doc: hooks.txt said post-commit default sends an email, it doesn't
[git/jrn.git] / sha1_file.c
blob2eff14ac87e3596194733086b94342a3112e2eee
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 struct stat st;
353 char path[PATH_MAX];
354 int fd;
356 sprintf(path, "%s/info/alternates", relative_base);
357 fd = open(path, O_RDONLY);
358 if (fd < 0)
359 return;
360 if (fstat(fd, &st) || (st.st_size == 0)) {
361 close(fd);
362 return;
364 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
365 close(fd);
367 link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
369 munmap(map, st.st_size);
372 void prepare_alt_odb(void)
374 const char *alt;
376 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
377 if (!alt) alt = "";
379 if (alt_odb_tail)
380 return;
381 alt_odb_tail = &alt_odb_list;
382 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
384 read_info_alternates(get_object_directory(), 0);
387 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
389 char *name = sha1_file_name(sha1);
390 struct alternate_object_database *alt;
392 if (!stat(name, st))
393 return name;
394 prepare_alt_odb();
395 for (alt = alt_odb_list; alt; alt = alt->next) {
396 name = alt->name;
397 fill_sha1_path(name, sha1);
398 if (!stat(alt->base, st))
399 return alt->base;
401 return NULL;
404 static unsigned int pack_used_ctr;
405 static unsigned int pack_mmap_calls;
406 static unsigned int peak_pack_open_windows;
407 static unsigned int pack_open_windows;
408 static size_t peak_pack_mapped;
409 static size_t pack_mapped;
410 static size_t page_size;
411 struct packed_git *packed_git;
413 void pack_report()
415 fprintf(stderr,
416 "pack_report: getpagesize() = %10" SZ_FMT "\n"
417 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
418 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
419 page_size,
420 packed_git_window_size,
421 packed_git_limit);
422 fprintf(stderr,
423 "pack_report: pack_used_ctr = %10u\n"
424 "pack_report: pack_mmap_calls = %10u\n"
425 "pack_report: pack_open_windows = %10u / %10u\n"
426 "pack_report: pack_mapped = "
427 "%10" SZ_FMT " / %10" SZ_FMT "\n",
428 pack_used_ctr,
429 pack_mmap_calls,
430 pack_open_windows, peak_pack_open_windows,
431 pack_mapped, peak_pack_mapped);
434 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
435 void **idx_map_)
437 void *idx_map;
438 uint32_t *index;
439 unsigned long idx_size;
440 int nr, i;
441 int fd = open(path, O_RDONLY);
442 struct stat st;
443 if (fd < 0)
444 return -1;
445 if (fstat(fd, &st)) {
446 close(fd);
447 return -1;
449 idx_size = st.st_size;
450 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
451 close(fd);
453 index = idx_map;
454 *idx_map_ = idx_map;
455 *idx_size_ = idx_size;
457 /* check index map */
458 if (idx_size < 4*256 + 20 + 20)
459 return error("index file %s is too small", path);
461 /* a future index format would start with this, as older git
462 * binaries would fail the non-monotonic index check below.
463 * give a nicer warning to the user if we can.
465 if (index[0] == htonl(PACK_IDX_SIGNATURE))
466 return error("index file %s is a newer version"
467 " and is not supported by this binary"
468 " (try upgrading GIT to a newer version)",
469 path);
471 nr = 0;
472 for (i = 0; i < 256; i++) {
473 unsigned int n = ntohl(index[i]);
474 if (n < nr)
475 return error("non-monotonic index %s", path);
476 nr = n;
480 * Total size:
481 * - 256 index entries 4 bytes each
482 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
483 * - 20-byte SHA1 of the packfile
484 * - 20-byte SHA1 file checksum
486 if (idx_size != 4*256 + nr * 24 + 20 + 20)
487 return error("wrong index file size in %s", path);
489 return 0;
492 static void scan_windows(struct packed_git *p,
493 struct packed_git **lru_p,
494 struct pack_window **lru_w,
495 struct pack_window **lru_l)
497 struct pack_window *w, *w_l;
499 for (w_l = NULL, w = p->windows; w; w = w->next) {
500 if (!w->inuse_cnt) {
501 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
502 *lru_p = p;
503 *lru_w = w;
504 *lru_l = w_l;
507 w_l = w;
511 static int unuse_one_window(struct packed_git *current)
513 struct packed_git *p, *lru_p = NULL;
514 struct pack_window *lru_w = NULL, *lru_l = NULL;
516 if (current)
517 scan_windows(current, &lru_p, &lru_w, &lru_l);
518 for (p = packed_git; p; p = p->next)
519 scan_windows(p, &lru_p, &lru_w, &lru_l);
520 if (lru_p) {
521 munmap(lru_w->base, lru_w->len);
522 pack_mapped -= lru_w->len;
523 if (lru_l)
524 lru_l->next = lru_w->next;
525 else {
526 lru_p->windows = lru_w->next;
527 if (!lru_p->windows && lru_p != current) {
528 close(lru_p->pack_fd);
529 lru_p->pack_fd = -1;
532 free(lru_w);
533 pack_open_windows--;
534 return 1;
536 return 0;
539 void release_pack_memory(size_t need)
541 size_t cur = pack_mapped;
542 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
543 ; /* nothing */
546 void unuse_pack(struct pack_window **w_cursor)
548 struct pack_window *w = *w_cursor;
549 if (w) {
550 w->inuse_cnt--;
551 *w_cursor = NULL;
555 static int open_packed_git(struct packed_git *p)
557 struct stat st;
558 struct pack_header hdr;
559 unsigned char sha1[20];
560 unsigned char *idx_sha1;
561 long fd_flag;
563 p->pack_fd = open(p->pack_name, O_RDONLY);
564 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
565 return -1;
567 /* If we created the struct before we had the pack we lack size. */
568 if (!p->pack_size) {
569 if (!S_ISREG(st.st_mode))
570 return error("packfile %s not a regular file", p->pack_name);
571 p->pack_size = st.st_size;
572 } else if (p->pack_size != st.st_size)
573 return error("packfile %s size changed", p->pack_name);
575 /* We leave these file descriptors open with sliding mmap;
576 * there is no point keeping them open across exec(), though.
578 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
579 if (fd_flag < 0)
580 return error("cannot determine file descriptor flags");
581 fd_flag |= FD_CLOEXEC;
582 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
583 return error("cannot set FD_CLOEXEC");
585 /* Verify we recognize this pack file format. */
586 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
587 return error("file %s is far too short to be a packfile", p->pack_name);
588 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
589 return error("file %s is not a GIT packfile", p->pack_name);
590 if (!pack_version_ok(hdr.hdr_version))
591 return error("packfile %s is version %u and not supported"
592 " (try upgrading GIT to a newer version)",
593 p->pack_name, ntohl(hdr.hdr_version));
595 /* Verify the pack matches its index. */
596 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
597 return error("packfile %s claims to have %u objects"
598 " while index size indicates %u objects",
599 p->pack_name, ntohl(hdr.hdr_entries),
600 num_packed_objects(p));
601 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
602 return error("end of packfile %s is unavailable", p->pack_name);
603 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
604 return error("packfile %s signature is unavailable", p->pack_name);
605 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
606 if (hashcmp(sha1, idx_sha1))
607 return error("packfile %s does not match index", p->pack_name);
608 return 0;
611 static int in_window(struct pack_window *win, unsigned long offset)
613 /* We must promise at least 20 bytes (one hash) after the
614 * offset is available from this window, otherwise the offset
615 * is not actually in this window and a different window (which
616 * has that one hash excess) must be used. This is to support
617 * the object header and delta base parsing routines below.
619 off_t win_off = win->offset;
620 return win_off <= offset
621 && (offset + 20) <= (win_off + win->len);
624 unsigned char* use_pack(struct packed_git *p,
625 struct pack_window **w_cursor,
626 unsigned long offset,
627 unsigned int *left)
629 struct pack_window *win = *w_cursor;
631 if (p->pack_fd == -1 && open_packed_git(p))
632 die("packfile %s cannot be accessed", p->pack_name);
634 /* Since packfiles end in a hash of their content and its
635 * pointless to ask for an offset into the middle of that
636 * hash, and the in_window function above wouldn't match
637 * don't allow an offset too close to the end of the file.
639 if (offset > (p->pack_size - 20))
640 die("offset beyond end of packfile (truncated pack?)");
642 if (!win || !in_window(win, offset)) {
643 if (win)
644 win->inuse_cnt--;
645 for (win = p->windows; win; win = win->next) {
646 if (in_window(win, offset))
647 break;
649 if (!win) {
650 if (!page_size)
651 page_size = getpagesize();
652 win = xcalloc(1, sizeof(*win));
653 win->offset = (offset / page_size) * page_size;
654 win->len = p->pack_size - win->offset;
655 if (win->len > packed_git_window_size)
656 win->len = packed_git_window_size;
657 pack_mapped += win->len;
658 while (packed_git_limit < pack_mapped
659 && unuse_one_window(p))
660 ; /* nothing */
661 win->base = xmmap(NULL, win->len,
662 PROT_READ, MAP_PRIVATE,
663 p->pack_fd, win->offset);
664 if (win->base == MAP_FAILED)
665 die("packfile %s cannot be mapped: %s",
666 p->pack_name,
667 strerror(errno));
668 pack_mmap_calls++;
669 pack_open_windows++;
670 if (pack_mapped > peak_pack_mapped)
671 peak_pack_mapped = pack_mapped;
672 if (pack_open_windows > peak_pack_open_windows)
673 peak_pack_open_windows = pack_open_windows;
674 win->next = p->windows;
675 p->windows = win;
678 if (win != *w_cursor) {
679 win->last_used = pack_used_ctr++;
680 win->inuse_cnt++;
681 *w_cursor = win;
683 offset -= win->offset;
684 if (left)
685 *left = win->len - offset;
686 return win->base + offset;
689 struct packed_git *add_packed_git(char *path, int path_len, int local)
691 struct stat st;
692 struct packed_git *p;
693 unsigned long idx_size;
694 void *idx_map;
695 unsigned char sha1[20];
697 if (check_packed_git_idx(path, &idx_size, &idx_map))
698 return NULL;
700 /* do we have a corresponding .pack file? */
701 strcpy(path + path_len - 4, ".pack");
702 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
703 munmap(idx_map, idx_size);
704 return NULL;
706 /* ok, it looks sane as far as we can check without
707 * actually mapping the pack file.
709 p = xmalloc(sizeof(*p) + path_len + 2);
710 strcpy(p->pack_name, path);
711 p->index_size = idx_size;
712 p->pack_size = st.st_size;
713 p->index_base = idx_map;
714 p->next = NULL;
715 p->windows = NULL;
716 p->pack_fd = -1;
717 p->pack_local = local;
718 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
719 hashcpy(p->sha1, sha1);
720 return p;
723 struct packed_git *parse_pack_index(unsigned char *sha1)
725 char *path = sha1_pack_index_name(sha1);
726 return parse_pack_index_file(sha1, path);
729 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
731 struct packed_git *p;
732 unsigned long idx_size;
733 void *idx_map;
734 char *path;
736 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
737 return NULL;
739 path = sha1_pack_name(sha1);
741 p = xmalloc(sizeof(*p) + strlen(path) + 2);
742 strcpy(p->pack_name, path);
743 p->index_size = idx_size;
744 p->pack_size = 0;
745 p->index_base = idx_map;
746 p->next = NULL;
747 p->windows = NULL;
748 p->pack_fd = -1;
749 hashcpy(p->sha1, sha1);
750 return p;
753 void install_packed_git(struct packed_git *pack)
755 pack->next = packed_git;
756 packed_git = pack;
759 static void prepare_packed_git_one(char *objdir, int local)
761 char path[PATH_MAX];
762 int len;
763 DIR *dir;
764 struct dirent *de;
766 sprintf(path, "%s/pack", objdir);
767 len = strlen(path);
768 dir = opendir(path);
769 if (!dir) {
770 if (errno != ENOENT)
771 error("unable to open object pack directory: %s: %s",
772 path, strerror(errno));
773 return;
775 path[len++] = '/';
776 while ((de = readdir(dir)) != NULL) {
777 int namelen = strlen(de->d_name);
778 struct packed_git *p;
780 if (!has_extension(de->d_name, ".idx"))
781 continue;
783 /* Don't reopen a pack we already have. */
784 strcpy(path + len, de->d_name);
785 for (p = packed_git; p; p = p->next) {
786 if (!memcmp(path, p->pack_name, len + namelen - 4))
787 break;
789 if (p)
790 continue;
791 /* See if it really is a valid .idx file with corresponding
792 * .pack file that we can map.
794 p = add_packed_git(path, len + namelen, local);
795 if (!p)
796 continue;
797 install_packed_git(p);
799 closedir(dir);
802 static int prepare_packed_git_run_once = 0;
803 void prepare_packed_git(void)
805 struct alternate_object_database *alt;
807 if (prepare_packed_git_run_once)
808 return;
809 prepare_packed_git_one(get_object_directory(), 1);
810 prepare_alt_odb();
811 for (alt = alt_odb_list; alt; alt = alt->next) {
812 alt->name[-1] = 0;
813 prepare_packed_git_one(alt->base, 0);
814 alt->name[-1] = '/';
816 prepare_packed_git_run_once = 1;
819 void reprepare_packed_git(void)
821 prepare_packed_git_run_once = 0;
822 prepare_packed_git();
825 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
827 unsigned char real_sha1[20];
828 hash_sha1_file(map, size, type, real_sha1);
829 return hashcmp(sha1, real_sha1) ? -1 : 0;
832 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
834 struct stat st;
835 void *map;
836 int fd;
837 char *filename = find_sha1_file(sha1, &st);
839 if (!filename) {
840 return NULL;
843 fd = open(filename, O_RDONLY | sha1_file_open_flag);
844 if (fd < 0) {
845 /* See if it works without O_NOATIME */
846 switch (sha1_file_open_flag) {
847 default:
848 fd = open(filename, O_RDONLY);
849 if (fd >= 0)
850 break;
851 /* Fallthrough */
852 case 0:
853 return NULL;
856 /* If it failed once, it will probably fail again.
857 * Stop using O_NOATIME
859 sha1_file_open_flag = 0;
861 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
862 close(fd);
863 *size = st.st_size;
864 return map;
867 int legacy_loose_object(unsigned char *map)
869 unsigned int word;
872 * Is it a zlib-compressed buffer? If so, the first byte
873 * must be 0x78 (15-bit window size, deflated), and the
874 * first 16-bit word is evenly divisible by 31
876 word = (map[0] << 8) + map[1];
877 if (map[0] == 0x78 && !(word % 31))
878 return 1;
879 else
880 return 0;
883 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
885 unsigned shift;
886 unsigned char c;
887 unsigned long size;
888 unsigned long used = 0;
890 c = buf[used++];
891 *type = (c >> 4) & 7;
892 size = c & 15;
893 shift = 4;
894 while (c & 0x80) {
895 if (len <= used)
896 return 0;
897 if (sizeof(long) * 8 <= shift)
898 return 0;
899 c = buf[used++];
900 size += (c & 0x7f) << shift;
901 shift += 7;
903 *sizep = size;
904 return used;
907 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
909 unsigned long size, used;
910 static const char valid_loose_object_type[8] = {
911 0, /* OBJ_EXT */
912 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
913 0, /* "delta" and others are invalid in a loose object */
915 enum object_type type;
917 /* Get the data stream */
918 memset(stream, 0, sizeof(*stream));
919 stream->next_in = map;
920 stream->avail_in = mapsize;
921 stream->next_out = buffer;
922 stream->avail_out = bufsiz;
924 if (legacy_loose_object(map)) {
925 inflateInit(stream);
926 return inflate(stream, 0);
929 used = unpack_object_header_gently(map, mapsize, &type, &size);
930 if (!used || !valid_loose_object_type[type])
931 return -1;
932 map += used;
933 mapsize -= used;
935 /* Set up the stream for the rest.. */
936 stream->next_in = map;
937 stream->avail_in = mapsize;
938 inflateInit(stream);
940 /* And generate the fake traditional header */
941 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
942 type_names[type], size);
943 return 0;
946 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
948 int bytes = strlen(buffer) + 1;
949 unsigned char *buf = xmalloc(1+size);
950 unsigned long n;
952 n = stream->total_out - bytes;
953 if (n > size)
954 n = size;
955 memcpy(buf, (char *) buffer + bytes, n);
956 bytes = n;
957 if (bytes < size) {
958 stream->next_out = buf + bytes;
959 stream->avail_out = size - bytes;
960 while (inflate(stream, Z_FINISH) == Z_OK)
961 /* nothing */;
963 buf[size] = 0;
964 inflateEnd(stream);
965 return buf;
969 * We used to just use "sscanf()", but that's actually way
970 * too permissive for what we want to check. So do an anal
971 * object header parse by hand.
973 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
975 int i;
976 unsigned long size;
979 * The type can be at most ten bytes (including the
980 * terminating '\0' that we add), and is followed by
981 * a space.
983 i = 10;
984 for (;;) {
985 char c = *hdr++;
986 if (c == ' ')
987 break;
988 if (!--i)
989 return -1;
990 *type++ = c;
992 *type = 0;
995 * The length must follow immediately, and be in canonical
996 * decimal format (ie "010" is not valid).
998 size = *hdr++ - '0';
999 if (size > 9)
1000 return -1;
1001 if (size) {
1002 for (;;) {
1003 unsigned long c = *hdr - '0';
1004 if (c > 9)
1005 break;
1006 hdr++;
1007 size = size * 10 + c;
1010 *sizep = size;
1013 * The length must be followed by a zero byte
1015 return *hdr ? -1 : 0;
1018 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1020 int ret;
1021 z_stream stream;
1022 char hdr[8192];
1024 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1025 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1026 return NULL;
1028 return unpack_sha1_rest(&stream, hdr, *size);
1031 static unsigned long get_delta_base(struct packed_git *p,
1032 struct pack_window **w_curs,
1033 unsigned long offset,
1034 enum object_type kind,
1035 unsigned long delta_obj_offset,
1036 unsigned long *base_obj_offset)
1038 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1039 unsigned long base_offset;
1041 /* use_pack() assured us we have [base_info, base_info + 20)
1042 * as a range that we can look at without walking off the
1043 * end of the mapped window. Its actually the hash size
1044 * that is assured. An OFS_DELTA longer than the hash size
1045 * is stupid, as then a REF_DELTA would be smaller to store.
1047 if (kind == OBJ_OFS_DELTA) {
1048 unsigned used = 0;
1049 unsigned char c = base_info[used++];
1050 base_offset = c & 127;
1051 while (c & 128) {
1052 base_offset += 1;
1053 if (!base_offset || base_offset & ~(~0UL >> 7))
1054 die("offset value overflow for delta base object");
1055 c = base_info[used++];
1056 base_offset = (base_offset << 7) + (c & 127);
1058 base_offset = delta_obj_offset - base_offset;
1059 if (base_offset >= delta_obj_offset)
1060 die("delta base offset out of bound");
1061 offset += used;
1062 } else if (kind == OBJ_REF_DELTA) {
1063 /* The base entry _must_ be in the same pack */
1064 base_offset = find_pack_entry_one(base_info, p);
1065 if (!base_offset)
1066 die("failed to find delta-pack base object %s",
1067 sha1_to_hex(base_info));
1068 offset += 20;
1069 } else
1070 die("I am totally screwed");
1071 *base_obj_offset = base_offset;
1072 return offset;
1075 /* forward declaration for a mutually recursive function */
1076 static int packed_object_info(struct packed_git *p, unsigned long offset,
1077 char *type, unsigned long *sizep);
1079 static int packed_delta_info(struct packed_git *p,
1080 struct pack_window **w_curs,
1081 unsigned long offset,
1082 enum object_type kind,
1083 unsigned long obj_offset,
1084 char *type,
1085 unsigned long *sizep)
1087 unsigned long base_offset;
1089 offset = get_delta_base(p, w_curs, offset, kind,
1090 obj_offset, &base_offset);
1092 /* We choose to only get the type of the base object and
1093 * ignore potentially corrupt pack file that expects the delta
1094 * based on a base with a wrong size. This saves tons of
1095 * inflate() calls.
1097 if (packed_object_info(p, base_offset, type, NULL))
1098 die("cannot get info for delta-pack base");
1100 if (sizep) {
1101 const unsigned char *data;
1102 unsigned char delta_head[20], *in;
1103 unsigned long result_size;
1104 z_stream stream;
1105 int st;
1107 memset(&stream, 0, sizeof(stream));
1108 stream.next_out = delta_head;
1109 stream.avail_out = sizeof(delta_head);
1111 inflateInit(&stream);
1112 do {
1113 in = use_pack(p, w_curs, offset, &stream.avail_in);
1114 stream.next_in = in;
1115 st = inflate(&stream, Z_FINISH);
1116 offset += stream.next_in - in;
1117 } while ((st == Z_OK || st == Z_BUF_ERROR)
1118 && stream.total_out < sizeof(delta_head));
1119 inflateEnd(&stream);
1120 if ((st != Z_STREAM_END) &&
1121 stream.total_out != sizeof(delta_head))
1122 die("delta data unpack-initial failed");
1124 /* Examine the initial part of the delta to figure out
1125 * the result size.
1127 data = delta_head;
1129 /* ignore base size */
1130 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1132 /* Read the result size */
1133 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1134 *sizep = result_size;
1136 return 0;
1139 static unsigned long unpack_object_header(struct packed_git *p,
1140 struct pack_window **w_curs,
1141 unsigned long offset,
1142 enum object_type *type,
1143 unsigned long *sizep)
1145 unsigned char *base;
1146 unsigned int left;
1147 unsigned long used;
1149 /* use_pack() assures us we have [base, base + 20) available
1150 * as a range that we can look at at. (Its actually the hash
1151 * size that is assurred.) With our object header encoding
1152 * the maximum deflated object size is 2^137, which is just
1153 * insane, so we know won't exceed what we have been given.
1155 base = use_pack(p, w_curs, offset, &left);
1156 used = unpack_object_header_gently(base, left, type, sizep);
1157 if (!used)
1158 die("object offset outside of pack file");
1160 return offset + used;
1163 void packed_object_info_detail(struct packed_git *p,
1164 unsigned long offset,
1165 char *type,
1166 unsigned long *size,
1167 unsigned long *store_size,
1168 unsigned int *delta_chain_length,
1169 unsigned char *base_sha1)
1171 struct pack_window *w_curs = NULL;
1172 unsigned long obj_offset, val;
1173 unsigned char *next_sha1;
1174 enum object_type kind;
1176 *delta_chain_length = 0;
1177 obj_offset = offset;
1178 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1180 for (;;) {
1181 switch (kind) {
1182 default:
1183 die("pack %s contains unknown object type %d",
1184 p->pack_name, kind);
1185 case OBJ_COMMIT:
1186 case OBJ_TREE:
1187 case OBJ_BLOB:
1188 case OBJ_TAG:
1189 strcpy(type, type_names[kind]);
1190 *store_size = 0; /* notyet */
1191 unuse_pack(&w_curs);
1192 return;
1193 case OBJ_OFS_DELTA:
1194 get_delta_base(p, &w_curs, offset, kind,
1195 obj_offset, &offset);
1196 if (*delta_chain_length == 0) {
1197 /* TODO: find base_sha1 as pointed by offset */
1199 break;
1200 case OBJ_REF_DELTA:
1201 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1202 if (*delta_chain_length == 0)
1203 hashcpy(base_sha1, next_sha1);
1204 offset = find_pack_entry_one(next_sha1, p);
1205 break;
1207 obj_offset = offset;
1208 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1209 (*delta_chain_length)++;
1213 static int packed_object_info(struct packed_git *p, unsigned long offset,
1214 char *type, unsigned long *sizep)
1216 struct pack_window *w_curs = NULL;
1217 unsigned long size, obj_offset = offset;
1218 enum object_type kind;
1219 int r;
1221 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1223 switch (kind) {
1224 case OBJ_OFS_DELTA:
1225 case OBJ_REF_DELTA:
1226 r = packed_delta_info(p, &w_curs, offset, kind,
1227 obj_offset, type, sizep);
1228 unuse_pack(&w_curs);
1229 return r;
1230 case OBJ_COMMIT:
1231 case OBJ_TREE:
1232 case OBJ_BLOB:
1233 case OBJ_TAG:
1234 strcpy(type, type_names[kind]);
1235 unuse_pack(&w_curs);
1236 break;
1237 default:
1238 die("pack %s contains unknown object type %d",
1239 p->pack_name, kind);
1241 if (sizep)
1242 *sizep = size;
1243 return 0;
1246 static void *unpack_compressed_entry(struct packed_git *p,
1247 struct pack_window **w_curs,
1248 unsigned long offset,
1249 unsigned long size)
1251 int st;
1252 z_stream stream;
1253 unsigned char *buffer, *in;
1255 buffer = xmalloc(size + 1);
1256 buffer[size] = 0;
1257 memset(&stream, 0, sizeof(stream));
1258 stream.next_out = buffer;
1259 stream.avail_out = size;
1261 inflateInit(&stream);
1262 do {
1263 in = use_pack(p, w_curs, offset, &stream.avail_in);
1264 stream.next_in = in;
1265 st = inflate(&stream, Z_FINISH);
1266 offset += stream.next_in - in;
1267 } while (st == Z_OK || st == Z_BUF_ERROR);
1268 inflateEnd(&stream);
1269 if ((st != Z_STREAM_END) || stream.total_out != size) {
1270 free(buffer);
1271 return NULL;
1274 return buffer;
1277 static void *unpack_delta_entry(struct packed_git *p,
1278 struct pack_window **w_curs,
1279 unsigned long offset,
1280 unsigned long delta_size,
1281 enum object_type kind,
1282 unsigned long obj_offset,
1283 char *type,
1284 unsigned long *sizep)
1286 void *delta_data, *result, *base;
1287 unsigned long result_size, base_size, base_offset;
1289 offset = get_delta_base(p, w_curs, offset, kind,
1290 obj_offset, &base_offset);
1291 base = unpack_entry(p, base_offset, type, &base_size);
1292 if (!base)
1293 die("failed to read delta base object at %lu from %s",
1294 base_offset, p->pack_name);
1296 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1297 result = patch_delta(base, base_size,
1298 delta_data, delta_size,
1299 &result_size);
1300 if (!result)
1301 die("failed to apply delta");
1302 free(delta_data);
1303 free(base);
1304 *sizep = result_size;
1305 return result;
1308 void *unpack_entry(struct packed_git *p, unsigned long offset,
1309 char *type, unsigned long *sizep)
1311 struct pack_window *w_curs = NULL;
1312 unsigned long size, obj_offset = offset;
1313 enum object_type kind;
1314 void *retval;
1316 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1317 switch (kind) {
1318 case OBJ_OFS_DELTA:
1319 case OBJ_REF_DELTA:
1320 retval = unpack_delta_entry(p, &w_curs, offset, size,
1321 kind, obj_offset, type, sizep);
1322 break;
1323 case OBJ_COMMIT:
1324 case OBJ_TREE:
1325 case OBJ_BLOB:
1326 case OBJ_TAG:
1327 strcpy(type, type_names[kind]);
1328 *sizep = size;
1329 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1330 break;
1331 default:
1332 die("unknown object type %i in %s", kind, p->pack_name);
1334 unuse_pack(&w_curs);
1335 return retval;
1338 int num_packed_objects(const struct packed_git *p)
1340 /* See check_packed_git_idx() */
1341 return (p->index_size - 20 - 20 - 4*256) / 24;
1344 int nth_packed_object_sha1(const struct packed_git *p, int n,
1345 unsigned char* sha1)
1347 void *index = p->index_base + 256;
1348 if (n < 0 || num_packed_objects(p) <= n)
1349 return -1;
1350 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1351 return 0;
1354 unsigned long find_pack_entry_one(const unsigned char *sha1,
1355 struct packed_git *p)
1357 uint32_t *level1_ofs = p->index_base;
1358 int hi = ntohl(level1_ofs[*sha1]);
1359 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1360 void *index = p->index_base + 256;
1362 do {
1363 int mi = (lo + hi) / 2;
1364 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1365 if (!cmp)
1366 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1367 if (cmp > 0)
1368 hi = mi;
1369 else
1370 lo = mi+1;
1371 } while (lo < hi);
1372 return 0;
1375 static int matches_pack_name(struct packed_git *p, const char *ig)
1377 const char *last_c, *c;
1379 if (!strcmp(p->pack_name, ig))
1380 return 0;
1382 for (c = p->pack_name, last_c = c; *c;)
1383 if (*c == '/')
1384 last_c = ++c;
1385 else
1386 ++c;
1387 if (!strcmp(last_c, ig))
1388 return 0;
1390 return 1;
1393 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1395 struct packed_git *p;
1396 unsigned long offset;
1398 prepare_packed_git();
1400 for (p = packed_git; p; p = p->next) {
1401 if (ignore_packed) {
1402 const char **ig;
1403 for (ig = ignore_packed; *ig; ig++)
1404 if (!matches_pack_name(p, *ig))
1405 break;
1406 if (*ig)
1407 continue;
1409 offset = find_pack_entry_one(sha1, p);
1410 if (offset) {
1412 * We are about to tell the caller where they can
1413 * locate the requested object. We better make
1414 * sure the packfile is still here and can be
1415 * accessed before supplying that answer, as
1416 * it may have been deleted since the index
1417 * was loaded!
1419 if (p->pack_fd == -1 && open_packed_git(p)) {
1420 error("packfile %s cannot be accessed", p->pack_name);
1421 continue;
1423 e->offset = offset;
1424 e->p = p;
1425 hashcpy(e->sha1, sha1);
1426 return 1;
1429 return 0;
1432 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1433 struct packed_git *packs)
1435 struct packed_git *p;
1437 for (p = packs; p; p = p->next) {
1438 if (find_pack_entry_one(sha1, p))
1439 return p;
1441 return NULL;
1445 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1447 int status;
1448 unsigned long mapsize, size;
1449 void *map;
1450 z_stream stream;
1451 char hdr[128];
1453 map = map_sha1_file(sha1, &mapsize);
1454 if (!map)
1455 return error("unable to find %s", sha1_to_hex(sha1));
1456 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1457 status = error("unable to unpack %s header",
1458 sha1_to_hex(sha1));
1459 if (parse_sha1_header(hdr, type, &size) < 0)
1460 status = error("unable to parse %s header", sha1_to_hex(sha1));
1461 else {
1462 status = 0;
1463 if (sizep)
1464 *sizep = size;
1466 inflateEnd(&stream);
1467 munmap(map, mapsize);
1468 return status;
1471 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1473 struct pack_entry e;
1475 if (!find_pack_entry(sha1, &e, NULL)) {
1476 reprepare_packed_git();
1477 if (!find_pack_entry(sha1, &e, NULL))
1478 return sha1_loose_object_info(sha1, type, sizep);
1480 return packed_object_info(e.p, e.offset, type, sizep);
1483 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1485 struct pack_entry e;
1487 if (!find_pack_entry(sha1, &e, NULL))
1488 return NULL;
1489 else
1490 return unpack_entry(e.p, e.offset, type, size);
1493 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1495 unsigned long mapsize;
1496 void *map, *buf;
1498 buf = read_packed_sha1(sha1, type, size);
1499 if (buf)
1500 return buf;
1501 map = map_sha1_file(sha1, &mapsize);
1502 if (map) {
1503 buf = unpack_sha1_file(map, mapsize, type, size);
1504 munmap(map, mapsize);
1505 return buf;
1507 reprepare_packed_git();
1508 return read_packed_sha1(sha1, type, size);
1511 void *read_object_with_reference(const unsigned char *sha1,
1512 const char *required_type,
1513 unsigned long *size,
1514 unsigned char *actual_sha1_return)
1516 char type[20];
1517 void *buffer;
1518 unsigned long isize;
1519 unsigned char actual_sha1[20];
1521 hashcpy(actual_sha1, sha1);
1522 while (1) {
1523 int ref_length = -1;
1524 const char *ref_type = NULL;
1526 buffer = read_sha1_file(actual_sha1, type, &isize);
1527 if (!buffer)
1528 return NULL;
1529 if (!strcmp(type, required_type)) {
1530 *size = isize;
1531 if (actual_sha1_return)
1532 hashcpy(actual_sha1_return, actual_sha1);
1533 return buffer;
1535 /* Handle references */
1536 else if (!strcmp(type, commit_type))
1537 ref_type = "tree ";
1538 else if (!strcmp(type, tag_type))
1539 ref_type = "object ";
1540 else {
1541 free(buffer);
1542 return NULL;
1544 ref_length = strlen(ref_type);
1546 if (memcmp(buffer, ref_type, ref_length) ||
1547 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1548 free(buffer);
1549 return NULL;
1551 free(buffer);
1552 /* Now we have the ID of the referred-to object in
1553 * actual_sha1. Check again. */
1557 static void write_sha1_file_prepare(void *buf, unsigned long len,
1558 const char *type, unsigned char *sha1,
1559 unsigned char *hdr, int *hdrlen)
1561 SHA_CTX c;
1563 /* Generate the header */
1564 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1566 /* Sha1.. */
1567 SHA1_Init(&c);
1568 SHA1_Update(&c, hdr, *hdrlen);
1569 SHA1_Update(&c, buf, len);
1570 SHA1_Final(sha1, &c);
1574 * Link the tempfile to the final place, possibly creating the
1575 * last directory level as you do so.
1577 * Returns the errno on failure, 0 on success.
1579 static int link_temp_to_file(const char *tmpfile, const char *filename)
1581 int ret;
1582 char *dir;
1584 if (!link(tmpfile, filename))
1585 return 0;
1588 * Try to mkdir the last path component if that failed.
1590 * Re-try the "link()" regardless of whether the mkdir
1591 * succeeds, since a race might mean that somebody
1592 * else succeeded.
1594 ret = errno;
1595 dir = strrchr(filename, '/');
1596 if (dir) {
1597 *dir = 0;
1598 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1599 *dir = '/';
1600 return -2;
1602 *dir = '/';
1603 if (!link(tmpfile, filename))
1604 return 0;
1605 ret = errno;
1607 return ret;
1611 * Move the just written object into its final resting place
1613 int move_temp_to_file(const char *tmpfile, const char *filename)
1615 int ret = link_temp_to_file(tmpfile, filename);
1618 * Coda hack - coda doesn't like cross-directory links,
1619 * so we fall back to a rename, which will mean that it
1620 * won't be able to check collisions, but that's not a
1621 * big deal.
1623 * The same holds for FAT formatted media.
1625 * When this succeeds, we just return 0. We have nothing
1626 * left to unlink.
1628 if (ret && ret != EEXIST) {
1629 if (!rename(tmpfile, filename))
1630 return 0;
1631 ret = errno;
1633 unlink(tmpfile);
1634 if (ret) {
1635 if (ret != EEXIST) {
1636 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1638 /* FIXME!!! Collision check here ? */
1641 return 0;
1644 static int write_buffer(int fd, const void *buf, size_t len)
1646 if (write_in_full(fd, buf, len) < 0)
1647 return error("file write error (%s)", strerror(errno));
1648 return 0;
1651 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1653 int hdr_len;
1654 unsigned char c;
1656 c = (type << 4) | (len & 15);
1657 len >>= 4;
1658 hdr_len = 1;
1659 while (len) {
1660 *hdr++ = c | 0x80;
1661 hdr_len++;
1662 c = (len & 0x7f);
1663 len >>= 7;
1665 *hdr = c;
1666 return hdr_len;
1669 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1671 int obj_type, hdr;
1673 if (use_legacy_headers) {
1674 while (deflate(stream, 0) == Z_OK)
1675 /* nothing */;
1676 return;
1678 if (!strcmp(type, blob_type))
1679 obj_type = OBJ_BLOB;
1680 else if (!strcmp(type, tree_type))
1681 obj_type = OBJ_TREE;
1682 else if (!strcmp(type, commit_type))
1683 obj_type = OBJ_COMMIT;
1684 else if (!strcmp(type, tag_type))
1685 obj_type = OBJ_TAG;
1686 else
1687 die("trying to generate bogus object of type '%s'", type);
1688 hdr = write_binary_header(stream->next_out, obj_type, len);
1689 stream->total_out = hdr;
1690 stream->next_out += hdr;
1691 stream->avail_out -= hdr;
1694 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1695 unsigned char *sha1)
1697 unsigned char hdr[50];
1698 int hdrlen;
1699 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1700 return 0;
1703 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1705 int size;
1706 unsigned char *compressed;
1707 z_stream stream;
1708 unsigned char sha1[20];
1709 char *filename;
1710 static char tmpfile[PATH_MAX];
1711 unsigned char hdr[50];
1712 int fd, hdrlen;
1714 /* Normally if we have it in the pack then we do not bother writing
1715 * it out into .git/objects/??/?{38} file.
1717 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1718 filename = sha1_file_name(sha1);
1719 if (returnsha1)
1720 hashcpy(returnsha1, sha1);
1721 if (has_sha1_file(sha1))
1722 return 0;
1723 fd = open(filename, O_RDONLY);
1724 if (fd >= 0) {
1726 * FIXME!!! We might do collision checking here, but we'd
1727 * need to uncompress the old file and check it. Later.
1729 close(fd);
1730 return 0;
1733 if (errno != ENOENT) {
1734 return error("sha1 file %s: %s\n", filename, strerror(errno));
1737 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1739 fd = mkstemp(tmpfile);
1740 if (fd < 0) {
1741 if (errno == EPERM)
1742 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1743 else
1744 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1747 /* Set it up */
1748 memset(&stream, 0, sizeof(stream));
1749 deflateInit(&stream, zlib_compression_level);
1750 size = 8 + deflateBound(&stream, len+hdrlen);
1751 compressed = xmalloc(size);
1753 /* Compress it */
1754 stream.next_out = compressed;
1755 stream.avail_out = size;
1757 /* First header.. */
1758 stream.next_in = hdr;
1759 stream.avail_in = hdrlen;
1760 setup_object_header(&stream, type, len);
1762 /* Then the data itself.. */
1763 stream.next_in = buf;
1764 stream.avail_in = len;
1765 while (deflate(&stream, Z_FINISH) == Z_OK)
1766 /* nothing */;
1767 deflateEnd(&stream);
1768 size = stream.total_out;
1770 if (write_buffer(fd, compressed, size) < 0)
1771 die("unable to write sha1 file");
1772 fchmod(fd, 0444);
1773 close(fd);
1774 free(compressed);
1776 return move_temp_to_file(tmpfile, filename);
1780 * We need to unpack and recompress the object for writing
1781 * it out to a different file.
1783 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1785 size_t size;
1786 z_stream stream;
1787 unsigned char *unpacked;
1788 unsigned long len;
1789 char type[20];
1790 char hdr[50];
1791 int hdrlen;
1792 void *buf;
1794 /* need to unpack and recompress it by itself */
1795 unpacked = read_packed_sha1(sha1, type, &len);
1796 if (!unpacked)
1797 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1799 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1801 /* Set it up */
1802 memset(&stream, 0, sizeof(stream));
1803 deflateInit(&stream, zlib_compression_level);
1804 size = deflateBound(&stream, len + hdrlen);
1805 buf = xmalloc(size);
1807 /* Compress it */
1808 stream.next_out = buf;
1809 stream.avail_out = size;
1811 /* First header.. */
1812 stream.next_in = (void *)hdr;
1813 stream.avail_in = hdrlen;
1814 while (deflate(&stream, 0) == Z_OK)
1815 /* nothing */;
1817 /* Then the data itself.. */
1818 stream.next_in = unpacked;
1819 stream.avail_in = len;
1820 while (deflate(&stream, Z_FINISH) == Z_OK)
1821 /* nothing */;
1822 deflateEnd(&stream);
1823 free(unpacked);
1825 *objsize = stream.total_out;
1826 return buf;
1829 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1831 int retval;
1832 unsigned long objsize;
1833 void *buf = map_sha1_file(sha1, &objsize);
1835 if (buf) {
1836 retval = write_buffer(fd, buf, objsize);
1837 munmap(buf, objsize);
1838 return retval;
1841 buf = repack_object(sha1, &objsize);
1842 retval = write_buffer(fd, buf, objsize);
1843 free(buf);
1844 return retval;
1847 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1848 size_t bufsize, size_t *bufposn)
1850 char tmpfile[PATH_MAX];
1851 int local;
1852 z_stream stream;
1853 unsigned char real_sha1[20];
1854 unsigned char discard[4096];
1855 int ret;
1856 SHA_CTX c;
1858 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1860 local = mkstemp(tmpfile);
1861 if (local < 0) {
1862 if (errno == EPERM)
1863 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1864 else
1865 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1868 memset(&stream, 0, sizeof(stream));
1870 inflateInit(&stream);
1872 SHA1_Init(&c);
1874 do {
1875 ssize_t size;
1876 if (*bufposn) {
1877 stream.avail_in = *bufposn;
1878 stream.next_in = (unsigned char *) buffer;
1879 do {
1880 stream.next_out = discard;
1881 stream.avail_out = sizeof(discard);
1882 ret = inflate(&stream, Z_SYNC_FLUSH);
1883 SHA1_Update(&c, discard, sizeof(discard) -
1884 stream.avail_out);
1885 } while (stream.avail_in && ret == Z_OK);
1886 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1887 die("unable to write sha1 file");
1888 memmove(buffer, buffer + *bufposn - stream.avail_in,
1889 stream.avail_in);
1890 *bufposn = stream.avail_in;
1891 if (ret != Z_OK)
1892 break;
1894 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1895 if (size <= 0) {
1896 close(local);
1897 unlink(tmpfile);
1898 if (!size)
1899 return error("Connection closed?");
1900 perror("Reading from connection");
1901 return -1;
1903 *bufposn += size;
1904 } while (1);
1905 inflateEnd(&stream);
1907 close(local);
1908 SHA1_Final(real_sha1, &c);
1909 if (ret != Z_STREAM_END) {
1910 unlink(tmpfile);
1911 return error("File %s corrupted", sha1_to_hex(sha1));
1913 if (hashcmp(sha1, real_sha1)) {
1914 unlink(tmpfile);
1915 return error("File %s has bad hash", sha1_to_hex(sha1));
1918 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1921 int has_pack_index(const unsigned char *sha1)
1923 struct stat st;
1924 if (stat(sha1_pack_index_name(sha1), &st))
1925 return 0;
1926 return 1;
1929 int has_pack_file(const unsigned char *sha1)
1931 struct stat st;
1932 if (stat(sha1_pack_name(sha1), &st))
1933 return 0;
1934 return 1;
1937 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1939 struct pack_entry e;
1940 return find_pack_entry(sha1, &e, ignore_packed);
1943 int has_sha1_file(const unsigned char *sha1)
1945 struct stat st;
1946 struct pack_entry e;
1948 if (find_pack_entry(sha1, &e, NULL))
1949 return 1;
1950 return find_sha1_file(sha1, &st) ? 1 : 0;
1954 * reads from fd as long as possible into a supplied buffer of size bytes.
1955 * If necessary the buffer's size is increased using realloc()
1957 * returns 0 if anything went fine and -1 otherwise
1959 * NOTE: both buf and size may change, but even when -1 is returned
1960 * you still have to free() it yourself.
1962 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1964 char* buf = *return_buf;
1965 unsigned long size = *return_size;
1966 int iret;
1967 unsigned long off = 0;
1969 do {
1970 iret = xread(fd, buf + off, size - off);
1971 if (iret > 0) {
1972 off += iret;
1973 if (off == size) {
1974 size *= 2;
1975 buf = xrealloc(buf, size);
1978 } while (iret > 0);
1980 *return_buf = buf;
1981 *return_size = off;
1983 if (iret < 0)
1984 return -1;
1985 return 0;
1988 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1990 unsigned long size = 4096;
1991 char *buf = xmalloc(size);
1992 int ret;
1994 if (read_pipe(fd, &buf, &size)) {
1995 free(buf);
1996 return -1;
1999 if (!type)
2000 type = blob_type;
2001 if (write_object)
2002 ret = write_sha1_file(buf, size, type, sha1);
2003 else
2004 ret = hash_sha1_file(buf, size, type, sha1);
2005 free(buf);
2006 return ret;
2009 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
2011 unsigned long size = st->st_size;
2012 void *buf;
2013 int ret;
2015 buf = "";
2016 if (size)
2017 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2018 close(fd);
2020 if (!type)
2021 type = blob_type;
2022 if (write_object)
2023 ret = write_sha1_file(buf, size, type, sha1);
2024 else
2025 ret = hash_sha1_file(buf, size, type, sha1);
2026 if (size)
2027 munmap(buf, size);
2028 return ret;
2031 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2033 int fd;
2034 char *target;
2036 switch (st->st_mode & S_IFMT) {
2037 case S_IFREG:
2038 fd = open(path, O_RDONLY);
2039 if (fd < 0)
2040 return error("open(\"%s\"): %s", path,
2041 strerror(errno));
2042 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2043 return error("%s: failed to insert into database",
2044 path);
2045 break;
2046 case S_IFLNK:
2047 target = xmalloc(st->st_size+1);
2048 if (readlink(path, target, st->st_size+1) != st->st_size) {
2049 char *errstr = strerror(errno);
2050 free(target);
2051 return error("readlink(\"%s\"): %s", path,
2052 errstr);
2054 if (!write_object)
2055 hash_sha1_file(target, st->st_size, blob_type, sha1);
2056 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2057 return error("%s: failed to insert into database",
2058 path);
2059 free(target);
2060 break;
2061 default:
2062 return error("%s: unsupported file type", path);
2064 return 0;
2067 int read_pack_header(int fd, struct pack_header *header)
2069 char *c = (char*)header;
2070 ssize_t remaining = sizeof(struct pack_header);
2071 do {
2072 ssize_t r = xread(fd, c, remaining);
2073 if (r <= 0)
2074 /* "eof before pack header was fully read" */
2075 return PH_ERROR_EOF;
2076 remaining -= r;
2077 c += r;
2078 } while (remaining > 0);
2079 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2080 /* "protocol error (pack signature mismatch detected)" */
2081 return PH_ERROR_PACK_SIGNATURE;
2082 if (!pack_version_ok(header->hdr_version))
2083 /* "protocol error (pack version unsupported)" */
2084 return PH_ERROR_PROTOCOL;
2085 return 0;