libzfs_core.c revision 254587
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) 2013 by Delphix. All rights reserved. 24 * Copyright (c) 2013 Steven Hartland. All rights reserved. 25 */ 26 27/* 28 * LibZFS_Core (lzc) is intended to replace most functionality in libzfs. 29 * It has the following characteristics: 30 * 31 * - Thread Safe. libzfs_core is accessible concurrently from multiple 32 * threads. This is accomplished primarily by avoiding global data 33 * (e.g. caching). Since it's thread-safe, there is no reason for a 34 * process to have multiple libzfs "instances". Therefore, we store 35 * our few pieces of data (e.g. the file descriptor) in global 36 * variables. The fd is reference-counted so that the libzfs_core 37 * library can be "initialized" multiple times (e.g. by different 38 * consumers within the same process). 39 * 40 * - Committed Interface. The libzfs_core interface will be committed, 41 * therefore consumers can compile against it and be confident that 42 * their code will continue to work on future releases of this code. 43 * Currently, the interface is Evolving (not Committed), but we intend 44 * to commit to it once it is more complete and we determine that it 45 * meets the needs of all consumers. 46 * 47 * - Programatic Error Handling. libzfs_core communicates errors with 48 * defined error numbers, and doesn't print anything to stdout/stderr. 49 * 50 * - Thin Layer. libzfs_core is a thin layer, marshaling arguments 51 * to/from the kernel ioctls. There is generally a 1:1 correspondence 52 * between libzfs_core functions and ioctls to /dev/zfs. 53 * 54 * - Clear Atomicity. Because libzfs_core functions are generally 1:1 55 * with kernel ioctls, and kernel ioctls are general atomic, each 56 * libzfs_core function is atomic. For example, creating multiple 57 * snapshots with a single call to lzc_snapshot() is atomic -- it 58 * can't fail with only some of the requested snapshots created, even 59 * in the event of power loss or system crash. 60 * 61 * - Continued libzfs Support. Some higher-level operations (e.g. 62 * support for "zfs send -R") are too complicated to fit the scope of 63 * libzfs_core. This functionality will continue to live in libzfs. 64 * Where appropriate, libzfs will use the underlying atomic operations 65 * of libzfs_core. For example, libzfs may implement "zfs send -R | 66 * zfs receive" by using individual "send one snapshot", rename, 67 * destroy, and "receive one snapshot" operations in libzfs_core. 68 * /sbin/zfs and /zbin/zpool will link with both libzfs and 69 * libzfs_core. Other consumers should aim to use only libzfs_core, 70 * since that will be the supported, stable interface going forwards. 71 */ 72 73#define _IN_LIBZFS_CORE_ 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#include "libzfs_core_compat.h" 89#include "libzfs_compat.h" 90 91#ifdef __FreeBSD__ 92extern int zfs_ioctl_version; 93#endif 94 95static int g_fd; 96static pthread_mutex_t g_lock = PTHREAD_MUTEX_INITIALIZER; 97static int g_refcount; 98 99int 100libzfs_core_init(void) 101{ 102 (void) pthread_mutex_lock(&g_lock); 103 if (g_refcount == 0) { 104 g_fd = open("/dev/zfs", O_RDWR); 105 if (g_fd < 0) { 106 (void) pthread_mutex_unlock(&g_lock); 107 return (errno); 108 } 109 } 110 g_refcount++; 111 (void) pthread_mutex_unlock(&g_lock); 112 113 return (0); 114} 115 116void 117libzfs_core_fini(void) 118{ 119 (void) pthread_mutex_lock(&g_lock); 120 ASSERT3S(g_refcount, >, 0); 121 g_refcount--; 122 if (g_refcount == 0) 123 (void) close(g_fd); 124 (void) pthread_mutex_unlock(&g_lock); 125} 126 127static int 128lzc_ioctl(zfs_ioc_t ioc, const char *name, 129 nvlist_t *source, nvlist_t **resultp) 130{ 131 zfs_cmd_t zc = { 0 }; 132 int error = 0; 133 char *packed; 134#ifdef __FreeBSD__ 135 nvlist_t *oldsource; 136#endif 137 size_t size; 138 139 ASSERT3S(g_refcount, >, 0); 140 141 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name)); 142 143#ifdef __FreeBSD__ 144 if (zfs_ioctl_version == ZFS_IOCVER_UNDEF) 145 zfs_ioctl_version = get_zfs_ioctl_version(); 146 147 if (zfs_ioctl_version < ZFS_IOCVER_LZC) { 148 oldsource = source; 149 error = lzc_compat_pre(&zc, &ioc, &source); 150 if (error) 151 return (error); 152 } 153#endif 154 155 packed = fnvlist_pack(source, &size); 156 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed; 157 zc.zc_nvlist_src_size = size; 158 159 if (resultp != NULL) { 160 *resultp = NULL; 161 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024); 162 zc.zc_nvlist_dst = (uint64_t)(uintptr_t) 163 malloc(zc.zc_nvlist_dst_size); 164#ifdef illumos 165 if (zc.zc_nvlist_dst == NULL) { 166#else 167 if (zc.zc_nvlist_dst == 0) { 168#endif 169 error = ENOMEM; 170 goto out; 171 } 172 } 173 174 while (ioctl(g_fd, ioc, &zc) != 0) { 175 if (errno == ENOMEM && resultp != NULL) { 176 free((void *)(uintptr_t)zc.zc_nvlist_dst); 177 zc.zc_nvlist_dst_size *= 2; 178 zc.zc_nvlist_dst = (uint64_t)(uintptr_t) 179 malloc(zc.zc_nvlist_dst_size); 180#ifdef illumos 181 if (zc.zc_nvlist_dst == NULL) { 182#else 183 if (zc.zc_nvlist_dst == 0) { 184#endif 185 error = ENOMEM; 186 goto out; 187 } 188 } else { 189 error = errno; 190 break; 191 } 192 } 193 194#ifdef __FreeBSD__ 195 if (zfs_ioctl_version < ZFS_IOCVER_LZC) 196 lzc_compat_post(&zc, ioc); 197#endif 198 if (zc.zc_nvlist_dst_filled) { 199 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst, 200 zc.zc_nvlist_dst_size); 201 } 202#ifdef __FreeBSD__ 203 if (zfs_ioctl_version < ZFS_IOCVER_LZC) 204 lzc_compat_outnvl(&zc, ioc, resultp); 205#endif 206out: 207#ifdef __FreeBSD__ 208 if (zfs_ioctl_version < ZFS_IOCVER_LZC) { 209 if (source != oldsource) 210 nvlist_free(source); 211 source = oldsource; 212 } 213#endif 214 fnvlist_pack_free(packed, size); 215 free((void *)(uintptr_t)zc.zc_nvlist_dst); 216 return (error); 217} 218 219int 220lzc_create(const char *fsname, dmu_objset_type_t type, nvlist_t *props) 221{ 222 int error; 223 nvlist_t *args = fnvlist_alloc(); 224 fnvlist_add_int32(args, "type", type); 225 if (props != NULL) 226 fnvlist_add_nvlist(args, "props", props); 227 error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL); 228 nvlist_free(args); 229 return (error); 230} 231 232int 233lzc_clone(const char *fsname, const char *origin, 234 nvlist_t *props) 235{ 236 int error; 237 nvlist_t *args = fnvlist_alloc(); 238 fnvlist_add_string(args, "origin", origin); 239 if (props != NULL) 240 fnvlist_add_nvlist(args, "props", props); 241 error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL); 242 nvlist_free(args); 243 return (error); 244} 245 246/* 247 * Creates snapshots. 248 * 249 * The keys in the snaps nvlist are the snapshots to be created. 250 * They must all be in the same pool. 251 * 252 * The props nvlist is properties to set. Currently only user properties 253 * are supported. { user:prop_name -> string value } 254 * 255 * The returned results nvlist will have an entry for each snapshot that failed. 256 * The value will be the (int32) error code. 257 * 258 * The return value will be 0 if all snapshots were created, otherwise it will 259 * be the errno of a (unspecified) snapshot that failed. 260 */ 261int 262lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist) 263{ 264 nvpair_t *elem; 265 nvlist_t *args; 266 int error; 267 char pool[MAXNAMELEN]; 268 269 *errlist = NULL; 270 271 /* determine the pool name */ 272 elem = nvlist_next_nvpair(snaps, NULL); 273 if (elem == NULL) 274 return (0); 275 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 276 pool[strcspn(pool, "/@")] = '\0'; 277 278 args = fnvlist_alloc(); 279 fnvlist_add_nvlist(args, "snaps", snaps); 280 if (props != NULL) 281 fnvlist_add_nvlist(args, "props", props); 282 283 error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist); 284 nvlist_free(args); 285 286 return (error); 287} 288 289/* 290 * Destroys snapshots. 291 * 292 * The keys in the snaps nvlist are the snapshots to be destroyed. 293 * They must all be in the same pool. 294 * 295 * Snapshots that do not exist will be silently ignored. 296 * 297 * If 'defer' is not set, and a snapshot has user holds or clones, the 298 * destroy operation will fail and none of the snapshots will be 299 * destroyed. 300 * 301 * If 'defer' is set, and a snapshot has user holds or clones, it will be 302 * marked for deferred destruction, and will be destroyed when the last hold 303 * or clone is removed/destroyed. 304 * 305 * The return value will be 0 if all snapshots were destroyed (or marked for 306 * later destruction if 'defer' is set) or didn't exist to begin with. 307 * 308 * Otherwise the return value will be the errno of a (unspecified) snapshot 309 * that failed, no snapshots will be destroyed, and the errlist will have an 310 * entry for each snapshot that failed. The value in the errlist will be 311 * the (int32) error code. 312 */ 313int 314lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist) 315{ 316 nvpair_t *elem; 317 nvlist_t *args; 318 int error; 319 char pool[MAXNAMELEN]; 320 321 /* determine the pool name */ 322 elem = nvlist_next_nvpair(snaps, NULL); 323 if (elem == NULL) 324 return (0); 325 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 326 pool[strcspn(pool, "/@")] = '\0'; 327 328 args = fnvlist_alloc(); 329 fnvlist_add_nvlist(args, "snaps", snaps); 330 if (defer) 331 fnvlist_add_boolean(args, "defer"); 332 333 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist); 334 nvlist_free(args); 335 336 return (error); 337} 338 339int 340lzc_snaprange_space(const char *firstsnap, const char *lastsnap, 341 uint64_t *usedp) 342{ 343 nvlist_t *args; 344 nvlist_t *result; 345 int err; 346 char fs[MAXNAMELEN]; 347 char *atp; 348 349 /* determine the fs name */ 350 (void) strlcpy(fs, firstsnap, sizeof (fs)); 351 atp = strchr(fs, '@'); 352 if (atp == NULL) 353 return (EINVAL); 354 *atp = '\0'; 355 356 args = fnvlist_alloc(); 357 fnvlist_add_string(args, "firstsnap", firstsnap); 358 359 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result); 360 nvlist_free(args); 361 if (err == 0) 362 *usedp = fnvlist_lookup_uint64(result, "used"); 363 fnvlist_free(result); 364 365 return (err); 366} 367 368boolean_t 369lzc_exists(const char *dataset) 370{ 371 /* 372 * The objset_stats ioctl is still legacy, so we need to construct our 373 * own zfs_cmd_t rather than using zfsc_ioctl(). 374 */ 375 zfs_cmd_t zc = { 0 }; 376 377 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name)); 378 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0); 379} 380 381/* 382 * Create "user holds" on snapshots. If there is a hold on a snapshot, 383 * the snapshot can not be destroyed. (However, it can be marked for deletion 384 * by lzc_destroy_snaps(defer=B_TRUE).) 385 * 386 * The keys in the nvlist are snapshot names. 387 * The snapshots must all be in the same pool. 388 * The value is the name of the hold (string type). 389 * 390 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL). 391 * In this case, when the cleanup_fd is closed (including on process 392 * termination), the holds will be released. If the system is shut down 393 * uncleanly, the holds will be released when the pool is next opened 394 * or imported. 395 * 396 * Holds for snapshots which don't exist will be skipped and have an entry 397 * added to errlist, but will not cause an overall failure. 398 * 399 * The return value will be 0 if all holds, for snapshots that existed, 400 * were succesfully created. 401 * 402 * Otherwise the return value will be the errno of a (unspecified) hold that 403 * failed and no holds will be created. 404 * 405 * In all cases the errlist will have an entry for each hold that failed 406 * (name = snapshot), with its value being the error code (int32). 407 */ 408int 409lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist) 410{ 411 char pool[MAXNAMELEN]; 412 nvlist_t *args; 413 nvpair_t *elem; 414 int error; 415 416 /* determine the pool name */ 417 elem = nvlist_next_nvpair(holds, NULL); 418 if (elem == NULL) 419 return (0); 420 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 421 pool[strcspn(pool, "/@")] = '\0'; 422 423 args = fnvlist_alloc(); 424 fnvlist_add_nvlist(args, "holds", holds); 425 if (cleanup_fd != -1) 426 fnvlist_add_int32(args, "cleanup_fd", cleanup_fd); 427 428 error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist); 429 nvlist_free(args); 430 return (error); 431} 432 433/* 434 * Release "user holds" on snapshots. If the snapshot has been marked for 435 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have 436 * any clones, and all the user holds are removed, then the snapshot will be 437 * destroyed. 438 * 439 * The keys in the nvlist are snapshot names. 440 * The snapshots must all be in the same pool. 441 * The value is a nvlist whose keys are the holds to remove. 442 * 443 * Holds which failed to release because they didn't exist will have an entry 444 * added to errlist, but will not cause an overall failure. 445 * 446 * The return value will be 0 if the nvl holds was empty or all holds that 447 * existed, were successfully removed. 448 * 449 * Otherwise the return value will be the errno of a (unspecified) hold that 450 * failed to release and no holds will be released. 451 * 452 * In all cases the errlist will have an entry for each hold that failed to 453 * to release. 454 */ 455int 456lzc_release(nvlist_t *holds, nvlist_t **errlist) 457{ 458 char pool[MAXNAMELEN]; 459 nvpair_t *elem; 460 461 /* determine the pool name */ 462 elem = nvlist_next_nvpair(holds, NULL); 463 if (elem == NULL) 464 return (0); 465 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 466 pool[strcspn(pool, "/@")] = '\0'; 467 468 return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist)); 469} 470 471/* 472 * Retrieve list of user holds on the specified snapshot. 473 * 474 * On success, *holdsp will be set to a nvlist which the caller must free. 475 * The keys are the names of the holds, and the value is the creation time 476 * of the hold (uint64) in seconds since the epoch. 477 */ 478int 479lzc_get_holds(const char *snapname, nvlist_t **holdsp) 480{ 481 int error; 482 nvlist_t *innvl = fnvlist_alloc(); 483 error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp); 484 fnvlist_free(innvl); 485 return (error); 486} 487 488/* 489 * If fromsnap is NULL, a full (non-incremental) stream will be sent. 490 */ 491int 492lzc_send(const char *snapname, const char *fromsnap, int fd) 493{ 494 nvlist_t *args; 495 int err; 496 497 args = fnvlist_alloc(); 498 fnvlist_add_int32(args, "fd", fd); 499 if (fromsnap != NULL) 500 fnvlist_add_string(args, "fromsnap", fromsnap); 501 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL); 502 nvlist_free(args); 503 return (err); 504} 505 506/* 507 * If fromsnap is NULL, a full (non-incremental) stream will be estimated. 508 */ 509int 510lzc_send_space(const char *snapname, const char *fromsnap, uint64_t *spacep) 511{ 512 nvlist_t *args; 513 nvlist_t *result; 514 int err; 515 516 args = fnvlist_alloc(); 517 if (fromsnap != NULL) 518 fnvlist_add_string(args, "fromsnap", fromsnap); 519 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result); 520 nvlist_free(args); 521 if (err == 0) 522 *spacep = fnvlist_lookup_uint64(result, "space"); 523 nvlist_free(result); 524 return (err); 525} 526 527static int 528recv_read(int fd, void *buf, int ilen) 529{ 530 char *cp = buf; 531 int rv; 532 int len = ilen; 533 534 do { 535 rv = read(fd, cp, len); 536 cp += rv; 537 len -= rv; 538 } while (rv > 0); 539 540 if (rv < 0 || len != 0) 541 return (EIO); 542 543 return (0); 544} 545 546/* 547 * The simplest receive case: receive from the specified fd, creating the 548 * specified snapshot. Apply the specified properties a "received" properties 549 * (which can be overridden by locally-set properties). If the stream is a 550 * clone, its origin snapshot must be specified by 'origin'. The 'force' 551 * flag will cause the target filesystem to be rolled back or destroyed if 552 * necessary to receive. 553 * 554 * Return 0 on success or an errno on failure. 555 * 556 * Note: this interface does not work on dedup'd streams 557 * (those with DMU_BACKUP_FEATURE_DEDUP). 558 */ 559int 560lzc_receive(const char *snapname, nvlist_t *props, const char *origin, 561 boolean_t force, int fd) 562{ 563 /* 564 * The receive ioctl is still legacy, so we need to construct our own 565 * zfs_cmd_t rather than using zfsc_ioctl(). 566 */ 567 zfs_cmd_t zc = { 0 }; 568 char *atp; 569 char *packed = NULL; 570 size_t size; 571 dmu_replay_record_t drr; 572 int error; 573 574 ASSERT3S(g_refcount, >, 0); 575 576 /* zc_name is name of containing filesystem */ 577 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name)); 578 atp = strchr(zc.zc_name, '@'); 579 if (atp == NULL) 580 return (EINVAL); 581 *atp = '\0'; 582 583 /* if the fs does not exist, try its parent. */ 584 if (!lzc_exists(zc.zc_name)) { 585 char *slashp = strrchr(zc.zc_name, '/'); 586 if (slashp == NULL) 587 return (ENOENT); 588 *slashp = '\0'; 589 590 } 591 592 /* zc_value is full name of the snapshot to create */ 593 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value)); 594 595 if (props != NULL) { 596 /* zc_nvlist_src is props to set */ 597 packed = fnvlist_pack(props, &size); 598 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed; 599 zc.zc_nvlist_src_size = size; 600 } 601 602 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */ 603 if (origin != NULL) 604 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string)); 605 606 /* zc_begin_record is non-byteswapped BEGIN record */ 607 error = recv_read(fd, &drr, sizeof (drr)); 608 if (error != 0) 609 goto out; 610 zc.zc_begin_record = drr.drr_u.drr_begin; 611 612 /* zc_cookie is fd to read from */ 613 zc.zc_cookie = fd; 614 615 /* zc guid is force flag */ 616 zc.zc_guid = force; 617 618 /* zc_cleanup_fd is unused */ 619 zc.zc_cleanup_fd = -1; 620 621 error = ioctl(g_fd, ZFS_IOC_RECV, &zc); 622 if (error != 0) 623 error = errno; 624 625out: 626 if (packed != NULL) 627 fnvlist_pack_free(packed, size); 628 free((void*)(uintptr_t)zc.zc_nvlist_dst); 629 return (error); 630} 631 632/* 633 * Roll back this filesystem or volume to its most recent snapshot. 634 * If snapnamebuf is not NULL, it will be filled in with the name 635 * of the most recent snapshot. 636 * 637 * Return 0 on success or an errno on failure. 638 */ 639int 640lzc_rollback(const char *fsname, char *snapnamebuf, int snapnamelen) 641{ 642 nvlist_t *args; 643 nvlist_t *result; 644 int err; 645 646 args = fnvlist_alloc(); 647 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result); 648 nvlist_free(args); 649 if (err == 0 && snapnamebuf != NULL) { 650 const char *snapname = fnvlist_lookup_string(result, "target"); 651 (void) strlcpy(snapnamebuf, snapname, snapnamelen); 652 } 653 return (err); 654} 655