xref: /dragonfly/sys/vfs/hammer/hammer.h (revision 87df5a71)
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 #include "hammer_crc.h"
73 
74 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
75 
76 MALLOC_DECLARE(M_HAMMER);
77 
78 /*
79  * Kernel trace
80  */
81 #if !defined(KTR_HAMMER)
82 #define KTR_HAMMER	KTR_ALL
83 #endif
84 /* KTR_INFO_MASTER_EXTERN(hammer); */
85 
86 /*
87  * Misc structures
88  */
89 struct hammer_mount;
90 struct hammer_inode;
91 struct hammer_volume;
92 struct hammer_buffer;
93 struct hammer_node;
94 struct hammer_undo;
95 struct hammer_reserve;
96 struct hammer_io;
97 
98 /*
99  * Key structure used for custom RB tree inode lookups.  This prototypes
100  * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
101  */
102 typedef struct hammer_inode_info {
103 	int64_t		obj_id;		/* (key) object identifier */
104 	hammer_tid_t	obj_asof;	/* (key) snapshot transid or 0 */
105 	uint32_t	obj_localization; /* (key) pseudo-fs id for upper 16 bits */
106 	union {
107 		hammer_btree_leaf_elm_t leaf;
108 	} u;
109 } *hammer_inode_info_t;
110 
111 typedef enum hammer_transaction_type {
112 	HAMMER_TRANS_RO,
113 	HAMMER_TRANS_STD,
114 	HAMMER_TRANS_FLS
115 } hammer_transaction_type_t;
116 
117 /*
118  * HAMMER Transaction tracking
119  */
120 typedef struct hammer_transaction {
121 	hammer_transaction_type_t type;
122 	struct hammer_mount *hmp;
123 	hammer_tid_t	tid;
124 	uint64_t	time;
125 	uint32_t	time32;
126 	int		sync_lock_refs;
127 	int		flags;
128 	struct hammer_volume *rootvol;
129 } *hammer_transaction_t;
130 
131 #define HAMMER_TRANSF_NEWINODE	0x0001
132 #define HAMMER_TRANSF_CRCDOM	0x0004	/* EDOM on CRC error, less critical */
133 
134 /*
135  * HAMMER locks
136  */
137 struct hammer_lock {
138 	volatile u_int	refs;		/* active references */
139 	volatile u_int	lockval;	/* lock count and control bits */
140 	struct thread	*lowner;	/* owner if exclusively held */
141 	struct thread	*rowner;	/* owner if exclusively held */
142 };
143 
144 #define HAMMER_REFS_LOCKED	0x40000000	/* transition check */
145 #define HAMMER_REFS_WANTED	0x20000000	/* transition check */
146 #define HAMMER_REFS_CHECK	0x10000000	/* transition check */
147 
148 #define HAMMER_REFS_FLAGS	(HAMMER_REFS_LOCKED | \
149 				 HAMMER_REFS_WANTED | \
150 				 HAMMER_REFS_CHECK)
151 
152 #define HAMMER_LOCKF_EXCLUSIVE	0x40000000
153 #define HAMMER_LOCKF_WANTED	0x20000000
154 
155 static __inline int
156 hammer_notlocked(struct hammer_lock *lock)
157 {
158 	return(lock->lockval == 0);
159 }
160 
161 static __inline int
162 hammer_islocked(struct hammer_lock *lock)
163 {
164 	return(lock->lockval != 0);
165 }
166 
167 /*
168  * Returns the number of refs on the object.
169  */
170 static __inline int
171 hammer_isactive(struct hammer_lock *lock)
172 {
173 	return(lock->refs & ~HAMMER_REFS_FLAGS);
174 }
175 
176 static __inline int
177 hammer_oneref(struct hammer_lock *lock)
178 {
179 	return((lock->refs & ~HAMMER_REFS_FLAGS) == 1);
180 }
181 
182 static __inline int
183 hammer_norefs(struct hammer_lock *lock)
184 {
185 	return((lock->refs & ~HAMMER_REFS_FLAGS) == 0);
186 }
187 
188 static __inline int
189 hammer_norefsorlock(struct hammer_lock *lock)
190 {
191 	return(lock->refs == 0);
192 }
193 
194 static __inline int
195 hammer_refsorlock(struct hammer_lock *lock)
196 {
197 	return(lock->refs != 0);
198 }
199 
200 /*
201  * Return if we specifically own the lock exclusively.
202  */
203 static __inline int
204 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
205 {
206 	if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) &&
207 	    lock->lowner == td) {
208 		return(1);
209 	}
210 	return(0);
211 }
212 
213 /*
214  * Flush state, used by various structures
215  */
216 typedef enum hammer_inode_state {
217 	HAMMER_FST_IDLE,
218 	HAMMER_FST_SETUP,
219 	HAMMER_FST_FLUSH
220 } hammer_inode_state_t;
221 
222 /*
223  * Pseudo-filesystem extended data tracking
224  */
225 struct hammer_pseudofs_inmem;
226 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
227 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
228 	      hammer_pfs_rb_compare, uint32_t);
229 
230 typedef struct hammer_pseudofs_inmem {
231 	RB_ENTRY(hammer_pseudofs_inmem)	rb_node;
232 	struct hammer_lock	lock;
233 	uint32_t		localization;
234 	hammer_tid_t		create_tid;
235 	int			flags;
236 	udev_t			fsid_udev;
237 	struct hammer_pseudofs_data pfsd;
238 } *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	((uint64_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 	uint32_t			bm0;
260 	uint32_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_cache;
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 RB_HEAD(hammer_dedup_off_rb_tree, hammer_dedup_cache);
283 RB_PROTOTYPE2(hammer_dedup_off_rb_tree, hammer_dedup_cache, off_entry,
284 		hammer_dedup_off_rb_compare, hammer_off_t);
285 
286 typedef struct hammer_dedup_cache {
287 	RB_ENTRY(hammer_dedup_cache) crc_entry;
288 	RB_ENTRY(hammer_dedup_cache) off_entry;
289 	TAILQ_ENTRY(hammer_dedup_cache) lru_entry;
290 	struct hammer_mount *hmp;
291 	int64_t obj_id;
292 	uint32_t localization;
293 	off_t file_offset;
294 	int bytes;
295 	hammer_off_t data_offset;
296 	hammer_crc_t crc;
297 } *hammer_dedup_cache_t;
298 
299 /*
300  * Structure used to organize flush groups.  Flush groups must be
301  * organized into chunks in order to avoid blowing out the UNDO FIFO.
302  * Without this a 'sync' could end up flushing 50,000 inodes in a single
303  * transaction.
304  */
305 RB_HEAD(hammer_fls_rb_tree, hammer_inode);
306 RB_PROTOTYPE(hammer_fls_rb_tree, hammer_inode, rb_flsnode,
307 	      hammer_ino_rb_compare);
308 
309 typedef struct hammer_flush_group {
310 	TAILQ_ENTRY(hammer_flush_group)	flush_entry;
311 	struct hammer_fls_rb_tree	flush_tree;
312 	int				seq;		/* our seq no */
313 	int				total_count;	/* record load */
314 	int				running;	/* group is running */
315 	int				closed;
316 	int				refs;
317 } *hammer_flush_group_t;
318 
319 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
320 
321 /*
322  * Structure used to represent an inode in-memory.
323  *
324  * The record and data associated with an inode may be out of sync with
325  * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
326  * clear).
327  *
328  * An inode may also hold a cache of unsynchronized records, used for
329  * database and directories only.  Unsynchronized regular file data is
330  * stored in the buffer cache.
331  *
332  * NOTE: A file which is created and destroyed within the initial
333  * synchronization period can wind up not doing any disk I/O at all.
334  *
335  * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
336  */
337 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
338 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
339 	      hammer_ino_rb_compare, hammer_inode_info_t);
340 
341 RB_HEAD(hammer_redo_rb_tree, hammer_inode);
342 RB_PROTOTYPE2(hammer_redo_rb_tree, hammer_inode, rb_redonode,
343 	      hammer_redo_rb_compare, hammer_off_t);
344 
345 struct hammer_record;
346 RB_HEAD(hammer_rec_rb_tree, hammer_record);
347 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
348 	      hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
349 
350 TAILQ_HEAD(hammer_record_list, hammer_record);
351 TAILQ_HEAD(hammer_node_list, hammer_node);
352 
353 typedef struct hammer_inode {
354 	RB_ENTRY(hammer_inode)	rb_node;
355 	hammer_inode_state_t	flush_state;
356 	hammer_flush_group_t	flush_group;
357 	RB_ENTRY(hammer_inode)	rb_flsnode;	/* when on flush list */
358 	RB_ENTRY(hammer_inode)	rb_redonode;	/* when INODE_RDIRTY is set */
359 	struct hammer_record_list target_list;	/* target of dependant recs */
360 	int64_t			obj_id;		/* (key) object identifier */
361 	hammer_tid_t		obj_asof;	/* (key) snapshot or 0 */
362 	uint32_t		obj_localization; /* (key) pseudo-fs id for upper 16 bits */
363 	struct hammer_mount	*hmp;
364 	hammer_objid_cache_t	objid_cache;
365 	int			flags;
366 	int			error;		/* flush error */
367 	int			cursor_ip_refs;	/* sanity */
368 #if 0
369 	int			cursor_exclreq_count;
370 #endif
371 	int			rsv_recs;
372 	struct vnode		*vp;
373 	hammer_pseudofs_inmem_t	pfsm;
374 	struct lockf		advlock;
375 	struct hammer_lock	lock;		/* sync copy interlock */
376 	off_t			trunc_off;
377 	struct hammer_btree_leaf_elm ino_leaf;  /* in-memory cache */
378 	struct hammer_inode_data ino_data;	/* in-memory cache */
379 	struct hammer_rec_rb_tree rec_tree;	/* in-memory cache */
380 	int			rec_generation;
381 
382 	/*
383 	 * search initiate cache
384 	 * cache[0] - this inode
385 	 * cache[1] - related data, the content depends on situations
386 	 * cache[2] - for dip to cache ip to shortcut B-Tree search
387 	 * cache[3] - related data copied from dip to a new ip's cache[1]
388 	 */
389 	struct hammer_node_cache cache[4];
390 
391 	/*
392 	 * When a demark is created to synchronize an inode to
393 	 * disk, certain fields are copied so the front-end VOPs
394 	 * can continue to run in parallel with the synchronization
395 	 * occuring in the background.
396 	 */
397 	int		sync_flags;		/* to-sync flags cache */
398 	off_t		sync_trunc_off;		/* to-sync truncation */
399 	off_t		save_trunc_off;		/* write optimization */
400 	struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
401 	struct hammer_inode_data sync_ino_data; /* to-sync cache */
402 	size_t		redo_count;
403 
404 	/*
405 	 * Track the earliest offset in the UNDO/REDO FIFO containing
406 	 * REDO records.  This is staged to the backend during flush
407 	 * sequences.  While the inode is staged redo_fifo_next is used
408 	 * to track the earliest offset for rotation into redo_fifo_start
409 	 * on completion of the flush.
410 	 */
411 	hammer_off_t	redo_fifo_start;
412 	hammer_off_t	redo_fifo_next;
413 } *hammer_inode_t;
414 
415 #define VTOI(vp)	((hammer_inode_t)(vp)->v_data)
416 
417 /*
418  * NOTE: DDIRTY does not include atime or mtime and does not include
419  *	 write-append size changes.  SDIRTY handles write-append size
420  *	 changes.
421  *
422  *	 REDO indicates that REDO logging is active, creating a definitive
423  *	 stream of REDO records in the UNDO/REDO log for writes and
424  *	 truncations, including boundary records when/if REDO is turned off.
425  *	 REDO is typically enabled by fsync() and turned off if excessive
426  *	 writes without an fsync() occurs.
427  *
428  *	 RDIRTY indicates that REDO records were laid down in the UNDO/REDO
429  *	 FIFO (even if REDO is turned off some might still be active) and
430  *	 still being tracked for this inode.  See hammer_redo.c
431  */
432 #define HAMMER_INODE_DDIRTY	0x0001	/* in-memory ino_data is dirty */
433 					/* (not including atime/mtime) */
434 #define HAMMER_INODE_RSV_INODES	0x0002	/* hmp->rsv_inodes bumped */
435 #define HAMMER_INODE_CONN_DOWN	0x0004	/* include in downward recursion */
436 #define HAMMER_INODE_XDIRTY	0x0008	/* in-memory records */
437 #define HAMMER_INODE_ONDISK	0x0010	/* inode is on-disk (else not yet) */
438 #define HAMMER_INODE_FLUSH	0x0020	/* flush on last ref */
439 #define HAMMER_INODE_DELETED	0x0080	/* inode delete (backend) */
440 #define HAMMER_INODE_DELONDISK	0x0100	/* delete synchronized to disk */
441 #define HAMMER_INODE_RO		0x0200	/* read-only (because of as-of) */
442 #define HAMMER_INODE_RECSW	0x0400	/* waiting on data record flush */
443 #define HAMMER_INODE_DONDISK	0x0800	/* data records may be on disk */
444 #define HAMMER_INODE_BUFS	0x1000	/* dirty high level bps present */
445 #define HAMMER_INODE_REFLUSH	0x2000	/* flush on dependancy / reflush */
446 #define HAMMER_INODE_RECLAIM	0x4000	/* trying to reclaim */
447 #define HAMMER_INODE_FLUSHW	0x8000	/* someone waiting for flush */
448 
449 #define HAMMER_INODE_TRUNCATED	0x00010000
450 #define HAMMER_INODE_DELETING	0x00020000 /* inode delete request (frontend)*/
451 #define HAMMER_INODE_RESIGNAL	0x00040000 /* re-signal on re-flush */
452 #define HAMMER_INODE_ATIME	0x00100000 /* in-memory atime modified */
453 #define HAMMER_INODE_MTIME	0x00200000 /* in-memory mtime modified */
454 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
455 #define HAMMER_INODE_DUMMY	0x00800000 /* dummy inode covering bad file */
456 #define HAMMER_INODE_SDIRTY	0x01000000 /* in-memory ino_data.size is dirty*/
457 #define HAMMER_INODE_REDO	0x02000000 /* REDO logging active */
458 #define HAMMER_INODE_RDIRTY	0x04000000 /* REDO records active in fifo */
459 #define HAMMER_INODE_SLAVEFLUSH	0x08000000 /* being flushed by slave */
460 
461 #define HAMMER_INODE_MODMASK	(HAMMER_INODE_DDIRTY|HAMMER_INODE_SDIRTY|   \
462 				 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS|	    \
463 				 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME|     \
464 				 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
465 
466 #define HAMMER_INODE_MODMASK_NOXDIRTY	\
467 				(HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
468 
469 #define HAMMER_INODE_MODMASK_NOREDO	\
470 				(HAMMER_INODE_DDIRTY|			    \
471 				 HAMMER_INODE_XDIRTY|			    \
472 				 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
473 
474 #define HAMMER_FLUSH_SIGNAL	0x0001
475 #define HAMMER_FLUSH_RECURSION	0x0002
476 
477 /*
478  * Used by the inode reclaim code to pipeline reclaims and avoid
479  * blowing out kernel memory or letting the flusher get too far
480  * behind.  The reclaim wakes up when count reaches 0 or the
481  * timer expires.
482  */
483 struct hammer_reclaim {
484 	TAILQ_ENTRY(hammer_reclaim) entry;
485 	int	count;
486 };
487 
488 /*
489  * Track who is creating the greatest burden on the
490  * inode cache.
491  */
492 struct hammer_inostats {
493 	pid_t		pid;	/* track user process */
494 	int		ltick;	/* last tick */
495 	int		count;	/* count (degenerates) */
496 };
497 
498 #define HAMMER_INOSTATS_HSIZE	32
499 #define HAMMER_INOSTATS_HMASK	(HAMMER_INOSTATS_HSIZE - 1)
500 
501 /*
502  * Structure used to represent an unsynchronized record in-memory.  These
503  * records typically represent directory entries.  Only non-historical
504  * records are kept in-memory.
505  *
506  * Records are organized as a per-inode RB-Tree.  If the inode is not
507  * on disk then neither are any records and the in-memory record tree
508  * represents the entire contents of the inode.  If the inode is on disk
509  * then the on-disk B-Tree is scanned in parallel with the in-memory
510  * RB-Tree to synthesize the current state of the file.
511  *
512  * Records are also used to enforce the ordering of directory create/delete
513  * operations.  A new inode will not be flushed to disk unless its related
514  * directory entry is also being flushed at the same time.  A directory entry
515  * will not be removed unless its related inode is also being removed at the
516  * same time.
517  */
518 typedef enum hammer_record_type {
519 	HAMMER_MEM_RECORD_GENERAL,	/* misc record */
520 	HAMMER_MEM_RECORD_INODE,	/* inode record */
521 	HAMMER_MEM_RECORD_ADD,		/* positive memory cache record */
522 	HAMMER_MEM_RECORD_DEL,		/* negative delete-on-disk record */
523 	HAMMER_MEM_RECORD_DATA		/* bulk-data record w/on-disk ref */
524 } hammer_record_type_t;
525 
526 typedef struct hammer_record {
527 	RB_ENTRY(hammer_record)		rb_node;
528 	TAILQ_ENTRY(hammer_record)	target_entry;
529 	hammer_inode_state_t		flush_state;
530 	hammer_flush_group_t		flush_group;
531 	hammer_record_type_t		type;
532 	struct hammer_lock		lock;
533 	struct hammer_reserve		*resv;
534 	hammer_inode_t			ip;
535 	hammer_inode_t			target_ip;
536 	struct hammer_btree_leaf_elm	leaf;
537 	hammer_data_ondisk_t		data;
538 	int				flags;
539 	int				gflags;
540 	hammer_off_t			zone2_offset;	/* direct-write only */
541 } *hammer_record_t;
542 
543 /*
544  * Record flags.  Note that FE can only be set by the frontend if the
545  * record has not been interlocked by the backend w/ BE.
546  */
547 #define HAMMER_RECF_ALLOCDATA		0x0001
548 #define HAMMER_RECF_ONRBTREE		0x0002
549 #define HAMMER_RECF_DELETED_FE		0x0004	/* deleted (frontend) */
550 #define HAMMER_RECF_DELETED_BE		0x0008	/* deleted (backend) */
551 #define HAMMER_RECF_COMMITTED		0x0010	/* committed to the B-Tree */
552 #define HAMMER_RECF_INTERLOCK_BE	0x0020	/* backend interlock */
553 #define HAMMER_RECF_WANTED		0x0040	/* wanted by the frontend */
554 #define HAMMER_RECF_DEDUPED		0x0080	/* will be live-dedup'ed */
555 #define HAMMER_RECF_CONVERT_DELETE	0x0100	/* special case */
556 #define HAMMER_RECF_REDO		0x1000	/* REDO was laid down */
557 
558 /*
559  * These flags must be separate to deal with SMP races
560  */
561 #define HAMMER_RECG_DIRECT_IO		0x0001	/* related direct I/O running*/
562 #define HAMMER_RECG_DIRECT_WAIT		0x0002	/* related direct I/O running*/
563 #define HAMMER_RECG_DIRECT_INVAL	0x0004	/* buffer alias invalidation */
564 /*
565  * hammer_create_at_cursor() and hammer_delete_at_cursor() flags.
566  */
567 #define HAMMER_CREATE_MODE_UMIRROR	0x0001
568 #define HAMMER_CREATE_MODE_SYS		0x0002
569 
570 #define HAMMER_DELETE_ADJUST		0x0001
571 #define HAMMER_DELETE_DESTROY		0x0002
572 
573 /*
574  * In-memory structures representing on-disk structures.
575  */
576 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
577 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
578 RB_HEAD(hammer_nod_rb_tree, hammer_node);
579 RB_HEAD(hammer_und_rb_tree, hammer_undo);
580 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
581 RB_HEAD(hammer_mod_rb_tree, hammer_io);
582 
583 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
584 	      hammer_vol_rb_compare, int32_t);
585 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
586 	      hammer_buf_rb_compare, hammer_off_t);
587 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
588 	      hammer_nod_rb_compare, hammer_off_t);
589 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
590 	      hammer_und_rb_compare, hammer_off_t);
591 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
592 	      hammer_res_rb_compare, hammer_off_t);
593 RB_PROTOTYPE2(hammer_mod_rb_tree, hammer_io, rb_node,
594 	      hammer_mod_rb_compare, hammer_off_t);
595 
596 /*
597  * IO management - embedded at the head of various in-memory structures
598  *
599  * VOLUME	- hammer_volume containing meta-data
600  * META_BUFFER	- hammer_buffer containing meta-data
601  * UNDO_BUFFER	- hammer_buffer containing undo-data
602  * DATA_BUFFER	- hammer_buffer containing pure-data
603  * DUMMY	- hammer_buffer not containing valid data
604  *
605  * Dirty volume headers and dirty meta-data buffers are locked until the
606  * flusher can sequence them out.  Dirty pure-data buffers can be written.
607  * Clean buffers can be passively released.
608  */
609 typedef enum hammer_io_type {
610 	HAMMER_STRUCTURE_VOLUME,
611 	HAMMER_STRUCTURE_META_BUFFER,
612 	HAMMER_STRUCTURE_UNDO_BUFFER,
613 	HAMMER_STRUCTURE_DATA_BUFFER,
614 	HAMMER_STRUCTURE_DUMMY
615 } hammer_io_type_t;
616 
617 typedef struct hammer_io {
618 	struct hammer_lock	lock;
619 	hammer_io_type_t	type;
620 	struct hammer_mount	*hmp;
621 	struct hammer_volume	*volume;
622 	RB_ENTRY(hammer_io)	rb_node;     /* if modified */
623 	TAILQ_ENTRY(hammer_io)	iorun_entry; /* iorun_list */
624 	struct hammer_mod_rb_tree *mod_root;
625 	struct buf		*bp;
626 	int64_t			offset;	   /* volume offset */
627 	int			bytes;	   /* buffer cache buffer size */
628 	int			modify_refs;
629 
630 	/*
631 	 * These can be modified at any time by the backend while holding
632 	 * io_token, due to bio_done and hammer_io_complete() callbacks.
633 	 */
634 	u_int		running : 1;	/* bp write IO in progress */
635 	u_int		waiting : 1;	/* someone is waiting on us */
636 	u_int		ioerror : 1;	/* abort on io-error */
637 	u_int		unusedA : 29;
638 
639 	/*
640 	 * These can only be modified by the frontend while holding
641 	 * fs_token, or by the backend while holding the io interlocked
642 	 * with no references (which will block the frontend when it
643 	 * tries to reference it).
644 	 *
645 	 * WARNING! SMP RACES will create havoc if the callbacks ever tried
646 	 *	    to modify any of these outside the above restrictions.
647 	 */
648 	u_int		modified : 1;	/* bp's data was modified */
649 	u_int		released : 1;	/* bp released (w/ B_LOCKED set) */
650 	u_int		waitdep : 1;	/* flush waits for dependancies */
651 	u_int		recovered : 1;	/* has recovery ref */
652 	u_int		waitmod : 1;	/* waiting for modify_refs */
653 	u_int		reclaim : 1;	/* reclaim requested */
654 	u_int		gencrc : 1;	/* crc needs to be generated */
655 	u_int		unusedB : 25;
656 } *hammer_io_t;
657 
658 #define HAMMER_CLUSTER_SIZE	(64 * 1024)
659 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
660 #undef  HAMMER_CLUSTER_SIZE
661 #define HAMMER_CLUSTER_SIZE	MAXBSIZE
662 #endif
663 
664 /*
665  * In-memory volume representing on-disk buffer
666  */
667 typedef struct hammer_volume {
668 	struct hammer_io io; /* must be at offset 0 */
669 	RB_ENTRY(hammer_volume) rb_node;
670 	hammer_volume_ondisk_t ondisk;
671 	int32_t	vol_no;
672 	hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
673 	char	*vol_name;
674 	struct vnode *devvp;
675 	int	vol_flags;
676 } *hammer_volume_t;
677 
678 #define HAMMER_ITOV(iop) ((hammer_volume_t)(iop))
679 
680 /*
681  * In-memory buffer representing an on-disk buffer.
682  */
683 typedef struct hammer_buffer {
684 	struct hammer_io io; /* must be at offset 0 */
685 	RB_ENTRY(hammer_buffer) rb_node;
686 	void *ondisk;
687 	hammer_off_t zoneX_offset;
688 	hammer_off_t zone2_offset;
689 	struct hammer_reserve *resv;
690 	struct hammer_node_list node_list;
691 } *hammer_buffer_t;
692 
693 #define HAMMER_ITOB(iop) ((hammer_buffer_t)(iop))
694 
695 /*
696  * In-memory B-Tree node, representing an on-disk B-Tree node.
697  *
698  * This is a hang-on structure which is backed by a hammer_buffer,
699  * and used for fine-grained locking of B-Tree nodes in order to
700  * properly control lock ordering.
701  */
702 typedef struct hammer_node {
703 	struct hammer_lock	lock;		/* node-by-node lock */
704 	TAILQ_ENTRY(hammer_node) entry;		/* per-buffer linkage */
705 	RB_ENTRY(hammer_node)	rb_node;	/* per-mount linkage */
706 	hammer_off_t		node_offset;	/* full offset spec */
707 	struct hammer_mount	*hmp;
708 	hammer_buffer_t		buffer;		/* backing buffer */
709 	hammer_node_ondisk_t	ondisk;		/* ptr to on-disk structure */
710 	TAILQ_HEAD(, hammer_cursor) cursor_list;  /* deadlock recovery */
711 	struct hammer_node_cache_list cache_list; /* passive caches */
712 	int			flags;
713 #if 0
714 	int			cursor_exclreq_count;
715 #endif
716 } *hammer_node_t;
717 
718 #define HAMMER_NODE_DELETED	0x0001
719 #define HAMMER_NODE_FLUSH	0x0002
720 #define HAMMER_NODE_CRCGOOD	0x0004
721 #define HAMMER_NODE_NEEDSCRC	0x0008
722 #define HAMMER_NODE_NEEDSMIRROR	0x0010
723 #define HAMMER_NODE_CRCBAD	0x0020
724 #define HAMMER_NODE_NONLINEAR	0x0040		/* linear heuristic */
725 
726 #define HAMMER_NODE_CRCANY	(HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD)
727 
728 /*
729  * List of locked nodes.  This structure is used to lock potentially large
730  * numbers of nodes as an aid for complex B-Tree operations.
731  */
732 struct hammer_node_lock;
733 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock);
734 
735 typedef struct hammer_node_lock {
736 	TAILQ_ENTRY(hammer_node_lock) entry;
737 	struct hammer_node_lock_list  list;
738 	struct hammer_node_lock	      *parent;
739 	hammer_node_t	node;
740 	hammer_node_ondisk_t copy;	/* copy of on-disk data */
741 	int		index;		/* index of this node in parent */
742 	int		count;		/* count children */
743 	int		flags;
744 } *hammer_node_lock_t;
745 
746 #define HAMMER_NODE_LOCK_UPDATED	0x0001
747 #define HAMMER_NODE_LOCK_LCACHE		0x0002
748 
749 /*
750  * The reserve structure prevents the blockmap from allocating
751  * out of a reserved big-block.  Such reservations are used by
752  * the direct-write mechanism.
753  *
754  * The structure is also used to hold off on reallocations of
755  * big-blocks from the freemap until flush dependancies have
756  * been dealt with.
757  */
758 typedef struct hammer_reserve {
759 	RB_ENTRY(hammer_reserve) rb_node;
760 	TAILQ_ENTRY(hammer_reserve) delay_entry;
761 	int		flg_no;
762 	int		flags;
763 	int		refs;
764 	int		zone;
765 	int		append_off;
766 	int32_t		bytes_free;
767 	hammer_off_t	zone_offset;
768 } *hammer_reserve_t;
769 
770 #define HAMMER_RESF_ONDELAY	0x0001
771 #define HAMMER_RESF_LAYER2FREE	0x0002
772 
773 #include "hammer_cursor.h"
774 
775 /*
776  * The undo structure tracks recent undos to avoid laying down duplicate
777  * undos within a flush group, saving us a significant amount of overhead.
778  *
779  * This is strictly a heuristic.
780  */
781 #define HAMMER_MAX_UNDOS		1024
782 #define HAMMER_MAX_FLUSHERS		4
783 
784 typedef struct hammer_undo {
785 	RB_ENTRY(hammer_undo)	rb_node;
786 	TAILQ_ENTRY(hammer_undo) lru_entry;
787 	hammer_off_t		offset;
788 	int			bytes;
789 } *hammer_undo_t;
790 
791 struct hammer_flusher_info;
792 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
793 
794 struct hammer_flusher {
795 	int		signal;		/* flusher thread sequencer */
796 	int		done;		/* last completed flush group */
797 	int		next;		/* next unallocated flg seqno */
798 	int		group_lock;	/* lock sequencing of the next flush */
799 	int		exiting;	/* request master exit */
800 	thread_t	td;		/* master flusher thread */
801 	hammer_tid_t	tid;		/* last flushed transaction id */
802 	int		finalize_want;		/* serialize finalization */
803 	struct hammer_lock finalize_lock;	/* serialize finalization */
804 	struct hammer_transaction trans;	/* shared transaction */
805 	struct hammer_flusher_info_list run_list;
806 	struct hammer_flusher_info_list ready_list;
807 };
808 
809 #define HAMMER_FLUSH_UNDOS_RELAXED	0
810 #define HAMMER_FLUSH_UNDOS_FORCED	1
811 #define HAMMER_FLUSH_UNDOS_AUTO		2
812 /*
813  * Internal hammer mount data structure
814  */
815 typedef struct hammer_mount {
816 	struct mount *mp;
817 	struct hammer_ino_rb_tree rb_inos_root;
818 	struct hammer_redo_rb_tree rb_redo_root;
819 	struct hammer_vol_rb_tree rb_vols_root;
820 	struct hammer_nod_rb_tree rb_nods_root;
821 	struct hammer_und_rb_tree rb_undo_root;
822 	struct hammer_res_rb_tree rb_resv_root;
823 	struct hammer_buf_rb_tree rb_bufs_root;
824 	struct hammer_pfs_rb_tree rb_pfsm_root;
825 
826 	struct hammer_dedup_crc_rb_tree rb_dedup_crc_root;
827 	struct hammer_dedup_off_rb_tree rb_dedup_off_root;
828 
829 	hammer_volume_t rootvol;
830 	struct hammer_base_elm root_btree_beg;
831 	struct hammer_base_elm root_btree_end;
832 
833 	struct malloc_type	*m_misc;
834 	struct malloc_type	*m_inodes;
835 
836 	int	flags;		/* HAMMER_MOUNT_xxx flags */
837 	int	hflags;
838 	int	ronly;
839 	int	nvolumes;
840 	int	master_id;	/* default 0, no-mirror -1, otherwise 0-15 */
841 	int	version;	/* hammer filesystem version to use */
842 	int	rsv_inodes;	/* reserved space due to dirty inodes */
843 	int64_t	rsv_databytes;	/* reserved space due to record data */
844 	int	rsv_recs;	/* reserved space due to dirty records */
845 	int	rsv_fromdelay;	/* big-blocks reserved due to flush delay */
846 	int	undo_rec_limit;	/* based on size of undo area */
847 
848 	int	volume_to_remove; /* volume that is currently being removed */
849 
850 	int	count_inodes;	/* total number of inodes */
851 	int	count_iqueued;	/* inodes queued to flusher */
852 	int	count_reclaims; /* inodes pending reclaim by flusher */
853 
854 	struct hammer_flusher flusher;
855 
856 	u_int	check_interrupt;
857 	u_int	check_yield;
858 	uuid_t	fsid;
859 	struct hammer_mod_rb_tree volu_root;	/* dirty undo buffers */
860 	struct hammer_mod_rb_tree undo_root;	/* dirty undo buffers */
861 	struct hammer_mod_rb_tree data_root;	/* dirty data buffers */
862 	struct hammer_mod_rb_tree meta_root;	/* dirty meta bufs    */
863 	struct hammer_mod_rb_tree lose_root;	/* loose buffers      */
864 	long	locked_dirty_space;		/* meta/volu count    */
865 	long	io_running_space;		/* io_token */
866 	int	objid_cache_count;
867 	int	dedup_cache_count;
868 	int	error;				/* critical I/O error */
869 	struct krate	krate;			/* rate limited kprintf */
870 	struct krate	kdiag;			/* rate limited kprintf */
871 	hammer_tid_t	asof;			/* snapshot mount */
872 	hammer_tid_t	next_tid;
873 	hammer_tid_t	flush_tid1;		/* flusher tid sequencing */
874 	hammer_tid_t	flush_tid2;		/* flusher tid sequencing */
875 	int64_t copy_stat_freebigblocks;	/* number of free big-blocks */
876 	uint32_t	undo_seqno;		/* UNDO/REDO FIFO seqno */
877 	uint32_t	recover_stage2_seqno;	/* REDO recovery seqno */
878 	hammer_off_t	recover_stage2_offset;	/* REDO recovery offset */
879 
880 	struct netexport export;
881 	struct hammer_lock sync_lock;
882 	struct hammer_lock undo_lock;
883 	struct hammer_lock blkmap_lock;
884 	struct hammer_lock snapshot_lock;
885 	struct hammer_lock volume_lock;
886 	struct hammer_blockmap  blockmap[HAMMER_MAX_ZONES];
887 	struct hammer_undo	undos[HAMMER_MAX_UNDOS];
888 	int			undo_alloc;
889 	TAILQ_HEAD(, hammer_undo)  undo_lru_list;
890 	TAILQ_HEAD(, hammer_reserve) delay_list;
891 	struct hammer_flush_group_list	flush_group_list;
892 	hammer_flush_group_t	fill_flush_group;
893 	hammer_flush_group_t	next_flush_group;
894 	TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
895 	TAILQ_HEAD(, hammer_dedup_cache) dedup_lru_list;
896 	hammer_dedup_cache_t	dedup_free_cache;
897 	TAILQ_HEAD(, hammer_reclaim) reclaim_list;
898 	TAILQ_HEAD(, hammer_io) iorun_list;
899 
900 	struct lwkt_token	fs_token;	/* high level */
901 	struct lwkt_token	io_token;	/* low level (IO callback) */
902 
903 	struct hammer_inostats	inostats[HAMMER_INOSTATS_HSIZE];
904 	uint64_t volume_map[4];  /* 256 bits bitfield */
905 } *hammer_mount_t;
906 
907 #define HAMMER_MOUNT_CRITICAL_ERROR	0x0001
908 #define HAMMER_MOUNT_FLUSH_RECOVERY	0x0002
909 #define HAMMER_MOUNT_REDO_SYNC		0x0004
910 #define HAMMER_MOUNT_REDO_RECOVERY_REQ	0x0008
911 #define HAMMER_MOUNT_REDO_RECOVERY_RUN	0x0010
912 
913 #define HAMMER_VOLUME_NUMBER_FOREACH(hmp, n)		\
914 	for (n = 0; n < HAMMER_MAX_VOLUMES; n++)	\
915 		if (hammer_volume_number_test(hmp, n))
916 
917 /*
918  * Minium buffer cache bufs required to rebalance the B-Tree.
919  * This is because we must hold the children and the children's children
920  * locked.  Even this might not be enough if things are horribly out
921  * of balance.
922  */
923 #define HAMMER_REBALANCE_MIN_BUFS	\
924 	(HAMMER_BTREE_LEAF_ELMS * HAMMER_BTREE_LEAF_ELMS)
925 
926 #endif  /* _KERNEL || _KERNEL_STRUCTURES */
927 
928 #if defined(_KERNEL)
929 /*
930  * checkspace slop (8MB chunks), higher numbers are more conservative.
931  */
932 #define HAMMER_CHKSPC_REBLOCK	25
933 #define HAMMER_CHKSPC_MIRROR	20
934 #define HAMMER_CHKSPC_WRITE	20
935 #define HAMMER_CHKSPC_CREATE	20
936 #define HAMMER_CHKSPC_REMOVE	10
937 #define HAMMER_CHKSPC_EMERGENCY	0
938 
939 extern struct vop_ops hammer_vnode_vops;
940 extern struct vop_ops hammer_spec_vops;
941 extern struct vop_ops hammer_fifo_vops;
942 
943 extern int hammer_debug_io;
944 extern int hammer_debug_general;
945 extern int hammer_debug_inode;
946 extern int hammer_debug_locks;
947 extern int hammer_debug_btree;
948 extern int hammer_debug_tid;
949 extern int hammer_debug_recover;
950 extern int hammer_debug_critical;
951 extern int hammer_cluster_enable;
952 extern int hammer_live_dedup;
953 extern int hammer_tdmux_ticks;
954 extern int hammer_count_fsyncs;
955 extern int hammer_count_inodes;
956 extern int hammer_count_iqueued;
957 extern int hammer_count_reclaims;
958 extern int hammer_count_records;
959 extern int hammer_count_record_datas;
960 extern int hammer_count_volumes;
961 extern int hammer_count_buffers;
962 extern int hammer_count_nodes;
963 extern int64_t hammer_stats_btree_lookups;
964 extern int64_t hammer_stats_btree_searches;
965 extern int64_t hammer_stats_btree_inserts;
966 extern int64_t hammer_stats_btree_deletes;
967 extern int64_t hammer_stats_btree_elements;
968 extern int64_t hammer_stats_btree_splits;
969 extern int64_t hammer_stats_btree_iterations;
970 extern int64_t hammer_stats_btree_root_iterations;
971 extern int64_t hammer_stats_record_iterations;
972 extern int64_t hammer_stats_file_read;
973 extern int64_t hammer_stats_file_write;
974 extern int64_t hammer_stats_file_iopsr;
975 extern int64_t hammer_stats_file_iopsw;
976 extern int64_t hammer_stats_disk_read;
977 extern int64_t hammer_stats_disk_write;
978 extern int64_t hammer_stats_inode_flushes;
979 extern int64_t hammer_stats_commits;
980 extern int64_t hammer_stats_undo;
981 extern int64_t hammer_stats_redo;
982 extern long hammer_count_dirtybufspace;
983 extern int hammer_count_refedbufs;
984 extern int hammer_count_reservations;
985 extern long hammer_count_io_running_read;
986 extern long hammer_count_io_running_write;
987 extern int hammer_count_io_locked;
988 extern long hammer_limit_dirtybufspace;
989 extern int hammer_limit_recs;
990 extern int hammer_limit_inode_recs;
991 extern int hammer_limit_reclaims;
992 extern int hammer_live_dedup_cache_size;
993 extern int hammer_limit_redo;
994 extern int hammer_verify_zone;
995 extern int hammer_verify_data;
996 extern int hammer_double_buffer;
997 extern int hammer_btree_full_undo;
998 extern int hammer_yield_check;
999 extern int hammer_fsync_mode;
1000 extern int hammer_autoflush;
1001 extern int64_t hammer_contention_count;
1002 
1003 extern int64_t hammer_live_dedup_vnode_bcmps;
1004 extern int64_t hammer_live_dedup_device_bcmps;
1005 extern int64_t hammer_live_dedup_findblk_failures;
1006 extern int64_t hammer_live_dedup_bmap_saves;
1007 
1008 void	hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
1009 			int error, const char *msg);
1010 int	hammer_vop_inactive(struct vop_inactive_args *);
1011 int	hammer_vop_reclaim(struct vop_reclaim_args *);
1012 int	hammer_get_vnode(hammer_inode_t ip, struct vnode **vpp);
1013 hammer_inode_t hammer_get_inode(hammer_transaction_t trans,
1014 			hammer_inode_t dip, int64_t obj_id,
1015 			hammer_tid_t asof, uint32_t localization,
1016 			int flags, int *errorp);
1017 hammer_inode_t hammer_get_dummy_inode(hammer_transaction_t trans,
1018 			hammer_inode_t dip, int64_t obj_id,
1019 			hammer_tid_t asof, uint32_t localization,
1020 			int flags, int *errorp);
1021 hammer_inode_t hammer_find_inode(hammer_transaction_t trans,
1022 			int64_t obj_id, hammer_tid_t asof,
1023 			uint32_t localization);
1024 void	hammer_scan_inode_snapshots(hammer_mount_t hmp,
1025 			hammer_inode_info_t iinfo,
1026 			int (*callback)(hammer_inode_t ip, void *data),
1027 			void *data);
1028 void	hammer_put_inode(hammer_inode_t ip);
1029 void	hammer_put_inode_ref(hammer_inode_t ip);
1030 void	hammer_inode_waitreclaims(hammer_transaction_t trans);
1031 void	hammer_inode_dirty(hammer_inode_t ip);
1032 
1033 int	hammer_unload_volume(hammer_volume_t volume, void *data);
1034 int	hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
1035 
1036 int	hammer_unload_buffer(hammer_buffer_t buffer, void *data);
1037 int	hammer_install_volume(hammer_mount_t hmp, const char *volname,
1038 			struct vnode *devvp, void *data);
1039 int	hammer_mountcheck_volumes(hammer_mount_t hmp);
1040 int	hammer_get_installed_volumes(hammer_mount_t hmp);
1041 
1042 int	hammer_mem_add(hammer_record_t record);
1043 int	hammer_ip_lookup(hammer_cursor_t cursor);
1044 int	hammer_ip_first(hammer_cursor_t cursor);
1045 int	hammer_ip_next(hammer_cursor_t cursor);
1046 int	hammer_ip_resolve_data(hammer_cursor_t cursor);
1047 int	hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
1048 			hammer_tid_t tid);
1049 int	hammer_create_at_cursor(hammer_cursor_t cursor,
1050 			hammer_btree_leaf_elm_t leaf, void *udata, int mode);
1051 int	hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
1052 			hammer_tid_t delete_tid, uint32_t delete_ts,
1053 			int track, int64_t *stat_bytes);
1054 int	hammer_ip_check_directory_empty(hammer_transaction_t trans,
1055 			hammer_inode_t ip);
1056 int	hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
1057 int	hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
1058 
1059 hammer_record_t
1060 	hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
1061 void	hammer_flush_record_done(hammer_record_t record, int error);
1062 void	hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
1063 void	hammer_rel_mem_record(hammer_record_t record);
1064 
1065 int	hammer_cursor_up(hammer_cursor_t cursor);
1066 int	hammer_cursor_up_locked(hammer_cursor_t cursor);
1067 int	hammer_cursor_down(hammer_cursor_t cursor);
1068 int	hammer_cursor_upgrade(hammer_cursor_t cursor);
1069 int	hammer_cursor_upgrade_node(hammer_cursor_t cursor);
1070 void	hammer_cursor_downgrade(hammer_cursor_t cursor);
1071 int	hammer_cursor_upgrade2(hammer_cursor_t c1, hammer_cursor_t c2);
1072 void	hammer_cursor_downgrade2(hammer_cursor_t c1, hammer_cursor_t c2);
1073 int	hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
1074 			int index);
1075 void	hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
1076 int	hammer_lock_ex_try(struct hammer_lock *lock);
1077 void	hammer_lock_sh(struct hammer_lock *lock);
1078 int	hammer_lock_sh_try(struct hammer_lock *lock);
1079 int	hammer_lock_upgrade(struct hammer_lock *lock, int shcount);
1080 void	hammer_lock_downgrade(struct hammer_lock *lock, int shcount);
1081 int	hammer_lock_status(struct hammer_lock *lock);
1082 void	hammer_unlock(struct hammer_lock *lock);
1083 void	hammer_ref(struct hammer_lock *lock);
1084 int	hammer_ref_interlock(struct hammer_lock *lock);
1085 int	hammer_ref_interlock_true(struct hammer_lock *lock);
1086 void	hammer_ref_interlock_done(struct hammer_lock *lock);
1087 void	hammer_rel(struct hammer_lock *lock);
1088 int	hammer_rel_interlock(struct hammer_lock *lock, int locked);
1089 void	hammer_rel_interlock_done(struct hammer_lock *lock, int orig_locked);
1090 int	hammer_get_interlock(struct hammer_lock *lock);
1091 int	hammer_try_interlock_norefs(struct hammer_lock *lock);
1092 void	hammer_put_interlock(struct hammer_lock *lock, int error);
1093 
1094 void	hammer_sync_lock_ex(hammer_transaction_t trans);
1095 void	hammer_sync_lock_sh(hammer_transaction_t trans);
1096 int	hammer_sync_lock_sh_try(hammer_transaction_t trans);
1097 void	hammer_sync_unlock(hammer_transaction_t trans);
1098 
1099 uint32_t hammer_to_unix_xid(uuid_t *uuid);
1100 void hammer_guid_to_uuid(uuid_t *uuid, uint32_t guid);
1101 void	hammer_time_to_timespec(uint64_t xtime, struct timespec *ts);
1102 uint64_t hammer_timespec_to_time(struct timespec *ts);
1103 int	hammer_str_to_tid(const char *str, int *ispfsp,
1104 			hammer_tid_t *tidp, uint32_t *localizationp);
1105 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip,
1106 			int64_t namekey);
1107 void hammer_clear_objid(hammer_inode_t dip);
1108 void hammer_destroy_objid_cache(hammer_mount_t hmp);
1109 
1110 int hammer_dedup_crc_rb_compare(hammer_dedup_cache_t dc1,
1111 			hammer_dedup_cache_t dc2);
1112 int hammer_dedup_off_rb_compare(hammer_dedup_cache_t dc1,
1113 			hammer_dedup_cache_t dc2);
1114 hammer_dedup_cache_t hammer_dedup_cache_add(hammer_inode_t ip,
1115 			hammer_btree_leaf_elm_t leaf);
1116 hammer_dedup_cache_t hammer_dedup_cache_lookup(hammer_mount_t hmp,
1117 			hammer_crc_t crc);
1118 void hammer_dedup_cache_inval(hammer_mount_t hmp, hammer_off_t base_offset);
1119 void hammer_destroy_dedup_cache(hammer_mount_t hmp);
1120 void hammer_dump_dedup_cache(hammer_mount_t hmp);
1121 int hammer_dedup_validate(hammer_dedup_cache_t dcp, int zone, int bytes,
1122 			void *data);
1123 
1124 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
1125 			int bytes);
1126 void hammer_clear_undo_history(hammer_mount_t hmp);
1127 enum vtype hammer_get_vnode_type(uint8_t obj_type);
1128 int hammer_get_dtype(uint8_t obj_type);
1129 uint8_t hammer_get_obj_type(enum vtype vtype);
1130 int64_t hammer_direntry_namekey(hammer_inode_t dip, const void *name, int len,
1131 			uint32_t *max_iterationsp);
1132 int	hammer_nohistory(hammer_inode_t ip);
1133 
1134 int	hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
1135 			hammer_node_cache_t cache, hammer_inode_t ip);
1136 void	hammer_normalize_cursor(hammer_cursor_t cursor);
1137 void	hammer_done_cursor(hammer_cursor_t cursor);
1138 int	hammer_recover_cursor(hammer_cursor_t cursor);
1139 void	hammer_unlock_cursor(hammer_cursor_t cursor);
1140 int	hammer_lock_cursor(hammer_cursor_t cursor);
1141 hammer_cursor_t	hammer_push_cursor(hammer_cursor_t ocursor);
1142 void	hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
1143 
1144 void	hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
1145 void	hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
1146 			int index);
1147 void	hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
1148 			int index);
1149 void	hammer_cursor_moved_element(hammer_node_t oparent, int pindex,
1150 			hammer_node_t onode, int oindex,
1151 			hammer_node_t nnode, int nindex);
1152 void	hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent,
1153 			hammer_node_t nparent, int nindex);
1154 void	hammer_cursor_inserted_element(hammer_node_t node, int index);
1155 void	hammer_cursor_deleted_element(hammer_node_t node, int index);
1156 void	hammer_cursor_invalidate_cache(hammer_cursor_t cursor);
1157 
1158 int	hammer_btree_lookup(hammer_cursor_t cursor);
1159 int	hammer_btree_first(hammer_cursor_t cursor);
1160 int	hammer_btree_last(hammer_cursor_t cursor);
1161 int	hammer_btree_extract(hammer_cursor_t cursor, int flags);
1162 int	hammer_btree_iterate(hammer_cursor_t cursor);
1163 int	hammer_btree_iterate_reverse(hammer_cursor_t cursor);
1164 int	hammer_btree_insert(hammer_cursor_t cursor,
1165 			    hammer_btree_leaf_elm_t elm, int *doprop);
1166 int	hammer_btree_delete(hammer_cursor_t cursor, int *ndelete);
1167 void	hammer_btree_do_propagation(hammer_cursor_t cursor,
1168 			    hammer_btree_leaf_elm_t leaf);
1169 int	hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
1170 int	hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
1171 int	hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
1172 int	hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
1173 
1174 int	btree_set_parent_of_child(hammer_transaction_t trans,
1175                         hammer_node_t node,
1176                         hammer_btree_elm_t elm);
1177 void	hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node);
1178 void	hammer_btree_lcache_init(hammer_mount_t hmp, hammer_node_lock_t lcache,
1179 			int depth);
1180 void	hammer_btree_lcache_free(hammer_mount_t hmp, hammer_node_lock_t lcache);
1181 int	hammer_btree_lock_children(hammer_cursor_t cursor, int depth,
1182 			hammer_node_lock_t parent,
1183 			hammer_node_lock_t lcache);
1184 void	hammer_btree_lock_copy(hammer_cursor_t cursor,
1185 			hammer_node_lock_t parent);
1186 int	hammer_btree_sync_copy(hammer_cursor_t cursor,
1187 			hammer_node_lock_t parent);
1188 void	hammer_btree_unlock_children(hammer_mount_t hmp,
1189 			hammer_node_lock_t parent,
1190 			hammer_node_lock_t lcache);
1191 int	hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
1192 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans,
1193 			hammer_node_t node, int *parent_indexp,
1194 			int *errorp, int try_exclusive);
1195 
1196 void	hammer_print_btree_node(hammer_node_ondisk_t ondisk);
1197 void	hammer_print_btree_elm(hammer_btree_elm_t elm);
1198 
1199 void	*hammer_bread(hammer_mount_t hmp, hammer_off_t off,
1200 			int *errorp, hammer_buffer_t *bufferp);
1201 void	*hammer_bnew(hammer_mount_t hmp, hammer_off_t off,
1202 			int *errorp, hammer_buffer_t *bufferp);
1203 void	*hammer_bread_ext(hammer_mount_t hmp, hammer_off_t off, int bytes,
1204 			int *errorp, hammer_buffer_t *bufferp);
1205 void	*hammer_bnew_ext(hammer_mount_t hmp, hammer_off_t off, int bytes,
1206 			int *errorp, hammer_buffer_t *bufferp);
1207 
1208 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
1209 
1210 hammer_volume_t	hammer_get_volume(hammer_mount_t hmp,
1211 			int32_t vol_no, int *errorp);
1212 hammer_buffer_t	hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
1213 			int bytes, int isnew, int *errorp);
1214 void		hammer_sync_buffers(hammer_mount_t hmp,
1215 			hammer_off_t base_offset, int bytes);
1216 int		hammer_del_buffers(hammer_mount_t hmp,
1217 			hammer_off_t base_offset,
1218 			hammer_off_t zone2_offset, int bytes,
1219 			int report_conflicts);
1220 
1221 int		hammer_ref_volume(hammer_volume_t volume);
1222 int		hammer_ref_buffer(hammer_buffer_t buffer);
1223 void		hammer_flush_buffer_nodes(hammer_buffer_t buffer);
1224 
1225 void		hammer_rel_volume(hammer_volume_t volume, int locked);
1226 void		hammer_rel_buffer(hammer_buffer_t buffer, int locked);
1227 
1228 int		hammer_vfs_export(struct mount *mp, int op,
1229 			const struct export_args *export);
1230 hammer_node_t	hammer_get_node(hammer_transaction_t trans,
1231 			hammer_off_t node_offset, int isnew, int *errorp);
1232 void		hammer_ref_node(hammer_node_t node);
1233 hammer_node_t	hammer_ref_node_safe(hammer_transaction_t trans,
1234 			hammer_node_cache_t cache, int *errorp);
1235 void		hammer_rel_node(hammer_node_t node);
1236 void		hammer_delete_node(hammer_transaction_t trans,
1237 			hammer_node_t node);
1238 void		hammer_cache_node(hammer_node_cache_t cache,
1239 			hammer_node_t node);
1240 void		hammer_uncache_node(hammer_node_cache_t cache);
1241 void		hammer_flush_node(hammer_node_t node, int locked);
1242 
1243 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans,
1244 			hammer_off_t hint, int *errorp);
1245 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1246 			uint16_t rec_type, hammer_off_t *data_offsetp,
1247 			hammer_buffer_t *data_bufferp,
1248 			hammer_off_t hint, int *errorp);
1249 
1250 int hammer_generate_undo(hammer_transaction_t trans,
1251 			hammer_off_t zone_offset, void *base, int len);
1252 int hammer_generate_redo(hammer_transaction_t trans, hammer_inode_t ip,
1253 			hammer_off_t file_offset, uint32_t flags,
1254 			void *base, int len);
1255 void hammer_generate_redo_sync(hammer_transaction_t trans);
1256 void hammer_redo_fifo_start_flush(hammer_inode_t ip);
1257 void hammer_redo_fifo_end_flush(hammer_inode_t ip);
1258 
1259 void hammer_format_undo(void *base, uint32_t seqno);
1260 int hammer_upgrade_undo_4(hammer_transaction_t trans);
1261 
1262 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1263 			hammer_off_t owner, int *errorp);
1264 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1265 			hammer_off_t owner, int *errorp);
1266 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp);
1267 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1268 			int bytes, hammer_off_t hint, int *errorp);
1269 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1270 			int bytes, hammer_off_t *zone_offp, int *errorp);
1271 hammer_reserve_t hammer_blockmap_reserve_dedup(hammer_mount_t hmp, int zone,
1272 			int bytes, hammer_off_t zone_offset, int *errorp);
1273 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1274 			hammer_reserve_t resv);
1275 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1276 void hammer_blockmap_free(hammer_transaction_t trans,
1277 			hammer_off_t zone_offset, int bytes);
1278 int hammer_blockmap_dedup(hammer_transaction_t trans,
1279 			hammer_off_t zone_offset, int bytes);
1280 int hammer_blockmap_finalize(hammer_transaction_t trans,
1281 			hammer_reserve_t resv,
1282 			hammer_off_t zone_offset, int bytes);
1283 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t zone_offset,
1284 			int *curp, int *errorp);
1285 hammer_off_t hammer_blockmap_lookup_verify(hammer_mount_t hmp,
1286 			hammer_off_t zone_offset, int *errorp);
1287 
1288 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t zone_offset,
1289 			int *errorp);
1290 int64_t hammer_undo_used(hammer_transaction_t trans);
1291 int64_t hammer_undo_space(hammer_transaction_t trans);
1292 int64_t hammer_undo_max(hammer_mount_t hmp);
1293 int hammer_undo_reclaim(hammer_io_t io);
1294 
1295 void hammer_start_transaction(hammer_transaction_t trans,
1296 			      hammer_mount_t hmp);
1297 void hammer_simple_transaction(hammer_transaction_t trans,
1298 			      hammer_mount_t hmp);
1299 void hammer_start_transaction_fls(hammer_transaction_t trans,
1300 			          hammer_mount_t hmp);
1301 void hammer_done_transaction(hammer_transaction_t trans);
1302 hammer_tid_t hammer_alloc_tid(hammer_mount_t hmp, int count);
1303 
1304 void hammer_modify_inode(hammer_transaction_t trans, hammer_inode_t ip, int flags);
1305 void hammer_flush_inode(hammer_inode_t ip, int flags);
1306 void hammer_flush_inode_done(hammer_inode_t ip, int error);
1307 void hammer_wait_inode(hammer_inode_t ip);
1308 
1309 int  hammer_create_inode(hammer_transaction_t trans, struct vattr *vap,
1310 			struct ucred *cred, hammer_inode_t dip,
1311 			const char *name, int namelen,
1312 			hammer_pseudofs_inmem_t pfsm,
1313 			hammer_inode_t *ipp);
1314 void hammer_rel_inode(hammer_inode_t ip, int flush);
1315 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1316 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1317 int hammer_redo_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1318 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1319 
1320 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1321 void hammer_test_inode(hammer_inode_t dip);
1322 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1323 int hammer_update_atime_quick(hammer_inode_t ip);
1324 
1325 int  hammer_ip_add_direntry(hammer_transaction_t trans,
1326 			hammer_inode_t dip, const char *name, int bytes,
1327 			hammer_inode_t nip);
1328 int  hammer_ip_del_direntry(hammer_transaction_t trans,
1329 			hammer_cursor_t cursor, hammer_inode_t dip,
1330 			hammer_inode_t ip);
1331 void hammer_ip_replace_bulk(hammer_mount_t hmp, hammer_record_t record);
1332 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1333 			void *data, int bytes, int *errorp);
1334 int  hammer_ip_frontend_trunc(hammer_inode_t ip, off_t file_size);
1335 int  hammer_ip_add_record(hammer_transaction_t trans,
1336 			hammer_record_t record);
1337 int  hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1338 			int64_t ran_beg, int64_t ran_end, int truncating);
1339 int  hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1340 			int *countp);
1341 int  hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1342 			int64_t offset, void *data, int bytes);
1343 int  hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1344 hammer_pseudofs_inmem_t  hammer_load_pseudofs(hammer_transaction_t trans,
1345 			uint32_t localization, int *errorp);
1346 int  hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1347 			hammer_pseudofs_inmem_t pfsm, hammer_inode_t dip);
1348 int  hammer_save_pseudofs(hammer_transaction_t trans,
1349 			hammer_pseudofs_inmem_t pfsm);
1350 int  hammer_unload_pseudofs(hammer_transaction_t trans, uint32_t localization);
1351 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1352 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1353 			struct ucred *cred);
1354 
1355 void hammer_io_init(hammer_io_t io, hammer_volume_t volume,
1356 			hammer_io_type_t type);
1357 int hammer_io_read(struct vnode *devvp, hammer_io_t io, int limit);
1358 void hammer_io_advance(hammer_io_t io);
1359 int hammer_io_new(struct vnode *devvp, hammer_io_t io);
1360 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1361 struct buf *hammer_io_release(hammer_io_t io, int flush);
1362 void hammer_io_flush(hammer_io_t io, int reclaim);
1363 void hammer_io_wait(hammer_io_t io);
1364 void hammer_io_waitdep(hammer_io_t io);
1365 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident, int doflush);
1366 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1367 			hammer_btree_leaf_elm_t leaf);
1368 int hammer_io_indirect_read(hammer_mount_t hmp, struct bio *bio,
1369 			hammer_btree_leaf_elm_t leaf);
1370 int hammer_io_direct_write(hammer_mount_t hmp, struct bio *bio,
1371 			hammer_record_t record);
1372 void hammer_io_direct_wait(hammer_record_t record);
1373 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1374 void hammer_io_write_interlock(hammer_io_t io);
1375 void hammer_io_done_interlock(hammer_io_t io);
1376 void hammer_io_clear_modify(hammer_io_t io, int inval);
1377 void hammer_io_clear_modlist(hammer_io_t io);
1378 void hammer_io_flush_sync(hammer_mount_t hmp);
1379 void hammer_io_clear_error(hammer_io_t io);
1380 void hammer_io_clear_error_noassert(hammer_io_t io);
1381 void hammer_io_notmeta(hammer_buffer_t buffer);
1382 void hammer_io_limit_backlog(hammer_mount_t hmp);
1383 
1384 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1385 			void *base, int len);
1386 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1387 			void *base, int len);
1388 void hammer_modify_volume_done(hammer_volume_t volume);
1389 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1390 
1391 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1392 			struct hammer_ioc_reblock *reblock);
1393 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip,
1394 			struct hammer_ioc_rebalance *rebal);
1395 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1396 			struct hammer_ioc_prune *prune);
1397 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1398 			struct hammer_ioc_mirror_rw *mirror);
1399 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1400 			struct hammer_ioc_mirror_rw *mirror);
1401 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1402 			struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1403 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1404                         struct hammer_ioc_pseudofs_rw *pfs);
1405 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1406                         struct hammer_ioc_pseudofs_rw *pfs);
1407 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1408                         struct hammer_ioc_pseudofs_rw *pfs);
1409 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1410                         struct hammer_ioc_pseudofs_rw *pfs);
1411 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1412                         struct hammer_ioc_pseudofs_rw *pfs);
1413 int hammer_ioc_scan_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1414                         struct hammer_ioc_pseudofs_rw *pfs);
1415 int hammer_ioc_volume_add(hammer_transaction_t trans, hammer_inode_t ip,
1416                         struct hammer_ioc_volume *ioc);
1417 int hammer_ioc_volume_del(hammer_transaction_t trans, hammer_inode_t ip,
1418                         struct hammer_ioc_volume *ioc);
1419 int hammer_ioc_volume_list(hammer_transaction_t trans, hammer_inode_t ip,
1420 			struct hammer_ioc_volume_list *ioc);
1421 int hammer_ioc_dedup(hammer_transaction_t trans, hammer_inode_t ip,
1422 			struct hammer_ioc_dedup *dedup);
1423 
1424 int hammer_signal_check(hammer_mount_t hmp);
1425 
1426 void hammer_flusher_create(hammer_mount_t hmp);
1427 void hammer_flusher_destroy(hammer_mount_t hmp);
1428 void hammer_flusher_sync(hammer_mount_t hmp);
1429 int  hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1430 int  hammer_flusher_async_one(hammer_mount_t hmp);
1431 int hammer_flusher_running(hammer_mount_t hmp);
1432 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1433 void hammer_flusher_wait_next(hammer_mount_t hmp);
1434 int  hammer_flusher_meta_limit(hammer_mount_t hmp);
1435 int  hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1436 int  hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1437 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1438 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1439 int  hammer_flusher_haswork(hammer_mount_t hmp);
1440 int  hammer_flush_dirty(hammer_mount_t hmp, int max_count);
1441 void hammer_flusher_flush_undos(hammer_mount_t hmp, int already_flushed);
1442 
1443 int hammer_recover_stage1(hammer_mount_t hmp, hammer_volume_t rootvol);
1444 int hammer_recover_stage2(hammer_mount_t hmp, hammer_volume_t rootvol);
1445 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1446 			hammer_volume_t root_volume, int final);
1447 
1448 udev_t hammer_fsid_to_udev(uuid_t *uuid);
1449 
1450 
1451 int hammer_blocksize(int64_t file_offset);
1452 int hammer_blockoff(int64_t file_offset);
1453 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1454 
1455 /*
1456  * Shortcut for _hammer_checkspace(), used all over the code.
1457  */
1458 static __inline int
1459 hammer_checkspace(hammer_mount_t hmp, int slop)
1460 {
1461 	return(_hammer_checkspace(hmp, slop, NULL));
1462 }
1463 
1464 static __inline void
1465 hammer_wait_mem_record(hammer_record_t record)
1466 {
1467 	hammer_wait_mem_record_ident(record, "hmmwai");
1468 }
1469 
1470 static __inline void
1471 hammer_lock_ex(struct hammer_lock *lock)
1472 {
1473 	hammer_lock_ex_ident(lock, "hmrlck");
1474 }
1475 
1476 static __inline void
1477 hammer_modify_volume_noundo(hammer_transaction_t trans, hammer_volume_t volume)
1478 {
1479 	hammer_modify_volume(trans, volume, NULL, 0);
1480 }
1481 
1482 static __inline void
1483 hammer_modify_buffer_noundo(hammer_transaction_t trans, hammer_buffer_t buffer)
1484 {
1485 	hammer_modify_buffer(trans, buffer, NULL, 0);
1486 }
1487 
1488 /*
1489  * Indicate that a B-Tree node is being modified.
1490  */
1491 static __inline void
1492 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1493 {
1494 	KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1495 	hammer_modify_buffer(trans, node->buffer, NULL, 0);
1496 }
1497 
1498 static __inline void
1499 hammer_modify_node_all(hammer_transaction_t trans, hammer_node_t node)
1500 {
1501 	KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1502 	hammer_modify_buffer(trans, node->buffer,
1503 			     node->ondisk, sizeof(*node->ondisk));
1504 }
1505 
1506 static __inline void
1507 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1508 		   void *base, int len)
1509 {
1510 	hammer_crc_t *crcptr;
1511 
1512 	KKASSERT((char *)base >= (char *)node->ondisk &&
1513 		 (char *)base + len <=
1514 		    (char *)node->ondisk + sizeof(*node->ondisk));
1515 	KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1516 
1517 	if (hammer_btree_full_undo) {
1518 		hammer_modify_node_all(trans, node);
1519 	} else {
1520 		hammer_modify_buffer(trans, node->buffer, base, len);
1521 		crcptr = &node->ondisk->crc;
1522 		hammer_modify_buffer(trans, node->buffer,
1523 				     crcptr, sizeof(hammer_crc_t));
1524 		--node->buffer->io.modify_refs;	/* only want one ref */
1525 	}
1526 }
1527 
1528 /*
1529  * Indicate that the specified modifications have been completed.
1530  *
1531  * Do not try to generate the crc here, it's very expensive to do and a
1532  * sequence of insertions or deletions can result in many calls to this
1533  * function on the same node.
1534  */
1535 static __inline void
1536 hammer_modify_node_done(hammer_node_t node)
1537 {
1538 	node->flags |= HAMMER_NODE_CRCGOOD;
1539 	if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1540 		node->flags |= HAMMER_NODE_NEEDSCRC;
1541 		node->buffer->io.gencrc = 1;
1542 		hammer_ref_node(node);
1543 	}
1544 	hammer_modify_buffer_done(node->buffer);
1545 }
1546 
1547 static __inline int
1548 hammer_btree_extract_leaf(hammer_cursor_t cursor)
1549 {
1550 	return(hammer_btree_extract(cursor, 0));
1551 }
1552 
1553 static __inline int
1554 hammer_btree_extract_data(hammer_cursor_t cursor)
1555 {
1556 	return(hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA));
1557 }
1558 
1559 /*
1560  * Lookup a blockmap offset.
1561  */
1562 static __inline hammer_off_t
1563 hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t zone_offset,
1564 			int *errorp)
1565 {
1566 #if defined INVARIANTS
1567 	int zone = HAMMER_ZONE_DECODE(zone_offset);
1568 	KKASSERT(hammer_is_zone2_mapped_index(zone));
1569 #endif
1570 
1571 	/*
1572 	 * We can actually skip blockmap verify by default,
1573 	 * as normal blockmaps are now direct-mapped onto the freemap
1574 	 * and so represent zone-2 addresses.
1575 	 */
1576 	if (hammer_verify_zone == 0) {
1577 		*errorp = 0;
1578 		return hammer_xlate_to_zone2(zone_offset);
1579 	}
1580 
1581 	return hammer_blockmap_lookup_verify(hmp, zone_offset, errorp);
1582 }
1583 
1584 #define hammer_modify_volume_field(trans, vol, field)		\
1585 	hammer_modify_volume(trans, vol, &(vol)->ondisk->field,	\
1586 			     sizeof((vol)->ondisk->field))
1587 
1588 #define hammer_modify_node_field(trans, node, field)		\
1589 	hammer_modify_node(trans, node, &(node)->ondisk->field,	\
1590 			     sizeof((node)->ondisk->field))
1591 
1592 /*
1593  * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1594  * created directories for HAMMER version 2 or greater and causes
1595  * directory entries to be placed the inode localization zone in
1596  * the B-Tree instead of the misc zone.
1597  *
1598  * This greatly improves localization between directory entries and
1599  * inodes
1600  */
1601 static __inline uint32_t
1602 hammer_dir_localization(hammer_inode_t dip)
1603 {
1604 	return(HAMMER_DIR_INODE_LOCALIZATION(&dip->ino_data));
1605 }
1606 
1607 static __inline
1608 hammer_io_t
1609 hammer_buf_peek_io(struct buf *bp)
1610 {
1611 	return((hammer_io_t)bp->b_priv);
1612 }
1613 
1614 static __inline
1615 void
1616 hammer_buf_attach_io(struct buf *bp, hammer_io_t io)
1617 {
1618 	/* struct buf and struct hammer_io are 1:1 */
1619 	KKASSERT(hammer_buf_peek_io(bp) == NULL);
1620 	bp->b_priv = io;
1621 }
1622 
1623 static __inline int
1624 __hammer_vol_index(int vol_no)
1625 {
1626 	return(vol_no >> 6);
1627 }
1628 
1629 static __inline uint64_t
1630 __hammer_vol_low(int vol_no)
1631 {
1632 	return((uint64_t)1 << (vol_no & ((1 << 6) - 1)));
1633 }
1634 
1635 static __inline void
1636 hammer_volume_number_add(hammer_mount_t hmp, hammer_volume_t vol)
1637 {
1638 	int i = __hammer_vol_index(vol->vol_no);
1639 	hmp->volume_map[i] |= __hammer_vol_low(vol->vol_no);
1640 }
1641 
1642 static __inline void
1643 hammer_volume_number_del(hammer_mount_t hmp, hammer_volume_t vol)
1644 {
1645 	int i = __hammer_vol_index(vol->vol_no);
1646 	hmp->volume_map[i] &= ~__hammer_vol_low(vol->vol_no);
1647 }
1648 
1649 static __inline int
1650 hammer_volume_number_test(hammer_mount_t hmp, int n)
1651 {
1652 	int i = __hammer_vol_index(n);
1653 	return((hmp->volume_map[i] & __hammer_vol_low(n)) != 0);
1654 }
1655 
1656 #define hkprintf(format, args...)			\
1657 	kprintf("HAMMER: "format,## args)
1658 #define hvkprintf(vol, format, args...)			\
1659 	kprintf("HAMMER(%s) "format, vol->ondisk->vol_label,## args)
1660 #define hmkprintf(hmp, format, args...)			\
1661 	kprintf("HAMMER(%s) "format, hmp->mp->mnt_stat.f_mntfromname,## args)
1662 #define hdkprintf(format, args...)			\
1663 	kprintf("%s: "format, __func__,## args)
1664 
1665 #define hkrateprintf(rate , format, args...)		\
1666 	krateprintf(rate, "HAMMER: "format,## args)
1667 #define hvkrateprintf(rate, vol, format, args...)	\
1668 	krateprintf(rate, "HAMMER(%s) "format, vol->ondisk->vol_label,## args)
1669 #define hmkrateprintf(rate, hmp, format, args...)	\
1670 	krateprintf(rate, "HAMMER(%s) "format, hmp->mp->mnt_stat.f_mntfromname,## args)
1671 #define hdkrateprintf(rate, format, args...)		\
1672 	krateprintf(rate, "%s: "format, __func__,## args)
1673 
1674 #define hpanic(format, args...)				\
1675 	panic("%s: "format, __func__,## args)
1676 #endif  /* _KERNEL */
1677 
1678 #endif /* !VFS_HAMMER_HAMMER_H_ */
1679