HAMMER VFS - REDO implementation base code part 2/many
[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
62 #include <sys/buf2.h>
63 #include <sys/signal2.h>
64 #include <sys/mplock2.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         int             refs;           /* active references delay writes */
130         volatile u_int  lockval;        /* lock count and control bits */
131         struct thread   *owner;         /* owner if exclusively held */
132 };
133
134 #define HAMMER_LOCKF_EXCLUSIVE  0x40000000
135 #define HAMMER_LOCKF_WANTED     0x80000000
136
137 static __inline int
138 hammer_notlocked(struct hammer_lock *lock)
139 {
140         return(lock->lockval == 0);
141 }
142
143 static __inline int
144 hammer_islocked(struct hammer_lock *lock)
145 {
146         return(lock->lockval != 0);
147 }
148
149 static __inline int
150 hammer_isactive(struct hammer_lock *lock)
151 {
152         return(lock->refs != 0);
153 }
154
155 static __inline int
156 hammer_islastref(struct hammer_lock *lock)
157 {
158         return(lock->refs == 1);
159 }
160
161 /*
162  * Return if we specifically own the lock exclusively.
163  */
164 static __inline int
165 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
166 {
167         if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) &&
168             lock->owner == td) {
169                 return(1);
170         }
171         return(0);
172 }
173
174 /*
175  * Flush state, used by various structures
176  */
177 typedef enum hammer_inode_state {
178         HAMMER_FST_IDLE,
179         HAMMER_FST_SETUP,
180         HAMMER_FST_FLUSH
181 } hammer_inode_state_t;
182
183 TAILQ_HEAD(hammer_record_list, hammer_record);
184
185 /*
186  * Pseudo-filesystem extended data tracking
187  */
188 struct hammer_pfs_rb_tree;
189 struct hammer_pseudofs_inmem;
190 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
191 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
192               hammer_pfs_rb_compare, u_int32_t);
193
194 struct hammer_pseudofs_inmem {
195         RB_ENTRY(hammer_pseudofs_inmem) rb_node;
196         struct hammer_lock      lock;
197         u_int32_t               localization;
198         hammer_tid_t            create_tid;
199         int                     flags;
200         udev_t                  fsid_udev;
201         struct hammer_pseudofs_data pfsd;
202 };
203
204 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t;
205
206 #define HAMMER_PFSM_DELETED     0x0001
207
208 /*
209  * Cache object ids.  A fixed number of objid cache structures are
210  * created to reserve object id's for newly created files in multiples
211  * of 100,000, localized to a particular directory, and recycled as
212  * needed.  This allows parallel create operations in different
213  * directories to retain fairly localized object ids which in turn
214  * improves reblocking performance and layout.
215  */
216 #define OBJID_CACHE_SIZE        1024
217 #define OBJID_CACHE_BULK_BITS   10              /* 10 bits (1024)       */
218 #define OBJID_CACHE_BULK        (32 * 32)       /* two level (1024)     */
219 #define OBJID_CACHE_BULK_MASK   (OBJID_CACHE_BULK - 1)
220 #define OBJID_CACHE_BULK_MASK64 ((u_int64_t)(OBJID_CACHE_BULK - 1))
221
222 typedef struct hammer_objid_cache {
223         TAILQ_ENTRY(hammer_objid_cache) entry;
224         struct hammer_inode             *dip;
225         hammer_tid_t                    base_tid;
226         int                             count;
227         u_int32_t                       bm0;
228         u_int32_t                       bm1[32];
229 } *hammer_objid_cache_t;
230
231 /*
232  * Associate an inode with a B-Tree node to cache search start positions
233  */
234 typedef struct hammer_node_cache {
235         TAILQ_ENTRY(hammer_node_cache)  entry;
236         struct hammer_node              *node;
237         struct hammer_inode             *ip;
238 } *hammer_node_cache_t;
239
240 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
241
242 /*
243  * Structure used to organize flush groups.  Flush groups must be
244  * organized into chunks in order to avoid blowing out the UNDO FIFO.
245  * Without this a 'sync' could end up flushing 50,000 inodes in a single
246  * transaction.
247  */
248 struct hammer_fls_rb_tree;
249 RB_HEAD(hammer_fls_rb_tree, hammer_inode);
250 RB_PROTOTYPE(hammer_fls_rb_tree, hammer_inode, rb_flsnode,
251               hammer_ino_rb_compare);
252
253 struct hammer_flush_group {
254         TAILQ_ENTRY(hammer_flush_group) flush_entry;
255         struct hammer_fls_rb_tree       flush_tree;
256         int                             unused01;
257         int                             total_count;    /* record load */
258         int                             running;        /* group is running */
259         int                             closed;
260         int                             refs;
261 };
262
263 typedef struct hammer_flush_group *hammer_flush_group_t;
264
265 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
266
267 /*
268  * Structure used to represent an inode in-memory.
269  *
270  * The record and data associated with an inode may be out of sync with
271  * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
272  * clear).
273  *
274  * An inode may also hold a cache of unsynchronized records, used for
275  * database and directories only.  Unsynchronized regular file data is
276  * stored in the buffer cache.
277  *
278  * NOTE: A file which is created and destroyed within the initial
279  * synchronization period can wind up not doing any disk I/O at all.
280  *
281  * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
282  */
283 struct hammer_ino_rb_tree;
284 struct hammer_inode;
285 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
286 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
287               hammer_ino_rb_compare, hammer_inode_info_t);
288
289 struct hammer_rec_rb_tree;
290 struct hammer_record;
291 RB_HEAD(hammer_rec_rb_tree, hammer_record);
292 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
293               hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
294
295 TAILQ_HEAD(hammer_node_list, hammer_node);
296
297 struct hammer_inode {
298         RB_ENTRY(hammer_inode)  rb_node;
299         hammer_inode_state_t    flush_state;
300         hammer_flush_group_t    flush_group;
301         RB_ENTRY(hammer_inode)  rb_flsnode;     /* when on flush list */
302         struct hammer_record_list target_list;  /* target of dependant recs */
303         int64_t                 obj_id;         /* (key) object identifier */
304         hammer_tid_t            obj_asof;       /* (key) snapshot or 0 */
305         u_int32_t               obj_localization; /* (key) pseudo-fs */
306         struct hammer_mount     *hmp;
307         hammer_objid_cache_t    objid_cache;
308         int                     flags;
309         int                     error;          /* flush error */
310         int                     cursor_ip_refs; /* sanity */
311         int                     rsv_recs;
312         struct vnode            *vp;
313         hammer_pseudofs_inmem_t pfsm;
314         struct lockf            advlock;
315         struct hammer_lock      lock;           /* sync copy interlock */
316         off_t                   trunc_off;
317         struct hammer_btree_leaf_elm ino_leaf;  /* in-memory cache */
318         struct hammer_inode_data ino_data;      /* in-memory cache */
319         struct hammer_rec_rb_tree rec_tree;     /* in-memory cache */
320         int                     rec_generation;
321         struct hammer_node_cache cache[4];      /* search initiate cache */
322
323         /*
324          * When a demark is created to synchronize an inode to
325          * disk, certain fields are copied so the front-end VOPs
326          * can continue to run in parallel with the synchronization
327          * occuring in the background.
328          */
329         int             sync_flags;             /* to-sync flags cache */
330         off_t           sync_trunc_off;         /* to-sync truncation */
331         off_t           save_trunc_off;         /* write optimization */
332         struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
333         struct hammer_inode_data sync_ino_data; /* to-sync cache */
334         size_t          redo_count;
335 };
336
337 typedef struct hammer_inode *hammer_inode_t;
338
339 #define VTOI(vp)        ((struct hammer_inode *)(vp)->v_data)
340
341 /*
342  * NOTE: DDIRTY does not include atime or mtime and does not include
343  *       write-append size changes.  SDIRTY handles write-append size
344  *       changes.
345  *
346  *       REDO indicates that REDO logging is active, creating a definitive
347  *       stream of REDO records in the UNDO/REDO log for writes and
348  *       truncations, including boundary records when/if REDO is turned off.
349  *       REDO is typically enabled by fsync() and turned off if excessive
350  *       writes without an fsync() occurs.
351  *
352  *       RDIRTY indicates that REDO records were laid down in the UNDO/REDO
353  *       FIFO (even if REDO is turned off some might still be active) and
354  *       still being tracked for this inode.  See hammer_redo.c
355  */
356                                         /* (not including atime/mtime) */
357 #define HAMMER_INODE_DDIRTY     0x0001  /* in-memory ino_data is dirty */
358 #define HAMMER_INODE_RSV_INODES 0x0002  /* hmp->rsv_inodes bumped */
359 #define HAMMER_INODE_CONN_DOWN  0x0004  /* include in downward recursion */
360 #define HAMMER_INODE_XDIRTY     0x0008  /* in-memory records */
361 #define HAMMER_INODE_ONDISK     0x0010  /* inode is on-disk (else not yet) */
362 #define HAMMER_INODE_FLUSH      0x0020  /* flush on last ref */
363 #define HAMMER_INODE_DELETED    0x0080  /* inode delete (backend) */
364 #define HAMMER_INODE_DELONDISK  0x0100  /* delete synchronized to disk */
365 #define HAMMER_INODE_RO         0x0200  /* read-only (because of as-of) */
366 #define HAMMER_INODE_VHELD      0x0400  /* vnode held on sync */
367 #define HAMMER_INODE_DONDISK    0x0800  /* data records may be on disk */
368 #define HAMMER_INODE_BUFS       0x1000  /* dirty high level bps present */
369 #define HAMMER_INODE_REFLUSH    0x2000  /* flush on dependancy / reflush */
370 #define HAMMER_INODE_RECLAIM    0x4000  /* trying to reclaim */
371 #define HAMMER_INODE_FLUSHW     0x8000  /* Someone waiting for flush */
372
373 #define HAMMER_INODE_TRUNCATED  0x00010000
374 #define HAMMER_INODE_DELETING   0x00020000 /* inode delete request (frontend)*/
375 #define HAMMER_INODE_RESIGNAL   0x00040000 /* re-signal on re-flush */
376 #define HAMMER_INODE_ATIME      0x00100000 /* in-memory atime modified */
377 #define HAMMER_INODE_MTIME      0x00200000 /* in-memory mtime modified */
378 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
379 #define HAMMER_INODE_DUMMY      0x00800000 /* dummy inode covering bad file */
380 #define HAMMER_INODE_SDIRTY     0x01000000 /* in-memory ino_data.size is dirty*/
381 #define HAMMER_INODE_REDO       0x02000000 /* REDO logging active */
382 #define HAMMER_INODE_RDIRTY     0x04000000 /* REDO records active in fifo */
383
384 #define HAMMER_INODE_MODMASK    (HAMMER_INODE_DDIRTY|HAMMER_INODE_SDIRTY|   \
385                                  HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS|     \
386                                  HAMMER_INODE_ATIME|HAMMER_INODE_MTIME|     \
387                                  HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
388
389 #define HAMMER_INODE_MODMASK_NOXDIRTY   \
390                                 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
391
392 #define HAMMER_INODE_MODMASK_NOREDO     \
393                                 (HAMMER_INODE_DDIRTY|                       \
394                                  HAMMER_INODE_XDIRTY|                       \
395                                  HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
396
397 #define HAMMER_FLUSH_SIGNAL     0x0001
398 #define HAMMER_FLUSH_RECURSION  0x0002
399
400 /*
401  * Used by the inode reclaim code to pipeline reclaims and avoid
402  * blowing out kernel memory or letting the flusher get too far
403  * behind.  The reclaim wakes up when count reaches 0 or the
404  * timer expires.
405  */
406 struct hammer_reclaim {
407         TAILQ_ENTRY(hammer_reclaim) entry;
408         int     count;
409 };
410
411 #define HAMMER_RECLAIM_WAIT     4000    /* default vfs.hammer.limit_reclaim */
412
413 /*
414  * Structure used to represent an unsynchronized record in-memory.  These
415  * records typically represent directory entries.  Only non-historical
416  * records are kept in-memory.
417  *
418  * Records are organized as a per-inode RB-Tree.  If the inode is not
419  * on disk then neither are any records and the in-memory record tree
420  * represents the entire contents of the inode.  If the inode is on disk
421  * then the on-disk B-Tree is scanned in parallel with the in-memory
422  * RB-Tree to synthesize the current state of the file.
423  *
424  * Records are also used to enforce the ordering of directory create/delete
425  * operations.  A new inode will not be flushed to disk unless its related
426  * directory entry is also being flushed at the same time.  A directory entry
427  * will not be removed unless its related inode is also being removed at the
428  * same time.
429  */
430 typedef enum hammer_record_type {
431         HAMMER_MEM_RECORD_GENERAL,      /* misc record */
432         HAMMER_MEM_RECORD_INODE,        /* inode record */
433         HAMMER_MEM_RECORD_ADD,          /* positive memory cache record */
434         HAMMER_MEM_RECORD_DEL,          /* negative delete-on-disk record */
435         HAMMER_MEM_RECORD_DATA          /* bulk-data record w/on-disk ref */
436 } hammer_record_type_t;
437
438 struct hammer_record {
439         RB_ENTRY(hammer_record)         rb_node;
440         TAILQ_ENTRY(hammer_record)      target_entry;
441         hammer_inode_state_t            flush_state;
442         hammer_flush_group_t            flush_group;
443         hammer_record_type_t            type;
444         struct hammer_lock              lock;
445         struct hammer_reserve           *resv;
446         struct hammer_inode             *ip;
447         struct hammer_inode             *target_ip;
448         struct hammer_btree_leaf_elm    leaf;
449         union hammer_data_ondisk        *data;
450         int                             flags;
451         hammer_off_t                    zone2_offset;   /* direct-write only */
452 };
453
454 typedef struct hammer_record *hammer_record_t;
455
456 /*
457  * Record flags.  Note that FE can only be set by the frontend if the
458  * record has not been interlocked by the backend w/ BE.
459  */
460 #define HAMMER_RECF_ALLOCDATA           0x0001
461 #define HAMMER_RECF_ONRBTREE            0x0002
462 #define HAMMER_RECF_DELETED_FE          0x0004  /* deleted (frontend) */
463 #define HAMMER_RECF_DELETED_BE          0x0008  /* deleted (backend) */
464 #define HAMMER_RECF_COMMITTED           0x0010  /* committed to the B-Tree */
465 #define HAMMER_RECF_INTERLOCK_BE        0x0020  /* backend interlock */
466 #define HAMMER_RECF_WANTED              0x0040  /* wanted by the frontend */
467 #define HAMMER_RECF_CONVERT_DELETE      0x0100  /* special case */
468 #define HAMMER_RECF_DIRECT_IO           0x0200  /* related direct I/O running*/
469 #define HAMMER_RECF_DIRECT_WAIT         0x0400  /* related direct I/O running*/
470 #define HAMMER_RECF_DIRECT_INVAL        0x0800  /* buffer alias invalidation */
471 #define HAMMER_RECF_REDO                0x1000  /* REDO was laid down */
472
473 /*
474  * hammer_create_at_cursor() and hammer_delete_at_cursor() flags.
475  */
476 #define HAMMER_CREATE_MODE_UMIRROR      0x0001
477 #define HAMMER_CREATE_MODE_SYS          0x0002
478
479 #define HAMMER_DELETE_ADJUST            0x0001
480 #define HAMMER_DELETE_DESTROY           0x0002
481
482 /*
483  * In-memory structures representing on-disk structures.
484  */
485 struct hammer_volume;
486 struct hammer_buffer;
487 struct hammer_node;
488 struct hammer_undo;
489 struct hammer_reserve;
490
491 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
492 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
493 RB_HEAD(hammer_nod_rb_tree, hammer_node);
494 RB_HEAD(hammer_und_rb_tree, hammer_undo);
495 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
496
497 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
498               hammer_vol_rb_compare, int32_t);
499 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
500               hammer_buf_rb_compare, hammer_off_t);
501 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
502               hammer_nod_rb_compare, hammer_off_t);
503 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
504               hammer_und_rb_compare, hammer_off_t);
505 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
506               hammer_res_rb_compare, hammer_off_t);
507
508 /*
509  * IO management - embedded at the head of various in-memory structures
510  *
511  * VOLUME       - hammer_volume containing meta-data
512  * META_BUFFER  - hammer_buffer containing meta-data
513  * DATA_BUFFER  - hammer_buffer containing pure-data
514  *
515  * Dirty volume headers and dirty meta-data buffers are locked until the
516  * flusher can sequence them out.  Dirty pure-data buffers can be written.
517  * Clean buffers can be passively released.
518  */
519 typedef enum hammer_io_type {
520         HAMMER_STRUCTURE_VOLUME,
521         HAMMER_STRUCTURE_META_BUFFER,
522         HAMMER_STRUCTURE_UNDO_BUFFER,
523         HAMMER_STRUCTURE_DATA_BUFFER
524 } hammer_io_type_t;
525
526 union hammer_io_structure;
527 struct hammer_io;
528
529 struct worklist {
530         LIST_ENTRY(worklist) node;
531 };
532
533 TAILQ_HEAD(hammer_io_list, hammer_io);
534 typedef struct hammer_io_list *hammer_io_list_t;
535
536 struct hammer_io {
537         struct worklist         worklist;
538         struct hammer_lock      lock;
539         enum hammer_io_type     type;
540         struct hammer_mount     *hmp;
541         struct hammer_volume    *volume;
542         TAILQ_ENTRY(hammer_io)  mod_entry; /* list entry if modified */
543         hammer_io_list_t        mod_list;
544         struct buf              *bp;
545         int64_t                 offset;    /* zone-2 offset */
546         int                     bytes;     /* buffer cache buffer size */
547         int                     loading;   /* loading/unloading interlock */
548         int                     modify_refs;
549
550         u_int           modified : 1;   /* bp's data was modified */
551         u_int           released : 1;   /* bp released (w/ B_LOCKED set) */
552         u_int           running : 1;    /* bp write IO in progress */
553         u_int           waiting : 1;    /* someone is waiting on us */
554         u_int           validated : 1;  /* ondisk has been validated */
555         u_int           waitdep : 1;    /* flush waits for dependancies */
556         u_int           recovered : 1;  /* has recovery ref */
557         u_int           waitmod : 1;    /* waiting for modify_refs */
558         u_int           reclaim : 1;    /* reclaim requested */
559         u_int           gencrc : 1;     /* crc needs to be generated */
560         u_int           ioerror : 1;    /* abort on io-error */
561 };
562
563 typedef struct hammer_io *hammer_io_t;
564
565 #define HAMMER_CLUSTER_SIZE     (64 * 1024)
566 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
567 #undef  HAMMER_CLUSTER_SIZE
568 #define HAMMER_CLUSTER_SIZE     MAXBSIZE
569 #endif
570 #define HAMMER_CLUSTER_BUFS     (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
571
572 /*
573  * In-memory volume representing on-disk buffer
574  */
575 struct hammer_volume {
576         struct hammer_io io;
577         RB_ENTRY(hammer_volume) rb_node;
578         struct hammer_volume_ondisk *ondisk;
579         int32_t vol_no;
580         int64_t nblocks;        /* note: special calculation for statfs */
581         int64_t buffer_base;    /* base offset of buffer 0 */
582         hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
583         hammer_off_t maxraw_off; /* Maximum raw offset for device */
584         char    *vol_name;
585         struct vnode *devvp;
586         int     vol_flags;
587 };
588
589 typedef struct hammer_volume *hammer_volume_t;
590
591 /*
592  * In-memory buffer (other then volume, super-cluster, or cluster),
593  * representing an on-disk buffer.
594  */
595 struct hammer_buffer {
596         struct hammer_io io;
597         RB_ENTRY(hammer_buffer) rb_node;
598         void *ondisk;
599         hammer_off_t zoneX_offset;
600         hammer_off_t zone2_offset;
601         struct hammer_reserve *resv;
602         struct hammer_node_list clist;
603 };
604
605 typedef struct hammer_buffer *hammer_buffer_t;
606
607 /*
608  * In-memory B-Tree node, representing an on-disk B-Tree node.
609  *
610  * This is a hang-on structure which is backed by a hammer_buffer,
611  * indexed by a hammer_cluster, and used for fine-grained locking of
612  * B-Tree nodes in order to properly control lock ordering.  A hammer_buffer
613  * can contain multiple nodes representing wildly disassociated portions
614  * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
615  *
616  * This structure uses a cluster-relative index to reduce the number
617  * of layers required to access it, and also because all on-disk B-Tree
618  * references are cluster-relative offsets.
619  */
620 struct hammer_node {
621         struct hammer_lock      lock;           /* node-by-node lock */
622         TAILQ_ENTRY(hammer_node) entry;         /* per-buffer linkage */
623         RB_ENTRY(hammer_node)   rb_node;        /* per-cluster linkage */
624         hammer_off_t            node_offset;    /* full offset spec */
625         struct hammer_mount     *hmp;
626         struct hammer_buffer    *buffer;        /* backing buffer */
627         hammer_node_ondisk_t    ondisk;         /* ptr to on-disk structure */
628         TAILQ_HEAD(, hammer_cursor) cursor_list;  /* deadlock recovery */
629         struct hammer_node_cache_list cache_list; /* passive caches */
630         int                     flags;
631         int                     loading;        /* load interlock */
632 };
633
634 #define HAMMER_NODE_DELETED     0x0001
635 #define HAMMER_NODE_FLUSH       0x0002
636 #define HAMMER_NODE_CRCGOOD     0x0004
637 #define HAMMER_NODE_NEEDSCRC    0x0008
638 #define HAMMER_NODE_NEEDSMIRROR 0x0010
639 #define HAMMER_NODE_CRCBAD      0x0020
640
641 #define HAMMER_NODE_CRCANY      (HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD)
642
643 typedef struct hammer_node      *hammer_node_t;
644
645 /*
646  * List of locked nodes.  This structure is used to lock potentially large
647  * numbers of nodes as an aid for complex B-Tree operations.
648  */
649 struct hammer_node_lock;
650 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock);
651
652 struct hammer_node_lock {
653         TAILQ_ENTRY(hammer_node_lock) entry;
654         struct hammer_node_lock_list  list;
655         struct hammer_node_lock       *parent;
656         hammer_node_t   node;
657         hammer_node_ondisk_t copy;      /* copy of on-disk data */
658         int             index;          /* index of this node in parent */
659         int             count;          /* count children */
660         int             flags;
661 };
662
663 typedef struct hammer_node_lock *hammer_node_lock_t;
664
665 #define HAMMER_NODE_LOCK_UPDATED        0x0001
666
667 /*
668  * Common I/O management structure - embedded in in-memory structures
669  * which are backed by filesystem buffers.
670  */
671 union hammer_io_structure {
672         struct hammer_io        io;
673         struct hammer_volume    volume;
674         struct hammer_buffer    buffer;
675 };
676
677 typedef union hammer_io_structure *hammer_io_structure_t;
678
679 /*
680  * The reserve structure prevents the blockmap from allocating
681  * out of a reserved bigblock.  Such reservations are used by
682  * the direct-write mechanism.
683  *
684  * The structure is also used to hold off on reallocations of
685  * big blocks from the freemap until flush dependancies have
686  * been dealt with.
687  */
688 struct hammer_reserve {
689         RB_ENTRY(hammer_reserve) rb_node;
690         TAILQ_ENTRY(hammer_reserve) delay_entry;
691         int             flush_group;
692         int             flags;
693         int             refs;
694         int             zone;
695         int             append_off;
696         hammer_off_t    zone_offset;
697 };
698
699 typedef struct hammer_reserve *hammer_reserve_t;
700
701 #define HAMMER_RESF_ONDELAY     0x0001
702 #define HAMMER_RESF_LAYER2FREE  0x0002
703
704 #include "hammer_cursor.h"
705
706 /*
707  * The undo structure tracks recent undos to avoid laying down duplicate
708  * undos within a flush group, saving us a significant amount of overhead.
709  *
710  * This is strictly a heuristic.
711  */
712 #define HAMMER_MAX_UNDOS                1024
713 #define HAMMER_MAX_FLUSHERS             4
714
715 struct hammer_undo {
716         RB_ENTRY(hammer_undo)   rb_node;
717         TAILQ_ENTRY(hammer_undo) lru_entry;
718         hammer_off_t            offset;
719         int                     bytes;
720 };
721
722 typedef struct hammer_undo *hammer_undo_t;
723
724 struct hammer_flusher_info;
725 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
726
727 struct hammer_flusher {
728         int             signal;         /* flusher thread sequencer */
729         int             act;            /* currently active flush group */
730         int             done;           /* set to act when complete */
731         int             next;           /* next flush group */
732         int             group_lock;     /* lock sequencing of the next flush */
733         int             exiting;        /* request master exit */
734         thread_t        td;             /* master flusher thread */
735         hammer_tid_t    tid;            /* last flushed transaction id */
736         int             finalize_want;          /* serialize finalization */
737         struct hammer_lock finalize_lock;       /* serialize finalization */
738         struct hammer_transaction trans;        /* shared transaction */
739         struct hammer_flusher_info_list run_list;
740         struct hammer_flusher_info_list ready_list;
741 };
742
743 #define HAMMER_FLUSH_UNDOS_RELAXED      0
744 #define HAMMER_FLUSH_UNDOS_FORCED       1
745 #define HAMMER_FLUSH_UNDOS_AUTO         2
746 /*
747  * Internal hammer mount data structure
748  */
749 struct hammer_mount {
750         struct mount *mp;
751         /*struct vnode *rootvp;*/
752         struct hammer_ino_rb_tree rb_inos_root;
753         struct hammer_vol_rb_tree rb_vols_root;
754         struct hammer_nod_rb_tree rb_nods_root;
755         struct hammer_und_rb_tree rb_undo_root;
756         struct hammer_res_rb_tree rb_resv_root;
757         struct hammer_buf_rb_tree rb_bufs_root;
758         struct hammer_pfs_rb_tree rb_pfsm_root;
759         struct hammer_volume *rootvol;
760         struct hammer_base_elm root_btree_beg;
761         struct hammer_base_elm root_btree_end;
762
763         struct malloc_type      *m_misc;
764         struct malloc_type      *m_inodes;
765
766         int     flags;          /* HAMMER_MOUNT_xxx flags */
767         int     hflags;
768         int     ronly;
769         int     nvolumes;
770         int     volume_iterator;
771         int     master_id;      /* -1 or 0-15 - clustering and mirroring */
772         int     version;        /* hammer filesystem version to use */
773         int     rsv_inodes;     /* reserved space due to dirty inodes */
774         int64_t rsv_databytes;  /* reserved space due to record data */
775         int     rsv_recs;       /* reserved space due to dirty records */
776         int     rsv_fromdelay;  /* bigblocks reserved due to flush delay */
777         int     undo_rec_limit; /* based on size of undo area */
778         int     last_newrecords;
779         int     count_newrecords;
780
781         int     volume_to_remove; /* volume that is currently being removed */
782
783         int     inode_reclaims; /* inodes pending reclaim by flusher */
784         int     count_inodes;   /* total number of inodes */
785         int     count_iqueued;  /* inodes queued to flusher */
786
787         struct hammer_flusher flusher;
788
789         u_int   check_interrupt;
790         u_int   check_yield;
791         uuid_t  fsid;
792         struct hammer_io_list volu_list;        /* dirty undo buffers */
793         struct hammer_io_list undo_list;        /* dirty undo buffers */
794         struct hammer_io_list data_list;        /* dirty data buffers */
795         struct hammer_io_list alt_data_list;    /* dirty data buffers */
796         struct hammer_io_list meta_list;        /* dirty meta bufs    */
797         struct hammer_io_list lose_list;        /* loose buffers      */
798         int     locked_dirty_space;             /* meta/volu count    */
799         int     io_running_space;
800         int     objid_cache_count;
801         int     error;                          /* critical I/O error */
802         struct krate    krate;                  /* rate limited kprintf */
803         hammer_tid_t    asof;                   /* snapshot mount */
804         hammer_tid_t    next_tid;
805         hammer_tid_t    flush_tid1;             /* flusher tid sequencing */
806         hammer_tid_t    flush_tid2;             /* flusher tid sequencing */
807         int64_t copy_stat_freebigblocks;        /* number of free bigblocks */
808         u_int32_t       undo_seqno;             /* UNDO/REDO FIFO seqno */
809
810         struct netexport export;
811         struct hammer_lock sync_lock;
812         struct hammer_lock free_lock;
813         struct hammer_lock undo_lock;
814         struct hammer_lock blkmap_lock;
815         struct hammer_lock snapshot_lock;
816         struct hammer_lock volume_lock;
817         struct hammer_blockmap  blockmap[HAMMER_MAX_ZONES];
818         struct hammer_undo      undos[HAMMER_MAX_UNDOS];
819         int                     undo_alloc;
820         TAILQ_HEAD(, hammer_undo)  undo_lru_list;
821         TAILQ_HEAD(, hammer_reserve) delay_list;
822         struct hammer_flush_group_list  flush_group_list;
823         hammer_flush_group_t    next_flush_group;
824         TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
825         TAILQ_HEAD(, hammer_reclaim) reclaim_list;
826 };
827
828 typedef struct hammer_mount     *hammer_mount_t;
829
830 #define HAMMER_MOUNT_CRITICAL_ERROR     0x0001
831 #define HAMMER_MOUNT_FLUSH_RECOVERY     0x0002
832 #define HAMMER_MOUNT_REDO_SYNC          0x0004
833
834 struct hammer_sync_info {
835         int error;
836         int waitfor;
837 };
838
839 #endif
840
841 /*
842  * checkspace slop (8MB chunks), higher numbers are more conservative.
843  */
844 #define HAMMER_CHKSPC_REBLOCK   25
845 #define HAMMER_CHKSPC_MIRROR    20
846 #define HAMMER_CHKSPC_WRITE     20
847 #define HAMMER_CHKSPC_CREATE    20
848 #define HAMMER_CHKSPC_REMOVE    10
849 #define HAMMER_CHKSPC_EMERGENCY 0
850
851 #if defined(_KERNEL)
852
853 extern struct vop_ops hammer_vnode_vops;
854 extern struct vop_ops hammer_spec_vops;
855 extern struct vop_ops hammer_fifo_vops;
856 extern struct bio_ops hammer_bioops;
857
858 extern int hammer_debug_io;
859 extern int hammer_debug_general;
860 extern int hammer_debug_debug;
861 extern int hammer_debug_inode;
862 extern int hammer_debug_locks;
863 extern int hammer_debug_btree;
864 extern int hammer_debug_tid;
865 extern int hammer_debug_recover;
866 extern int hammer_debug_recover_faults;
867 extern int hammer_debug_critical;
868 extern int hammer_cluster_enable;
869 extern int hammer_count_fsyncs;
870 extern int hammer_count_inodes;
871 extern int hammer_count_iqueued;
872 extern int hammer_count_reclaiming;
873 extern int hammer_count_records;
874 extern int hammer_count_record_datas;
875 extern int hammer_count_volumes;
876 extern int hammer_count_buffers;
877 extern int hammer_count_nodes;
878 extern int64_t hammer_count_extra_space_used;
879 extern int64_t hammer_stats_btree_lookups;
880 extern int64_t hammer_stats_btree_searches;
881 extern int64_t hammer_stats_btree_inserts;
882 extern int64_t hammer_stats_btree_deletes;
883 extern int64_t hammer_stats_btree_elements;
884 extern int64_t hammer_stats_btree_splits;
885 extern int64_t hammer_stats_btree_iterations;
886 extern int64_t hammer_stats_btree_root_iterations;
887 extern int64_t hammer_stats_record_iterations;
888 extern int64_t hammer_stats_file_read;
889 extern int64_t hammer_stats_file_write;
890 extern int64_t hammer_stats_file_iopsr;
891 extern int64_t hammer_stats_file_iopsw;
892 extern int64_t hammer_stats_disk_read;
893 extern int64_t hammer_stats_disk_write;
894 extern int64_t hammer_stats_inode_flushes;
895 extern int64_t hammer_stats_commits;
896 extern int64_t hammer_stats_undo;
897 extern int64_t hammer_stats_redo;
898 extern int hammer_count_dirtybufspace;
899 extern int hammer_count_refedbufs;
900 extern int hammer_count_reservations;
901 extern int hammer_count_io_running_read;
902 extern int hammer_count_io_running_write;
903 extern int hammer_count_io_locked;
904 extern int hammer_limit_dirtybufspace;
905 extern int hammer_limit_recs;
906 extern int hammer_limit_inode_recs;
907 extern int hammer_limit_reclaim;
908 extern int hammer_limit_redo;
909 extern int hammer_bio_count;
910 extern int hammer_verify_zone;
911 extern int hammer_verify_data;
912 extern int hammer_write_mode;
913 extern int hammer_yield_check;
914 extern int hammer_fsync_mode;
915 extern int hammer_autoflush;
916 extern int64_t hammer_contention_count;
917
918 void    hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
919                         int error, const char *msg);
920 int     hammer_vop_inactive(struct vop_inactive_args *);
921 int     hammer_vop_reclaim(struct vop_reclaim_args *);
922 int     hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
923 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
924                         hammer_inode_t dip, int64_t obj_id,
925                         hammer_tid_t asof, u_int32_t localization,
926                         int flags, int *errorp);
927 struct hammer_inode *hammer_get_dummy_inode(hammer_transaction_t trans,
928                         hammer_inode_t dip, int64_t obj_id,
929                         hammer_tid_t asof, u_int32_t localization,
930                         int flags, int *errorp);
931 struct hammer_inode *hammer_find_inode(hammer_transaction_t trans,
932                         int64_t obj_id, hammer_tid_t asof,
933                         u_int32_t localization);
934 void    hammer_scan_inode_snapshots(hammer_mount_t hmp,
935                         hammer_inode_info_t iinfo,
936                         int (*callback)(hammer_inode_t ip, void *data),
937                         void *data);
938 void    hammer_put_inode(struct hammer_inode *ip);
939 void    hammer_put_inode_ref(struct hammer_inode *ip);
940 void    hammer_inode_waitreclaims(hammer_mount_t hmp);
941 void    hammer_inode_waithard(hammer_mount_t hmp);
942
943 int     hammer_unload_volume(hammer_volume_t volume, void *data __unused);
944 int     hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
945
946 int     hammer_unload_buffer(hammer_buffer_t buffer, void *data);
947 int     hammer_install_volume(hammer_mount_t hmp, const char *volname,
948                         struct vnode *devvp);
949 int     hammer_mountcheck_volumes(hammer_mount_t hmp);
950
951 int     hammer_mem_add(hammer_record_t record);
952 int     hammer_ip_lookup(hammer_cursor_t cursor);
953 int     hammer_ip_first(hammer_cursor_t cursor);
954 int     hammer_ip_next(hammer_cursor_t cursor);
955 int     hammer_ip_resolve_data(hammer_cursor_t cursor);
956 int     hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
957                         hammer_tid_t tid);
958 int     hammer_create_at_cursor(hammer_cursor_t cursor,
959                         hammer_btree_leaf_elm_t leaf, void *udata, int mode);
960 int     hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
961                         hammer_tid_t delete_tid, u_int32_t delete_ts,
962                         int track, int64_t *stat_bytes);
963 int     hammer_ip_check_directory_empty(hammer_transaction_t trans,
964                         hammer_inode_t ip);
965 int     hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
966 int     hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
967
968 hammer_record_t
969         hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
970 void    hammer_flush_record_done(hammer_record_t record, int error);
971 void    hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
972 void    hammer_rel_mem_record(hammer_record_t record);
973
974 int     hammer_cursor_up(hammer_cursor_t cursor);
975 int     hammer_cursor_up_locked(hammer_cursor_t cursor);
976 int     hammer_cursor_down(hammer_cursor_t cursor);
977 int     hammer_cursor_upgrade(hammer_cursor_t cursor);
978 int     hammer_cursor_upgrade_node(hammer_cursor_t cursor);
979 void    hammer_cursor_downgrade(hammer_cursor_t cursor);
980 int     hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
981                         int index);
982 void    hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
983 int     hammer_lock_ex_try(struct hammer_lock *lock);
984 void    hammer_lock_sh(struct hammer_lock *lock);
985 int     hammer_lock_sh_try(struct hammer_lock *lock);
986 int     hammer_lock_upgrade(struct hammer_lock *lock);
987 void    hammer_lock_downgrade(struct hammer_lock *lock);
988 int     hammer_lock_status(struct hammer_lock *lock);
989 void    hammer_unlock(struct hammer_lock *lock);
990 void    hammer_ref(struct hammer_lock *lock);
991 void    hammer_unref(struct hammer_lock *lock);
992
993 void    hammer_sync_lock_ex(hammer_transaction_t trans);
994 void    hammer_sync_lock_sh(hammer_transaction_t trans);
995 int     hammer_sync_lock_sh_try(hammer_transaction_t trans);
996 void    hammer_sync_unlock(hammer_transaction_t trans);
997
998 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
999 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
1000 void    hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
1001 u_int64_t hammer_timespec_to_time(struct timespec *ts);
1002 int     hammer_str_to_tid(const char *str, int *ispfsp,
1003                         hammer_tid_t *tidp, u_int32_t *localizationp);
1004 int     hammer_is_atatext(const char *name, int len);
1005 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip,
1006                         int64_t namekey);
1007 void hammer_clear_objid(hammer_inode_t dip);
1008 void hammer_destroy_objid_cache(hammer_mount_t hmp);
1009
1010 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
1011                         int bytes);
1012 void hammer_clear_undo_history(hammer_mount_t hmp);
1013 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
1014 int hammer_get_dtype(u_int8_t obj_type);
1015 u_int8_t hammer_get_obj_type(enum vtype vtype);
1016 int64_t hammer_directory_namekey(hammer_inode_t dip, const void *name, int len,
1017                         u_int32_t *max_iterationsp);
1018 int     hammer_nohistory(hammer_inode_t ip);
1019
1020 int     hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
1021                         hammer_node_cache_t cache, hammer_inode_t ip);
1022 void    hammer_normalize_cursor(hammer_cursor_t cursor);
1023 void    hammer_done_cursor(hammer_cursor_t cursor);
1024 int     hammer_recover_cursor(hammer_cursor_t cursor);
1025 void    hammer_unlock_cursor(hammer_cursor_t cursor);
1026 int     hammer_lock_cursor(hammer_cursor_t cursor);
1027 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor);
1028 void    hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
1029
1030 void    hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
1031 void    hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
1032                         int index);
1033 void    hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
1034                         int index);
1035 void    hammer_cursor_moved_element(hammer_node_t onode, hammer_node_t nnode,
1036                         int oindex, int nindex);
1037 void    hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent,
1038                         hammer_node_t nparent, int nindex);
1039 void    hammer_cursor_inserted_element(hammer_node_t node, int index);
1040 void    hammer_cursor_deleted_element(hammer_node_t node, int index);
1041
1042 int     hammer_btree_lookup(hammer_cursor_t cursor);
1043 int     hammer_btree_first(hammer_cursor_t cursor);
1044 int     hammer_btree_last(hammer_cursor_t cursor);
1045 int     hammer_btree_extract(hammer_cursor_t cursor, int flags);
1046 int     hammer_btree_iterate(hammer_cursor_t cursor);
1047 int     hammer_btree_iterate_reverse(hammer_cursor_t cursor);
1048 int     hammer_btree_insert(hammer_cursor_t cursor,
1049                             hammer_btree_leaf_elm_t elm, int *doprop);
1050 int     hammer_btree_delete(hammer_cursor_t cursor);
1051 void    hammer_btree_do_propagation(hammer_cursor_t cursor,
1052                             hammer_pseudofs_inmem_t pfsm,
1053                             hammer_btree_leaf_elm_t leaf);
1054 int     hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
1055 int     hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
1056 int     hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
1057 int     hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
1058
1059 int     btree_set_parent(hammer_transaction_t trans, hammer_node_t node,
1060                         hammer_btree_elm_t elm);
1061 void    hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node);
1062 int     hammer_btree_lock_children(hammer_cursor_t cursor, int depth,
1063                         hammer_node_lock_t parent);
1064 void    hammer_btree_lock_copy(hammer_cursor_t cursor,
1065                         hammer_node_lock_t parent);
1066 int     hammer_btree_sync_copy(hammer_cursor_t cursor,
1067                         hammer_node_lock_t parent);
1068 void    hammer_btree_unlock_children(hammer_cursor_t cursor,
1069                         hammer_node_lock_t parent);
1070 int     hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
1071 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans,
1072                         hammer_node_t node, int *parent_indexp,
1073                         int *errorp, int try_exclusive);
1074
1075 void    hammer_print_btree_node(hammer_node_ondisk_t ondisk);
1076 void    hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
1077
1078 void    *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
1079                         int *errorp, struct hammer_buffer **bufferp);
1080 void    *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
1081                         int *errorp, struct hammer_buffer **bufferp);
1082 void    *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1083                         int *errorp, struct hammer_buffer **bufferp);
1084 void    *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1085                         int *errorp, struct hammer_buffer **bufferp);
1086
1087 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
1088
1089 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
1090                         int32_t vol_no, int *errorp);
1091 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
1092                         int bytes, int isnew, int *errorp);
1093 void            hammer_sync_buffers(hammer_mount_t hmp,
1094                         hammer_off_t base_offset, int bytes);
1095 int             hammer_del_buffers(hammer_mount_t hmp,
1096                         hammer_off_t base_offset,
1097                         hammer_off_t zone2_offset, int bytes,
1098                         int report_conflicts);
1099
1100 int             hammer_ref_volume(hammer_volume_t volume);
1101 int             hammer_ref_buffer(hammer_buffer_t buffer);
1102 void            hammer_flush_buffer_nodes(hammer_buffer_t buffer);
1103
1104 void            hammer_rel_volume(hammer_volume_t volume, int flush);
1105 void            hammer_rel_buffer(hammer_buffer_t buffer, int flush);
1106
1107 int             hammer_vfs_export(struct mount *mp, int op,
1108                         const struct export_args *export);
1109 hammer_node_t   hammer_get_node(hammer_transaction_t trans,
1110                         hammer_off_t node_offset, int isnew, int *errorp);
1111 void            hammer_ref_node(hammer_node_t node);
1112 hammer_node_t   hammer_ref_node_safe(hammer_transaction_t trans,
1113                         hammer_node_cache_t cache, int *errorp);
1114 void            hammer_rel_node(hammer_node_t node);
1115 void            hammer_delete_node(hammer_transaction_t trans,
1116                         hammer_node_t node);
1117 void            hammer_cache_node(hammer_node_cache_t cache,
1118                         hammer_node_t node);
1119 void            hammer_uncache_node(hammer_node_cache_t cache);
1120 void            hammer_flush_node(hammer_node_t node);
1121
1122 void hammer_dup_buffer(struct hammer_buffer **bufferp,
1123                         struct hammer_buffer *buffer);
1124 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans,
1125                         hammer_off_t hint, int *errorp);
1126 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1127                         u_int16_t rec_type, hammer_off_t *data_offsetp,
1128                         struct hammer_buffer **data_bufferp,
1129                         hammer_off_t hint, int *errorp);
1130
1131 int hammer_generate_undo(hammer_transaction_t trans,
1132                         hammer_off_t zone1_offset, void *base, int len);
1133 int hammer_generate_redo(hammer_transaction_t trans, hammer_inode_t ip,
1134                         hammer_off_t file_offset, u_int32_t flags,
1135                         void *base, int len);
1136 void hammer_generate_redo_sync(hammer_transaction_t trans);
1137 void hammer_format_undo(void *base, u_int32_t seqno);
1138 int hammer_upgrade_undo_4(hammer_transaction_t trans);
1139
1140 void hammer_put_volume(struct hammer_volume *volume, int flush);
1141 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
1142
1143 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1144                         hammer_off_t owner, int *errorp);
1145 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1146                         hammer_off_t owner, int *errorp);
1147 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp);
1148 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1149                         int bytes, hammer_off_t hint, int *errorp);
1150 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1151                         int bytes, hammer_off_t *zone_offp, int *errorp);
1152 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1153                         hammer_reserve_t resv);
1154 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1155 void hammer_blockmap_free(hammer_transaction_t trans,
1156                         hammer_off_t bmap_off, int bytes);
1157 int hammer_blockmap_finalize(hammer_transaction_t trans,
1158                         hammer_reserve_t resv,
1159                         hammer_off_t bmap_off, int bytes);
1160 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
1161                         int *curp, int *errorp);
1162 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1163                         int *errorp);
1164 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1165                         int *errorp);
1166 int64_t hammer_undo_used(hammer_transaction_t trans);
1167 int64_t hammer_undo_space(hammer_transaction_t trans);
1168 int64_t hammer_undo_max(hammer_mount_t hmp);
1169 int hammer_undo_reclaim(hammer_io_t io);
1170
1171 void hammer_start_transaction(struct hammer_transaction *trans,
1172                               struct hammer_mount *hmp);
1173 void hammer_simple_transaction(struct hammer_transaction *trans,
1174                               struct hammer_mount *hmp);
1175 void hammer_start_transaction_fls(struct hammer_transaction *trans,
1176                                   struct hammer_mount *hmp);
1177 void hammer_done_transaction(struct hammer_transaction *trans);
1178 hammer_tid_t hammer_alloc_tid(hammer_mount_t hmp, int count);
1179
1180 void hammer_modify_inode(hammer_inode_t ip, int flags);
1181 void hammer_flush_inode(hammer_inode_t ip, int flags);
1182 void hammer_flush_inode_done(hammer_inode_t ip, int error);
1183 void hammer_wait_inode(hammer_inode_t ip);
1184
1185 int  hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
1186                         struct ucred *cred, struct hammer_inode *dip,
1187                         const char *name, int namelen,
1188                         hammer_pseudofs_inmem_t pfsm,
1189                         struct hammer_inode **ipp);
1190 void hammer_rel_inode(hammer_inode_t ip, int flush);
1191 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1192 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1193 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1194
1195 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1196 void hammer_test_inode(hammer_inode_t dip);
1197 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1198
1199 int  hammer_ip_add_directory(struct hammer_transaction *trans,
1200                         hammer_inode_t dip, const char *name, int bytes,
1201                         hammer_inode_t nip);
1202 int  hammer_ip_del_directory(struct hammer_transaction *trans,
1203                         hammer_cursor_t cursor, hammer_inode_t dip,
1204                         hammer_inode_t ip);
1205 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1206                         void *data, int bytes, int *errorp);
1207 int  hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
1208 int  hammer_ip_add_record(struct hammer_transaction *trans,
1209                         hammer_record_t record);
1210 int  hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1211                         int64_t ran_beg, int64_t ran_end, int truncating);
1212 int  hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1213                         int *countp);
1214 int  hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1215                         int64_t offset, void *data, int bytes);
1216 int  hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
1217 int  hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1218 hammer_pseudofs_inmem_t  hammer_load_pseudofs(hammer_transaction_t trans,
1219                         u_int32_t localization, int *errorp);
1220 int  hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1221                         hammer_pseudofs_inmem_t pfsm);
1222 int  hammer_save_pseudofs(hammer_transaction_t trans,
1223                         hammer_pseudofs_inmem_t pfsm);
1224 int  hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization);
1225 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1226 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1227                         struct ucred *cred);
1228
1229 void hammer_io_init(hammer_io_t io, hammer_volume_t volume,
1230                         enum hammer_io_type type);
1231 int hammer_io_read(struct vnode *devvp, struct hammer_io *io,
1232                         hammer_off_t limit);
1233 void hammer_io_advance(struct hammer_io *io);
1234 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
1235 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1236 struct buf *hammer_io_release(struct hammer_io *io, int flush);
1237 void hammer_io_flush(struct hammer_io *io, int reclaim);
1238 void hammer_io_wait(struct hammer_io *io);
1239 void hammer_io_waitdep(struct hammer_io *io);
1240 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident);
1241 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1242                         hammer_btree_leaf_elm_t leaf);
1243 int hammer_io_direct_write(hammer_mount_t hmp, hammer_record_t record,
1244                         struct bio *bio);
1245 void hammer_io_direct_wait(hammer_record_t record);
1246 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1247 void hammer_io_write_interlock(hammer_io_t io);
1248 void hammer_io_done_interlock(hammer_io_t io);
1249 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1250 void hammer_io_clear_modlist(struct hammer_io *io);
1251 void hammer_io_flush_sync(hammer_mount_t hmp);
1252 void hammer_io_clear_error(struct hammer_io *io);
1253
1254 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1255                         void *base, int len);
1256 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1257                         void *base, int len);
1258 void hammer_modify_volume_done(hammer_volume_t volume);
1259 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1260
1261 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1262                         struct hammer_ioc_reblock *reblock);
1263 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip,
1264                         struct hammer_ioc_rebalance *rebal);
1265 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1266                         struct hammer_ioc_prune *prune);
1267 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1268                         struct hammer_ioc_mirror_rw *mirror);
1269 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1270                         struct hammer_ioc_mirror_rw *mirror);
1271 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1272                         struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1273 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1274                         struct hammer_ioc_pseudofs_rw *pfs);
1275 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1276                         struct hammer_ioc_pseudofs_rw *pfs);
1277 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1278                         struct hammer_ioc_pseudofs_rw *pfs);
1279 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1280                         struct hammer_ioc_pseudofs_rw *pfs);
1281 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1282                         struct hammer_ioc_pseudofs_rw *pfs);
1283 int hammer_ioc_volume_add(hammer_transaction_t trans, hammer_inode_t ip,
1284                         struct hammer_ioc_volume *ioc);
1285 int hammer_ioc_volume_del(hammer_transaction_t trans, hammer_inode_t ip,
1286                         struct hammer_ioc_volume *ioc);
1287
1288 int hammer_signal_check(hammer_mount_t hmp);
1289
1290 void hammer_flusher_create(hammer_mount_t hmp);
1291 void hammer_flusher_destroy(hammer_mount_t hmp);
1292 void hammer_flusher_sync(hammer_mount_t hmp);
1293 int  hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1294 int  hammer_flusher_async_one(hammer_mount_t hmp);
1295 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1296 void hammer_flusher_wait_next(hammer_mount_t hmp);
1297 int  hammer_flusher_meta_limit(hammer_mount_t hmp);
1298 int  hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1299 int  hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1300 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1301 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1302 int  hammer_flusher_haswork(hammer_mount_t hmp);
1303 void hammer_flusher_flush_undos(hammer_mount_t hmp, int already_flushed);
1304
1305 int hammer_recover_stage1(hammer_mount_t hmp, hammer_volume_t rootvol);
1306 int hammer_recover_stage2(hammer_mount_t hmp, hammer_volume_t rootvol);
1307 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1308                         hammer_volume_t root_volume, int final);
1309
1310 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1311 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1312 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1313
1314 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1315 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1316 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1317 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1318 void hkprintf(const char *ctl, ...);
1319 udev_t hammer_fsid_to_udev(uuid_t *uuid);
1320
1321
1322 int hammer_blocksize(int64_t file_offset);
1323 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1324
1325 /*
1326  * Shortcut for _hammer_checkspace(), used all over the code.
1327  */
1328 static __inline int
1329 hammer_checkspace(hammer_mount_t hmp, int slop)
1330 {
1331         return(_hammer_checkspace(hmp, slop, NULL));
1332 }
1333
1334 #endif
1335
1336 static __inline void
1337 hammer_wait_mem_record(hammer_record_t record)
1338 {
1339         hammer_wait_mem_record_ident(record, "hmmwai");
1340 }
1341
1342 static __inline void
1343 hammer_lock_ex(struct hammer_lock *lock)
1344 {
1345         hammer_lock_ex_ident(lock, "hmrlck");
1346 }
1347
1348 /*
1349  * Indicate that a B-Tree node is being modified.
1350  */
1351 static __inline void
1352 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1353 {
1354         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1355         hammer_modify_buffer(trans, node->buffer, NULL, 0);
1356 }
1357
1358 static __inline void
1359 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1360 {
1361         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1362         hammer_modify_buffer(trans, node->buffer,
1363                              node->ondisk, sizeof(*node->ondisk));
1364 }
1365
1366 static __inline void
1367 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1368                    void *base, int len)
1369 {
1370         hammer_crc_t *crcptr;
1371
1372         KKASSERT((char *)base >= (char *)node->ondisk &&
1373                  (char *)base + len <=
1374                     (char *)node->ondisk + sizeof(*node->ondisk));
1375         KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1376         hammer_modify_buffer(trans, node->buffer, base, len);
1377         crcptr = &node->ondisk->crc;
1378         hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t));
1379         --node->buffer->io.modify_refs; /* only want one ref */
1380 }
1381
1382 /*
1383  * Indicate that the specified modifications have been completed.
1384  *
1385  * Do not try to generate the crc here, it's very expensive to do and a
1386  * sequence of insertions or deletions can result in many calls to this
1387  * function on the same node.
1388  */
1389 static __inline void
1390 hammer_modify_node_done(hammer_node_t node)
1391 {
1392         node->flags |= HAMMER_NODE_CRCGOOD;
1393         if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1394                 node->flags |= HAMMER_NODE_NEEDSCRC;
1395                 node->buffer->io.gencrc = 1;
1396                 hammer_ref_node(node);
1397         }
1398         hammer_modify_buffer_done(node->buffer);
1399 }
1400
1401 #define hammer_modify_volume_field(trans, vol, field)           \
1402         hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1403                              sizeof((vol)->ondisk->field))
1404
1405 #define hammer_modify_node_field(trans, node, field)            \
1406         hammer_modify_node(trans, node, &(node)->ondisk->field, \
1407                              sizeof((node)->ondisk->field))
1408
1409 /*
1410  * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1411  * created directories for HAMMER version 2 or greater and causes
1412  * directory entries to be placed the inode localization zone in
1413  * the B-Tree instead of the misc zone.
1414  *
1415  * This greatly improves localization between directory entries and
1416  * inodes
1417  */
1418 static __inline u_int32_t
1419 hammer_dir_localization(hammer_inode_t dip)
1420 {
1421         if (dip->ino_data.cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO)
1422                 return(HAMMER_LOCALIZE_INODE);
1423         else
1424                 return(HAMMER_LOCALIZE_MISC);
1425 }