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, 2014 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 10 /* 1k */
61 #define DN_MAX_INDBLKSHIFT 14 /* 16k */
62 #define DNODE_BLOCK_SHIFT 14 /* 16k */
63 #define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */
64 #define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */
65 #define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */
70 * Note: a file will never ever have its
71 * ids moved from bonus->spill
72 * and only in a crypto environment would it be on spill
74 #define DN_ID_CHKED_BONUS 0x1
75 #define DN_ID_CHKED_SPILL 0x2
76 #define DN_ID_OLD_EXIST 0x4
77 #define DN_ID_NEW_EXIST 0x8
82 #define DNODE_SIZE (1 << DNODE_SHIFT)
83 #define DN_MAX_NBLKPTR ((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
84 #define DN_MAX_BONUSLEN (DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT))
85 #define DN_MAX_OBJECT (1ULL << DN_MAX_OBJECT_SHIFT)
86 #define DN_ZERO_BONUSLEN (DN_MAX_BONUSLEN + 1)
87 #define DN_KILL_SPILLBLK (1)
89 #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT)
90 #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT)
91 #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT)
92 #define DNODES_PER_LEVEL (1ULL << DNODES_PER_LEVEL_SHIFT)
94 /* The +2 here is a cheesy way to round up */
95 #define DN_MAX_LEVELS (2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \
96 (DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT)))
98 #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \
99 (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t))))
101 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \
102 (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT)
104 #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift))
110 enum dnode_dirtycontext
{
116 /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */
117 #define DNODE_FLAG_USED_BYTES (1<<0)
118 #define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1)
120 /* Does dnode have a SA spill blkptr in bonus? */
121 #define DNODE_FLAG_SPILL_BLKPTR (1<<2)
123 typedef struct dnode_phys
{
124 uint8_t dn_type
; /* dmu_object_type_t */
125 uint8_t dn_indblkshift
; /* ln2(indirect block size) */
126 uint8_t dn_nlevels
; /* 1=dn_blkptr->data blocks */
127 uint8_t dn_nblkptr
; /* length of dn_blkptr */
128 uint8_t dn_bonustype
; /* type of data in bonus buffer */
129 uint8_t dn_checksum
; /* ZIO_CHECKSUM type */
130 uint8_t dn_compress
; /* ZIO_COMPRESS type */
131 uint8_t dn_flags
; /* DNODE_FLAG_* */
132 uint16_t dn_datablkszsec
; /* data block size in 512b sectors */
133 uint16_t dn_bonuslen
; /* length of dn_bonus */
136 /* accounting is protected by dn_dirty_mtx */
137 uint64_t dn_maxblkid
; /* largest allocated block ID */
138 uint64_t dn_used
; /* bytes (or sectors) of disk space */
143 * The tail region is 448 bytes, and there are three ways to
146 * 0 64 128 192 256 320 384 448 (offset)
147 * +---------------+---------------+---------------+-------+
148 * | dn_blkptr[0] | dn_blkptr[1] | dn_blkptr[2] | / |
149 * +---------------+---------------+---------------+-------+
150 * | dn_blkptr[0] | dn_bonus[0..319] |
151 * +---------------+-----------------------+---------------+
152 * | dn_blkptr[0] | / | dn_spill |
153 * +---------------+-----------------------+---------------+
156 blkptr_t dn_blkptr
[1+DN_MAX_BONUSLEN
/sizeof (blkptr_t
)];
158 blkptr_t __dn_ignore1
;
159 uint8_t dn_bonus
[DN_MAX_BONUSLEN
];
162 blkptr_t __dn_ignore2
;
163 uint8_t __dn_ignore3
[DN_MAX_BONUSLEN
-sizeof (blkptr_t
)];
169 typedef struct dnode
{
171 * Protects the structure of the dnode, including the number of levels
172 * of indirection (dn_nlevels), dn_maxblkid, and dn_next_*
174 krwlock_t dn_struct_rwlock
;
176 /* Our link on dn_objset->os_dnodes list; protected by os_lock. */
180 struct objset
*dn_objset
;
182 struct dmu_buf_impl
*dn_dbuf
;
183 struct dnode_handle
*dn_handle
;
184 dnode_phys_t
*dn_phys
; /* pointer into dn->dn_dbuf->db.db_data */
187 * Copies of stuff in dn_phys. They're valid in the open
188 * context (eg. even before the dnode is first synced).
189 * Where necessary, these are protected by dn_struct_rwlock.
191 dmu_object_type_t dn_type
; /* object type */
192 uint16_t dn_bonuslen
; /* bonus length */
193 uint8_t dn_bonustype
; /* bonus type */
194 uint8_t dn_nblkptr
; /* number of blkptrs (immutable) */
195 uint8_t dn_checksum
; /* ZIO_CHECKSUM type */
196 uint8_t dn_compress
; /* ZIO_COMPRESS type */
198 uint8_t dn_indblkshift
;
199 uint8_t dn_datablkshift
; /* zero if blksz not power of 2! */
200 uint8_t dn_moved
; /* Has this dnode been moved? */
201 uint16_t dn_datablkszsec
; /* in 512b sectors */
202 uint32_t dn_datablksz
; /* in bytes */
203 uint64_t dn_maxblkid
;
204 uint8_t dn_next_type
[TXG_SIZE
];
205 uint8_t dn_next_nblkptr
[TXG_SIZE
];
206 uint8_t dn_next_nlevels
[TXG_SIZE
];
207 uint8_t dn_next_indblkshift
[TXG_SIZE
];
208 uint8_t dn_next_bonustype
[TXG_SIZE
];
209 uint8_t dn_rm_spillblk
[TXG_SIZE
]; /* for removing spill blk */
210 uint16_t dn_next_bonuslen
[TXG_SIZE
];
211 uint32_t dn_next_blksz
[TXG_SIZE
]; /* next block size in bytes */
213 /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */
214 uint32_t dn_dbufs_count
; /* count of dn_dbufs */
215 /* There are no level-0 blocks of this blkid or higher in dn_dbufs */
216 uint64_t dn_unlisted_l0_blkid
;
218 /* protected by os_lock: */
219 list_node_t dn_dirty_link
[TXG_SIZE
]; /* next on dataset's dirty */
221 /* protected by dn_mtx: */
223 list_t dn_dirty_records
[TXG_SIZE
];
224 struct range_tree
*dn_free_ranges
[TXG_SIZE
];
225 uint64_t dn_allocated_txg
;
226 uint64_t dn_free_txg
;
227 uint64_t dn_assigned_txg
;
228 kcondvar_t dn_notxholds
;
229 enum dnode_dirtycontext dn_dirtyctx
;
230 uint8_t *dn_dirtyctx_firstset
; /* dbg: contents meaningless */
232 /* protected by own devices */
233 refcount_t dn_tx_holds
;
236 kmutex_t dn_dbufs_mtx
;
238 * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs
239 * can contain multiple dbufs of the same (level, blkid) when a
240 * dbuf is marked DB_EVICTING without being removed from
241 * dn_dbufs. To maintain the avl invariant that there cannot be
242 * duplicate entries, we order the dbufs by an arbitrary value -
243 * their address in memory. This means that dn_dbufs cannot be used to
244 * directly look up a dbuf. Instead, callers must use avl_walk, have
245 * a reference to the dbuf, or look up a non-existant node with
246 * db_state = DB_SEARCH (see dbuf_free_range for an example).
250 /* protected by dn_struct_rwlock */
251 struct dmu_buf_impl
*dn_bonus
; /* bonus buffer dbuf */
253 boolean_t dn_have_spill
; /* have spill or are spilling */
255 /* parent IO for current sync write */
258 /* used in syncing context */
259 uint64_t dn_oldused
; /* old phys used bytes */
260 uint64_t dn_oldflags
; /* old phys dn_flags */
261 uint64_t dn_olduid
, dn_oldgid
;
262 uint64_t dn_newuid
, dn_newgid
;
265 /* holds prefetch structure */
266 struct zfetch dn_zfetch
;
270 * Adds a level of indirection between the dbuf and the dnode to avoid
271 * iterating descendent dbufs in dnode_move(). Handles are not allocated
272 * individually, but as an array of child dnodes in dnode_hold_impl().
274 typedef struct dnode_handle
{
275 /* Protects dnh_dnode from modification by dnode_move(). */
280 typedef struct dnode_children
{
281 dmu_buf_user_t dnc_dbu
; /* User evict data */
282 size_t dnc_count
; /* number of children */
283 dnode_handle_t dnc_children
[]; /* sized dynamically */
286 typedef struct free_range
{
292 void dnode_special_open(struct objset
*dd
, dnode_phys_t
*dnp
,
293 uint64_t object
, dnode_handle_t
*dnh
);
294 void dnode_special_close(dnode_handle_t
*dnh
);
296 void dnode_setbonuslen(dnode_t
*dn
, int newsize
, dmu_tx_t
*tx
);
297 void dnode_setbonus_type(dnode_t
*dn
, dmu_object_type_t
, dmu_tx_t
*tx
);
298 void dnode_rm_spill(dnode_t
*dn
, dmu_tx_t
*tx
);
300 int dnode_hold(struct objset
*dd
, uint64_t object
,
301 void *ref
, dnode_t
**dnp
);
302 int dnode_hold_impl(struct objset
*dd
, uint64_t object
, int flag
,
303 void *ref
, dnode_t
**dnp
);
304 boolean_t
dnode_add_ref(dnode_t
*dn
, void *ref
);
305 void dnode_rele(dnode_t
*dn
, void *ref
);
306 void dnode_rele_and_unlock(dnode_t
*dn
, void *tag
);
307 void dnode_setdirty(dnode_t
*dn
, dmu_tx_t
*tx
);
308 void dnode_sync(dnode_t
*dn
, dmu_tx_t
*tx
);
309 void dnode_allocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
, int ibs
,
310 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
);
311 void dnode_reallocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
,
312 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
);
313 void dnode_free(dnode_t
*dn
, dmu_tx_t
*tx
);
314 void dnode_byteswap(dnode_phys_t
*dnp
);
315 void dnode_buf_byteswap(void *buf
, size_t size
);
316 void dnode_verify(dnode_t
*dn
);
317 int dnode_set_blksz(dnode_t
*dn
, uint64_t size
, int ibs
, dmu_tx_t
*tx
);
318 void dnode_free_range(dnode_t
*dn
, uint64_t off
, uint64_t len
, dmu_tx_t
*tx
);
319 void dnode_diduse_space(dnode_t
*dn
, int64_t space
);
320 void dnode_willuse_space(dnode_t
*dn
, int64_t space
, dmu_tx_t
*tx
);
321 void dnode_new_blkid(dnode_t
*dn
, uint64_t blkid
, dmu_tx_t
*tx
, boolean_t
);
322 uint64_t dnode_block_freed(dnode_t
*dn
, uint64_t blkid
);
323 void dnode_init(void);
324 void dnode_fini(void);
325 int dnode_next_offset(dnode_t
*dn
, int flags
, uint64_t *off
,
326 int minlvl
, uint64_t blkfill
, uint64_t txg
);
327 void dnode_evict_dbufs(dnode_t
*dn
);
328 void dnode_evict_bonus(dnode_t
*dn
);
333 * There should be a ## between the string literal and fmt, to make it
334 * clear that we're joining two strings together, but that piece of shit
335 * gcc doesn't support that preprocessor token.
337 #define dprintf_dnode(dn, fmt, ...) do { \
338 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
340 uint64_t __db_obj = (dn)->dn_object; \
341 if (__db_obj == DMU_META_DNODE_OBJECT) \
342 (void) strcpy(__db_buf, "mdn"); \
344 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
345 (u_longlong_t)__db_obj);\
346 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \
347 __db_buf, __VA_ARGS__); \
349 _NOTE(CONSTCOND) } while (0)
351 #define DNODE_VERIFY(dn) dnode_verify(dn)
352 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx)
356 #define dprintf_dnode(db, fmt, ...)
357 #define DNODE_VERIFY(dn)
358 #define FREE_VERIFY(db, start, end, tx)
366 #endif /* _SYS_DNODE_H */