1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9#include "cache.h"
10#include "delta.h"
11#include "pack.h"
12#include "blob.h"
13#include "commit.h"
14#include "tag.h"
15#include "tree.h"
16
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
24
25#ifdef NO_C99_FORMAT
26#define SZ_FMT "lu"
27#else
28#define SZ_FMT "zu"
29#endif
30
31const unsigned char null_sha1[20];
32
33static unsigned int sha1_file_open_flag = O_NOATIME;
34
35signed 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 */
68};
69
70int get_sha1_hex(const char *hex, unsigned char *sha1)
71{
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;
79 }
80 return 0;
81}
82
83int safe_create_leading_directories(char *path)
84{
85 char *pos = path;
86 struct stat st;
87
88 if (*pos == '/')
89 pos++;
90
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;
101 }
102 }
103 else if (mkdir(path, 0777)) {
104 *pos = '/';
105 return -1;
106 }
107 else if (adjust_shared_perm(path)) {
108 *pos = '/';
109 return -2;
110 }
111 *pos++ = '/';
112 }
113 return 0;
114}
115
116char * sha1_to_hex(const unsigned char *sha1)
117{
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;
123
124 for (i = 0; i < 20; i++) {
125 unsigned int val = *sha1++;
126 *buf++ = hex[val >> 4];
127 *buf++ = hex[val & 0xf];
128 }
129 *buf = '\0';
130
131 return buffer;
132}
133
134static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
135{
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];
143 }
144}
145
146/*
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.
150 *
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.
155 */
156char *sha1_file_name(const unsigned char *sha1)
157{
158 static char *name, *base;
159
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;
169 }
170 fill_sha1_path(name, sha1);
171 return base;
172}
173
174char *sha1_pack_name(const unsigned char *sha1)
175{
176 static const char hex[] = "0123456789abcdef";
177 static char *name, *base, *buf;
178 int i;
179
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;
186 }
187
188 buf = name;
189
190 for (i = 0; i < 20; i++) {
191 unsigned int val = *sha1++;
192 *buf++ = hex[val >> 4];
193 *buf++ = hex[val & 0xf];
194 }
195
196 return base;
197}
198
199char *sha1_pack_index_name(const unsigned char *sha1)
200{
201 static const char hex[] = "0123456789abcdef";
202 static char *name, *base, *buf;
203 int i;
204
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;
211 }
212
213 buf = name;
214
215 for (i = 0; i < 20; i++) {
216 unsigned int val = *sha1++;
217 *buf++ = hex[val >> 4];
218 *buf++ = hex[val & 0xf];
219 }
220
221 return base;
222}
223
224struct alternate_object_database *alt_odb_list;
225static struct alternate_object_database **alt_odb_tail;
226
227static void read_info_alternates(const char * alternates, int depth);
228
229/*
230 * Prepare alternate object database registry.
231 *
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.
243 */
244static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
245{
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;
254
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;
261 }
262 ent = xmalloc(sizeof(*ent) + entlen);
263
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);
268 }
269 else
270 memcpy(ent->base, entry, pfxlen);
271
272 ent->name = ent->base + pfxlen + 1;
273 ent->base[pfxlen + 3] = '/';
274 ent->base[pfxlen] = ent->base[entlen-1] = 0;
275
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;
283 }
284
285 /* Prevent the common mistake of listing the same
286 * thing twice, or object directory itself.
287 */
288 for (alt = alt_odb_list; alt; alt = alt->next) {
289 if (!memcmp(ent->base, alt->base, pfxlen)) {
290 free(ent);
291 return -1;
292 }
293 }
294 if (!memcmp(ent->base, objdir, pfxlen)) {
295 free(ent);
296 return -1;
297 }
298
299 /* add the alternate entry */
300 *alt_odb_tail = ent;
301 alt_odb_tail = &(ent->next);
302 ent->next = NULL;
303
304 /* recursively add alternates */
305 read_info_alternates(ent->base, depth + 1);
306
307 ent->base[pfxlen] = '/';
308
309 return 0;
310}
311
312static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
313 const char *relative_base, int depth)
314{
315 const char *cp, *last;
316
317 if (depth > 5) {
318 error("%s: ignoring alternate object stores, nesting too deep.",
319 relative_base);
320 return;
321 }
322
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;
331 }
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);
341 }
342 }
343 while (cp < ep && *cp == sep)
344 cp++;
345 last = cp;
346 }
347}
348
349static void read_info_alternates(const char * relative_base, int depth)
350{
351 char *map;
352 struct stat st;
353 char path[PATH_MAX];
354 int fd;
355
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;
363 }
364 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
365 close(fd);
366
367 link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
368
369 munmap(map, st.st_size);
370}
371
372void prepare_alt_odb(void)
373{
374 const char *alt;
375
376 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
377 if (!alt) alt = "";
378
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);
383
384 read_info_alternates(get_object_directory(), 0);
385}
386
387static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
388{
389 char *name = sha1_file_name(sha1);
390 struct alternate_object_database *alt;
391
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;
400 }
401 return NULL;
402}
403
404static unsigned int pack_used_ctr;
405static unsigned int pack_mmap_calls;
406static unsigned int peak_pack_open_windows;
407static unsigned int pack_open_windows;
408static size_t peak_pack_mapped;
409static size_t pack_mapped;
410static size_t page_size;
411struct packed_git *packed_git;
412
413void pack_report()
414{
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);
432}
433
434static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
435 void **idx_map_)
436{
437 void *idx_map;
438 unsigned int *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;
448 }
449 idx_size = st.st_size;
450 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
451 close(fd);
452
453 index = idx_map;
454 *idx_map_ = idx_map;
455 *idx_size_ = idx_size;
456
457 /* check index map */
458 if (idx_size < 4*256 + 20 + 20)
459 return error("index file too small");
460 nr = 0;
461 for (i = 0; i < 256; i++) {
462 unsigned int n = ntohl(index[i]);
463 if (n < nr)
464 return error("non-monotonic index");
465 nr = n;
466 }
467
468 /*
469 * Total size:
470 * - 256 index entries 4 bytes each
471 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
472 * - 20-byte SHA1 of the packfile
473 * - 20-byte SHA1 file checksum
474 */
475 if (idx_size != 4*256 + nr * 24 + 20 + 20)
476 return error("wrong index file size");
477
478 return 0;
479}
480
481static void scan_windows(struct packed_git *p,
482 struct packed_git **lru_p,
483 struct pack_window **lru_w,
484 struct pack_window **lru_l)
485{
486 struct pack_window *w, *w_l;
487
488 for (w_l = NULL, w = p->windows; w; w = w->next) {
489 if (!w->inuse_cnt) {
490 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
491 *lru_p = p;
492 *lru_w = w;
493 *lru_l = w_l;
494 }
495 }
496 w_l = w;
497 }
498}
499
500static int unuse_one_window(struct packed_git *current)
501{
502 struct packed_git *p, *lru_p = NULL;
503 struct pack_window *lru_w = NULL, *lru_l = NULL;
504
505 if (current)
506 scan_windows(current, &lru_p, &lru_w, &lru_l);
507 for (p = packed_git; p; p = p->next)
508 scan_windows(p, &lru_p, &lru_w, &lru_l);
509 if (lru_p) {
510 munmap(lru_w->base, lru_w->len);
511 pack_mapped -= lru_w->len;
512 if (lru_l)
513 lru_l->next = lru_w->next;
514 else {
515 lru_p->windows = lru_w->next;
516 if (!lru_p->windows && lru_p != current) {
517 close(lru_p->pack_fd);
518 lru_p->pack_fd = -1;
519 }
520 }
521 free(lru_w);
522 pack_open_windows--;
523 return 1;
524 }
525 return 0;
526}
527
528void release_pack_memory(size_t need)
529{
530 size_t cur = pack_mapped;
531 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
532 ; /* nothing */
533}
534
535void unuse_pack(struct pack_window **w_cursor)
536{
537 struct pack_window *w = *w_cursor;
538 if (w) {
539 w->inuse_cnt--;
540 *w_cursor = NULL;
541 }
542}
543
544static void open_packed_git(struct packed_git *p)
545{
546 struct stat st;
547 struct pack_header hdr;
548 unsigned char sha1[20];
549 unsigned char *idx_sha1;
550 long fd_flag;
551
552 p->pack_fd = open(p->pack_name, O_RDONLY);
553 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
554 die("packfile %s cannot be opened", p->pack_name);
555
556 /* If we created the struct before we had the pack we lack size. */
557 if (!p->pack_size) {
558 if (!S_ISREG(st.st_mode))
559 die("packfile %s not a regular file", p->pack_name);
560 p->pack_size = st.st_size;
561 } else if (p->pack_size != st.st_size)
562 die("packfile %s size changed", p->pack_name);
563
564 /* We leave these file descriptors open with sliding mmap;
565 * there is no point keeping them open across exec(), though.
566 */
567 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
568 if (fd_flag < 0)
569 die("cannot determine file descriptor flags");
570 fd_flag |= FD_CLOEXEC;
571 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
572 die("cannot set FD_CLOEXEC");
573
574 /* Verify we recognize this pack file format. */
575 read_or_die(p->pack_fd, &hdr, sizeof(hdr));
576 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
577 die("file %s is not a GIT packfile", p->pack_name);
578 if (!pack_version_ok(hdr.hdr_version))
579 die("packfile %s is version %u and not supported"
580 " (try upgrading GIT to a newer version)",
581 p->pack_name, ntohl(hdr.hdr_version));
582
583 /* Verify the pack matches its index. */
584 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
585 die("packfile %s claims to have %u objects"
586 " while index size indicates %u objects",
587 p->pack_name, ntohl(hdr.hdr_entries),
588 num_packed_objects(p));
589 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
590 die("end of packfile %s is unavailable", p->pack_name);
591 read_or_die(p->pack_fd, sha1, sizeof(sha1));
592 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
593 if (hashcmp(sha1, idx_sha1))
594 die("packfile %s does not match index", p->pack_name);
595}
596
597static int in_window(struct pack_window *win, unsigned long offset)
598{
599 /* We must promise at least 20 bytes (one hash) after the
600 * offset is available from this window, otherwise the offset
601 * is not actually in this window and a different window (which
602 * has that one hash excess) must be used. This is to support
603 * the object header and delta base parsing routines below.
604 */
605 off_t win_off = win->offset;
606 return win_off <= offset
607 && (offset + 20) <= (win_off + win->len);
608}
609
610unsigned char* use_pack(struct packed_git *p,
611 struct pack_window **w_cursor,
612 unsigned long offset,
613 unsigned int *left)
614{
615 struct pack_window *win = *w_cursor;
616
617 if (p->pack_fd == -1)
618 open_packed_git(p);
619
620 /* Since packfiles end in a hash of their content and its
621 * pointless to ask for an offset into the middle of that
622 * hash, and the in_window function above wouldn't match
623 * don't allow an offset too close to the end of the file.
624 */
625 if (offset > (p->pack_size - 20))
626 die("offset beyond end of packfile (truncated pack?)");
627
628 if (!win || !in_window(win, offset)) {
629 if (win)
630 win->inuse_cnt--;
631 for (win = p->windows; win; win = win->next) {
632 if (in_window(win, offset))
633 break;
634 }
635 if (!win) {
636 if (!page_size)
637 page_size = getpagesize();
638 win = xcalloc(1, sizeof(*win));
639 win->offset = (offset / page_size) * page_size;
640 win->len = p->pack_size - win->offset;
641 if (win->len > packed_git_window_size)
642 win->len = packed_git_window_size;
643 pack_mapped += win->len;
644 while (packed_git_limit < pack_mapped
645 && unuse_one_window(p))
646 ; /* nothing */
647 win->base = xmmap(NULL, win->len,
648 PROT_READ, MAP_PRIVATE,
649 p->pack_fd, win->offset);
650 if (win->base == MAP_FAILED)
651 die("packfile %s cannot be mapped: %s",
652 p->pack_name,
653 strerror(errno));
654 pack_mmap_calls++;
655 pack_open_windows++;
656 if (pack_mapped > peak_pack_mapped)
657 peak_pack_mapped = pack_mapped;
658 if (pack_open_windows > peak_pack_open_windows)
659 peak_pack_open_windows = pack_open_windows;
660 win->next = p->windows;
661 p->windows = win;
662 }
663 }
664 if (win != *w_cursor) {
665 win->last_used = pack_used_ctr++;
666 win->inuse_cnt++;
667 *w_cursor = win;
668 }
669 offset -= win->offset;
670 if (left)
671 *left = win->len - offset;
672 return win->base + offset;
673}
674
675struct packed_git *add_packed_git(char *path, int path_len, int local)
676{
677 struct stat st;
678 struct packed_git *p;
679 unsigned long idx_size;
680 void *idx_map;
681 unsigned char sha1[20];
682
683 if (check_packed_git_idx(path, &idx_size, &idx_map))
684 return NULL;
685
686 /* do we have a corresponding .pack file? */
687 strcpy(path + path_len - 4, ".pack");
688 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
689 munmap(idx_map, idx_size);
690 return NULL;
691 }
692 /* ok, it looks sane as far as we can check without
693 * actually mapping the pack file.
694 */
695 p = xmalloc(sizeof(*p) + path_len + 2);
696 strcpy(p->pack_name, path);
697 p->index_size = idx_size;
698 p->pack_size = st.st_size;
699 p->index_base = idx_map;
700 p->next = NULL;
701 p->windows = NULL;
702 p->pack_fd = -1;
703 p->pack_local = local;
704 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
705 hashcpy(p->sha1, sha1);
706 return p;
707}
708
709struct packed_git *parse_pack_index(unsigned char *sha1)
710{
711 char *path = sha1_pack_index_name(sha1);
712 return parse_pack_index_file(sha1, path);
713}
714
715struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
716{
717 struct packed_git *p;
718 unsigned long idx_size;
719 void *idx_map;
720 char *path;
721
722 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
723 return NULL;
724
725 path = sha1_pack_name(sha1);
726
727 p = xmalloc(sizeof(*p) + strlen(path) + 2);
728 strcpy(p->pack_name, path);
729 p->index_size = idx_size;
730 p->pack_size = 0;
731 p->index_base = idx_map;
732 p->next = NULL;
733 p->windows = NULL;
734 p->pack_fd = -1;
735 hashcpy(p->sha1, sha1);
736 return p;
737}
738
739void install_packed_git(struct packed_git *pack)
740{
741 pack->next = packed_git;
742 packed_git = pack;
743}
744
745static void prepare_packed_git_one(char *objdir, int local)
746{
747 char path[PATH_MAX];
748 int len;
749 DIR *dir;
750 struct dirent *de;
751
752 sprintf(path, "%s/pack", objdir);
753 len = strlen(path);
754 dir = opendir(path);
755 if (!dir) {
756 if (errno != ENOENT)
757 error("unable to open object pack directory: %s: %s",
758 path, strerror(errno));
759 return;
760 }
761 path[len++] = '/';
762 while ((de = readdir(dir)) != NULL) {
763 int namelen = strlen(de->d_name);
764 struct packed_git *p;
765
766 if (!has_extension(de->d_name, ".idx"))
767 continue;
768
769 /* we have .idx. Is it a file we can map? */
770 strcpy(path + len, de->d_name);
771 for (p = packed_git; p; p = p->next) {
772 if (!memcmp(path, p->pack_name, len + namelen - 4))
773 break;
774 }
775 if (p)
776 continue;
777 p = add_packed_git(path, len + namelen, local);
778 if (!p)
779 continue;
780 p->next = packed_git;
781 packed_git = p;
782 }
783 closedir(dir);
784}
785
786static int prepare_packed_git_run_once = 0;
787void prepare_packed_git(void)
788{
789 struct alternate_object_database *alt;
790
791 if (prepare_packed_git_run_once)
792 return;
793 prepare_packed_git_one(get_object_directory(), 1);
794 prepare_alt_odb();
795 for (alt = alt_odb_list; alt; alt = alt->next) {
796 alt->name[-1] = 0;
797 prepare_packed_git_one(alt->base, 0);
798 alt->name[-1] = '/';
799 }
800 prepare_packed_git_run_once = 1;
801}
802
803void reprepare_packed_git(void)
804{
805 prepare_packed_git_run_once = 0;
806 prepare_packed_git();
807}
808
809int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
810{
811 unsigned char real_sha1[20];
812 hash_sha1_file(map, size, type, real_sha1);
813 return hashcmp(sha1, real_sha1) ? -1 : 0;
814}
815
816void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
817{
818 struct stat st;
819 void *map;
820 int fd;
821 char *filename = find_sha1_file(sha1, &st);
822
823 if (!filename) {
824 return NULL;
825 }
826
827 fd = open(filename, O_RDONLY | sha1_file_open_flag);
828 if (fd < 0) {
829 /* See if it works without O_NOATIME */
830 switch (sha1_file_open_flag) {
831 default:
832 fd = open(filename, O_RDONLY);
833 if (fd >= 0)
834 break;
835 /* Fallthrough */
836 case 0:
837 return NULL;
838 }
839
840 /* If it failed once, it will probably fail again.
841 * Stop using O_NOATIME
842 */
843 sha1_file_open_flag = 0;
844 }
845 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
846 close(fd);
847 *size = st.st_size;
848 return map;
849}
850
851int legacy_loose_object(unsigned char *map)
852{
853 unsigned int word;
854
855 /*
856 * Is it a zlib-compressed buffer? If so, the first byte
857 * must be 0x78 (15-bit window size, deflated), and the
858 * first 16-bit word is evenly divisible by 31
859 */
860 word = (map[0] << 8) + map[1];
861 if (map[0] == 0x78 && !(word % 31))
862 return 1;
863 else
864 return 0;
865}
866
867unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
868{
869 unsigned shift;
870 unsigned char c;
871 unsigned long size;
872 unsigned long used = 0;
873
874 c = buf[used++];
875 *type = (c >> 4) & 7;
876 size = c & 15;
877 shift = 4;
878 while (c & 0x80) {
879 if (len <= used)
880 return 0;
881 if (sizeof(long) * 8 <= shift)
882 return 0;
883 c = buf[used++];
884 size += (c & 0x7f) << shift;
885 shift += 7;
886 }
887 *sizep = size;
888 return used;
889}
890
891static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
892{
893 unsigned long size, used;
894 static const char valid_loose_object_type[8] = {
895 0, /* OBJ_EXT */
896 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
897 0, /* "delta" and others are invalid in a loose object */
898 };
899 enum object_type type;
900
901 /* Get the data stream */
902 memset(stream, 0, sizeof(*stream));
903 stream->next_in = map;
904 stream->avail_in = mapsize;
905 stream->next_out = buffer;
906 stream->avail_out = bufsiz;
907
908 if (legacy_loose_object(map)) {
909 inflateInit(stream);
910 return inflate(stream, 0);
911 }
912
913 used = unpack_object_header_gently(map, mapsize, &type, &size);
914 if (!used || !valid_loose_object_type[type])
915 return -1;
916 map += used;
917 mapsize -= used;
918
919 /* Set up the stream for the rest.. */
920 stream->next_in = map;
921 stream->avail_in = mapsize;
922 inflateInit(stream);
923
924 /* And generate the fake traditional header */
925 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
926 type_names[type], size);
927 return 0;
928}
929
930static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
931{
932 int bytes = strlen(buffer) + 1;
933 unsigned char *buf = xmalloc(1+size);
934 unsigned long n;
935
936 n = stream->total_out - bytes;
937 if (n > size)
938 n = size;
939 memcpy(buf, (char *) buffer + bytes, n);
940 bytes = n;
941 if (bytes < size) {
942 stream->next_out = buf + bytes;
943 stream->avail_out = size - bytes;
944 while (inflate(stream, Z_FINISH) == Z_OK)
945 /* nothing */;
946 }
947 buf[size] = 0;
948 inflateEnd(stream);
949 return buf;
950}
951
952/*
953 * We used to just use "sscanf()", but that's actually way
954 * too permissive for what we want to check. So do an anal
955 * object header parse by hand.
956 */
957static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
958{
959 int i;
960 unsigned long size;
961
962 /*
963 * The type can be at most ten bytes (including the
964 * terminating '\0' that we add), and is followed by
965 * a space.
966 */
967 i = 10;
968 for (;;) {
969 char c = *hdr++;
970 if (c == ' ')
971 break;
972 if (!--i)
973 return -1;
974 *type++ = c;
975 }
976 *type = 0;
977
978 /*
979 * The length must follow immediately, and be in canonical
980 * decimal format (ie "010" is not valid).
981 */
982 size = *hdr++ - '0';
983 if (size > 9)
984 return -1;
985 if (size) {
986 for (;;) {
987 unsigned long c = *hdr - '0';
988 if (c > 9)
989 break;
990 hdr++;
991 size = size * 10 + c;
992 }
993 }
994 *sizep = size;
995
996 /*
997 * The length must be followed by a zero byte
998 */
999 return *hdr ? -1 : 0;
1000}
1001
1002void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1003{
1004 int ret;
1005 z_stream stream;
1006 char hdr[8192];
1007
1008 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1009 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1010 return NULL;
1011
1012 return unpack_sha1_rest(&stream, hdr, *size);
1013}
1014
1015static unsigned long get_delta_base(struct packed_git *p,
1016 struct pack_window **w_curs,
1017 unsigned long offset,
1018 enum object_type kind,
1019 unsigned long delta_obj_offset,
1020 unsigned long *base_obj_offset)
1021{
1022 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1023 unsigned long base_offset;
1024
1025 /* use_pack() assured us we have [base_info, base_info + 20)
1026 * as a range that we can look at without walking off the
1027 * end of the mapped window. Its actually the hash size
1028 * that is assured. An OFS_DELTA longer than the hash size
1029 * is stupid, as then a REF_DELTA would be smaller to store.
1030 */
1031 if (kind == OBJ_OFS_DELTA) {
1032 unsigned used = 0;
1033 unsigned char c = base_info[used++];
1034 base_offset = c & 127;
1035 while (c & 128) {
1036 base_offset += 1;
1037 if (!base_offset || base_offset & ~(~0UL >> 7))
1038 die("offset value overflow for delta base object");
1039 c = base_info[used++];
1040 base_offset = (base_offset << 7) + (c & 127);
1041 }
1042 base_offset = delta_obj_offset - base_offset;
1043 if (base_offset >= delta_obj_offset)
1044 die("delta base offset out of bound");
1045 offset += used;
1046 } else if (kind == OBJ_REF_DELTA) {
1047 /* The base entry _must_ be in the same pack */
1048 base_offset = find_pack_entry_one(base_info, p);
1049 if (!base_offset)
1050 die("failed to find delta-pack base object %s",
1051 sha1_to_hex(base_info));
1052 offset += 20;
1053 } else
1054 die("I am totally screwed");
1055 *base_obj_offset = base_offset;
1056 return offset;
1057}
1058
1059/* forward declaration for a mutually recursive function */
1060static int packed_object_info(struct packed_git *p, unsigned long offset,
1061 char *type, unsigned long *sizep);
1062
1063static int packed_delta_info(struct packed_git *p,
1064 struct pack_window **w_curs,
1065 unsigned long offset,
1066 enum object_type kind,
1067 unsigned long obj_offset,
1068 char *type,
1069 unsigned long *sizep)
1070{
1071 unsigned long base_offset;
1072
1073 offset = get_delta_base(p, w_curs, offset, kind,
1074 obj_offset, &base_offset);
1075
1076 /* We choose to only get the type of the base object and
1077 * ignore potentially corrupt pack file that expects the delta
1078 * based on a base with a wrong size. This saves tons of
1079 * inflate() calls.
1080 */
1081 if (packed_object_info(p, base_offset, type, NULL))
1082 die("cannot get info for delta-pack base");
1083
1084 if (sizep) {
1085 const unsigned char *data;
1086 unsigned char delta_head[20], *in;
1087 unsigned long result_size;
1088 z_stream stream;
1089 int st;
1090
1091 memset(&stream, 0, sizeof(stream));
1092 stream.next_out = delta_head;
1093 stream.avail_out = sizeof(delta_head);
1094
1095 inflateInit(&stream);
1096 do {
1097 in = use_pack(p, w_curs, offset, &stream.avail_in);
1098 stream.next_in = in;
1099 st = inflate(&stream, Z_FINISH);
1100 offset += stream.next_in - in;
1101 } while ((st == Z_OK || st == Z_BUF_ERROR)
1102 && stream.total_out < sizeof(delta_head));
1103 inflateEnd(&stream);
1104 if ((st != Z_STREAM_END) &&
1105 stream.total_out != sizeof(delta_head))
1106 die("delta data unpack-initial failed");
1107
1108 /* Examine the initial part of the delta to figure out
1109 * the result size.
1110 */
1111 data = delta_head;
1112
1113 /* ignore base size */
1114 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1115
1116 /* Read the result size */
1117 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1118 *sizep = result_size;
1119 }
1120 return 0;
1121}
1122
1123static unsigned long unpack_object_header(struct packed_git *p,
1124 struct pack_window **w_curs,
1125 unsigned long offset,
1126 enum object_type *type,
1127 unsigned long *sizep)
1128{
1129 unsigned char *base;
1130 unsigned int left;
1131 unsigned long used;
1132
1133 /* use_pack() assures us we have [base, base + 20) available
1134 * as a range that we can look at at. (Its actually the hash
1135 * size that is assurred.) With our object header encoding
1136 * the maximum deflated object size is 2^137, which is just
1137 * insane, so we know won't exceed what we have been given.
1138 */
1139 base = use_pack(p, w_curs, offset, &left);
1140 used = unpack_object_header_gently(base, left, type, sizep);
1141 if (!used)
1142 die("object offset outside of pack file");
1143
1144 return offset + used;
1145}
1146
1147void packed_object_info_detail(struct packed_git *p,
1148 unsigned long offset,
1149 char *type,
1150 unsigned long *size,
1151 unsigned long *store_size,
1152 unsigned int *delta_chain_length,
1153 unsigned char *base_sha1)
1154{
1155 struct pack_window *w_curs = NULL;
1156 unsigned long obj_offset, val;
1157 unsigned char *next_sha1;
1158 enum object_type kind;
1159
1160 *delta_chain_length = 0;
1161 obj_offset = offset;
1162 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1163
1164 for (;;) {
1165 switch (kind) {
1166 default:
1167 die("pack %s contains unknown object type %d",
1168 p->pack_name, kind);
1169 case OBJ_COMMIT:
1170 case OBJ_TREE:
1171 case OBJ_BLOB:
1172 case OBJ_TAG:
1173 strcpy(type, type_names[kind]);
1174 *store_size = 0; /* notyet */
1175 unuse_pack(&w_curs);
1176 return;
1177 case OBJ_OFS_DELTA:
1178 get_delta_base(p, &w_curs, offset, kind,
1179 obj_offset, &offset);
1180 if (*delta_chain_length == 0) {
1181 /* TODO: find base_sha1 as pointed by offset */
1182 }
1183 break;
1184 case OBJ_REF_DELTA:
1185 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1186 if (*delta_chain_length == 0)
1187 hashcpy(base_sha1, next_sha1);
1188 offset = find_pack_entry_one(next_sha1, p);
1189 break;
1190 }
1191 obj_offset = offset;
1192 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1193 (*delta_chain_length)++;
1194 }
1195}
1196
1197static int packed_object_info(struct packed_git *p, unsigned long offset,
1198 char *type, unsigned long *sizep)
1199{
1200 struct pack_window *w_curs = NULL;
1201 unsigned long size, obj_offset = offset;
1202 enum object_type kind;
1203 int r;
1204
1205 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1206
1207 switch (kind) {
1208 case OBJ_OFS_DELTA:
1209 case OBJ_REF_DELTA:
1210 r = packed_delta_info(p, &w_curs, offset, kind,
1211 obj_offset, type, sizep);
1212 unuse_pack(&w_curs);
1213 return r;
1214 case OBJ_COMMIT:
1215 case OBJ_TREE:
1216 case OBJ_BLOB:
1217 case OBJ_TAG:
1218 strcpy(type, type_names[kind]);
1219 unuse_pack(&w_curs);
1220 break;
1221 default:
1222 die("pack %s contains unknown object type %d",
1223 p->pack_name, kind);
1224 }
1225 if (sizep)
1226 *sizep = size;
1227 return 0;
1228}
1229
1230static void *unpack_compressed_entry(struct packed_git *p,
1231 struct pack_window **w_curs,
1232 unsigned long offset,
1233 unsigned long size)
1234{
1235 int st;
1236 z_stream stream;
1237 unsigned char *buffer, *in;
1238
1239 buffer = xmalloc(size + 1);
1240 buffer[size] = 0;
1241 memset(&stream, 0, sizeof(stream));
1242 stream.next_out = buffer;
1243 stream.avail_out = size;
1244
1245 inflateInit(&stream);
1246 do {
1247 in = use_pack(p, w_curs, offset, &stream.avail_in);
1248 stream.next_in = in;
1249 st = inflate(&stream, Z_FINISH);
1250 offset += stream.next_in - in;
1251 } while (st == Z_OK || st == Z_BUF_ERROR);
1252 inflateEnd(&stream);
1253 if ((st != Z_STREAM_END) || stream.total_out != size) {
1254 free(buffer);
1255 return NULL;
1256 }
1257
1258 return buffer;
1259}
1260
1261static void *unpack_delta_entry(struct packed_git *p,
1262 struct pack_window **w_curs,
1263 unsigned long offset,
1264 unsigned long delta_size,
1265 enum object_type kind,
1266 unsigned long obj_offset,
1267 char *type,
1268 unsigned long *sizep)
1269{
1270 void *delta_data, *result, *base;
1271 unsigned long result_size, base_size, base_offset;
1272
1273 offset = get_delta_base(p, w_curs, offset, kind,
1274 obj_offset, &base_offset);
1275 base = unpack_entry(p, base_offset, type, &base_size);
1276 if (!base)
1277 die("failed to read delta base object at %lu from %s",
1278 base_offset, p->pack_name);
1279
1280 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1281 result = patch_delta(base, base_size,
1282 delta_data, delta_size,
1283 &result_size);
1284 if (!result)
1285 die("failed to apply delta");
1286 free(delta_data);
1287 free(base);
1288 *sizep = result_size;
1289 return result;
1290}
1291
1292void *unpack_entry(struct packed_git *p, unsigned long offset,
1293 char *type, unsigned long *sizep)
1294{
1295 struct pack_window *w_curs = NULL;
1296 unsigned long size, obj_offset = offset;
1297 enum object_type kind;
1298 void *retval;
1299
1300 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1301 switch (kind) {
1302 case OBJ_OFS_DELTA:
1303 case OBJ_REF_DELTA:
1304 retval = unpack_delta_entry(p, &w_curs, offset, size,
1305 kind, obj_offset, type, sizep);
1306 break;
1307 case OBJ_COMMIT:
1308 case OBJ_TREE:
1309 case OBJ_BLOB:
1310 case OBJ_TAG:
1311 strcpy(type, type_names[kind]);
1312 *sizep = size;
1313 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1314 break;
1315 default:
1316 die("unknown object type %i in %s", kind, p->pack_name);
1317 }
1318 unuse_pack(&w_curs);
1319 return retval;
1320}
1321
1322int num_packed_objects(const struct packed_git *p)
1323{
1324 /* See check_packed_git_idx() */
1325 return (p->index_size - 20 - 20 - 4*256) / 24;
1326}
1327
1328int nth_packed_object_sha1(const struct packed_git *p, int n,
1329 unsigned char* sha1)
1330{
1331 void *index = p->index_base + 256;
1332 if (n < 0 || num_packed_objects(p) <= n)
1333 return -1;
1334 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1335 return 0;
1336}
1337
1338unsigned long find_pack_entry_one(const unsigned char *sha1,
1339 struct packed_git *p)
1340{
1341 unsigned int *level1_ofs = p->index_base;
1342 int hi = ntohl(level1_ofs[*sha1]);
1343 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1344 void *index = p->index_base + 256;
1345
1346 do {
1347 int mi = (lo + hi) / 2;
1348 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1349 if (!cmp)
1350 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1351 if (cmp > 0)
1352 hi = mi;
1353 else
1354 lo = mi+1;
1355 } while (lo < hi);
1356 return 0;
1357}
1358
1359static int matches_pack_name(struct packed_git *p, const char *ig)
1360{
1361 const char *last_c, *c;
1362
1363 if (!strcmp(p->pack_name, ig))
1364 return 0;
1365
1366 for (c = p->pack_name, last_c = c; *c;)
1367 if (*c == '/')
1368 last_c = ++c;
1369 else
1370 ++c;
1371 if (!strcmp(last_c, ig))
1372 return 0;
1373
1374 return 1;
1375}
1376
1377static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1378{
1379 struct packed_git *p;
1380 unsigned long offset;
1381
1382 prepare_packed_git();
1383
1384 for (p = packed_git; p; p = p->next) {
1385 if (ignore_packed) {
1386 const char **ig;
1387 for (ig = ignore_packed; *ig; ig++)
1388 if (!matches_pack_name(p, *ig))
1389 break;
1390 if (*ig)
1391 continue;
1392 }
1393 offset = find_pack_entry_one(sha1, p);
1394 if (offset) {
1395 e->offset = offset;
1396 e->p = p;
1397 hashcpy(e->sha1, sha1);
1398 return 1;
1399 }
1400 }
1401 return 0;
1402}
1403
1404struct packed_git *find_sha1_pack(const unsigned char *sha1,
1405 struct packed_git *packs)
1406{
1407 struct packed_git *p;
1408
1409 for (p = packs; p; p = p->next) {
1410 if (find_pack_entry_one(sha1, p))
1411 return p;
1412 }
1413 return NULL;
1414
1415}
1416
1417static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1418{
1419 int status;
1420 unsigned long mapsize, size;
1421 void *map;
1422 z_stream stream;
1423 char hdr[128];
1424
1425 map = map_sha1_file(sha1, &mapsize);
1426 if (!map)
1427 return error("unable to find %s", sha1_to_hex(sha1));
1428 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1429 status = error("unable to unpack %s header",
1430 sha1_to_hex(sha1));
1431 if (parse_sha1_header(hdr, type, &size) < 0)
1432 status = error("unable to parse %s header", sha1_to_hex(sha1));
1433 else {
1434 status = 0;
1435 if (sizep)
1436 *sizep = size;
1437 }
1438 inflateEnd(&stream);
1439 munmap(map, mapsize);
1440 return status;
1441}
1442
1443int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1444{
1445 struct pack_entry e;
1446
1447 if (!find_pack_entry(sha1, &e, NULL)) {
1448 reprepare_packed_git();
1449 if (!find_pack_entry(sha1, &e, NULL))
1450 return sha1_loose_object_info(sha1, type, sizep);
1451 }
1452 return packed_object_info(e.p, e.offset, type, sizep);
1453}
1454
1455static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1456{
1457 struct pack_entry e;
1458
1459 if (!find_pack_entry(sha1, &e, NULL)) {
1460 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1461 return NULL;
1462 }
1463 return unpack_entry(e.p, e.offset, type, size);
1464}
1465
1466void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1467{
1468 unsigned long mapsize;
1469 void *map, *buf;
1470 struct pack_entry e;
1471
1472 if (find_pack_entry(sha1, &e, NULL))
1473 return read_packed_sha1(sha1, type, size);
1474 map = map_sha1_file(sha1, &mapsize);
1475 if (map) {
1476 buf = unpack_sha1_file(map, mapsize, type, size);
1477 munmap(map, mapsize);
1478 return buf;
1479 }
1480 reprepare_packed_git();
1481 if (find_pack_entry(sha1, &e, NULL))
1482 return read_packed_sha1(sha1, type, size);
1483 return NULL;
1484}
1485
1486void *read_object_with_reference(const unsigned char *sha1,
1487 const char *required_type,
1488 unsigned long *size,
1489 unsigned char *actual_sha1_return)
1490{
1491 char type[20];
1492 void *buffer;
1493 unsigned long isize;
1494 unsigned char actual_sha1[20];
1495
1496 hashcpy(actual_sha1, sha1);
1497 while (1) {
1498 int ref_length = -1;
1499 const char *ref_type = NULL;
1500
1501 buffer = read_sha1_file(actual_sha1, type, &isize);
1502 if (!buffer)
1503 return NULL;
1504 if (!strcmp(type, required_type)) {
1505 *size = isize;
1506 if (actual_sha1_return)
1507 hashcpy(actual_sha1_return, actual_sha1);
1508 return buffer;
1509 }
1510 /* Handle references */
1511 else if (!strcmp(type, commit_type))
1512 ref_type = "tree ";
1513 else if (!strcmp(type, tag_type))
1514 ref_type = "object ";
1515 else {
1516 free(buffer);
1517 return NULL;
1518 }
1519 ref_length = strlen(ref_type);
1520
1521 if (memcmp(buffer, ref_type, ref_length) ||
1522 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1523 free(buffer);
1524 return NULL;
1525 }
1526 free(buffer);
1527 /* Now we have the ID of the referred-to object in
1528 * actual_sha1. Check again. */
1529 }
1530}
1531
1532static void write_sha1_file_prepare(void *buf, unsigned long len,
1533 const char *type, unsigned char *sha1,
1534 unsigned char *hdr, int *hdrlen)
1535{
1536 SHA_CTX c;
1537
1538 /* Generate the header */
1539 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1540
1541 /* Sha1.. */
1542 SHA1_Init(&c);
1543 SHA1_Update(&c, hdr, *hdrlen);
1544 SHA1_Update(&c, buf, len);
1545 SHA1_Final(sha1, &c);
1546}
1547
1548/*
1549 * Link the tempfile to the final place, possibly creating the
1550 * last directory level as you do so.
1551 *
1552 * Returns the errno on failure, 0 on success.
1553 */
1554static int link_temp_to_file(const char *tmpfile, const char *filename)
1555{
1556 int ret;
1557 char *dir;
1558
1559 if (!link(tmpfile, filename))
1560 return 0;
1561
1562 /*
1563 * Try to mkdir the last path component if that failed.
1564 *
1565 * Re-try the "link()" regardless of whether the mkdir
1566 * succeeds, since a race might mean that somebody
1567 * else succeeded.
1568 */
1569 ret = errno;
1570 dir = strrchr(filename, '/');
1571 if (dir) {
1572 *dir = 0;
1573 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1574 *dir = '/';
1575 return -2;
1576 }
1577 *dir = '/';
1578 if (!link(tmpfile, filename))
1579 return 0;
1580 ret = errno;
1581 }
1582 return ret;
1583}
1584
1585/*
1586 * Move the just written object into its final resting place
1587 */
1588int move_temp_to_file(const char *tmpfile, const char *filename)
1589{
1590 int ret = link_temp_to_file(tmpfile, filename);
1591
1592 /*
1593 * Coda hack - coda doesn't like cross-directory links,
1594 * so we fall back to a rename, which will mean that it
1595 * won't be able to check collisions, but that's not a
1596 * big deal.
1597 *
1598 * The same holds for FAT formatted media.
1599 *
1600 * When this succeeds, we just return 0. We have nothing
1601 * left to unlink.
1602 */
1603 if (ret && ret != EEXIST) {
1604 if (!rename(tmpfile, filename))
1605 return 0;
1606 ret = errno;
1607 }
1608 unlink(tmpfile);
1609 if (ret) {
1610 if (ret != EEXIST) {
1611 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1612 }
1613 /* FIXME!!! Collision check here ? */
1614 }
1615
1616 return 0;
1617}
1618
1619static int write_buffer(int fd, const void *buf, size_t len)
1620{
1621 if (write_in_full(fd, buf, len) < 0)
1622 return error("file write error (%s)", strerror(errno));
1623 return 0;
1624}
1625
1626static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1627{
1628 int hdr_len;
1629 unsigned char c;
1630
1631 c = (type << 4) | (len & 15);
1632 len >>= 4;
1633 hdr_len = 1;
1634 while (len) {
1635 *hdr++ = c | 0x80;
1636 hdr_len++;
1637 c = (len & 0x7f);
1638 len >>= 7;
1639 }
1640 *hdr = c;
1641 return hdr_len;
1642}
1643
1644static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1645{
1646 int obj_type, hdr;
1647
1648 if (use_legacy_headers) {
1649 while (deflate(stream, 0) == Z_OK)
1650 /* nothing */;
1651 return;
1652 }
1653 if (!strcmp(type, blob_type))
1654 obj_type = OBJ_BLOB;
1655 else if (!strcmp(type, tree_type))
1656 obj_type = OBJ_TREE;
1657 else if (!strcmp(type, commit_type))
1658 obj_type = OBJ_COMMIT;
1659 else if (!strcmp(type, tag_type))
1660 obj_type = OBJ_TAG;
1661 else
1662 die("trying to generate bogus object of type '%s'", type);
1663 hdr = write_binary_header(stream->next_out, obj_type, len);
1664 stream->total_out = hdr;
1665 stream->next_out += hdr;
1666 stream->avail_out -= hdr;
1667}
1668
1669int hash_sha1_file(void *buf, unsigned long len, const char *type,
1670 unsigned char *sha1)
1671{
1672 unsigned char hdr[50];
1673 int hdrlen;
1674 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1675 return 0;
1676}
1677
1678int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1679{
1680 int size;
1681 unsigned char *compressed;
1682 z_stream stream;
1683 unsigned char sha1[20];
1684 char *filename;
1685 static char tmpfile[PATH_MAX];
1686 unsigned char hdr[50];
1687 int fd, hdrlen;
1688
1689 /* Normally if we have it in the pack then we do not bother writing
1690 * it out into .git/objects/??/?{38} file.
1691 */
1692 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1693 filename = sha1_file_name(sha1);
1694 if (returnsha1)
1695 hashcpy(returnsha1, sha1);
1696 if (has_sha1_file(sha1))
1697 return 0;
1698 fd = open(filename, O_RDONLY);
1699 if (fd >= 0) {
1700 /*
1701 * FIXME!!! We might do collision checking here, but we'd
1702 * need to uncompress the old file and check it. Later.
1703 */
1704 close(fd);
1705 return 0;
1706 }
1707
1708 if (errno != ENOENT) {
1709 return error("sha1 file %s: %s\n", filename, strerror(errno));
1710 }
1711
1712 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1713
1714 fd = mkstemp(tmpfile);
1715 if (fd < 0) {
1716 if (errno == EPERM)
1717 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1718 else
1719 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1720 }
1721
1722 /* Set it up */
1723 memset(&stream, 0, sizeof(stream));
1724 deflateInit(&stream, zlib_compression_level);
1725 size = 8 + deflateBound(&stream, len+hdrlen);
1726 compressed = xmalloc(size);
1727
1728 /* Compress it */
1729 stream.next_out = compressed;
1730 stream.avail_out = size;
1731
1732 /* First header.. */
1733 stream.next_in = hdr;
1734 stream.avail_in = hdrlen;
1735 setup_object_header(&stream, type, len);
1736
1737 /* Then the data itself.. */
1738 stream.next_in = buf;
1739 stream.avail_in = len;
1740 while (deflate(&stream, Z_FINISH) == Z_OK)
1741 /* nothing */;
1742 deflateEnd(&stream);
1743 size = stream.total_out;
1744
1745 if (write_buffer(fd, compressed, size) < 0)
1746 die("unable to write sha1 file");
1747 fchmod(fd, 0444);
1748 close(fd);
1749 free(compressed);
1750
1751 return move_temp_to_file(tmpfile, filename);
1752}
1753
1754/*
1755 * We need to unpack and recompress the object for writing
1756 * it out to a different file.
1757 */
1758static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1759{
1760 size_t size;
1761 z_stream stream;
1762 unsigned char *unpacked;
1763 unsigned long len;
1764 char type[20];
1765 char hdr[50];
1766 int hdrlen;
1767 void *buf;
1768
1769 /* need to unpack and recompress it by itself */
1770 unpacked = read_packed_sha1(sha1, type, &len);
1771
1772 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1773
1774 /* Set it up */
1775 memset(&stream, 0, sizeof(stream));
1776 deflateInit(&stream, zlib_compression_level);
1777 size = deflateBound(&stream, len + hdrlen);
1778 buf = xmalloc(size);
1779
1780 /* Compress it */
1781 stream.next_out = buf;
1782 stream.avail_out = size;
1783
1784 /* First header.. */
1785 stream.next_in = (void *)hdr;
1786 stream.avail_in = hdrlen;
1787 while (deflate(&stream, 0) == Z_OK)
1788 /* nothing */;
1789
1790 /* Then the data itself.. */
1791 stream.next_in = unpacked;
1792 stream.avail_in = len;
1793 while (deflate(&stream, Z_FINISH) == Z_OK)
1794 /* nothing */;
1795 deflateEnd(&stream);
1796 free(unpacked);
1797
1798 *objsize = stream.total_out;
1799 return buf;
1800}
1801
1802int write_sha1_to_fd(int fd, const unsigned char *sha1)
1803{
1804 int retval;
1805 unsigned long objsize;
1806 void *buf = map_sha1_file(sha1, &objsize);
1807
1808 if (buf) {
1809 retval = write_buffer(fd, buf, objsize);
1810 munmap(buf, objsize);
1811 return retval;
1812 }
1813
1814 buf = repack_object(sha1, &objsize);
1815 retval = write_buffer(fd, buf, objsize);
1816 free(buf);
1817 return retval;
1818}
1819
1820int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1821 size_t bufsize, size_t *bufposn)
1822{
1823 char tmpfile[PATH_MAX];
1824 int local;
1825 z_stream stream;
1826 unsigned char real_sha1[20];
1827 unsigned char discard[4096];
1828 int ret;
1829 SHA_CTX c;
1830
1831 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1832
1833 local = mkstemp(tmpfile);
1834 if (local < 0) {
1835 if (errno == EPERM)
1836 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1837 else
1838 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1839 }
1840
1841 memset(&stream, 0, sizeof(stream));
1842
1843 inflateInit(&stream);
1844
1845 SHA1_Init(&c);
1846
1847 do {
1848 ssize_t size;
1849 if (*bufposn) {
1850 stream.avail_in = *bufposn;
1851 stream.next_in = (unsigned char *) buffer;
1852 do {
1853 stream.next_out = discard;
1854 stream.avail_out = sizeof(discard);
1855 ret = inflate(&stream, Z_SYNC_FLUSH);
1856 SHA1_Update(&c, discard, sizeof(discard) -
1857 stream.avail_out);
1858 } while (stream.avail_in && ret == Z_OK);
1859 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1860 die("unable to write sha1 file");
1861 memmove(buffer, buffer + *bufposn - stream.avail_in,
1862 stream.avail_in);
1863 *bufposn = stream.avail_in;
1864 if (ret != Z_OK)
1865 break;
1866 }
1867 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1868 if (size <= 0) {
1869 close(local);
1870 unlink(tmpfile);
1871 if (!size)
1872 return error("Connection closed?");
1873 perror("Reading from connection");
1874 return -1;
1875 }
1876 *bufposn += size;
1877 } while (1);
1878 inflateEnd(&stream);
1879
1880 close(local);
1881 SHA1_Final(real_sha1, &c);
1882 if (ret != Z_STREAM_END) {
1883 unlink(tmpfile);
1884 return error("File %s corrupted", sha1_to_hex(sha1));
1885 }
1886 if (hashcmp(sha1, real_sha1)) {
1887 unlink(tmpfile);
1888 return error("File %s has bad hash", sha1_to_hex(sha1));
1889 }
1890
1891 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1892}
1893
1894int has_pack_index(const unsigned char *sha1)
1895{
1896 struct stat st;
1897 if (stat(sha1_pack_index_name(sha1), &st))
1898 return 0;
1899 return 1;
1900}
1901
1902int has_pack_file(const unsigned char *sha1)
1903{
1904 struct stat st;
1905 if (stat(sha1_pack_name(sha1), &st))
1906 return 0;
1907 return 1;
1908}
1909
1910int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1911{
1912 struct pack_entry e;
1913 return find_pack_entry(sha1, &e, ignore_packed);
1914}
1915
1916int has_sha1_file(const unsigned char *sha1)
1917{
1918 struct stat st;
1919 struct pack_entry e;
1920
1921 if (find_pack_entry(sha1, &e, NULL))
1922 return 1;
1923 return find_sha1_file(sha1, &st) ? 1 : 0;
1924}
1925
1926/*
1927 * reads from fd as long as possible into a supplied buffer of size bytes.
1928 * If necessary the buffer's size is increased using realloc()
1929 *
1930 * returns 0 if anything went fine and -1 otherwise
1931 *
1932 * NOTE: both buf and size may change, but even when -1 is returned
1933 * you still have to free() it yourself.
1934 */
1935int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1936{
1937 char* buf = *return_buf;
1938 unsigned long size = *return_size;
1939 int iret;
1940 unsigned long off = 0;
1941
1942 do {
1943 iret = xread(fd, buf + off, size - off);
1944 if (iret > 0) {
1945 off += iret;
1946 if (off == size) {
1947 size *= 2;
1948 buf = xrealloc(buf, size);
1949 }
1950 }
1951 } while (iret > 0);
1952
1953 *return_buf = buf;
1954 *return_size = off;
1955
1956 if (iret < 0)
1957 return -1;
1958 return 0;
1959}
1960
1961int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1962{
1963 unsigned long size = 4096;
1964 char *buf = xmalloc(size);
1965 int ret;
1966
1967 if (read_pipe(fd, &buf, &size)) {
1968 free(buf);
1969 return -1;
1970 }
1971
1972 if (!type)
1973 type = blob_type;
1974 if (write_object)
1975 ret = write_sha1_file(buf, size, type, sha1);
1976 else
1977 ret = hash_sha1_file(buf, size, type, sha1);
1978 free(buf);
1979 return ret;
1980}
1981
1982int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1983{
1984 unsigned long size = st->st_size;
1985 void *buf;
1986 int ret;
1987
1988 buf = "";
1989 if (size)
1990 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1991 close(fd);
1992
1993 if (!type)
1994 type = blob_type;
1995 if (write_object)
1996 ret = write_sha1_file(buf, size, type, sha1);
1997 else
1998 ret = hash_sha1_file(buf, size, type, sha1);
1999 if (size)
2000 munmap(buf, size);
2001 return ret;
2002}
2003
2004int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2005{
2006 int fd;
2007 char *target;
2008
2009 switch (st->st_mode & S_IFMT) {
2010 case S_IFREG:
2011 fd = open(path, O_RDONLY);
2012 if (fd < 0)
2013 return error("open(\"%s\"): %s", path,
2014 strerror(errno));
2015 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2016 return error("%s: failed to insert into database",
2017 path);
2018 break;
2019 case S_IFLNK:
2020 target = xmalloc(st->st_size+1);
2021 if (readlink(path, target, st->st_size+1) != st->st_size) {
2022 char *errstr = strerror(errno);
2023 free(target);
2024 return error("readlink(\"%s\"): %s", path,
2025 errstr);
2026 }
2027 if (!write_object)
2028 hash_sha1_file(target, st->st_size, blob_type, sha1);
2029 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2030 return error("%s: failed to insert into database",
2031 path);
2032 free(target);
2033 break;
2034 default:
2035 return error("%s: unsupported file type", path);
2036 }
2037 return 0;
2038}