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.
23 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
30 #include <sys/zfs_context.h>
35 #include <sys/refcount.h>
36 #include <sys/dmu_zfetch.h>
37 #include <sys/zrlock.h>
38 #include <sys/multilist.h>
47 #define DNODE_MUST_BE_ALLOCATED 1
48 #define DNODE_MUST_BE_FREE 2
51 * dnode_next_offset() flags.
53 #define DNODE_FIND_HOLE 1
54 #define DNODE_FIND_BACKWARDS 2
55 #define DNODE_FIND_HAVELOCK 4
60 #define DNODE_SHIFT 9 /* 512 bytes */
61 #define DN_MIN_INDBLKSHIFT 12 /* 4k */
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.
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 */
77 * Note: a file will never ever have its ids moved from bonus->spill
79 #define DN_ID_CHKED_BONUS 0x1
80 #define DN_ID_CHKED_SPILL 0x2
81 #define DN_ID_OLD_EXIST 0x4
82 #define DN_ID_NEW_EXIST 0x8
87 #define DNODE_MIN_SIZE (1 << DNODE_SHIFT)
88 #define DNODE_MAX_SIZE (1 << DNODE_BLOCK_SHIFT)
89 #define DNODE_BLOCK_SIZE (1 << DNODE_BLOCK_SHIFT)
90 #define DNODE_MIN_SLOTS (DNODE_MIN_SIZE >> DNODE_SHIFT)
91 #define DNODE_MAX_SLOTS (DNODE_MAX_SIZE >> DNODE_SHIFT)
92 #define DN_BONUS_SIZE(dnsize) ((dnsize) - DNODE_CORE_SIZE - \
93 (1 << SPA_BLKPTRSHIFT))
94 #define DN_SLOTS_TO_BONUSLEN(slots) DN_BONUS_SIZE((slots) << DNODE_SHIFT)
95 #define DN_OLD_MAX_BONUSLEN (DN_BONUS_SIZE(DNODE_MIN_SIZE))
96 #define DN_MAX_NBLKPTR ((DNODE_MIN_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
97 #define DN_MAX_OBJECT (1ULL << DN_MAX_OBJECT_SHIFT)
98 #define DN_ZERO_BONUSLEN (DN_BONUS_SIZE(DNODE_MAX_SIZE) + 1)
99 #define DN_KILL_SPILLBLK (1)
101 #define DN_SLOT_UNINIT ((void *)NULL) /* Uninitialized */
102 #define DN_SLOT_FREE ((void *)1UL) /* Free slot */
103 #define DN_SLOT_ALLOCATED ((void *)2UL) /* Allocated slot */
104 #define DN_SLOT_INTERIOR ((void *)3UL) /* Interior allocated slot */
105 #define DN_SLOT_IS_PTR(dn) ((void *)dn > DN_SLOT_INTERIOR)
106 #define DN_SLOT_IS_VALID(dn) ((void *)dn != NULL)
108 #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT)
109 #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT)
112 * This is inaccurate if the indblkshift of the particular object is not the
113 * max. But it's only used by userland to calculate the zvol reservation.
115 #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT)
116 #define DNODES_PER_LEVEL (1ULL << DNODES_PER_LEVEL_SHIFT)
118 #define DN_MAX_LEVELS (DIV_ROUND_UP(DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT, \
119 DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT) + 1)
121 #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \
122 (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t))))
123 #define DN_MAX_BONUS_LEN(dnp) \
124 ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ? \
125 (uint8_t *)DN_SPILL_BLKPTR(dnp) - (uint8_t *)DN_BONUS(dnp) : \
126 (uint8_t *)(dnp + (dnp->dn_extra_slots + 1)) - (uint8_t *)DN_BONUS(dnp))
128 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \
129 (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT)
131 #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift))
137 enum dnode_dirtycontext
{
143 /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */
144 #define DNODE_FLAG_USED_BYTES (1 << 0)
145 #define DNODE_FLAG_USERUSED_ACCOUNTED (1 << 1)
147 /* Does dnode have a SA spill blkptr in bonus? */
148 #define DNODE_FLAG_SPILL_BLKPTR (1 << 2)
150 /* User/Group/Project dnode accounting */
151 #define DNODE_FLAG_USEROBJUSED_ACCOUNTED (1 << 3)
153 #define DNODE_CRYPT_PORTABLE_FLAGS_MASK (DNODE_FLAG_SPILL_BLKPTR)
156 * VARIABLE-LENGTH (LARGE) DNODES
158 * The motivation for variable-length dnodes is to eliminate the overhead
159 * associated with using spill blocks. Spill blocks are used to store
160 * system attribute data (i.e. file metadata) that does not fit in the
161 * dnode's bonus buffer. By allowing a larger bonus buffer area the use of
162 * a spill block can be avoided. Spill blocks potentially incur an
163 * additional read I/O for every dnode in a dnode block. As a worst case
164 * example, reading 32 dnodes from a 16k dnode block and all of the spill
165 * blocks could issue 33 separate reads. Now suppose those dnodes have size
166 * 1024 and therefore don't need spill blocks. Then the worst case number
167 * of blocks read is reduced to from 33 to two--one per dnode block.
169 * ZFS-on-Linux systems that make heavy use of extended attributes benefit
170 * from this feature. In particular, ZFS-on-Linux supports the xattr=sa
171 * dataset property which allows file extended attribute data to be stored
172 * in the dnode bonus buffer as an alternative to the traditional
173 * directory-based format. Workloads such as SELinux and the Lustre
174 * distributed filesystem often store enough xattr data to force spill
175 * blocks when xattr=sa is in effect. Large dnodes may therefore provide a
176 * performance benefit to such systems. Other use cases that benefit from
177 * this feature include files with large ACLs and symbolic links with long
180 * The size of a dnode may be a multiple of 512 bytes up to the size of a
181 * dnode block (currently 16384 bytes). The dn_extra_slots field of the
182 * on-disk dnode_phys_t structure describes the size of the physical dnode
183 * on disk. The field represents how many "extra" dnode_phys_t slots a
184 * dnode consumes in its dnode block. This convention results in a value of
185 * 0 for 512 byte dnodes which preserves on-disk format compatibility with
186 * older software which doesn't support large dnodes.
188 * Similarly, the in-memory dnode_t structure has a dn_num_slots field
189 * to represent the total number of dnode_phys_t slots consumed on disk.
190 * Thus dn->dn_num_slots is 1 greater than the corresponding
191 * dnp->dn_extra_slots. This difference in convention was adopted
192 * because, unlike on-disk structures, backward compatibility is not a
193 * concern for in-memory objects, so we used a more natural way to
194 * represent size for a dnode_t.
196 * The default size for newly created dnodes is determined by the value of
197 * the "dnodesize" dataset property. By default the property is set to
198 * "legacy" which is compatible with older software. Setting the property
199 * to "auto" will allow the filesystem to choose the most suitable dnode
200 * size. Currently this just sets the default dnode size to 1k, but future
201 * code improvements could dynamically choose a size based on observed
202 * workload patterns. Dnodes of varying sizes can coexist within the same
203 * dataset and even within the same dnode block.
206 typedef struct dnode_phys
{
207 uint8_t dn_type
; /* dmu_object_type_t */
208 uint8_t dn_indblkshift
; /* ln2(indirect block size) */
209 uint8_t dn_nlevels
; /* 1=dn_blkptr->data blocks */
210 uint8_t dn_nblkptr
; /* length of dn_blkptr */
211 uint8_t dn_bonustype
; /* type of data in bonus buffer */
212 uint8_t dn_checksum
; /* ZIO_CHECKSUM type */
213 uint8_t dn_compress
; /* ZIO_COMPRESS type */
214 uint8_t dn_flags
; /* DNODE_FLAG_* */
215 uint16_t dn_datablkszsec
; /* data block size in 512b sectors */
216 uint16_t dn_bonuslen
; /* length of dn_bonus */
217 uint8_t dn_extra_slots
; /* # of subsequent slots consumed */
220 /* accounting is protected by dn_dirty_mtx */
221 uint64_t dn_maxblkid
; /* largest allocated block ID */
222 uint64_t dn_used
; /* bytes (or sectors) of disk space */
225 * Both dn_pad2 and dn_pad3 are protected by the block's MAC. This
226 * allows us to protect any fields that might be added here in the
227 * future. In either case, developers will want to check
228 * zio_crypt_init_uios_dnode() to ensure the new field is being
229 * protected properly.
234 * The tail region is 448 bytes for a 512 byte dnode, and
235 * correspondingly larger for larger dnode sizes. The spill
236 * block pointer, when present, is always at the end of the tail
237 * region. There are three ways this space may be used, using
238 * a 512 byte dnode for this diagram:
240 * 0 64 128 192 256 320 384 448 (offset)
241 * +---------------+---------------+---------------+-------+
242 * | dn_blkptr[0] | dn_blkptr[1] | dn_blkptr[2] | / |
243 * +---------------+---------------+---------------+-------+
244 * | dn_blkptr[0] | dn_bonus[0..319] |
245 * +---------------+-----------------------+---------------+
246 * | dn_blkptr[0] | dn_bonus[0..191] | dn_spill |
247 * +---------------+-----------------------+---------------+
250 blkptr_t dn_blkptr
[1+DN_OLD_MAX_BONUSLEN
/sizeof (blkptr_t
)];
252 blkptr_t __dn_ignore1
;
253 uint8_t dn_bonus
[DN_OLD_MAX_BONUSLEN
];
256 blkptr_t __dn_ignore2
;
257 uint8_t __dn_ignore3
[DN_OLD_MAX_BONUSLEN
-
264 #define DN_SPILL_BLKPTR(dnp) (blkptr_t *)((char *)(dnp) + \
265 (((dnp)->dn_extra_slots + 1) << DNODE_SHIFT) - (1 << SPA_BLKPTRSHIFT))
269 * Protects the structure of the dnode, including the number of levels
270 * of indirection (dn_nlevels), dn_maxblkid, and dn_next_*
272 krwlock_t dn_struct_rwlock
;
274 /* Our link on dn_objset->os_dnodes list; protected by os_lock. */
278 struct objset
*dn_objset
;
280 struct dmu_buf_impl
*dn_dbuf
;
281 struct dnode_handle
*dn_handle
;
282 dnode_phys_t
*dn_phys
; /* pointer into dn->dn_dbuf->db.db_data */
285 * Copies of stuff in dn_phys. They're valid in the open
286 * context (eg. even before the dnode is first synced).
287 * Where necessary, these are protected by dn_struct_rwlock.
289 dmu_object_type_t dn_type
; /* object type */
290 uint16_t dn_bonuslen
; /* bonus length */
291 uint8_t dn_bonustype
; /* bonus type */
292 uint8_t dn_nblkptr
; /* number of blkptrs (immutable) */
293 uint8_t dn_checksum
; /* ZIO_CHECKSUM type */
294 uint8_t dn_compress
; /* ZIO_COMPRESS type */
296 uint8_t dn_indblkshift
;
297 uint8_t dn_datablkshift
; /* zero if blksz not power of 2! */
298 uint8_t dn_moved
; /* Has this dnode been moved? */
299 uint16_t dn_datablkszsec
; /* in 512b sectors */
300 uint32_t dn_datablksz
; /* in bytes */
301 uint64_t dn_maxblkid
;
302 uint8_t dn_next_type
[TXG_SIZE
];
303 uint8_t dn_num_slots
; /* metadnode slots consumed on disk */
304 uint8_t dn_next_nblkptr
[TXG_SIZE
];
305 uint8_t dn_next_nlevels
[TXG_SIZE
];
306 uint8_t dn_next_indblkshift
[TXG_SIZE
];
307 uint8_t dn_next_bonustype
[TXG_SIZE
];
308 uint8_t dn_rm_spillblk
[TXG_SIZE
]; /* for removing spill blk */
309 uint16_t dn_next_bonuslen
[TXG_SIZE
];
310 uint32_t dn_next_blksz
[TXG_SIZE
]; /* next block size in bytes */
311 uint64_t dn_next_maxblkid
[TXG_SIZE
]; /* next maxblkid in bytes */
313 /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */
314 uint32_t dn_dbufs_count
; /* count of dn_dbufs */
316 /* protected by os_lock: */
317 multilist_node_t dn_dirty_link
[TXG_SIZE
]; /* next on dataset's dirty */
319 /* protected by dn_mtx: */
321 list_t dn_dirty_records
[TXG_SIZE
];
322 struct range_tree
*dn_free_ranges
[TXG_SIZE
];
323 uint64_t dn_allocated_txg
;
324 uint64_t dn_free_txg
;
325 uint64_t dn_assigned_txg
;
326 kcondvar_t dn_notxholds
;
327 enum dnode_dirtycontext dn_dirtyctx
;
328 uint8_t *dn_dirtyctx_firstset
; /* dbg: contents meaningless */
330 /* protected by own devices */
331 refcount_t dn_tx_holds
;
334 kmutex_t dn_dbufs_mtx
;
336 * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs
337 * can contain multiple dbufs of the same (level, blkid) when a
338 * dbuf is marked DB_EVICTING without being removed from
339 * dn_dbufs. To maintain the avl invariant that there cannot be
340 * duplicate entries, we order the dbufs by an arbitrary value -
341 * their address in memory. This means that dn_dbufs cannot be used to
342 * directly look up a dbuf. Instead, callers must use avl_walk, have
343 * a reference to the dbuf, or look up a non-existent node with
344 * db_state = DB_SEARCH (see dbuf_free_range for an example).
348 /* protected by dn_struct_rwlock */
349 struct dmu_buf_impl
*dn_bonus
; /* bonus buffer dbuf */
351 boolean_t dn_have_spill
; /* have spill or are spilling */
353 /* parent IO for current sync write */
356 /* used in syncing context */
357 uint64_t dn_oldused
; /* old phys used bytes */
358 uint64_t dn_oldflags
; /* old phys dn_flags */
359 uint64_t dn_olduid
, dn_oldgid
, dn_oldprojid
;
360 uint64_t dn_newuid
, dn_newgid
, dn_newprojid
;
363 /* holds prefetch structure */
364 struct zfetch dn_zfetch
;
368 * Adds a level of indirection between the dbuf and the dnode to avoid
369 * iterating descendent dbufs in dnode_move(). Handles are not allocated
370 * individually, but as an array of child dnodes in dnode_hold_impl().
372 typedef struct dnode_handle
{
373 /* Protects dnh_dnode from modification by dnode_move(). */
378 typedef struct dnode_children
{
379 dmu_buf_user_t dnc_dbu
; /* User evict data */
380 size_t dnc_count
; /* number of children */
381 dnode_handle_t dnc_children
[]; /* sized dynamically */
384 typedef struct free_range
{
390 void dnode_special_open(struct objset
*dd
, dnode_phys_t
*dnp
,
391 uint64_t object
, dnode_handle_t
*dnh
);
392 void dnode_special_close(dnode_handle_t
*dnh
);
394 void dnode_setbonuslen(dnode_t
*dn
, int newsize
, dmu_tx_t
*tx
);
395 void dnode_setbonus_type(dnode_t
*dn
, dmu_object_type_t
, dmu_tx_t
*tx
);
396 void dnode_rm_spill(dnode_t
*dn
, dmu_tx_t
*tx
);
398 int dnode_hold(struct objset
*dd
, uint64_t object
,
399 void *ref
, dnode_t
**dnp
);
400 int dnode_hold_impl(struct objset
*dd
, uint64_t object
, int flag
, int dn_slots
,
401 void *ref
, dnode_t
**dnp
);
402 boolean_t
dnode_add_ref(dnode_t
*dn
, void *ref
);
403 void dnode_rele(dnode_t
*dn
, void *ref
);
404 void dnode_rele_and_unlock(dnode_t
*dn
, void *tag
);
405 void dnode_setdirty(dnode_t
*dn
, dmu_tx_t
*tx
);
406 void dnode_sync(dnode_t
*dn
, dmu_tx_t
*tx
);
407 void dnode_allocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
, int ibs
,
408 dmu_object_type_t bonustype
, int bonuslen
, int dn_slots
, dmu_tx_t
*tx
);
409 void dnode_reallocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
,
410 dmu_object_type_t bonustype
, int bonuslen
, int dn_slots
, dmu_tx_t
*tx
);
411 void dnode_free(dnode_t
*dn
, dmu_tx_t
*tx
);
412 void dnode_byteswap(dnode_phys_t
*dnp
);
413 void dnode_buf_byteswap(void *buf
, size_t size
);
414 void dnode_verify(dnode_t
*dn
);
415 int dnode_set_nlevels(dnode_t
*dn
, int nlevels
, dmu_tx_t
*tx
);
416 int dnode_set_blksz(dnode_t
*dn
, uint64_t size
, int ibs
, dmu_tx_t
*tx
);
417 void dnode_free_range(dnode_t
*dn
, uint64_t off
, uint64_t len
, dmu_tx_t
*tx
);
418 void dnode_diduse_space(dnode_t
*dn
, int64_t space
);
419 void dnode_new_blkid(dnode_t
*dn
, uint64_t blkid
, dmu_tx_t
*tx
, boolean_t
);
420 uint64_t dnode_block_freed(dnode_t
*dn
, uint64_t blkid
);
421 void dnode_init(void);
422 void dnode_fini(void);
423 int dnode_next_offset(dnode_t
*dn
, int flags
, uint64_t *off
,
424 int minlvl
, uint64_t blkfill
, uint64_t txg
);
425 void dnode_evict_dbufs(dnode_t
*dn
);
426 void dnode_evict_bonus(dnode_t
*dn
);
427 void dnode_free_interior_slots(dnode_t
*dn
);
429 #define DNODE_IS_CACHEABLE(_dn) \
430 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \
431 (DMU_OT_IS_METADATA((_dn)->dn_type) && \
432 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA))
434 #define DNODE_META_IS_CACHEABLE(_dn) \
435 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \
436 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)
439 * Used for dnodestats kstat.
441 typedef struct dnode_stats
{
443 * Number of failed attempts to hold a meta dnode dbuf.
445 kstat_named_t dnode_hold_dbuf_hold
;
447 * Number of failed attempts to read a meta dnode dbuf.
449 kstat_named_t dnode_hold_dbuf_read
;
451 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was able
452 * to hold the requested object number which was allocated. This is
453 * the common case when looking up any allocated object number.
455 kstat_named_t dnode_hold_alloc_hits
;
457 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was not
458 * able to hold the request object number because it was not allocated.
460 kstat_named_t dnode_hold_alloc_misses
;
462 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was not
463 * able to hold the request object number because the object number
464 * refers to an interior large dnode slot.
466 kstat_named_t dnode_hold_alloc_interior
;
468 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) needed
469 * to retry acquiring slot zrl locks due to contention.
471 kstat_named_t dnode_hold_alloc_lock_retry
;
473 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) did not
474 * need to create the dnode because another thread did so after
475 * dropping the read lock but before acquiring the write lock.
477 kstat_named_t dnode_hold_alloc_lock_misses
;
479 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) found
480 * a free dnode instantiated by dnode_create() but not yet allocated
481 * by dnode_allocate().
483 kstat_named_t dnode_hold_alloc_type_none
;
485 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was able
486 * to hold the requested range of free dnode slots.
488 kstat_named_t dnode_hold_free_hits
;
490 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was not
491 * able to hold the requested range of free dnode slots because
492 * at least one slot was allocated.
494 kstat_named_t dnode_hold_free_misses
;
496 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was not
497 * able to hold the requested range of free dnode slots because
498 * after acquiring the zrl lock at least one slot was allocated.
500 kstat_named_t dnode_hold_free_lock_misses
;
502 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) needed
503 * to retry acquiring slot zrl locks due to contention.
505 kstat_named_t dnode_hold_free_lock_retry
;
507 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) requested
508 * a range of dnode slots which were held by another thread.
510 kstat_named_t dnode_hold_free_refcount
;
512 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) requested
513 * a range of dnode slots which would overflow the dnode_phys_t.
515 kstat_named_t dnode_hold_free_overflow
;
517 * Number of times a dnode_hold(...) was attempted on a dnode
518 * which had already been unlinked in an earlier txg.
520 kstat_named_t dnode_hold_free_txg
;
522 * Number of times dnode_free_interior_slots() needed to retry
523 * acquiring a slot zrl lock due to contention.
525 kstat_named_t dnode_free_interior_lock_retry
;
527 * Number of new dnodes allocated by dnode_allocate().
529 kstat_named_t dnode_allocate
;
531 * Number of dnodes re-allocated by dnode_reallocate().
533 kstat_named_t dnode_reallocate
;
535 * Number of meta dnode dbufs evicted.
537 kstat_named_t dnode_buf_evict
;
539 * Number of times dmu_object_alloc*() reached the end of the existing
540 * object ID chunk and advanced to a new one.
542 kstat_named_t dnode_alloc_next_chunk
;
544 * Number of times multiple threads attempted to allocate a dnode
545 * from the same block of free dnodes.
547 kstat_named_t dnode_alloc_race
;
549 * Number of times dmu_object_alloc*() was forced to advance to the
550 * next meta dnode dbuf due to an error from dmu_object_next().
552 kstat_named_t dnode_alloc_next_block
;
554 * Statistics for tracking dnodes which have been moved.
556 kstat_named_t dnode_move_invalid
;
557 kstat_named_t dnode_move_recheck1
;
558 kstat_named_t dnode_move_recheck2
;
559 kstat_named_t dnode_move_special
;
560 kstat_named_t dnode_move_handle
;
561 kstat_named_t dnode_move_rwlock
;
562 kstat_named_t dnode_move_active
;
565 extern dnode_stats_t dnode_stats
;
567 #define DNODE_STAT_INCR(stat, val) \
568 atomic_add_64(&dnode_stats.stat.value.ui64, (val));
569 #define DNODE_STAT_BUMP(stat) \
570 DNODE_STAT_INCR(stat, 1);
575 * There should be a ## between the string literal and fmt, to make it
576 * clear that we're joining two strings together, but that piece of shit
577 * gcc doesn't support that preprocessor token.
579 #define dprintf_dnode(dn, fmt, ...) do { \
580 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
582 uint64_t __db_obj = (dn)->dn_object; \
583 if (__db_obj == DMU_META_DNODE_OBJECT) \
584 (void) strcpy(__db_buf, "mdn"); \
586 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
587 (u_longlong_t)__db_obj);\
588 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \
589 __db_buf, __VA_ARGS__); \
591 _NOTE(CONSTCOND) } while (0)
593 #define DNODE_VERIFY(dn) dnode_verify(dn)
594 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx)
598 #define dprintf_dnode(db, fmt, ...)
599 #define DNODE_VERIFY(dn)
600 #define FREE_VERIFY(db, start, end, tx)
608 #endif /* _SYS_DNODE_H */