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 (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. |
25 | */ |
26 | |
27 | #ifndef _SYS_DNODE_H |
28 | #define _SYS_DNODE_H |
29 | |
30 | #include <sys/zfs_context.h> |
31 | #include <sys/avl.h> |
32 | #include <sys/spa.h> |
33 | #include <sys/txg.h> |
34 | #include <sys/zio.h> |
35 | #include <sys/refcount.h> |
36 | #include <sys/dmu_zfetch.h> |
37 | #include <sys/zrlock.h> |
38 | |
39 | #ifdef __cplusplus |
40 | extern "C" { |
41 | #endif |
42 | |
43 | /* |
44 | * dnode_hold() flags. |
45 | */ |
46 | #define DNODE_MUST_BE_ALLOCATED 1 |
47 | #define DNODE_MUST_BE_FREE 2 |
48 | |
49 | /* |
50 | * dnode_next_offset() flags. |
51 | */ |
52 | #define DNODE_FIND_HOLE 1 |
53 | #define DNODE_FIND_BACKWARDS 2 |
54 | #define DNODE_FIND_HAVELOCK 4 |
55 | |
56 | /* |
57 | * Fixed constants. |
58 | */ |
59 | #define DNODE_SHIFT 9 /* 512 bytes */ |
60 | #define DN_MIN_INDBLKSHIFT 12 /* 4k */ |
61 | #define DN_MAX_INDBLKSHIFT 17 /* 128k */ |
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 */ |
66 | |
67 | /* |
68 | * dnode id flags |
69 | * |
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 |
73 | */ |
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 |
78 | |
79 | /* |
80 | * Derived constants. |
81 | */ |
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) |
88 | |
89 | #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT) |
90 | #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT) |
91 | |
92 | /* |
93 | * This is inaccurate if the indblkshift of the particular object is not the |
94 | * max. But it's only used by userland to calculate the zvol reservation. |
95 | */ |
96 | #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT) |
97 | #define DNODES_PER_LEVEL (1ULL << DNODES_PER_LEVEL_SHIFT) |
98 | |
99 | /* The +2 here is a cheesy way to round up */ |
100 | #define DN_MAX_LEVELS (2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \ |
101 | (DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT))) |
102 | |
103 | #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \ |
104 | (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t)))) |
105 | |
106 | #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \ |
107 | (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT) |
108 | |
109 | #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift)) |
110 | |
111 | struct dmu_buf_impl; |
112 | struct objset; |
113 | struct zio; |
114 | |
115 | enum dnode_dirtycontext { |
116 | DN_UNDIRTIED, |
117 | DN_DIRTY_OPEN, |
118 | DN_DIRTY_SYNC |
119 | }; |
120 | |
121 | /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */ |
122 | #define DNODE_FLAG_USED_BYTES (1<<0) |
123 | #define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1) |
124 | |
125 | /* Does dnode have a SA spill blkptr in bonus? */ |
126 | #define DNODE_FLAG_SPILL_BLKPTR (1<<2) |
127 | |
128 | typedef struct dnode_phys { |
129 | uint8_t dn_type; /* dmu_object_type_t */ |
130 | uint8_t dn_indblkshift; /* ln2(indirect block size) */ |
131 | uint8_t dn_nlevels; /* 1=dn_blkptr->data blocks */ |
132 | uint8_t dn_nblkptr; /* length of dn_blkptr */ |
133 | uint8_t dn_bonustype; /* type of data in bonus buffer */ |
134 | uint8_t dn_checksum; /* ZIO_CHECKSUM type */ |
135 | uint8_t dn_compress; /* ZIO_COMPRESS type */ |
136 | uint8_t dn_flags; /* DNODE_FLAG_* */ |
137 | uint16_t dn_datablkszsec; /* data block size in 512b sectors */ |
138 | uint16_t dn_bonuslen; /* length of dn_bonus */ |
139 | uint8_t dn_pad2[4]; |
140 | |
141 | /* accounting is protected by dn_dirty_mtx */ |
142 | uint64_t dn_maxblkid; /* largest allocated block ID */ |
143 | uint64_t dn_used; /* bytes (or sectors) of disk space */ |
144 | |
145 | uint64_t dn_pad3[4]; |
146 | |
147 | blkptr_t dn_blkptr[1]; |
148 | uint8_t dn_bonus[DN_MAX_BONUSLEN - sizeof (blkptr_t)]; |
149 | blkptr_t dn_spill; |
150 | } dnode_phys_t; |
151 | |
152 | struct dnode { |
153 | /* |
154 | * Protects the structure of the dnode, including the number of levels |
155 | * of indirection (dn_nlevels), dn_maxblkid, and dn_next_* |
156 | */ |
157 | krwlock_t dn_struct_rwlock; |
158 | |
159 | /* Our link on dn_objset->os_dnodes list; protected by os_lock. */ |
160 | list_node_t dn_link; |
161 | |
162 | /* immutable: */ |
163 | struct objset *dn_objset; |
164 | uint64_t dn_object; |
165 | struct dmu_buf_impl *dn_dbuf; |
166 | struct dnode_handle *dn_handle; |
167 | dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */ |
168 | |
169 | /* |
170 | * Copies of stuff in dn_phys. They're valid in the open |
171 | * context (eg. even before the dnode is first synced). |
172 | * Where necessary, these are protected by dn_struct_rwlock. |
173 | */ |
174 | dmu_object_type_t dn_type; /* object type */ |
175 | uint16_t dn_bonuslen; /* bonus length */ |
176 | uint8_t dn_bonustype; /* bonus type */ |
177 | uint8_t dn_nblkptr; /* number of blkptrs (immutable) */ |
178 | uint8_t dn_checksum; /* ZIO_CHECKSUM type */ |
179 | uint8_t dn_compress; /* ZIO_COMPRESS type */ |
180 | uint8_t dn_nlevels; |
181 | uint8_t dn_indblkshift; |
182 | uint8_t dn_datablkshift; /* zero if blksz not power of 2! */ |
183 | uint8_t dn_moved; /* Has this dnode been moved? */ |
184 | uint16_t dn_datablkszsec; /* in 512b sectors */ |
185 | uint32_t dn_datablksz; /* in bytes */ |
186 | uint64_t dn_maxblkid; |
187 | uint8_t dn_next_type[TXG_SIZE]; |
188 | uint8_t dn_next_nblkptr[TXG_SIZE]; |
189 | uint8_t dn_next_nlevels[TXG_SIZE]; |
190 | uint8_t dn_next_indblkshift[TXG_SIZE]; |
191 | uint8_t dn_next_bonustype[TXG_SIZE]; |
192 | uint8_t dn_rm_spillblk[TXG_SIZE]; /* for removing spill blk */ |
193 | uint16_t dn_next_bonuslen[TXG_SIZE]; |
194 | uint32_t dn_next_blksz[TXG_SIZE]; /* next block size in bytes */ |
195 | |
196 | /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */ |
197 | uint32_t dn_dbufs_count; /* count of dn_dbufs */ |
198 | |
199 | /* protected by os_lock: */ |
200 | list_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */ |
201 | |
202 | /* protected by dn_mtx: */ |
203 | kmutex_t dn_mtx; |
204 | list_t dn_dirty_records[TXG_SIZE]; |
205 | struct range_tree *dn_free_ranges[TXG_SIZE]; |
206 | uint64_t dn_allocated_txg; |
207 | uint64_t dn_free_txg; |
208 | uint64_t dn_assigned_txg; |
209 | kcondvar_t dn_notxholds; |
210 | enum dnode_dirtycontext dn_dirtyctx; |
211 | uint8_t *dn_dirtyctx_firstset; /* dbg: contents meaningless */ |
212 | |
213 | /* protected by own devices */ |
214 | refcount_t dn_tx_holds; |
215 | refcount_t dn_holds; |
216 | |
217 | kmutex_t dn_dbufs_mtx; |
218 | /* |
219 | * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs |
220 | * can contain multiple dbufs of the same (level, blkid) when a |
221 | * dbuf is marked DB_EVICTING without being removed from |
222 | * dn_dbufs. To maintain the avl invariant that there cannot be |
223 | * duplicate entries, we order the dbufs by an arbitrary value - |
224 | * their address in memory. This means that dn_dbufs cannot be used to |
225 | * directly look up a dbuf. Instead, callers must use avl_walk, have |
226 | * a reference to the dbuf, or look up a non-existant node with |
227 | * db_state = DB_SEARCH (see dbuf_free_range for an example). |
228 | */ |
229 | avl_tree_t dn_dbufs; |
230 | |
231 | /* protected by dn_struct_rwlock */ |
232 | struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */ |
233 | |
234 | boolean_t dn_have_spill; /* have spill or are spilling */ |
235 | |
236 | /* parent IO for current sync write */ |
237 | zio_t *dn_zio; |
238 | |
239 | /* used in syncing context */ |
240 | uint64_t dn_oldused; /* old phys used bytes */ |
241 | uint64_t dn_oldflags; /* old phys dn_flags */ |
242 | uint64_t dn_olduid, dn_oldgid; |
243 | uint64_t dn_newuid, dn_newgid; |
244 | int dn_id_flags; |
245 | |
246 | /* holds prefetch structure */ |
247 | struct zfetch dn_zfetch; |
248 | }; |
249 | |
250 | /* |
251 | * Adds a level of indirection between the dbuf and the dnode to avoid |
252 | * iterating descendent dbufs in dnode_move(). Handles are not allocated |
253 | * individually, but as an array of child dnodes in dnode_hold_impl(). |
254 | */ |
255 | typedef struct dnode_handle { |
256 | /* Protects dnh_dnode from modification by dnode_move(). */ |
257 | zrlock_t dnh_zrlock; |
258 | dnode_t *dnh_dnode; |
259 | } dnode_handle_t; |
260 | |
261 | typedef struct dnode_children { |
262 | dmu_buf_user_t dnc_dbu; /* User evict data */ |
263 | size_t dnc_count; /* number of children */ |
264 | dnode_handle_t dnc_children[]; /* sized dynamically */ |
265 | } dnode_children_t; |
266 | |
267 | typedef struct free_range { |
268 | avl_node_t fr_node; |
269 | uint64_t fr_blkid; |
270 | uint64_t fr_nblks; |
271 | } free_range_t; |
272 | |
273 | void dnode_special_open(struct objset *dd, dnode_phys_t *dnp, |
274 | uint64_t object, dnode_handle_t *dnh); |
275 | void dnode_special_close(dnode_handle_t *dnh); |
276 | |
277 | void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx); |
278 | void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx); |
279 | void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx); |
280 | |
281 | int dnode_hold(struct objset *dd, uint64_t object, |
282 | void *ref, dnode_t **dnp); |
283 | int dnode_hold_impl(struct objset *dd, uint64_t object, int flag, |
284 | void *ref, dnode_t **dnp); |
285 | boolean_t dnode_add_ref(dnode_t *dn, void *ref); |
286 | void dnode_rele(dnode_t *dn, void *ref); |
287 | void dnode_rele_and_unlock(dnode_t *dn, void *tag); |
288 | void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx); |
289 | void dnode_sync(dnode_t *dn, dmu_tx_t *tx); |
290 | void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, |
291 | dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx); |
292 | void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, |
293 | dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx); |
294 | void dnode_free(dnode_t *dn, dmu_tx_t *tx); |
295 | void dnode_byteswap(dnode_phys_t *dnp); |
296 | void dnode_buf_byteswap(void *buf, size_t size); |
297 | void dnode_verify(dnode_t *dn); |
298 | int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx); |
299 | void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx); |
300 | void dnode_diduse_space(dnode_t *dn, int64_t space); |
301 | void dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx); |
302 | void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t); |
303 | uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid); |
304 | void dnode_init(void); |
305 | void dnode_fini(void); |
306 | int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off, |
307 | int minlvl, uint64_t blkfill, uint64_t txg); |
308 | void dnode_evict_dbufs(dnode_t *dn); |
309 | void dnode_evict_bonus(dnode_t *dn); |
310 | |
311 | #define DNODE_IS_CACHEABLE(_dn) \ |
312 | ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \ |
313 | (DMU_OT_IS_METADATA((_dn)->dn_type) && \ |
314 | (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)) |
315 | |
316 | #define DNODE_META_IS_CACHEABLE(_dn) \ |
317 | ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \ |
318 | (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA) |
319 | |
320 | #ifdef ZFS_DEBUG |
321 | |
322 | /* |
323 | * There should be a ## between the string literal and fmt, to make it |
324 | * clear that we're joining two strings together, but that piece of shit |
325 | * gcc doesn't support that preprocessor token. |
326 | */ |
327 | #define dprintf_dnode(dn, fmt, ...) do { \ |
328 | if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ |
329 | char __db_buf[32]; \ |
330 | uint64_t __db_obj = (dn)->dn_object; \ |
331 | if (__db_obj == DMU_META_DNODE_OBJECT) \ |
332 | (void) strcpy(__db_buf, "mdn"); \ |
333 | else \ |
334 | (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ |
335 | (u_longlong_t)__db_obj);\ |
336 | dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \ |
337 | __db_buf, __VA_ARGS__); \ |
338 | } \ |
339 | _NOTE(CONSTCOND) } while (0) |
340 | |
341 | #define DNODE_VERIFY(dn) dnode_verify(dn) |
342 | #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx) |
343 | |
344 | #else |
345 | |
346 | #define dprintf_dnode(db, fmt, ...) |
347 | #define DNODE_VERIFY(dn) |
348 | #define FREE_VERIFY(db, start, end, tx) |
349 | |
350 | #endif |
351 | |
352 | #ifdef __cplusplus |
353 | } |
354 | #endif |
355 | |
356 | #endif /* _SYS_DNODE_H */ |
357 | |