verify-pack: use strbuf_strip_suffix
[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
1b1005d1
MH
63/*
64 * A pointer to the last packed_git in which an object was found.
65 * When an object is sought, we look in this packfile first, because
66 * objects that are looked up at similar times are often in the same
67 * packfile as one another.
68 */
c01f51cc
NTND
69static struct packed_git *last_found_pack;
70
c597ba80
NTND
71static struct cached_object *find_cached_object(const unsigned char *sha1)
72{
73 int i;
74 struct cached_object *co = cached_objects;
75
76 for (i = 0; i < cached_object_nr; i++, co++) {
77 if (!hashcmp(co->sha1, sha1))
78 return co;
79 }
80 if (!hashcmp(sha1, empty_tree.sha1))
81 return &empty_tree;
82 return NULL;
83}
84
90a6464b
JH
85int mkdir_in_gitdir(const char *path)
86{
87 if (mkdir(path, 0777)) {
88 int saved_errno = errno;
89 struct stat st;
90 struct strbuf sb = STRBUF_INIT;
91
92 if (errno != EEXIST)
93 return -1;
94 /*
95 * Are we looking at a path in a symlinked worktree
96 * whose original repository does not yet have it?
97 * e.g. .git/rr-cache pointing at its original
98 * repository in which the user hasn't performed any
99 * conflict resolution yet?
100 */
101 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
102 strbuf_readlink(&sb, path, st.st_size) ||
103 !is_absolute_path(sb.buf) ||
104 mkdir(sb.buf, 0777)) {
105 strbuf_release(&sb);
106 errno = saved_errno;
107 return -1;
108 }
109 strbuf_release(&sb);
110 }
111 return adjust_shared_perm(path);
112}
113
0be0521b 114enum scld_error safe_create_leading_directories(char *path)
b2cb9425 115{
26c8ae2a 116 char *next_component = path + offset_1st_component(path);
0be0521b 117 enum scld_error ret = SCLD_OK;
67d42212 118
0be0521b 119 while (ret == SCLD_OK && next_component) {
f0502332 120 struct stat st;
0f527403 121 char *slash = next_component, slash_character;
f0502332 122
0f527403
MH
123 while (*slash && !is_dir_sep(*slash))
124 slash++;
125
126 if (!*slash)
b2cb9425 127 break;
bf10cf70 128
26c8ae2a 129 next_component = slash + 1;
0f527403 130 while (is_dir_sep(*next_component))
bf10cf70 131 next_component++;
26c8ae2a 132 if (!*next_component)
5f0bdf50 133 break;
831651fd 134
0f527403 135 slash_character = *slash;
831651fd 136 *slash = '\0';
67d42212
JR
137 if (!stat(path, &st)) {
138 /* path exists */
9e6f885d 139 if (!S_ISDIR(st.st_mode))
0be0521b 140 ret = SCLD_EXISTS;
53a39721 141 } else if (mkdir(path, 0777)) {
928734d9 142 if (errno == EEXIST &&
9e6f885d 143 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 144 ; /* somebody created it since we checked */
18d37e86
MH
145 else if (errno == ENOENT)
146 /*
147 * Either mkdir() failed because
148 * somebody just pruned the containing
149 * directory, or stat() failed because
150 * the file that was in our way was
151 * just removed. Either way, inform
152 * the caller that it might be worth
153 * trying again:
154 */
155 ret = SCLD_VANISHED;
9e6f885d 156 else
0be0521b 157 ret = SCLD_FAILED;
53a39721 158 } else if (adjust_shared_perm(path)) {
0be0521b 159 ret = SCLD_PERMS;
457f06d6 160 }
0f527403 161 *slash = slash_character;
b2cb9425 162 }
9e6f885d 163 return ret;
b2cb9425 164}
723c31fe 165
0be0521b 166enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b
JK
167{
168 /* path points to cache entries, so xstrdup before messing with it */
169 char *buf = xstrdup(path);
0be0521b 170 enum scld_error result = safe_create_leading_directories(buf);
8e21d63b
JK
171 free(buf);
172 return result;
173}
174
ace1534d
JH
175static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
176{
177 int i;
178 for (i = 0; i < 20; i++) {
179 static char hex[] = "0123456789abcdef";
180 unsigned int val = sha1[i];
181 char *pos = pathbuf + i*2 + (i > 0);
182 *pos++ = hex[val >> 4];
183 *pos = hex[val & 0xf];
184 }
185}
186
30d6c6ea 187const char *sha1_file_name(const unsigned char *sha1)
0fcfd160 188{
560fb6a1
JK
189 static char buf[PATH_MAX];
190 const char *objdir;
191 int len;
0fcfd160 192
560fb6a1
JK
193 objdir = get_object_directory();
194 len = strlen(objdir);
195
196 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
197 if (len + 43 > PATH_MAX)
198 die("insanely long object directory %s", objdir);
199 memcpy(buf, objdir, len);
200 buf[len] = '/';
201 buf[len+3] = '/';
202 buf[len+42] = '\0';
203 fill_sha1_path(buf + len + 1, sha1);
204 return buf;
0fcfd160
LT
205}
206
d40d535b
MH
207/*
208 * Return the name of the pack or index file with the specified sha1
209 * in its filename. *base and *name are scratch space that must be
210 * provided by the caller. which should be "pack" or "idx".
211 */
633f43e1 212static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 213 char **name, char **base, const char *which)
bf592c50
DB
214{
215 static const char hex[] = "0123456789abcdef";
633f43e1 216 char *buf;
bf592c50
DB
217 int i;
218
633f43e1 219 if (!*base) {
bf592c50
DB
220 const char *sha1_file_directory = get_object_directory();
221 int len = strlen(sha1_file_directory);
633f43e1 222 *base = xmalloc(len + 60);
6eec46bd
JH
223 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
224 sha1_file_directory, which);
633f43e1 225 *name = *base + len + 11;
bf592c50
DB
226 }
227
633f43e1 228 buf = *name;
bf592c50
DB
229
230 for (i = 0; i < 20; i++) {
231 unsigned int val = *sha1++;
232 *buf++ = hex[val >> 4];
233 *buf++ = hex[val & 0xf];
234 }
a6080a0a 235
633f43e1 236 return *base;
bf592c50
DB
237}
238
633f43e1 239char *sha1_pack_name(const unsigned char *sha1)
bf592c50 240{
633f43e1 241 static char *name, *base;
bf592c50 242
6eec46bd 243 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 244}
bf592c50 245
633f43e1
HO
246char *sha1_pack_index_name(const unsigned char *sha1)
247{
248 static char *name, *base;
a6080a0a 249
6eec46bd 250 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
251}
252
d5a63b99
JH
253struct alternate_object_database *alt_odb_list;
254static struct alternate_object_database **alt_odb_tail;
ace1534d 255
ddd5d056
JH
256/*
257 * Prepare alternate object database registry.
d5a63b99
JH
258 *
259 * The variable alt_odb_list points at the list of struct
260 * alternate_object_database. The elements on this list come from
261 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
262 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
263 * whose contents is similar to that environment variable but can be
264 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
265 * contains "/the/directory/corresponding/to/.git/objects/...", while
266 * its name points just after the slash at the end of ".git/objects/"
267 * in the example above, and has enough space to hold 40-byte hex
268 * SHA1, an extra slash for the first level indirection, and the
269 * terminating NUL.
ddd5d056 270 */
6eac50d8 271static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
ace1534d 272{
1494e038 273 const char *objdir = get_object_directory();
c2f493a4
MW
274 struct alternate_object_database *ent;
275 struct alternate_object_database *alt;
5bdf0a84
HW
276 int pfxlen, entlen;
277 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 278
85dadc38 279 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
280 strbuf_addstr(&pathbuf, real_path(relative_base));
281 strbuf_addch(&pathbuf, '/');
c2f493a4 282 }
6eac50d8 283 strbuf_addstr(&pathbuf, entry);
c2f493a4 284
5bdf0a84
HW
285 normalize_path_copy(pathbuf.buf, pathbuf.buf);
286
287 pfxlen = strlen(pathbuf.buf);
288
289 /*
290 * The trailing slash after the directory name is given by
291 * this function at the end. Remove duplicates.
292 */
293 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
294 pfxlen -= 1;
295
296 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
297 ent = xmalloc(sizeof(*ent) + entlen);
298 memcpy(ent->base, pathbuf.buf, pfxlen);
299 strbuf_release(&pathbuf);
c2f493a4
MW
300
301 ent->name = ent->base + pfxlen + 1;
302 ent->base[pfxlen + 3] = '/';
303 ent->base[pfxlen] = ent->base[entlen-1] = 0;
304
305 /* Detect cases where alternate disappeared */
90b4a71c 306 if (!is_directory(ent->base)) {
c2f493a4
MW
307 error("object directory %s does not exist; "
308 "check .git/objects/info/alternates.",
309 ent->base);
310 free(ent);
311 return -1;
312 }
313
314 /* Prevent the common mistake of listing the same
315 * thing twice, or object directory itself.
316 */
317 for (alt = alt_odb_list; alt; alt = alt->next) {
318 if (!memcmp(ent->base, alt->base, pfxlen)) {
319 free(ent);
320 return -1;
321 }
322 }
cb2912c3 323 if (!strcmp(ent->base, objdir)) {
c2f493a4
MW
324 free(ent);
325 return -1;
326 }
327
328 /* add the alternate entry */
329 *alt_odb_tail = ent;
330 alt_odb_tail = &(ent->next);
331 ent->next = NULL;
332
333 /* recursively add alternates */
334 read_info_alternates(ent->base, depth + 1);
335
336 ent->base[pfxlen] = '/';
337
338 return 0;
339}
340
c5950164 341static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
342 const char *relative_base, int depth)
343{
6eac50d8
MH
344 struct string_list entries = STRING_LIST_INIT_NODUP;
345 char *alt_copy;
346 int i;
c2f493a4
MW
347
348 if (depth > 5) {
349 error("%s: ignoring alternate object stores, nesting too deep.",
350 relative_base);
351 return;
352 }
353
c5950164 354 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
355 string_list_split_in_place(&entries, alt_copy, sep, -1);
356 for (i = 0; i < entries.nr; i++) {
357 const char *entry = entries.items[i].string;
358 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 359 continue;
6eac50d8
MH
360 if (!is_absolute_path(entry) && depth) {
361 error("%s: ignoring relative alternate object store %s",
362 relative_base, entry);
363 } else {
364 link_alt_odb_entry(entry, relative_base, depth);
9577e7e3 365 }
9577e7e3 366 }
6eac50d8
MH
367 string_list_clear(&entries, 0);
368 free(alt_copy);
d5a63b99
JH
369}
370
5e73633d 371void read_info_alternates(const char * relative_base, int depth)
d5a63b99 372{
9577e7e3 373 char *map;
dc49cd76 374 size_t mapsz;
d5a63b99 375 struct stat st;
9cb18f56
JM
376 const char alt_file_name[] = "info/alternates";
377 /* Given that relative_base is no longer than PATH_MAX,
378 ensure that "path" has enough space to append "/", the
379 file name, "info/alternates", and a trailing NUL. */
380 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 381 int fd;
d5a63b99 382
9cb18f56 383 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 384 fd = git_open_noatime(path);
d5a63b99
JH
385 if (fd < 0)
386 return;
387 if (fstat(fd, &st) || (st.st_size == 0)) {
388 close(fd);
9a217f2a 389 return;
ace1534d 390 }
dc49cd76
SP
391 mapsz = xsize_t(st.st_size);
392 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 393 close(fd);
d5a63b99 394
c5950164 395 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 396
dc49cd76 397 munmap(map, mapsz);
ace1534d
JH
398}
399
bef70b22
DB
400void add_to_alternates_file(const char *reference)
401{
402 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 403 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
e6baf4a1 404 char *alt = mkpath("%s\n", reference);
bef70b22
DB
405 write_or_die(fd, alt, strlen(alt));
406 if (commit_lock_file(lock))
407 die("could not close alternates file");
408 if (alt_odb_tail)
c5950164 409 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
bef70b22
DB
410}
411
d79796bc
JH
412void foreach_alt_odb(alt_odb_fn fn, void *cb)
413{
414 struct alternate_object_database *ent;
415
416 prepare_alt_odb();
417 for (ent = alt_odb_list; ent; ent = ent->next)
418 if (fn(ent, cb))
419 return;
420}
421
c2f493a4
MW
422void prepare_alt_odb(void)
423{
554fe20d 424 const char *alt;
c2f493a4 425
7dc24aa5
SP
426 if (alt_odb_tail)
427 return;
428
c2f493a4
MW
429 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
430 if (!alt) alt = "";
431
c2f493a4 432 alt_odb_tail = &alt_odb_list;
c5950164 433 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
434
435 read_info_alternates(get_object_directory(), 0);
436}
437
0f4dc14a 438static int has_loose_object_local(const unsigned char *sha1)
ace1534d 439{
30d6c6ea 440 return !access(sha1_file_name(sha1), F_OK);
0f4dc14a 441}
ace1534d 442
0f4dc14a
BC
443int has_loose_object_nonlocal(const unsigned char *sha1)
444{
445 struct alternate_object_database *alt;
9a217f2a 446 prepare_alt_odb();
d5a63b99 447 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 448 fill_sha1_path(alt->name, sha1);
c529d75a
LT
449 if (!access(alt->base, F_OK))
450 return 1;
ace1534d 451 }
c529d75a 452 return 0;
ace1534d
JH
453}
454
0f4dc14a
BC
455static int has_loose_object(const unsigned char *sha1)
456{
457 return has_loose_object_local(sha1) ||
458 has_loose_object_nonlocal(sha1);
459}
460
60bb8b14 461static unsigned int pack_used_ctr;
a53128b6
SP
462static unsigned int pack_mmap_calls;
463static unsigned int peak_pack_open_windows;
464static unsigned int pack_open_windows;
c7934306
SP
465static unsigned int pack_open_fds;
466static unsigned int pack_max_fds;
a53128b6 467static size_t peak_pack_mapped;
60bb8b14 468static size_t pack_mapped;
9a217f2a 469struct packed_git *packed_git;
1f688557 470
b79d18c9 471void pack_report(void)
a53128b6
SP
472{
473 fprintf(stderr,
e05db0fd
PR
474 "pack_report: getpagesize() = %10" SZ_FMT "\n"
475 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
476 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
477 sz_fmt(getpagesize()),
478 sz_fmt(packed_git_window_size),
479 sz_fmt(packed_git_limit));
a53128b6
SP
480 fprintf(stderr,
481 "pack_report: pack_used_ctr = %10u\n"
482 "pack_report: pack_mmap_calls = %10u\n"
483 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
484 "pack_report: pack_mapped = "
485 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
486 pack_used_ctr,
487 pack_mmap_calls,
488 pack_open_windows, peak_pack_open_windows,
9e42d6a1 489 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
490}
491
d40d535b
MH
492/*
493 * Open and mmap the index file at path, perform a couple of
494 * consistency checks, then record its information to p. Return 0 on
495 * success.
496 */
497static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
498{
499 void *idx_map;
42873078 500 struct pack_idx_header *hdr;
dc49cd76 501 size_t idx_size;
74e34e1f 502 uint32_t version, nr, i, *index;
f4e51683 503 int fd = git_open_noatime(path);
1f688557 504 struct stat st;
42873078 505
1f688557
JH
506 if (fd < 0)
507 return -1;
508 if (fstat(fd, &st)) {
509 close(fd);
510 return -1;
511 }
dc49cd76 512 idx_size = xsize_t(st.st_size);
2d88451b
SP
513 if (idx_size < 4 * 256 + 20 + 20) {
514 close(fd);
515 return error("index file %s is too small", path);
516 }
c4712e45 517 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 518 close(fd);
1f688557 519
42873078
NP
520 hdr = idx_map;
521 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
522 version = ntohl(hdr->idx_version);
523 if (version < 2 || version > 2) {
524 munmap(idx_map, idx_size);
6e1c2344 525 return error("index file %s is version %"PRIu32
74e34e1f
NP
526 " and is not supported by this binary"
527 " (try upgrading GIT to a newer version)",
528 path, version);
529 }
530 } else
531 version = 1;
df1b059d 532
1f688557 533 nr = 0;
42873078 534 index = idx_map;
74e34e1f
NP
535 if (version > 1)
536 index += 2; /* skip index header */
1f688557 537 for (i = 0; i < 256; i++) {
326bf396 538 uint32_t n = ntohl(index[i]);
2d88451b
SP
539 if (n < nr) {
540 munmap(idx_map, idx_size);
df1b059d 541 return error("non-monotonic index %s", path);
2d88451b 542 }
1f688557
JH
543 nr = n;
544 }
545
74e34e1f
NP
546 if (version == 1) {
547 /*
548 * Total size:
549 * - 256 index entries 4 bytes each
550 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
551 * - 20-byte SHA1 of the packfile
552 * - 20-byte SHA1 file checksum
553 */
554 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
555 munmap(idx_map, idx_size);
eef427a0 556 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
557 }
558 } else if (version == 2) {
559 /*
560 * Minimum size:
561 * - 8 bytes of header
562 * - 256 index entries 4 bytes each
563 * - 20-byte sha1 entry * nr
564 * - 4-byte crc entry * nr
565 * - 4-byte offset entry * nr
566 * - 20-byte SHA1 of the packfile
567 * - 20-byte SHA1 file checksum
568 * And after the 4-byte offset table might be a
569 * variable sized table containing 8-byte entries
570 * for offsets larger than 2^31.
571 */
572 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
573 unsigned long max_size = min_size;
574 if (nr)
575 max_size += (nr - 1)*8;
576 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 577 munmap(idx_map, idx_size);
eef427a0 578 return error("wrong index v2 file size in %s", path);
74e34e1f 579 }
7109c889
JH
580 if (idx_size != min_size &&
581 /*
582 * make sure we can deal with large pack offsets.
583 * 31-bit signed offset won't be enough, neither
584 * 32-bit unsigned one will be.
585 */
586 (sizeof(off_t) <= 4)) {
587 munmap(idx_map, idx_size);
588 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 589 }
2d88451b 590 }
1f688557 591
74e34e1f 592 p->index_version = version;
42873078
NP
593 p->index_data = idx_map;
594 p->index_size = idx_size;
57059091 595 p->num_objects = nr;
1f688557
JH
596 return 0;
597}
598
bc8e478a 599int open_pack_index(struct packed_git *p)
d079837e
SP
600{
601 char *idx_name;
602 int ret;
603
604 if (p->index_data)
605 return 0;
606
607 idx_name = xstrdup(p->pack_name);
608 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
609 ret = check_packed_git_idx(idx_name, p);
610 free(idx_name);
611 return ret;
612}
613
11daf39b
SP
614static void scan_windows(struct packed_git *p,
615 struct packed_git **lru_p,
616 struct pack_window **lru_w,
617 struct pack_window **lru_l)
1f688557 618{
11daf39b
SP
619 struct pack_window *w, *w_l;
620
621 for (w_l = NULL, w = p->windows; w; w = w->next) {
622 if (!w->inuse_cnt) {
623 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
624 *lru_p = p;
625 *lru_w = w;
626 *lru_l = w_l;
54044bf8 627 }
54044bf8 628 }
11daf39b 629 w_l = w;
f9253394 630 }
11daf39b
SP
631}
632
7c3ecb32 633static int unuse_one_window(struct packed_git *current)
11daf39b
SP
634{
635 struct packed_git *p, *lru_p = NULL;
636 struct pack_window *lru_w = NULL, *lru_l = NULL;
637
638 if (current)
639 scan_windows(current, &lru_p, &lru_w, &lru_l);
640 for (p = packed_git; p; p = p->next)
641 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
642 if (lru_p) {
643 munmap(lru_w->base, lru_w->len);
644 pack_mapped -= lru_w->len;
645 if (lru_l)
646 lru_l->next = lru_w->next;
7c3ecb32 647 else
54044bf8 648 lru_p->windows = lru_w->next;
54044bf8 649 free(lru_w);
a53128b6 650 pack_open_windows--;
54044bf8
SP
651 return 1;
652 }
653 return 0;
f9253394
JH
654}
655
7c3ecb32 656void release_pack_memory(size_t need)
97bfeb34
SP
657{
658 size_t cur = pack_mapped;
7c3ecb32 659 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
660 ; /* nothing */
661}
662
58ecbd5e
JN
663void *xmmap(void *start, size_t length,
664 int prot, int flags, int fd, off_t offset)
665{
666 void *ret = mmap(start, length, prot, flags, fd, offset);
667 if (ret == MAP_FAILED) {
668 if (!length)
669 return NULL;
7c3ecb32 670 release_pack_memory(length);
58ecbd5e
JN
671 ret = mmap(start, length, prot, flags, fd, offset);
672 if (ret == MAP_FAILED)
673 die_errno("Out of memory? mmap failed");
674 }
675 return ret;
676}
677
c9ced051
SP
678void close_pack_windows(struct packed_git *p)
679{
680 while (p->windows) {
681 struct pack_window *w = p->windows;
682
683 if (w->inuse_cnt)
684 die("pack '%s' still has open windows to it",
685 p->pack_name);
686 munmap(w->base, w->len);
687 pack_mapped -= w->len;
688 pack_open_windows--;
689 p->windows = w->next;
690 free(w);
691 }
692}
693
88d0db55
BC
694/*
695 * The LRU pack is the one with the oldest MRU window, preferring packs
696 * with no used windows, or the oldest mtime if it has no windows allocated.
697 */
698static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
699{
700 struct pack_window *w, *this_mru_w;
701 int has_windows_inuse = 0;
702
703 /*
704 * Reject this pack if it has windows and the previously selected
705 * one does not. If this pack does not have windows, reject
706 * it if the pack file is newer than the previously selected one.
707 */
708 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
709 return;
710
711 for (w = this_mru_w = p->windows; w; w = w->next) {
712 /*
713 * Reject this pack if any of its windows are in use,
714 * but the previously selected pack did not have any
715 * inuse windows. Otherwise, record that this pack
716 * has windows in use.
717 */
718 if (w->inuse_cnt) {
719 if (*accept_windows_inuse)
720 has_windows_inuse = 1;
721 else
722 return;
723 }
724
725 if (w->last_used > this_mru_w->last_used)
726 this_mru_w = w;
727
728 /*
729 * Reject this pack if it has windows that have been
730 * used more recently than the previously selected pack.
731 * If the previously selected pack had windows inuse and
732 * we have not encountered a window in this pack that is
733 * inuse, skip this check since we prefer a pack with no
734 * inuse windows to one that has inuse windows.
735 */
736 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
737 this_mru_w->last_used > (*mru_w)->last_used)
738 return;
739 }
740
741 /*
742 * Select this pack.
743 */
744 *mru_w = this_mru_w;
745 *lru_p = p;
746 *accept_windows_inuse = has_windows_inuse;
747}
748
749static int close_one_pack(void)
750{
751 struct packed_git *p, *lru_p = NULL;
752 struct pack_window *mru_w = NULL;
753 int accept_windows_inuse = 1;
754
755 for (p = packed_git; p; p = p->next) {
756 if (p->pack_fd == -1)
757 continue;
758 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
759 }
760
761 if (lru_p) {
762 close(lru_p->pack_fd);
763 pack_open_fds--;
764 lru_p->pack_fd = -1;
765 return 1;
766 }
767
768 return 0;
769}
770
03e79c88 771void unuse_pack(struct pack_window **w_cursor)
f9253394 772{
03e79c88
SP
773 struct pack_window *w = *w_cursor;
774 if (w) {
775 w->inuse_cnt--;
776 *w_cursor = NULL;
777 }
1f688557
JH
778}
779
fa5fc15d
SP
780void close_pack_index(struct packed_git *p)
781{
782 if (p->index_data) {
783 munmap((void *)p->index_data, p->index_size);
784 p->index_data = NULL;
785 }
786}
787
c74faea1
NP
788/*
789 * This is used by git-repack in case a newly created pack happens to
790 * contain the same set of objects as an existing one. In that case
791 * the resulting file might be different even if its name would be the
792 * same. It is best to close any reference to the old pack before it is
01689909 793 * replaced on disk. Of course no index pointers or windows for given pack
c74faea1
NP
794 * must subsist at this point. If ever objects from this pack are requested
795 * again, the new version of the pack will be reinitialized through
796 * reprepare_packed_git().
797 */
798void free_pack_by_name(const char *pack_name)
799{
800 struct packed_git *p, **pp = &packed_git;
801
802 while (*pp) {
803 p = *pp;
804 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 805 clear_delta_base_cache();
c74faea1 806 close_pack_windows(p);
c7934306 807 if (p->pack_fd != -1) {
c74faea1 808 close(p->pack_fd);
c7934306
SP
809 pack_open_fds--;
810 }
fa5fc15d 811 close_pack_index(p);
c74faea1
NP
812 free(p->bad_object_sha1);
813 *pp = p->next;
c01f51cc
NTND
814 if (last_found_pack == p)
815 last_found_pack = NULL;
c74faea1
NP
816 free(p);
817 return;
818 }
819 pp = &p->next;
820 }
821}
822
a0788266
JS
823static unsigned int get_max_fd_limit(void)
824{
825#ifdef RLIMIT_NOFILE
491a8dec
JH
826 {
827 struct rlimit lim;
a0788266 828
491a8dec
JH
829 if (!getrlimit(RLIMIT_NOFILE, &lim))
830 return lim.rlim_cur;
831 }
832#endif
833
834#ifdef _SC_OPEN_MAX
835 {
836 long open_max = sysconf(_SC_OPEN_MAX);
837 if (0 < open_max)
838 return open_max;
839 /*
840 * Otherwise, we got -1 for one of the two
841 * reasons:
842 *
843 * (1) sysconf() did not understand _SC_OPEN_MAX
844 * and signaled an error with -1; or
845 * (2) sysconf() said there is no limit.
846 *
847 * We _could_ clear errno before calling sysconf() to
848 * tell these two cases apart and return a huge number
849 * in the latter case to let the caller cap it to a
850 * value that is not so selfish, but letting the
851 * fallback OPEN_MAX codepath take care of these cases
852 * is a lot simpler.
853 */
854 }
855#endif
a0788266 856
491a8dec 857#ifdef OPEN_MAX
a0788266
JS
858 return OPEN_MAX;
859#else
860 return 1; /* see the caller ;-) */
861#endif
862}
863
3cf8b462
SP
864/*
865 * Do not call this directly as this leaks p->pack_fd on error return;
866 * call open_packed_git() instead.
867 */
868static int open_packed_git_1(struct packed_git *p)
1f688557 869{
9bc879c1
SP
870 struct stat st;
871 struct pack_header hdr;
872 unsigned char sha1[20];
873 unsigned char *idx_sha1;
2c039da8 874 long fd_flag;
9bc879c1 875
d079837e
SP
876 if (!p->index_data && open_pack_index(p))
877 return error("packfile %s index unavailable", p->pack_name);
878
c7934306 879 if (!pack_max_fds) {
a0788266 880 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
881
882 /* Save 3 for stdin/stdout/stderr, 22 for work */
883 if (25 < max_fds)
884 pack_max_fds = max_fds - 25;
885 else
886 pack_max_fds = 1;
887 }
888
88d0db55 889 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
890 ; /* nothing */
891
f4e51683 892 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 893 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 894 return -1;
c7934306 895 pack_open_fds++;
9bc879c1
SP
896
897 /* If we created the struct before we had the pack we lack size. */
bf592c50 898 if (!p->pack_size) {
bf592c50 899 if (!S_ISREG(st.st_mode))
072db278 900 return error("packfile %s not a regular file", p->pack_name);
bf592c50 901 p->pack_size = st.st_size;
9bc879c1 902 } else if (p->pack_size != st.st_size)
072db278 903 return error("packfile %s size changed", p->pack_name);
9bc879c1 904
2c039da8
JH
905 /* We leave these file descriptors open with sliding mmap;
906 * there is no point keeping them open across exec(), though.
907 */
908 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
909 if (fd_flag < 0)
072db278 910 return error("cannot determine file descriptor flags");
2c039da8
JH
911 fd_flag |= FD_CLOEXEC;
912 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 913 return error("cannot set FD_CLOEXEC");
2c039da8 914
9bc879c1 915 /* Verify we recognize this pack file format. */
e6e2bd62 916 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 917 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 918 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 919 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 920 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
921 return error("packfile %s is version %"PRIu32" and not"
922 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
923 p->pack_name, ntohl(hdr.hdr_version));
924
925 /* Verify the pack matches its index. */
57059091 926 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
927 return error("packfile %s claims to have %"PRIu32" objects"
928 " while index indicates %"PRIu32" objects",
57059091
NP
929 p->pack_name, ntohl(hdr.hdr_entries),
930 p->num_objects);
9bc879c1 931 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 932 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 933 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 934 return error("packfile %s signature is unavailable", p->pack_name);
42873078 935 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 936 if (hashcmp(sha1, idx_sha1))
072db278
SP
937 return error("packfile %s does not match index", p->pack_name);
938 return 0;
9bc879c1
SP
939}
940
3cf8b462
SP
941static int open_packed_git(struct packed_git *p)
942{
943 if (!open_packed_git_1(p))
944 return 0;
945 if (p->pack_fd != -1) {
946 close(p->pack_fd);
c7934306 947 pack_open_fds--;
3cf8b462
SP
948 p->pack_fd = -1;
949 }
950 return -1;
951}
952
c4001d92 953static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
954{
955 /* We must promise at least 20 bytes (one hash) after the
956 * offset is available from this window, otherwise the offset
957 * is not actually in this window and a different window (which
958 * has that one hash excess) must be used. This is to support
959 * the object header and delta base parsing routines below.
960 */
961 off_t win_off = win->offset;
962 return win_off <= offset
963 && (offset + 20) <= (win_off + win->len);
964}
965
4b25d091 966unsigned char *use_pack(struct packed_git *p,
03e79c88 967 struct pack_window **w_cursor,
c4001d92 968 off_t offset,
ef49a7a0 969 unsigned long *left)
9bc879c1 970{
60bb8b14 971 struct pack_window *win = *w_cursor;
03e79c88 972
a9d98a14 973 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
974 * pointless to ask for an offset into the middle of that
975 * hash, and the in_window function above wouldn't match
976 * don't allow an offset too close to the end of the file.
977 */
d131b7af
SP
978 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
979 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
980 if (offset > (p->pack_size - 20))
981 die("offset beyond end of packfile (truncated pack?)");
982
983 if (!win || !in_window(win, offset)) {
984 if (win)
985 win->inuse_cnt--;
986 for (win = p->windows; win; win = win->next) {
987 if (in_window(win, offset))
988 break;
989 }
990 if (!win) {
78a28df9 991 size_t window_align = packed_git_window_size / 2;
dc49cd76 992 off_t len;
d131b7af
SP
993
994 if (p->pack_fd == -1 && open_packed_git(p))
995 die("packfile %s cannot be accessed", p->pack_name);
996
60bb8b14 997 win = xcalloc(1, sizeof(*win));
78a28df9 998 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
999 len = p->pack_size - win->offset;
1000 if (len > packed_git_window_size)
1001 len = packed_git_window_size;
1002 win->len = (size_t)len;
60bb8b14 1003 pack_mapped += win->len;
11daf39b 1004 while (packed_git_limit < pack_mapped
7c3ecb32 1005 && unuse_one_window(p))
60bb8b14 1006 ; /* nothing */
c4712e45 1007 win->base = xmmap(NULL, win->len,
60bb8b14
SP
1008 PROT_READ, MAP_PRIVATE,
1009 p->pack_fd, win->offset);
1010 if (win->base == MAP_FAILED)
73b4e4be
SP
1011 die("packfile %s cannot be mapped: %s",
1012 p->pack_name,
1013 strerror(errno));
d131b7af
SP
1014 if (!win->offset && win->len == p->pack_size
1015 && !p->do_not_close) {
1016 close(p->pack_fd);
1017 pack_open_fds--;
1018 p->pack_fd = -1;
1019 }
a53128b6
SP
1020 pack_mmap_calls++;
1021 pack_open_windows++;
1022 if (pack_mapped > peak_pack_mapped)
1023 peak_pack_mapped = pack_mapped;
1024 if (pack_open_windows > peak_pack_open_windows)
1025 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
1026 win->next = p->windows;
1027 p->windows = win;
1028 }
1f688557 1029 }
03e79c88
SP
1030 if (win != *w_cursor) {
1031 win->last_used = pack_used_ctr++;
1032 win->inuse_cnt++;
1033 *w_cursor = win;
1034 }
60bb8b14 1035 offset -= win->offset;
03e79c88 1036 if (left)
dc49cd76 1037 *left = win->len - xsize_t(offset);
03e79c88 1038 return win->base + offset;
1f688557
JH
1039}
1040
27d69a46
NP
1041static struct packed_git *alloc_packed_git(int extra)
1042{
1043 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1044 memset(p, 0, sizeof(*p));
1045 p->pack_fd = -1;
1046 return p;
1047}
1048
e0500293
JN
1049static void try_to_free_pack_memory(size_t size)
1050{
7c3ecb32 1051 release_pack_memory(size);
e0500293
JN
1052}
1053
42873078 1054struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 1055{
e0500293 1056 static int have_set_try_to_free_routine;
1f688557 1057 struct stat st;
27d69a46 1058 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 1059
e0500293
JN
1060 if (!have_set_try_to_free_routine) {
1061 have_set_try_to_free_routine = 1;
1062 set_try_to_free_routine(try_to_free_pack_memory);
1063 }
1064
42873078
NP
1065 /*
1066 * Make sure a corresponding .pack file exists and that
1067 * the index looks sane.
1068 */
1069 path_len -= strlen(".idx");
27d69a46
NP
1070 if (path_len < 1) {
1071 free(p);
1f688557 1072 return NULL;
27d69a46 1073 }
42873078 1074 memcpy(p->pack_name, path, path_len);
8d25931d
BC
1075
1076 strcpy(p->pack_name + path_len, ".keep");
1077 if (!access(p->pack_name, F_OK))
1078 p->pack_keep = 1;
1079
42873078 1080 strcpy(p->pack_name + path_len, ".pack");
d079837e 1081 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1082 free(p);
1f688557
JH
1083 return NULL;
1084 }
42873078 1085
1f688557
JH
1086 /* ok, it looks sane as far as we can check without
1087 * actually mapping the pack file.
1088 */
1f688557 1089 p->pack_size = st.st_size;
9d835df2 1090 p->pack_local = local;
b867092f 1091 p->mtime = st.st_mtime;
42873078
NP
1092 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1093 hashclr(p->sha1);
1f688557
JH
1094 return p;
1095}
1096
7b64469a 1097struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1098{
42873078 1099 const char *path = sha1_pack_name(sha1);
27d69a46 1100 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 1101
27d69a46
NP
1102 strcpy(p->pack_name, path);
1103 hashcpy(p->sha1, sha1);
42873078
NP
1104 if (check_packed_git_idx(idx_path, p)) {
1105 free(p);
bf592c50 1106 return NULL;
42873078 1107 }
bf592c50 1108
bf592c50
DB
1109 return p;
1110}
1111
1112void install_packed_git(struct packed_git *pack)
1113{
c7934306
SP
1114 if (pack->pack_fd != -1)
1115 pack_open_fds++;
1116
bf592c50
DB
1117 pack->next = packed_git;
1118 packed_git = pack;
1119}
1120
543c5caa
NTND
1121void (*report_garbage)(const char *desc, const char *path);
1122
1123static void report_helper(const struct string_list *list,
1124 int seen_bits, int first, int last)
1125{
1126 const char *msg;
1127 switch (seen_bits) {
1128 case 0:
235e8d59 1129 msg = "no corresponding .idx or .pack";
543c5caa
NTND
1130 break;
1131 case 1:
1132 msg = "no corresponding .idx";
1133 break;
1134 case 2:
1135 msg = "no corresponding .pack";
1136 break;
1137 default:
1138 return;
1139 }
1140 for (; first < last; first++)
1141 report_garbage(msg, list->items[first].string);
1142}
1143
1144static void report_pack_garbage(struct string_list *list)
1145{
1146 int i, baselen = -1, first = 0, seen_bits = 0;
1147
1148 if (!report_garbage)
1149 return;
1150
1151 sort_string_list(list);
1152
1153 for (i = 0; i < list->nr; i++) {
1154 const char *path = list->items[i].string;
1155 if (baselen != -1 &&
1156 strncmp(path, list->items[first].string, baselen)) {
1157 report_helper(list, seen_bits, first, i);
1158 baselen = -1;
1159 seen_bits = 0;
1160 }
1161 if (baselen == -1) {
1162 const char *dot = strrchr(path, '.');
1163 if (!dot) {
1164 report_garbage("garbage found", path);
1165 continue;
1166 }
1167 baselen = dot - path + 1;
1168 first = i;
1169 }
1170 if (!strcmp(path + baselen, "pack"))
1171 seen_bits |= 1;
1172 else if (!strcmp(path + baselen, "idx"))
1173 seen_bits |= 2;
1174 }
1175 report_helper(list, seen_bits, first, list->nr);
1176}
1177
9d835df2 1178static void prepare_packed_git_one(char *objdir, int local)
1f688557 1179{
880fb8de
RS
1180 struct strbuf path = STRBUF_INIT;
1181 size_t dirnamelen;
1f688557
JH
1182 DIR *dir;
1183 struct dirent *de;
543c5caa 1184 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557 1185
880fb8de
RS
1186 strbuf_addstr(&path, objdir);
1187 strbuf_addstr(&path, "/pack");
1188 dir = opendir(path.buf);
b5b16990 1189 if (!dir) {
26125f6b 1190 if (errno != ENOENT)
bd2afde8 1191 error("unable to open object pack directory: %s: %s",
880fb8de
RS
1192 path.buf, strerror(errno));
1193 strbuf_release(&path);
1f688557 1194 return;
b5b16990 1195 }
880fb8de
RS
1196 strbuf_addch(&path, '/');
1197 dirnamelen = path.len;
1f688557 1198 while ((de = readdir(dir)) != NULL) {
1f688557
JH
1199 struct packed_git *p;
1200
543c5caa 1201 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1202 continue;
1203
880fb8de
RS
1204 strbuf_setlen(&path, dirnamelen);
1205 strbuf_addstr(&path, de->d_name);
d90906a9 1206
2975c770 1207 if (ends_with(de->d_name, ".idx")) {
d90906a9
NTND
1208 /* Don't reopen a pack we already have. */
1209 for (p = packed_git; p; p = p->next) {
880fb8de 1210 if (!memcmp(path.buf, p->pack_name, path.len - 4))
d90906a9
NTND
1211 break;
1212 }
1213 if (p == NULL &&
1214 /*
1215 * See if it really is a valid .idx file with
1216 * corresponding .pack file that we can map.
1217 */
880fb8de 1218 (p = add_packed_git(path.buf, path.len, local)) != NULL)
d90906a9 1219 install_packed_git(p);
86f7780c 1220 }
543c5caa
NTND
1221
1222 if (!report_garbage)
1223 continue;
1224
2975c770
JK
1225 if (ends_with(de->d_name, ".idx") ||
1226 ends_with(de->d_name, ".pack") ||
1227 ends_with(de->d_name, ".bitmap") ||
1228 ends_with(de->d_name, ".keep"))
880fb8de 1229 string_list_append(&garbage, path.buf);
543c5caa 1230 else
880fb8de 1231 report_garbage("garbage found", path.buf);
1f688557 1232 }
5b35bcd5 1233 closedir(dir);
543c5caa
NTND
1234 report_pack_garbage(&garbage);
1235 string_list_clear(&garbage, 0);
880fb8de 1236 strbuf_release(&path);
1f688557
JH
1237}
1238
b867092f
JH
1239static int sort_pack(const void *a_, const void *b_)
1240{
1241 struct packed_git *a = *((struct packed_git **)a_);
1242 struct packed_git *b = *((struct packed_git **)b_);
1243 int st;
1244
1245 /*
1246 * Local packs tend to contain objects specific to our
1247 * variant of the project than remote ones. In addition,
1248 * remote ones could be on a network mounted filesystem.
1249 * Favor local ones for these reasons.
1250 */
1251 st = a->pack_local - b->pack_local;
1252 if (st)
1253 return -st;
1254
1255 /*
1256 * Younger packs tend to contain more recent objects,
1257 * and more recent objects tend to get accessed more
1258 * often.
1259 */
1260 if (a->mtime < b->mtime)
1261 return 1;
1262 else if (a->mtime == b->mtime)
1263 return 0;
1264 return -1;
1265}
1266
1267static void rearrange_packed_git(void)
1268{
1269 struct packed_git **ary, *p;
1270 int i, n;
1271
1272 for (n = 0, p = packed_git; p; p = p->next)
1273 n++;
1274 if (n < 2)
1275 return;
1276
1277 /* prepare an array of packed_git for easier sorting */
1278 ary = xcalloc(n, sizeof(struct packed_git *));
1279 for (n = 0, p = packed_git; p; p = p->next)
1280 ary[n++] = p;
1281
1282 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1283
1284 /* link them back again */
1285 for (i = 0; i < n - 1; i++)
1286 ary[i]->next = ary[i + 1];
1287 ary[n - 1]->next = NULL;
1288 packed_git = ary[0];
1289
1290 free(ary);
1291}
1292
637cdd9d 1293static int prepare_packed_git_run_once = 0;
9a217f2a 1294void prepare_packed_git(void)
1f688557 1295{
d5a63b99 1296 struct alternate_object_database *alt;
1f688557 1297
637cdd9d 1298 if (prepare_packed_git_run_once)
1f688557 1299 return;
9d835df2 1300 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1301 prepare_alt_odb();
d5a63b99 1302 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1303 alt->name[-1] = 0;
9d835df2 1304 prepare_packed_git_one(alt->base, 0);
1494e038 1305 alt->name[-1] = '/';
1f688557 1306 }
b867092f 1307 rearrange_packed_git();
637cdd9d
JK
1308 prepare_packed_git_run_once = 1;
1309}
1310
fc04c412 1311void reprepare_packed_git(void)
637cdd9d
JK
1312{
1313 prepare_packed_git_run_once = 0;
1314 prepare_packed_git();
1f688557
JH
1315}
1316
8eca0b47
NP
1317static void mark_bad_packed_object(struct packed_git *p,
1318 const unsigned char *sha1)
1319{
1320 unsigned i;
1321 for (i = 0; i < p->num_bad_objects; i++)
1322 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1323 return;
1324 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1325 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1326 p->num_bad_objects++;
1327}
1328
b6c4cecc 1329static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1330{
1331 struct packed_git *p;
1332 unsigned i;
1333
1334 for (p = packed_git; p; p = p->next)
1335 for (i = 0; i < p->num_bad_objects; i++)
1336 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1337 return p;
1338 return NULL;
ac939109
NP
1339}
1340
090ea126
NTND
1341/*
1342 * With an in-core object data in "map", rehash it to make sure the
1343 * object name actually matches "sha1" to detect object corruption.
1344 * With "map" == NULL, try reading the object named with "sha1" using
1345 * the streaming interface and rehash it to do the same.
1346 */
1347int check_sha1_signature(const unsigned char *sha1, void *map,
1348 unsigned long size, const char *type)
0fcfd160
LT
1349{
1350 unsigned char real_sha1[20];
090ea126
NTND
1351 enum object_type obj_type;
1352 struct git_istream *st;
1353 git_SHA_CTX c;
1354 char hdr[32];
1355 int hdrlen;
1356
1357 if (map) {
1358 hash_sha1_file(map, size, type, real_sha1);
1359 return hashcmp(sha1, real_sha1) ? -1 : 0;
1360 }
1361
1362 st = open_istream(sha1, &obj_type, &size, NULL);
1363 if (!st)
1364 return -1;
1365
1366 /* Generate the header */
1367 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1368
1369 /* Sha1.. */
1370 git_SHA1_Init(&c);
1371 git_SHA1_Update(&c, hdr, hdrlen);
1372 for (;;) {
1373 char buf[1024 * 16];
1374 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1375
f54fac53
JK
1376 if (readlen < 0) {
1377 close_istream(st);
1378 return -1;
1379 }
090ea126
NTND
1380 if (!readlen)
1381 break;
1382 git_SHA1_Update(&c, buf, readlen);
1383 }
1384 git_SHA1_Final(real_sha1, &c);
1385 close_istream(st);
a89fccd2 1386 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1387}
1388
ec73f580 1389int git_open_noatime(const char *name)
44d1c19e
LT
1390{
1391 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1392
f2e872aa
SP
1393 for (;;) {
1394 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1395 if (fd >= 0)
f2e872aa
SP
1396 return fd;
1397
f2e872aa
SP
1398 /* Might the failure be due to O_NOATIME? */
1399 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1400 sha1_file_open_flag = 0;
f2e872aa
SP
1401 continue;
1402 }
1403
1404 return -1;
44d1c19e 1405 }
44d1c19e
LT
1406}
1407
052fe5ea
JK
1408static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1409{
052fe5ea
JK
1410 struct alternate_object_database *alt;
1411
30d6c6ea 1412 if (!lstat(sha1_file_name(sha1), st))
052fe5ea
JK
1413 return 0;
1414
1415 prepare_alt_odb();
1416 errno = ENOENT;
1417 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1418 fill_sha1_path(alt->name, sha1);
052fe5ea
JK
1419 if (!lstat(alt->base, st))
1420 return 0;
1421 }
1422
1423 return -1;
1424}
1425
44d1c19e
LT
1426static int open_sha1_file(const unsigned char *sha1)
1427{
1428 int fd;
44d1c19e 1429 struct alternate_object_database *alt;
d6c8a05b 1430 int most_interesting_errno;
44d1c19e 1431
30d6c6ea 1432 fd = git_open_noatime(sha1_file_name(sha1));
44d1c19e
LT
1433 if (fd >= 0)
1434 return fd;
d6c8a05b 1435 most_interesting_errno = errno;
44d1c19e
LT
1436
1437 prepare_alt_odb();
44d1c19e 1438 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1439 fill_sha1_path(alt->name, sha1);
f4e51683 1440 fd = git_open_noatime(alt->base);
44d1c19e
LT
1441 if (fd >= 0)
1442 return fd;
d6c8a05b
JK
1443 if (most_interesting_errno == ENOENT)
1444 most_interesting_errno = errno;
44d1c19e 1445 }
d6c8a05b 1446 errno = most_interesting_errno;
44d1c19e
LT
1447 return -1;
1448}
1449
f0270efd 1450void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1451{
0fcfd160 1452 void *map;
144bde78 1453 int fd;
ace1534d 1454
44d1c19e
LT
1455 fd = open_sha1_file(sha1);
1456 map = NULL;
1457 if (fd >= 0) {
1458 struct stat st;
0fcfd160 1459
44d1c19e
LT
1460 if (!fstat(fd, &st)) {
1461 *size = xsize_t(st.st_size);
33e42de0
MM
1462 if (!*size) {
1463 /* mmap() is forbidden on empty files */
1464 error("object file %s is empty", sha1_file_name(sha1));
1465 return NULL;
1466 }
44d1c19e 1467 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1468 }
44d1c19e 1469 close(fd);
0fcfd160 1470 }
0fcfd160
LT
1471 return map;
1472}
1473
09ded04b
NP
1474unsigned long unpack_object_header_buffer(const unsigned char *buf,
1475 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1476{
ad1ed5ee 1477 unsigned shift;
48fb7deb 1478 unsigned long size, c;
ad1ed5ee
JH
1479 unsigned long used = 0;
1480
1481 c = buf[used++];
1482 *type = (c >> 4) & 7;
1483 size = c & 15;
1484 shift = 4;
1485 while (c & 0x80) {
f630cfda 1486 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1487 error("bad object header");
ea4f9685
JH
1488 size = used = 0;
1489 break;
09ded04b 1490 }
ad1ed5ee
JH
1491 c = buf[used++];
1492 size += (c & 0x7f) << shift;
1493 shift += 7;
1494 }
1495 *sizep = size;
1496 return used;
1497}
1498
eb4f4076 1499int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee 1500{
c4483576
LT
1501 /* Get the data stream */
1502 memset(stream, 0, sizeof(*stream));
1503 stream->next_in = map;
1504 stream->avail_in = mapsize;
1505 stream->next_out = buffer;
93821bd9
LT
1506 stream->avail_out = bufsiz;
1507
39c68542 1508 git_inflate_init(stream);
cc5c54e7 1509 return git_inflate(stream, 0);
c4483576
LT
1510}
1511
ef49a7a0 1512static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1513{
1514 int bytes = strlen(buffer) + 1;
3aee68aa 1515 unsigned char *buf = xmallocz(size);
93821bd9 1516 unsigned long n;
7efbff75 1517 int status = Z_OK;
5180cacc 1518
93821bd9
LT
1519 n = stream->total_out - bytes;
1520 if (n > size)
1521 n = size;
1522 memcpy(buf, (char *) buffer + bytes, n);
1523 bytes = n;
456cdf6e
LT
1524 if (bytes <= size) {
1525 /*
1526 * The above condition must be (bytes <= size), not
1527 * (bytes < size). In other words, even though we
ccf5ace0 1528 * expect no more output and set avail_out to zero,
456cdf6e
LT
1529 * the input zlib stream may have bytes that express
1530 * "this concludes the stream", and we *do* want to
1531 * eat that input.
1532 *
1533 * Otherwise we would not be able to test that we
1534 * consumed all the input to reach the expected size;
1535 * we also want to check that zlib tells us that all
1536 * went well with status == Z_STREAM_END at the end.
1537 */
5180cacc
LT
1538 stream->next_out = buf + bytes;
1539 stream->avail_out = size - bytes;
7efbff75 1540 while (status == Z_OK)
39c68542 1541 status = git_inflate(stream, Z_FINISH);
5180cacc 1542 }
456cdf6e 1543 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1544 git_inflate_end(stream);
7efbff75
JH
1545 return buf;
1546 }
1547
1548 if (status < 0)
1549 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1550 else if (stream->avail_in)
1551 error("garbage at end of loose object '%s'",
1552 sha1_to_hex(sha1));
1553 free(buf);
1554 return NULL;
5180cacc
LT
1555}
1556
1557/*
1558 * We used to just use "sscanf()", but that's actually way
1559 * too permissive for what we want to check. So do an anal
1560 * object header parse by hand.
1561 */
f0270efd 1562int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1563{
21666f1a 1564 char type[10];
5180cacc
LT
1565 int i;
1566 unsigned long size;
1567
1568 /*
a6080a0a 1569 * The type can be at most ten bytes (including the
5180cacc 1570 * terminating '\0' that we add), and is followed by
21666f1a 1571 * a space.
5180cacc 1572 */
21666f1a 1573 i = 0;
5180cacc
LT
1574 for (;;) {
1575 char c = *hdr++;
1576 if (c == ' ')
1577 break;
21666f1a
NP
1578 type[i++] = c;
1579 if (i >= sizeof(type))
5180cacc 1580 return -1;
5180cacc 1581 }
21666f1a 1582 type[i] = 0;
5180cacc
LT
1583
1584 /*
1585 * The length must follow immediately, and be in canonical
1586 * decimal format (ie "010" is not valid).
1587 */
1588 size = *hdr++ - '0';
1589 if (size > 9)
1590 return -1;
1591 if (size) {
1592 for (;;) {
1593 unsigned long c = *hdr - '0';
1594 if (c > 9)
1595 break;
1596 hdr++;
1597 size = size * 10 + c;
1598 }
1599 }
1600 *sizep = size;
1601
1602 /*
1603 * The length must be followed by a zero byte
1604 */
21666f1a 1605 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1606}
1607
7efbff75 1608static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1609{
5180cacc 1610 int ret;
ef49a7a0 1611 git_zstream stream;
5180cacc 1612 char hdr[8192];
0fcfd160 1613
5180cacc 1614 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1615 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1616 return NULL;
1617
7efbff75 1618 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1619}
1620
54dab52a
NP
1621unsigned long get_size_from_delta(struct packed_git *p,
1622 struct pack_window **w_curs,
1623 off_t curpos)
1624{
1625 const unsigned char *data;
1626 unsigned char delta_head[20], *in;
ef49a7a0 1627 git_zstream stream;
54dab52a
NP
1628 int st;
1629
1630 memset(&stream, 0, sizeof(stream));
1631 stream.next_out = delta_head;
1632 stream.avail_out = sizeof(delta_head);
1633
39c68542 1634 git_inflate_init(&stream);
54dab52a
NP
1635 do {
1636 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1637 stream.next_in = in;
39c68542 1638 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1639 curpos += stream.next_in - in;
1640 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1641 stream.total_out < sizeof(delta_head));
39c68542 1642 git_inflate_end(&stream);
3d77d877
NP
1643 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1644 error("delta data unpack-initial failed");
1645 return 0;
1646 }
54dab52a
NP
1647
1648 /* Examine the initial part of the delta to figure out
1649 * the result size.
1650 */
1651 data = delta_head;
1652
1653 /* ignore base size */
1654 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1655
1656 /* Read the result size */
1657 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1658}
1659
c4001d92 1660static off_t get_delta_base(struct packed_git *p,
03e79c88 1661 struct pack_window **w_curs,
c4001d92 1662 off_t *curpos,
21666f1a 1663 enum object_type type,
c4001d92 1664 off_t delta_obj_offset)
eb32d236 1665{
2b87c45b 1666 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1667 off_t base_offset;
eb32d236 1668
8d8a4ea5
SP
1669 /* use_pack() assured us we have [base_info, base_info + 20)
1670 * as a range that we can look at without walking off the
1671 * end of the mapped window. Its actually the hash size
1672 * that is assured. An OFS_DELTA longer than the hash size
1673 * is stupid, as then a REF_DELTA would be smaller to store.
1674 */
21666f1a 1675 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1676 unsigned used = 0;
1677 unsigned char c = base_info[used++];
1678 base_offset = c & 127;
1679 while (c & 128) {
1680 base_offset += 1;
8723f216 1681 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1682 return 0; /* overflow */
eb32d236
NP
1683 c = base_info[used++];
1684 base_offset = (base_offset << 7) + (c & 127);
1685 }
1686 base_offset = delta_obj_offset - base_offset;
d8f32556 1687 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1688 return 0; /* out of bound */
2b87c45b 1689 *curpos += used;
21666f1a 1690 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1691 /* The base entry _must_ be in the same pack */
1692 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1693 *curpos += 20;
eb32d236
NP
1694 } else
1695 die("I am totally screwed");
2b87c45b 1696 return base_offset;
eb32d236
NP
1697}
1698
5d642e75
JK
1699/*
1700 * Like get_delta_base above, but we return the sha1 instead of the pack
1701 * offset. This means it is cheaper for REF deltas (we do not have to do
1702 * the final object lookup), but more expensive for OFS deltas (we
1703 * have to load the revidx to convert the offset back into a sha1).
1704 */
1705static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1706 struct pack_window **w_curs,
1707 off_t curpos,
1708 enum object_type type,
1709 off_t delta_obj_offset)
1710{
1711 if (type == OBJ_REF_DELTA) {
1712 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1713 return base;
1714 } else if (type == OBJ_OFS_DELTA) {
1715 struct revindex_entry *revidx;
1716 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1717 type, delta_obj_offset);
1718
1719 if (!base_offset)
1720 return NULL;
1721
1722 revidx = find_pack_revindex(p, base_offset);
1723 if (!revidx)
1724 return NULL;
1725
1726 return nth_packed_object_sha1(p, revidx->nr);
1727 } else
1728 return NULL;
1729}
1730
f8c8abc5
JH
1731int unpack_object_header(struct packed_git *p,
1732 struct pack_window **w_curs,
1733 off_t *curpos,
1734 unsigned long *sizep)
a733cb60 1735{
03e79c88 1736 unsigned char *base;
ef49a7a0 1737 unsigned long left;
ad1ed5ee 1738 unsigned long used;
2b87c45b 1739 enum object_type type;
a733cb60 1740
8d8a4ea5 1741 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1742 * as a range that we can look at. (Its actually the hash
3dff5379 1743 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1744 * the maximum deflated object size is 2^137, which is just
1745 * insane, so we know won't exceed what we have been given.
1746 */
2b87c45b 1747 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1748 used = unpack_object_header_buffer(base, left, &type, sizep);
1749 if (!used) {
1750 type = OBJ_BAD;
1751 } else
1752 *curpos += used;
ad1ed5ee 1753
2b87c45b 1754 return type;
a733cb60
LT
1755}
1756
790d96c0
TR
1757static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1758{
1759 int type;
1760 struct revindex_entry *revidx;
1761 const unsigned char *sha1;
1762 revidx = find_pack_revindex(p, obj_offset);
1763 if (!revidx)
1764 return OBJ_BAD;
1765 sha1 = nth_packed_object_sha1(p, revidx->nr);
1766 mark_bad_packed_object(p, sha1);
1767 type = sha1_object_info(sha1, NULL);
1768 if (type <= OBJ_NONE)
1769 return OBJ_BAD;
1770 return type;
1771}
1772
790d96c0
TR
1773#define POI_STACK_PREALLOC 64
1774
90191d37
JK
1775static enum object_type packed_to_object_type(struct packed_git *p,
1776 off_t obj_offset,
1777 enum object_type type,
1778 struct pack_window **w_curs,
1779 off_t curpos)
1f688557 1780{
790d96c0
TR
1781 off_t small_poi_stack[POI_STACK_PREALLOC];
1782 off_t *poi_stack = small_poi_stack;
1783 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1784
790d96c0
TR
1785 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1786 off_t base_offset;
90191d37 1787 unsigned long size;
790d96c0
TR
1788 /* Push the object we're going to leave behind */
1789 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1790 poi_stack_alloc = alloc_nr(poi_stack_nr);
1791 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1792 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1793 } else {
1794 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1795 }
1796 poi_stack[poi_stack_nr++] = obj_offset;
1797 /* If parsing the base offset fails, just unwind */
90191d37 1798 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1799 if (!base_offset)
1800 goto unwind;
1801 curpos = obj_offset = base_offset;
90191d37 1802 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1803 if (type <= OBJ_NONE) {
1804 /* If getting the base itself fails, we first
1805 * retry the base, otherwise unwind */
1806 type = retry_bad_packed_offset(p, base_offset);
1807 if (type > OBJ_NONE)
1808 goto out;
1809 goto unwind;
1810 }
1811 }
1812
21666f1a 1813 switch (type) {
790d96c0 1814 case OBJ_BAD:
a733cb60 1815 case OBJ_COMMIT:
a733cb60 1816 case OBJ_TREE:
a733cb60 1817 case OBJ_BLOB:
a733cb60 1818 case OBJ_TAG:
a69d0943 1819 break;
1f688557 1820 default:
3d77d877
NP
1821 error("unknown object type %i at offset %"PRIuMAX" in %s",
1822 type, (uintmax_t)obj_offset, p->pack_name);
1823 type = OBJ_BAD;
1f688557 1824 }
790d96c0
TR
1825
1826out:
1827 if (poi_stack != small_poi_stack)
1828 free(poi_stack);
21666f1a 1829 return type;
790d96c0
TR
1830
1831unwind:
1832 while (poi_stack_nr) {
1833 obj_offset = poi_stack[--poi_stack_nr];
1834 type = retry_bad_packed_offset(p, obj_offset);
1835 if (type > OBJ_NONE)
1836 goto out;
1837 }
1838 type = OBJ_BAD;
1839 goto out;
1f688557
JH
1840}
1841
90191d37 1842static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 1843 struct object_info *oi)
90191d37
JK
1844{
1845 struct pack_window *w_curs = NULL;
1846 unsigned long size;
1847 off_t curpos = obj_offset;
1848 enum object_type type;
1849
412916ee
JK
1850 /*
1851 * We always get the representation type, but only convert it to
1852 * a "real" type later if the caller is interested.
1853 */
90191d37
JK
1854 type = unpack_object_header(p, &w_curs, &curpos, &size);
1855
23c339c0 1856 if (oi->sizep) {
90191d37
JK
1857 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1858 off_t tmp_pos = curpos;
1859 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1860 type, obj_offset);
1861 if (!base_offset) {
1862 type = OBJ_BAD;
1863 goto out;
1864 }
23c339c0
JK
1865 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1866 if (*oi->sizep == 0) {
90191d37
JK
1867 type = OBJ_BAD;
1868 goto out;
1869 }
1870 } else {
23c339c0 1871 *oi->sizep = size;
90191d37
JK
1872 }
1873 }
1874
23c339c0 1875 if (oi->disk_sizep) {
90191d37 1876 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 1877 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
1878 }
1879
23c339c0
JK
1880 if (oi->typep) {
1881 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1882 if (*oi->typep < 0) {
412916ee
JK
1883 type = OBJ_BAD;
1884 goto out;
1885 }
1886 }
90191d37 1887
5d642e75
JK
1888 if (oi->delta_base_sha1) {
1889 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1890 const unsigned char *base;
1891
1892 base = get_delta_base_sha1(p, &w_curs, curpos,
1893 type, obj_offset);
1894 if (!base) {
1895 type = OBJ_BAD;
1896 goto out;
1897 }
1898
1899 hashcpy(oi->delta_base_sha1, base);
1900 } else
1901 hashclr(oi->delta_base_sha1);
1902 }
1903
90191d37
JK
1904out:
1905 unuse_pack(&w_curs);
1906 return type;
1907}
1908
eb950c19 1909static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1910 struct pack_window **w_curs,
c4001d92 1911 off_t curpos,
eb950c19 1912 unsigned long size)
de530aaa
SP
1913{
1914 int st;
ef49a7a0 1915 git_zstream stream;
079afb18 1916 unsigned char *buffer, *in;
de530aaa 1917
4ab07e4d 1918 buffer = xmallocz(size);
de530aaa 1919 memset(&stream, 0, sizeof(stream));
de530aaa 1920 stream.next_out = buffer;
39eea7bd 1921 stream.avail_out = size + 1;
de530aaa 1922
39c68542 1923 git_inflate_init(&stream);
079afb18 1924 do {
2b87c45b 1925 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1926 stream.next_in = in;
39c68542 1927 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1928 if (!stream.avail_out)
1929 break; /* the payload is larger than it should be */
2b87c45b 1930 curpos += stream.next_in - in;
079afb18 1931 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1932 git_inflate_end(&stream);
de530aaa
SP
1933 if ((st != Z_STREAM_END) || stream.total_out != size) {
1934 free(buffer);
1935 return NULL;
1936 }
1937
1938 return buffer;
1939}
1940
e5e01619
LT
1941#define MAX_DELTA_CACHE (256)
1942
18bdec11 1943static size_t delta_base_cached;
5e08ecbf
NP
1944
1945static struct delta_base_cache_lru_list {
1946 struct delta_base_cache_lru_list *prev;
1947 struct delta_base_cache_lru_list *next;
1948} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1949
e5e01619 1950static struct delta_base_cache_entry {
5e08ecbf
NP
1951 struct delta_base_cache_lru_list lru;
1952 void *data;
e5e01619
LT
1953 struct packed_git *p;
1954 off_t base_offset;
1955 unsigned long size;
e5e01619
LT
1956 enum object_type type;
1957} delta_base_cache[MAX_DELTA_CACHE];
1958
1959static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1960{
1961 unsigned long hash;
1962
1963 hash = (unsigned long)p + (unsigned long)base_offset;
1964 hash += (hash >> 8) + (hash >> 16);
3358004a 1965 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1966}
1967
84dd81c1
TR
1968static struct delta_base_cache_entry *
1969get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
1970{
1971 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
1972 return delta_base_cache + hash;
1973}
1974
1975static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1976 struct packed_git *p, off_t base_offset)
1977{
5266d369
JH
1978 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1979}
1980
84dd81c1
TR
1981static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1982{
1983 struct delta_base_cache_entry *ent;
1984 ent = get_delta_base_cache_entry(p, base_offset);
1985 return eq_delta_base_cache_entry(ent, p, base_offset);
1986}
1987
1988static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1989{
1990 ent->data = NULL;
1991 ent->lru.next->prev = ent->lru.prev;
1992 ent->lru.prev->next = ent->lru.next;
1993 delta_base_cached -= ent->size;
1994}
1995
62f255ad 1996static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1997 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1998{
84dd81c1 1999 struct delta_base_cache_entry *ent;
e5e01619 2000 void *ret;
e5e01619 2001
84dd81c1
TR
2002 ent = get_delta_base_cache_entry(p, base_offset);
2003
2004 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2005 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2006
84dd81c1
TR
2007 ret = ent->data;
2008
2009 if (!keep_cache)
2010 clear_delta_base_cache_entry(ent);
2011 else
182af834 2012 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
2013 *type = ent->type;
2014 *base_size = ent->size;
2015 return ret;
62f255ad
LT
2016}
2017
18bdec11
SP
2018static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2019{
2020 if (ent->data) {
2021 free(ent->data);
2022 ent->data = NULL;
5e08ecbf
NP
2023 ent->lru.next->prev = ent->lru.prev;
2024 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
2025 delta_base_cached -= ent->size;
2026 }
2027}
2028
3d20c636
SP
2029void clear_delta_base_cache(void)
2030{
2031 unsigned long p;
2032 for (p = 0; p < MAX_DELTA_CACHE; p++)
2033 release_delta_base_cache(&delta_base_cache[p]);
2034}
2035
62f255ad
LT
2036static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2037 void *base, unsigned long base_size, enum object_type type)
2038{
5e08ecbf 2039 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2040 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 2041 struct delta_base_cache_lru_list *lru;
e5e01619 2042
18bdec11
SP
2043 release_delta_base_cache(ent);
2044 delta_base_cached += base_size;
5e08ecbf
NP
2045
2046 for (lru = delta_base_cache_lru.next;
2047 delta_base_cached > delta_base_cache_limit
2048 && lru != &delta_base_cache_lru;
2049 lru = lru->next) {
2050 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
2051 if (f->type == OBJ_BLOB)
2052 release_delta_base_cache(f);
2053 }
5e08ecbf
NP
2054 for (lru = delta_base_cache_lru.next;
2055 delta_base_cached > delta_base_cache_limit
2056 && lru != &delta_base_cache_lru;
2057 lru = lru->next) {
2058 struct delta_base_cache_entry *f = (void *)lru;
2059 release_delta_base_cache(f);
2060 }
18bdec11 2061
e5e01619
LT
2062 ent->p = p;
2063 ent->base_offset = base_offset;
2064 ent->type = type;
2065 ent->data = base;
2066 ent->size = base_size;
5e08ecbf
NP
2067 ent->lru.next = &delta_base_cache_lru;
2068 ent->lru.prev = delta_base_cache_lru.prev;
2069 delta_base_cache_lru.prev->next = &ent->lru;
2070 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
2071}
2072
c2c5b270
CC
2073static void *read_object(const unsigned char *sha1, enum object_type *type,
2074 unsigned long *size);
2075
5f44324d
JH
2076static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2077{
2078 static FILE *log_file;
2079
b12ca963
NTND
2080 if (!log_pack_access)
2081 log_pack_access = getenv("GIT_TRACE_PACK_ACCESS");
2082 if (!log_pack_access)
2083 log_pack_access = no_log_pack_access;
2084 if (log_pack_access == no_log_pack_access)
2085 return;
2086
5f44324d
JH
2087 if (!log_file) {
2088 log_file = fopen(log_pack_access, "w");
2089 if (!log_file) {
2090 error("cannot open pack access log '%s' for writing: %s",
2091 log_pack_access, strerror(errno));
b12ca963 2092 log_pack_access = no_log_pack_access;
5f44324d
JH
2093 return;
2094 }
2095 }
2096 fprintf(log_file, "%s %"PRIuMAX"\n",
2097 p->pack_name, (uintmax_t)obj_offset);
2098 fflush(log_file);
2099}
2100
0e8189e2
NP
2101int do_check_packed_object_crc;
2102
abe601bb
TR
2103#define UNPACK_ENTRY_STACK_PREALLOC 64
2104struct unpack_entry_stack_ent {
2105 off_t obj_offset;
2106 off_t curpos;
2107 unsigned long size;
2108};
2109
c4001d92 2110void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2111 enum object_type *final_type, unsigned long *final_size)
1f688557 2112{
03e79c88 2113 struct pack_window *w_curs = NULL;
c4001d92 2114 off_t curpos = obj_offset;
abe601bb
TR
2115 void *data = NULL;
2116 unsigned long size;
2117 enum object_type type;
2118 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2119 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2120 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2121 int base_from_cache = 0;
1f688557 2122
b12ca963 2123 if (log_pack_access != no_log_pack_access)
5f44324d
JH
2124 write_pack_access_log(p, obj_offset);
2125
abe601bb
TR
2126 /* PHASE 1: drill down to the innermost base object */
2127 for (;;) {
2128 off_t base_offset;
2129 int i;
2130 struct delta_base_cache_entry *ent;
2131
77965f8b
NTND
2132 ent = get_delta_base_cache_entry(p, curpos);
2133 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2134 type = ent->type;
2135 data = ent->data;
2136 size = ent->size;
2137 clear_delta_base_cache_entry(ent);
2138 base_from_cache = 1;
2139 break;
2140 }
2141
abe601bb
TR
2142 if (do_check_packed_object_crc && p->index_version > 1) {
2143 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2144 unsigned long len = revidx[1].offset - obj_offset;
2145 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2146 const unsigned char *sha1 =
2147 nth_packed_object_sha1(p, revidx->nr);
2148 error("bad packed object CRC for %s",
2149 sha1_to_hex(sha1));
2150 mark_bad_packed_object(p, sha1);
2151 unuse_pack(&w_curs);
2152 return NULL;
2153 }
2154 }
2155
abe601bb
TR
2156 type = unpack_object_header(p, &w_curs, &curpos, &size);
2157 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2158 break;
2159
2160 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2161 if (!base_offset) {
2162 error("failed to validate delta base reference "
2163 "at offset %"PRIuMAX" from %s",
2164 (uintmax_t)curpos, p->pack_name);
2165 /* bail to phase 2, in hopes of recovery */
2166 data = NULL;
2167 break;
2168 }
2169
2170 /* push object, proceed to base */
2171 if (delta_stack_nr >= delta_stack_alloc
2172 && delta_stack == small_delta_stack) {
2173 delta_stack_alloc = alloc_nr(delta_stack_nr);
2174 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2175 memcpy(delta_stack, small_delta_stack,
2176 sizeof(*delta_stack)*delta_stack_nr);
2177 } else {
2178 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2179 }
2180 i = delta_stack_nr++;
2181 delta_stack[i].obj_offset = obj_offset;
2182 delta_stack[i].curpos = curpos;
2183 delta_stack[i].size = size;
2184
2185 curpos = obj_offset = base_offset;
0e8189e2
NP
2186 }
2187
abe601bb
TR
2188 /* PHASE 2: handle the base */
2189 switch (type) {
eb32d236
NP
2190 case OBJ_OFS_DELTA:
2191 case OBJ_REF_DELTA:
abe601bb
TR
2192 if (data)
2193 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2194 break;
a733cb60 2195 case OBJ_COMMIT:
a733cb60 2196 case OBJ_TREE:
a733cb60 2197 case OBJ_BLOB:
a733cb60 2198 case OBJ_TAG:
abe601bb
TR
2199 if (!base_from_cache)
2200 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2201 break;
1f688557 2202 default:
8eca0b47
NP
2203 data = NULL;
2204 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2205 type, (uintmax_t)obj_offset, p->pack_name);
2206 }
2207
2208 /* PHASE 3: apply deltas in order */
2209
2210 /* invariants:
2211 * 'data' holds the base data, or NULL if there was corruption
2212 */
2213 while (delta_stack_nr) {
2214 void *delta_data;
2215 void *base = data;
2216 unsigned long delta_size, base_size = size;
2217 int i;
2218
2219 data = NULL;
2220
2221 if (base)
2222 add_delta_base_cache(p, obj_offset, base, base_size, type);
2223
2224 if (!base) {
2225 /*
2226 * We're probably in deep shit, but let's try to fetch
2227 * the required base anyway from another pack or loose.
2228 * This is costly but should happen only in the presence
2229 * of a corrupted pack, and is better than failing outright.
2230 */
2231 struct revindex_entry *revidx;
2232 const unsigned char *base_sha1;
2233 revidx = find_pack_revindex(p, obj_offset);
2234 if (revidx) {
2235 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2236 error("failed to read delta base object %s"
2237 " at offset %"PRIuMAX" from %s",
2238 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2239 p->pack_name);
2240 mark_bad_packed_object(p, base_sha1);
2241 base = read_object(base_sha1, &type, &base_size);
2242 }
2243 }
2244
2245 i = --delta_stack_nr;
2246 obj_offset = delta_stack[i].obj_offset;
2247 curpos = delta_stack[i].curpos;
2248 delta_size = delta_stack[i].size;
2249
2250 if (!base)
2251 continue;
2252
2253 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2254
2255 if (!delta_data) {
2256 error("failed to unpack compressed delta "
2257 "at offset %"PRIuMAX" from %s",
2258 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2259 data = NULL;
2260 continue;
2261 }
2262
2263 data = patch_delta(base, base_size,
2264 delta_data, delta_size,
2265 &size);
1ee886c1
JK
2266
2267 /*
2268 * We could not apply the delta; warn the user, but keep going.
2269 * Our failure will be noticed either in the next iteration of
2270 * the loop, or if this is the final delta, in the caller when
2271 * we return NULL. Those code paths will take care of making
2272 * a more explicit warning and retrying with another copy of
2273 * the object.
2274 */
abe601bb 2275 if (!data)
1ee886c1 2276 error("failed to apply delta");
abe601bb 2277
4b8f772c 2278 free(delta_data);
1f688557 2279 }
abe601bb
TR
2280
2281 *final_type = type;
2282 *final_size = size;
2283
03e79c88 2284 unuse_pack(&w_curs);
019d1e65
NTND
2285
2286 if (delta_stack != small_delta_stack)
2287 free(delta_stack);
2288
21666f1a 2289 return data;
1f688557
JH
2290}
2291
d079837e 2292const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2293 uint32_t n)
9a217f2a 2294{
42873078 2295 const unsigned char *index = p->index_data;
d079837e
SP
2296 if (!index) {
2297 if (open_pack_index(p))
2298 return NULL;
2299 index = p->index_data;
2300 }
57059091 2301 if (n >= p->num_objects)
d72308e0 2302 return NULL;
74e34e1f
NP
2303 index += 4 * 256;
2304 if (p->index_version == 1) {
2305 return index + 24 * n + 4;
2306 } else {
2307 index += 8;
2308 return index + 20 * n;
2309 }
2310}
2311
99093238 2312off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2313{
2314 const unsigned char *index = p->index_data;
2315 index += 4 * 256;
2316 if (p->index_version == 1) {
2317 return ntohl(*((uint32_t *)(index + 24 * n)));
2318 } else {
2319 uint32_t off;
2320 index += 8 + p->num_objects * (20 + 4);
2321 off = ntohl(*((uint32_t *)(index + 4 * n)));
2322 if (!(off & 0x80000000))
2323 return off;
2324 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2325 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2326 ntohl(*((uint32_t *)(index + 4)));
2327 }
9a217f2a
JH
2328}
2329
c4001d92 2330off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2331 struct packed_git *p)
1f688557 2332{
42873078 2333 const uint32_t *level1_ofs = p->index_data;
42873078 2334 const unsigned char *index = p->index_data;
628522ec
JH
2335 unsigned hi, lo, stride;
2336 static int use_lookup = -1;
2337 static int debug_lookup = -1;
2338
2339 if (debug_lookup < 0)
2340 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2341
d079837e
SP
2342 if (!index) {
2343 if (open_pack_index(p))
2344 return 0;
2345 level1_ofs = p->index_data;
2346 index = p->index_data;
2347 }
74e34e1f
NP
2348 if (p->index_version > 1) {
2349 level1_ofs += 2;
2350 index += 8;
2351 }
42873078 2352 index += 4 * 256;
74e34e1f
NP
2353 hi = ntohl(level1_ofs[*sha1]);
2354 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2355 if (p->index_version > 1) {
2356 stride = 20;
2357 } else {
2358 stride = 24;
2359 index += 4;
2360 }
2361
2362 if (debug_lookup)
6e1c2344 2363 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2364 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2365
2366 if (use_lookup < 0)
2367 use_lookup = !!getenv("GIT_USE_LOOKUP");
2368 if (use_lookup) {
2369 int pos = sha1_entry_pos(index, stride, 0,
2370 lo, hi, p->num_objects, sha1);
2371 if (pos < 0)
2372 return 0;
2373 return nth_packed_object_offset(p, pos);
2374 }
1f688557
JH
2375
2376 do {
74e34e1f 2377 unsigned mi = (lo + hi) / 2;
628522ec
JH
2378 int cmp = hashcmp(index + mi * stride, sha1);
2379
2380 if (debug_lookup)
2381 printf("lo %u hi %u rg %u mi %u\n",
2382 lo, hi, hi - lo, mi);
43057304 2383 if (!cmp)
74e34e1f 2384 return nth_packed_object_offset(p, mi);
1f688557
JH
2385 if (cmp > 0)
2386 hi = mi;
2387 else
2388 lo = mi+1;
2389 } while (lo < hi);
2390 return 0;
2391}
2392
4c080182 2393int is_pack_valid(struct packed_git *p)
d131b7af
SP
2394{
2395 /* An already open pack is known to be valid. */
2396 if (p->pack_fd != -1)
2397 return 1;
2398
2399 /* If the pack has one window completely covering the
2400 * file size, the pack is known to be valid even if
2401 * the descriptor is not currently open.
2402 */
2403 if (p->windows) {
2404 struct pack_window *w = p->windows;
2405
2406 if (!w->offset && w->len == p->pack_size)
2407 return 1;
2408 }
2409
2410 /* Force the pack to open to prove its valid. */
2411 return !open_packed_git(p);
2412}
2413
95099731
NTND
2414static int fill_pack_entry(const unsigned char *sha1,
2415 struct pack_entry *e,
2416 struct packed_git *p)
2417{
2418 off_t offset;
2419
2420 if (p->num_bad_objects) {
2421 unsigned i;
2422 for (i = 0; i < p->num_bad_objects; i++)
2423 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2424 return 0;
2425 }
2426
2427 offset = find_pack_entry_one(sha1, p);
2428 if (!offset)
2429 return 0;
2430
2431 /*
2432 * We are about to tell the caller where they can locate the
2433 * requested object. We better make sure the packfile is
2434 * still here and can be accessed before supplying that
2435 * answer, as it may have been deleted since the index was
2436 * loaded!
2437 */
2438 if (!is_pack_valid(p)) {
2439 warning("packfile %s cannot be accessed", p->pack_name);
2440 return 0;
2441 }
2442 e->offset = offset;
2443 e->p = p;
2444 hashcpy(e->sha1, sha1);
2445 return 1;
2446}
2447
d40d535b
MH
2448/*
2449 * Iff a pack file contains the object named by sha1, return true and
2450 * store its location to e.
2451 */
4d6acb70 2452static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2453{
2454 struct packed_git *p;
43057304 2455
1f688557 2456 prepare_packed_git();
f7c22cc6
NP
2457 if (!packed_git)
2458 return 0;
1f688557 2459
c01f51cc
NTND
2460 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2461 return 1;
8eca0b47 2462
c01f51cc 2463 for (p = packed_git; p; p = p->next) {
1b1005d1
MH
2464 if (p == last_found_pack)
2465 continue; /* we already checked this one */
f7c22cc6 2466
1b1005d1
MH
2467 if (fill_pack_entry(sha1, e, p)) {
2468 last_found_pack = p;
2469 return 1;
2470 }
c01f51cc 2471 }
1f688557
JH
2472 return 0;
2473}
2474
a6080a0a 2475struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2476 struct packed_git *packs)
2477{
2478 struct packed_git *p;
bf592c50
DB
2479
2480 for (p = packs; p; p = p->next) {
43057304 2481 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2482 return p;
2483 }
2484 return NULL;
21666f1a 2485
bf592c50
DB
2486}
2487
052fe5ea 2488static int sha1_loose_object_info(const unsigned char *sha1,
23c339c0 2489 struct object_info *oi)
65c2e0c3 2490{
36e4d74a 2491 int status;
65c2e0c3
JH
2492 unsigned long mapsize, size;
2493 void *map;
ef49a7a0 2494 git_zstream stream;
d65a16f6 2495 char hdr[32];
65c2e0c3 2496
5d642e75
JK
2497 if (oi->delta_base_sha1)
2498 hashclr(oi->delta_base_sha1);
2499
052fe5ea
JK
2500 /*
2501 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2502 * need to look inside the object at all. Note that we
2503 * do not optimize out the stat call, even if the
2504 * caller doesn't care about the disk-size, since our
2505 * return value implicitly indicates whether the
2506 * object even exists.
052fe5ea 2507 */
23c339c0 2508 if (!oi->typep && !oi->sizep) {
4ef8d1dd
JH
2509 struct stat st;
2510 if (stat_sha1_file(sha1, &st) < 0)
2511 return -1;
2512 if (oi->disk_sizep)
23c339c0 2513 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2514 return 0;
2515 }
2516
bb6b8e4f 2517 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2518 if (!map)
dbea72a8 2519 return -1;
23c339c0
JK
2520 if (oi->disk_sizep)
2521 *oi->disk_sizep = mapsize;
36e4d74a
JH
2522 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2523 status = error("unable to unpack %s header",
2524 sha1_to_hex(sha1));
21666f1a 2525 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2526 status = error("unable to parse %s header", sha1_to_hex(sha1));
23c339c0
JK
2527 else if (oi->sizep)
2528 *oi->sizep = size;
39c68542 2529 git_inflate_end(&stream);
65c2e0c3 2530 munmap(map, mapsize);
23c339c0
JK
2531 if (oi->typep)
2532 *oi->typep = status;
052fe5ea 2533 return 0;
65c2e0c3
JH
2534}
2535
de7b5d62 2536int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 2537{
c4d9986f 2538 struct cached_object *co;
f0df4ed5 2539 struct pack_entry e;
5b086407 2540 int rtype;
1f7117ef 2541 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
f0df4ed5 2542
1f7117ef 2543 co = find_cached_object(real);
c4d9986f 2544 if (co) {
5b086407
JK
2545 if (oi->typep)
2546 *(oi->typep) = co->type;
9a490590
JH
2547 if (oi->sizep)
2548 *(oi->sizep) = co->size;
161f00e7
JK
2549 if (oi->disk_sizep)
2550 *(oi->disk_sizep) = 0;
5d642e75
JK
2551 if (oi->delta_base_sha1)
2552 hashclr(oi->delta_base_sha1);
9a490590 2553 oi->whence = OI_CACHED;
5b086407 2554 return 0;
c4d9986f
NTND
2555 }
2556
1f7117ef 2557 if (!find_pack_entry(real, &e)) {
ddd63e64 2558 /* Most likely it's a loose object. */
1f7117ef 2559 if (!sha1_loose_object_info(real, oi)) {
9a490590 2560 oi->whence = OI_LOOSE;
5b086407 2561 return 0;
9a490590 2562 }
ddd63e64
SG
2563
2564 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2565 reprepare_packed_git();
1f7117ef 2566 if (!find_pack_entry(real, &e))
052fe5ea 2567 return -1;
f0df4ed5 2568 }
3d77d877 2569
23c339c0 2570 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2571 if (rtype < 0) {
1f7117ef
CC
2572 mark_bad_packed_object(e.p, real);
2573 return sha1_object_info_extended(real, oi, 0);
5266d369
JH
2574 } else if (in_delta_base_cache(e.p, e.offset)) {
2575 oi->whence = OI_DBCACHED;
9a490590
JH
2576 } else {
2577 oi->whence = OI_PACKED;
2578 oi->u.packed.offset = e.offset;
2579 oi->u.packed.pack = e.p;
2580 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2581 rtype == OBJ_OFS_DELTA);
3d77d877
NP
2582 }
2583
5b086407 2584 return 0;
f0df4ed5
JS
2585}
2586
3fc0dca9 2587/* returns enum object_type or negative */
9a490590
JH
2588int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2589{
5b086407 2590 enum object_type type;
d099b717 2591 struct object_info oi = {NULL};
9a490590 2592
5b086407 2593 oi.typep = &type;
9a490590 2594 oi.sizep = sizep;
de7b5d62 2595 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
5b086407
JK
2596 return -1;
2597 return type;
9a490590
JH
2598}
2599
21666f1a
NP
2600static void *read_packed_sha1(const unsigned char *sha1,
2601 enum object_type *type, unsigned long *size)
1f688557
JH
2602{
2603 struct pack_entry e;
8eca0b47 2604 void *data;
1f688557 2605
cd673c1f 2606 if (!find_pack_entry(sha1, &e))
1f688557 2607 return NULL;
8eca0b47
NP
2608 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2609 if (!data) {
2610 /*
2611 * We're probably in deep shit, but let's try to fetch
2612 * the required object anyway from another pack or loose.
2613 * This should happen only in the presence of a corrupted
2614 * pack, and is better than failing outright.
2615 */
2616 error("failed to read object %s at offset %"PRIuMAX" from %s",
2617 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2618 mark_bad_packed_object(e.p, sha1);
ac939109 2619 data = read_object(sha1, type, size);
8eca0b47
NP
2620 }
2621 return data;
1f688557
JH
2622}
2623
21666f1a
NP
2624int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2625 unsigned char *sha1)
d66b37bb
JH
2626{
2627 struct cached_object *co;
2628
21666f1a 2629 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2630 if (has_sha1_file(sha1) || find_cached_object(sha1))
2631 return 0;
c7353967 2632 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
d66b37bb
JH
2633 co = &cached_objects[cached_object_nr++];
2634 co->size = len;
21666f1a 2635 co->type = type;
efa13f7b
JH
2636 co->buf = xmalloc(len);
2637 memcpy(co->buf, buf, len);
d66b37bb
JH
2638 hashcpy(co->sha1, sha1);
2639 return 0;
2640}
2641
c2c5b270
CC
2642static void *read_object(const unsigned char *sha1, enum object_type *type,
2643 unsigned long *size)
0fcfd160
LT
2644{
2645 unsigned long mapsize;
2646 void *map, *buf;
d66b37bb
JH
2647 struct cached_object *co;
2648
2649 co = find_cached_object(sha1);
2650 if (co) {
21666f1a 2651 *type = co->type;
d66b37bb 2652 *size = co->size;
182af834 2653 return xmemdupz(co->buf, co->size);
d66b37bb 2654 }
0fcfd160 2655
8276c007
PE
2656 buf = read_packed_sha1(sha1, type, size);
2657 if (buf)
2658 return buf;
bb6b8e4f 2659 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2660 if (map) {
7efbff75 2661 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2662 munmap(map, mapsize);
2663 return buf;
2664 }
637cdd9d 2665 reprepare_packed_git();
8276c007 2666 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2667}
2668
b6c4cecc
JH
2669/*
2670 * This function dies on corrupt objects; the callers who want to
2671 * deal with them should arrange to call read_object() and give error
2672 * messages themselves.
2673 */
5bf29b95
JH
2674void *read_sha1_file_extended(const unsigned char *sha1,
2675 enum object_type *type,
2676 unsigned long *size,
2677 unsigned flag)
ac939109 2678{
3ba7a065 2679 void *data;
b6c4cecc 2680 const struct packed_git *p;
bf93eea0 2681 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
b6c4cecc 2682
3ba7a065
JH
2683 errno = 0;
2684 data = read_object(repl, type, size);
4bbf5a26 2685 if (data)
b6c4cecc 2686 return data;
68095570 2687
25f3af3f 2688 if (errno && errno != ENOENT)
3ba7a065
JH
2689 die_errno("failed to read object %s", sha1_to_hex(sha1));
2690
68095570 2691 /* die if we replaced an object with one that does not exist */
b6c4cecc 2692 if (repl != sha1)
68095570
CC
2693 die("replacement %s not found for %s",
2694 sha1_to_hex(repl), sha1_to_hex(sha1));
2695
b6c4cecc 2696 if (has_loose_object(repl)) {
30d6c6ea
MH
2697 const char *path = sha1_file_name(sha1);
2698
b6c4cecc
JH
2699 die("loose object %s (stored in %s) is corrupt",
2700 sha1_to_hex(repl), path);
e8b15e61 2701 }
68095570 2702
b6c4cecc
JH
2703 if ((p = has_packed_and_bad(repl)) != NULL)
2704 die("packed object %s (stored in %s) is corrupt",
2705 sha1_to_hex(repl), p->pack_name);
f5552aee 2706
b6c4cecc 2707 return NULL;
ac939109
NP
2708}
2709
40469ee9 2710void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2711 const char *required_type_name,
40469ee9
JH
2712 unsigned long *size,
2713 unsigned char *actual_sha1_return)
f4913f91 2714{
21666f1a 2715 enum object_type type, required_type;
f4913f91
JH
2716 void *buffer;
2717 unsigned long isize;
40469ee9 2718 unsigned char actual_sha1[20];
f4913f91 2719
21666f1a 2720 required_type = type_from_string(required_type_name);
e702496e 2721 hashcpy(actual_sha1, sha1);
40469ee9
JH
2722 while (1) {
2723 int ref_length = -1;
2724 const char *ref_type = NULL;
f4913f91 2725
21666f1a 2726 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2727 if (!buffer)
2728 return NULL;
21666f1a 2729 if (type == required_type) {
40469ee9
JH
2730 *size = isize;
2731 if (actual_sha1_return)
e702496e 2732 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2733 return buffer;
2734 }
2735 /* Handle references */
21666f1a 2736 else if (type == OBJ_COMMIT)
40469ee9 2737 ref_type = "tree ";
21666f1a 2738 else if (type == OBJ_TAG)
40469ee9
JH
2739 ref_type = "object ";
2740 else {
2741 free(buffer);
2742 return NULL;
2743 }
2744 ref_length = strlen(ref_type);
f4913f91 2745
50974ec9
MK
2746 if (ref_length + 40 > isize ||
2747 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2748 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2749 free(buffer);
2750 return NULL;
2751 }
1cf58e72 2752 free(buffer);
40469ee9
JH
2753 /* Now we have the ID of the referred-to object in
2754 * actual_sha1. Check again. */
f4913f91 2755 }
f4913f91
JH
2756}
2757
ce9fbf16 2758static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2759 const char *type, unsigned char *sha1,
d65a16f6 2760 char *hdr, int *hdrlen)
d410c0f5 2761{
9126f009 2762 git_SHA_CTX c;
d410c0f5
JH
2763
2764 /* Generate the header */
d65a16f6 2765 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2766
2767 /* Sha1.. */
9126f009
NP
2768 git_SHA1_Init(&c);
2769 git_SHA1_Update(&c, hdr, *hdrlen);
2770 git_SHA1_Update(&c, buf, len);
2771 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2772}
2773
230f1322 2774/*
5a688fe4
JH
2775 * Move the just written object into its final resting place.
2776 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2777 * "moving" is only a part of what it does, when no patch between
2778 * master to pu changes the call sites of this function.
230f1322 2779 */
839837b9 2780int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2781{
e32c0a9c 2782 int ret = 0;
5a688fe4 2783
348df166 2784 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2785 goto try_rename;
2786 else if (link(tmpfile, filename))
e32c0a9c 2787 ret = errno;
7ebb6fca
LT
2788
2789 /*
2790 * Coda hack - coda doesn't like cross-directory links,
2791 * so we fall back to a rename, which will mean that it
2792 * won't be able to check collisions, but that's not a
2793 * big deal.
2794 *
2795 * The same holds for FAT formatted media.
2796 *
3be1f18e 2797 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2798 * left to unlink.
2799 */
2800 if (ret && ret != EEXIST) {
be66a6c4 2801 try_rename:
7ebb6fca 2802 if (!rename(tmpfile, filename))
3be1f18e 2803 goto out;
9e48b389 2804 ret = errno;
230f1322 2805 }
691f1a28 2806 unlink_or_warn(tmpfile);
230f1322
LT
2807 if (ret) {
2808 if (ret != EEXIST) {
82247e9b 2809 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2810 }
2811 /* FIXME!!! Collision check here ? */
2812 }
2813
3be1f18e 2814out:
5256b006 2815 if (adjust_shared_perm(filename))
5a688fe4 2816 return error("unable to set permission to '%s'", filename);
230f1322
LT
2817 return 0;
2818}
2819
4d548150
LT
2820static int write_buffer(int fd, const void *buf, size_t len)
2821{
d34cf19b 2822 if (write_in_full(fd, buf, len) < 0)
93822c22 2823 return error("file write error (%s)", strerror(errno));
4d548150
LT
2824 return 0;
2825}
2826
ce9fbf16 2827int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2828 unsigned char *sha1)
2829{
d65a16f6 2830 char hdr[32];
abdc3fc8
RS
2831 int hdrlen;
2832 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2833 return 0;
2834}
2835
e9039dd3
LT
2836/* Finalize a file on disk, and close it. */
2837static void close_sha1_file(int fd)
2838{
aafe9fba
LT
2839 if (fsync_object_files)
2840 fsync_or_die(fd, "sha1 file");
e9039dd3 2841 if (close(fd) != 0)
d824cbba 2842 die_errno("error when closing sha1 file");
e9039dd3
LT
2843}
2844
5723fe7e
LT
2845/* Size of directory component, including the ending '/' */
2846static inline int directory_size(const char *filename)
2847{
2848 const char *s = strrchr(filename, '/');
2849 if (!s)
2850 return 0;
2851 return s - filename + 1;
2852}
2853
2854/*
2855 * This creates a temporary file in the same directory as the final
2856 * 'filename'
2857 *
2858 * We want to avoid cross-directory filename renames, because those
2859 * can have problems on various filesystems (FAT, NFS, Coda).
2860 */
2861static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2862{
2863 int fd, dirlen = directory_size(filename);
2864
2865 if (dirlen + 20 > bufsiz) {
2866 errno = ENAMETOOLONG;
2867 return -1;
2868 }
2869 memcpy(buffer, filename, dirlen);
2870 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2871 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2872 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2873 /* Make sure the directory exists */
6ff6af62 2874 memcpy(buffer, filename, dirlen);
5723fe7e 2875 buffer[dirlen-1] = 0;
b2476a60
JH
2876 if (mkdir(buffer, 0777) && errno != EEXIST)
2877 return -1;
2878 if (adjust_shared_perm(buffer))
5723fe7e
LT
2879 return -1;
2880
2881 /* Try again */
2882 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2883 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2884 }
2885 return fd;
2886}
2887
bbac7311 2888static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2889 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2890{
915308b1 2891 int fd, ret;
9892beba 2892 unsigned char compressed[4096];
ef49a7a0 2893 git_zstream stream;
748af44c
NP
2894 git_SHA_CTX c;
2895 unsigned char parano_sha1[20];
ab1900a3 2896 static char tmp_file[PATH_MAX];
30d6c6ea 2897 const char *filename = sha1_file_name(sha1);
a44c9a5e 2898
ab1900a3 2899 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 2900 if (fd < 0) {
35243577 2901 if (errno == EACCES)
82247e9b 2902 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 2903 else
5eaeda70 2904 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
2905 }
2906
0fcfd160
LT
2907 /* Set it up */
2908 memset(&stream, 0, sizeof(stream));
55bb5c91 2909 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 2910 stream.next_out = compressed;
9892beba 2911 stream.avail_out = sizeof(compressed);
748af44c 2912 git_SHA1_Init(&c);
a44c9a5e
LT
2913
2914 /* First header.. */
d65a16f6 2915 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2916 stream.avail_in = hdrlen;
55bb5c91
JH
2917 while (git_deflate(&stream, 0) == Z_OK)
2918 ; /* nothing */
748af44c 2919 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2920
2921 /* Then the data itself.. */
c00e657d 2922 stream.next_in = (void *)buf;
a44c9a5e 2923 stream.avail_in = len;
9892beba 2924 do {
748af44c 2925 unsigned char *in0 = stream.next_in;
55bb5c91 2926 ret = git_deflate(&stream, Z_FINISH);
748af44c 2927 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2928 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2929 die("unable to write sha1 file");
2930 stream.next_out = compressed;
2931 stream.avail_out = sizeof(compressed);
2932 } while (ret == Z_OK);
2933
ac54c277
LT
2934 if (ret != Z_STREAM_END)
2935 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 2936 ret = git_deflate_end_gently(&stream);
ac54c277
LT
2937 if (ret != Z_OK)
2938 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2939 git_SHA1_Final(parano_sha1, &c);
2940 if (hashcmp(sha1, parano_sha1) != 0)
2941 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2942
e9039dd3 2943 close_sha1_file(fd);
0fcfd160 2944
bbac7311
NP
2945 if (mtime) {
2946 struct utimbuf utb;
2947 utb.actime = mtime;
2948 utb.modtime = mtime;
ab1900a3 2949 if (utime(tmp_file, &utb) < 0)
bbac7311 2950 warning("failed utime() on %s: %s",
ab1900a3 2951 tmp_file, strerror(errno));
bbac7311
NP
2952 }
2953
ab1900a3 2954 return move_temp_to_file(tmp_file, filename);
0fcfd160 2955}
8237b185 2956
c00e657d 2957int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2958{
2959 unsigned char sha1[20];
2960 char hdr[32];
2961 int hdrlen;
2962
2963 /* Normally if we have it in the pack then we do not bother writing
2964 * it out into .git/objects/??/?{38} file.
2965 */
2966 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2967 if (returnsha1)
2968 hashcpy(returnsha1, sha1);
2969 if (has_sha1_file(sha1))
2970 return 0;
2971 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2972}
2973
2974int force_object_loose(const unsigned char *sha1, time_t mtime)
2975{
bbac7311
NP
2976 void *buf;
2977 unsigned long len;
2978 enum object_type type;
2979 char hdr[32];
2980 int hdrlen;
1fb23e65 2981 int ret;
bbac7311 2982
c529d75a 2983 if (has_loose_object(sha1))
bbac7311
NP
2984 return 0;
2985 buf = read_packed_sha1(sha1, &type, &len);
2986 if (!buf)
2987 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2988 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2989 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2990 free(buf);
2991
2992 return ret;
bbac7311
NP
2993}
2994
bf592c50
DB
2995int has_pack_index(const unsigned char *sha1)
2996{
2997 struct stat st;
2998 if (stat(sha1_pack_index_name(sha1), &st))
2999 return 0;
3000 return 1;
bf592c50
DB
3001}
3002
cd673c1f
JH
3003int has_sha1_pack(const unsigned char *sha1)
3004{
3005 struct pack_entry e;
3006 return find_pack_entry(sha1, &e);
3007}
3008
8237b185
DB
3009int has_sha1_file(const unsigned char *sha1)
3010{
1f688557
JH
3011 struct pack_entry e;
3012
cd673c1f 3013 if (find_pack_entry(sha1, &e))
1f688557 3014 return 1;
45e8a748
JK
3015 if (has_loose_object(sha1))
3016 return 1;
3017 reprepare_packed_git();
3018 return find_pack_entry(sha1, &e);
8237b185 3019}
74400e71 3020
c879daa2
NTND
3021static void check_tree(const void *buf, size_t size)
3022{
3023 struct tree_desc desc;
3024 struct name_entry entry;
3025
3026 init_tree_desc(&desc, buf, size);
3027 while (tree_entry(&desc, &entry))
3028 /* do nothing
3029 * tree_entry() will die() on malformed entries */
3030 ;
3031}
3032
3033static void check_commit(const void *buf, size_t size)
3034{
3035 struct commit c;
3036 memset(&c, 0, sizeof(c));
3037 if (parse_commit_buffer(&c, buf, size))
3038 die("corrupt commit");
3039}
3040
3041static void check_tag(const void *buf, size_t size)
3042{
3043 struct tag t;
3044 memset(&t, 0, sizeof(t));
3045 if (parse_tag_buffer(&t, buf, size))
3046 die("corrupt tag");
3047}
3048
43df4f86 3049static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3050 enum object_type type,
3051 const char *path, unsigned flags)
e7332f96 3052{
6c510bee 3053 int ret, re_allocated = 0;
c4ce46fc 3054 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3055
7672db20 3056 if (!type)
edaec3fb 3057 type = OBJ_BLOB;
6c510bee
LT
3058
3059 /*
3060 * Convert blobs to git internal format
3061 */
43df4f86 3062 if ((type == OBJ_BLOB) && path) {
f285a2d7 3063 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3064 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3065 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3066 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3067 re_allocated = 1;
3068 }
3069 }
c4ce46fc 3070 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3071 if (type == OBJ_TREE)
3072 check_tree(buf, size);
3073 if (type == OBJ_COMMIT)
3074 check_commit(buf, size);
3075 if (type == OBJ_TAG)
3076 check_tag(buf, size);
3077 }
6c510bee 3078
7672db20 3079 if (write_object)
edaec3fb 3080 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3081 else
edaec3fb 3082 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3083 if (re_allocated)
6c510bee 3084 free(buf);
43df4f86
DP
3085 return ret;
3086}
3087
7b41e1e1
JH
3088static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3089 const char *path, unsigned flags)
3090{
3091 struct strbuf sbuf = STRBUF_INIT;
3092 int ret;
3093
3094 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3095 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3096 else
3097 ret = -1;
3098 strbuf_release(&sbuf);
3099 return ret;
3100}
3101
ea68b0ce
DP
3102#define SMALL_FILE_SIZE (32*1024)
3103
7b41e1e1
JH
3104static int index_core(unsigned char *sha1, int fd, size_t size,
3105 enum object_type type, const char *path,
3106 unsigned flags)
43df4f86
DP
3107{
3108 int ret;
43df4f86 3109
7b41e1e1 3110 if (!size) {
c4ce46fc 3111 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
3112 } else if (size <= SMALL_FILE_SIZE) {
3113 char *buf = xmalloc(size);
3114 if (size == read_in_full(fd, buf, size))
c4ce46fc 3115 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
3116 else
3117 ret = error("short read %s", strerror(errno));
3118 free(buf);
08bda208 3119 } else {
43df4f86 3120 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3121 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3122 munmap(buf, size);
08bda208 3123 }
7b41e1e1
JH
3124 return ret;
3125}
3126
4dd1fbc7 3127/*
568508e7
JH
3128 * This creates one packfile per large blob unless bulk-checkin
3129 * machinery is "plugged".
4dd1fbc7
JH
3130 *
3131 * This also bypasses the usual "convert-to-git" dance, and that is on
3132 * purpose. We could write a streaming version of the converting
3133 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3134 * (or equivalent in-core API described above). However, that is
3135 * somewhat complicated, as we do not know the size of the filter
3136 * result, which we need to know beforehand when writing a git object.
3137 * Since the primary motivation for trying to stream from the working
3138 * tree file and to avoid mmaping it in core is to deal with large
3139 * binary blobs, they generally do not want to get any conversion, and
3140 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3141 */
3142static int index_stream(unsigned char *sha1, int fd, size_t size,
3143 enum object_type type, const char *path,
3144 unsigned flags)
3145{
568508e7 3146 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3147}
3148
7b41e1e1
JH
3149int index_fd(unsigned char *sha1, int fd, struct stat *st,
3150 enum object_type type, const char *path, unsigned flags)
3151{
3152 int ret;
3153 size_t size = xsize_t(st->st_size);
3154
3155 if (!S_ISREG(st->st_mode))
3156 ret = index_pipe(sha1, fd, type, path, flags);
4f22b101
JK
3157 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3158 (path && would_convert_to_git(path, NULL, 0, 0)))
7b41e1e1 3159 ret = index_core(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3160 else
3161 ret = index_stream(sha1, fd, size, type, path, flags);
43df4f86 3162 close(fd);
aac17941 3163 return ret;
74400e71 3164}
ec1fcc16 3165
c4ce46fc 3166int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3167{
3168 int fd;
b760d3aa 3169 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3170
3171 switch (st->st_mode & S_IFMT) {
3172 case S_IFREG:
3173 fd = open(path, O_RDONLY);
3174 if (fd < 0)
3175 return error("open(\"%s\"): %s", path,
3176 strerror(errno));
c4ce46fc 3177 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3178 return error("%s: failed to insert into database",
3179 path);
3180 break;
3181 case S_IFLNK:
b760d3aa 3182 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 3183 char *errstr = strerror(errno);
ec1fcc16
JH
3184 return error("readlink(\"%s\"): %s", path,
3185 errstr);
3186 }
c4ce46fc 3187 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3188 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3189 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3190 return error("%s: failed to insert into database",
3191 path);
b760d3aa 3192 strbuf_release(&sb);
ec1fcc16 3193 break;
f35a6d3b
LT
3194 case S_IFDIR:
3195 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3196 default:
3197 return error("%s: unsupported file type", path);
3198 }
3199 return 0;
3200}
a69e5429
JH
3201
3202int read_pack_header(int fd, struct pack_header *header)
3203{
c697ad14
HO
3204 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3205 /* "eof before pack header was fully read" */
3206 return PH_ERROR_EOF;
3207
a69e5429
JH
3208 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3209 /* "protocol error (pack signature mismatch detected)" */
3210 return PH_ERROR_PACK_SIGNATURE;
3211 if (!pack_version_ok(header->hdr_version))
3212 /* "protocol error (pack version unsupported)" */
3213 return PH_ERROR_PROTOCOL;
3214 return 0;
3215}
40d52ff7
JK
3216
3217void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3218{
3219 enum object_type type = sha1_object_info(sha1, NULL);
3220 if (type < 0)
3221 die("%s is not a valid object", sha1_to_hex(sha1));
3222 if (type != expect)
3223 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3224 typename(expect));
3225}