xref: /dragonfly/sys/vfs/hammer/hammer.h (revision cad2e385)
1 /*
2  * Copyright (c) 2007-2008 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 #ifndef VFS_HAMMER_HAMMER_H_
36 #define VFS_HAMMER_HAMMER_H_
37 
38 /*
39  * This header file contains structures used internally by the HAMMERFS
40  * implementation.  See hammer_disk.h for on-disk structures.
41  */
42 
43 #include <sys/param.h>
44 #ifdef _KERNEL
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #endif
48 #include <sys/conf.h>
49 #include <sys/tree.h>
50 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/vnode.h>
53 #include <sys/proc.h>
54 #include <sys/priv.h>
55 #include <sys/dirent.h>
56 #include <sys/stat.h>
57 #include <sys/fcntl.h>
58 #include <sys/lockf.h>
59 #include <sys/file.h>
60 #include <sys/event.h>
61 #include <sys/buf.h>
62 #include <sys/queue.h>
63 #include <sys/ktr.h>
64 #include <sys/limits.h>
65 #include <sys/sysctl.h>
66 #include <vm/swap_pager.h>
67 #include <vm/vm_extern.h>
68 
69 #include "hammer_disk.h"
70 #include "hammer_mount.h"
71 #include "hammer_ioctl.h"
72 
73 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
74 
75 MALLOC_DECLARE(M_HAMMER);
76 
77 /*
78  * Kernel trace
79  */
80 #if !defined(KTR_HAMMER)
81 #define KTR_HAMMER	KTR_ALL
82 #endif
83 /* KTR_INFO_MASTER_EXTERN(hammer); */
84 
85 /*
86  * Misc structures
87  */
88 struct hammer_mount;
89 
90 /*
91  * Key structure used for custom RB tree inode lookups.  This prototypes
92  * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
93  */
94 typedef struct hammer_inode_info {
95 	int64_t		obj_id;		/* (key) object identifier */
96 	hammer_tid_t	obj_asof;	/* (key) snapshot transid or 0 */
97 	u_int32_t	obj_localization; /* (key) pseudo-fs */
98 	union {
99 		struct hammer_btree_leaf_elm *leaf;
100 	} u;
101 } *hammer_inode_info_t;
102 
103 typedef enum hammer_transaction_type {
104 	HAMMER_TRANS_RO,
105 	HAMMER_TRANS_STD,
106 	HAMMER_TRANS_FLS
107 } hammer_transaction_type_t;
108 
109 /*
110  * HAMMER Transaction tracking
111  */
112 struct hammer_transaction {
113 	hammer_transaction_type_t type;
114 	struct hammer_mount *hmp;
115 	hammer_tid_t	tid;
116 	u_int64_t	time;
117 	u_int32_t	time32;
118 	int		sync_lock_refs;
119 	int		flags;
120 	struct hammer_volume *rootvol;
121 };
122 
123 typedef struct hammer_transaction *hammer_transaction_t;
124 
125 #define HAMMER_TRANSF_NEWINODE	0x0001
126 #define HAMMER_TRANSF_DIDIO	0x0002
127 #define HAMMER_TRANSF_CRCDOM	0x0004	/* EDOM on CRC error, less critical */
128 
129 /*
130  * HAMMER locks
131  */
132 struct hammer_lock {
133 	volatile u_int	refs;		/* active references */
134 	volatile u_int	lockval;	/* lock count and control bits */
135 	struct thread	*lowner;	/* owner if exclusively held */
136 	struct thread	*rowner;	/* owner if exclusively held */
137 };
138 
139 #define HAMMER_REFS_LOCKED	0x40000000	/* transition check */
140 #define HAMMER_REFS_WANTED	0x20000000	/* transition check */
141 #define HAMMER_REFS_CHECK	0x10000000	/* transition check */
142 
143 #define HAMMER_REFS_FLAGS	(HAMMER_REFS_LOCKED | \
144 				 HAMMER_REFS_WANTED | \
145 				 HAMMER_REFS_CHECK)
146 
147 #define HAMMER_LOCKF_EXCLUSIVE	0x40000000
148 #define HAMMER_LOCKF_WANTED	0x20000000
149 
150 static __inline int
151 hammer_notlocked(struct hammer_lock *lock)
152 {
153 	return(lock->lockval == 0);
154 }
155 
156 static __inline int
157 hammer_islocked(struct hammer_lock *lock)
158 {
159 	return(lock->lockval != 0);
160 }
161 
162 /*
163  * Returns the number of refs on the object.
164  */
165 static __inline int
166 hammer_isactive(struct hammer_lock *lock)
167 {
168 	return(lock->refs & ~HAMMER_REFS_FLAGS);
169 }
170 
171 static __inline int
172 hammer_oneref(struct hammer_lock *lock)
173 {
174 	return((lock->refs & ~HAMMER_REFS_FLAGS) == 1);
175 }
176 
177 static __inline int
178 hammer_norefs(struct hammer_lock *lock)
179 {
180 	return((lock->refs & ~HAMMER_REFS_FLAGS) == 0);
181 }
182 
183 static __inline int
184 hammer_norefsorlock(struct hammer_lock *lock)
185 {
186 	return(lock->refs == 0);
187 }
188 
189 static __inline int
190 hammer_refsorlock(struct hammer_lock *lock)
191 {
192 	return(lock->refs != 0);
193 }
194 
195 /*
196  * Return if we specifically own the lock exclusively.
197  */
198 static __inline int
199 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
200 {
201 	if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) &&
202 	    lock->lowner == td) {
203 		return(1);
204 	}
205 	return(0);
206 }
207 
208 /*
209  * Flush state, used by various structures
210  */
211 typedef enum hammer_inode_state {
212 	HAMMER_FST_IDLE,
213 	HAMMER_FST_SETUP,
214 	HAMMER_FST_FLUSH
215 } hammer_inode_state_t;
216 
217 TAILQ_HEAD(hammer_record_list, hammer_record);
218 
219 /*
220  * Pseudo-filesystem extended data tracking
221  */
222 struct hammer_pfs_rb_tree;
223 struct hammer_pseudofs_inmem;
224 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
225 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
226 	      hammer_pfs_rb_compare, u_int32_t);
227 
228 struct hammer_pseudofs_inmem {
229 	RB_ENTRY(hammer_pseudofs_inmem)	rb_node;
230 	struct hammer_lock	lock;
231 	u_int32_t		localization;
232 	hammer_tid_t		create_tid;
233 	int			flags;
234 	udev_t			fsid_udev;
235 	struct hammer_pseudofs_data pfsd;
236 };
237 
238 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t;
239 
240 /*
241  * Cache object ids.  A fixed number of objid cache structures are
242  * created to reserve object id's for newly created files in multiples
243  * of 100,000, localized to a particular directory, and recycled as
244  * needed.  This allows parallel create operations in different
245  * directories to retain fairly localized object ids which in turn
246  * improves reblocking performance and layout.
247  */
248 #define OBJID_CACHE_SIZE	2048
249 #define OBJID_CACHE_BULK_BITS	10		/* 10 bits (1024)	*/
250 #define OBJID_CACHE_BULK	(32 * 32)	/* two level (1024)	*/
251 #define OBJID_CACHE_BULK_MASK	(OBJID_CACHE_BULK - 1)
252 #define OBJID_CACHE_BULK_MASK64	((u_int64_t)(OBJID_CACHE_BULK - 1))
253 
254 typedef struct hammer_objid_cache {
255 	TAILQ_ENTRY(hammer_objid_cache) entry;
256 	struct hammer_inode		*dip;
257 	hammer_tid_t			base_tid;
258 	int				count;
259 	u_int32_t			bm0;
260 	u_int32_t			bm1[32];
261 } *hammer_objid_cache_t;
262 
263 /*
264  * Associate an inode with a B-Tree node to cache search start positions
265  */
266 typedef struct hammer_node_cache {
267 	TAILQ_ENTRY(hammer_node_cache)	entry;
268 	struct hammer_node		*node;
269 	struct hammer_inode		*ip;
270 } *hammer_node_cache_t;
271 
272 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
273 
274 /*
275  * Live dedup cache
276  */
277 struct hammer_dedup_crc_rb_tree;
278 RB_HEAD(hammer_dedup_crc_rb_tree, hammer_dedup_cache);
279 RB_PROTOTYPE2(hammer_dedup_crc_rb_tree, hammer_dedup_cache, crc_entry,
280 		hammer_dedup_crc_rb_compare, hammer_crc_t);
281 
282 struct hammer_dedup_off_rb_tree;
283 RB_HEAD(hammer_dedup_off_rb_tree, hammer_dedup_cache);
284 RB_PROTOTYPE2(hammer_dedup_off_rb_tree, hammer_dedup_cache, off_entry,
285 		hammer_dedup_off_rb_compare, hammer_off_t);
286 
287 #define DEDUP_CACHE_SIZE	4096 /* XXX make it a dynamic tunable */
288 
289 typedef struct hammer_dedup_cache {
290 	RB_ENTRY(hammer_dedup_cache) crc_entry;
291 	RB_ENTRY(hammer_dedup_cache) off_entry;
292 	TAILQ_ENTRY(hammer_dedup_cache) lru_entry;
293 	struct hammer_mount *hmp;
294 	int64_t obj_id;
295 	u_int32_t localization;
296 	off_t file_offset;
297 	int bytes;
298 	hammer_off_t data_offset;
299 	hammer_crc_t crc;
300 } *hammer_dedup_cache_t;
301 
302 /*
303  * Structure used to organize flush groups.  Flush groups must be
304  * organized into chunks in order to avoid blowing out the UNDO FIFO.
305  * Without this a 'sync' could end up flushing 50,000 inodes in a single
306  * transaction.
307  */
308 struct hammer_fls_rb_tree;
309 RB_HEAD(hammer_fls_rb_tree, hammer_inode);
310 RB_PROTOTYPE(hammer_fls_rb_tree, hammer_inode, rb_flsnode,
311 	      hammer_ino_rb_compare);
312 
313 struct hammer_flush_group {
314 	TAILQ_ENTRY(hammer_flush_group)	flush_entry;
315 	struct hammer_fls_rb_tree	flush_tree;
316 	int				seq;		/* our seq no */
317 	int				total_count;	/* record load */
318 	int				running;	/* group is running */
319 	int				closed;
320 	int				refs;
321 };
322 
323 typedef struct hammer_flush_group *hammer_flush_group_t;
324 
325 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
326 
327 /*
328  * Structure used to represent an inode in-memory.
329  *
330  * The record and data associated with an inode may be out of sync with
331  * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
332  * clear).
333  *
334  * An inode may also hold a cache of unsynchronized records, used for
335  * database and directories only.  Unsynchronized regular file data is
336  * stored in the buffer cache.
337  *
338  * NOTE: A file which is created and destroyed within the initial
339  * synchronization period can wind up not doing any disk I/O at all.
340  *
341  * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
342  */
343 struct hammer_ino_rb_tree;
344 struct hammer_inode;
345 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
346 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
347 	      hammer_ino_rb_compare, hammer_inode_info_t);
348 
349 struct hammer_redo_rb_tree;
350 RB_HEAD(hammer_redo_rb_tree, hammer_inode);
351 RB_PROTOTYPE2(hammer_redo_rb_tree, hammer_inode, rb_redonode,
352 	      hammer_redo_rb_compare, hammer_off_t);
353 
354 struct hammer_rec_rb_tree;
355 struct hammer_record;
356 RB_HEAD(hammer_rec_rb_tree, hammer_record);
357 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
358 	      hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
359 
360 TAILQ_HEAD(hammer_node_list, hammer_node);
361 
362 struct hammer_inode {
363 	RB_ENTRY(hammer_inode)	rb_node;
364 	hammer_inode_state_t	flush_state;
365 	hammer_flush_group_t	flush_group;
366 	RB_ENTRY(hammer_inode)	rb_flsnode;	/* when on flush list */
367 	RB_ENTRY(hammer_inode)	rb_redonode;	/* when INODE_RDIRTY is set */
368 	struct hammer_record_list target_list;	/* target of dependant recs */
369 	int64_t			obj_id;		/* (key) object identifier */
370 	hammer_tid_t		obj_asof;	/* (key) snapshot or 0 */
371 	u_int32_t		obj_localization; /* (key) pseudo-fs */
372 	struct hammer_mount	*hmp;
373 	hammer_objid_cache_t	objid_cache;
374 	int			flags;
375 	int			error;		/* flush error */
376 	int			cursor_ip_refs;	/* sanity */
377 #if 0
378 	int			cursor_exclreq_count;
379 #endif
380 	int			rsv_recs;
381 	struct vnode		*vp;
382 	hammer_pseudofs_inmem_t	pfsm;
383 	struct lockf		advlock;
384 	struct hammer_lock	lock;		/* sync copy interlock */
385 	off_t			trunc_off;
386 	struct hammer_btree_leaf_elm ino_leaf;  /* in-memory cache */
387 	struct hammer_inode_data ino_data;	/* in-memory cache */
388 	struct hammer_rec_rb_tree rec_tree;	/* in-memory cache */
389 	int			rec_generation;
390 
391 	/*
392 	 * search initiate cache
393 	 * cache[0] - this inode
394 	 * cache[1] - related data, the content depends on situations
395 	 * cache[2] - for dip to cache ip to shortcut B-Tree search
396 	 * cache[3] - related data copied from dip to a new ip's cache[1]
397 	 */
398 	struct hammer_node_cache cache[4];
399 
400 	/*
401 	 * When a demark is created to synchronize an inode to
402 	 * disk, certain fields are copied so the front-end VOPs
403 	 * can continue to run in parallel with the synchronization
404 	 * occuring in the background.
405 	 */
406 	int		sync_flags;		/* to-sync flags cache */
407 	off_t		sync_trunc_off;		/* to-sync truncation */
408 	off_t		save_trunc_off;		/* write optimization */
409 	struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
410 	struct hammer_inode_data sync_ino_data; /* to-sync cache */
411 	size_t		redo_count;
412 
413 	/*
414 	 * Track the earliest offset in the UNDO/REDO FIFO containing
415 	 * REDO records.  This is staged to the backend during flush
416 	 * sequences.  While the inode is staged redo_fifo_next is used
417 	 * to track the earliest offset for rotation into redo_fifo_start
418 	 * on completion of the flush.
419 	 */
420 	hammer_off_t	redo_fifo_start;
421 	hammer_off_t	redo_fifo_next;
422 };
423 
424 typedef struct hammer_inode *hammer_inode_t;
425 
426 #define VTOI(vp)	((struct hammer_inode *)(vp)->v_data)
427 
428 /*
429  * NOTE: DDIRTY does not include atime or mtime and does not include
430  *	 write-append size changes.  SDIRTY handles write-append size
431  *	 changes.
432  *
433  *	 REDO indicates that REDO logging is active, creating a definitive
434  *	 stream of REDO records in the UNDO/REDO log for writes and
435  *	 truncations, including boundary records when/if REDO is turned off.
436  *	 REDO is typically enabled by fsync() and turned off if excessive
437  *	 writes without an fsync() occurs.
438  *
439  *	 RDIRTY indicates that REDO records were laid down in the UNDO/REDO
440  *	 FIFO (even if REDO is turned off some might still be active) and
441  *	 still being tracked for this inode.  See hammer_redo.c
442  */
443 					/* (not including atime/mtime) */
444 #define HAMMER_INODE_DDIRTY	0x0001	/* in-memory ino_data is dirty */
445 #define HAMMER_INODE_RSV_INODES	0x0002	/* hmp->rsv_inodes bumped */
446 #define HAMMER_INODE_CONN_DOWN	0x0004	/* include in downward recursion */
447 #define HAMMER_INODE_XDIRTY	0x0008	/* in-memory records */
448 #define HAMMER_INODE_ONDISK	0x0010	/* inode is on-disk (else not yet) */
449 #define HAMMER_INODE_FLUSH	0x0020	/* flush on last ref */
450 #define HAMMER_INODE_DELETED	0x0080	/* inode delete (backend) */
451 #define HAMMER_INODE_DELONDISK	0x0100	/* delete synchronized to disk */
452 #define HAMMER_INODE_RO		0x0200	/* read-only (because of as-of) */
453 #define HAMMER_INODE_RECSW	0x0400	/* waiting on data record flush */
454 #define HAMMER_INODE_DONDISK	0x0800	/* data records may be on disk */
455 #define HAMMER_INODE_BUFS	0x1000	/* dirty high level bps present */
456 #define HAMMER_INODE_REFLUSH	0x2000	/* flush on dependancy / reflush */
457 #define HAMMER_INODE_RECLAIM	0x4000	/* trying to reclaim */
458 #define HAMMER_INODE_FLUSHW	0x8000	/* Someone waiting for flush */
459 
460 #define HAMMER_INODE_TRUNCATED	0x00010000
461 #define HAMMER_INODE_DELETING	0x00020000 /* inode delete request (frontend)*/
462 #define HAMMER_INODE_RESIGNAL	0x00040000 /* re-signal on re-flush */
463 #define HAMMER_INODE_ATIME	0x00100000 /* in-memory atime modified */
464 #define HAMMER_INODE_MTIME	0x00200000 /* in-memory mtime modified */
465 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
466 #define HAMMER_INODE_DUMMY	0x00800000 /* dummy inode covering bad file */
467 #define HAMMER_INODE_SDIRTY	0x01000000 /* in-memory ino_data.size is dirty*/
468 #define HAMMER_INODE_REDO	0x02000000 /* REDO logging active */
469 #define HAMMER_INODE_RDIRTY	0x04000000 /* REDO records active in fifo */
470 #define HAMMER_INODE_SLAVEFLUSH	0x08000000 /* being flushed by slave */
471 
472 #define HAMMER_INODE_MODMASK	(HAMMER_INODE_DDIRTY|HAMMER_INODE_SDIRTY|   \
473 				 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS|	    \
474 				 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME|     \
475 				 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
476 
477 #define HAMMER_INODE_MODMASK_NOXDIRTY	\
478 				(HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
479 
480 #define HAMMER_INODE_MODMASK_NOREDO	\
481 				(HAMMER_INODE_DDIRTY|			    \
482 				 HAMMER_INODE_XDIRTY|			    \
483 				 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
484 
485 #define HAMMER_FLUSH_SIGNAL	0x0001
486 #define HAMMER_FLUSH_RECURSION	0x0002
487 
488 /*
489  * Used by the inode reclaim code to pipeline reclaims and avoid
490  * blowing out kernel memory or letting the flusher get too far
491  * behind.  The reclaim wakes up when count reaches 0 or the
492  * timer expires.
493  */
494 struct hammer_reclaim {
495 	TAILQ_ENTRY(hammer_reclaim) entry;
496 	int	count;
497 };
498 
499 /*
500  * Track who is creating the greatest burden on the
501  * inode cache.
502  */
503 struct hammer_inostats {
504 	pid_t		pid;	/* track user process */
505 	int		ltick;	/* last tick */
506 	int		count;	/* count (degenerates) */
507 };
508 
509 #define HAMMER_INOSTATS_HSIZE	32
510 #define HAMMER_INOSTATS_HMASK	(HAMMER_INOSTATS_HSIZE - 1)
511 
512 /*
513  * Structure used to represent an unsynchronized record in-memory.  These
514  * records typically represent directory entries.  Only non-historical
515  * records are kept in-memory.
516  *
517  * Records are organized as a per-inode RB-Tree.  If the inode is not
518  * on disk then neither are any records and the in-memory record tree
519  * represents the entire contents of the inode.  If the inode is on disk
520  * then the on-disk B-Tree is scanned in parallel with the in-memory
521  * RB-Tree to synthesize the current state of the file.
522  *
523  * Records are also used to enforce the ordering of directory create/delete
524  * operations.  A new inode will not be flushed to disk unless its related
525  * directory entry is also being flushed at the same time.  A directory entry
526  * will not be removed unless its related inode is also being removed at the
527  * same time.
528  */
529 typedef enum hammer_record_type {
530 	HAMMER_MEM_RECORD_GENERAL,	/* misc record */
531 	HAMMER_MEM_RECORD_INODE,	/* inode record */
532 	HAMMER_MEM_RECORD_ADD,		/* positive memory cache record */
533 	HAMMER_MEM_RECORD_DEL,		/* negative delete-on-disk record */
534 	HAMMER_MEM_RECORD_DATA		/* bulk-data record w/on-disk ref */
535 } hammer_record_type_t;
536 
537 struct hammer_record {
538 	RB_ENTRY(hammer_record)		rb_node;
539 	TAILQ_ENTRY(hammer_record)	target_entry;
540 	hammer_inode_state_t		flush_state;
541 	hammer_flush_group_t		flush_group;
542 	hammer_record_type_t		type;
543 	struct hammer_lock		lock;
544 	struct hammer_reserve		*resv;
545 	struct hammer_inode		*ip;
546 	struct hammer_inode		*target_ip;
547 	struct hammer_btree_leaf_elm	leaf;
548 	union hammer_data_ondisk	*data;
549 	int				flags;
550 	int				gflags;
551 	hammer_off_t			zone2_offset;	/* direct-write only */
552 };
553 
554 typedef struct hammer_record *hammer_record_t;
555 
556 /*
557  * Record flags.  Note that FE can only be set by the frontend if the
558  * record has not been interlocked by the backend w/ BE.
559  */
560 #define HAMMER_RECF_ALLOCDATA		0x0001
561 #define HAMMER_RECF_ONRBTREE		0x0002
562 #define HAMMER_RECF_DELETED_FE		0x0004	/* deleted (frontend) */
563 #define HAMMER_RECF_DELETED_BE		0x0008	/* deleted (backend) */
564 #define HAMMER_RECF_COMMITTED		0x0010	/* committed to the B-Tree */
565 #define HAMMER_RECF_INTERLOCK_BE	0x0020	/* backend interlock */
566 #define HAMMER_RECF_WANTED		0x0040	/* wanted by the frontend */
567 #define HAMMER_RECF_DEDUPED		0x0080	/* will be live-dedup'ed */
568 #define HAMMER_RECF_CONVERT_DELETE	0x0100	/* special case */
569 #define HAMMER_RECF_REDO		0x1000	/* REDO was laid down */
570 
571 /*
572  * These flags must be separate to deal with SMP races
573  */
574 #define HAMMER_RECG_DIRECT_IO		0x0001	/* related direct I/O running*/
575 #define HAMMER_RECG_DIRECT_WAIT		0x0002	/* related direct I/O running*/
576 #define HAMMER_RECG_DIRECT_INVAL	0x0004	/* buffer alias invalidation */
577 /*
578  * hammer_create_at_cursor() and hammer_delete_at_cursor() flags.
579  */
580 #define HAMMER_CREATE_MODE_UMIRROR	0x0001
581 #define HAMMER_CREATE_MODE_SYS		0x0002
582 
583 #define HAMMER_DELETE_ADJUST		0x0001
584 #define HAMMER_DELETE_DESTROY		0x0002
585 
586 /*
587  * In-memory structures representing on-disk structures.
588  */
589 struct hammer_volume;
590 struct hammer_buffer;
591 struct hammer_node;
592 struct hammer_undo;
593 struct hammer_reserve;
594 
595 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
596 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
597 RB_HEAD(hammer_nod_rb_tree, hammer_node);
598 RB_HEAD(hammer_und_rb_tree, hammer_undo);
599 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
600 RB_HEAD(hammer_mod_rb_tree, hammer_io);
601 
602 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
603 	      hammer_vol_rb_compare, int32_t);
604 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
605 	      hammer_buf_rb_compare, hammer_off_t);
606 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
607 	      hammer_nod_rb_compare, hammer_off_t);
608 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
609 	      hammer_und_rb_compare, hammer_off_t);
610 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
611 	      hammer_res_rb_compare, hammer_off_t);
612 RB_PROTOTYPE2(hammer_mod_rb_tree, hammer_io, rb_node,
613 	      hammer_mod_rb_compare, hammer_off_t);
614 
615 /*
616  * IO management - embedded at the head of various in-memory structures
617  *
618  * VOLUME	- hammer_volume containing meta-data
619  * META_BUFFER	- hammer_buffer containing meta-data
620  * UNDO_BUFFER	- hammer_buffer containing undo-data
621  * DATA_BUFFER	- hammer_buffer containing pure-data
622  * DUMMY	- hammer_buffer not containing valid data
623  *
624  * Dirty volume headers and dirty meta-data buffers are locked until the
625  * flusher can sequence them out.  Dirty pure-data buffers can be written.
626  * Clean buffers can be passively released.
627  */
628 typedef enum hammer_io_type {
629 	HAMMER_STRUCTURE_VOLUME,
630 	HAMMER_STRUCTURE_META_BUFFER,
631 	HAMMER_STRUCTURE_UNDO_BUFFER,
632 	HAMMER_STRUCTURE_DATA_BUFFER,
633 	HAMMER_STRUCTURE_DUMMY
634 } hammer_io_type_t;
635 
636 union hammer_io_structure;
637 struct hammer_io;
638 
639 struct worklist {
640 	LIST_ENTRY(worklist) node;
641 };
642 
643 TAILQ_HEAD(hammer_io_list, hammer_io);
644 typedef struct hammer_io_list *hammer_io_list_t;
645 
646 struct hammer_io {
647 	struct worklist		worklist;
648 	struct hammer_lock	lock;
649 	enum hammer_io_type	type;
650 	struct hammer_mount	*hmp;
651 	struct hammer_volume	*volume;
652 	RB_ENTRY(hammer_io)	rb_node;     /* if modified */
653 	TAILQ_ENTRY(hammer_io)	iorun_entry; /* iorun_list */
654 	struct hammer_mod_rb_tree *mod_root;
655 	struct buf		*bp;
656 	int64_t			offset;	   /* zone-2 offset */
657 	int			bytes;	   /* buffer cache buffer size */
658 	int			modify_refs;
659 
660 	/*
661 	 * These can be modified at any time by the backend while holding
662 	 * io_token, due to bio_done and hammer_io_complete() callbacks.
663 	 */
664 	u_int		running : 1;	/* bp write IO in progress */
665 	u_int		waiting : 1;	/* someone is waiting on us */
666 	u_int		ioerror : 1;	/* abort on io-error */
667 	u_int		unusedA : 29;
668 
669 	/*
670 	 * These can only be modified by the frontend while holding
671 	 * fs_token, or by the backend while holding the io interlocked
672 	 * with no references (which will block the frontend when it
673 	 * tries to reference it).
674 	 *
675 	 * WARNING! SMP RACES will create havoc if the callbacks ever tried
676 	 *	    to modify any of these outside the above restrictions.
677 	 */
678 	u_int		modified : 1;	/* bp's data was modified */
679 	u_int		released : 1;	/* bp released (w/ B_LOCKED set) */
680 	u_int		validated : 1;	/* ondisk has been validated */
681 	u_int		waitdep : 1;	/* flush waits for dependancies */
682 	u_int		recovered : 1;	/* has recovery ref */
683 	u_int		waitmod : 1;	/* waiting for modify_refs */
684 	u_int		reclaim : 1;	/* reclaim requested */
685 	u_int		gencrc : 1;	/* crc needs to be generated */
686 	u_int		unusedB : 24;
687 };
688 
689 typedef struct hammer_io *hammer_io_t;
690 
691 #define HAMMER_CLUSTER_SIZE	(64 * 1024)
692 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
693 #undef  HAMMER_CLUSTER_SIZE
694 #define HAMMER_CLUSTER_SIZE	MAXBSIZE
695 #endif
696 
697 /*
698  * In-memory volume representing on-disk buffer
699  */
700 struct hammer_volume {
701 	struct hammer_io io;
702 	RB_ENTRY(hammer_volume) rb_node;
703 	struct hammer_volume_ondisk *ondisk;
704 	int32_t	vol_no;
705 	int64_t nblocks;	/* note: special calculation for statfs */
706 	int64_t buffer_base;	/* base offset of buffer 0 */
707 	hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
708 	hammer_off_t maxraw_off; /* Maximum raw offset for device */
709 	char	*vol_name;
710 	struct vnode *devvp;
711 	int	vol_flags;
712 };
713 
714 typedef struct hammer_volume *hammer_volume_t;
715 
716 /*
717  * In-memory buffer representing an on-disk buffer.
718  */
719 struct hammer_buffer {
720 	struct hammer_io io;
721 	RB_ENTRY(hammer_buffer) rb_node;
722 	void *ondisk;
723 	hammer_off_t zoneX_offset;
724 	hammer_off_t zone2_offset;
725 	struct hammer_reserve *resv;
726 	struct hammer_node_list clist;
727 };
728 
729 typedef struct hammer_buffer *hammer_buffer_t;
730 
731 /*
732  * In-memory B-Tree node, representing an on-disk B-Tree node.
733  *
734  * This is a hang-on structure which is backed by a hammer_buffer,
735  * and used for fine-grained locking of B-Tree nodes in order to
736  * properly control lock ordering.
737  */
738 struct hammer_node {
739 	struct hammer_lock	lock;		/* node-by-node lock */
740 	TAILQ_ENTRY(hammer_node) entry;		/* per-buffer linkage */
741 	RB_ENTRY(hammer_node)	rb_node;	/* per-mount linkage */
742 	hammer_off_t		node_offset;	/* full offset spec */
743 	struct hammer_mount	*hmp;
744 	struct hammer_buffer	*buffer;	/* backing buffer */
745 	hammer_node_ondisk_t	ondisk;		/* ptr to on-disk structure */
746 	TAILQ_HEAD(, hammer_cursor) cursor_list;  /* deadlock recovery */
747 	struct hammer_node_cache_list cache_list; /* passive caches */
748 	int			flags;
749 #if 0
750 	int			cursor_exclreq_count;
751 #endif
752 };
753 
754 #define HAMMER_NODE_DELETED	0x0001
755 #define HAMMER_NODE_FLUSH	0x0002
756 #define HAMMER_NODE_CRCGOOD	0x0004
757 #define HAMMER_NODE_NEEDSCRC	0x0008
758 #define HAMMER_NODE_NEEDSMIRROR	0x0010
759 #define HAMMER_NODE_CRCBAD	0x0020
760 #define HAMMER_NODE_NONLINEAR	0x0040		/* linear heuristic */
761 
762 #define HAMMER_NODE_CRCANY	(HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD)
763 
764 typedef struct hammer_node	*hammer_node_t;
765 
766 /*
767  * List of locked nodes.  This structure is used to lock potentially large
768  * numbers of nodes as an aid for complex B-Tree operations.
769  */
770 struct hammer_node_lock;
771 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock);
772 
773 struct hammer_node_lock {
774 	TAILQ_ENTRY(hammer_node_lock) entry;
775 	struct hammer_node_lock_list  list;
776 	struct hammer_node_lock	      *parent;
777 	hammer_node_t	node;
778 	hammer_node_ondisk_t copy;	/* copy of on-disk data */
779 	int		index;		/* index of this node in parent */
780 	int		count;		/* count children */
781 	int		flags;
782 };
783 
784 typedef struct hammer_node_lock *hammer_node_lock_t;
785 
786 #define HAMMER_NODE_LOCK_UPDATED	0x0001
787 #define HAMMER_NODE_LOCK_LCACHE		0x0002
788 
789 /*
790  * Common I/O management structure - embedded in in-memory structures
791  * which are backed by filesystem buffers.
792  */
793 union hammer_io_structure {
794 	struct hammer_io	io;
795 	struct hammer_volume	volume;
796 	struct hammer_buffer	buffer;
797 };
798 
799 typedef union hammer_io_structure *hammer_io_structure_t;
800 
801 /*
802  * The reserve structure prevents the blockmap from allocating
803  * out of a reserved big-block.  Such reservations are used by
804  * the direct-write mechanism.
805  *
806  * The structure is also used to hold off on reallocations of
807  * big-blocks from the freemap until flush dependancies have
808  * been dealt with.
809  */
810 struct hammer_reserve {
811 	RB_ENTRY(hammer_reserve) rb_node;
812 	TAILQ_ENTRY(hammer_reserve) delay_entry;
813 	int		flush_group;
814 	int		flags;
815 	int		refs;
816 	int		zone;
817 	int		append_off;
818 	int32_t		bytes_free;
819 	hammer_off_t	zone_offset;
820 };
821 
822 typedef struct hammer_reserve *hammer_reserve_t;
823 
824 #define HAMMER_RESF_ONDELAY	0x0001
825 #define HAMMER_RESF_LAYER2FREE	0x0002
826 
827 #include "hammer_cursor.h"
828 
829 /*
830  * The undo structure tracks recent undos to avoid laying down duplicate
831  * undos within a flush group, saving us a significant amount of overhead.
832  *
833  * This is strictly a heuristic.
834  */
835 #define HAMMER_MAX_UNDOS		1024
836 #define HAMMER_MAX_FLUSHERS		4
837 
838 struct hammer_undo {
839 	RB_ENTRY(hammer_undo)	rb_node;
840 	TAILQ_ENTRY(hammer_undo) lru_entry;
841 	hammer_off_t		offset;
842 	int			bytes;
843 };
844 
845 typedef struct hammer_undo *hammer_undo_t;
846 
847 struct hammer_flusher_info;
848 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
849 
850 struct hammer_flusher {
851 	int		signal;		/* flusher thread sequencer */
852 	int		done;		/* last completed flush group */
853 	int		next;		/* next unallocated flg seqno */
854 	int		group_lock;	/* lock sequencing of the next flush */
855 	int		exiting;	/* request master exit */
856 	thread_t	td;		/* master flusher thread */
857 	hammer_tid_t	tid;		/* last flushed transaction id */
858 	int		finalize_want;		/* serialize finalization */
859 	struct hammer_lock finalize_lock;	/* serialize finalization */
860 	struct hammer_transaction trans;	/* shared transaction */
861 	struct hammer_flusher_info_list run_list;
862 	struct hammer_flusher_info_list ready_list;
863 };
864 
865 #define HAMMER_FLUSH_UNDOS_RELAXED	0
866 #define HAMMER_FLUSH_UNDOS_FORCED	1
867 #define HAMMER_FLUSH_UNDOS_AUTO		2
868 /*
869  * Internal hammer mount data structure
870  */
871 struct hammer_mount {
872 	struct mount *mp;
873 	struct hammer_ino_rb_tree rb_inos_root;
874 	struct hammer_redo_rb_tree rb_redo_root;
875 	struct hammer_vol_rb_tree rb_vols_root;
876 	struct hammer_nod_rb_tree rb_nods_root;
877 	struct hammer_und_rb_tree rb_undo_root;
878 	struct hammer_res_rb_tree rb_resv_root;
879 	struct hammer_buf_rb_tree rb_bufs_root;
880 	struct hammer_pfs_rb_tree rb_pfsm_root;
881 
882 	struct hammer_dedup_crc_rb_tree rb_dedup_crc_root;
883 	struct hammer_dedup_off_rb_tree rb_dedup_off_root;
884 
885 	struct hammer_volume *rootvol;
886 	struct hammer_base_elm root_btree_beg;
887 	struct hammer_base_elm root_btree_end;
888 
889 	struct malloc_type	*m_misc;
890 	struct malloc_type	*m_inodes;
891 
892 	int	flags;		/* HAMMER_MOUNT_xxx flags */
893 	int	hflags;
894 	int	ronly;
895 	int	nvolumes;
896 	int	volume_iterator;
897 	int	master_id;	/* -1 or 0-15 for mirroring */
898 	int	version;	/* hammer filesystem version to use */
899 	int	rsv_inodes;	/* reserved space due to dirty inodes */
900 	int64_t	rsv_databytes;	/* reserved space due to record data */
901 	int	rsv_recs;	/* reserved space due to dirty records */
902 	int	rsv_fromdelay;	/* big-blocks reserved due to flush delay */
903 	int	undo_rec_limit;	/* based on size of undo area */
904 	int	last_newrecords;
905 	int	count_newrecords;
906 
907 	int	volume_to_remove; /* volume that is currently being removed */
908 
909 	int	count_inodes;	/* total number of inodes */
910 	int	count_iqueued;	/* inodes queued to flusher */
911 	int	count_reclaims; /* inodes pending reclaim by flusher */
912 
913 	struct hammer_flusher flusher;
914 
915 	u_int	check_interrupt;
916 	u_int	check_yield;
917 	uuid_t	fsid;
918 	struct hammer_mod_rb_tree volu_root;	/* dirty undo buffers */
919 	struct hammer_mod_rb_tree undo_root;	/* dirty undo buffers */
920 	struct hammer_mod_rb_tree data_root;	/* dirty data buffers */
921 	struct hammer_mod_rb_tree meta_root;	/* dirty meta bufs    */
922 	struct hammer_mod_rb_tree lose_root;	/* loose buffers      */
923 	long	locked_dirty_space;		/* meta/volu count    */
924 	long	io_running_space;		/* io_token */
925 	int	unused01;
926 	int	objid_cache_count;
927 	int	dedup_cache_count;
928 	int	error;				/* critical I/O error */
929 	struct krate	krate;			/* rate limited kprintf */
930 	struct krate	kdiag;			/* rate limited kprintf */
931 	hammer_tid_t	asof;			/* snapshot mount */
932 	hammer_tid_t	next_tid;
933 	hammer_tid_t	flush_tid1;		/* flusher tid sequencing */
934 	hammer_tid_t	flush_tid2;		/* flusher tid sequencing */
935 	int64_t copy_stat_freebigblocks;	/* number of free big-blocks */
936 	u_int32_t	undo_seqno;		/* UNDO/REDO FIFO seqno */
937 	u_int32_t	recover_stage2_seqno;	/* REDO recovery seqno */
938 	hammer_off_t	recover_stage2_offset;	/* REDO recovery offset */
939 
940 	struct netexport export;
941 	struct hammer_lock sync_lock;
942 	struct hammer_lock free_lock;
943 	struct hammer_lock undo_lock;
944 	struct hammer_lock blkmap_lock;
945 	struct hammer_lock snapshot_lock;
946 	struct hammer_lock volume_lock;
947 	struct hammer_blockmap  blockmap[HAMMER_MAX_ZONES];
948 	struct hammer_undo	undos[HAMMER_MAX_UNDOS];
949 	int			undo_alloc;
950 	TAILQ_HEAD(, hammer_undo)  undo_lru_list;
951 	TAILQ_HEAD(, hammer_reserve) delay_list;
952 	struct hammer_flush_group_list	flush_group_list;
953 	hammer_flush_group_t	fill_flush_group;
954 	hammer_flush_group_t	next_flush_group;
955 	TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
956 	TAILQ_HEAD(, hammer_dedup_cache) dedup_lru_list;
957 	hammer_dedup_cache_t	dedup_free_cache;
958 	TAILQ_HEAD(, hammer_reclaim) reclaim_list;
959 	TAILQ_HEAD(, hammer_io) iorun_list;
960 
961 	struct lwkt_token	fs_token;	/* high level */
962 	struct lwkt_token	io_token;	/* low level (IO callback) */
963 
964 	struct hammer_inostats	inostats[HAMMER_INOSTATS_HSIZE];
965 	uint64_t volume_map[4];  /* 256 bits bitfield */
966 };
967 
968 typedef struct hammer_mount	*hammer_mount_t;
969 
970 #define HAMMER_MOUNT_CRITICAL_ERROR	0x0001
971 #define HAMMER_MOUNT_FLUSH_RECOVERY	0x0002
972 #define HAMMER_MOUNT_REDO_SYNC		0x0004
973 #define HAMMER_MOUNT_REDO_RECOVERY_REQ	0x0008
974 #define HAMMER_MOUNT_REDO_RECOVERY_RUN	0x0010
975 
976 #define HAMMER_VOLUME_NUMBER_ADD(hmp, vol)			\
977 	(hmp)->volume_map[(vol)->vol_no >> 6] |=		\
978 	((uint64_t)1 << ((vol)->vol_no & ((1 << 6) - 1)))
979 
980 #define HAMMER_VOLUME_NUMBER_DEL(hmp, vol)			\
981 	(hmp)->volume_map[(vol)->vol_no >> 6] &=		\
982 	~((uint64_t)1 << ((vol)->vol_no & ((1 << 6) - 1)))
983 
984 #define HAMMER_VOLUME_NUMBER_IS_SET(hmp, n)			\
985 	(((hmp)->volume_map[(n) >> 6] &				\
986 	((uint64_t)1 << ((n) & ((1 << 6) - 1)))) != 0)
987 
988 #define HAMMER_VOLUME_NUMBER_FOREACH(hmp, n)			\
989 	for (n = 0; n < HAMMER_MAX_VOLUMES; n++)		\
990 		if (HAMMER_VOLUME_NUMBER_IS_SET(hmp, n))
991 
992 struct hammer_sync_info {
993 	int error;
994 	int waitfor;
995 };
996 
997 /*
998  * Minium buffer cache bufs required to rebalance the B-Tree.
999  * This is because we must hold the children and the children's children
1000  * locked.  Even this might not be enough if things are horribly out
1001  * of balance.
1002  */
1003 #define HAMMER_REBALANCE_MIN_BUFS	\
1004 	(HAMMER_BTREE_LEAF_ELMS * HAMMER_BTREE_LEAF_ELMS)
1005 
1006 #endif  /* _KERNEL || _KERNEL_STRUCTURES */
1007 
1008 #if defined(_KERNEL)
1009 /*
1010  * checkspace slop (8MB chunks), higher numbers are more conservative.
1011  */
1012 #define HAMMER_CHKSPC_REBLOCK	25
1013 #define HAMMER_CHKSPC_MIRROR	20
1014 #define HAMMER_CHKSPC_WRITE	20
1015 #define HAMMER_CHKSPC_CREATE	20
1016 #define HAMMER_CHKSPC_REMOVE	10
1017 #define HAMMER_CHKSPC_EMERGENCY	0
1018 
1019 extern struct vop_ops hammer_vnode_vops;
1020 extern struct vop_ops hammer_spec_vops;
1021 extern struct vop_ops hammer_fifo_vops;
1022 extern struct bio_ops hammer_bioops;
1023 
1024 extern int hammer_debug_io;
1025 extern int hammer_debug_general;
1026 extern int hammer_debug_inode;
1027 extern int hammer_debug_locks;
1028 extern int hammer_debug_btree;
1029 extern int hammer_debug_tid;
1030 extern int hammer_debug_recover;
1031 extern int hammer_debug_recover_faults;
1032 extern int hammer_debug_critical;
1033 extern int hammer_cluster_enable;
1034 extern int hammer_live_dedup;
1035 extern int hammer_tdmux_ticks;
1036 extern int hammer_count_fsyncs;
1037 extern int hammer_count_inodes;
1038 extern int hammer_count_iqueued;
1039 extern int hammer_count_reclaims;
1040 extern int hammer_count_records;
1041 extern int hammer_count_record_datas;
1042 extern int hammer_count_volumes;
1043 extern int hammer_count_buffers;
1044 extern int hammer_count_nodes;
1045 extern int64_t hammer_count_extra_space_used;
1046 extern int64_t hammer_stats_btree_lookups;
1047 extern int64_t hammer_stats_btree_searches;
1048 extern int64_t hammer_stats_btree_inserts;
1049 extern int64_t hammer_stats_btree_deletes;
1050 extern int64_t hammer_stats_btree_elements;
1051 extern int64_t hammer_stats_btree_splits;
1052 extern int64_t hammer_stats_btree_iterations;
1053 extern int64_t hammer_stats_btree_root_iterations;
1054 extern int64_t hammer_stats_record_iterations;
1055 extern int64_t hammer_stats_file_read;
1056 extern int64_t hammer_stats_file_write;
1057 extern int64_t hammer_stats_file_iopsr;
1058 extern int64_t hammer_stats_file_iopsw;
1059 extern int64_t hammer_stats_disk_read;
1060 extern int64_t hammer_stats_disk_write;
1061 extern int64_t hammer_stats_inode_flushes;
1062 extern int64_t hammer_stats_commits;
1063 extern int64_t hammer_stats_undo;
1064 extern int64_t hammer_stats_redo;
1065 extern long hammer_count_dirtybufspace;
1066 extern int hammer_count_refedbufs;
1067 extern int hammer_count_reservations;
1068 extern long hammer_count_io_running_read;
1069 extern long hammer_count_io_running_write;
1070 extern int hammer_count_io_locked;
1071 extern long hammer_limit_dirtybufspace;
1072 extern int hammer_limit_recs;
1073 extern int hammer_limit_inode_recs;
1074 extern int hammer_limit_reclaims;
1075 extern int hammer_live_dedup_cache_size;
1076 extern int hammer_limit_redo;
1077 extern int hammer_bio_count;
1078 extern int hammer_verify_zone;
1079 extern int hammer_verify_data;
1080 extern int hammer_write_mode;
1081 extern int hammer_double_buffer;
1082 extern int hammer_btree_full_undo;
1083 extern int hammer_yield_check;
1084 extern int hammer_fsync_mode;
1085 extern int hammer_autoflush;
1086 extern int64_t hammer_contention_count;
1087 
1088 extern int64_t hammer_live_dedup_vnode_bcmps;
1089 extern int64_t hammer_live_dedup_device_bcmps;
1090 extern int64_t hammer_live_dedup_findblk_failures;
1091 extern int64_t hammer_live_dedup_bmap_saves;
1092 
1093 void	hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
1094 			int error, const char *msg);
1095 int	hammer_vop_inactive(struct vop_inactive_args *);
1096 int	hammer_vop_reclaim(struct vop_reclaim_args *);
1097 int	hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
1098 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
1099 			hammer_inode_t dip, int64_t obj_id,
1100 			hammer_tid_t asof, u_int32_t localization,
1101 			int flags, int *errorp);
1102 struct hammer_inode *hammer_get_dummy_inode(hammer_transaction_t trans,
1103 			hammer_inode_t dip, int64_t obj_id,
1104 			hammer_tid_t asof, u_int32_t localization,
1105 			int flags, int *errorp);
1106 struct hammer_inode *hammer_find_inode(hammer_transaction_t trans,
1107 			int64_t obj_id, hammer_tid_t asof,
1108 			u_int32_t localization);
1109 void	hammer_scan_inode_snapshots(hammer_mount_t hmp,
1110 			hammer_inode_info_t iinfo,
1111 			int (*callback)(hammer_inode_t ip, void *data),
1112 			void *data);
1113 void	hammer_put_inode(struct hammer_inode *ip);
1114 void	hammer_put_inode_ref(struct hammer_inode *ip);
1115 void	hammer_inode_waitreclaims(hammer_transaction_t trans);
1116 void	hammer_inode_dirty(struct hammer_inode *ip);
1117 
1118 int	hammer_unload_volume(hammer_volume_t volume, void *data);
1119 int	hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
1120 
1121 int	hammer_unload_buffer(hammer_buffer_t buffer, void *data);
1122 int	hammer_install_volume(hammer_mount_t hmp, const char *volname,
1123 			struct vnode *devvp, void *data);
1124 int	hammer_mountcheck_volumes(hammer_mount_t hmp);
1125 int	hammer_get_installed_volumes(hammer_mount_t hmp);
1126 
1127 int	hammer_mem_add(hammer_record_t record);
1128 int	hammer_ip_lookup(hammer_cursor_t cursor);
1129 int	hammer_ip_first(hammer_cursor_t cursor);
1130 int	hammer_ip_next(hammer_cursor_t cursor);
1131 int	hammer_ip_resolve_data(hammer_cursor_t cursor);
1132 int	hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
1133 			hammer_tid_t tid);
1134 int	hammer_create_at_cursor(hammer_cursor_t cursor,
1135 			hammer_btree_leaf_elm_t leaf, void *udata, int mode);
1136 int	hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
1137 			hammer_tid_t delete_tid, u_int32_t delete_ts,
1138 			int track, int64_t *stat_bytes);
1139 int	hammer_ip_check_directory_empty(hammer_transaction_t trans,
1140 			hammer_inode_t ip);
1141 int	hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
1142 int	hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
1143 
1144 hammer_record_t
1145 	hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
1146 void	hammer_flush_record_done(hammer_record_t record, int error);
1147 void	hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
1148 void	hammer_rel_mem_record(hammer_record_t record);
1149 
1150 int	hammer_cursor_up(hammer_cursor_t cursor);
1151 int	hammer_cursor_up_locked(hammer_cursor_t cursor);
1152 int	hammer_cursor_down(hammer_cursor_t cursor);
1153 int	hammer_cursor_upgrade(hammer_cursor_t cursor);
1154 int	hammer_cursor_upgrade_node(hammer_cursor_t cursor);
1155 void	hammer_cursor_downgrade(hammer_cursor_t cursor);
1156 int	hammer_cursor_upgrade2(hammer_cursor_t c1, hammer_cursor_t c2);
1157 void	hammer_cursor_downgrade2(hammer_cursor_t c1, hammer_cursor_t c2);
1158 int	hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
1159 			int index);
1160 void	hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
1161 int	hammer_lock_ex_try(struct hammer_lock *lock);
1162 void	hammer_lock_sh(struct hammer_lock *lock);
1163 int	hammer_lock_sh_try(struct hammer_lock *lock);
1164 int	hammer_lock_upgrade(struct hammer_lock *lock, int shcount);
1165 void	hammer_lock_downgrade(struct hammer_lock *lock, int shcount);
1166 int	hammer_lock_status(struct hammer_lock *lock);
1167 void	hammer_unlock(struct hammer_lock *lock);
1168 void	hammer_ref(struct hammer_lock *lock);
1169 int	hammer_ref_interlock(struct hammer_lock *lock);
1170 int	hammer_ref_interlock_true(struct hammer_lock *lock);
1171 void	hammer_ref_interlock_done(struct hammer_lock *lock);
1172 void	hammer_rel(struct hammer_lock *lock);
1173 int	hammer_rel_interlock(struct hammer_lock *lock, int locked);
1174 void	hammer_rel_interlock_done(struct hammer_lock *lock, int orig_locked);
1175 int	hammer_get_interlock(struct hammer_lock *lock);
1176 int	hammer_try_interlock_norefs(struct hammer_lock *lock);
1177 void	hammer_put_interlock(struct hammer_lock *lock, int error);
1178 
1179 void	hammer_sync_lock_ex(hammer_transaction_t trans);
1180 void	hammer_sync_lock_sh(hammer_transaction_t trans);
1181 int	hammer_sync_lock_sh_try(hammer_transaction_t trans);
1182 void	hammer_sync_unlock(hammer_transaction_t trans);
1183 
1184 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
1185 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
1186 void	hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
1187 u_int64_t hammer_timespec_to_time(struct timespec *ts);
1188 int	hammer_str_to_tid(const char *str, int *ispfsp,
1189 			hammer_tid_t *tidp, u_int32_t *localizationp);
1190 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip,
1191 			int64_t namekey);
1192 void hammer_clear_objid(hammer_inode_t dip);
1193 void hammer_destroy_objid_cache(hammer_mount_t hmp);
1194 
1195 int hammer_dedup_crc_rb_compare(hammer_dedup_cache_t dc1,
1196 			hammer_dedup_cache_t dc2);
1197 int hammer_dedup_off_rb_compare(hammer_dedup_cache_t dc1,
1198 			hammer_dedup_cache_t dc2);
1199 hammer_dedup_cache_t hammer_dedup_cache_add(hammer_inode_t ip,
1200 			hammer_btree_leaf_elm_t leaf);
1201 hammer_dedup_cache_t hammer_dedup_cache_lookup(hammer_mount_t hmp,
1202 			hammer_crc_t crc);
1203 void hammer_dedup_cache_inval(hammer_mount_t hmp, hammer_off_t base_offset);
1204 void hammer_destroy_dedup_cache(hammer_mount_t hmp);
1205 void hammer_dump_dedup_cache(hammer_mount_t hmp);
1206 int hammer_dedup_validate(hammer_dedup_cache_t dcp, int zone, int bytes,
1207 			void *data);
1208 
1209 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
1210 			int bytes);
1211 void hammer_clear_undo_history(hammer_mount_t hmp);
1212 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
1213 int hammer_get_dtype(u_int8_t obj_type);
1214 u_int8_t hammer_get_obj_type(enum vtype vtype);
1215 int64_t hammer_directory_namekey(hammer_inode_t dip, const void *name, int len,
1216 			u_int32_t *max_iterationsp);
1217 int	hammer_nohistory(hammer_inode_t ip);
1218 
1219 int	hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
1220 			hammer_node_cache_t cache, hammer_inode_t ip);
1221 void	hammer_normalize_cursor(hammer_cursor_t cursor);
1222 void	hammer_done_cursor(hammer_cursor_t cursor);
1223 int	hammer_recover_cursor(hammer_cursor_t cursor);
1224 void	hammer_unlock_cursor(hammer_cursor_t cursor);
1225 int	hammer_lock_cursor(hammer_cursor_t cursor);
1226 hammer_cursor_t	hammer_push_cursor(hammer_cursor_t ocursor);
1227 void	hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
1228 
1229 void	hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
1230 void	hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
1231 			int index);
1232 void	hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
1233 			int index);
1234 void	hammer_cursor_moved_element(hammer_node_t oparent, int pindex,
1235 			hammer_node_t onode, int oindex,
1236 			hammer_node_t nnode, int nindex);
1237 void	hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent,
1238 			hammer_node_t nparent, int nindex);
1239 void	hammer_cursor_inserted_element(hammer_node_t node, int index);
1240 void	hammer_cursor_deleted_element(hammer_node_t node, int index);
1241 void	hammer_cursor_invalidate_cache(hammer_cursor_t cursor);
1242 
1243 int	hammer_btree_lookup(hammer_cursor_t cursor);
1244 int	hammer_btree_first(hammer_cursor_t cursor);
1245 int	hammer_btree_last(hammer_cursor_t cursor);
1246 int	hammer_btree_extract(hammer_cursor_t cursor, int flags);
1247 int	hammer_btree_iterate(hammer_cursor_t cursor);
1248 int	hammer_btree_iterate_reverse(hammer_cursor_t cursor);
1249 int	hammer_btree_insert(hammer_cursor_t cursor,
1250 			    hammer_btree_leaf_elm_t elm, int *doprop);
1251 int	hammer_btree_delete(hammer_cursor_t cursor, int *ndelete);
1252 void	hammer_btree_do_propagation(hammer_cursor_t cursor,
1253 			    hammer_pseudofs_inmem_t pfsm,
1254 			    hammer_btree_leaf_elm_t leaf);
1255 int	hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
1256 int	hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
1257 int	hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
1258 int	hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
1259 
1260 int	btree_set_parent_of_child(hammer_transaction_t trans,
1261                         hammer_node_t node,
1262                         hammer_btree_elm_t elm);
1263 void	hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node);
1264 void	hammer_btree_lcache_init(hammer_mount_t hmp, hammer_node_lock_t lcache,
1265 			int depth);
1266 void	hammer_btree_lcache_free(hammer_mount_t hmp, hammer_node_lock_t lcache);
1267 int	hammer_btree_lock_children(hammer_cursor_t cursor, int depth,
1268 			hammer_node_lock_t parent,
1269 			hammer_node_lock_t lcache);
1270 void	hammer_btree_lock_copy(hammer_cursor_t cursor,
1271 			hammer_node_lock_t parent);
1272 int	hammer_btree_sync_copy(hammer_cursor_t cursor,
1273 			hammer_node_lock_t parent);
1274 void	hammer_btree_unlock_children(hammer_mount_t hmp,
1275 			hammer_node_lock_t parent,
1276 			hammer_node_lock_t lcache);
1277 int	hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
1278 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans,
1279 			hammer_node_t node, int *parent_indexp,
1280 			int *errorp, int try_exclusive);
1281 
1282 void	hammer_print_btree_node(hammer_node_ondisk_t ondisk);
1283 void	hammer_print_btree_elm(hammer_btree_elm_t elm);
1284 
1285 void	*hammer_bread(hammer_mount_t hmp, hammer_off_t off,
1286 			int *errorp, struct hammer_buffer **bufferp);
1287 void	*hammer_bnew(hammer_mount_t hmp, hammer_off_t off,
1288 			int *errorp, struct hammer_buffer **bufferp);
1289 void	*hammer_bread_ext(hammer_mount_t hmp, hammer_off_t off, int bytes,
1290 			int *errorp, struct hammer_buffer **bufferp);
1291 void	*hammer_bnew_ext(hammer_mount_t hmp, hammer_off_t off, int bytes,
1292 			int *errorp, struct hammer_buffer **bufferp);
1293 
1294 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
1295 
1296 hammer_volume_t	hammer_get_volume(hammer_mount_t hmp,
1297 			int32_t vol_no, int *errorp);
1298 hammer_buffer_t	hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
1299 			int bytes, int isnew, int *errorp);
1300 void		hammer_sync_buffers(hammer_mount_t hmp,
1301 			hammer_off_t base_offset, int bytes);
1302 int		hammer_del_buffers(hammer_mount_t hmp,
1303 			hammer_off_t base_offset,
1304 			hammer_off_t zone2_offset, int bytes,
1305 			int report_conflicts);
1306 
1307 int		hammer_ref_volume(hammer_volume_t volume);
1308 int		hammer_ref_buffer(hammer_buffer_t buffer);
1309 void		hammer_flush_buffer_nodes(hammer_buffer_t buffer);
1310 
1311 void		hammer_rel_volume(hammer_volume_t volume, int locked);
1312 void		hammer_rel_buffer(hammer_buffer_t buffer, int locked);
1313 
1314 int		hammer_vfs_export(struct mount *mp, int op,
1315 			const struct export_args *export);
1316 hammer_node_t	hammer_get_node(hammer_transaction_t trans,
1317 			hammer_off_t node_offset, int isnew, int *errorp);
1318 void		hammer_ref_node(hammer_node_t node);
1319 hammer_node_t	hammer_ref_node_safe(hammer_transaction_t trans,
1320 			hammer_node_cache_t cache, int *errorp);
1321 void		hammer_rel_node(hammer_node_t node);
1322 void		hammer_delete_node(hammer_transaction_t trans,
1323 			hammer_node_t node);
1324 void		hammer_cache_node(hammer_node_cache_t cache,
1325 			hammer_node_t node);
1326 void		hammer_uncache_node(hammer_node_cache_t cache);
1327 void		hammer_flush_node(hammer_node_t node, int locked);
1328 
1329 void hammer_dup_buffer(struct hammer_buffer **bufferp,
1330 			struct hammer_buffer *buffer);
1331 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans,
1332 			hammer_off_t hint, int *errorp);
1333 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1334 			u_int16_t rec_type, hammer_off_t *data_offsetp,
1335 			struct hammer_buffer **data_bufferp,
1336 			hammer_off_t hint, int *errorp);
1337 
1338 int hammer_generate_undo(hammer_transaction_t trans,
1339 			hammer_off_t zone_offset, void *base, int len);
1340 int hammer_generate_redo(hammer_transaction_t trans, hammer_inode_t ip,
1341 			hammer_off_t file_offset, u_int32_t flags,
1342 			void *base, int len);
1343 void hammer_generate_redo_sync(hammer_transaction_t trans);
1344 void hammer_redo_fifo_start_flush(hammer_inode_t ip);
1345 void hammer_redo_fifo_end_flush(hammer_inode_t ip);
1346 
1347 void hammer_format_undo(void *base, u_int32_t seqno);
1348 int hammer_upgrade_undo_4(hammer_transaction_t trans);
1349 
1350 void hammer_put_volume(struct hammer_volume *volume, int flush);
1351 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
1352 
1353 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1354 			hammer_off_t owner, int *errorp);
1355 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1356 			hammer_off_t owner, int *errorp);
1357 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp);
1358 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1359 			int bytes, hammer_off_t hint, int *errorp);
1360 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1361 			int bytes, hammer_off_t *zone_offp, int *errorp);
1362 hammer_reserve_t hammer_blockmap_reserve_dedup(hammer_mount_t hmp, int zone,
1363 			int bytes, hammer_off_t zone_offset, int *errorp);
1364 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1365 			hammer_reserve_t resv);
1366 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1367 void hammer_blockmap_free(hammer_transaction_t trans,
1368 			hammer_off_t zone_offset, int bytes);
1369 int hammer_blockmap_dedup(hammer_transaction_t trans,
1370 			hammer_off_t zone_offset, int bytes);
1371 int hammer_blockmap_finalize(hammer_transaction_t trans,
1372 			hammer_reserve_t resv,
1373 			hammer_off_t zone_offset, int bytes);
1374 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t zone_offset,
1375 			int *curp, int *errorp);
1376 hammer_off_t hammer_blockmap_lookup_verify(hammer_mount_t hmp,
1377 			hammer_off_t zone_offset, int *errorp);
1378 
1379 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t zone_offset,
1380 			int *errorp);
1381 int64_t hammer_undo_used(hammer_transaction_t trans);
1382 int64_t hammer_undo_space(hammer_transaction_t trans);
1383 int64_t hammer_undo_max(hammer_mount_t hmp);
1384 int hammer_undo_reclaim(hammer_io_t io);
1385 
1386 void hammer_start_transaction(struct hammer_transaction *trans,
1387 			      struct hammer_mount *hmp);
1388 void hammer_simple_transaction(struct hammer_transaction *trans,
1389 			      struct hammer_mount *hmp);
1390 void hammer_start_transaction_fls(struct hammer_transaction *trans,
1391 			          struct hammer_mount *hmp);
1392 void hammer_done_transaction(struct hammer_transaction *trans);
1393 hammer_tid_t hammer_alloc_tid(hammer_mount_t hmp, int count);
1394 
1395 void hammer_modify_inode(hammer_transaction_t trans, hammer_inode_t ip, int flags);
1396 void hammer_flush_inode(hammer_inode_t ip, int flags);
1397 void hammer_flush_inode_done(hammer_inode_t ip, int error);
1398 void hammer_wait_inode(hammer_inode_t ip);
1399 
1400 int  hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
1401 			struct ucred *cred, struct hammer_inode *dip,
1402 			const char *name, int namelen,
1403 			hammer_pseudofs_inmem_t pfsm,
1404 			struct hammer_inode **ipp);
1405 void hammer_rel_inode(hammer_inode_t ip, int flush);
1406 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1407 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1408 int hammer_redo_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1409 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1410 
1411 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1412 void hammer_test_inode(hammer_inode_t dip);
1413 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1414 int hammer_update_atime_quick(hammer_inode_t ip);
1415 
1416 int  hammer_ip_add_directory(struct hammer_transaction *trans,
1417 			hammer_inode_t dip, const char *name, int bytes,
1418 			hammer_inode_t nip);
1419 int  hammer_ip_del_directory(struct hammer_transaction *trans,
1420 			hammer_cursor_t cursor, hammer_inode_t dip,
1421 			hammer_inode_t ip);
1422 void hammer_ip_replace_bulk(hammer_mount_t hmp, hammer_record_t record);
1423 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1424 			void *data, int bytes, int *errorp);
1425 int  hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
1426 int  hammer_ip_add_record(struct hammer_transaction *trans,
1427 			hammer_record_t record);
1428 int  hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1429 			int64_t ran_beg, int64_t ran_end, int truncating);
1430 int  hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1431 			int *countp);
1432 int  hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1433 			int64_t offset, void *data, int bytes);
1434 int  hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1435 hammer_pseudofs_inmem_t  hammer_load_pseudofs(hammer_transaction_t trans,
1436 			u_int32_t localization, int *errorp);
1437 int  hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1438 			hammer_pseudofs_inmem_t pfsm);
1439 int  hammer_save_pseudofs(hammer_transaction_t trans,
1440 			hammer_pseudofs_inmem_t pfsm);
1441 int  hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization);
1442 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1443 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1444 			struct ucred *cred);
1445 
1446 void hammer_io_init(hammer_io_t io, hammer_volume_t volume,
1447 			enum hammer_io_type type);
1448 int hammer_io_read(struct vnode *devvp, struct hammer_io *io, int limit);
1449 void hammer_io_advance(struct hammer_io *io);
1450 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
1451 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1452 struct buf *hammer_io_release(struct hammer_io *io, int flush);
1453 void hammer_io_flush(struct hammer_io *io, int reclaim);
1454 void hammer_io_wait(struct hammer_io *io);
1455 void hammer_io_waitdep(struct hammer_io *io);
1456 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident, int doflush);
1457 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1458 			hammer_btree_leaf_elm_t leaf);
1459 int hammer_io_indirect_read(hammer_mount_t hmp, struct bio *bio,
1460 			hammer_btree_leaf_elm_t leaf);
1461 int hammer_io_direct_write(hammer_mount_t hmp, struct bio *bio,
1462 			hammer_record_t record);
1463 void hammer_io_direct_wait(hammer_record_t record);
1464 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1465 void hammer_io_write_interlock(hammer_io_t io);
1466 void hammer_io_done_interlock(hammer_io_t io);
1467 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1468 void hammer_io_clear_modlist(struct hammer_io *io);
1469 void hammer_io_flush_sync(hammer_mount_t hmp);
1470 void hammer_io_clear_error(struct hammer_io *io);
1471 void hammer_io_clear_error_noassert(struct hammer_io *io);
1472 void hammer_io_notmeta(hammer_buffer_t buffer);
1473 void hammer_io_limit_backlog(hammer_mount_t hmp);
1474 
1475 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1476 			void *base, int len);
1477 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1478 			void *base, int len);
1479 void hammer_modify_volume_done(hammer_volume_t volume);
1480 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1481 
1482 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1483 			struct hammer_ioc_reblock *reblock);
1484 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip,
1485 			struct hammer_ioc_rebalance *rebal);
1486 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1487 			struct hammer_ioc_prune *prune);
1488 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1489 			struct hammer_ioc_mirror_rw *mirror);
1490 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1491 			struct hammer_ioc_mirror_rw *mirror);
1492 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1493 			struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1494 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1495                         struct hammer_ioc_pseudofs_rw *pfs);
1496 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1497                         struct hammer_ioc_pseudofs_rw *pfs);
1498 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1499                         struct hammer_ioc_pseudofs_rw *pfs);
1500 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1501                         struct hammer_ioc_pseudofs_rw *pfs);
1502 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1503                         struct hammer_ioc_pseudofs_rw *pfs);
1504 int hammer_ioc_iterate_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1505 			struct hammer_ioc_pfs_iterate *pi);
1506 int hammer_ioc_volume_add(hammer_transaction_t trans, hammer_inode_t ip,
1507                         struct hammer_ioc_volume *ioc);
1508 int hammer_ioc_volume_del(hammer_transaction_t trans, hammer_inode_t ip,
1509                         struct hammer_ioc_volume *ioc);
1510 int hammer_ioc_volume_list(hammer_transaction_t trans, hammer_inode_t ip,
1511 			struct hammer_ioc_volume_list *ioc);
1512 int hammer_ioc_dedup(hammer_transaction_t trans, hammer_inode_t ip,
1513 			struct hammer_ioc_dedup *dedup);
1514 
1515 int hammer_signal_check(hammer_mount_t hmp);
1516 
1517 void hammer_flusher_create(hammer_mount_t hmp);
1518 void hammer_flusher_destroy(hammer_mount_t hmp);
1519 void hammer_flusher_sync(hammer_mount_t hmp);
1520 int  hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1521 int  hammer_flusher_async_one(hammer_mount_t hmp);
1522 int hammer_flusher_running(hammer_mount_t hmp);
1523 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1524 void hammer_flusher_wait_next(hammer_mount_t hmp);
1525 int  hammer_flusher_meta_limit(hammer_mount_t hmp);
1526 int  hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1527 int  hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1528 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1529 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1530 int  hammer_flusher_haswork(hammer_mount_t hmp);
1531 int  hammer_flush_dirty(hammer_mount_t hmp, int max_count);
1532 void hammer_flusher_flush_undos(hammer_mount_t hmp, int already_flushed);
1533 
1534 int hammer_recover_stage1(hammer_mount_t hmp, hammer_volume_t rootvol);
1535 int hammer_recover_stage2(hammer_mount_t hmp, hammer_volume_t rootvol);
1536 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1537 			hammer_volume_t root_volume, int final);
1538 
1539 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1540 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1541 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1542 
1543 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1544 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1545 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1546 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1547 udev_t hammer_fsid_to_udev(uuid_t *uuid);
1548 
1549 
1550 int hammer_blocksize(int64_t file_offset);
1551 int hammer_blockoff(int64_t file_offset);
1552 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1553 
1554 /*
1555  * Shortcut for _hammer_checkspace(), used all over the code.
1556  */
1557 static __inline int
1558 hammer_checkspace(hammer_mount_t hmp, int slop)
1559 {
1560 	return(_hammer_checkspace(hmp, slop, NULL));
1561 }
1562 
1563 static __inline void
1564 hammer_wait_mem_record(hammer_record_t record)
1565 {
1566 	hammer_wait_mem_record_ident(record, "hmmwai");
1567 }
1568 
1569 static __inline void
1570 hammer_lock_ex(struct hammer_lock *lock)
1571 {
1572 	hammer_lock_ex_ident(lock, "hmrlck");
1573 }
1574 
1575 static __inline void
1576 hammer_modify_volume_noundo(hammer_transaction_t trans, hammer_volume_t volume)
1577 {
1578 	hammer_modify_volume(trans, volume, NULL, 0);
1579 }
1580 
1581 static __inline void
1582 hammer_modify_buffer_noundo(hammer_transaction_t trans, hammer_buffer_t buffer)
1583 {
1584 	hammer_modify_buffer(trans, buffer, NULL, 0);
1585 }
1586 
1587 /*
1588  * Indicate that a B-Tree node is being modified.
1589  */
1590 static __inline void
1591 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1592 {
1593 	KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1594 	hammer_modify_buffer(trans, node->buffer, NULL, 0);
1595 }
1596 
1597 static __inline void
1598 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1599 {
1600 	KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1601 	hammer_modify_buffer(trans, node->buffer,
1602 			     node->ondisk, sizeof(*node->ondisk));
1603 }
1604 
1605 static __inline void
1606 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1607 		   void *base, int len)
1608 {
1609 	hammer_crc_t *crcptr;
1610 
1611 	KKASSERT((char *)base >= (char *)node->ondisk &&
1612 		 (char *)base + len <=
1613 		    (char *)node->ondisk + sizeof(*node->ondisk));
1614 	KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1615 
1616 	if (hammer_btree_full_undo) {
1617 		hammer_modify_node_all(trans, node);
1618 	} else {
1619 		hammer_modify_buffer(trans, node->buffer, base, len);
1620 		crcptr = &node->ondisk->crc;
1621 		hammer_modify_buffer(trans, node->buffer,
1622 				     crcptr, sizeof(hammer_crc_t));
1623 		--node->buffer->io.modify_refs;	/* only want one ref */
1624 	}
1625 }
1626 
1627 /*
1628  * Indicate that the specified modifications have been completed.
1629  *
1630  * Do not try to generate the crc here, it's very expensive to do and a
1631  * sequence of insertions or deletions can result in many calls to this
1632  * function on the same node.
1633  */
1634 static __inline void
1635 hammer_modify_node_done(hammer_node_t node)
1636 {
1637 	node->flags |= HAMMER_NODE_CRCGOOD;
1638 	if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1639 		node->flags |= HAMMER_NODE_NEEDSCRC;
1640 		node->buffer->io.gencrc = 1;
1641 		hammer_ref_node(node);
1642 	}
1643 	hammer_modify_buffer_done(node->buffer);
1644 }
1645 
1646 /*
1647  * Lookup a blockmap offset.
1648  */
1649 static __inline hammer_off_t
1650 hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t zone_offset,
1651 			int *errorp)
1652 {
1653 #if defined INVARIANTS
1654 	int zone = HAMMER_ZONE_DECODE(zone_offset);
1655 	KKASSERT(zone >= HAMMER_ZONE2_MAPPED_INDEX && zone < HAMMER_MAX_ZONES);
1656 #endif
1657 
1658 	/*
1659 	 * We can actually skip blockmap verify by default,
1660 	 * as normal blockmaps are now direct-mapped onto the freemap
1661 	 * and so represent zone-2 addresses.
1662 	 */
1663 	if (hammer_verify_zone == 0) {
1664 		*errorp = 0;
1665 		return hammer_xlate_to_zone2(zone_offset);
1666 	}
1667 
1668 	return hammer_blockmap_lookup_verify(hmp, zone_offset, errorp);
1669 }
1670 
1671 #define hammer_modify_volume_field(trans, vol, field)		\
1672 	hammer_modify_volume(trans, vol, &(vol)->ondisk->field,	\
1673 			     sizeof((vol)->ondisk->field))
1674 
1675 #define hammer_modify_node_field(trans, node, field)		\
1676 	hammer_modify_node(trans, node, &(node)->ondisk->field,	\
1677 			     sizeof((node)->ondisk->field))
1678 
1679 /*
1680  * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1681  * created directories for HAMMER version 2 or greater and causes
1682  * directory entries to be placed the inode localization zone in
1683  * the B-Tree instead of the misc zone.
1684  *
1685  * This greatly improves localization between directory entries and
1686  * inodes
1687  */
1688 static __inline u_int32_t
1689 hammer_dir_localization(hammer_inode_t dip)
1690 {
1691 	return(HAMMER_DIR_INODE_LOCALIZATION(&dip->ino_data));
1692 }
1693 
1694 #define hkprintf(format, args...)			\
1695 	kprintf("HAMMER: "format,## args)
1696 #define hvkprintf(vol, format, args...)			\
1697 	kprintf("HAMMER(%s) "format, vol->ondisk->vol_name,## args)
1698 #define hmkprintf(hmp, format, args...)			\
1699 	kprintf("HAMMER(%s) "format, hmp->mp->mnt_stat.f_mntfromname,## args)
1700 #define hdkprintf(format, args...)			\
1701 	kprintf("%s: "format, __func__,## args)
1702 
1703 #define hkrateprintf(rate , format, args...)		\
1704 	krateprintf(rate, "HAMMER: "format,## args)
1705 #define hvkrateprintf(rate, vol, format, args...)	\
1706 	krateprintf(rate, "HAMMER(%s) "format, vol->ondisk->vol_name,## args)
1707 #define hmkrateprintf(rate, hmp, format, args...)	\
1708 	krateprintf(rate, "HAMMER(%s) "format, hmp->mp->mnt_stat.f_mntfromname,## args)
1709 #define hdkrateprintf(rate, format, args...)		\
1710 	krateprintf(rate, "%s: "format, __func__,## args)
1711 
1712 #define hpanic(format, args...)				\
1713 	panic("%s: "format, __func__,## args)
1714 #endif  /* _KERNEL */
1715 
1716 #endif /* !VFS_HAMMER_HAMMER_H_ */
1717