Merge branch 'jc/report-tracking'
[git/dscho.git] / sha1_file.c
blob2df78b5afd57cf8e86f01134bf6c49d74defc52a
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"
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
20 #ifndef O_NOATIME
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
23 #else
24 #define O_NOATIME 0
25 #endif
26 #endif
28 #ifdef NO_C99_FORMAT
29 #define SZ_FMT "lu"
30 static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
31 #else
32 #define SZ_FMT "zu"
33 static size_t sz_fmt(size_t s) { return s; }
34 #endif
36 const unsigned char null_sha1[20];
38 const signed char hexval_table[256] = {
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
73 int get_sha1_hex(const char *hex, unsigned char *sha1)
75 int i;
76 for (i = 0; i < 20; i++) {
77 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
78 if (val & ~0xff)
79 return -1;
80 *sha1++ = val;
81 hex += 2;
83 return 0;
86 static inline int offset_1st_component(const char *path)
88 if (has_dos_drive_prefix(path))
89 return 2 + (path[2] == '/');
90 return *path == '/';
93 int safe_create_leading_directories(char *path)
95 char *pos = path + offset_1st_component(path);
96 struct stat st;
98 while (pos) {
99 pos = strchr(pos, '/');
100 if (!pos)
101 break;
102 *pos = 0;
103 if (!stat(path, &st)) {
104 /* path exists */
105 if (!S_ISDIR(st.st_mode)) {
106 *pos = '/';
107 return -3;
110 else if (mkdir(path, 0777)) {
111 *pos = '/';
112 return -1;
114 else if (adjust_shared_perm(path)) {
115 *pos = '/';
116 return -2;
118 *pos++ = '/';
120 return 0;
123 int safe_create_leading_directories_const(const char *path)
125 /* path points to cache entries, so xstrdup before messing with it */
126 char *buf = xstrdup(path);
127 int result = safe_create_leading_directories(buf);
128 free(buf);
129 return result;
132 char *sha1_to_hex(const unsigned char *sha1)
134 static int bufno;
135 static char hexbuffer[4][50];
136 static const char hex[] = "0123456789abcdef";
137 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
138 int i;
140 for (i = 0; i < 20; i++) {
141 unsigned int val = *sha1++;
142 *buf++ = hex[val >> 4];
143 *buf++ = hex[val & 0xf];
145 *buf = '\0';
147 return buffer;
150 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
152 int i;
153 for (i = 0; i < 20; i++) {
154 static char hex[] = "0123456789abcdef";
155 unsigned int val = sha1[i];
156 char *pos = pathbuf + i*2 + (i > 0);
157 *pos++ = hex[val >> 4];
158 *pos = hex[val & 0xf];
163 * NOTE! This returns a statically allocated buffer, so you have to be
164 * careful about using it. Do an "xstrdup()" if you need to save the
165 * filename.
167 * Also note that this returns the location for creating. Reading
168 * SHA1 file can happen from any alternate directory listed in the
169 * DB_ENVIRONMENT environment variable if it is not found in
170 * the primary object database.
172 char *sha1_file_name(const unsigned char *sha1)
174 static char *name, *base;
176 if (!base) {
177 const char *sha1_file_directory = get_object_directory();
178 int len = strlen(sha1_file_directory);
179 base = xmalloc(len + 60);
180 memcpy(base, sha1_file_directory, len);
181 memset(base+len, 0, 60);
182 base[len] = '/';
183 base[len+3] = '/';
184 name = base + len + 1;
186 fill_sha1_path(name, sha1);
187 return base;
190 static char *sha1_get_pack_name(const unsigned char *sha1,
191 char **name, char **base, const char *which)
193 static const char hex[] = "0123456789abcdef";
194 char *buf;
195 int i;
197 if (!*base) {
198 const char *sha1_file_directory = get_object_directory();
199 int len = strlen(sha1_file_directory);
200 *base = xmalloc(len + 60);
201 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
202 sha1_file_directory, which);
203 *name = *base + len + 11;
206 buf = *name;
208 for (i = 0; i < 20; i++) {
209 unsigned int val = *sha1++;
210 *buf++ = hex[val >> 4];
211 *buf++ = hex[val & 0xf];
214 return *base;
217 char *sha1_pack_name(const unsigned char *sha1)
219 static char *name, *base;
221 return sha1_get_pack_name(sha1, &name, &base, "pack");
224 char *sha1_pack_index_name(const unsigned char *sha1)
226 static char *name, *base;
228 return sha1_get_pack_name(sha1, &name, &base, "idx");
231 struct alternate_object_database *alt_odb_list;
232 static struct alternate_object_database **alt_odb_tail;
234 static void read_info_alternates(const char * alternates, int depth);
237 * Prepare alternate object database registry.
239 * The variable alt_odb_list points at the list of struct
240 * alternate_object_database. The elements on this list come from
241 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
242 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
243 * whose contents is similar to that environment variable but can be
244 * LF separated. Its base points at a statically allocated buffer that
245 * contains "/the/directory/corresponding/to/.git/objects/...", while
246 * its name points just after the slash at the end of ".git/objects/"
247 * in the example above, and has enough space to hold 40-byte hex
248 * SHA1, an extra slash for the first level indirection, and the
249 * terminating NUL.
251 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
253 struct stat st;
254 const char *objdir = get_object_directory();
255 struct alternate_object_database *ent;
256 struct alternate_object_database *alt;
257 /* 43 = 40-byte + 2 '/' + terminating NUL */
258 int pfxlen = len;
259 int entlen = pfxlen + 43;
260 int base_len = -1;
262 if (!is_absolute_path(entry) && relative_base) {
263 /* Relative alt-odb */
264 if (base_len < 0)
265 base_len = strlen(relative_base) + 1;
266 entlen += base_len;
267 pfxlen += base_len;
269 ent = xmalloc(sizeof(*ent) + entlen);
271 if (!is_absolute_path(entry) && relative_base) {
272 memcpy(ent->base, relative_base, base_len - 1);
273 ent->base[base_len - 1] = '/';
274 memcpy(ent->base + base_len, entry, len);
276 else
277 memcpy(ent->base, entry, pfxlen);
279 ent->name = ent->base + pfxlen + 1;
280 ent->base[pfxlen + 3] = '/';
281 ent->base[pfxlen] = ent->base[entlen-1] = 0;
283 /* Detect cases where alternate disappeared */
284 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
285 error("object directory %s does not exist; "
286 "check .git/objects/info/alternates.",
287 ent->base);
288 free(ent);
289 return -1;
292 /* Prevent the common mistake of listing the same
293 * thing twice, or object directory itself.
295 for (alt = alt_odb_list; alt; alt = alt->next) {
296 if (!memcmp(ent->base, alt->base, pfxlen)) {
297 free(ent);
298 return -1;
301 if (!memcmp(ent->base, objdir, pfxlen)) {
302 free(ent);
303 return -1;
306 /* add the alternate entry */
307 *alt_odb_tail = ent;
308 alt_odb_tail = &(ent->next);
309 ent->next = NULL;
311 /* recursively add alternates */
312 read_info_alternates(ent->base, depth + 1);
314 ent->base[pfxlen] = '/';
316 return 0;
319 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
320 const char *relative_base, int depth)
322 const char *cp, *last;
324 if (depth > 5) {
325 error("%s: ignoring alternate object stores, nesting too deep.",
326 relative_base);
327 return;
330 last = alt;
331 while (last < ep) {
332 cp = last;
333 if (cp < ep && *cp == '#') {
334 while (cp < ep && *cp != sep)
335 cp++;
336 last = cp + 1;
337 continue;
339 while (cp < ep && *cp != sep)
340 cp++;
341 if (last != cp) {
342 if (!is_absolute_path(last) && depth) {
343 error("%s: ignoring relative alternate object store %s",
344 relative_base, last);
345 } else {
346 link_alt_odb_entry(last, cp - last,
347 relative_base, depth);
350 while (cp < ep && *cp == sep)
351 cp++;
352 last = cp;
356 static void read_info_alternates(const char * relative_base, int depth)
358 char *map;
359 size_t mapsz;
360 struct stat st;
361 const char alt_file_name[] = "info/alternates";
362 /* Given that relative_base is no longer than PATH_MAX,
363 ensure that "path" has enough space to append "/", the
364 file name, "info/alternates", and a trailing NUL. */
365 char path[PATH_MAX + 1 + sizeof alt_file_name];
366 int fd;
368 sprintf(path, "%s/%s", relative_base, alt_file_name);
369 fd = open(path, O_RDONLY);
370 if (fd < 0)
371 return;
372 if (fstat(fd, &st) || (st.st_size == 0)) {
373 close(fd);
374 return;
376 mapsz = xsize_t(st.st_size);
377 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
378 close(fd);
380 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
382 munmap(map, mapsz);
385 void add_to_alternates_file(const char *reference)
387 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
388 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), 1);
389 char *alt = mkpath("%s/objects\n", reference);
390 write_or_die(fd, alt, strlen(alt));
391 if (commit_lock_file(lock))
392 die("could not close alternates file");
393 if (alt_odb_tail)
394 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
397 void prepare_alt_odb(void)
399 const char *alt;
401 if (alt_odb_tail)
402 return;
404 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
405 if (!alt) alt = "";
407 alt_odb_tail = &alt_odb_list;
408 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
410 read_info_alternates(get_object_directory(), 0);
413 static int has_loose_object(const unsigned char *sha1)
415 char *name = sha1_file_name(sha1);
416 struct alternate_object_database *alt;
418 if (!access(name, F_OK))
419 return 1;
420 prepare_alt_odb();
421 for (alt = alt_odb_list; alt; alt = alt->next) {
422 name = alt->name;
423 fill_sha1_path(name, sha1);
424 if (!access(alt->base, F_OK))
425 return 1;
427 return 0;
430 static unsigned int pack_used_ctr;
431 static unsigned int pack_mmap_calls;
432 static unsigned int peak_pack_open_windows;
433 static unsigned int pack_open_windows;
434 static size_t peak_pack_mapped;
435 static size_t pack_mapped;
436 struct packed_git *packed_git;
438 void pack_report(void)
440 fprintf(stderr,
441 "pack_report: getpagesize() = %10" SZ_FMT "\n"
442 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
443 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
444 sz_fmt(getpagesize()),
445 sz_fmt(packed_git_window_size),
446 sz_fmt(packed_git_limit));
447 fprintf(stderr,
448 "pack_report: pack_used_ctr = %10u\n"
449 "pack_report: pack_mmap_calls = %10u\n"
450 "pack_report: pack_open_windows = %10u / %10u\n"
451 "pack_report: pack_mapped = "
452 "%10" SZ_FMT " / %10" SZ_FMT "\n",
453 pack_used_ctr,
454 pack_mmap_calls,
455 pack_open_windows, peak_pack_open_windows,
456 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
459 static int check_packed_git_idx(const char *path, struct packed_git *p)
461 void *idx_map;
462 struct pack_idx_header *hdr;
463 size_t idx_size;
464 uint32_t version, nr, i, *index;
465 int fd = open(path, O_RDONLY);
466 struct stat st;
468 if (fd < 0)
469 return -1;
470 if (fstat(fd, &st)) {
471 close(fd);
472 return -1;
474 idx_size = xsize_t(st.st_size);
475 if (idx_size < 4 * 256 + 20 + 20) {
476 close(fd);
477 return error("index file %s is too small", path);
479 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
480 close(fd);
482 hdr = idx_map;
483 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
484 version = ntohl(hdr->idx_version);
485 if (version < 2 || version > 2) {
486 munmap(idx_map, idx_size);
487 return error("index file %s is version %"PRIu32
488 " and is not supported by this binary"
489 " (try upgrading GIT to a newer version)",
490 path, version);
492 } else
493 version = 1;
495 nr = 0;
496 index = idx_map;
497 if (version > 1)
498 index += 2; /* skip index header */
499 for (i = 0; i < 256; i++) {
500 uint32_t n = ntohl(index[i]);
501 if (n < nr) {
502 munmap(idx_map, idx_size);
503 return error("non-monotonic index %s", path);
505 nr = n;
508 if (version == 1) {
510 * Total size:
511 * - 256 index entries 4 bytes each
512 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
513 * - 20-byte SHA1 of the packfile
514 * - 20-byte SHA1 file checksum
516 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
517 munmap(idx_map, idx_size);
518 return error("wrong index v1 file size in %s", path);
520 } else if (version == 2) {
522 * Minimum size:
523 * - 8 bytes of header
524 * - 256 index entries 4 bytes each
525 * - 20-byte sha1 entry * nr
526 * - 4-byte crc entry * nr
527 * - 4-byte offset entry * nr
528 * - 20-byte SHA1 of the packfile
529 * - 20-byte SHA1 file checksum
530 * And after the 4-byte offset table might be a
531 * variable sized table containing 8-byte entries
532 * for offsets larger than 2^31.
534 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
535 unsigned long max_size = min_size;
536 if (nr)
537 max_size += (nr - 1)*8;
538 if (idx_size < min_size || idx_size > max_size) {
539 munmap(idx_map, idx_size);
540 return error("wrong index v2 file size in %s", path);
542 if (idx_size != min_size &&
544 * make sure we can deal with large pack offsets.
545 * 31-bit signed offset won't be enough, neither
546 * 32-bit unsigned one will be.
548 (sizeof(off_t) <= 4)) {
549 munmap(idx_map, idx_size);
550 return error("pack too large for current definition of off_t in %s", path);
554 p->index_version = version;
555 p->index_data = idx_map;
556 p->index_size = idx_size;
557 p->num_objects = nr;
558 return 0;
561 int open_pack_index(struct packed_git *p)
563 char *idx_name;
564 int ret;
566 if (p->index_data)
567 return 0;
569 idx_name = xstrdup(p->pack_name);
570 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
571 ret = check_packed_git_idx(idx_name, p);
572 free(idx_name);
573 return ret;
576 static void scan_windows(struct packed_git *p,
577 struct packed_git **lru_p,
578 struct pack_window **lru_w,
579 struct pack_window **lru_l)
581 struct pack_window *w, *w_l;
583 for (w_l = NULL, w = p->windows; w; w = w->next) {
584 if (!w->inuse_cnt) {
585 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
586 *lru_p = p;
587 *lru_w = w;
588 *lru_l = w_l;
591 w_l = w;
595 static int unuse_one_window(struct packed_git *current, int keep_fd)
597 struct packed_git *p, *lru_p = NULL;
598 struct pack_window *lru_w = NULL, *lru_l = NULL;
600 if (current)
601 scan_windows(current, &lru_p, &lru_w, &lru_l);
602 for (p = packed_git; p; p = p->next)
603 scan_windows(p, &lru_p, &lru_w, &lru_l);
604 if (lru_p) {
605 munmap(lru_w->base, lru_w->len);
606 pack_mapped -= lru_w->len;
607 if (lru_l)
608 lru_l->next = lru_w->next;
609 else {
610 lru_p->windows = lru_w->next;
611 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
612 close(lru_p->pack_fd);
613 lru_p->pack_fd = -1;
616 free(lru_w);
617 pack_open_windows--;
618 return 1;
620 return 0;
623 void release_pack_memory(size_t need, int fd)
625 size_t cur = pack_mapped;
626 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
627 ; /* nothing */
630 void close_pack_windows(struct packed_git *p)
632 while (p->windows) {
633 struct pack_window *w = p->windows;
635 if (w->inuse_cnt)
636 die("pack '%s' still has open windows to it",
637 p->pack_name);
638 munmap(w->base, w->len);
639 pack_mapped -= w->len;
640 pack_open_windows--;
641 p->windows = w->next;
642 free(w);
646 void unuse_pack(struct pack_window **w_cursor)
648 struct pack_window *w = *w_cursor;
649 if (w) {
650 w->inuse_cnt--;
651 *w_cursor = NULL;
656 * Do not call this directly as this leaks p->pack_fd on error return;
657 * call open_packed_git() instead.
659 static int open_packed_git_1(struct packed_git *p)
661 struct stat st;
662 struct pack_header hdr;
663 unsigned char sha1[20];
664 unsigned char *idx_sha1;
665 long fd_flag;
667 if (!p->index_data && open_pack_index(p))
668 return error("packfile %s index unavailable", p->pack_name);
670 p->pack_fd = open(p->pack_name, O_RDONLY);
671 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
672 return -1;
674 /* If we created the struct before we had the pack we lack size. */
675 if (!p->pack_size) {
676 if (!S_ISREG(st.st_mode))
677 return error("packfile %s not a regular file", p->pack_name);
678 p->pack_size = st.st_size;
679 } else if (p->pack_size != st.st_size)
680 return error("packfile %s size changed", p->pack_name);
682 /* We leave these file descriptors open with sliding mmap;
683 * there is no point keeping them open across exec(), though.
685 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
686 if (fd_flag < 0)
687 return error("cannot determine file descriptor flags");
688 fd_flag |= FD_CLOEXEC;
689 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
690 return error("cannot set FD_CLOEXEC");
692 /* Verify we recognize this pack file format. */
693 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
694 return error("file %s is far too short to be a packfile", p->pack_name);
695 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
696 return error("file %s is not a GIT packfile", p->pack_name);
697 if (!pack_version_ok(hdr.hdr_version))
698 return error("packfile %s is version %"PRIu32" and not"
699 " supported (try upgrading GIT to a newer version)",
700 p->pack_name, ntohl(hdr.hdr_version));
702 /* Verify the pack matches its index. */
703 if (p->num_objects != ntohl(hdr.hdr_entries))
704 return error("packfile %s claims to have %"PRIu32" objects"
705 " while index indicates %"PRIu32" objects",
706 p->pack_name, ntohl(hdr.hdr_entries),
707 p->num_objects);
708 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
709 return error("end of packfile %s is unavailable", p->pack_name);
710 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
711 return error("packfile %s signature is unavailable", p->pack_name);
712 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
713 if (hashcmp(sha1, idx_sha1))
714 return error("packfile %s does not match index", p->pack_name);
715 return 0;
718 static int open_packed_git(struct packed_git *p)
720 if (!open_packed_git_1(p))
721 return 0;
722 if (p->pack_fd != -1) {
723 close(p->pack_fd);
724 p->pack_fd = -1;
726 return -1;
729 static int in_window(struct pack_window *win, off_t offset)
731 /* We must promise at least 20 bytes (one hash) after the
732 * offset is available from this window, otherwise the offset
733 * is not actually in this window and a different window (which
734 * has that one hash excess) must be used. This is to support
735 * the object header and delta base parsing routines below.
737 off_t win_off = win->offset;
738 return win_off <= offset
739 && (offset + 20) <= (win_off + win->len);
742 unsigned char* use_pack(struct packed_git *p,
743 struct pack_window **w_cursor,
744 off_t offset,
745 unsigned int *left)
747 struct pack_window *win = *w_cursor;
749 if (p->pack_fd == -1 && open_packed_git(p))
750 die("packfile %s cannot be accessed", p->pack_name);
752 /* Since packfiles end in a hash of their content and its
753 * pointless to ask for an offset into the middle of that
754 * hash, and the in_window function above wouldn't match
755 * don't allow an offset too close to the end of the file.
757 if (offset > (p->pack_size - 20))
758 die("offset beyond end of packfile (truncated pack?)");
760 if (!win || !in_window(win, offset)) {
761 if (win)
762 win->inuse_cnt--;
763 for (win = p->windows; win; win = win->next) {
764 if (in_window(win, offset))
765 break;
767 if (!win) {
768 size_t window_align = packed_git_window_size / 2;
769 off_t len;
770 win = xcalloc(1, sizeof(*win));
771 win->offset = (offset / window_align) * window_align;
772 len = p->pack_size - win->offset;
773 if (len > packed_git_window_size)
774 len = packed_git_window_size;
775 win->len = (size_t)len;
776 pack_mapped += win->len;
777 while (packed_git_limit < pack_mapped
778 && unuse_one_window(p, p->pack_fd))
779 ; /* nothing */
780 win->base = xmmap(NULL, win->len,
781 PROT_READ, MAP_PRIVATE,
782 p->pack_fd, win->offset);
783 if (win->base == MAP_FAILED)
784 die("packfile %s cannot be mapped: %s",
785 p->pack_name,
786 strerror(errno));
787 pack_mmap_calls++;
788 pack_open_windows++;
789 if (pack_mapped > peak_pack_mapped)
790 peak_pack_mapped = pack_mapped;
791 if (pack_open_windows > peak_pack_open_windows)
792 peak_pack_open_windows = pack_open_windows;
793 win->next = p->windows;
794 p->windows = win;
797 if (win != *w_cursor) {
798 win->last_used = pack_used_ctr++;
799 win->inuse_cnt++;
800 *w_cursor = win;
802 offset -= win->offset;
803 if (left)
804 *left = win->len - xsize_t(offset);
805 return win->base + offset;
808 static struct packed_git *alloc_packed_git(int extra)
810 struct packed_git *p = xmalloc(sizeof(*p) + extra);
811 memset(p, 0, sizeof(*p));
812 p->pack_fd = -1;
813 return p;
816 struct packed_git *add_packed_git(const char *path, int path_len, int local)
818 struct stat st;
819 struct packed_git *p = alloc_packed_git(path_len + 2);
822 * Make sure a corresponding .pack file exists and that
823 * the index looks sane.
825 path_len -= strlen(".idx");
826 if (path_len < 1) {
827 free(p);
828 return NULL;
830 memcpy(p->pack_name, path, path_len);
831 strcpy(p->pack_name + path_len, ".pack");
832 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
833 free(p);
834 return NULL;
837 /* ok, it looks sane as far as we can check without
838 * actually mapping the pack file.
840 p->pack_size = st.st_size;
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 const char *idx_path = sha1_pack_index_name(sha1);
851 const char *path = sha1_pack_name(sha1);
852 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
854 strcpy(p->pack_name, path);
855 hashcpy(p->sha1, sha1);
856 if (check_packed_git_idx(idx_path, p)) {
857 free(p);
858 return NULL;
861 return p;
864 void install_packed_git(struct packed_git *pack)
866 pack->next = packed_git;
867 packed_git = pack;
870 static void prepare_packed_git_one(char *objdir, int local)
872 /* Ensure that this buffer is large enough so that we can
873 append "/pack/" without clobbering the stack even if
874 strlen(objdir) were PATH_MAX. */
875 char path[PATH_MAX + 1 + 4 + 1 + 1];
876 int len;
877 DIR *dir;
878 struct dirent *de;
880 sprintf(path, "%s/pack", objdir);
881 len = strlen(path);
882 dir = opendir(path);
883 if (!dir) {
884 if (errno != ENOENT)
885 error("unable to open object pack directory: %s: %s",
886 path, strerror(errno));
887 return;
889 path[len++] = '/';
890 while ((de = readdir(dir)) != NULL) {
891 int namelen = strlen(de->d_name);
892 struct packed_git *p;
894 if (!has_extension(de->d_name, ".idx"))
895 continue;
897 if (len + namelen + 1 > sizeof(path))
898 continue;
900 /* Don't reopen a pack we already have. */
901 strcpy(path + len, de->d_name);
902 for (p = packed_git; p; p = p->next) {
903 if (!memcmp(path, p->pack_name, len + namelen - 4))
904 break;
906 if (p)
907 continue;
908 /* See if it really is a valid .idx file with corresponding
909 * .pack file that we can map.
911 p = add_packed_git(path, len + namelen, local);
912 if (!p)
913 continue;
914 install_packed_git(p);
916 closedir(dir);
919 static int sort_pack(const void *a_, const void *b_)
921 struct packed_git *a = *((struct packed_git **)a_);
922 struct packed_git *b = *((struct packed_git **)b_);
923 int st;
926 * Local packs tend to contain objects specific to our
927 * variant of the project than remote ones. In addition,
928 * remote ones could be on a network mounted filesystem.
929 * Favor local ones for these reasons.
931 st = a->pack_local - b->pack_local;
932 if (st)
933 return -st;
936 * Younger packs tend to contain more recent objects,
937 * and more recent objects tend to get accessed more
938 * often.
940 if (a->mtime < b->mtime)
941 return 1;
942 else if (a->mtime == b->mtime)
943 return 0;
944 return -1;
947 static void rearrange_packed_git(void)
949 struct packed_git **ary, *p;
950 int i, n;
952 for (n = 0, p = packed_git; p; p = p->next)
953 n++;
954 if (n < 2)
955 return;
957 /* prepare an array of packed_git for easier sorting */
958 ary = xcalloc(n, sizeof(struct packed_git *));
959 for (n = 0, p = packed_git; p; p = p->next)
960 ary[n++] = p;
962 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
964 /* link them back again */
965 for (i = 0; i < n - 1; i++)
966 ary[i]->next = ary[i + 1];
967 ary[n - 1]->next = NULL;
968 packed_git = ary[0];
970 free(ary);
973 static int prepare_packed_git_run_once = 0;
974 void prepare_packed_git(void)
976 struct alternate_object_database *alt;
978 if (prepare_packed_git_run_once)
979 return;
980 prepare_packed_git_one(get_object_directory(), 1);
981 prepare_alt_odb();
982 for (alt = alt_odb_list; alt; alt = alt->next) {
983 alt->name[-1] = 0;
984 prepare_packed_git_one(alt->base, 0);
985 alt->name[-1] = '/';
987 rearrange_packed_git();
988 prepare_packed_git_run_once = 1;
991 void reprepare_packed_git(void)
993 prepare_packed_git_run_once = 0;
994 prepare_packed_git();
997 static void mark_bad_packed_object(struct packed_git *p,
998 const unsigned char *sha1)
1000 unsigned i;
1001 for (i = 0; i < p->num_bad_objects; i++)
1002 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1003 return;
1004 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1005 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1006 p->num_bad_objects++;
1009 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1011 unsigned char real_sha1[20];
1012 hash_sha1_file(map, size, type, real_sha1);
1013 return hashcmp(sha1, real_sha1) ? -1 : 0;
1016 static int git_open_noatime(const char *name)
1018 static int sha1_file_open_flag = O_NOATIME;
1019 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1021 /* Might the failure be due to O_NOATIME? */
1022 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1023 fd = open(name, O_RDONLY);
1024 if (fd >= 0)
1025 sha1_file_open_flag = 0;
1027 return fd;
1030 static int open_sha1_file(const unsigned char *sha1)
1032 int fd;
1033 char *name = sha1_file_name(sha1);
1034 struct alternate_object_database *alt;
1036 fd = git_open_noatime(name);
1037 if (fd >= 0)
1038 return fd;
1040 prepare_alt_odb();
1041 errno = ENOENT;
1042 for (alt = alt_odb_list; alt; alt = alt->next) {
1043 name = alt->name;
1044 fill_sha1_path(name, sha1);
1045 fd = git_open_noatime(alt->base);
1046 if (fd >= 0)
1047 return fd;
1049 return -1;
1052 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1054 void *map;
1055 int fd;
1057 fd = open_sha1_file(sha1);
1058 map = NULL;
1059 if (fd >= 0) {
1060 struct stat st;
1062 if (!fstat(fd, &st)) {
1063 *size = xsize_t(st.st_size);
1064 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1066 close(fd);
1068 return map;
1071 static int legacy_loose_object(unsigned char *map)
1073 unsigned int word;
1076 * Is it a zlib-compressed buffer? If so, the first byte
1077 * must be 0x78 (15-bit window size, deflated), and the
1078 * first 16-bit word is evenly divisible by 31
1080 word = (map[0] << 8) + map[1];
1081 if (map[0] == 0x78 && !(word % 31))
1082 return 1;
1083 else
1084 return 0;
1087 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1089 unsigned shift;
1090 unsigned char c;
1091 unsigned long size;
1092 unsigned long used = 0;
1094 c = buf[used++];
1095 *type = (c >> 4) & 7;
1096 size = c & 15;
1097 shift = 4;
1098 while (c & 0x80) {
1099 if (len <= used)
1100 return 0;
1101 if (sizeof(long) * 8 <= shift)
1102 return 0;
1103 c = buf[used++];
1104 size += (c & 0x7f) << shift;
1105 shift += 7;
1107 *sizep = size;
1108 return used;
1111 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1113 unsigned long size, used;
1114 static const char valid_loose_object_type[8] = {
1115 0, /* OBJ_EXT */
1116 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1117 0, /* "delta" and others are invalid in a loose object */
1119 enum object_type type;
1121 /* Get the data stream */
1122 memset(stream, 0, sizeof(*stream));
1123 stream->next_in = map;
1124 stream->avail_in = mapsize;
1125 stream->next_out = buffer;
1126 stream->avail_out = bufsiz;
1128 if (legacy_loose_object(map)) {
1129 inflateInit(stream);
1130 return inflate(stream, 0);
1135 * There used to be a second loose object header format which
1136 * was meant to mimic the in-pack format, allowing for direct
1137 * copy of the object data. This format turned up not to be
1138 * really worth it and we don't write it any longer. But we
1139 * can still read it.
1141 used = unpack_object_header_gently(map, mapsize, &type, &size);
1142 if (!used || !valid_loose_object_type[type])
1143 return -1;
1144 map += used;
1145 mapsize -= used;
1147 /* Set up the stream for the rest.. */
1148 stream->next_in = map;
1149 stream->avail_in = mapsize;
1150 inflateInit(stream);
1152 /* And generate the fake traditional header */
1153 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1154 typename(type), size);
1155 return 0;
1158 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1160 int bytes = strlen(buffer) + 1;
1161 unsigned char *buf = xmalloc(1+size);
1162 unsigned long n;
1163 int status = Z_OK;
1165 n = stream->total_out - bytes;
1166 if (n > size)
1167 n = size;
1168 memcpy(buf, (char *) buffer + bytes, n);
1169 bytes = n;
1170 if (bytes <= size) {
1172 * The above condition must be (bytes <= size), not
1173 * (bytes < size). In other words, even though we
1174 * expect no more output and set avail_out to zer0,
1175 * the input zlib stream may have bytes that express
1176 * "this concludes the stream", and we *do* want to
1177 * eat that input.
1179 * Otherwise we would not be able to test that we
1180 * consumed all the input to reach the expected size;
1181 * we also want to check that zlib tells us that all
1182 * went well with status == Z_STREAM_END at the end.
1184 stream->next_out = buf + bytes;
1185 stream->avail_out = size - bytes;
1186 while (status == Z_OK)
1187 status = inflate(stream, Z_FINISH);
1189 buf[size] = 0;
1190 if (status == Z_STREAM_END && !stream->avail_in) {
1191 inflateEnd(stream);
1192 return buf;
1195 if (status < 0)
1196 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1197 else if (stream->avail_in)
1198 error("garbage at end of loose object '%s'",
1199 sha1_to_hex(sha1));
1200 free(buf);
1201 return NULL;
1205 * We used to just use "sscanf()", but that's actually way
1206 * too permissive for what we want to check. So do an anal
1207 * object header parse by hand.
1209 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1211 char type[10];
1212 int i;
1213 unsigned long size;
1216 * The type can be at most ten bytes (including the
1217 * terminating '\0' that we add), and is followed by
1218 * a space.
1220 i = 0;
1221 for (;;) {
1222 char c = *hdr++;
1223 if (c == ' ')
1224 break;
1225 type[i++] = c;
1226 if (i >= sizeof(type))
1227 return -1;
1229 type[i] = 0;
1232 * The length must follow immediately, and be in canonical
1233 * decimal format (ie "010" is not valid).
1235 size = *hdr++ - '0';
1236 if (size > 9)
1237 return -1;
1238 if (size) {
1239 for (;;) {
1240 unsigned long c = *hdr - '0';
1241 if (c > 9)
1242 break;
1243 hdr++;
1244 size = size * 10 + c;
1247 *sizep = size;
1250 * The length must be followed by a zero byte
1252 return *hdr ? -1 : type_from_string(type);
1255 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1257 int ret;
1258 z_stream stream;
1259 char hdr[8192];
1261 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1262 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1263 return NULL;
1265 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1268 unsigned long get_size_from_delta(struct packed_git *p,
1269 struct pack_window **w_curs,
1270 off_t curpos)
1272 const unsigned char *data;
1273 unsigned char delta_head[20], *in;
1274 z_stream stream;
1275 int st;
1277 memset(&stream, 0, sizeof(stream));
1278 stream.next_out = delta_head;
1279 stream.avail_out = sizeof(delta_head);
1281 inflateInit(&stream);
1282 do {
1283 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1284 stream.next_in = in;
1285 st = inflate(&stream, Z_FINISH);
1286 curpos += stream.next_in - in;
1287 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1288 stream.total_out < sizeof(delta_head));
1289 inflateEnd(&stream);
1290 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1291 die("delta data unpack-initial failed");
1293 /* Examine the initial part of the delta to figure out
1294 * the result size.
1296 data = delta_head;
1298 /* ignore base size */
1299 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1301 /* Read the result size */
1302 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1305 static off_t get_delta_base(struct packed_git *p,
1306 struct pack_window **w_curs,
1307 off_t *curpos,
1308 enum object_type type,
1309 off_t delta_obj_offset)
1311 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1312 off_t base_offset;
1314 /* use_pack() assured us we have [base_info, base_info + 20)
1315 * as a range that we can look at without walking off the
1316 * end of the mapped window. Its actually the hash size
1317 * that is assured. An OFS_DELTA longer than the hash size
1318 * is stupid, as then a REF_DELTA would be smaller to store.
1320 if (type == OBJ_OFS_DELTA) {
1321 unsigned used = 0;
1322 unsigned char c = base_info[used++];
1323 base_offset = c & 127;
1324 while (c & 128) {
1325 base_offset += 1;
1326 if (!base_offset || MSB(base_offset, 7))
1327 return 0; /* overflow */
1328 c = base_info[used++];
1329 base_offset = (base_offset << 7) + (c & 127);
1331 base_offset = delta_obj_offset - base_offset;
1332 if (base_offset >= delta_obj_offset)
1333 return 0; /* out of bound */
1334 *curpos += used;
1335 } else if (type == OBJ_REF_DELTA) {
1336 /* The base entry _must_ be in the same pack */
1337 base_offset = find_pack_entry_one(base_info, p);
1338 *curpos += 20;
1339 } else
1340 die("I am totally screwed");
1341 return base_offset;
1344 /* forward declaration for a mutually recursive function */
1345 static int packed_object_info(struct packed_git *p, off_t offset,
1346 unsigned long *sizep);
1348 static int packed_delta_info(struct packed_git *p,
1349 struct pack_window **w_curs,
1350 off_t curpos,
1351 enum object_type type,
1352 off_t obj_offset,
1353 unsigned long *sizep)
1355 off_t base_offset;
1357 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1358 type = packed_object_info(p, base_offset, NULL);
1360 /* We choose to only get the type of the base object and
1361 * ignore potentially corrupt pack file that expects the delta
1362 * based on a base with a wrong size. This saves tons of
1363 * inflate() calls.
1365 if (sizep)
1366 *sizep = get_size_from_delta(p, w_curs, curpos);
1368 return type;
1371 static int unpack_object_header(struct packed_git *p,
1372 struct pack_window **w_curs,
1373 off_t *curpos,
1374 unsigned long *sizep)
1376 unsigned char *base;
1377 unsigned int left;
1378 unsigned long used;
1379 enum object_type type;
1381 /* use_pack() assures us we have [base, base + 20) available
1382 * as a range that we can look at at. (Its actually the hash
1383 * size that is assured.) With our object header encoding
1384 * the maximum deflated object size is 2^137, which is just
1385 * insane, so we know won't exceed what we have been given.
1387 base = use_pack(p, w_curs, *curpos, &left);
1388 used = unpack_object_header_gently(base, left, &type, sizep);
1389 if (!used)
1390 die("object offset outside of pack file");
1391 *curpos += used;
1393 return type;
1396 const char *packed_object_info_detail(struct packed_git *p,
1397 off_t obj_offset,
1398 unsigned long *size,
1399 unsigned long *store_size,
1400 unsigned int *delta_chain_length,
1401 unsigned char *base_sha1)
1403 struct pack_window *w_curs = NULL;
1404 off_t curpos;
1405 unsigned long dummy;
1406 unsigned char *next_sha1;
1407 enum object_type type;
1408 struct revindex_entry *revidx;
1410 *delta_chain_length = 0;
1411 curpos = obj_offset;
1412 type = unpack_object_header(p, &w_curs, &curpos, size);
1414 revidx = find_pack_revindex(p, obj_offset);
1415 *store_size = revidx[1].offset - obj_offset;
1417 for (;;) {
1418 switch (type) {
1419 default:
1420 die("pack %s contains unknown object type %d",
1421 p->pack_name, type);
1422 case OBJ_COMMIT:
1423 case OBJ_TREE:
1424 case OBJ_BLOB:
1425 case OBJ_TAG:
1426 unuse_pack(&w_curs);
1427 return typename(type);
1428 case OBJ_OFS_DELTA:
1429 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1430 if (!obj_offset)
1431 die("pack %s contains bad delta base reference of type %s",
1432 p->pack_name, typename(type));
1433 if (*delta_chain_length == 0) {
1434 revidx = find_pack_revindex(p, obj_offset);
1435 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1437 break;
1438 case OBJ_REF_DELTA:
1439 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1440 if (*delta_chain_length == 0)
1441 hashcpy(base_sha1, next_sha1);
1442 obj_offset = find_pack_entry_one(next_sha1, p);
1443 break;
1445 (*delta_chain_length)++;
1446 curpos = obj_offset;
1447 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1451 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1452 unsigned long *sizep)
1454 struct pack_window *w_curs = NULL;
1455 unsigned long size;
1456 off_t curpos = obj_offset;
1457 enum object_type type;
1459 type = unpack_object_header(p, &w_curs, &curpos, &size);
1461 switch (type) {
1462 case OBJ_OFS_DELTA:
1463 case OBJ_REF_DELTA:
1464 type = packed_delta_info(p, &w_curs, curpos,
1465 type, obj_offset, sizep);
1466 break;
1467 case OBJ_COMMIT:
1468 case OBJ_TREE:
1469 case OBJ_BLOB:
1470 case OBJ_TAG:
1471 if (sizep)
1472 *sizep = size;
1473 break;
1474 default:
1475 die("pack %s contains unknown object type %d",
1476 p->pack_name, type);
1478 unuse_pack(&w_curs);
1479 return type;
1482 static void *unpack_compressed_entry(struct packed_git *p,
1483 struct pack_window **w_curs,
1484 off_t curpos,
1485 unsigned long size)
1487 int st;
1488 z_stream stream;
1489 unsigned char *buffer, *in;
1491 buffer = xmalloc(size + 1);
1492 buffer[size] = 0;
1493 memset(&stream, 0, sizeof(stream));
1494 stream.next_out = buffer;
1495 stream.avail_out = size;
1497 inflateInit(&stream);
1498 do {
1499 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1500 stream.next_in = in;
1501 st = inflate(&stream, Z_FINISH);
1502 curpos += stream.next_in - in;
1503 } while (st == Z_OK || st == Z_BUF_ERROR);
1504 inflateEnd(&stream);
1505 if ((st != Z_STREAM_END) || stream.total_out != size) {
1506 free(buffer);
1507 return NULL;
1510 return buffer;
1513 #define MAX_DELTA_CACHE (256)
1515 static size_t delta_base_cached;
1517 static struct delta_base_cache_lru_list {
1518 struct delta_base_cache_lru_list *prev;
1519 struct delta_base_cache_lru_list *next;
1520 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1522 static struct delta_base_cache_entry {
1523 struct delta_base_cache_lru_list lru;
1524 void *data;
1525 struct packed_git *p;
1526 off_t base_offset;
1527 unsigned long size;
1528 enum object_type type;
1529 } delta_base_cache[MAX_DELTA_CACHE];
1531 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1533 unsigned long hash;
1535 hash = (unsigned long)p + (unsigned long)base_offset;
1536 hash += (hash >> 8) + (hash >> 16);
1537 return hash % MAX_DELTA_CACHE;
1540 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1541 unsigned long *base_size, enum object_type *type, int keep_cache)
1543 void *ret;
1544 unsigned long hash = pack_entry_hash(p, base_offset);
1545 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1547 ret = ent->data;
1548 if (ret && ent->p == p && ent->base_offset == base_offset)
1549 goto found_cache_entry;
1550 return unpack_entry(p, base_offset, type, base_size);
1552 found_cache_entry:
1553 if (!keep_cache) {
1554 ent->data = NULL;
1555 ent->lru.next->prev = ent->lru.prev;
1556 ent->lru.prev->next = ent->lru.next;
1557 delta_base_cached -= ent->size;
1558 } else {
1559 ret = xmemdupz(ent->data, ent->size);
1561 *type = ent->type;
1562 *base_size = ent->size;
1563 return ret;
1566 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1568 if (ent->data) {
1569 free(ent->data);
1570 ent->data = NULL;
1571 ent->lru.next->prev = ent->lru.prev;
1572 ent->lru.prev->next = ent->lru.next;
1573 delta_base_cached -= ent->size;
1577 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1578 void *base, unsigned long base_size, enum object_type type)
1580 unsigned long hash = pack_entry_hash(p, base_offset);
1581 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1582 struct delta_base_cache_lru_list *lru;
1584 release_delta_base_cache(ent);
1585 delta_base_cached += base_size;
1587 for (lru = delta_base_cache_lru.next;
1588 delta_base_cached > delta_base_cache_limit
1589 && lru != &delta_base_cache_lru;
1590 lru = lru->next) {
1591 struct delta_base_cache_entry *f = (void *)lru;
1592 if (f->type == OBJ_BLOB)
1593 release_delta_base_cache(f);
1595 for (lru = delta_base_cache_lru.next;
1596 delta_base_cached > delta_base_cache_limit
1597 && lru != &delta_base_cache_lru;
1598 lru = lru->next) {
1599 struct delta_base_cache_entry *f = (void *)lru;
1600 release_delta_base_cache(f);
1603 ent->p = p;
1604 ent->base_offset = base_offset;
1605 ent->type = type;
1606 ent->data = base;
1607 ent->size = base_size;
1608 ent->lru.next = &delta_base_cache_lru;
1609 ent->lru.prev = delta_base_cache_lru.prev;
1610 delta_base_cache_lru.prev->next = &ent->lru;
1611 delta_base_cache_lru.prev = &ent->lru;
1614 static void *unpack_delta_entry(struct packed_git *p,
1615 struct pack_window **w_curs,
1616 off_t curpos,
1617 unsigned long delta_size,
1618 off_t obj_offset,
1619 enum object_type *type,
1620 unsigned long *sizep)
1622 void *delta_data, *result, *base;
1623 unsigned long base_size;
1624 off_t base_offset;
1626 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1627 if (!base_offset) {
1628 error("failed to validate delta base reference "
1629 "at offset %"PRIuMAX" from %s",
1630 (uintmax_t)curpos, p->pack_name);
1631 return NULL;
1633 unuse_pack(w_curs);
1634 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1635 if (!base) {
1637 * We're probably in deep shit, but let's try to fetch
1638 * the required base anyway from another pack or loose.
1639 * This is costly but should happen only in the presence
1640 * of a corrupted pack, and is better than failing outright.
1642 struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1643 const unsigned char *base_sha1 =
1644 nth_packed_object_sha1(p, revidx->nr);
1645 error("failed to read delta base object %s"
1646 " at offset %"PRIuMAX" from %s",
1647 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1648 p->pack_name);
1649 mark_bad_packed_object(p, base_sha1);
1650 base = read_sha1_file(base_sha1, type, &base_size);
1651 if (!base)
1652 return NULL;
1655 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1656 if (!delta_data) {
1657 error("failed to unpack compressed delta "
1658 "at offset %"PRIuMAX" from %s",
1659 (uintmax_t)curpos, p->pack_name);
1660 free(base);
1661 return NULL;
1663 result = patch_delta(base, base_size,
1664 delta_data, delta_size,
1665 sizep);
1666 if (!result)
1667 die("failed to apply delta");
1668 free(delta_data);
1669 add_delta_base_cache(p, base_offset, base, base_size, *type);
1670 return result;
1673 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1674 enum object_type *type, unsigned long *sizep)
1676 struct pack_window *w_curs = NULL;
1677 off_t curpos = obj_offset;
1678 void *data;
1680 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1681 switch (*type) {
1682 case OBJ_OFS_DELTA:
1683 case OBJ_REF_DELTA:
1684 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1685 obj_offset, type, sizep);
1686 break;
1687 case OBJ_COMMIT:
1688 case OBJ_TREE:
1689 case OBJ_BLOB:
1690 case OBJ_TAG:
1691 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1692 break;
1693 default:
1694 data = NULL;
1695 error("unknown object type %i at offset %"PRIuMAX" in %s",
1696 *type, (uintmax_t)obj_offset, p->pack_name);
1698 unuse_pack(&w_curs);
1699 return data;
1702 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1703 uint32_t n)
1705 const unsigned char *index = p->index_data;
1706 if (!index) {
1707 if (open_pack_index(p))
1708 return NULL;
1709 index = p->index_data;
1711 if (n >= p->num_objects)
1712 return NULL;
1713 index += 4 * 256;
1714 if (p->index_version == 1) {
1715 return index + 24 * n + 4;
1716 } else {
1717 index += 8;
1718 return index + 20 * n;
1722 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1724 const unsigned char *index = p->index_data;
1725 index += 4 * 256;
1726 if (p->index_version == 1) {
1727 return ntohl(*((uint32_t *)(index + 24 * n)));
1728 } else {
1729 uint32_t off;
1730 index += 8 + p->num_objects * (20 + 4);
1731 off = ntohl(*((uint32_t *)(index + 4 * n)));
1732 if (!(off & 0x80000000))
1733 return off;
1734 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1735 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1736 ntohl(*((uint32_t *)(index + 4)));
1740 off_t find_pack_entry_one(const unsigned char *sha1,
1741 struct packed_git *p)
1743 const uint32_t *level1_ofs = p->index_data;
1744 const unsigned char *index = p->index_data;
1745 unsigned hi, lo, stride;
1746 static int use_lookup = -1;
1747 static int debug_lookup = -1;
1749 if (debug_lookup < 0)
1750 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1752 if (!index) {
1753 if (open_pack_index(p))
1754 return 0;
1755 level1_ofs = p->index_data;
1756 index = p->index_data;
1758 if (p->index_version > 1) {
1759 level1_ofs += 2;
1760 index += 8;
1762 index += 4 * 256;
1763 hi = ntohl(level1_ofs[*sha1]);
1764 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1765 if (p->index_version > 1) {
1766 stride = 20;
1767 } else {
1768 stride = 24;
1769 index += 4;
1772 if (debug_lookup)
1773 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1774 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1776 if (use_lookup < 0)
1777 use_lookup = !!getenv("GIT_USE_LOOKUP");
1778 if (use_lookup) {
1779 int pos = sha1_entry_pos(index, stride, 0,
1780 lo, hi, p->num_objects, sha1);
1781 if (pos < 0)
1782 return 0;
1783 return nth_packed_object_offset(p, pos);
1786 do {
1787 unsigned mi = (lo + hi) / 2;
1788 int cmp = hashcmp(index + mi * stride, sha1);
1790 if (debug_lookup)
1791 printf("lo %u hi %u rg %u mi %u\n",
1792 lo, hi, hi - lo, mi);
1793 if (!cmp)
1794 return nth_packed_object_offset(p, mi);
1795 if (cmp > 0)
1796 hi = mi;
1797 else
1798 lo = mi+1;
1799 } while (lo < hi);
1800 return 0;
1803 int matches_pack_name(struct packed_git *p, const char *name)
1805 const char *last_c, *c;
1807 if (!strcmp(p->pack_name, name))
1808 return 1;
1810 for (c = p->pack_name, last_c = c; *c;)
1811 if (*c == '/')
1812 last_c = ++c;
1813 else
1814 ++c;
1815 if (!strcmp(last_c, name))
1816 return 1;
1818 return 0;
1821 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1823 static struct packed_git *last_found = (void *)1;
1824 struct packed_git *p;
1825 off_t offset;
1827 prepare_packed_git();
1828 if (!packed_git)
1829 return 0;
1830 p = (last_found == (void *)1) ? packed_git : last_found;
1832 do {
1833 if (ignore_packed) {
1834 const char **ig;
1835 for (ig = ignore_packed; *ig; ig++)
1836 if (matches_pack_name(p, *ig))
1837 break;
1838 if (*ig)
1839 goto next;
1842 if (p->num_bad_objects) {
1843 unsigned i;
1844 for (i = 0; i < p->num_bad_objects; i++)
1845 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1846 goto next;
1849 offset = find_pack_entry_one(sha1, p);
1850 if (offset) {
1852 * We are about to tell the caller where they can
1853 * locate the requested object. We better make
1854 * sure the packfile is still here and can be
1855 * accessed before supplying that answer, as
1856 * it may have been deleted since the index
1857 * was loaded!
1859 if (p->pack_fd == -1 && open_packed_git(p)) {
1860 error("packfile %s cannot be accessed", p->pack_name);
1861 goto next;
1863 e->offset = offset;
1864 e->p = p;
1865 hashcpy(e->sha1, sha1);
1866 last_found = p;
1867 return 1;
1870 next:
1871 if (p == last_found)
1872 p = packed_git;
1873 else
1874 p = p->next;
1875 if (p == last_found)
1876 p = p->next;
1877 } while (p);
1878 return 0;
1881 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1882 struct packed_git *packs)
1884 struct packed_git *p;
1886 for (p = packs; p; p = p->next) {
1887 if (find_pack_entry_one(sha1, p))
1888 return p;
1890 return NULL;
1894 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1896 int status;
1897 unsigned long mapsize, size;
1898 void *map;
1899 z_stream stream;
1900 char hdr[32];
1902 map = map_sha1_file(sha1, &mapsize);
1903 if (!map)
1904 return error("unable to find %s", sha1_to_hex(sha1));
1905 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1906 status = error("unable to unpack %s header",
1907 sha1_to_hex(sha1));
1908 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1909 status = error("unable to parse %s header", sha1_to_hex(sha1));
1910 else if (sizep)
1911 *sizep = size;
1912 inflateEnd(&stream);
1913 munmap(map, mapsize);
1914 return status;
1917 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1919 struct pack_entry e;
1921 if (!find_pack_entry(sha1, &e, NULL)) {
1922 reprepare_packed_git();
1923 if (!find_pack_entry(sha1, &e, NULL))
1924 return sha1_loose_object_info(sha1, sizep);
1926 return packed_object_info(e.p, e.offset, sizep);
1929 static void *read_packed_sha1(const unsigned char *sha1,
1930 enum object_type *type, unsigned long *size)
1932 struct pack_entry e;
1933 void *data;
1935 if (!find_pack_entry(sha1, &e, NULL))
1936 return NULL;
1937 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1938 if (!data) {
1940 * We're probably in deep shit, but let's try to fetch
1941 * the required object anyway from another pack or loose.
1942 * This should happen only in the presence of a corrupted
1943 * pack, and is better than failing outright.
1945 error("failed to read object %s at offset %"PRIuMAX" from %s",
1946 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1947 mark_bad_packed_object(e.p, sha1);
1948 data = read_sha1_file(sha1, type, size);
1950 return data;
1954 * This is meant to hold a *small* number of objects that you would
1955 * want read_sha1_file() to be able to return, but yet you do not want
1956 * to write them into the object store (e.g. a browse-only
1957 * application).
1959 static struct cached_object {
1960 unsigned char sha1[20];
1961 enum object_type type;
1962 void *buf;
1963 unsigned long size;
1964 } *cached_objects;
1965 static int cached_object_nr, cached_object_alloc;
1967 static struct cached_object empty_tree = {
1968 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1969 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1970 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1971 OBJ_TREE,
1976 static struct cached_object *find_cached_object(const unsigned char *sha1)
1978 int i;
1979 struct cached_object *co = cached_objects;
1981 for (i = 0; i < cached_object_nr; i++, co++) {
1982 if (!hashcmp(co->sha1, sha1))
1983 return co;
1985 if (!hashcmp(sha1, empty_tree.sha1))
1986 return &empty_tree;
1987 return NULL;
1990 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1991 unsigned char *sha1)
1993 struct cached_object *co;
1995 hash_sha1_file(buf, len, typename(type), sha1);
1996 if (has_sha1_file(sha1) || find_cached_object(sha1))
1997 return 0;
1998 if (cached_object_alloc <= cached_object_nr) {
1999 cached_object_alloc = alloc_nr(cached_object_alloc);
2000 cached_objects = xrealloc(cached_objects,
2001 sizeof(*cached_objects) *
2002 cached_object_alloc);
2004 co = &cached_objects[cached_object_nr++];
2005 co->size = len;
2006 co->type = type;
2007 co->buf = xmalloc(len);
2008 memcpy(co->buf, buf, len);
2009 hashcpy(co->sha1, sha1);
2010 return 0;
2013 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2014 unsigned long *size)
2016 unsigned long mapsize;
2017 void *map, *buf;
2018 struct cached_object *co;
2020 co = find_cached_object(sha1);
2021 if (co) {
2022 *type = co->type;
2023 *size = co->size;
2024 return xmemdupz(co->buf, co->size);
2027 buf = read_packed_sha1(sha1, type, size);
2028 if (buf)
2029 return buf;
2030 map = map_sha1_file(sha1, &mapsize);
2031 if (map) {
2032 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2033 munmap(map, mapsize);
2034 return buf;
2036 reprepare_packed_git();
2037 return read_packed_sha1(sha1, type, size);
2040 void *read_object_with_reference(const unsigned char *sha1,
2041 const char *required_type_name,
2042 unsigned long *size,
2043 unsigned char *actual_sha1_return)
2045 enum object_type type, required_type;
2046 void *buffer;
2047 unsigned long isize;
2048 unsigned char actual_sha1[20];
2050 required_type = type_from_string(required_type_name);
2051 hashcpy(actual_sha1, sha1);
2052 while (1) {
2053 int ref_length = -1;
2054 const char *ref_type = NULL;
2056 buffer = read_sha1_file(actual_sha1, &type, &isize);
2057 if (!buffer)
2058 return NULL;
2059 if (type == required_type) {
2060 *size = isize;
2061 if (actual_sha1_return)
2062 hashcpy(actual_sha1_return, actual_sha1);
2063 return buffer;
2065 /* Handle references */
2066 else if (type == OBJ_COMMIT)
2067 ref_type = "tree ";
2068 else if (type == OBJ_TAG)
2069 ref_type = "object ";
2070 else {
2071 free(buffer);
2072 return NULL;
2074 ref_length = strlen(ref_type);
2076 if (ref_length + 40 > isize ||
2077 memcmp(buffer, ref_type, ref_length) ||
2078 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2079 free(buffer);
2080 return NULL;
2082 free(buffer);
2083 /* Now we have the ID of the referred-to object in
2084 * actual_sha1. Check again. */
2088 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2089 const char *type, unsigned char *sha1,
2090 char *hdr, int *hdrlen)
2092 SHA_CTX c;
2094 /* Generate the header */
2095 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2097 /* Sha1.. */
2098 SHA1_Init(&c);
2099 SHA1_Update(&c, hdr, *hdrlen);
2100 SHA1_Update(&c, buf, len);
2101 SHA1_Final(sha1, &c);
2105 * Move the just written object into its final resting place
2107 int move_temp_to_file(const char *tmpfile, const char *filename)
2109 int ret = link(tmpfile, filename);
2112 * Coda hack - coda doesn't like cross-directory links,
2113 * so we fall back to a rename, which will mean that it
2114 * won't be able to check collisions, but that's not a
2115 * big deal.
2117 * The same holds for FAT formatted media.
2119 * When this succeeds, we just return 0. We have nothing
2120 * left to unlink.
2122 if (ret && ret != EEXIST) {
2123 if (!rename(tmpfile, filename))
2124 return 0;
2125 ret = errno;
2127 unlink(tmpfile);
2128 if (ret) {
2129 if (ret != EEXIST) {
2130 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2132 /* FIXME!!! Collision check here ? */
2135 return 0;
2138 static int write_buffer(int fd, const void *buf, size_t len)
2140 if (write_in_full(fd, buf, len) < 0)
2141 return error("file write error (%s)", strerror(errno));
2142 return 0;
2145 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2146 unsigned char *sha1)
2148 char hdr[32];
2149 int hdrlen;
2150 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2151 return 0;
2154 /* Finalize a file on disk, and close it. */
2155 static void close_sha1_file(int fd)
2157 if (fsync_object_files)
2158 fsync_or_die(fd, "sha1 file");
2159 fchmod(fd, 0444);
2160 if (close(fd) != 0)
2161 die("unable to write sha1 file");
2164 /* Size of directory component, including the ending '/' */
2165 static inline int directory_size(const char *filename)
2167 const char *s = strrchr(filename, '/');
2168 if (!s)
2169 return 0;
2170 return s - filename + 1;
2174 * This creates a temporary file in the same directory as the final
2175 * 'filename'
2177 * We want to avoid cross-directory filename renames, because those
2178 * can have problems on various filesystems (FAT, NFS, Coda).
2180 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2182 int fd, dirlen = directory_size(filename);
2184 if (dirlen + 20 > bufsiz) {
2185 errno = ENAMETOOLONG;
2186 return -1;
2188 memcpy(buffer, filename, dirlen);
2189 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2190 fd = mkstemp(buffer);
2191 if (fd < 0 && dirlen) {
2192 /* Make sure the directory exists */
2193 memcpy(buffer, filename, dirlen);
2194 buffer[dirlen-1] = 0;
2195 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2196 return -1;
2198 /* Try again */
2199 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2200 fd = mkstemp(buffer);
2202 return fd;
2205 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2206 void *buf, unsigned long len, time_t mtime)
2208 int fd, size, ret;
2209 unsigned char *compressed;
2210 z_stream stream;
2211 char *filename;
2212 static char tmpfile[PATH_MAX];
2214 filename = sha1_file_name(sha1);
2215 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2216 if (fd < 0) {
2217 if (errno == EPERM)
2218 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2219 else
2220 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2223 /* Set it up */
2224 memset(&stream, 0, sizeof(stream));
2225 deflateInit(&stream, zlib_compression_level);
2226 size = 8 + deflateBound(&stream, len+hdrlen);
2227 compressed = xmalloc(size);
2229 /* Compress it */
2230 stream.next_out = compressed;
2231 stream.avail_out = size;
2233 /* First header.. */
2234 stream.next_in = (unsigned char *)hdr;
2235 stream.avail_in = hdrlen;
2236 while (deflate(&stream, 0) == Z_OK)
2237 /* nothing */;
2239 /* Then the data itself.. */
2240 stream.next_in = buf;
2241 stream.avail_in = len;
2242 ret = deflate(&stream, Z_FINISH);
2243 if (ret != Z_STREAM_END)
2244 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2246 ret = deflateEnd(&stream);
2247 if (ret != Z_OK)
2248 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2250 size = stream.total_out;
2252 if (write_buffer(fd, compressed, size) < 0)
2253 die("unable to write sha1 file");
2254 close_sha1_file(fd);
2255 free(compressed);
2257 if (mtime) {
2258 struct utimbuf utb;
2259 utb.actime = mtime;
2260 utb.modtime = mtime;
2261 if (utime(tmpfile, &utb) < 0)
2262 warning("failed utime() on %s: %s",
2263 tmpfile, strerror(errno));
2266 return move_temp_to_file(tmpfile, filename);
2269 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2271 unsigned char sha1[20];
2272 char hdr[32];
2273 int hdrlen;
2275 /* Normally if we have it in the pack then we do not bother writing
2276 * it out into .git/objects/??/?{38} file.
2278 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2279 if (returnsha1)
2280 hashcpy(returnsha1, sha1);
2281 if (has_sha1_file(sha1))
2282 return 0;
2283 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2286 int force_object_loose(const unsigned char *sha1, time_t mtime)
2288 void *buf;
2289 unsigned long len;
2290 enum object_type type;
2291 char hdr[32];
2292 int hdrlen;
2294 if (has_loose_object(sha1))
2295 return 0;
2296 buf = read_packed_sha1(sha1, &type, &len);
2297 if (!buf)
2298 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2299 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2300 return write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2303 int has_pack_index(const unsigned char *sha1)
2305 struct stat st;
2306 if (stat(sha1_pack_index_name(sha1), &st))
2307 return 0;
2308 return 1;
2311 int has_pack_file(const unsigned char *sha1)
2313 struct stat st;
2314 if (stat(sha1_pack_name(sha1), &st))
2315 return 0;
2316 return 1;
2319 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2321 struct pack_entry e;
2322 return find_pack_entry(sha1, &e, ignore_packed);
2325 int has_sha1_file(const unsigned char *sha1)
2327 struct pack_entry e;
2329 if (find_pack_entry(sha1, &e, NULL))
2330 return 1;
2331 return has_loose_object(sha1);
2334 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2336 struct strbuf buf;
2337 int ret;
2339 strbuf_init(&buf, 0);
2340 if (strbuf_read(&buf, fd, 4096) < 0) {
2341 strbuf_release(&buf);
2342 return -1;
2345 if (!type)
2346 type = blob_type;
2347 if (write_object)
2348 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2349 else
2350 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2351 strbuf_release(&buf);
2353 return ret;
2356 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2357 enum object_type type, const char *path)
2359 size_t size = xsize_t(st->st_size);
2360 void *buf = NULL;
2361 int ret, re_allocated = 0;
2363 if (size)
2364 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2365 close(fd);
2367 if (!type)
2368 type = OBJ_BLOB;
2371 * Convert blobs to git internal format
2373 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2374 struct strbuf nbuf;
2375 strbuf_init(&nbuf, 0);
2376 if (convert_to_git(path, buf, size, &nbuf,
2377 write_object ? safe_crlf : 0)) {
2378 munmap(buf, size);
2379 buf = strbuf_detach(&nbuf, &size);
2380 re_allocated = 1;
2384 if (write_object)
2385 ret = write_sha1_file(buf, size, typename(type), sha1);
2386 else
2387 ret = hash_sha1_file(buf, size, typename(type), sha1);
2388 if (re_allocated) {
2389 free(buf);
2390 return ret;
2392 if (size)
2393 munmap(buf, size);
2394 return ret;
2397 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2399 int fd;
2400 char *target;
2401 size_t len;
2403 switch (st->st_mode & S_IFMT) {
2404 case S_IFREG:
2405 fd = open(path, O_RDONLY);
2406 if (fd < 0)
2407 return error("open(\"%s\"): %s", path,
2408 strerror(errno));
2409 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2410 return error("%s: failed to insert into database",
2411 path);
2412 break;
2413 case S_IFLNK:
2414 len = xsize_t(st->st_size);
2415 target = xmalloc(len + 1);
2416 if (readlink(path, target, len + 1) != st->st_size) {
2417 char *errstr = strerror(errno);
2418 free(target);
2419 return error("readlink(\"%s\"): %s", path,
2420 errstr);
2422 if (!write_object)
2423 hash_sha1_file(target, len, blob_type, sha1);
2424 else if (write_sha1_file(target, len, blob_type, sha1))
2425 return error("%s: failed to insert into database",
2426 path);
2427 free(target);
2428 break;
2429 case S_IFDIR:
2430 return resolve_gitlink_ref(path, "HEAD", sha1);
2431 default:
2432 return error("%s: unsupported file type", path);
2434 return 0;
2437 int read_pack_header(int fd, struct pack_header *header)
2439 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2440 /* "eof before pack header was fully read" */
2441 return PH_ERROR_EOF;
2443 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2444 /* "protocol error (pack signature mismatch detected)" */
2445 return PH_ERROR_PACK_SIGNATURE;
2446 if (!pack_version_ok(header->hdr_version))
2447 /* "protocol error (pack version unsupported)" */
2448 return PH_ERROR_PROTOCOL;
2449 return 0;