HAMMER VFS - Version 4 part 1/many - UNDO FIFO layout work.
[dragonfly.git] / sys / vfs / hammer / hammer.h
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  * $DragonFly: src/sys/vfs/hammer/hammer.h,v 1.130 2008/11/13 02:18:43 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/priv.h>
53 #include <sys/stat.h>
54 #include <sys/globaldata.h>
55 #include <sys/lockf.h>
56 #include <sys/buf.h>
57 #include <sys/queue.h>
58 #include <sys/ktr.h>
59 #include <sys/globaldata.h>
60
61 #include <sys/buf2.h>
62 #include <sys/signal2.h>
63 #include "hammer_disk.h"
64 #include "hammer_mount.h"
65 #include "hammer_ioctl.h"
66
67 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
68
69 MALLOC_DECLARE(M_HAMMER);
70
71 /*
72  * Kernel trace
73  */
74 #if !defined(KTR_HAMMER)
75 #define KTR_HAMMER      KTR_ALL
76 #endif
77 KTR_INFO_MASTER_EXTERN(hammer);
78
79 /*
80  * Misc structures
81  */
82 struct hammer_mount;
83
84 /*
85  * Key structure used for custom RB tree inode lookups.  This prototypes
86  * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
87  */
88 typedef struct hammer_inode_info {
89         int64_t         obj_id;         /* (key) object identifier */
90         hammer_tid_t    obj_asof;       /* (key) snapshot transid or 0 */
91         u_int32_t       obj_localization; /* (key) pseudo-fs */
92         union {
93                 struct hammer_btree_leaf_elm *leaf;
94         } u;
95 } *hammer_inode_info_t;
96
97 typedef enum hammer_transaction_type {
98         HAMMER_TRANS_RO,
99         HAMMER_TRANS_STD,
100         HAMMER_TRANS_FLS
101 } hammer_transaction_type_t;
102
103 /*
104  * HAMMER Transaction tracking
105  */
106 struct hammer_transaction {
107         hammer_transaction_type_t type;
108         struct hammer_mount *hmp;
109         hammer_tid_t    tid;
110         u_int64_t       time;
111         u_int32_t       time32;
112         int             sync_lock_refs;
113         int             flags;
114         struct hammer_volume *rootvol;
115 };
116
117 typedef struct hammer_transaction *hammer_transaction_t;
118
119 #define HAMMER_TRANSF_NEWINODE  0x0001
120 #define HAMMER_TRANSF_DIDIO     0x0002
121 #define HAMMER_TRANSF_CRCDOM    0x0004  /* EDOM on CRC error, less critical */
122
123 /*
124  * HAMMER locks
125  */
126 struct hammer_lock {
127         int             refs;           /* active references delay writes */
128         volatile u_int  lockval;        /* lock count and control bits */
129         struct thread   *owner;         /* owner if exclusively held */
130 };
131
132 #define HAMMER_LOCKF_EXCLUSIVE  0x40000000
133 #define HAMMER_LOCKF_WANTED     0x80000000
134
135 static __inline int
136 hammer_notlocked(struct hammer_lock *lock)
137 {
138         return(lock->lockval == 0);
139 }
140
141 static __inline int
142 hammer_islocked(struct hammer_lock *lock)
143 {
144         return(lock->lockval != 0);
145 }
146
147 static __inline int
148 hammer_isactive(struct hammer_lock *lock)
149 {
150         return(lock->refs != 0);
151 }
152
153 static __inline int
154 hammer_islastref(struct hammer_lock *lock)
155 {
156         return(lock->refs == 1);
157 }
158
159 /*
160  * Return if we specifically own the lock exclusively.
161  */
162 static __inline int
163 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
164 {
165         if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) &&
166             lock->owner == td) {
167                 return(1);
168         }
169         return(0);
170 }
171
172 /*
173  * Flush state, used by various structures
174  */
175 typedef enum hammer_inode_state {
176         HAMMER_FST_IDLE,
177         HAMMER_FST_SETUP,
178         HAMMER_FST_FLUSH
179 } hammer_inode_state_t;
180
181 TAILQ_HEAD(hammer_record_list, hammer_record);
182
183 /*
184  * Pseudo-filesystem extended data tracking
185  */
186 struct hammer_pfs_rb_tree;
187 struct hammer_pseudofs_inmem;
188 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
189 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
190               hammer_pfs_rb_compare, u_int32_t);
191
192 struct hammer_pseudofs_inmem {
193         RB_ENTRY(hammer_pseudofs_inmem) rb_node;
194         struct hammer_lock      lock;
195         u_int32_t               localization;
196         hammer_tid_t            create_tid;
197         int                     flags;
198         udev_t                  fsid_udev;
199         struct hammer_pseudofs_data pfsd;
200 };
201
202 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t;
203
204 #define HAMMER_PFSM_DELETED     0x0001
205
206 /*
207  * Cache object ids.  A fixed number of objid cache structures are
208  * created to reserve object id's for newly created files in multiples
209  * of 100,000, localized to a particular directory, and recycled as
210  * needed.  This allows parallel create operations in different
211  * directories to retain fairly localized object ids which in turn
212  * improves reblocking performance and layout.
213  */
214 #define OBJID_CACHE_SIZE        1024
215 #define OBJID_CACHE_BULK_BITS   10              /* 10 bits (1024)       */
216 #define OBJID_CACHE_BULK        (32 * 32)       /* two level (1024)     */
217 #define OBJID_CACHE_BULK_MASK   (OBJID_CACHE_BULK - 1)
218 #define OBJID_CACHE_BULK_MASK64 ((u_int64_t)(OBJID_CACHE_BULK - 1))
219
220 typedef struct hammer_objid_cache {
221         TAILQ_ENTRY(hammer_objid_cache) entry;
222         struct hammer_inode             *dip;
223         hammer_tid_t                    base_tid;
224         int                             count;
225         u_int32_t                       bm0;
226         u_int32_t                       bm1[32];
227 } *hammer_objid_cache_t;
228
229 /*
230  * Associate an inode with a B-Tree node to cache search start positions
231  */
232 typedef struct hammer_node_cache {
233         TAILQ_ENTRY(hammer_node_cache)  entry;
234         struct hammer_node              *node;
235         struct hammer_inode             *ip;
236 } *hammer_node_cache_t;
237
238 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
239
240 /*
241  * Structure used to organize flush groups.  Flush groups must be
242  * organized into chunks in order to avoid blowing out the UNDO FIFO.
243  * Without this a 'sync' could end up flushing 50,000 inodes in a single
244  * transaction.
245  */
246 struct hammer_flush_group {
247         TAILQ_ENTRY(hammer_flush_group) flush_entry;
248         TAILQ_HEAD(, hammer_inode)      flush_list;
249         int                             unused01;       /* inode load */
250         int                             total_count;    /* record load */
251         int                             running;        /* group is running */
252         int                             closed;
253         int                             refs;
254 };
255
256 typedef struct hammer_flush_group *hammer_flush_group_t;
257
258 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
259
260 /*
261  * Structure used to represent an inode in-memory.
262  *
263  * The record and data associated with an inode may be out of sync with
264  * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
265  * clear).
266  *
267  * An inode may also hold a cache of unsynchronized records, used for
268  * database and directories only.  Unsynchronized regular file data is
269  * stored in the buffer cache.
270  *
271  * NOTE: A file which is created and destroyed within the initial
272  * synchronization period can wind up not doing any disk I/O at all.
273  *
274  * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
275  */
276 struct hammer_ino_rb_tree;
277 struct hammer_inode;
278 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
279 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
280               hammer_ino_rb_compare, hammer_inode_info_t);
281
282 struct hammer_rec_rb_tree;
283 struct hammer_record;
284 RB_HEAD(hammer_rec_rb_tree, hammer_record);
285 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
286               hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
287
288 TAILQ_HEAD(hammer_node_list, hammer_node);
289
290 struct hammer_inode {
291         RB_ENTRY(hammer_inode)  rb_node;
292         hammer_inode_state_t    flush_state;
293         hammer_flush_group_t    flush_group;
294         TAILQ_ENTRY(hammer_inode) flush_entry;
295         struct hammer_record_list target_list;  /* target of dependant recs */
296         int64_t                 obj_id;         /* (key) object identifier */
297         hammer_tid_t            obj_asof;       /* (key) snapshot or 0 */
298         u_int32_t               obj_localization; /* (key) pseudo-fs */
299         struct hammer_mount     *hmp;
300         hammer_objid_cache_t    objid_cache;
301         int                     flags;
302         int                     error;          /* flush error */
303         int                     cursor_ip_refs; /* sanity */
304         int                     rsv_recs;
305         struct vnode            *vp;
306         hammer_pseudofs_inmem_t pfsm;
307         struct lockf            advlock;
308         struct hammer_lock      lock;           /* sync copy interlock */
309         off_t                   trunc_off;
310         struct hammer_btree_leaf_elm ino_leaf;  /* in-memory cache */
311         struct hammer_inode_data ino_data;      /* in-memory cache */
312         struct hammer_rec_rb_tree rec_tree;     /* in-memory cache */
313         int                     rec_generation;
314         struct hammer_node_cache cache[4];      /* search initiate cache */
315
316         /*
317          * When a demark is created to synchronize an inode to
318          * disk, certain fields are copied so the front-end VOPs
319          * can continue to run in parallel with the synchronization
320          * occuring in the background.
321          */
322         int             sync_flags;             /* to-sync flags cache */
323         off_t           sync_trunc_off;         /* to-sync truncation */
324         off_t           save_trunc_off;         /* write optimization */
325         struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
326         struct hammer_inode_data sync_ino_data; /* to-sync cache */
327 };
328
329 typedef struct hammer_inode *hammer_inode_t;
330
331 #define VTOI(vp)        ((struct hammer_inode *)(vp)->v_data)
332
333 #define HAMMER_INODE_DDIRTY     0x0001  /* in-memory ino_data is dirty */
334                                         /* (not including atime/mtime) */
335 #define HAMMER_INODE_RSV_INODES 0x0002  /* hmp->rsv_inodes bumped */
336 #define HAMMER_INODE_CONN_DOWN  0x0004  /* include in downward recursion */
337 #define HAMMER_INODE_XDIRTY     0x0008  /* in-memory records */
338 #define HAMMER_INODE_ONDISK     0x0010  /* inode is on-disk (else not yet) */
339 #define HAMMER_INODE_FLUSH      0x0020  /* flush on last ref */
340 #define HAMMER_INODE_DELETED    0x0080  /* inode delete (backend) */
341 #define HAMMER_INODE_DELONDISK  0x0100  /* delete synchronized to disk */
342 #define HAMMER_INODE_RO         0x0200  /* read-only (because of as-of) */
343 #define HAMMER_INODE_VHELD      0x0400  /* vnode held on sync */
344 #define HAMMER_INODE_DONDISK    0x0800  /* data records may be on disk */
345 #define HAMMER_INODE_BUFS       0x1000  /* dirty high level bps present */
346 #define HAMMER_INODE_REFLUSH    0x2000  /* flush on dependancy / reflush */
347 #define HAMMER_INODE_RECLAIM    0x4000  /* trying to reclaim */
348 #define HAMMER_INODE_FLUSHW     0x8000  /* Someone waiting for flush */
349
350 #define HAMMER_INODE_TRUNCATED  0x00010000
351 #define HAMMER_INODE_DELETING   0x00020000 /* inode delete request (frontend)*/
352 #define HAMMER_INODE_RESIGNAL   0x00040000 /* re-signal on re-flush */
353 #define HAMMER_INODE_ATIME      0x00100000 /* in-memory atime modified */
354 #define HAMMER_INODE_MTIME      0x00200000 /* in-memory mtime modified */
355 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
356 #define HAMMER_INODE_DUMMY      0x00800000 /* dummy inode covering bad file */
357 #define HAMMER_INODE_CLOSESYNC  0x01000000 /* synchronously fsync on close */
358 #define HAMMER_INODE_CLOSEASYNC 0x02000000 /* asynchronously fsync on close */
359
360 #define HAMMER_INODE_MODMASK    (HAMMER_INODE_DDIRTY|                       \
361                                  HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS|     \
362                                  HAMMER_INODE_ATIME|HAMMER_INODE_MTIME|     \
363                                  HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
364
365 #define HAMMER_INODE_MODMASK_NOXDIRTY \
366                                 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
367
368 #define HAMMER_FLUSH_GROUP_SIZE 64
369
370 #define HAMMER_FLUSH_SIGNAL     0x0001
371 #define HAMMER_FLUSH_RECURSION  0x0002
372
373 /*
374  * Used by the inode reclaim code to pipeline reclaims and avoid
375  * blowing out kernel memory or letting the flusher get too far
376  * behind.  The reclaim wakes up when count reaches 0 or the
377  * timer expires.
378  */
379 struct hammer_reclaim {
380         TAILQ_ENTRY(hammer_reclaim) entry;
381         int     count;
382 };
383
384 #define HAMMER_RECLAIM_FLUSH    2000
385 #define HAMMER_RECLAIM_WAIT     4000
386
387 /*
388  * Structure used to represent an unsynchronized record in-memory.  These
389  * records typically represent directory entries.  Only non-historical
390  * records are kept in-memory.
391  *
392  * Records are organized as a per-inode RB-Tree.  If the inode is not
393  * on disk then neither are any records and the in-memory record tree
394  * represents the entire contents of the inode.  If the inode is on disk
395  * then the on-disk B-Tree is scanned in parallel with the in-memory
396  * RB-Tree to synthesize the current state of the file.
397  *
398  * Records are also used to enforce the ordering of directory create/delete
399  * operations.  A new inode will not be flushed to disk unless its related
400  * directory entry is also being flushed at the same time.  A directory entry
401  * will not be removed unless its related inode is also being removed at the
402  * same time.
403  */
404 typedef enum hammer_record_type {
405         HAMMER_MEM_RECORD_GENERAL,      /* misc record */
406         HAMMER_MEM_RECORD_INODE,        /* inode record */
407         HAMMER_MEM_RECORD_ADD,          /* positive memory cache record */
408         HAMMER_MEM_RECORD_DEL,          /* negative delete-on-disk record */
409         HAMMER_MEM_RECORD_DATA          /* bulk-data record w/on-disk ref */
410 } hammer_record_type_t;
411
412 struct hammer_record {
413         RB_ENTRY(hammer_record)         rb_node;
414         TAILQ_ENTRY(hammer_record)      target_entry;
415         hammer_inode_state_t            flush_state;
416         hammer_flush_group_t            flush_group;
417         hammer_record_type_t            type;
418         struct hammer_lock              lock;
419         struct hammer_reserve           *resv;
420         struct hammer_inode             *ip;
421         struct hammer_inode             *target_ip;
422         struct hammer_btree_leaf_elm    leaf;
423         union hammer_data_ondisk        *data;
424         int                             flags;
425         hammer_off_t                    zone2_offset;   /* direct-write only */
426 };
427
428 typedef struct hammer_record *hammer_record_t;
429
430 /*
431  * Record flags.  Note that FE can only be set by the frontend if the
432  * record has not been interlocked by the backend w/ BE.
433  */
434 #define HAMMER_RECF_ALLOCDATA           0x0001
435 #define HAMMER_RECF_ONRBTREE            0x0002
436 #define HAMMER_RECF_DELETED_FE          0x0004  /* deleted (frontend) */
437 #define HAMMER_RECF_DELETED_BE          0x0008  /* deleted (backend) */
438 #define HAMMER_RECF_COMMITTED           0x0010  /* committed to the B-Tree */
439 #define HAMMER_RECF_INTERLOCK_BE        0x0020  /* backend interlock */
440 #define HAMMER_RECF_WANTED              0x0040  /* wanted by the frontend */
441 #define HAMMER_RECF_CONVERT_DELETE      0x0100  /* special case */
442 #define HAMMER_RECF_DIRECT_IO           0x0200  /* related direct I/O running*/
443 #define HAMMER_RECF_DIRECT_WAIT         0x0400  /* related direct I/O running*/
444 #define HAMMER_RECF_DIRECT_INVAL        0x0800  /* buffer alias invalidation */
445
446 /*
447  * hammer_create_at_cursor() and hammer_delete_at_cursor() flags.
448  */
449 #define HAMMER_CREATE_MODE_UMIRROR      0x0001
450 #define HAMMER_CREATE_MODE_SYS          0x0002
451
452 #define HAMMER_DELETE_ADJUST            0x0001
453 #define HAMMER_DELETE_DESTROY           0x0002
454
455 /*
456  * In-memory structures representing on-disk structures.
457  */
458 struct hammer_volume;
459 struct hammer_buffer;
460 struct hammer_node;
461 struct hammer_undo;
462 struct hammer_reserve;
463
464 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
465 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
466 RB_HEAD(hammer_nod_rb_tree, hammer_node);
467 RB_HEAD(hammer_und_rb_tree, hammer_undo);
468 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
469
470 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
471               hammer_vol_rb_compare, int32_t);
472 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
473               hammer_buf_rb_compare, hammer_off_t);
474 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
475               hammer_nod_rb_compare, hammer_off_t);
476 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
477               hammer_und_rb_compare, hammer_off_t);
478 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
479               hammer_res_rb_compare, hammer_off_t);
480
481 /*
482  * IO management - embedded at the head of various in-memory structures
483  *
484  * VOLUME       - hammer_volume containing meta-data
485  * META_BUFFER  - hammer_buffer containing meta-data
486  * DATA_BUFFER  - hammer_buffer containing pure-data
487  *
488  * Dirty volume headers and dirty meta-data buffers are locked until the
489  * flusher can sequence them out.  Dirty pure-data buffers can be written.
490  * Clean buffers can be passively released.
491  */
492 typedef enum hammer_io_type {
493         HAMMER_STRUCTURE_VOLUME,
494         HAMMER_STRUCTURE_META_BUFFER,
495         HAMMER_STRUCTURE_UNDO_BUFFER,
496         HAMMER_STRUCTURE_DATA_BUFFER
497 } hammer_io_type_t;
498
499 union hammer_io_structure;
500 struct hammer_io;
501
502 struct worklist {
503         LIST_ENTRY(worklist) node;
504 };
505
506 TAILQ_HEAD(hammer_io_list, hammer_io);
507 typedef struct hammer_io_list *hammer_io_list_t;
508
509 struct hammer_io {
510         struct worklist         worklist;
511         struct hammer_lock      lock;
512         enum hammer_io_type     type;
513         struct hammer_mount     *hmp;
514         struct hammer_volume    *volume;
515         TAILQ_ENTRY(hammer_io)  mod_entry; /* list entry if modified */
516         hammer_io_list_t        mod_list;
517         struct buf              *bp;
518         int64_t                 offset;    /* zone-2 offset */
519         int                     bytes;     /* buffer cache buffer size */
520         int                     loading;   /* loading/unloading interlock */
521         int                     modify_refs;
522
523         u_int           modified : 1;   /* bp's data was modified */
524         u_int           released : 1;   /* bp released (w/ B_LOCKED set) */
525         u_int           running : 1;    /* bp write IO in progress */
526         u_int           waiting : 1;    /* someone is waiting on us */
527         u_int           validated : 1;  /* ondisk has been validated */
528         u_int           waitdep : 1;    /* flush waits for dependancies */
529         u_int           recovered : 1;  /* has recovery ref */
530         u_int           waitmod : 1;    /* waiting for modify_refs */
531         u_int           reclaim : 1;    /* reclaim requested */
532         u_int           gencrc : 1;     /* crc needs to be generated */
533         u_int           ioerror : 1;    /* abort on io-error */
534 };
535
536 typedef struct hammer_io *hammer_io_t;
537
538 #define HAMMER_CLUSTER_SIZE     (64 * 1024)
539 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
540 #undef  HAMMER_CLUSTER_SIZE
541 #define HAMMER_CLUSTER_SIZE     MAXBSIZE
542 #endif
543 #define HAMMER_CLUSTER_BUFS     (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
544
545 /*
546  * In-memory volume representing on-disk buffer
547  */
548 struct hammer_volume {
549         struct hammer_io io;
550         RB_ENTRY(hammer_volume) rb_node;
551         struct hammer_volume_ondisk *ondisk;
552         int32_t vol_no;
553         int64_t nblocks;        /* note: special calculation for statfs */
554         int64_t buffer_base;    /* base offset of buffer 0 */
555         hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
556         hammer_off_t maxraw_off; /* Maximum raw offset for device */
557         char    *vol_name;
558         struct vnode *devvp;
559         int     vol_flags;
560 };
561
562 typedef struct hammer_volume *hammer_volume_t;
563
564 /*
565  * In-memory buffer (other then volume, super-cluster, or cluster),
566  * representing an on-disk buffer.
567  */
568 struct hammer_buffer {
569         struct hammer_io io;
570         RB_ENTRY(hammer_buffer) rb_node;
571         void *ondisk;
572         hammer_off_t zoneX_offset;
573         hammer_off_t zone2_offset;
574         struct hammer_reserve *resv;
575         struct hammer_node_list clist;
576 };
577
578 typedef struct hammer_buffer *hammer_buffer_t;
579
580 /*
581  * In-memory B-Tree node, representing an on-disk B-Tree node.
582  *
583  * This is a hang-on structure which is backed by a hammer_buffer,
584  * indexed by a hammer_cluster, and used for fine-grained locking of
585  * B-Tree nodes in order to properly control lock ordering.  A hammer_buffer
586  * can contain multiple nodes representing wildly disassociated portions
587  * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
588  *
589  * This structure uses a cluster-relative index to reduce the number
590  * of layers required to access it, and also because all on-disk B-Tree
591  * references are cluster-relative offsets.
592  */
593 struct hammer_node {
594         struct hammer_lock      lock;           /* node-by-node lock */
595         TAILQ_ENTRY(hammer_node) entry;         /* per-buffer linkage */
596         RB_ENTRY(hammer_node)   rb_node;        /* per-cluster linkage */
597         hammer_off_t            node_offset;    /* full offset spec */
598         struct hammer_mount     *hmp;
599         struct hammer_buffer    *buffer;        /* backing buffer */
600         hammer_node_ondisk_t    ondisk;         /* ptr to on-disk structure */
601         TAILQ_HEAD(, hammer_cursor) cursor_list;  /* deadlock recovery */
602         struct hammer_node_cache_list cache_list; /* passive caches */
603         int                     flags;
604         int                     loading;        /* load interlock */
605 };
606
607 #define HAMMER_NODE_DELETED     0x0001
608 #define HAMMER_NODE_FLUSH       0x0002
609 #define HAMMER_NODE_CRCGOOD     0x0004
610 #define HAMMER_NODE_NEEDSCRC    0x0008
611 #define HAMMER_NODE_NEEDSMIRROR 0x0010
612 #define HAMMER_NODE_CRCBAD      0x0020
613
614 #define HAMMER_NODE_CRCANY      (HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD)
615
616 typedef struct hammer_node      *hammer_node_t;
617
618 /*
619  * List of locked nodes.  This structure is used to lock potentially large
620  * numbers of nodes as an aid for complex B-Tree operations.
621  */
622 struct hammer_node_lock;
623 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock);
624
625 struct hammer_node_lock {
626         TAILQ_ENTRY(hammer_node_lock) entry;
627         struct hammer_node_lock_list  list;
628         struct hammer_node_lock       *parent;
629         hammer_node_t   node;
630         hammer_node_ondisk_t copy;      /* copy of on-disk data */
631         int             index;          /* index of this node in parent */
632         int             count;          /* count children */
633         int             flags;
634 };
635
636 typedef struct hammer_node_lock *hammer_node_lock_t;
637
638 #define HAMMER_NODE_LOCK_UPDATED        0x0001
639
640 /*
641  * Common I/O management structure - embedded in in-memory structures
642  * which are backed by filesystem buffers.
643  */
644 union hammer_io_structure {
645         struct hammer_io        io;
646         struct hammer_volume    volume;
647         struct hammer_buffer    buffer;
648 };
649
650 typedef union hammer_io_structure *hammer_io_structure_t;
651
652 /*
653  * The reserve structure prevents the blockmap from allocating
654  * out of a reserved bigblock.  Such reservations are used by
655  * the direct-write mechanism.
656  *
657  * The structure is also used to hold off on reallocations of
658  * big blocks from the freemap until flush dependancies have
659  * been dealt with.
660  */
661 struct hammer_reserve {
662         RB_ENTRY(hammer_reserve) rb_node;
663         TAILQ_ENTRY(hammer_reserve) delay_entry;
664         int             flush_group;
665         int             flags;
666         int             refs;
667         int             zone;
668         int             append_off;
669         hammer_off_t    zone_offset;
670 };
671
672 typedef struct hammer_reserve *hammer_reserve_t;
673
674 #define HAMMER_RESF_ONDELAY     0x0001
675 #define HAMMER_RESF_LAYER2FREE  0x0002
676
677 #include "hammer_cursor.h"
678
679 /*
680  * The undo structure tracks recent undos to avoid laying down duplicate
681  * undos within a flush group, saving us a significant amount of overhead.
682  *
683  * This is strictly a heuristic.
684  */
685 #define HAMMER_MAX_UNDOS                1024
686 #define HAMMER_MAX_FLUSHERS             4
687
688 struct hammer_undo {
689         RB_ENTRY(hammer_undo)   rb_node;
690         TAILQ_ENTRY(hammer_undo) lru_entry;
691         hammer_off_t            offset;
692         int                     bytes;
693 };
694
695 typedef struct hammer_undo *hammer_undo_t;
696
697 struct hammer_flusher_info;
698 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
699
700 struct hammer_flusher {
701         int             signal;         /* flusher thread sequencer */
702         int             act;            /* currently active flush group */
703         int             done;           /* set to act when complete */
704         int             next;           /* next flush group */
705         int             group_lock;     /* lock sequencing of the next flush */
706         int             exiting;        /* request master exit */
707         thread_t        td;             /* master flusher thread */
708         hammer_tid_t    tid;            /* last flushed transaction id */
709         int             finalize_want;          /* serialize finalization */
710         struct hammer_lock finalize_lock;       /* serialize finalization */
711         struct hammer_transaction trans;        /* shared transaction */
712         struct hammer_flusher_info_list run_list;
713         struct hammer_flusher_info_list ready_list;
714 };
715
716 /*
717  * Internal hammer mount data structure
718  */
719 struct hammer_mount {
720         struct mount *mp;
721         /*struct vnode *rootvp;*/
722         struct hammer_ino_rb_tree rb_inos_root;
723         struct hammer_vol_rb_tree rb_vols_root;
724         struct hammer_nod_rb_tree rb_nods_root;
725         struct hammer_und_rb_tree rb_undo_root;
726         struct hammer_res_rb_tree rb_resv_root;
727         struct hammer_buf_rb_tree rb_bufs_root;
728         struct hammer_pfs_rb_tree rb_pfsm_root;
729         struct hammer_volume *rootvol;
730         struct hammer_base_elm root_btree_beg;
731         struct hammer_base_elm root_btree_end;
732
733         struct malloc_type      *m_misc;
734         struct malloc_type      *m_inodes;
735
736         int     flags;          /* HAMMER_MOUNT_xxx flags */
737         int     hflags;
738         int     ronly;
739         int     nvolumes;
740         int     volume_iterator;
741         int     master_id;      /* -1 or 0-15 - clustering and mirroring */
742         int     version;        /* hammer filesystem version to use */
743         int     rsv_inodes;     /* reserved space due to dirty inodes */
744         int64_t rsv_databytes;  /* reserved space due to record data */
745         int     rsv_recs;       /* reserved space due to dirty records */
746         int     rsv_fromdelay;  /* bigblocks reserved due to flush delay */
747         int     undo_rec_limit; /* based on size of undo area */
748         int     last_newrecords;
749         int     count_newrecords;
750
751         int     inode_reclaims; /* inodes pending reclaim by flusher */
752         int     count_inodes;   /* total number of inodes */
753         int     count_iqueued;  /* inodes queued to flusher */
754
755         struct hammer_flusher flusher;
756
757         u_int   check_interrupt;
758         u_int   check_yield;
759         uuid_t  fsid;
760         struct hammer_io_list volu_list;        /* dirty undo buffers */
761         struct hammer_io_list undo_list;        /* dirty undo buffers */
762         struct hammer_io_list data_list;        /* dirty data buffers */
763         struct hammer_io_list alt_data_list;    /* dirty data buffers */
764         struct hammer_io_list meta_list;        /* dirty meta bufs    */
765         struct hammer_io_list lose_list;        /* loose buffers      */
766         int     locked_dirty_space;             /* meta/volu count    */
767         int     io_running_space;
768         int     objid_cache_count;
769         int     error;                          /* critical I/O error */
770         struct krate    krate;                  /* rate limited kprintf */
771         hammer_tid_t    asof;                   /* snapshot mount */
772         hammer_tid_t    next_tid;
773         hammer_tid_t    flush_tid1;             /* flusher tid sequencing */
774         hammer_tid_t    flush_tid2;             /* flusher tid sequencing */
775         int64_t copy_stat_freebigblocks;        /* number of free bigblocks */
776         u_int32_t       undo_seqno;             /* UNDO/REDO FIFO seqno */
777
778         struct netexport export;
779         struct hammer_lock sync_lock;
780         struct hammer_lock free_lock;
781         struct hammer_lock undo_lock;
782         struct hammer_lock blkmap_lock;
783         struct hammer_lock snapshot_lock;
784         struct hammer_blockmap  blockmap[HAMMER_MAX_ZONES];
785         struct hammer_undo      undos[HAMMER_MAX_UNDOS];
786         int                     undo_alloc;
787         TAILQ_HEAD(, hammer_undo)  undo_lru_list;
788         TAILQ_HEAD(, hammer_reserve) delay_list;
789         struct hammer_flush_group_list  flush_group_list;
790         hammer_flush_group_t    next_flush_group;
791         TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
792         TAILQ_HEAD(, hammer_reclaim) reclaim_list;
793 };
794
795 typedef struct hammer_mount     *hammer_mount_t;
796
797 #define HAMMER_MOUNT_CRITICAL_ERROR     0x0001
798 #define HAMMER_MOUNT_FLUSH_RECOVERY     0x0002
799
800 struct hammer_sync_info {
801         int error;
802         int waitfor;
803 };
804
805 #endif
806
807 /*
808  * checkspace slop (8MB chunks), higher numbers are more conservative.
809  */
810 #define HAMMER_CHKSPC_REBLOCK   25
811 #define HAMMER_CHKSPC_MIRROR    20
812 #define HAMMER_CHKSPC_WRITE     20
813 #define HAMMER_CHKSPC_CREATE    20
814 #define HAMMER_CHKSPC_REMOVE    10
815 #define HAMMER_CHKSPC_EMERGENCY 0
816
817 #if defined(_KERNEL)
818
819 extern struct vop_ops hammer_vnode_vops;
820 extern struct vop_ops hammer_spec_vops;
821 extern struct vop_ops hammer_fifo_vops;
822 extern struct bio_ops hammer_bioops;
823
824 extern int hammer_debug_io;
825 extern int hammer_debug_general;
826 extern int hammer_debug_debug;
827 extern int hammer_debug_inode;
828 extern int hammer_debug_locks;
829 extern int hammer_debug_btree;
830 extern int hammer_debug_tid;
831 extern int hammer_debug_recover;
832 extern int hammer_debug_recover_faults;
833 extern int hammer_cluster_enable;
834 extern int hammer_count_fsyncs;
835 extern int hammer_count_inodes;
836 extern int hammer_count_iqueued;
837 extern int hammer_count_reclaiming;
838 extern int hammer_count_records;
839 extern int hammer_count_record_datas;
840 extern int hammer_count_volumes;
841 extern int hammer_count_buffers;
842 extern int hammer_count_nodes;
843 extern int64_t hammer_count_extra_space_used;
844 extern int64_t hammer_stats_btree_lookups;
845 extern int64_t hammer_stats_btree_searches;
846 extern int64_t hammer_stats_btree_inserts;
847 extern int64_t hammer_stats_btree_deletes;
848 extern int64_t hammer_stats_btree_elements;
849 extern int64_t hammer_stats_btree_splits;
850 extern int64_t hammer_stats_btree_iterations;
851 extern int64_t hammer_stats_btree_root_iterations;
852 extern int64_t hammer_stats_record_iterations;
853 extern int64_t hammer_stats_file_read;
854 extern int64_t hammer_stats_file_write;
855 extern int64_t hammer_stats_file_iopsr;
856 extern int64_t hammer_stats_file_iopsw;
857 extern int64_t hammer_stats_disk_read;
858 extern int64_t hammer_stats_disk_write;
859 extern int64_t hammer_stats_inode_flushes;
860 extern int64_t hammer_stats_commits;
861 extern int64_t hammer_stats_undo;
862 extern int hammer_count_dirtybufspace;
863 extern int hammer_count_refedbufs;
864 extern int hammer_count_reservations;
865 extern int hammer_count_io_running_read;
866 extern int hammer_count_io_running_write;
867 extern int hammer_count_io_locked;
868 extern int hammer_limit_dirtybufspace;
869 extern int hammer_limit_recs;
870 extern int hammer_limit_inode_recs;
871 extern int hammer_bio_count;
872 extern int hammer_verify_zone;
873 extern int hammer_verify_data;
874 extern int hammer_write_mode;
875 extern int hammer_yield_check;
876 extern int hammer_fsync_mode;
877 extern int hammer_autoflush;
878 extern int64_t hammer_contention_count;
879
880 void    hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
881                         int error, const char *msg);
882 int     hammer_vop_inactive(struct vop_inactive_args *);
883 int     hammer_vop_reclaim(struct vop_reclaim_args *);
884 int     hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
885 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
886                         hammer_inode_t dip, int64_t obj_id,
887                         hammer_tid_t asof, u_int32_t localization,
888                         int flags, int *errorp);
889 struct hammer_inode *hammer_get_dummy_inode(hammer_transaction_t trans,
890                         hammer_inode_t dip, int64_t obj_id,
891                         hammer_tid_t asof, u_int32_t localization,
892                         int flags, int *errorp);
893 struct hammer_inode *hammer_find_inode(hammer_transaction_t trans,
894                         int64_t obj_id, hammer_tid_t asof,
895                         u_int32_t localization);
896 void    hammer_scan_inode_snapshots(hammer_mount_t hmp,
897                         hammer_inode_info_t iinfo,
898                         int (*callback)(hammer_inode_t ip, void *data),
899                         void *data);
900 void    hammer_put_inode(struct hammer_inode *ip);
901 void    hammer_put_inode_ref(struct hammer_inode *ip);
902 void    hammer_inode_waitreclaims(hammer_mount_t hmp);
903 void    hammer_inode_waithard(hammer_mount_t hmp);
904
905 int     hammer_unload_volume(hammer_volume_t volume, void *data __unused);
906 int     hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
907
908 int     hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused);
909 int     hammer_install_volume(hammer_mount_t hmp, const char *volname,
910                         struct vnode *devvp);
911 int     hammer_mountcheck_volumes(hammer_mount_t hmp);
912
913 int     hammer_mem_add(hammer_record_t record);
914 int     hammer_ip_lookup(hammer_cursor_t cursor);
915 int     hammer_ip_first(hammer_cursor_t cursor);
916 int     hammer_ip_next(hammer_cursor_t cursor);
917 int     hammer_ip_resolve_data(hammer_cursor_t cursor);
918 int     hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
919                         hammer_tid_t tid);
920 int     hammer_create_at_cursor(hammer_cursor_t cursor,
921                         hammer_btree_leaf_elm_t leaf, void *udata, int mode);
922 int     hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
923                         hammer_tid_t delete_tid, u_int32_t delete_ts,
924                         int track, int64_t *stat_bytes);
925 int     hammer_ip_check_directory_empty(hammer_transaction_t trans,
926                         hammer_inode_t ip);
927 int     hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
928 int     hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
929
930 hammer_record_t
931         hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
932 void    hammer_flush_record_done(hammer_record_t record, int error);
933 void    hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
934 void    hammer_rel_mem_record(hammer_record_t record);
935
936 int     hammer_cursor_up(hammer_cursor_t cursor);
937 int     hammer_cursor_up_locked(hammer_cursor_t cursor);
938 int     hammer_cursor_down(hammer_cursor_t cursor);
939 int     hammer_cursor_upgrade(hammer_cursor_t cursor);
940 int     hammer_cursor_upgrade_node(hammer_cursor_t cursor);
941 void    hammer_cursor_downgrade(hammer_cursor_t cursor);
942 int     hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
943                         int index);
944 void    hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
945 int     hammer_lock_ex_try(struct hammer_lock *lock);
946 void    hammer_lock_sh(struct hammer_lock *lock);
947 int     hammer_lock_sh_try(struct hammer_lock *lock);
948 int     hammer_lock_upgrade(struct hammer_lock *lock);
949 void    hammer_lock_downgrade(struct hammer_lock *lock);
950 int     hammer_lock_status(struct hammer_lock *lock);
951 void    hammer_unlock(struct hammer_lock *lock);
952 void    hammer_ref(struct hammer_lock *lock);
953 void    hammer_unref(struct hammer_lock *lock);
954
955 void    hammer_sync_lock_ex(hammer_transaction_t trans);
956 void    hammer_sync_lock_sh(hammer_transaction_t trans);
957 int     hammer_sync_lock_sh_try(hammer_transaction_t trans);
958 void    hammer_sync_unlock(hammer_transaction_t trans);
959
960 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
961 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
962 void    hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
963 u_int64_t hammer_timespec_to_time(struct timespec *ts);
964 int     hammer_str_to_tid(const char *str, int *ispfsp,
965                         hammer_tid_t *tidp, u_int32_t *localizationp);
966 int     hammer_is_atatext(const char *name, int len);
967 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip,
968                         int64_t namekey);
969 void hammer_clear_objid(hammer_inode_t dip);
970 void hammer_destroy_objid_cache(hammer_mount_t hmp);
971
972 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
973                         int bytes);
974 void hammer_clear_undo_history(hammer_mount_t hmp);
975 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
976 int hammer_get_dtype(u_int8_t obj_type);
977 u_int8_t hammer_get_obj_type(enum vtype vtype);
978 int64_t hammer_directory_namekey(hammer_inode_t dip, const void *name, int len,
979                         u_int32_t *max_iterationsp);
980 int     hammer_nohistory(hammer_inode_t ip);
981
982 int     hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
983                         hammer_node_cache_t cache, hammer_inode_t ip);
984 void    hammer_normalize_cursor(hammer_cursor_t cursor);
985 void    hammer_done_cursor(hammer_cursor_t cursor);
986 int     hammer_recover_cursor(hammer_cursor_t cursor);
987 void    hammer_unlock_cursor(hammer_cursor_t cursor);
988 int     hammer_lock_cursor(hammer_cursor_t cursor);
989 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor);
990 void    hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
991
992 void    hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
993 void    hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
994                         int index);
995 void    hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
996                         int index);
997 void    hammer_cursor_moved_element(hammer_node_t onode, hammer_node_t nnode,
998                         int oindex, int nindex);
999 void    hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent,
1000                         hammer_node_t nparent, int nindex);
1001 void    hammer_cursor_inserted_element(hammer_node_t node, int index);
1002 void    hammer_cursor_deleted_element(hammer_node_t node, int index);
1003
1004 int     hammer_btree_lookup(hammer_cursor_t cursor);
1005 int     hammer_btree_first(hammer_cursor_t cursor);
1006 int     hammer_btree_last(hammer_cursor_t cursor);
1007 int     hammer_btree_extract(hammer_cursor_t cursor, int flags);
1008 int     hammer_btree_iterate(hammer_cursor_t cursor);
1009 int     hammer_btree_iterate_reverse(hammer_cursor_t cursor);
1010 int     hammer_btree_insert(hammer_cursor_t cursor,
1011                             hammer_btree_leaf_elm_t elm, int *doprop);
1012 int     hammer_btree_delete(hammer_cursor_t cursor);
1013 void    hammer_btree_do_propagation(hammer_cursor_t cursor,
1014                             hammer_pseudofs_inmem_t pfsm,
1015                             hammer_btree_leaf_elm_t leaf);
1016 int     hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
1017 int     hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
1018 int     hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
1019 int     hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
1020
1021 int     btree_set_parent(hammer_transaction_t trans, hammer_node_t node,
1022                         hammer_btree_elm_t elm);
1023 void    hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node);
1024 int     hammer_btree_lock_children(hammer_cursor_t cursor, int depth,
1025                         hammer_node_lock_t parent);
1026 void    hammer_btree_lock_copy(hammer_cursor_t cursor,
1027                         hammer_node_lock_t parent);
1028 int     hammer_btree_sync_copy(hammer_cursor_t cursor,
1029                         hammer_node_lock_t parent);
1030 void    hammer_btree_unlock_children(hammer_cursor_t cursor,
1031                         hammer_node_lock_t parent);
1032 int     hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
1033 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans,
1034                         hammer_node_t node, int *parent_indexp,
1035                         int *errorp, int try_exclusive);
1036
1037 void    hammer_print_btree_node(hammer_node_ondisk_t ondisk);
1038 void    hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
1039
1040 void    *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
1041                         int *errorp, struct hammer_buffer **bufferp);
1042 void    *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
1043                         int *errorp, struct hammer_buffer **bufferp);
1044 void    *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1045                         int *errorp, struct hammer_buffer **bufferp);
1046 void    *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1047                         int *errorp, struct hammer_buffer **bufferp);
1048
1049 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
1050
1051 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
1052                         int32_t vol_no, int *errorp);
1053 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
1054                         int bytes, int isnew, int *errorp);
1055 void            hammer_sync_buffers(hammer_mount_t hmp,
1056                         hammer_off_t base_offset, int bytes);
1057 int             hammer_del_buffers(hammer_mount_t hmp,
1058                         hammer_off_t base_offset,
1059                         hammer_off_t zone2_offset, int bytes,
1060                         int report_conflicts);
1061
1062 int             hammer_ref_volume(hammer_volume_t volume);
1063 int             hammer_ref_buffer(hammer_buffer_t buffer);
1064 void            hammer_flush_buffer_nodes(hammer_buffer_t buffer);
1065
1066 void            hammer_rel_volume(hammer_volume_t volume, int flush);
1067 void            hammer_rel_buffer(hammer_buffer_t buffer, int flush);
1068
1069 int             hammer_vfs_export(struct mount *mp, int op,
1070                         const struct export_args *export);
1071 hammer_node_t   hammer_get_node(hammer_transaction_t trans,
1072                         hammer_off_t node_offset, int isnew, int *errorp);
1073 void            hammer_ref_node(hammer_node_t node);
1074 hammer_node_t   hammer_ref_node_safe(hammer_transaction_t trans,
1075                         hammer_node_cache_t cache, int *errorp);
1076 void            hammer_rel_node(hammer_node_t node);
1077 void            hammer_delete_node(hammer_transaction_t trans,
1078                         hammer_node_t node);
1079 void            hammer_cache_node(hammer_node_cache_t cache,
1080                         hammer_node_t node);
1081 void            hammer_uncache_node(hammer_node_cache_t cache);
1082 void            hammer_flush_node(hammer_node_t node);
1083
1084 void hammer_dup_buffer(struct hammer_buffer **bufferp,
1085                         struct hammer_buffer *buffer);
1086 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans,
1087                         hammer_off_t hint, int *errorp);
1088 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1089                         u_int16_t rec_type, hammer_off_t *data_offsetp,
1090                         struct hammer_buffer **data_bufferp,
1091                         hammer_off_t hint, int *errorp);
1092
1093 int hammer_generate_undo(hammer_transaction_t trans,
1094                         hammer_off_t zone1_offset, void *base, int len);
1095 int hammer_upgrade_undo_4(hammer_transaction_t trans);
1096
1097 void hammer_put_volume(struct hammer_volume *volume, int flush);
1098 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
1099
1100 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1101                         hammer_off_t owner, int *errorp);
1102 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1103                         hammer_off_t owner, int *errorp);
1104 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp);
1105 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1106                         int bytes, hammer_off_t hint, int *errorp);
1107 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1108                         int bytes, hammer_off_t *zone_offp, int *errorp);
1109 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1110                         hammer_reserve_t resv);
1111 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1112 void hammer_blockmap_free(hammer_transaction_t trans,
1113                         hammer_off_t bmap_off, int bytes);
1114 int hammer_blockmap_finalize(hammer_transaction_t trans,
1115                         hammer_reserve_t resv,
1116                         hammer_off_t bmap_off, int bytes);
1117 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
1118                         int *curp, int *errorp);
1119 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1120                         int *errorp);
1121 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1122                         int *errorp);
1123 int64_t hammer_undo_used(hammer_transaction_t trans);
1124 int64_t hammer_undo_space(hammer_transaction_t trans);
1125 int64_t hammer_undo_max(hammer_mount_t hmp);
1126 int hammer_undo_reclaim(hammer_io_t io);
1127
1128 void hammer_start_transaction(struct hammer_transaction *trans,
1129                               struct hammer_mount *hmp);
1130 void hammer_simple_transaction(struct hammer_transaction *trans,
1131                               struct hammer_mount *hmp);
1132 void hammer_start_transaction_fls(struct hammer_transaction *trans,
1133                                   struct hammer_mount *hmp);
1134 void hammer_done_transaction(struct hammer_transaction *trans);
1135 hammer_tid_t hammer_alloc_tid(hammer_mount_t hmp, int count);
1136
1137 void hammer_modify_inode(hammer_inode_t ip, int flags);
1138 void hammer_flush_inode(hammer_inode_t ip, int flags);
1139 void hammer_flush_inode_done(hammer_inode_t ip, int error);
1140 void hammer_wait_inode(hammer_inode_t ip);
1141
1142 int  hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
1143                         struct ucred *cred, struct hammer_inode *dip,
1144                         const char *name, int namelen,
1145                         hammer_pseudofs_inmem_t pfsm,
1146                         struct hammer_inode **ipp);
1147 void hammer_rel_inode(hammer_inode_t ip, int flush);
1148 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1149 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1150 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1151
1152 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1153 void hammer_test_inode(hammer_inode_t dip);
1154 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1155
1156 int  hammer_ip_add_directory(struct hammer_transaction *trans,
1157                         hammer_inode_t dip, const char *name, int bytes,
1158                         hammer_inode_t nip);
1159 int  hammer_ip_del_directory(struct hammer_transaction *trans,
1160                         hammer_cursor_t cursor, hammer_inode_t dip,
1161                         hammer_inode_t ip);
1162 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1163                         void *data, int bytes, int *errorp);
1164 int  hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
1165 int  hammer_ip_add_record(struct hammer_transaction *trans,
1166                         hammer_record_t record);
1167 int  hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1168                         int64_t ran_beg, int64_t ran_end, int truncating);
1169 int  hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1170                         int *countp);
1171 int  hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1172                         int64_t offset, void *data, int bytes);
1173 int  hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
1174 int  hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1175 hammer_pseudofs_inmem_t  hammer_load_pseudofs(hammer_transaction_t trans,
1176                         u_int32_t localization, int *errorp);
1177 int  hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1178                         hammer_pseudofs_inmem_t pfsm);
1179 int  hammer_save_pseudofs(hammer_transaction_t trans,
1180                         hammer_pseudofs_inmem_t pfsm);
1181 int  hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization);
1182 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1183 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1184                         struct ucred *cred);
1185
1186 void hammer_io_init(hammer_io_t io, hammer_volume_t volume,
1187                         enum hammer_io_type type);
1188 int hammer_io_read(struct vnode *devvp, struct hammer_io *io,
1189                         hammer_off_t limit);
1190 void hammer_io_advance(struct hammer_io *io);
1191 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
1192 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1193 struct buf *hammer_io_release(struct hammer_io *io, int flush);
1194 void hammer_io_flush(struct hammer_io *io, int reclaim);
1195 void hammer_io_wait(struct hammer_io *io);
1196 void hammer_io_waitdep(struct hammer_io *io);
1197 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident);
1198 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1199                         hammer_btree_leaf_elm_t leaf);
1200 int hammer_io_direct_write(hammer_mount_t hmp, hammer_record_t record,
1201                         struct bio *bio);
1202 void hammer_io_direct_wait(hammer_record_t record);
1203 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1204 void hammer_io_write_interlock(hammer_io_t io);
1205 void hammer_io_done_interlock(hammer_io_t io);
1206 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1207 void hammer_io_clear_modlist(struct hammer_io *io);
1208 void hammer_io_flush_sync(hammer_mount_t hmp);
1209
1210 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1211                         void *base, int len);
1212 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1213                         void *base, int len);
1214 void hammer_modify_volume_done(hammer_volume_t volume);
1215 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1216
1217 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1218                         struct hammer_ioc_reblock *reblock);
1219 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip,
1220                         struct hammer_ioc_rebalance *rebal);
1221 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1222                         struct hammer_ioc_prune *prune);
1223 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1224                         struct hammer_ioc_mirror_rw *mirror);
1225 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1226                         struct hammer_ioc_mirror_rw *mirror);
1227 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1228                         struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1229 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1230                         struct hammer_ioc_pseudofs_rw *pfs);
1231 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1232                         struct hammer_ioc_pseudofs_rw *pfs);
1233 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1234                         struct hammer_ioc_pseudofs_rw *pfs);
1235 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1236                         struct hammer_ioc_pseudofs_rw *pfs);
1237 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1238                         struct hammer_ioc_pseudofs_rw *pfs);
1239 int hammer_ioc_expand(hammer_transaction_t trans, hammer_inode_t ip,
1240                         struct hammer_ioc_expand *expand);
1241
1242 int hammer_signal_check(hammer_mount_t hmp);
1243
1244 void hammer_flusher_create(hammer_mount_t hmp);
1245 void hammer_flusher_destroy(hammer_mount_t hmp);
1246 void hammer_flusher_sync(hammer_mount_t hmp);
1247 int  hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1248 int  hammer_flusher_async_one(hammer_mount_t hmp);
1249 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1250 void hammer_flusher_wait_next(hammer_mount_t hmp);
1251 int  hammer_flusher_meta_limit(hammer_mount_t hmp);
1252 int  hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1253 int  hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1254 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1255 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1256 int  hammer_flusher_haswork(hammer_mount_t hmp);
1257
1258
1259 int hammer_recover_stage1(hammer_mount_t hmp, hammer_volume_t rootvol);
1260 int hammer_recover_stage2(hammer_mount_t hmp, hammer_volume_t rootvol);
1261 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1262                         hammer_volume_t root_volume, int final);
1263
1264 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1265 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1266 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1267
1268 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1269 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1270 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1271 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1272 void hkprintf(const char *ctl, ...);
1273 udev_t hammer_fsid_to_udev(uuid_t *uuid);
1274
1275
1276 int hammer_blocksize(int64_t file_offset);
1277 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1278
1279 /*
1280  * Shortcut for _hammer_checkspace(), used all over the code.
1281  */
1282 static __inline int
1283 hammer_checkspace(hammer_mount_t hmp, int slop)
1284 {
1285         return(_hammer_checkspace(hmp, slop, NULL));
1286 }
1287
1288 #endif
1289
1290 static __inline void
1291 hammer_wait_mem_record(hammer_record_t record)
1292 {
1293         hammer_wait_mem_record_ident(record, "hmmwai");
1294 }
1295
1296 static __inline void
1297 hammer_lock_ex(struct hammer_lock *lock)
1298 {
1299         hammer_lock_ex_ident(lock, "hmrlck");
1300 }
1301
1302 /*
1303  * Indicate that a B-Tree node is being modified.
1304  */
1305 static __inline void
1306 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1307 {
1308         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1309         hammer_modify_buffer(trans, node->buffer, NULL, 0);
1310 }
1311
1312 static __inline void
1313 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1314 {
1315         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1316         hammer_modify_buffer(trans, node->buffer,
1317                              node->ondisk, sizeof(*node->ondisk));
1318 }
1319
1320 static __inline void
1321 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1322                    void *base, int len)
1323 {
1324         hammer_crc_t *crcptr;
1325
1326         KKASSERT((char *)base >= (char *)node->ondisk &&
1327                  (char *)base + len <=
1328                     (char *)node->ondisk + sizeof(*node->ondisk));
1329         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1330         hammer_modify_buffer(trans, node->buffer, base, len);
1331         crcptr = &node->ondisk->crc;
1332         hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t));
1333         --node->buffer->io.modify_refs; /* only want one ref */
1334 }
1335
1336 /*
1337  * Indicate that the specified modifications have been completed.
1338  *
1339  * Do not try to generate the crc here, it's very expensive to do and a
1340  * sequence of insertions or deletions can result in many calls to this
1341  * function on the same node.
1342  */
1343 static __inline void
1344 hammer_modify_node_done(hammer_node_t node)
1345 {
1346         node->flags |= HAMMER_NODE_CRCGOOD;
1347         if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1348                 node->flags |= HAMMER_NODE_NEEDSCRC;
1349                 node->buffer->io.gencrc = 1;
1350                 hammer_ref_node(node);
1351         }
1352         hammer_modify_buffer_done(node->buffer);
1353 }
1354
1355 #define hammer_modify_volume_field(trans, vol, field)           \
1356         hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1357                              sizeof((vol)->ondisk->field))
1358
1359 #define hammer_modify_node_field(trans, node, field)            \
1360         hammer_modify_node(trans, node, &(node)->ondisk->field, \
1361                              sizeof((node)->ondisk->field))
1362
1363 /*
1364  * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1365  * created directories for HAMMER version 2 or greater and causes
1366  * directory entries to be placed the inode localization zone in
1367  * the B-Tree instead of the misc zone.
1368  *
1369  * This greatly improves localization between directory entries and
1370  * inodes
1371  */
1372 static __inline u_int32_t
1373 hammer_dir_localization(hammer_inode_t dip)
1374 {
1375         if (dip->ino_data.cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO)
1376                 return(HAMMER_LOCALIZE_INODE);
1377         else
1378                 return(HAMMER_LOCALIZE_MISC);
1379 }