Merge branch 'master' of git://repo.or.cz/alt-git
[git/platforms.git] / sha1_file.c
blob951a5f61ca2f205ae2eb6d6d8c8295bd15cd22c1
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 static unsigned int sha1_file_open_flag = O_NOATIME;
40 const signed char hexval_table[256] = {
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
47 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
48 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
53 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
71 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
75 int get_sha1_hex(const char *hex, unsigned char *sha1)
77 int i;
78 for (i = 0; i < 20; i++) {
79 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
80 if (val & ~0xff)
81 return -1;
82 *sha1++ = val;
83 hex += 2;
85 return 0;
88 static inline int offset_1st_component(const char *path)
90 if (has_dos_drive_prefix(path))
91 return 2 + (path[2] == '/');
92 return *path == '/';
95 int safe_create_leading_directories(char *path)
97 char *pos = path + offset_1st_component(path);
98 struct stat st;
100 while (pos) {
101 pos = strchr(pos, '/');
102 if (!pos)
103 break;
104 *pos = 0;
105 if (!stat(path, &st)) {
106 /* path exists */
107 if (!S_ISDIR(st.st_mode)) {
108 *pos = '/';
109 return -3;
112 else if (mkdir(path, 0777)) {
113 *pos = '/';
114 return -1;
116 else if (adjust_shared_perm(path)) {
117 *pos = '/';
118 return -2;
120 *pos++ = '/';
122 return 0;
125 char * sha1_to_hex(const unsigned char *sha1)
127 static int bufno;
128 static char hexbuffer[4][50];
129 static const char hex[] = "0123456789abcdef";
130 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
131 int i;
133 for (i = 0; i < 20; i++) {
134 unsigned int val = *sha1++;
135 *buf++ = hex[val >> 4];
136 *buf++ = hex[val & 0xf];
138 *buf = '\0';
140 return buffer;
143 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
145 int i;
146 for (i = 0; i < 20; i++) {
147 static char hex[] = "0123456789abcdef";
148 unsigned int val = sha1[i];
149 char *pos = pathbuf + i*2 + (i > 0);
150 *pos++ = hex[val >> 4];
151 *pos = hex[val & 0xf];
156 * NOTE! This returns a statically allocated buffer, so you have to be
157 * careful about using it. Do an "xstrdup()" if you need to save the
158 * filename.
160 * Also note that this returns the location for creating. Reading
161 * SHA1 file can happen from any alternate directory listed in the
162 * DB_ENVIRONMENT environment variable if it is not found in
163 * the primary object database.
165 char *sha1_file_name(const unsigned char *sha1)
167 static char *name, *base;
169 if (!base) {
170 const char *sha1_file_directory = get_object_directory();
171 int len = strlen(sha1_file_directory);
172 base = xmalloc(len + 60);
173 memcpy(base, sha1_file_directory, len);
174 memset(base+len, 0, 60);
175 base[len] = '/';
176 base[len+3] = '/';
177 name = base + len + 1;
179 fill_sha1_path(name, sha1);
180 return base;
183 static char *sha1_get_pack_name(const unsigned char *sha1,
184 char **name, char **base)
186 static const char hex[] = "0123456789abcdef";
187 char *buf;
188 int i;
190 if (!*base) {
191 const char *sha1_file_directory = get_object_directory();
192 int len = strlen(sha1_file_directory);
193 *base = xmalloc(len + 60);
194 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
195 *name = *base + len + 11;
198 buf = *name;
200 for (i = 0; i < 20; i++) {
201 unsigned int val = *sha1++;
202 *buf++ = hex[val >> 4];
203 *buf++ = hex[val & 0xf];
206 return *base;
209 char *sha1_pack_name(const unsigned char *sha1)
211 static char *name, *base;
213 return sha1_get_pack_name(sha1, &name, &base);
216 char *sha1_pack_index_name(const unsigned char *sha1)
218 static char *name, *base;
220 return sha1_get_pack_name(sha1, &name, &base);
223 struct alternate_object_database *alt_odb_list;
224 static struct alternate_object_database **alt_odb_tail;
226 static void read_info_alternates(const char * alternates, int depth);
229 * Prepare alternate object database registry.
231 * The variable alt_odb_list points at the list of struct
232 * alternate_object_database. The elements on this list come from
233 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
234 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
235 * whose contents is similar to that environment variable but can be
236 * LF separated. Its base points at a statically allocated buffer that
237 * contains "/the/directory/corresponding/to/.git/objects/...", while
238 * its name points just after the slash at the end of ".git/objects/"
239 * in the example above, and has enough space to hold 40-byte hex
240 * SHA1, an extra slash for the first level indirection, and the
241 * terminating NUL.
243 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
245 struct stat st;
246 const char *objdir = get_object_directory();
247 struct alternate_object_database *ent;
248 struct alternate_object_database *alt;
249 /* 43 = 40-byte + 2 '/' + terminating NUL */
250 int pfxlen = len;
251 int entlen = pfxlen + 43;
252 int base_len = -1;
254 if (!is_absolute_path(entry) && relative_base) {
255 /* Relative alt-odb */
256 if (base_len < 0)
257 base_len = strlen(relative_base) + 1;
258 entlen += base_len;
259 pfxlen += base_len;
261 ent = xmalloc(sizeof(*ent) + entlen);
263 if (!is_absolute_path(entry) && relative_base) {
264 memcpy(ent->base, relative_base, base_len - 1);
265 ent->base[base_len - 1] = '/';
266 memcpy(ent->base + base_len, entry, len);
268 else
269 memcpy(ent->base, entry, pfxlen);
271 ent->name = ent->base + pfxlen + 1;
272 ent->base[pfxlen + 3] = '/';
273 ent->base[pfxlen] = ent->base[entlen-1] = 0;
275 /* Detect cases where alternate disappeared */
276 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
277 error("object directory %s does not exist; "
278 "check .git/objects/info/alternates.",
279 ent->base);
280 free(ent);
281 return -1;
284 /* Prevent the common mistake of listing the same
285 * thing twice, or object directory itself.
287 for (alt = alt_odb_list; alt; alt = alt->next) {
288 if (!memcmp(ent->base, alt->base, pfxlen)) {
289 free(ent);
290 return -1;
293 if (!memcmp(ent->base, objdir, pfxlen)) {
294 free(ent);
295 return -1;
298 /* add the alternate entry */
299 *alt_odb_tail = ent;
300 alt_odb_tail = &(ent->next);
301 ent->next = NULL;
303 /* recursively add alternates */
304 read_info_alternates(ent->base, depth + 1);
306 ent->base[pfxlen] = '/';
308 return 0;
311 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
312 const char *relative_base, int depth)
314 const char *cp, *last;
316 if (depth > 5) {
317 error("%s: ignoring alternate object stores, nesting too deep.",
318 relative_base);
319 return;
322 last = alt;
323 while (last < ep) {
324 cp = last;
325 if (cp < ep && *cp == '#') {
326 while (cp < ep && *cp != sep)
327 cp++;
328 last = cp + 1;
329 continue;
331 while (cp < ep && *cp != sep)
332 cp++;
333 if (last != cp) {
334 if (!is_absolute_path(last) && depth) {
335 error("%s: ignoring relative alternate object store %s",
336 relative_base, last);
337 } else {
338 link_alt_odb_entry(last, cp - last,
339 relative_base, depth);
342 while (cp < ep && *cp == sep)
343 cp++;
344 last = cp;
348 static void read_info_alternates(const char * relative_base, int depth)
350 char *map;
351 size_t mapsz;
352 struct stat st;
353 const char alt_file_name[] = "info/alternates";
354 /* Given that relative_base is no longer than PATH_MAX,
355 ensure that "path" has enough space to append "/", the
356 file name, "info/alternates", and a trailing NUL. */
357 char path[PATH_MAX + 1 + sizeof alt_file_name];
358 int fd;
360 sprintf(path, "%s/%s", relative_base, alt_file_name);
361 fd = open(path, O_RDONLY);
362 if (fd < 0)
363 return;
364 if (fstat(fd, &st) || (st.st_size == 0)) {
365 close(fd);
366 return;
368 mapsz = xsize_t(st.st_size);
369 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
370 close(fd);
372 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
374 munmap(map, mapsz);
377 void add_to_alternates_file(const char *reference)
379 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
380 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), 1);
381 char *alt = mkpath("%s/objects\n", reference);
382 write_or_die(fd, alt, strlen(alt));
383 if (commit_lock_file(lock))
384 die("could not close alternates file");
385 if (alt_odb_tail)
386 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
389 void prepare_alt_odb(void)
391 const char *alt;
393 if (alt_odb_tail)
394 return;
396 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
397 if (!alt) alt = "";
399 alt_odb_tail = &alt_odb_list;
400 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
402 read_info_alternates(get_object_directory(), 0);
405 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
407 char *name = sha1_file_name(sha1);
408 struct alternate_object_database *alt;
410 if (!stat(name, st))
411 return name;
412 prepare_alt_odb();
413 for (alt = alt_odb_list; alt; alt = alt->next) {
414 name = alt->name;
415 fill_sha1_path(name, sha1);
416 if (!stat(alt->base, st))
417 return alt->base;
419 return NULL;
422 static unsigned int pack_used_ctr;
423 static unsigned int pack_mmap_calls;
424 static unsigned int peak_pack_open_windows;
425 static unsigned int pack_open_windows;
426 static size_t peak_pack_mapped;
427 static size_t pack_mapped;
428 struct packed_git *packed_git;
430 void pack_report(void)
432 fprintf(stderr,
433 "pack_report: getpagesize() = %10" SZ_FMT "\n"
434 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
435 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
436 sz_fmt(getpagesize()),
437 sz_fmt(packed_git_window_size),
438 sz_fmt(packed_git_limit));
439 fprintf(stderr,
440 "pack_report: pack_used_ctr = %10u\n"
441 "pack_report: pack_mmap_calls = %10u\n"
442 "pack_report: pack_open_windows = %10u / %10u\n"
443 "pack_report: pack_mapped = "
444 "%10" SZ_FMT " / %10" SZ_FMT "\n",
445 pack_used_ctr,
446 pack_mmap_calls,
447 pack_open_windows, peak_pack_open_windows,
448 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
451 static int check_packed_git_idx(const char *path, struct packed_git *p)
453 void *idx_map;
454 struct pack_idx_header *hdr;
455 size_t idx_size;
456 uint32_t version, nr, i, *index;
457 int fd = open(path, O_RDONLY);
458 struct stat st;
460 if (fd < 0)
461 return -1;
462 if (fstat(fd, &st)) {
463 close(fd);
464 return -1;
466 idx_size = xsize_t(st.st_size);
467 if (idx_size < 4 * 256 + 20 + 20) {
468 close(fd);
469 return error("index file %s is too small", path);
471 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
472 close(fd);
474 hdr = idx_map;
475 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
476 version = ntohl(hdr->idx_version);
477 if (version < 2 || version > 2) {
478 munmap(idx_map, idx_size);
479 return error("index file %s is version %d"
480 " and is not supported by this binary"
481 " (try upgrading GIT to a newer version)",
482 path, version);
484 } else
485 version = 1;
487 nr = 0;
488 index = idx_map;
489 if (version > 1)
490 index += 2; /* skip index header */
491 for (i = 0; i < 256; i++) {
492 uint32_t n = ntohl(index[i]);
493 if (n < nr) {
494 munmap(idx_map, idx_size);
495 return error("non-monotonic index %s", path);
497 nr = n;
500 if (version == 1) {
502 * Total size:
503 * - 256 index entries 4 bytes each
504 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
505 * - 20-byte SHA1 of the packfile
506 * - 20-byte SHA1 file checksum
508 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
509 munmap(idx_map, idx_size);
510 return error("wrong index v1 file size in %s", path);
512 } else if (version == 2) {
514 * Minimum size:
515 * - 8 bytes of header
516 * - 256 index entries 4 bytes each
517 * - 20-byte sha1 entry * nr
518 * - 4-byte crc entry * nr
519 * - 4-byte offset entry * nr
520 * - 20-byte SHA1 of the packfile
521 * - 20-byte SHA1 file checksum
522 * And after the 4-byte offset table might be a
523 * variable sized table containing 8-byte entries
524 * for offsets larger than 2^31.
526 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
527 unsigned long max_size = min_size;
528 if (nr)
529 max_size += (nr - 1)*8;
530 if (idx_size < min_size || idx_size > max_size) {
531 munmap(idx_map, idx_size);
532 return error("wrong index v2 file size in %s", path);
534 if (idx_size != min_size &&
536 * make sure we can deal with large pack offsets.
537 * 31-bit signed offset won't be enough, neither
538 * 32-bit unsigned one will be.
540 (sizeof(off_t) <= 4)) {
541 munmap(idx_map, idx_size);
542 return error("pack too large for current definition of off_t in %s", path);
546 p->index_version = version;
547 p->index_data = idx_map;
548 p->index_size = idx_size;
549 p->num_objects = nr;
550 return 0;
553 int open_pack_index(struct packed_git *p)
555 char *idx_name;
556 int ret;
558 if (p->index_data)
559 return 0;
561 idx_name = xstrdup(p->pack_name);
562 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
563 ret = check_packed_git_idx(idx_name, p);
564 free(idx_name);
565 return ret;
568 static void scan_windows(struct packed_git *p,
569 struct packed_git **lru_p,
570 struct pack_window **lru_w,
571 struct pack_window **lru_l)
573 struct pack_window *w, *w_l;
575 for (w_l = NULL, w = p->windows; w; w = w->next) {
576 if (!w->inuse_cnt) {
577 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
578 *lru_p = p;
579 *lru_w = w;
580 *lru_l = w_l;
583 w_l = w;
587 static int unuse_one_window(struct packed_git *current, int keep_fd)
589 struct packed_git *p, *lru_p = NULL;
590 struct pack_window *lru_w = NULL, *lru_l = NULL;
592 if (current)
593 scan_windows(current, &lru_p, &lru_w, &lru_l);
594 for (p = packed_git; p; p = p->next)
595 scan_windows(p, &lru_p, &lru_w, &lru_l);
596 if (lru_p) {
597 munmap(lru_w->base, lru_w->len);
598 pack_mapped -= lru_w->len;
599 if (lru_l)
600 lru_l->next = lru_w->next;
601 else {
602 lru_p->windows = lru_w->next;
603 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
604 close(lru_p->pack_fd);
605 lru_p->pack_fd = -1;
608 free(lru_w);
609 pack_open_windows--;
610 return 1;
612 return 0;
615 void release_pack_memory(size_t need, int fd)
617 size_t cur = pack_mapped;
618 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
619 ; /* nothing */
622 void close_pack_windows(struct packed_git *p)
624 while (p->windows) {
625 struct pack_window *w = p->windows;
627 if (w->inuse_cnt)
628 die("pack '%s' still has open windows to it",
629 p->pack_name);
630 munmap(w->base, w->len);
631 pack_mapped -= w->len;
632 pack_open_windows--;
633 p->windows = w->next;
634 free(w);
638 void unuse_pack(struct pack_window **w_cursor)
640 struct pack_window *w = *w_cursor;
641 if (w) {
642 w->inuse_cnt--;
643 *w_cursor = NULL;
648 * Do not call this directly as this leaks p->pack_fd on error return;
649 * call open_packed_git() instead.
651 static int open_packed_git_1(struct packed_git *p)
653 struct stat st;
654 struct pack_header hdr;
655 unsigned char sha1[20];
656 unsigned char *idx_sha1;
657 long fd_flag;
659 if (!p->index_data && open_pack_index(p))
660 return error("packfile %s index unavailable", p->pack_name);
662 p->pack_fd = open(p->pack_name, O_RDONLY);
663 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
664 return -1;
666 /* If we created the struct before we had the pack we lack size. */
667 if (!p->pack_size) {
668 if (!S_ISREG(st.st_mode))
669 return error("packfile %s not a regular file", p->pack_name);
670 p->pack_size = st.st_size;
671 } else if (p->pack_size != st.st_size)
672 return error("packfile %s size changed", p->pack_name);
674 /* We leave these file descriptors open with sliding mmap;
675 * there is no point keeping them open across exec(), though.
677 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
678 if (fd_flag < 0)
679 return error("cannot determine file descriptor flags");
680 fd_flag |= FD_CLOEXEC;
681 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
682 return error("cannot set FD_CLOEXEC");
684 /* Verify we recognize this pack file format. */
685 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
686 return error("file %s is far too short to be a packfile", p->pack_name);
687 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
688 return error("file %s is not a GIT packfile", p->pack_name);
689 if (!pack_version_ok(hdr.hdr_version))
690 return error("packfile %s is version %u and not supported"
691 " (try upgrading GIT to a newer version)",
692 p->pack_name, ntohl(hdr.hdr_version));
694 /* Verify the pack matches its index. */
695 if (p->num_objects != ntohl(hdr.hdr_entries))
696 return error("packfile %s claims to have %u objects"
697 " while index indicates %u objects",
698 p->pack_name, ntohl(hdr.hdr_entries),
699 p->num_objects);
700 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
701 return error("end of packfile %s is unavailable", p->pack_name);
702 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
703 return error("packfile %s signature is unavailable", p->pack_name);
704 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
705 if (hashcmp(sha1, idx_sha1))
706 return error("packfile %s does not match index", p->pack_name);
707 return 0;
710 static int open_packed_git(struct packed_git *p)
712 if (!open_packed_git_1(p))
713 return 0;
714 if (p->pack_fd != -1) {
715 close(p->pack_fd);
716 p->pack_fd = -1;
718 return -1;
721 static int in_window(struct pack_window *win, off_t offset)
723 /* We must promise at least 20 bytes (one hash) after the
724 * offset is available from this window, otherwise the offset
725 * is not actually in this window and a different window (which
726 * has that one hash excess) must be used. This is to support
727 * the object header and delta base parsing routines below.
729 off_t win_off = win->offset;
730 return win_off <= offset
731 && (offset + 20) <= (win_off + win->len);
734 unsigned char* use_pack(struct packed_git *p,
735 struct pack_window **w_cursor,
736 off_t offset,
737 unsigned int *left)
739 struct pack_window *win = *w_cursor;
741 if (p->pack_fd == -1 && open_packed_git(p))
742 die("packfile %s cannot be accessed", p->pack_name);
744 /* Since packfiles end in a hash of their content and its
745 * pointless to ask for an offset into the middle of that
746 * hash, and the in_window function above wouldn't match
747 * don't allow an offset too close to the end of the file.
749 if (offset > (p->pack_size - 20))
750 die("offset beyond end of packfile (truncated pack?)");
752 if (!win || !in_window(win, offset)) {
753 if (win)
754 win->inuse_cnt--;
755 for (win = p->windows; win; win = win->next) {
756 if (in_window(win, offset))
757 break;
759 if (!win) {
760 size_t window_align = packed_git_window_size / 2;
761 off_t len;
762 win = xcalloc(1, sizeof(*win));
763 win->offset = (offset / window_align) * window_align;
764 len = p->pack_size - win->offset;
765 if (len > packed_git_window_size)
766 len = packed_git_window_size;
767 win->len = (size_t)len;
768 pack_mapped += win->len;
769 while (packed_git_limit < pack_mapped
770 && unuse_one_window(p, p->pack_fd))
771 ; /* nothing */
772 win->base = xmmap(NULL, win->len,
773 PROT_READ, MAP_PRIVATE,
774 p->pack_fd, win->offset);
775 if (win->base == MAP_FAILED)
776 die("packfile %s cannot be mapped: %s",
777 p->pack_name,
778 strerror(errno));
779 pack_mmap_calls++;
780 pack_open_windows++;
781 if (pack_mapped > peak_pack_mapped)
782 peak_pack_mapped = pack_mapped;
783 if (pack_open_windows > peak_pack_open_windows)
784 peak_pack_open_windows = pack_open_windows;
785 win->next = p->windows;
786 p->windows = win;
789 if (win != *w_cursor) {
790 win->last_used = pack_used_ctr++;
791 win->inuse_cnt++;
792 *w_cursor = win;
794 offset -= win->offset;
795 if (left)
796 *left = win->len - xsize_t(offset);
797 return win->base + offset;
800 struct packed_git *add_packed_git(const char *path, int path_len, int local)
802 struct stat st;
803 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
806 * Make sure a corresponding .pack file exists and that
807 * the index looks sane.
809 path_len -= strlen(".idx");
810 if (path_len < 1)
811 return NULL;
812 memcpy(p->pack_name, path, path_len);
813 strcpy(p->pack_name + path_len, ".pack");
814 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
815 free(p);
816 return NULL;
819 /* ok, it looks sane as far as we can check without
820 * actually mapping the pack file.
822 p->index_version = 0;
823 p->index_data = NULL;
824 p->index_size = 0;
825 p->num_objects = 0;
826 p->pack_size = st.st_size;
827 p->next = NULL;
828 p->windows = NULL;
829 p->pack_fd = -1;
830 p->pack_local = local;
831 p->mtime = st.st_mtime;
832 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
833 hashclr(p->sha1);
834 return p;
837 struct packed_git *parse_pack_index(unsigned char *sha1)
839 char *path = sha1_pack_index_name(sha1);
840 return parse_pack_index_file(sha1, path);
843 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
844 const char *idx_path)
846 const char *path = sha1_pack_name(sha1);
847 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
849 if (check_packed_git_idx(idx_path, p)) {
850 free(p);
851 return NULL;
854 strcpy(p->pack_name, path);
855 p->pack_size = 0;
856 p->next = NULL;
857 p->windows = NULL;
858 p->pack_fd = -1;
859 hashcpy(p->sha1, sha1);
860 return p;
863 void install_packed_git(struct packed_git *pack)
865 pack->next = packed_git;
866 packed_git = pack;
869 static void prepare_packed_git_one(char *objdir, int local)
871 /* Ensure that this buffer is large enough so that we can
872 append "/pack/" without clobbering the stack even if
873 strlen(objdir) were PATH_MAX. */
874 char path[PATH_MAX + 1 + 4 + 1 + 1];
875 int len;
876 DIR *dir;
877 struct dirent *de;
879 sprintf(path, "%s/pack", objdir);
880 len = strlen(path);
881 dir = opendir(path);
882 if (!dir) {
883 if (errno != ENOENT)
884 error("unable to open object pack directory: %s: %s",
885 path, strerror(errno));
886 return;
888 path[len++] = '/';
889 while ((de = readdir(dir)) != NULL) {
890 int namelen = strlen(de->d_name);
891 struct packed_git *p;
893 if (!has_extension(de->d_name, ".idx"))
894 continue;
896 if (len + namelen + 1 > sizeof(path))
897 continue;
899 /* Don't reopen a pack we already have. */
900 strcpy(path + len, de->d_name);
901 for (p = packed_git; p; p = p->next) {
902 if (!memcmp(path, p->pack_name, len + namelen - 4))
903 break;
905 if (p)
906 continue;
907 /* See if it really is a valid .idx file with corresponding
908 * .pack file that we can map.
910 p = add_packed_git(path, len + namelen, local);
911 if (!p)
912 continue;
913 install_packed_git(p);
915 closedir(dir);
918 static int sort_pack(const void *a_, const void *b_)
920 struct packed_git *a = *((struct packed_git **)a_);
921 struct packed_git *b = *((struct packed_git **)b_);
922 int st;
925 * Local packs tend to contain objects specific to our
926 * variant of the project than remote ones. In addition,
927 * remote ones could be on a network mounted filesystem.
928 * Favor local ones for these reasons.
930 st = a->pack_local - b->pack_local;
931 if (st)
932 return -st;
935 * Younger packs tend to contain more recent objects,
936 * and more recent objects tend to get accessed more
937 * often.
939 if (a->mtime < b->mtime)
940 return 1;
941 else if (a->mtime == b->mtime)
942 return 0;
943 return -1;
946 static void rearrange_packed_git(void)
948 struct packed_git **ary, *p;
949 int i, n;
951 for (n = 0, p = packed_git; p; p = p->next)
952 n++;
953 if (n < 2)
954 return;
956 /* prepare an array of packed_git for easier sorting */
957 ary = xcalloc(n, sizeof(struct packed_git *));
958 for (n = 0, p = packed_git; p; p = p->next)
959 ary[n++] = p;
961 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
963 /* link them back again */
964 for (i = 0; i < n - 1; i++)
965 ary[i]->next = ary[i + 1];
966 ary[n - 1]->next = NULL;
967 packed_git = ary[0];
969 free(ary);
972 static int prepare_packed_git_run_once = 0;
973 void prepare_packed_git(void)
975 struct alternate_object_database *alt;
977 if (prepare_packed_git_run_once)
978 return;
979 prepare_packed_git_one(get_object_directory(), 1);
980 prepare_alt_odb();
981 for (alt = alt_odb_list; alt; alt = alt->next) {
982 alt->name[-1] = 0;
983 prepare_packed_git_one(alt->base, 0);
984 alt->name[-1] = '/';
986 rearrange_packed_git();
987 prepare_packed_git_run_once = 1;
990 void reprepare_packed_git(void)
992 prepare_packed_git_run_once = 0;
993 prepare_packed_git();
996 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
998 unsigned char real_sha1[20];
999 hash_sha1_file(map, size, type, real_sha1);
1000 return hashcmp(sha1, real_sha1) ? -1 : 0;
1003 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1005 struct stat st;
1006 void *map;
1007 int fd;
1008 char *filename = find_sha1_file(sha1, &st);
1010 if (!filename) {
1011 return NULL;
1014 fd = open(filename, O_RDONLY | sha1_file_open_flag);
1015 if (fd < 0) {
1016 /* See if it works without O_NOATIME */
1017 switch (sha1_file_open_flag) {
1018 default:
1019 fd = open(filename, O_RDONLY);
1020 if (fd >= 0)
1021 break;
1022 /* Fallthrough */
1023 case 0:
1024 return NULL;
1027 /* If it failed once, it will probably fail again.
1028 * Stop using O_NOATIME
1030 sha1_file_open_flag = 0;
1032 *size = xsize_t(st.st_size);
1033 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1034 close(fd);
1035 return map;
1038 static int legacy_loose_object(unsigned char *map)
1040 unsigned int word;
1043 * Is it a zlib-compressed buffer? If so, the first byte
1044 * must be 0x78 (15-bit window size, deflated), and the
1045 * first 16-bit word is evenly divisible by 31
1047 word = (map[0] << 8) + map[1];
1048 if (map[0] == 0x78 && !(word % 31))
1049 return 1;
1050 else
1051 return 0;
1054 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1056 unsigned shift;
1057 unsigned char c;
1058 unsigned long size;
1059 unsigned long used = 0;
1061 c = buf[used++];
1062 *type = (c >> 4) & 7;
1063 size = c & 15;
1064 shift = 4;
1065 while (c & 0x80) {
1066 if (len <= used)
1067 return 0;
1068 if (sizeof(long) * 8 <= shift)
1069 return 0;
1070 c = buf[used++];
1071 size += (c & 0x7f) << shift;
1072 shift += 7;
1074 *sizep = size;
1075 return used;
1078 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1080 unsigned long size, used;
1081 static const char valid_loose_object_type[8] = {
1082 0, /* OBJ_EXT */
1083 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1084 0, /* "delta" and others are invalid in a loose object */
1086 enum object_type type;
1088 /* Get the data stream */
1089 memset(stream, 0, sizeof(*stream));
1090 stream->next_in = map;
1091 stream->avail_in = mapsize;
1092 stream->next_out = buffer;
1093 stream->avail_out = bufsiz;
1095 if (legacy_loose_object(map)) {
1096 inflateInit(stream);
1097 return inflate(stream, 0);
1102 * There used to be a second loose object header format which
1103 * was meant to mimic the in-pack format, allowing for direct
1104 * copy of the object data. This format turned up not to be
1105 * really worth it and we don't write it any longer. But we
1106 * can still read it.
1108 used = unpack_object_header_gently(map, mapsize, &type, &size);
1109 if (!used || !valid_loose_object_type[type])
1110 return -1;
1111 map += used;
1112 mapsize -= used;
1114 /* Set up the stream for the rest.. */
1115 stream->next_in = map;
1116 stream->avail_in = mapsize;
1117 inflateInit(stream);
1119 /* And generate the fake traditional header */
1120 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1121 typename(type), size);
1122 return 0;
1125 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1127 int bytes = strlen(buffer) + 1;
1128 unsigned char *buf = xmalloc(1+size);
1129 unsigned long n;
1130 int status = Z_OK;
1132 n = stream->total_out - bytes;
1133 if (n > size)
1134 n = size;
1135 memcpy(buf, (char *) buffer + bytes, n);
1136 bytes = n;
1137 if (bytes <= size) {
1139 * The above condition must be (bytes <= size), not
1140 * (bytes < size). In other words, even though we
1141 * expect no more output and set avail_out to zer0,
1142 * the input zlib stream may have bytes that express
1143 * "this concludes the stream", and we *do* want to
1144 * eat that input.
1146 * Otherwise we would not be able to test that we
1147 * consumed all the input to reach the expected size;
1148 * we also want to check that zlib tells us that all
1149 * went well with status == Z_STREAM_END at the end.
1151 stream->next_out = buf + bytes;
1152 stream->avail_out = size - bytes;
1153 while (status == Z_OK)
1154 status = inflate(stream, Z_FINISH);
1156 buf[size] = 0;
1157 if (status == Z_STREAM_END && !stream->avail_in) {
1158 inflateEnd(stream);
1159 return buf;
1162 if (status < 0)
1163 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1164 else if (stream->avail_in)
1165 error("garbage at end of loose object '%s'",
1166 sha1_to_hex(sha1));
1167 free(buf);
1168 return NULL;
1172 * We used to just use "sscanf()", but that's actually way
1173 * too permissive for what we want to check. So do an anal
1174 * object header parse by hand.
1176 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1178 char type[10];
1179 int i;
1180 unsigned long size;
1183 * The type can be at most ten bytes (including the
1184 * terminating '\0' that we add), and is followed by
1185 * a space.
1187 i = 0;
1188 for (;;) {
1189 char c = *hdr++;
1190 if (c == ' ')
1191 break;
1192 type[i++] = c;
1193 if (i >= sizeof(type))
1194 return -1;
1196 type[i] = 0;
1199 * The length must follow immediately, and be in canonical
1200 * decimal format (ie "010" is not valid).
1202 size = *hdr++ - '0';
1203 if (size > 9)
1204 return -1;
1205 if (size) {
1206 for (;;) {
1207 unsigned long c = *hdr - '0';
1208 if (c > 9)
1209 break;
1210 hdr++;
1211 size = size * 10 + c;
1214 *sizep = size;
1217 * The length must be followed by a zero byte
1219 return *hdr ? -1 : type_from_string(type);
1222 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1224 int ret;
1225 z_stream stream;
1226 char hdr[8192];
1228 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1229 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1230 return NULL;
1232 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1235 unsigned long get_size_from_delta(struct packed_git *p,
1236 struct pack_window **w_curs,
1237 off_t curpos)
1239 const unsigned char *data;
1240 unsigned char delta_head[20], *in;
1241 z_stream stream;
1242 int st;
1244 memset(&stream, 0, sizeof(stream));
1245 stream.next_out = delta_head;
1246 stream.avail_out = sizeof(delta_head);
1248 inflateInit(&stream);
1249 do {
1250 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1251 stream.next_in = in;
1252 st = inflate(&stream, Z_FINISH);
1253 curpos += stream.next_in - in;
1254 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1255 stream.total_out < sizeof(delta_head));
1256 inflateEnd(&stream);
1257 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1258 die("delta data unpack-initial failed");
1260 /* Examine the initial part of the delta to figure out
1261 * the result size.
1263 data = delta_head;
1265 /* ignore base size */
1266 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1268 /* Read the result size */
1269 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1272 static off_t get_delta_base(struct packed_git *p,
1273 struct pack_window **w_curs,
1274 off_t *curpos,
1275 enum object_type type,
1276 off_t delta_obj_offset)
1278 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1279 off_t base_offset;
1281 /* use_pack() assured us we have [base_info, base_info + 20)
1282 * as a range that we can look at without walking off the
1283 * end of the mapped window. Its actually the hash size
1284 * that is assured. An OFS_DELTA longer than the hash size
1285 * is stupid, as then a REF_DELTA would be smaller to store.
1287 if (type == OBJ_OFS_DELTA) {
1288 unsigned used = 0;
1289 unsigned char c = base_info[used++];
1290 base_offset = c & 127;
1291 while (c & 128) {
1292 base_offset += 1;
1293 if (!base_offset || MSB(base_offset, 7))
1294 die("offset value overflow for delta base object");
1295 c = base_info[used++];
1296 base_offset = (base_offset << 7) + (c & 127);
1298 base_offset = delta_obj_offset - base_offset;
1299 if (base_offset >= delta_obj_offset)
1300 die("delta base offset out of bound");
1301 *curpos += used;
1302 } else if (type == OBJ_REF_DELTA) {
1303 /* The base entry _must_ be in the same pack */
1304 base_offset = find_pack_entry_one(base_info, p);
1305 if (!base_offset)
1306 die("failed to find delta-pack base object %s",
1307 sha1_to_hex(base_info));
1308 *curpos += 20;
1309 } else
1310 die("I am totally screwed");
1311 return base_offset;
1314 /* forward declaration for a mutually recursive function */
1315 static int packed_object_info(struct packed_git *p, off_t offset,
1316 unsigned long *sizep);
1318 static int packed_delta_info(struct packed_git *p,
1319 struct pack_window **w_curs,
1320 off_t curpos,
1321 enum object_type type,
1322 off_t obj_offset,
1323 unsigned long *sizep)
1325 off_t base_offset;
1327 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1328 type = packed_object_info(p, base_offset, NULL);
1330 /* We choose to only get the type of the base object and
1331 * ignore potentially corrupt pack file that expects the delta
1332 * based on a base with a wrong size. This saves tons of
1333 * inflate() calls.
1335 if (sizep)
1336 *sizep = get_size_from_delta(p, w_curs, curpos);
1338 return type;
1341 static int unpack_object_header(struct packed_git *p,
1342 struct pack_window **w_curs,
1343 off_t *curpos,
1344 unsigned long *sizep)
1346 unsigned char *base;
1347 unsigned int left;
1348 unsigned long used;
1349 enum object_type type;
1351 /* use_pack() assures us we have [base, base + 20) available
1352 * as a range that we can look at at. (Its actually the hash
1353 * size that is assured.) With our object header encoding
1354 * the maximum deflated object size is 2^137, which is just
1355 * insane, so we know won't exceed what we have been given.
1357 base = use_pack(p, w_curs, *curpos, &left);
1358 used = unpack_object_header_gently(base, left, &type, sizep);
1359 if (!used)
1360 die("object offset outside of pack file");
1361 *curpos += used;
1363 return type;
1366 const char *packed_object_info_detail(struct packed_git *p,
1367 off_t obj_offset,
1368 unsigned long *size,
1369 unsigned long *store_size,
1370 unsigned int *delta_chain_length,
1371 unsigned char *base_sha1)
1373 struct pack_window *w_curs = NULL;
1374 off_t curpos;
1375 unsigned long dummy;
1376 unsigned char *next_sha1;
1377 enum object_type type;
1378 struct revindex_entry *revidx;
1380 *delta_chain_length = 0;
1381 curpos = obj_offset;
1382 type = unpack_object_header(p, &w_curs, &curpos, size);
1384 revidx = find_pack_revindex(p, obj_offset);
1385 *store_size = revidx[1].offset - obj_offset;
1387 for (;;) {
1388 switch (type) {
1389 default:
1390 die("pack %s contains unknown object type %d",
1391 p->pack_name, type);
1392 case OBJ_COMMIT:
1393 case OBJ_TREE:
1394 case OBJ_BLOB:
1395 case OBJ_TAG:
1396 unuse_pack(&w_curs);
1397 return typename(type);
1398 case OBJ_OFS_DELTA:
1399 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1400 if (*delta_chain_length == 0) {
1401 revidx = find_pack_revindex(p, obj_offset);
1402 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1404 break;
1405 case OBJ_REF_DELTA:
1406 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1407 if (*delta_chain_length == 0)
1408 hashcpy(base_sha1, next_sha1);
1409 obj_offset = find_pack_entry_one(next_sha1, p);
1410 break;
1412 (*delta_chain_length)++;
1413 curpos = obj_offset;
1414 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1418 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1419 unsigned long *sizep)
1421 struct pack_window *w_curs = NULL;
1422 unsigned long size;
1423 off_t curpos = obj_offset;
1424 enum object_type type;
1426 type = unpack_object_header(p, &w_curs, &curpos, &size);
1428 switch (type) {
1429 case OBJ_OFS_DELTA:
1430 case OBJ_REF_DELTA:
1431 type = packed_delta_info(p, &w_curs, curpos,
1432 type, obj_offset, sizep);
1433 break;
1434 case OBJ_COMMIT:
1435 case OBJ_TREE:
1436 case OBJ_BLOB:
1437 case OBJ_TAG:
1438 if (sizep)
1439 *sizep = size;
1440 break;
1441 default:
1442 die("pack %s contains unknown object type %d",
1443 p->pack_name, type);
1445 unuse_pack(&w_curs);
1446 return type;
1449 static void *unpack_compressed_entry(struct packed_git *p,
1450 struct pack_window **w_curs,
1451 off_t curpos,
1452 unsigned long size)
1454 int st;
1455 z_stream stream;
1456 unsigned char *buffer, *in;
1458 buffer = xmalloc(size + 1);
1459 buffer[size] = 0;
1460 memset(&stream, 0, sizeof(stream));
1461 stream.next_out = buffer;
1462 stream.avail_out = size;
1464 inflateInit(&stream);
1465 do {
1466 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1467 stream.next_in = in;
1468 st = inflate(&stream, Z_FINISH);
1469 curpos += stream.next_in - in;
1470 } while (st == Z_OK || st == Z_BUF_ERROR);
1471 inflateEnd(&stream);
1472 if ((st != Z_STREAM_END) || stream.total_out != size) {
1473 free(buffer);
1474 return NULL;
1477 return buffer;
1480 #define MAX_DELTA_CACHE (256)
1482 static size_t delta_base_cached;
1484 static struct delta_base_cache_lru_list {
1485 struct delta_base_cache_lru_list *prev;
1486 struct delta_base_cache_lru_list *next;
1487 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1489 static struct delta_base_cache_entry {
1490 struct delta_base_cache_lru_list lru;
1491 void *data;
1492 struct packed_git *p;
1493 off_t base_offset;
1494 unsigned long size;
1495 enum object_type type;
1496 } delta_base_cache[MAX_DELTA_CACHE];
1498 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1500 unsigned long hash;
1502 hash = (unsigned long)p + (unsigned long)base_offset;
1503 hash += (hash >> 8) + (hash >> 16);
1504 return hash % MAX_DELTA_CACHE;
1507 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1508 unsigned long *base_size, enum object_type *type, int keep_cache)
1510 void *ret;
1511 unsigned long hash = pack_entry_hash(p, base_offset);
1512 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1514 ret = ent->data;
1515 if (ret && ent->p == p && ent->base_offset == base_offset)
1516 goto found_cache_entry;
1517 return unpack_entry(p, base_offset, type, base_size);
1519 found_cache_entry:
1520 if (!keep_cache) {
1521 ent->data = NULL;
1522 ent->lru.next->prev = ent->lru.prev;
1523 ent->lru.prev->next = ent->lru.next;
1524 delta_base_cached -= ent->size;
1525 } else {
1526 ret = xmemdupz(ent->data, ent->size);
1528 *type = ent->type;
1529 *base_size = ent->size;
1530 return ret;
1533 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1535 if (ent->data) {
1536 free(ent->data);
1537 ent->data = NULL;
1538 ent->lru.next->prev = ent->lru.prev;
1539 ent->lru.prev->next = ent->lru.next;
1540 delta_base_cached -= ent->size;
1544 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1545 void *base, unsigned long base_size, enum object_type type)
1547 unsigned long hash = pack_entry_hash(p, base_offset);
1548 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1549 struct delta_base_cache_lru_list *lru;
1551 release_delta_base_cache(ent);
1552 delta_base_cached += base_size;
1554 for (lru = delta_base_cache_lru.next;
1555 delta_base_cached > delta_base_cache_limit
1556 && lru != &delta_base_cache_lru;
1557 lru = lru->next) {
1558 struct delta_base_cache_entry *f = (void *)lru;
1559 if (f->type == OBJ_BLOB)
1560 release_delta_base_cache(f);
1562 for (lru = delta_base_cache_lru.next;
1563 delta_base_cached > delta_base_cache_limit
1564 && lru != &delta_base_cache_lru;
1565 lru = lru->next) {
1566 struct delta_base_cache_entry *f = (void *)lru;
1567 release_delta_base_cache(f);
1570 ent->p = p;
1571 ent->base_offset = base_offset;
1572 ent->type = type;
1573 ent->data = base;
1574 ent->size = base_size;
1575 ent->lru.next = &delta_base_cache_lru;
1576 ent->lru.prev = delta_base_cache_lru.prev;
1577 delta_base_cache_lru.prev->next = &ent->lru;
1578 delta_base_cache_lru.prev = &ent->lru;
1581 static void *unpack_delta_entry(struct packed_git *p,
1582 struct pack_window **w_curs,
1583 off_t curpos,
1584 unsigned long delta_size,
1585 off_t obj_offset,
1586 enum object_type *type,
1587 unsigned long *sizep)
1589 void *delta_data, *result, *base;
1590 unsigned long base_size;
1591 off_t base_offset;
1593 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1594 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1595 if (!base)
1596 die("failed to read delta base object"
1597 " at %"PRIuMAX" from %s",
1598 (uintmax_t)base_offset, p->pack_name);
1600 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1601 if (!delta_data)
1602 die("failed to unpack compressed delta"
1603 " at %"PRIuMAX" from %s",
1604 (uintmax_t)curpos, p->pack_name);
1605 result = patch_delta(base, base_size,
1606 delta_data, delta_size,
1607 sizep);
1608 if (!result)
1609 die("failed to apply delta");
1610 free(delta_data);
1611 add_delta_base_cache(p, base_offset, base, base_size, *type);
1612 return result;
1615 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1616 enum object_type *type, unsigned long *sizep)
1618 struct pack_window *w_curs = NULL;
1619 off_t curpos = obj_offset;
1620 void *data;
1622 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1623 switch (*type) {
1624 case OBJ_OFS_DELTA:
1625 case OBJ_REF_DELTA:
1626 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1627 obj_offset, type, sizep);
1628 break;
1629 case OBJ_COMMIT:
1630 case OBJ_TREE:
1631 case OBJ_BLOB:
1632 case OBJ_TAG:
1633 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1634 break;
1635 default:
1636 die("unknown object type %i in %s", *type, p->pack_name);
1638 unuse_pack(&w_curs);
1639 return data;
1642 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1643 uint32_t n)
1645 const unsigned char *index = p->index_data;
1646 if (!index) {
1647 if (open_pack_index(p))
1648 return NULL;
1649 index = p->index_data;
1651 if (n >= p->num_objects)
1652 return NULL;
1653 index += 4 * 256;
1654 if (p->index_version == 1) {
1655 return index + 24 * n + 4;
1656 } else {
1657 index += 8;
1658 return index + 20 * n;
1662 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1664 const unsigned char *index = p->index_data;
1665 index += 4 * 256;
1666 if (p->index_version == 1) {
1667 return ntohl(*((uint32_t *)(index + 24 * n)));
1668 } else {
1669 uint32_t off;
1670 index += 8 + p->num_objects * (20 + 4);
1671 off = ntohl(*((uint32_t *)(index + 4 * n)));
1672 if (!(off & 0x80000000))
1673 return off;
1674 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1675 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1676 ntohl(*((uint32_t *)(index + 4)));
1680 off_t find_pack_entry_one(const unsigned char *sha1,
1681 struct packed_git *p)
1683 const uint32_t *level1_ofs = p->index_data;
1684 const unsigned char *index = p->index_data;
1685 unsigned hi, lo, stride;
1686 static int use_lookup = -1;
1687 static int debug_lookup = -1;
1689 if (debug_lookup < 0)
1690 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1692 if (!index) {
1693 if (open_pack_index(p))
1694 return 0;
1695 level1_ofs = p->index_data;
1696 index = p->index_data;
1698 if (p->index_version > 1) {
1699 level1_ofs += 2;
1700 index += 8;
1702 index += 4 * 256;
1703 hi = ntohl(level1_ofs[*sha1]);
1704 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1705 if (p->index_version > 1) {
1706 stride = 20;
1707 } else {
1708 stride = 24;
1709 index += 4;
1712 if (debug_lookup)
1713 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1714 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1716 if (use_lookup < 0)
1717 use_lookup = !!getenv("GIT_USE_LOOKUP");
1718 if (use_lookup) {
1719 int pos = sha1_entry_pos(index, stride, 0,
1720 lo, hi, p->num_objects, sha1);
1721 if (pos < 0)
1722 return 0;
1723 return nth_packed_object_offset(p, pos);
1726 do {
1727 unsigned mi = (lo + hi) / 2;
1728 int cmp = hashcmp(index + mi * stride, sha1);
1730 if (debug_lookup)
1731 printf("lo %u hi %u rg %u mi %u\n",
1732 lo, hi, hi - lo, mi);
1733 if (!cmp)
1734 return nth_packed_object_offset(p, mi);
1735 if (cmp > 0)
1736 hi = mi;
1737 else
1738 lo = mi+1;
1739 } while (lo < hi);
1740 return 0;
1743 int matches_pack_name(struct packed_git *p, const char *name)
1745 const char *last_c, *c;
1747 if (!strcmp(p->pack_name, name))
1748 return 1;
1750 for (c = p->pack_name, last_c = c; *c;)
1751 if (*c == '/')
1752 last_c = ++c;
1753 else
1754 ++c;
1755 if (!strcmp(last_c, name))
1756 return 1;
1758 return 0;
1761 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1763 static struct packed_git *last_found = (void *)1;
1764 struct packed_git *p;
1765 off_t offset;
1767 prepare_packed_git();
1768 if (!packed_git)
1769 return 0;
1770 p = (last_found == (void *)1) ? packed_git : last_found;
1772 do {
1773 if (ignore_packed) {
1774 const char **ig;
1775 for (ig = ignore_packed; *ig; ig++)
1776 if (matches_pack_name(p, *ig))
1777 break;
1778 if (*ig)
1779 goto next;
1782 offset = find_pack_entry_one(sha1, p);
1783 if (offset) {
1785 * We are about to tell the caller where they can
1786 * locate the requested object. We better make
1787 * sure the packfile is still here and can be
1788 * accessed before supplying that answer, as
1789 * it may have been deleted since the index
1790 * was loaded!
1792 if (p->pack_fd == -1 && open_packed_git(p)) {
1793 error("packfile %s cannot be accessed", p->pack_name);
1794 goto next;
1796 e->offset = offset;
1797 e->p = p;
1798 hashcpy(e->sha1, sha1);
1799 last_found = p;
1800 return 1;
1803 next:
1804 if (p == last_found)
1805 p = packed_git;
1806 else
1807 p = p->next;
1808 if (p == last_found)
1809 p = p->next;
1810 } while (p);
1811 return 0;
1814 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1815 struct packed_git *packs)
1817 struct packed_git *p;
1819 for (p = packs; p; p = p->next) {
1820 if (find_pack_entry_one(sha1, p))
1821 return p;
1823 return NULL;
1827 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1829 int status;
1830 unsigned long mapsize, size;
1831 void *map;
1832 z_stream stream;
1833 char hdr[32];
1835 map = map_sha1_file(sha1, &mapsize);
1836 if (!map)
1837 return error("unable to find %s", sha1_to_hex(sha1));
1838 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1839 status = error("unable to unpack %s header",
1840 sha1_to_hex(sha1));
1841 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1842 status = error("unable to parse %s header", sha1_to_hex(sha1));
1843 else if (sizep)
1844 *sizep = size;
1845 inflateEnd(&stream);
1846 munmap(map, mapsize);
1847 return status;
1850 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1852 struct pack_entry e;
1854 if (!find_pack_entry(sha1, &e, NULL)) {
1855 reprepare_packed_git();
1856 if (!find_pack_entry(sha1, &e, NULL))
1857 return sha1_loose_object_info(sha1, sizep);
1859 return packed_object_info(e.p, e.offset, sizep);
1862 static void *read_packed_sha1(const unsigned char *sha1,
1863 enum object_type *type, unsigned long *size)
1865 struct pack_entry e;
1867 if (!find_pack_entry(sha1, &e, NULL))
1868 return NULL;
1869 else
1870 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1874 * This is meant to hold a *small* number of objects that you would
1875 * want read_sha1_file() to be able to return, but yet you do not want
1876 * to write them into the object store (e.g. a browse-only
1877 * application).
1879 static struct cached_object {
1880 unsigned char sha1[20];
1881 enum object_type type;
1882 void *buf;
1883 unsigned long size;
1884 } *cached_objects;
1885 static int cached_object_nr, cached_object_alloc;
1887 static struct cached_object empty_tree = {
1888 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1889 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1890 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1891 OBJ_TREE,
1896 static struct cached_object *find_cached_object(const unsigned char *sha1)
1898 int i;
1899 struct cached_object *co = cached_objects;
1901 for (i = 0; i < cached_object_nr; i++, co++) {
1902 if (!hashcmp(co->sha1, sha1))
1903 return co;
1905 if (!hashcmp(sha1, empty_tree.sha1))
1906 return &empty_tree;
1907 return NULL;
1910 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1911 unsigned char *sha1)
1913 struct cached_object *co;
1915 hash_sha1_file(buf, len, typename(type), sha1);
1916 if (has_sha1_file(sha1) || find_cached_object(sha1))
1917 return 0;
1918 if (cached_object_alloc <= cached_object_nr) {
1919 cached_object_alloc = alloc_nr(cached_object_alloc);
1920 cached_objects = xrealloc(cached_objects,
1921 sizeof(*cached_objects) *
1922 cached_object_alloc);
1924 co = &cached_objects[cached_object_nr++];
1925 co->size = len;
1926 co->type = type;
1927 co->buf = xmalloc(len);
1928 memcpy(co->buf, buf, len);
1929 hashcpy(co->sha1, sha1);
1930 return 0;
1933 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1934 unsigned long *size)
1936 unsigned long mapsize;
1937 void *map, *buf;
1938 struct cached_object *co;
1940 co = find_cached_object(sha1);
1941 if (co) {
1942 *type = co->type;
1943 *size = co->size;
1944 return xmemdupz(co->buf, co->size);
1947 buf = read_packed_sha1(sha1, type, size);
1948 if (buf)
1949 return buf;
1950 map = map_sha1_file(sha1, &mapsize);
1951 if (map) {
1952 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1953 munmap(map, mapsize);
1954 return buf;
1956 reprepare_packed_git();
1957 return read_packed_sha1(sha1, type, size);
1960 void *read_object_with_reference(const unsigned char *sha1,
1961 const char *required_type_name,
1962 unsigned long *size,
1963 unsigned char *actual_sha1_return)
1965 enum object_type type, required_type;
1966 void *buffer;
1967 unsigned long isize;
1968 unsigned char actual_sha1[20];
1970 required_type = type_from_string(required_type_name);
1971 hashcpy(actual_sha1, sha1);
1972 while (1) {
1973 int ref_length = -1;
1974 const char *ref_type = NULL;
1976 buffer = read_sha1_file(actual_sha1, &type, &isize);
1977 if (!buffer)
1978 return NULL;
1979 if (type == required_type) {
1980 *size = isize;
1981 if (actual_sha1_return)
1982 hashcpy(actual_sha1_return, actual_sha1);
1983 return buffer;
1985 /* Handle references */
1986 else if (type == OBJ_COMMIT)
1987 ref_type = "tree ";
1988 else if (type == OBJ_TAG)
1989 ref_type = "object ";
1990 else {
1991 free(buffer);
1992 return NULL;
1994 ref_length = strlen(ref_type);
1996 if (ref_length + 40 > isize ||
1997 memcmp(buffer, ref_type, ref_length) ||
1998 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1999 free(buffer);
2000 return NULL;
2002 free(buffer);
2003 /* Now we have the ID of the referred-to object in
2004 * actual_sha1. Check again. */
2008 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2009 const char *type, unsigned char *sha1,
2010 char *hdr, int *hdrlen)
2012 SHA_CTX c;
2014 /* Generate the header */
2015 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2017 /* Sha1.. */
2018 SHA1_Init(&c);
2019 SHA1_Update(&c, hdr, *hdrlen);
2020 SHA1_Update(&c, buf, len);
2021 SHA1_Final(sha1, &c);
2025 * Link the tempfile to the final place, possibly creating the
2026 * last directory level as you do so.
2028 * Returns the errno on failure, 0 on success.
2030 static int link_temp_to_file(const char *tmpfile, const char *filename)
2032 int ret;
2033 char *dir;
2035 if (!link(tmpfile, filename))
2036 return 0;
2039 * Try to mkdir the last path component if that failed.
2041 * Re-try the "link()" regardless of whether the mkdir
2042 * succeeds, since a race might mean that somebody
2043 * else succeeded.
2045 ret = errno;
2046 dir = strrchr(filename, '/');
2047 if (dir) {
2048 *dir = 0;
2049 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
2050 *dir = '/';
2051 return -2;
2053 *dir = '/';
2054 if (!link(tmpfile, filename))
2055 return 0;
2056 ret = errno;
2058 return ret;
2062 * Move the just written object into its final resting place
2064 int move_temp_to_file(const char *tmpfile, const char *filename)
2066 int ret = link_temp_to_file(tmpfile, filename);
2069 * Coda hack - coda doesn't like cross-directory links,
2070 * so we fall back to a rename, which will mean that it
2071 * won't be able to check collisions, but that's not a
2072 * big deal.
2074 * The same holds for FAT formatted media.
2076 * When this succeeds, we just return 0. We have nothing
2077 * left to unlink.
2079 if (ret && ret != EEXIST) {
2080 if (!rename(tmpfile, filename))
2081 return 0;
2082 ret = errno;
2084 unlink(tmpfile);
2085 if (ret) {
2086 if (ret != EEXIST) {
2087 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2089 /* FIXME!!! Collision check here ? */
2092 return 0;
2095 static int write_buffer(int fd, const void *buf, size_t len)
2097 if (write_in_full(fd, buf, len) < 0)
2098 return error("file write error (%s)", strerror(errno));
2099 return 0;
2102 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2103 unsigned char *sha1)
2105 char hdr[32];
2106 int hdrlen;
2107 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2108 return 0;
2111 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2112 void *buf, unsigned long len, time_t mtime)
2114 int fd, size, ret;
2115 unsigned char *compressed;
2116 z_stream stream;
2117 char *filename;
2118 static char tmpfile[PATH_MAX];
2120 filename = sha1_file_name(sha1);
2121 fd = open(filename, O_RDONLY);
2122 if (fd >= 0) {
2124 * FIXME!!! We might do collision checking here, but we'd
2125 * need to uncompress the old file and check it. Later.
2127 close(fd);
2128 return 0;
2131 if (errno != ENOENT) {
2132 return error("sha1 file %s: %s\n", filename, strerror(errno));
2135 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2137 fd = mkstemp(tmpfile);
2138 if (fd < 0) {
2139 if (errno == EPERM)
2140 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2141 else
2142 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2145 /* Set it up */
2146 memset(&stream, 0, sizeof(stream));
2147 deflateInit(&stream, zlib_compression_level);
2148 size = 8 + deflateBound(&stream, len+hdrlen);
2149 compressed = xmalloc(size);
2151 /* Compress it */
2152 stream.next_out = compressed;
2153 stream.avail_out = size;
2155 /* First header.. */
2156 stream.next_in = (unsigned char *)hdr;
2157 stream.avail_in = hdrlen;
2158 while (deflate(&stream, 0) == Z_OK)
2159 /* nothing */;
2161 /* Then the data itself.. */
2162 stream.next_in = buf;
2163 stream.avail_in = len;
2164 ret = deflate(&stream, Z_FINISH);
2165 if (ret != Z_STREAM_END)
2166 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2168 ret = deflateEnd(&stream);
2169 if (ret != Z_OK)
2170 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2172 size = stream.total_out;
2174 if (write_buffer(fd, compressed, size) < 0)
2175 die("unable to write sha1 file");
2176 fchmod(fd, 0444);
2177 if (close(fd))
2178 die("unable to write sha1 file");
2179 free(compressed);
2181 if (mtime) {
2182 struct utimbuf utb;
2183 utb.actime = mtime;
2184 utb.modtime = mtime;
2185 if (utime(tmpfile, &utb) < 0)
2186 warning("failed utime() on %s: %s",
2187 tmpfile, strerror(errno));
2190 return move_temp_to_file(tmpfile, filename);
2193 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2195 unsigned char sha1[20];
2196 char hdr[32];
2197 int hdrlen;
2199 /* Normally if we have it in the pack then we do not bother writing
2200 * it out into .git/objects/??/?{38} file.
2202 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2203 if (returnsha1)
2204 hashcpy(returnsha1, sha1);
2205 if (has_sha1_file(sha1))
2206 return 0;
2207 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2210 int force_object_loose(const unsigned char *sha1, time_t mtime)
2212 struct stat st;
2213 void *buf;
2214 unsigned long len;
2215 enum object_type type;
2216 char hdr[32];
2217 int hdrlen;
2219 if (find_sha1_file(sha1, &st))
2220 return 0;
2221 buf = read_packed_sha1(sha1, &type, &len);
2222 if (!buf)
2223 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2224 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2225 return write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2229 * We need to unpack and recompress the object for writing
2230 * it out to a different file.
2232 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2234 size_t size;
2235 z_stream stream;
2236 unsigned char *unpacked;
2237 unsigned long len;
2238 enum object_type type;
2239 char hdr[32];
2240 int hdrlen;
2241 void *buf;
2243 /* need to unpack and recompress it by itself */
2244 unpacked = read_packed_sha1(sha1, &type, &len);
2245 if (!unpacked)
2246 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2248 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2250 /* Set it up */
2251 memset(&stream, 0, sizeof(stream));
2252 deflateInit(&stream, zlib_compression_level);
2253 size = deflateBound(&stream, len + hdrlen);
2254 buf = xmalloc(size);
2256 /* Compress it */
2257 stream.next_out = buf;
2258 stream.avail_out = size;
2260 /* First header.. */
2261 stream.next_in = (void *)hdr;
2262 stream.avail_in = hdrlen;
2263 while (deflate(&stream, 0) == Z_OK)
2264 /* nothing */;
2266 /* Then the data itself.. */
2267 stream.next_in = unpacked;
2268 stream.avail_in = len;
2269 while (deflate(&stream, Z_FINISH) == Z_OK)
2270 /* nothing */;
2271 deflateEnd(&stream);
2272 free(unpacked);
2274 *objsize = stream.total_out;
2275 return buf;
2278 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2280 int retval;
2281 unsigned long objsize;
2282 void *buf = map_sha1_file(sha1, &objsize);
2284 if (buf) {
2285 retval = write_buffer(fd, buf, objsize);
2286 munmap(buf, objsize);
2287 return retval;
2290 buf = repack_object(sha1, &objsize);
2291 retval = write_buffer(fd, buf, objsize);
2292 free(buf);
2293 return retval;
2296 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2297 size_t bufsize, size_t *bufposn)
2299 char tmpfile[PATH_MAX];
2300 int local;
2301 z_stream stream;
2302 unsigned char real_sha1[20];
2303 unsigned char discard[4096];
2304 int ret;
2305 SHA_CTX c;
2307 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2309 local = mkstemp(tmpfile);
2310 if (local < 0) {
2311 if (errno == EPERM)
2312 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2313 else
2314 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2317 memset(&stream, 0, sizeof(stream));
2319 inflateInit(&stream);
2321 SHA1_Init(&c);
2323 do {
2324 ssize_t size;
2325 if (*bufposn) {
2326 stream.avail_in = *bufposn;
2327 stream.next_in = (unsigned char *) buffer;
2328 do {
2329 stream.next_out = discard;
2330 stream.avail_out = sizeof(discard);
2331 ret = inflate(&stream, Z_SYNC_FLUSH);
2332 SHA1_Update(&c, discard, sizeof(discard) -
2333 stream.avail_out);
2334 } while (stream.avail_in && ret == Z_OK);
2335 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2336 die("unable to write sha1 file");
2337 memmove(buffer, buffer + *bufposn - stream.avail_in,
2338 stream.avail_in);
2339 *bufposn = stream.avail_in;
2340 if (ret != Z_OK)
2341 break;
2343 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2344 if (size <= 0) {
2345 close(local);
2346 unlink(tmpfile);
2347 if (!size)
2348 return error("Connection closed?");
2349 perror("Reading from connection");
2350 return -1;
2352 *bufposn += size;
2353 } while (1);
2354 inflateEnd(&stream);
2356 fchmod(local, 0444);
2357 if (close(local) != 0)
2358 die("unable to write sha1 file");
2359 SHA1_Final(real_sha1, &c);
2360 if (ret != Z_STREAM_END) {
2361 unlink(tmpfile);
2362 return error("File %s corrupted", sha1_to_hex(sha1));
2364 if (hashcmp(sha1, real_sha1)) {
2365 unlink(tmpfile);
2366 return error("File %s has bad hash", sha1_to_hex(sha1));
2369 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2372 int has_pack_index(const unsigned char *sha1)
2374 struct stat st;
2375 if (stat(sha1_pack_index_name(sha1), &st))
2376 return 0;
2377 return 1;
2380 int has_pack_file(const unsigned char *sha1)
2382 struct stat st;
2383 if (stat(sha1_pack_name(sha1), &st))
2384 return 0;
2385 return 1;
2388 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2390 struct pack_entry e;
2391 return find_pack_entry(sha1, &e, ignore_packed);
2394 int has_sha1_file(const unsigned char *sha1)
2396 struct stat st;
2397 struct pack_entry e;
2399 if (find_pack_entry(sha1, &e, NULL))
2400 return 1;
2401 return find_sha1_file(sha1, &st) ? 1 : 0;
2404 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2406 struct strbuf buf;
2407 int ret;
2409 strbuf_init(&buf, 0);
2410 if (strbuf_read(&buf, fd, 4096) < 0) {
2411 strbuf_release(&buf);
2412 return -1;
2415 if (!type)
2416 type = blob_type;
2417 if (write_object)
2418 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2419 else
2420 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2421 strbuf_release(&buf);
2423 return ret;
2426 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2427 enum object_type type, const char *path)
2429 size_t size = xsize_t(st->st_size);
2430 void *buf = NULL;
2431 int ret, re_allocated = 0;
2433 if (size)
2434 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2435 close(fd);
2437 if (!type)
2438 type = OBJ_BLOB;
2441 * Convert blobs to git internal format
2443 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2444 struct strbuf nbuf;
2445 strbuf_init(&nbuf, 0);
2446 if (convert_to_git(path, buf, size, &nbuf,
2447 write_object ? safe_crlf : 0)) {
2448 munmap(buf, size);
2449 buf = strbuf_detach(&nbuf, &size);
2450 re_allocated = 1;
2454 if (write_object)
2455 ret = write_sha1_file(buf, size, typename(type), sha1);
2456 else
2457 ret = hash_sha1_file(buf, size, typename(type), sha1);
2458 if (re_allocated) {
2459 free(buf);
2460 return ret;
2462 if (size)
2463 munmap(buf, size);
2464 return ret;
2467 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2469 int fd;
2470 char *target;
2471 size_t len;
2473 switch (st->st_mode & S_IFMT) {
2474 case S_IFREG:
2475 fd = open(path, O_RDONLY);
2476 if (fd < 0)
2477 return error("open(\"%s\"): %s", path,
2478 strerror(errno));
2479 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2480 return error("%s: failed to insert into database",
2481 path);
2482 break;
2483 case S_IFLNK:
2484 len = xsize_t(st->st_size);
2485 target = xmalloc(len + 1);
2486 if (readlink(path, target, len + 1) != st->st_size) {
2487 char *errstr = strerror(errno);
2488 free(target);
2489 return error("readlink(\"%s\"): %s", path,
2490 errstr);
2492 if (!write_object)
2493 hash_sha1_file(target, len, blob_type, sha1);
2494 else if (write_sha1_file(target, len, blob_type, sha1))
2495 return error("%s: failed to insert into database",
2496 path);
2497 free(target);
2498 break;
2499 case S_IFDIR:
2500 return resolve_gitlink_ref(path, "HEAD", sha1);
2501 default:
2502 return error("%s: unsupported file type", path);
2504 return 0;
2507 int read_pack_header(int fd, struct pack_header *header)
2509 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2510 /* "eof before pack header was fully read" */
2511 return PH_ERROR_EOF;
2513 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2514 /* "protocol error (pack signature mismatch detected)" */
2515 return PH_ERROR_PACK_SIGNATURE;
2516 if (!pack_version_ok(header->hdr_version))
2517 /* "protocol error (pack version unsupported)" */
2518 return PH_ERROR_PROTOCOL;
2519 return 0;