[PATCH] add test for OFS_DELTA objects
[git/git.git] / sha1_file.c
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
31 const unsigned char null_sha1[20];
32
33 static unsigned int sha1_file_open_flag = O_NOATIME;
34
35 signed char hexval_table[256] = {
36 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
42 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
43 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
44 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
48 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
68 };
69
70 int 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
83 int 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
116 char * 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
134 static 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 */
156 char *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
174 char *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
199 char *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
224 struct alternate_object_database *alt_odb_list;
225 static struct alternate_object_database **alt_odb_tail;
226
227 static 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 */
244 static 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
312 static 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
349 static void read_info_alternates(const char * relative_base, int depth)
350 {
351 char *map;
352 size_t mapsz;
353 struct stat st;
354 char path[PATH_MAX];
355 int fd;
356
357 sprintf(path, "%s/info/alternates", relative_base);
358 fd = open(path, O_RDONLY);
359 if (fd < 0)
360 return;
361 if (fstat(fd, &st) || (st.st_size == 0)) {
362 close(fd);
363 return;
364 }
365 mapsz = xsize_t(st.st_size);
366 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
367 close(fd);
368
369 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
370
371 munmap(map, mapsz);
372 }
373
374 void prepare_alt_odb(void)
375 {
376 const char *alt;
377
378 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
379 if (!alt) alt = "";
380
381 if (alt_odb_tail)
382 return;
383 alt_odb_tail = &alt_odb_list;
384 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
385
386 read_info_alternates(get_object_directory(), 0);
387 }
388
389 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
390 {
391 char *name = sha1_file_name(sha1);
392 struct alternate_object_database *alt;
393
394 if (!stat(name, st))
395 return name;
396 prepare_alt_odb();
397 for (alt = alt_odb_list; alt; alt = alt->next) {
398 name = alt->name;
399 fill_sha1_path(name, sha1);
400 if (!stat(alt->base, st))
401 return alt->base;
402 }
403 return NULL;
404 }
405
406 static unsigned int pack_used_ctr;
407 static unsigned int pack_mmap_calls;
408 static unsigned int peak_pack_open_windows;
409 static unsigned int pack_open_windows;
410 static size_t peak_pack_mapped;
411 static size_t pack_mapped;
412 struct packed_git *packed_git;
413
414 void pack_report()
415 {
416 fprintf(stderr,
417 "pack_report: getpagesize() = %10" SZ_FMT "\n"
418 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
419 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
420 (size_t) getpagesize(),
421 packed_git_window_size,
422 packed_git_limit);
423 fprintf(stderr,
424 "pack_report: pack_used_ctr = %10u\n"
425 "pack_report: pack_mmap_calls = %10u\n"
426 "pack_report: pack_open_windows = %10u / %10u\n"
427 "pack_report: pack_mapped = "
428 "%10" SZ_FMT " / %10" SZ_FMT "\n",
429 pack_used_ctr,
430 pack_mmap_calls,
431 pack_open_windows, peak_pack_open_windows,
432 pack_mapped, peak_pack_mapped);
433 }
434
435 static int check_packed_git_idx(const char *path,
436 unsigned long *idx_size_,
437 void **idx_map_)
438 {
439 void *idx_map;
440 uint32_t *index;
441 size_t idx_size;
442 uint32_t nr, i;
443 int fd = open(path, O_RDONLY);
444 struct stat st;
445 if (fd < 0)
446 return -1;
447 if (fstat(fd, &st)) {
448 close(fd);
449 return -1;
450 }
451 idx_size = xsize_t(st.st_size);
452 if (idx_size < 4 * 256 + 20 + 20) {
453 close(fd);
454 return error("index file %s is too small", path);
455 }
456 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
457 close(fd);
458
459 index = idx_map;
460 *idx_map_ = idx_map;
461 *idx_size_ = idx_size;
462
463 /* a future index format would start with this, as older git
464 * binaries would fail the non-monotonic index check below.
465 * give a nicer warning to the user if we can.
466 */
467 if (index[0] == htonl(PACK_IDX_SIGNATURE)) {
468 munmap(idx_map, idx_size);
469 return error("index file %s is a newer version"
470 " and is not supported by this binary"
471 " (try upgrading GIT to a newer version)",
472 path);
473 }
474
475 nr = 0;
476 for (i = 0; i < 256; i++) {
477 uint32_t n = ntohl(index[i]);
478 if (n < nr) {
479 munmap(idx_map, idx_size);
480 return error("non-monotonic index %s", path);
481 }
482 nr = n;
483 }
484
485 /*
486 * Total size:
487 * - 256 index entries 4 bytes each
488 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
489 * - 20-byte SHA1 of the packfile
490 * - 20-byte SHA1 file checksum
491 */
492 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
493 munmap(idx_map, idx_size);
494 return error("wrong index file size in %s", path);
495 }
496
497 return 0;
498 }
499
500 static void scan_windows(struct packed_git *p,
501 struct packed_git **lru_p,
502 struct pack_window **lru_w,
503 struct pack_window **lru_l)
504 {
505 struct pack_window *w, *w_l;
506
507 for (w_l = NULL, w = p->windows; w; w = w->next) {
508 if (!w->inuse_cnt) {
509 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
510 *lru_p = p;
511 *lru_w = w;
512 *lru_l = w_l;
513 }
514 }
515 w_l = w;
516 }
517 }
518
519 static int unuse_one_window(struct packed_git *current)
520 {
521 struct packed_git *p, *lru_p = NULL;
522 struct pack_window *lru_w = NULL, *lru_l = NULL;
523
524 if (current)
525 scan_windows(current, &lru_p, &lru_w, &lru_l);
526 for (p = packed_git; p; p = p->next)
527 scan_windows(p, &lru_p, &lru_w, &lru_l);
528 if (lru_p) {
529 munmap(lru_w->base, lru_w->len);
530 pack_mapped -= lru_w->len;
531 if (lru_l)
532 lru_l->next = lru_w->next;
533 else {
534 lru_p->windows = lru_w->next;
535 if (!lru_p->windows && lru_p != current) {
536 close(lru_p->pack_fd);
537 lru_p->pack_fd = -1;
538 }
539 }
540 free(lru_w);
541 pack_open_windows--;
542 return 1;
543 }
544 return 0;
545 }
546
547 void release_pack_memory(size_t need)
548 {
549 size_t cur = pack_mapped;
550 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
551 ; /* nothing */
552 }
553
554 void unuse_pack(struct pack_window **w_cursor)
555 {
556 struct pack_window *w = *w_cursor;
557 if (w) {
558 w->inuse_cnt--;
559 *w_cursor = NULL;
560 }
561 }
562
563 /*
564 * Do not call this directly as this leaks p->pack_fd on error return;
565 * call open_packed_git() instead.
566 */
567 static int open_packed_git_1(struct packed_git *p)
568 {
569 struct stat st;
570 struct pack_header hdr;
571 unsigned char sha1[20];
572 unsigned char *idx_sha1;
573 long fd_flag;
574
575 p->pack_fd = open(p->pack_name, O_RDONLY);
576 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
577 return -1;
578
579 /* If we created the struct before we had the pack we lack size. */
580 if (!p->pack_size) {
581 if (!S_ISREG(st.st_mode))
582 return error("packfile %s not a regular file", p->pack_name);
583 p->pack_size = st.st_size;
584 } else if (p->pack_size != st.st_size)
585 return error("packfile %s size changed", p->pack_name);
586
587 /* We leave these file descriptors open with sliding mmap;
588 * there is no point keeping them open across exec(), though.
589 */
590 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
591 if (fd_flag < 0)
592 return error("cannot determine file descriptor flags");
593 fd_flag |= FD_CLOEXEC;
594 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
595 return error("cannot set FD_CLOEXEC");
596
597 /* Verify we recognize this pack file format. */
598 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
599 return error("file %s is far too short to be a packfile", p->pack_name);
600 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
601 return error("file %s is not a GIT packfile", p->pack_name);
602 if (!pack_version_ok(hdr.hdr_version))
603 return error("packfile %s is version %u and not supported"
604 " (try upgrading GIT to a newer version)",
605 p->pack_name, ntohl(hdr.hdr_version));
606
607 /* Verify the pack matches its index. */
608 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
609 return error("packfile %s claims to have %u objects"
610 " while index size indicates %u objects",
611 p->pack_name, ntohl(hdr.hdr_entries),
612 num_packed_objects(p));
613 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
614 return error("end of packfile %s is unavailable", p->pack_name);
615 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
616 return error("packfile %s signature is unavailable", p->pack_name);
617 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
618 if (hashcmp(sha1, idx_sha1))
619 return error("packfile %s does not match index", p->pack_name);
620 return 0;
621 }
622
623 static int open_packed_git(struct packed_git *p)
624 {
625 if (!open_packed_git_1(p))
626 return 0;
627 if (p->pack_fd != -1) {
628 close(p->pack_fd);
629 p->pack_fd = -1;
630 }
631 return -1;
632 }
633
634 static int in_window(struct pack_window *win, off_t offset)
635 {
636 /* We must promise at least 20 bytes (one hash) after the
637 * offset is available from this window, otherwise the offset
638 * is not actually in this window and a different window (which
639 * has that one hash excess) must be used. This is to support
640 * the object header and delta base parsing routines below.
641 */
642 off_t win_off = win->offset;
643 return win_off <= offset
644 && (offset + 20) <= (win_off + win->len);
645 }
646
647 unsigned char* use_pack(struct packed_git *p,
648 struct pack_window **w_cursor,
649 off_t offset,
650 unsigned int *left)
651 {
652 struct pack_window *win = *w_cursor;
653
654 if (p->pack_fd == -1 && open_packed_git(p))
655 die("packfile %s cannot be accessed", p->pack_name);
656
657 /* Since packfiles end in a hash of their content and its
658 * pointless to ask for an offset into the middle of that
659 * hash, and the in_window function above wouldn't match
660 * don't allow an offset too close to the end of the file.
661 */
662 if (offset > (p->pack_size - 20))
663 die("offset beyond end of packfile (truncated pack?)");
664
665 if (!win || !in_window(win, offset)) {
666 if (win)
667 win->inuse_cnt--;
668 for (win = p->windows; win; win = win->next) {
669 if (in_window(win, offset))
670 break;
671 }
672 if (!win) {
673 size_t window_align = packed_git_window_size / 2;
674 off_t len;
675 win = xcalloc(1, sizeof(*win));
676 win->offset = (offset / window_align) * window_align;
677 len = p->pack_size - win->offset;
678 if (len > packed_git_window_size)
679 len = packed_git_window_size;
680 win->len = (size_t)len;
681 pack_mapped += win->len;
682 while (packed_git_limit < pack_mapped
683 && unuse_one_window(p))
684 ; /* nothing */
685 win->base = xmmap(NULL, win->len,
686 PROT_READ, MAP_PRIVATE,
687 p->pack_fd, win->offset);
688 if (win->base == MAP_FAILED)
689 die("packfile %s cannot be mapped: %s",
690 p->pack_name,
691 strerror(errno));
692 pack_mmap_calls++;
693 pack_open_windows++;
694 if (pack_mapped > peak_pack_mapped)
695 peak_pack_mapped = pack_mapped;
696 if (pack_open_windows > peak_pack_open_windows)
697 peak_pack_open_windows = pack_open_windows;
698 win->next = p->windows;
699 p->windows = win;
700 }
701 }
702 if (win != *w_cursor) {
703 win->last_used = pack_used_ctr++;
704 win->inuse_cnt++;
705 *w_cursor = win;
706 }
707 offset -= win->offset;
708 if (left)
709 *left = win->len - xsize_t(offset);
710 return win->base + offset;
711 }
712
713 struct packed_git *add_packed_git(char *path, int path_len, int local)
714 {
715 struct stat st;
716 struct packed_git *p;
717 unsigned long idx_size;
718 void *idx_map;
719 unsigned char sha1[20];
720
721 if (check_packed_git_idx(path, &idx_size, &idx_map))
722 return NULL;
723
724 /* do we have a corresponding .pack file? */
725 strcpy(path + path_len - 4, ".pack");
726 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
727 munmap(idx_map, idx_size);
728 return NULL;
729 }
730 /* ok, it looks sane as far as we can check without
731 * actually mapping the pack file.
732 */
733 p = xmalloc(sizeof(*p) + path_len + 2);
734 strcpy(p->pack_name, path);
735 p->index_size = idx_size;
736 p->pack_size = st.st_size;
737 p->index_base = idx_map;
738 p->next = NULL;
739 p->windows = NULL;
740 p->pack_fd = -1;
741 p->pack_local = local;
742 p->mtime = st.st_mtime;
743 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
744 hashcpy(p->sha1, sha1);
745 return p;
746 }
747
748 struct packed_git *parse_pack_index(unsigned char *sha1)
749 {
750 char *path = sha1_pack_index_name(sha1);
751 return parse_pack_index_file(sha1, path);
752 }
753
754 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
755 {
756 struct packed_git *p;
757 unsigned long idx_size;
758 void *idx_map;
759 char *path;
760
761 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
762 return NULL;
763
764 path = sha1_pack_name(sha1);
765
766 p = xmalloc(sizeof(*p) + strlen(path) + 2);
767 strcpy(p->pack_name, path);
768 p->index_size = idx_size;
769 p->pack_size = 0;
770 p->index_base = idx_map;
771 p->next = NULL;
772 p->windows = NULL;
773 p->pack_fd = -1;
774 hashcpy(p->sha1, sha1);
775 return p;
776 }
777
778 void install_packed_git(struct packed_git *pack)
779 {
780 pack->next = packed_git;
781 packed_git = pack;
782 }
783
784 static void prepare_packed_git_one(char *objdir, int local)
785 {
786 char path[PATH_MAX];
787 int len;
788 DIR *dir;
789 struct dirent *de;
790
791 sprintf(path, "%s/pack", objdir);
792 len = strlen(path);
793 dir = opendir(path);
794 if (!dir) {
795 if (errno != ENOENT)
796 error("unable to open object pack directory: %s: %s",
797 path, strerror(errno));
798 return;
799 }
800 path[len++] = '/';
801 while ((de = readdir(dir)) != NULL) {
802 int namelen = strlen(de->d_name);
803 struct packed_git *p;
804
805 if (!has_extension(de->d_name, ".idx"))
806 continue;
807
808 /* Don't reopen a pack we already have. */
809 strcpy(path + len, de->d_name);
810 for (p = packed_git; p; p = p->next) {
811 if (!memcmp(path, p->pack_name, len + namelen - 4))
812 break;
813 }
814 if (p)
815 continue;
816 /* See if it really is a valid .idx file with corresponding
817 * .pack file that we can map.
818 */
819 p = add_packed_git(path, len + namelen, local);
820 if (!p)
821 continue;
822 install_packed_git(p);
823 }
824 closedir(dir);
825 }
826
827 static int sort_pack(const void *a_, const void *b_)
828 {
829 struct packed_git *a = *((struct packed_git **)a_);
830 struct packed_git *b = *((struct packed_git **)b_);
831 int st;
832
833 /*
834 * Local packs tend to contain objects specific to our
835 * variant of the project than remote ones. In addition,
836 * remote ones could be on a network mounted filesystem.
837 * Favor local ones for these reasons.
838 */
839 st = a->pack_local - b->pack_local;
840 if (st)
841 return -st;
842
843 /*
844 * Younger packs tend to contain more recent objects,
845 * and more recent objects tend to get accessed more
846 * often.
847 */
848 if (a->mtime < b->mtime)
849 return 1;
850 else if (a->mtime == b->mtime)
851 return 0;
852 return -1;
853 }
854
855 static void rearrange_packed_git(void)
856 {
857 struct packed_git **ary, *p;
858 int i, n;
859
860 for (n = 0, p = packed_git; p; p = p->next)
861 n++;
862 if (n < 2)
863 return;
864
865 /* prepare an array of packed_git for easier sorting */
866 ary = xcalloc(n, sizeof(struct packed_git *));
867 for (n = 0, p = packed_git; p; p = p->next)
868 ary[n++] = p;
869
870 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
871
872 /* link them back again */
873 for (i = 0; i < n - 1; i++)
874 ary[i]->next = ary[i + 1];
875 ary[n - 1]->next = NULL;
876 packed_git = ary[0];
877
878 free(ary);
879 }
880
881 static int prepare_packed_git_run_once = 0;
882 void prepare_packed_git(void)
883 {
884 struct alternate_object_database *alt;
885
886 if (prepare_packed_git_run_once)
887 return;
888 prepare_packed_git_one(get_object_directory(), 1);
889 prepare_alt_odb();
890 for (alt = alt_odb_list; alt; alt = alt->next) {
891 alt->name[-1] = 0;
892 prepare_packed_git_one(alt->base, 0);
893 alt->name[-1] = '/';
894 }
895 rearrange_packed_git();
896 prepare_packed_git_run_once = 1;
897 }
898
899 void reprepare_packed_git(void)
900 {
901 prepare_packed_git_run_once = 0;
902 prepare_packed_git();
903 }
904
905 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
906 {
907 unsigned char real_sha1[20];
908 hash_sha1_file(map, size, type, real_sha1);
909 return hashcmp(sha1, real_sha1) ? -1 : 0;
910 }
911
912 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
913 {
914 struct stat st;
915 void *map;
916 int fd;
917 char *filename = find_sha1_file(sha1, &st);
918
919 if (!filename) {
920 return NULL;
921 }
922
923 fd = open(filename, O_RDONLY | sha1_file_open_flag);
924 if (fd < 0) {
925 /* See if it works without O_NOATIME */
926 switch (sha1_file_open_flag) {
927 default:
928 fd = open(filename, O_RDONLY);
929 if (fd >= 0)
930 break;
931 /* Fallthrough */
932 case 0:
933 return NULL;
934 }
935
936 /* If it failed once, it will probably fail again.
937 * Stop using O_NOATIME
938 */
939 sha1_file_open_flag = 0;
940 }
941 *size = xsize_t(st.st_size);
942 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
943 close(fd);
944 return map;
945 }
946
947 int legacy_loose_object(unsigned char *map)
948 {
949 unsigned int word;
950
951 /*
952 * Is it a zlib-compressed buffer? If so, the first byte
953 * must be 0x78 (15-bit window size, deflated), and the
954 * first 16-bit word is evenly divisible by 31
955 */
956 word = (map[0] << 8) + map[1];
957 if (map[0] == 0x78 && !(word % 31))
958 return 1;
959 else
960 return 0;
961 }
962
963 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
964 {
965 unsigned shift;
966 unsigned char c;
967 unsigned long size;
968 unsigned long used = 0;
969
970 c = buf[used++];
971 *type = (c >> 4) & 7;
972 size = c & 15;
973 shift = 4;
974 while (c & 0x80) {
975 if (len <= used)
976 return 0;
977 if (sizeof(long) * 8 <= shift)
978 return 0;
979 c = buf[used++];
980 size += (c & 0x7f) << shift;
981 shift += 7;
982 }
983 *sizep = size;
984 return used;
985 }
986
987 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
988 {
989 unsigned long size, used;
990 static const char valid_loose_object_type[8] = {
991 0, /* OBJ_EXT */
992 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
993 0, /* "delta" and others are invalid in a loose object */
994 };
995 enum object_type type;
996
997 /* Get the data stream */
998 memset(stream, 0, sizeof(*stream));
999 stream->next_in = map;
1000 stream->avail_in = mapsize;
1001 stream->next_out = buffer;
1002 stream->avail_out = bufsiz;
1003
1004 if (legacy_loose_object(map)) {
1005 inflateInit(stream);
1006 return inflate(stream, 0);
1007 }
1008
1009 used = unpack_object_header_gently(map, mapsize, &type, &size);
1010 if (!used || !valid_loose_object_type[type])
1011 return -1;
1012 map += used;
1013 mapsize -= used;
1014
1015 /* Set up the stream for the rest.. */
1016 stream->next_in = map;
1017 stream->avail_in = mapsize;
1018 inflateInit(stream);
1019
1020 /* And generate the fake traditional header */
1021 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1022 typename(type), size);
1023 return 0;
1024 }
1025
1026 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1027 {
1028 int bytes = strlen(buffer) + 1;
1029 unsigned char *buf = xmalloc(1+size);
1030 unsigned long n;
1031 int status = Z_OK;
1032
1033 n = stream->total_out - bytes;
1034 if (n > size)
1035 n = size;
1036 memcpy(buf, (char *) buffer + bytes, n);
1037 bytes = n;
1038 if (bytes < size) {
1039 stream->next_out = buf + bytes;
1040 stream->avail_out = size - bytes;
1041 while (status == Z_OK)
1042 status = inflate(stream, Z_FINISH);
1043 }
1044 buf[size] = 0;
1045 if ((status == Z_OK || status == Z_STREAM_END) && !stream->avail_in) {
1046 inflateEnd(stream);
1047 return buf;
1048 }
1049
1050 if (status < 0)
1051 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1052 else if (stream->avail_in)
1053 error("garbage at end of loose object '%s'",
1054 sha1_to_hex(sha1));
1055 free(buf);
1056 return NULL;
1057 }
1058
1059 /*
1060 * We used to just use "sscanf()", but that's actually way
1061 * too permissive for what we want to check. So do an anal
1062 * object header parse by hand.
1063 */
1064 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1065 {
1066 char type[10];
1067 int i;
1068 unsigned long size;
1069
1070 /*
1071 * The type can be at most ten bytes (including the
1072 * terminating '\0' that we add), and is followed by
1073 * a space.
1074 */
1075 i = 0;
1076 for (;;) {
1077 char c = *hdr++;
1078 if (c == ' ')
1079 break;
1080 type[i++] = c;
1081 if (i >= sizeof(type))
1082 return -1;
1083 }
1084 type[i] = 0;
1085
1086 /*
1087 * The length must follow immediately, and be in canonical
1088 * decimal format (ie "010" is not valid).
1089 */
1090 size = *hdr++ - '0';
1091 if (size > 9)
1092 return -1;
1093 if (size) {
1094 for (;;) {
1095 unsigned long c = *hdr - '0';
1096 if (c > 9)
1097 break;
1098 hdr++;
1099 size = size * 10 + c;
1100 }
1101 }
1102 *sizep = size;
1103
1104 /*
1105 * The length must be followed by a zero byte
1106 */
1107 return *hdr ? -1 : type_from_string(type);
1108 }
1109
1110 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1111 {
1112 int ret;
1113 z_stream stream;
1114 char hdr[8192];
1115
1116 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1117 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1118 return NULL;
1119
1120 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1121 }
1122
1123 static off_t get_delta_base(struct packed_git *p,
1124 struct pack_window **w_curs,
1125 off_t *curpos,
1126 enum object_type type,
1127 off_t delta_obj_offset)
1128 {
1129 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1130 off_t base_offset;
1131
1132 /* use_pack() assured us we have [base_info, base_info + 20)
1133 * as a range that we can look at without walking off the
1134 * end of the mapped window. Its actually the hash size
1135 * that is assured. An OFS_DELTA longer than the hash size
1136 * is stupid, as then a REF_DELTA would be smaller to store.
1137 */
1138 if (type == OBJ_OFS_DELTA) {
1139 unsigned used = 0;
1140 unsigned char c = base_info[used++];
1141 base_offset = c & 127;
1142 while (c & 128) {
1143 base_offset += 1;
1144 if (!base_offset || base_offset & ~(~0UL >> 7))
1145 die("offset value overflow for delta base object");
1146 c = base_info[used++];
1147 base_offset = (base_offset << 7) + (c & 127);
1148 }
1149 base_offset = delta_obj_offset - base_offset;
1150 if (base_offset >= delta_obj_offset)
1151 die("delta base offset out of bound");
1152 *curpos += used;
1153 } else if (type == OBJ_REF_DELTA) {
1154 /* The base entry _must_ be in the same pack */
1155 base_offset = find_pack_entry_one(base_info, p);
1156 if (!base_offset)
1157 die("failed to find delta-pack base object %s",
1158 sha1_to_hex(base_info));
1159 *curpos += 20;
1160 } else
1161 die("I am totally screwed");
1162 return base_offset;
1163 }
1164
1165 /* forward declaration for a mutually recursive function */
1166 static int packed_object_info(struct packed_git *p, off_t offset,
1167 unsigned long *sizep);
1168
1169 static int packed_delta_info(struct packed_git *p,
1170 struct pack_window **w_curs,
1171 off_t curpos,
1172 enum object_type type,
1173 off_t obj_offset,
1174 unsigned long *sizep)
1175 {
1176 off_t base_offset;
1177
1178 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1179 type = packed_object_info(p, base_offset, NULL);
1180
1181 /* We choose to only get the type of the base object and
1182 * ignore potentially corrupt pack file that expects the delta
1183 * based on a base with a wrong size. This saves tons of
1184 * inflate() calls.
1185 */
1186 if (sizep) {
1187 const unsigned char *data;
1188 unsigned char delta_head[20], *in;
1189 z_stream stream;
1190 int st;
1191
1192 memset(&stream, 0, sizeof(stream));
1193 stream.next_out = delta_head;
1194 stream.avail_out = sizeof(delta_head);
1195
1196 inflateInit(&stream);
1197 do {
1198 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1199 stream.next_in = in;
1200 st = inflate(&stream, Z_FINISH);
1201 curpos += stream.next_in - in;
1202 } while ((st == Z_OK || st == Z_BUF_ERROR)
1203 && stream.total_out < sizeof(delta_head));
1204 inflateEnd(&stream);
1205 if ((st != Z_STREAM_END) &&
1206 stream.total_out != sizeof(delta_head))
1207 die("delta data unpack-initial failed");
1208
1209 /* Examine the initial part of the delta to figure out
1210 * the result size.
1211 */
1212 data = delta_head;
1213
1214 /* ignore base size */
1215 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1216
1217 /* Read the result size */
1218 *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1219 }
1220
1221 return type;
1222 }
1223
1224 static int unpack_object_header(struct packed_git *p,
1225 struct pack_window **w_curs,
1226 off_t *curpos,
1227 unsigned long *sizep)
1228 {
1229 unsigned char *base;
1230 unsigned int left;
1231 unsigned long used;
1232 enum object_type type;
1233
1234 /* use_pack() assures us we have [base, base + 20) available
1235 * as a range that we can look at at. (Its actually the hash
1236 * size that is assured.) With our object header encoding
1237 * the maximum deflated object size is 2^137, which is just
1238 * insane, so we know won't exceed what we have been given.
1239 */
1240 base = use_pack(p, w_curs, *curpos, &left);
1241 used = unpack_object_header_gently(base, left, &type, sizep);
1242 if (!used)
1243 die("object offset outside of pack file");
1244 *curpos += used;
1245
1246 return type;
1247 }
1248
1249 const char *packed_object_info_detail(struct packed_git *p,
1250 off_t obj_offset,
1251 unsigned long *size,
1252 unsigned long *store_size,
1253 unsigned int *delta_chain_length,
1254 unsigned char *base_sha1)
1255 {
1256 struct pack_window *w_curs = NULL;
1257 off_t curpos;
1258 unsigned long dummy;
1259 unsigned char *next_sha1;
1260 enum object_type type;
1261
1262 *delta_chain_length = 0;
1263 curpos = obj_offset;
1264 type = unpack_object_header(p, &w_curs, &curpos, size);
1265
1266 for (;;) {
1267 switch (type) {
1268 default:
1269 die("pack %s contains unknown object type %d",
1270 p->pack_name, type);
1271 case OBJ_COMMIT:
1272 case OBJ_TREE:
1273 case OBJ_BLOB:
1274 case OBJ_TAG:
1275 *store_size = 0; /* notyet */
1276 unuse_pack(&w_curs);
1277 return typename(type);
1278 case OBJ_OFS_DELTA:
1279 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1280 if (*delta_chain_length == 0) {
1281 /* TODO: find base_sha1 as pointed by curpos */
1282 hashclr(base_sha1);
1283 }
1284 break;
1285 case OBJ_REF_DELTA:
1286 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1287 if (*delta_chain_length == 0)
1288 hashcpy(base_sha1, next_sha1);
1289 obj_offset = find_pack_entry_one(next_sha1, p);
1290 break;
1291 }
1292 (*delta_chain_length)++;
1293 curpos = obj_offset;
1294 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1295 }
1296 }
1297
1298 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1299 unsigned long *sizep)
1300 {
1301 struct pack_window *w_curs = NULL;
1302 unsigned long size;
1303 off_t curpos = obj_offset;
1304 enum object_type type;
1305
1306 type = unpack_object_header(p, &w_curs, &curpos, &size);
1307
1308 switch (type) {
1309 case OBJ_OFS_DELTA:
1310 case OBJ_REF_DELTA:
1311 type = packed_delta_info(p, &w_curs, curpos,
1312 type, obj_offset, sizep);
1313 break;
1314 case OBJ_COMMIT:
1315 case OBJ_TREE:
1316 case OBJ_BLOB:
1317 case OBJ_TAG:
1318 if (sizep)
1319 *sizep = size;
1320 break;
1321 default:
1322 die("pack %s contains unknown object type %d",
1323 p->pack_name, type);
1324 }
1325 unuse_pack(&w_curs);
1326 return type;
1327 }
1328
1329 static void *unpack_compressed_entry(struct packed_git *p,
1330 struct pack_window **w_curs,
1331 off_t curpos,
1332 unsigned long size)
1333 {
1334 int st;
1335 z_stream stream;
1336 unsigned char *buffer, *in;
1337
1338 buffer = xmalloc(size + 1);
1339 buffer[size] = 0;
1340 memset(&stream, 0, sizeof(stream));
1341 stream.next_out = buffer;
1342 stream.avail_out = size;
1343
1344 inflateInit(&stream);
1345 do {
1346 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1347 stream.next_in = in;
1348 st = inflate(&stream, Z_FINISH);
1349 curpos += stream.next_in - in;
1350 } while (st == Z_OK || st == Z_BUF_ERROR);
1351 inflateEnd(&stream);
1352 if ((st != Z_STREAM_END) || stream.total_out != size) {
1353 free(buffer);
1354 return NULL;
1355 }
1356
1357 return buffer;
1358 }
1359
1360 static void *unpack_delta_entry(struct packed_git *p,
1361 struct pack_window **w_curs,
1362 off_t curpos,
1363 unsigned long delta_size,
1364 off_t obj_offset,
1365 enum object_type *type,
1366 unsigned long *sizep)
1367 {
1368 void *delta_data, *result, *base;
1369 unsigned long base_size;
1370 off_t base_offset;
1371
1372 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1373 base = unpack_entry(p, base_offset, type, &base_size);
1374 if (!base)
1375 die("failed to read delta base object"
1376 " at %"PRIuMAX" from %s",
1377 (uintmax_t)base_offset, p->pack_name);
1378
1379 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1380 result = patch_delta(base, base_size,
1381 delta_data, delta_size,
1382 sizep);
1383 if (!result)
1384 die("failed to apply delta");
1385 free(delta_data);
1386 free(base);
1387 return result;
1388 }
1389
1390 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1391 enum object_type *type, unsigned long *sizep)
1392 {
1393 struct pack_window *w_curs = NULL;
1394 off_t curpos = obj_offset;
1395 void *data;
1396
1397 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1398 switch (*type) {
1399 case OBJ_OFS_DELTA:
1400 case OBJ_REF_DELTA:
1401 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1402 obj_offset, type, sizep);
1403 break;
1404 case OBJ_COMMIT:
1405 case OBJ_TREE:
1406 case OBJ_BLOB:
1407 case OBJ_TAG:
1408 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1409 break;
1410 default:
1411 die("unknown object type %i in %s", *type, p->pack_name);
1412 }
1413 unuse_pack(&w_curs);
1414 return data;
1415 }
1416
1417 uint32_t num_packed_objects(const struct packed_git *p)
1418 {
1419 /* See check_packed_git_idx() */
1420 return (uint32_t)((p->index_size - 20 - 20 - 4*256) / 24);
1421 }
1422
1423 int nth_packed_object_sha1(const struct packed_git *p, uint32_t n,
1424 unsigned char* sha1)
1425 {
1426 void *index = p->index_base + 256;
1427 if (num_packed_objects(p) <= n)
1428 return -1;
1429 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1430 return 0;
1431 }
1432
1433 off_t find_pack_entry_one(const unsigned char *sha1,
1434 struct packed_git *p)
1435 {
1436 uint32_t *level1_ofs = p->index_base;
1437 int hi = ntohl(level1_ofs[*sha1]);
1438 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1439 void *index = p->index_base + 256;
1440
1441 do {
1442 int mi = (lo + hi) / 2;
1443 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1444 if (!cmp)
1445 return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1446 if (cmp > 0)
1447 hi = mi;
1448 else
1449 lo = mi+1;
1450 } while (lo < hi);
1451 return 0;
1452 }
1453
1454 static int matches_pack_name(struct packed_git *p, const char *ig)
1455 {
1456 const char *last_c, *c;
1457
1458 if (!strcmp(p->pack_name, ig))
1459 return 0;
1460
1461 for (c = p->pack_name, last_c = c; *c;)
1462 if (*c == '/')
1463 last_c = ++c;
1464 else
1465 ++c;
1466 if (!strcmp(last_c, ig))
1467 return 0;
1468
1469 return 1;
1470 }
1471
1472 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1473 {
1474 struct packed_git *p;
1475 off_t offset;
1476
1477 prepare_packed_git();
1478
1479 for (p = packed_git; p; p = p->next) {
1480 if (ignore_packed) {
1481 const char **ig;
1482 for (ig = ignore_packed; *ig; ig++)
1483 if (!matches_pack_name(p, *ig))
1484 break;
1485 if (*ig)
1486 continue;
1487 }
1488 offset = find_pack_entry_one(sha1, p);
1489 if (offset) {
1490 /*
1491 * We are about to tell the caller where they can
1492 * locate the requested object. We better make
1493 * sure the packfile is still here and can be
1494 * accessed before supplying that answer, as
1495 * it may have been deleted since the index
1496 * was loaded!
1497 */
1498 if (p->pack_fd == -1 && open_packed_git(p)) {
1499 error("packfile %s cannot be accessed", p->pack_name);
1500 continue;
1501 }
1502 e->offset = offset;
1503 e->p = p;
1504 hashcpy(e->sha1, sha1);
1505 return 1;
1506 }
1507 }
1508 return 0;
1509 }
1510
1511 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1512 struct packed_git *packs)
1513 {
1514 struct packed_git *p;
1515
1516 for (p = packs; p; p = p->next) {
1517 if (find_pack_entry_one(sha1, p))
1518 return p;
1519 }
1520 return NULL;
1521
1522 }
1523
1524 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1525 {
1526 int status;
1527 unsigned long mapsize, size;
1528 void *map;
1529 z_stream stream;
1530 char hdr[32];
1531
1532 map = map_sha1_file(sha1, &mapsize);
1533 if (!map)
1534 return error("unable to find %s", sha1_to_hex(sha1));
1535 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1536 status = error("unable to unpack %s header",
1537 sha1_to_hex(sha1));
1538 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1539 status = error("unable to parse %s header", sha1_to_hex(sha1));
1540 else if (sizep)
1541 *sizep = size;
1542 inflateEnd(&stream);
1543 munmap(map, mapsize);
1544 return status;
1545 }
1546
1547 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1548 {
1549 struct pack_entry e;
1550
1551 if (!find_pack_entry(sha1, &e, NULL)) {
1552 reprepare_packed_git();
1553 if (!find_pack_entry(sha1, &e, NULL))
1554 return sha1_loose_object_info(sha1, sizep);
1555 }
1556 return packed_object_info(e.p, e.offset, sizep);
1557 }
1558
1559 static void *read_packed_sha1(const unsigned char *sha1,
1560 enum object_type *type, unsigned long *size)
1561 {
1562 struct pack_entry e;
1563
1564 if (!find_pack_entry(sha1, &e, NULL))
1565 return NULL;
1566 else
1567 return unpack_entry(e.p, e.offset, type, size);
1568 }
1569
1570 /*
1571 * This is meant to hold a *small* number of objects that you would
1572 * want read_sha1_file() to be able to return, but yet you do not want
1573 * to write them into the object store (e.g. a browse-only
1574 * application).
1575 */
1576 static struct cached_object {
1577 unsigned char sha1[20];
1578 enum object_type type;
1579 void *buf;
1580 unsigned long size;
1581 } *cached_objects;
1582 static int cached_object_nr, cached_object_alloc;
1583
1584 static struct cached_object *find_cached_object(const unsigned char *sha1)
1585 {
1586 int i;
1587 struct cached_object *co = cached_objects;
1588
1589 for (i = 0; i < cached_object_nr; i++, co++) {
1590 if (!hashcmp(co->sha1, sha1))
1591 return co;
1592 }
1593 return NULL;
1594 }
1595
1596 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1597 unsigned char *sha1)
1598 {
1599 struct cached_object *co;
1600
1601 hash_sha1_file(buf, len, typename(type), sha1);
1602 if (has_sha1_file(sha1) || find_cached_object(sha1))
1603 return 0;
1604 if (cached_object_alloc <= cached_object_nr) {
1605 cached_object_alloc = alloc_nr(cached_object_alloc);
1606 cached_objects = xrealloc(cached_objects,
1607 sizeof(*cached_objects) *
1608 cached_object_alloc);
1609 }
1610 co = &cached_objects[cached_object_nr++];
1611 co->size = len;
1612 co->type = type;
1613 co->buf = xmalloc(len);
1614 memcpy(co->buf, buf, len);
1615 hashcpy(co->sha1, sha1);
1616 return 0;
1617 }
1618
1619 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1620 unsigned long *size)
1621 {
1622 unsigned long mapsize;
1623 void *map, *buf;
1624 struct cached_object *co;
1625
1626 co = find_cached_object(sha1);
1627 if (co) {
1628 buf = xmalloc(co->size + 1);
1629 memcpy(buf, co->buf, co->size);
1630 ((char*)buf)[co->size] = 0;
1631 *type = co->type;
1632 *size = co->size;
1633 return buf;
1634 }
1635
1636 buf = read_packed_sha1(sha1, type, size);
1637 if (buf)
1638 return buf;
1639 map = map_sha1_file(sha1, &mapsize);
1640 if (map) {
1641 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1642 munmap(map, mapsize);
1643 return buf;
1644 }
1645 reprepare_packed_git();
1646 return read_packed_sha1(sha1, type, size);
1647 }
1648
1649 void *read_object_with_reference(const unsigned char *sha1,
1650 const char *required_type_name,
1651 unsigned long *size,
1652 unsigned char *actual_sha1_return)
1653 {
1654 enum object_type type, required_type;
1655 void *buffer;
1656 unsigned long isize;
1657 unsigned char actual_sha1[20];
1658
1659 required_type = type_from_string(required_type_name);
1660 hashcpy(actual_sha1, sha1);
1661 while (1) {
1662 int ref_length = -1;
1663 const char *ref_type = NULL;
1664
1665 buffer = read_sha1_file(actual_sha1, &type, &isize);
1666 if (!buffer)
1667 return NULL;
1668 if (type == required_type) {
1669 *size = isize;
1670 if (actual_sha1_return)
1671 hashcpy(actual_sha1_return, actual_sha1);
1672 return buffer;
1673 }
1674 /* Handle references */
1675 else if (type == OBJ_COMMIT)
1676 ref_type = "tree ";
1677 else if (type == OBJ_TAG)
1678 ref_type = "object ";
1679 else {
1680 free(buffer);
1681 return NULL;
1682 }
1683 ref_length = strlen(ref_type);
1684
1685 if (memcmp(buffer, ref_type, ref_length) ||
1686 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1687 free(buffer);
1688 return NULL;
1689 }
1690 free(buffer);
1691 /* Now we have the ID of the referred-to object in
1692 * actual_sha1. Check again. */
1693 }
1694 }
1695
1696 static void write_sha1_file_prepare(void *buf, unsigned long len,
1697 const char *type, unsigned char *sha1,
1698 char *hdr, int *hdrlen)
1699 {
1700 SHA_CTX c;
1701
1702 /* Generate the header */
1703 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1704
1705 /* Sha1.. */
1706 SHA1_Init(&c);
1707 SHA1_Update(&c, hdr, *hdrlen);
1708 SHA1_Update(&c, buf, len);
1709 SHA1_Final(sha1, &c);
1710 }
1711
1712 /*
1713 * Link the tempfile to the final place, possibly creating the
1714 * last directory level as you do so.
1715 *
1716 * Returns the errno on failure, 0 on success.
1717 */
1718 static int link_temp_to_file(const char *tmpfile, const char *filename)
1719 {
1720 int ret;
1721 char *dir;
1722
1723 if (!link(tmpfile, filename))
1724 return 0;
1725
1726 /*
1727 * Try to mkdir the last path component if that failed.
1728 *
1729 * Re-try the "link()" regardless of whether the mkdir
1730 * succeeds, since a race might mean that somebody
1731 * else succeeded.
1732 */
1733 ret = errno;
1734 dir = strrchr(filename, '/');
1735 if (dir) {
1736 *dir = 0;
1737 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1738 *dir = '/';
1739 return -2;
1740 }
1741 *dir = '/';
1742 if (!link(tmpfile, filename))
1743 return 0;
1744 ret = errno;
1745 }
1746 return ret;
1747 }
1748
1749 /*
1750 * Move the just written object into its final resting place
1751 */
1752 int move_temp_to_file(const char *tmpfile, const char *filename)
1753 {
1754 int ret = link_temp_to_file(tmpfile, filename);
1755
1756 /*
1757 * Coda hack - coda doesn't like cross-directory links,
1758 * so we fall back to a rename, which will mean that it
1759 * won't be able to check collisions, but that's not a
1760 * big deal.
1761 *
1762 * The same holds for FAT formatted media.
1763 *
1764 * When this succeeds, we just return 0. We have nothing
1765 * left to unlink.
1766 */
1767 if (ret && ret != EEXIST) {
1768 if (!rename(tmpfile, filename))
1769 return 0;
1770 ret = errno;
1771 }
1772 unlink(tmpfile);
1773 if (ret) {
1774 if (ret != EEXIST) {
1775 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1776 }
1777 /* FIXME!!! Collision check here ? */
1778 }
1779
1780 return 0;
1781 }
1782
1783 static int write_buffer(int fd, const void *buf, size_t len)
1784 {
1785 if (write_in_full(fd, buf, len) < 0)
1786 return error("file write error (%s)", strerror(errno));
1787 return 0;
1788 }
1789
1790 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1791 {
1792 int hdr_len;
1793 unsigned char c;
1794
1795 c = (type << 4) | (len & 15);
1796 len >>= 4;
1797 hdr_len = 1;
1798 while (len) {
1799 *hdr++ = c | 0x80;
1800 hdr_len++;
1801 c = (len & 0x7f);
1802 len >>= 7;
1803 }
1804 *hdr = c;
1805 return hdr_len;
1806 }
1807
1808 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1809 {
1810 int obj_type, hdrlen;
1811
1812 if (use_legacy_headers) {
1813 while (deflate(stream, 0) == Z_OK)
1814 /* nothing */;
1815 return;
1816 }
1817 obj_type = type_from_string(type);
1818 hdrlen = write_binary_header(stream->next_out, obj_type, len);
1819 stream->total_out = hdrlen;
1820 stream->next_out += hdrlen;
1821 stream->avail_out -= hdrlen;
1822 }
1823
1824 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1825 unsigned char *sha1)
1826 {
1827 char hdr[32];
1828 int hdrlen;
1829 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1830 return 0;
1831 }
1832
1833 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1834 {
1835 int size;
1836 unsigned char *compressed;
1837 z_stream stream;
1838 unsigned char sha1[20];
1839 char *filename;
1840 static char tmpfile[PATH_MAX];
1841 char hdr[32];
1842 int fd, hdrlen;
1843
1844 /* Normally if we have it in the pack then we do not bother writing
1845 * it out into .git/objects/??/?{38} file.
1846 */
1847 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1848 filename = sha1_file_name(sha1);
1849 if (returnsha1)
1850 hashcpy(returnsha1, sha1);
1851 if (has_sha1_file(sha1))
1852 return 0;
1853 fd = open(filename, O_RDONLY);
1854 if (fd >= 0) {
1855 /*
1856 * FIXME!!! We might do collision checking here, but we'd
1857 * need to uncompress the old file and check it. Later.
1858 */
1859 close(fd);
1860 return 0;
1861 }
1862
1863 if (errno != ENOENT) {
1864 return error("sha1 file %s: %s\n", filename, strerror(errno));
1865 }
1866
1867 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1868
1869 fd = mkstemp(tmpfile);
1870 if (fd < 0) {
1871 if (errno == EPERM)
1872 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1873 else
1874 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1875 }
1876
1877 /* Set it up */
1878 memset(&stream, 0, sizeof(stream));
1879 deflateInit(&stream, zlib_compression_level);
1880 size = 8 + deflateBound(&stream, len+hdrlen);
1881 compressed = xmalloc(size);
1882
1883 /* Compress it */
1884 stream.next_out = compressed;
1885 stream.avail_out = size;
1886
1887 /* First header.. */
1888 stream.next_in = (unsigned char *)hdr;
1889 stream.avail_in = hdrlen;
1890 setup_object_header(&stream, type, len);
1891
1892 /* Then the data itself.. */
1893 stream.next_in = buf;
1894 stream.avail_in = len;
1895 while (deflate(&stream, Z_FINISH) == Z_OK)
1896 /* nothing */;
1897 deflateEnd(&stream);
1898 size = stream.total_out;
1899
1900 if (write_buffer(fd, compressed, size) < 0)
1901 die("unable to write sha1 file");
1902 fchmod(fd, 0444);
1903 close(fd);
1904 free(compressed);
1905
1906 return move_temp_to_file(tmpfile, filename);
1907 }
1908
1909 /*
1910 * We need to unpack and recompress the object for writing
1911 * it out to a different file.
1912 */
1913 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1914 {
1915 size_t size;
1916 z_stream stream;
1917 unsigned char *unpacked;
1918 unsigned long len;
1919 enum object_type type;
1920 char hdr[32];
1921 int hdrlen;
1922 void *buf;
1923
1924 /* need to unpack and recompress it by itself */
1925 unpacked = read_packed_sha1(sha1, &type, &len);
1926 if (!unpacked)
1927 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1928
1929 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1930
1931 /* Set it up */
1932 memset(&stream, 0, sizeof(stream));
1933 deflateInit(&stream, zlib_compression_level);
1934 size = deflateBound(&stream, len + hdrlen);
1935 buf = xmalloc(size);
1936
1937 /* Compress it */
1938 stream.next_out = buf;
1939 stream.avail_out = size;
1940
1941 /* First header.. */
1942 stream.next_in = (void *)hdr;
1943 stream.avail_in = hdrlen;
1944 while (deflate(&stream, 0) == Z_OK)
1945 /* nothing */;
1946
1947 /* Then the data itself.. */
1948 stream.next_in = unpacked;
1949 stream.avail_in = len;
1950 while (deflate(&stream, Z_FINISH) == Z_OK)
1951 /* nothing */;
1952 deflateEnd(&stream);
1953 free(unpacked);
1954
1955 *objsize = stream.total_out;
1956 return buf;
1957 }
1958
1959 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1960 {
1961 int retval;
1962 unsigned long objsize;
1963 void *buf = map_sha1_file(sha1, &objsize);
1964
1965 if (buf) {
1966 retval = write_buffer(fd, buf, objsize);
1967 munmap(buf, objsize);
1968 return retval;
1969 }
1970
1971 buf = repack_object(sha1, &objsize);
1972 retval = write_buffer(fd, buf, objsize);
1973 free(buf);
1974 return retval;
1975 }
1976
1977 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1978 size_t bufsize, size_t *bufposn)
1979 {
1980 char tmpfile[PATH_MAX];
1981 int local;
1982 z_stream stream;
1983 unsigned char real_sha1[20];
1984 unsigned char discard[4096];
1985 int ret;
1986 SHA_CTX c;
1987
1988 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1989
1990 local = mkstemp(tmpfile);
1991 if (local < 0) {
1992 if (errno == EPERM)
1993 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1994 else
1995 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1996 }
1997
1998 memset(&stream, 0, sizeof(stream));
1999
2000 inflateInit(&stream);
2001
2002 SHA1_Init(&c);
2003
2004 do {
2005 ssize_t size;
2006 if (*bufposn) {
2007 stream.avail_in = *bufposn;
2008 stream.next_in = (unsigned char *) buffer;
2009 do {
2010 stream.next_out = discard;
2011 stream.avail_out = sizeof(discard);
2012 ret = inflate(&stream, Z_SYNC_FLUSH);
2013 SHA1_Update(&c, discard, sizeof(discard) -
2014 stream.avail_out);
2015 } while (stream.avail_in && ret == Z_OK);
2016 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2017 die("unable to write sha1 file");
2018 memmove(buffer, buffer + *bufposn - stream.avail_in,
2019 stream.avail_in);
2020 *bufposn = stream.avail_in;
2021 if (ret != Z_OK)
2022 break;
2023 }
2024 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2025 if (size <= 0) {
2026 close(local);
2027 unlink(tmpfile);
2028 if (!size)
2029 return error("Connection closed?");
2030 perror("Reading from connection");
2031 return -1;
2032 }
2033 *bufposn += size;
2034 } while (1);
2035 inflateEnd(&stream);
2036
2037 close(local);
2038 SHA1_Final(real_sha1, &c);
2039 if (ret != Z_STREAM_END) {
2040 unlink(tmpfile);
2041 return error("File %s corrupted", sha1_to_hex(sha1));
2042 }
2043 if (hashcmp(sha1, real_sha1)) {
2044 unlink(tmpfile);
2045 return error("File %s has bad hash", sha1_to_hex(sha1));
2046 }
2047
2048 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2049 }
2050
2051 int has_pack_index(const unsigned char *sha1)
2052 {
2053 struct stat st;
2054 if (stat(sha1_pack_index_name(sha1), &st))
2055 return 0;
2056 return 1;
2057 }
2058
2059 int has_pack_file(const unsigned char *sha1)
2060 {
2061 struct stat st;
2062 if (stat(sha1_pack_name(sha1), &st))
2063 return 0;
2064 return 1;
2065 }
2066
2067 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2068 {
2069 struct pack_entry e;
2070 return find_pack_entry(sha1, &e, ignore_packed);
2071 }
2072
2073 int has_sha1_file(const unsigned char *sha1)
2074 {
2075 struct stat st;
2076 struct pack_entry e;
2077
2078 if (find_pack_entry(sha1, &e, NULL))
2079 return 1;
2080 return find_sha1_file(sha1, &st) ? 1 : 0;
2081 }
2082
2083 /*
2084 * reads from fd as long as possible into a supplied buffer of size bytes.
2085 * If necessary the buffer's size is increased using realloc()
2086 *
2087 * returns 0 if anything went fine and -1 otherwise
2088 *
2089 * NOTE: both buf and size may change, but even when -1 is returned
2090 * you still have to free() it yourself.
2091 */
2092 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2093 {
2094 char* buf = *return_buf;
2095 unsigned long size = *return_size;
2096 int iret;
2097 unsigned long off = 0;
2098
2099 do {
2100 iret = xread(fd, buf + off, size - off);
2101 if (iret > 0) {
2102 off += iret;
2103 if (off == size) {
2104 size *= 2;
2105 buf = xrealloc(buf, size);
2106 }
2107 }
2108 } while (iret > 0);
2109
2110 *return_buf = buf;
2111 *return_size = off;
2112
2113 if (iret < 0)
2114 return -1;
2115 return 0;
2116 }
2117
2118 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2119 {
2120 unsigned long size = 4096;
2121 char *buf = xmalloc(size);
2122 int ret;
2123
2124 if (read_pipe(fd, &buf, &size)) {
2125 free(buf);
2126 return -1;
2127 }
2128
2129 if (!type)
2130 type = blob_type;
2131 if (write_object)
2132 ret = write_sha1_file(buf, size, type, sha1);
2133 else
2134 ret = hash_sha1_file(buf, size, type, sha1);
2135 free(buf);
2136 return ret;
2137 }
2138
2139 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2140 enum object_type type, const char *path)
2141 {
2142 size_t size = xsize_t(st->st_size);
2143 void *buf = NULL;
2144 int ret, re_allocated = 0;
2145
2146 if (size)
2147 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2148 close(fd);
2149
2150 if (!type)
2151 type = OBJ_BLOB;
2152
2153 /*
2154 * Convert blobs to git internal format
2155 */
2156 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2157 unsigned long nsize = size;
2158 char *nbuf = buf;
2159 if (convert_to_git(path, &nbuf, &nsize)) {
2160 if (size)
2161 munmap(buf, size);
2162 size = nsize;
2163 buf = nbuf;
2164 re_allocated = 1;
2165 }
2166 }
2167
2168 if (write_object)
2169 ret = write_sha1_file(buf, size, typename(type), sha1);
2170 else
2171 ret = hash_sha1_file(buf, size, typename(type), sha1);
2172 if (re_allocated) {
2173 free(buf);
2174 return ret;
2175 }
2176 if (size)
2177 munmap(buf, size);
2178 return ret;
2179 }
2180
2181 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2182 {
2183 int fd;
2184 char *target;
2185 size_t len;
2186
2187 switch (st->st_mode & S_IFMT) {
2188 case S_IFREG:
2189 fd = open(path, O_RDONLY);
2190 if (fd < 0)
2191 return error("open(\"%s\"): %s", path,
2192 strerror(errno));
2193 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2194 return error("%s: failed to insert into database",
2195 path);
2196 break;
2197 case S_IFLNK:
2198 len = xsize_t(st->st_size);
2199 target = xmalloc(len + 1);
2200 if (readlink(path, target, len + 1) != st->st_size) {
2201 char *errstr = strerror(errno);
2202 free(target);
2203 return error("readlink(\"%s\"): %s", path,
2204 errstr);
2205 }
2206 if (!write_object)
2207 hash_sha1_file(target, len, blob_type, sha1);
2208 else if (write_sha1_file(target, len, blob_type, sha1))
2209 return error("%s: failed to insert into database",
2210 path);
2211 free(target);
2212 break;
2213 default:
2214 return error("%s: unsupported file type", path);
2215 }
2216 return 0;
2217 }
2218
2219 int read_pack_header(int fd, struct pack_header *header)
2220 {
2221 char *c = (char*)header;
2222 ssize_t remaining = sizeof(struct pack_header);
2223 do {
2224 ssize_t r = xread(fd, c, remaining);
2225 if (r <= 0)
2226 /* "eof before pack header was fully read" */
2227 return PH_ERROR_EOF;
2228 remaining -= r;
2229 c += r;
2230 } while (remaining > 0);
2231 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2232 /* "protocol error (pack signature mismatch detected)" */
2233 return PH_ERROR_PACK_SIGNATURE;
2234 if (!pack_version_ok(header->hdr_version))
2235 /* "protocol error (pack version unsupported)" */
2236 return PH_ERROR_PROTOCOL;
2237 return 0;
2238 }