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 by Delphix. All rights reserved.
  24  */
  25 
  26 /*
  27  * LibZFS_Core (lzc) is intended to replace most functionality in libzfs.
  28  * It has the following characteristics:
  29  *
  30  *  - Thread Safe.  libzfs_core is accessible concurrently from multiple
  31  *  threads.  This is accomplished primarily by avoiding global data
  32  *  (e.g. caching).  Since it's thread-safe, there is no reason for a
  33  *  process to have multiple libzfs "instances".  Therefore, we store
  34  *  our few pieces of data (e.g. the file descriptor) in global
  35  *  variables.  The fd is reference-counted so that the libzfs_core
  36  *  library can be "initialized" multiple times (e.g. by different
  37  *  consumers within the same process).
  38  *
  39  *  - Committed Interface.  The libzfs_core interface will be committed,
  40  *  therefore consumers can compile against it and be confident that
  41  *  their code will continue to work on future releases of this code.
  42  *  Currently, the interface is Evolving (not Committed), but we intend
  43  *  to commit to it once it is more complete and we determine that it
  44  *  meets the needs of all consumers.
  45  *
  46  *  - Programatic Error Handling.  libzfs_core communicates errors with
  47  *  defined error numbers, and doesn't print anything to stdout/stderr.
  48  *
  49  *  - Thin Layer.  libzfs_core is a thin layer, marshaling arguments
  50  *  to/from the kernel ioctls.  There is generally a 1:1 correspondence
  51  *  between libzfs_core functions and ioctls to /dev/zfs.
  52  *
  53  *  - Clear Atomicity.  Because libzfs_core functions are generally 1:1
  54  *  with kernel ioctls, and kernel ioctls are general atomic, each
  55  *  libzfs_core function is atomic.  For example, creating multiple
  56  *  snapshots with a single call to lzc_snapshot() is atomic -- it
  57  *  can't fail with only some of the requested snapshots created, even
  58  *  in the event of power loss or system crash.
  59  *
  60  *  - Continued libzfs Support.  Some higher-level operations (e.g.
  61  *  support for "zfs send -R") are too complicated to fit the scope of
  62  *  libzfs_core.  This functionality will continue to live in libzfs.
  63  *  Where appropriate, libzfs will use the underlying atomic operations
  64  *  of libzfs_core.  For example, libzfs may implement "zfs send -R |
  65  *  zfs receive" by using individual "send one snapshot", rename,
  66  *  destroy, and "receive one snapshot" operations in libzfs_core.
  67  *  /sbin/zfs and /zbin/zpool will link with both libzfs and
  68  *  libzfs_core.  Other consumers should aim to use only libzfs_core,
  69  *  since that will be the supported, stable interface going forwards.
  70  */
  71 
  72 #include <libzfs_core.h>
  73 #include <ctype.h>
  74 #include <unistd.h>
  75 #include <stdlib.h>
  76 #include <string.h>
  77 #include <errno.h>
  78 #include <fcntl.h>
  79 #include <pthread.h>
  80 #include <sys/nvpair.h>
  81 #include <sys/param.h>
  82 #include <sys/types.h>
  83 #include <sys/stat.h>
  84 #include <sys/zfs_ioctl.h>
  85 
  86 static int g_fd;
  87 static pthread_mutex_t g_lock = PTHREAD_MUTEX_INITIALIZER;
  88 static int g_refcount;
  89 
  90 int
  91 libzfs_core_init(void)
  92 {
  93         (void) pthread_mutex_lock(&g_lock);
  94         if (g_refcount == 0) {
  95                 g_fd = open("/dev/zfs", O_RDWR);
  96                 if (g_fd < 0) {
  97                         (void) pthread_mutex_unlock(&g_lock);
  98                         return (errno);
  99                 }
 100         }
 101         g_refcount++;
 102         (void) pthread_mutex_unlock(&g_lock);
 103         return (0);
 104 }
 105 
 106 void
 107 libzfs_core_fini(void)
 108 {
 109         (void) pthread_mutex_lock(&g_lock);
 110         ASSERT3S(g_refcount, >, 0);
 111         g_refcount--;
 112         if (g_refcount == 0)
 113                 (void) close(g_fd);
 114         (void) pthread_mutex_unlock(&g_lock);
 115 }
 116 
 117 static int
 118 lzc_ioctl(zfs_ioc_t ioc, const char *name,
 119     nvlist_t *source, nvlist_t **resultp)
 120 {
 121         zfs_cmd_t zc = { 0 };
 122         int error = 0;
 123         char *packed;
 124         size_t size;
 125 
 126         ASSERT3S(g_refcount, >, 0);
 127 
 128         (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
 129 
 130         packed = fnvlist_pack(source, &size);
 131         zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
 132         zc.zc_nvlist_src_size = size;
 133 
 134         if (resultp != NULL) {
 135                 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024);
 136                 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
 137                     malloc(zc.zc_nvlist_dst_size);
 138                 if (zc.zc_nvlist_dst == NULL) {
 139                         error = ENOMEM;
 140                         goto out;
 141                 }
 142         }
 143 
 144         while (ioctl(g_fd, ioc, &zc) != 0) {
 145                 if (errno == ENOMEM && resultp != NULL) {
 146                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
 147                         zc.zc_nvlist_dst_size *= 2;
 148                         zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
 149                             malloc(zc.zc_nvlist_dst_size);
 150                         if (zc.zc_nvlist_dst == NULL) {
 151                                 error = ENOMEM;
 152                                 goto out;
 153                         }
 154                 } else {
 155                         error = errno;
 156                         break;
 157                 }
 158         }
 159         if (zc.zc_nvlist_dst_filled) {
 160                 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
 161                     zc.zc_nvlist_dst_size);
 162         } else if (resultp != NULL) {
 163                 *resultp = NULL;
 164         }
 165 
 166 out:
 167         fnvlist_pack_free(packed, size);
 168         free((void *)(uintptr_t)zc.zc_nvlist_dst);
 169         return (error);
 170 }
 171 
 172 int
 173 lzc_create(const char *fsname, dmu_objset_type_t type, nvlist_t *props)
 174 {
 175         int error;
 176         nvlist_t *args = fnvlist_alloc();
 177         fnvlist_add_int32(args, "type", type);
 178         if (props != NULL)
 179                 fnvlist_add_nvlist(args, "props", props);
 180         error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL);
 181         nvlist_free(args);
 182         return (error);
 183 }
 184 
 185 int
 186 lzc_clone(const char *fsname, const char *origin,
 187     nvlist_t *props)
 188 {
 189         int error;
 190         nvlist_t *args = fnvlist_alloc();
 191         fnvlist_add_string(args, "origin", origin);
 192         if (props != NULL)
 193                 fnvlist_add_nvlist(args, "props", props);
 194         error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL);
 195         nvlist_free(args);
 196         return (error);
 197 }
 198 
 199 /*
 200  * Creates snapshots.
 201  *
 202  * The keys in the snaps nvlist are the snapshots to be created.
 203  * They must all be in the same pool.
 204  *
 205  * The props nvlist is properties to set.  Currently only user properties
 206  * are supported.  { user:prop_name -> string value }
 207  *
 208  * The returned results nvlist will have an entry for each snapshot that failed.
 209  * The value will be the (int32) error code.
 210  *
 211  * The return value will be 0 if all snapshots were created, otherwise it will
 212  * be the errno of a (undetermined) snapshot that failed.
 213  */
 214 int
 215 lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist)
 216 {
 217         nvpair_t *elem;
 218         nvlist_t *args;
 219         int error;
 220         char pool[MAXNAMELEN];
 221 
 222         *errlist = NULL;
 223 
 224         /* determine the pool name */
 225         elem = nvlist_next_nvpair(snaps, NULL);
 226         if (elem == NULL)
 227                 return (0);
 228         (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
 229         pool[strcspn(pool, "/@")] = '\0';
 230 
 231         args = fnvlist_alloc();
 232         fnvlist_add_nvlist(args, "snaps", snaps);
 233         if (props != NULL)
 234                 fnvlist_add_nvlist(args, "props", props);
 235 
 236         error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist);
 237         nvlist_free(args);
 238 
 239         return (error);
 240 }
 241 
 242 /*
 243  * Destroys snapshots.
 244  *
 245  * The keys in the snaps nvlist are the snapshots to be destroyed.
 246  * They must all be in the same pool.
 247  *
 248  * Snapshots that do not exist will be silently ignored.
 249  *
 250  * If 'defer' is not set, and a snapshot has user holds or clones, the
 251  * destroy operation will fail and none of the snapshots will be
 252  * destroyed.
 253  *
 254  * If 'defer' is set, and a snapshot has user holds or clones, it will be
 255  * marked for deferred destruction, and will be destroyed when the last hold
 256  * or clone is removed/destroyed.
 257  *
 258  * The return value will be 0 if all snapshots were destroyed (or marked for
 259  * later destruction if 'defer' is set) or didn't exist to begin with.
 260  *
 261  * Otherwise the return value will be the errno of a (undetermined) snapshot
 262  * that failed, no snapshots will be destroyed, and the errlist will have an
 263  * entry for each snapshot that failed.  The value in the errlist will be
 264  * the (int32) error code.
 265  */
 266 int
 267 lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist)
 268 {
 269         nvpair_t *elem;
 270         nvlist_t *args;
 271         int error;
 272         char pool[MAXNAMELEN];
 273 
 274         /* determine the pool name */
 275         elem = nvlist_next_nvpair(snaps, NULL);
 276         if (elem == NULL)
 277                 return (0);
 278         (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
 279         pool[strcspn(pool, "/@")] = '\0';
 280 
 281         args = fnvlist_alloc();
 282         fnvlist_add_nvlist(args, "snaps", snaps);
 283         if (defer)
 284                 fnvlist_add_boolean(args, "defer");
 285 
 286         error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist);
 287         nvlist_free(args);
 288 
 289         return (error);
 290 
 291 }
 292 
 293 int
 294 lzc_snaprange_space(const char *firstsnap, const char *lastsnap,
 295     uint64_t *usedp)
 296 {
 297         nvlist_t *args;
 298         nvlist_t *result;
 299         int err;
 300         char fs[MAXNAMELEN];
 301         char *atp;
 302 
 303         /* determine the fs name */
 304         (void) strlcpy(fs, firstsnap, sizeof (fs));
 305         atp = strchr(fs, '@');
 306         if (atp == NULL)
 307                 return (EINVAL);
 308         *atp = '\0';
 309 
 310         args = fnvlist_alloc();
 311         fnvlist_add_string(args, "firstsnap", firstsnap);
 312 
 313         err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result);
 314         nvlist_free(args);
 315         if (err == 0)
 316                 *usedp = fnvlist_lookup_uint64(result, "used");
 317         fnvlist_free(result);
 318 
 319         return (err);
 320 }
 321 
 322 boolean_t
 323 lzc_exists(const char *dataset)
 324 {
 325         /*
 326          * The objset_stats ioctl is still legacy, so we need to construct our
 327          * own zfs_cmd_t rather than using zfsc_ioctl().
 328          */
 329         zfs_cmd_t zc = { 0 };
 330 
 331         (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
 332         return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0);
 333 }
 334 
 335 /*
 336  * If fromsnap is NULL, a full (non-incremental) stream will be sent.
 337  */
 338 int
 339 lzc_send(const char *snapname, const char *fromsnap, int fd)
 340 {
 341         nvlist_t *args;
 342         int err;
 343 
 344         args = fnvlist_alloc();
 345         fnvlist_add_int32(args, "fd", fd);
 346         if (fromsnap != NULL)
 347                 fnvlist_add_string(args, "fromsnap", fromsnap);
 348         err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL);
 349         nvlist_free(args);
 350         return (err);
 351 }
 352 
 353 /*
 354  * If fromsnap is NULL, a full (non-incremental) stream will be estimated.
 355  */
 356 int
 357 lzc_send_space(const char *snapname, const char *fromsnap, uint64_t *spacep)
 358 {
 359         nvlist_t *args;
 360         nvlist_t *result;
 361         int err;
 362 
 363         args = fnvlist_alloc();
 364         if (fromsnap != NULL)
 365                 fnvlist_add_string(args, "fromsnap", fromsnap);
 366         err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result);
 367         nvlist_free(args);
 368         if (err == 0)
 369                 *spacep = fnvlist_lookup_uint64(result, "space");
 370         nvlist_free(result);
 371         return (err);
 372 }
 373 
 374 static int
 375 recv_read(int fd, void *buf, int ilen)
 376 {
 377         char *cp = buf;
 378         int rv;
 379         int len = ilen;
 380 
 381         do {
 382                 rv = read(fd, cp, len);
 383                 cp += rv;
 384                 len -= rv;
 385         } while (rv > 0);
 386 
 387         if (rv < 0 || len != 0)
 388                 return (EIO);
 389 
 390         return (0);
 391 }
 392 
 393 /*
 394  * The simplest receive case: receive from the specified fd, creating the
 395  * specified snapshot.  Apply the specified properties a "received" properties
 396  * (which can be overridden by locally-set properties).  If the stream is a
 397  * clone, its origin snapshot must be specified by 'origin'.  The 'force'
 398  * flag will cause the target filesystem to be rolled back or destroyed if
 399  * necessary to receive.
 400  *
 401  * Return 0 on success or an errno on failure.
 402  *
 403  * Note: this interface does not work on dedup'd streams
 404  * (those with DMU_BACKUP_FEATURE_DEDUP).
 405  */
 406 int
 407 lzc_receive(const char *snapname, nvlist_t *props, const char *origin,
 408     boolean_t force, int fd)
 409 {
 410         /*
 411          * The receive ioctl is still legacy, so we need to construct our own
 412          * zfs_cmd_t rather than using zfsc_ioctl().
 413          */
 414         zfs_cmd_t zc = { 0 };
 415         char *atp;
 416         char *packed = NULL;
 417         size_t size;
 418         dmu_replay_record_t drr;
 419         int error;
 420 
 421         ASSERT3S(g_refcount, >, 0);
 422 
 423         /* zc_name is name of containing filesystem */
 424         (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name));
 425         atp = strchr(zc.zc_name, '@');
 426         if (atp == NULL)
 427                 return (EINVAL);
 428         *atp = '\0';
 429 
 430         /* if the fs does not exist, try its parent. */
 431         if (!lzc_exists(zc.zc_name)) {
 432                 char *slashp = strrchr(zc.zc_name, '/');
 433                 if (slashp == NULL)
 434                         return (ENOENT);
 435                 *slashp = '\0';
 436 
 437         }
 438 
 439         /* zc_value is full name of the snapshot to create */
 440         (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
 441 
 442         if (props != NULL) {
 443                 /* zc_nvlist_src is props to set */
 444                 packed = fnvlist_pack(props, &size);
 445                 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
 446                 zc.zc_nvlist_src_size = size;
 447         }
 448 
 449         /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */
 450         if (origin != NULL)
 451                 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string));
 452 
 453         /* zc_begin_record is non-byteswapped BEGIN record */
 454         error = recv_read(fd, &drr, sizeof (drr));
 455         if (error != 0)
 456                 goto out;
 457         zc.zc_begin_record = drr.drr_u.drr_begin;
 458 
 459         /* zc_cookie is fd to read from */
 460         zc.zc_cookie = fd;
 461 
 462         /* zc guid is force flag */
 463         zc.zc_guid = force;
 464 
 465         /* zc_cleanup_fd is unused */
 466         zc.zc_cleanup_fd = -1;
 467 
 468         error = ioctl(g_fd, ZFS_IOC_RECV, &zc);
 469         if (error != 0)
 470                 error = errno;
 471 
 472 out:
 473         if (packed != NULL)
 474                 fnvlist_pack_free(packed, size);
 475         free((void*)(uintptr_t)zc.zc_nvlist_dst);
 476         return (error);
 477 }