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