nth_packed_object_offset(): index v3 support
[git/packv4.git] / sha1_file.c
blob47179a5c3867c05c2d2b5b80c53d20dfc4e82b41
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "blob.h"
13 #include "commit.h"
14 #include "tag.h"
15 #include "tree.h"
16 #include "refs.h"
18 #ifndef O_NOATIME
19 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
20 #define O_NOATIME 01000000
21 #else
22 #define O_NOATIME 0
23 #endif
24 #endif
26 #ifdef NO_C99_FORMAT
27 #define SZ_FMT "lu"
28 #else
29 #define SZ_FMT "zu"
30 #endif
32 const unsigned char null_sha1[20];
34 static unsigned int sha1_file_open_flag = O_NOATIME;
36 const signed char hexval_table[256] = {
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
43 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
44 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
45 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
71 int get_sha1_hex(const char *hex, unsigned char *sha1)
73 int i;
74 for (i = 0; i < 20; i++) {
75 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
76 if (val & ~0xff)
77 return -1;
78 *sha1++ = val;
79 hex += 2;
81 return 0;
84 int safe_create_leading_directories(char *path)
86 char *pos = path;
87 struct stat st;
89 if (*pos == '/')
90 pos++;
92 while (pos) {
93 pos = strchr(pos, '/');
94 if (!pos)
95 break;
96 *pos = 0;
97 if (!stat(path, &st)) {
98 /* path exists */
99 if (!S_ISDIR(st.st_mode)) {
100 *pos = '/';
101 return -3;
104 else if (mkdir(path, 0777)) {
105 *pos = '/';
106 return -1;
108 else if (adjust_shared_perm(path)) {
109 *pos = '/';
110 return -2;
112 *pos++ = '/';
114 return 0;
117 char * sha1_to_hex(const unsigned char *sha1)
119 static int bufno;
120 static char hexbuffer[4][50];
121 static const char hex[] = "0123456789abcdef";
122 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
123 int i;
125 for (i = 0; i < 20; i++) {
126 unsigned int val = *sha1++;
127 *buf++ = hex[val >> 4];
128 *buf++ = hex[val & 0xf];
130 *buf = '\0';
132 return buffer;
135 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
137 int i;
138 for (i = 0; i < 20; i++) {
139 static char hex[] = "0123456789abcdef";
140 unsigned int val = sha1[i];
141 char *pos = pathbuf + i*2 + (i > 0);
142 *pos++ = hex[val >> 4];
143 *pos = hex[val & 0xf];
148 * NOTE! This returns a statically allocated buffer, so you have to be
149 * careful about using it. Do a "xstrdup()" if you need to save the
150 * filename.
152 * Also note that this returns the location for creating. Reading
153 * SHA1 file can happen from any alternate directory listed in the
154 * DB_ENVIRONMENT environment variable if it is not found in
155 * the primary object database.
157 char *sha1_file_name(const unsigned char *sha1)
159 static char *name, *base;
161 if (!base) {
162 const char *sha1_file_directory = get_object_directory();
163 int len = strlen(sha1_file_directory);
164 base = xmalloc(len + 60);
165 memcpy(base, sha1_file_directory, len);
166 memset(base+len, 0, 60);
167 base[len] = '/';
168 base[len+3] = '/';
169 name = base + len + 1;
171 fill_sha1_path(name, sha1);
172 return base;
175 char *sha1_pack_name(const unsigned char *sha1)
177 static const char hex[] = "0123456789abcdef";
178 static char *name, *base, *buf;
179 int i;
181 if (!base) {
182 const char *sha1_file_directory = get_object_directory();
183 int len = strlen(sha1_file_directory);
184 base = xmalloc(len + 60);
185 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
186 name = base + len + 11;
189 buf = name;
191 for (i = 0; i < 20; i++) {
192 unsigned int val = *sha1++;
193 *buf++ = hex[val >> 4];
194 *buf++ = hex[val & 0xf];
197 return base;
200 char *sha1_pack_index_name(const unsigned char *sha1)
202 static const char hex[] = "0123456789abcdef";
203 static char *name, *base, *buf;
204 int i;
206 if (!base) {
207 const char *sha1_file_directory = get_object_directory();
208 int len = strlen(sha1_file_directory);
209 base = xmalloc(len + 60);
210 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
211 name = base + len + 11;
214 buf = name;
216 for (i = 0; i < 20; i++) {
217 unsigned int val = *sha1++;
218 *buf++ = hex[val >> 4];
219 *buf++ = hex[val & 0xf];
222 return base;
225 struct alternate_object_database *alt_odb_list;
226 static struct alternate_object_database **alt_odb_tail;
228 static void read_info_alternates(const char * alternates, int depth);
231 * Prepare alternate object database registry.
233 * The variable alt_odb_list points at the list of struct
234 * alternate_object_database. The elements on this list come from
235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
237 * whose contents is similar to that environment variable but can be
238 * LF separated. Its base points at a statically allocated buffer that
239 * contains "/the/directory/corresponding/to/.git/objects/...", while
240 * its name points just after the slash at the end of ".git/objects/"
241 * in the example above, and has enough space to hold 40-byte hex
242 * SHA1, an extra slash for the first level indirection, and the
243 * terminating NUL.
245 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
247 struct stat st;
248 const char *objdir = get_object_directory();
249 struct alternate_object_database *ent;
250 struct alternate_object_database *alt;
251 /* 43 = 40-byte + 2 '/' + terminating NUL */
252 int pfxlen = len;
253 int entlen = pfxlen + 43;
254 int base_len = -1;
256 if (*entry != '/' && relative_base) {
257 /* Relative alt-odb */
258 if (base_len < 0)
259 base_len = strlen(relative_base) + 1;
260 entlen += base_len;
261 pfxlen += base_len;
263 ent = xmalloc(sizeof(*ent) + entlen);
265 if (*entry != '/' && relative_base) {
266 memcpy(ent->base, relative_base, base_len - 1);
267 ent->base[base_len - 1] = '/';
268 memcpy(ent->base + base_len, entry, len);
270 else
271 memcpy(ent->base, entry, pfxlen);
273 ent->name = ent->base + pfxlen + 1;
274 ent->base[pfxlen + 3] = '/';
275 ent->base[pfxlen] = ent->base[entlen-1] = 0;
277 /* Detect cases where alternate disappeared */
278 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
279 error("object directory %s does not exist; "
280 "check .git/objects/info/alternates.",
281 ent->base);
282 free(ent);
283 return -1;
286 /* Prevent the common mistake of listing the same
287 * thing twice, or object directory itself.
289 for (alt = alt_odb_list; alt; alt = alt->next) {
290 if (!memcmp(ent->base, alt->base, pfxlen)) {
291 free(ent);
292 return -1;
295 if (!memcmp(ent->base, objdir, pfxlen)) {
296 free(ent);
297 return -1;
300 /* add the alternate entry */
301 *alt_odb_tail = ent;
302 alt_odb_tail = &(ent->next);
303 ent->next = NULL;
305 /* recursively add alternates */
306 read_info_alternates(ent->base, depth + 1);
308 ent->base[pfxlen] = '/';
310 return 0;
313 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
314 const char *relative_base, int depth)
316 const char *cp, *last;
318 if (depth > 5) {
319 error("%s: ignoring alternate object stores, nesting too deep.",
320 relative_base);
321 return;
324 last = alt;
325 while (last < ep) {
326 cp = last;
327 if (cp < ep && *cp == '#') {
328 while (cp < ep && *cp != sep)
329 cp++;
330 last = cp + 1;
331 continue;
333 while (cp < ep && *cp != sep)
334 cp++;
335 if (last != cp) {
336 if ((*last != '/') && depth) {
337 error("%s: ignoring relative alternate object store %s",
338 relative_base, last);
339 } else {
340 link_alt_odb_entry(last, cp - last,
341 relative_base, depth);
344 while (cp < ep && *cp == sep)
345 cp++;
346 last = cp;
350 static void read_info_alternates(const char * relative_base, int depth)
352 char *map;
353 size_t mapsz;
354 struct stat st;
355 const char alt_file_name[] = "info/alternates";
356 /* Given that relative_base is no longer than PATH_MAX,
357 ensure that "path" has enough space to append "/", the
358 file name, "info/alternates", and a trailing NUL. */
359 char path[PATH_MAX + 1 + sizeof alt_file_name];
360 int fd;
362 sprintf(path, "%s/%s", relative_base, alt_file_name);
363 fd = open(path, O_RDONLY);
364 if (fd < 0)
365 return;
366 if (fstat(fd, &st) || (st.st_size == 0)) {
367 close(fd);
368 return;
370 mapsz = xsize_t(st.st_size);
371 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
372 close(fd);
374 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
376 munmap(map, mapsz);
379 void prepare_alt_odb(void)
381 const char *alt;
383 if (alt_odb_tail)
384 return;
386 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
387 if (!alt) alt = "";
389 alt_odb_tail = &alt_odb_list;
390 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
392 read_info_alternates(get_object_directory(), 0);
395 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
397 char *name = sha1_file_name(sha1);
398 struct alternate_object_database *alt;
400 if (!stat(name, st))
401 return name;
402 prepare_alt_odb();
403 for (alt = alt_odb_list; alt; alt = alt->next) {
404 name = alt->name;
405 fill_sha1_path(name, sha1);
406 if (!stat(alt->base, st))
407 return alt->base;
409 return NULL;
412 static unsigned int pack_used_ctr;
413 static unsigned int pack_mmap_calls;
414 static unsigned int peak_pack_open_windows;
415 static unsigned int pack_open_windows;
416 static size_t peak_pack_mapped;
417 static size_t pack_mapped;
418 struct packed_git *packed_git;
420 void pack_report(void)
422 fprintf(stderr,
423 "pack_report: getpagesize() = %10" SZ_FMT "\n"
424 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
425 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
426 (size_t) getpagesize(),
427 packed_git_window_size,
428 packed_git_limit);
429 fprintf(stderr,
430 "pack_report: pack_used_ctr = %10u\n"
431 "pack_report: pack_mmap_calls = %10u\n"
432 "pack_report: pack_open_windows = %10u / %10u\n"
433 "pack_report: pack_mapped = "
434 "%10" SZ_FMT " / %10" SZ_FMT "\n",
435 pack_used_ctr,
436 pack_mmap_calls,
437 pack_open_windows, peak_pack_open_windows,
438 pack_mapped, peak_pack_mapped);
441 static int check_packed_git_idx(const char *path, struct packed_git *p)
443 void *idx_map;
444 struct pack_idx_header *hdr;
445 size_t idx_size;
446 uint32_t version, nr, i, *index;
447 int fd = open(path, O_RDONLY);
448 struct stat st;
450 if (fd < 0)
451 return -1;
452 if (fstat(fd, &st)) {
453 close(fd);
454 return -1;
456 idx_size = xsize_t(st.st_size);
457 if (idx_size < 4 * 256 + 20 + 20) {
458 close(fd);
459 return error("index file %s is too small", path);
461 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
462 close(fd);
464 hdr = idx_map;
465 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
466 version = ntohl(hdr->idx_version);
467 if (version < 2 || version > 3) {
468 munmap(idx_map, idx_size);
469 return error("index file %s is version %d"
470 " and is not supported by this binary"
471 " (try upgrading GIT to a newer version)",
472 path, version);
474 } else
475 version = 1;
477 nr = 0;
478 index = idx_map;
479 if (version > 1)
480 index += 2; /* skip index header */
481 for (i = 0; i < 256; i++) {
482 uint32_t n = ntohl(index[i]);
483 if (n < nr) {
484 munmap(idx_map, idx_size);
485 return error("non-monotonic index %s", path);
487 nr = n;
490 if (version == 1) {
492 * Total size:
493 * - 256 index entries 4 bytes each
494 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
495 * - 20-byte SHA1 of the packfile
496 * - 20-byte SHA1 file checksum
498 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
499 munmap(idx_map, idx_size);
500 return error("wrong index v1 file size in %s", path);
502 } else if (version == 2) {
504 * Minimum size:
505 * - 8 bytes of header
506 * - 256 index entries 4 bytes each
507 * - 20-byte sha1 entry * nr
508 * - 4-byte crc entry * nr
509 * - 4-byte offset entry * nr
510 * - 20-byte SHA1 of the packfile
511 * - 20-byte SHA1 file checksum
512 * And after the 4-byte offset table might be a
513 * variable sized table containing 8-byte entries
514 * for offsets larger than 2^31.
516 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
517 unsigned long max_size = min_size;
518 if (nr)
519 max_size += (nr - 1)*8;
520 if (idx_size < min_size || idx_size > max_size) {
521 munmap(idx_map, idx_size);
522 return error("wrong index v2 file size in %s", path);
524 if (idx_size != min_size) {
525 /* make sure we can deal with large pack offsets */
526 off_t x = 0x7fffffffUL, y = 0xffffffffUL;
527 if (x > (x + 1) || y > (y + 1)) {
528 munmap(idx_map, idx_size);
529 return error("pack too large for current definition of off_t in %s", path);
532 } else if (version == 3) {
534 * Minimum size:
535 * - 8 bytes of header
536 * - 256 index entries 4 bytes each
537 * - 4-byte crc entry * nr
538 * - 4-byte offset entry * nr
539 * - 20-byte SHA1 of the packfile
540 * - 20-byte SHA1 file checksum
541 * And after the 4-byte offset table might be a
542 * variable sized table containing 8-byte entries
543 * for offsets larger than 2^31.
545 unsigned long min_size = 8 + 4*256 + nr*(4 + 4) + 20 + 20;
546 unsigned long max_size = min_size;
547 if (nr)
548 max_size += (nr - 1)*8;
549 if (idx_size < min_size || idx_size > max_size) {
550 munmap(idx_map, idx_size);
551 return error("wrong index v3 file size in %s", path);
553 if (idx_size != min_size) {
554 /* make sure we can deal with large pack offsets */
555 off_t x = 0x7fffffffUL, y = 0xffffffffUL;
556 if (x > (x + 1) || y > (y + 1)) {
557 munmap(idx_map, idx_size);
558 return error("pack too large for current definition of off_t in %s", path);
563 p->index_version = version;
564 p->index_data = idx_map;
565 p->index_size = idx_size;
566 p->num_objects = nr;
567 return 0;
570 int open_pack_index(struct packed_git *p)
572 char *idx_name;
573 int ret;
575 if (p->index_data)
576 return 0;
578 idx_name = xstrdup(p->pack_name);
579 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
580 ret = check_packed_git_idx(idx_name, p);
581 free(idx_name);
582 return ret;
585 static void scan_windows(struct packed_git *p,
586 struct packed_git **lru_p,
587 struct pack_window **lru_w,
588 struct pack_window **lru_l)
590 struct pack_window *w, *w_l;
592 for (w_l = NULL, w = p->windows; w; w = w->next) {
593 if (!w->inuse_cnt) {
594 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
595 *lru_p = p;
596 *lru_w = w;
597 *lru_l = w_l;
600 w_l = w;
604 static int unuse_one_window(struct packed_git *current, int keep_fd)
606 struct packed_git *p, *lru_p = NULL;
607 struct pack_window *lru_w = NULL, *lru_l = NULL;
609 if (current)
610 scan_windows(current, &lru_p, &lru_w, &lru_l);
611 for (p = packed_git; p; p = p->next)
612 scan_windows(p, &lru_p, &lru_w, &lru_l);
613 if (lru_p) {
614 munmap(lru_w->base, lru_w->len);
615 pack_mapped -= lru_w->len;
616 if (lru_l)
617 lru_l->next = lru_w->next;
618 else {
619 lru_p->windows = lru_w->next;
620 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
621 close(lru_p->pack_fd);
622 lru_p->pack_fd = -1;
625 free(lru_w);
626 pack_open_windows--;
627 return 1;
629 return 0;
632 void release_pack_memory(size_t need, int fd)
634 size_t cur = pack_mapped;
635 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
636 ; /* nothing */
639 void unuse_pack(struct pack_window **w_cursor)
641 struct pack_window *w = *w_cursor;
642 if (w) {
643 w->inuse_cnt--;
644 *w_cursor = NULL;
649 * Do not call this directly as this leaks p->pack_fd on error return;
650 * call open_packed_git() instead.
652 static int open_packed_git_1(struct packed_git *p)
654 struct stat st;
655 struct pack_header hdr;
656 unsigned char sha1[20];
657 unsigned char *idx_sha1;
658 long fd_flag;
660 if (!p->index_data && open_pack_index(p))
661 return error("packfile %s index unavailable", p->pack_name);
663 p->pack_fd = open(p->pack_name, O_RDONLY);
664 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
665 return -1;
667 /* If we created the struct before we had the pack we lack size. */
668 if (!p->pack_size) {
669 if (!S_ISREG(st.st_mode))
670 return error("packfile %s not a regular file", p->pack_name);
671 p->pack_size = st.st_size;
672 } else if (p->pack_size != st.st_size)
673 return error("packfile %s size changed", p->pack_name);
675 /* We leave these file descriptors open with sliding mmap;
676 * there is no point keeping them open across exec(), though.
678 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
679 if (fd_flag < 0)
680 return error("cannot determine file descriptor flags");
681 fd_flag |= FD_CLOEXEC;
682 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
683 return error("cannot set FD_CLOEXEC");
685 /* Verify we recognize this pack file format. */
686 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
687 return error("file %s is far too short to be a packfile", p->pack_name);
688 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
689 return error("file %s is not a GIT packfile", p->pack_name);
690 if (!pack_version_ok(hdr.hdr_version))
691 return error("packfile %s is version %u and not supported"
692 " (try upgrading GIT to a newer version)",
693 p->pack_name, ntohl(hdr.hdr_version));
695 /* Verify the pack matches its index. */
696 if (p->num_objects != ntohl(hdr.hdr_entries))
697 return error("packfile %s claims to have %u objects"
698 " while index indicates %u objects",
699 p->pack_name, ntohl(hdr.hdr_entries),
700 p->num_objects);
701 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
702 return error("end of packfile %s is unavailable", p->pack_name);
703 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
704 return error("packfile %s signature is unavailable", p->pack_name);
705 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
706 if (hashcmp(sha1, idx_sha1))
707 return error("packfile %s does not match index", p->pack_name);
708 return 0;
711 static int open_packed_git(struct packed_git *p)
713 if (!open_packed_git_1(p))
714 return 0;
715 if (p->pack_fd != -1) {
716 close(p->pack_fd);
717 p->pack_fd = -1;
719 return -1;
722 static void open_pack_sha1_dict(struct packed_git *p)
724 unsigned char *dict;
725 const size_t offset = sizeof(struct pack_header);
727 dict = xmmap(NULL, offset + p->num_objects * 20, PROT_READ,
728 MAP_PRIVATE, p->pack_fd, 0);
729 p->sha1_dict = dict + offset;
732 static int in_window(struct pack_window *win, off_t offset)
734 /* We must promise at least 20 bytes (one hash) after the
735 * offset is available from this window, otherwise the offset
736 * is not actually in this window and a different window (which
737 * has that one hash excess) must be used. This is to support
738 * the object header and delta base parsing routines below.
740 off_t win_off = win->offset;
741 return win_off <= offset
742 && (offset + 20) <= (win_off + win->len);
745 unsigned char* use_pack(struct packed_git *p,
746 struct pack_window **w_cursor,
747 off_t offset,
748 unsigned int *left)
750 struct pack_window *win = *w_cursor;
752 if (p->pack_fd == -1 && open_packed_git(p))
753 die("packfile %s cannot be accessed", p->pack_name);
755 /* Since packfiles end in a hash of their content and its
756 * pointless to ask for an offset into the middle of that
757 * hash, and the in_window function above wouldn't match
758 * don't allow an offset too close to the end of the file.
760 if (offset > (p->pack_size - 20))
761 die("offset beyond end of packfile (truncated pack?)");
763 if (!win || !in_window(win, offset)) {
764 if (win)
765 win->inuse_cnt--;
766 for (win = p->windows; win; win = win->next) {
767 if (in_window(win, offset))
768 break;
770 if (!win) {
771 size_t window_align = packed_git_window_size / 2;
772 off_t len;
773 win = xcalloc(1, sizeof(*win));
774 win->offset = (offset / window_align) * window_align;
775 len = p->pack_size - win->offset;
776 if (len > packed_git_window_size)
777 len = packed_git_window_size;
778 win->len = (size_t)len;
779 pack_mapped += win->len;
780 while (packed_git_limit < pack_mapped
781 && unuse_one_window(p, p->pack_fd))
782 ; /* nothing */
783 win->base = xmmap(NULL, win->len,
784 PROT_READ, MAP_PRIVATE,
785 p->pack_fd, win->offset);
786 if (win->base == MAP_FAILED)
787 die("packfile %s cannot be mapped: %s",
788 p->pack_name,
789 strerror(errno));
790 pack_mmap_calls++;
791 pack_open_windows++;
792 if (pack_mapped > peak_pack_mapped)
793 peak_pack_mapped = pack_mapped;
794 if (pack_open_windows > peak_pack_open_windows)
795 peak_pack_open_windows = pack_open_windows;
796 win->next = p->windows;
797 p->windows = win;
800 if (win != *w_cursor) {
801 win->last_used = pack_used_ctr++;
802 win->inuse_cnt++;
803 *w_cursor = win;
805 offset -= win->offset;
806 if (left)
807 *left = win->len - xsize_t(offset);
808 return win->base + offset;
811 struct packed_git *add_packed_git(const char *path, int path_len, int local)
813 struct stat st;
814 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
817 * Make sure a corresponding .pack file exists and that
818 * the index looks sane.
820 path_len -= strlen(".idx");
821 if (path_len < 1)
822 return NULL;
823 memcpy(p->pack_name, path, path_len);
824 strcpy(p->pack_name + path_len, ".pack");
825 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
826 free(p);
827 return NULL;
830 /* ok, it looks sane as far as we can check without
831 * actually mapping the pack file.
833 p->index_version = 0;
834 p->index_data = NULL;
835 p->index_size = 0;
836 p->num_objects = 0;
837 p->pack_size = st.st_size;
838 p->next = NULL;
839 p->windows = NULL;
840 p->pack_fd = -1;
841 p->pack_local = local;
842 p->mtime = st.st_mtime;
843 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
844 hashclr(p->sha1);
845 return p;
848 struct packed_git *parse_pack_index(unsigned char *sha1)
850 char *path = sha1_pack_index_name(sha1);
851 return parse_pack_index_file(sha1, path);
854 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
855 const char *idx_path)
857 const char *path = sha1_pack_name(sha1);
858 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
860 if (check_packed_git_idx(idx_path, p)) {
861 free(p);
862 return NULL;
865 strcpy(p->pack_name, path);
866 p->pack_size = 0;
867 p->next = NULL;
868 p->windows = NULL;
869 p->pack_fd = -1;
870 hashcpy(p->sha1, sha1);
871 return p;
874 void install_packed_git(struct packed_git *pack)
876 pack->next = packed_git;
877 packed_git = pack;
880 static void prepare_packed_git_one(char *objdir, int local)
882 /* Ensure that this buffer is large enough so that we can
883 append "/pack/" without clobbering the stack even if
884 strlen(objdir) were PATH_MAX. */
885 char path[PATH_MAX + 1 + 4 + 1 + 1];
886 int len;
887 DIR *dir;
888 struct dirent *de;
890 sprintf(path, "%s/pack", objdir);
891 len = strlen(path);
892 dir = opendir(path);
893 if (!dir) {
894 if (errno != ENOENT)
895 error("unable to open object pack directory: %s: %s",
896 path, strerror(errno));
897 return;
899 path[len++] = '/';
900 while ((de = readdir(dir)) != NULL) {
901 int namelen = strlen(de->d_name);
902 struct packed_git *p;
904 if (!has_extension(de->d_name, ".idx"))
905 continue;
907 if (len + namelen + 1 > sizeof(path))
908 continue;
910 /* Don't reopen a pack we already have. */
911 strcpy(path + len, de->d_name);
912 for (p = packed_git; p; p = p->next) {
913 if (!memcmp(path, p->pack_name, len + namelen - 4))
914 break;
916 if (p)
917 continue;
918 /* See if it really is a valid .idx file with corresponding
919 * .pack file that we can map.
921 p = add_packed_git(path, len + namelen, local);
922 if (!p)
923 continue;
924 install_packed_git(p);
926 closedir(dir);
929 static int sort_pack(const void *a_, const void *b_)
931 struct packed_git *a = *((struct packed_git **)a_);
932 struct packed_git *b = *((struct packed_git **)b_);
933 int st;
936 * Local packs tend to contain objects specific to our
937 * variant of the project than remote ones. In addition,
938 * remote ones could be on a network mounted filesystem.
939 * Favor local ones for these reasons.
941 st = a->pack_local - b->pack_local;
942 if (st)
943 return -st;
946 * Younger packs tend to contain more recent objects,
947 * and more recent objects tend to get accessed more
948 * often.
950 if (a->mtime < b->mtime)
951 return 1;
952 else if (a->mtime == b->mtime)
953 return 0;
954 return -1;
957 static void rearrange_packed_git(void)
959 struct packed_git **ary, *p;
960 int i, n;
962 for (n = 0, p = packed_git; p; p = p->next)
963 n++;
964 if (n < 2)
965 return;
967 /* prepare an array of packed_git for easier sorting */
968 ary = xcalloc(n, sizeof(struct packed_git *));
969 for (n = 0, p = packed_git; p; p = p->next)
970 ary[n++] = p;
972 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
974 /* link them back again */
975 for (i = 0; i < n - 1; i++)
976 ary[i]->next = ary[i + 1];
977 ary[n - 1]->next = NULL;
978 packed_git = ary[0];
980 free(ary);
983 static int prepare_packed_git_run_once = 0;
984 void prepare_packed_git(void)
986 struct alternate_object_database *alt;
988 if (prepare_packed_git_run_once)
989 return;
990 prepare_packed_git_one(get_object_directory(), 1);
991 prepare_alt_odb();
992 for (alt = alt_odb_list; alt; alt = alt->next) {
993 alt->name[-1] = 0;
994 prepare_packed_git_one(alt->base, 0);
995 alt->name[-1] = '/';
997 rearrange_packed_git();
998 prepare_packed_git_run_once = 1;
1001 void reprepare_packed_git(void)
1003 prepare_packed_git_run_once = 0;
1004 prepare_packed_git();
1007 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1009 unsigned char real_sha1[20];
1010 hash_sha1_file(map, size, type, real_sha1);
1011 return hashcmp(sha1, real_sha1) ? -1 : 0;
1014 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1016 struct stat st;
1017 void *map;
1018 int fd;
1019 char *filename = find_sha1_file(sha1, &st);
1021 if (!filename) {
1022 return NULL;
1025 fd = open(filename, O_RDONLY | sha1_file_open_flag);
1026 if (fd < 0) {
1027 /* See if it works without O_NOATIME */
1028 switch (sha1_file_open_flag) {
1029 default:
1030 fd = open(filename, O_RDONLY);
1031 if (fd >= 0)
1032 break;
1033 /* Fallthrough */
1034 case 0:
1035 return NULL;
1038 /* If it failed once, it will probably fail again.
1039 * Stop using O_NOATIME
1041 sha1_file_open_flag = 0;
1043 *size = xsize_t(st.st_size);
1044 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1045 close(fd);
1046 return map;
1049 static int legacy_loose_object(unsigned char *map)
1051 unsigned int word;
1054 * Is it a zlib-compressed buffer? If so, the first byte
1055 * must be 0x78 (15-bit window size, deflated), and the
1056 * first 16-bit word is evenly divisible by 31
1058 word = (map[0] << 8) + map[1];
1059 if (map[0] == 0x78 && !(word % 31))
1060 return 1;
1061 else
1062 return 0;
1065 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1067 unsigned shift;
1068 unsigned char c;
1069 unsigned long size;
1070 unsigned long used = 0;
1072 c = buf[used++];
1073 *type = (c >> 4) & 7;
1074 size = c & 15;
1075 shift = 4;
1076 while (c & 0x80) {
1077 if (len <= used)
1078 return 0;
1079 if (sizeof(long) * 8 <= shift)
1080 return 0;
1081 c = buf[used++];
1082 size += (c & 0x7f) << shift;
1083 shift += 7;
1085 *sizep = size;
1086 return used;
1089 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1091 unsigned long size, used;
1092 static const char valid_loose_object_type[8] = {
1093 0, /* OBJ_EXT */
1094 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1095 0, /* "delta" and others are invalid in a loose object */
1097 enum object_type type;
1099 /* Get the data stream */
1100 memset(stream, 0, sizeof(*stream));
1101 stream->next_in = map;
1102 stream->avail_in = mapsize;
1103 stream->next_out = buffer;
1104 stream->avail_out = bufsiz;
1106 if (legacy_loose_object(map)) {
1107 inflateInit(stream);
1108 return inflate(stream, 0);
1113 * There used to be a second loose object header format which
1114 * was meant to mimic the in-pack format, allowing for direct
1115 * copy of the object data. This format turned up not to be
1116 * really worth it and we don't write it any longer. But we
1117 * can still read it.
1119 used = unpack_object_header_gently(map, mapsize, &type, &size);
1120 if (!used || !valid_loose_object_type[type])
1121 return -1;
1122 map += used;
1123 mapsize -= used;
1125 /* Set up the stream for the rest.. */
1126 stream->next_in = map;
1127 stream->avail_in = mapsize;
1128 inflateInit(stream);
1130 /* And generate the fake traditional header */
1131 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1132 typename(type), size);
1133 return 0;
1136 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1138 int bytes = strlen(buffer) + 1;
1139 unsigned char *buf = xmalloc(1+size);
1140 unsigned long n;
1141 int status = Z_OK;
1143 n = stream->total_out - bytes;
1144 if (n > size)
1145 n = size;
1146 memcpy(buf, (char *) buffer + bytes, n);
1147 bytes = n;
1148 if (bytes <= size) {
1150 * The above condition must be (bytes <= size), not
1151 * (bytes < size). In other words, even though we
1152 * expect no more output and set avail_out to zer0,
1153 * the input zlib stream may have bytes that express
1154 * "this concludes the stream", and we *do* want to
1155 * eat that input.
1157 * Otherwise we would not be able to test that we
1158 * consumed all the input to reach the expected size;
1159 * we also want to check that zlib tells us that all
1160 * went well with status == Z_STREAM_END at the end.
1162 stream->next_out = buf + bytes;
1163 stream->avail_out = size - bytes;
1164 while (status == Z_OK)
1165 status = inflate(stream, Z_FINISH);
1167 buf[size] = 0;
1168 if (status == Z_STREAM_END && !stream->avail_in) {
1169 inflateEnd(stream);
1170 return buf;
1173 if (status < 0)
1174 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1175 else if (stream->avail_in)
1176 error("garbage at end of loose object '%s'",
1177 sha1_to_hex(sha1));
1178 free(buf);
1179 return NULL;
1183 * We used to just use "sscanf()", but that's actually way
1184 * too permissive for what we want to check. So do an anal
1185 * object header parse by hand.
1187 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1189 char type[10];
1190 int i;
1191 unsigned long size;
1194 * The type can be at most ten bytes (including the
1195 * terminating '\0' that we add), and is followed by
1196 * a space.
1198 i = 0;
1199 for (;;) {
1200 char c = *hdr++;
1201 if (c == ' ')
1202 break;
1203 type[i++] = c;
1204 if (i >= sizeof(type))
1205 return -1;
1207 type[i] = 0;
1210 * The length must follow immediately, and be in canonical
1211 * decimal format (ie "010" is not valid).
1213 size = *hdr++ - '0';
1214 if (size > 9)
1215 return -1;
1216 if (size) {
1217 for (;;) {
1218 unsigned long c = *hdr - '0';
1219 if (c > 9)
1220 break;
1221 hdr++;
1222 size = size * 10 + c;
1225 *sizep = size;
1228 * The length must be followed by a zero byte
1230 return *hdr ? -1 : type_from_string(type);
1233 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1235 int ret;
1236 z_stream stream;
1237 char hdr[8192];
1239 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1240 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1241 return NULL;
1243 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1246 unsigned long get_size_from_delta(struct packed_git *p,
1247 struct pack_window **w_curs,
1248 off_t curpos)
1250 const unsigned char *data;
1251 unsigned char delta_head[20], *in;
1252 z_stream stream;
1253 int st;
1255 memset(&stream, 0, sizeof(stream));
1256 stream.next_out = delta_head;
1257 stream.avail_out = sizeof(delta_head);
1259 inflateInit(&stream);
1260 do {
1261 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1262 stream.next_in = in;
1263 st = inflate(&stream, Z_FINISH);
1264 curpos += stream.next_in - in;
1265 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1266 stream.total_out < sizeof(delta_head));
1267 inflateEnd(&stream);
1268 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1269 die("delta data unpack-initial failed");
1271 /* Examine the initial part of the delta to figure out
1272 * the result size.
1274 data = delta_head;
1276 /* ignore base size */
1277 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1279 /* Read the result size */
1280 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1283 static off_t get_delta_base(struct packed_git *p,
1284 struct pack_window **w_curs,
1285 off_t *curpos,
1286 enum object_type type,
1287 off_t delta_obj_offset)
1289 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1290 off_t base_offset;
1292 /* use_pack() assured us we have [base_info, base_info + 20)
1293 * as a range that we can look at without walking off the
1294 * end of the mapped window. Its actually the hash size
1295 * that is assured. An OFS_DELTA longer than the hash size
1296 * is stupid, as then a REF_DELTA would be smaller to store.
1298 if (type == OBJ_OFS_DELTA) {
1299 unsigned used = 0;
1300 unsigned char c = base_info[used++];
1301 base_offset = c & 127;
1302 while (c & 128) {
1303 base_offset += 1;
1304 if (!base_offset || MSB(base_offset, 7))
1305 die("offset value overflow for delta base object");
1306 c = base_info[used++];
1307 base_offset = (base_offset << 7) + (c & 127);
1309 base_offset = delta_obj_offset - base_offset;
1310 if (base_offset >= delta_obj_offset)
1311 die("delta base offset out of bound");
1312 *curpos += used;
1313 } else if (type == OBJ_REF_DELTA) {
1314 /* The base entry _must_ be in the same pack */
1315 base_offset = find_pack_entry_one(base_info, p);
1316 if (!base_offset)
1317 die("failed to find delta-pack base object %s",
1318 sha1_to_hex(base_info));
1319 *curpos += 20;
1320 } else
1321 die("I am totally screwed");
1322 return base_offset;
1325 /* forward declaration for a mutually recursive function */
1326 static int packed_object_info(struct packed_git *p, off_t offset,
1327 unsigned long *sizep);
1329 static int packed_delta_info(struct packed_git *p,
1330 struct pack_window **w_curs,
1331 off_t curpos,
1332 enum object_type type,
1333 off_t obj_offset,
1334 unsigned long *sizep)
1336 off_t base_offset;
1338 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1339 type = packed_object_info(p, base_offset, NULL);
1341 /* We choose to only get the type of the base object and
1342 * ignore potentially corrupt pack file that expects the delta
1343 * based on a base with a wrong size. This saves tons of
1344 * inflate() calls.
1346 if (sizep)
1347 *sizep = get_size_from_delta(p, w_curs, curpos);
1349 return type;
1352 static int unpack_object_header(struct packed_git *p,
1353 struct pack_window **w_curs,
1354 off_t *curpos,
1355 unsigned long *sizep)
1357 unsigned char *base;
1358 unsigned int left;
1359 unsigned long used;
1360 enum object_type type;
1362 /* use_pack() assures us we have [base, base + 20) available
1363 * as a range that we can look at at. (Its actually the hash
1364 * size that is assured.) With our object header encoding
1365 * the maximum deflated object size is 2^137, which is just
1366 * insane, so we know won't exceed what we have been given.
1368 base = use_pack(p, w_curs, *curpos, &left);
1369 used = unpack_object_header_gently(base, left, &type, sizep);
1370 if (!used)
1371 die("object offset outside of pack file");
1372 *curpos += used;
1374 return type;
1377 const char *packed_object_info_detail(struct packed_git *p,
1378 off_t obj_offset,
1379 unsigned long *size,
1380 unsigned long *store_size,
1381 unsigned int *delta_chain_length,
1382 unsigned char *base_sha1)
1384 struct pack_window *w_curs = NULL;
1385 off_t curpos;
1386 unsigned long dummy;
1387 unsigned char *next_sha1;
1388 enum object_type type;
1390 *delta_chain_length = 0;
1391 curpos = obj_offset;
1392 type = unpack_object_header(p, &w_curs, &curpos, size);
1394 for (;;) {
1395 switch (type) {
1396 default:
1397 die("pack %s contains unknown object type %d",
1398 p->pack_name, type);
1399 case OBJ_COMMIT:
1400 case OBJ_TREE:
1401 case OBJ_BLOB:
1402 case OBJ_TAG:
1403 *store_size = 0; /* notyet */
1404 unuse_pack(&w_curs);
1405 return typename(type);
1406 case OBJ_OFS_DELTA:
1407 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1408 if (*delta_chain_length == 0) {
1409 /* TODO: find base_sha1 as pointed by curpos */
1410 hashclr(base_sha1);
1412 break;
1413 case OBJ_REF_DELTA:
1414 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1415 if (*delta_chain_length == 0)
1416 hashcpy(base_sha1, next_sha1);
1417 obj_offset = find_pack_entry_one(next_sha1, p);
1418 break;
1420 (*delta_chain_length)++;
1421 curpos = obj_offset;
1422 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1426 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1427 unsigned long *sizep)
1429 struct pack_window *w_curs = NULL;
1430 unsigned long size;
1431 off_t curpos = obj_offset;
1432 enum object_type type;
1434 type = unpack_object_header(p, &w_curs, &curpos, &size);
1436 switch (type) {
1437 case OBJ_OFS_DELTA:
1438 case OBJ_REF_DELTA:
1439 type = packed_delta_info(p, &w_curs, curpos,
1440 type, obj_offset, sizep);
1441 break;
1442 case OBJ_COMMIT:
1443 case OBJ_TREE:
1444 case OBJ_BLOB:
1445 case OBJ_TAG:
1446 if (sizep)
1447 *sizep = size;
1448 break;
1449 default:
1450 die("pack %s contains unknown object type %d",
1451 p->pack_name, type);
1453 unuse_pack(&w_curs);
1454 return type;
1457 static void *unpack_compressed_entry(struct packed_git *p,
1458 struct pack_window **w_curs,
1459 off_t curpos,
1460 unsigned long size)
1462 int st;
1463 z_stream stream;
1464 unsigned char *buffer, *in;
1466 buffer = xmalloc(size + 1);
1467 buffer[size] = 0;
1468 memset(&stream, 0, sizeof(stream));
1469 stream.next_out = buffer;
1470 stream.avail_out = size;
1472 inflateInit(&stream);
1473 do {
1474 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1475 stream.next_in = in;
1476 st = inflate(&stream, Z_FINISH);
1477 curpos += stream.next_in - in;
1478 } while (st == Z_OK || st == Z_BUF_ERROR);
1479 inflateEnd(&stream);
1480 if ((st != Z_STREAM_END) || stream.total_out != size) {
1481 free(buffer);
1482 return NULL;
1485 return buffer;
1488 #define MAX_DELTA_CACHE (256)
1490 static size_t delta_base_cached;
1492 static struct delta_base_cache_lru_list {
1493 struct delta_base_cache_lru_list *prev;
1494 struct delta_base_cache_lru_list *next;
1495 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1497 static struct delta_base_cache_entry {
1498 struct delta_base_cache_lru_list lru;
1499 void *data;
1500 struct packed_git *p;
1501 off_t base_offset;
1502 unsigned long size;
1503 enum object_type type;
1504 } delta_base_cache[MAX_DELTA_CACHE];
1506 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1508 unsigned long hash;
1510 hash = (unsigned long)p + (unsigned long)base_offset;
1511 hash += (hash >> 8) + (hash >> 16);
1512 return hash % MAX_DELTA_CACHE;
1515 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1516 unsigned long *base_size, enum object_type *type, int keep_cache)
1518 void *ret;
1519 unsigned long hash = pack_entry_hash(p, base_offset);
1520 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1522 ret = ent->data;
1523 if (ret && ent->p == p && ent->base_offset == base_offset)
1524 goto found_cache_entry;
1525 return unpack_entry(p, base_offset, type, base_size);
1527 found_cache_entry:
1528 if (!keep_cache) {
1529 ent->data = NULL;
1530 ent->lru.next->prev = ent->lru.prev;
1531 ent->lru.prev->next = ent->lru.next;
1532 delta_base_cached -= ent->size;
1534 else {
1535 ret = xmalloc(ent->size + 1);
1536 memcpy(ret, ent->data, ent->size);
1537 ((char *)ret)[ent->size] = 0;
1539 *type = ent->type;
1540 *base_size = ent->size;
1541 return ret;
1544 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1546 if (ent->data) {
1547 free(ent->data);
1548 ent->data = NULL;
1549 ent->lru.next->prev = ent->lru.prev;
1550 ent->lru.prev->next = ent->lru.next;
1551 delta_base_cached -= ent->size;
1555 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1556 void *base, unsigned long base_size, enum object_type type)
1558 unsigned long hash = pack_entry_hash(p, base_offset);
1559 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1560 struct delta_base_cache_lru_list *lru;
1562 release_delta_base_cache(ent);
1563 delta_base_cached += base_size;
1565 for (lru = delta_base_cache_lru.next;
1566 delta_base_cached > delta_base_cache_limit
1567 && lru != &delta_base_cache_lru;
1568 lru = lru->next) {
1569 struct delta_base_cache_entry *f = (void *)lru;
1570 if (f->type == OBJ_BLOB)
1571 release_delta_base_cache(f);
1573 for (lru = delta_base_cache_lru.next;
1574 delta_base_cached > delta_base_cache_limit
1575 && lru != &delta_base_cache_lru;
1576 lru = lru->next) {
1577 struct delta_base_cache_entry *f = (void *)lru;
1578 release_delta_base_cache(f);
1581 ent->p = p;
1582 ent->base_offset = base_offset;
1583 ent->type = type;
1584 ent->data = base;
1585 ent->size = base_size;
1586 ent->lru.next = &delta_base_cache_lru;
1587 ent->lru.prev = delta_base_cache_lru.prev;
1588 delta_base_cache_lru.prev->next = &ent->lru;
1589 delta_base_cache_lru.prev = &ent->lru;
1592 static void *unpack_delta_entry(struct packed_git *p,
1593 struct pack_window **w_curs,
1594 off_t curpos,
1595 unsigned long delta_size,
1596 off_t obj_offset,
1597 enum object_type *type,
1598 unsigned long *sizep)
1600 void *delta_data, *result, *base;
1601 unsigned long base_size;
1602 off_t base_offset;
1604 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1605 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1606 if (!base)
1607 die("failed to read delta base object"
1608 " at %"PRIuMAX" from %s",
1609 (uintmax_t)base_offset, p->pack_name);
1611 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1612 if (!delta_data)
1613 die("failed to unpack compressed delta"
1614 " at %"PRIuMAX" from %s",
1615 (uintmax_t)curpos, p->pack_name);
1616 result = patch_delta(base, base_size,
1617 delta_data, delta_size,
1618 sizep);
1619 if (!result)
1620 die("failed to apply delta");
1621 free(delta_data);
1622 add_delta_base_cache(p, base_offset, base, base_size, *type);
1623 return result;
1626 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1627 enum object_type *type, unsigned long *sizep)
1629 struct pack_window *w_curs = NULL;
1630 off_t curpos = obj_offset;
1631 void *data;
1633 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1634 switch (*type) {
1635 case OBJ_OFS_DELTA:
1636 case OBJ_REF_DELTA:
1637 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1638 obj_offset, type, sizep);
1639 break;
1640 case OBJ_COMMIT:
1641 case OBJ_TREE:
1642 case OBJ_BLOB:
1643 case OBJ_TAG:
1644 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1645 break;
1646 default:
1647 die("unknown object type %i in %s", *type, p->pack_name);
1649 unuse_pack(&w_curs);
1650 return data;
1653 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1654 uint32_t n)
1656 const unsigned char *index = p->index_data;
1657 if (!index) {
1658 if (open_pack_index(p))
1659 return NULL;
1660 index = p->index_data;
1662 if (n >= p->num_objects)
1663 return NULL;
1664 index += 4 * 256;
1665 if (p->index_version == 1) {
1666 return index + 24 * n + 4;
1667 } else {
1668 index += 8;
1669 return index + 20 * n;
1673 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1675 const unsigned char *index = p->index_data;
1676 index += 4 * 256;
1677 if (p->index_version == 1) {
1678 return ntohl(*((uint32_t *)(index + 24 * n)));
1679 } else if (p->index_version == 2) {
1680 uint32_t off;
1681 index += 8 + p->num_objects * (20 + 4);
1682 off = ntohl(*((uint32_t *)(index + 4 * n)));
1683 if (!(off & 0x80000000))
1684 return off;
1685 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1686 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1687 ntohl(*((uint32_t *)(index + 4)));
1688 } else if (p->index_version == 3) {
1689 uint32_t off;
1690 index += 8 + p->num_objects * 4;
1691 off = ntohl(*((uint32_t *)(index + 4 * n)));
1692 if (!(off & 0x80000000))
1693 return off;
1694 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1695 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1696 ntohl(*((uint32_t *)(index + 4)));
1700 off_t find_pack_entry_one(const unsigned char *sha1,
1701 struct packed_git *p)
1703 const uint32_t *level1_ofs = p->index_data;
1704 const unsigned char *index = p->index_data;
1705 unsigned hi, lo;
1707 if (!index) {
1708 if (open_pack_index(p))
1709 return 0;
1710 level1_ofs = p->index_data;
1711 index = p->index_data;
1713 if (p->index_version > 1) {
1714 level1_ofs += 2;
1715 index += 8;
1718 if (p->index_version > 2) {
1719 open_pack_sha1_dict(p);
1720 index = p->sha1_dict;
1721 } else
1722 index += 4 * 256;
1724 hi = ntohl(level1_ofs[*sha1]);
1725 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1727 do {
1728 unsigned mi = (lo + hi) / 2;
1729 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1730 int cmp = hashcmp(index + x, sha1);
1731 if (!cmp)
1732 return nth_packed_object_offset(p, mi);
1733 if (cmp > 0)
1734 hi = mi;
1735 else
1736 lo = mi+1;
1737 } while (lo < hi);
1738 return 0;
1741 static int matches_pack_name(struct packed_git *p, const char *ig)
1743 const char *last_c, *c;
1745 if (!strcmp(p->pack_name, ig))
1746 return 0;
1748 for (c = p->pack_name, last_c = c; *c;)
1749 if (*c == '/')
1750 last_c = ++c;
1751 else
1752 ++c;
1753 if (!strcmp(last_c, ig))
1754 return 0;
1756 return 1;
1759 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1761 static struct packed_git *last_found = (void *)1;
1762 struct packed_git *p;
1763 off_t offset;
1765 prepare_packed_git();
1766 if (!packed_git)
1767 return 0;
1768 p = (last_found == (void *)1) ? packed_git : last_found;
1770 do {
1771 if (ignore_packed) {
1772 const char **ig;
1773 for (ig = ignore_packed; *ig; ig++)
1774 if (!matches_pack_name(p, *ig))
1775 break;
1776 if (*ig)
1777 goto next;
1780 offset = find_pack_entry_one(sha1, p);
1781 if (offset) {
1783 * We are about to tell the caller where they can
1784 * locate the requested object. We better make
1785 * sure the packfile is still here and can be
1786 * accessed before supplying that answer, as
1787 * it may have been deleted since the index
1788 * was loaded!
1790 if (p->pack_fd == -1 && open_packed_git(p)) {
1791 error("packfile %s cannot be accessed", p->pack_name);
1792 goto next;
1794 e->offset = offset;
1795 e->p = p;
1796 hashcpy(e->sha1, sha1);
1797 last_found = p;
1798 return 1;
1801 next:
1802 if (p == last_found)
1803 p = packed_git;
1804 else
1805 p = p->next;
1806 if (p == last_found)
1807 p = p->next;
1808 } while (p);
1809 return 0;
1812 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1813 struct packed_git *packs)
1815 struct packed_git *p;
1817 for (p = packs; p; p = p->next) {
1818 if (find_pack_entry_one(sha1, p))
1819 return p;
1821 return NULL;
1825 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1827 int status;
1828 unsigned long mapsize, size;
1829 void *map;
1830 z_stream stream;
1831 char hdr[32];
1833 map = map_sha1_file(sha1, &mapsize);
1834 if (!map)
1835 return error("unable to find %s", sha1_to_hex(sha1));
1836 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1837 status = error("unable to unpack %s header",
1838 sha1_to_hex(sha1));
1839 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1840 status = error("unable to parse %s header", sha1_to_hex(sha1));
1841 else if (sizep)
1842 *sizep = size;
1843 inflateEnd(&stream);
1844 munmap(map, mapsize);
1845 return status;
1848 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1850 struct pack_entry e;
1852 if (!find_pack_entry(sha1, &e, NULL)) {
1853 reprepare_packed_git();
1854 if (!find_pack_entry(sha1, &e, NULL))
1855 return sha1_loose_object_info(sha1, sizep);
1857 return packed_object_info(e.p, e.offset, sizep);
1860 static void *read_packed_sha1(const unsigned char *sha1,
1861 enum object_type *type, unsigned long *size)
1863 struct pack_entry e;
1865 if (!find_pack_entry(sha1, &e, NULL))
1866 return NULL;
1867 else
1868 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1872 * This is meant to hold a *small* number of objects that you would
1873 * want read_sha1_file() to be able to return, but yet you do not want
1874 * to write them into the object store (e.g. a browse-only
1875 * application).
1877 static struct cached_object {
1878 unsigned char sha1[20];
1879 enum object_type type;
1880 void *buf;
1881 unsigned long size;
1882 } *cached_objects;
1883 static int cached_object_nr, cached_object_alloc;
1885 static struct cached_object *find_cached_object(const unsigned char *sha1)
1887 int i;
1888 struct cached_object *co = cached_objects;
1890 for (i = 0; i < cached_object_nr; i++, co++) {
1891 if (!hashcmp(co->sha1, sha1))
1892 return co;
1894 return NULL;
1897 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1898 unsigned char *sha1)
1900 struct cached_object *co;
1902 hash_sha1_file(buf, len, typename(type), sha1);
1903 if (has_sha1_file(sha1) || find_cached_object(sha1))
1904 return 0;
1905 if (cached_object_alloc <= cached_object_nr) {
1906 cached_object_alloc = alloc_nr(cached_object_alloc);
1907 cached_objects = xrealloc(cached_objects,
1908 sizeof(*cached_objects) *
1909 cached_object_alloc);
1911 co = &cached_objects[cached_object_nr++];
1912 co->size = len;
1913 co->type = type;
1914 co->buf = xmalloc(len);
1915 memcpy(co->buf, buf, len);
1916 hashcpy(co->sha1, sha1);
1917 return 0;
1920 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1921 unsigned long *size)
1923 unsigned long mapsize;
1924 void *map, *buf;
1925 struct cached_object *co;
1927 co = find_cached_object(sha1);
1928 if (co) {
1929 buf = xmalloc(co->size + 1);
1930 memcpy(buf, co->buf, co->size);
1931 ((char*)buf)[co->size] = 0;
1932 *type = co->type;
1933 *size = co->size;
1934 return buf;
1937 buf = read_packed_sha1(sha1, type, size);
1938 if (buf)
1939 return buf;
1940 map = map_sha1_file(sha1, &mapsize);
1941 if (map) {
1942 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1943 munmap(map, mapsize);
1944 return buf;
1946 reprepare_packed_git();
1947 return read_packed_sha1(sha1, type, size);
1950 void *read_object_with_reference(const unsigned char *sha1,
1951 const char *required_type_name,
1952 unsigned long *size,
1953 unsigned char *actual_sha1_return)
1955 enum object_type type, required_type;
1956 void *buffer;
1957 unsigned long isize;
1958 unsigned char actual_sha1[20];
1960 required_type = type_from_string(required_type_name);
1961 hashcpy(actual_sha1, sha1);
1962 while (1) {
1963 int ref_length = -1;
1964 const char *ref_type = NULL;
1966 buffer = read_sha1_file(actual_sha1, &type, &isize);
1967 if (!buffer)
1968 return NULL;
1969 if (type == required_type) {
1970 *size = isize;
1971 if (actual_sha1_return)
1972 hashcpy(actual_sha1_return, actual_sha1);
1973 return buffer;
1975 /* Handle references */
1976 else if (type == OBJ_COMMIT)
1977 ref_type = "tree ";
1978 else if (type == OBJ_TAG)
1979 ref_type = "object ";
1980 else {
1981 free(buffer);
1982 return NULL;
1984 ref_length = strlen(ref_type);
1986 if (memcmp(buffer, ref_type, ref_length) ||
1987 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1988 free(buffer);
1989 return NULL;
1991 free(buffer);
1992 /* Now we have the ID of the referred-to object in
1993 * actual_sha1. Check again. */
1997 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1998 const char *type, unsigned char *sha1,
1999 char *hdr, int *hdrlen)
2001 SHA_CTX c;
2003 /* Generate the header */
2004 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2006 /* Sha1.. */
2007 SHA1_Init(&c);
2008 SHA1_Update(&c, hdr, *hdrlen);
2009 SHA1_Update(&c, buf, len);
2010 SHA1_Final(sha1, &c);
2014 * Link the tempfile to the final place, possibly creating the
2015 * last directory level as you do so.
2017 * Returns the errno on failure, 0 on success.
2019 static int link_temp_to_file(const char *tmpfile, const char *filename)
2021 int ret;
2022 char *dir;
2024 if (!link(tmpfile, filename))
2025 return 0;
2028 * Try to mkdir the last path component if that failed.
2030 * Re-try the "link()" regardless of whether the mkdir
2031 * succeeds, since a race might mean that somebody
2032 * else succeeded.
2034 ret = errno;
2035 dir = strrchr(filename, '/');
2036 if (dir) {
2037 *dir = 0;
2038 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
2039 *dir = '/';
2040 return -2;
2042 *dir = '/';
2043 if (!link(tmpfile, filename))
2044 return 0;
2045 ret = errno;
2047 return ret;
2051 * Move the just written object into its final resting place
2053 int move_temp_to_file(const char *tmpfile, const char *filename)
2055 int ret = link_temp_to_file(tmpfile, filename);
2058 * Coda hack - coda doesn't like cross-directory links,
2059 * so we fall back to a rename, which will mean that it
2060 * won't be able to check collisions, but that's not a
2061 * big deal.
2063 * The same holds for FAT formatted media.
2065 * When this succeeds, we just return 0. We have nothing
2066 * left to unlink.
2068 if (ret && ret != EEXIST) {
2069 if (!rename(tmpfile, filename))
2070 return 0;
2071 ret = errno;
2073 unlink(tmpfile);
2074 if (ret) {
2075 if (ret != EEXIST) {
2076 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2078 /* FIXME!!! Collision check here ? */
2081 return 0;
2084 static int write_buffer(int fd, const void *buf, size_t len)
2086 if (write_in_full(fd, buf, len) < 0)
2087 return error("file write error (%s)", strerror(errno));
2088 return 0;
2091 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2092 unsigned char *sha1)
2094 char hdr[32];
2095 int hdrlen;
2096 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2097 return 0;
2100 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2102 int size, ret;
2103 unsigned char *compressed;
2104 z_stream stream;
2105 unsigned char sha1[20];
2106 char *filename;
2107 static char tmpfile[PATH_MAX];
2108 char hdr[32];
2109 int fd, hdrlen;
2111 /* Normally if we have it in the pack then we do not bother writing
2112 * it out into .git/objects/??/?{38} file.
2114 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2115 filename = sha1_file_name(sha1);
2116 if (returnsha1)
2117 hashcpy(returnsha1, sha1);
2118 if (has_sha1_file(sha1))
2119 return 0;
2120 fd = open(filename, O_RDONLY);
2121 if (fd >= 0) {
2123 * FIXME!!! We might do collision checking here, but we'd
2124 * need to uncompress the old file and check it. Later.
2126 close(fd);
2127 return 0;
2130 if (errno != ENOENT) {
2131 return error("sha1 file %s: %s\n", filename, strerror(errno));
2134 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2136 fd = mkstemp(tmpfile);
2137 if (fd < 0) {
2138 if (errno == EPERM)
2139 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2140 else
2141 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2144 /* Set it up */
2145 memset(&stream, 0, sizeof(stream));
2146 deflateInit(&stream, zlib_compression_level);
2147 size = 8 + deflateBound(&stream, len+hdrlen);
2148 compressed = xmalloc(size);
2150 /* Compress it */
2151 stream.next_out = compressed;
2152 stream.avail_out = size;
2154 /* First header.. */
2155 stream.next_in = (unsigned char *)hdr;
2156 stream.avail_in = hdrlen;
2157 while (deflate(&stream, 0) == Z_OK)
2158 /* nothing */;
2160 /* Then the data itself.. */
2161 stream.next_in = buf;
2162 stream.avail_in = len;
2163 ret = deflate(&stream, Z_FINISH);
2164 if (ret != Z_STREAM_END)
2165 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2167 ret = deflateEnd(&stream);
2168 if (ret != Z_OK)
2169 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2171 size = stream.total_out;
2173 if (write_buffer(fd, compressed, size) < 0)
2174 die("unable to write sha1 file");
2175 fchmod(fd, 0444);
2176 if (close(fd))
2177 die("unable to write sha1 file");
2178 free(compressed);
2180 return move_temp_to_file(tmpfile, filename);
2184 * We need to unpack and recompress the object for writing
2185 * it out to a different file.
2187 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2189 size_t size;
2190 z_stream stream;
2191 unsigned char *unpacked;
2192 unsigned long len;
2193 enum object_type type;
2194 char hdr[32];
2195 int hdrlen;
2196 void *buf;
2198 /* need to unpack and recompress it by itself */
2199 unpacked = read_packed_sha1(sha1, &type, &len);
2200 if (!unpacked)
2201 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2203 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2205 /* Set it up */
2206 memset(&stream, 0, sizeof(stream));
2207 deflateInit(&stream, zlib_compression_level);
2208 size = deflateBound(&stream, len + hdrlen);
2209 buf = xmalloc(size);
2211 /* Compress it */
2212 stream.next_out = buf;
2213 stream.avail_out = size;
2215 /* First header.. */
2216 stream.next_in = (void *)hdr;
2217 stream.avail_in = hdrlen;
2218 while (deflate(&stream, 0) == Z_OK)
2219 /* nothing */;
2221 /* Then the data itself.. */
2222 stream.next_in = unpacked;
2223 stream.avail_in = len;
2224 while (deflate(&stream, Z_FINISH) == Z_OK)
2225 /* nothing */;
2226 deflateEnd(&stream);
2227 free(unpacked);
2229 *objsize = stream.total_out;
2230 return buf;
2233 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2235 int retval;
2236 unsigned long objsize;
2237 void *buf = map_sha1_file(sha1, &objsize);
2239 if (buf) {
2240 retval = write_buffer(fd, buf, objsize);
2241 munmap(buf, objsize);
2242 return retval;
2245 buf = repack_object(sha1, &objsize);
2246 retval = write_buffer(fd, buf, objsize);
2247 free(buf);
2248 return retval;
2251 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2252 size_t bufsize, size_t *bufposn)
2254 char tmpfile[PATH_MAX];
2255 int local;
2256 z_stream stream;
2257 unsigned char real_sha1[20];
2258 unsigned char discard[4096];
2259 int ret;
2260 SHA_CTX c;
2262 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2264 local = mkstemp(tmpfile);
2265 if (local < 0) {
2266 if (errno == EPERM)
2267 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2268 else
2269 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2272 memset(&stream, 0, sizeof(stream));
2274 inflateInit(&stream);
2276 SHA1_Init(&c);
2278 do {
2279 ssize_t size;
2280 if (*bufposn) {
2281 stream.avail_in = *bufposn;
2282 stream.next_in = (unsigned char *) buffer;
2283 do {
2284 stream.next_out = discard;
2285 stream.avail_out = sizeof(discard);
2286 ret = inflate(&stream, Z_SYNC_FLUSH);
2287 SHA1_Update(&c, discard, sizeof(discard) -
2288 stream.avail_out);
2289 } while (stream.avail_in && ret == Z_OK);
2290 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2291 die("unable to write sha1 file");
2292 memmove(buffer, buffer + *bufposn - stream.avail_in,
2293 stream.avail_in);
2294 *bufposn = stream.avail_in;
2295 if (ret != Z_OK)
2296 break;
2298 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2299 if (size <= 0) {
2300 close(local);
2301 unlink(tmpfile);
2302 if (!size)
2303 return error("Connection closed?");
2304 perror("Reading from connection");
2305 return -1;
2307 *bufposn += size;
2308 } while (1);
2309 inflateEnd(&stream);
2311 fchmod(local, 0444);
2312 if (close(local) != 0)
2313 die("unable to write sha1 file");
2314 SHA1_Final(real_sha1, &c);
2315 if (ret != Z_STREAM_END) {
2316 unlink(tmpfile);
2317 return error("File %s corrupted", sha1_to_hex(sha1));
2319 if (hashcmp(sha1, real_sha1)) {
2320 unlink(tmpfile);
2321 return error("File %s has bad hash", sha1_to_hex(sha1));
2324 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2327 int has_pack_index(const unsigned char *sha1)
2329 struct stat st;
2330 if (stat(sha1_pack_index_name(sha1), &st))
2331 return 0;
2332 return 1;
2335 int has_pack_file(const unsigned char *sha1)
2337 struct stat st;
2338 if (stat(sha1_pack_name(sha1), &st))
2339 return 0;
2340 return 1;
2343 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2345 struct pack_entry e;
2346 return find_pack_entry(sha1, &e, ignore_packed);
2349 int has_sha1_file(const unsigned char *sha1)
2351 struct stat st;
2352 struct pack_entry e;
2354 if (find_pack_entry(sha1, &e, NULL))
2355 return 1;
2356 return find_sha1_file(sha1, &st) ? 1 : 0;
2360 * reads from fd as long as possible into a supplied buffer of size bytes.
2361 * If necessary the buffer's size is increased using realloc()
2363 * returns 0 if anything went fine and -1 otherwise
2365 * The buffer is always NUL-terminated, not including it in returned size.
2367 * NOTE: both buf and size may change, but even when -1 is returned
2368 * you still have to free() it yourself.
2370 int read_fd(int fd, char **return_buf, unsigned long *return_size)
2372 char *buf = *return_buf;
2373 unsigned long size = *return_size;
2374 ssize_t iret;
2375 unsigned long off = 0;
2377 if (!buf || size <= 1) {
2378 size = 1024;
2379 buf = xrealloc(buf, size);
2382 do {
2383 iret = xread(fd, buf + off, (size - 1) - off);
2384 if (iret > 0) {
2385 off += iret;
2386 if (off == size - 1) {
2387 size = alloc_nr(size);
2388 buf = xrealloc(buf, size);
2391 } while (iret > 0);
2393 buf[off] = '\0';
2395 *return_buf = buf;
2396 *return_size = off;
2398 if (iret < 0)
2399 return -1;
2400 return 0;
2403 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2405 unsigned long size = 4096;
2406 char *buf = xmalloc(size);
2407 int ret;
2409 if (read_fd(fd, &buf, &size)) {
2410 free(buf);
2411 return -1;
2414 if (!type)
2415 type = blob_type;
2416 if (write_object)
2417 ret = write_sha1_file(buf, size, type, sha1);
2418 else
2419 ret = hash_sha1_file(buf, size, type, sha1);
2420 free(buf);
2421 return ret;
2424 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2425 enum object_type type, const char *path)
2427 size_t size = xsize_t(st->st_size);
2428 void *buf = NULL;
2429 int ret, re_allocated = 0;
2431 if (size)
2432 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2433 close(fd);
2435 if (!type)
2436 type = OBJ_BLOB;
2439 * Convert blobs to git internal format
2441 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2442 unsigned long nsize = size;
2443 char *nbuf = convert_to_git(path, buf, &nsize);
2444 if (nbuf) {
2445 munmap(buf, size);
2446 size = nsize;
2447 buf = nbuf;
2448 re_allocated = 1;
2452 if (write_object)
2453 ret = write_sha1_file(buf, size, typename(type), sha1);
2454 else
2455 ret = hash_sha1_file(buf, size, typename(type), sha1);
2456 if (re_allocated) {
2457 free(buf);
2458 return ret;
2460 if (size)
2461 munmap(buf, size);
2462 return ret;
2465 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2467 int fd;
2468 char *target;
2469 size_t len;
2471 switch (st->st_mode & S_IFMT) {
2472 case S_IFREG:
2473 fd = open(path, O_RDONLY);
2474 if (fd < 0)
2475 return error("open(\"%s\"): %s", path,
2476 strerror(errno));
2477 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2478 return error("%s: failed to insert into database",
2479 path);
2480 break;
2481 case S_IFLNK:
2482 len = xsize_t(st->st_size);
2483 target = xmalloc(len + 1);
2484 if (readlink(path, target, len + 1) != st->st_size) {
2485 char *errstr = strerror(errno);
2486 free(target);
2487 return error("readlink(\"%s\"): %s", path,
2488 errstr);
2490 if (!write_object)
2491 hash_sha1_file(target, len, blob_type, sha1);
2492 else if (write_sha1_file(target, len, blob_type, sha1))
2493 return error("%s: failed to insert into database",
2494 path);
2495 free(target);
2496 break;
2497 case S_IFDIR:
2498 return resolve_gitlink_ref(path, "HEAD", sha1);
2499 default:
2500 return error("%s: unsupported file type", path);
2502 return 0;
2505 int read_pack_header(int fd, struct pack_header *header)
2507 char *c = (char*)header;
2508 ssize_t remaining = sizeof(struct pack_header);
2509 do {
2510 ssize_t r = xread(fd, c, remaining);
2511 if (r <= 0)
2512 /* "eof before pack header was fully read" */
2513 return PH_ERROR_EOF;
2514 remaining -= r;
2515 c += r;
2516 } while (remaining > 0);
2517 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2518 /* "protocol error (pack signature mismatch detected)" */
2519 return PH_ERROR_PACK_SIGNATURE;
2520 if (!pack_version_ok(header->hdr_version))
2521 /* "protocol error (pack version unsupported)" */
2522 return PH_ERROR_PROTOCOL;
2523 return 0;