sha1_file: remove recursion in unpack_entry
[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"
0fcfd160 24
144bde78
LT
25#ifndef O_NOATIME
26#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
27#define O_NOATIME 01000000
28#else
29#define O_NOATIME 0
30#endif
31#endif
32
28bd70d8
JN
33#define SZ_FMT PRIuMAX
34static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 35
96f1e58f 36const unsigned char null_sha1[20];
88cd621d 37
c597ba80
NTND
38/*
39 * This is meant to hold a *small* number of objects that you would
40 * want read_sha1_file() to be able to return, but yet you do not want
41 * to write them into the object store (e.g. a browse-only
42 * application).
43 */
44static struct cached_object {
45 unsigned char sha1[20];
46 enum object_type type;
47 void *buf;
48 unsigned long size;
49} *cached_objects;
50static int cached_object_nr, cached_object_alloc;
51
52static struct cached_object empty_tree = {
dab0d410 53 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
54 OBJ_TREE,
55 "",
56 0
57};
58
c01f51cc
NTND
59static struct packed_git *last_found_pack;
60
c597ba80
NTND
61static struct cached_object *find_cached_object(const unsigned char *sha1)
62{
63 int i;
64 struct cached_object *co = cached_objects;
65
66 for (i = 0; i < cached_object_nr; i++, co++) {
67 if (!hashcmp(co->sha1, sha1))
68 return co;
69 }
70 if (!hashcmp(sha1, empty_tree.sha1))
71 return &empty_tree;
72 return NULL;
73}
74
90a6464b
JH
75int mkdir_in_gitdir(const char *path)
76{
77 if (mkdir(path, 0777)) {
78 int saved_errno = errno;
79 struct stat st;
80 struct strbuf sb = STRBUF_INIT;
81
82 if (errno != EEXIST)
83 return -1;
84 /*
85 * Are we looking at a path in a symlinked worktree
86 * whose original repository does not yet have it?
87 * e.g. .git/rr-cache pointing at its original
88 * repository in which the user hasn't performed any
89 * conflict resolution yet?
90 */
91 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
92 strbuf_readlink(&sb, path, st.st_size) ||
93 !is_absolute_path(sb.buf) ||
94 mkdir(sb.buf, 0777)) {
95 strbuf_release(&sb);
96 errno = saved_errno;
97 return -1;
98 }
99 strbuf_release(&sb);
100 }
101 return adjust_shared_perm(path);
102}
103
b2cb9425
JH
104int safe_create_leading_directories(char *path)
105{
8385abfd 106 char *pos = path + offset_1st_component(path);
67d42212
JR
107 struct stat st;
108
b2cb9425
JH
109 while (pos) {
110 pos = strchr(pos, '/');
111 if (!pos)
112 break;
5f0bdf50
JH
113 while (*++pos == '/')
114 ;
115 if (!*pos)
116 break;
117 *--pos = '\0';
67d42212
JR
118 if (!stat(path, &st)) {
119 /* path exists */
120 if (!S_ISDIR(st.st_mode)) {
b2cb9425 121 *pos = '/';
67d42212 122 return -3;
b2cb9425 123 }
457f06d6 124 }
67d42212
JR
125 else if (mkdir(path, 0777)) {
126 *pos = '/';
127 return -1;
128 }
457f06d6
JS
129 else if (adjust_shared_perm(path)) {
130 *pos = '/';
131 return -2;
132 }
b2cb9425
JH
133 *pos++ = '/';
134 }
135 return 0;
136}
723c31fe 137
8e21d63b
JK
138int safe_create_leading_directories_const(const char *path)
139{
140 /* path points to cache entries, so xstrdup before messing with it */
141 char *buf = xstrdup(path);
142 int result = safe_create_leading_directories(buf);
143 free(buf);
144 return result;
145}
146
ace1534d
JH
147static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
148{
149 int i;
150 for (i = 0; i < 20; i++) {
151 static char hex[] = "0123456789abcdef";
152 unsigned int val = sha1[i];
153 char *pos = pathbuf + i*2 + (i > 0);
154 *pos++ = hex[val >> 4];
155 *pos = hex[val & 0xf];
156 }
157}
158
0fcfd160
LT
159/*
160 * NOTE! This returns a statically allocated buffer, so you have to be
790296fd 161 * careful about using it. Do an "xstrdup()" if you need to save the
0fcfd160 162 * filename.
ace1534d
JH
163 *
164 * Also note that this returns the location for creating. Reading
165 * SHA1 file can happen from any alternate directory listed in the
d19938ab 166 * DB_ENVIRONMENT environment variable if it is not found in
ace1534d 167 * the primary object database.
0fcfd160
LT
168 */
169char *sha1_file_name(const unsigned char *sha1)
170{
560fb6a1
JK
171 static char buf[PATH_MAX];
172 const char *objdir;
173 int len;
0fcfd160 174
560fb6a1
JK
175 objdir = get_object_directory();
176 len = strlen(objdir);
177
178 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
179 if (len + 43 > PATH_MAX)
180 die("insanely long object directory %s", objdir);
181 memcpy(buf, objdir, len);
182 buf[len] = '/';
183 buf[len+3] = '/';
184 buf[len+42] = '\0';
185 fill_sha1_path(buf + len + 1, sha1);
186 return buf;
0fcfd160
LT
187}
188
633f43e1 189static char *sha1_get_pack_name(const unsigned char *sha1,
6eec46bd 190 char **name, char **base, const char *which)
bf592c50
DB
191{
192 static const char hex[] = "0123456789abcdef";
633f43e1 193 char *buf;
bf592c50
DB
194 int i;
195
633f43e1 196 if (!*base) {
bf592c50
DB
197 const char *sha1_file_directory = get_object_directory();
198 int len = strlen(sha1_file_directory);
633f43e1 199 *base = xmalloc(len + 60);
6eec46bd
JH
200 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
201 sha1_file_directory, which);
633f43e1 202 *name = *base + len + 11;
bf592c50
DB
203 }
204
633f43e1 205 buf = *name;
bf592c50
DB
206
207 for (i = 0; i < 20; i++) {
208 unsigned int val = *sha1++;
209 *buf++ = hex[val >> 4];
210 *buf++ = hex[val & 0xf];
211 }
a6080a0a 212
633f43e1 213 return *base;
bf592c50
DB
214}
215
633f43e1 216char *sha1_pack_name(const unsigned char *sha1)
bf592c50 217{
633f43e1 218 static char *name, *base;
bf592c50 219
6eec46bd 220 return sha1_get_pack_name(sha1, &name, &base, "pack");
633f43e1 221}
bf592c50 222
633f43e1
HO
223char *sha1_pack_index_name(const unsigned char *sha1)
224{
225 static char *name, *base;
a6080a0a 226
6eec46bd 227 return sha1_get_pack_name(sha1, &name, &base, "idx");
bf592c50
DB
228}
229
d5a63b99
JH
230struct alternate_object_database *alt_odb_list;
231static struct alternate_object_database **alt_odb_tail;
ace1534d 232
f4e51683 233static int git_open_noatime(const char *name);
c2f493a4 234
ddd5d056
JH
235/*
236 * Prepare alternate object database registry.
d5a63b99
JH
237 *
238 * The variable alt_odb_list points at the list of struct
239 * alternate_object_database. The elements on this list come from
240 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
241 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
242 * whose contents is similar to that environment variable but can be
243 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
244 * contains "/the/directory/corresponding/to/.git/objects/...", while
245 * its name points just after the slash at the end of ".git/objects/"
246 * in the example above, and has enough space to hold 40-byte hex
247 * SHA1, an extra slash for the first level indirection, and the
248 * terminating NUL.
ddd5d056 249 */
6eac50d8 250static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
ace1534d 251{
1494e038 252 const char *objdir = get_object_directory();
c2f493a4
MW
253 struct alternate_object_database *ent;
254 struct alternate_object_database *alt;
5bdf0a84
HW
255 int pfxlen, entlen;
256 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 257
85dadc38 258 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
259 strbuf_addstr(&pathbuf, real_path(relative_base));
260 strbuf_addch(&pathbuf, '/');
c2f493a4 261 }
6eac50d8 262 strbuf_addstr(&pathbuf, entry);
c2f493a4 263
5bdf0a84
HW
264 normalize_path_copy(pathbuf.buf, pathbuf.buf);
265
266 pfxlen = strlen(pathbuf.buf);
267
268 /*
269 * The trailing slash after the directory name is given by
270 * this function at the end. Remove duplicates.
271 */
272 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
273 pfxlen -= 1;
274
275 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
276 ent = xmalloc(sizeof(*ent) + entlen);
277 memcpy(ent->base, pathbuf.buf, pfxlen);
278 strbuf_release(&pathbuf);
c2f493a4
MW
279
280 ent->name = ent->base + pfxlen + 1;
281 ent->base[pfxlen + 3] = '/';
282 ent->base[pfxlen] = ent->base[entlen-1] = 0;
283
284 /* Detect cases where alternate disappeared */
90b4a71c 285 if (!is_directory(ent->base)) {
c2f493a4
MW
286 error("object directory %s does not exist; "
287 "check .git/objects/info/alternates.",
288 ent->base);
289 free(ent);
290 return -1;
291 }
292
293 /* Prevent the common mistake of listing the same
294 * thing twice, or object directory itself.
295 */
296 for (alt = alt_odb_list; alt; alt = alt->next) {
297 if (!memcmp(ent->base, alt->base, pfxlen)) {
298 free(ent);
299 return -1;
300 }
301 }
cb2912c3 302 if (!strcmp(ent->base, objdir)) {
c2f493a4
MW
303 free(ent);
304 return -1;
305 }
306
307 /* add the alternate entry */
308 *alt_odb_tail = ent;
309 alt_odb_tail = &(ent->next);
310 ent->next = NULL;
311
312 /* recursively add alternates */
313 read_info_alternates(ent->base, depth + 1);
314
315 ent->base[pfxlen] = '/';
316
317 return 0;
318}
319
c5950164 320static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
321 const char *relative_base, int depth)
322{
6eac50d8
MH
323 struct string_list entries = STRING_LIST_INIT_NODUP;
324 char *alt_copy;
325 int i;
c2f493a4
MW
326
327 if (depth > 5) {
328 error("%s: ignoring alternate object stores, nesting too deep.",
329 relative_base);
330 return;
331 }
332
c5950164 333 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
334 string_list_split_in_place(&entries, alt_copy, sep, -1);
335 for (i = 0; i < entries.nr; i++) {
336 const char *entry = entries.items[i].string;
337 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 338 continue;
6eac50d8
MH
339 if (!is_absolute_path(entry) && depth) {
340 error("%s: ignoring relative alternate object store %s",
341 relative_base, entry);
342 } else {
343 link_alt_odb_entry(entry, relative_base, depth);
9577e7e3 344 }
9577e7e3 345 }
6eac50d8
MH
346 string_list_clear(&entries, 0);
347 free(alt_copy);
d5a63b99
JH
348}
349
5e73633d 350void read_info_alternates(const char * relative_base, int depth)
d5a63b99 351{
9577e7e3 352 char *map;
dc49cd76 353 size_t mapsz;
d5a63b99 354 struct stat st;
9cb18f56
JM
355 const char alt_file_name[] = "info/alternates";
356 /* Given that relative_base is no longer than PATH_MAX,
357 ensure that "path" has enough space to append "/", the
358 file name, "info/alternates", and a trailing NUL. */
359 char path[PATH_MAX + 1 + sizeof alt_file_name];
c2f493a4 360 int fd;
d5a63b99 361
9cb18f56 362 sprintf(path, "%s/%s", relative_base, alt_file_name);
f4e51683 363 fd = git_open_noatime(path);
d5a63b99
JH
364 if (fd < 0)
365 return;
366 if (fstat(fd, &st) || (st.st_size == 0)) {
367 close(fd);
9a217f2a 368 return;
ace1534d 369 }
dc49cd76
SP
370 mapsz = xsize_t(st.st_size);
371 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 372 close(fd);
d5a63b99 373
c5950164 374 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 375
dc49cd76 376 munmap(map, mapsz);
ace1534d
JH
377}
378
bef70b22
DB
379void add_to_alternates_file(const char *reference)
380{
381 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
acd3b9ec 382 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
e6baf4a1 383 char *alt = mkpath("%s\n", reference);
bef70b22
DB
384 write_or_die(fd, alt, strlen(alt));
385 if (commit_lock_file(lock))
386 die("could not close alternates file");
387 if (alt_odb_tail)
c5950164 388 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
bef70b22
DB
389}
390
d79796bc
JH
391void foreach_alt_odb(alt_odb_fn fn, void *cb)
392{
393 struct alternate_object_database *ent;
394
395 prepare_alt_odb();
396 for (ent = alt_odb_list; ent; ent = ent->next)
397 if (fn(ent, cb))
398 return;
399}
400
c2f493a4
MW
401void prepare_alt_odb(void)
402{
554fe20d 403 const char *alt;
c2f493a4 404
7dc24aa5
SP
405 if (alt_odb_tail)
406 return;
407
c2f493a4
MW
408 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
409 if (!alt) alt = "";
410
c2f493a4 411 alt_odb_tail = &alt_odb_list;
c5950164 412 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
413
414 read_info_alternates(get_object_directory(), 0);
415}
416
0f4dc14a 417static int has_loose_object_local(const unsigned char *sha1)
ace1534d 418{
ace1534d 419 char *name = sha1_file_name(sha1);
0f4dc14a
BC
420 return !access(name, F_OK);
421}
ace1534d 422
0f4dc14a
BC
423int has_loose_object_nonlocal(const unsigned char *sha1)
424{
425 struct alternate_object_database *alt;
9a217f2a 426 prepare_alt_odb();
d5a63b99 427 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 428 fill_sha1_path(alt->name, sha1);
c529d75a
LT
429 if (!access(alt->base, F_OK))
430 return 1;
ace1534d 431 }
c529d75a 432 return 0;
ace1534d
JH
433}
434
0f4dc14a
BC
435static int has_loose_object(const unsigned char *sha1)
436{
437 return has_loose_object_local(sha1) ||
438 has_loose_object_nonlocal(sha1);
439}
440
60bb8b14 441static unsigned int pack_used_ctr;
a53128b6
SP
442static unsigned int pack_mmap_calls;
443static unsigned int peak_pack_open_windows;
444static unsigned int pack_open_windows;
c7934306
SP
445static unsigned int pack_open_fds;
446static unsigned int pack_max_fds;
a53128b6 447static size_t peak_pack_mapped;
60bb8b14 448static size_t pack_mapped;
9a217f2a 449struct packed_git *packed_git;
1f688557 450
b79d18c9 451void pack_report(void)
a53128b6
SP
452{
453 fprintf(stderr,
e05db0fd
PR
454 "pack_report: getpagesize() = %10" SZ_FMT "\n"
455 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
456 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
457 sz_fmt(getpagesize()),
458 sz_fmt(packed_git_window_size),
459 sz_fmt(packed_git_limit));
a53128b6
SP
460 fprintf(stderr,
461 "pack_report: pack_used_ctr = %10u\n"
462 "pack_report: pack_mmap_calls = %10u\n"
463 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
464 "pack_report: pack_mapped = "
465 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
466 pack_used_ctr,
467 pack_mmap_calls,
468 pack_open_windows, peak_pack_open_windows,
9e42d6a1 469 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
470}
471
42873078 472static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
473{
474 void *idx_map;
42873078 475 struct pack_idx_header *hdr;
dc49cd76 476 size_t idx_size;
74e34e1f 477 uint32_t version, nr, i, *index;
f4e51683 478 int fd = git_open_noatime(path);
1f688557 479 struct stat st;
42873078 480
1f688557
JH
481 if (fd < 0)
482 return -1;
483 if (fstat(fd, &st)) {
484 close(fd);
485 return -1;
486 }
dc49cd76 487 idx_size = xsize_t(st.st_size);
2d88451b
SP
488 if (idx_size < 4 * 256 + 20 + 20) {
489 close(fd);
490 return error("index file %s is too small", path);
491 }
c4712e45 492 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 493 close(fd);
1f688557 494
42873078
NP
495 hdr = idx_map;
496 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
497 version = ntohl(hdr->idx_version);
498 if (version < 2 || version > 2) {
499 munmap(idx_map, idx_size);
6e1c2344 500 return error("index file %s is version %"PRIu32
74e34e1f
NP
501 " and is not supported by this binary"
502 " (try upgrading GIT to a newer version)",
503 path, version);
504 }
505 } else
506 version = 1;
df1b059d 507
1f688557 508 nr = 0;
42873078 509 index = idx_map;
74e34e1f
NP
510 if (version > 1)
511 index += 2; /* skip index header */
1f688557 512 for (i = 0; i < 256; i++) {
326bf396 513 uint32_t n = ntohl(index[i]);
2d88451b
SP
514 if (n < nr) {
515 munmap(idx_map, idx_size);
df1b059d 516 return error("non-monotonic index %s", path);
2d88451b 517 }
1f688557
JH
518 nr = n;
519 }
520
74e34e1f
NP
521 if (version == 1) {
522 /*
523 * Total size:
524 * - 256 index entries 4 bytes each
525 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
526 * - 20-byte SHA1 of the packfile
527 * - 20-byte SHA1 file checksum
528 */
529 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
530 munmap(idx_map, idx_size);
eef427a0 531 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
532 }
533 } else if (version == 2) {
534 /*
535 * Minimum size:
536 * - 8 bytes of header
537 * - 256 index entries 4 bytes each
538 * - 20-byte sha1 entry * nr
539 * - 4-byte crc entry * nr
540 * - 4-byte offset entry * nr
541 * - 20-byte SHA1 of the packfile
542 * - 20-byte SHA1 file checksum
543 * And after the 4-byte offset table might be a
544 * variable sized table containing 8-byte entries
545 * for offsets larger than 2^31.
546 */
547 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
548 unsigned long max_size = min_size;
549 if (nr)
550 max_size += (nr - 1)*8;
551 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 552 munmap(idx_map, idx_size);
eef427a0 553 return error("wrong index v2 file size in %s", path);
74e34e1f 554 }
7109c889
JH
555 if (idx_size != min_size &&
556 /*
557 * make sure we can deal with large pack offsets.
558 * 31-bit signed offset won't be enough, neither
559 * 32-bit unsigned one will be.
560 */
561 (sizeof(off_t) <= 4)) {
562 munmap(idx_map, idx_size);
563 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 564 }
2d88451b 565 }
1f688557 566
74e34e1f 567 p->index_version = version;
42873078
NP
568 p->index_data = idx_map;
569 p->index_size = idx_size;
57059091 570 p->num_objects = nr;
1f688557
JH
571 return 0;
572}
573
bc8e478a 574int open_pack_index(struct packed_git *p)
d079837e
SP
575{
576 char *idx_name;
577 int ret;
578
579 if (p->index_data)
580 return 0;
581
582 idx_name = xstrdup(p->pack_name);
583 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
584 ret = check_packed_git_idx(idx_name, p);
585 free(idx_name);
586 return ret;
587}
588
11daf39b
SP
589static void scan_windows(struct packed_git *p,
590 struct packed_git **lru_p,
591 struct pack_window **lru_w,
592 struct pack_window **lru_l)
1f688557 593{
11daf39b
SP
594 struct pack_window *w, *w_l;
595
596 for (w_l = NULL, w = p->windows; w; w = w->next) {
597 if (!w->inuse_cnt) {
598 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
599 *lru_p = p;
600 *lru_w = w;
601 *lru_l = w_l;
54044bf8 602 }
54044bf8 603 }
11daf39b 604 w_l = w;
f9253394 605 }
11daf39b
SP
606}
607
d1efefa4 608static int unuse_one_window(struct packed_git *current, int keep_fd)
11daf39b
SP
609{
610 struct packed_git *p, *lru_p = NULL;
611 struct pack_window *lru_w = NULL, *lru_l = NULL;
612
613 if (current)
614 scan_windows(current, &lru_p, &lru_w, &lru_l);
615 for (p = packed_git; p; p = p->next)
616 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
617 if (lru_p) {
618 munmap(lru_w->base, lru_w->len);
619 pack_mapped -= lru_w->len;
620 if (lru_l)
621 lru_l->next = lru_w->next;
622 else {
623 lru_p->windows = lru_w->next;
d131b7af
SP
624 if (!lru_p->windows && lru_p->pack_fd != -1
625 && lru_p->pack_fd != keep_fd) {
54044bf8 626 close(lru_p->pack_fd);
c7934306 627 pack_open_fds--;
54044bf8
SP
628 lru_p->pack_fd = -1;
629 }
630 }
631 free(lru_w);
a53128b6 632 pack_open_windows--;
54044bf8
SP
633 return 1;
634 }
635 return 0;
f9253394
JH
636}
637
d1efefa4 638void release_pack_memory(size_t need, int fd)
97bfeb34
SP
639{
640 size_t cur = pack_mapped;
d1efefa4 641 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
97bfeb34
SP
642 ; /* nothing */
643}
644
58ecbd5e
JN
645void *xmmap(void *start, size_t length,
646 int prot, int flags, int fd, off_t offset)
647{
648 void *ret = mmap(start, length, prot, flags, fd, offset);
649 if (ret == MAP_FAILED) {
650 if (!length)
651 return NULL;
652 release_pack_memory(length, fd);
653 ret = mmap(start, length, prot, flags, fd, offset);
654 if (ret == MAP_FAILED)
655 die_errno("Out of memory? mmap failed");
656 }
657 return ret;
658}
659
c9ced051
SP
660void close_pack_windows(struct packed_git *p)
661{
662 while (p->windows) {
663 struct pack_window *w = p->windows;
664
665 if (w->inuse_cnt)
666 die("pack '%s' still has open windows to it",
667 p->pack_name);
668 munmap(w->base, w->len);
669 pack_mapped -= w->len;
670 pack_open_windows--;
671 p->windows = w->next;
672 free(w);
673 }
674}
675
03e79c88 676void unuse_pack(struct pack_window **w_cursor)
f9253394 677{
03e79c88
SP
678 struct pack_window *w = *w_cursor;
679 if (w) {
680 w->inuse_cnt--;
681 *w_cursor = NULL;
682 }
1f688557
JH
683}
684
fa5fc15d
SP
685void close_pack_index(struct packed_git *p)
686{
687 if (p->index_data) {
688 munmap((void *)p->index_data, p->index_size);
689 p->index_data = NULL;
690 }
691}
692
c74faea1
NP
693/*
694 * This is used by git-repack in case a newly created pack happens to
695 * contain the same set of objects as an existing one. In that case
696 * the resulting file might be different even if its name would be the
697 * same. It is best to close any reference to the old pack before it is
698 * replaced on disk. Of course no index pointers nor windows for given pack
699 * must subsist at this point. If ever objects from this pack are requested
700 * again, the new version of the pack will be reinitialized through
701 * reprepare_packed_git().
702 */
703void free_pack_by_name(const char *pack_name)
704{
705 struct packed_git *p, **pp = &packed_git;
706
707 while (*pp) {
708 p = *pp;
709 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 710 clear_delta_base_cache();
c74faea1 711 close_pack_windows(p);
c7934306 712 if (p->pack_fd != -1) {
c74faea1 713 close(p->pack_fd);
c7934306
SP
714 pack_open_fds--;
715 }
fa5fc15d 716 close_pack_index(p);
c74faea1
NP
717 free(p->bad_object_sha1);
718 *pp = p->next;
c01f51cc
NTND
719 if (last_found_pack == p)
720 last_found_pack = NULL;
c74faea1
NP
721 free(p);
722 return;
723 }
724 pp = &p->next;
725 }
726}
727
a0788266
JS
728static unsigned int get_max_fd_limit(void)
729{
730#ifdef RLIMIT_NOFILE
731 struct rlimit lim;
732
733 if (getrlimit(RLIMIT_NOFILE, &lim))
734 die_errno("cannot get RLIMIT_NOFILE");
735
736 return lim.rlim_cur;
737#elif defined(_SC_OPEN_MAX)
738 return sysconf(_SC_OPEN_MAX);
739#elif defined(OPEN_MAX)
740 return OPEN_MAX;
741#else
742 return 1; /* see the caller ;-) */
743#endif
744}
745
3cf8b462
SP
746/*
747 * Do not call this directly as this leaks p->pack_fd on error return;
748 * call open_packed_git() instead.
749 */
750static int open_packed_git_1(struct packed_git *p)
1f688557 751{
9bc879c1
SP
752 struct stat st;
753 struct pack_header hdr;
754 unsigned char sha1[20];
755 unsigned char *idx_sha1;
2c039da8 756 long fd_flag;
9bc879c1 757
d079837e
SP
758 if (!p->index_data && open_pack_index(p))
759 return error("packfile %s index unavailable", p->pack_name);
760
c7934306 761 if (!pack_max_fds) {
a0788266 762 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
763
764 /* Save 3 for stdin/stdout/stderr, 22 for work */
765 if (25 < max_fds)
766 pack_max_fds = max_fds - 25;
767 else
768 pack_max_fds = 1;
769 }
770
771 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
772 ; /* nothing */
773
f4e51683 774 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 775 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 776 return -1;
c7934306 777 pack_open_fds++;
9bc879c1
SP
778
779 /* If we created the struct before we had the pack we lack size. */
bf592c50 780 if (!p->pack_size) {
bf592c50 781 if (!S_ISREG(st.st_mode))
072db278 782 return error("packfile %s not a regular file", p->pack_name);
bf592c50 783 p->pack_size = st.st_size;
9bc879c1 784 } else if (p->pack_size != st.st_size)
072db278 785 return error("packfile %s size changed", p->pack_name);
9bc879c1 786
2c039da8
JH
787 /* We leave these file descriptors open with sliding mmap;
788 * there is no point keeping them open across exec(), though.
789 */
790 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
791 if (fd_flag < 0)
072db278 792 return error("cannot determine file descriptor flags");
2c039da8
JH
793 fd_flag |= FD_CLOEXEC;
794 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 795 return error("cannot set FD_CLOEXEC");
2c039da8 796
9bc879c1 797 /* Verify we recognize this pack file format. */
e6e2bd62 798 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 799 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 800 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 801 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 802 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
803 return error("packfile %s is version %"PRIu32" and not"
804 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
805 p->pack_name, ntohl(hdr.hdr_version));
806
807 /* Verify the pack matches its index. */
57059091 808 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
809 return error("packfile %s claims to have %"PRIu32" objects"
810 " while index indicates %"PRIu32" objects",
57059091
NP
811 p->pack_name, ntohl(hdr.hdr_entries),
812 p->num_objects);
9bc879c1 813 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 814 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 815 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 816 return error("packfile %s signature is unavailable", p->pack_name);
42873078 817 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 818 if (hashcmp(sha1, idx_sha1))
072db278
SP
819 return error("packfile %s does not match index", p->pack_name);
820 return 0;
9bc879c1
SP
821}
822
3cf8b462
SP
823static int open_packed_git(struct packed_git *p)
824{
825 if (!open_packed_git_1(p))
826 return 0;
827 if (p->pack_fd != -1) {
828 close(p->pack_fd);
c7934306 829 pack_open_fds--;
3cf8b462
SP
830 p->pack_fd = -1;
831 }
832 return -1;
833}
834
c4001d92 835static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
836{
837 /* We must promise at least 20 bytes (one hash) after the
838 * offset is available from this window, otherwise the offset
839 * is not actually in this window and a different window (which
840 * has that one hash excess) must be used. This is to support
841 * the object header and delta base parsing routines below.
842 */
843 off_t win_off = win->offset;
844 return win_off <= offset
845 && (offset + 20) <= (win_off + win->len);
846}
847
4b25d091 848unsigned char *use_pack(struct packed_git *p,
03e79c88 849 struct pack_window **w_cursor,
c4001d92 850 off_t offset,
ef49a7a0 851 unsigned long *left)
9bc879c1 852{
60bb8b14 853 struct pack_window *win = *w_cursor;
03e79c88 854
a9d98a14 855 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
856 * pointless to ask for an offset into the middle of that
857 * hash, and the in_window function above wouldn't match
858 * don't allow an offset too close to the end of the file.
859 */
d131b7af
SP
860 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
861 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
862 if (offset > (p->pack_size - 20))
863 die("offset beyond end of packfile (truncated pack?)");
864
865 if (!win || !in_window(win, offset)) {
866 if (win)
867 win->inuse_cnt--;
868 for (win = p->windows; win; win = win->next) {
869 if (in_window(win, offset))
870 break;
871 }
872 if (!win) {
78a28df9 873 size_t window_align = packed_git_window_size / 2;
dc49cd76 874 off_t len;
d131b7af
SP
875
876 if (p->pack_fd == -1 && open_packed_git(p))
877 die("packfile %s cannot be accessed", p->pack_name);
878
60bb8b14 879 win = xcalloc(1, sizeof(*win));
78a28df9 880 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
881 len = p->pack_size - win->offset;
882 if (len > packed_git_window_size)
883 len = packed_git_window_size;
884 win->len = (size_t)len;
60bb8b14 885 pack_mapped += win->len;
11daf39b 886 while (packed_git_limit < pack_mapped
d1efefa4 887 && unuse_one_window(p, p->pack_fd))
60bb8b14 888 ; /* nothing */
c4712e45 889 win->base = xmmap(NULL, win->len,
60bb8b14
SP
890 PROT_READ, MAP_PRIVATE,
891 p->pack_fd, win->offset);
892 if (win->base == MAP_FAILED)
73b4e4be
SP
893 die("packfile %s cannot be mapped: %s",
894 p->pack_name,
895 strerror(errno));
d131b7af
SP
896 if (!win->offset && win->len == p->pack_size
897 && !p->do_not_close) {
898 close(p->pack_fd);
899 pack_open_fds--;
900 p->pack_fd = -1;
901 }
a53128b6
SP
902 pack_mmap_calls++;
903 pack_open_windows++;
904 if (pack_mapped > peak_pack_mapped)
905 peak_pack_mapped = pack_mapped;
906 if (pack_open_windows > peak_pack_open_windows)
907 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
908 win->next = p->windows;
909 p->windows = win;
910 }
1f688557 911 }
03e79c88
SP
912 if (win != *w_cursor) {
913 win->last_used = pack_used_ctr++;
914 win->inuse_cnt++;
915 *w_cursor = win;
916 }
60bb8b14 917 offset -= win->offset;
03e79c88 918 if (left)
dc49cd76 919 *left = win->len - xsize_t(offset);
03e79c88 920 return win->base + offset;
1f688557
JH
921}
922
27d69a46
NP
923static struct packed_git *alloc_packed_git(int extra)
924{
925 struct packed_git *p = xmalloc(sizeof(*p) + extra);
926 memset(p, 0, sizeof(*p));
927 p->pack_fd = -1;
928 return p;
929}
930
e0500293
JN
931static void try_to_free_pack_memory(size_t size)
932{
933 release_pack_memory(size, -1);
934}
935
42873078 936struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 937{
e0500293 938 static int have_set_try_to_free_routine;
1f688557 939 struct stat st;
27d69a46 940 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 941
e0500293
JN
942 if (!have_set_try_to_free_routine) {
943 have_set_try_to_free_routine = 1;
944 set_try_to_free_routine(try_to_free_pack_memory);
945 }
946
42873078
NP
947 /*
948 * Make sure a corresponding .pack file exists and that
949 * the index looks sane.
950 */
951 path_len -= strlen(".idx");
27d69a46
NP
952 if (path_len < 1) {
953 free(p);
1f688557 954 return NULL;
27d69a46 955 }
42873078 956 memcpy(p->pack_name, path, path_len);
8d25931d
BC
957
958 strcpy(p->pack_name + path_len, ".keep");
959 if (!access(p->pack_name, F_OK))
960 p->pack_keep = 1;
961
42873078 962 strcpy(p->pack_name + path_len, ".pack");
d079837e 963 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 964 free(p);
1f688557
JH
965 return NULL;
966 }
42873078 967
1f688557
JH
968 /* ok, it looks sane as far as we can check without
969 * actually mapping the pack file.
970 */
1f688557 971 p->pack_size = st.st_size;
9d835df2 972 p->pack_local = local;
b867092f 973 p->mtime = st.st_mtime;
42873078
NP
974 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
975 hashclr(p->sha1);
1f688557
JH
976 return p;
977}
978
7b64469a 979struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 980{
42873078 981 const char *path = sha1_pack_name(sha1);
27d69a46 982 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 983
27d69a46
NP
984 strcpy(p->pack_name, path);
985 hashcpy(p->sha1, sha1);
42873078
NP
986 if (check_packed_git_idx(idx_path, p)) {
987 free(p);
bf592c50 988 return NULL;
42873078 989 }
bf592c50 990
bf592c50
DB
991 return p;
992}
993
994void install_packed_git(struct packed_git *pack)
995{
c7934306
SP
996 if (pack->pack_fd != -1)
997 pack_open_fds++;
998
bf592c50
DB
999 pack->next = packed_git;
1000 packed_git = pack;
1001}
1002
9d835df2 1003static void prepare_packed_git_one(char *objdir, int local)
1f688557 1004{
9cb18f56
JM
1005 /* Ensure that this buffer is large enough so that we can
1006 append "/pack/" without clobbering the stack even if
1007 strlen(objdir) were PATH_MAX. */
1008 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
1009 int len;
1010 DIR *dir;
1011 struct dirent *de;
1012
1013 sprintf(path, "%s/pack", objdir);
1014 len = strlen(path);
1015 dir = opendir(path);
b5b16990 1016 if (!dir) {
26125f6b 1017 if (errno != ENOENT)
bd2afde8 1018 error("unable to open object pack directory: %s: %s",
26125f6b 1019 path, strerror(errno));
1f688557 1020 return;
b5b16990 1021 }
1f688557
JH
1022 path[len++] = '/';
1023 while ((de = readdir(dir)) != NULL) {
1024 int namelen = strlen(de->d_name);
1025 struct packed_git *p;
1026
5bb1cda5 1027 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
1028 continue;
1029
9cb18f56
JM
1030 if (len + namelen + 1 > sizeof(path))
1031 continue;
1032
54a15a8d 1033 /* Don't reopen a pack we already have. */
1f688557 1034 strcpy(path + len, de->d_name);
86f7780c
JK
1035 for (p = packed_git; p; p = p->next) {
1036 if (!memcmp(path, p->pack_name, len + namelen - 4))
1037 break;
1038 }
1039 if (p)
1040 continue;
54a15a8d
SP
1041 /* See if it really is a valid .idx file with corresponding
1042 * .pack file that we can map.
1043 */
9d835df2 1044 p = add_packed_git(path, len + namelen, local);
1f688557
JH
1045 if (!p)
1046 continue;
625e9421 1047 install_packed_git(p);
1f688557 1048 }
5b35bcd5 1049 closedir(dir);
1f688557
JH
1050}
1051
b867092f
JH
1052static int sort_pack(const void *a_, const void *b_)
1053{
1054 struct packed_git *a = *((struct packed_git **)a_);
1055 struct packed_git *b = *((struct packed_git **)b_);
1056 int st;
1057
1058 /*
1059 * Local packs tend to contain objects specific to our
1060 * variant of the project than remote ones. In addition,
1061 * remote ones could be on a network mounted filesystem.
1062 * Favor local ones for these reasons.
1063 */
1064 st = a->pack_local - b->pack_local;
1065 if (st)
1066 return -st;
1067
1068 /*
1069 * Younger packs tend to contain more recent objects,
1070 * and more recent objects tend to get accessed more
1071 * often.
1072 */
1073 if (a->mtime < b->mtime)
1074 return 1;
1075 else if (a->mtime == b->mtime)
1076 return 0;
1077 return -1;
1078}
1079
1080static void rearrange_packed_git(void)
1081{
1082 struct packed_git **ary, *p;
1083 int i, n;
1084
1085 for (n = 0, p = packed_git; p; p = p->next)
1086 n++;
1087 if (n < 2)
1088 return;
1089
1090 /* prepare an array of packed_git for easier sorting */
1091 ary = xcalloc(n, sizeof(struct packed_git *));
1092 for (n = 0, p = packed_git; p; p = p->next)
1093 ary[n++] = p;
1094
1095 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1096
1097 /* link them back again */
1098 for (i = 0; i < n - 1; i++)
1099 ary[i]->next = ary[i + 1];
1100 ary[n - 1]->next = NULL;
1101 packed_git = ary[0];
1102
1103 free(ary);
1104}
1105
637cdd9d 1106static int prepare_packed_git_run_once = 0;
9a217f2a 1107void prepare_packed_git(void)
1f688557 1108{
d5a63b99 1109 struct alternate_object_database *alt;
1f688557 1110
637cdd9d 1111 if (prepare_packed_git_run_once)
1f688557 1112 return;
9d835df2 1113 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1114 prepare_alt_odb();
d5a63b99 1115 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1116 alt->name[-1] = 0;
9d835df2 1117 prepare_packed_git_one(alt->base, 0);
1494e038 1118 alt->name[-1] = '/';
1f688557 1119 }
b867092f 1120 rearrange_packed_git();
637cdd9d
JK
1121 prepare_packed_git_run_once = 1;
1122}
1123
fc04c412 1124void reprepare_packed_git(void)
637cdd9d 1125{
4b480c67 1126 discard_revindex();
637cdd9d
JK
1127 prepare_packed_git_run_once = 0;
1128 prepare_packed_git();
1f688557
JH
1129}
1130
8eca0b47
NP
1131static void mark_bad_packed_object(struct packed_git *p,
1132 const unsigned char *sha1)
1133{
1134 unsigned i;
1135 for (i = 0; i < p->num_bad_objects; i++)
1136 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1137 return;
1138 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1139 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1140 p->num_bad_objects++;
1141}
1142
b6c4cecc 1143static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1144{
1145 struct packed_git *p;
1146 unsigned i;
1147
1148 for (p = packed_git; p; p = p->next)
1149 for (i = 0; i < p->num_bad_objects; i++)
1150 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1151 return p;
1152 return NULL;
ac939109
NP
1153}
1154
090ea126
NTND
1155/*
1156 * With an in-core object data in "map", rehash it to make sure the
1157 * object name actually matches "sha1" to detect object corruption.
1158 * With "map" == NULL, try reading the object named with "sha1" using
1159 * the streaming interface and rehash it to do the same.
1160 */
1161int check_sha1_signature(const unsigned char *sha1, void *map,
1162 unsigned long size, const char *type)
0fcfd160
LT
1163{
1164 unsigned char real_sha1[20];
090ea126
NTND
1165 enum object_type obj_type;
1166 struct git_istream *st;
1167 git_SHA_CTX c;
1168 char hdr[32];
1169 int hdrlen;
1170
1171 if (map) {
1172 hash_sha1_file(map, size, type, real_sha1);
1173 return hashcmp(sha1, real_sha1) ? -1 : 0;
1174 }
1175
1176 st = open_istream(sha1, &obj_type, &size, NULL);
1177 if (!st)
1178 return -1;
1179
1180 /* Generate the header */
1181 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1182
1183 /* Sha1.. */
1184 git_SHA1_Init(&c);
1185 git_SHA1_Update(&c, hdr, hdrlen);
1186 for (;;) {
1187 char buf[1024 * 16];
1188 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1189
1190 if (!readlen)
1191 break;
1192 git_SHA1_Update(&c, buf, readlen);
1193 }
1194 git_SHA1_Final(real_sha1, &c);
1195 close_istream(st);
a89fccd2 1196 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1197}
1198
f4e51683 1199static int git_open_noatime(const char *name)
44d1c19e
LT
1200{
1201 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1202
f2e872aa
SP
1203 for (;;) {
1204 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1205 if (fd >= 0)
f2e872aa
SP
1206 return fd;
1207
f2e872aa
SP
1208 /* Might the failure be due to O_NOATIME? */
1209 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1210 sha1_file_open_flag = 0;
f2e872aa
SP
1211 continue;
1212 }
1213
1214 return -1;
44d1c19e 1215 }
44d1c19e
LT
1216}
1217
1218static int open_sha1_file(const unsigned char *sha1)
1219{
1220 int fd;
1221 char *name = sha1_file_name(sha1);
1222 struct alternate_object_database *alt;
1223
f4e51683 1224 fd = git_open_noatime(name);
44d1c19e
LT
1225 if (fd >= 0)
1226 return fd;
1227
1228 prepare_alt_odb();
1229 errno = ENOENT;
1230 for (alt = alt_odb_list; alt; alt = alt->next) {
1231 name = alt->name;
1232 fill_sha1_path(name, sha1);
f4e51683 1233 fd = git_open_noatime(alt->base);
44d1c19e
LT
1234 if (fd >= 0)
1235 return fd;
1236 }
1237 return -1;
1238}
1239
f0270efd 1240void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1241{
0fcfd160 1242 void *map;
144bde78 1243 int fd;
ace1534d 1244
44d1c19e
LT
1245 fd = open_sha1_file(sha1);
1246 map = NULL;
1247 if (fd >= 0) {
1248 struct stat st;
0fcfd160 1249
44d1c19e
LT
1250 if (!fstat(fd, &st)) {
1251 *size = xsize_t(st.st_size);
33e42de0
MM
1252 if (!*size) {
1253 /* mmap() is forbidden on empty files */
1254 error("object file %s is empty", sha1_file_name(sha1));
1255 return NULL;
1256 }
44d1c19e 1257 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1258 }
44d1c19e 1259 close(fd);
0fcfd160 1260 }
0fcfd160
LT
1261 return map;
1262}
1263
cc5c54e7
JH
1264/*
1265 * There used to be a second loose object header format which
1266 * was meant to mimic the in-pack format, allowing for direct
1267 * copy of the object data. This format turned up not to be
1268 * really worth it and we no longer write loose objects in that
1269 * format.
1270 */
1271static int experimental_loose_object(unsigned char *map)
bb6b8e4f
JH
1272{
1273 unsigned int word;
1274
1275 /*
7f684a2a
RT
1276 * We must determine if the buffer contains the standard
1277 * zlib-deflated stream or the experimental format based
1278 * on the in-pack object format. Compare the header byte
1279 * for each format:
1280 *
1281 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1282 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1283 *
1284 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1285 * in standard loose-object format, UNLESS it is a Git-pack
1286 * format object *exactly* 8 bytes in size when inflated.
1287 *
1288 * However, RFC1950 also specifies that the 1st 16-bit word
1289 * must be divisible by 31 - this checksum tells us our buffer
1290 * is in the standard format, giving a false positive only if
1291 * the 1st word of the Git-pack format object happens to be
1292 * divisible by 31, ie:
1293 * ((byte0 * 256) + byte1) % 31 = 0
1294 * => 0ttt10000www1000 % 31 = 0
1295 *
1296 * As it happens, this case can only arise for www=3 & ttt=1
1297 * - ie, a Commit object, which would have to be 8 bytes in
1298 * size. As no Commit can be that small, we find that the
1299 * combination of these two criteria (bitmask & checksum)
1300 * can always correctly determine the buffer format.
bb6b8e4f
JH
1301 */
1302 word = (map[0] << 8) + map[1];
7f684a2a 1303 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
bb6b8e4f 1304 return 0;
cc5c54e7
JH
1305 else
1306 return 1;
bb6b8e4f
JH
1307}
1308
09ded04b
NP
1309unsigned long unpack_object_header_buffer(const unsigned char *buf,
1310 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1311{
ad1ed5ee 1312 unsigned shift;
48fb7deb 1313 unsigned long size, c;
ad1ed5ee
JH
1314 unsigned long used = 0;
1315
1316 c = buf[used++];
1317 *type = (c >> 4) & 7;
1318 size = c & 15;
1319 shift = 4;
1320 while (c & 0x80) {
f630cfda 1321 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1322 error("bad object header");
ea4f9685
JH
1323 size = used = 0;
1324 break;
09ded04b 1325 }
ad1ed5ee
JH
1326 c = buf[used++];
1327 size += (c & 0x7f) << shift;
1328 shift += 7;
1329 }
1330 *sizep = size;
1331 return used;
1332}
1333
eb4f4076 1334int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee
JH
1335{
1336 unsigned long size, used;
1337 static const char valid_loose_object_type[8] = {
1338 0, /* OBJ_EXT */
1339 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1340 0, /* "delta" and others are invalid in a loose object */
93821bd9 1341 };
ad1ed5ee 1342 enum object_type type;
93821bd9 1343
c4483576
LT
1344 /* Get the data stream */
1345 memset(stream, 0, sizeof(*stream));
1346 stream->next_in = map;
1347 stream->avail_in = mapsize;
1348 stream->next_out = buffer;
93821bd9
LT
1349 stream->avail_out = bufsiz;
1350
cc5c54e7
JH
1351 if (experimental_loose_object(map)) {
1352 /*
1353 * The old experimental format we no longer produce;
1354 * we can still read it.
1355 */
1356 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1357 if (!used || !valid_loose_object_type[type])
1358 return -1;
1359 map += used;
1360 mapsize -= used;
726f852b 1361
cc5c54e7
JH
1362 /* Set up the stream for the rest.. */
1363 stream->next_in = map;
1364 stream->avail_in = mapsize;
1365 git_inflate_init(stream);
93821bd9 1366
cc5c54e7
JH
1367 /* And generate the fake traditional header */
1368 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1369 typename(type), size);
1370 return 0;
1371 }
39c68542 1372 git_inflate_init(stream);
cc5c54e7 1373 return git_inflate(stream, 0);
c4483576
LT
1374}
1375
ef49a7a0 1376static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1377{
1378 int bytes = strlen(buffer) + 1;
3aee68aa 1379 unsigned char *buf = xmallocz(size);
93821bd9 1380 unsigned long n;
7efbff75 1381 int status = Z_OK;
5180cacc 1382
93821bd9
LT
1383 n = stream->total_out - bytes;
1384 if (n > size)
1385 n = size;
1386 memcpy(buf, (char *) buffer + bytes, n);
1387 bytes = n;
456cdf6e
LT
1388 if (bytes <= size) {
1389 /*
1390 * The above condition must be (bytes <= size), not
1391 * (bytes < size). In other words, even though we
ccf5ace0 1392 * expect no more output and set avail_out to zero,
456cdf6e
LT
1393 * the input zlib stream may have bytes that express
1394 * "this concludes the stream", and we *do* want to
1395 * eat that input.
1396 *
1397 * Otherwise we would not be able to test that we
1398 * consumed all the input to reach the expected size;
1399 * we also want to check that zlib tells us that all
1400 * went well with status == Z_STREAM_END at the end.
1401 */
5180cacc
LT
1402 stream->next_out = buf + bytes;
1403 stream->avail_out = size - bytes;
7efbff75 1404 while (status == Z_OK)
39c68542 1405 status = git_inflate(stream, Z_FINISH);
5180cacc 1406 }
456cdf6e 1407 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1408 git_inflate_end(stream);
7efbff75
JH
1409 return buf;
1410 }
1411
1412 if (status < 0)
1413 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1414 else if (stream->avail_in)
1415 error("garbage at end of loose object '%s'",
1416 sha1_to_hex(sha1));
1417 free(buf);
1418 return NULL;
5180cacc
LT
1419}
1420
1421/*
1422 * We used to just use "sscanf()", but that's actually way
1423 * too permissive for what we want to check. So do an anal
1424 * object header parse by hand.
1425 */
f0270efd 1426int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1427{
21666f1a 1428 char type[10];
5180cacc
LT
1429 int i;
1430 unsigned long size;
1431
1432 /*
a6080a0a 1433 * The type can be at most ten bytes (including the
5180cacc 1434 * terminating '\0' that we add), and is followed by
21666f1a 1435 * a space.
5180cacc 1436 */
21666f1a 1437 i = 0;
5180cacc
LT
1438 for (;;) {
1439 char c = *hdr++;
1440 if (c == ' ')
1441 break;
21666f1a
NP
1442 type[i++] = c;
1443 if (i >= sizeof(type))
5180cacc 1444 return -1;
5180cacc 1445 }
21666f1a 1446 type[i] = 0;
5180cacc
LT
1447
1448 /*
1449 * The length must follow immediately, and be in canonical
1450 * decimal format (ie "010" is not valid).
1451 */
1452 size = *hdr++ - '0';
1453 if (size > 9)
1454 return -1;
1455 if (size) {
1456 for (;;) {
1457 unsigned long c = *hdr - '0';
1458 if (c > 9)
1459 break;
1460 hdr++;
1461 size = size * 10 + c;
1462 }
1463 }
1464 *sizep = size;
1465
1466 /*
1467 * The length must be followed by a zero byte
1468 */
21666f1a 1469 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1470}
1471
7efbff75 1472static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1473{
5180cacc 1474 int ret;
ef49a7a0 1475 git_zstream stream;
5180cacc 1476 char hdr[8192];
0fcfd160 1477
5180cacc 1478 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1479 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1480 return NULL;
1481
7efbff75 1482 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1483}
1484
54dab52a
NP
1485unsigned long get_size_from_delta(struct packed_git *p,
1486 struct pack_window **w_curs,
1487 off_t curpos)
1488{
1489 const unsigned char *data;
1490 unsigned char delta_head[20], *in;
ef49a7a0 1491 git_zstream stream;
54dab52a
NP
1492 int st;
1493
1494 memset(&stream, 0, sizeof(stream));
1495 stream.next_out = delta_head;
1496 stream.avail_out = sizeof(delta_head);
1497
39c68542 1498 git_inflate_init(&stream);
54dab52a
NP
1499 do {
1500 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1501 stream.next_in = in;
39c68542 1502 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1503 curpos += stream.next_in - in;
1504 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1505 stream.total_out < sizeof(delta_head));
39c68542 1506 git_inflate_end(&stream);
3d77d877
NP
1507 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1508 error("delta data unpack-initial failed");
1509 return 0;
1510 }
54dab52a
NP
1511
1512 /* Examine the initial part of the delta to figure out
1513 * the result size.
1514 */
1515 data = delta_head;
1516
1517 /* ignore base size */
1518 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1519
1520 /* Read the result size */
1521 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1522}
1523
c4001d92 1524static off_t get_delta_base(struct packed_git *p,
03e79c88 1525 struct pack_window **w_curs,
c4001d92 1526 off_t *curpos,
21666f1a 1527 enum object_type type,
c4001d92 1528 off_t delta_obj_offset)
eb32d236 1529{
2b87c45b 1530 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1531 off_t base_offset;
eb32d236 1532
8d8a4ea5
SP
1533 /* use_pack() assured us we have [base_info, base_info + 20)
1534 * as a range that we can look at without walking off the
1535 * end of the mapped window. Its actually the hash size
1536 * that is assured. An OFS_DELTA longer than the hash size
1537 * is stupid, as then a REF_DELTA would be smaller to store.
1538 */
21666f1a 1539 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1540 unsigned used = 0;
1541 unsigned char c = base_info[used++];
1542 base_offset = c & 127;
1543 while (c & 128) {
1544 base_offset += 1;
8723f216 1545 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1546 return 0; /* overflow */
eb32d236
NP
1547 c = base_info[used++];
1548 base_offset = (base_offset << 7) + (c & 127);
1549 }
1550 base_offset = delta_obj_offset - base_offset;
d8f32556 1551 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1552 return 0; /* out of bound */
2b87c45b 1553 *curpos += used;
21666f1a 1554 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1555 /* The base entry _must_ be in the same pack */
1556 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1557 *curpos += 20;
eb32d236
NP
1558 } else
1559 die("I am totally screwed");
2b87c45b 1560 return base_offset;
eb32d236
NP
1561}
1562
f8c8abc5
JH
1563int unpack_object_header(struct packed_git *p,
1564 struct pack_window **w_curs,
1565 off_t *curpos,
1566 unsigned long *sizep)
a733cb60 1567{
03e79c88 1568 unsigned char *base;
ef49a7a0 1569 unsigned long left;
ad1ed5ee 1570 unsigned long used;
2b87c45b 1571 enum object_type type;
a733cb60 1572
8d8a4ea5 1573 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1574 * as a range that we can look at. (Its actually the hash
3dff5379 1575 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1576 * the maximum deflated object size is 2^137, which is just
1577 * insane, so we know won't exceed what we have been given.
1578 */
2b87c45b 1579 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1580 used = unpack_object_header_buffer(base, left, &type, sizep);
1581 if (!used) {
1582 type = OBJ_BAD;
1583 } else
1584 *curpos += used;
ad1ed5ee 1585
2b87c45b 1586 return type;
a733cb60
LT
1587}
1588
790d96c0
TR
1589static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1590{
1591 int type;
1592 struct revindex_entry *revidx;
1593 const unsigned char *sha1;
1594 revidx = find_pack_revindex(p, obj_offset);
1595 if (!revidx)
1596 return OBJ_BAD;
1597 sha1 = nth_packed_object_sha1(p, revidx->nr);
1598 mark_bad_packed_object(p, sha1);
1599 type = sha1_object_info(sha1, NULL);
1600 if (type <= OBJ_NONE)
1601 return OBJ_BAD;
1602 return type;
1603}
1604
1605
1606#define POI_STACK_PREALLOC 64
1607
c4001d92 1608static int packed_object_info(struct packed_git *p, off_t obj_offset,
9a490590 1609 unsigned long *sizep, int *rtype)
1f688557 1610{
03e79c88 1611 struct pack_window *w_curs = NULL;
c4001d92
SP
1612 unsigned long size;
1613 off_t curpos = obj_offset;
21666f1a 1614 enum object_type type;
790d96c0
TR
1615 off_t small_poi_stack[POI_STACK_PREALLOC];
1616 off_t *poi_stack = small_poi_stack;
1617 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1618
21666f1a 1619 type = unpack_object_header(p, &w_curs, &curpos, &size);
790d96c0 1620
9a490590
JH
1621 if (rtype)
1622 *rtype = type; /* representation type */
5db47c2b 1623
790d96c0
TR
1624 if (sizep) {
1625 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1626 off_t tmp_pos = curpos;
1627 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1628 type, obj_offset);
1629 if (!base_offset) {
1630 type = OBJ_BAD;
1631 goto out;
1632 }
1633 *sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1634 if (*sizep == 0) {
1635 type = OBJ_BAD;
1636 goto out;
1637 }
1638 } else {
1639 *sizep = size;
1640 }
1641 }
1642
1643 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1644 off_t base_offset;
1645 /* Push the object we're going to leave behind */
1646 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1647 poi_stack_alloc = alloc_nr(poi_stack_nr);
1648 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1649 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1650 } else {
1651 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1652 }
1653 poi_stack[poi_stack_nr++] = obj_offset;
1654 /* If parsing the base offset fails, just unwind */
1655 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1656 if (!base_offset)
1657 goto unwind;
1658 curpos = obj_offset = base_offset;
1659 type = unpack_object_header(p, &w_curs, &curpos, &size);
1660 if (type <= OBJ_NONE) {
1661 /* If getting the base itself fails, we first
1662 * retry the base, otherwise unwind */
1663 type = retry_bad_packed_offset(p, base_offset);
1664 if (type > OBJ_NONE)
1665 goto out;
1666 goto unwind;
1667 }
1668 }
1669
21666f1a 1670 switch (type) {
790d96c0 1671 case OBJ_BAD:
a733cb60 1672 case OBJ_COMMIT:
a733cb60 1673 case OBJ_TREE:
a733cb60 1674 case OBJ_BLOB:
a733cb60 1675 case OBJ_TAG:
a69d0943 1676 break;
1f688557 1677 default:
3d77d877
NP
1678 error("unknown object type %i at offset %"PRIuMAX" in %s",
1679 type, (uintmax_t)obj_offset, p->pack_name);
1680 type = OBJ_BAD;
1f688557 1681 }
790d96c0
TR
1682
1683out:
1684 if (poi_stack != small_poi_stack)
1685 free(poi_stack);
2b87c45b 1686 unuse_pack(&w_curs);
21666f1a 1687 return type;
790d96c0
TR
1688
1689unwind:
1690 while (poi_stack_nr) {
1691 obj_offset = poi_stack[--poi_stack_nr];
1692 type = retry_bad_packed_offset(p, obj_offset);
1693 if (type > OBJ_NONE)
1694 goto out;
1695 }
1696 type = OBJ_BAD;
1697 goto out;
1f688557
JH
1698}
1699
eb950c19 1700static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1701 struct pack_window **w_curs,
c4001d92 1702 off_t curpos,
eb950c19 1703 unsigned long size)
de530aaa
SP
1704{
1705 int st;
ef49a7a0 1706 git_zstream stream;
079afb18 1707 unsigned char *buffer, *in;
de530aaa 1708
4ab07e4d 1709 buffer = xmallocz(size);
de530aaa 1710 memset(&stream, 0, sizeof(stream));
de530aaa 1711 stream.next_out = buffer;
39eea7bd 1712 stream.avail_out = size + 1;
de530aaa 1713
39c68542 1714 git_inflate_init(&stream);
079afb18 1715 do {
2b87c45b 1716 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1717 stream.next_in = in;
39c68542 1718 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1719 if (!stream.avail_out)
1720 break; /* the payload is larger than it should be */
2b87c45b 1721 curpos += stream.next_in - in;
079afb18 1722 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1723 git_inflate_end(&stream);
de530aaa
SP
1724 if ((st != Z_STREAM_END) || stream.total_out != size) {
1725 free(buffer);
1726 return NULL;
1727 }
1728
1729 return buffer;
1730}
1731
e5e01619
LT
1732#define MAX_DELTA_CACHE (256)
1733
18bdec11 1734static size_t delta_base_cached;
5e08ecbf
NP
1735
1736static struct delta_base_cache_lru_list {
1737 struct delta_base_cache_lru_list *prev;
1738 struct delta_base_cache_lru_list *next;
1739} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1740
e5e01619 1741static struct delta_base_cache_entry {
5e08ecbf
NP
1742 struct delta_base_cache_lru_list lru;
1743 void *data;
e5e01619
LT
1744 struct packed_git *p;
1745 off_t base_offset;
1746 unsigned long size;
e5e01619
LT
1747 enum object_type type;
1748} delta_base_cache[MAX_DELTA_CACHE];
1749
1750static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1751{
1752 unsigned long hash;
1753
1754 hash = (unsigned long)p + (unsigned long)base_offset;
1755 hash += (hash >> 8) + (hash >> 16);
3358004a 1756 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1757}
1758
84dd81c1
TR
1759static struct delta_base_cache_entry *
1760get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
1761{
1762 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
1763 return delta_base_cache + hash;
1764}
1765
1766static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1767 struct packed_git *p, off_t base_offset)
1768{
5266d369
JH
1769 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1770}
1771
84dd81c1
TR
1772static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1773{
1774 struct delta_base_cache_entry *ent;
1775 ent = get_delta_base_cache_entry(p, base_offset);
1776 return eq_delta_base_cache_entry(ent, p, base_offset);
1777}
1778
1779static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1780{
1781 ent->data = NULL;
1782 ent->lru.next->prev = ent->lru.prev;
1783 ent->lru.prev->next = ent->lru.next;
1784 delta_base_cached -= ent->size;
1785}
1786
62f255ad 1787static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1788 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1789{
84dd81c1 1790 struct delta_base_cache_entry *ent;
e5e01619 1791 void *ret;
e5e01619 1792
84dd81c1
TR
1793 ent = get_delta_base_cache_entry(p, base_offset);
1794
1795 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 1796 return unpack_entry(p, base_offset, type, base_size);
e5e01619 1797
84dd81c1
TR
1798 ret = ent->data;
1799
1800 if (!keep_cache)
1801 clear_delta_base_cache_entry(ent);
1802 else
182af834 1803 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
1804 *type = ent->type;
1805 *base_size = ent->size;
1806 return ret;
62f255ad
LT
1807}
1808
18bdec11
SP
1809static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1810{
1811 if (ent->data) {
1812 free(ent->data);
1813 ent->data = NULL;
5e08ecbf
NP
1814 ent->lru.next->prev = ent->lru.prev;
1815 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1816 delta_base_cached -= ent->size;
1817 }
1818}
1819
3d20c636
SP
1820void clear_delta_base_cache(void)
1821{
1822 unsigned long p;
1823 for (p = 0; p < MAX_DELTA_CACHE; p++)
1824 release_delta_base_cache(&delta_base_cache[p]);
1825}
1826
62f255ad
LT
1827static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1828 void *base, unsigned long base_size, enum object_type type)
1829{
5e08ecbf 1830 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1831 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1832 struct delta_base_cache_lru_list *lru;
e5e01619 1833
18bdec11
SP
1834 release_delta_base_cache(ent);
1835 delta_base_cached += base_size;
5e08ecbf
NP
1836
1837 for (lru = delta_base_cache_lru.next;
1838 delta_base_cached > delta_base_cache_limit
1839 && lru != &delta_base_cache_lru;
1840 lru = lru->next) {
1841 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1842 if (f->type == OBJ_BLOB)
1843 release_delta_base_cache(f);
1844 }
5e08ecbf
NP
1845 for (lru = delta_base_cache_lru.next;
1846 delta_base_cached > delta_base_cache_limit
1847 && lru != &delta_base_cache_lru;
1848 lru = lru->next) {
1849 struct delta_base_cache_entry *f = (void *)lru;
1850 release_delta_base_cache(f);
1851 }
18bdec11 1852
e5e01619
LT
1853 ent->p = p;
1854 ent->base_offset = base_offset;
1855 ent->type = type;
1856 ent->data = base;
1857 ent->size = base_size;
5e08ecbf
NP
1858 ent->lru.next = &delta_base_cache_lru;
1859 ent->lru.prev = delta_base_cache_lru.prev;
1860 delta_base_cache_lru.prev->next = &ent->lru;
1861 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1862}
1863
c2c5b270
CC
1864static void *read_object(const unsigned char *sha1, enum object_type *type,
1865 unsigned long *size);
1866
5f44324d
JH
1867static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1868{
1869 static FILE *log_file;
1870
1871 if (!log_file) {
1872 log_file = fopen(log_pack_access, "w");
1873 if (!log_file) {
1874 error("cannot open pack access log '%s' for writing: %s",
1875 log_pack_access, strerror(errno));
1876 log_pack_access = NULL;
1877 return;
1878 }
1879 }
1880 fprintf(log_file, "%s %"PRIuMAX"\n",
1881 p->pack_name, (uintmax_t)obj_offset);
1882 fflush(log_file);
1883}
1884
0e8189e2
NP
1885int do_check_packed_object_crc;
1886
abe601bb
TR
1887#define UNPACK_ENTRY_STACK_PREALLOC 64
1888struct unpack_entry_stack_ent {
1889 off_t obj_offset;
1890 off_t curpos;
1891 unsigned long size;
1892};
1893
c4001d92 1894void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 1895 enum object_type *final_type, unsigned long *final_size)
1f688557 1896{
03e79c88 1897 struct pack_window *w_curs = NULL;
c4001d92 1898 off_t curpos = obj_offset;
abe601bb
TR
1899 void *data = NULL;
1900 unsigned long size;
1901 enum object_type type;
1902 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1903 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1904 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1905 int base_from_cache = 0;
1f688557 1906
5f44324d
JH
1907 if (log_pack_access)
1908 write_pack_access_log(p, obj_offset);
1909
abe601bb
TR
1910 /* PHASE 1: drill down to the innermost base object */
1911 for (;;) {
1912 off_t base_offset;
1913 int i;
1914 struct delta_base_cache_entry *ent;
1915
1916 if (do_check_packed_object_crc && p->index_version > 1) {
1917 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1918 unsigned long len = revidx[1].offset - obj_offset;
1919 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1920 const unsigned char *sha1 =
1921 nth_packed_object_sha1(p, revidx->nr);
1922 error("bad packed object CRC for %s",
1923 sha1_to_hex(sha1));
1924 mark_bad_packed_object(p, sha1);
1925 unuse_pack(&w_curs);
1926 return NULL;
1927 }
1928 }
1929
1930 ent = get_delta_base_cache_entry(p, curpos);
1931 if (eq_delta_base_cache_entry(ent, p, curpos)) {
1932 type = ent->type;
1933 data = ent->data;
1934 size = ent->size;
1935 clear_delta_base_cache_entry(ent);
1936 base_from_cache = 1;
1937 break;
0e8189e2 1938 }
abe601bb
TR
1939
1940 type = unpack_object_header(p, &w_curs, &curpos, &size);
1941 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1942 break;
1943
1944 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1945 if (!base_offset) {
1946 error("failed to validate delta base reference "
1947 "at offset %"PRIuMAX" from %s",
1948 (uintmax_t)curpos, p->pack_name);
1949 /* bail to phase 2, in hopes of recovery */
1950 data = NULL;
1951 break;
1952 }
1953
1954 /* push object, proceed to base */
1955 if (delta_stack_nr >= delta_stack_alloc
1956 && delta_stack == small_delta_stack) {
1957 delta_stack_alloc = alloc_nr(delta_stack_nr);
1958 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
1959 memcpy(delta_stack, small_delta_stack,
1960 sizeof(*delta_stack)*delta_stack_nr);
1961 } else {
1962 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1963 }
1964 i = delta_stack_nr++;
1965 delta_stack[i].obj_offset = obj_offset;
1966 delta_stack[i].curpos = curpos;
1967 delta_stack[i].size = size;
1968
1969 curpos = obj_offset = base_offset;
0e8189e2
NP
1970 }
1971
abe601bb
TR
1972 /* PHASE 2: handle the base */
1973 switch (type) {
eb32d236
NP
1974 case OBJ_OFS_DELTA:
1975 case OBJ_REF_DELTA:
abe601bb
TR
1976 if (data)
1977 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 1978 break;
a733cb60 1979 case OBJ_COMMIT:
a733cb60 1980 case OBJ_TREE:
a733cb60 1981 case OBJ_BLOB:
a733cb60 1982 case OBJ_TAG:
abe601bb
TR
1983 if (!base_from_cache)
1984 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 1985 break;
1f688557 1986 default:
8eca0b47
NP
1987 data = NULL;
1988 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
1989 type, (uintmax_t)obj_offset, p->pack_name);
1990 }
1991
1992 /* PHASE 3: apply deltas in order */
1993
1994 /* invariants:
1995 * 'data' holds the base data, or NULL if there was corruption
1996 */
1997 while (delta_stack_nr) {
1998 void *delta_data;
1999 void *base = data;
2000 unsigned long delta_size, base_size = size;
2001 int i;
2002
2003 data = NULL;
2004
2005 if (base)
2006 add_delta_base_cache(p, obj_offset, base, base_size, type);
2007
2008 if (!base) {
2009 /*
2010 * We're probably in deep shit, but let's try to fetch
2011 * the required base anyway from another pack or loose.
2012 * This is costly but should happen only in the presence
2013 * of a corrupted pack, and is better than failing outright.
2014 */
2015 struct revindex_entry *revidx;
2016 const unsigned char *base_sha1;
2017 revidx = find_pack_revindex(p, obj_offset);
2018 if (revidx) {
2019 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2020 error("failed to read delta base object %s"
2021 " at offset %"PRIuMAX" from %s",
2022 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2023 p->pack_name);
2024 mark_bad_packed_object(p, base_sha1);
2025 base = read_object(base_sha1, &type, &base_size);
2026 }
2027 }
2028
2029 i = --delta_stack_nr;
2030 obj_offset = delta_stack[i].obj_offset;
2031 curpos = delta_stack[i].curpos;
2032 delta_size = delta_stack[i].size;
2033
2034 if (!base)
2035 continue;
2036
2037 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2038
2039 if (!delta_data) {
2040 error("failed to unpack compressed delta "
2041 "at offset %"PRIuMAX" from %s",
2042 (uintmax_t)curpos, p->pack_name);
2043 free(base);
2044 data = NULL;
2045 continue;
2046 }
2047
2048 data = patch_delta(base, base_size,
2049 delta_data, delta_size,
2050 &size);
2051 if (!data)
2052 die("failed to apply delta");
2053
2054 free (delta_data);
1f688557 2055 }
abe601bb
TR
2056
2057 *final_type = type;
2058 *final_size = size;
2059
03e79c88 2060 unuse_pack(&w_curs);
21666f1a 2061 return data;
1f688557
JH
2062}
2063
d079837e 2064const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2065 uint32_t n)
9a217f2a 2066{
42873078 2067 const unsigned char *index = p->index_data;
d079837e
SP
2068 if (!index) {
2069 if (open_pack_index(p))
2070 return NULL;
2071 index = p->index_data;
2072 }
57059091 2073 if (n >= p->num_objects)
d72308e0 2074 return NULL;
74e34e1f
NP
2075 index += 4 * 256;
2076 if (p->index_version == 1) {
2077 return index + 24 * n + 4;
2078 } else {
2079 index += 8;
2080 return index + 20 * n;
2081 }
2082}
2083
99093238 2084off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2085{
2086 const unsigned char *index = p->index_data;
2087 index += 4 * 256;
2088 if (p->index_version == 1) {
2089 return ntohl(*((uint32_t *)(index + 24 * n)));
2090 } else {
2091 uint32_t off;
2092 index += 8 + p->num_objects * (20 + 4);
2093 off = ntohl(*((uint32_t *)(index + 4 * n)));
2094 if (!(off & 0x80000000))
2095 return off;
2096 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2097 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2098 ntohl(*((uint32_t *)(index + 4)));
2099 }
9a217f2a
JH
2100}
2101
c4001d92 2102off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2103 struct packed_git *p)
1f688557 2104{
42873078 2105 const uint32_t *level1_ofs = p->index_data;
42873078 2106 const unsigned char *index = p->index_data;
628522ec
JH
2107 unsigned hi, lo, stride;
2108 static int use_lookup = -1;
2109 static int debug_lookup = -1;
2110
2111 if (debug_lookup < 0)
2112 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2113
d079837e
SP
2114 if (!index) {
2115 if (open_pack_index(p))
2116 return 0;
2117 level1_ofs = p->index_data;
2118 index = p->index_data;
2119 }
74e34e1f
NP
2120 if (p->index_version > 1) {
2121 level1_ofs += 2;
2122 index += 8;
2123 }
42873078 2124 index += 4 * 256;
74e34e1f
NP
2125 hi = ntohl(level1_ofs[*sha1]);
2126 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2127 if (p->index_version > 1) {
2128 stride = 20;
2129 } else {
2130 stride = 24;
2131 index += 4;
2132 }
2133
2134 if (debug_lookup)
6e1c2344 2135 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2136 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2137
2138 if (use_lookup < 0)
2139 use_lookup = !!getenv("GIT_USE_LOOKUP");
2140 if (use_lookup) {
2141 int pos = sha1_entry_pos(index, stride, 0,
2142 lo, hi, p->num_objects, sha1);
2143 if (pos < 0)
2144 return 0;
2145 return nth_packed_object_offset(p, pos);
2146 }
1f688557
JH
2147
2148 do {
74e34e1f 2149 unsigned mi = (lo + hi) / 2;
628522ec
JH
2150 int cmp = hashcmp(index + mi * stride, sha1);
2151
2152 if (debug_lookup)
2153 printf("lo %u hi %u rg %u mi %u\n",
2154 lo, hi, hi - lo, mi);
43057304 2155 if (!cmp)
74e34e1f 2156 return nth_packed_object_offset(p, mi);
1f688557
JH
2157 if (cmp > 0)
2158 hi = mi;
2159 else
2160 lo = mi+1;
2161 } while (lo < hi);
2162 return 0;
2163}
2164
4c080182 2165int is_pack_valid(struct packed_git *p)
d131b7af
SP
2166{
2167 /* An already open pack is known to be valid. */
2168 if (p->pack_fd != -1)
2169 return 1;
2170
2171 /* If the pack has one window completely covering the
2172 * file size, the pack is known to be valid even if
2173 * the descriptor is not currently open.
2174 */
2175 if (p->windows) {
2176 struct pack_window *w = p->windows;
2177
2178 if (!w->offset && w->len == p->pack_size)
2179 return 1;
2180 }
2181
2182 /* Force the pack to open to prove its valid. */
2183 return !open_packed_git(p);
2184}
2185
95099731
NTND
2186static int fill_pack_entry(const unsigned char *sha1,
2187 struct pack_entry *e,
2188 struct packed_git *p)
2189{
2190 off_t offset;
2191
2192 if (p->num_bad_objects) {
2193 unsigned i;
2194 for (i = 0; i < p->num_bad_objects; i++)
2195 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2196 return 0;
2197 }
2198
2199 offset = find_pack_entry_one(sha1, p);
2200 if (!offset)
2201 return 0;
2202
2203 /*
2204 * We are about to tell the caller where they can locate the
2205 * requested object. We better make sure the packfile is
2206 * still here and can be accessed before supplying that
2207 * answer, as it may have been deleted since the index was
2208 * loaded!
2209 */
2210 if (!is_pack_valid(p)) {
2211 warning("packfile %s cannot be accessed", p->pack_name);
2212 return 0;
2213 }
2214 e->offset = offset;
2215 e->p = p;
2216 hashcpy(e->sha1, sha1);
2217 return 1;
2218}
2219
4d6acb70 2220static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2221{
2222 struct packed_git *p;
43057304 2223
1f688557 2224 prepare_packed_git();
f7c22cc6
NP
2225 if (!packed_git)
2226 return 0;
1f688557 2227
c01f51cc
NTND
2228 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2229 return 1;
8eca0b47 2230
c01f51cc
NTND
2231 for (p = packed_git; p; p = p->next) {
2232 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2233 continue;
f7c22cc6 2234
c01f51cc
NTND
2235 last_found_pack = p;
2236 return 1;
2237 }
1f688557
JH
2238 return 0;
2239}
2240
a6080a0a 2241struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2242 struct packed_git *packs)
2243{
2244 struct packed_git *p;
bf592c50
DB
2245
2246 for (p = packs; p; p = p->next) {
43057304 2247 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2248 return p;
2249 }
2250 return NULL;
21666f1a 2251
bf592c50
DB
2252}
2253
21666f1a 2254static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 2255{
36e4d74a 2256 int status;
65c2e0c3
JH
2257 unsigned long mapsize, size;
2258 void *map;
ef49a7a0 2259 git_zstream stream;
d65a16f6 2260 char hdr[32];
65c2e0c3 2261
bb6b8e4f 2262 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
2263 if (!map)
2264 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
2265 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2266 status = error("unable to unpack %s header",
2267 sha1_to_hex(sha1));
21666f1a 2268 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2269 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
2270 else if (sizep)
2271 *sizep = size;
39c68542 2272 git_inflate_end(&stream);
65c2e0c3
JH
2273 munmap(map, mapsize);
2274 return status;
2275}
2276
9a490590
JH
2277/* returns enum object_type or negative */
2278int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
f0df4ed5 2279{
c4d9986f 2280 struct cached_object *co;
f0df4ed5 2281 struct pack_entry e;
9a490590 2282 int status, rtype;
f0df4ed5 2283
c4d9986f
NTND
2284 co = find_cached_object(sha1);
2285 if (co) {
9a490590
JH
2286 if (oi->sizep)
2287 *(oi->sizep) = co->size;
2288 oi->whence = OI_CACHED;
c4d9986f
NTND
2289 return co->type;
2290 }
2291
cd673c1f 2292 if (!find_pack_entry(sha1, &e)) {
ddd63e64 2293 /* Most likely it's a loose object. */
9a490590
JH
2294 status = sha1_loose_object_info(sha1, oi->sizep);
2295 if (status >= 0) {
2296 oi->whence = OI_LOOSE;
ddd63e64 2297 return status;
9a490590 2298 }
ddd63e64
SG
2299
2300 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2301 reprepare_packed_git();
cd673c1f 2302 if (!find_pack_entry(sha1, &e))
ddd63e64 2303 return status;
f0df4ed5 2304 }
3d77d877 2305
9a490590 2306 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
3d77d877
NP
2307 if (status < 0) {
2308 mark_bad_packed_object(e.p, sha1);
9a490590 2309 status = sha1_object_info_extended(sha1, oi);
5266d369
JH
2310 } else if (in_delta_base_cache(e.p, e.offset)) {
2311 oi->whence = OI_DBCACHED;
9a490590
JH
2312 } else {
2313 oi->whence = OI_PACKED;
2314 oi->u.packed.offset = e.offset;
2315 oi->u.packed.pack = e.p;
2316 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2317 rtype == OBJ_OFS_DELTA);
3d77d877
NP
2318 }
2319
2320 return status;
f0df4ed5
JS
2321}
2322
9a490590
JH
2323int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2324{
2325 struct object_info oi;
2326
2327 oi.sizep = sizep;
2328 return sha1_object_info_extended(sha1, &oi);
2329}
2330
21666f1a
NP
2331static void *read_packed_sha1(const unsigned char *sha1,
2332 enum object_type *type, unsigned long *size)
1f688557
JH
2333{
2334 struct pack_entry e;
8eca0b47 2335 void *data;
1f688557 2336
cd673c1f 2337 if (!find_pack_entry(sha1, &e))
1f688557 2338 return NULL;
8eca0b47
NP
2339 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2340 if (!data) {
2341 /*
2342 * We're probably in deep shit, but let's try to fetch
2343 * the required object anyway from another pack or loose.
2344 * This should happen only in the presence of a corrupted
2345 * pack, and is better than failing outright.
2346 */
2347 error("failed to read object %s at offset %"PRIuMAX" from %s",
2348 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2349 mark_bad_packed_object(e.p, sha1);
ac939109 2350 data = read_object(sha1, type, size);
8eca0b47
NP
2351 }
2352 return data;
1f688557
JH
2353}
2354
21666f1a
NP
2355int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2356 unsigned char *sha1)
d66b37bb
JH
2357{
2358 struct cached_object *co;
2359
21666f1a 2360 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2361 if (has_sha1_file(sha1) || find_cached_object(sha1))
2362 return 0;
2363 if (cached_object_alloc <= cached_object_nr) {
2364 cached_object_alloc = alloc_nr(cached_object_alloc);
2365 cached_objects = xrealloc(cached_objects,
2366 sizeof(*cached_objects) *
2367 cached_object_alloc);
2368 }
2369 co = &cached_objects[cached_object_nr++];
2370 co->size = len;
21666f1a 2371 co->type = type;
efa13f7b
JH
2372 co->buf = xmalloc(len);
2373 memcpy(co->buf, buf, len);
d66b37bb
JH
2374 hashcpy(co->sha1, sha1);
2375 return 0;
2376}
2377
c2c5b270
CC
2378static void *read_object(const unsigned char *sha1, enum object_type *type,
2379 unsigned long *size)
0fcfd160
LT
2380{
2381 unsigned long mapsize;
2382 void *map, *buf;
d66b37bb
JH
2383 struct cached_object *co;
2384
2385 co = find_cached_object(sha1);
2386 if (co) {
21666f1a 2387 *type = co->type;
d66b37bb 2388 *size = co->size;
182af834 2389 return xmemdupz(co->buf, co->size);
d66b37bb 2390 }
0fcfd160 2391
8276c007
PE
2392 buf = read_packed_sha1(sha1, type, size);
2393 if (buf)
2394 return buf;
bb6b8e4f 2395 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2396 if (map) {
7efbff75 2397 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2398 munmap(map, mapsize);
2399 return buf;
2400 }
637cdd9d 2401 reprepare_packed_git();
8276c007 2402 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2403}
2404
b6c4cecc
JH
2405/*
2406 * This function dies on corrupt objects; the callers who want to
2407 * deal with them should arrange to call read_object() and give error
2408 * messages themselves.
2409 */
5bf29b95
JH
2410void *read_sha1_file_extended(const unsigned char *sha1,
2411 enum object_type *type,
2412 unsigned long *size,
2413 unsigned flag)
ac939109 2414{
3ba7a065 2415 void *data;
e8b15e61 2416 char *path;
b6c4cecc 2417 const struct packed_git *p;
5bf29b95
JH
2418 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2419 ? lookup_replace_object(sha1) : sha1;
b6c4cecc 2420
3ba7a065
JH
2421 errno = 0;
2422 data = read_object(repl, type, size);
4bbf5a26 2423 if (data)
b6c4cecc 2424 return data;
68095570 2425
25f3af3f 2426 if (errno && errno != ENOENT)
3ba7a065
JH
2427 die_errno("failed to read object %s", sha1_to_hex(sha1));
2428
68095570 2429 /* die if we replaced an object with one that does not exist */
b6c4cecc 2430 if (repl != sha1)
68095570
CC
2431 die("replacement %s not found for %s",
2432 sha1_to_hex(repl), sha1_to_hex(sha1));
2433
b6c4cecc
JH
2434 if (has_loose_object(repl)) {
2435 path = sha1_file_name(sha1);
2436 die("loose object %s (stored in %s) is corrupt",
2437 sha1_to_hex(repl), path);
e8b15e61 2438 }
68095570 2439
b6c4cecc
JH
2440 if ((p = has_packed_and_bad(repl)) != NULL)
2441 die("packed object %s (stored in %s) is corrupt",
2442 sha1_to_hex(repl), p->pack_name);
f5552aee 2443
b6c4cecc 2444 return NULL;
ac939109
NP
2445}
2446
40469ee9 2447void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2448 const char *required_type_name,
40469ee9
JH
2449 unsigned long *size,
2450 unsigned char *actual_sha1_return)
f4913f91 2451{
21666f1a 2452 enum object_type type, required_type;
f4913f91
JH
2453 void *buffer;
2454 unsigned long isize;
40469ee9 2455 unsigned char actual_sha1[20];
f4913f91 2456
21666f1a 2457 required_type = type_from_string(required_type_name);
e702496e 2458 hashcpy(actual_sha1, sha1);
40469ee9
JH
2459 while (1) {
2460 int ref_length = -1;
2461 const char *ref_type = NULL;
f4913f91 2462
21666f1a 2463 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2464 if (!buffer)
2465 return NULL;
21666f1a 2466 if (type == required_type) {
40469ee9
JH
2467 *size = isize;
2468 if (actual_sha1_return)
e702496e 2469 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2470 return buffer;
2471 }
2472 /* Handle references */
21666f1a 2473 else if (type == OBJ_COMMIT)
40469ee9 2474 ref_type = "tree ";
21666f1a 2475 else if (type == OBJ_TAG)
40469ee9
JH
2476 ref_type = "object ";
2477 else {
2478 free(buffer);
2479 return NULL;
2480 }
2481 ref_length = strlen(ref_type);
f4913f91 2482
50974ec9
MK
2483 if (ref_length + 40 > isize ||
2484 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2485 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2486 free(buffer);
2487 return NULL;
2488 }
1cf58e72 2489 free(buffer);
40469ee9
JH
2490 /* Now we have the ID of the referred-to object in
2491 * actual_sha1. Check again. */
f4913f91 2492 }
f4913f91
JH
2493}
2494
ce9fbf16 2495static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2496 const char *type, unsigned char *sha1,
d65a16f6 2497 char *hdr, int *hdrlen)
d410c0f5 2498{
9126f009 2499 git_SHA_CTX c;
d410c0f5
JH
2500
2501 /* Generate the header */
d65a16f6 2502 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2503
2504 /* Sha1.. */
9126f009
NP
2505 git_SHA1_Init(&c);
2506 git_SHA1_Update(&c, hdr, *hdrlen);
2507 git_SHA1_Update(&c, buf, len);
2508 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2509}
2510
230f1322 2511/*
5a688fe4
JH
2512 * Move the just written object into its final resting place.
2513 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2514 * "moving" is only a part of what it does, when no patch between
2515 * master to pu changes the call sites of this function.
230f1322 2516 */
839837b9 2517int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2518{
e32c0a9c 2519 int ret = 0;
5a688fe4 2520
348df166 2521 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2522 goto try_rename;
2523 else if (link(tmpfile, filename))
e32c0a9c 2524 ret = errno;
7ebb6fca
LT
2525
2526 /*
2527 * Coda hack - coda doesn't like cross-directory links,
2528 * so we fall back to a rename, which will mean that it
2529 * won't be able to check collisions, but that's not a
2530 * big deal.
2531 *
2532 * The same holds for FAT formatted media.
2533 *
3be1f18e 2534 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2535 * left to unlink.
2536 */
2537 if (ret && ret != EEXIST) {
be66a6c4 2538 try_rename:
7ebb6fca 2539 if (!rename(tmpfile, filename))
3be1f18e 2540 goto out;
9e48b389 2541 ret = errno;
230f1322 2542 }
691f1a28 2543 unlink_or_warn(tmpfile);
230f1322
LT
2544 if (ret) {
2545 if (ret != EEXIST) {
82247e9b 2546 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2547 }
2548 /* FIXME!!! Collision check here ? */
2549 }
2550
3be1f18e 2551out:
5256b006 2552 if (adjust_shared_perm(filename))
5a688fe4 2553 return error("unable to set permission to '%s'", filename);
230f1322
LT
2554 return 0;
2555}
2556
4d548150
LT
2557static int write_buffer(int fd, const void *buf, size_t len)
2558{
d34cf19b 2559 if (write_in_full(fd, buf, len) < 0)
93822c22 2560 return error("file write error (%s)", strerror(errno));
4d548150
LT
2561 return 0;
2562}
2563
ce9fbf16 2564int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2565 unsigned char *sha1)
2566{
d65a16f6 2567 char hdr[32];
abdc3fc8
RS
2568 int hdrlen;
2569 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2570 return 0;
2571}
2572
e9039dd3
LT
2573/* Finalize a file on disk, and close it. */
2574static void close_sha1_file(int fd)
2575{
aafe9fba
LT
2576 if (fsync_object_files)
2577 fsync_or_die(fd, "sha1 file");
e9039dd3 2578 if (close(fd) != 0)
d824cbba 2579 die_errno("error when closing sha1 file");
e9039dd3
LT
2580}
2581
5723fe7e
LT
2582/* Size of directory component, including the ending '/' */
2583static inline int directory_size(const char *filename)
2584{
2585 const char *s = strrchr(filename, '/');
2586 if (!s)
2587 return 0;
2588 return s - filename + 1;
2589}
2590
2591/*
2592 * This creates a temporary file in the same directory as the final
2593 * 'filename'
2594 *
2595 * We want to avoid cross-directory filename renames, because those
2596 * can have problems on various filesystems (FAT, NFS, Coda).
2597 */
2598static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2599{
2600 int fd, dirlen = directory_size(filename);
2601
2602 if (dirlen + 20 > bufsiz) {
2603 errno = ENAMETOOLONG;
2604 return -1;
2605 }
2606 memcpy(buffer, filename, dirlen);
2607 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2608 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2609 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2610 /* Make sure the directory exists */
6ff6af62 2611 memcpy(buffer, filename, dirlen);
5723fe7e 2612 buffer[dirlen-1] = 0;
3bfaf018 2613 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2614 return -1;
2615
2616 /* Try again */
2617 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2618 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2619 }
2620 return fd;
2621}
2622
bbac7311 2623static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2624 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2625{
915308b1 2626 int fd, ret;
9892beba 2627 unsigned char compressed[4096];
ef49a7a0 2628 git_zstream stream;
748af44c
NP
2629 git_SHA_CTX c;
2630 unsigned char parano_sha1[20];
706bc531 2631 char *filename;
ab1900a3 2632 static char tmp_file[PATH_MAX];
a44c9a5e 2633
972a9155 2634 filename = sha1_file_name(sha1);
ab1900a3 2635 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 2636 if (fd < 0) {
35243577 2637 if (errno == EACCES)
82247e9b 2638 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 2639 else
5eaeda70 2640 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
2641 }
2642
0fcfd160
LT
2643 /* Set it up */
2644 memset(&stream, 0, sizeof(stream));
55bb5c91 2645 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 2646 stream.next_out = compressed;
9892beba 2647 stream.avail_out = sizeof(compressed);
748af44c 2648 git_SHA1_Init(&c);
a44c9a5e
LT
2649
2650 /* First header.. */
d65a16f6 2651 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2652 stream.avail_in = hdrlen;
55bb5c91
JH
2653 while (git_deflate(&stream, 0) == Z_OK)
2654 ; /* nothing */
748af44c 2655 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2656
2657 /* Then the data itself.. */
c00e657d 2658 stream.next_in = (void *)buf;
a44c9a5e 2659 stream.avail_in = len;
9892beba 2660 do {
748af44c 2661 unsigned char *in0 = stream.next_in;
55bb5c91 2662 ret = git_deflate(&stream, Z_FINISH);
748af44c 2663 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2664 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2665 die("unable to write sha1 file");
2666 stream.next_out = compressed;
2667 stream.avail_out = sizeof(compressed);
2668 } while (ret == Z_OK);
2669
ac54c277
LT
2670 if (ret != Z_STREAM_END)
2671 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 2672 ret = git_deflate_end_gently(&stream);
ac54c277
LT
2673 if (ret != Z_OK)
2674 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2675 git_SHA1_Final(parano_sha1, &c);
2676 if (hashcmp(sha1, parano_sha1) != 0)
2677 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2678
e9039dd3 2679 close_sha1_file(fd);
0fcfd160 2680
bbac7311
NP
2681 if (mtime) {
2682 struct utimbuf utb;
2683 utb.actime = mtime;
2684 utb.modtime = mtime;
ab1900a3 2685 if (utime(tmp_file, &utb) < 0)
bbac7311 2686 warning("failed utime() on %s: %s",
ab1900a3 2687 tmp_file, strerror(errno));
bbac7311
NP
2688 }
2689
ab1900a3 2690 return move_temp_to_file(tmp_file, filename);
0fcfd160 2691}
8237b185 2692
c00e657d 2693int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2694{
2695 unsigned char sha1[20];
2696 char hdr[32];
2697 int hdrlen;
2698
2699 /* Normally if we have it in the pack then we do not bother writing
2700 * it out into .git/objects/??/?{38} file.
2701 */
2702 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2703 if (returnsha1)
2704 hashcpy(returnsha1, sha1);
2705 if (has_sha1_file(sha1))
2706 return 0;
2707 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2708}
2709
2710int force_object_loose(const unsigned char *sha1, time_t mtime)
2711{
bbac7311
NP
2712 void *buf;
2713 unsigned long len;
2714 enum object_type type;
2715 char hdr[32];
2716 int hdrlen;
1fb23e65 2717 int ret;
bbac7311 2718
c529d75a 2719 if (has_loose_object(sha1))
bbac7311
NP
2720 return 0;
2721 buf = read_packed_sha1(sha1, &type, &len);
2722 if (!buf)
2723 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2724 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2725 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2726 free(buf);
2727
2728 return ret;
bbac7311
NP
2729}
2730
bf592c50
DB
2731int has_pack_index(const unsigned char *sha1)
2732{
2733 struct stat st;
2734 if (stat(sha1_pack_index_name(sha1), &st))
2735 return 0;
2736 return 1;
bf592c50
DB
2737}
2738
cd673c1f
JH
2739int has_sha1_pack(const unsigned char *sha1)
2740{
2741 struct pack_entry e;
2742 return find_pack_entry(sha1, &e);
2743}
2744
8237b185
DB
2745int has_sha1_file(const unsigned char *sha1)
2746{
1f688557
JH
2747 struct pack_entry e;
2748
cd673c1f 2749 if (find_pack_entry(sha1, &e))
1f688557 2750 return 1;
c529d75a 2751 return has_loose_object(sha1);
8237b185 2752}
74400e71 2753
c879daa2
NTND
2754static void check_tree(const void *buf, size_t size)
2755{
2756 struct tree_desc desc;
2757 struct name_entry entry;
2758
2759 init_tree_desc(&desc, buf, size);
2760 while (tree_entry(&desc, &entry))
2761 /* do nothing
2762 * tree_entry() will die() on malformed entries */
2763 ;
2764}
2765
2766static void check_commit(const void *buf, size_t size)
2767{
2768 struct commit c;
2769 memset(&c, 0, sizeof(c));
2770 if (parse_commit_buffer(&c, buf, size))
2771 die("corrupt commit");
2772}
2773
2774static void check_tag(const void *buf, size_t size)
2775{
2776 struct tag t;
2777 memset(&t, 0, sizeof(t));
2778 if (parse_tag_buffer(&t, buf, size))
2779 die("corrupt tag");
2780}
2781
43df4f86 2782static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
2783 enum object_type type,
2784 const char *path, unsigned flags)
e7332f96 2785{
6c510bee 2786 int ret, re_allocated = 0;
c4ce46fc 2787 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 2788
7672db20 2789 if (!type)
edaec3fb 2790 type = OBJ_BLOB;
6c510bee
LT
2791
2792 /*
2793 * Convert blobs to git internal format
2794 */
43df4f86 2795 if ((type == OBJ_BLOB) && path) {
f285a2d7 2796 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 2797 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 2798 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 2799 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2800 re_allocated = 1;
2801 }
2802 }
c4ce46fc 2803 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
2804 if (type == OBJ_TREE)
2805 check_tree(buf, size);
2806 if (type == OBJ_COMMIT)
2807 check_commit(buf, size);
2808 if (type == OBJ_TAG)
2809 check_tag(buf, size);
2810 }
6c510bee 2811
7672db20 2812 if (write_object)
edaec3fb 2813 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2814 else
edaec3fb 2815 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2816 if (re_allocated)
6c510bee 2817 free(buf);
43df4f86
DP
2818 return ret;
2819}
2820
7b41e1e1
JH
2821static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2822 const char *path, unsigned flags)
2823{
2824 struct strbuf sbuf = STRBUF_INIT;
2825 int ret;
2826
2827 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2828 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2829 else
2830 ret = -1;
2831 strbuf_release(&sbuf);
2832 return ret;
2833}
2834
ea68b0ce
DP
2835#define SMALL_FILE_SIZE (32*1024)
2836
7b41e1e1
JH
2837static int index_core(unsigned char *sha1, int fd, size_t size,
2838 enum object_type type, const char *path,
2839 unsigned flags)
43df4f86
DP
2840{
2841 int ret;
43df4f86 2842
7b41e1e1 2843 if (!size) {
c4ce46fc 2844 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
2845 } else if (size <= SMALL_FILE_SIZE) {
2846 char *buf = xmalloc(size);
2847 if (size == read_in_full(fd, buf, size))
c4ce46fc 2848 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
2849 else
2850 ret = error("short read %s", strerror(errno));
2851 free(buf);
08bda208 2852 } else {
43df4f86 2853 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 2854 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 2855 munmap(buf, size);
08bda208 2856 }
7b41e1e1
JH
2857 return ret;
2858}
2859
4dd1fbc7 2860/*
568508e7
JH
2861 * This creates one packfile per large blob unless bulk-checkin
2862 * machinery is "plugged".
4dd1fbc7
JH
2863 *
2864 * This also bypasses the usual "convert-to-git" dance, and that is on
2865 * purpose. We could write a streaming version of the converting
2866 * functions and insert that before feeding the data to fast-import
4f22b101
JK
2867 * (or equivalent in-core API described above). However, that is
2868 * somewhat complicated, as we do not know the size of the filter
2869 * result, which we need to know beforehand when writing a git object.
2870 * Since the primary motivation for trying to stream from the working
2871 * tree file and to avoid mmaping it in core is to deal with large
2872 * binary blobs, they generally do not want to get any conversion, and
2873 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
2874 */
2875static int index_stream(unsigned char *sha1, int fd, size_t size,
2876 enum object_type type, const char *path,
2877 unsigned flags)
2878{
568508e7 2879 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
2880}
2881
7b41e1e1
JH
2882int index_fd(unsigned char *sha1, int fd, struct stat *st,
2883 enum object_type type, const char *path, unsigned flags)
2884{
2885 int ret;
2886 size_t size = xsize_t(st->st_size);
2887
2888 if (!S_ISREG(st->st_mode))
2889 ret = index_pipe(sha1, fd, type, path, flags);
4f22b101
JK
2890 else if (size <= big_file_threshold || type != OBJ_BLOB ||
2891 (path && would_convert_to_git(path, NULL, 0, 0)))
7b41e1e1 2892 ret = index_core(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
2893 else
2894 ret = index_stream(sha1, fd, size, type, path, flags);
43df4f86 2895 close(fd);
aac17941 2896 return ret;
74400e71 2897}
ec1fcc16 2898
c4ce46fc 2899int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
2900{
2901 int fd;
b760d3aa 2902 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
2903
2904 switch (st->st_mode & S_IFMT) {
2905 case S_IFREG:
2906 fd = open(path, O_RDONLY);
2907 if (fd < 0)
2908 return error("open(\"%s\"): %s", path,
2909 strerror(errno));
c4ce46fc 2910 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
2911 return error("%s: failed to insert into database",
2912 path);
2913 break;
2914 case S_IFLNK:
b760d3aa 2915 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 2916 char *errstr = strerror(errno);
ec1fcc16
JH
2917 return error("readlink(\"%s\"): %s", path,
2918 errstr);
2919 }
c4ce46fc 2920 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
2921 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2922 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
2923 return error("%s: failed to insert into database",
2924 path);
b760d3aa 2925 strbuf_release(&sb);
ec1fcc16 2926 break;
f35a6d3b
LT
2927 case S_IFDIR:
2928 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2929 default:
2930 return error("%s: unsupported file type", path);
2931 }
2932 return 0;
2933}
a69e5429
JH
2934
2935int read_pack_header(int fd, struct pack_header *header)
2936{
c697ad14
HO
2937 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2938 /* "eof before pack header was fully read" */
2939 return PH_ERROR_EOF;
2940
a69e5429
JH
2941 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2942 /* "protocol error (pack signature mismatch detected)" */
2943 return PH_ERROR_PACK_SIGNATURE;
2944 if (!pack_version_ok(header->hdr_version))
2945 /* "protocol error (pack version unsupported)" */
2946 return PH_ERROR_PROTOCOL;
2947 return 0;
2948}
40d52ff7
JK
2949
2950void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2951{
2952 enum object_type type = sha1_object_info(sha1, NULL);
2953 if (type < 0)
2954 die("%s is not a valid object", sha1_to_hex(sha1));
2955 if (type != expect)
2956 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2957 typename(expect));
2958}