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 /* 23 * Copyright (c) 2012, 2017 by Delphix. All rights reserved. 24 * Copyright (c) 2013 Steven Hartland. All rights reserved. 25 * Copyright (c) 2014 Integros [integros.com] 26 * Copyright 2017 RackTop Systems. 27 * Copyright (c) 2017 Datto Inc. 28 */ 29 30 /* 31 * LibZFS_Core (lzc) is intended to replace most functionality in libzfs. 32 * It has the following characteristics: 33 * 34 * - Thread Safe. libzfs_core is accessible concurrently from multiple 35 * threads. This is accomplished primarily by avoiding global data 36 * (e.g. caching). Since it's thread-safe, there is no reason for a 37 * process to have multiple libzfs "instances". Therefore, we store 38 * our few pieces of data (e.g. the file descriptor) in global 39 * variables. The fd is reference-counted so that the libzfs_core 40 * library can be "initialized" multiple times (e.g. by different 41 * consumers within the same process). 42 * 43 * - Committed Interface. The libzfs_core interface will be committed, 44 * therefore consumers can compile against it and be confident that 45 * their code will continue to work on future releases of this code. 46 * Currently, the interface is Evolving (not Committed), but we intend 47 * to commit to it once it is more complete and we determine that it 48 * meets the needs of all consumers. 49 * 50 * - Programatic Error Handling. libzfs_core communicates errors with 51 * defined error numbers, and doesn't print anything to stdout/stderr. 52 * 53 * - Thin Layer. libzfs_core is a thin layer, marshaling arguments 54 * to/from the kernel ioctls. There is generally a 1:1 correspondence 55 * between libzfs_core functions and ioctls to /dev/zfs. 56 * 57 * - Clear Atomicity. Because libzfs_core functions are generally 1:1 58 * with kernel ioctls, and kernel ioctls are general atomic, each 59 * libzfs_core function is atomic. For example, creating multiple 60 * snapshots with a single call to lzc_snapshot() is atomic -- it 61 * can't fail with only some of the requested snapshots created, even 62 * in the event of power loss or system crash. 63 * 64 * - Continued libzfs Support. Some higher-level operations (e.g. 65 * support for "zfs send -R") are too complicated to fit the scope of 66 * libzfs_core. This functionality will continue to live in libzfs. 67 * Where appropriate, libzfs will use the underlying atomic operations 68 * of libzfs_core. For example, libzfs may implement "zfs send -R | 69 * zfs receive" by using individual "send one snapshot", rename, 70 * destroy, and "receive one snapshot" operations in libzfs_core. 71 * /sbin/zfs and /zbin/zpool will link with both libzfs and 72 * libzfs_core. Other consumers should aim to use only libzfs_core, 73 * since that will be the supported, stable interface going forwards. 74 */ 75 76 #include <libzfs_core.h> 77 #include <ctype.h> 78 #include <unistd.h> 79 #include <stdlib.h> 80 #include <string.h> 81 #include <errno.h> 82 #include <fcntl.h> 83 #include <pthread.h> 84 #include <sys/nvpair.h> 85 #include <sys/param.h> 86 #include <sys/types.h> 87 #include <sys/stat.h> 88 #include <sys/zfs_ioctl.h> 89 90 static int g_fd = -1; 91 static pthread_mutex_t g_lock = PTHREAD_MUTEX_INITIALIZER; 92 static int g_refcount; 93 94 int 95 libzfs_core_init(void) 96 { 97 (void) pthread_mutex_lock(&g_lock); 98 if (g_refcount == 0) { 99 g_fd = open("/dev/zfs", O_RDWR); 100 if (g_fd < 0) { 101 (void) pthread_mutex_unlock(&g_lock); 102 return (errno); 103 } 104 } 105 g_refcount++; 106 (void) pthread_mutex_unlock(&g_lock); 107 return (0); 108 } 109 110 void 111 libzfs_core_fini(void) 112 { 113 (void) pthread_mutex_lock(&g_lock); 114 ASSERT3S(g_refcount, >, 0); 115 116 if (g_refcount > 0) 117 g_refcount--; 118 119 if (g_refcount == 0 && g_fd != -1) { 120 (void) close(g_fd); 121 g_fd = -1; 122 } 123 (void) pthread_mutex_unlock(&g_lock); 124 } 125 126 static int 127 lzc_ioctl(zfs_ioc_t ioc, const char *name, 128 nvlist_t *source, nvlist_t **resultp) 129 { 130 zfs_cmd_t zc = { 0 }; 131 int error = 0; 132 char *packed = NULL; 133 size_t size = 0; 134 135 ASSERT3S(g_refcount, >, 0); 136 VERIFY3S(g_fd, !=, -1); 137 138 if (name != NULL) 139 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name)); 140 141 if (source != NULL) { 142 packed = fnvlist_pack(source, &size); 143 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed; 144 zc.zc_nvlist_src_size = size; 145 } 146 147 if (resultp != NULL) { 148 *resultp = NULL; 149 if (ioc == ZFS_IOC_CHANNEL_PROGRAM) { 150 zc.zc_nvlist_dst_size = fnvlist_lookup_uint64(source, 151 ZCP_ARG_MEMLIMIT); 152 } else { 153 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024); 154 } 155 zc.zc_nvlist_dst = (uint64_t)(uintptr_t) 156 malloc(zc.zc_nvlist_dst_size); 157 if (zc.zc_nvlist_dst == NULL) { 158 error = ENOMEM; 159 goto out; 160 } 161 } 162 163 while (ioctl(g_fd, ioc, &zc) != 0) { 164 /* 165 * If ioctl exited with ENOMEM, we retry the ioctl after 166 * increasing the size of the destination nvlist. 167 * 168 * Channel programs that exit with ENOMEM ran over the 169 * lua memory sandbox; they should not be retried. 170 */ 171 if (errno == ENOMEM && resultp != NULL && 172 ioc != ZFS_IOC_CHANNEL_PROGRAM) { 173 free((void *)(uintptr_t)zc.zc_nvlist_dst); 174 zc.zc_nvlist_dst_size *= 2; 175 zc.zc_nvlist_dst = (uint64_t)(uintptr_t) 176 malloc(zc.zc_nvlist_dst_size); 177 if (zc.zc_nvlist_dst == NULL) { 178 error = ENOMEM; 179 goto out; 180 } 181 } else { 182 error = errno; 183 break; 184 } 185 } 186 if (zc.zc_nvlist_dst_filled) { 187 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst, 188 zc.zc_nvlist_dst_size); 189 } 190 191 out: 192 fnvlist_pack_free(packed, size); 193 free((void *)(uintptr_t)zc.zc_nvlist_dst); 194 return (error); 195 } 196 197 int 198 lzc_create(const char *fsname, enum lzc_dataset_type type, nvlist_t *props) 199 { 200 int error; 201 nvlist_t *args = fnvlist_alloc(); 202 fnvlist_add_int32(args, "type", (dmu_objset_type_t)type); 203 if (props != NULL) 204 fnvlist_add_nvlist(args, "props", props); 205 error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL); 206 nvlist_free(args); 207 return (error); 208 } 209 210 int 211 lzc_clone(const char *fsname, const char *origin, 212 nvlist_t *props) 213 { 214 int error; 215 nvlist_t *args = fnvlist_alloc(); 216 fnvlist_add_string(args, "origin", origin); 217 if (props != NULL) 218 fnvlist_add_nvlist(args, "props", props); 219 error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL); 220 nvlist_free(args); 221 return (error); 222 } 223 224 int 225 lzc_promote(const char *fsname, char *snapnamebuf, int snapnamelen) 226 { 227 /* 228 * The promote ioctl is still legacy, so we need to construct our 229 * own zfs_cmd_t rather than using lzc_ioctl(). 230 */ 231 zfs_cmd_t zc = { 0 }; 232 233 ASSERT3S(g_refcount, >, 0); 234 VERIFY3S(g_fd, !=, -1); 235 236 (void) strlcpy(zc.zc_name, fsname, sizeof (zc.zc_name)); 237 if (ioctl(g_fd, ZFS_IOC_PROMOTE, &zc) != 0) { 238 int error = errno; 239 if (error == EEXIST && snapnamebuf != NULL) 240 (void) strlcpy(snapnamebuf, zc.zc_string, snapnamelen); 241 return (error); 242 } 243 return (0); 244 } 245 246 int 247 lzc_remap(const char *fsname) 248 { 249 int error; 250 nvlist_t *args = fnvlist_alloc(); 251 error = lzc_ioctl(ZFS_IOC_REMAP, fsname, args, NULL); 252 nvlist_free(args); 253 return (error); 254 } 255 256 int 257 lzc_rename(const char *source, const char *target) 258 { 259 zfs_cmd_t zc = { 0 }; 260 int error; 261 262 ASSERT3S(g_refcount, >, 0); 263 VERIFY3S(g_fd, !=, -1); 264 265 (void) strlcpy(zc.zc_name, source, sizeof (zc.zc_name)); 266 (void) strlcpy(zc.zc_value, target, sizeof (zc.zc_value)); 267 error = ioctl(g_fd, ZFS_IOC_RENAME, &zc); 268 if (error != 0) 269 error = errno; 270 return (error); 271 } 272 273 int 274 lzc_destroy(const char *fsname) 275 { 276 int error; 277 278 nvlist_t *args = fnvlist_alloc(); 279 error = lzc_ioctl(ZFS_IOC_DESTROY, fsname, args, NULL); 280 nvlist_free(args); 281 return (error); 282 } 283 284 /* 285 * Creates snapshots. 286 * 287 * The keys in the snaps nvlist are the snapshots to be created. 288 * They must all be in the same pool. 289 * 290 * The props nvlist is properties to set. Currently only user properties 291 * are supported. { user:prop_name -> string value } 292 * 293 * The returned results nvlist will have an entry for each snapshot that failed. 294 * The value will be the (int32) error code. 295 * 296 * The return value will be 0 if all snapshots were created, otherwise it will 297 * be the errno of a (unspecified) snapshot that failed. 298 */ 299 int 300 lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist) 301 { 302 nvpair_t *elem; 303 nvlist_t *args; 304 int error; 305 char pool[ZFS_MAX_DATASET_NAME_LEN]; 306 307 *errlist = NULL; 308 309 /* determine the pool name */ 310 elem = nvlist_next_nvpair(snaps, NULL); 311 if (elem == NULL) 312 return (0); 313 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 314 pool[strcspn(pool, "/@")] = '\0'; 315 316 args = fnvlist_alloc(); 317 fnvlist_add_nvlist(args, "snaps", snaps); 318 if (props != NULL) 319 fnvlist_add_nvlist(args, "props", props); 320 321 error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist); 322 nvlist_free(args); 323 324 return (error); 325 } 326 327 /* 328 * Destroys snapshots. 329 * 330 * The keys in the snaps nvlist are the snapshots to be destroyed. 331 * They must all be in the same pool. 332 * 333 * Snapshots that do not exist will be silently ignored. 334 * 335 * If 'defer' is not set, and a snapshot has user holds or clones, the 336 * destroy operation will fail and none of the snapshots will be 337 * destroyed. 338 * 339 * If 'defer' is set, and a snapshot has user holds or clones, it will be 340 * marked for deferred destruction, and will be destroyed when the last hold 341 * or clone is removed/destroyed. 342 * 343 * The return value will be 0 if all snapshots were destroyed (or marked for 344 * later destruction if 'defer' is set) or didn't exist to begin with. 345 * 346 * Otherwise the return value will be the errno of a (unspecified) snapshot 347 * that failed, no snapshots will be destroyed, and the errlist will have an 348 * entry for each snapshot that failed. The value in the errlist will be 349 * the (int32) error code. 350 */ 351 int 352 lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist) 353 { 354 nvpair_t *elem; 355 nvlist_t *args; 356 int error; 357 char pool[ZFS_MAX_DATASET_NAME_LEN]; 358 359 /* determine the pool name */ 360 elem = nvlist_next_nvpair(snaps, NULL); 361 if (elem == NULL) 362 return (0); 363 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 364 pool[strcspn(pool, "/@")] = '\0'; 365 366 args = fnvlist_alloc(); 367 fnvlist_add_nvlist(args, "snaps", snaps); 368 if (defer) 369 fnvlist_add_boolean(args, "defer"); 370 371 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist); 372 nvlist_free(args); 373 374 return (error); 375 } 376 377 int 378 lzc_snaprange_space(const char *firstsnap, const char *lastsnap, 379 uint64_t *usedp) 380 { 381 nvlist_t *args; 382 nvlist_t *result; 383 int err; 384 char fs[ZFS_MAX_DATASET_NAME_LEN]; 385 char *atp; 386 387 /* determine the fs name */ 388 (void) strlcpy(fs, firstsnap, sizeof (fs)); 389 atp = strchr(fs, '@'); 390 if (atp == NULL) 391 return (EINVAL); 392 *atp = '\0'; 393 394 args = fnvlist_alloc(); 395 fnvlist_add_string(args, "firstsnap", firstsnap); 396 397 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result); 398 nvlist_free(args); 399 if (err == 0) 400 *usedp = fnvlist_lookup_uint64(result, "used"); 401 fnvlist_free(result); 402 403 return (err); 404 } 405 406 boolean_t 407 lzc_exists(const char *dataset) 408 { 409 /* 410 * The objset_stats ioctl is still legacy, so we need to construct our 411 * own zfs_cmd_t rather than using lzc_ioctl(). 412 */ 413 zfs_cmd_t zc = { 0 }; 414 415 ASSERT3S(g_refcount, >, 0); 416 VERIFY3S(g_fd, !=, -1); 417 418 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name)); 419 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0); 420 } 421 422 /* 423 * outnvl is unused. 424 * It was added to preserve the function signature in case it is 425 * needed in the future. 426 */ 427 /*ARGSUSED*/ 428 int 429 lzc_sync(const char *pool_name, nvlist_t *innvl, nvlist_t **outnvl) 430 { 431 return (lzc_ioctl(ZFS_IOC_POOL_SYNC, pool_name, innvl, NULL)); 432 } 433 434 /* 435 * Create "user holds" on snapshots. If there is a hold on a snapshot, 436 * the snapshot can not be destroyed. (However, it can be marked for deletion 437 * by lzc_destroy_snaps(defer=B_TRUE).) 438 * 439 * The keys in the nvlist are snapshot names. 440 * The snapshots must all be in the same pool. 441 * The value is the name of the hold (string type). 442 * 443 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL). 444 * In this case, when the cleanup_fd is closed (including on process 445 * termination), the holds will be released. If the system is shut down 446 * uncleanly, the holds will be released when the pool is next opened 447 * or imported. 448 * 449 * Holds for snapshots which don't exist will be skipped and have an entry 450 * added to errlist, but will not cause an overall failure. 451 * 452 * The return value will be 0 if all holds, for snapshots that existed, 453 * were succesfully created. 454 * 455 * Otherwise the return value will be the errno of a (unspecified) hold that 456 * failed and no holds will be created. 457 * 458 * In all cases the errlist will have an entry for each hold that failed 459 * (name = snapshot), with its value being the error code (int32). 460 */ 461 int 462 lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist) 463 { 464 char pool[ZFS_MAX_DATASET_NAME_LEN]; 465 nvlist_t *args; 466 nvpair_t *elem; 467 int error; 468 469 /* determine the pool name */ 470 elem = nvlist_next_nvpair(holds, NULL); 471 if (elem == NULL) 472 return (0); 473 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 474 pool[strcspn(pool, "/@")] = '\0'; 475 476 args = fnvlist_alloc(); 477 fnvlist_add_nvlist(args, "holds", holds); 478 if (cleanup_fd != -1) 479 fnvlist_add_int32(args, "cleanup_fd", cleanup_fd); 480 481 error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist); 482 nvlist_free(args); 483 return (error); 484 } 485 486 /* 487 * Release "user holds" on snapshots. If the snapshot has been marked for 488 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have 489 * any clones, and all the user holds are removed, then the snapshot will be 490 * destroyed. 491 * 492 * The keys in the nvlist are snapshot names. 493 * The snapshots must all be in the same pool. 494 * The value is a nvlist whose keys are the holds to remove. 495 * 496 * Holds which failed to release because they didn't exist will have an entry 497 * added to errlist, but will not cause an overall failure. 498 * 499 * The return value will be 0 if the nvl holds was empty or all holds that 500 * existed, were successfully removed. 501 * 502 * Otherwise the return value will be the errno of a (unspecified) hold that 503 * failed to release and no holds will be released. 504 * 505 * In all cases the errlist will have an entry for each hold that failed to 506 * to release. 507 */ 508 int 509 lzc_release(nvlist_t *holds, nvlist_t **errlist) 510 { 511 char pool[ZFS_MAX_DATASET_NAME_LEN]; 512 nvpair_t *elem; 513 514 /* determine the pool name */ 515 elem = nvlist_next_nvpair(holds, NULL); 516 if (elem == NULL) 517 return (0); 518 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 519 pool[strcspn(pool, "/@")] = '\0'; 520 521 return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist)); 522 } 523 524 /* 525 * Retrieve list of user holds on the specified snapshot. 526 * 527 * On success, *holdsp will be set to a nvlist which the caller must free. 528 * The keys are the names of the holds, and the value is the creation time 529 * of the hold (uint64) in seconds since the epoch. 530 */ 531 int 532 lzc_get_holds(const char *snapname, nvlist_t **holdsp) 533 { 534 return (lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, NULL, holdsp)); 535 } 536 537 /* 538 * Generate a zfs send stream for the specified snapshot and write it to 539 * the specified file descriptor. 540 * 541 * "snapname" is the full name of the snapshot to send (e.g. "pool/fs@snap") 542 * 543 * If "from" is NULL, a full (non-incremental) stream will be sent. 544 * If "from" is non-NULL, it must be the full name of a snapshot or 545 * bookmark to send an incremental from (e.g. "pool/fs@earlier_snap" or 546 * "pool/fs#earlier_bmark"). If non-NULL, the specified snapshot or 547 * bookmark must represent an earlier point in the history of "snapname"). 548 * It can be an earlier snapshot in the same filesystem or zvol as "snapname", 549 * or it can be the origin of "snapname"'s filesystem, or an earlier 550 * snapshot in the origin, etc. 551 * 552 * "fd" is the file descriptor to write the send stream to. 553 * 554 * If "flags" contains LZC_SEND_FLAG_LARGE_BLOCK, the stream is permitted 555 * to contain DRR_WRITE records with drr_length > 128K, and DRR_OBJECT 556 * records with drr_blksz > 128K. 557 * 558 * If "flags" contains LZC_SEND_FLAG_EMBED_DATA, the stream is permitted 559 * to contain DRR_WRITE_EMBEDDED records with drr_etype==BP_EMBEDDED_TYPE_DATA, 560 * which the receiving system must support (as indicated by support 561 * for the "embedded_data" feature). 562 */ 563 int 564 lzc_send(const char *snapname, const char *from, int fd, 565 enum lzc_send_flags flags) 566 { 567 return (lzc_send_resume(snapname, from, fd, flags, 0, 0)); 568 } 569 570 int 571 lzc_send_resume(const char *snapname, const char *from, int fd, 572 enum lzc_send_flags flags, uint64_t resumeobj, uint64_t resumeoff) 573 { 574 nvlist_t *args; 575 int err; 576 577 args = fnvlist_alloc(); 578 fnvlist_add_int32(args, "fd", fd); 579 if (from != NULL) 580 fnvlist_add_string(args, "fromsnap", from); 581 if (flags & LZC_SEND_FLAG_LARGE_BLOCK) 582 fnvlist_add_boolean(args, "largeblockok"); 583 if (flags & LZC_SEND_FLAG_EMBED_DATA) 584 fnvlist_add_boolean(args, "embedok"); 585 if (flags & LZC_SEND_FLAG_COMPRESS) 586 fnvlist_add_boolean(args, "compressok"); 587 if (resumeobj != 0 || resumeoff != 0) { 588 fnvlist_add_uint64(args, "resume_object", resumeobj); 589 fnvlist_add_uint64(args, "resume_offset", resumeoff); 590 } 591 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL); 592 nvlist_free(args); 593 return (err); 594 } 595 596 /* 597 * "from" can be NULL, a snapshot, or a bookmark. 598 * 599 * If from is NULL, a full (non-incremental) stream will be estimated. This 600 * is calculated very efficiently. 601 * 602 * If from is a snapshot, lzc_send_space uses the deadlists attached to 603 * each snapshot to efficiently estimate the stream size. 604 * 605 * If from is a bookmark, the indirect blocks in the destination snapshot 606 * are traversed, looking for blocks with a birth time since the creation TXG of 607 * the snapshot this bookmark was created from. This will result in 608 * significantly more I/O and be less efficient than a send space estimation on 609 * an equivalent snapshot. 610 */ 611 int 612 lzc_send_space(const char *snapname, const char *from, 613 enum lzc_send_flags flags, uint64_t *spacep) 614 { 615 nvlist_t *args; 616 nvlist_t *result; 617 int err; 618 619 args = fnvlist_alloc(); 620 if (from != NULL) 621 fnvlist_add_string(args, "from", from); 622 if (flags & LZC_SEND_FLAG_LARGE_BLOCK) 623 fnvlist_add_boolean(args, "largeblockok"); 624 if (flags & LZC_SEND_FLAG_EMBED_DATA) 625 fnvlist_add_boolean(args, "embedok"); 626 if (flags & LZC_SEND_FLAG_COMPRESS) 627 fnvlist_add_boolean(args, "compressok"); 628 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result); 629 nvlist_free(args); 630 if (err == 0) 631 *spacep = fnvlist_lookup_uint64(result, "space"); 632 nvlist_free(result); 633 return (err); 634 } 635 636 static int 637 recv_read(int fd, void *buf, int ilen) 638 { 639 char *cp = buf; 640 int rv; 641 int len = ilen; 642 643 do { 644 rv = read(fd, cp, len); 645 cp += rv; 646 len -= rv; 647 } while (rv > 0); 648 649 if (rv < 0 || len != 0) 650 return (EIO); 651 652 return (0); 653 } 654 655 static int 656 recv_impl(const char *snapname, nvlist_t *props, const char *origin, 657 boolean_t force, boolean_t resumable, int fd, 658 const dmu_replay_record_t *begin_record) 659 { 660 /* 661 * The receive ioctl is still legacy, so we need to construct our own 662 * zfs_cmd_t rather than using zfsc_ioctl(). 663 */ 664 zfs_cmd_t zc = { 0 }; 665 char *atp; 666 char *packed = NULL; 667 size_t size; 668 int error; 669 670 ASSERT3S(g_refcount, >, 0); 671 VERIFY3S(g_fd, !=, -1); 672 673 /* zc_name is name of containing filesystem */ 674 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name)); 675 atp = strchr(zc.zc_name, '@'); 676 if (atp == NULL) 677 return (EINVAL); 678 *atp = '\0'; 679 680 /* if the fs does not exist, try its parent. */ 681 if (!lzc_exists(zc.zc_name)) { 682 char *slashp = strrchr(zc.zc_name, '/'); 683 if (slashp == NULL) 684 return (ENOENT); 685 *slashp = '\0'; 686 687 } 688 689 /* zc_value is full name of the snapshot to create */ 690 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value)); 691 692 if (props != NULL) { 693 /* zc_nvlist_src is props to set */ 694 packed = fnvlist_pack(props, &size); 695 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed; 696 zc.zc_nvlist_src_size = size; 697 } 698 699 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */ 700 if (origin != NULL) 701 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string)); 702 703 /* zc_begin_record is non-byteswapped BEGIN record */ 704 if (begin_record == NULL) { 705 error = recv_read(fd, &zc.zc_begin_record, 706 sizeof (zc.zc_begin_record)); 707 if (error != 0) 708 goto out; 709 } else { 710 zc.zc_begin_record = *begin_record; 711 } 712 713 /* zc_cookie is fd to read from */ 714 zc.zc_cookie = fd; 715 716 /* zc guid is force flag */ 717 zc.zc_guid = force; 718 719 zc.zc_resumable = resumable; 720 721 /* zc_cleanup_fd is unused */ 722 zc.zc_cleanup_fd = -1; 723 724 error = ioctl(g_fd, ZFS_IOC_RECV, &zc); 725 if (error != 0) 726 error = errno; 727 728 out: 729 if (packed != NULL) 730 fnvlist_pack_free(packed, size); 731 free((void*)(uintptr_t)zc.zc_nvlist_dst); 732 return (error); 733 } 734 735 /* 736 * The simplest receive case: receive from the specified fd, creating the 737 * specified snapshot. Apply the specified properties as "received" properties 738 * (which can be overridden by locally-set properties). If the stream is a 739 * clone, its origin snapshot must be specified by 'origin'. The 'force' 740 * flag will cause the target filesystem to be rolled back or destroyed if 741 * necessary to receive. 742 * 743 * Return 0 on success or an errno on failure. 744 * 745 * Note: this interface does not work on dedup'd streams 746 * (those with DMU_BACKUP_FEATURE_DEDUP). 747 */ 748 int 749 lzc_receive(const char *snapname, nvlist_t *props, const char *origin, 750 boolean_t force, int fd) 751 { 752 return (recv_impl(snapname, props, origin, force, B_FALSE, fd, NULL)); 753 } 754 755 /* 756 * Like lzc_receive, but if the receive fails due to premature stream 757 * termination, the intermediate state will be preserved on disk. In this 758 * case, ECKSUM will be returned. The receive may subsequently be resumed 759 * with a resuming send stream generated by lzc_send_resume(). 760 */ 761 int 762 lzc_receive_resumable(const char *snapname, nvlist_t *props, const char *origin, 763 boolean_t force, int fd) 764 { 765 return (recv_impl(snapname, props, origin, force, B_TRUE, fd, NULL)); 766 } 767 768 /* 769 * Like lzc_receive, but allows the caller to read the begin record and then to 770 * pass it in. That could be useful if the caller wants to derive, for example, 771 * the snapname or the origin parameters based on the information contained in 772 * the begin record. 773 * The begin record must be in its original form as read from the stream, 774 * in other words, it should not be byteswapped. 775 * 776 * The 'resumable' parameter allows to obtain the same behavior as with 777 * lzc_receive_resumable. 778 */ 779 int 780 lzc_receive_with_header(const char *snapname, nvlist_t *props, 781 const char *origin, boolean_t force, boolean_t resumable, int fd, 782 const dmu_replay_record_t *begin_record) 783 { 784 if (begin_record == NULL) 785 return (EINVAL); 786 return (recv_impl(snapname, props, origin, force, resumable, fd, 787 begin_record)); 788 } 789 790 /* 791 * Roll back this filesystem or volume to its most recent snapshot. 792 * If snapnamebuf is not NULL, it will be filled in with the name 793 * of the most recent snapshot. 794 * Note that the latest snapshot may change if a new one is concurrently 795 * created or the current one is destroyed. lzc_rollback_to can be used 796 * to roll back to a specific latest snapshot. 797 * 798 * Return 0 on success or an errno on failure. 799 */ 800 int 801 lzc_rollback(const char *fsname, char *snapnamebuf, int snapnamelen) 802 { 803 nvlist_t *args; 804 nvlist_t *result; 805 int err; 806 807 args = fnvlist_alloc(); 808 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result); 809 nvlist_free(args); 810 if (err == 0 && snapnamebuf != NULL) { 811 const char *snapname = fnvlist_lookup_string(result, "target"); 812 (void) strlcpy(snapnamebuf, snapname, snapnamelen); 813 } 814 nvlist_free(result); 815 816 return (err); 817 } 818 819 /* 820 * Roll back this filesystem or volume to the specified snapshot, 821 * if possible. 822 * 823 * Return 0 on success or an errno on failure. 824 */ 825 int 826 lzc_rollback_to(const char *fsname, const char *snapname) 827 { 828 nvlist_t *args; 829 nvlist_t *result; 830 int err; 831 832 args = fnvlist_alloc(); 833 fnvlist_add_string(args, "target", snapname); 834 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result); 835 nvlist_free(args); 836 nvlist_free(result); 837 return (err); 838 } 839 840 /* 841 * Creates bookmarks. 842 * 843 * The bookmarks nvlist maps from name of the bookmark (e.g. "pool/fs#bmark") to 844 * the name of the snapshot (e.g. "pool/fs@snap"). All the bookmarks and 845 * snapshots must be in the same pool. 846 * 847 * The returned results nvlist will have an entry for each bookmark that failed. 848 * The value will be the (int32) error code. 849 * 850 * The return value will be 0 if all bookmarks were created, otherwise it will 851 * be the errno of a (undetermined) bookmarks that failed. 852 */ 853 int 854 lzc_bookmark(nvlist_t *bookmarks, nvlist_t **errlist) 855 { 856 nvpair_t *elem; 857 int error; 858 char pool[ZFS_MAX_DATASET_NAME_LEN]; 859 860 /* determine the pool name */ 861 elem = nvlist_next_nvpair(bookmarks, NULL); 862 if (elem == NULL) 863 return (0); 864 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 865 pool[strcspn(pool, "/#")] = '\0'; 866 867 error = lzc_ioctl(ZFS_IOC_BOOKMARK, pool, bookmarks, errlist); 868 869 return (error); 870 } 871 872 /* 873 * Retrieve bookmarks. 874 * 875 * Retrieve the list of bookmarks for the given file system. The props 876 * parameter is an nvlist of property names (with no values) that will be 877 * returned for each bookmark. 878 * 879 * The following are valid properties on bookmarks, all of which are numbers 880 * (represented as uint64 in the nvlist) 881 * 882 * "guid" - globally unique identifier of the snapshot it refers to 883 * "createtxg" - txg when the snapshot it refers to was created 884 * "creation" - timestamp when the snapshot it refers to was created 885 * 886 * The format of the returned nvlist as follows: 887 * <short name of bookmark> -> { 888 * <name of property> -> { 889 * "value" -> uint64 890 * } 891 * } 892 */ 893 int 894 lzc_get_bookmarks(const char *fsname, nvlist_t *props, nvlist_t **bmarks) 895 { 896 return (lzc_ioctl(ZFS_IOC_GET_BOOKMARKS, fsname, props, bmarks)); 897 } 898 899 /* 900 * Destroys bookmarks. 901 * 902 * The keys in the bmarks nvlist are the bookmarks to be destroyed. 903 * They must all be in the same pool. Bookmarks are specified as 904 * <fs>#<bmark>. 905 * 906 * Bookmarks that do not exist will be silently ignored. 907 * 908 * The return value will be 0 if all bookmarks that existed were destroyed. 909 * 910 * Otherwise the return value will be the errno of a (undetermined) bookmark 911 * that failed, no bookmarks will be destroyed, and the errlist will have an 912 * entry for each bookmarks that failed. The value in the errlist will be 913 * the (int32) error code. 914 */ 915 int 916 lzc_destroy_bookmarks(nvlist_t *bmarks, nvlist_t **errlist) 917 { 918 nvpair_t *elem; 919 int error; 920 char pool[ZFS_MAX_DATASET_NAME_LEN]; 921 922 /* determine the pool name */ 923 elem = nvlist_next_nvpair(bmarks, NULL); 924 if (elem == NULL) 925 return (0); 926 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 927 pool[strcspn(pool, "/#")] = '\0'; 928 929 error = lzc_ioctl(ZFS_IOC_DESTROY_BOOKMARKS, pool, bmarks, errlist); 930 931 return (error); 932 } 933 934 static int 935 lzc_channel_program_impl(const char *pool, const char *program, boolean_t sync, 936 uint64_t instrlimit, uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl) 937 { 938 int error; 939 nvlist_t *args; 940 941 args = fnvlist_alloc(); 942 fnvlist_add_string(args, ZCP_ARG_PROGRAM, program); 943 fnvlist_add_nvlist(args, ZCP_ARG_ARGLIST, argnvl); 944 fnvlist_add_boolean_value(args, ZCP_ARG_SYNC, sync); 945 fnvlist_add_uint64(args, ZCP_ARG_INSTRLIMIT, instrlimit); 946 fnvlist_add_uint64(args, ZCP_ARG_MEMLIMIT, memlimit); 947 error = lzc_ioctl(ZFS_IOC_CHANNEL_PROGRAM, pool, args, outnvl); 948 fnvlist_free(args); 949 950 return (error); 951 } 952 953 /* 954 * Executes a channel program. 955 * 956 * If this function returns 0 the channel program was successfully loaded and 957 * ran without failing. Note that individual commands the channel program ran 958 * may have failed and the channel program is responsible for reporting such 959 * errors through outnvl if they are important. 960 * 961 * This method may also return: 962 * 963 * EINVAL The program contains syntax errors, or an invalid memory or time 964 * limit was given. No part of the channel program was executed. 965 * If caused by syntax errors, 'outnvl' contains information about the 966 * errors. 967 * 968 * ECHRNG The program was executed, but encountered a runtime error, such as 969 * calling a function with incorrect arguments, invoking the error() 970 * function directly, failing an assert() command, etc. Some portion 971 * of the channel program may have executed and committed changes. 972 * Information about the failure can be found in 'outnvl'. 973 * 974 * ENOMEM The program fully executed, but the output buffer was not large 975 * enough to store the returned value. No output is returned through 976 * 'outnvl'. 977 * 978 * ENOSPC The program was terminated because it exceeded its memory usage 979 * limit. Some portion of the channel program may have executed and 980 * committed changes to disk. No output is returned through 'outnvl'. 981 * 982 * ETIME The program was terminated because it exceeded its Lua instruction 983 * limit. Some portion of the channel program may have executed and 984 * committed changes to disk. No output is returned through 'outnvl'. 985 */ 986 int 987 lzc_channel_program(const char *pool, const char *program, uint64_t instrlimit, 988 uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl) 989 { 990 return (lzc_channel_program_impl(pool, program, B_TRUE, instrlimit, 991 memlimit, argnvl, outnvl)); 992 } 993 994 /* 995 * Creates a checkpoint for the specified pool. 996 * 997 * If this function returns 0 the pool was successfully checkpointed. 998 * 999 * This method may also return: 1000 * 1001 * ZFS_ERR_CHECKPOINT_EXISTS 1002 * The pool already has a checkpoint. A pools can only have one 1003 * checkpoint at most, at any given time. 1004 * 1005 * ZFS_ERR_DISCARDING_CHECKPOINT 1006 * ZFS is in the middle of discarding a checkpoint for this pool. 1007 * The pool can be checkpointed again once the discard is done. 1008 * 1009 * ZFS_DEVRM_IN_PROGRESS 1010 * A vdev is currently being removed. The pool cannot be 1011 * checkpointed until the device removal is done. 1012 * 1013 * ZFS_VDEV_TOO_BIG 1014 * One or more top-level vdevs exceed the maximum vdev size 1015 * supported for this feature. 1016 */ 1017 int 1018 lzc_pool_checkpoint(const char *pool) 1019 { 1020 int error; 1021 1022 nvlist_t *result = NULL; 1023 nvlist_t *args = fnvlist_alloc(); 1024 1025 error = lzc_ioctl(ZFS_IOC_POOL_CHECKPOINT, pool, args, &result); 1026 1027 fnvlist_free(args); 1028 fnvlist_free(result); 1029 1030 return (error); 1031 } 1032 1033 /* 1034 * Discard the checkpoint from the specified pool. 1035 * 1036 * If this function returns 0 the checkpoint was successfully discarded. 1037 * 1038 * This method may also return: 1039 * 1040 * ZFS_ERR_NO_CHECKPOINT 1041 * The pool does not have a checkpoint. 1042 * 1043 * ZFS_ERR_DISCARDING_CHECKPOINT 1044 * ZFS is already in the middle of discarding the checkpoint. 1045 */ 1046 int 1047 lzc_pool_checkpoint_discard(const char *pool) 1048 { 1049 int error; 1050 1051 nvlist_t *result = NULL; 1052 nvlist_t *args = fnvlist_alloc(); 1053 1054 error = lzc_ioctl(ZFS_IOC_POOL_DISCARD_CHECKPOINT, pool, args, &result); 1055 1056 fnvlist_free(args); 1057 fnvlist_free(result); 1058 1059 return (error); 1060 } 1061 1062 /* 1063 * Executes a read-only channel program. 1064 * 1065 * A read-only channel program works programmatically the same way as a 1066 * normal channel program executed with lzc_channel_program(). The only 1067 * difference is it runs exclusively in open-context and therefore can 1068 * return faster. The downside to that, is that the program cannot change 1069 * on-disk state by calling functions from the zfs.sync submodule. 1070 * 1071 * The return values of this function (and their meaning) are exactly the 1072 * same as the ones described in lzc_channel_program(). 1073 */ 1074 int 1075 lzc_channel_program_nosync(const char *pool, const char *program, 1076 uint64_t timeout, uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl) 1077 { 1078 return (lzc_channel_program_impl(pool, program, B_FALSE, timeout, 1079 memlimit, argnvl, outnvl)); 1080 } 1081 1082 /* 1083 * Changes initializing state. 1084 * 1085 * vdevs should be a list of (<key>, guid) where guid is a uint64 vdev GUID. 1086 * The key is ignored. 1087 * 1088 * If there are errors related to vdev arguments, per-vdev errors are returned 1089 * in an nvlist with the key "vdevs". Each error is a (guid, errno) pair where 1090 * guid is stringified with PRIu64, and errno is one of the following as 1091 * an int64_t: 1092 * - ENODEV if the device was not found 1093 * - EINVAL if the devices is not a leaf or is not concrete (e.g. missing) 1094 * - EROFS if the device is not writeable 1095 * - EBUSY start requested but the device is already being initialized 1096 * - ESRCH cancel/suspend requested but device is not being initialized 1097 * 1098 * If the errlist is empty, then return value will be: 1099 * - EINVAL if one or more arguments was invalid 1100 * - Other spa_open failures 1101 * - 0 if the operation succeeded 1102 */ 1103 int 1104 lzc_initialize(const char *poolname, pool_initialize_func_t cmd_type, 1105 nvlist_t *vdevs, nvlist_t **errlist) 1106 { 1107 int error; 1108 nvlist_t *args = fnvlist_alloc(); 1109 fnvlist_add_uint64(args, ZPOOL_INITIALIZE_COMMAND, (uint64_t)cmd_type); 1110 fnvlist_add_nvlist(args, ZPOOL_INITIALIZE_VDEVS, vdevs); 1111 1112 error = lzc_ioctl(ZFS_IOC_POOL_INITIALIZE, poolname, args, errlist); 1113 1114 fnvlist_free(args); 1115 1116 return (error); 1117 } 1118