packv4: nth_packed_object_sha1() packv4 support
[git/packv4.git] / sha1_file.c
blobdc9e6a19f0a409b9d34d14734a83c8fd0030f522
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "blob.h"
13 #include "commit.h"
14 #include "tag.h"
15 #include "tree.h"
16 #include "refs.h"
18 #ifndef O_NOATIME
19 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
20 #define O_NOATIME 01000000
21 #else
22 #define O_NOATIME 0
23 #endif
24 #endif
26 #ifdef NO_C99_FORMAT
27 #define SZ_FMT "lu"
28 #else
29 #define SZ_FMT "zu"
30 #endif
32 const unsigned char null_sha1[20];
34 static unsigned int sha1_file_open_flag = O_NOATIME;
36 const signed char hexval_table[256] = {
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
43 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
44 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
45 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
71 int get_sha1_hex(const char *hex, unsigned char *sha1)
73 int i;
74 for (i = 0; i < 20; i++) {
75 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
76 if (val & ~0xff)
77 return -1;
78 *sha1++ = val;
79 hex += 2;
81 return 0;
84 int safe_create_leading_directories(char *path)
86 char *pos = path;
87 struct stat st;
89 if (*pos == '/')
90 pos++;
92 while (pos) {
93 pos = strchr(pos, '/');
94 if (!pos)
95 break;
96 *pos = 0;
97 if (!stat(path, &st)) {
98 /* path exists */
99 if (!S_ISDIR(st.st_mode)) {
100 *pos = '/';
101 return -3;
104 else if (mkdir(path, 0777)) {
105 *pos = '/';
106 return -1;
108 else if (adjust_shared_perm(path)) {
109 *pos = '/';
110 return -2;
112 *pos++ = '/';
114 return 0;
117 char * sha1_to_hex(const unsigned char *sha1)
119 static int bufno;
120 static char hexbuffer[4][50];
121 static const char hex[] = "0123456789abcdef";
122 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
123 int i;
125 for (i = 0; i < 20; i++) {
126 unsigned int val = *sha1++;
127 *buf++ = hex[val >> 4];
128 *buf++ = hex[val & 0xf];
130 *buf = '\0';
132 return buffer;
135 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
137 int i;
138 for (i = 0; i < 20; i++) {
139 static char hex[] = "0123456789abcdef";
140 unsigned int val = sha1[i];
141 char *pos = pathbuf + i*2 + (i > 0);
142 *pos++ = hex[val >> 4];
143 *pos = hex[val & 0xf];
148 * NOTE! This returns a statically allocated buffer, so you have to be
149 * careful about using it. Do a "xstrdup()" if you need to save the
150 * filename.
152 * Also note that this returns the location for creating. Reading
153 * SHA1 file can happen from any alternate directory listed in the
154 * DB_ENVIRONMENT environment variable if it is not found in
155 * the primary object database.
157 char *sha1_file_name(const unsigned char *sha1)
159 static char *name, *base;
161 if (!base) {
162 const char *sha1_file_directory = get_object_directory();
163 int len = strlen(sha1_file_directory);
164 base = xmalloc(len + 60);
165 memcpy(base, sha1_file_directory, len);
166 memset(base+len, 0, 60);
167 base[len] = '/';
168 base[len+3] = '/';
169 name = base + len + 1;
171 fill_sha1_path(name, sha1);
172 return base;
175 char *sha1_pack_name(const unsigned char *sha1)
177 static const char hex[] = "0123456789abcdef";
178 static char *name, *base, *buf;
179 int i;
181 if (!base) {
182 const char *sha1_file_directory = get_object_directory();
183 int len = strlen(sha1_file_directory);
184 base = xmalloc(len + 60);
185 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
186 name = base + len + 11;
189 buf = name;
191 for (i = 0; i < 20; i++) {
192 unsigned int val = *sha1++;
193 *buf++ = hex[val >> 4];
194 *buf++ = hex[val & 0xf];
197 return base;
200 char *sha1_pack_index_name(const unsigned char *sha1)
202 static const char hex[] = "0123456789abcdef";
203 static char *name, *base, *buf;
204 int i;
206 if (!base) {
207 const char *sha1_file_directory = get_object_directory();
208 int len = strlen(sha1_file_directory);
209 base = xmalloc(len + 60);
210 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
211 name = base + len + 11;
214 buf = name;
216 for (i = 0; i < 20; i++) {
217 unsigned int val = *sha1++;
218 *buf++ = hex[val >> 4];
219 *buf++ = hex[val & 0xf];
222 return base;
225 struct alternate_object_database *alt_odb_list;
226 static struct alternate_object_database **alt_odb_tail;
228 static void read_info_alternates(const char * alternates, int depth);
231 * Prepare alternate object database registry.
233 * The variable alt_odb_list points at the list of struct
234 * alternate_object_database. The elements on this list come from
235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
237 * whose contents is similar to that environment variable but can be
238 * LF separated. Its base points at a statically allocated buffer that
239 * contains "/the/directory/corresponding/to/.git/objects/...", while
240 * its name points just after the slash at the end of ".git/objects/"
241 * in the example above, and has enough space to hold 40-byte hex
242 * SHA1, an extra slash for the first level indirection, and the
243 * terminating NUL.
245 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
247 struct stat st;
248 const char *objdir = get_object_directory();
249 struct alternate_object_database *ent;
250 struct alternate_object_database *alt;
251 /* 43 = 40-byte + 2 '/' + terminating NUL */
252 int pfxlen = len;
253 int entlen = pfxlen + 43;
254 int base_len = -1;
256 if (*entry != '/' && relative_base) {
257 /* Relative alt-odb */
258 if (base_len < 0)
259 base_len = strlen(relative_base) + 1;
260 entlen += base_len;
261 pfxlen += base_len;
263 ent = xmalloc(sizeof(*ent) + entlen);
265 if (*entry != '/' && relative_base) {
266 memcpy(ent->base, relative_base, base_len - 1);
267 ent->base[base_len - 1] = '/';
268 memcpy(ent->base + base_len, entry, len);
270 else
271 memcpy(ent->base, entry, pfxlen);
273 ent->name = ent->base + pfxlen + 1;
274 ent->base[pfxlen + 3] = '/';
275 ent->base[pfxlen] = ent->base[entlen-1] = 0;
277 /* Detect cases where alternate disappeared */
278 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
279 error("object directory %s does not exist; "
280 "check .git/objects/info/alternates.",
281 ent->base);
282 free(ent);
283 return -1;
286 /* Prevent the common mistake of listing the same
287 * thing twice, or object directory itself.
289 for (alt = alt_odb_list; alt; alt = alt->next) {
290 if (!memcmp(ent->base, alt->base, pfxlen)) {
291 free(ent);
292 return -1;
295 if (!memcmp(ent->base, objdir, pfxlen)) {
296 free(ent);
297 return -1;
300 /* add the alternate entry */
301 *alt_odb_tail = ent;
302 alt_odb_tail = &(ent->next);
303 ent->next = NULL;
305 /* recursively add alternates */
306 read_info_alternates(ent->base, depth + 1);
308 ent->base[pfxlen] = '/';
310 return 0;
313 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
314 const char *relative_base, int depth)
316 const char *cp, *last;
318 if (depth > 5) {
319 error("%s: ignoring alternate object stores, nesting too deep.",
320 relative_base);
321 return;
324 last = alt;
325 while (last < ep) {
326 cp = last;
327 if (cp < ep && *cp == '#') {
328 while (cp < ep && *cp != sep)
329 cp++;
330 last = cp + 1;
331 continue;
333 while (cp < ep && *cp != sep)
334 cp++;
335 if (last != cp) {
336 if ((*last != '/') && depth) {
337 error("%s: ignoring relative alternate object store %s",
338 relative_base, last);
339 } else {
340 link_alt_odb_entry(last, cp - last,
341 relative_base, depth);
344 while (cp < ep && *cp == sep)
345 cp++;
346 last = cp;
350 static void read_info_alternates(const char * relative_base, int depth)
352 char *map;
353 size_t mapsz;
354 struct stat st;
355 const char alt_file_name[] = "info/alternates";
356 /* Given that relative_base is no longer than PATH_MAX,
357 ensure that "path" has enough space to append "/", the
358 file name, "info/alternates", and a trailing NUL. */
359 char path[PATH_MAX + 1 + sizeof alt_file_name];
360 int fd;
362 sprintf(path, "%s/%s", relative_base, alt_file_name);
363 fd = open(path, O_RDONLY);
364 if (fd < 0)
365 return;
366 if (fstat(fd, &st) || (st.st_size == 0)) {
367 close(fd);
368 return;
370 mapsz = xsize_t(st.st_size);
371 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
372 close(fd);
374 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
376 munmap(map, mapsz);
379 void prepare_alt_odb(void)
381 const char *alt;
383 if (alt_odb_tail)
384 return;
386 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
387 if (!alt) alt = "";
389 alt_odb_tail = &alt_odb_list;
390 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
392 read_info_alternates(get_object_directory(), 0);
395 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
397 char *name = sha1_file_name(sha1);
398 struct alternate_object_database *alt;
400 if (!stat(name, st))
401 return name;
402 prepare_alt_odb();
403 for (alt = alt_odb_list; alt; alt = alt->next) {
404 name = alt->name;
405 fill_sha1_path(name, sha1);
406 if (!stat(alt->base, st))
407 return alt->base;
409 return NULL;
412 static unsigned int pack_used_ctr;
413 static unsigned int pack_mmap_calls;
414 static unsigned int peak_pack_open_windows;
415 static unsigned int pack_open_windows;
416 static size_t peak_pack_mapped;
417 static size_t pack_mapped;
418 struct packed_git *packed_git;
420 void pack_report(void)
422 fprintf(stderr,
423 "pack_report: getpagesize() = %10" SZ_FMT "\n"
424 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
425 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
426 (size_t) getpagesize(),
427 packed_git_window_size,
428 packed_git_limit);
429 fprintf(stderr,
430 "pack_report: pack_used_ctr = %10u\n"
431 "pack_report: pack_mmap_calls = %10u\n"
432 "pack_report: pack_open_windows = %10u / %10u\n"
433 "pack_report: pack_mapped = "
434 "%10" SZ_FMT " / %10" SZ_FMT "\n",
435 pack_used_ctr,
436 pack_mmap_calls,
437 pack_open_windows, peak_pack_open_windows,
438 pack_mapped, peak_pack_mapped);
441 static int check_packed_git_idx(const char *path, struct packed_git *p)
443 void *idx_map;
444 struct pack_idx_header *hdr;
445 size_t idx_size;
446 uint32_t version, nr, i, *index;
447 int fd = open(path, O_RDONLY);
448 struct stat st;
450 if (fd < 0)
451 return -1;
452 if (fstat(fd, &st)) {
453 close(fd);
454 return -1;
456 idx_size = xsize_t(st.st_size);
457 if (idx_size < 4 * 256 + 20 + 20) {
458 close(fd);
459 return error("index file %s is too small", path);
461 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
462 close(fd);
464 hdr = idx_map;
465 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
466 version = ntohl(hdr->idx_version);
467 if (version < 2 || version > 3) {
468 munmap(idx_map, idx_size);
469 return error("index file %s is version %d"
470 " and is not supported by this binary"
471 " (try upgrading GIT to a newer version)",
472 path, version);
474 } else
475 version = 1;
477 nr = 0;
478 index = idx_map;
479 if (version > 1)
480 index += 2; /* skip index header */
481 for (i = 0; i < 256; i++) {
482 uint32_t n = ntohl(index[i]);
483 if (n < nr) {
484 munmap(idx_map, idx_size);
485 return error("non-monotonic index %s", path);
487 nr = n;
490 if (version == 1) {
492 * Total size:
493 * - 256 index entries 4 bytes each
494 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
495 * - 20-byte SHA1 of the packfile
496 * - 20-byte SHA1 file checksum
498 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
499 munmap(idx_map, idx_size);
500 return error("wrong index v1 file size in %s", path);
502 } else if (version > 1) {
504 * Minimum size:
505 * - 8 bytes of header
506 * - 256 index entries 4 bytes each
507 * - 20-byte sha1 entry * nr (only if version == 2)
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 + 20 + 20;
517 unsigned long max_size;
518 if (version == 2)
519 min_size += nr*(20 + 4 + 4);
520 else
521 min_size += nr*(4 + 4);
522 max_size = min_size;
523 if (nr)
524 max_size += (nr - 1)*8;
525 if (idx_size < min_size || idx_size > max_size) {
526 munmap(idx_map, idx_size);
527 return error("wrong index v%d file size in %s",
528 version, path);
530 if (idx_size != min_size) {
531 /* make sure we can deal with large pack offsets */
532 off_t x = 0x7fffffffUL, y = 0xffffffffUL;
533 if (x > (x + 1) || y > (y + 1)) {
534 munmap(idx_map, idx_size);
535 return error("pack too large for current definition of off_t in %s", path);
540 p->index_version = version;
541 p->index_data = idx_map;
542 p->index_size = idx_size;
543 p->num_objects = nr;
544 return 0;
547 int open_pack_index(struct packed_git *p)
549 char *idx_name;
550 int ret;
552 if (p->index_data)
553 return 0;
555 idx_name = xstrdup(p->pack_name);
556 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
557 ret = check_packed_git_idx(idx_name, p);
558 free(idx_name);
559 return ret;
562 static void scan_windows(struct packed_git *p,
563 struct packed_git **lru_p,
564 struct pack_window **lru_w,
565 struct pack_window **lru_l)
567 struct pack_window *w, *w_l;
569 for (w_l = NULL, w = p->windows; w; w = w->next) {
570 if (!w->inuse_cnt) {
571 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
572 *lru_p = p;
573 *lru_w = w;
574 *lru_l = w_l;
577 w_l = w;
581 static int unuse_one_window(struct packed_git *current, int keep_fd)
583 struct packed_git *p, *lru_p = NULL;
584 struct pack_window *lru_w = NULL, *lru_l = NULL;
586 if (current)
587 scan_windows(current, &lru_p, &lru_w, &lru_l);
588 for (p = packed_git; p; p = p->next)
589 scan_windows(p, &lru_p, &lru_w, &lru_l);
590 if (lru_p) {
591 munmap(lru_w->base, lru_w->len);
592 pack_mapped -= lru_w->len;
593 if (lru_l)
594 lru_l->next = lru_w->next;
595 else {
596 lru_p->windows = lru_w->next;
597 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
598 close(lru_p->pack_fd);
599 lru_p->pack_fd = -1;
602 free(lru_w);
603 pack_open_windows--;
604 return 1;
606 return 0;
609 void release_pack_memory(size_t need, int fd)
611 size_t cur = pack_mapped;
612 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
613 ; /* nothing */
616 void unuse_pack(struct pack_window **w_cursor)
618 struct pack_window *w = *w_cursor;
619 if (w) {
620 w->inuse_cnt--;
621 *w_cursor = NULL;
626 * Do not call this directly as this leaks p->pack_fd on error return;
627 * call open_packed_git() instead.
629 static int open_packed_git_1(struct packed_git *p)
631 struct stat st;
632 struct pack_header hdr;
633 unsigned char sha1[20];
634 unsigned char *idx_sha1;
635 long fd_flag;
637 if (!p->index_data && open_pack_index(p))
638 return error("packfile %s index unavailable", p->pack_name);
640 p->pack_fd = open(p->pack_name, O_RDONLY);
641 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
642 return -1;
644 /* If we created the struct before we had the pack we lack size. */
645 if (!p->pack_size) {
646 if (!S_ISREG(st.st_mode))
647 return error("packfile %s not a regular file", p->pack_name);
648 p->pack_size = st.st_size;
649 } else if (p->pack_size != st.st_size)
650 return error("packfile %s size changed", p->pack_name);
652 /* We leave these file descriptors open with sliding mmap;
653 * there is no point keeping them open across exec(), though.
655 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
656 if (fd_flag < 0)
657 return error("cannot determine file descriptor flags");
658 fd_flag |= FD_CLOEXEC;
659 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
660 return error("cannot set FD_CLOEXEC");
662 /* Verify we recognize this pack file format. */
663 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
664 return error("file %s is far too short to be a packfile", p->pack_name);
665 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
666 return error("file %s is not a GIT packfile", p->pack_name);
667 if (!pack_version_ok(hdr.hdr_version))
668 return error("packfile %s is version %u and not supported"
669 " (try upgrading GIT to a newer version)",
670 p->pack_name, ntohl(hdr.hdr_version));
672 /* Verify the pack matches its index. */
673 if (p->num_objects != ntohl(hdr.hdr_entries))
674 return error("packfile %s claims to have %u objects"
675 " while index indicates %u objects",
676 p->pack_name, ntohl(hdr.hdr_entries),
677 p->num_objects);
678 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
679 return error("end of packfile %s is unavailable", p->pack_name);
680 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
681 return error("packfile %s signature is unavailable", p->pack_name);
682 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
683 if (hashcmp(sha1, idx_sha1))
684 return error("packfile %s does not match index", p->pack_name);
685 return 0;
688 static int open_packed_git(struct packed_git *p)
690 if (!open_packed_git_1(p))
691 return 0;
692 if (p->pack_fd != -1) {
693 close(p->pack_fd);
694 p->pack_fd = -1;
696 return -1;
699 static void open_pack_sha1_dict(struct packed_git *p)
701 const size_t offset = sizeof(struct pack_header);
703 if (p->pack_fd < 0 && open_packed_git(p))
704 die("Could not open pack file");
706 p->sha1_dict_base = xmmap(NULL, offset + p->num_objects * 20,
707 PROT_READ, MAP_PRIVATE, p->pack_fd, 0);
708 p->sha1_dict = p->sha1_dict_base + offset;
710 close(p->pack_fd);
711 p->pack_fd = -1;
714 static void close_pack_sha1_dict(struct packed_git *p)
716 munmap(p->sha1_dict_base, sizeof(struct pack_header) +
717 p->num_objects * 20);
720 static int in_window(struct pack_window *win, off_t offset)
722 /* We must promise at least 20 bytes (one hash) after the
723 * offset is available from this window, otherwise the offset
724 * is not actually in this window and a different window (which
725 * has that one hash excess) must be used. This is to support
726 * the object header and delta base parsing routines below.
728 off_t win_off = win->offset;
729 return win_off <= offset
730 && (offset + 20) <= (win_off + win->len);
733 unsigned char* use_pack(struct packed_git *p,
734 struct pack_window **w_cursor,
735 off_t offset,
736 unsigned int *left)
738 struct pack_window *win = *w_cursor;
740 if (p->pack_fd == -1 && open_packed_git(p))
741 die("packfile %s cannot be accessed", p->pack_name);
743 /* Since packfiles end in a hash of their content and its
744 * pointless to ask for an offset into the middle of that
745 * hash, and the in_window function above wouldn't match
746 * don't allow an offset too close to the end of the file.
748 if (offset > (p->pack_size - 20))
749 die("offset beyond end of packfile (truncated pack?)");
751 if (!win || !in_window(win, offset)) {
752 if (win)
753 win->inuse_cnt--;
754 for (win = p->windows; win; win = win->next) {
755 if (in_window(win, offset))
756 break;
758 if (!win) {
759 size_t window_align = packed_git_window_size / 2;
760 off_t len;
761 win = xcalloc(1, sizeof(*win));
762 win->offset = (offset / window_align) * window_align;
763 len = p->pack_size - win->offset;
764 if (len > packed_git_window_size)
765 len = packed_git_window_size;
766 win->len = (size_t)len;
767 pack_mapped += win->len;
768 while (packed_git_limit < pack_mapped
769 && unuse_one_window(p, p->pack_fd))
770 ; /* nothing */
771 win->base = xmmap(NULL, win->len,
772 PROT_READ, MAP_PRIVATE,
773 p->pack_fd, win->offset);
774 if (win->base == MAP_FAILED)
775 die("packfile %s cannot be mapped: %s",
776 p->pack_name,
777 strerror(errno));
778 pack_mmap_calls++;
779 pack_open_windows++;
780 if (pack_mapped > peak_pack_mapped)
781 peak_pack_mapped = pack_mapped;
782 if (pack_open_windows > peak_pack_open_windows)
783 peak_pack_open_windows = pack_open_windows;
784 win->next = p->windows;
785 p->windows = win;
788 if (win != *w_cursor) {
789 win->last_used = pack_used_ctr++;
790 win->inuse_cnt++;
791 *w_cursor = win;
793 offset -= win->offset;
794 if (left)
795 *left = win->len - xsize_t(offset);
796 return win->base + offset;
799 struct packed_git *add_packed_git(const char *path, int path_len, int local)
801 struct stat st;
802 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
805 * Make sure a corresponding .pack file exists and that
806 * the index looks sane.
808 path_len -= strlen(".idx");
809 if (path_len < 1)
810 return NULL;
811 memcpy(p->pack_name, path, path_len);
812 strcpy(p->pack_name + path_len, ".pack");
813 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
814 free(p);
815 return NULL;
818 /* ok, it looks sane as far as we can check without
819 * actually mapping the pack file.
821 p->index_version = 0;
822 p->index_data = NULL;
823 p->index_size = 0;
824 p->num_objects = 0;
825 p->pack_size = st.st_size;
826 p->next = NULL;
827 p->windows = NULL;
828 p->pack_fd = -1;
829 p->pack_local = local;
830 p->mtime = st.st_mtime;
831 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
832 hashclr(p->sha1);
833 return p;
836 struct packed_git *parse_pack_index(unsigned char *sha1)
838 char *path = sha1_pack_index_name(sha1);
839 return parse_pack_index_file(sha1, path);
842 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
843 const char *idx_path)
845 const char *path = sha1_pack_name(sha1);
846 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
848 if (check_packed_git_idx(idx_path, p)) {
849 free(p);
850 return NULL;
853 strcpy(p->pack_name, path);
854 p->pack_size = 0;
855 p->next = NULL;
856 p->windows = NULL;
857 p->pack_fd = -1;
858 hashcpy(p->sha1, sha1);
859 return p;
862 void install_packed_git(struct packed_git *pack)
864 pack->next = packed_git;
865 packed_git = pack;
868 static void prepare_packed_git_one(char *objdir, int local)
870 /* Ensure that this buffer is large enough so that we can
871 append "/pack/" without clobbering the stack even if
872 strlen(objdir) were PATH_MAX. */
873 char path[PATH_MAX + 1 + 4 + 1 + 1];
874 int len;
875 DIR *dir;
876 struct dirent *de;
878 sprintf(path, "%s/pack", objdir);
879 len = strlen(path);
880 dir = opendir(path);
881 if (!dir) {
882 if (errno != ENOENT)
883 error("unable to open object pack directory: %s: %s",
884 path, strerror(errno));
885 return;
887 path[len++] = '/';
888 while ((de = readdir(dir)) != NULL) {
889 int namelen = strlen(de->d_name);
890 struct packed_git *p;
892 if (!has_extension(de->d_name, ".idx"))
893 continue;
895 if (len + namelen + 1 > sizeof(path))
896 continue;
898 /* Don't reopen a pack we already have. */
899 strcpy(path + len, de->d_name);
900 for (p = packed_git; p; p = p->next) {
901 if (!memcmp(path, p->pack_name, len + namelen - 4))
902 break;
904 if (p)
905 continue;
906 /* See if it really is a valid .idx file with corresponding
907 * .pack file that we can map.
909 p = add_packed_git(path, len + namelen, local);
910 if (!p)
911 continue;
912 install_packed_git(p);
914 closedir(dir);
917 static int sort_pack(const void *a_, const void *b_)
919 struct packed_git *a = *((struct packed_git **)a_);
920 struct packed_git *b = *((struct packed_git **)b_);
921 int st;
924 * Local packs tend to contain objects specific to our
925 * variant of the project than remote ones. In addition,
926 * remote ones could be on a network mounted filesystem.
927 * Favor local ones for these reasons.
929 st = a->pack_local - b->pack_local;
930 if (st)
931 return -st;
934 * Younger packs tend to contain more recent objects,
935 * and more recent objects tend to get accessed more
936 * often.
938 if (a->mtime < b->mtime)
939 return 1;
940 else if (a->mtime == b->mtime)
941 return 0;
942 return -1;
945 static void rearrange_packed_git(void)
947 struct packed_git **ary, *p;
948 int i, n;
950 for (n = 0, p = packed_git; p; p = p->next)
951 n++;
952 if (n < 2)
953 return;
955 /* prepare an array of packed_git for easier sorting */
956 ary = xcalloc(n, sizeof(struct packed_git *));
957 for (n = 0, p = packed_git; p; p = p->next)
958 ary[n++] = p;
960 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
962 /* link them back again */
963 for (i = 0; i < n - 1; i++)
964 ary[i]->next = ary[i + 1];
965 ary[n - 1]->next = NULL;
966 packed_git = ary[0];
968 free(ary);
971 static int prepare_packed_git_run_once = 0;
972 void prepare_packed_git(void)
974 struct alternate_object_database *alt;
976 if (prepare_packed_git_run_once)
977 return;
978 prepare_packed_git_one(get_object_directory(), 1);
979 prepare_alt_odb();
980 for (alt = alt_odb_list; alt; alt = alt->next) {
981 alt->name[-1] = 0;
982 prepare_packed_git_one(alt->base, 0);
983 alt->name[-1] = '/';
985 rearrange_packed_git();
986 prepare_packed_git_run_once = 1;
989 void reprepare_packed_git(void)
991 prepare_packed_git_run_once = 0;
992 prepare_packed_git();
995 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
997 unsigned char real_sha1[20];
998 hash_sha1_file(map, size, type, real_sha1);
999 return hashcmp(sha1, real_sha1) ? -1 : 0;
1002 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1004 struct stat st;
1005 void *map;
1006 int fd;
1007 char *filename = find_sha1_file(sha1, &st);
1009 if (!filename) {
1010 return NULL;
1013 fd = open(filename, O_RDONLY | sha1_file_open_flag);
1014 if (fd < 0) {
1015 /* See if it works without O_NOATIME */
1016 switch (sha1_file_open_flag) {
1017 default:
1018 fd = open(filename, O_RDONLY);
1019 if (fd >= 0)
1020 break;
1021 /* Fallthrough */
1022 case 0:
1023 return NULL;
1026 /* If it failed once, it will probably fail again.
1027 * Stop using O_NOATIME
1029 sha1_file_open_flag = 0;
1031 *size = xsize_t(st.st_size);
1032 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1033 close(fd);
1034 return map;
1037 static int legacy_loose_object(unsigned char *map)
1039 unsigned int word;
1042 * Is it a zlib-compressed buffer? If so, the first byte
1043 * must be 0x78 (15-bit window size, deflated), and the
1044 * first 16-bit word is evenly divisible by 31
1046 word = (map[0] << 8) + map[1];
1047 if (map[0] == 0x78 && !(word % 31))
1048 return 1;
1049 else
1050 return 0;
1053 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1055 unsigned shift;
1056 unsigned char c;
1057 unsigned long size;
1058 unsigned long used = 0;
1060 c = buf[used++];
1061 *type = (c >> 4) & 7;
1062 size = c & 15;
1063 shift = 4;
1064 while (c & 0x80) {
1065 if (len <= used)
1066 return 0;
1067 if (sizeof(long) * 8 <= shift)
1068 return 0;
1069 c = buf[used++];
1070 size += (c & 0x7f) << shift;
1071 shift += 7;
1073 *sizep = size;
1074 return used;
1077 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1079 unsigned long size, used;
1080 static const char valid_loose_object_type[8] = {
1081 0, /* OBJ_EXT */
1082 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1083 0, /* "delta" and others are invalid in a loose object */
1085 enum object_type type;
1087 /* Get the data stream */
1088 memset(stream, 0, sizeof(*stream));
1089 stream->next_in = map;
1090 stream->avail_in = mapsize;
1091 stream->next_out = buffer;
1092 stream->avail_out = bufsiz;
1094 if (legacy_loose_object(map)) {
1095 inflateInit(stream);
1096 return inflate(stream, 0);
1101 * There used to be a second loose object header format which
1102 * was meant to mimic the in-pack format, allowing for direct
1103 * copy of the object data. This format turned up not to be
1104 * really worth it and we don't write it any longer. But we
1105 * can still read it.
1107 used = unpack_object_header_gently(map, mapsize, &type, &size);
1108 if (!used || !valid_loose_object_type[type])
1109 return -1;
1110 map += used;
1111 mapsize -= used;
1113 /* Set up the stream for the rest.. */
1114 stream->next_in = map;
1115 stream->avail_in = mapsize;
1116 inflateInit(stream);
1118 /* And generate the fake traditional header */
1119 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1120 typename(type), size);
1121 return 0;
1124 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1126 int bytes = strlen(buffer) + 1;
1127 unsigned char *buf = xmalloc(1+size);
1128 unsigned long n;
1129 int status = Z_OK;
1131 n = stream->total_out - bytes;
1132 if (n > size)
1133 n = size;
1134 memcpy(buf, (char *) buffer + bytes, n);
1135 bytes = n;
1136 if (bytes <= size) {
1138 * The above condition must be (bytes <= size), not
1139 * (bytes < size). In other words, even though we
1140 * expect no more output and set avail_out to zer0,
1141 * the input zlib stream may have bytes that express
1142 * "this concludes the stream", and we *do* want to
1143 * eat that input.
1145 * Otherwise we would not be able to test that we
1146 * consumed all the input to reach the expected size;
1147 * we also want to check that zlib tells us that all
1148 * went well with status == Z_STREAM_END at the end.
1150 stream->next_out = buf + bytes;
1151 stream->avail_out = size - bytes;
1152 while (status == Z_OK)
1153 status = inflate(stream, Z_FINISH);
1155 buf[size] = 0;
1156 if (status == Z_STREAM_END && !stream->avail_in) {
1157 inflateEnd(stream);
1158 return buf;
1161 if (status < 0)
1162 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1163 else if (stream->avail_in)
1164 error("garbage at end of loose object '%s'",
1165 sha1_to_hex(sha1));
1166 free(buf);
1167 return NULL;
1171 * We used to just use "sscanf()", but that's actually way
1172 * too permissive for what we want to check. So do an anal
1173 * object header parse by hand.
1175 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1177 char type[10];
1178 int i;
1179 unsigned long size;
1182 * The type can be at most ten bytes (including the
1183 * terminating '\0' that we add), and is followed by
1184 * a space.
1186 i = 0;
1187 for (;;) {
1188 char c = *hdr++;
1189 if (c == ' ')
1190 break;
1191 type[i++] = c;
1192 if (i >= sizeof(type))
1193 return -1;
1195 type[i] = 0;
1198 * The length must follow immediately, and be in canonical
1199 * decimal format (ie "010" is not valid).
1201 size = *hdr++ - '0';
1202 if (size > 9)
1203 return -1;
1204 if (size) {
1205 for (;;) {
1206 unsigned long c = *hdr - '0';
1207 if (c > 9)
1208 break;
1209 hdr++;
1210 size = size * 10 + c;
1213 *sizep = size;
1216 * The length must be followed by a zero byte
1218 return *hdr ? -1 : type_from_string(type);
1221 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1223 int ret;
1224 z_stream stream;
1225 char hdr[8192];
1227 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1228 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1229 return NULL;
1231 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1234 unsigned long get_size_from_delta(struct packed_git *p,
1235 struct pack_window **w_curs,
1236 off_t curpos)
1238 const unsigned char *data;
1239 unsigned char delta_head[20], *in;
1240 z_stream stream;
1241 int st;
1243 memset(&stream, 0, sizeof(stream));
1244 stream.next_out = delta_head;
1245 stream.avail_out = sizeof(delta_head);
1247 inflateInit(&stream);
1248 do {
1249 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1250 stream.next_in = in;
1251 st = inflate(&stream, Z_FINISH);
1252 curpos += stream.next_in - in;
1253 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1254 stream.total_out < sizeof(delta_head));
1255 inflateEnd(&stream);
1256 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1257 die("delta data unpack-initial failed");
1259 /* Examine the initial part of the delta to figure out
1260 * the result size.
1262 data = delta_head;
1264 /* ignore base size */
1265 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1267 /* Read the result size */
1268 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1271 static off_t get_delta_base(struct packed_git *p,
1272 struct pack_window **w_curs,
1273 off_t *curpos,
1274 enum object_type type,
1275 off_t delta_obj_offset)
1277 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1278 off_t base_offset;
1280 /* use_pack() assured us we have [base_info, base_info + 20)
1281 * as a range that we can look at without walking off the
1282 * end of the mapped window. Its actually the hash size
1283 * that is assured. An OFS_DELTA longer than the hash size
1284 * is stupid, as then a REF_DELTA would be smaller to store.
1286 if (type == OBJ_OFS_DELTA) {
1287 unsigned used = 0;
1288 unsigned char c = base_info[used++];
1289 base_offset = c & 127;
1290 while (c & 128) {
1291 base_offset += 1;
1292 if (!base_offset || MSB(base_offset, 7))
1293 die("offset value overflow for delta base object");
1294 c = base_info[used++];
1295 base_offset = (base_offset << 7) + (c & 127);
1297 base_offset = delta_obj_offset - base_offset;
1298 if (base_offset >= delta_obj_offset)
1299 die("delta base offset out of bound");
1300 *curpos += used;
1301 } else if (type == OBJ_REF_DELTA) {
1302 /* The base entry _must_ be in the same pack */
1303 base_offset = find_pack_entry_one(base_info, p);
1304 if (!base_offset)
1305 die("failed to find delta-pack base object %s",
1306 sha1_to_hex(base_info));
1307 *curpos += 20;
1308 } else
1309 die("I am totally screwed");
1310 return base_offset;
1313 /* forward declaration for a mutually recursive function */
1314 static int packed_object_info(struct packed_git *p, off_t offset,
1315 unsigned long *sizep);
1317 static int packed_delta_info(struct packed_git *p,
1318 struct pack_window **w_curs,
1319 off_t curpos,
1320 enum object_type type,
1321 off_t obj_offset,
1322 unsigned long *sizep)
1324 off_t base_offset;
1326 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1327 type = packed_object_info(p, base_offset, NULL);
1329 /* We choose to only get the type of the base object and
1330 * ignore potentially corrupt pack file that expects the delta
1331 * based on a base with a wrong size. This saves tons of
1332 * inflate() calls.
1334 if (sizep)
1335 *sizep = get_size_from_delta(p, w_curs, curpos);
1337 return type;
1340 static int unpack_object_header(struct packed_git *p,
1341 struct pack_window **w_curs,
1342 off_t *curpos,
1343 unsigned long *sizep)
1345 unsigned char *base;
1346 unsigned int left;
1347 unsigned long used;
1348 enum object_type type;
1350 /* use_pack() assures us we have [base, base + 20) available
1351 * as a range that we can look at at. (Its actually the hash
1352 * size that is assured.) With our object header encoding
1353 * the maximum deflated object size is 2^137, which is just
1354 * insane, so we know won't exceed what we have been given.
1356 base = use_pack(p, w_curs, *curpos, &left);
1357 used = unpack_object_header_gently(base, left, &type, sizep);
1358 if (!used)
1359 die("object offset outside of pack file");
1360 *curpos += used;
1362 return type;
1365 const char *packed_object_info_detail(struct packed_git *p,
1366 off_t obj_offset,
1367 unsigned long *size,
1368 unsigned long *store_size,
1369 unsigned int *delta_chain_length,
1370 unsigned char *base_sha1)
1372 struct pack_window *w_curs = NULL;
1373 off_t curpos;
1374 unsigned long dummy;
1375 unsigned char *next_sha1;
1376 enum object_type type;
1378 *delta_chain_length = 0;
1379 curpos = obj_offset;
1380 type = unpack_object_header(p, &w_curs, &curpos, size);
1382 for (;;) {
1383 switch (type) {
1384 default:
1385 die("pack %s contains unknown object type %d",
1386 p->pack_name, type);
1387 case OBJ_COMMIT:
1388 case OBJ_TREE:
1389 case OBJ_BLOB:
1390 case OBJ_TAG:
1391 *store_size = 0; /* notyet */
1392 unuse_pack(&w_curs);
1393 return typename(type);
1394 case OBJ_OFS_DELTA:
1395 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1396 if (*delta_chain_length == 0) {
1397 /* TODO: find base_sha1 as pointed by curpos */
1398 hashclr(base_sha1);
1400 break;
1401 case OBJ_REF_DELTA:
1402 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1403 if (*delta_chain_length == 0)
1404 hashcpy(base_sha1, next_sha1);
1405 obj_offset = find_pack_entry_one(next_sha1, p);
1406 break;
1408 (*delta_chain_length)++;
1409 curpos = obj_offset;
1410 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1414 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1415 unsigned long *sizep)
1417 struct pack_window *w_curs = NULL;
1418 unsigned long size;
1419 off_t curpos = obj_offset;
1420 enum object_type type;
1422 type = unpack_object_header(p, &w_curs, &curpos, &size);
1424 switch (type) {
1425 case OBJ_OFS_DELTA:
1426 case OBJ_REF_DELTA:
1427 type = packed_delta_info(p, &w_curs, curpos,
1428 type, obj_offset, sizep);
1429 break;
1430 case OBJ_COMMIT:
1431 case OBJ_TREE:
1432 case OBJ_BLOB:
1433 case OBJ_TAG:
1434 if (sizep)
1435 *sizep = size;
1436 break;
1437 default:
1438 die("pack %s contains unknown object type %d",
1439 p->pack_name, type);
1441 unuse_pack(&w_curs);
1442 return type;
1445 static void *unpack_compressed_entry(struct packed_git *p,
1446 struct pack_window **w_curs,
1447 off_t curpos,
1448 unsigned long size)
1450 int st;
1451 z_stream stream;
1452 unsigned char *buffer, *in;
1454 buffer = xmalloc(size + 1);
1455 buffer[size] = 0;
1456 memset(&stream, 0, sizeof(stream));
1457 stream.next_out = buffer;
1458 stream.avail_out = size;
1460 inflateInit(&stream);
1461 do {
1462 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1463 stream.next_in = in;
1464 st = inflate(&stream, Z_FINISH);
1465 curpos += stream.next_in - in;
1466 } while (st == Z_OK || st == Z_BUF_ERROR);
1467 inflateEnd(&stream);
1468 if ((st != Z_STREAM_END) || stream.total_out != size) {
1469 free(buffer);
1470 return NULL;
1473 return buffer;
1476 #define MAX_DELTA_CACHE (256)
1478 static size_t delta_base_cached;
1480 static struct delta_base_cache_lru_list {
1481 struct delta_base_cache_lru_list *prev;
1482 struct delta_base_cache_lru_list *next;
1483 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1485 static struct delta_base_cache_entry {
1486 struct delta_base_cache_lru_list lru;
1487 void *data;
1488 struct packed_git *p;
1489 off_t base_offset;
1490 unsigned long size;
1491 enum object_type type;
1492 } delta_base_cache[MAX_DELTA_CACHE];
1494 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1496 unsigned long hash;
1498 hash = (unsigned long)p + (unsigned long)base_offset;
1499 hash += (hash >> 8) + (hash >> 16);
1500 return hash % MAX_DELTA_CACHE;
1503 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1504 unsigned long *base_size, enum object_type *type, int keep_cache)
1506 void *ret;
1507 unsigned long hash = pack_entry_hash(p, base_offset);
1508 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1510 ret = ent->data;
1511 if (ret && ent->p == p && ent->base_offset == base_offset)
1512 goto found_cache_entry;
1513 return unpack_entry(p, base_offset, type, base_size);
1515 found_cache_entry:
1516 if (!keep_cache) {
1517 ent->data = NULL;
1518 ent->lru.next->prev = ent->lru.prev;
1519 ent->lru.prev->next = ent->lru.next;
1520 delta_base_cached -= ent->size;
1522 else {
1523 ret = xmalloc(ent->size + 1);
1524 memcpy(ret, ent->data, ent->size);
1525 ((char *)ret)[ent->size] = 0;
1527 *type = ent->type;
1528 *base_size = ent->size;
1529 return ret;
1532 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1534 if (ent->data) {
1535 free(ent->data);
1536 ent->data = NULL;
1537 ent->lru.next->prev = ent->lru.prev;
1538 ent->lru.prev->next = ent->lru.next;
1539 delta_base_cached -= ent->size;
1543 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1544 void *base, unsigned long base_size, enum object_type type)
1546 unsigned long hash = pack_entry_hash(p, base_offset);
1547 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1548 struct delta_base_cache_lru_list *lru;
1550 release_delta_base_cache(ent);
1551 delta_base_cached += base_size;
1553 for (lru = delta_base_cache_lru.next;
1554 delta_base_cached > delta_base_cache_limit
1555 && lru != &delta_base_cache_lru;
1556 lru = lru->next) {
1557 struct delta_base_cache_entry *f = (void *)lru;
1558 if (f->type == OBJ_BLOB)
1559 release_delta_base_cache(f);
1561 for (lru = delta_base_cache_lru.next;
1562 delta_base_cached > delta_base_cache_limit
1563 && lru != &delta_base_cache_lru;
1564 lru = lru->next) {
1565 struct delta_base_cache_entry *f = (void *)lru;
1566 release_delta_base_cache(f);
1569 ent->p = p;
1570 ent->base_offset = base_offset;
1571 ent->type = type;
1572 ent->data = base;
1573 ent->size = base_size;
1574 ent->lru.next = &delta_base_cache_lru;
1575 ent->lru.prev = delta_base_cache_lru.prev;
1576 delta_base_cache_lru.prev->next = &ent->lru;
1577 delta_base_cache_lru.prev = &ent->lru;
1580 static void *unpack_delta_entry(struct packed_git *p,
1581 struct pack_window **w_curs,
1582 off_t curpos,
1583 unsigned long delta_size,
1584 off_t obj_offset,
1585 enum object_type *type,
1586 unsigned long *sizep)
1588 void *delta_data, *result, *base;
1589 unsigned long base_size;
1590 off_t base_offset;
1592 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1593 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1594 if (!base)
1595 die("failed to read delta base object"
1596 " at %"PRIuMAX" from %s",
1597 (uintmax_t)base_offset, p->pack_name);
1599 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1600 if (!delta_data)
1601 die("failed to unpack compressed delta"
1602 " at %"PRIuMAX" from %s",
1603 (uintmax_t)curpos, p->pack_name);
1604 result = patch_delta(base, base_size,
1605 delta_data, delta_size,
1606 sizep);
1607 if (!result)
1608 die("failed to apply delta");
1609 free(delta_data);
1610 add_delta_base_cache(p, base_offset, base, base_size, *type);
1611 return result;
1614 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1615 enum object_type *type, unsigned long *sizep)
1617 struct pack_window *w_curs = NULL;
1618 off_t curpos = obj_offset;
1619 void *data;
1621 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1622 switch (*type) {
1623 case OBJ_OFS_DELTA:
1624 case OBJ_REF_DELTA:
1625 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1626 obj_offset, type, sizep);
1627 break;
1628 case OBJ_COMMIT:
1629 case OBJ_TREE:
1630 case OBJ_BLOB:
1631 case OBJ_TAG:
1632 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1633 break;
1634 default:
1635 die("unknown object type %i in %s", *type, p->pack_name);
1637 unuse_pack(&w_curs);
1638 return data;
1641 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1642 uint32_t n)
1644 const unsigned char *index = p->index_data;
1645 if (!index) {
1646 if (open_pack_index(p))
1647 return NULL;
1648 index = p->index_data;
1650 if (n >= p->num_objects)
1651 return NULL;
1652 index += 4 * 256;
1653 if (p->index_version == 1) {
1654 return index + 24 * n + 4;
1655 } else if (p->index_version == 2) {
1656 index += 8;
1657 return index + 20 * n;
1658 } else {
1659 open_pack_sha1_dict(p);
1660 return p->sha1_dict + 20 * n;
1664 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1666 const unsigned char *index = p->index_data;
1667 index += 4 * 256;
1668 if (p->index_version == 1) {
1669 return ntohl(*((uint32_t *)(index + 24 * n)));
1670 } else { /* index_version > 1 */
1671 uint32_t off;
1672 if (p->index_version == 2)
1673 index += p->num_objects * (20 + 4);
1674 else
1675 index += p->num_objects * 4;
1676 index += 8;
1677 off = ntohl(*((uint32_t *)(index + 4 * n)));
1678 if (!(off & 0x80000000))
1679 return off;
1680 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1681 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1682 ntohl(*((uint32_t *)(index + 4)));
1686 off_t find_pack_entry_one(const unsigned char *sha1,
1687 struct packed_git *p)
1689 const uint32_t *level1_ofs = p->index_data;
1690 const unsigned char *index = p->index_data;
1691 unsigned hi, lo;
1693 if (!index) {
1694 if (open_pack_index(p))
1695 return 0;
1696 level1_ofs = p->index_data;
1697 index = p->index_data;
1699 if (p->index_version > 1) {
1700 level1_ofs += 2;
1701 index += 8;
1704 if (p->index_version > 2) {
1705 open_pack_sha1_dict(p);
1706 index = p->sha1_dict;
1707 } else
1708 index += 4 * 256;
1710 hi = ntohl(level1_ofs[*sha1]);
1711 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1713 do {
1714 unsigned mi = (lo + hi) / 2;
1715 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1716 int cmp = hashcmp(index + x, sha1);
1717 if (!cmp) {
1718 if (p->index_version > 2)
1719 close_pack_sha1_dict(p);
1720 return nth_packed_object_offset(p, mi);
1722 if (cmp > 0)
1723 hi = mi;
1724 else
1725 lo = mi+1;
1726 } while (lo < hi);
1728 if (p->index_version > 2)
1729 close_pack_sha1_dict(p);
1730 return 0;
1733 static int matches_pack_name(struct packed_git *p, const char *ig)
1735 const char *last_c, *c;
1737 if (!strcmp(p->pack_name, ig))
1738 return 0;
1740 for (c = p->pack_name, last_c = c; *c;)
1741 if (*c == '/')
1742 last_c = ++c;
1743 else
1744 ++c;
1745 if (!strcmp(last_c, ig))
1746 return 0;
1748 return 1;
1751 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1753 static struct packed_git *last_found = (void *)1;
1754 struct packed_git *p;
1755 off_t offset;
1757 prepare_packed_git();
1758 if (!packed_git)
1759 return 0;
1760 p = (last_found == (void *)1) ? packed_git : last_found;
1762 do {
1763 if (ignore_packed) {
1764 const char **ig;
1765 for (ig = ignore_packed; *ig; ig++)
1766 if (!matches_pack_name(p, *ig))
1767 break;
1768 if (*ig)
1769 goto next;
1772 offset = find_pack_entry_one(sha1, p);
1773 if (offset) {
1775 * We are about to tell the caller where they can
1776 * locate the requested object. We better make
1777 * sure the packfile is still here and can be
1778 * accessed before supplying that answer, as
1779 * it may have been deleted since the index
1780 * was loaded!
1782 if (p->pack_fd == -1 && open_packed_git(p)) {
1783 error("packfile %s cannot be accessed", p->pack_name);
1784 goto next;
1786 e->offset = offset;
1787 e->p = p;
1788 hashcpy(e->sha1, sha1);
1789 last_found = p;
1790 return 1;
1793 next:
1794 if (p == last_found)
1795 p = packed_git;
1796 else
1797 p = p->next;
1798 if (p == last_found)
1799 p = p->next;
1800 } while (p);
1801 return 0;
1804 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1805 struct packed_git *packs)
1807 struct packed_git *p;
1809 for (p = packs; p; p = p->next) {
1810 if (find_pack_entry_one(sha1, p))
1811 return p;
1813 return NULL;
1817 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1819 int status;
1820 unsigned long mapsize, size;
1821 void *map;
1822 z_stream stream;
1823 char hdr[32];
1825 map = map_sha1_file(sha1, &mapsize);
1826 if (!map)
1827 return error("unable to find %s", sha1_to_hex(sha1));
1828 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1829 status = error("unable to unpack %s header",
1830 sha1_to_hex(sha1));
1831 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1832 status = error("unable to parse %s header", sha1_to_hex(sha1));
1833 else if (sizep)
1834 *sizep = size;
1835 inflateEnd(&stream);
1836 munmap(map, mapsize);
1837 return status;
1840 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1842 struct pack_entry e;
1844 if (!find_pack_entry(sha1, &e, NULL)) {
1845 reprepare_packed_git();
1846 if (!find_pack_entry(sha1, &e, NULL))
1847 return sha1_loose_object_info(sha1, sizep);
1849 return packed_object_info(e.p, e.offset, sizep);
1852 static void *read_packed_sha1(const unsigned char *sha1,
1853 enum object_type *type, unsigned long *size)
1855 struct pack_entry e;
1857 if (!find_pack_entry(sha1, &e, NULL))
1858 return NULL;
1859 else
1860 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1864 * This is meant to hold a *small* number of objects that you would
1865 * want read_sha1_file() to be able to return, but yet you do not want
1866 * to write them into the object store (e.g. a browse-only
1867 * application).
1869 static struct cached_object {
1870 unsigned char sha1[20];
1871 enum object_type type;
1872 void *buf;
1873 unsigned long size;
1874 } *cached_objects;
1875 static int cached_object_nr, cached_object_alloc;
1877 static struct cached_object *find_cached_object(const unsigned char *sha1)
1879 int i;
1880 struct cached_object *co = cached_objects;
1882 for (i = 0; i < cached_object_nr; i++, co++) {
1883 if (!hashcmp(co->sha1, sha1))
1884 return co;
1886 return NULL;
1889 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1890 unsigned char *sha1)
1892 struct cached_object *co;
1894 hash_sha1_file(buf, len, typename(type), sha1);
1895 if (has_sha1_file(sha1) || find_cached_object(sha1))
1896 return 0;
1897 if (cached_object_alloc <= cached_object_nr) {
1898 cached_object_alloc = alloc_nr(cached_object_alloc);
1899 cached_objects = xrealloc(cached_objects,
1900 sizeof(*cached_objects) *
1901 cached_object_alloc);
1903 co = &cached_objects[cached_object_nr++];
1904 co->size = len;
1905 co->type = type;
1906 co->buf = xmalloc(len);
1907 memcpy(co->buf, buf, len);
1908 hashcpy(co->sha1, sha1);
1909 return 0;
1912 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1913 unsigned long *size)
1915 unsigned long mapsize;
1916 void *map, *buf;
1917 struct cached_object *co;
1919 co = find_cached_object(sha1);
1920 if (co) {
1921 buf = xmalloc(co->size + 1);
1922 memcpy(buf, co->buf, co->size);
1923 ((char*)buf)[co->size] = 0;
1924 *type = co->type;
1925 *size = co->size;
1926 return buf;
1929 buf = read_packed_sha1(sha1, type, size);
1930 if (buf)
1931 return buf;
1932 map = map_sha1_file(sha1, &mapsize);
1933 if (map) {
1934 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1935 munmap(map, mapsize);
1936 return buf;
1938 reprepare_packed_git();
1939 return read_packed_sha1(sha1, type, size);
1942 void *read_object_with_reference(const unsigned char *sha1,
1943 const char *required_type_name,
1944 unsigned long *size,
1945 unsigned char *actual_sha1_return)
1947 enum object_type type, required_type;
1948 void *buffer;
1949 unsigned long isize;
1950 unsigned char actual_sha1[20];
1952 required_type = type_from_string(required_type_name);
1953 hashcpy(actual_sha1, sha1);
1954 while (1) {
1955 int ref_length = -1;
1956 const char *ref_type = NULL;
1958 buffer = read_sha1_file(actual_sha1, &type, &isize);
1959 if (!buffer)
1960 return NULL;
1961 if (type == required_type) {
1962 *size = isize;
1963 if (actual_sha1_return)
1964 hashcpy(actual_sha1_return, actual_sha1);
1965 return buffer;
1967 /* Handle references */
1968 else if (type == OBJ_COMMIT)
1969 ref_type = "tree ";
1970 else if (type == OBJ_TAG)
1971 ref_type = "object ";
1972 else {
1973 free(buffer);
1974 return NULL;
1976 ref_length = strlen(ref_type);
1978 if (memcmp(buffer, ref_type, ref_length) ||
1979 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1980 free(buffer);
1981 return NULL;
1983 free(buffer);
1984 /* Now we have the ID of the referred-to object in
1985 * actual_sha1. Check again. */
1989 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1990 const char *type, unsigned char *sha1,
1991 char *hdr, int *hdrlen)
1993 SHA_CTX c;
1995 /* Generate the header */
1996 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1998 /* Sha1.. */
1999 SHA1_Init(&c);
2000 SHA1_Update(&c, hdr, *hdrlen);
2001 SHA1_Update(&c, buf, len);
2002 SHA1_Final(sha1, &c);
2006 * Link the tempfile to the final place, possibly creating the
2007 * last directory level as you do so.
2009 * Returns the errno on failure, 0 on success.
2011 static int link_temp_to_file(const char *tmpfile, const char *filename)
2013 int ret;
2014 char *dir;
2016 if (!link(tmpfile, filename))
2017 return 0;
2020 * Try to mkdir the last path component if that failed.
2022 * Re-try the "link()" regardless of whether the mkdir
2023 * succeeds, since a race might mean that somebody
2024 * else succeeded.
2026 ret = errno;
2027 dir = strrchr(filename, '/');
2028 if (dir) {
2029 *dir = 0;
2030 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
2031 *dir = '/';
2032 return -2;
2034 *dir = '/';
2035 if (!link(tmpfile, filename))
2036 return 0;
2037 ret = errno;
2039 return ret;
2043 * Move the just written object into its final resting place
2045 int move_temp_to_file(const char *tmpfile, const char *filename)
2047 int ret = link_temp_to_file(tmpfile, filename);
2050 * Coda hack - coda doesn't like cross-directory links,
2051 * so we fall back to a rename, which will mean that it
2052 * won't be able to check collisions, but that's not a
2053 * big deal.
2055 * The same holds for FAT formatted media.
2057 * When this succeeds, we just return 0. We have nothing
2058 * left to unlink.
2060 if (ret && ret != EEXIST) {
2061 if (!rename(tmpfile, filename))
2062 return 0;
2063 ret = errno;
2065 unlink(tmpfile);
2066 if (ret) {
2067 if (ret != EEXIST) {
2068 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2070 /* FIXME!!! Collision check here ? */
2073 return 0;
2076 static int write_buffer(int fd, const void *buf, size_t len)
2078 if (write_in_full(fd, buf, len) < 0)
2079 return error("file write error (%s)", strerror(errno));
2080 return 0;
2083 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2084 unsigned char *sha1)
2086 char hdr[32];
2087 int hdrlen;
2088 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2089 return 0;
2092 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2094 int size, ret;
2095 unsigned char *compressed;
2096 z_stream stream;
2097 unsigned char sha1[20];
2098 char *filename;
2099 static char tmpfile[PATH_MAX];
2100 char hdr[32];
2101 int fd, hdrlen;
2103 /* Normally if we have it in the pack then we do not bother writing
2104 * it out into .git/objects/??/?{38} file.
2106 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2107 filename = sha1_file_name(sha1);
2108 if (returnsha1)
2109 hashcpy(returnsha1, sha1);
2110 if (has_sha1_file(sha1))
2111 return 0;
2112 fd = open(filename, O_RDONLY);
2113 if (fd >= 0) {
2115 * FIXME!!! We might do collision checking here, but we'd
2116 * need to uncompress the old file and check it. Later.
2118 close(fd);
2119 return 0;
2122 if (errno != ENOENT) {
2123 return error("sha1 file %s: %s\n", filename, strerror(errno));
2126 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2128 fd = mkstemp(tmpfile);
2129 if (fd < 0) {
2130 if (errno == EPERM)
2131 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2132 else
2133 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2136 /* Set it up */
2137 memset(&stream, 0, sizeof(stream));
2138 deflateInit(&stream, zlib_compression_level);
2139 size = 8 + deflateBound(&stream, len+hdrlen);
2140 compressed = xmalloc(size);
2142 /* Compress it */
2143 stream.next_out = compressed;
2144 stream.avail_out = size;
2146 /* First header.. */
2147 stream.next_in = (unsigned char *)hdr;
2148 stream.avail_in = hdrlen;
2149 while (deflate(&stream, 0) == Z_OK)
2150 /* nothing */;
2152 /* Then the data itself.. */
2153 stream.next_in = buf;
2154 stream.avail_in = len;
2155 ret = deflate(&stream, Z_FINISH);
2156 if (ret != Z_STREAM_END)
2157 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2159 ret = deflateEnd(&stream);
2160 if (ret != Z_OK)
2161 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2163 size = stream.total_out;
2165 if (write_buffer(fd, compressed, size) < 0)
2166 die("unable to write sha1 file");
2167 fchmod(fd, 0444);
2168 if (close(fd))
2169 die("unable to write sha1 file");
2170 free(compressed);
2172 return move_temp_to_file(tmpfile, filename);
2176 * We need to unpack and recompress the object for writing
2177 * it out to a different file.
2179 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2181 size_t size;
2182 z_stream stream;
2183 unsigned char *unpacked;
2184 unsigned long len;
2185 enum object_type type;
2186 char hdr[32];
2187 int hdrlen;
2188 void *buf;
2190 /* need to unpack and recompress it by itself */
2191 unpacked = read_packed_sha1(sha1, &type, &len);
2192 if (!unpacked)
2193 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2195 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2197 /* Set it up */
2198 memset(&stream, 0, sizeof(stream));
2199 deflateInit(&stream, zlib_compression_level);
2200 size = deflateBound(&stream, len + hdrlen);
2201 buf = xmalloc(size);
2203 /* Compress it */
2204 stream.next_out = buf;
2205 stream.avail_out = size;
2207 /* First header.. */
2208 stream.next_in = (void *)hdr;
2209 stream.avail_in = hdrlen;
2210 while (deflate(&stream, 0) == Z_OK)
2211 /* nothing */;
2213 /* Then the data itself.. */
2214 stream.next_in = unpacked;
2215 stream.avail_in = len;
2216 while (deflate(&stream, Z_FINISH) == Z_OK)
2217 /* nothing */;
2218 deflateEnd(&stream);
2219 free(unpacked);
2221 *objsize = stream.total_out;
2222 return buf;
2225 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2227 int retval;
2228 unsigned long objsize;
2229 void *buf = map_sha1_file(sha1, &objsize);
2231 if (buf) {
2232 retval = write_buffer(fd, buf, objsize);
2233 munmap(buf, objsize);
2234 return retval;
2237 buf = repack_object(sha1, &objsize);
2238 retval = write_buffer(fd, buf, objsize);
2239 free(buf);
2240 return retval;
2243 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2244 size_t bufsize, size_t *bufposn)
2246 char tmpfile[PATH_MAX];
2247 int local;
2248 z_stream stream;
2249 unsigned char real_sha1[20];
2250 unsigned char discard[4096];
2251 int ret;
2252 SHA_CTX c;
2254 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2256 local = mkstemp(tmpfile);
2257 if (local < 0) {
2258 if (errno == EPERM)
2259 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2260 else
2261 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2264 memset(&stream, 0, sizeof(stream));
2266 inflateInit(&stream);
2268 SHA1_Init(&c);
2270 do {
2271 ssize_t size;
2272 if (*bufposn) {
2273 stream.avail_in = *bufposn;
2274 stream.next_in = (unsigned char *) buffer;
2275 do {
2276 stream.next_out = discard;
2277 stream.avail_out = sizeof(discard);
2278 ret = inflate(&stream, Z_SYNC_FLUSH);
2279 SHA1_Update(&c, discard, sizeof(discard) -
2280 stream.avail_out);
2281 } while (stream.avail_in && ret == Z_OK);
2282 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2283 die("unable to write sha1 file");
2284 memmove(buffer, buffer + *bufposn - stream.avail_in,
2285 stream.avail_in);
2286 *bufposn = stream.avail_in;
2287 if (ret != Z_OK)
2288 break;
2290 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2291 if (size <= 0) {
2292 close(local);
2293 unlink(tmpfile);
2294 if (!size)
2295 return error("Connection closed?");
2296 perror("Reading from connection");
2297 return -1;
2299 *bufposn += size;
2300 } while (1);
2301 inflateEnd(&stream);
2303 fchmod(local, 0444);
2304 if (close(local) != 0)
2305 die("unable to write sha1 file");
2306 SHA1_Final(real_sha1, &c);
2307 if (ret != Z_STREAM_END) {
2308 unlink(tmpfile);
2309 return error("File %s corrupted", sha1_to_hex(sha1));
2311 if (hashcmp(sha1, real_sha1)) {
2312 unlink(tmpfile);
2313 return error("File %s has bad hash", sha1_to_hex(sha1));
2316 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2319 int has_pack_index(const unsigned char *sha1)
2321 struct stat st;
2322 if (stat(sha1_pack_index_name(sha1), &st))
2323 return 0;
2324 return 1;
2327 int has_pack_file(const unsigned char *sha1)
2329 struct stat st;
2330 if (stat(sha1_pack_name(sha1), &st))
2331 return 0;
2332 return 1;
2335 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2337 struct pack_entry e;
2338 return find_pack_entry(sha1, &e, ignore_packed);
2341 int has_sha1_file(const unsigned char *sha1)
2343 struct stat st;
2344 struct pack_entry e;
2346 if (find_pack_entry(sha1, &e, NULL))
2347 return 1;
2348 return find_sha1_file(sha1, &st) ? 1 : 0;
2352 * reads from fd as long as possible into a supplied buffer of size bytes.
2353 * If necessary the buffer's size is increased using realloc()
2355 * returns 0 if anything went fine and -1 otherwise
2357 * The buffer is always NUL-terminated, not including it in returned size.
2359 * NOTE: both buf and size may change, but even when -1 is returned
2360 * you still have to free() it yourself.
2362 int read_fd(int fd, char **return_buf, unsigned long *return_size)
2364 char *buf = *return_buf;
2365 unsigned long size = *return_size;
2366 ssize_t iret;
2367 unsigned long off = 0;
2369 if (!buf || size <= 1) {
2370 size = 1024;
2371 buf = xrealloc(buf, size);
2374 do {
2375 iret = xread(fd, buf + off, (size - 1) - off);
2376 if (iret > 0) {
2377 off += iret;
2378 if (off == size - 1) {
2379 size = alloc_nr(size);
2380 buf = xrealloc(buf, size);
2383 } while (iret > 0);
2385 buf[off] = '\0';
2387 *return_buf = buf;
2388 *return_size = off;
2390 if (iret < 0)
2391 return -1;
2392 return 0;
2395 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2397 unsigned long size = 4096;
2398 char *buf = xmalloc(size);
2399 int ret;
2401 if (read_fd(fd, &buf, &size)) {
2402 free(buf);
2403 return -1;
2406 if (!type)
2407 type = blob_type;
2408 if (write_object)
2409 ret = write_sha1_file(buf, size, type, sha1);
2410 else
2411 ret = hash_sha1_file(buf, size, type, sha1);
2412 free(buf);
2413 return ret;
2416 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2417 enum object_type type, const char *path)
2419 size_t size = xsize_t(st->st_size);
2420 void *buf = NULL;
2421 int ret, re_allocated = 0;
2423 if (size)
2424 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2425 close(fd);
2427 if (!type)
2428 type = OBJ_BLOB;
2431 * Convert blobs to git internal format
2433 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2434 unsigned long nsize = size;
2435 char *nbuf = convert_to_git(path, buf, &nsize);
2436 if (nbuf) {
2437 munmap(buf, size);
2438 size = nsize;
2439 buf = nbuf;
2440 re_allocated = 1;
2444 if (write_object)
2445 ret = write_sha1_file(buf, size, typename(type), sha1);
2446 else
2447 ret = hash_sha1_file(buf, size, typename(type), sha1);
2448 if (re_allocated) {
2449 free(buf);
2450 return ret;
2452 if (size)
2453 munmap(buf, size);
2454 return ret;
2457 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2459 int fd;
2460 char *target;
2461 size_t len;
2463 switch (st->st_mode & S_IFMT) {
2464 case S_IFREG:
2465 fd = open(path, O_RDONLY);
2466 if (fd < 0)
2467 return error("open(\"%s\"): %s", path,
2468 strerror(errno));
2469 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2470 return error("%s: failed to insert into database",
2471 path);
2472 break;
2473 case S_IFLNK:
2474 len = xsize_t(st->st_size);
2475 target = xmalloc(len + 1);
2476 if (readlink(path, target, len + 1) != st->st_size) {
2477 char *errstr = strerror(errno);
2478 free(target);
2479 return error("readlink(\"%s\"): %s", path,
2480 errstr);
2482 if (!write_object)
2483 hash_sha1_file(target, len, blob_type, sha1);
2484 else if (write_sha1_file(target, len, blob_type, sha1))
2485 return error("%s: failed to insert into database",
2486 path);
2487 free(target);
2488 break;
2489 case S_IFDIR:
2490 return resolve_gitlink_ref(path, "HEAD", sha1);
2491 default:
2492 return error("%s: unsupported file type", path);
2494 return 0;
2497 int read_pack_header(int fd, struct pack_header *header)
2499 char *c = (char*)header;
2500 ssize_t remaining = sizeof(struct pack_header);
2501 do {
2502 ssize_t r = xread(fd, c, remaining);
2503 if (r <= 0)
2504 /* "eof before pack header was fully read" */
2505 return PH_ERROR_EOF;
2506 remaining -= r;
2507 c += r;
2508 } while (remaining > 0);
2509 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2510 /* "protocol error (pack signature mismatch detected)" */
2511 return PH_ERROR_PACK_SIGNATURE;
2512 if (!pack_version_ok(header->hdr_version))
2513 /* "protocol error (pack version unsupported)" */
2514 return PH_ERROR_PROTOCOL;
2515 return 0;