89318441e15d8d714f7b0f6b69f675516085a1d9
[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 <sys/mplock2.h>
64 #include "hammer_disk.h"
65 #include "hammer_mount.h"
66 #include "hammer_ioctl.h"
67
68 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
69
70 MALLOC_DECLARE(M_HAMMER);
71
72 /*
73  * Kernel trace
74  */
75 #if !defined(KTR_HAMMER)
76 #define KTR_HAMMER      KTR_ALL
77 #endif
78 KTR_INFO_MASTER_EXTERN(hammer);
79
80 /*
81  * Misc structures
82  */
83 struct hammer_mount;
84
85 /*
86  * Key structure used for custom RB tree inode lookups.  This prototypes
87  * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
88  */
89 typedef struct hammer_inode_info {
90         int64_t         obj_id;         /* (key) object identifier */
91         hammer_tid_t    obj_asof;       /* (key) snapshot transid or 0 */
92         u_int32_t       obj_localization; /* (key) pseudo-fs */
93         union {
94                 struct hammer_btree_leaf_elm *leaf;
95         } u;
96 } *hammer_inode_info_t;
97
98 typedef enum hammer_transaction_type {
99         HAMMER_TRANS_RO,
100         HAMMER_TRANS_STD,
101         HAMMER_TRANS_FLS
102 } hammer_transaction_type_t;
103
104 /*
105  * HAMMER Transaction tracking
106  */
107 struct hammer_transaction {
108         hammer_transaction_type_t type;
109         struct hammer_mount *hmp;
110         hammer_tid_t    tid;
111         u_int64_t       time;
112         u_int32_t       time32;
113         int             sync_lock_refs;
114         int             flags;
115         struct hammer_volume *rootvol;
116 };
117
118 typedef struct hammer_transaction *hammer_transaction_t;
119
120 #define HAMMER_TRANSF_NEWINODE  0x0001
121 #define HAMMER_TRANSF_DIDIO     0x0002
122 #define HAMMER_TRANSF_CRCDOM    0x0004  /* EDOM on CRC error, less critical */
123
124 /*
125  * HAMMER locks
126  */
127 struct hammer_lock {
128         int             refs;           /* active references delay writes */
129         volatile u_int  lockval;        /* lock count and control bits */
130         struct thread   *owner;         /* owner if exclusively held */
131 };
132
133 #define HAMMER_LOCKF_EXCLUSIVE  0x40000000
134 #define HAMMER_LOCKF_WANTED     0x80000000
135
136 static __inline int
137 hammer_notlocked(struct hammer_lock *lock)
138 {
139         return(lock->lockval == 0);
140 }
141
142 static __inline int
143 hammer_islocked(struct hammer_lock *lock)
144 {
145         return(lock->lockval != 0);
146 }
147
148 static __inline int
149 hammer_isactive(struct hammer_lock *lock)
150 {
151         return(lock->refs != 0);
152 }
153
154 static __inline int
155 hammer_islastref(struct hammer_lock *lock)
156 {
157         return(lock->refs == 1);
158 }
159
160 /*
161  * Return if we specifically own the lock exclusively.
162  */
163 static __inline int
164 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
165 {
166         if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) &&
167             lock->owner == td) {
168                 return(1);
169         }
170         return(0);
171 }
172
173 /*
174  * Flush state, used by various structures
175  */
176 typedef enum hammer_inode_state {
177         HAMMER_FST_IDLE,
178         HAMMER_FST_SETUP,
179         HAMMER_FST_FLUSH
180 } hammer_inode_state_t;
181
182 TAILQ_HEAD(hammer_record_list, hammer_record);
183
184 /*
185  * Pseudo-filesystem extended data tracking
186  */
187 struct hammer_pfs_rb_tree;
188 struct hammer_pseudofs_inmem;
189 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
190 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
191               hammer_pfs_rb_compare, u_int32_t);
192
193 struct hammer_pseudofs_inmem {
194         RB_ENTRY(hammer_pseudofs_inmem) rb_node;
195         struct hammer_lock      lock;
196         u_int32_t               localization;
197         hammer_tid_t            create_tid;
198         int                     flags;
199         udev_t                  fsid_udev;
200         struct hammer_pseudofs_data pfsd;
201 };
202
203 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t;
204
205 #define HAMMER_PFSM_DELETED     0x0001
206
207 /*
208  * Cache object ids.  A fixed number of objid cache structures are
209  * created to reserve object id's for newly created files in multiples
210  * of 100,000, localized to a particular directory, and recycled as
211  * needed.  This allows parallel create operations in different
212  * directories to retain fairly localized object ids which in turn
213  * improves reblocking performance and layout.
214  */
215 #define OBJID_CACHE_SIZE        1024
216 #define OBJID_CACHE_BULK_BITS   10              /* 10 bits (1024)       */
217 #define OBJID_CACHE_BULK        (32 * 32)       /* two level (1024)     */
218 #define OBJID_CACHE_BULK_MASK   (OBJID_CACHE_BULK - 1)
219 #define OBJID_CACHE_BULK_MASK64 ((u_int64_t)(OBJID_CACHE_BULK - 1))
220
221 typedef struct hammer_objid_cache {
222         TAILQ_ENTRY(hammer_objid_cache) entry;
223         struct hammer_inode             *dip;
224         hammer_tid_t                    base_tid;
225         int                             count;
226         u_int32_t                       bm0;
227         u_int32_t                       bm1[32];
228 } *hammer_objid_cache_t;
229
230 /*
231  * Associate an inode with a B-Tree node to cache search start positions
232  */
233 typedef struct hammer_node_cache {
234         TAILQ_ENTRY(hammer_node_cache)  entry;
235         struct hammer_node              *node;
236         struct hammer_inode             *ip;
237 } *hammer_node_cache_t;
238
239 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
240
241 /*
242  * Structure used to organize flush groups.  Flush groups must be
243  * organized into chunks in order to avoid blowing out the UNDO FIFO.
244  * Without this a 'sync' could end up flushing 50,000 inodes in a single
245  * transaction.
246  */
247 struct hammer_fls_rb_tree;
248 RB_HEAD(hammer_fls_rb_tree, hammer_inode);
249 RB_PROTOTYPE(hammer_fls_rb_tree, hammer_inode, rb_flsnode,
250               hammer_ino_rb_compare);
251
252 struct hammer_flush_group {
253         TAILQ_ENTRY(hammer_flush_group) flush_entry;
254         struct hammer_fls_rb_tree       flush_tree;
255         int                             unused01;
256         int                             total_count;    /* record load */
257         int                             running;        /* group is running */
258         int                             closed;
259         int                             refs;
260 };
261
262 typedef struct hammer_flush_group *hammer_flush_group_t;
263
264 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
265
266 /*
267  * Structure used to represent an inode in-memory.
268  *
269  * The record and data associated with an inode may be out of sync with
270  * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
271  * clear).
272  *
273  * An inode may also hold a cache of unsynchronized records, used for
274  * database and directories only.  Unsynchronized regular file data is
275  * stored in the buffer cache.
276  *
277  * NOTE: A file which is created and destroyed within the initial
278  * synchronization period can wind up not doing any disk I/O at all.
279  *
280  * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
281  */
282 struct hammer_ino_rb_tree;
283 struct hammer_inode;
284 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
285 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
286               hammer_ino_rb_compare, hammer_inode_info_t);
287
288 struct hammer_rec_rb_tree;
289 struct hammer_record;
290 RB_HEAD(hammer_rec_rb_tree, hammer_record);
291 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
292               hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
293
294 TAILQ_HEAD(hammer_node_list, hammer_node);
295
296 struct hammer_inode {
297         RB_ENTRY(hammer_inode)  rb_node;
298         hammer_inode_state_t    flush_state;
299         hammer_flush_group_t    flush_group;
300         RB_ENTRY(hammer_inode)  rb_flsnode;     /* when on flush list */
301         struct hammer_record_list target_list;  /* target of dependant recs */
302         int64_t                 obj_id;         /* (key) object identifier */
303         hammer_tid_t            obj_asof;       /* (key) snapshot or 0 */
304         u_int32_t               obj_localization; /* (key) pseudo-fs */
305         struct hammer_mount     *hmp;
306         hammer_objid_cache_t    objid_cache;
307         int                     flags;
308         int                     error;          /* flush error */
309         int                     cursor_ip_refs; /* sanity */
310         int                     rsv_recs;
311         struct vnode            *vp;
312         hammer_pseudofs_inmem_t pfsm;
313         struct lockf            advlock;
314         struct hammer_lock      lock;           /* sync copy interlock */
315         off_t                   trunc_off;
316         struct hammer_btree_leaf_elm ino_leaf;  /* in-memory cache */
317         struct hammer_inode_data ino_data;      /* in-memory cache */
318         struct hammer_rec_rb_tree rec_tree;     /* in-memory cache */
319         int                     rec_generation;
320         struct hammer_node_cache cache[4];      /* search initiate cache */
321
322         /*
323          * When a demark is created to synchronize an inode to
324          * disk, certain fields are copied so the front-end VOPs
325          * can continue to run in parallel with the synchronization
326          * occuring in the background.
327          */
328         int             sync_flags;             /* to-sync flags cache */
329         off_t           sync_trunc_off;         /* to-sync truncation */
330         off_t           save_trunc_off;         /* write optimization */
331         struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
332         struct hammer_inode_data sync_ino_data; /* to-sync cache */
333 };
334
335 typedef struct hammer_inode *hammer_inode_t;
336
337 #define VTOI(vp)        ((struct hammer_inode *)(vp)->v_data)
338
339 #define HAMMER_INODE_DDIRTY     0x0001  /* in-memory ino_data is dirty */
340                                         /* (not including atime/mtime) */
341 #define HAMMER_INODE_RSV_INODES 0x0002  /* hmp->rsv_inodes bumped */
342 #define HAMMER_INODE_CONN_DOWN  0x0004  /* include in downward recursion */
343 #define HAMMER_INODE_XDIRTY     0x0008  /* in-memory records */
344 #define HAMMER_INODE_ONDISK     0x0010  /* inode is on-disk (else not yet) */
345 #define HAMMER_INODE_FLUSH      0x0020  /* flush on last ref */
346 #define HAMMER_INODE_DELETED    0x0080  /* inode delete (backend) */
347 #define HAMMER_INODE_DELONDISK  0x0100  /* delete synchronized to disk */
348 #define HAMMER_INODE_RO         0x0200  /* read-only (because of as-of) */
349 #define HAMMER_INODE_VHELD      0x0400  /* vnode held on sync */
350 #define HAMMER_INODE_DONDISK    0x0800  /* data records may be on disk */
351 #define HAMMER_INODE_BUFS       0x1000  /* dirty high level bps present */
352 #define HAMMER_INODE_REFLUSH    0x2000  /* flush on dependancy / reflush */
353 #define HAMMER_INODE_RECLAIM    0x4000  /* trying to reclaim */
354 #define HAMMER_INODE_FLUSHW     0x8000  /* Someone waiting for flush */
355
356 #define HAMMER_INODE_TRUNCATED  0x00010000
357 #define HAMMER_INODE_DELETING   0x00020000 /* inode delete request (frontend)*/
358 #define HAMMER_INODE_RESIGNAL   0x00040000 /* re-signal on re-flush */
359 #define HAMMER_INODE_ATIME      0x00100000 /* in-memory atime modified */
360 #define HAMMER_INODE_MTIME      0x00200000 /* in-memory mtime modified */
361 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
362 #define HAMMER_INODE_DUMMY      0x00800000 /* dummy inode covering bad file */
363 #define HAMMER_INODE_CLOSESYNC  0x01000000 /* synchronously fsync on close */
364 #define HAMMER_INODE_CLOSEASYNC 0x02000000 /* asynchronously fsync on close */
365
366 #define HAMMER_INODE_MODMASK    (HAMMER_INODE_DDIRTY|                       \
367                                  HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS|     \
368                                  HAMMER_INODE_ATIME|HAMMER_INODE_MTIME|     \
369                                  HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
370
371 #define HAMMER_INODE_MODMASK_NOXDIRTY \
372                                 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
373
374 #define HAMMER_FLUSH_SIGNAL     0x0001
375 #define HAMMER_FLUSH_RECURSION  0x0002
376
377 /*
378  * Used by the inode reclaim code to pipeline reclaims and avoid
379  * blowing out kernel memory or letting the flusher get too far
380  * behind.  The reclaim wakes up when count reaches 0 or the
381  * timer expires.
382  */
383 struct hammer_reclaim {
384         TAILQ_ENTRY(hammer_reclaim) entry;
385         int     count;
386 };
387
388 #define HAMMER_RECLAIM_WAIT     4000    /* default vfs.hammer.limit_reclaim */
389
390 /*
391  * Structure used to represent an unsynchronized record in-memory.  These
392  * records typically represent directory entries.  Only non-historical
393  * records are kept in-memory.
394  *
395  * Records are organized as a per-inode RB-Tree.  If the inode is not
396  * on disk then neither are any records and the in-memory record tree
397  * represents the entire contents of the inode.  If the inode is on disk
398  * then the on-disk B-Tree is scanned in parallel with the in-memory
399  * RB-Tree to synthesize the current state of the file.
400  *
401  * Records are also used to enforce the ordering of directory create/delete
402  * operations.  A new inode will not be flushed to disk unless its related
403  * directory entry is also being flushed at the same time.  A directory entry
404  * will not be removed unless its related inode is also being removed at the
405  * same time.
406  */
407 typedef enum hammer_record_type {
408         HAMMER_MEM_RECORD_GENERAL,      /* misc record */
409         HAMMER_MEM_RECORD_INODE,        /* inode record */
410         HAMMER_MEM_RECORD_ADD,          /* positive memory cache record */
411         HAMMER_MEM_RECORD_DEL,          /* negative delete-on-disk record */
412         HAMMER_MEM_RECORD_DATA          /* bulk-data record w/on-disk ref */
413 } hammer_record_type_t;
414
415 struct hammer_record {
416         RB_ENTRY(hammer_record)         rb_node;
417         TAILQ_ENTRY(hammer_record)      target_entry;
418         hammer_inode_state_t            flush_state;
419         hammer_flush_group_t            flush_group;
420         hammer_record_type_t            type;
421         struct hammer_lock              lock;
422         struct hammer_reserve           *resv;
423         struct hammer_inode             *ip;
424         struct hammer_inode             *target_ip;
425         struct hammer_btree_leaf_elm    leaf;
426         union hammer_data_ondisk        *data;
427         int                             flags;
428         hammer_off_t                    zone2_offset;   /* direct-write only */
429 };
430
431 typedef struct hammer_record *hammer_record_t;
432
433 /*
434  * Record flags.  Note that FE can only be set by the frontend if the
435  * record has not been interlocked by the backend w/ BE.
436  */
437 #define HAMMER_RECF_ALLOCDATA           0x0001
438 #define HAMMER_RECF_ONRBTREE            0x0002
439 #define HAMMER_RECF_DELETED_FE          0x0004  /* deleted (frontend) */
440 #define HAMMER_RECF_DELETED_BE          0x0008  /* deleted (backend) */
441 #define HAMMER_RECF_COMMITTED           0x0010  /* committed to the B-Tree */
442 #define HAMMER_RECF_INTERLOCK_BE        0x0020  /* backend interlock */
443 #define HAMMER_RECF_WANTED              0x0040  /* wanted by the frontend */
444 #define HAMMER_RECF_CONVERT_DELETE      0x0100  /* special case */
445 #define HAMMER_RECF_DIRECT_IO           0x0200  /* related direct I/O running*/
446 #define HAMMER_RECF_DIRECT_WAIT         0x0400  /* related direct I/O running*/
447 #define HAMMER_RECF_DIRECT_INVAL        0x0800  /* buffer alias invalidation */
448
449 /*
450  * hammer_create_at_cursor() and hammer_delete_at_cursor() flags.
451  */
452 #define HAMMER_CREATE_MODE_UMIRROR      0x0001
453 #define HAMMER_CREATE_MODE_SYS          0x0002
454
455 #define HAMMER_DELETE_ADJUST            0x0001
456 #define HAMMER_DELETE_DESTROY           0x0002
457
458 /*
459  * In-memory structures representing on-disk structures.
460  */
461 struct hammer_volume;
462 struct hammer_buffer;
463 struct hammer_node;
464 struct hammer_undo;
465 struct hammer_reserve;
466
467 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
468 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
469 RB_HEAD(hammer_nod_rb_tree, hammer_node);
470 RB_HEAD(hammer_und_rb_tree, hammer_undo);
471 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
472
473 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
474               hammer_vol_rb_compare, int32_t);
475 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
476               hammer_buf_rb_compare, hammer_off_t);
477 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
478               hammer_nod_rb_compare, hammer_off_t);
479 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
480               hammer_und_rb_compare, hammer_off_t);
481 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
482               hammer_res_rb_compare, hammer_off_t);
483
484 /*
485  * IO management - embedded at the head of various in-memory structures
486  *
487  * VOLUME       - hammer_volume containing meta-data
488  * META_BUFFER  - hammer_buffer containing meta-data
489  * DATA_BUFFER  - hammer_buffer containing pure-data
490  *
491  * Dirty volume headers and dirty meta-data buffers are locked until the
492  * flusher can sequence them out.  Dirty pure-data buffers can be written.
493  * Clean buffers can be passively released.
494  */
495 typedef enum hammer_io_type {
496         HAMMER_STRUCTURE_VOLUME,
497         HAMMER_STRUCTURE_META_BUFFER,
498         HAMMER_STRUCTURE_UNDO_BUFFER,
499         HAMMER_STRUCTURE_DATA_BUFFER
500 } hammer_io_type_t;
501
502 union hammer_io_structure;
503 struct hammer_io;
504
505 struct worklist {
506         LIST_ENTRY(worklist) node;
507 };
508
509 TAILQ_HEAD(hammer_io_list, hammer_io);
510 typedef struct hammer_io_list *hammer_io_list_t;
511
512 struct hammer_io {
513         struct worklist         worklist;
514         struct hammer_lock      lock;
515         enum hammer_io_type     type;
516         struct hammer_mount     *hmp;
517         struct hammer_volume    *volume;
518         TAILQ_ENTRY(hammer_io)  mod_entry; /* list entry if modified */
519         hammer_io_list_t        mod_list;
520         struct buf              *bp;
521         int64_t                 offset;    /* zone-2 offset */
522         int                     bytes;     /* buffer cache buffer size */
523         int                     loading;   /* loading/unloading interlock */
524         int                     modify_refs;
525
526         u_int           modified : 1;   /* bp's data was modified */
527         u_int           released : 1;   /* bp released (w/ B_LOCKED set) */
528         u_int           running : 1;    /* bp write IO in progress */
529         u_int           waiting : 1;    /* someone is waiting on us */
530         u_int           validated : 1;  /* ondisk has been validated */
531         u_int           waitdep : 1;    /* flush waits for dependancies */
532         u_int           recovered : 1;  /* has recovery ref */
533         u_int           waitmod : 1;    /* waiting for modify_refs */
534         u_int           reclaim : 1;    /* reclaim requested */
535         u_int           gencrc : 1;     /* crc needs to be generated */
536         u_int           ioerror : 1;    /* abort on io-error */
537 };
538
539 typedef struct hammer_io *hammer_io_t;
540
541 #define HAMMER_CLUSTER_SIZE     (64 * 1024)
542 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
543 #undef  HAMMER_CLUSTER_SIZE
544 #define HAMMER_CLUSTER_SIZE     MAXBSIZE
545 #endif
546 #define HAMMER_CLUSTER_BUFS     (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
547
548 /*
549  * In-memory volume representing on-disk buffer
550  */
551 struct hammer_volume {
552         struct hammer_io io;
553         RB_ENTRY(hammer_volume) rb_node;
554         struct hammer_volume_ondisk *ondisk;
555         int32_t vol_no;
556         int64_t nblocks;        /* note: special calculation for statfs */
557         int64_t buffer_base;    /* base offset of buffer 0 */
558         hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
559         hammer_off_t maxraw_off; /* Maximum raw offset for device */
560         char    *vol_name;
561         struct vnode *devvp;
562         int     vol_flags;
563 };
564
565 typedef struct hammer_volume *hammer_volume_t;
566
567 /*
568  * In-memory buffer (other then volume, super-cluster, or cluster),
569  * representing an on-disk buffer.
570  */
571 struct hammer_buffer {
572         struct hammer_io io;
573         RB_ENTRY(hammer_buffer) rb_node;
574         void *ondisk;
575         hammer_off_t zoneX_offset;
576         hammer_off_t zone2_offset;
577         struct hammer_reserve *resv;
578         struct hammer_node_list clist;
579 };
580
581 typedef struct hammer_buffer *hammer_buffer_t;
582
583 /*
584  * In-memory B-Tree node, representing an on-disk B-Tree node.
585  *
586  * This is a hang-on structure which is backed by a hammer_buffer,
587  * indexed by a hammer_cluster, and used for fine-grained locking of
588  * B-Tree nodes in order to properly control lock ordering.  A hammer_buffer
589  * can contain multiple nodes representing wildly disassociated portions
590  * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
591  *
592  * This structure uses a cluster-relative index to reduce the number
593  * of layers required to access it, and also because all on-disk B-Tree
594  * references are cluster-relative offsets.
595  */
596 struct hammer_node {
597         struct hammer_lock      lock;           /* node-by-node lock */
598         TAILQ_ENTRY(hammer_node) entry;         /* per-buffer linkage */
599         RB_ENTRY(hammer_node)   rb_node;        /* per-cluster linkage */
600         hammer_off_t            node_offset;    /* full offset spec */
601         struct hammer_mount     *hmp;
602         struct hammer_buffer    *buffer;        /* backing buffer */
603         hammer_node_ondisk_t    ondisk;         /* ptr to on-disk structure */
604         TAILQ_HEAD(, hammer_cursor) cursor_list;  /* deadlock recovery */
605         struct hammer_node_cache_list cache_list; /* passive caches */
606         int                     flags;
607         int                     loading;        /* load interlock */
608 };
609
610 #define HAMMER_NODE_DELETED     0x0001
611 #define HAMMER_NODE_FLUSH       0x0002
612 #define HAMMER_NODE_CRCGOOD     0x0004
613 #define HAMMER_NODE_NEEDSCRC    0x0008
614 #define HAMMER_NODE_NEEDSMIRROR 0x0010
615 #define HAMMER_NODE_CRCBAD      0x0020
616
617 #define HAMMER_NODE_CRCANY      (HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD)
618
619 typedef struct hammer_node      *hammer_node_t;
620
621 /*
622  * List of locked nodes.  This structure is used to lock potentially large
623  * numbers of nodes as an aid for complex B-Tree operations.
624  */
625 struct hammer_node_lock;
626 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock);
627
628 struct hammer_node_lock {
629         TAILQ_ENTRY(hammer_node_lock) entry;
630         struct hammer_node_lock_list  list;
631         struct hammer_node_lock       *parent;
632         hammer_node_t   node;
633         hammer_node_ondisk_t copy;      /* copy of on-disk data */
634         int             index;          /* index of this node in parent */
635         int             count;          /* count children */
636         int             flags;
637 };
638
639 typedef struct hammer_node_lock *hammer_node_lock_t;
640
641 #define HAMMER_NODE_LOCK_UPDATED        0x0001
642
643 /*
644  * Common I/O management structure - embedded in in-memory structures
645  * which are backed by filesystem buffers.
646  */
647 union hammer_io_structure {
648         struct hammer_io        io;
649         struct hammer_volume    volume;
650         struct hammer_buffer    buffer;
651 };
652
653 typedef union hammer_io_structure *hammer_io_structure_t;
654
655 /*
656  * The reserve structure prevents the blockmap from allocating
657  * out of a reserved bigblock.  Such reservations are used by
658  * the direct-write mechanism.
659  *
660  * The structure is also used to hold off on reallocations of
661  * big blocks from the freemap until flush dependancies have
662  * been dealt with.
663  */
664 struct hammer_reserve {
665         RB_ENTRY(hammer_reserve) rb_node;
666         TAILQ_ENTRY(hammer_reserve) delay_entry;
667         int             flush_group;
668         int             flags;
669         int             refs;
670         int             zone;
671         int             append_off;
672         hammer_off_t    zone_offset;
673 };
674
675 typedef struct hammer_reserve *hammer_reserve_t;
676
677 #define HAMMER_RESF_ONDELAY     0x0001
678 #define HAMMER_RESF_LAYER2FREE  0x0002
679
680 #include "hammer_cursor.h"
681
682 /*
683  * The undo structure tracks recent undos to avoid laying down duplicate
684  * undos within a flush group, saving us a significant amount of overhead.
685  *
686  * This is strictly a heuristic.
687  */
688 #define HAMMER_MAX_UNDOS                1024
689 #define HAMMER_MAX_FLUSHERS             4
690
691 struct hammer_undo {
692         RB_ENTRY(hammer_undo)   rb_node;
693         TAILQ_ENTRY(hammer_undo) lru_entry;
694         hammer_off_t            offset;
695         int                     bytes;
696 };
697
698 typedef struct hammer_undo *hammer_undo_t;
699
700 struct hammer_flusher_info;
701 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
702
703 struct hammer_flusher {
704         int             signal;         /* flusher thread sequencer */
705         int             act;            /* currently active flush group */
706         int             done;           /* set to act when complete */
707         int             next;           /* next flush group */
708         int             group_lock;     /* lock sequencing of the next flush */
709         int             exiting;        /* request master exit */
710         thread_t        td;             /* master flusher thread */
711         hammer_tid_t    tid;            /* last flushed transaction id */
712         int             finalize_want;          /* serialize finalization */
713         struct hammer_lock finalize_lock;       /* serialize finalization */
714         struct hammer_transaction trans;        /* shared transaction */
715         struct hammer_flusher_info_list run_list;
716         struct hammer_flusher_info_list ready_list;
717 };
718
719 /*
720  * Internal hammer mount data structure
721  */
722 struct hammer_mount {
723         struct mount *mp;
724         /*struct vnode *rootvp;*/
725         struct hammer_ino_rb_tree rb_inos_root;
726         struct hammer_vol_rb_tree rb_vols_root;
727         struct hammer_nod_rb_tree rb_nods_root;
728         struct hammer_und_rb_tree rb_undo_root;
729         struct hammer_res_rb_tree rb_resv_root;
730         struct hammer_buf_rb_tree rb_bufs_root;
731         struct hammer_pfs_rb_tree rb_pfsm_root;
732         struct hammer_volume *rootvol;
733         struct hammer_base_elm root_btree_beg;
734         struct hammer_base_elm root_btree_end;
735
736         struct malloc_type      *m_misc;
737         struct malloc_type      *m_inodes;
738
739         int     flags;          /* HAMMER_MOUNT_xxx flags */
740         int     hflags;
741         int     ronly;
742         int     nvolumes;
743         int     volume_iterator;
744         int     master_id;      /* -1 or 0-15 - clustering and mirroring */
745         int     version;        /* hammer filesystem version to use */
746         int     rsv_inodes;     /* reserved space due to dirty inodes */
747         int64_t rsv_databytes;  /* reserved space due to record data */
748         int     rsv_recs;       /* reserved space due to dirty records */
749         int     rsv_fromdelay;  /* bigblocks reserved due to flush delay */
750         int     undo_rec_limit; /* based on size of undo area */
751         int     last_newrecords;
752         int     count_newrecords;
753
754         int     volume_to_remove; /* volume that is currently being removed */
755
756         int     inode_reclaims; /* inodes pending reclaim by flusher */
757         int     count_inodes;   /* total number of inodes */
758         int     count_iqueued;  /* inodes queued to flusher */
759
760         struct hammer_flusher flusher;
761
762         u_int   check_interrupt;
763         u_int   check_yield;
764         uuid_t  fsid;
765         struct hammer_io_list volu_list;        /* dirty undo buffers */
766         struct hammer_io_list undo_list;        /* dirty undo buffers */
767         struct hammer_io_list data_list;        /* dirty data buffers */
768         struct hammer_io_list alt_data_list;    /* dirty data buffers */
769         struct hammer_io_list meta_list;        /* dirty meta bufs    */
770         struct hammer_io_list lose_list;        /* loose buffers      */
771         int     locked_dirty_space;             /* meta/volu count    */
772         int     io_running_space;
773         int     objid_cache_count;
774         int     error;                          /* critical I/O error */
775         struct krate    krate;                  /* rate limited kprintf */
776         hammer_tid_t    asof;                   /* snapshot mount */
777         hammer_tid_t    next_tid;
778         hammer_tid_t    flush_tid1;             /* flusher tid sequencing */
779         hammer_tid_t    flush_tid2;             /* flusher tid sequencing */
780         int64_t copy_stat_freebigblocks;        /* number of free bigblocks */
781         u_int32_t       undo_seqno;             /* UNDO/REDO FIFO seqno */
782
783         struct netexport export;
784         struct hammer_lock sync_lock;
785         struct hammer_lock free_lock;
786         struct hammer_lock undo_lock;
787         struct hammer_lock blkmap_lock;
788         struct hammer_lock snapshot_lock;
789         struct hammer_lock volume_lock;
790         struct hammer_blockmap  blockmap[HAMMER_MAX_ZONES];
791         struct hammer_undo      undos[HAMMER_MAX_UNDOS];
792         int                     undo_alloc;
793         TAILQ_HEAD(, hammer_undo)  undo_lru_list;
794         TAILQ_HEAD(, hammer_reserve) delay_list;
795         struct hammer_flush_group_list  flush_group_list;
796         hammer_flush_group_t    next_flush_group;
797         TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
798         TAILQ_HEAD(, hammer_reclaim) reclaim_list;
799 };
800
801 typedef struct hammer_mount     *hammer_mount_t;
802
803 #define HAMMER_MOUNT_CRITICAL_ERROR     0x0001
804 #define HAMMER_MOUNT_FLUSH_RECOVERY     0x0002
805
806 struct hammer_sync_info {
807         int error;
808         int waitfor;
809 };
810
811 #endif
812
813 /*
814  * checkspace slop (8MB chunks), higher numbers are more conservative.
815  */
816 #define HAMMER_CHKSPC_REBLOCK   25
817 #define HAMMER_CHKSPC_MIRROR    20
818 #define HAMMER_CHKSPC_WRITE     20
819 #define HAMMER_CHKSPC_CREATE    20
820 #define HAMMER_CHKSPC_REMOVE    10
821 #define HAMMER_CHKSPC_EMERGENCY 0
822
823 #if defined(_KERNEL)
824
825 extern struct vop_ops hammer_vnode_vops;
826 extern struct vop_ops hammer_spec_vops;
827 extern struct vop_ops hammer_fifo_vops;
828 extern struct bio_ops hammer_bioops;
829
830 extern int hammer_debug_io;
831 extern int hammer_debug_general;
832 extern int hammer_debug_debug;
833 extern int hammer_debug_inode;
834 extern int hammer_debug_locks;
835 extern int hammer_debug_btree;
836 extern int hammer_debug_tid;
837 extern int hammer_debug_recover;
838 extern int hammer_debug_recover_faults;
839 extern int hammer_debug_critical;
840 extern int hammer_cluster_enable;
841 extern int hammer_count_fsyncs;
842 extern int hammer_count_inodes;
843 extern int hammer_count_iqueued;
844 extern int hammer_count_reclaiming;
845 extern int hammer_count_records;
846 extern int hammer_count_record_datas;
847 extern int hammer_count_volumes;
848 extern int hammer_count_buffers;
849 extern int hammer_count_nodes;
850 extern int64_t hammer_count_extra_space_used;
851 extern int64_t hammer_stats_btree_lookups;
852 extern int64_t hammer_stats_btree_searches;
853 extern int64_t hammer_stats_btree_inserts;
854 extern int64_t hammer_stats_btree_deletes;
855 extern int64_t hammer_stats_btree_elements;
856 extern int64_t hammer_stats_btree_splits;
857 extern int64_t hammer_stats_btree_iterations;
858 extern int64_t hammer_stats_btree_root_iterations;
859 extern int64_t hammer_stats_record_iterations;
860 extern int64_t hammer_stats_file_read;
861 extern int64_t hammer_stats_file_write;
862 extern int64_t hammer_stats_file_iopsr;
863 extern int64_t hammer_stats_file_iopsw;
864 extern int64_t hammer_stats_disk_read;
865 extern int64_t hammer_stats_disk_write;
866 extern int64_t hammer_stats_inode_flushes;
867 extern int64_t hammer_stats_commits;
868 extern int64_t hammer_stats_undo;
869 extern int64_t hammer_stats_redo;
870 extern int hammer_count_dirtybufspace;
871 extern int hammer_count_refedbufs;
872 extern int hammer_count_reservations;
873 extern int hammer_count_io_running_read;
874 extern int hammer_count_io_running_write;
875 extern int hammer_count_io_locked;
876 extern int hammer_limit_dirtybufspace;
877 extern int hammer_limit_recs;
878 extern int hammer_limit_inode_recs;
879 extern int hammer_limit_reclaim;
880 extern int hammer_bio_count;
881 extern int hammer_verify_zone;
882 extern int hammer_verify_data;
883 extern int hammer_write_mode;
884 extern int hammer_yield_check;
885 extern int hammer_fsync_mode;
886 extern int hammer_autoflush;
887 extern int64_t hammer_contention_count;
888
889 void    hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
890                         int error, const char *msg);
891 int     hammer_vop_inactive(struct vop_inactive_args *);
892 int     hammer_vop_reclaim(struct vop_reclaim_args *);
893 int     hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
894 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
895                         hammer_inode_t dip, int64_t obj_id,
896                         hammer_tid_t asof, u_int32_t localization,
897                         int flags, int *errorp);
898 struct hammer_inode *hammer_get_dummy_inode(hammer_transaction_t trans,
899                         hammer_inode_t dip, int64_t obj_id,
900                         hammer_tid_t asof, u_int32_t localization,
901                         int flags, int *errorp);
902 struct hammer_inode *hammer_find_inode(hammer_transaction_t trans,
903                         int64_t obj_id, hammer_tid_t asof,
904                         u_int32_t localization);
905 void    hammer_scan_inode_snapshots(hammer_mount_t hmp,
906                         hammer_inode_info_t iinfo,
907                         int (*callback)(hammer_inode_t ip, void *data),
908                         void *data);
909 void    hammer_put_inode(struct hammer_inode *ip);
910 void    hammer_put_inode_ref(struct hammer_inode *ip);
911 void    hammer_inode_waitreclaims(hammer_mount_t hmp);
912 void    hammer_inode_waithard(hammer_mount_t hmp);
913
914 int     hammer_unload_volume(hammer_volume_t volume, void *data __unused);
915 int     hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
916
917 int     hammer_unload_buffer(hammer_buffer_t buffer, void *data);
918 int     hammer_install_volume(hammer_mount_t hmp, const char *volname,
919                         struct vnode *devvp);
920 int     hammer_mountcheck_volumes(hammer_mount_t hmp);
921
922 int     hammer_mem_add(hammer_record_t record);
923 int     hammer_ip_lookup(hammer_cursor_t cursor);
924 int     hammer_ip_first(hammer_cursor_t cursor);
925 int     hammer_ip_next(hammer_cursor_t cursor);
926 int     hammer_ip_resolve_data(hammer_cursor_t cursor);
927 int     hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
928                         hammer_tid_t tid);
929 int     hammer_create_at_cursor(hammer_cursor_t cursor,
930                         hammer_btree_leaf_elm_t leaf, void *udata, int mode);
931 int     hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
932                         hammer_tid_t delete_tid, u_int32_t delete_ts,
933                         int track, int64_t *stat_bytes);
934 int     hammer_ip_check_directory_empty(hammer_transaction_t trans,
935                         hammer_inode_t ip);
936 int     hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
937 int     hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
938
939 hammer_record_t
940         hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
941 void    hammer_flush_record_done(hammer_record_t record, int error);
942 void    hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
943 void    hammer_rel_mem_record(hammer_record_t record);
944
945 int     hammer_cursor_up(hammer_cursor_t cursor);
946 int     hammer_cursor_up_locked(hammer_cursor_t cursor);
947 int     hammer_cursor_down(hammer_cursor_t cursor);
948 int     hammer_cursor_upgrade(hammer_cursor_t cursor);
949 int     hammer_cursor_upgrade_node(hammer_cursor_t cursor);
950 void    hammer_cursor_downgrade(hammer_cursor_t cursor);
951 int     hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
952                         int index);
953 void    hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
954 int     hammer_lock_ex_try(struct hammer_lock *lock);
955 void    hammer_lock_sh(struct hammer_lock *lock);
956 int     hammer_lock_sh_try(struct hammer_lock *lock);
957 int     hammer_lock_upgrade(struct hammer_lock *lock);
958 void    hammer_lock_downgrade(struct hammer_lock *lock);
959 int     hammer_lock_status(struct hammer_lock *lock);
960 void    hammer_unlock(struct hammer_lock *lock);
961 void    hammer_ref(struct hammer_lock *lock);
962 void    hammer_unref(struct hammer_lock *lock);
963
964 void    hammer_sync_lock_ex(hammer_transaction_t trans);
965 void    hammer_sync_lock_sh(hammer_transaction_t trans);
966 int     hammer_sync_lock_sh_try(hammer_transaction_t trans);
967 void    hammer_sync_unlock(hammer_transaction_t trans);
968
969 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
970 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
971 void    hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
972 u_int64_t hammer_timespec_to_time(struct timespec *ts);
973 int     hammer_str_to_tid(const char *str, int *ispfsp,
974                         hammer_tid_t *tidp, u_int32_t *localizationp);
975 int     hammer_is_atatext(const char *name, int len);
976 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip,
977                         int64_t namekey);
978 void hammer_clear_objid(hammer_inode_t dip);
979 void hammer_destroy_objid_cache(hammer_mount_t hmp);
980
981 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
982                         int bytes);
983 void hammer_clear_undo_history(hammer_mount_t hmp);
984 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
985 int hammer_get_dtype(u_int8_t obj_type);
986 u_int8_t hammer_get_obj_type(enum vtype vtype);
987 int64_t hammer_directory_namekey(hammer_inode_t dip, const void *name, int len,
988                         u_int32_t *max_iterationsp);
989 int     hammer_nohistory(hammer_inode_t ip);
990
991 int     hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
992                         hammer_node_cache_t cache, hammer_inode_t ip);
993 void    hammer_normalize_cursor(hammer_cursor_t cursor);
994 void    hammer_done_cursor(hammer_cursor_t cursor);
995 int     hammer_recover_cursor(hammer_cursor_t cursor);
996 void    hammer_unlock_cursor(hammer_cursor_t cursor);
997 int     hammer_lock_cursor(hammer_cursor_t cursor);
998 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor);
999 void    hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
1000
1001 void    hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
1002 void    hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
1003                         int index);
1004 void    hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
1005                         int index);
1006 void    hammer_cursor_moved_element(hammer_node_t onode, hammer_node_t nnode,
1007                         int oindex, int nindex);
1008 void    hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent,
1009                         hammer_node_t nparent, int nindex);
1010 void    hammer_cursor_inserted_element(hammer_node_t node, int index);
1011 void    hammer_cursor_deleted_element(hammer_node_t node, int index);
1012
1013 int     hammer_btree_lookup(hammer_cursor_t cursor);
1014 int     hammer_btree_first(hammer_cursor_t cursor);
1015 int     hammer_btree_last(hammer_cursor_t cursor);
1016 int     hammer_btree_extract(hammer_cursor_t cursor, int flags);
1017 int     hammer_btree_iterate(hammer_cursor_t cursor);
1018 int     hammer_btree_iterate_reverse(hammer_cursor_t cursor);
1019 int     hammer_btree_insert(hammer_cursor_t cursor,
1020                             hammer_btree_leaf_elm_t elm, int *doprop);
1021 int     hammer_btree_delete(hammer_cursor_t cursor);
1022 void    hammer_btree_do_propagation(hammer_cursor_t cursor,
1023                             hammer_pseudofs_inmem_t pfsm,
1024                             hammer_btree_leaf_elm_t leaf);
1025 int     hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
1026 int     hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
1027 int     hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
1028 int     hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
1029
1030 int     btree_set_parent(hammer_transaction_t trans, hammer_node_t node,
1031                         hammer_btree_elm_t elm);
1032 void    hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node);
1033 int     hammer_btree_lock_children(hammer_cursor_t cursor, int depth,
1034                         hammer_node_lock_t parent);
1035 void    hammer_btree_lock_copy(hammer_cursor_t cursor,
1036                         hammer_node_lock_t parent);
1037 int     hammer_btree_sync_copy(hammer_cursor_t cursor,
1038                         hammer_node_lock_t parent);
1039 void    hammer_btree_unlock_children(hammer_cursor_t cursor,
1040                         hammer_node_lock_t parent);
1041 int     hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
1042 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans,
1043                         hammer_node_t node, int *parent_indexp,
1044                         int *errorp, int try_exclusive);
1045
1046 void    hammer_print_btree_node(hammer_node_ondisk_t ondisk);
1047 void    hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
1048
1049 void    *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
1050                         int *errorp, struct hammer_buffer **bufferp);
1051 void    *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
1052                         int *errorp, struct hammer_buffer **bufferp);
1053 void    *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1054                         int *errorp, struct hammer_buffer **bufferp);
1055 void    *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1056                         int *errorp, struct hammer_buffer **bufferp);
1057
1058 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
1059
1060 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
1061                         int32_t vol_no, int *errorp);
1062 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
1063                         int bytes, int isnew, int *errorp);
1064 void            hammer_sync_buffers(hammer_mount_t hmp,
1065                         hammer_off_t base_offset, int bytes);
1066 int             hammer_del_buffers(hammer_mount_t hmp,
1067                         hammer_off_t base_offset,
1068                         hammer_off_t zone2_offset, int bytes,
1069                         int report_conflicts);
1070
1071 int             hammer_ref_volume(hammer_volume_t volume);
1072 int             hammer_ref_buffer(hammer_buffer_t buffer);
1073 void            hammer_flush_buffer_nodes(hammer_buffer_t buffer);
1074
1075 void            hammer_rel_volume(hammer_volume_t volume, int flush);
1076 void            hammer_rel_buffer(hammer_buffer_t buffer, int flush);
1077
1078 int             hammer_vfs_export(struct mount *mp, int op,
1079                         const struct export_args *export);
1080 hammer_node_t   hammer_get_node(hammer_transaction_t trans,
1081                         hammer_off_t node_offset, int isnew, int *errorp);
1082 void            hammer_ref_node(hammer_node_t node);
1083 hammer_node_t   hammer_ref_node_safe(hammer_transaction_t trans,
1084                         hammer_node_cache_t cache, int *errorp);
1085 void            hammer_rel_node(hammer_node_t node);
1086 void            hammer_delete_node(hammer_transaction_t trans,
1087                         hammer_node_t node);
1088 void            hammer_cache_node(hammer_node_cache_t cache,
1089                         hammer_node_t node);
1090 void            hammer_uncache_node(hammer_node_cache_t cache);
1091 void            hammer_flush_node(hammer_node_t node);
1092
1093 void hammer_dup_buffer(struct hammer_buffer **bufferp,
1094                         struct hammer_buffer *buffer);
1095 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans,
1096                         hammer_off_t hint, int *errorp);
1097 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1098                         u_int16_t rec_type, hammer_off_t *data_offsetp,
1099                         struct hammer_buffer **data_bufferp,
1100                         hammer_off_t hint, int *errorp);
1101
1102 int hammer_generate_undo(hammer_transaction_t trans,
1103                         hammer_off_t zone1_offset, void *base, int len);
1104 int hammer_generate_redo(hammer_transaction_t trans, hammer_inode_t ip,
1105                         hammer_off_t file_offset, void *base, int len);
1106 int hammer_upgrade_undo_4(hammer_transaction_t trans);
1107
1108 void hammer_put_volume(struct hammer_volume *volume, int flush);
1109 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
1110
1111 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1112                         hammer_off_t owner, int *errorp);
1113 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1114                         hammer_off_t owner, int *errorp);
1115 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp);
1116 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1117                         int bytes, hammer_off_t hint, int *errorp);
1118 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1119                         int bytes, hammer_off_t *zone_offp, int *errorp);
1120 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1121                         hammer_reserve_t resv);
1122 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1123 void hammer_blockmap_free(hammer_transaction_t trans,
1124                         hammer_off_t bmap_off, int bytes);
1125 int hammer_blockmap_finalize(hammer_transaction_t trans,
1126                         hammer_reserve_t resv,
1127                         hammer_off_t bmap_off, int bytes);
1128 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
1129                         int *curp, int *errorp);
1130 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1131                         int *errorp);
1132 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1133                         int *errorp);
1134 int64_t hammer_undo_used(hammer_transaction_t trans);
1135 int64_t hammer_undo_space(hammer_transaction_t trans);
1136 int64_t hammer_undo_max(hammer_mount_t hmp);
1137 int hammer_undo_reclaim(hammer_io_t io);
1138
1139 void hammer_start_transaction(struct hammer_transaction *trans,
1140                               struct hammer_mount *hmp);
1141 void hammer_simple_transaction(struct hammer_transaction *trans,
1142                               struct hammer_mount *hmp);
1143 void hammer_start_transaction_fls(struct hammer_transaction *trans,
1144                                   struct hammer_mount *hmp);
1145 void hammer_done_transaction(struct hammer_transaction *trans);
1146 hammer_tid_t hammer_alloc_tid(hammer_mount_t hmp, int count);
1147
1148 void hammer_modify_inode(hammer_inode_t ip, int flags);
1149 void hammer_flush_inode(hammer_inode_t ip, int flags);
1150 void hammer_flush_inode_done(hammer_inode_t ip, int error);
1151 void hammer_wait_inode(hammer_inode_t ip);
1152
1153 int  hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
1154                         struct ucred *cred, struct hammer_inode *dip,
1155                         const char *name, int namelen,
1156                         hammer_pseudofs_inmem_t pfsm,
1157                         struct hammer_inode **ipp);
1158 void hammer_rel_inode(hammer_inode_t ip, int flush);
1159 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1160 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1161 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1162
1163 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1164 void hammer_test_inode(hammer_inode_t dip);
1165 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1166
1167 int  hammer_ip_add_directory(struct hammer_transaction *trans,
1168                         hammer_inode_t dip, const char *name, int bytes,
1169                         hammer_inode_t nip);
1170 int  hammer_ip_del_directory(struct hammer_transaction *trans,
1171                         hammer_cursor_t cursor, hammer_inode_t dip,
1172                         hammer_inode_t ip);
1173 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1174                         void *data, int bytes, int *errorp);
1175 int  hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
1176 int  hammer_ip_add_record(struct hammer_transaction *trans,
1177                         hammer_record_t record);
1178 int  hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1179                         int64_t ran_beg, int64_t ran_end, int truncating);
1180 int  hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1181                         int *countp);
1182 int  hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1183                         int64_t offset, void *data, int bytes);
1184 int  hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
1185 int  hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1186 hammer_pseudofs_inmem_t  hammer_load_pseudofs(hammer_transaction_t trans,
1187                         u_int32_t localization, int *errorp);
1188 int  hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1189                         hammer_pseudofs_inmem_t pfsm);
1190 int  hammer_save_pseudofs(hammer_transaction_t trans,
1191                         hammer_pseudofs_inmem_t pfsm);
1192 int  hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization);
1193 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1194 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1195                         struct ucred *cred);
1196
1197 void hammer_io_init(hammer_io_t io, hammer_volume_t volume,
1198                         enum hammer_io_type type);
1199 int hammer_io_read(struct vnode *devvp, struct hammer_io *io,
1200                         hammer_off_t limit);
1201 void hammer_io_advance(struct hammer_io *io);
1202 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
1203 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1204 struct buf *hammer_io_release(struct hammer_io *io, int flush);
1205 void hammer_io_flush(struct hammer_io *io, int reclaim);
1206 void hammer_io_wait(struct hammer_io *io);
1207 void hammer_io_waitdep(struct hammer_io *io);
1208 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident);
1209 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1210                         hammer_btree_leaf_elm_t leaf);
1211 int hammer_io_direct_write(hammer_mount_t hmp, hammer_record_t record,
1212                         struct bio *bio);
1213 void hammer_io_direct_wait(hammer_record_t record);
1214 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1215 void hammer_io_write_interlock(hammer_io_t io);
1216 void hammer_io_done_interlock(hammer_io_t io);
1217 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1218 void hammer_io_clear_modlist(struct hammer_io *io);
1219 void hammer_io_flush_sync(hammer_mount_t hmp);
1220 void hammer_io_clear_error(struct hammer_io *io);
1221
1222 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1223                         void *base, int len);
1224 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1225                         void *base, int len);
1226 void hammer_modify_volume_done(hammer_volume_t volume);
1227 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1228
1229 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1230                         struct hammer_ioc_reblock *reblock);
1231 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip,
1232                         struct hammer_ioc_rebalance *rebal);
1233 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1234                         struct hammer_ioc_prune *prune);
1235 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1236                         struct hammer_ioc_mirror_rw *mirror);
1237 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1238                         struct hammer_ioc_mirror_rw *mirror);
1239 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1240                         struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1241 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1242                         struct hammer_ioc_pseudofs_rw *pfs);
1243 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1244                         struct hammer_ioc_pseudofs_rw *pfs);
1245 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1246                         struct hammer_ioc_pseudofs_rw *pfs);
1247 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1248                         struct hammer_ioc_pseudofs_rw *pfs);
1249 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1250                         struct hammer_ioc_pseudofs_rw *pfs);
1251 int hammer_ioc_volume_add(hammer_transaction_t trans, hammer_inode_t ip,
1252                         struct hammer_ioc_volume *ioc);
1253 int hammer_ioc_volume_del(hammer_transaction_t trans, hammer_inode_t ip,
1254                         struct hammer_ioc_volume *ioc);
1255
1256 int hammer_signal_check(hammer_mount_t hmp);
1257
1258 void hammer_flusher_create(hammer_mount_t hmp);
1259 void hammer_flusher_destroy(hammer_mount_t hmp);
1260 void hammer_flusher_sync(hammer_mount_t hmp);
1261 int  hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1262 int  hammer_flusher_async_one(hammer_mount_t hmp);
1263 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1264 void hammer_flusher_wait_next(hammer_mount_t hmp);
1265 int  hammer_flusher_meta_limit(hammer_mount_t hmp);
1266 int  hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1267 int  hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1268 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1269 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1270 int  hammer_flusher_haswork(hammer_mount_t hmp);
1271 void hammer_flusher_flush_undos(hammer_mount_t hmp, int already_flushed);
1272
1273 int hammer_recover_stage1(hammer_mount_t hmp, hammer_volume_t rootvol);
1274 int hammer_recover_stage2(hammer_mount_t hmp, hammer_volume_t rootvol);
1275 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1276                         hammer_volume_t root_volume, int final);
1277
1278 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1279 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1280 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1281
1282 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1283 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1284 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1285 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1286 void hkprintf(const char *ctl, ...);
1287 udev_t hammer_fsid_to_udev(uuid_t *uuid);
1288
1289
1290 int hammer_blocksize(int64_t file_offset);
1291 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1292
1293 /*
1294  * Shortcut for _hammer_checkspace(), used all over the code.
1295  */
1296 static __inline int
1297 hammer_checkspace(hammer_mount_t hmp, int slop)
1298 {
1299         return(_hammer_checkspace(hmp, slop, NULL));
1300 }
1301
1302 #endif
1303
1304 static __inline void
1305 hammer_wait_mem_record(hammer_record_t record)
1306 {
1307         hammer_wait_mem_record_ident(record, "hmmwai");
1308 }
1309
1310 static __inline void
1311 hammer_lock_ex(struct hammer_lock *lock)
1312 {
1313         hammer_lock_ex_ident(lock, "hmrlck");
1314 }
1315
1316 /*
1317  * Indicate that a B-Tree node is being modified.
1318  */
1319 static __inline void
1320 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1321 {
1322         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1323         hammer_modify_buffer(trans, node->buffer, NULL, 0);
1324 }
1325
1326 static __inline void
1327 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1328 {
1329         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1330         hammer_modify_buffer(trans, node->buffer,
1331                              node->ondisk, sizeof(*node->ondisk));
1332 }
1333
1334 static __inline void
1335 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1336                    void *base, int len)
1337 {
1338         hammer_crc_t *crcptr;
1339
1340         KKASSERT((char *)base >= (char *)node->ondisk &&
1341                  (char *)base + len <=
1342                     (char *)node->ondisk + sizeof(*node->ondisk));
1343         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1344         hammer_modify_buffer(trans, node->buffer, base, len);
1345         crcptr = &node->ondisk->crc;
1346         hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t));
1347         --node->buffer->io.modify_refs; /* only want one ref */
1348 }
1349
1350 /*
1351  * Indicate that the specified modifications have been completed.
1352  *
1353  * Do not try to generate the crc here, it's very expensive to do and a
1354  * sequence of insertions or deletions can result in many calls to this
1355  * function on the same node.
1356  */
1357 static __inline void
1358 hammer_modify_node_done(hammer_node_t node)
1359 {
1360         node->flags |= HAMMER_NODE_CRCGOOD;
1361         if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1362                 node->flags |= HAMMER_NODE_NEEDSCRC;
1363                 node->buffer->io.gencrc = 1;
1364                 hammer_ref_node(node);
1365         }
1366         hammer_modify_buffer_done(node->buffer);
1367 }
1368
1369 #define hammer_modify_volume_field(trans, vol, field)           \
1370         hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1371                              sizeof((vol)->ondisk->field))
1372
1373 #define hammer_modify_node_field(trans, node, field)            \
1374         hammer_modify_node(trans, node, &(node)->ondisk->field, \
1375                              sizeof((node)->ondisk->field))
1376
1377 /*
1378  * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1379  * created directories for HAMMER version 2 or greater and causes
1380  * directory entries to be placed the inode localization zone in
1381  * the B-Tree instead of the misc zone.
1382  *
1383  * This greatly improves localization between directory entries and
1384  * inodes
1385  */
1386 static __inline u_int32_t
1387 hammer_dir_localization(hammer_inode_t dip)
1388 {
1389         if (dip->ino_data.cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO)
1390                 return(HAMMER_LOCALIZE_INODE);
1391         else
1392                 return(HAMMER_LOCALIZE_MISC);
1393 }