Fix grammar nits in documentation and in code comments.
[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 #include "refs.h"
17
18 #ifndef O_NOATIME
19 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
20 #define O_NOATIME 01000000
21 #else
22 #define O_NOATIME 0
23 #endif
24 #endif
25
26 #ifdef NO_C99_FORMAT
27 #define SZ_FMT "lu"
28 static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
29 #else
30 #define SZ_FMT "zu"
31 static size_t sz_fmt(size_t s) { return s; }
32 #endif
33
34 const unsigned char null_sha1[20];
35
36 static unsigned int sha1_file_open_flag = O_NOATIME;
37
38 const signed char hexval_table[256] = {
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
71 };
72
73 int get_sha1_hex(const char *hex, unsigned char *sha1)
74 {
75 int i;
76 for (i = 0; i < 20; i++) {
77 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
78 if (val & ~0xff)
79 return -1;
80 *sha1++ = val;
81 hex += 2;
82 }
83 return 0;
84 }
85
86 int safe_create_leading_directories(char *path)
87 {
88 char *pos = path;
89 struct stat st;
90
91 if (is_absolute_path(path))
92 pos++;
93
94 while (pos) {
95 pos = strchr(pos, '/');
96 if (!pos)
97 break;
98 *pos = 0;
99 if (!stat(path, &st)) {
100 /* path exists */
101 if (!S_ISDIR(st.st_mode)) {
102 *pos = '/';
103 return -3;
104 }
105 }
106 else if (mkdir(path, 0777)) {
107 *pos = '/';
108 return -1;
109 }
110 else if (adjust_shared_perm(path)) {
111 *pos = '/';
112 return -2;
113 }
114 *pos++ = '/';
115 }
116 return 0;
117 }
118
119 char * sha1_to_hex(const unsigned char *sha1)
120 {
121 static int bufno;
122 static char hexbuffer[4][50];
123 static const char hex[] = "0123456789abcdef";
124 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
125 int i;
126
127 for (i = 0; i < 20; i++) {
128 unsigned int val = *sha1++;
129 *buf++ = hex[val >> 4];
130 *buf++ = hex[val & 0xf];
131 }
132 *buf = '\0';
133
134 return buffer;
135 }
136
137 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
138 {
139 int i;
140 for (i = 0; i < 20; i++) {
141 static char hex[] = "0123456789abcdef";
142 unsigned int val = sha1[i];
143 char *pos = pathbuf + i*2 + (i > 0);
144 *pos++ = hex[val >> 4];
145 *pos = hex[val & 0xf];
146 }
147 }
148
149 /*
150 * NOTE! This returns a statically allocated buffer, so you have to be
151 * careful about using it. Do an "xstrdup()" if you need to save the
152 * filename.
153 *
154 * Also note that this returns the location for creating. Reading
155 * SHA1 file can happen from any alternate directory listed in the
156 * DB_ENVIRONMENT environment variable if it is not found in
157 * the primary object database.
158 */
159 char *sha1_file_name(const unsigned char *sha1)
160 {
161 static char *name, *base;
162
163 if (!base) {
164 const char *sha1_file_directory = get_object_directory();
165 int len = strlen(sha1_file_directory);
166 base = xmalloc(len + 60);
167 memcpy(base, sha1_file_directory, len);
168 memset(base+len, 0, 60);
169 base[len] = '/';
170 base[len+3] = '/';
171 name = base + len + 1;
172 }
173 fill_sha1_path(name, sha1);
174 return base;
175 }
176
177 char *sha1_pack_name(const unsigned char *sha1)
178 {
179 static const char hex[] = "0123456789abcdef";
180 static char *name, *base, *buf;
181 int i;
182
183 if (!base) {
184 const char *sha1_file_directory = get_object_directory();
185 int len = strlen(sha1_file_directory);
186 base = xmalloc(len + 60);
187 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
188 name = base + len + 11;
189 }
190
191 buf = name;
192
193 for (i = 0; i < 20; i++) {
194 unsigned int val = *sha1++;
195 *buf++ = hex[val >> 4];
196 *buf++ = hex[val & 0xf];
197 }
198
199 return base;
200 }
201
202 char *sha1_pack_index_name(const unsigned char *sha1)
203 {
204 static const char hex[] = "0123456789abcdef";
205 static char *name, *base, *buf;
206 int i;
207
208 if (!base) {
209 const char *sha1_file_directory = get_object_directory();
210 int len = strlen(sha1_file_directory);
211 base = xmalloc(len + 60);
212 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
213 name = base + len + 11;
214 }
215
216 buf = name;
217
218 for (i = 0; i < 20; i++) {
219 unsigned int val = *sha1++;
220 *buf++ = hex[val >> 4];
221 *buf++ = hex[val & 0xf];
222 }
223
224 return base;
225 }
226
227 struct alternate_object_database *alt_odb_list;
228 static struct alternate_object_database **alt_odb_tail;
229
230 static void read_info_alternates(const char * alternates, int depth);
231
232 /*
233 * Prepare alternate object database registry.
234 *
235 * The variable alt_odb_list points at the list of struct
236 * alternate_object_database. The elements on this list come from
237 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
238 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
239 * whose contents is similar to that environment variable but can be
240 * LF separated. Its base points at a statically allocated buffer that
241 * contains "/the/directory/corresponding/to/.git/objects/...", while
242 * its name points just after the slash at the end of ".git/objects/"
243 * in the example above, and has enough space to hold 40-byte hex
244 * SHA1, an extra slash for the first level indirection, and the
245 * terminating NUL.
246 */
247 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
248 {
249 struct stat st;
250 const char *objdir = get_object_directory();
251 struct alternate_object_database *ent;
252 struct alternate_object_database *alt;
253 /* 43 = 40-byte + 2 '/' + terminating NUL */
254 int pfxlen = len;
255 int entlen = pfxlen + 43;
256 int base_len = -1;
257
258 if (!is_absolute_path(entry) && relative_base) {
259 /* Relative alt-odb */
260 if (base_len < 0)
261 base_len = strlen(relative_base) + 1;
262 entlen += base_len;
263 pfxlen += base_len;
264 }
265 ent = xmalloc(sizeof(*ent) + entlen);
266
267 if (!is_absolute_path(entry) && relative_base) {
268 memcpy(ent->base, relative_base, base_len - 1);
269 ent->base[base_len - 1] = '/';
270 memcpy(ent->base + base_len, entry, len);
271 }
272 else
273 memcpy(ent->base, entry, pfxlen);
274
275 ent->name = ent->base + pfxlen + 1;
276 ent->base[pfxlen + 3] = '/';
277 ent->base[pfxlen] = ent->base[entlen-1] = 0;
278
279 /* Detect cases where alternate disappeared */
280 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
281 error("object directory %s does not exist; "
282 "check .git/objects/info/alternates.",
283 ent->base);
284 free(ent);
285 return -1;
286 }
287
288 /* Prevent the common mistake of listing the same
289 * thing twice, or object directory itself.
290 */
291 for (alt = alt_odb_list; alt; alt = alt->next) {
292 if (!memcmp(ent->base, alt->base, pfxlen)) {
293 free(ent);
294 return -1;
295 }
296 }
297 if (!memcmp(ent->base, objdir, pfxlen)) {
298 free(ent);
299 return -1;
300 }
301
302 /* add the alternate entry */
303 *alt_odb_tail = ent;
304 alt_odb_tail = &(ent->next);
305 ent->next = NULL;
306
307 /* recursively add alternates */
308 read_info_alternates(ent->base, depth + 1);
309
310 ent->base[pfxlen] = '/';
311
312 return 0;
313 }
314
315 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
316 const char *relative_base, int depth)
317 {
318 const char *cp, *last;
319
320 if (depth > 5) {
321 error("%s: ignoring alternate object stores, nesting too deep.",
322 relative_base);
323 return;
324 }
325
326 last = alt;
327 while (last < ep) {
328 cp = last;
329 if (cp < ep && *cp == '#') {
330 while (cp < ep && *cp != sep)
331 cp++;
332 last = cp + 1;
333 continue;
334 }
335 while (cp < ep && *cp != sep)
336 cp++;
337 if (last != cp) {
338 if (!is_absolute_path(last) && depth) {
339 error("%s: ignoring relative alternate object store %s",
340 relative_base, last);
341 } else {
342 link_alt_odb_entry(last, cp - last,
343 relative_base, depth);
344 }
345 }
346 while (cp < ep && *cp == sep)
347 cp++;
348 last = cp;
349 }
350 }
351
352 static void read_info_alternates(const char * relative_base, int depth)
353 {
354 char *map;
355 size_t mapsz;
356 struct stat st;
357 const char alt_file_name[] = "info/alternates";
358 /* Given that relative_base is no longer than PATH_MAX,
359 ensure that "path" has enough space to append "/", the
360 file name, "info/alternates", and a trailing NUL. */
361 char path[PATH_MAX + 1 + sizeof alt_file_name];
362 int fd;
363
364 sprintf(path, "%s/%s", relative_base, alt_file_name);
365 fd = open(path, O_RDONLY);
366 if (fd < 0)
367 return;
368 if (fstat(fd, &st) || (st.st_size == 0)) {
369 close(fd);
370 return;
371 }
372 mapsz = xsize_t(st.st_size);
373 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
374 close(fd);
375
376 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
377
378 munmap(map, mapsz);
379 }
380
381 void prepare_alt_odb(void)
382 {
383 const char *alt;
384
385 if (alt_odb_tail)
386 return;
387
388 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
389 if (!alt) alt = "";
390
391 alt_odb_tail = &alt_odb_list;
392 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
393
394 read_info_alternates(get_object_directory(), 0);
395 }
396
397 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
398 {
399 char *name = sha1_file_name(sha1);
400 struct alternate_object_database *alt;
401
402 if (!stat(name, st))
403 return name;
404 prepare_alt_odb();
405 for (alt = alt_odb_list; alt; alt = alt->next) {
406 name = alt->name;
407 fill_sha1_path(name, sha1);
408 if (!stat(alt->base, st))
409 return alt->base;
410 }
411 return NULL;
412 }
413
414 static unsigned int pack_used_ctr;
415 static unsigned int pack_mmap_calls;
416 static unsigned int peak_pack_open_windows;
417 static unsigned int pack_open_windows;
418 static size_t peak_pack_mapped;
419 static size_t pack_mapped;
420 struct packed_git *packed_git;
421
422 void pack_report(void)
423 {
424 fprintf(stderr,
425 "pack_report: getpagesize() = %10" SZ_FMT "\n"
426 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
427 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
428 sz_fmt(getpagesize()),
429 sz_fmt(packed_git_window_size),
430 sz_fmt(packed_git_limit));
431 fprintf(stderr,
432 "pack_report: pack_used_ctr = %10u\n"
433 "pack_report: pack_mmap_calls = %10u\n"
434 "pack_report: pack_open_windows = %10u / %10u\n"
435 "pack_report: pack_mapped = "
436 "%10" SZ_FMT " / %10" SZ_FMT "\n",
437 pack_used_ctr,
438 pack_mmap_calls,
439 pack_open_windows, peak_pack_open_windows,
440 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
441 }
442
443 static int check_packed_git_idx(const char *path, struct packed_git *p)
444 {
445 void *idx_map;
446 struct pack_idx_header *hdr;
447 size_t idx_size;
448 uint32_t version, nr, i, *index;
449 int fd = open(path, O_RDONLY);
450 struct stat st;
451
452 if (fd < 0)
453 return -1;
454 if (fstat(fd, &st)) {
455 close(fd);
456 return -1;
457 }
458 idx_size = xsize_t(st.st_size);
459 if (idx_size < 4 * 256 + 20 + 20) {
460 close(fd);
461 return error("index file %s is too small", path);
462 }
463 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
464 close(fd);
465
466 hdr = idx_map;
467 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
468 version = ntohl(hdr->idx_version);
469 if (version < 2 || version > 2) {
470 munmap(idx_map, idx_size);
471 return error("index file %s is version %d"
472 " and is not supported by this binary"
473 " (try upgrading GIT to a newer version)",
474 path, version);
475 }
476 } else
477 version = 1;
478
479 nr = 0;
480 index = idx_map;
481 if (version > 1)
482 index += 2; /* skip index header */
483 for (i = 0; i < 256; i++) {
484 uint32_t n = ntohl(index[i]);
485 if (n < nr) {
486 munmap(idx_map, idx_size);
487 return error("non-monotonic index %s", path);
488 }
489 nr = n;
490 }
491
492 if (version == 1) {
493 /*
494 * Total size:
495 * - 256 index entries 4 bytes each
496 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
497 * - 20-byte SHA1 of the packfile
498 * - 20-byte SHA1 file checksum
499 */
500 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
501 munmap(idx_map, idx_size);
502 return error("wrong index v1 file size in %s", path);
503 }
504 } else if (version == 2) {
505 /*
506 * Minimum size:
507 * - 8 bytes of header
508 * - 256 index entries 4 bytes each
509 * - 20-byte sha1 entry * nr
510 * - 4-byte crc entry * nr
511 * - 4-byte offset entry * nr
512 * - 20-byte SHA1 of the packfile
513 * - 20-byte SHA1 file checksum
514 * And after the 4-byte offset table might be a
515 * variable sized table containing 8-byte entries
516 * for offsets larger than 2^31.
517 */
518 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
519 unsigned long max_size = min_size;
520 if (nr)
521 max_size += (nr - 1)*8;
522 if (idx_size < min_size || idx_size > max_size) {
523 munmap(idx_map, idx_size);
524 return error("wrong index v2 file size in %s", path);
525 }
526 if (idx_size != min_size &&
527 /*
528 * make sure we can deal with large pack offsets.
529 * 31-bit signed offset won't be enough, neither
530 * 32-bit unsigned one will be.
531 */
532 (sizeof(off_t) <= 4)) {
533 munmap(idx_map, idx_size);
534 return error("pack too large for current definition of off_t in %s", path);
535 }
536 }
537
538 p->index_version = version;
539 p->index_data = idx_map;
540 p->index_size = idx_size;
541 p->num_objects = nr;
542 return 0;
543 }
544
545 int open_pack_index(struct packed_git *p)
546 {
547 char *idx_name;
548 int ret;
549
550 if (p->index_data)
551 return 0;
552
553 idx_name = xstrdup(p->pack_name);
554 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
555 ret = check_packed_git_idx(idx_name, p);
556 free(idx_name);
557 return ret;
558 }
559
560 static void scan_windows(struct packed_git *p,
561 struct packed_git **lru_p,
562 struct pack_window **lru_w,
563 struct pack_window **lru_l)
564 {
565 struct pack_window *w, *w_l;
566
567 for (w_l = NULL, w = p->windows; w; w = w->next) {
568 if (!w->inuse_cnt) {
569 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
570 *lru_p = p;
571 *lru_w = w;
572 *lru_l = w_l;
573 }
574 }
575 w_l = w;
576 }
577 }
578
579 static int unuse_one_window(struct packed_git *current, int keep_fd)
580 {
581 struct packed_git *p, *lru_p = NULL;
582 struct pack_window *lru_w = NULL, *lru_l = NULL;
583
584 if (current)
585 scan_windows(current, &lru_p, &lru_w, &lru_l);
586 for (p = packed_git; p; p = p->next)
587 scan_windows(p, &lru_p, &lru_w, &lru_l);
588 if (lru_p) {
589 munmap(lru_w->base, lru_w->len);
590 pack_mapped -= lru_w->len;
591 if (lru_l)
592 lru_l->next = lru_w->next;
593 else {
594 lru_p->windows = lru_w->next;
595 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
596 close(lru_p->pack_fd);
597 lru_p->pack_fd = -1;
598 }
599 }
600 free(lru_w);
601 pack_open_windows--;
602 return 1;
603 }
604 return 0;
605 }
606
607 void release_pack_memory(size_t need, int fd)
608 {
609 size_t cur = pack_mapped;
610 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
611 ; /* nothing */
612 }
613
614 void unuse_pack(struct pack_window **w_cursor)
615 {
616 struct pack_window *w = *w_cursor;
617 if (w) {
618 w->inuse_cnt--;
619 *w_cursor = NULL;
620 }
621 }
622
623 /*
624 * Do not call this directly as this leaks p->pack_fd on error return;
625 * call open_packed_git() instead.
626 */
627 static int open_packed_git_1(struct packed_git *p)
628 {
629 struct stat st;
630 struct pack_header hdr;
631 unsigned char sha1[20];
632 unsigned char *idx_sha1;
633 long fd_flag;
634
635 if (!p->index_data && open_pack_index(p))
636 return error("packfile %s index unavailable", p->pack_name);
637
638 p->pack_fd = open(p->pack_name, O_RDONLY);
639 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
640 return -1;
641
642 /* If we created the struct before we had the pack we lack size. */
643 if (!p->pack_size) {
644 if (!S_ISREG(st.st_mode))
645 return error("packfile %s not a regular file", p->pack_name);
646 p->pack_size = st.st_size;
647 } else if (p->pack_size != st.st_size)
648 return error("packfile %s size changed", p->pack_name);
649
650 /* We leave these file descriptors open with sliding mmap;
651 * there is no point keeping them open across exec(), though.
652 */
653 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
654 if (fd_flag < 0)
655 return error("cannot determine file descriptor flags");
656 fd_flag |= FD_CLOEXEC;
657 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
658 return error("cannot set FD_CLOEXEC");
659
660 /* Verify we recognize this pack file format. */
661 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
662 return error("file %s is far too short to be a packfile", p->pack_name);
663 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
664 return error("file %s is not a GIT packfile", p->pack_name);
665 if (!pack_version_ok(hdr.hdr_version))
666 return error("packfile %s is version %u and not supported"
667 " (try upgrading GIT to a newer version)",
668 p->pack_name, ntohl(hdr.hdr_version));
669
670 /* Verify the pack matches its index. */
671 if (p->num_objects != ntohl(hdr.hdr_entries))
672 return error("packfile %s claims to have %u objects"
673 " while index indicates %u objects",
674 p->pack_name, ntohl(hdr.hdr_entries),
675 p->num_objects);
676 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
677 return error("end of packfile %s is unavailable", p->pack_name);
678 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
679 return error("packfile %s signature is unavailable", p->pack_name);
680 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
681 if (hashcmp(sha1, idx_sha1))
682 return error("packfile %s does not match index", p->pack_name);
683 return 0;
684 }
685
686 static int open_packed_git(struct packed_git *p)
687 {
688 if (!open_packed_git_1(p))
689 return 0;
690 if (p->pack_fd != -1) {
691 close(p->pack_fd);
692 p->pack_fd = -1;
693 }
694 return -1;
695 }
696
697 static int in_window(struct pack_window *win, off_t offset)
698 {
699 /* We must promise at least 20 bytes (one hash) after the
700 * offset is available from this window, otherwise the offset
701 * is not actually in this window and a different window (which
702 * has that one hash excess) must be used. This is to support
703 * the object header and delta base parsing routines below.
704 */
705 off_t win_off = win->offset;
706 return win_off <= offset
707 && (offset + 20) <= (win_off + win->len);
708 }
709
710 unsigned char* use_pack(struct packed_git *p,
711 struct pack_window **w_cursor,
712 off_t offset,
713 unsigned int *left)
714 {
715 struct pack_window *win = *w_cursor;
716
717 if (p->pack_fd == -1 && open_packed_git(p))
718 die("packfile %s cannot be accessed", p->pack_name);
719
720 /* Since packfiles end in a hash of their content and its
721 * pointless to ask for an offset into the middle of that
722 * hash, and the in_window function above wouldn't match
723 * don't allow an offset too close to the end of the file.
724 */
725 if (offset > (p->pack_size - 20))
726 die("offset beyond end of packfile (truncated pack?)");
727
728 if (!win || !in_window(win, offset)) {
729 if (win)
730 win->inuse_cnt--;
731 for (win = p->windows; win; win = win->next) {
732 if (in_window(win, offset))
733 break;
734 }
735 if (!win) {
736 size_t window_align = packed_git_window_size / 2;
737 off_t len;
738 win = xcalloc(1, sizeof(*win));
739 win->offset = (offset / window_align) * window_align;
740 len = p->pack_size - win->offset;
741 if (len > packed_git_window_size)
742 len = packed_git_window_size;
743 win->len = (size_t)len;
744 pack_mapped += win->len;
745 while (packed_git_limit < pack_mapped
746 && unuse_one_window(p, p->pack_fd))
747 ; /* nothing */
748 win->base = xmmap(NULL, win->len,
749 PROT_READ, MAP_PRIVATE,
750 p->pack_fd, win->offset);
751 if (win->base == MAP_FAILED)
752 die("packfile %s cannot be mapped: %s",
753 p->pack_name,
754 strerror(errno));
755 pack_mmap_calls++;
756 pack_open_windows++;
757 if (pack_mapped > peak_pack_mapped)
758 peak_pack_mapped = pack_mapped;
759 if (pack_open_windows > peak_pack_open_windows)
760 peak_pack_open_windows = pack_open_windows;
761 win->next = p->windows;
762 p->windows = win;
763 }
764 }
765 if (win != *w_cursor) {
766 win->last_used = pack_used_ctr++;
767 win->inuse_cnt++;
768 *w_cursor = win;
769 }
770 offset -= win->offset;
771 if (left)
772 *left = win->len - xsize_t(offset);
773 return win->base + offset;
774 }
775
776 struct packed_git *add_packed_git(const char *path, int path_len, int local)
777 {
778 struct stat st;
779 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
780
781 /*
782 * Make sure a corresponding .pack file exists and that
783 * the index looks sane.
784 */
785 path_len -= strlen(".idx");
786 if (path_len < 1)
787 return NULL;
788 memcpy(p->pack_name, path, path_len);
789 strcpy(p->pack_name + path_len, ".pack");
790 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
791 free(p);
792 return NULL;
793 }
794
795 /* ok, it looks sane as far as we can check without
796 * actually mapping the pack file.
797 */
798 p->index_version = 0;
799 p->index_data = NULL;
800 p->index_size = 0;
801 p->num_objects = 0;
802 p->pack_size = st.st_size;
803 p->next = NULL;
804 p->windows = NULL;
805 p->pack_fd = -1;
806 p->pack_local = local;
807 p->mtime = st.st_mtime;
808 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
809 hashclr(p->sha1);
810 return p;
811 }
812
813 struct packed_git *parse_pack_index(unsigned char *sha1)
814 {
815 char *path = sha1_pack_index_name(sha1);
816 return parse_pack_index_file(sha1, path);
817 }
818
819 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
820 const char *idx_path)
821 {
822 const char *path = sha1_pack_name(sha1);
823 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
824
825 if (check_packed_git_idx(idx_path, p)) {
826 free(p);
827 return NULL;
828 }
829
830 strcpy(p->pack_name, path);
831 p->pack_size = 0;
832 p->next = NULL;
833 p->windows = NULL;
834 p->pack_fd = -1;
835 hashcpy(p->sha1, sha1);
836 return p;
837 }
838
839 void install_packed_git(struct packed_git *pack)
840 {
841 pack->next = packed_git;
842 packed_git = pack;
843 }
844
845 static void prepare_packed_git_one(char *objdir, int local)
846 {
847 /* Ensure that this buffer is large enough so that we can
848 append "/pack/" without clobbering the stack even if
849 strlen(objdir) were PATH_MAX. */
850 char path[PATH_MAX + 1 + 4 + 1 + 1];
851 int len;
852 DIR *dir;
853 struct dirent *de;
854
855 sprintf(path, "%s/pack", objdir);
856 len = strlen(path);
857 dir = opendir(path);
858 if (!dir) {
859 if (errno != ENOENT)
860 error("unable to open object pack directory: %s: %s",
861 path, strerror(errno));
862 return;
863 }
864 path[len++] = '/';
865 while ((de = readdir(dir)) != NULL) {
866 int namelen = strlen(de->d_name);
867 struct packed_git *p;
868
869 if (!has_extension(de->d_name, ".idx"))
870 continue;
871
872 if (len + namelen + 1 > sizeof(path))
873 continue;
874
875 /* Don't reopen a pack we already have. */
876 strcpy(path + len, de->d_name);
877 for (p = packed_git; p; p = p->next) {
878 if (!memcmp(path, p->pack_name, len + namelen - 4))
879 break;
880 }
881 if (p)
882 continue;
883 /* See if it really is a valid .idx file with corresponding
884 * .pack file that we can map.
885 */
886 p = add_packed_git(path, len + namelen, local);
887 if (!p)
888 continue;
889 install_packed_git(p);
890 }
891 closedir(dir);
892 }
893
894 static int sort_pack(const void *a_, const void *b_)
895 {
896 struct packed_git *a = *((struct packed_git **)a_);
897 struct packed_git *b = *((struct packed_git **)b_);
898 int st;
899
900 /*
901 * Local packs tend to contain objects specific to our
902 * variant of the project than remote ones. In addition,
903 * remote ones could be on a network mounted filesystem.
904 * Favor local ones for these reasons.
905 */
906 st = a->pack_local - b->pack_local;
907 if (st)
908 return -st;
909
910 /*
911 * Younger packs tend to contain more recent objects,
912 * and more recent objects tend to get accessed more
913 * often.
914 */
915 if (a->mtime < b->mtime)
916 return 1;
917 else if (a->mtime == b->mtime)
918 return 0;
919 return -1;
920 }
921
922 static void rearrange_packed_git(void)
923 {
924 struct packed_git **ary, *p;
925 int i, n;
926
927 for (n = 0, p = packed_git; p; p = p->next)
928 n++;
929 if (n < 2)
930 return;
931
932 /* prepare an array of packed_git for easier sorting */
933 ary = xcalloc(n, sizeof(struct packed_git *));
934 for (n = 0, p = packed_git; p; p = p->next)
935 ary[n++] = p;
936
937 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
938
939 /* link them back again */
940 for (i = 0; i < n - 1; i++)
941 ary[i]->next = ary[i + 1];
942 ary[n - 1]->next = NULL;
943 packed_git = ary[0];
944
945 free(ary);
946 }
947
948 static int prepare_packed_git_run_once = 0;
949 void prepare_packed_git(void)
950 {
951 struct alternate_object_database *alt;
952
953 if (prepare_packed_git_run_once)
954 return;
955 prepare_packed_git_one(get_object_directory(), 1);
956 prepare_alt_odb();
957 for (alt = alt_odb_list; alt; alt = alt->next) {
958 alt->name[-1] = 0;
959 prepare_packed_git_one(alt->base, 0);
960 alt->name[-1] = '/';
961 }
962 rearrange_packed_git();
963 prepare_packed_git_run_once = 1;
964 }
965
966 void reprepare_packed_git(void)
967 {
968 prepare_packed_git_run_once = 0;
969 prepare_packed_git();
970 }
971
972 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
973 {
974 unsigned char real_sha1[20];
975 hash_sha1_file(map, size, type, real_sha1);
976 return hashcmp(sha1, real_sha1) ? -1 : 0;
977 }
978
979 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
980 {
981 struct stat st;
982 void *map;
983 int fd;
984 char *filename = find_sha1_file(sha1, &st);
985
986 if (!filename) {
987 return NULL;
988 }
989
990 fd = open(filename, O_RDONLY | sha1_file_open_flag);
991 if (fd < 0) {
992 /* See if it works without O_NOATIME */
993 switch (sha1_file_open_flag) {
994 default:
995 fd = open(filename, O_RDONLY);
996 if (fd >= 0)
997 break;
998 /* Fallthrough */
999 case 0:
1000 return NULL;
1001 }
1002
1003 /* If it failed once, it will probably fail again.
1004 * Stop using O_NOATIME
1005 */
1006 sha1_file_open_flag = 0;
1007 }
1008 *size = xsize_t(st.st_size);
1009 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1010 close(fd);
1011 return map;
1012 }
1013
1014 static int legacy_loose_object(unsigned char *map)
1015 {
1016 unsigned int word;
1017
1018 /*
1019 * Is it a zlib-compressed buffer? If so, the first byte
1020 * must be 0x78 (15-bit window size, deflated), and the
1021 * first 16-bit word is evenly divisible by 31
1022 */
1023 word = (map[0] << 8) + map[1];
1024 if (map[0] == 0x78 && !(word % 31))
1025 return 1;
1026 else
1027 return 0;
1028 }
1029
1030 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1031 {
1032 unsigned shift;
1033 unsigned char c;
1034 unsigned long size;
1035 unsigned long used = 0;
1036
1037 c = buf[used++];
1038 *type = (c >> 4) & 7;
1039 size = c & 15;
1040 shift = 4;
1041 while (c & 0x80) {
1042 if (len <= used)
1043 return 0;
1044 if (sizeof(long) * 8 <= shift)
1045 return 0;
1046 c = buf[used++];
1047 size += (c & 0x7f) << shift;
1048 shift += 7;
1049 }
1050 *sizep = size;
1051 return used;
1052 }
1053
1054 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1055 {
1056 unsigned long size, used;
1057 static const char valid_loose_object_type[8] = {
1058 0, /* OBJ_EXT */
1059 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1060 0, /* "delta" and others are invalid in a loose object */
1061 };
1062 enum object_type type;
1063
1064 /* Get the data stream */
1065 memset(stream, 0, sizeof(*stream));
1066 stream->next_in = map;
1067 stream->avail_in = mapsize;
1068 stream->next_out = buffer;
1069 stream->avail_out = bufsiz;
1070
1071 if (legacy_loose_object(map)) {
1072 inflateInit(stream);
1073 return inflate(stream, 0);
1074 }
1075
1076
1077 /*
1078 * There used to be a second loose object header format which
1079 * was meant to mimic the in-pack format, allowing for direct
1080 * copy of the object data. This format turned up not to be
1081 * really worth it and we don't write it any longer. But we
1082 * can still read it.
1083 */
1084 used = unpack_object_header_gently(map, mapsize, &type, &size);
1085 if (!used || !valid_loose_object_type[type])
1086 return -1;
1087 map += used;
1088 mapsize -= used;
1089
1090 /* Set up the stream for the rest.. */
1091 stream->next_in = map;
1092 stream->avail_in = mapsize;
1093 inflateInit(stream);
1094
1095 /* And generate the fake traditional header */
1096 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1097 typename(type), size);
1098 return 0;
1099 }
1100
1101 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1102 {
1103 int bytes = strlen(buffer) + 1;
1104 unsigned char *buf = xmalloc(1+size);
1105 unsigned long n;
1106 int status = Z_OK;
1107
1108 n = stream->total_out - bytes;
1109 if (n > size)
1110 n = size;
1111 memcpy(buf, (char *) buffer + bytes, n);
1112 bytes = n;
1113 if (bytes <= size) {
1114 /*
1115 * The above condition must be (bytes <= size), not
1116 * (bytes < size). In other words, even though we
1117 * expect no more output and set avail_out to zer0,
1118 * the input zlib stream may have bytes that express
1119 * "this concludes the stream", and we *do* want to
1120 * eat that input.
1121 *
1122 * Otherwise we would not be able to test that we
1123 * consumed all the input to reach the expected size;
1124 * we also want to check that zlib tells us that all
1125 * went well with status == Z_STREAM_END at the end.
1126 */
1127 stream->next_out = buf + bytes;
1128 stream->avail_out = size - bytes;
1129 while (status == Z_OK)
1130 status = inflate(stream, Z_FINISH);
1131 }
1132 buf[size] = 0;
1133 if (status == Z_STREAM_END && !stream->avail_in) {
1134 inflateEnd(stream);
1135 return buf;
1136 }
1137
1138 if (status < 0)
1139 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1140 else if (stream->avail_in)
1141 error("garbage at end of loose object '%s'",
1142 sha1_to_hex(sha1));
1143 free(buf);
1144 return NULL;
1145 }
1146
1147 /*
1148 * We used to just use "sscanf()", but that's actually way
1149 * too permissive for what we want to check. So do an anal
1150 * object header parse by hand.
1151 */
1152 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1153 {
1154 char type[10];
1155 int i;
1156 unsigned long size;
1157
1158 /*
1159 * The type can be at most ten bytes (including the
1160 * terminating '\0' that we add), and is followed by
1161 * a space.
1162 */
1163 i = 0;
1164 for (;;) {
1165 char c = *hdr++;
1166 if (c == ' ')
1167 break;
1168 type[i++] = c;
1169 if (i >= sizeof(type))
1170 return -1;
1171 }
1172 type[i] = 0;
1173
1174 /*
1175 * The length must follow immediately, and be in canonical
1176 * decimal format (ie "010" is not valid).
1177 */
1178 size = *hdr++ - '0';
1179 if (size > 9)
1180 return -1;
1181 if (size) {
1182 for (;;) {
1183 unsigned long c = *hdr - '0';
1184 if (c > 9)
1185 break;
1186 hdr++;
1187 size = size * 10 + c;
1188 }
1189 }
1190 *sizep = size;
1191
1192 /*
1193 * The length must be followed by a zero byte
1194 */
1195 return *hdr ? -1 : type_from_string(type);
1196 }
1197
1198 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1199 {
1200 int ret;
1201 z_stream stream;
1202 char hdr[8192];
1203
1204 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1205 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1206 return NULL;
1207
1208 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1209 }
1210
1211 unsigned long get_size_from_delta(struct packed_git *p,
1212 struct pack_window **w_curs,
1213 off_t curpos)
1214 {
1215 const unsigned char *data;
1216 unsigned char delta_head[20], *in;
1217 z_stream stream;
1218 int st;
1219
1220 memset(&stream, 0, sizeof(stream));
1221 stream.next_out = delta_head;
1222 stream.avail_out = sizeof(delta_head);
1223
1224 inflateInit(&stream);
1225 do {
1226 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1227 stream.next_in = in;
1228 st = inflate(&stream, Z_FINISH);
1229 curpos += stream.next_in - in;
1230 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1231 stream.total_out < sizeof(delta_head));
1232 inflateEnd(&stream);
1233 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1234 die("delta data unpack-initial failed");
1235
1236 /* Examine the initial part of the delta to figure out
1237 * the result size.
1238 */
1239 data = delta_head;
1240
1241 /* ignore base size */
1242 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1243
1244 /* Read the result size */
1245 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1246 }
1247
1248 static off_t get_delta_base(struct packed_git *p,
1249 struct pack_window **w_curs,
1250 off_t *curpos,
1251 enum object_type type,
1252 off_t delta_obj_offset)
1253 {
1254 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1255 off_t base_offset;
1256
1257 /* use_pack() assured us we have [base_info, base_info + 20)
1258 * as a range that we can look at without walking off the
1259 * end of the mapped window. Its actually the hash size
1260 * that is assured. An OFS_DELTA longer than the hash size
1261 * is stupid, as then a REF_DELTA would be smaller to store.
1262 */
1263 if (type == OBJ_OFS_DELTA) {
1264 unsigned used = 0;
1265 unsigned char c = base_info[used++];
1266 base_offset = c & 127;
1267 while (c & 128) {
1268 base_offset += 1;
1269 if (!base_offset || MSB(base_offset, 7))
1270 die("offset value overflow for delta base object");
1271 c = base_info[used++];
1272 base_offset = (base_offset << 7) + (c & 127);
1273 }
1274 base_offset = delta_obj_offset - base_offset;
1275 if (base_offset >= delta_obj_offset)
1276 die("delta base offset out of bound");
1277 *curpos += used;
1278 } else if (type == OBJ_REF_DELTA) {
1279 /* The base entry _must_ be in the same pack */
1280 base_offset = find_pack_entry_one(base_info, p);
1281 if (!base_offset)
1282 die("failed to find delta-pack base object %s",
1283 sha1_to_hex(base_info));
1284 *curpos += 20;
1285 } else
1286 die("I am totally screwed");
1287 return base_offset;
1288 }
1289
1290 /* forward declaration for a mutually recursive function */
1291 static int packed_object_info(struct packed_git *p, off_t offset,
1292 unsigned long *sizep);
1293
1294 static int packed_delta_info(struct packed_git *p,
1295 struct pack_window **w_curs,
1296 off_t curpos,
1297 enum object_type type,
1298 off_t obj_offset,
1299 unsigned long *sizep)
1300 {
1301 off_t base_offset;
1302
1303 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1304 type = packed_object_info(p, base_offset, NULL);
1305
1306 /* We choose to only get the type of the base object and
1307 * ignore potentially corrupt pack file that expects the delta
1308 * based on a base with a wrong size. This saves tons of
1309 * inflate() calls.
1310 */
1311 if (sizep)
1312 *sizep = get_size_from_delta(p, w_curs, curpos);
1313
1314 return type;
1315 }
1316
1317 static int unpack_object_header(struct packed_git *p,
1318 struct pack_window **w_curs,
1319 off_t *curpos,
1320 unsigned long *sizep)
1321 {
1322 unsigned char *base;
1323 unsigned int left;
1324 unsigned long used;
1325 enum object_type type;
1326
1327 /* use_pack() assures us we have [base, base + 20) available
1328 * as a range that we can look at at. (Its actually the hash
1329 * size that is assured.) With our object header encoding
1330 * the maximum deflated object size is 2^137, which is just
1331 * insane, so we know won't exceed what we have been given.
1332 */
1333 base = use_pack(p, w_curs, *curpos, &left);
1334 used = unpack_object_header_gently(base, left, &type, sizep);
1335 if (!used)
1336 die("object offset outside of pack file");
1337 *curpos += used;
1338
1339 return type;
1340 }
1341
1342 const char *packed_object_info_detail(struct packed_git *p,
1343 off_t obj_offset,
1344 unsigned long *size,
1345 unsigned long *store_size,
1346 unsigned int *delta_chain_length,
1347 unsigned char *base_sha1)
1348 {
1349 struct pack_window *w_curs = NULL;
1350 off_t curpos;
1351 unsigned long dummy;
1352 unsigned char *next_sha1;
1353 enum object_type type;
1354
1355 *delta_chain_length = 0;
1356 curpos = obj_offset;
1357 type = unpack_object_header(p, &w_curs, &curpos, size);
1358
1359 for (;;) {
1360 switch (type) {
1361 default:
1362 die("pack %s contains unknown object type %d",
1363 p->pack_name, type);
1364 case OBJ_COMMIT:
1365 case OBJ_TREE:
1366 case OBJ_BLOB:
1367 case OBJ_TAG:
1368 *store_size = 0; /* notyet */
1369 unuse_pack(&w_curs);
1370 return typename(type);
1371 case OBJ_OFS_DELTA:
1372 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1373 if (*delta_chain_length == 0) {
1374 /* TODO: find base_sha1 as pointed by curpos */
1375 hashclr(base_sha1);
1376 }
1377 break;
1378 case OBJ_REF_DELTA:
1379 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1380 if (*delta_chain_length == 0)
1381 hashcpy(base_sha1, next_sha1);
1382 obj_offset = find_pack_entry_one(next_sha1, p);
1383 break;
1384 }
1385 (*delta_chain_length)++;
1386 curpos = obj_offset;
1387 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1388 }
1389 }
1390
1391 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1392 unsigned long *sizep)
1393 {
1394 struct pack_window *w_curs = NULL;
1395 unsigned long size;
1396 off_t curpos = obj_offset;
1397 enum object_type type;
1398
1399 type = unpack_object_header(p, &w_curs, &curpos, &size);
1400
1401 switch (type) {
1402 case OBJ_OFS_DELTA:
1403 case OBJ_REF_DELTA:
1404 type = packed_delta_info(p, &w_curs, curpos,
1405 type, obj_offset, sizep);
1406 break;
1407 case OBJ_COMMIT:
1408 case OBJ_TREE:
1409 case OBJ_BLOB:
1410 case OBJ_TAG:
1411 if (sizep)
1412 *sizep = size;
1413 break;
1414 default:
1415 die("pack %s contains unknown object type %d",
1416 p->pack_name, type);
1417 }
1418 unuse_pack(&w_curs);
1419 return type;
1420 }
1421
1422 static void *unpack_compressed_entry(struct packed_git *p,
1423 struct pack_window **w_curs,
1424 off_t curpos,
1425 unsigned long size)
1426 {
1427 int st;
1428 z_stream stream;
1429 unsigned char *buffer, *in;
1430
1431 buffer = xmalloc(size + 1);
1432 buffer[size] = 0;
1433 memset(&stream, 0, sizeof(stream));
1434 stream.next_out = buffer;
1435 stream.avail_out = size;
1436
1437 inflateInit(&stream);
1438 do {
1439 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1440 stream.next_in = in;
1441 st = inflate(&stream, Z_FINISH);
1442 curpos += stream.next_in - in;
1443 } while (st == Z_OK || st == Z_BUF_ERROR);
1444 inflateEnd(&stream);
1445 if ((st != Z_STREAM_END) || stream.total_out != size) {
1446 free(buffer);
1447 return NULL;
1448 }
1449
1450 return buffer;
1451 }
1452
1453 #define MAX_DELTA_CACHE (256)
1454
1455 static size_t delta_base_cached;
1456
1457 static struct delta_base_cache_lru_list {
1458 struct delta_base_cache_lru_list *prev;
1459 struct delta_base_cache_lru_list *next;
1460 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1461
1462 static struct delta_base_cache_entry {
1463 struct delta_base_cache_lru_list lru;
1464 void *data;
1465 struct packed_git *p;
1466 off_t base_offset;
1467 unsigned long size;
1468 enum object_type type;
1469 } delta_base_cache[MAX_DELTA_CACHE];
1470
1471 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1472 {
1473 unsigned long hash;
1474
1475 hash = (unsigned long)p + (unsigned long)base_offset;
1476 hash += (hash >> 8) + (hash >> 16);
1477 return hash % MAX_DELTA_CACHE;
1478 }
1479
1480 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1481 unsigned long *base_size, enum object_type *type, int keep_cache)
1482 {
1483 void *ret;
1484 unsigned long hash = pack_entry_hash(p, base_offset);
1485 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1486
1487 ret = ent->data;
1488 if (ret && ent->p == p && ent->base_offset == base_offset)
1489 goto found_cache_entry;
1490 return unpack_entry(p, base_offset, type, base_size);
1491
1492 found_cache_entry:
1493 if (!keep_cache) {
1494 ent->data = NULL;
1495 ent->lru.next->prev = ent->lru.prev;
1496 ent->lru.prev->next = ent->lru.next;
1497 delta_base_cached -= ent->size;
1498 } else {
1499 ret = xmemdupz(ent->data, ent->size);
1500 }
1501 *type = ent->type;
1502 *base_size = ent->size;
1503 return ret;
1504 }
1505
1506 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1507 {
1508 if (ent->data) {
1509 free(ent->data);
1510 ent->data = NULL;
1511 ent->lru.next->prev = ent->lru.prev;
1512 ent->lru.prev->next = ent->lru.next;
1513 delta_base_cached -= ent->size;
1514 }
1515 }
1516
1517 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1518 void *base, unsigned long base_size, enum object_type type)
1519 {
1520 unsigned long hash = pack_entry_hash(p, base_offset);
1521 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1522 struct delta_base_cache_lru_list *lru;
1523
1524 release_delta_base_cache(ent);
1525 delta_base_cached += base_size;
1526
1527 for (lru = delta_base_cache_lru.next;
1528 delta_base_cached > delta_base_cache_limit
1529 && lru != &delta_base_cache_lru;
1530 lru = lru->next) {
1531 struct delta_base_cache_entry *f = (void *)lru;
1532 if (f->type == OBJ_BLOB)
1533 release_delta_base_cache(f);
1534 }
1535 for (lru = delta_base_cache_lru.next;
1536 delta_base_cached > delta_base_cache_limit
1537 && lru != &delta_base_cache_lru;
1538 lru = lru->next) {
1539 struct delta_base_cache_entry *f = (void *)lru;
1540 release_delta_base_cache(f);
1541 }
1542
1543 ent->p = p;
1544 ent->base_offset = base_offset;
1545 ent->type = type;
1546 ent->data = base;
1547 ent->size = base_size;
1548 ent->lru.next = &delta_base_cache_lru;
1549 ent->lru.prev = delta_base_cache_lru.prev;
1550 delta_base_cache_lru.prev->next = &ent->lru;
1551 delta_base_cache_lru.prev = &ent->lru;
1552 }
1553
1554 static void *unpack_delta_entry(struct packed_git *p,
1555 struct pack_window **w_curs,
1556 off_t curpos,
1557 unsigned long delta_size,
1558 off_t obj_offset,
1559 enum object_type *type,
1560 unsigned long *sizep)
1561 {
1562 void *delta_data, *result, *base;
1563 unsigned long base_size;
1564 off_t base_offset;
1565
1566 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1567 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1568 if (!base)
1569 die("failed to read delta base object"
1570 " at %"PRIuMAX" from %s",
1571 (uintmax_t)base_offset, p->pack_name);
1572
1573 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1574 if (!delta_data)
1575 die("failed to unpack compressed delta"
1576 " at %"PRIuMAX" from %s",
1577 (uintmax_t)curpos, p->pack_name);
1578 result = patch_delta(base, base_size,
1579 delta_data, delta_size,
1580 sizep);
1581 if (!result)
1582 die("failed to apply delta");
1583 free(delta_data);
1584 add_delta_base_cache(p, base_offset, base, base_size, *type);
1585 return result;
1586 }
1587
1588 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1589 enum object_type *type, unsigned long *sizep)
1590 {
1591 struct pack_window *w_curs = NULL;
1592 off_t curpos = obj_offset;
1593 void *data;
1594
1595 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1596 switch (*type) {
1597 case OBJ_OFS_DELTA:
1598 case OBJ_REF_DELTA:
1599 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1600 obj_offset, type, sizep);
1601 break;
1602 case OBJ_COMMIT:
1603 case OBJ_TREE:
1604 case OBJ_BLOB:
1605 case OBJ_TAG:
1606 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1607 break;
1608 default:
1609 die("unknown object type %i in %s", *type, p->pack_name);
1610 }
1611 unuse_pack(&w_curs);
1612 return data;
1613 }
1614
1615 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1616 uint32_t n)
1617 {
1618 const unsigned char *index = p->index_data;
1619 if (!index) {
1620 if (open_pack_index(p))
1621 return NULL;
1622 index = p->index_data;
1623 }
1624 if (n >= p->num_objects)
1625 return NULL;
1626 index += 4 * 256;
1627 if (p->index_version == 1) {
1628 return index + 24 * n + 4;
1629 } else {
1630 index += 8;
1631 return index + 20 * n;
1632 }
1633 }
1634
1635 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1636 {
1637 const unsigned char *index = p->index_data;
1638 index += 4 * 256;
1639 if (p->index_version == 1) {
1640 return ntohl(*((uint32_t *)(index + 24 * n)));
1641 } else {
1642 uint32_t off;
1643 index += 8 + p->num_objects * (20 + 4);
1644 off = ntohl(*((uint32_t *)(index + 4 * n)));
1645 if (!(off & 0x80000000))
1646 return off;
1647 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1648 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1649 ntohl(*((uint32_t *)(index + 4)));
1650 }
1651 }
1652
1653 off_t find_pack_entry_one(const unsigned char *sha1,
1654 struct packed_git *p)
1655 {
1656 const uint32_t *level1_ofs = p->index_data;
1657 const unsigned char *index = p->index_data;
1658 unsigned hi, lo;
1659
1660 if (!index) {
1661 if (open_pack_index(p))
1662 return 0;
1663 level1_ofs = p->index_data;
1664 index = p->index_data;
1665 }
1666 if (p->index_version > 1) {
1667 level1_ofs += 2;
1668 index += 8;
1669 }
1670 index += 4 * 256;
1671 hi = ntohl(level1_ofs[*sha1]);
1672 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1673
1674 do {
1675 unsigned mi = (lo + hi) / 2;
1676 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1677 int cmp = hashcmp(index + x, sha1);
1678 if (!cmp)
1679 return nth_packed_object_offset(p, mi);
1680 if (cmp > 0)
1681 hi = mi;
1682 else
1683 lo = mi+1;
1684 } while (lo < hi);
1685 return 0;
1686 }
1687
1688 int matches_pack_name(struct packed_git *p, const char *name)
1689 {
1690 const char *last_c, *c;
1691
1692 if (!strcmp(p->pack_name, name))
1693 return 1;
1694
1695 for (c = p->pack_name, last_c = c; *c;)
1696 if (*c == '/')
1697 last_c = ++c;
1698 else
1699 ++c;
1700 if (!strcmp(last_c, name))
1701 return 1;
1702
1703 return 0;
1704 }
1705
1706 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1707 {
1708 static struct packed_git *last_found = (void *)1;
1709 struct packed_git *p;
1710 off_t offset;
1711
1712 prepare_packed_git();
1713 if (!packed_git)
1714 return 0;
1715 p = (last_found == (void *)1) ? packed_git : last_found;
1716
1717 do {
1718 if (ignore_packed) {
1719 const char **ig;
1720 for (ig = ignore_packed; *ig; ig++)
1721 if (matches_pack_name(p, *ig))
1722 break;
1723 if (*ig)
1724 goto next;
1725 }
1726
1727 offset = find_pack_entry_one(sha1, p);
1728 if (offset) {
1729 /*
1730 * We are about to tell the caller where they can
1731 * locate the requested object. We better make
1732 * sure the packfile is still here and can be
1733 * accessed before supplying that answer, as
1734 * it may have been deleted since the index
1735 * was loaded!
1736 */
1737 if (p->pack_fd == -1 && open_packed_git(p)) {
1738 error("packfile %s cannot be accessed", p->pack_name);
1739 goto next;
1740 }
1741 e->offset = offset;
1742 e->p = p;
1743 hashcpy(e->sha1, sha1);
1744 last_found = p;
1745 return 1;
1746 }
1747
1748 next:
1749 if (p == last_found)
1750 p = packed_git;
1751 else
1752 p = p->next;
1753 if (p == last_found)
1754 p = p->next;
1755 } while (p);
1756 return 0;
1757 }
1758
1759 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1760 struct packed_git *packs)
1761 {
1762 struct packed_git *p;
1763
1764 for (p = packs; p; p = p->next) {
1765 if (find_pack_entry_one(sha1, p))
1766 return p;
1767 }
1768 return NULL;
1769
1770 }
1771
1772 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1773 {
1774 int status;
1775 unsigned long mapsize, size;
1776 void *map;
1777 z_stream stream;
1778 char hdr[32];
1779
1780 map = map_sha1_file(sha1, &mapsize);
1781 if (!map)
1782 return error("unable to find %s", sha1_to_hex(sha1));
1783 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1784 status = error("unable to unpack %s header",
1785 sha1_to_hex(sha1));
1786 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1787 status = error("unable to parse %s header", sha1_to_hex(sha1));
1788 else if (sizep)
1789 *sizep = size;
1790 inflateEnd(&stream);
1791 munmap(map, mapsize);
1792 return status;
1793 }
1794
1795 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1796 {
1797 struct pack_entry e;
1798
1799 if (!find_pack_entry(sha1, &e, NULL)) {
1800 reprepare_packed_git();
1801 if (!find_pack_entry(sha1, &e, NULL))
1802 return sha1_loose_object_info(sha1, sizep);
1803 }
1804 return packed_object_info(e.p, e.offset, sizep);
1805 }
1806
1807 static void *read_packed_sha1(const unsigned char *sha1,
1808 enum object_type *type, unsigned long *size)
1809 {
1810 struct pack_entry e;
1811
1812 if (!find_pack_entry(sha1, &e, NULL))
1813 return NULL;
1814 else
1815 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1816 }
1817
1818 /*
1819 * This is meant to hold a *small* number of objects that you would
1820 * want read_sha1_file() to be able to return, but yet you do not want
1821 * to write them into the object store (e.g. a browse-only
1822 * application).
1823 */
1824 static struct cached_object {
1825 unsigned char sha1[20];
1826 enum object_type type;
1827 void *buf;
1828 unsigned long size;
1829 } *cached_objects;
1830 static int cached_object_nr, cached_object_alloc;
1831
1832 static struct cached_object *find_cached_object(const unsigned char *sha1)
1833 {
1834 int i;
1835 struct cached_object *co = cached_objects;
1836
1837 for (i = 0; i < cached_object_nr; i++, co++) {
1838 if (!hashcmp(co->sha1, sha1))
1839 return co;
1840 }
1841 return NULL;
1842 }
1843
1844 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1845 unsigned char *sha1)
1846 {
1847 struct cached_object *co;
1848
1849 hash_sha1_file(buf, len, typename(type), sha1);
1850 if (has_sha1_file(sha1) || find_cached_object(sha1))
1851 return 0;
1852 if (cached_object_alloc <= cached_object_nr) {
1853 cached_object_alloc = alloc_nr(cached_object_alloc);
1854 cached_objects = xrealloc(cached_objects,
1855 sizeof(*cached_objects) *
1856 cached_object_alloc);
1857 }
1858 co = &cached_objects[cached_object_nr++];
1859 co->size = len;
1860 co->type = type;
1861 co->buf = xmalloc(len);
1862 memcpy(co->buf, buf, len);
1863 hashcpy(co->sha1, sha1);
1864 return 0;
1865 }
1866
1867 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1868 unsigned long *size)
1869 {
1870 unsigned long mapsize;
1871 void *map, *buf;
1872 struct cached_object *co;
1873
1874 co = find_cached_object(sha1);
1875 if (co) {
1876 *type = co->type;
1877 *size = co->size;
1878 return xmemdupz(co->buf, co->size);
1879 }
1880
1881 buf = read_packed_sha1(sha1, type, size);
1882 if (buf)
1883 return buf;
1884 map = map_sha1_file(sha1, &mapsize);
1885 if (map) {
1886 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1887 munmap(map, mapsize);
1888 return buf;
1889 }
1890 reprepare_packed_git();
1891 return read_packed_sha1(sha1, type, size);
1892 }
1893
1894 void *read_object_with_reference(const unsigned char *sha1,
1895 const char *required_type_name,
1896 unsigned long *size,
1897 unsigned char *actual_sha1_return)
1898 {
1899 enum object_type type, required_type;
1900 void *buffer;
1901 unsigned long isize;
1902 unsigned char actual_sha1[20];
1903
1904 required_type = type_from_string(required_type_name);
1905 hashcpy(actual_sha1, sha1);
1906 while (1) {
1907 int ref_length = -1;
1908 const char *ref_type = NULL;
1909
1910 buffer = read_sha1_file(actual_sha1, &type, &isize);
1911 if (!buffer)
1912 return NULL;
1913 if (type == required_type) {
1914 *size = isize;
1915 if (actual_sha1_return)
1916 hashcpy(actual_sha1_return, actual_sha1);
1917 return buffer;
1918 }
1919 /* Handle references */
1920 else if (type == OBJ_COMMIT)
1921 ref_type = "tree ";
1922 else if (type == OBJ_TAG)
1923 ref_type = "object ";
1924 else {
1925 free(buffer);
1926 return NULL;
1927 }
1928 ref_length = strlen(ref_type);
1929
1930 if (memcmp(buffer, ref_type, ref_length) ||
1931 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1932 free(buffer);
1933 return NULL;
1934 }
1935 free(buffer);
1936 /* Now we have the ID of the referred-to object in
1937 * actual_sha1. Check again. */
1938 }
1939 }
1940
1941 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1942 const char *type, unsigned char *sha1,
1943 char *hdr, int *hdrlen)
1944 {
1945 SHA_CTX c;
1946
1947 /* Generate the header */
1948 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1949
1950 /* Sha1.. */
1951 SHA1_Init(&c);
1952 SHA1_Update(&c, hdr, *hdrlen);
1953 SHA1_Update(&c, buf, len);
1954 SHA1_Final(sha1, &c);
1955 }
1956
1957 /*
1958 * Link the tempfile to the final place, possibly creating the
1959 * last directory level as you do so.
1960 *
1961 * Returns the errno on failure, 0 on success.
1962 */
1963 static int link_temp_to_file(const char *tmpfile, const char *filename)
1964 {
1965 int ret;
1966 char *dir;
1967
1968 if (!link(tmpfile, filename))
1969 return 0;
1970
1971 /*
1972 * Try to mkdir the last path component if that failed.
1973 *
1974 * Re-try the "link()" regardless of whether the mkdir
1975 * succeeds, since a race might mean that somebody
1976 * else succeeded.
1977 */
1978 ret = errno;
1979 dir = strrchr(filename, '/');
1980 if (dir) {
1981 *dir = 0;
1982 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1983 *dir = '/';
1984 return -2;
1985 }
1986 *dir = '/';
1987 if (!link(tmpfile, filename))
1988 return 0;
1989 ret = errno;
1990 }
1991 return ret;
1992 }
1993
1994 /*
1995 * Move the just written object into its final resting place
1996 */
1997 int move_temp_to_file(const char *tmpfile, const char *filename)
1998 {
1999 int ret = link_temp_to_file(tmpfile, filename);
2000
2001 /*
2002 * Coda hack - coda doesn't like cross-directory links,
2003 * so we fall back to a rename, which will mean that it
2004 * won't be able to check collisions, but that's not a
2005 * big deal.
2006 *
2007 * The same holds for FAT formatted media.
2008 *
2009 * When this succeeds, we just return 0. We have nothing
2010 * left to unlink.
2011 */
2012 if (ret && ret != EEXIST) {
2013 if (!rename(tmpfile, filename))
2014 return 0;
2015 ret = errno;
2016 }
2017 unlink(tmpfile);
2018 if (ret) {
2019 if (ret != EEXIST) {
2020 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2021 }
2022 /* FIXME!!! Collision check here ? */
2023 }
2024
2025 return 0;
2026 }
2027
2028 static int write_buffer(int fd, const void *buf, size_t len)
2029 {
2030 if (write_in_full(fd, buf, len) < 0)
2031 return error("file write error (%s)", strerror(errno));
2032 return 0;
2033 }
2034
2035 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2036 unsigned char *sha1)
2037 {
2038 char hdr[32];
2039 int hdrlen;
2040 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2041 return 0;
2042 }
2043
2044 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2045 {
2046 int size, ret;
2047 unsigned char *compressed;
2048 z_stream stream;
2049 unsigned char sha1[20];
2050 char *filename;
2051 static char tmpfile[PATH_MAX];
2052 char hdr[32];
2053 int fd, hdrlen;
2054
2055 /* Normally if we have it in the pack then we do not bother writing
2056 * it out into .git/objects/??/?{38} file.
2057 */
2058 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2059 filename = sha1_file_name(sha1);
2060 if (returnsha1)
2061 hashcpy(returnsha1, sha1);
2062 if (has_sha1_file(sha1))
2063 return 0;
2064 fd = open(filename, O_RDONLY);
2065 if (fd >= 0) {
2066 /*
2067 * FIXME!!! We might do collision checking here, but we'd
2068 * need to uncompress the old file and check it. Later.
2069 */
2070 close(fd);
2071 return 0;
2072 }
2073
2074 if (errno != ENOENT) {
2075 return error("sha1 file %s: %s\n", filename, strerror(errno));
2076 }
2077
2078 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2079
2080 fd = mkstemp(tmpfile);
2081 if (fd < 0) {
2082 if (errno == EPERM)
2083 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2084 else
2085 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2086 }
2087
2088 /* Set it up */
2089 memset(&stream, 0, sizeof(stream));
2090 deflateInit(&stream, zlib_compression_level);
2091 size = 8 + deflateBound(&stream, len+hdrlen);
2092 compressed = xmalloc(size);
2093
2094 /* Compress it */
2095 stream.next_out = compressed;
2096 stream.avail_out = size;
2097
2098 /* First header.. */
2099 stream.next_in = (unsigned char *)hdr;
2100 stream.avail_in = hdrlen;
2101 while (deflate(&stream, 0) == Z_OK)
2102 /* nothing */;
2103
2104 /* Then the data itself.. */
2105 stream.next_in = buf;
2106 stream.avail_in = len;
2107 ret = deflate(&stream, Z_FINISH);
2108 if (ret != Z_STREAM_END)
2109 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2110
2111 ret = deflateEnd(&stream);
2112 if (ret != Z_OK)
2113 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2114
2115 size = stream.total_out;
2116
2117 if (write_buffer(fd, compressed, size) < 0)
2118 die("unable to write sha1 file");
2119 fchmod(fd, 0444);
2120 if (close(fd))
2121 die("unable to write sha1 file");
2122 free(compressed);
2123
2124 return move_temp_to_file(tmpfile, filename);
2125 }
2126
2127 /*
2128 * We need to unpack and recompress the object for writing
2129 * it out to a different file.
2130 */
2131 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2132 {
2133 size_t size;
2134 z_stream stream;
2135 unsigned char *unpacked;
2136 unsigned long len;
2137 enum object_type type;
2138 char hdr[32];
2139 int hdrlen;
2140 void *buf;
2141
2142 /* need to unpack and recompress it by itself */
2143 unpacked = read_packed_sha1(sha1, &type, &len);
2144 if (!unpacked)
2145 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2146
2147 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2148
2149 /* Set it up */
2150 memset(&stream, 0, sizeof(stream));
2151 deflateInit(&stream, zlib_compression_level);
2152 size = deflateBound(&stream, len + hdrlen);
2153 buf = xmalloc(size);
2154
2155 /* Compress it */
2156 stream.next_out = buf;
2157 stream.avail_out = size;
2158
2159 /* First header.. */
2160 stream.next_in = (void *)hdr;
2161 stream.avail_in = hdrlen;
2162 while (deflate(&stream, 0) == Z_OK)
2163 /* nothing */;
2164
2165 /* Then the data itself.. */
2166 stream.next_in = unpacked;
2167 stream.avail_in = len;
2168 while (deflate(&stream, Z_FINISH) == Z_OK)
2169 /* nothing */;
2170 deflateEnd(&stream);
2171 free(unpacked);
2172
2173 *objsize = stream.total_out;
2174 return buf;
2175 }
2176
2177 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2178 {
2179 int retval;
2180 unsigned long objsize;
2181 void *buf = map_sha1_file(sha1, &objsize);
2182
2183 if (buf) {
2184 retval = write_buffer(fd, buf, objsize);
2185 munmap(buf, objsize);
2186 return retval;
2187 }
2188
2189 buf = repack_object(sha1, &objsize);
2190 retval = write_buffer(fd, buf, objsize);
2191 free(buf);
2192 return retval;
2193 }
2194
2195 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2196 size_t bufsize, size_t *bufposn)
2197 {
2198 char tmpfile[PATH_MAX];
2199 int local;
2200 z_stream stream;
2201 unsigned char real_sha1[20];
2202 unsigned char discard[4096];
2203 int ret;
2204 SHA_CTX c;
2205
2206 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2207
2208 local = mkstemp(tmpfile);
2209 if (local < 0) {
2210 if (errno == EPERM)
2211 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2212 else
2213 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2214 }
2215
2216 memset(&stream, 0, sizeof(stream));
2217
2218 inflateInit(&stream);
2219
2220 SHA1_Init(&c);
2221
2222 do {
2223 ssize_t size;
2224 if (*bufposn) {
2225 stream.avail_in = *bufposn;
2226 stream.next_in = (unsigned char *) buffer;
2227 do {
2228 stream.next_out = discard;
2229 stream.avail_out = sizeof(discard);
2230 ret = inflate(&stream, Z_SYNC_FLUSH);
2231 SHA1_Update(&c, discard, sizeof(discard) -
2232 stream.avail_out);
2233 } while (stream.avail_in && ret == Z_OK);
2234 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2235 die("unable to write sha1 file");
2236 memmove(buffer, buffer + *bufposn - stream.avail_in,
2237 stream.avail_in);
2238 *bufposn = stream.avail_in;
2239 if (ret != Z_OK)
2240 break;
2241 }
2242 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2243 if (size <= 0) {
2244 close(local);
2245 unlink(tmpfile);
2246 if (!size)
2247 return error("Connection closed?");
2248 perror("Reading from connection");
2249 return -1;
2250 }
2251 *bufposn += size;
2252 } while (1);
2253 inflateEnd(&stream);
2254
2255 fchmod(local, 0444);
2256 if (close(local) != 0)
2257 die("unable to write sha1 file");
2258 SHA1_Final(real_sha1, &c);
2259 if (ret != Z_STREAM_END) {
2260 unlink(tmpfile);
2261 return error("File %s corrupted", sha1_to_hex(sha1));
2262 }
2263 if (hashcmp(sha1, real_sha1)) {
2264 unlink(tmpfile);
2265 return error("File %s has bad hash", sha1_to_hex(sha1));
2266 }
2267
2268 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2269 }
2270
2271 int has_pack_index(const unsigned char *sha1)
2272 {
2273 struct stat st;
2274 if (stat(sha1_pack_index_name(sha1), &st))
2275 return 0;
2276 return 1;
2277 }
2278
2279 int has_pack_file(const unsigned char *sha1)
2280 {
2281 struct stat st;
2282 if (stat(sha1_pack_name(sha1), &st))
2283 return 0;
2284 return 1;
2285 }
2286
2287 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2288 {
2289 struct pack_entry e;
2290 return find_pack_entry(sha1, &e, ignore_packed);
2291 }
2292
2293 int has_sha1_file(const unsigned char *sha1)
2294 {
2295 struct stat st;
2296 struct pack_entry e;
2297
2298 if (find_pack_entry(sha1, &e, NULL))
2299 return 1;
2300 return find_sha1_file(sha1, &st) ? 1 : 0;
2301 }
2302
2303 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2304 {
2305 struct strbuf buf;
2306 int ret;
2307
2308 strbuf_init(&buf, 0);
2309 if (strbuf_read(&buf, fd, 4096) < 0) {
2310 strbuf_release(&buf);
2311 return -1;
2312 }
2313
2314 if (!type)
2315 type = blob_type;
2316 if (write_object)
2317 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2318 else
2319 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2320 strbuf_release(&buf);
2321
2322 return ret;
2323 }
2324
2325 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2326 enum object_type type, const char *path)
2327 {
2328 size_t size = xsize_t(st->st_size);
2329 void *buf = NULL;
2330 int ret, re_allocated = 0;
2331
2332 if (size)
2333 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2334 close(fd);
2335
2336 if (!type)
2337 type = OBJ_BLOB;
2338
2339 /*
2340 * Convert blobs to git internal format
2341 */
2342 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2343 struct strbuf nbuf;
2344 strbuf_init(&nbuf, 0);
2345 if (convert_to_git(path, buf, size, &nbuf)) {
2346 munmap(buf, size);
2347 buf = strbuf_detach(&nbuf, &size);
2348 re_allocated = 1;
2349 }
2350 }
2351
2352 if (write_object)
2353 ret = write_sha1_file(buf, size, typename(type), sha1);
2354 else
2355 ret = hash_sha1_file(buf, size, typename(type), sha1);
2356 if (re_allocated) {
2357 free(buf);
2358 return ret;
2359 }
2360 if (size)
2361 munmap(buf, size);
2362 return ret;
2363 }
2364
2365 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2366 {
2367 int fd;
2368 char *target;
2369 size_t len;
2370
2371 switch (st->st_mode & S_IFMT) {
2372 case S_IFREG:
2373 fd = open(path, O_RDONLY);
2374 if (fd < 0)
2375 return error("open(\"%s\"): %s", path,
2376 strerror(errno));
2377 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2378 return error("%s: failed to insert into database",
2379 path);
2380 break;
2381 case S_IFLNK:
2382 len = xsize_t(st->st_size);
2383 target = xmalloc(len + 1);
2384 if (readlink(path, target, len + 1) != st->st_size) {
2385 char *errstr = strerror(errno);
2386 free(target);
2387 return error("readlink(\"%s\"): %s", path,
2388 errstr);
2389 }
2390 if (!write_object)
2391 hash_sha1_file(target, len, blob_type, sha1);
2392 else if (write_sha1_file(target, len, blob_type, sha1))
2393 return error("%s: failed to insert into database",
2394 path);
2395 free(target);
2396 break;
2397 case S_IFDIR:
2398 return resolve_gitlink_ref(path, "HEAD", sha1);
2399 default:
2400 return error("%s: unsupported file type", path);
2401 }
2402 return 0;
2403 }
2404
2405 int read_pack_header(int fd, struct pack_header *header)
2406 {
2407 char *c = (char*)header;
2408 ssize_t remaining = sizeof(struct pack_header);
2409 do {
2410 ssize_t r = xread(fd, c, remaining);
2411 if (r <= 0)
2412 /* "eof before pack header was fully read" */
2413 return PH_ERROR_EOF;
2414 remaining -= r;
2415 c += r;
2416 } while (remaining > 0);
2417 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2418 /* "protocol error (pack signature mismatch detected)" */
2419 return PH_ERROR_PACK_SIGNATURE;
2420 if (!pack_version_ok(header->hdr_version))
2421 /* "protocol error (pack version unsupported)" */
2422 return PH_ERROR_PROTOCOL;
2423 return 0;
2424 }