sha1_file: introduce close_one_pack() to close packs on fd pressure
[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
88d0db55
BC
676/*
677 * The LRU pack is the one with the oldest MRU window, preferring packs
678 * with no used windows, or the oldest mtime if it has no windows allocated.
679 */
680static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
681{
682 struct pack_window *w, *this_mru_w;
683 int has_windows_inuse = 0;
684
685 /*
686 * Reject this pack if it has windows and the previously selected
687 * one does not. If this pack does not have windows, reject
688 * it if the pack file is newer than the previously selected one.
689 */
690 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
691 return;
692
693 for (w = this_mru_w = p->windows; w; w = w->next) {
694 /*
695 * Reject this pack if any of its windows are in use,
696 * but the previously selected pack did not have any
697 * inuse windows. Otherwise, record that this pack
698 * has windows in use.
699 */
700 if (w->inuse_cnt) {
701 if (*accept_windows_inuse)
702 has_windows_inuse = 1;
703 else
704 return;
705 }
706
707 if (w->last_used > this_mru_w->last_used)
708 this_mru_w = w;
709
710 /*
711 * Reject this pack if it has windows that have been
712 * used more recently than the previously selected pack.
713 * If the previously selected pack had windows inuse and
714 * we have not encountered a window in this pack that is
715 * inuse, skip this check since we prefer a pack with no
716 * inuse windows to one that has inuse windows.
717 */
718 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
719 this_mru_w->last_used > (*mru_w)->last_used)
720 return;
721 }
722
723 /*
724 * Select this pack.
725 */
726 *mru_w = this_mru_w;
727 *lru_p = p;
728 *accept_windows_inuse = has_windows_inuse;
729}
730
731static int close_one_pack(void)
732{
733 struct packed_git *p, *lru_p = NULL;
734 struct pack_window *mru_w = NULL;
735 int accept_windows_inuse = 1;
736
737 for (p = packed_git; p; p = p->next) {
738 if (p->pack_fd == -1)
739 continue;
740 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
741 }
742
743 if (lru_p) {
744 close(lru_p->pack_fd);
745 pack_open_fds--;
746 lru_p->pack_fd = -1;
747 return 1;
748 }
749
750 return 0;
751}
752
03e79c88 753void unuse_pack(struct pack_window **w_cursor)
f9253394 754{
03e79c88
SP
755 struct pack_window *w = *w_cursor;
756 if (w) {
757 w->inuse_cnt--;
758 *w_cursor = NULL;
759 }
1f688557
JH
760}
761
fa5fc15d
SP
762void close_pack_index(struct packed_git *p)
763{
764 if (p->index_data) {
765 munmap((void *)p->index_data, p->index_size);
766 p->index_data = NULL;
767 }
768}
769
c74faea1
NP
770/*
771 * This is used by git-repack in case a newly created pack happens to
772 * contain the same set of objects as an existing one. In that case
773 * the resulting file might be different even if its name would be the
774 * same. It is best to close any reference to the old pack before it is
775 * replaced on disk. Of course no index pointers nor windows for given pack
776 * must subsist at this point. If ever objects from this pack are requested
777 * again, the new version of the pack will be reinitialized through
778 * reprepare_packed_git().
779 */
780void free_pack_by_name(const char *pack_name)
781{
782 struct packed_git *p, **pp = &packed_git;
783
784 while (*pp) {
785 p = *pp;
786 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 787 clear_delta_base_cache();
c74faea1 788 close_pack_windows(p);
c7934306 789 if (p->pack_fd != -1) {
c74faea1 790 close(p->pack_fd);
c7934306
SP
791 pack_open_fds--;
792 }
fa5fc15d 793 close_pack_index(p);
c74faea1
NP
794 free(p->bad_object_sha1);
795 *pp = p->next;
c01f51cc
NTND
796 if (last_found_pack == p)
797 last_found_pack = NULL;
c74faea1
NP
798 free(p);
799 return;
800 }
801 pp = &p->next;
802 }
803}
804
a0788266
JS
805static unsigned int get_max_fd_limit(void)
806{
807#ifdef RLIMIT_NOFILE
808 struct rlimit lim;
809
810 if (getrlimit(RLIMIT_NOFILE, &lim))
811 die_errno("cannot get RLIMIT_NOFILE");
812
813 return lim.rlim_cur;
814#elif defined(_SC_OPEN_MAX)
815 return sysconf(_SC_OPEN_MAX);
816#elif defined(OPEN_MAX)
817 return OPEN_MAX;
818#else
819 return 1; /* see the caller ;-) */
820#endif
821}
822
3cf8b462
SP
823/*
824 * Do not call this directly as this leaks p->pack_fd on error return;
825 * call open_packed_git() instead.
826 */
827static int open_packed_git_1(struct packed_git *p)
1f688557 828{
9bc879c1
SP
829 struct stat st;
830 struct pack_header hdr;
831 unsigned char sha1[20];
832 unsigned char *idx_sha1;
2c039da8 833 long fd_flag;
9bc879c1 834
d079837e
SP
835 if (!p->index_data && open_pack_index(p))
836 return error("packfile %s index unavailable", p->pack_name);
837
c7934306 838 if (!pack_max_fds) {
a0788266 839 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
840
841 /* Save 3 for stdin/stdout/stderr, 22 for work */
842 if (25 < max_fds)
843 pack_max_fds = max_fds - 25;
844 else
845 pack_max_fds = 1;
846 }
847
88d0db55 848 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
849 ; /* nothing */
850
f4e51683 851 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 852 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 853 return -1;
c7934306 854 pack_open_fds++;
9bc879c1
SP
855
856 /* If we created the struct before we had the pack we lack size. */
bf592c50 857 if (!p->pack_size) {
bf592c50 858 if (!S_ISREG(st.st_mode))
072db278 859 return error("packfile %s not a regular file", p->pack_name);
bf592c50 860 p->pack_size = st.st_size;
9bc879c1 861 } else if (p->pack_size != st.st_size)
072db278 862 return error("packfile %s size changed", p->pack_name);
9bc879c1 863
2c039da8
JH
864 /* We leave these file descriptors open with sliding mmap;
865 * there is no point keeping them open across exec(), though.
866 */
867 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
868 if (fd_flag < 0)
072db278 869 return error("cannot determine file descriptor flags");
2c039da8
JH
870 fd_flag |= FD_CLOEXEC;
871 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 872 return error("cannot set FD_CLOEXEC");
2c039da8 873
9bc879c1 874 /* Verify we recognize this pack file format. */
e6e2bd62 875 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 876 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 877 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 878 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 879 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
880 return error("packfile %s is version %"PRIu32" and not"
881 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
882 p->pack_name, ntohl(hdr.hdr_version));
883
884 /* Verify the pack matches its index. */
57059091 885 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
886 return error("packfile %s claims to have %"PRIu32" objects"
887 " while index indicates %"PRIu32" objects",
57059091
NP
888 p->pack_name, ntohl(hdr.hdr_entries),
889 p->num_objects);
9bc879c1 890 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 891 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 892 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 893 return error("packfile %s signature is unavailable", p->pack_name);
42873078 894 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 895 if (hashcmp(sha1, idx_sha1))
072db278
SP
896 return error("packfile %s does not match index", p->pack_name);
897 return 0;
9bc879c1
SP
898}
899
3cf8b462
SP
900static int open_packed_git(struct packed_git *p)
901{
902 if (!open_packed_git_1(p))
903 return 0;
904 if (p->pack_fd != -1) {
905 close(p->pack_fd);
c7934306 906 pack_open_fds--;
3cf8b462
SP
907 p->pack_fd = -1;
908 }
909 return -1;
910}
911
c4001d92 912static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
913{
914 /* We must promise at least 20 bytes (one hash) after the
915 * offset is available from this window, otherwise the offset
916 * is not actually in this window and a different window (which
917 * has that one hash excess) must be used. This is to support
918 * the object header and delta base parsing routines below.
919 */
920 off_t win_off = win->offset;
921 return win_off <= offset
922 && (offset + 20) <= (win_off + win->len);
923}
924
4b25d091 925unsigned char *use_pack(struct packed_git *p,
03e79c88 926 struct pack_window **w_cursor,
c4001d92 927 off_t offset,
ef49a7a0 928 unsigned long *left)
9bc879c1 929{
60bb8b14 930 struct pack_window *win = *w_cursor;
03e79c88 931
a9d98a14 932 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
933 * pointless to ask for an offset into the middle of that
934 * hash, and the in_window function above wouldn't match
935 * don't allow an offset too close to the end of the file.
936 */
d131b7af
SP
937 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
938 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
939 if (offset > (p->pack_size - 20))
940 die("offset beyond end of packfile (truncated pack?)");
941
942 if (!win || !in_window(win, offset)) {
943 if (win)
944 win->inuse_cnt--;
945 for (win = p->windows; win; win = win->next) {
946 if (in_window(win, offset))
947 break;
948 }
949 if (!win) {
78a28df9 950 size_t window_align = packed_git_window_size / 2;
dc49cd76 951 off_t len;
d131b7af
SP
952
953 if (p->pack_fd == -1 && open_packed_git(p))
954 die("packfile %s cannot be accessed", p->pack_name);
955
60bb8b14 956 win = xcalloc(1, sizeof(*win));
78a28df9 957 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
958 len = p->pack_size - win->offset;
959 if (len > packed_git_window_size)
960 len = packed_git_window_size;
961 win->len = (size_t)len;
60bb8b14 962 pack_mapped += win->len;
11daf39b 963 while (packed_git_limit < pack_mapped
d1efefa4 964 && unuse_one_window(p, p->pack_fd))
60bb8b14 965 ; /* nothing */
c4712e45 966 win->base = xmmap(NULL, win->len,
60bb8b14
SP
967 PROT_READ, MAP_PRIVATE,
968 p->pack_fd, win->offset);
969 if (win->base == MAP_FAILED)
73b4e4be
SP
970 die("packfile %s cannot be mapped: %s",
971 p->pack_name,
972 strerror(errno));
d131b7af
SP
973 if (!win->offset && win->len == p->pack_size
974 && !p->do_not_close) {
975 close(p->pack_fd);
976 pack_open_fds--;
977 p->pack_fd = -1;
978 }
a53128b6
SP
979 pack_mmap_calls++;
980 pack_open_windows++;
981 if (pack_mapped > peak_pack_mapped)
982 peak_pack_mapped = pack_mapped;
983 if (pack_open_windows > peak_pack_open_windows)
984 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
985 win->next = p->windows;
986 p->windows = win;
987 }
1f688557 988 }
03e79c88
SP
989 if (win != *w_cursor) {
990 win->last_used = pack_used_ctr++;
991 win->inuse_cnt++;
992 *w_cursor = win;
993 }
60bb8b14 994 offset -= win->offset;
03e79c88 995 if (left)
dc49cd76 996 *left = win->len - xsize_t(offset);
03e79c88 997 return win->base + offset;
1f688557
JH
998}
999
27d69a46
NP
1000static struct packed_git *alloc_packed_git(int extra)
1001{
1002 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1003 memset(p, 0, sizeof(*p));
1004 p->pack_fd = -1;
1005 return p;
1006}
1007
e0500293
JN
1008static void try_to_free_pack_memory(size_t size)
1009{
1010 release_pack_memory(size, -1);
1011}
1012
42873078 1013struct packed_git *add_packed_git(const char *path, int path_len, int local)
1f688557 1014{
e0500293 1015 static int have_set_try_to_free_routine;
1f688557 1016 struct stat st;
27d69a46 1017 struct packed_git *p = alloc_packed_git(path_len + 2);
1f688557 1018
e0500293
JN
1019 if (!have_set_try_to_free_routine) {
1020 have_set_try_to_free_routine = 1;
1021 set_try_to_free_routine(try_to_free_pack_memory);
1022 }
1023
42873078
NP
1024 /*
1025 * Make sure a corresponding .pack file exists and that
1026 * the index looks sane.
1027 */
1028 path_len -= strlen(".idx");
27d69a46
NP
1029 if (path_len < 1) {
1030 free(p);
1f688557 1031 return NULL;
27d69a46 1032 }
42873078 1033 memcpy(p->pack_name, path, path_len);
8d25931d
BC
1034
1035 strcpy(p->pack_name + path_len, ".keep");
1036 if (!access(p->pack_name, F_OK))
1037 p->pack_keep = 1;
1038
42873078 1039 strcpy(p->pack_name + path_len, ".pack");
d079837e 1040 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1041 free(p);
1f688557
JH
1042 return NULL;
1043 }
42873078 1044
1f688557
JH
1045 /* ok, it looks sane as far as we can check without
1046 * actually mapping the pack file.
1047 */
1f688557 1048 p->pack_size = st.st_size;
9d835df2 1049 p->pack_local = local;
b867092f 1050 p->mtime = st.st_mtime;
42873078
NP
1051 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1052 hashclr(p->sha1);
1f688557
JH
1053 return p;
1054}
1055
7b64469a 1056struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1057{
42873078 1058 const char *path = sha1_pack_name(sha1);
27d69a46 1059 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
bf592c50 1060
27d69a46
NP
1061 strcpy(p->pack_name, path);
1062 hashcpy(p->sha1, sha1);
42873078
NP
1063 if (check_packed_git_idx(idx_path, p)) {
1064 free(p);
bf592c50 1065 return NULL;
42873078 1066 }
bf592c50 1067
bf592c50
DB
1068 return p;
1069}
1070
1071void install_packed_git(struct packed_git *pack)
1072{
c7934306
SP
1073 if (pack->pack_fd != -1)
1074 pack_open_fds++;
1075
bf592c50
DB
1076 pack->next = packed_git;
1077 packed_git = pack;
1078}
1079
9d835df2 1080static void prepare_packed_git_one(char *objdir, int local)
1f688557 1081{
9cb18f56
JM
1082 /* Ensure that this buffer is large enough so that we can
1083 append "/pack/" without clobbering the stack even if
1084 strlen(objdir) were PATH_MAX. */
1085 char path[PATH_MAX + 1 + 4 + 1 + 1];
1f688557
JH
1086 int len;
1087 DIR *dir;
1088 struct dirent *de;
1089
1090 sprintf(path, "%s/pack", objdir);
1091 len = strlen(path);
1092 dir = opendir(path);
b5b16990 1093 if (!dir) {
26125f6b 1094 if (errno != ENOENT)
bd2afde8 1095 error("unable to open object pack directory: %s: %s",
26125f6b 1096 path, strerror(errno));
1f688557 1097 return;
b5b16990 1098 }
1f688557
JH
1099 path[len++] = '/';
1100 while ((de = readdir(dir)) != NULL) {
1101 int namelen = strlen(de->d_name);
1102 struct packed_git *p;
1103
5bb1cda5 1104 if (!has_extension(de->d_name, ".idx"))
1f688557
JH
1105 continue;
1106
9cb18f56
JM
1107 if (len + namelen + 1 > sizeof(path))
1108 continue;
1109
54a15a8d 1110 /* Don't reopen a pack we already have. */
1f688557 1111 strcpy(path + len, de->d_name);
86f7780c
JK
1112 for (p = packed_git; p; p = p->next) {
1113 if (!memcmp(path, p->pack_name, len + namelen - 4))
1114 break;
1115 }
1116 if (p)
1117 continue;
54a15a8d
SP
1118 /* See if it really is a valid .idx file with corresponding
1119 * .pack file that we can map.
1120 */
9d835df2 1121 p = add_packed_git(path, len + namelen, local);
1f688557
JH
1122 if (!p)
1123 continue;
625e9421 1124 install_packed_git(p);
1f688557 1125 }
5b35bcd5 1126 closedir(dir);
1f688557
JH
1127}
1128
b867092f
JH
1129static int sort_pack(const void *a_, const void *b_)
1130{
1131 struct packed_git *a = *((struct packed_git **)a_);
1132 struct packed_git *b = *((struct packed_git **)b_);
1133 int st;
1134
1135 /*
1136 * Local packs tend to contain objects specific to our
1137 * variant of the project than remote ones. In addition,
1138 * remote ones could be on a network mounted filesystem.
1139 * Favor local ones for these reasons.
1140 */
1141 st = a->pack_local - b->pack_local;
1142 if (st)
1143 return -st;
1144
1145 /*
1146 * Younger packs tend to contain more recent objects,
1147 * and more recent objects tend to get accessed more
1148 * often.
1149 */
1150 if (a->mtime < b->mtime)
1151 return 1;
1152 else if (a->mtime == b->mtime)
1153 return 0;
1154 return -1;
1155}
1156
1157static void rearrange_packed_git(void)
1158{
1159 struct packed_git **ary, *p;
1160 int i, n;
1161
1162 for (n = 0, p = packed_git; p; p = p->next)
1163 n++;
1164 if (n < 2)
1165 return;
1166
1167 /* prepare an array of packed_git for easier sorting */
1168 ary = xcalloc(n, sizeof(struct packed_git *));
1169 for (n = 0, p = packed_git; p; p = p->next)
1170 ary[n++] = p;
1171
1172 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1173
1174 /* link them back again */
1175 for (i = 0; i < n - 1; i++)
1176 ary[i]->next = ary[i + 1];
1177 ary[n - 1]->next = NULL;
1178 packed_git = ary[0];
1179
1180 free(ary);
1181}
1182
637cdd9d 1183static int prepare_packed_git_run_once = 0;
9a217f2a 1184void prepare_packed_git(void)
1f688557 1185{
d5a63b99 1186 struct alternate_object_database *alt;
1f688557 1187
637cdd9d 1188 if (prepare_packed_git_run_once)
1f688557 1189 return;
9d835df2 1190 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1191 prepare_alt_odb();
d5a63b99 1192 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1193 alt->name[-1] = 0;
9d835df2 1194 prepare_packed_git_one(alt->base, 0);
1494e038 1195 alt->name[-1] = '/';
1f688557 1196 }
b867092f 1197 rearrange_packed_git();
637cdd9d
JK
1198 prepare_packed_git_run_once = 1;
1199}
1200
fc04c412 1201void reprepare_packed_git(void)
637cdd9d 1202{
4b480c67 1203 discard_revindex();
637cdd9d
JK
1204 prepare_packed_git_run_once = 0;
1205 prepare_packed_git();
1f688557
JH
1206}
1207
8eca0b47
NP
1208static void mark_bad_packed_object(struct packed_git *p,
1209 const unsigned char *sha1)
1210{
1211 unsigned i;
1212 for (i = 0; i < p->num_bad_objects; i++)
1213 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1214 return;
1215 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1216 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1217 p->num_bad_objects++;
1218}
1219
b6c4cecc 1220static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1221{
1222 struct packed_git *p;
1223 unsigned i;
1224
1225 for (p = packed_git; p; p = p->next)
1226 for (i = 0; i < p->num_bad_objects; i++)
1227 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1228 return p;
1229 return NULL;
ac939109
NP
1230}
1231
090ea126
NTND
1232/*
1233 * With an in-core object data in "map", rehash it to make sure the
1234 * object name actually matches "sha1" to detect object corruption.
1235 * With "map" == NULL, try reading the object named with "sha1" using
1236 * the streaming interface and rehash it to do the same.
1237 */
1238int check_sha1_signature(const unsigned char *sha1, void *map,
1239 unsigned long size, const char *type)
0fcfd160
LT
1240{
1241 unsigned char real_sha1[20];
090ea126
NTND
1242 enum object_type obj_type;
1243 struct git_istream *st;
1244 git_SHA_CTX c;
1245 char hdr[32];
1246 int hdrlen;
1247
1248 if (map) {
1249 hash_sha1_file(map, size, type, real_sha1);
1250 return hashcmp(sha1, real_sha1) ? -1 : 0;
1251 }
1252
1253 st = open_istream(sha1, &obj_type, &size, NULL);
1254 if (!st)
1255 return -1;
1256
1257 /* Generate the header */
1258 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1259
1260 /* Sha1.. */
1261 git_SHA1_Init(&c);
1262 git_SHA1_Update(&c, hdr, hdrlen);
1263 for (;;) {
1264 char buf[1024 * 16];
1265 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1266
1267 if (!readlen)
1268 break;
1269 git_SHA1_Update(&c, buf, readlen);
1270 }
1271 git_SHA1_Final(real_sha1, &c);
1272 close_istream(st);
a89fccd2 1273 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1274}
1275
f4e51683 1276static int git_open_noatime(const char *name)
44d1c19e
LT
1277{
1278 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1279
f2e872aa
SP
1280 for (;;) {
1281 int fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1282 if (fd >= 0)
f2e872aa
SP
1283 return fd;
1284
f2e872aa
SP
1285 /* Might the failure be due to O_NOATIME? */
1286 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1287 sha1_file_open_flag = 0;
f2e872aa
SP
1288 continue;
1289 }
1290
1291 return -1;
44d1c19e 1292 }
44d1c19e
LT
1293}
1294
1295static int open_sha1_file(const unsigned char *sha1)
1296{
1297 int fd;
1298 char *name = sha1_file_name(sha1);
1299 struct alternate_object_database *alt;
1300
f4e51683 1301 fd = git_open_noatime(name);
44d1c19e
LT
1302 if (fd >= 0)
1303 return fd;
1304
1305 prepare_alt_odb();
1306 errno = ENOENT;
1307 for (alt = alt_odb_list; alt; alt = alt->next) {
1308 name = alt->name;
1309 fill_sha1_path(name, sha1);
f4e51683 1310 fd = git_open_noatime(alt->base);
44d1c19e
LT
1311 if (fd >= 0)
1312 return fd;
1313 }
1314 return -1;
1315}
1316
f0270efd 1317void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1318{
0fcfd160 1319 void *map;
144bde78 1320 int fd;
ace1534d 1321
44d1c19e
LT
1322 fd = open_sha1_file(sha1);
1323 map = NULL;
1324 if (fd >= 0) {
1325 struct stat st;
0fcfd160 1326
44d1c19e
LT
1327 if (!fstat(fd, &st)) {
1328 *size = xsize_t(st.st_size);
33e42de0
MM
1329 if (!*size) {
1330 /* mmap() is forbidden on empty files */
1331 error("object file %s is empty", sha1_file_name(sha1));
1332 return NULL;
1333 }
44d1c19e 1334 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1335 }
44d1c19e 1336 close(fd);
0fcfd160 1337 }
0fcfd160
LT
1338 return map;
1339}
1340
cc5c54e7
JH
1341/*
1342 * There used to be a second loose object header format which
1343 * was meant to mimic the in-pack format, allowing for direct
1344 * copy of the object data. This format turned up not to be
1345 * really worth it and we no longer write loose objects in that
1346 * format.
1347 */
1348static int experimental_loose_object(unsigned char *map)
bb6b8e4f
JH
1349{
1350 unsigned int word;
1351
1352 /*
7f684a2a
RT
1353 * We must determine if the buffer contains the standard
1354 * zlib-deflated stream or the experimental format based
1355 * on the in-pack object format. Compare the header byte
1356 * for each format:
1357 *
1358 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1359 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1360 *
1361 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1362 * in standard loose-object format, UNLESS it is a Git-pack
1363 * format object *exactly* 8 bytes in size when inflated.
1364 *
1365 * However, RFC1950 also specifies that the 1st 16-bit word
1366 * must be divisible by 31 - this checksum tells us our buffer
1367 * is in the standard format, giving a false positive only if
1368 * the 1st word of the Git-pack format object happens to be
1369 * divisible by 31, ie:
1370 * ((byte0 * 256) + byte1) % 31 = 0
1371 * => 0ttt10000www1000 % 31 = 0
1372 *
1373 * As it happens, this case can only arise for www=3 & ttt=1
1374 * - ie, a Commit object, which would have to be 8 bytes in
1375 * size. As no Commit can be that small, we find that the
1376 * combination of these two criteria (bitmask & checksum)
1377 * can always correctly determine the buffer format.
bb6b8e4f
JH
1378 */
1379 word = (map[0] << 8) + map[1];
7f684a2a 1380 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
bb6b8e4f 1381 return 0;
cc5c54e7
JH
1382 else
1383 return 1;
bb6b8e4f
JH
1384}
1385
09ded04b
NP
1386unsigned long unpack_object_header_buffer(const unsigned char *buf,
1387 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1388{
ad1ed5ee 1389 unsigned shift;
48fb7deb 1390 unsigned long size, c;
ad1ed5ee
JH
1391 unsigned long used = 0;
1392
1393 c = buf[used++];
1394 *type = (c >> 4) & 7;
1395 size = c & 15;
1396 shift = 4;
1397 while (c & 0x80) {
f630cfda 1398 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1399 error("bad object header");
ea4f9685
JH
1400 size = used = 0;
1401 break;
09ded04b 1402 }
ad1ed5ee
JH
1403 c = buf[used++];
1404 size += (c & 0x7f) << shift;
1405 shift += 7;
1406 }
1407 *sizep = size;
1408 return used;
1409}
1410
eb4f4076 1411int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee
JH
1412{
1413 unsigned long size, used;
1414 static const char valid_loose_object_type[8] = {
1415 0, /* OBJ_EXT */
1416 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1417 0, /* "delta" and others are invalid in a loose object */
93821bd9 1418 };
ad1ed5ee 1419 enum object_type type;
93821bd9 1420
c4483576
LT
1421 /* Get the data stream */
1422 memset(stream, 0, sizeof(*stream));
1423 stream->next_in = map;
1424 stream->avail_in = mapsize;
1425 stream->next_out = buffer;
93821bd9
LT
1426 stream->avail_out = bufsiz;
1427
cc5c54e7
JH
1428 if (experimental_loose_object(map)) {
1429 /*
1430 * The old experimental format we no longer produce;
1431 * we can still read it.
1432 */
1433 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1434 if (!used || !valid_loose_object_type[type])
1435 return -1;
1436 map += used;
1437 mapsize -= used;
726f852b 1438
cc5c54e7
JH
1439 /* Set up the stream for the rest.. */
1440 stream->next_in = map;
1441 stream->avail_in = mapsize;
1442 git_inflate_init(stream);
93821bd9 1443
cc5c54e7
JH
1444 /* And generate the fake traditional header */
1445 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1446 typename(type), size);
1447 return 0;
1448 }
39c68542 1449 git_inflate_init(stream);
cc5c54e7 1450 return git_inflate(stream, 0);
c4483576
LT
1451}
1452
ef49a7a0 1453static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1454{
1455 int bytes = strlen(buffer) + 1;
3aee68aa 1456 unsigned char *buf = xmallocz(size);
93821bd9 1457 unsigned long n;
7efbff75 1458 int status = Z_OK;
5180cacc 1459
93821bd9
LT
1460 n = stream->total_out - bytes;
1461 if (n > size)
1462 n = size;
1463 memcpy(buf, (char *) buffer + bytes, n);
1464 bytes = n;
456cdf6e
LT
1465 if (bytes <= size) {
1466 /*
1467 * The above condition must be (bytes <= size), not
1468 * (bytes < size). In other words, even though we
ccf5ace0 1469 * expect no more output and set avail_out to zero,
456cdf6e
LT
1470 * the input zlib stream may have bytes that express
1471 * "this concludes the stream", and we *do* want to
1472 * eat that input.
1473 *
1474 * Otherwise we would not be able to test that we
1475 * consumed all the input to reach the expected size;
1476 * we also want to check that zlib tells us that all
1477 * went well with status == Z_STREAM_END at the end.
1478 */
5180cacc
LT
1479 stream->next_out = buf + bytes;
1480 stream->avail_out = size - bytes;
7efbff75 1481 while (status == Z_OK)
39c68542 1482 status = git_inflate(stream, Z_FINISH);
5180cacc 1483 }
456cdf6e 1484 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1485 git_inflate_end(stream);
7efbff75
JH
1486 return buf;
1487 }
1488
1489 if (status < 0)
1490 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1491 else if (stream->avail_in)
1492 error("garbage at end of loose object '%s'",
1493 sha1_to_hex(sha1));
1494 free(buf);
1495 return NULL;
5180cacc
LT
1496}
1497
1498/*
1499 * We used to just use "sscanf()", but that's actually way
1500 * too permissive for what we want to check. So do an anal
1501 * object header parse by hand.
1502 */
f0270efd 1503int parse_sha1_header(const char *hdr, unsigned long *sizep)
5180cacc 1504{
21666f1a 1505 char type[10];
5180cacc
LT
1506 int i;
1507 unsigned long size;
1508
1509 /*
a6080a0a 1510 * The type can be at most ten bytes (including the
5180cacc 1511 * terminating '\0' that we add), and is followed by
21666f1a 1512 * a space.
5180cacc 1513 */
21666f1a 1514 i = 0;
5180cacc
LT
1515 for (;;) {
1516 char c = *hdr++;
1517 if (c == ' ')
1518 break;
21666f1a
NP
1519 type[i++] = c;
1520 if (i >= sizeof(type))
5180cacc 1521 return -1;
5180cacc 1522 }
21666f1a 1523 type[i] = 0;
5180cacc
LT
1524
1525 /*
1526 * The length must follow immediately, and be in canonical
1527 * decimal format (ie "010" is not valid).
1528 */
1529 size = *hdr++ - '0';
1530 if (size > 9)
1531 return -1;
1532 if (size) {
1533 for (;;) {
1534 unsigned long c = *hdr - '0';
1535 if (c > 9)
1536 break;
1537 hdr++;
1538 size = size * 10 + c;
1539 }
1540 }
1541 *sizep = size;
1542
1543 /*
1544 * The length must be followed by a zero byte
1545 */
21666f1a 1546 return *hdr ? -1 : type_from_string(type);
5180cacc
LT
1547}
1548
7efbff75 1549static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1550{
5180cacc 1551 int ret;
ef49a7a0 1552 git_zstream stream;
5180cacc 1553 char hdr[8192];
0fcfd160 1554
5180cacc 1555 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1556 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1557 return NULL;
1558
7efbff75 1559 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1560}
1561
54dab52a
NP
1562unsigned long get_size_from_delta(struct packed_git *p,
1563 struct pack_window **w_curs,
1564 off_t curpos)
1565{
1566 const unsigned char *data;
1567 unsigned char delta_head[20], *in;
ef49a7a0 1568 git_zstream stream;
54dab52a
NP
1569 int st;
1570
1571 memset(&stream, 0, sizeof(stream));
1572 stream.next_out = delta_head;
1573 stream.avail_out = sizeof(delta_head);
1574
39c68542 1575 git_inflate_init(&stream);
54dab52a
NP
1576 do {
1577 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1578 stream.next_in = in;
39c68542 1579 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1580 curpos += stream.next_in - in;
1581 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1582 stream.total_out < sizeof(delta_head));
39c68542 1583 git_inflate_end(&stream);
3d77d877
NP
1584 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1585 error("delta data unpack-initial failed");
1586 return 0;
1587 }
54dab52a
NP
1588
1589 /* Examine the initial part of the delta to figure out
1590 * the result size.
1591 */
1592 data = delta_head;
1593
1594 /* ignore base size */
1595 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1596
1597 /* Read the result size */
1598 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1599}
1600
c4001d92 1601static off_t get_delta_base(struct packed_git *p,
03e79c88 1602 struct pack_window **w_curs,
c4001d92 1603 off_t *curpos,
21666f1a 1604 enum object_type type,
c4001d92 1605 off_t delta_obj_offset)
eb32d236 1606{
2b87c45b 1607 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1608 off_t base_offset;
eb32d236 1609
8d8a4ea5
SP
1610 /* use_pack() assured us we have [base_info, base_info + 20)
1611 * as a range that we can look at without walking off the
1612 * end of the mapped window. Its actually the hash size
1613 * that is assured. An OFS_DELTA longer than the hash size
1614 * is stupid, as then a REF_DELTA would be smaller to store.
1615 */
21666f1a 1616 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1617 unsigned used = 0;
1618 unsigned char c = base_info[used++];
1619 base_offset = c & 127;
1620 while (c & 128) {
1621 base_offset += 1;
8723f216 1622 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1623 return 0; /* overflow */
eb32d236
NP
1624 c = base_info[used++];
1625 base_offset = (base_offset << 7) + (c & 127);
1626 }
1627 base_offset = delta_obj_offset - base_offset;
d8f32556 1628 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1629 return 0; /* out of bound */
2b87c45b 1630 *curpos += used;
21666f1a 1631 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1632 /* The base entry _must_ be in the same pack */
1633 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1634 *curpos += 20;
eb32d236
NP
1635 } else
1636 die("I am totally screwed");
2b87c45b 1637 return base_offset;
eb32d236
NP
1638}
1639
f3bf9224 1640/* forward declaration for a mutually recursive function */
c4001d92 1641static int packed_object_info(struct packed_git *p, off_t offset,
9a490590 1642 unsigned long *sizep, int *rtype);
f3bf9224 1643
43057304 1644static int packed_delta_info(struct packed_git *p,
03e79c88 1645 struct pack_window **w_curs,
c4001d92 1646 off_t curpos,
21666f1a 1647 enum object_type type,
c4001d92 1648 off_t obj_offset,
43057304 1649 unsigned long *sizep)
5db47c2b 1650{
c4001d92 1651 off_t base_offset;
f3bf9224 1652
21666f1a 1653 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
3d77d877
NP
1654 if (!base_offset)
1655 return OBJ_BAD;
9a490590 1656 type = packed_object_info(p, base_offset, NULL, NULL);
3d77d877 1657 if (type <= OBJ_NONE) {
08698b1e
NP
1658 struct revindex_entry *revidx;
1659 const unsigned char *base_sha1;
1660 revidx = find_pack_revindex(p, base_offset);
1661 if (!revidx)
1662 return OBJ_BAD;
1663 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
3d77d877
NP
1664 mark_bad_packed_object(p, base_sha1);
1665 type = sha1_object_info(base_sha1, NULL);
1666 if (type <= OBJ_NONE)
1667 return OBJ_BAD;
1668 }
f3bf9224 1669
c62266f3
JH
1670 /* We choose to only get the type of the base object and
1671 * ignore potentially corrupt pack file that expects the delta
1672 * based on a base with a wrong size. This saves tons of
1673 * inflate() calls.
1674 */
3d77d877 1675 if (sizep) {
54dab52a 1676 *sizep = get_size_from_delta(p, w_curs, curpos);
3d77d877
NP
1677 if (*sizep == 0)
1678 type = OBJ_BAD;
1679 }
21666f1a
NP
1680
1681 return type;
5db47c2b
JH
1682}
1683
f8c8abc5
JH
1684int unpack_object_header(struct packed_git *p,
1685 struct pack_window **w_curs,
1686 off_t *curpos,
1687 unsigned long *sizep)
a733cb60 1688{
03e79c88 1689 unsigned char *base;
ef49a7a0 1690 unsigned long left;
ad1ed5ee 1691 unsigned long used;
2b87c45b 1692 enum object_type type;
a733cb60 1693
8d8a4ea5 1694 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1695 * as a range that we can look at. (Its actually the hash
3dff5379 1696 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1697 * the maximum deflated object size is 2^137, which is just
1698 * insane, so we know won't exceed what we have been given.
1699 */
2b87c45b 1700 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1701 used = unpack_object_header_buffer(base, left, &type, sizep);
1702 if (!used) {
1703 type = OBJ_BAD;
1704 } else
1705 *curpos += used;
ad1ed5ee 1706
2b87c45b 1707 return type;
a733cb60
LT
1708}
1709
c4001d92 1710static int packed_object_info(struct packed_git *p, off_t obj_offset,
9a490590 1711 unsigned long *sizep, int *rtype)
1f688557 1712{
03e79c88 1713 struct pack_window *w_curs = NULL;
c4001d92
SP
1714 unsigned long size;
1715 off_t curpos = obj_offset;
21666f1a 1716 enum object_type type;
5db47c2b 1717
21666f1a 1718 type = unpack_object_header(p, &w_curs, &curpos, &size);
9a490590
JH
1719 if (rtype)
1720 *rtype = type; /* representation type */
5db47c2b 1721
21666f1a 1722 switch (type) {
eb32d236
NP
1723 case OBJ_OFS_DELTA:
1724 case OBJ_REF_DELTA:
21666f1a
NP
1725 type = packed_delta_info(p, &w_curs, curpos,
1726 type, obj_offset, sizep);
2b87c45b 1727 break;
a733cb60 1728 case OBJ_COMMIT:
a733cb60 1729 case OBJ_TREE:
a733cb60 1730 case OBJ_BLOB:
a733cb60 1731 case OBJ_TAG:
2b87c45b
NP
1732 if (sizep)
1733 *sizep = size;
a69d0943 1734 break;
1f688557 1735 default:
3d77d877
NP
1736 error("unknown object type %i at offset %"PRIuMAX" in %s",
1737 type, (uintmax_t)obj_offset, p->pack_name);
1738 type = OBJ_BAD;
1f688557 1739 }
2b87c45b 1740 unuse_pack(&w_curs);
21666f1a 1741 return type;
1f688557
JH
1742}
1743
eb950c19 1744static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 1745 struct pack_window **w_curs,
c4001d92 1746 off_t curpos,
eb950c19 1747 unsigned long size)
de530aaa
SP
1748{
1749 int st;
ef49a7a0 1750 git_zstream stream;
079afb18 1751 unsigned char *buffer, *in;
de530aaa 1752
4ab07e4d 1753 buffer = xmallocz(size);
de530aaa 1754 memset(&stream, 0, sizeof(stream));
de530aaa 1755 stream.next_out = buffer;
39eea7bd 1756 stream.avail_out = size + 1;
de530aaa 1757
39c68542 1758 git_inflate_init(&stream);
079afb18 1759 do {
2b87c45b 1760 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 1761 stream.next_in = in;
39c68542 1762 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
1763 if (!stream.avail_out)
1764 break; /* the payload is larger than it should be */
2b87c45b 1765 curpos += stream.next_in - in;
079afb18 1766 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 1767 git_inflate_end(&stream);
de530aaa
SP
1768 if ((st != Z_STREAM_END) || stream.total_out != size) {
1769 free(buffer);
1770 return NULL;
1771 }
1772
1773 return buffer;
1774}
1775
e5e01619
LT
1776#define MAX_DELTA_CACHE (256)
1777
18bdec11 1778static size_t delta_base_cached;
5e08ecbf
NP
1779
1780static struct delta_base_cache_lru_list {
1781 struct delta_base_cache_lru_list *prev;
1782 struct delta_base_cache_lru_list *next;
1783} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1784
e5e01619 1785static struct delta_base_cache_entry {
5e08ecbf
NP
1786 struct delta_base_cache_lru_list lru;
1787 void *data;
e5e01619
LT
1788 struct packed_git *p;
1789 off_t base_offset;
1790 unsigned long size;
e5e01619
LT
1791 enum object_type type;
1792} delta_base_cache[MAX_DELTA_CACHE];
1793
1794static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1795{
1796 unsigned long hash;
1797
1798 hash = (unsigned long)p + (unsigned long)base_offset;
1799 hash += (hash >> 8) + (hash >> 16);
3358004a 1800 return hash % MAX_DELTA_CACHE;
e5e01619
LT
1801}
1802
5266d369
JH
1803static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1804{
1805 unsigned long hash = pack_entry_hash(p, base_offset);
1806 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1807 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1808}
1809
62f255ad 1810static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 1811 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 1812{
e5e01619
LT
1813 void *ret;
1814 unsigned long hash = pack_entry_hash(p, base_offset);
1815 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1816
1817 ret = ent->data;
749bc58c
MV
1818 if (!ret || ent->p != p || ent->base_offset != base_offset)
1819 return unpack_entry(p, base_offset, type, base_size);
e5e01619 1820
18bdec11 1821 if (!keep_cache) {
a0cba108 1822 ent->data = NULL;
5e08ecbf
NP
1823 ent->lru.next->prev = ent->lru.prev;
1824 ent->lru.prev->next = ent->lru.next;
18bdec11 1825 delta_base_cached -= ent->size;
182af834
PH
1826 } else {
1827 ret = xmemdupz(ent->data, ent->size);
a0cba108 1828 }
e5e01619
LT
1829 *type = ent->type;
1830 *base_size = ent->size;
1831 return ret;
62f255ad
LT
1832}
1833
18bdec11
SP
1834static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1835{
1836 if (ent->data) {
1837 free(ent->data);
1838 ent->data = NULL;
5e08ecbf
NP
1839 ent->lru.next->prev = ent->lru.prev;
1840 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
1841 delta_base_cached -= ent->size;
1842 }
1843}
1844
3d20c636
SP
1845void clear_delta_base_cache(void)
1846{
1847 unsigned long p;
1848 for (p = 0; p < MAX_DELTA_CACHE; p++)
1849 release_delta_base_cache(&delta_base_cache[p]);
1850}
1851
62f255ad
LT
1852static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1853 void *base, unsigned long base_size, enum object_type type)
1854{
5e08ecbf 1855 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 1856 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 1857 struct delta_base_cache_lru_list *lru;
e5e01619 1858
18bdec11
SP
1859 release_delta_base_cache(ent);
1860 delta_base_cached += base_size;
5e08ecbf
NP
1861
1862 for (lru = delta_base_cache_lru.next;
1863 delta_base_cached > delta_base_cache_limit
1864 && lru != &delta_base_cache_lru;
1865 lru = lru->next) {
1866 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
1867 if (f->type == OBJ_BLOB)
1868 release_delta_base_cache(f);
1869 }
5e08ecbf
NP
1870 for (lru = delta_base_cache_lru.next;
1871 delta_base_cached > delta_base_cache_limit
1872 && lru != &delta_base_cache_lru;
1873 lru = lru->next) {
1874 struct delta_base_cache_entry *f = (void *)lru;
1875 release_delta_base_cache(f);
1876 }
18bdec11 1877
e5e01619
LT
1878 ent->p = p;
1879 ent->base_offset = base_offset;
1880 ent->type = type;
1881 ent->data = base;
1882 ent->size = base_size;
5e08ecbf
NP
1883 ent->lru.next = &delta_base_cache_lru;
1884 ent->lru.prev = delta_base_cache_lru.prev;
1885 delta_base_cache_lru.prev->next = &ent->lru;
1886 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
1887}
1888
c2c5b270
CC
1889static void *read_object(const unsigned char *sha1, enum object_type *type,
1890 unsigned long *size);
1891
eb950c19 1892static void *unpack_delta_entry(struct packed_git *p,
03e79c88 1893 struct pack_window **w_curs,
c4001d92 1894 off_t curpos,
1f688557 1895 unsigned long delta_size,
c4001d92 1896 off_t obj_offset,
21666f1a 1897 enum object_type *type,
eb950c19 1898 unsigned long *sizep)
1f688557 1899{
7c3e8be3 1900 void *delta_data, *result, *base;
c4001d92
SP
1901 unsigned long base_size;
1902 off_t base_offset;
43057304 1903
21666f1a 1904 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
8eca0b47
NP
1905 if (!base_offset) {
1906 error("failed to validate delta base reference "
1907 "at offset %"PRIuMAX" from %s",
1908 (uintmax_t)curpos, p->pack_name);
1909 return NULL;
1910 }
eac12e2d 1911 unuse_pack(w_curs);
a0cba108 1912 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
8eca0b47
NP
1913 if (!base) {
1914 /*
1915 * We're probably in deep shit, but let's try to fetch
1916 * the required base anyway from another pack or loose.
1917 * This is costly but should happen only in the presence
1918 * of a corrupted pack, and is better than failing outright.
1919 */
08698b1e
NP
1920 struct revindex_entry *revidx;
1921 const unsigned char *base_sha1;
1922 revidx = find_pack_revindex(p, base_offset);
1923 if (!revidx)
1924 return NULL;
1925 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
8eca0b47
NP
1926 error("failed to read delta base object %s"
1927 " at offset %"PRIuMAX" from %s",
1928 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1929 p->pack_name);
1930 mark_bad_packed_object(p, base_sha1);
ac939109 1931 base = read_object(base_sha1, type, &base_size);
8eca0b47
NP
1932 if (!base)
1933 return NULL;
1934 }
67686d95 1935
2b87c45b 1936 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
8eca0b47
NP
1937 if (!delta_data) {
1938 error("failed to unpack compressed delta "
1939 "at offset %"PRIuMAX" from %s",
1940 (uintmax_t)curpos, p->pack_name);
1941 free(base);
1942 return NULL;
1943 }
1f688557
JH
1944 result = patch_delta(base, base_size,
1945 delta_data, delta_size,
21666f1a 1946 sizep);
1f688557
JH
1947 if (!result)
1948 die("failed to apply delta");
1949 free(delta_data);
62f255ad 1950 add_delta_base_cache(p, base_offset, base, base_size, *type);
1f688557
JH
1951 return result;
1952}
1953
5f44324d
JH
1954static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1955{
1956 static FILE *log_file;
1957
1958 if (!log_file) {
1959 log_file = fopen(log_pack_access, "w");
1960 if (!log_file) {
1961 error("cannot open pack access log '%s' for writing: %s",
1962 log_pack_access, strerror(errno));
1963 log_pack_access = NULL;
1964 return;
1965 }
1966 }
1967 fprintf(log_file, "%s %"PRIuMAX"\n",
1968 p->pack_name, (uintmax_t)obj_offset);
1969 fflush(log_file);
1970}
1971
0e8189e2
NP
1972int do_check_packed_object_crc;
1973
c4001d92 1974void *unpack_entry(struct packed_git *p, off_t obj_offset,
21666f1a 1975 enum object_type *type, unsigned long *sizep)
1f688557 1976{
03e79c88 1977 struct pack_window *w_curs = NULL;
c4001d92 1978 off_t curpos = obj_offset;
21666f1a 1979 void *data;
1f688557 1980
5f44324d
JH
1981 if (log_pack_access)
1982 write_pack_access_log(p, obj_offset);
1983
0e8189e2
NP
1984 if (do_check_packed_object_crc && p->index_version > 1) {
1985 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1986 unsigned long len = revidx[1].offset - obj_offset;
1987 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1988 const unsigned char *sha1 =
1989 nth_packed_object_sha1(p, revidx->nr);
1990 error("bad packed object CRC for %s",
1991 sha1_to_hex(sha1));
1992 mark_bad_packed_object(p, sha1);
f755bb99 1993 unuse_pack(&w_curs);
0e8189e2
NP
1994 return NULL;
1995 }
1996 }
1997
21666f1a
NP
1998 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1999 switch (*type) {
eb32d236
NP
2000 case OBJ_OFS_DELTA:
2001 case OBJ_REF_DELTA:
21666f1a
NP
2002 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
2003 obj_offset, type, sizep);
4d703a1a 2004 break;
a733cb60 2005 case OBJ_COMMIT:
a733cb60 2006 case OBJ_TREE:
a733cb60 2007 case OBJ_BLOB:
a733cb60 2008 case OBJ_TAG:
21666f1a 2009 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
4d703a1a 2010 break;
1f688557 2011 default:
8eca0b47
NP
2012 data = NULL;
2013 error("unknown object type %i at offset %"PRIuMAX" in %s",
2014 *type, (uintmax_t)obj_offset, p->pack_name);
1f688557 2015 }
03e79c88 2016 unuse_pack(&w_curs);
21666f1a 2017 return data;
1f688557
JH
2018}
2019
d079837e 2020const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2021 uint32_t n)
9a217f2a 2022{
42873078 2023 const unsigned char *index = p->index_data;
d079837e
SP
2024 if (!index) {
2025 if (open_pack_index(p))
2026 return NULL;
2027 index = p->index_data;
2028 }
57059091 2029 if (n >= p->num_objects)
d72308e0 2030 return NULL;
74e34e1f
NP
2031 index += 4 * 256;
2032 if (p->index_version == 1) {
2033 return index + 24 * n + 4;
2034 } else {
2035 index += 8;
2036 return index + 20 * n;
2037 }
2038}
2039
99093238 2040off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2041{
2042 const unsigned char *index = p->index_data;
2043 index += 4 * 256;
2044 if (p->index_version == 1) {
2045 return ntohl(*((uint32_t *)(index + 24 * n)));
2046 } else {
2047 uint32_t off;
2048 index += 8 + p->num_objects * (20 + 4);
2049 off = ntohl(*((uint32_t *)(index + 4 * n)));
2050 if (!(off & 0x80000000))
2051 return off;
2052 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2053 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2054 ntohl(*((uint32_t *)(index + 4)));
2055 }
9a217f2a
JH
2056}
2057
c4001d92 2058off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2059 struct packed_git *p)
1f688557 2060{
42873078 2061 const uint32_t *level1_ofs = p->index_data;
42873078 2062 const unsigned char *index = p->index_data;
628522ec
JH
2063 unsigned hi, lo, stride;
2064 static int use_lookup = -1;
2065 static int debug_lookup = -1;
2066
2067 if (debug_lookup < 0)
2068 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2069
d079837e
SP
2070 if (!index) {
2071 if (open_pack_index(p))
2072 return 0;
2073 level1_ofs = p->index_data;
2074 index = p->index_data;
2075 }
74e34e1f
NP
2076 if (p->index_version > 1) {
2077 level1_ofs += 2;
2078 index += 8;
2079 }
42873078 2080 index += 4 * 256;
74e34e1f
NP
2081 hi = ntohl(level1_ofs[*sha1]);
2082 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2083 if (p->index_version > 1) {
2084 stride = 20;
2085 } else {
2086 stride = 24;
2087 index += 4;
2088 }
2089
2090 if (debug_lookup)
6e1c2344 2091 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2092 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2093
2094 if (use_lookup < 0)
2095 use_lookup = !!getenv("GIT_USE_LOOKUP");
2096 if (use_lookup) {
2097 int pos = sha1_entry_pos(index, stride, 0,
2098 lo, hi, p->num_objects, sha1);
2099 if (pos < 0)
2100 return 0;
2101 return nth_packed_object_offset(p, pos);
2102 }
1f688557
JH
2103
2104 do {
74e34e1f 2105 unsigned mi = (lo + hi) / 2;
628522ec
JH
2106 int cmp = hashcmp(index + mi * stride, sha1);
2107
2108 if (debug_lookup)
2109 printf("lo %u hi %u rg %u mi %u\n",
2110 lo, hi, hi - lo, mi);
43057304 2111 if (!cmp)
74e34e1f 2112 return nth_packed_object_offset(p, mi);
1f688557
JH
2113 if (cmp > 0)
2114 hi = mi;
2115 else
2116 lo = mi+1;
2117 } while (lo < hi);
2118 return 0;
2119}
2120
4c080182 2121int is_pack_valid(struct packed_git *p)
d131b7af
SP
2122{
2123 /* An already open pack is known to be valid. */
2124 if (p->pack_fd != -1)
2125 return 1;
2126
2127 /* If the pack has one window completely covering the
2128 * file size, the pack is known to be valid even if
2129 * the descriptor is not currently open.
2130 */
2131 if (p->windows) {
2132 struct pack_window *w = p->windows;
2133
2134 if (!w->offset && w->len == p->pack_size)
2135 return 1;
2136 }
2137
2138 /* Force the pack to open to prove its valid. */
2139 return !open_packed_git(p);
2140}
2141
95099731
NTND
2142static int fill_pack_entry(const unsigned char *sha1,
2143 struct pack_entry *e,
2144 struct packed_git *p)
2145{
2146 off_t offset;
2147
2148 if (p->num_bad_objects) {
2149 unsigned i;
2150 for (i = 0; i < p->num_bad_objects; i++)
2151 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2152 return 0;
2153 }
2154
2155 offset = find_pack_entry_one(sha1, p);
2156 if (!offset)
2157 return 0;
2158
2159 /*
2160 * We are about to tell the caller where they can locate the
2161 * requested object. We better make sure the packfile is
2162 * still here and can be accessed before supplying that
2163 * answer, as it may have been deleted since the index was
2164 * loaded!
2165 */
2166 if (!is_pack_valid(p)) {
2167 warning("packfile %s cannot be accessed", p->pack_name);
2168 return 0;
2169 }
2170 e->offset = offset;
2171 e->p = p;
2172 hashcpy(e->sha1, sha1);
2173 return 1;
2174}
2175
4d6acb70 2176static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2177{
2178 struct packed_git *p;
43057304 2179
1f688557 2180 prepare_packed_git();
f7c22cc6
NP
2181 if (!packed_git)
2182 return 0;
1f688557 2183
c01f51cc
NTND
2184 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2185 return 1;
8eca0b47 2186
c01f51cc
NTND
2187 for (p = packed_git; p; p = p->next) {
2188 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2189 continue;
f7c22cc6 2190
c01f51cc
NTND
2191 last_found_pack = p;
2192 return 1;
2193 }
1f688557
JH
2194 return 0;
2195}
2196
a6080a0a 2197struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2198 struct packed_git *packs)
2199{
2200 struct packed_git *p;
bf592c50
DB
2201
2202 for (p = packs; p; p = p->next) {
43057304 2203 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2204 return p;
2205 }
2206 return NULL;
21666f1a 2207
bf592c50
DB
2208}
2209
21666f1a 2210static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
65c2e0c3 2211{
36e4d74a 2212 int status;
65c2e0c3
JH
2213 unsigned long mapsize, size;
2214 void *map;
ef49a7a0 2215 git_zstream stream;
d65a16f6 2216 char hdr[32];
65c2e0c3 2217
bb6b8e4f 2218 map = map_sha1_file(sha1, &mapsize);
f0df4ed5
JS
2219 if (!map)
2220 return error("unable to find %s", sha1_to_hex(sha1));
36e4d74a
JH
2221 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2222 status = error("unable to unpack %s header",
2223 sha1_to_hex(sha1));
21666f1a 2224 else if ((status = parse_sha1_header(hdr, &size)) < 0)
36e4d74a 2225 status = error("unable to parse %s header", sha1_to_hex(sha1));
21666f1a
NP
2226 else if (sizep)
2227 *sizep = size;
39c68542 2228 git_inflate_end(&stream);
65c2e0c3
JH
2229 munmap(map, mapsize);
2230 return status;
2231}
2232
9a490590
JH
2233/* returns enum object_type or negative */
2234int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
f0df4ed5 2235{
c4d9986f 2236 struct cached_object *co;
f0df4ed5 2237 struct pack_entry e;
9a490590 2238 int status, rtype;
f0df4ed5 2239
c4d9986f
NTND
2240 co = find_cached_object(sha1);
2241 if (co) {
9a490590
JH
2242 if (oi->sizep)
2243 *(oi->sizep) = co->size;
2244 oi->whence = OI_CACHED;
c4d9986f
NTND
2245 return co->type;
2246 }
2247
cd673c1f 2248 if (!find_pack_entry(sha1, &e)) {
ddd63e64 2249 /* Most likely it's a loose object. */
9a490590
JH
2250 status = sha1_loose_object_info(sha1, oi->sizep);
2251 if (status >= 0) {
2252 oi->whence = OI_LOOSE;
ddd63e64 2253 return status;
9a490590 2254 }
ddd63e64
SG
2255
2256 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2257 reprepare_packed_git();
cd673c1f 2258 if (!find_pack_entry(sha1, &e))
ddd63e64 2259 return status;
f0df4ed5 2260 }
3d77d877 2261
9a490590 2262 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
3d77d877
NP
2263 if (status < 0) {
2264 mark_bad_packed_object(e.p, sha1);
9a490590 2265 status = sha1_object_info_extended(sha1, oi);
5266d369
JH
2266 } else if (in_delta_base_cache(e.p, e.offset)) {
2267 oi->whence = OI_DBCACHED;
9a490590
JH
2268 } else {
2269 oi->whence = OI_PACKED;
2270 oi->u.packed.offset = e.offset;
2271 oi->u.packed.pack = e.p;
2272 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2273 rtype == OBJ_OFS_DELTA);
3d77d877
NP
2274 }
2275
2276 return status;
f0df4ed5
JS
2277}
2278
9a490590
JH
2279int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2280{
2281 struct object_info oi;
2282
2283 oi.sizep = sizep;
2284 return sha1_object_info_extended(sha1, &oi);
2285}
2286
21666f1a
NP
2287static void *read_packed_sha1(const unsigned char *sha1,
2288 enum object_type *type, unsigned long *size)
1f688557
JH
2289{
2290 struct pack_entry e;
8eca0b47 2291 void *data;
1f688557 2292
cd673c1f 2293 if (!find_pack_entry(sha1, &e))
1f688557 2294 return NULL;
8eca0b47
NP
2295 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2296 if (!data) {
2297 /*
2298 * We're probably in deep shit, but let's try to fetch
2299 * the required object anyway from another pack or loose.
2300 * This should happen only in the presence of a corrupted
2301 * pack, and is better than failing outright.
2302 */
2303 error("failed to read object %s at offset %"PRIuMAX" from %s",
2304 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2305 mark_bad_packed_object(e.p, sha1);
ac939109 2306 data = read_object(sha1, type, size);
8eca0b47
NP
2307 }
2308 return data;
1f688557
JH
2309}
2310
21666f1a
NP
2311int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2312 unsigned char *sha1)
d66b37bb
JH
2313{
2314 struct cached_object *co;
2315
21666f1a 2316 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2317 if (has_sha1_file(sha1) || find_cached_object(sha1))
2318 return 0;
2319 if (cached_object_alloc <= cached_object_nr) {
2320 cached_object_alloc = alloc_nr(cached_object_alloc);
2321 cached_objects = xrealloc(cached_objects,
2322 sizeof(*cached_objects) *
2323 cached_object_alloc);
2324 }
2325 co = &cached_objects[cached_object_nr++];
2326 co->size = len;
21666f1a 2327 co->type = type;
efa13f7b
JH
2328 co->buf = xmalloc(len);
2329 memcpy(co->buf, buf, len);
d66b37bb
JH
2330 hashcpy(co->sha1, sha1);
2331 return 0;
2332}
2333
c2c5b270
CC
2334static void *read_object(const unsigned char *sha1, enum object_type *type,
2335 unsigned long *size)
0fcfd160
LT
2336{
2337 unsigned long mapsize;
2338 void *map, *buf;
d66b37bb
JH
2339 struct cached_object *co;
2340
2341 co = find_cached_object(sha1);
2342 if (co) {
21666f1a 2343 *type = co->type;
d66b37bb 2344 *size = co->size;
182af834 2345 return xmemdupz(co->buf, co->size);
d66b37bb 2346 }
0fcfd160 2347
8276c007
PE
2348 buf = read_packed_sha1(sha1, type, size);
2349 if (buf)
2350 return buf;
bb6b8e4f 2351 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2352 if (map) {
7efbff75 2353 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2354 munmap(map, mapsize);
2355 return buf;
2356 }
637cdd9d 2357 reprepare_packed_git();
8276c007 2358 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2359}
2360
b6c4cecc
JH
2361/*
2362 * This function dies on corrupt objects; the callers who want to
2363 * deal with them should arrange to call read_object() and give error
2364 * messages themselves.
2365 */
5bf29b95
JH
2366void *read_sha1_file_extended(const unsigned char *sha1,
2367 enum object_type *type,
2368 unsigned long *size,
2369 unsigned flag)
ac939109 2370{
3ba7a065 2371 void *data;
e8b15e61 2372 char *path;
b6c4cecc 2373 const struct packed_git *p;
5bf29b95
JH
2374 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2375 ? lookup_replace_object(sha1) : sha1;
b6c4cecc 2376
3ba7a065
JH
2377 errno = 0;
2378 data = read_object(repl, type, size);
4bbf5a26 2379 if (data)
b6c4cecc 2380 return data;
68095570 2381
25f3af3f 2382 if (errno && errno != ENOENT)
3ba7a065
JH
2383 die_errno("failed to read object %s", sha1_to_hex(sha1));
2384
68095570 2385 /* die if we replaced an object with one that does not exist */
b6c4cecc 2386 if (repl != sha1)
68095570
CC
2387 die("replacement %s not found for %s",
2388 sha1_to_hex(repl), sha1_to_hex(sha1));
2389
b6c4cecc
JH
2390 if (has_loose_object(repl)) {
2391 path = sha1_file_name(sha1);
2392 die("loose object %s (stored in %s) is corrupt",
2393 sha1_to_hex(repl), path);
e8b15e61 2394 }
68095570 2395
b6c4cecc
JH
2396 if ((p = has_packed_and_bad(repl)) != NULL)
2397 die("packed object %s (stored in %s) is corrupt",
2398 sha1_to_hex(repl), p->pack_name);
f5552aee 2399
b6c4cecc 2400 return NULL;
ac939109
NP
2401}
2402
40469ee9 2403void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2404 const char *required_type_name,
40469ee9
JH
2405 unsigned long *size,
2406 unsigned char *actual_sha1_return)
f4913f91 2407{
21666f1a 2408 enum object_type type, required_type;
f4913f91
JH
2409 void *buffer;
2410 unsigned long isize;
40469ee9 2411 unsigned char actual_sha1[20];
f4913f91 2412
21666f1a 2413 required_type = type_from_string(required_type_name);
e702496e 2414 hashcpy(actual_sha1, sha1);
40469ee9
JH
2415 while (1) {
2416 int ref_length = -1;
2417 const char *ref_type = NULL;
f4913f91 2418
21666f1a 2419 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2420 if (!buffer)
2421 return NULL;
21666f1a 2422 if (type == required_type) {
40469ee9
JH
2423 *size = isize;
2424 if (actual_sha1_return)
e702496e 2425 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2426 return buffer;
2427 }
2428 /* Handle references */
21666f1a 2429 else if (type == OBJ_COMMIT)
40469ee9 2430 ref_type = "tree ";
21666f1a 2431 else if (type == OBJ_TAG)
40469ee9
JH
2432 ref_type = "object ";
2433 else {
2434 free(buffer);
2435 return NULL;
2436 }
2437 ref_length = strlen(ref_type);
f4913f91 2438
50974ec9
MK
2439 if (ref_length + 40 > isize ||
2440 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2441 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2442 free(buffer);
2443 return NULL;
2444 }
1cf58e72 2445 free(buffer);
40469ee9
JH
2446 /* Now we have the ID of the referred-to object in
2447 * actual_sha1. Check again. */
f4913f91 2448 }
f4913f91
JH
2449}
2450
ce9fbf16 2451static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2452 const char *type, unsigned char *sha1,
d65a16f6 2453 char *hdr, int *hdrlen)
d410c0f5 2454{
9126f009 2455 git_SHA_CTX c;
d410c0f5
JH
2456
2457 /* Generate the header */
d65a16f6 2458 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
d410c0f5
JH
2459
2460 /* Sha1.. */
9126f009
NP
2461 git_SHA1_Init(&c);
2462 git_SHA1_Update(&c, hdr, *hdrlen);
2463 git_SHA1_Update(&c, buf, len);
2464 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2465}
2466
230f1322 2467/*
5a688fe4
JH
2468 * Move the just written object into its final resting place.
2469 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2470 * "moving" is only a part of what it does, when no patch between
2471 * master to pu changes the call sites of this function.
230f1322 2472 */
839837b9 2473int move_temp_to_file(const char *tmpfile, const char *filename)
230f1322 2474{
e32c0a9c 2475 int ret = 0;
5a688fe4 2476
348df166 2477 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2478 goto try_rename;
2479 else if (link(tmpfile, filename))
e32c0a9c 2480 ret = errno;
7ebb6fca
LT
2481
2482 /*
2483 * Coda hack - coda doesn't like cross-directory links,
2484 * so we fall back to a rename, which will mean that it
2485 * won't be able to check collisions, but that's not a
2486 * big deal.
2487 *
2488 * The same holds for FAT formatted media.
2489 *
3be1f18e 2490 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2491 * left to unlink.
2492 */
2493 if (ret && ret != EEXIST) {
be66a6c4 2494 try_rename:
7ebb6fca 2495 if (!rename(tmpfile, filename))
3be1f18e 2496 goto out;
9e48b389 2497 ret = errno;
230f1322 2498 }
691f1a28 2499 unlink_or_warn(tmpfile);
230f1322
LT
2500 if (ret) {
2501 if (ret != EEXIST) {
82247e9b 2502 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2503 }
2504 /* FIXME!!! Collision check here ? */
2505 }
2506
3be1f18e 2507out:
5256b006 2508 if (adjust_shared_perm(filename))
5a688fe4 2509 return error("unable to set permission to '%s'", filename);
230f1322
LT
2510 return 0;
2511}
2512
4d548150
LT
2513static int write_buffer(int fd, const void *buf, size_t len)
2514{
d34cf19b 2515 if (write_in_full(fd, buf, len) < 0)
93822c22 2516 return error("file write error (%s)", strerror(errno));
4d548150
LT
2517 return 0;
2518}
2519
ce9fbf16 2520int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2521 unsigned char *sha1)
2522{
d65a16f6 2523 char hdr[32];
abdc3fc8
RS
2524 int hdrlen;
2525 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2526 return 0;
2527}
2528
e9039dd3
LT
2529/* Finalize a file on disk, and close it. */
2530static void close_sha1_file(int fd)
2531{
aafe9fba
LT
2532 if (fsync_object_files)
2533 fsync_or_die(fd, "sha1 file");
e9039dd3 2534 if (close(fd) != 0)
d824cbba 2535 die_errno("error when closing sha1 file");
e9039dd3
LT
2536}
2537
5723fe7e
LT
2538/* Size of directory component, including the ending '/' */
2539static inline int directory_size(const char *filename)
2540{
2541 const char *s = strrchr(filename, '/');
2542 if (!s)
2543 return 0;
2544 return s - filename + 1;
2545}
2546
2547/*
2548 * This creates a temporary file in the same directory as the final
2549 * 'filename'
2550 *
2551 * We want to avoid cross-directory filename renames, because those
2552 * can have problems on various filesystems (FAT, NFS, Coda).
2553 */
2554static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2555{
2556 int fd, dirlen = directory_size(filename);
2557
2558 if (dirlen + 20 > bufsiz) {
2559 errno = ENAMETOOLONG;
2560 return -1;
2561 }
2562 memcpy(buffer, filename, dirlen);
2563 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
5256b006 2564 fd = git_mkstemp_mode(buffer, 0444);
cbacbf4e 2565 if (fd < 0 && dirlen && errno == ENOENT) {
5723fe7e 2566 /* Make sure the directory exists */
6ff6af62 2567 memcpy(buffer, filename, dirlen);
5723fe7e 2568 buffer[dirlen-1] = 0;
3bfaf018 2569 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
5723fe7e
LT
2570 return -1;
2571
2572 /* Try again */
2573 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
5256b006 2574 fd = git_mkstemp_mode(buffer, 0444);
5723fe7e
LT
2575 }
2576 return fd;
2577}
2578
bbac7311 2579static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 2580 const void *buf, unsigned long len, time_t mtime)
0fcfd160 2581{
915308b1 2582 int fd, ret;
9892beba 2583 unsigned char compressed[4096];
ef49a7a0 2584 git_zstream stream;
748af44c
NP
2585 git_SHA_CTX c;
2586 unsigned char parano_sha1[20];
706bc531 2587 char *filename;
ab1900a3 2588 static char tmp_file[PATH_MAX];
a44c9a5e 2589
972a9155 2590 filename = sha1_file_name(sha1);
ab1900a3 2591 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
aac17941 2592 if (fd < 0) {
35243577 2593 if (errno == EACCES)
82247e9b 2594 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 2595 else
5eaeda70 2596 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
2597 }
2598
0fcfd160
LT
2599 /* Set it up */
2600 memset(&stream, 0, sizeof(stream));
55bb5c91 2601 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 2602 stream.next_out = compressed;
9892beba 2603 stream.avail_out = sizeof(compressed);
748af44c 2604 git_SHA1_Init(&c);
a44c9a5e
LT
2605
2606 /* First header.. */
d65a16f6 2607 stream.next_in = (unsigned char *)hdr;
a44c9a5e 2608 stream.avail_in = hdrlen;
55bb5c91
JH
2609 while (git_deflate(&stream, 0) == Z_OK)
2610 ; /* nothing */
748af44c 2611 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
2612
2613 /* Then the data itself.. */
c00e657d 2614 stream.next_in = (void *)buf;
a44c9a5e 2615 stream.avail_in = len;
9892beba 2616 do {
748af44c 2617 unsigned char *in0 = stream.next_in;
55bb5c91 2618 ret = git_deflate(&stream, Z_FINISH);
748af44c 2619 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
2620 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2621 die("unable to write sha1 file");
2622 stream.next_out = compressed;
2623 stream.avail_out = sizeof(compressed);
2624 } while (ret == Z_OK);
2625
ac54c277
LT
2626 if (ret != Z_STREAM_END)
2627 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 2628 ret = git_deflate_end_gently(&stream);
ac54c277
LT
2629 if (ret != Z_OK)
2630 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
2631 git_SHA1_Final(parano_sha1, &c);
2632 if (hashcmp(sha1, parano_sha1) != 0)
2633 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 2634
e9039dd3 2635 close_sha1_file(fd);
0fcfd160 2636
bbac7311
NP
2637 if (mtime) {
2638 struct utimbuf utb;
2639 utb.actime = mtime;
2640 utb.modtime = mtime;
ab1900a3 2641 if (utime(tmp_file, &utb) < 0)
bbac7311 2642 warning("failed utime() on %s: %s",
ab1900a3 2643 tmp_file, strerror(errno));
bbac7311
NP
2644 }
2645
ab1900a3 2646 return move_temp_to_file(tmp_file, filename);
0fcfd160 2647}
8237b185 2648
c00e657d 2649int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
bbac7311
NP
2650{
2651 unsigned char sha1[20];
2652 char hdr[32];
2653 int hdrlen;
2654
2655 /* Normally if we have it in the pack then we do not bother writing
2656 * it out into .git/objects/??/?{38} file.
2657 */
2658 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2659 if (returnsha1)
2660 hashcpy(returnsha1, sha1);
2661 if (has_sha1_file(sha1))
2662 return 0;
2663 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2664}
2665
2666int force_object_loose(const unsigned char *sha1, time_t mtime)
2667{
bbac7311
NP
2668 void *buf;
2669 unsigned long len;
2670 enum object_type type;
2671 char hdr[32];
2672 int hdrlen;
1fb23e65 2673 int ret;
bbac7311 2674
c529d75a 2675 if (has_loose_object(sha1))
bbac7311
NP
2676 return 0;
2677 buf = read_packed_sha1(sha1, &type, &len);
2678 if (!buf)
2679 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2680 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
1fb23e65
BS
2681 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2682 free(buf);
2683
2684 return ret;
bbac7311
NP
2685}
2686
bf592c50
DB
2687int has_pack_index(const unsigned char *sha1)
2688{
2689 struct stat st;
2690 if (stat(sha1_pack_index_name(sha1), &st))
2691 return 0;
2692 return 1;
bf592c50
DB
2693}
2694
cd673c1f
JH
2695int has_sha1_pack(const unsigned char *sha1)
2696{
2697 struct pack_entry e;
2698 return find_pack_entry(sha1, &e);
2699}
2700
8237b185
DB
2701int has_sha1_file(const unsigned char *sha1)
2702{
1f688557
JH
2703 struct pack_entry e;
2704
cd673c1f 2705 if (find_pack_entry(sha1, &e))
1f688557 2706 return 1;
c529d75a 2707 return has_loose_object(sha1);
8237b185 2708}
74400e71 2709
c879daa2
NTND
2710static void check_tree(const void *buf, size_t size)
2711{
2712 struct tree_desc desc;
2713 struct name_entry entry;
2714
2715 init_tree_desc(&desc, buf, size);
2716 while (tree_entry(&desc, &entry))
2717 /* do nothing
2718 * tree_entry() will die() on malformed entries */
2719 ;
2720}
2721
2722static void check_commit(const void *buf, size_t size)
2723{
2724 struct commit c;
2725 memset(&c, 0, sizeof(c));
2726 if (parse_commit_buffer(&c, buf, size))
2727 die("corrupt commit");
2728}
2729
2730static void check_tag(const void *buf, size_t size)
2731{
2732 struct tag t;
2733 memset(&t, 0, sizeof(t));
2734 if (parse_tag_buffer(&t, buf, size))
2735 die("corrupt tag");
2736}
2737
43df4f86 2738static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
2739 enum object_type type,
2740 const char *path, unsigned flags)
e7332f96 2741{
6c510bee 2742 int ret, re_allocated = 0;
c4ce46fc 2743 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 2744
7672db20 2745 if (!type)
edaec3fb 2746 type = OBJ_BLOB;
6c510bee
LT
2747
2748 /*
2749 * Convert blobs to git internal format
2750 */
43df4f86 2751 if ((type == OBJ_BLOB) && path) {
f285a2d7 2752 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 2753 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 2754 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 2755 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
2756 re_allocated = 1;
2757 }
2758 }
c4ce46fc 2759 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
2760 if (type == OBJ_TREE)
2761 check_tree(buf, size);
2762 if (type == OBJ_COMMIT)
2763 check_commit(buf, size);
2764 if (type == OBJ_TAG)
2765 check_tag(buf, size);
2766 }
6c510bee 2767
7672db20 2768 if (write_object)
edaec3fb 2769 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 2770 else
edaec3fb 2771 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 2772 if (re_allocated)
6c510bee 2773 free(buf);
43df4f86
DP
2774 return ret;
2775}
2776
7b41e1e1
JH
2777static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2778 const char *path, unsigned flags)
2779{
2780 struct strbuf sbuf = STRBUF_INIT;
2781 int ret;
2782
2783 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2784 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2785 else
2786 ret = -1;
2787 strbuf_release(&sbuf);
2788 return ret;
2789}
2790
ea68b0ce
DP
2791#define SMALL_FILE_SIZE (32*1024)
2792
7b41e1e1
JH
2793static int index_core(unsigned char *sha1, int fd, size_t size,
2794 enum object_type type, const char *path,
2795 unsigned flags)
43df4f86
DP
2796{
2797 int ret;
43df4f86 2798
7b41e1e1 2799 if (!size) {
c4ce46fc 2800 ret = index_mem(sha1, NULL, size, type, path, flags);
ea68b0ce
DP
2801 } else if (size <= SMALL_FILE_SIZE) {
2802 char *buf = xmalloc(size);
2803 if (size == read_in_full(fd, buf, size))
c4ce46fc 2804 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
2805 else
2806 ret = error("short read %s", strerror(errno));
2807 free(buf);
08bda208 2808 } else {
43df4f86 2809 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 2810 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 2811 munmap(buf, size);
08bda208 2812 }
7b41e1e1
JH
2813 return ret;
2814}
2815
4dd1fbc7 2816/*
568508e7
JH
2817 * This creates one packfile per large blob unless bulk-checkin
2818 * machinery is "plugged".
4dd1fbc7
JH
2819 *
2820 * This also bypasses the usual "convert-to-git" dance, and that is on
2821 * purpose. We could write a streaming version of the converting
2822 * functions and insert that before feeding the data to fast-import
4f22b101
JK
2823 * (or equivalent in-core API described above). However, that is
2824 * somewhat complicated, as we do not know the size of the filter
2825 * result, which we need to know beforehand when writing a git object.
2826 * Since the primary motivation for trying to stream from the working
2827 * tree file and to avoid mmaping it in core is to deal with large
2828 * binary blobs, they generally do not want to get any conversion, and
2829 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
2830 */
2831static int index_stream(unsigned char *sha1, int fd, size_t size,
2832 enum object_type type, const char *path,
2833 unsigned flags)
2834{
568508e7 2835 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
2836}
2837
7b41e1e1
JH
2838int index_fd(unsigned char *sha1, int fd, struct stat *st,
2839 enum object_type type, const char *path, unsigned flags)
2840{
2841 int ret;
2842 size_t size = xsize_t(st->st_size);
2843
2844 if (!S_ISREG(st->st_mode))
2845 ret = index_pipe(sha1, fd, type, path, flags);
4f22b101
JK
2846 else if (size <= big_file_threshold || type != OBJ_BLOB ||
2847 (path && would_convert_to_git(path, NULL, 0, 0)))
7b41e1e1 2848 ret = index_core(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
2849 else
2850 ret = index_stream(sha1, fd, size, type, path, flags);
43df4f86 2851 close(fd);
aac17941 2852 return ret;
74400e71 2853}
ec1fcc16 2854
c4ce46fc 2855int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
2856{
2857 int fd;
b760d3aa 2858 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
2859
2860 switch (st->st_mode & S_IFMT) {
2861 case S_IFREG:
2862 fd = open(path, O_RDONLY);
2863 if (fd < 0)
2864 return error("open(\"%s\"): %s", path,
2865 strerror(errno));
c4ce46fc 2866 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
2867 return error("%s: failed to insert into database",
2868 path);
2869 break;
2870 case S_IFLNK:
b760d3aa 2871 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 2872 char *errstr = strerror(errno);
ec1fcc16
JH
2873 return error("readlink(\"%s\"): %s", path,
2874 errstr);
2875 }
c4ce46fc 2876 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
2877 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2878 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
2879 return error("%s: failed to insert into database",
2880 path);
b760d3aa 2881 strbuf_release(&sb);
ec1fcc16 2882 break;
f35a6d3b
LT
2883 case S_IFDIR:
2884 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
2885 default:
2886 return error("%s: unsupported file type", path);
2887 }
2888 return 0;
2889}
a69e5429
JH
2890
2891int read_pack_header(int fd, struct pack_header *header)
2892{
c697ad14
HO
2893 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2894 /* "eof before pack header was fully read" */
2895 return PH_ERROR_EOF;
2896
a69e5429
JH
2897 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2898 /* "protocol error (pack signature mismatch detected)" */
2899 return PH_ERROR_PACK_SIGNATURE;
2900 if (!pack_version_ok(header->hdr_version))
2901 /* "protocol error (pack version unsupported)" */
2902 return PH_ERROR_PROTOCOL;
2903 return 0;
2904}
40d52ff7
JK
2905
2906void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2907{
2908 enum object_type type = sha1_object_info(sha1, NULL);
2909 if (type < 0)
2910 die("%s is not a valid object", sha1_to_hex(sha1));
2911 if (type != expect)
2912 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2913 typename(expect));
2914}