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) 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 list_t zvol_state_list; 94 krwlock_t zvol_state_lock; 95 const zvol_platform_ops_t *ops; 96 97 typedef enum { 98 ZVOL_ASYNC_REMOVE_MINORS, 99 ZVOL_ASYNC_RENAME_MINORS, 100 ZVOL_ASYNC_SET_SNAPDEV, 101 ZVOL_ASYNC_SET_VOLMODE, 102 ZVOL_ASYNC_MAX 103 } zvol_async_op_t; 104 105 typedef struct { 106 zvol_async_op_t op; 107 char name1[MAXNAMELEN]; 108 char name2[MAXNAMELEN]; 109 uint64_t value; 110 } zvol_task_t; 111 112 uint64_t 113 zvol_name_hash(const char *name) 114 { 115 int i; 116 uint64_t crc = -1ULL; 117 const uint8_t *p = (const uint8_t *)name; 118 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); 119 for (i = 0; i < MAXNAMELEN - 1 && *p; i++, p++) { 120 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (*p)) & 0xFF]; 121 } 122 return (crc); 123 } 124 125 /* 126 * Find a zvol_state_t given the name and hash generated by zvol_name_hash. 127 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise, 128 * return (NULL) without the taking locks. The zv_suspend_lock is always taken 129 * before zv_state_lock. The mode argument indicates the mode (including none) 130 * for zv_suspend_lock to be taken. 131 */ 132 zvol_state_t * 133 zvol_find_by_name_hash(const char *name, uint64_t hash, int mode) 134 { 135 zvol_state_t *zv; 136 struct hlist_node *p = NULL; 137 138 rw_enter(&zvol_state_lock, RW_READER); 139 hlist_for_each(p, ZVOL_HT_HEAD(hash)) { 140 zv = hlist_entry(p, zvol_state_t, zv_hlink); 141 mutex_enter(&zv->zv_state_lock); 142 if (zv->zv_hash == hash && 143 strncmp(zv->zv_name, name, MAXNAMELEN) == 0) { 144 /* 145 * this is the right zvol, take the locks in the 146 * right order 147 */ 148 if (mode != RW_NONE && 149 !rw_tryenter(&zv->zv_suspend_lock, mode)) { 150 mutex_exit(&zv->zv_state_lock); 151 rw_enter(&zv->zv_suspend_lock, mode); 152 mutex_enter(&zv->zv_state_lock); 153 /* 154 * zvol cannot be renamed as we continue 155 * to hold zvol_state_lock 156 */ 157 ASSERT(zv->zv_hash == hash && 158 strncmp(zv->zv_name, name, MAXNAMELEN) 159 == 0); 160 } 161 rw_exit(&zvol_state_lock); 162 return (zv); 163 } 164 mutex_exit(&zv->zv_state_lock); 165 } 166 rw_exit(&zvol_state_lock); 167 168 return (NULL); 169 } 170 171 /* 172 * Find a zvol_state_t given the name. 173 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise, 174 * return (NULL) without the taking locks. The zv_suspend_lock is always taken 175 * before zv_state_lock. The mode argument indicates the mode (including none) 176 * for zv_suspend_lock to be taken. 177 */ 178 static zvol_state_t * 179 zvol_find_by_name(const char *name, int mode) 180 { 181 return (zvol_find_by_name_hash(name, zvol_name_hash(name), mode)); 182 } 183 184 /* 185 * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation. 186 */ 187 void 188 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 189 { 190 zfs_creat_t *zct = arg; 191 nvlist_t *nvprops = zct->zct_props; 192 int error; 193 uint64_t volblocksize, volsize; 194 195 VERIFY(nvlist_lookup_uint64(nvprops, 196 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); 197 if (nvlist_lookup_uint64(nvprops, 198 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) 199 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); 200 201 /* 202 * These properties must be removed from the list so the generic 203 * property setting step won't apply to them. 204 */ 205 VERIFY(nvlist_remove_all(nvprops, 206 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); 207 (void) nvlist_remove_all(nvprops, 208 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); 209 210 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, 211 DMU_OT_NONE, 0, tx); 212 ASSERT(error == 0); 213 214 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, 215 DMU_OT_NONE, 0, tx); 216 ASSERT(error == 0); 217 218 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); 219 ASSERT(error == 0); 220 } 221 222 /* 223 * ZFS_IOC_OBJSET_STATS entry point. 224 */ 225 int 226 zvol_get_stats(objset_t *os, nvlist_t *nv) 227 { 228 int error; 229 dmu_object_info_t *doi; 230 uint64_t val; 231 232 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); 233 if (error) 234 return (SET_ERROR(error)); 235 236 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); 237 doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP); 238 error = dmu_object_info(os, ZVOL_OBJ, doi); 239 240 if (error == 0) { 241 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, 242 doi->doi_data_block_size); 243 } 244 245 kmem_free(doi, sizeof (dmu_object_info_t)); 246 247 return (SET_ERROR(error)); 248 } 249 250 /* 251 * Sanity check volume size. 252 */ 253 int 254 zvol_check_volsize(uint64_t volsize, uint64_t blocksize) 255 { 256 if (volsize == 0) 257 return (SET_ERROR(EINVAL)); 258 259 if (volsize % blocksize != 0) 260 return (SET_ERROR(EINVAL)); 261 262 #ifdef _ILP32 263 if (volsize - 1 > SPEC_MAXOFFSET_T) 264 return (SET_ERROR(EOVERFLOW)); 265 #endif 266 return (0); 267 } 268 269 /* 270 * Ensure the zap is flushed then inform the VFS of the capacity change. 271 */ 272 static int 273 zvol_update_volsize(uint64_t volsize, objset_t *os) 274 { 275 dmu_tx_t *tx; 276 int error; 277 uint64_t txg; 278 279 tx = dmu_tx_create(os); 280 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); 281 dmu_tx_mark_netfree(tx); 282 error = dmu_tx_assign(tx, TXG_WAIT); 283 if (error) { 284 dmu_tx_abort(tx); 285 return (SET_ERROR(error)); 286 } 287 txg = dmu_tx_get_txg(tx); 288 289 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, 290 &volsize, tx); 291 dmu_tx_commit(tx); 292 293 txg_wait_synced(dmu_objset_pool(os), txg); 294 295 if (error == 0) 296 error = dmu_free_long_range(os, 297 ZVOL_OBJ, volsize, DMU_OBJECT_END); 298 299 return (error); 300 } 301 302 /* 303 * Set ZFS_PROP_VOLSIZE set entry point. Note that modifying the volume 304 * size will result in a udev "change" event being generated. 305 */ 306 int 307 zvol_set_volsize(const char *name, uint64_t volsize) 308 { 309 objset_t *os = NULL; 310 uint64_t readonly; 311 int error; 312 boolean_t owned = B_FALSE; 313 314 error = dsl_prop_get_integer(name, 315 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL); 316 if (error != 0) 317 return (SET_ERROR(error)); 318 if (readonly) 319 return (SET_ERROR(EROFS)); 320 321 zvol_state_t *zv = zvol_find_by_name(name, RW_READER); 322 323 ASSERT(zv == NULL || (MUTEX_HELD(&zv->zv_state_lock) && 324 RW_READ_HELD(&zv->zv_suspend_lock))); 325 326 if (zv == NULL || zv->zv_objset == NULL) { 327 if (zv != NULL) 328 rw_exit(&zv->zv_suspend_lock); 329 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, B_TRUE, 330 FTAG, &os)) != 0) { 331 if (zv != NULL) 332 mutex_exit(&zv->zv_state_lock); 333 return (SET_ERROR(error)); 334 } 335 owned = B_TRUE; 336 if (zv != NULL) 337 zv->zv_objset = os; 338 } else { 339 os = zv->zv_objset; 340 } 341 342 dmu_object_info_t *doi = kmem_alloc(sizeof (*doi), KM_SLEEP); 343 344 if ((error = dmu_object_info(os, ZVOL_OBJ, doi)) || 345 (error = zvol_check_volsize(volsize, doi->doi_data_block_size))) 346 goto out; 347 348 error = zvol_update_volsize(volsize, os); 349 if (error == 0 && zv != NULL) { 350 zv->zv_volsize = volsize; 351 zv->zv_changed = 1; 352 } 353 out: 354 kmem_free(doi, sizeof (dmu_object_info_t)); 355 356 if (owned) { 357 dmu_objset_disown(os, B_TRUE, FTAG); 358 if (zv != NULL) 359 zv->zv_objset = NULL; 360 } else { 361 rw_exit(&zv->zv_suspend_lock); 362 } 363 364 if (zv != NULL) 365 mutex_exit(&zv->zv_state_lock); 366 367 if (error == 0 && zv != NULL) 368 ops->zv_update_volsize(zv, volsize); 369 370 return (SET_ERROR(error)); 371 } 372 373 /* 374 * Sanity check volume block size. 375 */ 376 int 377 zvol_check_volblocksize(const char *name, uint64_t volblocksize) 378 { 379 /* Record sizes above 128k need the feature to be enabled */ 380 if (volblocksize > SPA_OLD_MAXBLOCKSIZE) { 381 spa_t *spa; 382 int error; 383 384 if ((error = spa_open(name, &spa, FTAG)) != 0) 385 return (error); 386 387 if (!spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 388 spa_close(spa, FTAG); 389 return (SET_ERROR(ENOTSUP)); 390 } 391 392 /* 393 * We don't allow setting the property above 1MB, 394 * unless the tunable has been changed. 395 */ 396 if (volblocksize > zfs_max_recordsize) 397 return (SET_ERROR(EDOM)); 398 399 spa_close(spa, FTAG); 400 } 401 402 if (volblocksize < SPA_MINBLOCKSIZE || 403 volblocksize > SPA_MAXBLOCKSIZE || 404 !ISP2(volblocksize)) 405 return (SET_ERROR(EDOM)); 406 407 return (0); 408 } 409 410 /* 411 * Set ZFS_PROP_VOLBLOCKSIZE set entry point. 412 */ 413 int 414 zvol_set_volblocksize(const char *name, uint64_t volblocksize) 415 { 416 zvol_state_t *zv; 417 dmu_tx_t *tx; 418 int error; 419 420 zv = zvol_find_by_name(name, RW_READER); 421 422 if (zv == NULL) 423 return (SET_ERROR(ENXIO)); 424 425 ASSERT(MUTEX_HELD(&zv->zv_state_lock)); 426 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock)); 427 428 if (zv->zv_flags & ZVOL_RDONLY) { 429 mutex_exit(&zv->zv_state_lock); 430 rw_exit(&zv->zv_suspend_lock); 431 return (SET_ERROR(EROFS)); 432 } 433 434 tx = dmu_tx_create(zv->zv_objset); 435 dmu_tx_hold_bonus(tx, ZVOL_OBJ); 436 error = dmu_tx_assign(tx, TXG_WAIT); 437 if (error) { 438 dmu_tx_abort(tx); 439 } else { 440 error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ, 441 volblocksize, 0, tx); 442 if (error == ENOTSUP) 443 error = SET_ERROR(EBUSY); 444 dmu_tx_commit(tx); 445 if (error == 0) 446 zv->zv_volblocksize = volblocksize; 447 } 448 449 mutex_exit(&zv->zv_state_lock); 450 rw_exit(&zv->zv_suspend_lock); 451 452 return (SET_ERROR(error)); 453 } 454 455 /* 456 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we 457 * implement DKIOCFREE/free-long-range. 458 */ 459 static int 460 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap) 461 { 462 zvol_state_t *zv = arg1; 463 lr_truncate_t *lr = arg2; 464 uint64_t offset, length; 465 466 if (byteswap) 467 byteswap_uint64_array(lr, sizeof (*lr)); 468 469 offset = lr->lr_offset; 470 length = lr->lr_length; 471 472 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); 473 dmu_tx_mark_netfree(tx); 474 int error = dmu_tx_assign(tx, TXG_WAIT); 475 if (error != 0) { 476 dmu_tx_abort(tx); 477 } else { 478 zil_replaying(zv->zv_zilog, tx); 479 dmu_tx_commit(tx); 480 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, 481 length); 482 } 483 484 return (error); 485 } 486 487 /* 488 * Replay a TX_WRITE ZIL transaction that didn't get committed 489 * after a system failure 490 */ 491 static int 492 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap) 493 { 494 zvol_state_t *zv = arg1; 495 lr_write_t *lr = arg2; 496 objset_t *os = zv->zv_objset; 497 char *data = (char *)(lr + 1); /* data follows lr_write_t */ 498 uint64_t offset, length; 499 dmu_tx_t *tx; 500 int error; 501 502 if (byteswap) 503 byteswap_uint64_array(lr, sizeof (*lr)); 504 505 offset = lr->lr_offset; 506 length = lr->lr_length; 507 508 /* If it's a dmu_sync() block, write the whole block */ 509 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { 510 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); 511 if (length < blocksize) { 512 offset -= offset % blocksize; 513 length = blocksize; 514 } 515 } 516 517 tx = dmu_tx_create(os); 518 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); 519 error = dmu_tx_assign(tx, TXG_WAIT); 520 if (error) { 521 dmu_tx_abort(tx); 522 } else { 523 dmu_write(os, ZVOL_OBJ, offset, length, data, tx); 524 zil_replaying(zv->zv_zilog, tx); 525 dmu_tx_commit(tx); 526 } 527 528 return (error); 529 } 530 531 static int 532 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap) 533 { 534 return (SET_ERROR(ENOTSUP)); 535 } 536 537 /* 538 * Callback vectors for replaying records. 539 * Only TX_WRITE and TX_TRUNCATE are needed for zvol. 540 */ 541 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { 542 zvol_replay_err, /* no such transaction type */ 543 zvol_replay_err, /* TX_CREATE */ 544 zvol_replay_err, /* TX_MKDIR */ 545 zvol_replay_err, /* TX_MKXATTR */ 546 zvol_replay_err, /* TX_SYMLINK */ 547 zvol_replay_err, /* TX_REMOVE */ 548 zvol_replay_err, /* TX_RMDIR */ 549 zvol_replay_err, /* TX_LINK */ 550 zvol_replay_err, /* TX_RENAME */ 551 zvol_replay_write, /* TX_WRITE */ 552 zvol_replay_truncate, /* TX_TRUNCATE */ 553 zvol_replay_err, /* TX_SETATTR */ 554 zvol_replay_err, /* TX_ACL */ 555 zvol_replay_err, /* TX_CREATE_ATTR */ 556 zvol_replay_err, /* TX_CREATE_ACL_ATTR */ 557 zvol_replay_err, /* TX_MKDIR_ACL */ 558 zvol_replay_err, /* TX_MKDIR_ATTR */ 559 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ 560 zvol_replay_err, /* TX_WRITE2 */ 561 }; 562 563 /* 564 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. 565 * 566 * We store data in the log buffers if it's small enough. 567 * Otherwise we will later flush the data out via dmu_sync(). 568 */ 569 ssize_t zvol_immediate_write_sz = 32768; 570 571 void 572 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset, 573 uint64_t size, int sync) 574 { 575 uint32_t blocksize = zv->zv_volblocksize; 576 zilog_t *zilog = zv->zv_zilog; 577 itx_wr_state_t write_state; 578 uint64_t sz = size; 579 580 if (zil_replaying(zilog, tx)) 581 return; 582 583 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) 584 write_state = WR_INDIRECT; 585 else if (!spa_has_slogs(zilog->zl_spa) && 586 size >= blocksize && blocksize > zvol_immediate_write_sz) 587 write_state = WR_INDIRECT; 588 else if (sync) 589 write_state = WR_COPIED; 590 else 591 write_state = WR_NEED_COPY; 592 593 while (size) { 594 itx_t *itx; 595 lr_write_t *lr; 596 itx_wr_state_t wr_state = write_state; 597 ssize_t len = size; 598 599 if (wr_state == WR_COPIED && size > zil_max_copied_data(zilog)) 600 wr_state = WR_NEED_COPY; 601 else if (wr_state == WR_INDIRECT) 602 len = MIN(blocksize - P2PHASE(offset, blocksize), size); 603 604 itx = zil_itx_create(TX_WRITE, sizeof (*lr) + 605 (wr_state == WR_COPIED ? len : 0)); 606 lr = (lr_write_t *)&itx->itx_lr; 607 if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn, 608 offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) { 609 zil_itx_destroy(itx); 610 itx = zil_itx_create(TX_WRITE, sizeof (*lr)); 611 lr = (lr_write_t *)&itx->itx_lr; 612 wr_state = WR_NEED_COPY; 613 } 614 615 itx->itx_wr_state = wr_state; 616 lr->lr_foid = ZVOL_OBJ; 617 lr->lr_offset = offset; 618 lr->lr_length = len; 619 lr->lr_blkoff = 0; 620 BP_ZERO(&lr->lr_blkptr); 621 622 itx->itx_private = zv; 623 itx->itx_sync = sync; 624 625 (void) zil_itx_assign(zilog, itx, tx); 626 627 offset += len; 628 size -= len; 629 } 630 631 if (write_state == WR_COPIED || write_state == WR_NEED_COPY) { 632 dsl_pool_wrlog_count(zilog->zl_dmu_pool, sz, tx->tx_txg); 633 } 634 } 635 636 /* 637 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. 638 */ 639 void 640 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, 641 boolean_t sync) 642 { 643 itx_t *itx; 644 lr_truncate_t *lr; 645 zilog_t *zilog = zv->zv_zilog; 646 647 if (zil_replaying(zilog, tx)) 648 return; 649 650 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); 651 lr = (lr_truncate_t *)&itx->itx_lr; 652 lr->lr_foid = ZVOL_OBJ; 653 lr->lr_offset = off; 654 lr->lr_length = len; 655 656 itx->itx_sync = sync; 657 zil_itx_assign(zilog, itx, tx); 658 } 659 660 661 /* ARGSUSED */ 662 static void 663 zvol_get_done(zgd_t *zgd, int error) 664 { 665 if (zgd->zgd_db) 666 dmu_buf_rele(zgd->zgd_db, zgd); 667 668 zfs_rangelock_exit(zgd->zgd_lr); 669 670 kmem_free(zgd, sizeof (zgd_t)); 671 } 672 673 /* 674 * Get data to generate a TX_WRITE intent log record. 675 */ 676 int 677 zvol_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf, 678 struct lwb *lwb, zio_t *zio) 679 { 680 zvol_state_t *zv = arg; 681 uint64_t offset = lr->lr_offset; 682 uint64_t size = lr->lr_length; 683 dmu_buf_t *db; 684 zgd_t *zgd; 685 int error; 686 687 ASSERT3P(lwb, !=, NULL); 688 ASSERT3P(zio, !=, NULL); 689 ASSERT3U(size, !=, 0); 690 691 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 692 zgd->zgd_lwb = lwb; 693 694 /* 695 * Write records come in two flavors: immediate and indirect. 696 * For small writes it's cheaper to store the data with the 697 * log record (immediate); for large writes it's cheaper to 698 * sync the data and get a pointer to it (indirect) so that 699 * we don't have to write the data twice. 700 */ 701 if (buf != NULL) { /* immediate write */ 702 zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset, 703 size, RL_READER); 704 error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf, 705 DMU_READ_NO_PREFETCH); 706 } else { /* indirect write */ 707 /* 708 * Have to lock the whole block to ensure when it's written out 709 * and its checksum is being calculated that no one can change 710 * the data. Contrarily to zfs_get_data we need not re-check 711 * blocksize after we get the lock because it cannot be changed. 712 */ 713 size = zv->zv_volblocksize; 714 offset = P2ALIGN_TYPED(offset, size, uint64_t); 715 zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset, 716 size, RL_READER); 717 error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db, 718 DMU_READ_NO_PREFETCH); 719 if (error == 0) { 720 blkptr_t *bp = &lr->lr_blkptr; 721 722 zgd->zgd_db = db; 723 zgd->zgd_bp = bp; 724 725 ASSERT(db != NULL); 726 ASSERT(db->db_offset == offset); 727 ASSERT(db->db_size == size); 728 729 error = dmu_sync(zio, lr->lr_common.lrc_txg, 730 zvol_get_done, zgd); 731 732 if (error == 0) 733 return (0); 734 } 735 } 736 737 zvol_get_done(zgd, error); 738 739 return (SET_ERROR(error)); 740 } 741 742 /* 743 * The zvol_state_t's are inserted into zvol_state_list and zvol_htable. 744 */ 745 746 void 747 zvol_insert(zvol_state_t *zv) 748 { 749 ASSERT(RW_WRITE_HELD(&zvol_state_lock)); 750 list_insert_head(&zvol_state_list, zv); 751 hlist_add_head(&zv->zv_hlink, ZVOL_HT_HEAD(zv->zv_hash)); 752 } 753 754 /* 755 * Simply remove the zvol from to list of zvols. 756 */ 757 static void 758 zvol_remove(zvol_state_t *zv) 759 { 760 ASSERT(RW_WRITE_HELD(&zvol_state_lock)); 761 list_remove(&zvol_state_list, zv); 762 hlist_del(&zv->zv_hlink); 763 } 764 765 /* 766 * Setup zv after we just own the zv->objset 767 */ 768 static int 769 zvol_setup_zv(zvol_state_t *zv) 770 { 771 uint64_t volsize; 772 int error; 773 uint64_t ro; 774 objset_t *os = zv->zv_objset; 775 776 ASSERT(MUTEX_HELD(&zv->zv_state_lock)); 777 ASSERT(RW_LOCK_HELD(&zv->zv_suspend_lock)); 778 779 zv->zv_zilog = NULL; 780 zv->zv_flags &= ~ZVOL_WRITTEN_TO; 781 782 error = dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL); 783 if (error) 784 return (SET_ERROR(error)); 785 786 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); 787 if (error) 788 return (SET_ERROR(error)); 789 790 error = dnode_hold(os, ZVOL_OBJ, zv, &zv->zv_dn); 791 if (error) 792 return (SET_ERROR(error)); 793 794 ops->zv_set_capacity(zv, volsize >> 9); 795 zv->zv_volsize = volsize; 796 797 if (ro || dmu_objset_is_snapshot(os) || 798 !spa_writeable(dmu_objset_spa(os))) { 799 ops->zv_set_disk_ro(zv, 1); 800 zv->zv_flags |= ZVOL_RDONLY; 801 } else { 802 ops->zv_set_disk_ro(zv, 0); 803 zv->zv_flags &= ~ZVOL_RDONLY; 804 } 805 return (0); 806 } 807 808 /* 809 * Shutdown every zv_objset related stuff except zv_objset itself. 810 * The is the reverse of zvol_setup_zv. 811 */ 812 static void 813 zvol_shutdown_zv(zvol_state_t *zv) 814 { 815 ASSERT(MUTEX_HELD(&zv->zv_state_lock) && 816 RW_LOCK_HELD(&zv->zv_suspend_lock)); 817 818 if (zv->zv_flags & ZVOL_WRITTEN_TO) { 819 ASSERT(zv->zv_zilog != NULL); 820 zil_close(zv->zv_zilog); 821 } 822 823 zv->zv_zilog = NULL; 824 825 dnode_rele(zv->zv_dn, zv); 826 zv->zv_dn = NULL; 827 828 /* 829 * Evict cached data. We must write out any dirty data before 830 * disowning the dataset. 831 */ 832 if (zv->zv_flags & ZVOL_WRITTEN_TO) 833 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); 834 (void) dmu_objset_evict_dbufs(zv->zv_objset); 835 } 836 837 /* 838 * return the proper tag for rollback and recv 839 */ 840 void * 841 zvol_tag(zvol_state_t *zv) 842 { 843 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock)); 844 return (zv->zv_open_count > 0 ? zv : NULL); 845 } 846 847 /* 848 * Suspend the zvol for recv and rollback. 849 */ 850 zvol_state_t * 851 zvol_suspend(const char *name) 852 { 853 zvol_state_t *zv; 854 855 zv = zvol_find_by_name(name, RW_WRITER); 856 857 if (zv == NULL) 858 return (NULL); 859 860 /* block all I/O, release in zvol_resume. */ 861 ASSERT(MUTEX_HELD(&zv->zv_state_lock)); 862 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock)); 863 864 atomic_inc(&zv->zv_suspend_ref); 865 866 if (zv->zv_open_count > 0) 867 zvol_shutdown_zv(zv); 868 869 /* 870 * do not hold zv_state_lock across suspend/resume to 871 * avoid locking up zvol lookups 872 */ 873 mutex_exit(&zv->zv_state_lock); 874 875 /* zv_suspend_lock is released in zvol_resume() */ 876 return (zv); 877 } 878 879 int 880 zvol_resume(zvol_state_t *zv) 881 { 882 int error = 0; 883 884 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock)); 885 886 mutex_enter(&zv->zv_state_lock); 887 888 if (zv->zv_open_count > 0) { 889 VERIFY0(dmu_objset_hold(zv->zv_name, zv, &zv->zv_objset)); 890 VERIFY3P(zv->zv_objset->os_dsl_dataset->ds_owner, ==, zv); 891 VERIFY(dsl_dataset_long_held(zv->zv_objset->os_dsl_dataset)); 892 dmu_objset_rele(zv->zv_objset, zv); 893 894 error = zvol_setup_zv(zv); 895 } 896 897 mutex_exit(&zv->zv_state_lock); 898 899 rw_exit(&zv->zv_suspend_lock); 900 /* 901 * We need this because we don't hold zvol_state_lock while releasing 902 * zv_suspend_lock. zvol_remove_minors_impl thus cannot check 903 * zv_suspend_lock to determine it is safe to free because rwlock is 904 * not inherent atomic. 905 */ 906 atomic_dec(&zv->zv_suspend_ref); 907 908 return (SET_ERROR(error)); 909 } 910 911 int 912 zvol_first_open(zvol_state_t *zv, boolean_t readonly) 913 { 914 objset_t *os; 915 int error, locked = 0; 916 boolean_t ro; 917 918 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock)); 919 ASSERT(MUTEX_HELD(&zv->zv_state_lock)); 920 921 /* 922 * In all other cases the spa_namespace_lock is taken before the 923 * bdev->bd_mutex lock. But in this case the Linux __blkdev_get() 924 * function calls fops->open() with the bdev->bd_mutex lock held. 925 * This deadlock can be easily observed with zvols used as vdevs. 926 * 927 * To avoid a potential lock inversion deadlock we preemptively 928 * try to take the spa_namespace_lock(). Normally it will not 929 * be contended and this is safe because spa_open_common() handles 930 * the case where the caller already holds the spa_namespace_lock. 931 * 932 * When it is contended we risk a lock inversion if we were to 933 * block waiting for the lock. Luckily, the __blkdev_get() 934 * function allows us to return -ERESTARTSYS which will result in 935 * bdev->bd_mutex being dropped, reacquired, and fops->open() being 936 * called again. This process can be repeated safely until both 937 * locks are acquired. 938 */ 939 if (!mutex_owned(&spa_namespace_lock)) { 940 locked = mutex_tryenter(&spa_namespace_lock); 941 if (!locked) 942 return (SET_ERROR(EINTR)); 943 } 944 945 ro = (readonly || (strchr(zv->zv_name, '@') != NULL)); 946 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, ro, B_TRUE, zv, &os); 947 if (error) 948 goto out_mutex; 949 950 zv->zv_objset = os; 951 952 error = zvol_setup_zv(zv); 953 954 if (error) { 955 dmu_objset_disown(os, 1, zv); 956 zv->zv_objset = NULL; 957 } 958 959 out_mutex: 960 if (locked) 961 mutex_exit(&spa_namespace_lock); 962 return (SET_ERROR(error)); 963 } 964 965 void 966 zvol_last_close(zvol_state_t *zv) 967 { 968 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock)); 969 ASSERT(MUTEX_HELD(&zv->zv_state_lock)); 970 971 zvol_shutdown_zv(zv); 972 973 dmu_objset_disown(zv->zv_objset, 1, zv); 974 zv->zv_objset = NULL; 975 } 976 977 typedef struct minors_job { 978 list_t *list; 979 list_node_t link; 980 /* input */ 981 char *name; 982 /* output */ 983 int error; 984 } minors_job_t; 985 986 /* 987 * Prefetch zvol dnodes for the minors_job 988 */ 989 static void 990 zvol_prefetch_minors_impl(void *arg) 991 { 992 minors_job_t *job = arg; 993 char *dsname = job->name; 994 objset_t *os = NULL; 995 996 job->error = dmu_objset_own(dsname, DMU_OST_ZVOL, B_TRUE, B_TRUE, 997 FTAG, &os); 998 if (job->error == 0) { 999 dmu_prefetch(os, ZVOL_OBJ, 0, 0, 0, ZIO_PRIORITY_SYNC_READ); 1000 dmu_objset_disown(os, B_TRUE, FTAG); 1001 } 1002 } 1003 1004 /* 1005 * Mask errors to continue dmu_objset_find() traversal 1006 */ 1007 static int 1008 zvol_create_snap_minor_cb(const char *dsname, void *arg) 1009 { 1010 minors_job_t *j = arg; 1011 list_t *minors_list = j->list; 1012 const char *name = j->name; 1013 1014 ASSERT0(MUTEX_HELD(&spa_namespace_lock)); 1015 1016 /* skip the designated dataset */ 1017 if (name && strcmp(dsname, name) == 0) 1018 return (0); 1019 1020 /* at this point, the dsname should name a snapshot */ 1021 if (strchr(dsname, '@') == 0) { 1022 dprintf("zvol_create_snap_minor_cb(): " 1023 "%s is not a snapshot name\n", dsname); 1024 } else { 1025 minors_job_t *job; 1026 char *n = kmem_strdup(dsname); 1027 if (n == NULL) 1028 return (0); 1029 1030 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP); 1031 job->name = n; 1032 job->list = minors_list; 1033 job->error = 0; 1034 list_insert_tail(minors_list, job); 1035 /* don't care if dispatch fails, because job->error is 0 */ 1036 taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job, 1037 TQ_SLEEP); 1038 } 1039 1040 return (0); 1041 } 1042 1043 /* 1044 * If spa_keystore_load_wkey() is called for an encrypted zvol, 1045 * we need to look for any clones also using the key. This function 1046 * is "best effort" - so we just skip over it if there are failures. 1047 */ 1048 static void 1049 zvol_add_clones(const char *dsname, list_t *minors_list) 1050 { 1051 /* Also check if it has clones */ 1052 dsl_dir_t *dd = NULL; 1053 dsl_pool_t *dp = NULL; 1054 1055 if (dsl_pool_hold(dsname, FTAG, &dp) != 0) 1056 return; 1057 1058 if (!spa_feature_is_enabled(dp->dp_spa, 1059 SPA_FEATURE_ENCRYPTION)) 1060 goto out; 1061 1062 if (dsl_dir_hold(dp, dsname, FTAG, &dd, NULL) != 0) 1063 goto out; 1064 1065 if (dsl_dir_phys(dd)->dd_clones == 0) 1066 goto out; 1067 1068 zap_cursor_t *zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP); 1069 zap_attribute_t *za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 1070 objset_t *mos = dd->dd_pool->dp_meta_objset; 1071 1072 for (zap_cursor_init(zc, mos, dsl_dir_phys(dd)->dd_clones); 1073 zap_cursor_retrieve(zc, za) == 0; 1074 zap_cursor_advance(zc)) { 1075 dsl_dataset_t *clone; 1076 minors_job_t *job; 1077 1078 if (dsl_dataset_hold_obj(dd->dd_pool, 1079 za->za_first_integer, FTAG, &clone) == 0) { 1080 1081 char name[ZFS_MAX_DATASET_NAME_LEN]; 1082 dsl_dataset_name(clone, name); 1083 1084 char *n = kmem_strdup(name); 1085 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP); 1086 job->name = n; 1087 job->list = minors_list; 1088 job->error = 0; 1089 list_insert_tail(minors_list, job); 1090 1091 dsl_dataset_rele(clone, FTAG); 1092 } 1093 } 1094 zap_cursor_fini(zc); 1095 kmem_free(za, sizeof (zap_attribute_t)); 1096 kmem_free(zc, sizeof (zap_cursor_t)); 1097 1098 out: 1099 if (dd != NULL) 1100 dsl_dir_rele(dd, FTAG); 1101 if (dp != NULL) 1102 dsl_pool_rele(dp, FTAG); 1103 } 1104 1105 /* 1106 * Mask errors to continue dmu_objset_find() traversal 1107 */ 1108 static int 1109 zvol_create_minors_cb(const char *dsname, void *arg) 1110 { 1111 uint64_t snapdev; 1112 int error; 1113 list_t *minors_list = arg; 1114 1115 ASSERT0(MUTEX_HELD(&spa_namespace_lock)); 1116 1117 error = dsl_prop_get_integer(dsname, "snapdev", &snapdev, NULL); 1118 if (error) 1119 return (0); 1120 1121 /* 1122 * Given the name and the 'snapdev' property, create device minor nodes 1123 * with the linkages to zvols/snapshots as needed. 1124 * If the name represents a zvol, create a minor node for the zvol, then 1125 * check if its snapshots are 'visible', and if so, iterate over the 1126 * snapshots and create device minor nodes for those. 1127 */ 1128 if (strchr(dsname, '@') == 0) { 1129 minors_job_t *job; 1130 char *n = kmem_strdup(dsname); 1131 if (n == NULL) 1132 return (0); 1133 1134 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP); 1135 job->name = n; 1136 job->list = minors_list; 1137 job->error = 0; 1138 list_insert_tail(minors_list, job); 1139 /* don't care if dispatch fails, because job->error is 0 */ 1140 taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job, 1141 TQ_SLEEP); 1142 1143 zvol_add_clones(dsname, minors_list); 1144 1145 if (snapdev == ZFS_SNAPDEV_VISIBLE) { 1146 /* 1147 * traverse snapshots only, do not traverse children, 1148 * and skip the 'dsname' 1149 */ 1150 error = dmu_objset_find(dsname, 1151 zvol_create_snap_minor_cb, (void *)job, 1152 DS_FIND_SNAPSHOTS); 1153 } 1154 } else { 1155 dprintf("zvol_create_minors_cb(): %s is not a zvol name\n", 1156 dsname); 1157 } 1158 1159 return (0); 1160 } 1161 1162 /* 1163 * Create minors for the specified dataset, including children and snapshots. 1164 * Pay attention to the 'snapdev' property and iterate over the snapshots 1165 * only if they are 'visible'. This approach allows one to assure that the 1166 * snapshot metadata is read from disk only if it is needed. 1167 * 1168 * The name can represent a dataset to be recursively scanned for zvols and 1169 * their snapshots, or a single zvol snapshot. If the name represents a 1170 * dataset, the scan is performed in two nested stages: 1171 * - scan the dataset for zvols, and 1172 * - for each zvol, create a minor node, then check if the zvol's snapshots 1173 * are 'visible', and only then iterate over the snapshots if needed 1174 * 1175 * If the name represents a snapshot, a check is performed if the snapshot is 1176 * 'visible' (which also verifies that the parent is a zvol), and if so, 1177 * a minor node for that snapshot is created. 1178 */ 1179 void 1180 zvol_create_minors_recursive(const char *name) 1181 { 1182 list_t minors_list; 1183 minors_job_t *job; 1184 1185 if (zvol_inhibit_dev) 1186 return; 1187 1188 /* 1189 * This is the list for prefetch jobs. Whenever we found a match 1190 * during dmu_objset_find, we insert a minors_job to the list and do 1191 * taskq_dispatch to parallel prefetch zvol dnodes. Note we don't need 1192 * any lock because all list operation is done on the current thread. 1193 * 1194 * We will use this list to do zvol_create_minor_impl after prefetch 1195 * so we don't have to traverse using dmu_objset_find again. 1196 */ 1197 list_create(&minors_list, sizeof (minors_job_t), 1198 offsetof(minors_job_t, link)); 1199 1200 1201 if (strchr(name, '@') != NULL) { 1202 uint64_t snapdev; 1203 1204 int error = dsl_prop_get_integer(name, "snapdev", 1205 &snapdev, NULL); 1206 1207 if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE) 1208 (void) ops->zv_create_minor(name); 1209 } else { 1210 fstrans_cookie_t cookie = spl_fstrans_mark(); 1211 (void) dmu_objset_find(name, zvol_create_minors_cb, 1212 &minors_list, DS_FIND_CHILDREN); 1213 spl_fstrans_unmark(cookie); 1214 } 1215 1216 taskq_wait_outstanding(system_taskq, 0); 1217 1218 /* 1219 * Prefetch is completed, we can do zvol_create_minor_impl 1220 * sequentially. 1221 */ 1222 while ((job = list_head(&minors_list)) != NULL) { 1223 list_remove(&minors_list, job); 1224 if (!job->error) 1225 (void) ops->zv_create_minor(job->name); 1226 kmem_strfree(job->name); 1227 kmem_free(job, sizeof (minors_job_t)); 1228 } 1229 1230 list_destroy(&minors_list); 1231 } 1232 1233 void 1234 zvol_create_minor(const char *name) 1235 { 1236 /* 1237 * Note: the dsl_pool_config_lock must not be held. 1238 * Minor node creation needs to obtain the zvol_state_lock. 1239 * zvol_open() obtains the zvol_state_lock and then the dsl pool 1240 * config lock. Therefore, we can't have the config lock now if 1241 * we are going to wait for the zvol_state_lock, because it 1242 * would be a lock order inversion which could lead to deadlock. 1243 */ 1244 1245 if (zvol_inhibit_dev) 1246 return; 1247 1248 if (strchr(name, '@') != NULL) { 1249 uint64_t snapdev; 1250 1251 int error = dsl_prop_get_integer(name, 1252 "snapdev", &snapdev, NULL); 1253 1254 if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE) 1255 (void) ops->zv_create_minor(name); 1256 } else { 1257 (void) ops->zv_create_minor(name); 1258 } 1259 } 1260 1261 /* 1262 * Remove minors for specified dataset including children and snapshots. 1263 */ 1264 1265 static void 1266 zvol_free_task(void *arg) 1267 { 1268 ops->zv_free(arg); 1269 } 1270 1271 void 1272 zvol_remove_minors_impl(const char *name) 1273 { 1274 zvol_state_t *zv, *zv_next; 1275 int namelen = ((name) ? strlen(name) : 0); 1276 taskqid_t t; 1277 list_t free_list; 1278 1279 if (zvol_inhibit_dev) 1280 return; 1281 1282 list_create(&free_list, sizeof (zvol_state_t), 1283 offsetof(zvol_state_t, zv_next)); 1284 1285 rw_enter(&zvol_state_lock, RW_WRITER); 1286 1287 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { 1288 zv_next = list_next(&zvol_state_list, zv); 1289 1290 mutex_enter(&zv->zv_state_lock); 1291 if (name == NULL || strcmp(zv->zv_name, name) == 0 || 1292 (strncmp(zv->zv_name, name, namelen) == 0 && 1293 (zv->zv_name[namelen] == '/' || 1294 zv->zv_name[namelen] == '@'))) { 1295 /* 1296 * By holding zv_state_lock here, we guarantee that no 1297 * one is currently using this zv 1298 */ 1299 1300 /* If in use, leave alone */ 1301 if (zv->zv_open_count > 0 || 1302 atomic_read(&zv->zv_suspend_ref)) { 1303 mutex_exit(&zv->zv_state_lock); 1304 continue; 1305 } 1306 1307 zvol_remove(zv); 1308 1309 /* 1310 * Cleared while holding zvol_state_lock as a writer 1311 * which will prevent zvol_open() from opening it. 1312 */ 1313 ops->zv_clear_private(zv); 1314 1315 /* Drop zv_state_lock before zvol_free() */ 1316 mutex_exit(&zv->zv_state_lock); 1317 1318 /* Try parallel zv_free, if failed do it in place */ 1319 t = taskq_dispatch(system_taskq, zvol_free_task, zv, 1320 TQ_SLEEP); 1321 if (t == TASKQID_INVALID) 1322 list_insert_head(&free_list, zv); 1323 } else { 1324 mutex_exit(&zv->zv_state_lock); 1325 } 1326 } 1327 rw_exit(&zvol_state_lock); 1328 1329 /* Drop zvol_state_lock before calling zvol_free() */ 1330 while ((zv = list_head(&free_list)) != NULL) { 1331 list_remove(&free_list, zv); 1332 ops->zv_free(zv); 1333 } 1334 } 1335 1336 /* Remove minor for this specific volume only */ 1337 static void 1338 zvol_remove_minor_impl(const char *name) 1339 { 1340 zvol_state_t *zv = NULL, *zv_next; 1341 1342 if (zvol_inhibit_dev) 1343 return; 1344 1345 rw_enter(&zvol_state_lock, RW_WRITER); 1346 1347 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { 1348 zv_next = list_next(&zvol_state_list, zv); 1349 1350 mutex_enter(&zv->zv_state_lock); 1351 if (strcmp(zv->zv_name, name) == 0) { 1352 /* 1353 * By holding zv_state_lock here, we guarantee that no 1354 * one is currently using this zv 1355 */ 1356 1357 /* If in use, leave alone */ 1358 if (zv->zv_open_count > 0 || 1359 atomic_read(&zv->zv_suspend_ref)) { 1360 mutex_exit(&zv->zv_state_lock); 1361 continue; 1362 } 1363 zvol_remove(zv); 1364 1365 ops->zv_clear_private(zv); 1366 mutex_exit(&zv->zv_state_lock); 1367 break; 1368 } else { 1369 mutex_exit(&zv->zv_state_lock); 1370 } 1371 } 1372 1373 /* Drop zvol_state_lock before calling zvol_free() */ 1374 rw_exit(&zvol_state_lock); 1375 1376 if (zv != NULL) 1377 ops->zv_free(zv); 1378 } 1379 1380 /* 1381 * Rename minors for specified dataset including children and snapshots. 1382 */ 1383 static void 1384 zvol_rename_minors_impl(const char *oldname, const char *newname) 1385 { 1386 zvol_state_t *zv, *zv_next; 1387 int oldnamelen, newnamelen; 1388 1389 if (zvol_inhibit_dev) 1390 return; 1391 1392 oldnamelen = strlen(oldname); 1393 newnamelen = strlen(newname); 1394 1395 rw_enter(&zvol_state_lock, RW_READER); 1396 1397 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { 1398 zv_next = list_next(&zvol_state_list, zv); 1399 1400 mutex_enter(&zv->zv_state_lock); 1401 1402 if (strcmp(zv->zv_name, oldname) == 0) { 1403 ops->zv_rename_minor(zv, newname); 1404 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && 1405 (zv->zv_name[oldnamelen] == '/' || 1406 zv->zv_name[oldnamelen] == '@')) { 1407 char *name = kmem_asprintf("%s%c%s", newname, 1408 zv->zv_name[oldnamelen], 1409 zv->zv_name + oldnamelen + 1); 1410 ops->zv_rename_minor(zv, name); 1411 kmem_strfree(name); 1412 } 1413 1414 mutex_exit(&zv->zv_state_lock); 1415 } 1416 1417 rw_exit(&zvol_state_lock); 1418 } 1419 1420 typedef struct zvol_snapdev_cb_arg { 1421 uint64_t snapdev; 1422 } zvol_snapdev_cb_arg_t; 1423 1424 static int 1425 zvol_set_snapdev_cb(const char *dsname, void *param) 1426 { 1427 zvol_snapdev_cb_arg_t *arg = param; 1428 1429 if (strchr(dsname, '@') == NULL) 1430 return (0); 1431 1432 switch (arg->snapdev) { 1433 case ZFS_SNAPDEV_VISIBLE: 1434 (void) ops->zv_create_minor(dsname); 1435 break; 1436 case ZFS_SNAPDEV_HIDDEN: 1437 (void) zvol_remove_minor_impl(dsname); 1438 break; 1439 } 1440 1441 return (0); 1442 } 1443 1444 static void 1445 zvol_set_snapdev_impl(char *name, uint64_t snapdev) 1446 { 1447 zvol_snapdev_cb_arg_t arg = {snapdev}; 1448 fstrans_cookie_t cookie = spl_fstrans_mark(); 1449 /* 1450 * The zvol_set_snapdev_sync() sets snapdev appropriately 1451 * in the dataset hierarchy. Here, we only scan snapshots. 1452 */ 1453 dmu_objset_find(name, zvol_set_snapdev_cb, &arg, DS_FIND_SNAPSHOTS); 1454 spl_fstrans_unmark(cookie); 1455 } 1456 1457 static void 1458 zvol_set_volmode_impl(char *name, uint64_t volmode) 1459 { 1460 fstrans_cookie_t cookie; 1461 uint64_t old_volmode; 1462 zvol_state_t *zv; 1463 1464 if (strchr(name, '@') != NULL) 1465 return; 1466 1467 /* 1468 * It's unfortunate we need to remove minors before we create new ones: 1469 * this is necessary because our backing gendisk (zvol_state->zv_disk) 1470 * could be different when we set, for instance, volmode from "geom" 1471 * to "dev" (or vice versa). 1472 */ 1473 zv = zvol_find_by_name(name, RW_NONE); 1474 if (zv == NULL && volmode == ZFS_VOLMODE_NONE) 1475 return; 1476 if (zv != NULL) { 1477 old_volmode = zv->zv_volmode; 1478 mutex_exit(&zv->zv_state_lock); 1479 if (old_volmode == volmode) 1480 return; 1481 zvol_wait_close(zv); 1482 } 1483 cookie = spl_fstrans_mark(); 1484 switch (volmode) { 1485 case ZFS_VOLMODE_NONE: 1486 (void) zvol_remove_minor_impl(name); 1487 break; 1488 case ZFS_VOLMODE_GEOM: 1489 case ZFS_VOLMODE_DEV: 1490 (void) zvol_remove_minor_impl(name); 1491 (void) ops->zv_create_minor(name); 1492 break; 1493 case ZFS_VOLMODE_DEFAULT: 1494 (void) zvol_remove_minor_impl(name); 1495 if (zvol_volmode == ZFS_VOLMODE_NONE) 1496 break; 1497 else /* if zvol_volmode is invalid defaults to "geom" */ 1498 (void) ops->zv_create_minor(name); 1499 break; 1500 } 1501 spl_fstrans_unmark(cookie); 1502 } 1503 1504 static zvol_task_t * 1505 zvol_task_alloc(zvol_async_op_t op, const char *name1, const char *name2, 1506 uint64_t value) 1507 { 1508 zvol_task_t *task; 1509 1510 /* Never allow tasks on hidden names. */ 1511 if (name1[0] == '$') 1512 return (NULL); 1513 1514 task = kmem_zalloc(sizeof (zvol_task_t), KM_SLEEP); 1515 task->op = op; 1516 task->value = value; 1517 1518 strlcpy(task->name1, name1, MAXNAMELEN); 1519 if (name2 != NULL) 1520 strlcpy(task->name2, name2, MAXNAMELEN); 1521 1522 return (task); 1523 } 1524 1525 static void 1526 zvol_task_free(zvol_task_t *task) 1527 { 1528 kmem_free(task, sizeof (zvol_task_t)); 1529 } 1530 1531 /* 1532 * The worker thread function performed asynchronously. 1533 */ 1534 static void 1535 zvol_task_cb(void *arg) 1536 { 1537 zvol_task_t *task = arg; 1538 1539 switch (task->op) { 1540 case ZVOL_ASYNC_REMOVE_MINORS: 1541 zvol_remove_minors_impl(task->name1); 1542 break; 1543 case ZVOL_ASYNC_RENAME_MINORS: 1544 zvol_rename_minors_impl(task->name1, task->name2); 1545 break; 1546 case ZVOL_ASYNC_SET_SNAPDEV: 1547 zvol_set_snapdev_impl(task->name1, task->value); 1548 break; 1549 case ZVOL_ASYNC_SET_VOLMODE: 1550 zvol_set_volmode_impl(task->name1, task->value); 1551 break; 1552 default: 1553 VERIFY(0); 1554 break; 1555 } 1556 1557 zvol_task_free(task); 1558 } 1559 1560 typedef struct zvol_set_prop_int_arg { 1561 const char *zsda_name; 1562 uint64_t zsda_value; 1563 zprop_source_t zsda_source; 1564 dmu_tx_t *zsda_tx; 1565 } zvol_set_prop_int_arg_t; 1566 1567 /* 1568 * Sanity check the dataset for safe use by the sync task. No additional 1569 * conditions are imposed. 1570 */ 1571 static int 1572 zvol_set_snapdev_check(void *arg, dmu_tx_t *tx) 1573 { 1574 zvol_set_prop_int_arg_t *zsda = arg; 1575 dsl_pool_t *dp = dmu_tx_pool(tx); 1576 dsl_dir_t *dd; 1577 int error; 1578 1579 error = dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL); 1580 if (error != 0) 1581 return (error); 1582 1583 dsl_dir_rele(dd, FTAG); 1584 1585 return (error); 1586 } 1587 1588 /* ARGSUSED */ 1589 static int 1590 zvol_set_snapdev_sync_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 1591 { 1592 char dsname[MAXNAMELEN]; 1593 zvol_task_t *task; 1594 uint64_t snapdev; 1595 1596 dsl_dataset_name(ds, dsname); 1597 if (dsl_prop_get_int_ds(ds, "snapdev", &snapdev) != 0) 1598 return (0); 1599 task = zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV, dsname, NULL, snapdev); 1600 if (task == NULL) 1601 return (0); 1602 1603 (void) taskq_dispatch(dp->dp_spa->spa_zvol_taskq, zvol_task_cb, 1604 task, TQ_SLEEP); 1605 return (0); 1606 } 1607 1608 /* 1609 * Traverse all child datasets and apply snapdev appropriately. 1610 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel 1611 * dataset and read the effective "snapdev" on every child in the callback 1612 * function: this is because the value is not guaranteed to be the same in the 1613 * whole dataset hierarchy. 1614 */ 1615 static void 1616 zvol_set_snapdev_sync(void *arg, dmu_tx_t *tx) 1617 { 1618 zvol_set_prop_int_arg_t *zsda = arg; 1619 dsl_pool_t *dp = dmu_tx_pool(tx); 1620 dsl_dir_t *dd; 1621 dsl_dataset_t *ds; 1622 int error; 1623 1624 VERIFY0(dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL)); 1625 zsda->zsda_tx = tx; 1626 1627 error = dsl_dataset_hold(dp, zsda->zsda_name, FTAG, &ds); 1628 if (error == 0) { 1629 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_SNAPDEV), 1630 zsda->zsda_source, sizeof (zsda->zsda_value), 1, 1631 &zsda->zsda_value, zsda->zsda_tx); 1632 dsl_dataset_rele(ds, FTAG); 1633 } 1634 dmu_objset_find_dp(dp, dd->dd_object, zvol_set_snapdev_sync_cb, 1635 zsda, DS_FIND_CHILDREN); 1636 1637 dsl_dir_rele(dd, FTAG); 1638 } 1639 1640 int 1641 zvol_set_snapdev(const char *ddname, zprop_source_t source, uint64_t snapdev) 1642 { 1643 zvol_set_prop_int_arg_t zsda; 1644 1645 zsda.zsda_name = ddname; 1646 zsda.zsda_source = source; 1647 zsda.zsda_value = snapdev; 1648 1649 return (dsl_sync_task(ddname, zvol_set_snapdev_check, 1650 zvol_set_snapdev_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE)); 1651 } 1652 1653 /* 1654 * Sanity check the dataset for safe use by the sync task. No additional 1655 * conditions are imposed. 1656 */ 1657 static int 1658 zvol_set_volmode_check(void *arg, dmu_tx_t *tx) 1659 { 1660 zvol_set_prop_int_arg_t *zsda = arg; 1661 dsl_pool_t *dp = dmu_tx_pool(tx); 1662 dsl_dir_t *dd; 1663 int error; 1664 1665 error = dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL); 1666 if (error != 0) 1667 return (error); 1668 1669 dsl_dir_rele(dd, FTAG); 1670 1671 return (error); 1672 } 1673 1674 /* ARGSUSED */ 1675 static int 1676 zvol_set_volmode_sync_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 1677 { 1678 char dsname[MAXNAMELEN]; 1679 zvol_task_t *task; 1680 uint64_t volmode; 1681 1682 dsl_dataset_name(ds, dsname); 1683 if (dsl_prop_get_int_ds(ds, "volmode", &volmode) != 0) 1684 return (0); 1685 task = zvol_task_alloc(ZVOL_ASYNC_SET_VOLMODE, dsname, NULL, volmode); 1686 if (task == NULL) 1687 return (0); 1688 1689 (void) taskq_dispatch(dp->dp_spa->spa_zvol_taskq, zvol_task_cb, 1690 task, TQ_SLEEP); 1691 return (0); 1692 } 1693 1694 /* 1695 * Traverse all child datasets and apply volmode appropriately. 1696 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel 1697 * dataset and read the effective "volmode" on every child in the callback 1698 * function: this is because the value is not guaranteed to be the same in the 1699 * whole dataset hierarchy. 1700 */ 1701 static void 1702 zvol_set_volmode_sync(void *arg, dmu_tx_t *tx) 1703 { 1704 zvol_set_prop_int_arg_t *zsda = arg; 1705 dsl_pool_t *dp = dmu_tx_pool(tx); 1706 dsl_dir_t *dd; 1707 dsl_dataset_t *ds; 1708 int error; 1709 1710 VERIFY0(dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL)); 1711 zsda->zsda_tx = tx; 1712 1713 error = dsl_dataset_hold(dp, zsda->zsda_name, FTAG, &ds); 1714 if (error == 0) { 1715 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_VOLMODE), 1716 zsda->zsda_source, sizeof (zsda->zsda_value), 1, 1717 &zsda->zsda_value, zsda->zsda_tx); 1718 dsl_dataset_rele(ds, FTAG); 1719 } 1720 1721 dmu_objset_find_dp(dp, dd->dd_object, zvol_set_volmode_sync_cb, 1722 zsda, DS_FIND_CHILDREN); 1723 1724 dsl_dir_rele(dd, FTAG); 1725 } 1726 1727 int 1728 zvol_set_volmode(const char *ddname, zprop_source_t source, uint64_t volmode) 1729 { 1730 zvol_set_prop_int_arg_t zsda; 1731 1732 zsda.zsda_name = ddname; 1733 zsda.zsda_source = source; 1734 zsda.zsda_value = volmode; 1735 1736 return (dsl_sync_task(ddname, zvol_set_volmode_check, 1737 zvol_set_volmode_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE)); 1738 } 1739 1740 void 1741 zvol_remove_minors(spa_t *spa, const char *name, boolean_t async) 1742 { 1743 zvol_task_t *task; 1744 taskqid_t id; 1745 1746 task = zvol_task_alloc(ZVOL_ASYNC_REMOVE_MINORS, name, NULL, ~0ULL); 1747 if (task == NULL) 1748 return; 1749 1750 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP); 1751 if ((async == B_FALSE) && (id != TASKQID_INVALID)) 1752 taskq_wait_id(spa->spa_zvol_taskq, id); 1753 } 1754 1755 void 1756 zvol_rename_minors(spa_t *spa, const char *name1, const char *name2, 1757 boolean_t async) 1758 { 1759 zvol_task_t *task; 1760 taskqid_t id; 1761 1762 task = zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS, name1, name2, ~0ULL); 1763 if (task == NULL) 1764 return; 1765 1766 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP); 1767 if ((async == B_FALSE) && (id != TASKQID_INVALID)) 1768 taskq_wait_id(spa->spa_zvol_taskq, id); 1769 } 1770 1771 boolean_t 1772 zvol_is_zvol(const char *name) 1773 { 1774 1775 return (ops->zv_is_zvol(name)); 1776 } 1777 1778 void 1779 zvol_register_ops(const zvol_platform_ops_t *zvol_ops) 1780 { 1781 ops = zvol_ops; 1782 } 1783 1784 int 1785 zvol_init_impl(void) 1786 { 1787 int i; 1788 1789 list_create(&zvol_state_list, sizeof (zvol_state_t), 1790 offsetof(zvol_state_t, zv_next)); 1791 rw_init(&zvol_state_lock, NULL, RW_DEFAULT, NULL); 1792 1793 zvol_htable = kmem_alloc(ZVOL_HT_SIZE * sizeof (struct hlist_head), 1794 KM_SLEEP); 1795 for (i = 0; i < ZVOL_HT_SIZE; i++) 1796 INIT_HLIST_HEAD(&zvol_htable[i]); 1797 1798 return (0); 1799 } 1800 1801 void 1802 zvol_fini_impl(void) 1803 { 1804 zvol_remove_minors_impl(NULL); 1805 1806 /* 1807 * The call to "zvol_remove_minors_impl" may dispatch entries to 1808 * the system_taskq, but it doesn't wait for those entries to 1809 * complete before it returns. Thus, we must wait for all of the 1810 * removals to finish, before we can continue. 1811 */ 1812 taskq_wait_outstanding(system_taskq, 0); 1813 1814 kmem_free(zvol_htable, ZVOL_HT_SIZE * sizeof (struct hlist_head)); 1815 list_destroy(&zvol_state_list); 1816 rw_destroy(&zvol_state_lock); 1817 } 1818