HAMMER 39/Many: Parallel operations optimizations
[dragonfly.git] / sys / vfs / hammer / hammer.h
1 /*
2  * Copyright (c) 2007 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  * $DragonFly: src/sys/vfs/hammer/hammer.h,v 1.54 2008/04/29 01:10:37 dillon Exp $
35  */
36 /*
37  * This header file contains structures used internally by the HAMMERFS
38  * implementation.  See hammer_disk.h for on-disk structures.
39  */
40
41 #include <sys/param.h>
42 #include <sys/types.h>
43 #include <sys/kernel.h>
44 #include <sys/conf.h>
45 #include <sys/systm.h>
46 #include <sys/tree.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/mountctl.h>
50 #include <sys/vnode.h>
51 #include <sys/proc.h>
52 #include <sys/globaldata.h>
53 #include <sys/lockf.h>
54 #include <sys/buf.h>
55 #include <sys/queue.h>
56 #include <sys/globaldata.h>
57
58 #include <sys/buf2.h>
59 #include <sys/signal2.h>
60 #include "hammer_disk.h"
61 #include "hammer_mount.h"
62 #include "hammer_ioctl.h"
63
64 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
65
66 MALLOC_DECLARE(M_HAMMER);
67
68 struct hammer_mount;
69
70 /*
71  * Key structure used for custom RB tree inode lookups.  This prototypes
72  * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
73  */
74 typedef struct hammer_inode_info {
75         int64_t         obj_id;         /* (key) object identifier */
76         hammer_tid_t    obj_asof;       /* (key) snapshot transid or 0 */
77 } *hammer_inode_info_t;
78
79 typedef enum hammer_transaction_type {
80         HAMMER_TRANS_RO,
81         HAMMER_TRANS_STD,
82         HAMMER_TRANS_FLS
83 } hammer_transaction_type_t;
84
85 /*
86  * HAMMER Transaction tracking
87  */
88 struct hammer_transaction {
89         hammer_transaction_type_t type;
90         struct hammer_mount *hmp;
91         hammer_tid_t    tid;
92         hammer_tid_t    time;
93         struct hammer_volume *rootvol;
94 };
95
96 typedef struct hammer_transaction *hammer_transaction_t;
97
98 /*
99  * HAMMER locks
100  */
101 struct hammer_lock {
102         int     refs;           /* active references delay writes */
103         int     lockcount;      /* lock count for exclusive/shared access */
104         int     wanted;
105         struct thread *locktd;
106 };
107
108 static __inline int
109 hammer_islocked(struct hammer_lock *lock)
110 {
111         return(lock->lockcount != 0);
112 }
113
114 static __inline int
115 hammer_isactive(struct hammer_lock *lock)
116 {
117         return(lock->refs != 0);
118 }
119
120 static __inline int
121 hammer_islastref(struct hammer_lock *lock)
122 {
123         return(lock->refs == 1);
124 }
125
126 /*
127  * Return if we specifically own the lock exclusively.
128  */
129 static __inline int
130 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
131 {
132         if (lock->lockcount > 0 && lock->locktd == td)
133                 return(1);
134         return(0);
135 }
136
137 /*
138  * inode->inode dependancy
139  */
140 typedef struct hammer_depend {
141         TAILQ_ENTRY(hammer_depend) ip_entry;
142         TAILQ_ENTRY(hammer_depend) rec_entry;
143         struct hammer_inode *ip;
144         struct hammer_record *record;
145 } *hammer_depend_t;
146
147 TAILQ_HEAD(hammer_depend_list, hammer_depend);
148
149 /*
150  * Cache object ids.  A fixed number of objid cache structures are
151  * created to reserve object id's for newly created files in multiples
152  * of 100,000, localized to a particular directory, and recycled as
153  * needed.  This allows parallel create operations in different
154  * directories to retain fairly localized object ids which in turn
155  * improves reblocking performance and layout.
156  */
157 #define OBJID_CACHE_SIZE        128
158 #define OBJID_CACHE_BULK        100000
159
160 typedef struct hammer_objid_cache {
161         TAILQ_ENTRY(hammer_objid_cache) entry;
162         struct hammer_inode             *dip;
163         hammer_tid_t                    next_tid;
164         int                             count;
165 } *hammer_objid_cache_t;
166
167 /*
168  * Structure used to represent an inode in-memory.
169  *
170  * The record and data associated with an inode may be out of sync with
171  * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
172  * clear).
173  *
174  * An inode may also hold a cache of unsynchronized records, used for
175  * database and directories only.  Unsynchronized regular file data is
176  * stored in the buffer cache.
177  *
178  * NOTE: A file which is created and destroyed within the initial
179  * synchronization period can wind up not doing any disk I/O at all.
180  *
181  * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
182  */
183 struct hammer_ino_rb_tree;
184 struct hammer_inode;
185 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
186 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
187               hammer_ino_rb_compare, hammer_inode_info_t);
188
189 struct hammer_rec_rb_tree;
190 struct hammer_record;
191 RB_HEAD(hammer_rec_rb_tree, hammer_record);
192 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
193               hammer_rec_rb_compare, hammer_base_elm_t);
194
195 TAILQ_HEAD(hammer_node_list, hammer_node);
196
197 typedef enum hammer_inode_state {
198         HAMMER_FST_IDLE,
199         HAMMER_FST_SETUP,
200         HAMMER_FST_FLUSH
201 } hammer_inode_state_t;
202
203 struct hammer_inode {
204         RB_ENTRY(hammer_inode) rb_node;
205         hammer_inode_state_t flush_state;
206         TAILQ_ENTRY(hammer_inode) flush_entry;
207         struct hammer_depend_list depend_list;
208         u_int64_t               obj_id;         /* (key) object identifier */
209         hammer_tid_t            obj_asof;       /* (key) snapshot or 0 */
210         struct hammer_mount     *hmp;
211         hammer_objid_cache_t    objid_cache;
212         int                     flags;
213         int                     error;          /* flush error */
214         int                     depend_count;
215         int                     cursor_ip_refs; /* sanity */
216         struct vnode            *vp;
217         struct lockf            advlock;
218         struct hammer_lock      lock;           /* sync copy interlock */
219         TAILQ_HEAD(, bio)       bio_list;       /* BIOs to flush out */
220         TAILQ_HEAD(, bio)       bio_alt_list;   /* BIOs to flush out */
221         off_t                   trunc_off;
222         struct hammer_inode_record ino_rec;     /* in-memory cache */
223         struct hammer_inode_data ino_data;      /* in-memory cache */
224         struct hammer_rec_rb_tree rec_tree;     /* in-memory cache */
225         struct hammer_node      *cache[2];      /* search initiate cache */
226
227         /*
228          * When a demark is created to synchronize an inode to
229          * disk, certain fields are copied so the front-end VOPs
230          * can continue to run in parallel with the synchronization
231          * occuring in the background.
232          */
233         int             sync_flags;             /* to-sync flags cache */
234         off_t           sync_trunc_off;         /* to-sync truncation */
235         struct hammer_inode_record sync_ino_rec;/* to-sync cache */
236         struct hammer_inode_data sync_ino_data; /* to-sync cache */
237 };
238
239 typedef struct hammer_inode *hammer_inode_t;
240
241 #define VTOI(vp)        ((struct hammer_inode *)(vp)->v_data)
242
243 #define HAMMER_INODE_DDIRTY     0x0001  /* in-memory ino_data is dirty */
244 #define HAMMER_INODE_RDIRTY     0x0002  /* in-memory ino_rec is dirty */
245 #define HAMMER_INODE_ITIMES     0x0004  /* in-memory mtime/atime modified */
246 #define HAMMER_INODE_XDIRTY     0x0008  /* in-memory records/flsbufs present */
247 #define HAMMER_INODE_ONDISK     0x0010  /* inode is on-disk (else not yet) */
248 #define HAMMER_INODE_FLUSH      0x0020  /* flush on last ref */
249 #define HAMMER_INODE_DELETED    0x0080  /* inode ready for deletion */
250 #define HAMMER_INODE_DELONDISK  0x0100  /* delete synchronized to disk */
251 #define HAMMER_INODE_RO         0x0200  /* read-only (because of as-of) */
252 #define HAMMER_INODE_GONE       0x0400  /* delete flushed out */
253 #define HAMMER_INODE_DONDISK    0x0800  /* data records may be on disk */
254 #define HAMMER_INODE_BUFS       0x1000  /* dirty high level bps present */
255 #define HAMMER_INODE_REFLUSH    0x2000  /* pipelined flush during flush */
256 #define HAMMER_INODE_UNUSED4000 0x4000
257 #define HAMMER_INODE_FLUSHW     0x8000  /* Someone waiting for flush */
258
259 #define HAMMER_INODE_TRUNCATED  0x00010000
260 #define HAMMER_INODE_NEW        0x00020000
261
262 #define HAMMER_INODE_MODMASK    (HAMMER_INODE_DDIRTY|HAMMER_INODE_RDIRTY| \
263                                  HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS|   \
264                                  HAMMER_INODE_ITIMES|HAMMER_INODE_TRUNCATED)
265
266 #define HAMMER_MAX_INODE_CURSORS        4
267
268 #define HAMMER_FLUSH_SIGNAL     0x0001
269 #define HAMMER_FLUSH_FORCE      0x0002
270 #define HAMMER_FLUSH_RELEASE    0x0004
271
272 /*
273  * Structure used to represent an unsynchronized record in-memory.  This
274  * structure is orgranized in a per-inode RB-tree.  If the inode is not
275  * on disk then neither are any records and the in-memory record tree
276  * represents the entire contents of the inode.  If the inode is on disk
277  * then the on-disk B-Tree is scanned in parallel with the in-memory
278  * RB-Tree to synthesize the current state of the file.
279  *
280  * Only current (delete_tid == 0) unsynchronized records are kept in-memory.
281  *
282  * blocked is the count of the number of cursors (ip_first/ip_next) blocked
283  * on the record waiting for a synchronization to complete.
284  */
285 struct hammer_record {
286         RB_ENTRY(hammer_record)         rb_node;
287         hammer_inode_state_t            state;
288         struct hammer_lock              lock;
289         struct hammer_inode             *ip;
290         union hammer_record_ondisk      rec;
291         union hammer_data_ondisk        *data;
292         int                             flags;
293         struct hammer_depend_list       depend_list;
294 };
295
296 typedef struct hammer_record *hammer_record_t;
297
298 /*
299  * Record flags.  Note that FE can only be set by the frontend if the
300  * record has not been interlocked by the backend w/ BE.
301  */
302 #define HAMMER_RECF_ALLOCDATA           0x0001
303 #define HAMMER_RECF_ONRBTREE            0x0002
304 #define HAMMER_RECF_DELETED_FE          0x0004  /* deleted (frontend) */
305 #define HAMMER_RECF_DELETED_BE          0x0008  /* deleted (backend) */
306 #define HAMMER_RECF_INBAND              0x0010
307 #define HAMMER_RECF_INTERLOCK_BE        0x0020  /* backend interlock */
308 #define HAMMER_RECF_WANTED              0x0040
309 #define HAMMER_RECF_DELETE_ONDISK       0x0080
310 #define HAMMER_RECF_CONVERT_DELETE_ONDISK 0x0100 /* special case */
311
312 /*
313  * In-memory structures representing on-disk structures.
314  */
315 struct hammer_volume;
316 struct hammer_buffer;
317 struct hammer_node;
318 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
319 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
320 RB_HEAD(hammer_nod_rb_tree, hammer_node);
321
322 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
323               hammer_vol_rb_compare, int32_t);
324 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
325               hammer_buf_rb_compare, hammer_off_t);
326 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
327               hammer_nod_rb_compare, hammer_off_t);
328
329 /*
330  * IO management - embedded at the head of various in-memory structures
331  *
332  * VOLUME       - hammer_volume containing meta-data
333  * META_BUFFER  - hammer_buffer containing meta-data
334  * DATA_BUFFER  - hammer_buffer containing pure-data
335  *
336  * Dirty volume headers and dirty meta-data buffers are locked until the
337  * flusher can sequence them out.  Dirty pure-data buffers can be written.
338  * Clean buffers can be passively released.
339  */
340 typedef enum hammer_io_type {
341         HAMMER_STRUCTURE_VOLUME,
342         HAMMER_STRUCTURE_META_BUFFER,
343         HAMMER_STRUCTURE_UNDO_BUFFER,
344         HAMMER_STRUCTURE_DATA_BUFFER
345 } hammer_io_type_t;
346
347 union hammer_io_structure;
348 struct hammer_io;
349
350 struct worklist {
351         LIST_ENTRY(worklist) node;
352 };
353
354 TAILQ_HEAD(hammer_io_list, hammer_io);
355 typedef struct hammer_io_list *hammer_io_list_t;
356
357 struct hammer_io {
358         struct worklist         worklist;
359         struct hammer_lock      lock;
360         enum hammer_io_type     type;
361         struct hammer_mount     *hmp;
362         TAILQ_ENTRY(hammer_io)  mod_entry; /* list entry if modified */
363         hammer_io_list_t        mod_list;
364         struct buf              *bp;
365         int64_t                 offset;
366         int                     loading;   /* loading/unloading interlock */
367         int                     modify_refs;
368
369         u_int           modified : 1;   /* bp's data was modified */
370         u_int           released : 1;   /* bp released (w/ B_LOCKED set) */
371         u_int           running : 1;    /* bp write IO in progress */
372         u_int           waiting : 1;    /* someone is waiting on us */
373         u_int           validated : 1;  /* ondisk has been validated */
374         u_int           flush : 1;      /* flush on last release */
375         u_int           waitdep : 1;    /* flush waits for dependancies */
376 };
377
378 typedef struct hammer_io *hammer_io_t;
379
380 /*
381  * In-memory volume representing on-disk buffer
382  */
383 struct hammer_volume {
384         struct hammer_io io;
385         RB_ENTRY(hammer_volume) rb_node;
386         struct hammer_buf_rb_tree rb_bufs_root;
387         struct hammer_volume_ondisk *ondisk;
388         int32_t vol_no;
389         int64_t nblocks;        /* note: special calculation for statfs */
390         int64_t buffer_base;    /* base offset of buffer 0 */
391         hammer_off_t maxbuf_off; /* Maximum buffer offset */
392         char    *vol_name;
393         struct vnode *devvp;
394         int     vol_flags;
395 };
396
397 typedef struct hammer_volume *hammer_volume_t;
398
399 /*
400  * In-memory buffer (other then volume, super-cluster, or cluster),
401  * representing an on-disk buffer.
402  */
403 struct hammer_buffer {
404         struct hammer_io io;
405         RB_ENTRY(hammer_buffer) rb_node;
406         void *ondisk;
407         struct hammer_volume *volume;
408         hammer_off_t zone2_offset;
409         hammer_off_t zoneX_offset;
410         struct hammer_node_list clist;
411 };
412
413 typedef struct hammer_buffer *hammer_buffer_t;
414
415 /*
416  * In-memory B-Tree node, representing an on-disk B-Tree node.
417  *
418  * This is a hang-on structure which is backed by a hammer_buffer,
419  * indexed by a hammer_cluster, and used for fine-grained locking of
420  * B-Tree nodes in order to properly control lock ordering.  A hammer_buffer
421  * can contain multiple nodes representing wildly disassociated portions
422  * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
423  *
424  * This structure uses a cluster-relative index to reduce the number
425  * of layers required to access it, and also because all on-disk B-Tree
426  * references are cluster-relative offsets.
427  */
428 struct hammer_node {
429         struct hammer_lock      lock;           /* node-by-node lock */
430         TAILQ_ENTRY(hammer_node) entry;         /* per-buffer linkage */
431         RB_ENTRY(hammer_node)   rb_node;        /* per-cluster linkage */
432         hammer_off_t            node_offset;    /* full offset spec */
433         struct hammer_mount     *hmp;
434         struct hammer_buffer    *buffer;        /* backing buffer */
435         hammer_node_ondisk_t    ondisk;         /* ptr to on-disk structure */
436         struct hammer_node      **cache1;       /* passive cache(s) */
437         struct hammer_node      **cache2;
438         int                     flags;
439         int                     loading;        /* load interlock */
440 };
441
442 #define HAMMER_NODE_DELETED     0x0001
443 #define HAMMER_NODE_FLUSH       0x0002
444
445 typedef struct hammer_node      *hammer_node_t;
446
447 /*
448  * List of locked nodes.
449  */
450 struct hammer_node_locklist {
451         struct hammer_node_locklist *next;
452         hammer_node_t   node;
453 };
454
455 typedef struct hammer_node_locklist *hammer_node_locklist_t;
456
457
458 /*
459  * Common I/O management structure - embedded in in-memory structures
460  * which are backed by filesystem buffers.
461  */
462 union hammer_io_structure {
463         struct hammer_io        io;
464         struct hammer_volume    volume;
465         struct hammer_buffer    buffer;
466 };
467
468 typedef union hammer_io_structure *hammer_io_structure_t;
469
470 /*
471  * Allocation holes are recorded for a short period of time in an attempt
472  * to use up the space.
473  */
474
475 #define HAMMER_MAX_HOLES        8
476
477 struct hammer_hole;
478
479 struct hammer_holes {
480         TAILQ_HEAD(, hammer_hole) list;
481         int     count;
482 };
483
484 typedef struct hammer_holes *hammer_holes_t;
485
486 struct hammer_hole {
487         TAILQ_ENTRY(hammer_hole) entry;
488         hammer_off_t    offset;
489         int             bytes;
490 };
491
492 typedef struct hammer_hole *hammer_hole_t;
493
494 #include "hammer_cursor.h"
495
496 /*
497  * Internal hammer mount data structure
498  */
499 struct hammer_mount {
500         struct mount *mp;
501         /*struct vnode *rootvp;*/
502         struct hammer_ino_rb_tree rb_inos_root;
503         struct hammer_vol_rb_tree rb_vols_root;
504         struct hammer_nod_rb_tree rb_nods_root;
505         struct hammer_volume *rootvol;
506         struct hammer_base_elm root_btree_beg;
507         struct hammer_base_elm root_btree_end;
508         char    *zbuf;  /* HAMMER_BUFSIZE bytes worth of all-zeros */
509         int     hflags;
510         int     ronly;
511         int     nvolumes;
512         int     volume_iterator;
513         int     flusher_seq;
514         int     flusher_act;
515         int     flusher_exiting;
516         int     reclaim_count;
517         thread_t flusher_td;
518         u_int   check_interrupt;
519         uuid_t  fsid;
520         udev_t  fsid_udev;
521         struct hammer_io_list volu_list;        /* dirty undo buffers */
522         struct hammer_io_list undo_list;        /* dirty undo buffers */
523         struct hammer_io_list data_list;        /* dirty data buffers */
524         struct hammer_io_list meta_list;        /* dirty meta bufs    */
525         struct hammer_io_list lose_list;        /* loose buffers      */
526         int     locked_dirty_count;             /* meta/volu count    */
527         int     io_running_count;
528         int     objid_cache_count;
529         hammer_tid_t asof;
530         hammer_off_t next_tid;
531         u_int32_t namekey_iterator;
532         hammer_off_t zone_limits[HAMMER_MAX_ZONES];
533         struct netexport export;
534         struct hammer_lock sync_lock;
535         struct lock blockmap_lock;
536         hammer_inode_t  flusher_demark;
537         struct hammer_blockmap  blockmap[HAMMER_MAX_ZONES];
538         struct hammer_holes holes[HAMMER_MAX_ZONES];
539         TAILQ_HEAD(, hammer_inode) flush_list;
540         TAILQ_HEAD(, hammer_inode) flush_alt_list;
541         TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
542 };
543
544 typedef struct hammer_mount     *hammer_mount_t;
545
546 struct hammer_sync_info {
547         int error;
548         int waitfor;
549 };
550
551 #endif
552
553 #if defined(_KERNEL)
554
555 extern struct vop_ops hammer_vnode_vops;
556 extern struct vop_ops hammer_spec_vops;
557 extern struct vop_ops hammer_fifo_vops;
558 extern struct bio_ops hammer_bioops;
559
560 extern int hammer_debug_general;
561 extern int hammer_debug_locks;
562 extern int hammer_debug_btree;
563 extern int hammer_debug_tid;
564 extern int hammer_debug_recover;
565 extern int hammer_debug_recover_faults;
566 extern int hammer_count_inodes;
567 extern int hammer_count_records;
568 extern int hammer_count_record_datas;
569 extern int hammer_count_volumes;
570 extern int hammer_count_buffers;
571 extern int hammer_count_nodes;
572 extern int hammer_count_dirtybufs;
573 extern int hammer_limit_dirtybufs;
574 extern int64_t hammer_contention_count;
575
576 int     hammer_vop_inactive(struct vop_inactive_args *);
577 int     hammer_vop_reclaim(struct vop_reclaim_args *);
578 int     hammer_get_vnode(struct hammer_inode *ip, int lktype,
579                         struct vnode **vpp);
580 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
581                         struct hammer_node **cache,
582                         u_int64_t obj_id, hammer_tid_t asof, int flags,
583                         int *errorp);
584 void    hammer_put_inode(struct hammer_inode *ip);
585 void    hammer_put_inode_ref(struct hammer_inode *ip);
586
587 int     hammer_unload_volume(hammer_volume_t volume, void *data __unused);
588 int     hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused);
589 int     hammer_install_volume(hammer_mount_t hmp, const char *volname);
590
591 int     hammer_ip_lookup(hammer_cursor_t cursor, hammer_inode_t ip);
592 int     hammer_ip_first(hammer_cursor_t cursor, hammer_inode_t ip);
593 int     hammer_ip_next(hammer_cursor_t cursor);
594 int     hammer_ip_resolve_record_and_data(hammer_cursor_t cursor);
595 int     hammer_ip_resolve_data(hammer_cursor_t cursor);
596 int     hammer_ip_delete_record(hammer_cursor_t cursor, hammer_tid_t tid);
597 int     hammer_delete_at_cursor(hammer_cursor_t cursor, int64_t *stat_bytes);
598 int     hammer_ip_check_directory_empty(hammer_transaction_t trans,
599                         hammer_inode_t ip);
600 int     hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
601
602 hammer_record_t
603         hammer_alloc_mem_record(hammer_inode_t ip);
604 void    hammer_flush_record_done(hammer_record_t record, int error);
605 void    hammer_wait_mem_record(hammer_record_t record);
606 void    hammer_rel_mem_record(hammer_record_t record);
607 void    hammer_cleardep_mem_record(struct hammer_record *record);
608
609 int     hammer_cursor_up(hammer_cursor_t cursor);
610 int     hammer_cursor_down(hammer_cursor_t cursor);
611 int     hammer_cursor_upgrade(hammer_cursor_t cursor);
612 void    hammer_cursor_downgrade(hammer_cursor_t cursor);
613 int     hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
614                         int index);
615 void    hammer_lock_ex(struct hammer_lock *lock);
616 int     hammer_lock_ex_try(struct hammer_lock *lock);
617 void    hammer_lock_sh(struct hammer_lock *lock);
618 int     hammer_lock_upgrade(struct hammer_lock *lock);
619 void    hammer_lock_downgrade(struct hammer_lock *lock);
620 void    hammer_unlock(struct hammer_lock *lock);
621 void    hammer_ref(struct hammer_lock *lock);
622 void    hammer_unref(struct hammer_lock *lock);
623
624 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
625 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
626 void    hammer_to_timespec(hammer_tid_t tid, struct timespec *ts);
627 hammer_tid_t hammer_timespec_to_transid(struct timespec *ts);
628 hammer_tid_t hammer_now_tid(void);
629 hammer_tid_t hammer_str_to_tid(const char *str);
630 hammer_tid_t hammer_alloc_objid(hammer_transaction_t trans, hammer_inode_t dip);
631 void hammer_clear_objid(hammer_inode_t dip);
632 void hammer_destroy_objid_cache(hammer_mount_t hmp);
633
634 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
635 int hammer_get_dtype(u_int8_t obj_type);
636 u_int8_t hammer_get_obj_type(enum vtype vtype);
637 int64_t hammer_directory_namekey(void *name, int len);
638
639 int     hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
640                            struct hammer_node **cache);
641
642 void    hammer_done_cursor(hammer_cursor_t cursor);
643 void    hammer_mem_done(hammer_cursor_t cursor);
644
645 int     hammer_btree_lookup(hammer_cursor_t cursor);
646 int     hammer_btree_first(hammer_cursor_t cursor);
647 int     hammer_btree_last(hammer_cursor_t cursor);
648 int     hammer_btree_extract(hammer_cursor_t cursor, int flags);
649 int     hammer_btree_iterate(hammer_cursor_t cursor);
650 int     hammer_btree_iterate_reverse(hammer_cursor_t cursor);
651 int     hammer_btree_insert(hammer_cursor_t cursor, hammer_btree_elm_t elm);
652 int     hammer_btree_delete(hammer_cursor_t cursor);
653 int     hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
654 int     hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
655 int     hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
656 int     hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
657
658
659 int     hammer_btree_lock_children(hammer_cursor_t cursor,
660                         struct hammer_node_locklist **locklistp);
661
662 void    hammer_print_btree_node(hammer_node_ondisk_t ondisk);
663 void    hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
664
665 void    *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
666                         int *errorp, struct hammer_buffer **bufferp);
667 void    *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
668                         int *errorp, struct hammer_buffer **bufferp);
669
670 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
671 int     hammer_dowrite(hammer_transaction_t trans, hammer_inode_t ip,
672                         struct bio *bio);
673
674 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
675                         int32_t vol_no, int *errorp);
676 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp,
677                         hammer_off_t buf_offset, int isnew, int *errorp);
678 void    hammer_uncache_buffer(struct hammer_mount *hmp, hammer_off_t off);
679
680 int             hammer_ref_volume(hammer_volume_t volume);
681 int             hammer_ref_buffer(hammer_buffer_t buffer);
682 void            hammer_flush_buffer_nodes(hammer_buffer_t buffer);
683
684 void            hammer_rel_volume(hammer_volume_t volume, int flush);
685 void            hammer_rel_buffer(hammer_buffer_t buffer, int flush);
686
687 int             hammer_vfs_export(struct mount *mp, int op,
688                         const struct export_args *export);
689 hammer_node_t   hammer_get_node(hammer_mount_t hmp,
690                         hammer_off_t node_offset, int *errorp);
691 void            hammer_ref_node(hammer_node_t node);
692 hammer_node_t   hammer_ref_node_safe(struct hammer_mount *hmp,
693                         struct hammer_node **cache, int *errorp);
694 void            hammer_rel_node(hammer_node_t node);
695 void            hammer_delete_node(hammer_transaction_t trans,
696                         hammer_node_t node);
697 void            hammer_cache_node(hammer_node_t node,
698                         struct hammer_node **cache);
699 void            hammer_uncache_node(struct hammer_node **cache);
700 void            hammer_flush_node(hammer_node_t node);
701
702 void hammer_dup_buffer(struct hammer_buffer **bufferp,
703                         struct hammer_buffer *buffer);
704 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans, int *errorp);
705 void *hammer_alloc_record(hammer_transaction_t trans,
706                         hammer_off_t *rec_offp, u_int16_t rec_type,
707                         struct hammer_buffer **rec_bufferp,
708                         int32_t data_len, void **datap,
709                         struct hammer_buffer **data_bufferp, int *errorp);
710 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
711                         hammer_off_t *data_offsetp,
712                         struct hammer_buffer **data_bufferp, int *errorp);
713
714 int hammer_generate_undo(hammer_transaction_t trans, hammer_io_t io,
715                         hammer_off_t zone1_offset, void *base, int len);
716
717 void hammer_put_volume(struct hammer_volume *volume, int flush);
718 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
719
720 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
721                         hammer_off_t owner, int *errorp);
722 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
723                         hammer_off_t owner, int *errorp);
724 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
725                         int bytes, int *errorp);
726 void hammer_blockmap_free(hammer_transaction_t trans,
727                         hammer_off_t bmap_off, int bytes);
728 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
729                         int *curp, int *errorp);
730 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
731                         int *errorp);
732 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
733                         int *errorp);
734
735 void hammer_start_transaction(struct hammer_transaction *trans,
736                               struct hammer_mount *hmp);
737 void hammer_simple_transaction(struct hammer_transaction *trans,
738                               struct hammer_mount *hmp);
739 void hammer_start_transaction_fls(struct hammer_transaction *trans,
740                                   struct hammer_mount *hmp);
741 void hammer_done_transaction(struct hammer_transaction *trans);
742
743 void hammer_modify_inode(struct hammer_transaction *trans,
744                         hammer_inode_t ip, int flags);
745 void hammer_flush_inode(hammer_inode_t ip, int flags);
746 void hammer_flush_inode_done(hammer_inode_t ip);
747 void hammer_wait_inode(hammer_inode_t ip);
748
749 int  hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
750                         struct ucred *cred, struct hammer_inode *dip,
751                         struct hammer_inode **ipp);
752 void  hammer_finalize_inode(hammer_transaction_t trans, hammer_inode_t ip,
753                         int error);
754 void hammer_rel_inode(hammer_inode_t ip, int flush);
755 int hammer_sync_inode(hammer_inode_t ip, int handle_delete);
756
757 int  hammer_ip_add_directory(struct hammer_transaction *trans,
758                         hammer_inode_t dip, struct namecache *ncp,
759                         hammer_inode_t nip);
760 int  hammer_ip_del_directory(struct hammer_transaction *trans,
761                         hammer_cursor_t cursor, hammer_inode_t dip,
762                         hammer_inode_t ip);
763 int  hammer_ip_add_record(struct hammer_transaction *trans,
764                         hammer_record_t record);
765 int  hammer_ip_delete_range(struct hammer_transaction *trans,
766                         hammer_inode_t ip, int64_t ran_beg, int64_t ran_end);
767 int  hammer_ip_delete_range_all(struct hammer_transaction *trans,
768                         hammer_inode_t ip);
769 int  hammer_ip_sync_data(struct hammer_transaction *trans,
770                         hammer_inode_t ip, int64_t offset,
771                         void *data, int bytes);
772 int  hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
773
774 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
775                         struct ucred *cred);
776
777 void hammer_io_init(hammer_io_t io, hammer_mount_t hmp,
778                         enum hammer_io_type type);
779 void hammer_io_reinit(hammer_io_t io, enum hammer_io_type type);
780 int hammer_io_read(struct vnode *devvp, struct hammer_io *io);
781 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
782 void hammer_io_release(struct hammer_io *io);
783 void hammer_io_flush(struct hammer_io *io);
784 int hammer_io_checkflush(hammer_io_t io);
785 void hammer_io_clear_modify(struct hammer_io *io);
786 void hammer_io_waitdep(struct hammer_io *io);
787
788 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
789                         void *base, int len);
790 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
791                         void *base, int len);
792 void hammer_modify_volume_done(hammer_volume_t volume);
793 void hammer_modify_buffer_done(hammer_buffer_t buffer);
794
795 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
796                         struct hammer_ioc_reblock *reblock);
797
798 void hammer_init_holes(hammer_mount_t hmp, hammer_holes_t holes);
799 void hammer_free_holes(hammer_mount_t hmp, hammer_holes_t holes);
800 int hammer_signal_check(hammer_mount_t hmp);
801
802 void hammer_flusher_create(hammer_mount_t hmp);
803 void hammer_flusher_destroy(hammer_mount_t hmp);
804 void hammer_flusher_sync(hammer_mount_t hmp);
805 void hammer_flusher_async(hammer_mount_t hmp);
806
807 int hammer_recover(hammer_mount_t hmp, hammer_volume_t rootvol);
808
809 #endif
810
811 static __inline void
812 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
813 {
814         hammer_modify_buffer(trans, node->buffer, NULL, 0);
815 }
816
817 static __inline void
818 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
819 {
820         hammer_modify_buffer(trans, node->buffer,
821                              node->ondisk, sizeof(*node->ondisk));
822 }
823
824 static __inline void
825 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
826                    void *base, int len)
827 {
828         KKASSERT((char *)base >= (char *)node->ondisk &&
829                  (char *)base + len <=
830                     (char *)node->ondisk + sizeof(*node->ondisk));
831         hammer_modify_buffer(trans, node->buffer, base, len);
832 }
833
834 static __inline void
835 hammer_modify_node_done(hammer_node_t node)
836 {
837         hammer_modify_buffer_done(node->buffer);
838 }
839
840 static __inline void
841 hammer_modify_record(hammer_transaction_t trans, hammer_buffer_t buffer,
842                      void *base, int len)
843 {
844         KKASSERT((char *)base >= (char *)buffer->ondisk &&
845                  (char *)base + len <= (char *)buffer->ondisk + HAMMER_BUFSIZE);
846         hammer_modify_buffer(trans, buffer, base, len);
847 }
848
849 static __inline void
850 hammer_modify_record_done(hammer_buffer_t buffer)
851 {
852         hammer_modify_buffer_done(buffer);
853 }
854