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
25const unsigned char null_sha1[20];
26
27static unsigned int sha1_file_open_flag = O_NOATIME;
28
29signed char hexval_table[256] = {
30 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
31 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
32 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
33 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
34 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
35 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
36 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
37 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
38 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
42 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
62};
63
64int get_sha1_hex(const char *hex, unsigned char *sha1)
65{
66 int i;
67 for (i = 0; i < 20; i++) {
68 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
69 if (val & ~0xff)
70 return -1;
71 *sha1++ = val;
72 hex += 2;
73 }
74 return 0;
75}
76
77int safe_create_leading_directories(char *path)
78{
79 char *pos = path;
80 struct stat st;
81
82 if (*pos == '/')
83 pos++;
84
85 while (pos) {
86 pos = strchr(pos, '/');
87 if (!pos)
88 break;
89 *pos = 0;
90 if (!stat(path, &st)) {
91 /* path exists */
92 if (!S_ISDIR(st.st_mode)) {
93 *pos = '/';
94 return -3;
95 }
96 }
97 else if (mkdir(path, 0777)) {
98 *pos = '/';
99 return -1;
100 }
101 else if (adjust_shared_perm(path)) {
102 *pos = '/';
103 return -2;
104 }
105 *pos++ = '/';
106 }
107 return 0;
108}
109
110char * sha1_to_hex(const unsigned char *sha1)
111{
112 static int bufno;
113 static char hexbuffer[4][50];
114 static const char hex[] = "0123456789abcdef";
115 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
116 int i;
117
118 for (i = 0; i < 20; i++) {
119 unsigned int val = *sha1++;
120 *buf++ = hex[val >> 4];
121 *buf++ = hex[val & 0xf];
122 }
123 *buf = '\0';
124
125 return buffer;
126}
127
128static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
129{
130 int i;
131 for (i = 0; i < 20; i++) {
132 static char hex[] = "0123456789abcdef";
133 unsigned int val = sha1[i];
134 char *pos = pathbuf + i*2 + (i > 0);
135 *pos++ = hex[val >> 4];
136 *pos = hex[val & 0xf];
137 }
138}
139
140/*
141 * NOTE! This returns a statically allocated buffer, so you have to be
142 * careful about using it. Do a "xstrdup()" if you need to save the
143 * filename.
144 *
145 * Also note that this returns the location for creating. Reading
146 * SHA1 file can happen from any alternate directory listed in the
147 * DB_ENVIRONMENT environment variable if it is not found in
148 * the primary object database.
149 */
150char *sha1_file_name(const unsigned char *sha1)
151{
152 static char *name, *base;
153
154 if (!base) {
155 const char *sha1_file_directory = get_object_directory();
156 int len = strlen(sha1_file_directory);
157 base = xmalloc(len + 60);
158 memcpy(base, sha1_file_directory, len);
159 memset(base+len, 0, 60);
160 base[len] = '/';
161 base[len+3] = '/';
162 name = base + len + 1;
163 }
164 fill_sha1_path(name, sha1);
165 return base;
166}
167
168char *sha1_pack_name(const unsigned char *sha1)
169{
170 static const char hex[] = "0123456789abcdef";
171 static char *name, *base, *buf;
172 int i;
173
174 if (!base) {
175 const char *sha1_file_directory = get_object_directory();
176 int len = strlen(sha1_file_directory);
177 base = xmalloc(len + 60);
178 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
179 name = base + len + 11;
180 }
181
182 buf = name;
183
184 for (i = 0; i < 20; i++) {
185 unsigned int val = *sha1++;
186 *buf++ = hex[val >> 4];
187 *buf++ = hex[val & 0xf];
188 }
189
190 return base;
191}
192
193char *sha1_pack_index_name(const unsigned char *sha1)
194{
195 static const char hex[] = "0123456789abcdef";
196 static char *name, *base, *buf;
197 int i;
198
199 if (!base) {
200 const char *sha1_file_directory = get_object_directory();
201 int len = strlen(sha1_file_directory);
202 base = xmalloc(len + 60);
203 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
204 name = base + len + 11;
205 }
206
207 buf = name;
208
209 for (i = 0; i < 20; i++) {
210 unsigned int val = *sha1++;
211 *buf++ = hex[val >> 4];
212 *buf++ = hex[val & 0xf];
213 }
214
215 return base;
216}
217
218struct alternate_object_database *alt_odb_list;
219static struct alternate_object_database **alt_odb_tail;
220
221static void read_info_alternates(const char * alternates, int depth);
222
223/*
224 * Prepare alternate object database registry.
225 *
226 * The variable alt_odb_list points at the list of struct
227 * alternate_object_database. The elements on this list come from
228 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
229 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
230 * whose contents is similar to that environment variable but can be
231 * LF separated. Its base points at a statically allocated buffer that
232 * contains "/the/directory/corresponding/to/.git/objects/...", while
233 * its name points just after the slash at the end of ".git/objects/"
234 * in the example above, and has enough space to hold 40-byte hex
235 * SHA1, an extra slash for the first level indirection, and the
236 * terminating NUL.
237 */
238static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
239{
240 struct stat st;
241 const char *objdir = get_object_directory();
242 struct alternate_object_database *ent;
243 struct alternate_object_database *alt;
244 /* 43 = 40-byte + 2 '/' + terminating NUL */
245 int pfxlen = len;
246 int entlen = pfxlen + 43;
247 int base_len = -1;
248
249 if (*entry != '/' && relative_base) {
250 /* Relative alt-odb */
251 if (base_len < 0)
252 base_len = strlen(relative_base) + 1;
253 entlen += base_len;
254 pfxlen += base_len;
255 }
256 ent = xmalloc(sizeof(*ent) + entlen);
257
258 if (*entry != '/' && relative_base) {
259 memcpy(ent->base, relative_base, base_len - 1);
260 ent->base[base_len - 1] = '/';
261 memcpy(ent->base + base_len, entry, len);
262 }
263 else
264 memcpy(ent->base, entry, pfxlen);
265
266 ent->name = ent->base + pfxlen + 1;
267 ent->base[pfxlen + 3] = '/';
268 ent->base[pfxlen] = ent->base[entlen-1] = 0;
269
270 /* Detect cases where alternate disappeared */
271 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
272 error("object directory %s does not exist; "
273 "check .git/objects/info/alternates.",
274 ent->base);
275 free(ent);
276 return -1;
277 }
278
279 /* Prevent the common mistake of listing the same
280 * thing twice, or object directory itself.
281 */
282 for (alt = alt_odb_list; alt; alt = alt->next) {
283 if (!memcmp(ent->base, alt->base, pfxlen)) {
284 free(ent);
285 return -1;
286 }
287 }
288 if (!memcmp(ent->base, objdir, pfxlen)) {
289 free(ent);
290 return -1;
291 }
292
293 /* add the alternate entry */
294 *alt_odb_tail = ent;
295 alt_odb_tail = &(ent->next);
296 ent->next = NULL;
297
298 /* recursively add alternates */
299 read_info_alternates(ent->base, depth + 1);
300
301 ent->base[pfxlen] = '/';
302
303 return 0;
304}
305
306static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
307 const char *relative_base, int depth)
308{
309 const char *cp, *last;
310
311 if (depth > 5) {
312 error("%s: ignoring alternate object stores, nesting too deep.",
313 relative_base);
314 return;
315 }
316
317 last = alt;
318 while (last < ep) {
319 cp = last;
320 if (cp < ep && *cp == '#') {
321 while (cp < ep && *cp != sep)
322 cp++;
323 last = cp + 1;
324 continue;
325 }
326 while (cp < ep && *cp != sep)
327 cp++;
328 if (last != cp) {
329 if ((*last != '/') && depth) {
330 error("%s: ignoring relative alternate object store %s",
331 relative_base, last);
332 } else {
333 link_alt_odb_entry(last, cp - last,
334 relative_base, depth);
335 }
336 }
337 while (cp < ep && *cp == sep)
338 cp++;
339 last = cp;
340 }
341}
342
343static void read_info_alternates(const char * relative_base, int depth)
344{
345 char *map;
346 struct stat st;
347 char path[PATH_MAX];
348 int fd;
349
350 sprintf(path, "%s/info/alternates", relative_base);
351 fd = open(path, O_RDONLY);
352 if (fd < 0)
353 return;
354 if (fstat(fd, &st) || (st.st_size == 0)) {
355 close(fd);
356 return;
357 }
358 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
359 close(fd);
360 if (map == MAP_FAILED)
361 return;
362
363 link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
364
365 munmap(map, st.st_size);
366}
367
368void prepare_alt_odb(void)
369{
370 const char *alt;
371
372 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
373 if (!alt) alt = "";
374
375 if (alt_odb_tail)
376 return;
377 alt_odb_tail = &alt_odb_list;
378 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
379
380 read_info_alternates(get_object_directory(), 0);
381}
382
383static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
384{
385 char *name = sha1_file_name(sha1);
386 struct alternate_object_database *alt;
387
388 if (!stat(name, st))
389 return name;
390 prepare_alt_odb();
391 for (alt = alt_odb_list; alt; alt = alt->next) {
392 name = alt->name;
393 fill_sha1_path(name, sha1);
394 if (!stat(alt->base, st))
395 return alt->base;
396 }
397 return NULL;
398}
399
400static unsigned int pack_used_ctr;
401static size_t pack_mapped;
402static size_t page_size;
403struct packed_git *packed_git;
404
405static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
406 void **idx_map_)
407{
408 void *idx_map;
409 unsigned int *index;
410 unsigned long idx_size;
411 int nr, i;
412 int fd = open(path, O_RDONLY);
413 struct stat st;
414 if (fd < 0)
415 return -1;
416 if (fstat(fd, &st)) {
417 close(fd);
418 return -1;
419 }
420 idx_size = st.st_size;
421 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
422 close(fd);
423 if (idx_map == MAP_FAILED)
424 return -1;
425
426 index = idx_map;
427 *idx_map_ = idx_map;
428 *idx_size_ = idx_size;
429
430 /* check index map */
431 if (idx_size < 4*256 + 20 + 20)
432 return error("index file too small");
433 nr = 0;
434 for (i = 0; i < 256; i++) {
435 unsigned int n = ntohl(index[i]);
436 if (n < nr)
437 return error("non-monotonic index");
438 nr = n;
439 }
440
441 /*
442 * Total size:
443 * - 256 index entries 4 bytes each
444 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
445 * - 20-byte SHA1 of the packfile
446 * - 20-byte SHA1 file checksum
447 */
448 if (idx_size != 4*256 + nr * 24 + 20 + 20)
449 return error("wrong index file size");
450
451 return 0;
452}
453
454static int unuse_one_window(void)
455{
456 struct packed_git *p, *lru_p = NULL;
457 struct pack_window *w, *w_l, *lru_w = NULL, *lru_l = NULL;
458
459 for (p = packed_git; p; p = p->next) {
460 for (w_l = NULL, w = p->windows; w; w = w->next) {
461 if (!w->inuse_cnt) {
462 if (!lru_w || w->last_used < lru_w->last_used) {
463 lru_p = p;
464 lru_w = w;
465 lru_l = w_l;
466 }
467 }
468 w_l = w;
469 }
470 }
471 if (lru_p) {
472 munmap(lru_w->base, lru_w->len);
473 pack_mapped -= lru_w->len;
474 if (lru_l)
475 lru_l->next = lru_w->next;
476 else {
477 lru_p->windows = lru_w->next;
478 if (!lru_p->windows) {
479 close(lru_p->pack_fd);
480 lru_p->pack_fd = -1;
481 }
482 }
483 free(lru_w);
484 return 1;
485 }
486 return 0;
487}
488
489void unuse_pack(struct pack_window **w_cursor)
490{
491 struct pack_window *w = *w_cursor;
492 if (w) {
493 w->inuse_cnt--;
494 *w_cursor = NULL;
495 }
496}
497
498static void open_packed_git(struct packed_git *p)
499{
500 struct stat st;
501 struct pack_header hdr;
502 unsigned char sha1[20];
503 unsigned char *idx_sha1;
504
505 p->pack_fd = open(p->pack_name, O_RDONLY);
506 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
507 die("packfile %s cannot be opened", p->pack_name);
508
509 /* If we created the struct before we had the pack we lack size. */
510 if (!p->pack_size) {
511 if (!S_ISREG(st.st_mode))
512 die("packfile %s not a regular file", p->pack_name);
513 p->pack_size = st.st_size;
514 } else if (p->pack_size != st.st_size)
515 die("packfile %s size changed", p->pack_name);
516
517 /* Verify we recognize this pack file format. */
518 read_or_die(p->pack_fd, &hdr, sizeof(hdr));
519 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
520 die("file %s is not a GIT packfile", p->pack_name);
521 if (!pack_version_ok(hdr.hdr_version))
522 die("packfile %s is version %u and not supported"
523 " (try upgrading GIT to a newer version)",
524 p->pack_name, ntohl(hdr.hdr_version));
525
526 /* Verify the pack matches its index. */
527 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
528 die("packfile %s claims to have %u objects"
529 " while index size indicates %u objects",
530 p->pack_name, ntohl(hdr.hdr_entries),
531 num_packed_objects(p));
532 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
533 die("end of packfile %s is unavailable", p->pack_name);
534 read_or_die(p->pack_fd, sha1, sizeof(sha1));
535 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
536 if (hashcmp(sha1, idx_sha1))
537 die("packfile %s does not match index", p->pack_name);
538}
539
540static int in_window(struct pack_window *win, unsigned long offset)
541{
542 /* We must promise at least 20 bytes (one hash) after the
543 * offset is available from this window, otherwise the offset
544 * is not actually in this window and a different window (which
545 * has that one hash excess) must be used. This is to support
546 * the object header and delta base parsing routines below.
547 */
548 off_t win_off = win->offset;
549 return win_off <= offset
550 && (offset + 20) <= (win_off + win->len);
551}
552
553unsigned char* use_pack(struct packed_git *p,
554 struct pack_window **w_cursor,
555 unsigned long offset,
556 unsigned int *left)
557{
558 struct pack_window *win = *w_cursor;
559
560 if (p->pack_fd == -1)
561 open_packed_git(p);
562
563 /* Since packfiles end in a hash of their content and its
564 * pointless to ask for an offset into the middle of that
565 * hash, and the in_window function above wouldn't match
566 * don't allow an offset too close to the end of the file.
567 */
568 if (offset > (p->pack_size - 20))
569 die("offset beyond end of packfile (truncated pack?)");
570
571 if (!win || !in_window(win, offset)) {
572 if (win)
573 win->inuse_cnt--;
574 for (win = p->windows; win; win = win->next) {
575 if (in_window(win, offset))
576 break;
577 }
578 if (!win) {
579 if (!page_size)
580 page_size = getpagesize();
581 win = xcalloc(1, sizeof(*win));
582 win->offset = (offset / page_size) * page_size;
583 win->len = p->pack_size - win->offset;
584 if (win->len > packed_git_window_size)
585 win->len = packed_git_window_size;
586 pack_mapped += win->len;
587 while (packed_git_limit < pack_mapped && unuse_one_window())
588 ; /* nothing */
589 win->base = mmap(NULL, win->len,
590 PROT_READ, MAP_PRIVATE,
591 p->pack_fd, win->offset);
592 if (win->base == MAP_FAILED)
593 die("packfile %s cannot be mapped: %s",
594 p->pack_name,
595 strerror(errno));
596 win->next = p->windows;
597 p->windows = win;
598 }
599 }
600 if (win != *w_cursor) {
601 win->last_used = pack_used_ctr++;
602 win->inuse_cnt++;
603 *w_cursor = win;
604 }
605 offset -= win->offset;
606 if (left)
607 *left = win->len - offset;
608 return win->base + offset;
609}
610
611struct packed_git *add_packed_git(char *path, int path_len, int local)
612{
613 struct stat st;
614 struct packed_git *p;
615 unsigned long idx_size;
616 void *idx_map;
617 unsigned char sha1[20];
618
619 if (check_packed_git_idx(path, &idx_size, &idx_map))
620 return NULL;
621
622 /* do we have a corresponding .pack file? */
623 strcpy(path + path_len - 4, ".pack");
624 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
625 munmap(idx_map, idx_size);
626 return NULL;
627 }
628 /* ok, it looks sane as far as we can check without
629 * actually mapping the pack file.
630 */
631 p = xmalloc(sizeof(*p) + path_len + 2);
632 strcpy(p->pack_name, path);
633 p->index_size = idx_size;
634 p->pack_size = st.st_size;
635 p->index_base = idx_map;
636 p->next = NULL;
637 p->windows = NULL;
638 p->pack_fd = -1;
639 p->pack_local = local;
640 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
641 hashcpy(p->sha1, sha1);
642 return p;
643}
644
645struct packed_git *parse_pack_index(unsigned char *sha1)
646{
647 char *path = sha1_pack_index_name(sha1);
648 return parse_pack_index_file(sha1, path);
649}
650
651struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
652{
653 struct packed_git *p;
654 unsigned long idx_size;
655 void *idx_map;
656 char *path;
657
658 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
659 return NULL;
660
661 path = sha1_pack_name(sha1);
662
663 p = xmalloc(sizeof(*p) + strlen(path) + 2);
664 strcpy(p->pack_name, path);
665 p->index_size = idx_size;
666 p->pack_size = 0;
667 p->index_base = idx_map;
668 p->next = NULL;
669 p->windows = NULL;
670 p->pack_fd = -1;
671 hashcpy(p->sha1, sha1);
672 return p;
673}
674
675void install_packed_git(struct packed_git *pack)
676{
677 pack->next = packed_git;
678 packed_git = pack;
679}
680
681static void prepare_packed_git_one(char *objdir, int local)
682{
683 char path[PATH_MAX];
684 int len;
685 DIR *dir;
686 struct dirent *de;
687
688 sprintf(path, "%s/pack", objdir);
689 len = strlen(path);
690 dir = opendir(path);
691 if (!dir) {
692 if (errno != ENOENT)
693 error("unable to open object pack directory: %s: %s",
694 path, strerror(errno));
695 return;
696 }
697 path[len++] = '/';
698 while ((de = readdir(dir)) != NULL) {
699 int namelen = strlen(de->d_name);
700 struct packed_git *p;
701
702 if (!has_extension(de->d_name, ".idx"))
703 continue;
704
705 /* we have .idx. Is it a file we can map? */
706 strcpy(path + len, de->d_name);
707 for (p = packed_git; p; p = p->next) {
708 if (!memcmp(path, p->pack_name, len + namelen - 4))
709 break;
710 }
711 if (p)
712 continue;
713 p = add_packed_git(path, len + namelen, local);
714 if (!p)
715 continue;
716 p->next = packed_git;
717 packed_git = p;
718 }
719 closedir(dir);
720}
721
722static int prepare_packed_git_run_once = 0;
723void prepare_packed_git(void)
724{
725 struct alternate_object_database *alt;
726
727 if (prepare_packed_git_run_once)
728 return;
729 prepare_packed_git_one(get_object_directory(), 1);
730 prepare_alt_odb();
731 for (alt = alt_odb_list; alt; alt = alt->next) {
732 alt->name[-1] = 0;
733 prepare_packed_git_one(alt->base, 0);
734 alt->name[-1] = '/';
735 }
736 prepare_packed_git_run_once = 1;
737}
738
739void reprepare_packed_git(void)
740{
741 prepare_packed_git_run_once = 0;
742 prepare_packed_git();
743}
744
745int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
746{
747 unsigned char real_sha1[20];
748 hash_sha1_file(map, size, type, real_sha1);
749 return hashcmp(sha1, real_sha1) ? -1 : 0;
750}
751
752void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
753{
754 struct stat st;
755 void *map;
756 int fd;
757 char *filename = find_sha1_file(sha1, &st);
758
759 if (!filename) {
760 return NULL;
761 }
762
763 fd = open(filename, O_RDONLY | sha1_file_open_flag);
764 if (fd < 0) {
765 /* See if it works without O_NOATIME */
766 switch (sha1_file_open_flag) {
767 default:
768 fd = open(filename, O_RDONLY);
769 if (fd >= 0)
770 break;
771 /* Fallthrough */
772 case 0:
773 return NULL;
774 }
775
776 /* If it failed once, it will probably fail again.
777 * Stop using O_NOATIME
778 */
779 sha1_file_open_flag = 0;
780 }
781 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
782 close(fd);
783 if (map == MAP_FAILED)
784 return NULL;
785 *size = st.st_size;
786 return map;
787}
788
789int legacy_loose_object(unsigned char *map)
790{
791 unsigned int word;
792
793 /*
794 * Is it a zlib-compressed buffer? If so, the first byte
795 * must be 0x78 (15-bit window size, deflated), and the
796 * first 16-bit word is evenly divisible by 31
797 */
798 word = (map[0] << 8) + map[1];
799 if (map[0] == 0x78 && !(word % 31))
800 return 1;
801 else
802 return 0;
803}
804
805unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
806{
807 unsigned shift;
808 unsigned char c;
809 unsigned long size;
810 unsigned long used = 0;
811
812 c = buf[used++];
813 *type = (c >> 4) & 7;
814 size = c & 15;
815 shift = 4;
816 while (c & 0x80) {
817 if (len <= used)
818 return 0;
819 if (sizeof(long) * 8 <= shift)
820 return 0;
821 c = buf[used++];
822 size += (c & 0x7f) << shift;
823 shift += 7;
824 }
825 *sizep = size;
826 return used;
827}
828
829static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
830{
831 unsigned long size, used;
832 static const char valid_loose_object_type[8] = {
833 0, /* OBJ_EXT */
834 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
835 0, /* "delta" and others are invalid in a loose object */
836 };
837 enum object_type type;
838
839 /* Get the data stream */
840 memset(stream, 0, sizeof(*stream));
841 stream->next_in = map;
842 stream->avail_in = mapsize;
843 stream->next_out = buffer;
844 stream->avail_out = bufsiz;
845
846 if (legacy_loose_object(map)) {
847 inflateInit(stream);
848 return inflate(stream, 0);
849 }
850
851 used = unpack_object_header_gently(map, mapsize, &type, &size);
852 if (!used || !valid_loose_object_type[type])
853 return -1;
854 map += used;
855 mapsize -= used;
856
857 /* Set up the stream for the rest.. */
858 stream->next_in = map;
859 stream->avail_in = mapsize;
860 inflateInit(stream);
861
862 /* And generate the fake traditional header */
863 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
864 type_names[type], size);
865 return 0;
866}
867
868static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
869{
870 int bytes = strlen(buffer) + 1;
871 unsigned char *buf = xmalloc(1+size);
872 unsigned long n;
873
874 n = stream->total_out - bytes;
875 if (n > size)
876 n = size;
877 memcpy(buf, (char *) buffer + bytes, n);
878 bytes = n;
879 if (bytes < size) {
880 stream->next_out = buf + bytes;
881 stream->avail_out = size - bytes;
882 while (inflate(stream, Z_FINISH) == Z_OK)
883 /* nothing */;
884 }
885 buf[size] = 0;
886 inflateEnd(stream);
887 return buf;
888}
889
890/*
891 * We used to just use "sscanf()", but that's actually way
892 * too permissive for what we want to check. So do an anal
893 * object header parse by hand.
894 */
895static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
896{
897 int i;
898 unsigned long size;
899
900 /*
901 * The type can be at most ten bytes (including the
902 * terminating '\0' that we add), and is followed by
903 * a space.
904 */
905 i = 10;
906 for (;;) {
907 char c = *hdr++;
908 if (c == ' ')
909 break;
910 if (!--i)
911 return -1;
912 *type++ = c;
913 }
914 *type = 0;
915
916 /*
917 * The length must follow immediately, and be in canonical
918 * decimal format (ie "010" is not valid).
919 */
920 size = *hdr++ - '0';
921 if (size > 9)
922 return -1;
923 if (size) {
924 for (;;) {
925 unsigned long c = *hdr - '0';
926 if (c > 9)
927 break;
928 hdr++;
929 size = size * 10 + c;
930 }
931 }
932 *sizep = size;
933
934 /*
935 * The length must be followed by a zero byte
936 */
937 return *hdr ? -1 : 0;
938}
939
940void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
941{
942 int ret;
943 z_stream stream;
944 char hdr[8192];
945
946 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
947 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
948 return NULL;
949
950 return unpack_sha1_rest(&stream, hdr, *size);
951}
952
953static unsigned long get_delta_base(struct packed_git *p,
954 struct pack_window **w_curs,
955 unsigned long offset,
956 enum object_type kind,
957 unsigned long delta_obj_offset,
958 unsigned long *base_obj_offset)
959{
960 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
961 unsigned long base_offset;
962
963 /* use_pack() assured us we have [base_info, base_info + 20)
964 * as a range that we can look at without walking off the
965 * end of the mapped window. Its actually the hash size
966 * that is assured. An OFS_DELTA longer than the hash size
967 * is stupid, as then a REF_DELTA would be smaller to store.
968 */
969 if (kind == OBJ_OFS_DELTA) {
970 unsigned used = 0;
971 unsigned char c = base_info[used++];
972 base_offset = c & 127;
973 while (c & 128) {
974 base_offset += 1;
975 if (!base_offset || base_offset & ~(~0UL >> 7))
976 die("offset value overflow for delta base object");
977 c = base_info[used++];
978 base_offset = (base_offset << 7) + (c & 127);
979 }
980 base_offset = delta_obj_offset - base_offset;
981 if (base_offset >= delta_obj_offset)
982 die("delta base offset out of bound");
983 offset += used;
984 } else if (kind == OBJ_REF_DELTA) {
985 /* The base entry _must_ be in the same pack */
986 base_offset = find_pack_entry_one(base_info, p);
987 if (!base_offset)
988 die("failed to find delta-pack base object %s",
989 sha1_to_hex(base_info));
990 offset += 20;
991 } else
992 die("I am totally screwed");
993 *base_obj_offset = base_offset;
994 return offset;
995}
996
997/* forward declaration for a mutually recursive function */
998static int packed_object_info(struct packed_git *p, unsigned long offset,
999 char *type, unsigned long *sizep);
1000
1001static int packed_delta_info(struct packed_git *p,
1002 struct pack_window **w_curs,
1003 unsigned long offset,
1004 enum object_type kind,
1005 unsigned long obj_offset,
1006 char *type,
1007 unsigned long *sizep)
1008{
1009 unsigned long base_offset;
1010
1011 offset = get_delta_base(p, w_curs, offset, kind,
1012 obj_offset, &base_offset);
1013
1014 /* We choose to only get the type of the base object and
1015 * ignore potentially corrupt pack file that expects the delta
1016 * based on a base with a wrong size. This saves tons of
1017 * inflate() calls.
1018 */
1019 if (packed_object_info(p, base_offset, type, NULL))
1020 die("cannot get info for delta-pack base");
1021
1022 if (sizep) {
1023 const unsigned char *data;
1024 unsigned char delta_head[20], *in;
1025 unsigned long result_size;
1026 z_stream stream;
1027 int st;
1028
1029 memset(&stream, 0, sizeof(stream));
1030 stream.next_out = delta_head;
1031 stream.avail_out = sizeof(delta_head);
1032
1033 inflateInit(&stream);
1034 do {
1035 in = use_pack(p, w_curs, offset, &stream.avail_in);
1036 stream.next_in = in;
1037 st = inflate(&stream, Z_FINISH);
1038 offset += stream.next_in - in;
1039 } while ((st == Z_OK || st == Z_BUF_ERROR)
1040 && stream.total_out < sizeof(delta_head));
1041 inflateEnd(&stream);
1042 if ((st != Z_STREAM_END) &&
1043 stream.total_out != sizeof(delta_head))
1044 die("delta data unpack-initial failed");
1045
1046 /* Examine the initial part of the delta to figure out
1047 * the result size.
1048 */
1049 data = delta_head;
1050
1051 /* ignore base size */
1052 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1053
1054 /* Read the result size */
1055 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1056 *sizep = result_size;
1057 }
1058 return 0;
1059}
1060
1061static unsigned long unpack_object_header(struct packed_git *p,
1062 struct pack_window **w_curs,
1063 unsigned long offset,
1064 enum object_type *type,
1065 unsigned long *sizep)
1066{
1067 unsigned char *base;
1068 unsigned int left;
1069 unsigned long used;
1070
1071 /* use_pack() assures us we have [base, base + 20) available
1072 * as a range that we can look at at. (Its actually the hash
1073 * size that is assurred.) With our object header encoding
1074 * the maximum deflated object size is 2^137, which is just
1075 * insane, so we know won't exceed what we have been given.
1076 */
1077 base = use_pack(p, w_curs, offset, &left);
1078 used = unpack_object_header_gently(base, left, type, sizep);
1079 if (!used)
1080 die("object offset outside of pack file");
1081
1082 return offset + used;
1083}
1084
1085void packed_object_info_detail(struct packed_git *p,
1086 unsigned long offset,
1087 char *type,
1088 unsigned long *size,
1089 unsigned long *store_size,
1090 unsigned int *delta_chain_length,
1091 unsigned char *base_sha1)
1092{
1093 struct pack_window *w_curs = NULL;
1094 unsigned long obj_offset, val;
1095 unsigned char *next_sha1;
1096 enum object_type kind;
1097
1098 *delta_chain_length = 0;
1099 obj_offset = offset;
1100 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1101
1102 for (;;) {
1103 switch (kind) {
1104 default:
1105 die("pack %s contains unknown object type %d",
1106 p->pack_name, kind);
1107 case OBJ_COMMIT:
1108 case OBJ_TREE:
1109 case OBJ_BLOB:
1110 case OBJ_TAG:
1111 strcpy(type, type_names[kind]);
1112 *store_size = 0; /* notyet */
1113 unuse_pack(&w_curs);
1114 return;
1115 case OBJ_OFS_DELTA:
1116 get_delta_base(p, &w_curs, offset, kind,
1117 obj_offset, &offset);
1118 if (*delta_chain_length == 0) {
1119 /* TODO: find base_sha1 as pointed by offset */
1120 }
1121 break;
1122 case OBJ_REF_DELTA:
1123 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1124 if (*delta_chain_length == 0)
1125 hashcpy(base_sha1, next_sha1);
1126 offset = find_pack_entry_one(next_sha1, p);
1127 break;
1128 }
1129 obj_offset = offset;
1130 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1131 (*delta_chain_length)++;
1132 }
1133}
1134
1135static int packed_object_info(struct packed_git *p, unsigned long offset,
1136 char *type, unsigned long *sizep)
1137{
1138 struct pack_window *w_curs = NULL;
1139 unsigned long size, obj_offset = offset;
1140 enum object_type kind;
1141 int r;
1142
1143 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1144
1145 switch (kind) {
1146 case OBJ_OFS_DELTA:
1147 case OBJ_REF_DELTA:
1148 r = packed_delta_info(p, &w_curs, offset, kind,
1149 obj_offset, type, sizep);
1150 unuse_pack(&w_curs);
1151 return r;
1152 case OBJ_COMMIT:
1153 case OBJ_TREE:
1154 case OBJ_BLOB:
1155 case OBJ_TAG:
1156 strcpy(type, type_names[kind]);
1157 unuse_pack(&w_curs);
1158 break;
1159 default:
1160 die("pack %s contains unknown object type %d",
1161 p->pack_name, kind);
1162 }
1163 if (sizep)
1164 *sizep = size;
1165 return 0;
1166}
1167
1168static void *unpack_compressed_entry(struct packed_git *p,
1169 struct pack_window **w_curs,
1170 unsigned long offset,
1171 unsigned long size)
1172{
1173 int st;
1174 z_stream stream;
1175 unsigned char *buffer, *in;
1176
1177 buffer = xmalloc(size + 1);
1178 buffer[size] = 0;
1179 memset(&stream, 0, sizeof(stream));
1180 stream.next_out = buffer;
1181 stream.avail_out = size;
1182
1183 inflateInit(&stream);
1184 do {
1185 in = use_pack(p, w_curs, offset, &stream.avail_in);
1186 stream.next_in = in;
1187 st = inflate(&stream, Z_FINISH);
1188 offset += stream.next_in - in;
1189 } while (st == Z_OK || st == Z_BUF_ERROR);
1190 inflateEnd(&stream);
1191 if ((st != Z_STREAM_END) || stream.total_out != size) {
1192 free(buffer);
1193 return NULL;
1194 }
1195
1196 return buffer;
1197}
1198
1199static void *unpack_delta_entry(struct packed_git *p,
1200 struct pack_window **w_curs,
1201 unsigned long offset,
1202 unsigned long delta_size,
1203 enum object_type kind,
1204 unsigned long obj_offset,
1205 char *type,
1206 unsigned long *sizep)
1207{
1208 void *delta_data, *result, *base;
1209 unsigned long result_size, base_size, base_offset;
1210
1211 offset = get_delta_base(p, w_curs, offset, kind,
1212 obj_offset, &base_offset);
1213 base = unpack_entry(p, base_offset, type, &base_size);
1214 if (!base)
1215 die("failed to read delta base object at %lu from %s",
1216 base_offset, p->pack_name);
1217
1218 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1219 result = patch_delta(base, base_size,
1220 delta_data, delta_size,
1221 &result_size);
1222 if (!result)
1223 die("failed to apply delta");
1224 free(delta_data);
1225 free(base);
1226 *sizep = result_size;
1227 return result;
1228}
1229
1230void *unpack_entry(struct packed_git *p, unsigned long offset,
1231 char *type, unsigned long *sizep)
1232{
1233 struct pack_window *w_curs = NULL;
1234 unsigned long size, obj_offset = offset;
1235 enum object_type kind;
1236 void *retval;
1237
1238 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1239 switch (kind) {
1240 case OBJ_OFS_DELTA:
1241 case OBJ_REF_DELTA:
1242 retval = unpack_delta_entry(p, &w_curs, offset, size,
1243 kind, obj_offset, type, sizep);
1244 break;
1245 case OBJ_COMMIT:
1246 case OBJ_TREE:
1247 case OBJ_BLOB:
1248 case OBJ_TAG:
1249 strcpy(type, type_names[kind]);
1250 *sizep = size;
1251 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1252 break;
1253 default:
1254 die("unknown object type %i in %s", kind, p->pack_name);
1255 }
1256 unuse_pack(&w_curs);
1257 return retval;
1258}
1259
1260int num_packed_objects(const struct packed_git *p)
1261{
1262 /* See check_packed_git_idx() */
1263 return (p->index_size - 20 - 20 - 4*256) / 24;
1264}
1265
1266int nth_packed_object_sha1(const struct packed_git *p, int n,
1267 unsigned char* sha1)
1268{
1269 void *index = p->index_base + 256;
1270 if (n < 0 || num_packed_objects(p) <= n)
1271 return -1;
1272 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1273 return 0;
1274}
1275
1276unsigned long find_pack_entry_one(const unsigned char *sha1,
1277 struct packed_git *p)
1278{
1279 unsigned int *level1_ofs = p->index_base;
1280 int hi = ntohl(level1_ofs[*sha1]);
1281 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1282 void *index = p->index_base + 256;
1283
1284 do {
1285 int mi = (lo + hi) / 2;
1286 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1287 if (!cmp)
1288 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1289 if (cmp > 0)
1290 hi = mi;
1291 else
1292 lo = mi+1;
1293 } while (lo < hi);
1294 return 0;
1295}
1296
1297static int matches_pack_name(struct packed_git *p, const char *ig)
1298{
1299 const char *last_c, *c;
1300
1301 if (!strcmp(p->pack_name, ig))
1302 return 0;
1303
1304 for (c = p->pack_name, last_c = c; *c;)
1305 if (*c == '/')
1306 last_c = ++c;
1307 else
1308 ++c;
1309 if (!strcmp(last_c, ig))
1310 return 0;
1311
1312 return 1;
1313}
1314
1315static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1316{
1317 struct packed_git *p;
1318 unsigned long offset;
1319
1320 prepare_packed_git();
1321
1322 for (p = packed_git; p; p = p->next) {
1323 if (ignore_packed) {
1324 const char **ig;
1325 for (ig = ignore_packed; *ig; ig++)
1326 if (!matches_pack_name(p, *ig))
1327 break;
1328 if (*ig)
1329 continue;
1330 }
1331 offset = find_pack_entry_one(sha1, p);
1332 if (offset) {
1333 e->offset = offset;
1334 e->p = p;
1335 hashcpy(e->sha1, sha1);
1336 return 1;
1337 }
1338 }
1339 return 0;
1340}
1341
1342struct packed_git *find_sha1_pack(const unsigned char *sha1,
1343 struct packed_git *packs)
1344{
1345 struct packed_git *p;
1346
1347 for (p = packs; p; p = p->next) {
1348 if (find_pack_entry_one(sha1, p))
1349 return p;
1350 }
1351 return NULL;
1352
1353}
1354
1355static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1356{
1357 int status;
1358 unsigned long mapsize, size;
1359 void *map;
1360 z_stream stream;
1361 char hdr[128];
1362
1363 map = map_sha1_file(sha1, &mapsize);
1364 if (!map)
1365 return error("unable to find %s", sha1_to_hex(sha1));
1366 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1367 status = error("unable to unpack %s header",
1368 sha1_to_hex(sha1));
1369 if (parse_sha1_header(hdr, type, &size) < 0)
1370 status = error("unable to parse %s header", sha1_to_hex(sha1));
1371 else {
1372 status = 0;
1373 if (sizep)
1374 *sizep = size;
1375 }
1376 inflateEnd(&stream);
1377 munmap(map, mapsize);
1378 return status;
1379}
1380
1381int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1382{
1383 struct pack_entry e;
1384
1385 if (!find_pack_entry(sha1, &e, NULL)) {
1386 reprepare_packed_git();
1387 if (!find_pack_entry(sha1, &e, NULL))
1388 return sha1_loose_object_info(sha1, type, sizep);
1389 }
1390 return packed_object_info(e.p, e.offset, type, sizep);
1391}
1392
1393static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1394{
1395 struct pack_entry e;
1396
1397 if (!find_pack_entry(sha1, &e, NULL)) {
1398 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1399 return NULL;
1400 }
1401 return unpack_entry(e.p, e.offset, type, size);
1402}
1403
1404void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1405{
1406 unsigned long mapsize;
1407 void *map, *buf;
1408 struct pack_entry e;
1409
1410 if (find_pack_entry(sha1, &e, NULL))
1411 return read_packed_sha1(sha1, type, size);
1412 map = map_sha1_file(sha1, &mapsize);
1413 if (map) {
1414 buf = unpack_sha1_file(map, mapsize, type, size);
1415 munmap(map, mapsize);
1416 return buf;
1417 }
1418 reprepare_packed_git();
1419 if (find_pack_entry(sha1, &e, NULL))
1420 return read_packed_sha1(sha1, type, size);
1421 return NULL;
1422}
1423
1424void *read_object_with_reference(const unsigned char *sha1,
1425 const char *required_type,
1426 unsigned long *size,
1427 unsigned char *actual_sha1_return)
1428{
1429 char type[20];
1430 void *buffer;
1431 unsigned long isize;
1432 unsigned char actual_sha1[20];
1433
1434 hashcpy(actual_sha1, sha1);
1435 while (1) {
1436 int ref_length = -1;
1437 const char *ref_type = NULL;
1438
1439 buffer = read_sha1_file(actual_sha1, type, &isize);
1440 if (!buffer)
1441 return NULL;
1442 if (!strcmp(type, required_type)) {
1443 *size = isize;
1444 if (actual_sha1_return)
1445 hashcpy(actual_sha1_return, actual_sha1);
1446 return buffer;
1447 }
1448 /* Handle references */
1449 else if (!strcmp(type, commit_type))
1450 ref_type = "tree ";
1451 else if (!strcmp(type, tag_type))
1452 ref_type = "object ";
1453 else {
1454 free(buffer);
1455 return NULL;
1456 }
1457 ref_length = strlen(ref_type);
1458
1459 if (memcmp(buffer, ref_type, ref_length) ||
1460 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1461 free(buffer);
1462 return NULL;
1463 }
1464 free(buffer);
1465 /* Now we have the ID of the referred-to object in
1466 * actual_sha1. Check again. */
1467 }
1468}
1469
1470static void write_sha1_file_prepare(void *buf, unsigned long len,
1471 const char *type, unsigned char *sha1,
1472 unsigned char *hdr, int *hdrlen)
1473{
1474 SHA_CTX c;
1475
1476 /* Generate the header */
1477 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1478
1479 /* Sha1.. */
1480 SHA1_Init(&c);
1481 SHA1_Update(&c, hdr, *hdrlen);
1482 SHA1_Update(&c, buf, len);
1483 SHA1_Final(sha1, &c);
1484}
1485
1486/*
1487 * Link the tempfile to the final place, possibly creating the
1488 * last directory level as you do so.
1489 *
1490 * Returns the errno on failure, 0 on success.
1491 */
1492static int link_temp_to_file(const char *tmpfile, const char *filename)
1493{
1494 int ret;
1495 char *dir;
1496
1497 if (!link(tmpfile, filename))
1498 return 0;
1499
1500 /*
1501 * Try to mkdir the last path component if that failed.
1502 *
1503 * Re-try the "link()" regardless of whether the mkdir
1504 * succeeds, since a race might mean that somebody
1505 * else succeeded.
1506 */
1507 ret = errno;
1508 dir = strrchr(filename, '/');
1509 if (dir) {
1510 *dir = 0;
1511 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1512 *dir = '/';
1513 return -2;
1514 }
1515 *dir = '/';
1516 if (!link(tmpfile, filename))
1517 return 0;
1518 ret = errno;
1519 }
1520 return ret;
1521}
1522
1523/*
1524 * Move the just written object into its final resting place
1525 */
1526int move_temp_to_file(const char *tmpfile, const char *filename)
1527{
1528 int ret = link_temp_to_file(tmpfile, filename);
1529
1530 /*
1531 * Coda hack - coda doesn't like cross-directory links,
1532 * so we fall back to a rename, which will mean that it
1533 * won't be able to check collisions, but that's not a
1534 * big deal.
1535 *
1536 * The same holds for FAT formatted media.
1537 *
1538 * When this succeeds, we just return 0. We have nothing
1539 * left to unlink.
1540 */
1541 if (ret && ret != EEXIST) {
1542 if (!rename(tmpfile, filename))
1543 return 0;
1544 ret = errno;
1545 }
1546 unlink(tmpfile);
1547 if (ret) {
1548 if (ret != EEXIST) {
1549 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1550 }
1551 /* FIXME!!! Collision check here ? */
1552 }
1553
1554 return 0;
1555}
1556
1557static int write_buffer(int fd, const void *buf, size_t len)
1558{
1559 while (len) {
1560 ssize_t size;
1561
1562 size = write(fd, buf, len);
1563 if (!size)
1564 return error("file write: disk full");
1565 if (size < 0) {
1566 if (errno == EINTR || errno == EAGAIN)
1567 continue;
1568 return error("file write error (%s)", strerror(errno));
1569 }
1570 len -= size;
1571 buf = (char *) buf + size;
1572 }
1573 return 0;
1574}
1575
1576static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1577{
1578 int hdr_len;
1579 unsigned char c;
1580
1581 c = (type << 4) | (len & 15);
1582 len >>= 4;
1583 hdr_len = 1;
1584 while (len) {
1585 *hdr++ = c | 0x80;
1586 hdr_len++;
1587 c = (len & 0x7f);
1588 len >>= 7;
1589 }
1590 *hdr = c;
1591 return hdr_len;
1592}
1593
1594static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1595{
1596 int obj_type, hdr;
1597
1598 if (use_legacy_headers) {
1599 while (deflate(stream, 0) == Z_OK)
1600 /* nothing */;
1601 return;
1602 }
1603 if (!strcmp(type, blob_type))
1604 obj_type = OBJ_BLOB;
1605 else if (!strcmp(type, tree_type))
1606 obj_type = OBJ_TREE;
1607 else if (!strcmp(type, commit_type))
1608 obj_type = OBJ_COMMIT;
1609 else if (!strcmp(type, tag_type))
1610 obj_type = OBJ_TAG;
1611 else
1612 die("trying to generate bogus object of type '%s'", type);
1613 hdr = write_binary_header(stream->next_out, obj_type, len);
1614 stream->total_out = hdr;
1615 stream->next_out += hdr;
1616 stream->avail_out -= hdr;
1617}
1618
1619int hash_sha1_file(void *buf, unsigned long len, const char *type,
1620 unsigned char *sha1)
1621{
1622 unsigned char hdr[50];
1623 int hdrlen;
1624 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1625 return 0;
1626}
1627
1628int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1629{
1630 int size;
1631 unsigned char *compressed;
1632 z_stream stream;
1633 unsigned char sha1[20];
1634 char *filename;
1635 static char tmpfile[PATH_MAX];
1636 unsigned char hdr[50];
1637 int fd, hdrlen;
1638
1639 /* Normally if we have it in the pack then we do not bother writing
1640 * it out into .git/objects/??/?{38} file.
1641 */
1642 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1643 filename = sha1_file_name(sha1);
1644 if (returnsha1)
1645 hashcpy(returnsha1, sha1);
1646 if (has_sha1_file(sha1))
1647 return 0;
1648 fd = open(filename, O_RDONLY);
1649 if (fd >= 0) {
1650 /*
1651 * FIXME!!! We might do collision checking here, but we'd
1652 * need to uncompress the old file and check it. Later.
1653 */
1654 close(fd);
1655 return 0;
1656 }
1657
1658 if (errno != ENOENT) {
1659 return error("sha1 file %s: %s\n", filename, strerror(errno));
1660 }
1661
1662 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1663
1664 fd = mkstemp(tmpfile);
1665 if (fd < 0) {
1666 if (errno == EPERM)
1667 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1668 else
1669 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1670 }
1671
1672 /* Set it up */
1673 memset(&stream, 0, sizeof(stream));
1674 deflateInit(&stream, zlib_compression_level);
1675 size = 8 + deflateBound(&stream, len+hdrlen);
1676 compressed = xmalloc(size);
1677
1678 /* Compress it */
1679 stream.next_out = compressed;
1680 stream.avail_out = size;
1681
1682 /* First header.. */
1683 stream.next_in = hdr;
1684 stream.avail_in = hdrlen;
1685 setup_object_header(&stream, type, len);
1686
1687 /* Then the data itself.. */
1688 stream.next_in = buf;
1689 stream.avail_in = len;
1690 while (deflate(&stream, Z_FINISH) == Z_OK)
1691 /* nothing */;
1692 deflateEnd(&stream);
1693 size = stream.total_out;
1694
1695 if (write_buffer(fd, compressed, size) < 0)
1696 die("unable to write sha1 file");
1697 fchmod(fd, 0444);
1698 close(fd);
1699 free(compressed);
1700
1701 return move_temp_to_file(tmpfile, filename);
1702}
1703
1704/*
1705 * We need to unpack and recompress the object for writing
1706 * it out to a different file.
1707 */
1708static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1709{
1710 size_t size;
1711 z_stream stream;
1712 unsigned char *unpacked;
1713 unsigned long len;
1714 char type[20];
1715 char hdr[50];
1716 int hdrlen;
1717 void *buf;
1718
1719 /* need to unpack and recompress it by itself */
1720 unpacked = read_packed_sha1(sha1, type, &len);
1721
1722 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1723
1724 /* Set it up */
1725 memset(&stream, 0, sizeof(stream));
1726 deflateInit(&stream, zlib_compression_level);
1727 size = deflateBound(&stream, len + hdrlen);
1728 buf = xmalloc(size);
1729
1730 /* Compress it */
1731 stream.next_out = buf;
1732 stream.avail_out = size;
1733
1734 /* First header.. */
1735 stream.next_in = (void *)hdr;
1736 stream.avail_in = hdrlen;
1737 while (deflate(&stream, 0) == Z_OK)
1738 /* nothing */;
1739
1740 /* Then the data itself.. */
1741 stream.next_in = unpacked;
1742 stream.avail_in = len;
1743 while (deflate(&stream, Z_FINISH) == Z_OK)
1744 /* nothing */;
1745 deflateEnd(&stream);
1746 free(unpacked);
1747
1748 *objsize = stream.total_out;
1749 return buf;
1750}
1751
1752int write_sha1_to_fd(int fd, const unsigned char *sha1)
1753{
1754 int retval;
1755 unsigned long objsize;
1756 void *buf = map_sha1_file(sha1, &objsize);
1757
1758 if (buf) {
1759 retval = write_buffer(fd, buf, objsize);
1760 munmap(buf, objsize);
1761 return retval;
1762 }
1763
1764 buf = repack_object(sha1, &objsize);
1765 retval = write_buffer(fd, buf, objsize);
1766 free(buf);
1767 return retval;
1768}
1769
1770int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1771 size_t bufsize, size_t *bufposn)
1772{
1773 char tmpfile[PATH_MAX];
1774 int local;
1775 z_stream stream;
1776 unsigned char real_sha1[20];
1777 unsigned char discard[4096];
1778 int ret;
1779 SHA_CTX c;
1780
1781 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1782
1783 local = mkstemp(tmpfile);
1784 if (local < 0) {
1785 if (errno == EPERM)
1786 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1787 else
1788 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1789 }
1790
1791 memset(&stream, 0, sizeof(stream));
1792
1793 inflateInit(&stream);
1794
1795 SHA1_Init(&c);
1796
1797 do {
1798 ssize_t size;
1799 if (*bufposn) {
1800 stream.avail_in = *bufposn;
1801 stream.next_in = (unsigned char *) buffer;
1802 do {
1803 stream.next_out = discard;
1804 stream.avail_out = sizeof(discard);
1805 ret = inflate(&stream, Z_SYNC_FLUSH);
1806 SHA1_Update(&c, discard, sizeof(discard) -
1807 stream.avail_out);
1808 } while (stream.avail_in && ret == Z_OK);
1809 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1810 die("unable to write sha1 file");
1811 memmove(buffer, buffer + *bufposn - stream.avail_in,
1812 stream.avail_in);
1813 *bufposn = stream.avail_in;
1814 if (ret != Z_OK)
1815 break;
1816 }
1817 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1818 if (size <= 0) {
1819 close(local);
1820 unlink(tmpfile);
1821 if (!size)
1822 return error("Connection closed?");
1823 perror("Reading from connection");
1824 return -1;
1825 }
1826 *bufposn += size;
1827 } while (1);
1828 inflateEnd(&stream);
1829
1830 close(local);
1831 SHA1_Final(real_sha1, &c);
1832 if (ret != Z_STREAM_END) {
1833 unlink(tmpfile);
1834 return error("File %s corrupted", sha1_to_hex(sha1));
1835 }
1836 if (hashcmp(sha1, real_sha1)) {
1837 unlink(tmpfile);
1838 return error("File %s has bad hash", sha1_to_hex(sha1));
1839 }
1840
1841 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1842}
1843
1844int has_pack_index(const unsigned char *sha1)
1845{
1846 struct stat st;
1847 if (stat(sha1_pack_index_name(sha1), &st))
1848 return 0;
1849 return 1;
1850}
1851
1852int has_pack_file(const unsigned char *sha1)
1853{
1854 struct stat st;
1855 if (stat(sha1_pack_name(sha1), &st))
1856 return 0;
1857 return 1;
1858}
1859
1860int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1861{
1862 struct pack_entry e;
1863 return find_pack_entry(sha1, &e, ignore_packed);
1864}
1865
1866int has_sha1_file(const unsigned char *sha1)
1867{
1868 struct stat st;
1869 struct pack_entry e;
1870
1871 if (find_pack_entry(sha1, &e, NULL))
1872 return 1;
1873 return find_sha1_file(sha1, &st) ? 1 : 0;
1874}
1875
1876/*
1877 * reads from fd as long as possible into a supplied buffer of size bytes.
1878 * If necessary the buffer's size is increased using realloc()
1879 *
1880 * returns 0 if anything went fine and -1 otherwise
1881 *
1882 * NOTE: both buf and size may change, but even when -1 is returned
1883 * you still have to free() it yourself.
1884 */
1885int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1886{
1887 char* buf = *return_buf;
1888 unsigned long size = *return_size;
1889 int iret;
1890 unsigned long off = 0;
1891
1892 do {
1893 iret = xread(fd, buf + off, size - off);
1894 if (iret > 0) {
1895 off += iret;
1896 if (off == size) {
1897 size *= 2;
1898 buf = xrealloc(buf, size);
1899 }
1900 }
1901 } while (iret > 0);
1902
1903 *return_buf = buf;
1904 *return_size = off;
1905
1906 if (iret < 0)
1907 return -1;
1908 return 0;
1909}
1910
1911int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1912{
1913 unsigned long size = 4096;
1914 char *buf = xmalloc(size);
1915 int ret;
1916
1917 if (read_pipe(fd, &buf, &size)) {
1918 free(buf);
1919 return -1;
1920 }
1921
1922 if (!type)
1923 type = blob_type;
1924 if (write_object)
1925 ret = write_sha1_file(buf, size, type, sha1);
1926 else
1927 ret = hash_sha1_file(buf, size, type, sha1);
1928 free(buf);
1929 return ret;
1930}
1931
1932int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1933{
1934 unsigned long size = st->st_size;
1935 void *buf;
1936 int ret;
1937
1938 buf = "";
1939 if (size)
1940 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1941 close(fd);
1942 if (buf == MAP_FAILED)
1943 return -1;
1944
1945 if (!type)
1946 type = blob_type;
1947 if (write_object)
1948 ret = write_sha1_file(buf, size, type, sha1);
1949 else
1950 ret = hash_sha1_file(buf, size, type, sha1);
1951 if (size)
1952 munmap(buf, size);
1953 return ret;
1954}
1955
1956int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1957{
1958 int fd;
1959 char *target;
1960
1961 switch (st->st_mode & S_IFMT) {
1962 case S_IFREG:
1963 fd = open(path, O_RDONLY);
1964 if (fd < 0)
1965 return error("open(\"%s\"): %s", path,
1966 strerror(errno));
1967 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1968 return error("%s: failed to insert into database",
1969 path);
1970 break;
1971 case S_IFLNK:
1972 target = xmalloc(st->st_size+1);
1973 if (readlink(path, target, st->st_size+1) != st->st_size) {
1974 char *errstr = strerror(errno);
1975 free(target);
1976 return error("readlink(\"%s\"): %s", path,
1977 errstr);
1978 }
1979 if (!write_object)
1980 hash_sha1_file(target, st->st_size, blob_type, sha1);
1981 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
1982 return error("%s: failed to insert into database",
1983 path);
1984 free(target);
1985 break;
1986 default:
1987 return error("%s: unsupported file type", path);
1988 }
1989 return 0;
1990}