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 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
29 #include <sys/zfs_context.h>
34 #include <sys/refcount.h>
35 #include <sys/dmu_zfetch.h>
44 #define DNODE_MUST_BE_ALLOCATED 1
45 #define DNODE_MUST_BE_FREE 2
48 * dnode_next_offset() flags.
50 #define DNODE_FIND_HOLE 1
51 #define DNODE_FIND_BACKWARDS 2
52 #define DNODE_FIND_HAVELOCK 4
57 #define DNODE_SHIFT 9 /* 512 bytes */
58 #define DN_MIN_INDBLKSHIFT 10 /* 1k */
59 #define DN_MAX_INDBLKSHIFT 14 /* 16k */
60 #define DNODE_BLOCK_SHIFT 14 /* 16k */
61 #define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */
62 #define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */
63 #define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */
68 #define DNODE_SIZE (1 << DNODE_SHIFT)
69 #define DN_MAX_NBLKPTR ((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
70 #define DN_MAX_BONUSLEN (DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT))
71 #define DN_MAX_OBJECT (1ULL << DN_MAX_OBJECT_SHIFT)
72 #define DN_ZERO_BONUSLEN (DN_MAX_BONUSLEN + 1)
74 #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT)
75 #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT)
76 #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT)
78 /* The +2 here is a cheesy way to round up */
79 #define DN_MAX_LEVELS (2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \
80 (DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT)))
82 #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \
83 (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t))))
85 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \
86 (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT)
88 #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift))
94 enum dnode_dirtycontext
{
100 /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */
101 #define DNODE_FLAG_USED_BYTES (1<<0)
103 typedef struct dnode_phys
{
104 uint8_t dn_type
; /* dmu_object_type_t */
105 uint8_t dn_indblkshift
; /* ln2(indirect block size) */
106 uint8_t dn_nlevels
; /* 1=dn_blkptr->data blocks */
107 uint8_t dn_nblkptr
; /* length of dn_blkptr */
108 uint8_t dn_bonustype
; /* type of data in bonus buffer */
109 uint8_t dn_checksum
; /* ZIO_CHECKSUM type */
110 uint8_t dn_compress
; /* ZIO_COMPRESS type */
111 uint8_t dn_flags
; /* DNODE_FLAG_* */
112 uint16_t dn_datablkszsec
; /* data block size in 512b sectors */
113 uint16_t dn_bonuslen
; /* length of dn_bonus */
116 /* accounting is protected by dn_dirty_mtx */
117 uint64_t dn_maxblkid
; /* largest allocated block ID */
118 uint64_t dn_used
; /* bytes (or sectors) of disk space */
122 blkptr_t dn_blkptr
[1];
123 uint8_t dn_bonus
[DN_MAX_BONUSLEN
];
126 typedef struct dnode
{
128 * dn_struct_rwlock protects the structure of the dnode,
129 * including the number of levels of indirection (dn_nlevels),
130 * dn_maxblkid, and dn_next_*
132 krwlock_t dn_struct_rwlock
;
135 * Our link on dataset's dd_dnodes list.
136 * Protected by dd_accounting_mtx.
141 struct objset_impl
*dn_objset
;
143 struct dmu_buf_impl
*dn_dbuf
;
144 dnode_phys_t
*dn_phys
; /* pointer into dn->dn_dbuf->db.db_data */
147 * Copies of stuff in dn_phys. They're valid in the open
148 * context (eg. even before the dnode is first synced).
149 * Where necessary, these are protected by dn_struct_rwlock.
151 dmu_object_type_t dn_type
; /* object type */
152 uint16_t dn_bonuslen
; /* bonus length */
153 uint8_t dn_bonustype
; /* bonus type */
154 uint8_t dn_nblkptr
; /* number of blkptrs (immutable) */
155 uint8_t dn_checksum
; /* ZIO_CHECKSUM type */
156 uint8_t dn_compress
; /* ZIO_COMPRESS type */
158 uint8_t dn_indblkshift
;
159 uint8_t dn_datablkshift
; /* zero if blksz not power of 2! */
160 uint16_t dn_datablkszsec
; /* in 512b sectors */
161 uint32_t dn_datablksz
; /* in bytes */
162 uint64_t dn_maxblkid
;
163 uint8_t dn_next_nlevels
[TXG_SIZE
];
164 uint8_t dn_next_indblkshift
[TXG_SIZE
];
165 uint16_t dn_next_bonuslen
[TXG_SIZE
];
166 uint32_t dn_next_blksz
[TXG_SIZE
]; /* next block size in bytes */
168 /* protected by os_lock: */
169 list_node_t dn_dirty_link
[TXG_SIZE
]; /* next on dataset's dirty */
171 /* protected by dn_mtx: */
173 list_t dn_dirty_records
[TXG_SIZE
];
174 avl_tree_t dn_ranges
[TXG_SIZE
];
175 uint64_t dn_allocated_txg
;
176 uint64_t dn_free_txg
;
177 uint64_t dn_assigned_txg
;
178 kcondvar_t dn_notxholds
;
179 enum dnode_dirtycontext dn_dirtyctx
;
180 uint8_t *dn_dirtyctx_firstset
; /* dbg: contents meaningless */
182 /* protected by own devices */
183 refcount_t dn_tx_holds
;
186 kmutex_t dn_dbufs_mtx
;
187 list_t dn_dbufs
; /* linked list of descendent dbuf_t's */
188 struct dmu_buf_impl
*dn_bonus
; /* bonus buffer dbuf */
190 /* parent IO for current sync write */
193 /* holds prefetch structure */
194 struct zfetch dn_zfetch
;
197 typedef struct free_range
{
203 dnode_t
*dnode_special_open(struct objset_impl
*dd
, dnode_phys_t
*dnp
,
205 void dnode_special_close(dnode_t
*dn
);
207 void dnode_setbonuslen(dnode_t
*dn
, int newsize
, dmu_tx_t
*tx
);
208 int dnode_hold(struct objset_impl
*dd
, uint64_t object
,
209 void *ref
, dnode_t
**dnp
);
210 int dnode_hold_impl(struct objset_impl
*dd
, uint64_t object
, int flag
,
211 void *ref
, dnode_t
**dnp
);
212 boolean_t
dnode_add_ref(dnode_t
*dn
, void *ref
);
213 void dnode_rele(dnode_t
*dn
, void *ref
);
214 void dnode_setdirty(dnode_t
*dn
, dmu_tx_t
*tx
);
215 void dnode_sync(dnode_t
*dn
, dmu_tx_t
*tx
);
216 void dnode_allocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
, int ibs
,
217 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
);
218 void dnode_reallocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
,
219 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
);
220 void dnode_free(dnode_t
*dn
, dmu_tx_t
*tx
);
221 void dnode_byteswap(dnode_phys_t
*dnp
);
222 void dnode_buf_byteswap(void *buf
, size_t size
);
223 void dnode_verify(dnode_t
*dn
);
224 int dnode_set_blksz(dnode_t
*dn
, uint64_t size
, int ibs
, dmu_tx_t
*tx
);
225 uint64_t dnode_current_max_length(dnode_t
*dn
);
226 void dnode_free_range(dnode_t
*dn
, uint64_t off
, uint64_t len
, dmu_tx_t
*tx
);
227 void dnode_clear_range(dnode_t
*dn
, uint64_t blkid
,
228 uint64_t nblks
, dmu_tx_t
*tx
);
229 void dnode_diduse_space(dnode_t
*dn
, int64_t space
);
230 void dnode_willuse_space(dnode_t
*dn
, int64_t space
, dmu_tx_t
*tx
);
231 void dnode_new_blkid(dnode_t
*dn
, uint64_t blkid
, dmu_tx_t
*tx
, boolean_t
);
232 uint64_t dnode_block_freed(dnode_t
*dn
, uint64_t blkid
);
233 void dnode_init(void);
234 void dnode_fini(void);
235 int dnode_next_offset(dnode_t
*dn
, int flags
, uint64_t *off
,
236 int minlvl
, uint64_t blkfill
, uint64_t txg
);
237 void dnode_evict_dbufs(dnode_t
*dn
);
242 * There should be a ## between the string literal and fmt, to make it
243 * clear that we're joining two strings together, but that piece of shit
244 * gcc doesn't support that preprocessor token.
246 #define dprintf_dnode(dn, fmt, ...) do { \
247 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
249 uint64_t __db_obj = (dn)->dn_object; \
250 if (__db_obj == DMU_META_DNODE_OBJECT) \
251 (void) strcpy(__db_buf, "mdn"); \
253 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
254 (u_longlong_t)__db_obj);\
255 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \
256 __db_buf, __VA_ARGS__); \
258 _NOTE(CONSTCOND) } while (0)
260 #define DNODE_VERIFY(dn) dnode_verify(dn)
261 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx)
265 #define dprintf_dnode(db, fmt, ...)
266 #define DNODE_VERIFY(dn)
267 #define FREE_VERIFY(db, start, end, tx)
275 #endif /* _SYS_DNODE_H */