sbin/hammer: Remove alloc_data_element()
[dragonfly.git] / sbin / hammer / ondisk.c
1 /*
2  * Copyright (c) 2007 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
35 #include <sys/diskslice.h>
36 #include <sys/diskmbr.h>
37
38 #include "hammer_util.h"
39
40 static void check_volume(struct volume_info *vol);
41 static void get_buffer_readahead(struct buffer_info *base);
42 static __inline int readhammervol(struct volume_info *vol);
43 static __inline int readhammerbuf(struct buffer_info *buf);
44 static __inline int writehammervol(struct volume_info *vol);
45 static __inline int writehammerbuf(struct buffer_info *buf);
46
47 uuid_t Hammer_FSType;
48 uuid_t Hammer_FSId;
49 int UseReadBehind = -4;
50 int UseReadAhead = 4;
51 int DebugOpt;
52
53 TAILQ_HEAD(volume_list, volume_info);
54 static struct volume_list VolList = TAILQ_HEAD_INITIALIZER(VolList);
55 static int valid_hammer_volumes;
56
57 static __inline
58 int
59 buffer_hash(hammer_off_t buf_offset)
60 {
61         int hi;
62
63         hi = (int)(buf_offset / HAMMER_BUFSIZE) & HAMMER_BUFLISTMASK;
64         return(hi);
65 }
66
67 static struct buffer_info*
68 find_buffer(struct volume_info *volume, hammer_off_t buf_offset)
69 {
70         int hi;
71         struct buffer_info *buf;
72
73         hi = buffer_hash(buf_offset);
74         TAILQ_FOREACH(buf, &volume->buffer_lists[hi], entry)
75                 if (buf->buf_offset == buf_offset)
76                         return(buf);
77         return(NULL);
78 }
79
80 static
81 struct volume_info *
82 __alloc_volume(const char *volname, int oflags)
83 {
84         struct volume_info *vol;
85         int i;
86
87         vol = malloc(sizeof(*vol));
88         if (vol == NULL)
89                 err(1, "alloc_volume");
90         bzero(vol, sizeof(*vol));
91
92         vol->vol_no = -1;
93         vol->rdonly = (oflags == O_RDONLY);
94         vol->name = strdup(volname);
95         vol->fd = open(vol->name, oflags);
96         if (vol->fd < 0)
97                 err(1, "alloc_volume: Failed to open %s", vol->name);
98         check_volume(vol);
99
100         vol->ondisk = malloc(HAMMER_BUFSIZE);
101         if (vol->ondisk == NULL)
102                 err(1, "alloc_volume");
103         bzero(vol->ondisk, HAMMER_BUFSIZE);
104
105         for (i = 0; i < HAMMER_BUFLISTS; ++i)
106                 TAILQ_INIT(&vol->buffer_lists[i]);
107
108         return(vol);
109 }
110
111 static void
112 __add_volume(struct volume_info *vol)
113 {
114         struct volume_info *scan;
115         struct stat st1, st2;
116
117         if (fstat(vol->fd, &st1) != 0)
118                 errx(1, "add_volume: %s: Failed to stat", vol->name);
119
120         TAILQ_FOREACH(scan, &VolList, entry) {
121                 if (scan->vol_no == vol->vol_no) {
122                         errx(1, "add_volume: %s: Duplicate volume number %d "
123                                 "against %s",
124                                 vol->name, vol->vol_no, scan->name);
125                 }
126                 if (fstat(scan->fd, &st2) != 0) {
127                         errx(1, "add_volume: %s: Failed to stat %s",
128                                 vol->name, scan->name);
129                 }
130                 if ((st1.st_ino == st2.st_ino) && (st1.st_dev == st2.st_dev)) {
131                         errx(1, "add_volume: %s: Specified more than once",
132                                 vol->name);
133                 }
134         }
135
136         TAILQ_INSERT_TAIL(&VolList, vol, entry);
137 }
138
139 /*
140  * Initialize a volume structure and ondisk vol_no field.
141  */
142 struct volume_info *
143 init_volume(int32_t vol_no, const char *filename, int oflags)
144 {
145         struct volume_info *vol;
146
147         vol = __alloc_volume(filename, oflags);
148         vol->vol_no = vol->ondisk->vol_no = vol_no;
149
150         __add_volume(vol);
151
152         return(vol);
153 }
154
155 /*
156  * Initialize a volume structure and read ondisk volume header.
157  */
158 struct volume_info*
159 load_volume(const char *filename, int oflags)
160 {
161         struct volume_info *vol;
162         hammer_volume_ondisk_t ondisk;
163         int n;
164
165         vol = __alloc_volume(filename, oflags);
166
167         n = readhammervol(vol);
168         if (n == -1) {
169                 err(1, "load_volume: %s: Read failed at offset 0", vol->name);
170         }
171         ondisk = vol->ondisk;
172         vol->vol_no = ondisk->vol_no;
173
174         if (ondisk->vol_rootvol != HAMMER_ROOT_VOLNO) {
175                 errx(1, "load_volume: Invalid root volume# %d",
176                         ondisk->vol_rootvol);
177         }
178
179         if (bcmp(&Hammer_FSType, &ondisk->vol_fstype, sizeof(Hammer_FSType))) {
180                 errx(1, "load_volume: %s: Header does not indicate "
181                         "that this is a hammer volume", vol->name);
182         }
183
184         if (valid_hammer_volumes++ == 0) {
185                 Hammer_FSId = ondisk->vol_fsid;
186         } else if (bcmp(&Hammer_FSId, &ondisk->vol_fsid, sizeof(Hammer_FSId))) {
187                 errx(1, "load_volume: %s: FSId does match other volumes!",
188                         vol->name);
189         }
190
191         __add_volume(vol);
192
193         return(vol);
194 }
195
196 /*
197  * Check basic volume characteristics.
198  */
199 static void
200 check_volume(struct volume_info *vol)
201 {
202         struct partinfo pinfo;
203         struct stat st;
204
205         /*
206          * Get basic information about the volume
207          */
208         if (ioctl(vol->fd, DIOCGPART, &pinfo) < 0) {
209                 /*
210                  * Allow the formatting of regular files as HAMMER volumes
211                  */
212                 if (fstat(vol->fd, &st) < 0)
213                         err(1, "Unable to stat %s", vol->name);
214                 vol->size = st.st_size;
215                 vol->type = "REGFILE";
216         } else {
217                 /*
218                  * When formatting a block device as a HAMMER volume the
219                  * sector size must be compatible.  HAMMER uses 16384 byte
220                  * filesystem buffers.
221                  */
222                 if (pinfo.reserved_blocks) {
223                         errx(1, "HAMMER cannot be placed in a partition "
224                                 "which overlaps the disklabel or MBR");
225                 }
226                 if (pinfo.media_blksize > HAMMER_BUFSIZE ||
227                     HAMMER_BUFSIZE % pinfo.media_blksize) {
228                         errx(1, "A media sector size of %d is not supported",
229                              pinfo.media_blksize);
230                 }
231
232                 vol->size = pinfo.media_size;
233                 vol->device_offset = pinfo.media_offset;
234                 vol->type = "DEVICE";
235         }
236 }
237
238 void
239 assert_volume_offset(struct volume_info *vol)
240 {
241         assert(hammer_is_zone_raw_buffer(vol->vol_free_off));
242         assert(hammer_is_zone_raw_buffer(vol->vol_free_end));
243 }
244
245 struct volume_info *
246 get_volume(int32_t vol_no)
247 {
248         struct volume_info *vol;
249
250         TAILQ_FOREACH(vol, &VolList, entry) {
251                 if (vol->vol_no == vol_no)
252                         break;
253         }
254
255         return(vol);
256 }
257
258 struct volume_info *
259 get_root_volume(void)
260 {
261         struct volume_info *root_vol;
262
263         root_vol = get_volume(HAMMER_ROOT_VOLNO);
264         assert(root_vol != NULL);
265
266         return(root_vol);
267 }
268
269 /*
270  * Acquire the specified buffer.  isnew is -1 only when called
271  * via get_buffer_readahead() to prevent another readahead.
272  */
273 static struct buffer_info *
274 get_buffer(hammer_off_t buf_offset, int isnew)
275 {
276         struct buffer_info *buf;
277         struct volume_info *volume;
278         int vol_no;
279         int zone;
280         int hi;
281         int dora = 0;
282         int error = 0;
283
284         zone = HAMMER_ZONE_DECODE(buf_offset);
285         if (zone > HAMMER_ZONE_RAW_BUFFER_INDEX)
286                 buf_offset = blockmap_lookup(buf_offset, NULL, NULL, &error);
287         if (error || buf_offset == HAMMER_OFF_BAD)
288                 return(NULL);
289         assert(hammer_is_zone_raw_buffer(buf_offset));
290
291         vol_no = HAMMER_VOL_DECODE(buf_offset);
292         volume = get_volume(vol_no);
293         assert(volume != NULL);
294
295         buf_offset &= ~HAMMER_BUFMASK64;
296         buf = find_buffer(volume, buf_offset);
297
298         if (buf == NULL) {
299                 buf = malloc(sizeof(*buf));
300                 bzero(buf, sizeof(*buf));
301                 buf->buf_offset = buf_offset;
302                 buf->raw_offset = hammer_xlate_to_phys(volume->ondisk,
303                                                         buf_offset);
304                 buf->volume = volume;
305                 buf->ondisk = malloc(HAMMER_BUFSIZE);
306                 if (isnew <= 0) {
307                         if (readhammerbuf(buf) == -1) {
308                                 err(1, "get_buffer: %s:%016jx "
309                                     "Read failed at offset %016jx",
310                                     volume->name,
311                                     (intmax_t)buf->buf_offset,
312                                     (intmax_t)buf->raw_offset);
313                         }
314                 }
315
316                 hi = buffer_hash(buf_offset);
317                 TAILQ_INSERT_TAIL(&volume->buffer_lists[hi], buf, entry);
318                 hammer_cache_add(&buf->cache);
319                 dora = (isnew == 0);
320         } else {
321                 assert(buf->ondisk != NULL);
322                 assert(isnew != -1);
323                 hammer_cache_used(&buf->cache);
324         }
325
326         ++buf->cache.refs;
327         hammer_cache_flush();
328
329         if (isnew > 0) {
330                 assert(buf->cache.modified == 0);
331                 bzero(buf->ondisk, HAMMER_BUFSIZE);
332                 buf->cache.modified = 1;
333         }
334         if (dora)
335                 get_buffer_readahead(buf);
336         return(buf);
337 }
338
339 static void
340 get_buffer_readahead(struct buffer_info *base)
341 {
342         struct buffer_info *buf;
343         struct volume_info *vol;
344         hammer_off_t buf_offset;
345         int64_t raw_offset;
346         int ri = UseReadBehind;
347         int re = UseReadAhead;
348
349         raw_offset = base->raw_offset + ri * HAMMER_BUFSIZE;
350         vol = base->volume;
351
352         while (ri < re) {
353                 if (raw_offset >= vol->ondisk->vol_buf_end)
354                         break;
355                 if (raw_offset < vol->ondisk->vol_buf_beg || ri == 0) {
356                         ++ri;
357                         raw_offset += HAMMER_BUFSIZE;
358                         continue;
359                 }
360                 buf_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no,
361                         raw_offset - vol->ondisk->vol_buf_beg);
362                 buf = find_buffer(vol, buf_offset);
363                 if (buf == NULL) {
364                         buf = get_buffer(buf_offset, -1);
365                         rel_buffer(buf);
366                 }
367                 ++ri;
368                 raw_offset += HAMMER_BUFSIZE;
369         }
370 }
371
372 void
373 rel_buffer(struct buffer_info *buffer)
374 {
375         struct volume_info *volume;
376         int hi;
377
378         if (buffer == NULL)
379                 return;
380         assert(buffer->cache.refs > 0);
381         if (--buffer->cache.refs == 0) {
382                 if (buffer->cache.delete) {
383                         hi = buffer_hash(buffer->buf_offset);
384                         volume = buffer->volume;
385                         if (buffer->cache.modified)
386                                 flush_buffer(buffer);
387                         TAILQ_REMOVE(&volume->buffer_lists[hi], buffer, entry);
388                         hammer_cache_del(&buffer->cache);
389                         free(buffer->ondisk);
390                         free(buffer);
391                 }
392         }
393 }
394
395 /*
396  * Retrieve a pointer to a buffer data given a buffer offset.  The underlying
397  * bufferp is freed if isnew or the offset is out of range of the cached data.
398  * If bufferp is freed a referenced buffer is loaded into it.
399  */
400 void *
401 get_buffer_data(hammer_off_t buf_offset, struct buffer_info **bufferp,
402                 int isnew)
403 {
404         if (*bufferp != NULL) {
405                 if (isnew > 0 ||
406                     (((*bufferp)->buf_offset ^ buf_offset) & ~HAMMER_BUFMASK64)) {
407                         rel_buffer(*bufferp);
408                         *bufferp = NULL;
409                 }
410         }
411
412         if (*bufferp == NULL) {
413                 *bufferp = get_buffer(buf_offset, isnew);
414                 if (*bufferp == NULL)
415                         return(NULL);
416         }
417
418         return(((char *)(*bufferp)->ondisk) +
419                 ((int32_t)buf_offset & HAMMER_BUFMASK));
420 }
421
422 /*
423  * Allocate HAMMER elements - B-Tree nodes
424  */
425 hammer_node_ondisk_t
426 alloc_btree_node(hammer_off_t *offp, struct buffer_info **data_bufferp)
427 {
428         hammer_node_ondisk_t node;
429
430         node = alloc_blockmap(HAMMER_ZONE_BTREE_INDEX, sizeof(*node),
431                               offp, data_bufferp);
432         bzero(node, sizeof(*node));
433         return(node);
434 }
435
436 /*
437  * Allocate HAMMER elements - meta data (inode, direntry, PFS, etc)
438  */
439 void *
440 alloc_meta_element(hammer_off_t *offp, int32_t data_len,
441                    struct buffer_info **data_bufferp)
442 {
443         void *data;
444
445         data = alloc_blockmap(HAMMER_ZONE_META_INDEX, data_len,
446                               offp, data_bufferp);
447         bzero(data, data_len);
448         return(data);
449 }
450
451 /*
452  * Format a new blockmap.  This is mostly a degenerate case because
453  * all allocations are now actually done from the freemap.
454  */
455 void
456 format_blockmap(struct volume_info *root_vol, int zone, hammer_off_t offset)
457 {
458         hammer_blockmap_t blockmap;
459         hammer_off_t zone_base;
460
461         /* Only root volume needs formatting */
462         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
463
464         assert(hammer_is_zone2_mapped_index(zone));
465
466         blockmap = &root_vol->ondisk->vol0_blockmap[zone];
467         zone_base = HAMMER_ZONE_ENCODE(zone, offset);
468
469         bzero(blockmap, sizeof(*blockmap));
470         blockmap->phys_offset = 0;
471         blockmap->first_offset = zone_base;
472         blockmap->next_offset = zone_base;
473         blockmap->alloc_offset = HAMMER_ENCODE(zone, 255, -1);
474         hammer_crc_set_blockmap(blockmap);
475 }
476
477 /*
478  * Format a new freemap.  Set all layer1 entries to UNAVAIL.  The initialize
479  * code will load each volume's freemap.
480  */
481 void
482 format_freemap(struct volume_info *root_vol)
483 {
484         struct buffer_info *buffer = NULL;
485         hammer_off_t layer1_offset;
486         hammer_blockmap_t blockmap;
487         hammer_blockmap_layer1_t layer1;
488         int i, isnew;
489
490         /* Only root volume needs formatting */
491         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
492
493         layer1_offset = alloc_bigblock(root_vol, HAMMER_ZONE_FREEMAP_INDEX);
494         for (i = 0; i < HAMMER_BIGBLOCK_SIZE; i += sizeof(*layer1)) {
495                 isnew = ((i % HAMMER_BUFSIZE) == 0);
496                 layer1 = get_buffer_data(layer1_offset + i, &buffer, isnew);
497                 bzero(layer1, sizeof(*layer1));
498                 layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
499                 layer1->blocks_free = 0;
500                 hammer_crc_set_layer1(layer1);
501         }
502         assert(i == HAMMER_BIGBLOCK_SIZE);
503         rel_buffer(buffer);
504
505         blockmap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
506         bzero(blockmap, sizeof(*blockmap));
507         blockmap->phys_offset = layer1_offset;
508         blockmap->first_offset = 0;
509         blockmap->next_offset = HAMMER_ENCODE_RAW_BUFFER(0, 0);
510         blockmap->alloc_offset = HAMMER_ENCODE_RAW_BUFFER(255, -1);
511         hammer_crc_set_blockmap(blockmap);
512 }
513
514 /*
515  * Load the volume's remaining free space into the freemap.
516  *
517  * Returns the number of big-blocks available.
518  */
519 int64_t
520 initialize_freemap(struct volume_info *vol)
521 {
522         struct volume_info *root_vol;
523         struct buffer_info *buffer1 = NULL;
524         struct buffer_info *buffer2 = NULL;
525         hammer_blockmap_layer1_t layer1;
526         hammer_blockmap_layer2_t layer2;
527         hammer_off_t layer1_offset;
528         hammer_off_t layer2_offset;
529         hammer_off_t phys_offset;
530         hammer_off_t block_offset;
531         hammer_off_t aligned_vol_free_end;
532         hammer_blockmap_t freemap;
533         int64_t count = 0;
534         int64_t layer1_count = 0;
535
536         root_vol = get_root_volume();
537
538         assert_volume_offset(vol);
539         aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK)
540                                 & ~HAMMER_BLOCKMAP_LAYER2_MASK;
541
542         printf("initialize freemap volume %d\n", vol->vol_no);
543
544         /*
545          * Initialize the freemap.  First preallocate the big-blocks required
546          * to implement layer2.   This preallocation is a bootstrap allocation
547          * using blocks from the target volume.
548          */
549         freemap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
550
551         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
552              phys_offset < aligned_vol_free_end;
553              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
554                 layer1_offset = freemap->phys_offset +
555                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
556                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
557                 if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
558                         layer1->phys_offset = alloc_bigblock(vol,
559                                                 HAMMER_ZONE_FREEMAP_INDEX);
560                         layer1->blocks_free = 0;
561                         buffer1->cache.modified = 1;
562                         hammer_crc_set_layer1(layer1);
563                 }
564         }
565
566         /*
567          * Now fill everything in.
568          */
569         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
570              phys_offset < aligned_vol_free_end;
571              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
572                 layer1_count = 0;
573                 layer1_offset = freemap->phys_offset +
574                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
575                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
576                 assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
577
578                 for (block_offset = 0;
579                      block_offset < HAMMER_BLOCKMAP_LAYER2;
580                      block_offset += HAMMER_BIGBLOCK_SIZE) {
581                         layer2_offset = layer1->phys_offset +
582                                         HAMMER_BLOCKMAP_LAYER2_OFFSET(block_offset);
583                         layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
584                         bzero(layer2, sizeof(*layer2));
585
586                         if (phys_offset + block_offset < vol->vol_free_off) {
587                                 /*
588                                  * Fixups XXX - big-blocks already allocated as part
589                                  * of the freemap bootstrap.
590                                  */
591                                 layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
592                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
593                                 layer2->bytes_free = 0;
594                         } else if (phys_offset + block_offset < vol->vol_free_end) {
595                                 layer2->zone = 0;
596                                 layer2->append_off = 0;
597                                 layer2->bytes_free = HAMMER_BIGBLOCK_SIZE;
598                                 ++count;
599                                 ++layer1_count;
600                         } else {
601                                 layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
602                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
603                                 layer2->bytes_free = 0;
604                         }
605                         hammer_crc_set_layer2(layer2);
606                         buffer2->cache.modified = 1;
607                 }
608
609                 layer1->blocks_free += layer1_count;
610                 hammer_crc_set_layer1(layer1);
611                 buffer1->cache.modified = 1;
612         }
613
614         rel_buffer(buffer1);
615         rel_buffer(buffer2);
616         return(count);
617 }
618
619 /*
620  * Returns the number of big-blocks available for filesystem data and undos
621  * without formatting.
622  */
623 int64_t
624 count_freemap(struct volume_info *vol)
625 {
626         hammer_off_t phys_offset;
627         hammer_off_t vol_free_off;
628         hammer_off_t aligned_vol_free_end;
629         int64_t count = 0;
630
631         vol_free_off = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
632
633         assert_volume_offset(vol);
634         aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK)
635                                 & ~HAMMER_BLOCKMAP_LAYER2_MASK;
636
637         if (vol->vol_no == HAMMER_ROOT_VOLNO)
638                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
639
640         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
641              phys_offset < aligned_vol_free_end;
642              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
643                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
644         }
645
646         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
647              phys_offset < aligned_vol_free_end;
648              phys_offset += HAMMER_BIGBLOCK_SIZE) {
649                 if (phys_offset < vol_free_off) {
650                         ;
651                 } else if (phys_offset < vol->vol_free_end) {
652                         ++count;
653                 }
654         }
655
656         return(count);
657 }
658
659 /*
660  * Format the undomap for the root volume.
661  */
662 void
663 format_undomap(struct volume_info *root_vol, int64_t *undo_buffer_size)
664 {
665         const int undo_zone = HAMMER_ZONE_UNDO_INDEX;
666         hammer_off_t undo_limit;
667         hammer_blockmap_t blockmap;
668         hammer_volume_ondisk_t ondisk;
669         struct buffer_info *buffer = NULL;
670         hammer_off_t scan;
671         int n;
672         int limit_index;
673         uint32_t seqno;
674
675         /* Only root volume needs formatting */
676         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
677         ondisk = root_vol->ondisk;
678
679         /*
680          * Size the undo buffer in multiples of HAMMER_BIGBLOCK_SIZE,
681          * up to HAMMER_UNDO_LAYER2 big-blocks.  Size to approximately
682          * 0.1% of the disk.
683          *
684          * The minimum UNDO fifo size is 500MB, or approximately 1% of
685          * the recommended 50G disk.
686          *
687          * Changing this minimum is rather dangerous as complex filesystem
688          * operations can cause the UNDO FIFO to fill up otherwise.
689          */
690         undo_limit = *undo_buffer_size;
691         if (undo_limit == 0) {
692                 undo_limit = HAMMER_VOL_BUF_SIZE(ondisk) / 1000;
693                 if (undo_limit < 500*1024*1024)
694                         undo_limit = 500*1024*1024;
695         }
696         undo_limit = (undo_limit + HAMMER_BIGBLOCK_MASK64) &
697                      ~HAMMER_BIGBLOCK_MASK64;
698         if (undo_limit < HAMMER_BIGBLOCK_SIZE)
699                 undo_limit = HAMMER_BIGBLOCK_SIZE;
700         if (undo_limit > HAMMER_BIGBLOCK_SIZE * HAMMER_UNDO_LAYER2)
701                 undo_limit = HAMMER_BIGBLOCK_SIZE * HAMMER_UNDO_LAYER2;
702         *undo_buffer_size = undo_limit;
703
704         blockmap = &ondisk->vol0_blockmap[undo_zone];
705         bzero(blockmap, sizeof(*blockmap));
706         blockmap->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
707         blockmap->first_offset = HAMMER_ZONE_ENCODE(undo_zone, 0);
708         blockmap->next_offset = blockmap->first_offset;
709         blockmap->alloc_offset = HAMMER_ZONE_ENCODE(undo_zone, undo_limit);
710         hammer_crc_set_blockmap(blockmap);
711
712         limit_index = undo_limit / HAMMER_BIGBLOCK_SIZE;
713         assert(limit_index <= HAMMER_UNDO_LAYER2);
714
715         for (n = 0; n < limit_index; ++n) {
716                 ondisk->vol0_undo_array[n] = alloc_bigblock(root_vol,
717                                                         HAMMER_ZONE_UNDO_INDEX);
718         }
719         while (n < HAMMER_UNDO_LAYER2) {
720                 ondisk->vol0_undo_array[n++] = HAMMER_BLOCKMAP_UNAVAIL;
721         }
722
723         /*
724          * Pre-initialize the UNDO blocks (HAMMER version 4+)
725          */
726         printf("initializing the undo map (%jd MB)\n",
727                 (intmax_t)(blockmap->alloc_offset & HAMMER_OFF_LONG_MASK) /
728                 (1024 * 1024));
729
730         scan = blockmap->first_offset;
731         seqno = 0;
732
733         while (scan < blockmap->alloc_offset) {
734                 hammer_fifo_head_t head;
735                 hammer_fifo_tail_t tail;
736                 int isnew;
737                 int bytes = HAMMER_UNDO_ALIGN;
738
739                 isnew = ((scan & HAMMER_BUFMASK64) == 0);
740                 head = get_buffer_data(scan, &buffer, isnew);
741                 buffer->cache.modified = 1;
742                 tail = (void *)((char *)head + bytes - sizeof(*tail));
743
744                 bzero(head, bytes);
745                 head->hdr_signature = HAMMER_HEAD_SIGNATURE;
746                 head->hdr_type = HAMMER_HEAD_TYPE_DUMMY;
747                 head->hdr_size = bytes;
748                 head->hdr_seq = seqno++;
749
750                 tail->tail_signature = HAMMER_TAIL_SIGNATURE;
751                 tail->tail_type = HAMMER_HEAD_TYPE_DUMMY;
752                 tail->tail_size = bytes;
753
754                 hammer_crc_set_fifo_head(head, bytes);
755
756                 scan += bytes;
757         }
758         rel_buffer(buffer);
759 }
760
761 const char *zone_labels[] = {
762         "",             /* 0 */
763         "raw_volume",   /* 1 */
764         "raw_buffer",   /* 2 */
765         "undo",         /* 3 */
766         "freemap",      /* 4 */
767         "",             /* 5 */
768         "",             /* 6 */
769         "",             /* 7 */
770         "btree",        /* 8 */
771         "meta",         /* 9 */
772         "large_data",   /* 10 */
773         "small_data",   /* 11 */
774         "",             /* 12 */
775         "",             /* 13 */
776         "",             /* 14 */
777         "unavail",      /* 15 */
778 };
779
780 void
781 print_blockmap(const struct volume_info *root_vol)
782 {
783         hammer_blockmap_t blockmap;
784         hammer_volume_ondisk_t ondisk;
785         int64_t size, used;
786         int i;
787 #define INDENT ""
788
789         ondisk = root_vol->ondisk;
790         printf(INDENT"vol_label\t%s\n", ondisk->vol_label);
791         printf(INDENT"vol_count\t%d\n", ondisk->vol_count);
792         printf(INDENT"vol_bot_beg\t%s\n", sizetostr(ondisk->vol_bot_beg));
793         printf(INDENT"vol_mem_beg\t%s\n", sizetostr(ondisk->vol_mem_beg));
794         printf(INDENT"vol_buf_beg\t%s\n", sizetostr(ondisk->vol_buf_beg));
795         printf(INDENT"vol_buf_end\t%s\n", sizetostr(ondisk->vol_buf_end));
796         printf(INDENT"vol0_next_tid\t%016jx\n",
797                (uintmax_t)ondisk->vol0_next_tid);
798
799         blockmap = &ondisk->vol0_blockmap[HAMMER_ZONE_UNDO_INDEX];
800         size = blockmap->alloc_offset & HAMMER_OFF_LONG_MASK;
801         if (blockmap->first_offset <= blockmap->next_offset)
802                 used = blockmap->next_offset - blockmap->first_offset;
803         else
804                 used = blockmap->alloc_offset - blockmap->first_offset +
805                         (blockmap->next_offset & HAMMER_OFF_LONG_MASK);
806         printf(INDENT"undo_size\t%s\n", sizetostr(size));
807         printf(INDENT"undo_used\t%s\n", sizetostr(used));
808
809         printf(INDENT"zone #             "
810                "phys             first            next             alloc\n");
811         for (i = 0; i < HAMMER_MAX_ZONES; i++) {
812                 blockmap = &ondisk->vol0_blockmap[i];
813                 printf(INDENT"zone %-2d %-10s %016jx %016jx %016jx %016jx\n",
814                         i, zone_labels[i],
815                         (uintmax_t)blockmap->phys_offset,
816                         (uintmax_t)blockmap->first_offset,
817                         (uintmax_t)blockmap->next_offset,
818                         (uintmax_t)blockmap->alloc_offset);
819         }
820 }
821
822 /*
823  * Flush various tracking structures to disk
824  */
825 void
826 flush_all_volumes(void)
827 {
828         struct volume_info *vol;
829
830         TAILQ_FOREACH(vol, &VolList, entry)
831                 flush_volume(vol);
832 }
833
834 void
835 flush_volume(struct volume_info *volume)
836 {
837         struct buffer_info *buffer;
838         int i;
839
840         for (i = 0; i < HAMMER_BUFLISTS; ++i) {
841                 TAILQ_FOREACH(buffer, &volume->buffer_lists[i], entry)
842                         flush_buffer(buffer);
843         }
844         if (writehammervol(volume) == -1)
845                 err(1, "Write volume %d (%s)", volume->vol_no, volume->name);
846 }
847
848 void
849 flush_buffer(struct buffer_info *buffer)
850 {
851         struct volume_info *vol;
852
853         vol = buffer->volume;
854         if (writehammerbuf(buffer) == -1)
855                 err(1, "Write volume %d (%s)", vol->vol_no, vol->name);
856         buffer->cache.modified = 0;
857 }
858
859 /*
860  * Core I/O operations
861  */
862 static int
863 __read(struct volume_info *vol, void *data, int64_t offset, int size)
864 {
865         ssize_t n;
866
867         n = pread(vol->fd, data, size, offset);
868         if (n != size)
869                 return(-1);
870         return(0);
871 }
872
873 static __inline int
874 readhammervol(struct volume_info *vol)
875 {
876         return(__read(vol, vol->ondisk, 0, HAMMER_BUFSIZE));
877 }
878
879 static __inline int
880 readhammerbuf(struct buffer_info *buf)
881 {
882         return(__read(buf->volume, buf->ondisk, buf->raw_offset, HAMMER_BUFSIZE));
883 }
884
885 static int
886 __write(struct volume_info *vol, const void *data, int64_t offset, int size)
887 {
888         ssize_t n;
889
890         if (vol->rdonly)
891                 return(0);
892
893         n = pwrite(vol->fd, data, size, offset);
894         if (n != size)
895                 return(-1);
896         return(0);
897 }
898
899 static __inline int
900 writehammervol(struct volume_info *vol)
901 {
902         return(__write(vol, vol->ondisk, 0, HAMMER_BUFSIZE));
903 }
904
905 static __inline int
906 writehammerbuf(struct buffer_info *buf)
907 {
908         return(__write(buf->volume, buf->ondisk, buf->raw_offset, HAMMER_BUFSIZE));
909 }
910
911 int64_t init_boot_area_size(int64_t value, off_t avg_vol_size)
912 {
913         if (value == 0) {
914                 value = HAMMER_BOOT_NOMBYTES;
915                 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
916                         value >>= 1;
917                 if (value < HAMMER_BOOT_MINBYTES)
918                         value = 0;
919         } else if (value < HAMMER_BOOT_MINBYTES) {
920                 value = HAMMER_BOOT_MINBYTES;
921         }
922
923         return(value);
924 }
925
926 int64_t init_mem_area_size(int64_t value, off_t avg_vol_size)
927 {
928         if (value == 0) {
929                 value = HAMMER_MEM_NOMBYTES;
930                 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
931                         value >>= 1;
932                 if (value < HAMMER_MEM_MINBYTES)
933                         value = 0;
934         } else if (value < HAMMER_MEM_MINBYTES) {
935                 value = HAMMER_MEM_MINBYTES;
936         }
937
938         return(value);
939 }