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