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.
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.
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]
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.
30 #include <sys/zfs_context.h>
35 #include <sys/refcount.h>
36 #include <sys/dmu_zfetch.h>
37 #include <sys/zrlock.h>
46 #define DNODE_MUST_BE_ALLOCATED 1
47 #define DNODE_MUST_BE_FREE 2
50 * dnode_next_offset() flags.
52 #define DNODE_FIND_HOLE 1
53 #define DNODE_FIND_BACKWARDS 2
54 #define DNODE_FIND_HAVELOCK 4
59 #define DNODE_SHIFT 9 /* 512 bytes */
60 #define DN_MIN_INDBLKSHIFT 12 /* 4k */
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.
67 #define DN_MAX_INDBLKSHIFT 14 /* 16k */
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 */
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
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
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)
102 #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT)
103 #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT)
104 #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT)
105 #define DNODES_PER_LEVEL (1ULL << DNODES_PER_LEVEL_SHIFT)
107 #define DN_MAX_LEVELS (DIV_ROUND_UP(DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT, \
108 DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT) + 1)
110 #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \
111 (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t))))
113 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \
114 (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT)
116 #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift))
122 enum dnode_dirtycontext
{
128 /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */
129 #define DNODE_FLAG_USED_BYTES (1<<0)
130 #define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1)
132 /* Does dnode have a SA spill blkptr in bonus? */
133 #define DNODE_FLAG_SPILL_BLKPTR (1<<2)
135 typedef struct dnode_phys
{
136 uint8_t dn_type
; /* dmu_object_type_t */
137 uint8_t dn_indblkshift
; /* ln2(indirect block size) */
138 uint8_t dn_nlevels
; /* 1=dn_blkptr->data blocks */
139 uint8_t dn_nblkptr
; /* length of dn_blkptr */
140 uint8_t dn_bonustype
; /* type of data in bonus buffer */
141 uint8_t dn_checksum
; /* ZIO_CHECKSUM type */
142 uint8_t dn_compress
; /* ZIO_COMPRESS type */
143 uint8_t dn_flags
; /* DNODE_FLAG_* */
144 uint16_t dn_datablkszsec
; /* data block size in 512b sectors */
145 uint16_t dn_bonuslen
; /* length of dn_bonus */
146 uint8_t dn_extra_slots
; /* # of subsequent slots consumed */
149 /* accounting is protected by dn_dirty_mtx */
150 uint64_t dn_maxblkid
; /* largest allocated block ID */
151 uint64_t dn_used
; /* bytes (or sectors) of disk space */
156 * The tail region is 448 bytes for a 512 byte dnode, and
157 * correspondingly larger for larger dnode sizes. The spill
158 * block pointer, when present, is always at the end of the tail
159 * region. There are three ways this space may be used, using
160 * a 512 byte dnode for this diagram:
162 * 0 64 128 192 256 320 384 448 (offset)
163 * +---------------+---------------+---------------+-------+
164 * | dn_blkptr[0] | dn_blkptr[1] | dn_blkptr[2] | / |
165 * +---------------+---------------+---------------+-------+
166 * | dn_blkptr[0] | dn_bonus[0..319] |
167 * +---------------+-----------------------+---------------+
168 * | dn_blkptr[0] | dn_bonus[0..191] | dn_spill |
169 * +---------------+-----------------------+---------------+
172 blkptr_t dn_blkptr
[1+DN_OLD_MAX_BONUSLEN
/sizeof (blkptr_t
)];
174 blkptr_t __dn_ignore1
;
175 uint8_t dn_bonus
[DN_OLD_MAX_BONUSLEN
];
178 blkptr_t __dn_ignore2
;
179 uint8_t __dn_ignore3
[DN_OLD_MAX_BONUSLEN
-
186 #define DN_SPILL_BLKPTR(dnp) (blkptr_t *)((char *)(dnp) + \
187 (((dnp)->dn_extra_slots + 1) << DNODE_SHIFT) - (1 << SPA_BLKPTRSHIFT))
191 * Protects the structure of the dnode, including the number of levels
192 * of indirection (dn_nlevels), dn_maxblkid, and dn_next_*
194 krwlock_t dn_struct_rwlock
;
196 /* Our link on dn_objset->os_dnodes list; protected by os_lock. */
200 struct objset
*dn_objset
;
202 struct dmu_buf_impl
*dn_dbuf
;
203 struct dnode_handle
*dn_handle
;
204 dnode_phys_t
*dn_phys
; /* pointer into dn->dn_dbuf->db.db_data */
207 * Copies of stuff in dn_phys. They're valid in the open
208 * context (eg. even before the dnode is first synced).
209 * Where necessary, these are protected by dn_struct_rwlock.
211 dmu_object_type_t dn_type
; /* object type */
212 uint16_t dn_bonuslen
; /* bonus length */
213 uint8_t dn_bonustype
; /* bonus type */
214 uint8_t dn_nblkptr
; /* number of blkptrs (immutable) */
215 uint8_t dn_checksum
; /* ZIO_CHECKSUM type */
216 uint8_t dn_compress
; /* ZIO_COMPRESS type */
218 uint8_t dn_indblkshift
;
219 uint8_t dn_datablkshift
; /* zero if blksz not power of 2! */
220 uint8_t dn_moved
; /* Has this dnode been moved? */
221 uint16_t dn_datablkszsec
; /* in 512b sectors */
222 uint32_t dn_datablksz
; /* in bytes */
223 uint64_t dn_maxblkid
;
224 uint8_t dn_next_type
[TXG_SIZE
];
225 uint8_t dn_num_slots
; /* metadnode slots consumed on disk */
226 uint8_t dn_next_nblkptr
[TXG_SIZE
];
227 uint8_t dn_next_nlevels
[TXG_SIZE
];
228 uint8_t dn_next_indblkshift
[TXG_SIZE
];
229 uint8_t dn_next_bonustype
[TXG_SIZE
];
230 uint8_t dn_rm_spillblk
[TXG_SIZE
]; /* for removing spill blk */
231 uint16_t dn_next_bonuslen
[TXG_SIZE
];
232 uint32_t dn_next_blksz
[TXG_SIZE
]; /* next block size in bytes */
234 /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */
235 uint32_t dn_dbufs_count
; /* count of dn_dbufs */
236 /* There are no level-0 blocks of this blkid or higher in dn_dbufs */
237 uint64_t dn_unlisted_l0_blkid
;
239 /* protected by os_lock: */
240 list_node_t dn_dirty_link
[TXG_SIZE
]; /* next on dataset's dirty */
242 /* protected by dn_mtx: */
244 list_t dn_dirty_records
[TXG_SIZE
];
245 struct range_tree
*dn_free_ranges
[TXG_SIZE
];
246 uint64_t dn_allocated_txg
;
247 uint64_t dn_free_txg
;
248 uint64_t dn_assigned_txg
;
249 kcondvar_t dn_notxholds
;
250 enum dnode_dirtycontext dn_dirtyctx
;
251 uint8_t *dn_dirtyctx_firstset
; /* dbg: contents meaningless */
253 /* protected by own devices */
254 refcount_t dn_tx_holds
;
257 kmutex_t dn_dbufs_mtx
;
259 * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs
260 * can contain multiple dbufs of the same (level, blkid) when a
261 * dbuf is marked DB_EVICTING without being removed from
262 * dn_dbufs. To maintain the avl invariant that there cannot be
263 * duplicate entries, we order the dbufs by an arbitrary value -
264 * their address in memory. This means that dn_dbufs cannot be used to
265 * directly look up a dbuf. Instead, callers must use avl_walk, have
266 * a reference to the dbuf, or look up a non-existant node with
267 * db_state = DB_SEARCH (see dbuf_free_range for an example).
271 /* protected by dn_struct_rwlock */
272 struct dmu_buf_impl
*dn_bonus
; /* bonus buffer dbuf */
274 boolean_t dn_have_spill
; /* have spill or are spilling */
276 /* parent IO for current sync write */
279 /* used in syncing context */
280 uint64_t dn_oldused
; /* old phys used bytes */
281 uint64_t dn_oldflags
; /* old phys dn_flags */
282 uint64_t dn_olduid
, dn_oldgid
;
283 uint64_t dn_newuid
, dn_newgid
;
286 /* holds prefetch structure */
287 struct zfetch dn_zfetch
;
291 * Adds a level of indirection between the dbuf and the dnode to avoid
292 * iterating descendent dbufs in dnode_move(). Handles are not allocated
293 * individually, but as an array of child dnodes in dnode_hold_impl().
295 typedef struct dnode_handle
{
296 /* Protects dnh_dnode from modification by dnode_move(). */
301 typedef struct dnode_children
{
302 dmu_buf_user_t dnc_dbu
; /* User evict data */
303 size_t dnc_count
; /* number of children */
304 dnode_handle_t dnc_children
[]; /* sized dynamically */
307 typedef struct free_range
{
313 void dnode_special_open(struct objset
*dd
, dnode_phys_t
*dnp
,
314 uint64_t object
, dnode_handle_t
*dnh
);
315 void dnode_special_close(dnode_handle_t
*dnh
);
317 void dnode_setbonuslen(dnode_t
*dn
, int newsize
, dmu_tx_t
*tx
);
318 void dnode_setbonus_type(dnode_t
*dn
, dmu_object_type_t
, dmu_tx_t
*tx
);
319 void dnode_rm_spill(dnode_t
*dn
, dmu_tx_t
*tx
);
321 int dnode_hold(struct objset
*dd
, uint64_t object
,
322 void *ref
, dnode_t
**dnp
);
323 int dnode_hold_impl(struct objset
*dd
, uint64_t object
, int flag
, int dn_slots
,
324 void *ref
, dnode_t
**dnp
);
325 boolean_t
dnode_add_ref(dnode_t
*dn
, void *ref
);
326 void dnode_rele(dnode_t
*dn
, void *ref
);
327 void dnode_rele_and_unlock(dnode_t
*dn
, void *tag
);
328 void dnode_setdirty(dnode_t
*dn
, dmu_tx_t
*tx
);
329 void dnode_sync(dnode_t
*dn
, dmu_tx_t
*tx
);
330 void dnode_allocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
, int ibs
,
331 dmu_object_type_t bonustype
, int bonuslen
, int dn_slots
, dmu_tx_t
*tx
);
332 void dnode_reallocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
,
333 dmu_object_type_t bonustype
, int bonuslen
, int dn_slots
, dmu_tx_t
*tx
);
334 void dnode_free(dnode_t
*dn
, dmu_tx_t
*tx
);
335 void dnode_byteswap(dnode_phys_t
*dnp
);
336 void dnode_buf_byteswap(void *buf
, size_t size
);
337 void dnode_verify(dnode_t
*dn
);
338 int dnode_set_blksz(dnode_t
*dn
, uint64_t size
, int ibs
, dmu_tx_t
*tx
);
339 void dnode_free_range(dnode_t
*dn
, uint64_t off
, uint64_t len
, dmu_tx_t
*tx
);
340 void dnode_diduse_space(dnode_t
*dn
, int64_t space
);
341 void dnode_willuse_space(dnode_t
*dn
, int64_t space
, dmu_tx_t
*tx
);
342 void dnode_new_blkid(dnode_t
*dn
, uint64_t blkid
, dmu_tx_t
*tx
, boolean_t
);
343 uint64_t dnode_block_freed(dnode_t
*dn
, uint64_t blkid
);
344 void dnode_init(void);
345 void dnode_fini(void);
346 int dnode_next_offset(dnode_t
*dn
, int flags
, uint64_t *off
,
347 int minlvl
, uint64_t blkfill
, uint64_t txg
);
348 void dnode_evict_dbufs(dnode_t
*dn
);
349 void dnode_evict_bonus(dnode_t
*dn
);
351 #define DNODE_IS_CACHEABLE(_dn) \
352 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \
353 (DMU_OT_IS_METADATA((_dn)->dn_type) && \
354 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA))
356 #define DNODE_META_IS_CACHEABLE(_dn) \
357 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \
358 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)
363 * There should be a ## between the string literal and fmt, to make it
364 * clear that we're joining two strings together, but that piece of shit
365 * gcc doesn't support that preprocessor token.
367 #define dprintf_dnode(dn, fmt, ...) do { \
368 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
370 uint64_t __db_obj = (dn)->dn_object; \
371 if (__db_obj == DMU_META_DNODE_OBJECT) \
372 (void) strcpy(__db_buf, "mdn"); \
374 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
375 (u_longlong_t)__db_obj);\
376 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \
377 __db_buf, __VA_ARGS__); \
379 _NOTE(CONSTCOND) } while (0)
381 #define DNODE_VERIFY(dn) dnode_verify(dn)
382 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx)
386 #define dprintf_dnode(db, fmt, ...)
387 #define DNODE_VERIFY(dn)
388 #define FREE_VERIFY(db, start, end, tx)
396 #endif /* _SYS_DNODE_H */