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