include/sys/dnode.h

   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) 2012, 2017 by Delphix. All rights reserved.
  24  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  25  */
  26
  27 #ifndef _SYS_DNODE_H
  28 #define _SYS_DNODE_H
  29
  30 #include <sys/zfs_context.h>
  31 #include <sys/avl.h>
  32 #include <sys/spa.h>
  33 #include <sys/txg.h>
  34 #include <sys/zio.h>
  35 #include <sys/refcount.h>
  36 #include <sys/dmu_zfetch.h>
  37 #include <sys/zrlock.h>
  38 #include <sys/multilist.h>
  39
  40 #ifdef  __cplusplus
  41 extern "C" {
  42 #endif
  43
  44 /*
  45  * dnode_hold() flags.
  46  */
  47 #define DNODE_MUST_BE_ALLOCATED 1
  48 #define DNODE_MUST_BE_FREE      2
  49
  50 /*
  51  * dnode_next_offset() flags.
  52  */
  53 #define DNODE_FIND_HOLE         1
  54 #define DNODE_FIND_BACKWARDS    2
  55 #define DNODE_FIND_HAVELOCK     4
  56
  57 /*
  58  * Fixed constants.
  59  */
  60 #define DNODE_SHIFT             9       /* 512 bytes */
  61 #define DN_MIN_INDBLKSHIFT      12      /* 4k */
  62 /*
  63  * If we ever increase this value beyond 20, we need to revisit all logic that
  64  * does x << level * ebps to handle overflow.  With a 1M indirect block size,
  65  * 4 levels of indirect blocks would not be able to guarantee addressing an
  66  * entire object, so 5 levels will be used, but 5 * (20 - 7) = 65.
  67  */
  68 #define DN_MAX_INDBLKSHIFT      17      /* 128k */
  69 #define DNODE_BLOCK_SHIFT       14      /* 16k */
  70 #define DNODE_CORE_SIZE         64      /* 64 bytes for dnode sans blkptrs */
  71 #define DN_MAX_OBJECT_SHIFT     48      /* 256 trillion (zfs_fid_t limit) */
  72 #define DN_MAX_OFFSET_SHIFT     64      /* 2^64 bytes in a dnode */
  73
  74 /*
  75  * dnode id flags
  76  *
  77  * Note: a file will never ever have its
  78  * ids moved from bonus->spill
  79  * and only in a crypto environment would it be on spill
  80  */
  81 #define DN_ID_CHKED_BONUS       0x1
  82 #define DN_ID_CHKED_SPILL       0x2
  83 #define DN_ID_OLD_EXIST         0x4
  84 #define DN_ID_NEW_EXIST         0x8
  85
  86 /*
  87  * Derived constants.
  88  */
  89 #define DNODE_MIN_SIZE          (1 << DNODE_SHIFT)
  90 #define DNODE_MAX_SIZE          (1 << DNODE_BLOCK_SHIFT)
  91 #define DNODE_BLOCK_SIZE        (1 << DNODE_BLOCK_SHIFT)
  92 #define DNODE_MIN_SLOTS         (DNODE_MIN_SIZE >> DNODE_SHIFT)
  93 #define DNODE_MAX_SLOTS         (DNODE_MAX_SIZE >> DNODE_SHIFT)
  94 #define DN_BONUS_SIZE(dnsize)   ((dnsize) - DNODE_CORE_SIZE - \
  95         (1 << SPA_BLKPTRSHIFT))
  96 #define DN_SLOTS_TO_BONUSLEN(slots)     DN_BONUS_SIZE((slots) << DNODE_SHIFT)
  97 #define DN_OLD_MAX_BONUSLEN     (DN_BONUS_SIZE(DNODE_MIN_SIZE))
  98 #define DN_MAX_NBLKPTR  ((DNODE_MIN_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
  99 #define DN_MAX_OBJECT   (1ULL << DN_MAX_OBJECT_SHIFT)
 100 #define DN_ZERO_BONUSLEN        (DN_BONUS_SIZE(DNODE_MAX_SIZE) + 1)
 101 #define DN_KILL_SPILLBLK (1)
 102
 103 #define DNODES_PER_BLOCK_SHIFT  (DNODE_BLOCK_SHIFT - DNODE_SHIFT)
 104 #define DNODES_PER_BLOCK        (1ULL << DNODES_PER_BLOCK_SHIFT)
 105
 106 /*
 107  * This is inaccurate if the indblkshift of the particular object is not the
 108  * max.  But it's only used by userland to calculate the zvol reservation.
 109  */
 110 #define DNODES_PER_LEVEL_SHIFT  (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT)
 111 #define DNODES_PER_LEVEL        (1ULL << DNODES_PER_LEVEL_SHIFT)
 112
 113 #define DN_MAX_LEVELS   (DIV_ROUND_UP(DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT, \
 114         DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT) + 1)
 115
 116 #define DN_BONUS(dnp)   ((void*)((dnp)->dn_bonus + \
 117         (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t))))
 118
 119 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \
 120         (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT)
 121
 122 #define EPB(blkshift, typeshift)        (1 << (blkshift - typeshift))
 123
 124 struct dmu_buf_impl;
 125 struct objset;
 126 struct zio;
 127
 128 enum dnode_dirtycontext {
 129         DN_UNDIRTIED,
 130         DN_DIRTY_OPEN,
 131         DN_DIRTY_SYNC
 132 };
 133
 134 /* Is dn_used in bytes?  if not, it's in multiples of SPA_MINBLOCKSIZE */
 135 #define DNODE_FLAG_USED_BYTES                   (1 << 0)
 136 #define DNODE_FLAG_USERUSED_ACCOUNTED           (1 << 1)
 137
 138 /* Does dnode have a SA spill blkptr in bonus? */
 139 #define DNODE_FLAG_SPILL_BLKPTR                 (1 << 2)
 140
 141 /* User/Group dnode accounting */
 142 #define DNODE_FLAG_USEROBJUSED_ACCOUNTED        (1 << 3)
 143
 144 typedef struct dnode_phys {
 145         uint8_t dn_type;                /* dmu_object_type_t */
 146         uint8_t dn_indblkshift;         /* ln2(indirect block size) */
 147         uint8_t dn_nlevels;             /* 1=dn_blkptr->data blocks */
 148         uint8_t dn_nblkptr;             /* length of dn_blkptr */
 149         uint8_t dn_bonustype;           /* type of data in bonus buffer */
 150         uint8_t dn_checksum;            /* ZIO_CHECKSUM type */
 151         uint8_t dn_compress;            /* ZIO_COMPRESS type */
 152         uint8_t dn_flags;               /* DNODE_FLAG_* */
 153         uint16_t dn_datablkszsec;       /* data block size in 512b sectors */
 154         uint16_t dn_bonuslen;           /* length of dn_bonus */
 155         uint8_t dn_extra_slots;         /* # of subsequent slots consumed */
 156         uint8_t dn_pad2[3];
 157
 158         /* accounting is protected by dn_dirty_mtx */
 159         uint64_t dn_maxblkid;           /* largest allocated block ID */
 160         uint64_t dn_used;               /* bytes (or sectors) of disk space */
 161
 162         uint64_t dn_pad3[4];
 163
 164         /*
 165          * The tail region is 448 bytes for a 512 byte dnode, and
 166          * correspondingly larger for larger dnode sizes. The spill
 167          * block pointer, when present, is always at the end of the tail
 168          * region. There are three ways this space may be used, using
 169          * a 512 byte dnode for this diagram:
 170          *
 171          * 0       64      128     192     256     320     384     448 (offset)
 172          * +---------------+---------------+---------------+-------+
 173          * | dn_blkptr[0]  | dn_blkptr[1]  | dn_blkptr[2]  | /     |
 174          * +---------------+---------------+---------------+-------+
 175          * | dn_blkptr[0]  | dn_bonus[0..319]                      |
 176          * +---------------+-----------------------+---------------+
 177          * | dn_blkptr[0]  | dn_bonus[0..191]      | dn_spill      |
 178          * +---------------+-----------------------+---------------+
 179          */
 180         union {
 181                 blkptr_t dn_blkptr[1+DN_OLD_MAX_BONUSLEN/sizeof (blkptr_t)];
 182                 struct {
 183                         blkptr_t __dn_ignore1;
 184                         uint8_t dn_bonus[DN_OLD_MAX_BONUSLEN];
 185                 };
 186                 struct {
 187                         blkptr_t __dn_ignore2;
 188                         uint8_t __dn_ignore3[DN_OLD_MAX_BONUSLEN -
 189                             sizeof (blkptr_t)];
 190                         blkptr_t dn_spill;
 191                 };
 192         };
 193 } dnode_phys_t;
 194
 195 #define DN_SPILL_BLKPTR(dnp)    (blkptr_t *)((char *)(dnp) + \
 196         (((dnp)->dn_extra_slots + 1) << DNODE_SHIFT) - (1 << SPA_BLKPTRSHIFT))
 197
 198 struct dnode {
 199         /*
 200          * Protects the structure of the dnode, including the number of levels
 201          * of indirection (dn_nlevels), dn_maxblkid, and dn_next_*
 202          */
 203         krwlock_t dn_struct_rwlock;
 204
 205         /* Our link on dn_objset->os_dnodes list; protected by os_lock.  */
 206         list_node_t dn_link;
 207
 208         /* immutable: */
 209         struct objset *dn_objset;
 210         uint64_t dn_object;
 211         struct dmu_buf_impl *dn_dbuf;
 212         struct dnode_handle *dn_handle;
 213         dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */
 214
 215         /*
 216          * Copies of stuff in dn_phys.  They're valid in the open
 217          * context (eg. even before the dnode is first synced).
 218          * Where necessary, these are protected by dn_struct_rwlock.
 219          */
 220         dmu_object_type_t dn_type;      /* object type */
 221         uint16_t dn_bonuslen;           /* bonus length */
 222         uint8_t dn_bonustype;           /* bonus type */
 223         uint8_t dn_nblkptr;             /* number of blkptrs (immutable) */
 224         uint8_t dn_checksum;            /* ZIO_CHECKSUM type */
 225         uint8_t dn_compress;            /* ZIO_COMPRESS type */
 226         uint8_t dn_nlevels;
 227         uint8_t dn_indblkshift;
 228         uint8_t dn_datablkshift;        /* zero if blksz not power of 2! */
 229         uint8_t dn_moved;               /* Has this dnode been moved? */
 230         uint16_t dn_datablkszsec;       /* in 512b sectors */
 231         uint32_t dn_datablksz;          /* in bytes */
 232         uint64_t dn_maxblkid;
 233         uint8_t dn_next_type[TXG_SIZE];
 234         uint8_t dn_num_slots;           /* metadnode slots consumed on disk */
 235         uint8_t dn_next_nblkptr[TXG_SIZE];
 236         uint8_t dn_next_nlevels[TXG_SIZE];
 237         uint8_t dn_next_indblkshift[TXG_SIZE];
 238         uint8_t dn_next_bonustype[TXG_SIZE];
 239         uint8_t dn_rm_spillblk[TXG_SIZE];       /* for removing spill blk */
 240         uint16_t dn_next_bonuslen[TXG_SIZE];
 241         uint32_t dn_next_blksz[TXG_SIZE];       /* next block size in bytes */
 242
 243         /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */
 244         uint32_t dn_dbufs_count;        /* count of dn_dbufs */
 245
 246         /* protected by os_lock: */
 247         multilist_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */
 248
 249         /* protected by dn_mtx: */
 250         kmutex_t dn_mtx;
 251         list_t dn_dirty_records[TXG_SIZE];
 252         struct range_tree *dn_free_ranges[TXG_SIZE];
 253         uint64_t dn_allocated_txg;
 254         uint64_t dn_free_txg;
 255         uint64_t dn_assigned_txg;
 256         kcondvar_t dn_notxholds;
 257         enum dnode_dirtycontext dn_dirtyctx;
 258         uint8_t *dn_dirtyctx_firstset;          /* dbg: contents meaningless */
 259
 260         /* protected by own devices */
 261         refcount_t dn_tx_holds;
 262         refcount_t dn_holds;
 263
 264         kmutex_t dn_dbufs_mtx;
 265         /*
 266          * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs
 267          * can contain multiple dbufs of the same (level, blkid) when a
 268          * dbuf is marked DB_EVICTING without being removed from
 269          * dn_dbufs. To maintain the avl invariant that there cannot be
 270          * duplicate entries, we order the dbufs by an arbitrary value -
 271          * their address in memory. This means that dn_dbufs cannot be used to
 272          * directly look up a dbuf. Instead, callers must use avl_walk, have
 273          * a reference to the dbuf, or look up a non-existent node with
 274          * db_state = DB_SEARCH (see dbuf_free_range for an example).
 275          */
 276         avl_tree_t dn_dbufs;
 277
 278         /* protected by dn_struct_rwlock */
 279         struct dmu_buf_impl *dn_bonus;  /* bonus buffer dbuf */
 280
 281         boolean_t dn_have_spill;        /* have spill or are spilling */
 282
 283         /* parent IO for current sync write */
 284         zio_t *dn_zio;
 285
 286         /* used in syncing context */
 287         uint64_t dn_oldused;    /* old phys used bytes */
 288         uint64_t dn_oldflags;   /* old phys dn_flags */
 289         uint64_t dn_olduid, dn_oldgid;
 290         uint64_t dn_newuid, dn_newgid;
 291         int dn_id_flags;
 292
 293         /* holds prefetch structure */
 294         struct zfetch   dn_zfetch;
 295 };
 296
 297 /*
 298  * Adds a level of indirection between the dbuf and the dnode to avoid
 299  * iterating descendent dbufs in dnode_move(). Handles are not allocated
 300  * individually, but as an array of child dnodes in dnode_hold_impl().
 301  */
 302 typedef struct dnode_handle {
 303         /* Protects dnh_dnode from modification by dnode_move(). */
 304         zrlock_t dnh_zrlock;
 305         dnode_t *dnh_dnode;
 306 } dnode_handle_t;
 307
 308 typedef struct dnode_children {
 309         dmu_buf_user_t dnc_dbu;         /* User evict data */
 310         size_t dnc_count;               /* number of children */
 311         dnode_handle_t dnc_children[];  /* sized dynamically */
 312 } dnode_children_t;
 313
 314 typedef struct free_range {
 315         avl_node_t fr_node;
 316         uint64_t fr_blkid;
 317         uint64_t fr_nblks;
 318 } free_range_t;
 319
 320 void dnode_special_open(struct objset *dd, dnode_phys_t *dnp,
 321     uint64_t object, dnode_handle_t *dnh);
 322 void dnode_special_close(dnode_handle_t *dnh);
 323
 324 void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx);
 325 void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx);
 326 void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx);
 327
 328 int dnode_hold(struct objset *dd, uint64_t object,
 329     void *ref, dnode_t **dnp);
 330 int dnode_hold_impl(struct objset *dd, uint64_t object, int flag, int dn_slots,
 331     void *ref, dnode_t **dnp);
 332 boolean_t dnode_add_ref(dnode_t *dn, void *ref);
 333 void dnode_rele(dnode_t *dn, void *ref);
 334 void dnode_rele_and_unlock(dnode_t *dn, void *tag);
 335 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx);
 336 void dnode_sync(dnode_t *dn, dmu_tx_t *tx);
 337 void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
 338     dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx);
 339 void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
 340     dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx);
 341 void dnode_free(dnode_t *dn, dmu_tx_t *tx);
 342 void dnode_byteswap(dnode_phys_t *dnp);
 343 void dnode_buf_byteswap(void *buf, size_t size);
 344 void dnode_verify(dnode_t *dn);
 345 int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx);
 346 void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx);
 347 void dnode_diduse_space(dnode_t *dn, int64_t space);
 348 void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t);
 349 uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid);
 350 void dnode_init(void);
 351 void dnode_fini(void);
 352 int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off,
 353     int minlvl, uint64_t blkfill, uint64_t txg);
 354 void dnode_evict_dbufs(dnode_t *dn);
 355 void dnode_evict_bonus(dnode_t *dn);
 356
 357 #define DNODE_IS_CACHEABLE(_dn)                                         \
 358         ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL ||         \
 359         (DMU_OT_IS_METADATA((_dn)->dn_type) &&                          \
 360         (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA))
 361
 362 #define DNODE_META_IS_CACHEABLE(_dn)                                    \
 363         ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL ||         \
 364         (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)
 365
 366 #ifdef ZFS_DEBUG
 367
 368 /*
 369  * There should be a ## between the string literal and fmt, to make it
 370  * clear that we're joining two strings together, but that piece of shit
 371  * gcc doesn't support that preprocessor token.
 372  */
 373 #define dprintf_dnode(dn, fmt, ...) do { \
 374         if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
 375         char __db_buf[32]; \
 376         uint64_t __db_obj = (dn)->dn_object; \
 377         if (__db_obj == DMU_META_DNODE_OBJECT) \
 378                 (void) strcpy(__db_buf, "mdn"); \
 379         else \
 380                 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
 381                     (u_longlong_t)__db_obj);\
 382         dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \
 383             __db_buf, __VA_ARGS__); \
 384         } \
 385 _NOTE(CONSTCOND) } while (0)
 386
 387 #define DNODE_VERIFY(dn)                dnode_verify(dn)
 388 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx)
 389
 390 #else
 391
 392 #define dprintf_dnode(db, fmt, ...)
 393 #define DNODE_VERIFY(dn)
 394 #define FREE_VERIFY(db, start, end, tx)
 395
 396 #endif
 397
 398 #ifdef  __cplusplus
 399 }
 400 #endif
 401
 402 #endif  /* _SYS_DNODE_H */