11f20b46ae53e8075bda77e951b9eac2d445effa
[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 "hammer_disk.h"
66 #include "hammer_mount.h"
67 #include "hammer_ioctl.h"
68
69 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
70
71 MALLOC_DECLARE(M_HAMMER);
72
73 /*
74  * Kernel trace
75  */
76 #if !defined(KTR_HAMMER)
77 #define KTR_HAMMER      KTR_ALL
78 #endif
79 KTR_INFO_MASTER_EXTERN(hammer);
80
81 /*
82  * Misc structures
83  */
84 struct hammer_mount;
85
86 /*
87  * Key structure used for custom RB tree inode lookups.  This prototypes
88  * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
89  */
90 typedef struct hammer_inode_info {
91         int64_t         obj_id;         /* (key) object identifier */
92         hammer_tid_t    obj_asof;       /* (key) snapshot transid or 0 */
93         u_int32_t       obj_localization; /* (key) pseudo-fs */
94         union {
95                 struct hammer_btree_leaf_elm *leaf;
96         } u;
97 } *hammer_inode_info_t;
98
99 typedef enum hammer_transaction_type {
100         HAMMER_TRANS_RO,
101         HAMMER_TRANS_STD,
102         HAMMER_TRANS_FLS
103 } hammer_transaction_type_t;
104
105 /*
106  * HAMMER Transaction tracking
107  */
108 struct hammer_transaction {
109         hammer_transaction_type_t type;
110         struct hammer_mount *hmp;
111         hammer_tid_t    tid;
112         u_int64_t       time;
113         u_int32_t       time32;
114         int             sync_lock_refs;
115         int             flags;
116         struct hammer_volume *rootvol;
117 };
118
119 typedef struct hammer_transaction *hammer_transaction_t;
120
121 #define HAMMER_TRANSF_NEWINODE  0x0001
122 #define HAMMER_TRANSF_DIDIO     0x0002
123 #define HAMMER_TRANSF_CRCDOM    0x0004  /* EDOM on CRC error, less critical */
124
125 /*
126  * HAMMER locks
127  */
128 struct hammer_lock {
129         volatile u_int  refs;           /* active references */
130         volatile u_int  lockval;        /* lock count and control bits */
131         struct thread   *lowner;        /* owner if exclusively held */
132         struct thread   *rowner;        /* owner if exclusively held */
133 };
134
135 #define HAMMER_REFS_LOCKED      0x40000000      /* transition check */
136 #define HAMMER_REFS_WANTED      0x20000000      /* transition check */
137 #define HAMMER_REFS_CHECK       0x10000000      /* transition check */
138
139 #define HAMMER_REFS_FLAGS       (HAMMER_REFS_LOCKED | \
140                                  HAMMER_REFS_WANTED | \
141                                  HAMMER_REFS_CHECK)
142
143 #define HAMMER_LOCKF_EXCLUSIVE  0x40000000
144 #define HAMMER_LOCKF_WANTED     0x20000000
145
146 static __inline int
147 hammer_notlocked(struct hammer_lock *lock)
148 {
149         return(lock->lockval == 0);
150 }
151
152 static __inline int
153 hammer_islocked(struct hammer_lock *lock)
154 {
155         return(lock->lockval != 0);
156 }
157
158 /*
159  * Returns the number of refs on the object.
160  */
161 static __inline int
162 hammer_isactive(struct hammer_lock *lock)
163 {
164         return(lock->refs & ~HAMMER_REFS_FLAGS);
165 }
166
167 static __inline int
168 hammer_oneref(struct hammer_lock *lock)
169 {
170         return((lock->refs & ~HAMMER_REFS_FLAGS) == 1);
171 }
172
173 static __inline int
174 hammer_norefs(struct hammer_lock *lock)
175 {
176         return((lock->refs & ~HAMMER_REFS_FLAGS) == 0);
177 }
178
179 static __inline int
180 hammer_norefsorlock(struct hammer_lock *lock)
181 {
182         return(lock->refs == 0);
183 }
184
185 static __inline int
186 hammer_refsorlock(struct hammer_lock *lock)
187 {
188         return(lock->refs != 0);
189 }
190
191 /*
192  * Return if we specifically own the lock exclusively.
193  */
194 static __inline int
195 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
196 {
197         if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) &&
198             lock->lowner == td) {
199                 return(1);
200         }
201         return(0);
202 }
203
204 /*
205  * Flush state, used by various structures
206  */
207 typedef enum hammer_inode_state {
208         HAMMER_FST_IDLE,
209         HAMMER_FST_SETUP,
210         HAMMER_FST_FLUSH
211 } hammer_inode_state_t;
212
213 TAILQ_HEAD(hammer_record_list, hammer_record);
214
215 /*
216  * Pseudo-filesystem extended data tracking
217  */
218 struct hammer_pfs_rb_tree;
219 struct hammer_pseudofs_inmem;
220 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
221 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
222               hammer_pfs_rb_compare, u_int32_t);
223
224 struct hammer_pseudofs_inmem {
225         RB_ENTRY(hammer_pseudofs_inmem) rb_node;
226         struct hammer_lock      lock;
227         u_int32_t               localization;
228         hammer_tid_t            create_tid;
229         int                     flags;
230         udev_t                  fsid_udev;
231         struct hammer_pseudofs_data pfsd;
232 };
233
234 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t;
235
236 #define HAMMER_PFSM_DELETED     0x0001
237
238 /*
239  * Cache object ids.  A fixed number of objid cache structures are
240  * created to reserve object id's for newly created files in multiples
241  * of 100,000, localized to a particular directory, and recycled as
242  * needed.  This allows parallel create operations in different
243  * directories to retain fairly localized object ids which in turn
244  * improves reblocking performance and layout.
245  */
246 #define OBJID_CACHE_SIZE        2048
247 #define OBJID_CACHE_BULK_BITS   10              /* 10 bits (1024)       */
248 #define OBJID_CACHE_BULK        (32 * 32)       /* two level (1024)     */
249 #define OBJID_CACHE_BULK_MASK   (OBJID_CACHE_BULK - 1)
250 #define OBJID_CACHE_BULK_MASK64 ((u_int64_t)(OBJID_CACHE_BULK - 1))
251
252 typedef struct hammer_objid_cache {
253         TAILQ_ENTRY(hammer_objid_cache) entry;
254         struct hammer_inode             *dip;
255         hammer_tid_t                    base_tid;
256         int                             count;
257         u_int32_t                       bm0;
258         u_int32_t                       bm1[32];
259 } *hammer_objid_cache_t;
260
261 /*
262  * Associate an inode with a B-Tree node to cache search start positions
263  */
264 typedef struct hammer_node_cache {
265         TAILQ_ENTRY(hammer_node_cache)  entry;
266         struct hammer_node              *node;
267         struct hammer_inode             *ip;
268 } *hammer_node_cache_t;
269
270 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
271
272 /*
273  * Live dedup cache
274  */
275 struct hammer_dedup_crc_rb_tree;
276 RB_HEAD(hammer_dedup_crc_rb_tree, hammer_dedup_cache);
277 RB_PROTOTYPE2(hammer_dedup_crc_rb_tree, hammer_dedup_cache, crc_entry,
278                 hammer_dedup_crc_rb_compare, hammer_crc_t);
279
280 struct hammer_dedup_off_rb_tree;
281 RB_HEAD(hammer_dedup_off_rb_tree, hammer_dedup_cache);
282 RB_PROTOTYPE2(hammer_dedup_off_rb_tree, hammer_dedup_cache, off_entry,
283                 hammer_dedup_off_rb_compare, hammer_off_t);
284
285 #define DEDUP_CACHE_SIZE        4096 /* XXX make it a dynamic tunable */
286
287 typedef struct hammer_dedup_cache {
288         RB_ENTRY(hammer_dedup_cache) crc_entry;
289         RB_ENTRY(hammer_dedup_cache) off_entry;
290         TAILQ_ENTRY(hammer_dedup_cache) lru_entry;
291         struct hammer_mount *hmp;
292         int64_t obj_id;
293         u_int32_t localization;
294         off_t file_offset;
295         int bytes;
296         hammer_off_t data_offset; 
297         hammer_crc_t crc;
298 } *hammer_dedup_cache_t;
299
300 /*
301  * Structure used to organize flush groups.  Flush groups must be
302  * organized into chunks in order to avoid blowing out the UNDO FIFO.
303  * Without this a 'sync' could end up flushing 50,000 inodes in a single
304  * transaction.
305  */
306 struct hammer_fls_rb_tree;
307 RB_HEAD(hammer_fls_rb_tree, hammer_inode);
308 RB_PROTOTYPE(hammer_fls_rb_tree, hammer_inode, rb_flsnode,
309               hammer_ino_rb_compare);
310
311 struct hammer_flush_group {
312         TAILQ_ENTRY(hammer_flush_group) flush_entry;
313         struct hammer_fls_rb_tree       flush_tree;
314         int                             seq;            /* our seq no */
315         int                             total_count;    /* record load */
316         int                             running;        /* group is running */
317         int                             closed;
318         int                             refs;
319 };
320
321 typedef struct hammer_flush_group *hammer_flush_group_t;
322
323 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
324
325 /*
326  * Structure used to represent an inode in-memory.
327  *
328  * The record and data associated with an inode may be out of sync with
329  * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
330  * clear).
331  *
332  * An inode may also hold a cache of unsynchronized records, used for
333  * database and directories only.  Unsynchronized regular file data is
334  * stored in the buffer cache.
335  *
336  * NOTE: A file which is created and destroyed within the initial
337  * synchronization period can wind up not doing any disk I/O at all.
338  *
339  * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
340  */
341 struct hammer_ino_rb_tree;
342 struct hammer_inode;
343 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
344 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
345               hammer_ino_rb_compare, hammer_inode_info_t);
346
347 struct hammer_redo_rb_tree;
348 RB_HEAD(hammer_redo_rb_tree, hammer_inode);
349 RB_PROTOTYPE2(hammer_redo_rb_tree, hammer_inode, rb_redonode,
350               hammer_redo_rb_compare, hammer_off_t);
351
352 struct hammer_rec_rb_tree;
353 struct hammer_record;
354 RB_HEAD(hammer_rec_rb_tree, hammer_record);
355 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
356               hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
357
358 TAILQ_HEAD(hammer_node_list, hammer_node);
359
360 struct hammer_inode {
361         RB_ENTRY(hammer_inode)  rb_node;
362         hammer_inode_state_t    flush_state;
363         hammer_flush_group_t    flush_group;
364         RB_ENTRY(hammer_inode)  rb_flsnode;     /* when on flush list */
365         RB_ENTRY(hammer_inode)  rb_redonode;    /* when INODE_RDIRTY is set */
366         struct hammer_record_list target_list;  /* target of dependant recs */
367         int64_t                 obj_id;         /* (key) object identifier */
368         hammer_tid_t            obj_asof;       /* (key) snapshot or 0 */
369         u_int32_t               obj_localization; /* (key) pseudo-fs */
370         struct hammer_mount     *hmp;
371         hammer_objid_cache_t    objid_cache;
372         int                     flags;
373         int                     error;          /* flush error */
374         int                     cursor_ip_refs; /* sanity */
375         int                     rsv_recs;
376         struct vnode            *vp;
377         hammer_pseudofs_inmem_t pfsm;
378         struct lockf            advlock;
379         struct hammer_lock      lock;           /* sync copy interlock */
380         off_t                   trunc_off;
381         struct hammer_btree_leaf_elm ino_leaf;  /* in-memory cache */
382         struct hammer_inode_data ino_data;      /* in-memory cache */
383         struct hammer_rec_rb_tree rec_tree;     /* in-memory cache */
384         int                     rec_generation;
385         struct hammer_node_cache cache[4];      /* search initiate cache */
386
387         /*
388          * When a demark is created to synchronize an inode to
389          * disk, certain fields are copied so the front-end VOPs
390          * can continue to run in parallel with the synchronization
391          * occuring in the background.
392          */
393         int             sync_flags;             /* to-sync flags cache */
394         off_t           sync_trunc_off;         /* to-sync truncation */
395         off_t           save_trunc_off;         /* write optimization */
396         struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
397         struct hammer_inode_data sync_ino_data; /* to-sync cache */
398         size_t          redo_count;
399
400         /*
401          * Track the earliest offset in the UNDO/REDO FIFO containing
402          * REDO records.  This is staged to the backend during flush
403          * sequences.  While the inode is staged redo_fifo_next is used
404          * to track the earliest offset for rotation into redo_fifo_start
405          * on completion of the flush.
406          */
407         hammer_off_t    redo_fifo_start;
408         hammer_off_t    redo_fifo_next;
409 };
410
411 typedef struct hammer_inode *hammer_inode_t;
412
413 #define VTOI(vp)        ((struct hammer_inode *)(vp)->v_data)
414
415 /*
416  * NOTE: DDIRTY does not include atime or mtime and does not include
417  *       write-append size changes.  SDIRTY handles write-append size
418  *       changes.
419  *
420  *       REDO indicates that REDO logging is active, creating a definitive
421  *       stream of REDO records in the UNDO/REDO log for writes and
422  *       truncations, including boundary records when/if REDO is turned off.
423  *       REDO is typically enabled by fsync() and turned off if excessive
424  *       writes without an fsync() occurs.
425  *
426  *       RDIRTY indicates that REDO records were laid down in the UNDO/REDO
427  *       FIFO (even if REDO is turned off some might still be active) and
428  *       still being tracked for this inode.  See hammer_redo.c
429  */
430                                         /* (not including atime/mtime) */
431 #define HAMMER_INODE_DDIRTY     0x0001  /* in-memory ino_data is dirty */
432 #define HAMMER_INODE_RSV_INODES 0x0002  /* hmp->rsv_inodes bumped */
433 #define HAMMER_INODE_CONN_DOWN  0x0004  /* include in downward recursion */
434 #define HAMMER_INODE_XDIRTY     0x0008  /* in-memory records */
435 #define HAMMER_INODE_ONDISK     0x0010  /* inode is on-disk (else not yet) */
436 #define HAMMER_INODE_FLUSH      0x0020  /* flush on last ref */
437 #define HAMMER_INODE_DELETED    0x0080  /* inode delete (backend) */
438 #define HAMMER_INODE_DELONDISK  0x0100  /* delete synchronized to disk */
439 #define HAMMER_INODE_RO         0x0200  /* read-only (because of as-of) */
440 #define HAMMER_INODE_RECSW      0x0400  /* waiting on data record flush */
441 #define HAMMER_INODE_DONDISK    0x0800  /* data records may be on disk */
442 #define HAMMER_INODE_BUFS       0x1000  /* dirty high level bps present */
443 #define HAMMER_INODE_REFLUSH    0x2000  /* flush on dependancy / reflush */
444 #define HAMMER_INODE_RECLAIM    0x4000  /* trying to reclaim */
445 #define HAMMER_INODE_FLUSHW     0x8000  /* Someone waiting for flush */
446
447 #define HAMMER_INODE_TRUNCATED  0x00010000
448 #define HAMMER_INODE_DELETING   0x00020000 /* inode delete request (frontend)*/
449 #define HAMMER_INODE_RESIGNAL   0x00040000 /* re-signal on re-flush */
450 #define HAMMER_INODE_ATIME      0x00100000 /* in-memory atime modified */
451 #define HAMMER_INODE_MTIME      0x00200000 /* in-memory mtime modified */
452 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
453 #define HAMMER_INODE_DUMMY      0x00800000 /* dummy inode covering bad file */
454 #define HAMMER_INODE_SDIRTY     0x01000000 /* in-memory ino_data.size is dirty*/
455 #define HAMMER_INODE_REDO       0x02000000 /* REDO logging active */
456 #define HAMMER_INODE_RDIRTY     0x04000000 /* REDO records active in fifo */
457
458 #define HAMMER_INODE_MODMASK    (HAMMER_INODE_DDIRTY|HAMMER_INODE_SDIRTY|   \
459                                  HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS|     \
460                                  HAMMER_INODE_ATIME|HAMMER_INODE_MTIME|     \
461                                  HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
462
463 #define HAMMER_INODE_MODMASK_NOXDIRTY   \
464                                 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
465
466 #define HAMMER_INODE_MODMASK_NOREDO     \
467                                 (HAMMER_INODE_DDIRTY|                       \
468                                  HAMMER_INODE_XDIRTY|                       \
469                                  HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
470
471 #define HAMMER_FLUSH_SIGNAL     0x0001
472 #define HAMMER_FLUSH_RECURSION  0x0002
473
474 /*
475  * Used by the inode reclaim code to pipeline reclaims and avoid
476  * blowing out kernel memory or letting the flusher get too far
477  * behind.  The reclaim wakes up when count reaches 0 or the
478  * timer expires.
479  */
480 struct hammer_reclaim {
481         TAILQ_ENTRY(hammer_reclaim) entry;
482         int     count;
483 };
484
485 /*
486  * Track who is creating the greatest burden on the
487  * inode cache.
488  */
489 struct hammer_inostats {
490         pid_t           pid;    /* track user process */
491         int             ltick;  /* last tick */
492         int             count;  /* count (degenerates) */
493 };
494
495 #define HAMMER_INOSTATS_HSIZE   32
496 #define HAMMER_INOSTATS_HMASK   (HAMMER_INOSTATS_HSIZE - 1)
497
498 /*
499  * Structure used to represent an unsynchronized record in-memory.  These
500  * records typically represent directory entries.  Only non-historical
501  * records are kept in-memory.
502  *
503  * Records are organized as a per-inode RB-Tree.  If the inode is not
504  * on disk then neither are any records and the in-memory record tree
505  * represents the entire contents of the inode.  If the inode is on disk
506  * then the on-disk B-Tree is scanned in parallel with the in-memory
507  * RB-Tree to synthesize the current state of the file.
508  *
509  * Records are also used to enforce the ordering of directory create/delete
510  * operations.  A new inode will not be flushed to disk unless its related
511  * directory entry is also being flushed at the same time.  A directory entry
512  * will not be removed unless its related inode is also being removed at the
513  * same time.
514  */
515 typedef enum hammer_record_type {
516         HAMMER_MEM_RECORD_GENERAL,      /* misc record */
517         HAMMER_MEM_RECORD_INODE,        /* inode record */
518         HAMMER_MEM_RECORD_ADD,          /* positive memory cache record */
519         HAMMER_MEM_RECORD_DEL,          /* negative delete-on-disk record */
520         HAMMER_MEM_RECORD_DATA          /* bulk-data record w/on-disk ref */
521 } hammer_record_type_t;
522
523 struct hammer_record {
524         RB_ENTRY(hammer_record)         rb_node;
525         TAILQ_ENTRY(hammer_record)      target_entry;
526         hammer_inode_state_t            flush_state;
527         hammer_flush_group_t            flush_group;
528         hammer_record_type_t            type;
529         struct hammer_lock              lock;
530         struct hammer_reserve           *resv;
531         struct hammer_inode             *ip;
532         struct hammer_inode             *target_ip;
533         struct hammer_btree_leaf_elm    leaf;
534         union hammer_data_ondisk        *data;
535         int                             flags;
536         int                             gflags;
537         hammer_off_t                    zone2_offset;   /* direct-write only */
538 };
539
540 typedef struct hammer_record *hammer_record_t;
541
542 /*
543  * Record flags.  Note that FE can only be set by the frontend if the
544  * record has not been interlocked by the backend w/ BE.
545  */
546 #define HAMMER_RECF_ALLOCDATA           0x0001
547 #define HAMMER_RECF_ONRBTREE            0x0002
548 #define HAMMER_RECF_DELETED_FE          0x0004  /* deleted (frontend) */
549 #define HAMMER_RECF_DELETED_BE          0x0008  /* deleted (backend) */
550 #define HAMMER_RECF_COMMITTED           0x0010  /* committed to the B-Tree */
551 #define HAMMER_RECF_INTERLOCK_BE        0x0020  /* backend interlock */
552 #define HAMMER_RECF_WANTED              0x0040  /* wanted by the frontend */
553 #define HAMMER_RECF_DEDUPED             0x0080  /* will be live-dedup'ed */
554 #define HAMMER_RECF_CONVERT_DELETE      0x0100  /* special case */
555 #define HAMMER_RECF_REDO                0x1000  /* REDO was laid down */
556
557 /*
558  * These flags must be separate to deal with SMP races
559  */
560 #define HAMMER_RECG_DIRECT_IO           0x0001  /* related direct I/O running*/
561 #define HAMMER_RECG_DIRECT_WAIT         0x0002  /* related direct I/O running*/
562 #define HAMMER_RECG_DIRECT_INVAL        0x0004  /* buffer alias invalidation */
563 /*
564  * hammer_create_at_cursor() and hammer_delete_at_cursor() flags.
565  */
566 #define HAMMER_CREATE_MODE_UMIRROR      0x0001
567 #define HAMMER_CREATE_MODE_SYS          0x0002
568
569 #define HAMMER_DELETE_ADJUST            0x0001
570 #define HAMMER_DELETE_DESTROY           0x0002
571
572 /*
573  * In-memory structures representing on-disk structures.
574  */
575 struct hammer_volume;
576 struct hammer_buffer;
577 struct hammer_node;
578 struct hammer_undo;
579 struct hammer_reserve;
580
581 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
582 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
583 RB_HEAD(hammer_nod_rb_tree, hammer_node);
584 RB_HEAD(hammer_und_rb_tree, hammer_undo);
585 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
586 RB_HEAD(hammer_mod_rb_tree, hammer_io);
587
588 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
589               hammer_vol_rb_compare, int32_t);
590 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
591               hammer_buf_rb_compare, hammer_off_t);
592 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
593               hammer_nod_rb_compare, hammer_off_t);
594 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
595               hammer_und_rb_compare, hammer_off_t);
596 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
597               hammer_res_rb_compare, hammer_off_t);
598 RB_PROTOTYPE2(hammer_mod_rb_tree, hammer_io, rb_node,
599               hammer_mod_rb_compare, hammer_off_t);
600
601 /*
602  * IO management - embedded at the head of various in-memory structures
603  *
604  * VOLUME       - hammer_volume containing meta-data
605  * META_BUFFER  - hammer_buffer containing meta-data
606  * DATA_BUFFER  - hammer_buffer containing pure-data
607  *
608  * Dirty volume headers and dirty meta-data buffers are locked until the
609  * flusher can sequence them out.  Dirty pure-data buffers can be written.
610  * Clean buffers can be passively released.
611  */
612 typedef enum hammer_io_type {
613         HAMMER_STRUCTURE_VOLUME,
614         HAMMER_STRUCTURE_META_BUFFER,
615         HAMMER_STRUCTURE_UNDO_BUFFER,
616         HAMMER_STRUCTURE_DATA_BUFFER,
617         HAMMER_STRUCTURE_DUMMY
618 } hammer_io_type_t;
619
620 union hammer_io_structure;
621 struct hammer_io;
622
623 struct worklist {
624         LIST_ENTRY(worklist) node;
625 };
626
627 TAILQ_HEAD(hammer_io_list, hammer_io);
628 typedef struct hammer_io_list *hammer_io_list_t;
629
630 struct hammer_io {
631         struct worklist         worklist;
632         struct hammer_lock      lock;
633         enum hammer_io_type     type;
634         struct hammer_mount     *hmp;
635         struct hammer_volume    *volume;
636         RB_ENTRY(hammer_io)     rb_node;     /* if modified */
637         TAILQ_ENTRY(hammer_io)  iorun_entry; /* iorun_list */
638         struct hammer_mod_rb_tree *mod_root;
639         struct buf              *bp;
640         int64_t                 offset;    /* zone-2 offset */
641         int                     bytes;     /* buffer cache buffer size */
642         int                     modify_refs;
643
644         /*
645          * These can be modified at any time by the backend while holding
646          * io_token, due to bio_done and hammer_io_complete() callbacks.
647          */
648         u_int           running : 1;    /* bp write IO in progress */
649         u_int           waiting : 1;    /* someone is waiting on us */
650         u_int           ioerror : 1;    /* abort on io-error */
651         u_int           unusedA : 29;
652
653         /*
654          * These can only be modified by the frontend while holding
655          * fs_token, or by the backend while holding the io interlocked
656          * with no references (which will block the frontend when it
657          * tries to reference it).
658          *
659          * WARNING! SMP RACES will create havoc if the callbacks ever tried
660          *          to modify any of these outside the above restrictions.
661          */
662         u_int           modified : 1;   /* bp's data was modified */
663         u_int           released : 1;   /* bp released (w/ B_LOCKED set) */
664         u_int           validated : 1;  /* ondisk has been validated */
665         u_int           waitdep : 1;    /* flush waits for dependancies */
666         u_int           recovered : 1;  /* has recovery ref */
667         u_int           waitmod : 1;    /* waiting for modify_refs */
668         u_int           reclaim : 1;    /* reclaim requested */
669         u_int           gencrc : 1;     /* crc needs to be generated */
670         u_int           unusedB : 24;
671 };
672
673 typedef struct hammer_io *hammer_io_t;
674
675 #define HAMMER_CLUSTER_SIZE     (64 * 1024)
676 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
677 #undef  HAMMER_CLUSTER_SIZE
678 #define HAMMER_CLUSTER_SIZE     MAXBSIZE
679 #endif
680 #define HAMMER_CLUSTER_BUFS     (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
681
682 /*
683  * In-memory volume representing on-disk buffer
684  */
685 struct hammer_volume {
686         struct hammer_io io;
687         RB_ENTRY(hammer_volume) rb_node;
688         struct hammer_volume_ondisk *ondisk;
689         int32_t vol_no;
690         int64_t nblocks;        /* note: special calculation for statfs */
691         int64_t buffer_base;    /* base offset of buffer 0 */
692         hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
693         hammer_off_t maxraw_off; /* Maximum raw offset for device */
694         char    *vol_name;
695         struct vnode *devvp;
696         int     vol_flags;
697 };
698
699 typedef struct hammer_volume *hammer_volume_t;
700
701 /*
702  * In-memory buffer (other then volume, super-cluster, or cluster),
703  * representing an on-disk buffer.
704  */
705 struct hammer_buffer {
706         struct hammer_io io;
707         RB_ENTRY(hammer_buffer) rb_node;
708         void *ondisk;
709         hammer_off_t zoneX_offset;
710         hammer_off_t zone2_offset;
711         struct hammer_reserve *resv;
712         struct hammer_node_list clist;
713 };
714
715 typedef struct hammer_buffer *hammer_buffer_t;
716
717 /*
718  * In-memory B-Tree node, representing an on-disk B-Tree node.
719  *
720  * This is a hang-on structure which is backed by a hammer_buffer,
721  * indexed by a hammer_cluster, and used for fine-grained locking of
722  * B-Tree nodes in order to properly control lock ordering.  A hammer_buffer
723  * can contain multiple nodes representing wildly disassociated portions
724  * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
725  *
726  * This structure uses a cluster-relative index to reduce the number
727  * of layers required to access it, and also because all on-disk B-Tree
728  * references are cluster-relative offsets.
729  */
730 struct hammer_node {
731         struct hammer_lock      lock;           /* node-by-node lock */
732         TAILQ_ENTRY(hammer_node) entry;         /* per-buffer linkage */
733         RB_ENTRY(hammer_node)   rb_node;        /* per-cluster linkage */
734         hammer_off_t            node_offset;    /* full offset spec */
735         struct hammer_mount     *hmp;
736         struct hammer_buffer    *buffer;        /* backing buffer */
737         hammer_node_ondisk_t    ondisk;         /* ptr to on-disk structure */
738         TAILQ_HEAD(, hammer_cursor) cursor_list;  /* deadlock recovery */
739         struct hammer_node_cache_list cache_list; /* passive caches */
740         int                     flags;
741 };
742
743 #define HAMMER_NODE_DELETED     0x0001
744 #define HAMMER_NODE_FLUSH       0x0002
745 #define HAMMER_NODE_CRCGOOD     0x0004
746 #define HAMMER_NODE_NEEDSCRC    0x0008
747 #define HAMMER_NODE_NEEDSMIRROR 0x0010
748 #define HAMMER_NODE_CRCBAD      0x0020
749 #define HAMMER_NODE_NONLINEAR   0x0040          /* linear heuristic */
750
751 #define HAMMER_NODE_CRCANY      (HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD)
752
753 typedef struct hammer_node      *hammer_node_t;
754
755 /*
756  * List of locked nodes.  This structure is used to lock potentially large
757  * numbers of nodes as an aid for complex B-Tree operations.
758  */
759 struct hammer_node_lock;
760 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock);
761
762 struct hammer_node_lock {
763         TAILQ_ENTRY(hammer_node_lock) entry;
764         struct hammer_node_lock_list  list;
765         struct hammer_node_lock       *parent;
766         hammer_node_t   node;
767         hammer_node_ondisk_t copy;      /* copy of on-disk data */
768         int             index;          /* index of this node in parent */
769         int             count;          /* count children */
770         int             flags;
771 };
772
773 typedef struct hammer_node_lock *hammer_node_lock_t;
774
775 #define HAMMER_NODE_LOCK_UPDATED        0x0001
776 #define HAMMER_NODE_LOCK_LCACHE         0x0002
777
778 /*
779  * Common I/O management structure - embedded in in-memory structures
780  * which are backed by filesystem buffers.
781  */
782 union hammer_io_structure {
783         struct hammer_io        io;
784         struct hammer_volume    volume;
785         struct hammer_buffer    buffer;
786 };
787
788 typedef union hammer_io_structure *hammer_io_structure_t;
789
790 /*
791  * The reserve structure prevents the blockmap from allocating
792  * out of a reserved bigblock.  Such reservations are used by
793  * the direct-write mechanism.
794  *
795  * The structure is also used to hold off on reallocations of
796  * big blocks from the freemap until flush dependancies have
797  * been dealt with.
798  */
799 struct hammer_reserve {
800         RB_ENTRY(hammer_reserve) rb_node;
801         TAILQ_ENTRY(hammer_reserve) delay_entry;
802         int             flush_group;
803         int             flags;
804         int             refs;
805         int             zone;
806         int             append_off;
807         int32_t         bytes_free;
808         hammer_off_t    zone_offset;
809 };
810
811 typedef struct hammer_reserve *hammer_reserve_t;
812
813 #define HAMMER_RESF_ONDELAY     0x0001
814 #define HAMMER_RESF_LAYER2FREE  0x0002
815
816 #include "hammer_cursor.h"
817
818 /*
819  * The undo structure tracks recent undos to avoid laying down duplicate
820  * undos within a flush group, saving us a significant amount of overhead.
821  *
822  * This is strictly a heuristic.
823  */
824 #define HAMMER_MAX_UNDOS                1024
825 #define HAMMER_MAX_FLUSHERS             4
826
827 struct hammer_undo {
828         RB_ENTRY(hammer_undo)   rb_node;
829         TAILQ_ENTRY(hammer_undo) lru_entry;
830         hammer_off_t            offset;
831         int                     bytes;
832 };
833
834 typedef struct hammer_undo *hammer_undo_t;
835
836 struct hammer_flusher_info;
837 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
838
839 struct hammer_flusher {
840         int             signal;         /* flusher thread sequencer */
841         int             act;            /* currently active flush group */
842         int             done;           /* set to act when complete */
843         int             next;           /* next unallocated flg seqno */
844         int             group_lock;     /* lock sequencing of the next flush */
845         int             exiting;        /* request master exit */
846         thread_t        td;             /* master flusher thread */
847         hammer_tid_t    tid;            /* last flushed transaction id */
848         int             finalize_want;          /* serialize finalization */
849         struct hammer_lock finalize_lock;       /* serialize finalization */
850         struct hammer_transaction trans;        /* shared transaction */
851         struct hammer_flusher_info_list run_list;
852         struct hammer_flusher_info_list ready_list;
853 };
854
855 #define HAMMER_FLUSH_UNDOS_RELAXED      0
856 #define HAMMER_FLUSH_UNDOS_FORCED       1
857 #define HAMMER_FLUSH_UNDOS_AUTO         2
858 /*
859  * Internal hammer mount data structure
860  */
861 struct hammer_mount {
862         struct mount *mp;
863         /*struct vnode *rootvp;*/
864         struct hammer_ino_rb_tree rb_inos_root;
865         struct hammer_redo_rb_tree rb_redo_root;
866         struct hammer_vol_rb_tree rb_vols_root;
867         struct hammer_nod_rb_tree rb_nods_root;
868         struct hammer_und_rb_tree rb_undo_root;
869         struct hammer_res_rb_tree rb_resv_root;
870         struct hammer_buf_rb_tree rb_bufs_root;
871         struct hammer_pfs_rb_tree rb_pfsm_root;
872
873         struct hammer_dedup_crc_rb_tree rb_dedup_crc_root;
874         struct hammer_dedup_off_rb_tree rb_dedup_off_root;
875
876         struct hammer_volume *rootvol;
877         struct hammer_base_elm root_btree_beg;
878         struct hammer_base_elm root_btree_end;
879
880         struct malloc_type      *m_misc;
881         struct malloc_type      *m_inodes;
882
883         int     flags;          /* HAMMER_MOUNT_xxx flags */
884         int     hflags;
885         int     ronly;
886         int     nvolumes;
887         int     volume_iterator;
888         int     master_id;      /* -1 or 0-15 - clustering and mirroring */
889         int     version;        /* hammer filesystem version to use */
890         int     rsv_inodes;     /* reserved space due to dirty inodes */
891         int64_t rsv_databytes;  /* reserved space due to record data */
892         int     rsv_recs;       /* reserved space due to dirty records */
893         int     rsv_fromdelay;  /* bigblocks reserved due to flush delay */
894         int     undo_rec_limit; /* based on size of undo area */
895         int     last_newrecords;
896         int     count_newrecords;
897
898         int     volume_to_remove; /* volume that is currently being removed */
899
900         int     inode_reclaims; /* inodes pending reclaim by flusher */
901         int     count_inodes;   /* total number of inodes */
902         int     count_iqueued;  /* inodes queued to flusher */
903
904         struct hammer_flusher flusher;
905
906         u_int   check_interrupt;
907         u_int   check_yield;
908         uuid_t  fsid;
909         struct hammer_mod_rb_tree volu_root;    /* dirty undo buffers */
910         struct hammer_mod_rb_tree undo_root;    /* dirty undo buffers */
911         struct hammer_mod_rb_tree data_root;    /* dirty data buffers */
912         struct hammer_mod_rb_tree meta_root;    /* dirty meta bufs    */
913         struct hammer_mod_rb_tree lose_root;    /* loose buffers      */
914         int     locked_dirty_space;             /* meta/volu count    */
915         int     io_running_space;               /* io_token */
916         int     io_running_wakeup;              /* io_token */
917         int     objid_cache_count;
918         int     dedup_cache_count;
919         int     error;                          /* critical I/O error */
920         struct krate    krate;                  /* rate limited kprintf */
921         hammer_tid_t    asof;                   /* snapshot mount */
922         hammer_tid_t    next_tid;
923         hammer_tid_t    flush_tid1;             /* flusher tid sequencing */
924         hammer_tid_t    flush_tid2;             /* flusher tid sequencing */
925         int64_t copy_stat_freebigblocks;        /* number of free bigblocks */
926         u_int32_t       undo_seqno;             /* UNDO/REDO FIFO seqno */
927         u_int32_t       recover_stage2_seqno;   /* REDO recovery seqno */
928         hammer_off_t    recover_stage2_offset;  /* REDO recovery offset */
929
930         struct netexport export;
931         struct hammer_lock sync_lock;
932         struct hammer_lock free_lock;
933         struct hammer_lock undo_lock;
934         struct hammer_lock blkmap_lock;
935         struct hammer_lock snapshot_lock;
936         struct hammer_lock volume_lock;
937         struct hammer_blockmap  blockmap[HAMMER_MAX_ZONES];
938         struct hammer_undo      undos[HAMMER_MAX_UNDOS];
939         int                     undo_alloc;
940         TAILQ_HEAD(, hammer_undo)  undo_lru_list;
941         TAILQ_HEAD(, hammer_reserve) delay_list;
942         struct hammer_flush_group_list  flush_group_list;
943         hammer_flush_group_t    fill_flush_group;
944         hammer_flush_group_t    next_flush_group;
945         TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
946         TAILQ_HEAD(, hammer_dedup_cache) dedup_lru_list;
947         hammer_dedup_cache_t    dedup_free_cache;
948         TAILQ_HEAD(, hammer_reclaim) reclaim_list;
949         TAILQ_HEAD(, hammer_io) iorun_list;
950
951         struct lwkt_token       fs_token;       /* high level */
952         struct lwkt_token       io_token;       /* low level (IO callback) */
953
954         struct hammer_inostats  inostats[HAMMER_INOSTATS_HSIZE];
955 };
956
957 typedef struct hammer_mount     *hammer_mount_t;
958
959 #define HAMMER_MOUNT_CRITICAL_ERROR     0x0001
960 #define HAMMER_MOUNT_FLUSH_RECOVERY     0x0002
961 #define HAMMER_MOUNT_REDO_SYNC          0x0004
962 #define HAMMER_MOUNT_REDO_RECOVERY_REQ  0x0008
963 #define HAMMER_MOUNT_REDO_RECOVERY_RUN  0x0010
964
965 struct hammer_sync_info {
966         int error;
967         int waitfor;
968 };
969
970 /*
971  * Minium buffer cache bufs required to rebalance the B-Tree.
972  * This is because we must hold the children and the children's children
973  * locked.  Even this might not be enough if things are horribly out
974  * of balance.
975  */
976 #define HAMMER_REBALANCE_MIN_BUFS       \
977         (HAMMER_BTREE_LEAF_ELMS * HAMMER_BTREE_LEAF_ELMS)
978
979
980 #endif
981
982 /*
983  * checkspace slop (8MB chunks), higher numbers are more conservative.
984  */
985 #define HAMMER_CHKSPC_REBLOCK   25
986 #define HAMMER_CHKSPC_MIRROR    20
987 #define HAMMER_CHKSPC_WRITE     20
988 #define HAMMER_CHKSPC_CREATE    20
989 #define HAMMER_CHKSPC_REMOVE    10
990 #define HAMMER_CHKSPC_EMERGENCY 0
991
992 #if defined(_KERNEL)
993
994 extern struct vop_ops hammer_vnode_vops;
995 extern struct vop_ops hammer_spec_vops;
996 extern struct vop_ops hammer_fifo_vops;
997 extern struct bio_ops hammer_bioops;
998
999 extern int hammer_debug_io;
1000 extern int hammer_debug_general;
1001 extern int hammer_debug_debug;
1002 extern int hammer_debug_inode;
1003 extern int hammer_debug_locks;
1004 extern int hammer_debug_btree;
1005 extern int hammer_debug_tid;
1006 extern int hammer_debug_recover;
1007 extern int hammer_debug_recover_faults;
1008 extern int hammer_debug_critical;
1009 extern int hammer_cluster_enable;
1010 extern int hammer_live_dedup;
1011 extern int hammer_count_fsyncs;
1012 extern int hammer_count_inodes;
1013 extern int hammer_count_iqueued;
1014 extern int hammer_count_reclaiming;
1015 extern int hammer_count_records;
1016 extern int hammer_count_record_datas;
1017 extern int hammer_count_volumes;
1018 extern int hammer_count_buffers;
1019 extern int hammer_count_nodes;
1020 extern int64_t hammer_count_extra_space_used;
1021 extern int64_t hammer_stats_btree_lookups;
1022 extern int64_t hammer_stats_btree_searches;
1023 extern int64_t hammer_stats_btree_inserts;
1024 extern int64_t hammer_stats_btree_deletes;
1025 extern int64_t hammer_stats_btree_elements;
1026 extern int64_t hammer_stats_btree_splits;
1027 extern int64_t hammer_stats_btree_iterations;
1028 extern int64_t hammer_stats_btree_root_iterations;
1029 extern int64_t hammer_stats_record_iterations;
1030 extern int64_t hammer_stats_file_read;
1031 extern int64_t hammer_stats_file_write;
1032 extern int64_t hammer_stats_file_iopsr;
1033 extern int64_t hammer_stats_file_iopsw;
1034 extern int64_t hammer_stats_disk_read;
1035 extern int64_t hammer_stats_disk_write;
1036 extern int64_t hammer_stats_inode_flushes;
1037 extern int64_t hammer_stats_commits;
1038 extern int64_t hammer_stats_undo;
1039 extern int64_t hammer_stats_redo;
1040 extern int hammer_count_dirtybufspace;
1041 extern int hammer_count_refedbufs;
1042 extern int hammer_count_reservations;
1043 extern int hammer_count_io_running_read;
1044 extern int hammer_count_io_running_write;
1045 extern int hammer_count_io_locked;
1046 extern int hammer_limit_dirtybufspace;
1047 extern int hammer_limit_running_io;
1048 extern int hammer_limit_recs;
1049 extern int hammer_limit_inode_recs;
1050 extern int hammer_limit_reclaim;
1051 extern int hammer_live_dedup_cache_size;
1052 extern int hammer_limit_redo;
1053 extern int hammer_bio_count;
1054 extern int hammer_verify_zone;
1055 extern int hammer_verify_data;
1056 extern int hammer_write_mode;
1057 extern int hammer_double_buffer;
1058 extern int hammer_yield_check;
1059 extern int hammer_fsync_mode;
1060 extern int hammer_autoflush;
1061 extern int64_t hammer_contention_count;
1062
1063 extern int64_t hammer_live_dedup_vnode_bcmps;
1064 extern int64_t hammer_live_dedup_device_bcmps;
1065 extern int64_t hammer_live_dedup_findblk_failures;
1066 extern int64_t hammer_live_dedup_bmap_saves;
1067
1068 void    hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
1069                         int error, const char *msg);
1070 int     hammer_vop_inactive(struct vop_inactive_args *);
1071 int     hammer_vop_reclaim(struct vop_reclaim_args *);
1072 int     hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
1073 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
1074                         hammer_inode_t dip, int64_t obj_id,
1075                         hammer_tid_t asof, u_int32_t localization,
1076                         int flags, int *errorp);
1077 struct hammer_inode *hammer_get_dummy_inode(hammer_transaction_t trans,
1078                         hammer_inode_t dip, int64_t obj_id,
1079                         hammer_tid_t asof, u_int32_t localization,
1080                         int flags, int *errorp);
1081 struct hammer_inode *hammer_find_inode(hammer_transaction_t trans,
1082                         int64_t obj_id, hammer_tid_t asof,
1083                         u_int32_t localization);
1084 void    hammer_scan_inode_snapshots(hammer_mount_t hmp,
1085                         hammer_inode_info_t iinfo,
1086                         int (*callback)(hammer_inode_t ip, void *data),
1087                         void *data);
1088 void    hammer_put_inode(struct hammer_inode *ip);
1089 void    hammer_put_inode_ref(struct hammer_inode *ip);
1090 void    hammer_inode_waitreclaims(hammer_transaction_t trans);
1091
1092 int     hammer_unload_volume(hammer_volume_t volume, void *data __unused);
1093 int     hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
1094
1095 int     hammer_unload_buffer(hammer_buffer_t buffer, void *data);
1096 int     hammer_install_volume(hammer_mount_t hmp, const char *volname,
1097                         struct vnode *devvp);
1098 int     hammer_mountcheck_volumes(hammer_mount_t hmp);
1099
1100 int     hammer_mem_add(hammer_record_t record);
1101 int     hammer_ip_lookup(hammer_cursor_t cursor);
1102 int     hammer_ip_first(hammer_cursor_t cursor);
1103 int     hammer_ip_next(hammer_cursor_t cursor);
1104 int     hammer_ip_resolve_data(hammer_cursor_t cursor);
1105 int     hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
1106                         hammer_tid_t tid);
1107 int     hammer_create_at_cursor(hammer_cursor_t cursor,
1108                         hammer_btree_leaf_elm_t leaf, void *udata, int mode);
1109 int     hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
1110                         hammer_tid_t delete_tid, u_int32_t delete_ts,
1111                         int track, int64_t *stat_bytes);
1112 int     hammer_ip_check_directory_empty(hammer_transaction_t trans,
1113                         hammer_inode_t ip);
1114 int     hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
1115 int     hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
1116
1117 hammer_record_t
1118         hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
1119 void    hammer_flush_record_done(hammer_record_t record, int error);
1120 void    hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
1121 void    hammer_rel_mem_record(hammer_record_t record);
1122
1123 int     hammer_cursor_up(hammer_cursor_t cursor);
1124 int     hammer_cursor_up_locked(hammer_cursor_t cursor);
1125 int     hammer_cursor_down(hammer_cursor_t cursor);
1126 int     hammer_cursor_upgrade(hammer_cursor_t cursor);
1127 int     hammer_cursor_upgrade_node(hammer_cursor_t cursor);
1128 void    hammer_cursor_downgrade(hammer_cursor_t cursor);
1129 int     hammer_cursor_upgrade2(hammer_cursor_t c1, hammer_cursor_t c2);
1130 void    hammer_cursor_downgrade2(hammer_cursor_t c1, hammer_cursor_t c2);
1131 int     hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
1132                         int index);
1133 void    hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
1134 int     hammer_lock_ex_try(struct hammer_lock *lock);
1135 void    hammer_lock_sh(struct hammer_lock *lock);
1136 int     hammer_lock_sh_try(struct hammer_lock *lock);
1137 int     hammer_lock_upgrade(struct hammer_lock *lock, int shcount);
1138 void    hammer_lock_downgrade(struct hammer_lock *lock, int shcount);
1139 int     hammer_lock_status(struct hammer_lock *lock);
1140 void    hammer_unlock(struct hammer_lock *lock);
1141 void    hammer_ref(struct hammer_lock *lock);
1142 int     hammer_ref_interlock(struct hammer_lock *lock);
1143 int     hammer_ref_interlock_true(struct hammer_lock *lock);
1144 void    hammer_ref_interlock_done(struct hammer_lock *lock);
1145 void    hammer_rel(struct hammer_lock *lock);
1146 int     hammer_rel_interlock(struct hammer_lock *lock, int locked);
1147 void    hammer_rel_interlock_done(struct hammer_lock *lock, int orig_locked);
1148 int     hammer_get_interlock(struct hammer_lock *lock);
1149 int     hammer_try_interlock_norefs(struct hammer_lock *lock);
1150 void    hammer_put_interlock(struct hammer_lock *lock, int error);
1151
1152 void    hammer_sync_lock_ex(hammer_transaction_t trans);
1153 void    hammer_sync_lock_sh(hammer_transaction_t trans);
1154 int     hammer_sync_lock_sh_try(hammer_transaction_t trans);
1155 void    hammer_sync_unlock(hammer_transaction_t trans);
1156
1157 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
1158 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
1159 void    hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
1160 u_int64_t hammer_timespec_to_time(struct timespec *ts);
1161 int     hammer_str_to_tid(const char *str, int *ispfsp,
1162                         hammer_tid_t *tidp, u_int32_t *localizationp);
1163 int     hammer_is_atatext(const char *name, int len);
1164 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip,
1165                         int64_t namekey);
1166 void hammer_clear_objid(hammer_inode_t dip);
1167 void hammer_destroy_objid_cache(hammer_mount_t hmp);
1168
1169 int hammer_dedup_crc_rb_compare(hammer_dedup_cache_t dc1,
1170                         hammer_dedup_cache_t dc2);
1171 int hammer_dedup_off_rb_compare(hammer_dedup_cache_t dc1,
1172                         hammer_dedup_cache_t dc2);
1173 hammer_dedup_cache_t hammer_dedup_cache_add(hammer_inode_t ip,
1174                         hammer_btree_leaf_elm_t leaf);
1175 hammer_dedup_cache_t hammer_dedup_cache_lookup(hammer_mount_t hmp,
1176                         hammer_crc_t crc);
1177 void hammer_dedup_cache_inval(hammer_mount_t hmp, hammer_off_t base_offset);
1178 void hammer_destroy_dedup_cache(hammer_mount_t hmp);
1179 void hammer_dump_dedup_cache(hammer_mount_t hmp);
1180 int hammer_dedup_validate(hammer_dedup_cache_t dcp, int zone, int bytes,
1181                         void *data);
1182
1183 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
1184                         int bytes);
1185 void hammer_clear_undo_history(hammer_mount_t hmp);
1186 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
1187 int hammer_get_dtype(u_int8_t obj_type);
1188 u_int8_t hammer_get_obj_type(enum vtype vtype);
1189 int64_t hammer_directory_namekey(hammer_inode_t dip, const void *name, int len,
1190                         u_int32_t *max_iterationsp);
1191 int     hammer_nohistory(hammer_inode_t ip);
1192
1193 int     hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
1194                         hammer_node_cache_t cache, hammer_inode_t ip);
1195 void    hammer_normalize_cursor(hammer_cursor_t cursor);
1196 void    hammer_done_cursor(hammer_cursor_t cursor);
1197 int     hammer_recover_cursor(hammer_cursor_t cursor);
1198 void    hammer_unlock_cursor(hammer_cursor_t cursor);
1199 int     hammer_lock_cursor(hammer_cursor_t cursor);
1200 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor);
1201 void    hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
1202
1203 void    hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
1204 void    hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
1205                         int index);
1206 void    hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
1207                         int index);
1208 void    hammer_cursor_moved_element(hammer_node_t oparent, int pindex,
1209                         hammer_node_t onode, int oindex,
1210                         hammer_node_t nnode, int nindex);
1211 void    hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent,
1212                         hammer_node_t nparent, int nindex);
1213 void    hammer_cursor_inserted_element(hammer_node_t node, int index);
1214 void    hammer_cursor_deleted_element(hammer_node_t node, int index);
1215 void    hammer_cursor_invalidate_cache(hammer_cursor_t cursor);
1216
1217 int     hammer_btree_lookup(hammer_cursor_t cursor);
1218 int     hammer_btree_first(hammer_cursor_t cursor);
1219 int     hammer_btree_last(hammer_cursor_t cursor);
1220 int     hammer_btree_extract(hammer_cursor_t cursor, int flags);
1221 int     hammer_btree_iterate(hammer_cursor_t cursor);
1222 int     hammer_btree_iterate_reverse(hammer_cursor_t cursor);
1223 int     hammer_btree_insert(hammer_cursor_t cursor,
1224                             hammer_btree_leaf_elm_t elm, int *doprop);
1225 int     hammer_btree_delete(hammer_cursor_t cursor);
1226 void    hammer_btree_do_propagation(hammer_cursor_t cursor,
1227                             hammer_pseudofs_inmem_t pfsm,
1228                             hammer_btree_leaf_elm_t leaf);
1229 int     hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
1230 int     hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
1231 int     hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
1232 int     hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
1233
1234 int     btree_set_parent(hammer_transaction_t trans, hammer_node_t node,
1235                         hammer_btree_elm_t elm);
1236 void    hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node);
1237 void    hammer_btree_lcache_init(hammer_mount_t hmp, hammer_node_lock_t lcache,
1238                         int depth);
1239 void    hammer_btree_lcache_free(hammer_mount_t hmp, hammer_node_lock_t lcache);
1240 int     hammer_btree_lock_children(hammer_cursor_t cursor, int depth,
1241                         hammer_node_lock_t parent,
1242                         hammer_node_lock_t lcache);
1243 void    hammer_btree_lock_copy(hammer_cursor_t cursor,
1244                         hammer_node_lock_t parent);
1245 int     hammer_btree_sync_copy(hammer_cursor_t cursor,
1246                         hammer_node_lock_t parent);
1247 void    hammer_btree_unlock_children(hammer_mount_t hmp,
1248                         hammer_node_lock_t parent,
1249                         hammer_node_lock_t lcache);
1250 int     hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
1251 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans,
1252                         hammer_node_t node, int *parent_indexp,
1253                         int *errorp, int try_exclusive);
1254
1255 void    hammer_print_btree_node(hammer_node_ondisk_t ondisk);
1256 void    hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
1257
1258 void    *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
1259                         int *errorp, struct hammer_buffer **bufferp);
1260 void    *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
1261                         int *errorp, struct hammer_buffer **bufferp);
1262 void    *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1263                         int *errorp, struct hammer_buffer **bufferp);
1264 void    *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1265                         int *errorp, struct hammer_buffer **bufferp);
1266
1267 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
1268
1269 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
1270                         int32_t vol_no, int *errorp);
1271 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
1272                         int bytes, int isnew, int *errorp);
1273 void            hammer_sync_buffers(hammer_mount_t hmp,
1274                         hammer_off_t base_offset, int bytes);
1275 int             hammer_del_buffers(hammer_mount_t hmp,
1276                         hammer_off_t base_offset,
1277                         hammer_off_t zone2_offset, int bytes,
1278                         int report_conflicts);
1279
1280 int             hammer_ref_volume(hammer_volume_t volume);
1281 int             hammer_ref_buffer(hammer_buffer_t buffer);
1282 void            hammer_flush_buffer_nodes(hammer_buffer_t buffer);
1283
1284 void            hammer_rel_volume(hammer_volume_t volume, int locked);
1285 void            hammer_rel_buffer(hammer_buffer_t buffer, int locked);
1286
1287 int             hammer_vfs_export(struct mount *mp, int op,
1288                         const struct export_args *export);
1289 hammer_node_t   hammer_get_node(hammer_transaction_t trans,
1290                         hammer_off_t node_offset, int isnew, int *errorp);
1291 void            hammer_ref_node(hammer_node_t node);
1292 hammer_node_t   hammer_ref_node_safe(hammer_transaction_t trans,
1293                         hammer_node_cache_t cache, int *errorp);
1294 void            hammer_rel_node(hammer_node_t node);
1295 void            hammer_delete_node(hammer_transaction_t trans,
1296                         hammer_node_t node);
1297 void            hammer_cache_node(hammer_node_cache_t cache,
1298                         hammer_node_t node);
1299 void            hammer_uncache_node(hammer_node_cache_t cache);
1300 void            hammer_flush_node(hammer_node_t node, int locked);
1301
1302 void hammer_dup_buffer(struct hammer_buffer **bufferp,
1303                         struct hammer_buffer *buffer);
1304 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans,
1305                         hammer_off_t hint, int *errorp);
1306 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1307                         u_int16_t rec_type, hammer_off_t *data_offsetp,
1308                         struct hammer_buffer **data_bufferp,
1309                         hammer_off_t hint, int *errorp);
1310
1311 int hammer_generate_undo(hammer_transaction_t trans,
1312                         hammer_off_t zone1_offset, void *base, int len);
1313 int hammer_generate_redo(hammer_transaction_t trans, hammer_inode_t ip,
1314                         hammer_off_t file_offset, u_int32_t flags,
1315                         void *base, int len);
1316 void hammer_generate_redo_sync(hammer_transaction_t trans);
1317 void hammer_redo_fifo_start_flush(hammer_inode_t ip);
1318 void hammer_redo_fifo_end_flush(hammer_inode_t ip);
1319
1320 void hammer_format_undo(void *base, u_int32_t seqno);
1321 int hammer_upgrade_undo_4(hammer_transaction_t trans);
1322
1323 void hammer_put_volume(struct hammer_volume *volume, int flush);
1324 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
1325
1326 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1327                         hammer_off_t owner, int *errorp);
1328 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1329                         hammer_off_t owner, int *errorp);
1330 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp);
1331 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1332                         int bytes, hammer_off_t hint, int *errorp);
1333 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1334                         int bytes, hammer_off_t *zone_offp, int *errorp);
1335 hammer_reserve_t hammer_blockmap_reserve_dedup(hammer_mount_t hmp, int zone,
1336                         int bytes, hammer_off_t zone_offset, int *errorp);
1337 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1338                         hammer_reserve_t resv);
1339 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1340 void hammer_blockmap_free(hammer_transaction_t trans,
1341                         hammer_off_t bmap_off, int bytes);
1342 int hammer_blockmap_dedup(hammer_transaction_t trans,
1343                         hammer_off_t bmap_off, int bytes);
1344 int hammer_blockmap_finalize(hammer_transaction_t trans,
1345                         hammer_reserve_t resv,
1346                         hammer_off_t bmap_off, int bytes);
1347 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
1348                         int *curp, int *errorp);
1349 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1350                         int *errorp);
1351 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1352                         int *errorp);
1353 int64_t hammer_undo_used(hammer_transaction_t trans);
1354 int64_t hammer_undo_space(hammer_transaction_t trans);
1355 int64_t hammer_undo_max(hammer_mount_t hmp);
1356 int hammer_undo_reclaim(hammer_io_t io);
1357
1358 void hammer_start_transaction(struct hammer_transaction *trans,
1359                               struct hammer_mount *hmp);
1360 void hammer_simple_transaction(struct hammer_transaction *trans,
1361                               struct hammer_mount *hmp);
1362 void hammer_start_transaction_fls(struct hammer_transaction *trans,
1363                                   struct hammer_mount *hmp);
1364 void hammer_done_transaction(struct hammer_transaction *trans);
1365 hammer_tid_t hammer_alloc_tid(hammer_mount_t hmp, int count);
1366
1367 void hammer_modify_inode(hammer_transaction_t trans, hammer_inode_t ip, int flags);
1368 void hammer_flush_inode(hammer_inode_t ip, int flags);
1369 void hammer_flush_inode_done(hammer_inode_t ip, int error);
1370 void hammer_wait_inode(hammer_inode_t ip);
1371
1372 int  hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
1373                         struct ucred *cred, struct hammer_inode *dip,
1374                         const char *name, int namelen,
1375                         hammer_pseudofs_inmem_t pfsm,
1376                         struct hammer_inode **ipp);
1377 void hammer_rel_inode(hammer_inode_t ip, int flush);
1378 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1379 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1380 int hammer_redo_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1381 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1382
1383 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1384 void hammer_test_inode(hammer_inode_t dip);
1385 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1386
1387 int  hammer_ip_add_directory(struct hammer_transaction *trans,
1388                         hammer_inode_t dip, const char *name, int bytes,
1389                         hammer_inode_t nip);
1390 int  hammer_ip_del_directory(struct hammer_transaction *trans,
1391                         hammer_cursor_t cursor, hammer_inode_t dip,
1392                         hammer_inode_t ip);
1393 void hammer_ip_replace_bulk(hammer_mount_t hmp, hammer_record_t record);
1394 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1395                         void *data, int bytes, int *errorp);
1396 int  hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
1397 int  hammer_ip_add_record(struct hammer_transaction *trans,
1398                         hammer_record_t record);
1399 int  hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1400                         int64_t ran_beg, int64_t ran_end, int truncating);
1401 int  hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1402                         int *countp);
1403 int  hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1404                         int64_t offset, void *data, int bytes);
1405 int  hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
1406 int  hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1407 hammer_pseudofs_inmem_t  hammer_load_pseudofs(hammer_transaction_t trans,
1408                         u_int32_t localization, int *errorp);
1409 int  hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1410                         hammer_pseudofs_inmem_t pfsm);
1411 int  hammer_save_pseudofs(hammer_transaction_t trans,
1412                         hammer_pseudofs_inmem_t pfsm);
1413 int  hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization);
1414 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1415 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1416                         struct ucred *cred);
1417
1418 void hammer_io_init(hammer_io_t io, hammer_volume_t volume,
1419                         enum hammer_io_type type);
1420 int hammer_io_read(struct vnode *devvp, struct hammer_io *io, int limit);
1421 void hammer_io_advance(struct hammer_io *io);
1422 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
1423 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1424 struct buf *hammer_io_release(struct hammer_io *io, int flush);
1425 void hammer_io_flush(struct hammer_io *io, int reclaim);
1426 void hammer_io_wait(struct hammer_io *io);
1427 void hammer_io_waitdep(struct hammer_io *io);
1428 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident, int doflush);
1429 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1430                         hammer_btree_leaf_elm_t leaf);
1431 int hammer_io_indirect_read(hammer_mount_t hmp, struct bio *bio,
1432                         hammer_btree_leaf_elm_t leaf);
1433 int hammer_io_direct_write(hammer_mount_t hmp, struct bio *bio,
1434                         hammer_record_t record);
1435 void hammer_io_direct_wait(hammer_record_t record);
1436 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1437 void hammer_io_write_interlock(hammer_io_t io);
1438 void hammer_io_done_interlock(hammer_io_t io);
1439 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1440 void hammer_io_clear_modlist(struct hammer_io *io);
1441 void hammer_io_flush_sync(hammer_mount_t hmp);
1442 void hammer_io_clear_error(struct hammer_io *io);
1443 void hammer_io_clear_error_noassert(struct hammer_io *io);
1444 void hammer_io_notmeta(hammer_buffer_t buffer);
1445 void hammer_io_limit_backlog(hammer_mount_t hmp);
1446
1447 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1448                         void *base, int len);
1449 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1450                         void *base, int len);
1451 void hammer_modify_volume_done(hammer_volume_t volume);
1452 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1453
1454 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1455                         struct hammer_ioc_reblock *reblock);
1456 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip,
1457                         struct hammer_ioc_rebalance *rebal);
1458 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1459                         struct hammer_ioc_prune *prune);
1460 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1461                         struct hammer_ioc_mirror_rw *mirror);
1462 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1463                         struct hammer_ioc_mirror_rw *mirror);
1464 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1465                         struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1466 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1467                         struct hammer_ioc_pseudofs_rw *pfs);
1468 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1469                         struct hammer_ioc_pseudofs_rw *pfs);
1470 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1471                         struct hammer_ioc_pseudofs_rw *pfs);
1472 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1473                         struct hammer_ioc_pseudofs_rw *pfs);
1474 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1475                         struct hammer_ioc_pseudofs_rw *pfs);
1476 int hammer_ioc_volume_add(hammer_transaction_t trans, hammer_inode_t ip,
1477                         struct hammer_ioc_volume *ioc);
1478 int hammer_ioc_volume_del(hammer_transaction_t trans, hammer_inode_t ip,
1479                         struct hammer_ioc_volume *ioc);
1480 int hammer_ioc_volume_list(hammer_transaction_t trans, hammer_inode_t ip,
1481                         struct hammer_ioc_volume_list *ioc);
1482 int hammer_ioc_dedup(hammer_transaction_t trans, hammer_inode_t ip,
1483                         struct hammer_ioc_dedup *dedup);
1484
1485 int hammer_signal_check(hammer_mount_t hmp);
1486
1487 void hammer_flusher_create(hammer_mount_t hmp);
1488 void hammer_flusher_destroy(hammer_mount_t hmp);
1489 void hammer_flusher_sync(hammer_mount_t hmp);
1490 int  hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1491 int  hammer_flusher_async_one(hammer_mount_t hmp);
1492 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1493 void hammer_flusher_wait_next(hammer_mount_t hmp);
1494 int  hammer_flusher_meta_limit(hammer_mount_t hmp);
1495 int  hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1496 int  hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1497 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1498 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1499 int  hammer_flusher_haswork(hammer_mount_t hmp);
1500 void hammer_flusher_flush_undos(hammer_mount_t hmp, int already_flushed);
1501
1502 int hammer_recover_stage1(hammer_mount_t hmp, hammer_volume_t rootvol);
1503 int hammer_recover_stage2(hammer_mount_t hmp, hammer_volume_t rootvol);
1504 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1505                         hammer_volume_t root_volume, int final);
1506
1507 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1508 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1509 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1510
1511 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1512 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1513 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1514 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1515 void hkprintf(const char *ctl, ...) __printflike(1, 2);
1516 udev_t hammer_fsid_to_udev(uuid_t *uuid);
1517
1518
1519 int hammer_blocksize(int64_t file_offset);
1520 int hammer_blockoff(int64_t file_offset);
1521 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1522
1523 /*
1524  * Shortcut for _hammer_checkspace(), used all over the code.
1525  */
1526 static __inline int
1527 hammer_checkspace(hammer_mount_t hmp, int slop)
1528 {
1529         return(_hammer_checkspace(hmp, slop, NULL));
1530 }
1531
1532 #endif
1533
1534 static __inline void
1535 hammer_wait_mem_record(hammer_record_t record)
1536 {
1537         hammer_wait_mem_record_ident(record, "hmmwai");
1538 }
1539
1540 static __inline void
1541 hammer_lock_ex(struct hammer_lock *lock)
1542 {
1543         hammer_lock_ex_ident(lock, "hmrlck");
1544 }
1545
1546 /*
1547  * Indicate that a B-Tree node is being modified.
1548  */
1549 static __inline void
1550 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1551 {
1552         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1553         hammer_modify_buffer(trans, node->buffer, NULL, 0);
1554 }
1555
1556 static __inline void
1557 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1558 {
1559         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1560         hammer_modify_buffer(trans, node->buffer,
1561                              node->ondisk, sizeof(*node->ondisk));
1562 }
1563
1564 static __inline void
1565 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1566                    void *base, int len)
1567 {
1568         hammer_crc_t *crcptr;
1569
1570         KKASSERT((char *)base >= (char *)node->ondisk &&
1571                  (char *)base + len <=
1572                     (char *)node->ondisk + sizeof(*node->ondisk));
1573         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1574         hammer_modify_buffer(trans, node->buffer, base, len);
1575         crcptr = &node->ondisk->crc;
1576         hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t));
1577         --node->buffer->io.modify_refs; /* only want one ref */
1578 }
1579
1580 /*
1581  * Indicate that the specified modifications have been completed.
1582  *
1583  * Do not try to generate the crc here, it's very expensive to do and a
1584  * sequence of insertions or deletions can result in many calls to this
1585  * function on the same node.
1586  */
1587 static __inline void
1588 hammer_modify_node_done(hammer_node_t node)
1589 {
1590         node->flags |= HAMMER_NODE_CRCGOOD;
1591         if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1592                 node->flags |= HAMMER_NODE_NEEDSCRC;
1593                 node->buffer->io.gencrc = 1;
1594                 hammer_ref_node(node);
1595         }
1596         hammer_modify_buffer_done(node->buffer);
1597 }
1598
1599 #define hammer_modify_volume_field(trans, vol, field)           \
1600         hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1601                              sizeof((vol)->ondisk->field))
1602
1603 #define hammer_modify_node_field(trans, node, field)            \
1604         hammer_modify_node(trans, node, &(node)->ondisk->field, \
1605                              sizeof((node)->ondisk->field))
1606
1607 /*
1608  * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1609  * created directories for HAMMER version 2 or greater and causes
1610  * directory entries to be placed the inode localization zone in
1611  * the B-Tree instead of the misc zone.
1612  *
1613  * This greatly improves localization between directory entries and
1614  * inodes
1615  */
1616 static __inline u_int32_t
1617 hammer_dir_localization(hammer_inode_t dip)
1618 {
1619         if (dip->ino_data.cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO)
1620                 return(HAMMER_LOCALIZE_INODE);
1621         else
1622                 return(HAMMER_LOCALIZE_MISC);
1623 }