find_pack_entry(): document last_found_pack
[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
0fcfd160
LT
187/*
188 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 189 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 190 * filename.
ace1534d
JH
191 *
192 * Also note that this returns the location for creating. Reading
193 * SHA1 file can happen from any alternate directory listed in the
d19938ab 194 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 195 * the primary object database.
0fcfd160
LT
196 */
197char *sha1_file_name(const unsigned char *sha1)
198{
560fb6a1
JK
199 static char buf[PATH_MAX];
200 const char *objdir;
201 int len;
0fcfd160 202
560fb6a1
JK
203 objdir = get_object_directory();
204 len = strlen(objdir);
205
206 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
207 if (len + 43 > PATH_MAX)
208 die("insanely long object directory %s", objdir);
209 memcpy(buf, objdir, len);
210 buf[len] = '/';
211 buf[len+3] = '/';
212 buf[len+42] = '\0';
213 fill_sha1_path(buf + len + 1, sha1);
214 return buf;
0fcfd160
LT
215}
216
633f43e1 217static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 218 char **name, char **base, const char *which)
bf592c50
DB
219{
220 static const char hex[] = "0123456789abcdef";
633f43e1 221 char *buf;
bf592c50
DB
222 int i;
223
633f43e1 224 if (!*base) {
bf592c50
DB
225 const char *sha1_file_directory = get_object_directory();
226 int len = strlen(sha1_file_directory);
633f43e1 227 *base = xmalloc(len + 60);
6eec46bd
JH
228 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
229 sha1_file_directory, which);
633f43e1 230 *name = *base + len + 11;
bf592c50
DB
231 }
232
633f43e1 233 buf = *name;
bf592c50
DB
234
235 for (i = 0; i < 20; i++) {
236 unsigned int val = *sha1++;
237 *buf++ = hex[val >> 4];
238 *buf++ = hex[val & 0xf];
239 }
a6080a0a 240
633f43e1 241 return *base;
bf592c50
DB
242}
243
633f43e1 244char *sha1_pack_name(const unsigned char *sha1)
bf592c50 245{
633f43e1 246 static char *name, *base;
bf592c50 247
6eec46bd 248 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 249}
bf592c50 250
633f43e1
HO
251char *sha1_pack_index_name(const unsigned char *sha1)
252{
253 static char *name, *base;
a6080a0a 254
6eec46bd 255 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
256}
257
d5a63b99
JH
258struct alternate_object_database *alt_odb_list;
259static struct alternate_object_database **alt_odb_tail;
ace1534d 260
f4e51683 261static int git_open_noatime(const char *name);
c2f493a4 262
ddd5d056
JH
263/*
264 * Prepare alternate object database registry.
d5a63b99
JH
265 *
266 * The variable alt_odb_list points at the list of struct
267 * alternate_object_database. The elements on this list come from
268 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
269 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
270 * whose contents is similar to that environment variable but can be
271 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
272 * contains "/the/directory/corresponding/to/.git/objects/...", while
273 * its name points just after the slash at the end of ".git/objects/"
274 * in the example above, and has enough space to hold 40-byte hex
275 * SHA1, an extra slash for the first level indirection, and the
276 * terminating NUL.
ddd5d056 277 */
6eac50d8 278static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
ace1534d 279{
1494e038 280 const char *objdir = get_object_directory();
c2f493a4
MW
281 struct alternate_object_database *ent;
282 struct alternate_object_database *alt;
5bdf0a84
HW
283 int pfxlen, entlen;
284 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 285
85dadc38 286 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
287 strbuf_addstr(&pathbuf, real_path(relative_base));
288 strbuf_addch(&pathbuf, '/');
c2f493a4 289 }
6eac50d8 290 strbuf_addstr(&pathbuf, entry);
c2f493a4 291
5bdf0a84
HW
292 normalize_path_copy(pathbuf.buf, pathbuf.buf);
293
294 pfxlen = strlen(pathbuf.buf);
295
296 /*
297 * The trailing slash after the directory name is given by
298 * this function at the end. Remove duplicates.
299 */
300 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
301 pfxlen -= 1;
302
303 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
304 ent = xmalloc(sizeof(*ent) + entlen);
305 memcpy(ent->base, pathbuf.buf, pfxlen);
306 strbuf_release(&pathbuf);
c2f493a4
MW
307
308 ent->name = ent->base + pfxlen + 1;
309 ent->base[pfxlen + 3] = '/';
310 ent->base[pfxlen] = ent->base[entlen-1] = 0;
311
312 /* Detect cases where alternate disappeared */
90b4a71c 313 if (!is_directory(ent->base)) {
c2f493a4
MW
314 error("object directory %s does not exist; "
315 "check .git/objects/info/alternates.",
316 ent->base);
317 free(ent);
318 return -1;
319 }
320
321 /* Prevent the common mistake of listing the same
322 * thing twice, or object directory itself.
323 */
324 for (alt = alt_odb_list; alt; alt = alt->next) {
325 if (!memcmp(ent->base, alt->base, pfxlen)) {
326 free(ent);
327 return -1;
328 }
329 }
cb2912c3 330 if (!strcmp(ent->base, objdir)) {
c2f493a4
MW
331 free(ent);
332 return -1;
333 }
334
335 /* add the alternate entry */
336 *alt_odb_tail = ent;
337 alt_odb_tail = &(ent->next);
338 ent->next = NULL;
339
340 /* recursively add alternates */
341 read_info_alternates(ent->base, depth + 1);
342
343 ent->base[pfxlen] = '/';
344
345 return 0;
346}
347
c5950164 348static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
349 const char *relative_base, int depth)
350{
6eac50d8
MH
351 struct string_list entries = STRING_LIST_INIT_NODUP;
352 char *alt_copy;
353 int i;
c2f493a4
MW
354
355 if (depth > 5) {
356 error("%s: ignoring alternate object stores, nesting too deep.",
357 relative_base);
358 return;
359 }
360
c5950164 361 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
362 string_list_split_in_place(&entries, alt_copy, sep, -1);
363 for (i = 0; i < entries.nr; i++) {
364 const char *entry = entries.items[i].string;
365 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 366 continue;
6eac50d8
MH
367 if (!is_absolute_path(entry) && depth) {
368 error("%s: ignoring relative alternate object store %s",
369 relative_base, entry);
370 } else {
371 link_alt_odb_entry(entry, relative_base, depth);
9577e7e3 372 }
9577e7e3 373 }
6eac50d8
MH
374 string_list_clear(&entries, 0);
375 free(alt_copy);
d5a63b99
JH
376}
377
5e73633d 378void read_info_alternates(const char * relative_base, int depth)
d5a63b99 379{
9577e7e3 380 char *map;
dc49cd76 381 size_t mapsz;
d5a63b99 382 struct stat st;
9cb18f56
JM
383 const char alt_file_name[] = "info/alternates";
384 /* Given that relative_base is no longer than PATH_MAX,
385 ensure that "path" has enough space to append "/", the
386 file name, "info/alternates", and a trailing NUL. */
387 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 388 int fd;
d5a63b99 389
9cb18f56 390 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 391 fd = git_open_noatime(path);
d5a63b99
JH
392 if (fd < 0)
393 return;
394 if (fstat(fd, &st) || (st.st_size == 0)) {
395 close(fd);
9a217f2a 396 return;
ace1534d 397 }
dc49cd76
SP
398 mapsz = xsize_t(st.st_size);
399 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 400 close(fd);
d5a63b99 401
c5950164 402 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 403
dc49cd76 404 munmap(map, mapsz);
ace1534d
JH
405}
406
bef70b22
DB
407void add_to_alternates_file(const char *reference)
408{
409 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 410 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
e6baf4a1 411 char *alt = mkpath("%s\n", reference);
bef70b22
DB
412 write_or_die(fd, alt, strlen(alt));
413 if (commit_lock_file(lock))
414 die("could not close alternates file");
415 if (alt_odb_tail)
c5950164 416 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
bef70b22
DB
417}
418
d79796bc
JH
419void foreach_alt_odb(alt_odb_fn fn, void *cb)
420{
421 struct alternate_object_database *ent;
422
423 prepare_alt_odb();
424 for (ent = alt_odb_list; ent; ent = ent->next)
425 if (fn(ent, cb))
426 return;
427}
428
c2f493a4
MW
429void prepare_alt_odb(void)
430{
554fe20d 431 const char *alt;
c2f493a4 432
7dc24aa5
SP
433 if (alt_odb_tail)
434 return;
435
c2f493a4
MW
436 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
437 if (!alt) alt = "";
438
c2f493a4 439 alt_odb_tail = &alt_odb_list;
c5950164 440 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
441
442 read_info_alternates(get_object_directory(), 0);
443}
444
0f4dc14a 445static int has_loose_object_local(const unsigned char *sha1)
ace1534d 446{
ace1534d 447 char *name = sha1_file_name(sha1);
0f4dc14a
BC
448 return !access(name, F_OK);
449}
ace1534d 450
0f4dc14a
BC
451int has_loose_object_nonlocal(const unsigned char *sha1)
452{
453 struct alternate_object_database *alt;
9a217f2a 454 prepare_alt_odb();
d5a63b99 455 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 456 fill_sha1_path(alt->name, sha1);
c529d75a
LT
457 if (!access(alt->base, F_OK))
458 return 1;
ace1534d 459 }
c529d75a 460 return 0;
ace1534d
JH
461}
462
0f4dc14a
BC
463static int has_loose_object(const unsigned char *sha1)
464{
465 return has_loose_object_local(sha1) ||
466 has_loose_object_nonlocal(sha1);
467}
468
60bb8b14 469static unsigned int pack_used_ctr;
a53128b6
SP
470static unsigned int pack_mmap_calls;
471static unsigned int peak_pack_open_windows;
472static unsigned int pack_open_windows;
c7934306
SP
473static unsigned int pack_open_fds;
474static unsigned int pack_max_fds;
a53128b6 475static size_t peak_pack_mapped;
60bb8b14 476static size_t pack_mapped;
9a217f2a 477struct packed_git *packed_git;
1f688557 478
b79d18c9 479void pack_report(void)
a53128b6
SP
480{
481 fprintf(stderr,
e05db0fd
PR
482 "pack_report: getpagesize() = %10" SZ_FMT "\n"
483 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
484 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
485 sz_fmt(getpagesize()),
486 sz_fmt(packed_git_window_size),
487 sz_fmt(packed_git_limit));
a53128b6
SP
488 fprintf(stderr,
489 "pack_report: pack_used_ctr = %10u\n"
490 "pack_report: pack_mmap_calls = %10u\n"
491 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
492 "pack_report: pack_mapped = "
493 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
494 pack_used_ctr,
495 pack_mmap_calls,
496 pack_open_windows, peak_pack_open_windows,
9e42d6a1 497 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
498}
499
42873078 500static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
501{
502 void *idx_map;
42873078 503 struct pack_idx_header *hdr;
dc49cd76 504 size_t idx_size;
74e34e1f 505 uint32_t version, nr, i, *index;
f4e51683 506 int fd = git_open_noatime(path);
1f688557 507 struct stat st;
42873078 508
1f688557
JH
509 if (fd < 0)
510 return -1;
511 if (fstat(fd, &st)) {
512 close(fd);
513 return -1;
514 }
dc49cd76 515 idx_size = xsize_t(st.st_size);
2d88451b
SP
516 if (idx_size < 4 * 256 + 20 + 20) {
517 close(fd);
518 return error("index file %s is too small", path);
519 }
c4712e45 520 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 521 close(fd);
1f688557 522
42873078
NP
523 hdr = idx_map;
524 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
525 version = ntohl(hdr->idx_version);
526 if (version < 2 || version > 2) {
527 munmap(idx_map, idx_size);
6e1c2344 528 return error("index file %s is version %"PRIu32
74e34e1f
NP
529 " and is not supported by this binary"
530 " (try upgrading GIT to a newer version)",
531 path, version);
532 }
533 } else
534 version = 1;
df1b059d 535
1f688557 536 nr = 0;
42873078 537 index = idx_map;
74e34e1f
NP
538 if (version > 1)
539 index += 2; /* skip index header */
1f688557 540 for (i = 0; i < 256; i++) {
326bf396 541 uint32_t n = ntohl(index[i]);
2d88451b
SP
542 if (n < nr) {
543 munmap(idx_map, idx_size);
df1b059d 544 return error("non-monotonic index %s", path);
2d88451b 545 }
1f688557
JH
546 nr = n;
547 }
548
74e34e1f
NP
549 if (version == 1) {
550 /*
551 * Total size:
552 * - 256 index entries 4 bytes each
553 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
554 * - 20-byte SHA1 of the packfile
555 * - 20-byte SHA1 file checksum
556 */
557 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
558 munmap(idx_map, idx_size);
eef427a0 559 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
560 }
561 } else if (version == 2) {
562 /*
563 * Minimum size:
564 * - 8 bytes of header
565 * - 256 index entries 4 bytes each
566 * - 20-byte sha1 entry * nr
567 * - 4-byte crc entry * nr
568 * - 4-byte offset entry * nr
569 * - 20-byte SHA1 of the packfile
570 * - 20-byte SHA1 file checksum
571 * And after the 4-byte offset table might be a
572 * variable sized table containing 8-byte entries
573 * for offsets larger than 2^31.
574 */
575 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
576 unsigned long max_size = min_size;
577 if (nr)
578 max_size += (nr - 1)*8;
579 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 580 munmap(idx_map, idx_size);
eef427a0 581 return error("wrong index v2 file size in %s", path);
74e34e1f 582 }
7109c889
JH
583 if (idx_size != min_size &&
584 /*
585 * make sure we can deal with large pack offsets.
586 * 31-bit signed offset won't be enough, neither
587 * 32-bit unsigned one will be.
588 */
589 (sizeof(off_t) <= 4)) {
590 munmap(idx_map, idx_size);
591 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 592 }
2d88451b 593 }
1f688557 594
74e34e1f 595 p->index_version = version;
42873078
NP
596 p->index_data = idx_map;
597 p->index_size = idx_size;
57059091 598 p->num_objects = nr;
1f688557
JH
599 return 0;
600}
601
bc8e478a 602int open_pack_index(struct packed_git *p)
d079837e
SP
603{
604 char *idx_name;
605 int ret;
606
607 if (p->index_data)
608 return 0;
609
610 idx_name = xstrdup(p->pack_name);
611 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
612 ret = check_packed_git_idx(idx_name, p);
613 free(idx_name);
614 return ret;
615}
616
11daf39b
SP
617static void scan_windows(struct packed_git *p,
618 struct packed_git **lru_p,
619 struct pack_window **lru_w,
620 struct pack_window **lru_l)
1f688557 621{
11daf39b
SP
622 struct pack_window *w, *w_l;
623
624 for (w_l = NULL, w = p->windows; w; w = w->next) {
625 if (!w->inuse_cnt) {
626 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
627 *lru_p = p;
628 *lru_w = w;
629 *lru_l = w_l;
54044bf8 630 }
54044bf8 631 }
11daf39b 632 w_l = w;
f9253394 633 }
11daf39b
SP
634}
635
7c3ecb32 636static int unuse_one_window(struct packed_git *current)
11daf39b
SP
637{
638 struct packed_git *p, *lru_p = NULL;
639 struct pack_window *lru_w = NULL, *lru_l = NULL;
640
641 if (current)
642 scan_windows(current, &lru_p, &lru_w, &lru_l);
643 for (p = packed_git; p; p = p->next)
644 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
645 if (lru_p) {
646 munmap(lru_w->base, lru_w->len);
647 pack_mapped -= lru_w->len;
648 if (lru_l)
649 lru_l->next = lru_w->next;
7c3ecb32 650 else
54044bf8 651 lru_p->windows = lru_w->next;
54044bf8 652 free(lru_w);
a53128b6 653 pack_open_windows--;
54044bf8
SP
654 return 1;
655 }
656 return 0;
f9253394
JH
657}
658
7c3ecb32 659void release_pack_memory(size_t need)
97bfeb34
SP
660{
661 size_t cur = pack_mapped;
7c3ecb32 662 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
663 ; /* nothing */
664}
665
58ecbd5e
JN
666void *xmmap(void *start, size_t length,
667 int prot, int flags, int fd, off_t offset)
668{
669 void *ret = mmap(start, length, prot, flags, fd, offset);
670 if (ret == MAP_FAILED) {
671 if (!length)
672 return NULL;
7c3ecb32 673 release_pack_memory(length);
58ecbd5e
JN
674 ret = mmap(start, length, prot, flags, fd, offset);
675 if (ret == MAP_FAILED)
676 die_errno("Out of memory? mmap failed");
677 }
678 return ret;
679}
680
c9ced051
SP
681void close_pack_windows(struct packed_git *p)
682{
683 while (p->windows) {
684 struct pack_window *w = p->windows;
685
686 if (w->inuse_cnt)
687 die("pack '%s' still has open windows to it",
688 p->pack_name);
689 munmap(w->base, w->len);
690 pack_mapped -= w->len;
691 pack_open_windows--;
692 p->windows = w->next;
693 free(w);
694 }
695}
696
88d0db55
BC
697/*
698 * The LRU pack is the one with the oldest MRU window, preferring packs
699 * with no used windows, or the oldest mtime if it has no windows allocated.
700 */
701static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
702{
703 struct pack_window *w, *this_mru_w;
704 int has_windows_inuse = 0;
705
706 /*
707 * Reject this pack if it has windows and the previously selected
708 * one does not. If this pack does not have windows, reject
709 * it if the pack file is newer than the previously selected one.
710 */
711 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
712 return;
713
714 for (w = this_mru_w = p->windows; w; w = w->next) {
715 /*
716 * Reject this pack if any of its windows are in use,
717 * but the previously selected pack did not have any
718 * inuse windows. Otherwise, record that this pack
719 * has windows in use.
720 */
721 if (w->inuse_cnt) {
722 if (*accept_windows_inuse)
723 has_windows_inuse = 1;
724 else
725 return;
726 }
727
728 if (w->last_used > this_mru_w->last_used)
729 this_mru_w = w;
730
731 /*
732 * Reject this pack if it has windows that have been
733 * used more recently than the previously selected pack.
734 * If the previously selected pack had windows inuse and
735 * we have not encountered a window in this pack that is
736 * inuse, skip this check since we prefer a pack with no
737 * inuse windows to one that has inuse windows.
738 */
739 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
740 this_mru_w->last_used > (*mru_w)->last_used)
741 return;
742 }
743
744 /*
745 * Select this pack.
746 */
747 *mru_w = this_mru_w;
748 *lru_p = p;
749 *accept_windows_inuse = has_windows_inuse;
750}
751
752static int close_one_pack(void)
753{
754 struct packed_git *p, *lru_p = NULL;
755 struct pack_window *mru_w = NULL;
756 int accept_windows_inuse = 1;
757
758 for (p = packed_git; p; p = p->next) {
759 if (p->pack_fd == -1)
760 continue;
761 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
762 }
763
764 if (lru_p) {
765 close(lru_p->pack_fd);
766 pack_open_fds--;
767 lru_p->pack_fd = -1;
768 return 1;
769 }
770
771 return 0;
772}
773
03e79c88 774void unuse_pack(struct pack_window **w_cursor)
f9253394 775{
03e79c88
SP
776 struct pack_window *w = *w_cursor;
777 if (w) {
778 w->inuse_cnt--;
779 *w_cursor = NULL;
780 }
1f688557
JH
781}
782
fa5fc15d
SP
783void close_pack_index(struct packed_git *p)
784{
785 if (p->index_data) {
786 munmap((void *)p->index_data, p->index_size);
787 p->index_data = NULL;
788 }
789}
790
c74faea1
NP
791/*
792 * This is used by git-repack in case a newly created pack happens to
793 * contain the same set of objects as an existing one. In that case
794 * the resulting file might be different even if its name would be the
795 * same. It is best to close any reference to the old pack before it is
796 * replaced on disk. Of course no index pointers nor windows for given pack
797 * must subsist at this point. If ever objects from this pack are requested
798 * again, the new version of the pack will be reinitialized through
799 * reprepare_packed_git().
800 */
801void free_pack_by_name(const char *pack_name)
802{
803 struct packed_git *p, **pp = &packed_git;
804
805 while (*pp) {
806 p = *pp;
807 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 808 clear_delta_base_cache();
c74faea1 809 close_pack_windows(p);
c7934306 810 if (p->pack_fd != -1) {
c74faea1 811 close(p->pack_fd);
c7934306
SP
812 pack_open_fds--;
813 }
fa5fc15d 814 close_pack_index(p);
c74faea1
NP
815 free(p->bad_object_sha1);
816 *pp = p->next;
c01f51cc
NTND
817 if (last_found_pack == p)
818 last_found_pack = NULL;
c74faea1
NP
819 free(p);
820 return;
821 }
822 pp = &p->next;
823 }
824}
825
a0788266
JS
826static unsigned int get_max_fd_limit(void)
827{
828#ifdef RLIMIT_NOFILE
491a8dec
JH
829 {
830 struct rlimit lim;
a0788266 831
491a8dec
JH
832 if (!getrlimit(RLIMIT_NOFILE, &lim))
833 return lim.rlim_cur;
834 }
835#endif
836
837#ifdef _SC_OPEN_MAX
838 {
839 long open_max = sysconf(_SC_OPEN_MAX);
840 if (0 < open_max)
841 return open_max;
842 /*
843 * Otherwise, we got -1 for one of the two
844 * reasons:
845 *
846 * (1) sysconf() did not understand _SC_OPEN_MAX
847 * and signaled an error with -1; or
848 * (2) sysconf() said there is no limit.
849 *
850 * We _could_ clear errno before calling sysconf() to
851 * tell these two cases apart and return a huge number
852 * in the latter case to let the caller cap it to a
853 * value that is not so selfish, but letting the
854 * fallback OPEN_MAX codepath take care of these cases
855 * is a lot simpler.
856 */
857 }
858#endif
a0788266 859
491a8dec 860#ifdef OPEN_MAX
a0788266
JS
861 return OPEN_MAX;
862#else
863 return 1; /* see the caller ;-) */
864#endif
865}
866
3cf8b462
SP
867/*
868 * Do not call this directly as this leaks p->pack_fd on error return;
869 * call open_packed_git() instead.
870 */
871static int open_packed_git_1(struct packed_git *p)
1f688557 872{
9bc879c1
SP
873 struct stat st;
874 struct pack_header hdr;
875 unsigned char sha1[20];
876 unsigned char *idx_sha1;
2c039da8 877 long fd_flag;
9bc879c1 878
d079837e
SP
879 if (!p->index_data && open_pack_index(p))
880 return error("packfile %s index unavailable", p->pack_name);
881
c7934306 882 if (!pack_max_fds) {
a0788266 883 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
884
885 /* Save 3 for stdin/stdout/stderr, 22 for work */
886 if (25 < max_fds)
887 pack_max_fds = max_fds - 25;
888 else
889 pack_max_fds = 1;
890 }
891
88d0db55 892 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
893 ; /* nothing */
894
f4e51683 895 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 896 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 897 return -1;
c7934306 898 pack_open_fds++;
9bc879c1
SP
899
900 /* If we created the struct before we had the pack we lack size. */
bf592c50 901 if (!p->pack_size) {
bf592c50 902 if (!S_ISREG(st.st_mode))
072db278 903 return error("packfile %s not a regular file", p->pack_name);
bf592c50 904 p->pack_size = st.st_size;
9bc879c1 905 } else if (p->pack_size != st.st_size)
072db278 906 return error("packfile %s size changed", p->pack_name);
9bc879c1 907
2c039da8
JH
908 /* We leave these file descriptors open with sliding mmap;
909 * there is no point keeping them open across exec(), though.
910 */
911 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
912 if (fd_flag < 0)
072db278 913 return error("cannot determine file descriptor flags");
2c039da8
JH
914 fd_flag |= FD_CLOEXEC;
915 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 916 return error("cannot set FD_CLOEXEC");
2c039da8 917
9bc879c1 918 /* Verify we recognize this pack file format. */
e6e2bd62 919 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 920 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 921 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 922 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 923 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
924 return error("packfile %s is version %"PRIu32" and not"
925 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
926 p->pack_name, ntohl(hdr.hdr_version));
927
928 /* Verify the pack matches its index. */
57059091 929 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
930 return error("packfile %s claims to have %"PRIu32" objects"
931 " while index indicates %"PRIu32" objects",
57059091
NP
932 p->pack_name, ntohl(hdr.hdr_entries),
933 p->num_objects);
9bc879c1 934 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 935 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 936 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 937 return error("packfile %s signature is unavailable", p->pack_name);
42873078 938 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 939 if (hashcmp(sha1, idx_sha1))
072db278
SP
940 return error("packfile %s does not match index", p->pack_name);
941 return 0;
9bc879c1
SP
942}
943
3cf8b462
SP
944static int open_packed_git(struct packed_git *p)
945{
946 if (!open_packed_git_1(p))
947 return 0;
948 if (p->pack_fd != -1) {
949 close(p->pack_fd);
c7934306 950 pack_open_fds--;
3cf8b462
SP
951 p->pack_fd = -1;
952 }
953 return -1;
954}
955
c4001d92 956static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
957{
958 /* We must promise at least 20 bytes (one hash) after the
959 * offset is available from this window, otherwise the offset
960 * is not actually in this window and a different window (which
961 * has that one hash excess) must be used. This is to support
962 * the object header and delta base parsing routines below.
963 */
964 off_t win_off = win->offset;
965 return win_off <= offset
966 && (offset + 20) <= (win_off + win->len);
967}
968
4b25d091 969unsigned char *use_pack(struct packed_git *p,
03e79c88 970 struct pack_window **w_cursor,
c4001d92 971 off_t offset,
ef49a7a0 972 unsigned long *left)
9bc879c1 973{
60bb8b14 974 struct pack_window *win = *w_cursor;
03e79c88 975
a9d98a14 976 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
977 * pointless to ask for an offset into the middle of that
978 * hash, and the in_window function above wouldn't match
979 * don't allow an offset too close to the end of the file.
980 */
d131b7af
SP
981 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
982 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
983 if (offset > (p->pack_size - 20))
984 die("offset beyond end of packfile (truncated pack?)");
985
986 if (!win || !in_window(win, offset)) {
987 if (win)
988 win->inuse_cnt--;
989 for (win = p->windows; win; win = win->next) {
990 if (in_window(win, offset))
991 break;
992 }
993 if (!win) {
78a28df9 994 size_t window_align = packed_git_window_size / 2;
dc49cd76 995 off_t len;
d131b7af
SP
996
997 if (p->pack_fd == -1 && open_packed_git(p))
998 die("packfile %s cannot be accessed", p->pack_name);
999
60bb8b14 1000 win = xcalloc(1, sizeof(*win));
78a28df9 1001 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
1002 len = p->pack_size - win->offset;
1003 if (len > packed_git_window_size)
1004 len = packed_git_window_size;
1005 win->len = (size_t)len;
60bb8b14 1006 pack_mapped += win->len;
11daf39b 1007 while (packed_git_limit < pack_mapped
7c3ecb32 1008 && unuse_one_window(p))
60bb8b14 1009 ; /* nothing */
c4712e45 1010 win->base = xmmap(NULL, win->len,
60bb8b14
SP
1011 PROT_READ, MAP_PRIVATE,
1012 p->pack_fd, win->offset);
1013 if (win->base == MAP_FAILED)
73b4e4be
SP
1014 die("packfile %s cannot be mapped: %s",
1015 p->pack_name,
1016 strerror(errno));
d131b7af
SP
1017 if (!win->offset && win->len == p->pack_size
1018 && !p->do_not_close) {
1019 close(p->pack_fd);
1020 pack_open_fds--;
1021 p->pack_fd = -1;
1022 }
a53128b6
SP
1023 pack_mmap_calls++;
1024 pack_open_windows++;
1025 if (pack_mapped > peak_pack_mapped)
1026 peak_pack_mapped = pack_mapped;
1027 if (pack_open_windows > peak_pack_open_windows)
1028 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
1029 win->next = p->windows;
1030 p->windows = win;
1031 }
1f688557 1032 }
03e79c88
SP
1033 if (win != *w_cursor) {
1034 win->last_used = pack_used_ctr++;
1035 win->inuse_cnt++;
1036 *w_cursor = win;
1037 }
60bb8b14 1038 offset -= win->offset;
03e79c88 1039 if (left)
dc49cd76 1040 *left = win->len - xsize_t(offset);
03e79c88 1041 return win->base + offset;
1f688557
JH
1042}
1043
27d69a46
NP
1044static struct packed_git *alloc_packed_git(int extra)
1045{
1046 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1047 memset(p, 0, sizeof(*p));
1048 p->pack_fd = -1;
1049 return p;
1050}
1051
e0500293
JN
1052static void try_to_free_pack_memory(size_t size)
1053{
7c3ecb32 1054 release_pack_memory(size);
e0500293
JN
1055}
1056
42873078 1057struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 1058{
e0500293 1059 static int have_set_try_to_free_routine;
1f688557 1060 struct stat st;
27d69a46 1061 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 1062
e0500293
JN
1063 if (!have_set_try_to_free_routine) {
1064 have_set_try_to_free_routine = 1;
1065 set_try_to_free_routine(try_to_free_pack_memory);
1066 }
1067
42873078
NP
1068 /*
1069 * Make sure a corresponding .pack file exists and that
1070 * the index looks sane.
1071 */
1072 path_len -= strlen(".idx");
27d69a46
NP
1073 if (path_len < 1) {
1074 free(p);
1f688557 1075 return NULL;
27d69a46 1076 }
42873078 1077 memcpy(p->pack_name, path, path_len);
8d25931d
BC
1078
1079 strcpy(p->pack_name + path_len, ".keep");
1080 if (!access(p->pack_name, F_OK))
1081 p->pack_keep = 1;
1082
42873078 1083 strcpy(p->pack_name + path_len, ".pack");
d079837e 1084 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1085 free(p);
1f688557
JH
1086 return NULL;
1087 }
42873078 1088
1f688557
JH
1089 /* ok, it looks sane as far as we can check without
1090 * actually mapping the pack file.
1091 */
1f688557 1092 p->pack_size = st.st_size;
9d835df2 1093 p->pack_local = local;
b867092f 1094 p->mtime = st.st_mtime;
42873078
NP
1095 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1096 hashclr(p->sha1);
1f688557
JH
1097 return p;
1098}
1099
7b64469a 1100struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1101{
42873078 1102 const char *path = sha1_pack_name(sha1);
27d69a46 1103 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 1104
27d69a46
NP
1105 strcpy(p->pack_name, path);
1106 hashcpy(p->sha1, sha1);
42873078
NP
1107 if (check_packed_git_idx(idx_path, p)) {
1108 free(p);
bf592c50 1109 return NULL;
42873078 1110 }
bf592c50 1111
bf592c50
DB
1112 return p;
1113}
1114
1115void install_packed_git(struct packed_git *pack)
1116{
c7934306
SP
1117 if (pack->pack_fd != -1)
1118 pack_open_fds++;
1119
bf592c50
DB
1120 pack->next = packed_git;
1121 packed_git = pack;
1122}
1123
543c5caa
NTND
1124void (*report_garbage)(const char *desc, const char *path);
1125
1126static void report_helper(const struct string_list *list,
1127 int seen_bits, int first, int last)
1128{
1129 const char *msg;
1130 switch (seen_bits) {
1131 case 0:
1132 msg = "no corresponding .idx nor .pack";
1133 break;
1134 case 1:
1135 msg = "no corresponding .idx";
1136 break;
1137 case 2:
1138 msg = "no corresponding .pack";
1139 break;
1140 default:
1141 return;
1142 }
1143 for (; first < last; first++)
1144 report_garbage(msg, list->items[first].string);
1145}
1146
1147static void report_pack_garbage(struct string_list *list)
1148{
1149 int i, baselen = -1, first = 0, seen_bits = 0;
1150
1151 if (!report_garbage)
1152 return;
1153
1154 sort_string_list(list);
1155
1156 for (i = 0; i < list->nr; i++) {
1157 const char *path = list->items[i].string;
1158 if (baselen != -1 &&
1159 strncmp(path, list->items[first].string, baselen)) {
1160 report_helper(list, seen_bits, first, i);
1161 baselen = -1;
1162 seen_bits = 0;
1163 }
1164 if (baselen == -1) {
1165 const char *dot = strrchr(path, '.');
1166 if (!dot) {
1167 report_garbage("garbage found", path);
1168 continue;
1169 }
1170 baselen = dot - path + 1;
1171 first = i;
1172 }
1173 if (!strcmp(path + baselen, "pack"))
1174 seen_bits |= 1;
1175 else if (!strcmp(path + baselen, "idx"))
1176 seen_bits |= 2;
1177 }
1178 report_helper(list, seen_bits, first, list->nr);
1179}
1180
9d835df2 1181static void prepare_packed_git_one(char *objdir, int local)
1f688557 1182{
9cb18f56
JM
1183 /* Ensure that this buffer is large enough so that we can
1184 append "/pack/" without clobbering the stack even if
1185 strlen(objdir) were PATH_MAX. */
1186 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
1187 int len;
1188 DIR *dir;
1189 struct dirent *de;
543c5caa 1190 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557
JH
1191
1192 sprintf(path, "%s/pack", objdir);
1193 len = strlen(path);
1194 dir = opendir(path);
b5b16990 1195 if (!dir) {
26125f6b 1196 if (errno != ENOENT)
bd2afde8 1197 error("unable to open object pack directory: %s: %s",
26125f6b 1198 path, strerror(errno));
1f688557 1199 return;
b5b16990 1200 }
1f688557
JH
1201 path[len++] = '/';
1202 while ((de = readdir(dir)) != NULL) {
1203 int namelen = strlen(de->d_name);
1204 struct packed_git *p;
1205
543c5caa
NTND
1206 if (len + namelen + 1 > sizeof(path)) {
1207 if (report_garbage) {
1208 struct strbuf sb = STRBUF_INIT;
1209 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1210 report_garbage("path too long", sb.buf);
1211 strbuf_release(&sb);
1212 }
1213 continue;
1214 }
1215
1216 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1217 continue;
1218
1f688557 1219 strcpy(path + len, de->d_name);
d90906a9
NTND
1220
1221 if (has_extension(de->d_name, ".idx")) {
1222 /* Don't reopen a pack we already have. */
1223 for (p = packed_git; p; p = p->next) {
1224 if (!memcmp(path, p->pack_name, len + namelen - 4))
1225 break;
1226 }
1227 if (p == NULL &&
1228 /*
1229 * See if it really is a valid .idx file with
1230 * corresponding .pack file that we can map.
1231 */
1232 (p = add_packed_git(path, len + namelen, local)) != NULL)
1233 install_packed_git(p);
86f7780c 1234 }
543c5caa
NTND
1235
1236 if (!report_garbage)
1237 continue;
1238
1239 if (has_extension(de->d_name, ".idx") ||
1240 has_extension(de->d_name, ".pack") ||
1241 has_extension(de->d_name, ".keep"))
1242 string_list_append(&garbage, path);
1243 else
1244 report_garbage("garbage found", path);
1f688557 1245 }
5b35bcd5 1246 closedir(dir);
543c5caa
NTND
1247 report_pack_garbage(&garbage);
1248 string_list_clear(&garbage, 0);
1f688557
JH
1249}
1250
b867092f
JH
1251static int sort_pack(const void *a_, const void *b_)
1252{
1253 struct packed_git *a = *((struct packed_git **)a_);
1254 struct packed_git *b = *((struct packed_git **)b_);
1255 int st;
1256
1257 /*
1258 * Local packs tend to contain objects specific to our
1259 * variant of the project than remote ones. In addition,
1260 * remote ones could be on a network mounted filesystem.
1261 * Favor local ones for these reasons.
1262 */
1263 st = a->pack_local - b->pack_local;
1264 if (st)
1265 return -st;
1266
1267 /*
1268 * Younger packs tend to contain more recent objects,
1269 * and more recent objects tend to get accessed more
1270 * often.
1271 */
1272 if (a->mtime < b->mtime)
1273 return 1;
1274 else if (a->mtime == b->mtime)
1275 return 0;
1276 return -1;
1277}
1278
1279static void rearrange_packed_git(void)
1280{
1281 struct packed_git **ary, *p;
1282 int i, n;
1283
1284 for (n = 0, p = packed_git; p; p = p->next)
1285 n++;
1286 if (n < 2)
1287 return;
1288
1289 /* prepare an array of packed_git for easier sorting */
1290 ary = xcalloc(n, sizeof(struct packed_git *));
1291 for (n = 0, p = packed_git; p; p = p->next)
1292 ary[n++] = p;
1293
1294 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1295
1296 /* link them back again */
1297 for (i = 0; i < n - 1; i++)
1298 ary[i]->next = ary[i + 1];
1299 ary[n - 1]->next = NULL;
1300 packed_git = ary[0];
1301
1302 free(ary);
1303}
1304
637cdd9d 1305static int prepare_packed_git_run_once = 0;
9a217f2a 1306void prepare_packed_git(void)
1f688557 1307{
d5a63b99 1308 struct alternate_object_database *alt;
1f688557 1309
637cdd9d 1310 if (prepare_packed_git_run_once)
1f688557 1311 return;
9d835df2 1312 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1313 prepare_alt_odb();
d5a63b99 1314 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1315 alt->name[-1] = 0;
9d835df2 1316 prepare_packed_git_one(alt->base, 0);
1494e038 1317 alt->name[-1] = '/';
1f688557 1318 }
b867092f 1319 rearrange_packed_git();
637cdd9d
JK
1320 prepare_packed_git_run_once = 1;
1321}
1322
fc04c412 1323void reprepare_packed_git(void)
637cdd9d 1324{
4b480c67 1325 discard_revindex();
637cdd9d
JK
1326 prepare_packed_git_run_once = 0;
1327 prepare_packed_git();
1f688557
JH
1328}
1329
8eca0b47
NP
1330static void mark_bad_packed_object(struct packed_git *p,
1331 const unsigned char *sha1)
1332{
1333 unsigned i;
1334 for (i = 0; i < p->num_bad_objects; i++)
1335 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1336 return;
1337 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1338 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1339 p->num_bad_objects++;
1340}
1341
b6c4cecc 1342static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1343{
1344 struct packed_git *p;
1345 unsigned i;
1346
1347 for (p = packed_git; p; p = p->next)
1348 for (i = 0; i < p->num_bad_objects; i++)
1349 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1350 return p;
1351 return NULL;
ac939109
NP
1352}
1353
090ea126
NTND
1354/*
1355 * With an in-core object data in "map", rehash it to make sure the
1356 * object name actually matches "sha1" to detect object corruption.
1357 * With "map" == NULL, try reading the object named with "sha1" using
1358 * the streaming interface and rehash it to do the same.
1359 */
1360int check_sha1_signature(const unsigned char *sha1, void *map,
1361 unsigned long size, const char *type)
0fcfd160
LT
1362{
1363 unsigned char real_sha1[20];
090ea126
NTND
1364 enum object_type obj_type;
1365 struct git_istream *st;
1366 git_SHA_CTX c;
1367 char hdr[32];
1368 int hdrlen;
1369
1370 if (map) {
1371 hash_sha1_file(map, size, type, real_sha1);
1372 return hashcmp(sha1, real_sha1) ? -1 : 0;
1373 }
1374
1375 st = open_istream(sha1, &obj_type, &size, NULL);
1376 if (!st)
1377 return -1;
1378
1379 /* Generate the header */
1380 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1381
1382 /* Sha1.. */
1383 git_SHA1_Init(&c);
1384 git_SHA1_Update(&c, hdr, hdrlen);
1385 for (;;) {
1386 char buf[1024 * 16];
1387 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1388
f54fac53
JK
1389 if (readlen < 0) {
1390 close_istream(st);
1391 return -1;
1392 }
090ea126
NTND
1393 if (!readlen)
1394 break;
1395 git_SHA1_Update(&c, buf, readlen);
1396 }
1397 git_SHA1_Final(real_sha1, &c);
1398 close_istream(st);
a89fccd2 1399 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1400}
1401
f4e51683 1402static int git_open_noatime(const char *name)
44d1c19e
LT
1403{
1404 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1405
f2e872aa
SP
1406 for (;;) {
1407 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1408 if (fd >= 0)
f2e872aa
SP
1409 return fd;
1410
f2e872aa
SP
1411 /* Might the failure be due to O_NOATIME? */
1412 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1413 sha1_file_open_flag = 0;
f2e872aa
SP
1414 continue;
1415 }
1416
1417 return -1;
44d1c19e 1418 }
44d1c19e
LT
1419}
1420
052fe5ea
JK
1421static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1422{
1423 char *name = sha1_file_name(sha1);
1424 struct alternate_object_database *alt;
1425
1426 if (!lstat(name, st))
1427 return 0;
1428
1429 prepare_alt_odb();
1430 errno = ENOENT;
1431 for (alt = alt_odb_list; alt; alt = alt->next) {
1432 name = alt->name;
1433 fill_sha1_path(name, sha1);
1434 if (!lstat(alt->base, st))
1435 return 0;
1436 }
1437
1438 return -1;
1439}
1440
44d1c19e
LT
1441static int open_sha1_file(const unsigned char *sha1)
1442{
1443 int fd;
1444 char *name = sha1_file_name(sha1);
1445 struct alternate_object_database *alt;
1446
f4e51683 1447 fd = git_open_noatime(name);
44d1c19e
LT
1448 if (fd >= 0)
1449 return fd;
1450
1451 prepare_alt_odb();
1452 errno = ENOENT;
1453 for (alt = alt_odb_list; alt; alt = alt->next) {
1454 name = alt->name;
1455 fill_sha1_path(name, sha1);
f4e51683 1456 fd = git_open_noatime(alt->base);
44d1c19e
LT
1457 if (fd >= 0)
1458 return fd;
1459 }
1460 return -1;
1461}
1462
f0270efd 1463void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1464{
0fcfd160 1465 void *map;
144bde78 1466 int fd;
ace1534d 1467
44d1c19e
LT
1468 fd = open_sha1_file(sha1);
1469 map = NULL;
1470 if (fd >= 0) {
1471 struct stat st;
0fcfd160 1472
44d1c19e
LT
1473 if (!fstat(fd, &st)) {
1474 *size = xsize_t(st.st_size);
33e42de0
MM
1475 if (!*size) {
1476 /* mmap() is forbidden on empty files */
1477 error("object file %s is empty", sha1_file_name(sha1));
1478 return NULL;
1479 }
44d1c19e 1480 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1481 }
44d1c19e 1482 close(fd);
0fcfd160 1483 }
0fcfd160
LT
1484 return map;
1485}
1486
09ded04b
NP
1487unsigned long unpack_object_header_buffer(const unsigned char *buf,
1488 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1489{
ad1ed5ee 1490 unsigned shift;
48fb7deb 1491 unsigned long size, c;
ad1ed5ee
JH
1492 unsigned long used = 0;
1493
1494 c = buf[used++];
1495 *type = (c >> 4) & 7;
1496 size = c & 15;
1497 shift = 4;
1498 while (c & 0x80) {
f630cfda 1499 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1500 error("bad object header");
ea4f9685
JH
1501 size = used = 0;
1502 break;
09ded04b 1503 }
ad1ed5ee
JH
1504 c = buf[used++];
1505 size += (c & 0x7f) << shift;
1506 shift += 7;
1507 }
1508 *sizep = size;
1509 return used;
1510}
1511
eb4f4076 1512int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee 1513{
c4483576
LT
1514 /* Get the data stream */
1515 memset(stream, 0, sizeof(*stream));
1516 stream->next_in = map;
1517 stream->avail_in = mapsize;
1518 stream->next_out = buffer;
93821bd9
LT
1519 stream->avail_out = bufsiz;
1520
39c68542 1521 git_inflate_init(stream);
cc5c54e7 1522 return git_inflate(stream, 0);
c4483576
LT
1523}
1524
ef49a7a0 1525static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1526{
1527 int bytes = strlen(buffer) + 1;
3aee68aa 1528 unsigned char *buf = xmallocz(size);
93821bd9 1529 unsigned long n;
7efbff75 1530 int status = Z_OK;
5180cacc 1531
93821bd9
LT
1532 n = stream->total_out - bytes;
1533 if (n > size)
1534 n = size;
1535 memcpy(buf, (char *) buffer + bytes, n);
1536 bytes = n;
456cdf6e
LT
1537 if (bytes <= size) {
1538 /*
1539 * The above condition must be (bytes <= size), not
1540 * (bytes < size). In other words, even though we
ccf5ace0 1541 * expect no more output and set avail_out to zero,
456cdf6e
LT
1542 * the input zlib stream may have bytes that express
1543 * "this concludes the stream", and we *do* want to
1544 * eat that input.
1545 *
1546 * Otherwise we would not be able to test that we
1547 * consumed all the input to reach the expected size;
1548 * we also want to check that zlib tells us that all
1549 * went well with status == Z_STREAM_END at the end.
1550 */
5180cacc
LT
1551 stream->next_out = buf + bytes;
1552 stream->avail_out = size - bytes;
7efbff75 1553 while (status == Z_OK)
39c68542 1554 status = git_inflate(stream, Z_FINISH);
5180cacc 1555 }
456cdf6e 1556 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1557 git_inflate_end(stream);
7efbff75
JH
1558 return buf;
1559 }
1560
1561 if (status < 0)
1562 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1563 else if (stream->avail_in)
1564 error("garbage at end of loose object '%s'",
1565 sha1_to_hex(sha1));
1566 free(buf);
1567 return NULL;
5180cacc
LT
1568}
1569
1570/*
1571 * We used to just use "sscanf()", but that's actually way
1572 * too permissive for what we want to check. So do an anal
1573 * object header parse by hand.
1574 */
f0270efd 1575int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1576{
21666f1a 1577 char type[10];
5180cacc
LT
1578 int i;
1579 unsigned long size;
1580
1581 /*
a6080a0a 1582 * The type can be at most ten bytes (including the
5180cacc 1583 * terminating '\0' that we add), and is followed by
21666f1a 1584 * a space.
5180cacc 1585 */
21666f1a 1586 i = 0;
5180cacc
LT
1587 for (;;) {
1588 char c = *hdr++;
1589 if (c == ' ')
1590 break;
21666f1a
NP
1591 type[i++] = c;
1592 if (i >= sizeof(type))
5180cacc 1593 return -1;
5180cacc 1594 }
21666f1a 1595 type[i] = 0;
5180cacc
LT
1596
1597 /*
1598 * The length must follow immediately, and be in canonical
1599 * decimal format (ie "010" is not valid).
1600 */
1601 size = *hdr++ - '0';
1602 if (size > 9)
1603 return -1;
1604 if (size) {
1605 for (;;) {
1606 unsigned long c = *hdr - '0';
1607 if (c > 9)
1608 break;
1609 hdr++;
1610 size = size * 10 + c;
1611 }
1612 }
1613 *sizep = size;
1614
1615 /*
1616 * The length must be followed by a zero byte
1617 */
21666f1a 1618 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1619}
1620
7efbff75 1621static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1622{
5180cacc 1623 int ret;
ef49a7a0 1624 git_zstream stream;
5180cacc 1625 char hdr[8192];
0fcfd160 1626
5180cacc 1627 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1628 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1629 return NULL;
1630
7efbff75 1631 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1632}
1633
54dab52a
NP
1634unsigned long get_size_from_delta(struct packed_git *p,
1635 struct pack_window **w_curs,
1636 off_t curpos)
1637{
1638 const unsigned char *data;
1639 unsigned char delta_head[20], *in;
ef49a7a0 1640 git_zstream stream;
54dab52a
NP
1641 int st;
1642
1643 memset(&stream, 0, sizeof(stream));
1644 stream.next_out = delta_head;
1645 stream.avail_out = sizeof(delta_head);
1646
39c68542 1647 git_inflate_init(&stream);
54dab52a
NP
1648 do {
1649 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1650 stream.next_in = in;
39c68542 1651 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1652 curpos += stream.next_in - in;
1653 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1654 stream.total_out < sizeof(delta_head));
39c68542 1655 git_inflate_end(&stream);
3d77d877
NP
1656 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1657 error("delta data unpack-initial failed");
1658 return 0;
1659 }
54dab52a
NP
1660
1661 /* Examine the initial part of the delta to figure out
1662 * the result size.
1663 */
1664 data = delta_head;
1665
1666 /* ignore base size */
1667 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1668
1669 /* Read the result size */
1670 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1671}
1672
c4001d92 1673static off_t get_delta_base(struct packed_git *p,
03e79c88 1674 struct pack_window **w_curs,
c4001d92 1675 off_t *curpos,
21666f1a 1676 enum object_type type,
c4001d92 1677 off_t delta_obj_offset)
eb32d236 1678{
2b87c45b 1679 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1680 off_t base_offset;
eb32d236 1681
8d8a4ea5
SP
1682 /* use_pack() assured us we have [base_info, base_info + 20)
1683 * as a range that we can look at without walking off the
1684 * end of the mapped window. Its actually the hash size
1685 * that is assured. An OFS_DELTA longer than the hash size
1686 * is stupid, as then a REF_DELTA would be smaller to store.
1687 */
21666f1a 1688 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1689 unsigned used = 0;
1690 unsigned char c = base_info[used++];
1691 base_offset = c & 127;
1692 while (c & 128) {
1693 base_offset += 1;
8723f216 1694 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1695 return 0; /* overflow */
eb32d236
NP
1696 c = base_info[used++];
1697 base_offset = (base_offset << 7) + (c & 127);
1698 }
1699 base_offset = delta_obj_offset - base_offset;
d8f32556 1700 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1701 return 0; /* out of bound */
2b87c45b 1702 *curpos += used;
21666f1a 1703 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1704 /* The base entry _must_ be in the same pack */
1705 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1706 *curpos += 20;
eb32d236
NP
1707 } else
1708 die("I am totally screwed");
2b87c45b 1709 return base_offset;
eb32d236
NP
1710}
1711
5d642e75
JK
1712/*
1713 * Like get_delta_base above, but we return the sha1 instead of the pack
1714 * offset. This means it is cheaper for REF deltas (we do not have to do
1715 * the final object lookup), but more expensive for OFS deltas (we
1716 * have to load the revidx to convert the offset back into a sha1).
1717 */
1718static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1719 struct pack_window **w_curs,
1720 off_t curpos,
1721 enum object_type type,
1722 off_t delta_obj_offset)
1723{
1724 if (type == OBJ_REF_DELTA) {
1725 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1726 return base;
1727 } else if (type == OBJ_OFS_DELTA) {
1728 struct revindex_entry *revidx;
1729 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1730 type, delta_obj_offset);
1731
1732 if (!base_offset)
1733 return NULL;
1734
1735 revidx = find_pack_revindex(p, base_offset);
1736 if (!revidx)
1737 return NULL;
1738
1739 return nth_packed_object_sha1(p, revidx->nr);
1740 } else
1741 return NULL;
1742}
1743
f8c8abc5
JH
1744int unpack_object_header(struct packed_git *p,
1745 struct pack_window **w_curs,
1746 off_t *curpos,
1747 unsigned long *sizep)
a733cb60 1748{
03e79c88 1749 unsigned char *base;
ef49a7a0 1750 unsigned long left;
ad1ed5ee 1751 unsigned long used;
2b87c45b 1752 enum object_type type;
a733cb60 1753
8d8a4ea5 1754 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1755 * as a range that we can look at. (Its actually the hash
3dff5379 1756 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1757 * the maximum deflated object size is 2^137, which is just
1758 * insane, so we know won't exceed what we have been given.
1759 */
2b87c45b 1760 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1761 used = unpack_object_header_buffer(base, left, &type, sizep);
1762 if (!used) {
1763 type = OBJ_BAD;
1764 } else
1765 *curpos += used;
ad1ed5ee 1766
2b87c45b 1767 return type;
a733cb60
LT
1768}
1769
790d96c0
TR
1770static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1771{
1772 int type;
1773 struct revindex_entry *revidx;
1774 const unsigned char *sha1;
1775 revidx = find_pack_revindex(p, obj_offset);
1776 if (!revidx)
1777 return OBJ_BAD;
1778 sha1 = nth_packed_object_sha1(p, revidx->nr);
1779 mark_bad_packed_object(p, sha1);
1780 type = sha1_object_info(sha1, NULL);
1781 if (type <= OBJ_NONE)
1782 return OBJ_BAD;
1783 return type;
1784}
1785
790d96c0
TR
1786#define POI_STACK_PREALLOC 64
1787
90191d37
JK
1788static enum object_type packed_to_object_type(struct packed_git *p,
1789 off_t obj_offset,
1790 enum object_type type,
1791 struct pack_window **w_curs,
1792 off_t curpos)
1f688557 1793{
790d96c0
TR
1794 off_t small_poi_stack[POI_STACK_PREALLOC];
1795 off_t *poi_stack = small_poi_stack;
1796 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1797
790d96c0
TR
1798 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1799 off_t base_offset;
90191d37 1800 unsigned long size;
790d96c0
TR
1801 /* Push the object we're going to leave behind */
1802 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1803 poi_stack_alloc = alloc_nr(poi_stack_nr);
1804 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1805 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1806 } else {
1807 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1808 }
1809 poi_stack[poi_stack_nr++] = obj_offset;
1810 /* If parsing the base offset fails, just unwind */
90191d37 1811 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1812 if (!base_offset)
1813 goto unwind;
1814 curpos = obj_offset = base_offset;
90191d37 1815 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1816 if (type <= OBJ_NONE) {
1817 /* If getting the base itself fails, we first
1818 * retry the base, otherwise unwind */
1819 type = retry_bad_packed_offset(p, base_offset);
1820 if (type > OBJ_NONE)
1821 goto out;
1822 goto unwind;
1823 }
1824 }
1825
21666f1a 1826 switch (type) {
790d96c0 1827 case OBJ_BAD:
a733cb60 1828 case OBJ_COMMIT:
a733cb60 1829 case OBJ_TREE:
a733cb60 1830 case OBJ_BLOB:
a733cb60 1831 case OBJ_TAG:
a69d0943 1832 break;
1f688557 1833 default:
3d77d877
NP
1834 error("unknown object type %i at offset %"PRIuMAX" in %s",
1835 type, (uintmax_t)obj_offset, p->pack_name);
1836 type = OBJ_BAD;
1f688557 1837 }
790d96c0
TR
1838
1839out:
1840 if (poi_stack != small_poi_stack)
1841 free(poi_stack);
21666f1a 1842 return type;
790d96c0
TR
1843
1844unwind:
1845 while (poi_stack_nr) {
1846 obj_offset = poi_stack[--poi_stack_nr];
1847 type = retry_bad_packed_offset(p, obj_offset);
1848 if (type > OBJ_NONE)
1849 goto out;
1850 }
1851 type = OBJ_BAD;
1852 goto out;
1f688557
JH
1853}
1854
90191d37 1855static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 1856 struct object_info *oi)
90191d37
JK
1857{
1858 struct pack_window *w_curs = NULL;
1859 unsigned long size;
1860 off_t curpos = obj_offset;
1861 enum object_type type;
1862
412916ee
JK
1863 /*
1864 * We always get the representation type, but only convert it to
1865 * a "real" type later if the caller is interested.
1866 */
90191d37
JK
1867 type = unpack_object_header(p, &w_curs, &curpos, &size);
1868
23c339c0 1869 if (oi->sizep) {
90191d37
JK
1870 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1871 off_t tmp_pos = curpos;
1872 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1873 type, obj_offset);
1874 if (!base_offset) {
1875 type = OBJ_BAD;
1876 goto out;
1877 }
23c339c0
JK
1878 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1879 if (*oi->sizep == 0) {
90191d37
JK
1880 type = OBJ_BAD;
1881 goto out;
1882 }
1883 } else {
23c339c0 1884 *oi->sizep = size;
90191d37
JK
1885 }
1886 }
1887
23c339c0 1888 if (oi->disk_sizep) {
90191d37 1889 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 1890 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
1891 }
1892
23c339c0
JK
1893 if (oi->typep) {
1894 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1895 if (*oi->typep < 0) {
412916ee
JK
1896 type = OBJ_BAD;
1897 goto out;
1898 }
1899 }
90191d37 1900
5d642e75
JK
1901 if (oi->delta_base_sha1) {
1902 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1903 const unsigned char *base;
1904
1905 base = get_delta_base_sha1(p, &w_curs, curpos,
1906 type, obj_offset);
1907 if (!base) {
1908 type = OBJ_BAD;
1909 goto out;
1910 }
1911
1912 hashcpy(oi->delta_base_sha1, base);
1913 } else
1914 hashclr(oi->delta_base_sha1);
1915 }
1916
90191d37
JK
1917out:
1918 unuse_pack(&w_curs);
1919 return type;
1920}
1921
eb950c19 1922static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1923 struct pack_window **w_curs,
c4001d92 1924 off_t curpos,
eb950c19 1925 unsigned long size)
de530aaa
SP
1926{
1927 int st;
ef49a7a0 1928 git_zstream stream;
079afb18 1929 unsigned char *buffer, *in;
de530aaa 1930
4ab07e4d 1931 buffer = xmallocz(size);
de530aaa 1932 memset(&stream, 0, sizeof(stream));
de530aaa 1933 stream.next_out = buffer;
39eea7bd 1934 stream.avail_out = size + 1;
de530aaa 1935
39c68542 1936 git_inflate_init(&stream);
079afb18 1937 do {
2b87c45b 1938 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1939 stream.next_in = in;
39c68542 1940 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1941 if (!stream.avail_out)
1942 break; /* the payload is larger than it should be */
2b87c45b 1943 curpos += stream.next_in - in;
079afb18 1944 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1945 git_inflate_end(&stream);
de530aaa
SP
1946 if ((st != Z_STREAM_END) || stream.total_out != size) {
1947 free(buffer);
1948 return NULL;
1949 }
1950
1951 return buffer;
1952}
1953
e5e01619
LT
1954#define MAX_DELTA_CACHE (256)
1955
18bdec11 1956static size_t delta_base_cached;
5e08ecbf
NP
1957
1958static struct delta_base_cache_lru_list {
1959 struct delta_base_cache_lru_list *prev;
1960 struct delta_base_cache_lru_list *next;
1961} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1962
e5e01619 1963static struct delta_base_cache_entry {
5e08ecbf
NP
1964 struct delta_base_cache_lru_list lru;
1965 void *data;
e5e01619
LT
1966 struct packed_git *p;
1967 off_t base_offset;
1968 unsigned long size;
e5e01619
LT
1969 enum object_type type;
1970} delta_base_cache[MAX_DELTA_CACHE];
1971
1972static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1973{
1974 unsigned long hash;
1975
1976 hash = (unsigned long)p + (unsigned long)base_offset;
1977 hash += (hash >> 8) + (hash >> 16);
3358004a 1978 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1979}
1980
84dd81c1
TR
1981static struct delta_base_cache_entry *
1982get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
1983{
1984 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
1985 return delta_base_cache + hash;
1986}
1987
1988static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1989 struct packed_git *p, off_t base_offset)
1990{
5266d369
JH
1991 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1992}
1993
84dd81c1
TR
1994static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1995{
1996 struct delta_base_cache_entry *ent;
1997 ent = get_delta_base_cache_entry(p, base_offset);
1998 return eq_delta_base_cache_entry(ent, p, base_offset);
1999}
2000
2001static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2002{
2003 ent->data = NULL;
2004 ent->lru.next->prev = ent->lru.prev;
2005 ent->lru.prev->next = ent->lru.next;
2006 delta_base_cached -= ent->size;
2007}
2008
62f255ad 2009static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 2010 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 2011{
84dd81c1 2012 struct delta_base_cache_entry *ent;
e5e01619 2013 void *ret;
e5e01619 2014
84dd81c1
TR
2015 ent = get_delta_base_cache_entry(p, base_offset);
2016
2017 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2018 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2019
84dd81c1
TR
2020 ret = ent->data;
2021
2022 if (!keep_cache)
2023 clear_delta_base_cache_entry(ent);
2024 else
182af834 2025 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
2026 *type = ent->type;
2027 *base_size = ent->size;
2028 return ret;
62f255ad
LT
2029}
2030
18bdec11
SP
2031static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2032{
2033 if (ent->data) {
2034 free(ent->data);
2035 ent->data = NULL;
5e08ecbf
NP
2036 ent->lru.next->prev = ent->lru.prev;
2037 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
2038 delta_base_cached -= ent->size;
2039 }
2040}
2041
3d20c636
SP
2042void clear_delta_base_cache(void)
2043{
2044 unsigned long p;
2045 for (p = 0; p < MAX_DELTA_CACHE; p++)
2046 release_delta_base_cache(&delta_base_cache[p]);
2047}
2048
62f255ad
LT
2049static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2050 void *base, unsigned long base_size, enum object_type type)
2051{
5e08ecbf 2052 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2053 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 2054 struct delta_base_cache_lru_list *lru;
e5e01619 2055
18bdec11
SP
2056 release_delta_base_cache(ent);
2057 delta_base_cached += base_size;
5e08ecbf
NP
2058
2059 for (lru = delta_base_cache_lru.next;
2060 delta_base_cached > delta_base_cache_limit
2061 && lru != &delta_base_cache_lru;
2062 lru = lru->next) {
2063 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
2064 if (f->type == OBJ_BLOB)
2065 release_delta_base_cache(f);
2066 }
5e08ecbf
NP
2067 for (lru = delta_base_cache_lru.next;
2068 delta_base_cached > delta_base_cache_limit
2069 && lru != &delta_base_cache_lru;
2070 lru = lru->next) {
2071 struct delta_base_cache_entry *f = (void *)lru;
2072 release_delta_base_cache(f);
2073 }
18bdec11 2074
e5e01619
LT
2075 ent->p = p;
2076 ent->base_offset = base_offset;
2077 ent->type = type;
2078 ent->data = base;
2079 ent->size = base_size;
5e08ecbf
NP
2080 ent->lru.next = &delta_base_cache_lru;
2081 ent->lru.prev = delta_base_cache_lru.prev;
2082 delta_base_cache_lru.prev->next = &ent->lru;
2083 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
2084}
2085
c2c5b270
CC
2086static void *read_object(const unsigned char *sha1, enum object_type *type,
2087 unsigned long *size);
2088
5f44324d
JH
2089static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2090{
2091 static FILE *log_file;
2092
b12ca963
NTND
2093 if (!log_pack_access)
2094 log_pack_access = getenv("GIT_TRACE_PACK_ACCESS");
2095 if (!log_pack_access)
2096 log_pack_access = no_log_pack_access;
2097 if (log_pack_access == no_log_pack_access)
2098 return;
2099
5f44324d
JH
2100 if (!log_file) {
2101 log_file = fopen(log_pack_access, "w");
2102 if (!log_file) {
2103 error("cannot open pack access log '%s' for writing: %s",
2104 log_pack_access, strerror(errno));
b12ca963 2105 log_pack_access = no_log_pack_access;
5f44324d
JH
2106 return;
2107 }
2108 }
2109 fprintf(log_file, "%s %"PRIuMAX"\n",
2110 p->pack_name, (uintmax_t)obj_offset);
2111 fflush(log_file);
2112}
2113
0e8189e2
NP
2114int do_check_packed_object_crc;
2115
abe601bb
TR
2116#define UNPACK_ENTRY_STACK_PREALLOC 64
2117struct unpack_entry_stack_ent {
2118 off_t obj_offset;
2119 off_t curpos;
2120 unsigned long size;
2121};
2122
c4001d92 2123void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2124 enum object_type *final_type, unsigned long *final_size)
1f688557 2125{
03e79c88 2126 struct pack_window *w_curs = NULL;
c4001d92 2127 off_t curpos = obj_offset;
abe601bb
TR
2128 void *data = NULL;
2129 unsigned long size;
2130 enum object_type type;
2131 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2132 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2133 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2134 int base_from_cache = 0;
1f688557 2135
b12ca963 2136 if (log_pack_access != no_log_pack_access)
5f44324d
JH
2137 write_pack_access_log(p, obj_offset);
2138
abe601bb
TR
2139 /* PHASE 1: drill down to the innermost base object */
2140 for (;;) {
2141 off_t base_offset;
2142 int i;
2143 struct delta_base_cache_entry *ent;
2144
77965f8b
NTND
2145 ent = get_delta_base_cache_entry(p, curpos);
2146 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2147 type = ent->type;
2148 data = ent->data;
2149 size = ent->size;
2150 clear_delta_base_cache_entry(ent);
2151 base_from_cache = 1;
2152 break;
2153 }
2154
abe601bb
TR
2155 if (do_check_packed_object_crc && p->index_version > 1) {
2156 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2157 unsigned long len = revidx[1].offset - obj_offset;
2158 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2159 const unsigned char *sha1 =
2160 nth_packed_object_sha1(p, revidx->nr);
2161 error("bad packed object CRC for %s",
2162 sha1_to_hex(sha1));
2163 mark_bad_packed_object(p, sha1);
2164 unuse_pack(&w_curs);
2165 return NULL;
2166 }
2167 }
2168
abe601bb
TR
2169 type = unpack_object_header(p, &w_curs, &curpos, &size);
2170 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2171 break;
2172
2173 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2174 if (!base_offset) {
2175 error("failed to validate delta base reference "
2176 "at offset %"PRIuMAX" from %s",
2177 (uintmax_t)curpos, p->pack_name);
2178 /* bail to phase 2, in hopes of recovery */
2179 data = NULL;
2180 break;
2181 }
2182
2183 /* push object, proceed to base */
2184 if (delta_stack_nr >= delta_stack_alloc
2185 && delta_stack == small_delta_stack) {
2186 delta_stack_alloc = alloc_nr(delta_stack_nr);
2187 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2188 memcpy(delta_stack, small_delta_stack,
2189 sizeof(*delta_stack)*delta_stack_nr);
2190 } else {
2191 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2192 }
2193 i = delta_stack_nr++;
2194 delta_stack[i].obj_offset = obj_offset;
2195 delta_stack[i].curpos = curpos;
2196 delta_stack[i].size = size;
2197
2198 curpos = obj_offset = base_offset;
0e8189e2
NP
2199 }
2200
abe601bb
TR
2201 /* PHASE 2: handle the base */
2202 switch (type) {
eb32d236
NP
2203 case OBJ_OFS_DELTA:
2204 case OBJ_REF_DELTA:
abe601bb
TR
2205 if (data)
2206 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2207 break;
a733cb60 2208 case OBJ_COMMIT:
a733cb60 2209 case OBJ_TREE:
a733cb60 2210 case OBJ_BLOB:
a733cb60 2211 case OBJ_TAG:
abe601bb
TR
2212 if (!base_from_cache)
2213 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2214 break;
1f688557 2215 default:
8eca0b47
NP
2216 data = NULL;
2217 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2218 type, (uintmax_t)obj_offset, p->pack_name);
2219 }
2220
2221 /* PHASE 3: apply deltas in order */
2222
2223 /* invariants:
2224 * 'data' holds the base data, or NULL if there was corruption
2225 */
2226 while (delta_stack_nr) {
2227 void *delta_data;
2228 void *base = data;
2229 unsigned long delta_size, base_size = size;
2230 int i;
2231
2232 data = NULL;
2233
2234 if (base)
2235 add_delta_base_cache(p, obj_offset, base, base_size, type);
2236
2237 if (!base) {
2238 /*
2239 * We're probably in deep shit, but let's try to fetch
2240 * the required base anyway from another pack or loose.
2241 * This is costly but should happen only in the presence
2242 * of a corrupted pack, and is better than failing outright.
2243 */
2244 struct revindex_entry *revidx;
2245 const unsigned char *base_sha1;
2246 revidx = find_pack_revindex(p, obj_offset);
2247 if (revidx) {
2248 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2249 error("failed to read delta base object %s"
2250 " at offset %"PRIuMAX" from %s",
2251 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2252 p->pack_name);
2253 mark_bad_packed_object(p, base_sha1);
2254 base = read_object(base_sha1, &type, &base_size);
2255 }
2256 }
2257
2258 i = --delta_stack_nr;
2259 obj_offset = delta_stack[i].obj_offset;
2260 curpos = delta_stack[i].curpos;
2261 delta_size = delta_stack[i].size;
2262
2263 if (!base)
2264 continue;
2265
2266 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2267
2268 if (!delta_data) {
2269 error("failed to unpack compressed delta "
2270 "at offset %"PRIuMAX" from %s",
2271 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2272 data = NULL;
2273 continue;
2274 }
2275
2276 data = patch_delta(base, base_size,
2277 delta_data, delta_size,
2278 &size);
1ee886c1
JK
2279
2280 /*
2281 * We could not apply the delta; warn the user, but keep going.
2282 * Our failure will be noticed either in the next iteration of
2283 * the loop, or if this is the final delta, in the caller when
2284 * we return NULL. Those code paths will take care of making
2285 * a more explicit warning and retrying with another copy of
2286 * the object.
2287 */
abe601bb 2288 if (!data)
1ee886c1 2289 error("failed to apply delta");
abe601bb 2290
4b8f772c 2291 free(delta_data);
1f688557 2292 }
abe601bb
TR
2293
2294 *final_type = type;
2295 *final_size = size;
2296
03e79c88 2297 unuse_pack(&w_curs);
21666f1a 2298 return data;
1f688557
JH
2299}
2300
d079837e 2301const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2302 uint32_t n)
9a217f2a 2303{
42873078 2304 const unsigned char *index = p->index_data;
d079837e
SP
2305 if (!index) {
2306 if (open_pack_index(p))
2307 return NULL;
2308 index = p->index_data;
2309 }
57059091 2310 if (n >= p->num_objects)
d72308e0 2311 return NULL;
74e34e1f
NP
2312 index += 4 * 256;
2313 if (p->index_version == 1) {
2314 return index + 24 * n + 4;
2315 } else {
2316 index += 8;
2317 return index + 20 * n;
2318 }
2319}
2320
99093238 2321off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2322{
2323 const unsigned char *index = p->index_data;
2324 index += 4 * 256;
2325 if (p->index_version == 1) {
2326 return ntohl(*((uint32_t *)(index + 24 * n)));
2327 } else {
2328 uint32_t off;
2329 index += 8 + p->num_objects * (20 + 4);
2330 off = ntohl(*((uint32_t *)(index + 4 * n)));
2331 if (!(off & 0x80000000))
2332 return off;
2333 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2334 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2335 ntohl(*((uint32_t *)(index + 4)));
2336 }
9a217f2a
JH
2337}
2338
c4001d92 2339off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2340 struct packed_git *p)
1f688557 2341{
42873078 2342 const uint32_t *level1_ofs = p->index_data;
42873078 2343 const unsigned char *index = p->index_data;
628522ec
JH
2344 unsigned hi, lo, stride;
2345 static int use_lookup = -1;
2346 static int debug_lookup = -1;
2347
2348 if (debug_lookup < 0)
2349 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2350
d079837e
SP
2351 if (!index) {
2352 if (open_pack_index(p))
2353 return 0;
2354 level1_ofs = p->index_data;
2355 index = p->index_data;
2356 }
74e34e1f
NP
2357 if (p->index_version > 1) {
2358 level1_ofs += 2;
2359 index += 8;
2360 }
42873078 2361 index += 4 * 256;
74e34e1f
NP
2362 hi = ntohl(level1_ofs[*sha1]);
2363 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2364 if (p->index_version > 1) {
2365 stride = 20;
2366 } else {
2367 stride = 24;
2368 index += 4;
2369 }
2370
2371 if (debug_lookup)
6e1c2344 2372 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2373 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2374
2375 if (use_lookup < 0)
2376 use_lookup = !!getenv("GIT_USE_LOOKUP");
2377 if (use_lookup) {
2378 int pos = sha1_entry_pos(index, stride, 0,
2379 lo, hi, p->num_objects, sha1);
2380 if (pos < 0)
2381 return 0;
2382 return nth_packed_object_offset(p, pos);
2383 }
1f688557
JH
2384
2385 do {
74e34e1f 2386 unsigned mi = (lo + hi) / 2;
628522ec
JH
2387 int cmp = hashcmp(index + mi * stride, sha1);
2388
2389 if (debug_lookup)
2390 printf("lo %u hi %u rg %u mi %u\n",
2391 lo, hi, hi - lo, mi);
43057304 2392 if (!cmp)
74e34e1f 2393 return nth_packed_object_offset(p, mi);
1f688557
JH
2394 if (cmp > 0)
2395 hi = mi;
2396 else
2397 lo = mi+1;
2398 } while (lo < hi);
2399 return 0;
2400}
2401
4c080182 2402int is_pack_valid(struct packed_git *p)
d131b7af
SP
2403{
2404 /* An already open pack is known to be valid. */
2405 if (p->pack_fd != -1)
2406 return 1;
2407
2408 /* If the pack has one window completely covering the
2409 * file size, the pack is known to be valid even if
2410 * the descriptor is not currently open.
2411 */
2412 if (p->windows) {
2413 struct pack_window *w = p->windows;
2414
2415 if (!w->offset && w->len == p->pack_size)
2416 return 1;
2417 }
2418
2419 /* Force the pack to open to prove its valid. */
2420 return !open_packed_git(p);
2421}
2422
95099731
NTND
2423static int fill_pack_entry(const unsigned char *sha1,
2424 struct pack_entry *e,
2425 struct packed_git *p)
2426{
2427 off_t offset;
2428
2429 if (p->num_bad_objects) {
2430 unsigned i;
2431 for (i = 0; i < p->num_bad_objects; i++)
2432 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2433 return 0;
2434 }
2435
2436 offset = find_pack_entry_one(sha1, p);
2437 if (!offset)
2438 return 0;
2439
2440 /*
2441 * We are about to tell the caller where they can locate the
2442 * requested object. We better make sure the packfile is
2443 * still here and can be accessed before supplying that
2444 * answer, as it may have been deleted since the index was
2445 * loaded!
2446 */
2447 if (!is_pack_valid(p)) {
2448 warning("packfile %s cannot be accessed", p->pack_name);
2449 return 0;
2450 }
2451 e->offset = offset;
2452 e->p = p;
2453 hashcpy(e->sha1, sha1);
2454 return 1;
2455}
2456
4d6acb70 2457static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2458{
2459 struct packed_git *p;
43057304 2460
1f688557 2461 prepare_packed_git();
f7c22cc6
NP
2462 if (!packed_git)
2463 return 0;
1f688557 2464
c01f51cc
NTND
2465 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2466 return 1;
8eca0b47 2467
c01f51cc 2468 for (p = packed_git; p; p = p->next) {
1b1005d1
MH
2469 if (p == last_found_pack)
2470 continue; /* we already checked this one */
f7c22cc6 2471
1b1005d1
MH
2472 if (fill_pack_entry(sha1, e, p)) {
2473 last_found_pack = p;
2474 return 1;
2475 }
c01f51cc 2476 }
1f688557
JH
2477 return 0;
2478}
2479
a6080a0a 2480struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2481 struct packed_git *packs)
2482{
2483 struct packed_git *p;
bf592c50
DB
2484
2485 for (p = packs; p; p = p->next) {
43057304 2486 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2487 return p;
2488 }
2489 return NULL;
21666f1a 2490
bf592c50
DB
2491}
2492
052fe5ea 2493static int sha1_loose_object_info(const unsigned char *sha1,
23c339c0 2494 struct object_info *oi)
65c2e0c3 2495{
36e4d74a 2496 int status;
65c2e0c3
JH
2497 unsigned long mapsize, size;
2498 void *map;
ef49a7a0 2499 git_zstream stream;
d65a16f6 2500 char hdr[32];
65c2e0c3 2501
5d642e75
JK
2502 if (oi->delta_base_sha1)
2503 hashclr(oi->delta_base_sha1);
2504
052fe5ea
JK
2505 /*
2506 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2507 * need to look inside the object at all. Note that we
2508 * do not optimize out the stat call, even if the
2509 * caller doesn't care about the disk-size, since our
2510 * return value implicitly indicates whether the
2511 * object even exists.
052fe5ea 2512 */
23c339c0 2513 if (!oi->typep && !oi->sizep) {
4ef8d1dd
JH
2514 struct stat st;
2515 if (stat_sha1_file(sha1, &st) < 0)
2516 return -1;
2517 if (oi->disk_sizep)
23c339c0 2518 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2519 return 0;
2520 }
2521
bb6b8e4f 2522 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2523 if (!map)
dbea72a8 2524 return -1;
23c339c0
JK
2525 if (oi->disk_sizep)
2526 *oi->disk_sizep = mapsize;
36e4d74a
JH
2527 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2528 status = error("unable to unpack %s header",
2529 sha1_to_hex(sha1));
21666f1a 2530 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2531 status = error("unable to parse %s header", sha1_to_hex(sha1));
23c339c0
JK
2532 else if (oi->sizep)
2533 *oi->sizep = size;
39c68542 2534 git_inflate_end(&stream);
65c2e0c3 2535 munmap(map, mapsize);
23c339c0
JK
2536 if (oi->typep)
2537 *oi->typep = status;
052fe5ea 2538 return 0;
65c2e0c3
JH
2539}
2540
de7b5d62 2541int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 2542{
c4d9986f 2543 struct cached_object *co;
f0df4ed5 2544 struct pack_entry e;
5b086407 2545 int rtype;
1f7117ef 2546 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
f0df4ed5 2547
1f7117ef 2548 co = find_cached_object(real);
c4d9986f 2549 if (co) {
5b086407
JK
2550 if (oi->typep)
2551 *(oi->typep) = co->type;
9a490590
JH
2552 if (oi->sizep)
2553 *(oi->sizep) = co->size;
161f00e7
JK
2554 if (oi->disk_sizep)
2555 *(oi->disk_sizep) = 0;
5d642e75
JK
2556 if (oi->delta_base_sha1)
2557 hashclr(oi->delta_base_sha1);
9a490590 2558 oi->whence = OI_CACHED;
5b086407 2559 return 0;
c4d9986f
NTND
2560 }
2561
1f7117ef 2562 if (!find_pack_entry(real, &e)) {
ddd63e64 2563 /* Most likely it's a loose object. */
1f7117ef 2564 if (!sha1_loose_object_info(real, oi)) {
9a490590 2565 oi->whence = OI_LOOSE;
5b086407 2566 return 0;
9a490590 2567 }
ddd63e64
SG
2568
2569 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2570 reprepare_packed_git();
1f7117ef 2571 if (!find_pack_entry(real, &e))
052fe5ea 2572 return -1;
f0df4ed5 2573 }
3d77d877 2574
23c339c0 2575 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2576 if (rtype < 0) {
1f7117ef
CC
2577 mark_bad_packed_object(e.p, real);
2578 return sha1_object_info_extended(real, oi, 0);
5266d369
JH
2579 } else if (in_delta_base_cache(e.p, e.offset)) {
2580 oi->whence = OI_DBCACHED;
9a490590
JH
2581 } else {
2582 oi->whence = OI_PACKED;
2583 oi->u.packed.offset = e.offset;
2584 oi->u.packed.pack = e.p;
2585 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2586 rtype == OBJ_OFS_DELTA);
3d77d877
NP
2587 }
2588
5b086407 2589 return 0;
f0df4ed5
JS
2590}
2591
3fc0dca9 2592/* returns enum object_type or negative */
9a490590
JH
2593int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2594{
5b086407 2595 enum object_type type;
d099b717 2596 struct object_info oi = {NULL};
9a490590 2597
5b086407 2598 oi.typep = &type;
9a490590 2599 oi.sizep = sizep;
de7b5d62 2600 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
5b086407
JK
2601 return -1;
2602 return type;
9a490590
JH
2603}
2604
21666f1a
NP
2605static void *read_packed_sha1(const unsigned char *sha1,
2606 enum object_type *type, unsigned long *size)
1f688557
JH
2607{
2608 struct pack_entry e;
8eca0b47 2609 void *data;
1f688557 2610
cd673c1f 2611 if (!find_pack_entry(sha1, &e))
1f688557 2612 return NULL;
8eca0b47
NP
2613 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2614 if (!data) {
2615 /*
2616 * We're probably in deep shit, but let's try to fetch
2617 * the required object anyway from another pack or loose.
2618 * This should happen only in the presence of a corrupted
2619 * pack, and is better than failing outright.
2620 */
2621 error("failed to read object %s at offset %"PRIuMAX" from %s",
2622 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2623 mark_bad_packed_object(e.p, sha1);
ac939109 2624 data = read_object(sha1, type, size);
8eca0b47
NP
2625 }
2626 return data;
1f688557
JH
2627}
2628
21666f1a
NP
2629int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2630 unsigned char *sha1)
d66b37bb
JH
2631{
2632 struct cached_object *co;
2633
21666f1a 2634 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2635 if (has_sha1_file(sha1) || find_cached_object(sha1))
2636 return 0;
2637 if (cached_object_alloc <= cached_object_nr) {
2638 cached_object_alloc = alloc_nr(cached_object_alloc);
2639 cached_objects = xrealloc(cached_objects,
2640 sizeof(*cached_objects) *
2641 cached_object_alloc);
2642 }
2643 co = &cached_objects[cached_object_nr++];
2644 co->size = len;
21666f1a 2645 co->type = type;
efa13f7b
JH
2646 co->buf = xmalloc(len);
2647 memcpy(co->buf, buf, len);
d66b37bb
JH
2648 hashcpy(co->sha1, sha1);
2649 return 0;
2650}
2651
c2c5b270
CC
2652static void *read_object(const unsigned char *sha1, enum object_type *type,
2653 unsigned long *size)
0fcfd160
LT
2654{
2655 unsigned long mapsize;
2656 void *map, *buf;
d66b37bb
JH
2657 struct cached_object *co;
2658
2659 co = find_cached_object(sha1);
2660 if (co) {
21666f1a 2661 *type = co->type;
d66b37bb 2662 *size = co->size;
182af834 2663 return xmemdupz(co->buf, co->size);
d66b37bb 2664 }
0fcfd160 2665
8276c007
PE
2666 buf = read_packed_sha1(sha1, type, size);
2667 if (buf)
2668 return buf;
bb6b8e4f 2669 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2670 if (map) {
7efbff75 2671 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2672 munmap(map, mapsize);
2673 return buf;
2674 }
637cdd9d 2675 reprepare_packed_git();
8276c007 2676 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2677}
2678
b6c4cecc
JH
2679/*
2680 * This function dies on corrupt objects; the callers who want to
2681 * deal with them should arrange to call read_object() and give error
2682 * messages themselves.
2683 */
5bf29b95
JH
2684void *read_sha1_file_extended(const unsigned char *sha1,
2685 enum object_type *type,
2686 unsigned long *size,
2687 unsigned flag)
ac939109 2688{
3ba7a065 2689 void *data;
e8b15e61 2690 char *path;
b6c4cecc 2691 const struct packed_git *p;
bf93eea0 2692 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
b6c4cecc 2693
3ba7a065
JH
2694 errno = 0;
2695 data = read_object(repl, type, size);
4bbf5a26 2696 if (data)
b6c4cecc 2697 return data;
68095570 2698
25f3af3f 2699 if (errno && errno != ENOENT)
3ba7a065
JH
2700 die_errno("failed to read object %s", sha1_to_hex(sha1));
2701
68095570 2702 /* die if we replaced an object with one that does not exist */
b6c4cecc 2703 if (repl != sha1)
68095570
CC
2704 die("replacement %s not found for %s",
2705 sha1_to_hex(repl), sha1_to_hex(sha1));
2706
b6c4cecc
JH
2707 if (has_loose_object(repl)) {
2708 path = sha1_file_name(sha1);
2709 die("loose object %s (stored in %s) is corrupt",
2710 sha1_to_hex(repl), path);
e8b15e61 2711 }
68095570 2712
b6c4cecc
JH
2713 if ((p = has_packed_and_bad(repl)) != NULL)
2714 die("packed object %s (stored in %s) is corrupt",
2715 sha1_to_hex(repl), p->pack_name);
f5552aee 2716
b6c4cecc 2717 return NULL;
ac939109
NP
2718}
2719
40469ee9 2720void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2721 const char *required_type_name,
40469ee9
JH
2722 unsigned long *size,
2723 unsigned char *actual_sha1_return)
f4913f91 2724{
21666f1a 2725 enum object_type type, required_type;
f4913f91
JH
2726 void *buffer;
2727 unsigned long isize;
40469ee9 2728 unsigned char actual_sha1[20];
f4913f91 2729
21666f1a 2730 required_type = type_from_string(required_type_name);
e702496e 2731 hashcpy(actual_sha1, sha1);
40469ee9
JH
2732 while (1) {
2733 int ref_length = -1;
2734 const char *ref_type = NULL;
f4913f91 2735
21666f1a 2736 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2737 if (!buffer)
2738 return NULL;
21666f1a 2739 if (type == required_type) {
40469ee9
JH
2740 *size = isize;
2741 if (actual_sha1_return)
e702496e 2742 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2743 return buffer;
2744 }
2745 /* Handle references */
21666f1a 2746 else if (type == OBJ_COMMIT)
40469ee9 2747 ref_type = "tree ";
21666f1a 2748 else if (type == OBJ_TAG)
40469ee9
JH
2749 ref_type = "object ";
2750 else {
2751 free(buffer);
2752 return NULL;
2753 }
2754 ref_length = strlen(ref_type);
f4913f91 2755
50974ec9
MK
2756 if (ref_length + 40 > isize ||
2757 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2758 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2759 free(buffer);
2760 return NULL;
2761 }
1cf58e72 2762 free(buffer);
40469ee9
JH
2763 /* Now we have the ID of the referred-to object in
2764 * actual_sha1. Check again. */
f4913f91 2765 }
f4913f91
JH
2766}
2767
ce9fbf16 2768static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2769 const char *type, unsigned char *sha1,
d65a16f6 2770 char *hdr, int *hdrlen)
d410c0f5 2771{
9126f009 2772 git_SHA_CTX c;
d410c0f5
JH
2773
2774 /* Generate the header */
d65a16f6 2775 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2776
2777 /* Sha1.. */
9126f009
NP
2778 git_SHA1_Init(&c);
2779 git_SHA1_Update(&c, hdr, *hdrlen);
2780 git_SHA1_Update(&c, buf, len);
2781 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2782}
2783
230f1322 2784/*
5a688fe4
JH
2785 * Move the just written object into its final resting place.
2786 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2787 * "moving" is only a part of what it does, when no patch between
2788 * master to pu changes the call sites of this function.
230f1322 2789 */
839837b9 2790int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2791{
e32c0a9c 2792 int ret = 0;
5a688fe4 2793
348df166 2794 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2795 goto try_rename;
2796 else if (link(tmpfile, filename))
e32c0a9c 2797 ret = errno;
7ebb6fca
LT
2798
2799 /*
2800 * Coda hack - coda doesn't like cross-directory links,
2801 * so we fall back to a rename, which will mean that it
2802 * won't be able to check collisions, but that's not a
2803 * big deal.
2804 *
2805 * The same holds for FAT formatted media.
2806 *
3be1f18e 2807 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2808 * left to unlink.
2809 */
2810 if (ret && ret != EEXIST) {
be66a6c4 2811 try_rename:
7ebb6fca 2812 if (!rename(tmpfile, filename))
3be1f18e 2813 goto out;
9e48b389 2814 ret = errno;
230f1322 2815 }
691f1a28 2816 unlink_or_warn(tmpfile);
230f1322
LT
2817 if (ret) {
2818 if (ret != EEXIST) {
82247e9b 2819 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2820 }
2821 /* FIXME!!! Collision check here ? */
2822 }
2823
3be1f18e 2824out:
5256b006 2825 if (adjust_shared_perm(filename))
5a688fe4 2826 return error("unable to set permission to '%s'", filename);
230f1322
LT
2827 return 0;
2828}
2829
4d548150
LT
2830static int write_buffer(int fd, const void *buf, size_t len)
2831{
d34cf19b 2832 if (write_in_full(fd, buf, len) < 0)
93822c22 2833 return error("file write error (%s)", strerror(errno));
4d548150
LT
2834 return 0;
2835}
2836
ce9fbf16 2837int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2838 unsigned char *sha1)
2839{
d65a16f6 2840 char hdr[32];
abdc3fc8
RS
2841 int hdrlen;
2842 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2843 return 0;
2844}
2845
e9039dd3
LT
2846/* Finalize a file on disk, and close it. */
2847static void close_sha1_file(int fd)
2848{
aafe9fba
LT
2849 if (fsync_object_files)
2850 fsync_or_die(fd, "sha1 file");
e9039dd3 2851 if (close(fd) != 0)
d824cbba 2852 die_errno("error when closing sha1 file");
e9039dd3
LT
2853}
2854
5723fe7e
LT
2855/* Size of directory component, including the ending '/' */
2856static inline int directory_size(const char *filename)
2857{
2858 const char *s = strrchr(filename, '/');
2859 if (!s)
2860 return 0;
2861 return s - filename + 1;
2862}
2863
2864/*
2865 * This creates a temporary file in the same directory as the final
2866 * 'filename'
2867 *
2868 * We want to avoid cross-directory filename renames, because those
2869 * can have problems on various filesystems (FAT, NFS, Coda).
2870 */
2871static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2872{
2873 int fd, dirlen = directory_size(filename);
2874
2875 if (dirlen + 20 > bufsiz) {
2876 errno = ENAMETOOLONG;
2877 return -1;
2878 }
2879 memcpy(buffer, filename, dirlen);
2880 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2881 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2882 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2883 /* Make sure the directory exists */
6ff6af62 2884 memcpy(buffer, filename, dirlen);
5723fe7e 2885 buffer[dirlen-1] = 0;
b2476a60
JH
2886 if (mkdir(buffer, 0777) && errno != EEXIST)
2887 return -1;
2888 if (adjust_shared_perm(buffer))
5723fe7e
LT
2889 return -1;
2890
2891 /* Try again */
2892 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2893 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2894 }
2895 return fd;
2896}
2897
bbac7311 2898static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2899 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2900{
915308b1 2901 int fd, ret;
9892beba 2902 unsigned char compressed[4096];
ef49a7a0 2903 git_zstream stream;
748af44c
NP
2904 git_SHA_CTX c;
2905 unsigned char parano_sha1[20];
706bc531 2906 char *filename;
ab1900a3 2907 static char tmp_file[PATH_MAX];
a44c9a5e 2908
972a9155 2909 filename = sha1_file_name(sha1);
ab1900a3 2910 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 2911 if (fd < 0) {
35243577 2912 if (errno == EACCES)
82247e9b 2913 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 2914 else
5eaeda70 2915 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
2916 }
2917
0fcfd160
LT
2918 /* Set it up */
2919 memset(&stream, 0, sizeof(stream));
55bb5c91 2920 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 2921 stream.next_out = compressed;
9892beba 2922 stream.avail_out = sizeof(compressed);
748af44c 2923 git_SHA1_Init(&c);
a44c9a5e
LT
2924
2925 /* First header.. */
d65a16f6 2926 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2927 stream.avail_in = hdrlen;
55bb5c91
JH
2928 while (git_deflate(&stream, 0) == Z_OK)
2929 ; /* nothing */
748af44c 2930 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2931
2932 /* Then the data itself.. */
c00e657d 2933 stream.next_in = (void *)buf;
a44c9a5e 2934 stream.avail_in = len;
9892beba 2935 do {
748af44c 2936 unsigned char *in0 = stream.next_in;
55bb5c91 2937 ret = git_deflate(&stream, Z_FINISH);
748af44c 2938 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2939 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2940 die("unable to write sha1 file");
2941 stream.next_out = compressed;
2942 stream.avail_out = sizeof(compressed);
2943 } while (ret == Z_OK);
2944
ac54c277
LT
2945 if (ret != Z_STREAM_END)
2946 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 2947 ret = git_deflate_end_gently(&stream);
ac54c277
LT
2948 if (ret != Z_OK)
2949 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2950 git_SHA1_Final(parano_sha1, &c);
2951 if (hashcmp(sha1, parano_sha1) != 0)
2952 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2953
e9039dd3 2954 close_sha1_file(fd);
0fcfd160 2955
bbac7311
NP
2956 if (mtime) {
2957 struct utimbuf utb;
2958 utb.actime = mtime;
2959 utb.modtime = mtime;
ab1900a3 2960 if (utime(tmp_file, &utb) < 0)
bbac7311 2961 warning("failed utime() on %s: %s",
ab1900a3 2962 tmp_file, strerror(errno));
bbac7311
NP
2963 }
2964
ab1900a3 2965 return move_temp_to_file(tmp_file, filename);
0fcfd160 2966}
8237b185 2967
c00e657d 2968int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2969{
2970 unsigned char sha1[20];
2971 char hdr[32];
2972 int hdrlen;
2973
2974 /* Normally if we have it in the pack then we do not bother writing
2975 * it out into .git/objects/??/?{38} file.
2976 */
2977 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2978 if (returnsha1)
2979 hashcpy(returnsha1, sha1);
2980 if (has_sha1_file(sha1))
2981 return 0;
2982 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2983}
2984
2985int force_object_loose(const unsigned char *sha1, time_t mtime)
2986{
bbac7311
NP
2987 void *buf;
2988 unsigned long len;
2989 enum object_type type;
2990 char hdr[32];
2991 int hdrlen;
1fb23e65 2992 int ret;
bbac7311 2993
c529d75a 2994 if (has_loose_object(sha1))
bbac7311
NP
2995 return 0;
2996 buf = read_packed_sha1(sha1, &type, &len);
2997 if (!buf)
2998 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2999 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
3000 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3001 free(buf);
3002
3003 return ret;
bbac7311
NP
3004}
3005
bf592c50
DB
3006int has_pack_index(const unsigned char *sha1)
3007{
3008 struct stat st;
3009 if (stat(sha1_pack_index_name(sha1), &st))
3010 return 0;
3011 return 1;
bf592c50
DB
3012}
3013
cd673c1f
JH
3014int has_sha1_pack(const unsigned char *sha1)
3015{
3016 struct pack_entry e;
3017 return find_pack_entry(sha1, &e);
3018}
3019
8237b185
DB
3020int has_sha1_file(const unsigned char *sha1)
3021{
1f688557
JH
3022 struct pack_entry e;
3023
cd673c1f 3024 if (find_pack_entry(sha1, &e))
1f688557 3025 return 1;
45e8a748
JK
3026 if (has_loose_object(sha1))
3027 return 1;
3028 reprepare_packed_git();
3029 return find_pack_entry(sha1, &e);
8237b185 3030}
74400e71 3031
c879daa2
NTND
3032static void check_tree(const void *buf, size_t size)
3033{
3034 struct tree_desc desc;
3035 struct name_entry entry;
3036
3037 init_tree_desc(&desc, buf, size);
3038 while (tree_entry(&desc, &entry))
3039 /* do nothing
3040 * tree_entry() will die() on malformed entries */
3041 ;
3042}
3043
3044static void check_commit(const void *buf, size_t size)
3045{
3046 struct commit c;
3047 memset(&c, 0, sizeof(c));
3048 if (parse_commit_buffer(&c, buf, size))
3049 die("corrupt commit");
3050}
3051
3052static void check_tag(const void *buf, size_t size)
3053{
3054 struct tag t;
3055 memset(&t, 0, sizeof(t));
3056 if (parse_tag_buffer(&t, buf, size))
3057 die("corrupt tag");
3058}
3059
43df4f86 3060static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3061 enum object_type type,
3062 const char *path, unsigned flags)
e7332f96 3063{
6c510bee 3064 int ret, re_allocated = 0;
c4ce46fc 3065 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3066
7672db20 3067 if (!type)
edaec3fb 3068 type = OBJ_BLOB;
6c510bee
LT
3069
3070 /*
3071 * Convert blobs to git internal format
3072 */
43df4f86 3073 if ((type == OBJ_BLOB) && path) {
f285a2d7 3074 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3075 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3076 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3077 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3078 re_allocated = 1;
3079 }
3080 }
c4ce46fc 3081 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3082 if (type == OBJ_TREE)
3083 check_tree(buf, size);
3084 if (type == OBJ_COMMIT)
3085 check_commit(buf, size);
3086 if (type == OBJ_TAG)
3087 check_tag(buf, size);
3088 }
6c510bee 3089
7672db20 3090 if (write_object)
edaec3fb 3091 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3092 else
edaec3fb 3093 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3094 if (re_allocated)
6c510bee 3095 free(buf);
43df4f86
DP
3096 return ret;
3097}
3098
7b41e1e1
JH
3099static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3100 const char *path, unsigned flags)
3101{
3102 struct strbuf sbuf = STRBUF_INIT;
3103 int ret;
3104
3105 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3106 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3107 else
3108 ret = -1;
3109 strbuf_release(&sbuf);
3110 return ret;
3111}
3112
ea68b0ce
DP
3113#define SMALL_FILE_SIZE (32*1024)
3114
7b41e1e1
JH
3115static int index_core(unsigned char *sha1, int fd, size_t size,
3116 enum object_type type, const char *path,
3117 unsigned flags)
43df4f86
DP
3118{
3119 int ret;
43df4f86 3120
7b41e1e1 3121 if (!size) {
c4ce46fc 3122 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
3123 } else if (size <= SMALL_FILE_SIZE) {
3124 char *buf = xmalloc(size);
3125 if (size == read_in_full(fd, buf, size))
c4ce46fc 3126 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
3127 else
3128 ret = error("short read %s", strerror(errno));
3129 free(buf);
08bda208 3130 } else {
43df4f86 3131 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3132 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3133 munmap(buf, size);
08bda208 3134 }
7b41e1e1
JH
3135 return ret;
3136}
3137
4dd1fbc7 3138/*
568508e7
JH
3139 * This creates one packfile per large blob unless bulk-checkin
3140 * machinery is "plugged".
4dd1fbc7
JH
3141 *
3142 * This also bypasses the usual "convert-to-git" dance, and that is on
3143 * purpose. We could write a streaming version of the converting
3144 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3145 * (or equivalent in-core API described above). However, that is
3146 * somewhat complicated, as we do not know the size of the filter
3147 * result, which we need to know beforehand when writing a git object.
3148 * Since the primary motivation for trying to stream from the working
3149 * tree file and to avoid mmaping it in core is to deal with large
3150 * binary blobs, they generally do not want to get any conversion, and
3151 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3152 */
3153static int index_stream(unsigned char *sha1, int fd, size_t size,
3154 enum object_type type, const char *path,
3155 unsigned flags)
3156{
568508e7 3157 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3158}
3159
7b41e1e1
JH
3160int index_fd(unsigned char *sha1, int fd, struct stat *st,
3161 enum object_type type, const char *path, unsigned flags)
3162{
3163 int ret;
3164 size_t size = xsize_t(st->st_size);
3165
3166 if (!S_ISREG(st->st_mode))
3167 ret = index_pipe(sha1, fd, type, path, flags);
4f22b101
JK
3168 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3169 (path && would_convert_to_git(path, NULL, 0, 0)))
7b41e1e1 3170 ret = index_core(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3171 else
3172 ret = index_stream(sha1, fd, size, type, path, flags);
43df4f86 3173 close(fd);
aac17941 3174 return ret;
74400e71 3175}
ec1fcc16 3176
c4ce46fc 3177int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3178{
3179 int fd;
b760d3aa 3180 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3181
3182 switch (st->st_mode & S_IFMT) {
3183 case S_IFREG:
3184 fd = open(path, O_RDONLY);
3185 if (fd < 0)
3186 return error("open(\"%s\"): %s", path,
3187 strerror(errno));
c4ce46fc 3188 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3189 return error("%s: failed to insert into database",
3190 path);
3191 break;
3192 case S_IFLNK:
b760d3aa 3193 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 3194 char *errstr = strerror(errno);
ec1fcc16
JH
3195 return error("readlink(\"%s\"): %s", path,
3196 errstr);
3197 }
c4ce46fc 3198 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3199 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3200 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3201 return error("%s: failed to insert into database",
3202 path);
b760d3aa 3203 strbuf_release(&sb);
ec1fcc16 3204 break;
f35a6d3b
LT
3205 case S_IFDIR:
3206 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3207 default:
3208 return error("%s: unsupported file type", path);
3209 }
3210 return 0;
3211}
a69e5429
JH
3212
3213int read_pack_header(int fd, struct pack_header *header)
3214{
c697ad14
HO
3215 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3216 /* "eof before pack header was fully read" */
3217 return PH_ERROR_EOF;
3218
a69e5429
JH
3219 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3220 /* "protocol error (pack signature mismatch detected)" */
3221 return PH_ERROR_PACK_SIGNATURE;
3222 if (!pack_version_ok(header->hdr_version))
3223 /* "protocol error (pack version unsupported)" */
3224 return PH_ERROR_PROTOCOL;
3225 return 0;
3226}
40d52ff7
JK
3227
3228void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3229{
3230 enum object_type type = sha1_object_info(sha1, NULL);
3231 if (type < 0)
3232 die("%s is not a valid object", sha1_to_hex(sha1));
3233 if (type != expect)
3234 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3235 typename(expect));
3236}