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) 2013 by Delphix. All rights reserved.
  24  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  25  */
  26 
  27 #ifndef _SYS_DBUF_H
  28 #define _SYS_DBUF_H
  29 
  30 #include <sys/dmu.h>
  31 #include <sys/spa.h>
  32 #include <sys/txg.h>
  33 #include <sys/zio.h>
  34 #include <sys/arc.h>
  35 #include <sys/zfs_context.h>
  36 #include <sys/refcount.h>
  37 #include <sys/zrlock.h>
  38 
  39 #ifdef  __cplusplus
  40 extern "C" {
  41 #endif
  42 
  43 #define IN_DMU_SYNC 2
  44 
  45 /*
  46  * define flags for dbuf_read
  47  */
  48 
  49 #define DB_RF_MUST_SUCCEED      (1 << 0)
  50 #define DB_RF_CANFAIL           (1 << 1)
  51 #define DB_RF_HAVESTRUCT        (1 << 2)
  52 #define DB_RF_NOPREFETCH        (1 << 3)
  53 #define DB_RF_NEVERWAIT         (1 << 4)
  54 #define DB_RF_CACHED            (1 << 5)
  55 
  56 /*
  57  * The simplified state transition diagram for dbufs looks like:
  58  *
  59  *              +----> READ ----+
  60  *              |               |
  61  *              |               V
  62  *  (alloc)-->UNCACHED            CACHED-->EVICTING-->(free)
  63  *              |               ^        ^
  64  *              |               |        |
  65  *              +----> FILL ----+     |
  66  *              |                        |
  67  *              |                        |
  68  *              +--------> NOFILL -------+
  69  */
  70 typedef enum dbuf_states {
  71         DB_UNCACHED,
  72         DB_FILL,
  73         DB_NOFILL,
  74         DB_READ,
  75         DB_CACHED,
  76         DB_EVICTING
  77 } dbuf_states_t;
  78 
  79 struct dnode;
  80 struct dmu_tx;
  81 
  82 /*
  83  * level = 0 means the user data
  84  * level = 1 means the single indirect block
  85  * etc.
  86  */
  87 
  88 struct dmu_buf_impl;
  89 
  90 typedef enum override_states {
  91         DR_NOT_OVERRIDDEN,
  92         DR_IN_DMU_SYNC,
  93         DR_OVERRIDDEN
  94 } override_states_t;
  95 
  96 typedef struct dbuf_dirty_record {
  97         /* link on our parents dirty list */
  98         list_node_t dr_dirty_node;
  99 
 100         /* transaction group this data will sync in */
 101         uint64_t dr_txg;
 102 
 103         /* zio of outstanding write IO */
 104         zio_t *dr_zio;
 105 
 106         /* pointer back to our dbuf */
 107         struct dmu_buf_impl *dr_dbuf;
 108 
 109         /* pointer to next dirty record */
 110         struct dbuf_dirty_record *dr_next;
 111 
 112         /* pointer to parent dirty record */
 113         struct dbuf_dirty_record *dr_parent;
 114 
 115         /* How much space was changed to dsl_pool_dirty_space() for this? */
 116         unsigned int dr_accounted;
 117 
 118         union dirty_types {
 119                 struct dirty_indirect {
 120 
 121                         /* protect access to list */
 122                         kmutex_t dr_mtx;
 123 
 124                         /* Our list of dirty children */
 125                         list_t dr_children;
 126                 } di;
 127                 struct dirty_leaf {
 128 
 129                         /*
 130                          * dr_data is set when we dirty the buffer
 131                          * so that we can retain the pointer even if it
 132                          * gets COW'd in a subsequent transaction group.
 133                          */
 134                         arc_buf_t *dr_data;
 135                         blkptr_t dr_overridden_by;
 136                         override_states_t dr_override_state;
 137                         uint8_t dr_copies;
 138                         boolean_t dr_nopwrite;
 139                 } dl;
 140         } dt;
 141 } dbuf_dirty_record_t;
 142 
 143 typedef struct dmu_buf_impl {
 144         /*
 145          * The following members are immutable, with the exception of
 146          * db.db_data, which is protected by db_mtx.
 147          */
 148 
 149         /* the publicly visible structure */
 150         dmu_buf_t db;
 151 
 152         /* the objset we belong to */
 153         struct objset *db_objset;
 154 
 155         /*
 156          * handle to safely access the dnode we belong to (NULL when evicted)
 157          */
 158         struct dnode_handle *db_dnode_handle;
 159 
 160         /*
 161          * our parent buffer; if the dnode points to us directly,
 162          * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
 163          * only accessed by sync thread ???
 164          * (NULL when evicted)
 165          * May change from NULL to non-NULL under the protection of db_mtx
 166          * (see dbuf_check_blkptr())
 167          */
 168         struct dmu_buf_impl *db_parent;
 169 
 170         /*
 171          * link for hash table of all dmu_buf_impl_t's
 172          */
 173         struct dmu_buf_impl *db_hash_next;
 174 
 175         /* our block number */
 176         uint64_t db_blkid;
 177 
 178         /*
 179          * Pointer to the blkptr_t which points to us. May be NULL if we
 180          * don't have one yet. (NULL when evicted)
 181          */
 182         blkptr_t *db_blkptr;
 183 
 184         /*
 185          * Our indirection level.  Data buffers have db_level==0.
 186          * Indirect buffers which point to data buffers have
 187          * db_level==1. etc.  Buffers which contain dnodes have
 188          * db_level==0, since the dnodes are stored in a file.
 189          */
 190         uint8_t db_level;
 191 
 192         /* db_mtx protects the members below */
 193         kmutex_t db_mtx;
 194 
 195         /*
 196          * Current state of the buffer
 197          */
 198         dbuf_states_t db_state;
 199 
 200         /*
 201          * Refcount accessed by dmu_buf_{hold,rele}.
 202          * If nonzero, the buffer can't be destroyed.
 203          * Protected by db_mtx.
 204          */
 205         refcount_t db_holds;
 206 
 207         /* buffer holding our data */
 208         arc_buf_t *db_buf;
 209 
 210         kcondvar_t db_changed;
 211         dbuf_dirty_record_t *db_data_pending;
 212 
 213         /* pointer to most recent dirty record for this buffer */
 214         dbuf_dirty_record_t *db_last_dirty;
 215 
 216         /*
 217          * Our link on the owner dnodes's dn_dbufs list.
 218          * Protected by its dn_dbufs_mtx.
 219          */
 220         list_node_t db_link;
 221 
 222         /* Data which is unique to data (leaf) blocks: */
 223 
 224         /* stuff we store for the user (see dmu_buf_set_user) */
 225         void *db_user_ptr;
 226         void **db_user_data_ptr_ptr;
 227         dmu_buf_evict_func_t *db_evict_func;
 228 
 229         uint8_t db_immediate_evict;
 230         uint8_t db_freed_in_flight;
 231 
 232         uint8_t db_dirtycnt;
 233 } dmu_buf_impl_t;
 234 
 235 /* Note: the dbuf hash table is exposed only for the mdb module */
 236 #define DBUF_MUTEXES 256
 237 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
 238 typedef struct dbuf_hash_table {
 239         uint64_t hash_table_mask;
 240         dmu_buf_impl_t **hash_table;
 241         kmutex_t hash_mutexes[DBUF_MUTEXES];
 242 } dbuf_hash_table_t;
 243 
 244 
 245 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset);
 246 
 247 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
 248 void dbuf_create_bonus(struct dnode *dn);
 249 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
 250 void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag);
 251 
 252 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
 253 
 254 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
 255 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
 256     void *tag);
 257 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create,
 258     void *tag, dmu_buf_impl_t **dbp);
 259 
 260 void dbuf_prefetch(struct dnode *dn, uint64_t blkid, zio_priority_t prio);
 261 
 262 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
 263 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
 264 
 265 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
 266 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
 267 
 268 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid);
 269 
 270 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
 271 void dbuf_will_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
 272 void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx);
 273 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
 274 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
 275 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
 276 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
 277 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
 278 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
 279 
 280 void dbuf_clear(dmu_buf_impl_t *db);
 281 void dbuf_evict(dmu_buf_impl_t *db);
 282 
 283 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
 284 void dbuf_unoverride(dbuf_dirty_record_t *dr);
 285 void dbuf_sync_list(list_t *list, dmu_tx_t *tx);
 286 void dbuf_release_bp(dmu_buf_impl_t *db);
 287 
 288 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
 289     struct dmu_tx *);
 290 
 291 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
 292 
 293 #define DB_DNODE(_db)           ((_db)->db_dnode_handle->dnh_dnode)
 294 #define DB_DNODE_LOCK(_db)      ((_db)->db_dnode_handle->dnh_zrlock)
 295 #define DB_DNODE_ENTER(_db)     (zrl_add(&DB_DNODE_LOCK(_db)))
 296 #define DB_DNODE_EXIT(_db)      (zrl_remove(&DB_DNODE_LOCK(_db)))
 297 #define DB_DNODE_HELD(_db)      (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
 298 #define DB_GET_SPA(_spa_p, _db) {               \
 299         dnode_t *__dn;                          \
 300         DB_DNODE_ENTER(_db);                    \
 301         __dn = DB_DNODE(_db);                   \
 302         *(_spa_p) = __dn->dn_objset->os_spa;      \
 303         DB_DNODE_EXIT(_db);                     \
 304 }
 305 #define DB_GET_OBJSET(_os_p, _db) {             \
 306         dnode_t *__dn;                          \
 307         DB_DNODE_ENTER(_db);                    \
 308         __dn = DB_DNODE(_db);                   \
 309         *(_os_p) = __dn->dn_objset;          \
 310         DB_DNODE_EXIT(_db);                     \
 311 }
 312 
 313 void dbuf_init(void);
 314 void dbuf_fini(void);
 315 
 316 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
 317 
 318 #define DBUF_GET_BUFC_TYPE(_db) \
 319         (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
 320 
 321 #define DBUF_IS_CACHEABLE(_db)                                          \
 322         ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL ||           \
 323         (dbuf_is_metadata(_db) &&                                       \
 324         ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
 325 
 326 #define DBUF_IS_L2CACHEABLE(_db)                                        \
 327         ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \
 328         (dbuf_is_metadata(_db) &&                                       \
 329         ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
 330 
 331 #define DBUF_IS_L2COMPRESSIBLE(_db)                                     \
 332         ((_db)->db_objset->os_compress != ZIO_COMPRESS_OFF)
 333 
 334 #ifdef ZFS_DEBUG
 335 
 336 /*
 337  * There should be a ## between the string literal and fmt, to make it
 338  * clear that we're joining two strings together, but gcc does not
 339  * support that preprocessor token.
 340  */
 341 #define dprintf_dbuf(dbuf, fmt, ...) do { \
 342         if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
 343         char __db_buf[32]; \
 344         uint64_t __db_obj = (dbuf)->db.db_object; \
 345         if (__db_obj == DMU_META_DNODE_OBJECT) \
 346                 (void) strcpy(__db_buf, "mdn"); \
 347         else \
 348                 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
 349                     (u_longlong_t)__db_obj); \
 350         dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
 351             "obj=%s lvl=%u blkid=%lld " fmt, \
 352             __db_buf, (dbuf)->db_level, \
 353             (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
 354         } \
 355 _NOTE(CONSTCOND) } while (0)
 356 
 357 #define dprintf_dbuf_bp(db, bp, fmt, ...) do {                  \
 358         if (zfs_flags & ZFS_DEBUG_DPRINTF) {                        \
 359         char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP);  \
 360         sprintf_blkptr(__blkbuf, bp);                           \
 361         dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf);   \
 362         kmem_free(__blkbuf, BP_SPRINTF_LEN);                    \
 363         }                                                       \
 364 _NOTE(CONSTCOND) } while (0)
 365 
 366 #define DBUF_VERIFY(db) dbuf_verify(db)
 367 
 368 #else
 369 
 370 #define dprintf_dbuf(db, fmt, ...)
 371 #define dprintf_dbuf_bp(db, bp, fmt, ...)
 372 #define DBUF_VERIFY(db)
 373 
 374 #endif
 375 
 376 
 377 #ifdef  __cplusplus
 378 }
 379 #endif
 380 
 381 #endif /* _SYS_DBUF_H */