kernel - Fix excessive mbuf use in nfs_realign()
[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 #include <sys/limits.h>
61 #include <vm/vm_extern.h>
62
63 #include <sys/buf2.h>
64 #include <sys/signal2.h>
65 #include <sys/mplock2.h>
66 #include "hammer_disk.h"
67 #include "hammer_mount.h"
68 #include "hammer_ioctl.h"
69
70 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
71
72 MALLOC_DECLARE(M_HAMMER);
73
74 /*
75  * Kernel trace
76  */
77 #if !defined(KTR_HAMMER)
78 #define KTR_HAMMER      KTR_ALL
79 #endif
80 KTR_INFO_MASTER_EXTERN(hammer);
81
82 /*
83  * Misc structures
84  */
85 struct hammer_mount;
86
87 /*
88  * Key structure used for custom RB tree inode lookups.  This prototypes
89  * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
90  */
91 typedef struct hammer_inode_info {
92         int64_t         obj_id;         /* (key) object identifier */
93         hammer_tid_t    obj_asof;       /* (key) snapshot transid or 0 */
94         u_int32_t       obj_localization; /* (key) pseudo-fs */
95         union {
96                 struct hammer_btree_leaf_elm *leaf;
97         } u;
98 } *hammer_inode_info_t;
99
100 typedef enum hammer_transaction_type {
101         HAMMER_TRANS_RO,
102         HAMMER_TRANS_STD,
103         HAMMER_TRANS_FLS
104 } hammer_transaction_type_t;
105
106 /*
107  * HAMMER Transaction tracking
108  */
109 struct hammer_transaction {
110         hammer_transaction_type_t type;
111         struct hammer_mount *hmp;
112         hammer_tid_t    tid;
113         u_int64_t       time;
114         u_int32_t       time32;
115         int             sync_lock_refs;
116         int             flags;
117         struct hammer_volume *rootvol;
118 };
119
120 typedef struct hammer_transaction *hammer_transaction_t;
121
122 #define HAMMER_TRANSF_NEWINODE  0x0001
123 #define HAMMER_TRANSF_DIDIO     0x0002
124 #define HAMMER_TRANSF_CRCDOM    0x0004  /* EDOM on CRC error, less critical */
125
126 /*
127  * HAMMER locks
128  */
129 struct hammer_lock {
130         volatile u_int  refs;           /* active references */
131         volatile u_int  lockval;        /* lock count and control bits */
132         struct thread   *lowner;        /* owner if exclusively held */
133         struct thread   *rowner;        /* owner if exclusively held */
134 };
135
136 #define HAMMER_REFS_LOCKED      0x40000000      /* transition check */
137 #define HAMMER_REFS_WANTED      0x20000000      /* transition check */
138 #define HAMMER_REFS_CHECK       0x10000000      /* transition check */
139
140 #define HAMMER_REFS_FLAGS       (HAMMER_REFS_LOCKED | \
141                                  HAMMER_REFS_WANTED | \
142                                  HAMMER_REFS_CHECK)
143
144 #define HAMMER_LOCKF_EXCLUSIVE  0x40000000
145 #define HAMMER_LOCKF_WANTED     0x20000000
146
147 static __inline int
148 hammer_notlocked(struct hammer_lock *lock)
149 {
150         return(lock->lockval == 0);
151 }
152
153 static __inline int
154 hammer_islocked(struct hammer_lock *lock)
155 {
156         return(lock->lockval != 0);
157 }
158
159 /*
160  * Returns the number of refs on the object.
161  */
162 static __inline int
163 hammer_isactive(struct hammer_lock *lock)
164 {
165         return(lock->refs & ~HAMMER_REFS_FLAGS);
166 }
167
168 static __inline int
169 hammer_oneref(struct hammer_lock *lock)
170 {
171         return((lock->refs & ~HAMMER_REFS_FLAGS) == 1);
172 }
173
174 static __inline int
175 hammer_norefs(struct hammer_lock *lock)
176 {
177         return((lock->refs & ~HAMMER_REFS_FLAGS) == 0);
178 }
179
180 static __inline int
181 hammer_norefsorlock(struct hammer_lock *lock)
182 {
183         return(lock->refs == 0);
184 }
185
186 static __inline int
187 hammer_refsorlock(struct hammer_lock *lock)
188 {
189         return(lock->refs != 0);
190 }
191
192 /*
193  * Return if we specifically own the lock exclusively.
194  */
195 static __inline int
196 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
197 {
198         if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) &&
199             lock->lowner == td) {
200                 return(1);
201         }
202         return(0);
203 }
204
205 /*
206  * Flush state, used by various structures
207  */
208 typedef enum hammer_inode_state {
209         HAMMER_FST_IDLE,
210         HAMMER_FST_SETUP,
211         HAMMER_FST_FLUSH
212 } hammer_inode_state_t;
213
214 TAILQ_HEAD(hammer_record_list, hammer_record);
215
216 /*
217  * Pseudo-filesystem extended data tracking
218  */
219 struct hammer_pfs_rb_tree;
220 struct hammer_pseudofs_inmem;
221 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
222 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
223               hammer_pfs_rb_compare, u_int32_t);
224
225 struct hammer_pseudofs_inmem {
226         RB_ENTRY(hammer_pseudofs_inmem) rb_node;
227         struct hammer_lock      lock;
228         u_int32_t               localization;
229         hammer_tid_t            create_tid;
230         int                     flags;
231         udev_t                  fsid_udev;
232         struct hammer_pseudofs_data pfsd;
233 };
234
235 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t;
236
237 #define HAMMER_PFSM_DELETED     0x0001
238
239 /*
240  * Cache object ids.  A fixed number of objid cache structures are
241  * created to reserve object id's for newly created files in multiples
242  * of 100,000, localized to a particular directory, and recycled as
243  * needed.  This allows parallel create operations in different
244  * directories to retain fairly localized object ids which in turn
245  * improves reblocking performance and layout.
246  */
247 #define OBJID_CACHE_SIZE        1024
248 #define OBJID_CACHE_BULK_BITS   10              /* 10 bits (1024)       */
249 #define OBJID_CACHE_BULK        (32 * 32)       /* two level (1024)     */
250 #define OBJID_CACHE_BULK_MASK   (OBJID_CACHE_BULK - 1)
251 #define OBJID_CACHE_BULK_MASK64 ((u_int64_t)(OBJID_CACHE_BULK - 1))
252
253 typedef struct hammer_objid_cache {
254         TAILQ_ENTRY(hammer_objid_cache) entry;
255         struct hammer_inode             *dip;
256         hammer_tid_t                    base_tid;
257         int                             count;
258         u_int32_t                       bm0;
259         u_int32_t                       bm1[32];
260 } *hammer_objid_cache_t;
261
262 /*
263  * Associate an inode with a B-Tree node to cache search start positions
264  */
265 typedef struct hammer_node_cache {
266         TAILQ_ENTRY(hammer_node_cache)  entry;
267         struct hammer_node              *node;
268         struct hammer_inode             *ip;
269 } *hammer_node_cache_t;
270
271 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
272
273 /*
274  * Structure used to organize flush groups.  Flush groups must be
275  * organized into chunks in order to avoid blowing out the UNDO FIFO.
276  * Without this a 'sync' could end up flushing 50,000 inodes in a single
277  * transaction.
278  */
279 struct hammer_fls_rb_tree;
280 RB_HEAD(hammer_fls_rb_tree, hammer_inode);
281 RB_PROTOTYPE(hammer_fls_rb_tree, hammer_inode, rb_flsnode,
282               hammer_ino_rb_compare);
283
284 struct hammer_flush_group {
285         TAILQ_ENTRY(hammer_flush_group) flush_entry;
286         struct hammer_fls_rb_tree       flush_tree;
287         int                             unused01;
288         int                             total_count;    /* record load */
289         int                             running;        /* group is running */
290         int                             closed;
291         int                             refs;
292 };
293
294 typedef struct hammer_flush_group *hammer_flush_group_t;
295
296 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
297
298 /*
299  * Structure used to represent an inode in-memory.
300  *
301  * The record and data associated with an inode may be out of sync with
302  * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
303  * clear).
304  *
305  * An inode may also hold a cache of unsynchronized records, used for
306  * database and directories only.  Unsynchronized regular file data is
307  * stored in the buffer cache.
308  *
309  * NOTE: A file which is created and destroyed within the initial
310  * synchronization period can wind up not doing any disk I/O at all.
311  *
312  * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
313  */
314 struct hammer_ino_rb_tree;
315 struct hammer_inode;
316 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
317 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
318               hammer_ino_rb_compare, hammer_inode_info_t);
319
320 struct hammer_redo_rb_tree;
321 RB_HEAD(hammer_redo_rb_tree, hammer_inode);
322 RB_PROTOTYPE2(hammer_redo_rb_tree, hammer_inode, rb_redonode,
323               hammer_redo_rb_compare, hammer_off_t);
324
325 struct hammer_rec_rb_tree;
326 struct hammer_record;
327 RB_HEAD(hammer_rec_rb_tree, hammer_record);
328 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
329               hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
330
331 TAILQ_HEAD(hammer_node_list, hammer_node);
332
333 struct hammer_inode {
334         RB_ENTRY(hammer_inode)  rb_node;
335         hammer_inode_state_t    flush_state;
336         hammer_flush_group_t    flush_group;
337         RB_ENTRY(hammer_inode)  rb_flsnode;     /* when on flush list */
338         RB_ENTRY(hammer_inode)  rb_redonode;    /* when INODE_RDIRTY is set */
339         struct hammer_record_list target_list;  /* target of dependant recs */
340         int64_t                 obj_id;         /* (key) object identifier */
341         hammer_tid_t            obj_asof;       /* (key) snapshot or 0 */
342         u_int32_t               obj_localization; /* (key) pseudo-fs */
343         struct hammer_mount     *hmp;
344         hammer_objid_cache_t    objid_cache;
345         int                     flags;
346         int                     error;          /* flush error */
347         int                     cursor_ip_refs; /* sanity */
348         int                     rsv_recs;
349         struct vnode            *vp;
350         hammer_pseudofs_inmem_t pfsm;
351         struct lockf            advlock;
352         struct hammer_lock      lock;           /* sync copy interlock */
353         off_t                   trunc_off;
354         struct hammer_btree_leaf_elm ino_leaf;  /* in-memory cache */
355         struct hammer_inode_data ino_data;      /* in-memory cache */
356         struct hammer_rec_rb_tree rec_tree;     /* in-memory cache */
357         int                     rec_generation;
358         struct hammer_node_cache cache[4];      /* search initiate cache */
359
360         /*
361          * When a demark is created to synchronize an inode to
362          * disk, certain fields are copied so the front-end VOPs
363          * can continue to run in parallel with the synchronization
364          * occuring in the background.
365          */
366         int             sync_flags;             /* to-sync flags cache */
367         off_t           sync_trunc_off;         /* to-sync truncation */
368         off_t           save_trunc_off;         /* write optimization */
369         struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
370         struct hammer_inode_data sync_ino_data; /* to-sync cache */
371         size_t          redo_count;
372
373         /*
374          * Track the earliest offset in the UNDO/REDO FIFO containing
375          * REDO records.  This is staged to the backend during flush
376          * sequences.  While the inode is staged redo_fifo_next is used
377          * to track the earliest offset for rotation into redo_fifo_start
378          * on completion of the flush.
379          */
380         hammer_off_t    redo_fifo_start;
381         hammer_off_t    redo_fifo_next;
382 };
383
384 typedef struct hammer_inode *hammer_inode_t;
385
386 #define VTOI(vp)        ((struct hammer_inode *)(vp)->v_data)
387
388 /*
389  * NOTE: DDIRTY does not include atime or mtime and does not include
390  *       write-append size changes.  SDIRTY handles write-append size
391  *       changes.
392  *
393  *       REDO indicates that REDO logging is active, creating a definitive
394  *       stream of REDO records in the UNDO/REDO log for writes and
395  *       truncations, including boundary records when/if REDO is turned off.
396  *       REDO is typically enabled by fsync() and turned off if excessive
397  *       writes without an fsync() occurs.
398  *
399  *       RDIRTY indicates that REDO records were laid down in the UNDO/REDO
400  *       FIFO (even if REDO is turned off some might still be active) and
401  *       still being tracked for this inode.  See hammer_redo.c
402  */
403                                         /* (not including atime/mtime) */
404 #define HAMMER_INODE_DDIRTY     0x0001  /* in-memory ino_data is dirty */
405 #define HAMMER_INODE_RSV_INODES 0x0002  /* hmp->rsv_inodes bumped */
406 #define HAMMER_INODE_CONN_DOWN  0x0004  /* include in downward recursion */
407 #define HAMMER_INODE_XDIRTY     0x0008  /* in-memory records */
408 #define HAMMER_INODE_ONDISK     0x0010  /* inode is on-disk (else not yet) */
409 #define HAMMER_INODE_FLUSH      0x0020  /* flush on last ref */
410 #define HAMMER_INODE_DELETED    0x0080  /* inode delete (backend) */
411 #define HAMMER_INODE_DELONDISK  0x0100  /* delete synchronized to disk */
412 #define HAMMER_INODE_RO         0x0200  /* read-only (because of as-of) */
413 #define HAMMER_INODE_UNUSED0400 0x0400
414 #define HAMMER_INODE_DONDISK    0x0800  /* data records may be on disk */
415 #define HAMMER_INODE_BUFS       0x1000  /* dirty high level bps present */
416 #define HAMMER_INODE_REFLUSH    0x2000  /* flush on dependancy / reflush */
417 #define HAMMER_INODE_RECLAIM    0x4000  /* trying to reclaim */
418 #define HAMMER_INODE_FLUSHW     0x8000  /* Someone waiting for flush */
419
420 #define HAMMER_INODE_TRUNCATED  0x00010000
421 #define HAMMER_INODE_DELETING   0x00020000 /* inode delete request (frontend)*/
422 #define HAMMER_INODE_RESIGNAL   0x00040000 /* re-signal on re-flush */
423 #define HAMMER_INODE_ATIME      0x00100000 /* in-memory atime modified */
424 #define HAMMER_INODE_MTIME      0x00200000 /* in-memory mtime modified */
425 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
426 #define HAMMER_INODE_DUMMY      0x00800000 /* dummy inode covering bad file */
427 #define HAMMER_INODE_SDIRTY     0x01000000 /* in-memory ino_data.size is dirty*/
428 #define HAMMER_INODE_REDO       0x02000000 /* REDO logging active */
429 #define HAMMER_INODE_RDIRTY     0x04000000 /* REDO records active in fifo */
430
431 #define HAMMER_INODE_MODMASK    (HAMMER_INODE_DDIRTY|HAMMER_INODE_SDIRTY|   \
432                                  HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS|     \
433                                  HAMMER_INODE_ATIME|HAMMER_INODE_MTIME|     \
434                                  HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
435
436 #define HAMMER_INODE_MODMASK_NOXDIRTY   \
437                                 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
438
439 #define HAMMER_INODE_MODMASK_NOREDO     \
440                                 (HAMMER_INODE_DDIRTY|                       \
441                                  HAMMER_INODE_XDIRTY|                       \
442                                  HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
443
444 #define HAMMER_FLUSH_SIGNAL     0x0001
445 #define HAMMER_FLUSH_RECURSION  0x0002
446
447 /*
448  * Used by the inode reclaim code to pipeline reclaims and avoid
449  * blowing out kernel memory or letting the flusher get too far
450  * behind.  The reclaim wakes up when count reaches 0 or the
451  * timer expires.
452  */
453 struct hammer_reclaim {
454         TAILQ_ENTRY(hammer_reclaim) entry;
455         int     count;
456 };
457
458 #define HAMMER_RECLAIM_WAIT     4000    /* default vfs.hammer.limit_reclaim */
459
460 /*
461  * Track who is creating the greatest burden on the
462  * inode cache.
463  */
464 struct hammer_inostats {
465         pid_t           pid;    /* track user process */
466         int             ltick;  /* last tick */
467         int             count;  /* count (degenerates) */
468 };
469
470 #define HAMMER_INOSTATS_HSIZE   32
471 #define HAMMER_INOSTATS_HMASK   (HAMMER_INOSTATS_HSIZE - 1)
472
473 /*
474  * Structure used to represent an unsynchronized record in-memory.  These
475  * records typically represent directory entries.  Only non-historical
476  * records are kept in-memory.
477  *
478  * Records are organized as a per-inode RB-Tree.  If the inode is not
479  * on disk then neither are any records and the in-memory record tree
480  * represents the entire contents of the inode.  If the inode is on disk
481  * then the on-disk B-Tree is scanned in parallel with the in-memory
482  * RB-Tree to synthesize the current state of the file.
483  *
484  * Records are also used to enforce the ordering of directory create/delete
485  * operations.  A new inode will not be flushed to disk unless its related
486  * directory entry is also being flushed at the same time.  A directory entry
487  * will not be removed unless its related inode is also being removed at the
488  * same time.
489  */
490 typedef enum hammer_record_type {
491         HAMMER_MEM_RECORD_GENERAL,      /* misc record */
492         HAMMER_MEM_RECORD_INODE,        /* inode record */
493         HAMMER_MEM_RECORD_ADD,          /* positive memory cache record */
494         HAMMER_MEM_RECORD_DEL,          /* negative delete-on-disk record */
495         HAMMER_MEM_RECORD_DATA          /* bulk-data record w/on-disk ref */
496 } hammer_record_type_t;
497
498 struct hammer_record {
499         RB_ENTRY(hammer_record)         rb_node;
500         TAILQ_ENTRY(hammer_record)      target_entry;
501         hammer_inode_state_t            flush_state;
502         hammer_flush_group_t            flush_group;
503         hammer_record_type_t            type;
504         struct hammer_lock              lock;
505         struct hammer_reserve           *resv;
506         struct hammer_inode             *ip;
507         struct hammer_inode             *target_ip;
508         struct hammer_btree_leaf_elm    leaf;
509         union hammer_data_ondisk        *data;
510         int                             flags;
511         hammer_off_t                    zone2_offset;   /* direct-write only */
512 };
513
514 typedef struct hammer_record *hammer_record_t;
515
516 /*
517  * Record flags.  Note that FE can only be set by the frontend if the
518  * record has not been interlocked by the backend w/ BE.
519  */
520 #define HAMMER_RECF_ALLOCDATA           0x0001
521 #define HAMMER_RECF_ONRBTREE            0x0002
522 #define HAMMER_RECF_DELETED_FE          0x0004  /* deleted (frontend) */
523 #define HAMMER_RECF_DELETED_BE          0x0008  /* deleted (backend) */
524 #define HAMMER_RECF_COMMITTED           0x0010  /* committed to the B-Tree */
525 #define HAMMER_RECF_INTERLOCK_BE        0x0020  /* backend interlock */
526 #define HAMMER_RECF_WANTED              0x0040  /* wanted by the frontend */
527 #define HAMMER_RECF_CONVERT_DELETE      0x0100  /* special case */
528 #define HAMMER_RECF_DIRECT_IO           0x0200  /* related direct I/O running*/
529 #define HAMMER_RECF_DIRECT_WAIT         0x0400  /* related direct I/O running*/
530 #define HAMMER_RECF_DIRECT_INVAL        0x0800  /* buffer alias invalidation */
531 #define HAMMER_RECF_REDO                0x1000  /* REDO was laid down */
532
533 /*
534  * hammer_create_at_cursor() and hammer_delete_at_cursor() flags.
535  */
536 #define HAMMER_CREATE_MODE_UMIRROR      0x0001
537 #define HAMMER_CREATE_MODE_SYS          0x0002
538
539 #define HAMMER_DELETE_ADJUST            0x0001
540 #define HAMMER_DELETE_DESTROY           0x0002
541
542 /*
543  * In-memory structures representing on-disk structures.
544  */
545 struct hammer_volume;
546 struct hammer_buffer;
547 struct hammer_node;
548 struct hammer_undo;
549 struct hammer_reserve;
550
551 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
552 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
553 RB_HEAD(hammer_nod_rb_tree, hammer_node);
554 RB_HEAD(hammer_und_rb_tree, hammer_undo);
555 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
556
557 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
558               hammer_vol_rb_compare, int32_t);
559 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
560               hammer_buf_rb_compare, hammer_off_t);
561 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
562               hammer_nod_rb_compare, hammer_off_t);
563 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
564               hammer_und_rb_compare, hammer_off_t);
565 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
566               hammer_res_rb_compare, hammer_off_t);
567
568 /*
569  * IO management - embedded at the head of various in-memory structures
570  *
571  * VOLUME       - hammer_volume containing meta-data
572  * META_BUFFER  - hammer_buffer containing meta-data
573  * DATA_BUFFER  - hammer_buffer containing pure-data
574  *
575  * Dirty volume headers and dirty meta-data buffers are locked until the
576  * flusher can sequence them out.  Dirty pure-data buffers can be written.
577  * Clean buffers can be passively released.
578  */
579 typedef enum hammer_io_type {
580         HAMMER_STRUCTURE_VOLUME,
581         HAMMER_STRUCTURE_META_BUFFER,
582         HAMMER_STRUCTURE_UNDO_BUFFER,
583         HAMMER_STRUCTURE_DATA_BUFFER,
584         HAMMER_STRUCTURE_DUMMY
585 } hammer_io_type_t;
586
587 union hammer_io_structure;
588 struct hammer_io;
589
590 struct worklist {
591         LIST_ENTRY(worklist) node;
592 };
593
594 TAILQ_HEAD(hammer_io_list, hammer_io);
595 typedef struct hammer_io_list *hammer_io_list_t;
596
597 struct hammer_io {
598         struct worklist         worklist;
599         struct hammer_lock      lock;
600         enum hammer_io_type     type;
601         struct hammer_mount     *hmp;
602         struct hammer_volume    *volume;
603         TAILQ_ENTRY(hammer_io)  mod_entry; /* list entry if modified */
604         TAILQ_ENTRY(hammer_io)  iorun_entry; /* iorun_list */
605         hammer_io_list_t        mod_list;
606         struct buf              *bp;
607         int64_t                 offset;    /* zone-2 offset */
608         int                     bytes;     /* buffer cache buffer size */
609         int                     modify_refs;
610
611         u_int           modified : 1;   /* bp's data was modified */
612         u_int           released : 1;   /* bp released (w/ B_LOCKED set) */
613         u_int           running : 1;    /* bp write IO in progress */
614         u_int           waiting : 1;    /* someone is waiting on us */
615         u_int           validated : 1;  /* ondisk has been validated */
616         u_int           waitdep : 1;    /* flush waits for dependancies */
617         u_int           recovered : 1;  /* has recovery ref */
618         u_int           waitmod : 1;    /* waiting for modify_refs */
619         u_int           reclaim : 1;    /* reclaim requested */
620         u_int           gencrc : 1;     /* crc needs to be generated */
621         u_int           ioerror : 1;    /* abort on io-error */
622 };
623
624 typedef struct hammer_io *hammer_io_t;
625
626 #define HAMMER_CLUSTER_SIZE     (64 * 1024)
627 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
628 #undef  HAMMER_CLUSTER_SIZE
629 #define HAMMER_CLUSTER_SIZE     MAXBSIZE
630 #endif
631 #define HAMMER_CLUSTER_BUFS     (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
632
633 /*
634  * In-memory volume representing on-disk buffer
635  */
636 struct hammer_volume {
637         struct hammer_io io;
638         RB_ENTRY(hammer_volume) rb_node;
639         struct hammer_volume_ondisk *ondisk;
640         int32_t vol_no;
641         int64_t nblocks;        /* note: special calculation for statfs */
642         int64_t buffer_base;    /* base offset of buffer 0 */
643         hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
644         hammer_off_t maxraw_off; /* Maximum raw offset for device */
645         char    *vol_name;
646         struct vnode *devvp;
647         int     vol_flags;
648 };
649
650 typedef struct hammer_volume *hammer_volume_t;
651
652 /*
653  * In-memory buffer (other then volume, super-cluster, or cluster),
654  * representing an on-disk buffer.
655  */
656 struct hammer_buffer {
657         struct hammer_io io;
658         RB_ENTRY(hammer_buffer) rb_node;
659         void *ondisk;
660         hammer_off_t zoneX_offset;
661         hammer_off_t zone2_offset;
662         struct hammer_reserve *resv;
663         struct hammer_node_list clist;
664 };
665
666 typedef struct hammer_buffer *hammer_buffer_t;
667
668 /*
669  * In-memory B-Tree node, representing an on-disk B-Tree node.
670  *
671  * This is a hang-on structure which is backed by a hammer_buffer,
672  * indexed by a hammer_cluster, and used for fine-grained locking of
673  * B-Tree nodes in order to properly control lock ordering.  A hammer_buffer
674  * can contain multiple nodes representing wildly disassociated portions
675  * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
676  *
677  * This structure uses a cluster-relative index to reduce the number
678  * of layers required to access it, and also because all on-disk B-Tree
679  * references are cluster-relative offsets.
680  */
681 struct hammer_node {
682         struct hammer_lock      lock;           /* node-by-node lock */
683         TAILQ_ENTRY(hammer_node) entry;         /* per-buffer linkage */
684         RB_ENTRY(hammer_node)   rb_node;        /* per-cluster linkage */
685         hammer_off_t            node_offset;    /* full offset spec */
686         struct hammer_mount     *hmp;
687         struct hammer_buffer    *buffer;        /* backing buffer */
688         hammer_node_ondisk_t    ondisk;         /* ptr to on-disk structure */
689         TAILQ_HEAD(, hammer_cursor) cursor_list;  /* deadlock recovery */
690         struct hammer_node_cache_list cache_list; /* passive caches */
691         int                     flags;
692 };
693
694 #define HAMMER_NODE_DELETED     0x0001
695 #define HAMMER_NODE_FLUSH       0x0002
696 #define HAMMER_NODE_CRCGOOD     0x0004
697 #define HAMMER_NODE_NEEDSCRC    0x0008
698 #define HAMMER_NODE_NEEDSMIRROR 0x0010
699 #define HAMMER_NODE_CRCBAD      0x0020
700 #define HAMMER_NODE_NONLINEAR   0x0040          /* linear heuristic */
701
702 #define HAMMER_NODE_CRCANY      (HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD)
703
704 typedef struct hammer_node      *hammer_node_t;
705
706 /*
707  * List of locked nodes.  This structure is used to lock potentially large
708  * numbers of nodes as an aid for complex B-Tree operations.
709  */
710 struct hammer_node_lock;
711 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock);
712
713 struct hammer_node_lock {
714         TAILQ_ENTRY(hammer_node_lock) entry;
715         struct hammer_node_lock_list  list;
716         struct hammer_node_lock       *parent;
717         hammer_node_t   node;
718         hammer_node_ondisk_t copy;      /* copy of on-disk data */
719         int             index;          /* index of this node in parent */
720         int             count;          /* count children */
721         int             flags;
722 };
723
724 typedef struct hammer_node_lock *hammer_node_lock_t;
725
726 #define HAMMER_NODE_LOCK_UPDATED        0x0001
727 #define HAMMER_NODE_LOCK_LCACHE         0x0002
728
729 /*
730  * Common I/O management structure - embedded in in-memory structures
731  * which are backed by filesystem buffers.
732  */
733 union hammer_io_structure {
734         struct hammer_io        io;
735         struct hammer_volume    volume;
736         struct hammer_buffer    buffer;
737 };
738
739 typedef union hammer_io_structure *hammer_io_structure_t;
740
741 /*
742  * The reserve structure prevents the blockmap from allocating
743  * out of a reserved bigblock.  Such reservations are used by
744  * the direct-write mechanism.
745  *
746  * The structure is also used to hold off on reallocations of
747  * big blocks from the freemap until flush dependancies have
748  * been dealt with.
749  */
750 struct hammer_reserve {
751         RB_ENTRY(hammer_reserve) rb_node;
752         TAILQ_ENTRY(hammer_reserve) delay_entry;
753         int             flush_group;
754         int             flags;
755         int             refs;
756         int             zone;
757         int             append_off;
758         hammer_off_t    zone_offset;
759 };
760
761 typedef struct hammer_reserve *hammer_reserve_t;
762
763 #define HAMMER_RESF_ONDELAY     0x0001
764 #define HAMMER_RESF_LAYER2FREE  0x0002
765
766 #include "hammer_cursor.h"
767
768 /*
769  * The undo structure tracks recent undos to avoid laying down duplicate
770  * undos within a flush group, saving us a significant amount of overhead.
771  *
772  * This is strictly a heuristic.
773  */
774 #define HAMMER_MAX_UNDOS                1024
775 #define HAMMER_MAX_FLUSHERS             4
776
777 struct hammer_undo {
778         RB_ENTRY(hammer_undo)   rb_node;
779         TAILQ_ENTRY(hammer_undo) lru_entry;
780         hammer_off_t            offset;
781         int                     bytes;
782 };
783
784 typedef struct hammer_undo *hammer_undo_t;
785
786 struct hammer_flusher_info;
787 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
788
789 struct hammer_flusher {
790         int             signal;         /* flusher thread sequencer */
791         int             act;            /* currently active flush group */
792         int             done;           /* set to act when complete */
793         int             next;           /* next flush group */
794         int             group_lock;     /* lock sequencing of the next flush */
795         int             exiting;        /* request master exit */
796         thread_t        td;             /* master flusher thread */
797         hammer_tid_t    tid;            /* last flushed transaction id */
798         int             finalize_want;          /* serialize finalization */
799         struct hammer_lock finalize_lock;       /* serialize finalization */
800         struct hammer_transaction trans;        /* shared transaction */
801         struct hammer_flusher_info_list run_list;
802         struct hammer_flusher_info_list ready_list;
803 };
804
805 #define HAMMER_FLUSH_UNDOS_RELAXED      0
806 #define HAMMER_FLUSH_UNDOS_FORCED       1
807 #define HAMMER_FLUSH_UNDOS_AUTO         2
808 /*
809  * Internal hammer mount data structure
810  */
811 struct hammer_mount {
812         struct mount *mp;
813         /*struct vnode *rootvp;*/
814         struct hammer_ino_rb_tree rb_inos_root;
815         struct hammer_redo_rb_tree rb_redo_root;
816         struct hammer_vol_rb_tree rb_vols_root;
817         struct hammer_nod_rb_tree rb_nods_root;
818         struct hammer_und_rb_tree rb_undo_root;
819         struct hammer_res_rb_tree rb_resv_root;
820         struct hammer_buf_rb_tree rb_bufs_root;
821         struct hammer_pfs_rb_tree rb_pfsm_root;
822         struct hammer_volume *rootvol;
823         struct hammer_base_elm root_btree_beg;
824         struct hammer_base_elm root_btree_end;
825
826         struct malloc_type      *m_misc;
827         struct malloc_type      *m_inodes;
828
829         int     flags;          /* HAMMER_MOUNT_xxx flags */
830         int     hflags;
831         int     ronly;
832         int     nvolumes;
833         int     volume_iterator;
834         int     master_id;      /* -1 or 0-15 - clustering and mirroring */
835         int     version;        /* hammer filesystem version to use */
836         int     rsv_inodes;     /* reserved space due to dirty inodes */
837         int64_t rsv_databytes;  /* reserved space due to record data */
838         int     rsv_recs;       /* reserved space due to dirty records */
839         int     rsv_fromdelay;  /* bigblocks reserved due to flush delay */
840         int     undo_rec_limit; /* based on size of undo area */
841         int     last_newrecords;
842         int     count_newrecords;
843
844         int     volume_to_remove; /* volume that is currently being removed */
845
846         int     inode_reclaims; /* inodes pending reclaim by flusher */
847         int     count_inodes;   /* total number of inodes */
848         int     count_iqueued;  /* inodes queued to flusher */
849
850         struct hammer_flusher flusher;
851
852         u_int   check_interrupt;
853         u_int   check_yield;
854         uuid_t  fsid;
855         struct hammer_io_list volu_list;        /* dirty undo buffers */
856         struct hammer_io_list undo_list;        /* dirty undo buffers */
857         struct hammer_io_list data_list;        /* dirty data buffers */
858         struct hammer_io_list alt_data_list;    /* dirty data buffers */
859         struct hammer_io_list meta_list;        /* dirty meta bufs    */
860         struct hammer_io_list lose_list;        /* loose buffers      */
861         int     locked_dirty_space;             /* meta/volu count    */
862         int     io_running_space;               /* track I/O in progress */
863         int     io_running_wakeup;
864         int     objid_cache_count;
865         int     error;                          /* critical I/O error */
866         struct krate    krate;                  /* rate limited kprintf */
867         hammer_tid_t    asof;                   /* snapshot mount */
868         hammer_tid_t    next_tid;
869         hammer_tid_t    flush_tid1;             /* flusher tid sequencing */
870         hammer_tid_t    flush_tid2;             /* flusher tid sequencing */
871         int64_t copy_stat_freebigblocks;        /* number of free bigblocks */
872         u_int32_t       undo_seqno;             /* UNDO/REDO FIFO seqno */
873         u_int32_t       recover_stage2_seqno;   /* REDO recovery seqno */
874         hammer_off_t    recover_stage2_offset;  /* REDO recovery offset */
875
876         struct netexport export;
877         struct hammer_lock sync_lock;
878         struct hammer_lock free_lock;
879         struct hammer_lock undo_lock;
880         struct hammer_lock blkmap_lock;
881         struct hammer_lock snapshot_lock;
882         struct hammer_lock volume_lock;
883         struct hammer_blockmap  blockmap[HAMMER_MAX_ZONES];
884         struct hammer_undo      undos[HAMMER_MAX_UNDOS];
885         int                     undo_alloc;
886         TAILQ_HEAD(, hammer_undo)  undo_lru_list;
887         TAILQ_HEAD(, hammer_reserve) delay_list;
888         struct hammer_flush_group_list  flush_group_list;
889         hammer_flush_group_t    next_flush_group;
890         TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
891         TAILQ_HEAD(, hammer_reclaim) reclaim_list;
892         TAILQ_HEAD(, hammer_io) iorun_list;
893
894         struct hammer_inostats  inostats[HAMMER_INOSTATS_HSIZE];
895 };
896
897 typedef struct hammer_mount     *hammer_mount_t;
898
899 #define HAMMER_MOUNT_CRITICAL_ERROR     0x0001
900 #define HAMMER_MOUNT_FLUSH_RECOVERY     0x0002
901 #define HAMMER_MOUNT_REDO_SYNC          0x0004
902 #define HAMMER_MOUNT_REDO_RECOVERY_REQ  0x0008
903 #define HAMMER_MOUNT_REDO_RECOVERY_RUN  0x0010
904
905 struct hammer_sync_info {
906         int error;
907         int waitfor;
908 };
909
910 /*
911  * Minium buffer cache bufs required to rebalance the B-Tree.
912  * This is because we must hold the children and the children's children
913  * locked.  Even this might not be enough if things are horribly out
914  * of balance.
915  */
916 #define HAMMER_REBALANCE_MIN_BUFS       \
917         (HAMMER_BTREE_LEAF_ELMS * HAMMER_BTREE_LEAF_ELMS)
918
919
920 #endif
921
922 /*
923  * checkspace slop (8MB chunks), higher numbers are more conservative.
924  */
925 #define HAMMER_CHKSPC_REBLOCK   25
926 #define HAMMER_CHKSPC_MIRROR    20
927 #define HAMMER_CHKSPC_WRITE     20
928 #define HAMMER_CHKSPC_CREATE    20
929 #define HAMMER_CHKSPC_REMOVE    10
930 #define HAMMER_CHKSPC_EMERGENCY 0
931
932 #if defined(_KERNEL)
933
934 extern struct vop_ops hammer_vnode_vops;
935 extern struct vop_ops hammer_spec_vops;
936 extern struct vop_ops hammer_fifo_vops;
937 extern struct bio_ops hammer_bioops;
938
939 extern int hammer_debug_io;
940 extern int hammer_debug_general;
941 extern int hammer_debug_debug;
942 extern int hammer_debug_inode;
943 extern int hammer_debug_locks;
944 extern int hammer_debug_btree;
945 extern int hammer_debug_tid;
946 extern int hammer_debug_recover;
947 extern int hammer_debug_recover_faults;
948 extern int hammer_debug_critical;
949 extern int hammer_cluster_enable;
950 extern int hammer_count_fsyncs;
951 extern int hammer_count_inodes;
952 extern int hammer_count_iqueued;
953 extern int hammer_count_reclaiming;
954 extern int hammer_count_records;
955 extern int hammer_count_record_datas;
956 extern int hammer_count_volumes;
957 extern int hammer_count_buffers;
958 extern int hammer_count_nodes;
959 extern int64_t hammer_count_extra_space_used;
960 extern int64_t hammer_stats_btree_lookups;
961 extern int64_t hammer_stats_btree_searches;
962 extern int64_t hammer_stats_btree_inserts;
963 extern int64_t hammer_stats_btree_deletes;
964 extern int64_t hammer_stats_btree_elements;
965 extern int64_t hammer_stats_btree_splits;
966 extern int64_t hammer_stats_btree_iterations;
967 extern int64_t hammer_stats_btree_root_iterations;
968 extern int64_t hammer_stats_record_iterations;
969 extern int64_t hammer_stats_file_read;
970 extern int64_t hammer_stats_file_write;
971 extern int64_t hammer_stats_file_iopsr;
972 extern int64_t hammer_stats_file_iopsw;
973 extern int64_t hammer_stats_disk_read;
974 extern int64_t hammer_stats_disk_write;
975 extern int64_t hammer_stats_inode_flushes;
976 extern int64_t hammer_stats_commits;
977 extern int64_t hammer_stats_undo;
978 extern int64_t hammer_stats_redo;
979 extern int hammer_count_dirtybufspace;
980 extern int hammer_count_refedbufs;
981 extern int hammer_count_reservations;
982 extern int hammer_count_io_running_read;
983 extern int hammer_count_io_running_write;
984 extern int hammer_count_io_locked;
985 extern int hammer_limit_dirtybufspace;
986 extern int hammer_limit_running_io;
987 extern int hammer_limit_recs;
988 extern int hammer_limit_inode_recs;
989 extern int hammer_limit_reclaim;
990 extern int hammer_limit_redo;
991 extern int hammer_bio_count;
992 extern int hammer_verify_zone;
993 extern int hammer_verify_data;
994 extern int hammer_write_mode;
995 extern int hammer_yield_check;
996 extern int hammer_fsync_mode;
997 extern int hammer_autoflush;
998 extern int64_t hammer_contention_count;
999
1000 void    hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
1001                         int error, const char *msg);
1002 int     hammer_vop_inactive(struct vop_inactive_args *);
1003 int     hammer_vop_reclaim(struct vop_reclaim_args *);
1004 int     hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
1005 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
1006                         hammer_inode_t dip, int64_t obj_id,
1007                         hammer_tid_t asof, u_int32_t localization,
1008                         int flags, int *errorp);
1009 struct hammer_inode *hammer_get_dummy_inode(hammer_transaction_t trans,
1010                         hammer_inode_t dip, int64_t obj_id,
1011                         hammer_tid_t asof, u_int32_t localization,
1012                         int flags, int *errorp);
1013 struct hammer_inode *hammer_find_inode(hammer_transaction_t trans,
1014                         int64_t obj_id, hammer_tid_t asof,
1015                         u_int32_t localization);
1016 void    hammer_scan_inode_snapshots(hammer_mount_t hmp,
1017                         hammer_inode_info_t iinfo,
1018                         int (*callback)(hammer_inode_t ip, void *data),
1019                         void *data);
1020 void    hammer_put_inode(struct hammer_inode *ip);
1021 void    hammer_put_inode_ref(struct hammer_inode *ip);
1022 void    hammer_inode_waitreclaims(hammer_transaction_t trans);
1023
1024 int     hammer_unload_volume(hammer_volume_t volume, void *data __unused);
1025 int     hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
1026
1027 int     hammer_unload_buffer(hammer_buffer_t buffer, void *data);
1028 int     hammer_install_volume(hammer_mount_t hmp, const char *volname,
1029                         struct vnode *devvp);
1030 int     hammer_mountcheck_volumes(hammer_mount_t hmp);
1031
1032 int     hammer_mem_add(hammer_record_t record);
1033 int     hammer_ip_lookup(hammer_cursor_t cursor);
1034 int     hammer_ip_first(hammer_cursor_t cursor);
1035 int     hammer_ip_next(hammer_cursor_t cursor);
1036 int     hammer_ip_resolve_data(hammer_cursor_t cursor);
1037 int     hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
1038                         hammer_tid_t tid);
1039 int     hammer_create_at_cursor(hammer_cursor_t cursor,
1040                         hammer_btree_leaf_elm_t leaf, void *udata, int mode);
1041 int     hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
1042                         hammer_tid_t delete_tid, u_int32_t delete_ts,
1043                         int track, int64_t *stat_bytes);
1044 int     hammer_ip_check_directory_empty(hammer_transaction_t trans,
1045                         hammer_inode_t ip);
1046 int     hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
1047 int     hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
1048
1049 hammer_record_t
1050         hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
1051 void    hammer_flush_record_done(hammer_record_t record, int error);
1052 void    hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
1053 void    hammer_rel_mem_record(hammer_record_t record);
1054
1055 int     hammer_cursor_up(hammer_cursor_t cursor);
1056 int     hammer_cursor_up_locked(hammer_cursor_t cursor);
1057 int     hammer_cursor_down(hammer_cursor_t cursor);
1058 int     hammer_cursor_upgrade(hammer_cursor_t cursor);
1059 int     hammer_cursor_upgrade_node(hammer_cursor_t cursor);
1060 void    hammer_cursor_downgrade(hammer_cursor_t cursor);
1061 int     hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
1062                         int index);
1063 void    hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
1064 int     hammer_lock_ex_try(struct hammer_lock *lock);
1065 void    hammer_lock_sh(struct hammer_lock *lock);
1066 int     hammer_lock_sh_try(struct hammer_lock *lock);
1067 int     hammer_lock_upgrade(struct hammer_lock *lock);
1068 void    hammer_lock_downgrade(struct hammer_lock *lock);
1069 int     hammer_lock_status(struct hammer_lock *lock);
1070 void    hammer_unlock(struct hammer_lock *lock);
1071 void    hammer_ref(struct hammer_lock *lock);
1072 int     hammer_ref_interlock(struct hammer_lock *lock);
1073 int     hammer_ref_interlock_true(struct hammer_lock *lock);
1074 void    hammer_ref_interlock_done(struct hammer_lock *lock);
1075 void    hammer_rel(struct hammer_lock *lock);
1076 int     hammer_rel_interlock(struct hammer_lock *lock, int locked);
1077 void    hammer_rel_interlock_done(struct hammer_lock *lock, int orig_locked);
1078 int     hammer_get_interlock(struct hammer_lock *lock);
1079 int     hammer_try_interlock_norefs(struct hammer_lock *lock);
1080 void    hammer_put_interlock(struct hammer_lock *lock, int error);
1081
1082 void    hammer_sync_lock_ex(hammer_transaction_t trans);
1083 void    hammer_sync_lock_sh(hammer_transaction_t trans);
1084 int     hammer_sync_lock_sh_try(hammer_transaction_t trans);
1085 void    hammer_sync_unlock(hammer_transaction_t trans);
1086
1087 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
1088 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
1089 void    hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
1090 u_int64_t hammer_timespec_to_time(struct timespec *ts);
1091 int     hammer_str_to_tid(const char *str, int *ispfsp,
1092                         hammer_tid_t *tidp, u_int32_t *localizationp);
1093 int     hammer_is_atatext(const char *name, int len);
1094 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip,
1095                         int64_t namekey);
1096 void hammer_clear_objid(hammer_inode_t dip);
1097 void hammer_destroy_objid_cache(hammer_mount_t hmp);
1098
1099 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
1100                         int bytes);
1101 void hammer_clear_undo_history(hammer_mount_t hmp);
1102 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
1103 int hammer_get_dtype(u_int8_t obj_type);
1104 u_int8_t hammer_get_obj_type(enum vtype vtype);
1105 int64_t hammer_directory_namekey(hammer_inode_t dip, const void *name, int len,
1106                         u_int32_t *max_iterationsp);
1107 int     hammer_nohistory(hammer_inode_t ip);
1108
1109 int     hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
1110                         hammer_node_cache_t cache, hammer_inode_t ip);
1111 void    hammer_normalize_cursor(hammer_cursor_t cursor);
1112 void    hammer_done_cursor(hammer_cursor_t cursor);
1113 int     hammer_recover_cursor(hammer_cursor_t cursor);
1114 void    hammer_unlock_cursor(hammer_cursor_t cursor);
1115 int     hammer_lock_cursor(hammer_cursor_t cursor);
1116 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor);
1117 void    hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
1118
1119 void    hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
1120 void    hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
1121                         int index);
1122 void    hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
1123                         int index);
1124 void    hammer_cursor_moved_element(hammer_node_t oparent, int pindex,
1125                         hammer_node_t onode, int oindex,
1126                         hammer_node_t nnode, int nindex);
1127 void    hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent,
1128                         hammer_node_t nparent, int nindex);
1129 void    hammer_cursor_inserted_element(hammer_node_t node, int index);
1130 void    hammer_cursor_deleted_element(hammer_node_t node, int index);
1131 void    hammer_cursor_invalidate_cache(hammer_cursor_t cursor);
1132
1133 int     hammer_btree_lookup(hammer_cursor_t cursor);
1134 int     hammer_btree_first(hammer_cursor_t cursor);
1135 int     hammer_btree_last(hammer_cursor_t cursor);
1136 int     hammer_btree_extract(hammer_cursor_t cursor, int flags);
1137 int     hammer_btree_iterate(hammer_cursor_t cursor);
1138 int     hammer_btree_iterate_reverse(hammer_cursor_t cursor);
1139 int     hammer_btree_insert(hammer_cursor_t cursor,
1140                             hammer_btree_leaf_elm_t elm, int *doprop);
1141 int     hammer_btree_delete(hammer_cursor_t cursor);
1142 void    hammer_btree_do_propagation(hammer_cursor_t cursor,
1143                             hammer_pseudofs_inmem_t pfsm,
1144                             hammer_btree_leaf_elm_t leaf);
1145 int     hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
1146 int     hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
1147 int     hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
1148 int     hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
1149
1150 int     btree_set_parent(hammer_transaction_t trans, hammer_node_t node,
1151                         hammer_btree_elm_t elm);
1152 void    hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node);
1153 void    hammer_btree_lcache_init(hammer_mount_t hmp, hammer_node_lock_t lcache,
1154                         int depth);
1155 void    hammer_btree_lcache_free(hammer_mount_t hmp, hammer_node_lock_t lcache);
1156 int     hammer_btree_lock_children(hammer_cursor_t cursor, int depth,
1157                         hammer_node_lock_t parent,
1158                         hammer_node_lock_t lcache);
1159 void    hammer_btree_lock_copy(hammer_cursor_t cursor,
1160                         hammer_node_lock_t parent);
1161 int     hammer_btree_sync_copy(hammer_cursor_t cursor,
1162                         hammer_node_lock_t parent);
1163 void    hammer_btree_unlock_children(hammer_mount_t hmp,
1164                         hammer_node_lock_t parent,
1165                         hammer_node_lock_t lcache);
1166 int     hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
1167 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans,
1168                         hammer_node_t node, int *parent_indexp,
1169                         int *errorp, int try_exclusive);
1170
1171 void    hammer_print_btree_node(hammer_node_ondisk_t ondisk);
1172 void    hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
1173
1174 void    *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
1175                         int *errorp, struct hammer_buffer **bufferp);
1176 void    *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
1177                         int *errorp, struct hammer_buffer **bufferp);
1178 void    *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1179                         int *errorp, struct hammer_buffer **bufferp);
1180 void    *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1181                         int *errorp, struct hammer_buffer **bufferp);
1182
1183 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
1184
1185 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
1186                         int32_t vol_no, int *errorp);
1187 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
1188                         int bytes, int isnew, int *errorp);
1189 void            hammer_sync_buffers(hammer_mount_t hmp,
1190                         hammer_off_t base_offset, int bytes);
1191 int             hammer_del_buffers(hammer_mount_t hmp,
1192                         hammer_off_t base_offset,
1193                         hammer_off_t zone2_offset, int bytes,
1194                         int report_conflicts);
1195
1196 int             hammer_ref_volume(hammer_volume_t volume);
1197 int             hammer_ref_buffer(hammer_buffer_t buffer);
1198 void            hammer_flush_buffer_nodes(hammer_buffer_t buffer);
1199
1200 void            hammer_rel_volume(hammer_volume_t volume, int locked);
1201 void            hammer_rel_buffer(hammer_buffer_t buffer, int locked);
1202
1203 int             hammer_vfs_export(struct mount *mp, int op,
1204                         const struct export_args *export);
1205 hammer_node_t   hammer_get_node(hammer_transaction_t trans,
1206                         hammer_off_t node_offset, int isnew, int *errorp);
1207 void            hammer_ref_node(hammer_node_t node);
1208 hammer_node_t   hammer_ref_node_safe(hammer_transaction_t trans,
1209                         hammer_node_cache_t cache, int *errorp);
1210 void            hammer_rel_node(hammer_node_t node);
1211 void            hammer_delete_node(hammer_transaction_t trans,
1212                         hammer_node_t node);
1213 void            hammer_cache_node(hammer_node_cache_t cache,
1214                         hammer_node_t node);
1215 void            hammer_uncache_node(hammer_node_cache_t cache);
1216 void            hammer_flush_node(hammer_node_t node, int locked);
1217
1218 void hammer_dup_buffer(struct hammer_buffer **bufferp,
1219                         struct hammer_buffer *buffer);
1220 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans,
1221                         hammer_off_t hint, int *errorp);
1222 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1223                         u_int16_t rec_type, hammer_off_t *data_offsetp,
1224                         struct hammer_buffer **data_bufferp,
1225                         hammer_off_t hint, int *errorp);
1226
1227 int hammer_generate_undo(hammer_transaction_t trans,
1228                         hammer_off_t zone1_offset, void *base, int len);
1229 int hammer_generate_redo(hammer_transaction_t trans, hammer_inode_t ip,
1230                         hammer_off_t file_offset, u_int32_t flags,
1231                         void *base, int len);
1232 void hammer_generate_redo_sync(hammer_transaction_t trans);
1233 void hammer_redo_fifo_start_flush(hammer_inode_t ip);
1234 void hammer_redo_fifo_end_flush(hammer_inode_t ip);
1235
1236 void hammer_format_undo(void *base, u_int32_t seqno);
1237 int hammer_upgrade_undo_4(hammer_transaction_t trans);
1238
1239 void hammer_put_volume(struct hammer_volume *volume, int flush);
1240 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
1241
1242 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1243                         hammer_off_t owner, int *errorp);
1244 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1245                         hammer_off_t owner, int *errorp);
1246 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp);
1247 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1248                         int bytes, hammer_off_t hint, int *errorp);
1249 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1250                         int bytes, hammer_off_t *zone_offp, int *errorp);
1251 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1252                         hammer_reserve_t resv);
1253 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1254 void hammer_blockmap_free(hammer_transaction_t trans,
1255                         hammer_off_t bmap_off, int bytes);
1256 int hammer_blockmap_finalize(hammer_transaction_t trans,
1257                         hammer_reserve_t resv,
1258                         hammer_off_t bmap_off, int bytes);
1259 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
1260                         int *curp, int *errorp);
1261 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1262                         int *errorp);
1263 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1264                         int *errorp);
1265 int64_t hammer_undo_used(hammer_transaction_t trans);
1266 int64_t hammer_undo_space(hammer_transaction_t trans);
1267 int64_t hammer_undo_max(hammer_mount_t hmp);
1268 int hammer_undo_reclaim(hammer_io_t io);
1269
1270 void hammer_start_transaction(struct hammer_transaction *trans,
1271                               struct hammer_mount *hmp);
1272 void hammer_simple_transaction(struct hammer_transaction *trans,
1273                               struct hammer_mount *hmp);
1274 void hammer_start_transaction_fls(struct hammer_transaction *trans,
1275                                   struct hammer_mount *hmp);
1276 void hammer_done_transaction(struct hammer_transaction *trans);
1277 hammer_tid_t hammer_alloc_tid(hammer_mount_t hmp, int count);
1278
1279 void hammer_modify_inode(hammer_transaction_t trans, hammer_inode_t ip, int flags);
1280 void hammer_flush_inode(hammer_inode_t ip, int flags);
1281 void hammer_flush_inode_done(hammer_inode_t ip, int error);
1282 void hammer_wait_inode(hammer_inode_t ip);
1283
1284 int  hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
1285                         struct ucred *cred, struct hammer_inode *dip,
1286                         const char *name, int namelen,
1287                         hammer_pseudofs_inmem_t pfsm,
1288                         struct hammer_inode **ipp);
1289 void hammer_rel_inode(hammer_inode_t ip, int flush);
1290 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1291 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1292 int hammer_redo_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1293 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1294
1295 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1296 void hammer_test_inode(hammer_inode_t dip);
1297 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1298
1299 int  hammer_ip_add_directory(struct hammer_transaction *trans,
1300                         hammer_inode_t dip, const char *name, int bytes,
1301                         hammer_inode_t nip);
1302 int  hammer_ip_del_directory(struct hammer_transaction *trans,
1303                         hammer_cursor_t cursor, hammer_inode_t dip,
1304                         hammer_inode_t ip);
1305 void hammer_ip_replace_bulk(hammer_mount_t hmp, hammer_record_t record);
1306 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1307                         void *data, int bytes, int *errorp);
1308 int  hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
1309 int  hammer_ip_add_record(struct hammer_transaction *trans,
1310                         hammer_record_t record);
1311 int  hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1312                         int64_t ran_beg, int64_t ran_end, int truncating);
1313 int  hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1314                         int *countp);
1315 int  hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1316                         int64_t offset, void *data, int bytes);
1317 int  hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
1318 int  hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1319 hammer_pseudofs_inmem_t  hammer_load_pseudofs(hammer_transaction_t trans,
1320                         u_int32_t localization, int *errorp);
1321 int  hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1322                         hammer_pseudofs_inmem_t pfsm);
1323 int  hammer_save_pseudofs(hammer_transaction_t trans,
1324                         hammer_pseudofs_inmem_t pfsm);
1325 int  hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization);
1326 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1327 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1328                         struct ucred *cred);
1329
1330 void hammer_io_init(hammer_io_t io, hammer_volume_t volume,
1331                         enum hammer_io_type type);
1332 int hammer_io_read(struct vnode *devvp, struct hammer_io *io,
1333                         hammer_off_t limit);
1334 void hammer_io_advance(struct hammer_io *io);
1335 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
1336 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1337 struct buf *hammer_io_release(struct hammer_io *io, int flush);
1338 void hammer_io_flush(struct hammer_io *io, int reclaim);
1339 void hammer_io_wait(struct hammer_io *io);
1340 void hammer_io_waitdep(struct hammer_io *io);
1341 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident, int doflush);
1342 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1343                         hammer_btree_leaf_elm_t leaf);
1344 int hammer_io_direct_write(hammer_mount_t hmp, struct bio *bio,
1345                         hammer_record_t record);
1346 void hammer_io_direct_wait(hammer_record_t record);
1347 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1348 void hammer_io_write_interlock(hammer_io_t io);
1349 void hammer_io_done_interlock(hammer_io_t io);
1350 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1351 void hammer_io_clear_modlist(struct hammer_io *io);
1352 void hammer_io_flush_sync(hammer_mount_t hmp);
1353 void hammer_io_clear_error(struct hammer_io *io);
1354 void hammer_io_notmeta(hammer_buffer_t buffer);
1355 void hammer_io_limit_backlog(hammer_mount_t hmp);
1356
1357 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1358                         void *base, int len);
1359 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1360                         void *base, int len);
1361 void hammer_modify_volume_done(hammer_volume_t volume);
1362 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1363
1364 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1365                         struct hammer_ioc_reblock *reblock);
1366 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip,
1367                         struct hammer_ioc_rebalance *rebal);
1368 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1369                         struct hammer_ioc_prune *prune);
1370 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1371                         struct hammer_ioc_mirror_rw *mirror);
1372 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1373                         struct hammer_ioc_mirror_rw *mirror);
1374 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1375                         struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1376 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1377                         struct hammer_ioc_pseudofs_rw *pfs);
1378 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1379                         struct hammer_ioc_pseudofs_rw *pfs);
1380 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1381                         struct hammer_ioc_pseudofs_rw *pfs);
1382 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1383                         struct hammer_ioc_pseudofs_rw *pfs);
1384 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1385                         struct hammer_ioc_pseudofs_rw *pfs);
1386 int hammer_ioc_volume_add(hammer_transaction_t trans, hammer_inode_t ip,
1387                         struct hammer_ioc_volume *ioc);
1388 int hammer_ioc_volume_del(hammer_transaction_t trans, hammer_inode_t ip,
1389                         struct hammer_ioc_volume *ioc);
1390
1391 int hammer_signal_check(hammer_mount_t hmp);
1392
1393 void hammer_flusher_create(hammer_mount_t hmp);
1394 void hammer_flusher_destroy(hammer_mount_t hmp);
1395 void hammer_flusher_sync(hammer_mount_t hmp);
1396 int  hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1397 int  hammer_flusher_async_one(hammer_mount_t hmp);
1398 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1399 void hammer_flusher_wait_next(hammer_mount_t hmp);
1400 int  hammer_flusher_meta_limit(hammer_mount_t hmp);
1401 int  hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1402 int  hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1403 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1404 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1405 int  hammer_flusher_haswork(hammer_mount_t hmp);
1406 void hammer_flusher_flush_undos(hammer_mount_t hmp, int already_flushed);
1407
1408 int hammer_recover_stage1(hammer_mount_t hmp, hammer_volume_t rootvol);
1409 int hammer_recover_stage2(hammer_mount_t hmp, hammer_volume_t rootvol);
1410 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1411                         hammer_volume_t root_volume, int final);
1412
1413 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1414 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1415 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1416
1417 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1418 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1419 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1420 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1421 void hkprintf(const char *ctl, ...);
1422 udev_t hammer_fsid_to_udev(uuid_t *uuid);
1423
1424
1425 int hammer_blocksize(int64_t file_offset);
1426 int hammer_blockoff(int64_t file_offset);
1427 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1428
1429 /*
1430  * Shortcut for _hammer_checkspace(), used all over the code.
1431  */
1432 static __inline int
1433 hammer_checkspace(hammer_mount_t hmp, int slop)
1434 {
1435         return(_hammer_checkspace(hmp, slop, NULL));
1436 }
1437
1438 #endif
1439
1440 static __inline void
1441 hammer_wait_mem_record(hammer_record_t record)
1442 {
1443         hammer_wait_mem_record_ident(record, "hmmwai");
1444 }
1445
1446 static __inline void
1447 hammer_lock_ex(struct hammer_lock *lock)
1448 {
1449         hammer_lock_ex_ident(lock, "hmrlck");
1450 }
1451
1452 /*
1453  * Indicate that a B-Tree node is being modified.
1454  */
1455 static __inline void
1456 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1457 {
1458         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1459         hammer_modify_buffer(trans, node->buffer, NULL, 0);
1460 }
1461
1462 static __inline void
1463 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1464 {
1465         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1466         hammer_modify_buffer(trans, node->buffer,
1467                              node->ondisk, sizeof(*node->ondisk));
1468 }
1469
1470 static __inline void
1471 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1472                    void *base, int len)
1473 {
1474         hammer_crc_t *crcptr;
1475
1476         KKASSERT((char *)base >= (char *)node->ondisk &&
1477                  (char *)base + len <=
1478                     (char *)node->ondisk + sizeof(*node->ondisk));
1479         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1480         hammer_modify_buffer(trans, node->buffer, base, len);
1481         crcptr = &node->ondisk->crc;
1482         hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t));
1483         --node->buffer->io.modify_refs; /* only want one ref */
1484 }
1485
1486 /*
1487  * Indicate that the specified modifications have been completed.
1488  *
1489  * Do not try to generate the crc here, it's very expensive to do and a
1490  * sequence of insertions or deletions can result in many calls to this
1491  * function on the same node.
1492  */
1493 static __inline void
1494 hammer_modify_node_done(hammer_node_t node)
1495 {
1496         node->flags |= HAMMER_NODE_CRCGOOD;
1497         if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1498                 node->flags |= HAMMER_NODE_NEEDSCRC;
1499                 node->buffer->io.gencrc = 1;
1500                 hammer_ref_node(node);
1501         }
1502         hammer_modify_buffer_done(node->buffer);
1503 }
1504
1505 #define hammer_modify_volume_field(trans, vol, field)           \
1506         hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1507                              sizeof((vol)->ondisk->field))
1508
1509 #define hammer_modify_node_field(trans, node, field)            \
1510         hammer_modify_node(trans, node, &(node)->ondisk->field, \
1511                              sizeof((node)->ondisk->field))
1512
1513 /*
1514  * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1515  * created directories for HAMMER version 2 or greater and causes
1516  * directory entries to be placed the inode localization zone in
1517  * the B-Tree instead of the misc zone.
1518  *
1519  * This greatly improves localization between directory entries and
1520  * inodes
1521  */
1522 static __inline u_int32_t
1523 hammer_dir_localization(hammer_inode_t dip)
1524 {
1525         if (dip->ino_data.cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO)
1526                 return(HAMMER_LOCALIZE_INODE);
1527         else
1528                 return(HAMMER_LOCALIZE_MISC);
1529 }