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"), 1);
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(const unsigned char *sha1)
427{
428 char *name = sha1_file_name(sha1);
429 struct alternate_object_database *alt;
430
431 if (!access(name, F_OK))
432 return 1;
433 prepare_alt_odb();
434 for (alt = alt_odb_list; alt; alt = alt->next) {
435 name = alt->name;
436 fill_sha1_path(name, sha1);
437 if (!access(alt->base, F_OK))
438 return 1;
439 }
440 return 0;
441}
442
443static unsigned int pack_used_ctr;
444static unsigned int pack_mmap_calls;
445static unsigned int peak_pack_open_windows;
446static unsigned int pack_open_windows;
447static size_t peak_pack_mapped;
448static size_t pack_mapped;
449struct packed_git *packed_git;
450
451void pack_report(void)
452{
453 fprintf(stderr,
454 "pack_report: getpagesize() = %10" SZ_FMT "\n"
455 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
456 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
457 sz_fmt(getpagesize()),
458 sz_fmt(packed_git_window_size),
459 sz_fmt(packed_git_limit));
460 fprintf(stderr,
461 "pack_report: pack_used_ctr = %10u\n"
462 "pack_report: pack_mmap_calls = %10u\n"
463 "pack_report: pack_open_windows = %10u / %10u\n"
464 "pack_report: pack_mapped = "
465 "%10" SZ_FMT " / %10" SZ_FMT "\n",
466 pack_used_ctr,
467 pack_mmap_calls,
468 pack_open_windows, peak_pack_open_windows,
469 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
470}
471
472static int check_packed_git_idx(const char *path, struct packed_git *p)
473{
474 void *idx_map;
475 struct pack_idx_header *hdr;
476 size_t idx_size;
477 uint32_t version, nr, i, *index;
478 int fd = open(path, O_RDONLY);
479 struct stat st;
480
481 if (fd < 0)
482 return -1;
483 if (fstat(fd, &st)) {
484 close(fd);
485 return -1;
486 }
487 idx_size = xsize_t(st.st_size);
488 if (idx_size < 4 * 256 + 20 + 20) {
489 close(fd);
490 return error("index file %s is too small", path);
491 }
492 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
493 close(fd);
494
495 hdr = idx_map;
496 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
497 version = ntohl(hdr->idx_version);
498 if (version < 2 || version > 2) {
499 munmap(idx_map, idx_size);
500 return error("index file %s is version %"PRIu32
501 " and is not supported by this binary"
502 " (try upgrading GIT to a newer version)",
503 path, version);
504 }
505 } else
506 version = 1;
507
508 nr = 0;
509 index = idx_map;
510 if (version > 1)
511 index += 2; /* skip index header */
512 for (i = 0; i < 256; i++) {
513 uint32_t n = ntohl(index[i]);
514 if (n < nr) {
515 munmap(idx_map, idx_size);
516 return error("non-monotonic index %s", path);
517 }
518 nr = n;
519 }
520
521 if (version == 1) {
522 /*
523 * Total size:
524 * - 256 index entries 4 bytes each
525 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
526 * - 20-byte SHA1 of the packfile
527 * - 20-byte SHA1 file checksum
528 */
529 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
530 munmap(idx_map, idx_size);
531 return error("wrong index v1 file size in %s", path);
532 }
533 } else if (version == 2) {
534 /*
535 * Minimum size:
536 * - 8 bytes of header
537 * - 256 index entries 4 bytes each
538 * - 20-byte sha1 entry * nr
539 * - 4-byte crc entry * nr
540 * - 4-byte offset entry * nr
541 * - 20-byte SHA1 of the packfile
542 * - 20-byte SHA1 file checksum
543 * And after the 4-byte offset table might be a
544 * variable sized table containing 8-byte entries
545 * for offsets larger than 2^31.
546 */
547 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
548 unsigned long max_size = min_size;
549 if (nr)
550 max_size += (nr - 1)*8;
551 if (idx_size < min_size || idx_size > max_size) {
552 munmap(idx_map, idx_size);
553 return error("wrong index v2 file size in %s", path);
554 }
555 if (idx_size != min_size &&
556 /*
557 * make sure we can deal with large pack offsets.
558 * 31-bit signed offset won't be enough, neither
559 * 32-bit unsigned one will be.
560 */
561 (sizeof(off_t) <= 4)) {
562 munmap(idx_map, idx_size);
563 return error("pack too large for current definition of off_t in %s", path);
564 }
565 }
566
567 p->index_version = version;
568 p->index_data = idx_map;
569 p->index_size = idx_size;
570 p->num_objects = nr;
571 return 0;
572}
573
574int open_pack_index(struct packed_git *p)
575{
576 char *idx_name;
577 int ret;
578
579 if (p->index_data)
580 return 0;
581
582 idx_name = xstrdup(p->pack_name);
583 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
584 ret = check_packed_git_idx(idx_name, p);
585 free(idx_name);
586 return ret;
587}
588
589static void scan_windows(struct packed_git *p,
590 struct packed_git **lru_p,
591 struct pack_window **lru_w,
592 struct pack_window **lru_l)
593{
594 struct pack_window *w, *w_l;
595
596 for (w_l = NULL, w = p->windows; w; w = w->next) {
597 if (!w->inuse_cnt) {
598 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
599 *lru_p = p;
600 *lru_w = w;
601 *lru_l = w_l;
602 }
603 }
604 w_l = w;
605 }
606}
607
608static int unuse_one_window(struct packed_git *current, int keep_fd)
609{
610 struct packed_git *p, *lru_p = NULL;
611 struct pack_window *lru_w = NULL, *lru_l = NULL;
612
613 if (current)
614 scan_windows(current, &lru_p, &lru_w, &lru_l);
615 for (p = packed_git; p; p = p->next)
616 scan_windows(p, &lru_p, &lru_w, &lru_l);
617 if (lru_p) {
618 munmap(lru_w->base, lru_w->len);
619 pack_mapped -= lru_w->len;
620 if (lru_l)
621 lru_l->next = lru_w->next;
622 else {
623 lru_p->windows = lru_w->next;
624 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
625 close(lru_p->pack_fd);
626 lru_p->pack_fd = -1;
627 }
628 }
629 free(lru_w);
630 pack_open_windows--;
631 return 1;
632 }
633 return 0;
634}
635
636void release_pack_memory(size_t need, int fd)
637{
638 size_t cur = pack_mapped;
639 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
640 ; /* nothing */
641}
642
643void close_pack_windows(struct packed_git *p)
644{
645 while (p->windows) {
646 struct pack_window *w = p->windows;
647
648 if (w->inuse_cnt)
649 die("pack '%s' still has open windows to it",
650 p->pack_name);
651 munmap(w->base, w->len);
652 pack_mapped -= w->len;
653 pack_open_windows--;
654 p->windows = w->next;
655 free(w);
656 }
657}
658
659void unuse_pack(struct pack_window **w_cursor)
660{
661 struct pack_window *w = *w_cursor;
662 if (w) {
663 w->inuse_cnt--;
664 *w_cursor = NULL;
665 }
666}
667
668/*
669 * Do not call this directly as this leaks p->pack_fd on error return;
670 * call open_packed_git() instead.
671 */
672static int open_packed_git_1(struct packed_git *p)
673{
674 struct stat st;
675 struct pack_header hdr;
676 unsigned char sha1[20];
677 unsigned char *idx_sha1;
678 long fd_flag;
679
680 if (!p->index_data && open_pack_index(p))
681 return error("packfile %s index unavailable", p->pack_name);
682
683 p->pack_fd = open(p->pack_name, O_RDONLY);
684 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
685 return -1;
686
687 /* If we created the struct before we had the pack we lack size. */
688 if (!p->pack_size) {
689 if (!S_ISREG(st.st_mode))
690 return error("packfile %s not a regular file", p->pack_name);
691 p->pack_size = st.st_size;
692 } else if (p->pack_size != st.st_size)
693 return error("packfile %s size changed", p->pack_name);
694
695 /* We leave these file descriptors open with sliding mmap;
696 * there is no point keeping them open across exec(), though.
697 */
698 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
699 if (fd_flag < 0)
700 return error("cannot determine file descriptor flags");
701 fd_flag |= FD_CLOEXEC;
702 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
703 return error("cannot set FD_CLOEXEC");
704
705 /* Verify we recognize this pack file format. */
706 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
707 return error("file %s is far too short to be a packfile", p->pack_name);
708 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
709 return error("file %s is not a GIT packfile", p->pack_name);
710 if (!pack_version_ok(hdr.hdr_version))
711 return error("packfile %s is version %"PRIu32" and not"
712 " supported (try upgrading GIT to a newer version)",
713 p->pack_name, ntohl(hdr.hdr_version));
714
715 /* Verify the pack matches its index. */
716 if (p->num_objects != ntohl(hdr.hdr_entries))
717 return error("packfile %s claims to have %"PRIu32" objects"
718 " while index indicates %"PRIu32" objects",
719 p->pack_name, ntohl(hdr.hdr_entries),
720 p->num_objects);
721 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
722 return error("end of packfile %s is unavailable", p->pack_name);
723 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
724 return error("packfile %s signature is unavailable", p->pack_name);
725 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
726 if (hashcmp(sha1, idx_sha1))
727 return error("packfile %s does not match index", p->pack_name);
728 return 0;
729}
730
731static int open_packed_git(struct packed_git *p)
732{
733 if (!open_packed_git_1(p))
734 return 0;
735 if (p->pack_fd != -1) {
736 close(p->pack_fd);
737 p->pack_fd = -1;
738 }
739 return -1;
740}
741
742static int in_window(struct pack_window *win, off_t offset)
743{
744 /* We must promise at least 20 bytes (one hash) after the
745 * offset is available from this window, otherwise the offset
746 * is not actually in this window and a different window (which
747 * has that one hash excess) must be used. This is to support
748 * the object header and delta base parsing routines below.
749 */
750 off_t win_off = win->offset;
751 return win_off <= offset
752 && (offset + 20) <= (win_off + win->len);
753}
754
755unsigned char* use_pack(struct packed_git *p,
756 struct pack_window **w_cursor,
757 off_t offset,
758 unsigned int *left)
759{
760 struct pack_window *win = *w_cursor;
761
762 if (p->pack_fd == -1 && open_packed_git(p))
763 die("packfile %s cannot be accessed", p->pack_name);
764
765 /* Since packfiles end in a hash of their content and its
766 * pointless to ask for an offset into the middle of that
767 * hash, and the in_window function above wouldn't match
768 * don't allow an offset too close to the end of the file.
769 */
770 if (offset > (p->pack_size - 20))
771 die("offset beyond end of packfile (truncated pack?)");
772
773 if (!win || !in_window(win, offset)) {
774 if (win)
775 win->inuse_cnt--;
776 for (win = p->windows; win; win = win->next) {
777 if (in_window(win, offset))
778 break;
779 }
780 if (!win) {
781 size_t window_align = packed_git_window_size / 2;
782 off_t len;
783 win = xcalloc(1, sizeof(*win));
784 win->offset = (offset / window_align) * window_align;
785 len = p->pack_size - win->offset;
786 if (len > packed_git_window_size)
787 len = packed_git_window_size;
788 win->len = (size_t)len;
789 pack_mapped += win->len;
790 while (packed_git_limit < pack_mapped
791 && unuse_one_window(p, p->pack_fd))
792 ; /* nothing */
793 win->base = xmmap(NULL, win->len,
794 PROT_READ, MAP_PRIVATE,
795 p->pack_fd, win->offset);
796 if (win->base == MAP_FAILED)
797 die("packfile %s cannot be mapped: %s",
798 p->pack_name,
799 strerror(errno));
800 pack_mmap_calls++;
801 pack_open_windows++;
802 if (pack_mapped > peak_pack_mapped)
803 peak_pack_mapped = pack_mapped;
804 if (pack_open_windows > peak_pack_open_windows)
805 peak_pack_open_windows = pack_open_windows;
806 win->next = p->windows;
807 p->windows = win;
808 }
809 }
810 if (win != *w_cursor) {
811 win->last_used = pack_used_ctr++;
812 win->inuse_cnt++;
813 *w_cursor = win;
814 }
815 offset -= win->offset;
816 if (left)
817 *left = win->len - xsize_t(offset);
818 return win->base + offset;
819}
820
821static struct packed_git *alloc_packed_git(int extra)
822{
823 struct packed_git *p = xmalloc(sizeof(*p) + extra);
824 memset(p, 0, sizeof(*p));
825 p->pack_fd = -1;
826 return p;
827}
828
829struct packed_git *add_packed_git(const char *path, int path_len, int local)
830{
831 struct stat st;
832 struct packed_git *p = alloc_packed_git(path_len + 2);
833
834 /*
835 * Make sure a corresponding .pack file exists and that
836 * the index looks sane.
837 */
838 path_len -= strlen(".idx");
839 if (path_len < 1) {
840 free(p);
841 return NULL;
842 }
843 memcpy(p->pack_name, path, path_len);
844 strcpy(p->pack_name + path_len, ".pack");
845 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
846 free(p);
847 return NULL;
848 }
849
850 /* ok, it looks sane as far as we can check without
851 * actually mapping the pack file.
852 */
853 p->pack_size = st.st_size;
854 p->pack_local = local;
855 p->mtime = st.st_mtime;
856 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
857 hashclr(p->sha1);
858 return p;
859}
860
861struct packed_git *parse_pack_index(unsigned char *sha1)
862{
863 const char *idx_path = sha1_pack_index_name(sha1);
864 const char *path = sha1_pack_name(sha1);
865 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
866
867 strcpy(p->pack_name, path);
868 hashcpy(p->sha1, sha1);
869 if (check_packed_git_idx(idx_path, p)) {
870 free(p);
871 return NULL;
872 }
873
874 return p;
875}
876
877void install_packed_git(struct packed_git *pack)
878{
879 pack->next = packed_git;
880 packed_git = pack;
881}
882
883static void prepare_packed_git_one(char *objdir, int local)
884{
885 /* Ensure that this buffer is large enough so that we can
886 append "/pack/" without clobbering the stack even if
887 strlen(objdir) were PATH_MAX. */
888 char path[PATH_MAX + 1 + 4 + 1 + 1];
889 int len;
890 DIR *dir;
891 struct dirent *de;
892
893 sprintf(path, "%s/pack", objdir);
894 len = strlen(path);
895 dir = opendir(path);
896 if (!dir) {
897 if (errno != ENOENT)
898 error("unable to open object pack directory: %s: %s",
899 path, strerror(errno));
900 return;
901 }
902 path[len++] = '/';
903 while ((de = readdir(dir)) != NULL) {
904 int namelen = strlen(de->d_name);
905 struct packed_git *p;
906
907 if (!has_extension(de->d_name, ".idx"))
908 continue;
909
910 if (len + namelen + 1 > sizeof(path))
911 continue;
912
913 /* Don't reopen a pack we already have. */
914 strcpy(path + len, de->d_name);
915 for (p = packed_git; p; p = p->next) {
916 if (!memcmp(path, p->pack_name, len + namelen - 4))
917 break;
918 }
919 if (p)
920 continue;
921 /* See if it really is a valid .idx file with corresponding
922 * .pack file that we can map.
923 */
924 p = add_packed_git(path, len + namelen, local);
925 if (!p)
926 continue;
927 install_packed_git(p);
928 }
929 closedir(dir);
930}
931
932static int sort_pack(const void *a_, const void *b_)
933{
934 struct packed_git *a = *((struct packed_git **)a_);
935 struct packed_git *b = *((struct packed_git **)b_);
936 int st;
937
938 /*
939 * Local packs tend to contain objects specific to our
940 * variant of the project than remote ones. In addition,
941 * remote ones could be on a network mounted filesystem.
942 * Favor local ones for these reasons.
943 */
944 st = a->pack_local - b->pack_local;
945 if (st)
946 return -st;
947
948 /*
949 * Younger packs tend to contain more recent objects,
950 * and more recent objects tend to get accessed more
951 * often.
952 */
953 if (a->mtime < b->mtime)
954 return 1;
955 else if (a->mtime == b->mtime)
956 return 0;
957 return -1;
958}
959
960static void rearrange_packed_git(void)
961{
962 struct packed_git **ary, *p;
963 int i, n;
964
965 for (n = 0, p = packed_git; p; p = p->next)
966 n++;
967 if (n < 2)
968 return;
969
970 /* prepare an array of packed_git for easier sorting */
971 ary = xcalloc(n, sizeof(struct packed_git *));
972 for (n = 0, p = packed_git; p; p = p->next)
973 ary[n++] = p;
974
975 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
976
977 /* link them back again */
978 for (i = 0; i < n - 1; i++)
979 ary[i]->next = ary[i + 1];
980 ary[n - 1]->next = NULL;
981 packed_git = ary[0];
982
983 free(ary);
984}
985
986static int prepare_packed_git_run_once = 0;
987void prepare_packed_git(void)
988{
989 struct alternate_object_database *alt;
990
991 if (prepare_packed_git_run_once)
992 return;
993 prepare_packed_git_one(get_object_directory(), 1);
994 prepare_alt_odb();
995 for (alt = alt_odb_list; alt; alt = alt->next) {
996 alt->name[-1] = 0;
997 prepare_packed_git_one(alt->base, 0);
998 alt->name[-1] = '/';
999 }
1000 rearrange_packed_git();
1001 prepare_packed_git_run_once = 1;
1002}
1003
1004void reprepare_packed_git(void)
1005{
1006 discard_revindex();
1007 prepare_packed_git_run_once = 0;
1008 prepare_packed_git();
1009}
1010
1011static void mark_bad_packed_object(struct packed_git *p,
1012 const unsigned char *sha1)
1013{
1014 unsigned i;
1015 for (i = 0; i < p->num_bad_objects; i++)
1016 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1017 return;
1018 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1019 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1020 p->num_bad_objects++;
1021}
1022
1023static int has_packed_and_bad(const unsigned char *sha1)
1024{
1025 struct packed_git *p;
1026 unsigned i;
1027
1028 for (p = packed_git; p; p = p->next)
1029 for (i = 0; i < p->num_bad_objects; i++)
1030 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1031 return 1;
1032 return 0;
1033}
1034
1035int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1036{
1037 unsigned char real_sha1[20];
1038 hash_sha1_file(map, size, type, real_sha1);
1039 return hashcmp(sha1, real_sha1) ? -1 : 0;
1040}
1041
1042static int git_open_noatime(const char *name)
1043{
1044 static int sha1_file_open_flag = O_NOATIME;
1045 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1046
1047 /* Might the failure be due to O_NOATIME? */
1048 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1049 fd = open(name, O_RDONLY);
1050 if (fd >= 0)
1051 sha1_file_open_flag = 0;
1052 }
1053 return fd;
1054}
1055
1056static int open_sha1_file(const unsigned char *sha1)
1057{
1058 int fd;
1059 char *name = sha1_file_name(sha1);
1060 struct alternate_object_database *alt;
1061
1062 fd = git_open_noatime(name);
1063 if (fd >= 0)
1064 return fd;
1065
1066 prepare_alt_odb();
1067 errno = ENOENT;
1068 for (alt = alt_odb_list; alt; alt = alt->next) {
1069 name = alt->name;
1070 fill_sha1_path(name, sha1);
1071 fd = git_open_noatime(alt->base);
1072 if (fd >= 0)
1073 return fd;
1074 }
1075 return -1;
1076}
1077
1078static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1079{
1080 void *map;
1081 int fd;
1082
1083 fd = open_sha1_file(sha1);
1084 map = NULL;
1085 if (fd >= 0) {
1086 struct stat st;
1087
1088 if (!fstat(fd, &st)) {
1089 *size = xsize_t(st.st_size);
1090 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1091 }
1092 close(fd);
1093 }
1094 return map;
1095}
1096
1097static int legacy_loose_object(unsigned char *map)
1098{
1099 unsigned int word;
1100
1101 /*
1102 * Is it a zlib-compressed buffer? If so, the first byte
1103 * must be 0x78 (15-bit window size, deflated), and the
1104 * first 16-bit word is evenly divisible by 31
1105 */
1106 word = (map[0] << 8) + map[1];
1107 if (map[0] == 0x78 && !(word % 31))
1108 return 1;
1109 else
1110 return 0;
1111}
1112
1113unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1114{
1115 unsigned shift;
1116 unsigned char c;
1117 unsigned long size;
1118 unsigned long used = 0;
1119
1120 c = buf[used++];
1121 *type = (c >> 4) & 7;
1122 size = c & 15;
1123 shift = 4;
1124 while (c & 0x80) {
1125 if (len <= used)
1126 return 0;
1127 if (sizeof(long) * 8 <= shift)
1128 return 0;
1129 c = buf[used++];
1130 size += (c & 0x7f) << shift;
1131 shift += 7;
1132 }
1133 *sizep = size;
1134 return used;
1135}
1136
1137static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1138{
1139 unsigned long size, used;
1140 static const char valid_loose_object_type[8] = {
1141 0, /* OBJ_EXT */
1142 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1143 0, /* "delta" and others are invalid in a loose object */
1144 };
1145 enum object_type type;
1146
1147 /* Get the data stream */
1148 memset(stream, 0, sizeof(*stream));
1149 stream->next_in = map;
1150 stream->avail_in = mapsize;
1151 stream->next_out = buffer;
1152 stream->avail_out = bufsiz;
1153
1154 if (legacy_loose_object(map)) {
1155 inflateInit(stream);
1156 return inflate(stream, 0);
1157 }
1158
1159
1160 /*
1161 * There used to be a second loose object header format which
1162 * was meant to mimic the in-pack format, allowing for direct
1163 * copy of the object data. This format turned up not to be
1164 * really worth it and we don't write it any longer. But we
1165 * can still read it.
1166 */
1167 used = unpack_object_header_gently(map, mapsize, &type, &size);
1168 if (!used || !valid_loose_object_type[type])
1169 return -1;
1170 map += used;
1171 mapsize -= used;
1172
1173 /* Set up the stream for the rest.. */
1174 stream->next_in = map;
1175 stream->avail_in = mapsize;
1176 inflateInit(stream);
1177
1178 /* And generate the fake traditional header */
1179 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1180 typename(type), size);
1181 return 0;
1182}
1183
1184static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1185{
1186 int bytes = strlen(buffer) + 1;
1187 unsigned char *buf = xmalloc(1+size);
1188 unsigned long n;
1189 int status = Z_OK;
1190
1191 n = stream->total_out - bytes;
1192 if (n > size)
1193 n = size;
1194 memcpy(buf, (char *) buffer + bytes, n);
1195 bytes = n;
1196 if (bytes <= size) {
1197 /*
1198 * The above condition must be (bytes <= size), not
1199 * (bytes < size). In other words, even though we
1200 * expect no more output and set avail_out to zer0,
1201 * the input zlib stream may have bytes that express
1202 * "this concludes the stream", and we *do* want to
1203 * eat that input.
1204 *
1205 * Otherwise we would not be able to test that we
1206 * consumed all the input to reach the expected size;
1207 * we also want to check that zlib tells us that all
1208 * went well with status == Z_STREAM_END at the end.
1209 */
1210 stream->next_out = buf + bytes;
1211 stream->avail_out = size - bytes;
1212 while (status == Z_OK)
1213 status = inflate(stream, Z_FINISH);
1214 }
1215 buf[size] = 0;
1216 if (status == Z_STREAM_END && !stream->avail_in) {
1217 inflateEnd(stream);
1218 return buf;
1219 }
1220
1221 if (status < 0)
1222 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1223 else if (stream->avail_in)
1224 error("garbage at end of loose object '%s'",
1225 sha1_to_hex(sha1));
1226 free(buf);
1227 return NULL;
1228}
1229
1230/*
1231 * We used to just use "sscanf()", but that's actually way
1232 * too permissive for what we want to check. So do an anal
1233 * object header parse by hand.
1234 */
1235static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1236{
1237 char type[10];
1238 int i;
1239 unsigned long size;
1240
1241 /*
1242 * The type can be at most ten bytes (including the
1243 * terminating '\0' that we add), and is followed by
1244 * a space.
1245 */
1246 i = 0;
1247 for (;;) {
1248 char c = *hdr++;
1249 if (c == ' ')
1250 break;
1251 type[i++] = c;
1252 if (i >= sizeof(type))
1253 return -1;
1254 }
1255 type[i] = 0;
1256
1257 /*
1258 * The length must follow immediately, and be in canonical
1259 * decimal format (ie "010" is not valid).
1260 */
1261 size = *hdr++ - '0';
1262 if (size > 9)
1263 return -1;
1264 if (size) {
1265 for (;;) {
1266 unsigned long c = *hdr - '0';
1267 if (c > 9)
1268 break;
1269 hdr++;
1270 size = size * 10 + c;
1271 }
1272 }
1273 *sizep = size;
1274
1275 /*
1276 * The length must be followed by a zero byte
1277 */
1278 return *hdr ? -1 : type_from_string(type);
1279}
1280
1281static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1282{
1283 int ret;
1284 z_stream stream;
1285 char hdr[8192];
1286
1287 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1288 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1289 return NULL;
1290
1291 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1292}
1293
1294unsigned long get_size_from_delta(struct packed_git *p,
1295 struct pack_window **w_curs,
1296 off_t curpos)
1297{
1298 const unsigned char *data;
1299 unsigned char delta_head[20], *in;
1300 z_stream stream;
1301 int st;
1302
1303 memset(&stream, 0, sizeof(stream));
1304 stream.next_out = delta_head;
1305 stream.avail_out = sizeof(delta_head);
1306
1307 inflateInit(&stream);
1308 do {
1309 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1310 stream.next_in = in;
1311 st = inflate(&stream, Z_FINISH);
1312 curpos += stream.next_in - in;
1313 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1314 stream.total_out < sizeof(delta_head));
1315 inflateEnd(&stream);
1316 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1317 die("delta data unpack-initial failed");
1318
1319 /* Examine the initial part of the delta to figure out
1320 * the result size.
1321 */
1322 data = delta_head;
1323
1324 /* ignore base size */
1325 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1326
1327 /* Read the result size */
1328 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1329}
1330
1331static off_t get_delta_base(struct packed_git *p,
1332 struct pack_window **w_curs,
1333 off_t *curpos,
1334 enum object_type type,
1335 off_t delta_obj_offset)
1336{
1337 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1338 off_t base_offset;
1339
1340 /* use_pack() assured us we have [base_info, base_info + 20)
1341 * as a range that we can look at without walking off the
1342 * end of the mapped window. Its actually the hash size
1343 * that is assured. An OFS_DELTA longer than the hash size
1344 * is stupid, as then a REF_DELTA would be smaller to store.
1345 */
1346 if (type == OBJ_OFS_DELTA) {
1347 unsigned used = 0;
1348 unsigned char c = base_info[used++];
1349 base_offset = c & 127;
1350 while (c & 128) {
1351 base_offset += 1;
1352 if (!base_offset || MSB(base_offset, 7))
1353 return 0; /* overflow */
1354 c = base_info[used++];
1355 base_offset = (base_offset << 7) + (c & 127);
1356 }
1357 base_offset = delta_obj_offset - base_offset;
1358 if (base_offset >= delta_obj_offset)
1359 return 0; /* out of bound */
1360 *curpos += used;
1361 } else if (type == OBJ_REF_DELTA) {
1362 /* The base entry _must_ be in the same pack */
1363 base_offset = find_pack_entry_one(base_info, p);
1364 *curpos += 20;
1365 } else
1366 die("I am totally screwed");
1367 return base_offset;
1368}
1369
1370/* forward declaration for a mutually recursive function */
1371static int packed_object_info(struct packed_git *p, off_t offset,
1372 unsigned long *sizep);
1373
1374static int packed_delta_info(struct packed_git *p,
1375 struct pack_window **w_curs,
1376 off_t curpos,
1377 enum object_type type,
1378 off_t obj_offset,
1379 unsigned long *sizep)
1380{
1381 off_t base_offset;
1382
1383 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1384 type = packed_object_info(p, base_offset, NULL);
1385
1386 /* We choose to only get the type of the base object and
1387 * ignore potentially corrupt pack file that expects the delta
1388 * based on a base with a wrong size. This saves tons of
1389 * inflate() calls.
1390 */
1391 if (sizep)
1392 *sizep = get_size_from_delta(p, w_curs, curpos);
1393
1394 return type;
1395}
1396
1397static int unpack_object_header(struct packed_git *p,
1398 struct pack_window **w_curs,
1399 off_t *curpos,
1400 unsigned long *sizep)
1401{
1402 unsigned char *base;
1403 unsigned int left;
1404 unsigned long used;
1405 enum object_type type;
1406
1407 /* use_pack() assures us we have [base, base + 20) available
1408 * as a range that we can look at at. (Its actually the hash
1409 * size that is assured.) With our object header encoding
1410 * the maximum deflated object size is 2^137, which is just
1411 * insane, so we know won't exceed what we have been given.
1412 */
1413 base = use_pack(p, w_curs, *curpos, &left);
1414 used = unpack_object_header_gently(base, left, &type, sizep);
1415 if (!used)
1416 die("object offset outside of pack file");
1417 *curpos += used;
1418
1419 return type;
1420}
1421
1422const char *packed_object_info_detail(struct packed_git *p,
1423 off_t obj_offset,
1424 unsigned long *size,
1425 unsigned long *store_size,
1426 unsigned int *delta_chain_length,
1427 unsigned char *base_sha1)
1428{
1429 struct pack_window *w_curs = NULL;
1430 off_t curpos;
1431 unsigned long dummy;
1432 unsigned char *next_sha1;
1433 enum object_type type;
1434 struct revindex_entry *revidx;
1435
1436 *delta_chain_length = 0;
1437 curpos = obj_offset;
1438 type = unpack_object_header(p, &w_curs, &curpos, size);
1439
1440 revidx = find_pack_revindex(p, obj_offset);
1441 *store_size = revidx[1].offset - obj_offset;
1442
1443 for (;;) {
1444 switch (type) {
1445 default:
1446 die("pack %s contains unknown object type %d",
1447 p->pack_name, type);
1448 case OBJ_COMMIT:
1449 case OBJ_TREE:
1450 case OBJ_BLOB:
1451 case OBJ_TAG:
1452 unuse_pack(&w_curs);
1453 return typename(type);
1454 case OBJ_OFS_DELTA:
1455 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1456 if (!obj_offset)
1457 die("pack %s contains bad delta base reference of type %s",
1458 p->pack_name, typename(type));
1459 if (*delta_chain_length == 0) {
1460 revidx = find_pack_revindex(p, obj_offset);
1461 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1462 }
1463 break;
1464 case OBJ_REF_DELTA:
1465 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1466 if (*delta_chain_length == 0)
1467 hashcpy(base_sha1, next_sha1);
1468 obj_offset = find_pack_entry_one(next_sha1, p);
1469 break;
1470 }
1471 (*delta_chain_length)++;
1472 curpos = obj_offset;
1473 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1474 }
1475}
1476
1477static int packed_object_info(struct packed_git *p, off_t obj_offset,
1478 unsigned long *sizep)
1479{
1480 struct pack_window *w_curs = NULL;
1481 unsigned long size;
1482 off_t curpos = obj_offset;
1483 enum object_type type;
1484
1485 type = unpack_object_header(p, &w_curs, &curpos, &size);
1486
1487 switch (type) {
1488 case OBJ_OFS_DELTA:
1489 case OBJ_REF_DELTA:
1490 type = packed_delta_info(p, &w_curs, curpos,
1491 type, obj_offset, sizep);
1492 break;
1493 case OBJ_COMMIT:
1494 case OBJ_TREE:
1495 case OBJ_BLOB:
1496 case OBJ_TAG:
1497 if (sizep)
1498 *sizep = size;
1499 break;
1500 default:
1501 die("pack %s contains unknown object type %d",
1502 p->pack_name, type);
1503 }
1504 unuse_pack(&w_curs);
1505 return type;
1506}
1507
1508static void *unpack_compressed_entry(struct packed_git *p,
1509 struct pack_window **w_curs,
1510 off_t curpos,
1511 unsigned long size)
1512{
1513 int st;
1514 z_stream stream;
1515 unsigned char *buffer, *in;
1516
1517 buffer = xmalloc(size + 1);
1518 buffer[size] = 0;
1519 memset(&stream, 0, sizeof(stream));
1520 stream.next_out = buffer;
1521 stream.avail_out = size;
1522
1523 inflateInit(&stream);
1524 do {
1525 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1526 stream.next_in = in;
1527 st = inflate(&stream, Z_FINISH);
1528 curpos += stream.next_in - in;
1529 } while (st == Z_OK || st == Z_BUF_ERROR);
1530 inflateEnd(&stream);
1531 if ((st != Z_STREAM_END) || stream.total_out != size) {
1532 free(buffer);
1533 return NULL;
1534 }
1535
1536 return buffer;
1537}
1538
1539#define MAX_DELTA_CACHE (256)
1540
1541static size_t delta_base_cached;
1542
1543static struct delta_base_cache_lru_list {
1544 struct delta_base_cache_lru_list *prev;
1545 struct delta_base_cache_lru_list *next;
1546} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1547
1548static struct delta_base_cache_entry {
1549 struct delta_base_cache_lru_list lru;
1550 void *data;
1551 struct packed_git *p;
1552 off_t base_offset;
1553 unsigned long size;
1554 enum object_type type;
1555} delta_base_cache[MAX_DELTA_CACHE];
1556
1557static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1558{
1559 unsigned long hash;
1560
1561 hash = (unsigned long)p + (unsigned long)base_offset;
1562 hash += (hash >> 8) + (hash >> 16);
1563 return hash % MAX_DELTA_CACHE;
1564}
1565
1566static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1567 unsigned long *base_size, enum object_type *type, int keep_cache)
1568{
1569 void *ret;
1570 unsigned long hash = pack_entry_hash(p, base_offset);
1571 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1572
1573 ret = ent->data;
1574 if (!ret || ent->p != p || ent->base_offset != base_offset)
1575 return unpack_entry(p, base_offset, type, base_size);
1576
1577 if (!keep_cache) {
1578 ent->data = NULL;
1579 ent->lru.next->prev = ent->lru.prev;
1580 ent->lru.prev->next = ent->lru.next;
1581 delta_base_cached -= ent->size;
1582 } else {
1583 ret = xmemdupz(ent->data, ent->size);
1584 }
1585 *type = ent->type;
1586 *base_size = ent->size;
1587 return ret;
1588}
1589
1590static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1591{
1592 if (ent->data) {
1593 free(ent->data);
1594 ent->data = NULL;
1595 ent->lru.next->prev = ent->lru.prev;
1596 ent->lru.prev->next = ent->lru.next;
1597 delta_base_cached -= ent->size;
1598 }
1599}
1600
1601static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1602 void *base, unsigned long base_size, enum object_type type)
1603{
1604 unsigned long hash = pack_entry_hash(p, base_offset);
1605 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1606 struct delta_base_cache_lru_list *lru;
1607
1608 release_delta_base_cache(ent);
1609 delta_base_cached += base_size;
1610
1611 for (lru = delta_base_cache_lru.next;
1612 delta_base_cached > delta_base_cache_limit
1613 && lru != &delta_base_cache_lru;
1614 lru = lru->next) {
1615 struct delta_base_cache_entry *f = (void *)lru;
1616 if (f->type == OBJ_BLOB)
1617 release_delta_base_cache(f);
1618 }
1619 for (lru = delta_base_cache_lru.next;
1620 delta_base_cached > delta_base_cache_limit
1621 && lru != &delta_base_cache_lru;
1622 lru = lru->next) {
1623 struct delta_base_cache_entry *f = (void *)lru;
1624 release_delta_base_cache(f);
1625 }
1626
1627 ent->p = p;
1628 ent->base_offset = base_offset;
1629 ent->type = type;
1630 ent->data = base;
1631 ent->size = base_size;
1632 ent->lru.next = &delta_base_cache_lru;
1633 ent->lru.prev = delta_base_cache_lru.prev;
1634 delta_base_cache_lru.prev->next = &ent->lru;
1635 delta_base_cache_lru.prev = &ent->lru;
1636}
1637
1638static void *unpack_delta_entry(struct packed_git *p,
1639 struct pack_window **w_curs,
1640 off_t curpos,
1641 unsigned long delta_size,
1642 off_t obj_offset,
1643 enum object_type *type,
1644 unsigned long *sizep)
1645{
1646 void *delta_data, *result, *base;
1647 unsigned long base_size;
1648 off_t base_offset;
1649
1650 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1651 if (!base_offset) {
1652 error("failed to validate delta base reference "
1653 "at offset %"PRIuMAX" from %s",
1654 (uintmax_t)curpos, p->pack_name);
1655 return NULL;
1656 }
1657 unuse_pack(w_curs);
1658 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1659 if (!base) {
1660 /*
1661 * We're probably in deep shit, but let's try to fetch
1662 * the required base anyway from another pack or loose.
1663 * This is costly but should happen only in the presence
1664 * of a corrupted pack, and is better than failing outright.
1665 */
1666 struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1667 const unsigned char *base_sha1 =
1668 nth_packed_object_sha1(p, revidx->nr);
1669 error("failed to read delta base object %s"
1670 " at offset %"PRIuMAX" from %s",
1671 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1672 p->pack_name);
1673 mark_bad_packed_object(p, base_sha1);
1674 base = read_object(base_sha1, type, &base_size);
1675 if (!base)
1676 return NULL;
1677 }
1678
1679 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1680 if (!delta_data) {
1681 error("failed to unpack compressed delta "
1682 "at offset %"PRIuMAX" from %s",
1683 (uintmax_t)curpos, p->pack_name);
1684 free(base);
1685 return NULL;
1686 }
1687 result = patch_delta(base, base_size,
1688 delta_data, delta_size,
1689 sizep);
1690 if (!result)
1691 die("failed to apply delta");
1692 free(delta_data);
1693 add_delta_base_cache(p, base_offset, base, base_size, *type);
1694 return result;
1695}
1696
1697void *unpack_entry(struct packed_git *p, off_t obj_offset,
1698 enum object_type *type, unsigned long *sizep)
1699{
1700 struct pack_window *w_curs = NULL;
1701 off_t curpos = obj_offset;
1702 void *data;
1703
1704 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1705 switch (*type) {
1706 case OBJ_OFS_DELTA:
1707 case OBJ_REF_DELTA:
1708 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1709 obj_offset, type, sizep);
1710 break;
1711 case OBJ_COMMIT:
1712 case OBJ_TREE:
1713 case OBJ_BLOB:
1714 case OBJ_TAG:
1715 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1716 break;
1717 default:
1718 data = NULL;
1719 error("unknown object type %i at offset %"PRIuMAX" in %s",
1720 *type, (uintmax_t)obj_offset, p->pack_name);
1721 }
1722 unuse_pack(&w_curs);
1723 return data;
1724}
1725
1726const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1727 uint32_t n)
1728{
1729 const unsigned char *index = p->index_data;
1730 if (!index) {
1731 if (open_pack_index(p))
1732 return NULL;
1733 index = p->index_data;
1734 }
1735 if (n >= p->num_objects)
1736 return NULL;
1737 index += 4 * 256;
1738 if (p->index_version == 1) {
1739 return index + 24 * n + 4;
1740 } else {
1741 index += 8;
1742 return index + 20 * n;
1743 }
1744}
1745
1746off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1747{
1748 const unsigned char *index = p->index_data;
1749 index += 4 * 256;
1750 if (p->index_version == 1) {
1751 return ntohl(*((uint32_t *)(index + 24 * n)));
1752 } else {
1753 uint32_t off;
1754 index += 8 + p->num_objects * (20 + 4);
1755 off = ntohl(*((uint32_t *)(index + 4 * n)));
1756 if (!(off & 0x80000000))
1757 return off;
1758 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1759 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1760 ntohl(*((uint32_t *)(index + 4)));
1761 }
1762}
1763
1764off_t find_pack_entry_one(const unsigned char *sha1,
1765 struct packed_git *p)
1766{
1767 const uint32_t *level1_ofs = p->index_data;
1768 const unsigned char *index = p->index_data;
1769 unsigned hi, lo, stride;
1770 static int use_lookup = -1;
1771 static int debug_lookup = -1;
1772
1773 if (debug_lookup < 0)
1774 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1775
1776 if (!index) {
1777 if (open_pack_index(p))
1778 return 0;
1779 level1_ofs = p->index_data;
1780 index = p->index_data;
1781 }
1782 if (p->index_version > 1) {
1783 level1_ofs += 2;
1784 index += 8;
1785 }
1786 index += 4 * 256;
1787 hi = ntohl(level1_ofs[*sha1]);
1788 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1789 if (p->index_version > 1) {
1790 stride = 20;
1791 } else {
1792 stride = 24;
1793 index += 4;
1794 }
1795
1796 if (debug_lookup)
1797 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1798 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1799
1800 if (use_lookup < 0)
1801 use_lookup = !!getenv("GIT_USE_LOOKUP");
1802 if (use_lookup) {
1803 int pos = sha1_entry_pos(index, stride, 0,
1804 lo, hi, p->num_objects, sha1);
1805 if (pos < 0)
1806 return 0;
1807 return nth_packed_object_offset(p, pos);
1808 }
1809
1810 do {
1811 unsigned mi = (lo + hi) / 2;
1812 int cmp = hashcmp(index + mi * stride, sha1);
1813
1814 if (debug_lookup)
1815 printf("lo %u hi %u rg %u mi %u\n",
1816 lo, hi, hi - lo, mi);
1817 if (!cmp)
1818 return nth_packed_object_offset(p, mi);
1819 if (cmp > 0)
1820 hi = mi;
1821 else
1822 lo = mi+1;
1823 } while (lo < hi);
1824 return 0;
1825}
1826
1827int matches_pack_name(struct packed_git *p, const char *name)
1828{
1829 const char *last_c, *c;
1830
1831 if (!strcmp(p->pack_name, name))
1832 return 1;
1833
1834 for (c = p->pack_name, last_c = c; *c;)
1835 if (*c == '/')
1836 last_c = ++c;
1837 else
1838 ++c;
1839 if (!strcmp(last_c, name))
1840 return 1;
1841
1842 return 0;
1843}
1844
1845static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1846{
1847 static struct packed_git *last_found = (void *)1;
1848 struct packed_git *p;
1849 off_t offset;
1850
1851 prepare_packed_git();
1852 if (!packed_git)
1853 return 0;
1854 p = (last_found == (void *)1) ? packed_git : last_found;
1855
1856 do {
1857 if (ignore_packed) {
1858 const char **ig;
1859 for (ig = ignore_packed; *ig; ig++)
1860 if (matches_pack_name(p, *ig))
1861 break;
1862 if (*ig)
1863 goto next;
1864 }
1865
1866 if (p->num_bad_objects) {
1867 unsigned i;
1868 for (i = 0; i < p->num_bad_objects; i++)
1869 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1870 goto next;
1871 }
1872
1873 offset = find_pack_entry_one(sha1, p);
1874 if (offset) {
1875 /*
1876 * We are about to tell the caller where they can
1877 * locate the requested object. We better make
1878 * sure the packfile is still here and can be
1879 * accessed before supplying that answer, as
1880 * it may have been deleted since the index
1881 * was loaded!
1882 */
1883 if (p->pack_fd == -1 && open_packed_git(p)) {
1884 error("packfile %s cannot be accessed", p->pack_name);
1885 goto next;
1886 }
1887 e->offset = offset;
1888 e->p = p;
1889 hashcpy(e->sha1, sha1);
1890 last_found = p;
1891 return 1;
1892 }
1893
1894 next:
1895 if (p == last_found)
1896 p = packed_git;
1897 else
1898 p = p->next;
1899 if (p == last_found)
1900 p = p->next;
1901 } while (p);
1902 return 0;
1903}
1904
1905struct packed_git *find_sha1_pack(const unsigned char *sha1,
1906 struct packed_git *packs)
1907{
1908 struct packed_git *p;
1909
1910 for (p = packs; p; p = p->next) {
1911 if (find_pack_entry_one(sha1, p))
1912 return p;
1913 }
1914 return NULL;
1915
1916}
1917
1918static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1919{
1920 int status;
1921 unsigned long mapsize, size;
1922 void *map;
1923 z_stream stream;
1924 char hdr[32];
1925
1926 map = map_sha1_file(sha1, &mapsize);
1927 if (!map)
1928 return error("unable to find %s", sha1_to_hex(sha1));
1929 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1930 status = error("unable to unpack %s header",
1931 sha1_to_hex(sha1));
1932 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1933 status = error("unable to parse %s header", sha1_to_hex(sha1));
1934 else if (sizep)
1935 *sizep = size;
1936 inflateEnd(&stream);
1937 munmap(map, mapsize);
1938 return status;
1939}
1940
1941int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1942{
1943 struct pack_entry e;
1944 int status;
1945
1946 if (!find_pack_entry(sha1, &e, NULL)) {
1947 /* Most likely it's a loose object. */
1948 status = sha1_loose_object_info(sha1, sizep);
1949 if (status >= 0)
1950 return status;
1951
1952 /* Not a loose object; someone else may have just packed it. */
1953 reprepare_packed_git();
1954 if (!find_pack_entry(sha1, &e, NULL))
1955 return status;
1956 }
1957 return packed_object_info(e.p, e.offset, sizep);
1958}
1959
1960static void *read_packed_sha1(const unsigned char *sha1,
1961 enum object_type *type, unsigned long *size)
1962{
1963 struct pack_entry e;
1964 void *data;
1965
1966 if (!find_pack_entry(sha1, &e, NULL))
1967 return NULL;
1968 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1969 if (!data) {
1970 /*
1971 * We're probably in deep shit, but let's try to fetch
1972 * the required object anyway from another pack or loose.
1973 * This should happen only in the presence of a corrupted
1974 * pack, and is better than failing outright.
1975 */
1976 error("failed to read object %s at offset %"PRIuMAX" from %s",
1977 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1978 mark_bad_packed_object(e.p, sha1);
1979 data = read_object(sha1, type, size);
1980 }
1981 return data;
1982}
1983
1984/*
1985 * This is meant to hold a *small* number of objects that you would
1986 * want read_sha1_file() to be able to return, but yet you do not want
1987 * to write them into the object store (e.g. a browse-only
1988 * application).
1989 */
1990static struct cached_object {
1991 unsigned char sha1[20];
1992 enum object_type type;
1993 void *buf;
1994 unsigned long size;
1995} *cached_objects;
1996static int cached_object_nr, cached_object_alloc;
1997
1998static struct cached_object empty_tree = {
1999 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
2000 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
2001 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
2002 OBJ_TREE,
2003 "",
2004 0
2005};
2006
2007static struct cached_object *find_cached_object(const unsigned char *sha1)
2008{
2009 int i;
2010 struct cached_object *co = cached_objects;
2011
2012 for (i = 0; i < cached_object_nr; i++, co++) {
2013 if (!hashcmp(co->sha1, sha1))
2014 return co;
2015 }
2016 if (!hashcmp(sha1, empty_tree.sha1))
2017 return &empty_tree;
2018 return NULL;
2019}
2020
2021int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2022 unsigned char *sha1)
2023{
2024 struct cached_object *co;
2025
2026 hash_sha1_file(buf, len, typename(type), sha1);
2027 if (has_sha1_file(sha1) || find_cached_object(sha1))
2028 return 0;
2029 if (cached_object_alloc <= cached_object_nr) {
2030 cached_object_alloc = alloc_nr(cached_object_alloc);
2031 cached_objects = xrealloc(cached_objects,
2032 sizeof(*cached_objects) *
2033 cached_object_alloc);
2034 }
2035 co = &cached_objects[cached_object_nr++];
2036 co->size = len;
2037 co->type = type;
2038 co->buf = xmalloc(len);
2039 memcpy(co->buf, buf, len);
2040 hashcpy(co->sha1, sha1);
2041 return 0;
2042}
2043
2044void *read_object(const unsigned char *sha1, enum object_type *type,
2045 unsigned long *size)
2046{
2047 unsigned long mapsize;
2048 void *map, *buf;
2049 struct cached_object *co;
2050
2051 co = find_cached_object(sha1);
2052 if (co) {
2053 *type = co->type;
2054 *size = co->size;
2055 return xmemdupz(co->buf, co->size);
2056 }
2057
2058 buf = read_packed_sha1(sha1, type, size);
2059 if (buf)
2060 return buf;
2061 map = map_sha1_file(sha1, &mapsize);
2062 if (map) {
2063 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2064 munmap(map, mapsize);
2065 return buf;
2066 }
2067 reprepare_packed_git();
2068 return read_packed_sha1(sha1, type, size);
2069}
2070
2071void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2072 unsigned long *size)
2073{
2074 void *data = read_object(sha1, type, size);
2075 /* legacy behavior is to die on corrupted objects */
2076 if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2077 die("object %s is corrupted", sha1_to_hex(sha1));
2078 return data;
2079}
2080
2081void *read_object_with_reference(const unsigned char *sha1,
2082 const char *required_type_name,
2083 unsigned long *size,
2084 unsigned char *actual_sha1_return)
2085{
2086 enum object_type type, required_type;
2087 void *buffer;
2088 unsigned long isize;
2089 unsigned char actual_sha1[20];
2090
2091 required_type = type_from_string(required_type_name);
2092 hashcpy(actual_sha1, sha1);
2093 while (1) {
2094 int ref_length = -1;
2095 const char *ref_type = NULL;
2096
2097 buffer = read_sha1_file(actual_sha1, &type, &isize);
2098 if (!buffer)
2099 return NULL;
2100 if (type == required_type) {
2101 *size = isize;
2102 if (actual_sha1_return)
2103 hashcpy(actual_sha1_return, actual_sha1);
2104 return buffer;
2105 }
2106 /* Handle references */
2107 else if (type == OBJ_COMMIT)
2108 ref_type = "tree ";
2109 else if (type == OBJ_TAG)
2110 ref_type = "object ";
2111 else {
2112 free(buffer);
2113 return NULL;
2114 }
2115 ref_length = strlen(ref_type);
2116
2117 if (ref_length + 40 > isize ||
2118 memcmp(buffer, ref_type, ref_length) ||
2119 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2120 free(buffer);
2121 return NULL;
2122 }
2123 free(buffer);
2124 /* Now we have the ID of the referred-to object in
2125 * actual_sha1. Check again. */
2126 }
2127}
2128
2129static void write_sha1_file_prepare(const void *buf, unsigned long len,
2130 const char *type, unsigned char *sha1,
2131 char *hdr, int *hdrlen)
2132{
2133 git_SHA_CTX c;
2134
2135 /* Generate the header */
2136 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2137
2138 /* Sha1.. */
2139 git_SHA1_Init(&c);
2140 git_SHA1_Update(&c, hdr, *hdrlen);
2141 git_SHA1_Update(&c, buf, len);
2142 git_SHA1_Final(sha1, &c);
2143}
2144
2145/*
2146 * Move the just written object into its final resting place
2147 */
2148int move_temp_to_file(const char *tmpfile, const char *filename)
2149{
2150 int ret = 0;
2151 if (link(tmpfile, filename))
2152 ret = errno;
2153
2154 /*
2155 * Coda hack - coda doesn't like cross-directory links,
2156 * so we fall back to a rename, which will mean that it
2157 * won't be able to check collisions, but that's not a
2158 * big deal.
2159 *
2160 * The same holds for FAT formatted media.
2161 *
2162 * When this succeeds, we just return 0. We have nothing
2163 * left to unlink.
2164 */
2165 if (ret && ret != EEXIST) {
2166 if (!rename(tmpfile, filename))
2167 return 0;
2168 ret = errno;
2169 }
2170 unlink(tmpfile);
2171 if (ret) {
2172 if (ret != EEXIST) {
2173 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2174 }
2175 /* FIXME!!! Collision check here ? */
2176 }
2177
2178 return 0;
2179}
2180
2181static int write_buffer(int fd, const void *buf, size_t len)
2182{
2183 if (write_in_full(fd, buf, len) < 0)
2184 return error("file write error (%s)", strerror(errno));
2185 return 0;
2186}
2187
2188int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2189 unsigned char *sha1)
2190{
2191 char hdr[32];
2192 int hdrlen;
2193 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2194 return 0;
2195}
2196
2197/* Finalize a file on disk, and close it. */
2198static void close_sha1_file(int fd)
2199{
2200 if (fsync_object_files)
2201 fsync_or_die(fd, "sha1 file");
2202 fchmod(fd, 0444);
2203 if (close(fd) != 0)
2204 die("unable to write sha1 file");
2205}
2206
2207/* Size of directory component, including the ending '/' */
2208static inline int directory_size(const char *filename)
2209{
2210 const char *s = strrchr(filename, '/');
2211 if (!s)
2212 return 0;
2213 return s - filename + 1;
2214}
2215
2216/*
2217 * This creates a temporary file in the same directory as the final
2218 * 'filename'
2219 *
2220 * We want to avoid cross-directory filename renames, because those
2221 * can have problems on various filesystems (FAT, NFS, Coda).
2222 */
2223static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2224{
2225 int fd, dirlen = directory_size(filename);
2226
2227 if (dirlen + 20 > bufsiz) {
2228 errno = ENAMETOOLONG;
2229 return -1;
2230 }
2231 memcpy(buffer, filename, dirlen);
2232 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2233 fd = mkstemp(buffer);
2234 if (fd < 0 && dirlen) {
2235 /* Make sure the directory exists */
2236 memcpy(buffer, filename, dirlen);
2237 buffer[dirlen-1] = 0;
2238 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2239 return -1;
2240
2241 /* Try again */
2242 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2243 fd = mkstemp(buffer);
2244 }
2245 return fd;
2246}
2247
2248static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2249 void *buf, unsigned long len, time_t mtime)
2250{
2251 int fd, size, ret;
2252 unsigned char *compressed;
2253 z_stream stream;
2254 char *filename;
2255 static char tmpfile[PATH_MAX];
2256
2257 filename = sha1_file_name(sha1);
2258 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2259 if (fd < 0) {
2260 if (errno == EPERM)
2261 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2262 else
2263 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2264 }
2265
2266 /* Set it up */
2267 memset(&stream, 0, sizeof(stream));
2268 deflateInit(&stream, zlib_compression_level);
2269 size = 8 + deflateBound(&stream, len+hdrlen);
2270 compressed = xmalloc(size);
2271
2272 /* Compress it */
2273 stream.next_out = compressed;
2274 stream.avail_out = size;
2275
2276 /* First header.. */
2277 stream.next_in = (unsigned char *)hdr;
2278 stream.avail_in = hdrlen;
2279 while (deflate(&stream, 0) == Z_OK)
2280 /* nothing */;
2281
2282 /* Then the data itself.. */
2283 stream.next_in = buf;
2284 stream.avail_in = len;
2285 ret = deflate(&stream, Z_FINISH);
2286 if (ret != Z_STREAM_END)
2287 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2288
2289 ret = deflateEnd(&stream);
2290 if (ret != Z_OK)
2291 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2292
2293 size = stream.total_out;
2294
2295 if (write_buffer(fd, compressed, size) < 0)
2296 die("unable to write sha1 file");
2297 close_sha1_file(fd);
2298 free(compressed);
2299
2300 if (mtime) {
2301 struct utimbuf utb;
2302 utb.actime = mtime;
2303 utb.modtime = mtime;
2304 if (utime(tmpfile, &utb) < 0)
2305 warning("failed utime() on %s: %s",
2306 tmpfile, strerror(errno));
2307 }
2308
2309 return move_temp_to_file(tmpfile, filename);
2310}
2311
2312int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2313{
2314 unsigned char sha1[20];
2315 char hdr[32];
2316 int hdrlen;
2317
2318 /* Normally if we have it in the pack then we do not bother writing
2319 * it out into .git/objects/??/?{38} file.
2320 */
2321 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2322 if (returnsha1)
2323 hashcpy(returnsha1, sha1);
2324 if (has_sha1_file(sha1))
2325 return 0;
2326 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2327}
2328
2329int force_object_loose(const unsigned char *sha1, time_t mtime)
2330{
2331 void *buf;
2332 unsigned long len;
2333 enum object_type type;
2334 char hdr[32];
2335 int hdrlen;
2336 int ret;
2337
2338 if (has_loose_object(sha1))
2339 return 0;
2340 buf = read_packed_sha1(sha1, &type, &len);
2341 if (!buf)
2342 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2343 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2344 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2345 free(buf);
2346
2347 return ret;
2348}
2349
2350int has_pack_index(const unsigned char *sha1)
2351{
2352 struct stat st;
2353 if (stat(sha1_pack_index_name(sha1), &st))
2354 return 0;
2355 return 1;
2356}
2357
2358int has_pack_file(const unsigned char *sha1)
2359{
2360 struct stat st;
2361 if (stat(sha1_pack_name(sha1), &st))
2362 return 0;
2363 return 1;
2364}
2365
2366int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2367{
2368 struct pack_entry e;
2369 return find_pack_entry(sha1, &e, ignore_packed);
2370}
2371
2372int has_sha1_file(const unsigned char *sha1)
2373{
2374 struct pack_entry e;
2375
2376 if (find_pack_entry(sha1, &e, NULL))
2377 return 1;
2378 return has_loose_object(sha1);
2379}
2380
2381static int index_mem(unsigned char *sha1, void *buf, size_t size,
2382 int write_object, enum object_type type, const char *path)
2383{
2384 int ret, re_allocated = 0;
2385
2386 if (!type)
2387 type = OBJ_BLOB;
2388
2389 /*
2390 * Convert blobs to git internal format
2391 */
2392 if ((type == OBJ_BLOB) && path) {
2393 struct strbuf nbuf = STRBUF_INIT;
2394 if (convert_to_git(path, buf, size, &nbuf,
2395 write_object ? safe_crlf : 0)) {
2396 buf = strbuf_detach(&nbuf, &size);
2397 re_allocated = 1;
2398 }
2399 }
2400
2401 if (write_object)
2402 ret = write_sha1_file(buf, size, typename(type), sha1);
2403 else
2404 ret = hash_sha1_file(buf, size, typename(type), sha1);
2405 if (re_allocated)
2406 free(buf);
2407 return ret;
2408}
2409
2410int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2411 enum object_type type, const char *path)
2412{
2413 int ret;
2414 size_t size = xsize_t(st->st_size);
2415
2416 if (!S_ISREG(st->st_mode)) {
2417 struct strbuf sbuf = STRBUF_INIT;
2418 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2419 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2420 type, path);
2421 else
2422 ret = -1;
2423 strbuf_release(&sbuf);
2424 } else if (size) {
2425 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2426 ret = index_mem(sha1, buf, size, write_object, type, path);
2427 munmap(buf, size);
2428 } else
2429 ret = index_mem(sha1, NULL, size, write_object, type, path);
2430 close(fd);
2431 return ret;
2432}
2433
2434int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2435{
2436 int fd;
2437 char *target;
2438 size_t len;
2439
2440 switch (st->st_mode & S_IFMT) {
2441 case S_IFREG:
2442 fd = open(path, O_RDONLY);
2443 if (fd < 0)
2444 return error("open(\"%s\"): %s", path,
2445 strerror(errno));
2446 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2447 return error("%s: failed to insert into database",
2448 path);
2449 break;
2450 case S_IFLNK:
2451 len = xsize_t(st->st_size);
2452 target = xmalloc(len + 1);
2453 if (readlink(path, target, len + 1) != st->st_size) {
2454 char *errstr = strerror(errno);
2455 free(target);
2456 return error("readlink(\"%s\"): %s", path,
2457 errstr);
2458 }
2459 if (!write_object)
2460 hash_sha1_file(target, len, blob_type, sha1);
2461 else if (write_sha1_file(target, len, blob_type, sha1))
2462 return error("%s: failed to insert into database",
2463 path);
2464 free(target);
2465 break;
2466 case S_IFDIR:
2467 return resolve_gitlink_ref(path, "HEAD", sha1);
2468 default:
2469 return error("%s: unsupported file type", path);
2470 }
2471 return 0;
2472}
2473
2474int read_pack_header(int fd, struct pack_header *header)
2475{
2476 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2477 /* "eof before pack header was fully read" */
2478 return PH_ERROR_EOF;
2479
2480 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2481 /* "protocol error (pack signature mismatch detected)" */
2482 return PH_ERROR_PACK_SIGNATURE;
2483 if (!pack_version_ok(header->hdr_version))
2484 /* "protocol error (pack version unsupported)" */
2485 return PH_ERROR_PROTOCOL;
2486 return 0;
2487}