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 clear_delta_base_cache();
693 close_pack_windows(p);
694 if (p->pack_fd != -1)
695 close(p->pack_fd);
696 if (p->index_data)
697 munmap((void *)p->index_data, p->index_size);
698 free(p->bad_object_sha1);
699 *pp = p->next;
700 free(p);
701 return;
702 }
703 pp = &p->next;
704 }
705}
706
707/*
708 * Do not call this directly as this leaks p->pack_fd on error return;
709 * call open_packed_git() instead.
710 */
711static int open_packed_git_1(struct packed_git *p)
712{
713 struct stat st;
714 struct pack_header hdr;
715 unsigned char sha1[20];
716 unsigned char *idx_sha1;
717 long fd_flag;
718
719 if (!p->index_data && open_pack_index(p))
720 return error("packfile %s index unavailable", p->pack_name);
721
722 p->pack_fd = open(p->pack_name, O_RDONLY);
723 while (p->pack_fd < 0 && errno == EMFILE && unuse_one_window(p, -1))
724 p->pack_fd = open(p->pack_name, O_RDONLY);
725 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
726 return -1;
727
728 /* If we created the struct before we had the pack we lack size. */
729 if (!p->pack_size) {
730 if (!S_ISREG(st.st_mode))
731 return error("packfile %s not a regular file", p->pack_name);
732 p->pack_size = st.st_size;
733 } else if (p->pack_size != st.st_size)
734 return error("packfile %s size changed", p->pack_name);
735
736 /* We leave these file descriptors open with sliding mmap;
737 * there is no point keeping them open across exec(), though.
738 */
739 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
740 if (fd_flag < 0)
741 return error("cannot determine file descriptor flags");
742 fd_flag |= FD_CLOEXEC;
743 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
744 return error("cannot set FD_CLOEXEC");
745
746 /* Verify we recognize this pack file format. */
747 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
748 return error("file %s is far too short to be a packfile", p->pack_name);
749 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
750 return error("file %s is not a GIT packfile", p->pack_name);
751 if (!pack_version_ok(hdr.hdr_version))
752 return error("packfile %s is version %"PRIu32" and not"
753 " supported (try upgrading GIT to a newer version)",
754 p->pack_name, ntohl(hdr.hdr_version));
755
756 /* Verify the pack matches its index. */
757 if (p->num_objects != ntohl(hdr.hdr_entries))
758 return error("packfile %s claims to have %"PRIu32" objects"
759 " while index indicates %"PRIu32" objects",
760 p->pack_name, ntohl(hdr.hdr_entries),
761 p->num_objects);
762 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
763 return error("end of packfile %s is unavailable", p->pack_name);
764 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
765 return error("packfile %s signature is unavailable", p->pack_name);
766 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
767 if (hashcmp(sha1, idx_sha1))
768 return error("packfile %s does not match index", p->pack_name);
769 return 0;
770}
771
772static int open_packed_git(struct packed_git *p)
773{
774 if (!open_packed_git_1(p))
775 return 0;
776 if (p->pack_fd != -1) {
777 close(p->pack_fd);
778 p->pack_fd = -1;
779 }
780 return -1;
781}
782
783static int in_window(struct pack_window *win, off_t offset)
784{
785 /* We must promise at least 20 bytes (one hash) after the
786 * offset is available from this window, otherwise the offset
787 * is not actually in this window and a different window (which
788 * has that one hash excess) must be used. This is to support
789 * the object header and delta base parsing routines below.
790 */
791 off_t win_off = win->offset;
792 return win_off <= offset
793 && (offset + 20) <= (win_off + win->len);
794}
795
796unsigned char *use_pack(struct packed_git *p,
797 struct pack_window **w_cursor,
798 off_t offset,
799 unsigned int *left)
800{
801 struct pack_window *win = *w_cursor;
802
803 if (p->pack_fd == -1 && open_packed_git(p))
804 die("packfile %s cannot be accessed", p->pack_name);
805
806 /* Since packfiles end in a hash of their content and it's
807 * pointless to ask for an offset into the middle of that
808 * hash, and the in_window function above wouldn't match
809 * don't allow an offset too close to the end of the file.
810 */
811 if (offset > (p->pack_size - 20))
812 die("offset beyond end of packfile (truncated pack?)");
813
814 if (!win || !in_window(win, offset)) {
815 if (win)
816 win->inuse_cnt--;
817 for (win = p->windows; win; win = win->next) {
818 if (in_window(win, offset))
819 break;
820 }
821 if (!win) {
822 size_t window_align = packed_git_window_size / 2;
823 off_t len;
824 win = xcalloc(1, sizeof(*win));
825 win->offset = (offset / window_align) * window_align;
826 len = p->pack_size - win->offset;
827 if (len > packed_git_window_size)
828 len = packed_git_window_size;
829 win->len = (size_t)len;
830 pack_mapped += win->len;
831 while (packed_git_limit < pack_mapped
832 && unuse_one_window(p, p->pack_fd))
833 ; /* nothing */
834 win->base = xmmap(NULL, win->len,
835 PROT_READ, MAP_PRIVATE,
836 p->pack_fd, win->offset);
837 if (win->base == MAP_FAILED)
838 die("packfile %s cannot be mapped: %s",
839 p->pack_name,
840 strerror(errno));
841 pack_mmap_calls++;
842 pack_open_windows++;
843 if (pack_mapped > peak_pack_mapped)
844 peak_pack_mapped = pack_mapped;
845 if (pack_open_windows > peak_pack_open_windows)
846 peak_pack_open_windows = pack_open_windows;
847 win->next = p->windows;
848 p->windows = win;
849 }
850 }
851 if (win != *w_cursor) {
852 win->last_used = pack_used_ctr++;
853 win->inuse_cnt++;
854 *w_cursor = win;
855 }
856 offset -= win->offset;
857 if (left)
858 *left = win->len - xsize_t(offset);
859 return win->base + offset;
860}
861
862static struct packed_git *alloc_packed_git(int extra)
863{
864 struct packed_git *p = xmalloc(sizeof(*p) + extra);
865 memset(p, 0, sizeof(*p));
866 p->pack_fd = -1;
867 return p;
868}
869
870struct packed_git *add_packed_git(const char *path, int path_len, int local)
871{
872 struct stat st;
873 struct packed_git *p = alloc_packed_git(path_len + 2);
874
875 /*
876 * Make sure a corresponding .pack file exists and that
877 * the index looks sane.
878 */
879 path_len -= strlen(".idx");
880 if (path_len < 1) {
881 free(p);
882 return NULL;
883 }
884 memcpy(p->pack_name, path, path_len);
885
886 strcpy(p->pack_name + path_len, ".keep");
887 if (!access(p->pack_name, F_OK))
888 p->pack_keep = 1;
889
890 strcpy(p->pack_name + path_len, ".pack");
891 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
892 free(p);
893 return NULL;
894 }
895
896 /* ok, it looks sane as far as we can check without
897 * actually mapping the pack file.
898 */
899 p->pack_size = st.st_size;
900 p->pack_local = local;
901 p->mtime = st.st_mtime;
902 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
903 hashclr(p->sha1);
904 return p;
905}
906
907struct packed_git *parse_pack_index(unsigned char *sha1)
908{
909 const char *idx_path = sha1_pack_index_name(sha1);
910 const char *path = sha1_pack_name(sha1);
911 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
912
913 strcpy(p->pack_name, path);
914 hashcpy(p->sha1, sha1);
915 if (check_packed_git_idx(idx_path, p)) {
916 free(p);
917 return NULL;
918 }
919
920 return p;
921}
922
923void install_packed_git(struct packed_git *pack)
924{
925 pack->next = packed_git;
926 packed_git = pack;
927}
928
929static void prepare_packed_git_one(char *objdir, int local)
930{
931 /* Ensure that this buffer is large enough so that we can
932 append "/pack/" without clobbering the stack even if
933 strlen(objdir) were PATH_MAX. */
934 char path[PATH_MAX + 1 + 4 + 1 + 1];
935 int len;
936 DIR *dir;
937 struct dirent *de;
938
939 sprintf(path, "%s/pack", objdir);
940 len = strlen(path);
941 dir = opendir(path);
942 while (!dir && errno == EMFILE && unuse_one_window(packed_git, -1))
943 dir = opendir(path);
944 if (!dir) {
945 if (errno != ENOENT)
946 error("unable to open object pack directory: %s: %s",
947 path, strerror(errno));
948 return;
949 }
950 path[len++] = '/';
951 while ((de = readdir(dir)) != NULL) {
952 int namelen = strlen(de->d_name);
953 struct packed_git *p;
954
955 if (!has_extension(de->d_name, ".idx"))
956 continue;
957
958 if (len + namelen + 1 > sizeof(path))
959 continue;
960
961 /* Don't reopen a pack we already have. */
962 strcpy(path + len, de->d_name);
963 for (p = packed_git; p; p = p->next) {
964 if (!memcmp(path, p->pack_name, len + namelen - 4))
965 break;
966 }
967 if (p)
968 continue;
969 /* See if it really is a valid .idx file with corresponding
970 * .pack file that we can map.
971 */
972 p = add_packed_git(path, len + namelen, local);
973 if (!p)
974 continue;
975 install_packed_git(p);
976 }
977 closedir(dir);
978}
979
980static int sort_pack(const void *a_, const void *b_)
981{
982 struct packed_git *a = *((struct packed_git **)a_);
983 struct packed_git *b = *((struct packed_git **)b_);
984 int st;
985
986 /*
987 * Local packs tend to contain objects specific to our
988 * variant of the project than remote ones. In addition,
989 * remote ones could be on a network mounted filesystem.
990 * Favor local ones for these reasons.
991 */
992 st = a->pack_local - b->pack_local;
993 if (st)
994 return -st;
995
996 /*
997 * Younger packs tend to contain more recent objects,
998 * and more recent objects tend to get accessed more
999 * often.
1000 */
1001 if (a->mtime < b->mtime)
1002 return 1;
1003 else if (a->mtime == b->mtime)
1004 return 0;
1005 return -1;
1006}
1007
1008static void rearrange_packed_git(void)
1009{
1010 struct packed_git **ary, *p;
1011 int i, n;
1012
1013 for (n = 0, p = packed_git; p; p = p->next)
1014 n++;
1015 if (n < 2)
1016 return;
1017
1018 /* prepare an array of packed_git for easier sorting */
1019 ary = xcalloc(n, sizeof(struct packed_git *));
1020 for (n = 0, p = packed_git; p; p = p->next)
1021 ary[n++] = p;
1022
1023 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1024
1025 /* link them back again */
1026 for (i = 0; i < n - 1; i++)
1027 ary[i]->next = ary[i + 1];
1028 ary[n - 1]->next = NULL;
1029 packed_git = ary[0];
1030
1031 free(ary);
1032}
1033
1034static int prepare_packed_git_run_once = 0;
1035void prepare_packed_git(void)
1036{
1037 struct alternate_object_database *alt;
1038
1039 if (prepare_packed_git_run_once)
1040 return;
1041 prepare_packed_git_one(get_object_directory(), 1);
1042 prepare_alt_odb();
1043 for (alt = alt_odb_list; alt; alt = alt->next) {
1044 alt->name[-1] = 0;
1045 prepare_packed_git_one(alt->base, 0);
1046 alt->name[-1] = '/';
1047 }
1048 rearrange_packed_git();
1049 prepare_packed_git_run_once = 1;
1050}
1051
1052void reprepare_packed_git(void)
1053{
1054 discard_revindex();
1055 prepare_packed_git_run_once = 0;
1056 prepare_packed_git();
1057}
1058
1059static void mark_bad_packed_object(struct packed_git *p,
1060 const unsigned char *sha1)
1061{
1062 unsigned i;
1063 for (i = 0; i < p->num_bad_objects; i++)
1064 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1065 return;
1066 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1067 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1068 p->num_bad_objects++;
1069}
1070
1071static int has_packed_and_bad(const unsigned char *sha1)
1072{
1073 struct packed_git *p;
1074 unsigned i;
1075
1076 for (p = packed_git; p; p = p->next)
1077 for (i = 0; i < p->num_bad_objects; i++)
1078 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1079 return 1;
1080 return 0;
1081}
1082
1083int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1084{
1085 unsigned char real_sha1[20];
1086 hash_sha1_file(map, size, type, real_sha1);
1087 return hashcmp(sha1, real_sha1) ? -1 : 0;
1088}
1089
1090static int git_open_noatime(const char *name)
1091{
1092 static int sha1_file_open_flag = O_NOATIME;
1093 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1094
1095 /* Might the failure be due to O_NOATIME? */
1096 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1097 fd = open(name, O_RDONLY);
1098 if (fd >= 0)
1099 sha1_file_open_flag = 0;
1100 }
1101 return fd;
1102}
1103
1104static int open_sha1_file(const unsigned char *sha1)
1105{
1106 int fd;
1107 char *name = sha1_file_name(sha1);
1108 struct alternate_object_database *alt;
1109
1110 fd = git_open_noatime(name);
1111 if (fd >= 0)
1112 return fd;
1113
1114 prepare_alt_odb();
1115 errno = ENOENT;
1116 for (alt = alt_odb_list; alt; alt = alt->next) {
1117 name = alt->name;
1118 fill_sha1_path(name, sha1);
1119 fd = git_open_noatime(alt->base);
1120 if (fd >= 0)
1121 return fd;
1122 }
1123 return -1;
1124}
1125
1126static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1127{
1128 void *map;
1129 int fd;
1130
1131 fd = open_sha1_file(sha1);
1132 map = NULL;
1133 if (fd >= 0) {
1134 struct stat st;
1135
1136 if (!fstat(fd, &st)) {
1137 *size = xsize_t(st.st_size);
1138 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1139 }
1140 close(fd);
1141 }
1142 return map;
1143}
1144
1145static int legacy_loose_object(unsigned char *map)
1146{
1147 unsigned int word;
1148
1149 /*
1150 * Is it a zlib-compressed buffer? If so, the first byte
1151 * must be 0x78 (15-bit window size, deflated), and the
1152 * first 16-bit word is evenly divisible by 31
1153 */
1154 word = (map[0] << 8) + map[1];
1155 if (map[0] == 0x78 && !(word % 31))
1156 return 1;
1157 else
1158 return 0;
1159}
1160
1161unsigned long unpack_object_header_buffer(const unsigned char *buf,
1162 unsigned long len, enum object_type *type, unsigned long *sizep)
1163{
1164 unsigned shift;
1165 unsigned long size, c;
1166 unsigned long used = 0;
1167
1168 c = buf[used++];
1169 *type = (c >> 4) & 7;
1170 size = c & 15;
1171 shift = 4;
1172 while (c & 0x80) {
1173 if (len <= used || bitsizeof(long) <= shift) {
1174 error("bad object header");
1175 return 0;
1176 }
1177 c = buf[used++];
1178 size += (c & 0x7f) << shift;
1179 shift += 7;
1180 }
1181 *sizep = size;
1182 return used;
1183}
1184
1185static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1186{
1187 unsigned long size, used;
1188 static const char valid_loose_object_type[8] = {
1189 0, /* OBJ_EXT */
1190 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1191 0, /* "delta" and others are invalid in a loose object */
1192 };
1193 enum object_type type;
1194
1195 /* Get the data stream */
1196 memset(stream, 0, sizeof(*stream));
1197 stream->next_in = map;
1198 stream->avail_in = mapsize;
1199 stream->next_out = buffer;
1200 stream->avail_out = bufsiz;
1201
1202 if (legacy_loose_object(map)) {
1203 git_inflate_init(stream);
1204 return git_inflate(stream, 0);
1205 }
1206
1207
1208 /*
1209 * There used to be a second loose object header format which
1210 * was meant to mimic the in-pack format, allowing for direct
1211 * copy of the object data. This format turned up not to be
1212 * really worth it and we don't write it any longer. But we
1213 * can still read it.
1214 */
1215 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1216 if (!used || !valid_loose_object_type[type])
1217 return -1;
1218 map += used;
1219 mapsize -= used;
1220
1221 /* Set up the stream for the rest.. */
1222 stream->next_in = map;
1223 stream->avail_in = mapsize;
1224 git_inflate_init(stream);
1225
1226 /* And generate the fake traditional header */
1227 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1228 typename(type), size);
1229 return 0;
1230}
1231
1232static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1233{
1234 int bytes = strlen(buffer) + 1;
1235 unsigned char *buf = xmallocz(size);
1236 unsigned long n;
1237 int status = Z_OK;
1238
1239 n = stream->total_out - bytes;
1240 if (n > size)
1241 n = size;
1242 memcpy(buf, (char *) buffer + bytes, n);
1243 bytes = n;
1244 if (bytes <= size) {
1245 /*
1246 * The above condition must be (bytes <= size), not
1247 * (bytes < size). In other words, even though we
1248 * expect no more output and set avail_out to zer0,
1249 * the input zlib stream may have bytes that express
1250 * "this concludes the stream", and we *do* want to
1251 * eat that input.
1252 *
1253 * Otherwise we would not be able to test that we
1254 * consumed all the input to reach the expected size;
1255 * we also want to check that zlib tells us that all
1256 * went well with status == Z_STREAM_END at the end.
1257 */
1258 stream->next_out = buf + bytes;
1259 stream->avail_out = size - bytes;
1260 while (status == Z_OK)
1261 status = git_inflate(stream, Z_FINISH);
1262 }
1263 if (status == Z_STREAM_END && !stream->avail_in) {
1264 git_inflate_end(stream);
1265 return buf;
1266 }
1267
1268 if (status < 0)
1269 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1270 else if (stream->avail_in)
1271 error("garbage at end of loose object '%s'",
1272 sha1_to_hex(sha1));
1273 free(buf);
1274 return NULL;
1275}
1276
1277/*
1278 * We used to just use "sscanf()", but that's actually way
1279 * too permissive for what we want to check. So do an anal
1280 * object header parse by hand.
1281 */
1282static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1283{
1284 char type[10];
1285 int i;
1286 unsigned long size;
1287
1288 /*
1289 * The type can be at most ten bytes (including the
1290 * terminating '\0' that we add), and is followed by
1291 * a space.
1292 */
1293 i = 0;
1294 for (;;) {
1295 char c = *hdr++;
1296 if (c == ' ')
1297 break;
1298 type[i++] = c;
1299 if (i >= sizeof(type))
1300 return -1;
1301 }
1302 type[i] = 0;
1303
1304 /*
1305 * The length must follow immediately, and be in canonical
1306 * decimal format (ie "010" is not valid).
1307 */
1308 size = *hdr++ - '0';
1309 if (size > 9)
1310 return -1;
1311 if (size) {
1312 for (;;) {
1313 unsigned long c = *hdr - '0';
1314 if (c > 9)
1315 break;
1316 hdr++;
1317 size = size * 10 + c;
1318 }
1319 }
1320 *sizep = size;
1321
1322 /*
1323 * The length must be followed by a zero byte
1324 */
1325 return *hdr ? -1 : type_from_string(type);
1326}
1327
1328static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1329{
1330 int ret;
1331 z_stream stream;
1332 char hdr[8192];
1333
1334 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1335 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1336 return NULL;
1337
1338 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1339}
1340
1341unsigned long get_size_from_delta(struct packed_git *p,
1342 struct pack_window **w_curs,
1343 off_t curpos)
1344{
1345 const unsigned char *data;
1346 unsigned char delta_head[20], *in;
1347 z_stream stream;
1348 int st;
1349
1350 memset(&stream, 0, sizeof(stream));
1351 stream.next_out = delta_head;
1352 stream.avail_out = sizeof(delta_head);
1353
1354 git_inflate_init(&stream);
1355 do {
1356 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1357 stream.next_in = in;
1358 st = git_inflate(&stream, Z_FINISH);
1359 curpos += stream.next_in - in;
1360 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1361 stream.total_out < sizeof(delta_head));
1362 git_inflate_end(&stream);
1363 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1364 error("delta data unpack-initial failed");
1365 return 0;
1366 }
1367
1368 /* Examine the initial part of the delta to figure out
1369 * the result size.
1370 */
1371 data = delta_head;
1372
1373 /* ignore base size */
1374 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1375
1376 /* Read the result size */
1377 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1378}
1379
1380static off_t get_delta_base(struct packed_git *p,
1381 struct pack_window **w_curs,
1382 off_t *curpos,
1383 enum object_type type,
1384 off_t delta_obj_offset)
1385{
1386 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1387 off_t base_offset;
1388
1389 /* use_pack() assured us we have [base_info, base_info + 20)
1390 * as a range that we can look at without walking off the
1391 * end of the mapped window. Its actually the hash size
1392 * that is assured. An OFS_DELTA longer than the hash size
1393 * is stupid, as then a REF_DELTA would be smaller to store.
1394 */
1395 if (type == OBJ_OFS_DELTA) {
1396 unsigned used = 0;
1397 unsigned char c = base_info[used++];
1398 base_offset = c & 127;
1399 while (c & 128) {
1400 base_offset += 1;
1401 if (!base_offset || MSB(base_offset, 7))
1402 return 0; /* overflow */
1403 c = base_info[used++];
1404 base_offset = (base_offset << 7) + (c & 127);
1405 }
1406 base_offset = delta_obj_offset - base_offset;
1407 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1408 return 0; /* out of bound */
1409 *curpos += used;
1410 } else if (type == OBJ_REF_DELTA) {
1411 /* The base entry _must_ be in the same pack */
1412 base_offset = find_pack_entry_one(base_info, p);
1413 *curpos += 20;
1414 } else
1415 die("I am totally screwed");
1416 return base_offset;
1417}
1418
1419/* forward declaration for a mutually recursive function */
1420static int packed_object_info(struct packed_git *p, off_t offset,
1421 unsigned long *sizep);
1422
1423static int packed_delta_info(struct packed_git *p,
1424 struct pack_window **w_curs,
1425 off_t curpos,
1426 enum object_type type,
1427 off_t obj_offset,
1428 unsigned long *sizep)
1429{
1430 off_t base_offset;
1431
1432 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1433 if (!base_offset)
1434 return OBJ_BAD;
1435 type = packed_object_info(p, base_offset, NULL);
1436 if (type <= OBJ_NONE) {
1437 struct revindex_entry *revidx;
1438 const unsigned char *base_sha1;
1439 revidx = find_pack_revindex(p, base_offset);
1440 if (!revidx)
1441 return OBJ_BAD;
1442 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1443 mark_bad_packed_object(p, base_sha1);
1444 type = sha1_object_info(base_sha1, NULL);
1445 if (type <= OBJ_NONE)
1446 return OBJ_BAD;
1447 }
1448
1449 /* We choose to only get the type of the base object and
1450 * ignore potentially corrupt pack file that expects the delta
1451 * based on a base with a wrong size. This saves tons of
1452 * inflate() calls.
1453 */
1454 if (sizep) {
1455 *sizep = get_size_from_delta(p, w_curs, curpos);
1456 if (*sizep == 0)
1457 type = OBJ_BAD;
1458 }
1459
1460 return type;
1461}
1462
1463static int unpack_object_header(struct packed_git *p,
1464 struct pack_window **w_curs,
1465 off_t *curpos,
1466 unsigned long *sizep)
1467{
1468 unsigned char *base;
1469 unsigned int left;
1470 unsigned long used;
1471 enum object_type type;
1472
1473 /* use_pack() assures us we have [base, base + 20) available
1474 * as a range that we can look at at. (Its actually the hash
1475 * size that is assured.) With our object header encoding
1476 * the maximum deflated object size is 2^137, which is just
1477 * insane, so we know won't exceed what we have been given.
1478 */
1479 base = use_pack(p, w_curs, *curpos, &left);
1480 used = unpack_object_header_buffer(base, left, &type, sizep);
1481 if (!used) {
1482 type = OBJ_BAD;
1483 } else
1484 *curpos += used;
1485
1486 return type;
1487}
1488
1489const char *packed_object_info_detail(struct packed_git *p,
1490 off_t obj_offset,
1491 unsigned long *size,
1492 unsigned long *store_size,
1493 unsigned int *delta_chain_length,
1494 unsigned char *base_sha1)
1495{
1496 struct pack_window *w_curs = NULL;
1497 off_t curpos;
1498 unsigned long dummy;
1499 unsigned char *next_sha1;
1500 enum object_type type;
1501 struct revindex_entry *revidx;
1502
1503 *delta_chain_length = 0;
1504 curpos = obj_offset;
1505 type = unpack_object_header(p, &w_curs, &curpos, size);
1506
1507 revidx = find_pack_revindex(p, obj_offset);
1508 *store_size = revidx[1].offset - obj_offset;
1509
1510 for (;;) {
1511 switch (type) {
1512 default:
1513 die("pack %s contains unknown object type %d",
1514 p->pack_name, type);
1515 case OBJ_COMMIT:
1516 case OBJ_TREE:
1517 case OBJ_BLOB:
1518 case OBJ_TAG:
1519 unuse_pack(&w_curs);
1520 return typename(type);
1521 case OBJ_OFS_DELTA:
1522 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1523 if (!obj_offset)
1524 die("pack %s contains bad delta base reference of type %s",
1525 p->pack_name, typename(type));
1526 if (*delta_chain_length == 0) {
1527 revidx = find_pack_revindex(p, obj_offset);
1528 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1529 }
1530 break;
1531 case OBJ_REF_DELTA:
1532 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1533 if (*delta_chain_length == 0)
1534 hashcpy(base_sha1, next_sha1);
1535 obj_offset = find_pack_entry_one(next_sha1, p);
1536 break;
1537 }
1538 (*delta_chain_length)++;
1539 curpos = obj_offset;
1540 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1541 }
1542}
1543
1544static int packed_object_info(struct packed_git *p, off_t obj_offset,
1545 unsigned long *sizep)
1546{
1547 struct pack_window *w_curs = NULL;
1548 unsigned long size;
1549 off_t curpos = obj_offset;
1550 enum object_type type;
1551
1552 type = unpack_object_header(p, &w_curs, &curpos, &size);
1553
1554 switch (type) {
1555 case OBJ_OFS_DELTA:
1556 case OBJ_REF_DELTA:
1557 type = packed_delta_info(p, &w_curs, curpos,
1558 type, obj_offset, sizep);
1559 break;
1560 case OBJ_COMMIT:
1561 case OBJ_TREE:
1562 case OBJ_BLOB:
1563 case OBJ_TAG:
1564 if (sizep)
1565 *sizep = size;
1566 break;
1567 default:
1568 error("unknown object type %i at offset %"PRIuMAX" in %s",
1569 type, (uintmax_t)obj_offset, p->pack_name);
1570 type = OBJ_BAD;
1571 }
1572 unuse_pack(&w_curs);
1573 return type;
1574}
1575
1576static void *unpack_compressed_entry(struct packed_git *p,
1577 struct pack_window **w_curs,
1578 off_t curpos,
1579 unsigned long size)
1580{
1581 int st;
1582 z_stream stream;
1583 unsigned char *buffer, *in;
1584
1585 buffer = xmallocz(size);
1586 memset(&stream, 0, sizeof(stream));
1587 stream.next_out = buffer;
1588 stream.avail_out = size + 1;
1589
1590 git_inflate_init(&stream);
1591 do {
1592 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1593 stream.next_in = in;
1594 st = git_inflate(&stream, Z_FINISH);
1595 if (!stream.avail_out)
1596 break; /* the payload is larger than it should be */
1597 curpos += stream.next_in - in;
1598 } while (st == Z_OK || st == Z_BUF_ERROR);
1599 git_inflate_end(&stream);
1600 if ((st != Z_STREAM_END) || stream.total_out != size) {
1601 free(buffer);
1602 return NULL;
1603 }
1604
1605 return buffer;
1606}
1607
1608#define MAX_DELTA_CACHE (256)
1609
1610static size_t delta_base_cached;
1611
1612static struct delta_base_cache_lru_list {
1613 struct delta_base_cache_lru_list *prev;
1614 struct delta_base_cache_lru_list *next;
1615} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1616
1617static struct delta_base_cache_entry {
1618 struct delta_base_cache_lru_list lru;
1619 void *data;
1620 struct packed_git *p;
1621 off_t base_offset;
1622 unsigned long size;
1623 enum object_type type;
1624} delta_base_cache[MAX_DELTA_CACHE];
1625
1626static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1627{
1628 unsigned long hash;
1629
1630 hash = (unsigned long)p + (unsigned long)base_offset;
1631 hash += (hash >> 8) + (hash >> 16);
1632 return hash % MAX_DELTA_CACHE;
1633}
1634
1635static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1636 unsigned long *base_size, enum object_type *type, int keep_cache)
1637{
1638 void *ret;
1639 unsigned long hash = pack_entry_hash(p, base_offset);
1640 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1641
1642 ret = ent->data;
1643 if (!ret || ent->p != p || ent->base_offset != base_offset)
1644 return unpack_entry(p, base_offset, type, base_size);
1645
1646 if (!keep_cache) {
1647 ent->data = NULL;
1648 ent->lru.next->prev = ent->lru.prev;
1649 ent->lru.prev->next = ent->lru.next;
1650 delta_base_cached -= ent->size;
1651 } else {
1652 ret = xmemdupz(ent->data, ent->size);
1653 }
1654 *type = ent->type;
1655 *base_size = ent->size;
1656 return ret;
1657}
1658
1659static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1660{
1661 if (ent->data) {
1662 free(ent->data);
1663 ent->data = NULL;
1664 ent->lru.next->prev = ent->lru.prev;
1665 ent->lru.prev->next = ent->lru.next;
1666 delta_base_cached -= ent->size;
1667 }
1668}
1669
1670void clear_delta_base_cache(void)
1671{
1672 unsigned long p;
1673 for (p = 0; p < MAX_DELTA_CACHE; p++)
1674 release_delta_base_cache(&delta_base_cache[p]);
1675}
1676
1677static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1678 void *base, unsigned long base_size, enum object_type type)
1679{
1680 unsigned long hash = pack_entry_hash(p, base_offset);
1681 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1682 struct delta_base_cache_lru_list *lru;
1683
1684 release_delta_base_cache(ent);
1685 delta_base_cached += base_size;
1686
1687 for (lru = delta_base_cache_lru.next;
1688 delta_base_cached > delta_base_cache_limit
1689 && lru != &delta_base_cache_lru;
1690 lru = lru->next) {
1691 struct delta_base_cache_entry *f = (void *)lru;
1692 if (f->type == OBJ_BLOB)
1693 release_delta_base_cache(f);
1694 }
1695 for (lru = delta_base_cache_lru.next;
1696 delta_base_cached > delta_base_cache_limit
1697 && lru != &delta_base_cache_lru;
1698 lru = lru->next) {
1699 struct delta_base_cache_entry *f = (void *)lru;
1700 release_delta_base_cache(f);
1701 }
1702
1703 ent->p = p;
1704 ent->base_offset = base_offset;
1705 ent->type = type;
1706 ent->data = base;
1707 ent->size = base_size;
1708 ent->lru.next = &delta_base_cache_lru;
1709 ent->lru.prev = delta_base_cache_lru.prev;
1710 delta_base_cache_lru.prev->next = &ent->lru;
1711 delta_base_cache_lru.prev = &ent->lru;
1712}
1713
1714static void *read_object(const unsigned char *sha1, enum object_type *type,
1715 unsigned long *size);
1716
1717static void *unpack_delta_entry(struct packed_git *p,
1718 struct pack_window **w_curs,
1719 off_t curpos,
1720 unsigned long delta_size,
1721 off_t obj_offset,
1722 enum object_type *type,
1723 unsigned long *sizep)
1724{
1725 void *delta_data, *result, *base;
1726 unsigned long base_size;
1727 off_t base_offset;
1728
1729 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1730 if (!base_offset) {
1731 error("failed to validate delta base reference "
1732 "at offset %"PRIuMAX" from %s",
1733 (uintmax_t)curpos, p->pack_name);
1734 return NULL;
1735 }
1736 unuse_pack(w_curs);
1737 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1738 if (!base) {
1739 /*
1740 * We're probably in deep shit, but let's try to fetch
1741 * the required base anyway from another pack or loose.
1742 * This is costly but should happen only in the presence
1743 * of a corrupted pack, and is better than failing outright.
1744 */
1745 struct revindex_entry *revidx;
1746 const unsigned char *base_sha1;
1747 revidx = find_pack_revindex(p, base_offset);
1748 if (!revidx)
1749 return NULL;
1750 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1751 error("failed to read delta base object %s"
1752 " at offset %"PRIuMAX" from %s",
1753 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1754 p->pack_name);
1755 mark_bad_packed_object(p, base_sha1);
1756 base = read_object(base_sha1, type, &base_size);
1757 if (!base)
1758 return NULL;
1759 }
1760
1761 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1762 if (!delta_data) {
1763 error("failed to unpack compressed delta "
1764 "at offset %"PRIuMAX" from %s",
1765 (uintmax_t)curpos, p->pack_name);
1766 free(base);
1767 return NULL;
1768 }
1769 result = patch_delta(base, base_size,
1770 delta_data, delta_size,
1771 sizep);
1772 if (!result)
1773 die("failed to apply delta");
1774 free(delta_data);
1775 add_delta_base_cache(p, base_offset, base, base_size, *type);
1776 return result;
1777}
1778
1779int do_check_packed_object_crc;
1780
1781void *unpack_entry(struct packed_git *p, off_t obj_offset,
1782 enum object_type *type, unsigned long *sizep)
1783{
1784 struct pack_window *w_curs = NULL;
1785 off_t curpos = obj_offset;
1786 void *data;
1787
1788 if (do_check_packed_object_crc && p->index_version > 1) {
1789 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1790 unsigned long len = revidx[1].offset - obj_offset;
1791 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1792 const unsigned char *sha1 =
1793 nth_packed_object_sha1(p, revidx->nr);
1794 error("bad packed object CRC for %s",
1795 sha1_to_hex(sha1));
1796 mark_bad_packed_object(p, sha1);
1797 unuse_pack(&w_curs);
1798 return NULL;
1799 }
1800 }
1801
1802 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1803 switch (*type) {
1804 case OBJ_OFS_DELTA:
1805 case OBJ_REF_DELTA:
1806 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1807 obj_offset, type, sizep);
1808 break;
1809 case OBJ_COMMIT:
1810 case OBJ_TREE:
1811 case OBJ_BLOB:
1812 case OBJ_TAG:
1813 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1814 break;
1815 default:
1816 data = NULL;
1817 error("unknown object type %i at offset %"PRIuMAX" in %s",
1818 *type, (uintmax_t)obj_offset, p->pack_name);
1819 }
1820 unuse_pack(&w_curs);
1821 return data;
1822}
1823
1824const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1825 uint32_t n)
1826{
1827 const unsigned char *index = p->index_data;
1828 if (!index) {
1829 if (open_pack_index(p))
1830 return NULL;
1831 index = p->index_data;
1832 }
1833 if (n >= p->num_objects)
1834 return NULL;
1835 index += 4 * 256;
1836 if (p->index_version == 1) {
1837 return index + 24 * n + 4;
1838 } else {
1839 index += 8;
1840 return index + 20 * n;
1841 }
1842}
1843
1844off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1845{
1846 const unsigned char *index = p->index_data;
1847 index += 4 * 256;
1848 if (p->index_version == 1) {
1849 return ntohl(*((uint32_t *)(index + 24 * n)));
1850 } else {
1851 uint32_t off;
1852 index += 8 + p->num_objects * (20 + 4);
1853 off = ntohl(*((uint32_t *)(index + 4 * n)));
1854 if (!(off & 0x80000000))
1855 return off;
1856 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1857 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1858 ntohl(*((uint32_t *)(index + 4)));
1859 }
1860}
1861
1862off_t find_pack_entry_one(const unsigned char *sha1,
1863 struct packed_git *p)
1864{
1865 const uint32_t *level1_ofs = p->index_data;
1866 const unsigned char *index = p->index_data;
1867 unsigned hi, lo, stride;
1868 static int use_lookup = -1;
1869 static int debug_lookup = -1;
1870
1871 if (debug_lookup < 0)
1872 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1873
1874 if (!index) {
1875 if (open_pack_index(p))
1876 return 0;
1877 level1_ofs = p->index_data;
1878 index = p->index_data;
1879 }
1880 if (p->index_version > 1) {
1881 level1_ofs += 2;
1882 index += 8;
1883 }
1884 index += 4 * 256;
1885 hi = ntohl(level1_ofs[*sha1]);
1886 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1887 if (p->index_version > 1) {
1888 stride = 20;
1889 } else {
1890 stride = 24;
1891 index += 4;
1892 }
1893
1894 if (debug_lookup)
1895 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1896 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1897
1898 if (use_lookup < 0)
1899 use_lookup = !!getenv("GIT_USE_LOOKUP");
1900 if (use_lookup) {
1901 int pos = sha1_entry_pos(index, stride, 0,
1902 lo, hi, p->num_objects, sha1);
1903 if (pos < 0)
1904 return 0;
1905 return nth_packed_object_offset(p, pos);
1906 }
1907
1908 do {
1909 unsigned mi = (lo + hi) / 2;
1910 int cmp = hashcmp(index + mi * stride, sha1);
1911
1912 if (debug_lookup)
1913 printf("lo %u hi %u rg %u mi %u\n",
1914 lo, hi, hi - lo, mi);
1915 if (!cmp)
1916 return nth_packed_object_offset(p, mi);
1917 if (cmp > 0)
1918 hi = mi;
1919 else
1920 lo = mi+1;
1921 } while (lo < hi);
1922 return 0;
1923}
1924
1925static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1926{
1927 static struct packed_git *last_found = (void *)1;
1928 struct packed_git *p;
1929 off_t offset;
1930
1931 prepare_packed_git();
1932 if (!packed_git)
1933 return 0;
1934 p = (last_found == (void *)1) ? packed_git : last_found;
1935
1936 do {
1937 if (p->num_bad_objects) {
1938 unsigned i;
1939 for (i = 0; i < p->num_bad_objects; i++)
1940 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1941 goto next;
1942 }
1943
1944 offset = find_pack_entry_one(sha1, p);
1945 if (offset) {
1946 /*
1947 * We are about to tell the caller where they can
1948 * locate the requested object. We better make
1949 * sure the packfile is still here and can be
1950 * accessed before supplying that answer, as
1951 * it may have been deleted since the index
1952 * was loaded!
1953 */
1954 if (p->pack_fd == -1 && open_packed_git(p)) {
1955 error("packfile %s cannot be accessed", p->pack_name);
1956 goto next;
1957 }
1958 e->offset = offset;
1959 e->p = p;
1960 hashcpy(e->sha1, sha1);
1961 last_found = p;
1962 return 1;
1963 }
1964
1965 next:
1966 if (p == last_found)
1967 p = packed_git;
1968 else
1969 p = p->next;
1970 if (p == last_found)
1971 p = p->next;
1972 } while (p);
1973 return 0;
1974}
1975
1976struct packed_git *find_sha1_pack(const unsigned char *sha1,
1977 struct packed_git *packs)
1978{
1979 struct packed_git *p;
1980
1981 for (p = packs; p; p = p->next) {
1982 if (find_pack_entry_one(sha1, p))
1983 return p;
1984 }
1985 return NULL;
1986
1987}
1988
1989static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1990{
1991 int status;
1992 unsigned long mapsize, size;
1993 void *map;
1994 z_stream stream;
1995 char hdr[32];
1996
1997 map = map_sha1_file(sha1, &mapsize);
1998 if (!map)
1999 return error("unable to find %s", sha1_to_hex(sha1));
2000 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2001 status = error("unable to unpack %s header",
2002 sha1_to_hex(sha1));
2003 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2004 status = error("unable to parse %s header", sha1_to_hex(sha1));
2005 else if (sizep)
2006 *sizep = size;
2007 git_inflate_end(&stream);
2008 munmap(map, mapsize);
2009 return status;
2010}
2011
2012int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2013{
2014 struct pack_entry e;
2015 int status;
2016
2017 if (!find_pack_entry(sha1, &e)) {
2018 /* Most likely it's a loose object. */
2019 status = sha1_loose_object_info(sha1, sizep);
2020 if (status >= 0)
2021 return status;
2022
2023 /* Not a loose object; someone else may have just packed it. */
2024 reprepare_packed_git();
2025 if (!find_pack_entry(sha1, &e))
2026 return status;
2027 }
2028
2029 status = packed_object_info(e.p, e.offset, sizep);
2030 if (status < 0) {
2031 mark_bad_packed_object(e.p, sha1);
2032 status = sha1_object_info(sha1, sizep);
2033 }
2034
2035 return status;
2036}
2037
2038static void *read_packed_sha1(const unsigned char *sha1,
2039 enum object_type *type, unsigned long *size)
2040{
2041 struct pack_entry e;
2042 void *data;
2043
2044 if (!find_pack_entry(sha1, &e))
2045 return NULL;
2046 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2047 if (!data) {
2048 /*
2049 * We're probably in deep shit, but let's try to fetch
2050 * the required object anyway from another pack or loose.
2051 * This should happen only in the presence of a corrupted
2052 * pack, and is better than failing outright.
2053 */
2054 error("failed to read object %s at offset %"PRIuMAX" from %s",
2055 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2056 mark_bad_packed_object(e.p, sha1);
2057 data = read_object(sha1, type, size);
2058 }
2059 return data;
2060}
2061
2062/*
2063 * This is meant to hold a *small* number of objects that you would
2064 * want read_sha1_file() to be able to return, but yet you do not want
2065 * to write them into the object store (e.g. a browse-only
2066 * application).
2067 */
2068static struct cached_object {
2069 unsigned char sha1[20];
2070 enum object_type type;
2071 void *buf;
2072 unsigned long size;
2073} *cached_objects;
2074static int cached_object_nr, cached_object_alloc;
2075
2076static struct cached_object empty_tree = {
2077 EMPTY_TREE_SHA1_BIN,
2078 OBJ_TREE,
2079 "",
2080 0
2081};
2082
2083static struct cached_object *find_cached_object(const unsigned char *sha1)
2084{
2085 int i;
2086 struct cached_object *co = cached_objects;
2087
2088 for (i = 0; i < cached_object_nr; i++, co++) {
2089 if (!hashcmp(co->sha1, sha1))
2090 return co;
2091 }
2092 if (!hashcmp(sha1, empty_tree.sha1))
2093 return &empty_tree;
2094 return NULL;
2095}
2096
2097int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2098 unsigned char *sha1)
2099{
2100 struct cached_object *co;
2101
2102 hash_sha1_file(buf, len, typename(type), sha1);
2103 if (has_sha1_file(sha1) || find_cached_object(sha1))
2104 return 0;
2105 if (cached_object_alloc <= cached_object_nr) {
2106 cached_object_alloc = alloc_nr(cached_object_alloc);
2107 cached_objects = xrealloc(cached_objects,
2108 sizeof(*cached_objects) *
2109 cached_object_alloc);
2110 }
2111 co = &cached_objects[cached_object_nr++];
2112 co->size = len;
2113 co->type = type;
2114 co->buf = xmalloc(len);
2115 memcpy(co->buf, buf, len);
2116 hashcpy(co->sha1, sha1);
2117 return 0;
2118}
2119
2120static void *read_object(const unsigned char *sha1, enum object_type *type,
2121 unsigned long *size)
2122{
2123 unsigned long mapsize;
2124 void *map, *buf;
2125 struct cached_object *co;
2126
2127 co = find_cached_object(sha1);
2128 if (co) {
2129 *type = co->type;
2130 *size = co->size;
2131 return xmemdupz(co->buf, co->size);
2132 }
2133
2134 buf = read_packed_sha1(sha1, type, size);
2135 if (buf)
2136 return buf;
2137 map = map_sha1_file(sha1, &mapsize);
2138 if (map) {
2139 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2140 munmap(map, mapsize);
2141 return buf;
2142 }
2143 reprepare_packed_git();
2144 return read_packed_sha1(sha1, type, size);
2145}
2146
2147void *read_sha1_file_repl(const unsigned char *sha1,
2148 enum object_type *type,
2149 unsigned long *size,
2150 const unsigned char **replacement)
2151{
2152 const unsigned char *repl = lookup_replace_object(sha1);
2153 void *data = read_object(repl, type, size);
2154
2155 /* die if we replaced an object with one that does not exist */
2156 if (!data && repl != sha1)
2157 die("replacement %s not found for %s",
2158 sha1_to_hex(repl), sha1_to_hex(sha1));
2159
2160 /* legacy behavior is to die on corrupted objects */
2161 if (!data && (has_loose_object(repl) || has_packed_and_bad(repl)))
2162 die("object %s is corrupted", sha1_to_hex(repl));
2163
2164 if (replacement)
2165 *replacement = repl;
2166
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 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2237 * "moving" is only a part of what it does, when no patch between
2238 * master to pu changes the call sites of this function.
2239 */
2240int move_temp_to_file(const char *tmpfile, const char *filename)
2241{
2242 int ret = 0;
2243
2244 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2245 goto try_rename;
2246 else if (link(tmpfile, filename))
2247 ret = errno;
2248
2249 /*
2250 * Coda hack - coda doesn't like cross-directory links,
2251 * so we fall back to a rename, which will mean that it
2252 * won't be able to check collisions, but that's not a
2253 * big deal.
2254 *
2255 * The same holds for FAT formatted media.
2256 *
2257 * When this succeeds, we just return. We have nothing
2258 * left to unlink.
2259 */
2260 if (ret && ret != EEXIST) {
2261 try_rename:
2262 if (!rename(tmpfile, filename))
2263 goto out;
2264 ret = errno;
2265 }
2266 unlink_or_warn(tmpfile);
2267 if (ret) {
2268 if (ret != EEXIST) {
2269 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2270 }
2271 /* FIXME!!! Collision check here ? */
2272 }
2273
2274out:
2275 if (set_shared_perm(filename, (S_IFREG|0444)))
2276 return error("unable to set permission to '%s'", filename);
2277 return 0;
2278}
2279
2280static int write_buffer(int fd, const void *buf, size_t len)
2281{
2282 if (write_in_full(fd, buf, len) < 0)
2283 return error("file write error (%s)", strerror(errno));
2284 return 0;
2285}
2286
2287int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2288 unsigned char *sha1)
2289{
2290 char hdr[32];
2291 int hdrlen;
2292 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2293 return 0;
2294}
2295
2296/* Finalize a file on disk, and close it. */
2297static void close_sha1_file(int fd)
2298{
2299 if (fsync_object_files)
2300 fsync_or_die(fd, "sha1 file");
2301 if (close(fd) != 0)
2302 die_errno("error when closing sha1 file");
2303}
2304
2305/* Size of directory component, including the ending '/' */
2306static inline int directory_size(const char *filename)
2307{
2308 const char *s = strrchr(filename, '/');
2309 if (!s)
2310 return 0;
2311 return s - filename + 1;
2312}
2313
2314/*
2315 * This creates a temporary file in the same directory as the final
2316 * 'filename'
2317 *
2318 * We want to avoid cross-directory filename renames, because those
2319 * can have problems on various filesystems (FAT, NFS, Coda).
2320 */
2321static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2322{
2323 int fd, dirlen = directory_size(filename);
2324
2325 if (dirlen + 20 > bufsiz) {
2326 errno = ENAMETOOLONG;
2327 return -1;
2328 }
2329 memcpy(buffer, filename, dirlen);
2330 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2331 fd = mkstemp(buffer);
2332 if (fd < 0 && dirlen && errno == ENOENT) {
2333 /* Make sure the directory exists */
2334 memcpy(buffer, filename, dirlen);
2335 buffer[dirlen-1] = 0;
2336 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2337 return -1;
2338
2339 /* Try again */
2340 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2341 fd = mkstemp(buffer);
2342 }
2343 return fd;
2344}
2345
2346static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2347 void *buf, unsigned long len, time_t mtime)
2348{
2349 int fd, ret;
2350 size_t size;
2351 unsigned char *compressed;
2352 z_stream stream;
2353 char *filename;
2354 static char tmpfile[PATH_MAX];
2355
2356 filename = sha1_file_name(sha1);
2357 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2358 while (fd < 0 && errno == EMFILE && unuse_one_window(packed_git, -1))
2359 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2360 if (fd < 0) {
2361 if (errno == EACCES)
2362 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2363 else
2364 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2365 }
2366
2367 /* Set it up */
2368 memset(&stream, 0, sizeof(stream));
2369 deflateInit(&stream, zlib_compression_level);
2370 size = 8 + deflateBound(&stream, len+hdrlen);
2371 compressed = xmalloc(size);
2372
2373 /* Compress it */
2374 stream.next_out = compressed;
2375 stream.avail_out = size;
2376
2377 /* First header.. */
2378 stream.next_in = (unsigned char *)hdr;
2379 stream.avail_in = hdrlen;
2380 while (deflate(&stream, 0) == Z_OK)
2381 /* nothing */;
2382
2383 /* Then the data itself.. */
2384 stream.next_in = buf;
2385 stream.avail_in = len;
2386 ret = deflate(&stream, Z_FINISH);
2387 if (ret != Z_STREAM_END)
2388 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2389
2390 ret = deflateEnd(&stream);
2391 if (ret != Z_OK)
2392 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2393
2394 size = stream.total_out;
2395
2396 if (write_buffer(fd, compressed, size) < 0)
2397 die("unable to write sha1 file");
2398 close_sha1_file(fd);
2399 free(compressed);
2400
2401 if (mtime) {
2402 struct utimbuf utb;
2403 utb.actime = mtime;
2404 utb.modtime = mtime;
2405 if (utime(tmpfile, &utb) < 0)
2406 warning("failed utime() on %s: %s",
2407 tmpfile, strerror(errno));
2408 }
2409
2410 return move_temp_to_file(tmpfile, filename);
2411}
2412
2413int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2414{
2415 unsigned char sha1[20];
2416 char hdr[32];
2417 int hdrlen;
2418
2419 /* Normally if we have it in the pack then we do not bother writing
2420 * it out into .git/objects/??/?{38} file.
2421 */
2422 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2423 if (returnsha1)
2424 hashcpy(returnsha1, sha1);
2425 if (has_sha1_file(sha1))
2426 return 0;
2427 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2428}
2429
2430int force_object_loose(const unsigned char *sha1, time_t mtime)
2431{
2432 void *buf;
2433 unsigned long len;
2434 enum object_type type;
2435 char hdr[32];
2436 int hdrlen;
2437 int ret;
2438
2439 if (has_loose_object(sha1))
2440 return 0;
2441 buf = read_packed_sha1(sha1, &type, &len);
2442 if (!buf)
2443 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2444 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2445 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2446 free(buf);
2447
2448 return ret;
2449}
2450
2451int has_pack_index(const unsigned char *sha1)
2452{
2453 struct stat st;
2454 if (stat(sha1_pack_index_name(sha1), &st))
2455 return 0;
2456 return 1;
2457}
2458
2459int has_pack_file(const unsigned char *sha1)
2460{
2461 struct stat st;
2462 if (stat(sha1_pack_name(sha1), &st))
2463 return 0;
2464 return 1;
2465}
2466
2467int has_sha1_pack(const unsigned char *sha1)
2468{
2469 struct pack_entry e;
2470 return find_pack_entry(sha1, &e);
2471}
2472
2473int has_sha1_file(const unsigned char *sha1)
2474{
2475 struct pack_entry e;
2476
2477 if (find_pack_entry(sha1, &e))
2478 return 1;
2479 return has_loose_object(sha1);
2480}
2481
2482static int index_mem(unsigned char *sha1, void *buf, size_t size,
2483 int write_object, enum object_type type, const char *path)
2484{
2485 int ret, re_allocated = 0;
2486
2487 if (!type)
2488 type = OBJ_BLOB;
2489
2490 /*
2491 * Convert blobs to git internal format
2492 */
2493 if ((type == OBJ_BLOB) && path) {
2494 struct strbuf nbuf = STRBUF_INIT;
2495 if (convert_to_git(path, buf, size, &nbuf,
2496 write_object ? safe_crlf : 0)) {
2497 buf = strbuf_detach(&nbuf, &size);
2498 re_allocated = 1;
2499 }
2500 }
2501
2502 if (write_object)
2503 ret = write_sha1_file(buf, size, typename(type), sha1);
2504 else
2505 ret = hash_sha1_file(buf, size, typename(type), sha1);
2506 if (re_allocated)
2507 free(buf);
2508 return ret;
2509}
2510
2511int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2512 enum object_type type, const char *path)
2513{
2514 int ret;
2515 size_t size = xsize_t(st->st_size);
2516
2517 if (!S_ISREG(st->st_mode)) {
2518 struct strbuf sbuf = STRBUF_INIT;
2519 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2520 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2521 type, path);
2522 else
2523 ret = -1;
2524 strbuf_release(&sbuf);
2525 } else if (size) {
2526 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2527 ret = index_mem(sha1, buf, size, write_object, type, path);
2528 munmap(buf, size);
2529 } else
2530 ret = index_mem(sha1, NULL, size, write_object, type, path);
2531 close(fd);
2532 return ret;
2533}
2534
2535int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2536{
2537 int fd;
2538 struct strbuf sb = STRBUF_INIT;
2539
2540 switch (st->st_mode & S_IFMT) {
2541 case S_IFREG:
2542 fd = open(path, O_RDONLY);
2543 if (fd < 0)
2544 return error("open(\"%s\"): %s", path,
2545 strerror(errno));
2546 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2547 return error("%s: failed to insert into database",
2548 path);
2549 break;
2550 case S_IFLNK:
2551 if (strbuf_readlink(&sb, path, st->st_size)) {
2552 char *errstr = strerror(errno);
2553 return error("readlink(\"%s\"): %s", path,
2554 errstr);
2555 }
2556 if (!write_object)
2557 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2558 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2559 return error("%s: failed to insert into database",
2560 path);
2561 strbuf_release(&sb);
2562 break;
2563 case S_IFDIR:
2564 return resolve_gitlink_ref(path, "HEAD", sha1);
2565 default:
2566 return error("%s: unsupported file type", path);
2567 }
2568 return 0;
2569}
2570
2571int read_pack_header(int fd, struct pack_header *header)
2572{
2573 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2574 /* "eof before pack header was fully read" */
2575 return PH_ERROR_EOF;
2576
2577 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2578 /* "protocol error (pack signature mismatch detected)" */
2579 return PH_ERROR_PACK_SIGNATURE;
2580 if (!pack_version_ok(header->hdr_version))
2581 /* "protocol error (pack version unsupported)" */
2582 return PH_ERROR_PROTOCOL;
2583 return 0;
2584}