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, 2015 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_EMBEDOK "com.delphix:resume_embedok"
100 
101 /*
102  * DS_FLAG_CI_DATASET is set if the dataset contains a file system whose
103  * name lookups should be performed case-insensitively.
104  */
105 #define   DS_FLAG_CI_DATASET  (1ULL<<16)
106 
107 #define   DS_CREATE_FLAG_NODIRTY        (1ULL<<24)
108 
109 typedef struct dsl_dataset_phys {
110           uint64_t ds_dir_obj;                    /* DMU_OT_DSL_DIR */
111           uint64_t ds_prev_snap_obj;    /* DMU_OT_DSL_DATASET */
112           uint64_t ds_prev_snap_txg;
113           uint64_t ds_next_snap_obj;    /* DMU_OT_DSL_DATASET */
114           uint64_t ds_snapnames_zapobj; /* DMU_OT_DSL_DS_SNAP_MAP 0 for snaps */
115           uint64_t ds_num_children;     /* clone/snap children; ==0 for head */
116           uint64_t ds_creation_time;    /* seconds since 1970 */
117           uint64_t ds_creation_txg;
118           uint64_t ds_deadlist_obj;     /* DMU_OT_DEADLIST */
119           /*
120            * ds_referenced_bytes, ds_compressed_bytes, and ds_uncompressed_bytes
121            * include all blocks referenced by this dataset, including those
122            * shared with any other datasets.
123            */
124           uint64_t ds_referenced_bytes;
125           uint64_t ds_compressed_bytes;
126           uint64_t ds_uncompressed_bytes;
127           uint64_t ds_unique_bytes;     /* only relevant to snapshots */
128           /*
129            * The ds_fsid_guid is a 56-bit ID that can change to avoid
130            * collisions.  The ds_guid is a 64-bit ID that will never
131            * change, so there is a small probability that it will collide.
132            */
133           uint64_t ds_fsid_guid;
134           uint64_t ds_guid;
135           uint64_t ds_flags;            /* DS_FLAG_* */
136           blkptr_t ds_bp;
137           uint64_t ds_next_clones_obj;  /* DMU_OT_DSL_CLONES */
138           uint64_t ds_props_obj;                  /* DMU_OT_DSL_PROPS for snaps */
139           uint64_t ds_userrefs_obj;     /* DMU_OT_USERREFS */
140           uint64_t ds_pad[5]; /* pad out to 320 bytes for good measure */
141 } dsl_dataset_phys_t;
142 
143 typedef struct dsl_dataset {
144           dmu_buf_user_t ds_dbu;
145           rrwlock_t ds_bp_rwlock; /* Protects ds_phys->ds_bp */
146 
147           /* Immutable: */
148           struct dsl_dir *ds_dir;
149           dmu_buf_t *ds_dbuf;
150           uint64_t ds_object;
151           uint64_t ds_fsid_guid;
152           boolean_t ds_is_snapshot;
153 
154           /* only used in syncing context, only valid for non-snapshots: */
155           struct dsl_dataset *ds_prev;
156           uint64_t ds_bookmarks;  /* DMU_OTN_ZAP_METADATA */
157 
158           /* has internal locking: */
159           dsl_deadlist_t ds_deadlist;
160           bplist_t ds_pending_deadlist;
161 
162           /* protected by lock on pool's dp_dirty_datasets list */
163           txg_node_t ds_dirty_link;
164           list_node_t ds_synced_link;
165 
166           /*
167            * ds_phys->ds_<accounting> is also protected by ds_lock.
168            * Protected by ds_lock:
169            */
170           kmutex_t ds_lock;
171           objset_t *ds_objset;
172           uint64_t ds_userrefs;
173           void *ds_owner;
174 
175           /*
176            * Long holds prevent the ds from being destroyed; they allow the
177            * ds to remain held even after dropping the dp_config_rwlock.
178            * Owning counts as a long hold.  See the comments above
179            * dsl_pool_hold() for details.
180            */
181           refcount_t ds_longholds;
182 
183           /* no locking; only for making guesses */
184           uint64_t ds_trysnap_txg;
185 
186           /* for objset_open() */
187           kmutex_t ds_opening_lock;
188 
189           uint64_t ds_reserved;         /* cached refreservation */
190           uint64_t ds_quota;  /* cached refquota */
191 
192           kmutex_t ds_sendstream_lock;
193           list_t ds_sendstreams;
194 
195           /*
196            * When in the middle of a resumable receive, tracks how much
197            * progress we have made.
198            */
199           uint64_t ds_resume_object[TXG_SIZE];
200           uint64_t ds_resume_offset[TXG_SIZE];
201           uint64_t ds_resume_bytes[TXG_SIZE];
202 
203           /* Protected by our dsl_dir's dd_lock */
204           list_t ds_prop_cbs;
205 
206           /*
207            * For ZFEATURE_FLAG_PER_DATASET features, set if this dataset
208            * uses this feature.
209            */
210           uint8_t ds_feature_inuse[SPA_FEATURES];
211 
212           /*
213            * Set if we need to activate the feature on this dataset this txg
214            * (used only in syncing context).
215            */
216           uint8_t ds_feature_activation_needed[SPA_FEATURES];
217 
218           /* Protected by ds_lock; keep at end of struct for better locality */
219           char ds_snapname[ZFS_MAX_DATASET_NAME_LEN];
220 } dsl_dataset_t;
221 
222 inline dsl_dataset_phys_t *
dsl_dataset_phys(dsl_dataset_t * ds)223 dsl_dataset_phys(dsl_dataset_t *ds)
224 {
225           return (ds->ds_dbuf->db_data);
226 }
227 
228 /*
229  * The max length of a temporary tag prefix is the number of hex digits
230  * required to express UINT64_MAX plus one for the hyphen.
231  */
232 #define   MAX_TAG_PREFIX_LEN  17
233 
234 #define   dsl_dataset_is_snapshot(ds) \
235           (dsl_dataset_phys(ds)->ds_num_children != 0)
236 
237 #define   DS_UNIQUE_IS_ACCURATE(ds)     \
238           ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_UNIQUE_ACCURATE) != 0)
239 
240 int dsl_dataset_hold(struct dsl_pool *dp, const char *name, void *tag,
241     dsl_dataset_t **dsp);
242 boolean_t dsl_dataset_try_add_ref(struct dsl_pool *dp, dsl_dataset_t *ds,
243     void *tag);
244 int dsl_dataset_hold_obj(struct dsl_pool *dp, uint64_t dsobj, void *tag,
245     dsl_dataset_t **);
246 void dsl_dataset_rele(dsl_dataset_t *ds, void *tag);
247 int dsl_dataset_own(struct dsl_pool *dp, const char *name,
248     void *tag, dsl_dataset_t **dsp);
249 int dsl_dataset_own_obj(struct dsl_pool *dp, uint64_t dsobj,
250     void *tag, dsl_dataset_t **dsp);
251 void dsl_dataset_disown(dsl_dataset_t *ds, void *tag);
252 void dsl_dataset_name(dsl_dataset_t *ds, char *name);
253 boolean_t dsl_dataset_tryown(dsl_dataset_t *ds, void *tag);
254 int dsl_dataset_namelen(dsl_dataset_t *ds);
255 boolean_t dsl_dataset_has_owner(dsl_dataset_t *ds);
256 uint64_t dsl_dataset_create_sync(dsl_dir_t *pds, const char *lastname,
257     dsl_dataset_t *origin, uint64_t flags, cred_t *, dmu_tx_t *);
258 uint64_t dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
259     uint64_t flags, dmu_tx_t *tx);
260 int dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors);
261 int dsl_dataset_promote(const char *name, char *conflsnap);
262 int dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
263     boolean_t force);
264 int dsl_dataset_rename_snapshot(const char *fsname,
265     const char *oldsnapname, const char *newsnapname, boolean_t recursive);
266 int dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
267     minor_t cleanup_minor, const char *htag);
268 
269 blkptr_t *dsl_dataset_get_blkptr(dsl_dataset_t *ds);
270 
271 spa_t *dsl_dataset_get_spa(dsl_dataset_t *ds);
272 
273 boolean_t dsl_dataset_modified_since_snap(dsl_dataset_t *ds,
274     dsl_dataset_t *snap);
275 
276 void dsl_dataset_sync(dsl_dataset_t *os, zio_t *zio, dmu_tx_t *tx);
277 void dsl_dataset_sync_done(dsl_dataset_t *os, dmu_tx_t *tx);
278 
279 void dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp,
280     dmu_tx_t *tx);
281 int dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp,
282     dmu_tx_t *tx, boolean_t async);
283 boolean_t dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
284     uint64_t blk_birth);
285 uint64_t dsl_dataset_prev_snap_txg(dsl_dataset_t *ds);
286 
287 void dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx);
288 void dsl_dataset_stats(dsl_dataset_t *os, nvlist_t *nv);
289 void dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat);
290 void dsl_dataset_space(dsl_dataset_t *ds,
291     uint64_t *refdbytesp, uint64_t *availbytesp,
292     uint64_t *usedobjsp, uint64_t *availobjsp);
293 uint64_t dsl_dataset_fsid_guid(dsl_dataset_t *ds);
294 int dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
295     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp);
296 int dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, dsl_dataset_t *last,
297     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp);
298 boolean_t dsl_dataset_is_dirty(dsl_dataset_t *ds);
299 
300 int dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf);
301 
302 int dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
303     uint64_t asize, uint64_t inflight, uint64_t *used,
304     uint64_t *ref_rsrv);
305 int dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
306     uint64_t quota);
307 int dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
308     uint64_t reservation);
309 
310 boolean_t dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
311     uint64_t earlier_txg);
312 void dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag);
313 void dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag);
314 boolean_t dsl_dataset_long_held(dsl_dataset_t *ds);
315 
316 int dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
317     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx);
318 void dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
319     dsl_dataset_t *origin_head, dmu_tx_t *tx);
320 int dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
321     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr);
322 void dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
323     dmu_tx_t *tx);
324 
325 void dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
326     dmu_tx_t *tx);
327 void dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds);
328 int dsl_dataset_get_snapname(dsl_dataset_t *ds);
329 int dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name,
330     uint64_t *value);
331 int dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
332     boolean_t adj_cnt);
333 void dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
334     zprop_source_t source, uint64_t value, dmu_tx_t *tx);
335 void dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx);
336 boolean_t dsl_dataset_is_zapified(dsl_dataset_t *ds);
337 boolean_t dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds);
338 int dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result);
339 
340 void dsl_dataset_deactivate_feature(uint64_t dsobj,
341     spa_feature_t f, dmu_tx_t *tx);
342 
343 #ifdef ZFS_DEBUG
344 #define   dprintf_ds(ds, fmt, ...) do { \
345           if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
346           char *__ds_name = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); \
347           dsl_dataset_name(ds, __ds_name); \
348           dprintf("ds=%s " fmt, __ds_name, __VA_ARGS__); \
349           kmem_free(__ds_name, ZFS_MAX_DATASET_NAME_LEN); \
350           } \
351 _NOTE(CONSTCOND) } while (0)
352 #else
353 #define   dprintf_ds(dd, fmt, ...)
354 #endif
355 
356 #ifdef    __cplusplus
357 }
358 #endif
359 
360 #endif /* _SYS_DSL_DATASET_H */
361