1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2011, 2017 by Delphix. All rights reserved. 24 * Copyright (c) 2013, Joyent, Inc. All rights reserved. 25 * Copyright (c) 2013 Steven Hartland. All rights reserved. 26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 27 * Copyright (c) 2014 Integros [integros.com] 28 */ 29 30 #ifndef _SYS_DSL_DATASET_H 31 #define _SYS_DSL_DATASET_H 32 33 #include <sys/dmu.h> 34 #include <sys/spa.h> 35 #include <sys/txg.h> 36 #include <sys/zio.h> 37 #include <sys/bplist.h> 38 #include <sys/dsl_synctask.h> 39 #include <sys/zfs_context.h> 40 #include <sys/dsl_deadlist.h> 41 #include <sys/refcount.h> 42 #include <sys/rrwlock.h> 43 #include <zfeature_common.h> 44 45 #ifdef __cplusplus 46 extern "C" { 47 #endif 48 49 struct dsl_dataset; 50 struct dsl_dir; 51 struct dsl_pool; 52 53 #define DS_FLAG_INCONSISTENT (1ULL<<0) 54 #define DS_IS_INCONSISTENT(ds) \ 55 (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) 56 57 /* 58 * Do not allow this dataset to be promoted. 59 */ 60 #define DS_FLAG_NOPROMOTE (1ULL<<1) 61 62 /* 63 * DS_FLAG_UNIQUE_ACCURATE is set if ds_unique_bytes has been correctly 64 * calculated for head datasets (starting with SPA_VERSION_UNIQUE_ACCURATE, 65 * refquota/refreservations). 66 */ 67 #define DS_FLAG_UNIQUE_ACCURATE (1ULL<<2) 68 69 /* 70 * DS_FLAG_DEFER_DESTROY is set after 'zfs destroy -d' has been called 71 * on a dataset. This allows the dataset to be destroyed using 'zfs release'. 72 */ 73 #define DS_FLAG_DEFER_DESTROY (1ULL<<3) 74 #define DS_IS_DEFER_DESTROY(ds) \ 75 (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_DEFER_DESTROY) 76 77 /* 78 * DS_FIELD_* are strings that are used in the "extensified" dataset zap object. 79 * They should be of the format <reverse-dns>:<field>. 80 */ 81 82 /* 83 * This field's value is the object ID of a zap object which contains the 84 * bookmarks of this dataset. If it is present, then this dataset is counted 85 * in the refcount of the SPA_FEATURES_BOOKMARKS feature. 86 */ 87 #define DS_FIELD_BOOKMARK_NAMES "com.delphix:bookmarks" 88 89 /* 90 * These fields are set on datasets that are in the middle of a resumable 91 * receive, and allow the sender to resume the send if it is interrupted. 92 */ 93 #define DS_FIELD_RESUME_FROMGUID "com.delphix:resume_fromguid" 94 #define DS_FIELD_RESUME_TONAME "com.delphix:resume_toname" 95 #define DS_FIELD_RESUME_TOGUID "com.delphix:resume_toguid" 96 #define DS_FIELD_RESUME_OBJECT "com.delphix:resume_object" 97 #define DS_FIELD_RESUME_OFFSET "com.delphix:resume_offset" 98 #define DS_FIELD_RESUME_BYTES "com.delphix:resume_bytes" 99 #define DS_FIELD_RESUME_LARGEBLOCK "com.delphix:resume_largeblockok" 100 #define DS_FIELD_RESUME_EMBEDOK "com.delphix:resume_embedok" 101 #define DS_FIELD_RESUME_COMPRESSOK "com.delphix:resume_compressok" 102 103 /* 104 * This field is set to the object number of the remap deadlist if one exists. 105 */ 106 #define DS_FIELD_REMAP_DEADLIST "com.delphix:remap_deadlist" 107 108 /* 109 * DS_FLAG_CI_DATASET is set if the dataset contains a file system whose 110 * name lookups should be performed case-insensitively. 111 */ 112 #define DS_FLAG_CI_DATASET (1ULL<<16) 113 114 #define DS_CREATE_FLAG_NODIRTY (1ULL<<24) 115 116 typedef struct dsl_dataset_phys { 117 uint64_t ds_dir_obj; /* DMU_OT_DSL_DIR */ 118 uint64_t ds_prev_snap_obj; /* DMU_OT_DSL_DATASET */ 119 uint64_t ds_prev_snap_txg; 120 uint64_t ds_next_snap_obj; /* DMU_OT_DSL_DATASET */ 121 uint64_t ds_snapnames_zapobj; /* DMU_OT_DSL_DS_SNAP_MAP 0 for snaps */ 122 uint64_t ds_num_children; /* clone/snap children; ==0 for head */ 123 uint64_t ds_creation_time; /* seconds since 1970 */ 124 uint64_t ds_creation_txg; 125 uint64_t ds_deadlist_obj; /* DMU_OT_DEADLIST */ 126 /* 127 * ds_referenced_bytes, ds_compressed_bytes, and ds_uncompressed_bytes 128 * include all blocks referenced by this dataset, including those 129 * shared with any other datasets. 130 */ 131 uint64_t ds_referenced_bytes; 132 uint64_t ds_compressed_bytes; 133 uint64_t ds_uncompressed_bytes; 134 uint64_t ds_unique_bytes; /* only relevant to snapshots */ 135 /* 136 * The ds_fsid_guid is a 56-bit ID that can change to avoid 137 * collisions. The ds_guid is a 64-bit ID that will never 138 * change, so there is a small probability that it will collide. 139 */ 140 uint64_t ds_fsid_guid; 141 uint64_t ds_guid; 142 uint64_t ds_flags; /* DS_FLAG_* */ 143 blkptr_t ds_bp; 144 uint64_t ds_next_clones_obj; /* DMU_OT_DSL_CLONES */ 145 uint64_t ds_props_obj; /* DMU_OT_DSL_PROPS for snaps */ 146 uint64_t ds_userrefs_obj; /* DMU_OT_USERREFS */ 147 uint64_t ds_pad[5]; /* pad out to 320 bytes for good measure */ 148 } dsl_dataset_phys_t; 149 150 typedef struct dsl_dataset { 151 dmu_buf_user_t ds_dbu; 152 rrwlock_t ds_bp_rwlock; /* Protects ds_phys->ds_bp */ 153 154 /* Immutable: */ 155 struct dsl_dir *ds_dir; 156 dmu_buf_t *ds_dbuf; 157 uint64_t ds_object; 158 uint64_t ds_fsid_guid; 159 boolean_t ds_is_snapshot; 160 161 /* only used in syncing context, only valid for non-snapshots: */ 162 struct dsl_dataset *ds_prev; 163 uint64_t ds_bookmarks; /* DMU_OTN_ZAP_METADATA */ 164 165 /* has internal locking: */ 166 dsl_deadlist_t ds_deadlist; 167 bplist_t ds_pending_deadlist; 168 169 /* 170 * The remap deadlist contains blocks (DVA's, really) that are 171 * referenced by the previous snapshot and point to indirect vdevs, 172 * but in this dataset they have been remapped to point to concrete 173 * (or at least, less-indirect) vdevs. In other words, the 174 * physical DVA is referenced by the previous snapshot but not by 175 * this dataset. Logically, the DVA continues to be referenced, 176 * but we are using a different (less indirect) physical DVA. 177 * This deadlist is used to determine when physical DVAs that 178 * point to indirect vdevs are no longer referenced anywhere, 179 * and thus should be marked obsolete. 180 * 181 * This is only used if SPA_FEATURE_OBSOLETE_COUNTS is enabled. 182 */ 183 dsl_deadlist_t ds_remap_deadlist; 184 /* protects creation of the ds_remap_deadlist */ 185 kmutex_t ds_remap_deadlist_lock; 186 187 /* protected by lock on pool's dp_dirty_datasets list */ 188 txg_node_t ds_dirty_link; 189 list_node_t ds_synced_link; 190 191 /* 192 * ds_phys->ds_<accounting> is also protected by ds_lock. 193 * Protected by ds_lock: 194 */ 195 kmutex_t ds_lock; 196 objset_t *ds_objset; 197 uint64_t ds_userrefs; 198 void *ds_owner; 199 200 /* 201 * Long holds prevent the ds from being destroyed; they allow the 202 * ds to remain held even after dropping the dp_config_rwlock. 203 * Owning counts as a long hold. See the comments above 204 * dsl_pool_hold() for details. 205 */ 206 refcount_t ds_longholds; 207 208 /* no locking; only for making guesses */ 209 uint64_t ds_trysnap_txg; 210 211 /* for objset_open() */ 212 kmutex_t ds_opening_lock; 213 214 uint64_t ds_reserved; /* cached refreservation */ 215 uint64_t ds_quota; /* cached refquota */ 216 217 kmutex_t ds_sendstream_lock; 218 list_t ds_sendstreams; 219 220 /* 221 * When in the middle of a resumable receive, tracks how much 222 * progress we have made. 223 */ 224 uint64_t ds_resume_object[TXG_SIZE]; 225 uint64_t ds_resume_offset[TXG_SIZE]; 226 uint64_t ds_resume_bytes[TXG_SIZE]; 227 228 /* Protected by our dsl_dir's dd_lock */ 229 list_t ds_prop_cbs; 230 231 /* 232 * For ZFEATURE_FLAG_PER_DATASET features, set if this dataset 233 * uses this feature. 234 */ 235 uint8_t ds_feature_inuse[SPA_FEATURES]; 236 237 /* 238 * Set if we need to activate the feature on this dataset this txg 239 * (used only in syncing context). 240 */ 241 uint8_t ds_feature_activation_needed[SPA_FEATURES]; 242 243 /* Protected by ds_lock; keep at end of struct for better locality */ 244 char ds_snapname[ZFS_MAX_DATASET_NAME_LEN]; 245 } dsl_dataset_t; 246 247 inline dsl_dataset_phys_t * 248 dsl_dataset_phys(dsl_dataset_t *ds) 249 { 250 return (ds->ds_dbuf->db_data); 251 } 252 253 typedef struct dsl_dataset_promote_arg { 254 const char *ddpa_clonename; 255 dsl_dataset_t *ddpa_clone; 256 list_t shared_snaps, origin_snaps, clone_snaps; 257 dsl_dataset_t *origin_origin; /* origin of the origin */ 258 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 259 nvlist_t *err_ds; 260 cred_t *cr; 261 } dsl_dataset_promote_arg_t; 262 263 typedef struct dsl_dataset_rollback_arg { 264 const char *ddra_fsname; 265 const char *ddra_tosnap; 266 void *ddra_owner; 267 nvlist_t *ddra_result; 268 } dsl_dataset_rollback_arg_t; 269 270 typedef struct dsl_dataset_snapshot_arg { 271 nvlist_t *ddsa_snaps; 272 nvlist_t *ddsa_props; 273 nvlist_t *ddsa_errors; 274 cred_t *ddsa_cr; 275 } dsl_dataset_snapshot_arg_t; 276 277 /* 278 * The max length of a temporary tag prefix is the number of hex digits 279 * required to express UINT64_MAX plus one for the hyphen. 280 */ 281 #define MAX_TAG_PREFIX_LEN 17 282 283 #define dsl_dataset_is_snapshot(ds) \ 284 (dsl_dataset_phys(ds)->ds_num_children != 0) 285 286 #define DS_UNIQUE_IS_ACCURATE(ds) \ 287 ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_UNIQUE_ACCURATE) != 0) 288 289 int dsl_dataset_hold(struct dsl_pool *dp, const char *name, void *tag, 290 dsl_dataset_t **dsp); 291 boolean_t dsl_dataset_try_add_ref(struct dsl_pool *dp, dsl_dataset_t *ds, 292 void *tag); 293 int dsl_dataset_hold_obj(struct dsl_pool *dp, uint64_t dsobj, void *tag, 294 dsl_dataset_t **); 295 void dsl_dataset_rele(dsl_dataset_t *ds, void *tag); 296 int dsl_dataset_own(struct dsl_pool *dp, const char *name, 297 void *tag, dsl_dataset_t **dsp); 298 int dsl_dataset_own_obj(struct dsl_pool *dp, uint64_t dsobj, 299 void *tag, dsl_dataset_t **dsp); 300 void dsl_dataset_disown(dsl_dataset_t *ds, void *tag); 301 void dsl_dataset_name(dsl_dataset_t *ds, char *name); 302 boolean_t dsl_dataset_tryown(dsl_dataset_t *ds, void *tag); 303 int dsl_dataset_namelen(dsl_dataset_t *ds); 304 boolean_t dsl_dataset_has_owner(dsl_dataset_t *ds); 305 uint64_t dsl_dataset_create_sync(dsl_dir_t *pds, const char *lastname, 306 dsl_dataset_t *origin, uint64_t flags, cred_t *, dmu_tx_t *); 307 uint64_t dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 308 uint64_t flags, dmu_tx_t *tx); 309 void dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx); 310 int dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx); 311 int dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors); 312 void dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx); 313 int dsl_dataset_promote_check(void *arg, dmu_tx_t *tx); 314 int dsl_dataset_promote(const char *name, char *conflsnap); 315 int dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head, 316 boolean_t force); 317 int dsl_dataset_rename_snapshot(const char *fsname, 318 const char *oldsnapname, const char *newsnapname, boolean_t recursive); 319 int dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 320 minor_t cleanup_minor, const char *htag); 321 322 blkptr_t *dsl_dataset_get_blkptr(dsl_dataset_t *ds); 323 324 spa_t *dsl_dataset_get_spa(dsl_dataset_t *ds); 325 326 boolean_t dsl_dataset_modified_since_snap(dsl_dataset_t *ds, 327 dsl_dataset_t *snap); 328 329 void dsl_dataset_sync(dsl_dataset_t *os, zio_t *zio, dmu_tx_t *tx); 330 void dsl_dataset_sync_done(dsl_dataset_t *os, dmu_tx_t *tx); 331 332 void dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, 333 dmu_tx_t *tx); 334 int dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, 335 dmu_tx_t *tx, boolean_t async); 336 void dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, 337 uint64_t offset, uint64_t size, uint64_t birth, dmu_tx_t *tx); 338 339 void dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx); 340 341 int get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val); 342 char *get_receive_resume_stats_impl(dsl_dataset_t *ds); 343 char *get_child_receive_stats(dsl_dataset_t *ds); 344 uint64_t dsl_get_refratio(dsl_dataset_t *ds); 345 uint64_t dsl_get_logicalreferenced(dsl_dataset_t *ds); 346 uint64_t dsl_get_compressratio(dsl_dataset_t *ds); 347 uint64_t dsl_get_used(dsl_dataset_t *ds); 348 uint64_t dsl_get_creation(dsl_dataset_t *ds); 349 uint64_t dsl_get_creationtxg(dsl_dataset_t *ds); 350 uint64_t dsl_get_refquota(dsl_dataset_t *ds); 351 uint64_t dsl_get_refreservation(dsl_dataset_t *ds); 352 uint64_t dsl_get_guid(dsl_dataset_t *ds); 353 uint64_t dsl_get_unique(dsl_dataset_t *ds); 354 uint64_t dsl_get_objsetid(dsl_dataset_t *ds); 355 uint64_t dsl_get_userrefs(dsl_dataset_t *ds); 356 uint64_t dsl_get_defer_destroy(dsl_dataset_t *ds); 357 uint64_t dsl_get_referenced(dsl_dataset_t *ds); 358 uint64_t dsl_get_numclones(dsl_dataset_t *ds); 359 uint64_t dsl_get_inconsistent(dsl_dataset_t *ds); 360 uint64_t dsl_get_available(dsl_dataset_t *ds); 361 int dsl_get_written(dsl_dataset_t *ds, uint64_t *written); 362 int dsl_get_prev_snap(dsl_dataset_t *ds, char *snap); 363 int dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value, 364 char *source); 365 366 void get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv); 367 368 void dsl_dataset_stats(dsl_dataset_t *os, nvlist_t *nv); 369 370 void dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat); 371 void dsl_dataset_space(dsl_dataset_t *ds, 372 uint64_t *refdbytesp, uint64_t *availbytesp, 373 uint64_t *usedobjsp, uint64_t *availobjsp); 374 uint64_t dsl_dataset_fsid_guid(dsl_dataset_t *ds); 375 int dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 376 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp); 377 int dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, dsl_dataset_t *last, 378 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp); 379 boolean_t dsl_dataset_is_dirty(dsl_dataset_t *ds); 380 381 int dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf); 382 383 int dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 384 uint64_t asize, uint64_t inflight, uint64_t *used, 385 uint64_t *ref_rsrv); 386 int dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 387 uint64_t quota); 388 int dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 389 uint64_t reservation); 390 391 boolean_t dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 392 uint64_t earlier_txg); 393 void dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag); 394 void dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag); 395 boolean_t dsl_dataset_long_held(dsl_dataset_t *ds); 396 397 int dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 398 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx); 399 void dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 400 dsl_dataset_t *origin_head, dmu_tx_t *tx); 401 int dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 402 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr); 403 void dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 404 dmu_tx_t *tx); 405 406 void dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 407 dmu_tx_t *tx); 408 void dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds); 409 int dsl_dataset_get_snapname(dsl_dataset_t *ds); 410 int dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, 411 uint64_t *value); 412 int dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 413 boolean_t adj_cnt); 414 void dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 415 zprop_source_t source, uint64_t value, dmu_tx_t *tx); 416 void dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx); 417 boolean_t dsl_dataset_is_zapified(dsl_dataset_t *ds); 418 boolean_t dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds); 419 420 int dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx); 421 void dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx); 422 int dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner, 423 nvlist_t *result); 424 425 uint64_t dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds); 426 void dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx); 427 boolean_t dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds); 428 void dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx); 429 430 void dsl_dataset_deactivate_feature(uint64_t dsobj, 431 spa_feature_t f, dmu_tx_t *tx); 432 433 #ifdef ZFS_DEBUG 434 #define dprintf_ds(ds, fmt, ...) do { \ 435 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 436 char *__ds_name = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); \ 437 dsl_dataset_name(ds, __ds_name); \ 438 dprintf("ds=%s " fmt, __ds_name, __VA_ARGS__); \ 439 kmem_free(__ds_name, ZFS_MAX_DATASET_NAME_LEN); \ 440 } \ 441 _NOTE(CONSTCOND) } while (0) 442 #else 443 #define dprintf_ds(dd, fmt, ...) 444 #endif 445 446 #ifdef __cplusplus 447 } 448 #endif 449 450 #endif /* _SYS_DSL_DATASET_H */ 451