sbin/hammer: Refactor get_buffer()
[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 int readhammerbuf(struct volume_info *vol, void *data, int64_t offset);
44 static int writehammerbuf(struct volume_info *vol, const void *data,
45                         int64_t offset);
46
47 int DebugOpt;
48
49 uuid_t Hammer_FSType;
50 uuid_t Hammer_FSId;
51 int     UseReadBehind = -4;
52 int     UseReadAhead = 4;
53 int     AssertOnFailure = 1;
54 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
99         vol->size = 0;
100         vol->device_offset = 0;
101         vol->type = NULL;
102
103         vol->ondisk = malloc(HAMMER_BUFSIZE);
104         if (vol->ondisk == NULL)
105                 err(1, "alloc_volume");
106         bzero(vol->ondisk, HAMMER_BUFSIZE);
107
108         for (i = 0; i < HAMMER_BUFLISTS; ++i)
109                 TAILQ_INIT(&vol->buffer_lists[i]);
110
111         return(vol);
112 }
113
114 static void
115 __add_volume(struct volume_info *vol)
116 {
117         struct volume_info *scan;
118         struct stat st1, st2;
119
120         if (fstat(vol->fd, &st1) != 0)
121                 errx(1, "add_volume: %s: Failed to stat", vol->name);
122
123         TAILQ_FOREACH(scan, &VolList, entry) {
124                 if (scan->vol_no == vol->vol_no) {
125                         errx(1, "add_volume: %s: Duplicate volume number %d "
126                                 "against %s",
127                                 vol->name, vol->vol_no, scan->name);
128                 }
129                 if (fstat(scan->fd, &st2) != 0) {
130                         errx(1, "add_volume: %s: Failed to stat %s",
131                                 vol->name, scan->name);
132                 }
133                 if ((st1.st_ino == st2.st_ino) && (st1.st_dev == st2.st_dev)) {
134                         errx(1, "add_volume: %s: Specified more than once",
135                                 vol->name);
136                 }
137         }
138
139         TAILQ_INSERT_TAIL(&VolList, vol, entry);
140 }
141
142 /*
143  * Initialize a volume structure and ondisk vol_no field.
144  */
145 struct volume_info *
146 init_volume(int32_t vol_no, const char *filename, int oflags)
147 {
148         struct volume_info *vol;
149
150         vol = __alloc_volume(filename, oflags);
151         vol->vol_no = vol->ondisk->vol_no = vol_no;
152
153         __add_volume(vol);
154
155         return(vol);
156 }
157
158 /*
159  * Initialize a volume structure and read ondisk volume header.
160  */
161 struct volume_info*
162 load_volume(const char *filename, int oflags)
163 {
164         struct volume_info *vol;
165         struct hammer_volume_ondisk *ondisk;
166         int n;
167
168         vol = __alloc_volume(filename, oflags);
169         ondisk = vol->ondisk;
170
171         n = readhammerbuf(vol, ondisk, 0);
172         if (n == -1) {
173                 err(1, "load_volume: %s: Read failed at offset 0", vol->name);
174         }
175         vol->vol_no = ondisk->vol_no;
176
177         if (ondisk->vol_rootvol != HAMMER_ROOT_VOLNO) {
178                 errx(1, "load_volume: Invalid root volume# %d",
179                         ondisk->vol_rootvol);
180         }
181
182         if (bcmp(&Hammer_FSType, &ondisk->vol_fstype, sizeof(Hammer_FSType))) {
183                 errx(1, "load_volume: %s: Header does not indicate "
184                         "that this is a hammer volume", vol->name);
185         }
186
187         if (valid_hammer_volumes++ == 0) {
188                 Hammer_FSId = ondisk->vol_fsid;
189         } else if (bcmp(&Hammer_FSId, &ondisk->vol_fsid, sizeof(Hammer_FSId))) {
190                 errx(1, "load_volume: %s: FSId does match other volumes!",
191                         vol->name);
192         }
193
194         __add_volume(vol);
195
196         return(vol);
197 }
198
199 /*
200  * Check basic volume characteristics.
201  */
202 void
203 check_volume(struct volume_info *vol)
204 {
205         struct partinfo pinfo;
206         struct stat st;
207
208         /*
209          * Get basic information about the volume
210          */
211         if (ioctl(vol->fd, DIOCGPART, &pinfo) < 0) {
212                 /*
213                  * Allow the formatting of regular files as HAMMER volumes
214                  */
215                 if (fstat(vol->fd, &st) < 0)
216                         err(1, "Unable to stat %s", vol->name);
217                 vol->size = st.st_size;
218                 vol->type = "REGFILE";
219         } else {
220                 /*
221                  * When formatting a block device as a HAMMER volume the
222                  * sector size must be compatible.  HAMMER uses 16384 byte
223                  * filesystem buffers.
224                  */
225                 if (pinfo.reserved_blocks) {
226                         errx(1, "HAMMER cannot be placed in a partition "
227                                 "which overlaps the disklabel or MBR");
228                 }
229                 if (pinfo.media_blksize > HAMMER_BUFSIZE ||
230                     HAMMER_BUFSIZE % pinfo.media_blksize) {
231                         errx(1, "A media sector size of %d is not supported",
232                              pinfo.media_blksize);
233                 }
234
235                 vol->size = pinfo.media_size;
236                 vol->device_offset = pinfo.media_offset;
237                 vol->type = "DEVICE";
238         }
239 }
240
241 struct volume_info *
242 get_volume(int32_t vol_no)
243 {
244         struct volume_info *vol;
245
246         TAILQ_FOREACH(vol, &VolList, entry) {
247                 if (vol->vol_no == vol_no)
248                         break;
249         }
250         if (vol == NULL)
251                 errx(1, "get_volume: Volume %d does not exist!", vol_no);
252
253         /* not added to or removed from hammer cache */
254         return(vol);
255 }
256
257 struct volume_info *
258 get_root_volume(void)
259 {
260         return(get_volume(HAMMER_ROOT_VOLNO));
261 }
262
263 void
264 rel_volume(struct volume_info *volume __unused)
265 {
266         /* nothing to do */
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 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         hammer_off_t orig_offset = buf_offset;
279         int vol_no;
280         int zone;
281         int hi, n;
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                 if (DebugOpt > 1) {
302                         fprintf(stderr, "get_buffer: %016jx %016jx at %p\n",
303                                 (intmax_t)orig_offset, (intmax_t)buf_offset,
304                                 buf);
305                 }
306                 buf->buf_offset = buf_offset;
307                 buf->raw_offset = hammer_xlate_to_phys(volume->ondisk,
308                                                         buf_offset);
309                 buf->volume = volume;
310                 buf->ondisk = malloc(HAMMER_BUFSIZE);
311                 if (isnew <= 0) {
312                         n = readhammerbuf(volume, buf->ondisk, buf->raw_offset);
313                         if (n == -1) {
314                                 if (AssertOnFailure)
315                                         err(1, "get_buffer: %s:%016jx "
316                                             "Read failed at offset %016jx",
317                                             volume->name,
318                                             (intmax_t)buf->buf_offset,
319                                             (intmax_t)buf->raw_offset);
320                                 bzero(buf->ondisk, HAMMER_BUFSIZE);
321                         }
322                 }
323
324                 hi = buffer_hash(buf_offset);
325                 TAILQ_INSERT_TAIL(&volume->buffer_lists[hi], buf, entry);
326                 buf->cache.buffer = buf;
327                 hammer_cache_add(&buf->cache);
328                 dora = (isnew == 0);
329         } else {
330                 if (DebugOpt > 1) {
331                         fprintf(stderr, "get_buffer: %016jx %016jx at %p *\n",
332                                 (intmax_t)orig_offset, (intmax_t)buf_offset,
333                                 buf);
334                 }
335                 assert(buf->ondisk != NULL);
336                 hammer_cache_used(&buf->cache);
337         }
338
339         ++buf->cache.refs;
340         hammer_cache_flush();
341
342         if (isnew > 0) {
343                 assert(buf->cache.modified == 0);
344                 bzero(buf->ondisk, HAMMER_BUFSIZE);
345                 buf->cache.modified = 1;
346         }
347         if (dora)
348                 get_buffer_readahead(buf);
349         return(buf);
350 }
351
352 static void
353 get_buffer_readahead(struct buffer_info *base)
354 {
355         struct buffer_info *buf;
356         struct volume_info *vol;
357         hammer_off_t buf_offset;
358         int64_t raw_offset;
359         int ri = UseReadBehind;
360         int re = UseReadAhead;
361
362         raw_offset = base->raw_offset + ri * HAMMER_BUFSIZE;
363         vol = base->volume;
364
365         while (ri < re) {
366                 if (raw_offset >= vol->ondisk->vol_buf_end)
367                         break;
368                 if (raw_offset < vol->ondisk->vol_buf_beg || ri == 0) {
369                         ++ri;
370                         raw_offset += HAMMER_BUFSIZE;
371                         continue;
372                 }
373                 buf_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no,
374                         raw_offset - vol->ondisk->vol_buf_beg);
375                 buf = find_buffer(vol, buf_offset);
376                 if (buf == NULL) {
377                         buf = get_buffer(buf_offset, -1);
378                         rel_buffer(buf);
379                 }
380                 ++ri;
381                 raw_offset += HAMMER_BUFSIZE;
382         }
383 }
384
385 void
386 rel_buffer(struct buffer_info *buffer)
387 {
388         struct volume_info *volume;
389         int hi;
390
391         if (buffer == NULL)
392                 return;
393         assert(buffer->cache.refs > 0);
394         if (--buffer->cache.refs == 0) {
395                 if (buffer->cache.delete) {
396                         hi = buffer_hash(buffer->buf_offset);
397                         volume = buffer->volume;
398                         if (buffer->cache.modified)
399                                 flush_buffer(buffer);
400                         TAILQ_REMOVE(&volume->buffer_lists[hi], buffer, entry);
401                         hammer_cache_del(&buffer->cache);
402                         free(buffer->ondisk);
403                         free(buffer);
404                         rel_volume(volume);
405                 }
406         }
407 }
408
409 /*
410  * Retrieve a pointer to a buffer data given a buffer offset.  The underlying
411  * bufferp is freed if isnew or the offset is out of range of the cached data.
412  * If bufferp is freed a referenced buffer is loaded into it.
413  */
414 void *
415 get_buffer_data(hammer_off_t buf_offset, struct buffer_info **bufferp,
416                 int isnew)
417 {
418         if (*bufferp != NULL) {
419                 if (isnew > 0 ||
420                     (((*bufferp)->buf_offset ^ buf_offset) & ~HAMMER_BUFMASK64)) {
421                         rel_buffer(*bufferp);
422                         *bufferp = NULL;
423                 }
424         }
425         return(get_ondisk(buf_offset, bufferp, isnew));
426 }
427
428 /*
429  * Retrieve a pointer to a B-Tree node given a zone offset.  The underlying
430  * bufferp is freed if non-NULL and a referenced buffer is loaded into it.
431  */
432 hammer_node_ondisk_t
433 get_node(hammer_off_t node_offset, struct buffer_info **bufferp)
434 {
435         if (*bufferp != NULL) {
436                 rel_buffer(*bufferp);
437                 *bufferp = NULL;
438         }
439         return(get_ondisk(node_offset, bufferp, 0));
440 }
441
442 /*
443  * Return a pointer to a buffer data given a buffer offset.
444  * If *bufferp is NULL acquire the buffer otherwise use that buffer.
445  */
446 static void *
447 get_ondisk(hammer_off_t buf_offset, struct buffer_info **bufferp, int isnew)
448 {
449         struct buffer_info *buffer;
450
451         buffer = *bufferp;
452         if (buffer == NULL) {
453                 buffer = *bufferp = get_buffer(buf_offset, isnew);
454                 if (buffer == NULL)
455                         return(NULL);
456         }
457
458         return((char *)buffer->ondisk +
459                 ((int32_t)buf_offset & HAMMER_BUFMASK));
460 }
461
462 /*
463  * Allocate HAMMER elements - B-Tree nodse
464  */
465 void *
466 alloc_btree_element(hammer_off_t *offp, struct buffer_info **data_bufferp)
467 {
468         hammer_node_ondisk_t node;
469
470         node = alloc_blockmap(HAMMER_ZONE_BTREE_INDEX, sizeof(*node),
471                               offp, data_bufferp);
472         bzero(node, sizeof(*node));
473         return (node);
474 }
475
476 /*
477  * Allocate HAMMER elements - meta data (inode, direntry, PFS, etc)
478  */
479 void *
480 alloc_meta_element(hammer_off_t *offp, int32_t data_len,
481                    struct buffer_info **data_bufferp)
482 {
483         void *data;
484
485         data = alloc_blockmap(HAMMER_ZONE_META_INDEX, data_len,
486                               offp, data_bufferp);
487         bzero(data, data_len);
488         return (data);
489 }
490
491 /*
492  * Allocate HAMMER elements - data storage
493  *
494  * The only data_len supported by HAMMER userspace for large data zone
495  * (zone 10) is HAMMER_BUFSIZE which is 16KB.  >16KB data does not fit
496  * in a buffer allocated by get_buffer().  Also alloc_blockmap() does
497  * not consider >16KB buffer size.
498  */
499 void *
500 alloc_data_element(hammer_off_t *offp, int32_t data_len,
501                    struct buffer_info **data_bufferp)
502 {
503         void *data;
504         int zone;
505
506         if (data_len == 0)
507                 return(NULL);
508
509         zone = hammer_data_zone_index(data_len);
510         assert(data_len <= HAMMER_BUFSIZE); /* just one buffer */
511         assert(zone == HAMMER_ZONE_LARGE_DATA_INDEX ||
512                zone == HAMMER_ZONE_SMALL_DATA_INDEX);
513
514         data = alloc_blockmap(zone, data_len, offp, data_bufferp);
515         bzero(data, data_len);
516         return(data);
517 }
518
519 /*
520  * Format a new blockmap.  This is mostly a degenerate case because
521  * all allocations are now actually done from the freemap.
522  */
523 void
524 format_blockmap(struct volume_info *root_vol, int zone, hammer_off_t offset)
525 {
526         hammer_blockmap_t blockmap;
527         hammer_off_t zone_base;
528
529         /* Only root volume needs formatting */
530         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
531
532         assert(hammer_is_zone2_mapped_index(zone));
533
534         blockmap = &root_vol->ondisk->vol0_blockmap[zone];
535         zone_base = HAMMER_ZONE_ENCODE(zone, offset);
536
537         bzero(blockmap, sizeof(*blockmap));
538         blockmap->phys_offset = 0;
539         blockmap->first_offset = zone_base;
540         blockmap->next_offset = zone_base;
541         blockmap->alloc_offset = HAMMER_ENCODE(zone, 255, -1);
542         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
543 }
544
545 /*
546  * Format a new freemap.  Set all layer1 entries to UNAVAIL.  The initialize
547  * code will load each volume's freemap.
548  */
549 void
550 format_freemap(struct volume_info *root_vol)
551 {
552         struct buffer_info *buffer = NULL;
553         hammer_off_t layer1_offset;
554         hammer_blockmap_t blockmap;
555         struct hammer_blockmap_layer1 *layer1;
556         int i, isnew;
557
558         /* Only root volume needs formatting */
559         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
560
561         layer1_offset = alloc_bigblock(root_vol, HAMMER_ZONE_FREEMAP_INDEX);
562         for (i = 0; i < HAMMER_BIGBLOCK_SIZE; i += sizeof(*layer1)) {
563                 isnew = ((i % HAMMER_BUFSIZE) == 0);
564                 layer1 = get_buffer_data(layer1_offset + i, &buffer, isnew);
565                 bzero(layer1, sizeof(*layer1));
566                 layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
567                 layer1->blocks_free = 0;
568                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
569         }
570         assert(i == HAMMER_BIGBLOCK_SIZE);
571         rel_buffer(buffer);
572
573         blockmap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
574         bzero(blockmap, sizeof(*blockmap));
575         blockmap->phys_offset = layer1_offset;
576         blockmap->first_offset = 0;
577         blockmap->next_offset = HAMMER_ENCODE_RAW_BUFFER(0, 0);
578         blockmap->alloc_offset = HAMMER_ENCODE_RAW_BUFFER(255, -1);
579         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
580 }
581
582 /*
583  * Load the volume's remaining free space into the freemap.
584  *
585  * Returns the number of big-blocks available.
586  */
587 int64_t
588 initialize_freemap(struct volume_info *vol)
589 {
590         struct volume_info *root_vol;
591         struct buffer_info *buffer1 = NULL;
592         struct buffer_info *buffer2 = NULL;
593         struct hammer_blockmap_layer1 *layer1;
594         struct hammer_blockmap_layer2 *layer2;
595         hammer_off_t layer1_base;
596         hammer_off_t layer1_offset;
597         hammer_off_t layer2_offset;
598         hammer_off_t phys_offset;
599         hammer_off_t block_offset;
600         hammer_off_t aligned_vol_free_end;
601         hammer_blockmap_t freemap;
602         int64_t count = 0;
603         int64_t layer1_count = 0;
604
605         root_vol = get_root_volume();
606         aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK)
607                                 & ~HAMMER_BLOCKMAP_LAYER2_MASK;
608
609         printf("initialize freemap volume %d\n", vol->vol_no);
610
611         /*
612          * Initialize the freemap.  First preallocate the big-blocks required
613          * to implement layer2.   This preallocation is a bootstrap allocation
614          * using blocks from the target volume.
615          */
616         freemap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
617         layer1_base = freemap->phys_offset;
618
619         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
620              phys_offset < aligned_vol_free_end;
621              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
622                 layer1_offset = layer1_base +
623                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
624                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
625                 if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
626                         layer1->phys_offset = alloc_bigblock(vol,
627                                                 HAMMER_ZONE_FREEMAP_INDEX);
628                         layer1->blocks_free = 0;
629                         buffer1->cache.modified = 1;
630                         layer1->layer1_crc = crc32(layer1,
631                                                    HAMMER_LAYER1_CRCSIZE);
632                 }
633         }
634
635         /*
636          * Now fill everything in.
637          */
638         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
639              phys_offset < aligned_vol_free_end;
640              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
641                 layer1_count = 0;
642                 layer1_offset = layer1_base +
643                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
644                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
645                 assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
646
647                 for (block_offset = 0;
648                      block_offset < HAMMER_BLOCKMAP_LAYER2;
649                      block_offset += HAMMER_BIGBLOCK_SIZE) {
650                         layer2_offset = layer1->phys_offset +
651                                         HAMMER_BLOCKMAP_LAYER2_OFFSET(block_offset);
652                         layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
653                         bzero(layer2, sizeof(*layer2));
654
655                         if (phys_offset + block_offset < vol->vol_free_off) {
656                                 /*
657                                  * Fixups XXX - big-blocks already allocated as part
658                                  * of the freemap bootstrap.
659                                  */
660                                 layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
661                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
662                                 layer2->bytes_free = 0;
663                         } else if (phys_offset + block_offset < vol->vol_free_end) {
664                                 layer2->zone = 0;
665                                 layer2->append_off = 0;
666                                 layer2->bytes_free = HAMMER_BIGBLOCK_SIZE;
667                                 ++count;
668                                 ++layer1_count;
669                         } else {
670                                 layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
671                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
672                                 layer2->bytes_free = 0;
673                         }
674                         layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
675                         buffer2->cache.modified = 1;
676                 }
677
678                 layer1->blocks_free += layer1_count;
679                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
680                 buffer1->cache.modified = 1;
681         }
682
683         rel_buffer(buffer1);
684         rel_buffer(buffer2);
685         rel_volume(root_vol);
686         return(count);
687 }
688
689 /*
690  * Returns the number of big-blocks available for filesystem data and undos
691  * without formatting.
692  */
693 int64_t
694 count_freemap(struct volume_info *vol)
695 {
696         hammer_off_t phys_offset;
697         hammer_off_t vol_free_off;
698         hammer_off_t aligned_vol_free_end;
699         int64_t count = 0;
700
701         vol_free_off = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
702         aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK)
703                                 & ~HAMMER_BLOCKMAP_LAYER2_MASK;
704
705         if (vol->vol_no == HAMMER_ROOT_VOLNO)
706                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
707
708         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
709              phys_offset < aligned_vol_free_end;
710              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
711                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
712         }
713
714         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
715              phys_offset < aligned_vol_free_end;
716              phys_offset += HAMMER_BIGBLOCK_SIZE) {
717                 if (phys_offset < vol_free_off) {
718                         ;
719                 } else if (phys_offset < vol->vol_free_end) {
720                         ++count;
721                 }
722         }
723
724         return(count);
725 }
726
727 /*
728  * Format the undomap for the root volume.
729  */
730 void
731 format_undomap(struct volume_info *root_vol, int64_t *undo_buffer_size)
732 {
733         const int undo_zone = HAMMER_ZONE_UNDO_INDEX;
734         hammer_off_t undo_limit;
735         hammer_blockmap_t blockmap;
736         struct hammer_volume_ondisk *ondisk;
737         struct buffer_info *buffer = NULL;
738         hammer_off_t scan;
739         int n;
740         int limit_index;
741         uint32_t seqno;
742
743         /* Only root volume needs formatting */
744         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
745         ondisk = root_vol->ondisk;
746
747         /*
748          * Size the undo buffer in multiples of HAMMER_BIGBLOCK_SIZE,
749          * up to HAMMER_UNDO_LAYER2 big-blocks.  Size to approximately
750          * 0.1% of the disk.
751          *
752          * The minimum UNDO fifo size is 500MB, or approximately 1% of
753          * the recommended 50G disk.
754          *
755          * Changing this minimum is rather dangerous as complex filesystem
756          * operations can cause the UNDO FIFO to fill up otherwise.
757          */
758         undo_limit = *undo_buffer_size;
759         if (undo_limit == 0) {
760                 undo_limit = (ondisk->vol_buf_end - ondisk->vol_buf_beg) / 1000;
761                 if (undo_limit < 500*1024*1024)
762                         undo_limit = 500*1024*1024;
763         }
764         undo_limit = (undo_limit + HAMMER_BIGBLOCK_MASK64) &
765                      ~HAMMER_BIGBLOCK_MASK64;
766         if (undo_limit < HAMMER_BIGBLOCK_SIZE)
767                 undo_limit = HAMMER_BIGBLOCK_SIZE;
768         if (undo_limit > HAMMER_BIGBLOCK_SIZE * HAMMER_UNDO_LAYER2)
769                 undo_limit = HAMMER_BIGBLOCK_SIZE * HAMMER_UNDO_LAYER2;
770         *undo_buffer_size = undo_limit;
771
772         blockmap = &ondisk->vol0_blockmap[undo_zone];
773         bzero(blockmap, sizeof(*blockmap));
774         blockmap->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
775         blockmap->first_offset = HAMMER_ZONE_ENCODE(undo_zone, 0);
776         blockmap->next_offset = blockmap->first_offset;
777         blockmap->alloc_offset = HAMMER_ZONE_ENCODE(undo_zone, undo_limit);
778         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
779
780         limit_index = undo_limit / HAMMER_BIGBLOCK_SIZE;
781         assert(limit_index <= HAMMER_UNDO_LAYER2);
782
783         for (n = 0; n < limit_index; ++n) {
784                 ondisk->vol0_undo_array[n] = alloc_bigblock(root_vol,
785                                                         HAMMER_ZONE_UNDO_INDEX);
786         }
787         while (n < HAMMER_UNDO_LAYER2) {
788                 ondisk->vol0_undo_array[n++] = HAMMER_BLOCKMAP_UNAVAIL;
789         }
790
791         /*
792          * Pre-initialize the UNDO blocks (HAMMER version 4+)
793          */
794         printf("initializing the undo map (%jd MB)\n",
795                 (intmax_t)(blockmap->alloc_offset & HAMMER_OFF_LONG_MASK) /
796                 (1024 * 1024));
797
798         scan = blockmap->first_offset;
799         seqno = 0;
800
801         while (scan < blockmap->alloc_offset) {
802                 hammer_fifo_head_t head;
803                 hammer_fifo_tail_t tail;
804                 int isnew;
805                 int bytes = HAMMER_UNDO_ALIGN;
806
807                 isnew = ((scan & HAMMER_BUFMASK64) == 0);
808                 head = get_buffer_data(scan, &buffer, isnew);
809                 buffer->cache.modified = 1;
810                 tail = (void *)((char *)head + bytes - sizeof(*tail));
811
812                 bzero(head, bytes);
813                 head->hdr_signature = HAMMER_HEAD_SIGNATURE;
814                 head->hdr_type = HAMMER_HEAD_TYPE_DUMMY;
815                 head->hdr_size = bytes;
816                 head->hdr_seq = seqno++;
817
818                 tail->tail_signature = HAMMER_TAIL_SIGNATURE;
819                 tail->tail_type = HAMMER_HEAD_TYPE_DUMMY;
820                 tail->tail_size = bytes;
821
822                 head->hdr_crc = crc32(head, HAMMER_FIFO_HEAD_CRCOFF) ^
823                                 crc32(head + 1, bytes - sizeof(*head));
824
825                 scan += bytes;
826         }
827         rel_buffer(buffer);
828 }
829
830 /*
831  * Flush various tracking structures to disk
832  */
833 void
834 flush_all_volumes(void)
835 {
836         struct volume_info *vol;
837
838         TAILQ_FOREACH(vol, &VolList, entry)
839                 flush_volume(vol);
840 }
841
842 void
843 flush_volume(struct volume_info *volume)
844 {
845         struct buffer_info *buffer;
846         int i;
847
848         for (i = 0; i < HAMMER_BUFLISTS; ++i) {
849                 TAILQ_FOREACH(buffer, &volume->buffer_lists[i], entry)
850                         flush_buffer(buffer);
851         }
852         if (writehammerbuf(volume, volume->ondisk, 0) == -1)
853                 err(1, "Write volume %d (%s)", volume->vol_no, volume->name);
854 }
855
856 void
857 flush_buffer(struct buffer_info *buffer)
858 {
859         struct volume_info *vol;
860
861         vol = buffer->volume;
862         if (writehammerbuf(vol, buffer->ondisk, buffer->raw_offset) == -1)
863                 err(1, "Write volume %d (%s)", vol->vol_no, vol->name);
864         buffer->cache.modified = 0;
865 }
866
867 /*
868  * Core I/O operations
869  */
870 static int
871 readhammerbuf(struct volume_info *vol, void *data, int64_t offset)
872 {
873         ssize_t n;
874
875         n = pread(vol->fd, data, HAMMER_BUFSIZE, offset);
876         if (n != HAMMER_BUFSIZE)
877                 return(-1);
878         return(0);
879 }
880
881 static int
882 writehammerbuf(struct volume_info *vol, const void *data, int64_t offset)
883 {
884         ssize_t n;
885
886         if (vol->rdonly)
887                 return(0);
888
889         n = pwrite(vol->fd, data, HAMMER_BUFSIZE, offset);
890         if (n != HAMMER_BUFSIZE)
891                 return(-1);
892         return(0);
893 }
894
895 int64_t init_boot_area_size(int64_t value, off_t avg_vol_size)
896 {
897         if (value == 0) {
898                 value = HAMMER_BOOT_NOMBYTES;
899                 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
900                         value >>= 1;
901                 if (value < HAMMER_BOOT_MINBYTES)
902                         value = 0;
903         } else if (value < HAMMER_BOOT_MINBYTES) {
904                 value = HAMMER_BOOT_MINBYTES;
905         }
906
907         return(value);
908 }
909
910 int64_t init_mem_area_size(int64_t value, off_t avg_vol_size)
911 {
912         if (value == 0) {
913                 value = HAMMER_MEM_NOMBYTES;
914                 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
915                         value >>= 1;
916                 if (value < HAMMER_MEM_MINBYTES)
917                         value = 0;
918         } else if (value < HAMMER_MEM_MINBYTES) {
919                 value = HAMMER_MEM_MINBYTES;
920         }
921
922         return(value);
923 }