Mercurial > illumos > illumos-gate
annotate usr/src/uts/common/fs/zfs/dsl_dataset.c @ 3978:2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
| author | mmusante |
|---|---|
| date | Thu, 05 Apr 2007 20:34:28 -0700 |
| parents | f6891a60bd72 |
| children | c6f5c6753018 |
| 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 /* | |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
22 * Copyright 2007 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 |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
48 #define DS_REF_MAX (1ULL << 62) |
| 789 | 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 | |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
55 * is DS_REF_MAX, and a PRIMARY open is little more than half of an EXCLUSIVE. |
| 789 | 56 * This makes the exclusion logic simple: the total refcnt for all opens cannot |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
57 * exceed DS_REF_MAX. For example, EXCLUSIVE opens are exclusive because their |
|
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
58 * weight (DS_REF_MAX) consumes the entire refcnt space. PRIMARY opens consume |
| 789 | 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] = { | |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
63 0, /* DS_MODE_NONE - invalid */ |
|
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
64 1, /* DS_MODE_STANDARD - unlimited number */ |
|
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
65 (DS_REF_MAX >> 1) + 1, /* DS_MODE_PRIMARY - only one of these */ |
|
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
66 DS_REF_MAX /* DS_MODE_EXCLUSIVE - no other opens */ |
| 789 | 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 | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
108 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio, |
|
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
109 dmu_tx_t *tx) |
| 789 | 110 { |
| 2082 | 111 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp); |
| 789 | 112 int compressed = BP_GET_PSIZE(bp); |
| 113 int uncompressed = BP_GET_UCSIZE(bp); | |
| 114 | |
| 115 ASSERT(dmu_tx_is_syncing(tx)); | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
116 /* No block pointer => nothing to free */ |
| 789 | 117 if (BP_IS_HOLE(bp)) |
| 118 return; | |
| 119 | |
| 120 ASSERT(used > 0); | |
| 121 if (ds == NULL) { | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
122 int err; |
| 789 | 123 /* |
| 124 * Account for the meta-objset space in its placeholder | |
| 125 * dataset. | |
| 126 */ | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
127 err = arc_free(pio, tx->tx_pool->dp_spa, |
|
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
128 tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT); |
|
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
129 ASSERT(err == 0); |
| 789 | 130 |
| 131 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, | |
| 132 -used, -compressed, -uncompressed, tx); | |
| 133 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); | |
| 134 return; | |
| 135 } | |
| 136 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); | |
| 137 | |
| 138 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 139 | |
| 140 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
141 int err; |
|
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
142 |
| 789 | 143 dprintf_bp(bp, "freeing: %s", ""); |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
144 err = arc_free(pio, tx->tx_pool->dp_spa, |
|
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
145 tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT); |
|
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
146 ASSERT(err == 0); |
| 789 | 147 |
| 148 mutex_enter(&ds->ds_lock); | |
| 149 /* XXX unique_bytes is not accurate for head datasets */ | |
| 150 /* ASSERT3U(ds->ds_phys->ds_unique_bytes, >=, used); */ | |
| 151 ds->ds_phys->ds_unique_bytes -= used; | |
| 152 mutex_exit(&ds->ds_lock); | |
| 153 dsl_dir_diduse_space(ds->ds_dir, | |
| 154 -used, -compressed, -uncompressed, tx); | |
| 155 } else { | |
| 156 dprintf_bp(bp, "putting on dead list: %s", ""); | |
| 1544 | 157 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx)); |
| 789 | 158 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ |
| 159 if (ds->ds_phys->ds_prev_snap_obj != 0) { | |
| 160 ASSERT3U(ds->ds_prev->ds_object, ==, | |
| 161 ds->ds_phys->ds_prev_snap_obj); | |
| 162 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); | |
| 163 if (ds->ds_prev->ds_phys->ds_next_snap_obj == | |
| 2082 | 164 ds->ds_object && bp->blk_birth > |
| 789 | 165 ds->ds_prev->ds_phys->ds_prev_snap_txg) { |
| 166 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
| 167 mutex_enter(&ds->ds_prev->ds_lock); | |
| 168 ds->ds_prev->ds_phys->ds_unique_bytes += | |
| 169 used; | |
| 170 mutex_exit(&ds->ds_prev->ds_lock); | |
| 171 } | |
| 172 } | |
| 173 } | |
| 174 mutex_enter(&ds->ds_lock); | |
| 175 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used); | |
| 176 ds->ds_phys->ds_used_bytes -= used; | |
| 177 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); | |
| 178 ds->ds_phys->ds_compressed_bytes -= compressed; | |
| 179 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); | |
| 180 ds->ds_phys->ds_uncompressed_bytes -= uncompressed; | |
| 181 mutex_exit(&ds->ds_lock); | |
| 182 } | |
| 183 | |
| 1544 | 184 uint64_t |
| 185 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) | |
| 789 | 186 { |
| 2885 | 187 uint64_t trysnap = 0; |
| 188 | |
| 789 | 189 if (ds == NULL) |
| 1544 | 190 return (0); |
| 789 | 191 /* |
| 192 * The snapshot creation could fail, but that would cause an | |
| 193 * incorrect FALSE return, which would only result in an | |
| 194 * overestimation of the amount of space that an operation would | |
| 195 * consume, which is OK. | |
| 196 * | |
| 197 * There's also a small window where we could miss a pending | |
| 198 * snapshot, because we could set the sync task in the quiescing | |
| 199 * phase. So this should only be used as a guess. | |
| 200 */ | |
| 2885 | 201 if (ds->ds_trysnap_txg > |
| 202 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) | |
| 203 trysnap = ds->ds_trysnap_txg; | |
| 204 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap)); | |
| 1544 | 205 } |
| 206 | |
| 207 int | |
| 208 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth) | |
| 209 { | |
| 210 return (blk_birth > dsl_dataset_prev_snap_txg(ds)); | |
| 789 | 211 } |
| 212 | |
| 213 /* ARGSUSED */ | |
| 214 static void | |
| 215 dsl_dataset_evict(dmu_buf_t *db, void *dsv) | |
| 216 { | |
| 217 dsl_dataset_t *ds = dsv; | |
| 218 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 219 | |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
220 /* open_refcount == DS_REF_MAX when deleting */ |
| 789 | 221 ASSERT(ds->ds_open_refcount == 0 || |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
222 ds->ds_open_refcount == DS_REF_MAX); |
| 789 | 223 |
| 224 dprintf_ds(ds, "evicting %s\n", ""); | |
| 225 | |
| 226 unique_remove(ds->ds_phys->ds_fsid_guid); | |
| 227 | |
| 228 if (ds->ds_user_ptr != NULL) | |
| 229 ds->ds_user_evict_func(ds, ds->ds_user_ptr); | |
| 230 | |
| 231 if (ds->ds_prev) { | |
| 232 dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds); | |
| 233 ds->ds_prev = NULL; | |
| 234 } | |
| 235 | |
| 236 bplist_close(&ds->ds_deadlist); | |
| 237 dsl_dir_close(ds->ds_dir, ds); | |
| 238 | |
| 239 if (list_link_active(&ds->ds_synced_link)) | |
| 240 list_remove(&dp->dp_synced_objsets, ds); | |
| 241 | |
| 2856 | 242 mutex_destroy(&ds->ds_lock); |
| 243 mutex_destroy(&ds->ds_deadlist.bpl_lock); | |
| 244 | |
| 789 | 245 kmem_free(ds, sizeof (dsl_dataset_t)); |
| 246 } | |
| 247 | |
| 1544 | 248 static int |
| 789 | 249 dsl_dataset_get_snapname(dsl_dataset_t *ds) |
| 250 { | |
| 251 dsl_dataset_phys_t *headphys; | |
| 252 int err; | |
| 253 dmu_buf_t *headdbuf; | |
| 254 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 255 objset_t *mos = dp->dp_meta_objset; | |
| 256 | |
| 257 if (ds->ds_snapname[0]) | |
| 1544 | 258 return (0); |
| 789 | 259 if (ds->ds_phys->ds_next_snap_obj == 0) |
| 1544 | 260 return (0); |
| 789 | 261 |
| 1544 | 262 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, |
| 263 FTAG, &headdbuf); | |
| 264 if (err) | |
| 265 return (err); | |
| 789 | 266 headphys = headdbuf->db_data; |
| 267 err = zap_value_search(dp->dp_meta_objset, | |
| 268 headphys->ds_snapnames_zapobj, ds->ds_object, ds->ds_snapname); | |
| 1544 | 269 dmu_buf_rele(headdbuf, FTAG); |
| 270 return (err); | |
| 789 | 271 } |
| 272 | |
| 1544 | 273 int |
| 789 | 274 dsl_dataset_open_obj(dsl_pool_t *dp, uint64_t dsobj, const char *snapname, |
| 1544 | 275 int mode, void *tag, dsl_dataset_t **dsp) |
| 789 | 276 { |
| 277 uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)]; | |
| 278 objset_t *mos = dp->dp_meta_objset; | |
| 279 dmu_buf_t *dbuf; | |
| 280 dsl_dataset_t *ds; | |
| 1544 | 281 int err; |
| 789 | 282 |
| 283 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || | |
| 284 dsl_pool_sync_context(dp)); | |
| 285 | |
| 1544 | 286 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); |
| 287 if (err) | |
| 288 return (err); | |
| 789 | 289 ds = dmu_buf_get_user(dbuf); |
| 290 if (ds == NULL) { | |
| 291 dsl_dataset_t *winner; | |
| 292 | |
| 293 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); | |
| 294 ds->ds_dbuf = dbuf; | |
| 295 ds->ds_object = dsobj; | |
| 296 ds->ds_phys = dbuf->db_data; | |
| 297 | |
| 2856 | 298 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); |
| 299 mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT, | |
| 300 NULL); | |
| 301 | |
| 1544 | 302 err = bplist_open(&ds->ds_deadlist, |
| 789 | 303 mos, ds->ds_phys->ds_deadlist_obj); |
| 1544 | 304 if (err == 0) { |
| 305 err = dsl_dir_open_obj(dp, | |
| 306 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); | |
| 307 } | |
| 308 if (err) { | |
| 309 /* | |
| 310 * we don't really need to close the blist if we | |
| 311 * just opened it. | |
| 312 */ | |
| 2856 | 313 mutex_destroy(&ds->ds_lock); |
| 314 mutex_destroy(&ds->ds_deadlist.bpl_lock); | |
| 1544 | 315 kmem_free(ds, sizeof (dsl_dataset_t)); |
| 316 dmu_buf_rele(dbuf, tag); | |
| 317 return (err); | |
| 318 } | |
| 789 | 319 |
| 320 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == dsobj) { | |
| 321 ds->ds_snapname[0] = '\0'; | |
| 322 if (ds->ds_phys->ds_prev_snap_obj) { | |
| 1544 | 323 err = dsl_dataset_open_obj(dp, |
| 789 | 324 ds->ds_phys->ds_prev_snap_obj, NULL, |
| 1544 | 325 DS_MODE_NONE, ds, &ds->ds_prev); |
| 789 | 326 } |
| 327 } else { | |
| 328 if (snapname) { | |
| 329 #ifdef ZFS_DEBUG | |
| 330 dsl_dataset_phys_t *headphys; | |
| 1544 | 331 dmu_buf_t *headdbuf; |
| 332 err = dmu_bonus_hold(mos, | |
| 333 ds->ds_dir->dd_phys->dd_head_dataset_obj, | |
| 334 FTAG, &headdbuf); | |
| 335 if (err == 0) { | |
| 336 headphys = headdbuf->db_data; | |
| 337 uint64_t foundobj; | |
| 338 err = zap_lookup(dp->dp_meta_objset, | |
| 339 headphys->ds_snapnames_zapobj, | |
| 340 snapname, sizeof (foundobj), 1, | |
| 341 &foundobj); | |
| 342 ASSERT3U(foundobj, ==, dsobj); | |
| 343 dmu_buf_rele(headdbuf, FTAG); | |
| 344 } | |
| 789 | 345 #endif |
| 346 (void) strcat(ds->ds_snapname, snapname); | |
| 347 } else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) { | |
| 1544 | 348 err = dsl_dataset_get_snapname(ds); |
| 789 | 349 } |
| 350 } | |
| 351 | |
| 1544 | 352 if (err == 0) { |
| 353 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys, | |
| 354 dsl_dataset_evict); | |
| 355 } | |
| 356 if (err || winner) { | |
| 789 | 357 bplist_close(&ds->ds_deadlist); |
| 358 if (ds->ds_prev) { | |
| 359 dsl_dataset_close(ds->ds_prev, | |
| 360 DS_MODE_NONE, ds); | |
| 361 } | |
| 362 dsl_dir_close(ds->ds_dir, ds); | |
| 2856 | 363 mutex_destroy(&ds->ds_lock); |
| 364 mutex_destroy(&ds->ds_deadlist.bpl_lock); | |
| 789 | 365 kmem_free(ds, sizeof (dsl_dataset_t)); |
| 1544 | 366 if (err) { |
| 367 dmu_buf_rele(dbuf, tag); | |
| 368 return (err); | |
| 369 } | |
| 789 | 370 ds = winner; |
| 371 } else { | |
| 372 uint64_t new = | |
| 373 unique_insert(ds->ds_phys->ds_fsid_guid); | |
| 374 if (new != ds->ds_phys->ds_fsid_guid) { | |
| 375 /* XXX it won't necessarily be synced... */ | |
| 376 ds->ds_phys->ds_fsid_guid = new; | |
| 377 } | |
| 378 } | |
| 379 } | |
| 380 ASSERT3P(ds->ds_dbuf, ==, dbuf); | |
| 381 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); | |
| 382 | |
| 383 mutex_enter(&ds->ds_lock); | |
| 384 if ((DS_MODE_LEVEL(mode) == DS_MODE_PRIMARY && | |
| 2082 | 385 (ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) && |
| 386 !DS_MODE_IS_INCONSISTENT(mode)) || | |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
387 (ds->ds_open_refcount + weight > DS_REF_MAX)) { |
| 789 | 388 mutex_exit(&ds->ds_lock); |
| 389 dsl_dataset_close(ds, DS_MODE_NONE, tag); | |
| 1544 | 390 return (EBUSY); |
| 789 | 391 } |
| 392 ds->ds_open_refcount += weight; | |
| 393 mutex_exit(&ds->ds_lock); | |
| 394 | |
| 1544 | 395 *dsp = ds; |
| 396 return (0); | |
| 789 | 397 } |
| 398 | |
| 399 int | |
| 400 dsl_dataset_open_spa(spa_t *spa, const char *name, int mode, | |
| 401 void *tag, dsl_dataset_t **dsp) | |
| 402 { | |
| 403 dsl_dir_t *dd; | |
| 404 dsl_pool_t *dp; | |
| 405 const char *tail; | |
| 406 uint64_t obj; | |
| 407 dsl_dataset_t *ds = NULL; | |
| 408 int err = 0; | |
| 409 | |
| 1544 | 410 err = dsl_dir_open_spa(spa, name, FTAG, &dd, &tail); |
| 411 if (err) | |
| 412 return (err); | |
| 789 | 413 |
| 414 dp = dd->dd_pool; | |
| 415 obj = dd->dd_phys->dd_head_dataset_obj; | |
| 416 rw_enter(&dp->dp_config_rwlock, RW_READER); | |
| 417 if (obj == 0) { | |
| 418 /* A dataset with no associated objset */ | |
| 419 err = ENOENT; | |
| 420 goto out; | |
| 421 } | |
| 422 | |
| 423 if (tail != NULL) { | |
| 424 objset_t *mos = dp->dp_meta_objset; | |
| 425 | |
| 1544 | 426 err = dsl_dataset_open_obj(dp, obj, NULL, |
| 427 DS_MODE_NONE, tag, &ds); | |
| 428 if (err) | |
| 429 goto out; | |
| 789 | 430 obj = ds->ds_phys->ds_snapnames_zapobj; |
| 431 dsl_dataset_close(ds, DS_MODE_NONE, tag); | |
| 432 ds = NULL; | |
| 433 | |
| 434 if (tail[0] != '@') { | |
| 435 err = ENOENT; | |
| 436 goto out; | |
| 437 } | |
| 438 tail++; | |
| 439 | |
| 440 /* Look for a snapshot */ | |
| 441 if (!DS_MODE_IS_READONLY(mode)) { | |
| 442 err = EROFS; | |
| 443 goto out; | |
| 444 } | |
| 445 dprintf("looking for snapshot '%s'\n", tail); | |
| 446 err = zap_lookup(mos, obj, tail, 8, 1, &obj); | |
| 447 if (err) | |
| 448 goto out; | |
| 449 } | |
| 1544 | 450 err = dsl_dataset_open_obj(dp, obj, tail, mode, tag, &ds); |
| 789 | 451 |
| 452 out: | |
| 453 rw_exit(&dp->dp_config_rwlock); | |
| 454 dsl_dir_close(dd, FTAG); | |
| 455 | |
| 456 ASSERT3U((err == 0), ==, (ds != NULL)); | |
| 457 /* ASSERT(ds == NULL || strcmp(name, ds->ds_name) == 0); */ | |
| 458 | |
| 459 *dsp = ds; | |
| 460 return (err); | |
| 461 } | |
| 462 | |
| 463 int | |
| 464 dsl_dataset_open(const char *name, int mode, void *tag, dsl_dataset_t **dsp) | |
| 465 { | |
| 466 return (dsl_dataset_open_spa(NULL, name, mode, tag, dsp)); | |
| 467 } | |
| 468 | |
| 469 void | |
| 470 dsl_dataset_name(dsl_dataset_t *ds, char *name) | |
| 471 { | |
| 472 if (ds == NULL) { | |
| 473 (void) strcpy(name, "mos"); | |
| 474 } else { | |
| 475 dsl_dir_name(ds->ds_dir, name); | |
| 1544 | 476 VERIFY(0 == dsl_dataset_get_snapname(ds)); |
| 789 | 477 if (ds->ds_snapname[0]) { |
| 478 (void) strcat(name, "@"); | |
| 479 if (!MUTEX_HELD(&ds->ds_lock)) { | |
| 480 /* | |
| 481 * We use a "recursive" mutex so that we | |
| 482 * can call dprintf_ds() with ds_lock held. | |
| 483 */ | |
| 484 mutex_enter(&ds->ds_lock); | |
| 485 (void) strcat(name, ds->ds_snapname); | |
| 486 mutex_exit(&ds->ds_lock); | |
| 487 } else { | |
| 488 (void) strcat(name, ds->ds_snapname); | |
| 489 } | |
| 490 } | |
| 491 } | |
| 492 } | |
| 493 | |
|
3978
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
494 static int |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
495 dsl_dataset_namelen(dsl_dataset_t *ds) |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
496 { |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
497 int result; |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
498 |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
499 if (ds == NULL) { |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
500 result = 3; /* "mos" */ |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
501 } else { |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
502 result = dsl_dir_namelen(ds->ds_dir); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
503 VERIFY(0 == dsl_dataset_get_snapname(ds)); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
504 if (ds->ds_snapname[0]) { |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
505 ++result; /* adding one for the @-sign */ |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
506 if (!MUTEX_HELD(&ds->ds_lock)) { |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
507 /* see dsl_datset_name */ |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
508 mutex_enter(&ds->ds_lock); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
509 result += strlen(ds->ds_snapname); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
510 mutex_exit(&ds->ds_lock); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
511 } else { |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
512 result += strlen(ds->ds_snapname); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
513 } |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
514 } |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
515 } |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
516 |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
517 return (result); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
518 } |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
519 |
| 789 | 520 void |
| 521 dsl_dataset_close(dsl_dataset_t *ds, int mode, void *tag) | |
| 522 { | |
| 523 uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)]; | |
| 524 mutex_enter(&ds->ds_lock); | |
| 525 ASSERT3U(ds->ds_open_refcount, >=, weight); | |
| 526 ds->ds_open_refcount -= weight; | |
| 527 dprintf_ds(ds, "closing mode %u refcount now 0x%llx\n", | |
| 528 mode, ds->ds_open_refcount); | |
| 529 mutex_exit(&ds->ds_lock); | |
| 530 | |
| 1544 | 531 dmu_buf_rele(ds->ds_dbuf, tag); |
| 789 | 532 } |
| 533 | |
| 534 void | |
| 535 dsl_dataset_create_root(dsl_pool_t *dp, uint64_t *ddobjp, dmu_tx_t *tx) | |
| 536 { | |
| 537 objset_t *mos = dp->dp_meta_objset; | |
| 538 dmu_buf_t *dbuf; | |
| 539 dsl_dataset_phys_t *dsphys; | |
| 540 dsl_dataset_t *ds; | |
| 541 uint64_t dsobj; | |
| 542 dsl_dir_t *dd; | |
| 543 | |
| 544 dsl_dir_create_root(mos, ddobjp, tx); | |
| 1544 | 545 VERIFY(0 == dsl_dir_open_obj(dp, *ddobjp, NULL, FTAG, &dd)); |
| 789 | 546 |
|
928
36d72fe4da29
6349314 dmu_object_type names incorrect for DSL Directories and DSL Datasets
tabriz
parents:
885
diff
changeset
|
547 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
|
548 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); |
| 1544 | 549 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); |
| 789 | 550 dmu_buf_will_dirty(dbuf, tx); |
| 551 dsphys = dbuf->db_data; | |
| 552 dsphys->ds_dir_obj = dd->dd_object; | |
| 553 dsphys->ds_fsid_guid = unique_create(); | |
| 1544 | 554 unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */ |
| 789 | 555 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, |
| 556 sizeof (dsphys->ds_guid)); | |
| 557 dsphys->ds_snapnames_zapobj = | |
|
885
d925b21dba78
6347493 tar of 25K empty directory entries in ZFS takes 30+ seconds ...
ahrens
parents:
789
diff
changeset
|
558 zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx); |
| 789 | 559 dsphys->ds_creation_time = gethrestime_sec(); |
| 560 dsphys->ds_creation_txg = tx->tx_txg; | |
| 561 dsphys->ds_deadlist_obj = | |
| 562 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 1544 | 563 dmu_buf_rele(dbuf, FTAG); |
| 789 | 564 |
| 565 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 566 dd->dd_phys->dd_head_dataset_obj = dsobj; | |
| 567 dsl_dir_close(dd, FTAG); | |
| 568 | |
| 1544 | 569 VERIFY(0 == |
| 570 dsl_dataset_open_obj(dp, dsobj, NULL, DS_MODE_NONE, FTAG, &ds)); | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
571 (void) dmu_objset_create_impl(dp->dp_spa, ds, |
|
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
572 &ds->ds_phys->ds_bp, DMU_OST_ZFS, tx); |
| 789 | 573 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); |
| 574 } | |
| 575 | |
| 2199 | 576 uint64_t |
| 577 dsl_dataset_create_sync(dsl_dir_t *pdd, | |
| 789 | 578 const char *lastname, dsl_dataset_t *clone_parent, dmu_tx_t *tx) |
| 579 { | |
| 2199 | 580 dsl_pool_t *dp = pdd->dd_pool; |
| 789 | 581 dmu_buf_t *dbuf; |
| 582 dsl_dataset_phys_t *dsphys; | |
| 2199 | 583 uint64_t dsobj, ddobj; |
| 789 | 584 objset_t *mos = dp->dp_meta_objset; |
| 585 dsl_dir_t *dd; | |
| 586 | |
| 2199 | 587 ASSERT(clone_parent == NULL || clone_parent->ds_dir->dd_pool == dp); |
| 588 ASSERT(clone_parent == NULL || | |
| 589 clone_parent->ds_phys->ds_num_children > 0); | |
| 789 | 590 ASSERT(lastname[0] != '@'); |
| 591 ASSERT(dmu_tx_is_syncing(tx)); | |
| 592 | |
| 2199 | 593 ddobj = dsl_dir_create_sync(pdd, lastname, tx); |
| 594 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd)); | |
| 789 | 595 |
|
928
36d72fe4da29
6349314 dmu_object_type names incorrect for DSL Directories and DSL Datasets
tabriz
parents:
885
diff
changeset
|
596 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
|
597 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); |
| 1544 | 598 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); |
| 789 | 599 dmu_buf_will_dirty(dbuf, tx); |
| 600 dsphys = dbuf->db_data; | |
| 601 dsphys->ds_dir_obj = dd->dd_object; | |
| 602 dsphys->ds_fsid_guid = unique_create(); | |
| 603 unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */ | |
| 604 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, | |
| 605 sizeof (dsphys->ds_guid)); | |
| 606 dsphys->ds_snapnames_zapobj = | |
|
885
d925b21dba78
6347493 tar of 25K empty directory entries in ZFS takes 30+ seconds ...
ahrens
parents:
789
diff
changeset
|
607 zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx); |
| 789 | 608 dsphys->ds_creation_time = gethrestime_sec(); |
| 609 dsphys->ds_creation_txg = tx->tx_txg; | |
| 610 dsphys->ds_deadlist_obj = | |
| 611 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 612 if (clone_parent) { | |
| 613 dsphys->ds_prev_snap_obj = clone_parent->ds_object; | |
| 614 dsphys->ds_prev_snap_txg = | |
| 615 clone_parent->ds_phys->ds_creation_txg; | |
| 616 dsphys->ds_used_bytes = | |
| 617 clone_parent->ds_phys->ds_used_bytes; | |
| 618 dsphys->ds_compressed_bytes = | |
| 619 clone_parent->ds_phys->ds_compressed_bytes; | |
| 620 dsphys->ds_uncompressed_bytes = | |
| 621 clone_parent->ds_phys->ds_uncompressed_bytes; | |
| 622 dsphys->ds_bp = clone_parent->ds_phys->ds_bp; | |
| 623 | |
| 624 dmu_buf_will_dirty(clone_parent->ds_dbuf, tx); | |
| 625 clone_parent->ds_phys->ds_num_children++; | |
| 626 | |
| 627 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 628 dd->dd_phys->dd_clone_parent_obj = clone_parent->ds_object; | |
| 629 } | |
| 1544 | 630 dmu_buf_rele(dbuf, FTAG); |
| 789 | 631 |
| 632 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 633 dd->dd_phys->dd_head_dataset_obj = dsobj; | |
| 634 dsl_dir_close(dd, FTAG); | |
| 635 | |
| 2199 | 636 return (dsobj); |
| 637 } | |
| 638 | |
| 639 struct destroyarg { | |
| 640 dsl_sync_task_group_t *dstg; | |
| 641 char *snapname; | |
| 642 void *tag; | |
| 643 char *failed; | |
| 644 }; | |
| 645 | |
| 646 static int | |
| 647 dsl_snapshot_destroy_one(char *name, void *arg) | |
| 648 { | |
| 649 struct destroyarg *da = arg; | |
| 650 dsl_dataset_t *ds; | |
| 651 char *cp; | |
| 652 int err; | |
| 653 | |
| 654 (void) strcat(name, "@"); | |
| 655 (void) strcat(name, da->snapname); | |
| 656 err = dsl_dataset_open(name, | |
| 657 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT, | |
| 658 da->tag, &ds); | |
| 659 cp = strchr(name, '@'); | |
| 660 *cp = '\0'; | |
| 661 if (err == ENOENT) | |
| 662 return (0); | |
| 663 if (err) { | |
| 664 (void) strcpy(da->failed, name); | |
| 665 return (err); | |
| 666 } | |
| 667 | |
| 668 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check, | |
| 669 dsl_dataset_destroy_sync, ds, da->tag, 0); | |
| 789 | 670 return (0); |
| 671 } | |
| 672 | |
| 2199 | 673 /* |
| 674 * Destroy 'snapname' in all descendants of 'fsname'. | |
| 675 */ | |
| 676 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy | |
| 677 int | |
| 678 dsl_snapshots_destroy(char *fsname, char *snapname) | |
| 679 { | |
| 680 int err; | |
| 681 struct destroyarg da; | |
| 682 dsl_sync_task_t *dst; | |
| 683 spa_t *spa; | |
| 684 char *cp; | |
| 685 | |
| 686 cp = strchr(fsname, '/'); | |
| 687 if (cp) { | |
| 688 *cp = '\0'; | |
| 689 err = spa_open(fsname, &spa, FTAG); | |
| 690 *cp = '/'; | |
| 691 } else { | |
| 692 err = spa_open(fsname, &spa, FTAG); | |
| 693 } | |
| 694 if (err) | |
| 695 return (err); | |
| 696 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); | |
| 697 da.snapname = snapname; | |
| 698 da.tag = FTAG; | |
| 699 da.failed = fsname; | |
| 700 | |
| 701 err = dmu_objset_find(fsname, | |
| 2417 | 702 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN); |
| 2199 | 703 |
| 704 if (err == 0) | |
| 705 err = dsl_sync_task_group_wait(da.dstg); | |
| 706 | |
| 707 for (dst = list_head(&da.dstg->dstg_tasks); dst; | |
| 708 dst = list_next(&da.dstg->dstg_tasks, dst)) { | |
| 709 dsl_dataset_t *ds = dst->dst_arg1; | |
| 710 if (dst->dst_err) { | |
| 711 dsl_dataset_name(ds, fsname); | |
| 712 cp = strchr(fsname, '@'); | |
| 713 *cp = '\0'; | |
| 714 } | |
| 715 /* | |
| 716 * If it was successful, destroy_sync would have | |
| 717 * closed the ds | |
| 718 */ | |
| 719 if (err) | |
| 720 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 721 } | |
| 722 | |
| 723 dsl_sync_task_group_destroy(da.dstg); | |
| 724 spa_close(spa, FTAG); | |
| 725 return (err); | |
| 726 } | |
| 727 | |
| 789 | 728 int |
| 729 dsl_dataset_destroy(const char *name) | |
| 730 { | |
| 731 int err; | |
| 2199 | 732 dsl_sync_task_group_t *dstg; |
| 733 objset_t *os; | |
| 734 dsl_dataset_t *ds; | |
| 789 | 735 dsl_dir_t *dd; |
| 2199 | 736 uint64_t obj; |
| 737 | |
| 738 if (strchr(name, '@')) { | |
| 739 /* Destroying a snapshot is simpler */ | |
| 740 err = dsl_dataset_open(name, | |
| 741 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT, | |
| 742 FTAG, &ds); | |
| 743 if (err) | |
| 744 return (err); | |
| 745 err = dsl_sync_task_do(ds->ds_dir->dd_pool, | |
| 746 dsl_dataset_destroy_check, dsl_dataset_destroy_sync, | |
| 747 ds, FTAG, 0); | |
| 748 if (err) | |
| 749 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 750 return (err); | |
| 751 } | |
| 752 | |
| 753 err = dmu_objset_open(name, DMU_OST_ANY, | |
| 754 DS_MODE_EXCLUSIVE | DS_MODE_INCONSISTENT, &os); | |
| 755 if (err) | |
| 756 return (err); | |
| 757 ds = os->os->os_dsl_dataset; | |
| 758 dd = ds->ds_dir; | |
| 789 | 759 |
| 2199 | 760 /* |
| 761 * Check for errors and mark this ds as inconsistent, in | |
| 762 * case we crash while freeing the objects. | |
| 763 */ | |
| 764 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check, | |
| 765 dsl_dataset_destroy_begin_sync, ds, NULL, 0); | |
| 766 if (err) { | |
| 767 dmu_objset_close(os); | |
| 768 return (err); | |
| 769 } | |
| 770 | |
| 771 /* | |
| 772 * remove the objects in open context, so that we won't | |
| 773 * have too much to do in syncing context. | |
| 774 */ | |
|
3025
4e5ee8301d84
6424466 "panic: data after EOF" when unmounting abused pool
ahrens
parents:
2885
diff
changeset
|
775 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, |
|
4e5ee8301d84
6424466 "panic: data after EOF" when unmounting abused pool
ahrens
parents:
2885
diff
changeset
|
776 ds->ds_phys->ds_prev_snap_txg)) { |
| 2199 | 777 dmu_tx_t *tx = dmu_tx_create(os); |
| 778 dmu_tx_hold_free(tx, obj, 0, DMU_OBJECT_END); | |
| 779 dmu_tx_hold_bonus(tx, obj); | |
| 780 err = dmu_tx_assign(tx, TXG_WAIT); | |
| 781 if (err) { | |
| 782 /* | |
| 783 * Perhaps there is not enough disk | |
| 784 * space. Just deal with it from | |
| 785 * dsl_dataset_destroy_sync(). | |
| 786 */ | |
| 787 dmu_tx_abort(tx); | |
| 788 continue; | |
| 789 } | |
| 790 VERIFY(0 == dmu_object_free(os, obj, tx)); | |
| 791 dmu_tx_commit(tx); | |
| 792 } | |
| 793 /* Make sure it's not dirty before we finish destroying it. */ | |
| 794 txg_wait_synced(dd->dd_pool, 0); | |
| 795 | |
| 796 dmu_objset_close(os); | |
| 797 if (err != ESRCH) | |
| 798 return (err); | |
| 799 | |
| 800 err = dsl_dataset_open(name, | |
| 801 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT, | |
| 802 FTAG, &ds); | |
| 1544 | 803 if (err) |
| 804 return (err); | |
| 789 | 805 |
| 2199 | 806 err = dsl_dir_open(name, FTAG, &dd, NULL); |
| 807 if (err) { | |
| 808 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 809 return (err); | |
| 789 | 810 } |
| 811 | |
| 2199 | 812 /* |
| 813 * Blow away the dsl_dir + head dataset. | |
| 814 */ | |
| 815 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool); | |
| 816 dsl_sync_task_create(dstg, dsl_dataset_destroy_check, | |
| 817 dsl_dataset_destroy_sync, ds, FTAG, 0); | |
| 818 dsl_sync_task_create(dstg, dsl_dir_destroy_check, | |
| 819 dsl_dir_destroy_sync, dd, FTAG, 0); | |
| 820 err = dsl_sync_task_group_wait(dstg); | |
| 821 dsl_sync_task_group_destroy(dstg); | |
| 822 /* if it is successful, *destroy_sync will close the ds+dd */ | |
| 823 if (err) { | |
| 824 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 825 dsl_dir_close(dd, FTAG); | |
| 826 } | |
| 789 | 827 return (err); |
| 828 } | |
| 829 | |
| 830 int | |
| 2199 | 831 dsl_dataset_rollback(dsl_dataset_t *ds) |
| 789 | 832 { |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
833 ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX); |
| 2199 | 834 return (dsl_sync_task_do(ds->ds_dir->dd_pool, |
| 835 dsl_dataset_rollback_check, dsl_dataset_rollback_sync, | |
| 836 ds, NULL, 0)); | |
| 789 | 837 } |
| 838 | |
| 839 void * | |
| 840 dsl_dataset_set_user_ptr(dsl_dataset_t *ds, | |
| 841 void *p, dsl_dataset_evict_func_t func) | |
| 842 { | |
| 843 void *old; | |
| 844 | |
| 845 mutex_enter(&ds->ds_lock); | |
| 846 old = ds->ds_user_ptr; | |
| 847 if (old == NULL) { | |
| 848 ds->ds_user_ptr = p; | |
| 849 ds->ds_user_evict_func = func; | |
| 850 } | |
| 851 mutex_exit(&ds->ds_lock); | |
| 852 return (old); | |
| 853 } | |
| 854 | |
| 855 void * | |
| 856 dsl_dataset_get_user_ptr(dsl_dataset_t *ds) | |
| 857 { | |
| 858 return (ds->ds_user_ptr); | |
| 859 } | |
| 860 | |
| 861 | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
862 blkptr_t * |
|
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
863 dsl_dataset_get_blkptr(dsl_dataset_t *ds) |
| 789 | 864 { |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
865 return (&ds->ds_phys->ds_bp); |
| 789 | 866 } |
| 867 | |
| 868 void | |
| 869 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) | |
| 870 { | |
| 871 ASSERT(dmu_tx_is_syncing(tx)); | |
| 872 /* If it's the meta-objset, set dp_meta_rootbp */ | |
| 873 if (ds == NULL) { | |
| 874 tx->tx_pool->dp_meta_rootbp = *bp; | |
| 875 } else { | |
| 876 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 877 ds->ds_phys->ds_bp = *bp; | |
| 878 } | |
| 879 } | |
| 880 | |
| 881 spa_t * | |
| 882 dsl_dataset_get_spa(dsl_dataset_t *ds) | |
| 883 { | |
| 884 return (ds->ds_dir->dd_pool->dp_spa); | |
| 885 } | |
| 886 | |
| 887 void | |
| 888 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) | |
| 889 { | |
| 890 dsl_pool_t *dp; | |
| 891 | |
| 892 if (ds == NULL) /* this is the meta-objset */ | |
| 893 return; | |
| 894 | |
| 895 ASSERT(ds->ds_user_ptr != NULL); | |
| 2885 | 896 |
| 897 if (ds->ds_phys->ds_next_snap_obj != 0) | |
| 898 panic("dirtying snapshot!"); | |
| 789 | 899 |
| 900 dp = ds->ds_dir->dd_pool; | |
| 901 | |
| 902 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) { | |
| 903 /* up the hold count until we can be written out */ | |
| 904 dmu_buf_add_ref(ds->ds_dbuf, ds); | |
| 905 } | |
| 906 } | |
| 907 | |
| 908 struct killarg { | |
| 909 uint64_t *usedp; | |
| 910 uint64_t *compressedp; | |
| 911 uint64_t *uncompressedp; | |
| 912 zio_t *zio; | |
| 913 dmu_tx_t *tx; | |
| 914 }; | |
| 915 | |
| 916 static int | |
| 917 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg) | |
| 918 { | |
| 919 struct killarg *ka = arg; | |
| 920 blkptr_t *bp = &bc->bc_blkptr; | |
| 921 | |
| 922 ASSERT3U(bc->bc_errno, ==, 0); | |
| 923 | |
| 924 /* | |
| 925 * Since this callback is not called concurrently, no lock is | |
| 926 * needed on the accounting values. | |
| 927 */ | |
| 2082 | 928 *ka->usedp += bp_get_dasize(spa, bp); |
| 789 | 929 *ka->compressedp += BP_GET_PSIZE(bp); |
| 930 *ka->uncompressedp += BP_GET_UCSIZE(bp); | |
| 931 /* XXX check for EIO? */ | |
| 932 (void) arc_free(ka->zio, spa, ka->tx->tx_txg, bp, NULL, NULL, | |
| 933 ARC_NOWAIT); | |
| 934 return (0); | |
| 935 } | |
| 936 | |
| 937 /* ARGSUSED */ | |
| 2199 | 938 static int |
| 939 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 789 | 940 { |
| 2199 | 941 dsl_dataset_t *ds = arg1; |
| 789 | 942 |
| 2199 | 943 /* |
| 944 * There must be a previous snapshot. I suppose we could roll | |
| 945 * it back to being empty (and re-initialize the upper (ZPL) | |
| 946 * layer). But for now there's no way to do this via the user | |
| 947 * interface. | |
| 948 */ | |
| 949 if (ds->ds_phys->ds_prev_snap_txg == 0) | |
| 789 | 950 return (EINVAL); |
| 951 | |
| 2199 | 952 /* |
| 953 * This must not be a snapshot. | |
| 954 */ | |
| 955 if (ds->ds_phys->ds_next_snap_obj != 0) | |
| 956 return (EINVAL); | |
| 789 | 957 |
| 958 /* | |
| 959 * If we made changes this txg, traverse_dsl_dataset won't find | |
| 960 * them. Try again. | |
| 961 */ | |
| 2199 | 962 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) |
| 789 | 963 return (EAGAIN); |
| 2199 | 964 |
| 965 return (0); | |
| 966 } | |
| 789 | 967 |
| 2199 | 968 /* ARGSUSED */ |
| 969 static void | |
| 970 dsl_dataset_rollback_sync(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 971 { | |
| 972 dsl_dataset_t *ds = arg1; | |
| 973 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; | |
| 789 | 974 |
| 975 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 976 | |
| 977 /* Zero out the deadlist. */ | |
| 978 bplist_close(&ds->ds_deadlist); | |
| 979 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); | |
| 980 ds->ds_phys->ds_deadlist_obj = | |
| 981 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 1544 | 982 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos, |
| 983 ds->ds_phys->ds_deadlist_obj)); | |
| 789 | 984 |
| 985 { | |
| 986 /* Free blkptrs that we gave birth to */ | |
| 987 zio_t *zio; | |
| 988 uint64_t used = 0, compressed = 0, uncompressed = 0; | |
| 989 struct killarg ka; | |
| 990 | |
| 991 zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL, | |
| 992 ZIO_FLAG_MUSTSUCCEED); | |
| 993 ka.usedp = &used; | |
| 994 ka.compressedp = &compressed; | |
| 995 ka.uncompressedp = &uncompressed; | |
| 996 ka.zio = zio; | |
| 997 ka.tx = tx; | |
| 998 (void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg, | |
| 999 ADVANCE_POST, kill_blkptr, &ka); | |
| 1000 (void) zio_wait(zio); | |
| 1001 | |
| 2199 | 1002 dsl_dir_diduse_space(ds->ds_dir, |
| 789 | 1003 -used, -compressed, -uncompressed, tx); |
| 1004 } | |
| 1005 | |
| 2199 | 1006 /* Change our contents to that of the prev snapshot */ |
| 789 | 1007 ASSERT3U(ds->ds_prev->ds_object, ==, ds->ds_phys->ds_prev_snap_obj); |
| 1008 ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp; | |
| 1009 ds->ds_phys->ds_used_bytes = ds->ds_prev->ds_phys->ds_used_bytes; | |
| 1010 ds->ds_phys->ds_compressed_bytes = | |
| 1011 ds->ds_prev->ds_phys->ds_compressed_bytes; | |
| 1012 ds->ds_phys->ds_uncompressed_bytes = | |
| 1013 ds->ds_prev->ds_phys->ds_uncompressed_bytes; | |
| 2082 | 1014 ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags; |
| 789 | 1015 ds->ds_phys->ds_unique_bytes = 0; |
| 1016 | |
|
2532
752725c22841
6448999 panic: used == ds->ds_phys->ds_unique_bytes
ahrens
parents:
2417
diff
changeset
|
1017 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { |
|
752725c22841
6448999 panic: used == ds->ds_phys->ds_unique_bytes
ahrens
parents:
2417
diff
changeset
|
1018 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); |
|
752725c22841
6448999 panic: used == ds->ds_phys->ds_unique_bytes
ahrens
parents:
2417
diff
changeset
|
1019 ds->ds_prev->ds_phys->ds_unique_bytes = 0; |
|
752725c22841
6448999 panic: used == ds->ds_phys->ds_unique_bytes
ahrens
parents:
2417
diff
changeset
|
1020 } |
| 789 | 1021 } |
| 1022 | |
|
1731
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
1023 /* ARGSUSED */ |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
1024 static int |
| 2199 | 1025 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
|
1026 { |
| 2199 | 1027 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
|
1028 |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
1029 /* |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
1030 * 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
|
1031 * (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
|
1032 * from.) |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
1033 */ |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
1034 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
|
1035 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
|
1036 return (EINVAL); |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
1037 |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
1038 return (0); |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
1039 } |
|
1efa8b3d1296
6402598 'zfs destroy <fs>' can take a long time, stopping up the txg train
bonwick
parents:
1544
diff
changeset
|
1040 |
| 2199 | 1041 /* ARGSUSED */ |
| 1042 static void | |
| 1043 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 789 | 1044 { |
| 2199 | 1045 dsl_dataset_t *ds = arg1; |
| 789 | 1046 |
| 2199 | 1047 /* Mark it as inconsistent on-disk, in case we crash */ |
| 1048 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 1049 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; | |
| 1050 } | |
| 789 | 1051 |
| 2199 | 1052 /* ARGSUSED */ |
| 1053 static int | |
| 1054 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 1055 { | |
| 1056 dsl_dataset_t *ds = arg1; | |
| 789 | 1057 |
| 1058 /* Can't delete a branch point. */ | |
| 2199 | 1059 if (ds->ds_phys->ds_num_children > 1) |
| 1060 return (EEXIST); | |
| 789 | 1061 |
| 1062 /* | |
| 1063 * Can't delete a head dataset if there are snapshots of it. | |
| 1064 * (Except if the only snapshots are from the branch we cloned | |
| 1065 * from.) | |
| 1066 */ | |
| 1067 if (ds->ds_prev != NULL && | |
| 2199 | 1068 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) |
| 789 | 1069 return (EINVAL); |
| 1070 | |
| 1071 /* | |
| 1072 * If we made changes this txg, traverse_dsl_dataset won't find | |
| 1073 * them. Try again. | |
| 1074 */ | |
| 2199 | 1075 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) |
| 789 | 1076 return (EAGAIN); |
| 2199 | 1077 |
| 1078 /* XXX we should do some i/o error checking... */ | |
| 1079 return (0); | |
| 1080 } | |
| 1081 | |
| 1082 static void | |
| 1083 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) | |
| 1084 { | |
| 1085 dsl_dataset_t *ds = arg1; | |
| 1086 uint64_t used = 0, compressed = 0, uncompressed = 0; | |
| 1087 zio_t *zio; | |
| 1088 int err; | |
| 1089 int after_branch_point = FALSE; | |
| 1090 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 1091 objset_t *mos = dp->dp_meta_objset; | |
| 1092 dsl_dataset_t *ds_prev = NULL; | |
| 1093 uint64_t obj; | |
| 1094 | |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
1095 ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX); |
| 2199 | 1096 ASSERT3U(ds->ds_phys->ds_num_children, <=, 1); |
| 1097 ASSERT(ds->ds_prev == NULL || | |
| 1098 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object); | |
| 1099 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); | |
| 1100 | |
| 1101 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); | |
| 1102 | |
| 1103 obj = ds->ds_object; | |
| 789 | 1104 |
| 1105 if (ds->ds_phys->ds_prev_snap_obj != 0) { | |
| 1106 if (ds->ds_prev) { | |
| 1107 ds_prev = ds->ds_prev; | |
| 1108 } else { | |
| 2199 | 1109 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 789 | 1110 ds->ds_phys->ds_prev_snap_obj, NULL, |
| 2199 | 1111 DS_MODE_NONE, FTAG, &ds_prev)); |
| 789 | 1112 } |
| 1113 after_branch_point = | |
| 1114 (ds_prev->ds_phys->ds_next_snap_obj != obj); | |
| 1115 | |
| 1116 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); | |
| 1117 if (after_branch_point && | |
| 1118 ds->ds_phys->ds_next_snap_obj == 0) { | |
| 1119 /* This clone is toast. */ | |
| 1120 ASSERT(ds_prev->ds_phys->ds_num_children > 1); | |
| 1121 ds_prev->ds_phys->ds_num_children--; | |
| 1122 } else if (!after_branch_point) { | |
| 1123 ds_prev->ds_phys->ds_next_snap_obj = | |
| 1124 ds->ds_phys->ds_next_snap_obj; | |
| 1125 } | |
| 1126 } | |
| 1127 | |
| 1128 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
| 1129 | |
| 1130 if (ds->ds_phys->ds_next_snap_obj != 0) { | |
| 2199 | 1131 blkptr_t bp; |
| 789 | 1132 dsl_dataset_t *ds_next; |
| 1133 uint64_t itor = 0; | |
| 1134 | |
| 1135 spa_scrub_restart(dp->dp_spa, tx->tx_txg); | |
| 1136 | |
| 2199 | 1137 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 1544 | 1138 ds->ds_phys->ds_next_snap_obj, NULL, |
| 1139 DS_MODE_NONE, FTAG, &ds_next)); | |
| 789 | 1140 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); |
| 1141 | |
| 1142 dmu_buf_will_dirty(ds_next->ds_dbuf, tx); | |
| 1143 ds_next->ds_phys->ds_prev_snap_obj = | |
| 1144 ds->ds_phys->ds_prev_snap_obj; | |
| 1145 ds_next->ds_phys->ds_prev_snap_txg = | |
| 1146 ds->ds_phys->ds_prev_snap_txg; | |
| 1147 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, | |
| 1148 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); | |
| 1149 | |
| 1150 /* | |
| 1151 * Transfer to our deadlist (which will become next's | |
| 1152 * new deadlist) any entries from next's current | |
| 1153 * deadlist which were born before prev, and free the | |
| 1154 * other entries. | |
| 1155 * | |
| 1156 * XXX we're doing this long task with the config lock held | |
| 1157 */ | |
| 1158 while (bplist_iterate(&ds_next->ds_deadlist, &itor, | |
| 1159 &bp) == 0) { | |
| 1160 if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) { | |
| 1544 | 1161 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, |
| 1162 &bp, tx)); | |
| 789 | 1163 if (ds_prev && !after_branch_point && |
| 1164 bp.blk_birth > | |
| 1165 ds_prev->ds_phys->ds_prev_snap_txg) { | |
| 1166 ds_prev->ds_phys->ds_unique_bytes += | |
| 2082 | 1167 bp_get_dasize(dp->dp_spa, &bp); |
| 789 | 1168 } |
| 1169 } else { | |
| 2082 | 1170 used += bp_get_dasize(dp->dp_spa, &bp); |
| 789 | 1171 compressed += BP_GET_PSIZE(&bp); |
| 1172 uncompressed += BP_GET_UCSIZE(&bp); | |
| 1173 /* XXX check return value? */ | |
| 1174 (void) arc_free(zio, dp->dp_spa, tx->tx_txg, | |
| 1175 &bp, NULL, NULL, ARC_NOWAIT); | |
| 1176 } | |
| 1177 } | |
| 1178 | |
| 1179 /* free next's deadlist */ | |
| 1180 bplist_close(&ds_next->ds_deadlist); | |
| 1181 bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx); | |
| 1182 | |
| 1183 /* set next's deadlist to our deadlist */ | |
| 1184 ds_next->ds_phys->ds_deadlist_obj = | |
| 1185 ds->ds_phys->ds_deadlist_obj; | |
| 1544 | 1186 VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos, |
| 1187 ds_next->ds_phys->ds_deadlist_obj)); | |
| 789 | 1188 ds->ds_phys->ds_deadlist_obj = 0; |
| 1189 | |
| 1190 if (ds_next->ds_phys->ds_next_snap_obj != 0) { | |
| 1191 /* | |
| 1192 * Update next's unique to include blocks which | |
| 1193 * were previously shared by only this snapshot | |
| 1194 * and it. Those blocks will be born after the | |
| 1195 * prev snap and before this snap, and will have | |
| 1196 * died after the next snap and before the one | |
| 1197 * after that (ie. be on the snap after next's | |
| 1198 * deadlist). | |
| 1199 * | |
| 1200 * XXX we're doing this long task with the | |
| 1201 * config lock held | |
| 1202 */ | |
| 1203 dsl_dataset_t *ds_after_next; | |
| 1204 | |
| 2199 | 1205 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 789 | 1206 ds_next->ds_phys->ds_next_snap_obj, NULL, |
| 1544 | 1207 DS_MODE_NONE, FTAG, &ds_after_next)); |
| 789 | 1208 itor = 0; |
| 1209 while (bplist_iterate(&ds_after_next->ds_deadlist, | |
| 1210 &itor, &bp) == 0) { | |
| 1211 if (bp.blk_birth > | |
| 1212 ds->ds_phys->ds_prev_snap_txg && | |
| 1213 bp.blk_birth <= | |
| 1214 ds->ds_phys->ds_creation_txg) { | |
| 1215 ds_next->ds_phys->ds_unique_bytes += | |
| 2082 | 1216 bp_get_dasize(dp->dp_spa, &bp); |
| 789 | 1217 } |
| 1218 } | |
| 1219 | |
| 1220 dsl_dataset_close(ds_after_next, DS_MODE_NONE, FTAG); | |
| 1221 ASSERT3P(ds_next->ds_prev, ==, NULL); | |
| 1222 } else { | |
| 1223 /* | |
| 1224 * It would be nice to update the head dataset's | |
| 1225 * unique. To do so we would have to traverse | |
| 1226 * it for blocks born after ds_prev, which is | |
| 1227 * pretty expensive just to maintain something | |
| 1228 * for debugging purposes. | |
| 1229 */ | |
| 1230 ASSERT3P(ds_next->ds_prev, ==, ds); | |
| 1231 dsl_dataset_close(ds_next->ds_prev, DS_MODE_NONE, | |
| 1232 ds_next); | |
| 1233 if (ds_prev) { | |
| 2199 | 1234 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 1544 | 1235 ds->ds_phys->ds_prev_snap_obj, NULL, |
| 1236 DS_MODE_NONE, ds_next, &ds_next->ds_prev)); | |
| 789 | 1237 } else { |
| 1238 ds_next->ds_prev = NULL; | |
| 1239 } | |
| 1240 } | |
| 1241 dsl_dataset_close(ds_next, DS_MODE_NONE, FTAG); | |
| 1242 | |
| 1243 /* | |
| 1244 * NB: unique_bytes is not accurate for head objsets | |
| 1245 * because we don't update it when we delete the most | |
| 1246 * recent snapshot -- see above comment. | |
| 1247 */ | |
| 1248 ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes); | |
| 1249 } else { | |
| 1250 /* | |
| 1251 * There's no next snapshot, so this is a head dataset. | |
| 1252 * Destroy the deadlist. Unless it's a clone, the | |
| 1253 * deadlist should be empty. (If it's a clone, it's | |
| 1254 * safe to ignore the deadlist contents.) | |
| 1255 */ | |
| 1256 struct killarg ka; | |
| 1257 | |
| 1258 ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist)); | |
| 1259 bplist_close(&ds->ds_deadlist); | |
| 1260 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); | |
| 1261 ds->ds_phys->ds_deadlist_obj = 0; | |
| 1262 | |
| 1263 /* | |
| 1264 * Free everything that we point to (that's born after | |
| 1265 * the previous snapshot, if we are a clone) | |
| 1266 * | |
| 1267 * XXX we're doing this long task with the config lock held | |
| 1268 */ | |
| 1269 ka.usedp = &used; | |
| 1270 ka.compressedp = &compressed; | |
| 1271 ka.uncompressedp = &uncompressed; | |
| 1272 ka.zio = zio; | |
| 1273 ka.tx = tx; | |
| 1274 err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg, | |
| 1275 ADVANCE_POST, kill_blkptr, &ka); | |
| 1276 ASSERT3U(err, ==, 0); | |
| 1277 } | |
| 1278 | |
| 1279 err = zio_wait(zio); | |
| 1280 ASSERT3U(err, ==, 0); | |
| 1281 | |
| 2199 | 1282 dsl_dir_diduse_space(ds->ds_dir, -used, -compressed, -uncompressed, tx); |
| 789 | 1283 |
| 1284 if (ds->ds_phys->ds_snapnames_zapobj) { | |
| 1285 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx); | |
| 1286 ASSERT(err == 0); | |
| 1287 } | |
| 1288 | |
| 2199 | 1289 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) { |
| 789 | 1290 /* Erase the link in the dataset */ |
| 2199 | 1291 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); |
| 1292 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0; | |
| 789 | 1293 /* |
| 1294 * dsl_dir_sync_destroy() called us, they'll destroy | |
| 1295 * the dataset. | |
| 1296 */ | |
| 1297 } else { | |
| 1298 /* remove from snapshot namespace */ | |
| 1299 dsl_dataset_t *ds_head; | |
| 2199 | 1300 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 1301 ds->ds_dir->dd_phys->dd_head_dataset_obj, NULL, | |
| 1544 | 1302 DS_MODE_NONE, FTAG, &ds_head)); |
|
2207
47efcb3433a7
6439370 assertion failures possible in dsl_dataset_destroy_sync()
ahrens
parents:
2199
diff
changeset
|
1303 VERIFY(0 == dsl_dataset_get_snapname(ds)); |
| 789 | 1304 #ifdef ZFS_DEBUG |
| 1305 { | |
| 1306 uint64_t val; | |
| 1307 err = zap_lookup(mos, | |
| 1308 ds_head->ds_phys->ds_snapnames_zapobj, | |
| 2199 | 1309 ds->ds_snapname, 8, 1, &val); |
| 789 | 1310 ASSERT3U(err, ==, 0); |
| 1311 ASSERT3U(val, ==, obj); | |
| 1312 } | |
| 1313 #endif | |
| 1314 err = zap_remove(mos, ds_head->ds_phys->ds_snapnames_zapobj, | |
| 2199 | 1315 ds->ds_snapname, tx); |
| 789 | 1316 ASSERT(err == 0); |
| 1317 dsl_dataset_close(ds_head, DS_MODE_NONE, FTAG); | |
| 1318 } | |
| 1319 | |
| 1320 if (ds_prev && ds->ds_prev != ds_prev) | |
| 1321 dsl_dataset_close(ds_prev, DS_MODE_NONE, FTAG); | |
| 1322 | |
| 3912 | 1323 spa_clear_bootfs(dp->dp_spa, ds->ds_object, tx); |
| 2199 | 1324 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, tag); |
| 1325 VERIFY(0 == dmu_object_free(mos, obj, tx)); | |
| 3912 | 1326 |
| 2199 | 1327 } |
| 1328 | |
| 1329 /* ARGSUSED */ | |
| 1330 int | |
| 1331 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 1332 { | |
| 1333 objset_t *os = arg1; | |
| 1334 dsl_dataset_t *ds = os->os->os_dsl_dataset; | |
| 1335 const char *snapname = arg2; | |
| 1336 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; | |
| 1337 int err; | |
| 1338 uint64_t value; | |
| 789 | 1339 |
| 1340 /* | |
| 2199 | 1341 * We don't allow multiple snapshots of the same txg. If there |
| 1342 * is already one, try again. | |
| 1343 */ | |
| 1344 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) | |
| 1345 return (EAGAIN); | |
| 1346 | |
| 1347 /* | |
| 1348 * Check for conflicting name snapshot name. | |
| 789 | 1349 */ |
| 2199 | 1350 err = zap_lookup(mos, ds->ds_phys->ds_snapnames_zapobj, |
| 1351 snapname, 8, 1, &value); | |
| 1352 if (err == 0) | |
| 1353 return (EEXIST); | |
| 1354 if (err != ENOENT) | |
| 1355 return (err); | |
| 789 | 1356 |
|
3978
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
1357 /* |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
1358 * Check that the dataset's name is not too long. Name consists |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
1359 * of the dataset's length + 1 for the @-sign + snapshot name's length |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
1360 */ |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
1361 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN) |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
1362 return (ENAMETOOLONG); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
1363 |
| 2199 | 1364 ds->ds_trysnap_txg = tx->tx_txg; |
| 789 | 1365 return (0); |
| 1366 } | |
| 1367 | |
| 2199 | 1368 void |
| 1369 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 789 | 1370 { |
| 2199 | 1371 objset_t *os = arg1; |
| 1372 dsl_dataset_t *ds = os->os->os_dsl_dataset; | |
| 1373 const char *snapname = arg2; | |
| 1374 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 789 | 1375 dmu_buf_t *dbuf; |
| 1376 dsl_dataset_phys_t *dsphys; | |
| 2199 | 1377 uint64_t dsobj; |
| 789 | 1378 objset_t *mos = dp->dp_meta_objset; |
| 1379 int err; | |
| 1380 | |
| 1381 spa_scrub_restart(dp->dp_spa, tx->tx_txg); | |
| 2199 | 1382 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); |
| 789 | 1383 |
|
928
36d72fe4da29
6349314 dmu_object_type names incorrect for DSL Directories and DSL Datasets
tabriz
parents:
885
diff
changeset
|
1384 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
|
1385 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); |
| 1544 | 1386 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); |
| 789 | 1387 dmu_buf_will_dirty(dbuf, tx); |
| 1388 dsphys = dbuf->db_data; | |
| 2199 | 1389 dsphys->ds_dir_obj = ds->ds_dir->dd_object; |
| 789 | 1390 dsphys->ds_fsid_guid = unique_create(); |
| 1391 unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */ | |
| 1392 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, | |
| 1393 sizeof (dsphys->ds_guid)); | |
| 1394 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; | |
| 1395 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; | |
| 1396 dsphys->ds_next_snap_obj = ds->ds_object; | |
| 1397 dsphys->ds_num_children = 1; | |
| 1398 dsphys->ds_creation_time = gethrestime_sec(); | |
| 1399 dsphys->ds_creation_txg = tx->tx_txg; | |
| 1400 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; | |
| 1401 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes; | |
| 1402 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; | |
| 1403 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; | |
| 2082 | 1404 dsphys->ds_flags = ds->ds_phys->ds_flags; |
| 789 | 1405 dsphys->ds_bp = ds->ds_phys->ds_bp; |
| 1544 | 1406 dmu_buf_rele(dbuf, FTAG); |
| 789 | 1407 |
| 2199 | 1408 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); |
| 1409 if (ds->ds_prev) { | |
| 1410 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == | |
| 789 | 1411 ds->ds_object || |
| 2199 | 1412 ds->ds_prev->ds_phys->ds_num_children > 1); |
| 1413 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { | |
| 1414 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
| 789 | 1415 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, |
| 2199 | 1416 ds->ds_prev->ds_phys->ds_creation_txg); |
| 1417 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; | |
| 789 | 1418 } |
| 1419 } | |
| 1420 | |
| 1421 bplist_close(&ds->ds_deadlist); | |
| 1422 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 1423 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, dsphys->ds_creation_txg); | |
| 1424 ds->ds_phys->ds_prev_snap_obj = dsobj; | |
| 1425 ds->ds_phys->ds_prev_snap_txg = dsphys->ds_creation_txg; | |
| 1426 ds->ds_phys->ds_unique_bytes = 0; | |
| 1427 ds->ds_phys->ds_deadlist_obj = | |
| 1428 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 1544 | 1429 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos, |
| 1430 ds->ds_phys->ds_deadlist_obj)); | |
| 789 | 1431 |
| 1432 dprintf("snap '%s' -> obj %llu\n", snapname, dsobj); | |
| 1433 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, | |
| 1434 snapname, 8, 1, &dsobj, tx); | |
| 1435 ASSERT(err == 0); | |
| 1436 | |
| 1437 if (ds->ds_prev) | |
| 1438 dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds); | |
| 1544 | 1439 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 1440 ds->ds_phys->ds_prev_snap_obj, snapname, | |
| 1441 DS_MODE_NONE, ds, &ds->ds_prev)); | |
| 789 | 1442 } |
| 1443 | |
| 1444 void | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1445 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) |
| 789 | 1446 { |
| 1447 ASSERT(dmu_tx_is_syncing(tx)); | |
| 1448 ASSERT(ds->ds_user_ptr != NULL); | |
| 1449 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); | |
| 1450 | |
| 1451 dsl_dir_dirty(ds->ds_dir, tx); | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1452 dmu_objset_sync(ds->ds_user_ptr, zio, tx); |
|
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1453 /* Unneeded? bplist_close(&ds->ds_deadlist); */ |
| 789 | 1454 } |
| 1455 | |
| 1456 void | |
| 2885 | 1457 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) |
| 789 | 1458 { |
| 2885 | 1459 dsl_dir_stats(ds->ds_dir, nv); |
| 789 | 1460 |
| 2885 | 1461 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, |
| 1462 ds->ds_phys->ds_creation_time); | |
| 1463 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, | |
| 1464 ds->ds_phys->ds_creation_txg); | |
| 1465 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, | |
| 1466 ds->ds_phys->ds_used_bytes); | |
| 789 | 1467 |
| 1468 if (ds->ds_phys->ds_next_snap_obj) { | |
| 1469 /* | |
| 1470 * This is a snapshot; override the dd's space used with | |
| 2885 | 1471 * our unique space and compression ratio. |
| 789 | 1472 */ |
| 2885 | 1473 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, |
| 1474 ds->ds_phys->ds_unique_bytes); | |
| 1475 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, | |
| 1476 ds->ds_phys->ds_compressed_bytes == 0 ? 100 : | |
| 1477 (ds->ds_phys->ds_uncompressed_bytes * 100 / | |
| 1478 ds->ds_phys->ds_compressed_bytes)); | |
| 789 | 1479 } |
| 1480 } | |
| 1481 | |
| 2885 | 1482 void |
| 1483 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) | |
| 789 | 1484 { |
| 2885 | 1485 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; |
| 1486 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; | |
| 1487 if (ds->ds_phys->ds_next_snap_obj) { | |
| 1488 stat->dds_is_snapshot = B_TRUE; | |
| 1489 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; | |
| 1490 } | |
| 1491 | |
| 1492 /* clone origin is really a dsl_dir thing... */ | |
| 1493 if (ds->ds_dir->dd_phys->dd_clone_parent_obj) { | |
| 1494 dsl_dataset_t *ods; | |
| 1495 | |
| 1496 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); | |
| 1497 VERIFY(0 == dsl_dataset_open_obj(ds->ds_dir->dd_pool, | |
| 1498 ds->ds_dir->dd_phys->dd_clone_parent_obj, | |
| 1499 NULL, DS_MODE_NONE, FTAG, &ods)); | |
| 1500 dsl_dataset_name(ods, stat->dds_clone_of); | |
| 1501 dsl_dataset_close(ods, DS_MODE_NONE, FTAG); | |
| 1502 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); | |
| 1503 } | |
| 1504 } | |
| 1505 | |
| 1506 uint64_t | |
| 1507 dsl_dataset_fsid_guid(dsl_dataset_t *ds) | |
| 1508 { | |
| 1509 return (ds->ds_phys->ds_fsid_guid); | |
| 1510 } | |
| 1511 | |
| 1512 void | |
| 1513 dsl_dataset_space(dsl_dataset_t *ds, | |
| 1514 uint64_t *refdbytesp, uint64_t *availbytesp, | |
| 1515 uint64_t *usedobjsp, uint64_t *availobjsp) | |
| 1516 { | |
| 1517 *refdbytesp = ds->ds_phys->ds_used_bytes; | |
| 1518 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); | |
| 1519 *usedobjsp = ds->ds_phys->ds_bp.blk_fill; | |
| 1520 *availobjsp = DN_MAX_OBJECT - *usedobjsp; | |
| 789 | 1521 } |
| 1522 | |
| 2199 | 1523 /* ARGSUSED */ |
| 789 | 1524 static int |
| 2199 | 1525 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) |
| 789 | 1526 { |
| 2199 | 1527 dsl_dataset_t *ds = arg1; |
| 1528 char *newsnapname = arg2; | |
| 1529 dsl_dir_t *dd = ds->ds_dir; | |
| 789 | 1530 objset_t *mos = dd->dd_pool->dp_meta_objset; |
| 2199 | 1531 dsl_dataset_t *hds; |
| 1532 uint64_t val; | |
| 789 | 1533 int err; |
| 1534 | |
| 2199 | 1535 err = dsl_dataset_open_obj(dd->dd_pool, |
| 1536 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds); | |
| 789 | 1537 if (err) |
| 1538 return (err); | |
| 1539 | |
| 2199 | 1540 /* new name better not be in use */ |
| 1541 err = zap_lookup(mos, hds->ds_phys->ds_snapnames_zapobj, | |
| 1542 newsnapname, 8, 1, &val); | |
| 1543 dsl_dataset_close(hds, DS_MODE_NONE, FTAG); | |
| 789 | 1544 |
| 2199 | 1545 if (err == 0) |
| 1546 err = EEXIST; | |
| 1547 else if (err == ENOENT) | |
| 1548 err = 0; | |
| 1549 return (err); | |
| 1550 } | |
| 789 | 1551 |
| 2199 | 1552 static void |
| 1553 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 1554 { | |
| 1555 dsl_dataset_t *ds = arg1; | |
| 1556 char *newsnapname = arg2; | |
| 1557 dsl_dir_t *dd = ds->ds_dir; | |
| 1558 objset_t *mos = dd->dd_pool->dp_meta_objset; | |
| 1559 dsl_dataset_t *hds; | |
| 1560 int err; | |
| 789 | 1561 |
| 2199 | 1562 ASSERT(ds->ds_phys->ds_next_snap_obj != 0); |
| 789 | 1563 |
| 2199 | 1564 VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool, |
| 1565 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds)); | |
| 789 | 1566 |
| 2199 | 1567 VERIFY(0 == dsl_dataset_get_snapname(ds)); |
| 1568 err = zap_remove(mos, hds->ds_phys->ds_snapnames_zapobj, | |
| 1569 ds->ds_snapname, tx); | |
| 789 | 1570 ASSERT3U(err, ==, 0); |
| 2199 | 1571 mutex_enter(&ds->ds_lock); |
| 1572 (void) strcpy(ds->ds_snapname, newsnapname); | |
| 1573 mutex_exit(&ds->ds_lock); | |
| 1574 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj, | |
| 1575 ds->ds_snapname, 8, 1, &ds->ds_object, tx); | |
| 789 | 1576 ASSERT3U(err, ==, 0); |
| 1577 | |
| 2199 | 1578 dsl_dataset_close(hds, DS_MODE_NONE, FTAG); |
| 789 | 1579 } |
| 1580 | |
| 1581 #pragma weak dmu_objset_rename = dsl_dataset_rename | |
| 1582 int | |
| 2199 | 1583 dsl_dataset_rename(const char *oldname, const char *newname) |
| 789 | 1584 { |
| 1585 dsl_dir_t *dd; | |
| 2199 | 1586 dsl_dataset_t *ds; |
| 789 | 1587 const char *tail; |
| 1588 int err; | |
| 1589 | |
| 2199 | 1590 err = dsl_dir_open(oldname, FTAG, &dd, &tail); |
| 1544 | 1591 if (err) |
| 1592 return (err); | |
| 789 | 1593 if (tail == NULL) { |
| 2199 | 1594 err = dsl_dir_rename(dd, newname); |
| 789 | 1595 dsl_dir_close(dd, FTAG); |
| 1596 return (err); | |
| 1597 } | |
| 1598 if (tail[0] != '@') { | |
| 1599 /* the name ended in a nonexistant component */ | |
| 1600 dsl_dir_close(dd, FTAG); | |
| 1601 return (ENOENT); | |
| 1602 } | |
| 1603 | |
| 2199 | 1604 dsl_dir_close(dd, FTAG); |
| 1605 | |
| 1606 /* new name must be snapshot in same filesystem */ | |
| 1607 tail = strchr(newname, '@'); | |
| 1608 if (tail == NULL) | |
| 1609 return (EINVAL); | |
| 1610 tail++; | |
| 1611 if (strncmp(oldname, newname, tail - newname) != 0) | |
| 1612 return (EXDEV); | |
| 789 | 1613 |
| 2199 | 1614 err = dsl_dataset_open(oldname, |
| 1615 DS_MODE_READONLY | DS_MODE_STANDARD, FTAG, &ds); | |
| 1616 if (err) | |
| 1617 return (err); | |
| 1618 | |
| 1619 err = dsl_sync_task_do(ds->ds_dir->dd_pool, | |
| 1620 dsl_dataset_snapshot_rename_check, | |
| 1621 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1); | |
| 1622 | |
| 1623 dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG); | |
| 1624 | |
| 789 | 1625 return (err); |
| 1626 } | |
| 2082 | 1627 |
| 2199 | 1628 struct promotearg { |
| 1629 uint64_t used, comp, uncomp, unique; | |
| 1630 uint64_t newnext_obj, snapnames_obj; | |
| 1631 }; | |
| 1632 | |
| 2082 | 1633 static int |
| 2199 | 1634 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) |
| 2082 | 1635 { |
| 2199 | 1636 dsl_dataset_t *hds = arg1; |
| 1637 struct promotearg *pa = arg2; | |
| 1638 dsl_dir_t *dd = hds->ds_dir; | |
| 1639 dsl_pool_t *dp = hds->ds_dir->dd_pool; | |
| 2082 | 1640 dsl_dir_t *pdd = NULL; |
| 1641 dsl_dataset_t *ds = NULL; | |
| 1642 dsl_dataset_t *pivot_ds = NULL; | |
| 1643 dsl_dataset_t *newnext_ds = NULL; | |
| 1644 int err; | |
| 1645 char *name = NULL; | |
| 2199 | 1646 uint64_t itor = 0; |
| 2082 | 1647 blkptr_t bp; |
| 1648 | |
| 2199 | 1649 bzero(pa, sizeof (*pa)); |
| 1650 | |
| 2082 | 1651 /* Check that it is a clone */ |
| 1652 if (dd->dd_phys->dd_clone_parent_obj == 0) | |
| 1653 return (EINVAL); | |
| 1654 | |
| 2199 | 1655 /* Since this is so expensive, don't do the preliminary check */ |
| 1656 if (!dmu_tx_is_syncing(tx)) | |
| 1657 return (0); | |
| 1658 | |
| 1659 if (err = dsl_dataset_open_obj(dp, | |
| 2082 | 1660 dd->dd_phys->dd_clone_parent_obj, |
| 1661 NULL, DS_MODE_EXCLUSIVE, FTAG, &pivot_ds)) | |
| 1662 goto out; | |
| 1663 pdd = pivot_ds->ds_dir; | |
| 2199 | 1664 |
| 1665 { | |
| 1666 dsl_dataset_t *phds; | |
| 1667 if (err = dsl_dataset_open_obj(dd->dd_pool, | |
| 1668 pdd->dd_phys->dd_head_dataset_obj, | |
| 1669 NULL, DS_MODE_NONE, FTAG, &phds)) | |
| 1670 goto out; | |
| 1671 pa->snapnames_obj = phds->ds_phys->ds_snapnames_zapobj; | |
| 1672 dsl_dataset_close(phds, DS_MODE_NONE, FTAG); | |
| 1673 } | |
| 2082 | 1674 |
| 1675 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) { | |
| 1676 err = EXDEV; | |
| 1677 goto out; | |
| 1678 } | |
| 1679 | |
| 1680 /* find pivot point's new next ds */ | |
| 1681 VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool, hds->ds_object, | |
| 1682 NULL, DS_MODE_NONE, FTAG, &newnext_ds)); | |
| 1683 while (newnext_ds->ds_phys->ds_prev_snap_obj != pivot_ds->ds_object) { | |
| 1684 dsl_dataset_t *prev; | |
| 1685 | |
| 1686 if (err = dsl_dataset_open_obj(dd->dd_pool, | |
| 2199 | 1687 newnext_ds->ds_phys->ds_prev_snap_obj, |
| 1688 NULL, DS_MODE_NONE, FTAG, &prev)) | |
| 2082 | 1689 goto out; |
| 1690 dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG); | |
| 1691 newnext_ds = prev; | |
| 1692 } | |
| 2199 | 1693 pa->newnext_obj = newnext_ds->ds_object; |
| 2082 | 1694 |
| 1695 /* compute pivot point's new unique space */ | |
| 1696 while ((err = bplist_iterate(&newnext_ds->ds_deadlist, | |
| 1697 &itor, &bp)) == 0) { | |
| 1698 if (bp.blk_birth > pivot_ds->ds_phys->ds_prev_snap_txg) | |
| 2199 | 1699 pa->unique += bp_get_dasize(dd->dd_pool->dp_spa, &bp); |
| 2082 | 1700 } |
| 1701 if (err != ENOENT) | |
| 1702 goto out; | |
| 1703 | |
| 1704 /* Walk the snapshots that we are moving */ | |
| 1705 name = kmem_alloc(MAXPATHLEN, KM_SLEEP); | |
| 1706 ds = pivot_ds; | |
| 1707 /* CONSTCOND */ | |
| 1708 while (TRUE) { | |
| 1709 uint64_t val, dlused, dlcomp, dluncomp; | |
| 1710 dsl_dataset_t *prev; | |
| 1711 | |
| 1712 /* Check that the snapshot name does not conflict */ | |
| 1713 dsl_dataset_name(ds, name); | |
| 1714 err = zap_lookup(dd->dd_pool->dp_meta_objset, | |
| 1715 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, | |
| 1716 8, 1, &val); | |
| 1717 if (err != ENOENT) { | |
| 1718 if (err == 0) | |
| 1719 err = EEXIST; | |
| 1720 goto out; | |
| 1721 } | |
| 1722 | |
| 1723 /* | |
| 1724 * compute space to transfer. Each snapshot gave birth to: | |
| 1725 * (my used) - (prev's used) + (deadlist's used) | |
| 1726 */ | |
| 2199 | 1727 pa->used += ds->ds_phys->ds_used_bytes; |
| 1728 pa->comp += ds->ds_phys->ds_compressed_bytes; | |
| 1729 pa->uncomp += ds->ds_phys->ds_uncompressed_bytes; | |
| 2082 | 1730 |
| 1731 /* If we reach the first snapshot, we're done. */ | |
| 1732 if (ds->ds_phys->ds_prev_snap_obj == 0) | |
| 1733 break; | |
| 1734 | |
| 1735 if (err = bplist_space(&ds->ds_deadlist, | |
| 1736 &dlused, &dlcomp, &dluncomp)) | |
| 1737 goto out; | |
| 1738 if (err = dsl_dataset_open_obj(dd->dd_pool, | |
| 1739 ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE, | |
| 1740 FTAG, &prev)) | |
| 1741 goto out; | |
| 2199 | 1742 pa->used += dlused - prev->ds_phys->ds_used_bytes; |
| 1743 pa->comp += dlcomp - prev->ds_phys->ds_compressed_bytes; | |
| 1744 pa->uncomp += dluncomp - prev->ds_phys->ds_uncompressed_bytes; | |
| 2082 | 1745 |
| 1746 /* | |
| 1747 * We could be a clone of a clone. If we reach our | |
| 1748 * parent's branch point, we're done. | |
| 1749 */ | |
| 1750 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) { | |
| 1751 dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG); | |
| 1752 break; | |
| 1753 } | |
| 1754 if (ds != pivot_ds) | |
| 1755 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 1756 ds = prev; | |
| 1757 } | |
| 1758 | |
| 1759 /* Check that there is enough space here */ | |
| 2199 | 1760 err = dsl_dir_transfer_possible(pdd, dd, pa->used); |
| 1761 | |
| 1762 out: | |
| 1763 if (ds && ds != pivot_ds) | |
| 1764 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 1765 if (pivot_ds) | |
| 1766 dsl_dataset_close(pivot_ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 1767 if (newnext_ds) | |
| 1768 dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG); | |
| 1769 if (name) | |
| 1770 kmem_free(name, MAXPATHLEN); | |
| 1771 return (err); | |
| 1772 } | |
| 2082 | 1773 |
| 2199 | 1774 static void |
| 1775 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 1776 { | |
| 1777 dsl_dataset_t *hds = arg1; | |
| 1778 struct promotearg *pa = arg2; | |
| 1779 dsl_dir_t *dd = hds->ds_dir; | |
| 1780 dsl_pool_t *dp = hds->ds_dir->dd_pool; | |
| 1781 dsl_dir_t *pdd = NULL; | |
| 1782 dsl_dataset_t *ds, *pivot_ds; | |
| 1783 char *name; | |
| 1784 | |
| 1785 ASSERT(dd->dd_phys->dd_clone_parent_obj != 0); | |
| 1786 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)); | |
| 1787 | |
| 1788 VERIFY(0 == dsl_dataset_open_obj(dp, | |
| 1789 dd->dd_phys->dd_clone_parent_obj, | |
| 1790 NULL, DS_MODE_EXCLUSIVE, FTAG, &pivot_ds)); | |
| 2417 | 1791 /* |
| 1792 * We need to explicitly open pdd, since pivot_ds's pdd will be | |
| 1793 * changing. | |
| 1794 */ | |
| 1795 VERIFY(0 == dsl_dir_open_obj(dp, pivot_ds->ds_dir->dd_object, | |
| 1796 NULL, FTAG, &pdd)); | |
| 2082 | 1797 |
| 1798 /* move snapshots to this dir */ | |
| 2199 | 1799 name = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
| 2082 | 1800 ds = pivot_ds; |
| 1801 /* CONSTCOND */ | |
| 1802 while (TRUE) { | |
| 1803 dsl_dataset_t *prev; | |
| 1804 | |
| 1805 /* move snap name entry */ | |
| 1806 dsl_dataset_name(ds, name); | |
| 2199 | 1807 VERIFY(0 == zap_remove(dp->dp_meta_objset, |
| 1808 pa->snapnames_obj, ds->ds_snapname, tx)); | |
| 1809 VERIFY(0 == zap_add(dp->dp_meta_objset, | |
| 2082 | 1810 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, |
| 1811 8, 1, &ds->ds_object, tx)); | |
| 1812 | |
| 1813 /* change containing dsl_dir */ | |
| 1814 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 1815 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, pdd->dd_object); | |
| 1816 ds->ds_phys->ds_dir_obj = dd->dd_object; | |
| 1817 ASSERT3P(ds->ds_dir, ==, pdd); | |
| 1818 dsl_dir_close(ds->ds_dir, ds); | |
| 2199 | 1819 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object, |
| 2082 | 1820 NULL, ds, &ds->ds_dir)); |
| 1821 | |
| 1822 ASSERT3U(dsl_prop_numcb(ds), ==, 0); | |
| 1823 | |
| 1824 if (ds->ds_phys->ds_prev_snap_obj == 0) | |
| 1825 break; | |
| 1826 | |
| 2199 | 1827 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 2082 | 1828 ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE, |
| 1829 FTAG, &prev)); | |
| 1830 | |
| 1831 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) { | |
| 1832 dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG); | |
| 1833 break; | |
| 1834 } | |
| 1835 if (ds != pivot_ds) | |
| 1836 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 1837 ds = prev; | |
| 1838 } | |
| 2199 | 1839 if (ds != pivot_ds) |
| 1840 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 2082 | 1841 |
| 1842 /* change pivot point's next snap */ | |
| 1843 dmu_buf_will_dirty(pivot_ds->ds_dbuf, tx); | |
| 2199 | 1844 pivot_ds->ds_phys->ds_next_snap_obj = pa->newnext_obj; |
| 2082 | 1845 |
| 1846 /* change clone_parent-age */ | |
| 1847 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 1848 ASSERT3U(dd->dd_phys->dd_clone_parent_obj, ==, pivot_ds->ds_object); | |
| 1849 dd->dd_phys->dd_clone_parent_obj = pdd->dd_phys->dd_clone_parent_obj; | |
| 1850 dmu_buf_will_dirty(pdd->dd_dbuf, tx); | |
| 1851 pdd->dd_phys->dd_clone_parent_obj = pivot_ds->ds_object; | |
| 1852 | |
| 1853 /* change space accounting */ | |
| 2199 | 1854 dsl_dir_diduse_space(pdd, -pa->used, -pa->comp, -pa->uncomp, tx); |
| 1855 dsl_dir_diduse_space(dd, pa->used, pa->comp, pa->uncomp, tx); | |
| 1856 pivot_ds->ds_phys->ds_unique_bytes = pa->unique; | |
| 2082 | 1857 |
| 2417 | 1858 dsl_dir_close(pdd, FTAG); |
| 2199 | 1859 dsl_dataset_close(pivot_ds, DS_MODE_EXCLUSIVE, FTAG); |
| 1860 kmem_free(name, MAXPATHLEN); | |
| 2082 | 1861 } |
| 1862 | |
| 1863 int | |
| 1864 dsl_dataset_promote(const char *name) | |
| 1865 { | |
| 1866 dsl_dataset_t *ds; | |
| 1867 int err; | |
| 1868 dmu_object_info_t doi; | |
| 2199 | 1869 struct promotearg pa; |
| 2082 | 1870 |
| 1871 err = dsl_dataset_open(name, DS_MODE_NONE, FTAG, &ds); | |
| 1872 if (err) | |
| 1873 return (err); | |
| 1874 | |
| 1875 err = dmu_object_info(ds->ds_dir->dd_pool->dp_meta_objset, | |
| 1876 ds->ds_phys->ds_snapnames_zapobj, &doi); | |
| 1877 if (err) { | |
| 1878 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 1879 return (err); | |
| 1880 } | |
| 1881 | |
| 1882 /* | |
| 1883 * Add in 128x the snapnames zapobj size, since we will be moving | |
| 1884 * a bunch of snapnames to the promoted ds, and dirtying their | |
| 1885 * bonus buffers. | |
| 1886 */ | |
| 2199 | 1887 err = dsl_sync_task_do(ds->ds_dir->dd_pool, |
| 1888 dsl_dataset_promote_check, | |
| 1889 dsl_dataset_promote_sync, ds, &pa, 2 + 2 * doi.doi_physical_blks); | |
| 2082 | 1890 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); |
| 1891 return (err); | |
| 1892 } | |
| 3912 | 1893 |
| 1894 /* | |
| 1895 * Given a pool name and a dataset object number in that pool, | |
| 1896 * return the name of that dataset. | |
| 1897 */ | |
| 1898 int | |
| 1899 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) | |
| 1900 { | |
| 1901 spa_t *spa; | |
| 1902 dsl_pool_t *dp; | |
| 1903 dsl_dataset_t *ds = NULL; | |
| 1904 int error; | |
| 1905 | |
| 1906 if ((error = spa_open(pname, &spa, FTAG)) != 0) | |
| 1907 return (error); | |
| 1908 dp = spa_get_dsl(spa); | |
| 1909 rw_enter(&dp->dp_config_rwlock, RW_READER); | |
| 1910 if ((error = dsl_dataset_open_obj(dp, obj, | |
| 1911 NULL, DS_MODE_NONE, FTAG, &ds)) != 0) { | |
| 1912 rw_exit(&dp->dp_config_rwlock); | |
| 1913 spa_close(spa, FTAG); | |
| 1914 return (error); | |
| 1915 } | |
| 1916 dsl_dataset_name(ds, buf); | |
| 1917 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 1918 rw_exit(&dp->dp_config_rwlock); | |
| 1919 spa_close(spa, FTAG); | |
| 1920 | |
| 1921 return (0); | |
| 1922 } |