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 2015 Nexenta Systems, Inc. All rights reserved.
24 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 */
27
28 #ifndef _SYS_DNODE_H
29 #define _SYS_DNODE_H
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/zio.h>
36 #include <sys/refcount.h>
37 #include <sys/dmu_zfetch.h>
38 #include <sys/zrlock.h>
39 #include <sys/multilist.h>
40
41 #ifdef __cplusplus
42 extern "C" {
43 #endif
44
45 /*
46 * dnode_hold() flags.
47 */
48 #define DNODE_MUST_BE_ALLOCATED 1
49 #define DNODE_MUST_BE_FREE 2
50
51 /*
52 * dnode_next_offset() flags.
53 */
54 #define DNODE_FIND_HOLE 1
55 #define DNODE_FIND_BACKWARDS 2
56 #define DNODE_FIND_HAVELOCK 4
57
58 /*
59 * Fixed constants.
60 */
61 #define DNODE_SHIFT 9 /* 512 bytes */
62 #define DN_MIN_INDBLKSHIFT 12 /* 4k */
63 #define DN_DFL_INDBLKSHIFT 14 /* 16k */
64 /*
65 * If we ever increase this value beyond 20, we need to revisit all logic that
66 * does x << level * ebps to handle overflow. With a 1M indirect block size,
67 * 4 levels of indirect blocks would not be able to guarantee addressing an
68 * entire object, so 5 levels will be used, but 5 * (20 - 7) = 65.
69 */
70 #define DN_MAX_INDBLKSHIFT 17 /* 128k */
71 #define DNODE_BLOCK_SHIFT 14 /* 16k */
72 #define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */
73 #define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */
74 #define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */
75
76 /*
77 * dnode id flags
78 *
79 * Note: a file will never ever have its
80 * ids moved from bonus->spill
81 * and only in a crypto environment would it be on spill
82 */
83 #define DN_ID_CHKED_BONUS 0x1
84 #define DN_ID_CHKED_SPILL 0x2
85 #define DN_ID_OLD_EXIST 0x4
86 #define DN_ID_NEW_EXIST 0x8
87
88 /*
89 * Derived constants.
90 */
91 #define DNODE_SIZE (1 << DNODE_SHIFT)
92 #define DN_MAX_NBLKPTR ((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
93 #define DN_MAX_BONUSLEN (DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT))
94 #define DN_MAX_OBJECT (1ULL << DN_MAX_OBJECT_SHIFT)
95 #define DN_ZERO_BONUSLEN (DN_MAX_BONUSLEN + 1)
96 #define DN_KILL_SPILLBLK (1)
97
98 #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT)
99 #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT)
100
101 /*
102 * This is inaccurate if the indblkshift of the particular object is not the
103 * max. But it's only used by userland to calculate the zvol reservation.
104 */
105 #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT)
106 #define DNODES_PER_LEVEL (1ULL << DNODES_PER_LEVEL_SHIFT)
107
108 /* The +2 here is a cheesy way to round up */
109 #define DN_MAX_LEVELS (2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \
110 (DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT)))
111
112 #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \
113 (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t))))
114
115 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \
116 (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT)
117
118 #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift))
119
120 struct dmu_buf_impl;
121 struct objset;
122 struct zio;
123
124 enum dnode_dirtycontext {
125 DN_UNDIRTIED,
126 DN_DIRTY_OPEN,
127 DN_DIRTY_SYNC
128 };
129
130 /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */
131 #define DNODE_FLAG_USED_BYTES (1<<0)
132 #define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1)
133
134 /* Does dnode have a SA spill blkptr in bonus? */
135 #define DNODE_FLAG_SPILL_BLKPTR (1<<2)
136
137 /*
138 * Smart Compression
139 *
140 * Smart compression is a simple heuristic algorithm that automatically
141 * tries to avoid compression on incompressible objects by continuously
142 * monitoring per-object compression performance.
143 *
144 * The smart compression system has two states for each object:
145 * COMPRESSING: compression is applied to the object and results are
146 * reevaluated at fixed check intervals to see if we should
147 * continue attempting to compress.
148 * DENYING: compression has failed too often, so we give up trying
149 * for a while and retry at a later time.
150 *
151 * Each time compression succeeds, we bump down the sc_compression_failures
152 * counter down to a minimum of -5. This helps us to prevent starting
153 * denying in case of short incompressible transients. Each time
154 * compression fails, we bump up the sc_compression_failures counter up to
155 * a maximum of 5 (zfs_smartcomp_interval_shift). If the failure counter
156 * is > 0 at the time the compression failed, we transition from the
157 * COMPRESSING to the DENYING state and calculate a deny interval by
158 * multiplying zfs_smartcomp_interval and (1 << sc_compression_failures).
159 * This means that successive failures at compression will have us retry
160 * compression progressively less often down to approx 32x less often
161 * (by default) than the compression test interval. To avoid potential
162 * data patterns confusing us, the deny interval is randomized by +-10%.
163 */
164 typedef enum dnode_smartcomp_state {
165 DNODE_SMARTCOMP_COMPRESSING = 0,
166 DNODE_SMARTCOMP_DENYING
167 } dnode_smartcomp_state_t;
168
169 typedef struct dnode_smartcomp {
170 kmutex_t sc_lock;
171 dnode_smartcomp_state_t sc_state;
172 uint64_t sc_size;
173 uint64_t sc_orig_size;
174 uint64_t sc_deny_interval;
175 int64_t sc_comp_failures;
176 } dnode_smartcomp_t;
177
178 typedef struct dnode_phys {
179 uint8_t dn_type; /* dmu_object_type_t */
180 uint8_t dn_indblkshift; /* ln2(indirect block size) */
181 uint8_t dn_nlevels; /* 1=dn_blkptr->data blocks */
182 uint8_t dn_nblkptr; /* length of dn_blkptr */
183 uint8_t dn_bonustype; /* type of data in bonus buffer */
184 uint8_t dn_checksum; /* ZIO_CHECKSUM type */
185 uint8_t dn_compress; /* ZIO_COMPRESS type */
186 uint8_t dn_flags; /* DNODE_FLAG_* */
187 uint16_t dn_datablkszsec; /* data block size in 512b sectors */
188 uint16_t dn_bonuslen; /* length of dn_bonus */
189 uint8_t dn_pad2[4];
190
191 /* accounting is protected by dn_dirty_mtx */
192 uint64_t dn_maxblkid; /* largest allocated block ID */
193 uint64_t dn_used; /* bytes (or sectors) of disk space */
194
195 uint64_t dn_pad3[4];
196
197 blkptr_t dn_blkptr[1];
198 uint8_t dn_bonus[DN_MAX_BONUSLEN - sizeof (blkptr_t)];
199 blkptr_t dn_spill;
200 } dnode_phys_t;
201
202 struct dnode {
203 /*
204 * Protects the structure of the dnode, including the number of levels
205 * of indirection (dn_nlevels), dn_maxblkid, and dn_next_*
206 */
207 krwlock_t dn_struct_rwlock;
208
209 /* Our link on dn_objset->os_dnodes list; protected by os_lock. */
210 list_node_t dn_link;
211
212 /* immutable: */
213 struct objset *dn_objset;
214 uint64_t dn_object;
215 struct dmu_buf_impl *dn_dbuf;
216 struct dnode_handle *dn_handle;
217 dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */
218
219 /*
220 * Copies of stuff in dn_phys. They're valid in the open
221 * context (eg. even before the dnode is first synced).
222 * Where necessary, these are protected by dn_struct_rwlock.
223 */
224 dmu_object_type_t dn_type; /* object type */
225 uint16_t dn_bonuslen; /* bonus length */
226 uint8_t dn_bonustype; /* bonus type */
227 uint8_t dn_nblkptr; /* number of blkptrs (immutable) */
228 uint8_t dn_checksum; /* ZIO_CHECKSUM type */
229 uint8_t dn_compress; /* ZIO_COMPRESS type */
230 uint8_t dn_nlevels;
231 uint8_t dn_indblkshift;
232 uint8_t dn_datablkshift; /* zero if blksz not power of 2! */
233 uint8_t dn_moved; /* Has this dnode been moved? */
234 uint16_t dn_datablkszsec; /* in 512b sectors */
235 uint32_t dn_datablksz; /* in bytes */
236 uint64_t dn_maxblkid;
237 uint8_t dn_next_type[TXG_SIZE];
238 uint8_t dn_next_nblkptr[TXG_SIZE];
239 uint8_t dn_next_nlevels[TXG_SIZE];
240 uint8_t dn_next_indblkshift[TXG_SIZE];
241 uint8_t dn_next_bonustype[TXG_SIZE];
242 uint8_t dn_rm_spillblk[TXG_SIZE]; /* for removing spill blk */
243 uint16_t dn_next_bonuslen[TXG_SIZE];
244 uint32_t dn_next_blksz[TXG_SIZE]; /* next block size in bytes */
245
246 /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */
247 uint32_t dn_dbufs_count; /* count of dn_dbufs */
248
249 /* protected by os_lock: */
250 multilist_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */
251
252 /* protected by dn_mtx: */
253 kmutex_t dn_mtx;
254 list_t dn_dirty_records[TXG_SIZE];
255 struct range_tree *dn_free_ranges[TXG_SIZE];
256 uint64_t dn_allocated_txg;
257 uint64_t dn_free_txg;
258 uint64_t dn_assigned_txg;
259 kcondvar_t dn_notxholds;
260 enum dnode_dirtycontext dn_dirtyctx;
261 uint8_t *dn_dirtyctx_firstset; /* dbg: contents meaningless */
262
263 /* protected by own devices */
264 refcount_t dn_tx_holds;
265 refcount_t dn_holds;
266
267 kmutex_t dn_dbufs_mtx;
268 /*
269 * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs
270 * can contain multiple dbufs of the same (level, blkid) when a
271 * dbuf is marked DB_EVICTING without being removed from
272 * dn_dbufs. To maintain the avl invariant that there cannot be
273 * duplicate entries, we order the dbufs by an arbitrary value -
274 * their address in memory. This means that dn_dbufs cannot be used to
275 * directly look up a dbuf. Instead, callers must use avl_walk, have
276 * a reference to the dbuf, or look up a non-existant node with
277 * db_state = DB_SEARCH (see dbuf_free_range for an example).
278 */
279 avl_tree_t dn_dbufs;
280
281 /* protected by dn_struct_rwlock */
282 struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */
283
284 boolean_t dn_have_spill; /* have spill or are spilling */
285
286 /* parent IO for current sync write */
287 zio_t *dn_zio;
288
289 /* used in syncing context */
290 uint64_t dn_oldused; /* old phys used bytes */
291 uint64_t dn_oldflags; /* old phys dn_flags */
292 uint64_t dn_olduid, dn_oldgid;
293 uint64_t dn_newuid, dn_newgid;
294 int dn_id_flags;
295
296 /* holds prefetch structure */
297 struct zfetch dn_zfetch;
298
299 dnode_smartcomp_t dn_smartcomp; /* smart compression performance */
300 };
301
302 /*
303 * Adds a level of indirection between the dbuf and the dnode to avoid
304 * iterating descendent dbufs in dnode_move(). Handles are not allocated
305 * individually, but as an array of child dnodes in dnode_hold_impl().
306 */
307 typedef struct dnode_handle {
308 /* Protects dnh_dnode from modification by dnode_move(). */
309 zrlock_t dnh_zrlock;
310 dnode_t *dnh_dnode;
311 } dnode_handle_t;
312
313 typedef struct dnode_children {
314 dmu_buf_user_t dnc_dbu; /* User evict data */
315 size_t dnc_count; /* number of children */
316 dnode_handle_t dnc_children[]; /* sized dynamically */
317 } dnode_children_t;
318
319 typedef struct free_range {
320 avl_node_t fr_node;
321 uint64_t fr_blkid;
322 uint64_t fr_nblks;
323 } free_range_t;
324
325 void dnode_special_open(struct objset *dd, dnode_phys_t *dnp,
326 uint64_t object, dnode_handle_t *dnh);
327 void dnode_special_close(dnode_handle_t *dnh);
328
329 void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx);
330 void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx);
331 void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx);
332
333 int dnode_hold(struct objset *dd, uint64_t object,
334 void *ref, dnode_t **dnp);
335 int dnode_hold_impl(struct objset *dd, uint64_t object, int flag,
336 void *ref, dnode_t **dnp);
337 boolean_t dnode_add_ref(dnode_t *dn, void *ref);
338 void dnode_rele(dnode_t *dn, void *ref);
339 void dnode_rele_and_unlock(dnode_t *dn, void *tag);
340 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx);
341 void dnode_setdirty_sc(dnode_t *dn, dmu_tx_t *tx, boolean_t usesc);
342 void dnode_sync(dnode_t *dn, dmu_tx_t *tx);
343 void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
344 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
345 void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
346 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
347 void dnode_free(dnode_t *dn, dmu_tx_t *tx);
348 void dnode_byteswap(dnode_phys_t *dnp);
349 void dnode_buf_byteswap(void *buf, size_t size);
350 void dnode_verify(dnode_t *dn);
351 int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx);
352 void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx);
353 void dnode_diduse_space(dnode_t *dn, int64_t space);
354 void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx,
355 boolean_t usesc, boolean_t);
356 uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid);
357 void dnode_init(void);
358 void dnode_fini(void);
359 int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off,
360 int minlvl, uint64_t blkfill, uint64_t txg);
361 void dnode_evict_dbufs(dnode_t *dn, int level);
362 void dnode_evict_bonus(dnode_t *dn);
363
364 /* Smart compression callbacks (see zio_smartcomp_info_t). */
365 extern boolean_t dnode_smartcomp_ask_cb(void *userinfo, const zio_t *zio);
366 extern void dnode_smartcomp_result_cb(void *userinfo, const zio_t *zio);
367 extern void dnode_setup_zio_smartcomp(struct dmu_buf_impl *db,
368 zio_smartcomp_info_t *sc);
369
370 #define DNODE_IS_CACHEABLE(_dn) \
371 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \
372 (DMU_OT_IS_METADATA((_dn)->dn_type) && \
373 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA))
374
375 #define DNODE_META_IS_CACHEABLE(_dn) \
376 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \
377 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)
378
379 #ifdef ZFS_DEBUG
380
381 /*
382 * There should be a ## between the string literal and fmt, to make it
383 * clear that we're joining two strings together, but that piece of shit
384 * gcc doesn't support that preprocessor token.
385 */
386 #define dprintf_dnode(dn, fmt, ...) do { \
387 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
388 char __db_buf[32]; \
389 uint64_t __db_obj = (dn)->dn_object; \
390 if (__db_obj == DMU_META_DNODE_OBJECT) \
391 (void) strcpy(__db_buf, "mdn"); \
392 else \
393 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
394 (u_longlong_t)__db_obj);\
395 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \
396 __db_buf, __VA_ARGS__); \
397 } \
398 _NOTE(CONSTCOND) } while (0)
399
400 #define DNODE_VERIFY(dn) dnode_verify(dn)
401 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx)
402
403 #else
404
405 #define dprintf_dnode(db, fmt, ...)
406 #define DNODE_VERIFY(dn)
407 #define FREE_VERIFY(db, start, end, tx)
408
409 #endif
410
411 #ifdef __cplusplus
412 }
413 #endif
414
415 #endif /* _SYS_DNODE_H */