1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2013 by Delphix. All rights reserved.
  24  */
  25 
  26 #include <sys/zfs_context.h>
  27 #include <sys/dsl_userhold.h>
  28 #include <sys/dsl_dataset.h>
  29 #include <sys/dsl_destroy.h>
  30 #include <sys/dsl_synctask.h>
  31 #include <sys/dmu_tx.h>
  32 #include <sys/zfs_onexit.h>
  33 #include <sys/dsl_pool.h>
  34 #include <sys/dsl_dir.h>
  35 #include <sys/zfs_ioctl.h>
  36 #include <sys/zap.h>
  37 
  38 typedef struct dsl_dataset_user_hold_arg {
  39         nvlist_t *dduha_holds;
  40         nvlist_t *dduha_errlist;
  41         minor_t dduha_minor;
  42 } dsl_dataset_user_hold_arg_t;
  43 
  44 /*
  45  * If you add new checks here, you may need to add additional checks to the
  46  * "temporary" case in snapshot_check() in dmu_objset.c.
  47  */
  48 int
  49 dsl_dataset_user_hold_check_one(dsl_dataset_t *ds, const char *htag,
  50     boolean_t temphold, dmu_tx_t *tx)
  51 {
  52         dsl_pool_t *dp = dmu_tx_pool(tx);
  53         objset_t *mos = dp->dp_meta_objset;
  54         int error = 0;
  55 
  56         if (strlen(htag) > MAXNAMELEN)
  57                 return (E2BIG);
  58         /* Tempholds have a more restricted length */
  59         if (temphold && strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
  60                 return (E2BIG);
  61 
  62         /* tags must be unique (if ds already exists) */
  63         if (ds != NULL) {
  64                 mutex_enter(&ds->ds_lock);
  65                 if (ds->ds_phys->ds_userrefs_obj != 0) {
  66                         uint64_t value;
  67                         error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj,
  68                             htag, 8, 1, &value);
  69                         if (error == 0)
  70                                 error = SET_ERROR(EEXIST);
  71                         else if (error == ENOENT)
  72                                 error = 0;
  73                 }
  74                 mutex_exit(&ds->ds_lock);
  75         }
  76 
  77         return (error);
  78 }
  79 
  80 static int
  81 dsl_dataset_user_hold_check(void *arg, dmu_tx_t *tx)
  82 {
  83         dsl_dataset_user_hold_arg_t *dduha = arg;
  84         dsl_pool_t *dp = dmu_tx_pool(tx);
  85         nvpair_t *pair;
  86         int rv = 0;
  87 
  88         if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS)
  89                 return (SET_ERROR(ENOTSUP));
  90 
  91         for (pair = nvlist_next_nvpair(dduha->dduha_holds, NULL); pair != NULL;
  92             pair = nvlist_next_nvpair(dduha->dduha_holds, pair)) {
  93                 int error = 0;
  94                 dsl_dataset_t *ds;
  95                 char *htag;
  96 
  97                 /* must be a snapshot */
  98                 if (strchr(nvpair_name(pair), '@') == NULL)
  99                         error = SET_ERROR(EINVAL);
 100 
 101                 if (error == 0)
 102                         error = nvpair_value_string(pair, &htag);
 103                 if (error == 0) {
 104                         error = dsl_dataset_hold(dp,
 105                             nvpair_name(pair), FTAG, &ds);
 106                 }
 107                 if (error == 0) {
 108                         error = dsl_dataset_user_hold_check_one(ds, htag,
 109                             dduha->dduha_minor != 0, tx);
 110                         dsl_dataset_rele(ds, FTAG);
 111                 }
 112 
 113                 if (error != 0) {
 114                         rv = error;
 115                         fnvlist_add_int32(dduha->dduha_errlist,
 116                             nvpair_name(pair), error);
 117                 }
 118         }
 119         return (rv);
 120 }
 121 
 122 void
 123 dsl_dataset_user_hold_sync_one(dsl_dataset_t *ds, const char *htag,
 124     minor_t minor, uint64_t now, dmu_tx_t *tx)
 125 {
 126         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 127         objset_t *mos = dp->dp_meta_objset;
 128         uint64_t zapobj;
 129 
 130         mutex_enter(&ds->ds_lock);
 131         if (ds->ds_phys->ds_userrefs_obj == 0) {
 132                 /*
 133                  * This is the first user hold for this dataset.  Create
 134                  * the userrefs zap object.
 135                  */
 136                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
 137                 zapobj = ds->ds_phys->ds_userrefs_obj =
 138                     zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
 139         } else {
 140                 zapobj = ds->ds_phys->ds_userrefs_obj;
 141         }
 142         ds->ds_userrefs++;
 143         mutex_exit(&ds->ds_lock);
 144 
 145         VERIFY0(zap_add(mos, zapobj, htag, 8, 1, &now, tx));
 146 
 147         if (minor != 0) {
 148                 VERIFY0(dsl_pool_user_hold(dp, ds->ds_object,
 149                     htag, now, tx));
 150                 dsl_register_onexit_hold_cleanup(ds, htag, minor);
 151         }
 152 
 153         spa_history_log_internal_ds(ds, "hold", tx,
 154             "tag=%s temp=%d refs=%llu",
 155             htag, minor != 0, ds->ds_userrefs);
 156 }
 157 
 158 static void
 159 dsl_dataset_user_hold_sync(void *arg, dmu_tx_t *tx)
 160 {
 161         dsl_dataset_user_hold_arg_t *dduha = arg;
 162         dsl_pool_t *dp = dmu_tx_pool(tx);
 163         nvpair_t *pair;
 164         uint64_t now = gethrestime_sec();
 165 
 166         for (pair = nvlist_next_nvpair(dduha->dduha_holds, NULL); pair != NULL;
 167             pair = nvlist_next_nvpair(dduha->dduha_holds, pair)) {
 168                 dsl_dataset_t *ds;
 169                 VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
 170                 dsl_dataset_user_hold_sync_one(ds, fnvpair_value_string(pair),
 171                     dduha->dduha_minor, now, tx);
 172                 dsl_dataset_rele(ds, FTAG);
 173         }
 174 }
 175 
 176 /*
 177  * holds is nvl of snapname -> holdname
 178  * errlist will be filled in with snapname -> error
 179  * if cleanup_minor is not 0, the holds will be temporary, cleaned up
 180  * when the process exits.
 181  *
 182  * if any fails, all will fail.
 183  */
 184 int
 185 dsl_dataset_user_hold(nvlist_t *holds, minor_t cleanup_minor, nvlist_t *errlist)
 186 {
 187         dsl_dataset_user_hold_arg_t dduha;
 188         nvpair_t *pair;
 189 
 190         pair = nvlist_next_nvpair(holds, NULL);
 191         if (pair == NULL)
 192                 return (0);
 193 
 194         dduha.dduha_holds = holds;
 195         dduha.dduha_errlist = errlist;
 196         dduha.dduha_minor = cleanup_minor;
 197 
 198         return (dsl_sync_task(nvpair_name(pair), dsl_dataset_user_hold_check,
 199             dsl_dataset_user_hold_sync, &dduha, fnvlist_num_pairs(holds)));
 200 }
 201 
 202 typedef struct dsl_dataset_user_release_arg {
 203         nvlist_t *ddura_holds;
 204         nvlist_t *ddura_todelete;
 205         nvlist_t *ddura_errlist;
 206 } dsl_dataset_user_release_arg_t;
 207 
 208 static int
 209 dsl_dataset_user_release_check_one(dsl_dataset_t *ds,
 210     nvlist_t *holds, boolean_t *todelete)
 211 {
 212         uint64_t zapobj;
 213         nvpair_t *pair;
 214         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 215         int error;
 216         int numholds = 0;
 217 
 218         *todelete = B_FALSE;
 219 
 220         if (!dsl_dataset_is_snapshot(ds))
 221                 return (SET_ERROR(EINVAL));
 222 
 223         zapobj = ds->ds_phys->ds_userrefs_obj;
 224         if (zapobj == 0)
 225                 return (SET_ERROR(ESRCH));
 226 
 227         for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
 228             pair = nvlist_next_nvpair(holds, pair)) {
 229                 /* Make sure the hold exists */
 230                 uint64_t tmp;
 231                 error = zap_lookup(mos, zapobj, nvpair_name(pair), 8, 1, &tmp);
 232                 if (error == ENOENT)
 233                         error = SET_ERROR(ESRCH);
 234                 if (error != 0)
 235                         return (error);
 236                 numholds++;
 237         }
 238 
 239         if (DS_IS_DEFER_DESTROY(ds) && ds->ds_phys->ds_num_children == 1 &&
 240             ds->ds_userrefs == numholds) {
 241                 /* we need to destroy the snapshot as well */
 242 
 243                 if (dsl_dataset_long_held(ds))
 244                         return (SET_ERROR(EBUSY));
 245                 *todelete = B_TRUE;
 246         }
 247         return (0);
 248 }
 249 
 250 static int
 251 dsl_dataset_user_release_check(void *arg, dmu_tx_t *tx)
 252 {
 253         dsl_dataset_user_release_arg_t *ddura = arg;
 254         dsl_pool_t *dp = dmu_tx_pool(tx);
 255         nvpair_t *pair;
 256         int rv = 0;
 257 
 258         if (!dmu_tx_is_syncing(tx))
 259                 return (0);
 260 
 261         for (pair = nvlist_next_nvpair(ddura->ddura_holds, NULL); pair != NULL;
 262             pair = nvlist_next_nvpair(ddura->ddura_holds, pair)) {
 263                 const char *name = nvpair_name(pair);
 264                 int error;
 265                 dsl_dataset_t *ds;
 266                 nvlist_t *holds;
 267 
 268                 error = nvpair_value_nvlist(pair, &holds);
 269                 if (error != 0)
 270                         return (SET_ERROR(EINVAL));
 271 
 272                 error = dsl_dataset_hold(dp, name, FTAG, &ds);
 273                 if (error == 0) {
 274                         boolean_t deleteme;
 275                         error = dsl_dataset_user_release_check_one(ds,
 276                             holds, &deleteme);
 277                         if (error == 0 && deleteme) {
 278                                 fnvlist_add_boolean(ddura->ddura_todelete,
 279                                     name);
 280                         }
 281                         dsl_dataset_rele(ds, FTAG);
 282                 }
 283                 if (error != 0) {
 284                         if (ddura->ddura_errlist != NULL) {
 285                                 fnvlist_add_int32(ddura->ddura_errlist,
 286                                     name, error);
 287                         }
 288                         rv = error;
 289                 }
 290         }
 291         return (rv);
 292 }
 293 
 294 static void
 295 dsl_dataset_user_release_sync_one(dsl_dataset_t *ds, nvlist_t *holds,
 296     dmu_tx_t *tx)
 297 {
 298         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 299         objset_t *mos = dp->dp_meta_objset;
 300         uint64_t zapobj;
 301         int error;
 302         nvpair_t *pair;
 303 
 304         for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
 305             pair = nvlist_next_nvpair(holds, pair)) {
 306                 ds->ds_userrefs--;
 307                 error = dsl_pool_user_release(dp, ds->ds_object,
 308                     nvpair_name(pair), tx);
 309                 VERIFY(error == 0 || error == ENOENT);
 310                 zapobj = ds->ds_phys->ds_userrefs_obj;
 311                 VERIFY0(zap_remove(mos, zapobj, nvpair_name(pair), tx));
 312 
 313                 spa_history_log_internal_ds(ds, "release", tx,
 314                     "tag=%s refs=%lld", nvpair_name(pair),
 315                     (longlong_t)ds->ds_userrefs);
 316         }
 317 }
 318 
 319 static void
 320 dsl_dataset_user_release_sync(void *arg, dmu_tx_t *tx)
 321 {
 322         dsl_dataset_user_release_arg_t *ddura = arg;
 323         dsl_pool_t *dp = dmu_tx_pool(tx);
 324         nvpair_t *pair;
 325 
 326         for (pair = nvlist_next_nvpair(ddura->ddura_holds, NULL); pair != NULL;
 327             pair = nvlist_next_nvpair(ddura->ddura_holds, pair)) {
 328                 dsl_dataset_t *ds;
 329 
 330                 VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
 331                 dsl_dataset_user_release_sync_one(ds,
 332                     fnvpair_value_nvlist(pair), tx);
 333                 if (nvlist_exists(ddura->ddura_todelete,
 334                     nvpair_name(pair))) {
 335                         ASSERT(ds->ds_userrefs == 0 &&
 336                             ds->ds_phys->ds_num_children == 1 &&
 337                             DS_IS_DEFER_DESTROY(ds));
 338                         dsl_destroy_snapshot_sync_impl(ds, B_FALSE, tx);
 339                 }
 340                 dsl_dataset_rele(ds, FTAG);
 341         }
 342 }
 343 
 344 /*
 345  * holds is nvl of snapname -> { holdname, ... }
 346  * errlist will be filled in with snapname -> error
 347  *
 348  * if any fails, all will fail.
 349  */
 350 int
 351 dsl_dataset_user_release(nvlist_t *holds, nvlist_t *errlist)
 352 {
 353         dsl_dataset_user_release_arg_t ddura;
 354         nvpair_t *pair;
 355         int error;
 356 
 357         pair = nvlist_next_nvpair(holds, NULL);
 358         if (pair == NULL)
 359                 return (0);
 360 
 361         ddura.ddura_holds = holds;
 362         ddura.ddura_errlist = errlist;
 363         ddura.ddura_todelete = fnvlist_alloc();
 364 
 365         error = dsl_sync_task(nvpair_name(pair), dsl_dataset_user_release_check,
 366             dsl_dataset_user_release_sync, &ddura, fnvlist_num_pairs(holds));
 367         fnvlist_free(ddura.ddura_todelete);
 368         return (error);
 369 }
 370 
 371 typedef struct dsl_dataset_user_release_tmp_arg {
 372         uint64_t ddurta_dsobj;
 373         nvlist_t *ddurta_holds;
 374         boolean_t ddurta_deleteme;
 375 } dsl_dataset_user_release_tmp_arg_t;
 376 
 377 static int
 378 dsl_dataset_user_release_tmp_check(void *arg, dmu_tx_t *tx)
 379 {
 380         dsl_dataset_user_release_tmp_arg_t *ddurta = arg;
 381         dsl_pool_t *dp = dmu_tx_pool(tx);
 382         dsl_dataset_t *ds;
 383         int error;
 384 
 385         if (!dmu_tx_is_syncing(tx))
 386                 return (0);
 387 
 388         error = dsl_dataset_hold_obj(dp, ddurta->ddurta_dsobj, FTAG, &ds);
 389         if (error)
 390                 return (error);
 391 
 392         error = dsl_dataset_user_release_check_one(ds,
 393             ddurta->ddurta_holds, &ddurta->ddurta_deleteme);
 394         dsl_dataset_rele(ds, FTAG);
 395         return (error);
 396 }
 397 
 398 static void
 399 dsl_dataset_user_release_tmp_sync(void *arg, dmu_tx_t *tx)
 400 {
 401         dsl_dataset_user_release_tmp_arg_t *ddurta = arg;
 402         dsl_pool_t *dp = dmu_tx_pool(tx);
 403         dsl_dataset_t *ds;
 404 
 405         VERIFY0(dsl_dataset_hold_obj(dp, ddurta->ddurta_dsobj, FTAG, &ds));
 406         dsl_dataset_user_release_sync_one(ds, ddurta->ddurta_holds, tx);
 407         if (ddurta->ddurta_deleteme) {
 408                 ASSERT(ds->ds_userrefs == 0 &&
 409                     ds->ds_phys->ds_num_children == 1 &&
 410                     DS_IS_DEFER_DESTROY(ds));
 411                 dsl_destroy_snapshot_sync_impl(ds, B_FALSE, tx);
 412         }
 413         dsl_dataset_rele(ds, FTAG);
 414 }
 415 
 416 /*
 417  * Called at spa_load time to release a stale temporary user hold.
 418  * Also called by the onexit code.
 419  */
 420 void
 421 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, const char *htag)
 422 {
 423         dsl_dataset_user_release_tmp_arg_t ddurta;
 424         dsl_dataset_t *ds;
 425         int error;
 426 
 427 #ifdef _KERNEL
 428         /* Make sure it is not mounted. */
 429         dsl_pool_config_enter(dp, FTAG);
 430         error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
 431         if (error == 0) {
 432                 char name[MAXNAMELEN];
 433                 dsl_dataset_name(ds, name);
 434                 dsl_dataset_rele(ds, FTAG);
 435                 dsl_pool_config_exit(dp, FTAG);
 436                 (void) zfs_unmount_snap(name);
 437         } else {
 438                 dsl_pool_config_exit(dp, FTAG);
 439         }
 440 #endif
 441 
 442         ddurta.ddurta_dsobj = dsobj;
 443         ddurta.ddurta_holds = fnvlist_alloc();
 444         fnvlist_add_boolean(ddurta.ddurta_holds, htag);
 445 
 446         (void) dsl_sync_task(spa_name(dp->dp_spa),
 447             dsl_dataset_user_release_tmp_check,
 448             dsl_dataset_user_release_tmp_sync, &ddurta, 1);
 449         fnvlist_free(ddurta.ddurta_holds);
 450 }
 451 
 452 typedef struct zfs_hold_cleanup_arg {
 453         char zhca_spaname[MAXNAMELEN];
 454         uint64_t zhca_spa_load_guid;
 455         uint64_t zhca_dsobj;
 456         char zhca_htag[MAXNAMELEN];
 457 } zfs_hold_cleanup_arg_t;
 458 
 459 static void
 460 dsl_dataset_user_release_onexit(void *arg)
 461 {
 462         zfs_hold_cleanup_arg_t *ca = arg;
 463         spa_t *spa;
 464         int error;
 465 
 466         error = spa_open(ca->zhca_spaname, &spa, FTAG);
 467         if (error != 0) {
 468                 zfs_dbgmsg("couldn't release hold on pool=%s ds=%llu tag=%s "
 469                     "because pool is no longer loaded",
 470                     ca->zhca_spaname, ca->zhca_dsobj, ca->zhca_htag);
 471                 return;
 472         }
 473         if (spa_load_guid(spa) != ca->zhca_spa_load_guid) {
 474                 zfs_dbgmsg("couldn't release hold on pool=%s ds=%llu tag=%s "
 475                     "because pool is no longer loaded (guid doesn't match)",
 476                     ca->zhca_spaname, ca->zhca_dsobj, ca->zhca_htag);
 477                 spa_close(spa, FTAG);
 478                 return;
 479         }
 480 
 481         dsl_dataset_user_release_tmp(spa_get_dsl(spa),
 482             ca->zhca_dsobj, ca->zhca_htag);
 483         kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
 484         spa_close(spa, FTAG);
 485 }
 486 
 487 void
 488 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
 489     minor_t minor)
 490 {
 491         zfs_hold_cleanup_arg_t *ca = kmem_alloc(sizeof (*ca), KM_SLEEP);
 492         spa_t *spa = dsl_dataset_get_spa(ds);
 493         (void) strlcpy(ca->zhca_spaname, spa_name(spa),
 494             sizeof (ca->zhca_spaname));
 495         ca->zhca_spa_load_guid = spa_load_guid(spa);
 496         ca->zhca_dsobj = ds->ds_object;
 497         (void) strlcpy(ca->zhca_htag, htag, sizeof (ca->zhca_htag));
 498         VERIFY0(zfs_onexit_add_cb(minor,
 499             dsl_dataset_user_release_onexit, ca, NULL));
 500 }
 501 
 502 int
 503 dsl_dataset_get_holds(const char *dsname, nvlist_t *nvl)
 504 {
 505         dsl_pool_t *dp;
 506         dsl_dataset_t *ds;
 507         int err;
 508 
 509         err = dsl_pool_hold(dsname, FTAG, &dp);
 510         if (err != 0)
 511                 return (err);
 512         err = dsl_dataset_hold(dp, dsname, FTAG, &ds);
 513         if (err != 0) {
 514                 dsl_pool_rele(dp, FTAG);
 515                 return (err);
 516         }
 517 
 518         if (ds->ds_phys->ds_userrefs_obj != 0) {
 519                 zap_attribute_t *za;
 520                 zap_cursor_t zc;
 521 
 522                 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
 523                 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
 524                     ds->ds_phys->ds_userrefs_obj);
 525                     zap_cursor_retrieve(&zc, za) == 0;
 526                     zap_cursor_advance(&zc)) {
 527                         fnvlist_add_uint64(nvl, za->za_name,
 528                             za->za_first_integer);
 529                 }
 530                 zap_cursor_fini(&zc);
 531                 kmem_free(za, sizeof (zap_attribute_t));
 532         }
 533         dsl_dataset_rele(ds, FTAG);
 534         dsl_pool_rele(dp, FTAG);
 535         return (0);
 536 }