Merge branch 'vendor/OPENSSH'
[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 get_buffer_readahead(struct buffer_info *base);
41 static void *get_ondisk(hammer_off_t buf_offset, struct buffer_info **bufferp,
42                         int isnew);
43 static __inline int readhammervol(struct volume_info *vol);
44 static __inline int readhammerbuf(struct buffer_info *buf);
45 static __inline int writehammervol(struct volume_info *vol);
46 static __inline int writehammerbuf(struct buffer_info *buf);
47
48 int DebugOpt;
49
50 uuid_t Hammer_FSType;
51 uuid_t Hammer_FSId;
52 int     UseReadBehind = -4;
53 int     UseReadAhead = 4;
54 int     AssertOnFailure = 1;
55 struct volume_list VolList = TAILQ_HEAD_INITIALIZER(VolList);
56 static int valid_hammer_volumes;
57
58 static __inline
59 int
60 buffer_hash(hammer_off_t buf_offset)
61 {
62         int hi;
63
64         hi = (int)(buf_offset / HAMMER_BUFSIZE) & HAMMER_BUFLISTMASK;
65         return(hi);
66 }
67
68 static struct buffer_info*
69 find_buffer(struct volume_info *volume, hammer_off_t buf_offset)
70 {
71         int hi;
72         struct buffer_info *buf;
73
74         hi = buffer_hash(buf_offset);
75         TAILQ_FOREACH(buf, &volume->buffer_lists[hi], entry)
76                 if (buf->buf_offset == buf_offset)
77                         return(buf);
78         return(NULL);
79 }
80
81 static
82 struct volume_info *
83 __alloc_volume(const char *volname, int oflags)
84 {
85         struct volume_info *vol;
86         int i;
87
88         vol = malloc(sizeof(*vol));
89         if (vol == NULL)
90                 err(1, "alloc_volume");
91         bzero(vol, sizeof(*vol));
92
93         vol->vol_no = -1;
94         vol->rdonly = (oflags == O_RDONLY);
95         vol->name = strdup(volname);
96         vol->fd = open(vol->name, oflags);
97         if (vol->fd < 0)
98                 err(1, "alloc_volume: Failed to open %s", vol->name);
99
100         vol->size = 0;
101         vol->device_offset = 0;
102         vol->type = NULL;
103
104         vol->ondisk = malloc(HAMMER_BUFSIZE);
105         if (vol->ondisk == NULL)
106                 err(1, "alloc_volume");
107         bzero(vol->ondisk, HAMMER_BUFSIZE);
108
109         for (i = 0; i < HAMMER_BUFLISTS; ++i)
110                 TAILQ_INIT(&vol->buffer_lists[i]);
111
112         return(vol);
113 }
114
115 static void
116 __add_volume(struct volume_info *vol)
117 {
118         struct volume_info *scan;
119         struct stat st1, st2;
120
121         if (fstat(vol->fd, &st1) != 0)
122                 errx(1, "add_volume: %s: Failed to stat", vol->name);
123
124         TAILQ_FOREACH(scan, &VolList, entry) {
125                 if (scan->vol_no == vol->vol_no) {
126                         errx(1, "add_volume: %s: Duplicate volume number %d "
127                                 "against %s",
128                                 vol->name, vol->vol_no, scan->name);
129                 }
130                 if (fstat(scan->fd, &st2) != 0) {
131                         errx(1, "add_volume: %s: Failed to stat %s",
132                                 vol->name, scan->name);
133                 }
134                 if ((st1.st_ino == st2.st_ino) && (st1.st_dev == st2.st_dev)) {
135                         errx(1, "add_volume: %s: Specified more than once",
136                                 vol->name);
137                 }
138         }
139
140         TAILQ_INSERT_TAIL(&VolList, vol, entry);
141 }
142
143 /*
144  * Initialize a volume structure and ondisk vol_no field.
145  */
146 struct volume_info *
147 init_volume(int32_t vol_no, const char *filename, int oflags)
148 {
149         struct volume_info *vol;
150
151         vol = __alloc_volume(filename, oflags);
152         vol->vol_no = vol->ondisk->vol_no = vol_no;
153
154         __add_volume(vol);
155
156         return(vol);
157 }
158
159 /*
160  * Initialize a volume structure and read ondisk volume header.
161  */
162 struct volume_info*
163 load_volume(const char *filename, int oflags)
164 {
165         struct volume_info *vol;
166         struct hammer_volume_ondisk *ondisk;
167         int n;
168
169         vol = __alloc_volume(filename, oflags);
170
171         n = readhammervol(vol);
172         if (n == -1) {
173                 err(1, "load_volume: %s: Read failed at offset 0", vol->name);
174         }
175         ondisk = vol->ondisk;
176         vol->vol_no = ondisk->vol_no;
177
178         if (ondisk->vol_rootvol != HAMMER_ROOT_VOLNO) {
179                 errx(1, "load_volume: Invalid root volume# %d",
180                         ondisk->vol_rootvol);
181         }
182
183         if (bcmp(&Hammer_FSType, &ondisk->vol_fstype, sizeof(Hammer_FSType))) {
184                 errx(1, "load_volume: %s: Header does not indicate "
185                         "that this is a hammer volume", vol->name);
186         }
187
188         if (valid_hammer_volumes++ == 0) {
189                 Hammer_FSId = ondisk->vol_fsid;
190         } else if (bcmp(&Hammer_FSId, &ondisk->vol_fsid, sizeof(Hammer_FSId))) {
191                 errx(1, "load_volume: %s: FSId does match other volumes!",
192                         vol->name);
193         }
194
195         __add_volume(vol);
196
197         return(vol);
198 }
199
200 /*
201  * Check basic volume characteristics.
202  */
203 void
204 check_volume(struct volume_info *vol)
205 {
206         struct partinfo pinfo;
207         struct stat st;
208
209         /*
210          * Get basic information about the volume
211          */
212         if (ioctl(vol->fd, DIOCGPART, &pinfo) < 0) {
213                 /*
214                  * Allow the formatting of regular files as HAMMER volumes
215                  */
216                 if (fstat(vol->fd, &st) < 0)
217                         err(1, "Unable to stat %s", vol->name);
218                 vol->size = st.st_size;
219                 vol->type = "REGFILE";
220         } else {
221                 /*
222                  * When formatting a block device as a HAMMER volume the
223                  * sector size must be compatible.  HAMMER uses 16384 byte
224                  * filesystem buffers.
225                  */
226                 if (pinfo.reserved_blocks) {
227                         errx(1, "HAMMER cannot be placed in a partition "
228                                 "which overlaps the disklabel or MBR");
229                 }
230                 if (pinfo.media_blksize > HAMMER_BUFSIZE ||
231                     HAMMER_BUFSIZE % pinfo.media_blksize) {
232                         errx(1, "A media sector size of %d is not supported",
233                              pinfo.media_blksize);
234                 }
235
236                 vol->size = pinfo.media_size;
237                 vol->device_offset = pinfo.media_offset;
238                 vol->type = "DEVICE";
239         }
240 }
241
242 struct volume_info *
243 get_volume(int32_t vol_no)
244 {
245         struct volume_info *vol;
246
247         TAILQ_FOREACH(vol, &VolList, entry) {
248                 if (vol->vol_no == vol_no)
249                         break;
250         }
251         if (vol == NULL)
252                 errx(1, "get_volume: Volume %d does not exist!", vol_no);
253
254         /* not added to or removed from hammer cache */
255         return(vol);
256 }
257
258 struct volume_info *
259 get_root_volume(void)
260 {
261         return(get_volume(HAMMER_ROOT_VOLNO));
262 }
263
264 void
265 rel_volume(struct volume_info *volume __unused)
266 {
267         /* nothing to do */
268 }
269
270 /*
271  * Acquire the specified buffer.  isnew is -1 only when called
272  * via get_buffer_readahead() to prevent another readahead.
273  */
274 struct buffer_info *
275 get_buffer(hammer_off_t buf_offset, int isnew)
276 {
277         struct buffer_info *buf;
278         struct volume_info *volume;
279         int vol_no;
280         int zone;
281         int hi;
282         int dora = 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, NULL);
287         }
288         if (buf_offset == HAMMER_OFF_BAD)
289                 return(NULL);
290         assert(hammer_is_zone_raw_buffer(buf_offset));
291
292         vol_no = HAMMER_VOL_DECODE(buf_offset);
293         volume = get_volume(vol_no);
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                 buf->cache.buffer = buf;
319                 hammer_cache_add(&buf->cache);
320                 dora = (isnew == 0);
321         } else {
322                 assert(buf->ondisk != NULL);
323                 assert(isnew != -1);
324                 hammer_cache_used(&buf->cache);
325         }
326
327         ++buf->cache.refs;
328         hammer_cache_flush();
329
330         if (isnew > 0) {
331                 assert(buf->cache.modified == 0);
332                 bzero(buf->ondisk, HAMMER_BUFSIZE);
333                 buf->cache.modified = 1;
334         }
335         if (dora)
336                 get_buffer_readahead(buf);
337         return(buf);
338 }
339
340 static void
341 get_buffer_readahead(struct buffer_info *base)
342 {
343         struct buffer_info *buf;
344         struct volume_info *vol;
345         hammer_off_t buf_offset;
346         int64_t raw_offset;
347         int ri = UseReadBehind;
348         int re = UseReadAhead;
349
350         raw_offset = base->raw_offset + ri * HAMMER_BUFSIZE;
351         vol = base->volume;
352
353         while (ri < re) {
354                 if (raw_offset >= vol->ondisk->vol_buf_end)
355                         break;
356                 if (raw_offset < vol->ondisk->vol_buf_beg || ri == 0) {
357                         ++ri;
358                         raw_offset += HAMMER_BUFSIZE;
359                         continue;
360                 }
361                 buf_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no,
362                         raw_offset - vol->ondisk->vol_buf_beg);
363                 buf = find_buffer(vol, buf_offset);
364                 if (buf == NULL) {
365                         buf = get_buffer(buf_offset, -1);
366                         rel_buffer(buf);
367                 }
368                 ++ri;
369                 raw_offset += HAMMER_BUFSIZE;
370         }
371 }
372
373 void
374 rel_buffer(struct buffer_info *buffer)
375 {
376         struct volume_info *volume;
377         int hi;
378
379         if (buffer == NULL)
380                 return;
381         assert(buffer->cache.refs > 0);
382         if (--buffer->cache.refs == 0) {
383                 if (buffer->cache.delete) {
384                         hi = buffer_hash(buffer->buf_offset);
385                         volume = buffer->volume;
386                         if (buffer->cache.modified)
387                                 flush_buffer(buffer);
388                         TAILQ_REMOVE(&volume->buffer_lists[hi], buffer, entry);
389                         hammer_cache_del(&buffer->cache);
390                         free(buffer->ondisk);
391                         free(buffer);
392                         rel_volume(volume);
393                 }
394         }
395 }
396
397 /*
398  * Retrieve a pointer to a buffer data given a buffer offset.  The underlying
399  * bufferp is freed if isnew or the offset is out of range of the cached data.
400  * If bufferp is freed a referenced buffer is loaded into it.
401  */
402 void *
403 get_buffer_data(hammer_off_t buf_offset, struct buffer_info **bufferp,
404                 int isnew)
405 {
406         if (*bufferp != NULL) {
407                 if (isnew > 0 ||
408                     (((*bufferp)->buf_offset ^ buf_offset) & ~HAMMER_BUFMASK64)) {
409                         rel_buffer(*bufferp);
410                         *bufferp = NULL;
411                 }
412         }
413         return(get_ondisk(buf_offset, bufferp, isnew));
414 }
415
416 /*
417  * Retrieve a pointer to a B-Tree node given a zone offset.  The underlying
418  * bufferp is freed if non-NULL and a referenced buffer is loaded into it.
419  */
420 hammer_node_ondisk_t
421 get_node(hammer_off_t node_offset, struct buffer_info **bufferp)
422 {
423         if (*bufferp != NULL) {
424                 rel_buffer(*bufferp);
425                 *bufferp = NULL;
426         }
427         return(get_ondisk(node_offset, bufferp, 0));
428 }
429
430 /*
431  * Return a pointer to a buffer data given a buffer offset.
432  * If *bufferp is NULL acquire the buffer otherwise use that buffer.
433  */
434 static void *
435 get_ondisk(hammer_off_t buf_offset, struct buffer_info **bufferp, int isnew)
436 {
437         struct buffer_info *buffer;
438
439         buffer = *bufferp;
440         if (buffer == NULL) {
441                 buffer = *bufferp = get_buffer(buf_offset, isnew);
442                 if (buffer == NULL)
443                         return(NULL);
444         }
445
446         return((char *)buffer->ondisk +
447                 ((int32_t)buf_offset & HAMMER_BUFMASK));
448 }
449
450 /*
451  * Allocate HAMMER elements - B-Tree nodes
452  */
453 void *
454 alloc_btree_element(hammer_off_t *offp, struct buffer_info **data_bufferp)
455 {
456         hammer_node_ondisk_t node;
457
458         node = alloc_blockmap(HAMMER_ZONE_BTREE_INDEX, sizeof(*node),
459                               offp, data_bufferp);
460         bzero(node, sizeof(*node));
461         return (node);
462 }
463
464 /*
465  * Allocate HAMMER elements - meta data (inode, direntry, PFS, etc)
466  */
467 void *
468 alloc_meta_element(hammer_off_t *offp, int32_t data_len,
469                    struct buffer_info **data_bufferp)
470 {
471         void *data;
472
473         data = alloc_blockmap(HAMMER_ZONE_META_INDEX, data_len,
474                               offp, data_bufferp);
475         bzero(data, data_len);
476         return (data);
477 }
478
479 /*
480  * Allocate HAMMER elements - data storage
481  *
482  * The only data_len supported by HAMMER userspace for large data zone
483  * (zone 10) is HAMMER_BUFSIZE which is 16KB.  >16KB data does not fit
484  * in a buffer allocated by get_buffer().  Also alloc_blockmap() does
485  * not consider >16KB buffer size.
486  */
487 void *
488 alloc_data_element(hammer_off_t *offp, int32_t data_len,
489                    struct buffer_info **data_bufferp)
490 {
491         void *data;
492         int zone;
493
494         if (data_len == 0)
495                 return(NULL);
496
497         zone = hammer_data_zone_index(data_len);
498         assert(data_len <= HAMMER_BUFSIZE); /* just one buffer */
499         assert(zone == HAMMER_ZONE_LARGE_DATA_INDEX ||
500                zone == HAMMER_ZONE_SMALL_DATA_INDEX);
501
502         data = alloc_blockmap(zone, data_len, offp, data_bufferp);
503         bzero(data, data_len);
504         return(data);
505 }
506
507 /*
508  * Format a new blockmap.  This is mostly a degenerate case because
509  * all allocations are now actually done from the freemap.
510  */
511 void
512 format_blockmap(struct volume_info *root_vol, int zone, hammer_off_t offset)
513 {
514         hammer_blockmap_t blockmap;
515         hammer_off_t zone_base;
516
517         /* Only root volume needs formatting */
518         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
519
520         assert(hammer_is_zone2_mapped_index(zone));
521
522         blockmap = &root_vol->ondisk->vol0_blockmap[zone];
523         zone_base = HAMMER_ZONE_ENCODE(zone, offset);
524
525         bzero(blockmap, sizeof(*blockmap));
526         blockmap->phys_offset = 0;
527         blockmap->first_offset = zone_base;
528         blockmap->next_offset = zone_base;
529         blockmap->alloc_offset = HAMMER_ENCODE(zone, 255, -1);
530         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
531 }
532
533 /*
534  * Format a new freemap.  Set all layer1 entries to UNAVAIL.  The initialize
535  * code will load each volume's freemap.
536  */
537 void
538 format_freemap(struct volume_info *root_vol)
539 {
540         struct buffer_info *buffer = NULL;
541         hammer_off_t layer1_offset;
542         hammer_blockmap_t blockmap;
543         struct hammer_blockmap_layer1 *layer1;
544         int i, isnew;
545
546         /* Only root volume needs formatting */
547         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
548
549         layer1_offset = alloc_bigblock(root_vol, HAMMER_ZONE_FREEMAP_INDEX);
550         for (i = 0; i < HAMMER_BIGBLOCK_SIZE; i += sizeof(*layer1)) {
551                 isnew = ((i % HAMMER_BUFSIZE) == 0);
552                 layer1 = get_buffer_data(layer1_offset + i, &buffer, isnew);
553                 bzero(layer1, sizeof(*layer1));
554                 layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
555                 layer1->blocks_free = 0;
556                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
557         }
558         assert(i == HAMMER_BIGBLOCK_SIZE);
559         rel_buffer(buffer);
560
561         blockmap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
562         bzero(blockmap, sizeof(*blockmap));
563         blockmap->phys_offset = layer1_offset;
564         blockmap->first_offset = 0;
565         blockmap->next_offset = HAMMER_ENCODE_RAW_BUFFER(0, 0);
566         blockmap->alloc_offset = HAMMER_ENCODE_RAW_BUFFER(255, -1);
567         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
568 }
569
570 /*
571  * Load the volume's remaining free space into the freemap.
572  *
573  * Returns the number of big-blocks available.
574  */
575 int64_t
576 initialize_freemap(struct volume_info *vol)
577 {
578         struct volume_info *root_vol;
579         struct buffer_info *buffer1 = NULL;
580         struct buffer_info *buffer2 = NULL;
581         struct hammer_blockmap_layer1 *layer1;
582         struct hammer_blockmap_layer2 *layer2;
583         hammer_off_t layer1_offset;
584         hammer_off_t layer2_offset;
585         hammer_off_t phys_offset;
586         hammer_off_t block_offset;
587         hammer_off_t aligned_vol_free_end;
588         hammer_blockmap_t freemap;
589         int64_t count = 0;
590         int64_t layer1_count = 0;
591
592         root_vol = get_root_volume();
593         aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK)
594                                 & ~HAMMER_BLOCKMAP_LAYER2_MASK;
595
596         printf("initialize freemap volume %d\n", vol->vol_no);
597
598         /*
599          * Initialize the freemap.  First preallocate the big-blocks required
600          * to implement layer2.   This preallocation is a bootstrap allocation
601          * using blocks from the target volume.
602          */
603         freemap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
604
605         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
606              phys_offset < aligned_vol_free_end;
607              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
608                 layer1_offset = freemap->phys_offset +
609                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
610                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
611                 if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
612                         layer1->phys_offset = alloc_bigblock(vol,
613                                                 HAMMER_ZONE_FREEMAP_INDEX);
614                         layer1->blocks_free = 0;
615                         buffer1->cache.modified = 1;
616                         layer1->layer1_crc = crc32(layer1,
617                                                    HAMMER_LAYER1_CRCSIZE);
618                 }
619         }
620
621         /*
622          * Now fill everything in.
623          */
624         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
625              phys_offset < aligned_vol_free_end;
626              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
627                 layer1_count = 0;
628                 layer1_offset = freemap->phys_offset +
629                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
630                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
631                 assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
632
633                 for (block_offset = 0;
634                      block_offset < HAMMER_BLOCKMAP_LAYER2;
635                      block_offset += HAMMER_BIGBLOCK_SIZE) {
636                         layer2_offset = layer1->phys_offset +
637                                         HAMMER_BLOCKMAP_LAYER2_OFFSET(block_offset);
638                         layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
639                         bzero(layer2, sizeof(*layer2));
640
641                         if (phys_offset + block_offset < vol->vol_free_off) {
642                                 /*
643                                  * Fixups XXX - big-blocks already allocated as part
644                                  * of the freemap bootstrap.
645                                  */
646                                 layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
647                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
648                                 layer2->bytes_free = 0;
649                         } else if (phys_offset + block_offset < vol->vol_free_end) {
650                                 layer2->zone = 0;
651                                 layer2->append_off = 0;
652                                 layer2->bytes_free = HAMMER_BIGBLOCK_SIZE;
653                                 ++count;
654                                 ++layer1_count;
655                         } else {
656                                 layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
657                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
658                                 layer2->bytes_free = 0;
659                         }
660                         layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
661                         buffer2->cache.modified = 1;
662                 }
663
664                 layer1->blocks_free += layer1_count;
665                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
666                 buffer1->cache.modified = 1;
667         }
668
669         rel_buffer(buffer1);
670         rel_buffer(buffer2);
671         rel_volume(root_vol);
672         return(count);
673 }
674
675 /*
676  * Returns the number of big-blocks available for filesystem data and undos
677  * without formatting.
678  */
679 int64_t
680 count_freemap(struct volume_info *vol)
681 {
682         hammer_off_t phys_offset;
683         hammer_off_t vol_free_off;
684         hammer_off_t aligned_vol_free_end;
685         int64_t count = 0;
686
687         vol_free_off = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
688         aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK)
689                                 & ~HAMMER_BLOCKMAP_LAYER2_MASK;
690
691         if (vol->vol_no == HAMMER_ROOT_VOLNO)
692                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
693
694         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
695              phys_offset < aligned_vol_free_end;
696              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
697                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
698         }
699
700         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
701              phys_offset < aligned_vol_free_end;
702              phys_offset += HAMMER_BIGBLOCK_SIZE) {
703                 if (phys_offset < vol_free_off) {
704                         ;
705                 } else if (phys_offset < vol->vol_free_end) {
706                         ++count;
707                 }
708         }
709
710         return(count);
711 }
712
713 /*
714  * Format the undomap for the root volume.
715  */
716 void
717 format_undomap(struct volume_info *root_vol, int64_t *undo_buffer_size)
718 {
719         const int undo_zone = HAMMER_ZONE_UNDO_INDEX;
720         hammer_off_t undo_limit;
721         hammer_blockmap_t blockmap;
722         struct hammer_volume_ondisk *ondisk;
723         struct buffer_info *buffer = NULL;
724         hammer_off_t scan;
725         int n;
726         int limit_index;
727         uint32_t seqno;
728
729         /* Only root volume needs formatting */
730         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
731         ondisk = root_vol->ondisk;
732
733         /*
734          * Size the undo buffer in multiples of HAMMER_BIGBLOCK_SIZE,
735          * up to HAMMER_UNDO_LAYER2 big-blocks.  Size to approximately
736          * 0.1% of the disk.
737          *
738          * The minimum UNDO fifo size is 500MB, or approximately 1% of
739          * the recommended 50G disk.
740          *
741          * Changing this minimum is rather dangerous as complex filesystem
742          * operations can cause the UNDO FIFO to fill up otherwise.
743          */
744         undo_limit = *undo_buffer_size;
745         if (undo_limit == 0) {
746                 undo_limit = (ondisk->vol_buf_end - ondisk->vol_buf_beg) / 1000;
747                 if (undo_limit < 500*1024*1024)
748                         undo_limit = 500*1024*1024;
749         }
750         undo_limit = (undo_limit + HAMMER_BIGBLOCK_MASK64) &
751                      ~HAMMER_BIGBLOCK_MASK64;
752         if (undo_limit < HAMMER_BIGBLOCK_SIZE)
753                 undo_limit = HAMMER_BIGBLOCK_SIZE;
754         if (undo_limit > HAMMER_BIGBLOCK_SIZE * HAMMER_UNDO_LAYER2)
755                 undo_limit = HAMMER_BIGBLOCK_SIZE * HAMMER_UNDO_LAYER2;
756         *undo_buffer_size = undo_limit;
757
758         blockmap = &ondisk->vol0_blockmap[undo_zone];
759         bzero(blockmap, sizeof(*blockmap));
760         blockmap->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
761         blockmap->first_offset = HAMMER_ZONE_ENCODE(undo_zone, 0);
762         blockmap->next_offset = blockmap->first_offset;
763         blockmap->alloc_offset = HAMMER_ZONE_ENCODE(undo_zone, undo_limit);
764         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
765
766         limit_index = undo_limit / HAMMER_BIGBLOCK_SIZE;
767         assert(limit_index <= HAMMER_UNDO_LAYER2);
768
769         for (n = 0; n < limit_index; ++n) {
770                 ondisk->vol0_undo_array[n] = alloc_bigblock(root_vol,
771                                                         HAMMER_ZONE_UNDO_INDEX);
772         }
773         while (n < HAMMER_UNDO_LAYER2) {
774                 ondisk->vol0_undo_array[n++] = HAMMER_BLOCKMAP_UNAVAIL;
775         }
776
777         /*
778          * Pre-initialize the UNDO blocks (HAMMER version 4+)
779          */
780         printf("initializing the undo map (%jd MB)\n",
781                 (intmax_t)(blockmap->alloc_offset & HAMMER_OFF_LONG_MASK) /
782                 (1024 * 1024));
783
784         scan = blockmap->first_offset;
785         seqno = 0;
786
787         while (scan < blockmap->alloc_offset) {
788                 hammer_fifo_head_t head;
789                 hammer_fifo_tail_t tail;
790                 int isnew;
791                 int bytes = HAMMER_UNDO_ALIGN;
792
793                 isnew = ((scan & HAMMER_BUFMASK64) == 0);
794                 head = get_buffer_data(scan, &buffer, isnew);
795                 buffer->cache.modified = 1;
796                 tail = (void *)((char *)head + bytes - sizeof(*tail));
797
798                 bzero(head, bytes);
799                 head->hdr_signature = HAMMER_HEAD_SIGNATURE;
800                 head->hdr_type = HAMMER_HEAD_TYPE_DUMMY;
801                 head->hdr_size = bytes;
802                 head->hdr_seq = seqno++;
803
804                 tail->tail_signature = HAMMER_TAIL_SIGNATURE;
805                 tail->tail_type = HAMMER_HEAD_TYPE_DUMMY;
806                 tail->tail_size = bytes;
807
808                 head->hdr_crc = crc32(head, HAMMER_FIFO_HEAD_CRCOFF) ^
809                                 crc32(head + 1, bytes - sizeof(*head));
810
811                 scan += bytes;
812         }
813         rel_buffer(buffer);
814 }
815
816 /*
817  * Flush various tracking structures to disk
818  */
819 void
820 flush_all_volumes(void)
821 {
822         struct volume_info *vol;
823
824         TAILQ_FOREACH(vol, &VolList, entry)
825                 flush_volume(vol);
826 }
827
828 void
829 flush_volume(struct volume_info *volume)
830 {
831         struct buffer_info *buffer;
832         int i;
833
834         for (i = 0; i < HAMMER_BUFLISTS; ++i) {
835                 TAILQ_FOREACH(buffer, &volume->buffer_lists[i], entry)
836                         flush_buffer(buffer);
837         }
838         if (writehammervol(volume) == -1)
839                 err(1, "Write volume %d (%s)", volume->vol_no, volume->name);
840 }
841
842 void
843 flush_buffer(struct buffer_info *buffer)
844 {
845         struct volume_info *vol;
846
847         vol = buffer->volume;
848         if (writehammerbuf(buffer) == -1)
849                 err(1, "Write volume %d (%s)", vol->vol_no, vol->name);
850         buffer->cache.modified = 0;
851 }
852
853 /*
854  * Core I/O operations
855  */
856 static int
857 __read(struct volume_info *vol, void *data, int64_t offset, int size)
858 {
859         ssize_t n;
860
861         n = pread(vol->fd, data, size, offset);
862         if (n != size)
863                 return(-1);
864         return(0);
865 }
866
867 static __inline int
868 readhammervol(struct volume_info *vol)
869 {
870         return(__read(vol, vol->ondisk, 0, HAMMER_BUFSIZE));
871 }
872
873 static __inline int
874 readhammerbuf(struct buffer_info *buf)
875 {
876         return(__read(buf->volume, buf->ondisk, buf->raw_offset, HAMMER_BUFSIZE));
877 }
878
879 static int
880 __write(struct volume_info *vol, const void *data, int64_t offset, int size)
881 {
882         ssize_t n;
883
884         if (vol->rdonly)
885                 return(0);
886
887         n = pwrite(vol->fd, data, size, offset);
888         if (n != size)
889                 return(-1);
890         return(0);
891 }
892
893 static __inline int
894 writehammervol(struct volume_info *vol)
895 {
896         return(__write(vol, vol->ondisk, 0, HAMMER_BUFSIZE));
897 }
898
899 static __inline int
900 writehammerbuf(struct buffer_info *buf)
901 {
902         return(__write(buf->volume, buf->ondisk, buf->raw_offset, HAMMER_BUFSIZE));
903 }
904
905 int64_t init_boot_area_size(int64_t value, off_t avg_vol_size)
906 {
907         if (value == 0) {
908                 value = HAMMER_BOOT_NOMBYTES;
909                 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
910                         value >>= 1;
911                 if (value < HAMMER_BOOT_MINBYTES)
912                         value = 0;
913         } else if (value < HAMMER_BOOT_MINBYTES) {
914                 value = HAMMER_BOOT_MINBYTES;
915         }
916
917         return(value);
918 }
919
920 int64_t init_mem_area_size(int64_t value, off_t avg_vol_size)
921 {
922         if (value == 0) {
923                 value = HAMMER_MEM_NOMBYTES;
924                 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
925                         value >>= 1;
926                 if (value < HAMMER_MEM_MINBYTES)
927                         value = 0;
928         } else if (value < HAMMER_MEM_MINBYTES) {
929                 value = HAMMER_MEM_MINBYTES;
930         }
931
932         return(value);
933 }