Cleanup prepare_packed_git_one to reuse install_packed_git.
[git/gitweb.git] / sha1_file.c
bloba42f94ac94fac2186cbcb0eed395ea3befd8fced
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 void 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 die("packfile %s cannot be opened", p->pack_name);
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 die("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 die("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 die("cannot determine file descriptor flags");
581 fd_flag |= FD_CLOEXEC;
582 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
583 die("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 die("file %s is far too short to be a packfile", p->pack_name);
588 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
589 die("file %s is not a GIT packfile", p->pack_name);
590 if (!pack_version_ok(hdr.hdr_version))
591 die("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 die("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 die("end of packfile %s is unavailable", p->pack_name);
603 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
604 die("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 die("packfile %s does not match index", p->pack_name);
610 static int in_window(struct pack_window *win, unsigned long offset)
612 /* We must promise at least 20 bytes (one hash) after the
613 * offset is available from this window, otherwise the offset
614 * is not actually in this window and a different window (which
615 * has that one hash excess) must be used. This is to support
616 * the object header and delta base parsing routines below.
618 off_t win_off = win->offset;
619 return win_off <= offset
620 && (offset + 20) <= (win_off + win->len);
623 unsigned char* use_pack(struct packed_git *p,
624 struct pack_window **w_cursor,
625 unsigned long offset,
626 unsigned int *left)
628 struct pack_window *win = *w_cursor;
630 if (p->pack_fd == -1)
631 open_packed_git(p);
633 /* Since packfiles end in a hash of their content and its
634 * pointless to ask for an offset into the middle of that
635 * hash, and the in_window function above wouldn't match
636 * don't allow an offset too close to the end of the file.
638 if (offset > (p->pack_size - 20))
639 die("offset beyond end of packfile (truncated pack?)");
641 if (!win || !in_window(win, offset)) {
642 if (win)
643 win->inuse_cnt--;
644 for (win = p->windows; win; win = win->next) {
645 if (in_window(win, offset))
646 break;
648 if (!win) {
649 if (!page_size)
650 page_size = getpagesize();
651 win = xcalloc(1, sizeof(*win));
652 win->offset = (offset / page_size) * page_size;
653 win->len = p->pack_size - win->offset;
654 if (win->len > packed_git_window_size)
655 win->len = packed_git_window_size;
656 pack_mapped += win->len;
657 while (packed_git_limit < pack_mapped
658 && unuse_one_window(p))
659 ; /* nothing */
660 win->base = xmmap(NULL, win->len,
661 PROT_READ, MAP_PRIVATE,
662 p->pack_fd, win->offset);
663 if (win->base == MAP_FAILED)
664 die("packfile %s cannot be mapped: %s",
665 p->pack_name,
666 strerror(errno));
667 pack_mmap_calls++;
668 pack_open_windows++;
669 if (pack_mapped > peak_pack_mapped)
670 peak_pack_mapped = pack_mapped;
671 if (pack_open_windows > peak_pack_open_windows)
672 peak_pack_open_windows = pack_open_windows;
673 win->next = p->windows;
674 p->windows = win;
677 if (win != *w_cursor) {
678 win->last_used = pack_used_ctr++;
679 win->inuse_cnt++;
680 *w_cursor = win;
682 offset -= win->offset;
683 if (left)
684 *left = win->len - offset;
685 return win->base + offset;
688 struct packed_git *add_packed_git(char *path, int path_len, int local)
690 struct stat st;
691 struct packed_git *p;
692 unsigned long idx_size;
693 void *idx_map;
694 unsigned char sha1[20];
696 if (check_packed_git_idx(path, &idx_size, &idx_map))
697 return NULL;
699 /* do we have a corresponding .pack file? */
700 strcpy(path + path_len - 4, ".pack");
701 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
702 munmap(idx_map, idx_size);
703 return NULL;
705 /* ok, it looks sane as far as we can check without
706 * actually mapping the pack file.
708 p = xmalloc(sizeof(*p) + path_len + 2);
709 strcpy(p->pack_name, path);
710 p->index_size = idx_size;
711 p->pack_size = st.st_size;
712 p->index_base = idx_map;
713 p->next = NULL;
714 p->windows = NULL;
715 p->pack_fd = -1;
716 p->pack_local = local;
717 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
718 hashcpy(p->sha1, sha1);
719 return p;
722 struct packed_git *parse_pack_index(unsigned char *sha1)
724 char *path = sha1_pack_index_name(sha1);
725 return parse_pack_index_file(sha1, path);
728 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
730 struct packed_git *p;
731 unsigned long idx_size;
732 void *idx_map;
733 char *path;
735 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
736 return NULL;
738 path = sha1_pack_name(sha1);
740 p = xmalloc(sizeof(*p) + strlen(path) + 2);
741 strcpy(p->pack_name, path);
742 p->index_size = idx_size;
743 p->pack_size = 0;
744 p->index_base = idx_map;
745 p->next = NULL;
746 p->windows = NULL;
747 p->pack_fd = -1;
748 hashcpy(p->sha1, sha1);
749 return p;
752 void install_packed_git(struct packed_git *pack)
754 pack->next = packed_git;
755 packed_git = pack;
758 static void prepare_packed_git_one(char *objdir, int local)
760 char path[PATH_MAX];
761 int len;
762 DIR *dir;
763 struct dirent *de;
765 sprintf(path, "%s/pack", objdir);
766 len = strlen(path);
767 dir = opendir(path);
768 if (!dir) {
769 if (errno != ENOENT)
770 error("unable to open object pack directory: %s: %s",
771 path, strerror(errno));
772 return;
774 path[len++] = '/';
775 while ((de = readdir(dir)) != NULL) {
776 int namelen = strlen(de->d_name);
777 struct packed_git *p;
779 if (!has_extension(de->d_name, ".idx"))
780 continue;
782 /* we have .idx. Is it a file we can map? */
783 strcpy(path + len, de->d_name);
784 for (p = packed_git; p; p = p->next) {
785 if (!memcmp(path, p->pack_name, len + namelen - 4))
786 break;
788 if (p)
789 continue;
790 p = add_packed_git(path, len + namelen, local);
791 if (!p)
792 continue;
793 install_packed_git(p);
795 closedir(dir);
798 static int prepare_packed_git_run_once = 0;
799 void prepare_packed_git(void)
801 struct alternate_object_database *alt;
803 if (prepare_packed_git_run_once)
804 return;
805 prepare_packed_git_one(get_object_directory(), 1);
806 prepare_alt_odb();
807 for (alt = alt_odb_list; alt; alt = alt->next) {
808 alt->name[-1] = 0;
809 prepare_packed_git_one(alt->base, 0);
810 alt->name[-1] = '/';
812 prepare_packed_git_run_once = 1;
815 void reprepare_packed_git(void)
817 prepare_packed_git_run_once = 0;
818 prepare_packed_git();
821 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
823 unsigned char real_sha1[20];
824 hash_sha1_file(map, size, type, real_sha1);
825 return hashcmp(sha1, real_sha1) ? -1 : 0;
828 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
830 struct stat st;
831 void *map;
832 int fd;
833 char *filename = find_sha1_file(sha1, &st);
835 if (!filename) {
836 return NULL;
839 fd = open(filename, O_RDONLY | sha1_file_open_flag);
840 if (fd < 0) {
841 /* See if it works without O_NOATIME */
842 switch (sha1_file_open_flag) {
843 default:
844 fd = open(filename, O_RDONLY);
845 if (fd >= 0)
846 break;
847 /* Fallthrough */
848 case 0:
849 return NULL;
852 /* If it failed once, it will probably fail again.
853 * Stop using O_NOATIME
855 sha1_file_open_flag = 0;
857 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
858 close(fd);
859 *size = st.st_size;
860 return map;
863 int legacy_loose_object(unsigned char *map)
865 unsigned int word;
868 * Is it a zlib-compressed buffer? If so, the first byte
869 * must be 0x78 (15-bit window size, deflated), and the
870 * first 16-bit word is evenly divisible by 31
872 word = (map[0] << 8) + map[1];
873 if (map[0] == 0x78 && !(word % 31))
874 return 1;
875 else
876 return 0;
879 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
881 unsigned shift;
882 unsigned char c;
883 unsigned long size;
884 unsigned long used = 0;
886 c = buf[used++];
887 *type = (c >> 4) & 7;
888 size = c & 15;
889 shift = 4;
890 while (c & 0x80) {
891 if (len <= used)
892 return 0;
893 if (sizeof(long) * 8 <= shift)
894 return 0;
895 c = buf[used++];
896 size += (c & 0x7f) << shift;
897 shift += 7;
899 *sizep = size;
900 return used;
903 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
905 unsigned long size, used;
906 static const char valid_loose_object_type[8] = {
907 0, /* OBJ_EXT */
908 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
909 0, /* "delta" and others are invalid in a loose object */
911 enum object_type type;
913 /* Get the data stream */
914 memset(stream, 0, sizeof(*stream));
915 stream->next_in = map;
916 stream->avail_in = mapsize;
917 stream->next_out = buffer;
918 stream->avail_out = bufsiz;
920 if (legacy_loose_object(map)) {
921 inflateInit(stream);
922 return inflate(stream, 0);
925 used = unpack_object_header_gently(map, mapsize, &type, &size);
926 if (!used || !valid_loose_object_type[type])
927 return -1;
928 map += used;
929 mapsize -= used;
931 /* Set up the stream for the rest.. */
932 stream->next_in = map;
933 stream->avail_in = mapsize;
934 inflateInit(stream);
936 /* And generate the fake traditional header */
937 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
938 type_names[type], size);
939 return 0;
942 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
944 int bytes = strlen(buffer) + 1;
945 unsigned char *buf = xmalloc(1+size);
946 unsigned long n;
948 n = stream->total_out - bytes;
949 if (n > size)
950 n = size;
951 memcpy(buf, (char *) buffer + bytes, n);
952 bytes = n;
953 if (bytes < size) {
954 stream->next_out = buf + bytes;
955 stream->avail_out = size - bytes;
956 while (inflate(stream, Z_FINISH) == Z_OK)
957 /* nothing */;
959 buf[size] = 0;
960 inflateEnd(stream);
961 return buf;
965 * We used to just use "sscanf()", but that's actually way
966 * too permissive for what we want to check. So do an anal
967 * object header parse by hand.
969 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
971 int i;
972 unsigned long size;
975 * The type can be at most ten bytes (including the
976 * terminating '\0' that we add), and is followed by
977 * a space.
979 i = 10;
980 for (;;) {
981 char c = *hdr++;
982 if (c == ' ')
983 break;
984 if (!--i)
985 return -1;
986 *type++ = c;
988 *type = 0;
991 * The length must follow immediately, and be in canonical
992 * decimal format (ie "010" is not valid).
994 size = *hdr++ - '0';
995 if (size > 9)
996 return -1;
997 if (size) {
998 for (;;) {
999 unsigned long c = *hdr - '0';
1000 if (c > 9)
1001 break;
1002 hdr++;
1003 size = size * 10 + c;
1006 *sizep = size;
1009 * The length must be followed by a zero byte
1011 return *hdr ? -1 : 0;
1014 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1016 int ret;
1017 z_stream stream;
1018 char hdr[8192];
1020 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1021 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1022 return NULL;
1024 return unpack_sha1_rest(&stream, hdr, *size);
1027 static unsigned long get_delta_base(struct packed_git *p,
1028 struct pack_window **w_curs,
1029 unsigned long offset,
1030 enum object_type kind,
1031 unsigned long delta_obj_offset,
1032 unsigned long *base_obj_offset)
1034 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1035 unsigned long base_offset;
1037 /* use_pack() assured us we have [base_info, base_info + 20)
1038 * as a range that we can look at without walking off the
1039 * end of the mapped window. Its actually the hash size
1040 * that is assured. An OFS_DELTA longer than the hash size
1041 * is stupid, as then a REF_DELTA would be smaller to store.
1043 if (kind == OBJ_OFS_DELTA) {
1044 unsigned used = 0;
1045 unsigned char c = base_info[used++];
1046 base_offset = c & 127;
1047 while (c & 128) {
1048 base_offset += 1;
1049 if (!base_offset || base_offset & ~(~0UL >> 7))
1050 die("offset value overflow for delta base object");
1051 c = base_info[used++];
1052 base_offset = (base_offset << 7) + (c & 127);
1054 base_offset = delta_obj_offset - base_offset;
1055 if (base_offset >= delta_obj_offset)
1056 die("delta base offset out of bound");
1057 offset += used;
1058 } else if (kind == OBJ_REF_DELTA) {
1059 /* The base entry _must_ be in the same pack */
1060 base_offset = find_pack_entry_one(base_info, p);
1061 if (!base_offset)
1062 die("failed to find delta-pack base object %s",
1063 sha1_to_hex(base_info));
1064 offset += 20;
1065 } else
1066 die("I am totally screwed");
1067 *base_obj_offset = base_offset;
1068 return offset;
1071 /* forward declaration for a mutually recursive function */
1072 static int packed_object_info(struct packed_git *p, unsigned long offset,
1073 char *type, unsigned long *sizep);
1075 static int packed_delta_info(struct packed_git *p,
1076 struct pack_window **w_curs,
1077 unsigned long offset,
1078 enum object_type kind,
1079 unsigned long obj_offset,
1080 char *type,
1081 unsigned long *sizep)
1083 unsigned long base_offset;
1085 offset = get_delta_base(p, w_curs, offset, kind,
1086 obj_offset, &base_offset);
1088 /* We choose to only get the type of the base object and
1089 * ignore potentially corrupt pack file that expects the delta
1090 * based on a base with a wrong size. This saves tons of
1091 * inflate() calls.
1093 if (packed_object_info(p, base_offset, type, NULL))
1094 die("cannot get info for delta-pack base");
1096 if (sizep) {
1097 const unsigned char *data;
1098 unsigned char delta_head[20], *in;
1099 unsigned long result_size;
1100 z_stream stream;
1101 int st;
1103 memset(&stream, 0, sizeof(stream));
1104 stream.next_out = delta_head;
1105 stream.avail_out = sizeof(delta_head);
1107 inflateInit(&stream);
1108 do {
1109 in = use_pack(p, w_curs, offset, &stream.avail_in);
1110 stream.next_in = in;
1111 st = inflate(&stream, Z_FINISH);
1112 offset += stream.next_in - in;
1113 } while ((st == Z_OK || st == Z_BUF_ERROR)
1114 && stream.total_out < sizeof(delta_head));
1115 inflateEnd(&stream);
1116 if ((st != Z_STREAM_END) &&
1117 stream.total_out != sizeof(delta_head))
1118 die("delta data unpack-initial failed");
1120 /* Examine the initial part of the delta to figure out
1121 * the result size.
1123 data = delta_head;
1125 /* ignore base size */
1126 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1128 /* Read the result size */
1129 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1130 *sizep = result_size;
1132 return 0;
1135 static unsigned long unpack_object_header(struct packed_git *p,
1136 struct pack_window **w_curs,
1137 unsigned long offset,
1138 enum object_type *type,
1139 unsigned long *sizep)
1141 unsigned char *base;
1142 unsigned int left;
1143 unsigned long used;
1145 /* use_pack() assures us we have [base, base + 20) available
1146 * as a range that we can look at at. (Its actually the hash
1147 * size that is assurred.) With our object header encoding
1148 * the maximum deflated object size is 2^137, which is just
1149 * insane, so we know won't exceed what we have been given.
1151 base = use_pack(p, w_curs, offset, &left);
1152 used = unpack_object_header_gently(base, left, type, sizep);
1153 if (!used)
1154 die("object offset outside of pack file");
1156 return offset + used;
1159 void packed_object_info_detail(struct packed_git *p,
1160 unsigned long offset,
1161 char *type,
1162 unsigned long *size,
1163 unsigned long *store_size,
1164 unsigned int *delta_chain_length,
1165 unsigned char *base_sha1)
1167 struct pack_window *w_curs = NULL;
1168 unsigned long obj_offset, val;
1169 unsigned char *next_sha1;
1170 enum object_type kind;
1172 *delta_chain_length = 0;
1173 obj_offset = offset;
1174 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1176 for (;;) {
1177 switch (kind) {
1178 default:
1179 die("pack %s contains unknown object type %d",
1180 p->pack_name, kind);
1181 case OBJ_COMMIT:
1182 case OBJ_TREE:
1183 case OBJ_BLOB:
1184 case OBJ_TAG:
1185 strcpy(type, type_names[kind]);
1186 *store_size = 0; /* notyet */
1187 unuse_pack(&w_curs);
1188 return;
1189 case OBJ_OFS_DELTA:
1190 get_delta_base(p, &w_curs, offset, kind,
1191 obj_offset, &offset);
1192 if (*delta_chain_length == 0) {
1193 /* TODO: find base_sha1 as pointed by offset */
1195 break;
1196 case OBJ_REF_DELTA:
1197 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1198 if (*delta_chain_length == 0)
1199 hashcpy(base_sha1, next_sha1);
1200 offset = find_pack_entry_one(next_sha1, p);
1201 break;
1203 obj_offset = offset;
1204 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1205 (*delta_chain_length)++;
1209 static int packed_object_info(struct packed_git *p, unsigned long offset,
1210 char *type, unsigned long *sizep)
1212 struct pack_window *w_curs = NULL;
1213 unsigned long size, obj_offset = offset;
1214 enum object_type kind;
1215 int r;
1217 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1219 switch (kind) {
1220 case OBJ_OFS_DELTA:
1221 case OBJ_REF_DELTA:
1222 r = packed_delta_info(p, &w_curs, offset, kind,
1223 obj_offset, type, sizep);
1224 unuse_pack(&w_curs);
1225 return r;
1226 case OBJ_COMMIT:
1227 case OBJ_TREE:
1228 case OBJ_BLOB:
1229 case OBJ_TAG:
1230 strcpy(type, type_names[kind]);
1231 unuse_pack(&w_curs);
1232 break;
1233 default:
1234 die("pack %s contains unknown object type %d",
1235 p->pack_name, kind);
1237 if (sizep)
1238 *sizep = size;
1239 return 0;
1242 static void *unpack_compressed_entry(struct packed_git *p,
1243 struct pack_window **w_curs,
1244 unsigned long offset,
1245 unsigned long size)
1247 int st;
1248 z_stream stream;
1249 unsigned char *buffer, *in;
1251 buffer = xmalloc(size + 1);
1252 buffer[size] = 0;
1253 memset(&stream, 0, sizeof(stream));
1254 stream.next_out = buffer;
1255 stream.avail_out = size;
1257 inflateInit(&stream);
1258 do {
1259 in = use_pack(p, w_curs, offset, &stream.avail_in);
1260 stream.next_in = in;
1261 st = inflate(&stream, Z_FINISH);
1262 offset += stream.next_in - in;
1263 } while (st == Z_OK || st == Z_BUF_ERROR);
1264 inflateEnd(&stream);
1265 if ((st != Z_STREAM_END) || stream.total_out != size) {
1266 free(buffer);
1267 return NULL;
1270 return buffer;
1273 static void *unpack_delta_entry(struct packed_git *p,
1274 struct pack_window **w_curs,
1275 unsigned long offset,
1276 unsigned long delta_size,
1277 enum object_type kind,
1278 unsigned long obj_offset,
1279 char *type,
1280 unsigned long *sizep)
1282 void *delta_data, *result, *base;
1283 unsigned long result_size, base_size, base_offset;
1285 offset = get_delta_base(p, w_curs, offset, kind,
1286 obj_offset, &base_offset);
1287 base = unpack_entry(p, base_offset, type, &base_size);
1288 if (!base)
1289 die("failed to read delta base object at %lu from %s",
1290 base_offset, p->pack_name);
1292 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1293 result = patch_delta(base, base_size,
1294 delta_data, delta_size,
1295 &result_size);
1296 if (!result)
1297 die("failed to apply delta");
1298 free(delta_data);
1299 free(base);
1300 *sizep = result_size;
1301 return result;
1304 void *unpack_entry(struct packed_git *p, unsigned long offset,
1305 char *type, unsigned long *sizep)
1307 struct pack_window *w_curs = NULL;
1308 unsigned long size, obj_offset = offset;
1309 enum object_type kind;
1310 void *retval;
1312 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1313 switch (kind) {
1314 case OBJ_OFS_DELTA:
1315 case OBJ_REF_DELTA:
1316 retval = unpack_delta_entry(p, &w_curs, offset, size,
1317 kind, obj_offset, type, sizep);
1318 break;
1319 case OBJ_COMMIT:
1320 case OBJ_TREE:
1321 case OBJ_BLOB:
1322 case OBJ_TAG:
1323 strcpy(type, type_names[kind]);
1324 *sizep = size;
1325 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1326 break;
1327 default:
1328 die("unknown object type %i in %s", kind, p->pack_name);
1330 unuse_pack(&w_curs);
1331 return retval;
1334 int num_packed_objects(const struct packed_git *p)
1336 /* See check_packed_git_idx() */
1337 return (p->index_size - 20 - 20 - 4*256) / 24;
1340 int nth_packed_object_sha1(const struct packed_git *p, int n,
1341 unsigned char* sha1)
1343 void *index = p->index_base + 256;
1344 if (n < 0 || num_packed_objects(p) <= n)
1345 return -1;
1346 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1347 return 0;
1350 unsigned long find_pack_entry_one(const unsigned char *sha1,
1351 struct packed_git *p)
1353 uint32_t *level1_ofs = p->index_base;
1354 int hi = ntohl(level1_ofs[*sha1]);
1355 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1356 void *index = p->index_base + 256;
1358 do {
1359 int mi = (lo + hi) / 2;
1360 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1361 if (!cmp)
1362 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1363 if (cmp > 0)
1364 hi = mi;
1365 else
1366 lo = mi+1;
1367 } while (lo < hi);
1368 return 0;
1371 static int matches_pack_name(struct packed_git *p, const char *ig)
1373 const char *last_c, *c;
1375 if (!strcmp(p->pack_name, ig))
1376 return 0;
1378 for (c = p->pack_name, last_c = c; *c;)
1379 if (*c == '/')
1380 last_c = ++c;
1381 else
1382 ++c;
1383 if (!strcmp(last_c, ig))
1384 return 0;
1386 return 1;
1389 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1391 struct packed_git *p;
1392 unsigned long offset;
1394 prepare_packed_git();
1396 for (p = packed_git; p; p = p->next) {
1397 if (ignore_packed) {
1398 const char **ig;
1399 for (ig = ignore_packed; *ig; ig++)
1400 if (!matches_pack_name(p, *ig))
1401 break;
1402 if (*ig)
1403 continue;
1405 offset = find_pack_entry_one(sha1, p);
1406 if (offset) {
1407 e->offset = offset;
1408 e->p = p;
1409 hashcpy(e->sha1, sha1);
1410 return 1;
1413 return 0;
1416 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1417 struct packed_git *packs)
1419 struct packed_git *p;
1421 for (p = packs; p; p = p->next) {
1422 if (find_pack_entry_one(sha1, p))
1423 return p;
1425 return NULL;
1429 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1431 int status;
1432 unsigned long mapsize, size;
1433 void *map;
1434 z_stream stream;
1435 char hdr[128];
1437 map = map_sha1_file(sha1, &mapsize);
1438 if (!map)
1439 return error("unable to find %s", sha1_to_hex(sha1));
1440 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1441 status = error("unable to unpack %s header",
1442 sha1_to_hex(sha1));
1443 if (parse_sha1_header(hdr, type, &size) < 0)
1444 status = error("unable to parse %s header", sha1_to_hex(sha1));
1445 else {
1446 status = 0;
1447 if (sizep)
1448 *sizep = size;
1450 inflateEnd(&stream);
1451 munmap(map, mapsize);
1452 return status;
1455 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1457 struct pack_entry e;
1459 if (!find_pack_entry(sha1, &e, NULL)) {
1460 reprepare_packed_git();
1461 if (!find_pack_entry(sha1, &e, NULL))
1462 return sha1_loose_object_info(sha1, type, sizep);
1464 return packed_object_info(e.p, e.offset, type, sizep);
1467 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1469 struct pack_entry e;
1471 if (!find_pack_entry(sha1, &e, NULL))
1472 return NULL;
1473 else
1474 return unpack_entry(e.p, e.offset, type, size);
1477 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1479 unsigned long mapsize;
1480 void *map, *buf;
1482 buf = read_packed_sha1(sha1, type, size);
1483 if (buf)
1484 return buf;
1485 map = map_sha1_file(sha1, &mapsize);
1486 if (map) {
1487 buf = unpack_sha1_file(map, mapsize, type, size);
1488 munmap(map, mapsize);
1489 return buf;
1491 reprepare_packed_git();
1492 return read_packed_sha1(sha1, type, size);
1495 void *read_object_with_reference(const unsigned char *sha1,
1496 const char *required_type,
1497 unsigned long *size,
1498 unsigned char *actual_sha1_return)
1500 char type[20];
1501 void *buffer;
1502 unsigned long isize;
1503 unsigned char actual_sha1[20];
1505 hashcpy(actual_sha1, sha1);
1506 while (1) {
1507 int ref_length = -1;
1508 const char *ref_type = NULL;
1510 buffer = read_sha1_file(actual_sha1, type, &isize);
1511 if (!buffer)
1512 return NULL;
1513 if (!strcmp(type, required_type)) {
1514 *size = isize;
1515 if (actual_sha1_return)
1516 hashcpy(actual_sha1_return, actual_sha1);
1517 return buffer;
1519 /* Handle references */
1520 else if (!strcmp(type, commit_type))
1521 ref_type = "tree ";
1522 else if (!strcmp(type, tag_type))
1523 ref_type = "object ";
1524 else {
1525 free(buffer);
1526 return NULL;
1528 ref_length = strlen(ref_type);
1530 if (memcmp(buffer, ref_type, ref_length) ||
1531 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1532 free(buffer);
1533 return NULL;
1535 free(buffer);
1536 /* Now we have the ID of the referred-to object in
1537 * actual_sha1. Check again. */
1541 static void write_sha1_file_prepare(void *buf, unsigned long len,
1542 const char *type, unsigned char *sha1,
1543 unsigned char *hdr, int *hdrlen)
1545 SHA_CTX c;
1547 /* Generate the header */
1548 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1550 /* Sha1.. */
1551 SHA1_Init(&c);
1552 SHA1_Update(&c, hdr, *hdrlen);
1553 SHA1_Update(&c, buf, len);
1554 SHA1_Final(sha1, &c);
1558 * Link the tempfile to the final place, possibly creating the
1559 * last directory level as you do so.
1561 * Returns the errno on failure, 0 on success.
1563 static int link_temp_to_file(const char *tmpfile, const char *filename)
1565 int ret;
1566 char *dir;
1568 if (!link(tmpfile, filename))
1569 return 0;
1572 * Try to mkdir the last path component if that failed.
1574 * Re-try the "link()" regardless of whether the mkdir
1575 * succeeds, since a race might mean that somebody
1576 * else succeeded.
1578 ret = errno;
1579 dir = strrchr(filename, '/');
1580 if (dir) {
1581 *dir = 0;
1582 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1583 *dir = '/';
1584 return -2;
1586 *dir = '/';
1587 if (!link(tmpfile, filename))
1588 return 0;
1589 ret = errno;
1591 return ret;
1595 * Move the just written object into its final resting place
1597 int move_temp_to_file(const char *tmpfile, const char *filename)
1599 int ret = link_temp_to_file(tmpfile, filename);
1602 * Coda hack - coda doesn't like cross-directory links,
1603 * so we fall back to a rename, which will mean that it
1604 * won't be able to check collisions, but that's not a
1605 * big deal.
1607 * The same holds for FAT formatted media.
1609 * When this succeeds, we just return 0. We have nothing
1610 * left to unlink.
1612 if (ret && ret != EEXIST) {
1613 if (!rename(tmpfile, filename))
1614 return 0;
1615 ret = errno;
1617 unlink(tmpfile);
1618 if (ret) {
1619 if (ret != EEXIST) {
1620 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1622 /* FIXME!!! Collision check here ? */
1625 return 0;
1628 static int write_buffer(int fd, const void *buf, size_t len)
1630 if (write_in_full(fd, buf, len) < 0)
1631 return error("file write error (%s)", strerror(errno));
1632 return 0;
1635 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1637 int hdr_len;
1638 unsigned char c;
1640 c = (type << 4) | (len & 15);
1641 len >>= 4;
1642 hdr_len = 1;
1643 while (len) {
1644 *hdr++ = c | 0x80;
1645 hdr_len++;
1646 c = (len & 0x7f);
1647 len >>= 7;
1649 *hdr = c;
1650 return hdr_len;
1653 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1655 int obj_type, hdr;
1657 if (use_legacy_headers) {
1658 while (deflate(stream, 0) == Z_OK)
1659 /* nothing */;
1660 return;
1662 if (!strcmp(type, blob_type))
1663 obj_type = OBJ_BLOB;
1664 else if (!strcmp(type, tree_type))
1665 obj_type = OBJ_TREE;
1666 else if (!strcmp(type, commit_type))
1667 obj_type = OBJ_COMMIT;
1668 else if (!strcmp(type, tag_type))
1669 obj_type = OBJ_TAG;
1670 else
1671 die("trying to generate bogus object of type '%s'", type);
1672 hdr = write_binary_header(stream->next_out, obj_type, len);
1673 stream->total_out = hdr;
1674 stream->next_out += hdr;
1675 stream->avail_out -= hdr;
1678 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1679 unsigned char *sha1)
1681 unsigned char hdr[50];
1682 int hdrlen;
1683 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1684 return 0;
1687 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1689 int size;
1690 unsigned char *compressed;
1691 z_stream stream;
1692 unsigned char sha1[20];
1693 char *filename;
1694 static char tmpfile[PATH_MAX];
1695 unsigned char hdr[50];
1696 int fd, hdrlen;
1698 /* Normally if we have it in the pack then we do not bother writing
1699 * it out into .git/objects/??/?{38} file.
1701 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1702 filename = sha1_file_name(sha1);
1703 if (returnsha1)
1704 hashcpy(returnsha1, sha1);
1705 if (has_sha1_file(sha1))
1706 return 0;
1707 fd = open(filename, O_RDONLY);
1708 if (fd >= 0) {
1710 * FIXME!!! We might do collision checking here, but we'd
1711 * need to uncompress the old file and check it. Later.
1713 close(fd);
1714 return 0;
1717 if (errno != ENOENT) {
1718 return error("sha1 file %s: %s\n", filename, strerror(errno));
1721 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1723 fd = mkstemp(tmpfile);
1724 if (fd < 0) {
1725 if (errno == EPERM)
1726 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1727 else
1728 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1731 /* Set it up */
1732 memset(&stream, 0, sizeof(stream));
1733 deflateInit(&stream, zlib_compression_level);
1734 size = 8 + deflateBound(&stream, len+hdrlen);
1735 compressed = xmalloc(size);
1737 /* Compress it */
1738 stream.next_out = compressed;
1739 stream.avail_out = size;
1741 /* First header.. */
1742 stream.next_in = hdr;
1743 stream.avail_in = hdrlen;
1744 setup_object_header(&stream, type, len);
1746 /* Then the data itself.. */
1747 stream.next_in = buf;
1748 stream.avail_in = len;
1749 while (deflate(&stream, Z_FINISH) == Z_OK)
1750 /* nothing */;
1751 deflateEnd(&stream);
1752 size = stream.total_out;
1754 if (write_buffer(fd, compressed, size) < 0)
1755 die("unable to write sha1 file");
1756 fchmod(fd, 0444);
1757 close(fd);
1758 free(compressed);
1760 return move_temp_to_file(tmpfile, filename);
1764 * We need to unpack and recompress the object for writing
1765 * it out to a different file.
1767 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1769 size_t size;
1770 z_stream stream;
1771 unsigned char *unpacked;
1772 unsigned long len;
1773 char type[20];
1774 char hdr[50];
1775 int hdrlen;
1776 void *buf;
1778 /* need to unpack and recompress it by itself */
1779 unpacked = read_packed_sha1(sha1, type, &len);
1780 if (!unpacked)
1781 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1783 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1785 /* Set it up */
1786 memset(&stream, 0, sizeof(stream));
1787 deflateInit(&stream, zlib_compression_level);
1788 size = deflateBound(&stream, len + hdrlen);
1789 buf = xmalloc(size);
1791 /* Compress it */
1792 stream.next_out = buf;
1793 stream.avail_out = size;
1795 /* First header.. */
1796 stream.next_in = (void *)hdr;
1797 stream.avail_in = hdrlen;
1798 while (deflate(&stream, 0) == Z_OK)
1799 /* nothing */;
1801 /* Then the data itself.. */
1802 stream.next_in = unpacked;
1803 stream.avail_in = len;
1804 while (deflate(&stream, Z_FINISH) == Z_OK)
1805 /* nothing */;
1806 deflateEnd(&stream);
1807 free(unpacked);
1809 *objsize = stream.total_out;
1810 return buf;
1813 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1815 int retval;
1816 unsigned long objsize;
1817 void *buf = map_sha1_file(sha1, &objsize);
1819 if (buf) {
1820 retval = write_buffer(fd, buf, objsize);
1821 munmap(buf, objsize);
1822 return retval;
1825 buf = repack_object(sha1, &objsize);
1826 retval = write_buffer(fd, buf, objsize);
1827 free(buf);
1828 return retval;
1831 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1832 size_t bufsize, size_t *bufposn)
1834 char tmpfile[PATH_MAX];
1835 int local;
1836 z_stream stream;
1837 unsigned char real_sha1[20];
1838 unsigned char discard[4096];
1839 int ret;
1840 SHA_CTX c;
1842 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1844 local = mkstemp(tmpfile);
1845 if (local < 0) {
1846 if (errno == EPERM)
1847 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1848 else
1849 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1852 memset(&stream, 0, sizeof(stream));
1854 inflateInit(&stream);
1856 SHA1_Init(&c);
1858 do {
1859 ssize_t size;
1860 if (*bufposn) {
1861 stream.avail_in = *bufposn;
1862 stream.next_in = (unsigned char *) buffer;
1863 do {
1864 stream.next_out = discard;
1865 stream.avail_out = sizeof(discard);
1866 ret = inflate(&stream, Z_SYNC_FLUSH);
1867 SHA1_Update(&c, discard, sizeof(discard) -
1868 stream.avail_out);
1869 } while (stream.avail_in && ret == Z_OK);
1870 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1871 die("unable to write sha1 file");
1872 memmove(buffer, buffer + *bufposn - stream.avail_in,
1873 stream.avail_in);
1874 *bufposn = stream.avail_in;
1875 if (ret != Z_OK)
1876 break;
1878 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1879 if (size <= 0) {
1880 close(local);
1881 unlink(tmpfile);
1882 if (!size)
1883 return error("Connection closed?");
1884 perror("Reading from connection");
1885 return -1;
1887 *bufposn += size;
1888 } while (1);
1889 inflateEnd(&stream);
1891 close(local);
1892 SHA1_Final(real_sha1, &c);
1893 if (ret != Z_STREAM_END) {
1894 unlink(tmpfile);
1895 return error("File %s corrupted", sha1_to_hex(sha1));
1897 if (hashcmp(sha1, real_sha1)) {
1898 unlink(tmpfile);
1899 return error("File %s has bad hash", sha1_to_hex(sha1));
1902 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1905 int has_pack_index(const unsigned char *sha1)
1907 struct stat st;
1908 if (stat(sha1_pack_index_name(sha1), &st))
1909 return 0;
1910 return 1;
1913 int has_pack_file(const unsigned char *sha1)
1915 struct stat st;
1916 if (stat(sha1_pack_name(sha1), &st))
1917 return 0;
1918 return 1;
1921 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1923 struct pack_entry e;
1924 return find_pack_entry(sha1, &e, ignore_packed);
1927 int has_sha1_file(const unsigned char *sha1)
1929 struct stat st;
1930 struct pack_entry e;
1932 if (find_pack_entry(sha1, &e, NULL))
1933 return 1;
1934 return find_sha1_file(sha1, &st) ? 1 : 0;
1938 * reads from fd as long as possible into a supplied buffer of size bytes.
1939 * If necessary the buffer's size is increased using realloc()
1941 * returns 0 if anything went fine and -1 otherwise
1943 * NOTE: both buf and size may change, but even when -1 is returned
1944 * you still have to free() it yourself.
1946 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1948 char* buf = *return_buf;
1949 unsigned long size = *return_size;
1950 int iret;
1951 unsigned long off = 0;
1953 do {
1954 iret = xread(fd, buf + off, size - off);
1955 if (iret > 0) {
1956 off += iret;
1957 if (off == size) {
1958 size *= 2;
1959 buf = xrealloc(buf, size);
1962 } while (iret > 0);
1964 *return_buf = buf;
1965 *return_size = off;
1967 if (iret < 0)
1968 return -1;
1969 return 0;
1972 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1974 unsigned long size = 4096;
1975 char *buf = xmalloc(size);
1976 int ret;
1978 if (read_pipe(fd, &buf, &size)) {
1979 free(buf);
1980 return -1;
1983 if (!type)
1984 type = blob_type;
1985 if (write_object)
1986 ret = write_sha1_file(buf, size, type, sha1);
1987 else
1988 ret = hash_sha1_file(buf, size, type, sha1);
1989 free(buf);
1990 return ret;
1993 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1995 unsigned long size = st->st_size;
1996 void *buf;
1997 int ret;
1999 buf = "";
2000 if (size)
2001 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2002 close(fd);
2004 if (!type)
2005 type = blob_type;
2006 if (write_object)
2007 ret = write_sha1_file(buf, size, type, sha1);
2008 else
2009 ret = hash_sha1_file(buf, size, type, sha1);
2010 if (size)
2011 munmap(buf, size);
2012 return ret;
2015 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2017 int fd;
2018 char *target;
2020 switch (st->st_mode & S_IFMT) {
2021 case S_IFREG:
2022 fd = open(path, O_RDONLY);
2023 if (fd < 0)
2024 return error("open(\"%s\"): %s", path,
2025 strerror(errno));
2026 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2027 return error("%s: failed to insert into database",
2028 path);
2029 break;
2030 case S_IFLNK:
2031 target = xmalloc(st->st_size+1);
2032 if (readlink(path, target, st->st_size+1) != st->st_size) {
2033 char *errstr = strerror(errno);
2034 free(target);
2035 return error("readlink(\"%s\"): %s", path,
2036 errstr);
2038 if (!write_object)
2039 hash_sha1_file(target, st->st_size, blob_type, sha1);
2040 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2041 return error("%s: failed to insert into database",
2042 path);
2043 free(target);
2044 break;
2045 default:
2046 return error("%s: unsupported file type", path);
2048 return 0;
2051 int read_pack_header(int fd, struct pack_header *header)
2053 char *c = (char*)header;
2054 ssize_t remaining = sizeof(struct pack_header);
2055 do {
2056 ssize_t r = xread(fd, c, remaining);
2057 if (r <= 0)
2058 /* "eof before pack header was fully read" */
2059 return PH_ERROR_EOF;
2060 remaining -= r;
2061 c += r;
2062 } while (remaining > 0);
2063 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2064 /* "protocol error (pack signature mismatch detected)" */
2065 return PH_ERROR_PACK_SIGNATURE;
2066 if (!pack_version_ok(header->hdr_version))
2067 /* "protocol error (pack version unsupported)" */
2068 return PH_ERROR_PROTOCOL;
2069 return 0;