * not inherent atomic. 918 */ 919 atomic_dec(&zv->zv_suspend_ref); 920 921 return (SET_ERROR(error)); 922 } 923 924 int 925 zvol_first_open(zvol_state_t *zv, boolean_t readonly) 926 { 927 objset_t *os; 928 int error; 929 930 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock)); 931 ASSERT(MUTEX_HELD(&zv->zv_state_lock)); 932 ASSERT(mutex_owned(&spa_namespace_lock)); 933 934 boolean_t ro = (readonly || (strchr(zv->zv_name, '@') != NULL)); 935 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, ro, B_TRUE, zv, &os); 936 if (error) 937 return (SET_ERROR(error)); 938 939 zv->zv_objset = os; 940 941 error = zvol_setup_zv(zv); 942 if (error) { 943 dmu_objset_disown(os, 1, zv); 944 zv->zv_objset = NULL; 945 } 946 947 return (error); 948 } 949 950 void 951 zvol_last_close(zvol_state_t *zv) 952 { 953 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock)); 954 ASSERT(MUTEX_HELD(&zv->zv_state_lock)); 955 956 zvol_shutdown_zv(zv); 957 958 dmu_objset_disown(zv->zv_objset, 1, zv); 959 zv->zv_objset = NULL; 960 } 961 962 typedef struct minors_job { 963 list_t *list; 964 list_node_t link; 965 /* input */ 966 char *name; 967 /* output */ 968 int error; 969 } minors_job_t; 970 971 /* 972 * Prefetch zvol dnodes for the minors_job 973 */ 974 static void 975 zvol_prefetch_minors_impl(void *arg) 976 { 977 minors_job_t *job = arg; 978 char *dsname = job->name; 979 objset_t *os = NULL; 980 981 job->error = dmu_objset_own(dsname, DMU_OST_ZVOL, B_TRUE, B_TRUE, 982 FTAG, &os); 983 if (job->error == 0) { 984 dmu_prefetch(os, ZVOL_OBJ, 0, 0, 0, ZIO_PRIORITY_SYNC_READ); 985 dmu_objset_disown(os, B_TRUE, FTAG); 986 } 987 } 988 989 /* 990 * Mask errors to continue dmu_objset_find() traversal 991 */ 992 static int 993 zvol_create_snap_minor_cb(const char *dsname, void *arg) 994 { 995 minors_job_t *j = arg; 996 list_t *minors_list = j->list; 997 const char *name = j->name; 998 999 ASSERT0(MUTEX_HELD(&spa_namespace_lock)); 1000 1001 /* skip the designated dataset */ 1002 if (name && strcmp(dsname, name) == 0) 1003 return (0); 1004 1005 /* at this point, the dsname should name a snapshot */ 1006 if (strchr(dsname, '@') == 0) { 1007 dprintf("zvol_create_snap_minor_cb(): " 1008 "%s is not a snapshot name\n", dsname); 1009 } else { 1010 minors_job_t *job; 1011 char *n = kmem_strdup(dsname); 1012 if (n == NULL) 1013 return (0); 1014 1015 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP); 1016 job->name = n; 1017 job->list = minors_list; 1018 job->error = 0; 1019 list_insert_tail(minors_list, job); 1020 /* don't care if dispatch fails, because job->error is 0 */ 1021 taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job, 1022 TQ_SLEEP); 1023 } 1024 1025 return (0); 1026 } 1027 1028 /* 1029 * If spa_keystore_load_wkey() is called for an encrypted zvol, 1030 * we need to look for any clones also using the key. This function 1031 * is "best effort" - so we just skip over it if there are failures. 1032 */ 1033 static void 1034 zvol_add_clones(const char *dsname, list_t *minors_list) 1035 { 1036 /* Also check if it has clones */ 1037 dsl_dir_t *dd = NULL; 1038 dsl_pool_t *dp = NULL; 1039 1040 if (dsl_pool_hold(dsname, FTAG, &dp) != 0) 1041 return; 1042 1043 if (!spa_feature_is_enabled(dp->dp_spa, 1044 SPA_FEATURE_ENCRYPTION)) 1045 goto out; 1046 1047 if (dsl_dir_hold(dp, dsname, FTAG, &dd, NULL) != 0) 1048 goto out; 1049 1050 if (dsl_dir_phys(dd)->dd_clones == 0) 1051 goto out; 1052 1053 zap_cursor_t *zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP); 1054 zap_attribute_t *za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 1055 objset_t *mos = dd->dd_pool->dp_meta_objset; 1056 1057 for (zap_cursor_init(zc, mos, dsl_dir_phys(dd)->dd_clones); 1058 zap_cursor_retrieve(zc, za) == 0; 1059 zap_cursor_advance(zc)) { 1060 dsl_dataset_t *clone; 1061 minors_job_t *job; 1062 1063 if (dsl_dataset_hold_obj(dd->dd_pool, 1064 za->za_first_integer, FTAG, &clone) == 0) { 1065 1066 char name[ZFS_MAX_DATASET_NAME_LEN]; 1067 dsl_dataset_name(clone, name); 1068 1069 char *n = kmem_strdup(name); 1070 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP); 1071 job->name = n; 1072 job->list = minors_list; 1073 job->error = 0; 1074 list_insert_tail(minors_list, job); 1075 1076 dsl_dataset_rele(clone, FTAG); 1077 } 1078 } 1079 zap_cursor_fini(zc); 1080 kmem_free(za, sizeof (zap_attribute_t)); 1081 kmem_free(zc, sizeof (zap_cursor_t)); 1082 1083 out: 1084 if (dd != NULL) 1085 dsl_dir_rele(dd, FTAG); 1086 dsl_pool_rele(dp, FTAG); 1087 } 1088 1089 /* 1090 * Mask errors to continue dmu_objset_find() traversal 1091 */ 1092 static int 1093 zvol_create_minors_cb(const char *dsname, void *arg) 1094 { 1095 uint64_t snapdev; 1096 int error; 1097 list_t *minors_list = arg; 1098 1099 ASSERT0(MUTEX_HELD(&spa_namespace_lock)); 1100 1101 error = dsl_prop_get_integer(dsname, "snapdev", &snapdev, NULL); 1102 if (error) 1103 return (0); 1104 1105 /* 1106 * Given the name and the 'snapdev' property, create device minor nodes 1107 * with the linkages to zvols/snapshots as needed. 1108 * If the name represents a zvol, create a minor node for the zvol, then 1109 * check if its snapshots are 'visible', and if so, iterate over the 1110 * snapshots and create device minor nodes for those. 1111 */ 1112 if (strchr(dsname, '@') == 0) { 1113 minors_job_t *job; 1114 char *n = kmem_strdup(dsname); 1115 if (n == NULL) 1116 return (0); 1117 1118 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP); 1119 job->name = n; 1120 job->list = minors_list; 1121 job->error = 0; 1122 list_insert_tail(minors_list, job); 1123 /* don't care if dispatch fails, because job->error is 0 */ 1124 taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job, 1125 TQ_SLEEP); 1126 1127 zvol_add_clones(dsname, minors_list); 1128 1129 if (snapdev == ZFS_SNAPDEV_VISIBLE) { 1130 /* 1131 * traverse snapshots only, do not traverse children, 1132 * and skip the 'dsname' 1133 */ 1134 (void) dmu_objset_find(dsname, 1135 zvol_create_snap_minor_cb, (void *)job, 1136 DS_FIND_SNAPSHOTS); 1137 } 1138 } else { 1139 dprintf("zvol_create_minors_cb(): %s is not a zvol name\n", 1140 dsname); 1141 } 1142 1143 return (0); 1144 } 1145 1146 /* 1147 * Create minors for the specified dataset, including children and snapshots. 1148 * Pay attention to the 'snapdev' property and iterate over the snapshots 1149 * only if they are 'visible'. This approach allows one to assure that the 1150 * snapshot metadata is read from disk only if it is needed. 1151 * 1152 * The name can represent a dataset to be recursively scanned for zvols and 1153 * their snapshots, or a single zvol snapshot. If the name represents a 1154 * dataset, the scan is performed in two nested stages: 1155 * - scan the dataset for zvols, and 1156 * - for each zvol, create a minor node, then check if the zvol's snapshots 1157 * are 'visible', and only then iterate over the snapshots if needed 1158 * 1159 * If the name represents a snapshot, a check is performed if the snapshot is 1160 * 'visible' (which also verifies that the parent is a zvol), and if so, 1161 * a minor node for that snapshot is created. 1162 */ 1163 void 1164 zvol_create_minors_recursive(const char *name) 1165 { 1166 list_t minors_list; 1167 minors_job_t *job; 1168 1169 if (zvol_inhibit_dev) 1170 return; 1171 1172 /* 1173 * This is the list for prefetch jobs. Whenever we found a match 1174 * during dmu_objset_find, we insert a minors_job to the list and do 1175 * taskq_dispatch to parallel prefetch zvol dnodes. Note we don't need 1176 * any lock because all list operation is done on the current thread. 1177 * 1178 * We will use this list to do zvol_os_create_minor after prefetch 1179 * so we don't have to traverse using dmu_objset_find again. 1180 */ 1181 list_create(&minors_list, sizeof (minors_job_t), 1182 offsetof(minors_job_t, link)); 1183 1184 1185 if (strchr(name, '@') != NULL) { 1186 uint64_t snapdev; 1187 1188 int error = dsl_prop_get_integer(name, "snapdev", 1189 &snapdev, NULL); 1190 1191 if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE) 1192 (void) zvol_os_create_minor(name); 1193 } else { 1194 fstrans_cookie_t cookie = spl_fstrans_mark(); 1195 (void) dmu_objset_find(name, zvol_create_minors_cb, 1196 &minors_list, DS_FIND_CHILDREN); 1197 spl_fstrans_unmark(cookie); 1198 } 1199 1200 taskq_wait_outstanding(system_taskq, 0); 1201 1202 /* 1203 * Prefetch is completed, we can do zvol_os_create_minor 1204 * sequentially. 1205 */ 1206 while ((job = list_head(&minors_list)) != NULL) { 1207 list_remove(&minors_list, job); 1208 if (!job->error) 1209 (void) zvol_os_create_minor(job->name); 1210 kmem_strfree(job->name); 1211 kmem_free(job, sizeof (minors_job_t)); 1212 } 1213 1214 list_destroy(&minors_list); 1215 } 1216 1217 void 1218 zvol_create_minor(const char *name) 1219 { 1220 /* 1221 * Note: the dsl_pool_config_lock must not be held. 1222 * Minor node creation needs to obtain the zvol_state_lock. 1223 * zvol_open() obtains the zvol_state_lock and then the dsl pool 1224 * config lock. Therefore, we can't have the config lock now if 1225 * we are going to wait for the zvol_state_lock, because it 1226 * would be a lock order inversion which could lead to deadlock. 1227 */ 1228 1229 if (zvol_inhibit_dev) 1230 return; 1231 1232 if (strchr(name, '@') != NULL) { 1233 uint64_t snapdev; 1234 1235 int error = dsl_prop_get_integer(name, 1236 "snapdev", &snapdev, NULL); 1237 1238 if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE) 1239 (void) zvol_os_create_minor(name); 1240 } else { 1241 (void) zvol_os_create_minor(name); 1242 } 1243 } 1244 1245 /* 1246 * Remove minors for specified dataset including children and snapshots. 1247 */ 1248 1249 static void 1250 zvol_free_task(void *arg) 1251 { 1252 zvol_os_free(arg); 1253 } 1254 1255 void 1256 zvol_remove_minors_impl(const char *name) 1257 { 1258 zvol_state_t *zv, *zv_next; 1259 int namelen = ((name) ? strlen(name) : 0); 1260 taskqid_t t; 1261 list_t free_list; 1262 1263 if (zvol_inhibit_dev) 1264 return; 1265 1266 list_create(&free_list, sizeof (zvol_state_t), 1267 offsetof(zvol_state_t, zv_next)); 1268 1269 rw_enter(&zvol_state_lock, RW_WRITER); 1270 1271 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { 1272 zv_next = list_next(&zvol_state_list, zv); 1273 1274 mutex_enter(&zv->zv_state_lock); 1275 if (name == NULL || strcmp(zv->zv_name, name) == 0 || 1276 (strncmp(zv->zv_name, name, namelen) == 0 && 1277 (zv->zv_name[namelen] == '/' || 1278 zv->zv_name[namelen] == '@'))) { 1279 /* 1280 * By holding zv_state_lock here, we guarantee that no 1281 * one is currently using this zv 1282 */ 1283 1284 /* If in use, leave alone */ 1285 if (zv->zv_open_count > 0 || 1286 atomic_read(&zv->zv_suspend_ref)) { 1287 mutex_exit(&zv->zv_state_lock); 1288 continue; 1289 } 1290 1291 zvol_remove(zv); 1292 1293 /* 1294 * Cleared while holding zvol_state_lock as a writer 1295 * which will prevent zvol_open() from opening it. 1296 */ 1297 zvol_os_clear_private(zv); 1298 1299 /* Drop zv_state_lock before zvol_free() */ 1300 mutex_exit(&zv->zv_state_lock); 1301 1302 /* Try parallel zv_free, if failed do it in place */ 1303 t = taskq_dispatch(system_taskq, zvol_free_task, zv, 1304 TQ_SLEEP); 1305 if (t == TASKQID_INVALID) 1306 list_insert_head(&free_list, zv); 1307 } else { 1308 mutex_exit(&zv->zv_state_lock); 1309 } 1310 } 1311 rw_exit(&zvol_state_lock); 1312 1313 /* Drop zvol_state_lock before calling zvol_free() */ 1314 while ((zv = list_head(&free_list)) != NULL) { 1315 list_remove(&free_list, zv); 1316 zvol_os_free(zv); 1317 } 1318 } 1319 1320 /* Remove minor for this specific volume only */ 1321 static void 1322 zvol_remove_minor_impl(const char *name) 1323 { 1324 zvol_state_t *zv = NULL, *zv_next; 1325 1326 if (zvol_inhibit_dev) 1327 return; 1328 1329 rw_enter(&zvol_state_lock, RW_WRITER); 1330 1331 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { 1332 zv_next = list_next(&zvol_state_list, zv); 1333 1334 mutex_enter(&zv->zv_state_lock); 1335 if (strcmp(zv->zv_name, name) == 0) { 1336 /* 1337 * By holding zv_state_lock here, we guarantee that no 1338 * one is currently using this zv 1339 */ 1340 1341 /* If in use, leave alone */ 1342 if (zv->zv_open_count > 0 || 1343 atomic_read(&zv->zv_suspend_ref)) { 1344 mutex_exit(&zv->zv_state_lock); 1345 continue; 1346 } 1347 zvol_remove(zv); 1348 1349 zvol_os_clear_private(zv); 1350 mutex_exit(&zv->zv_state_lock); 1351 break; 1352 } else { 1353 mutex_exit(&zv->zv_state_lock); 1354 } 1355 } 1356 1357 /* Drop zvol_state_lock before calling zvol_free() */ 1358 rw_exit(&zvol_state_lock); 1359 1360 if (zv != NULL) 1361 zvol_os_free(zv); 1362 } 1363 1364 /* 1365 * Rename minors for specified dataset including children and snapshots. 1366 */ 1367 static void 1368 zvol_rename_minors_impl(const char *oldname, const char *newname) 1369 { 1370 zvol_state_t *zv, *zv_next; 1371 int oldnamelen; 1372 1373 if (zvol_inhibit_dev) 1374 return; 1375 1376 oldnamelen = strlen(oldname); 1377 1378 rw_enter(&zvol_state_lock, RW_READER); 1379 1380 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { 1381 zv_next = list_next(&zvol_state_list, zv); 1382 1383 mutex_enter(&zv->zv_state_lock); 1384 1385 if (strcmp(zv->zv_name, oldname) == 0) { 1386 zvol_os_rename_minor(zv, newname); 1387 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && 1388 (zv->zv_name[oldnamelen] == '/' || 1389 zv->zv_name[oldnamelen] == '@')) { 1390 char *name = kmem_asprintf("%s%c%s", newname, 1391 zv->zv_name[oldnamelen], 1392 zv->zv_name + oldnamelen + 1); 1393 zvol_os_rename_minor(zv, name); 1394 kmem_strfree(name); 1395 } 1396 1397 mutex_exit(&zv->zv_state_lock); 1398 } 1399 1400 rw_exit(&zvol_state_lock); 1401 } 1402 1403 typedef struct zvol_snapdev_cb_arg { 1404 uint64_t snapdev; 1405 } zvol_snapdev_cb_arg_t; 1406 1407 static int 1408 zvol_set_snapdev_cb(const char *dsname, void *param) 1409 { 1410 zvol_snapdev_cb_arg_t *arg = param; 1411 1412 if (strchr(dsname, '@') == NULL) 1413 return (0); 1414 1415 switch (arg->snapdev) { 1416 case ZFS_SNAPDEV_VISIBLE: 1417 (void) zvol_os_create_minor(dsname); 1418 break; 1419 case ZFS_SNAPDEV_HIDDEN: 1420 (void) zvol_remove_minor_impl(dsname); 1421 break; 1422 } 1423 1424 return (0); 1425 } 1426 1427 static void 1428 zvol_set_snapdev_impl(char *name, uint64_t snapdev) 1429 { 1430 zvol_snapdev_cb_arg_t arg = {snapdev}; 1431 fstrans_cookie_t cookie = spl_fstrans_mark(); 1432 /* 1433 * The zvol_set_snapdev_sync() sets snapdev appropriately 1434 * in the dataset hierarchy. Here, we only scan snapshots. 1435 */ 1436 dmu_objset_find(name, zvol_set_snapdev_cb, &arg, DS_FIND_SNAPSHOTS); 1437 spl_fstrans_unmark(cookie); 1438 } 1439 1440 static void 1441 zvol_set_volmode_impl(char *name, uint64_t volmode) 1442 { 1443 fstrans_cookie_t cookie; 1444 uint64_t old_volmode; 1445 zvol_state_t *zv; 1446 1447 if (strchr(name, '@') != NULL) 1448 return; 1449 1450 /* 1451 * It's unfortunate we need to remove minors before we create new ones: 1452 * this is necessary because our backing gendisk (zvol_state->zv_disk) 1453 * could be different when we set, for instance, volmode from "geom" 1454 * to "dev" (or vice versa). 1455 */ 1456 zv = zvol_find_by_name(name, RW_NONE); 1457 if (zv == NULL && volmode == ZFS_VOLMODE_NONE) 1458 return; 1459 if (zv != NULL) { 1460 old_volmode = zv->zv_volmode; 1461 mutex_exit(&zv->zv_state_lock); 1462 if (old_volmode == volmode) 1463 return; 1464 zvol_wait_close(zv); 1465 } 1466 cookie = spl_fstrans_mark(); 1467 switch (volmode) { 1468 case ZFS_VOLMODE_NONE: 1469 (void) zvol_remove_minor_impl(name); 1470 break; 1471 case ZFS_VOLMODE_GEOM: 1472 case ZFS_VOLMODE_DEV: 1473 (void) zvol_remove_minor_impl(name); 1474 (void) zvol_os_create_minor(name); 1475 break; 1476 case ZFS_VOLMODE_DEFAULT: 1477 (void) zvol_remove_minor_impl(name); 1478 if (zvol_volmode == ZFS_VOLMODE_NONE) 1479 break; 1480 else /* if zvol_volmode is invalid defaults to "geom" */ 1481 (void) zvol_os_create_minor(name); 1482 break; 1483 } 1484 spl_fstrans_unmark(cookie); 1485 } 1486 1487 static zvol_task_t * 1488 zvol_task_alloc(zvol_async_op_t op, const char *name1, const char *name2, 1489 uint64_t value) 1490 { 1491 zvol_task_t *task; 1492 1493 /* Never allow tasks on hidden names. */ 1494 if (name1[0] == '$') 1495 return (NULL); 1496 1497 task = kmem_zalloc(sizeof (zvol_task_t), KM_SLEEP); 1498 task->op = op; 1499 task->value = value; 1500 1501 strlcpy(task->name1, name1, MAXNAMELEN); 1502 if (name2 != NULL) 1503 strlcpy(task->name2, name2, MAXNAMELEN); 1504 1505 return (task); 1506 } 1507 1508 static void 1509 zvol_task_free(zvol_task_t *task) 1510 { 1511 kmem_free(task, sizeof (zvol_task_t)); 1512 } 1513 1514 /* 1515 * The worker thread function performed asynchronously. 1516 */ 1517 static void 1518 zvol_task_cb(void *arg) 1519 { 1520 zvol_task_t *task = arg; 1521 1522 switch (task->op) { 1523 case ZVOL_ASYNC_REMOVE_MINORS: 1524 zvol_remove_minors_impl(task->name1); 1525 break; 1526 case ZVOL_ASYNC_RENAME_MINORS: 1527 zvol_rename_minors_impl(task->name1, task->name2); 1528 break; 1529 case ZVOL_ASYNC_SET_SNAPDEV: 1530 zvol_set_snapdev_impl(task->name1, task->value); 1531 break; 1532 case ZVOL_ASYNC_SET_VOLMODE: 1533 zvol_set_volmode_impl(task->name1, task->value); 1534 break; 1535 default: 1536 VERIFY(0); 1537 break; 1538 } 1539 1540 zvol_task_free(task); 1541 } 1542 1543 typedef struct zvol_set_prop_int_arg { 1544 const char *zsda_name; 1545 uint64_t zsda_value; 1546 zprop_source_t zsda_source; 1547 dmu_tx_t *zsda_tx; 1548 } zvol_set_prop_int_arg_t; 1549 1550 /* 1551 * Sanity check the dataset for safe use by the sync task. No additional 1552 * conditions are imposed. 1553 */ 1554 static int 1555 zvol_set_snapdev_check(void *arg, dmu_tx_t *tx) 1556 { 1557 zvol_set_prop_int_arg_t *zsda = arg; 1558 dsl_pool_t *dp = dmu_tx_pool(tx); 1559 dsl_dir_t *dd; 1560 int error; 1561 1562 error = dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL); 1563 if (error != 0) 1564 return (error); 1565 1566 dsl_dir_rele(dd, FTAG); 1567 1568 return (error); 1569 } 1570 1571 static int 1572 zvol_set_snapdev_sync_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 1573 { 1574 (void) arg; 1575 char dsname[MAXNAMELEN]; 1576 zvol_task_t *task; 1577 uint64_t snapdev; 1578 1579 dsl_dataset_name(ds, dsname); 1580 if (dsl_prop_get_int_ds(ds, "snapdev", &snapdev) != 0) 1581 return (0); 1582 task = zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV, dsname, NULL, snapdev); 1583 if (task == NULL) 1584 return (0); 1585 1586 (void) taskq_dispatch(dp->dp_spa->spa_zvol_taskq, zvol_task_cb, 1587 task, TQ_SLEEP); 1588 return (0); 1589 } 1590 1591 /* 1592 * Traverse all child datasets and apply snapdev appropriately. 1593 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel 1594 * dataset and read the effective "snapdev" on every child in the callback 1595 * function: this is because the value is not guaranteed to be the same in the 1596 * whole dataset hierarchy. 1597 */ 1598 static void 1599 zvol_set_snapdev_sync(void *arg, dmu_tx_t *tx) 1600 { 1601 zvol_set_prop_int_arg_t *zsda = arg; 1602 dsl_pool_t *dp = dmu_tx_pool(tx); 1603 dsl_dir_t *dd; 1604 dsl_dataset_t *ds; 1605 int error; 1606 1607 VERIFY0(dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL)); 1608 zsda->zsda_tx = tx; 1609 1610 error = dsl_dataset_hold(dp, zsda->zsda_name, FTAG, &ds); 1611 if (error == 0) { 1612 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_SNAPDEV), 1613 zsda->zsda_source, sizeof (zsda->zsda_value), 1, 1614 &zsda->zsda_value, zsda->zsda_tx); 1615 dsl_dataset_rele(ds, FTAG); 1616 } 1617 dmu_objset_find_dp(dp, dd->dd_object, zvol_set_snapdev_sync_cb, 1618 zsda, DS_FIND_CHILDREN); 1619 1620 dsl_dir_rele(dd, FTAG); 1621 } 1622 1623 int 1624 zvol_set_snapdev(const char *ddname, zprop_source_t source, uint64_t snapdev) 1625 { 1626 zvol_set_prop_int_arg_t zsda; 1627 1628 zsda.zsda_name = ddname; 1629 zsda.zsda_source = source; 1630 zsda.zsda_value = snapdev; 1631 1632 return (dsl_sync_task(ddname, zvol_set_snapdev_check, 1633 zvol_set_snapdev_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE)); 1634 } 1635 1636 /* 1637 * Sanity check the dataset for safe use by the sync task. No additional 1638 * conditions are imposed. 1639 */ 1640 static int 1641 zvol_set_volmode_check(void *arg, dmu_tx_t *tx) 1642 { 1643 zvol_set_prop_int_arg_t *zsda = arg; 1644 dsl_pool_t *dp = dmu_tx_pool(tx); 1645 dsl_dir_t *dd; 1646 int error; 1647 1648 error = dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL); 1649 if (error != 0) 1650 return (error); 1651 1652 dsl_dir_rele(dd, FTAG); 1653 1654 return (error); 1655 } 1656 1657 static int 1658 zvol_set_volmode_sync_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 1659 { 1660 (void) arg; 1661 char dsname[MAXNAMELEN]; 1662 zvol_task_t *task; 1663 uint64_t volmode; 1664 1665 dsl_dataset_name(ds, dsname); 1666 if (dsl_prop_get_int_ds(ds, "volmode", &volmode) != 0) 1667 return (0); 1668 task = zvol_task_alloc(ZVOL_ASYNC_SET_VOLMODE, dsname, NULL, volmode); 1669 if (task == NULL) 1670 return (0); 1671 1672 (void) taskq_dispatch(dp->dp_spa->spa_zvol_taskq, zvol_task_cb, 1673 task, TQ_SLEEP); 1674 return (0); 1675 } 1676 1677 /* 1678 * Traverse all child datasets and apply volmode appropriately. 1679 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel 1680 * dataset and read the effective "volmode" on every child in the callback 1681 * function: this is because the value is not guaranteed to be the same in the 1682 * whole dataset hierarchy. 1683 */ 1684 static void 1685 zvol_set_volmode_sync(void *arg, dmu_tx_t *tx) 1686 { 1687 zvol_set_prop_int_arg_t *zsda = arg; 1688 dsl_pool_t *dp = dmu_tx_pool(tx); 1689 dsl_dir_t *dd; 1690 dsl_dataset_t *ds; 1691 int error; 1692 1693 VERIFY0(dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL)); 1694 zsda->zsda_tx = tx; 1695 1696 error = dsl_dataset_hold(dp, zsda->zsda_name, FTAG, &ds); 1697 if (error == 0) { 1698 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_VOLMODE), 1699 zsda->zsda_source, sizeof (zsda->zsda_value), 1, 1700 &zsda->zsda_value, zsda->zsda_tx); 1701 dsl_dataset_rele(ds, FTAG); 1702 } 1703 1704 dmu_objset_find_dp(dp, dd->dd_object, zvol_set_volmode_sync_cb, 1705 zsda, DS_FIND_CHILDREN); 1706 1707 dsl_dir_rele(dd, FTAG); 1708 } 1709 1710 int 1711 zvol_set_volmode(const char *ddname, zprop_source_t source, uint64_t volmode) 1712 { 1713 zvol_set_prop_int_arg_t zsda; 1714 1715 zsda.zsda_name = ddname; 1716 zsda.zsda_source = source; 1717 zsda.zsda_value = volmode; 1718 1719 return (dsl_sync_task(ddname, zvol_set_volmode_check, 1720 zvol_set_volmode_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE)); 1721 } 1722 1723 void 1724 zvol_remove_minors(spa_t *spa, const char *name, boolean_t async) 1725 { 1726 zvol_task_t *task; 1727 taskqid_t id; 1728 1729 task = zvol_task_alloc(ZVOL_ASYNC_REMOVE_MINORS, name, NULL, ~0ULL); 1730 if (task == NULL) 1731 return; 1732 1733 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP); 1734 if ((async == B_FALSE) && (id != TASKQID_INVALID)) 1735 taskq_wait_id(spa->spa_zvol_taskq, id); 1736 } 1737 1738 void 1739 zvol_rename_minors(spa_t *spa, const char *name1, const char *name2, 1740 boolean_t async) 1741 { 1742 zvol_task_t *task; 1743 taskqid_t id; 1744 1745 task = zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS, name1, name2, ~0ULL); 1746 if (task == NULL) 1747 return; 1748 1749 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP); 1750 if ((async == B_FALSE) && (id != TASKQID_INVALID)) 1751 taskq_wait_id(spa->spa_zvol_taskq, id); 1752 } 1753 1754 boolean_t 1755 zvol_is_zvol(const char *name) 1756 { 1757 1758 return (zvol_os_is_zvol(name)); 1759 } 1760 1761 int 1762 zvol_init_impl(void) 1763 { 1764 int i; 1765 1766 list_create(&zvol_state_list, sizeof (zvol_state_t), 1767 offsetof(zvol_state_t, zv_next)); 1768 rw_init(&zvol_state_lock, NULL, RW_DEFAULT, NULL); 1769 1770 zvol_htable = kmem_alloc(ZVOL_HT_SIZE * sizeof (struct hlist_head), 1771 KM_SLEEP); 1772 for (i = 0; i < ZVOL_HT_SIZE; i++) 1773 INIT_HLIST_HEAD(&zvol_htable[i]); 1774 1775 return (0); 1776 } 1777 1778 void 1779 zvol_fini_impl(void) 1780 { 1781 zvol_remove_minors_impl(NULL); 1782 1783 /* 1784 * The call to "zvol_remove_minors_impl" may dispatch entries to 1785 * the system_taskq, but it doesn't wait for those entries to 1786 * complete before it returns. Thus, we must wait for all of the 1787 * removals to finish, before we can continue. 1788 */ 1789 taskq_wait_outstanding(system_taskq, 0); 1790 1791 kmem_free(zvol_htable, ZVOL_HT_SIZE * sizeof (struct hlist_head)); 1792 list_destroy(&zvol_state_list); 1793 rw_destroy(&zvol_state_lock); 1794 } 1795