xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/dnode.h (revision 8a2f1b9190d1dc288470a1fd2776d79ce82cb129)
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 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #ifndef	_SYS_DNODE_H
27 #define	_SYS_DNODE_H
28 
29 #pragma ident	"%Z%%M%	%I%	%E% SMI"
30 
31 #include <sys/zfs_context.h>
32 #include <sys/avl.h>
33 #include <sys/spa.h>
34 #include <sys/txg.h>
35 #include <sys/refcount.h>
36 #include <sys/dmu_zfetch.h>
37 
38 #ifdef	__cplusplus
39 extern "C" {
40 #endif
41 
42 /*
43  * Flags.
44  */
45 #define	DNODE_MUST_BE_ALLOCATED	1
46 #define	DNODE_MUST_BE_FREE	2
47 
48 /*
49  * Fixed constants.
50  */
51 #define	DNODE_SHIFT		9	/* 512 bytes */
52 #define	DN_MIN_INDBLKSHIFT	10	/* 1k */
53 #define	DN_MAX_INDBLKSHIFT	14	/* 16k */
54 #define	DNODE_BLOCK_SHIFT	14	/* 16k */
55 #define	DNODE_CORE_SIZE		64	/* 64 bytes for dnode sans blkptrs */
56 #define	DN_MAX_OBJECT_SHIFT	48	/* 256 trillion (zfs_fid_t limit) */
57 #define	DN_MAX_OFFSET_SHIFT	64	/* 2^64 bytes in a dnode */
58 
59 /*
60  * Derived constants.
61  */
62 #define	DNODE_SIZE	(1 << DNODE_SHIFT)
63 #define	DN_MAX_NBLKPTR	((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
64 #define	DN_MAX_BONUSLEN	(DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT))
65 #define	DN_MAX_OBJECT	(1ULL << DN_MAX_OBJECT_SHIFT)
66 
67 #define	DNODES_PER_BLOCK_SHIFT	(DNODE_BLOCK_SHIFT - DNODE_SHIFT)
68 #define	DNODES_PER_BLOCK	(1ULL << DNODES_PER_BLOCK_SHIFT)
69 #define	DNODES_PER_LEVEL_SHIFT	(DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT)
70 
71 /* The +2 here is a cheesy way to round up */
72 #define	DN_MAX_LEVELS	(2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \
73 	(DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT)))
74 
75 #define	DN_BONUS(dnp)	((void*)((dnp)->dn_bonus + \
76 	(((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t))))
77 
78 #define	DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \
79 	(dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT)
80 
81 #define	EPB(blkshift, typeshift)	(1 << (blkshift - typeshift))
82 
83 struct dmu_buf_impl;
84 struct objset_impl;
85 struct zio;
86 
87 enum dnode_dirtycontext {
88 	DN_UNDIRTIED,
89 	DN_DIRTY_OPEN,
90 	DN_DIRTY_SYNC
91 };
92 
93 /* Is dn_used in bytes?  if not, it's in multiples of SPA_MINBLOCKSIZE */
94 #define	DNODE_FLAG_USED_BYTES	(1<<0)
95 
96 typedef struct dnode_phys {
97 	uint8_t dn_type;		/* dmu_object_type_t */
98 	uint8_t dn_indblkshift;		/* ln2(indirect block size) */
99 	uint8_t dn_nlevels;		/* 1=dn_blkptr->data blocks */
100 	uint8_t dn_nblkptr;		/* length of dn_blkptr */
101 	uint8_t dn_bonustype;		/* type of data in bonus buffer */
102 	uint8_t	dn_checksum;		/* ZIO_CHECKSUM type */
103 	uint8_t	dn_compress;		/* ZIO_COMPRESS type */
104 	uint8_t dn_flags;		/* DNODE_FLAG_* */
105 	uint16_t dn_datablkszsec;	/* data block size in 512b sectors */
106 	uint16_t dn_bonuslen;		/* length of dn_bonus */
107 	uint8_t dn_pad2[4];
108 
109 	/* accounting is protected by dn_dirty_mtx */
110 	uint64_t dn_maxblkid;		/* largest allocated block ID */
111 	uint64_t dn_used;		/* bytes (or sectors) of disk space */
112 
113 	uint64_t dn_pad3[4];
114 
115 	blkptr_t dn_blkptr[1];
116 	uint8_t dn_bonus[DN_MAX_BONUSLEN];
117 } dnode_phys_t;
118 
119 typedef struct dnode {
120 	/*
121 	 * dn_struct_rwlock protects the structure of the dnode,
122 	 * including the number of levels of indirection (dn_nlevels),
123 	 * dn_maxblkid, and dn_next_*
124 	 */
125 	krwlock_t dn_struct_rwlock;
126 
127 	/*
128 	 * Our link on dataset's dd_dnodes list.
129 	 * Protected by dd_accounting_mtx.
130 	 */
131 	list_node_t dn_link;
132 
133 	/* immutable: */
134 	struct objset_impl *dn_objset;
135 	uint64_t dn_object;
136 	struct dmu_buf_impl *dn_dbuf;
137 	dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */
138 
139 	/*
140 	 * Copies of stuff in dn_phys.  They're valid in the open
141 	 * context (eg. even before the dnode is first synced).
142 	 * Where necessary, these are protected by dn_struct_rwlock.
143 	 */
144 	dmu_object_type_t dn_type;	/* object type */
145 	uint16_t dn_bonuslen;		/* bonus length */
146 	uint8_t dn_bonustype;		/* bonus type */
147 	uint8_t dn_nblkptr;		/* number of blkptrs (immutable) */
148 	uint8_t dn_checksum;		/* ZIO_CHECKSUM type */
149 	uint8_t dn_compress;		/* ZIO_COMPRESS type */
150 	uint8_t dn_nlevels;
151 	uint8_t dn_indblkshift;
152 	uint8_t dn_datablkshift;	/* zero if blksz not power of 2! */
153 	uint16_t dn_datablkszsec;	/* in 512b sectors */
154 	uint32_t dn_datablksz;		/* in bytes */
155 	uint64_t dn_maxblkid;
156 	uint8_t dn_next_nlevels[TXG_SIZE];
157 	uint8_t dn_next_indblkshift[TXG_SIZE];
158 	uint32_t dn_next_blksz[TXG_SIZE];	/* next block size in bytes */
159 
160 	/* protected by os_lock: */
161 	list_node_t dn_dirty_link[TXG_SIZE];	/* next on dataset's dirty */
162 
163 	/* protected by dn_mtx: */
164 	kmutex_t dn_mtx;
165 	list_t dn_dirty_dbufs[TXG_SIZE];
166 	avl_tree_t dn_ranges[TXG_SIZE];
167 	uint64_t dn_allocated_txg;
168 	uint64_t dn_free_txg;
169 	uint64_t dn_assigned_txg;
170 	kcondvar_t dn_notxholds;
171 	enum dnode_dirtycontext dn_dirtyctx;
172 	uint8_t *dn_dirtyctx_firstset;		/* dbg: contents meaningless */
173 
174 	/* protected by own devices */
175 	refcount_t dn_tx_holds;
176 	refcount_t dn_holds;
177 
178 	kmutex_t dn_dbufs_mtx;
179 	list_t dn_dbufs;		/* linked list of descendent dbuf_t's */
180 	struct dmu_buf_impl *dn_bonus;	/* bonus buffer dbuf */
181 
182 	/* holds prefetch structure */
183 	struct zfetch	dn_zfetch;
184 } dnode_t;
185 
186 typedef struct free_range {
187 	avl_node_t fr_node;
188 	uint64_t fr_blkid;
189 	uint64_t fr_nblks;
190 } free_range_t;
191 
192 dnode_t *dnode_special_open(struct objset_impl *dd, dnode_phys_t *dnp,
193     uint64_t object);
194 void dnode_special_close(dnode_t *dn);
195 
196 int dnode_hold(struct objset_impl *dd, uint64_t object,
197     void *ref, dnode_t **dnp);
198 int dnode_hold_impl(struct objset_impl *dd, uint64_t object, int flag,
199     void *ref, dnode_t **dnp);
200 void dnode_add_ref(dnode_t *dn, void *ref);
201 void dnode_rele(dnode_t *dn, void *ref);
202 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx);
203 int dnode_sync(dnode_t *dn, int level, struct zio *zio, dmu_tx_t *tx);
204 void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
205     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
206 void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
207     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
208 void dnode_free(dnode_t *dn, dmu_tx_t *tx);
209 void dnode_byteswap(dnode_phys_t *dnp);
210 void dnode_buf_byteswap(void *buf, size_t size);
211 void dnode_verify(dnode_t *dn);
212 int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx);
213 uint64_t dnode_current_max_length(dnode_t *dn);
214 uint64_t dnode_max_nonzero_offset(dnode_t *dn);
215 void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx);
216 void dnode_clear_range(dnode_t *dn, uint64_t blkid,
217     uint64_t nblks, dmu_tx_t *tx);
218 void dnode_diduse_space(dnode_t *dn, int64_t space);
219 void dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx);
220 void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx);
221 uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid);
222 void dnode_init(void);
223 void dnode_fini(void);
224 int dnode_next_offset(dnode_t *dn, boolean_t hole, uint64_t *off, int minlvl,
225     uint64_t blkfill);
226 int dnode_evict_dbufs(dnode_t *dn, int try);
227 
228 #ifdef ZFS_DEBUG
229 
230 /*
231  * There should be a ## between the string literal and fmt, to make it
232  * clear that we're joining two strings together, but that piece of shit
233  * gcc doesn't support that preprocessor token.
234  */
235 #define	dprintf_dnode(dn, fmt, ...) do { \
236 	if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
237 	char __db_buf[32]; \
238 	uint64_t __db_obj = (dn)->dn_object; \
239 	if (__db_obj == DMU_META_DNODE_OBJECT) \
240 		(void) strcpy(__db_buf, "mdn"); \
241 	else \
242 		(void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
243 		    (u_longlong_t)__db_obj);\
244 	dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \
245 	    __db_buf, __VA_ARGS__); \
246 	} \
247 _NOTE(CONSTCOND) } while (0)
248 
249 #define	DNODE_VERIFY(dn)		dnode_verify(dn)
250 #define	FREE_VERIFY(db, start, end, tx)	free_verify(db, start, end, tx)
251 
252 #else
253 
254 #define	dprintf_dnode(db, fmt, ...)
255 #define	DNODE_VERIFY(dn)
256 #define	FREE_VERIFY(db, start, end, tx)
257 
258 #endif
259 
260 #ifdef	__cplusplus
261 }
262 #endif
263 
264 #endif	/* _SYS_DNODE_H */
265