Mercurial > illumos > illumos-gate
annotate usr/src/uts/common/fs/zfs/dsl_dataset.c @ 885:d925b21dba78
6347493 tar of 25K empty directory entries in ZFS takes 30+ seconds ...
6348409 'zfs rename' process hangs after assigning a very long name ...
6348464 a few DMU object type macros are misnamed
| author | ahrens |
|---|---|
| date | Thu, 10 Nov 2005 18:43:50 -0800 |
| parents | b348f31ed315 |
| children | 36d72fe4da29 |
| rev | line source |
|---|---|
| 789 | 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, Version 1.0 only | |
| 6 * (the "License"). You may not use this file except in compliance | |
| 7 * with the License. | |
| 8 * | |
| 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
| 10 * or http://www.opensolaris.org/os/licensing. | |
| 11 * See the License for the specific language governing permissions | |
| 12 * and limitations under the License. | |
| 13 * | |
| 14 * When distributing Covered Code, include this CDDL HEADER in each | |
| 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
| 16 * If applicable, add the following below this CDDL HEADER, with the | |
| 17 * fields enclosed by brackets "[]" replaced with your own identifying | |
| 18 * information: Portions Copyright [yyyy] [name of copyright owner] | |
| 19 * | |
| 20 * CDDL HEADER END | |
| 21 */ | |
| 22 /* | |
| 23 * Copyright 2005 Sun Microsystems, Inc. All rights reserved. | |
| 24 * Use is subject to license terms. | |
| 25 */ | |
| 26 | |
| 27 #pragma ident "%Z%%M% %I% %E% SMI" | |
| 28 | |
| 29 #include <sys/dmu_objset.h> | |
| 30 #include <sys/dsl_dataset.h> | |
| 31 #include <sys/dsl_dir.h> | |
| 32 #include <sys/dmu_traverse.h> | |
| 33 #include <sys/dmu_tx.h> | |
| 34 #include <sys/arc.h> | |
| 35 #include <sys/zio.h> | |
| 36 #include <sys/zap.h> | |
| 37 #include <sys/unique.h> | |
| 38 #include <sys/zfs_context.h> | |
| 39 | |
| 40 #define DOS_REF_MAX (1ULL << 62) | |
| 41 | |
| 42 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE | |
| 43 | |
| 44 #define BP_GET_UCSIZE(bp) \ | |
| 45 ((BP_GET_LEVEL(bp) > 0 || dmu_ot[BP_GET_TYPE(bp)].ot_metadata) ? \ | |
| 46 BP_GET_PSIZE(bp) : BP_GET_LSIZE(bp)); | |
| 47 | |
| 48 /* | |
| 49 * We use weighted reference counts to express the various forms of exclusion | |
| 50 * between different open modes. A STANDARD open is 1 point, an EXCLUSIVE open | |
| 51 * is DOS_REF_MAX, and a PRIMARY open is little more than half of an EXCLUSIVE. | |
| 52 * This makes the exclusion logic simple: the total refcnt for all opens cannot | |
| 53 * exceed DOS_REF_MAX. For example, EXCLUSIVE opens are exclusive because their | |
| 54 * weight (DOS_REF_MAX) consumes the entire refcnt space. PRIMARY opens consume | |
| 55 * just over half of the refcnt space, so there can't be more than one, but it | |
| 56 * can peacefully coexist with any number of STANDARD opens. | |
| 57 */ | |
| 58 static uint64_t ds_refcnt_weight[DS_MODE_LEVELS] = { | |
| 59 0, /* DOS_MODE_NONE - invalid */ | |
| 60 1, /* DOS_MODE_STANDARD - unlimited number */ | |
| 61 (DOS_REF_MAX >> 1) + 1, /* DOS_MODE_PRIMARY - only one of these */ | |
| 62 DOS_REF_MAX /* DOS_MODE_EXCLUSIVE - no other opens */ | |
| 63 }; | |
| 64 | |
| 65 | |
| 66 void | |
| 67 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) | |
| 68 { | |
| 69 int used = BP_GET_ASIZE(bp); | |
| 70 int compressed = BP_GET_PSIZE(bp); | |
| 71 int uncompressed = BP_GET_UCSIZE(bp); | |
| 72 | |
| 73 dprintf_bp(bp, "born, ds=%p\n", ds); | |
| 74 | |
| 75 ASSERT(dmu_tx_is_syncing(tx)); | |
| 76 /* It could have been compressed away to nothing */ | |
| 77 if (BP_IS_HOLE(bp)) | |
| 78 return; | |
| 79 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); | |
| 80 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES); | |
| 81 if (ds == NULL) { | |
| 82 /* | |
| 83 * Account for the meta-objset space in its placeholder | |
| 84 * dsl_dir. | |
| 85 */ | |
| 86 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */ | |
| 87 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, | |
| 88 used, compressed, uncompressed, tx); | |
| 89 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); | |
| 90 return; | |
| 91 } | |
| 92 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 93 mutex_enter(&ds->ds_lock); | |
| 94 ds->ds_phys->ds_used_bytes += used; | |
| 95 ds->ds_phys->ds_compressed_bytes += compressed; | |
| 96 ds->ds_phys->ds_uncompressed_bytes += uncompressed; | |
| 97 ds->ds_phys->ds_unique_bytes += used; | |
| 98 mutex_exit(&ds->ds_lock); | |
| 99 dsl_dir_diduse_space(ds->ds_dir, | |
| 100 used, compressed, uncompressed, tx); | |
| 101 } | |
| 102 | |
| 103 void | |
| 104 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) | |
| 105 { | |
| 106 int used = BP_GET_ASIZE(bp); | |
| 107 int compressed = BP_GET_PSIZE(bp); | |
| 108 int uncompressed = BP_GET_UCSIZE(bp); | |
| 109 | |
| 110 ASSERT(dmu_tx_is_syncing(tx)); | |
| 111 if (BP_IS_HOLE(bp)) | |
| 112 return; | |
| 113 | |
| 114 ASSERT(used > 0); | |
| 115 if (ds == NULL) { | |
| 116 /* | |
| 117 * Account for the meta-objset space in its placeholder | |
| 118 * dataset. | |
| 119 */ | |
| 120 /* XXX this can fail, what do we do when it does? */ | |
| 121 (void) arc_free(NULL, tx->tx_pool->dp_spa, | |
| 122 tx->tx_txg, bp, NULL, NULL, ARC_WAIT); | |
| 123 bzero(bp, sizeof (blkptr_t)); | |
| 124 | |
| 125 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, | |
| 126 -used, -compressed, -uncompressed, tx); | |
| 127 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); | |
| 128 return; | |
| 129 } | |
| 130 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); | |
| 131 | |
| 132 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 133 | |
| 134 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { | |
| 135 dprintf_bp(bp, "freeing: %s", ""); | |
| 136 /* XXX check return code? */ | |
| 137 (void) arc_free(NULL, tx->tx_pool->dp_spa, | |
| 138 tx->tx_txg, bp, NULL, NULL, ARC_WAIT); | |
| 139 | |
| 140 mutex_enter(&ds->ds_lock); | |
| 141 /* XXX unique_bytes is not accurate for head datasets */ | |
| 142 /* ASSERT3U(ds->ds_phys->ds_unique_bytes, >=, used); */ | |
| 143 ds->ds_phys->ds_unique_bytes -= used; | |
| 144 mutex_exit(&ds->ds_lock); | |
| 145 dsl_dir_diduse_space(ds->ds_dir, | |
| 146 -used, -compressed, -uncompressed, tx); | |
| 147 } else { | |
| 148 dprintf_bp(bp, "putting on dead list: %s", ""); | |
| 149 bplist_enqueue(&ds->ds_deadlist, bp, tx); | |
| 150 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ | |
| 151 if (ds->ds_phys->ds_prev_snap_obj != 0) { | |
| 152 ASSERT3U(ds->ds_prev->ds_object, ==, | |
| 153 ds->ds_phys->ds_prev_snap_obj); | |
| 154 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); | |
| 155 if (ds->ds_prev->ds_phys->ds_next_snap_obj == | |
| 156 ds->ds_object && | |
| 157 bp->blk_birth > | |
| 158 ds->ds_prev->ds_phys->ds_prev_snap_txg) { | |
| 159 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
| 160 mutex_enter(&ds->ds_prev->ds_lock); | |
| 161 ds->ds_prev->ds_phys->ds_unique_bytes += | |
| 162 used; | |
| 163 mutex_exit(&ds->ds_prev->ds_lock); | |
| 164 } | |
| 165 } | |
| 166 } | |
| 167 bzero(bp, sizeof (blkptr_t)); | |
| 168 mutex_enter(&ds->ds_lock); | |
| 169 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used); | |
| 170 ds->ds_phys->ds_used_bytes -= used; | |
| 171 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); | |
| 172 ds->ds_phys->ds_compressed_bytes -= compressed; | |
| 173 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); | |
| 174 ds->ds_phys->ds_uncompressed_bytes -= uncompressed; | |
| 175 mutex_exit(&ds->ds_lock); | |
| 176 } | |
| 177 | |
| 178 int | |
| 179 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth, dmu_tx_t *tx) | |
| 180 { | |
| 181 uint64_t prev_snap_txg; | |
| 182 dsl_dir_t *dd; | |
| 183 /* ASSERT that it is not a snapshot */ | |
| 184 if (ds == NULL) | |
| 185 return (TRUE); | |
| 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 */ | |
| 196 dd = ds->ds_dir; | |
| 197 mutex_enter(&dd->dd_lock); | |
| 198 if (dd->dd_sync_func == dsl_dataset_snapshot_sync && | |
| 199 dd->dd_sync_txg < tx->tx_txg) | |
| 200 prev_snap_txg = dd->dd_sync_txg; | |
| 201 else | |
| 202 prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; | |
| 203 mutex_exit(&dd->dd_lock); | |
| 204 return (blk_birth > prev_snap_txg); | |
| 205 } | |
| 206 | |
| 207 /* ARGSUSED */ | |
| 208 static void | |
| 209 dsl_dataset_evict(dmu_buf_t *db, void *dsv) | |
| 210 { | |
| 211 dsl_dataset_t *ds = dsv; | |
| 212 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 213 | |
| 214 /* open_refcount == DOS_REF_MAX when deleting */ | |
| 215 ASSERT(ds->ds_open_refcount == 0 || | |
| 216 ds->ds_open_refcount == DOS_REF_MAX); | |
| 217 | |
| 218 dprintf_ds(ds, "evicting %s\n", ""); | |
| 219 | |
| 220 unique_remove(ds->ds_phys->ds_fsid_guid); | |
| 221 | |
| 222 if (ds->ds_user_ptr != NULL) | |
| 223 ds->ds_user_evict_func(ds, ds->ds_user_ptr); | |
| 224 | |
| 225 if (ds->ds_prev) { | |
| 226 dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds); | |
| 227 ds->ds_prev = NULL; | |
| 228 } | |
| 229 | |
| 230 bplist_close(&ds->ds_deadlist); | |
| 231 dsl_dir_close(ds->ds_dir, ds); | |
| 232 | |
| 233 if (list_link_active(&ds->ds_synced_link)) | |
| 234 list_remove(&dp->dp_synced_objsets, ds); | |
| 235 | |
| 236 kmem_free(ds, sizeof (dsl_dataset_t)); | |
| 237 } | |
| 238 | |
| 239 static void | |
| 240 dsl_dataset_get_snapname(dsl_dataset_t *ds) | |
| 241 { | |
| 242 dsl_dataset_phys_t *headphys; | |
| 243 int err; | |
| 244 dmu_buf_t *headdbuf; | |
| 245 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 246 objset_t *mos = dp->dp_meta_objset; | |
| 247 | |
| 248 if (ds->ds_snapname[0]) | |
| 249 return; | |
| 250 if (ds->ds_phys->ds_next_snap_obj == 0) | |
| 251 return; | |
| 252 | |
| 253 headdbuf = dmu_bonus_hold_tag(mos, | |
| 254 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG); | |
| 255 dmu_buf_read(headdbuf); | |
| 256 headphys = headdbuf->db_data; | |
| 257 err = zap_value_search(dp->dp_meta_objset, | |
| 258 headphys->ds_snapnames_zapobj, ds->ds_object, ds->ds_snapname); | |
| 259 ASSERT(err == 0); | |
| 260 dmu_buf_rele_tag(headdbuf, FTAG); | |
| 261 } | |
| 262 | |
| 263 dsl_dataset_t * | |
| 264 dsl_dataset_open_obj(dsl_pool_t *dp, uint64_t dsobj, const char *snapname, | |
| 265 int mode, void *tag) | |
| 266 { | |
| 267 uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)]; | |
| 268 objset_t *mos = dp->dp_meta_objset; | |
| 269 dmu_buf_t *dbuf; | |
| 270 dsl_dataset_t *ds; | |
| 271 | |
| 272 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || | |
| 273 dsl_pool_sync_context(dp)); | |
| 274 | |
| 275 dbuf = dmu_bonus_hold_tag(mos, dsobj, tag); | |
| 276 dmu_buf_read(dbuf); | |
| 277 ds = dmu_buf_get_user(dbuf); | |
| 278 if (ds == NULL) { | |
| 279 dsl_dataset_t *winner; | |
| 280 | |
| 281 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); | |
| 282 ds->ds_dbuf = dbuf; | |
| 283 ds->ds_object = dsobj; | |
| 284 ds->ds_phys = dbuf->db_data; | |
| 285 ds->ds_dir = dsl_dir_open_obj(dp, | |
| 286 ds->ds_phys->ds_dir_obj, NULL, ds); | |
| 287 | |
| 288 bplist_open(&ds->ds_deadlist, | |
| 289 mos, ds->ds_phys->ds_deadlist_obj); | |
| 290 | |
| 291 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == dsobj) { | |
| 292 ds->ds_snapname[0] = '\0'; | |
| 293 if (ds->ds_phys->ds_prev_snap_obj) { | |
| 294 ds->ds_prev = | |
| 295 dsl_dataset_open_obj(dp, | |
| 296 ds->ds_phys->ds_prev_snap_obj, NULL, | |
| 297 DS_MODE_NONE, ds); | |
| 298 } | |
| 299 } else { | |
| 300 if (snapname) { | |
| 301 #ifdef ZFS_DEBUG | |
| 302 dsl_dataset_phys_t *headphys; | |
| 303 int err; | |
| 304 dmu_buf_t *headdbuf = dmu_bonus_hold_tag(mos, | |
| 305 ds->ds_dir->dd_phys-> | |
| 306 dd_head_dataset_obj, FTAG); | |
| 307 dmu_buf_read(headdbuf); | |
| 308 headphys = headdbuf->db_data; | |
| 309 uint64_t foundobj; | |
| 310 err = zap_lookup(dp->dp_meta_objset, | |
| 311 headphys->ds_snapnames_zapobj, | |
| 312 snapname, sizeof (foundobj), 1, &foundobj); | |
| 313 ASSERT3U(err, ==, 0); | |
| 314 ASSERT3U(foundobj, ==, dsobj); | |
| 315 dmu_buf_rele_tag(headdbuf, FTAG); | |
| 316 #endif | |
| 317 (void) strcat(ds->ds_snapname, snapname); | |
| 318 } else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) { | |
| 319 dsl_dataset_get_snapname(ds); | |
| 320 } | |
| 321 } | |
| 322 | |
| 323 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys, | |
| 324 dsl_dataset_evict); | |
| 325 if (winner) { | |
| 326 bplist_close(&ds->ds_deadlist); | |
| 327 if (ds->ds_prev) { | |
| 328 dsl_dataset_close(ds->ds_prev, | |
| 329 DS_MODE_NONE, ds); | |
| 330 } | |
| 331 dsl_dir_close(ds->ds_dir, ds); | |
| 332 kmem_free(ds, sizeof (dsl_dataset_t)); | |
| 333 ds = winner; | |
| 334 } else { | |
| 335 uint64_t new = | |
| 336 unique_insert(ds->ds_phys->ds_fsid_guid); | |
| 337 if (new != ds->ds_phys->ds_fsid_guid) { | |
| 338 /* XXX it won't necessarily be synced... */ | |
| 339 ds->ds_phys->ds_fsid_guid = new; | |
| 340 } | |
| 341 } | |
| 342 } | |
| 343 ASSERT3P(ds->ds_dbuf, ==, dbuf); | |
| 344 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); | |
| 345 | |
| 346 mutex_enter(&ds->ds_lock); | |
| 347 if ((DS_MODE_LEVEL(mode) == DS_MODE_PRIMARY && | |
| 348 ds->ds_phys->ds_restoring && !DS_MODE_IS_RESTORE(mode)) || | |
| 349 (ds->ds_open_refcount + weight > DOS_REF_MAX)) { | |
| 350 mutex_exit(&ds->ds_lock); | |
| 351 dsl_dataset_close(ds, DS_MODE_NONE, tag); | |
| 352 return (NULL); | |
| 353 } | |
| 354 ds->ds_open_refcount += weight; | |
| 355 mutex_exit(&ds->ds_lock); | |
| 356 | |
| 357 return (ds); | |
| 358 } | |
| 359 | |
| 360 int | |
| 361 dsl_dataset_open_spa(spa_t *spa, const char *name, int mode, | |
| 362 void *tag, dsl_dataset_t **dsp) | |
| 363 { | |
| 364 dsl_dir_t *dd; | |
| 365 dsl_pool_t *dp; | |
| 366 const char *tail; | |
| 367 uint64_t obj; | |
| 368 dsl_dataset_t *ds = NULL; | |
| 369 int err = 0; | |
| 370 | |
| 371 dd = dsl_dir_open_spa(spa, name, FTAG, &tail); | |
| 372 if (dd == NULL) | |
| 373 return (ENOENT); | |
| 374 | |
| 375 dp = dd->dd_pool; | |
| 376 obj = dd->dd_phys->dd_head_dataset_obj; | |
| 377 rw_enter(&dp->dp_config_rwlock, RW_READER); | |
| 378 if (obj == 0) { | |
| 379 /* A dataset with no associated objset */ | |
| 380 err = ENOENT; | |
| 381 goto out; | |
| 382 } | |
| 383 | |
| 384 if (tail != NULL) { | |
| 385 objset_t *mos = dp->dp_meta_objset; | |
| 386 | |
| 387 ds = dsl_dataset_open_obj(dp, obj, NULL, DS_MODE_NONE, tag); | |
| 388 obj = ds->ds_phys->ds_snapnames_zapobj; | |
| 389 dsl_dataset_close(ds, DS_MODE_NONE, tag); | |
| 390 ds = NULL; | |
| 391 | |
| 392 if (tail[0] != '@') { | |
| 393 err = ENOENT; | |
| 394 goto out; | |
| 395 } | |
| 396 tail++; | |
| 397 | |
| 398 /* Look for a snapshot */ | |
| 399 if (!DS_MODE_IS_READONLY(mode)) { | |
| 400 err = EROFS; | |
| 401 goto out; | |
| 402 } | |
| 403 dprintf("looking for snapshot '%s'\n", tail); | |
| 404 err = zap_lookup(mos, obj, tail, 8, 1, &obj); | |
| 405 if (err) | |
| 406 goto out; | |
| 407 } | |
| 408 ds = dsl_dataset_open_obj(dp, obj, tail, mode, tag); | |
| 409 if (ds == NULL) | |
| 410 err = EBUSY; | |
| 411 | |
| 412 out: | |
| 413 rw_exit(&dp->dp_config_rwlock); | |
| 414 dsl_dir_close(dd, FTAG); | |
| 415 | |
| 416 ASSERT3U((err == 0), ==, (ds != NULL)); | |
| 417 /* ASSERT(ds == NULL || strcmp(name, ds->ds_name) == 0); */ | |
| 418 | |
| 419 *dsp = ds; | |
| 420 return (err); | |
| 421 } | |
| 422 | |
| 423 int | |
| 424 dsl_dataset_open(const char *name, int mode, void *tag, dsl_dataset_t **dsp) | |
| 425 { | |
| 426 return (dsl_dataset_open_spa(NULL, name, mode, tag, dsp)); | |
| 427 } | |
| 428 | |
| 429 void | |
| 430 dsl_dataset_name(dsl_dataset_t *ds, char *name) | |
| 431 { | |
| 432 if (ds == NULL) { | |
| 433 (void) strcpy(name, "mos"); | |
| 434 } else { | |
| 435 dsl_dir_name(ds->ds_dir, name); | |
| 436 dsl_dataset_get_snapname(ds); | |
| 437 if (ds->ds_snapname[0]) { | |
| 438 (void) strcat(name, "@"); | |
| 439 if (!MUTEX_HELD(&ds->ds_lock)) { | |
| 440 /* | |
| 441 * We use a "recursive" mutex so that we | |
| 442 * can call dprintf_ds() with ds_lock held. | |
| 443 */ | |
| 444 mutex_enter(&ds->ds_lock); | |
| 445 (void) strcat(name, ds->ds_snapname); | |
| 446 mutex_exit(&ds->ds_lock); | |
| 447 } else { | |
| 448 (void) strcat(name, ds->ds_snapname); | |
| 449 } | |
| 450 } | |
| 451 } | |
| 452 } | |
| 453 | |
| 454 void | |
| 455 dsl_dataset_close(dsl_dataset_t *ds, int mode, void *tag) | |
| 456 { | |
| 457 uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)]; | |
| 458 mutex_enter(&ds->ds_lock); | |
| 459 ASSERT3U(ds->ds_open_refcount, >=, weight); | |
| 460 ds->ds_open_refcount -= weight; | |
| 461 dprintf_ds(ds, "closing mode %u refcount now 0x%llx\n", | |
| 462 mode, ds->ds_open_refcount); | |
| 463 mutex_exit(&ds->ds_lock); | |
| 464 | |
| 465 dmu_buf_rele_tag(ds->ds_dbuf, tag); | |
| 466 } | |
| 467 | |
| 468 void | |
| 469 dsl_dataset_create_root(dsl_pool_t *dp, uint64_t *ddobjp, dmu_tx_t *tx) | |
| 470 { | |
| 471 objset_t *mos = dp->dp_meta_objset; | |
| 472 dmu_buf_t *dbuf; | |
| 473 dsl_dataset_phys_t *dsphys; | |
| 474 dsl_dataset_t *ds; | |
| 475 uint64_t dsobj; | |
| 476 dsl_dir_t *dd; | |
| 477 | |
| 478 dsl_dir_create_root(mos, ddobjp, tx); | |
| 479 dd = dsl_dir_open_obj(dp, *ddobjp, NULL, FTAG); | |
| 480 ASSERT(dd != NULL); | |
| 481 | |
| 482 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_OBJSET, 0, | |
| 483 DMU_OT_DSL_OBJSET, sizeof (dsl_dataset_phys_t), tx); | |
| 484 dbuf = dmu_bonus_hold(mos, dsobj); | |
| 485 dmu_buf_will_dirty(dbuf, tx); | |
| 486 dsphys = dbuf->db_data; | |
| 487 dsphys->ds_dir_obj = dd->dd_object; | |
| 488 dsphys->ds_fsid_guid = unique_create(); | |
| 489 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, | |
| 490 sizeof (dsphys->ds_guid)); | |
| 491 dsphys->ds_snapnames_zapobj = | |
|
885
d925b21dba78
6347493 tar of 25K empty directory entries in ZFS takes 30+ seconds ...
ahrens
parents:
789
diff
changeset
|
492 zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx); |
| 789 | 493 dsphys->ds_creation_time = gethrestime_sec(); |
| 494 dsphys->ds_creation_txg = tx->tx_txg; | |
| 495 dsphys->ds_deadlist_obj = | |
| 496 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 497 dmu_buf_rele(dbuf); | |
| 498 | |
| 499 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 500 dd->dd_phys->dd_head_dataset_obj = dsobj; | |
| 501 dsl_dir_close(dd, FTAG); | |
| 502 | |
| 503 ds = dsl_dataset_open_obj(dp, dsobj, NULL, DS_MODE_NONE, FTAG); | |
| 504 (void) dmu_objset_create_impl(dp->dp_spa, ds, DMU_OST_ZFS, tx); | |
| 505 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 506 } | |
| 507 | |
| 508 int | |
| 509 dsl_dataset_create_sync(dsl_dir_t *pds, const char *fullname, | |
| 510 const char *lastname, dsl_dataset_t *clone_parent, dmu_tx_t *tx) | |
| 511 { | |
| 512 int err; | |
| 513 dsl_pool_t *dp = pds->dd_pool; | |
| 514 dmu_buf_t *dbuf; | |
| 515 dsl_dataset_phys_t *dsphys; | |
| 516 uint64_t dsobj; | |
| 517 objset_t *mos = dp->dp_meta_objset; | |
| 518 dsl_dir_t *dd; | |
| 519 | |
| 520 if (clone_parent != NULL) { | |
| 521 /* | |
| 522 * You can't clone across pools. | |
| 523 */ | |
| 524 if (clone_parent->ds_dir->dd_pool != dp) | |
| 525 return (EXDEV); | |
| 526 | |
| 527 /* | |
| 528 * You can only clone snapshots, not the head datasets. | |
| 529 */ | |
| 530 if (clone_parent->ds_phys->ds_num_children == 0) | |
| 531 return (EINVAL); | |
| 532 } | |
| 533 | |
| 534 ASSERT(lastname[0] != '@'); | |
| 535 ASSERT(dmu_tx_is_syncing(tx)); | |
| 536 | |
| 537 err = dsl_dir_create_sync(pds, lastname, tx); | |
| 538 if (err) | |
| 539 return (err); | |
| 540 dd = dsl_dir_open_spa(dp->dp_spa, fullname, FTAG, NULL); | |
| 541 ASSERT(dd != NULL); | |
| 542 | |
| 543 /* This is the point of no (unsuccessful) return */ | |
| 544 | |
| 545 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_OBJSET, 0, | |
| 546 DMU_OT_DSL_OBJSET, sizeof (dsl_dataset_phys_t), tx); | |
| 547 dbuf = dmu_bonus_hold(mos, dsobj); | |
| 548 dmu_buf_will_dirty(dbuf, tx); | |
| 549 dsphys = dbuf->db_data; | |
| 550 dsphys->ds_dir_obj = dd->dd_object; | |
| 551 dsphys->ds_fsid_guid = unique_create(); | |
| 552 unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */ | |
| 553 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, | |
| 554 sizeof (dsphys->ds_guid)); | |
| 555 dsphys->ds_snapnames_zapobj = | |
|
885
d925b21dba78
6347493 tar of 25K empty directory entries in ZFS takes 30+ seconds ...
ahrens
parents:
789
diff
changeset
|
556 zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx); |
| 789 | 557 dsphys->ds_creation_time = gethrestime_sec(); |
| 558 dsphys->ds_creation_txg = tx->tx_txg; | |
| 559 dsphys->ds_deadlist_obj = | |
| 560 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 561 if (clone_parent) { | |
| 562 dsphys->ds_prev_snap_obj = clone_parent->ds_object; | |
| 563 dsphys->ds_prev_snap_txg = | |
| 564 clone_parent->ds_phys->ds_creation_txg; | |
| 565 dsphys->ds_used_bytes = | |
| 566 clone_parent->ds_phys->ds_used_bytes; | |
| 567 dsphys->ds_compressed_bytes = | |
| 568 clone_parent->ds_phys->ds_compressed_bytes; | |
| 569 dsphys->ds_uncompressed_bytes = | |
| 570 clone_parent->ds_phys->ds_uncompressed_bytes; | |
| 571 dsphys->ds_bp = clone_parent->ds_phys->ds_bp; | |
| 572 | |
| 573 dmu_buf_will_dirty(clone_parent->ds_dbuf, tx); | |
| 574 clone_parent->ds_phys->ds_num_children++; | |
| 575 | |
| 576 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 577 dd->dd_phys->dd_clone_parent_obj = clone_parent->ds_object; | |
| 578 } | |
| 579 dmu_buf_rele(dbuf); | |
| 580 | |
| 581 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 582 dd->dd_phys->dd_head_dataset_obj = dsobj; | |
| 583 dsl_dir_close(dd, FTAG); | |
| 584 | |
| 585 return (0); | |
| 586 } | |
| 587 | |
| 588 | |
| 589 int | |
| 590 dsl_dataset_destroy(const char *name) | |
| 591 { | |
| 592 int err; | |
| 593 dsl_pool_t *dp; | |
| 594 dsl_dir_t *dd; | |
| 595 const char *tail; | |
| 596 | |
| 597 dd = dsl_dir_open(name, FTAG, &tail); | |
| 598 if (dd == NULL) | |
| 599 return (ENOENT); | |
| 600 | |
| 601 dp = dd->dd_pool; | |
| 602 if (tail != NULL) { | |
| 603 if (tail[0] != '@') { | |
| 604 dsl_dir_close(dd, FTAG); | |
| 605 return (ENOENT); | |
| 606 } | |
| 607 tail++; | |
| 608 /* Just blow away the snapshot */ | |
| 609 do { | |
| 610 txg_wait_synced(dp, 0); | |
| 611 err = dsl_dir_sync_task(dd, | |
| 612 dsl_dataset_destroy_sync, (void*)tail, 0); | |
| 613 } while (err == EAGAIN); | |
| 614 dsl_dir_close(dd, FTAG); | |
| 615 } else { | |
| 616 char buf[MAXNAMELEN]; | |
| 617 char *cp; | |
| 618 | |
| 619 dsl_dir_t *pds; | |
| 620 if (dd->dd_phys->dd_parent_obj == 0) { | |
| 621 dsl_dir_close(dd, FTAG); | |
| 622 return (EINVAL); | |
| 623 } | |
| 624 /* | |
| 625 * Make sure it's not dirty before we destroy it. | |
| 626 */ | |
| 627 txg_wait_synced(dd->dd_pool, 0); | |
| 628 /* | |
| 629 * Blow away the dsl_dir + head dataset. | |
| 630 * dsl_dir_destroy_sync() will call | |
| 631 * dsl_dataset_destroy_sync() to destroy the head dataset. | |
| 632 */ | |
| 633 rw_enter(&dp->dp_config_rwlock, RW_READER); | |
| 634 pds = dsl_dir_open_obj(dd->dd_pool, | |
| 635 dd->dd_phys->dd_parent_obj, NULL, FTAG); | |
| 636 dsl_dir_close(dd, FTAG); | |
| 637 rw_exit(&dp->dp_config_rwlock); | |
| 638 | |
| 639 (void) strcpy(buf, name); | |
| 640 cp = strrchr(buf, '/') + 1; | |
| 641 ASSERT(cp[0] != '\0'); | |
| 642 do { | |
| 643 txg_wait_synced(dp, 0); | |
| 644 err = dsl_dir_sync_task(pds, | |
| 645 dsl_dir_destroy_sync, cp, 0); | |
| 646 } while (err == EAGAIN); | |
| 647 dsl_dir_close(pds, FTAG); | |
| 648 } | |
| 649 | |
| 650 return (err); | |
| 651 } | |
| 652 | |
| 653 int | |
| 654 dsl_dataset_rollback(const char *name) | |
| 655 { | |
| 656 int err; | |
| 657 dsl_dir_t *dd; | |
| 658 const char *tail; | |
| 659 | |
| 660 dd = dsl_dir_open(name, FTAG, &tail); | |
| 661 if (dd == NULL) | |
| 662 return (ENOENT); | |
| 663 | |
| 664 if (tail != NULL) { | |
| 665 dsl_dir_close(dd, FTAG); | |
| 666 return (EINVAL); | |
| 667 } | |
| 668 do { | |
| 669 txg_wait_synced(dd->dd_pool, 0); | |
| 670 err = dsl_dir_sync_task(dd, | |
| 671 dsl_dataset_rollback_sync, NULL, 0); | |
| 672 } while (err == EAGAIN); | |
| 673 dsl_dir_close(dd, FTAG); | |
| 674 | |
| 675 return (err); | |
| 676 } | |
| 677 | |
| 678 void * | |
| 679 dsl_dataset_set_user_ptr(dsl_dataset_t *ds, | |
| 680 void *p, dsl_dataset_evict_func_t func) | |
| 681 { | |
| 682 void *old; | |
| 683 | |
| 684 mutex_enter(&ds->ds_lock); | |
| 685 old = ds->ds_user_ptr; | |
| 686 if (old == NULL) { | |
| 687 ds->ds_user_ptr = p; | |
| 688 ds->ds_user_evict_func = func; | |
| 689 } | |
| 690 mutex_exit(&ds->ds_lock); | |
| 691 return (old); | |
| 692 } | |
| 693 | |
| 694 void * | |
| 695 dsl_dataset_get_user_ptr(dsl_dataset_t *ds) | |
| 696 { | |
| 697 return (ds->ds_user_ptr); | |
| 698 } | |
| 699 | |
| 700 | |
| 701 void | |
| 702 dsl_dataset_get_blkptr(dsl_dataset_t *ds, blkptr_t *bp) | |
| 703 { | |
| 704 *bp = ds->ds_phys->ds_bp; | |
| 705 } | |
| 706 | |
| 707 void | |
| 708 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) | |
| 709 { | |
| 710 ASSERT(dmu_tx_is_syncing(tx)); | |
| 711 /* If it's the meta-objset, set dp_meta_rootbp */ | |
| 712 if (ds == NULL) { | |
| 713 tx->tx_pool->dp_meta_rootbp = *bp; | |
| 714 } else { | |
| 715 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 716 ds->ds_phys->ds_bp = *bp; | |
| 717 } | |
| 718 } | |
| 719 | |
| 720 spa_t * | |
| 721 dsl_dataset_get_spa(dsl_dataset_t *ds) | |
| 722 { | |
| 723 return (ds->ds_dir->dd_pool->dp_spa); | |
| 724 } | |
| 725 | |
| 726 void | |
| 727 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) | |
| 728 { | |
| 729 dsl_pool_t *dp; | |
| 730 | |
| 731 if (ds == NULL) /* this is the meta-objset */ | |
| 732 return; | |
| 733 | |
| 734 ASSERT(ds->ds_user_ptr != NULL); | |
| 735 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); | |
| 736 | |
| 737 dp = ds->ds_dir->dd_pool; | |
| 738 | |
| 739 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) { | |
| 740 /* up the hold count until we can be written out */ | |
| 741 dmu_buf_add_ref(ds->ds_dbuf, ds); | |
| 742 } | |
| 743 } | |
| 744 | |
| 745 struct killarg { | |
| 746 uint64_t *usedp; | |
| 747 uint64_t *compressedp; | |
| 748 uint64_t *uncompressedp; | |
| 749 zio_t *zio; | |
| 750 dmu_tx_t *tx; | |
| 751 }; | |
| 752 | |
| 753 static int | |
| 754 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg) | |
| 755 { | |
| 756 struct killarg *ka = arg; | |
| 757 blkptr_t *bp = &bc->bc_blkptr; | |
| 758 | |
| 759 ASSERT3U(bc->bc_errno, ==, 0); | |
| 760 | |
| 761 /* | |
| 762 * Since this callback is not called concurrently, no lock is | |
| 763 * needed on the accounting values. | |
| 764 */ | |
| 765 *ka->usedp += BP_GET_ASIZE(bp); | |
| 766 *ka->compressedp += BP_GET_PSIZE(bp); | |
| 767 *ka->uncompressedp += BP_GET_UCSIZE(bp); | |
| 768 /* XXX check for EIO? */ | |
| 769 (void) arc_free(ka->zio, spa, ka->tx->tx_txg, bp, NULL, NULL, | |
| 770 ARC_NOWAIT); | |
| 771 return (0); | |
| 772 } | |
| 773 | |
| 774 /* ARGSUSED */ | |
| 775 int | |
| 776 dsl_dataset_rollback_sync(dsl_dir_t *dd, void *arg, dmu_tx_t *tx) | |
| 777 { | |
| 778 objset_t *mos = dd->dd_pool->dp_meta_objset; | |
| 779 dsl_dataset_t *ds; | |
| 780 | |
| 781 if (dd->dd_phys->dd_head_dataset_obj == 0) | |
| 782 return (EINVAL); | |
| 783 ds = dsl_dataset_open_obj(dd->dd_pool, | |
| 784 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG); | |
| 785 | |
| 786 if (ds->ds_phys->ds_prev_snap_txg == 0) { | |
| 787 /* | |
| 788 * There's no previous snapshot. I suppose we could | |
| 789 * roll it back to being empty (and re-initialize the | |
| 790 * upper (ZPL) layer). But for now there's no way to do | |
| 791 * this via the user interface. | |
| 792 */ | |
| 793 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 794 return (EINVAL); | |
| 795 } | |
| 796 | |
| 797 mutex_enter(&ds->ds_lock); | |
| 798 if (ds->ds_open_refcount > 0) { | |
| 799 mutex_exit(&ds->ds_lock); | |
| 800 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 801 return (EBUSY); | |
| 802 } | |
| 803 | |
| 804 /* | |
| 805 * If we made changes this txg, traverse_dsl_dataset won't find | |
| 806 * them. Try again. | |
| 807 */ | |
| 808 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) { | |
| 809 mutex_exit(&ds->ds_lock); | |
| 810 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 811 return (EAGAIN); | |
| 812 } | |
| 813 | |
| 814 /* THE POINT OF NO (unsuccessful) RETURN */ | |
| 815 ds->ds_open_refcount = DOS_REF_MAX; | |
| 816 mutex_exit(&ds->ds_lock); | |
| 817 | |
| 818 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 819 | |
| 820 /* Zero out the deadlist. */ | |
| 821 dprintf("old deadlist obj = %llx\n", ds->ds_phys->ds_deadlist_obj); | |
| 822 bplist_close(&ds->ds_deadlist); | |
| 823 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); | |
| 824 ds->ds_phys->ds_deadlist_obj = | |
| 825 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 826 bplist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); | |
| 827 dprintf("new deadlist obj = %llx\n", ds->ds_phys->ds_deadlist_obj); | |
| 828 | |
| 829 { | |
| 830 /* Free blkptrs that we gave birth to */ | |
| 831 zio_t *zio; | |
| 832 uint64_t used = 0, compressed = 0, uncompressed = 0; | |
| 833 struct killarg ka; | |
| 834 | |
| 835 zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL, | |
| 836 ZIO_FLAG_MUSTSUCCEED); | |
| 837 ka.usedp = &used; | |
| 838 ka.compressedp = &compressed; | |
| 839 ka.uncompressedp = &uncompressed; | |
| 840 ka.zio = zio; | |
| 841 ka.tx = tx; | |
| 842 (void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg, | |
| 843 ADVANCE_POST, kill_blkptr, &ka); | |
| 844 (void) zio_wait(zio); | |
| 845 | |
| 846 dsl_dir_diduse_space(dd, | |
| 847 -used, -compressed, -uncompressed, tx); | |
| 848 } | |
| 849 | |
| 850 /* Change our contents to that of the prev snapshot (finally!) */ | |
| 851 ASSERT3U(ds->ds_prev->ds_object, ==, ds->ds_phys->ds_prev_snap_obj); | |
| 852 ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp; | |
| 853 ds->ds_phys->ds_used_bytes = ds->ds_prev->ds_phys->ds_used_bytes; | |
| 854 ds->ds_phys->ds_compressed_bytes = | |
| 855 ds->ds_prev->ds_phys->ds_compressed_bytes; | |
| 856 ds->ds_phys->ds_uncompressed_bytes = | |
| 857 ds->ds_prev->ds_phys->ds_uncompressed_bytes; | |
| 858 ds->ds_phys->ds_restoring = ds->ds_prev->ds_phys->ds_restoring; | |
| 859 ds->ds_phys->ds_unique_bytes = 0; | |
| 860 | |
| 861 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
| 862 ds->ds_prev->ds_phys->ds_unique_bytes = 0; | |
| 863 | |
| 864 dprintf("new deadlist obj = %llx\n", ds->ds_phys->ds_deadlist_obj); | |
| 865 ds->ds_open_refcount = 0; | |
| 866 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 867 | |
| 868 return (0); | |
| 869 } | |
| 870 | |
| 871 int | |
| 872 dsl_dataset_destroy_sync(dsl_dir_t *dd, void *arg, dmu_tx_t *tx) | |
| 873 { | |
| 874 const char *snapname = arg; | |
| 875 uint64_t used = 0, compressed = 0, uncompressed = 0; | |
| 876 blkptr_t bp; | |
| 877 zio_t *zio; | |
| 878 int err; | |
| 879 int after_branch_point = FALSE; | |
| 880 int drop_lock = FALSE; | |
| 881 dsl_pool_t *dp = dd->dd_pool; | |
| 882 objset_t *mos = dp->dp_meta_objset; | |
| 883 dsl_dataset_t *ds, *ds_prev = NULL; | |
| 884 uint64_t obj; | |
| 885 | |
| 886 if (dd->dd_phys->dd_head_dataset_obj == 0) | |
| 887 return (EINVAL); | |
| 888 | |
| 889 if (!RW_WRITE_HELD(&dp->dp_config_rwlock)) { | |
| 890 rw_enter(&dp->dp_config_rwlock, RW_WRITER); | |
| 891 drop_lock = TRUE; | |
| 892 } | |
| 893 | |
| 894 ds = dsl_dataset_open_obj(dd->dd_pool, | |
| 895 dd->dd_phys->dd_head_dataset_obj, NULL, | |
| 896 snapname ? DS_MODE_NONE : DS_MODE_EXCLUSIVE, FTAG); | |
| 897 | |
| 898 if (snapname) { | |
| 899 err = zap_lookup(mos, ds->ds_phys->ds_snapnames_zapobj, | |
| 900 snapname, 8, 1, &obj); | |
| 901 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 902 if (err) { | |
| 903 if (drop_lock) | |
| 904 rw_exit(&dp->dp_config_rwlock); | |
| 905 return (err); | |
| 906 } | |
| 907 | |
| 908 ds = dsl_dataset_open_obj(dd->dd_pool, obj, NULL, | |
| 909 DS_MODE_EXCLUSIVE, FTAG); | |
| 910 } | |
| 911 if (ds == NULL) { | |
| 912 if (drop_lock) | |
| 913 rw_exit(&dp->dp_config_rwlock); | |
| 914 return (EBUSY); | |
| 915 } | |
| 916 | |
| 917 obj = ds->ds_object; | |
| 918 | |
| 919 /* Can't delete a branch point. */ | |
| 920 if (ds->ds_phys->ds_num_children > 1) { | |
| 921 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 922 if (drop_lock) | |
| 923 rw_exit(&dp->dp_config_rwlock); | |
| 924 return (EINVAL); | |
| 925 } | |
| 926 | |
| 927 /* | |
| 928 * Can't delete a head dataset if there are snapshots of it. | |
| 929 * (Except if the only snapshots are from the branch we cloned | |
| 930 * from.) | |
| 931 */ | |
| 932 if (ds->ds_prev != NULL && | |
| 933 ds->ds_prev->ds_phys->ds_next_snap_obj == obj) { | |
| 934 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 935 if (drop_lock) | |
| 936 rw_exit(&dp->dp_config_rwlock); | |
| 937 return (EINVAL); | |
| 938 } | |
| 939 | |
| 940 /* | |
| 941 * If we made changes this txg, traverse_dsl_dataset won't find | |
| 942 * them. Try again. | |
| 943 */ | |
| 944 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) { | |
| 945 mutex_exit(&ds->ds_lock); | |
| 946 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
|
885
d925b21dba78
6347493 tar of 25K empty directory entries in ZFS takes 30+ seconds ...
ahrens
parents:
789
diff
changeset
|
947 if (drop_lock) |
|
d925b21dba78
6347493 tar of 25K empty directory entries in ZFS takes 30+ seconds ...
ahrens
parents:
789
diff
changeset
|
948 rw_exit(&dp->dp_config_rwlock); |
| 789 | 949 return (EAGAIN); |
| 950 } | |
| 951 | |
| 952 /* THE POINT OF NO (unsuccessful) RETURN */ | |
| 953 | |
| 954 if (ds->ds_phys->ds_prev_snap_obj != 0) { | |
| 955 if (ds->ds_prev) { | |
| 956 ds_prev = ds->ds_prev; | |
| 957 } else { | |
| 958 ds_prev = dsl_dataset_open_obj(dd->dd_pool, | |
| 959 ds->ds_phys->ds_prev_snap_obj, NULL, | |
| 960 DS_MODE_NONE, FTAG); | |
| 961 } | |
| 962 after_branch_point = | |
| 963 (ds_prev->ds_phys->ds_next_snap_obj != obj); | |
| 964 | |
| 965 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); | |
| 966 if (after_branch_point && | |
| 967 ds->ds_phys->ds_next_snap_obj == 0) { | |
| 968 /* This clone is toast. */ | |
| 969 ASSERT(ds_prev->ds_phys->ds_num_children > 1); | |
| 970 ds_prev->ds_phys->ds_num_children--; | |
| 971 } else if (!after_branch_point) { | |
| 972 ds_prev->ds_phys->ds_next_snap_obj = | |
| 973 ds->ds_phys->ds_next_snap_obj; | |
| 974 } | |
| 975 } | |
| 976 | |
| 977 ASSERT3P(tx->tx_pool, ==, dd->dd_pool); | |
| 978 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
| 979 | |
| 980 if (ds->ds_phys->ds_next_snap_obj != 0) { | |
| 981 dsl_dataset_t *ds_next; | |
| 982 uint64_t itor = 0; | |
| 983 | |
| 984 spa_scrub_restart(dp->dp_spa, tx->tx_txg); | |
| 985 | |
| 986 ds_next = dsl_dataset_open_obj(dd->dd_pool, | |
| 987 ds->ds_phys->ds_next_snap_obj, NULL, DS_MODE_NONE, FTAG); | |
| 988 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); | |
| 989 | |
| 990 dmu_buf_will_dirty(ds_next->ds_dbuf, tx); | |
| 991 ds_next->ds_phys->ds_prev_snap_obj = | |
| 992 ds->ds_phys->ds_prev_snap_obj; | |
| 993 ds_next->ds_phys->ds_prev_snap_txg = | |
| 994 ds->ds_phys->ds_prev_snap_txg; | |
| 995 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, | |
| 996 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); | |
| 997 | |
| 998 /* | |
| 999 * Transfer to our deadlist (which will become next's | |
| 1000 * new deadlist) any entries from next's current | |
| 1001 * deadlist which were born before prev, and free the | |
| 1002 * other entries. | |
| 1003 * | |
| 1004 * XXX we're doing this long task with the config lock held | |
| 1005 */ | |
| 1006 while (bplist_iterate(&ds_next->ds_deadlist, &itor, | |
| 1007 &bp) == 0) { | |
| 1008 if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) { | |
| 1009 bplist_enqueue(&ds->ds_deadlist, &bp, tx); | |
| 1010 if (ds_prev && !after_branch_point && | |
| 1011 bp.blk_birth > | |
| 1012 ds_prev->ds_phys->ds_prev_snap_txg) { | |
| 1013 ds_prev->ds_phys->ds_unique_bytes += | |
| 1014 BP_GET_ASIZE(&bp); | |
| 1015 } | |
| 1016 } else { | |
| 1017 used += BP_GET_ASIZE(&bp); | |
| 1018 compressed += BP_GET_PSIZE(&bp); | |
| 1019 uncompressed += BP_GET_UCSIZE(&bp); | |
| 1020 /* XXX check return value? */ | |
| 1021 (void) arc_free(zio, dp->dp_spa, tx->tx_txg, | |
| 1022 &bp, NULL, NULL, ARC_NOWAIT); | |
| 1023 } | |
| 1024 } | |
| 1025 | |
| 1026 /* free next's deadlist */ | |
| 1027 bplist_close(&ds_next->ds_deadlist); | |
| 1028 bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx); | |
| 1029 | |
| 1030 /* set next's deadlist to our deadlist */ | |
| 1031 ds_next->ds_phys->ds_deadlist_obj = | |
| 1032 ds->ds_phys->ds_deadlist_obj; | |
| 1033 bplist_open(&ds_next->ds_deadlist, mos, | |
| 1034 ds_next->ds_phys->ds_deadlist_obj); | |
| 1035 ds->ds_phys->ds_deadlist_obj = 0; | |
| 1036 | |
| 1037 if (ds_next->ds_phys->ds_next_snap_obj != 0) { | |
| 1038 /* | |
| 1039 * Update next's unique to include blocks which | |
| 1040 * were previously shared by only this snapshot | |
| 1041 * and it. Those blocks will be born after the | |
| 1042 * prev snap and before this snap, and will have | |
| 1043 * died after the next snap and before the one | |
| 1044 * after that (ie. be on the snap after next's | |
| 1045 * deadlist). | |
| 1046 * | |
| 1047 * XXX we're doing this long task with the | |
| 1048 * config lock held | |
| 1049 */ | |
| 1050 dsl_dataset_t *ds_after_next; | |
| 1051 | |
| 1052 ds_after_next = dsl_dataset_open_obj(dd->dd_pool, | |
| 1053 ds_next->ds_phys->ds_next_snap_obj, NULL, | |
| 1054 DS_MODE_NONE, FTAG); | |
| 1055 itor = 0; | |
| 1056 while (bplist_iterate(&ds_after_next->ds_deadlist, | |
| 1057 &itor, &bp) == 0) { | |
| 1058 if (bp.blk_birth > | |
| 1059 ds->ds_phys->ds_prev_snap_txg && | |
| 1060 bp.blk_birth <= | |
| 1061 ds->ds_phys->ds_creation_txg) { | |
| 1062 ds_next->ds_phys->ds_unique_bytes += | |
| 1063 BP_GET_ASIZE(&bp); | |
| 1064 } | |
| 1065 } | |
| 1066 | |
| 1067 dsl_dataset_close(ds_after_next, DS_MODE_NONE, FTAG); | |
| 1068 ASSERT3P(ds_next->ds_prev, ==, NULL); | |
| 1069 } else { | |
| 1070 /* | |
| 1071 * It would be nice to update the head dataset's | |
| 1072 * unique. To do so we would have to traverse | |
| 1073 * it for blocks born after ds_prev, which is | |
| 1074 * pretty expensive just to maintain something | |
| 1075 * for debugging purposes. | |
| 1076 */ | |
| 1077 ASSERT3P(ds_next->ds_prev, ==, ds); | |
| 1078 dsl_dataset_close(ds_next->ds_prev, DS_MODE_NONE, | |
| 1079 ds_next); | |
| 1080 if (ds_prev) { | |
| 1081 ds_next->ds_prev = dsl_dataset_open_obj( | |
| 1082 dd->dd_pool, ds->ds_phys->ds_prev_snap_obj, | |
| 1083 NULL, DS_MODE_NONE, ds_next); | |
| 1084 } else { | |
| 1085 ds_next->ds_prev = NULL; | |
| 1086 } | |
| 1087 } | |
| 1088 dsl_dataset_close(ds_next, DS_MODE_NONE, FTAG); | |
| 1089 | |
| 1090 /* | |
| 1091 * NB: unique_bytes is not accurate for head objsets | |
| 1092 * because we don't update it when we delete the most | |
| 1093 * recent snapshot -- see above comment. | |
| 1094 */ | |
| 1095 ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes); | |
| 1096 } else { | |
| 1097 /* | |
| 1098 * There's no next snapshot, so this is a head dataset. | |
| 1099 * Destroy the deadlist. Unless it's a clone, the | |
| 1100 * deadlist should be empty. (If it's a clone, it's | |
| 1101 * safe to ignore the deadlist contents.) | |
| 1102 */ | |
| 1103 struct killarg ka; | |
| 1104 | |
| 1105 ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist)); | |
| 1106 bplist_close(&ds->ds_deadlist); | |
| 1107 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); | |
| 1108 ds->ds_phys->ds_deadlist_obj = 0; | |
| 1109 | |
| 1110 /* | |
| 1111 * Free everything that we point to (that's born after | |
| 1112 * the previous snapshot, if we are a clone) | |
| 1113 * | |
| 1114 * XXX we're doing this long task with the config lock held | |
| 1115 */ | |
| 1116 ka.usedp = &used; | |
| 1117 ka.compressedp = &compressed; | |
| 1118 ka.uncompressedp = &uncompressed; | |
| 1119 ka.zio = zio; | |
| 1120 ka.tx = tx; | |
| 1121 err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg, | |
| 1122 ADVANCE_POST, kill_blkptr, &ka); | |
| 1123 ASSERT3U(err, ==, 0); | |
| 1124 } | |
| 1125 | |
| 1126 err = zio_wait(zio); | |
| 1127 ASSERT3U(err, ==, 0); | |
| 1128 | |
| 1129 dsl_dir_diduse_space(dd, -used, -compressed, -uncompressed, tx); | |
| 1130 | |
| 1131 if (ds->ds_phys->ds_snapnames_zapobj) { | |
| 1132 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx); | |
| 1133 ASSERT(err == 0); | |
| 1134 } | |
| 1135 | |
| 1136 if (dd->dd_phys->dd_head_dataset_obj == ds->ds_object) { | |
| 1137 /* Erase the link in the dataset */ | |
| 1138 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 1139 dd->dd_phys->dd_head_dataset_obj = 0; | |
| 1140 /* | |
| 1141 * dsl_dir_sync_destroy() called us, they'll destroy | |
| 1142 * the dataset. | |
| 1143 */ | |
| 1144 } else { | |
| 1145 /* remove from snapshot namespace */ | |
| 1146 dsl_dataset_t *ds_head; | |
| 1147 ds_head = dsl_dataset_open_obj(dd->dd_pool, | |
| 1148 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG); | |
| 1149 #ifdef ZFS_DEBUG | |
| 1150 { | |
| 1151 uint64_t val; | |
| 1152 err = zap_lookup(mos, | |
| 1153 ds_head->ds_phys->ds_snapnames_zapobj, | |
| 1154 snapname, 8, 1, &val); | |
| 1155 ASSERT3U(err, ==, 0); | |
| 1156 ASSERT3U(val, ==, obj); | |
| 1157 } | |
| 1158 #endif | |
| 1159 err = zap_remove(mos, ds_head->ds_phys->ds_snapnames_zapobj, | |
| 1160 snapname, tx); | |
| 1161 ASSERT(err == 0); | |
| 1162 dsl_dataset_close(ds_head, DS_MODE_NONE, FTAG); | |
| 1163 } | |
| 1164 | |
| 1165 if (ds_prev && ds->ds_prev != ds_prev) | |
| 1166 dsl_dataset_close(ds_prev, DS_MODE_NONE, FTAG); | |
| 1167 | |
| 1168 err = dmu_object_free(mos, obj, tx); | |
| 1169 ASSERT(err == 0); | |
| 1170 | |
| 1171 /* | |
| 1172 * Close the objset with mode NONE, thus leaving it with | |
| 1173 * DOS_REF_MAX set, so that noone can access it. | |
| 1174 */ | |
| 1175 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 1176 | |
| 1177 if (drop_lock) | |
| 1178 rw_exit(&dp->dp_config_rwlock); | |
| 1179 return (0); | |
| 1180 } | |
| 1181 | |
| 1182 int | |
| 1183 dsl_dataset_snapshot_sync(dsl_dir_t *dd, void *arg, dmu_tx_t *tx) | |
| 1184 { | |
| 1185 const char *snapname = arg; | |
| 1186 dsl_pool_t *dp = dd->dd_pool; | |
| 1187 dmu_buf_t *dbuf; | |
| 1188 dsl_dataset_phys_t *dsphys; | |
| 1189 uint64_t dsobj, value; | |
| 1190 objset_t *mos = dp->dp_meta_objset; | |
| 1191 dsl_dataset_t *ds; | |
| 1192 int err; | |
| 1193 | |
| 1194 ASSERT(dmu_tx_is_syncing(tx)); | |
| 1195 | |
| 1196 if (dd->dd_phys->dd_head_dataset_obj == 0) | |
| 1197 return (EINVAL); | |
| 1198 ds = dsl_dataset_open_obj(dp, dd->dd_phys->dd_head_dataset_obj, NULL, | |
| 1199 DS_MODE_NONE, FTAG); | |
| 1200 | |
| 1201 err = zap_lookup(mos, ds->ds_phys->ds_snapnames_zapobj, | |
| 1202 snapname, 8, 1, &value); | |
| 1203 if (err == 0) { | |
| 1204 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 1205 return (EEXIST); | |
| 1206 } | |
| 1207 ASSERT(err == ENOENT); | |
| 1208 | |
| 1209 /* The point of no (unsuccessful) return */ | |
| 1210 | |
| 1211 dprintf_dd(dd, "taking snapshot %s in txg %llu\n", | |
| 1212 snapname, tx->tx_txg); | |
| 1213 | |
| 1214 spa_scrub_restart(dp->dp_spa, tx->tx_txg); | |
| 1215 | |
| 1216 rw_enter(&dp->dp_config_rwlock, RW_WRITER); | |
| 1217 | |
| 1218 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_OBJSET, 0, | |
| 1219 DMU_OT_DSL_OBJSET, sizeof (dsl_dataset_phys_t), tx); | |
| 1220 dbuf = dmu_bonus_hold(mos, dsobj); | |
| 1221 dmu_buf_will_dirty(dbuf, tx); | |
| 1222 dsphys = dbuf->db_data; | |
| 1223 dsphys->ds_dir_obj = dd->dd_object; | |
| 1224 dsphys->ds_fsid_guid = unique_create(); | |
| 1225 unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */ | |
| 1226 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, | |
| 1227 sizeof (dsphys->ds_guid)); | |
| 1228 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; | |
| 1229 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; | |
| 1230 dsphys->ds_next_snap_obj = ds->ds_object; | |
| 1231 dsphys->ds_num_children = 1; | |
| 1232 dsphys->ds_creation_time = gethrestime_sec(); | |
| 1233 dsphys->ds_creation_txg = tx->tx_txg; | |
| 1234 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; | |
| 1235 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes; | |
| 1236 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; | |
| 1237 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; | |
| 1238 dsphys->ds_restoring = ds->ds_phys->ds_restoring; | |
| 1239 dsphys->ds_bp = ds->ds_phys->ds_bp; | |
| 1240 dmu_buf_rele(dbuf); | |
| 1241 | |
| 1242 if (ds->ds_phys->ds_prev_snap_obj != 0) { | |
| 1243 dsl_dataset_t *ds_prev; | |
| 1244 | |
| 1245 ds_prev = dsl_dataset_open_obj(dp, | |
| 1246 ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_NONE, FTAG); | |
| 1247 ASSERT(ds_prev->ds_phys->ds_next_snap_obj == | |
| 1248 ds->ds_object || | |
| 1249 ds_prev->ds_phys->ds_num_children > 1); | |
| 1250 if (ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { | |
| 1251 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); | |
| 1252 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, | |
| 1253 ds_prev->ds_phys->ds_creation_txg); | |
| 1254 ds_prev->ds_phys->ds_next_snap_obj = dsobj; | |
| 1255 } | |
| 1256 dsl_dataset_close(ds_prev, DS_MODE_NONE, FTAG); | |
| 1257 } else { | |
| 1258 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 0); | |
| 1259 } | |
| 1260 | |
| 1261 bplist_close(&ds->ds_deadlist); | |
| 1262 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 1263 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, dsphys->ds_creation_txg); | |
| 1264 ds->ds_phys->ds_prev_snap_obj = dsobj; | |
| 1265 ds->ds_phys->ds_prev_snap_txg = dsphys->ds_creation_txg; | |
| 1266 ds->ds_phys->ds_unique_bytes = 0; | |
| 1267 ds->ds_phys->ds_deadlist_obj = | |
| 1268 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 1269 bplist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); | |
| 1270 | |
| 1271 dprintf("snap '%s' -> obj %llu\n", snapname, dsobj); | |
| 1272 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, | |
| 1273 snapname, 8, 1, &dsobj, tx); | |
| 1274 ASSERT(err == 0); | |
| 1275 | |
| 1276 if (ds->ds_prev) | |
| 1277 dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds); | |
| 1278 ds->ds_prev = dsl_dataset_open_obj(dp, | |
| 1279 ds->ds_phys->ds_prev_snap_obj, snapname, DS_MODE_NONE, ds); | |
| 1280 | |
| 1281 rw_exit(&dp->dp_config_rwlock); | |
| 1282 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 1283 | |
| 1284 return (0); | |
| 1285 } | |
| 1286 | |
| 1287 void | |
| 1288 dsl_dataset_sync(dsl_dataset_t *ds, dmu_tx_t *tx) | |
| 1289 { | |
| 1290 ASSERT(dmu_tx_is_syncing(tx)); | |
| 1291 ASSERT(ds->ds_user_ptr != NULL); | |
| 1292 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); | |
| 1293 | |
| 1294 dmu_objset_sync(ds->ds_user_ptr, tx); | |
| 1295 dsl_dir_dirty(ds->ds_dir, tx); | |
| 1296 bplist_close(&ds->ds_deadlist); | |
| 1297 | |
| 1298 dmu_buf_remove_ref(ds->ds_dbuf, ds); | |
| 1299 } | |
| 1300 | |
| 1301 void | |
| 1302 dsl_dataset_stats(dsl_dataset_t *ds, dmu_objset_stats_t *dds) | |
| 1303 { | |
| 1304 /* fill in properties crap */ | |
| 1305 dsl_dir_stats(ds->ds_dir, dds); | |
| 1306 | |
| 1307 if (ds->ds_phys->ds_num_children != 0) { | |
| 1308 dds->dds_is_snapshot = TRUE; | |
| 1309 dds->dds_num_clones = ds->ds_phys->ds_num_children - 1; | |
| 1310 } | |
| 1311 | |
| 1312 dds->dds_last_txg = ds->ds_phys->ds_bp.blk_birth; | |
| 1313 | |
| 1314 dds->dds_objects_used = ds->ds_phys->ds_bp.blk_fill; | |
| 1315 dds->dds_objects_avail = DN_MAX_OBJECT - dds->dds_objects_used; | |
| 1316 | |
| 1317 /* We override the dataset's creation time... they should be the same */ | |
| 1318 dds->dds_creation_time = ds->ds_phys->ds_creation_time; | |
| 1319 dds->dds_creation_txg = ds->ds_phys->ds_creation_txg; | |
| 1320 dds->dds_space_refd = ds->ds_phys->ds_used_bytes; | |
| 1321 dds->dds_fsid_guid = ds->ds_phys->ds_fsid_guid; | |
| 1322 dds->dds_guid = ds->ds_phys->ds_guid; | |
| 1323 | |
| 1324 if (ds->ds_phys->ds_next_snap_obj) { | |
| 1325 /* | |
| 1326 * This is a snapshot; override the dd's space used with | |
| 1327 * our unique space | |
| 1328 */ | |
| 1329 dds->dds_space_used = ds->ds_phys->ds_unique_bytes; | |
| 1330 dds->dds_compressed_bytes = | |
| 1331 ds->ds_phys->ds_compressed_bytes; | |
| 1332 dds->dds_uncompressed_bytes = | |
| 1333 ds->ds_phys->ds_uncompressed_bytes; | |
| 1334 } | |
| 1335 | |
| 1336 dds->dds_objset_obj = ds->ds_object; | |
| 1337 } | |
| 1338 | |
| 1339 dsl_pool_t * | |
| 1340 dsl_dataset_pool(dsl_dataset_t *ds) | |
| 1341 { | |
| 1342 return (ds->ds_dir->dd_pool); | |
| 1343 } | |
| 1344 | |
| 1345 struct osrenamearg { | |
| 1346 const char *oldname; | |
| 1347 const char *newname; | |
| 1348 }; | |
| 1349 | |
| 1350 static int | |
| 1351 dsl_dataset_snapshot_rename_sync(dsl_dir_t *dd, void *arg, dmu_tx_t *tx) | |
| 1352 { | |
| 1353 struct osrenamearg *ora = arg; | |
| 1354 objset_t *mos = dd->dd_pool->dp_meta_objset; | |
| 1355 dsl_dir_t *nds; | |
| 1356 const char *tail; | |
| 1357 int err; | |
| 1358 dsl_dataset_t *snds, *fsds; | |
| 1359 uint64_t val; | |
| 1360 | |
| 1361 err = dsl_dataset_open_spa(dd->dd_pool->dp_spa, ora->oldname, | |
| 1362 DS_MODE_READONLY | DS_MODE_STANDARD, FTAG, &snds); | |
| 1363 if (err) | |
| 1364 return (err); | |
| 1365 | |
| 1366 if (snds->ds_dir != dd) { | |
| 1367 dsl_dataset_close(snds, DS_MODE_STANDARD, FTAG); | |
| 1368 return (EINVAL); | |
| 1369 } | |
| 1370 | |
| 1371 /* better be changing a snapshot */ | |
| 1372 if (snds->ds_phys->ds_next_snap_obj == 0) { | |
| 1373 dsl_dataset_close(snds, DS_MODE_STANDARD, FTAG); | |
| 1374 return (EINVAL); | |
| 1375 } | |
| 1376 | |
| 1377 /* new fs better exist */ | |
| 1378 nds = dsl_dir_open_spa(dd->dd_pool->dp_spa, ora->newname, FTAG, &tail); | |
| 1379 if (nds == NULL) { | |
| 1380 dsl_dataset_close(snds, DS_MODE_STANDARD, FTAG); | |
| 1381 return (ENOENT); | |
| 1382 } | |
| 1383 | |
| 1384 dsl_dir_close(nds, FTAG); | |
| 1385 | |
| 1386 /* new name better be in same fs */ | |
| 1387 if (nds != dd) { | |
| 1388 dsl_dataset_close(snds, DS_MODE_STANDARD, FTAG); | |
| 1389 return (EINVAL); | |
| 1390 } | |
| 1391 | |
| 1392 /* new name better be a snapshot */ | |
| 1393 if (tail == NULL || tail[0] != '@') { | |
| 1394 dsl_dataset_close(snds, DS_MODE_STANDARD, FTAG); | |
| 1395 return (EINVAL); | |
| 1396 } | |
| 1397 | |
| 1398 tail++; | |
| 1399 | |
| 1400 fsds = dsl_dataset_open_obj(dd->dd_pool, | |
| 1401 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG); | |
| 1402 | |
| 1403 /* new name better not be in use */ | |
| 1404 err = zap_lookup(mos, fsds->ds_phys->ds_snapnames_zapobj, | |
| 1405 tail, 8, 1, &val); | |
| 1406 if (err != ENOENT) { | |
| 1407 if (err == 0) | |
| 1408 err = EEXIST; | |
| 1409 dsl_dataset_close(fsds, DS_MODE_NONE, FTAG); | |
| 1410 dsl_dataset_close(snds, DS_MODE_STANDARD, FTAG); | |
| 1411 return (EEXIST); | |
| 1412 } | |
| 1413 | |
| 1414 /* The point of no (unsuccessful) return */ | |
| 1415 | |
| 1416 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_WRITER); | |
| 1417 dsl_dataset_get_snapname(snds); | |
| 1418 err = zap_remove(mos, fsds->ds_phys->ds_snapnames_zapobj, | |
| 1419 snds->ds_snapname, tx); | |
| 1420 ASSERT3U(err, ==, 0); | |
| 1421 mutex_enter(&snds->ds_lock); | |
| 1422 (void) strcpy(snds->ds_snapname, tail); | |
| 1423 mutex_exit(&snds->ds_lock); | |
| 1424 err = zap_add(mos, fsds->ds_phys->ds_snapnames_zapobj, | |
| 1425 snds->ds_snapname, 8, 1, &snds->ds_object, tx); | |
| 1426 ASSERT3U(err, ==, 0); | |
| 1427 rw_exit(&dd->dd_pool->dp_config_rwlock); | |
| 1428 | |
| 1429 dsl_dataset_close(fsds, DS_MODE_NONE, FTAG); | |
| 1430 dsl_dataset_close(snds, DS_MODE_STANDARD, FTAG); | |
| 1431 return (0); | |
| 1432 } | |
| 1433 | |
| 1434 #pragma weak dmu_objset_rename = dsl_dataset_rename | |
| 1435 int | |
| 1436 dsl_dataset_rename(const char *osname, const char *newname) | |
| 1437 { | |
| 1438 dsl_dir_t *dd; | |
| 1439 const char *tail; | |
| 1440 struct osrenamearg ora; | |
| 1441 int err; | |
| 1442 | |
| 1443 dd = dsl_dir_open(osname, FTAG, &tail); | |
| 1444 if (dd == NULL) | |
| 1445 return (ENOENT); | |
| 1446 if (tail == NULL) { | |
| 1447 err = dsl_dir_sync_task(dd, | |
| 1448 dsl_dir_rename_sync, (void*)newname, 1<<12); | |
| 1449 dsl_dir_close(dd, FTAG); | |
| 1450 return (err); | |
| 1451 } | |
| 1452 if (tail[0] != '@') { | |
| 1453 /* the name ended in a nonexistant component */ | |
| 1454 dsl_dir_close(dd, FTAG); | |
| 1455 return (ENOENT); | |
| 1456 } | |
| 1457 | |
| 1458 ora.oldname = osname; | |
| 1459 ora.newname = newname; | |
| 1460 | |
| 1461 err = dsl_dir_sync_task(dd, | |
| 1462 dsl_dataset_snapshot_rename_sync, &ora, 1<<12); | |
| 1463 dsl_dir_close(dd, FTAG); | |
| 1464 return (err); | |
| 1465 } |