Mercurial > illumos > illumos-gate
annotate usr/src/uts/common/fs/zfs/dsl_dataset.c @ 2199:712a788c2dfd
PSARC 2006/388 snapshot -r
6373978 want to take lots of snapshots quickly ('zfs snapshot -r')
| author | ahrens |
|---|---|
| date | Wed, 14 Jun 2006 23:16:39 -0700 |
| parents | 76b439ec3ac1 |
| children | 47efcb3433a7 |
| rev | line source |
|---|---|
| 789 | 1 /* |
| 2 * CDDL HEADER START | |
| 3 * | |
| 4 * The contents of this file are subject to the terms of the | |
| 1544 | 5 * Common Development and Distribution License (the "License"). |
| 6 * You may not use this file except in compliance with the License. | |
| 789 | 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 /* | |
| 1544 | 22 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. |
| 789 | 23 * Use is subject to license terms. |
| 24 */ | |
| 25 | |
| 26 #pragma ident "%Z%%M% %I% %E% SMI" | |
| 27 | |
| 28 #include <sys/dmu_objset.h> | |
| 29 #include <sys/dsl_dataset.h> | |
| 30 #include <sys/dsl_dir.h> | |
| 2082 | 31 #include <sys/dsl_prop.h> |
| 2199 | 32 #include <sys/dsl_synctask.h> |
| 789 | 33 #include <sys/dmu_traverse.h> |
| 34 #include <sys/dmu_tx.h> | |
| 35 #include <sys/arc.h> | |
| 36 #include <sys/zio.h> | |
| 37 #include <sys/zap.h> | |
| 38 #include <sys/unique.h> | |
| 39 #include <sys/zfs_context.h> | |
| 40 | |
| 2199 | 41 static dsl_checkfunc_t dsl_dataset_destroy_begin_check; |
| 42 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync; | |
| 43 static dsl_checkfunc_t dsl_dataset_rollback_check; | |
| 44 static dsl_syncfunc_t dsl_dataset_rollback_sync; | |
| 45 static dsl_checkfunc_t dsl_dataset_destroy_check; | |
| 46 static dsl_syncfunc_t dsl_dataset_destroy_sync; | |
|
1731
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
47 |
| 789 | 48 #define DOS_REF_MAX (1ULL << 62) |
| 49 | |
| 50 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE | |
| 51 | |
| 52 /* | |
| 53 * We use weighted reference counts to express the various forms of exclusion | |
| 54 * between different open modes. A STANDARD open is 1 point, an EXCLUSIVE open | |
| 55 * is DOS_REF_MAX, and a PRIMARY open is little more than half of an EXCLUSIVE. | |
| 56 * This makes the exclusion logic simple: the total refcnt for all opens cannot | |
| 57 * exceed DOS_REF_MAX. For example, EXCLUSIVE opens are exclusive because their | |
| 58 * weight (DOS_REF_MAX) consumes the entire refcnt space. PRIMARY opens consume | |
| 59 * just over half of the refcnt space, so there can't be more than one, but it | |
| 60 * can peacefully coexist with any number of STANDARD opens. | |
| 61 */ | |
| 62 static uint64_t ds_refcnt_weight[DS_MODE_LEVELS] = { | |
| 63 0, /* DOS_MODE_NONE - invalid */ | |
| 64 1, /* DOS_MODE_STANDARD - unlimited number */ | |
| 65 (DOS_REF_MAX >> 1) + 1, /* DOS_MODE_PRIMARY - only one of these */ | |
| 66 DOS_REF_MAX /* DOS_MODE_EXCLUSIVE - no other opens */ | |
| 67 }; | |
| 68 | |
| 69 | |
| 70 void | |
| 71 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) | |
| 72 { | |
| 2082 | 73 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp); |
| 789 | 74 int compressed = BP_GET_PSIZE(bp); |
| 75 int uncompressed = BP_GET_UCSIZE(bp); | |
| 76 | |
| 77 dprintf_bp(bp, "born, ds=%p\n", ds); | |
| 78 | |
| 79 ASSERT(dmu_tx_is_syncing(tx)); | |
| 80 /* It could have been compressed away to nothing */ | |
| 81 if (BP_IS_HOLE(bp)) | |
| 82 return; | |
| 83 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); | |
| 84 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES); | |
| 85 if (ds == NULL) { | |
| 86 /* | |
| 87 * Account for the meta-objset space in its placeholder | |
| 88 * dsl_dir. | |
| 89 */ | |
| 90 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */ | |
| 91 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, | |
| 92 used, compressed, uncompressed, tx); | |
| 93 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); | |
| 94 return; | |
| 95 } | |
| 96 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 97 mutex_enter(&ds->ds_lock); | |
| 98 ds->ds_phys->ds_used_bytes += used; | |
| 99 ds->ds_phys->ds_compressed_bytes += compressed; | |
| 100 ds->ds_phys->ds_uncompressed_bytes += uncompressed; | |
| 101 ds->ds_phys->ds_unique_bytes += used; | |
| 102 mutex_exit(&ds->ds_lock); | |
| 103 dsl_dir_diduse_space(ds->ds_dir, | |
| 104 used, compressed, uncompressed, tx); | |
| 105 } | |
| 106 | |
| 107 void | |
| 108 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) | |
| 109 { | |
| 2082 | 110 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp); |
| 789 | 111 int compressed = BP_GET_PSIZE(bp); |
| 112 int uncompressed = BP_GET_UCSIZE(bp); | |
| 113 | |
| 114 ASSERT(dmu_tx_is_syncing(tx)); | |
| 115 if (BP_IS_HOLE(bp)) | |
| 116 return; | |
| 117 | |
| 118 ASSERT(used > 0); | |
| 119 if (ds == NULL) { | |
| 120 /* | |
| 121 * Account for the meta-objset space in its placeholder | |
| 122 * dataset. | |
| 123 */ | |
| 124 /* XXX this can fail, what do we do when it does? */ | |
| 125 (void) arc_free(NULL, tx->tx_pool->dp_spa, | |
| 126 tx->tx_txg, bp, NULL, NULL, ARC_WAIT); | |
| 127 bzero(bp, sizeof (blkptr_t)); | |
| 128 | |
| 129 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, | |
| 130 -used, -compressed, -uncompressed, tx); | |
| 131 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); | |
| 132 return; | |
| 133 } | |
| 134 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); | |
| 135 | |
| 136 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 137 | |
| 138 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { | |
| 139 dprintf_bp(bp, "freeing: %s", ""); | |
| 140 /* XXX check return code? */ | |
| 141 (void) arc_free(NULL, tx->tx_pool->dp_spa, | |
| 142 tx->tx_txg, bp, NULL, NULL, ARC_WAIT); | |
| 143 | |
| 144 mutex_enter(&ds->ds_lock); | |
| 145 /* XXX unique_bytes is not accurate for head datasets */ | |
| 146 /* ASSERT3U(ds->ds_phys->ds_unique_bytes, >=, used); */ | |
| 147 ds->ds_phys->ds_unique_bytes -= used; | |
| 148 mutex_exit(&ds->ds_lock); | |
| 149 dsl_dir_diduse_space(ds->ds_dir, | |
| 150 -used, -compressed, -uncompressed, tx); | |
| 151 } else { | |
| 152 dprintf_bp(bp, "putting on dead list: %s", ""); | |
| 1544 | 153 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx)); |
| 789 | 154 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ |
| 155 if (ds->ds_phys->ds_prev_snap_obj != 0) { | |
| 156 ASSERT3U(ds->ds_prev->ds_object, ==, | |
| 157 ds->ds_phys->ds_prev_snap_obj); | |
| 158 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); | |
| 159 if (ds->ds_prev->ds_phys->ds_next_snap_obj == | |
| 2082 | 160 ds->ds_object && bp->blk_birth > |
| 789 | 161 ds->ds_prev->ds_phys->ds_prev_snap_txg) { |
| 162 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
| 163 mutex_enter(&ds->ds_prev->ds_lock); | |
| 164 ds->ds_prev->ds_phys->ds_unique_bytes += | |
| 165 used; | |
| 166 mutex_exit(&ds->ds_prev->ds_lock); | |
| 167 } | |
| 168 } | |
| 169 } | |
| 170 bzero(bp, sizeof (blkptr_t)); | |
| 171 mutex_enter(&ds->ds_lock); | |
| 172 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used); | |
| 173 ds->ds_phys->ds_used_bytes -= used; | |
| 174 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); | |
| 175 ds->ds_phys->ds_compressed_bytes -= compressed; | |
| 176 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); | |
| 177 ds->ds_phys->ds_uncompressed_bytes -= uncompressed; | |
| 178 mutex_exit(&ds->ds_lock); | |
| 179 } | |
| 180 | |
| 1544 | 181 uint64_t |
| 182 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) | |
| 789 | 183 { |
| 184 if (ds == NULL) | |
| 1544 | 185 return (0); |
| 789 | 186 /* |
| 187 * The snapshot creation could fail, but that would cause an | |
| 188 * incorrect FALSE return, which would only result in an | |
| 189 * overestimation of the amount of space that an operation would | |
| 190 * consume, which is OK. | |
| 191 * | |
| 192 * There's also a small window where we could miss a pending | |
| 193 * snapshot, because we could set the sync task in the quiescing | |
| 194 * phase. So this should only be used as a guess. | |
| 195 */ | |
| 2199 | 196 return (MAX(ds->ds_phys->ds_prev_snap_txg, ds->ds_trysnap_txg)); |
| 1544 | 197 } |
| 198 | |
| 199 int | |
| 200 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth) | |
| 201 { | |
| 202 return (blk_birth > dsl_dataset_prev_snap_txg(ds)); | |
| 789 | 203 } |
| 204 | |
| 205 /* ARGSUSED */ | |
| 206 static void | |
| 207 dsl_dataset_evict(dmu_buf_t *db, void *dsv) | |
| 208 { | |
| 209 dsl_dataset_t *ds = dsv; | |
| 210 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 211 | |
| 212 /* open_refcount == DOS_REF_MAX when deleting */ | |
| 213 ASSERT(ds->ds_open_refcount == 0 || | |
| 214 ds->ds_open_refcount == DOS_REF_MAX); | |
| 215 | |
| 216 dprintf_ds(ds, "evicting %s\n", ""); | |
| 217 | |
| 218 unique_remove(ds->ds_phys->ds_fsid_guid); | |
| 219 | |
| 220 if (ds->ds_user_ptr != NULL) | |
| 221 ds->ds_user_evict_func(ds, ds->ds_user_ptr); | |
| 222 | |
| 223 if (ds->ds_prev) { | |
| 224 dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds); | |
| 225 ds->ds_prev = NULL; | |
| 226 } | |
| 227 | |
| 228 bplist_close(&ds->ds_deadlist); | |
| 229 dsl_dir_close(ds->ds_dir, ds); | |
| 230 | |
| 231 if (list_link_active(&ds->ds_synced_link)) | |
| 232 list_remove(&dp->dp_synced_objsets, ds); | |
| 233 | |
| 234 kmem_free(ds, sizeof (dsl_dataset_t)); | |
| 235 } | |
| 236 | |
| 1544 | 237 static int |
| 789 | 238 dsl_dataset_get_snapname(dsl_dataset_t *ds) |
| 239 { | |
| 240 dsl_dataset_phys_t *headphys; | |
| 241 int err; | |
| 242 dmu_buf_t *headdbuf; | |
| 243 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 244 objset_t *mos = dp->dp_meta_objset; | |
| 245 | |
| 246 if (ds->ds_snapname[0]) | |
| 1544 | 247 return (0); |
| 789 | 248 if (ds->ds_phys->ds_next_snap_obj == 0) |
| 1544 | 249 return (0); |
| 789 | 250 |
| 1544 | 251 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, |
| 252 FTAG, &headdbuf); | |
| 253 if (err) | |
| 254 return (err); | |
| 789 | 255 headphys = headdbuf->db_data; |
| 256 err = zap_value_search(dp->dp_meta_objset, | |
| 257 headphys->ds_snapnames_zapobj, ds->ds_object, ds->ds_snapname); | |
| 1544 | 258 dmu_buf_rele(headdbuf, FTAG); |
| 259 return (err); | |
| 789 | 260 } |
| 261 | |
| 1544 | 262 int |
| 789 | 263 dsl_dataset_open_obj(dsl_pool_t *dp, uint64_t dsobj, const char *snapname, |
| 1544 | 264 int mode, void *tag, dsl_dataset_t **dsp) |
| 789 | 265 { |
| 266 uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)]; | |
| 267 objset_t *mos = dp->dp_meta_objset; | |
| 268 dmu_buf_t *dbuf; | |
| 269 dsl_dataset_t *ds; | |
| 1544 | 270 int err; |
| 789 | 271 |
| 272 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || | |
| 273 dsl_pool_sync_context(dp)); | |
| 274 | |
| 1544 | 275 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); |
| 276 if (err) | |
| 277 return (err); | |
| 789 | 278 ds = dmu_buf_get_user(dbuf); |
| 279 if (ds == NULL) { | |
| 280 dsl_dataset_t *winner; | |
| 281 | |
| 282 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); | |
| 283 ds->ds_dbuf = dbuf; | |
| 284 ds->ds_object = dsobj; | |
| 285 ds->ds_phys = dbuf->db_data; | |
| 286 | |
| 1544 | 287 err = bplist_open(&ds->ds_deadlist, |
| 789 | 288 mos, ds->ds_phys->ds_deadlist_obj); |
| 1544 | 289 if (err == 0) { |
| 290 err = dsl_dir_open_obj(dp, | |
| 291 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); | |
| 292 } | |
| 293 if (err) { | |
| 294 /* | |
| 295 * we don't really need to close the blist if we | |
| 296 * just opened it. | |
| 297 */ | |
| 298 kmem_free(ds, sizeof (dsl_dataset_t)); | |
| 299 dmu_buf_rele(dbuf, tag); | |
| 300 return (err); | |
| 301 } | |
| 789 | 302 |
| 303 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == dsobj) { | |
| 304 ds->ds_snapname[0] = '\0'; | |
| 305 if (ds->ds_phys->ds_prev_snap_obj) { | |
| 1544 | 306 err = dsl_dataset_open_obj(dp, |
| 789 | 307 ds->ds_phys->ds_prev_snap_obj, NULL, |
| 1544 | 308 DS_MODE_NONE, ds, &ds->ds_prev); |
| 789 | 309 } |
| 310 } else { | |
| 311 if (snapname) { | |
| 312 #ifdef ZFS_DEBUG | |
| 313 dsl_dataset_phys_t *headphys; | |
| 1544 | 314 dmu_buf_t *headdbuf; |
| 315 err = dmu_bonus_hold(mos, | |
| 316 ds->ds_dir->dd_phys->dd_head_dataset_obj, | |
| 317 FTAG, &headdbuf); | |
| 318 if (err == 0) { | |
| 319 headphys = headdbuf->db_data; | |
| 320 uint64_t foundobj; | |
| 321 err = zap_lookup(dp->dp_meta_objset, | |
| 322 headphys->ds_snapnames_zapobj, | |
| 323 snapname, sizeof (foundobj), 1, | |
| 324 &foundobj); | |
| 325 ASSERT3U(foundobj, ==, dsobj); | |
| 326 dmu_buf_rele(headdbuf, FTAG); | |
| 327 } | |
| 789 | 328 #endif |
| 329 (void) strcat(ds->ds_snapname, snapname); | |
| 330 } else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) { | |
| 1544 | 331 err = dsl_dataset_get_snapname(ds); |
| 789 | 332 } |
| 333 } | |
| 334 | |
| 1544 | 335 if (err == 0) { |
| 336 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys, | |
| 337 dsl_dataset_evict); | |
| 338 } | |
| 339 if (err || winner) { | |
| 789 | 340 bplist_close(&ds->ds_deadlist); |
| 341 if (ds->ds_prev) { | |
| 342 dsl_dataset_close(ds->ds_prev, | |
| 343 DS_MODE_NONE, ds); | |
| 344 } | |
| 345 dsl_dir_close(ds->ds_dir, ds); | |
| 346 kmem_free(ds, sizeof (dsl_dataset_t)); | |
| 1544 | 347 if (err) { |
| 348 dmu_buf_rele(dbuf, tag); | |
| 349 return (err); | |
| 350 } | |
| 789 | 351 ds = winner; |
| 352 } else { | |
| 353 uint64_t new = | |
| 354 unique_insert(ds->ds_phys->ds_fsid_guid); | |
| 355 if (new != ds->ds_phys->ds_fsid_guid) { | |
| 356 /* XXX it won't necessarily be synced... */ | |
| 357 ds->ds_phys->ds_fsid_guid = new; | |
| 358 } | |
| 359 } | |
| 360 } | |
| 361 ASSERT3P(ds->ds_dbuf, ==, dbuf); | |
| 362 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); | |
| 363 | |
| 364 mutex_enter(&ds->ds_lock); | |
| 365 if ((DS_MODE_LEVEL(mode) == DS_MODE_PRIMARY && | |
| 2082 | 366 (ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) && |
| 367 !DS_MODE_IS_INCONSISTENT(mode)) || | |
| 789 | 368 (ds->ds_open_refcount + weight > DOS_REF_MAX)) { |
| 369 mutex_exit(&ds->ds_lock); | |
| 370 dsl_dataset_close(ds, DS_MODE_NONE, tag); | |
| 1544 | 371 return (EBUSY); |
| 789 | 372 } |
| 373 ds->ds_open_refcount += weight; | |
| 374 mutex_exit(&ds->ds_lock); | |
| 375 | |
| 1544 | 376 *dsp = ds; |
| 377 return (0); | |
| 789 | 378 } |
| 379 | |
| 380 int | |
| 381 dsl_dataset_open_spa(spa_t *spa, const char *name, int mode, | |
| 382 void *tag, dsl_dataset_t **dsp) | |
| 383 { | |
| 384 dsl_dir_t *dd; | |
| 385 dsl_pool_t *dp; | |
| 386 const char *tail; | |
| 387 uint64_t obj; | |
| 388 dsl_dataset_t *ds = NULL; | |
| 389 int err = 0; | |
| 390 | |
| 1544 | 391 err = dsl_dir_open_spa(spa, name, FTAG, &dd, &tail); |
| 392 if (err) | |
| 393 return (err); | |
| 789 | 394 |
| 395 dp = dd->dd_pool; | |
| 396 obj = dd->dd_phys->dd_head_dataset_obj; | |
| 397 rw_enter(&dp->dp_config_rwlock, RW_READER); | |
| 398 if (obj == 0) { | |
| 399 /* A dataset with no associated objset */ | |
| 400 err = ENOENT; | |
| 401 goto out; | |
| 402 } | |
| 403 | |
| 404 if (tail != NULL) { | |
| 405 objset_t *mos = dp->dp_meta_objset; | |
| 406 | |
| 1544 | 407 err = dsl_dataset_open_obj(dp, obj, NULL, |
| 408 DS_MODE_NONE, tag, &ds); | |
| 409 if (err) | |
| 410 goto out; | |
| 789 | 411 obj = ds->ds_phys->ds_snapnames_zapobj; |
| 412 dsl_dataset_close(ds, DS_MODE_NONE, tag); | |
| 413 ds = NULL; | |
| 414 | |
| 415 if (tail[0] != '@') { | |
| 416 err = ENOENT; | |
| 417 goto out; | |
| 418 } | |
| 419 tail++; | |
| 420 | |
| 421 /* Look for a snapshot */ | |
| 422 if (!DS_MODE_IS_READONLY(mode)) { | |
| 423 err = EROFS; | |
| 424 goto out; | |
| 425 } | |
| 426 dprintf("looking for snapshot '%s'\n", tail); | |
| 427 err = zap_lookup(mos, obj, tail, 8, 1, &obj); | |
| 428 if (err) | |
| 429 goto out; | |
| 430 } | |
| 1544 | 431 err = dsl_dataset_open_obj(dp, obj, tail, mode, tag, &ds); |
| 789 | 432 |
| 433 out: | |
| 434 rw_exit(&dp->dp_config_rwlock); | |
| 435 dsl_dir_close(dd, FTAG); | |
| 436 | |
| 437 ASSERT3U((err == 0), ==, (ds != NULL)); | |
| 438 /* ASSERT(ds == NULL || strcmp(name, ds->ds_name) == 0); */ | |
| 439 | |
| 440 *dsp = ds; | |
| 441 return (err); | |
| 442 } | |
| 443 | |
| 444 int | |
| 445 dsl_dataset_open(const char *name, int mode, void *tag, dsl_dataset_t **dsp) | |
| 446 { | |
| 447 return (dsl_dataset_open_spa(NULL, name, mode, tag, dsp)); | |
| 448 } | |
| 449 | |
| 450 void | |
| 451 dsl_dataset_name(dsl_dataset_t *ds, char *name) | |
| 452 { | |
| 453 if (ds == NULL) { | |
| 454 (void) strcpy(name, "mos"); | |
| 455 } else { | |
| 456 dsl_dir_name(ds->ds_dir, name); | |
| 1544 | 457 VERIFY(0 == dsl_dataset_get_snapname(ds)); |
| 789 | 458 if (ds->ds_snapname[0]) { |
| 459 (void) strcat(name, "@"); | |
| 460 if (!MUTEX_HELD(&ds->ds_lock)) { | |
| 461 /* | |
| 462 * We use a "recursive" mutex so that we | |
| 463 * can call dprintf_ds() with ds_lock held. | |
| 464 */ | |
| 465 mutex_enter(&ds->ds_lock); | |
| 466 (void) strcat(name, ds->ds_snapname); | |
| 467 mutex_exit(&ds->ds_lock); | |
| 468 } else { | |
| 469 (void) strcat(name, ds->ds_snapname); | |
| 470 } | |
| 471 } | |
| 472 } | |
| 473 } | |
| 474 | |
| 475 void | |
| 476 dsl_dataset_close(dsl_dataset_t *ds, int mode, void *tag) | |
| 477 { | |
| 478 uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)]; | |
| 479 mutex_enter(&ds->ds_lock); | |
| 480 ASSERT3U(ds->ds_open_refcount, >=, weight); | |
| 481 ds->ds_open_refcount -= weight; | |
| 482 dprintf_ds(ds, "closing mode %u refcount now 0x%llx\n", | |
| 483 mode, ds->ds_open_refcount); | |
| 484 mutex_exit(&ds->ds_lock); | |
| 485 | |
| 1544 | 486 dmu_buf_rele(ds->ds_dbuf, tag); |
| 789 | 487 } |
| 488 | |
| 489 void | |
| 490 dsl_dataset_create_root(dsl_pool_t *dp, uint64_t *ddobjp, dmu_tx_t *tx) | |
| 491 { | |
| 492 objset_t *mos = dp->dp_meta_objset; | |
| 493 dmu_buf_t *dbuf; | |
| 494 dsl_dataset_phys_t *dsphys; | |
| 495 dsl_dataset_t *ds; | |
| 496 uint64_t dsobj; | |
| 497 dsl_dir_t *dd; | |
| 498 | |
| 499 dsl_dir_create_root(mos, ddobjp, tx); | |
| 1544 | 500 VERIFY(0 == dsl_dir_open_obj(dp, *ddobjp, NULL, FTAG, &dd)); |
| 789 | 501 |
|
928
36d72fe4da29
6349314 dmu_object_type names incorrect for DSL Directories and DSL Datasets
tabriz
parents:
885
diff
changeset
|
502 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, |
|
36d72fe4da29
6349314 dmu_object_type names incorrect for DSL Directories and DSL Datasets
tabriz
parents:
885
diff
changeset
|
503 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); |
| 1544 | 504 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); |
| 789 | 505 dmu_buf_will_dirty(dbuf, tx); |
| 506 dsphys = dbuf->db_data; | |
| 507 dsphys->ds_dir_obj = dd->dd_object; | |
| 508 dsphys->ds_fsid_guid = unique_create(); | |
| 1544 | 509 unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */ |
| 789 | 510 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, |
| 511 sizeof (dsphys->ds_guid)); | |
| 512 dsphys->ds_snapnames_zapobj = | |
|
885
d925b21dba78
6347493 tar of 25K empty directory entries in ZFS takes 30+ seconds ...
ahrens
parents:
789
diff
changeset
|
513 zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx); |
| 789 | 514 dsphys->ds_creation_time = gethrestime_sec(); |
| 515 dsphys->ds_creation_txg = tx->tx_txg; | |
| 516 dsphys->ds_deadlist_obj = | |
| 517 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 1544 | 518 dmu_buf_rele(dbuf, FTAG); |
| 789 | 519 |
| 520 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 521 dd->dd_phys->dd_head_dataset_obj = dsobj; | |
| 522 dsl_dir_close(dd, FTAG); | |
| 523 | |
| 1544 | 524 VERIFY(0 == |
| 525 dsl_dataset_open_obj(dp, dsobj, NULL, DS_MODE_NONE, FTAG, &ds)); | |
| 789 | 526 (void) dmu_objset_create_impl(dp->dp_spa, ds, DMU_OST_ZFS, tx); |
| 527 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 528 } | |
| 529 | |
| 2199 | 530 uint64_t |
| 531 dsl_dataset_create_sync(dsl_dir_t *pdd, | |
| 789 | 532 const char *lastname, dsl_dataset_t *clone_parent, dmu_tx_t *tx) |
| 533 { | |
| 2199 | 534 dsl_pool_t *dp = pdd->dd_pool; |
| 789 | 535 dmu_buf_t *dbuf; |
| 536 dsl_dataset_phys_t *dsphys; | |
| 2199 | 537 uint64_t dsobj, ddobj; |
| 789 | 538 objset_t *mos = dp->dp_meta_objset; |
| 539 dsl_dir_t *dd; | |
| 540 | |
| 2199 | 541 ASSERT(clone_parent == NULL || clone_parent->ds_dir->dd_pool == dp); |
| 542 ASSERT(clone_parent == NULL || | |
| 543 clone_parent->ds_phys->ds_num_children > 0); | |
| 789 | 544 ASSERT(lastname[0] != '@'); |
| 545 ASSERT(dmu_tx_is_syncing(tx)); | |
| 546 | |
| 2199 | 547 ddobj = dsl_dir_create_sync(pdd, lastname, tx); |
| 548 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd)); | |
| 789 | 549 |
|
928
36d72fe4da29
6349314 dmu_object_type names incorrect for DSL Directories and DSL Datasets
tabriz
parents:
885
diff
changeset
|
550 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, |
|
36d72fe4da29
6349314 dmu_object_type names incorrect for DSL Directories and DSL Datasets
tabriz
parents:
885
diff
changeset
|
551 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); |
| 1544 | 552 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); |
| 789 | 553 dmu_buf_will_dirty(dbuf, tx); |
| 554 dsphys = dbuf->db_data; | |
| 555 dsphys->ds_dir_obj = dd->dd_object; | |
| 556 dsphys->ds_fsid_guid = unique_create(); | |
| 557 unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */ | |
| 558 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, | |
| 559 sizeof (dsphys->ds_guid)); | |
| 560 dsphys->ds_snapnames_zapobj = | |
|
885
d925b21dba78
6347493 tar of 25K empty directory entries in ZFS takes 30+ seconds ...
ahrens
parents:
789
diff
changeset
|
561 zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx); |
| 789 | 562 dsphys->ds_creation_time = gethrestime_sec(); |
| 563 dsphys->ds_creation_txg = tx->tx_txg; | |
| 564 dsphys->ds_deadlist_obj = | |
| 565 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 566 if (clone_parent) { | |
| 567 dsphys->ds_prev_snap_obj = clone_parent->ds_object; | |
| 568 dsphys->ds_prev_snap_txg = | |
| 569 clone_parent->ds_phys->ds_creation_txg; | |
| 570 dsphys->ds_used_bytes = | |
| 571 clone_parent->ds_phys->ds_used_bytes; | |
| 572 dsphys->ds_compressed_bytes = | |
| 573 clone_parent->ds_phys->ds_compressed_bytes; | |
| 574 dsphys->ds_uncompressed_bytes = | |
| 575 clone_parent->ds_phys->ds_uncompressed_bytes; | |
| 576 dsphys->ds_bp = clone_parent->ds_phys->ds_bp; | |
| 577 | |
| 578 dmu_buf_will_dirty(clone_parent->ds_dbuf, tx); | |
| 579 clone_parent->ds_phys->ds_num_children++; | |
| 580 | |
| 581 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 582 dd->dd_phys->dd_clone_parent_obj = clone_parent->ds_object; | |
| 583 } | |
| 1544 | 584 dmu_buf_rele(dbuf, FTAG); |
| 789 | 585 |
| 586 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 587 dd->dd_phys->dd_head_dataset_obj = dsobj; | |
| 588 dsl_dir_close(dd, FTAG); | |
| 589 | |
| 2199 | 590 return (dsobj); |
| 591 } | |
| 592 | |
| 593 struct destroyarg { | |
| 594 dsl_sync_task_group_t *dstg; | |
| 595 char *snapname; | |
| 596 void *tag; | |
| 597 char *failed; | |
| 598 }; | |
| 599 | |
| 600 static int | |
| 601 dsl_snapshot_destroy_one(char *name, void *arg) | |
| 602 { | |
| 603 struct destroyarg *da = arg; | |
| 604 dsl_dataset_t *ds; | |
| 605 char *cp; | |
| 606 int err; | |
| 607 | |
| 608 (void) strcat(name, "@"); | |
| 609 (void) strcat(name, da->snapname); | |
| 610 err = dsl_dataset_open(name, | |
| 611 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT, | |
| 612 da->tag, &ds); | |
| 613 cp = strchr(name, '@'); | |
| 614 *cp = '\0'; | |
| 615 if (err == ENOENT) | |
| 616 return (0); | |
| 617 if (err) { | |
| 618 (void) strcpy(da->failed, name); | |
| 619 return (err); | |
| 620 } | |
| 621 | |
| 622 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check, | |
| 623 dsl_dataset_destroy_sync, ds, da->tag, 0); | |
| 789 | 624 return (0); |
| 625 } | |
| 626 | |
| 2199 | 627 /* |
| 628 * Destroy 'snapname' in all descendants of 'fsname'. | |
| 629 */ | |
| 630 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy | |
| 631 int | |
| 632 dsl_snapshots_destroy(char *fsname, char *snapname) | |
| 633 { | |
| 634 int err; | |
| 635 struct destroyarg da; | |
| 636 dsl_sync_task_t *dst; | |
| 637 spa_t *spa; | |
| 638 char *cp; | |
| 639 | |
| 640 cp = strchr(fsname, '/'); | |
| 641 if (cp) { | |
| 642 *cp = '\0'; | |
| 643 err = spa_open(fsname, &spa, FTAG); | |
| 644 *cp = '/'; | |
| 645 } else { | |
| 646 err = spa_open(fsname, &spa, FTAG); | |
| 647 } | |
| 648 if (err) | |
| 649 return (err); | |
| 650 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); | |
| 651 da.snapname = snapname; | |
| 652 da.tag = FTAG; | |
| 653 da.failed = fsname; | |
| 654 | |
| 655 err = dmu_objset_find(fsname, | |
| 656 dsl_snapshot_destroy_one, &da, 0); | |
| 657 | |
| 658 if (err == 0) | |
| 659 err = dsl_sync_task_group_wait(da.dstg); | |
| 660 | |
| 661 for (dst = list_head(&da.dstg->dstg_tasks); dst; | |
| 662 dst = list_next(&da.dstg->dstg_tasks, dst)) { | |
| 663 dsl_dataset_t *ds = dst->dst_arg1; | |
| 664 if (dst->dst_err) { | |
| 665 dsl_dataset_name(ds, fsname); | |
| 666 cp = strchr(fsname, '@'); | |
| 667 *cp = '\0'; | |
| 668 } | |
| 669 /* | |
| 670 * If it was successful, destroy_sync would have | |
| 671 * closed the ds | |
| 672 */ | |
| 673 if (err) | |
| 674 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 675 } | |
| 676 | |
| 677 dsl_sync_task_group_destroy(da.dstg); | |
| 678 spa_close(spa, FTAG); | |
| 679 return (err); | |
| 680 } | |
| 681 | |
| 789 | 682 int |
| 683 dsl_dataset_destroy(const char *name) | |
| 684 { | |
| 685 int err; | |
| 2199 | 686 dsl_sync_task_group_t *dstg; |
| 687 objset_t *os; | |
| 688 dsl_dataset_t *ds; | |
| 789 | 689 dsl_dir_t *dd; |
| 2199 | 690 uint64_t obj; |
| 691 | |
| 692 if (strchr(name, '@')) { | |
| 693 /* Destroying a snapshot is simpler */ | |
| 694 err = dsl_dataset_open(name, | |
| 695 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT, | |
| 696 FTAG, &ds); | |
| 697 if (err) | |
| 698 return (err); | |
| 699 err = dsl_sync_task_do(ds->ds_dir->dd_pool, | |
| 700 dsl_dataset_destroy_check, dsl_dataset_destroy_sync, | |
| 701 ds, FTAG, 0); | |
| 702 if (err) | |
| 703 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 704 return (err); | |
| 705 } | |
| 706 | |
| 707 err = dmu_objset_open(name, DMU_OST_ANY, | |
| 708 DS_MODE_EXCLUSIVE | DS_MODE_INCONSISTENT, &os); | |
| 709 if (err) | |
| 710 return (err); | |
| 711 ds = os->os->os_dsl_dataset; | |
| 712 dd = ds->ds_dir; | |
| 789 | 713 |
| 2199 | 714 /* |
| 715 * Check for errors and mark this ds as inconsistent, in | |
| 716 * case we crash while freeing the objects. | |
| 717 */ | |
| 718 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check, | |
| 719 dsl_dataset_destroy_begin_sync, ds, NULL, 0); | |
| 720 if (err) { | |
| 721 dmu_objset_close(os); | |
| 722 return (err); | |
| 723 } | |
| 724 | |
| 725 /* | |
| 726 * remove the objects in open context, so that we won't | |
| 727 * have too much to do in syncing context. | |
| 728 */ | |
| 729 for (obj = 0; err == 0; | |
| 730 err = dmu_object_next(os, &obj, FALSE)) { | |
| 731 dmu_tx_t *tx = dmu_tx_create(os); | |
| 732 dmu_tx_hold_free(tx, obj, 0, DMU_OBJECT_END); | |
| 733 dmu_tx_hold_bonus(tx, obj); | |
| 734 err = dmu_tx_assign(tx, TXG_WAIT); | |
| 735 if (err) { | |
| 736 /* | |
| 737 * Perhaps there is not enough disk | |
| 738 * space. Just deal with it from | |
| 739 * dsl_dataset_destroy_sync(). | |
| 740 */ | |
| 741 dmu_tx_abort(tx); | |
| 742 continue; | |
| 743 } | |
| 744 VERIFY(0 == dmu_object_free(os, obj, tx)); | |
| 745 dmu_tx_commit(tx); | |
| 746 } | |
| 747 /* Make sure it's not dirty before we finish destroying it. */ | |
| 748 txg_wait_synced(dd->dd_pool, 0); | |
| 749 | |
| 750 dmu_objset_close(os); | |
| 751 if (err != ESRCH) | |
| 752 return (err); | |
| 753 | |
| 754 err = dsl_dataset_open(name, | |
| 755 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT, | |
| 756 FTAG, &ds); | |
| 1544 | 757 if (err) |
| 758 return (err); | |
| 789 | 759 |
| 2199 | 760 err = dsl_dir_open(name, FTAG, &dd, NULL); |
| 761 if (err) { | |
| 762 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 763 return (err); | |
| 789 | 764 } |
| 765 | |
| 2199 | 766 /* |
| 767 * Blow away the dsl_dir + head dataset. | |
| 768 */ | |
| 769 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool); | |
| 770 dsl_sync_task_create(dstg, dsl_dataset_destroy_check, | |
| 771 dsl_dataset_destroy_sync, ds, FTAG, 0); | |
| 772 dsl_sync_task_create(dstg, dsl_dir_destroy_check, | |
| 773 dsl_dir_destroy_sync, dd, FTAG, 0); | |
| 774 err = dsl_sync_task_group_wait(dstg); | |
| 775 dsl_sync_task_group_destroy(dstg); | |
| 776 /* if it is successful, *destroy_sync will close the ds+dd */ | |
| 777 if (err) { | |
| 778 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 779 dsl_dir_close(dd, FTAG); | |
| 780 } | |
| 789 | 781 return (err); |
| 782 } | |
| 783 | |
| 784 int | |
| 2199 | 785 dsl_dataset_rollback(dsl_dataset_t *ds) |
| 789 | 786 { |
| 2199 | 787 ASSERT3U(ds->ds_open_refcount, ==, DOS_REF_MAX); |
| 788 return (dsl_sync_task_do(ds->ds_dir->dd_pool, | |
| 789 dsl_dataset_rollback_check, dsl_dataset_rollback_sync, | |
| 790 ds, NULL, 0)); | |
| 789 | 791 } |
| 792 | |
| 793 void * | |
| 794 dsl_dataset_set_user_ptr(dsl_dataset_t *ds, | |
| 795 void *p, dsl_dataset_evict_func_t func) | |
| 796 { | |
| 797 void *old; | |
| 798 | |
| 799 mutex_enter(&ds->ds_lock); | |
| 800 old = ds->ds_user_ptr; | |
| 801 if (old == NULL) { | |
| 802 ds->ds_user_ptr = p; | |
| 803 ds->ds_user_evict_func = func; | |
| 804 } | |
| 805 mutex_exit(&ds->ds_lock); | |
| 806 return (old); | |
| 807 } | |
| 808 | |
| 809 void * | |
| 810 dsl_dataset_get_user_ptr(dsl_dataset_t *ds) | |
| 811 { | |
| 812 return (ds->ds_user_ptr); | |
| 813 } | |
| 814 | |
| 815 | |
| 816 void | |
| 817 dsl_dataset_get_blkptr(dsl_dataset_t *ds, blkptr_t *bp) | |
| 818 { | |
| 819 *bp = ds->ds_phys->ds_bp; | |
| 820 } | |
| 821 | |
| 822 void | |
| 823 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) | |
| 824 { | |
| 825 ASSERT(dmu_tx_is_syncing(tx)); | |
| 826 /* If it's the meta-objset, set dp_meta_rootbp */ | |
| 827 if (ds == NULL) { | |
| 828 tx->tx_pool->dp_meta_rootbp = *bp; | |
| 829 } else { | |
| 830 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 831 ds->ds_phys->ds_bp = *bp; | |
| 832 } | |
| 833 } | |
| 834 | |
| 835 spa_t * | |
| 836 dsl_dataset_get_spa(dsl_dataset_t *ds) | |
| 837 { | |
| 838 return (ds->ds_dir->dd_pool->dp_spa); | |
| 839 } | |
| 840 | |
| 841 void | |
| 842 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) | |
| 843 { | |
| 844 dsl_pool_t *dp; | |
| 845 | |
| 846 if (ds == NULL) /* this is the meta-objset */ | |
| 847 return; | |
| 848 | |
| 849 ASSERT(ds->ds_user_ptr != NULL); | |
| 850 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); | |
| 851 | |
| 852 dp = ds->ds_dir->dd_pool; | |
| 853 | |
| 854 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) { | |
| 855 /* up the hold count until we can be written out */ | |
| 856 dmu_buf_add_ref(ds->ds_dbuf, ds); | |
| 857 } | |
| 858 } | |
| 859 | |
| 860 struct killarg { | |
| 861 uint64_t *usedp; | |
| 862 uint64_t *compressedp; | |
| 863 uint64_t *uncompressedp; | |
| 864 zio_t *zio; | |
| 865 dmu_tx_t *tx; | |
| 866 }; | |
| 867 | |
| 868 static int | |
| 869 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg) | |
| 870 { | |
| 871 struct killarg *ka = arg; | |
| 872 blkptr_t *bp = &bc->bc_blkptr; | |
| 873 | |
| 874 ASSERT3U(bc->bc_errno, ==, 0); | |
| 875 | |
| 876 /* | |
| 877 * Since this callback is not called concurrently, no lock is | |
| 878 * needed on the accounting values. | |
| 879 */ | |
| 2082 | 880 *ka->usedp += bp_get_dasize(spa, bp); |
| 789 | 881 *ka->compressedp += BP_GET_PSIZE(bp); |
| 882 *ka->uncompressedp += BP_GET_UCSIZE(bp); | |
| 883 /* XXX check for EIO? */ | |
| 884 (void) arc_free(ka->zio, spa, ka->tx->tx_txg, bp, NULL, NULL, | |
| 885 ARC_NOWAIT); | |
| 886 return (0); | |
| 887 } | |
| 888 | |
| 889 /* ARGSUSED */ | |
| 2199 | 890 static int |
| 891 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 789 | 892 { |
| 2199 | 893 dsl_dataset_t *ds = arg1; |
| 789 | 894 |
| 2199 | 895 /* |
| 896 * There must be a previous snapshot. I suppose we could roll | |
| 897 * it back to being empty (and re-initialize the upper (ZPL) | |
| 898 * layer). But for now there's no way to do this via the user | |
| 899 * interface. | |
| 900 */ | |
| 901 if (ds->ds_phys->ds_prev_snap_txg == 0) | |
| 789 | 902 return (EINVAL); |
| 903 | |
| 2199 | 904 /* |
| 905 * This must not be a snapshot. | |
| 906 */ | |
| 907 if (ds->ds_phys->ds_next_snap_obj != 0) | |
| 908 return (EINVAL); | |
| 789 | 909 |
| 910 /* | |
| 911 * If we made changes this txg, traverse_dsl_dataset won't find | |
| 912 * them. Try again. | |
| 913 */ | |
| 2199 | 914 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) |
| 789 | 915 return (EAGAIN); |
| 2199 | 916 |
| 917 return (0); | |
| 918 } | |
| 789 | 919 |
| 2199 | 920 /* ARGSUSED */ |
| 921 static void | |
| 922 dsl_dataset_rollback_sync(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 923 { | |
| 924 dsl_dataset_t *ds = arg1; | |
| 925 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; | |
| 789 | 926 |
| 927 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 928 | |
| 929 /* Zero out the deadlist. */ | |
| 930 bplist_close(&ds->ds_deadlist); | |
| 931 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); | |
| 932 ds->ds_phys->ds_deadlist_obj = | |
| 933 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 1544 | 934 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos, |
| 935 ds->ds_phys->ds_deadlist_obj)); | |
| 789 | 936 |
| 937 { | |
| 938 /* Free blkptrs that we gave birth to */ | |
| 939 zio_t *zio; | |
| 940 uint64_t used = 0, compressed = 0, uncompressed = 0; | |
| 941 struct killarg ka; | |
| 942 | |
| 943 zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL, | |
| 944 ZIO_FLAG_MUSTSUCCEED); | |
| 945 ka.usedp = &used; | |
| 946 ka.compressedp = &compressed; | |
| 947 ka.uncompressedp = &uncompressed; | |
| 948 ka.zio = zio; | |
| 949 ka.tx = tx; | |
| 950 (void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg, | |
| 951 ADVANCE_POST, kill_blkptr, &ka); | |
| 952 (void) zio_wait(zio); | |
| 953 | |
| 2199 | 954 dsl_dir_diduse_space(ds->ds_dir, |
| 789 | 955 -used, -compressed, -uncompressed, tx); |
| 956 } | |
| 957 | |
| 2199 | 958 /* Change our contents to that of the prev snapshot */ |
| 789 | 959 ASSERT3U(ds->ds_prev->ds_object, ==, ds->ds_phys->ds_prev_snap_obj); |
| 960 ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp; | |
| 961 ds->ds_phys->ds_used_bytes = ds->ds_prev->ds_phys->ds_used_bytes; | |
| 962 ds->ds_phys->ds_compressed_bytes = | |
| 963 ds->ds_prev->ds_phys->ds_compressed_bytes; | |
| 964 ds->ds_phys->ds_uncompressed_bytes = | |
| 965 ds->ds_prev->ds_phys->ds_uncompressed_bytes; | |
| 2082 | 966 ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags; |
| 789 | 967 ds->ds_phys->ds_unique_bytes = 0; |
| 968 | |
| 969 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
| 970 ds->ds_prev->ds_phys->ds_unique_bytes = 0; | |
| 971 } | |
| 972 | |
|
1731
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
973 /* ARGSUSED */ |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
974 static int |
| 2199 | 975 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx) |
|
1731
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
976 { |
| 2199 | 977 dsl_dataset_t *ds = arg1; |
|
1731
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
978 |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
979 /* |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
980 * Can't delete a head dataset if there are snapshots of it. |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
981 * (Except if the only snapshots are from the branch we cloned |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
982 * from.) |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
983 */ |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
984 if (ds->ds_prev != NULL && |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
985 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
986 return (EINVAL); |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
987 |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
988 return (0); |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
989 } |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
990 |
| 2199 | 991 /* ARGSUSED */ |
| 992 static void | |
| 993 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 789 | 994 { |
| 2199 | 995 dsl_dataset_t *ds = arg1; |
| 789 | 996 |
| 2199 | 997 /* Mark it as inconsistent on-disk, in case we crash */ |
| 998 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 999 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; | |
| 1000 } | |
| 789 | 1001 |
| 2199 | 1002 /* ARGSUSED */ |
| 1003 static int | |
| 1004 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 1005 { | |
| 1006 dsl_dataset_t *ds = arg1; | |
| 789 | 1007 |
| 1008 /* Can't delete a branch point. */ | |
| 2199 | 1009 if (ds->ds_phys->ds_num_children > 1) |
| 1010 return (EEXIST); | |
| 789 | 1011 |
| 1012 /* | |
| 1013 * Can't delete a head dataset if there are snapshots of it. | |
| 1014 * (Except if the only snapshots are from the branch we cloned | |
| 1015 * from.) | |
| 1016 */ | |
| 1017 if (ds->ds_prev != NULL && | |
| 2199 | 1018 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) |
| 789 | 1019 return (EINVAL); |
| 1020 | |
| 1021 /* | |
| 1022 * If we made changes this txg, traverse_dsl_dataset won't find | |
| 1023 * them. Try again. | |
| 1024 */ | |
| 2199 | 1025 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) |
| 789 | 1026 return (EAGAIN); |
| 2199 | 1027 |
| 1028 /* XXX we should do some i/o error checking... */ | |
| 1029 return (0); | |
| 1030 } | |
| 1031 | |
| 1032 static void | |
| 1033 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) | |
| 1034 { | |
| 1035 dsl_dataset_t *ds = arg1; | |
| 1036 uint64_t used = 0, compressed = 0, uncompressed = 0; | |
| 1037 zio_t *zio; | |
| 1038 int err; | |
| 1039 int after_branch_point = FALSE; | |
| 1040 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 1041 objset_t *mos = dp->dp_meta_objset; | |
| 1042 dsl_dataset_t *ds_prev = NULL; | |
| 1043 uint64_t obj; | |
| 1044 | |
| 1045 ASSERT3U(ds->ds_open_refcount, ==, DOS_REF_MAX); | |
| 1046 ASSERT3U(ds->ds_phys->ds_num_children, <=, 1); | |
| 1047 ASSERT(ds->ds_prev == NULL || | |
| 1048 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object); | |
| 1049 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); | |
| 1050 | |
| 1051 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); | |
| 1052 | |
| 1053 obj = ds->ds_object; | |
| 789 | 1054 |
| 1055 if (ds->ds_phys->ds_prev_snap_obj != 0) { | |
| 1056 if (ds->ds_prev) { | |
| 1057 ds_prev = ds->ds_prev; | |
| 1058 } else { | |
| 2199 | 1059 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 789 | 1060 ds->ds_phys->ds_prev_snap_obj, NULL, |
| 2199 | 1061 DS_MODE_NONE, FTAG, &ds_prev)); |
| 789 | 1062 } |
| 1063 after_branch_point = | |
| 1064 (ds_prev->ds_phys->ds_next_snap_obj != obj); | |
| 1065 | |
| 1066 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); | |
| 1067 if (after_branch_point && | |
| 1068 ds->ds_phys->ds_next_snap_obj == 0) { | |
| 1069 /* This clone is toast. */ | |
| 1070 ASSERT(ds_prev->ds_phys->ds_num_children > 1); | |
| 1071 ds_prev->ds_phys->ds_num_children--; | |
| 1072 } else if (!after_branch_point) { | |
| 1073 ds_prev->ds_phys->ds_next_snap_obj = | |
| 1074 ds->ds_phys->ds_next_snap_obj; | |
| 1075 } | |
| 1076 } | |
| 1077 | |
| 1078 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
| 1079 | |
| 1080 if (ds->ds_phys->ds_next_snap_obj != 0) { | |
| 2199 | 1081 blkptr_t bp; |
| 789 | 1082 dsl_dataset_t *ds_next; |
| 1083 uint64_t itor = 0; | |
| 1084 | |
| 1085 spa_scrub_restart(dp->dp_spa, tx->tx_txg); | |
| 1086 | |
| 2199 | 1087 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 1544 | 1088 ds->ds_phys->ds_next_snap_obj, NULL, |
| 1089 DS_MODE_NONE, FTAG, &ds_next)); | |
| 789 | 1090 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); |
| 1091 | |
| 1092 dmu_buf_will_dirty(ds_next->ds_dbuf, tx); | |
| 1093 ds_next->ds_phys->ds_prev_snap_obj = | |
| 1094 ds->ds_phys->ds_prev_snap_obj; | |
| 1095 ds_next->ds_phys->ds_prev_snap_txg = | |
| 1096 ds->ds_phys->ds_prev_snap_txg; | |
| 1097 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, | |
| 1098 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); | |
| 1099 | |
| 1100 /* | |
| 1101 * Transfer to our deadlist (which will become next's | |
| 1102 * new deadlist) any entries from next's current | |
| 1103 * deadlist which were born before prev, and free the | |
| 1104 * other entries. | |
| 1105 * | |
| 1106 * XXX we're doing this long task with the config lock held | |
| 1107 */ | |
| 1108 while (bplist_iterate(&ds_next->ds_deadlist, &itor, | |
| 1109 &bp) == 0) { | |
| 1110 if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) { | |
| 1544 | 1111 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, |
| 1112 &bp, tx)); | |
| 789 | 1113 if (ds_prev && !after_branch_point && |
| 1114 bp.blk_birth > | |
| 1115 ds_prev->ds_phys->ds_prev_snap_txg) { | |
| 1116 ds_prev->ds_phys->ds_unique_bytes += | |
| 2082 | 1117 bp_get_dasize(dp->dp_spa, &bp); |
| 789 | 1118 } |
| 1119 } else { | |
| 2082 | 1120 used += bp_get_dasize(dp->dp_spa, &bp); |
| 789 | 1121 compressed += BP_GET_PSIZE(&bp); |
| 1122 uncompressed += BP_GET_UCSIZE(&bp); | |
| 1123 /* XXX check return value? */ | |
| 1124 (void) arc_free(zio, dp->dp_spa, tx->tx_txg, | |
| 1125 &bp, NULL, NULL, ARC_NOWAIT); | |
| 1126 } | |
| 1127 } | |
| 1128 | |
| 1129 /* free next's deadlist */ | |
| 1130 bplist_close(&ds_next->ds_deadlist); | |
| 1131 bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx); | |
| 1132 | |
| 1133 /* set next's deadlist to our deadlist */ | |
| 1134 ds_next->ds_phys->ds_deadlist_obj = | |
| 1135 ds->ds_phys->ds_deadlist_obj; | |
| 1544 | 1136 VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos, |
| 1137 ds_next->ds_phys->ds_deadlist_obj)); | |
| 789 | 1138 ds->ds_phys->ds_deadlist_obj = 0; |
| 1139 | |
| 1140 if (ds_next->ds_phys->ds_next_snap_obj != 0) { | |
| 1141 /* | |
| 1142 * Update next's unique to include blocks which | |
| 1143 * were previously shared by only this snapshot | |
| 1144 * and it. Those blocks will be born after the | |
| 1145 * prev snap and before this snap, and will have | |
| 1146 * died after the next snap and before the one | |
| 1147 * after that (ie. be on the snap after next's | |
| 1148 * deadlist). | |
| 1149 * | |
| 1150 * XXX we're doing this long task with the | |
| 1151 * config lock held | |
| 1152 */ | |
| 1153 dsl_dataset_t *ds_after_next; | |
| 1154 | |
| 2199 | 1155 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 789 | 1156 ds_next->ds_phys->ds_next_snap_obj, NULL, |
| 1544 | 1157 DS_MODE_NONE, FTAG, &ds_after_next)); |
| 789 | 1158 itor = 0; |
| 1159 while (bplist_iterate(&ds_after_next->ds_deadlist, | |
| 1160 &itor, &bp) == 0) { | |
| 1161 if (bp.blk_birth > | |
| 1162 ds->ds_phys->ds_prev_snap_txg && | |
| 1163 bp.blk_birth <= | |
| 1164 ds->ds_phys->ds_creation_txg) { | |
| 1165 ds_next->ds_phys->ds_unique_bytes += | |
| 2082 | 1166 bp_get_dasize(dp->dp_spa, &bp); |
| 789 | 1167 } |
| 1168 } | |
| 1169 | |
| 1170 dsl_dataset_close(ds_after_next, DS_MODE_NONE, FTAG); | |
| 1171 ASSERT3P(ds_next->ds_prev, ==, NULL); | |
| 1172 } else { | |
| 1173 /* | |
| 1174 * It would be nice to update the head dataset's | |
| 1175 * unique. To do so we would have to traverse | |
| 1176 * it for blocks born after ds_prev, which is | |
| 1177 * pretty expensive just to maintain something | |
| 1178 * for debugging purposes. | |
| 1179 */ | |
| 1180 ASSERT3P(ds_next->ds_prev, ==, ds); | |
| 1181 dsl_dataset_close(ds_next->ds_prev, DS_MODE_NONE, | |
| 1182 ds_next); | |
| 1183 if (ds_prev) { | |
| 2199 | 1184 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 1544 | 1185 ds->ds_phys->ds_prev_snap_obj, NULL, |
| 1186 DS_MODE_NONE, ds_next, &ds_next->ds_prev)); | |
| 789 | 1187 } else { |
| 1188 ds_next->ds_prev = NULL; | |
| 1189 } | |
| 1190 } | |
| 1191 dsl_dataset_close(ds_next, DS_MODE_NONE, FTAG); | |
| 1192 | |
| 1193 /* | |
| 1194 * NB: unique_bytes is not accurate for head objsets | |
| 1195 * because we don't update it when we delete the most | |
| 1196 * recent snapshot -- see above comment. | |
| 1197 */ | |
| 1198 ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes); | |
| 1199 } else { | |
| 1200 /* | |
| 1201 * There's no next snapshot, so this is a head dataset. | |
| 1202 * Destroy the deadlist. Unless it's a clone, the | |
| 1203 * deadlist should be empty. (If it's a clone, it's | |
| 1204 * safe to ignore the deadlist contents.) | |
| 1205 */ | |
| 1206 struct killarg ka; | |
| 1207 | |
| 1208 ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist)); | |
| 1209 bplist_close(&ds->ds_deadlist); | |
| 1210 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); | |
| 1211 ds->ds_phys->ds_deadlist_obj = 0; | |
| 1212 | |
| 1213 /* | |
| 1214 * Free everything that we point to (that's born after | |
| 1215 * the previous snapshot, if we are a clone) | |
| 1216 * | |
| 1217 * XXX we're doing this long task with the config lock held | |
| 1218 */ | |
| 1219 ka.usedp = &used; | |
| 1220 ka.compressedp = &compressed; | |
| 1221 ka.uncompressedp = &uncompressed; | |
| 1222 ka.zio = zio; | |
| 1223 ka.tx = tx; | |
| 1224 err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg, | |
| 1225 ADVANCE_POST, kill_blkptr, &ka); | |
| 1226 ASSERT3U(err, ==, 0); | |
| 1227 } | |
| 1228 | |
| 1229 err = zio_wait(zio); | |
| 1230 ASSERT3U(err, ==, 0); | |
| 1231 | |
| 2199 | 1232 dsl_dir_diduse_space(ds->ds_dir, -used, -compressed, -uncompressed, tx); |
| 789 | 1233 |
| 1234 if (ds->ds_phys->ds_snapnames_zapobj) { | |
| 1235 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx); | |
| 1236 ASSERT(err == 0); | |
| 1237 } | |
| 1238 | |
| 2199 | 1239 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) { |
| 789 | 1240 /* Erase the link in the dataset */ |
| 2199 | 1241 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); |
| 1242 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0; | |
| 789 | 1243 /* |
| 1244 * dsl_dir_sync_destroy() called us, they'll destroy | |
| 1245 * the dataset. | |
| 1246 */ | |
| 1247 } else { | |
| 1248 /* remove from snapshot namespace */ | |
| 1249 dsl_dataset_t *ds_head; | |
| 2199 | 1250 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 1251 ds->ds_dir->dd_phys->dd_head_dataset_obj, NULL, | |
| 1544 | 1252 DS_MODE_NONE, FTAG, &ds_head)); |
| 789 | 1253 #ifdef ZFS_DEBUG |
| 1254 { | |
| 1255 uint64_t val; | |
| 1256 err = zap_lookup(mos, | |
| 1257 ds_head->ds_phys->ds_snapnames_zapobj, | |
| 2199 | 1258 ds->ds_snapname, 8, 1, &val); |
| 789 | 1259 ASSERT3U(err, ==, 0); |
| 1260 ASSERT3U(val, ==, obj); | |
| 1261 } | |
| 1262 #endif | |
| 1263 err = zap_remove(mos, ds_head->ds_phys->ds_snapnames_zapobj, | |
| 2199 | 1264 ds->ds_snapname, tx); |
| 789 | 1265 ASSERT(err == 0); |
| 1266 dsl_dataset_close(ds_head, DS_MODE_NONE, FTAG); | |
| 1267 } | |
| 1268 | |
| 1269 if (ds_prev && ds->ds_prev != ds_prev) | |
| 1270 dsl_dataset_close(ds_prev, DS_MODE_NONE, FTAG); | |
| 1271 | |
| 2199 | 1272 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, tag); |
| 1273 VERIFY(0 == dmu_object_free(mos, obj, tx)); | |
| 1274 } | |
| 1275 | |
| 1276 /* ARGSUSED */ | |
| 1277 int | |
| 1278 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 1279 { | |
| 1280 objset_t *os = arg1; | |
| 1281 dsl_dataset_t *ds = os->os->os_dsl_dataset; | |
| 1282 const char *snapname = arg2; | |
| 1283 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; | |
| 1284 int err; | |
| 1285 uint64_t value; | |
| 789 | 1286 |
| 1287 /* | |
| 2199 | 1288 * We don't allow multiple snapshots of the same txg. If there |
| 1289 * is already one, try again. | |
| 1290 */ | |
| 1291 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) | |
| 1292 return (EAGAIN); | |
| 1293 | |
| 1294 /* | |
| 1295 * Check for conflicting name snapshot name. | |
| 789 | 1296 */ |
| 2199 | 1297 err = zap_lookup(mos, ds->ds_phys->ds_snapnames_zapobj, |
| 1298 snapname, 8, 1, &value); | |
| 1299 if (err == 0) | |
| 1300 return (EEXIST); | |
| 1301 if (err != ENOENT) | |
| 1302 return (err); | |
| 789 | 1303 |
| 2199 | 1304 ds->ds_trysnap_txg = tx->tx_txg; |
| 789 | 1305 return (0); |
| 1306 } | |
| 1307 | |
| 2199 | 1308 void |
| 1309 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 789 | 1310 { |
| 2199 | 1311 objset_t *os = arg1; |
| 1312 dsl_dataset_t *ds = os->os->os_dsl_dataset; | |
| 1313 const char *snapname = arg2; | |
| 1314 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 789 | 1315 dmu_buf_t *dbuf; |
| 1316 dsl_dataset_phys_t *dsphys; | |
| 2199 | 1317 uint64_t dsobj; |
| 789 | 1318 objset_t *mos = dp->dp_meta_objset; |
| 1319 int err; | |
| 1320 | |
| 1321 spa_scrub_restart(dp->dp_spa, tx->tx_txg); | |
| 2199 | 1322 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); |
| 789 | 1323 |
|
928
36d72fe4da29
6349314 dmu_object_type names incorrect for DSL Directories and DSL Datasets
tabriz
parents:
885
diff
changeset
|
1324 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, |
|
36d72fe4da29
6349314 dmu_object_type names incorrect for DSL Directories and DSL Datasets
tabriz
parents:
885
diff
changeset
|
1325 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); |
| 1544 | 1326 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); |
| 789 | 1327 dmu_buf_will_dirty(dbuf, tx); |
| 1328 dsphys = dbuf->db_data; | |
| 2199 | 1329 dsphys->ds_dir_obj = ds->ds_dir->dd_object; |
| 789 | 1330 dsphys->ds_fsid_guid = unique_create(); |
| 1331 unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */ | |
| 1332 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, | |
| 1333 sizeof (dsphys->ds_guid)); | |
| 1334 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; | |
| 1335 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; | |
| 1336 dsphys->ds_next_snap_obj = ds->ds_object; | |
| 1337 dsphys->ds_num_children = 1; | |
| 1338 dsphys->ds_creation_time = gethrestime_sec(); | |
| 1339 dsphys->ds_creation_txg = tx->tx_txg; | |
| 1340 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; | |
| 1341 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes; | |
| 1342 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; | |
| 1343 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; | |
| 2082 | 1344 dsphys->ds_flags = ds->ds_phys->ds_flags; |
| 789 | 1345 dsphys->ds_bp = ds->ds_phys->ds_bp; |
| 1544 | 1346 dmu_buf_rele(dbuf, FTAG); |
| 789 | 1347 |
| 2199 | 1348 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); |
| 1349 if (ds->ds_prev) { | |
| 1350 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == | |
| 789 | 1351 ds->ds_object || |
| 2199 | 1352 ds->ds_prev->ds_phys->ds_num_children > 1); |
| 1353 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { | |
| 1354 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
| 789 | 1355 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, |
| 2199 | 1356 ds->ds_prev->ds_phys->ds_creation_txg); |
| 1357 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; | |
| 789 | 1358 } |
| 1359 } | |
| 1360 | |
| 1361 bplist_close(&ds->ds_deadlist); | |
| 1362 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 1363 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, dsphys->ds_creation_txg); | |
| 1364 ds->ds_phys->ds_prev_snap_obj = dsobj; | |
| 1365 ds->ds_phys->ds_prev_snap_txg = dsphys->ds_creation_txg; | |
| 1366 ds->ds_phys->ds_unique_bytes = 0; | |
| 1367 ds->ds_phys->ds_deadlist_obj = | |
| 1368 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 1544 | 1369 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos, |
| 1370 ds->ds_phys->ds_deadlist_obj)); | |
| 789 | 1371 |
| 1372 dprintf("snap '%s' -> obj %llu\n", snapname, dsobj); | |
| 1373 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, | |
| 1374 snapname, 8, 1, &dsobj, tx); | |
| 1375 ASSERT(err == 0); | |
| 1376 | |
| 1377 if (ds->ds_prev) | |
| 1378 dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds); | |
| 1544 | 1379 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 1380 ds->ds_phys->ds_prev_snap_obj, snapname, | |
| 1381 DS_MODE_NONE, ds, &ds->ds_prev)); | |
| 789 | 1382 } |
| 1383 | |
| 1384 void | |
| 1385 dsl_dataset_sync(dsl_dataset_t *ds, dmu_tx_t *tx) | |
| 1386 { | |
| 1387 ASSERT(dmu_tx_is_syncing(tx)); | |
| 1388 ASSERT(ds->ds_user_ptr != NULL); | |
| 1389 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); | |
| 1390 | |
| 1391 dmu_objset_sync(ds->ds_user_ptr, tx); | |
| 1392 dsl_dir_dirty(ds->ds_dir, tx); | |
| 1393 bplist_close(&ds->ds_deadlist); | |
| 1394 | |
| 1544 | 1395 dmu_buf_rele(ds->ds_dbuf, ds); |
| 789 | 1396 } |
| 1397 | |
| 1398 void | |
| 1399 dsl_dataset_stats(dsl_dataset_t *ds, dmu_objset_stats_t *dds) | |
| 1400 { | |
| 1401 /* fill in properties crap */ | |
| 1402 dsl_dir_stats(ds->ds_dir, dds); | |
| 1403 | |
| 1404 if (ds->ds_phys->ds_num_children != 0) { | |
| 1405 dds->dds_is_snapshot = TRUE; | |
| 1406 dds->dds_num_clones = ds->ds_phys->ds_num_children - 1; | |
| 1407 } | |
| 1408 | |
| 2082 | 1409 dds->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; |
| 789 | 1410 dds->dds_last_txg = ds->ds_phys->ds_bp.blk_birth; |
| 1411 | |
| 1412 dds->dds_objects_used = ds->ds_phys->ds_bp.blk_fill; | |
| 1413 dds->dds_objects_avail = DN_MAX_OBJECT - dds->dds_objects_used; | |
| 1414 | |
| 1415 /* We override the dataset's creation time... they should be the same */ | |
| 1416 dds->dds_creation_time = ds->ds_phys->ds_creation_time; | |
| 1417 dds->dds_creation_txg = ds->ds_phys->ds_creation_txg; | |
| 1418 dds->dds_space_refd = ds->ds_phys->ds_used_bytes; | |
| 1419 dds->dds_fsid_guid = ds->ds_phys->ds_fsid_guid; | |
| 1420 | |
| 1421 if (ds->ds_phys->ds_next_snap_obj) { | |
| 1422 /* | |
| 1423 * This is a snapshot; override the dd's space used with | |
| 1424 * our unique space | |
| 1425 */ | |
| 1426 dds->dds_space_used = ds->ds_phys->ds_unique_bytes; | |
| 1427 dds->dds_compressed_bytes = | |
| 1428 ds->ds_phys->ds_compressed_bytes; | |
| 1429 dds->dds_uncompressed_bytes = | |
| 1430 ds->ds_phys->ds_uncompressed_bytes; | |
| 1431 } | |
| 1432 } | |
| 1433 | |
| 1434 dsl_pool_t * | |
| 1435 dsl_dataset_pool(dsl_dataset_t *ds) | |
| 1436 { | |
| 1437 return (ds->ds_dir->dd_pool); | |
| 1438 } | |
| 1439 | |
| 2199 | 1440 /* ARGSUSED */ |
| 789 | 1441 static int |
| 2199 | 1442 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) |
| 789 | 1443 { |
| 2199 | 1444 dsl_dataset_t *ds = arg1; |
| 1445 char *newsnapname = arg2; | |
| 1446 dsl_dir_t *dd = ds->ds_dir; | |
| 789 | 1447 objset_t *mos = dd->dd_pool->dp_meta_objset; |
| 2199 | 1448 dsl_dataset_t *hds; |
| 1449 uint64_t val; | |
| 789 | 1450 int err; |
| 1451 | |
| 2199 | 1452 err = dsl_dataset_open_obj(dd->dd_pool, |
| 1453 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds); | |
| 789 | 1454 if (err) |
| 1455 return (err); | |
| 1456 | |
| 2199 | 1457 /* new name better not be in use */ |
| 1458 err = zap_lookup(mos, hds->ds_phys->ds_snapnames_zapobj, | |
| 1459 newsnapname, 8, 1, &val); | |
| 1460 dsl_dataset_close(hds, DS_MODE_NONE, FTAG); | |
| 789 | 1461 |
| 2199 | 1462 if (err == 0) |
| 1463 err = EEXIST; | |
| 1464 else if (err == ENOENT) | |
| 1465 err = 0; | |
| 1466 return (err); | |
| 1467 } | |
| 789 | 1468 |
| 2199 | 1469 static void |
| 1470 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 1471 { | |
| 1472 dsl_dataset_t *ds = arg1; | |
| 1473 char *newsnapname = arg2; | |
| 1474 dsl_dir_t *dd = ds->ds_dir; | |
| 1475 objset_t *mos = dd->dd_pool->dp_meta_objset; | |
| 1476 dsl_dataset_t *hds; | |
| 1477 int err; | |
| 789 | 1478 |
| 2199 | 1479 ASSERT(ds->ds_phys->ds_next_snap_obj != 0); |
| 789 | 1480 |
| 2199 | 1481 VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool, |
| 1482 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds)); | |
| 789 | 1483 |
| 2199 | 1484 VERIFY(0 == dsl_dataset_get_snapname(ds)); |
| 1485 err = zap_remove(mos, hds->ds_phys->ds_snapnames_zapobj, | |
| 1486 ds->ds_snapname, tx); | |
| 789 | 1487 ASSERT3U(err, ==, 0); |
| 2199 | 1488 mutex_enter(&ds->ds_lock); |
| 1489 (void) strcpy(ds->ds_snapname, newsnapname); | |
| 1490 mutex_exit(&ds->ds_lock); | |
| 1491 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj, | |
| 1492 ds->ds_snapname, 8, 1, &ds->ds_object, tx); | |
| 789 | 1493 ASSERT3U(err, ==, 0); |
| 1494 | |
| 2199 | 1495 dsl_dataset_close(hds, DS_MODE_NONE, FTAG); |
| 789 | 1496 } |
| 1497 | |
| 1498 #pragma weak dmu_objset_rename = dsl_dataset_rename | |
| 1499 int | |
| 2199 | 1500 dsl_dataset_rename(const char *oldname, const char *newname) |
| 789 | 1501 { |
| 1502 dsl_dir_t *dd; | |
| 2199 | 1503 dsl_dataset_t *ds; |
| 789 | 1504 const char *tail; |
| 1505 int err; | |
| 1506 | |
| 2199 | 1507 err = dsl_dir_open(oldname, FTAG, &dd, &tail); |
| 1544 | 1508 if (err) |
| 1509 return (err); | |
| 789 | 1510 if (tail == NULL) { |
| 2199 | 1511 err = dsl_dir_rename(dd, newname); |
| 789 | 1512 dsl_dir_close(dd, FTAG); |
| 1513 return (err); | |
| 1514 } | |
| 1515 if (tail[0] != '@') { | |
| 1516 /* the name ended in a nonexistant component */ | |
| 1517 dsl_dir_close(dd, FTAG); | |
| 1518 return (ENOENT); | |
| 1519 } | |
| 1520 | |
| 2199 | 1521 dsl_dir_close(dd, FTAG); |
| 1522 | |
| 1523 /* new name must be snapshot in same filesystem */ | |
| 1524 tail = strchr(newname, '@'); | |
| 1525 if (tail == NULL) | |
| 1526 return (EINVAL); | |
| 1527 tail++; | |
| 1528 if (strncmp(oldname, newname, tail - newname) != 0) | |
| 1529 return (EXDEV); | |
| 789 | 1530 |
| 2199 | 1531 err = dsl_dataset_open(oldname, |
| 1532 DS_MODE_READONLY | DS_MODE_STANDARD, FTAG, &ds); | |
| 1533 if (err) | |
| 1534 return (err); | |
| 1535 | |
| 1536 err = dsl_sync_task_do(ds->ds_dir->dd_pool, | |
| 1537 dsl_dataset_snapshot_rename_check, | |
| 1538 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1); | |
| 1539 | |
| 1540 dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG); | |
| 1541 | |
| 789 | 1542 return (err); |
| 1543 } | |
| 2082 | 1544 |
| 2199 | 1545 struct promotearg { |
| 1546 uint64_t used, comp, uncomp, unique; | |
| 1547 uint64_t newnext_obj, snapnames_obj; | |
| 1548 }; | |
| 1549 | |
| 2082 | 1550 static int |
| 2199 | 1551 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) |
| 2082 | 1552 { |
| 2199 | 1553 dsl_dataset_t *hds = arg1; |
| 1554 struct promotearg *pa = arg2; | |
| 1555 dsl_dir_t *dd = hds->ds_dir; | |
| 1556 dsl_pool_t *dp = hds->ds_dir->dd_pool; | |
| 2082 | 1557 dsl_dir_t *pdd = NULL; |
| 1558 dsl_dataset_t *ds = NULL; | |
| 1559 dsl_dataset_t *pivot_ds = NULL; | |
| 1560 dsl_dataset_t *newnext_ds = NULL; | |
| 1561 int err; | |
| 1562 char *name = NULL; | |
| 2199 | 1563 uint64_t itor = 0; |
| 2082 | 1564 blkptr_t bp; |
| 1565 | |
| 2199 | 1566 bzero(pa, sizeof (*pa)); |
| 1567 | |
| 2082 | 1568 /* Check that it is a clone */ |
| 1569 if (dd->dd_phys->dd_clone_parent_obj == 0) | |
| 1570 return (EINVAL); | |
| 1571 | |
| 2199 | 1572 /* Since this is so expensive, don't do the preliminary check */ |
| 1573 if (!dmu_tx_is_syncing(tx)) | |
| 1574 return (0); | |
| 1575 | |
| 1576 if (err = dsl_dataset_open_obj(dp, | |
| 2082 | 1577 dd->dd_phys->dd_clone_parent_obj, |
| 1578 NULL, DS_MODE_EXCLUSIVE, FTAG, &pivot_ds)) | |
| 1579 goto out; | |
| 1580 pdd = pivot_ds->ds_dir; | |
| 2199 | 1581 |
| 1582 { | |
| 1583 dsl_dataset_t *phds; | |
| 1584 if (err = dsl_dataset_open_obj(dd->dd_pool, | |
| 1585 pdd->dd_phys->dd_head_dataset_obj, | |
| 1586 NULL, DS_MODE_NONE, FTAG, &phds)) | |
| 1587 goto out; | |
| 1588 pa->snapnames_obj = phds->ds_phys->ds_snapnames_zapobj; | |
| 1589 dsl_dataset_close(phds, DS_MODE_NONE, FTAG); | |
| 1590 } | |
| 2082 | 1591 |
| 1592 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) { | |
| 1593 err = EXDEV; | |
| 1594 goto out; | |
| 1595 } | |
| 1596 | |
| 1597 /* find pivot point's new next ds */ | |
| 1598 VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool, hds->ds_object, | |
| 1599 NULL, DS_MODE_NONE, FTAG, &newnext_ds)); | |
| 1600 while (newnext_ds->ds_phys->ds_prev_snap_obj != pivot_ds->ds_object) { | |
| 1601 dsl_dataset_t *prev; | |
| 1602 | |
| 1603 if (err = dsl_dataset_open_obj(dd->dd_pool, | |
| 2199 | 1604 newnext_ds->ds_phys->ds_prev_snap_obj, |
| 1605 NULL, DS_MODE_NONE, FTAG, &prev)) | |
| 2082 | 1606 goto out; |
| 1607 dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG); | |
| 1608 newnext_ds = prev; | |
| 1609 } | |
| 2199 | 1610 pa->newnext_obj = newnext_ds->ds_object; |
| 2082 | 1611 |
| 1612 /* compute pivot point's new unique space */ | |
| 1613 while ((err = bplist_iterate(&newnext_ds->ds_deadlist, | |
| 1614 &itor, &bp)) == 0) { | |
| 1615 if (bp.blk_birth > pivot_ds->ds_phys->ds_prev_snap_txg) | |
| 2199 | 1616 pa->unique += bp_get_dasize(dd->dd_pool->dp_spa, &bp); |
| 2082 | 1617 } |
| 1618 if (err != ENOENT) | |
| 1619 goto out; | |
| 1620 | |
| 1621 /* Walk the snapshots that we are moving */ | |
| 1622 name = kmem_alloc(MAXPATHLEN, KM_SLEEP); | |
| 1623 ds = pivot_ds; | |
| 1624 /* CONSTCOND */ | |
| 1625 while (TRUE) { | |
| 1626 uint64_t val, dlused, dlcomp, dluncomp; | |
| 1627 dsl_dataset_t *prev; | |
| 1628 | |
| 1629 /* Check that the snapshot name does not conflict */ | |
| 1630 dsl_dataset_name(ds, name); | |
| 1631 err = zap_lookup(dd->dd_pool->dp_meta_objset, | |
| 1632 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, | |
| 1633 8, 1, &val); | |
| 1634 if (err != ENOENT) { | |
| 1635 if (err == 0) | |
| 1636 err = EEXIST; | |
| 1637 goto out; | |
| 1638 } | |
| 1639 | |
| 1640 /* | |
| 1641 * compute space to transfer. Each snapshot gave birth to: | |
| 1642 * (my used) - (prev's used) + (deadlist's used) | |
| 1643 */ | |
| 2199 | 1644 pa->used += ds->ds_phys->ds_used_bytes; |
| 1645 pa->comp += ds->ds_phys->ds_compressed_bytes; | |
| 1646 pa->uncomp += ds->ds_phys->ds_uncompressed_bytes; | |
| 2082 | 1647 |
| 1648 /* If we reach the first snapshot, we're done. */ | |
| 1649 if (ds->ds_phys->ds_prev_snap_obj == 0) | |
| 1650 break; | |
| 1651 | |
| 1652 if (err = bplist_space(&ds->ds_deadlist, | |
| 1653 &dlused, &dlcomp, &dluncomp)) | |
| 1654 goto out; | |
| 1655 if (err = dsl_dataset_open_obj(dd->dd_pool, | |
| 1656 ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE, | |
| 1657 FTAG, &prev)) | |
| 1658 goto out; | |
| 2199 | 1659 pa->used += dlused - prev->ds_phys->ds_used_bytes; |
| 1660 pa->comp += dlcomp - prev->ds_phys->ds_compressed_bytes; | |
| 1661 pa->uncomp += dluncomp - prev->ds_phys->ds_uncompressed_bytes; | |
| 2082 | 1662 |
| 1663 /* | |
| 1664 * We could be a clone of a clone. If we reach our | |
| 1665 * parent's branch point, we're done. | |
| 1666 */ | |
| 1667 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) { | |
| 1668 dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG); | |
| 1669 break; | |
| 1670 } | |
| 1671 if (ds != pivot_ds) | |
| 1672 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 1673 ds = prev; | |
| 1674 } | |
| 1675 | |
| 1676 /* Check that there is enough space here */ | |
| 2199 | 1677 err = dsl_dir_transfer_possible(pdd, dd, pa->used); |
| 1678 | |
| 1679 out: | |
| 1680 if (ds && ds != pivot_ds) | |
| 1681 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 1682 if (pivot_ds) | |
| 1683 dsl_dataset_close(pivot_ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 1684 if (newnext_ds) | |
| 1685 dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG); | |
| 1686 if (name) | |
| 1687 kmem_free(name, MAXPATHLEN); | |
| 1688 return (err); | |
| 1689 } | |
| 2082 | 1690 |
| 2199 | 1691 static void |
| 1692 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 1693 { | |
| 1694 dsl_dataset_t *hds = arg1; | |
| 1695 struct promotearg *pa = arg2; | |
| 1696 dsl_dir_t *dd = hds->ds_dir; | |
| 1697 dsl_pool_t *dp = hds->ds_dir->dd_pool; | |
| 1698 dsl_dir_t *pdd = NULL; | |
| 1699 dsl_dataset_t *ds, *pivot_ds; | |
| 1700 char *name; | |
| 1701 | |
| 1702 ASSERT(dd->dd_phys->dd_clone_parent_obj != 0); | |
| 1703 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)); | |
| 1704 | |
| 1705 VERIFY(0 == dsl_dataset_open_obj(dp, | |
| 1706 dd->dd_phys->dd_clone_parent_obj, | |
| 1707 NULL, DS_MODE_EXCLUSIVE, FTAG, &pivot_ds)); | |
| 1708 pdd = pivot_ds->ds_dir; | |
| 2082 | 1709 |
| 1710 /* move snapshots to this dir */ | |
| 2199 | 1711 name = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
| 2082 | 1712 ds = pivot_ds; |
| 1713 /* CONSTCOND */ | |
| 1714 while (TRUE) { | |
| 1715 dsl_dataset_t *prev; | |
| 1716 | |
| 1717 /* move snap name entry */ | |
| 1718 dsl_dataset_name(ds, name); | |
| 2199 | 1719 VERIFY(0 == zap_remove(dp->dp_meta_objset, |
| 1720 pa->snapnames_obj, ds->ds_snapname, tx)); | |
| 1721 VERIFY(0 == zap_add(dp->dp_meta_objset, | |
| 2082 | 1722 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, |
| 1723 8, 1, &ds->ds_object, tx)); | |
| 1724 | |
| 1725 /* change containing dsl_dir */ | |
| 1726 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 1727 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, pdd->dd_object); | |
| 1728 ds->ds_phys->ds_dir_obj = dd->dd_object; | |
| 1729 ASSERT3P(ds->ds_dir, ==, pdd); | |
| 1730 dsl_dir_close(ds->ds_dir, ds); | |
| 2199 | 1731 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object, |
| 2082 | 1732 NULL, ds, &ds->ds_dir)); |
| 1733 | |
| 1734 ASSERT3U(dsl_prop_numcb(ds), ==, 0); | |
| 1735 | |
| 1736 if (ds->ds_phys->ds_prev_snap_obj == 0) | |
| 1737 break; | |
| 1738 | |
| 2199 | 1739 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 2082 | 1740 ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE, |
| 1741 FTAG, &prev)); | |
| 1742 | |
| 1743 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) { | |
| 1744 dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG); | |
| 1745 break; | |
| 1746 } | |
| 1747 if (ds != pivot_ds) | |
| 1748 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 1749 ds = prev; | |
| 1750 } | |
| 2199 | 1751 if (ds != pivot_ds) |
| 1752 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 2082 | 1753 |
| 1754 /* change pivot point's next snap */ | |
| 1755 dmu_buf_will_dirty(pivot_ds->ds_dbuf, tx); | |
| 2199 | 1756 pivot_ds->ds_phys->ds_next_snap_obj = pa->newnext_obj; |
| 2082 | 1757 |
| 1758 /* change clone_parent-age */ | |
| 1759 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 1760 ASSERT3U(dd->dd_phys->dd_clone_parent_obj, ==, pivot_ds->ds_object); | |
| 1761 dd->dd_phys->dd_clone_parent_obj = pdd->dd_phys->dd_clone_parent_obj; | |
| 1762 dmu_buf_will_dirty(pdd->dd_dbuf, tx); | |
| 1763 pdd->dd_phys->dd_clone_parent_obj = pivot_ds->ds_object; | |
| 1764 | |
| 1765 /* change space accounting */ | |
| 2199 | 1766 dsl_dir_diduse_space(pdd, -pa->used, -pa->comp, -pa->uncomp, tx); |
| 1767 dsl_dir_diduse_space(dd, pa->used, pa->comp, pa->uncomp, tx); | |
| 1768 pivot_ds->ds_phys->ds_unique_bytes = pa->unique; | |
| 2082 | 1769 |
| 2199 | 1770 dsl_dataset_close(pivot_ds, DS_MODE_EXCLUSIVE, FTAG); |
| 1771 kmem_free(name, MAXPATHLEN); | |
| 2082 | 1772 } |
| 1773 | |
| 1774 int | |
| 1775 dsl_dataset_promote(const char *name) | |
| 1776 { | |
| 1777 dsl_dataset_t *ds; | |
| 1778 int err; | |
| 1779 dmu_object_info_t doi; | |
| 2199 | 1780 struct promotearg pa; |
| 2082 | 1781 |
| 1782 err = dsl_dataset_open(name, DS_MODE_NONE, FTAG, &ds); | |
| 1783 if (err) | |
| 1784 return (err); | |
| 1785 | |
| 1786 err = dmu_object_info(ds->ds_dir->dd_pool->dp_meta_objset, | |
| 1787 ds->ds_phys->ds_snapnames_zapobj, &doi); | |
| 1788 if (err) { | |
| 1789 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 1790 return (err); | |
| 1791 } | |
| 1792 | |
| 1793 /* | |
| 1794 * Add in 128x the snapnames zapobj size, since we will be moving | |
| 1795 * a bunch of snapnames to the promoted ds, and dirtying their | |
| 1796 * bonus buffers. | |
| 1797 */ | |
| 2199 | 1798 err = dsl_sync_task_do(ds->ds_dir->dd_pool, |
| 1799 dsl_dataset_promote_check, | |
| 1800 dsl_dataset_promote_sync, ds, &pa, 2 + 2 * doi.doi_physical_blks); | |
| 2082 | 1801 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); |
| 1802 return (err); | |
| 1803 } |