More static
[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 #else
29 #define SZ_FMT "zu"
30 #endif
31
32 const unsigned char null_sha1[20];
33
34 static unsigned int sha1_file_open_flag = O_NOATIME;
35
36 const signed char hexval_table[256] = {
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
43 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
44 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
45 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
69 };
70
71 int get_sha1_hex(const char *hex, unsigned char *sha1)
72 {
73 int i;
74 for (i = 0; i < 20; i++) {
75 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
76 if (val & ~0xff)
77 return -1;
78 *sha1++ = val;
79 hex += 2;
80 }
81 return 0;
82 }
83
84 int safe_create_leading_directories(char *path)
85 {
86 char *pos = path;
87 struct stat st;
88
89 if (*pos == '/')
90 pos++;
91
92 while (pos) {
93 pos = strchr(pos, '/');
94 if (!pos)
95 break;
96 *pos = 0;
97 if (!stat(path, &st)) {
98 /* path exists */
99 if (!S_ISDIR(st.st_mode)) {
100 *pos = '/';
101 return -3;
102 }
103 }
104 else if (mkdir(path, 0777)) {
105 *pos = '/';
106 return -1;
107 }
108 else if (adjust_shared_perm(path)) {
109 *pos = '/';
110 return -2;
111 }
112 *pos++ = '/';
113 }
114 return 0;
115 }
116
117 char * sha1_to_hex(const unsigned char *sha1)
118 {
119 static int bufno;
120 static char hexbuffer[4][50];
121 static const char hex[] = "0123456789abcdef";
122 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
123 int i;
124
125 for (i = 0; i < 20; i++) {
126 unsigned int val = *sha1++;
127 *buf++ = hex[val >> 4];
128 *buf++ = hex[val & 0xf];
129 }
130 *buf = '\0';
131
132 return buffer;
133 }
134
135 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
136 {
137 int i;
138 for (i = 0; i < 20; i++) {
139 static char hex[] = "0123456789abcdef";
140 unsigned int val = sha1[i];
141 char *pos = pathbuf + i*2 + (i > 0);
142 *pos++ = hex[val >> 4];
143 *pos = hex[val & 0xf];
144 }
145 }
146
147 /*
148 * NOTE! This returns a statically allocated buffer, so you have to be
149 * careful about using it. Do a "xstrdup()" if you need to save the
150 * filename.
151 *
152 * Also note that this returns the location for creating. Reading
153 * SHA1 file can happen from any alternate directory listed in the
154 * DB_ENVIRONMENT environment variable if it is not found in
155 * the primary object database.
156 */
157 char *sha1_file_name(const unsigned char *sha1)
158 {
159 static char *name, *base;
160
161 if (!base) {
162 const char *sha1_file_directory = get_object_directory();
163 int len = strlen(sha1_file_directory);
164 base = xmalloc(len + 60);
165 memcpy(base, sha1_file_directory, len);
166 memset(base+len, 0, 60);
167 base[len] = '/';
168 base[len+3] = '/';
169 name = base + len + 1;
170 }
171 fill_sha1_path(name, sha1);
172 return base;
173 }
174
175 char *sha1_pack_name(const unsigned char *sha1)
176 {
177 static const char hex[] = "0123456789abcdef";
178 static char *name, *base, *buf;
179 int i;
180
181 if (!base) {
182 const char *sha1_file_directory = get_object_directory();
183 int len = strlen(sha1_file_directory);
184 base = xmalloc(len + 60);
185 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
186 name = base + len + 11;
187 }
188
189 buf = name;
190
191 for (i = 0; i < 20; i++) {
192 unsigned int val = *sha1++;
193 *buf++ = hex[val >> 4];
194 *buf++ = hex[val & 0xf];
195 }
196
197 return base;
198 }
199
200 char *sha1_pack_index_name(const unsigned char *sha1)
201 {
202 static const char hex[] = "0123456789abcdef";
203 static char *name, *base, *buf;
204 int i;
205
206 if (!base) {
207 const char *sha1_file_directory = get_object_directory();
208 int len = strlen(sha1_file_directory);
209 base = xmalloc(len + 60);
210 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
211 name = base + len + 11;
212 }
213
214 buf = name;
215
216 for (i = 0; i < 20; i++) {
217 unsigned int val = *sha1++;
218 *buf++ = hex[val >> 4];
219 *buf++ = hex[val & 0xf];
220 }
221
222 return base;
223 }
224
225 struct alternate_object_database *alt_odb_list;
226 static struct alternate_object_database **alt_odb_tail;
227
228 static void read_info_alternates(const char * alternates, int depth);
229
230 /*
231 * Prepare alternate object database registry.
232 *
233 * The variable alt_odb_list points at the list of struct
234 * alternate_object_database. The elements on this list come from
235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
237 * whose contents is similar to that environment variable but can be
238 * LF separated. Its base points at a statically allocated buffer that
239 * contains "/the/directory/corresponding/to/.git/objects/...", while
240 * its name points just after the slash at the end of ".git/objects/"
241 * in the example above, and has enough space to hold 40-byte hex
242 * SHA1, an extra slash for the first level indirection, and the
243 * terminating NUL.
244 */
245 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
246 {
247 struct stat st;
248 const char *objdir = get_object_directory();
249 struct alternate_object_database *ent;
250 struct alternate_object_database *alt;
251 /* 43 = 40-byte + 2 '/' + terminating NUL */
252 int pfxlen = len;
253 int entlen = pfxlen + 43;
254 int base_len = -1;
255
256 if (*entry != '/' && relative_base) {
257 /* Relative alt-odb */
258 if (base_len < 0)
259 base_len = strlen(relative_base) + 1;
260 entlen += base_len;
261 pfxlen += base_len;
262 }
263 ent = xmalloc(sizeof(*ent) + entlen);
264
265 if (*entry != '/' && relative_base) {
266 memcpy(ent->base, relative_base, base_len - 1);
267 ent->base[base_len - 1] = '/';
268 memcpy(ent->base + base_len, entry, len);
269 }
270 else
271 memcpy(ent->base, entry, pfxlen);
272
273 ent->name = ent->base + pfxlen + 1;
274 ent->base[pfxlen + 3] = '/';
275 ent->base[pfxlen] = ent->base[entlen-1] = 0;
276
277 /* Detect cases where alternate disappeared */
278 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
279 error("object directory %s does not exist; "
280 "check .git/objects/info/alternates.",
281 ent->base);
282 free(ent);
283 return -1;
284 }
285
286 /* Prevent the common mistake of listing the same
287 * thing twice, or object directory itself.
288 */
289 for (alt = alt_odb_list; alt; alt = alt->next) {
290 if (!memcmp(ent->base, alt->base, pfxlen)) {
291 free(ent);
292 return -1;
293 }
294 }
295 if (!memcmp(ent->base, objdir, pfxlen)) {
296 free(ent);
297 return -1;
298 }
299
300 /* add the alternate entry */
301 *alt_odb_tail = ent;
302 alt_odb_tail = &(ent->next);
303 ent->next = NULL;
304
305 /* recursively add alternates */
306 read_info_alternates(ent->base, depth + 1);
307
308 ent->base[pfxlen] = '/';
309
310 return 0;
311 }
312
313 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
314 const char *relative_base, int depth)
315 {
316 const char *cp, *last;
317
318 if (depth > 5) {
319 error("%s: ignoring alternate object stores, nesting too deep.",
320 relative_base);
321 return;
322 }
323
324 last = alt;
325 while (last < ep) {
326 cp = last;
327 if (cp < ep && *cp == '#') {
328 while (cp < ep && *cp != sep)
329 cp++;
330 last = cp + 1;
331 continue;
332 }
333 while (cp < ep && *cp != sep)
334 cp++;
335 if (last != cp) {
336 if ((*last != '/') && depth) {
337 error("%s: ignoring relative alternate object store %s",
338 relative_base, last);
339 } else {
340 link_alt_odb_entry(last, cp - last,
341 relative_base, depth);
342 }
343 }
344 while (cp < ep && *cp == sep)
345 cp++;
346 last = cp;
347 }
348 }
349
350 static void read_info_alternates(const char * relative_base, int depth)
351 {
352 char *map;
353 size_t mapsz;
354 struct stat st;
355 char path[PATH_MAX];
356 int fd;
357
358 sprintf(path, "%s/info/alternates", relative_base);
359 fd = open(path, O_RDONLY);
360 if (fd < 0)
361 return;
362 if (fstat(fd, &st) || (st.st_size == 0)) {
363 close(fd);
364 return;
365 }
366 mapsz = xsize_t(st.st_size);
367 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
368 close(fd);
369
370 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
371
372 munmap(map, mapsz);
373 }
374
375 void prepare_alt_odb(void)
376 {
377 const char *alt;
378
379 if (alt_odb_tail)
380 return;
381
382 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
383 if (!alt) alt = "";
384
385 alt_odb_tail = &alt_odb_list;
386 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
387
388 read_info_alternates(get_object_directory(), 0);
389 }
390
391 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
392 {
393 char *name = sha1_file_name(sha1);
394 struct alternate_object_database *alt;
395
396 if (!stat(name, st))
397 return name;
398 prepare_alt_odb();
399 for (alt = alt_odb_list; alt; alt = alt->next) {
400 name = alt->name;
401 fill_sha1_path(name, sha1);
402 if (!stat(alt->base, st))
403 return alt->base;
404 }
405 return NULL;
406 }
407
408 static unsigned int pack_used_ctr;
409 static unsigned int pack_mmap_calls;
410 static unsigned int peak_pack_open_windows;
411 static unsigned int pack_open_windows;
412 static size_t peak_pack_mapped;
413 static size_t pack_mapped;
414 struct packed_git *packed_git;
415
416 void pack_report(void)
417 {
418 fprintf(stderr,
419 "pack_report: getpagesize() = %10" SZ_FMT "\n"
420 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
421 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
422 (size_t) getpagesize(),
423 packed_git_window_size,
424 packed_git_limit);
425 fprintf(stderr,
426 "pack_report: pack_used_ctr = %10u\n"
427 "pack_report: pack_mmap_calls = %10u\n"
428 "pack_report: pack_open_windows = %10u / %10u\n"
429 "pack_report: pack_mapped = "
430 "%10" SZ_FMT " / %10" SZ_FMT "\n",
431 pack_used_ctr,
432 pack_mmap_calls,
433 pack_open_windows, peak_pack_open_windows,
434 pack_mapped, peak_pack_mapped);
435 }
436
437 static int check_packed_git_idx(const char *path, struct packed_git *p)
438 {
439 void *idx_map;
440 struct pack_idx_header *hdr;
441 size_t idx_size;
442 uint32_t version, nr, i, *index;
443 int fd = open(path, O_RDONLY);
444 struct stat st;
445
446 if (fd < 0)
447 return -1;
448 if (fstat(fd, &st)) {
449 close(fd);
450 return -1;
451 }
452 idx_size = xsize_t(st.st_size);
453 if (idx_size < 4 * 256 + 20 + 20) {
454 close(fd);
455 return error("index file %s is too small", path);
456 }
457 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
458 close(fd);
459
460 hdr = idx_map;
461 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
462 version = ntohl(hdr->idx_version);
463 if (version < 2 || version > 2) {
464 munmap(idx_map, idx_size);
465 return error("index file %s is version %d"
466 " and is not supported by this binary"
467 " (try upgrading GIT to a newer version)",
468 path, version);
469 }
470 } else
471 version = 1;
472
473 nr = 0;
474 index = idx_map;
475 if (version > 1)
476 index += 2; /* skip index header */
477 for (i = 0; i < 256; i++) {
478 uint32_t n = ntohl(index[i]);
479 if (n < nr) {
480 munmap(idx_map, idx_size);
481 return error("non-monotonic index %s", path);
482 }
483 nr = n;
484 }
485
486 if (version == 1) {
487 /*
488 * Total size:
489 * - 256 index entries 4 bytes each
490 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
491 * - 20-byte SHA1 of the packfile
492 * - 20-byte SHA1 file checksum
493 */
494 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
495 munmap(idx_map, idx_size);
496 return error("wrong index file size in %s", path);
497 }
498 } else if (version == 2) {
499 /*
500 * Minimum size:
501 * - 8 bytes of header
502 * - 256 index entries 4 bytes each
503 * - 20-byte sha1 entry * nr
504 * - 4-byte crc entry * nr
505 * - 4-byte offset entry * nr
506 * - 20-byte SHA1 of the packfile
507 * - 20-byte SHA1 file checksum
508 * And after the 4-byte offset table might be a
509 * variable sized table containing 8-byte entries
510 * for offsets larger than 2^31.
511 */
512 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
513 if (idx_size < min_size || idx_size > min_size + (nr - 1)*8) {
514 munmap(idx_map, idx_size);
515 return error("wrong index file size in %s", path);
516 }
517 if (idx_size != min_size) {
518 /* make sure we can deal with large pack offsets */
519 off_t x = 0x7fffffffUL, y = 0xffffffffUL;
520 if (x > (x + 1) || y > (y + 1)) {
521 munmap(idx_map, idx_size);
522 return error("pack too large for current definition of off_t in %s", path);
523 }
524 }
525 }
526
527 p->index_version = version;
528 p->index_data = idx_map;
529 p->index_size = idx_size;
530 p->num_objects = nr;
531 return 0;
532 }
533
534 int open_pack_index(struct packed_git *p)
535 {
536 char *idx_name;
537 int ret;
538
539 if (p->index_data)
540 return 0;
541
542 idx_name = xstrdup(p->pack_name);
543 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
544 ret = check_packed_git_idx(idx_name, p);
545 free(idx_name);
546 return ret;
547 }
548
549 static void scan_windows(struct packed_git *p,
550 struct packed_git **lru_p,
551 struct pack_window **lru_w,
552 struct pack_window **lru_l)
553 {
554 struct pack_window *w, *w_l;
555
556 for (w_l = NULL, w = p->windows; w; w = w->next) {
557 if (!w->inuse_cnt) {
558 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
559 *lru_p = p;
560 *lru_w = w;
561 *lru_l = w_l;
562 }
563 }
564 w_l = w;
565 }
566 }
567
568 static int unuse_one_window(struct packed_git *current, int keep_fd)
569 {
570 struct packed_git *p, *lru_p = NULL;
571 struct pack_window *lru_w = NULL, *lru_l = NULL;
572
573 if (current)
574 scan_windows(current, &lru_p, &lru_w, &lru_l);
575 for (p = packed_git; p; p = p->next)
576 scan_windows(p, &lru_p, &lru_w, &lru_l);
577 if (lru_p) {
578 munmap(lru_w->base, lru_w->len);
579 pack_mapped -= lru_w->len;
580 if (lru_l)
581 lru_l->next = lru_w->next;
582 else {
583 lru_p->windows = lru_w->next;
584 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
585 close(lru_p->pack_fd);
586 lru_p->pack_fd = -1;
587 }
588 }
589 free(lru_w);
590 pack_open_windows--;
591 return 1;
592 }
593 return 0;
594 }
595
596 void release_pack_memory(size_t need, int fd)
597 {
598 size_t cur = pack_mapped;
599 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
600 ; /* nothing */
601 }
602
603 void unuse_pack(struct pack_window **w_cursor)
604 {
605 struct pack_window *w = *w_cursor;
606 if (w) {
607 w->inuse_cnt--;
608 *w_cursor = NULL;
609 }
610 }
611
612 /*
613 * Do not call this directly as this leaks p->pack_fd on error return;
614 * call open_packed_git() instead.
615 */
616 static int open_packed_git_1(struct packed_git *p)
617 {
618 struct stat st;
619 struct pack_header hdr;
620 unsigned char sha1[20];
621 unsigned char *idx_sha1;
622 long fd_flag;
623
624 if (!p->index_data && open_pack_index(p))
625 return error("packfile %s index unavailable", p->pack_name);
626
627 p->pack_fd = open(p->pack_name, O_RDONLY);
628 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
629 return -1;
630
631 /* If we created the struct before we had the pack we lack size. */
632 if (!p->pack_size) {
633 if (!S_ISREG(st.st_mode))
634 return error("packfile %s not a regular file", p->pack_name);
635 p->pack_size = st.st_size;
636 } else if (p->pack_size != st.st_size)
637 return error("packfile %s size changed", p->pack_name);
638
639 /* We leave these file descriptors open with sliding mmap;
640 * there is no point keeping them open across exec(), though.
641 */
642 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
643 if (fd_flag < 0)
644 return error("cannot determine file descriptor flags");
645 fd_flag |= FD_CLOEXEC;
646 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
647 return error("cannot set FD_CLOEXEC");
648
649 /* Verify we recognize this pack file format. */
650 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
651 return error("file %s is far too short to be a packfile", p->pack_name);
652 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
653 return error("file %s is not a GIT packfile", p->pack_name);
654 if (!pack_version_ok(hdr.hdr_version))
655 return error("packfile %s is version %u and not supported"
656 " (try upgrading GIT to a newer version)",
657 p->pack_name, ntohl(hdr.hdr_version));
658
659 /* Verify the pack matches its index. */
660 if (p->num_objects != ntohl(hdr.hdr_entries))
661 return error("packfile %s claims to have %u objects"
662 " while index indicates %u objects",
663 p->pack_name, ntohl(hdr.hdr_entries),
664 p->num_objects);
665 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
666 return error("end of packfile %s is unavailable", p->pack_name);
667 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
668 return error("packfile %s signature is unavailable", p->pack_name);
669 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
670 if (hashcmp(sha1, idx_sha1))
671 return error("packfile %s does not match index", p->pack_name);
672 return 0;
673 }
674
675 static int open_packed_git(struct packed_git *p)
676 {
677 if (!open_packed_git_1(p))
678 return 0;
679 if (p->pack_fd != -1) {
680 close(p->pack_fd);
681 p->pack_fd = -1;
682 }
683 return -1;
684 }
685
686 static int in_window(struct pack_window *win, off_t offset)
687 {
688 /* We must promise at least 20 bytes (one hash) after the
689 * offset is available from this window, otherwise the offset
690 * is not actually in this window and a different window (which
691 * has that one hash excess) must be used. This is to support
692 * the object header and delta base parsing routines below.
693 */
694 off_t win_off = win->offset;
695 return win_off <= offset
696 && (offset + 20) <= (win_off + win->len);
697 }
698
699 unsigned char* use_pack(struct packed_git *p,
700 struct pack_window **w_cursor,
701 off_t offset,
702 unsigned int *left)
703 {
704 struct pack_window *win = *w_cursor;
705
706 if (p->pack_fd == -1 && open_packed_git(p))
707 die("packfile %s cannot be accessed", p->pack_name);
708
709 /* Since packfiles end in a hash of their content and its
710 * pointless to ask for an offset into the middle of that
711 * hash, and the in_window function above wouldn't match
712 * don't allow an offset too close to the end of the file.
713 */
714 if (offset > (p->pack_size - 20))
715 die("offset beyond end of packfile (truncated pack?)");
716
717 if (!win || !in_window(win, offset)) {
718 if (win)
719 win->inuse_cnt--;
720 for (win = p->windows; win; win = win->next) {
721 if (in_window(win, offset))
722 break;
723 }
724 if (!win) {
725 size_t window_align = packed_git_window_size / 2;
726 off_t len;
727 win = xcalloc(1, sizeof(*win));
728 win->offset = (offset / window_align) * window_align;
729 len = p->pack_size - win->offset;
730 if (len > packed_git_window_size)
731 len = packed_git_window_size;
732 win->len = (size_t)len;
733 pack_mapped += win->len;
734 while (packed_git_limit < pack_mapped
735 && unuse_one_window(p, p->pack_fd))
736 ; /* nothing */
737 win->base = xmmap(NULL, win->len,
738 PROT_READ, MAP_PRIVATE,
739 p->pack_fd, win->offset);
740 if (win->base == MAP_FAILED)
741 die("packfile %s cannot be mapped: %s",
742 p->pack_name,
743 strerror(errno));
744 pack_mmap_calls++;
745 pack_open_windows++;
746 if (pack_mapped > peak_pack_mapped)
747 peak_pack_mapped = pack_mapped;
748 if (pack_open_windows > peak_pack_open_windows)
749 peak_pack_open_windows = pack_open_windows;
750 win->next = p->windows;
751 p->windows = win;
752 }
753 }
754 if (win != *w_cursor) {
755 win->last_used = pack_used_ctr++;
756 win->inuse_cnt++;
757 *w_cursor = win;
758 }
759 offset -= win->offset;
760 if (left)
761 *left = win->len - xsize_t(offset);
762 return win->base + offset;
763 }
764
765 struct packed_git *add_packed_git(const char *path, int path_len, int local)
766 {
767 struct stat st;
768 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
769
770 /*
771 * Make sure a corresponding .pack file exists and that
772 * the index looks sane.
773 */
774 path_len -= strlen(".idx");
775 if (path_len < 1)
776 return NULL;
777 memcpy(p->pack_name, path, path_len);
778 strcpy(p->pack_name + path_len, ".pack");
779 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
780 free(p);
781 return NULL;
782 }
783
784 /* ok, it looks sane as far as we can check without
785 * actually mapping the pack file.
786 */
787 p->index_version = 0;
788 p->index_data = NULL;
789 p->index_size = 0;
790 p->num_objects = 0;
791 p->pack_size = st.st_size;
792 p->next = NULL;
793 p->windows = NULL;
794 p->pack_fd = -1;
795 p->pack_local = local;
796 p->mtime = st.st_mtime;
797 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
798 hashclr(p->sha1);
799 return p;
800 }
801
802 struct packed_git *parse_pack_index(unsigned char *sha1)
803 {
804 char *path = sha1_pack_index_name(sha1);
805 return parse_pack_index_file(sha1, path);
806 }
807
808 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
809 const char *idx_path)
810 {
811 const char *path = sha1_pack_name(sha1);
812 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
813
814 if (check_packed_git_idx(idx_path, p)) {
815 free(p);
816 return NULL;
817 }
818
819 strcpy(p->pack_name, path);
820 p->pack_size = 0;
821 p->next = NULL;
822 p->windows = NULL;
823 p->pack_fd = -1;
824 hashcpy(p->sha1, sha1);
825 return p;
826 }
827
828 void install_packed_git(struct packed_git *pack)
829 {
830 pack->next = packed_git;
831 packed_git = pack;
832 }
833
834 static void prepare_packed_git_one(char *objdir, int local)
835 {
836 char path[PATH_MAX];
837 int len;
838 DIR *dir;
839 struct dirent *de;
840
841 sprintf(path, "%s/pack", objdir);
842 len = strlen(path);
843 dir = opendir(path);
844 if (!dir) {
845 if (errno != ENOENT)
846 error("unable to open object pack directory: %s: %s",
847 path, strerror(errno));
848 return;
849 }
850 path[len++] = '/';
851 while ((de = readdir(dir)) != NULL) {
852 int namelen = strlen(de->d_name);
853 struct packed_git *p;
854
855 if (!has_extension(de->d_name, ".idx"))
856 continue;
857
858 /* Don't reopen a pack we already have. */
859 strcpy(path + len, de->d_name);
860 for (p = packed_git; p; p = p->next) {
861 if (!memcmp(path, p->pack_name, len + namelen - 4))
862 break;
863 }
864 if (p)
865 continue;
866 /* See if it really is a valid .idx file with corresponding
867 * .pack file that we can map.
868 */
869 p = add_packed_git(path, len + namelen, local);
870 if (!p)
871 continue;
872 install_packed_git(p);
873 }
874 closedir(dir);
875 }
876
877 static int sort_pack(const void *a_, const void *b_)
878 {
879 struct packed_git *a = *((struct packed_git **)a_);
880 struct packed_git *b = *((struct packed_git **)b_);
881 int st;
882
883 /*
884 * Local packs tend to contain objects specific to our
885 * variant of the project than remote ones. In addition,
886 * remote ones could be on a network mounted filesystem.
887 * Favor local ones for these reasons.
888 */
889 st = a->pack_local - b->pack_local;
890 if (st)
891 return -st;
892
893 /*
894 * Younger packs tend to contain more recent objects,
895 * and more recent objects tend to get accessed more
896 * often.
897 */
898 if (a->mtime < b->mtime)
899 return 1;
900 else if (a->mtime == b->mtime)
901 return 0;
902 return -1;
903 }
904
905 static void rearrange_packed_git(void)
906 {
907 struct packed_git **ary, *p;
908 int i, n;
909
910 for (n = 0, p = packed_git; p; p = p->next)
911 n++;
912 if (n < 2)
913 return;
914
915 /* prepare an array of packed_git for easier sorting */
916 ary = xcalloc(n, sizeof(struct packed_git *));
917 for (n = 0, p = packed_git; p; p = p->next)
918 ary[n++] = p;
919
920 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
921
922 /* link them back again */
923 for (i = 0; i < n - 1; i++)
924 ary[i]->next = ary[i + 1];
925 ary[n - 1]->next = NULL;
926 packed_git = ary[0];
927
928 free(ary);
929 }
930
931 static int prepare_packed_git_run_once = 0;
932 void prepare_packed_git(void)
933 {
934 struct alternate_object_database *alt;
935
936 if (prepare_packed_git_run_once)
937 return;
938 prepare_packed_git_one(get_object_directory(), 1);
939 prepare_alt_odb();
940 for (alt = alt_odb_list; alt; alt = alt->next) {
941 alt->name[-1] = 0;
942 prepare_packed_git_one(alt->base, 0);
943 alt->name[-1] = '/';
944 }
945 rearrange_packed_git();
946 prepare_packed_git_run_once = 1;
947 }
948
949 void reprepare_packed_git(void)
950 {
951 prepare_packed_git_run_once = 0;
952 prepare_packed_git();
953 }
954
955 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
956 {
957 unsigned char real_sha1[20];
958 hash_sha1_file(map, size, type, real_sha1);
959 return hashcmp(sha1, real_sha1) ? -1 : 0;
960 }
961
962 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
963 {
964 struct stat st;
965 void *map;
966 int fd;
967 char *filename = find_sha1_file(sha1, &st);
968
969 if (!filename) {
970 return NULL;
971 }
972
973 fd = open(filename, O_RDONLY | sha1_file_open_flag);
974 if (fd < 0) {
975 /* See if it works without O_NOATIME */
976 switch (sha1_file_open_flag) {
977 default:
978 fd = open(filename, O_RDONLY);
979 if (fd >= 0)
980 break;
981 /* Fallthrough */
982 case 0:
983 return NULL;
984 }
985
986 /* If it failed once, it will probably fail again.
987 * Stop using O_NOATIME
988 */
989 sha1_file_open_flag = 0;
990 }
991 *size = xsize_t(st.st_size);
992 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
993 close(fd);
994 return map;
995 }
996
997 static int legacy_loose_object(unsigned char *map)
998 {
999 unsigned int word;
1000
1001 /*
1002 * Is it a zlib-compressed buffer? If so, the first byte
1003 * must be 0x78 (15-bit window size, deflated), and the
1004 * first 16-bit word is evenly divisible by 31
1005 */
1006 word = (map[0] << 8) + map[1];
1007 if (map[0] == 0x78 && !(word % 31))
1008 return 1;
1009 else
1010 return 0;
1011 }
1012
1013 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1014 {
1015 unsigned shift;
1016 unsigned char c;
1017 unsigned long size;
1018 unsigned long used = 0;
1019
1020 c = buf[used++];
1021 *type = (c >> 4) & 7;
1022 size = c & 15;
1023 shift = 4;
1024 while (c & 0x80) {
1025 if (len <= used)
1026 return 0;
1027 if (sizeof(long) * 8 <= shift)
1028 return 0;
1029 c = buf[used++];
1030 size += (c & 0x7f) << shift;
1031 shift += 7;
1032 }
1033 *sizep = size;
1034 return used;
1035 }
1036
1037 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1038 {
1039 unsigned long size, used;
1040 static const char valid_loose_object_type[8] = {
1041 0, /* OBJ_EXT */
1042 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1043 0, /* "delta" and others are invalid in a loose object */
1044 };
1045 enum object_type type;
1046
1047 /* Get the data stream */
1048 memset(stream, 0, sizeof(*stream));
1049 stream->next_in = map;
1050 stream->avail_in = mapsize;
1051 stream->next_out = buffer;
1052 stream->avail_out = bufsiz;
1053
1054 if (legacy_loose_object(map)) {
1055 inflateInit(stream);
1056 return inflate(stream, 0);
1057 }
1058
1059
1060 /*
1061 * There used to be a second loose object header format which
1062 * was meant to mimic the in-pack format, allowing for direct
1063 * copy of the object data. This format turned up not to be
1064 * really worth it and we don't write it any longer. But we
1065 * can still read it.
1066 */
1067 used = unpack_object_header_gently(map, mapsize, &type, &size);
1068 if (!used || !valid_loose_object_type[type])
1069 return -1;
1070 map += used;
1071 mapsize -= used;
1072
1073 /* Set up the stream for the rest.. */
1074 stream->next_in = map;
1075 stream->avail_in = mapsize;
1076 inflateInit(stream);
1077
1078 /* And generate the fake traditional header */
1079 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1080 typename(type), size);
1081 return 0;
1082 }
1083
1084 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1085 {
1086 int bytes = strlen(buffer) + 1;
1087 unsigned char *buf = xmalloc(1+size);
1088 unsigned long n;
1089 int status = Z_OK;
1090
1091 n = stream->total_out - bytes;
1092 if (n > size)
1093 n = size;
1094 memcpy(buf, (char *) buffer + bytes, n);
1095 bytes = n;
1096 if (bytes <= size) {
1097 /*
1098 * The above condition must be (bytes <= size), not
1099 * (bytes < size). In other words, even though we
1100 * expect no more output and set avail_out to zer0,
1101 * the input zlib stream may have bytes that express
1102 * "this concludes the stream", and we *do* want to
1103 * eat that input.
1104 *
1105 * Otherwise we would not be able to test that we
1106 * consumed all the input to reach the expected size;
1107 * we also want to check that zlib tells us that all
1108 * went well with status == Z_STREAM_END at the end.
1109 */
1110 stream->next_out = buf + bytes;
1111 stream->avail_out = size - bytes;
1112 while (status == Z_OK)
1113 status = inflate(stream, Z_FINISH);
1114 }
1115 buf[size] = 0;
1116 if (status == Z_STREAM_END && !stream->avail_in) {
1117 inflateEnd(stream);
1118 return buf;
1119 }
1120
1121 if (status < 0)
1122 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1123 else if (stream->avail_in)
1124 error("garbage at end of loose object '%s'",
1125 sha1_to_hex(sha1));
1126 free(buf);
1127 return NULL;
1128 }
1129
1130 /*
1131 * We used to just use "sscanf()", but that's actually way
1132 * too permissive for what we want to check. So do an anal
1133 * object header parse by hand.
1134 */
1135 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1136 {
1137 char type[10];
1138 int i;
1139 unsigned long size;
1140
1141 /*
1142 * The type can be at most ten bytes (including the
1143 * terminating '\0' that we add), and is followed by
1144 * a space.
1145 */
1146 i = 0;
1147 for (;;) {
1148 char c = *hdr++;
1149 if (c == ' ')
1150 break;
1151 type[i++] = c;
1152 if (i >= sizeof(type))
1153 return -1;
1154 }
1155 type[i] = 0;
1156
1157 /*
1158 * The length must follow immediately, and be in canonical
1159 * decimal format (ie "010" is not valid).
1160 */
1161 size = *hdr++ - '0';
1162 if (size > 9)
1163 return -1;
1164 if (size) {
1165 for (;;) {
1166 unsigned long c = *hdr - '0';
1167 if (c > 9)
1168 break;
1169 hdr++;
1170 size = size * 10 + c;
1171 }
1172 }
1173 *sizep = size;
1174
1175 /*
1176 * The length must be followed by a zero byte
1177 */
1178 return *hdr ? -1 : type_from_string(type);
1179 }
1180
1181 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1182 {
1183 int ret;
1184 z_stream stream;
1185 char hdr[8192];
1186
1187 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1188 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1189 return NULL;
1190
1191 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1192 }
1193
1194 unsigned long get_size_from_delta(struct packed_git *p,
1195 struct pack_window **w_curs,
1196 off_t curpos)
1197 {
1198 const unsigned char *data;
1199 unsigned char delta_head[20], *in;
1200 z_stream stream;
1201 int st;
1202
1203 memset(&stream, 0, sizeof(stream));
1204 stream.next_out = delta_head;
1205 stream.avail_out = sizeof(delta_head);
1206
1207 inflateInit(&stream);
1208 do {
1209 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1210 stream.next_in = in;
1211 st = inflate(&stream, Z_FINISH);
1212 curpos += stream.next_in - in;
1213 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1214 stream.total_out < sizeof(delta_head));
1215 inflateEnd(&stream);
1216 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1217 die("delta data unpack-initial failed");
1218
1219 /* Examine the initial part of the delta to figure out
1220 * the result size.
1221 */
1222 data = delta_head;
1223
1224 /* ignore base size */
1225 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1226
1227 /* Read the result size */
1228 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1229 }
1230
1231 static off_t get_delta_base(struct packed_git *p,
1232 struct pack_window **w_curs,
1233 off_t *curpos,
1234 enum object_type type,
1235 off_t delta_obj_offset)
1236 {
1237 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1238 off_t base_offset;
1239
1240 /* use_pack() assured us we have [base_info, base_info + 20)
1241 * as a range that we can look at without walking off the
1242 * end of the mapped window. Its actually the hash size
1243 * that is assured. An OFS_DELTA longer than the hash size
1244 * is stupid, as then a REF_DELTA would be smaller to store.
1245 */
1246 if (type == OBJ_OFS_DELTA) {
1247 unsigned used = 0;
1248 unsigned char c = base_info[used++];
1249 base_offset = c & 127;
1250 while (c & 128) {
1251 base_offset += 1;
1252 if (!base_offset || MSB(base_offset, 7))
1253 die("offset value overflow for delta base object");
1254 c = base_info[used++];
1255 base_offset = (base_offset << 7) + (c & 127);
1256 }
1257 base_offset = delta_obj_offset - base_offset;
1258 if (base_offset >= delta_obj_offset)
1259 die("delta base offset out of bound");
1260 *curpos += used;
1261 } else if (type == OBJ_REF_DELTA) {
1262 /* The base entry _must_ be in the same pack */
1263 base_offset = find_pack_entry_one(base_info, p);
1264 if (!base_offset)
1265 die("failed to find delta-pack base object %s",
1266 sha1_to_hex(base_info));
1267 *curpos += 20;
1268 } else
1269 die("I am totally screwed");
1270 return base_offset;
1271 }
1272
1273 /* forward declaration for a mutually recursive function */
1274 static int packed_object_info(struct packed_git *p, off_t offset,
1275 unsigned long *sizep);
1276
1277 static int packed_delta_info(struct packed_git *p,
1278 struct pack_window **w_curs,
1279 off_t curpos,
1280 enum object_type type,
1281 off_t obj_offset,
1282 unsigned long *sizep)
1283 {
1284 off_t base_offset;
1285
1286 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1287 type = packed_object_info(p, base_offset, NULL);
1288
1289 /* We choose to only get the type of the base object and
1290 * ignore potentially corrupt pack file that expects the delta
1291 * based on a base with a wrong size. This saves tons of
1292 * inflate() calls.
1293 */
1294 if (sizep)
1295 *sizep = get_size_from_delta(p, w_curs, curpos);
1296
1297 return type;
1298 }
1299
1300 static int unpack_object_header(struct packed_git *p,
1301 struct pack_window **w_curs,
1302 off_t *curpos,
1303 unsigned long *sizep)
1304 {
1305 unsigned char *base;
1306 unsigned int left;
1307 unsigned long used;
1308 enum object_type type;
1309
1310 /* use_pack() assures us we have [base, base + 20) available
1311 * as a range that we can look at at. (Its actually the hash
1312 * size that is assured.) With our object header encoding
1313 * the maximum deflated object size is 2^137, which is just
1314 * insane, so we know won't exceed what we have been given.
1315 */
1316 base = use_pack(p, w_curs, *curpos, &left);
1317 used = unpack_object_header_gently(base, left, &type, sizep);
1318 if (!used)
1319 die("object offset outside of pack file");
1320 *curpos += used;
1321
1322 return type;
1323 }
1324
1325 const char *packed_object_info_detail(struct packed_git *p,
1326 off_t obj_offset,
1327 unsigned long *size,
1328 unsigned long *store_size,
1329 unsigned int *delta_chain_length,
1330 unsigned char *base_sha1)
1331 {
1332 struct pack_window *w_curs = NULL;
1333 off_t curpos;
1334 unsigned long dummy;
1335 unsigned char *next_sha1;
1336 enum object_type type;
1337
1338 *delta_chain_length = 0;
1339 curpos = obj_offset;
1340 type = unpack_object_header(p, &w_curs, &curpos, size);
1341
1342 for (;;) {
1343 switch (type) {
1344 default:
1345 die("pack %s contains unknown object type %d",
1346 p->pack_name, type);
1347 case OBJ_COMMIT:
1348 case OBJ_TREE:
1349 case OBJ_BLOB:
1350 case OBJ_TAG:
1351 *store_size = 0; /* notyet */
1352 unuse_pack(&w_curs);
1353 return typename(type);
1354 case OBJ_OFS_DELTA:
1355 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1356 if (*delta_chain_length == 0) {
1357 /* TODO: find base_sha1 as pointed by curpos */
1358 hashclr(base_sha1);
1359 }
1360 break;
1361 case OBJ_REF_DELTA:
1362 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1363 if (*delta_chain_length == 0)
1364 hashcpy(base_sha1, next_sha1);
1365 obj_offset = find_pack_entry_one(next_sha1, p);
1366 break;
1367 }
1368 (*delta_chain_length)++;
1369 curpos = obj_offset;
1370 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1371 }
1372 }
1373
1374 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1375 unsigned long *sizep)
1376 {
1377 struct pack_window *w_curs = NULL;
1378 unsigned long size;
1379 off_t curpos = obj_offset;
1380 enum object_type type;
1381
1382 type = unpack_object_header(p, &w_curs, &curpos, &size);
1383
1384 switch (type) {
1385 case OBJ_OFS_DELTA:
1386 case OBJ_REF_DELTA:
1387 type = packed_delta_info(p, &w_curs, curpos,
1388 type, obj_offset, sizep);
1389 break;
1390 case OBJ_COMMIT:
1391 case OBJ_TREE:
1392 case OBJ_BLOB:
1393 case OBJ_TAG:
1394 if (sizep)
1395 *sizep = size;
1396 break;
1397 default:
1398 die("pack %s contains unknown object type %d",
1399 p->pack_name, type);
1400 }
1401 unuse_pack(&w_curs);
1402 return type;
1403 }
1404
1405 static void *unpack_compressed_entry(struct packed_git *p,
1406 struct pack_window **w_curs,
1407 off_t curpos,
1408 unsigned long size)
1409 {
1410 int st;
1411 z_stream stream;
1412 unsigned char *buffer, *in;
1413
1414 buffer = xmalloc(size + 1);
1415 buffer[size] = 0;
1416 memset(&stream, 0, sizeof(stream));
1417 stream.next_out = buffer;
1418 stream.avail_out = size;
1419
1420 inflateInit(&stream);
1421 do {
1422 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1423 stream.next_in = in;
1424 st = inflate(&stream, Z_FINISH);
1425 curpos += stream.next_in - in;
1426 } while (st == Z_OK || st == Z_BUF_ERROR);
1427 inflateEnd(&stream);
1428 if ((st != Z_STREAM_END) || stream.total_out != size) {
1429 free(buffer);
1430 return NULL;
1431 }
1432
1433 return buffer;
1434 }
1435
1436 #define MAX_DELTA_CACHE (256)
1437
1438 static size_t delta_base_cached;
1439
1440 static struct delta_base_cache_lru_list {
1441 struct delta_base_cache_lru_list *prev;
1442 struct delta_base_cache_lru_list *next;
1443 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1444
1445 static struct delta_base_cache_entry {
1446 struct delta_base_cache_lru_list lru;
1447 void *data;
1448 struct packed_git *p;
1449 off_t base_offset;
1450 unsigned long size;
1451 enum object_type type;
1452 } delta_base_cache[MAX_DELTA_CACHE];
1453
1454 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1455 {
1456 unsigned long hash;
1457
1458 hash = (unsigned long)p + (unsigned long)base_offset;
1459 hash += (hash >> 8) + (hash >> 16);
1460 return hash % MAX_DELTA_CACHE;
1461 }
1462
1463 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1464 unsigned long *base_size, enum object_type *type, int keep_cache)
1465 {
1466 void *ret;
1467 unsigned long hash = pack_entry_hash(p, base_offset);
1468 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1469
1470 ret = ent->data;
1471 if (ret && ent->p == p && ent->base_offset == base_offset)
1472 goto found_cache_entry;
1473 return unpack_entry(p, base_offset, type, base_size);
1474
1475 found_cache_entry:
1476 if (!keep_cache) {
1477 ent->data = NULL;
1478 ent->lru.next->prev = ent->lru.prev;
1479 ent->lru.prev->next = ent->lru.next;
1480 delta_base_cached -= ent->size;
1481 }
1482 else {
1483 ret = xmalloc(ent->size + 1);
1484 memcpy(ret, ent->data, ent->size);
1485 ((char *)ret)[ent->size] = 0;
1486 }
1487 *type = ent->type;
1488 *base_size = ent->size;
1489 return ret;
1490 }
1491
1492 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1493 {
1494 if (ent->data) {
1495 free(ent->data);
1496 ent->data = NULL;
1497 ent->lru.next->prev = ent->lru.prev;
1498 ent->lru.prev->next = ent->lru.next;
1499 delta_base_cached -= ent->size;
1500 }
1501 }
1502
1503 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1504 void *base, unsigned long base_size, enum object_type type)
1505 {
1506 unsigned long hash = pack_entry_hash(p, base_offset);
1507 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1508 struct delta_base_cache_lru_list *lru;
1509
1510 release_delta_base_cache(ent);
1511 delta_base_cached += base_size;
1512
1513 for (lru = delta_base_cache_lru.next;
1514 delta_base_cached > delta_base_cache_limit
1515 && lru != &delta_base_cache_lru;
1516 lru = lru->next) {
1517 struct delta_base_cache_entry *f = (void *)lru;
1518 if (f->type == OBJ_BLOB)
1519 release_delta_base_cache(f);
1520 }
1521 for (lru = delta_base_cache_lru.next;
1522 delta_base_cached > delta_base_cache_limit
1523 && lru != &delta_base_cache_lru;
1524 lru = lru->next) {
1525 struct delta_base_cache_entry *f = (void *)lru;
1526 release_delta_base_cache(f);
1527 }
1528
1529 ent->p = p;
1530 ent->base_offset = base_offset;
1531 ent->type = type;
1532 ent->data = base;
1533 ent->size = base_size;
1534 ent->lru.next = &delta_base_cache_lru;
1535 ent->lru.prev = delta_base_cache_lru.prev;
1536 delta_base_cache_lru.prev->next = &ent->lru;
1537 delta_base_cache_lru.prev = &ent->lru;
1538 }
1539
1540 static void *unpack_delta_entry(struct packed_git *p,
1541 struct pack_window **w_curs,
1542 off_t curpos,
1543 unsigned long delta_size,
1544 off_t obj_offset,
1545 enum object_type *type,
1546 unsigned long *sizep)
1547 {
1548 void *delta_data, *result, *base;
1549 unsigned long base_size;
1550 off_t base_offset;
1551
1552 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1553 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1554 if (!base)
1555 die("failed to read delta base object"
1556 " at %"PRIuMAX" from %s",
1557 (uintmax_t)base_offset, p->pack_name);
1558
1559 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1560 result = patch_delta(base, base_size,
1561 delta_data, delta_size,
1562 sizep);
1563 if (!result)
1564 die("failed to apply delta");
1565 free(delta_data);
1566 add_delta_base_cache(p, base_offset, base, base_size, *type);
1567 return result;
1568 }
1569
1570 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1571 enum object_type *type, unsigned long *sizep)
1572 {
1573 struct pack_window *w_curs = NULL;
1574 off_t curpos = obj_offset;
1575 void *data;
1576
1577 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1578 switch (*type) {
1579 case OBJ_OFS_DELTA:
1580 case OBJ_REF_DELTA:
1581 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1582 obj_offset, type, sizep);
1583 break;
1584 case OBJ_COMMIT:
1585 case OBJ_TREE:
1586 case OBJ_BLOB:
1587 case OBJ_TAG:
1588 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1589 break;
1590 default:
1591 die("unknown object type %i in %s", *type, p->pack_name);
1592 }
1593 unuse_pack(&w_curs);
1594 return data;
1595 }
1596
1597 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1598 uint32_t n)
1599 {
1600 const unsigned char *index = p->index_data;
1601 if (!index) {
1602 if (open_pack_index(p))
1603 return NULL;
1604 index = p->index_data;
1605 }
1606 if (n >= p->num_objects)
1607 return NULL;
1608 index += 4 * 256;
1609 if (p->index_version == 1) {
1610 return index + 24 * n + 4;
1611 } else {
1612 index += 8;
1613 return index + 20 * n;
1614 }
1615 }
1616
1617 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1618 {
1619 const unsigned char *index = p->index_data;
1620 index += 4 * 256;
1621 if (p->index_version == 1) {
1622 return ntohl(*((uint32_t *)(index + 24 * n)));
1623 } else {
1624 uint32_t off;
1625 index += 8 + p->num_objects * (20 + 4);
1626 off = ntohl(*((uint32_t *)(index + 4 * n)));
1627 if (!(off & 0x80000000))
1628 return off;
1629 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1630 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1631 ntohl(*((uint32_t *)(index + 4)));
1632 }
1633 }
1634
1635 off_t find_pack_entry_one(const unsigned char *sha1,
1636 struct packed_git *p)
1637 {
1638 const uint32_t *level1_ofs = p->index_data;
1639 const unsigned char *index = p->index_data;
1640 unsigned hi, lo;
1641
1642 if (!index) {
1643 if (open_pack_index(p))
1644 return 0;
1645 level1_ofs = p->index_data;
1646 index = p->index_data;
1647 }
1648 if (p->index_version > 1) {
1649 level1_ofs += 2;
1650 index += 8;
1651 }
1652 index += 4 * 256;
1653 hi = ntohl(level1_ofs[*sha1]);
1654 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1655
1656 do {
1657 unsigned mi = (lo + hi) / 2;
1658 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1659 int cmp = hashcmp(index + x, sha1);
1660 if (!cmp)
1661 return nth_packed_object_offset(p, mi);
1662 if (cmp > 0)
1663 hi = mi;
1664 else
1665 lo = mi+1;
1666 } while (lo < hi);
1667 return 0;
1668 }
1669
1670 static int matches_pack_name(struct packed_git *p, const char *ig)
1671 {
1672 const char *last_c, *c;
1673
1674 if (!strcmp(p->pack_name, ig))
1675 return 0;
1676
1677 for (c = p->pack_name, last_c = c; *c;)
1678 if (*c == '/')
1679 last_c = ++c;
1680 else
1681 ++c;
1682 if (!strcmp(last_c, ig))
1683 return 0;
1684
1685 return 1;
1686 }
1687
1688 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1689 {
1690 static struct packed_git *last_found = (void *)1;
1691 struct packed_git *p;
1692 off_t offset;
1693
1694 prepare_packed_git();
1695 if (!packed_git)
1696 return 0;
1697 p = (last_found == (void *)1) ? packed_git : last_found;
1698
1699 do {
1700 if (ignore_packed) {
1701 const char **ig;
1702 for (ig = ignore_packed; *ig; ig++)
1703 if (!matches_pack_name(p, *ig))
1704 break;
1705 if (*ig)
1706 goto next;
1707 }
1708
1709 offset = find_pack_entry_one(sha1, p);
1710 if (offset) {
1711 /*
1712 * We are about to tell the caller where they can
1713 * locate the requested object. We better make
1714 * sure the packfile is still here and can be
1715 * accessed before supplying that answer, as
1716 * it may have been deleted since the index
1717 * was loaded!
1718 */
1719 if (p->pack_fd == -1 && open_packed_git(p)) {
1720 error("packfile %s cannot be accessed", p->pack_name);
1721 goto next;
1722 }
1723 e->offset = offset;
1724 e->p = p;
1725 hashcpy(e->sha1, sha1);
1726 last_found = p;
1727 return 1;
1728 }
1729
1730 next:
1731 if (p == last_found)
1732 p = packed_git;
1733 else
1734 p = p->next;
1735 if (p == last_found)
1736 p = p->next;
1737 } while (p);
1738 return 0;
1739 }
1740
1741 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1742 struct packed_git *packs)
1743 {
1744 struct packed_git *p;
1745
1746 for (p = packs; p; p = p->next) {
1747 if (find_pack_entry_one(sha1, p))
1748 return p;
1749 }
1750 return NULL;
1751
1752 }
1753
1754 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1755 {
1756 int status;
1757 unsigned long mapsize, size;
1758 void *map;
1759 z_stream stream;
1760 char hdr[32];
1761
1762 map = map_sha1_file(sha1, &mapsize);
1763 if (!map)
1764 return error("unable to find %s", sha1_to_hex(sha1));
1765 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1766 status = error("unable to unpack %s header",
1767 sha1_to_hex(sha1));
1768 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1769 status = error("unable to parse %s header", sha1_to_hex(sha1));
1770 else if (sizep)
1771 *sizep = size;
1772 inflateEnd(&stream);
1773 munmap(map, mapsize);
1774 return status;
1775 }
1776
1777 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1778 {
1779 struct pack_entry e;
1780
1781 if (!find_pack_entry(sha1, &e, NULL)) {
1782 reprepare_packed_git();
1783 if (!find_pack_entry(sha1, &e, NULL))
1784 return sha1_loose_object_info(sha1, sizep);
1785 }
1786 return packed_object_info(e.p, e.offset, sizep);
1787 }
1788
1789 static void *read_packed_sha1(const unsigned char *sha1,
1790 enum object_type *type, unsigned long *size)
1791 {
1792 struct pack_entry e;
1793
1794 if (!find_pack_entry(sha1, &e, NULL))
1795 return NULL;
1796 else
1797 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1798 }
1799
1800 /*
1801 * This is meant to hold a *small* number of objects that you would
1802 * want read_sha1_file() to be able to return, but yet you do not want
1803 * to write them into the object store (e.g. a browse-only
1804 * application).
1805 */
1806 static struct cached_object {
1807 unsigned char sha1[20];
1808 enum object_type type;
1809 void *buf;
1810 unsigned long size;
1811 } *cached_objects;
1812 static int cached_object_nr, cached_object_alloc;
1813
1814 static struct cached_object *find_cached_object(const unsigned char *sha1)
1815 {
1816 int i;
1817 struct cached_object *co = cached_objects;
1818
1819 for (i = 0; i < cached_object_nr; i++, co++) {
1820 if (!hashcmp(co->sha1, sha1))
1821 return co;
1822 }
1823 return NULL;
1824 }
1825
1826 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1827 unsigned char *sha1)
1828 {
1829 struct cached_object *co;
1830
1831 hash_sha1_file(buf, len, typename(type), sha1);
1832 if (has_sha1_file(sha1) || find_cached_object(sha1))
1833 return 0;
1834 if (cached_object_alloc <= cached_object_nr) {
1835 cached_object_alloc = alloc_nr(cached_object_alloc);
1836 cached_objects = xrealloc(cached_objects,
1837 sizeof(*cached_objects) *
1838 cached_object_alloc);
1839 }
1840 co = &cached_objects[cached_object_nr++];
1841 co->size = len;
1842 co->type = type;
1843 co->buf = xmalloc(len);
1844 memcpy(co->buf, buf, len);
1845 hashcpy(co->sha1, sha1);
1846 return 0;
1847 }
1848
1849 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1850 unsigned long *size)
1851 {
1852 unsigned long mapsize;
1853 void *map, *buf;
1854 struct cached_object *co;
1855
1856 co = find_cached_object(sha1);
1857 if (co) {
1858 buf = xmalloc(co->size + 1);
1859 memcpy(buf, co->buf, co->size);
1860 ((char*)buf)[co->size] = 0;
1861 *type = co->type;
1862 *size = co->size;
1863 return buf;
1864 }
1865
1866 buf = read_packed_sha1(sha1, type, size);
1867 if (buf)
1868 return buf;
1869 map = map_sha1_file(sha1, &mapsize);
1870 if (map) {
1871 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1872 munmap(map, mapsize);
1873 return buf;
1874 }
1875 reprepare_packed_git();
1876 return read_packed_sha1(sha1, type, size);
1877 }
1878
1879 void *read_object_with_reference(const unsigned char *sha1,
1880 const char *required_type_name,
1881 unsigned long *size,
1882 unsigned char *actual_sha1_return)
1883 {
1884 enum object_type type, required_type;
1885 void *buffer;
1886 unsigned long isize;
1887 unsigned char actual_sha1[20];
1888
1889 required_type = type_from_string(required_type_name);
1890 hashcpy(actual_sha1, sha1);
1891 while (1) {
1892 int ref_length = -1;
1893 const char *ref_type = NULL;
1894
1895 buffer = read_sha1_file(actual_sha1, &type, &isize);
1896 if (!buffer)
1897 return NULL;
1898 if (type == required_type) {
1899 *size = isize;
1900 if (actual_sha1_return)
1901 hashcpy(actual_sha1_return, actual_sha1);
1902 return buffer;
1903 }
1904 /* Handle references */
1905 else if (type == OBJ_COMMIT)
1906 ref_type = "tree ";
1907 else if (type == OBJ_TAG)
1908 ref_type = "object ";
1909 else {
1910 free(buffer);
1911 return NULL;
1912 }
1913 ref_length = strlen(ref_type);
1914
1915 if (memcmp(buffer, ref_type, ref_length) ||
1916 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1917 free(buffer);
1918 return NULL;
1919 }
1920 free(buffer);
1921 /* Now we have the ID of the referred-to object in
1922 * actual_sha1. Check again. */
1923 }
1924 }
1925
1926 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1927 const char *type, unsigned char *sha1,
1928 char *hdr, int *hdrlen)
1929 {
1930 SHA_CTX c;
1931
1932 /* Generate the header */
1933 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1934
1935 /* Sha1.. */
1936 SHA1_Init(&c);
1937 SHA1_Update(&c, hdr, *hdrlen);
1938 SHA1_Update(&c, buf, len);
1939 SHA1_Final(sha1, &c);
1940 }
1941
1942 /*
1943 * Link the tempfile to the final place, possibly creating the
1944 * last directory level as you do so.
1945 *
1946 * Returns the errno on failure, 0 on success.
1947 */
1948 static int link_temp_to_file(const char *tmpfile, const char *filename)
1949 {
1950 int ret;
1951 char *dir;
1952
1953 if (!link(tmpfile, filename))
1954 return 0;
1955
1956 /*
1957 * Try to mkdir the last path component if that failed.
1958 *
1959 * Re-try the "link()" regardless of whether the mkdir
1960 * succeeds, since a race might mean that somebody
1961 * else succeeded.
1962 */
1963 ret = errno;
1964 dir = strrchr(filename, '/');
1965 if (dir) {
1966 *dir = 0;
1967 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1968 *dir = '/';
1969 return -2;
1970 }
1971 *dir = '/';
1972 if (!link(tmpfile, filename))
1973 return 0;
1974 ret = errno;
1975 }
1976 return ret;
1977 }
1978
1979 /*
1980 * Move the just written object into its final resting place
1981 */
1982 int move_temp_to_file(const char *tmpfile, const char *filename)
1983 {
1984 int ret = link_temp_to_file(tmpfile, filename);
1985
1986 /*
1987 * Coda hack - coda doesn't like cross-directory links,
1988 * so we fall back to a rename, which will mean that it
1989 * won't be able to check collisions, but that's not a
1990 * big deal.
1991 *
1992 * The same holds for FAT formatted media.
1993 *
1994 * When this succeeds, we just return 0. We have nothing
1995 * left to unlink.
1996 */
1997 if (ret && ret != EEXIST) {
1998 if (!rename(tmpfile, filename))
1999 return 0;
2000 ret = errno;
2001 }
2002 unlink(tmpfile);
2003 if (ret) {
2004 if (ret != EEXIST) {
2005 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2006 }
2007 /* FIXME!!! Collision check here ? */
2008 }
2009
2010 return 0;
2011 }
2012
2013 static int write_buffer(int fd, const void *buf, size_t len)
2014 {
2015 if (write_in_full(fd, buf, len) < 0)
2016 return error("file write error (%s)", strerror(errno));
2017 return 0;
2018 }
2019
2020 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2021 unsigned char *sha1)
2022 {
2023 char hdr[32];
2024 int hdrlen;
2025 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2026 return 0;
2027 }
2028
2029 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2030 {
2031 int size, ret;
2032 unsigned char *compressed;
2033 z_stream stream;
2034 unsigned char sha1[20];
2035 char *filename;
2036 static char tmpfile[PATH_MAX];
2037 char hdr[32];
2038 int fd, hdrlen;
2039
2040 /* Normally if we have it in the pack then we do not bother writing
2041 * it out into .git/objects/??/?{38} file.
2042 */
2043 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2044 filename = sha1_file_name(sha1);
2045 if (returnsha1)
2046 hashcpy(returnsha1, sha1);
2047 if (has_sha1_file(sha1))
2048 return 0;
2049 fd = open(filename, O_RDONLY);
2050 if (fd >= 0) {
2051 /*
2052 * FIXME!!! We might do collision checking here, but we'd
2053 * need to uncompress the old file and check it. Later.
2054 */
2055 close(fd);
2056 return 0;
2057 }
2058
2059 if (errno != ENOENT) {
2060 return error("sha1 file %s: %s\n", filename, strerror(errno));
2061 }
2062
2063 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2064
2065 fd = mkstemp(tmpfile);
2066 if (fd < 0) {
2067 if (errno == EPERM)
2068 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2069 else
2070 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2071 }
2072
2073 /* Set it up */
2074 memset(&stream, 0, sizeof(stream));
2075 deflateInit(&stream, zlib_compression_level);
2076 size = 8 + deflateBound(&stream, len+hdrlen);
2077 compressed = xmalloc(size);
2078
2079 /* Compress it */
2080 stream.next_out = compressed;
2081 stream.avail_out = size;
2082
2083 /* First header.. */
2084 stream.next_in = (unsigned char *)hdr;
2085 stream.avail_in = hdrlen;
2086 while (deflate(&stream, 0) == Z_OK)
2087 /* nothing */;
2088
2089 /* Then the data itself.. */
2090 stream.next_in = buf;
2091 stream.avail_in = len;
2092 ret = deflate(&stream, Z_FINISH);
2093 if (ret != Z_STREAM_END)
2094 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2095
2096 ret = deflateEnd(&stream);
2097 if (ret != Z_OK)
2098 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2099
2100 size = stream.total_out;
2101
2102 if (write_buffer(fd, compressed, size) < 0)
2103 die("unable to write sha1 file");
2104 fchmod(fd, 0444);
2105 if (close(fd))
2106 die("unable to write sha1 file");
2107 free(compressed);
2108
2109 return move_temp_to_file(tmpfile, filename);
2110 }
2111
2112 /*
2113 * We need to unpack and recompress the object for writing
2114 * it out to a different file.
2115 */
2116 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2117 {
2118 size_t size;
2119 z_stream stream;
2120 unsigned char *unpacked;
2121 unsigned long len;
2122 enum object_type type;
2123 char hdr[32];
2124 int hdrlen;
2125 void *buf;
2126
2127 /* need to unpack and recompress it by itself */
2128 unpacked = read_packed_sha1(sha1, &type, &len);
2129 if (!unpacked)
2130 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2131
2132 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2133
2134 /* Set it up */
2135 memset(&stream, 0, sizeof(stream));
2136 deflateInit(&stream, zlib_compression_level);
2137 size = deflateBound(&stream, len + hdrlen);
2138 buf = xmalloc(size);
2139
2140 /* Compress it */
2141 stream.next_out = buf;
2142 stream.avail_out = size;
2143
2144 /* First header.. */
2145 stream.next_in = (void *)hdr;
2146 stream.avail_in = hdrlen;
2147 while (deflate(&stream, 0) == Z_OK)
2148 /* nothing */;
2149
2150 /* Then the data itself.. */
2151 stream.next_in = unpacked;
2152 stream.avail_in = len;
2153 while (deflate(&stream, Z_FINISH) == Z_OK)
2154 /* nothing */;
2155 deflateEnd(&stream);
2156 free(unpacked);
2157
2158 *objsize = stream.total_out;
2159 return buf;
2160 }
2161
2162 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2163 {
2164 int retval;
2165 unsigned long objsize;
2166 void *buf = map_sha1_file(sha1, &objsize);
2167
2168 if (buf) {
2169 retval = write_buffer(fd, buf, objsize);
2170 munmap(buf, objsize);
2171 return retval;
2172 }
2173
2174 buf = repack_object(sha1, &objsize);
2175 retval = write_buffer(fd, buf, objsize);
2176 free(buf);
2177 return retval;
2178 }
2179
2180 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2181 size_t bufsize, size_t *bufposn)
2182 {
2183 char tmpfile[PATH_MAX];
2184 int local;
2185 z_stream stream;
2186 unsigned char real_sha1[20];
2187 unsigned char discard[4096];
2188 int ret;
2189 SHA_CTX c;
2190
2191 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2192
2193 local = mkstemp(tmpfile);
2194 if (local < 0) {
2195 if (errno == EPERM)
2196 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2197 else
2198 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2199 }
2200
2201 memset(&stream, 0, sizeof(stream));
2202
2203 inflateInit(&stream);
2204
2205 SHA1_Init(&c);
2206
2207 do {
2208 ssize_t size;
2209 if (*bufposn) {
2210 stream.avail_in = *bufposn;
2211 stream.next_in = (unsigned char *) buffer;
2212 do {
2213 stream.next_out = discard;
2214 stream.avail_out = sizeof(discard);
2215 ret = inflate(&stream, Z_SYNC_FLUSH);
2216 SHA1_Update(&c, discard, sizeof(discard) -
2217 stream.avail_out);
2218 } while (stream.avail_in && ret == Z_OK);
2219 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2220 die("unable to write sha1 file");
2221 memmove(buffer, buffer + *bufposn - stream.avail_in,
2222 stream.avail_in);
2223 *bufposn = stream.avail_in;
2224 if (ret != Z_OK)
2225 break;
2226 }
2227 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2228 if (size <= 0) {
2229 close(local);
2230 unlink(tmpfile);
2231 if (!size)
2232 return error("Connection closed?");
2233 perror("Reading from connection");
2234 return -1;
2235 }
2236 *bufposn += size;
2237 } while (1);
2238 inflateEnd(&stream);
2239
2240 fchmod(local, 0444);
2241 if (close(local) != 0)
2242 die("unable to write sha1 file");
2243 SHA1_Final(real_sha1, &c);
2244 if (ret != Z_STREAM_END) {
2245 unlink(tmpfile);
2246 return error("File %s corrupted", sha1_to_hex(sha1));
2247 }
2248 if (hashcmp(sha1, real_sha1)) {
2249 unlink(tmpfile);
2250 return error("File %s has bad hash", sha1_to_hex(sha1));
2251 }
2252
2253 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2254 }
2255
2256 int has_pack_index(const unsigned char *sha1)
2257 {
2258 struct stat st;
2259 if (stat(sha1_pack_index_name(sha1), &st))
2260 return 0;
2261 return 1;
2262 }
2263
2264 int has_pack_file(const unsigned char *sha1)
2265 {
2266 struct stat st;
2267 if (stat(sha1_pack_name(sha1), &st))
2268 return 0;
2269 return 1;
2270 }
2271
2272 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2273 {
2274 struct pack_entry e;
2275 return find_pack_entry(sha1, &e, ignore_packed);
2276 }
2277
2278 int has_sha1_file(const unsigned char *sha1)
2279 {
2280 struct stat st;
2281 struct pack_entry e;
2282
2283 if (find_pack_entry(sha1, &e, NULL))
2284 return 1;
2285 return find_sha1_file(sha1, &st) ? 1 : 0;
2286 }
2287
2288 /*
2289 * reads from fd as long as possible into a supplied buffer of size bytes.
2290 * If necessary the buffer's size is increased using realloc()
2291 *
2292 * returns 0 if anything went fine and -1 otherwise
2293 *
2294 * NOTE: both buf and size may change, but even when -1 is returned
2295 * you still have to free() it yourself.
2296 */
2297 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2298 {
2299 char* buf = *return_buf;
2300 unsigned long size = *return_size;
2301 ssize_t iret;
2302 unsigned long off = 0;
2303
2304 do {
2305 iret = xread(fd, buf + off, size - off);
2306 if (iret > 0) {
2307 off += iret;
2308 if (off == size) {
2309 size *= 2;
2310 buf = xrealloc(buf, size);
2311 }
2312 }
2313 } while (iret > 0);
2314
2315 *return_buf = buf;
2316 *return_size = off;
2317
2318 if (iret < 0)
2319 return -1;
2320 return 0;
2321 }
2322
2323 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2324 {
2325 unsigned long size = 4096;
2326 char *buf = xmalloc(size);
2327 int ret;
2328
2329 if (read_pipe(fd, &buf, &size)) {
2330 free(buf);
2331 return -1;
2332 }
2333
2334 if (!type)
2335 type = blob_type;
2336 if (write_object)
2337 ret = write_sha1_file(buf, size, type, sha1);
2338 else
2339 ret = hash_sha1_file(buf, size, type, sha1);
2340 free(buf);
2341 return ret;
2342 }
2343
2344 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2345 enum object_type type, const char *path)
2346 {
2347 size_t size = xsize_t(st->st_size);
2348 void *buf = NULL;
2349 int ret, re_allocated = 0;
2350
2351 if (size)
2352 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2353 close(fd);
2354
2355 if (!type)
2356 type = OBJ_BLOB;
2357
2358 /*
2359 * Convert blobs to git internal format
2360 */
2361 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2362 unsigned long nsize = size;
2363 char *nbuf = convert_to_git(path, buf, &nsize);
2364 if (nbuf) {
2365 munmap(buf, size);
2366 size = nsize;
2367 buf = nbuf;
2368 re_allocated = 1;
2369 }
2370 }
2371
2372 if (write_object)
2373 ret = write_sha1_file(buf, size, typename(type), sha1);
2374 else
2375 ret = hash_sha1_file(buf, size, typename(type), sha1);
2376 if (re_allocated) {
2377 free(buf);
2378 return ret;
2379 }
2380 if (size)
2381 munmap(buf, size);
2382 return ret;
2383 }
2384
2385 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2386 {
2387 int fd;
2388 char *target;
2389 size_t len;
2390
2391 switch (st->st_mode & S_IFMT) {
2392 case S_IFREG:
2393 fd = open(path, O_RDONLY);
2394 if (fd < 0)
2395 return error("open(\"%s\"): %s", path,
2396 strerror(errno));
2397 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2398 return error("%s: failed to insert into database",
2399 path);
2400 break;
2401 case S_IFLNK:
2402 len = xsize_t(st->st_size);
2403 target = xmalloc(len + 1);
2404 if (readlink(path, target, len + 1) != st->st_size) {
2405 char *errstr = strerror(errno);
2406 free(target);
2407 return error("readlink(\"%s\"): %s", path,
2408 errstr);
2409 }
2410 if (!write_object)
2411 hash_sha1_file(target, len, blob_type, sha1);
2412 else if (write_sha1_file(target, len, blob_type, sha1))
2413 return error("%s: failed to insert into database",
2414 path);
2415 free(target);
2416 break;
2417 case S_IFDIR:
2418 return resolve_gitlink_ref(path, "HEAD", sha1);
2419 default:
2420 return error("%s: unsupported file type", path);
2421 }
2422 return 0;
2423 }
2424
2425 int read_pack_header(int fd, struct pack_header *header)
2426 {
2427 char *c = (char*)header;
2428 ssize_t remaining = sizeof(struct pack_header);
2429 do {
2430 ssize_t r = xread(fd, c, remaining);
2431 if (r <= 0)
2432 /* "eof before pack header was fully read" */
2433 return PH_ERROR_EOF;
2434 remaining -= r;
2435 c += r;
2436 } while (remaining > 0);
2437 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2438 /* "protocol error (pack signature mismatch detected)" */
2439 return PH_ERROR_PACK_SIGNATURE;
2440 if (!pack_version_ok(header->hdr_version))
2441 /* "protocol error (pack version unsupported)" */
2442 return PH_ERROR_PROTOCOL;
2443 return 0;
2444 }