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, 2017 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 #include <sys/multilist.h>
39
40 #ifdef __cplusplus
41 extern "C" {
42 #endif
43
44 /*
45 * dnode_hold() flags.
46 */
47 #define DNODE_MUST_BE_ALLOCATED 1
48 #define DNODE_MUST_BE_FREE 2
49
50 /*
51 * dnode_next_offset() flags.
52 */
53 #define DNODE_FIND_HOLE 1
54 #define DNODE_FIND_BACKWARDS 2
55 #define DNODE_FIND_HAVELOCK 4
56
57 /*
58 * Fixed constants.
59 */
60 #define DNODE_SHIFT 9 /* 512 bytes */
61 #define DN_MIN_INDBLKSHIFT 12 /* 4k */
62 /*
63 * If we ever increase this value beyond 20, we need to revisit all logic that
64 * does x << level * ebps to handle overflow. With a 1M indirect block size,
65 * 4 levels of indirect blocks would not be able to guarantee addressing an
66 * entire object, so 5 levels will be used, but 5 * (20 - 7) = 65.
67 */
68 #define DN_MAX_INDBLKSHIFT 17 /* 128k */
69 #define DNODE_BLOCK_SHIFT 14 /* 16k */
70 #define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */
71 #define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */
72 #define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */
73
74 /*
75 * dnode id flags
76 *
77 * Note: a file will never ever have its
78 * ids moved from bonus->spill
79 * and only in a crypto environment would it be on spill
80 */
81 #define DN_ID_CHKED_BONUS 0x1
115
116 #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift))
117
118 struct dmu_buf_impl;
119 struct objset;
120 struct zio;
121
122 enum dnode_dirtycontext {
123 DN_UNDIRTIED,
124 DN_DIRTY_OPEN,
125 DN_DIRTY_SYNC
126 };
127
128 /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */
129 #define DNODE_FLAG_USED_BYTES (1<<0)
130 #define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1)
131
132 /* Does dnode have a SA spill blkptr in bonus? */
133 #define DNODE_FLAG_SPILL_BLKPTR (1<<2)
134
135 typedef struct dnode_phys {
136 uint8_t dn_type; /* dmu_object_type_t */
137 uint8_t dn_indblkshift; /* ln2(indirect block size) */
138 uint8_t dn_nlevels; /* 1=dn_blkptr->data blocks */
139 uint8_t dn_nblkptr; /* length of dn_blkptr */
140 uint8_t dn_bonustype; /* type of data in bonus buffer */
141 uint8_t dn_checksum; /* ZIO_CHECKSUM type */
142 uint8_t dn_compress; /* ZIO_COMPRESS type */
143 uint8_t dn_flags; /* DNODE_FLAG_* */
144 uint16_t dn_datablkszsec; /* data block size in 512b sectors */
145 uint16_t dn_bonuslen; /* length of dn_bonus */
146 uint8_t dn_pad2[4];
147
148 /* accounting is protected by dn_dirty_mtx */
149 uint64_t dn_maxblkid; /* largest allocated block ID */
150 uint64_t dn_used; /* bytes (or sectors) of disk space */
151
152 uint64_t dn_pad3[4];
153
154 blkptr_t dn_blkptr[1];
235 */
236 avl_tree_t dn_dbufs;
237
238 /* protected by dn_struct_rwlock */
239 struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */
240
241 boolean_t dn_have_spill; /* have spill or are spilling */
242
243 /* parent IO for current sync write */
244 zio_t *dn_zio;
245
246 /* used in syncing context */
247 uint64_t dn_oldused; /* old phys used bytes */
248 uint64_t dn_oldflags; /* old phys dn_flags */
249 uint64_t dn_olduid, dn_oldgid;
250 uint64_t dn_newuid, dn_newgid;
251 int dn_id_flags;
252
253 /* holds prefetch structure */
254 struct zfetch dn_zfetch;
255 };
256
257 /*
258 * Adds a level of indirection between the dbuf and the dnode to avoid
259 * iterating descendent dbufs in dnode_move(). Handles are not allocated
260 * individually, but as an array of child dnodes in dnode_hold_impl().
261 */
262 typedef struct dnode_handle {
263 /* Protects dnh_dnode from modification by dnode_move(). */
264 zrlock_t dnh_zrlock;
265 dnode_t *dnh_dnode;
266 } dnode_handle_t;
267
268 typedef struct dnode_children {
269 dmu_buf_user_t dnc_dbu; /* User evict data */
270 size_t dnc_count; /* number of children */
271 dnode_handle_t dnc_children[]; /* sized dynamically */
272 } dnode_children_t;
273
274 typedef struct free_range {
276 uint64_t fr_blkid;
277 uint64_t fr_nblks;
278 } free_range_t;
279
280 void dnode_special_open(struct objset *dd, dnode_phys_t *dnp,
281 uint64_t object, dnode_handle_t *dnh);
282 void dnode_special_close(dnode_handle_t *dnh);
283
284 void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx);
285 void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx);
286 void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx);
287
288 int dnode_hold(struct objset *dd, uint64_t object,
289 void *ref, dnode_t **dnp);
290 int dnode_hold_impl(struct objset *dd, uint64_t object, int flag,
291 void *ref, dnode_t **dnp);
292 boolean_t dnode_add_ref(dnode_t *dn, void *ref);
293 void dnode_rele(dnode_t *dn, void *ref);
294 void dnode_rele_and_unlock(dnode_t *dn, void *tag);
295 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx);
296 void dnode_sync(dnode_t *dn, dmu_tx_t *tx);
297 void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
298 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
299 void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
300 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
301 void dnode_free(dnode_t *dn, dmu_tx_t *tx);
302 void dnode_byteswap(dnode_phys_t *dnp);
303 void dnode_buf_byteswap(void *buf, size_t size);
304 void dnode_verify(dnode_t *dn);
305 int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx);
306 void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx);
307 void dnode_diduse_space(dnode_t *dn, int64_t space);
308 void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t);
309 uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid);
310 void dnode_init(void);
311 void dnode_fini(void);
312 int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off,
313 int minlvl, uint64_t blkfill, uint64_t txg);
314 void dnode_evict_dbufs(dnode_t *dn);
315 void dnode_evict_bonus(dnode_t *dn);
316 boolean_t dnode_needs_remap(const dnode_t *dn);
317
318 #define DNODE_IS_CACHEABLE(_dn) \
319 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \
320 (DMU_OT_IS_METADATA((_dn)->dn_type) && \
321 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA))
322
323 #define DNODE_META_IS_CACHEABLE(_dn) \
324 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \
325 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)
326
327 #ifdef ZFS_DEBUG
328
329 /*
330 * There should be a ## between the string literal and fmt, to make it
331 * clear that we're joining two strings together, but that piece of shit
332 * gcc doesn't support that preprocessor token.
333 */
334 #define dprintf_dnode(dn, fmt, ...) do { \
335 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
336 char __db_buf[32]; \
337 uint64_t __db_obj = (dn)->dn_object; \
|
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
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];
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 {
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; \
|