Pre-2.0 release: Sync with HAMMER 64 - simplify PFS operations, fix pfs-upgrade
[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 1.117.2.4 2008/07/19 18:46:20 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/stat.h>
53 #include <sys/globaldata.h>
54 #include <sys/lockf.h>
55 #include <sys/buf.h>
56 #include <sys/queue.h>
57 #include <sys/ktr.h>
58 #include <sys/globaldata.h>
59
60 #include <sys/buf2.h>
61 #include <sys/signal2.h>
62 #include "hammer_disk.h"
63 #include "hammer_mount.h"
64 #include "hammer_ioctl.h"
65
66 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
67
68 MALLOC_DECLARE(M_HAMMER);
69
70 /*
71  * Kernel trace
72  */
73 #if !defined(KTR_HAMMER)
74 #define KTR_HAMMER      KTR_ALL
75 #endif
76 KTR_INFO_MASTER_EXTERN(hammer);
77
78 /*
79  * Misc structures
80  */
81 struct hammer_mount;
82
83 /*
84  * Key structure used for custom RB tree inode lookups.  This prototypes
85  * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
86  */
87 typedef struct hammer_inode_info {
88         int64_t         obj_id;         /* (key) object identifier */
89         hammer_tid_t    obj_asof;       /* (key) snapshot transid or 0 */
90         u_int32_t       obj_localization; /* (key) pseudo-fs */
91         union {
92                 struct hammer_btree_leaf_elm *leaf;
93         } u;
94 } *hammer_inode_info_t;
95
96 typedef enum hammer_transaction_type {
97         HAMMER_TRANS_RO,
98         HAMMER_TRANS_STD,
99         HAMMER_TRANS_FLS
100 } hammer_transaction_type_t;
101
102 /*
103  * HAMMER Transaction tracking
104  */
105 struct hammer_transaction {
106         hammer_transaction_type_t type;
107         struct hammer_mount *hmp;
108         hammer_tid_t    tid;
109         u_int64_t       time;
110         u_int32_t       time32;
111         int             sync_lock_refs;
112         struct hammer_volume *rootvol;
113 };
114
115 typedef struct hammer_transaction *hammer_transaction_t;
116
117 /*
118  * HAMMER locks
119  */
120 struct hammer_lock {
121         int     refs;           /* active references delay writes */
122         int     lockcount;      /* lock count for exclusive/shared access */
123         int     wanted;
124         int     exwanted;       /* number of threads waiting for ex lock */
125         struct thread *locktd;
126 };
127
128 static __inline int
129 hammer_islocked(struct hammer_lock *lock)
130 {
131         return(lock->lockcount != 0);
132 }
133
134 static __inline int
135 hammer_isactive(struct hammer_lock *lock)
136 {
137         return(lock->refs != 0);
138 }
139
140 static __inline int
141 hammer_islastref(struct hammer_lock *lock)
142 {
143         return(lock->refs == 1);
144 }
145
146 /*
147  * Return if we specifically own the lock exclusively.
148  */
149 static __inline int
150 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
151 {
152         if (lock->lockcount > 0 && lock->locktd == td)
153                 return(1);
154         return(0);
155 }
156
157 /*
158  * Flush state, used by various structures
159  */
160 typedef enum hammer_inode_state {
161         HAMMER_FST_IDLE,
162         HAMMER_FST_SETUP,
163         HAMMER_FST_FLUSH
164 } hammer_inode_state_t;
165
166 TAILQ_HEAD(hammer_record_list, hammer_record);
167
168 /*
169  * Pseudo-filesystem extended data tracking
170  */
171 struct hammer_pfs_rb_tree;
172 struct hammer_pseudofs_inmem;
173 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
174 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
175               hammer_pfs_rb_compare, u_int32_t);
176
177 struct hammer_pseudofs_inmem {
178         RB_ENTRY(hammer_pseudofs_inmem) rb_node;
179         struct hammer_lock      lock;
180         u_int32_t               localization;
181         hammer_tid_t            create_tid;
182         int                     flags;
183         udev_t                  fsid_udev;
184         struct hammer_pseudofs_data pfsd;
185 };
186
187 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t;
188
189 #define HAMMER_PFSM_DELETED     0x0001
190
191 /*
192  * Cache object ids.  A fixed number of objid cache structures are
193  * created to reserve object id's for newly created files in multiples
194  * of 100,000, localized to a particular directory, and recycled as
195  * needed.  This allows parallel create operations in different
196  * directories to retain fairly localized object ids which in turn
197  * improves reblocking performance and layout.
198  */
199 #define OBJID_CACHE_SIZE        1024
200 #define OBJID_CACHE_BULK        100000
201
202 typedef struct hammer_objid_cache {
203         TAILQ_ENTRY(hammer_objid_cache) entry;
204         struct hammer_inode             *dip;
205         hammer_tid_t                    next_tid;
206         int                             count;
207 } *hammer_objid_cache_t;
208
209 /*
210  * Associate an inode with a B-Tree node to cache search start positions
211  */
212 typedef struct hammer_node_cache {
213         TAILQ_ENTRY(hammer_node_cache)  entry;
214         struct hammer_node              *node;
215         struct hammer_inode             *ip;
216 } *hammer_node_cache_t;
217
218 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
219
220 /*
221  * Structure used to organize flush groups.  Flush groups must be
222  * organized into chunks in order to avoid blowing out the UNDO FIFO.
223  * Without this a 'sync' could end up flushing 50,000 inodes in a single
224  * transaction.
225  */
226 struct hammer_flush_group {
227         TAILQ_ENTRY(hammer_flush_group) flush_entry;
228         TAILQ_HEAD(, hammer_inode)      flush_list;
229         int                             inode_count;    /* inode load */
230         int                             total_count;    /* record load */
231         int                             running;        /* group is running */
232         int                             closed;
233         int                             refs;
234 };
235
236 typedef struct hammer_flush_group *hammer_flush_group_t;
237
238 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
239
240 /*
241  * Structure used to represent an inode in-memory.
242  *
243  * The record and data associated with an inode may be out of sync with
244  * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
245  * clear).
246  *
247  * An inode may also hold a cache of unsynchronized records, used for
248  * database and directories only.  Unsynchronized regular file data is
249  * stored in the buffer cache.
250  *
251  * NOTE: A file which is created and destroyed within the initial
252  * synchronization period can wind up not doing any disk I/O at all.
253  *
254  * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
255  */
256 struct hammer_ino_rb_tree;
257 struct hammer_inode;
258 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
259 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
260               hammer_ino_rb_compare, hammer_inode_info_t);
261
262 struct hammer_rec_rb_tree;
263 struct hammer_record;
264 RB_HEAD(hammer_rec_rb_tree, hammer_record);
265 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
266               hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
267
268 TAILQ_HEAD(hammer_node_list, hammer_node);
269
270 struct hammer_inode {
271         RB_ENTRY(hammer_inode)  rb_node;
272         hammer_inode_state_t    flush_state;
273         hammer_flush_group_t    flush_group;
274         TAILQ_ENTRY(hammer_inode) flush_entry;
275         struct hammer_record_list target_list;  /* target of dependant recs */
276         int64_t                 obj_id;         /* (key) object identifier */
277         hammer_tid_t            obj_asof;       /* (key) snapshot or 0 */
278         u_int32_t               obj_localization; /* (key) pseudo-fs */
279         struct hammer_mount     *hmp;
280         hammer_objid_cache_t    objid_cache;
281         int                     flags;
282         int                     error;          /* flush error */
283         int                     cursor_ip_refs; /* sanity */
284         int                     rsv_recs;
285         struct vnode            *vp;
286         hammer_pseudofs_inmem_t pfsm;
287         struct lockf            advlock;
288         struct hammer_lock      lock;           /* sync copy interlock */
289         off_t                   trunc_off;
290         struct hammer_btree_leaf_elm ino_leaf;  /* in-memory cache */
291         struct hammer_inode_data ino_data;      /* in-memory cache */
292         struct hammer_rec_rb_tree rec_tree;     /* in-memory cache */
293         struct hammer_node_cache cache[2];      /* search initiate cache */
294
295         /*
296          * When a demark is created to synchronize an inode to
297          * disk, certain fields are copied so the front-end VOPs
298          * can continue to run in parallel with the synchronization
299          * occuring in the background.
300          */
301         int             sync_flags;             /* to-sync flags cache */
302         off_t           sync_trunc_off;         /* to-sync truncation */
303         off_t           save_trunc_off;         /* write optimization */
304         struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
305         struct hammer_inode_data sync_ino_data; /* to-sync cache */
306 };
307
308 typedef struct hammer_inode *hammer_inode_t;
309
310 #define VTOI(vp)        ((struct hammer_inode *)(vp)->v_data)
311
312 #define HAMMER_INODE_DDIRTY     0x0001  /* in-memory ino_data is dirty */
313                                         /* (not including atime/mtime) */
314 #define HAMMER_INODE_RSV_INODES 0x0002  /* hmp->rsv_inodes bumped */
315 #define HAMMER_INODE_CONN_DOWN  0x0004  /* include in downward recursion */
316 #define HAMMER_INODE_XDIRTY     0x0008  /* in-memory records */
317 #define HAMMER_INODE_ONDISK     0x0010  /* inode is on-disk (else not yet) */
318 #define HAMMER_INODE_FLUSH      0x0020  /* flush on last ref */
319 #define HAMMER_INODE_DELETED    0x0080  /* inode delete (backend) */
320 #define HAMMER_INODE_DELONDISK  0x0100  /* delete synchronized to disk */
321 #define HAMMER_INODE_RO         0x0200  /* read-only (because of as-of) */
322 #define HAMMER_INODE_VHELD      0x0400  /* vnode held on sync */
323 #define HAMMER_INODE_DONDISK    0x0800  /* data records may be on disk */
324 #define HAMMER_INODE_BUFS       0x1000  /* dirty high level bps present */
325 #define HAMMER_INODE_REFLUSH    0x2000  /* pipelined flush during flush */
326 #define HAMMER_INODE_RECLAIM    0x4000  /* trying to reclaim */
327 #define HAMMER_INODE_FLUSHW     0x8000  /* Someone waiting for flush */
328
329 #define HAMMER_INODE_TRUNCATED  0x00010000
330 #define HAMMER_INODE_DELETING   0x00020000 /* inode delete request (frontend)*/
331 #define HAMMER_INODE_RESIGNAL   0x00040000 /* re-signal on re-flush */
332 #define HAMMER_INODE_ATIME      0x00100000 /* in-memory atime modified */
333 #define HAMMER_INODE_MTIME      0x00200000 /* in-memory mtime modified */
334 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
335
336 #define HAMMER_INODE_MODMASK    (HAMMER_INODE_DDIRTY|                       \
337                                  HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS|     \
338                                  HAMMER_INODE_ATIME|HAMMER_INODE_MTIME|     \
339                                  HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
340
341 #define HAMMER_INODE_MODMASK_NOXDIRTY \
342                                 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
343
344 #define HAMMER_FLUSH_GROUP_SIZE 64
345
346 #define HAMMER_FLUSH_SIGNAL     0x0001
347 #define HAMMER_FLUSH_RECURSION  0x0002
348
349 /*
350  * Used by the inode reclaim code to pipeline reclaims and avoid
351  * blowing out kernel memory or letting the flusher get too far
352  * behind.
353  */
354 struct hammer_reclaim {
355         TAILQ_ENTRY(hammer_reclaim) entry;
356         int     okydoky;
357 };
358
359 #define HAMMER_RECLAIM_FLUSH    2000
360 #define HAMMER_RECLAIM_WAIT     4000
361
362 /*
363  * Structure used to represent an unsynchronized record in-memory.  These
364  * records typically represent directory entries.  Only non-historical
365  * records are kept in-memory.
366  *
367  * Records are organized as a per-inode RB-Tree.  If the inode is not
368  * on disk then neither are any records and the in-memory record tree
369  * represents the entire contents of the inode.  If the inode is on disk
370  * then the on-disk B-Tree is scanned in parallel with the in-memory
371  * RB-Tree to synthesize the current state of the file.
372  *
373  * Records are also used to enforce the ordering of directory create/delete
374  * operations.  A new inode will not be flushed to disk unless its related
375  * directory entry is also being flushed at the same time.  A directory entry
376  * will not be removed unless its related inode is also being removed at the
377  * same time.
378  */
379 typedef enum hammer_record_type {
380         HAMMER_MEM_RECORD_GENERAL,      /* misc record */
381         HAMMER_MEM_RECORD_INODE,        /* inode record */
382         HAMMER_MEM_RECORD_ADD,          /* positive memory cache record */
383         HAMMER_MEM_RECORD_DEL,          /* negative delete-on-disk record */
384         HAMMER_MEM_RECORD_DATA          /* bulk-data record w/on-disk ref */
385 } hammer_record_type_t;
386
387 struct hammer_record {
388         RB_ENTRY(hammer_record)         rb_node;
389         TAILQ_ENTRY(hammer_record)      target_entry;
390         hammer_inode_state_t            flush_state;
391         hammer_flush_group_t            flush_group;
392         hammer_record_type_t            type;
393         struct hammer_lock              lock;
394         struct hammer_reserve           *resv;
395         struct hammer_inode             *ip;
396         struct hammer_inode             *target_ip;
397         struct hammer_btree_leaf_elm    leaf;
398         union hammer_data_ondisk        *data;
399         int                             flags;
400 };
401
402 typedef struct hammer_record *hammer_record_t;
403
404 /*
405  * Record flags.  Note that FE can only be set by the frontend if the
406  * record has not been interlocked by the backend w/ BE.
407  */
408 #define HAMMER_RECF_ALLOCDATA           0x0001
409 #define HAMMER_RECF_ONRBTREE            0x0002
410 #define HAMMER_RECF_DELETED_FE          0x0004  /* deleted (frontend) */
411 #define HAMMER_RECF_DELETED_BE          0x0008  /* deleted (backend) */
412 #define HAMMER_RECF_COMMITTED           0x0010  /* committed to the B-Tree */
413 #define HAMMER_RECF_INTERLOCK_BE        0x0020  /* backend interlock */
414 #define HAMMER_RECF_WANTED              0x0040  /* wanted by the frontend */
415 #define HAMMER_RECF_CONVERT_DELETE      0x0100  /* special case */
416 #define HAMMER_RECF_DIRECT_IO           0x0200  /* related direct I/O running*/
417 #define HAMMER_RECF_DIRECT_WAIT         0x0400  /* related direct I/O running*/
418
419 /*
420  * hammer_delete_at_cursor() flags
421  */
422 #define HAMMER_DELETE_ADJUST            0x0001
423 #define HAMMER_DELETE_DESTROY           0x0002
424
425 /*
426  * In-memory structures representing on-disk structures.
427  */
428 struct hammer_volume;
429 struct hammer_buffer;
430 struct hammer_node;
431 struct hammer_undo;
432 struct hammer_reserve;
433
434 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
435 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
436 RB_HEAD(hammer_nod_rb_tree, hammer_node);
437 RB_HEAD(hammer_und_rb_tree, hammer_undo);
438 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
439
440 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
441               hammer_vol_rb_compare, int32_t);
442 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
443               hammer_buf_rb_compare, hammer_off_t);
444 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
445               hammer_nod_rb_compare, hammer_off_t);
446 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
447               hammer_und_rb_compare, hammer_off_t);
448 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
449               hammer_res_rb_compare, hammer_off_t);
450
451 /*
452  * IO management - embedded at the head of various in-memory structures
453  *
454  * VOLUME       - hammer_volume containing meta-data
455  * META_BUFFER  - hammer_buffer containing meta-data
456  * DATA_BUFFER  - hammer_buffer containing pure-data
457  *
458  * Dirty volume headers and dirty meta-data buffers are locked until the
459  * flusher can sequence them out.  Dirty pure-data buffers can be written.
460  * Clean buffers can be passively released.
461  */
462 typedef enum hammer_io_type {
463         HAMMER_STRUCTURE_VOLUME,
464         HAMMER_STRUCTURE_META_BUFFER,
465         HAMMER_STRUCTURE_UNDO_BUFFER,
466         HAMMER_STRUCTURE_DATA_BUFFER
467 } hammer_io_type_t;
468
469 union hammer_io_structure;
470 struct hammer_io;
471
472 struct worklist {
473         LIST_ENTRY(worklist) node;
474 };
475
476 TAILQ_HEAD(hammer_io_list, hammer_io);
477 typedef struct hammer_io_list *hammer_io_list_t;
478
479 struct hammer_io {
480         struct worklist         worklist;
481         struct hammer_lock      lock;
482         enum hammer_io_type     type;
483         struct hammer_mount     *hmp;
484         TAILQ_ENTRY(hammer_io)  mod_entry; /* list entry if modified */
485         hammer_io_list_t        mod_list;
486         struct buf              *bp;
487         int64_t                 offset;    /* zone-2 offset */
488         int                     bytes;     /* buffer cache buffer size */
489         int                     loading;   /* loading/unloading interlock */
490         int                     modify_refs;
491
492         u_int           modified : 1;   /* bp's data was modified */
493         u_int           released : 1;   /* bp released (w/ B_LOCKED set) */
494         u_int           running : 1;    /* bp write IO in progress */
495         u_int           waiting : 1;    /* someone is waiting on us */
496         u_int           validated : 1;  /* ondisk has been validated */
497         u_int           waitdep : 1;    /* flush waits for dependancies */
498         u_int           recovered : 1;  /* has recovery ref */
499         u_int           waitmod : 1;    /* waiting for modify_refs */
500         u_int           reclaim : 1;    /* reclaim requested */
501         u_int           gencrc : 1;     /* crc needs to be generated */
502         u_int           ioerror : 1;    /* abort on io-error */
503 };
504
505 typedef struct hammer_io *hammer_io_t;
506
507 #define HAMMER_CLUSTER_SIZE     (64 * 1024)
508 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
509 #undef  HAMMER_CLUSTER_SIZE
510 #define HAMMER_CLUSTER_SIZE     MAXBSIZE
511 #endif
512 #define HAMMER_CLUSTER_BUFS     (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
513
514 /*
515  * In-memory volume representing on-disk buffer
516  */
517 struct hammer_volume {
518         struct hammer_io io;
519         RB_ENTRY(hammer_volume) rb_node;
520         struct hammer_volume_ondisk *ondisk;
521         int32_t vol_no;
522         int64_t nblocks;        /* note: special calculation for statfs */
523         int64_t buffer_base;    /* base offset of buffer 0 */
524         hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
525         hammer_off_t maxraw_off; /* Maximum raw offset for device */
526         char    *vol_name;
527         struct vnode *devvp;
528         int     vol_flags;
529 };
530
531 typedef struct hammer_volume *hammer_volume_t;
532
533 /*
534  * In-memory buffer (other then volume, super-cluster, or cluster),
535  * representing an on-disk buffer.
536  */
537 struct hammer_buffer {
538         struct hammer_io io;
539         RB_ENTRY(hammer_buffer) rb_node;
540         void *ondisk;
541         struct hammer_volume *volume;
542         hammer_off_t zoneX_offset;
543         hammer_off_t zone2_offset;
544         struct hammer_reserve *resv;
545         struct hammer_node_list clist;
546 };
547
548 typedef struct hammer_buffer *hammer_buffer_t;
549
550 /*
551  * In-memory B-Tree node, representing an on-disk B-Tree node.
552  *
553  * This is a hang-on structure which is backed by a hammer_buffer,
554  * indexed by a hammer_cluster, and used for fine-grained locking of
555  * B-Tree nodes in order to properly control lock ordering.  A hammer_buffer
556  * can contain multiple nodes representing wildly disassociated portions
557  * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
558  *
559  * This structure uses a cluster-relative index to reduce the number
560  * of layers required to access it, and also because all on-disk B-Tree
561  * references are cluster-relative offsets.
562  */
563 struct hammer_node {
564         struct hammer_lock      lock;           /* node-by-node lock */
565         TAILQ_ENTRY(hammer_node) entry;         /* per-buffer linkage */
566         RB_ENTRY(hammer_node)   rb_node;        /* per-cluster linkage */
567         hammer_off_t            node_offset;    /* full offset spec */
568         struct hammer_mount     *hmp;
569         struct hammer_buffer    *buffer;        /* backing buffer */
570         hammer_node_ondisk_t    ondisk;         /* ptr to on-disk structure */
571         TAILQ_HEAD(, hammer_cursor) cursor_list;  /* deadlock recovery */
572         struct hammer_node_cache_list cache_list; /* passive caches */
573         int                     flags;
574         int                     loading;        /* load interlock */
575 };
576
577 #define HAMMER_NODE_DELETED     0x0001
578 #define HAMMER_NODE_FLUSH       0x0002
579 #define HAMMER_NODE_CRCGOOD     0x0004
580 #define HAMMER_NODE_NEEDSCRC    0x0008
581 #define HAMMER_NODE_NEEDSMIRROR 0x0010
582
583 typedef struct hammer_node      *hammer_node_t;
584
585 /*
586  * List of locked nodes.
587  */
588 struct hammer_node_locklist {
589         struct hammer_node_locklist *next;
590         hammer_node_t   node;
591 };
592
593 typedef struct hammer_node_locklist *hammer_node_locklist_t;
594
595
596 /*
597  * Common I/O management structure - embedded in in-memory structures
598  * which are backed by filesystem buffers.
599  */
600 union hammer_io_structure {
601         struct hammer_io        io;
602         struct hammer_volume    volume;
603         struct hammer_buffer    buffer;
604 };
605
606 typedef union hammer_io_structure *hammer_io_structure_t;
607
608 /*
609  * The reserve structure prevents the blockmap from allocating
610  * out of a reserved bigblock.  Such reservations are used by
611  * the direct-write mechanism.
612  *
613  * The structure is also used to hold off on reallocations of
614  * big blocks from the freemap until flush dependancies have
615  * been dealt with.
616  */
617 struct hammer_reserve {
618         RB_ENTRY(hammer_reserve) rb_node;
619         TAILQ_ENTRY(hammer_reserve) delay_entry;
620         int             flush_group;
621         int             flags;
622         int             refs;
623         int             zone;
624         int             bytes_freed;
625         int             append_off;
626         hammer_off_t    zone_offset;
627 };
628
629 typedef struct hammer_reserve *hammer_reserve_t;
630
631 #define HAMMER_RESF_ONDELAY     0x0001
632
633 #include "hammer_cursor.h"
634
635 /*
636  * The undo structure tracks recent undos to avoid laying down duplicate
637  * undos within a flush group, saving us a significant amount of overhead.
638  *
639  * This is strictly a heuristic.
640  */
641 #define HAMMER_MAX_UNDOS                1024
642 #define HAMMER_MAX_FLUSHERS             4
643
644 struct hammer_undo {
645         RB_ENTRY(hammer_undo)   rb_node;
646         TAILQ_ENTRY(hammer_undo) lru_entry;
647         hammer_off_t            offset;
648         int                     bytes;
649 };
650
651 typedef struct hammer_undo *hammer_undo_t;
652
653 struct hammer_flusher_info;
654 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
655
656 struct hammer_flusher {
657         int             signal;         /* flusher thread sequencer */
658         int             act;            /* currently active flush group */
659         int             done;           /* set to act when complete */
660         int             next;           /* next flush group */
661         int             group_lock;     /* lock sequencing of the next flush */
662         int             exiting;        /* request master exit */
663         thread_t        td;             /* master flusher thread */
664         hammer_tid_t    tid;            /* last flushed transaction id */
665         int             finalize_want;          /* serialize finalization */
666         struct hammer_lock finalize_lock;       /* serialize finalization */
667         struct hammer_transaction trans;        /* shared transaction */
668         struct hammer_flusher_info_list run_list;
669         struct hammer_flusher_info_list ready_list;
670 };
671
672 /*
673  * Internal hammer mount data structure
674  */
675 struct hammer_mount {
676         struct mount *mp;
677         /*struct vnode *rootvp;*/
678         struct hammer_ino_rb_tree rb_inos_root;
679         struct hammer_vol_rb_tree rb_vols_root;
680         struct hammer_nod_rb_tree rb_nods_root;
681         struct hammer_und_rb_tree rb_undo_root;
682         struct hammer_res_rb_tree rb_resv_root;
683         struct hammer_buf_rb_tree rb_bufs_root;
684         struct hammer_pfs_rb_tree rb_pfsm_root;
685         struct hammer_volume *rootvol;
686         struct hammer_base_elm root_btree_beg;
687         struct hammer_base_elm root_btree_end;
688         int     flags;          /* HAMMER_MOUNT_xxx flags */
689         int     hflags;
690         int     ronly;
691         int     nvolumes;
692         int     volume_iterator;
693         int     master_id;      /* -1 or 0-15 - clustering and mirroring */
694         int     rsv_inodes;     /* reserved space due to dirty inodes */
695         int64_t rsv_databytes;  /* reserved space due to record data */
696         int     rsv_recs;       /* reserved space due to dirty records */
697         int     rsv_fromdelay;  /* bigblocks reserved due to flush delay */
698         int     undo_rec_limit; /* based on size of undo area */
699         int     last_newrecords;
700         int     count_newrecords;
701
702         int     inode_reclaims; /* inodes pending reclaim by flusher */
703         int     count_inodes;   /* total number of inodes */
704         int     count_iqueued;  /* inodes queued to flusher */
705
706         struct hammer_flusher flusher;
707
708         u_int   check_interrupt;
709         uuid_t  fsid;
710         struct hammer_io_list volu_list;        /* dirty undo buffers */
711         struct hammer_io_list undo_list;        /* dirty undo buffers */
712         struct hammer_io_list data_list;        /* dirty data buffers */
713         struct hammer_io_list alt_data_list;    /* dirty data buffers */
714         struct hammer_io_list meta_list;        /* dirty meta bufs    */
715         struct hammer_io_list lose_list;        /* loose buffers      */
716         int     locked_dirty_space;             /* meta/volu count    */
717         int     io_running_space;
718         int     objid_cache_count;
719         int     error;                          /* critical I/O error */
720         struct krate    krate;                  /* rate limited kprintf */
721         hammer_tid_t    asof;                   /* snapshot mount */
722         hammer_off_t    next_tid;
723         int64_t copy_stat_freebigblocks;        /* number of free bigblocks */
724
725         u_int32_t namekey_iterator;
726         struct netexport export;
727         struct hammer_lock sync_lock;
728         struct hammer_lock free_lock;
729         struct hammer_lock undo_lock;
730         struct hammer_lock blkmap_lock;
731         struct hammer_blockmap  blockmap[HAMMER_MAX_ZONES];
732         struct hammer_undo      undos[HAMMER_MAX_UNDOS];
733         int                     undo_alloc;
734         TAILQ_HEAD(, hammer_undo)  undo_lru_list;
735         TAILQ_HEAD(, hammer_reserve) delay_list;
736         struct hammer_flush_group_list  flush_group_list;
737         hammer_flush_group_t    next_flush_group;
738         TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
739         TAILQ_HEAD(, hammer_reclaim) reclaim_list;
740 };
741
742 typedef struct hammer_mount     *hammer_mount_t;
743
744 #define HAMMER_MOUNT_CRITICAL_ERROR     0x0001
745
746 struct hammer_sync_info {
747         int error;
748         int waitfor;
749 };
750
751 #endif
752
753 /*
754  * checkspace slop (8MB chunks), higher numbers are more conservative.
755  */
756 #define HAMMER_CHKSPC_REBLOCK   25
757 #define HAMMER_CHKSPC_MIRROR    20
758 #define HAMMER_CHKSPC_WRITE     20
759 #define HAMMER_CHKSPC_CREATE    20
760 #define HAMMER_CHKSPC_REMOVE    10
761 #define HAMMER_CHKSPC_EMERGENCY 0
762
763 #if defined(_KERNEL)
764
765 extern struct vop_ops hammer_vnode_vops;
766 extern struct vop_ops hammer_spec_vops;
767 extern struct vop_ops hammer_fifo_vops;
768 extern struct bio_ops hammer_bioops;
769
770 extern int hammer_debug_io;
771 extern int hammer_debug_general;
772 extern int hammer_debug_debug;
773 extern int hammer_debug_inode;
774 extern int hammer_debug_locks;
775 extern int hammer_debug_btree;
776 extern int hammer_debug_tid;
777 extern int hammer_debug_recover;
778 extern int hammer_debug_recover_faults;
779 extern int hammer_cluster_enable;
780 extern int hammer_count_fsyncs;
781 extern int hammer_count_inodes;
782 extern int hammer_count_iqueued;
783 extern int hammer_count_reclaiming;
784 extern int hammer_count_records;
785 extern int hammer_count_record_datas;
786 extern int hammer_count_volumes;
787 extern int hammer_count_buffers;
788 extern int hammer_count_nodes;
789 extern int64_t hammer_count_extra_space_used;
790 extern int64_t hammer_stats_btree_lookups;
791 extern int64_t hammer_stats_btree_searches;
792 extern int64_t hammer_stats_btree_inserts;
793 extern int64_t hammer_stats_btree_deletes;
794 extern int64_t hammer_stats_btree_elements;
795 extern int64_t hammer_stats_btree_splits;
796 extern int64_t hammer_stats_btree_iterations;
797 extern int64_t hammer_stats_record_iterations;
798 extern int64_t hammer_stats_file_read;
799 extern int64_t hammer_stats_file_write;
800 extern int64_t hammer_stats_file_iopsr;
801 extern int64_t hammer_stats_file_iopsw;
802 extern int64_t hammer_stats_disk_read;
803 extern int64_t hammer_stats_disk_write;
804 extern int64_t hammer_stats_inode_flushes;
805 extern int64_t hammer_stats_commits;
806 extern int hammer_count_dirtybufspace;
807 extern int hammer_count_refedbufs;
808 extern int hammer_count_reservations;
809 extern int hammer_count_io_running_read;
810 extern int hammer_count_io_running_write;
811 extern int hammer_count_io_locked;
812 extern int hammer_limit_dirtybufspace;
813 extern int hammer_limit_iqueued;
814 extern int hammer_limit_recs;
815 extern int hammer_bio_count;
816 extern int hammer_verify_zone;
817 extern int hammer_verify_data;
818 extern int hammer_write_mode;
819 extern int64_t hammer_contention_count;
820
821 void    hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
822                         int error, const char *msg);
823 int     hammer_vop_inactive(struct vop_inactive_args *);
824 int     hammer_vop_reclaim(struct vop_reclaim_args *);
825 int     hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
826 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
827                         hammer_inode_t dip, int64_t obj_id,
828                         hammer_tid_t asof, u_int32_t localization,
829                         int flags, int *errorp);
830 void    hammer_scan_inode_snapshots(hammer_mount_t hmp,
831                         hammer_inode_info_t iinfo,
832                         int (*callback)(hammer_inode_t ip, void *data),
833                         void *data);
834 void    hammer_put_inode(struct hammer_inode *ip);
835 void    hammer_put_inode_ref(struct hammer_inode *ip);
836 void    hammer_inode_waitreclaims(hammer_mount_t hmp);
837
838 int     hammer_unload_volume(hammer_volume_t volume, void *data __unused);
839 int     hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
840
841 int     hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused);
842 int     hammer_install_volume(hammer_mount_t hmp, const char *volname);
843 int     hammer_mountcheck_volumes(hammer_mount_t hmp);
844
845 int     hammer_ip_lookup(hammer_cursor_t cursor);
846 int     hammer_ip_first(hammer_cursor_t cursor);
847 int     hammer_ip_next(hammer_cursor_t cursor);
848 int     hammer_ip_resolve_data(hammer_cursor_t cursor);
849 int     hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
850                         hammer_tid_t tid);
851 int     hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
852                         hammer_tid_t delete_tid, u_int32_t delete_ts,
853                         int track, int64_t *stat_bytes);
854 int     hammer_ip_check_directory_empty(hammer_transaction_t trans,
855                         hammer_inode_t ip);
856 int     hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
857 int     hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
858
859 hammer_record_t
860         hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
861 void    hammer_flush_record_done(hammer_record_t record, int error);
862 void    hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
863 void    hammer_rel_mem_record(hammer_record_t record);
864
865 int     hammer_cursor_up(hammer_cursor_t cursor);
866 int     hammer_cursor_up_locked(hammer_cursor_t cursor);
867 int     hammer_cursor_down(hammer_cursor_t cursor);
868 int     hammer_cursor_upgrade(hammer_cursor_t cursor);
869 int     hammer_cursor_upgrade_node(hammer_cursor_t cursor);
870 void    hammer_cursor_downgrade(hammer_cursor_t cursor);
871 int     hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
872                         int index);
873 void    hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
874 int     hammer_lock_ex_try(struct hammer_lock *lock);
875 void    hammer_lock_sh(struct hammer_lock *lock);
876 int     hammer_lock_sh_try(struct hammer_lock *lock);
877 int     hammer_lock_upgrade(struct hammer_lock *lock);
878 void    hammer_lock_downgrade(struct hammer_lock *lock);
879 int     hammer_lock_status(struct hammer_lock *lock);
880 void    hammer_unlock(struct hammer_lock *lock);
881 void    hammer_ref(struct hammer_lock *lock);
882 void    hammer_unref(struct hammer_lock *lock);
883
884 void    hammer_sync_lock_ex(hammer_transaction_t trans);
885 void    hammer_sync_lock_sh(hammer_transaction_t trans);
886 int     hammer_sync_lock_sh_try(hammer_transaction_t trans);
887 void    hammer_sync_unlock(hammer_transaction_t trans);
888
889 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
890 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
891 void    hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
892 u_int64_t hammer_timespec_to_time(struct timespec *ts);
893 hammer_tid_t hammer_str_to_tid(const char *str, int *ispfs,
894                         u_int32_t *localizationp);
895 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip);
896 void hammer_clear_objid(hammer_inode_t dip);
897 void hammer_destroy_objid_cache(hammer_mount_t hmp);
898
899 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
900                         int bytes);
901 void hammer_clear_undo_history(hammer_mount_t hmp);
902 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
903 int hammer_get_dtype(u_int8_t obj_type);
904 u_int8_t hammer_get_obj_type(enum vtype vtype);
905 int64_t hammer_directory_namekey(const void *name, int len);
906 int     hammer_nohistory(hammer_inode_t ip);
907
908 int     hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
909                         hammer_node_cache_t cache, hammer_inode_t ip);
910 void    hammer_normalize_cursor(hammer_cursor_t cursor);
911 void    hammer_done_cursor(hammer_cursor_t cursor);
912 int     hammer_recover_cursor(hammer_cursor_t cursor);
913 void    hammer_unlock_cursor(hammer_cursor_t cursor, int also_ip);
914 int     hammer_lock_cursor(hammer_cursor_t cursor, int also_ip);
915 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor);
916 void    hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
917
918 void    hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
919 void    hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
920                         int index);
921 void    hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
922                         int index);
923 void    hammer_cursor_inserted_element(hammer_node_t node, int index);
924 void    hammer_cursor_deleted_element(hammer_node_t node, int index);
925
926 int     hammer_btree_lookup(hammer_cursor_t cursor);
927 int     hammer_btree_first(hammer_cursor_t cursor);
928 int     hammer_btree_last(hammer_cursor_t cursor);
929 int     hammer_btree_extract(hammer_cursor_t cursor, int flags);
930 int     hammer_btree_iterate(hammer_cursor_t cursor);
931 int     hammer_btree_iterate_reverse(hammer_cursor_t cursor);
932 int     hammer_btree_insert(hammer_cursor_t cursor,
933                             hammer_btree_leaf_elm_t elm, int *doprop);
934 int     hammer_btree_delete(hammer_cursor_t cursor);
935 void    hammer_btree_do_propagation(hammer_cursor_t cursor,
936                             hammer_pseudofs_inmem_t pfsm,
937                             hammer_btree_leaf_elm_t leaf);
938 int     hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
939 int     hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
940 int     hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
941 int     hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
942
943 int     btree_set_parent(hammer_transaction_t trans, hammer_node_t node,
944                         hammer_btree_elm_t elm);
945 int     hammer_btree_lock_children(hammer_cursor_t cursor,
946                         struct hammer_node_locklist **locklistp);
947 void    hammer_btree_unlock_children(struct hammer_node_locklist **locklistp);
948 int     hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
949 hammer_node_t hammer_btree_get_parent(hammer_node_t node, int *parent_indexp,
950                         int *errorp, int try_exclusive);
951
952 void    hammer_print_btree_node(hammer_node_ondisk_t ondisk);
953 void    hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
954
955 void    *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
956                         int *errorp, struct hammer_buffer **bufferp);
957 void    *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
958                         int *errorp, struct hammer_buffer **bufferp);
959 void    *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
960                         int *errorp, struct hammer_buffer **bufferp);
961 void    *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
962                         int *errorp, struct hammer_buffer **bufferp);
963
964 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
965
966 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
967                         int32_t vol_no, int *errorp);
968 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
969                         int bytes, int isnew, int *errorp);
970 void            hammer_sync_buffers(hammer_mount_t hmp,
971                         hammer_off_t base_offset, int bytes);
972 void            hammer_del_buffers(hammer_mount_t hmp,
973                         hammer_off_t base_offset,
974                         hammer_off_t zone2_offset, int bytes);
975
976 int             hammer_ref_volume(hammer_volume_t volume);
977 int             hammer_ref_buffer(hammer_buffer_t buffer);
978 void            hammer_flush_buffer_nodes(hammer_buffer_t buffer);
979
980 void            hammer_rel_volume(hammer_volume_t volume, int flush);
981 void            hammer_rel_buffer(hammer_buffer_t buffer, int flush);
982
983 int             hammer_vfs_export(struct mount *mp, int op,
984                         const struct export_args *export);
985 hammer_node_t   hammer_get_node(hammer_mount_t hmp, hammer_off_t node_offset,
986                         int isnew, int *errorp);
987 void            hammer_ref_node(hammer_node_t node);
988 hammer_node_t   hammer_ref_node_safe(struct hammer_mount *hmp,
989                         hammer_node_cache_t cache, int *errorp);
990 void            hammer_rel_node(hammer_node_t node);
991 void            hammer_delete_node(hammer_transaction_t trans,
992                         hammer_node_t node);
993 void            hammer_cache_node(hammer_node_cache_t cache,
994                         hammer_node_t node);
995 void            hammer_uncache_node(hammer_node_cache_t cache);
996 void            hammer_flush_node(hammer_node_t node);
997
998 void hammer_dup_buffer(struct hammer_buffer **bufferp,
999                         struct hammer_buffer *buffer);
1000 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans, int *errorp);
1001 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1002                         u_int16_t rec_type, hammer_off_t *data_offsetp,
1003                         struct hammer_buffer **data_bufferp, int *errorp);
1004
1005 int hammer_generate_undo(hammer_transaction_t trans, hammer_io_t io,
1006                         hammer_off_t zone1_offset, void *base, int len);
1007
1008 void hammer_put_volume(struct hammer_volume *volume, int flush);
1009 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
1010
1011 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1012                         hammer_off_t owner, int *errorp);
1013 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1014                         hammer_off_t owner, int *errorp);
1015 int hammer_checkspace(hammer_mount_t hmp, int slop);
1016 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1017                         int bytes, int *errorp);
1018 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1019                         int bytes, hammer_off_t *zone_offp, int *errorp);
1020 void hammer_blockmap_reserve_undo(hammer_reserve_t resv,
1021                         hammer_off_t zone_offset, int bytes);
1022 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1023                         hammer_reserve_t resv);
1024 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1025 void hammer_blockmap_free(hammer_transaction_t trans,
1026                         hammer_off_t bmap_off, int bytes);
1027 int hammer_blockmap_finalize(hammer_transaction_t trans,
1028                         hammer_off_t bmap_off, int bytes);
1029 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
1030                         int *curp, int *errorp);
1031 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1032                         int *errorp);
1033 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1034                         int *errorp);
1035 int64_t hammer_undo_used(hammer_transaction_t trans);
1036 int64_t hammer_undo_space(hammer_transaction_t trans);
1037 int64_t hammer_undo_max(hammer_mount_t hmp);
1038
1039 void hammer_start_transaction(struct hammer_transaction *trans,
1040                               struct hammer_mount *hmp);
1041 void hammer_simple_transaction(struct hammer_transaction *trans,
1042                               struct hammer_mount *hmp);
1043 void hammer_start_transaction_fls(struct hammer_transaction *trans,
1044                                   struct hammer_mount *hmp);
1045 void hammer_done_transaction(struct hammer_transaction *trans);
1046
1047 void hammer_modify_inode(hammer_inode_t ip, int flags);
1048 void hammer_flush_inode(hammer_inode_t ip, int flags);
1049 void hammer_flush_inode_done(hammer_inode_t ip, int error);
1050 void hammer_wait_inode(hammer_inode_t ip);
1051
1052 int  hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
1053                         struct ucred *cred, struct hammer_inode *dip,
1054                         hammer_pseudofs_inmem_t pfsm,
1055                         struct hammer_inode **ipp);
1056 void hammer_rel_inode(hammer_inode_t ip, int flush);
1057 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1058 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1059 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1060
1061 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1062 void hammer_test_inode(hammer_inode_t dip);
1063 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1064
1065 int  hammer_ip_add_directory(struct hammer_transaction *trans,
1066                         hammer_inode_t dip, const char *name, int bytes,
1067                         hammer_inode_t nip);
1068 int  hammer_ip_del_directory(struct hammer_transaction *trans,
1069                         hammer_cursor_t cursor, hammer_inode_t dip,
1070                         hammer_inode_t ip);
1071 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1072                         void *data, int bytes, int *errorp);
1073 int  hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
1074 int  hammer_ip_add_record(struct hammer_transaction *trans,
1075                         hammer_record_t record);
1076 int  hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1077                         int64_t ran_beg, int64_t ran_end, int truncating);
1078 int  hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1079                         int *countp);
1080 int  hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1081                         int64_t offset, void *data, int bytes);
1082 int  hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
1083 int  hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1084 hammer_pseudofs_inmem_t  hammer_load_pseudofs(hammer_transaction_t trans,
1085                         u_int32_t localization, int *errorp);
1086 int  hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1087                         hammer_pseudofs_inmem_t pfsm);
1088 int  hammer_save_pseudofs(hammer_transaction_t trans,
1089                         hammer_pseudofs_inmem_t pfsm);
1090 int  hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization);
1091 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1092 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1093                         struct ucred *cred);
1094
1095 void hammer_io_init(hammer_io_t io, hammer_mount_t hmp,
1096                         enum hammer_io_type type);
1097 int hammer_io_read(struct vnode *devvp, struct hammer_io *io,
1098                         hammer_off_t limit);
1099 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
1100 void hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1101 struct buf *hammer_io_release(struct hammer_io *io, int flush);
1102 void hammer_io_flush(struct hammer_io *io);
1103 void hammer_io_wait(struct hammer_io *io);
1104 void hammer_io_waitdep(struct hammer_io *io);
1105 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident);
1106 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1107                         hammer_btree_leaf_elm_t leaf);
1108 int hammer_io_direct_write(hammer_mount_t hmp, hammer_record_t record,
1109                         struct bio *bio);
1110 void hammer_io_direct_wait(hammer_record_t record);
1111 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1112 void hammer_io_write_interlock(hammer_io_t io);
1113 void hammer_io_done_interlock(hammer_io_t io);
1114 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1115 void hammer_io_clear_modlist(struct hammer_io *io);
1116 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1117                         void *base, int len);
1118 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1119                         void *base, int len);
1120 void hammer_modify_volume_done(hammer_volume_t volume);
1121 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1122
1123 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1124                         struct hammer_ioc_reblock *reblock);
1125 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1126                         struct hammer_ioc_prune *prune);
1127 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1128                         struct hammer_ioc_mirror_rw *mirror);
1129 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1130                         struct hammer_ioc_mirror_rw *mirror);
1131 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1132                         struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1133 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1134                         struct hammer_ioc_pseudofs_rw *pfs);
1135 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1136                         struct hammer_ioc_pseudofs_rw *pfs);
1137 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1138                         struct hammer_ioc_pseudofs_rw *pfs);
1139 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1140                         struct hammer_ioc_pseudofs_rw *pfs);
1141
1142 int hammer_signal_check(hammer_mount_t hmp);
1143
1144 void hammer_flusher_create(hammer_mount_t hmp);
1145 void hammer_flusher_destroy(hammer_mount_t hmp);
1146 void hammer_flusher_sync(hammer_mount_t hmp);
1147 int  hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1148 int  hammer_flusher_async_one(hammer_mount_t hmp);
1149 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1150 int  hammer_flusher_meta_limit(hammer_mount_t hmp);
1151 int  hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1152 int  hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1153 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1154 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1155 int  hammer_flusher_haswork(hammer_mount_t hmp);
1156
1157
1158 int hammer_recover(hammer_mount_t hmp, hammer_volume_t rootvol);
1159 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1160                         hammer_volume_t root_volume, int final);
1161
1162 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1163 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1164 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1165
1166 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1167 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1168 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1169 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1170 void hkprintf(const char *ctl, ...);
1171 udev_t hammer_fsid_to_udev(uuid_t *uuid);
1172
1173
1174 int hammer_blocksize(int64_t file_offset);
1175 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1176
1177 #endif
1178
1179 static __inline void
1180 hammer_wait_mem_record(hammer_record_t record)
1181 {
1182         hammer_wait_mem_record_ident(record, "hmmwai");
1183 }
1184
1185 static __inline void
1186 hammer_lock_ex(struct hammer_lock *lock)
1187 {
1188         hammer_lock_ex_ident(lock, "hmrlck");
1189 }
1190
1191 /*
1192  * Indicate that a B-Tree node is being modified.
1193  */
1194 static __inline void
1195 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1196 {
1197         hammer_modify_buffer(trans, node->buffer, NULL, 0);
1198 }
1199
1200 static __inline void
1201 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1202 {
1203         hammer_modify_buffer(trans, node->buffer,
1204                              node->ondisk, sizeof(*node->ondisk));
1205 }
1206
1207 static __inline void
1208 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1209                    void *base, int len)
1210 {
1211         hammer_crc_t *crcptr;
1212
1213         KKASSERT((char *)base >= (char *)node->ondisk &&
1214                  (char *)base + len <=
1215                     (char *)node->ondisk + sizeof(*node->ondisk));
1216         hammer_modify_buffer(trans, node->buffer, base, len);
1217         crcptr = &node->ondisk->crc;
1218         hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t));
1219         --node->buffer->io.modify_refs; /* only want one ref */
1220 }
1221
1222 /*
1223  * Indicate that the specified modifications have been completed.
1224  *
1225  * Do not try to generate the crc here, it's very expensive to do and a
1226  * sequence of insertions or deletions can result in many calls to this
1227  * function on the same node.
1228  */
1229 static __inline void
1230 hammer_modify_node_done(hammer_node_t node)
1231 {
1232         node->flags |= HAMMER_NODE_CRCGOOD;
1233         if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1234                 node->flags |= HAMMER_NODE_NEEDSCRC;
1235                 node->buffer->io.gencrc = 1;
1236                 hammer_ref_node(node);
1237         }
1238         hammer_modify_buffer_done(node->buffer);
1239 }
1240
1241 #define hammer_modify_volume_field(trans, vol, field)           \
1242         hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1243                              sizeof((vol)->ondisk->field))
1244
1245 #define hammer_modify_node_field(trans, node, field)            \
1246         hammer_modify_node(trans, node, &(node)->ondisk->field, \
1247                              sizeof((node)->ondisk->field))
1248