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, 2016 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
  39 #ifdef  __cplusplus
  40 extern "C" {
  41 #endif
  42
  43 /*
  44  * dnode_hold() flags.
  45  */
  46 #define DNODE_MUST_BE_ALLOCATED 1
  47 #define DNODE_MUST_BE_FREE      2
  48
  49 /*
  50  * dnode_next_offset() flags.
  51  */
  52 #define DNODE_FIND_HOLE         1
  53 #define DNODE_FIND_BACKWARDS    2
  54 #define DNODE_FIND_HAVELOCK     4
  55
  56 /*
  57  * Fixed constants.
  58  */
  59 #define DNODE_SHIFT             9       /* 512 bytes */
  60 #define DN_MIN_INDBLKSHIFT      12      /* 4k */
  61 /*
  62  * If we ever increase this value beyond 20, we need to revisit all logic that
  63  * does x << level * ebps to handle overflow.  With a 1M indirect block size,
  64  * 4 levels of indirect blocks would not be able to guarantee addressing an
  65  * entire object, so 5 levels will be used, but 5 * (20 - 7) = 65.
  66  */
  67 #define DN_MAX_INDBLKSHIFT      17      /* 128k */
  68 #define DNODE_BLOCK_SHIFT       14      /* 16k */
  69 #define DNODE_CORE_SIZE         64      /* 64 bytes for dnode sans blkptrs */
  70 #define DN_MAX_OBJECT_SHIFT     48      /* 256 trillion (zfs_fid_t limit) */
  71 #define DN_MAX_OFFSET_SHIFT     64      /* 2^64 bytes in a dnode */
  72
  73 /*
  74  * dnode id flags
  75  *
  76  * Note: a file will never ever have its
  77  * ids moved from bonus->spill
  78  * and only in a crypto environment would it be on spill
  79  */
  80 #define DN_ID_CHKED_BONUS       0x1
  81 #define DN_ID_CHKED_SPILL       0x2
  82 #define DN_ID_OLD_EXIST         0x4
  83 #define DN_ID_NEW_EXIST         0x8
  84
  85 /*
  86  * Derived constants.
  87  */
  88 #define DNODE_MIN_SIZE          (1 << DNODE_SHIFT)
  89 #define DNODE_MAX_SIZE          (1 << DNODE_BLOCK_SHIFT)
  90 #define DNODE_BLOCK_SIZE        (1 << DNODE_BLOCK_SHIFT)
  91 #define DNODE_MIN_SLOTS         (DNODE_MIN_SIZE >> DNODE_SHIFT)
  92 #define DNODE_MAX_SLOTS         (DNODE_MAX_SIZE >> DNODE_SHIFT)
  93 #define DN_BONUS_SIZE(dnsize)   ((dnsize) - DNODE_CORE_SIZE - \
  94         (1 << SPA_BLKPTRSHIFT))
  95 #define DN_SLOTS_TO_BONUSLEN(slots)     DN_BONUS_SIZE((slots) << DNODE_SHIFT)
  96 #define DN_OLD_MAX_BONUSLEN     (DN_BONUS_SIZE(DNODE_MIN_SIZE))
  97 #define DN_MAX_NBLKPTR  ((DNODE_MIN_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
  98 #define DN_MAX_OBJECT   (1ULL << DN_MAX_OBJECT_SHIFT)
  99 #define DN_ZERO_BONUSLEN        (DN_BONUS_SIZE(DNODE_MAX_SIZE) + 1)
 100 #define DN_KILL_SPILLBLK (1)
 101
 102 #define DNODES_PER_BLOCK_SHIFT  (DNODE_BLOCK_SHIFT - DNODE_SHIFT)
 103 #define DNODES_PER_BLOCK        (1ULL << DNODES_PER_BLOCK_SHIFT)
 104
 105 /*
 106  * This is inaccurate if the indblkshift of the particular object is not the
 107  * max.  But it's only used by userland to calculate the zvol reservation.
 108  */
 109 #define DNODES_PER_LEVEL_SHIFT  (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT)
 110 #define DNODES_PER_LEVEL        (1ULL << DNODES_PER_LEVEL_SHIFT)
 111
 112 #define DN_MAX_LEVELS   (DIV_ROUND_UP(DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT, \
 113         DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT) + 1)
 114
 115 #define DN_BONUS(dnp)   ((void*)((dnp)->dn_bonus + \
 116         (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t))))
 117
 118 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \
 119         (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT)
 120
 121 #define EPB(blkshift, typeshift)        (1 << (blkshift - typeshift))
 122
 123 struct dmu_buf_impl;
 124 struct objset;
 125 struct zio;
 126
 127 enum dnode_dirtycontext {
 128         DN_UNDIRTIED,
 129         DN_DIRTY_OPEN,
 130         DN_DIRTY_SYNC
 131 };
 132
 133 /* Is dn_used in bytes?  if not, it's in multiples of SPA_MINBLOCKSIZE */
 134 #define DNODE_FLAG_USED_BYTES                   (1 << 0)
 135 #define DNODE_FLAG_USERUSED_ACCOUNTED           (1 << 1)
 136
 137 /* Does dnode have a SA spill blkptr in bonus? */
 138 #define DNODE_FLAG_SPILL_BLKPTR                 (1 << 2)
 139
 140 /* User/Group dnode accounting */
 141 #define DNODE_FLAG_USEROBJUSED_ACCOUNTED        (1 << 3)
 142
 143 typedef struct dnode_phys {
 144         uint8_t dn_type;                /* dmu_object_type_t */
 145         uint8_t dn_indblkshift;         /* ln2(indirect block size) */
 146         uint8_t dn_nlevels;             /* 1=dn_blkptr->data blocks */
 147         uint8_t dn_nblkptr;             /* length of dn_blkptr */
 148         uint8_t dn_bonustype;           /* type of data in bonus buffer */
 149         uint8_t dn_checksum;            /* ZIO_CHECKSUM type */
 150         uint8_t dn_compress;            /* ZIO_COMPRESS type */
 151         uint8_t dn_flags;               /* DNODE_FLAG_* */
 152         uint16_t dn_datablkszsec;       /* data block size in 512b sectors */
 153         uint16_t dn_bonuslen;           /* length of dn_bonus */
 154         uint8_t dn_extra_slots;         /* # of subsequent slots consumed */
 155         uint8_t dn_pad2[3];
 156
 157         /* accounting is protected by dn_dirty_mtx */
 158         uint64_t dn_maxblkid;           /* largest allocated block ID */
 159         uint64_t dn_used;               /* bytes (or sectors) of disk space */
 160
 161         uint64_t dn_pad3[4];
 162
 163         /*
 164          * The tail region is 448 bytes for a 512 byte dnode, and
 165          * correspondingly larger for larger dnode sizes. The spill
 166          * block pointer, when present, is always at the end of the tail
 167          * region. There are three ways this space may be used, using
 168          * a 512 byte dnode for this diagram:
 169          *
 170          * 0       64      128     192     256     320     384     448 (offset)
 171          * +---------------+---------------+---------------+-------+
 172          * | dn_blkptr[0]  | dn_blkptr[1]  | dn_blkptr[2]  | /     |
 173          * +---------------+---------------+---------------+-------+
 174          * | dn_blkptr[0]  | dn_bonus[0..319]                      |
 175          * +---------------+-----------------------+---------------+
 176          * | dn_blkptr[0]  | dn_bonus[0..191]      | dn_spill      |
 177          * +---------------+-----------------------+---------------+
 178          */
 179         union {
 180                 blkptr_t dn_blkptr[1+DN_OLD_MAX_BONUSLEN/sizeof (blkptr_t)];
 181                 struct {
 182                         blkptr_t __dn_ignore1;
 183                         uint8_t dn_bonus[DN_OLD_MAX_BONUSLEN];
 184                 };
 185                 struct {
 186                         blkptr_t __dn_ignore2;
 187                         uint8_t __dn_ignore3[DN_OLD_MAX_BONUSLEN -
 188                             sizeof (blkptr_t)];
 189                         blkptr_t dn_spill;
 190                 };
 191         };
 192 } dnode_phys_t;
 193
 194 #define DN_SPILL_BLKPTR(dnp)    (blkptr_t *)((char *)(dnp) + \
 195         (((dnp)->dn_extra_slots + 1) << DNODE_SHIFT) - (1 << SPA_BLKPTRSHIFT))
 196
 197 struct dnode {
 198         /*
 199          * Protects the structure of the dnode, including the number of levels
 200          * of indirection (dn_nlevels), dn_maxblkid, and dn_next_*
 201          */
 202         krwlock_t dn_struct_rwlock;
 203
 204         /* Our link on dn_objset->os_dnodes list; protected by os_lock.  */
 205         list_node_t dn_link;
 206
 207         /* immutable: */
 208         struct objset *dn_objset;
 209         uint64_t dn_object;
 210         struct dmu_buf_impl *dn_dbuf;
 211         struct dnode_handle *dn_handle;
 212         dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */
 213
 214         /*
 215          * Copies of stuff in dn_phys.  They're valid in the open
 216          * context (eg. even before the dnode is first synced).
 217          * Where necessary, these are protected by dn_struct_rwlock.
 218          */
 219         dmu_object_type_t dn_type;      /* object type */
 220         uint16_t dn_bonuslen;           /* bonus length */
 221         uint8_t dn_bonustype;           /* bonus type */
 222         uint8_t dn_nblkptr;             /* number of blkptrs (immutable) */
 223         uint8_t dn_checksum;            /* ZIO_CHECKSUM type */
 224         uint8_t dn_compress;            /* ZIO_COMPRESS type */
 225         uint8_t dn_nlevels;
 226         uint8_t dn_indblkshift;
 227         uint8_t dn_datablkshift;        /* zero if blksz not power of 2! */
 228         uint8_t dn_moved;               /* Has this dnode been moved? */
 229         uint16_t dn_datablkszsec;       /* in 512b sectors */
 230         uint32_t dn_datablksz;          /* in bytes */
 231         uint64_t dn_maxblkid;
 232         uint8_t dn_next_type[TXG_SIZE];
 233         uint8_t dn_num_slots;           /* metadnode slots consumed on disk */
 234         uint8_t dn_next_nblkptr[TXG_SIZE];
 235         uint8_t dn_next_nlevels[TXG_SIZE];
 236         uint8_t dn_next_indblkshift[TXG_SIZE];
 237         uint8_t dn_next_bonustype[TXG_SIZE];
 238         uint8_t dn_rm_spillblk[TXG_SIZE];       /* for removing spill blk */
 239         uint16_t dn_next_bonuslen[TXG_SIZE];
 240         uint32_t dn_next_blksz[TXG_SIZE];       /* next block size in bytes */
 241
 242         /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */
 243         uint32_t dn_dbufs_count;        /* count of dn_dbufs */
 244
 245         /* protected by os_lock: */
 246         list_node_t dn_dirty_link[TXG_SIZE];    /* next on dataset's dirty */
 247
 248         /* protected by dn_mtx: */
 249         kmutex_t dn_mtx;
 250         list_t dn_dirty_records[TXG_SIZE];
 251         struct range_tree *dn_free_ranges[TXG_SIZE];
 252         uint64_t dn_allocated_txg;
 253         uint64_t dn_free_txg;
 254         uint64_t dn_assigned_txg;
 255         kcondvar_t dn_notxholds;
 256         enum dnode_dirtycontext dn_dirtyctx;
 257         uint8_t *dn_dirtyctx_firstset;          /* dbg: contents meaningless */
 258
 259         /* protected by own devices */
 260         refcount_t dn_tx_holds;
 261         refcount_t dn_holds;
 262
 263         kmutex_t dn_dbufs_mtx;
 264         /*
 265          * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs
 266          * can contain multiple dbufs of the same (level, blkid) when a
 267          * dbuf is marked DB_EVICTING without being removed from
 268          * dn_dbufs. To maintain the avl invariant that there cannot be
 269          * duplicate entries, we order the dbufs by an arbitrary value -
 270          * their address in memory. This means that dn_dbufs cannot be used to
 271          * directly look up a dbuf. Instead, callers must use avl_walk, have
 272          * a reference to the dbuf, or look up a non-existent node with
 273          * db_state = DB_SEARCH (see dbuf_free_range for an example).
 274          */
 275         avl_tree_t dn_dbufs;
 276
 277         /* protected by dn_struct_rwlock */
 278         struct dmu_buf_impl *dn_bonus;  /* bonus buffer dbuf */
 279
 280         boolean_t dn_have_spill;        /* have spill or are spilling */
 281
 282         /* parent IO for current sync write */
 283         zio_t *dn_zio;
 284
 285         /* used in syncing context */
 286         uint64_t dn_oldused;    /* old phys used bytes */
 287         uint64_t dn_oldflags;   /* old phys dn_flags */
 288         uint64_t dn_olduid, dn_oldgid;
 289         uint64_t dn_newuid, dn_newgid;
 290         int dn_id_flags;
 291
 292         /* holds prefetch structure */
 293         struct zfetch   dn_zfetch;
 294 };
 295
 296 /*
 297  * Adds a level of indirection between the dbuf and the dnode to avoid
 298  * iterating descendent dbufs in dnode_move(). Handles are not allocated
 299  * individually, but as an array of child dnodes in dnode_hold_impl().
 300  */
 301 typedef struct dnode_handle {
 302         /* Protects dnh_dnode from modification by dnode_move(). */
 303         zrlock_t dnh_zrlock;
 304         dnode_t *dnh_dnode;
 305 } dnode_handle_t;
 306
 307 typedef struct dnode_children {
 308         dmu_buf_user_t dnc_dbu;         /* User evict data */
 309         size_t dnc_count;               /* number of children */
 310         dnode_handle_t dnc_children[];  /* sized dynamically */
 311 } dnode_children_t;
 312
 313 typedef struct free_range {
 314         avl_node_t fr_node;
 315         uint64_t fr_blkid;
 316         uint64_t fr_nblks;
 317 } free_range_t;
 318
 319 void dnode_special_open(struct objset *dd, dnode_phys_t *dnp,
 320     uint64_t object, dnode_handle_t *dnh);
 321 void dnode_special_close(dnode_handle_t *dnh);
 322
 323 void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx);
 324 void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx);
 325 void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx);
 326
 327 int dnode_hold(struct objset *dd, uint64_t object,
 328     void *ref, dnode_t **dnp);
 329 int dnode_hold_impl(struct objset *dd, uint64_t object, int flag, int dn_slots,
 330     void *ref, dnode_t **dnp);
 331 boolean_t dnode_add_ref(dnode_t *dn, void *ref);
 332 void dnode_rele(dnode_t *dn, void *ref);
 333 void dnode_rele_and_unlock(dnode_t *dn, void *tag);
 334 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx);
 335 void dnode_sync(dnode_t *dn, dmu_tx_t *tx);
 336 void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
 337     dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx);
 338 void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
 339     dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx);
 340 void dnode_free(dnode_t *dn, dmu_tx_t *tx);
 341 void dnode_byteswap(dnode_phys_t *dnp);
 342 void dnode_buf_byteswap(void *buf, size_t size);
 343 void dnode_verify(dnode_t *dn);
 344 int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx);
 345 void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx);
 346 void dnode_diduse_space(dnode_t *dn, int64_t space);
 347 void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t);
 348 uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid);
 349 void dnode_init(void);
 350 void dnode_fini(void);
 351 int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off,
 352     int minlvl, uint64_t blkfill, uint64_t txg);
 353 void dnode_evict_dbufs(dnode_t *dn);
 354 void dnode_evict_bonus(dnode_t *dn);
 355
 356 #define DNODE_IS_CACHEABLE(_dn)                                         \
 357         ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL ||         \
 358         (DMU_OT_IS_METADATA((_dn)->dn_type) &&                          \
 359         (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA))
 360
 361 #define DNODE_META_IS_CACHEABLE(_dn)                                    \
 362         ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL ||         \
 363         (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)
 364
 365 #ifdef ZFS_DEBUG
 366
 367 /*
 368  * There should be a ## between the string literal and fmt, to make it
 369  * clear that we're joining two strings together, but that piece of shit
 370  * gcc doesn't support that preprocessor token.
 371  */
 372 #define dprintf_dnode(dn, fmt, ...) do { \
 373         if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
 374         char __db_buf[32]; \
 375         uint64_t __db_obj = (dn)->dn_object; \
 376         if (__db_obj == DMU_META_DNODE_OBJECT) \
 377                 (void) strcpy(__db_buf, "mdn"); \
 378         else \
 379                 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
 380                     (u_longlong_t)__db_obj);\
 381         dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \
 382             __db_buf, __VA_ARGS__); \
 383         } \
 384 _NOTE(CONSTCOND) } while (0)
 385
 386 #define DNODE_VERIFY(dn)                dnode_verify(dn)
 387 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx)
 388
 389 #else
 390
 391 #define dprintf_dnode(db, fmt, ...)
 392 #define DNODE_VERIFY(dn)
 393 #define FREE_VERIFY(db, start, end, tx)
 394
 395 #endif
 396
 397 #ifdef  __cplusplus
 398 }
 399 #endif
 400
 401 #endif  /* _SYS_DNODE_H */