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.
28 #include <sys/zfs_context.h>
33 #include <sys/refcount.h>
34 #include <sys/dmu_zfetch.h>
43 #define DNODE_MUST_BE_ALLOCATED 1
44 #define DNODE_MUST_BE_FREE 2
47 * dnode_next_offset() flags.
49 #define DNODE_FIND_HOLE 1
50 #define DNODE_FIND_BACKWARDS 2
51 #define DNODE_FIND_HAVELOCK 4
56 #define DNODE_SHIFT 9 /* 512 bytes */
57 #define DN_MIN_INDBLKSHIFT 10 /* 1k */
58 #define DN_MAX_INDBLKSHIFT 14 /* 16k */
59 #define DNODE_BLOCK_SHIFT 14 /* 16k */
60 #define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */
61 #define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */
62 #define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */
67 * Note: a file will never ever have its
68 * ids moved from bonus->spill
69 * and only in a crypto environment would it be on spill
71 #define DN_ID_CHKED_BONUS 0x1
72 #define DN_ID_CHKED_SPILL 0x2
73 #define DN_ID_OLD_EXIST 0x4
74 #define DN_ID_NEW_EXIST 0x8
79 #define DNODE_SIZE (1 << DNODE_SHIFT)
80 #define DN_MAX_NBLKPTR ((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
81 #define DN_MAX_BONUSLEN (DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT))
82 #define DN_MAX_OBJECT (1ULL << DN_MAX_OBJECT_SHIFT)
83 #define DN_ZERO_BONUSLEN (DN_MAX_BONUSLEN + 1)
84 #define DN_KILL_SPILLBLK (1)
86 #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT)
87 #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT)
88 #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT)
89 #define DNODES_PER_LEVEL (1ULL << DNODES_PER_LEVEL_SHIFT)
91 /* The +2 here is a cheesy way to round up */
92 #define DN_MAX_LEVELS (2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \
93 (DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT)))
95 #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \
96 (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t))))
98 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \
99 (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT)
101 #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift))
107 enum dnode_dirtycontext {
113 /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */
114 #define DNODE_FLAG_USED_BYTES (1<<0)
115 #define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1)
117 /* Does dnode have a SA spill blkptr in bonus? */
118 #define DNODE_FLAG_SPILL_BLKPTR (1<<2)
120 typedef struct dnode_phys {
121 uint8_t dn_type; /* dmu_object_type_t */
122 uint8_t dn_indblkshift; /* ln2(indirect block size) */
123 uint8_t dn_nlevels; /* 1=dn_blkptr->data blocks */
124 uint8_t dn_nblkptr; /* length of dn_blkptr */
125 uint8_t dn_bonustype; /* type of data in bonus buffer */
126 uint8_t dn_checksum; /* ZIO_CHECKSUM type */
127 uint8_t dn_compress; /* ZIO_COMPRESS type */
128 uint8_t dn_flags; /* DNODE_FLAG_* */
129 uint16_t dn_datablkszsec; /* data block size in 512b sectors */
130 uint16_t dn_bonuslen; /* length of dn_bonus */
133 /* accounting is protected by dn_dirty_mtx */
134 uint64_t dn_maxblkid; /* largest allocated block ID */
135 uint64_t dn_used; /* bytes (or sectors) of disk space */
139 blkptr_t dn_blkptr[1];
140 uint8_t dn_bonus[DN_MAX_BONUSLEN - sizeof (blkptr_t)];
144 typedef struct dnode {
146 * dn_struct_rwlock protects the structure of the dnode,
147 * including the number of levels of indirection (dn_nlevels),
148 * dn_maxblkid, and dn_next_*
150 krwlock_t dn_struct_rwlock;
152 /* Our link on dn_objset->os_dnodes list; protected by os_lock. */
156 struct objset *dn_objset;
158 struct dmu_buf_impl *dn_dbuf;
159 dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */
162 * Copies of stuff in dn_phys. They're valid in the open
163 * context (eg. even before the dnode is first synced).
164 * Where necessary, these are protected by dn_struct_rwlock.
166 dmu_object_type_t dn_type; /* object type */
167 uint16_t dn_bonuslen; /* bonus length */
168 uint8_t dn_bonustype; /* bonus type */
169 uint8_t dn_nblkptr; /* number of blkptrs (immutable) */
170 uint8_t dn_checksum; /* ZIO_CHECKSUM type */
171 uint8_t dn_compress; /* ZIO_COMPRESS type */
173 uint8_t dn_indblkshift;
174 uint8_t dn_datablkshift; /* zero if blksz not power of 2! */
175 uint16_t dn_datablkszsec; /* in 512b sectors */
176 uint32_t dn_datablksz; /* in bytes */
177 uint64_t dn_maxblkid;
178 uint8_t dn_next_nblkptr[TXG_SIZE];
179 uint8_t dn_next_nlevels[TXG_SIZE];
180 uint8_t dn_next_indblkshift[TXG_SIZE];
181 uint8_t dn_next_bonustype[TXG_SIZE];
182 uint8_t dn_rm_spillblk[TXG_SIZE]; /* for removing spill blk */
183 uint16_t dn_next_bonuslen[TXG_SIZE];
184 uint32_t dn_next_blksz[TXG_SIZE]; /* next block size in bytes */
186 /* protected by os_lock: */
187 list_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */
189 /* protected by dn_mtx: */
191 list_t dn_dirty_records[TXG_SIZE];
192 avl_tree_t dn_ranges[TXG_SIZE];
193 uint64_t dn_allocated_txg;
194 uint64_t dn_free_txg;
195 uint64_t dn_assigned_txg;
196 kcondvar_t dn_notxholds;
197 enum dnode_dirtycontext dn_dirtyctx;
198 uint8_t *dn_dirtyctx_firstset; /* dbg: contents meaningless */
200 /* protected by own devices */
201 refcount_t dn_tx_holds;
204 kmutex_t dn_dbufs_mtx;
205 list_t dn_dbufs; /* linked list of descendent dbuf_t's */
206 struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */
207 boolean_t dn_have_spill; /* have spill or are spilling */
209 /* parent IO for current sync write */
212 /* used in syncing context */
213 uint64_t dn_oldused; /* old phys used bytes */
214 uint64_t dn_oldflags; /* old phys dn_flags */
215 uint64_t dn_olduid, dn_oldgid;
216 uint64_t dn_newuid, dn_newgid;
219 /* holds prefetch structure */
220 struct zfetch dn_zfetch;
223 typedef struct free_range {
229 dnode_t *dnode_special_open(struct objset *dd, dnode_phys_t *dnp,
231 void dnode_special_close(dnode_t *dn);
233 void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx);
234 void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx);
235 void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx);
237 int dnode_hold(struct objset *dd, uint64_t object,
238 void *ref, dnode_t **dnp);
239 int dnode_hold_impl(struct objset *dd, uint64_t object, int flag,
240 void *ref, dnode_t **dnp);
241 boolean_t dnode_add_ref(dnode_t *dn, void *ref);
242 void dnode_rele(dnode_t *dn, void *ref);
243 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx);
244 void dnode_sync(dnode_t *dn, dmu_tx_t *tx);
245 void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
246 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
247 void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
248 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
249 void dnode_free(dnode_t *dn, dmu_tx_t *tx);
250 void dnode_byteswap(dnode_phys_t *dnp);
251 void dnode_buf_byteswap(void *buf, size_t size);
252 void dnode_verify(dnode_t *dn);
253 int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx);
254 uint64_t dnode_current_max_length(dnode_t *dn);
255 void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx);
256 void dnode_clear_range(dnode_t *dn, uint64_t blkid,
257 uint64_t nblks, dmu_tx_t *tx);
258 void dnode_diduse_space(dnode_t *dn, int64_t space);
259 void dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx);
260 void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t);
261 uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid);
262 void dnode_init(void);
263 void dnode_fini(void);
264 int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off,
265 int minlvl, uint64_t blkfill, uint64_t txg);
266 void dnode_evict_dbufs(dnode_t *dn);
271 * There should be a ## between the string literal and fmt, to make it
272 * clear that we're joining two strings together, but that piece of shit
273 * gcc doesn't support that preprocessor token.
275 #define dprintf_dnode(dn, fmt, ...) do { \
276 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
278 uint64_t __db_obj = (dn)->dn_object; \
279 if (__db_obj == DMU_META_DNODE_OBJECT) \
280 (void) strcpy(__db_buf, "mdn"); \
282 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
283 (u_longlong_t)__db_obj);\
284 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \
285 __db_buf, __VA_ARGS__); \
287 _NOTE(CONSTCOND) } while (0)
289 #define DNODE_VERIFY(dn) dnode_verify(dn)
290 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx)
294 #define dprintf_dnode(db, fmt, ...)
295 #define DNODE_VERIFY(dn)
296 #define FREE_VERIFY(db, start, end, tx)
304 #endif /* _SYS_DNODE_H */