Rewrite convert_to_{git,working_tree} to use strbuf's.
[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 "strbuf.h"
18
19 #ifndef O_NOATIME
20 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
21 #define O_NOATIME 01000000
22 #else
23 #define O_NOATIME 0
24 #endif
25 #endif
26
27 #ifdef NO_C99_FORMAT
28 #define SZ_FMT "lu"
29 #else
30 #define SZ_FMT "zu"
31 #endif
32
33 const unsigned char null_sha1[20];
34
35 static unsigned int sha1_file_open_flag = O_NOATIME;
36
37 const signed char hexval_table[256] = {
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
44 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
45 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
46 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
50 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
70 };
71
72 int get_sha1_hex(const char *hex, unsigned char *sha1)
73 {
74 int i;
75 for (i = 0; i < 20; i++) {
76 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
77 if (val & ~0xff)
78 return -1;
79 *sha1++ = val;
80 hex += 2;
81 }
82 return 0;
83 }
84
85 int safe_create_leading_directories(char *path)
86 {
87 char *pos = path;
88 struct stat st;
89
90 if (*pos == '/')
91 pos++;
92
93 while (pos) {
94 pos = strchr(pos, '/');
95 if (!pos)
96 break;
97 *pos = 0;
98 if (!stat(path, &st)) {
99 /* path exists */
100 if (!S_ISDIR(st.st_mode)) {
101 *pos = '/';
102 return -3;
103 }
104 }
105 else if (mkdir(path, 0777)) {
106 *pos = '/';
107 return -1;
108 }
109 else if (adjust_shared_perm(path)) {
110 *pos = '/';
111 return -2;
112 }
113 *pos++ = '/';
114 }
115 return 0;
116 }
117
118 char * sha1_to_hex(const unsigned char *sha1)
119 {
120 static int bufno;
121 static char hexbuffer[4][50];
122 static const char hex[] = "0123456789abcdef";
123 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
124 int i;
125
126 for (i = 0; i < 20; i++) {
127 unsigned int val = *sha1++;
128 *buf++ = hex[val >> 4];
129 *buf++ = hex[val & 0xf];
130 }
131 *buf = '\0';
132
133 return buffer;
134 }
135
136 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
137 {
138 int i;
139 for (i = 0; i < 20; i++) {
140 static char hex[] = "0123456789abcdef";
141 unsigned int val = sha1[i];
142 char *pos = pathbuf + i*2 + (i > 0);
143 *pos++ = hex[val >> 4];
144 *pos = hex[val & 0xf];
145 }
146 }
147
148 /*
149 * NOTE! This returns a statically allocated buffer, so you have to be
150 * careful about using it. Do a "xstrdup()" if you need to save the
151 * filename.
152 *
153 * Also note that this returns the location for creating. Reading
154 * SHA1 file can happen from any alternate directory listed in the
155 * DB_ENVIRONMENT environment variable if it is not found in
156 * the primary object database.
157 */
158 char *sha1_file_name(const unsigned char *sha1)
159 {
160 static char *name, *base;
161
162 if (!base) {
163 const char *sha1_file_directory = get_object_directory();
164 int len = strlen(sha1_file_directory);
165 base = xmalloc(len + 60);
166 memcpy(base, sha1_file_directory, len);
167 memset(base+len, 0, 60);
168 base[len] = '/';
169 base[len+3] = '/';
170 name = base + len + 1;
171 }
172 fill_sha1_path(name, sha1);
173 return base;
174 }
175
176 char *sha1_pack_name(const unsigned char *sha1)
177 {
178 static const char hex[] = "0123456789abcdef";
179 static char *name, *base, *buf;
180 int i;
181
182 if (!base) {
183 const char *sha1_file_directory = get_object_directory();
184 int len = strlen(sha1_file_directory);
185 base = xmalloc(len + 60);
186 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
187 name = base + len + 11;
188 }
189
190 buf = name;
191
192 for (i = 0; i < 20; i++) {
193 unsigned int val = *sha1++;
194 *buf++ = hex[val >> 4];
195 *buf++ = hex[val & 0xf];
196 }
197
198 return base;
199 }
200
201 char *sha1_pack_index_name(const unsigned char *sha1)
202 {
203 static const char hex[] = "0123456789abcdef";
204 static char *name, *base, *buf;
205 int i;
206
207 if (!base) {
208 const char *sha1_file_directory = get_object_directory();
209 int len = strlen(sha1_file_directory);
210 base = xmalloc(len + 60);
211 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
212 name = base + len + 11;
213 }
214
215 buf = name;
216
217 for (i = 0; i < 20; i++) {
218 unsigned int val = *sha1++;
219 *buf++ = hex[val >> 4];
220 *buf++ = hex[val & 0xf];
221 }
222
223 return base;
224 }
225
226 struct alternate_object_database *alt_odb_list;
227 static struct alternate_object_database **alt_odb_tail;
228
229 static void read_info_alternates(const char * alternates, int depth);
230
231 /*
232 * Prepare alternate object database registry.
233 *
234 * The variable alt_odb_list points at the list of struct
235 * alternate_object_database. The elements on this list come from
236 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
237 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
238 * whose contents is similar to that environment variable but can be
239 * LF separated. Its base points at a statically allocated buffer that
240 * contains "/the/directory/corresponding/to/.git/objects/...", while
241 * its name points just after the slash at the end of ".git/objects/"
242 * in the example above, and has enough space to hold 40-byte hex
243 * SHA1, an extra slash for the first level indirection, and the
244 * terminating NUL.
245 */
246 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
247 {
248 struct stat st;
249 const char *objdir = get_object_directory();
250 struct alternate_object_database *ent;
251 struct alternate_object_database *alt;
252 /* 43 = 40-byte + 2 '/' + terminating NUL */
253 int pfxlen = len;
254 int entlen = pfxlen + 43;
255 int base_len = -1;
256
257 if (*entry != '/' && relative_base) {
258 /* Relative alt-odb */
259 if (base_len < 0)
260 base_len = strlen(relative_base) + 1;
261 entlen += base_len;
262 pfxlen += base_len;
263 }
264 ent = xmalloc(sizeof(*ent) + entlen);
265
266 if (*entry != '/' && relative_base) {
267 memcpy(ent->base, relative_base, base_len - 1);
268 ent->base[base_len - 1] = '/';
269 memcpy(ent->base + base_len, entry, len);
270 }
271 else
272 memcpy(ent->base, entry, pfxlen);
273
274 ent->name = ent->base + pfxlen + 1;
275 ent->base[pfxlen + 3] = '/';
276 ent->base[pfxlen] = ent->base[entlen-1] = 0;
277
278 /* Detect cases where alternate disappeared */
279 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
280 error("object directory %s does not exist; "
281 "check .git/objects/info/alternates.",
282 ent->base);
283 free(ent);
284 return -1;
285 }
286
287 /* Prevent the common mistake of listing the same
288 * thing twice, or object directory itself.
289 */
290 for (alt = alt_odb_list; alt; alt = alt->next) {
291 if (!memcmp(ent->base, alt->base, pfxlen)) {
292 free(ent);
293 return -1;
294 }
295 }
296 if (!memcmp(ent->base, objdir, pfxlen)) {
297 free(ent);
298 return -1;
299 }
300
301 /* add the alternate entry */
302 *alt_odb_tail = ent;
303 alt_odb_tail = &(ent->next);
304 ent->next = NULL;
305
306 /* recursively add alternates */
307 read_info_alternates(ent->base, depth + 1);
308
309 ent->base[pfxlen] = '/';
310
311 return 0;
312 }
313
314 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
315 const char *relative_base, int depth)
316 {
317 const char *cp, *last;
318
319 if (depth > 5) {
320 error("%s: ignoring alternate object stores, nesting too deep.",
321 relative_base);
322 return;
323 }
324
325 last = alt;
326 while (last < ep) {
327 cp = last;
328 if (cp < ep && *cp == '#') {
329 while (cp < ep && *cp != sep)
330 cp++;
331 last = cp + 1;
332 continue;
333 }
334 while (cp < ep && *cp != sep)
335 cp++;
336 if (last != cp) {
337 if ((*last != '/') && depth) {
338 error("%s: ignoring relative alternate object store %s",
339 relative_base, last);
340 } else {
341 link_alt_odb_entry(last, cp - last,
342 relative_base, depth);
343 }
344 }
345 while (cp < ep && *cp == sep)
346 cp++;
347 last = cp;
348 }
349 }
350
351 static void read_info_alternates(const char * relative_base, int depth)
352 {
353 char *map;
354 size_t mapsz;
355 struct stat st;
356 const char alt_file_name[] = "info/alternates";
357 /* Given that relative_base is no longer than PATH_MAX,
358 ensure that "path" has enough space to append "/", the
359 file name, "info/alternates", and a trailing NUL. */
360 char path[PATH_MAX + 1 + sizeof alt_file_name];
361 int fd;
362
363 sprintf(path, "%s/%s", relative_base, alt_file_name);
364 fd = open(path, O_RDONLY);
365 if (fd < 0)
366 return;
367 if (fstat(fd, &st) || (st.st_size == 0)) {
368 close(fd);
369 return;
370 }
371 mapsz = xsize_t(st.st_size);
372 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
373 close(fd);
374
375 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
376
377 munmap(map, mapsz);
378 }
379
380 void prepare_alt_odb(void)
381 {
382 const char *alt;
383
384 if (alt_odb_tail)
385 return;
386
387 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
388 if (!alt) alt = "";
389
390 alt_odb_tail = &alt_odb_list;
391 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
392
393 read_info_alternates(get_object_directory(), 0);
394 }
395
396 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
397 {
398 char *name = sha1_file_name(sha1);
399 struct alternate_object_database *alt;
400
401 if (!stat(name, st))
402 return name;
403 prepare_alt_odb();
404 for (alt = alt_odb_list; alt; alt = alt->next) {
405 name = alt->name;
406 fill_sha1_path(name, sha1);
407 if (!stat(alt->base, st))
408 return alt->base;
409 }
410 return NULL;
411 }
412
413 static unsigned int pack_used_ctr;
414 static unsigned int pack_mmap_calls;
415 static unsigned int peak_pack_open_windows;
416 static unsigned int pack_open_windows;
417 static size_t peak_pack_mapped;
418 static size_t pack_mapped;
419 struct packed_git *packed_git;
420
421 void pack_report(void)
422 {
423 fprintf(stderr,
424 "pack_report: getpagesize() = %10" SZ_FMT "\n"
425 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
426 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
427 (size_t) getpagesize(),
428 packed_git_window_size,
429 packed_git_limit);
430 fprintf(stderr,
431 "pack_report: pack_used_ctr = %10u\n"
432 "pack_report: pack_mmap_calls = %10u\n"
433 "pack_report: pack_open_windows = %10u / %10u\n"
434 "pack_report: pack_mapped = "
435 "%10" SZ_FMT " / %10" SZ_FMT "\n",
436 pack_used_ctr,
437 pack_mmap_calls,
438 pack_open_windows, peak_pack_open_windows,
439 pack_mapped, peak_pack_mapped);
440 }
441
442 static int check_packed_git_idx(const char *path, struct packed_git *p)
443 {
444 void *idx_map;
445 struct pack_idx_header *hdr;
446 size_t idx_size;
447 uint32_t version, nr, i, *index;
448 int fd = open(path, O_RDONLY);
449 struct stat st;
450
451 if (fd < 0)
452 return -1;
453 if (fstat(fd, &st)) {
454 close(fd);
455 return -1;
456 }
457 idx_size = xsize_t(st.st_size);
458 if (idx_size < 4 * 256 + 20 + 20) {
459 close(fd);
460 return error("index file %s is too small", path);
461 }
462 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
463 close(fd);
464
465 hdr = idx_map;
466 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
467 version = ntohl(hdr->idx_version);
468 if (version < 2 || version > 2) {
469 munmap(idx_map, idx_size);
470 return error("index file %s is version %d"
471 " and is not supported by this binary"
472 " (try upgrading GIT to a newer version)",
473 path, version);
474 }
475 } else
476 version = 1;
477
478 nr = 0;
479 index = idx_map;
480 if (version > 1)
481 index += 2; /* skip index header */
482 for (i = 0; i < 256; i++) {
483 uint32_t n = ntohl(index[i]);
484 if (n < nr) {
485 munmap(idx_map, idx_size);
486 return error("non-monotonic index %s", path);
487 }
488 nr = n;
489 }
490
491 if (version == 1) {
492 /*
493 * Total size:
494 * - 256 index entries 4 bytes each
495 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
496 * - 20-byte SHA1 of the packfile
497 * - 20-byte SHA1 file checksum
498 */
499 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
500 munmap(idx_map, idx_size);
501 return error("wrong index v1 file size in %s", path);
502 }
503 } else if (version == 2) {
504 /*
505 * Minimum size:
506 * - 8 bytes of header
507 * - 256 index entries 4 bytes each
508 * - 20-byte sha1 entry * nr
509 * - 4-byte crc entry * nr
510 * - 4-byte offset entry * nr
511 * - 20-byte SHA1 of the packfile
512 * - 20-byte SHA1 file checksum
513 * And after the 4-byte offset table might be a
514 * variable sized table containing 8-byte entries
515 * for offsets larger than 2^31.
516 */
517 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
518 unsigned long max_size = min_size;
519 if (nr)
520 max_size += (nr - 1)*8;
521 if (idx_size < min_size || idx_size > max_size) {
522 munmap(idx_map, idx_size);
523 return error("wrong index v2 file size in %s", path);
524 }
525 if (idx_size != min_size) {
526 /* make sure we can deal with large pack offsets */
527 off_t x = 0x7fffffffUL, y = 0xffffffffUL;
528 if (x > (x + 1) || y > (y + 1)) {
529 munmap(idx_map, idx_size);
530 return error("pack too large for current definition of off_t in %s", path);
531 }
532 }
533 }
534
535 p->index_version = version;
536 p->index_data = idx_map;
537 p->index_size = idx_size;
538 p->num_objects = nr;
539 return 0;
540 }
541
542 int open_pack_index(struct packed_git *p)
543 {
544 char *idx_name;
545 int ret;
546
547 if (p->index_data)
548 return 0;
549
550 idx_name = xstrdup(p->pack_name);
551 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
552 ret = check_packed_git_idx(idx_name, p);
553 free(idx_name);
554 return ret;
555 }
556
557 static void scan_windows(struct packed_git *p,
558 struct packed_git **lru_p,
559 struct pack_window **lru_w,
560 struct pack_window **lru_l)
561 {
562 struct pack_window *w, *w_l;
563
564 for (w_l = NULL, w = p->windows; w; w = w->next) {
565 if (!w->inuse_cnt) {
566 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
567 *lru_p = p;
568 *lru_w = w;
569 *lru_l = w_l;
570 }
571 }
572 w_l = w;
573 }
574 }
575
576 static int unuse_one_window(struct packed_git *current, int keep_fd)
577 {
578 struct packed_git *p, *lru_p = NULL;
579 struct pack_window *lru_w = NULL, *lru_l = NULL;
580
581 if (current)
582 scan_windows(current, &lru_p, &lru_w, &lru_l);
583 for (p = packed_git; p; p = p->next)
584 scan_windows(p, &lru_p, &lru_w, &lru_l);
585 if (lru_p) {
586 munmap(lru_w->base, lru_w->len);
587 pack_mapped -= lru_w->len;
588 if (lru_l)
589 lru_l->next = lru_w->next;
590 else {
591 lru_p->windows = lru_w->next;
592 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
593 close(lru_p->pack_fd);
594 lru_p->pack_fd = -1;
595 }
596 }
597 free(lru_w);
598 pack_open_windows--;
599 return 1;
600 }
601 return 0;
602 }
603
604 void release_pack_memory(size_t need, int fd)
605 {
606 size_t cur = pack_mapped;
607 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
608 ; /* nothing */
609 }
610
611 void unuse_pack(struct pack_window **w_cursor)
612 {
613 struct pack_window *w = *w_cursor;
614 if (w) {
615 w->inuse_cnt--;
616 *w_cursor = NULL;
617 }
618 }
619
620 /*
621 * Do not call this directly as this leaks p->pack_fd on error return;
622 * call open_packed_git() instead.
623 */
624 static int open_packed_git_1(struct packed_git *p)
625 {
626 struct stat st;
627 struct pack_header hdr;
628 unsigned char sha1[20];
629 unsigned char *idx_sha1;
630 long fd_flag;
631
632 if (!p->index_data && open_pack_index(p))
633 return error("packfile %s index unavailable", p->pack_name);
634
635 p->pack_fd = open(p->pack_name, O_RDONLY);
636 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
637 return -1;
638
639 /* If we created the struct before we had the pack we lack size. */
640 if (!p->pack_size) {
641 if (!S_ISREG(st.st_mode))
642 return error("packfile %s not a regular file", p->pack_name);
643 p->pack_size = st.st_size;
644 } else if (p->pack_size != st.st_size)
645 return error("packfile %s size changed", p->pack_name);
646
647 /* We leave these file descriptors open with sliding mmap;
648 * there is no point keeping them open across exec(), though.
649 */
650 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
651 if (fd_flag < 0)
652 return error("cannot determine file descriptor flags");
653 fd_flag |= FD_CLOEXEC;
654 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
655 return error("cannot set FD_CLOEXEC");
656
657 /* Verify we recognize this pack file format. */
658 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
659 return error("file %s is far too short to be a packfile", p->pack_name);
660 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
661 return error("file %s is not a GIT packfile", p->pack_name);
662 if (!pack_version_ok(hdr.hdr_version))
663 return error("packfile %s is version %u and not supported"
664 " (try upgrading GIT to a newer version)",
665 p->pack_name, ntohl(hdr.hdr_version));
666
667 /* Verify the pack matches its index. */
668 if (p->num_objects != ntohl(hdr.hdr_entries))
669 return error("packfile %s claims to have %u objects"
670 " while index indicates %u objects",
671 p->pack_name, ntohl(hdr.hdr_entries),
672 p->num_objects);
673 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
674 return error("end of packfile %s is unavailable", p->pack_name);
675 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
676 return error("packfile %s signature is unavailable", p->pack_name);
677 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
678 if (hashcmp(sha1, idx_sha1))
679 return error("packfile %s does not match index", p->pack_name);
680 return 0;
681 }
682
683 static int open_packed_git(struct packed_git *p)
684 {
685 if (!open_packed_git_1(p))
686 return 0;
687 if (p->pack_fd != -1) {
688 close(p->pack_fd);
689 p->pack_fd = -1;
690 }
691 return -1;
692 }
693
694 static int in_window(struct pack_window *win, off_t offset)
695 {
696 /* We must promise at least 20 bytes (one hash) after the
697 * offset is available from this window, otherwise the offset
698 * is not actually in this window and a different window (which
699 * has that one hash excess) must be used. This is to support
700 * the object header and delta base parsing routines below.
701 */
702 off_t win_off = win->offset;
703 return win_off <= offset
704 && (offset + 20) <= (win_off + win->len);
705 }
706
707 unsigned char* use_pack(struct packed_git *p,
708 struct pack_window **w_cursor,
709 off_t offset,
710 unsigned int *left)
711 {
712 struct pack_window *win = *w_cursor;
713
714 if (p->pack_fd == -1 && open_packed_git(p))
715 die("packfile %s cannot be accessed", p->pack_name);
716
717 /* Since packfiles end in a hash of their content and its
718 * pointless to ask for an offset into the middle of that
719 * hash, and the in_window function above wouldn't match
720 * don't allow an offset too close to the end of the file.
721 */
722 if (offset > (p->pack_size - 20))
723 die("offset beyond end of packfile (truncated pack?)");
724
725 if (!win || !in_window(win, offset)) {
726 if (win)
727 win->inuse_cnt--;
728 for (win = p->windows; win; win = win->next) {
729 if (in_window(win, offset))
730 break;
731 }
732 if (!win) {
733 size_t window_align = packed_git_window_size / 2;
734 off_t len;
735 win = xcalloc(1, sizeof(*win));
736 win->offset = (offset / window_align) * window_align;
737 len = p->pack_size - win->offset;
738 if (len > packed_git_window_size)
739 len = packed_git_window_size;
740 win->len = (size_t)len;
741 pack_mapped += win->len;
742 while (packed_git_limit < pack_mapped
743 && unuse_one_window(p, p->pack_fd))
744 ; /* nothing */
745 win->base = xmmap(NULL, win->len,
746 PROT_READ, MAP_PRIVATE,
747 p->pack_fd, win->offset);
748 if (win->base == MAP_FAILED)
749 die("packfile %s cannot be mapped: %s",
750 p->pack_name,
751 strerror(errno));
752 pack_mmap_calls++;
753 pack_open_windows++;
754 if (pack_mapped > peak_pack_mapped)
755 peak_pack_mapped = pack_mapped;
756 if (pack_open_windows > peak_pack_open_windows)
757 peak_pack_open_windows = pack_open_windows;
758 win->next = p->windows;
759 p->windows = win;
760 }
761 }
762 if (win != *w_cursor) {
763 win->last_used = pack_used_ctr++;
764 win->inuse_cnt++;
765 *w_cursor = win;
766 }
767 offset -= win->offset;
768 if (left)
769 *left = win->len - xsize_t(offset);
770 return win->base + offset;
771 }
772
773 struct packed_git *add_packed_git(const char *path, int path_len, int local)
774 {
775 struct stat st;
776 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
777
778 /*
779 * Make sure a corresponding .pack file exists and that
780 * the index looks sane.
781 */
782 path_len -= strlen(".idx");
783 if (path_len < 1)
784 return NULL;
785 memcpy(p->pack_name, path, path_len);
786 strcpy(p->pack_name + path_len, ".pack");
787 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
788 free(p);
789 return NULL;
790 }
791
792 /* ok, it looks sane as far as we can check without
793 * actually mapping the pack file.
794 */
795 p->index_version = 0;
796 p->index_data = NULL;
797 p->index_size = 0;
798 p->num_objects = 0;
799 p->pack_size = st.st_size;
800 p->next = NULL;
801 p->windows = NULL;
802 p->pack_fd = -1;
803 p->pack_local = local;
804 p->mtime = st.st_mtime;
805 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
806 hashclr(p->sha1);
807 return p;
808 }
809
810 struct packed_git *parse_pack_index(unsigned char *sha1)
811 {
812 char *path = sha1_pack_index_name(sha1);
813 return parse_pack_index_file(sha1, path);
814 }
815
816 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
817 const char *idx_path)
818 {
819 const char *path = sha1_pack_name(sha1);
820 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
821
822 if (check_packed_git_idx(idx_path, p)) {
823 free(p);
824 return NULL;
825 }
826
827 strcpy(p->pack_name, path);
828 p->pack_size = 0;
829 p->next = NULL;
830 p->windows = NULL;
831 p->pack_fd = -1;
832 hashcpy(p->sha1, sha1);
833 return p;
834 }
835
836 void install_packed_git(struct packed_git *pack)
837 {
838 pack->next = packed_git;
839 packed_git = pack;
840 }
841
842 static void prepare_packed_git_one(char *objdir, int local)
843 {
844 /* Ensure that this buffer is large enough so that we can
845 append "/pack/" without clobbering the stack even if
846 strlen(objdir) were PATH_MAX. */
847 char path[PATH_MAX + 1 + 4 + 1 + 1];
848 int len;
849 DIR *dir;
850 struct dirent *de;
851
852 sprintf(path, "%s/pack", objdir);
853 len = strlen(path);
854 dir = opendir(path);
855 if (!dir) {
856 if (errno != ENOENT)
857 error("unable to open object pack directory: %s: %s",
858 path, strerror(errno));
859 return;
860 }
861 path[len++] = '/';
862 while ((de = readdir(dir)) != NULL) {
863 int namelen = strlen(de->d_name);
864 struct packed_git *p;
865
866 if (!has_extension(de->d_name, ".idx"))
867 continue;
868
869 if (len + namelen + 1 > sizeof(path))
870 continue;
871
872 /* Don't reopen a pack we already have. */
873 strcpy(path + len, de->d_name);
874 for (p = packed_git; p; p = p->next) {
875 if (!memcmp(path, p->pack_name, len + namelen - 4))
876 break;
877 }
878 if (p)
879 continue;
880 /* See if it really is a valid .idx file with corresponding
881 * .pack file that we can map.
882 */
883 p = add_packed_git(path, len + namelen, local);
884 if (!p)
885 continue;
886 install_packed_git(p);
887 }
888 closedir(dir);
889 }
890
891 static int sort_pack(const void *a_, const void *b_)
892 {
893 struct packed_git *a = *((struct packed_git **)a_);
894 struct packed_git *b = *((struct packed_git **)b_);
895 int st;
896
897 /*
898 * Local packs tend to contain objects specific to our
899 * variant of the project than remote ones. In addition,
900 * remote ones could be on a network mounted filesystem.
901 * Favor local ones for these reasons.
902 */
903 st = a->pack_local - b->pack_local;
904 if (st)
905 return -st;
906
907 /*
908 * Younger packs tend to contain more recent objects,
909 * and more recent objects tend to get accessed more
910 * often.
911 */
912 if (a->mtime < b->mtime)
913 return 1;
914 else if (a->mtime == b->mtime)
915 return 0;
916 return -1;
917 }
918
919 static void rearrange_packed_git(void)
920 {
921 struct packed_git **ary, *p;
922 int i, n;
923
924 for (n = 0, p = packed_git; p; p = p->next)
925 n++;
926 if (n < 2)
927 return;
928
929 /* prepare an array of packed_git for easier sorting */
930 ary = xcalloc(n, sizeof(struct packed_git *));
931 for (n = 0, p = packed_git; p; p = p->next)
932 ary[n++] = p;
933
934 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
935
936 /* link them back again */
937 for (i = 0; i < n - 1; i++)
938 ary[i]->next = ary[i + 1];
939 ary[n - 1]->next = NULL;
940 packed_git = ary[0];
941
942 free(ary);
943 }
944
945 static int prepare_packed_git_run_once = 0;
946 void prepare_packed_git(void)
947 {
948 struct alternate_object_database *alt;
949
950 if (prepare_packed_git_run_once)
951 return;
952 prepare_packed_git_one(get_object_directory(), 1);
953 prepare_alt_odb();
954 for (alt = alt_odb_list; alt; alt = alt->next) {
955 alt->name[-1] = 0;
956 prepare_packed_git_one(alt->base, 0);
957 alt->name[-1] = '/';
958 }
959 rearrange_packed_git();
960 prepare_packed_git_run_once = 1;
961 }
962
963 void reprepare_packed_git(void)
964 {
965 prepare_packed_git_run_once = 0;
966 prepare_packed_git();
967 }
968
969 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
970 {
971 unsigned char real_sha1[20];
972 hash_sha1_file(map, size, type, real_sha1);
973 return hashcmp(sha1, real_sha1) ? -1 : 0;
974 }
975
976 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
977 {
978 struct stat st;
979 void *map;
980 int fd;
981 char *filename = find_sha1_file(sha1, &st);
982
983 if (!filename) {
984 return NULL;
985 }
986
987 fd = open(filename, O_RDONLY | sha1_file_open_flag);
988 if (fd < 0) {
989 /* See if it works without O_NOATIME */
990 switch (sha1_file_open_flag) {
991 default:
992 fd = open(filename, O_RDONLY);
993 if (fd >= 0)
994 break;
995 /* Fallthrough */
996 case 0:
997 return NULL;
998 }
999
1000 /* If it failed once, it will probably fail again.
1001 * Stop using O_NOATIME
1002 */
1003 sha1_file_open_flag = 0;
1004 }
1005 *size = xsize_t(st.st_size);
1006 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1007 close(fd);
1008 return map;
1009 }
1010
1011 static int legacy_loose_object(unsigned char *map)
1012 {
1013 unsigned int word;
1014
1015 /*
1016 * Is it a zlib-compressed buffer? If so, the first byte
1017 * must be 0x78 (15-bit window size, deflated), and the
1018 * first 16-bit word is evenly divisible by 31
1019 */
1020 word = (map[0] << 8) + map[1];
1021 if (map[0] == 0x78 && !(word % 31))
1022 return 1;
1023 else
1024 return 0;
1025 }
1026
1027 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1028 {
1029 unsigned shift;
1030 unsigned char c;
1031 unsigned long size;
1032 unsigned long used = 0;
1033
1034 c = buf[used++];
1035 *type = (c >> 4) & 7;
1036 size = c & 15;
1037 shift = 4;
1038 while (c & 0x80) {
1039 if (len <= used)
1040 return 0;
1041 if (sizeof(long) * 8 <= shift)
1042 return 0;
1043 c = buf[used++];
1044 size += (c & 0x7f) << shift;
1045 shift += 7;
1046 }
1047 *sizep = size;
1048 return used;
1049 }
1050
1051 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1052 {
1053 unsigned long size, used;
1054 static const char valid_loose_object_type[8] = {
1055 0, /* OBJ_EXT */
1056 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1057 0, /* "delta" and others are invalid in a loose object */
1058 };
1059 enum object_type type;
1060
1061 /* Get the data stream */
1062 memset(stream, 0, sizeof(*stream));
1063 stream->next_in = map;
1064 stream->avail_in = mapsize;
1065 stream->next_out = buffer;
1066 stream->avail_out = bufsiz;
1067
1068 if (legacy_loose_object(map)) {
1069 inflateInit(stream);
1070 return inflate(stream, 0);
1071 }
1072
1073
1074 /*
1075 * There used to be a second loose object header format which
1076 * was meant to mimic the in-pack format, allowing for direct
1077 * copy of the object data. This format turned up not to be
1078 * really worth it and we don't write it any longer. But we
1079 * can still read it.
1080 */
1081 used = unpack_object_header_gently(map, mapsize, &type, &size);
1082 if (!used || !valid_loose_object_type[type])
1083 return -1;
1084 map += used;
1085 mapsize -= used;
1086
1087 /* Set up the stream for the rest.. */
1088 stream->next_in = map;
1089 stream->avail_in = mapsize;
1090 inflateInit(stream);
1091
1092 /* And generate the fake traditional header */
1093 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1094 typename(type), size);
1095 return 0;
1096 }
1097
1098 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1099 {
1100 int bytes = strlen(buffer) + 1;
1101 unsigned char *buf = xmalloc(1+size);
1102 unsigned long n;
1103 int status = Z_OK;
1104
1105 n = stream->total_out - bytes;
1106 if (n > size)
1107 n = size;
1108 memcpy(buf, (char *) buffer + bytes, n);
1109 bytes = n;
1110 if (bytes <= size) {
1111 /*
1112 * The above condition must be (bytes <= size), not
1113 * (bytes < size). In other words, even though we
1114 * expect no more output and set avail_out to zer0,
1115 * the input zlib stream may have bytes that express
1116 * "this concludes the stream", and we *do* want to
1117 * eat that input.
1118 *
1119 * Otherwise we would not be able to test that we
1120 * consumed all the input to reach the expected size;
1121 * we also want to check that zlib tells us that all
1122 * went well with status == Z_STREAM_END at the end.
1123 */
1124 stream->next_out = buf + bytes;
1125 stream->avail_out = size - bytes;
1126 while (status == Z_OK)
1127 status = inflate(stream, Z_FINISH);
1128 }
1129 buf[size] = 0;
1130 if (status == Z_STREAM_END && !stream->avail_in) {
1131 inflateEnd(stream);
1132 return buf;
1133 }
1134
1135 if (status < 0)
1136 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1137 else if (stream->avail_in)
1138 error("garbage at end of loose object '%s'",
1139 sha1_to_hex(sha1));
1140 free(buf);
1141 return NULL;
1142 }
1143
1144 /*
1145 * We used to just use "sscanf()", but that's actually way
1146 * too permissive for what we want to check. So do an anal
1147 * object header parse by hand.
1148 */
1149 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1150 {
1151 char type[10];
1152 int i;
1153 unsigned long size;
1154
1155 /*
1156 * The type can be at most ten bytes (including the
1157 * terminating '\0' that we add), and is followed by
1158 * a space.
1159 */
1160 i = 0;
1161 for (;;) {
1162 char c = *hdr++;
1163 if (c == ' ')
1164 break;
1165 type[i++] = c;
1166 if (i >= sizeof(type))
1167 return -1;
1168 }
1169 type[i] = 0;
1170
1171 /*
1172 * The length must follow immediately, and be in canonical
1173 * decimal format (ie "010" is not valid).
1174 */
1175 size = *hdr++ - '0';
1176 if (size > 9)
1177 return -1;
1178 if (size) {
1179 for (;;) {
1180 unsigned long c = *hdr - '0';
1181 if (c > 9)
1182 break;
1183 hdr++;
1184 size = size * 10 + c;
1185 }
1186 }
1187 *sizep = size;
1188
1189 /*
1190 * The length must be followed by a zero byte
1191 */
1192 return *hdr ? -1 : type_from_string(type);
1193 }
1194
1195 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1196 {
1197 int ret;
1198 z_stream stream;
1199 char hdr[8192];
1200
1201 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1202 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1203 return NULL;
1204
1205 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1206 }
1207
1208 unsigned long get_size_from_delta(struct packed_git *p,
1209 struct pack_window **w_curs,
1210 off_t curpos)
1211 {
1212 const unsigned char *data;
1213 unsigned char delta_head[20], *in;
1214 z_stream stream;
1215 int st;
1216
1217 memset(&stream, 0, sizeof(stream));
1218 stream.next_out = delta_head;
1219 stream.avail_out = sizeof(delta_head);
1220
1221 inflateInit(&stream);
1222 do {
1223 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1224 stream.next_in = in;
1225 st = inflate(&stream, Z_FINISH);
1226 curpos += stream.next_in - in;
1227 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1228 stream.total_out < sizeof(delta_head));
1229 inflateEnd(&stream);
1230 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1231 die("delta data unpack-initial failed");
1232
1233 /* Examine the initial part of the delta to figure out
1234 * the result size.
1235 */
1236 data = delta_head;
1237
1238 /* ignore base size */
1239 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1240
1241 /* Read the result size */
1242 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1243 }
1244
1245 static off_t get_delta_base(struct packed_git *p,
1246 struct pack_window **w_curs,
1247 off_t *curpos,
1248 enum object_type type,
1249 off_t delta_obj_offset)
1250 {
1251 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1252 off_t base_offset;
1253
1254 /* use_pack() assured us we have [base_info, base_info + 20)
1255 * as a range that we can look at without walking off the
1256 * end of the mapped window. Its actually the hash size
1257 * that is assured. An OFS_DELTA longer than the hash size
1258 * is stupid, as then a REF_DELTA would be smaller to store.
1259 */
1260 if (type == OBJ_OFS_DELTA) {
1261 unsigned used = 0;
1262 unsigned char c = base_info[used++];
1263 base_offset = c & 127;
1264 while (c & 128) {
1265 base_offset += 1;
1266 if (!base_offset || MSB(base_offset, 7))
1267 die("offset value overflow for delta base object");
1268 c = base_info[used++];
1269 base_offset = (base_offset << 7) + (c & 127);
1270 }
1271 base_offset = delta_obj_offset - base_offset;
1272 if (base_offset >= delta_obj_offset)
1273 die("delta base offset out of bound");
1274 *curpos += used;
1275 } else if (type == OBJ_REF_DELTA) {
1276 /* The base entry _must_ be in the same pack */
1277 base_offset = find_pack_entry_one(base_info, p);
1278 if (!base_offset)
1279 die("failed to find delta-pack base object %s",
1280 sha1_to_hex(base_info));
1281 *curpos += 20;
1282 } else
1283 die("I am totally screwed");
1284 return base_offset;
1285 }
1286
1287 /* forward declaration for a mutually recursive function */
1288 static int packed_object_info(struct packed_git *p, off_t offset,
1289 unsigned long *sizep);
1290
1291 static int packed_delta_info(struct packed_git *p,
1292 struct pack_window **w_curs,
1293 off_t curpos,
1294 enum object_type type,
1295 off_t obj_offset,
1296 unsigned long *sizep)
1297 {
1298 off_t base_offset;
1299
1300 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1301 type = packed_object_info(p, base_offset, NULL);
1302
1303 /* We choose to only get the type of the base object and
1304 * ignore potentially corrupt pack file that expects the delta
1305 * based on a base with a wrong size. This saves tons of
1306 * inflate() calls.
1307 */
1308 if (sizep)
1309 *sizep = get_size_from_delta(p, w_curs, curpos);
1310
1311 return type;
1312 }
1313
1314 static int unpack_object_header(struct packed_git *p,
1315 struct pack_window **w_curs,
1316 off_t *curpos,
1317 unsigned long *sizep)
1318 {
1319 unsigned char *base;
1320 unsigned int left;
1321 unsigned long used;
1322 enum object_type type;
1323
1324 /* use_pack() assures us we have [base, base + 20) available
1325 * as a range that we can look at at. (Its actually the hash
1326 * size that is assured.) With our object header encoding
1327 * the maximum deflated object size is 2^137, which is just
1328 * insane, so we know won't exceed what we have been given.
1329 */
1330 base = use_pack(p, w_curs, *curpos, &left);
1331 used = unpack_object_header_gently(base, left, &type, sizep);
1332 if (!used)
1333 die("object offset outside of pack file");
1334 *curpos += used;
1335
1336 return type;
1337 }
1338
1339 const char *packed_object_info_detail(struct packed_git *p,
1340 off_t obj_offset,
1341 unsigned long *size,
1342 unsigned long *store_size,
1343 unsigned int *delta_chain_length,
1344 unsigned char *base_sha1)
1345 {
1346 struct pack_window *w_curs = NULL;
1347 off_t curpos;
1348 unsigned long dummy;
1349 unsigned char *next_sha1;
1350 enum object_type type;
1351
1352 *delta_chain_length = 0;
1353 curpos = obj_offset;
1354 type = unpack_object_header(p, &w_curs, &curpos, size);
1355
1356 for (;;) {
1357 switch (type) {
1358 default:
1359 die("pack %s contains unknown object type %d",
1360 p->pack_name, type);
1361 case OBJ_COMMIT:
1362 case OBJ_TREE:
1363 case OBJ_BLOB:
1364 case OBJ_TAG:
1365 *store_size = 0; /* notyet */
1366 unuse_pack(&w_curs);
1367 return typename(type);
1368 case OBJ_OFS_DELTA:
1369 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1370 if (*delta_chain_length == 0) {
1371 /* TODO: find base_sha1 as pointed by curpos */
1372 hashclr(base_sha1);
1373 }
1374 break;
1375 case OBJ_REF_DELTA:
1376 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1377 if (*delta_chain_length == 0)
1378 hashcpy(base_sha1, next_sha1);
1379 obj_offset = find_pack_entry_one(next_sha1, p);
1380 break;
1381 }
1382 (*delta_chain_length)++;
1383 curpos = obj_offset;
1384 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1385 }
1386 }
1387
1388 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1389 unsigned long *sizep)
1390 {
1391 struct pack_window *w_curs = NULL;
1392 unsigned long size;
1393 off_t curpos = obj_offset;
1394 enum object_type type;
1395
1396 type = unpack_object_header(p, &w_curs, &curpos, &size);
1397
1398 switch (type) {
1399 case OBJ_OFS_DELTA:
1400 case OBJ_REF_DELTA:
1401 type = packed_delta_info(p, &w_curs, curpos,
1402 type, obj_offset, sizep);
1403 break;
1404 case OBJ_COMMIT:
1405 case OBJ_TREE:
1406 case OBJ_BLOB:
1407 case OBJ_TAG:
1408 if (sizep)
1409 *sizep = size;
1410 break;
1411 default:
1412 die("pack %s contains unknown object type %d",
1413 p->pack_name, type);
1414 }
1415 unuse_pack(&w_curs);
1416 return type;
1417 }
1418
1419 static void *unpack_compressed_entry(struct packed_git *p,
1420 struct pack_window **w_curs,
1421 off_t curpos,
1422 unsigned long size)
1423 {
1424 int st;
1425 z_stream stream;
1426 unsigned char *buffer, *in;
1427
1428 buffer = xmalloc(size + 1);
1429 buffer[size] = 0;
1430 memset(&stream, 0, sizeof(stream));
1431 stream.next_out = buffer;
1432 stream.avail_out = size;
1433
1434 inflateInit(&stream);
1435 do {
1436 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1437 stream.next_in = in;
1438 st = inflate(&stream, Z_FINISH);
1439 curpos += stream.next_in - in;
1440 } while (st == Z_OK || st == Z_BUF_ERROR);
1441 inflateEnd(&stream);
1442 if ((st != Z_STREAM_END) || stream.total_out != size) {
1443 free(buffer);
1444 return NULL;
1445 }
1446
1447 return buffer;
1448 }
1449
1450 #define MAX_DELTA_CACHE (256)
1451
1452 static size_t delta_base_cached;
1453
1454 static struct delta_base_cache_lru_list {
1455 struct delta_base_cache_lru_list *prev;
1456 struct delta_base_cache_lru_list *next;
1457 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1458
1459 static struct delta_base_cache_entry {
1460 struct delta_base_cache_lru_list lru;
1461 void *data;
1462 struct packed_git *p;
1463 off_t base_offset;
1464 unsigned long size;
1465 enum object_type type;
1466 } delta_base_cache[MAX_DELTA_CACHE];
1467
1468 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1469 {
1470 unsigned long hash;
1471
1472 hash = (unsigned long)p + (unsigned long)base_offset;
1473 hash += (hash >> 8) + (hash >> 16);
1474 return hash % MAX_DELTA_CACHE;
1475 }
1476
1477 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1478 unsigned long *base_size, enum object_type *type, int keep_cache)
1479 {
1480 void *ret;
1481 unsigned long hash = pack_entry_hash(p, base_offset);
1482 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1483
1484 ret = ent->data;
1485 if (ret && ent->p == p && ent->base_offset == base_offset)
1486 goto found_cache_entry;
1487 return unpack_entry(p, base_offset, type, base_size);
1488
1489 found_cache_entry:
1490 if (!keep_cache) {
1491 ent->data = NULL;
1492 ent->lru.next->prev = ent->lru.prev;
1493 ent->lru.prev->next = ent->lru.next;
1494 delta_base_cached -= ent->size;
1495 }
1496 else {
1497 ret = xmalloc(ent->size + 1);
1498 memcpy(ret, ent->data, ent->size);
1499 ((char *)ret)[ent->size] = 0;
1500 }
1501 *type = ent->type;
1502 *base_size = ent->size;
1503 return ret;
1504 }
1505
1506 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1507 {
1508 if (ent->data) {
1509 free(ent->data);
1510 ent->data = NULL;
1511 ent->lru.next->prev = ent->lru.prev;
1512 ent->lru.prev->next = ent->lru.next;
1513 delta_base_cached -= ent->size;
1514 }
1515 }
1516
1517 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1518 void *base, unsigned long base_size, enum object_type type)
1519 {
1520 unsigned long hash = pack_entry_hash(p, base_offset);
1521 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1522 struct delta_base_cache_lru_list *lru;
1523
1524 release_delta_base_cache(ent);
1525 delta_base_cached += base_size;
1526
1527 for (lru = delta_base_cache_lru.next;
1528 delta_base_cached > delta_base_cache_limit
1529 && lru != &delta_base_cache_lru;
1530 lru = lru->next) {
1531 struct delta_base_cache_entry *f = (void *)lru;
1532 if (f->type == OBJ_BLOB)
1533 release_delta_base_cache(f);
1534 }
1535 for (lru = delta_base_cache_lru.next;
1536 delta_base_cached > delta_base_cache_limit
1537 && lru != &delta_base_cache_lru;
1538 lru = lru->next) {
1539 struct delta_base_cache_entry *f = (void *)lru;
1540 release_delta_base_cache(f);
1541 }
1542
1543 ent->p = p;
1544 ent->base_offset = base_offset;
1545 ent->type = type;
1546 ent->data = base;
1547 ent->size = base_size;
1548 ent->lru.next = &delta_base_cache_lru;
1549 ent->lru.prev = delta_base_cache_lru.prev;
1550 delta_base_cache_lru.prev->next = &ent->lru;
1551 delta_base_cache_lru.prev = &ent->lru;
1552 }
1553
1554 static void *unpack_delta_entry(struct packed_git *p,
1555 struct pack_window **w_curs,
1556 off_t curpos,
1557 unsigned long delta_size,
1558 off_t obj_offset,
1559 enum object_type *type,
1560 unsigned long *sizep)
1561 {
1562 void *delta_data, *result, *base;
1563 unsigned long base_size;
1564 off_t base_offset;
1565
1566 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1567 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1568 if (!base)
1569 die("failed to read delta base object"
1570 " at %"PRIuMAX" from %s",
1571 (uintmax_t)base_offset, p->pack_name);
1572
1573 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1574 if (!delta_data)
1575 die("failed to unpack compressed delta"
1576 " at %"PRIuMAX" from %s",
1577 (uintmax_t)curpos, p->pack_name);
1578 result = patch_delta(base, base_size,
1579 delta_data, delta_size,
1580 sizep);
1581 if (!result)
1582 die("failed to apply delta");
1583 free(delta_data);
1584 add_delta_base_cache(p, base_offset, base, base_size, *type);
1585 return result;
1586 }
1587
1588 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1589 enum object_type *type, unsigned long *sizep)
1590 {
1591 struct pack_window *w_curs = NULL;
1592 off_t curpos = obj_offset;
1593 void *data;
1594
1595 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1596 switch (*type) {
1597 case OBJ_OFS_DELTA:
1598 case OBJ_REF_DELTA:
1599 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1600 obj_offset, type, sizep);
1601 break;
1602 case OBJ_COMMIT:
1603 case OBJ_TREE:
1604 case OBJ_BLOB:
1605 case OBJ_TAG:
1606 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1607 break;
1608 default:
1609 die("unknown object type %i in %s", *type, p->pack_name);
1610 }
1611 unuse_pack(&w_curs);
1612 return data;
1613 }
1614
1615 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1616 uint32_t n)
1617 {
1618 const unsigned char *index = p->index_data;
1619 if (!index) {
1620 if (open_pack_index(p))
1621 return NULL;
1622 index = p->index_data;
1623 }
1624 if (n >= p->num_objects)
1625 return NULL;
1626 index += 4 * 256;
1627 if (p->index_version == 1) {
1628 return index + 24 * n + 4;
1629 } else {
1630 index += 8;
1631 return index + 20 * n;
1632 }
1633 }
1634
1635 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1636 {
1637 const unsigned char *index = p->index_data;
1638 index += 4 * 256;
1639 if (p->index_version == 1) {
1640 return ntohl(*((uint32_t *)(index + 24 * n)));
1641 } else {
1642 uint32_t off;
1643 index += 8 + p->num_objects * (20 + 4);
1644 off = ntohl(*((uint32_t *)(index + 4 * n)));
1645 if (!(off & 0x80000000))
1646 return off;
1647 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1648 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1649 ntohl(*((uint32_t *)(index + 4)));
1650 }
1651 }
1652
1653 off_t find_pack_entry_one(const unsigned char *sha1,
1654 struct packed_git *p)
1655 {
1656 const uint32_t *level1_ofs = p->index_data;
1657 const unsigned char *index = p->index_data;
1658 unsigned hi, lo;
1659
1660 if (!index) {
1661 if (open_pack_index(p))
1662 return 0;
1663 level1_ofs = p->index_data;
1664 index = p->index_data;
1665 }
1666 if (p->index_version > 1) {
1667 level1_ofs += 2;
1668 index += 8;
1669 }
1670 index += 4 * 256;
1671 hi = ntohl(level1_ofs[*sha1]);
1672 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1673
1674 do {
1675 unsigned mi = (lo + hi) / 2;
1676 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1677 int cmp = hashcmp(index + x, sha1);
1678 if (!cmp)
1679 return nth_packed_object_offset(p, mi);
1680 if (cmp > 0)
1681 hi = mi;
1682 else
1683 lo = mi+1;
1684 } while (lo < hi);
1685 return 0;
1686 }
1687
1688 static int matches_pack_name(struct packed_git *p, const char *ig)
1689 {
1690 const char *last_c, *c;
1691
1692 if (!strcmp(p->pack_name, ig))
1693 return 0;
1694
1695 for (c = p->pack_name, last_c = c; *c;)
1696 if (*c == '/')
1697 last_c = ++c;
1698 else
1699 ++c;
1700 if (!strcmp(last_c, ig))
1701 return 0;
1702
1703 return 1;
1704 }
1705
1706 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1707 {
1708 static struct packed_git *last_found = (void *)1;
1709 struct packed_git *p;
1710 off_t offset;
1711
1712 prepare_packed_git();
1713 if (!packed_git)
1714 return 0;
1715 p = (last_found == (void *)1) ? packed_git : last_found;
1716
1717 do {
1718 if (ignore_packed) {
1719 const char **ig;
1720 for (ig = ignore_packed; *ig; ig++)
1721 if (!matches_pack_name(p, *ig))
1722 break;
1723 if (*ig)
1724 goto next;
1725 }
1726
1727 offset = find_pack_entry_one(sha1, p);
1728 if (offset) {
1729 /*
1730 * We are about to tell the caller where they can
1731 * locate the requested object. We better make
1732 * sure the packfile is still here and can be
1733 * accessed before supplying that answer, as
1734 * it may have been deleted since the index
1735 * was loaded!
1736 */
1737 if (p->pack_fd == -1 && open_packed_git(p)) {
1738 error("packfile %s cannot be accessed", p->pack_name);
1739 goto next;
1740 }
1741 e->offset = offset;
1742 e->p = p;
1743 hashcpy(e->sha1, sha1);
1744 last_found = p;
1745 return 1;
1746 }
1747
1748 next:
1749 if (p == last_found)
1750 p = packed_git;
1751 else
1752 p = p->next;
1753 if (p == last_found)
1754 p = p->next;
1755 } while (p);
1756 return 0;
1757 }
1758
1759 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1760 struct packed_git *packs)
1761 {
1762 struct packed_git *p;
1763
1764 for (p = packs; p; p = p->next) {
1765 if (find_pack_entry_one(sha1, p))
1766 return p;
1767 }
1768 return NULL;
1769
1770 }
1771
1772 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1773 {
1774 int status;
1775 unsigned long mapsize, size;
1776 void *map;
1777 z_stream stream;
1778 char hdr[32];
1779
1780 map = map_sha1_file(sha1, &mapsize);
1781 if (!map)
1782 return error("unable to find %s", sha1_to_hex(sha1));
1783 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1784 status = error("unable to unpack %s header",
1785 sha1_to_hex(sha1));
1786 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1787 status = error("unable to parse %s header", sha1_to_hex(sha1));
1788 else if (sizep)
1789 *sizep = size;
1790 inflateEnd(&stream);
1791 munmap(map, mapsize);
1792 return status;
1793 }
1794
1795 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1796 {
1797 struct pack_entry e;
1798
1799 if (!find_pack_entry(sha1, &e, NULL)) {
1800 reprepare_packed_git();
1801 if (!find_pack_entry(sha1, &e, NULL))
1802 return sha1_loose_object_info(sha1, sizep);
1803 }
1804 return packed_object_info(e.p, e.offset, sizep);
1805 }
1806
1807 static void *read_packed_sha1(const unsigned char *sha1,
1808 enum object_type *type, unsigned long *size)
1809 {
1810 struct pack_entry e;
1811
1812 if (!find_pack_entry(sha1, &e, NULL))
1813 return NULL;
1814 else
1815 return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1816 }
1817
1818 /*
1819 * This is meant to hold a *small* number of objects that you would
1820 * want read_sha1_file() to be able to return, but yet you do not want
1821 * to write them into the object store (e.g. a browse-only
1822 * application).
1823 */
1824 static struct cached_object {
1825 unsigned char sha1[20];
1826 enum object_type type;
1827 void *buf;
1828 unsigned long size;
1829 } *cached_objects;
1830 static int cached_object_nr, cached_object_alloc;
1831
1832 static struct cached_object *find_cached_object(const unsigned char *sha1)
1833 {
1834 int i;
1835 struct cached_object *co = cached_objects;
1836
1837 for (i = 0; i < cached_object_nr; i++, co++) {
1838 if (!hashcmp(co->sha1, sha1))
1839 return co;
1840 }
1841 return NULL;
1842 }
1843
1844 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1845 unsigned char *sha1)
1846 {
1847 struct cached_object *co;
1848
1849 hash_sha1_file(buf, len, typename(type), sha1);
1850 if (has_sha1_file(sha1) || find_cached_object(sha1))
1851 return 0;
1852 if (cached_object_alloc <= cached_object_nr) {
1853 cached_object_alloc = alloc_nr(cached_object_alloc);
1854 cached_objects = xrealloc(cached_objects,
1855 sizeof(*cached_objects) *
1856 cached_object_alloc);
1857 }
1858 co = &cached_objects[cached_object_nr++];
1859 co->size = len;
1860 co->type = type;
1861 co->buf = xmalloc(len);
1862 memcpy(co->buf, buf, len);
1863 hashcpy(co->sha1, sha1);
1864 return 0;
1865 }
1866
1867 void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1868 unsigned long *size)
1869 {
1870 unsigned long mapsize;
1871 void *map, *buf;
1872 struct cached_object *co;
1873
1874 co = find_cached_object(sha1);
1875 if (co) {
1876 buf = xmalloc(co->size + 1);
1877 memcpy(buf, co->buf, co->size);
1878 ((char*)buf)[co->size] = 0;
1879 *type = co->type;
1880 *size = co->size;
1881 return buf;
1882 }
1883
1884 buf = read_packed_sha1(sha1, type, size);
1885 if (buf)
1886 return buf;
1887 map = map_sha1_file(sha1, &mapsize);
1888 if (map) {
1889 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1890 munmap(map, mapsize);
1891 return buf;
1892 }
1893 reprepare_packed_git();
1894 return read_packed_sha1(sha1, type, size);
1895 }
1896
1897 void *read_object_with_reference(const unsigned char *sha1,
1898 const char *required_type_name,
1899 unsigned long *size,
1900 unsigned char *actual_sha1_return)
1901 {
1902 enum object_type type, required_type;
1903 void *buffer;
1904 unsigned long isize;
1905 unsigned char actual_sha1[20];
1906
1907 required_type = type_from_string(required_type_name);
1908 hashcpy(actual_sha1, sha1);
1909 while (1) {
1910 int ref_length = -1;
1911 const char *ref_type = NULL;
1912
1913 buffer = read_sha1_file(actual_sha1, &type, &isize);
1914 if (!buffer)
1915 return NULL;
1916 if (type == required_type) {
1917 *size = isize;
1918 if (actual_sha1_return)
1919 hashcpy(actual_sha1_return, actual_sha1);
1920 return buffer;
1921 }
1922 /* Handle references */
1923 else if (type == OBJ_COMMIT)
1924 ref_type = "tree ";
1925 else if (type == OBJ_TAG)
1926 ref_type = "object ";
1927 else {
1928 free(buffer);
1929 return NULL;
1930 }
1931 ref_length = strlen(ref_type);
1932
1933 if (memcmp(buffer, ref_type, ref_length) ||
1934 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1935 free(buffer);
1936 return NULL;
1937 }
1938 free(buffer);
1939 /* Now we have the ID of the referred-to object in
1940 * actual_sha1. Check again. */
1941 }
1942 }
1943
1944 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1945 const char *type, unsigned char *sha1,
1946 char *hdr, int *hdrlen)
1947 {
1948 SHA_CTX c;
1949
1950 /* Generate the header */
1951 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1952
1953 /* Sha1.. */
1954 SHA1_Init(&c);
1955 SHA1_Update(&c, hdr, *hdrlen);
1956 SHA1_Update(&c, buf, len);
1957 SHA1_Final(sha1, &c);
1958 }
1959
1960 /*
1961 * Link the tempfile to the final place, possibly creating the
1962 * last directory level as you do so.
1963 *
1964 * Returns the errno on failure, 0 on success.
1965 */
1966 static int link_temp_to_file(const char *tmpfile, const char *filename)
1967 {
1968 int ret;
1969 char *dir;
1970
1971 if (!link(tmpfile, filename))
1972 return 0;
1973
1974 /*
1975 * Try to mkdir the last path component if that failed.
1976 *
1977 * Re-try the "link()" regardless of whether the mkdir
1978 * succeeds, since a race might mean that somebody
1979 * else succeeded.
1980 */
1981 ret = errno;
1982 dir = strrchr(filename, '/');
1983 if (dir) {
1984 *dir = 0;
1985 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1986 *dir = '/';
1987 return -2;
1988 }
1989 *dir = '/';
1990 if (!link(tmpfile, filename))
1991 return 0;
1992 ret = errno;
1993 }
1994 return ret;
1995 }
1996
1997 /*
1998 * Move the just written object into its final resting place
1999 */
2000 int move_temp_to_file(const char *tmpfile, const char *filename)
2001 {
2002 int ret = link_temp_to_file(tmpfile, filename);
2003
2004 /*
2005 * Coda hack - coda doesn't like cross-directory links,
2006 * so we fall back to a rename, which will mean that it
2007 * won't be able to check collisions, but that's not a
2008 * big deal.
2009 *
2010 * The same holds for FAT formatted media.
2011 *
2012 * When this succeeds, we just return 0. We have nothing
2013 * left to unlink.
2014 */
2015 if (ret && ret != EEXIST) {
2016 if (!rename(tmpfile, filename))
2017 return 0;
2018 ret = errno;
2019 }
2020 unlink(tmpfile);
2021 if (ret) {
2022 if (ret != EEXIST) {
2023 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2024 }
2025 /* FIXME!!! Collision check here ? */
2026 }
2027
2028 return 0;
2029 }
2030
2031 static int write_buffer(int fd, const void *buf, size_t len)
2032 {
2033 if (write_in_full(fd, buf, len) < 0)
2034 return error("file write error (%s)", strerror(errno));
2035 return 0;
2036 }
2037
2038 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2039 unsigned char *sha1)
2040 {
2041 char hdr[32];
2042 int hdrlen;
2043 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2044 return 0;
2045 }
2046
2047 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2048 {
2049 int size, ret;
2050 unsigned char *compressed;
2051 z_stream stream;
2052 unsigned char sha1[20];
2053 char *filename;
2054 static char tmpfile[PATH_MAX];
2055 char hdr[32];
2056 int fd, hdrlen;
2057
2058 /* Normally if we have it in the pack then we do not bother writing
2059 * it out into .git/objects/??/?{38} file.
2060 */
2061 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2062 filename = sha1_file_name(sha1);
2063 if (returnsha1)
2064 hashcpy(returnsha1, sha1);
2065 if (has_sha1_file(sha1))
2066 return 0;
2067 fd = open(filename, O_RDONLY);
2068 if (fd >= 0) {
2069 /*
2070 * FIXME!!! We might do collision checking here, but we'd
2071 * need to uncompress the old file and check it. Later.
2072 */
2073 close(fd);
2074 return 0;
2075 }
2076
2077 if (errno != ENOENT) {
2078 return error("sha1 file %s: %s\n", filename, strerror(errno));
2079 }
2080
2081 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2082
2083 fd = mkstemp(tmpfile);
2084 if (fd < 0) {
2085 if (errno == EPERM)
2086 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2087 else
2088 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2089 }
2090
2091 /* Set it up */
2092 memset(&stream, 0, sizeof(stream));
2093 deflateInit(&stream, zlib_compression_level);
2094 size = 8 + deflateBound(&stream, len+hdrlen);
2095 compressed = xmalloc(size);
2096
2097 /* Compress it */
2098 stream.next_out = compressed;
2099 stream.avail_out = size;
2100
2101 /* First header.. */
2102 stream.next_in = (unsigned char *)hdr;
2103 stream.avail_in = hdrlen;
2104 while (deflate(&stream, 0) == Z_OK)
2105 /* nothing */;
2106
2107 /* Then the data itself.. */
2108 stream.next_in = buf;
2109 stream.avail_in = len;
2110 ret = deflate(&stream, Z_FINISH);
2111 if (ret != Z_STREAM_END)
2112 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2113
2114 ret = deflateEnd(&stream);
2115 if (ret != Z_OK)
2116 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2117
2118 size = stream.total_out;
2119
2120 if (write_buffer(fd, compressed, size) < 0)
2121 die("unable to write sha1 file");
2122 fchmod(fd, 0444);
2123 if (close(fd))
2124 die("unable to write sha1 file");
2125 free(compressed);
2126
2127 return move_temp_to_file(tmpfile, filename);
2128 }
2129
2130 /*
2131 * We need to unpack and recompress the object for writing
2132 * it out to a different file.
2133 */
2134 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2135 {
2136 size_t size;
2137 z_stream stream;
2138 unsigned char *unpacked;
2139 unsigned long len;
2140 enum object_type type;
2141 char hdr[32];
2142 int hdrlen;
2143 void *buf;
2144
2145 /* need to unpack and recompress it by itself */
2146 unpacked = read_packed_sha1(sha1, &type, &len);
2147 if (!unpacked)
2148 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2149
2150 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2151
2152 /* Set it up */
2153 memset(&stream, 0, sizeof(stream));
2154 deflateInit(&stream, zlib_compression_level);
2155 size = deflateBound(&stream, len + hdrlen);
2156 buf = xmalloc(size);
2157
2158 /* Compress it */
2159 stream.next_out = buf;
2160 stream.avail_out = size;
2161
2162 /* First header.. */
2163 stream.next_in = (void *)hdr;
2164 stream.avail_in = hdrlen;
2165 while (deflate(&stream, 0) == Z_OK)
2166 /* nothing */;
2167
2168 /* Then the data itself.. */
2169 stream.next_in = unpacked;
2170 stream.avail_in = len;
2171 while (deflate(&stream, Z_FINISH) == Z_OK)
2172 /* nothing */;
2173 deflateEnd(&stream);
2174 free(unpacked);
2175
2176 *objsize = stream.total_out;
2177 return buf;
2178 }
2179
2180 int write_sha1_to_fd(int fd, const unsigned char *sha1)
2181 {
2182 int retval;
2183 unsigned long objsize;
2184 void *buf = map_sha1_file(sha1, &objsize);
2185
2186 if (buf) {
2187 retval = write_buffer(fd, buf, objsize);
2188 munmap(buf, objsize);
2189 return retval;
2190 }
2191
2192 buf = repack_object(sha1, &objsize);
2193 retval = write_buffer(fd, buf, objsize);
2194 free(buf);
2195 return retval;
2196 }
2197
2198 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2199 size_t bufsize, size_t *bufposn)
2200 {
2201 char tmpfile[PATH_MAX];
2202 int local;
2203 z_stream stream;
2204 unsigned char real_sha1[20];
2205 unsigned char discard[4096];
2206 int ret;
2207 SHA_CTX c;
2208
2209 snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2210
2211 local = mkstemp(tmpfile);
2212 if (local < 0) {
2213 if (errno == EPERM)
2214 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2215 else
2216 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2217 }
2218
2219 memset(&stream, 0, sizeof(stream));
2220
2221 inflateInit(&stream);
2222
2223 SHA1_Init(&c);
2224
2225 do {
2226 ssize_t size;
2227 if (*bufposn) {
2228 stream.avail_in = *bufposn;
2229 stream.next_in = (unsigned char *) buffer;
2230 do {
2231 stream.next_out = discard;
2232 stream.avail_out = sizeof(discard);
2233 ret = inflate(&stream, Z_SYNC_FLUSH);
2234 SHA1_Update(&c, discard, sizeof(discard) -
2235 stream.avail_out);
2236 } while (stream.avail_in && ret == Z_OK);
2237 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2238 die("unable to write sha1 file");
2239 memmove(buffer, buffer + *bufposn - stream.avail_in,
2240 stream.avail_in);
2241 *bufposn = stream.avail_in;
2242 if (ret != Z_OK)
2243 break;
2244 }
2245 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2246 if (size <= 0) {
2247 close(local);
2248 unlink(tmpfile);
2249 if (!size)
2250 return error("Connection closed?");
2251 perror("Reading from connection");
2252 return -1;
2253 }
2254 *bufposn += size;
2255 } while (1);
2256 inflateEnd(&stream);
2257
2258 fchmod(local, 0444);
2259 if (close(local) != 0)
2260 die("unable to write sha1 file");
2261 SHA1_Final(real_sha1, &c);
2262 if (ret != Z_STREAM_END) {
2263 unlink(tmpfile);
2264 return error("File %s corrupted", sha1_to_hex(sha1));
2265 }
2266 if (hashcmp(sha1, real_sha1)) {
2267 unlink(tmpfile);
2268 return error("File %s has bad hash", sha1_to_hex(sha1));
2269 }
2270
2271 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2272 }
2273
2274 int has_pack_index(const unsigned char *sha1)
2275 {
2276 struct stat st;
2277 if (stat(sha1_pack_index_name(sha1), &st))
2278 return 0;
2279 return 1;
2280 }
2281
2282 int has_pack_file(const unsigned char *sha1)
2283 {
2284 struct stat st;
2285 if (stat(sha1_pack_name(sha1), &st))
2286 return 0;
2287 return 1;
2288 }
2289
2290 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2291 {
2292 struct pack_entry e;
2293 return find_pack_entry(sha1, &e, ignore_packed);
2294 }
2295
2296 int has_sha1_file(const unsigned char *sha1)
2297 {
2298 struct stat st;
2299 struct pack_entry e;
2300
2301 if (find_pack_entry(sha1, &e, NULL))
2302 return 1;
2303 return find_sha1_file(sha1, &st) ? 1 : 0;
2304 }
2305
2306 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2307 {
2308 struct strbuf buf;
2309 int ret;
2310
2311 strbuf_init(&buf, 0);
2312 if (strbuf_read(&buf, fd, 4096) < 0) {
2313 strbuf_release(&buf);
2314 return -1;
2315 }
2316
2317 if (!type)
2318 type = blob_type;
2319 if (write_object)
2320 ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2321 else
2322 ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2323 strbuf_release(&buf);
2324
2325 return ret;
2326 }
2327
2328 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2329 enum object_type type, const char *path)
2330 {
2331 size_t size = xsize_t(st->st_size);
2332 void *buf = NULL;
2333 int ret, re_allocated = 0;
2334
2335 if (size)
2336 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2337 close(fd);
2338
2339 if (!type)
2340 type = OBJ_BLOB;
2341
2342 /*
2343 * Convert blobs to git internal format
2344 */
2345 if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2346 struct strbuf nbuf;
2347 strbuf_init(&nbuf, 0);
2348 if (convert_to_git(path, buf, size, &nbuf)) {
2349 munmap(buf, size);
2350 size = nbuf.len;
2351 buf = nbuf.buf;
2352 re_allocated = 1;
2353 }
2354 }
2355
2356 if (write_object)
2357 ret = write_sha1_file(buf, size, typename(type), sha1);
2358 else
2359 ret = hash_sha1_file(buf, size, typename(type), sha1);
2360 if (re_allocated) {
2361 free(buf);
2362 return ret;
2363 }
2364 if (size)
2365 munmap(buf, size);
2366 return ret;
2367 }
2368
2369 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2370 {
2371 int fd;
2372 char *target;
2373 size_t len;
2374
2375 switch (st->st_mode & S_IFMT) {
2376 case S_IFREG:
2377 fd = open(path, O_RDONLY);
2378 if (fd < 0)
2379 return error("open(\"%s\"): %s", path,
2380 strerror(errno));
2381 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2382 return error("%s: failed to insert into database",
2383 path);
2384 break;
2385 case S_IFLNK:
2386 len = xsize_t(st->st_size);
2387 target = xmalloc(len + 1);
2388 if (readlink(path, target, len + 1) != st->st_size) {
2389 char *errstr = strerror(errno);
2390 free(target);
2391 return error("readlink(\"%s\"): %s", path,
2392 errstr);
2393 }
2394 if (!write_object)
2395 hash_sha1_file(target, len, blob_type, sha1);
2396 else if (write_sha1_file(target, len, blob_type, sha1))
2397 return error("%s: failed to insert into database",
2398 path);
2399 free(target);
2400 break;
2401 case S_IFDIR:
2402 return resolve_gitlink_ref(path, "HEAD", sha1);
2403 default:
2404 return error("%s: unsupported file type", path);
2405 }
2406 return 0;
2407 }
2408
2409 int read_pack_header(int fd, struct pack_header *header)
2410 {
2411 char *c = (char*)header;
2412 ssize_t remaining = sizeof(struct pack_header);
2413 do {
2414 ssize_t r = xread(fd, c, remaining);
2415 if (r <= 0)
2416 /* "eof before pack header was fully read" */
2417 return PH_ERROR_EOF;
2418 remaining -= r;
2419 c += r;
2420 } while (remaining > 0);
2421 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2422 /* "protocol error (pack signature mismatch detected)" */
2423 return PH_ERROR_PACK_SIGNATURE;
2424 if (!pack_version_ok(header->hdr_version))
2425 /* "protocol error (pack version unsupported)" */
2426 return PH_ERROR_PROTOCOL;
2427 return 0;
2428 }