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"
30static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
31#else
32#define SZ_FMT "zu"
33static size_t sz_fmt(size_t s) { return s; }
34#endif
35
36const unsigned char null_sha1[20];
37
38const signed char hexval_table[256] = {
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
45 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
46 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
47 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
51 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
71};
72
73int get_sha1_hex(const char *hex, unsigned char *sha1)
74{
75 int i;
76 for (i = 0; i < 20; i++) {
77 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
78 if (val & ~0xff)
79 return -1;
80 *sha1++ = val;
81 hex += 2;
82 }
83 return 0;
84}
85
86static inline int offset_1st_component(const char *path)
87{
88 if (has_dos_drive_prefix(path))
89 return 2 + (path[2] == '/');
90 return *path == '/';
91}
92
93int safe_create_leading_directories(char *path)
94{
95 char *pos = path + offset_1st_component(path);
96 struct stat st;
97
98 while (pos) {
99 pos = strchr(pos, '/');
100 if (!pos)
101 break;
102 while (*++pos == '/')
103 ;
104 if (!*pos)
105 break;
106 *--pos = '\0';
107 if (!stat(path, &st)) {
108 /* path exists */
109 if (!S_ISDIR(st.st_mode)) {
110 *pos = '/';
111 return -3;
112 }
113 }
114 else if (mkdir(path, 0777)) {
115 *pos = '/';
116 return -1;
117 }
118 else if (adjust_shared_perm(path)) {
119 *pos = '/';
120 return -2;
121 }
122 *pos++ = '/';
123 }
124 return 0;
125}
126
127int safe_create_leading_directories_const(const char *path)
128{
129 /* path points to cache entries, so xstrdup before messing with it */
130 char *buf = xstrdup(path);
131 int result = safe_create_leading_directories(buf);
132 free(buf);
133 return result;
134}
135
136char *sha1_to_hex(const unsigned char *sha1)
137{
138 static int bufno;
139 static char hexbuffer[4][50];
140 static const char hex[] = "0123456789abcdef";
141 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
142 int i;
143
144 for (i = 0; i < 20; i++) {
145 unsigned int val = *sha1++;
146 *buf++ = hex[val >> 4];
147 *buf++ = hex[val & 0xf];
148 }
149 *buf = '\0';
150
151 return buffer;
152}
153
154static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
155{
156 int i;
157 for (i = 0; i < 20; i++) {
158 static char hex[] = "0123456789abcdef";
159 unsigned int val = sha1[i];
160 char *pos = pathbuf + i*2 + (i > 0);
161 *pos++ = hex[val >> 4];
162 *pos = hex[val & 0xf];
163 }
164}
165
166/*
167 * NOTE! This returns a statically allocated buffer, so you have to be
168 * careful about using it. Do an "xstrdup()" if you need to save the
169 * filename.
170 *
171 * Also note that this returns the location for creating. Reading
172 * SHA1 file can happen from any alternate directory listed in the
173 * DB_ENVIRONMENT environment variable if it is not found in
174 * the primary object database.
175 */
176char *sha1_file_name(const unsigned char *sha1)
177{
178 static char *name, *base;
179
180 if (!base) {
181 const char *sha1_file_directory = get_object_directory();
182 int len = strlen(sha1_file_directory);
183 base = xmalloc(len + 60);
184 memcpy(base, sha1_file_directory, len);
185 memset(base+len, 0, 60);
186 base[len] = '/';
187 base[len+3] = '/';
188 name = base + len + 1;
189 }
190 fill_sha1_path(name, sha1);
191 return base;
192}
193
194static char *sha1_get_pack_name(const unsigned char *sha1,
195 char **name, char **base, const char *which)
196{
197 static const char hex[] = "0123456789abcdef";
198 char *buf;
199 int i;
200
201 if (!*base) {
202 const char *sha1_file_directory = get_object_directory();
203 int len = strlen(sha1_file_directory);
204 *base = xmalloc(len + 60);
205 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
206 sha1_file_directory, which);
207 *name = *base + len + 11;
208 }
209
210 buf = *name;
211
212 for (i = 0; i < 20; i++) {
213 unsigned int val = *sha1++;
214 *buf++ = hex[val >> 4];
215 *buf++ = hex[val & 0xf];
216 }
217
218 return *base;
219}
220
221char *sha1_pack_name(const unsigned char *sha1)
222{
223 static char *name, *base;
224
225 return sha1_get_pack_name(sha1, &name, &base, "pack");
226}
227
228char *sha1_pack_index_name(const unsigned char *sha1)
229{
230 static char *name, *base;
231
232 return sha1_get_pack_name(sha1, &name, &base, "idx");
233}
234
235struct alternate_object_database *alt_odb_list;
236static struct alternate_object_database **alt_odb_tail;
237
238static void read_info_alternates(const char * alternates, int depth);
239
240/*
241 * Prepare alternate object database registry.
242 *
243 * The variable alt_odb_list points at the list of struct
244 * alternate_object_database. The elements on this list come from
245 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
246 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
247 * whose contents is similar to that environment variable but can be
248 * LF separated. Its base points at a statically allocated buffer that
249 * contains "/the/directory/corresponding/to/.git/objects/...", while
250 * its name points just after the slash at the end of ".git/objects/"
251 * in the example above, and has enough space to hold 40-byte hex
252 * SHA1, an extra slash for the first level indirection, and the
253 * terminating NUL.
254 */
255static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
256{
257 const char *objdir = get_object_directory();
258 struct alternate_object_database *ent;
259 struct alternate_object_database *alt;
260 /* 43 = 40-byte + 2 '/' + terminating NUL */
261 int pfxlen = len;
262 int entlen = pfxlen + 43;
263 int base_len = -1;
264
265 if (!is_absolute_path(entry) && relative_base) {
266 /* Relative alt-odb */
267 if (base_len < 0)
268 base_len = strlen(relative_base) + 1;
269 entlen += base_len;
270 pfxlen += base_len;
271 }
272 ent = xmalloc(sizeof(*ent) + entlen);
273
274 if (!is_absolute_path(entry) && relative_base) {
275 memcpy(ent->base, relative_base, base_len - 1);
276 ent->base[base_len - 1] = '/';
277 memcpy(ent->base + base_len, entry, len);
278 }
279 else
280 memcpy(ent->base, entry, pfxlen);
281
282 ent->name = ent->base + pfxlen + 1;
283 ent->base[pfxlen + 3] = '/';
284 ent->base[pfxlen] = ent->base[entlen-1] = 0;
285
286 /* Detect cases where alternate disappeared */
287 if (!is_directory(ent->base)) {
288 error("object directory %s does not exist; "
289 "check .git/objects/info/alternates.",
290 ent->base);
291 free(ent);
292 return -1;
293 }
294
295 /* Prevent the common mistake of listing the same
296 * thing twice, or object directory itself.
297 */
298 for (alt = alt_odb_list; alt; alt = alt->next) {
299 if (!memcmp(ent->base, alt->base, pfxlen)) {
300 free(ent);
301 return -1;
302 }
303 }
304 if (!memcmp(ent->base, objdir, pfxlen)) {
305 free(ent);
306 return -1;
307 }
308
309 /* add the alternate entry */
310 *alt_odb_tail = ent;
311 alt_odb_tail = &(ent->next);
312 ent->next = NULL;
313
314 /* recursively add alternates */
315 read_info_alternates(ent->base, depth + 1);
316
317 ent->base[pfxlen] = '/';
318
319 return 0;
320}
321
322static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
323 const char *relative_base, int depth)
324{
325 const char *cp, *last;
326
327 if (depth > 5) {
328 error("%s: ignoring alternate object stores, nesting too deep.",
329 relative_base);
330 return;
331 }
332
333 last = alt;
334 while (last < ep) {
335 cp = last;
336 if (cp < ep && *cp == '#') {
337 while (cp < ep && *cp != sep)
338 cp++;
339 last = cp + 1;
340 continue;
341 }
342 while (cp < ep && *cp != sep)
343 cp++;
344 if (last != cp) {
345 if (!is_absolute_path(last) && depth) {
346 error("%s: ignoring relative alternate object store %s",
347 relative_base, last);
348 } else {
349 link_alt_odb_entry(last, cp - last,
350 relative_base, depth);
351 }
352 }
353 while (cp < ep && *cp == sep)
354 cp++;
355 last = cp;
356 }
357}
358
359static void read_info_alternates(const char * relative_base, int depth)
360{
361 char *map;
362 size_t mapsz;
363 struct stat st;
364 const char alt_file_name[] = "info/alternates";
365 /* Given that relative_base is no longer than PATH_MAX,
366 ensure that "path" has enough space to append "/", the
367 file name, "info/alternates", and a trailing NUL. */
368 char path[PATH_MAX + 1 + sizeof alt_file_name];
369 int fd;
370
371 sprintf(path, "%s/%s", relative_base, alt_file_name);
372 fd = open(path, O_RDONLY);
373 if (fd < 0)
374 return;
375 if (fstat(fd, &st) || (st.st_size == 0)) {
376 close(fd);
377 return;
378 }
379 mapsz = xsize_t(st.st_size);
380 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
381 close(fd);
382
383 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
384
385 munmap(map, mapsz);
386}
387
388void add_to_alternates_file(const char *reference)
389{
390 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
391 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
392 char *alt = mkpath("%s/objects\n", reference);
393 write_or_die(fd, alt, strlen(alt));
394 if (commit_lock_file(lock))
395 die("could not close alternates file");
396 if (alt_odb_tail)
397 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
398}
399
400void foreach_alt_odb(alt_odb_fn fn, void *cb)
401{
402 struct alternate_object_database *ent;
403
404 prepare_alt_odb();
405 for (ent = alt_odb_list; ent; ent = ent->next)
406 if (fn(ent, cb))
407 return;
408}
409
410void prepare_alt_odb(void)
411{
412 const char *alt;
413
414 if (alt_odb_tail)
415 return;
416
417 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
418 if (!alt) alt = "";
419
420 alt_odb_tail = &alt_odb_list;
421 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
422
423 read_info_alternates(get_object_directory(), 0);
424}
425
426static int has_loose_object_local(const unsigned char *sha1)
427{
428 char *name = sha1_file_name(sha1);
429 return !access(name, F_OK);
430}
431
432int has_loose_object_nonlocal(const unsigned char *sha1)
433{
434 struct alternate_object_database *alt;
435 prepare_alt_odb();
436 for (alt = alt_odb_list; alt; alt = alt->next) {
437 fill_sha1_path(alt->name, sha1);
438 if (!access(alt->base, F_OK))
439 return 1;
440 }
441 return 0;
442}
443
444static int has_loose_object(const unsigned char *sha1)
445{
446 return has_loose_object_local(sha1) ||
447 has_loose_object_nonlocal(sha1);
448}
449
450static unsigned int pack_used_ctr;
451static unsigned int pack_mmap_calls;
452static unsigned int peak_pack_open_windows;
453static unsigned int pack_open_windows;
454static size_t peak_pack_mapped;
455static size_t pack_mapped;
456struct packed_git *packed_git;
457
458void pack_report(void)
459{
460 fprintf(stderr,
461 "pack_report: getpagesize() = %10" SZ_FMT "\n"
462 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
463 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
464 sz_fmt(getpagesize()),
465 sz_fmt(packed_git_window_size),
466 sz_fmt(packed_git_limit));
467 fprintf(stderr,
468 "pack_report: pack_used_ctr = %10u\n"
469 "pack_report: pack_mmap_calls = %10u\n"
470 "pack_report: pack_open_windows = %10u / %10u\n"
471 "pack_report: pack_mapped = "
472 "%10" SZ_FMT " / %10" SZ_FMT "\n",
473 pack_used_ctr,
474 pack_mmap_calls,
475 pack_open_windows, peak_pack_open_windows,
476 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
477}
478
479static int check_packed_git_idx(const char *path, struct packed_git *p)
480{
481 void *idx_map;
482 struct pack_idx_header *hdr;
483 size_t idx_size;
484 uint32_t version, nr, i, *index;
485 int fd = open(path, O_RDONLY);
486 struct stat st;
487
488 if (fd < 0)
489 return -1;
490 if (fstat(fd, &st)) {
491 close(fd);
492 return -1;
493 }
494 idx_size = xsize_t(st.st_size);
495 if (idx_size < 4 * 256 + 20 + 20) {
496 close(fd);
497 return error("index file %s is too small", path);
498 }
499 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
500 close(fd);
501
502 hdr = idx_map;
503 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
504 version = ntohl(hdr->idx_version);
505 if (version < 2 || version > 2) {
506 munmap(idx_map, idx_size);
507 return error("index file %s is version %"PRIu32
508 " and is not supported by this binary"
509 " (try upgrading GIT to a newer version)",
510 path, version);
511 }
512 } else
513 version = 1;
514
515 nr = 0;
516 index = idx_map;
517 if (version > 1)
518 index += 2; /* skip index header */
519 for (i = 0; i < 256; i++) {
520 uint32_t n = ntohl(index[i]);
521 if (n < nr) {
522 munmap(idx_map, idx_size);
523 return error("non-monotonic index %s", path);
524 }
525 nr = n;
526 }
527
528 if (version == 1) {
529 /*
530 * Total size:
531 * - 256 index entries 4 bytes each
532 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
533 * - 20-byte SHA1 of the packfile
534 * - 20-byte SHA1 file checksum
535 */
536 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
537 munmap(idx_map, idx_size);
538 return error("wrong index v1 file size in %s", path);
539 }
540 } else if (version == 2) {
541 /*
542 * Minimum size:
543 * - 8 bytes of header
544 * - 256 index entries 4 bytes each
545 * - 20-byte sha1 entry * nr
546 * - 4-byte crc entry * nr
547 * - 4-byte offset entry * nr
548 * - 20-byte SHA1 of the packfile
549 * - 20-byte SHA1 file checksum
550 * And after the 4-byte offset table might be a
551 * variable sized table containing 8-byte entries
552 * for offsets larger than 2^31.
553 */
554 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
555 unsigned long max_size = min_size;
556 if (nr)
557 max_size += (nr - 1)*8;
558 if (idx_size < min_size || idx_size > max_size) {
559 munmap(idx_map, idx_size);
560 return error("wrong index v2 file size in %s", path);
561 }
562 if (idx_size != min_size &&
563 /*
564 * make sure we can deal with large pack offsets.
565 * 31-bit signed offset won't be enough, neither
566 * 32-bit unsigned one will be.
567 */
568 (sizeof(off_t) <= 4)) {
569 munmap(idx_map, idx_size);
570 return error("pack too large for current definition of off_t in %s", path);
571 }
572 }
573
574 p->index_version = version;
575 p->index_data = idx_map;
576 p->index_size = idx_size;
577 p->num_objects = nr;
578 return 0;
579}
580
581int open_pack_index(struct packed_git *p)
582{
583 char *idx_name;
584 int ret;
585
586 if (p->index_data)
587 return 0;
588
589 idx_name = xstrdup(p->pack_name);
590 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
591 ret = check_packed_git_idx(idx_name, p);
592 free(idx_name);
593 return ret;
594}
595
596static void scan_windows(struct packed_git *p,
597 struct packed_git **lru_p,
598 struct pack_window **lru_w,
599 struct pack_window **lru_l)
600{
601 struct pack_window *w, *w_l;
602
603 for (w_l = NULL, w = p->windows; w; w = w->next) {
604 if (!w->inuse_cnt) {
605 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
606 *lru_p = p;
607 *lru_w = w;
608 *lru_l = w_l;
609 }
610 }
611 w_l = w;
612 }
613}
614
615static int unuse_one_window(struct packed_git *current, int keep_fd)
616{
617 struct packed_git *p, *lru_p = NULL;
618 struct pack_window *lru_w = NULL, *lru_l = NULL;
619
620 if (current)
621 scan_windows(current, &lru_p, &lru_w, &lru_l);
622 for (p = packed_git; p; p = p->next)
623 scan_windows(p, &lru_p, &lru_w, &lru_l);
624 if (lru_p) {
625 munmap(lru_w->base, lru_w->len);
626 pack_mapped -= lru_w->len;
627 if (lru_l)
628 lru_l->next = lru_w->next;
629 else {
630 lru_p->windows = lru_w->next;
631 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
632 close(lru_p->pack_fd);
633 lru_p->pack_fd = -1;
634 }
635 }
636 free(lru_w);
637 pack_open_windows--;
638 return 1;
639 }
640 return 0;
641}
642
643void release_pack_memory(size_t need, int fd)
644{
645 size_t cur = pack_mapped;
646 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
647 ; /* nothing */
648}
649
650void close_pack_windows(struct packed_git *p)
651{
652 while (p->windows) {
653 struct pack_window *w = p->windows;
654
655 if (w->inuse_cnt)
656 die("pack '%s' still has open windows to it",
657 p->pack_name);
658 munmap(w->base, w->len);
659 pack_mapped -= w->len;
660 pack_open_windows--;
661 p->windows = w->next;
662 free(w);
663 }
664}
665
666void unuse_pack(struct pack_window **w_cursor)
667{
668 struct pack_window *w = *w_cursor;
669 if (w) {
670 w->inuse_cnt--;
671 *w_cursor = NULL;
672 }
673}
674
675/*
676 * This is used by git-repack in case a newly created pack happens to
677 * contain the same set of objects as an existing one. In that case
678 * the resulting file might be different even if its name would be the
679 * same. It is best to close any reference to the old pack before it is
680 * replaced on disk. Of course no index pointers nor windows for given pack
681 * must subsist at this point. If ever objects from this pack are requested
682 * again, the new version of the pack will be reinitialized through
683 * reprepare_packed_git().
684 */
685void free_pack_by_name(const char *pack_name)
686{
687 struct packed_git *p, **pp = &packed_git;
688
689 while (*pp) {
690 p = *pp;
691 if (strcmp(pack_name, p->pack_name) == 0) {
692 close_pack_windows(p);
693 if (p->pack_fd != -1)
694 close(p->pack_fd);
695 if (p->index_data)
696 munmap((void *)p->index_data, p->index_size);
697 free(p->bad_object_sha1);
698 *pp = p->next;
699 free(p);
700 return;
701 }
702 pp = &p->next;
703 }
704}
705
706/*
707 * Do not call this directly as this leaks p->pack_fd on error return;
708 * call open_packed_git() instead.
709 */
710static int open_packed_git_1(struct packed_git *p)
711{
712 struct stat st;
713 struct pack_header hdr;
714 unsigned char sha1[20];
715 unsigned char *idx_sha1;
716 long fd_flag;
717
718 if (!p->index_data && open_pack_index(p))
719 return error("packfile %s index unavailable", p->pack_name);
720
721 p->pack_fd = open(p->pack_name, O_RDONLY);
722 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
723 return -1;
724
725 /* If we created the struct before we had the pack we lack size. */
726 if (!p->pack_size) {
727 if (!S_ISREG(st.st_mode))
728 return error("packfile %s not a regular file", p->pack_name);
729 p->pack_size = st.st_size;
730 } else if (p->pack_size != st.st_size)
731 return error("packfile %s size changed", p->pack_name);
732
733 /* We leave these file descriptors open with sliding mmap;
734 * there is no point keeping them open across exec(), though.
735 */
736 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
737 if (fd_flag < 0)
738 return error("cannot determine file descriptor flags");
739 fd_flag |= FD_CLOEXEC;
740 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
741 return error("cannot set FD_CLOEXEC");
742
743 /* Verify we recognize this pack file format. */
744 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
745 return error("file %s is far too short to be a packfile", p->pack_name);
746 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
747 return error("file %s is not a GIT packfile", p->pack_name);
748 if (!pack_version_ok(hdr.hdr_version))
749 return error("packfile %s is version %"PRIu32" and not"
750 " supported (try upgrading GIT to a newer version)",
751 p->pack_name, ntohl(hdr.hdr_version));
752
753 /* Verify the pack matches its index. */
754 if (p->num_objects != ntohl(hdr.hdr_entries))
755 return error("packfile %s claims to have %"PRIu32" objects"
756 " while index indicates %"PRIu32" objects",
757 p->pack_name, ntohl(hdr.hdr_entries),
758 p->num_objects);
759 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
760 return error("end of packfile %s is unavailable", p->pack_name);
761 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
762 return error("packfile %s signature is unavailable", p->pack_name);
763 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
764 if (hashcmp(sha1, idx_sha1))
765 return error("packfile %s does not match index", p->pack_name);
766 return 0;
767}
768
769static int open_packed_git(struct packed_git *p)
770{
771 if (!open_packed_git_1(p))
772 return 0;
773 if (p->pack_fd != -1) {
774 close(p->pack_fd);
775 p->pack_fd = -1;
776 }
777 return -1;
778}
779
780static int in_window(struct pack_window *win, off_t offset)
781{
782 /* We must promise at least 20 bytes (one hash) after the
783 * offset is available from this window, otherwise the offset
784 * is not actually in this window and a different window (which
785 * has that one hash excess) must be used. This is to support
786 * the object header and delta base parsing routines below.
787 */
788 off_t win_off = win->offset;
789 return win_off <= offset
790 && (offset + 20) <= (win_off + win->len);
791}
792
793unsigned char* use_pack(struct packed_git *p,
794 struct pack_window **w_cursor,
795 off_t offset,
796 unsigned int *left)
797{
798 struct pack_window *win = *w_cursor;
799
800 if (p->pack_fd == -1 && open_packed_git(p))
801 die("packfile %s cannot be accessed", p->pack_name);
802
803 /* Since packfiles end in a hash of their content and its
804 * pointless to ask for an offset into the middle of that
805 * hash, and the in_window function above wouldn't match
806 * don't allow an offset too close to the end of the file.
807 */
808 if (offset > (p->pack_size - 20))
809 die("offset beyond end of packfile (truncated pack?)");
810
811 if (!win || !in_window(win, offset)) {
812 if (win)
813 win->inuse_cnt--;
814 for (win = p->windows; win; win = win->next) {
815 if (in_window(win, offset))
816 break;
817 }
818 if (!win) {
819 size_t window_align = packed_git_window_size / 2;
820 off_t len;
821 win = xcalloc(1, sizeof(*win));
822 win->offset = (offset / window_align) * window_align;
823 len = p->pack_size - win->offset;
824 if (len > packed_git_window_size)
825 len = packed_git_window_size;
826 win->len = (size_t)len;
827 pack_mapped += win->len;
828 while (packed_git_limit < pack_mapped
829 && unuse_one_window(p, p->pack_fd))
830 ; /* nothing */
831 win->base = xmmap(NULL, win->len,
832 PROT_READ, MAP_PRIVATE,
833 p->pack_fd, win->offset);
834 if (win->base == MAP_FAILED)
835 die("packfile %s cannot be mapped: %s",
836 p->pack_name,
837 strerror(errno));
838 pack_mmap_calls++;
839 pack_open_windows++;
840 if (pack_mapped > peak_pack_mapped)
841 peak_pack_mapped = pack_mapped;
842 if (pack_open_windows > peak_pack_open_windows)
843 peak_pack_open_windows = pack_open_windows;
844 win->next = p->windows;
845 p->windows = win;
846 }
847 }
848 if (win != *w_cursor) {
849 win->last_used = pack_used_ctr++;
850 win->inuse_cnt++;
851 *w_cursor = win;
852 }
853 offset -= win->offset;
854 if (left)
855 *left = win->len - xsize_t(offset);
856 return win->base + offset;
857}
858
859static struct packed_git *alloc_packed_git(int extra)
860{
861 struct packed_git *p = xmalloc(sizeof(*p) + extra);
862 memset(p, 0, sizeof(*p));
863 p->pack_fd = -1;
864 return p;
865}
866
867struct packed_git *add_packed_git(const char *path, int path_len, int local)
868{
869 struct stat st;
870 struct packed_git *p = alloc_packed_git(path_len + 2);
871
872 /*
873 * Make sure a corresponding .pack file exists and that
874 * the index looks sane.
875 */
876 path_len -= strlen(".idx");
877 if (path_len < 1) {
878 free(p);
879 return NULL;
880 }
881 memcpy(p->pack_name, path, path_len);
882
883 strcpy(p->pack_name + path_len, ".keep");
884 if (!access(p->pack_name, F_OK))
885 p->pack_keep = 1;
886
887 strcpy(p->pack_name + path_len, ".pack");
888 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
889 free(p);
890 return NULL;
891 }
892
893 /* ok, it looks sane as far as we can check without
894 * actually mapping the pack file.
895 */
896 p->pack_size = st.st_size;
897 p->pack_local = local;
898 p->mtime = st.st_mtime;
899 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
900 hashclr(p->sha1);
901 return p;
902}
903
904struct packed_git *parse_pack_index(unsigned char *sha1)
905{
906 const char *idx_path = sha1_pack_index_name(sha1);
907 const char *path = sha1_pack_name(sha1);
908 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
909
910 strcpy(p->pack_name, path);
911 hashcpy(p->sha1, sha1);
912 if (check_packed_git_idx(idx_path, p)) {
913 free(p);
914 return NULL;
915 }
916
917 return p;
918}
919
920void install_packed_git(struct packed_git *pack)
921{
922 pack->next = packed_git;
923 packed_git = pack;
924}
925
926static void prepare_packed_git_one(char *objdir, int local)
927{
928 /* Ensure that this buffer is large enough so that we can
929 append "/pack/" without clobbering the stack even if
930 strlen(objdir) were PATH_MAX. */
931 char path[PATH_MAX + 1 + 4 + 1 + 1];
932 int len;
933 DIR *dir;
934 struct dirent *de;
935
936 sprintf(path, "%s/pack", objdir);
937 len = strlen(path);
938 dir = opendir(path);
939 if (!dir) {
940 if (errno != ENOENT)
941 error("unable to open object pack directory: %s: %s",
942 path, strerror(errno));
943 return;
944 }
945 path[len++] = '/';
946 while ((de = readdir(dir)) != NULL) {
947 int namelen = strlen(de->d_name);
948 struct packed_git *p;
949
950 if (!has_extension(de->d_name, ".idx"))
951 continue;
952
953 if (len + namelen + 1 > sizeof(path))
954 continue;
955
956 /* Don't reopen a pack we already have. */
957 strcpy(path + len, de->d_name);
958 for (p = packed_git; p; p = p->next) {
959 if (!memcmp(path, p->pack_name, len + namelen - 4))
960 break;
961 }
962 if (p)
963 continue;
964 /* See if it really is a valid .idx file with corresponding
965 * .pack file that we can map.
966 */
967 p = add_packed_git(path, len + namelen, local);
968 if (!p)
969 continue;
970 install_packed_git(p);
971 }
972 closedir(dir);
973}
974
975static int sort_pack(const void *a_, const void *b_)
976{
977 struct packed_git *a = *((struct packed_git **)a_);
978 struct packed_git *b = *((struct packed_git **)b_);
979 int st;
980
981 /*
982 * Local packs tend to contain objects specific to our
983 * variant of the project than remote ones. In addition,
984 * remote ones could be on a network mounted filesystem.
985 * Favor local ones for these reasons.
986 */
987 st = a->pack_local - b->pack_local;
988 if (st)
989 return -st;
990
991 /*
992 * Younger packs tend to contain more recent objects,
993 * and more recent objects tend to get accessed more
994 * often.
995 */
996 if (a->mtime < b->mtime)
997 return 1;
998 else if (a->mtime == b->mtime)
999 return 0;
1000 return -1;
1001}
1002
1003static void rearrange_packed_git(void)
1004{
1005 struct packed_git **ary, *p;
1006 int i, n;
1007
1008 for (n = 0, p = packed_git; p; p = p->next)
1009 n++;
1010 if (n < 2)
1011 return;
1012
1013 /* prepare an array of packed_git for easier sorting */
1014 ary = xcalloc(n, sizeof(struct packed_git *));
1015 for (n = 0, p = packed_git; p; p = p->next)
1016 ary[n++] = p;
1017
1018 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1019
1020 /* link them back again */
1021 for (i = 0; i < n - 1; i++)
1022 ary[i]->next = ary[i + 1];
1023 ary[n - 1]->next = NULL;
1024 packed_git = ary[0];
1025
1026 free(ary);
1027}
1028
1029static int prepare_packed_git_run_once = 0;
1030void prepare_packed_git(void)
1031{
1032 struct alternate_object_database *alt;
1033
1034 if (prepare_packed_git_run_once)
1035 return;
1036 prepare_packed_git_one(get_object_directory(), 1);
1037 prepare_alt_odb();
1038 for (alt = alt_odb_list; alt; alt = alt->next) {
1039 alt->name[-1] = 0;
1040 prepare_packed_git_one(alt->base, 0);
1041 alt->name[-1] = '/';
1042 }
1043 rearrange_packed_git();
1044 prepare_packed_git_run_once = 1;
1045}
1046
1047void reprepare_packed_git(void)
1048{
1049 discard_revindex();
1050 prepare_packed_git_run_once = 0;
1051 prepare_packed_git();
1052}
1053
1054static void mark_bad_packed_object(struct packed_git *p,
1055 const unsigned char *sha1)
1056{
1057 unsigned i;
1058 for (i = 0; i < p->num_bad_objects; i++)
1059 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1060 return;
1061 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1062 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1063 p->num_bad_objects++;
1064}
1065
1066static int has_packed_and_bad(const unsigned char *sha1)
1067{
1068 struct packed_git *p;
1069 unsigned i;
1070
1071 for (p = packed_git; p; p = p->next)
1072 for (i = 0; i < p->num_bad_objects; i++)
1073 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1074 return 1;
1075 return 0;
1076}
1077
1078int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1079{
1080 unsigned char real_sha1[20];
1081 hash_sha1_file(map, size, type, real_sha1);
1082 return hashcmp(sha1, real_sha1) ? -1 : 0;
1083}
1084
1085static int git_open_noatime(const char *name)
1086{
1087 static int sha1_file_open_flag = O_NOATIME;
1088 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1089
1090 /* Might the failure be due to O_NOATIME? */
1091 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1092 fd = open(name, O_RDONLY);
1093 if (fd >= 0)
1094 sha1_file_open_flag = 0;
1095 }
1096 return fd;
1097}
1098
1099static int open_sha1_file(const unsigned char *sha1)
1100{
1101 int fd;
1102 char *name = sha1_file_name(sha1);
1103 struct alternate_object_database *alt;
1104
1105 fd = git_open_noatime(name);
1106 if (fd >= 0)
1107 return fd;
1108
1109 prepare_alt_odb();
1110 errno = ENOENT;
1111 for (alt = alt_odb_list; alt; alt = alt->next) {
1112 name = alt->name;
1113 fill_sha1_path(name, sha1);
1114 fd = git_open_noatime(alt->base);
1115 if (fd >= 0)
1116 return fd;
1117 }
1118 return -1;
1119}
1120
1121static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1122{
1123 void *map;
1124 int fd;
1125
1126 fd = open_sha1_file(sha1);
1127 map = NULL;
1128 if (fd >= 0) {
1129 struct stat st;
1130
1131 if (!fstat(fd, &st)) {
1132 *size = xsize_t(st.st_size);
1133 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1134 }
1135 close(fd);
1136 }
1137 return map;
1138}
1139
1140static int legacy_loose_object(unsigned char *map)
1141{
1142 unsigned int word;
1143
1144 /*
1145 * Is it a zlib-compressed buffer? If so, the first byte
1146 * must be 0x78 (15-bit window size, deflated), and the
1147 * first 16-bit word is evenly divisible by 31
1148 */
1149 word = (map[0] << 8) + map[1];
1150 if (map[0] == 0x78 && !(word % 31))
1151 return 1;
1152 else
1153 return 0;
1154}
1155
1156unsigned long unpack_object_header_buffer(const unsigned char *buf,
1157 unsigned long len, enum object_type *type, unsigned long *sizep)
1158{
1159 unsigned shift;
1160 unsigned char c;
1161 unsigned long size;
1162 unsigned long used = 0;
1163
1164 c = buf[used++];
1165 *type = (c >> 4) & 7;
1166 size = c & 15;
1167 shift = 4;
1168 while (c & 0x80) {
1169 if (len <= used || sizeof(long) * 8 <= shift) {
1170 error("bad object header");
1171 return 0;
1172 }
1173 c = buf[used++];
1174 size += (c & 0x7f) << shift;
1175 shift += 7;
1176 }
1177 *sizep = size;
1178 return used;
1179}
1180
1181static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1182{
1183 unsigned long size, used;
1184 static const char valid_loose_object_type[8] = {
1185 0, /* OBJ_EXT */
1186 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1187 0, /* "delta" and others are invalid in a loose object */
1188 };
1189 enum object_type type;
1190
1191 /* Get the data stream */
1192 memset(stream, 0, sizeof(*stream));
1193 stream->next_in = map;
1194 stream->avail_in = mapsize;
1195 stream->next_out = buffer;
1196 stream->avail_out = bufsiz;
1197
1198 if (legacy_loose_object(map)) {
1199 inflateInit(stream);
1200 return inflate(stream, 0);
1201 }
1202
1203
1204 /*
1205 * There used to be a second loose object header format which
1206 * was meant to mimic the in-pack format, allowing for direct
1207 * copy of the object data. This format turned up not to be
1208 * really worth it and we don't write it any longer. But we
1209 * can still read it.
1210 */
1211 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1212 if (!used || !valid_loose_object_type[type])
1213 return -1;
1214 map += used;
1215 mapsize -= used;
1216
1217 /* Set up the stream for the rest.. */
1218 stream->next_in = map;
1219 stream->avail_in = mapsize;
1220 inflateInit(stream);
1221
1222 /* And generate the fake traditional header */
1223 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1224 typename(type), size);
1225 return 0;
1226}
1227
1228static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1229{
1230 int bytes = strlen(buffer) + 1;
1231 unsigned char *buf = xmalloc(1+size);
1232 unsigned long n;
1233 int status = Z_OK;
1234
1235 n = stream->total_out - bytes;
1236 if (n > size)
1237 n = size;
1238 memcpy(buf, (char *) buffer + bytes, n);
1239 bytes = n;
1240 if (bytes <= size) {
1241 /*
1242 * The above condition must be (bytes <= size), not
1243 * (bytes < size). In other words, even though we
1244 * expect no more output and set avail_out to zer0,
1245 * the input zlib stream may have bytes that express
1246 * "this concludes the stream", and we *do* want to
1247 * eat that input.
1248 *
1249 * Otherwise we would not be able to test that we
1250 * consumed all the input to reach the expected size;
1251 * we also want to check that zlib tells us that all
1252 * went well with status == Z_STREAM_END at the end.
1253 */
1254 stream->next_out = buf + bytes;
1255 stream->avail_out = size - bytes;
1256 while (status == Z_OK)
1257 status = inflate(stream, Z_FINISH);
1258 }
1259 buf[size] = 0;
1260 if (status == Z_STREAM_END && !stream->avail_in) {
1261 inflateEnd(stream);
1262 return buf;
1263 }
1264
1265 if (status < 0)
1266 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1267 else if (stream->avail_in)
1268 error("garbage at end of loose object '%s'",
1269 sha1_to_hex(sha1));
1270 free(buf);
1271 return NULL;
1272}
1273
1274/*
1275 * We used to just use "sscanf()", but that's actually way
1276 * too permissive for what we want to check. So do an anal
1277 * object header parse by hand.
1278 */
1279static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1280{
1281 char type[10];
1282 int i;
1283 unsigned long size;
1284
1285 /*
1286 * The type can be at most ten bytes (including the
1287 * terminating '\0' that we add), and is followed by
1288 * a space.
1289 */
1290 i = 0;
1291 for (;;) {
1292 char c = *hdr++;
1293 if (c == ' ')
1294 break;
1295 type[i++] = c;
1296 if (i >= sizeof(type))
1297 return -1;
1298 }
1299 type[i] = 0;
1300
1301 /*
1302 * The length must follow immediately, and be in canonical
1303 * decimal format (ie "010" is not valid).
1304 */
1305 size = *hdr++ - '0';
1306 if (size > 9)
1307 return -1;
1308 if (size) {
1309 for (;;) {
1310 unsigned long c = *hdr - '0';
1311 if (c > 9)
1312 break;
1313 hdr++;
1314 size = size * 10 + c;
1315 }
1316 }
1317 *sizep = size;
1318
1319 /*
1320 * The length must be followed by a zero byte
1321 */
1322 return *hdr ? -1 : type_from_string(type);
1323}
1324
1325static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1326{
1327 int ret;
1328 z_stream stream;
1329 char hdr[8192];
1330
1331 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1332 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1333 return NULL;
1334
1335 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1336}
1337
1338unsigned long get_size_from_delta(struct packed_git *p,
1339 struct pack_window **w_curs,
1340 off_t curpos)
1341{
1342 const unsigned char *data;
1343 unsigned char delta_head[20], *in;
1344 z_stream stream;
1345 int st;
1346
1347 memset(&stream, 0, sizeof(stream));
1348 stream.next_out = delta_head;
1349 stream.avail_out = sizeof(delta_head);
1350
1351 inflateInit(&stream);
1352 do {
1353 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1354 stream.next_in = in;
1355 st = inflate(&stream, Z_FINISH);
1356 curpos += stream.next_in - in;
1357 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1358 stream.total_out < sizeof(delta_head));
1359 inflateEnd(&stream);
1360 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1361 error("delta data unpack-initial failed");
1362 return 0;
1363 }
1364
1365 /* Examine the initial part of the delta to figure out
1366 * the result size.
1367 */
1368 data = delta_head;
1369
1370 /* ignore base size */
1371 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1372
1373 /* Read the result size */
1374 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1375}
1376
1377static off_t get_delta_base(struct packed_git *p,
1378 struct pack_window **w_curs,
1379 off_t *curpos,
1380 enum object_type type,
1381 off_t delta_obj_offset)
1382{
1383 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1384 off_t base_offset;
1385
1386 /* use_pack() assured us we have [base_info, base_info + 20)
1387 * as a range that we can look at without walking off the
1388 * end of the mapped window. Its actually the hash size
1389 * that is assured. An OFS_DELTA longer than the hash size
1390 * is stupid, as then a REF_DELTA would be smaller to store.
1391 */
1392 if (type == OBJ_OFS_DELTA) {
1393 unsigned used = 0;
1394 unsigned char c = base_info[used++];
1395 base_offset = c & 127;
1396 while (c & 128) {
1397 base_offset += 1;
1398 if (!base_offset || MSB(base_offset, 7))
1399 return 0; /* overflow */
1400 c = base_info[used++];
1401 base_offset = (base_offset << 7) + (c & 127);
1402 }
1403 base_offset = delta_obj_offset - base_offset;
1404 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1405 return 0; /* out of bound */
1406 *curpos += used;
1407 } else if (type == OBJ_REF_DELTA) {
1408 /* The base entry _must_ be in the same pack */
1409 base_offset = find_pack_entry_one(base_info, p);
1410 *curpos += 20;
1411 } else
1412 die("I am totally screwed");
1413 return base_offset;
1414}
1415
1416/* forward declaration for a mutually recursive function */
1417static int packed_object_info(struct packed_git *p, off_t offset,
1418 unsigned long *sizep);
1419
1420static int packed_delta_info(struct packed_git *p,
1421 struct pack_window **w_curs,
1422 off_t curpos,
1423 enum object_type type,
1424 off_t obj_offset,
1425 unsigned long *sizep)
1426{
1427 off_t base_offset;
1428
1429 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1430 if (!base_offset)
1431 return OBJ_BAD;
1432 type = packed_object_info(p, base_offset, NULL);
1433 if (type <= OBJ_NONE) {
1434 struct revindex_entry *revidx;
1435 const unsigned char *base_sha1;
1436 revidx = find_pack_revindex(p, base_offset);
1437 if (!revidx)
1438 return OBJ_BAD;
1439 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1440 mark_bad_packed_object(p, base_sha1);
1441 type = sha1_object_info(base_sha1, NULL);
1442 if (type <= OBJ_NONE)
1443 return OBJ_BAD;
1444 }
1445
1446 /* We choose to only get the type of the base object and
1447 * ignore potentially corrupt pack file that expects the delta
1448 * based on a base with a wrong size. This saves tons of
1449 * inflate() calls.
1450 */
1451 if (sizep) {
1452 *sizep = get_size_from_delta(p, w_curs, curpos);
1453 if (*sizep == 0)
1454 type = OBJ_BAD;
1455 }
1456
1457 return type;
1458}
1459
1460static int unpack_object_header(struct packed_git *p,
1461 struct pack_window **w_curs,
1462 off_t *curpos,
1463 unsigned long *sizep)
1464{
1465 unsigned char *base;
1466 unsigned int left;
1467 unsigned long used;
1468 enum object_type type;
1469
1470 /* use_pack() assures us we have [base, base + 20) available
1471 * as a range that we can look at at. (Its actually the hash
1472 * size that is assured.) With our object header encoding
1473 * the maximum deflated object size is 2^137, which is just
1474 * insane, so we know won't exceed what we have been given.
1475 */
1476 base = use_pack(p, w_curs, *curpos, &left);
1477 used = unpack_object_header_buffer(base, left, &type, sizep);
1478 if (!used) {
1479 type = OBJ_BAD;
1480 } else
1481 *curpos += used;
1482
1483 return type;
1484}
1485
1486const char *packed_object_info_detail(struct packed_git *p,
1487 off_t obj_offset,
1488 unsigned long *size,
1489 unsigned long *store_size,
1490 unsigned int *delta_chain_length,
1491 unsigned char *base_sha1)
1492{
1493 struct pack_window *w_curs = NULL;
1494 off_t curpos;
1495 unsigned long dummy;
1496 unsigned char *next_sha1;
1497 enum object_type type;
1498 struct revindex_entry *revidx;
1499
1500 *delta_chain_length = 0;
1501 curpos = obj_offset;
1502 type = unpack_object_header(p, &w_curs, &curpos, size);
1503
1504 revidx = find_pack_revindex(p, obj_offset);
1505 *store_size = revidx[1].offset - obj_offset;
1506
1507 for (;;) {
1508 switch (type) {
1509 default:
1510 die("pack %s contains unknown object type %d",
1511 p->pack_name, type);
1512 case OBJ_COMMIT:
1513 case OBJ_TREE:
1514 case OBJ_BLOB:
1515 case OBJ_TAG:
1516 unuse_pack(&w_curs);
1517 return typename(type);
1518 case OBJ_OFS_DELTA:
1519 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1520 if (!obj_offset)
1521 die("pack %s contains bad delta base reference of type %s",
1522 p->pack_name, typename(type));
1523 if (*delta_chain_length == 0) {
1524 revidx = find_pack_revindex(p, obj_offset);
1525 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1526 }
1527 break;
1528 case OBJ_REF_DELTA:
1529 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1530 if (*delta_chain_length == 0)
1531 hashcpy(base_sha1, next_sha1);
1532 obj_offset = find_pack_entry_one(next_sha1, p);
1533 break;
1534 }
1535 (*delta_chain_length)++;
1536 curpos = obj_offset;
1537 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1538 }
1539}
1540
1541static int packed_object_info(struct packed_git *p, off_t obj_offset,
1542 unsigned long *sizep)
1543{
1544 struct pack_window *w_curs = NULL;
1545 unsigned long size;
1546 off_t curpos = obj_offset;
1547 enum object_type type;
1548
1549 type = unpack_object_header(p, &w_curs, &curpos, &size);
1550
1551 switch (type) {
1552 case OBJ_OFS_DELTA:
1553 case OBJ_REF_DELTA:
1554 type = packed_delta_info(p, &w_curs, curpos,
1555 type, obj_offset, sizep);
1556 break;
1557 case OBJ_COMMIT:
1558 case OBJ_TREE:
1559 case OBJ_BLOB:
1560 case OBJ_TAG:
1561 if (sizep)
1562 *sizep = size;
1563 break;
1564 default:
1565 error("unknown object type %i at offset %"PRIuMAX" in %s",
1566 type, (uintmax_t)obj_offset, p->pack_name);
1567 type = OBJ_BAD;
1568 }
1569 unuse_pack(&w_curs);
1570 return type;
1571}
1572
1573static void *unpack_compressed_entry(struct packed_git *p,
1574 struct pack_window **w_curs,
1575 off_t curpos,
1576 unsigned long size)
1577{
1578 int st;
1579 z_stream stream;
1580 unsigned char *buffer, *in;
1581
1582 buffer = xmalloc(size + 1);
1583 buffer[size] = 0;
1584 memset(&stream, 0, sizeof(stream));
1585 stream.next_out = buffer;
1586 stream.avail_out = size;
1587
1588 inflateInit(&stream);
1589 do {
1590 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1591 stream.next_in = in;
1592 st = inflate(&stream, Z_FINISH);
1593 curpos += stream.next_in - in;
1594 } while (st == Z_OK || st == Z_BUF_ERROR);
1595 inflateEnd(&stream);
1596 if ((st != Z_STREAM_END) || stream.total_out != size) {
1597 free(buffer);
1598 return NULL;
1599 }
1600
1601 return buffer;
1602}
1603
1604#define MAX_DELTA_CACHE (256)
1605
1606static size_t delta_base_cached;
1607
1608static struct delta_base_cache_lru_list {
1609 struct delta_base_cache_lru_list *prev;
1610 struct delta_base_cache_lru_list *next;
1611} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1612
1613static struct delta_base_cache_entry {
1614 struct delta_base_cache_lru_list lru;
1615 void *data;
1616 struct packed_git *p;
1617 off_t base_offset;
1618 unsigned long size;
1619 enum object_type type;
1620} delta_base_cache[MAX_DELTA_CACHE];
1621
1622static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1623{
1624 unsigned long hash;
1625
1626 hash = (unsigned long)p + (unsigned long)base_offset;
1627 hash += (hash >> 8) + (hash >> 16);
1628 return hash % MAX_DELTA_CACHE;
1629}
1630
1631static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1632 unsigned long *base_size, enum object_type *type, int keep_cache)
1633{
1634 void *ret;
1635 unsigned long hash = pack_entry_hash(p, base_offset);
1636 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1637
1638 ret = ent->data;
1639 if (!ret || ent->p != p || ent->base_offset != base_offset)
1640 return unpack_entry(p, base_offset, type, base_size);
1641
1642 if (!keep_cache) {
1643 ent->data = NULL;
1644 ent->lru.next->prev = ent->lru.prev;
1645 ent->lru.prev->next = ent->lru.next;
1646 delta_base_cached -= ent->size;
1647 } else {
1648 ret = xmemdupz(ent->data, ent->size);
1649 }
1650 *type = ent->type;
1651 *base_size = ent->size;
1652 return ret;
1653}
1654
1655static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1656{
1657 if (ent->data) {
1658 free(ent->data);
1659 ent->data = NULL;
1660 ent->lru.next->prev = ent->lru.prev;
1661 ent->lru.prev->next = ent->lru.next;
1662 delta_base_cached -= ent->size;
1663 }
1664}
1665
1666static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1667 void *base, unsigned long base_size, enum object_type type)
1668{
1669 unsigned long hash = pack_entry_hash(p, base_offset);
1670 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1671 struct delta_base_cache_lru_list *lru;
1672
1673 release_delta_base_cache(ent);
1674 delta_base_cached += base_size;
1675
1676 for (lru = delta_base_cache_lru.next;
1677 delta_base_cached > delta_base_cache_limit
1678 && lru != &delta_base_cache_lru;
1679 lru = lru->next) {
1680 struct delta_base_cache_entry *f = (void *)lru;
1681 if (f->type == OBJ_BLOB)
1682 release_delta_base_cache(f);
1683 }
1684 for (lru = delta_base_cache_lru.next;
1685 delta_base_cached > delta_base_cache_limit
1686 && lru != &delta_base_cache_lru;
1687 lru = lru->next) {
1688 struct delta_base_cache_entry *f = (void *)lru;
1689 release_delta_base_cache(f);
1690 }
1691
1692 ent->p = p;
1693 ent->base_offset = base_offset;
1694 ent->type = type;
1695 ent->data = base;
1696 ent->size = base_size;
1697 ent->lru.next = &delta_base_cache_lru;
1698 ent->lru.prev = delta_base_cache_lru.prev;
1699 delta_base_cache_lru.prev->next = &ent->lru;
1700 delta_base_cache_lru.prev = &ent->lru;
1701}
1702
1703static void *unpack_delta_entry(struct packed_git *p,
1704 struct pack_window **w_curs,
1705 off_t curpos,
1706 unsigned long delta_size,
1707 off_t obj_offset,
1708 enum object_type *type,
1709 unsigned long *sizep)
1710{
1711 void *delta_data, *result, *base;
1712 unsigned long base_size;
1713 off_t base_offset;
1714
1715 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1716 if (!base_offset) {
1717 error("failed to validate delta base reference "
1718 "at offset %"PRIuMAX" from %s",
1719 (uintmax_t)curpos, p->pack_name);
1720 return NULL;
1721 }
1722 unuse_pack(w_curs);
1723 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1724 if (!base) {
1725 /*
1726 * We're probably in deep shit, but let's try to fetch
1727 * the required base anyway from another pack or loose.
1728 * This is costly but should happen only in the presence
1729 * of a corrupted pack, and is better than failing outright.
1730 */
1731 struct revindex_entry *revidx;
1732 const unsigned char *base_sha1;
1733 revidx = find_pack_revindex(p, base_offset);
1734 if (!revidx)
1735 return NULL;
1736 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1737 error("failed to read delta base object %s"
1738 " at offset %"PRIuMAX" from %s",
1739 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1740 p->pack_name);
1741 mark_bad_packed_object(p, base_sha1);
1742 base = read_object(base_sha1, type, &base_size);
1743 if (!base)
1744 return NULL;
1745 }
1746
1747 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1748 if (!delta_data) {
1749 error("failed to unpack compressed delta "
1750 "at offset %"PRIuMAX" from %s",
1751 (uintmax_t)curpos, p->pack_name);
1752 free(base);
1753 return NULL;
1754 }
1755 result = patch_delta(base, base_size,
1756 delta_data, delta_size,
1757 sizep);
1758 if (!result)
1759 die("failed to apply delta");
1760 free(delta_data);
1761 add_delta_base_cache(p, base_offset, base, base_size, *type);
1762 return result;
1763}
1764
1765int do_check_packed_object_crc;
1766
1767void *unpack_entry(struct packed_git *p, off_t obj_offset,
1768 enum object_type *type, unsigned long *sizep)
1769{
1770 struct pack_window *w_curs = NULL;
1771 off_t curpos = obj_offset;
1772 void *data;
1773
1774 if (do_check_packed_object_crc && p->index_version > 1) {
1775 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1776 unsigned long len = revidx[1].offset - obj_offset;
1777 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1778 const unsigned char *sha1 =
1779 nth_packed_object_sha1(p, revidx->nr);
1780 error("bad packed object CRC for %s",
1781 sha1_to_hex(sha1));
1782 mark_bad_packed_object(p, sha1);
1783 unuse_pack(&w_curs);
1784 return NULL;
1785 }
1786 }
1787
1788 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1789 switch (*type) {
1790 case OBJ_OFS_DELTA:
1791 case OBJ_REF_DELTA:
1792 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1793 obj_offset, type, sizep);
1794 break;
1795 case OBJ_COMMIT:
1796 case OBJ_TREE:
1797 case OBJ_BLOB:
1798 case OBJ_TAG:
1799 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1800 break;
1801 default:
1802 data = NULL;
1803 error("unknown object type %i at offset %"PRIuMAX" in %s",
1804 *type, (uintmax_t)obj_offset, p->pack_name);
1805 }
1806 unuse_pack(&w_curs);
1807 return data;
1808}
1809
1810const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1811 uint32_t n)
1812{
1813 const unsigned char *index = p->index_data;
1814 if (!index) {
1815 if (open_pack_index(p))
1816 return NULL;
1817 index = p->index_data;
1818 }
1819 if (n >= p->num_objects)
1820 return NULL;
1821 index += 4 * 256;
1822 if (p->index_version == 1) {
1823 return index + 24 * n + 4;
1824 } else {
1825 index += 8;
1826 return index + 20 * n;
1827 }
1828}
1829
1830off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1831{
1832 const unsigned char *index = p->index_data;
1833 index += 4 * 256;
1834 if (p->index_version == 1) {
1835 return ntohl(*((uint32_t *)(index + 24 * n)));
1836 } else {
1837 uint32_t off;
1838 index += 8 + p->num_objects * (20 + 4);
1839 off = ntohl(*((uint32_t *)(index + 4 * n)));
1840 if (!(off & 0x80000000))
1841 return off;
1842 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1843 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1844 ntohl(*((uint32_t *)(index + 4)));
1845 }
1846}
1847
1848off_t find_pack_entry_one(const unsigned char *sha1,
1849 struct packed_git *p)
1850{
1851 const uint32_t *level1_ofs = p->index_data;
1852 const unsigned char *index = p->index_data;
1853 unsigned hi, lo, stride;
1854 static int use_lookup = -1;
1855 static int debug_lookup = -1;
1856
1857 if (debug_lookup < 0)
1858 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1859
1860 if (!index) {
1861 if (open_pack_index(p))
1862 return 0;
1863 level1_ofs = p->index_data;
1864 index = p->index_data;
1865 }
1866 if (p->index_version > 1) {
1867 level1_ofs += 2;
1868 index += 8;
1869 }
1870 index += 4 * 256;
1871 hi = ntohl(level1_ofs[*sha1]);
1872 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1873 if (p->index_version > 1) {
1874 stride = 20;
1875 } else {
1876 stride = 24;
1877 index += 4;
1878 }
1879
1880 if (debug_lookup)
1881 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1882 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1883
1884 if (use_lookup < 0)
1885 use_lookup = !!getenv("GIT_USE_LOOKUP");
1886 if (use_lookup) {
1887 int pos = sha1_entry_pos(index, stride, 0,
1888 lo, hi, p->num_objects, sha1);
1889 if (pos < 0)
1890 return 0;
1891 return nth_packed_object_offset(p, pos);
1892 }
1893
1894 do {
1895 unsigned mi = (lo + hi) / 2;
1896 int cmp = hashcmp(index + mi * stride, sha1);
1897
1898 if (debug_lookup)
1899 printf("lo %u hi %u rg %u mi %u\n",
1900 lo, hi, hi - lo, mi);
1901 if (!cmp)
1902 return nth_packed_object_offset(p, mi);
1903 if (cmp > 0)
1904 hi = mi;
1905 else
1906 lo = mi+1;
1907 } while (lo < hi);
1908 return 0;
1909}
1910
1911int matches_pack_name(struct packed_git *p, const char *name)
1912{
1913 const char *last_c, *c;
1914
1915 if (!strcmp(p->pack_name, name))
1916 return 1;
1917
1918 for (c = p->pack_name, last_c = c; *c;)
1919 if (*c == '/')
1920 last_c = ++c;
1921 else
1922 ++c;
1923 if (!strcmp(last_c, name))
1924 return 1;
1925
1926 return 0;
1927}
1928
1929static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1930{
1931 static struct packed_git *last_found = (void *)1;
1932 struct packed_git *p;
1933 off_t offset;
1934
1935 prepare_packed_git();
1936 if (!packed_git)
1937 return 0;
1938 p = (last_found == (void *)1) ? packed_git : last_found;
1939
1940 do {
1941 if (ignore_packed) {
1942 const char **ig;
1943 for (ig = ignore_packed; *ig; ig++)
1944 if (matches_pack_name(p, *ig))
1945 break;
1946 if (*ig)
1947 goto next;
1948 }
1949
1950 if (p->num_bad_objects) {
1951 unsigned i;
1952 for (i = 0; i < p->num_bad_objects; i++)
1953 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1954 goto next;
1955 }
1956
1957 offset = find_pack_entry_one(sha1, p);
1958 if (offset) {
1959 /*
1960 * We are about to tell the caller where they can
1961 * locate the requested object. We better make
1962 * sure the packfile is still here and can be
1963 * accessed before supplying that answer, as
1964 * it may have been deleted since the index
1965 * was loaded!
1966 */
1967 if (p->pack_fd == -1 && open_packed_git(p)) {
1968 error("packfile %s cannot be accessed", p->pack_name);
1969 goto next;
1970 }
1971 e->offset = offset;
1972 e->p = p;
1973 hashcpy(e->sha1, sha1);
1974 last_found = p;
1975 return 1;
1976 }
1977
1978 next:
1979 if (p == last_found)
1980 p = packed_git;
1981 else
1982 p = p->next;
1983 if (p == last_found)
1984 p = p->next;
1985 } while (p);
1986 return 0;
1987}
1988
1989struct packed_git *find_sha1_pack(const unsigned char *sha1,
1990 struct packed_git *packs)
1991{
1992 struct packed_git *p;
1993
1994 for (p = packs; p; p = p->next) {
1995 if (find_pack_entry_one(sha1, p))
1996 return p;
1997 }
1998 return NULL;
1999
2000}
2001
2002static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2003{
2004 int status;
2005 unsigned long mapsize, size;
2006 void *map;
2007 z_stream stream;
2008 char hdr[32];
2009
2010 map = map_sha1_file(sha1, &mapsize);
2011 if (!map)
2012 return error("unable to find %s", sha1_to_hex(sha1));
2013 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2014 status = error("unable to unpack %s header",
2015 sha1_to_hex(sha1));
2016 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2017 status = error("unable to parse %s header", sha1_to_hex(sha1));
2018 else if (sizep)
2019 *sizep = size;
2020 inflateEnd(&stream);
2021 munmap(map, mapsize);
2022 return status;
2023}
2024
2025int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2026{
2027 struct pack_entry e;
2028 int status;
2029
2030 if (!find_pack_entry(sha1, &e, NULL)) {
2031 /* Most likely it's a loose object. */
2032 status = sha1_loose_object_info(sha1, sizep);
2033 if (status >= 0)
2034 return status;
2035
2036 /* Not a loose object; someone else may have just packed it. */
2037 reprepare_packed_git();
2038 if (!find_pack_entry(sha1, &e, NULL))
2039 return status;
2040 }
2041
2042 status = packed_object_info(e.p, e.offset, sizep);
2043 if (status < 0) {
2044 mark_bad_packed_object(e.p, sha1);
2045 status = sha1_object_info(sha1, sizep);
2046 }
2047
2048 return status;
2049}
2050
2051static void *read_packed_sha1(const unsigned char *sha1,
2052 enum object_type *type, unsigned long *size)
2053{
2054 struct pack_entry e;
2055 void *data;
2056
2057 if (!find_pack_entry(sha1, &e, NULL))
2058 return NULL;
2059 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2060 if (!data) {
2061 /*
2062 * We're probably in deep shit, but let's try to fetch
2063 * the required object anyway from another pack or loose.
2064 * This should happen only in the presence of a corrupted
2065 * pack, and is better than failing outright.
2066 */
2067 error("failed to read object %s at offset %"PRIuMAX" from %s",
2068 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2069 mark_bad_packed_object(e.p, sha1);
2070 data = read_object(sha1, type, size);
2071 }
2072 return data;
2073}
2074
2075/*
2076 * This is meant to hold a *small* number of objects that you would
2077 * want read_sha1_file() to be able to return, but yet you do not want
2078 * to write them into the object store (e.g. a browse-only
2079 * application).
2080 */
2081static struct cached_object {
2082 unsigned char sha1[20];
2083 enum object_type type;
2084 void *buf;
2085 unsigned long size;
2086} *cached_objects;
2087static int cached_object_nr, cached_object_alloc;
2088
2089static struct cached_object empty_tree = {
2090 EMPTY_TREE_SHA1_BIN,
2091 OBJ_TREE,
2092 "",
2093 0
2094};
2095
2096static struct cached_object *find_cached_object(const unsigned char *sha1)
2097{
2098 int i;
2099 struct cached_object *co = cached_objects;
2100
2101 for (i = 0; i < cached_object_nr; i++, co++) {
2102 if (!hashcmp(co->sha1, sha1))
2103 return co;
2104 }
2105 if (!hashcmp(sha1, empty_tree.sha1))
2106 return &empty_tree;
2107 return NULL;
2108}
2109
2110int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2111 unsigned char *sha1)
2112{
2113 struct cached_object *co;
2114
2115 hash_sha1_file(buf, len, typename(type), sha1);
2116 if (has_sha1_file(sha1) || find_cached_object(sha1))
2117 return 0;
2118 if (cached_object_alloc <= cached_object_nr) {
2119 cached_object_alloc = alloc_nr(cached_object_alloc);
2120 cached_objects = xrealloc(cached_objects,
2121 sizeof(*cached_objects) *
2122 cached_object_alloc);
2123 }
2124 co = &cached_objects[cached_object_nr++];
2125 co->size = len;
2126 co->type = type;
2127 co->buf = xmalloc(len);
2128 memcpy(co->buf, buf, len);
2129 hashcpy(co->sha1, sha1);
2130 return 0;
2131}
2132
2133void *read_object(const unsigned char *sha1, enum object_type *type,
2134 unsigned long *size)
2135{
2136 unsigned long mapsize;
2137 void *map, *buf;
2138 struct cached_object *co;
2139
2140 co = find_cached_object(sha1);
2141 if (co) {
2142 *type = co->type;
2143 *size = co->size;
2144 return xmemdupz(co->buf, co->size);
2145 }
2146
2147 buf = read_packed_sha1(sha1, type, size);
2148 if (buf)
2149 return buf;
2150 map = map_sha1_file(sha1, &mapsize);
2151 if (map) {
2152 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2153 munmap(map, mapsize);
2154 return buf;
2155 }
2156 reprepare_packed_git();
2157 return read_packed_sha1(sha1, type, size);
2158}
2159
2160void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2161 unsigned long *size)
2162{
2163 void *data = read_object(sha1, type, size);
2164 /* legacy behavior is to die on corrupted objects */
2165 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2166 die("object %s is corrupted", sha1_to_hex(sha1));
2167 return data;
2168}
2169
2170void *read_object_with_reference(const unsigned char *sha1,
2171 const char *required_type_name,
2172 unsigned long *size,
2173 unsigned char *actual_sha1_return)
2174{
2175 enum object_type type, required_type;
2176 void *buffer;
2177 unsigned long isize;
2178 unsigned char actual_sha1[20];
2179
2180 required_type = type_from_string(required_type_name);
2181 hashcpy(actual_sha1, sha1);
2182 while (1) {
2183 int ref_length = -1;
2184 const char *ref_type = NULL;
2185
2186 buffer = read_sha1_file(actual_sha1, &type, &isize);
2187 if (!buffer)
2188 return NULL;
2189 if (type == required_type) {
2190 *size = isize;
2191 if (actual_sha1_return)
2192 hashcpy(actual_sha1_return, actual_sha1);
2193 return buffer;
2194 }
2195 /* Handle references */
2196 else if (type == OBJ_COMMIT)
2197 ref_type = "tree ";
2198 else if (type == OBJ_TAG)
2199 ref_type = "object ";
2200 else {
2201 free(buffer);
2202 return NULL;
2203 }
2204 ref_length = strlen(ref_type);
2205
2206 if (ref_length + 40 > isize ||
2207 memcmp(buffer, ref_type, ref_length) ||
2208 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2209 free(buffer);
2210 return NULL;
2211 }
2212 free(buffer);
2213 /* Now we have the ID of the referred-to object in
2214 * actual_sha1. Check again. */
2215 }
2216}
2217
2218static void write_sha1_file_prepare(const void *buf, unsigned long len,
2219 const char *type, unsigned char *sha1,
2220 char *hdr, int *hdrlen)
2221{
2222 git_SHA_CTX c;
2223
2224 /* Generate the header */
2225 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2226
2227 /* Sha1.. */
2228 git_SHA1_Init(&c);
2229 git_SHA1_Update(&c, hdr, *hdrlen);
2230 git_SHA1_Update(&c, buf, len);
2231 git_SHA1_Final(sha1, &c);
2232}
2233
2234/*
2235 * Move the just written object into its final resting place
2236 */
2237int move_temp_to_file(const char *tmpfile, const char *filename)
2238{
2239 int ret = 0;
2240 if (link(tmpfile, filename))
2241 ret = errno;
2242
2243 /*
2244 * Coda hack - coda doesn't like cross-directory links,
2245 * so we fall back to a rename, which will mean that it
2246 * won't be able to check collisions, but that's not a
2247 * big deal.
2248 *
2249 * The same holds for FAT formatted media.
2250 *
2251 * When this succeeds, we just return 0. We have nothing
2252 * left to unlink.
2253 */
2254 if (ret && ret != EEXIST) {
2255 if (!rename(tmpfile, filename))
2256 return 0;
2257 ret = errno;
2258 }
2259 unlink(tmpfile);
2260 if (ret) {
2261 if (ret != EEXIST) {
2262 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2263 }
2264 /* FIXME!!! Collision check here ? */
2265 }
2266
2267 return 0;
2268}
2269
2270static int write_buffer(int fd, const void *buf, size_t len)
2271{
2272 if (write_in_full(fd, buf, len) < 0)
2273 return error("file write error (%s)", strerror(errno));
2274 return 0;
2275}
2276
2277int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2278 unsigned char *sha1)
2279{
2280 char hdr[32];
2281 int hdrlen;
2282 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2283 return 0;
2284}
2285
2286/* Finalize a file on disk, and close it. */
2287static void close_sha1_file(int fd)
2288{
2289 if (fsync_object_files)
2290 fsync_or_die(fd, "sha1 file");
2291 fchmod(fd, 0444);
2292 if (close(fd) != 0)
2293 die("unable to write sha1 file");
2294}
2295
2296/* Size of directory component, including the ending '/' */
2297static inline int directory_size(const char *filename)
2298{
2299 const char *s = strrchr(filename, '/');
2300 if (!s)
2301 return 0;
2302 return s - filename + 1;
2303}
2304
2305/*
2306 * This creates a temporary file in the same directory as the final
2307 * 'filename'
2308 *
2309 * We want to avoid cross-directory filename renames, because those
2310 * can have problems on various filesystems (FAT, NFS, Coda).
2311 */
2312static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2313{
2314 int fd, dirlen = directory_size(filename);
2315
2316 if (dirlen + 20 > bufsiz) {
2317 errno = ENAMETOOLONG;
2318 return -1;
2319 }
2320 memcpy(buffer, filename, dirlen);
2321 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2322 fd = mkstemp(buffer);
2323 if (fd < 0 && dirlen && errno == ENOENT) {
2324 /* Make sure the directory exists */
2325 memcpy(buffer, filename, dirlen);
2326 buffer[dirlen-1] = 0;
2327 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2328 return -1;
2329
2330 /* Try again */
2331 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2332 fd = mkstemp(buffer);
2333 }
2334 return fd;
2335}
2336
2337static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2338 void *buf, unsigned long len, time_t mtime)
2339{
2340 int fd, size, ret;
2341 unsigned char *compressed;
2342 z_stream stream;
2343 char *filename;
2344 static char tmpfile[PATH_MAX];
2345
2346 filename = sha1_file_name(sha1);
2347 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2348 if (fd < 0) {
2349 if (errno == EACCES)
2350 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2351 else
2352 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2353 }
2354
2355 /* Set it up */
2356 memset(&stream, 0, sizeof(stream));
2357 deflateInit(&stream, zlib_compression_level);
2358 size = 8 + deflateBound(&stream, len+hdrlen);
2359 compressed = xmalloc(size);
2360
2361 /* Compress it */
2362 stream.next_out = compressed;
2363 stream.avail_out = size;
2364
2365 /* First header.. */
2366 stream.next_in = (unsigned char *)hdr;
2367 stream.avail_in = hdrlen;
2368 while (deflate(&stream, 0) == Z_OK)
2369 /* nothing */;
2370
2371 /* Then the data itself.. */
2372 stream.next_in = buf;
2373 stream.avail_in = len;
2374 ret = deflate(&stream, Z_FINISH);
2375 if (ret != Z_STREAM_END)
2376 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2377
2378 ret = deflateEnd(&stream);
2379 if (ret != Z_OK)
2380 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2381
2382 size = stream.total_out;
2383
2384 if (write_buffer(fd, compressed, size) < 0)
2385 die("unable to write sha1 file");
2386 close_sha1_file(fd);
2387 free(compressed);
2388
2389 if (mtime) {
2390 struct utimbuf utb;
2391 utb.actime = mtime;
2392 utb.modtime = mtime;
2393 if (utime(tmpfile, &utb) < 0)
2394 warning("failed utime() on %s: %s",
2395 tmpfile, strerror(errno));
2396 }
2397
2398 return move_temp_to_file(tmpfile, filename);
2399}
2400
2401int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2402{
2403 unsigned char sha1[20];
2404 char hdr[32];
2405 int hdrlen;
2406
2407 /* Normally if we have it in the pack then we do not bother writing
2408 * it out into .git/objects/??/?{38} file.
2409 */
2410 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2411 if (returnsha1)
2412 hashcpy(returnsha1, sha1);
2413 if (has_sha1_file(sha1))
2414 return 0;
2415 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2416}
2417
2418int force_object_loose(const unsigned char *sha1, time_t mtime)
2419{
2420 void *buf;
2421 unsigned long len;
2422 enum object_type type;
2423 char hdr[32];
2424 int hdrlen;
2425 int ret;
2426
2427 if (has_loose_object(sha1))
2428 return 0;
2429 buf = read_packed_sha1(sha1, &type, &len);
2430 if (!buf)
2431 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2432 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2433 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2434 free(buf);
2435
2436 return ret;
2437}
2438
2439int has_pack_index(const unsigned char *sha1)
2440{
2441 struct stat st;
2442 if (stat(sha1_pack_index_name(sha1), &st))
2443 return 0;
2444 return 1;
2445}
2446
2447int has_pack_file(const unsigned char *sha1)
2448{
2449 struct stat st;
2450 if (stat(sha1_pack_name(sha1), &st))
2451 return 0;
2452 return 1;
2453}
2454
2455int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2456{
2457 struct pack_entry e;
2458 return find_pack_entry(sha1, &e, ignore_packed);
2459}
2460
2461int has_sha1_file(const unsigned char *sha1)
2462{
2463 struct pack_entry e;
2464
2465 if (find_pack_entry(sha1, &e, NULL))
2466 return 1;
2467 return has_loose_object(sha1);
2468}
2469
2470static int index_mem(unsigned char *sha1, void *buf, size_t size,
2471 int write_object, enum object_type type, const char *path)
2472{
2473 int ret, re_allocated = 0;
2474
2475 if (!type)
2476 type = OBJ_BLOB;
2477
2478 /*
2479 * Convert blobs to git internal format
2480 */
2481 if ((type == OBJ_BLOB) && path) {
2482 struct strbuf nbuf = STRBUF_INIT;
2483 if (convert_to_git(path, buf, size, &nbuf,
2484 write_object ? safe_crlf : 0)) {
2485 buf = strbuf_detach(&nbuf, &size);
2486 re_allocated = 1;
2487 }
2488 }
2489
2490 if (write_object)
2491 ret = write_sha1_file(buf, size, typename(type), sha1);
2492 else
2493 ret = hash_sha1_file(buf, size, typename(type), sha1);
2494 if (re_allocated)
2495 free(buf);
2496 return ret;
2497}
2498
2499int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2500 enum object_type type, const char *path)
2501{
2502 int ret;
2503 size_t size = xsize_t(st->st_size);
2504
2505 if (!S_ISREG(st->st_mode)) {
2506 struct strbuf sbuf = STRBUF_INIT;
2507 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2508 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2509 type, path);
2510 else
2511 ret = -1;
2512 strbuf_release(&sbuf);
2513 } else if (size) {
2514 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2515 ret = index_mem(sha1, buf, size, write_object, type, path);
2516 munmap(buf, size);
2517 } else
2518 ret = index_mem(sha1, NULL, size, write_object, type, path);
2519 close(fd);
2520 return ret;
2521}
2522
2523int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2524{
2525 int fd;
2526 struct strbuf sb = STRBUF_INIT;
2527
2528 switch (st->st_mode & S_IFMT) {
2529 case S_IFREG:
2530 fd = open(path, O_RDONLY);
2531 if (fd < 0)
2532 return error("open(\"%s\"): %s", path,
2533 strerror(errno));
2534 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2535 return error("%s: failed to insert into database",
2536 path);
2537 break;
2538 case S_IFLNK:
2539 if (strbuf_readlink(&sb, path, st->st_size)) {
2540 char *errstr = strerror(errno);
2541 return error("readlink(\"%s\"): %s", path,
2542 errstr);
2543 }
2544 if (!write_object)
2545 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2546 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2547 return error("%s: failed to insert into database",
2548 path);
2549 strbuf_release(&sb);
2550 break;
2551 case S_IFDIR:
2552 return resolve_gitlink_ref(path, "HEAD", sha1);
2553 default:
2554 return error("%s: unsupported file type", path);
2555 }
2556 return 0;
2557}
2558
2559int read_pack_header(int fd, struct pack_header *header)
2560{
2561 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2562 /* "eof before pack header was fully read" */
2563 return PH_ERROR_EOF;
2564
2565 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2566 /* "protocol error (pack signature mismatch detected)" */
2567 return PH_ERROR_PACK_SIGNATURE;
2568 if (!pack_version_ok(header->hdr_version))
2569 /* "protocol error (pack version unsupported)" */
2570 return PH_ERROR_PROTOCOL;
2571 return 0;
2572}