safe_create_leading_directories(): split on first of multiple slashes
[git/git.git] / sha1_file.c
CommitLineData
0fcfd160
LT
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 */
0fcfd160 9#include "cache.h"
6eac50d8 10#include "string-list.h"
1f688557 11#include "delta.h"
a733cb60 12#include "pack.h"
8e440259
PE
13#include "blob.h"
14#include "commit.h"
4dd1fbc7 15#include "run-command.h"
8e440259
PE
16#include "tag.h"
17#include "tree.h"
c879daa2 18#include "tree-walk.h"
f35a6d3b 19#include "refs.h"
70f5d5d3 20#include "pack-revindex.h"
628522ec 21#include "sha1-lookup.h"
568508e7 22#include "bulk-checkin.h"
090ea126 23#include "streaming.h"
543c5caa 24#include "dir.h"
0fcfd160 25
144bde78
LT
26#ifndef O_NOATIME
27#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
28#define O_NOATIME 01000000
29#else
30#define O_NOATIME 0
31#endif
32#endif
33
28bd70d8
JN
34#define SZ_FMT PRIuMAX
35static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 36
96f1e58f 37const unsigned char null_sha1[20];
88cd621d 38
b12ca963
NTND
39static const char *no_log_pack_access = "no_log_pack_access";
40static const char *log_pack_access;
41
c597ba80
NTND
42/*
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
46 * application).
47 */
48static struct cached_object {
49 unsigned char sha1[20];
50 enum object_type type;
51 void *buf;
52 unsigned long size;
53} *cached_objects;
54static int cached_object_nr, cached_object_alloc;
55
56static struct cached_object empty_tree = {
dab0d410 57 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
58 OBJ_TREE,
59 "",
60 0
61};
62
c01f51cc
NTND
63static struct packed_git *last_found_pack;
64
c597ba80
NTND
65static struct cached_object *find_cached_object(const unsigned char *sha1)
66{
67 int i;
68 struct cached_object *co = cached_objects;
69
70 for (i = 0; i < cached_object_nr; i++, co++) {
71 if (!hashcmp(co->sha1, sha1))
72 return co;
73 }
74 if (!hashcmp(sha1, empty_tree.sha1))
75 return &empty_tree;
76 return NULL;
77}
78
90a6464b
JH
79int mkdir_in_gitdir(const char *path)
80{
81 if (mkdir(path, 0777)) {
82 int saved_errno = errno;
83 struct stat st;
84 struct strbuf sb = STRBUF_INIT;
85
86 if (errno != EEXIST)
87 return -1;
88 /*
89 * Are we looking at a path in a symlinked worktree
90 * whose original repository does not yet have it?
91 * e.g. .git/rr-cache pointing at its original
92 * repository in which the user hasn't performed any
93 * conflict resolution yet?
94 */
95 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
96 strbuf_readlink(&sb, path, st.st_size) ||
97 !is_absolute_path(sb.buf) ||
98 mkdir(sb.buf, 0777)) {
99 strbuf_release(&sb);
100 errno = saved_errno;
101 return -1;
102 }
103 strbuf_release(&sb);
104 }
105 return adjust_shared_perm(path);
106}
107
b2cb9425
JH
108int safe_create_leading_directories(char *path)
109{
26c8ae2a 110 char *next_component = path + offset_1st_component(path);
67d42212 111
26c8ae2a 112 while (next_component) {
f0502332 113 struct stat st;
26c8ae2a 114 char *slash = strchr(next_component, '/');
f0502332 115
831651fd 116 if (!slash)
b2cb9425 117 break;
bf10cf70 118
26c8ae2a 119 next_component = slash + 1;
bf10cf70
MH
120 while (*next_component == '/')
121 next_component++;
26c8ae2a 122 if (!*next_component)
5f0bdf50 123 break;
831651fd
MH
124
125 *slash = '\0';
67d42212
JR
126 if (!stat(path, &st)) {
127 /* path exists */
128 if (!S_ISDIR(st.st_mode)) {
831651fd 129 *slash = '/';
67d42212 130 return -3;
b2cb9425 131 }
53a39721 132 } else if (mkdir(path, 0777)) {
928734d9
SW
133 if (errno == EEXIST &&
134 !stat(path, &st) && S_ISDIR(st.st_mode)) {
135 ; /* somebody created it since we checked */
136 } else {
831651fd 137 *slash = '/';
928734d9
SW
138 return -1;
139 }
53a39721 140 } else if (adjust_shared_perm(path)) {
831651fd 141 *slash = '/';
457f06d6
JS
142 return -2;
143 }
831651fd 144 *slash = '/';
b2cb9425
JH
145 }
146 return 0;
147}
723c31fe 148
8e21d63b
JK
149int safe_create_leading_directories_const(const char *path)
150{
151 /* path points to cache entries, so xstrdup before messing with it */
152 char *buf = xstrdup(path);
153 int result = safe_create_leading_directories(buf);
154 free(buf);
155 return result;
156}
157
ace1534d
JH
158static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
159{
160 int i;
161 for (i = 0; i < 20; i++) {
162 static char hex[] = "0123456789abcdef";
163 unsigned int val = sha1[i];
164 char *pos = pathbuf + i*2 + (i > 0);
165 *pos++ = hex[val >> 4];
166 *pos = hex[val & 0xf];
167 }
168}
169
0fcfd160
LT
170/*
171 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 172 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 173 * filename.
ace1534d
JH
174 *
175 * Also note that this returns the location for creating. Reading
176 * SHA1 file can happen from any alternate directory listed in the
d19938ab 177 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 178 * the primary object database.
0fcfd160
LT
179 */
180char *sha1_file_name(const unsigned char *sha1)
181{
560fb6a1
JK
182 static char buf[PATH_MAX];
183 const char *objdir;
184 int len;
0fcfd160 185
560fb6a1
JK
186 objdir = get_object_directory();
187 len = strlen(objdir);
188
189 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
190 if (len + 43 > PATH_MAX)
191 die("insanely long object directory %s", objdir);
192 memcpy(buf, objdir, len);
193 buf[len] = '/';
194 buf[len+3] = '/';
195 buf[len+42] = '\0';
196 fill_sha1_path(buf + len + 1, sha1);
197 return buf;
0fcfd160
LT
198}
199
633f43e1 200static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 201 char **name, char **base, const char *which)
bf592c50
DB
202{
203 static const char hex[] = "0123456789abcdef";
633f43e1 204 char *buf;
bf592c50
DB
205 int i;
206
633f43e1 207 if (!*base) {
bf592c50
DB
208 const char *sha1_file_directory = get_object_directory();
209 int len = strlen(sha1_file_directory);
633f43e1 210 *base = xmalloc(len + 60);
6eec46bd
JH
211 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
212 sha1_file_directory, which);
633f43e1 213 *name = *base + len + 11;
bf592c50
DB
214 }
215
633f43e1 216 buf = *name;
bf592c50
DB
217
218 for (i = 0; i < 20; i++) {
219 unsigned int val = *sha1++;
220 *buf++ = hex[val >> 4];
221 *buf++ = hex[val & 0xf];
222 }
a6080a0a 223
633f43e1 224 return *base;
bf592c50
DB
225}
226
633f43e1 227char *sha1_pack_name(const unsigned char *sha1)
bf592c50 228{
633f43e1 229 static char *name, *base;
bf592c50 230
6eec46bd 231 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 232}
bf592c50 233
633f43e1
HO
234char *sha1_pack_index_name(const unsigned char *sha1)
235{
236 static char *name, *base;
a6080a0a 237
6eec46bd 238 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
239}
240
d5a63b99
JH
241struct alternate_object_database *alt_odb_list;
242static struct alternate_object_database **alt_odb_tail;
ace1534d 243
f4e51683 244static int git_open_noatime(const char *name);
c2f493a4 245
ddd5d056
JH
246/*
247 * Prepare alternate object database registry.
d5a63b99
JH
248 *
249 * The variable alt_odb_list points at the list of struct
250 * alternate_object_database. The elements on this list come from
251 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
252 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
253 * whose contents is similar to that environment variable but can be
254 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
255 * contains "/the/directory/corresponding/to/.git/objects/...", while
256 * its name points just after the slash at the end of ".git/objects/"
257 * in the example above, and has enough space to hold 40-byte hex
258 * SHA1, an extra slash for the first level indirection, and the
259 * terminating NUL.
ddd5d056 260 */
6eac50d8 261static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
ace1534d 262{
1494e038 263 const char *objdir = get_object_directory();
c2f493a4
MW
264 struct alternate_object_database *ent;
265 struct alternate_object_database *alt;
5bdf0a84
HW
266 int pfxlen, entlen;
267 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 268
85dadc38 269 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
270 strbuf_addstr(&pathbuf, real_path(relative_base));
271 strbuf_addch(&pathbuf, '/');
c2f493a4 272 }
6eac50d8 273 strbuf_addstr(&pathbuf, entry);
c2f493a4 274
5bdf0a84
HW
275 normalize_path_copy(pathbuf.buf, pathbuf.buf);
276
277 pfxlen = strlen(pathbuf.buf);
278
279 /*
280 * The trailing slash after the directory name is given by
281 * this function at the end. Remove duplicates.
282 */
283 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
284 pfxlen -= 1;
285
286 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
287 ent = xmalloc(sizeof(*ent) + entlen);
288 memcpy(ent->base, pathbuf.buf, pfxlen);
289 strbuf_release(&pathbuf);
c2f493a4
MW
290
291 ent->name = ent->base + pfxlen + 1;
292 ent->base[pfxlen + 3] = '/';
293 ent->base[pfxlen] = ent->base[entlen-1] = 0;
294
295 /* Detect cases where alternate disappeared */
90b4a71c 296 if (!is_directory(ent->base)) {
c2f493a4
MW
297 error("object directory %s does not exist; "
298 "check .git/objects/info/alternates.",
299 ent->base);
300 free(ent);
301 return -1;
302 }
303
304 /* Prevent the common mistake of listing the same
305 * thing twice, or object directory itself.
306 */
307 for (alt = alt_odb_list; alt; alt = alt->next) {
308 if (!memcmp(ent->base, alt->base, pfxlen)) {
309 free(ent);
310 return -1;
311 }
312 }
cb2912c3 313 if (!strcmp(ent->base, objdir)) {
c2f493a4
MW
314 free(ent);
315 return -1;
316 }
317
318 /* add the alternate entry */
319 *alt_odb_tail = ent;
320 alt_odb_tail = &(ent->next);
321 ent->next = NULL;
322
323 /* recursively add alternates */
324 read_info_alternates(ent->base, depth + 1);
325
326 ent->base[pfxlen] = '/';
327
328 return 0;
329}
330
c5950164 331static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
332 const char *relative_base, int depth)
333{
6eac50d8
MH
334 struct string_list entries = STRING_LIST_INIT_NODUP;
335 char *alt_copy;
336 int i;
c2f493a4
MW
337
338 if (depth > 5) {
339 error("%s: ignoring alternate object stores, nesting too deep.",
340 relative_base);
341 return;
342 }
343
c5950164 344 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
345 string_list_split_in_place(&entries, alt_copy, sep, -1);
346 for (i = 0; i < entries.nr; i++) {
347 const char *entry = entries.items[i].string;
348 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 349 continue;
6eac50d8
MH
350 if (!is_absolute_path(entry) && depth) {
351 error("%s: ignoring relative alternate object store %s",
352 relative_base, entry);
353 } else {
354 link_alt_odb_entry(entry, relative_base, depth);
9577e7e3 355 }
9577e7e3 356 }
6eac50d8
MH
357 string_list_clear(&entries, 0);
358 free(alt_copy);
d5a63b99
JH
359}
360
5e73633d 361void read_info_alternates(const char * relative_base, int depth)
d5a63b99 362{
9577e7e3 363 char *map;
dc49cd76 364 size_t mapsz;
d5a63b99 365 struct stat st;
9cb18f56
JM
366 const char alt_file_name[] = "info/alternates";
367 /* Given that relative_base is no longer than PATH_MAX,
368 ensure that "path" has enough space to append "/", the
369 file name, "info/alternates", and a trailing NUL. */
370 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 371 int fd;
d5a63b99 372
9cb18f56 373 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 374 fd = git_open_noatime(path);
d5a63b99
JH
375 if (fd < 0)
376 return;
377 if (fstat(fd, &st) || (st.st_size == 0)) {
378 close(fd);
9a217f2a 379 return;
ace1534d 380 }
dc49cd76
SP
381 mapsz = xsize_t(st.st_size);
382 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 383 close(fd);
d5a63b99 384
c5950164 385 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 386
dc49cd76 387 munmap(map, mapsz);
ace1534d
JH
388}
389
bef70b22
DB
390void add_to_alternates_file(const char *reference)
391{
392 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 393 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
e6baf4a1 394 char *alt = mkpath("%s\n", reference);
bef70b22
DB
395 write_or_die(fd, alt, strlen(alt));
396 if (commit_lock_file(lock))
397 die("could not close alternates file");
398 if (alt_odb_tail)
c5950164 399 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
bef70b22
DB
400}
401
d79796bc
JH
402void foreach_alt_odb(alt_odb_fn fn, void *cb)
403{
404 struct alternate_object_database *ent;
405
406 prepare_alt_odb();
407 for (ent = alt_odb_list; ent; ent = ent->next)
408 if (fn(ent, cb))
409 return;
410}
411
c2f493a4
MW
412void prepare_alt_odb(void)
413{
554fe20d 414 const char *alt;
c2f493a4 415
7dc24aa5
SP
416 if (alt_odb_tail)
417 return;
418
c2f493a4
MW
419 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
420 if (!alt) alt = "";
421
c2f493a4 422 alt_odb_tail = &alt_odb_list;
c5950164 423 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
424
425 read_info_alternates(get_object_directory(), 0);
426}
427
0f4dc14a 428static int has_loose_object_local(const unsigned char *sha1)
ace1534d 429{
ace1534d 430 char *name = sha1_file_name(sha1);
0f4dc14a
BC
431 return !access(name, F_OK);
432}
ace1534d 433
0f4dc14a
BC
434int has_loose_object_nonlocal(const unsigned char *sha1)
435{
436 struct alternate_object_database *alt;
9a217f2a 437 prepare_alt_odb();
d5a63b99 438 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 439 fill_sha1_path(alt->name, sha1);
c529d75a
LT
440 if (!access(alt->base, F_OK))
441 return 1;
ace1534d 442 }
c529d75a 443 return 0;
ace1534d
JH
444}
445
0f4dc14a
BC
446static int has_loose_object(const unsigned char *sha1)
447{
448 return has_loose_object_local(sha1) ||
449 has_loose_object_nonlocal(sha1);
450}
451
60bb8b14 452static unsigned int pack_used_ctr;
a53128b6
SP
453static unsigned int pack_mmap_calls;
454static unsigned int peak_pack_open_windows;
455static unsigned int pack_open_windows;
c7934306
SP
456static unsigned int pack_open_fds;
457static unsigned int pack_max_fds;
a53128b6 458static size_t peak_pack_mapped;
60bb8b14 459static size_t pack_mapped;
9a217f2a 460struct packed_git *packed_git;
1f688557 461
b79d18c9 462void pack_report(void)
a53128b6
SP
463{
464 fprintf(stderr,
e05db0fd
PR
465 "pack_report: getpagesize() = %10" SZ_FMT "\n"
466 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
467 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
468 sz_fmt(getpagesize()),
469 sz_fmt(packed_git_window_size),
470 sz_fmt(packed_git_limit));
a53128b6
SP
471 fprintf(stderr,
472 "pack_report: pack_used_ctr = %10u\n"
473 "pack_report: pack_mmap_calls = %10u\n"
474 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
475 "pack_report: pack_mapped = "
476 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
477 pack_used_ctr,
478 pack_mmap_calls,
479 pack_open_windows, peak_pack_open_windows,
9e42d6a1 480 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
481}
482
42873078 483static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
484{
485 void *idx_map;
42873078 486 struct pack_idx_header *hdr;
dc49cd76 487 size_t idx_size;
74e34e1f 488 uint32_t version, nr, i, *index;
f4e51683 489 int fd = git_open_noatime(path);
1f688557 490 struct stat st;
42873078 491
1f688557
JH
492 if (fd < 0)
493 return -1;
494 if (fstat(fd, &st)) {
495 close(fd);
496 return -1;
497 }
dc49cd76 498 idx_size = xsize_t(st.st_size);
2d88451b
SP
499 if (idx_size < 4 * 256 + 20 + 20) {
500 close(fd);
501 return error("index file %s is too small", path);
502 }
c4712e45 503 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 504 close(fd);
1f688557 505
42873078
NP
506 hdr = idx_map;
507 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
508 version = ntohl(hdr->idx_version);
509 if (version < 2 || version > 2) {
510 munmap(idx_map, idx_size);
6e1c2344 511 return error("index file %s is version %"PRIu32
74e34e1f
NP
512 " and is not supported by this binary"
513 " (try upgrading GIT to a newer version)",
514 path, version);
515 }
516 } else
517 version = 1;
df1b059d 518
1f688557 519 nr = 0;
42873078 520 index = idx_map;
74e34e1f
NP
521 if (version > 1)
522 index += 2; /* skip index header */
1f688557 523 for (i = 0; i < 256; i++) {
326bf396 524 uint32_t n = ntohl(index[i]);
2d88451b
SP
525 if (n < nr) {
526 munmap(idx_map, idx_size);
df1b059d 527 return error("non-monotonic index %s", path);
2d88451b 528 }
1f688557
JH
529 nr = n;
530 }
531
74e34e1f
NP
532 if (version == 1) {
533 /*
534 * Total size:
535 * - 256 index entries 4 bytes each
536 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
537 * - 20-byte SHA1 of the packfile
538 * - 20-byte SHA1 file checksum
539 */
540 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
541 munmap(idx_map, idx_size);
eef427a0 542 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
543 }
544 } else if (version == 2) {
545 /*
546 * Minimum size:
547 * - 8 bytes of header
548 * - 256 index entries 4 bytes each
549 * - 20-byte sha1 entry * nr
550 * - 4-byte crc entry * nr
551 * - 4-byte offset entry * nr
552 * - 20-byte SHA1 of the packfile
553 * - 20-byte SHA1 file checksum
554 * And after the 4-byte offset table might be a
555 * variable sized table containing 8-byte entries
556 * for offsets larger than 2^31.
557 */
558 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
559 unsigned long max_size = min_size;
560 if (nr)
561 max_size += (nr - 1)*8;
562 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 563 munmap(idx_map, idx_size);
eef427a0 564 return error("wrong index v2 file size in %s", path);
74e34e1f 565 }
7109c889
JH
566 if (idx_size != min_size &&
567 /*
568 * make sure we can deal with large pack offsets.
569 * 31-bit signed offset won't be enough, neither
570 * 32-bit unsigned one will be.
571 */
572 (sizeof(off_t) <= 4)) {
573 munmap(idx_map, idx_size);
574 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 575 }
2d88451b 576 }
1f688557 577
74e34e1f 578 p->index_version = version;
42873078
NP
579 p->index_data = idx_map;
580 p->index_size = idx_size;
57059091 581 p->num_objects = nr;
1f688557
JH
582 return 0;
583}
584
bc8e478a 585int open_pack_index(struct packed_git *p)
d079837e
SP
586{
587 char *idx_name;
588 int ret;
589
590 if (p->index_data)
591 return 0;
592
593 idx_name = xstrdup(p->pack_name);
594 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
595 ret = check_packed_git_idx(idx_name, p);
596 free(idx_name);
597 return ret;
598}
599
11daf39b
SP
600static void scan_windows(struct packed_git *p,
601 struct packed_git **lru_p,
602 struct pack_window **lru_w,
603 struct pack_window **lru_l)
1f688557 604{
11daf39b
SP
605 struct pack_window *w, *w_l;
606
607 for (w_l = NULL, w = p->windows; w; w = w->next) {
608 if (!w->inuse_cnt) {
609 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
610 *lru_p = p;
611 *lru_w = w;
612 *lru_l = w_l;
54044bf8 613 }
54044bf8 614 }
11daf39b 615 w_l = w;
f9253394 616 }
11daf39b
SP
617}
618
7c3ecb32 619static int unuse_one_window(struct packed_git *current)
11daf39b
SP
620{
621 struct packed_git *p, *lru_p = NULL;
622 struct pack_window *lru_w = NULL, *lru_l = NULL;
623
624 if (current)
625 scan_windows(current, &lru_p, &lru_w, &lru_l);
626 for (p = packed_git; p; p = p->next)
627 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
628 if (lru_p) {
629 munmap(lru_w->base, lru_w->len);
630 pack_mapped -= lru_w->len;
631 if (lru_l)
632 lru_l->next = lru_w->next;
7c3ecb32 633 else
54044bf8 634 lru_p->windows = lru_w->next;
54044bf8 635 free(lru_w);
a53128b6 636 pack_open_windows--;
54044bf8
SP
637 return 1;
638 }
639 return 0;
f9253394
JH
640}
641
7c3ecb32 642void release_pack_memory(size_t need)
97bfeb34
SP
643{
644 size_t cur = pack_mapped;
7c3ecb32 645 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
646 ; /* nothing */
647}
648
58ecbd5e
JN
649void *xmmap(void *start, size_t length,
650 int prot, int flags, int fd, off_t offset)
651{
652 void *ret = mmap(start, length, prot, flags, fd, offset);
653 if (ret == MAP_FAILED) {
654 if (!length)
655 return NULL;
7c3ecb32 656 release_pack_memory(length);
58ecbd5e
JN
657 ret = mmap(start, length, prot, flags, fd, offset);
658 if (ret == MAP_FAILED)
659 die_errno("Out of memory? mmap failed");
660 }
661 return ret;
662}
663
c9ced051
SP
664void close_pack_windows(struct packed_git *p)
665{
666 while (p->windows) {
667 struct pack_window *w = p->windows;
668
669 if (w->inuse_cnt)
670 die("pack '%s' still has open windows to it",
671 p->pack_name);
672 munmap(w->base, w->len);
673 pack_mapped -= w->len;
674 pack_open_windows--;
675 p->windows = w->next;
676 free(w);
677 }
678}
679
88d0db55
BC
680/*
681 * The LRU pack is the one with the oldest MRU window, preferring packs
682 * with no used windows, or the oldest mtime if it has no windows allocated.
683 */
684static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
685{
686 struct pack_window *w, *this_mru_w;
687 int has_windows_inuse = 0;
688
689 /*
690 * Reject this pack if it has windows and the previously selected
691 * one does not. If this pack does not have windows, reject
692 * it if the pack file is newer than the previously selected one.
693 */
694 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
695 return;
696
697 for (w = this_mru_w = p->windows; w; w = w->next) {
698 /*
699 * Reject this pack if any of its windows are in use,
700 * but the previously selected pack did not have any
701 * inuse windows. Otherwise, record that this pack
702 * has windows in use.
703 */
704 if (w->inuse_cnt) {
705 if (*accept_windows_inuse)
706 has_windows_inuse = 1;
707 else
708 return;
709 }
710
711 if (w->last_used > this_mru_w->last_used)
712 this_mru_w = w;
713
714 /*
715 * Reject this pack if it has windows that have been
716 * used more recently than the previously selected pack.
717 * If the previously selected pack had windows inuse and
718 * we have not encountered a window in this pack that is
719 * inuse, skip this check since we prefer a pack with no
720 * inuse windows to one that has inuse windows.
721 */
722 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
723 this_mru_w->last_used > (*mru_w)->last_used)
724 return;
725 }
726
727 /*
728 * Select this pack.
729 */
730 *mru_w = this_mru_w;
731 *lru_p = p;
732 *accept_windows_inuse = has_windows_inuse;
733}
734
735static int close_one_pack(void)
736{
737 struct packed_git *p, *lru_p = NULL;
738 struct pack_window *mru_w = NULL;
739 int accept_windows_inuse = 1;
740
741 for (p = packed_git; p; p = p->next) {
742 if (p->pack_fd == -1)
743 continue;
744 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
745 }
746
747 if (lru_p) {
748 close(lru_p->pack_fd);
749 pack_open_fds--;
750 lru_p->pack_fd = -1;
751 return 1;
752 }
753
754 return 0;
755}
756
03e79c88 757void unuse_pack(struct pack_window **w_cursor)
f9253394 758{
03e79c88
SP
759 struct pack_window *w = *w_cursor;
760 if (w) {
761 w->inuse_cnt--;
762 *w_cursor = NULL;
763 }
1f688557
JH
764}
765
fa5fc15d
SP
766void close_pack_index(struct packed_git *p)
767{
768 if (p->index_data) {
769 munmap((void *)p->index_data, p->index_size);
770 p->index_data = NULL;
771 }
772}
773
c74faea1
NP
774/*
775 * This is used by git-repack in case a newly created pack happens to
776 * contain the same set of objects as an existing one. In that case
777 * the resulting file might be different even if its name would be the
778 * same. It is best to close any reference to the old pack before it is
779 * replaced on disk. Of course no index pointers nor windows for given pack
780 * must subsist at this point. If ever objects from this pack are requested
781 * again, the new version of the pack will be reinitialized through
782 * reprepare_packed_git().
783 */
784void free_pack_by_name(const char *pack_name)
785{
786 struct packed_git *p, **pp = &packed_git;
787
788 while (*pp) {
789 p = *pp;
790 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 791 clear_delta_base_cache();
c74faea1 792 close_pack_windows(p);
c7934306 793 if (p->pack_fd != -1) {
c74faea1 794 close(p->pack_fd);
c7934306
SP
795 pack_open_fds--;
796 }
fa5fc15d 797 close_pack_index(p);
c74faea1
NP
798 free(p->bad_object_sha1);
799 *pp = p->next;
c01f51cc
NTND
800 if (last_found_pack == p)
801 last_found_pack = NULL;
c74faea1
NP
802 free(p);
803 return;
804 }
805 pp = &p->next;
806 }
807}
808
a0788266
JS
809static unsigned int get_max_fd_limit(void)
810{
811#ifdef RLIMIT_NOFILE
812 struct rlimit lim;
813
814 if (getrlimit(RLIMIT_NOFILE, &lim))
815 die_errno("cannot get RLIMIT_NOFILE");
816
817 return lim.rlim_cur;
818#elif defined(_SC_OPEN_MAX)
819 return sysconf(_SC_OPEN_MAX);
820#elif defined(OPEN_MAX)
821 return OPEN_MAX;
822#else
823 return 1; /* see the caller ;-) */
824#endif
825}
826
3cf8b462
SP
827/*
828 * Do not call this directly as this leaks p->pack_fd on error return;
829 * call open_packed_git() instead.
830 */
831static int open_packed_git_1(struct packed_git *p)
1f688557 832{
9bc879c1
SP
833 struct stat st;
834 struct pack_header hdr;
835 unsigned char sha1[20];
836 unsigned char *idx_sha1;
2c039da8 837 long fd_flag;
9bc879c1 838
d079837e
SP
839 if (!p->index_data && open_pack_index(p))
840 return error("packfile %s index unavailable", p->pack_name);
841
c7934306 842 if (!pack_max_fds) {
a0788266 843 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
844
845 /* Save 3 for stdin/stdout/stderr, 22 for work */
846 if (25 < max_fds)
847 pack_max_fds = max_fds - 25;
848 else
849 pack_max_fds = 1;
850 }
851
88d0db55 852 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
853 ; /* nothing */
854
f4e51683 855 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 856 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 857 return -1;
c7934306 858 pack_open_fds++;
9bc879c1
SP
859
860 /* If we created the struct before we had the pack we lack size. */
bf592c50 861 if (!p->pack_size) {
bf592c50 862 if (!S_ISREG(st.st_mode))
072db278 863 return error("packfile %s not a regular file", p->pack_name);
bf592c50 864 p->pack_size = st.st_size;
9bc879c1 865 } else if (p->pack_size != st.st_size)
072db278 866 return error("packfile %s size changed", p->pack_name);
9bc879c1 867
2c039da8
JH
868 /* We leave these file descriptors open with sliding mmap;
869 * there is no point keeping them open across exec(), though.
870 */
871 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
872 if (fd_flag < 0)
072db278 873 return error("cannot determine file descriptor flags");
2c039da8
JH
874 fd_flag |= FD_CLOEXEC;
875 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 876 return error("cannot set FD_CLOEXEC");
2c039da8 877
9bc879c1 878 /* Verify we recognize this pack file format. */
e6e2bd62 879 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 880 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 881 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 882 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 883 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
884 return error("packfile %s is version %"PRIu32" and not"
885 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
886 p->pack_name, ntohl(hdr.hdr_version));
887
888 /* Verify the pack matches its index. */
57059091 889 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
890 return error("packfile %s claims to have %"PRIu32" objects"
891 " while index indicates %"PRIu32" objects",
57059091
NP
892 p->pack_name, ntohl(hdr.hdr_entries),
893 p->num_objects);
9bc879c1 894 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 895 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 896 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 897 return error("packfile %s signature is unavailable", p->pack_name);
42873078 898 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 899 if (hashcmp(sha1, idx_sha1))
072db278
SP
900 return error("packfile %s does not match index", p->pack_name);
901 return 0;
9bc879c1
SP
902}
903
3cf8b462
SP
904static int open_packed_git(struct packed_git *p)
905{
906 if (!open_packed_git_1(p))
907 return 0;
908 if (p->pack_fd != -1) {
909 close(p->pack_fd);
c7934306 910 pack_open_fds--;
3cf8b462
SP
911 p->pack_fd = -1;
912 }
913 return -1;
914}
915
c4001d92 916static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
917{
918 /* We must promise at least 20 bytes (one hash) after the
919 * offset is available from this window, otherwise the offset
920 * is not actually in this window and a different window (which
921 * has that one hash excess) must be used. This is to support
922 * the object header and delta base parsing routines below.
923 */
924 off_t win_off = win->offset;
925 return win_off <= offset
926 && (offset + 20) <= (win_off + win->len);
927}
928
4b25d091 929unsigned char *use_pack(struct packed_git *p,
03e79c88 930 struct pack_window **w_cursor,
c4001d92 931 off_t offset,
ef49a7a0 932 unsigned long *left)
9bc879c1 933{
60bb8b14 934 struct pack_window *win = *w_cursor;
03e79c88 935
a9d98a14 936 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
937 * pointless to ask for an offset into the middle of that
938 * hash, and the in_window function above wouldn't match
939 * don't allow an offset too close to the end of the file.
940 */
d131b7af
SP
941 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
942 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
943 if (offset > (p->pack_size - 20))
944 die("offset beyond end of packfile (truncated pack?)");
945
946 if (!win || !in_window(win, offset)) {
947 if (win)
948 win->inuse_cnt--;
949 for (win = p->windows; win; win = win->next) {
950 if (in_window(win, offset))
951 break;
952 }
953 if (!win) {
78a28df9 954 size_t window_align = packed_git_window_size / 2;
dc49cd76 955 off_t len;
d131b7af
SP
956
957 if (p->pack_fd == -1 && open_packed_git(p))
958 die("packfile %s cannot be accessed", p->pack_name);
959
60bb8b14 960 win = xcalloc(1, sizeof(*win));
78a28df9 961 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
962 len = p->pack_size - win->offset;
963 if (len > packed_git_window_size)
964 len = packed_git_window_size;
965 win->len = (size_t)len;
60bb8b14 966 pack_mapped += win->len;
11daf39b 967 while (packed_git_limit < pack_mapped
7c3ecb32 968 && unuse_one_window(p))
60bb8b14 969 ; /* nothing */
c4712e45 970 win->base = xmmap(NULL, win->len,
60bb8b14
SP
971 PROT_READ, MAP_PRIVATE,
972 p->pack_fd, win->offset);
973 if (win->base == MAP_FAILED)
73b4e4be
SP
974 die("packfile %s cannot be mapped: %s",
975 p->pack_name,
976 strerror(errno));
d131b7af
SP
977 if (!win->offset && win->len == p->pack_size
978 && !p->do_not_close) {
979 close(p->pack_fd);
980 pack_open_fds--;
981 p->pack_fd = -1;
982 }
a53128b6
SP
983 pack_mmap_calls++;
984 pack_open_windows++;
985 if (pack_mapped > peak_pack_mapped)
986 peak_pack_mapped = pack_mapped;
987 if (pack_open_windows > peak_pack_open_windows)
988 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
989 win->next = p->windows;
990 p->windows = win;
991 }
1f688557 992 }
03e79c88
SP
993 if (win != *w_cursor) {
994 win->last_used = pack_used_ctr++;
995 win->inuse_cnt++;
996 *w_cursor = win;
997 }
60bb8b14 998 offset -= win->offset;
03e79c88 999 if (left)
dc49cd76 1000 *left = win->len - xsize_t(offset);
03e79c88 1001 return win->base + offset;
1f688557
JH
1002}
1003
27d69a46
NP
1004static struct packed_git *alloc_packed_git(int extra)
1005{
1006 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1007 memset(p, 0, sizeof(*p));
1008 p->pack_fd = -1;
1009 return p;
1010}
1011
e0500293
JN
1012static void try_to_free_pack_memory(size_t size)
1013{
7c3ecb32 1014 release_pack_memory(size);
e0500293
JN
1015}
1016
42873078 1017struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 1018{
e0500293 1019 static int have_set_try_to_free_routine;
1f688557 1020 struct stat st;
27d69a46 1021 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 1022
e0500293
JN
1023 if (!have_set_try_to_free_routine) {
1024 have_set_try_to_free_routine = 1;
1025 set_try_to_free_routine(try_to_free_pack_memory);
1026 }
1027
42873078
NP
1028 /*
1029 * Make sure a corresponding .pack file exists and that
1030 * the index looks sane.
1031 */
1032 path_len -= strlen(".idx");
27d69a46
NP
1033 if (path_len < 1) {
1034 free(p);
1f688557 1035 return NULL;
27d69a46 1036 }
42873078 1037 memcpy(p->pack_name, path, path_len);
8d25931d
BC
1038
1039 strcpy(p->pack_name + path_len, ".keep");
1040 if (!access(p->pack_name, F_OK))
1041 p->pack_keep = 1;
1042
42873078 1043 strcpy(p->pack_name + path_len, ".pack");
d079837e 1044 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1045 free(p);
1f688557
JH
1046 return NULL;
1047 }
42873078 1048
1f688557
JH
1049 /* ok, it looks sane as far as we can check without
1050 * actually mapping the pack file.
1051 */
1f688557 1052 p->pack_size = st.st_size;
9d835df2 1053 p->pack_local = local;
b867092f 1054 p->mtime = st.st_mtime;
42873078
NP
1055 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1056 hashclr(p->sha1);
1f688557
JH
1057 return p;
1058}
1059
7b64469a 1060struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1061{
42873078 1062 const char *path = sha1_pack_name(sha1);
27d69a46 1063 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 1064
27d69a46
NP
1065 strcpy(p->pack_name, path);
1066 hashcpy(p->sha1, sha1);
42873078
NP
1067 if (check_packed_git_idx(idx_path, p)) {
1068 free(p);
bf592c50 1069 return NULL;
42873078 1070 }
bf592c50 1071
bf592c50
DB
1072 return p;
1073}
1074
1075void install_packed_git(struct packed_git *pack)
1076{
c7934306
SP
1077 if (pack->pack_fd != -1)
1078 pack_open_fds++;
1079
bf592c50
DB
1080 pack->next = packed_git;
1081 packed_git = pack;
1082}
1083
543c5caa
NTND
1084void (*report_garbage)(const char *desc, const char *path);
1085
1086static void report_helper(const struct string_list *list,
1087 int seen_bits, int first, int last)
1088{
1089 const char *msg;
1090 switch (seen_bits) {
1091 case 0:
1092 msg = "no corresponding .idx nor .pack";
1093 break;
1094 case 1:
1095 msg = "no corresponding .idx";
1096 break;
1097 case 2:
1098 msg = "no corresponding .pack";
1099 break;
1100 default:
1101 return;
1102 }
1103 for (; first < last; first++)
1104 report_garbage(msg, list->items[first].string);
1105}
1106
1107static void report_pack_garbage(struct string_list *list)
1108{
1109 int i, baselen = -1, first = 0, seen_bits = 0;
1110
1111 if (!report_garbage)
1112 return;
1113
1114 sort_string_list(list);
1115
1116 for (i = 0; i < list->nr; i++) {
1117 const char *path = list->items[i].string;
1118 if (baselen != -1 &&
1119 strncmp(path, list->items[first].string, baselen)) {
1120 report_helper(list, seen_bits, first, i);
1121 baselen = -1;
1122 seen_bits = 0;
1123 }
1124 if (baselen == -1) {
1125 const char *dot = strrchr(path, '.');
1126 if (!dot) {
1127 report_garbage("garbage found", path);
1128 continue;
1129 }
1130 baselen = dot - path + 1;
1131 first = i;
1132 }
1133 if (!strcmp(path + baselen, "pack"))
1134 seen_bits |= 1;
1135 else if (!strcmp(path + baselen, "idx"))
1136 seen_bits |= 2;
1137 }
1138 report_helper(list, seen_bits, first, list->nr);
1139}
1140
9d835df2 1141static void prepare_packed_git_one(char *objdir, int local)
1f688557 1142{
9cb18f56
JM
1143 /* Ensure that this buffer is large enough so that we can
1144 append "/pack/" without clobbering the stack even if
1145 strlen(objdir) were PATH_MAX. */
1146 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
1147 int len;
1148 DIR *dir;
1149 struct dirent *de;
543c5caa 1150 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557
JH
1151
1152 sprintf(path, "%s/pack", objdir);
1153 len = strlen(path);
1154 dir = opendir(path);
b5b16990 1155 if (!dir) {
26125f6b 1156 if (errno != ENOENT)
bd2afde8 1157 error("unable to open object pack directory: %s: %s",
26125f6b 1158 path, strerror(errno));
1f688557 1159 return;
b5b16990 1160 }
1f688557
JH
1161 path[len++] = '/';
1162 while ((de = readdir(dir)) != NULL) {
1163 int namelen = strlen(de->d_name);
1164 struct packed_git *p;
1165
543c5caa
NTND
1166 if (len + namelen + 1 > sizeof(path)) {
1167 if (report_garbage) {
1168 struct strbuf sb = STRBUF_INIT;
1169 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1170 report_garbage("path too long", sb.buf);
1171 strbuf_release(&sb);
1172 }
1173 continue;
1174 }
1175
1176 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1177 continue;
1178
1f688557 1179 strcpy(path + len, de->d_name);
d90906a9
NTND
1180
1181 if (has_extension(de->d_name, ".idx")) {
1182 /* Don't reopen a pack we already have. */
1183 for (p = packed_git; p; p = p->next) {
1184 if (!memcmp(path, p->pack_name, len + namelen - 4))
1185 break;
1186 }
1187 if (p == NULL &&
1188 /*
1189 * See if it really is a valid .idx file with
1190 * corresponding .pack file that we can map.
1191 */
1192 (p = add_packed_git(path, len + namelen, local)) != NULL)
1193 install_packed_git(p);
86f7780c 1194 }
543c5caa
NTND
1195
1196 if (!report_garbage)
1197 continue;
1198
1199 if (has_extension(de->d_name, ".idx") ||
1200 has_extension(de->d_name, ".pack") ||
1201 has_extension(de->d_name, ".keep"))
1202 string_list_append(&garbage, path);
1203 else
1204 report_garbage("garbage found", path);
1f688557 1205 }
5b35bcd5 1206 closedir(dir);
543c5caa
NTND
1207 report_pack_garbage(&garbage);
1208 string_list_clear(&garbage, 0);
1f688557
JH
1209}
1210
b867092f
JH
1211static int sort_pack(const void *a_, const void *b_)
1212{
1213 struct packed_git *a = *((struct packed_git **)a_);
1214 struct packed_git *b = *((struct packed_git **)b_);
1215 int st;
1216
1217 /*
1218 * Local packs tend to contain objects specific to our
1219 * variant of the project than remote ones. In addition,
1220 * remote ones could be on a network mounted filesystem.
1221 * Favor local ones for these reasons.
1222 */
1223 st = a->pack_local - b->pack_local;
1224 if (st)
1225 return -st;
1226
1227 /*
1228 * Younger packs tend to contain more recent objects,
1229 * and more recent objects tend to get accessed more
1230 * often.
1231 */
1232 if (a->mtime < b->mtime)
1233 return 1;
1234 else if (a->mtime == b->mtime)
1235 return 0;
1236 return -1;
1237}
1238
1239static void rearrange_packed_git(void)
1240{
1241 struct packed_git **ary, *p;
1242 int i, n;
1243
1244 for (n = 0, p = packed_git; p; p = p->next)
1245 n++;
1246 if (n < 2)
1247 return;
1248
1249 /* prepare an array of packed_git for easier sorting */
1250 ary = xcalloc(n, sizeof(struct packed_git *));
1251 for (n = 0, p = packed_git; p; p = p->next)
1252 ary[n++] = p;
1253
1254 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1255
1256 /* link them back again */
1257 for (i = 0; i < n - 1; i++)
1258 ary[i]->next = ary[i + 1];
1259 ary[n - 1]->next = NULL;
1260 packed_git = ary[0];
1261
1262 free(ary);
1263}
1264
637cdd9d 1265static int prepare_packed_git_run_once = 0;
9a217f2a 1266void prepare_packed_git(void)
1f688557 1267{
d5a63b99 1268 struct alternate_object_database *alt;
1f688557 1269
637cdd9d 1270 if (prepare_packed_git_run_once)
1f688557 1271 return;
9d835df2 1272 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1273 prepare_alt_odb();
d5a63b99 1274 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1275 alt->name[-1] = 0;
9d835df2 1276 prepare_packed_git_one(alt->base, 0);
1494e038 1277 alt->name[-1] = '/';
1f688557 1278 }
b867092f 1279 rearrange_packed_git();
637cdd9d
JK
1280 prepare_packed_git_run_once = 1;
1281}
1282
fc04c412 1283void reprepare_packed_git(void)
637cdd9d 1284{
4b480c67 1285 discard_revindex();
637cdd9d
JK
1286 prepare_packed_git_run_once = 0;
1287 prepare_packed_git();
1f688557
JH
1288}
1289
8eca0b47
NP
1290static void mark_bad_packed_object(struct packed_git *p,
1291 const unsigned char *sha1)
1292{
1293 unsigned i;
1294 for (i = 0; i < p->num_bad_objects; i++)
1295 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1296 return;
1297 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1298 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1299 p->num_bad_objects++;
1300}
1301
b6c4cecc 1302static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1303{
1304 struct packed_git *p;
1305 unsigned i;
1306
1307 for (p = packed_git; p; p = p->next)
1308 for (i = 0; i < p->num_bad_objects; i++)
1309 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1310 return p;
1311 return NULL;
ac939109
NP
1312}
1313
090ea126
NTND
1314/*
1315 * With an in-core object data in "map", rehash it to make sure the
1316 * object name actually matches "sha1" to detect object corruption.
1317 * With "map" == NULL, try reading the object named with "sha1" using
1318 * the streaming interface and rehash it to do the same.
1319 */
1320int check_sha1_signature(const unsigned char *sha1, void *map,
1321 unsigned long size, const char *type)
0fcfd160
LT
1322{
1323 unsigned char real_sha1[20];
090ea126
NTND
1324 enum object_type obj_type;
1325 struct git_istream *st;
1326 git_SHA_CTX c;
1327 char hdr[32];
1328 int hdrlen;
1329
1330 if (map) {
1331 hash_sha1_file(map, size, type, real_sha1);
1332 return hashcmp(sha1, real_sha1) ? -1 : 0;
1333 }
1334
1335 st = open_istream(sha1, &obj_type, &size, NULL);
1336 if (!st)
1337 return -1;
1338
1339 /* Generate the header */
1340 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1341
1342 /* Sha1.. */
1343 git_SHA1_Init(&c);
1344 git_SHA1_Update(&c, hdr, hdrlen);
1345 for (;;) {
1346 char buf[1024 * 16];
1347 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1348
f54fac53
JK
1349 if (readlen < 0) {
1350 close_istream(st);
1351 return -1;
1352 }
090ea126
NTND
1353 if (!readlen)
1354 break;
1355 git_SHA1_Update(&c, buf, readlen);
1356 }
1357 git_SHA1_Final(real_sha1, &c);
1358 close_istream(st);
a89fccd2 1359 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1360}
1361
f4e51683 1362static int git_open_noatime(const char *name)
44d1c19e
LT
1363{
1364 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1365
f2e872aa
SP
1366 for (;;) {
1367 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1368 if (fd >= 0)
f2e872aa
SP
1369 return fd;
1370
f2e872aa
SP
1371 /* Might the failure be due to O_NOATIME? */
1372 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1373 sha1_file_open_flag = 0;
f2e872aa
SP
1374 continue;
1375 }
1376
1377 return -1;
44d1c19e 1378 }
44d1c19e
LT
1379}
1380
052fe5ea
JK
1381static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1382{
1383 char *name = sha1_file_name(sha1);
1384 struct alternate_object_database *alt;
1385
1386 if (!lstat(name, st))
1387 return 0;
1388
1389 prepare_alt_odb();
1390 errno = ENOENT;
1391 for (alt = alt_odb_list; alt; alt = alt->next) {
1392 name = alt->name;
1393 fill_sha1_path(name, sha1);
1394 if (!lstat(alt->base, st))
1395 return 0;
1396 }
1397
1398 return -1;
1399}
1400
44d1c19e
LT
1401static int open_sha1_file(const unsigned char *sha1)
1402{
1403 int fd;
1404 char *name = sha1_file_name(sha1);
1405 struct alternate_object_database *alt;
1406
f4e51683 1407 fd = git_open_noatime(name);
44d1c19e
LT
1408 if (fd >= 0)
1409 return fd;
1410
1411 prepare_alt_odb();
1412 errno = ENOENT;
1413 for (alt = alt_odb_list; alt; alt = alt->next) {
1414 name = alt->name;
1415 fill_sha1_path(name, sha1);
f4e51683 1416 fd = git_open_noatime(alt->base);
44d1c19e
LT
1417 if (fd >= 0)
1418 return fd;
1419 }
1420 return -1;
1421}
1422
f0270efd 1423void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1424{
0fcfd160 1425 void *map;
144bde78 1426 int fd;
ace1534d 1427
44d1c19e
LT
1428 fd = open_sha1_file(sha1);
1429 map = NULL;
1430 if (fd >= 0) {
1431 struct stat st;
0fcfd160 1432
44d1c19e
LT
1433 if (!fstat(fd, &st)) {
1434 *size = xsize_t(st.st_size);
33e42de0
MM
1435 if (!*size) {
1436 /* mmap() is forbidden on empty files */
1437 error("object file %s is empty", sha1_file_name(sha1));
1438 return NULL;
1439 }
44d1c19e 1440 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1441 }
44d1c19e 1442 close(fd);
0fcfd160 1443 }
0fcfd160
LT
1444 return map;
1445}
1446
cc5c54e7
JH
1447/*
1448 * There used to be a second loose object header format which
1449 * was meant to mimic the in-pack format, allowing for direct
1450 * copy of the object data. This format turned up not to be
1451 * really worth it and we no longer write loose objects in that
1452 * format.
1453 */
1454static int experimental_loose_object(unsigned char *map)
bb6b8e4f
JH
1455{
1456 unsigned int word;
1457
1458 /*
7f684a2a
RT
1459 * We must determine if the buffer contains the standard
1460 * zlib-deflated stream or the experimental format based
1461 * on the in-pack object format. Compare the header byte
1462 * for each format:
1463 *
1464 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1465 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1466 *
1467 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1468 * in standard loose-object format, UNLESS it is a Git-pack
1469 * format object *exactly* 8 bytes in size when inflated.
1470 *
1471 * However, RFC1950 also specifies that the 1st 16-bit word
1472 * must be divisible by 31 - this checksum tells us our buffer
1473 * is in the standard format, giving a false positive only if
1474 * the 1st word of the Git-pack format object happens to be
1475 * divisible by 31, ie:
1476 * ((byte0 * 256) + byte1) % 31 = 0
1477 * => 0ttt10000www1000 % 31 = 0
1478 *
1479 * As it happens, this case can only arise for www=3 & ttt=1
1480 * - ie, a Commit object, which would have to be 8 bytes in
1481 * size. As no Commit can be that small, we find that the
1482 * combination of these two criteria (bitmask & checksum)
1483 * can always correctly determine the buffer format.
bb6b8e4f
JH
1484 */
1485 word = (map[0] << 8) + map[1];
7f684a2a 1486 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
bb6b8e4f 1487 return 0;
cc5c54e7
JH
1488 else
1489 return 1;
bb6b8e4f
JH
1490}
1491
09ded04b
NP
1492unsigned long unpack_object_header_buffer(const unsigned char *buf,
1493 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1494{
ad1ed5ee 1495 unsigned shift;
48fb7deb 1496 unsigned long size, c;
ad1ed5ee
JH
1497 unsigned long used = 0;
1498
1499 c = buf[used++];
1500 *type = (c >> 4) & 7;
1501 size = c & 15;
1502 shift = 4;
1503 while (c & 0x80) {
f630cfda 1504 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1505 error("bad object header");
ea4f9685
JH
1506 size = used = 0;
1507 break;
09ded04b 1508 }
ad1ed5ee
JH
1509 c = buf[used++];
1510 size += (c & 0x7f) << shift;
1511 shift += 7;
1512 }
1513 *sizep = size;
1514 return used;
1515}
1516
eb4f4076 1517int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee
JH
1518{
1519 unsigned long size, used;
1520 static const char valid_loose_object_type[8] = {
1521 0, /* OBJ_EXT */
1522 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1523 0, /* "delta" and others are invalid in a loose object */
93821bd9 1524 };
ad1ed5ee 1525 enum object_type type;
93821bd9 1526
c4483576
LT
1527 /* Get the data stream */
1528 memset(stream, 0, sizeof(*stream));
1529 stream->next_in = map;
1530 stream->avail_in = mapsize;
1531 stream->next_out = buffer;
93821bd9
LT
1532 stream->avail_out = bufsiz;
1533
cc5c54e7
JH
1534 if (experimental_loose_object(map)) {
1535 /*
1536 * The old experimental format we no longer produce;
1537 * we can still read it.
1538 */
1539 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1540 if (!used || !valid_loose_object_type[type])
1541 return -1;
1542 map += used;
1543 mapsize -= used;
726f852b 1544
cc5c54e7
JH
1545 /* Set up the stream for the rest.. */
1546 stream->next_in = map;
1547 stream->avail_in = mapsize;
1548 git_inflate_init(stream);
93821bd9 1549
cc5c54e7
JH
1550 /* And generate the fake traditional header */
1551 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1552 typename(type), size);
1553 return 0;
1554 }
39c68542 1555 git_inflate_init(stream);
cc5c54e7 1556 return git_inflate(stream, 0);
c4483576
LT
1557}
1558
ef49a7a0 1559static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1560{
1561 int bytes = strlen(buffer) + 1;
3aee68aa 1562 unsigned char *buf = xmallocz(size);
93821bd9 1563 unsigned long n;
7efbff75 1564 int status = Z_OK;
5180cacc 1565
93821bd9
LT
1566 n = stream->total_out - bytes;
1567 if (n > size)
1568 n = size;
1569 memcpy(buf, (char *) buffer + bytes, n);
1570 bytes = n;
456cdf6e
LT
1571 if (bytes <= size) {
1572 /*
1573 * The above condition must be (bytes <= size), not
1574 * (bytes < size). In other words, even though we
ccf5ace0 1575 * expect no more output and set avail_out to zero,
456cdf6e
LT
1576 * the input zlib stream may have bytes that express
1577 * "this concludes the stream", and we *do* want to
1578 * eat that input.
1579 *
1580 * Otherwise we would not be able to test that we
1581 * consumed all the input to reach the expected size;
1582 * we also want to check that zlib tells us that all
1583 * went well with status == Z_STREAM_END at the end.
1584 */
5180cacc
LT
1585 stream->next_out = buf + bytes;
1586 stream->avail_out = size - bytes;
7efbff75 1587 while (status == Z_OK)
39c68542 1588 status = git_inflate(stream, Z_FINISH);
5180cacc 1589 }
456cdf6e 1590 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1591 git_inflate_end(stream);
7efbff75
JH
1592 return buf;
1593 }
1594
1595 if (status < 0)
1596 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1597 else if (stream->avail_in)
1598 error("garbage at end of loose object '%s'",
1599 sha1_to_hex(sha1));
1600 free(buf);
1601 return NULL;
5180cacc
LT
1602}
1603
1604/*
1605 * We used to just use "sscanf()", but that's actually way
1606 * too permissive for what we want to check. So do an anal
1607 * object header parse by hand.
1608 */
f0270efd 1609int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1610{
21666f1a 1611 char type[10];
5180cacc
LT
1612 int i;
1613 unsigned long size;
1614
1615 /*
a6080a0a 1616 * The type can be at most ten bytes (including the
5180cacc 1617 * terminating '\0' that we add), and is followed by
21666f1a 1618 * a space.
5180cacc 1619 */
21666f1a 1620 i = 0;
5180cacc
LT
1621 for (;;) {
1622 char c = *hdr++;
1623 if (c == ' ')
1624 break;
21666f1a
NP
1625 type[i++] = c;
1626 if (i >= sizeof(type))
5180cacc 1627 return -1;
5180cacc 1628 }
21666f1a 1629 type[i] = 0;
5180cacc
LT
1630
1631 /*
1632 * The length must follow immediately, and be in canonical
1633 * decimal format (ie "010" is not valid).
1634 */
1635 size = *hdr++ - '0';
1636 if (size > 9)
1637 return -1;
1638 if (size) {
1639 for (;;) {
1640 unsigned long c = *hdr - '0';
1641 if (c > 9)
1642 break;
1643 hdr++;
1644 size = size * 10 + c;
1645 }
1646 }
1647 *sizep = size;
1648
1649 /*
1650 * The length must be followed by a zero byte
1651 */
21666f1a 1652 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1653}
1654
7efbff75 1655static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1656{
5180cacc 1657 int ret;
ef49a7a0 1658 git_zstream stream;
5180cacc 1659 char hdr[8192];
0fcfd160 1660
5180cacc 1661 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1662 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1663 return NULL;
1664
7efbff75 1665 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1666}
1667
54dab52a
NP
1668unsigned long get_size_from_delta(struct packed_git *p,
1669 struct pack_window **w_curs,
1670 off_t curpos)
1671{
1672 const unsigned char *data;
1673 unsigned char delta_head[20], *in;
ef49a7a0 1674 git_zstream stream;
54dab52a
NP
1675 int st;
1676
1677 memset(&stream, 0, sizeof(stream));
1678 stream.next_out = delta_head;
1679 stream.avail_out = sizeof(delta_head);
1680
39c68542 1681 git_inflate_init(&stream);
54dab52a
NP
1682 do {
1683 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1684 stream.next_in = in;
39c68542 1685 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1686 curpos += stream.next_in - in;
1687 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1688 stream.total_out < sizeof(delta_head));
39c68542 1689 git_inflate_end(&stream);
3d77d877
NP
1690 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1691 error("delta data unpack-initial failed");
1692 return 0;
1693 }
54dab52a
NP
1694
1695 /* Examine the initial part of the delta to figure out
1696 * the result size.
1697 */
1698 data = delta_head;
1699
1700 /* ignore base size */
1701 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1702
1703 /* Read the result size */
1704 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1705}
1706
c4001d92 1707static off_t get_delta_base(struct packed_git *p,
03e79c88 1708 struct pack_window **w_curs,
c4001d92 1709 off_t *curpos,
21666f1a 1710 enum object_type type,
c4001d92 1711 off_t delta_obj_offset)
eb32d236 1712{
2b87c45b 1713 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1714 off_t base_offset;
eb32d236 1715
8d8a4ea5
SP
1716 /* use_pack() assured us we have [base_info, base_info + 20)
1717 * as a range that we can look at without walking off the
1718 * end of the mapped window. Its actually the hash size
1719 * that is assured. An OFS_DELTA longer than the hash size
1720 * is stupid, as then a REF_DELTA would be smaller to store.
1721 */
21666f1a 1722 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1723 unsigned used = 0;
1724 unsigned char c = base_info[used++];
1725 base_offset = c & 127;
1726 while (c & 128) {
1727 base_offset += 1;
8723f216 1728 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1729 return 0; /* overflow */
eb32d236
NP
1730 c = base_info[used++];
1731 base_offset = (base_offset << 7) + (c & 127);
1732 }
1733 base_offset = delta_obj_offset - base_offset;
d8f32556 1734 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1735 return 0; /* out of bound */
2b87c45b 1736 *curpos += used;
21666f1a 1737 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1738 /* The base entry _must_ be in the same pack */
1739 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1740 *curpos += 20;
eb32d236
NP
1741 } else
1742 die("I am totally screwed");
2b87c45b 1743 return base_offset;
eb32d236
NP
1744}
1745
f8c8abc5
JH
1746int unpack_object_header(struct packed_git *p,
1747 struct pack_window **w_curs,
1748 off_t *curpos,
1749 unsigned long *sizep)
a733cb60 1750{
03e79c88 1751 unsigned char *base;
ef49a7a0 1752 unsigned long left;
ad1ed5ee 1753 unsigned long used;
2b87c45b 1754 enum object_type type;
a733cb60 1755
8d8a4ea5 1756 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1757 * as a range that we can look at. (Its actually the hash
3dff5379 1758 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1759 * the maximum deflated object size is 2^137, which is just
1760 * insane, so we know won't exceed what we have been given.
1761 */
2b87c45b 1762 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1763 used = unpack_object_header_buffer(base, left, &type, sizep);
1764 if (!used) {
1765 type = OBJ_BAD;
1766 } else
1767 *curpos += used;
ad1ed5ee 1768
2b87c45b 1769 return type;
a733cb60
LT
1770}
1771
790d96c0
TR
1772static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1773{
1774 int type;
1775 struct revindex_entry *revidx;
1776 const unsigned char *sha1;
1777 revidx = find_pack_revindex(p, obj_offset);
1778 if (!revidx)
1779 return OBJ_BAD;
1780 sha1 = nth_packed_object_sha1(p, revidx->nr);
1781 mark_bad_packed_object(p, sha1);
1782 type = sha1_object_info(sha1, NULL);
1783 if (type <= OBJ_NONE)
1784 return OBJ_BAD;
1785 return type;
1786}
1787
790d96c0
TR
1788#define POI_STACK_PREALLOC 64
1789
90191d37
JK
1790static enum object_type packed_to_object_type(struct packed_git *p,
1791 off_t obj_offset,
1792 enum object_type type,
1793 struct pack_window **w_curs,
1794 off_t curpos)
1f688557 1795{
790d96c0
TR
1796 off_t small_poi_stack[POI_STACK_PREALLOC];
1797 off_t *poi_stack = small_poi_stack;
1798 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1799
790d96c0
TR
1800 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1801 off_t base_offset;
90191d37 1802 unsigned long size;
790d96c0
TR
1803 /* Push the object we're going to leave behind */
1804 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1805 poi_stack_alloc = alloc_nr(poi_stack_nr);
1806 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1807 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1808 } else {
1809 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1810 }
1811 poi_stack[poi_stack_nr++] = obj_offset;
1812 /* If parsing the base offset fails, just unwind */
90191d37 1813 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1814 if (!base_offset)
1815 goto unwind;
1816 curpos = obj_offset = base_offset;
90191d37 1817 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1818 if (type <= OBJ_NONE) {
1819 /* If getting the base itself fails, we first
1820 * retry the base, otherwise unwind */
1821 type = retry_bad_packed_offset(p, base_offset);
1822 if (type > OBJ_NONE)
1823 goto out;
1824 goto unwind;
1825 }
1826 }
1827
21666f1a 1828 switch (type) {
790d96c0 1829 case OBJ_BAD:
a733cb60 1830 case OBJ_COMMIT:
a733cb60 1831 case OBJ_TREE:
a733cb60 1832 case OBJ_BLOB:
a733cb60 1833 case OBJ_TAG:
a69d0943 1834 break;
1f688557 1835 default:
3d77d877
NP
1836 error("unknown object type %i at offset %"PRIuMAX" in %s",
1837 type, (uintmax_t)obj_offset, p->pack_name);
1838 type = OBJ_BAD;
1f688557 1839 }
790d96c0
TR
1840
1841out:
1842 if (poi_stack != small_poi_stack)
1843 free(poi_stack);
21666f1a 1844 return type;
790d96c0
TR
1845
1846unwind:
1847 while (poi_stack_nr) {
1848 obj_offset = poi_stack[--poi_stack_nr];
1849 type = retry_bad_packed_offset(p, obj_offset);
1850 if (type > OBJ_NONE)
1851 goto out;
1852 }
1853 type = OBJ_BAD;
1854 goto out;
1f688557
JH
1855}
1856
90191d37 1857static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 1858 struct object_info *oi)
90191d37
JK
1859{
1860 struct pack_window *w_curs = NULL;
1861 unsigned long size;
1862 off_t curpos = obj_offset;
1863 enum object_type type;
1864
412916ee
JK
1865 /*
1866 * We always get the representation type, but only convert it to
1867 * a "real" type later if the caller is interested.
1868 */
90191d37
JK
1869 type = unpack_object_header(p, &w_curs, &curpos, &size);
1870
23c339c0 1871 if (oi->sizep) {
90191d37
JK
1872 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1873 off_t tmp_pos = curpos;
1874 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1875 type, obj_offset);
1876 if (!base_offset) {
1877 type = OBJ_BAD;
1878 goto out;
1879 }
23c339c0
JK
1880 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1881 if (*oi->sizep == 0) {
90191d37
JK
1882 type = OBJ_BAD;
1883 goto out;
1884 }
1885 } else {
23c339c0 1886 *oi->sizep = size;
90191d37
JK
1887 }
1888 }
1889
23c339c0 1890 if (oi->disk_sizep) {
90191d37 1891 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 1892 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
1893 }
1894
23c339c0
JK
1895 if (oi->typep) {
1896 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1897 if (*oi->typep < 0) {
412916ee
JK
1898 type = OBJ_BAD;
1899 goto out;
1900 }
1901 }
90191d37
JK
1902
1903out:
1904 unuse_pack(&w_curs);
1905 return type;
1906}
1907
eb950c19 1908static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1909 struct pack_window **w_curs,
c4001d92 1910 off_t curpos,
eb950c19 1911 unsigned long size)
de530aaa
SP
1912{
1913 int st;
ef49a7a0 1914 git_zstream stream;
079afb18 1915 unsigned char *buffer, *in;
de530aaa 1916
4ab07e4d 1917 buffer = xmallocz(size);
de530aaa 1918 memset(&stream, 0, sizeof(stream));
de530aaa 1919 stream.next_out = buffer;
39eea7bd 1920 stream.avail_out = size + 1;
de530aaa 1921
39c68542 1922 git_inflate_init(&stream);
079afb18 1923 do {
2b87c45b 1924 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1925 stream.next_in = in;
39c68542 1926 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1927 if (!stream.avail_out)
1928 break; /* the payload is larger than it should be */
2b87c45b 1929 curpos += stream.next_in - in;
079afb18 1930 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1931 git_inflate_end(&stream);
de530aaa
SP
1932 if ((st != Z_STREAM_END) || stream.total_out != size) {
1933 free(buffer);
1934 return NULL;
1935 }
1936
1937 return buffer;
1938}
1939
e5e01619
LT
1940#define MAX_DELTA_CACHE (256)
1941
18bdec11 1942static size_t delta_base_cached;
5e08ecbf
NP
1943
1944static struct delta_base_cache_lru_list {
1945 struct delta_base_cache_lru_list *prev;
1946 struct delta_base_cache_lru_list *next;
1947} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1948
e5e01619 1949static struct delta_base_cache_entry {
5e08ecbf
NP
1950 struct delta_base_cache_lru_list lru;
1951 void *data;
e5e01619
LT
1952 struct packed_git *p;
1953 off_t base_offset;
1954 unsigned long size;
e5e01619
LT
1955 enum object_type type;
1956} delta_base_cache[MAX_DELTA_CACHE];
1957
1958static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1959{
1960 unsigned long hash;
1961
1962 hash = (unsigned long)p + (unsigned long)base_offset;
1963 hash += (hash >> 8) + (hash >> 16);
3358004a 1964 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1965}
1966
84dd81c1
TR
1967static struct delta_base_cache_entry *
1968get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
1969{
1970 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
1971 return delta_base_cache + hash;
1972}
1973
1974static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1975 struct packed_git *p, off_t base_offset)
1976{
5266d369
JH
1977 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1978}
1979
84dd81c1
TR
1980static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1981{
1982 struct delta_base_cache_entry *ent;
1983 ent = get_delta_base_cache_entry(p, base_offset);
1984 return eq_delta_base_cache_entry(ent, p, base_offset);
1985}
1986
1987static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1988{
1989 ent->data = NULL;
1990 ent->lru.next->prev = ent->lru.prev;
1991 ent->lru.prev->next = ent->lru.next;
1992 delta_base_cached -= ent->size;
1993}
1994
62f255ad 1995static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1996 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1997{
84dd81c1 1998 struct delta_base_cache_entry *ent;
e5e01619 1999 void *ret;
e5e01619 2000
84dd81c1
TR
2001 ent = get_delta_base_cache_entry(p, base_offset);
2002
2003 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2004 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2005
84dd81c1
TR
2006 ret = ent->data;
2007
2008 if (!keep_cache)
2009 clear_delta_base_cache_entry(ent);
2010 else
182af834 2011 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
2012 *type = ent->type;
2013 *base_size = ent->size;
2014 return ret;
62f255ad
LT
2015}
2016
18bdec11
SP
2017static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2018{
2019 if (ent->data) {
2020 free(ent->data);
2021 ent->data = NULL;
5e08ecbf
NP
2022 ent->lru.next->prev = ent->lru.prev;
2023 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
2024 delta_base_cached -= ent->size;
2025 }
2026}
2027
3d20c636
SP
2028void clear_delta_base_cache(void)
2029{
2030 unsigned long p;
2031 for (p = 0; p < MAX_DELTA_CACHE; p++)
2032 release_delta_base_cache(&delta_base_cache[p]);
2033}
2034
62f255ad
LT
2035static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2036 void *base, unsigned long base_size, enum object_type type)
2037{
5e08ecbf 2038 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2039 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 2040 struct delta_base_cache_lru_list *lru;
e5e01619 2041
18bdec11
SP
2042 release_delta_base_cache(ent);
2043 delta_base_cached += base_size;
5e08ecbf
NP
2044
2045 for (lru = delta_base_cache_lru.next;
2046 delta_base_cached > delta_base_cache_limit
2047 && lru != &delta_base_cache_lru;
2048 lru = lru->next) {
2049 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
2050 if (f->type == OBJ_BLOB)
2051 release_delta_base_cache(f);
2052 }
5e08ecbf
NP
2053 for (lru = delta_base_cache_lru.next;
2054 delta_base_cached > delta_base_cache_limit
2055 && lru != &delta_base_cache_lru;
2056 lru = lru->next) {
2057 struct delta_base_cache_entry *f = (void *)lru;
2058 release_delta_base_cache(f);
2059 }
18bdec11 2060
e5e01619
LT
2061 ent->p = p;
2062 ent->base_offset = base_offset;
2063 ent->type = type;
2064 ent->data = base;
2065 ent->size = base_size;
5e08ecbf
NP
2066 ent->lru.next = &delta_base_cache_lru;
2067 ent->lru.prev = delta_base_cache_lru.prev;
2068 delta_base_cache_lru.prev->next = &ent->lru;
2069 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
2070}
2071
c2c5b270
CC
2072static void *read_object(const unsigned char *sha1, enum object_type *type,
2073 unsigned long *size);
2074
5f44324d
JH
2075static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2076{
2077 static FILE *log_file;
2078
b12ca963
NTND
2079 if (!log_pack_access)
2080 log_pack_access = getenv("GIT_TRACE_PACK_ACCESS");
2081 if (!log_pack_access)
2082 log_pack_access = no_log_pack_access;
2083 if (log_pack_access == no_log_pack_access)
2084 return;
2085
5f44324d
JH
2086 if (!log_file) {
2087 log_file = fopen(log_pack_access, "w");
2088 if (!log_file) {
2089 error("cannot open pack access log '%s' for writing: %s",
2090 log_pack_access, strerror(errno));
b12ca963 2091 log_pack_access = no_log_pack_access;
5f44324d
JH
2092 return;
2093 }
2094 }
2095 fprintf(log_file, "%s %"PRIuMAX"\n",
2096 p->pack_name, (uintmax_t)obj_offset);
2097 fflush(log_file);
2098}
2099
0e8189e2
NP
2100int do_check_packed_object_crc;
2101
abe601bb
TR
2102#define UNPACK_ENTRY_STACK_PREALLOC 64
2103struct unpack_entry_stack_ent {
2104 off_t obj_offset;
2105 off_t curpos;
2106 unsigned long size;
2107};
2108
c4001d92 2109void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2110 enum object_type *final_type, unsigned long *final_size)
1f688557 2111{
03e79c88 2112 struct pack_window *w_curs = NULL;
c4001d92 2113 off_t curpos = obj_offset;
abe601bb
TR
2114 void *data = NULL;
2115 unsigned long size;
2116 enum object_type type;
2117 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2118 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2119 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2120 int base_from_cache = 0;
1f688557 2121
b12ca963 2122 if (log_pack_access != no_log_pack_access)
5f44324d
JH
2123 write_pack_access_log(p, obj_offset);
2124
abe601bb
TR
2125 /* PHASE 1: drill down to the innermost base object */
2126 for (;;) {
2127 off_t base_offset;
2128 int i;
2129 struct delta_base_cache_entry *ent;
2130
77965f8b
NTND
2131 ent = get_delta_base_cache_entry(p, curpos);
2132 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2133 type = ent->type;
2134 data = ent->data;
2135 size = ent->size;
2136 clear_delta_base_cache_entry(ent);
2137 base_from_cache = 1;
2138 break;
2139 }
2140
abe601bb
TR
2141 if (do_check_packed_object_crc && p->index_version > 1) {
2142 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2143 unsigned long len = revidx[1].offset - obj_offset;
2144 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2145 const unsigned char *sha1 =
2146 nth_packed_object_sha1(p, revidx->nr);
2147 error("bad packed object CRC for %s",
2148 sha1_to_hex(sha1));
2149 mark_bad_packed_object(p, sha1);
2150 unuse_pack(&w_curs);
2151 return NULL;
2152 }
2153 }
2154
abe601bb
TR
2155 type = unpack_object_header(p, &w_curs, &curpos, &size);
2156 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2157 break;
2158
2159 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2160 if (!base_offset) {
2161 error("failed to validate delta base reference "
2162 "at offset %"PRIuMAX" from %s",
2163 (uintmax_t)curpos, p->pack_name);
2164 /* bail to phase 2, in hopes of recovery */
2165 data = NULL;
2166 break;
2167 }
2168
2169 /* push object, proceed to base */
2170 if (delta_stack_nr >= delta_stack_alloc
2171 && delta_stack == small_delta_stack) {
2172 delta_stack_alloc = alloc_nr(delta_stack_nr);
2173 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2174 memcpy(delta_stack, small_delta_stack,
2175 sizeof(*delta_stack)*delta_stack_nr);
2176 } else {
2177 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2178 }
2179 i = delta_stack_nr++;
2180 delta_stack[i].obj_offset = obj_offset;
2181 delta_stack[i].curpos = curpos;
2182 delta_stack[i].size = size;
2183
2184 curpos = obj_offset = base_offset;
0e8189e2
NP
2185 }
2186
abe601bb
TR
2187 /* PHASE 2: handle the base */
2188 switch (type) {
eb32d236
NP
2189 case OBJ_OFS_DELTA:
2190 case OBJ_REF_DELTA:
abe601bb
TR
2191 if (data)
2192 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2193 break;
a733cb60 2194 case OBJ_COMMIT:
a733cb60 2195 case OBJ_TREE:
a733cb60 2196 case OBJ_BLOB:
a733cb60 2197 case OBJ_TAG:
abe601bb
TR
2198 if (!base_from_cache)
2199 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2200 break;
1f688557 2201 default:
8eca0b47
NP
2202 data = NULL;
2203 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2204 type, (uintmax_t)obj_offset, p->pack_name);
2205 }
2206
2207 /* PHASE 3: apply deltas in order */
2208
2209 /* invariants:
2210 * 'data' holds the base data, or NULL if there was corruption
2211 */
2212 while (delta_stack_nr) {
2213 void *delta_data;
2214 void *base = data;
2215 unsigned long delta_size, base_size = size;
2216 int i;
2217
2218 data = NULL;
2219
2220 if (base)
2221 add_delta_base_cache(p, obj_offset, base, base_size, type);
2222
2223 if (!base) {
2224 /*
2225 * We're probably in deep shit, but let's try to fetch
2226 * the required base anyway from another pack or loose.
2227 * This is costly but should happen only in the presence
2228 * of a corrupted pack, and is better than failing outright.
2229 */
2230 struct revindex_entry *revidx;
2231 const unsigned char *base_sha1;
2232 revidx = find_pack_revindex(p, obj_offset);
2233 if (revidx) {
2234 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2235 error("failed to read delta base object %s"
2236 " at offset %"PRIuMAX" from %s",
2237 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2238 p->pack_name);
2239 mark_bad_packed_object(p, base_sha1);
2240 base = read_object(base_sha1, &type, &base_size);
2241 }
2242 }
2243
2244 i = --delta_stack_nr;
2245 obj_offset = delta_stack[i].obj_offset;
2246 curpos = delta_stack[i].curpos;
2247 delta_size = delta_stack[i].size;
2248
2249 if (!base)
2250 continue;
2251
2252 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2253
2254 if (!delta_data) {
2255 error("failed to unpack compressed delta "
2256 "at offset %"PRIuMAX" from %s",
2257 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2258 data = NULL;
2259 continue;
2260 }
2261
2262 data = patch_delta(base, base_size,
2263 delta_data, delta_size,
2264 &size);
1ee886c1
JK
2265
2266 /*
2267 * We could not apply the delta; warn the user, but keep going.
2268 * Our failure will be noticed either in the next iteration of
2269 * the loop, or if this is the final delta, in the caller when
2270 * we return NULL. Those code paths will take care of making
2271 * a more explicit warning and retrying with another copy of
2272 * the object.
2273 */
abe601bb 2274 if (!data)
1ee886c1 2275 error("failed to apply delta");
abe601bb 2276
4b8f772c 2277 free(delta_data);
1f688557 2278 }
abe601bb
TR
2279
2280 *final_type = type;
2281 *final_size = size;
2282
03e79c88 2283 unuse_pack(&w_curs);
21666f1a 2284 return data;
1f688557
JH
2285}
2286
d079837e 2287const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2288 uint32_t n)
9a217f2a 2289{
42873078 2290 const unsigned char *index = p->index_data;
d079837e
SP
2291 if (!index) {
2292 if (open_pack_index(p))
2293 return NULL;
2294 index = p->index_data;
2295 }
57059091 2296 if (n >= p->num_objects)
d72308e0 2297 return NULL;
74e34e1f
NP
2298 index += 4 * 256;
2299 if (p->index_version == 1) {
2300 return index + 24 * n + 4;
2301 } else {
2302 index += 8;
2303 return index + 20 * n;
2304 }
2305}
2306
99093238 2307off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2308{
2309 const unsigned char *index = p->index_data;
2310 index += 4 * 256;
2311 if (p->index_version == 1) {
2312 return ntohl(*((uint32_t *)(index + 24 * n)));
2313 } else {
2314 uint32_t off;
2315 index += 8 + p->num_objects * (20 + 4);
2316 off = ntohl(*((uint32_t *)(index + 4 * n)));
2317 if (!(off & 0x80000000))
2318 return off;
2319 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2320 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2321 ntohl(*((uint32_t *)(index + 4)));
2322 }
9a217f2a
JH
2323}
2324
c4001d92 2325off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2326 struct packed_git *p)
1f688557 2327{
42873078 2328 const uint32_t *level1_ofs = p->index_data;
42873078 2329 const unsigned char *index = p->index_data;
628522ec
JH
2330 unsigned hi, lo, stride;
2331 static int use_lookup = -1;
2332 static int debug_lookup = -1;
2333
2334 if (debug_lookup < 0)
2335 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2336
d079837e
SP
2337 if (!index) {
2338 if (open_pack_index(p))
2339 return 0;
2340 level1_ofs = p->index_data;
2341 index = p->index_data;
2342 }
74e34e1f
NP
2343 if (p->index_version > 1) {
2344 level1_ofs += 2;
2345 index += 8;
2346 }
42873078 2347 index += 4 * 256;
74e34e1f
NP
2348 hi = ntohl(level1_ofs[*sha1]);
2349 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2350 if (p->index_version > 1) {
2351 stride = 20;
2352 } else {
2353 stride = 24;
2354 index += 4;
2355 }
2356
2357 if (debug_lookup)
6e1c2344 2358 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2359 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2360
2361 if (use_lookup < 0)
2362 use_lookup = !!getenv("GIT_USE_LOOKUP");
2363 if (use_lookup) {
2364 int pos = sha1_entry_pos(index, stride, 0,
2365 lo, hi, p->num_objects, sha1);
2366 if (pos < 0)
2367 return 0;
2368 return nth_packed_object_offset(p, pos);
2369 }
1f688557
JH
2370
2371 do {
74e34e1f 2372 unsigned mi = (lo + hi) / 2;
628522ec
JH
2373 int cmp = hashcmp(index + mi * stride, sha1);
2374
2375 if (debug_lookup)
2376 printf("lo %u hi %u rg %u mi %u\n",
2377 lo, hi, hi - lo, mi);
43057304 2378 if (!cmp)
74e34e1f 2379 return nth_packed_object_offset(p, mi);
1f688557
JH
2380 if (cmp > 0)
2381 hi = mi;
2382 else
2383 lo = mi+1;
2384 } while (lo < hi);
2385 return 0;
2386}
2387
4c080182 2388int is_pack_valid(struct packed_git *p)
d131b7af
SP
2389{
2390 /* An already open pack is known to be valid. */
2391 if (p->pack_fd != -1)
2392 return 1;
2393
2394 /* If the pack has one window completely covering the
2395 * file size, the pack is known to be valid even if
2396 * the descriptor is not currently open.
2397 */
2398 if (p->windows) {
2399 struct pack_window *w = p->windows;
2400
2401 if (!w->offset && w->len == p->pack_size)
2402 return 1;
2403 }
2404
2405 /* Force the pack to open to prove its valid. */
2406 return !open_packed_git(p);
2407}
2408
95099731
NTND
2409static int fill_pack_entry(const unsigned char *sha1,
2410 struct pack_entry *e,
2411 struct packed_git *p)
2412{
2413 off_t offset;
2414
2415 if (p->num_bad_objects) {
2416 unsigned i;
2417 for (i = 0; i < p->num_bad_objects; i++)
2418 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2419 return 0;
2420 }
2421
2422 offset = find_pack_entry_one(sha1, p);
2423 if (!offset)
2424 return 0;
2425
2426 /*
2427 * We are about to tell the caller where they can locate the
2428 * requested object. We better make sure the packfile is
2429 * still here and can be accessed before supplying that
2430 * answer, as it may have been deleted since the index was
2431 * loaded!
2432 */
2433 if (!is_pack_valid(p)) {
2434 warning("packfile %s cannot be accessed", p->pack_name);
2435 return 0;
2436 }
2437 e->offset = offset;
2438 e->p = p;
2439 hashcpy(e->sha1, sha1);
2440 return 1;
2441}
2442
4d6acb70 2443static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2444{
2445 struct packed_git *p;
43057304 2446
1f688557 2447 prepare_packed_git();
f7c22cc6
NP
2448 if (!packed_git)
2449 return 0;
1f688557 2450
c01f51cc
NTND
2451 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2452 return 1;
8eca0b47 2453
c01f51cc
NTND
2454 for (p = packed_git; p; p = p->next) {
2455 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2456 continue;
f7c22cc6 2457
c01f51cc
NTND
2458 last_found_pack = p;
2459 return 1;
2460 }
1f688557
JH
2461 return 0;
2462}
2463
a6080a0a 2464struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2465 struct packed_git *packs)
2466{
2467 struct packed_git *p;
bf592c50
DB
2468
2469 for (p = packs; p; p = p->next) {
43057304 2470 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2471 return p;
2472 }
2473 return NULL;
21666f1a 2474
bf592c50
DB
2475}
2476
052fe5ea 2477static int sha1_loose_object_info(const unsigned char *sha1,
23c339c0 2478 struct object_info *oi)
65c2e0c3 2479{
36e4d74a 2480 int status;
65c2e0c3
JH
2481 unsigned long mapsize, size;
2482 void *map;
ef49a7a0 2483 git_zstream stream;
d65a16f6 2484 char hdr[32];
65c2e0c3 2485
052fe5ea
JK
2486 /*
2487 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2488 * need to look inside the object at all. Note that we
2489 * do not optimize out the stat call, even if the
2490 * caller doesn't care about the disk-size, since our
2491 * return value implicitly indicates whether the
2492 * object even exists.
052fe5ea 2493 */
23c339c0 2494 if (!oi->typep && !oi->sizep) {
4ef8d1dd
JH
2495 struct stat st;
2496 if (stat_sha1_file(sha1, &st) < 0)
2497 return -1;
2498 if (oi->disk_sizep)
23c339c0 2499 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2500 return 0;
2501 }
2502
bb6b8e4f 2503 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2504 if (!map)
dbea72a8 2505 return -1;
23c339c0
JK
2506 if (oi->disk_sizep)
2507 *oi->disk_sizep = mapsize;
36e4d74a
JH
2508 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2509 status = error("unable to unpack %s header",
2510 sha1_to_hex(sha1));
21666f1a 2511 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2512 status = error("unable to parse %s header", sha1_to_hex(sha1));
23c339c0
JK
2513 else if (oi->sizep)
2514 *oi->sizep = size;
39c68542 2515 git_inflate_end(&stream);
65c2e0c3 2516 munmap(map, mapsize);
23c339c0
JK
2517 if (oi->typep)
2518 *oi->typep = status;
052fe5ea 2519 return 0;
65c2e0c3
JH
2520}
2521
9a490590 2522int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
f0df4ed5 2523{
c4d9986f 2524 struct cached_object *co;
f0df4ed5 2525 struct pack_entry e;
5b086407 2526 int rtype;
f0df4ed5 2527
c4d9986f
NTND
2528 co = find_cached_object(sha1);
2529 if (co) {
5b086407
JK
2530 if (oi->typep)
2531 *(oi->typep) = co->type;
9a490590
JH
2532 if (oi->sizep)
2533 *(oi->sizep) = co->size;
161f00e7
JK
2534 if (oi->disk_sizep)
2535 *(oi->disk_sizep) = 0;
9a490590 2536 oi->whence = OI_CACHED;
5b086407 2537 return 0;
c4d9986f
NTND
2538 }
2539
cd673c1f 2540 if (!find_pack_entry(sha1, &e)) {
ddd63e64 2541 /* Most likely it's a loose object. */
23c339c0 2542 if (!sha1_loose_object_info(sha1, oi)) {
9a490590 2543 oi->whence = OI_LOOSE;
5b086407 2544 return 0;
9a490590 2545 }
ddd63e64
SG
2546
2547 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2548 reprepare_packed_git();
cd673c1f 2549 if (!find_pack_entry(sha1, &e))
052fe5ea 2550 return -1;
f0df4ed5 2551 }
3d77d877 2552
23c339c0 2553 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2554 if (rtype < 0) {
3d77d877 2555 mark_bad_packed_object(e.p, sha1);
412916ee 2556 return sha1_object_info_extended(sha1, oi);
5266d369
JH
2557 } else if (in_delta_base_cache(e.p, e.offset)) {
2558 oi->whence = OI_DBCACHED;
9a490590
JH
2559 } else {
2560 oi->whence = OI_PACKED;
2561 oi->u.packed.offset = e.offset;
2562 oi->u.packed.pack = e.p;
2563 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2564 rtype == OBJ_OFS_DELTA);
3d77d877
NP
2565 }
2566
5b086407 2567 return 0;
f0df4ed5
JS
2568}
2569
3fc0dca9 2570/* returns enum object_type or negative */
9a490590
JH
2571int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2572{
5b086407 2573 enum object_type type;
d099b717 2574 struct object_info oi = {NULL};
9a490590 2575
5b086407 2576 oi.typep = &type;
9a490590 2577 oi.sizep = sizep;
5b086407
JK
2578 if (sha1_object_info_extended(sha1, &oi) < 0)
2579 return -1;
2580 return type;
9a490590
JH
2581}
2582
21666f1a
NP
2583static void *read_packed_sha1(const unsigned char *sha1,
2584 enum object_type *type, unsigned long *size)
1f688557
JH
2585{
2586 struct pack_entry e;
8eca0b47 2587 void *data;
1f688557 2588
cd673c1f 2589 if (!find_pack_entry(sha1, &e))
1f688557 2590 return NULL;
8eca0b47
NP
2591 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2592 if (!data) {
2593 /*
2594 * We're probably in deep shit, but let's try to fetch
2595 * the required object anyway from another pack or loose.
2596 * This should happen only in the presence of a corrupted
2597 * pack, and is better than failing outright.
2598 */
2599 error("failed to read object %s at offset %"PRIuMAX" from %s",
2600 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2601 mark_bad_packed_object(e.p, sha1);
ac939109 2602 data = read_object(sha1, type, size);
8eca0b47
NP
2603 }
2604 return data;
1f688557
JH
2605}
2606
21666f1a
NP
2607int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2608 unsigned char *sha1)
d66b37bb
JH
2609{
2610 struct cached_object *co;
2611
21666f1a 2612 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2613 if (has_sha1_file(sha1) || find_cached_object(sha1))
2614 return 0;
2615 if (cached_object_alloc <= cached_object_nr) {
2616 cached_object_alloc = alloc_nr(cached_object_alloc);
2617 cached_objects = xrealloc(cached_objects,
2618 sizeof(*cached_objects) *
2619 cached_object_alloc);
2620 }
2621 co = &cached_objects[cached_object_nr++];
2622 co->size = len;
21666f1a 2623 co->type = type;
efa13f7b
JH
2624 co->buf = xmalloc(len);
2625 memcpy(co->buf, buf, len);
d66b37bb
JH
2626 hashcpy(co->sha1, sha1);
2627 return 0;
2628}
2629
c2c5b270
CC
2630static void *read_object(const unsigned char *sha1, enum object_type *type,
2631 unsigned long *size)
0fcfd160
LT
2632{
2633 unsigned long mapsize;
2634 void *map, *buf;
d66b37bb
JH
2635 struct cached_object *co;
2636
2637 co = find_cached_object(sha1);
2638 if (co) {
21666f1a 2639 *type = co->type;
d66b37bb 2640 *size = co->size;
182af834 2641 return xmemdupz(co->buf, co->size);
d66b37bb 2642 }
0fcfd160 2643
8276c007
PE
2644 buf = read_packed_sha1(sha1, type, size);
2645 if (buf)
2646 return buf;
bb6b8e4f 2647 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2648 if (map) {
7efbff75 2649 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2650 munmap(map, mapsize);
2651 return buf;
2652 }
637cdd9d 2653 reprepare_packed_git();
8276c007 2654 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2655}
2656
b6c4cecc
JH
2657/*
2658 * This function dies on corrupt objects; the callers who want to
2659 * deal with them should arrange to call read_object() and give error
2660 * messages themselves.
2661 */
5bf29b95
JH
2662void *read_sha1_file_extended(const unsigned char *sha1,
2663 enum object_type *type,
2664 unsigned long *size,
2665 unsigned flag)
ac939109 2666{
3ba7a065 2667 void *data;
e8b15e61 2668 char *path;
b6c4cecc 2669 const struct packed_git *p;
5bf29b95
JH
2670 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2671 ? lookup_replace_object(sha1) : sha1;
b6c4cecc 2672
3ba7a065
JH
2673 errno = 0;
2674 data = read_object(repl, type, size);
4bbf5a26 2675 if (data)
b6c4cecc 2676 return data;
68095570 2677
25f3af3f 2678 if (errno && errno != ENOENT)
3ba7a065
JH
2679 die_errno("failed to read object %s", sha1_to_hex(sha1));
2680
68095570 2681 /* die if we replaced an object with one that does not exist */
b6c4cecc 2682 if (repl != sha1)
68095570
CC
2683 die("replacement %s not found for %s",
2684 sha1_to_hex(repl), sha1_to_hex(sha1));
2685
b6c4cecc
JH
2686 if (has_loose_object(repl)) {
2687 path = sha1_file_name(sha1);
2688 die("loose object %s (stored in %s) is corrupt",
2689 sha1_to_hex(repl), path);
e8b15e61 2690 }
68095570 2691
b6c4cecc
JH
2692 if ((p = has_packed_and_bad(repl)) != NULL)
2693 die("packed object %s (stored in %s) is corrupt",
2694 sha1_to_hex(repl), p->pack_name);
f5552aee 2695
b6c4cecc 2696 return NULL;
ac939109
NP
2697}
2698
40469ee9 2699void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2700 const char *required_type_name,
40469ee9
JH
2701 unsigned long *size,
2702 unsigned char *actual_sha1_return)
f4913f91 2703{
21666f1a 2704 enum object_type type, required_type;
f4913f91
JH
2705 void *buffer;
2706 unsigned long isize;
40469ee9 2707 unsigned char actual_sha1[20];
f4913f91 2708
21666f1a 2709 required_type = type_from_string(required_type_name);
e702496e 2710 hashcpy(actual_sha1, sha1);
40469ee9
JH
2711 while (1) {
2712 int ref_length = -1;
2713 const char *ref_type = NULL;
f4913f91 2714
21666f1a 2715 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2716 if (!buffer)
2717 return NULL;
21666f1a 2718 if (type == required_type) {
40469ee9
JH
2719 *size = isize;
2720 if (actual_sha1_return)
e702496e 2721 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2722 return buffer;
2723 }
2724 /* Handle references */
21666f1a 2725 else if (type == OBJ_COMMIT)
40469ee9 2726 ref_type = "tree ";
21666f1a 2727 else if (type == OBJ_TAG)
40469ee9
JH
2728 ref_type = "object ";
2729 else {
2730 free(buffer);
2731 return NULL;
2732 }
2733 ref_length = strlen(ref_type);
f4913f91 2734
50974ec9
MK
2735 if (ref_length + 40 > isize ||
2736 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2737 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2738 free(buffer);
2739 return NULL;
2740 }
1cf58e72 2741 free(buffer);
40469ee9
JH
2742 /* Now we have the ID of the referred-to object in
2743 * actual_sha1. Check again. */
f4913f91 2744 }
f4913f91
JH
2745}
2746
ce9fbf16 2747static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2748 const char *type, unsigned char *sha1,
d65a16f6 2749 char *hdr, int *hdrlen)
d410c0f5 2750{
9126f009 2751 git_SHA_CTX c;
d410c0f5
JH
2752
2753 /* Generate the header */
d65a16f6 2754 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2755
2756 /* Sha1.. */
9126f009
NP
2757 git_SHA1_Init(&c);
2758 git_SHA1_Update(&c, hdr, *hdrlen);
2759 git_SHA1_Update(&c, buf, len);
2760 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2761}
2762
230f1322 2763/*
5a688fe4
JH
2764 * Move the just written object into its final resting place.
2765 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2766 * "moving" is only a part of what it does, when no patch between
2767 * master to pu changes the call sites of this function.
230f1322 2768 */
839837b9 2769int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2770{
e32c0a9c 2771 int ret = 0;
5a688fe4 2772
348df166 2773 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2774 goto try_rename;
2775 else if (link(tmpfile, filename))
e32c0a9c 2776 ret = errno;
7ebb6fca
LT
2777
2778 /*
2779 * Coda hack - coda doesn't like cross-directory links,
2780 * so we fall back to a rename, which will mean that it
2781 * won't be able to check collisions, but that's not a
2782 * big deal.
2783 *
2784 * The same holds for FAT formatted media.
2785 *
3be1f18e 2786 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2787 * left to unlink.
2788 */
2789 if (ret && ret != EEXIST) {
be66a6c4 2790 try_rename:
7ebb6fca 2791 if (!rename(tmpfile, filename))
3be1f18e 2792 goto out;
9e48b389 2793 ret = errno;
230f1322 2794 }
691f1a28 2795 unlink_or_warn(tmpfile);
230f1322
LT
2796 if (ret) {
2797 if (ret != EEXIST) {
82247e9b 2798 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2799 }
2800 /* FIXME!!! Collision check here ? */
2801 }
2802
3be1f18e 2803out:
5256b006 2804 if (adjust_shared_perm(filename))
5a688fe4 2805 return error("unable to set permission to '%s'", filename);
230f1322
LT
2806 return 0;
2807}
2808
4d548150
LT
2809static int write_buffer(int fd, const void *buf, size_t len)
2810{
d34cf19b 2811 if (write_in_full(fd, buf, len) < 0)
93822c22 2812 return error("file write error (%s)", strerror(errno));
4d548150
LT
2813 return 0;
2814}
2815
ce9fbf16 2816int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2817 unsigned char *sha1)
2818{
d65a16f6 2819 char hdr[32];
abdc3fc8
RS
2820 int hdrlen;
2821 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2822 return 0;
2823}
2824
e9039dd3
LT
2825/* Finalize a file on disk, and close it. */
2826static void close_sha1_file(int fd)
2827{
aafe9fba
LT
2828 if (fsync_object_files)
2829 fsync_or_die(fd, "sha1 file");
e9039dd3 2830 if (close(fd) != 0)
d824cbba 2831 die_errno("error when closing sha1 file");
e9039dd3
LT
2832}
2833
5723fe7e
LT
2834/* Size of directory component, including the ending '/' */
2835static inline int directory_size(const char *filename)
2836{
2837 const char *s = strrchr(filename, '/');
2838 if (!s)
2839 return 0;
2840 return s - filename + 1;
2841}
2842
2843/*
2844 * This creates a temporary file in the same directory as the final
2845 * 'filename'
2846 *
2847 * We want to avoid cross-directory filename renames, because those
2848 * can have problems on various filesystems (FAT, NFS, Coda).
2849 */
2850static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2851{
2852 int fd, dirlen = directory_size(filename);
2853
2854 if (dirlen + 20 > bufsiz) {
2855 errno = ENAMETOOLONG;
2856 return -1;
2857 }
2858 memcpy(buffer, filename, dirlen);
2859 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2860 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2861 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2862 /* Make sure the directory exists */
6ff6af62 2863 memcpy(buffer, filename, dirlen);
5723fe7e 2864 buffer[dirlen-1] = 0;
b2476a60
JH
2865 if (mkdir(buffer, 0777) && errno != EEXIST)
2866 return -1;
2867 if (adjust_shared_perm(buffer))
5723fe7e
LT
2868 return -1;
2869
2870 /* Try again */
2871 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2872 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2873 }
2874 return fd;
2875}
2876
bbac7311 2877static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2878 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2879{
915308b1 2880 int fd, ret;
9892beba 2881 unsigned char compressed[4096];
ef49a7a0 2882 git_zstream stream;
748af44c
NP
2883 git_SHA_CTX c;
2884 unsigned char parano_sha1[20];
706bc531 2885 char *filename;
ab1900a3 2886 static char tmp_file[PATH_MAX];
a44c9a5e 2887
972a9155 2888 filename = sha1_file_name(sha1);
ab1900a3 2889 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 2890 if (fd < 0) {
35243577 2891 if (errno == EACCES)
82247e9b 2892 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 2893 else
5eaeda70 2894 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
2895 }
2896
0fcfd160
LT
2897 /* Set it up */
2898 memset(&stream, 0, sizeof(stream));
55bb5c91 2899 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 2900 stream.next_out = compressed;
9892beba 2901 stream.avail_out = sizeof(compressed);
748af44c 2902 git_SHA1_Init(&c);
a44c9a5e
LT
2903
2904 /* First header.. */
d65a16f6 2905 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2906 stream.avail_in = hdrlen;
55bb5c91
JH
2907 while (git_deflate(&stream, 0) == Z_OK)
2908 ; /* nothing */
748af44c 2909 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2910
2911 /* Then the data itself.. */
c00e657d 2912 stream.next_in = (void *)buf;
a44c9a5e 2913 stream.avail_in = len;
9892beba 2914 do {
748af44c 2915 unsigned char *in0 = stream.next_in;
55bb5c91 2916 ret = git_deflate(&stream, Z_FINISH);
748af44c 2917 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2918 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2919 die("unable to write sha1 file");
2920 stream.next_out = compressed;
2921 stream.avail_out = sizeof(compressed);
2922 } while (ret == Z_OK);
2923
ac54c277
LT
2924 if (ret != Z_STREAM_END)
2925 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 2926 ret = git_deflate_end_gently(&stream);
ac54c277
LT
2927 if (ret != Z_OK)
2928 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2929 git_SHA1_Final(parano_sha1, &c);
2930 if (hashcmp(sha1, parano_sha1) != 0)
2931 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2932
e9039dd3 2933 close_sha1_file(fd);
0fcfd160 2934
bbac7311
NP
2935 if (mtime) {
2936 struct utimbuf utb;
2937 utb.actime = mtime;
2938 utb.modtime = mtime;
ab1900a3 2939 if (utime(tmp_file, &utb) < 0)
bbac7311 2940 warning("failed utime() on %s: %s",
ab1900a3 2941 tmp_file, strerror(errno));
bbac7311
NP
2942 }
2943
ab1900a3 2944 return move_temp_to_file(tmp_file, filename);
0fcfd160 2945}
8237b185 2946
c00e657d 2947int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2948{
2949 unsigned char sha1[20];
2950 char hdr[32];
2951 int hdrlen;
2952
2953 /* Normally if we have it in the pack then we do not bother writing
2954 * it out into .git/objects/??/?{38} file.
2955 */
2956 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2957 if (returnsha1)
2958 hashcpy(returnsha1, sha1);
2959 if (has_sha1_file(sha1))
2960 return 0;
2961 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2962}
2963
2964int force_object_loose(const unsigned char *sha1, time_t mtime)
2965{
bbac7311
NP
2966 void *buf;
2967 unsigned long len;
2968 enum object_type type;
2969 char hdr[32];
2970 int hdrlen;
1fb23e65 2971 int ret;
bbac7311 2972
c529d75a 2973 if (has_loose_object(sha1))
bbac7311
NP
2974 return 0;
2975 buf = read_packed_sha1(sha1, &type, &len);
2976 if (!buf)
2977 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2978 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2979 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2980 free(buf);
2981
2982 return ret;
bbac7311
NP
2983}
2984
bf592c50
DB
2985int has_pack_index(const unsigned char *sha1)
2986{
2987 struct stat st;
2988 if (stat(sha1_pack_index_name(sha1), &st))
2989 return 0;
2990 return 1;
bf592c50
DB
2991}
2992
cd673c1f
JH
2993int has_sha1_pack(const unsigned char *sha1)
2994{
2995 struct pack_entry e;
2996 return find_pack_entry(sha1, &e);
2997}
2998
8237b185
DB
2999int has_sha1_file(const unsigned char *sha1)
3000{
1f688557
JH
3001 struct pack_entry e;
3002
cd673c1f 3003 if (find_pack_entry(sha1, &e))
1f688557 3004 return 1;
45e8a748
JK
3005 if (has_loose_object(sha1))
3006 return 1;
3007 reprepare_packed_git();
3008 return find_pack_entry(sha1, &e);
8237b185 3009}
74400e71 3010
c879daa2
NTND
3011static void check_tree(const void *buf, size_t size)
3012{
3013 struct tree_desc desc;
3014 struct name_entry entry;
3015
3016 init_tree_desc(&desc, buf, size);
3017 while (tree_entry(&desc, &entry))
3018 /* do nothing
3019 * tree_entry() will die() on malformed entries */
3020 ;
3021}
3022
3023static void check_commit(const void *buf, size_t size)
3024{
3025 struct commit c;
3026 memset(&c, 0, sizeof(c));
3027 if (parse_commit_buffer(&c, buf, size))
3028 die("corrupt commit");
3029}
3030
3031static void check_tag(const void *buf, size_t size)
3032{
3033 struct tag t;
3034 memset(&t, 0, sizeof(t));
3035 if (parse_tag_buffer(&t, buf, size))
3036 die("corrupt tag");
3037}
3038
43df4f86 3039static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3040 enum object_type type,
3041 const char *path, unsigned flags)
e7332f96 3042{
6c510bee 3043 int ret, re_allocated = 0;
c4ce46fc 3044 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3045
7672db20 3046 if (!type)
edaec3fb 3047 type = OBJ_BLOB;
6c510bee
LT
3048
3049 /*
3050 * Convert blobs to git internal format
3051 */
43df4f86 3052 if ((type == OBJ_BLOB) && path) {
f285a2d7 3053 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3054 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3055 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3056 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3057 re_allocated = 1;
3058 }
3059 }
c4ce46fc 3060 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3061 if (type == OBJ_TREE)
3062 check_tree(buf, size);
3063 if (type == OBJ_COMMIT)
3064 check_commit(buf, size);
3065 if (type == OBJ_TAG)
3066 check_tag(buf, size);
3067 }
6c510bee 3068
7672db20 3069 if (write_object)
edaec3fb 3070 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3071 else
edaec3fb 3072 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3073 if (re_allocated)
6c510bee 3074 free(buf);
43df4f86
DP
3075 return ret;
3076}
3077
7b41e1e1
JH
3078static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3079 const char *path, unsigned flags)
3080{
3081 struct strbuf sbuf = STRBUF_INIT;
3082 int ret;
3083
3084 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3085 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3086 else
3087 ret = -1;
3088 strbuf_release(&sbuf);
3089 return ret;
3090}
3091
ea68b0ce
DP
3092#define SMALL_FILE_SIZE (32*1024)
3093
7b41e1e1
JH
3094static int index_core(unsigned char *sha1, int fd, size_t size,
3095 enum object_type type, const char *path,
3096 unsigned flags)
43df4f86
DP
3097{
3098 int ret;
43df4f86 3099
7b41e1e1 3100 if (!size) {
c4ce46fc 3101 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
3102 } else if (size <= SMALL_FILE_SIZE) {
3103 char *buf = xmalloc(size);
3104 if (size == read_in_full(fd, buf, size))
c4ce46fc 3105 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
3106 else
3107 ret = error("short read %s", strerror(errno));
3108 free(buf);
08bda208 3109 } else {
43df4f86 3110 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3111 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3112 munmap(buf, size);
08bda208 3113 }
7b41e1e1
JH
3114 return ret;
3115}
3116
4dd1fbc7 3117/*
568508e7
JH
3118 * This creates one packfile per large blob unless bulk-checkin
3119 * machinery is "plugged".
4dd1fbc7
JH
3120 *
3121 * This also bypasses the usual "convert-to-git" dance, and that is on
3122 * purpose. We could write a streaming version of the converting
3123 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3124 * (or equivalent in-core API described above). However, that is
3125 * somewhat complicated, as we do not know the size of the filter
3126 * result, which we need to know beforehand when writing a git object.
3127 * Since the primary motivation for trying to stream from the working
3128 * tree file and to avoid mmaping it in core is to deal with large
3129 * binary blobs, they generally do not want to get any conversion, and
3130 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3131 */
3132static int index_stream(unsigned char *sha1, int fd, size_t size,
3133 enum object_type type, const char *path,
3134 unsigned flags)
3135{
568508e7 3136 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3137}
3138
7b41e1e1
JH
3139int index_fd(unsigned char *sha1, int fd, struct stat *st,
3140 enum object_type type, const char *path, unsigned flags)
3141{
3142 int ret;
3143 size_t size = xsize_t(st->st_size);
3144
3145 if (!S_ISREG(st->st_mode))
3146 ret = index_pipe(sha1, fd, type, path, flags);
4f22b101
JK
3147 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3148 (path && would_convert_to_git(path, NULL, 0, 0)))
7b41e1e1 3149 ret = index_core(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3150 else
3151 ret = index_stream(sha1, fd, size, type, path, flags);
43df4f86 3152 close(fd);
aac17941 3153 return ret;
74400e71 3154}
ec1fcc16 3155
c4ce46fc 3156int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3157{
3158 int fd;
b760d3aa 3159 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3160
3161 switch (st->st_mode & S_IFMT) {
3162 case S_IFREG:
3163 fd = open(path, O_RDONLY);
3164 if (fd < 0)
3165 return error("open(\"%s\"): %s", path,
3166 strerror(errno));
c4ce46fc 3167 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3168 return error("%s: failed to insert into database",
3169 path);
3170 break;
3171 case S_IFLNK:
b760d3aa 3172 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 3173 char *errstr = strerror(errno);
ec1fcc16
JH
3174 return error("readlink(\"%s\"): %s", path,
3175 errstr);
3176 }
c4ce46fc 3177 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3178 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3179 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3180 return error("%s: failed to insert into database",
3181 path);
b760d3aa 3182 strbuf_release(&sb);
ec1fcc16 3183 break;
f35a6d3b
LT
3184 case S_IFDIR:
3185 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3186 default:
3187 return error("%s: unsupported file type", path);
3188 }
3189 return 0;
3190}
a69e5429
JH
3191
3192int read_pack_header(int fd, struct pack_header *header)
3193{
c697ad14
HO
3194 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3195 /* "eof before pack header was fully read" */
3196 return PH_ERROR_EOF;
3197
a69e5429
JH
3198 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3199 /* "protocol error (pack signature mismatch detected)" */
3200 return PH_ERROR_PACK_SIGNATURE;
3201 if (!pack_version_ok(header->hdr_version))
3202 /* "protocol error (pack version unsupported)" */
3203 return PH_ERROR_PROTOCOL;
3204 return 0;
3205}
40d52ff7
JK
3206
3207void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3208{
3209 enum object_type type = sha1_object_info(sha1, NULL);
3210 if (type < 0)
3211 die("%s is not a valid object", sha1_to_hex(sha1));
3212 if (type != expect)
3213 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3214 typename(expect));
3215}