Merge branch 'ar/diff'
[git/repo.git] / sha1_file.c
blobee64865b607c70ff42f851b91c19316ef90ce7b9
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"
17 #ifndef O_NOATIME
18 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
19 #define O_NOATIME 01000000
20 #else
21 #define O_NOATIME 0
22 #endif
23 #endif
25 #ifdef NO_C99_FORMAT
26 #define SZ_FMT "lu"
27 #else
28 #define SZ_FMT "zu"
29 #endif
31 const unsigned char null_sha1[20];
33 static unsigned int sha1_file_open_flag = O_NOATIME;
35 signed char hexval_table[256] = {
36 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
42 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
43 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
44 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
48 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
70 int get_sha1_hex(const char *hex, unsigned char *sha1)
72 int i;
73 for (i = 0; i < 20; i++) {
74 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
75 if (val & ~0xff)
76 return -1;
77 *sha1++ = val;
78 hex += 2;
80 return 0;
83 int safe_create_leading_directories(char *path)
85 char *pos = path;
86 struct stat st;
88 if (*pos == '/')
89 pos++;
91 while (pos) {
92 pos = strchr(pos, '/');
93 if (!pos)
94 break;
95 *pos = 0;
96 if (!stat(path, &st)) {
97 /* path exists */
98 if (!S_ISDIR(st.st_mode)) {
99 *pos = '/';
100 return -3;
103 else if (mkdir(path, 0777)) {
104 *pos = '/';
105 return -1;
107 else if (adjust_shared_perm(path)) {
108 *pos = '/';
109 return -2;
111 *pos++ = '/';
113 return 0;
116 char * sha1_to_hex(const unsigned char *sha1)
118 static int bufno;
119 static char hexbuffer[4][50];
120 static const char hex[] = "0123456789abcdef";
121 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
122 int i;
124 for (i = 0; i < 20; i++) {
125 unsigned int val = *sha1++;
126 *buf++ = hex[val >> 4];
127 *buf++ = hex[val & 0xf];
129 *buf = '\0';
131 return buffer;
134 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
136 int i;
137 for (i = 0; i < 20; i++) {
138 static char hex[] = "0123456789abcdef";
139 unsigned int val = sha1[i];
140 char *pos = pathbuf + i*2 + (i > 0);
141 *pos++ = hex[val >> 4];
142 *pos = hex[val & 0xf];
147 * NOTE! This returns a statically allocated buffer, so you have to be
148 * careful about using it. Do a "xstrdup()" if you need to save the
149 * filename.
151 * Also note that this returns the location for creating. Reading
152 * SHA1 file can happen from any alternate directory listed in the
153 * DB_ENVIRONMENT environment variable if it is not found in
154 * the primary object database.
156 char *sha1_file_name(const unsigned char *sha1)
158 static char *name, *base;
160 if (!base) {
161 const char *sha1_file_directory = get_object_directory();
162 int len = strlen(sha1_file_directory);
163 base = xmalloc(len + 60);
164 memcpy(base, sha1_file_directory, len);
165 memset(base+len, 0, 60);
166 base[len] = '/';
167 base[len+3] = '/';
168 name = base + len + 1;
170 fill_sha1_path(name, sha1);
171 return base;
174 char *sha1_pack_name(const unsigned char *sha1)
176 static const char hex[] = "0123456789abcdef";
177 static char *name, *base, *buf;
178 int i;
180 if (!base) {
181 const char *sha1_file_directory = get_object_directory();
182 int len = strlen(sha1_file_directory);
183 base = xmalloc(len + 60);
184 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
185 name = base + len + 11;
188 buf = name;
190 for (i = 0; i < 20; i++) {
191 unsigned int val = *sha1++;
192 *buf++ = hex[val >> 4];
193 *buf++ = hex[val & 0xf];
196 return base;
199 char *sha1_pack_index_name(const unsigned char *sha1)
201 static const char hex[] = "0123456789abcdef";
202 static char *name, *base, *buf;
203 int i;
205 if (!base) {
206 const char *sha1_file_directory = get_object_directory();
207 int len = strlen(sha1_file_directory);
208 base = xmalloc(len + 60);
209 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
210 name = base + len + 11;
213 buf = name;
215 for (i = 0; i < 20; i++) {
216 unsigned int val = *sha1++;
217 *buf++ = hex[val >> 4];
218 *buf++ = hex[val & 0xf];
221 return base;
224 struct alternate_object_database *alt_odb_list;
225 static struct alternate_object_database **alt_odb_tail;
227 static void read_info_alternates(const char * alternates, int depth);
230 * Prepare alternate object database registry.
232 * The variable alt_odb_list points at the list of struct
233 * alternate_object_database. The elements on this list come from
234 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
235 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
236 * whose contents is similar to that environment variable but can be
237 * LF separated. Its base points at a statically allocated buffer that
238 * contains "/the/directory/corresponding/to/.git/objects/...", while
239 * its name points just after the slash at the end of ".git/objects/"
240 * in the example above, and has enough space to hold 40-byte hex
241 * SHA1, an extra slash for the first level indirection, and the
242 * terminating NUL.
244 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
246 struct stat st;
247 const char *objdir = get_object_directory();
248 struct alternate_object_database *ent;
249 struct alternate_object_database *alt;
250 /* 43 = 40-byte + 2 '/' + terminating NUL */
251 int pfxlen = len;
252 int entlen = pfxlen + 43;
253 int base_len = -1;
255 if (*entry != '/' && relative_base) {
256 /* Relative alt-odb */
257 if (base_len < 0)
258 base_len = strlen(relative_base) + 1;
259 entlen += base_len;
260 pfxlen += base_len;
262 ent = xmalloc(sizeof(*ent) + entlen);
264 if (*entry != '/' && relative_base) {
265 memcpy(ent->base, relative_base, base_len - 1);
266 ent->base[base_len - 1] = '/';
267 memcpy(ent->base + base_len, entry, len);
269 else
270 memcpy(ent->base, entry, pfxlen);
272 ent->name = ent->base + pfxlen + 1;
273 ent->base[pfxlen + 3] = '/';
274 ent->base[pfxlen] = ent->base[entlen-1] = 0;
276 /* Detect cases where alternate disappeared */
277 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
278 error("object directory %s does not exist; "
279 "check .git/objects/info/alternates.",
280 ent->base);
281 free(ent);
282 return -1;
285 /* Prevent the common mistake of listing the same
286 * thing twice, or object directory itself.
288 for (alt = alt_odb_list; alt; alt = alt->next) {
289 if (!memcmp(ent->base, alt->base, pfxlen)) {
290 free(ent);
291 return -1;
294 if (!memcmp(ent->base, objdir, pfxlen)) {
295 free(ent);
296 return -1;
299 /* add the alternate entry */
300 *alt_odb_tail = ent;
301 alt_odb_tail = &(ent->next);
302 ent->next = NULL;
304 /* recursively add alternates */
305 read_info_alternates(ent->base, depth + 1);
307 ent->base[pfxlen] = '/';
309 return 0;
312 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
313 const char *relative_base, int depth)
315 const char *cp, *last;
317 if (depth > 5) {
318 error("%s: ignoring alternate object stores, nesting too deep.",
319 relative_base);
320 return;
323 last = alt;
324 while (last < ep) {
325 cp = last;
326 if (cp < ep && *cp == '#') {
327 while (cp < ep && *cp != sep)
328 cp++;
329 last = cp + 1;
330 continue;
332 while (cp < ep && *cp != sep)
333 cp++;
334 if (last != cp) {
335 if ((*last != '/') && depth) {
336 error("%s: ignoring relative alternate object store %s",
337 relative_base, last);
338 } else {
339 link_alt_odb_entry(last, cp - last,
340 relative_base, depth);
343 while (cp < ep && *cp == sep)
344 cp++;
345 last = cp;
349 static void read_info_alternates(const char * relative_base, int depth)
351 char *map;
352 size_t mapsz;
353 struct stat st;
354 char path[PATH_MAX];
355 int fd;
357 sprintf(path, "%s/info/alternates", relative_base);
358 fd = open(path, O_RDONLY);
359 if (fd < 0)
360 return;
361 if (fstat(fd, &st) || (st.st_size == 0)) {
362 close(fd);
363 return;
365 mapsz = xsize_t(st.st_size);
366 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
367 close(fd);
369 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
371 munmap(map, mapsz);
374 void prepare_alt_odb(void)
376 const char *alt;
378 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
379 if (!alt) alt = "";
381 if (alt_odb_tail)
382 return;
383 alt_odb_tail = &alt_odb_list;
384 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
386 read_info_alternates(get_object_directory(), 0);
389 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
391 char *name = sha1_file_name(sha1);
392 struct alternate_object_database *alt;
394 if (!stat(name, st))
395 return name;
396 prepare_alt_odb();
397 for (alt = alt_odb_list; alt; alt = alt->next) {
398 name = alt->name;
399 fill_sha1_path(name, sha1);
400 if (!stat(alt->base, st))
401 return alt->base;
403 return NULL;
406 static unsigned int pack_used_ctr;
407 static unsigned int pack_mmap_calls;
408 static unsigned int peak_pack_open_windows;
409 static unsigned int pack_open_windows;
410 static size_t peak_pack_mapped;
411 static size_t pack_mapped;
412 struct packed_git *packed_git;
414 void pack_report()
416 fprintf(stderr,
417 "pack_report: getpagesize() = %10" SZ_FMT "\n"
418 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
419 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
420 (size_t) getpagesize(),
421 packed_git_window_size,
422 packed_git_limit);
423 fprintf(stderr,
424 "pack_report: pack_used_ctr = %10u\n"
425 "pack_report: pack_mmap_calls = %10u\n"
426 "pack_report: pack_open_windows = %10u / %10u\n"
427 "pack_report: pack_mapped = "
428 "%10" SZ_FMT " / %10" SZ_FMT "\n",
429 pack_used_ctr,
430 pack_mmap_calls,
431 pack_open_windows, peak_pack_open_windows,
432 pack_mapped, peak_pack_mapped);
435 static int check_packed_git_idx(const char *path, struct packed_git *p)
437 void *idx_map;
438 struct pack_idx_header *hdr;
439 size_t idx_size;
440 uint32_t nr, i, *index;
441 int fd = open(path, O_RDONLY);
442 struct stat st;
444 if (fd < 0)
445 return -1;
446 if (fstat(fd, &st)) {
447 close(fd);
448 return -1;
450 idx_size = xsize_t(st.st_size);
451 if (idx_size < 4 * 256 + 20 + 20) {
452 close(fd);
453 return error("index file %s is too small", path);
455 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
456 close(fd);
458 /* a future index format would start with this, as older git
459 * binaries would fail the non-monotonic index check below.
460 * give a nicer warning to the user if we can.
462 hdr = idx_map;
463 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
464 munmap(idx_map, idx_size);
465 return error("index file %s is a newer version"
466 " and is not supported by this binary"
467 " (try upgrading GIT to a newer version)",
468 path);
471 nr = 0;
472 index = idx_map;
473 for (i = 0; i < 256; i++) {
474 uint32_t n = ntohl(index[i]);
475 if (n < nr) {
476 munmap(idx_map, idx_size);
477 return error("non-monotonic index %s", path);
479 nr = n;
483 * Total size:
484 * - 256 index entries 4 bytes each
485 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
486 * - 20-byte SHA1 of the packfile
487 * - 20-byte SHA1 file checksum
489 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
490 munmap(idx_map, idx_size);
491 return error("wrong index file size in %s", path);
494 p->index_version = 1;
495 p->index_data = idx_map;
496 p->index_size = idx_size;
497 return 0;
500 static void scan_windows(struct packed_git *p,
501 struct packed_git **lru_p,
502 struct pack_window **lru_w,
503 struct pack_window **lru_l)
505 struct pack_window *w, *w_l;
507 for (w_l = NULL, w = p->windows; w; w = w->next) {
508 if (!w->inuse_cnt) {
509 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
510 *lru_p = p;
511 *lru_w = w;
512 *lru_l = w_l;
515 w_l = w;
519 static int unuse_one_window(struct packed_git *current)
521 struct packed_git *p, *lru_p = NULL;
522 struct pack_window *lru_w = NULL, *lru_l = NULL;
524 if (current)
525 scan_windows(current, &lru_p, &lru_w, &lru_l);
526 for (p = packed_git; p; p = p->next)
527 scan_windows(p, &lru_p, &lru_w, &lru_l);
528 if (lru_p) {
529 munmap(lru_w->base, lru_w->len);
530 pack_mapped -= lru_w->len;
531 if (lru_l)
532 lru_l->next = lru_w->next;
533 else {
534 lru_p->windows = lru_w->next;
535 if (!lru_p->windows && lru_p != current) {
536 close(lru_p->pack_fd);
537 lru_p->pack_fd = -1;
540 free(lru_w);
541 pack_open_windows--;
542 return 1;
544 return 0;
547 void release_pack_memory(size_t need)
549 size_t cur = pack_mapped;
550 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
551 ; /* nothing */
554 void unuse_pack(struct pack_window **w_cursor)
556 struct pack_window *w = *w_cursor;
557 if (w) {
558 w->inuse_cnt--;
559 *w_cursor = NULL;
564 * Do not call this directly as this leaks p->pack_fd on error return;
565 * call open_packed_git() instead.
567 static int open_packed_git_1(struct packed_git *p)
569 struct stat st;
570 struct pack_header hdr;
571 unsigned char sha1[20];
572 unsigned char *idx_sha1;
573 long fd_flag;
575 p->pack_fd = open(p->pack_name, O_RDONLY);
576 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
577 return -1;
579 /* If we created the struct before we had the pack we lack size. */
580 if (!p->pack_size) {
581 if (!S_ISREG(st.st_mode))
582 return error("packfile %s not a regular file", p->pack_name);
583 p->pack_size = st.st_size;
584 } else if (p->pack_size != st.st_size)
585 return error("packfile %s size changed", p->pack_name);
587 /* We leave these file descriptors open with sliding mmap;
588 * there is no point keeping them open across exec(), though.
590 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
591 if (fd_flag < 0)
592 return error("cannot determine file descriptor flags");
593 fd_flag |= FD_CLOEXEC;
594 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
595 return error("cannot set FD_CLOEXEC");
597 /* Verify we recognize this pack file format. */
598 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
599 return error("file %s is far too short to be a packfile", p->pack_name);
600 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
601 return error("file %s is not a GIT packfile", p->pack_name);
602 if (!pack_version_ok(hdr.hdr_version))
603 return error("packfile %s is version %u and not supported"
604 " (try upgrading GIT to a newer version)",
605 p->pack_name, ntohl(hdr.hdr_version));
607 /* Verify the pack matches its index. */
608 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
609 return error("packfile %s claims to have %u objects"
610 " while index size indicates %u objects",
611 p->pack_name, ntohl(hdr.hdr_entries),
612 num_packed_objects(p));
613 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
614 return error("end of packfile %s is unavailable", p->pack_name);
615 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
616 return error("packfile %s signature is unavailable", p->pack_name);
617 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
618 if (hashcmp(sha1, idx_sha1))
619 return error("packfile %s does not match index", p->pack_name);
620 return 0;
623 static int open_packed_git(struct packed_git *p)
625 if (!open_packed_git_1(p))
626 return 0;
627 if (p->pack_fd != -1) {
628 close(p->pack_fd);
629 p->pack_fd = -1;
631 return -1;
634 static int in_window(struct pack_window *win, off_t offset)
636 /* We must promise at least 20 bytes (one hash) after the
637 * offset is available from this window, otherwise the offset
638 * is not actually in this window and a different window (which
639 * has that one hash excess) must be used. This is to support
640 * the object header and delta base parsing routines below.
642 off_t win_off = win->offset;
643 return win_off <= offset
644 && (offset + 20) <= (win_off + win->len);
647 unsigned char* use_pack(struct packed_git *p,
648 struct pack_window **w_cursor,
649 off_t offset,
650 unsigned int *left)
652 struct pack_window *win = *w_cursor;
654 if (p->pack_fd == -1 && open_packed_git(p))
655 die("packfile %s cannot be accessed", p->pack_name);
657 /* Since packfiles end in a hash of their content and its
658 * pointless to ask for an offset into the middle of that
659 * hash, and the in_window function above wouldn't match
660 * don't allow an offset too close to the end of the file.
662 if (offset > (p->pack_size - 20))
663 die("offset beyond end of packfile (truncated pack?)");
665 if (!win || !in_window(win, offset)) {
666 if (win)
667 win->inuse_cnt--;
668 for (win = p->windows; win; win = win->next) {
669 if (in_window(win, offset))
670 break;
672 if (!win) {
673 size_t window_align = packed_git_window_size / 2;
674 off_t len;
675 win = xcalloc(1, sizeof(*win));
676 win->offset = (offset / window_align) * window_align;
677 len = p->pack_size - win->offset;
678 if (len > packed_git_window_size)
679 len = packed_git_window_size;
680 win->len = (size_t)len;
681 pack_mapped += win->len;
682 while (packed_git_limit < pack_mapped
683 && unuse_one_window(p))
684 ; /* nothing */
685 win->base = xmmap(NULL, win->len,
686 PROT_READ, MAP_PRIVATE,
687 p->pack_fd, win->offset);
688 if (win->base == MAP_FAILED)
689 die("packfile %s cannot be mapped: %s",
690 p->pack_name,
691 strerror(errno));
692 pack_mmap_calls++;
693 pack_open_windows++;
694 if (pack_mapped > peak_pack_mapped)
695 peak_pack_mapped = pack_mapped;
696 if (pack_open_windows > peak_pack_open_windows)
697 peak_pack_open_windows = pack_open_windows;
698 win->next = p->windows;
699 p->windows = win;
702 if (win != *w_cursor) {
703 win->last_used = pack_used_ctr++;
704 win->inuse_cnt++;
705 *w_cursor = win;
707 offset -= win->offset;
708 if (left)
709 *left = win->len - xsize_t(offset);
710 return win->base + offset;
713 struct packed_git *add_packed_git(const char *path, int path_len, int local)
715 struct stat st;
716 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
719 * Make sure a corresponding .pack file exists and that
720 * the index looks sane.
722 path_len -= strlen(".idx");
723 if (path_len < 1)
724 return NULL;
725 memcpy(p->pack_name, path, path_len);
726 strcpy(p->pack_name + path_len, ".pack");
727 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode) ||
728 check_packed_git_idx(path, p)) {
729 free(p);
730 return NULL;
733 /* ok, it looks sane as far as we can check without
734 * actually mapping the pack file.
736 p->pack_size = st.st_size;
737 p->next = NULL;
738 p->windows = NULL;
739 p->pack_fd = -1;
740 p->pack_local = local;
741 p->mtime = st.st_mtime;
742 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
743 hashclr(p->sha1);
744 return p;
747 struct packed_git *parse_pack_index(unsigned char *sha1)
749 char *path = sha1_pack_index_name(sha1);
750 return parse_pack_index_file(sha1, path);
753 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
754 const char *idx_path)
756 const char *path = sha1_pack_name(sha1);
757 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
759 if (check_packed_git_idx(idx_path, p)) {
760 free(p);
761 return NULL;
764 strcpy(p->pack_name, path);
765 p->pack_size = 0;
766 p->next = NULL;
767 p->windows = NULL;
768 p->pack_fd = -1;
769 hashcpy(p->sha1, sha1);
770 return p;
773 void install_packed_git(struct packed_git *pack)
775 pack->next = packed_git;
776 packed_git = pack;
779 static void prepare_packed_git_one(char *objdir, int local)
781 char path[PATH_MAX];
782 int len;
783 DIR *dir;
784 struct dirent *de;
786 sprintf(path, "%s/pack", objdir);
787 len = strlen(path);
788 dir = opendir(path);
789 if (!dir) {
790 if (errno != ENOENT)
791 error("unable to open object pack directory: %s: %s",
792 path, strerror(errno));
793 return;
795 path[len++] = '/';
796 while ((de = readdir(dir)) != NULL) {
797 int namelen = strlen(de->d_name);
798 struct packed_git *p;
800 if (!has_extension(de->d_name, ".idx"))
801 continue;
803 /* Don't reopen a pack we already have. */
804 strcpy(path + len, de->d_name);
805 for (p = packed_git; p; p = p->next) {
806 if (!memcmp(path, p->pack_name, len + namelen - 4))
807 break;
809 if (p)
810 continue;
811 /* See if it really is a valid .idx file with corresponding
812 * .pack file that we can map.
814 p = add_packed_git(path, len + namelen, local);
815 if (!p)
816 continue;
817 install_packed_git(p);
819 closedir(dir);
822 static int sort_pack(const void *a_, const void *b_)
824 struct packed_git *a = *((struct packed_git **)a_);
825 struct packed_git *b = *((struct packed_git **)b_);
826 int st;
829 * Local packs tend to contain objects specific to our
830 * variant of the project than remote ones. In addition,
831 * remote ones could be on a network mounted filesystem.
832 * Favor local ones for these reasons.
834 st = a->pack_local - b->pack_local;
835 if (st)
836 return -st;
839 * Younger packs tend to contain more recent objects,
840 * and more recent objects tend to get accessed more
841 * often.
843 if (a->mtime < b->mtime)
844 return 1;
845 else if (a->mtime == b->mtime)
846 return 0;
847 return -1;
850 static void rearrange_packed_git(void)
852 struct packed_git **ary, *p;
853 int i, n;
855 for (n = 0, p = packed_git; p; p = p->next)
856 n++;
857 if (n < 2)
858 return;
860 /* prepare an array of packed_git for easier sorting */
861 ary = xcalloc(n, sizeof(struct packed_git *));
862 for (n = 0, p = packed_git; p; p = p->next)
863 ary[n++] = p;
865 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
867 /* link them back again */
868 for (i = 0; i < n - 1; i++)
869 ary[i]->next = ary[i + 1];
870 ary[n - 1]->next = NULL;
871 packed_git = ary[0];
873 free(ary);
876 static int prepare_packed_git_run_once = 0;
877 void prepare_packed_git(void)
879 struct alternate_object_database *alt;
881 if (prepare_packed_git_run_once)
882 return;
883 prepare_packed_git_one(get_object_directory(), 1);
884 prepare_alt_odb();
885 for (alt = alt_odb_list; alt; alt = alt->next) {
886 alt->name[-1] = 0;
887 prepare_packed_git_one(alt->base, 0);
888 alt->name[-1] = '/';
890 rearrange_packed_git();
891 prepare_packed_git_run_once = 1;
894 void reprepare_packed_git(void)
896 prepare_packed_git_run_once = 0;
897 prepare_packed_git();
900 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
902 unsigned char real_sha1[20];
903 hash_sha1_file(map, size, type, real_sha1);
904 return hashcmp(sha1, real_sha1) ? -1 : 0;
907 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
909 struct stat st;
910 void *map;
911 int fd;
912 char *filename = find_sha1_file(sha1, &st);
914 if (!filename) {
915 return NULL;
918 fd = open(filename, O_RDONLY | sha1_file_open_flag);
919 if (fd < 0) {
920 /* See if it works without O_NOATIME */
921 switch (sha1_file_open_flag) {
922 default:
923 fd = open(filename, O_RDONLY);
924 if (fd >= 0)
925 break;
926 /* Fallthrough */
927 case 0:
928 return NULL;
931 /* If it failed once, it will probably fail again.
932 * Stop using O_NOATIME
934 sha1_file_open_flag = 0;
936 *size = xsize_t(st.st_size);
937 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
938 close(fd);
939 return map;
942 int legacy_loose_object(unsigned char *map)
944 unsigned int word;
947 * Is it a zlib-compressed buffer? If so, the first byte
948 * must be 0x78 (15-bit window size, deflated), and the
949 * first 16-bit word is evenly divisible by 31
951 word = (map[0] << 8) + map[1];
952 if (map[0] == 0x78 && !(word % 31))
953 return 1;
954 else
955 return 0;
958 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
960 unsigned shift;
961 unsigned char c;
962 unsigned long size;
963 unsigned long used = 0;
965 c = buf[used++];
966 *type = (c >> 4) & 7;
967 size = c & 15;
968 shift = 4;
969 while (c & 0x80) {
970 if (len <= used)
971 return 0;
972 if (sizeof(long) * 8 <= shift)
973 return 0;
974 c = buf[used++];
975 size += (c & 0x7f) << shift;
976 shift += 7;
978 *sizep = size;
979 return used;
982 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
984 unsigned long size, used;
985 static const char valid_loose_object_type[8] = {
986 0, /* OBJ_EXT */
987 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
988 0, /* "delta" and others are invalid in a loose object */
990 enum object_type type;
992 /* Get the data stream */
993 memset(stream, 0, sizeof(*stream));
994 stream->next_in = map;
995 stream->avail_in = mapsize;
996 stream->next_out = buffer;
997 stream->avail_out = bufsiz;
999 if (legacy_loose_object(map)) {
1000 inflateInit(stream);
1001 return inflate(stream, 0);
1004 used = unpack_object_header_gently(map, mapsize, &type, &size);
1005 if (!used || !valid_loose_object_type[type])
1006 return -1;
1007 map += used;
1008 mapsize -= used;
1010 /* Set up the stream for the rest.. */
1011 stream->next_in = map;
1012 stream->avail_in = mapsize;
1013 inflateInit(stream);
1015 /* And generate the fake traditional header */
1016 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1017 typename(type), size);
1018 return 0;
1021 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1023 int bytes = strlen(buffer) + 1;
1024 unsigned char *buf = xmalloc(1+size);
1025 unsigned long n;
1026 int status = Z_OK;
1028 n = stream->total_out - bytes;
1029 if (n > size)
1030 n = size;
1031 memcpy(buf, (char *) buffer + bytes, n);
1032 bytes = n;
1033 if (bytes < size) {
1034 stream->next_out = buf + bytes;
1035 stream->avail_out = size - bytes;
1036 while (status == Z_OK)
1037 status = inflate(stream, Z_FINISH);
1039 buf[size] = 0;
1040 if ((status == Z_OK || status == Z_STREAM_END) && !stream->avail_in) {
1041 inflateEnd(stream);
1042 return buf;
1045 if (status < 0)
1046 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1047 else if (stream->avail_in)
1048 error("garbage at end of loose object '%s'",
1049 sha1_to_hex(sha1));
1050 free(buf);
1051 return NULL;
1055 * We used to just use "sscanf()", but that's actually way
1056 * too permissive for what we want to check. So do an anal
1057 * object header parse by hand.
1059 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1061 char type[10];
1062 int i;
1063 unsigned long size;
1066 * The type can be at most ten bytes (including the
1067 * terminating '\0' that we add), and is followed by
1068 * a space.
1070 i = 0;
1071 for (;;) {
1072 char c = *hdr++;
1073 if (c == ' ')
1074 break;
1075 type[i++] = c;
1076 if (i >= sizeof(type))
1077 return -1;
1079 type[i] = 0;
1082 * The length must follow immediately, and be in canonical
1083 * decimal format (ie "010" is not valid).
1085 size = *hdr++ - '0';
1086 if (size > 9)
1087 return -1;
1088 if (size) {
1089 for (;;) {
1090 unsigned long c = *hdr - '0';
1091 if (c > 9)
1092 break;
1093 hdr++;
1094 size = size * 10 + c;
1097 *sizep = size;
1100 * The length must be followed by a zero byte
1102 return *hdr ? -1 : type_from_string(type);
1105 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1107 int ret;
1108 z_stream stream;
1109 char hdr[8192];
1111 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1112 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1113 return NULL;
1115 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1118 static off_t get_delta_base(struct packed_git *p,
1119 struct pack_window **w_curs,
1120 off_t *curpos,
1121 enum object_type type,
1122 off_t delta_obj_offset)
1124 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1125 off_t base_offset;
1127 /* use_pack() assured us we have [base_info, base_info + 20)
1128 * as a range that we can look at without walking off the
1129 * end of the mapped window. Its actually the hash size
1130 * that is assured. An OFS_DELTA longer than the hash size
1131 * is stupid, as then a REF_DELTA would be smaller to store.
1133 if (type == OBJ_OFS_DELTA) {
1134 unsigned used = 0;
1135 unsigned char c = base_info[used++];
1136 base_offset = c & 127;
1137 while (c & 128) {
1138 base_offset += 1;
1139 if (!base_offset || base_offset & ~(~0UL >> 7))
1140 die("offset value overflow for delta base object");
1141 c = base_info[used++];
1142 base_offset = (base_offset << 7) + (c & 127);
1144 base_offset = delta_obj_offset - base_offset;
1145 if (base_offset >= delta_obj_offset)
1146 die("delta base offset out of bound");
1147 *curpos += used;
1148 } else if (type == OBJ_REF_DELTA) {
1149 /* The base entry _must_ be in the same pack */
1150 base_offset = find_pack_entry_one(base_info, p);
1151 if (!base_offset)
1152 die("failed to find delta-pack base object %s",
1153 sha1_to_hex(base_info));
1154 *curpos += 20;
1155 } else
1156 die("I am totally screwed");
1157 return base_offset;
1160 /* forward declaration for a mutually recursive function */
1161 static int packed_object_info(struct packed_git *p, off_t offset,
1162 unsigned long *sizep);
1164 static int packed_delta_info(struct packed_git *p,
1165 struct pack_window **w_curs,
1166 off_t curpos,
1167 enum object_type type,
1168 off_t obj_offset,
1169 unsigned long *sizep)
1171 off_t base_offset;
1173 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1174 type = packed_object_info(p, base_offset, NULL);
1176 /* We choose to only get the type of the base object and
1177 * ignore potentially corrupt pack file that expects the delta
1178 * based on a base with a wrong size. This saves tons of
1179 * inflate() calls.
1181 if (sizep) {
1182 const unsigned char *data;
1183 unsigned char delta_head[20], *in;
1184 z_stream stream;
1185 int st;
1187 memset(&stream, 0, sizeof(stream));
1188 stream.next_out = delta_head;
1189 stream.avail_out = sizeof(delta_head);
1191 inflateInit(&stream);
1192 do {
1193 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1194 stream.next_in = in;
1195 st = inflate(&stream, Z_FINISH);
1196 curpos += stream.next_in - in;
1197 } while ((st == Z_OK || st == Z_BUF_ERROR)
1198 && stream.total_out < sizeof(delta_head));
1199 inflateEnd(&stream);
1200 if ((st != Z_STREAM_END) &&
1201 stream.total_out != sizeof(delta_head))
1202 die("delta data unpack-initial failed");
1204 /* Examine the initial part of the delta to figure out
1205 * the result size.
1207 data = delta_head;
1209 /* ignore base size */
1210 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1212 /* Read the result size */
1213 *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1216 return type;
1219 static int unpack_object_header(struct packed_git *p,
1220 struct pack_window **w_curs,
1221 off_t *curpos,
1222 unsigned long *sizep)
1224 unsigned char *base;
1225 unsigned int left;
1226 unsigned long used;
1227 enum object_type type;
1229 /* use_pack() assures us we have [base, base + 20) available
1230 * as a range that we can look at at. (Its actually the hash
1231 * size that is assured.) With our object header encoding
1232 * the maximum deflated object size is 2^137, which is just
1233 * insane, so we know won't exceed what we have been given.
1235 base = use_pack(p, w_curs, *curpos, &left);
1236 used = unpack_object_header_gently(base, left, &type, sizep);
1237 if (!used)
1238 die("object offset outside of pack file");
1239 *curpos += used;
1241 return type;
1244 const char *packed_object_info_detail(struct packed_git *p,
1245 off_t obj_offset,
1246 unsigned long *size,
1247 unsigned long *store_size,
1248 unsigned int *delta_chain_length,
1249 unsigned char *base_sha1)
1251 struct pack_window *w_curs = NULL;
1252 off_t curpos;
1253 unsigned long dummy;
1254 unsigned char *next_sha1;
1255 enum object_type type;
1257 *delta_chain_length = 0;
1258 curpos = obj_offset;
1259 type = unpack_object_header(p, &w_curs, &curpos, size);
1261 for (;;) {
1262 switch (type) {
1263 default:
1264 die("pack %s contains unknown object type %d",
1265 p->pack_name, type);
1266 case OBJ_COMMIT:
1267 case OBJ_TREE:
1268 case OBJ_BLOB:
1269 case OBJ_TAG:
1270 *store_size = 0; /* notyet */
1271 unuse_pack(&w_curs);
1272 return typename(type);
1273 case OBJ_OFS_DELTA:
1274 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1275 if (*delta_chain_length == 0) {
1276 /* TODO: find base_sha1 as pointed by curpos */
1277 hashclr(base_sha1);
1279 break;
1280 case OBJ_REF_DELTA:
1281 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1282 if (*delta_chain_length == 0)
1283 hashcpy(base_sha1, next_sha1);
1284 obj_offset = find_pack_entry_one(next_sha1, p);
1285 break;
1287 (*delta_chain_length)++;
1288 curpos = obj_offset;
1289 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1293 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1294 unsigned long *sizep)
1296 struct pack_window *w_curs = NULL;
1297 unsigned long size;
1298 off_t curpos = obj_offset;
1299 enum object_type type;
1301 type = unpack_object_header(p, &w_curs, &curpos, &size);
1303 switch (type) {
1304 case OBJ_OFS_DELTA:
1305 case OBJ_REF_DELTA:
1306 type = packed_delta_info(p, &w_curs, curpos,
1307 type, obj_offset, sizep);
1308 break;
1309 case OBJ_COMMIT:
1310 case OBJ_TREE:
1311 case OBJ_BLOB:
1312 case OBJ_TAG:
1313 if (sizep)
1314 *sizep = size;
1315 break;
1316 default:
1317 die("pack %s contains unknown object type %d",
1318 p->pack_name, type);
1320 unuse_pack(&w_curs);
1321 return type;
1324 static void *unpack_compressed_entry(struct packed_git *p,
1325 struct pack_window **w_curs,
1326 off_t curpos,
1327 unsigned long size)
1329 int st;
1330 z_stream stream;
1331 unsigned char *buffer, *in;
1333 buffer = xmalloc(size + 1);
1334 buffer[size] = 0;
1335 memset(&stream, 0, sizeof(stream));
1336 stream.next_out = buffer;
1337 stream.avail_out = size;
1339 inflateInit(&stream);
1340 do {
1341 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1342 stream.next_in = in;
1343 st = inflate(&stream, Z_FINISH);
1344 curpos += stream.next_in - in;
1345 } while (st == Z_OK || st == Z_BUF_ERROR);
1346 inflateEnd(&stream);
1347 if ((st != Z_STREAM_END) || stream.total_out != size) {
1348 free(buffer);
1349 return NULL;
1352 return buffer;
1355 #define MAX_DELTA_CACHE (256)
1357 static struct delta_base_cache_entry {
1358 struct packed_git *p;
1359 off_t base_offset;
1360 unsigned long size;
1361 void *data;
1362 enum object_type type;
1363 } delta_base_cache[MAX_DELTA_CACHE];
1365 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1367 unsigned long hash;
1369 hash = (unsigned long)p + (unsigned long)base_offset;
1370 hash += (hash >> 8) + (hash >> 16);
1371 return hash & 0xff;
1374 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1375 unsigned long *base_size, enum object_type *type, int keep_cache)
1377 void *ret;
1378 unsigned long hash = pack_entry_hash(p, base_offset);
1379 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1381 ret = ent->data;
1382 if (ret && ent->p == p && ent->base_offset == base_offset)
1383 goto found_cache_entry;
1384 return unpack_entry(p, base_offset, type, base_size);
1386 found_cache_entry:
1387 if (!keep_cache)
1388 ent->data = NULL;
1389 else {
1390 ret = xmalloc(ent->size + 1);
1391 memcpy(ret, ent->data, ent->size);
1392 ((char *)ret)[ent->size] = 0;
1394 *type = ent->type;
1395 *base_size = ent->size;
1396 return ret;
1399 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1400 void *base, unsigned long base_size, enum object_type type)
1402 unsigned long hash = pack_entry_hash(p, base_offset);
1403 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1405 if (ent->data)
1406 free(ent->data);
1407 ent->p = p;
1408 ent->base_offset = base_offset;
1409 ent->type = type;
1410 ent->data = base;
1411 ent->size = base_size;
1414 static void *unpack_delta_entry(struct packed_git *p,
1415 struct pack_window **w_curs,
1416 off_t curpos,
1417 unsigned long delta_size,
1418 off_t obj_offset,
1419 enum object_type *type,
1420 unsigned long *sizep)
1422 void *delta_data, *result, *base;
1423 unsigned long base_size;
1424 off_t base_offset;
1426 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1427 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1428 if (!base)
1429 die("failed to read delta base object"
1430 " at %"PRIuMAX" from %s",
1431 (uintmax_t)base_offset, p->pack_name);
1433 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1434 result = patch_delta(base, base_size,
1435 delta_data, delta_size,
1436 sizep);
1437 if (!result)
1438 die("failed to apply delta");
1439 free(delta_data);
1440 add_delta_base_cache(p, base_offset, base, base_size, *type);
1441 return result;
1444 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1445 enum object_type *type, unsigned long *sizep)
1447 struct pack_window *w_curs = NULL;
1448 off_t curpos = obj_offset;
1449 void *data;
1451 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1452 switch (*type) {
1453 case OBJ_OFS_DELTA:
1454 case OBJ_REF_DELTA:
1455 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1456 obj_offset, type, sizep);
1457 break;
1458 case OBJ_COMMIT:
1459 case OBJ_TREE:
1460 case OBJ_BLOB:
1461 case OBJ_TAG:
1462 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1463 break;
1464 default:
1465 die("unknown object type %i in %s", *type, p->pack_name);
1467 unuse_pack(&w_curs);
1468 return data;
1471 uint32_t num_packed_objects(const struct packed_git *p)
1473 /* See check_packed_git_idx() */
1474 return (uint32_t)((p->index_size - 20 - 20 - 4*256) / 24);
1477 int nth_packed_object_sha1(const struct packed_git *p, uint32_t n,
1478 unsigned char* sha1)
1480 const unsigned char *index = p->index_data;
1481 index += 4 * 256;
1482 if (num_packed_objects(p) <= n)
1483 return -1;
1484 hashcpy(sha1, index + 24 * n + 4);
1485 return 0;
1488 off_t find_pack_entry_one(const unsigned char *sha1,
1489 struct packed_git *p)
1491 const uint32_t *level1_ofs = p->index_data;
1492 int hi = ntohl(level1_ofs[*sha1]);
1493 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1494 const unsigned char *index = p->index_data;
1496 index += 4 * 256;
1498 do {
1499 int mi = (lo + hi) / 2;
1500 int cmp = hashcmp(index + 24 * mi + 4, sha1);
1501 if (!cmp)
1502 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1503 if (cmp > 0)
1504 hi = mi;
1505 else
1506 lo = mi+1;
1507 } while (lo < hi);
1508 return 0;
1511 static int matches_pack_name(struct packed_git *p, const char *ig)
1513 const char *last_c, *c;
1515 if (!strcmp(p->pack_name, ig))
1516 return 0;
1518 for (c = p->pack_name, last_c = c; *c;)
1519 if (*c == '/')
1520 last_c = ++c;
1521 else
1522 ++c;
1523 if (!strcmp(last_c, ig))
1524 return 0;
1526 return 1;
1529 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1531 struct packed_git *p;
1532 off_t offset;
1534 prepare_packed_git();
1536 for (p = packed_git; p; p = p->next) {
1537 if (ignore_packed) {
1538 const char **ig;
1539 for (ig = ignore_packed; *ig; ig++)
1540 if (!matches_pack_name(p, *ig))
1541 break;
1542 if (*ig)
1543 continue;
1545 offset = find_pack_entry_one(sha1, p);
1546 if (offset) {
1548 * We are about to tell the caller where they can
1549 * locate the requested object. We better make
1550 * sure the packfile is still here and can be
1551 * accessed before supplying that answer, as
1552 * it may have been deleted since the index
1553 * was loaded!
1555 if (p->pack_fd == -1 && open_packed_git(p)) {
1556 error("packfile %s cannot be accessed", p->pack_name);
1557 continue;
1559 e->offset = offset;
1560 e->p = p;
1561 hashcpy(e->sha1, sha1);
1562 return 1;
1565 return 0;
1568 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1569 struct packed_git *packs)
1571 struct packed_git *p;
1573 for (p = packs; p; p = p->next) {
1574 if (find_pack_entry_one(sha1, p))
1575 return p;
1577 return NULL;
1581 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1583 int status;
1584 unsigned long mapsize, size;
1585 void *map;
1586 z_stream stream;
1587 char hdr[32];
1589 map = map_sha1_file(sha1, &mapsize);
1590 if (!map)
1591 return error("unable to find %s", sha1_to_hex(sha1));
1592 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1593 status = error("unable to unpack %s header",
1594 sha1_to_hex(sha1));
1595 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1596 status = error("unable to parse %s header", sha1_to_hex(sha1));
1597 else if (sizep)
1598 *sizep = size;
1599 inflateEnd(&stream);
1600 munmap(map, mapsize);
1601 return status;
1604 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1606 struct pack_entry e;
1608 if (!find_pack_entry(sha1, &e, NULL)) {
1609 reprepare_packed_git();
1610 if (!find_pack_entry(sha1, &e, NULL))
1611 return sha1_loose_object_info(sha1, sizep);
1613 return packed_object_info(e.p, e.offset, sizep);
1616 static void *read_packed_sha1(const unsigned char *sha1,
1617 enum object_type *type, unsigned long *size)
1619 struct pack_entry e;
1621 if (!find_pack_entry(sha1, &e, NULL))
1622 return NULL;
1623 else
1624 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1628 * This is meant to hold a *small* number of objects that you would
1629 * want read_sha1_file() to be able to return, but yet you do not want
1630 * to write them into the object store (e.g. a browse-only
1631 * application).
1633 static struct cached_object {
1634 unsigned char sha1[20];
1635 enum object_type type;
1636 void *buf;
1637 unsigned long size;
1638 } *cached_objects;
1639 static int cached_object_nr, cached_object_alloc;
1641 static struct cached_object *find_cached_object(const unsigned char *sha1)
1643 int i;
1644 struct cached_object *co = cached_objects;
1646 for (i = 0; i < cached_object_nr; i++, co++) {
1647 if (!hashcmp(co->sha1, sha1))
1648 return co;
1650 return NULL;
1653 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1654 unsigned char *sha1)
1656 struct cached_object *co;
1658 hash_sha1_file(buf, len, typename(type), sha1);
1659 if (has_sha1_file(sha1) || find_cached_object(sha1))
1660 return 0;
1661 if (cached_object_alloc <= cached_object_nr) {
1662 cached_object_alloc = alloc_nr(cached_object_alloc);
1663 cached_objects = xrealloc(cached_objects,
1664 sizeof(*cached_objects) *
1665 cached_object_alloc);
1667 co = &cached_objects[cached_object_nr++];
1668 co->size = len;
1669 co->type = type;
1670 co->buf = xmalloc(len);
1671 memcpy(co->buf, buf, len);
1672 hashcpy(co->sha1, sha1);
1673 return 0;
1676 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1677 unsigned long *size)
1679 unsigned long mapsize;
1680 void *map, *buf;
1681 struct cached_object *co;
1683 co = find_cached_object(sha1);
1684 if (co) {
1685 buf = xmalloc(co->size + 1);
1686 memcpy(buf, co->buf, co->size);
1687 ((char*)buf)[co->size] = 0;
1688 *type = co->type;
1689 *size = co->size;
1690 return buf;
1693 buf = read_packed_sha1(sha1, type, size);
1694 if (buf)
1695 return buf;
1696 map = map_sha1_file(sha1, &mapsize);
1697 if (map) {
1698 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1699 munmap(map, mapsize);
1700 return buf;
1702 reprepare_packed_git();
1703 return read_packed_sha1(sha1, type, size);
1706 void *read_object_with_reference(const unsigned char *sha1,
1707 const char *required_type_name,
1708 unsigned long *size,
1709 unsigned char *actual_sha1_return)
1711 enum object_type type, required_type;
1712 void *buffer;
1713 unsigned long isize;
1714 unsigned char actual_sha1[20];
1716 required_type = type_from_string(required_type_name);
1717 hashcpy(actual_sha1, sha1);
1718 while (1) {
1719 int ref_length = -1;
1720 const char *ref_type = NULL;
1722 buffer = read_sha1_file(actual_sha1, &type, &isize);
1723 if (!buffer)
1724 return NULL;
1725 if (type == required_type) {
1726 *size = isize;
1727 if (actual_sha1_return)
1728 hashcpy(actual_sha1_return, actual_sha1);
1729 return buffer;
1731 /* Handle references */
1732 else if (type == OBJ_COMMIT)
1733 ref_type = "tree ";
1734 else if (type == OBJ_TAG)
1735 ref_type = "object ";
1736 else {
1737 free(buffer);
1738 return NULL;
1740 ref_length = strlen(ref_type);
1742 if (memcmp(buffer, ref_type, ref_length) ||
1743 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1744 free(buffer);
1745 return NULL;
1747 free(buffer);
1748 /* Now we have the ID of the referred-to object in
1749 * actual_sha1. Check again. */
1753 static void write_sha1_file_prepare(void *buf, unsigned long len,
1754 const char *type, unsigned char *sha1,
1755 char *hdr, int *hdrlen)
1757 SHA_CTX c;
1759 /* Generate the header */
1760 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1762 /* Sha1.. */
1763 SHA1_Init(&c);
1764 SHA1_Update(&c, hdr, *hdrlen);
1765 SHA1_Update(&c, buf, len);
1766 SHA1_Final(sha1, &c);
1770 * Link the tempfile to the final place, possibly creating the
1771 * last directory level as you do so.
1773 * Returns the errno on failure, 0 on success.
1775 static int link_temp_to_file(const char *tmpfile, const char *filename)
1777 int ret;
1778 char *dir;
1780 if (!link(tmpfile, filename))
1781 return 0;
1784 * Try to mkdir the last path component if that failed.
1786 * Re-try the "link()" regardless of whether the mkdir
1787 * succeeds, since a race might mean that somebody
1788 * else succeeded.
1790 ret = errno;
1791 dir = strrchr(filename, '/');
1792 if (dir) {
1793 *dir = 0;
1794 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1795 *dir = '/';
1796 return -2;
1798 *dir = '/';
1799 if (!link(tmpfile, filename))
1800 return 0;
1801 ret = errno;
1803 return ret;
1807 * Move the just written object into its final resting place
1809 int move_temp_to_file(const char *tmpfile, const char *filename)
1811 int ret = link_temp_to_file(tmpfile, filename);
1814 * Coda hack - coda doesn't like cross-directory links,
1815 * so we fall back to a rename, which will mean that it
1816 * won't be able to check collisions, but that's not a
1817 * big deal.
1819 * The same holds for FAT formatted media.
1821 * When this succeeds, we just return 0. We have nothing
1822 * left to unlink.
1824 if (ret && ret != EEXIST) {
1825 if (!rename(tmpfile, filename))
1826 return 0;
1827 ret = errno;
1829 unlink(tmpfile);
1830 if (ret) {
1831 if (ret != EEXIST) {
1832 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1834 /* FIXME!!! Collision check here ? */
1837 return 0;
1840 static int write_buffer(int fd, const void *buf, size_t len)
1842 if (write_in_full(fd, buf, len) < 0)
1843 return error("file write error (%s)", strerror(errno));
1844 return 0;
1847 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1849 int hdr_len;
1850 unsigned char c;
1852 c = (type << 4) | (len & 15);
1853 len >>= 4;
1854 hdr_len = 1;
1855 while (len) {
1856 *hdr++ = c | 0x80;
1857 hdr_len++;
1858 c = (len & 0x7f);
1859 len >>= 7;
1861 *hdr = c;
1862 return hdr_len;
1865 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1867 int obj_type, hdrlen;
1869 if (use_legacy_headers) {
1870 while (deflate(stream, 0) == Z_OK)
1871 /* nothing */;
1872 return;
1874 obj_type = type_from_string(type);
1875 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1876 stream->total_out = hdrlen;
1877 stream->next_out += hdrlen;
1878 stream->avail_out -= hdrlen;
1881 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1882 unsigned char *sha1)
1884 char hdr[32];
1885 int hdrlen;
1886 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1887 return 0;
1890 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1892 int size;
1893 unsigned char *compressed;
1894 z_stream stream;
1895 unsigned char sha1[20];
1896 char *filename;
1897 static char tmpfile[PATH_MAX];
1898 char hdr[32];
1899 int fd, hdrlen;
1901 /* Normally if we have it in the pack then we do not bother writing
1902 * it out into .git/objects/??/?{38} file.
1904 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1905 filename = sha1_file_name(sha1);
1906 if (returnsha1)
1907 hashcpy(returnsha1, sha1);
1908 if (has_sha1_file(sha1))
1909 return 0;
1910 fd = open(filename, O_RDONLY);
1911 if (fd >= 0) {
1913 * FIXME!!! We might do collision checking here, but we'd
1914 * need to uncompress the old file and check it. Later.
1916 close(fd);
1917 return 0;
1920 if (errno != ENOENT) {
1921 return error("sha1 file %s: %s\n", filename, strerror(errno));
1924 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1926 fd = mkstemp(tmpfile);
1927 if (fd < 0) {
1928 if (errno == EPERM)
1929 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1930 else
1931 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1934 /* Set it up */
1935 memset(&stream, 0, sizeof(stream));
1936 deflateInit(&stream, zlib_compression_level);
1937 size = 8 + deflateBound(&stream, len+hdrlen);
1938 compressed = xmalloc(size);
1940 /* Compress it */
1941 stream.next_out = compressed;
1942 stream.avail_out = size;
1944 /* First header.. */
1945 stream.next_in = (unsigned char *)hdr;
1946 stream.avail_in = hdrlen;
1947 setup_object_header(&stream, type, len);
1949 /* Then the data itself.. */
1950 stream.next_in = buf;
1951 stream.avail_in = len;
1952 while (deflate(&stream, Z_FINISH) == Z_OK)
1953 /* nothing */;
1954 deflateEnd(&stream);
1955 size = stream.total_out;
1957 if (write_buffer(fd, compressed, size) < 0)
1958 die("unable to write sha1 file");
1959 fchmod(fd, 0444);
1960 close(fd);
1961 free(compressed);
1963 return move_temp_to_file(tmpfile, filename);
1967 * We need to unpack and recompress the object for writing
1968 * it out to a different file.
1970 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1972 size_t size;
1973 z_stream stream;
1974 unsigned char *unpacked;
1975 unsigned long len;
1976 enum object_type type;
1977 char hdr[32];
1978 int hdrlen;
1979 void *buf;
1981 /* need to unpack and recompress it by itself */
1982 unpacked = read_packed_sha1(sha1, &type, &len);
1983 if (!unpacked)
1984 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1986 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1988 /* Set it up */
1989 memset(&stream, 0, sizeof(stream));
1990 deflateInit(&stream, zlib_compression_level);
1991 size = deflateBound(&stream, len + hdrlen);
1992 buf = xmalloc(size);
1994 /* Compress it */
1995 stream.next_out = buf;
1996 stream.avail_out = size;
1998 /* First header.. */
1999 stream.next_in = (void *)hdr;
2000 stream.avail_in = hdrlen;
2001 while (deflate(&stream, 0) == Z_OK)
2002 /* nothing */;
2004 /* Then the data itself.. */
2005 stream.next_in = unpacked;
2006 stream.avail_in = len;
2007 while (deflate(&stream, Z_FINISH) == Z_OK)
2008 /* nothing */;
2009 deflateEnd(&stream);
2010 free(unpacked);
2012 *objsize = stream.total_out;
2013 return buf;
2016 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2018 int retval;
2019 unsigned long objsize;
2020 void *buf = map_sha1_file(sha1, &objsize);
2022 if (buf) {
2023 retval = write_buffer(fd, buf, objsize);
2024 munmap(buf, objsize);
2025 return retval;
2028 buf = repack_object(sha1, &objsize);
2029 retval = write_buffer(fd, buf, objsize);
2030 free(buf);
2031 return retval;
2034 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2035 size_t bufsize, size_t *bufposn)
2037 char tmpfile[PATH_MAX];
2038 int local;
2039 z_stream stream;
2040 unsigned char real_sha1[20];
2041 unsigned char discard[4096];
2042 int ret;
2043 SHA_CTX c;
2045 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
2047 local = mkstemp(tmpfile);
2048 if (local < 0) {
2049 if (errno == EPERM)
2050 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2051 else
2052 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2055 memset(&stream, 0, sizeof(stream));
2057 inflateInit(&stream);
2059 SHA1_Init(&c);
2061 do {
2062 ssize_t size;
2063 if (*bufposn) {
2064 stream.avail_in = *bufposn;
2065 stream.next_in = (unsigned char *) buffer;
2066 do {
2067 stream.next_out = discard;
2068 stream.avail_out = sizeof(discard);
2069 ret = inflate(&stream, Z_SYNC_FLUSH);
2070 SHA1_Update(&c, discard, sizeof(discard) -
2071 stream.avail_out);
2072 } while (stream.avail_in && ret == Z_OK);
2073 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2074 die("unable to write sha1 file");
2075 memmove(buffer, buffer + *bufposn - stream.avail_in,
2076 stream.avail_in);
2077 *bufposn = stream.avail_in;
2078 if (ret != Z_OK)
2079 break;
2081 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2082 if (size <= 0) {
2083 close(local);
2084 unlink(tmpfile);
2085 if (!size)
2086 return error("Connection closed?");
2087 perror("Reading from connection");
2088 return -1;
2090 *bufposn += size;
2091 } while (1);
2092 inflateEnd(&stream);
2094 close(local);
2095 SHA1_Final(real_sha1, &c);
2096 if (ret != Z_STREAM_END) {
2097 unlink(tmpfile);
2098 return error("File %s corrupted", sha1_to_hex(sha1));
2100 if (hashcmp(sha1, real_sha1)) {
2101 unlink(tmpfile);
2102 return error("File %s has bad hash", sha1_to_hex(sha1));
2105 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2108 int has_pack_index(const unsigned char *sha1)
2110 struct stat st;
2111 if (stat(sha1_pack_index_name(sha1), &st))
2112 return 0;
2113 return 1;
2116 int has_pack_file(const unsigned char *sha1)
2118 struct stat st;
2119 if (stat(sha1_pack_name(sha1), &st))
2120 return 0;
2121 return 1;
2124 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2126 struct pack_entry e;
2127 return find_pack_entry(sha1, &e, ignore_packed);
2130 int has_sha1_file(const unsigned char *sha1)
2132 struct stat st;
2133 struct pack_entry e;
2135 if (find_pack_entry(sha1, &e, NULL))
2136 return 1;
2137 return find_sha1_file(sha1, &st) ? 1 : 0;
2141 * reads from fd as long as possible into a supplied buffer of size bytes.
2142 * If necessary the buffer's size is increased using realloc()
2144 * returns 0 if anything went fine and -1 otherwise
2146 * NOTE: both buf and size may change, but even when -1 is returned
2147 * you still have to free() it yourself.
2149 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2151 char* buf = *return_buf;
2152 unsigned long size = *return_size;
2153 int iret;
2154 unsigned long off = 0;
2156 do {
2157 iret = xread(fd, buf + off, size - off);
2158 if (iret > 0) {
2159 off += iret;
2160 if (off == size) {
2161 size *= 2;
2162 buf = xrealloc(buf, size);
2165 } while (iret > 0);
2167 *return_buf = buf;
2168 *return_size = off;
2170 if (iret < 0)
2171 return -1;
2172 return 0;
2175 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2177 unsigned long size = 4096;
2178 char *buf = xmalloc(size);
2179 int ret;
2181 if (read_pipe(fd, &buf, &size)) {
2182 free(buf);
2183 return -1;
2186 if (!type)
2187 type = blob_type;
2188 if (write_object)
2189 ret = write_sha1_file(buf, size, type, sha1);
2190 else
2191 ret = hash_sha1_file(buf, size, type, sha1);
2192 free(buf);
2193 return ret;
2196 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2197 enum object_type type, const char *path)
2199 size_t size = xsize_t(st->st_size);
2200 void *buf = NULL;
2201 int ret, re_allocated = 0;
2203 if (size)
2204 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2205 close(fd);
2207 if (!type)
2208 type = OBJ_BLOB;
2211 * Convert blobs to git internal format
2213 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2214 unsigned long nsize = size;
2215 char *nbuf = buf;
2216 if (convert_to_git(path, &nbuf, &nsize)) {
2217 if (size)
2218 munmap(buf, size);
2219 size = nsize;
2220 buf = nbuf;
2221 re_allocated = 1;
2225 if (write_object)
2226 ret = write_sha1_file(buf, size, typename(type), sha1);
2227 else
2228 ret = hash_sha1_file(buf, size, typename(type), sha1);
2229 if (re_allocated) {
2230 free(buf);
2231 return ret;
2233 if (size)
2234 munmap(buf, size);
2235 return ret;
2238 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2240 int fd;
2241 char *target;
2242 size_t len;
2244 switch (st->st_mode & S_IFMT) {
2245 case S_IFREG:
2246 fd = open(path, O_RDONLY);
2247 if (fd < 0)
2248 return error("open(\"%s\"): %s", path,
2249 strerror(errno));
2250 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2251 return error("%s: failed to insert into database",
2252 path);
2253 break;
2254 case S_IFLNK:
2255 len = xsize_t(st->st_size);
2256 target = xmalloc(len + 1);
2257 if (readlink(path, target, len + 1) != st->st_size) {
2258 char *errstr = strerror(errno);
2259 free(target);
2260 return error("readlink(\"%s\"): %s", path,
2261 errstr);
2263 if (!write_object)
2264 hash_sha1_file(target, len, blob_type, sha1);
2265 else if (write_sha1_file(target, len, blob_type, sha1))
2266 return error("%s: failed to insert into database",
2267 path);
2268 free(target);
2269 break;
2270 default:
2271 return error("%s: unsupported file type", path);
2273 return 0;
2276 int read_pack_header(int fd, struct pack_header *header)
2278 char *c = (char*)header;
2279 ssize_t remaining = sizeof(struct pack_header);
2280 do {
2281 ssize_t r = xread(fd, c, remaining);
2282 if (r <= 0)
2283 /* "eof before pack header was fully read" */
2284 return PH_ERROR_EOF;
2285 remaining -= r;
2286 c += r;
2287 } while (remaining > 0);
2288 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2289 /* "protocol error (pack signature mismatch detected)" */
2290 return PH_ERROR_PACK_SIGNATURE;
2291 if (!pack_version_ok(header->hdr_version))
2292 /* "protocol error (pack version unsupported)" */
2293 return PH_ERROR_PROTOCOL;
2294 return 0;