Merge branch 'lt/case-insensitive'
[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 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
19
20 #ifndef O_NOATIME
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
23 #else
24 #define O_NOATIME 0
25 #endif
26 #endif
27
28 #ifdef NO_C99_FORMAT
29 #define SZ_FMT "lu"
30 static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
31 #else
32 #define SZ_FMT "zu"
33 static size_t sz_fmt(size_t s) { return s; }
34 #endif
35
36 const unsigned char null_sha1[20];
37
38 static unsigned int sha1_file_open_flag = O_NOATIME;
39
40 const signed char hexval_table[256] = {
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
47 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
48 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
53 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
71 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
73 };
74
75 int get_sha1_hex(const char *hex, unsigned char *sha1)
76 {
77 int i;
78 for (i = 0; i < 20; i++) {
79 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
80 if (val & ~0xff)
81 return -1;
82 *sha1++ = val;
83 hex += 2;
84 }
85 return 0;
86 }
87
88 int safe_create_leading_directories(char *path)
89 {
90 char *pos = path;
91 struct stat st;
92
93 if (is_absolute_path(path))
94 pos++;
95
96 while (pos) {
97 pos = strchr(pos, '/');
98 if (!pos)
99 break;
100 *pos = 0;
101 if (!stat(path, &st)) {
102 /* path exists */
103 if (!S_ISDIR(st.st_mode)) {
104 *pos = '/';
105 return -3;
106 }
107 }
108 else if (mkdir(path, 0777)) {
109 *pos = '/';
110 return -1;
111 }
112 else if (adjust_shared_perm(path)) {
113 *pos = '/';
114 return -2;
115 }
116 *pos++ = '/';
117 }
118 return 0;
119 }
120
121 char * sha1_to_hex(const unsigned char *sha1)
122 {
123 static int bufno;
124 static char hexbuffer[4][50];
125 static const char hex[] = "0123456789abcdef";
126 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
127 int i;
128
129 for (i = 0; i < 20; i++) {
130 unsigned int val = *sha1++;
131 *buf++ = hex[val >> 4];
132 *buf++ = hex[val & 0xf];
133 }
134 *buf = '\0';
135
136 return buffer;
137 }
138
139 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
140 {
141 int i;
142 for (i = 0; i < 20; i++) {
143 static char hex[] = "0123456789abcdef";
144 unsigned int val = sha1[i];
145 char *pos = pathbuf + i*2 + (i > 0);
146 *pos++ = hex[val >> 4];
147 *pos = hex[val & 0xf];
148 }
149 }
150
151 /*
152 * NOTE! This returns a statically allocated buffer, so you have to be
153 * careful about using it. Do an "xstrdup()" if you need to save the
154 * filename.
155 *
156 * Also note that this returns the location for creating. Reading
157 * SHA1 file can happen from any alternate directory listed in the
158 * DB_ENVIRONMENT environment variable if it is not found in
159 * the primary object database.
160 */
161 char *sha1_file_name(const unsigned char *sha1)
162 {
163 static char *name, *base;
164
165 if (!base) {
166 const char *sha1_file_directory = get_object_directory();
167 int len = strlen(sha1_file_directory);
168 base = xmalloc(len + 60);
169 memcpy(base, sha1_file_directory, len);
170 memset(base+len, 0, 60);
171 base[len] = '/';
172 base[len+3] = '/';
173 name = base + len + 1;
174 }
175 fill_sha1_path(name, sha1);
176 return base;
177 }
178
179 char *sha1_pack_name(const unsigned char *sha1)
180 {
181 static const char hex[] = "0123456789abcdef";
182 static char *name, *base, *buf;
183 int i;
184
185 if (!base) {
186 const char *sha1_file_directory = get_object_directory();
187 int len = strlen(sha1_file_directory);
188 base = xmalloc(len + 60);
189 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
190 name = base + len + 11;
191 }
192
193 buf = name;
194
195 for (i = 0; i < 20; i++) {
196 unsigned int val = *sha1++;
197 *buf++ = hex[val >> 4];
198 *buf++ = hex[val & 0xf];
199 }
200
201 return base;
202 }
203
204 char *sha1_pack_index_name(const unsigned char *sha1)
205 {
206 static const char hex[] = "0123456789abcdef";
207 static char *name, *base, *buf;
208 int i;
209
210 if (!base) {
211 const char *sha1_file_directory = get_object_directory();
212 int len = strlen(sha1_file_directory);
213 base = xmalloc(len + 60);
214 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
215 name = base + len + 11;
216 }
217
218 buf = name;
219
220 for (i = 0; i < 20; i++) {
221 unsigned int val = *sha1++;
222 *buf++ = hex[val >> 4];
223 *buf++ = hex[val & 0xf];
224 }
225
226 return base;
227 }
228
229 struct alternate_object_database *alt_odb_list;
230 static struct alternate_object_database **alt_odb_tail;
231
232 static void read_info_alternates(const char * alternates, int depth);
233
234 /*
235 * Prepare alternate object database registry.
236 *
237 * The variable alt_odb_list points at the list of struct
238 * alternate_object_database. The elements on this list come from
239 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
240 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
241 * whose contents is similar to that environment variable but can be
242 * LF separated. Its base points at a statically allocated buffer that
243 * contains "/the/directory/corresponding/to/.git/objects/...", while
244 * its name points just after the slash at the end of ".git/objects/"
245 * in the example above, and has enough space to hold 40-byte hex
246 * SHA1, an extra slash for the first level indirection, and the
247 * terminating NUL.
248 */
249 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
250 {
251 struct stat st;
252 const char *objdir = get_object_directory();
253 struct alternate_object_database *ent;
254 struct alternate_object_database *alt;
255 /* 43 = 40-byte + 2 '/' + terminating NUL */
256 int pfxlen = len;
257 int entlen = pfxlen + 43;
258 int base_len = -1;
259
260 if (!is_absolute_path(entry) && relative_base) {
261 /* Relative alt-odb */
262 if (base_len < 0)
263 base_len = strlen(relative_base) + 1;
264 entlen += base_len;
265 pfxlen += base_len;
266 }
267 ent = xmalloc(sizeof(*ent) + entlen);
268
269 if (!is_absolute_path(entry) && relative_base) {
270 memcpy(ent->base, relative_base, base_len - 1);
271 ent->base[base_len - 1] = '/';
272 memcpy(ent->base + base_len, entry, len);
273 }
274 else
275 memcpy(ent->base, entry, pfxlen);
276
277 ent->name = ent->base + pfxlen + 1;
278 ent->base[pfxlen + 3] = '/';
279 ent->base[pfxlen] = ent->base[entlen-1] = 0;
280
281 /* Detect cases where alternate disappeared */
282 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
283 error("object directory %s does not exist; "
284 "check .git/objects/info/alternates.",
285 ent->base);
286 free(ent);
287 return -1;
288 }
289
290 /* Prevent the common mistake of listing the same
291 * thing twice, or object directory itself.
292 */
293 for (alt = alt_odb_list; alt; alt = alt->next) {
294 if (!memcmp(ent->base, alt->base, pfxlen)) {
295 free(ent);
296 return -1;
297 }
298 }
299 if (!memcmp(ent->base, objdir, pfxlen)) {
300 free(ent);
301 return -1;
302 }
303
304 /* add the alternate entry */
305 *alt_odb_tail = ent;
306 alt_odb_tail = &(ent->next);
307 ent->next = NULL;
308
309 /* recursively add alternates */
310 read_info_alternates(ent->base, depth + 1);
311
312 ent->base[pfxlen] = '/';
313
314 return 0;
315 }
316
317 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
318 const char *relative_base, int depth)
319 {
320 const char *cp, *last;
321
322 if (depth > 5) {
323 error("%s: ignoring alternate object stores, nesting too deep.",
324 relative_base);
325 return;
326 }
327
328 last = alt;
329 while (last < ep) {
330 cp = last;
331 if (cp < ep && *cp == '#') {
332 while (cp < ep && *cp != sep)
333 cp++;
334 last = cp + 1;
335 continue;
336 }
337 while (cp < ep && *cp != sep)
338 cp++;
339 if (last != cp) {
340 if (!is_absolute_path(last) && depth) {
341 error("%s: ignoring relative alternate object store %s",
342 relative_base, last);
343 } else {
344 link_alt_odb_entry(last, cp - last,
345 relative_base, depth);
346 }
347 }
348 while (cp < ep && *cp == sep)
349 cp++;
350 last = cp;
351 }
352 }
353
354 static void read_info_alternates(const char * relative_base, int depth)
355 {
356 char *map;
357 size_t mapsz;
358 struct stat st;
359 const char alt_file_name[] = "info/alternates";
360 /* Given that relative_base is no longer than PATH_MAX,
361 ensure that "path" has enough space to append "/", the
362 file name, "info/alternates", and a trailing NUL. */
363 char path[PATH_MAX + 1 + sizeof alt_file_name];
364 int fd;
365
366 sprintf(path, "%s/%s", relative_base, alt_file_name);
367 fd = open(path, O_RDONLY);
368 if (fd < 0)
369 return;
370 if (fstat(fd, &st) || (st.st_size == 0)) {
371 close(fd);
372 return;
373 }
374 mapsz = xsize_t(st.st_size);
375 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
376 close(fd);
377
378 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
379
380 munmap(map, mapsz);
381 }
382
383 void prepare_alt_odb(void)
384 {
385 const char *alt;
386
387 if (alt_odb_tail)
388 return;
389
390 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
391 if (!alt) alt = "";
392
393 alt_odb_tail = &alt_odb_list;
394 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
395
396 read_info_alternates(get_object_directory(), 0);
397 }
398
399 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
400 {
401 char *name = sha1_file_name(sha1);
402 struct alternate_object_database *alt;
403
404 if (!stat(name, st))
405 return name;
406 prepare_alt_odb();
407 for (alt = alt_odb_list; alt; alt = alt->next) {
408 name = alt->name;
409 fill_sha1_path(name, sha1);
410 if (!stat(alt->base, st))
411 return alt->base;
412 }
413 return NULL;
414 }
415
416 static unsigned int pack_used_ctr;
417 static unsigned int pack_mmap_calls;
418 static unsigned int peak_pack_open_windows;
419 static unsigned int pack_open_windows;
420 static size_t peak_pack_mapped;
421 static size_t pack_mapped;
422 struct packed_git *packed_git;
423
424 void pack_report(void)
425 {
426 fprintf(stderr,
427 "pack_report: getpagesize() = %10" SZ_FMT "\n"
428 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
429 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
430 sz_fmt(getpagesize()),
431 sz_fmt(packed_git_window_size),
432 sz_fmt(packed_git_limit));
433 fprintf(stderr,
434 "pack_report: pack_used_ctr = %10u\n"
435 "pack_report: pack_mmap_calls = %10u\n"
436 "pack_report: pack_open_windows = %10u / %10u\n"
437 "pack_report: pack_mapped = "
438 "%10" SZ_FMT " / %10" SZ_FMT "\n",
439 pack_used_ctr,
440 pack_mmap_calls,
441 pack_open_windows, peak_pack_open_windows,
442 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
443 }
444
445 static int check_packed_git_idx(const char *path, struct packed_git *p)
446 {
447 void *idx_map;
448 struct pack_idx_header *hdr;
449 size_t idx_size;
450 uint32_t version, nr, i, *index;
451 int fd = open(path, O_RDONLY);
452 struct stat st;
453
454 if (fd < 0)
455 return -1;
456 if (fstat(fd, &st)) {
457 close(fd);
458 return -1;
459 }
460 idx_size = xsize_t(st.st_size);
461 if (idx_size < 4 * 256 + 20 + 20) {
462 close(fd);
463 return error("index file %s is too small", path);
464 }
465 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
466 close(fd);
467
468 hdr = idx_map;
469 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
470 version = ntohl(hdr->idx_version);
471 if (version < 2 || version > 2) {
472 munmap(idx_map, idx_size);
473 return error("index file %s is version %d"
474 " and is not supported by this binary"
475 " (try upgrading GIT to a newer version)",
476 path, version);
477 }
478 } else
479 version = 1;
480
481 nr = 0;
482 index = idx_map;
483 if (version > 1)
484 index += 2; /* skip index header */
485 for (i = 0; i < 256; i++) {
486 uint32_t n = ntohl(index[i]);
487 if (n < nr) {
488 munmap(idx_map, idx_size);
489 return error("non-monotonic index %s", path);
490 }
491 nr = n;
492 }
493
494 if (version == 1) {
495 /*
496 * Total size:
497 * - 256 index entries 4 bytes each
498 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
499 * - 20-byte SHA1 of the packfile
500 * - 20-byte SHA1 file checksum
501 */
502 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
503 munmap(idx_map, idx_size);
504 return error("wrong index v1 file size in %s", path);
505 }
506 } else if (version == 2) {
507 /*
508 * Minimum size:
509 * - 8 bytes of header
510 * - 256 index entries 4 bytes each
511 * - 20-byte sha1 entry * nr
512 * - 4-byte crc entry * nr
513 * - 4-byte offset entry * nr
514 * - 20-byte SHA1 of the packfile
515 * - 20-byte SHA1 file checksum
516 * And after the 4-byte offset table might be a
517 * variable sized table containing 8-byte entries
518 * for offsets larger than 2^31.
519 */
520 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
521 unsigned long max_size = min_size;
522 if (nr)
523 max_size += (nr - 1)*8;
524 if (idx_size < min_size || idx_size > max_size) {
525 munmap(idx_map, idx_size);
526 return error("wrong index v2 file size in %s", path);
527 }
528 if (idx_size != min_size &&
529 /*
530 * make sure we can deal with large pack offsets.
531 * 31-bit signed offset won't be enough, neither
532 * 32-bit unsigned one will be.
533 */
534 (sizeof(off_t) <= 4)) {
535 munmap(idx_map, idx_size);
536 return error("pack too large for current definition of off_t in %s", path);
537 }
538 }
539
540 p->index_version = version;
541 p->index_data = idx_map;
542 p->index_size = idx_size;
543 p->num_objects = nr;
544 return 0;
545 }
546
547 int open_pack_index(struct packed_git *p)
548 {
549 char *idx_name;
550 int ret;
551
552 if (p->index_data)
553 return 0;
554
555 idx_name = xstrdup(p->pack_name);
556 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
557 ret = check_packed_git_idx(idx_name, p);
558 free(idx_name);
559 return ret;
560 }
561
562 static void scan_windows(struct packed_git *p,
563 struct packed_git **lru_p,
564 struct pack_window **lru_w,
565 struct pack_window **lru_l)
566 {
567 struct pack_window *w, *w_l;
568
569 for (w_l = NULL, w = p->windows; w; w = w->next) {
570 if (!w->inuse_cnt) {
571 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
572 *lru_p = p;
573 *lru_w = w;
574 *lru_l = w_l;
575 }
576 }
577 w_l = w;
578 }
579 }
580
581 static int unuse_one_window(struct packed_git *current, int keep_fd)
582 {
583 struct packed_git *p, *lru_p = NULL;
584 struct pack_window *lru_w = NULL, *lru_l = NULL;
585
586 if (current)
587 scan_windows(current, &lru_p, &lru_w, &lru_l);
588 for (p = packed_git; p; p = p->next)
589 scan_windows(p, &lru_p, &lru_w, &lru_l);
590 if (lru_p) {
591 munmap(lru_w->base, lru_w->len);
592 pack_mapped -= lru_w->len;
593 if (lru_l)
594 lru_l->next = lru_w->next;
595 else {
596 lru_p->windows = lru_w->next;
597 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
598 close(lru_p->pack_fd);
599 lru_p->pack_fd = -1;
600 }
601 }
602 free(lru_w);
603 pack_open_windows--;
604 return 1;
605 }
606 return 0;
607 }
608
609 void release_pack_memory(size_t need, int fd)
610 {
611 size_t cur = pack_mapped;
612 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
613 ; /* nothing */
614 }
615
616 void close_pack_windows(struct packed_git *p)
617 {
618 while (p->windows) {
619 struct pack_window *w = p->windows;
620
621 if (w->inuse_cnt)
622 die("pack '%s' still has open windows to it",
623 p->pack_name);
624 munmap(w->base, w->len);
625 pack_mapped -= w->len;
626 pack_open_windows--;
627 p->windows = w->next;
628 free(w);
629 }
630 }
631
632 void unuse_pack(struct pack_window **w_cursor)
633 {
634 struct pack_window *w = *w_cursor;
635 if (w) {
636 w->inuse_cnt--;
637 *w_cursor = NULL;
638 }
639 }
640
641 /*
642 * Do not call this directly as this leaks p->pack_fd on error return;
643 * call open_packed_git() instead.
644 */
645 static int open_packed_git_1(struct packed_git *p)
646 {
647 struct stat st;
648 struct pack_header hdr;
649 unsigned char sha1[20];
650 unsigned char *idx_sha1;
651 long fd_flag;
652
653 if (!p->index_data && open_pack_index(p))
654 return error("packfile %s index unavailable", p->pack_name);
655
656 p->pack_fd = open(p->pack_name, O_RDONLY);
657 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
658 return -1;
659
660 /* If we created the struct before we had the pack we lack size. */
661 if (!p->pack_size) {
662 if (!S_ISREG(st.st_mode))
663 return error("packfile %s not a regular file", p->pack_name);
664 p->pack_size = st.st_size;
665 } else if (p->pack_size != st.st_size)
666 return error("packfile %s size changed", p->pack_name);
667
668 /* We leave these file descriptors open with sliding mmap;
669 * there is no point keeping them open across exec(), though.
670 */
671 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
672 if (fd_flag < 0)
673 return error("cannot determine file descriptor flags");
674 fd_flag |= FD_CLOEXEC;
675 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
676 return error("cannot set FD_CLOEXEC");
677
678 /* Verify we recognize this pack file format. */
679 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
680 return error("file %s is far too short to be a packfile", p->pack_name);
681 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
682 return error("file %s is not a GIT packfile", p->pack_name);
683 if (!pack_version_ok(hdr.hdr_version))
684 return error("packfile %s is version %u and not supported"
685 " (try upgrading GIT to a newer version)",
686 p->pack_name, ntohl(hdr.hdr_version));
687
688 /* Verify the pack matches its index. */
689 if (p->num_objects != ntohl(hdr.hdr_entries))
690 return error("packfile %s claims to have %u objects"
691 " while index indicates %u objects",
692 p->pack_name, ntohl(hdr.hdr_entries),
693 p->num_objects);
694 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
695 return error("end of packfile %s is unavailable", p->pack_name);
696 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
697 return error("packfile %s signature is unavailable", p->pack_name);
698 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
699 if (hashcmp(sha1, idx_sha1))
700 return error("packfile %s does not match index", p->pack_name);
701 return 0;
702 }
703
704 static int open_packed_git(struct packed_git *p)
705 {
706 if (!open_packed_git_1(p))
707 return 0;
708 if (p->pack_fd != -1) {
709 close(p->pack_fd);
710 p->pack_fd = -1;
711 }
712 return -1;
713 }
714
715 static int in_window(struct pack_window *win, off_t offset)
716 {
717 /* We must promise at least 20 bytes (one hash) after the
718 * offset is available from this window, otherwise the offset
719 * is not actually in this window and a different window (which
720 * has that one hash excess) must be used. This is to support
721 * the object header and delta base parsing routines below.
722 */
723 off_t win_off = win->offset;
724 return win_off <= offset
725 && (offset + 20) <= (win_off + win->len);
726 }
727
728 unsigned char* use_pack(struct packed_git *p,
729 struct pack_window **w_cursor,
730 off_t offset,
731 unsigned int *left)
732 {
733 struct pack_window *win = *w_cursor;
734
735 if (p->pack_fd == -1 && open_packed_git(p))
736 die("packfile %s cannot be accessed", p->pack_name);
737
738 /* Since packfiles end in a hash of their content and its
739 * pointless to ask for an offset into the middle of that
740 * hash, and the in_window function above wouldn't match
741 * don't allow an offset too close to the end of the file.
742 */
743 if (offset > (p->pack_size - 20))
744 die("offset beyond end of packfile (truncated pack?)");
745
746 if (!win || !in_window(win, offset)) {
747 if (win)
748 win->inuse_cnt--;
749 for (win = p->windows; win; win = win->next) {
750 if (in_window(win, offset))
751 break;
752 }
753 if (!win) {
754 size_t window_align = packed_git_window_size / 2;
755 off_t len;
756 win = xcalloc(1, sizeof(*win));
757 win->offset = (offset / window_align) * window_align;
758 len = p->pack_size - win->offset;
759 if (len > packed_git_window_size)
760 len = packed_git_window_size;
761 win->len = (size_t)len;
762 pack_mapped += win->len;
763 while (packed_git_limit < pack_mapped
764 && unuse_one_window(p, p->pack_fd))
765 ; /* nothing */
766 win->base = xmmap(NULL, win->len,
767 PROT_READ, MAP_PRIVATE,
768 p->pack_fd, win->offset);
769 if (win->base == MAP_FAILED)
770 die("packfile %s cannot be mapped: %s",
771 p->pack_name,
772 strerror(errno));
773 pack_mmap_calls++;
774 pack_open_windows++;
775 if (pack_mapped > peak_pack_mapped)
776 peak_pack_mapped = pack_mapped;
777 if (pack_open_windows > peak_pack_open_windows)
778 peak_pack_open_windows = pack_open_windows;
779 win->next = p->windows;
780 p->windows = win;
781 }
782 }
783 if (win != *w_cursor) {
784 win->last_used = pack_used_ctr++;
785 win->inuse_cnt++;
786 *w_cursor = win;
787 }
788 offset -= win->offset;
789 if (left)
790 *left = win->len - xsize_t(offset);
791 return win->base + offset;
792 }
793
794 struct packed_git *add_packed_git(const char *path, int path_len, int local)
795 {
796 struct stat st;
797 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
798
799 /*
800 * Make sure a corresponding .pack file exists and that
801 * the index looks sane.
802 */
803 path_len -= strlen(".idx");
804 if (path_len < 1)
805 return NULL;
806 memcpy(p->pack_name, path, path_len);
807 strcpy(p->pack_name + path_len, ".pack");
808 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
809 free(p);
810 return NULL;
811 }
812
813 /* ok, it looks sane as far as we can check without
814 * actually mapping the pack file.
815 */
816 p->index_version = 0;
817 p->index_data = NULL;
818 p->index_size = 0;
819 p->num_objects = 0;
820 p->pack_size = st.st_size;
821 p->next = NULL;
822 p->windows = NULL;
823 p->pack_fd = -1;
824 p->pack_local = local;
825 p->mtime = st.st_mtime;
826 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
827 hashclr(p->sha1);
828 return p;
829 }
830
831 struct packed_git *parse_pack_index(unsigned char *sha1)
832 {
833 char *path = sha1_pack_index_name(sha1);
834 return parse_pack_index_file(sha1, path);
835 }
836
837 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
838 const char *idx_path)
839 {
840 const char *path = sha1_pack_name(sha1);
841 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
842
843 if (check_packed_git_idx(idx_path, p)) {
844 free(p);
845 return NULL;
846 }
847
848 strcpy(p->pack_name, path);
849 p->pack_size = 0;
850 p->next = NULL;
851 p->windows = NULL;
852 p->pack_fd = -1;
853 hashcpy(p->sha1, sha1);
854 return p;
855 }
856
857 void install_packed_git(struct packed_git *pack)
858 {
859 pack->next = packed_git;
860 packed_git = pack;
861 }
862
863 static void prepare_packed_git_one(char *objdir, int local)
864 {
865 /* Ensure that this buffer is large enough so that we can
866 append "/pack/" without clobbering the stack even if
867 strlen(objdir) were PATH_MAX. */
868 char path[PATH_MAX + 1 + 4 + 1 + 1];
869 int len;
870 DIR *dir;
871 struct dirent *de;
872
873 sprintf(path, "%s/pack", objdir);
874 len = strlen(path);
875 dir = opendir(path);
876 if (!dir) {
877 if (errno != ENOENT)
878 error("unable to open object pack directory: %s: %s",
879 path, strerror(errno));
880 return;
881 }
882 path[len++] = '/';
883 while ((de = readdir(dir)) != NULL) {
884 int namelen = strlen(de->d_name);
885 struct packed_git *p;
886
887 if (!has_extension(de->d_name, ".idx"))
888 continue;
889
890 if (len + namelen + 1 > sizeof(path))
891 continue;
892
893 /* Don't reopen a pack we already have. */
894 strcpy(path + len, de->d_name);
895 for (p = packed_git; p; p = p->next) {
896 if (!memcmp(path, p->pack_name, len + namelen - 4))
897 break;
898 }
899 if (p)
900 continue;
901 /* See if it really is a valid .idx file with corresponding
902 * .pack file that we can map.
903 */
904 p = add_packed_git(path, len + namelen, local);
905 if (!p)
906 continue;
907 install_packed_git(p);
908 }
909 closedir(dir);
910 }
911
912 static int sort_pack(const void *a_, const void *b_)
913 {
914 struct packed_git *a = *((struct packed_git **)a_);
915 struct packed_git *b = *((struct packed_git **)b_);
916 int st;
917
918 /*
919 * Local packs tend to contain objects specific to our
920 * variant of the project than remote ones. In addition,
921 * remote ones could be on a network mounted filesystem.
922 * Favor local ones for these reasons.
923 */
924 st = a->pack_local - b->pack_local;
925 if (st)
926 return -st;
927
928 /*
929 * Younger packs tend to contain more recent objects,
930 * and more recent objects tend to get accessed more
931 * often.
932 */
933 if (a->mtime < b->mtime)
934 return 1;
935 else if (a->mtime == b->mtime)
936 return 0;
937 return -1;
938 }
939
940 static void rearrange_packed_git(void)
941 {
942 struct packed_git **ary, *p;
943 int i, n;
944
945 for (n = 0, p = packed_git; p; p = p->next)
946 n++;
947 if (n < 2)
948 return;
949
950 /* prepare an array of packed_git for easier sorting */
951 ary = xcalloc(n, sizeof(struct packed_git *));
952 for (n = 0, p = packed_git; p; p = p->next)
953 ary[n++] = p;
954
955 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
956
957 /* link them back again */
958 for (i = 0; i < n - 1; i++)
959 ary[i]->next = ary[i + 1];
960 ary[n - 1]->next = NULL;
961 packed_git = ary[0];
962
963 free(ary);
964 }
965
966 static int prepare_packed_git_run_once = 0;
967 void prepare_packed_git(void)
968 {
969 struct alternate_object_database *alt;
970
971 if (prepare_packed_git_run_once)
972 return;
973 prepare_packed_git_one(get_object_directory(), 1);
974 prepare_alt_odb();
975 for (alt = alt_odb_list; alt; alt = alt->next) {
976 alt->name[-1] = 0;
977 prepare_packed_git_one(alt->base, 0);
978 alt->name[-1] = '/';
979 }
980 rearrange_packed_git();
981 prepare_packed_git_run_once = 1;
982 }
983
984 void reprepare_packed_git(void)
985 {
986 prepare_packed_git_run_once = 0;
987 prepare_packed_git();
988 }
989
990 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
991 {
992 unsigned char real_sha1[20];
993 hash_sha1_file(map, size, type, real_sha1);
994 return hashcmp(sha1, real_sha1) ? -1 : 0;
995 }
996
997 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
998 {
999 struct stat st;
1000 void *map;
1001 int fd;
1002 char *filename = find_sha1_file(sha1, &st);
1003
1004 if (!filename) {
1005 return NULL;
1006 }
1007
1008 fd = open(filename, O_RDONLY | sha1_file_open_flag);
1009 if (fd < 0) {
1010 /* See if it works without O_NOATIME */
1011 switch (sha1_file_open_flag) {
1012 default:
1013 fd = open(filename, O_RDONLY);
1014 if (fd >= 0)
1015 break;
1016 /* Fallthrough */
1017 case 0:
1018 return NULL;
1019 }
1020
1021 /* If it failed once, it will probably fail again.
1022 * Stop using O_NOATIME
1023 */
1024 sha1_file_open_flag = 0;
1025 }
1026 *size = xsize_t(st.st_size);
1027 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1028 close(fd);
1029 return map;
1030 }
1031
1032 static int legacy_loose_object(unsigned char *map)
1033 {
1034 unsigned int word;
1035
1036 /*
1037 * Is it a zlib-compressed buffer? If so, the first byte
1038 * must be 0x78 (15-bit window size, deflated), and the
1039 * first 16-bit word is evenly divisible by 31
1040 */
1041 word = (map[0] << 8) + map[1];
1042 if (map[0] == 0x78 && !(word % 31))
1043 return 1;
1044 else
1045 return 0;
1046 }
1047
1048 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1049 {
1050 unsigned shift;
1051 unsigned char c;
1052 unsigned long size;
1053 unsigned long used = 0;
1054
1055 c = buf[used++];
1056 *type = (c >> 4) & 7;
1057 size = c & 15;
1058 shift = 4;
1059 while (c & 0x80) {
1060 if (len <= used)
1061 return 0;
1062 if (sizeof(long) * 8 <= shift)
1063 return 0;
1064 c = buf[used++];
1065 size += (c & 0x7f) << shift;
1066 shift += 7;
1067 }
1068 *sizep = size;
1069 return used;
1070 }
1071
1072 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1073 {
1074 unsigned long size, used;
1075 static const char valid_loose_object_type[8] = {
1076 0, /* OBJ_EXT */
1077 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1078 0, /* "delta" and others are invalid in a loose object */
1079 };
1080 enum object_type type;
1081
1082 /* Get the data stream */
1083 memset(stream, 0, sizeof(*stream));
1084 stream->next_in = map;
1085 stream->avail_in = mapsize;
1086 stream->next_out = buffer;
1087 stream->avail_out = bufsiz;
1088
1089 if (legacy_loose_object(map)) {
1090 inflateInit(stream);
1091 return inflate(stream, 0);
1092 }
1093
1094
1095 /*
1096 * There used to be a second loose object header format which
1097 * was meant to mimic the in-pack format, allowing for direct
1098 * copy of the object data. This format turned up not to be
1099 * really worth it and we don't write it any longer. But we
1100 * can still read it.
1101 */
1102 used = unpack_object_header_gently(map, mapsize, &type, &size);
1103 if (!used || !valid_loose_object_type[type])
1104 return -1;
1105 map += used;
1106 mapsize -= used;
1107
1108 /* Set up the stream for the rest.. */
1109 stream->next_in = map;
1110 stream->avail_in = mapsize;
1111 inflateInit(stream);
1112
1113 /* And generate the fake traditional header */
1114 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1115 typename(type), size);
1116 return 0;
1117 }
1118
1119 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1120 {
1121 int bytes = strlen(buffer) + 1;
1122 unsigned char *buf = xmalloc(1+size);
1123 unsigned long n;
1124 int status = Z_OK;
1125
1126 n = stream->total_out - bytes;
1127 if (n > size)
1128 n = size;
1129 memcpy(buf, (char *) buffer + bytes, n);
1130 bytes = n;
1131 if (bytes <= size) {
1132 /*
1133 * The above condition must be (bytes <= size), not
1134 * (bytes < size). In other words, even though we
1135 * expect no more output and set avail_out to zer0,
1136 * the input zlib stream may have bytes that express
1137 * "this concludes the stream", and we *do* want to
1138 * eat that input.
1139 *
1140 * Otherwise we would not be able to test that we
1141 * consumed all the input to reach the expected size;
1142 * we also want to check that zlib tells us that all
1143 * went well with status == Z_STREAM_END at the end.
1144 */
1145 stream->next_out = buf + bytes;
1146 stream->avail_out = size - bytes;
1147 while (status == Z_OK)
1148 status = inflate(stream, Z_FINISH);
1149 }
1150 buf[size] = 0;
1151 if (status == Z_STREAM_END && !stream->avail_in) {
1152 inflateEnd(stream);
1153 return buf;
1154 }
1155
1156 if (status < 0)
1157 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1158 else if (stream->avail_in)
1159 error("garbage at end of loose object '%s'",
1160 sha1_to_hex(sha1));
1161 free(buf);
1162 return NULL;
1163 }
1164
1165 /*
1166 * We used to just use "sscanf()", but that's actually way
1167 * too permissive for what we want to check. So do an anal
1168 * object header parse by hand.
1169 */
1170 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1171 {
1172 char type[10];
1173 int i;
1174 unsigned long size;
1175
1176 /*
1177 * The type can be at most ten bytes (including the
1178 * terminating '\0' that we add), and is followed by
1179 * a space.
1180 */
1181 i = 0;
1182 for (;;) {
1183 char c = *hdr++;
1184 if (c == ' ')
1185 break;
1186 type[i++] = c;
1187 if (i >= sizeof(type))
1188 return -1;
1189 }
1190 type[i] = 0;
1191
1192 /*
1193 * The length must follow immediately, and be in canonical
1194 * decimal format (ie "010" is not valid).
1195 */
1196 size = *hdr++ - '0';
1197 if (size > 9)
1198 return -1;
1199 if (size) {
1200 for (;;) {
1201 unsigned long c = *hdr - '0';
1202 if (c > 9)
1203 break;
1204 hdr++;
1205 size = size * 10 + c;
1206 }
1207 }
1208 *sizep = size;
1209
1210 /*
1211 * The length must be followed by a zero byte
1212 */
1213 return *hdr ? -1 : type_from_string(type);
1214 }
1215
1216 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1217 {
1218 int ret;
1219 z_stream stream;
1220 char hdr[8192];
1221
1222 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1223 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1224 return NULL;
1225
1226 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1227 }
1228
1229 unsigned long get_size_from_delta(struct packed_git *p,
1230 struct pack_window **w_curs,
1231 off_t curpos)
1232 {
1233 const unsigned char *data;
1234 unsigned char delta_head[20], *in;
1235 z_stream stream;
1236 int st;
1237
1238 memset(&stream, 0, sizeof(stream));
1239 stream.next_out = delta_head;
1240 stream.avail_out = sizeof(delta_head);
1241
1242 inflateInit(&stream);
1243 do {
1244 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1245 stream.next_in = in;
1246 st = inflate(&stream, Z_FINISH);
1247 curpos += stream.next_in - in;
1248 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1249 stream.total_out < sizeof(delta_head));
1250 inflateEnd(&stream);
1251 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1252 die("delta data unpack-initial failed");
1253
1254 /* Examine the initial part of the delta to figure out
1255 * the result size.
1256 */
1257 data = delta_head;
1258
1259 /* ignore base size */
1260 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1261
1262 /* Read the result size */
1263 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1264 }
1265
1266 static off_t get_delta_base(struct packed_git *p,
1267 struct pack_window **w_curs,
1268 off_t *curpos,
1269 enum object_type type,
1270 off_t delta_obj_offset)
1271 {
1272 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1273 off_t base_offset;
1274
1275 /* use_pack() assured us we have [base_info, base_info + 20)
1276 * as a range that we can look at without walking off the
1277 * end of the mapped window. Its actually the hash size
1278 * that is assured. An OFS_DELTA longer than the hash size
1279 * is stupid, as then a REF_DELTA would be smaller to store.
1280 */
1281 if (type == OBJ_OFS_DELTA) {
1282 unsigned used = 0;
1283 unsigned char c = base_info[used++];
1284 base_offset = c & 127;
1285 while (c & 128) {
1286 base_offset += 1;
1287 if (!base_offset || MSB(base_offset, 7))
1288 die("offset value overflow for delta base object");
1289 c = base_info[used++];
1290 base_offset = (base_offset << 7) + (c & 127);
1291 }
1292 base_offset = delta_obj_offset - base_offset;
1293 if (base_offset >= delta_obj_offset)
1294 die("delta base offset out of bound");
1295 *curpos += used;
1296 } else if (type == OBJ_REF_DELTA) {
1297 /* The base entry _must_ be in the same pack */
1298 base_offset = find_pack_entry_one(base_info, p);
1299 if (!base_offset)
1300 die("failed to find delta-pack base object %s",
1301 sha1_to_hex(base_info));
1302 *curpos += 20;
1303 } else
1304 die("I am totally screwed");
1305 return base_offset;
1306 }
1307
1308 /* forward declaration for a mutually recursive function */
1309 static int packed_object_info(struct packed_git *p, off_t offset,
1310 unsigned long *sizep);
1311
1312 static int packed_delta_info(struct packed_git *p,
1313 struct pack_window **w_curs,
1314 off_t curpos,
1315 enum object_type type,
1316 off_t obj_offset,
1317 unsigned long *sizep)
1318 {
1319 off_t base_offset;
1320
1321 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1322 type = packed_object_info(p, base_offset, NULL);
1323
1324 /* We choose to only get the type of the base object and
1325 * ignore potentially corrupt pack file that expects the delta
1326 * based on a base with a wrong size. This saves tons of
1327 * inflate() calls.
1328 */
1329 if (sizep)
1330 *sizep = get_size_from_delta(p, w_curs, curpos);
1331
1332 return type;
1333 }
1334
1335 static int unpack_object_header(struct packed_git *p,
1336 struct pack_window **w_curs,
1337 off_t *curpos,
1338 unsigned long *sizep)
1339 {
1340 unsigned char *base;
1341 unsigned int left;
1342 unsigned long used;
1343 enum object_type type;
1344
1345 /* use_pack() assures us we have [base, base + 20) available
1346 * as a range that we can look at at. (Its actually the hash
1347 * size that is assured.) With our object header encoding
1348 * the maximum deflated object size is 2^137, which is just
1349 * insane, so we know won't exceed what we have been given.
1350 */
1351 base = use_pack(p, w_curs, *curpos, &left);
1352 used = unpack_object_header_gently(base, left, &type, sizep);
1353 if (!used)
1354 die("object offset outside of pack file");
1355 *curpos += used;
1356
1357 return type;
1358 }
1359
1360 const char *packed_object_info_detail(struct packed_git *p,
1361 off_t obj_offset,
1362 unsigned long *size,
1363 unsigned long *store_size,
1364 unsigned int *delta_chain_length,
1365 unsigned char *base_sha1)
1366 {
1367 struct pack_window *w_curs = NULL;
1368 off_t curpos;
1369 unsigned long dummy;
1370 unsigned char *next_sha1;
1371 enum object_type type;
1372 struct revindex_entry *revidx;
1373
1374 *delta_chain_length = 0;
1375 curpos = obj_offset;
1376 type = unpack_object_header(p, &w_curs, &curpos, size);
1377
1378 revidx = find_pack_revindex(p, obj_offset);
1379 *store_size = revidx[1].offset - obj_offset;
1380
1381 for (;;) {
1382 switch (type) {
1383 default:
1384 die("pack %s contains unknown object type %d",
1385 p->pack_name, type);
1386 case OBJ_COMMIT:
1387 case OBJ_TREE:
1388 case OBJ_BLOB:
1389 case OBJ_TAG:
1390 unuse_pack(&w_curs);
1391 return typename(type);
1392 case OBJ_OFS_DELTA:
1393 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1394 if (*delta_chain_length == 0) {
1395 revidx = find_pack_revindex(p, obj_offset);
1396 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1397 }
1398 break;
1399 case OBJ_REF_DELTA:
1400 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1401 if (*delta_chain_length == 0)
1402 hashcpy(base_sha1, next_sha1);
1403 obj_offset = find_pack_entry_one(next_sha1, p);
1404 break;
1405 }
1406 (*delta_chain_length)++;
1407 curpos = obj_offset;
1408 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1409 }
1410 }
1411
1412 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1413 unsigned long *sizep)
1414 {
1415 struct pack_window *w_curs = NULL;
1416 unsigned long size;
1417 off_t curpos = obj_offset;
1418 enum object_type type;
1419
1420 type = unpack_object_header(p, &w_curs, &curpos, &size);
1421
1422 switch (type) {
1423 case OBJ_OFS_DELTA:
1424 case OBJ_REF_DELTA:
1425 type = packed_delta_info(p, &w_curs, curpos,
1426 type, obj_offset, sizep);
1427 break;
1428 case OBJ_COMMIT:
1429 case OBJ_TREE:
1430 case OBJ_BLOB:
1431 case OBJ_TAG:
1432 if (sizep)
1433 *sizep = size;
1434 break;
1435 default:
1436 die("pack %s contains unknown object type %d",
1437 p->pack_name, type);
1438 }
1439 unuse_pack(&w_curs);
1440 return type;
1441 }
1442
1443 static void *unpack_compressed_entry(struct packed_git *p,
1444 struct pack_window **w_curs,
1445 off_t curpos,
1446 unsigned long size)
1447 {
1448 int st;
1449 z_stream stream;
1450 unsigned char *buffer, *in;
1451
1452 buffer = xmalloc(size + 1);
1453 buffer[size] = 0;
1454 memset(&stream, 0, sizeof(stream));
1455 stream.next_out = buffer;
1456 stream.avail_out = size;
1457
1458 inflateInit(&stream);
1459 do {
1460 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1461 stream.next_in = in;
1462 st = inflate(&stream, Z_FINISH);
1463 curpos += stream.next_in - in;
1464 } while (st == Z_OK || st == Z_BUF_ERROR);
1465 inflateEnd(&stream);
1466 if ((st != Z_STREAM_END) || stream.total_out != size) {
1467 free(buffer);
1468 return NULL;
1469 }
1470
1471 return buffer;
1472 }
1473
1474 #define MAX_DELTA_CACHE (256)
1475
1476 static size_t delta_base_cached;
1477
1478 static struct delta_base_cache_lru_list {
1479 struct delta_base_cache_lru_list *prev;
1480 struct delta_base_cache_lru_list *next;
1481 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1482
1483 static struct delta_base_cache_entry {
1484 struct delta_base_cache_lru_list lru;
1485 void *data;
1486 struct packed_git *p;
1487 off_t base_offset;
1488 unsigned long size;
1489 enum object_type type;
1490 } delta_base_cache[MAX_DELTA_CACHE];
1491
1492 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1493 {
1494 unsigned long hash;
1495
1496 hash = (unsigned long)p + (unsigned long)base_offset;
1497 hash += (hash >> 8) + (hash >> 16);
1498 return hash % MAX_DELTA_CACHE;
1499 }
1500
1501 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1502 unsigned long *base_size, enum object_type *type, int keep_cache)
1503 {
1504 void *ret;
1505 unsigned long hash = pack_entry_hash(p, base_offset);
1506 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1507
1508 ret = ent->data;
1509 if (ret && ent->p == p && ent->base_offset == base_offset)
1510 goto found_cache_entry;
1511 return unpack_entry(p, base_offset, type, base_size);
1512
1513 found_cache_entry:
1514 if (!keep_cache) {
1515 ent->data = NULL;
1516 ent->lru.next->prev = ent->lru.prev;
1517 ent->lru.prev->next = ent->lru.next;
1518 delta_base_cached -= ent->size;
1519 } else {
1520 ret = xmemdupz(ent->data, ent->size);
1521 }
1522 *type = ent->type;
1523 *base_size = ent->size;
1524 return ret;
1525 }
1526
1527 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1528 {
1529 if (ent->data) {
1530 free(ent->data);
1531 ent->data = NULL;
1532 ent->lru.next->prev = ent->lru.prev;
1533 ent->lru.prev->next = ent->lru.next;
1534 delta_base_cached -= ent->size;
1535 }
1536 }
1537
1538 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1539 void *base, unsigned long base_size, enum object_type type)
1540 {
1541 unsigned long hash = pack_entry_hash(p, base_offset);
1542 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1543 struct delta_base_cache_lru_list *lru;
1544
1545 release_delta_base_cache(ent);
1546 delta_base_cached += base_size;
1547
1548 for (lru = delta_base_cache_lru.next;
1549 delta_base_cached > delta_base_cache_limit
1550 && lru != &delta_base_cache_lru;
1551 lru = lru->next) {
1552 struct delta_base_cache_entry *f = (void *)lru;
1553 if (f->type == OBJ_BLOB)
1554 release_delta_base_cache(f);
1555 }
1556 for (lru = delta_base_cache_lru.next;
1557 delta_base_cached > delta_base_cache_limit
1558 && lru != &delta_base_cache_lru;
1559 lru = lru->next) {
1560 struct delta_base_cache_entry *f = (void *)lru;
1561 release_delta_base_cache(f);
1562 }
1563
1564 ent->p = p;
1565 ent->base_offset = base_offset;
1566 ent->type = type;
1567 ent->data = base;
1568 ent->size = base_size;
1569 ent->lru.next = &delta_base_cache_lru;
1570 ent->lru.prev = delta_base_cache_lru.prev;
1571 delta_base_cache_lru.prev->next = &ent->lru;
1572 delta_base_cache_lru.prev = &ent->lru;
1573 }
1574
1575 static void *unpack_delta_entry(struct packed_git *p,
1576 struct pack_window **w_curs,
1577 off_t curpos,
1578 unsigned long delta_size,
1579 off_t obj_offset,
1580 enum object_type *type,
1581 unsigned long *sizep)
1582 {
1583 void *delta_data, *result, *base;
1584 unsigned long base_size;
1585 off_t base_offset;
1586
1587 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1588 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1589 if (!base)
1590 die("failed to read delta base object"
1591 " at %"PRIuMAX" from %s",
1592 (uintmax_t)base_offset, p->pack_name);
1593
1594 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1595 if (!delta_data)
1596 die("failed to unpack compressed delta"
1597 " at %"PRIuMAX" from %s",
1598 (uintmax_t)curpos, p->pack_name);
1599 result = patch_delta(base, base_size,
1600 delta_data, delta_size,
1601 sizep);
1602 if (!result)
1603 die("failed to apply delta");
1604 free(delta_data);
1605 add_delta_base_cache(p, base_offset, base, base_size, *type);
1606 return result;
1607 }
1608
1609 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1610 enum object_type *type, unsigned long *sizep)
1611 {
1612 struct pack_window *w_curs = NULL;
1613 off_t curpos = obj_offset;
1614 void *data;
1615
1616 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1617 switch (*type) {
1618 case OBJ_OFS_DELTA:
1619 case OBJ_REF_DELTA:
1620 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1621 obj_offset, type, sizep);
1622 break;
1623 case OBJ_COMMIT:
1624 case OBJ_TREE:
1625 case OBJ_BLOB:
1626 case OBJ_TAG:
1627 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1628 break;
1629 default:
1630 die("unknown object type %i in %s", *type, p->pack_name);
1631 }
1632 unuse_pack(&w_curs);
1633 return data;
1634 }
1635
1636 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1637 uint32_t n)
1638 {
1639 const unsigned char *index = p->index_data;
1640 if (!index) {
1641 if (open_pack_index(p))
1642 return NULL;
1643 index = p->index_data;
1644 }
1645 if (n >= p->num_objects)
1646 return NULL;
1647 index += 4 * 256;
1648 if (p->index_version == 1) {
1649 return index + 24 * n + 4;
1650 } else {
1651 index += 8;
1652 return index + 20 * n;
1653 }
1654 }
1655
1656 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1657 {
1658 const unsigned char *index = p->index_data;
1659 index += 4 * 256;
1660 if (p->index_version == 1) {
1661 return ntohl(*((uint32_t *)(index + 24 * n)));
1662 } else {
1663 uint32_t off;
1664 index += 8 + p->num_objects * (20 + 4);
1665 off = ntohl(*((uint32_t *)(index + 4 * n)));
1666 if (!(off & 0x80000000))
1667 return off;
1668 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1669 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1670 ntohl(*((uint32_t *)(index + 4)));
1671 }
1672 }
1673
1674 off_t find_pack_entry_one(const unsigned char *sha1,
1675 struct packed_git *p)
1676 {
1677 const uint32_t *level1_ofs = p->index_data;
1678 const unsigned char *index = p->index_data;
1679 unsigned hi, lo, stride;
1680 static int use_lookup = -1;
1681 static int debug_lookup = -1;
1682
1683 if (debug_lookup < 0)
1684 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1685
1686 if (!index) {
1687 if (open_pack_index(p))
1688 return 0;
1689 level1_ofs = p->index_data;
1690 index = p->index_data;
1691 }
1692 if (p->index_version > 1) {
1693 level1_ofs += 2;
1694 index += 8;
1695 }
1696 index += 4 * 256;
1697 hi = ntohl(level1_ofs[*sha1]);
1698 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1699 if (p->index_version > 1) {
1700 stride = 20;
1701 } else {
1702 stride = 24;
1703 index += 4;
1704 }
1705
1706 if (debug_lookup)
1707 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1708 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1709
1710 if (use_lookup < 0)
1711 use_lookup = !!getenv("GIT_USE_LOOKUP");
1712 if (use_lookup) {
1713 int pos = sha1_entry_pos(index, stride, 0,
1714 lo, hi, p->num_objects, sha1);
1715 if (pos < 0)
1716 return 0;
1717 return nth_packed_object_offset(p, pos);
1718 }
1719
1720 do {
1721 unsigned mi = (lo + hi) / 2;
1722 int cmp = hashcmp(index + mi * stride, sha1);
1723
1724 if (debug_lookup)
1725 printf("lo %u hi %u rg %u mi %u\n",
1726 lo, hi, hi - lo, mi);
1727 if (!cmp)
1728 return nth_packed_object_offset(p, mi);
1729 if (cmp > 0)
1730 hi = mi;
1731 else
1732 lo = mi+1;
1733 } while (lo < hi);
1734 return 0;
1735 }
1736
1737 int matches_pack_name(struct packed_git *p, const char *name)
1738 {
1739 const char *last_c, *c;
1740
1741 if (!strcmp(p->pack_name, name))
1742 return 1;
1743
1744 for (c = p->pack_name, last_c = c; *c;)
1745 if (*c == '/')
1746 last_c = ++c;
1747 else
1748 ++c;
1749 if (!strcmp(last_c, name))
1750 return 1;
1751
1752 return 0;
1753 }
1754
1755 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1756 {
1757 static struct packed_git *last_found = (void *)1;
1758 struct packed_git *p;
1759 off_t offset;
1760
1761 prepare_packed_git();
1762 if (!packed_git)
1763 return 0;
1764 p = (last_found == (void *)1) ? packed_git : last_found;
1765
1766 do {
1767 if (ignore_packed) {
1768 const char **ig;
1769 for (ig = ignore_packed; *ig; ig++)
1770 if (matches_pack_name(p, *ig))
1771 break;
1772 if (*ig)
1773 goto next;
1774 }
1775
1776 offset = find_pack_entry_one(sha1, p);
1777 if (offset) {
1778 /*
1779 * We are about to tell the caller where they can
1780 * locate the requested object. We better make
1781 * sure the packfile is still here and can be
1782 * accessed before supplying that answer, as
1783 * it may have been deleted since the index
1784 * was loaded!
1785 */
1786 if (p->pack_fd == -1 && open_packed_git(p)) {
1787 error("packfile %s cannot be accessed", p->pack_name);
1788 goto next;
1789 }
1790 e->offset = offset;
1791 e->p = p;
1792 hashcpy(e->sha1, sha1);
1793 last_found = p;
1794 return 1;
1795 }
1796
1797 next:
1798 if (p == last_found)
1799 p = packed_git;
1800 else
1801 p = p->next;
1802 if (p == last_found)
1803 p = p->next;
1804 } while (p);
1805 return 0;
1806 }
1807
1808 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1809 struct packed_git *packs)
1810 {
1811 struct packed_git *p;
1812
1813 for (p = packs; p; p = p->next) {
1814 if (find_pack_entry_one(sha1, p))
1815 return p;
1816 }
1817 return NULL;
1818
1819 }
1820
1821 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1822 {
1823 int status;
1824 unsigned long mapsize, size;
1825 void *map;
1826 z_stream stream;
1827 char hdr[32];
1828
1829 map = map_sha1_file(sha1, &mapsize);
1830 if (!map)
1831 return error("unable to find %s", sha1_to_hex(sha1));
1832 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1833 status = error("unable to unpack %s header",
1834 sha1_to_hex(sha1));
1835 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1836 status = error("unable to parse %s header", sha1_to_hex(sha1));
1837 else if (sizep)
1838 *sizep = size;
1839 inflateEnd(&stream);
1840 munmap(map, mapsize);
1841 return status;
1842 }
1843
1844 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1845 {
1846 struct pack_entry e;
1847
1848 if (!find_pack_entry(sha1, &e, NULL)) {
1849 reprepare_packed_git();
1850 if (!find_pack_entry(sha1, &e, NULL))
1851 return sha1_loose_object_info(sha1, sizep);
1852 }
1853 return packed_object_info(e.p, e.offset, sizep);
1854 }
1855
1856 static void *read_packed_sha1(const unsigned char *sha1,
1857 enum object_type *type, unsigned long *size)
1858 {
1859 struct pack_entry e;
1860
1861 if (!find_pack_entry(sha1, &e, NULL))
1862 return NULL;
1863 else
1864 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1865 }
1866
1867 /*
1868 * This is meant to hold a *small* number of objects that you would
1869 * want read_sha1_file() to be able to return, but yet you do not want
1870 * to write them into the object store (e.g. a browse-only
1871 * application).
1872 */
1873 static struct cached_object {
1874 unsigned char sha1[20];
1875 enum object_type type;
1876 void *buf;
1877 unsigned long size;
1878 } *cached_objects;
1879 static int cached_object_nr, cached_object_alloc;
1880
1881 static struct cached_object empty_tree = {
1882 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1883 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1884 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1885 OBJ_TREE,
1886 "",
1887 0
1888 };
1889
1890 static struct cached_object *find_cached_object(const unsigned char *sha1)
1891 {
1892 int i;
1893 struct cached_object *co = cached_objects;
1894
1895 for (i = 0; i < cached_object_nr; i++, co++) {
1896 if (!hashcmp(co->sha1, sha1))
1897 return co;
1898 }
1899 if (!hashcmp(sha1, empty_tree.sha1))
1900 return &empty_tree;
1901 return NULL;
1902 }
1903
1904 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1905 unsigned char *sha1)
1906 {
1907 struct cached_object *co;
1908
1909 hash_sha1_file(buf, len, typename(type), sha1);
1910 if (has_sha1_file(sha1) || find_cached_object(sha1))
1911 return 0;
1912 if (cached_object_alloc <= cached_object_nr) {
1913 cached_object_alloc = alloc_nr(cached_object_alloc);
1914 cached_objects = xrealloc(cached_objects,
1915 sizeof(*cached_objects) *
1916 cached_object_alloc);
1917 }
1918 co = &cached_objects[cached_object_nr++];
1919 co->size = len;
1920 co->type = type;
1921 co->buf = xmalloc(len);
1922 memcpy(co->buf, buf, len);
1923 hashcpy(co->sha1, sha1);
1924 return 0;
1925 }
1926
1927 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1928 unsigned long *size)
1929 {
1930 unsigned long mapsize;
1931 void *map, *buf;
1932 struct cached_object *co;
1933
1934 co = find_cached_object(sha1);
1935 if (co) {
1936 *type = co->type;
1937 *size = co->size;
1938 return xmemdupz(co->buf, co->size);
1939 }
1940
1941 buf = read_packed_sha1(sha1, type, size);
1942 if (buf)
1943 return buf;
1944 map = map_sha1_file(sha1, &mapsize);
1945 if (map) {
1946 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1947 munmap(map, mapsize);
1948 return buf;
1949 }
1950 reprepare_packed_git();
1951 return read_packed_sha1(sha1, type, size);
1952 }
1953
1954 void *read_object_with_reference(const unsigned char *sha1,
1955 const char *required_type_name,
1956 unsigned long *size,
1957 unsigned char *actual_sha1_return)
1958 {
1959 enum object_type type, required_type;
1960 void *buffer;
1961 unsigned long isize;
1962 unsigned char actual_sha1[20];
1963
1964 required_type = type_from_string(required_type_name);
1965 hashcpy(actual_sha1, sha1);
1966 while (1) {
1967 int ref_length = -1;
1968 const char *ref_type = NULL;
1969
1970 buffer = read_sha1_file(actual_sha1, &type, &isize);
1971 if (!buffer)
1972 return NULL;
1973 if (type == required_type) {
1974 *size = isize;
1975 if (actual_sha1_return)
1976 hashcpy(actual_sha1_return, actual_sha1);
1977 return buffer;
1978 }
1979 /* Handle references */
1980 else if (type == OBJ_COMMIT)
1981 ref_type = "tree ";
1982 else if (type == OBJ_TAG)
1983 ref_type = "object ";
1984 else {
1985 free(buffer);
1986 return NULL;
1987 }
1988 ref_length = strlen(ref_type);
1989
1990 if (ref_length + 40 > isize ||
1991 memcmp(buffer, ref_type, ref_length) ||
1992 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1993 free(buffer);
1994 return NULL;
1995 }
1996 free(buffer);
1997 /* Now we have the ID of the referred-to object in
1998 * actual_sha1. Check again. */
1999 }
2000 }
2001
2002 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2003 const char *type, unsigned char *sha1,
2004 char *hdr, int *hdrlen)
2005 {
2006 SHA_CTX c;
2007
2008 /* Generate the header */
2009 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2010
2011 /* Sha1.. */
2012 SHA1_Init(&c);
2013 SHA1_Update(&c, hdr, *hdrlen);
2014 SHA1_Update(&c, buf, len);
2015 SHA1_Final(sha1, &c);
2016 }
2017
2018 /*
2019 * Link the tempfile to the final place, possibly creating the
2020 * last directory level as you do so.
2021 *
2022 * Returns the errno on failure, 0 on success.
2023 */
2024 static int link_temp_to_file(const char *tmpfile, const char *filename)
2025 {
2026 int ret;
2027 char *dir;
2028
2029 if (!link(tmpfile, filename))
2030 return 0;
2031
2032 /*
2033 * Try to mkdir the last path component if that failed.
2034 *
2035 * Re-try the "link()" regardless of whether the mkdir
2036 * succeeds, since a race might mean that somebody
2037 * else succeeded.
2038 */
2039 ret = errno;
2040 dir = strrchr(filename, '/');
2041 if (dir) {
2042 *dir = 0;
2043 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
2044 *dir = '/';
2045 return -2;
2046 }
2047 *dir = '/';
2048 if (!link(tmpfile, filename))
2049 return 0;
2050 ret = errno;
2051 }
2052 return ret;
2053 }
2054
2055 /*
2056 * Move the just written object into its final resting place
2057 */
2058 int move_temp_to_file(const char *tmpfile, const char *filename)
2059 {
2060 int ret = link_temp_to_file(tmpfile, filename);
2061
2062 /*
2063 * Coda hack - coda doesn't like cross-directory links,
2064 * so we fall back to a rename, which will mean that it
2065 * won't be able to check collisions, but that's not a
2066 * big deal.
2067 *
2068 * The same holds for FAT formatted media.
2069 *
2070 * When this succeeds, we just return 0. We have nothing
2071 * left to unlink.
2072 */
2073 if (ret && ret != EEXIST) {
2074 if (!rename(tmpfile, filename))
2075 return 0;
2076 ret = errno;
2077 }
2078 unlink(tmpfile);
2079 if (ret) {
2080 if (ret != EEXIST) {
2081 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2082 }
2083 /* FIXME!!! Collision check here ? */
2084 }
2085
2086 return 0;
2087 }
2088
2089 static int write_buffer(int fd, const void *buf, size_t len)
2090 {
2091 if (write_in_full(fd, buf, len) < 0)
2092 return error("file write error (%s)", strerror(errno));
2093 return 0;
2094 }
2095
2096 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2097 unsigned char *sha1)
2098 {
2099 char hdr[32];
2100 int hdrlen;
2101 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2102 return 0;
2103 }
2104
2105 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2106 {
2107 int size, ret;
2108 unsigned char *compressed;
2109 z_stream stream;
2110 unsigned char sha1[20];
2111 char *filename;
2112 static char tmpfile[PATH_MAX];
2113 char hdr[32];
2114 int fd, hdrlen;
2115
2116 /* Normally if we have it in the pack then we do not bother writing
2117 * it out into .git/objects/??/?{38} file.
2118 */
2119 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2120 filename = sha1_file_name(sha1);
2121 if (returnsha1)
2122 hashcpy(returnsha1, sha1);
2123 if (has_sha1_file(sha1))
2124 return 0;
2125 fd = open(filename, O_RDONLY);
2126 if (fd >= 0) {
2127 /*
2128 * FIXME!!! We might do collision checking here, but we'd
2129 * need to uncompress the old file and check it. Later.
2130 */
2131 close(fd);
2132 return 0;
2133 }
2134
2135 if (errno != ENOENT) {
2136 return error("sha1 file %s: %s\n", filename, strerror(errno));
2137 }
2138
2139 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2140
2141 fd = mkstemp(tmpfile);
2142 if (fd < 0) {
2143 if (errno == EPERM)
2144 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2145 else
2146 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2147 }
2148
2149 /* Set it up */
2150 memset(&stream, 0, sizeof(stream));
2151 deflateInit(&stream, zlib_compression_level);
2152 size = 8 + deflateBound(&stream, len+hdrlen);
2153 compressed = xmalloc(size);
2154
2155 /* Compress it */
2156 stream.next_out = compressed;
2157 stream.avail_out = size;
2158
2159 /* First header.. */
2160 stream.next_in = (unsigned char *)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 = buf;
2167 stream.avail_in = len;
2168 ret = deflate(&stream, Z_FINISH);
2169 if (ret != Z_STREAM_END)
2170 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2171
2172 ret = deflateEnd(&stream);
2173 if (ret != Z_OK)
2174 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2175
2176 size = stream.total_out;
2177
2178 if (write_buffer(fd, compressed, size) < 0)
2179 die("unable to write sha1 file");
2180 fchmod(fd, 0444);
2181 if (close(fd))
2182 die("unable to write sha1 file");
2183 free(compressed);
2184
2185 return move_temp_to_file(tmpfile, filename);
2186 }
2187
2188 /*
2189 * We need to unpack and recompress the object for writing
2190 * it out to a different file.
2191 */
2192 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2193 {
2194 size_t size;
2195 z_stream stream;
2196 unsigned char *unpacked;
2197 unsigned long len;
2198 enum object_type type;
2199 char hdr[32];
2200 int hdrlen;
2201 void *buf;
2202
2203 /* need to unpack and recompress it by itself */
2204 unpacked = read_packed_sha1(sha1, &type, &len);
2205 if (!unpacked)
2206 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2207
2208 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2209
2210 /* Set it up */
2211 memset(&stream, 0, sizeof(stream));
2212 deflateInit(&stream, zlib_compression_level);
2213 size = deflateBound(&stream, len + hdrlen);
2214 buf = xmalloc(size);
2215
2216 /* Compress it */
2217 stream.next_out = buf;
2218 stream.avail_out = size;
2219
2220 /* First header.. */
2221 stream.next_in = (void *)hdr;
2222 stream.avail_in = hdrlen;
2223 while (deflate(&stream, 0) == Z_OK)
2224 /* nothing */;
2225
2226 /* Then the data itself.. */
2227 stream.next_in = unpacked;
2228 stream.avail_in = len;
2229 while (deflate(&stream, Z_FINISH) == Z_OK)
2230 /* nothing */;
2231 deflateEnd(&stream);
2232 free(unpacked);
2233
2234 *objsize = stream.total_out;
2235 return buf;
2236 }
2237
2238 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2239 {
2240 int retval;
2241 unsigned long objsize;
2242 void *buf = map_sha1_file(sha1, &objsize);
2243
2244 if (buf) {
2245 retval = write_buffer(fd, buf, objsize);
2246 munmap(buf, objsize);
2247 return retval;
2248 }
2249
2250 buf = repack_object(sha1, &objsize);
2251 retval = write_buffer(fd, buf, objsize);
2252 free(buf);
2253 return retval;
2254 }
2255
2256 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2257 size_t bufsize, size_t *bufposn)
2258 {
2259 char tmpfile[PATH_MAX];
2260 int local;
2261 z_stream stream;
2262 unsigned char real_sha1[20];
2263 unsigned char discard[4096];
2264 int ret;
2265 SHA_CTX c;
2266
2267 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2268
2269 local = mkstemp(tmpfile);
2270 if (local < 0) {
2271 if (errno == EPERM)
2272 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2273 else
2274 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2275 }
2276
2277 memset(&stream, 0, sizeof(stream));
2278
2279 inflateInit(&stream);
2280
2281 SHA1_Init(&c);
2282
2283 do {
2284 ssize_t size;
2285 if (*bufposn) {
2286 stream.avail_in = *bufposn;
2287 stream.next_in = (unsigned char *) buffer;
2288 do {
2289 stream.next_out = discard;
2290 stream.avail_out = sizeof(discard);
2291 ret = inflate(&stream, Z_SYNC_FLUSH);
2292 SHA1_Update(&c, discard, sizeof(discard) -
2293 stream.avail_out);
2294 } while (stream.avail_in && ret == Z_OK);
2295 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2296 die("unable to write sha1 file");
2297 memmove(buffer, buffer + *bufposn - stream.avail_in,
2298 stream.avail_in);
2299 *bufposn = stream.avail_in;
2300 if (ret != Z_OK)
2301 break;
2302 }
2303 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2304 if (size <= 0) {
2305 close(local);
2306 unlink(tmpfile);
2307 if (!size)
2308 return error("Connection closed?");
2309 perror("Reading from connection");
2310 return -1;
2311 }
2312 *bufposn += size;
2313 } while (1);
2314 inflateEnd(&stream);
2315
2316 fchmod(local, 0444);
2317 if (close(local) != 0)
2318 die("unable to write sha1 file");
2319 SHA1_Final(real_sha1, &c);
2320 if (ret != Z_STREAM_END) {
2321 unlink(tmpfile);
2322 return error("File %s corrupted", sha1_to_hex(sha1));
2323 }
2324 if (hashcmp(sha1, real_sha1)) {
2325 unlink(tmpfile);
2326 return error("File %s has bad hash", sha1_to_hex(sha1));
2327 }
2328
2329 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2330 }
2331
2332 int has_pack_index(const unsigned char *sha1)
2333 {
2334 struct stat st;
2335 if (stat(sha1_pack_index_name(sha1), &st))
2336 return 0;
2337 return 1;
2338 }
2339
2340 int has_pack_file(const unsigned char *sha1)
2341 {
2342 struct stat st;
2343 if (stat(sha1_pack_name(sha1), &st))
2344 return 0;
2345 return 1;
2346 }
2347
2348 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2349 {
2350 struct pack_entry e;
2351 return find_pack_entry(sha1, &e, ignore_packed);
2352 }
2353
2354 int has_sha1_file(const unsigned char *sha1)
2355 {
2356 struct stat st;
2357 struct pack_entry e;
2358
2359 if (find_pack_entry(sha1, &e, NULL))
2360 return 1;
2361 return find_sha1_file(sha1, &st) ? 1 : 0;
2362 }
2363
2364 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2365 {
2366 struct strbuf buf;
2367 int ret;
2368
2369 strbuf_init(&buf, 0);
2370 if (strbuf_read(&buf, fd, 4096) < 0) {
2371 strbuf_release(&buf);
2372 return -1;
2373 }
2374
2375 if (!type)
2376 type = blob_type;
2377 if (write_object)
2378 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2379 else
2380 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2381 strbuf_release(&buf);
2382
2383 return ret;
2384 }
2385
2386 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2387 enum object_type type, const char *path)
2388 {
2389 size_t size = xsize_t(st->st_size);
2390 void *buf = NULL;
2391 int ret, re_allocated = 0;
2392
2393 if (size)
2394 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2395 close(fd);
2396
2397 if (!type)
2398 type = OBJ_BLOB;
2399
2400 /*
2401 * Convert blobs to git internal format
2402 */
2403 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2404 struct strbuf nbuf;
2405 strbuf_init(&nbuf, 0);
2406 if (convert_to_git(path, buf, size, &nbuf,
2407 write_object ? safe_crlf : 0)) {
2408 munmap(buf, size);
2409 buf = strbuf_detach(&nbuf, &size);
2410 re_allocated = 1;
2411 }
2412 }
2413
2414 if (write_object)
2415 ret = write_sha1_file(buf, size, typename(type), sha1);
2416 else
2417 ret = hash_sha1_file(buf, size, typename(type), sha1);
2418 if (re_allocated) {
2419 free(buf);
2420 return ret;
2421 }
2422 if (size)
2423 munmap(buf, size);
2424 return ret;
2425 }
2426
2427 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2428 {
2429 int fd;
2430 char *target;
2431 size_t len;
2432
2433 switch (st->st_mode & S_IFMT) {
2434 case S_IFREG:
2435 fd = open(path, O_RDONLY);
2436 if (fd < 0)
2437 return error("open(\"%s\"): %s", path,
2438 strerror(errno));
2439 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2440 return error("%s: failed to insert into database",
2441 path);
2442 break;
2443 case S_IFLNK:
2444 len = xsize_t(st->st_size);
2445 target = xmalloc(len + 1);
2446 if (readlink(path, target, len + 1) != st->st_size) {
2447 char *errstr = strerror(errno);
2448 free(target);
2449 return error("readlink(\"%s\"): %s", path,
2450 errstr);
2451 }
2452 if (!write_object)
2453 hash_sha1_file(target, len, blob_type, sha1);
2454 else if (write_sha1_file(target, len, blob_type, sha1))
2455 return error("%s: failed to insert into database",
2456 path);
2457 free(target);
2458 break;
2459 case S_IFDIR:
2460 return resolve_gitlink_ref(path, "HEAD", sha1);
2461 default:
2462 return error("%s: unsupported file type", path);
2463 }
2464 return 0;
2465 }
2466
2467 int read_pack_header(int fd, struct pack_header *header)
2468 {
2469 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2470 /* "eof before pack header was fully read" */
2471 return PH_ERROR_EOF;
2472
2473 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2474 /* "protocol error (pack signature mismatch detected)" */
2475 return PH_ERROR_PACK_SIGNATURE;
2476 if (!pack_version_ok(header->hdr_version))
2477 /* "protocol error (pack version unsupported)" */
2478 return PH_ERROR_PROTOCOL;
2479 return 0;
2480 }