Mercurial > illumos > illumos-gate
annotate usr/src/uts/common/fs/zfs/dsl_dataset.c @ 4967:8d0de61ff354
6597381 BAD TRAP: type=e (#pf Page fault) rp=ffffff00043a48f0 addr=28 occurred in module "<unknown>" due to
| author | perrin |
|---|---|
| date | Thu, 30 Aug 2007 15:15:28 -0700 |
| parents | c80bf0e6f4aa |
| children | 71a3e95fb9e2 |
| 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> | |
| 4007 | 40 #include <sys/zfs_ioctl.h> |
| 4543 | 41 #include <sys/spa.h> |
| 42 #include <sys/sunddi.h> | |
| 789 | 43 |
| 2199 | 44 static dsl_checkfunc_t dsl_dataset_destroy_begin_check; |
| 45 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync; | |
| 46 static dsl_checkfunc_t dsl_dataset_rollback_check; | |
| 47 static dsl_syncfunc_t dsl_dataset_rollback_sync; | |
| 48 static dsl_checkfunc_t dsl_dataset_destroy_check; | |
| 49 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
|
50 |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
51 #define DS_REF_MAX (1ULL << 62) |
| 789 | 52 |
| 53 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE | |
| 54 | |
| 55 /* | |
| 56 * We use weighted reference counts to express the various forms of exclusion | |
| 57 * 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
|
58 * is DS_REF_MAX, and a PRIMARY open is little more than half of an EXCLUSIVE. |
| 789 | 59 * 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
|
60 * 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
|
61 * weight (DS_REF_MAX) consumes the entire refcnt space. PRIMARY opens consume |
| 789 | 62 * just over half of the refcnt space, so there can't be more than one, but it |
| 63 * can peacefully coexist with any number of STANDARD opens. | |
| 64 */ | |
| 65 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
|
66 0, /* DS_MODE_NONE - invalid */ |
|
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
67 1, /* DS_MODE_STANDARD - unlimited number */ |
|
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
68 (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
|
69 DS_REF_MAX /* DS_MODE_EXCLUSIVE - no other opens */ |
| 789 | 70 }; |
| 71 | |
| 72 | |
| 73 void | |
| 74 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) | |
| 75 { | |
| 2082 | 76 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp); |
| 789 | 77 int compressed = BP_GET_PSIZE(bp); |
| 78 int uncompressed = BP_GET_UCSIZE(bp); | |
| 79 | |
| 80 dprintf_bp(bp, "born, ds=%p\n", ds); | |
| 81 | |
| 82 ASSERT(dmu_tx_is_syncing(tx)); | |
| 83 /* It could have been compressed away to nothing */ | |
| 84 if (BP_IS_HOLE(bp)) | |
| 85 return; | |
| 86 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); | |
| 87 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES); | |
| 88 if (ds == NULL) { | |
| 89 /* | |
| 90 * Account for the meta-objset space in its placeholder | |
| 91 * dsl_dir. | |
| 92 */ | |
| 93 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */ | |
| 94 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, | |
| 95 used, compressed, uncompressed, tx); | |
| 96 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); | |
| 97 return; | |
| 98 } | |
| 99 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 100 mutex_enter(&ds->ds_lock); | |
| 101 ds->ds_phys->ds_used_bytes += used; | |
| 102 ds->ds_phys->ds_compressed_bytes += compressed; | |
| 103 ds->ds_phys->ds_uncompressed_bytes += uncompressed; | |
| 104 ds->ds_phys->ds_unique_bytes += used; | |
| 105 mutex_exit(&ds->ds_lock); | |
| 106 dsl_dir_diduse_space(ds->ds_dir, | |
| 107 used, compressed, uncompressed, tx); | |
| 108 } | |
| 109 | |
| 110 void | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
111 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
|
112 dmu_tx_t *tx) |
| 789 | 113 { |
| 2082 | 114 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp); |
| 789 | 115 int compressed = BP_GET_PSIZE(bp); |
| 116 int uncompressed = BP_GET_UCSIZE(bp); | |
| 117 | |
| 118 ASSERT(dmu_tx_is_syncing(tx)); | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
119 /* No block pointer => nothing to free */ |
| 789 | 120 if (BP_IS_HOLE(bp)) |
| 121 return; | |
| 122 | |
| 123 ASSERT(used > 0); | |
| 124 if (ds == NULL) { | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
125 int err; |
| 789 | 126 /* |
| 127 * Account for the meta-objset space in its placeholder | |
| 128 * dataset. | |
| 129 */ | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
130 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
|
131 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
|
132 ASSERT(err == 0); |
| 789 | 133 |
| 134 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, | |
| 135 -used, -compressed, -uncompressed, tx); | |
| 136 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); | |
| 137 return; | |
| 138 } | |
| 139 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); | |
| 140 | |
| 141 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 142 | |
| 143 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
|
144 int err; |
|
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
145 |
| 789 | 146 dprintf_bp(bp, "freeing: %s", ""); |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
147 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
|
148 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
|
149 ASSERT(err == 0); |
| 789 | 150 |
| 151 mutex_enter(&ds->ds_lock); | |
| 152 /* XXX unique_bytes is not accurate for head datasets */ | |
| 153 /* ASSERT3U(ds->ds_phys->ds_unique_bytes, >=, used); */ | |
| 154 ds->ds_phys->ds_unique_bytes -= used; | |
| 155 mutex_exit(&ds->ds_lock); | |
| 156 dsl_dir_diduse_space(ds->ds_dir, | |
| 157 -used, -compressed, -uncompressed, tx); | |
| 158 } else { | |
| 159 dprintf_bp(bp, "putting on dead list: %s", ""); | |
| 1544 | 160 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx)); |
| 789 | 161 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ |
| 162 if (ds->ds_phys->ds_prev_snap_obj != 0) { | |
| 163 ASSERT3U(ds->ds_prev->ds_object, ==, | |
| 164 ds->ds_phys->ds_prev_snap_obj); | |
| 165 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); | |
| 166 if (ds->ds_prev->ds_phys->ds_next_snap_obj == | |
| 2082 | 167 ds->ds_object && bp->blk_birth > |
| 789 | 168 ds->ds_prev->ds_phys->ds_prev_snap_txg) { |
| 169 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
| 170 mutex_enter(&ds->ds_prev->ds_lock); | |
| 171 ds->ds_prev->ds_phys->ds_unique_bytes += | |
| 172 used; | |
| 173 mutex_exit(&ds->ds_prev->ds_lock); | |
| 174 } | |
| 175 } | |
| 176 } | |
| 177 mutex_enter(&ds->ds_lock); | |
| 178 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used); | |
| 179 ds->ds_phys->ds_used_bytes -= used; | |
| 180 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); | |
| 181 ds->ds_phys->ds_compressed_bytes -= compressed; | |
| 182 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); | |
| 183 ds->ds_phys->ds_uncompressed_bytes -= uncompressed; | |
| 184 mutex_exit(&ds->ds_lock); | |
| 185 } | |
| 186 | |
| 1544 | 187 uint64_t |
| 188 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) | |
| 789 | 189 { |
| 2885 | 190 uint64_t trysnap = 0; |
| 191 | |
| 789 | 192 if (ds == NULL) |
| 1544 | 193 return (0); |
| 789 | 194 /* |
| 195 * The snapshot creation could fail, but that would cause an | |
| 196 * incorrect FALSE return, which would only result in an | |
| 197 * overestimation of the amount of space that an operation would | |
| 198 * consume, which is OK. | |
| 199 * | |
| 200 * There's also a small window where we could miss a pending | |
| 201 * snapshot, because we could set the sync task in the quiescing | |
| 202 * phase. So this should only be used as a guess. | |
| 203 */ | |
| 2885 | 204 if (ds->ds_trysnap_txg > |
| 205 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) | |
| 206 trysnap = ds->ds_trysnap_txg; | |
| 207 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap)); | |
| 1544 | 208 } |
| 209 | |
| 210 int | |
| 211 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth) | |
| 212 { | |
| 213 return (blk_birth > dsl_dataset_prev_snap_txg(ds)); | |
| 789 | 214 } |
| 215 | |
| 216 /* ARGSUSED */ | |
| 217 static void | |
| 218 dsl_dataset_evict(dmu_buf_t *db, void *dsv) | |
| 219 { | |
| 220 dsl_dataset_t *ds = dsv; | |
| 221 | |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
222 /* open_refcount == DS_REF_MAX when deleting */ |
| 789 | 223 ASSERT(ds->ds_open_refcount == 0 || |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
224 ds->ds_open_refcount == DS_REF_MAX); |
| 789 | 225 |
| 226 dprintf_ds(ds, "evicting %s\n", ""); | |
| 227 | |
| 4787 | 228 unique_remove(ds->ds_fsid_guid); |
| 789 | 229 |
| 230 if (ds->ds_user_ptr != NULL) | |
| 231 ds->ds_user_evict_func(ds, ds->ds_user_ptr); | |
| 232 | |
| 233 if (ds->ds_prev) { | |
| 234 dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds); | |
| 235 ds->ds_prev = NULL; | |
| 236 } | |
| 237 | |
| 238 bplist_close(&ds->ds_deadlist); | |
| 239 dsl_dir_close(ds->ds_dir, ds); | |
| 240 | |
| 4787 | 241 ASSERT(!list_link_active(&ds->ds_synced_link)); |
| 789 | 242 |
| 2856 | 243 mutex_destroy(&ds->ds_lock); |
| 4787 | 244 mutex_destroy(&ds->ds_opening_lock); |
| 2856 | 245 mutex_destroy(&ds->ds_deadlist.bpl_lock); |
| 246 | |
| 789 | 247 kmem_free(ds, sizeof (dsl_dataset_t)); |
| 248 } | |
| 249 | |
| 1544 | 250 static int |
| 789 | 251 dsl_dataset_get_snapname(dsl_dataset_t *ds) |
| 252 { | |
| 253 dsl_dataset_phys_t *headphys; | |
| 254 int err; | |
| 255 dmu_buf_t *headdbuf; | |
| 256 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 257 objset_t *mos = dp->dp_meta_objset; | |
| 258 | |
| 259 if (ds->ds_snapname[0]) | |
| 1544 | 260 return (0); |
| 789 | 261 if (ds->ds_phys->ds_next_snap_obj == 0) |
| 1544 | 262 return (0); |
| 789 | 263 |
| 1544 | 264 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, |
| 265 FTAG, &headdbuf); | |
| 266 if (err) | |
| 267 return (err); | |
| 789 | 268 headphys = headdbuf->db_data; |
| 269 err = zap_value_search(dp->dp_meta_objset, | |
| 4577 | 270 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); |
| 1544 | 271 dmu_buf_rele(headdbuf, FTAG); |
| 272 return (err); | |
| 789 | 273 } |
| 274 | |
| 1544 | 275 int |
| 789 | 276 dsl_dataset_open_obj(dsl_pool_t *dp, uint64_t dsobj, const char *snapname, |
| 1544 | 277 int mode, void *tag, dsl_dataset_t **dsp) |
| 789 | 278 { |
| 279 uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)]; | |
| 280 objset_t *mos = dp->dp_meta_objset; | |
| 281 dmu_buf_t *dbuf; | |
| 282 dsl_dataset_t *ds; | |
| 1544 | 283 int err; |
| 789 | 284 |
| 285 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || | |
| 286 dsl_pool_sync_context(dp)); | |
| 287 | |
| 1544 | 288 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); |
| 289 if (err) | |
| 290 return (err); | |
| 789 | 291 ds = dmu_buf_get_user(dbuf); |
| 292 if (ds == NULL) { | |
| 293 dsl_dataset_t *winner; | |
| 294 | |
| 295 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); | |
| 296 ds->ds_dbuf = dbuf; | |
| 297 ds->ds_object = dsobj; | |
| 298 ds->ds_phys = dbuf->db_data; | |
| 299 | |
| 2856 | 300 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); |
| 4787 | 301 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); |
| 2856 | 302 mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT, |
| 303 NULL); | |
| 304 | |
| 1544 | 305 err = bplist_open(&ds->ds_deadlist, |
| 789 | 306 mos, ds->ds_phys->ds_deadlist_obj); |
| 1544 | 307 if (err == 0) { |
| 308 err = dsl_dir_open_obj(dp, | |
| 309 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); | |
| 310 } | |
| 311 if (err) { | |
| 312 /* | |
| 313 * we don't really need to close the blist if we | |
| 314 * just opened it. | |
| 315 */ | |
| 2856 | 316 mutex_destroy(&ds->ds_lock); |
| 4787 | 317 mutex_destroy(&ds->ds_opening_lock); |
| 2856 | 318 mutex_destroy(&ds->ds_deadlist.bpl_lock); |
| 1544 | 319 kmem_free(ds, sizeof (dsl_dataset_t)); |
| 320 dmu_buf_rele(dbuf, tag); | |
| 321 return (err); | |
| 322 } | |
| 789 | 323 |
| 324 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == dsobj) { | |
| 325 ds->ds_snapname[0] = '\0'; | |
| 326 if (ds->ds_phys->ds_prev_snap_obj) { | |
| 1544 | 327 err = dsl_dataset_open_obj(dp, |
| 789 | 328 ds->ds_phys->ds_prev_snap_obj, NULL, |
| 1544 | 329 DS_MODE_NONE, ds, &ds->ds_prev); |
| 789 | 330 } |
| 331 } else { | |
| 332 if (snapname) { | |
| 333 #ifdef ZFS_DEBUG | |
| 334 dsl_dataset_phys_t *headphys; | |
| 1544 | 335 dmu_buf_t *headdbuf; |
| 336 err = dmu_bonus_hold(mos, | |
| 337 ds->ds_dir->dd_phys->dd_head_dataset_obj, | |
| 338 FTAG, &headdbuf); | |
| 339 if (err == 0) { | |
| 340 headphys = headdbuf->db_data; | |
| 341 uint64_t foundobj; | |
| 342 err = zap_lookup(dp->dp_meta_objset, | |
| 343 headphys->ds_snapnames_zapobj, | |
| 344 snapname, sizeof (foundobj), 1, | |
| 345 &foundobj); | |
| 346 ASSERT3U(foundobj, ==, dsobj); | |
| 347 dmu_buf_rele(headdbuf, FTAG); | |
| 348 } | |
| 789 | 349 #endif |
| 350 (void) strcat(ds->ds_snapname, snapname); | |
| 351 } else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) { | |
| 1544 | 352 err = dsl_dataset_get_snapname(ds); |
| 789 | 353 } |
| 354 } | |
| 355 | |
| 1544 | 356 if (err == 0) { |
| 357 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys, | |
| 358 dsl_dataset_evict); | |
| 359 } | |
| 360 if (err || winner) { | |
| 789 | 361 bplist_close(&ds->ds_deadlist); |
| 362 if (ds->ds_prev) { | |
| 363 dsl_dataset_close(ds->ds_prev, | |
| 364 DS_MODE_NONE, ds); | |
| 365 } | |
| 366 dsl_dir_close(ds->ds_dir, ds); | |
| 2856 | 367 mutex_destroy(&ds->ds_lock); |
| 4787 | 368 mutex_destroy(&ds->ds_opening_lock); |
| 2856 | 369 mutex_destroy(&ds->ds_deadlist.bpl_lock); |
| 789 | 370 kmem_free(ds, sizeof (dsl_dataset_t)); |
| 1544 | 371 if (err) { |
| 372 dmu_buf_rele(dbuf, tag); | |
| 373 return (err); | |
| 374 } | |
| 789 | 375 ds = winner; |
| 376 } else { | |
| 4787 | 377 ds->ds_fsid_guid = |
| 789 | 378 unique_insert(ds->ds_phys->ds_fsid_guid); |
| 379 } | |
| 380 } | |
| 381 ASSERT3P(ds->ds_dbuf, ==, dbuf); | |
| 382 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); | |
| 383 | |
| 384 mutex_enter(&ds->ds_lock); | |
| 385 if ((DS_MODE_LEVEL(mode) == DS_MODE_PRIMARY && | |
| 2082 | 386 (ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) && |
| 387 !DS_MODE_IS_INCONSISTENT(mode)) || | |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
388 (ds->ds_open_refcount + weight > DS_REF_MAX)) { |
| 789 | 389 mutex_exit(&ds->ds_lock); |
| 390 dsl_dataset_close(ds, DS_MODE_NONE, tag); | |
| 1544 | 391 return (EBUSY); |
| 789 | 392 } |
| 393 ds->ds_open_refcount += weight; | |
| 394 mutex_exit(&ds->ds_lock); | |
| 395 | |
| 1544 | 396 *dsp = ds; |
| 397 return (0); | |
| 789 | 398 } |
| 399 | |
| 400 int | |
| 401 dsl_dataset_open_spa(spa_t *spa, const char *name, int mode, | |
| 402 void *tag, dsl_dataset_t **dsp) | |
| 403 { | |
| 404 dsl_dir_t *dd; | |
| 405 dsl_pool_t *dp; | |
| 406 const char *tail; | |
| 407 uint64_t obj; | |
| 408 dsl_dataset_t *ds = NULL; | |
| 409 int err = 0; | |
| 410 | |
| 1544 | 411 err = dsl_dir_open_spa(spa, name, FTAG, &dd, &tail); |
| 412 if (err) | |
| 413 return (err); | |
| 789 | 414 |
| 415 dp = dd->dd_pool; | |
| 416 obj = dd->dd_phys->dd_head_dataset_obj; | |
| 417 rw_enter(&dp->dp_config_rwlock, RW_READER); | |
| 418 if (obj == 0) { | |
| 419 /* A dataset with no associated objset */ | |
| 420 err = ENOENT; | |
| 421 goto out; | |
| 422 } | |
| 423 | |
| 424 if (tail != NULL) { | |
| 425 objset_t *mos = dp->dp_meta_objset; | |
| 426 | |
| 1544 | 427 err = dsl_dataset_open_obj(dp, obj, NULL, |
| 428 DS_MODE_NONE, tag, &ds); | |
| 429 if (err) | |
| 430 goto out; | |
| 789 | 431 obj = ds->ds_phys->ds_snapnames_zapobj; |
| 432 dsl_dataset_close(ds, DS_MODE_NONE, tag); | |
| 433 ds = NULL; | |
| 434 | |
| 435 if (tail[0] != '@') { | |
| 436 err = ENOENT; | |
| 437 goto out; | |
| 438 } | |
| 439 tail++; | |
| 440 | |
| 441 /* Look for a snapshot */ | |
| 442 if (!DS_MODE_IS_READONLY(mode)) { | |
| 443 err = EROFS; | |
| 444 goto out; | |
| 445 } | |
| 446 dprintf("looking for snapshot '%s'\n", tail); | |
| 447 err = zap_lookup(mos, obj, tail, 8, 1, &obj); | |
| 448 if (err) | |
| 449 goto out; | |
| 450 } | |
| 1544 | 451 err = dsl_dataset_open_obj(dp, obj, tail, mode, tag, &ds); |
| 789 | 452 |
| 453 out: | |
| 454 rw_exit(&dp->dp_config_rwlock); | |
| 455 dsl_dir_close(dd, FTAG); | |
| 456 | |
| 457 ASSERT3U((err == 0), ==, (ds != NULL)); | |
| 458 /* ASSERT(ds == NULL || strcmp(name, ds->ds_name) == 0); */ | |
| 459 | |
| 460 *dsp = ds; | |
| 461 return (err); | |
| 462 } | |
| 463 | |
| 464 int | |
| 465 dsl_dataset_open(const char *name, int mode, void *tag, dsl_dataset_t **dsp) | |
| 466 { | |
| 467 return (dsl_dataset_open_spa(NULL, name, mode, tag, dsp)); | |
| 468 } | |
| 469 | |
| 470 void | |
| 471 dsl_dataset_name(dsl_dataset_t *ds, char *name) | |
| 472 { | |
| 473 if (ds == NULL) { | |
| 474 (void) strcpy(name, "mos"); | |
| 475 } else { | |
| 476 dsl_dir_name(ds->ds_dir, name); | |
| 1544 | 477 VERIFY(0 == dsl_dataset_get_snapname(ds)); |
| 789 | 478 if (ds->ds_snapname[0]) { |
| 479 (void) strcat(name, "@"); | |
| 480 if (!MUTEX_HELD(&ds->ds_lock)) { | |
| 481 /* | |
| 482 * We use a "recursive" mutex so that we | |
| 483 * can call dprintf_ds() with ds_lock held. | |
| 484 */ | |
| 485 mutex_enter(&ds->ds_lock); | |
| 486 (void) strcat(name, ds->ds_snapname); | |
| 487 mutex_exit(&ds->ds_lock); | |
| 488 } else { | |
| 489 (void) strcat(name, ds->ds_snapname); | |
| 490 } | |
| 491 } | |
| 492 } | |
| 493 } | |
| 494 | |
|
3978
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
495 static int |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
496 dsl_dataset_namelen(dsl_dataset_t *ds) |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
497 { |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
498 int result; |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
499 |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
500 if (ds == NULL) { |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
501 result = 3; /* "mos" */ |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
502 } else { |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
503 result = dsl_dir_namelen(ds->ds_dir); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
504 VERIFY(0 == dsl_dataset_get_snapname(ds)); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
505 if (ds->ds_snapname[0]) { |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
506 ++result; /* adding one for the @-sign */ |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
507 if (!MUTEX_HELD(&ds->ds_lock)) { |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
508 /* see dsl_datset_name */ |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
509 mutex_enter(&ds->ds_lock); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
510 result += strlen(ds->ds_snapname); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
511 mutex_exit(&ds->ds_lock); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
512 } else { |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
513 result += strlen(ds->ds_snapname); |
|
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 |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
518 return (result); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
519 } |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
520 |
| 789 | 521 void |
| 522 dsl_dataset_close(dsl_dataset_t *ds, int mode, void *tag) | |
| 523 { | |
| 524 uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)]; | |
| 525 mutex_enter(&ds->ds_lock); | |
| 526 ASSERT3U(ds->ds_open_refcount, >=, weight); | |
| 527 ds->ds_open_refcount -= weight; | |
| 528 dprintf_ds(ds, "closing mode %u refcount now 0x%llx\n", | |
| 529 mode, ds->ds_open_refcount); | |
| 530 mutex_exit(&ds->ds_lock); | |
| 531 | |
| 1544 | 532 dmu_buf_rele(ds->ds_dbuf, tag); |
| 789 | 533 } |
| 534 | |
| 535 void | |
| 536 dsl_dataset_create_root(dsl_pool_t *dp, uint64_t *ddobjp, dmu_tx_t *tx) | |
| 537 { | |
| 538 objset_t *mos = dp->dp_meta_objset; | |
| 539 dmu_buf_t *dbuf; | |
| 540 dsl_dataset_phys_t *dsphys; | |
| 541 dsl_dataset_t *ds; | |
| 542 uint64_t dsobj; | |
| 543 dsl_dir_t *dd; | |
| 544 | |
| 545 dsl_dir_create_root(mos, ddobjp, tx); | |
| 1544 | 546 VERIFY(0 == dsl_dir_open_obj(dp, *ddobjp, NULL, FTAG, &dd)); |
| 789 | 547 |
|
928
36d72fe4da29
6349314 dmu_object_type names incorrect for DSL Directories and DSL Datasets
tabriz
parents:
885
diff
changeset
|
548 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
|
549 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); |
| 1544 | 550 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); |
| 789 | 551 dmu_buf_will_dirty(dbuf, tx); |
| 552 dsphys = dbuf->db_data; | |
| 553 dsphys->ds_dir_obj = dd->dd_object; | |
| 554 dsphys->ds_fsid_guid = unique_create(); | |
| 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 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, | |
| 604 sizeof (dsphys->ds_guid)); | |
| 605 dsphys->ds_snapnames_zapobj = | |
|
885
d925b21dba78
6347493 tar of 25K empty directory entries in ZFS takes 30+ seconds ...
ahrens
parents:
789
diff
changeset
|
606 zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx); |
| 789 | 607 dsphys->ds_creation_time = gethrestime_sec(); |
| 608 dsphys->ds_creation_txg = tx->tx_txg; | |
| 609 dsphys->ds_deadlist_obj = | |
| 610 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 611 if (clone_parent) { | |
| 612 dsphys->ds_prev_snap_obj = clone_parent->ds_object; | |
| 613 dsphys->ds_prev_snap_txg = | |
| 614 clone_parent->ds_phys->ds_creation_txg; | |
| 615 dsphys->ds_used_bytes = | |
| 616 clone_parent->ds_phys->ds_used_bytes; | |
| 617 dsphys->ds_compressed_bytes = | |
| 618 clone_parent->ds_phys->ds_compressed_bytes; | |
| 619 dsphys->ds_uncompressed_bytes = | |
| 620 clone_parent->ds_phys->ds_uncompressed_bytes; | |
| 621 dsphys->ds_bp = clone_parent->ds_phys->ds_bp; | |
| 622 | |
| 623 dmu_buf_will_dirty(clone_parent->ds_dbuf, tx); | |
| 624 clone_parent->ds_phys->ds_num_children++; | |
| 625 | |
| 626 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 627 dd->dd_phys->dd_clone_parent_obj = clone_parent->ds_object; | |
| 628 } | |
| 1544 | 629 dmu_buf_rele(dbuf, FTAG); |
| 789 | 630 |
| 631 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 632 dd->dd_phys->dd_head_dataset_obj = dsobj; | |
| 633 dsl_dir_close(dd, FTAG); | |
| 634 | |
| 2199 | 635 return (dsobj); |
| 636 } | |
| 637 | |
| 638 struct destroyarg { | |
| 639 dsl_sync_task_group_t *dstg; | |
| 640 char *snapname; | |
| 641 char *failed; | |
| 642 }; | |
| 643 | |
| 644 static int | |
| 645 dsl_snapshot_destroy_one(char *name, void *arg) | |
| 646 { | |
| 647 struct destroyarg *da = arg; | |
| 648 dsl_dataset_t *ds; | |
| 649 char *cp; | |
| 650 int err; | |
| 651 | |
| 652 (void) strcat(name, "@"); | |
| 653 (void) strcat(name, da->snapname); | |
| 654 err = dsl_dataset_open(name, | |
| 655 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT, | |
| 4007 | 656 da->dstg, &ds); |
| 2199 | 657 cp = strchr(name, '@'); |
| 658 *cp = '\0'; | |
| 659 if (err == ENOENT) | |
| 660 return (0); | |
| 661 if (err) { | |
| 662 (void) strcpy(da->failed, name); | |
| 663 return (err); | |
| 664 } | |
| 665 | |
| 666 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check, | |
| 4007 | 667 dsl_dataset_destroy_sync, ds, da->dstg, 0); |
| 789 | 668 return (0); |
| 669 } | |
| 670 | |
| 2199 | 671 /* |
| 672 * Destroy 'snapname' in all descendants of 'fsname'. | |
| 673 */ | |
| 674 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy | |
| 675 int | |
| 676 dsl_snapshots_destroy(char *fsname, char *snapname) | |
| 677 { | |
| 678 int err; | |
| 679 struct destroyarg da; | |
| 680 dsl_sync_task_t *dst; | |
| 681 spa_t *spa; | |
| 682 | |
|
4603
c7840c367d00
6494569 zfs recv -d pool/<doesn't exist> core dumps for top-level filesystem backups
ahrens
parents:
4577
diff
changeset
|
683 err = spa_open(fsname, &spa, FTAG); |
| 2199 | 684 if (err) |
| 685 return (err); | |
| 686 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); | |
| 687 da.snapname = snapname; | |
| 688 da.failed = fsname; | |
| 689 | |
| 690 err = dmu_objset_find(fsname, | |
| 2417 | 691 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN); |
| 2199 | 692 |
| 693 if (err == 0) | |
| 694 err = dsl_sync_task_group_wait(da.dstg); | |
| 695 | |
| 696 for (dst = list_head(&da.dstg->dstg_tasks); dst; | |
| 697 dst = list_next(&da.dstg->dstg_tasks, dst)) { | |
| 698 dsl_dataset_t *ds = dst->dst_arg1; | |
| 699 if (dst->dst_err) { | |
| 700 dsl_dataset_name(ds, fsname); | |
|
4603
c7840c367d00
6494569 zfs recv -d pool/<doesn't exist> core dumps for top-level filesystem backups
ahrens
parents:
4577
diff
changeset
|
701 *strchr(fsname, '@') = '\0'; |
| 2199 | 702 } |
| 703 /* | |
| 704 * If it was successful, destroy_sync would have | |
| 705 * closed the ds | |
| 706 */ | |
| 707 if (err) | |
| 4007 | 708 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, da.dstg); |
| 2199 | 709 } |
| 710 | |
| 711 dsl_sync_task_group_destroy(da.dstg); | |
| 712 spa_close(spa, FTAG); | |
| 713 return (err); | |
| 714 } | |
| 715 | |
| 789 | 716 int |
| 717 dsl_dataset_destroy(const char *name) | |
| 718 { | |
| 719 int err; | |
| 2199 | 720 dsl_sync_task_group_t *dstg; |
| 721 objset_t *os; | |
| 722 dsl_dataset_t *ds; | |
| 789 | 723 dsl_dir_t *dd; |
| 2199 | 724 uint64_t obj; |
| 725 | |
| 726 if (strchr(name, '@')) { | |
| 727 /* Destroying a snapshot is simpler */ | |
| 728 err = dsl_dataset_open(name, | |
| 729 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT, | |
| 730 FTAG, &ds); | |
| 731 if (err) | |
| 732 return (err); | |
| 733 err = dsl_sync_task_do(ds->ds_dir->dd_pool, | |
| 734 dsl_dataset_destroy_check, dsl_dataset_destroy_sync, | |
| 735 ds, FTAG, 0); | |
| 736 if (err) | |
| 737 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 738 return (err); | |
| 739 } | |
| 740 | |
| 741 err = dmu_objset_open(name, DMU_OST_ANY, | |
| 742 DS_MODE_EXCLUSIVE | DS_MODE_INCONSISTENT, &os); | |
| 743 if (err) | |
| 744 return (err); | |
| 745 ds = os->os->os_dsl_dataset; | |
| 746 dd = ds->ds_dir; | |
| 789 | 747 |
| 2199 | 748 /* |
| 749 * Check for errors and mark this ds as inconsistent, in | |
| 750 * case we crash while freeing the objects. | |
| 751 */ | |
| 752 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check, | |
| 753 dsl_dataset_destroy_begin_sync, ds, NULL, 0); | |
| 754 if (err) { | |
| 755 dmu_objset_close(os); | |
| 756 return (err); | |
| 757 } | |
| 758 | |
| 759 /* | |
| 760 * remove the objects in open context, so that we won't | |
| 761 * have too much to do in syncing context. | |
| 762 */ | |
|
3025
4e5ee8301d84
6424466 "panic: data after EOF" when unmounting abused pool
ahrens
parents:
2885
diff
changeset
|
763 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
|
764 ds->ds_phys->ds_prev_snap_txg)) { |
| 2199 | 765 dmu_tx_t *tx = dmu_tx_create(os); |
| 766 dmu_tx_hold_free(tx, obj, 0, DMU_OBJECT_END); | |
| 767 dmu_tx_hold_bonus(tx, obj); | |
| 768 err = dmu_tx_assign(tx, TXG_WAIT); | |
| 769 if (err) { | |
| 770 /* | |
| 771 * Perhaps there is not enough disk | |
| 772 * space. Just deal with it from | |
| 773 * dsl_dataset_destroy_sync(). | |
| 774 */ | |
| 775 dmu_tx_abort(tx); | |
| 776 continue; | |
| 777 } | |
| 778 VERIFY(0 == dmu_object_free(os, obj, tx)); | |
| 779 dmu_tx_commit(tx); | |
| 780 } | |
| 781 /* Make sure it's not dirty before we finish destroying it. */ | |
| 782 txg_wait_synced(dd->dd_pool, 0); | |
| 783 | |
| 784 dmu_objset_close(os); | |
| 785 if (err != ESRCH) | |
| 786 return (err); | |
| 787 | |
| 788 err = dsl_dataset_open(name, | |
| 789 DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT, | |
| 790 FTAG, &ds); | |
| 1544 | 791 if (err) |
| 792 return (err); | |
| 789 | 793 |
| 2199 | 794 err = dsl_dir_open(name, FTAG, &dd, NULL); |
| 795 if (err) { | |
| 796 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 797 return (err); | |
| 789 | 798 } |
| 799 | |
| 2199 | 800 /* |
| 801 * Blow away the dsl_dir + head dataset. | |
| 802 */ | |
| 803 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool); | |
| 804 dsl_sync_task_create(dstg, dsl_dataset_destroy_check, | |
| 805 dsl_dataset_destroy_sync, ds, FTAG, 0); | |
| 806 dsl_sync_task_create(dstg, dsl_dir_destroy_check, | |
| 807 dsl_dir_destroy_sync, dd, FTAG, 0); | |
| 808 err = dsl_sync_task_group_wait(dstg); | |
| 809 dsl_sync_task_group_destroy(dstg); | |
| 810 /* if it is successful, *destroy_sync will close the ds+dd */ | |
| 811 if (err) { | |
| 812 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 813 dsl_dir_close(dd, FTAG); | |
| 814 } | |
| 789 | 815 return (err); |
| 816 } | |
| 817 | |
| 818 int | |
| 2199 | 819 dsl_dataset_rollback(dsl_dataset_t *ds) |
| 789 | 820 { |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
821 ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX); |
| 2199 | 822 return (dsl_sync_task_do(ds->ds_dir->dd_pool, |
| 823 dsl_dataset_rollback_check, dsl_dataset_rollback_sync, | |
| 824 ds, NULL, 0)); | |
| 789 | 825 } |
| 826 | |
| 827 void * | |
| 828 dsl_dataset_set_user_ptr(dsl_dataset_t *ds, | |
| 829 void *p, dsl_dataset_evict_func_t func) | |
| 830 { | |
| 831 void *old; | |
| 832 | |
| 833 mutex_enter(&ds->ds_lock); | |
| 834 old = ds->ds_user_ptr; | |
| 835 if (old == NULL) { | |
| 836 ds->ds_user_ptr = p; | |
| 837 ds->ds_user_evict_func = func; | |
| 838 } | |
| 839 mutex_exit(&ds->ds_lock); | |
| 840 return (old); | |
| 841 } | |
| 842 | |
| 843 void * | |
| 844 dsl_dataset_get_user_ptr(dsl_dataset_t *ds) | |
| 845 { | |
| 846 return (ds->ds_user_ptr); | |
| 847 } | |
| 848 | |
| 849 | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
850 blkptr_t * |
|
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
851 dsl_dataset_get_blkptr(dsl_dataset_t *ds) |
| 789 | 852 { |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
853 return (&ds->ds_phys->ds_bp); |
| 789 | 854 } |
| 855 | |
| 856 void | |
| 857 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) | |
| 858 { | |
| 859 ASSERT(dmu_tx_is_syncing(tx)); | |
| 860 /* If it's the meta-objset, set dp_meta_rootbp */ | |
| 861 if (ds == NULL) { | |
| 862 tx->tx_pool->dp_meta_rootbp = *bp; | |
| 863 } else { | |
| 864 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 865 ds->ds_phys->ds_bp = *bp; | |
| 866 } | |
| 867 } | |
| 868 | |
| 869 spa_t * | |
| 870 dsl_dataset_get_spa(dsl_dataset_t *ds) | |
| 871 { | |
| 872 return (ds->ds_dir->dd_pool->dp_spa); | |
| 873 } | |
| 874 | |
| 875 void | |
| 876 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) | |
| 877 { | |
| 878 dsl_pool_t *dp; | |
| 879 | |
| 880 if (ds == NULL) /* this is the meta-objset */ | |
| 881 return; | |
| 882 | |
| 883 ASSERT(ds->ds_user_ptr != NULL); | |
| 2885 | 884 |
| 885 if (ds->ds_phys->ds_next_snap_obj != 0) | |
| 886 panic("dirtying snapshot!"); | |
| 789 | 887 |
| 888 dp = ds->ds_dir->dd_pool; | |
| 889 | |
| 890 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) { | |
| 891 /* up the hold count until we can be written out */ | |
| 892 dmu_buf_add_ref(ds->ds_dbuf, ds); | |
| 893 } | |
| 894 } | |
| 895 | |
| 896 struct killarg { | |
| 897 uint64_t *usedp; | |
| 898 uint64_t *compressedp; | |
| 899 uint64_t *uncompressedp; | |
| 900 zio_t *zio; | |
| 901 dmu_tx_t *tx; | |
| 902 }; | |
| 903 | |
| 904 static int | |
| 905 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg) | |
| 906 { | |
| 907 struct killarg *ka = arg; | |
| 908 blkptr_t *bp = &bc->bc_blkptr; | |
| 909 | |
| 910 ASSERT3U(bc->bc_errno, ==, 0); | |
| 911 | |
| 912 /* | |
| 913 * Since this callback is not called concurrently, no lock is | |
| 914 * needed on the accounting values. | |
| 915 */ | |
| 2082 | 916 *ka->usedp += bp_get_dasize(spa, bp); |
| 789 | 917 *ka->compressedp += BP_GET_PSIZE(bp); |
| 918 *ka->uncompressedp += BP_GET_UCSIZE(bp); | |
| 919 /* XXX check for EIO? */ | |
| 920 (void) arc_free(ka->zio, spa, ka->tx->tx_txg, bp, NULL, NULL, | |
| 921 ARC_NOWAIT); | |
| 922 return (0); | |
| 923 } | |
| 924 | |
| 925 /* ARGSUSED */ | |
| 2199 | 926 static int |
| 927 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 789 | 928 { |
| 2199 | 929 dsl_dataset_t *ds = arg1; |
| 789 | 930 |
| 2199 | 931 /* |
| 932 * There must be a previous snapshot. I suppose we could roll | |
| 933 * it back to being empty (and re-initialize the upper (ZPL) | |
| 934 * layer). But for now there's no way to do this via the user | |
| 935 * interface. | |
| 936 */ | |
| 937 if (ds->ds_phys->ds_prev_snap_txg == 0) | |
| 789 | 938 return (EINVAL); |
| 939 | |
| 2199 | 940 /* |
| 941 * This must not be a snapshot. | |
| 942 */ | |
| 943 if (ds->ds_phys->ds_next_snap_obj != 0) | |
| 944 return (EINVAL); | |
| 789 | 945 |
| 946 /* | |
| 947 * If we made changes this txg, traverse_dsl_dataset won't find | |
| 948 * them. Try again. | |
| 949 */ | |
| 2199 | 950 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) |
| 789 | 951 return (EAGAIN); |
| 2199 | 952 |
| 953 return (0); | |
| 954 } | |
| 789 | 955 |
| 2199 | 956 /* ARGSUSED */ |
| 957 static void | |
| 4543 | 958 dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) |
| 2199 | 959 { |
| 960 dsl_dataset_t *ds = arg1; | |
| 961 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; | |
| 789 | 962 |
| 963 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 964 | |
|
4967
8d0de61ff354
6597381 BAD TRAP: type=e (#pf Page fault) rp=ffffff00043a48f0 addr=28 occurred in module "<unknown>" due to
perrin
parents:
4935
diff
changeset
|
965 /* |
|
8d0de61ff354
6597381 BAD TRAP: type=e (#pf Page fault) rp=ffffff00043a48f0 addr=28 occurred in module "<unknown>" due to
perrin
parents:
4935
diff
changeset
|
966 * Before the roll back destroy the zil. |
|
8d0de61ff354
6597381 BAD TRAP: type=e (#pf Page fault) rp=ffffff00043a48f0 addr=28 occurred in module "<unknown>" due to
perrin
parents:
4935
diff
changeset
|
967 * Note, ds_user_ptr can be null if we are doing a "zfs receive -F" |
|
8d0de61ff354
6597381 BAD TRAP: type=e (#pf Page fault) rp=ffffff00043a48f0 addr=28 occurred in module "<unknown>" due to
perrin
parents:
4935
diff
changeset
|
968 */ |
|
8d0de61ff354
6597381 BAD TRAP: type=e (#pf Page fault) rp=ffffff00043a48f0 addr=28 occurred in module "<unknown>" due to
perrin
parents:
4935
diff
changeset
|
969 if (ds->ds_user_ptr != NULL) { |
|
8d0de61ff354
6597381 BAD TRAP: type=e (#pf Page fault) rp=ffffff00043a48f0 addr=28 occurred in module "<unknown>" due to
perrin
parents:
4935
diff
changeset
|
970 zil_rollback_destroy( |
|
8d0de61ff354
6597381 BAD TRAP: type=e (#pf Page fault) rp=ffffff00043a48f0 addr=28 occurred in module "<unknown>" due to
perrin
parents:
4935
diff
changeset
|
971 ((objset_impl_t *)ds->ds_user_ptr)->os_zil, tx); |
|
8d0de61ff354
6597381 BAD TRAP: type=e (#pf Page fault) rp=ffffff00043a48f0 addr=28 occurred in module "<unknown>" due to
perrin
parents:
4935
diff
changeset
|
972 } |
|
4935
c80bf0e6f4aa
6534949 Stale need for range locking comment in zvol.c
perrin
parents:
4787
diff
changeset
|
973 |
| 789 | 974 /* Zero out the deadlist. */ |
| 975 bplist_close(&ds->ds_deadlist); | |
| 976 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); | |
| 977 ds->ds_phys->ds_deadlist_obj = | |
| 978 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 1544 | 979 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos, |
| 980 ds->ds_phys->ds_deadlist_obj)); | |
| 789 | 981 |
| 982 { | |
| 983 /* Free blkptrs that we gave birth to */ | |
| 984 zio_t *zio; | |
| 985 uint64_t used = 0, compressed = 0, uncompressed = 0; | |
| 986 struct killarg ka; | |
| 987 | |
| 988 zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL, | |
| 989 ZIO_FLAG_MUSTSUCCEED); | |
| 990 ka.usedp = &used; | |
| 991 ka.compressedp = &compressed; | |
| 992 ka.uncompressedp = &uncompressed; | |
| 993 ka.zio = zio; | |
| 994 ka.tx = tx; | |
| 995 (void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg, | |
| 996 ADVANCE_POST, kill_blkptr, &ka); | |
| 997 (void) zio_wait(zio); | |
| 998 | |
| 2199 | 999 dsl_dir_diduse_space(ds->ds_dir, |
| 789 | 1000 -used, -compressed, -uncompressed, tx); |
| 1001 } | |
| 1002 | |
| 2199 | 1003 /* Change our contents to that of the prev snapshot */ |
| 789 | 1004 ASSERT3U(ds->ds_prev->ds_object, ==, ds->ds_phys->ds_prev_snap_obj); |
| 1005 ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp; | |
| 1006 ds->ds_phys->ds_used_bytes = ds->ds_prev->ds_phys->ds_used_bytes; | |
| 1007 ds->ds_phys->ds_compressed_bytes = | |
| 1008 ds->ds_prev->ds_phys->ds_compressed_bytes; | |
| 1009 ds->ds_phys->ds_uncompressed_bytes = | |
| 1010 ds->ds_prev->ds_phys->ds_uncompressed_bytes; | |
| 2082 | 1011 ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags; |
| 789 | 1012 ds->ds_phys->ds_unique_bytes = 0; |
| 1013 | |
|
2532
752725c22841
6448999 panic: used == ds->ds_phys->ds_unique_bytes
ahrens
parents:
2417
diff
changeset
|
1014 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
|
1015 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
|
1016 ds->ds_prev->ds_phys->ds_unique_bytes = 0; |
|
752725c22841
6448999 panic: used == ds->ds_phys->ds_unique_bytes
ahrens
parents:
2417
diff
changeset
|
1017 } |
| 4543 | 1018 |
| 1019 spa_history_internal_log(LOG_DS_ROLLBACK, ds->ds_dir->dd_pool->dp_spa, | |
| 1020 tx, cr, "dataset = %llu", ds->ds_object); | |
| 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 | |
| 4543 | 1043 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) |
| 789 | 1044 { |
| 2199 | 1045 dsl_dataset_t *ds = arg1; |
| 4543 | 1046 dsl_pool_t *dp = ds->ds_dir->dd_pool; |
| 789 | 1047 |
| 2199 | 1048 /* Mark it as inconsistent on-disk, in case we crash */ |
| 1049 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 1050 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; | |
| 4543 | 1051 |
| 1052 spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx, | |
| 1053 cr, "dataset = %llu", ds->ds_object); | |
| 2199 | 1054 } |
| 789 | 1055 |
| 2199 | 1056 /* ARGSUSED */ |
| 1057 static int | |
| 1058 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 1059 { | |
| 1060 dsl_dataset_t *ds = arg1; | |
| 789 | 1061 |
| 1062 /* Can't delete a branch point. */ | |
| 2199 | 1063 if (ds->ds_phys->ds_num_children > 1) |
| 1064 return (EEXIST); | |
| 789 | 1065 |
| 1066 /* | |
| 1067 * Can't delete a head dataset if there are snapshots of it. | |
| 1068 * (Except if the only snapshots are from the branch we cloned | |
| 1069 * from.) | |
| 1070 */ | |
| 1071 if (ds->ds_prev != NULL && | |
| 2199 | 1072 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) |
| 789 | 1073 return (EINVAL); |
| 1074 | |
| 1075 /* | |
| 1076 * If we made changes this txg, traverse_dsl_dataset won't find | |
| 1077 * them. Try again. | |
| 1078 */ | |
| 2199 | 1079 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) |
| 789 | 1080 return (EAGAIN); |
| 2199 | 1081 |
| 1082 /* XXX we should do some i/o error checking... */ | |
| 1083 return (0); | |
| 1084 } | |
| 1085 | |
| 1086 static void | |
| 4543 | 1087 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx) |
| 2199 | 1088 { |
| 1089 dsl_dataset_t *ds = arg1; | |
| 1090 uint64_t used = 0, compressed = 0, uncompressed = 0; | |
| 1091 zio_t *zio; | |
| 1092 int err; | |
| 1093 int after_branch_point = FALSE; | |
| 1094 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 1095 objset_t *mos = dp->dp_meta_objset; | |
| 1096 dsl_dataset_t *ds_prev = NULL; | |
| 1097 uint64_t obj; | |
| 1098 | |
|
3444
dc160a70a50d
6410433 'zpool status -v' would be more useful with filenames
ek110237
parents:
3025
diff
changeset
|
1099 ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX); |
| 2199 | 1100 ASSERT3U(ds->ds_phys->ds_num_children, <=, 1); |
| 1101 ASSERT(ds->ds_prev == NULL || | |
| 1102 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object); | |
| 1103 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); | |
| 1104 | |
| 1105 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); | |
| 1106 | |
| 1107 obj = ds->ds_object; | |
| 789 | 1108 |
| 1109 if (ds->ds_phys->ds_prev_snap_obj != 0) { | |
| 1110 if (ds->ds_prev) { | |
| 1111 ds_prev = ds->ds_prev; | |
| 1112 } else { | |
| 2199 | 1113 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 789 | 1114 ds->ds_phys->ds_prev_snap_obj, NULL, |
| 2199 | 1115 DS_MODE_NONE, FTAG, &ds_prev)); |
| 789 | 1116 } |
| 1117 after_branch_point = | |
| 1118 (ds_prev->ds_phys->ds_next_snap_obj != obj); | |
| 1119 | |
| 1120 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); | |
| 1121 if (after_branch_point && | |
| 1122 ds->ds_phys->ds_next_snap_obj == 0) { | |
| 1123 /* This clone is toast. */ | |
| 1124 ASSERT(ds_prev->ds_phys->ds_num_children > 1); | |
| 1125 ds_prev->ds_phys->ds_num_children--; | |
| 1126 } else if (!after_branch_point) { | |
| 1127 ds_prev->ds_phys->ds_next_snap_obj = | |
| 1128 ds->ds_phys->ds_next_snap_obj; | |
| 1129 } | |
| 1130 } | |
| 1131 | |
| 1132 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
| 1133 | |
| 1134 if (ds->ds_phys->ds_next_snap_obj != 0) { | |
| 2199 | 1135 blkptr_t bp; |
| 789 | 1136 dsl_dataset_t *ds_next; |
| 1137 uint64_t itor = 0; | |
| 1138 | |
| 1139 spa_scrub_restart(dp->dp_spa, tx->tx_txg); | |
| 1140 | |
| 2199 | 1141 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 1544 | 1142 ds->ds_phys->ds_next_snap_obj, NULL, |
| 1143 DS_MODE_NONE, FTAG, &ds_next)); | |
| 789 | 1144 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); |
| 1145 | |
| 1146 dmu_buf_will_dirty(ds_next->ds_dbuf, tx); | |
| 1147 ds_next->ds_phys->ds_prev_snap_obj = | |
| 1148 ds->ds_phys->ds_prev_snap_obj; | |
| 1149 ds_next->ds_phys->ds_prev_snap_txg = | |
| 1150 ds->ds_phys->ds_prev_snap_txg; | |
| 1151 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, | |
| 1152 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); | |
| 1153 | |
| 1154 /* | |
| 1155 * Transfer to our deadlist (which will become next's | |
| 1156 * new deadlist) any entries from next's current | |
| 1157 * deadlist which were born before prev, and free the | |
| 1158 * other entries. | |
| 1159 * | |
| 1160 * XXX we're doing this long task with the config lock held | |
| 1161 */ | |
| 1162 while (bplist_iterate(&ds_next->ds_deadlist, &itor, | |
| 1163 &bp) == 0) { | |
| 1164 if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) { | |
| 1544 | 1165 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, |
| 1166 &bp, tx)); | |
| 789 | 1167 if (ds_prev && !after_branch_point && |
| 1168 bp.blk_birth > | |
| 1169 ds_prev->ds_phys->ds_prev_snap_txg) { | |
| 1170 ds_prev->ds_phys->ds_unique_bytes += | |
| 2082 | 1171 bp_get_dasize(dp->dp_spa, &bp); |
| 789 | 1172 } |
| 1173 } else { | |
| 2082 | 1174 used += bp_get_dasize(dp->dp_spa, &bp); |
| 789 | 1175 compressed += BP_GET_PSIZE(&bp); |
| 1176 uncompressed += BP_GET_UCSIZE(&bp); | |
| 1177 /* XXX check return value? */ | |
| 1178 (void) arc_free(zio, dp->dp_spa, tx->tx_txg, | |
| 1179 &bp, NULL, NULL, ARC_NOWAIT); | |
| 1180 } | |
| 1181 } | |
| 1182 | |
| 1183 /* free next's deadlist */ | |
| 1184 bplist_close(&ds_next->ds_deadlist); | |
| 1185 bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx); | |
| 1186 | |
| 1187 /* set next's deadlist to our deadlist */ | |
| 1188 ds_next->ds_phys->ds_deadlist_obj = | |
| 1189 ds->ds_phys->ds_deadlist_obj; | |
| 1544 | 1190 VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos, |
| 1191 ds_next->ds_phys->ds_deadlist_obj)); | |
| 789 | 1192 ds->ds_phys->ds_deadlist_obj = 0; |
| 1193 | |
| 1194 if (ds_next->ds_phys->ds_next_snap_obj != 0) { | |
| 1195 /* | |
| 1196 * Update next's unique to include blocks which | |
| 1197 * were previously shared by only this snapshot | |
| 1198 * and it. Those blocks will be born after the | |
| 1199 * prev snap and before this snap, and will have | |
| 1200 * died after the next snap and before the one | |
| 1201 * after that (ie. be on the snap after next's | |
| 1202 * deadlist). | |
| 1203 * | |
| 1204 * XXX we're doing this long task with the | |
| 1205 * config lock held | |
| 1206 */ | |
| 1207 dsl_dataset_t *ds_after_next; | |
| 1208 | |
| 2199 | 1209 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 789 | 1210 ds_next->ds_phys->ds_next_snap_obj, NULL, |
| 1544 | 1211 DS_MODE_NONE, FTAG, &ds_after_next)); |
| 789 | 1212 itor = 0; |
| 1213 while (bplist_iterate(&ds_after_next->ds_deadlist, | |
| 1214 &itor, &bp) == 0) { | |
| 1215 if (bp.blk_birth > | |
| 1216 ds->ds_phys->ds_prev_snap_txg && | |
| 1217 bp.blk_birth <= | |
| 1218 ds->ds_phys->ds_creation_txg) { | |
| 1219 ds_next->ds_phys->ds_unique_bytes += | |
| 2082 | 1220 bp_get_dasize(dp->dp_spa, &bp); |
| 789 | 1221 } |
| 1222 } | |
| 1223 | |
| 1224 dsl_dataset_close(ds_after_next, DS_MODE_NONE, FTAG); | |
| 1225 ASSERT3P(ds_next->ds_prev, ==, NULL); | |
| 1226 } else { | |
| 1227 /* | |
| 1228 * It would be nice to update the head dataset's | |
| 1229 * unique. To do so we would have to traverse | |
| 1230 * it for blocks born after ds_prev, which is | |
| 1231 * pretty expensive just to maintain something | |
| 1232 * for debugging purposes. | |
| 1233 */ | |
| 1234 ASSERT3P(ds_next->ds_prev, ==, ds); | |
| 1235 dsl_dataset_close(ds_next->ds_prev, DS_MODE_NONE, | |
| 1236 ds_next); | |
| 1237 if (ds_prev) { | |
| 2199 | 1238 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 1544 | 1239 ds->ds_phys->ds_prev_snap_obj, NULL, |
| 1240 DS_MODE_NONE, ds_next, &ds_next->ds_prev)); | |
| 789 | 1241 } else { |
| 1242 ds_next->ds_prev = NULL; | |
| 1243 } | |
| 1244 } | |
| 1245 dsl_dataset_close(ds_next, DS_MODE_NONE, FTAG); | |
| 1246 | |
| 1247 /* | |
| 1248 * NB: unique_bytes is not accurate for head objsets | |
| 1249 * because we don't update it when we delete the most | |
| 1250 * recent snapshot -- see above comment. | |
| 1251 */ | |
| 1252 ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes); | |
| 1253 } else { | |
| 1254 /* | |
| 1255 * There's no next snapshot, so this is a head dataset. | |
| 1256 * Destroy the deadlist. Unless it's a clone, the | |
| 1257 * deadlist should be empty. (If it's a clone, it's | |
| 1258 * safe to ignore the deadlist contents.) | |
| 1259 */ | |
| 1260 struct killarg ka; | |
| 1261 | |
| 1262 ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist)); | |
| 1263 bplist_close(&ds->ds_deadlist); | |
| 1264 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); | |
| 1265 ds->ds_phys->ds_deadlist_obj = 0; | |
| 1266 | |
| 1267 /* | |
| 1268 * Free everything that we point to (that's born after | |
| 1269 * the previous snapshot, if we are a clone) | |
| 1270 * | |
| 1271 * XXX we're doing this long task with the config lock held | |
| 1272 */ | |
| 1273 ka.usedp = &used; | |
| 1274 ka.compressedp = &compressed; | |
| 1275 ka.uncompressedp = &uncompressed; | |
| 1276 ka.zio = zio; | |
| 1277 ka.tx = tx; | |
| 1278 err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg, | |
| 1279 ADVANCE_POST, kill_blkptr, &ka); | |
| 1280 ASSERT3U(err, ==, 0); | |
| 1281 } | |
| 1282 | |
| 1283 err = zio_wait(zio); | |
| 1284 ASSERT3U(err, ==, 0); | |
| 1285 | |
| 2199 | 1286 dsl_dir_diduse_space(ds->ds_dir, -used, -compressed, -uncompressed, tx); |
| 789 | 1287 |
| 1288 if (ds->ds_phys->ds_snapnames_zapobj) { | |
| 1289 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx); | |
| 1290 ASSERT(err == 0); | |
| 1291 } | |
| 1292 | |
| 2199 | 1293 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) { |
| 789 | 1294 /* Erase the link in the dataset */ |
| 2199 | 1295 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); |
| 1296 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0; | |
| 789 | 1297 /* |
| 1298 * dsl_dir_sync_destroy() called us, they'll destroy | |
| 1299 * the dataset. | |
| 1300 */ | |
| 1301 } else { | |
| 1302 /* remove from snapshot namespace */ | |
| 1303 dsl_dataset_t *ds_head; | |
| 2199 | 1304 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 1305 ds->ds_dir->dd_phys->dd_head_dataset_obj, NULL, | |
| 1544 | 1306 DS_MODE_NONE, FTAG, &ds_head)); |
|
2207
47efcb3433a7
6439370 assertion failures possible in dsl_dataset_destroy_sync()
ahrens
parents:
2199
diff
changeset
|
1307 VERIFY(0 == dsl_dataset_get_snapname(ds)); |
| 789 | 1308 #ifdef ZFS_DEBUG |
| 1309 { | |
| 1310 uint64_t val; | |
| 1311 err = zap_lookup(mos, | |
| 1312 ds_head->ds_phys->ds_snapnames_zapobj, | |
| 2199 | 1313 ds->ds_snapname, 8, 1, &val); |
| 789 | 1314 ASSERT3U(err, ==, 0); |
| 1315 ASSERT3U(val, ==, obj); | |
| 1316 } | |
| 1317 #endif | |
| 1318 err = zap_remove(mos, ds_head->ds_phys->ds_snapnames_zapobj, | |
| 2199 | 1319 ds->ds_snapname, tx); |
| 789 | 1320 ASSERT(err == 0); |
| 1321 dsl_dataset_close(ds_head, DS_MODE_NONE, FTAG); | |
| 1322 } | |
| 1323 | |
| 1324 if (ds_prev && ds->ds_prev != ds_prev) | |
| 1325 dsl_dataset_close(ds_prev, DS_MODE_NONE, FTAG); | |
| 1326 | |
| 3912 | 1327 spa_clear_bootfs(dp->dp_spa, ds->ds_object, tx); |
| 4543 | 1328 spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx, |
| 1329 cr, "dataset = %llu", ds->ds_object); | |
| 1330 | |
| 2199 | 1331 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, tag); |
| 1332 VERIFY(0 == dmu_object_free(mos, obj, tx)); | |
| 3912 | 1333 |
| 2199 | 1334 } |
| 1335 | |
| 1336 /* ARGSUSED */ | |
| 1337 int | |
| 1338 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
| 1339 { | |
| 1340 objset_t *os = arg1; | |
| 1341 dsl_dataset_t *ds = os->os->os_dsl_dataset; | |
| 1342 const char *snapname = arg2; | |
| 1343 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; | |
| 1344 int err; | |
| 1345 uint64_t value; | |
| 789 | 1346 |
| 1347 /* | |
| 2199 | 1348 * We don't allow multiple snapshots of the same txg. If there |
| 1349 * is already one, try again. | |
| 1350 */ | |
| 1351 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) | |
| 1352 return (EAGAIN); | |
| 1353 | |
| 1354 /* | |
| 1355 * Check for conflicting name snapshot name. | |
| 789 | 1356 */ |
| 2199 | 1357 err = zap_lookup(mos, ds->ds_phys->ds_snapnames_zapobj, |
| 1358 snapname, 8, 1, &value); | |
| 1359 if (err == 0) | |
| 1360 return (EEXIST); | |
| 1361 if (err != ENOENT) | |
| 1362 return (err); | |
| 789 | 1363 |
|
3978
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
1364 /* |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
1365 * 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
|
1366 * 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
|
1367 */ |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
1368 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN) |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
1369 return (ENAMETOOLONG); |
|
2dd668007b7a
6533813 recursive snapshotting resulted in a bad stack overflow
mmusante
parents:
3912
diff
changeset
|
1370 |
| 2199 | 1371 ds->ds_trysnap_txg = tx->tx_txg; |
| 789 | 1372 return (0); |
| 1373 } | |
| 1374 | |
| 2199 | 1375 void |
| 4543 | 1376 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) |
| 789 | 1377 { |
| 2199 | 1378 objset_t *os = arg1; |
| 1379 dsl_dataset_t *ds = os->os->os_dsl_dataset; | |
| 1380 const char *snapname = arg2; | |
| 1381 dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
| 789 | 1382 dmu_buf_t *dbuf; |
| 1383 dsl_dataset_phys_t *dsphys; | |
| 2199 | 1384 uint64_t dsobj; |
| 789 | 1385 objset_t *mos = dp->dp_meta_objset; |
| 1386 int err; | |
| 1387 | |
| 1388 spa_scrub_restart(dp->dp_spa, tx->tx_txg); | |
| 2199 | 1389 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); |
| 789 | 1390 |
|
928
36d72fe4da29
6349314 dmu_object_type names incorrect for DSL Directories and DSL Datasets
tabriz
parents:
885
diff
changeset
|
1391 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
|
1392 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); |
| 1544 | 1393 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); |
| 789 | 1394 dmu_buf_will_dirty(dbuf, tx); |
| 1395 dsphys = dbuf->db_data; | |
| 2199 | 1396 dsphys->ds_dir_obj = ds->ds_dir->dd_object; |
| 789 | 1397 dsphys->ds_fsid_guid = unique_create(); |
| 1398 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, | |
| 1399 sizeof (dsphys->ds_guid)); | |
| 1400 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; | |
| 1401 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; | |
| 1402 dsphys->ds_next_snap_obj = ds->ds_object; | |
| 1403 dsphys->ds_num_children = 1; | |
| 1404 dsphys->ds_creation_time = gethrestime_sec(); | |
| 1405 dsphys->ds_creation_txg = tx->tx_txg; | |
| 1406 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; | |
| 1407 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes; | |
| 1408 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; | |
| 1409 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; | |
| 2082 | 1410 dsphys->ds_flags = ds->ds_phys->ds_flags; |
| 789 | 1411 dsphys->ds_bp = ds->ds_phys->ds_bp; |
| 1544 | 1412 dmu_buf_rele(dbuf, FTAG); |
| 789 | 1413 |
| 2199 | 1414 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); |
| 1415 if (ds->ds_prev) { | |
| 1416 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == | |
| 789 | 1417 ds->ds_object || |
| 2199 | 1418 ds->ds_prev->ds_phys->ds_num_children > 1); |
| 1419 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { | |
| 1420 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
| 789 | 1421 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, |
| 2199 | 1422 ds->ds_prev->ds_phys->ds_creation_txg); |
| 1423 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; | |
| 789 | 1424 } |
| 1425 } | |
| 1426 | |
| 1427 bplist_close(&ds->ds_deadlist); | |
| 1428 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 1429 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, dsphys->ds_creation_txg); | |
| 1430 ds->ds_phys->ds_prev_snap_obj = dsobj; | |
| 1431 ds->ds_phys->ds_prev_snap_txg = dsphys->ds_creation_txg; | |
| 1432 ds->ds_phys->ds_unique_bytes = 0; | |
| 1433 ds->ds_phys->ds_deadlist_obj = | |
| 1434 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); | |
| 1544 | 1435 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos, |
| 1436 ds->ds_phys->ds_deadlist_obj)); | |
| 789 | 1437 |
| 1438 dprintf("snap '%s' -> obj %llu\n", snapname, dsobj); | |
| 1439 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, | |
| 1440 snapname, 8, 1, &dsobj, tx); | |
| 1441 ASSERT(err == 0); | |
| 1442 | |
| 1443 if (ds->ds_prev) | |
| 1444 dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds); | |
| 1544 | 1445 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 1446 ds->ds_phys->ds_prev_snap_obj, snapname, | |
| 1447 DS_MODE_NONE, ds, &ds->ds_prev)); | |
| 4543 | 1448 |
| 1449 spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr, | |
|
4603
c7840c367d00
6494569 zfs recv -d pool/<doesn't exist> core dumps for top-level filesystem backups
ahrens
parents:
4577
diff
changeset
|
1450 "dataset = %llu", dsobj); |
| 789 | 1451 } |
| 1452 | |
| 1453 void | |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1454 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) |
| 789 | 1455 { |
| 1456 ASSERT(dmu_tx_is_syncing(tx)); | |
| 1457 ASSERT(ds->ds_user_ptr != NULL); | |
| 1458 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); | |
| 1459 | |
| 4787 | 1460 /* |
| 1461 * in case we had to change ds_fsid_guid when we opened it, | |
| 1462 * sync it out now. | |
| 1463 */ | |
| 1464 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 1465 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; | |
| 1466 | |
| 789 | 1467 dsl_dir_dirty(ds->ds_dir, tx); |
|
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1468 dmu_objset_sync(ds->ds_user_ptr, zio, tx); |
| 789 | 1469 } |
| 1470 | |
| 1471 void | |
| 2885 | 1472 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) |
| 789 | 1473 { |
| 2885 | 1474 dsl_dir_stats(ds->ds_dir, nv); |
| 789 | 1475 |
| 2885 | 1476 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, |
| 1477 ds->ds_phys->ds_creation_time); | |
| 1478 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, | |
| 1479 ds->ds_phys->ds_creation_txg); | |
| 1480 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, | |
| 1481 ds->ds_phys->ds_used_bytes); | |
| 789 | 1482 |
| 1483 if (ds->ds_phys->ds_next_snap_obj) { | |
| 1484 /* | |
| 1485 * This is a snapshot; override the dd's space used with | |
| 2885 | 1486 * our unique space and compression ratio. |
| 789 | 1487 */ |
| 2885 | 1488 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, |
| 1489 ds->ds_phys->ds_unique_bytes); | |
| 1490 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, | |
| 1491 ds->ds_phys->ds_compressed_bytes == 0 ? 100 : | |
| 1492 (ds->ds_phys->ds_uncompressed_bytes * 100 / | |
| 1493 ds->ds_phys->ds_compressed_bytes)); | |
| 789 | 1494 } |
| 1495 } | |
| 1496 | |
| 2885 | 1497 void |
| 1498 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) | |
| 789 | 1499 { |
| 2885 | 1500 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; |
| 1501 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; | |
| 1502 if (ds->ds_phys->ds_next_snap_obj) { | |
| 1503 stat->dds_is_snapshot = B_TRUE; | |
| 1504 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; | |
| 1505 } | |
| 1506 | |
| 1507 /* clone origin is really a dsl_dir thing... */ | |
| 1508 if (ds->ds_dir->dd_phys->dd_clone_parent_obj) { | |
| 1509 dsl_dataset_t *ods; | |
| 1510 | |
| 1511 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); | |
| 1512 VERIFY(0 == dsl_dataset_open_obj(ds->ds_dir->dd_pool, | |
| 1513 ds->ds_dir->dd_phys->dd_clone_parent_obj, | |
| 1514 NULL, DS_MODE_NONE, FTAG, &ods)); | |
| 1515 dsl_dataset_name(ods, stat->dds_clone_of); | |
| 1516 dsl_dataset_close(ods, DS_MODE_NONE, FTAG); | |
| 1517 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); | |
| 1518 } | |
| 1519 } | |
| 1520 | |
| 1521 uint64_t | |
| 1522 dsl_dataset_fsid_guid(dsl_dataset_t *ds) | |
| 1523 { | |
| 4787 | 1524 return (ds->ds_fsid_guid); |
| 2885 | 1525 } |
| 1526 | |
| 1527 void | |
| 1528 dsl_dataset_space(dsl_dataset_t *ds, | |
| 1529 uint64_t *refdbytesp, uint64_t *availbytesp, | |
| 1530 uint64_t *usedobjsp, uint64_t *availobjsp) | |
| 1531 { | |
| 1532 *refdbytesp = ds->ds_phys->ds_used_bytes; | |
| 1533 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); | |
| 1534 *usedobjsp = ds->ds_phys->ds_bp.blk_fill; | |
| 1535 *availobjsp = DN_MAX_OBJECT - *usedobjsp; | |
| 789 | 1536 } |
| 1537 | |
| 2199 | 1538 /* ARGSUSED */ |
| 789 | 1539 static int |
| 2199 | 1540 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) |
| 789 | 1541 { |
| 2199 | 1542 dsl_dataset_t *ds = arg1; |
| 1543 char *newsnapname = arg2; | |
| 1544 dsl_dir_t *dd = ds->ds_dir; | |
| 789 | 1545 objset_t *mos = dd->dd_pool->dp_meta_objset; |
| 2199 | 1546 dsl_dataset_t *hds; |
| 1547 uint64_t val; | |
| 789 | 1548 int err; |
| 1549 | |
| 2199 | 1550 err = dsl_dataset_open_obj(dd->dd_pool, |
| 1551 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds); | |
| 789 | 1552 if (err) |
| 1553 return (err); | |
| 1554 | |
| 2199 | 1555 /* new name better not be in use */ |
| 1556 err = zap_lookup(mos, hds->ds_phys->ds_snapnames_zapobj, | |
| 1557 newsnapname, 8, 1, &val); | |
| 1558 dsl_dataset_close(hds, DS_MODE_NONE, FTAG); | |
| 789 | 1559 |
| 2199 | 1560 if (err == 0) |
| 1561 err = EEXIST; | |
| 1562 else if (err == ENOENT) | |
| 1563 err = 0; | |
| 4007 | 1564 |
| 1565 /* dataset name + 1 for the "@" + the new snapshot name must fit */ | |
| 1566 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN) | |
| 1567 err = ENAMETOOLONG; | |
| 1568 | |
| 2199 | 1569 return (err); |
| 1570 } | |
| 789 | 1571 |
| 2199 | 1572 static void |
| 4543 | 1573 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, |
| 1574 cred_t *cr, dmu_tx_t *tx) | |
| 2199 | 1575 { |
| 1576 dsl_dataset_t *ds = arg1; | |
| 4543 | 1577 const char *newsnapname = arg2; |
| 2199 | 1578 dsl_dir_t *dd = ds->ds_dir; |
| 1579 objset_t *mos = dd->dd_pool->dp_meta_objset; | |
| 1580 dsl_dataset_t *hds; | |
| 1581 int err; | |
| 789 | 1582 |
| 2199 | 1583 ASSERT(ds->ds_phys->ds_next_snap_obj != 0); |
| 789 | 1584 |
| 2199 | 1585 VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool, |
| 1586 dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds)); | |
| 789 | 1587 |
| 2199 | 1588 VERIFY(0 == dsl_dataset_get_snapname(ds)); |
| 1589 err = zap_remove(mos, hds->ds_phys->ds_snapnames_zapobj, | |
| 1590 ds->ds_snapname, tx); | |
| 789 | 1591 ASSERT3U(err, ==, 0); |
| 2199 | 1592 mutex_enter(&ds->ds_lock); |
| 1593 (void) strcpy(ds->ds_snapname, newsnapname); | |
| 1594 mutex_exit(&ds->ds_lock); | |
| 1595 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj, | |
| 1596 ds->ds_snapname, 8, 1, &ds->ds_object, tx); | |
| 789 | 1597 ASSERT3U(err, ==, 0); |
| 1598 | |
| 4543 | 1599 spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx, |
| 1600 cr, "dataset = %llu", ds->ds_object); | |
| 2199 | 1601 dsl_dataset_close(hds, DS_MODE_NONE, FTAG); |
| 789 | 1602 } |
| 1603 | |
| 4007 | 1604 struct renamearg { |
| 1605 dsl_sync_task_group_t *dstg; | |
| 1606 char failed[MAXPATHLEN]; | |
| 1607 char *oldsnap; | |
| 1608 char *newsnap; | |
| 1609 }; | |
| 1610 | |
| 1611 static int | |
| 1612 dsl_snapshot_rename_one(char *name, void *arg) | |
| 1613 { | |
| 1614 struct renamearg *ra = arg; | |
| 1615 dsl_dataset_t *ds = NULL; | |
| 1616 char *cp; | |
| 1617 int err; | |
| 1618 | |
| 1619 cp = name + strlen(name); | |
| 1620 *cp = '@'; | |
| 1621 (void) strcpy(cp + 1, ra->oldsnap); | |
| 4543 | 1622 |
| 1623 /* | |
| 1624 * For recursive snapshot renames the parent won't be changing | |
| 1625 * so we just pass name for both the to/from argument. | |
| 1626 */ | |
| 1627 if (err = zfs_secpolicy_rename_perms(name, name, CRED())) { | |
| 1628 (void) strcpy(ra->failed, name); | |
| 1629 return (err); | |
| 1630 } | |
| 1631 | |
| 4007 | 1632 err = dsl_dataset_open(name, DS_MODE_READONLY | DS_MODE_STANDARD, |
| 1633 ra->dstg, &ds); | |
| 1634 if (err == ENOENT) { | |
| 1635 *cp = '\0'; | |
| 1636 return (0); | |
| 1637 } | |
| 1638 if (err) { | |
| 1639 (void) strcpy(ra->failed, name); | |
| 1640 *cp = '\0'; | |
| 1641 dsl_dataset_close(ds, DS_MODE_STANDARD, ra->dstg); | |
| 1642 return (err); | |
| 1643 } | |
| 1644 | |
| 1645 #ifdef _KERNEL | |
| 1646 /* for all filesystems undergoing rename, we'll need to unmount it */ | |
| 1647 (void) zfs_unmount_snap(name, NULL); | |
| 1648 #endif | |
| 1649 | |
| 1650 *cp = '\0'; | |
| 1651 | |
| 1652 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check, | |
| 1653 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0); | |
| 1654 | |
| 1655 return (0); | |
| 1656 } | |
| 1657 | |
| 1658 static int | |
| 1659 dsl_recursive_rename(char *oldname, const char *newname) | |
| 1660 { | |
| 1661 int err; | |
| 1662 struct renamearg *ra; | |
| 1663 dsl_sync_task_t *dst; | |
| 1664 spa_t *spa; | |
| 1665 char *cp, *fsname = spa_strdup(oldname); | |
| 1666 int len = strlen(oldname); | |
| 1667 | |
| 1668 /* truncate the snapshot name to get the fsname */ | |
| 1669 cp = strchr(fsname, '@'); | |
| 1670 *cp = '\0'; | |
| 1671 | |
|
4603
c7840c367d00
6494569 zfs recv -d pool/<doesn't exist> core dumps for top-level filesystem backups
ahrens
parents:
4577
diff
changeset
|
1672 err = spa_open(fsname, &spa, FTAG); |
| 4007 | 1673 if (err) { |
| 1674 kmem_free(fsname, len + 1); | |
| 1675 return (err); | |
| 1676 } | |
| 1677 ra = kmem_alloc(sizeof (struct renamearg), KM_SLEEP); | |
| 1678 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); | |
| 1679 | |
| 1680 ra->oldsnap = strchr(oldname, '@') + 1; | |
| 1681 ra->newsnap = strchr(newname, '@') + 1; | |
| 1682 *ra->failed = '\0'; | |
| 1683 | |
| 1684 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra, | |
| 1685 DS_FIND_CHILDREN); | |
| 1686 kmem_free(fsname, len + 1); | |
| 1687 | |
| 1688 if (err == 0) { | |
| 1689 err = dsl_sync_task_group_wait(ra->dstg); | |
| 1690 } | |
| 1691 | |
| 1692 for (dst = list_head(&ra->dstg->dstg_tasks); dst; | |
| 1693 dst = list_next(&ra->dstg->dstg_tasks, dst)) { | |
| 1694 dsl_dataset_t *ds = dst->dst_arg1; | |
| 1695 if (dst->dst_err) { | |
| 1696 dsl_dir_name(ds->ds_dir, ra->failed); | |
|
4009
1a9b4fbc0d2a
6479884 want 'zfs rename -r' to recursively rename snapshots (fix lint)
mmusante
parents:
4007
diff
changeset
|
1697 (void) strcat(ra->failed, "@"); |
|
1a9b4fbc0d2a
6479884 want 'zfs rename -r' to recursively rename snapshots (fix lint)
mmusante
parents:
4007
diff
changeset
|
1698 (void) strcat(ra->failed, ra->newsnap); |
| 4007 | 1699 } |
| 1700 dsl_dataset_close(ds, DS_MODE_STANDARD, ra->dstg); | |
| 1701 } | |
| 1702 | |
| 4543 | 1703 if (err) |
| 1704 (void) strcpy(oldname, ra->failed); | |
| 4007 | 1705 |
| 1706 dsl_sync_task_group_destroy(ra->dstg); | |
| 1707 kmem_free(ra, sizeof (struct renamearg)); | |
| 1708 spa_close(spa, FTAG); | |
| 1709 return (err); | |
| 1710 } | |
| 1711 | |
|
4569
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1712 static int |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1713 dsl_valid_rename(char *oldname, void *arg) |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1714 { |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1715 int delta = *(int *)arg; |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1716 |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1717 if (strlen(oldname) + delta >= MAXNAMELEN) |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1718 return (ENAMETOOLONG); |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1719 |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1720 return (0); |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1721 } |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1722 |
| 789 | 1723 #pragma weak dmu_objset_rename = dsl_dataset_rename |
| 1724 int | |
| 4007 | 1725 dsl_dataset_rename(char *oldname, const char *newname, |
| 1726 boolean_t recursive) | |
| 789 | 1727 { |
| 1728 dsl_dir_t *dd; | |
| 2199 | 1729 dsl_dataset_t *ds; |
| 789 | 1730 const char *tail; |
| 1731 int err; | |
| 1732 | |
| 2199 | 1733 err = dsl_dir_open(oldname, FTAG, &dd, &tail); |
| 1544 | 1734 if (err) |
| 1735 return (err); | |
| 789 | 1736 if (tail == NULL) { |
|
4569
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1737 int delta = strlen(newname) - strlen(oldname); |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1738 |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1739 /* if we're growing, validate child size lengths */ |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1740 if (delta > 0) |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1741 err = dmu_objset_find(oldname, dsl_valid_rename, |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1742 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1743 |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1744 if (!err) |
|
e80d40447f27
6355623 zfs rename to a valid dataset name, but its snapshot name becomes too long, panics the system
mmusante
parents:
4543
diff
changeset
|
1745 err = dsl_dir_rename(dd, newname); |
| 789 | 1746 dsl_dir_close(dd, FTAG); |
| 1747 return (err); | |
| 1748 } | |
| 1749 if (tail[0] != '@') { | |
| 1750 /* the name ended in a nonexistant component */ | |
| 1751 dsl_dir_close(dd, FTAG); | |
| 1752 return (ENOENT); | |
| 1753 } | |
| 1754 | |
| 2199 | 1755 dsl_dir_close(dd, FTAG); |
| 1756 | |
| 1757 /* new name must be snapshot in same filesystem */ | |
| 1758 tail = strchr(newname, '@'); | |
| 1759 if (tail == NULL) | |
| 1760 return (EINVAL); | |
| 1761 tail++; | |
| 1762 if (strncmp(oldname, newname, tail - newname) != 0) | |
| 1763 return (EXDEV); | |
| 789 | 1764 |
| 4007 | 1765 if (recursive) { |
| 1766 err = dsl_recursive_rename(oldname, newname); | |
| 1767 } else { | |
| 1768 err = dsl_dataset_open(oldname, | |
| 1769 DS_MODE_READONLY | DS_MODE_STANDARD, FTAG, &ds); | |
| 1770 if (err) | |
| 1771 return (err); | |
| 2199 | 1772 |
| 4007 | 1773 err = dsl_sync_task_do(ds->ds_dir->dd_pool, |
| 1774 dsl_dataset_snapshot_rename_check, | |
| 1775 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1); | |
| 2199 | 1776 |
| 4007 | 1777 dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG); |
| 1778 } | |
| 2199 | 1779 |
| 789 | 1780 return (err); |
| 1781 } | |
| 2082 | 1782 |
| 2199 | 1783 struct promotearg { |
| 1784 uint64_t used, comp, uncomp, unique; | |
| 1785 uint64_t newnext_obj, snapnames_obj; | |
| 1786 }; | |
| 1787 | |
| 4543 | 1788 /* ARGSUSED */ |
| 2082 | 1789 static int |
| 2199 | 1790 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) |
| 2082 | 1791 { |
| 2199 | 1792 dsl_dataset_t *hds = arg1; |
| 1793 struct promotearg *pa = arg2; | |
| 1794 dsl_dir_t *dd = hds->ds_dir; | |
| 1795 dsl_pool_t *dp = hds->ds_dir->dd_pool; | |
| 2082 | 1796 dsl_dir_t *pdd = NULL; |
| 1797 dsl_dataset_t *ds = NULL; | |
| 1798 dsl_dataset_t *pivot_ds = NULL; | |
| 1799 dsl_dataset_t *newnext_ds = NULL; | |
| 1800 int err; | |
| 1801 char *name = NULL; | |
| 2199 | 1802 uint64_t itor = 0; |
| 2082 | 1803 blkptr_t bp; |
| 1804 | |
| 2199 | 1805 bzero(pa, sizeof (*pa)); |
| 1806 | |
| 2082 | 1807 /* Check that it is a clone */ |
| 1808 if (dd->dd_phys->dd_clone_parent_obj == 0) | |
| 1809 return (EINVAL); | |
| 1810 | |
| 2199 | 1811 /* Since this is so expensive, don't do the preliminary check */ |
| 1812 if (!dmu_tx_is_syncing(tx)) | |
| 1813 return (0); | |
| 1814 | |
| 1815 if (err = dsl_dataset_open_obj(dp, | |
| 2082 | 1816 dd->dd_phys->dd_clone_parent_obj, |
| 1817 NULL, DS_MODE_EXCLUSIVE, FTAG, &pivot_ds)) | |
| 1818 goto out; | |
| 1819 pdd = pivot_ds->ds_dir; | |
| 2199 | 1820 |
| 1821 { | |
| 1822 dsl_dataset_t *phds; | |
| 1823 if (err = dsl_dataset_open_obj(dd->dd_pool, | |
| 1824 pdd->dd_phys->dd_head_dataset_obj, | |
| 1825 NULL, DS_MODE_NONE, FTAG, &phds)) | |
| 1826 goto out; | |
| 1827 pa->snapnames_obj = phds->ds_phys->ds_snapnames_zapobj; | |
| 1828 dsl_dataset_close(phds, DS_MODE_NONE, FTAG); | |
| 1829 } | |
| 2082 | 1830 |
| 1831 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) { | |
| 1832 err = EXDEV; | |
| 1833 goto out; | |
| 1834 } | |
| 1835 | |
| 1836 /* find pivot point's new next ds */ | |
| 1837 VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool, hds->ds_object, | |
| 1838 NULL, DS_MODE_NONE, FTAG, &newnext_ds)); | |
| 1839 while (newnext_ds->ds_phys->ds_prev_snap_obj != pivot_ds->ds_object) { | |
| 1840 dsl_dataset_t *prev; | |
| 1841 | |
| 1842 if (err = dsl_dataset_open_obj(dd->dd_pool, | |
| 2199 | 1843 newnext_ds->ds_phys->ds_prev_snap_obj, |
| 1844 NULL, DS_MODE_NONE, FTAG, &prev)) | |
| 2082 | 1845 goto out; |
| 1846 dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG); | |
| 1847 newnext_ds = prev; | |
| 1848 } | |
| 2199 | 1849 pa->newnext_obj = newnext_ds->ds_object; |
| 2082 | 1850 |
| 1851 /* compute pivot point's new unique space */ | |
| 1852 while ((err = bplist_iterate(&newnext_ds->ds_deadlist, | |
| 1853 &itor, &bp)) == 0) { | |
| 1854 if (bp.blk_birth > pivot_ds->ds_phys->ds_prev_snap_txg) | |
| 2199 | 1855 pa->unique += bp_get_dasize(dd->dd_pool->dp_spa, &bp); |
| 2082 | 1856 } |
| 1857 if (err != ENOENT) | |
| 1858 goto out; | |
| 1859 | |
| 1860 /* Walk the snapshots that we are moving */ | |
| 1861 name = kmem_alloc(MAXPATHLEN, KM_SLEEP); | |
| 1862 ds = pivot_ds; | |
| 1863 /* CONSTCOND */ | |
| 1864 while (TRUE) { | |
| 1865 uint64_t val, dlused, dlcomp, dluncomp; | |
| 1866 dsl_dataset_t *prev; | |
| 1867 | |
| 1868 /* Check that the snapshot name does not conflict */ | |
| 1869 dsl_dataset_name(ds, name); | |
| 1870 err = zap_lookup(dd->dd_pool->dp_meta_objset, | |
| 1871 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, | |
| 1872 8, 1, &val); | |
| 1873 if (err != ENOENT) { | |
| 1874 if (err == 0) | |
| 1875 err = EEXIST; | |
| 1876 goto out; | |
| 1877 } | |
| 1878 | |
| 1879 /* | |
| 1880 * compute space to transfer. Each snapshot gave birth to: | |
| 1881 * (my used) - (prev's used) + (deadlist's used) | |
| 1882 */ | |
| 2199 | 1883 pa->used += ds->ds_phys->ds_used_bytes; |
| 1884 pa->comp += ds->ds_phys->ds_compressed_bytes; | |
| 1885 pa->uncomp += ds->ds_phys->ds_uncompressed_bytes; | |
| 2082 | 1886 |
| 1887 /* If we reach the first snapshot, we're done. */ | |
| 1888 if (ds->ds_phys->ds_prev_snap_obj == 0) | |
| 1889 break; | |
| 1890 | |
| 1891 if (err = bplist_space(&ds->ds_deadlist, | |
| 1892 &dlused, &dlcomp, &dluncomp)) | |
| 1893 goto out; | |
| 1894 if (err = dsl_dataset_open_obj(dd->dd_pool, | |
| 1895 ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE, | |
| 1896 FTAG, &prev)) | |
| 1897 goto out; | |
| 2199 | 1898 pa->used += dlused - prev->ds_phys->ds_used_bytes; |
| 1899 pa->comp += dlcomp - prev->ds_phys->ds_compressed_bytes; | |
| 1900 pa->uncomp += dluncomp - prev->ds_phys->ds_uncompressed_bytes; | |
| 2082 | 1901 |
| 1902 /* | |
| 1903 * We could be a clone of a clone. If we reach our | |
| 1904 * parent's branch point, we're done. | |
| 1905 */ | |
| 1906 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) { | |
| 1907 dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG); | |
| 1908 break; | |
| 1909 } | |
| 1910 if (ds != pivot_ds) | |
| 1911 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 1912 ds = prev; | |
| 1913 } | |
| 1914 | |
| 1915 /* Check that there is enough space here */ | |
| 2199 | 1916 err = dsl_dir_transfer_possible(pdd, dd, pa->used); |
| 1917 | |
| 1918 out: | |
| 1919 if (ds && ds != pivot_ds) | |
| 1920 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 1921 if (pivot_ds) | |
| 1922 dsl_dataset_close(pivot_ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 1923 if (newnext_ds) | |
| 1924 dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG); | |
| 1925 if (name) | |
| 1926 kmem_free(name, MAXPATHLEN); | |
| 1927 return (err); | |
| 1928 } | |
| 2082 | 1929 |
| 2199 | 1930 static void |
| 4543 | 1931 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) |
| 2199 | 1932 { |
| 1933 dsl_dataset_t *hds = arg1; | |
| 1934 struct promotearg *pa = arg2; | |
| 1935 dsl_dir_t *dd = hds->ds_dir; | |
| 1936 dsl_pool_t *dp = hds->ds_dir->dd_pool; | |
| 1937 dsl_dir_t *pdd = NULL; | |
| 1938 dsl_dataset_t *ds, *pivot_ds; | |
| 1939 char *name; | |
| 1940 | |
| 1941 ASSERT(dd->dd_phys->dd_clone_parent_obj != 0); | |
| 1942 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)); | |
| 1943 | |
| 1944 VERIFY(0 == dsl_dataset_open_obj(dp, | |
| 1945 dd->dd_phys->dd_clone_parent_obj, | |
| 1946 NULL, DS_MODE_EXCLUSIVE, FTAG, &pivot_ds)); | |
| 2417 | 1947 /* |
| 1948 * We need to explicitly open pdd, since pivot_ds's pdd will be | |
| 1949 * changing. | |
| 1950 */ | |
| 1951 VERIFY(0 == dsl_dir_open_obj(dp, pivot_ds->ds_dir->dd_object, | |
| 1952 NULL, FTAG, &pdd)); | |
| 2082 | 1953 |
| 1954 /* move snapshots to this dir */ | |
| 2199 | 1955 name = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
| 2082 | 1956 ds = pivot_ds; |
| 1957 /* CONSTCOND */ | |
| 1958 while (TRUE) { | |
| 1959 dsl_dataset_t *prev; | |
| 1960 | |
| 1961 /* move snap name entry */ | |
| 1962 dsl_dataset_name(ds, name); | |
| 2199 | 1963 VERIFY(0 == zap_remove(dp->dp_meta_objset, |
| 1964 pa->snapnames_obj, ds->ds_snapname, tx)); | |
| 1965 VERIFY(0 == zap_add(dp->dp_meta_objset, | |
| 2082 | 1966 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, |
| 1967 8, 1, &ds->ds_object, tx)); | |
| 1968 | |
| 1969 /* change containing dsl_dir */ | |
| 1970 dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
| 1971 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, pdd->dd_object); | |
| 1972 ds->ds_phys->ds_dir_obj = dd->dd_object; | |
| 1973 ASSERT3P(ds->ds_dir, ==, pdd); | |
| 1974 dsl_dir_close(ds->ds_dir, ds); | |
| 2199 | 1975 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object, |
| 2082 | 1976 NULL, ds, &ds->ds_dir)); |
| 1977 | |
| 1978 ASSERT3U(dsl_prop_numcb(ds), ==, 0); | |
| 1979 | |
| 1980 if (ds->ds_phys->ds_prev_snap_obj == 0) | |
| 1981 break; | |
| 1982 | |
| 2199 | 1983 VERIFY(0 == dsl_dataset_open_obj(dp, |
| 2082 | 1984 ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE, |
| 1985 FTAG, &prev)); | |
| 1986 | |
| 1987 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) { | |
| 1988 dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG); | |
| 1989 break; | |
| 1990 } | |
| 1991 if (ds != pivot_ds) | |
| 1992 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 1993 ds = prev; | |
| 1994 } | |
| 2199 | 1995 if (ds != pivot_ds) |
| 1996 dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG); | |
| 2082 | 1997 |
| 1998 /* change pivot point's next snap */ | |
| 1999 dmu_buf_will_dirty(pivot_ds->ds_dbuf, tx); | |
| 2199 | 2000 pivot_ds->ds_phys->ds_next_snap_obj = pa->newnext_obj; |
| 2082 | 2001 |
| 2002 /* change clone_parent-age */ | |
| 2003 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
| 2004 ASSERT3U(dd->dd_phys->dd_clone_parent_obj, ==, pivot_ds->ds_object); | |
| 2005 dd->dd_phys->dd_clone_parent_obj = pdd->dd_phys->dd_clone_parent_obj; | |
| 2006 dmu_buf_will_dirty(pdd->dd_dbuf, tx); | |
| 2007 pdd->dd_phys->dd_clone_parent_obj = pivot_ds->ds_object; | |
| 2008 | |
| 2009 /* change space accounting */ | |
| 2199 | 2010 dsl_dir_diduse_space(pdd, -pa->used, -pa->comp, -pa->uncomp, tx); |
| 2011 dsl_dir_diduse_space(dd, pa->used, pa->comp, pa->uncomp, tx); | |
| 2012 pivot_ds->ds_phys->ds_unique_bytes = pa->unique; | |
| 2082 | 2013 |
| 4543 | 2014 /* log history record */ |
| 2015 spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx, | |
| 2016 cr, "dataset = %llu", ds->ds_object); | |
| 2017 | |
| 2417 | 2018 dsl_dir_close(pdd, FTAG); |
| 2199 | 2019 dsl_dataset_close(pivot_ds, DS_MODE_EXCLUSIVE, FTAG); |
| 2020 kmem_free(name, MAXPATHLEN); | |
| 2082 | 2021 } |
| 2022 | |
| 2023 int | |
| 2024 dsl_dataset_promote(const char *name) | |
| 2025 { | |
| 2026 dsl_dataset_t *ds; | |
| 2027 int err; | |
| 2028 dmu_object_info_t doi; | |
| 2199 | 2029 struct promotearg pa; |
| 2082 | 2030 |
| 2031 err = dsl_dataset_open(name, DS_MODE_NONE, FTAG, &ds); | |
| 2032 if (err) | |
| 2033 return (err); | |
| 2034 | |
| 2035 err = dmu_object_info(ds->ds_dir->dd_pool->dp_meta_objset, | |
| 2036 ds->ds_phys->ds_snapnames_zapobj, &doi); | |
| 2037 if (err) { | |
| 2038 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 2039 return (err); | |
| 2040 } | |
| 2041 | |
| 2042 /* | |
| 2043 * Add in 128x the snapnames zapobj size, since we will be moving | |
| 2044 * a bunch of snapnames to the promoted ds, and dirtying their | |
| 2045 * bonus buffers. | |
| 2046 */ | |
| 2199 | 2047 err = dsl_sync_task_do(ds->ds_dir->dd_pool, |
| 2048 dsl_dataset_promote_check, | |
| 2049 dsl_dataset_promote_sync, ds, &pa, 2 + 2 * doi.doi_physical_blks); | |
| 2082 | 2050 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); |
| 2051 return (err); | |
| 2052 } | |
| 3912 | 2053 |
| 2054 /* | |
| 2055 * Given a pool name and a dataset object number in that pool, | |
| 2056 * return the name of that dataset. | |
| 2057 */ | |
| 2058 int | |
| 2059 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) | |
| 2060 { | |
| 2061 spa_t *spa; | |
| 2062 dsl_pool_t *dp; | |
| 2063 dsl_dataset_t *ds = NULL; | |
| 2064 int error; | |
| 2065 | |
| 2066 if ((error = spa_open(pname, &spa, FTAG)) != 0) | |
| 2067 return (error); | |
| 2068 dp = spa_get_dsl(spa); | |
| 2069 rw_enter(&dp->dp_config_rwlock, RW_READER); | |
| 2070 if ((error = dsl_dataset_open_obj(dp, obj, | |
| 2071 NULL, DS_MODE_NONE, FTAG, &ds)) != 0) { | |
| 2072 rw_exit(&dp->dp_config_rwlock); | |
| 2073 spa_close(spa, FTAG); | |
| 2074 return (error); | |
| 2075 } | |
| 2076 dsl_dataset_name(ds, buf); | |
| 2077 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); | |
| 2078 rw_exit(&dp->dp_config_rwlock); | |
| 2079 spa_close(spa, FTAG); | |
| 2080 | |
| 2081 return (0); | |
| 2082 } |