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, 2018 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
41extern "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#define	DNODE_DRY_RUN		4
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/*
64 * If we ever increase this value beyond 20, we need to revisit all logic that
65 * does x << level * ebps to handle overflow.  With a 1M indirect block size,
66 * 4 levels of indirect blocks would not be able to guarantee addressing an
67 * entire object, so 5 levels will be used, but 5 * (20 - 7) = 65.
68 */
69#define	DN_MAX_INDBLKSHIFT	17	/* 128k */
70#define	DNODE_BLOCK_SHIFT	14	/* 16k */
71#define	DNODE_CORE_SIZE		64	/* 64 bytes for dnode sans blkptrs */
72#define	DN_MAX_OBJECT_SHIFT	48	/* 256 trillion (zfs_fid_t limit) */
73#define	DN_MAX_OFFSET_SHIFT	64	/* 2^64 bytes in a dnode */
74
75/*
76 * dnode id flags
77 *
78 * Note: a file will never ever have its ids moved from bonus->spill
79 */
80#define	DN_ID_CHKED_BONUS	0x1
81#define	DN_ID_CHKED_SPILL	0x2
82#define	DN_ID_OLD_EXIST		0x4
83#define	DN_ID_NEW_EXIST		0x8
84
85/*
86 * Derived constants.
87 */
88#define	DNODE_MIN_SIZE		(1 << DNODE_SHIFT)
89#define	DNODE_MAX_SIZE		(1 << DNODE_BLOCK_SHIFT)
90#define	DNODE_BLOCK_SIZE	(1 << DNODE_BLOCK_SHIFT)
91#define	DNODE_MIN_SLOTS		(DNODE_MIN_SIZE >> DNODE_SHIFT)
92#define	DNODE_MAX_SLOTS		(DNODE_MAX_SIZE >> DNODE_SHIFT)
93#define	DN_BONUS_SIZE(dnsize)	((dnsize) - DNODE_CORE_SIZE - \
94	(1 << SPA_BLKPTRSHIFT))
95#define	DN_SLOTS_TO_BONUSLEN(slots)	DN_BONUS_SIZE((slots) << DNODE_SHIFT)
96#define	DN_OLD_MAX_BONUSLEN	(DN_BONUS_SIZE(DNODE_MIN_SIZE))
97#define	DN_MAX_NBLKPTR	((DNODE_MIN_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
98#define	DN_MAX_OBJECT		(1ULL << DN_MAX_OBJECT_SHIFT)
99#define	DN_ZERO_BONUSLEN	(DN_BONUS_SIZE(DNODE_MAX_SIZE) + 1)
100#define	DN_KILL_SPILLBLK	(1)
101
102#define	DN_SLOT_UNINIT		((void *)NULL)	/* Uninitialized */
103#define	DN_SLOT_FREE		((void *)1UL)	/* Free slot */
104#define	DN_SLOT_ALLOCATED	((void *)2UL)	/* Allocated slot */
105#define	DN_SLOT_INTERIOR	((void *)3UL)	/* Interior allocated slot */
106#define	DN_SLOT_IS_PTR(dn)	((void *)dn > DN_SLOT_INTERIOR)
107#define	DN_SLOT_IS_VALID(dn)	((void *)dn != NULL)
108
109#define	DNODES_PER_BLOCK_SHIFT	(DNODE_BLOCK_SHIFT - DNODE_SHIFT)
110#define	DNODES_PER_BLOCK	(1ULL << DNODES_PER_BLOCK_SHIFT)
111
112/*
113 * This is inaccurate if the indblkshift of the particular object is not the
114 * max.  But it's only used by userland to calculate the zvol reservation.
115 */
116#define	DNODES_PER_LEVEL_SHIFT	(DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT)
117#define	DNODES_PER_LEVEL	(1ULL << DNODES_PER_LEVEL_SHIFT)
118
119/* The +2 here is a cheesy way to round up */
120#define	DN_MAX_LEVELS	(2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \
121	(DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT)))
122
123#define	DN_BONUS(dnp)	((void*)((dnp)->dn_bonus + \
124	(((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t))))
125#define	DN_MAX_BONUS_LEN(dnp) \
126	((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ? \
127	(uint8_t *)DN_SPILL_BLKPTR(dnp) - (uint8_t *)DN_BONUS(dnp) : \
128	(uint8_t *)(dnp + (dnp->dn_extra_slots + 1)) - (uint8_t *)DN_BONUS(dnp))
129
130#define	DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \
131	(dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT)
132
133#define	EPB(blkshift, typeshift)	(1 << (blkshift - typeshift))
134
135struct dmu_buf_impl;
136struct objset;
137struct zio;
138
139enum dnode_dirtycontext {
140	DN_UNDIRTIED,
141	DN_DIRTY_OPEN,
142	DN_DIRTY_SYNC
143};
144
145/* Is dn_used in bytes?  if not, it's in multiples of SPA_MINBLOCKSIZE */
146#define	DNODE_FLAG_USED_BYTES			(1 << 0)
147#define	DNODE_FLAG_USERUSED_ACCOUNTED		(1 << 1)
148
149/* Does dnode have a SA spill blkptr in bonus? */
150#define	DNODE_FLAG_SPILL_BLKPTR			(1 << 2)
151
152/* User/Group/Project dnode accounting */
153#define	DNODE_FLAG_USEROBJUSED_ACCOUNTED	(1 << 3)
154
155/*
156 * VARIABLE-LENGTH (LARGE) DNODES
157 *
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.
168 *
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
178 * target names.
179 *
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.
187 *
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.
195 *
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.
204 */
205
206#define	DNODE_CRYPT_PORTABLE_FLAGS_MASK		(DNODE_FLAG_SPILL_BLKPTR)
207
208typedef struct dnode_phys {
209	uint8_t dn_type;		/* dmu_object_type_t */
210	uint8_t dn_indblkshift;		/* ln2(indirect block size) */
211	uint8_t dn_nlevels;		/* 1=dn_blkptr->data blocks */
212	uint8_t dn_nblkptr;		/* length of dn_blkptr */
213	uint8_t dn_bonustype;		/* type of data in bonus buffer */
214	uint8_t	dn_checksum;		/* ZIO_CHECKSUM type */
215	uint8_t	dn_compress;		/* ZIO_COMPRESS type */
216	uint8_t dn_flags;		/* DNODE_FLAG_* */
217	uint16_t dn_datablkszsec;	/* data block size in 512b sectors */
218	uint16_t dn_bonuslen;		/* length of dn_bonus */
219	uint8_t dn_extra_slots;		/* # of subsequent slots consumed */
220	uint8_t dn_pad2[3];
221
222	/* accounting is protected by dn_dirty_mtx */
223	uint64_t dn_maxblkid;		/* largest allocated block ID */
224	uint64_t dn_used;		/* bytes (or sectors) of disk space */
225
226	/*
227	 * Both dn_pad2 and dn_pad3 are protected by the block's MAC. This
228	 * allows us to protect any fields that might be added here in the
229	 * future. In either case, developers will want to check
230	 * zio_crypt_init_uios_dnode() to ensure the new field is being
231	 * protected properly.
232	 */
233	uint64_t dn_pad3[4];
234	union {
235		blkptr_t dn_blkptr[1+DN_OLD_MAX_BONUSLEN/sizeof (blkptr_t)];
236		struct {
237			blkptr_t __dn_ignore1;
238			uint8_t dn_bonus[DN_OLD_MAX_BONUSLEN];
239		};
240		struct {
241			blkptr_t __dn_ignore2;
242			uint8_t __dn_ignore3[DN_OLD_MAX_BONUSLEN -
243			    sizeof (blkptr_t)];
244			blkptr_t dn_spill;
245		};
246	};
247} dnode_phys_t;
248
249#define	DN_SPILL_BLKPTR(dnp)	((blkptr_t *)((char *)(dnp) +		\
250	(((dnp)->dn_extra_slots + 1) << DNODE_SHIFT) - (1 << SPA_BLKPTRSHIFT)))
251
252struct dnode {
253	/*
254	 * Protects the structure of the dnode, including the number of levels
255	 * of indirection (dn_nlevels), dn_maxblkid, and dn_next_*
256	 */
257	krwlock_t dn_struct_rwlock;
258
259	/* Our link on dn_objset->os_dnodes list; protected by os_lock.  */
260	list_node_t dn_link;
261
262	/* immutable: */
263	struct objset *dn_objset;
264	uint64_t dn_object;
265	struct dmu_buf_impl *dn_dbuf;
266	struct dnode_handle *dn_handle;
267	dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */
268
269	/*
270	 * Copies of stuff in dn_phys.  They're valid in the open
271	 * context (eg. even before the dnode is first synced).
272	 * Where necessary, these are protected by dn_struct_rwlock.
273	 */
274	dmu_object_type_t dn_type;	/* object type */
275	uint16_t dn_bonuslen;		/* bonus length */
276	uint8_t dn_bonustype;		/* bonus type */
277	uint8_t dn_nblkptr;		/* number of blkptrs (immutable) */
278	uint8_t dn_checksum;		/* ZIO_CHECKSUM type */
279	uint8_t dn_compress;		/* ZIO_COMPRESS type */
280	uint8_t dn_nlevels;
281	uint8_t dn_indblkshift;
282	uint8_t dn_datablkshift;	/* zero if blksz not power of 2! */
283	uint8_t dn_moved;		/* Has this dnode been moved? */
284	uint16_t dn_datablkszsec;	/* in 512b sectors */
285	uint32_t dn_datablksz;		/* in bytes */
286	uint64_t dn_maxblkid;
287	uint8_t dn_next_type[TXG_SIZE];
288	uint8_t dn_num_slots;		/* metadnode slots consumed on disk */
289	uint8_t dn_next_nblkptr[TXG_SIZE];
290	uint8_t dn_next_nlevels[TXG_SIZE];
291	uint8_t dn_next_indblkshift[TXG_SIZE];
292	uint8_t dn_next_bonustype[TXG_SIZE];
293	uint8_t dn_rm_spillblk[TXG_SIZE];	/* for removing spill blk */
294	uint16_t dn_next_bonuslen[TXG_SIZE];
295	uint32_t dn_next_blksz[TXG_SIZE];	/* next block size in bytes */
296	uint64_t dn_next_maxblkid[TXG_SIZE];	/* next maxblkid in bytes */
297
298	/* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */
299	uint32_t dn_dbufs_count;	/* count of dn_dbufs */
300
301	/* protected by os_lock: */
302	multilist_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */
303
304	/* protected by dn_mtx: */
305	kmutex_t dn_mtx;
306	list_t dn_dirty_records[TXG_SIZE];
307	struct range_tree *dn_free_ranges[TXG_SIZE];
308	uint64_t dn_allocated_txg;
309	uint64_t dn_free_txg;
310	uint64_t dn_assigned_txg;
311	uint64_t dn_dirty_txg;			/* txg dnode was last dirtied */
312	kcondvar_t dn_notxholds;
313	enum dnode_dirtycontext dn_dirtyctx;
314	uint8_t *dn_dirtyctx_firstset;		/* dbg: contents meaningless */
315
316	/* protected by own devices */
317	zfs_refcount_t dn_tx_holds;
318	zfs_refcount_t dn_holds;
319
320	kmutex_t dn_dbufs_mtx;
321	/*
322	 * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs
323	 * can contain multiple dbufs of the same (level, blkid) when a
324	 * dbuf is marked DB_EVICTING without being removed from
325	 * dn_dbufs. To maintain the avl invariant that there cannot be
326	 * duplicate entries, we order the dbufs by an arbitrary value -
327	 * their address in memory. This means that dn_dbufs cannot be used to
328	 * directly look up a dbuf. Instead, callers must use avl_walk, have
329	 * a reference to the dbuf, or look up a non-existant node with
330	 * db_state = DB_SEARCH (see dbuf_free_range for an example).
331	 */
332	avl_tree_t dn_dbufs;
333
334	/* protected by dn_struct_rwlock */
335	struct dmu_buf_impl *dn_bonus;	/* bonus buffer dbuf */
336
337	boolean_t dn_have_spill;	/* have spill or are spilling */
338
339	/* parent IO for current sync write */
340	zio_t *dn_zio;
341
342	/* used in syncing context */
343	uint64_t dn_oldused;	/* old phys used bytes */
344	uint64_t dn_oldflags;	/* old phys dn_flags */
345	uint64_t dn_olduid, dn_oldgid, dn_oldprojid;
346	uint64_t dn_newuid, dn_newgid, dn_newprojid;
347	int dn_id_flags;
348
349	/* holds prefetch structure */
350	struct zfetch	dn_zfetch;
351};
352
353/*
354 * We use this (otherwise unused) bit to indicate if the value of
355 * dn_next_maxblkid[txgoff] is valid to use in dnode_sync().
356 */
357#define	DMU_NEXT_MAXBLKID_SET		(1ULL << 63)
358
359/*
360 * Adds a level of indirection between the dbuf and the dnode to avoid
361 * iterating descendent dbufs in dnode_move(). Handles are not allocated
362 * individually, but as an array of child dnodes in dnode_hold_impl().
363 */
364typedef struct dnode_handle {
365	/* Protects dnh_dnode from modification by dnode_move(). */
366	zrlock_t dnh_zrlock;
367	dnode_t *dnh_dnode;
368} dnode_handle_t;
369
370typedef struct dnode_children {
371	dmu_buf_user_t dnc_dbu;		/* User evict data */
372	size_t dnc_count;		/* number of children */
373	dnode_handle_t dnc_children[];	/* sized dynamically */
374} dnode_children_t;
375
376typedef struct free_range {
377	avl_node_t fr_node;
378	uint64_t fr_blkid;
379	uint64_t fr_nblks;
380} free_range_t;
381
382void dnode_special_open(struct objset *dd, dnode_phys_t *dnp,
383    uint64_t object, dnode_handle_t *dnh);
384void dnode_special_close(dnode_handle_t *dnh);
385
386void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx);
387void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx);
388void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx);
389
390int dnode_hold(struct objset *dd, uint64_t object,
391    void *ref, dnode_t **dnp);
392int dnode_hold_impl(struct objset *dd, uint64_t object, int flag, int dn_slots,
393    void *ref, dnode_t **dnp);
394boolean_t dnode_add_ref(dnode_t *dn, void *ref);
395void dnode_rele(dnode_t *dn, void *ref);
396void dnode_rele_and_unlock(dnode_t *dn, void *tag, boolean_t evicting);
397int dnode_try_claim(objset_t *os, uint64_t object, int slots);
398void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx);
399void dnode_sync(dnode_t *dn, dmu_tx_t *tx);
400void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
401    dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx);
402void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
403    dmu_object_type_t bonustype, int bonuslen, int dn_slots,
404    boolean_t keep_spill, dmu_tx_t *tx);
405void dnode_free(dnode_t *dn, dmu_tx_t *tx);
406void dnode_byteswap(dnode_phys_t *dnp);
407void dnode_buf_byteswap(void *buf, size_t size);
408void dnode_verify(dnode_t *dn);
409int dnode_set_nlevels(dnode_t *dn, int nlevels, dmu_tx_t *tx);
410int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx);
411void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx);
412void dnode_diduse_space(dnode_t *dn, int64_t space);
413void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx,
414    boolean_t have_read, boolean_t force);
415uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid);
416void dnode_init(void);
417void dnode_fini(void);
418int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off,
419    int minlvl, uint64_t blkfill, uint64_t txg);
420void dnode_evict_dbufs(dnode_t *dn);
421void dnode_evict_bonus(dnode_t *dn);
422void dnode_free_interior_slots(dnode_t *dn);
423boolean_t dnode_needs_remap(const dnode_t *dn);
424
425#define	DNODE_IS_DIRTY(_dn)						\
426	((_dn)->dn_dirty_txg >= spa_syncing_txg((_dn)->dn_objset->os_spa))
427
428#define	DNODE_IS_CACHEABLE(_dn)						\
429	((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL ||		\
430	(DMU_OT_IS_METADATA((_dn)->dn_type) &&				\
431	(_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA))
432
433#define	DNODE_META_IS_CACHEABLE(_dn)					\
434	((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL ||		\
435	(_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)
436
437/*
438 * Used for dnodestats kstat.
439 */
440typedef struct dnode_stats {
441	/*
442	 * Number of failed attempts to hold a meta dnode dbuf.
443	 */
444	kstat_named_t dnode_hold_dbuf_hold;
445	/*
446	 * Number of failed attempts to read a meta dnode dbuf.
447	 */
448	kstat_named_t dnode_hold_dbuf_read;
449	/*
450	 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was able
451	 * to hold the requested object number which was allocated.  This is
452	 * the common case when looking up any allocated object number.
453	 */
454	kstat_named_t dnode_hold_alloc_hits;
455	/*
456	 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was not
457	 * able to hold the request object number because it was not allocated.
458	 */
459	kstat_named_t dnode_hold_alloc_misses;
460	/*
461	 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) was not
462	 * able to hold the request object number because the object number
463	 * refers to an interior large dnode slot.
464	 */
465	kstat_named_t dnode_hold_alloc_interior;
466	/*
467	 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) needed
468	 * to retry acquiring slot zrl locks due to contention.
469	 */
470	kstat_named_t dnode_hold_alloc_lock_retry;
471	/*
472	 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) did not
473	 * need to create the dnode because another thread did so after
474	 * dropping the read lock but before acquiring the write lock.
475	 */
476	kstat_named_t dnode_hold_alloc_lock_misses;
477	/*
478	 * Number of times dnode_hold(..., DNODE_MUST_BE_ALLOCATED) found
479	 * a free dnode instantiated by dnode_create() but not yet allocated
480	 * by dnode_allocate().
481	 */
482	kstat_named_t dnode_hold_alloc_type_none;
483	/*
484	 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was able
485	 * to hold the requested range of free dnode slots.
486	 */
487	kstat_named_t dnode_hold_free_hits;
488	/*
489	 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was not
490	 * able to hold the requested range of free dnode slots because
491	 * at least one slot was allocated.
492	 */
493	kstat_named_t dnode_hold_free_misses;
494	/*
495	 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) was not
496	 * able to hold the requested range of free dnode slots because
497	 * after acquiring the zrl lock at least one slot was allocated.
498	 */
499	kstat_named_t dnode_hold_free_lock_misses;
500	/*
501	 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) needed
502	 * to retry acquiring slot zrl locks due to contention.
503	 */
504	kstat_named_t dnode_hold_free_lock_retry;
505	/*
506	 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) requested
507	 * a range of dnode slots which were held by another thread.
508	 */
509	kstat_named_t dnode_hold_free_refcount;
510	/*
511	 * Number of times dnode_hold(..., DNODE_MUST_BE_FREE) requested
512	 * a range of dnode slots which would overflow the dnode_phys_t.
513	 */
514	kstat_named_t dnode_hold_free_overflow;
515	/*
516	 * Number of times dnode_free_interior_slots() needed to retry
517	 * acquiring a slot zrl lock due to contention.
518	 */
519	kstat_named_t dnode_free_interior_lock_retry;
520	/*
521	 * Number of new dnodes allocated by dnode_allocate().
522	 */
523	kstat_named_t dnode_allocate;
524	/*
525	 * Number of dnodes re-allocated by dnode_reallocate().
526	 */
527	kstat_named_t dnode_reallocate;
528	/*
529	 * Number of meta dnode dbufs evicted.
530	 */
531	kstat_named_t dnode_buf_evict;
532	/*
533	 * Number of times dmu_object_alloc*() reached the end of the existing
534	 * object ID chunk and advanced to a new one.
535	 */
536	kstat_named_t dnode_alloc_next_chunk;
537	/*
538	 * Number of times multiple threads attempted to allocate a dnode
539	 * from the same block of free dnodes.
540	 */
541	kstat_named_t dnode_alloc_race;
542	/*
543	 * Number of times dmu_object_alloc*() was forced to advance to the
544	 * next meta dnode dbuf due to an error from  dmu_object_next().
545	 */
546	kstat_named_t dnode_alloc_next_block;
547	/*
548	 * Statistics for tracking dnodes which have been moved.
549	 */
550	kstat_named_t dnode_move_invalid;
551	kstat_named_t dnode_move_recheck1;
552	kstat_named_t dnode_move_recheck2;
553	kstat_named_t dnode_move_special;
554	kstat_named_t dnode_move_handle;
555	kstat_named_t dnode_move_rwlock;
556	kstat_named_t dnode_move_active;
557} dnode_stats_t;
558
559extern dnode_stats_t dnode_stats;
560
561#define	DNODE_STAT_INCR(stat, val) \
562    atomic_add_64(&dnode_stats.stat.value.ui64, (val));
563#define	DNODE_STAT_BUMP(stat) \
564    DNODE_STAT_INCR(stat, 1);
565
566#ifdef ZFS_DEBUG
567
568/*
569 * There should be a ## between the string literal and fmt, to make it
570 * clear that we're joining two strings together, but that piece of shit
571 * gcc doesn't support that preprocessor token.
572 */
573#define	dprintf_dnode(dn, fmt, ...) do { \
574	if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
575	char __db_buf[32]; \
576	uint64_t __db_obj = (dn)->dn_object; \
577	if (__db_obj == DMU_META_DNODE_OBJECT) \
578		(void) strcpy(__db_buf, "mdn"); \
579	else \
580		(void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
581		    (u_longlong_t)__db_obj);\
582	dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \
583	    __db_buf, __VA_ARGS__); \
584	} \
585_NOTE(CONSTCOND) } while (0)
586
587#define	DNODE_VERIFY(dn)		dnode_verify(dn)
588#define	FREE_VERIFY(db, start, end, tx)	free_verify(db, start, end, tx)
589
590#else
591
592#define	dprintf_dnode(db, fmt, ...)
593#define	DNODE_VERIFY(dn)
594#define	FREE_VERIFY(db, start, end, tx)
595
596#endif
597
598#ifdef	__cplusplus
599}
600#endif
601
602#endif	/* _SYS_DNODE_H */
603