?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ispfw.4: Remove extra .Pp
[dragonfly.git] / sbin / hammer / ondisk.c
bloba35613b5c21c4a3f2a0a2b10dd5be7543dd2595b
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(volume_info_t volume);
41 static void get_buffer_readahead(buffer_info_t base);
42 static __inline int readhammervol(volume_info_t volume);
43 static __inline int readhammerbuf(buffer_info_t buffer);
44 static __inline int writehammervol(volume_info_t volume);
45 static __inline int writehammerbuf(buffer_info_t buffer);
46
47 uuid_t Hammer_FSType;
48 uuid_t Hammer_FSId;
49 int UseReadBehind = -4;
50 int UseReadAhead = 4;
51 int DebugOpt;
52 uint32_t HammerVersion = -1;
53
54 TAILQ_HEAD(volume_list, volume_info);
55 static 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 zone2_offset)
61 {
62 int hi;
63
64 hi = (int)(zone2_offset / HAMMER_BUFSIZE) & HAMMER_BUFLISTMASK;
65 return(hi);
66 }
67
68 static
69 buffer_info_t
70 find_buffer(hammer_off_t zone2_offset)
71 {
72 volume_info_t volume;
73 buffer_info_t buffer;
74 int hi;
75
76 volume = get_volume(HAMMER_VOL_DECODE(zone2_offset));
77 assert(volume);
78
79 hi = buffer_hash(zone2_offset);
80 TAILQ_FOREACH(buffer, &volume->buffer_lists[hi], entry) {
81 if (buffer->zone2_offset == zone2_offset)
82 return(buffer);
83 }
84 return(NULL);
85 }
86
87 static
88 volume_info_t
89 __alloc_volume(const char *volname, int oflags)
90 {
91 volume_info_t volume;
92 int i;
93
94 volume = calloc(1, sizeof(*volume));
95 volume->vol_no = -1;
96 volume->rdonly = (oflags == O_RDONLY);
97 volume->name = strdup(volname);
98 volume->fd = open(volume->name, oflags);
99 if (volume->fd < 0) {
100 err(1, "alloc_volume: Failed to open %s", volume->name);
101 /* not reached */
102 }
103 check_volume(volume);
104
105 volume->ondisk = calloc(1, HAMMER_BUFSIZE);
106
107 for (i = 0; i < HAMMER_BUFLISTS; ++i)
108 TAILQ_INIT(&volume->buffer_lists[i]);
109
110 return(volume);
111 }
112
113 static
114 void
115 __add_volume(const volume_info_t volume)
116 {
117 volume_info_t scan;
118 struct stat st1, st2;
119
120 if (fstat(volume->fd, &st1) != 0) {
121 errx(1, "add_volume: %s: Failed to stat", volume->name);
122 /* not reached */
123 }
124
125 TAILQ_FOREACH(scan, &VolList, entry) {
126 if (scan->vol_no == volume->vol_no) {
127 errx(1, "add_volume: %s: Duplicate volume number %d "
128 "against %s",
129 volume->name, volume->vol_no, scan->name);
130 /* not reached */
131 }
132 if (fstat(scan->fd, &st2) != 0) {
133 errx(1, "add_volume: %s: Failed to stat %s",
134 volume->name, scan->name);
135 /* not reached */
136 }
137 if ((st1.st_ino == st2.st_ino) && (st1.st_dev == st2.st_dev)) {
138 errx(1, "add_volume: %s: Specified more than once",
139 volume->name);
140 /* not reached */
141 }
142 }
143
144 TAILQ_INSERT_TAIL(&VolList, volume, entry);
145 }
146
147 static
148 void
149 __verify_volume(const volume_info_t volume)
150 {
151 hammer_volume_ondisk_t ondisk = volume->ondisk;
152
153 if (ondisk->vol_signature != HAMMER_FSBUF_VOLUME) {
154 errx(1, "verify_volume: Invalid volume signature %016jx",
155 ondisk->vol_signature);
156 /* not reached */
157 }
158 if (ondisk->vol_rootvol != HAMMER_ROOT_VOLNO) {
159 errx(1, "verify_volume: Invalid root volume# %d",
160 ondisk->vol_rootvol);
161 /* not reached */
162 }
163 if (bcmp(&Hammer_FSType, &ondisk->vol_fstype, sizeof(Hammer_FSType))) {
164 errx(1, "verify_volume: %s: Header does not indicate "
165 "that this is a HAMMER volume", volume->name);
166 /* not reached */
167 }
168 if (bcmp(&Hammer_FSId, &ondisk->vol_fsid, sizeof(Hammer_FSId))) {
169 errx(1, "verify_volume: %s: FSId does not match other volumes!",
170 volume->name);
171 /* not reached */
172 }
173 if (ondisk->vol_version < HAMMER_VOL_VERSION_MIN ||
174 ondisk->vol_version >= HAMMER_VOL_VERSION_WIP) {
175 errx(1, "verify_volume: %s: Invalid volume version %u",
176 volume->name, ondisk->vol_version);
177 /* not reached */
178 }
179 }
180
181 /*
182 * Initialize a volume structure and ondisk vol_no field.
183 */
184 volume_info_t
185 init_volume(const char *filename, int oflags, int32_t vol_no)
186 {
187 volume_info_t volume;
188
189 volume = __alloc_volume(filename, oflags);
190 volume->vol_no = volume->ondisk->vol_no = vol_no;
191
192 __add_volume(volume);
193
194 return(volume);
195 }
196
197 /*
198 * Initialize a volume structure and read ondisk volume header.
199 */
200 volume_info_t
201 load_volume(const char *filename, int oflags, int verify_volume)
202 {
203 volume_info_t volume;
204 int n;
205
206 volume = __alloc_volume(filename, oflags);
207
208 n = readhammervol(volume);
209 if (n == -1) {
210 err(1, "load_volume: %s: Read failed at offset 0",
211 volume->name);
212 /* not reached */
213 }
214 volume->vol_no = volume->ondisk->vol_no;
215 if (volume->vol_no == HAMMER_ROOT_VOLNO)
216 HammerVersion = volume->ondisk->vol_version;
217
218 if (valid_hammer_volumes++ == 0)
219 Hammer_FSId = volume->ondisk->vol_fsid;
220 if (verify_volume)
221 __verify_volume(volume);
222
223 __add_volume(volume);
224
225 return(volume);
226 }
227
228 /*
229 * Check basic volume characteristics.
230 */
231 static
232 void
233 check_volume(volume_info_t volume)
234 {
235 struct partinfo pinfo;
236 struct stat st;
237
238 /*
239 * Allow the formatting of block devices or regular files
240 */
241 if (ioctl(volume->fd, DIOCGPART, &pinfo) < 0) {
242 if (fstat(volume->fd, &st) < 0) {
243 err(1, "Unable to stat %s", volume->name);
244 /* not reached */
245 }
246 if (S_ISREG(st.st_mode)) {
247 volume->size = st.st_size;
248 volume->type = "REGFILE";
249 } else {
250 errx(1, "Unsupported file type for %s", volume->name);
251 /* not reached */
252 }
253 } else {
254 /*
255 * When formatting a block device as a HAMMER volume the
256 * sector size must be compatible. HAMMER uses 16384 byte
257 * filesystem buffers.
258 */
259 if (pinfo.reserved_blocks) {
260 errx(1, "HAMMER cannot be placed in a partition "
261 "which overlaps the disklabel or MBR");
262 /* not reached */
263 }
264 if (pinfo.media_blksize > HAMMER_BUFSIZE ||
265 HAMMER_BUFSIZE % pinfo.media_blksize) {
266 errx(1, "A media sector size of %d is not supported",
267 pinfo.media_blksize);
268 /* not reached */
269 }
270
271 volume->size = pinfo.media_size;
272 volume->device_offset = pinfo.media_offset;
273 volume->type = "DEVICE";
274 }
275 }
276
277 int
278 is_regfile(const volume_info_t volume)
279 {
280 return(strcmp(volume->type, "REGFILE") ? 0 : 1);
281 }
282
283 void
284 assert_volume_offset(const volume_info_t volume)
285 {
286 assert(hammer_is_zone_raw_buffer(volume->vol_free_off));
287 assert(hammer_is_zone_raw_buffer(volume->vol_free_end));
288 if (volume->vol_free_off >= volume->vol_free_end) {
289 errx(1, "Ran out of room, filesystem too small");
290 /* not reached */
291 }
292 }
293
294 volume_info_t
295 get_volume(int32_t vol_no)
296 {
297 volume_info_t volume;
298
299 TAILQ_FOREACH(volume, &VolList, entry) {
300 if (volume->vol_no == vol_no)
301 break;
302 }
303
304 return(volume);
305 }
306
307 volume_info_t
308 get_root_volume(void)
309 {
310 return(get_volume(HAMMER_ROOT_VOLNO));
311 }
312
313 static
314 hammer_off_t
315 __blockmap_xlate_to_zone2(hammer_off_t buf_offset)
316 {
317 hammer_off_t zone2_offset;
318 int error = 0;
319
320 if (hammer_is_zone_raw_buffer(buf_offset))
321 zone2_offset = buf_offset;
322 else
323 zone2_offset = blockmap_lookup(buf_offset, &error);
324
325 if (error)
326 return(HAMMER_OFF_BAD);
327 assert(hammer_is_zone_raw_buffer(zone2_offset));
328
329 return(zone2_offset);
330 }
331
332 static
333 buffer_info_t
334 __alloc_buffer(hammer_off_t zone2_offset, int isnew)
335 {
336 volume_info_t volume;
337 buffer_info_t buffer;
338 int hi;
339
340 volume = get_volume(HAMMER_VOL_DECODE(zone2_offset));
341 assert(volume != NULL);
342
343 buffer = calloc(1, sizeof(*buffer));
344 buffer->zone2_offset = zone2_offset;
345 buffer->raw_offset = hammer_xlate_to_phys(volume->ondisk, zone2_offset);
346 buffer->volume = volume;
347 buffer->ondisk = calloc(1, HAMMER_BUFSIZE);
348
349 if (isnew <= 0) {
350 if (readhammerbuf(buffer) == -1) {
351 err(1, "Failed to read %s:%016jx at %016jx",
352 volume->name,
353 (intmax_t)buffer->zone2_offset,
354 (intmax_t)buffer->raw_offset);
355 /* not reached */
356 }
357 }
358
359 hi = buffer_hash(zone2_offset);
360 TAILQ_INSERT_TAIL(&volume->buffer_lists[hi], buffer, entry);
361 hammer_cache_add(&buffer->cache);
362
363 return(buffer);
364 }
365
366 /*
367 * Acquire the 16KB buffer for specified zone offset.
368 */
369 static
370 buffer_info_t
371 get_buffer(hammer_off_t buf_offset, int isnew)
372 {
373 buffer_info_t buffer;
374 hammer_off_t zone2_offset;
375 int dora = 0;
376
377 zone2_offset = __blockmap_xlate_to_zone2(buf_offset);
378 if (zone2_offset == HAMMER_OFF_BAD)
379 return(NULL);
380
381 zone2_offset &= ~HAMMER_BUFMASK64;
382 buffer = find_buffer(zone2_offset);
383
384 if (buffer == NULL) {
385 buffer = __alloc_buffer(zone2_offset, isnew);
386 dora = (isnew == 0);
387 } else {
388 assert(isnew != -1);
389 hammer_cache_used(&buffer->cache);
390 }
391 assert(buffer->ondisk != NULL);
392
393 ++buffer->cache.refs;
394 hammer_cache_flush();
395
396 if (isnew > 0) {
397 assert(buffer->cache.modified == 0);
398 bzero(buffer->ondisk, HAMMER_BUFSIZE);
399 buffer->cache.modified = 1;
400 }
401 if (dora)
402 get_buffer_readahead(buffer);
403 return(buffer);
404 }
405
406 static
407 void
408 get_buffer_readahead(const buffer_info_t base)
409 {
410 buffer_info_t buffer;
411 volume_info_t volume;
412 hammer_off_t zone2_offset;
413 int64_t raw_offset;
414 int ri = UseReadBehind;
415 int re = UseReadAhead;
416
417 raw_offset = base->raw_offset + ri * HAMMER_BUFSIZE;
418 volume = base->volume;
419
420 while (ri < re) {
421 if (raw_offset >= volume->ondisk->vol_buf_end)
422 break;
423 if (raw_offset < volume->ondisk->vol_buf_beg || ri == 0) {
424 ++ri;
425 raw_offset += HAMMER_BUFSIZE;
426 continue;
427 }
428 zone2_offset = HAMMER_ENCODE_RAW_BUFFER(volume->vol_no,
429 raw_offset - volume->ondisk->vol_buf_beg);
430 buffer = find_buffer(zone2_offset);
431 if (buffer == NULL) {
432 /* call with -1 to prevent another readahead */
433 buffer = get_buffer(zone2_offset, -1);
434 rel_buffer(buffer);
435 }
436 ++ri;
437 raw_offset += HAMMER_BUFSIZE;
438 }
439 }
440
441 void
442 rel_buffer(buffer_info_t buffer)
443 {
444 volume_info_t volume;
445 int hi;
446
447 if (buffer == NULL)
448 return;
449 assert(buffer->cache.refs > 0);
450 if (--buffer->cache.refs == 0) {
451 if (buffer->cache.delete) {
452 hi = buffer_hash(buffer->zone2_offset);
453 volume = buffer->volume;
454 if (buffer->cache.modified)
455 flush_buffer(buffer);
456 TAILQ_REMOVE(&volume->buffer_lists[hi], buffer, entry);
457 hammer_cache_del(&buffer->cache);
458 free(buffer->ondisk);
459 free(buffer);
460 }
461 }
462 }
463
464 /*
465 * Retrieve a pointer to a buffer data given a buffer offset. The underlying
466 * bufferp is freed if isnew or the offset is out of range of the cached data.
467 * If bufferp is freed a referenced buffer is loaded into it.
468 */
469 void *
470 get_buffer_data(hammer_off_t buf_offset, buffer_info_t *bufferp, int isnew)
471 {
472 hammer_off_t xor;
473
474 if (*bufferp != NULL) {
475 /* XXX xor is always non zero for indirect zones */
476 xor = HAMMER_OFF_LONG_ENCODE(buf_offset) ^
477 HAMMER_OFF_LONG_ENCODE((*bufferp)->zone2_offset);
478 if (isnew > 0 || (xor & ~HAMMER_BUFMASK64)) {
479 rel_buffer(*bufferp);
480 *bufferp = NULL;
481 }
482 }
483
484 if (*bufferp == NULL) {
485 *bufferp = get_buffer(buf_offset, isnew);
486 if (*bufferp == NULL)
487 return(NULL);
488 }
489
490 return(((char *)(*bufferp)->ondisk) +
491 ((int32_t)buf_offset & HAMMER_BUFMASK));
492 }
493
494 /*
495 * Allocate HAMMER elements - B-Tree nodes
496 */
497 hammer_node_ondisk_t
498 alloc_btree_node(hammer_off_t *offp, buffer_info_t *data_bufferp)
499 {
500 hammer_node_ondisk_t node;
501
502 node = alloc_blockmap(HAMMER_ZONE_BTREE_INDEX, sizeof(*node),
503 offp, data_bufferp);
504 bzero(node, sizeof(*node));
505 return(node);
506 }
507
508 /*
509 * Allocate HAMMER elements - meta data (inode, direntry, PFS, etc)
510 */
511 void *
512 alloc_meta_element(hammer_off_t *offp, int32_t data_len,
513 buffer_info_t *data_bufferp)
514 {
515 void *data;
516
517 data = alloc_blockmap(HAMMER_ZONE_META_INDEX, data_len,
518 offp, data_bufferp);
519 bzero(data, data_len);
520 return(data);
521 }
522
523 /*
524 * Format a new blockmap. This is mostly a degenerate case because
525 * all allocations are now actually done from the freemap.
526 */
527 void
528 format_blockmap(volume_info_t root_vol, int zone, hammer_off_t offset)
529 {
530 hammer_blockmap_t blockmap;
531 hammer_off_t zone_base;
532
533 /* Only root volume needs formatting */
534 assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
535
536 assert(hammer_is_index_record(zone));
537
538 blockmap = &root_vol->ondisk->vol0_blockmap[zone];
539 zone_base = HAMMER_ZONE_ENCODE(zone, offset);
540
541 bzero(blockmap, sizeof(*blockmap));
542 blockmap->phys_offset = 0;
543 blockmap->first_offset = zone_base;
544 blockmap->next_offset = zone_base;
545 blockmap->alloc_offset = HAMMER_ENCODE(zone, 255, -1);
546 hammer_crc_set_blockmap(HammerVersion, blockmap);
547 }
548
549 /*
550 * Format a new freemap. Set all layer1 entries to UNAVAIL. The initialize
551 * code will load each volume's freemap.
552 */
553 void
554 format_freemap(volume_info_t root_vol)
555 {
556 buffer_info_t buffer = NULL;
557 hammer_off_t layer1_offset;
558 hammer_blockmap_t blockmap;
559 hammer_blockmap_layer1_t layer1;
560 int i, isnew;
561
562 /* Only root volume needs formatting */
563 assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
564
565 layer1_offset = bootstrap_bigblock(root_vol);
566 for (i = 0; i < HAMMER_BIGBLOCK_SIZE; i += sizeof(*layer1)) {
567 isnew = ((i % HAMMER_BUFSIZE) == 0);
568 layer1 = get_buffer_data(layer1_offset + i, &buffer, isnew);
569 bzero(layer1, sizeof(*layer1));
570 layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
571 layer1->blocks_free = 0;
572 hammer_crc_set_layer1(HammerVersion, layer1);
573 }
574 assert(i == HAMMER_BIGBLOCK_SIZE);
575 rel_buffer(buffer);
576
577 blockmap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
578 bzero(blockmap, sizeof(*blockmap));
579 blockmap->phys_offset = layer1_offset;
580 blockmap->first_offset = 0;
581 blockmap->next_offset = HAMMER_ENCODE_RAW_BUFFER(0, 0);
582 blockmap->alloc_offset = HAMMER_ENCODE_RAW_BUFFER(255, -1);
583 hammer_crc_set_blockmap(HammerVersion, blockmap);
584 }
585
586 /*
587 * Load the volume's remaining free space into the freemap.
588 *
589 * Returns the number of big-blocks available.
590 */
591 int64_t
592 initialize_freemap(volume_info_t volume)
593 {
594 volume_info_t root_vol;
595 buffer_info_t buffer1 = NULL;
596 buffer_info_t buffer2 = NULL;
597 hammer_blockmap_layer1_t layer1;
598 hammer_blockmap_layer2_t layer2;
599 hammer_off_t layer1_offset;
600 hammer_off_t layer2_offset;
601 hammer_off_t phys_offset;
602 hammer_off_t block_offset;
603 hammer_off_t aligned_vol_free_end;
604 hammer_blockmap_t freemap;
605 int64_t count = 0;
606 int64_t layer1_count = 0;
607
608 root_vol = get_root_volume();
609
610 assert_volume_offset(volume);
611 aligned_vol_free_end = HAMMER_BLOCKMAP_LAYER2_DOALIGN(volume->vol_free_end);
612
613 printf("initialize freemap volume %d\n", volume->vol_no);
614
615 /*
616 * Initialize the freemap. First preallocate the big-blocks required
617 * to implement layer2. This preallocation is a bootstrap allocation
618 * using blocks from the target volume.
619 */
620 freemap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
621
622 for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(volume->vol_no, 0);
623 phys_offset < aligned_vol_free_end;
624 phys_offset += HAMMER_BLOCKMAP_LAYER2) {
625 layer1_offset = freemap->phys_offset +
626 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
627 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
628 if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
629 layer1->phys_offset = bootstrap_bigblock(volume);
630 layer1->blocks_free = 0;
631 buffer1->cache.modified = 1;
632 hammer_crc_set_layer1(HammerVersion, layer1);
633 }
634 }
635
636 /*
637 * Now fill everything in.
638 */
639 for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(volume->vol_no, 0);
640 phys_offset < aligned_vol_free_end;
641 phys_offset += HAMMER_BLOCKMAP_LAYER2) {
642 layer1_count = 0;
643 layer1_offset = freemap->phys_offset +
644 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
645 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
646 assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
647
648 for (block_offset = 0;
649 block_offset < HAMMER_BLOCKMAP_LAYER2;
650 block_offset += HAMMER_BIGBLOCK_SIZE) {
651 layer2_offset = layer1->phys_offset +
652 HAMMER_BLOCKMAP_LAYER2_OFFSET(block_offset);
653 layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
654 bzero(layer2, sizeof(*layer2));
655
656 if (phys_offset + block_offset < volume->vol_free_off) {
657 /*
658 * Big-blocks already allocated as part
659 * of the freemap bootstrap.
660 */
661 layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
662 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
663 layer2->bytes_free = 0;
664 } else if (phys_offset + block_offset < volume->vol_free_end) {
665 layer2->zone = 0;
666 layer2->append_off = 0;
667 layer2->bytes_free = HAMMER_BIGBLOCK_SIZE;
668 ++count;
669 ++layer1_count;
670 } else {
671 layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
672 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
673 layer2->bytes_free = 0;
674 }
675 hammer_crc_set_layer2(HammerVersion, layer2);
676 buffer2->cache.modified = 1;
677 }
678
679 layer1->blocks_free += layer1_count;
680 hammer_crc_set_layer1(HammerVersion, layer1);
681 buffer1->cache.modified = 1;
682 }
683
684 rel_buffer(buffer1);
685 rel_buffer(buffer2);
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(const volume_info_t volume)
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(volume->vol_no, 0);
702
703 assert_volume_offset(volume);
704 aligned_vol_free_end = HAMMER_BLOCKMAP_LAYER2_DOALIGN(volume->vol_free_end);
705
706 if (volume->vol_no == HAMMER_ROOT_VOLNO)
707 vol_free_off += HAMMER_BIGBLOCK_SIZE;
708
709 for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(volume->vol_no, 0);
710 phys_offset < aligned_vol_free_end;
711 phys_offset += HAMMER_BLOCKMAP_LAYER2) {
712 vol_free_off += HAMMER_BIGBLOCK_SIZE;
713 }
714
715 for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(volume->vol_no, 0);
716 phys_offset < aligned_vol_free_end;
717 phys_offset += HAMMER_BIGBLOCK_SIZE) {
718 if (phys_offset < vol_free_off)
719 ;
720 else if (phys_offset < volume->vol_free_end)
721 ++count;
722 }
723
724 return(count);
725 }
726
727 /*
728 * Format the undomap for the root volume.
729 */
730 void
731 format_undomap(volume_info_t root_vol, int64_t *undo_buffer_size)
732 {
733 hammer_off_t undo_limit;
734 hammer_blockmap_t blockmap;
735 hammer_volume_ondisk_t ondisk;
736 buffer_info_t buffer = NULL;
737 hammer_off_t scan;
738 int n;
739 int limit_index;
740 uint32_t seqno;
741
742 /* Only root volume needs formatting */
743 assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
744 ondisk = root_vol->ondisk;
745
746 /*
747 * Size the undo buffer in multiples of HAMMER_BIGBLOCK_SIZE,
748 * up to HAMMER_MAX_UNDO_BIGBLOCKS big-blocks.
749 * Size to approximately 0.1% of the disk.
750 *
751 * The minimum UNDO fifo size is 512MB, or approximately 1% of
752 * the recommended 50G disk.
753 *
754 * Changing this minimum is rather dangerous as complex filesystem
755 * operations can cause the UNDO FIFO to fill up otherwise.
756 */
757 undo_limit = *undo_buffer_size;
758 if (undo_limit == 0) {
759 undo_limit = HAMMER_VOL_BUF_SIZE(ondisk) / 1000;
760 if (undo_limit < HAMMER_BIGBLOCK_SIZE * HAMMER_MIN_UNDO_BIGBLOCKS)
761 undo_limit = HAMMER_BIGBLOCK_SIZE * HAMMER_MIN_UNDO_BIGBLOCKS;
762 }
763 undo_limit = HAMMER_BIGBLOCK_DOALIGN(undo_limit);
764 if (undo_limit < HAMMER_BIGBLOCK_SIZE)
765 undo_limit = HAMMER_BIGBLOCK_SIZE;
766 if (undo_limit > HAMMER_BIGBLOCK_SIZE * HAMMER_MAX_UNDO_BIGBLOCKS)
767 undo_limit = HAMMER_BIGBLOCK_SIZE * HAMMER_MAX_UNDO_BIGBLOCKS;
768 *undo_buffer_size = undo_limit;
769
770 blockmap = &ondisk->vol0_blockmap[HAMMER_ZONE_UNDO_INDEX];
771 bzero(blockmap, sizeof(*blockmap));
772 blockmap->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
773 blockmap->first_offset = HAMMER_ENCODE_UNDO(0);
774 blockmap->next_offset = blockmap->first_offset;
775 blockmap->alloc_offset = HAMMER_ENCODE_UNDO(undo_limit);
776 hammer_crc_set_blockmap(HammerVersion, blockmap);
777
778 limit_index = undo_limit / HAMMER_BIGBLOCK_SIZE;
779 assert(limit_index <= HAMMER_MAX_UNDO_BIGBLOCKS);
780
781 for (n = 0; n < limit_index; ++n)
782 ondisk->vol0_undo_array[n] = alloc_undo_bigblock(root_vol);
783 while (n < HAMMER_MAX_UNDO_BIGBLOCKS)
784 ondisk->vol0_undo_array[n++] = HAMMER_BLOCKMAP_UNAVAIL;
785
786 /*
787 * Pre-initialize the UNDO blocks (HAMMER version 4+)
788 */
789 printf("initializing the undo map (%jd MB)\n",
790 (intmax_t)HAMMER_OFF_LONG_ENCODE(blockmap->alloc_offset) /
791 (1024 * 1024));
792
793 scan = blockmap->first_offset;
794 seqno = 0;
795
796 while (scan < blockmap->alloc_offset) {
797 hammer_fifo_head_t head;
798 hammer_fifo_tail_t tail;
799 int bytes = HAMMER_UNDO_ALIGN;
800 int isnew = ((scan & HAMMER_BUFMASK64) == 0);
801
802 head = get_buffer_data(scan, &buffer, isnew);
803 buffer->cache.modified = 1;
804 tail = (void *)((char *)head + bytes - sizeof(*tail));
805
806 bzero(head, bytes);
807 head->hdr_signature = HAMMER_HEAD_SIGNATURE;
808 head->hdr_type = HAMMER_HEAD_TYPE_DUMMY;
809 head->hdr_size = bytes;
810 head->hdr_seq = seqno++;
811
812 tail->tail_signature = HAMMER_TAIL_SIGNATURE;
813 tail->tail_type = HAMMER_HEAD_TYPE_DUMMY;
814 tail->tail_size = bytes;
815
816 hammer_crc_set_fifo_head(HammerVersion, head, bytes);
817
818 scan += bytes;
819 }
820 rel_buffer(buffer);
821 }
822
823 const char *zone_labels[] = {
824 "", /* 0 */
825 "raw_volume", /* 1 */
826 "raw_buffer", /* 2 */
827 "undo", /* 3 */
828 "freemap", /* 4 */
829 "", /* 5 */
830 "", /* 6 */
831 "", /* 7 */
832 "btree", /* 8 */
833 "meta", /* 9 */
834 "large_data", /* 10 */
835 "small_data", /* 11 */
836 "", /* 12 */
837 "", /* 13 */
838 "", /* 14 */
839 "unavail", /* 15 */
840 };
841
842 void
843 print_blockmap(const volume_info_t volume)
844 {
845 hammer_blockmap_t blockmap;
846 hammer_volume_ondisk_t ondisk;
847 int64_t size, used;
848 int i;
849 #define INDENT ""
850
851 ondisk = volume->ondisk;
852 printf(INDENT"vol_label\t%s\n", ondisk->vol_label);
853 printf(INDENT"vol_count\t%d\n", ondisk->vol_count);
854 printf(INDENT"vol_bot_beg\t%s\n", sizetostr(ondisk->vol_bot_beg));
855 printf(INDENT"vol_mem_beg\t%s\n", sizetostr(ondisk->vol_mem_beg));
856 printf(INDENT"vol_buf_beg\t%s\n", sizetostr(ondisk->vol_buf_beg));
857 printf(INDENT"vol_buf_end\t%s\n", sizetostr(ondisk->vol_buf_end));
858 printf(INDENT"vol0_next_tid\t%016jx\n",
859 (uintmax_t)ondisk->vol0_next_tid);
860
861 blockmap = &ondisk->vol0_blockmap[HAMMER_ZONE_UNDO_INDEX];
862 size = HAMMER_OFF_LONG_ENCODE(blockmap->alloc_offset);
863 if (blockmap->first_offset <= blockmap->next_offset)
864 used = blockmap->next_offset - blockmap->first_offset;
865 else
866 used = blockmap->alloc_offset - blockmap->first_offset +
867 HAMMER_OFF_LONG_ENCODE(blockmap->next_offset);
868 printf(INDENT"undo_size\t%s\n", sizetostr(size));
869 printf(INDENT"undo_used\t%s\n", sizetostr(used));
870
871 printf(INDENT"zone # "
872 "phys first next alloc\n");
873 for (i = 0; i < HAMMER_MAX_ZONES; i++) {
874 blockmap = &ondisk->vol0_blockmap[i];
875 printf(INDENT"zone %-2d %-10s %016jx %016jx %016jx %016jx\n",
876 i, zone_labels[i],
877 (uintmax_t)blockmap->phys_offset,
878 (uintmax_t)blockmap->first_offset,
879 (uintmax_t)blockmap->next_offset,
880 (uintmax_t)blockmap->alloc_offset);
881 }
882 }
883
884 /*
885 * Flush various tracking structures to disk
886 */
887 void
888 flush_all_volumes(void)
889 {
890 volume_info_t volume;
891
892 TAILQ_FOREACH(volume, &VolList, entry)
893 flush_volume(volume);
894 }
895
896 void
897 flush_volume(volume_info_t volume)
898 {
899 buffer_info_t buffer;
900 int i;
901
902 for (i = 0; i < HAMMER_BUFLISTS; ++i) {
903 TAILQ_FOREACH(buffer, &volume->buffer_lists[i], entry)
904 flush_buffer(buffer);
905 }
906 if (writehammervol(volume) == -1) {
907 err(1, "Write volume %d (%s)", volume->vol_no, volume->name);
908 /* not reached */
909 }
910 }
911
912 void
913 flush_buffer(buffer_info_t buffer)
914 {
915 volume_info_t volume;
916
917 volume = buffer->volume;
918 if (writehammerbuf(buffer) == -1) {
919 err(1, "Write volume %d (%s)", volume->vol_no, volume->name);
920 /* not reached */
921 }
922 buffer->cache.modified = 0;
923 }
924
925 /*
926 * Core I/O operations
927 */
928 static
929 int
930 __read(volume_info_t volume, void *data, int64_t offset, int size)
931 {
932 ssize_t n;
933
934 n = pread(volume->fd, data, size, offset);
935 if (n != size)
936 return(-1);
937 return(0);
938 }
939
940 static __inline
941 int
942 readhammervol(volume_info_t volume)
943 {
944 return(__read(volume, volume->ondisk, 0, HAMMER_BUFSIZE));
945 }
946
947 static __inline
948 int
949 readhammerbuf(buffer_info_t buffer)
950 {
951 return(__read(buffer->volume, buffer->ondisk, buffer->raw_offset,
952 HAMMER_BUFSIZE));
953 }
954
955 static
956 int
957 __write(volume_info_t volume, const void *data, int64_t offset, int size)
958 {
959 ssize_t n;
960
961 if (volume->rdonly)
962 return(0);
963
964 n = pwrite(volume->fd, data, size, offset);
965 if (n != size)
966 return(-1);
967 return(0);
968 }
969
970 static __inline
971 int
972 writehammervol(volume_info_t volume)
973 {
974 return(__write(volume, volume->ondisk, 0, HAMMER_BUFSIZE));
975 }
976
977 static __inline
978 int
979 writehammerbuf(buffer_info_t buffer)
980 {
981 return(__write(buffer->volume, buffer->ondisk, buffer->raw_offset,
982 HAMMER_BUFSIZE));
983 }
984
985 int64_t
986 init_boot_area_size(int64_t value, off_t avg_vol_size)
987 {
988 if (value == 0) {
989 value = HAMMER_BOOT_NOMBYTES;
990 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
991 value >>= 1;
992 }
993
994 if (value < HAMMER_BOOT_MINBYTES)
995 value = HAMMER_BOOT_MINBYTES;
996 else if (value > HAMMER_BOOT_MAXBYTES)
997 value = HAMMER_BOOT_MAXBYTES;
998
999 return(value);
1000 }
1001
1002 int64_t
1003 init_memory_log_size(int64_t value, off_t avg_vol_size)
1004 {
1005 if (value == 0) {
1006 value = HAMMER_MEM_NOMBYTES;
1007 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
1008 value >>= 1;
1009 }
1010
1011 if (value < HAMMER_MEM_MINBYTES)
1012 value = HAMMER_MEM_MINBYTES;
1013 else if (value > HAMMER_MEM_MAXBYTES)
1014 value = HAMMER_MEM_MAXBYTES;
1015
1016 return(value);
1017 }
DragonFly BSD