Allow git-svnimport to take "" as the trunk directory.
[git/dscho.git] / sha1_file.c
blobaca741b79ce3cdc28f0eb5bdc6848ddba6da8f35
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "blob.h"
13 #include "commit.h"
14 #include "tag.h"
15 #include "tree.h"
16 #include "refs.h"
18 #ifndef O_NOATIME
19 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
20 #define O_NOATIME 01000000
21 #else
22 #define O_NOATIME 0
23 #endif
24 #endif
26 #ifdef NO_C99_FORMAT
27 #define SZ_FMT "lu"
28 #else
29 #define SZ_FMT "zu"
30 #endif
32 const unsigned char null_sha1[20];
34 static unsigned int sha1_file_open_flag = O_NOATIME;
36 const signed char hexval_table[256] = {
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
43 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
44 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
45 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
71 int get_sha1_hex(const char *hex, unsigned char *sha1)
73 int i;
74 for (i = 0; i < 20; i++) {
75 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
76 if (val & ~0xff)
77 return -1;
78 *sha1++ = val;
79 hex += 2;
81 return 0;
84 int safe_create_leading_directories(char *path)
86 char *pos = path;
87 struct stat st;
89 if (*pos == '/')
90 pos++;
92 while (pos) {
93 pos = strchr(pos, '/');
94 if (!pos)
95 break;
96 *pos = 0;
97 if (!stat(path, &st)) {
98 /* path exists */
99 if (!S_ISDIR(st.st_mode)) {
100 *pos = '/';
101 return -3;
104 else if (mkdir(path, 0777)) {
105 *pos = '/';
106 return -1;
108 else if (adjust_shared_perm(path)) {
109 *pos = '/';
110 return -2;
112 *pos++ = '/';
114 return 0;
117 char * sha1_to_hex(const unsigned char *sha1)
119 static int bufno;
120 static char hexbuffer[4][50];
121 static const char hex[] = "0123456789abcdef";
122 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
123 int i;
125 for (i = 0; i < 20; i++) {
126 unsigned int val = *sha1++;
127 *buf++ = hex[val >> 4];
128 *buf++ = hex[val & 0xf];
130 *buf = '\0';
132 return buffer;
135 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
137 int i;
138 for (i = 0; i < 20; i++) {
139 static char hex[] = "0123456789abcdef";
140 unsigned int val = sha1[i];
141 char *pos = pathbuf + i*2 + (i > 0);
142 *pos++ = hex[val >> 4];
143 *pos = hex[val & 0xf];
148 * NOTE! This returns a statically allocated buffer, so you have to be
149 * careful about using it. Do a "xstrdup()" if you need to save the
150 * filename.
152 * Also note that this returns the location for creating. Reading
153 * SHA1 file can happen from any alternate directory listed in the
154 * DB_ENVIRONMENT environment variable if it is not found in
155 * the primary object database.
157 char *sha1_file_name(const unsigned char *sha1)
159 static char *name, *base;
161 if (!base) {
162 const char *sha1_file_directory = get_object_directory();
163 int len = strlen(sha1_file_directory);
164 base = xmalloc(len + 60);
165 memcpy(base, sha1_file_directory, len);
166 memset(base+len, 0, 60);
167 base[len] = '/';
168 base[len+3] = '/';
169 name = base + len + 1;
171 fill_sha1_path(name, sha1);
172 return base;
175 char *sha1_pack_name(const unsigned char *sha1)
177 static const char hex[] = "0123456789abcdef";
178 static char *name, *base, *buf;
179 int i;
181 if (!base) {
182 const char *sha1_file_directory = get_object_directory();
183 int len = strlen(sha1_file_directory);
184 base = xmalloc(len + 60);
185 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
186 name = base + len + 11;
189 buf = name;
191 for (i = 0; i < 20; i++) {
192 unsigned int val = *sha1++;
193 *buf++ = hex[val >> 4];
194 *buf++ = hex[val & 0xf];
197 return base;
200 char *sha1_pack_index_name(const unsigned char *sha1)
202 static const char hex[] = "0123456789abcdef";
203 static char *name, *base, *buf;
204 int i;
206 if (!base) {
207 const char *sha1_file_directory = get_object_directory();
208 int len = strlen(sha1_file_directory);
209 base = xmalloc(len + 60);
210 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
211 name = base + len + 11;
214 buf = name;
216 for (i = 0; i < 20; i++) {
217 unsigned int val = *sha1++;
218 *buf++ = hex[val >> 4];
219 *buf++ = hex[val & 0xf];
222 return base;
225 struct alternate_object_database *alt_odb_list;
226 static struct alternate_object_database **alt_odb_tail;
228 static void read_info_alternates(const char * alternates, int depth);
231 * Prepare alternate object database registry.
233 * The variable alt_odb_list points at the list of struct
234 * alternate_object_database. The elements on this list come from
235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
237 * whose contents is similar to that environment variable but can be
238 * LF separated. Its base points at a statically allocated buffer that
239 * contains "/the/directory/corresponding/to/.git/objects/...", while
240 * its name points just after the slash at the end of ".git/objects/"
241 * in the example above, and has enough space to hold 40-byte hex
242 * SHA1, an extra slash for the first level indirection, and the
243 * terminating NUL.
245 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
247 struct stat st;
248 const char *objdir = get_object_directory();
249 struct alternate_object_database *ent;
250 struct alternate_object_database *alt;
251 /* 43 = 40-byte + 2 '/' + terminating NUL */
252 int pfxlen = len;
253 int entlen = pfxlen + 43;
254 int base_len = -1;
256 if (*entry != '/' && relative_base) {
257 /* Relative alt-odb */
258 if (base_len < 0)
259 base_len = strlen(relative_base) + 1;
260 entlen += base_len;
261 pfxlen += base_len;
263 ent = xmalloc(sizeof(*ent) + entlen);
265 if (*entry != '/' && relative_base) {
266 memcpy(ent->base, relative_base, base_len - 1);
267 ent->base[base_len - 1] = '/';
268 memcpy(ent->base + base_len, entry, len);
270 else
271 memcpy(ent->base, entry, pfxlen);
273 ent->name = ent->base + pfxlen + 1;
274 ent->base[pfxlen + 3] = '/';
275 ent->base[pfxlen] = ent->base[entlen-1] = 0;
277 /* Detect cases where alternate disappeared */
278 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
279 error("object directory %s does not exist; "
280 "check .git/objects/info/alternates.",
281 ent->base);
282 free(ent);
283 return -1;
286 /* Prevent the common mistake of listing the same
287 * thing twice, or object directory itself.
289 for (alt = alt_odb_list; alt; alt = alt->next) {
290 if (!memcmp(ent->base, alt->base, pfxlen)) {
291 free(ent);
292 return -1;
295 if (!memcmp(ent->base, objdir, pfxlen)) {
296 free(ent);
297 return -1;
300 /* add the alternate entry */
301 *alt_odb_tail = ent;
302 alt_odb_tail = &(ent->next);
303 ent->next = NULL;
305 /* recursively add alternates */
306 read_info_alternates(ent->base, depth + 1);
308 ent->base[pfxlen] = '/';
310 return 0;
313 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
314 const char *relative_base, int depth)
316 const char *cp, *last;
318 if (depth > 5) {
319 error("%s: ignoring alternate object stores, nesting too deep.",
320 relative_base);
321 return;
324 last = alt;
325 while (last < ep) {
326 cp = last;
327 if (cp < ep && *cp == '#') {
328 while (cp < ep && *cp != sep)
329 cp++;
330 last = cp + 1;
331 continue;
333 while (cp < ep && *cp != sep)
334 cp++;
335 if (last != cp) {
336 if ((*last != '/') && depth) {
337 error("%s: ignoring relative alternate object store %s",
338 relative_base, last);
339 } else {
340 link_alt_odb_entry(last, cp - last,
341 relative_base, depth);
344 while (cp < ep && *cp == sep)
345 cp++;
346 last = cp;
350 static void read_info_alternates(const char * relative_base, int depth)
352 char *map;
353 size_t mapsz;
354 struct stat st;
355 const char alt_file_name[] = "info/alternates";
356 /* Given that relative_base is no longer than PATH_MAX,
357 ensure that "path" has enough space to append "/", the
358 file name, "info/alternates", and a trailing NUL. */
359 char path[PATH_MAX + 1 + sizeof alt_file_name];
360 int fd;
362 sprintf(path, "%s/%s", relative_base, alt_file_name);
363 fd = open(path, O_RDONLY);
364 if (fd < 0)
365 return;
366 if (fstat(fd, &st) || (st.st_size == 0)) {
367 close(fd);
368 return;
370 mapsz = xsize_t(st.st_size);
371 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
372 close(fd);
374 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
376 munmap(map, mapsz);
379 void prepare_alt_odb(void)
381 const char *alt;
383 if (alt_odb_tail)
384 return;
386 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
387 if (!alt) alt = "";
389 alt_odb_tail = &alt_odb_list;
390 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
392 read_info_alternates(get_object_directory(), 0);
395 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
397 char *name = sha1_file_name(sha1);
398 struct alternate_object_database *alt;
400 if (!stat(name, st))
401 return name;
402 prepare_alt_odb();
403 for (alt = alt_odb_list; alt; alt = alt->next) {
404 name = alt->name;
405 fill_sha1_path(name, sha1);
406 if (!stat(alt->base, st))
407 return alt->base;
409 return NULL;
412 static unsigned int pack_used_ctr;
413 static unsigned int pack_mmap_calls;
414 static unsigned int peak_pack_open_windows;
415 static unsigned int pack_open_windows;
416 static size_t peak_pack_mapped;
417 static size_t pack_mapped;
418 struct packed_git *packed_git;
420 void pack_report(void)
422 fprintf(stderr,
423 "pack_report: getpagesize() = %10" SZ_FMT "\n"
424 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
425 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
426 (size_t) getpagesize(),
427 packed_git_window_size,
428 packed_git_limit);
429 fprintf(stderr,
430 "pack_report: pack_used_ctr = %10u\n"
431 "pack_report: pack_mmap_calls = %10u\n"
432 "pack_report: pack_open_windows = %10u / %10u\n"
433 "pack_report: pack_mapped = "
434 "%10" SZ_FMT " / %10" SZ_FMT "\n",
435 pack_used_ctr,
436 pack_mmap_calls,
437 pack_open_windows, peak_pack_open_windows,
438 pack_mapped, peak_pack_mapped);
441 static int check_packed_git_idx(const char *path, struct packed_git *p)
443 void *idx_map;
444 struct pack_idx_header *hdr;
445 size_t idx_size;
446 uint32_t version, nr, i, *index;
447 int fd = open(path, O_RDONLY);
448 struct stat st;
450 if (fd < 0)
451 return -1;
452 if (fstat(fd, &st)) {
453 close(fd);
454 return -1;
456 idx_size = xsize_t(st.st_size);
457 if (idx_size < 4 * 256 + 20 + 20) {
458 close(fd);
459 return error("index file %s is too small", path);
461 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
462 close(fd);
464 hdr = idx_map;
465 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
466 version = ntohl(hdr->idx_version);
467 if (version < 2 || version > 2) {
468 munmap(idx_map, idx_size);
469 return error("index file %s is version %d"
470 " and is not supported by this binary"
471 " (try upgrading GIT to a newer version)",
472 path, version);
474 } else
475 version = 1;
477 nr = 0;
478 index = idx_map;
479 if (version > 1)
480 index += 2; /* skip index header */
481 for (i = 0; i < 256; i++) {
482 uint32_t n = ntohl(index[i]);
483 if (n < nr) {
484 munmap(idx_map, idx_size);
485 return error("non-monotonic index %s", path);
487 nr = n;
490 if (version == 1) {
492 * Total size:
493 * - 256 index entries 4 bytes each
494 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
495 * - 20-byte SHA1 of the packfile
496 * - 20-byte SHA1 file checksum
498 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
499 munmap(idx_map, idx_size);
500 return error("wrong index file size in %s", path);
502 } else if (version == 2) {
504 * Minimum size:
505 * - 8 bytes of header
506 * - 256 index entries 4 bytes each
507 * - 20-byte sha1 entry * nr
508 * - 4-byte crc entry * nr
509 * - 4-byte offset entry * nr
510 * - 20-byte SHA1 of the packfile
511 * - 20-byte SHA1 file checksum
512 * And after the 4-byte offset table might be a
513 * variable sized table containing 8-byte entries
514 * for offsets larger than 2^31.
516 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
517 unsigned long max_size = min_size;
518 if (nr)
519 max_size += (nr - 1)*8;
520 if (idx_size < min_size || idx_size > max_size) {
521 munmap(idx_map, idx_size);
522 return error("wrong index file size in %s", path);
524 if (idx_size != min_size) {
525 /* make sure we can deal with large pack offsets */
526 off_t x = 0x7fffffffUL, y = 0xffffffffUL;
527 if (x > (x + 1) || y > (y + 1)) {
528 munmap(idx_map, idx_size);
529 return error("pack too large for current definition of off_t in %s", path);
534 p->index_version = version;
535 p->index_data = idx_map;
536 p->index_size = idx_size;
537 p->num_objects = nr;
538 return 0;
541 int open_pack_index(struct packed_git *p)
543 char *idx_name;
544 int ret;
546 if (p->index_data)
547 return 0;
549 idx_name = xstrdup(p->pack_name);
550 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
551 ret = check_packed_git_idx(idx_name, p);
552 free(idx_name);
553 return ret;
556 static void scan_windows(struct packed_git *p,
557 struct packed_git **lru_p,
558 struct pack_window **lru_w,
559 struct pack_window **lru_l)
561 struct pack_window *w, *w_l;
563 for (w_l = NULL, w = p->windows; w; w = w->next) {
564 if (!w->inuse_cnt) {
565 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
566 *lru_p = p;
567 *lru_w = w;
568 *lru_l = w_l;
571 w_l = w;
575 static int unuse_one_window(struct packed_git *current, int keep_fd)
577 struct packed_git *p, *lru_p = NULL;
578 struct pack_window *lru_w = NULL, *lru_l = NULL;
580 if (current)
581 scan_windows(current, &lru_p, &lru_w, &lru_l);
582 for (p = packed_git; p; p = p->next)
583 scan_windows(p, &lru_p, &lru_w, &lru_l);
584 if (lru_p) {
585 munmap(lru_w->base, lru_w->len);
586 pack_mapped -= lru_w->len;
587 if (lru_l)
588 lru_l->next = lru_w->next;
589 else {
590 lru_p->windows = lru_w->next;
591 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
592 close(lru_p->pack_fd);
593 lru_p->pack_fd = -1;
596 free(lru_w);
597 pack_open_windows--;
598 return 1;
600 return 0;
603 void release_pack_memory(size_t need, int fd)
605 size_t cur = pack_mapped;
606 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
607 ; /* nothing */
610 void unuse_pack(struct pack_window **w_cursor)
612 struct pack_window *w = *w_cursor;
613 if (w) {
614 w->inuse_cnt--;
615 *w_cursor = NULL;
620 * Do not call this directly as this leaks p->pack_fd on error return;
621 * call open_packed_git() instead.
623 static int open_packed_git_1(struct packed_git *p)
625 struct stat st;
626 struct pack_header hdr;
627 unsigned char sha1[20];
628 unsigned char *idx_sha1;
629 long fd_flag;
631 if (!p->index_data && open_pack_index(p))
632 return error("packfile %s index unavailable", p->pack_name);
634 p->pack_fd = open(p->pack_name, O_RDONLY);
635 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
636 return -1;
638 /* If we created the struct before we had the pack we lack size. */
639 if (!p->pack_size) {
640 if (!S_ISREG(st.st_mode))
641 return error("packfile %s not a regular file", p->pack_name);
642 p->pack_size = st.st_size;
643 } else if (p->pack_size != st.st_size)
644 return error("packfile %s size changed", p->pack_name);
646 /* We leave these file descriptors open with sliding mmap;
647 * there is no point keeping them open across exec(), though.
649 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
650 if (fd_flag < 0)
651 return error("cannot determine file descriptor flags");
652 fd_flag |= FD_CLOEXEC;
653 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
654 return error("cannot set FD_CLOEXEC");
656 /* Verify we recognize this pack file format. */
657 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
658 return error("file %s is far too short to be a packfile", p->pack_name);
659 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
660 return error("file %s is not a GIT packfile", p->pack_name);
661 if (!pack_version_ok(hdr.hdr_version))
662 return error("packfile %s is version %u and not supported"
663 " (try upgrading GIT to a newer version)",
664 p->pack_name, ntohl(hdr.hdr_version));
666 /* Verify the pack matches its index. */
667 if (p->num_objects != ntohl(hdr.hdr_entries))
668 return error("packfile %s claims to have %u objects"
669 " while index indicates %u objects",
670 p->pack_name, ntohl(hdr.hdr_entries),
671 p->num_objects);
672 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
673 return error("end of packfile %s is unavailable", p->pack_name);
674 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
675 return error("packfile %s signature is unavailable", p->pack_name);
676 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
677 if (hashcmp(sha1, idx_sha1))
678 return error("packfile %s does not match index", p->pack_name);
679 return 0;
682 static int open_packed_git(struct packed_git *p)
684 if (!open_packed_git_1(p))
685 return 0;
686 if (p->pack_fd != -1) {
687 close(p->pack_fd);
688 p->pack_fd = -1;
690 return -1;
693 static int in_window(struct pack_window *win, off_t offset)
695 /* We must promise at least 20 bytes (one hash) after the
696 * offset is available from this window, otherwise the offset
697 * is not actually in this window and a different window (which
698 * has that one hash excess) must be used. This is to support
699 * the object header and delta base parsing routines below.
701 off_t win_off = win->offset;
702 return win_off <= offset
703 && (offset + 20) <= (win_off + win->len);
706 unsigned char* use_pack(struct packed_git *p,
707 struct pack_window **w_cursor,
708 off_t offset,
709 unsigned int *left)
711 struct pack_window *win = *w_cursor;
713 if (p->pack_fd == -1 && open_packed_git(p))
714 die("packfile %s cannot be accessed", p->pack_name);
716 /* Since packfiles end in a hash of their content and its
717 * pointless to ask for an offset into the middle of that
718 * hash, and the in_window function above wouldn't match
719 * don't allow an offset too close to the end of the file.
721 if (offset > (p->pack_size - 20))
722 die("offset beyond end of packfile (truncated pack?)");
724 if (!win || !in_window(win, offset)) {
725 if (win)
726 win->inuse_cnt--;
727 for (win = p->windows; win; win = win->next) {
728 if (in_window(win, offset))
729 break;
731 if (!win) {
732 size_t window_align = packed_git_window_size / 2;
733 off_t len;
734 win = xcalloc(1, sizeof(*win));
735 win->offset = (offset / window_align) * window_align;
736 len = p->pack_size - win->offset;
737 if (len > packed_git_window_size)
738 len = packed_git_window_size;
739 win->len = (size_t)len;
740 pack_mapped += win->len;
741 while (packed_git_limit < pack_mapped
742 && unuse_one_window(p, p->pack_fd))
743 ; /* nothing */
744 win->base = xmmap(NULL, win->len,
745 PROT_READ, MAP_PRIVATE,
746 p->pack_fd, win->offset);
747 if (win->base == MAP_FAILED)
748 die("packfile %s cannot be mapped: %s",
749 p->pack_name,
750 strerror(errno));
751 pack_mmap_calls++;
752 pack_open_windows++;
753 if (pack_mapped > peak_pack_mapped)
754 peak_pack_mapped = pack_mapped;
755 if (pack_open_windows > peak_pack_open_windows)
756 peak_pack_open_windows = pack_open_windows;
757 win->next = p->windows;
758 p->windows = win;
761 if (win != *w_cursor) {
762 win->last_used = pack_used_ctr++;
763 win->inuse_cnt++;
764 *w_cursor = win;
766 offset -= win->offset;
767 if (left)
768 *left = win->len - xsize_t(offset);
769 return win->base + offset;
772 struct packed_git *add_packed_git(const char *path, int path_len, int local)
774 struct stat st;
775 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
778 * Make sure a corresponding .pack file exists and that
779 * the index looks sane.
781 path_len -= strlen(".idx");
782 if (path_len < 1)
783 return NULL;
784 memcpy(p->pack_name, path, path_len);
785 strcpy(p->pack_name + path_len, ".pack");
786 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
787 free(p);
788 return NULL;
791 /* ok, it looks sane as far as we can check without
792 * actually mapping the pack file.
794 p->index_version = 0;
795 p->index_data = NULL;
796 p->index_size = 0;
797 p->num_objects = 0;
798 p->pack_size = st.st_size;
799 p->next = NULL;
800 p->windows = NULL;
801 p->pack_fd = -1;
802 p->pack_local = local;
803 p->mtime = st.st_mtime;
804 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
805 hashclr(p->sha1);
806 return p;
809 struct packed_git *parse_pack_index(unsigned char *sha1)
811 char *path = sha1_pack_index_name(sha1);
812 return parse_pack_index_file(sha1, path);
815 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
816 const char *idx_path)
818 const char *path = sha1_pack_name(sha1);
819 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
821 if (check_packed_git_idx(idx_path, p)) {
822 free(p);
823 return NULL;
826 strcpy(p->pack_name, path);
827 p->pack_size = 0;
828 p->next = NULL;
829 p->windows = NULL;
830 p->pack_fd = -1;
831 hashcpy(p->sha1, sha1);
832 return p;
835 void install_packed_git(struct packed_git *pack)
837 pack->next = packed_git;
838 packed_git = pack;
841 static void prepare_packed_git_one(char *objdir, int local)
843 /* Ensure that this buffer is large enough so that we can
844 append "/pack/" without clobbering the stack even if
845 strlen(objdir) were PATH_MAX. */
846 char path[PATH_MAX + 1 + 4 + 1 + 1];
847 int len;
848 DIR *dir;
849 struct dirent *de;
851 sprintf(path, "%s/pack", objdir);
852 len = strlen(path);
853 dir = opendir(path);
854 if (!dir) {
855 if (errno != ENOENT)
856 error("unable to open object pack directory: %s: %s",
857 path, strerror(errno));
858 return;
860 path[len++] = '/';
861 while ((de = readdir(dir)) != NULL) {
862 int namelen = strlen(de->d_name);
863 struct packed_git *p;
865 if (!has_extension(de->d_name, ".idx"))
866 continue;
868 if (len + namelen + 1 > sizeof(path))
869 continue;
871 /* Don't reopen a pack we already have. */
872 strcpy(path + len, de->d_name);
873 for (p = packed_git; p; p = p->next) {
874 if (!memcmp(path, p->pack_name, len + namelen - 4))
875 break;
877 if (p)
878 continue;
879 /* See if it really is a valid .idx file with corresponding
880 * .pack file that we can map.
882 p = add_packed_git(path, len + namelen, local);
883 if (!p)
884 continue;
885 install_packed_git(p);
887 closedir(dir);
890 static int sort_pack(const void *a_, const void *b_)
892 struct packed_git *a = *((struct packed_git **)a_);
893 struct packed_git *b = *((struct packed_git **)b_);
894 int st;
897 * Local packs tend to contain objects specific to our
898 * variant of the project than remote ones. In addition,
899 * remote ones could be on a network mounted filesystem.
900 * Favor local ones for these reasons.
902 st = a->pack_local - b->pack_local;
903 if (st)
904 return -st;
907 * Younger packs tend to contain more recent objects,
908 * and more recent objects tend to get accessed more
909 * often.
911 if (a->mtime < b->mtime)
912 return 1;
913 else if (a->mtime == b->mtime)
914 return 0;
915 return -1;
918 static void rearrange_packed_git(void)
920 struct packed_git **ary, *p;
921 int i, n;
923 for (n = 0, p = packed_git; p; p = p->next)
924 n++;
925 if (n < 2)
926 return;
928 /* prepare an array of packed_git for easier sorting */
929 ary = xcalloc(n, sizeof(struct packed_git *));
930 for (n = 0, p = packed_git; p; p = p->next)
931 ary[n++] = p;
933 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
935 /* link them back again */
936 for (i = 0; i < n - 1; i++)
937 ary[i]->next = ary[i + 1];
938 ary[n - 1]->next = NULL;
939 packed_git = ary[0];
941 free(ary);
944 static int prepare_packed_git_run_once = 0;
945 void prepare_packed_git(void)
947 struct alternate_object_database *alt;
949 if (prepare_packed_git_run_once)
950 return;
951 prepare_packed_git_one(get_object_directory(), 1);
952 prepare_alt_odb();
953 for (alt = alt_odb_list; alt; alt = alt->next) {
954 alt->name[-1] = 0;
955 prepare_packed_git_one(alt->base, 0);
956 alt->name[-1] = '/';
958 rearrange_packed_git();
959 prepare_packed_git_run_once = 1;
962 void reprepare_packed_git(void)
964 prepare_packed_git_run_once = 0;
965 prepare_packed_git();
968 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
970 unsigned char real_sha1[20];
971 hash_sha1_file(map, size, type, real_sha1);
972 return hashcmp(sha1, real_sha1) ? -1 : 0;
975 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
977 struct stat st;
978 void *map;
979 int fd;
980 char *filename = find_sha1_file(sha1, &st);
982 if (!filename) {
983 return NULL;
986 fd = open(filename, O_RDONLY | sha1_file_open_flag);
987 if (fd < 0) {
988 /* See if it works without O_NOATIME */
989 switch (sha1_file_open_flag) {
990 default:
991 fd = open(filename, O_RDONLY);
992 if (fd >= 0)
993 break;
994 /* Fallthrough */
995 case 0:
996 return NULL;
999 /* If it failed once, it will probably fail again.
1000 * Stop using O_NOATIME
1002 sha1_file_open_flag = 0;
1004 *size = xsize_t(st.st_size);
1005 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1006 close(fd);
1007 return map;
1010 static int legacy_loose_object(unsigned char *map)
1012 unsigned int word;
1015 * Is it a zlib-compressed buffer? If so, the first byte
1016 * must be 0x78 (15-bit window size, deflated), and the
1017 * first 16-bit word is evenly divisible by 31
1019 word = (map[0] << 8) + map[1];
1020 if (map[0] == 0x78 && !(word % 31))
1021 return 1;
1022 else
1023 return 0;
1026 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1028 unsigned shift;
1029 unsigned char c;
1030 unsigned long size;
1031 unsigned long used = 0;
1033 c = buf[used++];
1034 *type = (c >> 4) & 7;
1035 size = c & 15;
1036 shift = 4;
1037 while (c & 0x80) {
1038 if (len <= used)
1039 return 0;
1040 if (sizeof(long) * 8 <= shift)
1041 return 0;
1042 c = buf[used++];
1043 size += (c & 0x7f) << shift;
1044 shift += 7;
1046 *sizep = size;
1047 return used;
1050 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1052 unsigned long size, used;
1053 static const char valid_loose_object_type[8] = {
1054 0, /* OBJ_EXT */
1055 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1056 0, /* "delta" and others are invalid in a loose object */
1058 enum object_type type;
1060 /* Get the data stream */
1061 memset(stream, 0, sizeof(*stream));
1062 stream->next_in = map;
1063 stream->avail_in = mapsize;
1064 stream->next_out = buffer;
1065 stream->avail_out = bufsiz;
1067 if (legacy_loose_object(map)) {
1068 inflateInit(stream);
1069 return inflate(stream, 0);
1074 * There used to be a second loose object header format which
1075 * was meant to mimic the in-pack format, allowing for direct
1076 * copy of the object data. This format turned up not to be
1077 * really worth it and we don't write it any longer. But we
1078 * can still read it.
1080 used = unpack_object_header_gently(map, mapsize, &type, &size);
1081 if (!used || !valid_loose_object_type[type])
1082 return -1;
1083 map += used;
1084 mapsize -= used;
1086 /* Set up the stream for the rest.. */
1087 stream->next_in = map;
1088 stream->avail_in = mapsize;
1089 inflateInit(stream);
1091 /* And generate the fake traditional header */
1092 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1093 typename(type), size);
1094 return 0;
1097 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1099 int bytes = strlen(buffer) + 1;
1100 unsigned char *buf = xmalloc(1+size);
1101 unsigned long n;
1102 int status = Z_OK;
1104 n = stream->total_out - bytes;
1105 if (n > size)
1106 n = size;
1107 memcpy(buf, (char *) buffer + bytes, n);
1108 bytes = n;
1109 if (bytes <= size) {
1111 * The above condition must be (bytes <= size), not
1112 * (bytes < size). In other words, even though we
1113 * expect no more output and set avail_out to zer0,
1114 * the input zlib stream may have bytes that express
1115 * "this concludes the stream", and we *do* want to
1116 * eat that input.
1118 * Otherwise we would not be able to test that we
1119 * consumed all the input to reach the expected size;
1120 * we also want to check that zlib tells us that all
1121 * went well with status == Z_STREAM_END at the end.
1123 stream->next_out = buf + bytes;
1124 stream->avail_out = size - bytes;
1125 while (status == Z_OK)
1126 status = inflate(stream, Z_FINISH);
1128 buf[size] = 0;
1129 if (status == Z_STREAM_END && !stream->avail_in) {
1130 inflateEnd(stream);
1131 return buf;
1134 if (status < 0)
1135 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1136 else if (stream->avail_in)
1137 error("garbage at end of loose object '%s'",
1138 sha1_to_hex(sha1));
1139 free(buf);
1140 return NULL;
1144 * We used to just use "sscanf()", but that's actually way
1145 * too permissive for what we want to check. So do an anal
1146 * object header parse by hand.
1148 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1150 char type[10];
1151 int i;
1152 unsigned long size;
1155 * The type can be at most ten bytes (including the
1156 * terminating '\0' that we add), and is followed by
1157 * a space.
1159 i = 0;
1160 for (;;) {
1161 char c = *hdr++;
1162 if (c == ' ')
1163 break;
1164 type[i++] = c;
1165 if (i >= sizeof(type))
1166 return -1;
1168 type[i] = 0;
1171 * The length must follow immediately, and be in canonical
1172 * decimal format (ie "010" is not valid).
1174 size = *hdr++ - '0';
1175 if (size > 9)
1176 return -1;
1177 if (size) {
1178 for (;;) {
1179 unsigned long c = *hdr - '0';
1180 if (c > 9)
1181 break;
1182 hdr++;
1183 size = size * 10 + c;
1186 *sizep = size;
1189 * The length must be followed by a zero byte
1191 return *hdr ? -1 : type_from_string(type);
1194 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1196 int ret;
1197 z_stream stream;
1198 char hdr[8192];
1200 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1201 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1202 return NULL;
1204 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1207 unsigned long get_size_from_delta(struct packed_git *p,
1208 struct pack_window **w_curs,
1209 off_t curpos)
1211 const unsigned char *data;
1212 unsigned char delta_head[20], *in;
1213 z_stream stream;
1214 int st;
1216 memset(&stream, 0, sizeof(stream));
1217 stream.next_out = delta_head;
1218 stream.avail_out = sizeof(delta_head);
1220 inflateInit(&stream);
1221 do {
1222 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1223 stream.next_in = in;
1224 st = inflate(&stream, Z_FINISH);
1225 curpos += stream.next_in - in;
1226 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1227 stream.total_out < sizeof(delta_head));
1228 inflateEnd(&stream);
1229 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1230 die("delta data unpack-initial failed");
1232 /* Examine the initial part of the delta to figure out
1233 * the result size.
1235 data = delta_head;
1237 /* ignore base size */
1238 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1240 /* Read the result size */
1241 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1244 static off_t get_delta_base(struct packed_git *p,
1245 struct pack_window **w_curs,
1246 off_t *curpos,
1247 enum object_type type,
1248 off_t delta_obj_offset)
1250 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1251 off_t base_offset;
1253 /* use_pack() assured us we have [base_info, base_info + 20)
1254 * as a range that we can look at without walking off the
1255 * end of the mapped window. Its actually the hash size
1256 * that is assured. An OFS_DELTA longer than the hash size
1257 * is stupid, as then a REF_DELTA would be smaller to store.
1259 if (type == OBJ_OFS_DELTA) {
1260 unsigned used = 0;
1261 unsigned char c = base_info[used++];
1262 base_offset = c & 127;
1263 while (c & 128) {
1264 base_offset += 1;
1265 if (!base_offset || MSB(base_offset, 7))
1266 die("offset value overflow for delta base object");
1267 c = base_info[used++];
1268 base_offset = (base_offset << 7) + (c & 127);
1270 base_offset = delta_obj_offset - base_offset;
1271 if (base_offset >= delta_obj_offset)
1272 die("delta base offset out of bound");
1273 *curpos += used;
1274 } else if (type == OBJ_REF_DELTA) {
1275 /* The base entry _must_ be in the same pack */
1276 base_offset = find_pack_entry_one(base_info, p);
1277 if (!base_offset)
1278 die("failed to find delta-pack base object %s",
1279 sha1_to_hex(base_info));
1280 *curpos += 20;
1281 } else
1282 die("I am totally screwed");
1283 return base_offset;
1286 /* forward declaration for a mutually recursive function */
1287 static int packed_object_info(struct packed_git *p, off_t offset,
1288 unsigned long *sizep);
1290 static int packed_delta_info(struct packed_git *p,
1291 struct pack_window **w_curs,
1292 off_t curpos,
1293 enum object_type type,
1294 off_t obj_offset,
1295 unsigned long *sizep)
1297 off_t base_offset;
1299 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1300 type = packed_object_info(p, base_offset, NULL);
1302 /* We choose to only get the type of the base object and
1303 * ignore potentially corrupt pack file that expects the delta
1304 * based on a base with a wrong size. This saves tons of
1305 * inflate() calls.
1307 if (sizep)
1308 *sizep = get_size_from_delta(p, w_curs, curpos);
1310 return type;
1313 static int unpack_object_header(struct packed_git *p,
1314 struct pack_window **w_curs,
1315 off_t *curpos,
1316 unsigned long *sizep)
1318 unsigned char *base;
1319 unsigned int left;
1320 unsigned long used;
1321 enum object_type type;
1323 /* use_pack() assures us we have [base, base + 20) available
1324 * as a range that we can look at at. (Its actually the hash
1325 * size that is assured.) With our object header encoding
1326 * the maximum deflated object size is 2^137, which is just
1327 * insane, so we know won't exceed what we have been given.
1329 base = use_pack(p, w_curs, *curpos, &left);
1330 used = unpack_object_header_gently(base, left, &type, sizep);
1331 if (!used)
1332 die("object offset outside of pack file");
1333 *curpos += used;
1335 return type;
1338 const char *packed_object_info_detail(struct packed_git *p,
1339 off_t obj_offset,
1340 unsigned long *size,
1341 unsigned long *store_size,
1342 unsigned int *delta_chain_length,
1343 unsigned char *base_sha1)
1345 struct pack_window *w_curs = NULL;
1346 off_t curpos;
1347 unsigned long dummy;
1348 unsigned char *next_sha1;
1349 enum object_type type;
1351 *delta_chain_length = 0;
1352 curpos = obj_offset;
1353 type = unpack_object_header(p, &w_curs, &curpos, size);
1355 for (;;) {
1356 switch (type) {
1357 default:
1358 die("pack %s contains unknown object type %d",
1359 p->pack_name, type);
1360 case OBJ_COMMIT:
1361 case OBJ_TREE:
1362 case OBJ_BLOB:
1363 case OBJ_TAG:
1364 *store_size = 0; /* notyet */
1365 unuse_pack(&w_curs);
1366 return typename(type);
1367 case OBJ_OFS_DELTA:
1368 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1369 if (*delta_chain_length == 0) {
1370 /* TODO: find base_sha1 as pointed by curpos */
1371 hashclr(base_sha1);
1373 break;
1374 case OBJ_REF_DELTA:
1375 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1376 if (*delta_chain_length == 0)
1377 hashcpy(base_sha1, next_sha1);
1378 obj_offset = find_pack_entry_one(next_sha1, p);
1379 break;
1381 (*delta_chain_length)++;
1382 curpos = obj_offset;
1383 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1387 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1388 unsigned long *sizep)
1390 struct pack_window *w_curs = NULL;
1391 unsigned long size;
1392 off_t curpos = obj_offset;
1393 enum object_type type;
1395 type = unpack_object_header(p, &w_curs, &curpos, &size);
1397 switch (type) {
1398 case OBJ_OFS_DELTA:
1399 case OBJ_REF_DELTA:
1400 type = packed_delta_info(p, &w_curs, curpos,
1401 type, obj_offset, sizep);
1402 break;
1403 case OBJ_COMMIT:
1404 case OBJ_TREE:
1405 case OBJ_BLOB:
1406 case OBJ_TAG:
1407 if (sizep)
1408 *sizep = size;
1409 break;
1410 default:
1411 die("pack %s contains unknown object type %d",
1412 p->pack_name, type);
1414 unuse_pack(&w_curs);
1415 return type;
1418 static void *unpack_compressed_entry(struct packed_git *p,
1419 struct pack_window **w_curs,
1420 off_t curpos,
1421 unsigned long size)
1423 int st;
1424 z_stream stream;
1425 unsigned char *buffer, *in;
1427 buffer = xmalloc(size + 1);
1428 buffer[size] = 0;
1429 memset(&stream, 0, sizeof(stream));
1430 stream.next_out = buffer;
1431 stream.avail_out = size;
1433 inflateInit(&stream);
1434 do {
1435 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1436 stream.next_in = in;
1437 st = inflate(&stream, Z_FINISH);
1438 curpos += stream.next_in - in;
1439 } while (st == Z_OK || st == Z_BUF_ERROR);
1440 inflateEnd(&stream);
1441 if ((st != Z_STREAM_END) || stream.total_out != size) {
1442 free(buffer);
1443 return NULL;
1446 return buffer;
1449 #define MAX_DELTA_CACHE (256)
1451 static size_t delta_base_cached;
1453 static struct delta_base_cache_lru_list {
1454 struct delta_base_cache_lru_list *prev;
1455 struct delta_base_cache_lru_list *next;
1456 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1458 static struct delta_base_cache_entry {
1459 struct delta_base_cache_lru_list lru;
1460 void *data;
1461 struct packed_git *p;
1462 off_t base_offset;
1463 unsigned long size;
1464 enum object_type type;
1465 } delta_base_cache[MAX_DELTA_CACHE];
1467 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1469 unsigned long hash;
1471 hash = (unsigned long)p + (unsigned long)base_offset;
1472 hash += (hash >> 8) + (hash >> 16);
1473 return hash % MAX_DELTA_CACHE;
1476 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1477 unsigned long *base_size, enum object_type *type, int keep_cache)
1479 void *ret;
1480 unsigned long hash = pack_entry_hash(p, base_offset);
1481 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1483 ret = ent->data;
1484 if (ret && ent->p == p && ent->base_offset == base_offset)
1485 goto found_cache_entry;
1486 return unpack_entry(p, base_offset, type, base_size);
1488 found_cache_entry:
1489 if (!keep_cache) {
1490 ent->data = NULL;
1491 ent->lru.next->prev = ent->lru.prev;
1492 ent->lru.prev->next = ent->lru.next;
1493 delta_base_cached -= ent->size;
1495 else {
1496 ret = xmalloc(ent->size + 1);
1497 memcpy(ret, ent->data, ent->size);
1498 ((char *)ret)[ent->size] = 0;
1500 *type = ent->type;
1501 *base_size = ent->size;
1502 return ret;
1505 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1507 if (ent->data) {
1508 free(ent->data);
1509 ent->data = NULL;
1510 ent->lru.next->prev = ent->lru.prev;
1511 ent->lru.prev->next = ent->lru.next;
1512 delta_base_cached -= ent->size;
1516 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1517 void *base, unsigned long base_size, enum object_type type)
1519 unsigned long hash = pack_entry_hash(p, base_offset);
1520 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1521 struct delta_base_cache_lru_list *lru;
1523 release_delta_base_cache(ent);
1524 delta_base_cached += base_size;
1526 for (lru = delta_base_cache_lru.next;
1527 delta_base_cached > delta_base_cache_limit
1528 && lru != &delta_base_cache_lru;
1529 lru = lru->next) {
1530 struct delta_base_cache_entry *f = (void *)lru;
1531 if (f->type == OBJ_BLOB)
1532 release_delta_base_cache(f);
1534 for (lru = delta_base_cache_lru.next;
1535 delta_base_cached > delta_base_cache_limit
1536 && lru != &delta_base_cache_lru;
1537 lru = lru->next) {
1538 struct delta_base_cache_entry *f = (void *)lru;
1539 release_delta_base_cache(f);
1542 ent->p = p;
1543 ent->base_offset = base_offset;
1544 ent->type = type;
1545 ent->data = base;
1546 ent->size = base_size;
1547 ent->lru.next = &delta_base_cache_lru;
1548 ent->lru.prev = delta_base_cache_lru.prev;
1549 delta_base_cache_lru.prev->next = &ent->lru;
1550 delta_base_cache_lru.prev = &ent->lru;
1553 static void *unpack_delta_entry(struct packed_git *p,
1554 struct pack_window **w_curs,
1555 off_t curpos,
1556 unsigned long delta_size,
1557 off_t obj_offset,
1558 enum object_type *type,
1559 unsigned long *sizep)
1561 void *delta_data, *result, *base;
1562 unsigned long base_size;
1563 off_t base_offset;
1565 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1566 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1567 if (!base)
1568 die("failed to read delta base object"
1569 " at %"PRIuMAX" from %s",
1570 (uintmax_t)base_offset, p->pack_name);
1572 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1573 result = patch_delta(base, base_size,
1574 delta_data, delta_size,
1575 sizep);
1576 if (!result)
1577 die("failed to apply delta");
1578 free(delta_data);
1579 add_delta_base_cache(p, base_offset, base, base_size, *type);
1580 return result;
1583 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1584 enum object_type *type, unsigned long *sizep)
1586 struct pack_window *w_curs = NULL;
1587 off_t curpos = obj_offset;
1588 void *data;
1590 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1591 switch (*type) {
1592 case OBJ_OFS_DELTA:
1593 case OBJ_REF_DELTA:
1594 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1595 obj_offset, type, sizep);
1596 break;
1597 case OBJ_COMMIT:
1598 case OBJ_TREE:
1599 case OBJ_BLOB:
1600 case OBJ_TAG:
1601 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1602 break;
1603 default:
1604 die("unknown object type %i in %s", *type, p->pack_name);
1606 unuse_pack(&w_curs);
1607 return data;
1610 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1611 uint32_t n)
1613 const unsigned char *index = p->index_data;
1614 if (!index) {
1615 if (open_pack_index(p))
1616 return NULL;
1617 index = p->index_data;
1619 if (n >= p->num_objects)
1620 return NULL;
1621 index += 4 * 256;
1622 if (p->index_version == 1) {
1623 return index + 24 * n + 4;
1624 } else {
1625 index += 8;
1626 return index + 20 * n;
1630 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1632 const unsigned char *index = p->index_data;
1633 index += 4 * 256;
1634 if (p->index_version == 1) {
1635 return ntohl(*((uint32_t *)(index + 24 * n)));
1636 } else {
1637 uint32_t off;
1638 index += 8 + p->num_objects * (20 + 4);
1639 off = ntohl(*((uint32_t *)(index + 4 * n)));
1640 if (!(off & 0x80000000))
1641 return off;
1642 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1643 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1644 ntohl(*((uint32_t *)(index + 4)));
1648 off_t find_pack_entry_one(const unsigned char *sha1,
1649 struct packed_git *p)
1651 const uint32_t *level1_ofs = p->index_data;
1652 const unsigned char *index = p->index_data;
1653 unsigned hi, lo;
1655 if (!index) {
1656 if (open_pack_index(p))
1657 return 0;
1658 level1_ofs = p->index_data;
1659 index = p->index_data;
1661 if (p->index_version > 1) {
1662 level1_ofs += 2;
1663 index += 8;
1665 index += 4 * 256;
1666 hi = ntohl(level1_ofs[*sha1]);
1667 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1669 do {
1670 unsigned mi = (lo + hi) / 2;
1671 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1672 int cmp = hashcmp(index + x, sha1);
1673 if (!cmp)
1674 return nth_packed_object_offset(p, mi);
1675 if (cmp > 0)
1676 hi = mi;
1677 else
1678 lo = mi+1;
1679 } while (lo < hi);
1680 return 0;
1683 static int matches_pack_name(struct packed_git *p, const char *ig)
1685 const char *last_c, *c;
1687 if (!strcmp(p->pack_name, ig))
1688 return 0;
1690 for (c = p->pack_name, last_c = c; *c;)
1691 if (*c == '/')
1692 last_c = ++c;
1693 else
1694 ++c;
1695 if (!strcmp(last_c, ig))
1696 return 0;
1698 return 1;
1701 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1703 static struct packed_git *last_found = (void *)1;
1704 struct packed_git *p;
1705 off_t offset;
1707 prepare_packed_git();
1708 if (!packed_git)
1709 return 0;
1710 p = (last_found == (void *)1) ? packed_git : last_found;
1712 do {
1713 if (ignore_packed) {
1714 const char **ig;
1715 for (ig = ignore_packed; *ig; ig++)
1716 if (!matches_pack_name(p, *ig))
1717 break;
1718 if (*ig)
1719 goto next;
1722 offset = find_pack_entry_one(sha1, p);
1723 if (offset) {
1725 * We are about to tell the caller where they can
1726 * locate the requested object. We better make
1727 * sure the packfile is still here and can be
1728 * accessed before supplying that answer, as
1729 * it may have been deleted since the index
1730 * was loaded!
1732 if (p->pack_fd == -1 && open_packed_git(p)) {
1733 error("packfile %s cannot be accessed", p->pack_name);
1734 goto next;
1736 e->offset = offset;
1737 e->p = p;
1738 hashcpy(e->sha1, sha1);
1739 last_found = p;
1740 return 1;
1743 next:
1744 if (p == last_found)
1745 p = packed_git;
1746 else
1747 p = p->next;
1748 if (p == last_found)
1749 p = p->next;
1750 } while (p);
1751 return 0;
1754 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1755 struct packed_git *packs)
1757 struct packed_git *p;
1759 for (p = packs; p; p = p->next) {
1760 if (find_pack_entry_one(sha1, p))
1761 return p;
1763 return NULL;
1767 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1769 int status;
1770 unsigned long mapsize, size;
1771 void *map;
1772 z_stream stream;
1773 char hdr[32];
1775 map = map_sha1_file(sha1, &mapsize);
1776 if (!map)
1777 return error("unable to find %s", sha1_to_hex(sha1));
1778 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1779 status = error("unable to unpack %s header",
1780 sha1_to_hex(sha1));
1781 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1782 status = error("unable to parse %s header", sha1_to_hex(sha1));
1783 else if (sizep)
1784 *sizep = size;
1785 inflateEnd(&stream);
1786 munmap(map, mapsize);
1787 return status;
1790 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1792 struct pack_entry e;
1794 if (!find_pack_entry(sha1, &e, NULL)) {
1795 reprepare_packed_git();
1796 if (!find_pack_entry(sha1, &e, NULL))
1797 return sha1_loose_object_info(sha1, sizep);
1799 return packed_object_info(e.p, e.offset, sizep);
1802 static void *read_packed_sha1(const unsigned char *sha1,
1803 enum object_type *type, unsigned long *size)
1805 struct pack_entry e;
1807 if (!find_pack_entry(sha1, &e, NULL))
1808 return NULL;
1809 else
1810 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1814 * This is meant to hold a *small* number of objects that you would
1815 * want read_sha1_file() to be able to return, but yet you do not want
1816 * to write them into the object store (e.g. a browse-only
1817 * application).
1819 static struct cached_object {
1820 unsigned char sha1[20];
1821 enum object_type type;
1822 void *buf;
1823 unsigned long size;
1824 } *cached_objects;
1825 static int cached_object_nr, cached_object_alloc;
1827 static struct cached_object *find_cached_object(const unsigned char *sha1)
1829 int i;
1830 struct cached_object *co = cached_objects;
1832 for (i = 0; i < cached_object_nr; i++, co++) {
1833 if (!hashcmp(co->sha1, sha1))
1834 return co;
1836 return NULL;
1839 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1840 unsigned char *sha1)
1842 struct cached_object *co;
1844 hash_sha1_file(buf, len, typename(type), sha1);
1845 if (has_sha1_file(sha1) || find_cached_object(sha1))
1846 return 0;
1847 if (cached_object_alloc <= cached_object_nr) {
1848 cached_object_alloc = alloc_nr(cached_object_alloc);
1849 cached_objects = xrealloc(cached_objects,
1850 sizeof(*cached_objects) *
1851 cached_object_alloc);
1853 co = &cached_objects[cached_object_nr++];
1854 co->size = len;
1855 co->type = type;
1856 co->buf = xmalloc(len);
1857 memcpy(co->buf, buf, len);
1858 hashcpy(co->sha1, sha1);
1859 return 0;
1862 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1863 unsigned long *size)
1865 unsigned long mapsize;
1866 void *map, *buf;
1867 struct cached_object *co;
1869 co = find_cached_object(sha1);
1870 if (co) {
1871 buf = xmalloc(co->size + 1);
1872 memcpy(buf, co->buf, co->size);
1873 ((char*)buf)[co->size] = 0;
1874 *type = co->type;
1875 *size = co->size;
1876 return buf;
1879 buf = read_packed_sha1(sha1, type, size);
1880 if (buf)
1881 return buf;
1882 map = map_sha1_file(sha1, &mapsize);
1883 if (map) {
1884 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1885 munmap(map, mapsize);
1886 return buf;
1888 reprepare_packed_git();
1889 return read_packed_sha1(sha1, type, size);
1892 void *read_object_with_reference(const unsigned char *sha1,
1893 const char *required_type_name,
1894 unsigned long *size,
1895 unsigned char *actual_sha1_return)
1897 enum object_type type, required_type;
1898 void *buffer;
1899 unsigned long isize;
1900 unsigned char actual_sha1[20];
1902 required_type = type_from_string(required_type_name);
1903 hashcpy(actual_sha1, sha1);
1904 while (1) {
1905 int ref_length = -1;
1906 const char *ref_type = NULL;
1908 buffer = read_sha1_file(actual_sha1, &type, &isize);
1909 if (!buffer)
1910 return NULL;
1911 if (type == required_type) {
1912 *size = isize;
1913 if (actual_sha1_return)
1914 hashcpy(actual_sha1_return, actual_sha1);
1915 return buffer;
1917 /* Handle references */
1918 else if (type == OBJ_COMMIT)
1919 ref_type = "tree ";
1920 else if (type == OBJ_TAG)
1921 ref_type = "object ";
1922 else {
1923 free(buffer);
1924 return NULL;
1926 ref_length = strlen(ref_type);
1928 if (memcmp(buffer, ref_type, ref_length) ||
1929 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1930 free(buffer);
1931 return NULL;
1933 free(buffer);
1934 /* Now we have the ID of the referred-to object in
1935 * actual_sha1. Check again. */
1939 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1940 const char *type, unsigned char *sha1,
1941 char *hdr, int *hdrlen)
1943 SHA_CTX c;
1945 /* Generate the header */
1946 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1948 /* Sha1.. */
1949 SHA1_Init(&c);
1950 SHA1_Update(&c, hdr, *hdrlen);
1951 SHA1_Update(&c, buf, len);
1952 SHA1_Final(sha1, &c);
1956 * Link the tempfile to the final place, possibly creating the
1957 * last directory level as you do so.
1959 * Returns the errno on failure, 0 on success.
1961 static int link_temp_to_file(const char *tmpfile, const char *filename)
1963 int ret;
1964 char *dir;
1966 if (!link(tmpfile, filename))
1967 return 0;
1970 * Try to mkdir the last path component if that failed.
1972 * Re-try the "link()" regardless of whether the mkdir
1973 * succeeds, since a race might mean that somebody
1974 * else succeeded.
1976 ret = errno;
1977 dir = strrchr(filename, '/');
1978 if (dir) {
1979 *dir = 0;
1980 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1981 *dir = '/';
1982 return -2;
1984 *dir = '/';
1985 if (!link(tmpfile, filename))
1986 return 0;
1987 ret = errno;
1989 return ret;
1993 * Move the just written object into its final resting place
1995 int move_temp_to_file(const char *tmpfile, const char *filename)
1997 int ret = link_temp_to_file(tmpfile, filename);
2000 * Coda hack - coda doesn't like cross-directory links,
2001 * so we fall back to a rename, which will mean that it
2002 * won't be able to check collisions, but that's not a
2003 * big deal.
2005 * The same holds for FAT formatted media.
2007 * When this succeeds, we just return 0. We have nothing
2008 * left to unlink.
2010 if (ret && ret != EEXIST) {
2011 if (!rename(tmpfile, filename))
2012 return 0;
2013 ret = errno;
2015 unlink(tmpfile);
2016 if (ret) {
2017 if (ret != EEXIST) {
2018 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2020 /* FIXME!!! Collision check here ? */
2023 return 0;
2026 static int write_buffer(int fd, const void *buf, size_t len)
2028 if (write_in_full(fd, buf, len) < 0)
2029 return error("file write error (%s)", strerror(errno));
2030 return 0;
2033 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2034 unsigned char *sha1)
2036 char hdr[32];
2037 int hdrlen;
2038 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2039 return 0;
2042 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2044 int size, ret;
2045 unsigned char *compressed;
2046 z_stream stream;
2047 unsigned char sha1[20];
2048 char *filename;
2049 static char tmpfile[PATH_MAX];
2050 char hdr[32];
2051 int fd, hdrlen;
2053 /* Normally if we have it in the pack then we do not bother writing
2054 * it out into .git/objects/??/?{38} file.
2056 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2057 filename = sha1_file_name(sha1);
2058 if (returnsha1)
2059 hashcpy(returnsha1, sha1);
2060 if (has_sha1_file(sha1))
2061 return 0;
2062 fd = open(filename, O_RDONLY);
2063 if (fd >= 0) {
2065 * FIXME!!! We might do collision checking here, but we'd
2066 * need to uncompress the old file and check it. Later.
2068 close(fd);
2069 return 0;
2072 if (errno != ENOENT) {
2073 return error("sha1 file %s: %s\n", filename, strerror(errno));
2076 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2078 fd = mkstemp(tmpfile);
2079 if (fd < 0) {
2080 if (errno == EPERM)
2081 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2082 else
2083 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2086 /* Set it up */
2087 memset(&stream, 0, sizeof(stream));
2088 deflateInit(&stream, zlib_compression_level);
2089 size = 8 + deflateBound(&stream, len+hdrlen);
2090 compressed = xmalloc(size);
2092 /* Compress it */
2093 stream.next_out = compressed;
2094 stream.avail_out = size;
2096 /* First header.. */
2097 stream.next_in = (unsigned char *)hdr;
2098 stream.avail_in = hdrlen;
2099 while (deflate(&stream, 0) == Z_OK)
2100 /* nothing */;
2102 /* Then the data itself.. */
2103 stream.next_in = buf;
2104 stream.avail_in = len;
2105 ret = deflate(&stream, Z_FINISH);
2106 if (ret != Z_STREAM_END)
2107 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2109 ret = deflateEnd(&stream);
2110 if (ret != Z_OK)
2111 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2113 size = stream.total_out;
2115 if (write_buffer(fd, compressed, size) < 0)
2116 die("unable to write sha1 file");
2117 fchmod(fd, 0444);
2118 if (close(fd))
2119 die("unable to write sha1 file");
2120 free(compressed);
2122 return move_temp_to_file(tmpfile, filename);
2126 * We need to unpack and recompress the object for writing
2127 * it out to a different file.
2129 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2131 size_t size;
2132 z_stream stream;
2133 unsigned char *unpacked;
2134 unsigned long len;
2135 enum object_type type;
2136 char hdr[32];
2137 int hdrlen;
2138 void *buf;
2140 /* need to unpack and recompress it by itself */
2141 unpacked = read_packed_sha1(sha1, &type, &len);
2142 if (!unpacked)
2143 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2145 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2147 /* Set it up */
2148 memset(&stream, 0, sizeof(stream));
2149 deflateInit(&stream, zlib_compression_level);
2150 size = deflateBound(&stream, len + hdrlen);
2151 buf = xmalloc(size);
2153 /* Compress it */
2154 stream.next_out = buf;
2155 stream.avail_out = size;
2157 /* First header.. */
2158 stream.next_in = (void *)hdr;
2159 stream.avail_in = hdrlen;
2160 while (deflate(&stream, 0) == Z_OK)
2161 /* nothing */;
2163 /* Then the data itself.. */
2164 stream.next_in = unpacked;
2165 stream.avail_in = len;
2166 while (deflate(&stream, Z_FINISH) == Z_OK)
2167 /* nothing */;
2168 deflateEnd(&stream);
2169 free(unpacked);
2171 *objsize = stream.total_out;
2172 return buf;
2175 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2177 int retval;
2178 unsigned long objsize;
2179 void *buf = map_sha1_file(sha1, &objsize);
2181 if (buf) {
2182 retval = write_buffer(fd, buf, objsize);
2183 munmap(buf, objsize);
2184 return retval;
2187 buf = repack_object(sha1, &objsize);
2188 retval = write_buffer(fd, buf, objsize);
2189 free(buf);
2190 return retval;
2193 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2194 size_t bufsize, size_t *bufposn)
2196 char tmpfile[PATH_MAX];
2197 int local;
2198 z_stream stream;
2199 unsigned char real_sha1[20];
2200 unsigned char discard[4096];
2201 int ret;
2202 SHA_CTX c;
2204 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2206 local = mkstemp(tmpfile);
2207 if (local < 0) {
2208 if (errno == EPERM)
2209 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2210 else
2211 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2214 memset(&stream, 0, sizeof(stream));
2216 inflateInit(&stream);
2218 SHA1_Init(&c);
2220 do {
2221 ssize_t size;
2222 if (*bufposn) {
2223 stream.avail_in = *bufposn;
2224 stream.next_in = (unsigned char *) buffer;
2225 do {
2226 stream.next_out = discard;
2227 stream.avail_out = sizeof(discard);
2228 ret = inflate(&stream, Z_SYNC_FLUSH);
2229 SHA1_Update(&c, discard, sizeof(discard) -
2230 stream.avail_out);
2231 } while (stream.avail_in && ret == Z_OK);
2232 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2233 die("unable to write sha1 file");
2234 memmove(buffer, buffer + *bufposn - stream.avail_in,
2235 stream.avail_in);
2236 *bufposn = stream.avail_in;
2237 if (ret != Z_OK)
2238 break;
2240 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2241 if (size <= 0) {
2242 close(local);
2243 unlink(tmpfile);
2244 if (!size)
2245 return error("Connection closed?");
2246 perror("Reading from connection");
2247 return -1;
2249 *bufposn += size;
2250 } while (1);
2251 inflateEnd(&stream);
2253 fchmod(local, 0444);
2254 if (close(local) != 0)
2255 die("unable to write sha1 file");
2256 SHA1_Final(real_sha1, &c);
2257 if (ret != Z_STREAM_END) {
2258 unlink(tmpfile);
2259 return error("File %s corrupted", sha1_to_hex(sha1));
2261 if (hashcmp(sha1, real_sha1)) {
2262 unlink(tmpfile);
2263 return error("File %s has bad hash", sha1_to_hex(sha1));
2266 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2269 int has_pack_index(const unsigned char *sha1)
2271 struct stat st;
2272 if (stat(sha1_pack_index_name(sha1), &st))
2273 return 0;
2274 return 1;
2277 int has_pack_file(const unsigned char *sha1)
2279 struct stat st;
2280 if (stat(sha1_pack_name(sha1), &st))
2281 return 0;
2282 return 1;
2285 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2287 struct pack_entry e;
2288 return find_pack_entry(sha1, &e, ignore_packed);
2291 int has_sha1_file(const unsigned char *sha1)
2293 struct stat st;
2294 struct pack_entry e;
2296 if (find_pack_entry(sha1, &e, NULL))
2297 return 1;
2298 return find_sha1_file(sha1, &st) ? 1 : 0;
2302 * reads from fd as long as possible into a supplied buffer of size bytes.
2303 * If necessary the buffer's size is increased using realloc()
2305 * returns 0 if anything went fine and -1 otherwise
2307 * The buffer is always NUL-terminated, not including it in returned size.
2309 * NOTE: both buf and size may change, but even when -1 is returned
2310 * you still have to free() it yourself.
2312 int read_fd(int fd, char **return_buf, unsigned long *return_size)
2314 char *buf = *return_buf;
2315 unsigned long size = *return_size;
2316 ssize_t iret;
2317 unsigned long off = 0;
2319 if (!buf || size <= 1) {
2320 size = 1024;
2321 buf = xrealloc(buf, size);
2324 do {
2325 iret = xread(fd, buf + off, (size - 1) - off);
2326 if (iret > 0) {
2327 off += iret;
2328 if (off == size - 1) {
2329 size = alloc_nr(size);
2330 buf = xrealloc(buf, size);
2333 } while (iret > 0);
2335 buf[off] = '\0';
2337 *return_buf = buf;
2338 *return_size = off;
2340 if (iret < 0)
2341 return -1;
2342 return 0;
2345 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2347 unsigned long size = 4096;
2348 char *buf = xmalloc(size);
2349 int ret;
2351 if (read_fd(fd, &buf, &size)) {
2352 free(buf);
2353 return -1;
2356 if (!type)
2357 type = blob_type;
2358 if (write_object)
2359 ret = write_sha1_file(buf, size, type, sha1);
2360 else
2361 ret = hash_sha1_file(buf, size, type, sha1);
2362 free(buf);
2363 return ret;
2366 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2367 enum object_type type, const char *path)
2369 size_t size = xsize_t(st->st_size);
2370 void *buf = NULL;
2371 int ret, re_allocated = 0;
2373 if (size)
2374 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2375 close(fd);
2377 if (!type)
2378 type = OBJ_BLOB;
2381 * Convert blobs to git internal format
2383 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2384 unsigned long nsize = size;
2385 char *nbuf = convert_to_git(path, buf, &nsize);
2386 if (nbuf) {
2387 munmap(buf, size);
2388 size = nsize;
2389 buf = nbuf;
2390 re_allocated = 1;
2394 if (write_object)
2395 ret = write_sha1_file(buf, size, typename(type), sha1);
2396 else
2397 ret = hash_sha1_file(buf, size, typename(type), sha1);
2398 if (re_allocated) {
2399 free(buf);
2400 return ret;
2402 if (size)
2403 munmap(buf, size);
2404 return ret;
2407 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2409 int fd;
2410 char *target;
2411 size_t len;
2413 switch (st->st_mode & S_IFMT) {
2414 case S_IFREG:
2415 fd = open(path, O_RDONLY);
2416 if (fd < 0)
2417 return error("open(\"%s\"): %s", path,
2418 strerror(errno));
2419 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2420 return error("%s: failed to insert into database",
2421 path);
2422 break;
2423 case S_IFLNK:
2424 len = xsize_t(st->st_size);
2425 target = xmalloc(len + 1);
2426 if (readlink(path, target, len + 1) != st->st_size) {
2427 char *errstr = strerror(errno);
2428 free(target);
2429 return error("readlink(\"%s\"): %s", path,
2430 errstr);
2432 if (!write_object)
2433 hash_sha1_file(target, len, blob_type, sha1);
2434 else if (write_sha1_file(target, len, blob_type, sha1))
2435 return error("%s: failed to insert into database",
2436 path);
2437 free(target);
2438 break;
2439 case S_IFDIR:
2440 return resolve_gitlink_ref(path, "HEAD", sha1);
2441 default:
2442 return error("%s: unsupported file type", path);
2444 return 0;
2447 int read_pack_header(int fd, struct pack_header *header)
2449 char *c = (char*)header;
2450 ssize_t remaining = sizeof(struct pack_header);
2451 do {
2452 ssize_t r = xread(fd, c, remaining);
2453 if (r <= 0)
2454 /* "eof before pack header was fully read" */
2455 return PH_ERROR_EOF;
2456 remaining -= r;
2457 c += r;
2458 } while (remaining > 0);
2459 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2460 /* "protocol error (pack signature mismatch detected)" */
2461 return PH_ERROR_PACK_SIGNATURE;
2462 if (!pack_version_ok(header->hdr_version))
2463 /* "protocol error (pack version unsupported)" */
2464 return PH_ERROR_PROTOCOL;
2465 return 0;