blob: 613839db544dc852ee49d026bb1b561e6e1ce4bf [file] [log] [blame]
Linus Torvalds0fcfd162005-04-18 13:04:43 -07001/*
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 */
Linus Torvalds0fcfd162005-04-18 13:04:43 -07009#include "cache.h"
Michael Haggerty6eac50d2012-11-05 09:41:22 +010010#include "string-list.h"
Junio C Hamano1f688552005-06-27 03:35:33 -070011#include "delta.h"
Linus Torvaldsa733cb62005-06-28 14:21:02 -070012#include "pack.h"
Peter Eriksen8e440252006-04-02 14:44:09 +020013#include "blob.h"
14#include "commit.h"
Junio C Hamano4dd1fbc2011-05-08 01:47:35 -070015#include "run-command.h"
Peter Eriksen8e440252006-04-02 14:44:09 +020016#include "tag.h"
17#include "tree.h"
Nguyễn Thái Ngọc Duyc879daa2011-02-05 17:52:21 +070018#include "tree-walk.h"
Linus Torvaldsf35a6d32007-04-09 21:20:29 -070019#include "refs.h"
Nicolas Pitre70f5d5d2008-02-28 00:25:19 -050020#include "pack-revindex.h"
Junio C Hamano628522e2007-12-29 02:05:47 -080021#include "sha1-lookup.h"
Junio C Hamano568508e2011-10-28 14:48:40 -070022#include "bulk-checkin.h"
Nguyễn Thái Ngọc Duy090ea122012-03-07 17:54:18 +070023#include "streaming.h"
Nguyễn Thái Ngọc Duy543c5ca2013-02-15 19:07:10 +070024#include "dir.h"
Linus Torvalds0fcfd162005-04-18 13:04:43 -070025
Linus Torvalds144bde72005-04-23 11:09:32 -070026#ifndef O_NOATIME
27#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
28#define O_NOATIME 01000000
29#else
30#define O_NOATIME 0
31#endif
32#endif
33
Jonathan Nieder28bd70d2011-03-16 00:15:31 -050034#define SZ_FMT PRIuMAX
35static inline uintmax_t sz_fmt(size_t s) { return s; }
Pavel Roskine05db0f2007-01-09 23:07:11 -050036
David Rientjes96f1e582006-08-15 10:23:48 -070037const unsigned char null_sha1[20];
Junio C Hamano88cd6212005-09-30 14:02:47 -070038
Nguyễn Thái Ngọc Duyb12ca962013-06-09 12:22:48 +070039static const char *no_log_pack_access = "no_log_pack_access";
40static const char *log_pack_access;
41
Nguyễn Thái Ngọc Duyc597ba82011-02-05 21:03:01 +070042/*
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
46 * application).
47 */
48static struct cached_object {
49 unsigned char sha1[20];
50 enum object_type type;
51 void *buf;
52 unsigned long size;
53} *cached_objects;
54static int cached_object_nr, cached_object_alloc;
55
56static struct cached_object empty_tree = {
Jonathan Niederdab0d412011-02-07 02:17:27 -060057 EMPTY_TREE_SHA1_BIN_LITERAL,
Nguyễn Thái Ngọc Duyc597ba82011-02-05 21:03:01 +070058 OBJ_TREE,
59 "",
60 0
61};
62
Nguyễn Thái Ngọc Duyc01f51c2012-02-01 20:48:55 +070063static struct packed_git *last_found_pack;
64
Nguyễn Thái Ngọc Duyc597ba82011-02-05 21:03:01 +070065static struct cached_object *find_cached_object(const unsigned char *sha1)
66{
67 int i;
68 struct cached_object *co = cached_objects;
69
70 for (i = 0; i < cached_object_nr; i++, co++) {
71 if (!hashcmp(co->sha1, sha1))
72 return co;
73 }
74 if (!hashcmp(sha1, empty_tree.sha1))
75 return &empty_tree;
76 return NULL;
77}
78
Junio C Hamano90a64642011-03-10 16:02:50 -080079int mkdir_in_gitdir(const char *path)
80{
81 if (mkdir(path, 0777)) {
82 int saved_errno = errno;
83 struct stat st;
84 struct strbuf sb = STRBUF_INIT;
85
86 if (errno != EEXIST)
87 return -1;
88 /*
89 * Are we looking at a path in a symlinked worktree
90 * whose original repository does not yet have it?
91 * e.g. .git/rr-cache pointing at its original
92 * repository in which the user hasn't performed any
93 * conflict resolution yet?
94 */
95 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
96 strbuf_readlink(&sb, path, st.st_size) ||
97 !is_absolute_path(sb.buf) ||
98 mkdir(sb.buf, 0777)) {
99 strbuf_release(&sb);
100 errno = saved_errno;
101 return -1;
102 }
103 strbuf_release(&sb);
104 }
105 return adjust_shared_perm(path);
106}
107
Junio C Hamanob2cb9422005-07-06 01:11:52 -0700108int safe_create_leading_directories(char *path)
109{
Johannes Sixt8385abf2007-11-30 21:36:00 +0100110 char *pos = path + offset_1st_component(path);
Jason Riedy67d42212006-02-09 17:56:13 -0800111 struct stat st;
112
Junio C Hamanob2cb9422005-07-06 01:11:52 -0700113 while (pos) {
114 pos = strchr(pos, '/');
115 if (!pos)
116 break;
Junio C Hamano5f0bdf52008-09-02 14:10:15 -0700117 while (*++pos == '/')
118 ;
119 if (!*pos)
120 break;
121 *--pos = '\0';
Jason Riedy67d42212006-02-09 17:56:13 -0800122 if (!stat(path, &st)) {
123 /* path exists */
124 if (!S_ISDIR(st.st_mode)) {
Junio C Hamanob2cb9422005-07-06 01:11:52 -0700125 *pos = '/';
Jason Riedy67d42212006-02-09 17:56:13 -0800126 return -3;
Junio C Hamanob2cb9422005-07-06 01:11:52 -0700127 }
Johannes Schindelin457f06d2005-12-22 23:13:56 +0100128 }
Jason Riedy67d42212006-02-09 17:56:13 -0800129 else if (mkdir(path, 0777)) {
Steven Walter928734d2013-03-17 10:09:27 -0400130 if (errno == EEXIST &&
131 !stat(path, &st) && S_ISDIR(st.st_mode)) {
132 ; /* somebody created it since we checked */
133 } else {
134 *pos = '/';
135 return -1;
136 }
Jason Riedy67d42212006-02-09 17:56:13 -0800137 }
Johannes Schindelin457f06d2005-12-22 23:13:56 +0100138 else if (adjust_shared_perm(path)) {
139 *pos = '/';
140 return -2;
141 }
Junio C Hamanob2cb9422005-07-06 01:11:52 -0700142 *pos++ = '/';
143 }
144 return 0;
145}
146
Jeff King8e21d632008-06-25 01:41:34 -0400147int safe_create_leading_directories_const(const char *path)
148{
149 /* path points to cache entries, so xstrdup before messing with it */
150 char *buf = xstrdup(path);
151 int result = safe_create_leading_directories(buf);
152 free(buf);
153 return result;
154}
155
Junio C Hamanoace15342005-05-07 00:38:04 -0700156static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
157{
158 int i;
159 for (i = 0; i < 20; i++) {
160 static char hex[] = "0123456789abcdef";
161 unsigned int val = sha1[i];
162 char *pos = pathbuf + i*2 + (i > 0);
163 *pos++ = hex[val >> 4];
164 *pos = hex[val & 0xf];
165 }
166}
167
Linus Torvalds0fcfd162005-04-18 13:04:43 -0700168/*
169 * NOTE! This returns a statically allocated buffer, so you have to be
Jim Meyering790296f2008-01-03 15:18:07 +0100170 * careful about using it. Do an "xstrdup()" if you need to save the
Linus Torvalds0fcfd162005-04-18 13:04:43 -0700171 * filename.
Junio C Hamanoace15342005-05-07 00:38:04 -0700172 *
173 * Also note that this returns the location for creating. Reading
174 * SHA1 file can happen from any alternate directory listed in the
Junio C Hamanod19938a2005-05-09 17:57:56 -0700175 * DB_ENVIRONMENT environment variable if it is not found in
Junio C Hamanoace15342005-05-07 00:38:04 -0700176 * the primary object database.
Linus Torvalds0fcfd162005-04-18 13:04:43 -0700177 */
178char *sha1_file_name(const unsigned char *sha1)
179{
Jeff King560fb6a2010-05-22 02:59:42 -0400180 static char buf[PATH_MAX];
181 const char *objdir;
182 int len;
Linus Torvalds0fcfd162005-04-18 13:04:43 -0700183
Jeff King560fb6a2010-05-22 02:59:42 -0400184 objdir = get_object_directory();
185 len = strlen(objdir);
186
187 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
188 if (len + 43 > PATH_MAX)
189 die("insanely long object directory %s", objdir);
190 memcpy(buf, objdir, len);
191 buf[len] = '/';
192 buf[len+3] = '/';
193 buf[len+42] = '\0';
194 fill_sha1_path(buf + len + 1, sha1);
195 return buf;
Linus Torvalds0fcfd162005-04-18 13:04:43 -0700196}
197
Heikki Orsila633f43e2008-05-24 01:43:55 +0300198static char *sha1_get_pack_name(const unsigned char *sha1,
Junio C Hamano6eec46b2008-05-28 09:47:43 -0700199 char **name, char **base, const char *which)
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400200{
201 static const char hex[] = "0123456789abcdef";
Heikki Orsila633f43e2008-05-24 01:43:55 +0300202 char *buf;
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400203 int i;
204
Heikki Orsila633f43e2008-05-24 01:43:55 +0300205 if (!*base) {
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400206 const char *sha1_file_directory = get_object_directory();
207 int len = strlen(sha1_file_directory);
Heikki Orsila633f43e2008-05-24 01:43:55 +0300208 *base = xmalloc(len + 60);
Junio C Hamano6eec46b2008-05-28 09:47:43 -0700209 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
210 sha1_file_directory, which);
Heikki Orsila633f43e2008-05-24 01:43:55 +0300211 *name = *base + len + 11;
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400212 }
213
Heikki Orsila633f43e2008-05-24 01:43:55 +0300214 buf = *name;
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400215
216 for (i = 0; i < 20; i++) {
217 unsigned int val = *sha1++;
218 *buf++ = hex[val >> 4];
219 *buf++ = hex[val & 0xf];
220 }
Junio C Hamanoa6080a02007-06-07 00:04:01 -0700221
Heikki Orsila633f43e2008-05-24 01:43:55 +0300222 return *base;
223}
224
225char *sha1_pack_name(const unsigned char *sha1)
226{
227 static char *name, *base;
228
Junio C Hamano6eec46b2008-05-28 09:47:43 -0700229 return sha1_get_pack_name(sha1, &name, &base, "pack");
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400230}
231
232char *sha1_pack_index_name(const unsigned char *sha1)
233{
Heikki Orsila633f43e2008-05-24 01:43:55 +0300234 static char *name, *base;
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400235
Junio C Hamano6eec46b2008-05-28 09:47:43 -0700236 return sha1_get_pack_name(sha1, &name, &base, "idx");
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400237}
238
Junio C Hamanod5a63b92005-08-14 17:25:57 -0700239struct alternate_object_database *alt_odb_list;
240static struct alternate_object_database **alt_odb_tail;
Junio C Hamanoace15342005-05-07 00:38:04 -0700241
Junio C Hamanof4e51682011-05-15 12:16:29 -0700242static int git_open_noatime(const char *name);
Martin Waitzc2f493a2006-05-07 20:19:21 +0200243
Junio C Hamanoddd5d052005-05-08 13:51:13 -0700244/*
245 * Prepare alternate object database registry.
Junio C Hamanod5a63b92005-08-14 17:25:57 -0700246 *
247 * The variable alt_odb_list points at the list of struct
248 * alternate_object_database. The elements on this list come from
249 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
250 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
Junio C Hamano1494e032005-12-04 22:48:43 -0800251 * whose contents is similar to that environment variable but can be
252 * LF separated. Its base points at a statically allocated buffer that
Junio C Hamanod5a63b92005-08-14 17:25:57 -0700253 * contains "/the/directory/corresponding/to/.git/objects/...", while
254 * its name points just after the slash at the end of ".git/objects/"
255 * in the example above, and has enough space to hold 40-byte hex
256 * SHA1, an extra slash for the first level indirection, and the
257 * terminating NUL.
Junio C Hamanoddd5d052005-05-08 13:51:13 -0700258 */
Michael Haggerty6eac50d2012-11-05 09:41:22 +0100259static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
Martin Waitzc2f493a2006-05-07 20:19:21 +0200260{
Martin Waitzc2f493a2006-05-07 20:19:21 +0200261 const char *objdir = get_object_directory();
262 struct alternate_object_database *ent;
263 struct alternate_object_database *alt;
Hui Wang5bdf0a82011-09-07 18:37:47 +0800264 int pfxlen, entlen;
265 struct strbuf pathbuf = STRBUF_INIT;
Martin Waitzc2f493a2006-05-07 20:19:21 +0200266
Johannes Sixt85dadc32007-11-13 21:05:00 +0100267 if (!is_absolute_path(entry) && relative_base) {
Hui Wang5bdf0a82011-09-07 18:37:47 +0800268 strbuf_addstr(&pathbuf, real_path(relative_base));
269 strbuf_addch(&pathbuf, '/');
Martin Waitzc2f493a2006-05-07 20:19:21 +0200270 }
Michael Haggerty6eac50d2012-11-05 09:41:22 +0100271 strbuf_addstr(&pathbuf, entry);
Hui Wang5bdf0a82011-09-07 18:37:47 +0800272
273 normalize_path_copy(pathbuf.buf, pathbuf.buf);
274
275 pfxlen = strlen(pathbuf.buf);
276
277 /*
278 * The trailing slash after the directory name is given by
279 * this function at the end. Remove duplicates.
280 */
281 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
282 pfxlen -= 1;
283
284 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
Martin Waitzc2f493a2006-05-07 20:19:21 +0200285 ent = xmalloc(sizeof(*ent) + entlen);
Hui Wang5bdf0a82011-09-07 18:37:47 +0800286 memcpy(ent->base, pathbuf.buf, pfxlen);
287 strbuf_release(&pathbuf);
Martin Waitzc2f493a2006-05-07 20:19:21 +0200288
289 ent->name = ent->base + pfxlen + 1;
290 ent->base[pfxlen + 3] = '/';
291 ent->base[pfxlen] = ent->base[entlen-1] = 0;
292
293 /* Detect cases where alternate disappeared */
Junio C Hamano90b4a712008-09-09 01:27:07 -0700294 if (!is_directory(ent->base)) {
Martin Waitzc2f493a2006-05-07 20:19:21 +0200295 error("object directory %s does not exist; "
296 "check .git/objects/info/alternates.",
297 ent->base);
298 free(ent);
299 return -1;
300 }
301
302 /* Prevent the common mistake of listing the same
303 * thing twice, or object directory itself.
304 */
305 for (alt = alt_odb_list; alt; alt = alt->next) {
306 if (!memcmp(ent->base, alt->base, pfxlen)) {
307 free(ent);
308 return -1;
309 }
310 }
Heiko Voigtcb2912c2012-07-28 17:46:36 +0200311 if (!strcmp(ent->base, objdir)) {
Martin Waitzc2f493a2006-05-07 20:19:21 +0200312 free(ent);
313 return -1;
314 }
315
316 /* add the alternate entry */
317 *alt_odb_tail = ent;
318 alt_odb_tail = &(ent->next);
319 ent->next = NULL;
320
321 /* recursively add alternates */
322 read_info_alternates(ent->base, depth + 1);
323
324 ent->base[pfxlen] = '/';
325
326 return 0;
327}
328
Michael Haggertyc5950162012-11-05 09:41:23 +0100329static void link_alt_odb_entries(const char *alt, int len, int sep,
Martin Waitzc2f493a2006-05-07 20:19:21 +0200330 const char *relative_base, int depth)
Junio C Hamanod5a63b92005-08-14 17:25:57 -0700331{
Michael Haggerty6eac50d2012-11-05 09:41:22 +0100332 struct string_list entries = STRING_LIST_INIT_NODUP;
333 char *alt_copy;
334 int i;
Martin Waitzc2f493a2006-05-07 20:19:21 +0200335
336 if (depth > 5) {
337 error("%s: ignoring alternate object stores, nesting too deep.",
338 relative_base);
339 return;
340 }
Junio C Hamanod5a63b92005-08-14 17:25:57 -0700341
Michael Haggertyc5950162012-11-05 09:41:23 +0100342 alt_copy = xmemdupz(alt, len);
Michael Haggerty6eac50d2012-11-05 09:41:22 +0100343 string_list_split_in_place(&entries, alt_copy, sep, -1);
344 for (i = 0; i < entries.nr; i++) {
345 const char *entry = entries.items[i].string;
346 if (entry[0] == '\0' || entry[0] == '#')
Junio C Hamano9577e7e2005-08-16 18:22:05 -0700347 continue;
Michael Haggerty6eac50d2012-11-05 09:41:22 +0100348 if (!is_absolute_path(entry) && depth) {
349 error("%s: ignoring relative alternate object store %s",
350 relative_base, entry);
351 } else {
352 link_alt_odb_entry(entry, relative_base, depth);
Junio C Hamano9577e7e2005-08-16 18:22:05 -0700353 }
Junio C Hamano9577e7e2005-08-16 18:22:05 -0700354 }
Michael Haggerty6eac50d2012-11-05 09:41:22 +0100355 string_list_clear(&entries, 0);
356 free(alt_copy);
Junio C Hamanod5a63b92005-08-14 17:25:57 -0700357}
358
Heiko Voigt5e736332012-05-14 18:24:45 +0200359void read_info_alternates(const char * relative_base, int depth)
Junio C Hamanoace15342005-05-07 00:38:04 -0700360{
Junio C Hamano9577e7e2005-08-16 18:22:05 -0700361 char *map;
Shawn O. Pearcedc49cd72007-03-06 20:44:37 -0500362 size_t mapsz;
Junio C Hamanod5a63b92005-08-14 17:25:57 -0700363 struct stat st;
Jim Meyering9cb18f52007-07-03 12:40:20 +0200364 const char alt_file_name[] = "info/alternates";
365 /* Given that relative_base is no longer than PATH_MAX,
366 ensure that "path" has enough space to append "/", the
367 file name, "info/alternates", and a trailing NUL. */
368 char path[PATH_MAX + 1 + sizeof alt_file_name];
Martin Waitzc2f493a2006-05-07 20:19:21 +0200369 int fd;
Jason Riedyc7c81b32005-08-23 13:34:07 -0700370
Jim Meyering9cb18f52007-07-03 12:40:20 +0200371 sprintf(path, "%s/%s", relative_base, alt_file_name);
Junio C Hamanof4e51682011-05-15 12:16:29 -0700372 fd = git_open_noatime(path);
Junio C Hamanod5a63b92005-08-14 17:25:57 -0700373 if (fd < 0)
374 return;
375 if (fstat(fd, &st) || (st.st_size == 0)) {
376 close(fd);
377 return;
Junio C Hamanoace15342005-05-07 00:38:04 -0700378 }
Shawn O. Pearcedc49cd72007-03-06 20:44:37 -0500379 mapsz = xsize_t(st.st_size);
380 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
Junio C Hamanod5a63b92005-08-14 17:25:57 -0700381 close(fd);
Junio C Hamanod5a63b92005-08-14 17:25:57 -0700382
Michael Haggertyc5950162012-11-05 09:41:23 +0100383 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
Martin Waitzc2f493a2006-05-07 20:19:21 +0200384
Shawn O. Pearcedc49cd72007-03-06 20:44:37 -0500385 munmap(map, mapsz);
Junio C Hamanoace15342005-05-07 00:38:04 -0700386}
387
Daniel Barkalowbef70b22008-04-17 19:32:30 -0400388void add_to_alternates_file(const char *reference)
389{
390 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
Junio C Hamanoacd3b9e2008-10-17 15:44:39 -0700391 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
Junio C Hamanoe6baf4a2011-08-22 18:05:16 -0700392 char *alt = mkpath("%s\n", reference);
Daniel Barkalowbef70b22008-04-17 19:32:30 -0400393 write_or_die(fd, alt, strlen(alt));
394 if (commit_lock_file(lock))
395 die("could not close alternates file");
396 if (alt_odb_tail)
Michael Haggertyc5950162012-11-05 09:41:23 +0100397 link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
Daniel Barkalowbef70b22008-04-17 19:32:30 -0400398}
399
Junio C Hamanod79796b2008-09-09 01:27:10 -0700400void foreach_alt_odb(alt_odb_fn fn, void *cb)
401{
402 struct alternate_object_database *ent;
403
404 prepare_alt_odb();
405 for (ent = alt_odb_list; ent; ent = ent->next)
406 if (fn(ent, cb))
407 return;
408}
409
Martin Waitzc2f493a2006-05-07 20:19:21 +0200410void prepare_alt_odb(void)
411{
Timo Hirvonen554fe202006-06-28 12:04:39 +0300412 const char *alt;
Martin Waitzc2f493a2006-05-07 20:19:21 +0200413
Shawn O. Pearce7dc24aa2007-05-26 01:24:40 -0400414 if (alt_odb_tail)
415 return;
416
Martin Waitzc2f493a2006-05-07 20:19:21 +0200417 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
418 if (!alt) alt = "";
419
Martin Waitzc2f493a2006-05-07 20:19:21 +0200420 alt_odb_tail = &alt_odb_list;
Michael Haggertyc5950162012-11-05 09:41:23 +0100421 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
Martin Waitzc2f493a2006-05-07 20:19:21 +0200422
423 read_info_alternates(get_object_directory(), 0);
424}
425
Brandon Casey0f4dc142008-11-09 23:59:57 -0600426static int has_loose_object_local(const unsigned char *sha1)
Junio C Hamanoace15342005-05-07 00:38:04 -0700427{
Junio C Hamanoace15342005-05-07 00:38:04 -0700428 char *name = sha1_file_name(sha1);
Brandon Casey0f4dc142008-11-09 23:59:57 -0600429 return !access(name, F_OK);
430}
Junio C Hamanoace15342005-05-07 00:38:04 -0700431
Brandon Casey0f4dc142008-11-09 23:59:57 -0600432int has_loose_object_nonlocal(const unsigned char *sha1)
433{
434 struct alternate_object_database *alt;
Junio C Hamano9a217f22005-06-28 14:56:57 -0700435 prepare_alt_odb();
Junio C Hamanod5a63b92005-08-14 17:25:57 -0700436 for (alt = alt_odb_list; alt; alt = alt->next) {
Brandon Casey0f4dc142008-11-09 23:59:57 -0600437 fill_sha1_path(alt->name, sha1);
Linus Torvaldsc529d752008-06-14 11:43:01 -0700438 if (!access(alt->base, F_OK))
439 return 1;
Junio C Hamanoace15342005-05-07 00:38:04 -0700440 }
Linus Torvaldsc529d752008-06-14 11:43:01 -0700441 return 0;
Junio C Hamanoace15342005-05-07 00:38:04 -0700442}
443
Brandon Casey0f4dc142008-11-09 23:59:57 -0600444static int has_loose_object(const unsigned char *sha1)
445{
446 return has_loose_object_local(sha1) ||
447 has_loose_object_nonlocal(sha1);
448}
449
Shawn O. Pearce60bb8b12006-12-23 02:34:28 -0500450static unsigned int pack_used_ctr;
Shawn O. Pearcea53128b2006-12-23 02:34:47 -0500451static unsigned int pack_mmap_calls;
452static unsigned int peak_pack_open_windows;
453static unsigned int pack_open_windows;
Shawn O. Pearcec7934302011-02-28 12:52:39 -0800454static unsigned int pack_open_fds;
455static unsigned int pack_max_fds;
Shawn O. Pearcea53128b2006-12-23 02:34:47 -0500456static size_t peak_pack_mapped;
Shawn O. Pearce60bb8b12006-12-23 02:34:28 -0500457static size_t pack_mapped;
Junio C Hamano9a217f22005-06-28 14:56:57 -0700458struct packed_git *packed_git;
Junio C Hamano1f688552005-06-27 03:35:33 -0700459
Junio C Hamanob79d18c2007-06-13 01:22:51 -0700460void pack_report(void)
Shawn O. Pearcea53128b2006-12-23 02:34:47 -0500461{
462 fprintf(stderr,
Pavel Roskine05db0f2007-01-09 23:07:11 -0500463 "pack_report: getpagesize() = %10" SZ_FMT "\n"
464 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
465 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
Steffen Prohaska9e42d6a2007-11-21 21:27:19 +0100466 sz_fmt(getpagesize()),
467 sz_fmt(packed_git_window_size),
468 sz_fmt(packed_git_limit));
Shawn O. Pearcea53128b2006-12-23 02:34:47 -0500469 fprintf(stderr,
470 "pack_report: pack_used_ctr = %10u\n"
471 "pack_report: pack_mmap_calls = %10u\n"
472 "pack_report: pack_open_windows = %10u / %10u\n"
Pavel Roskine05db0f2007-01-09 23:07:11 -0500473 "pack_report: pack_mapped = "
474 "%10" SZ_FMT " / %10" SZ_FMT "\n",
Shawn O. Pearcea53128b2006-12-23 02:34:47 -0500475 pack_used_ctr,
476 pack_mmap_calls,
477 pack_open_windows, peak_pack_open_windows,
Steffen Prohaska9e42d6a2007-11-21 21:27:19 +0100478 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
Shawn O. Pearcea53128b2006-12-23 02:34:47 -0500479}
480
Nicolas Pitre42873072007-03-16 16:42:50 -0400481static int check_packed_git_idx(const char *path, struct packed_git *p)
Junio C Hamano1f688552005-06-27 03:35:33 -0700482{
483 void *idx_map;
Nicolas Pitre42873072007-03-16 16:42:50 -0400484 struct pack_idx_header *hdr;
Shawn O. Pearcedc49cd72007-03-06 20:44:37 -0500485 size_t idx_size;
Nicolas Pitre74e34e12007-04-09 01:06:35 -0400486 uint32_t version, nr, i, *index;
Junio C Hamanof4e51682011-05-15 12:16:29 -0700487 int fd = git_open_noatime(path);
Junio C Hamano1f688552005-06-27 03:35:33 -0700488 struct stat st;
Nicolas Pitre42873072007-03-16 16:42:50 -0400489
Junio C Hamano1f688552005-06-27 03:35:33 -0700490 if (fd < 0)
491 return -1;
492 if (fstat(fd, &st)) {
493 close(fd);
494 return -1;
495 }
Shawn O. Pearcedc49cd72007-03-06 20:44:37 -0500496 idx_size = xsize_t(st.st_size);
Shawn O. Pearce2d884512007-03-06 20:44:11 -0500497 if (idx_size < 4 * 256 + 20 + 20) {
498 close(fd);
499 return error("index file %s is too small", path);
500 }
Shawn O. Pearcec4712e42006-12-24 00:47:23 -0500501 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
Junio C Hamano1f688552005-06-27 03:35:33 -0700502 close(fd);
Junio C Hamano1f688552005-06-27 03:35:33 -0700503
Nicolas Pitre42873072007-03-16 16:42:50 -0400504 hdr = idx_map;
505 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
Nicolas Pitre74e34e12007-04-09 01:06:35 -0400506 version = ntohl(hdr->idx_version);
507 if (version < 2 || version > 2) {
508 munmap(idx_map, idx_size);
Ramsay Jones6e1c2342008-07-03 16:52:09 +0100509 return error("index file %s is version %"PRIu32
Nicolas Pitre74e34e12007-04-09 01:06:35 -0400510 " and is not supported by this binary"
511 " (try upgrading GIT to a newer version)",
512 path, version);
513 }
514 } else
515 version = 1;
Shawn O. Pearcedf1b0592007-01-17 20:43:57 -0500516
Junio C Hamano1f688552005-06-27 03:35:33 -0700517 nr = 0;
Nicolas Pitre42873072007-03-16 16:42:50 -0400518 index = idx_map;
Nicolas Pitre74e34e12007-04-09 01:06:35 -0400519 if (version > 1)
520 index += 2; /* skip index header */
Junio C Hamano1f688552005-06-27 03:35:33 -0700521 for (i = 0; i < 256; i++) {
Shawn O. Pearce326bf392007-03-06 20:44:19 -0500522 uint32_t n = ntohl(index[i]);
Shawn O. Pearce2d884512007-03-06 20:44:11 -0500523 if (n < nr) {
524 munmap(idx_map, idx_size);
Shawn O. Pearcedf1b0592007-01-17 20:43:57 -0500525 return error("non-monotonic index %s", path);
Shawn O. Pearce2d884512007-03-06 20:44:11 -0500526 }
Junio C Hamano1f688552005-06-27 03:35:33 -0700527 nr = n;
528 }
529
Nicolas Pitre74e34e12007-04-09 01:06:35 -0400530 if (version == 1) {
531 /*
532 * Total size:
533 * - 256 index entries 4 bytes each
534 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
535 * - 20-byte SHA1 of the packfile
536 * - 20-byte SHA1 file checksum
537 */
538 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
539 munmap(idx_map, idx_size);
Luiz Fernando N. Capitulinoeef427a2007-08-14 16:42:37 -0300540 return error("wrong index v1 file size in %s", path);
Nicolas Pitre74e34e12007-04-09 01:06:35 -0400541 }
542 } else if (version == 2) {
543 /*
544 * Minimum size:
545 * - 8 bytes of header
546 * - 256 index entries 4 bytes each
547 * - 20-byte sha1 entry * nr
548 * - 4-byte crc entry * nr
549 * - 4-byte offset entry * nr
550 * - 20-byte SHA1 of the packfile
551 * - 20-byte SHA1 file checksum
552 * And after the 4-byte offset table might be a
553 * variable sized table containing 8-byte entries
554 * for offsets larger than 2^31.
555 */
556 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
Linus Torvalds1164f1e2007-06-26 14:34:02 -0700557 unsigned long max_size = min_size;
558 if (nr)
559 max_size += (nr - 1)*8;
560 if (idx_size < min_size || idx_size > max_size) {
Nicolas Pitre74e34e12007-04-09 01:06:35 -0400561 munmap(idx_map, idx_size);
Luiz Fernando N. Capitulinoeef427a2007-08-14 16:42:37 -0300562 return error("wrong index v2 file size in %s", path);
Nicolas Pitre74e34e12007-04-09 01:06:35 -0400563 }
Junio C Hamano7109c882007-10-29 11:53:55 -0700564 if (idx_size != min_size &&
565 /*
566 * make sure we can deal with large pack offsets.
567 * 31-bit signed offset won't be enough, neither
568 * 32-bit unsigned one will be.
569 */
570 (sizeof(off_t) <= 4)) {
571 munmap(idx_map, idx_size);
572 return error("pack too large for current definition of off_t in %s", path);
Nicolas Pitre74e34e12007-04-09 01:06:35 -0400573 }
Shawn O. Pearce2d884512007-03-06 20:44:11 -0500574 }
Junio C Hamano1f688552005-06-27 03:35:33 -0700575
Nicolas Pitre74e34e12007-04-09 01:06:35 -0400576 p->index_version = version;
Nicolas Pitre42873072007-03-16 16:42:50 -0400577 p->index_data = idx_map;
578 p->index_size = idx_size;
Nicolas Pitre57059092007-04-09 01:06:28 -0400579 p->num_objects = nr;
Junio C Hamano1f688552005-06-27 03:35:33 -0700580 return 0;
581}
582
Shawn O. Pearcebc8e4782007-05-30 02:13:42 -0400583int open_pack_index(struct packed_git *p)
Shawn O. Pearced0798372007-05-26 01:24:19 -0400584{
585 char *idx_name;
586 int ret;
587
588 if (p->index_data)
589 return 0;
590
591 idx_name = xstrdup(p->pack_name);
592 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
593 ret = check_packed_git_idx(idx_name, p);
594 free(idx_name);
595 return ret;
596}
597
Shawn O. Pearce11daf392006-12-23 02:34:44 -0500598static void scan_windows(struct packed_git *p,
599 struct packed_git **lru_p,
600 struct pack_window **lru_w,
601 struct pack_window **lru_l)
602{
603 struct pack_window *w, *w_l;
604
605 for (w_l = NULL, w = p->windows; w; w = w->next) {
606 if (!w->inuse_cnt) {
607 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
608 *lru_p = p;
609 *lru_w = w;
610 *lru_l = w_l;
611 }
612 }
613 w_l = w;
614 }
615}
616
Shawn O. Pearced1efefa2007-04-25 04:02:27 -0400617static int unuse_one_window(struct packed_git *current, int keep_fd)
Junio C Hamano1f688552005-06-27 03:35:33 -0700618{
Shawn O. Pearce54044bf2006-12-23 02:34:23 -0500619 struct packed_git *p, *lru_p = NULL;
Shawn O. Pearce11daf392006-12-23 02:34:44 -0500620 struct pack_window *lru_w = NULL, *lru_l = NULL;
Junio C Hamanof9253392005-06-29 02:51:27 -0700621
Shawn O. Pearce11daf392006-12-23 02:34:44 -0500622 if (current)
623 scan_windows(current, &lru_p, &lru_w, &lru_l);
624 for (p = packed_git; p; p = p->next)
625 scan_windows(p, &lru_p, &lru_w, &lru_l);
Shawn O. Pearce54044bf2006-12-23 02:34:23 -0500626 if (lru_p) {
627 munmap(lru_w->base, lru_w->len);
628 pack_mapped -= lru_w->len;
629 if (lru_l)
630 lru_l->next = lru_w->next;
631 else {
632 lru_p->windows = lru_w->next;
Shawn O. Pearced131b7a2011-03-02 10:01:54 -0800633 if (!lru_p->windows && lru_p->pack_fd != -1
634 && lru_p->pack_fd != keep_fd) {
Shawn O. Pearce54044bf2006-12-23 02:34:23 -0500635 close(lru_p->pack_fd);
Shawn O. Pearcec7934302011-02-28 12:52:39 -0800636 pack_open_fds--;
Shawn O. Pearce54044bf2006-12-23 02:34:23 -0500637 lru_p->pack_fd = -1;
638 }
639 }
640 free(lru_w);
Shawn O. Pearcea53128b2006-12-23 02:34:47 -0500641 pack_open_windows--;
Shawn O. Pearce54044bf2006-12-23 02:34:23 -0500642 return 1;
643 }
644 return 0;
Junio C Hamano1f688552005-06-27 03:35:33 -0700645}
646
Shawn O. Pearced1efefa2007-04-25 04:02:27 -0400647void release_pack_memory(size_t need, int fd)
Shawn O. Pearce97bfeb32006-12-24 00:47:19 -0500648{
649 size_t cur = pack_mapped;
Shawn O. Pearced1efefa2007-04-25 04:02:27 -0400650 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
Shawn O. Pearce97bfeb32006-12-24 00:47:19 -0500651 ; /* nothing */
652}
653
Jonathan Nieder58ecbd52010-11-06 06:44:11 -0500654void *xmmap(void *start, size_t length,
655 int prot, int flags, int fd, off_t offset)
656{
657 void *ret = mmap(start, length, prot, flags, fd, offset);
658 if (ret == MAP_FAILED) {
659 if (!length)
660 return NULL;
661 release_pack_memory(length, fd);
662 ret = mmap(start, length, prot, flags, fd, offset);
663 if (ret == MAP_FAILED)
664 die_errno("Out of memory? mmap failed");
665 }
666 return ret;
667}
668
Shawn O. Pearcec9ced052008-01-17 22:57:00 -0500669void close_pack_windows(struct packed_git *p)
670{
671 while (p->windows) {
672 struct pack_window *w = p->windows;
673
674 if (w->inuse_cnt)
675 die("pack '%s' still has open windows to it",
676 p->pack_name);
677 munmap(w->base, w->len);
678 pack_mapped -= w->len;
679 pack_open_windows--;
680 p->windows = w->next;
681 free(w);
682 }
683}
684
Shawn O. Pearce03e79c82006-12-23 02:34:08 -0500685void unuse_pack(struct pack_window **w_cursor)
Junio C Hamanof9253392005-06-29 02:51:27 -0700686{
Shawn O. Pearce03e79c82006-12-23 02:34:08 -0500687 struct pack_window *w = *w_cursor;
688 if (w) {
689 w->inuse_cnt--;
690 *w_cursor = NULL;
691 }
Junio C Hamanof9253392005-06-29 02:51:27 -0700692}
693
Shawn O. Pearcefa5fc152010-04-19 07:23:06 -0700694void close_pack_index(struct packed_git *p)
695{
696 if (p->index_data) {
697 munmap((void *)p->index_data, p->index_size);
698 p->index_data = NULL;
699 }
700}
701
Shawn O. Pearce3cf8b462007-02-02 03:00:03 -0500702/*
Nicolas Pitrec74faea2008-12-09 14:26:52 -0500703 * This is used by git-repack in case a newly created pack happens to
704 * contain the same set of objects as an existing one. In that case
705 * the resulting file might be different even if its name would be the
706 * same. It is best to close any reference to the old pack before it is
707 * replaced on disk. Of course no index pointers nor windows for given pack
708 * must subsist at this point. If ever objects from this pack are requested
709 * again, the new version of the pack will be reinitialized through
710 * reprepare_packed_git().
711 */
712void free_pack_by_name(const char *pack_name)
713{
714 struct packed_git *p, **pp = &packed_git;
715
716 while (*pp) {
717 p = *pp;
718 if (strcmp(pack_name, p->pack_name) == 0) {
Shawn O. Pearcefa3a0c92009-02-11 10:15:30 -0800719 clear_delta_base_cache();
Nicolas Pitrec74faea2008-12-09 14:26:52 -0500720 close_pack_windows(p);
Shawn O. Pearcec7934302011-02-28 12:52:39 -0800721 if (p->pack_fd != -1) {
Nicolas Pitrec74faea2008-12-09 14:26:52 -0500722 close(p->pack_fd);
Shawn O. Pearcec7934302011-02-28 12:52:39 -0800723 pack_open_fds--;
724 }
Shawn O. Pearcefa5fc152010-04-19 07:23:06 -0700725 close_pack_index(p);
Nicolas Pitrec74faea2008-12-09 14:26:52 -0500726 free(p->bad_object_sha1);
727 *pp = p->next;
Nguyễn Thái Ngọc Duyc01f51c2012-02-01 20:48:55 +0700728 if (last_found_pack == p)
729 last_found_pack = NULL;
Nicolas Pitrec74faea2008-12-09 14:26:52 -0500730 free(p);
731 return;
732 }
733 pp = &p->next;
734 }
735}
736
Joachim Schmitza0788262012-08-24 11:52:22 +0200737static unsigned int get_max_fd_limit(void)
738{
739#ifdef RLIMIT_NOFILE
740 struct rlimit lim;
741
742 if (getrlimit(RLIMIT_NOFILE, &lim))
743 die_errno("cannot get RLIMIT_NOFILE");
744
745 return lim.rlim_cur;
746#elif defined(_SC_OPEN_MAX)
747 return sysconf(_SC_OPEN_MAX);
748#elif defined(OPEN_MAX)
749 return OPEN_MAX;
750#else
751 return 1; /* see the caller ;-) */
752#endif
753}
754
Nicolas Pitrec74faea2008-12-09 14:26:52 -0500755/*
Shawn O. Pearce3cf8b462007-02-02 03:00:03 -0500756 * Do not call this directly as this leaks p->pack_fd on error return;
757 * call open_packed_git() instead.
758 */
759static int open_packed_git_1(struct packed_git *p)
Junio C Hamano1f688552005-06-27 03:35:33 -0700760{
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500761 struct stat st;
762 struct pack_header hdr;
763 unsigned char sha1[20];
764 unsigned char *idx_sha1;
Junio C Hamano2c039da2006-12-29 00:30:01 -0800765 long fd_flag;
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500766
Shawn O. Pearced0798372007-05-26 01:24:19 -0400767 if (!p->index_data && open_pack_index(p))
768 return error("packfile %s index unavailable", p->pack_name);
769
Shawn O. Pearcec7934302011-02-28 12:52:39 -0800770 if (!pack_max_fds) {
Joachim Schmitza0788262012-08-24 11:52:22 +0200771 unsigned int max_fds = get_max_fd_limit();
Shawn O. Pearcec7934302011-02-28 12:52:39 -0800772
773 /* Save 3 for stdin/stdout/stderr, 22 for work */
774 if (25 < max_fds)
775 pack_max_fds = max_fds - 25;
776 else
777 pack_max_fds = 1;
778 }
779
780 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
781 ; /* nothing */
782
Junio C Hamanof4e51682011-05-15 12:16:29 -0700783 p->pack_fd = git_open_noatime(p->pack_name);
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500784 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
Shawn O. Pearce072db272007-02-01 15:52:33 -0500785 return -1;
Shawn O. Pearcec7934302011-02-28 12:52:39 -0800786 pack_open_fds++;
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500787
788 /* If we created the struct before we had the pack we lack size. */
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400789 if (!p->pack_size) {
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400790 if (!S_ISREG(st.st_mode))
Shawn O. Pearce072db272007-02-01 15:52:33 -0500791 return error("packfile %s not a regular file", p->pack_name);
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400792 p->pack_size = st.st_size;
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500793 } else if (p->pack_size != st.st_size)
Shawn O. Pearce072db272007-02-01 15:52:33 -0500794 return error("packfile %s size changed", p->pack_name);
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500795
Junio C Hamano2c039da2006-12-29 00:30:01 -0800796 /* We leave these file descriptors open with sliding mmap;
797 * there is no point keeping them open across exec(), though.
798 */
799 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
800 if (fd_flag < 0)
Shawn O. Pearce072db272007-02-01 15:52:33 -0500801 return error("cannot determine file descriptor flags");
Junio C Hamano2c039da2006-12-29 00:30:01 -0800802 fd_flag |= FD_CLOEXEC;
803 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
Shawn O. Pearce072db272007-02-01 15:52:33 -0500804 return error("cannot set FD_CLOEXEC");
Junio C Hamano2c039da2006-12-29 00:30:01 -0800805
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500806 /* Verify we recognize this pack file format. */
Shawn O. Pearcee6e2bd62007-01-14 01:01:49 -0500807 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
Shawn O. Pearce072db272007-02-01 15:52:33 -0500808 return error("file %s is far too short to be a packfile", p->pack_name);
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500809 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
Shawn O. Pearce072db272007-02-01 15:52:33 -0500810 return error("file %s is not a GIT packfile", p->pack_name);
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500811 if (!pack_version_ok(hdr.hdr_version))
Ramsay Jones6e1c2342008-07-03 16:52:09 +0100812 return error("packfile %s is version %"PRIu32" and not"
813 " supported (try upgrading GIT to a newer version)",
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500814 p->pack_name, ntohl(hdr.hdr_version));
815
816 /* Verify the pack matches its index. */
Nicolas Pitre57059092007-04-09 01:06:28 -0400817 if (p->num_objects != ntohl(hdr.hdr_entries))
Ramsay Jones6e1c2342008-07-03 16:52:09 +0100818 return error("packfile %s claims to have %"PRIu32" objects"
819 " while index indicates %"PRIu32" objects",
Nicolas Pitre57059092007-04-09 01:06:28 -0400820 p->pack_name, ntohl(hdr.hdr_entries),
821 p->num_objects);
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500822 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
Shawn O. Pearce072db272007-02-01 15:52:33 -0500823 return error("end of packfile %s is unavailable", p->pack_name);
Shawn O. Pearcee6e2bd62007-01-14 01:01:49 -0500824 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
Shawn O. Pearce072db272007-02-01 15:52:33 -0500825 return error("packfile %s signature is unavailable", p->pack_name);
Nicolas Pitre42873072007-03-16 16:42:50 -0400826 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500827 if (hashcmp(sha1, idx_sha1))
Shawn O. Pearce072db272007-02-01 15:52:33 -0500828 return error("packfile %s does not match index", p->pack_name);
829 return 0;
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500830}
831
Shawn O. Pearce3cf8b462007-02-02 03:00:03 -0500832static int open_packed_git(struct packed_git *p)
833{
834 if (!open_packed_git_1(p))
835 return 0;
836 if (p->pack_fd != -1) {
837 close(p->pack_fd);
Shawn O. Pearcec7934302011-02-28 12:52:39 -0800838 pack_open_fds--;
Shawn O. Pearce3cf8b462007-02-02 03:00:03 -0500839 p->pack_fd = -1;
840 }
841 return -1;
842}
843
Shawn O. Pearcec4001d92007-03-06 20:44:30 -0500844static int in_window(struct pack_window *win, off_t offset)
Shawn O. Pearce60bb8b12006-12-23 02:34:28 -0500845{
846 /* We must promise at least 20 bytes (one hash) after the
847 * offset is available from this window, otherwise the offset
848 * is not actually in this window and a different window (which
849 * has that one hash excess) must be used. This is to support
850 * the object header and delta base parsing routines below.
851 */
852 off_t win_off = win->offset;
853 return win_off <= offset
854 && (offset + 20) <= (win_off + win->len);
855}
856
Felipe Contreras4b25d092009-05-01 12:06:36 +0300857unsigned char *use_pack(struct packed_git *p,
Shawn O. Pearce03e79c82006-12-23 02:34:08 -0500858 struct pack_window **w_cursor,
Shawn O. Pearcec4001d92007-03-06 20:44:30 -0500859 off_t offset,
Junio C Hamanoef49a7a2011-06-10 11:52:15 -0700860 unsigned long *left)
Shawn O. Pearce9bc879c2006-12-23 02:34:01 -0500861{
Shawn O. Pearce60bb8b12006-12-23 02:34:28 -0500862 struct pack_window *win = *w_cursor;
Shawn O. Pearce03e79c82006-12-23 02:34:08 -0500863
Felipe Contrerasa9d98a12009-02-24 23:59:05 +0200864 /* Since packfiles end in a hash of their content and it's
Shawn O. Pearce60bb8b12006-12-23 02:34:28 -0500865 * pointless to ask for an offset into the middle of that
866 * hash, and the in_window function above wouldn't match
867 * don't allow an offset too close to the end of the file.
868 */
Shawn O. Pearced131b7a2011-03-02 10:01:54 -0800869 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
870 die("packfile %s cannot be accessed", p->pack_name);
Shawn O. Pearce60bb8b12006-12-23 02:34:28 -0500871 if (offset > (p->pack_size - 20))
872 die("offset beyond end of packfile (truncated pack?)");
873
874 if (!win || !in_window(win, offset)) {
875 if (win)
876 win->inuse_cnt--;
877 for (win = p->windows; win; win = win->next) {
878 if (in_window(win, offset))
879 break;
880 }
881 if (!win) {
Alexandre Julliard78a28df2007-02-14 18:11:40 +0100882 size_t window_align = packed_git_window_size / 2;
Shawn O. Pearcedc49cd72007-03-06 20:44:37 -0500883 off_t len;
Shawn O. Pearced131b7a2011-03-02 10:01:54 -0800884
885 if (p->pack_fd == -1 && open_packed_git(p))
886 die("packfile %s cannot be accessed", p->pack_name);
887
Shawn O. Pearce60bb8b12006-12-23 02:34:28 -0500888 win = xcalloc(1, sizeof(*win));
Alexandre Julliard78a28df2007-02-14 18:11:40 +0100889 win->offset = (offset / window_align) * window_align;
Shawn O. Pearcedc49cd72007-03-06 20:44:37 -0500890 len = p->pack_size - win->offset;
891 if (len > packed_git_window_size)
892 len = packed_git_window_size;
893 win->len = (size_t)len;
Shawn O. Pearce60bb8b12006-12-23 02:34:28 -0500894 pack_mapped += win->len;
Shawn O. Pearce11daf392006-12-23 02:34:44 -0500895 while (packed_git_limit < pack_mapped
Shawn O. Pearced1efefa2007-04-25 04:02:27 -0400896 && unuse_one_window(p, p->pack_fd))
Shawn O. Pearce60bb8b12006-12-23 02:34:28 -0500897 ; /* nothing */
Shawn O. Pearcec4712e42006-12-24 00:47:23 -0500898 win->base = xmmap(NULL, win->len,
Shawn O. Pearce60bb8b12006-12-23 02:34:28 -0500899 PROT_READ, MAP_PRIVATE,
900 p->pack_fd, win->offset);
901 if (win->base == MAP_FAILED)
Shawn O. Pearce73b4e4b2006-12-23 02:34:41 -0500902 die("packfile %s cannot be mapped: %s",
903 p->pack_name,
904 strerror(errno));
Shawn O. Pearced131b7a2011-03-02 10:01:54 -0800905 if (!win->offset && win->len == p->pack_size
906 && !p->do_not_close) {
907 close(p->pack_fd);
908 pack_open_fds--;
909 p->pack_fd = -1;
910 }
Shawn O. Pearcea53128b2006-12-23 02:34:47 -0500911 pack_mmap_calls++;
912 pack_open_windows++;
913 if (pack_mapped > peak_pack_mapped)
914 peak_pack_mapped = pack_mapped;
915 if (pack_open_windows > peak_pack_open_windows)
916 peak_pack_open_windows = pack_open_windows;
Shawn O. Pearce60bb8b12006-12-23 02:34:28 -0500917 win->next = p->windows;
918 p->windows = win;
919 }
Junio C Hamano1f688552005-06-27 03:35:33 -0700920 }
Shawn O. Pearce03e79c82006-12-23 02:34:08 -0500921 if (win != *w_cursor) {
922 win->last_used = pack_used_ctr++;
923 win->inuse_cnt++;
924 *w_cursor = win;
925 }
Shawn O. Pearce60bb8b12006-12-23 02:34:28 -0500926 offset -= win->offset;
Shawn O. Pearce03e79c82006-12-23 02:34:08 -0500927 if (left)
Shawn O. Pearcedc49cd72007-03-06 20:44:37 -0500928 *left = win->len - xsize_t(offset);
Shawn O. Pearce03e79c82006-12-23 02:34:08 -0500929 return win->base + offset;
Junio C Hamano1f688552005-06-27 03:35:33 -0700930}
931
Nicolas Pitre27d69a42008-06-24 18:58:06 -0400932static struct packed_git *alloc_packed_git(int extra)
933{
934 struct packed_git *p = xmalloc(sizeof(*p) + extra);
935 memset(p, 0, sizeof(*p));
936 p->pack_fd = -1;
937 return p;
938}
939
Jonathan Niedere0500292010-11-06 14:00:38 -0500940static void try_to_free_pack_memory(size_t size)
941{
942 release_pack_memory(size, -1);
943}
944
Nicolas Pitre42873072007-03-16 16:42:50 -0400945struct packed_git *add_packed_git(const char *path, int path_len, int local)
Junio C Hamano1f688552005-06-27 03:35:33 -0700946{
Jonathan Niedere0500292010-11-06 14:00:38 -0500947 static int have_set_try_to_free_routine;
Junio C Hamano1f688552005-06-27 03:35:33 -0700948 struct stat st;
Nicolas Pitre27d69a42008-06-24 18:58:06 -0400949 struct packed_git *p = alloc_packed_git(path_len + 2);
Junio C Hamano1f688552005-06-27 03:35:33 -0700950
Jonathan Niedere0500292010-11-06 14:00:38 -0500951 if (!have_set_try_to_free_routine) {
952 have_set_try_to_free_routine = 1;
953 set_try_to_free_routine(try_to_free_pack_memory);
954 }
955
Nicolas Pitre42873072007-03-16 16:42:50 -0400956 /*
957 * Make sure a corresponding .pack file exists and that
958 * the index looks sane.
959 */
960 path_len -= strlen(".idx");
Nicolas Pitre27d69a42008-06-24 18:58:06 -0400961 if (path_len < 1) {
962 free(p);
Junio C Hamano1f688552005-06-27 03:35:33 -0700963 return NULL;
Nicolas Pitre27d69a42008-06-24 18:58:06 -0400964 }
Nicolas Pitre42873072007-03-16 16:42:50 -0400965 memcpy(p->pack_name, path, path_len);
Brandon Casey8d259312008-11-12 11:59:03 -0600966
967 strcpy(p->pack_name + path_len, ".keep");
968 if (!access(p->pack_name, F_OK))
969 p->pack_keep = 1;
970
Nicolas Pitre42873072007-03-16 16:42:50 -0400971 strcpy(p->pack_name + path_len, ".pack");
Shawn O. Pearced0798372007-05-26 01:24:19 -0400972 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
Nicolas Pitre42873072007-03-16 16:42:50 -0400973 free(p);
Junio C Hamano1f688552005-06-27 03:35:33 -0700974 return NULL;
975 }
Nicolas Pitre42873072007-03-16 16:42:50 -0400976
Junio C Hamano1f688552005-06-27 03:35:33 -0700977 /* ok, it looks sane as far as we can check without
978 * actually mapping the pack file.
979 */
Junio C Hamano1f688552005-06-27 03:35:33 -0700980 p->pack_size = st.st_size;
Linus Torvalds9d835df2005-10-13 15:38:28 -0700981 p->pack_local = local;
Junio C Hamanob8670922007-03-09 03:52:12 -0800982 p->mtime = st.st_mtime;
Nicolas Pitre42873072007-03-16 16:42:50 -0400983 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
984 hashclr(p->sha1);
Junio C Hamano1f688552005-06-27 03:35:33 -0700985 return p;
986}
987
Shawn O. Pearce7b644692010-04-19 07:23:08 -0700988struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400989{
Nicolas Pitre42873072007-03-16 16:42:50 -0400990 const char *path = sha1_pack_name(sha1);
Nicolas Pitre27d69a42008-06-24 18:58:06 -0400991 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400992
Nicolas Pitre27d69a42008-06-24 18:58:06 -0400993 strcpy(p->pack_name, path);
994 hashcpy(p->sha1, sha1);
Nicolas Pitre42873072007-03-16 16:42:50 -0400995 if (check_packed_git_idx(idx_path, p)) {
996 free(p);
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400997 return NULL;
Nicolas Pitre42873072007-03-16 16:42:50 -0400998 }
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -0400999
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -04001000 return p;
1001}
1002
1003void install_packed_git(struct packed_git *pack)
1004{
Shawn O. Pearcec7934302011-02-28 12:52:39 -08001005 if (pack->pack_fd != -1)
1006 pack_open_fds++;
1007
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -04001008 pack->next = packed_git;
1009 packed_git = pack;
1010}
1011
Nguyễn Thái Ngọc Duy543c5ca2013-02-15 19:07:10 +07001012void (*report_garbage)(const char *desc, const char *path);
1013
1014static void report_helper(const struct string_list *list,
1015 int seen_bits, int first, int last)
1016{
1017 const char *msg;
1018 switch (seen_bits) {
1019 case 0:
1020 msg = "no corresponding .idx nor .pack";
1021 break;
1022 case 1:
1023 msg = "no corresponding .idx";
1024 break;
1025 case 2:
1026 msg = "no corresponding .pack";
1027 break;
1028 default:
1029 return;
1030 }
1031 for (; first < last; first++)
1032 report_garbage(msg, list->items[first].string);
1033}
1034
1035static void report_pack_garbage(struct string_list *list)
1036{
1037 int i, baselen = -1, first = 0, seen_bits = 0;
1038
1039 if (!report_garbage)
1040 return;
1041
1042 sort_string_list(list);
1043
1044 for (i = 0; i < list->nr; i++) {
1045 const char *path = list->items[i].string;
1046 if (baselen != -1 &&
1047 strncmp(path, list->items[first].string, baselen)) {
1048 report_helper(list, seen_bits, first, i);
1049 baselen = -1;
1050 seen_bits = 0;
1051 }
1052 if (baselen == -1) {
1053 const char *dot = strrchr(path, '.');
1054 if (!dot) {
1055 report_garbage("garbage found", path);
1056 continue;
1057 }
1058 baselen = dot - path + 1;
1059 first = i;
1060 }
1061 if (!strcmp(path + baselen, "pack"))
1062 seen_bits |= 1;
1063 else if (!strcmp(path + baselen, "idx"))
1064 seen_bits |= 2;
1065 }
1066 report_helper(list, seen_bits, first, list->nr);
1067}
1068
Linus Torvalds9d835df2005-10-13 15:38:28 -07001069static void prepare_packed_git_one(char *objdir, int local)
Junio C Hamano1f688552005-06-27 03:35:33 -07001070{
Jim Meyering9cb18f52007-07-03 12:40:20 +02001071 /* Ensure that this buffer is large enough so that we can
1072 append "/pack/" without clobbering the stack even if
1073 strlen(objdir) were PATH_MAX. */
1074 char path[PATH_MAX + 1 + 4 + 1 + 1];
Junio C Hamano1f688552005-06-27 03:35:33 -07001075 int len;
1076 DIR *dir;
1077 struct dirent *de;
Nguyễn Thái Ngọc Duy543c5ca2013-02-15 19:07:10 +07001078 struct string_list garbage = STRING_LIST_INIT_DUP;
Junio C Hamano1f688552005-06-27 03:35:33 -07001079
1080 sprintf(path, "%s/pack", objdir);
1081 len = strlen(path);
1082 dir = opendir(path);
Carl Worthb5b16992006-02-17 16:14:52 -08001083 if (!dir) {
Junio C Hamano26125f62006-02-22 11:16:38 -08001084 if (errno != ENOENT)
Junio C Hamanobd2afde2006-02-22 17:47:10 -08001085 error("unable to open object pack directory: %s: %s",
Junio C Hamano26125f62006-02-22 11:16:38 -08001086 path, strerror(errno));
Junio C Hamano1f688552005-06-27 03:35:33 -07001087 return;
Carl Worthb5b16992006-02-17 16:14:52 -08001088 }
Junio C Hamano1f688552005-06-27 03:35:33 -07001089 path[len++] = '/';
1090 while ((de = readdir(dir)) != NULL) {
1091 int namelen = strlen(de->d_name);
1092 struct packed_git *p;
1093
Nguyễn Thái Ngọc Duy543c5ca2013-02-15 19:07:10 +07001094 if (len + namelen + 1 > sizeof(path)) {
1095 if (report_garbage) {
1096 struct strbuf sb = STRBUF_INIT;
1097 strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1098 report_garbage("path too long", sb.buf);
1099 strbuf_release(&sb);
1100 }
1101 continue;
1102 }
1103
1104 if (is_dot_or_dotdot(de->d_name))
Jim Meyering9cb18f52007-07-03 12:40:20 +02001105 continue;
1106
Junio C Hamano1f688552005-06-27 03:35:33 -07001107 strcpy(path + len, de->d_name);
Nguyễn Thái Ngọc Duyd90906a2013-02-13 16:13:17 +07001108
1109 if (has_extension(de->d_name, ".idx")) {
1110 /* Don't reopen a pack we already have. */
1111 for (p = packed_git; p; p = p->next) {
1112 if (!memcmp(path, p->pack_name, len + namelen - 4))
1113 break;
1114 }
1115 if (p == NULL &&
1116 /*
1117 * See if it really is a valid .idx file with
1118 * corresponding .pack file that we can map.
1119 */
1120 (p = add_packed_git(path, len + namelen, local)) != NULL)
1121 install_packed_git(p);
Jeff King86f77802006-06-02 12:49:32 -04001122 }
Nguyễn Thái Ngọc Duy543c5ca2013-02-15 19:07:10 +07001123
1124 if (!report_garbage)
1125 continue;
1126
1127 if (has_extension(de->d_name, ".idx") ||
1128 has_extension(de->d_name, ".pack") ||
1129 has_extension(de->d_name, ".keep"))
1130 string_list_append(&garbage, path);
1131 else
1132 report_garbage("garbage found", path);
Junio C Hamano1f688552005-06-27 03:35:33 -07001133 }
Junio C Hamano5b35bcd2005-07-05 23:52:17 -07001134 closedir(dir);
Nguyễn Thái Ngọc Duy543c5ca2013-02-15 19:07:10 +07001135 report_pack_garbage(&garbage);
1136 string_list_clear(&garbage, 0);
Junio C Hamano1f688552005-06-27 03:35:33 -07001137}
1138
Junio C Hamanob8670922007-03-09 03:52:12 -08001139static int sort_pack(const void *a_, const void *b_)
1140{
1141 struct packed_git *a = *((struct packed_git **)a_);
1142 struct packed_git *b = *((struct packed_git **)b_);
1143 int st;
1144
1145 /*
1146 * Local packs tend to contain objects specific to our
1147 * variant of the project than remote ones. In addition,
1148 * remote ones could be on a network mounted filesystem.
1149 * Favor local ones for these reasons.
1150 */
1151 st = a->pack_local - b->pack_local;
1152 if (st)
1153 return -st;
1154
1155 /*
1156 * Younger packs tend to contain more recent objects,
1157 * and more recent objects tend to get accessed more
1158 * often.
1159 */
1160 if (a->mtime < b->mtime)
1161 return 1;
1162 else if (a->mtime == b->mtime)
1163 return 0;
1164 return -1;
1165}
1166
1167static void rearrange_packed_git(void)
1168{
1169 struct packed_git **ary, *p;
1170 int i, n;
1171
1172 for (n = 0, p = packed_git; p; p = p->next)
1173 n++;
1174 if (n < 2)
1175 return;
1176
1177 /* prepare an array of packed_git for easier sorting */
1178 ary = xcalloc(n, sizeof(struct packed_git *));
1179 for (n = 0, p = packed_git; p; p = p->next)
1180 ary[n++] = p;
1181
1182 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1183
1184 /* link them back again */
1185 for (i = 0; i < n - 1; i++)
1186 ary[i]->next = ary[i + 1];
1187 ary[n - 1]->next = NULL;
1188 packed_git = ary[0];
1189
1190 free(ary);
1191}
1192
Jeff King637cdd92006-06-02 11:32:23 -04001193static int prepare_packed_git_run_once = 0;
Junio C Hamano9a217f22005-06-28 14:56:57 -07001194void prepare_packed_git(void)
Junio C Hamano1f688552005-06-27 03:35:33 -07001195{
Junio C Hamanod5a63b92005-08-14 17:25:57 -07001196 struct alternate_object_database *alt;
Junio C Hamano1f688552005-06-27 03:35:33 -07001197
Jeff King637cdd92006-06-02 11:32:23 -04001198 if (prepare_packed_git_run_once)
Junio C Hamano1f688552005-06-27 03:35:33 -07001199 return;
Linus Torvalds9d835df2005-10-13 15:38:28 -07001200 prepare_packed_git_one(get_object_directory(), 1);
Junio C Hamano9a217f22005-06-28 14:56:57 -07001201 prepare_alt_odb();
Junio C Hamanod5a63b92005-08-14 17:25:57 -07001202 for (alt = alt_odb_list; alt; alt = alt->next) {
Junio C Hamano1494e032005-12-04 22:48:43 -08001203 alt->name[-1] = 0;
Linus Torvalds9d835df2005-10-13 15:38:28 -07001204 prepare_packed_git_one(alt->base, 0);
Junio C Hamano1494e032005-12-04 22:48:43 -08001205 alt->name[-1] = '/';
Junio C Hamano1f688552005-06-27 03:35:33 -07001206 }
Junio C Hamanob8670922007-03-09 03:52:12 -08001207 rearrange_packed_git();
Jeff King637cdd92006-06-02 11:32:23 -04001208 prepare_packed_git_run_once = 1;
1209}
1210
Shawn Pearcefc04c412006-11-01 17:06:21 -05001211void reprepare_packed_git(void)
Jeff King637cdd92006-06-02 11:32:23 -04001212{
Nicolas Pitre4b480c62008-08-22 15:45:53 -04001213 discard_revindex();
Jeff King637cdd92006-06-02 11:32:23 -04001214 prepare_packed_git_run_once = 0;
1215 prepare_packed_git();
Junio C Hamano1f688552005-06-27 03:35:33 -07001216}
1217
Nicolas Pitre8eca0b42008-06-23 21:23:39 -04001218static void mark_bad_packed_object(struct packed_git *p,
1219 const unsigned char *sha1)
1220{
1221 unsigned i;
1222 for (i = 0; i < p->num_bad_objects; i++)
1223 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1224 return;
1225 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1226 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1227 p->num_bad_objects++;
1228}
1229
Junio C Hamanob6c4cec2010-10-28 11:13:06 -07001230static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
Nicolas Pitreac939102008-07-14 21:46:48 -04001231{
1232 struct packed_git *p;
1233 unsigned i;
1234
1235 for (p = packed_git; p; p = p->next)
1236 for (i = 0; i < p->num_bad_objects; i++)
1237 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
Junio C Hamanob6c4cec2010-10-28 11:13:06 -07001238 return p;
1239 return NULL;
Nicolas Pitreac939102008-07-14 21:46:48 -04001240}
1241
Nguyễn Thái Ngọc Duy090ea122012-03-07 17:54:18 +07001242/*
1243 * With an in-core object data in "map", rehash it to make sure the
1244 * object name actually matches "sha1" to detect object corruption.
1245 * With "map" == NULL, try reading the object named with "sha1" using
1246 * the streaming interface and rehash it to do the same.
1247 */
1248int check_sha1_signature(const unsigned char *sha1, void *map,
1249 unsigned long size, const char *type)
Linus Torvalds0fcfd162005-04-18 13:04:43 -07001250{
1251 unsigned char real_sha1[20];
Nguyễn Thái Ngọc Duy090ea122012-03-07 17:54:18 +07001252 enum object_type obj_type;
1253 struct git_istream *st;
1254 git_SHA_CTX c;
1255 char hdr[32];
1256 int hdrlen;
1257
1258 if (map) {
1259 hash_sha1_file(map, size, type, real_sha1);
1260 return hashcmp(sha1, real_sha1) ? -1 : 0;
1261 }
1262
1263 st = open_istream(sha1, &obj_type, &size, NULL);
1264 if (!st)
1265 return -1;
1266
1267 /* Generate the header */
1268 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1269
1270 /* Sha1.. */
1271 git_SHA1_Init(&c);
1272 git_SHA1_Update(&c, hdr, hdrlen);
1273 for (;;) {
1274 char buf[1024 * 16];
1275 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1276
Jeff Kingf54fac52013-03-25 16:17:17 -04001277 if (readlen < 0) {
1278 close_istream(st);
1279 return -1;
1280 }
Nguyễn Thái Ngọc Duy090ea122012-03-07 17:54:18 +07001281 if (!readlen)
1282 break;
1283 git_SHA1_Update(&c, buf, readlen);
1284 }
1285 git_SHA1_Final(real_sha1, &c);
1286 close_istream(st);
David Rientjesa89fccd2006-08-17 11:54:57 -07001287 return hashcmp(sha1, real_sha1) ? -1 : 0;
Linus Torvalds0fcfd162005-04-18 13:04:43 -07001288}
1289
Junio C Hamanof4e51682011-05-15 12:16:29 -07001290static int git_open_noatime(const char *name)
Linus Torvalds44d1c192008-06-14 11:32:37 -07001291{
1292 static int sha1_file_open_flag = O_NOATIME;
Linus Torvalds44d1c192008-06-14 11:32:37 -07001293
Shawn O. Pearcef2e872a2010-11-01 15:54:21 -07001294 for (;;) {
1295 int fd = open(name, O_RDONLY | sha1_file_open_flag);
Linus Torvalds44d1c192008-06-14 11:32:37 -07001296 if (fd >= 0)
Shawn O. Pearcef2e872a2010-11-01 15:54:21 -07001297 return fd;
1298
Shawn O. Pearcef2e872a2010-11-01 15:54:21 -07001299 /* Might the failure be due to O_NOATIME? */
1300 if (errno != ENOENT && sha1_file_open_flag) {
Linus Torvalds44d1c192008-06-14 11:32:37 -07001301 sha1_file_open_flag = 0;
Shawn O. Pearcef2e872a2010-11-01 15:54:21 -07001302 continue;
1303 }
1304
1305 return -1;
Linus Torvalds44d1c192008-06-14 11:32:37 -07001306 }
Linus Torvalds44d1c192008-06-14 11:32:37 -07001307}
1308
Jeff King052fe5e2013-07-12 02:30:48 -04001309static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1310{
1311 char *name = sha1_file_name(sha1);
1312 struct alternate_object_database *alt;
1313
1314 if (!lstat(name, st))
1315 return 0;
1316
1317 prepare_alt_odb();
1318 errno = ENOENT;
1319 for (alt = alt_odb_list; alt; alt = alt->next) {
1320 name = alt->name;
1321 fill_sha1_path(name, sha1);
1322 if (!lstat(alt->base, st))
1323 return 0;
1324 }
1325
1326 return -1;
1327}
1328
Linus Torvalds44d1c192008-06-14 11:32:37 -07001329static int open_sha1_file(const unsigned char *sha1)
1330{
1331 int fd;
1332 char *name = sha1_file_name(sha1);
1333 struct alternate_object_database *alt;
1334
Junio C Hamanof4e51682011-05-15 12:16:29 -07001335 fd = git_open_noatime(name);
Linus Torvalds44d1c192008-06-14 11:32:37 -07001336 if (fd >= 0)
1337 return fd;
1338
1339 prepare_alt_odb();
1340 errno = ENOENT;
1341 for (alt = alt_odb_list; alt; alt = alt->next) {
1342 name = alt->name;
1343 fill_sha1_path(name, sha1);
Junio C Hamanof4e51682011-05-15 12:16:29 -07001344 fd = git_open_noatime(alt->base);
Linus Torvalds44d1c192008-06-14 11:32:37 -07001345 if (fd >= 0)
1346 return fd;
1347 }
1348 return -1;
1349}
1350
Junio C Hamanof0270ef2011-05-14 19:42:10 -07001351void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
Linus Torvalds0fcfd162005-04-18 13:04:43 -07001352{
Linus Torvalds0fcfd162005-04-18 13:04:43 -07001353 void *map;
Linus Torvalds144bde72005-04-23 11:09:32 -07001354 int fd;
Junio C Hamanoace15342005-05-07 00:38:04 -07001355
Linus Torvalds44d1c192008-06-14 11:32:37 -07001356 fd = open_sha1_file(sha1);
1357 map = NULL;
1358 if (fd >= 0) {
1359 struct stat st;
Linus Torvalds0fcfd162005-04-18 13:04:43 -07001360
Linus Torvalds44d1c192008-06-14 11:32:37 -07001361 if (!fstat(fd, &st)) {
1362 *size = xsize_t(st.st_size);
Matthieu Moy33e42de2012-02-06 17:24:52 +01001363 if (!*size) {
1364 /* mmap() is forbidden on empty files */
1365 error("object file %s is empty", sha1_file_name(sha1));
1366 return NULL;
1367 }
Linus Torvalds44d1c192008-06-14 11:32:37 -07001368 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
Linus Torvalds144bde72005-04-23 11:09:32 -07001369 }
Linus Torvalds44d1c192008-06-14 11:32:37 -07001370 close(fd);
Linus Torvalds0fcfd162005-04-18 13:04:43 -07001371 }
Linus Torvalds0fcfd162005-04-18 13:04:43 -07001372 return map;
1373}
1374
Junio C Hamanocc5c54e2011-06-08 11:29:01 -07001375/*
1376 * There used to be a second loose object header format which
1377 * was meant to mimic the in-pack format, allowing for direct
1378 * copy of the object data. This format turned up not to be
1379 * really worth it and we no longer write loose objects in that
1380 * format.
1381 */
1382static int experimental_loose_object(unsigned char *map)
Junio C Hamanobb6b8e42006-07-17 15:04:47 -07001383{
1384 unsigned int word;
1385
1386 /*
Roberto Tyley7f684a22011-08-07 19:46:13 +01001387 * We must determine if the buffer contains the standard
1388 * zlib-deflated stream or the experimental format based
1389 * on the in-pack object format. Compare the header byte
1390 * for each format:
1391 *
1392 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1393 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1394 *
1395 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1396 * in standard loose-object format, UNLESS it is a Git-pack
1397 * format object *exactly* 8 bytes in size when inflated.
1398 *
1399 * However, RFC1950 also specifies that the 1st 16-bit word
1400 * must be divisible by 31 - this checksum tells us our buffer
1401 * is in the standard format, giving a false positive only if
1402 * the 1st word of the Git-pack format object happens to be
1403 * divisible by 31, ie:
1404 * ((byte0 * 256) + byte1) % 31 = 0
1405 * => 0ttt10000www1000 % 31 = 0
1406 *
1407 * As it happens, this case can only arise for www=3 & ttt=1
1408 * - ie, a Commit object, which would have to be 8 bytes in
1409 * size. As no Commit can be that small, we find that the
1410 * combination of these two criteria (bitmask & checksum)
1411 * can always correctly determine the buffer format.
Junio C Hamanobb6b8e42006-07-17 15:04:47 -07001412 */
1413 word = (map[0] << 8) + map[1];
Roberto Tyley7f684a22011-08-07 19:46:13 +01001414 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
Junio C Hamanobb6b8e42006-07-17 15:04:47 -07001415 return 0;
Junio C Hamanocc5c54e2011-06-08 11:29:01 -07001416 else
1417 return 1;
Junio C Hamanobb6b8e42006-07-17 15:04:47 -07001418}
1419
Nicolas Pitre09ded042008-10-29 19:02:46 -04001420unsigned long unpack_object_header_buffer(const unsigned char *buf,
1421 unsigned long len, enum object_type *type, unsigned long *sizep)
Junio C Hamanoad1ed5e2006-09-01 15:17:01 -07001422{
1423 unsigned shift;
Linus Torvalds48fb7de2009-06-17 17:22:27 -07001424 unsigned long size, c;
Junio C Hamanoad1ed5e2006-09-01 15:17:01 -07001425 unsigned long used = 0;
1426
1427 c = buf[used++];
1428 *type = (c >> 4) & 7;
1429 size = c & 15;
1430 shift = 4;
1431 while (c & 0x80) {
Pierre Habouzitf630cfd2009-07-22 23:34:34 +02001432 if (len <= used || bitsizeof(long) <= shift) {
Nicolas Pitre09ded042008-10-29 19:02:46 -04001433 error("bad object header");
Junio C Hamanoea4f9682011-10-27 11:42:57 -07001434 size = used = 0;
1435 break;
Nicolas Pitre09ded042008-10-29 19:02:46 -04001436 }
Junio C Hamanoad1ed5e2006-09-01 15:17:01 -07001437 c = buf[used++];
1438 size += (c & 0x7f) << shift;
1439 shift += 7;
1440 }
1441 *sizep = size;
1442 return used;
1443}
1444
Junio C Hamanoeb4f4072011-07-19 09:33:03 -07001445int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
Linus Torvaldsc4483572005-06-01 17:54:59 -07001446{
Junio C Hamanoad1ed5e2006-09-01 15:17:01 -07001447 unsigned long size, used;
1448 static const char valid_loose_object_type[8] = {
1449 0, /* OBJ_EXT */
1450 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1451 0, /* "delta" and others are invalid in a loose object */
Linus Torvalds93821bd2006-07-11 12:48:08 -07001452 };
Junio C Hamanoad1ed5e2006-09-01 15:17:01 -07001453 enum object_type type;
Linus Torvalds93821bd2006-07-11 12:48:08 -07001454
Linus Torvaldsc4483572005-06-01 17:54:59 -07001455 /* Get the data stream */
1456 memset(stream, 0, sizeof(*stream));
1457 stream->next_in = map;
1458 stream->avail_in = mapsize;
1459 stream->next_out = buffer;
Linus Torvalds93821bd2006-07-11 12:48:08 -07001460 stream->avail_out = bufsiz;
Linus Torvaldsc4483572005-06-01 17:54:59 -07001461
Junio C Hamanocc5c54e2011-06-08 11:29:01 -07001462 if (experimental_loose_object(map)) {
1463 /*
1464 * The old experimental format we no longer produce;
1465 * we can still read it.
1466 */
1467 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1468 if (!used || !valid_loose_object_type[type])
1469 return -1;
1470 map += used;
1471 mapsize -= used;
1472
1473 /* Set up the stream for the rest.. */
1474 stream->next_in = map;
1475 stream->avail_in = mapsize;
Linus Torvalds39c68542009-01-07 19:54:47 -08001476 git_inflate_init(stream);
Junio C Hamanocc5c54e2011-06-08 11:29:01 -07001477
1478 /* And generate the fake traditional header */
1479 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1480 typename(type), size);
1481 return 0;
Linus Torvalds93821bd2006-07-11 12:48:08 -07001482 }
Linus Torvalds39c68542009-01-07 19:54:47 -08001483 git_inflate_init(stream);
Junio C Hamanocc5c54e2011-06-08 11:29:01 -07001484 return git_inflate(stream, 0);
Linus Torvaldsc4483572005-06-01 17:54:59 -07001485}
1486
Junio C Hamanoef49a7a2011-06-10 11:52:15 -07001487static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
Linus Torvalds0fcfd162005-04-18 13:04:43 -07001488{
Linus Torvalds5180cac2005-06-02 07:57:25 -07001489 int bytes = strlen(buffer) + 1;
Ilari Liusvaara3aee68a2010-01-26 20:24:14 +02001490 unsigned char *buf = xmallocz(size);
Linus Torvalds93821bd2006-07-11 12:48:08 -07001491 unsigned long n;
Junio C Hamano7efbff72007-03-05 00:21:37 -08001492 int status = Z_OK;
Linus Torvalds0fcfd162005-04-18 13:04:43 -07001493
Linus Torvalds93821bd2006-07-11 12:48:08 -07001494 n = stream->total_out - bytes;
1495 if (n > size)
1496 n = size;
1497 memcpy(buf, (char *) buffer + bytes, n);
1498 bytes = n;
Linus Torvalds456cdf62007-03-19 22:49:53 -07001499 if (bytes <= size) {
1500 /*
1501 * The above condition must be (bytes <= size), not
1502 * (bytes < size). In other words, even though we
Junio C Hamanoccf5ace2011-05-15 12:16:03 -07001503 * expect no more output and set avail_out to zero,
Linus Torvalds456cdf62007-03-19 22:49:53 -07001504 * the input zlib stream may have bytes that express
1505 * "this concludes the stream", and we *do* want to
1506 * eat that input.
1507 *
1508 * Otherwise we would not be able to test that we
1509 * consumed all the input to reach the expected size;
1510 * we also want to check that zlib tells us that all
1511 * went well with status == Z_STREAM_END at the end.
1512 */
Linus Torvalds5180cac2005-06-02 07:57:25 -07001513 stream->next_out = buf + bytes;
1514 stream->avail_out = size - bytes;
Junio C Hamano7efbff72007-03-05 00:21:37 -08001515 while (status == Z_OK)
Linus Torvalds39c68542009-01-07 19:54:47 -08001516 status = git_inflate(stream, Z_FINISH);
Linus Torvalds0fcfd162005-04-18 13:04:43 -07001517 }
Linus Torvalds456cdf62007-03-19 22:49:53 -07001518 if (status == Z_STREAM_END && !stream->avail_in) {
Linus Torvalds39c68542009-01-07 19:54:47 -08001519 git_inflate_end(stream);
Junio C Hamano7efbff72007-03-05 00:21:37 -08001520 return buf;
1521 }
1522
1523 if (status < 0)
1524 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1525 else if (stream->avail_in)
1526 error("garbage at end of loose object '%s'",
1527 sha1_to_hex(sha1));
1528 free(buf);
1529 return NULL;
Linus Torvalds0fcfd162005-04-18 13:04:43 -07001530}
1531
Linus Torvalds5180cac2005-06-02 07:57:25 -07001532/*
1533 * We used to just use "sscanf()", but that's actually way
1534 * too permissive for what we want to check. So do an anal
1535 * object header parse by hand.
1536 */
Junio C Hamanof0270ef2011-05-14 19:42:10 -07001537int parse_sha1_header(const char *hdr, unsigned long *sizep)
Linus Torvalds5180cac2005-06-02 07:57:25 -07001538{
Nicolas Pitre21666f12007-02-26 14:55:59 -05001539 char type[10];
Linus Torvalds5180cac2005-06-02 07:57:25 -07001540 int i;
1541 unsigned long size;
1542
1543 /*
Junio C Hamanoa6080a02007-06-07 00:04:01 -07001544 * The type can be at most ten bytes (including the
Linus Torvalds5180cac2005-06-02 07:57:25 -07001545 * terminating '\0' that we add), and is followed by
Nicolas Pitre21666f12007-02-26 14:55:59 -05001546 * a space.
Linus Torvalds5180cac2005-06-02 07:57:25 -07001547 */
Nicolas Pitre21666f12007-02-26 14:55:59 -05001548 i = 0;
Linus Torvalds5180cac2005-06-02 07:57:25 -07001549 for (;;) {
1550 char c = *hdr++;
1551 if (c == ' ')
1552 break;
Nicolas Pitre21666f12007-02-26 14:55:59 -05001553 type[i++] = c;
1554 if (i >= sizeof(type))
Linus Torvalds5180cac2005-06-02 07:57:25 -07001555 return -1;
Linus Torvalds5180cac2005-06-02 07:57:25 -07001556 }
Nicolas Pitre21666f12007-02-26 14:55:59 -05001557 type[i] = 0;
Linus Torvalds5180cac2005-06-02 07:57:25 -07001558
1559 /*
1560 * The length must follow immediately, and be in canonical
1561 * decimal format (ie "010" is not valid).
1562 */
1563 size = *hdr++ - '0';
1564 if (size > 9)
1565 return -1;
1566 if (size) {
1567 for (;;) {
1568 unsigned long c = *hdr - '0';
1569 if (c > 9)
1570 break;
1571 hdr++;
1572 size = size * 10 + c;
1573 }
1574 }
1575 *sizep = size;
1576
1577 /*
1578 * The length must be followed by a zero byte
1579 */
Nicolas Pitre21666f12007-02-26 14:55:59 -05001580 return *hdr ? -1 : type_from_string(type);
Linus Torvalds5180cac2005-06-02 07:57:25 -07001581}
1582
Junio C Hamano7efbff72007-03-05 00:21:37 -08001583static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
Linus Torvalds5180cac2005-06-02 07:57:25 -07001584{
1585 int ret;
Junio C Hamanoef49a7a2011-06-10 11:52:15 -07001586 git_zstream stream;
Linus Torvalds5180cac2005-06-02 07:57:25 -07001587 char hdr[8192];
1588
1589 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
Nicolas Pitre21666f12007-02-26 14:55:59 -05001590 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
Linus Torvalds5180cac2005-06-02 07:57:25 -07001591 return NULL;
1592
Junio C Hamano7efbff72007-03-05 00:21:37 -08001593 return unpack_sha1_rest(&stream, hdr, *size, sha1);
Linus Torvalds5180cac2005-06-02 07:57:25 -07001594}
1595
Nicolas Pitre54dab522007-04-16 12:31:56 -04001596unsigned long get_size_from_delta(struct packed_git *p,
1597 struct pack_window **w_curs,
1598 off_t curpos)
1599{
1600 const unsigned char *data;
1601 unsigned char delta_head[20], *in;
Junio C Hamanoef49a7a2011-06-10 11:52:15 -07001602 git_zstream stream;
Nicolas Pitre54dab522007-04-16 12:31:56 -04001603 int st;
1604
1605 memset(&stream, 0, sizeof(stream));
1606 stream.next_out = delta_head;
1607 stream.avail_out = sizeof(delta_head);
1608
Linus Torvalds39c68542009-01-07 19:54:47 -08001609 git_inflate_init(&stream);
Nicolas Pitre54dab522007-04-16 12:31:56 -04001610 do {
1611 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1612 stream.next_in = in;
Linus Torvalds39c68542009-01-07 19:54:47 -08001613 st = git_inflate(&stream, Z_FINISH);
Nicolas Pitre54dab522007-04-16 12:31:56 -04001614 curpos += stream.next_in - in;
1615 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1616 stream.total_out < sizeof(delta_head));
Linus Torvalds39c68542009-01-07 19:54:47 -08001617 git_inflate_end(&stream);
Nicolas Pitre3d77d872008-10-29 19:02:47 -04001618 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1619 error("delta data unpack-initial failed");
1620 return 0;
1621 }
Nicolas Pitre54dab522007-04-16 12:31:56 -04001622
1623 /* Examine the initial part of the delta to figure out
1624 * the result size.
1625 */
1626 data = delta_head;
1627
1628 /* ignore base size */
1629 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1630
1631 /* Read the result size */
1632 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1633}
1634
Shawn O. Pearcec4001d92007-03-06 20:44:30 -05001635static off_t get_delta_base(struct packed_git *p,
Shawn O. Pearce03e79c82006-12-23 02:34:08 -05001636 struct pack_window **w_curs,
Shawn O. Pearcec4001d92007-03-06 20:44:30 -05001637 off_t *curpos,
Nicolas Pitre21666f12007-02-26 14:55:59 -05001638 enum object_type type,
Shawn O. Pearcec4001d92007-03-06 20:44:30 -05001639 off_t delta_obj_offset)
Nicolas Pitreeb32d232006-09-21 00:06:49 -04001640{
Nicolas Pitre2b87c452007-02-26 14:55:56 -05001641 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
Shawn O. Pearcec4001d92007-03-06 20:44:30 -05001642 off_t base_offset;
Nicolas Pitreeb32d232006-09-21 00:06:49 -04001643
Shawn O. Pearce8d8a4ea2006-12-23 02:34:18 -05001644 /* use_pack() assured us we have [base_info, base_info + 20)
1645 * as a range that we can look at without walking off the
1646 * end of the mapped window. Its actually the hash size
1647 * that is assured. An OFS_DELTA longer than the hash size
1648 * is stupid, as then a REF_DELTA would be smaller to store.
1649 */
Nicolas Pitre21666f12007-02-26 14:55:59 -05001650 if (type == OBJ_OFS_DELTA) {
Nicolas Pitreeb32d232006-09-21 00:06:49 -04001651 unsigned used = 0;
1652 unsigned char c = base_info[used++];
1653 base_offset = c & 127;
1654 while (c & 128) {
1655 base_offset += 1;
Nicolas Pitre8723f212007-04-09 01:06:29 -04001656 if (!base_offset || MSB(base_offset, 7))
Nicolas Pitre8eca0b42008-06-23 21:23:39 -04001657 return 0; /* overflow */
Nicolas Pitreeb32d232006-09-21 00:06:49 -04001658 c = base_info[used++];
1659 base_offset = (base_offset << 7) + (c & 127);
1660 }
1661 base_offset = delta_obj_offset - base_offset;
Nicolas Pitred8f32552008-10-29 19:02:45 -04001662 if (base_offset <= 0 || base_offset >= delta_obj_offset)
Nicolas Pitre8eca0b42008-06-23 21:23:39 -04001663 return 0; /* out of bound */
Nicolas Pitre2b87c452007-02-26 14:55:56 -05001664 *curpos += used;
Nicolas Pitre21666f12007-02-26 14:55:59 -05001665 } else if (type == OBJ_REF_DELTA) {
Nicolas Pitreeb32d232006-09-21 00:06:49 -04001666 /* The base entry _must_ be in the same pack */
1667 base_offset = find_pack_entry_one(base_info, p);
Nicolas Pitre2b87c452007-02-26 14:55:56 -05001668 *curpos += 20;
Nicolas Pitreeb32d232006-09-21 00:06:49 -04001669 } else
1670 die("I am totally screwed");
Nicolas Pitre2b87c452007-02-26 14:55:56 -05001671 return base_offset;
Nicolas Pitreeb32d232006-09-21 00:06:49 -04001672}
1673
Junio C Hamanof8c8abc2011-05-13 15:33:33 -07001674int unpack_object_header(struct packed_git *p,
1675 struct pack_window **w_curs,
1676 off_t *curpos,
1677 unsigned long *sizep)
Linus Torvaldsa733cb62005-06-28 14:21:02 -07001678{
Shawn O. Pearce03e79c82006-12-23 02:34:08 -05001679 unsigned char *base;
Junio C Hamanoef49a7a2011-06-10 11:52:15 -07001680 unsigned long left;
Junio C Hamanoad1ed5e2006-09-01 15:17:01 -07001681 unsigned long used;
Nicolas Pitre2b87c452007-02-26 14:55:56 -05001682 enum object_type type;
Linus Torvaldsa733cb62005-06-28 14:21:02 -07001683
Shawn O. Pearce8d8a4ea2006-12-23 02:34:18 -05001684 /* use_pack() assures us we have [base, base + 20) available
Jim Meyering0353a0c2011-04-13 17:39:40 +02001685 * as a range that we can look at. (Its actually the hash
Pavel Roskin3dff5372007-02-03 23:49:16 -05001686 * size that is assured.) With our object header encoding
Shawn O. Pearce8d8a4ea2006-12-23 02:34:18 -05001687 * the maximum deflated object size is 2^137, which is just
1688 * insane, so we know won't exceed what we have been given.
1689 */
Nicolas Pitre2b87c452007-02-26 14:55:56 -05001690 base = use_pack(p, w_curs, *curpos, &left);
Nicolas Pitre09ded042008-10-29 19:02:46 -04001691 used = unpack_object_header_buffer(base, left, &type, sizep);
1692 if (!used) {
1693 type = OBJ_BAD;
1694 } else
1695 *curpos += used;
Junio C Hamanoad1ed5e2006-09-01 15:17:01 -07001696
Nicolas Pitre2b87c452007-02-26 14:55:56 -05001697 return type;
Linus Torvaldsa733cb62005-06-28 14:21:02 -07001698}
1699
Thomas Rast790d96c2013-03-25 19:07:39 +01001700static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1701{
1702 int type;
1703 struct revindex_entry *revidx;
1704 const unsigned char *sha1;
1705 revidx = find_pack_revindex(p, obj_offset);
1706 if (!revidx)
1707 return OBJ_BAD;
1708 sha1 = nth_packed_object_sha1(p, revidx->nr);
1709 mark_bad_packed_object(p, sha1);
1710 type = sha1_object_info(sha1, NULL);
1711 if (type <= OBJ_NONE)
1712 return OBJ_BAD;
1713 return type;
1714}
1715
Thomas Rast790d96c2013-03-25 19:07:39 +01001716#define POI_STACK_PREALLOC 64
1717
Jeff King90191d32013-07-12 02:31:57 -04001718static enum object_type packed_to_object_type(struct packed_git *p,
1719 off_t obj_offset,
1720 enum object_type type,
1721 struct pack_window **w_curs,
1722 off_t curpos)
Junio C Hamano1f688552005-06-27 03:35:33 -07001723{
Thomas Rast790d96c2013-03-25 19:07:39 +01001724 off_t small_poi_stack[POI_STACK_PREALLOC];
1725 off_t *poi_stack = small_poi_stack;
1726 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
Junio C Hamano5db47c22005-06-27 23:58:08 -07001727
Thomas Rast790d96c2013-03-25 19:07:39 +01001728 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1729 off_t base_offset;
Jeff King90191d32013-07-12 02:31:57 -04001730 unsigned long size;
Thomas Rast790d96c2013-03-25 19:07:39 +01001731 /* Push the object we're going to leave behind */
1732 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1733 poi_stack_alloc = alloc_nr(poi_stack_nr);
1734 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1735 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1736 } else {
1737 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1738 }
1739 poi_stack[poi_stack_nr++] = obj_offset;
1740 /* If parsing the base offset fails, just unwind */
Jeff King90191d32013-07-12 02:31:57 -04001741 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
Thomas Rast790d96c2013-03-25 19:07:39 +01001742 if (!base_offset)
1743 goto unwind;
1744 curpos = obj_offset = base_offset;
Jeff King90191d32013-07-12 02:31:57 -04001745 type = unpack_object_header(p, w_curs, &curpos, &size);
Thomas Rast790d96c2013-03-25 19:07:39 +01001746 if (type <= OBJ_NONE) {
1747 /* If getting the base itself fails, we first
1748 * retry the base, otherwise unwind */
1749 type = retry_bad_packed_offset(p, base_offset);
1750 if (type > OBJ_NONE)
1751 goto out;
1752 goto unwind;
1753 }
1754 }
1755
Nicolas Pitre21666f12007-02-26 14:55:59 -05001756 switch (type) {
Thomas Rast790d96c2013-03-25 19:07:39 +01001757 case OBJ_BAD:
Linus Torvaldsa733cb62005-06-28 14:21:02 -07001758 case OBJ_COMMIT:
Linus Torvaldsa733cb62005-06-28 14:21:02 -07001759 case OBJ_TREE:
Linus Torvaldsa733cb62005-06-28 14:21:02 -07001760 case OBJ_BLOB:
Linus Torvaldsa733cb62005-06-28 14:21:02 -07001761 case OBJ_TAG:
Linus Torvaldsa69d0942005-06-28 09:58:23 -07001762 break;
Junio C Hamano1f688552005-06-27 03:35:33 -07001763 default:
Nicolas Pitre3d77d872008-10-29 19:02:47 -04001764 error("unknown object type %i at offset %"PRIuMAX" in %s",
1765 type, (uintmax_t)obj_offset, p->pack_name);
1766 type = OBJ_BAD;
Junio C Hamano1f688552005-06-27 03:35:33 -07001767 }
Thomas Rast790d96c2013-03-25 19:07:39 +01001768
1769out:
1770 if (poi_stack != small_poi_stack)
1771 free(poi_stack);
Nicolas Pitre21666f12007-02-26 14:55:59 -05001772 return type;
Thomas Rast790d96c2013-03-25 19:07:39 +01001773
1774unwind:
1775 while (poi_stack_nr) {
1776 obj_offset = poi_stack[--poi_stack_nr];
1777 type = retry_bad_packed_offset(p, obj_offset);
1778 if (type > OBJ_NONE)
1779 goto out;
1780 }
1781 type = OBJ_BAD;
1782 goto out;
Junio C Hamano1f688552005-06-27 03:35:33 -07001783}
1784
Junio C Hamano1f688552005-06-27 03:35:33 -07001785static int packed_object_info(struct packed_git *p, off_t obj_offset,
Jeff King23c339c2013-07-12 02:37:53 -04001786 struct object_info *oi)
Junio C Hamano1f688552005-06-27 03:35:33 -07001787{
1788 struct pack_window *w_curs = NULL;
1789 unsigned long size;
1790 off_t curpos = obj_offset;
1791 enum object_type type;
Junio C Hamano1f688552005-06-27 03:35:33 -07001792
Jeff King412916e2013-07-12 02:32:25 -04001793 /*
1794 * We always get the representation type, but only convert it to
1795 * a "real" type later if the caller is interested.
1796 */
Junio C Hamano1f688552005-06-27 03:35:33 -07001797 type = unpack_object_header(p, &w_curs, &curpos, &size);
1798
Jeff King23c339c2013-07-12 02:37:53 -04001799 if (oi->sizep) {
Junio C Hamano1f688552005-06-27 03:35:33 -07001800 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1801 off_t tmp_pos = curpos;
1802 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1803 type, obj_offset);
1804 if (!base_offset) {
1805 type = OBJ_BAD;
1806 goto out;
1807 }
Jeff King23c339c2013-07-12 02:37:53 -04001808 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1809 if (*oi->sizep == 0) {
Junio C Hamano1f688552005-06-27 03:35:33 -07001810 type = OBJ_BAD;
1811 goto out;
1812 }
1813 } else {
Jeff King23c339c2013-07-12 02:37:53 -04001814 *oi->sizep = size;
Junio C Hamano1f688552005-06-27 03:35:33 -07001815 }
1816 }
1817
Jeff King23c339c2013-07-12 02:37:53 -04001818 if (oi->disk_sizep) {
Jeff King161f00e2013-07-07 06:04:00 -04001819 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
Jeff King23c339c2013-07-12 02:37:53 -04001820 *oi->disk_sizep = revidx[1].offset - obj_offset;
Jeff King161f00e2013-07-07 06:04:00 -04001821 }
1822
Jeff King23c339c2013-07-12 02:37:53 -04001823 if (oi->typep) {
1824 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1825 if (*oi->typep < 0) {
Jeff King412916e2013-07-12 02:32:25 -04001826 type = OBJ_BAD;
1827 goto out;
1828 }
1829 }
Junio C Hamano1f688552005-06-27 03:35:33 -07001830
1831out:
Junio C Hamano1f688552005-06-27 03:35:33 -07001832 unuse_pack(&w_curs);
1833 return type;
Junio C Hamano1f688552005-06-27 03:35:33 -07001834}
1835
Shawn Pearceeb950c12006-08-26 04:12:27 -04001836static void *unpack_compressed_entry(struct packed_git *p,
Shawn O. Pearce03e79c82006-12-23 02:34:08 -05001837 struct pack_window **w_curs,
Shawn O. Pearcec4001d92007-03-06 20:44:30 -05001838 off_t curpos,
Shawn Pearceeb950c12006-08-26 04:12:27 -04001839 unsigned long size)
Shawn Pearcede530aa2006-08-26 04:10:43 -04001840{
1841 int st;
Junio C Hamanoef49a7a2011-06-10 11:52:15 -07001842 git_zstream stream;
Shawn O. Pearce079afb12006-12-23 02:34:13 -05001843 unsigned char *buffer, *in;
Shawn Pearcede530aa2006-08-26 04:10:43 -04001844
Ilari Liusvaara4ab07e42010-01-26 20:24:15 +02001845 buffer = xmallocz(size);
Shawn Pearcede530aa2006-08-26 04:10:43 -04001846 memset(&stream, 0, sizeof(stream));
Shawn Pearcede530aa2006-08-26 04:10:43 -04001847 stream.next_out = buffer;
Junio C Hamano39eea7b2009-10-21 23:06:14 -07001848 stream.avail_out = size + 1;
Shawn Pearcede530aa2006-08-26 04:10:43 -04001849
Linus Torvalds39c68542009-01-07 19:54:47 -08001850 git_inflate_init(&stream);
Shawn O. Pearce079afb12006-12-23 02:34:13 -05001851 do {
Nicolas Pitre2b87c452007-02-26 14:55:56 -05001852 in = use_pack(p, w_curs, curpos, &stream.avail_in);
Shawn O. Pearce079afb12006-12-23 02:34:13 -05001853 stream.next_in = in;
Linus Torvalds39c68542009-01-07 19:54:47 -08001854 st = git_inflate(&stream, Z_FINISH);
Junio C Hamano39eea7b2009-10-21 23:06:14 -07001855 if (!stream.avail_out)
1856 break; /* the payload is larger than it should be */
Nicolas Pitre2b87c452007-02-26 14:55:56 -05001857 curpos += stream.next_in - in;
Shawn O. Pearce079afb12006-12-23 02:34:13 -05001858 } while (st == Z_OK || st == Z_BUF_ERROR);
Linus Torvalds39c68542009-01-07 19:54:47 -08001859 git_inflate_end(&stream);
Shawn Pearcede530aa2006-08-26 04:10:43 -04001860 if ((st != Z_STREAM_END) || stream.total_out != size) {
1861 free(buffer);
1862 return NULL;
1863 }
1864
1865 return buffer;
1866}
1867
Linus Torvaldse5e01612007-03-17 12:44:06 -07001868#define MAX_DELTA_CACHE (256)
1869
Shawn O. Pearce18bdec12007-03-19 01:14:37 -04001870static size_t delta_base_cached;
Nicolas Pitre5e08ecb2007-03-19 16:31:04 -04001871
1872static struct delta_base_cache_lru_list {
1873 struct delta_base_cache_lru_list *prev;
1874 struct delta_base_cache_lru_list *next;
1875} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1876
Linus Torvaldse5e01612007-03-17 12:44:06 -07001877static struct delta_base_cache_entry {
Nicolas Pitre5e08ecb2007-03-19 16:31:04 -04001878 struct delta_base_cache_lru_list lru;
1879 void *data;
Linus Torvaldse5e01612007-03-17 12:44:06 -07001880 struct packed_git *p;
1881 off_t base_offset;
1882 unsigned long size;
Linus Torvaldse5e01612007-03-17 12:44:06 -07001883 enum object_type type;
1884} delta_base_cache[MAX_DELTA_CACHE];
1885
1886static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1887{
1888 unsigned long hash;
1889
1890 hash = (unsigned long)p + (unsigned long)base_offset;
1891 hash += (hash >> 8) + (hash >> 16);
Nicolas Pitre33580042007-03-19 16:28:51 -04001892 return hash % MAX_DELTA_CACHE;
Linus Torvaldse5e01612007-03-17 12:44:06 -07001893}
1894
Thomas Rast84dd81c2013-03-27 21:03:41 +01001895static struct delta_base_cache_entry *
1896get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
Junio C Hamano5266d362011-05-13 13:20:43 -07001897{
1898 unsigned long hash = pack_entry_hash(p, base_offset);
Thomas Rast84dd81c2013-03-27 21:03:41 +01001899 return delta_base_cache + hash;
1900}
1901
1902static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1903 struct packed_git *p, off_t base_offset)
1904{
Junio C Hamano5266d362011-05-13 13:20:43 -07001905 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1906}
1907
Thomas Rast84dd81c2013-03-27 21:03:41 +01001908static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1909{
1910 struct delta_base_cache_entry *ent;
1911 ent = get_delta_base_cache_entry(p, base_offset);
1912 return eq_delta_base_cache_entry(ent, p, base_offset);
1913}
1914
1915static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1916{
1917 ent->data = NULL;
1918 ent->lru.next->prev = ent->lru.prev;
1919 ent->lru.prev->next = ent->lru.next;
1920 delta_base_cached -= ent->size;
1921}
1922
Linus Torvalds62f255a2007-03-17 12:42:15 -07001923static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
Nicolas Pitrea0cba102007-03-17 21:13:57 -04001924 unsigned long *base_size, enum object_type *type, int keep_cache)
Linus Torvalds62f255a2007-03-17 12:42:15 -07001925{
Thomas Rast84dd81c2013-03-27 21:03:41 +01001926 struct delta_base_cache_entry *ent;
Linus Torvaldse5e01612007-03-17 12:44:06 -07001927 void *ret;
Linus Torvaldse5e01612007-03-17 12:44:06 -07001928
Thomas Rast84dd81c2013-03-27 21:03:41 +01001929 ent = get_delta_base_cache_entry(p, base_offset);
1930
1931 if (!eq_delta_base_cache_entry(ent, p, base_offset))
Miklos Vajna749bc582008-10-09 02:11:24 +02001932 return unpack_entry(p, base_offset, type, base_size);
Linus Torvaldse5e01612007-03-17 12:44:06 -07001933
Thomas Rast84dd81c2013-03-27 21:03:41 +01001934 ret = ent->data;
1935
1936 if (!keep_cache)
1937 clear_delta_base_cache_entry(ent);
1938 else
Pierre Habouzit182af832007-09-16 00:32:36 +02001939 ret = xmemdupz(ent->data, ent->size);
Linus Torvaldse5e01612007-03-17 12:44:06 -07001940 *type = ent->type;
1941 *base_size = ent->size;
1942 return ret;
Linus Torvalds62f255a2007-03-17 12:42:15 -07001943}
1944
Shawn O. Pearce18bdec12007-03-19 01:14:37 -04001945static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1946{
1947 if (ent->data) {
1948 free(ent->data);
1949 ent->data = NULL;
Nicolas Pitre5e08ecb2007-03-19 16:31:04 -04001950 ent->lru.next->prev = ent->lru.prev;
1951 ent->lru.prev->next = ent->lru.next;
Shawn O. Pearce18bdec12007-03-19 01:14:37 -04001952 delta_base_cached -= ent->size;
1953 }
1954}
1955
Shawn O. Pearce3d20c632009-02-10 13:36:12 -08001956void clear_delta_base_cache(void)
1957{
1958 unsigned long p;
1959 for (p = 0; p < MAX_DELTA_CACHE; p++)
1960 release_delta_base_cache(&delta_base_cache[p]);
1961}
1962
Linus Torvalds62f255a2007-03-17 12:42:15 -07001963static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1964 void *base, unsigned long base_size, enum object_type type)
1965{
Nicolas Pitre5e08ecb2007-03-19 16:31:04 -04001966 unsigned long hash = pack_entry_hash(p, base_offset);
Linus Torvaldse5e01612007-03-17 12:44:06 -07001967 struct delta_base_cache_entry *ent = delta_base_cache + hash;
Nicolas Pitre5e08ecb2007-03-19 16:31:04 -04001968 struct delta_base_cache_lru_list *lru;
Linus Torvaldse5e01612007-03-17 12:44:06 -07001969
Shawn O. Pearce18bdec12007-03-19 01:14:37 -04001970 release_delta_base_cache(ent);
1971 delta_base_cached += base_size;
Nicolas Pitre5e08ecb2007-03-19 16:31:04 -04001972
1973 for (lru = delta_base_cache_lru.next;
1974 delta_base_cached > delta_base_cache_limit
1975 && lru != &delta_base_cache_lru;
1976 lru = lru->next) {
1977 struct delta_base_cache_entry *f = (void *)lru;
Shawn O. Pearce18bdec12007-03-19 01:14:37 -04001978 if (f->type == OBJ_BLOB)
1979 release_delta_base_cache(f);
1980 }
Nicolas Pitre5e08ecb2007-03-19 16:31:04 -04001981 for (lru = delta_base_cache_lru.next;
1982 delta_base_cached > delta_base_cache_limit
1983 && lru != &delta_base_cache_lru;
1984 lru = lru->next) {
1985 struct delta_base_cache_entry *f = (void *)lru;
1986 release_delta_base_cache(f);
1987 }
Shawn O. Pearce18bdec12007-03-19 01:14:37 -04001988
Linus Torvaldse5e01612007-03-17 12:44:06 -07001989 ent->p = p;
1990 ent->base_offset = base_offset;
1991 ent->type = type;
1992 ent->data = base;
1993 ent->size = base_size;
Nicolas Pitre5e08ecb2007-03-19 16:31:04 -04001994 ent->lru.next = &delta_base_cache_lru;
1995 ent->lru.prev = delta_base_cache_lru.prev;
1996 delta_base_cache_lru.prev->next = &ent->lru;
1997 delta_base_cache_lru.prev = &ent->lru;
Linus Torvalds62f255a2007-03-17 12:42:15 -07001998}
1999
Christian Couderc2c5b272009-01-12 18:42:24 +01002000static void *read_object(const unsigned char *sha1, enum object_type *type,
2001 unsigned long *size);
2002
Junio C Hamano5f443242011-07-06 19:08:55 -07002003static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2004{
2005 static FILE *log_file;
2006
Nguyễn Thái Ngọc Duyb12ca962013-06-09 12:22:48 +07002007 if (!log_pack_access)
2008 log_pack_access = getenv("GIT_TRACE_PACK_ACCESS");
2009 if (!log_pack_access)
2010 log_pack_access = no_log_pack_access;
2011 if (log_pack_access == no_log_pack_access)
2012 return;
2013
Junio C Hamano5f443242011-07-06 19:08:55 -07002014 if (!log_file) {
2015 log_file = fopen(log_pack_access, "w");
2016 if (!log_file) {
2017 error("cannot open pack access log '%s' for writing: %s",
2018 log_pack_access, strerror(errno));
Nguyễn Thái Ngọc Duyb12ca962013-06-09 12:22:48 +07002019 log_pack_access = no_log_pack_access;
Junio C Hamano5f443242011-07-06 19:08:55 -07002020 return;
2021 }
2022 }
2023 fprintf(log_file, "%s %"PRIuMAX"\n",
2024 p->pack_name, (uintmax_t)obj_offset);
2025 fflush(log_file);
2026}
2027
Nicolas Pitre0e8189e2008-10-31 11:31:08 -04002028int do_check_packed_object_crc;
2029
Thomas Rastabe601b2013-03-27 21:03:42 +01002030#define UNPACK_ENTRY_STACK_PREALLOC 64
2031struct unpack_entry_stack_ent {
2032 off_t obj_offset;
2033 off_t curpos;
2034 unsigned long size;
2035};
2036
Shawn O. Pearcec4001d92007-03-06 20:44:30 -05002037void *unpack_entry(struct packed_git *p, off_t obj_offset,
Thomas Rastabe601b2013-03-27 21:03:42 +01002038 enum object_type *final_type, unsigned long *final_size)
Junio C Hamanof3bf9222005-06-30 17:15:39 -07002039{
Shawn O. Pearce03e79c82006-12-23 02:34:08 -05002040 struct pack_window *w_curs = NULL;
Shawn O. Pearcec4001d92007-03-06 20:44:30 -05002041 off_t curpos = obj_offset;
Thomas Rastabe601b2013-03-27 21:03:42 +01002042 void *data = NULL;
2043 unsigned long size;
2044 enum object_type type;
2045 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2046 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2047 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2048 int base_from_cache = 0;
Junio C Hamano1f688552005-06-27 03:35:33 -07002049
Nguyễn Thái Ngọc Duyb12ca962013-06-09 12:22:48 +07002050 if (log_pack_access != no_log_pack_access)
Junio C Hamano5f443242011-07-06 19:08:55 -07002051 write_pack_access_log(p, obj_offset);
2052
Thomas Rastabe601b2013-03-27 21:03:42 +01002053 /* PHASE 1: drill down to the innermost base object */
2054 for (;;) {
2055 off_t base_offset;
2056 int i;
2057 struct delta_base_cache_entry *ent;
2058
2059 if (do_check_packed_object_crc && p->index_version > 1) {
2060 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2061 unsigned long len = revidx[1].offset - obj_offset;
2062 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2063 const unsigned char *sha1 =
2064 nth_packed_object_sha1(p, revidx->nr);
2065 error("bad packed object CRC for %s",
2066 sha1_to_hex(sha1));
2067 mark_bad_packed_object(p, sha1);
2068 unuse_pack(&w_curs);
2069 return NULL;
2070 }
Nicolas Pitre0e8189e2008-10-31 11:31:08 -04002071 }
Thomas Rastabe601b2013-03-27 21:03:42 +01002072
2073 ent = get_delta_base_cache_entry(p, curpos);
2074 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2075 type = ent->type;
2076 data = ent->data;
2077 size = ent->size;
2078 clear_delta_base_cache_entry(ent);
2079 base_from_cache = 1;
2080 break;
2081 }
2082
2083 type = unpack_object_header(p, &w_curs, &curpos, &size);
2084 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2085 break;
2086
2087 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2088 if (!base_offset) {
2089 error("failed to validate delta base reference "
2090 "at offset %"PRIuMAX" from %s",
2091 (uintmax_t)curpos, p->pack_name);
2092 /* bail to phase 2, in hopes of recovery */
2093 data = NULL;
2094 break;
2095 }
2096
2097 /* push object, proceed to base */
2098 if (delta_stack_nr >= delta_stack_alloc
2099 && delta_stack == small_delta_stack) {
2100 delta_stack_alloc = alloc_nr(delta_stack_nr);
2101 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2102 memcpy(delta_stack, small_delta_stack,
2103 sizeof(*delta_stack)*delta_stack_nr);
2104 } else {
2105 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2106 }
2107 i = delta_stack_nr++;
2108 delta_stack[i].obj_offset = obj_offset;
2109 delta_stack[i].curpos = curpos;
2110 delta_stack[i].size = size;
2111
2112 curpos = obj_offset = base_offset;
Nicolas Pitre0e8189e2008-10-31 11:31:08 -04002113 }
2114
Thomas Rastabe601b2013-03-27 21:03:42 +01002115 /* PHASE 2: handle the base */
2116 switch (type) {
Nicolas Pitreeb32d232006-09-21 00:06:49 -04002117 case OBJ_OFS_DELTA:
2118 case OBJ_REF_DELTA:
Thomas Rastabe601b2013-03-27 21:03:42 +01002119 if (data)
2120 die("BUG in unpack_entry: left loop at a valid delta");
Shawn O. Pearce4d703a12006-12-23 02:33:25 -05002121 break;
Linus Torvaldsa733cb62005-06-28 14:21:02 -07002122 case OBJ_COMMIT:
Linus Torvaldsa733cb62005-06-28 14:21:02 -07002123 case OBJ_TREE:
Linus Torvaldsa733cb62005-06-28 14:21:02 -07002124 case OBJ_BLOB:
Linus Torvaldsa733cb62005-06-28 14:21:02 -07002125 case OBJ_TAG:
Thomas Rastabe601b2013-03-27 21:03:42 +01002126 if (!base_from_cache)
2127 data = unpack_compressed_entry(p, &w_curs, curpos, size);
Shawn O. Pearce4d703a12006-12-23 02:33:25 -05002128 break;
Junio C Hamano1f688552005-06-27 03:35:33 -07002129 default:
Nicolas Pitre8eca0b42008-06-23 21:23:39 -04002130 data = NULL;
2131 error("unknown object type %i at offset %"PRIuMAX" in %s",
Thomas Rastabe601b2013-03-27 21:03:42 +01002132 type, (uintmax_t)obj_offset, p->pack_name);
Junio C Hamano1f688552005-06-27 03:35:33 -07002133 }
Thomas Rastabe601b2013-03-27 21:03:42 +01002134
2135 /* PHASE 3: apply deltas in order */
2136
2137 /* invariants:
2138 * 'data' holds the base data, or NULL if there was corruption
2139 */
2140 while (delta_stack_nr) {
2141 void *delta_data;
2142 void *base = data;
2143 unsigned long delta_size, base_size = size;
2144 int i;
2145
2146 data = NULL;
2147
2148 if (base)
2149 add_delta_base_cache(p, obj_offset, base, base_size, type);
2150
2151 if (!base) {
2152 /*
2153 * We're probably in deep shit, but let's try to fetch
2154 * the required base anyway from another pack or loose.
2155 * This is costly but should happen only in the presence
2156 * of a corrupted pack, and is better than failing outright.
2157 */
2158 struct revindex_entry *revidx;
2159 const unsigned char *base_sha1;
2160 revidx = find_pack_revindex(p, obj_offset);
2161 if (revidx) {
2162 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2163 error("failed to read delta base object %s"
2164 " at offset %"PRIuMAX" from %s",
2165 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2166 p->pack_name);
2167 mark_bad_packed_object(p, base_sha1);
2168 base = read_object(base_sha1, &type, &base_size);
2169 }
2170 }
2171
2172 i = --delta_stack_nr;
2173 obj_offset = delta_stack[i].obj_offset;
2174 curpos = delta_stack[i].curpos;
2175 delta_size = delta_stack[i].size;
2176
2177 if (!base)
2178 continue;
2179
2180 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2181
2182 if (!delta_data) {
2183 error("failed to unpack compressed delta "
2184 "at offset %"PRIuMAX" from %s",
2185 (uintmax_t)curpos, p->pack_name);
Thomas Rastabe601b2013-03-27 21:03:42 +01002186 data = NULL;
2187 continue;
2188 }
2189
2190 data = patch_delta(base, base_size,
2191 delta_data, delta_size,
2192 &size);
Jeff King1ee886c2013-06-14 17:53:34 -04002193
2194 /*
2195 * We could not apply the delta; warn the user, but keep going.
2196 * Our failure will be noticed either in the next iteration of
2197 * the loop, or if this is the final delta, in the caller when
2198 * we return NULL. Those code paths will take care of making
2199 * a more explicit warning and retrying with another copy of
2200 * the object.
2201 */
Thomas Rastabe601b2013-03-27 21:03:42 +01002202 if (!data)
Jeff King1ee886c2013-06-14 17:53:34 -04002203 error("failed to apply delta");
Thomas Rastabe601b2013-03-27 21:03:42 +01002204
Felipe Contreras4b8f7722013-05-30 08:56:21 -05002205 free(delta_data);
Thomas Rastabe601b2013-03-27 21:03:42 +01002206 }
2207
2208 *final_type = type;
2209 *final_size = size;
2210
Shawn O. Pearce03e79c82006-12-23 02:34:08 -05002211 unuse_pack(&w_curs);
Nicolas Pitre21666f12007-02-26 14:55:59 -05002212 return data;
Junio C Hamano1f688552005-06-27 03:35:33 -07002213}
2214
Shawn O. Pearced0798372007-05-26 01:24:19 -04002215const unsigned char *nth_packed_object_sha1(struct packed_git *p,
Nicolas Pitred72308e2007-04-04 16:49:04 -04002216 uint32_t n)
Junio C Hamano9a217f22005-06-28 14:56:57 -07002217{
Nicolas Pitre42873072007-03-16 16:42:50 -04002218 const unsigned char *index = p->index_data;
Shawn O. Pearced0798372007-05-26 01:24:19 -04002219 if (!index) {
2220 if (open_pack_index(p))
2221 return NULL;
2222 index = p->index_data;
2223 }
Nicolas Pitre57059092007-04-09 01:06:28 -04002224 if (n >= p->num_objects)
Nicolas Pitred72308e2007-04-04 16:49:04 -04002225 return NULL;
Nicolas Pitre74e34e12007-04-09 01:06:35 -04002226 index += 4 * 256;
2227 if (p->index_version == 1) {
2228 return index + 24 * n + 4;
2229 } else {
2230 index += 8;
2231 return index + 20 * n;
2232 }
2233}
2234
Nicolas Pitre99093232008-06-24 23:17:12 -04002235off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
Nicolas Pitre74e34e12007-04-09 01:06:35 -04002236{
2237 const unsigned char *index = p->index_data;
2238 index += 4 * 256;
2239 if (p->index_version == 1) {
2240 return ntohl(*((uint32_t *)(index + 24 * n)));
2241 } else {
2242 uint32_t off;
2243 index += 8 + p->num_objects * (20 + 4);
2244 off = ntohl(*((uint32_t *)(index + 4 * n)));
2245 if (!(off & 0x80000000))
2246 return off;
2247 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2248 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2249 ntohl(*((uint32_t *)(index + 4)));
2250 }
Junio C Hamano9a217f22005-06-28 14:56:57 -07002251}
2252
Shawn O. Pearcec4001d92007-03-06 20:44:30 -05002253off_t find_pack_entry_one(const unsigned char *sha1,
Nicolas Pitre43057302006-09-21 00:05:37 -04002254 struct packed_git *p)
Junio C Hamano1f688552005-06-27 03:35:33 -07002255{
Nicolas Pitre42873072007-03-16 16:42:50 -04002256 const uint32_t *level1_ofs = p->index_data;
Nicolas Pitre42873072007-03-16 16:42:50 -04002257 const unsigned char *index = p->index_data;
Junio C Hamano628522e2007-12-29 02:05:47 -08002258 unsigned hi, lo, stride;
2259 static int use_lookup = -1;
2260 static int debug_lookup = -1;
2261
2262 if (debug_lookup < 0)
2263 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
Nicolas Pitre42873072007-03-16 16:42:50 -04002264
Shawn O. Pearced0798372007-05-26 01:24:19 -04002265 if (!index) {
2266 if (open_pack_index(p))
2267 return 0;
2268 level1_ofs = p->index_data;
2269 index = p->index_data;
2270 }
Nicolas Pitre74e34e12007-04-09 01:06:35 -04002271 if (p->index_version > 1) {
2272 level1_ofs += 2;
2273 index += 8;
2274 }
Nicolas Pitre42873072007-03-16 16:42:50 -04002275 index += 4 * 256;
Nicolas Pitre74e34e12007-04-09 01:06:35 -04002276 hi = ntohl(level1_ofs[*sha1]);
2277 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
Junio C Hamano628522e2007-12-29 02:05:47 -08002278 if (p->index_version > 1) {
2279 stride = 20;
2280 } else {
2281 stride = 24;
2282 index += 4;
2283 }
2284
2285 if (debug_lookup)
Ramsay Jones6e1c2342008-07-03 16:52:09 +01002286 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
Junio C Hamano628522e2007-12-29 02:05:47 -08002287 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2288
2289 if (use_lookup < 0)
2290 use_lookup = !!getenv("GIT_USE_LOOKUP");
2291 if (use_lookup) {
2292 int pos = sha1_entry_pos(index, stride, 0,
2293 lo, hi, p->num_objects, sha1);
2294 if (pos < 0)
2295 return 0;
2296 return nth_packed_object_offset(p, pos);
2297 }
Junio C Hamano1f688552005-06-27 03:35:33 -07002298
2299 do {
Nicolas Pitre74e34e12007-04-09 01:06:35 -04002300 unsigned mi = (lo + hi) / 2;
Junio C Hamano628522e2007-12-29 02:05:47 -08002301 int cmp = hashcmp(index + mi * stride, sha1);
2302
2303 if (debug_lookup)
2304 printf("lo %u hi %u rg %u mi %u\n",
2305 lo, hi, hi - lo, mi);
Nicolas Pitre43057302006-09-21 00:05:37 -04002306 if (!cmp)
Nicolas Pitre74e34e12007-04-09 01:06:35 -04002307 return nth_packed_object_offset(p, mi);
Junio C Hamano1f688552005-06-27 03:35:33 -07002308 if (cmp > 0)
2309 hi = mi;
2310 else
2311 lo = mi+1;
2312 } while (lo < hi);
2313 return 0;
2314}
2315
Jeff King4c080182011-10-14 14:03:48 -04002316int is_pack_valid(struct packed_git *p)
Shawn O. Pearced131b7a2011-03-02 10:01:54 -08002317{
2318 /* An already open pack is known to be valid. */
2319 if (p->pack_fd != -1)
2320 return 1;
2321
2322 /* If the pack has one window completely covering the
2323 * file size, the pack is known to be valid even if
2324 * the descriptor is not currently open.
2325 */
2326 if (p->windows) {
2327 struct pack_window *w = p->windows;
2328
2329 if (!w->offset && w->len == p->pack_size)
2330 return 1;
2331 }
2332
2333 /* Force the pack to open to prove its valid. */
2334 return !open_packed_git(p);
2335}
2336
Nguyễn Thái Ngọc Duy95099732012-02-01 20:48:54 +07002337static int fill_pack_entry(const unsigned char *sha1,
2338 struct pack_entry *e,
2339 struct packed_git *p)
2340{
2341 off_t offset;
2342
2343 if (p->num_bad_objects) {
2344 unsigned i;
2345 for (i = 0; i < p->num_bad_objects; i++)
2346 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2347 return 0;
2348 }
2349
2350 offset = find_pack_entry_one(sha1, p);
2351 if (!offset)
2352 return 0;
2353
2354 /*
2355 * We are about to tell the caller where they can locate the
2356 * requested object. We better make sure the packfile is
2357 * still here and can be accessed before supplying that
2358 * answer, as it may have been deleted since the index was
2359 * loaded!
2360 */
2361 if (!is_pack_valid(p)) {
2362 warning("packfile %s cannot be accessed", p->pack_name);
2363 return 0;
2364 }
2365 e->offset = offset;
2366 e->p = p;
2367 hashcpy(e->sha1, sha1);
2368 return 1;
2369}
2370
Brandon Casey4d6acb72009-03-19 22:47:54 -05002371static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
Junio C Hamano1f688552005-06-27 03:35:33 -07002372{
2373 struct packed_git *p;
Nicolas Pitre43057302006-09-21 00:05:37 -04002374
Junio C Hamano1f688552005-06-27 03:35:33 -07002375 prepare_packed_git();
Nicolas Pitref7c22cc2007-05-30 22:48:13 -04002376 if (!packed_git)
2377 return 0;
Junio C Hamano1f688552005-06-27 03:35:33 -07002378
Nguyễn Thái Ngọc Duyc01f51c2012-02-01 20:48:55 +07002379 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2380 return 1;
Nicolas Pitre8eca0b42008-06-23 21:23:39 -04002381
Nguyễn Thái Ngọc Duyc01f51c2012-02-01 20:48:55 +07002382 for (p = packed_git; p; p = p->next) {
2383 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2384 continue;
Nicolas Pitref7c22cc2007-05-30 22:48:13 -04002385
Nguyễn Thái Ngọc Duyc01f51c2012-02-01 20:48:55 +07002386 last_found_pack = p;
2387 return 1;
2388 }
Junio C Hamano1f688552005-06-27 03:35:33 -07002389 return 0;
2390}
2391
Junio C Hamanoa6080a02007-06-07 00:04:01 -07002392struct packed_git *find_sha1_pack(const unsigned char *sha1,
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -04002393 struct packed_git *packs)
2394{
2395 struct packed_git *p;
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -04002396
2397 for (p = packs; p; p = p->next) {
Nicolas Pitre43057302006-09-21 00:05:37 -04002398 if (find_pack_entry_one(sha1, p))
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -04002399 return p;
2400 }
2401 return NULL;
Nicolas Pitre21666f12007-02-26 14:55:59 -05002402
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -04002403}
2404
Jeff King052fe5e2013-07-12 02:30:48 -04002405static int sha1_loose_object_info(const unsigned char *sha1,
Jeff King23c339c2013-07-12 02:37:53 -04002406 struct object_info *oi)
Junio C Hamano65c2e0c2005-06-02 15:20:54 -07002407{
Junio C Hamano36e4d742005-06-27 03:34:06 -07002408 int status;
Junio C Hamano65c2e0c2005-06-02 15:20:54 -07002409 unsigned long mapsize, size;
2410 void *map;
Junio C Hamanoef49a7a2011-06-10 11:52:15 -07002411 git_zstream stream;
Nicolas Pitred65a16f2007-02-26 14:55:55 -05002412 char hdr[32];
Junio C Hamano65c2e0c2005-06-02 15:20:54 -07002413
Jeff King052fe5e2013-07-12 02:30:48 -04002414 /*
2415 * If we don't care about type or size, then we don't
2416 * need to look inside the object at all.
2417 */
Jeff King23c339c2013-07-12 02:37:53 -04002418 if (!oi->typep && !oi->sizep) {
2419 if (oi->disk_sizep) {
Jeff King052fe5e2013-07-12 02:30:48 -04002420 struct stat st;
2421 if (stat_sha1_file(sha1, &st) < 0)
2422 return -1;
Jeff King23c339c2013-07-12 02:37:53 -04002423 *oi->disk_sizep = st.st_size;
Jeff King052fe5e2013-07-12 02:30:48 -04002424 }
2425 return 0;
2426 }
2427
Junio C Hamanobb6b8e42006-07-17 15:04:47 -07002428 map = map_sha1_file(sha1, &mapsize);
Johannes Schindelinf0df4ed2006-11-28 00:18:55 +01002429 if (!map)
Thomas Rastdbea72a2013-05-30 22:00:22 +02002430 return -1;
Jeff King23c339c2013-07-12 02:37:53 -04002431 if (oi->disk_sizep)
2432 *oi->disk_sizep = mapsize;
Junio C Hamano36e4d742005-06-27 03:34:06 -07002433 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2434 status = error("unable to unpack %s header",
2435 sha1_to_hex(sha1));
Nicolas Pitre21666f12007-02-26 14:55:59 -05002436 else if ((status = parse_sha1_header(hdr, &size)) < 0)
Junio C Hamano36e4d742005-06-27 03:34:06 -07002437 status = error("unable to parse %s header", sha1_to_hex(sha1));
Jeff King23c339c2013-07-12 02:37:53 -04002438 else if (oi->sizep)
2439 *oi->sizep = size;
Linus Torvalds39c68542009-01-07 19:54:47 -08002440 git_inflate_end(&stream);
Junio C Hamano65c2e0c2005-06-02 15:20:54 -07002441 munmap(map, mapsize);
Jeff King23c339c2013-07-12 02:37:53 -04002442 if (oi->typep)
2443 *oi->typep = status;
Jeff King052fe5e2013-07-12 02:30:48 -04002444 return 0;
Junio C Hamano65c2e0c2005-06-02 15:20:54 -07002445}
2446
Junio C Hamano9a490592011-05-12 15:51:38 -07002447int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
Johannes Schindelinf0df4ed2006-11-28 00:18:55 +01002448{
Nguyễn Thái Ngọc Duyc4d99862011-02-05 21:03:02 +07002449 struct cached_object *co;
Johannes Schindelinf0df4ed2006-11-28 00:18:55 +01002450 struct pack_entry e;
Jeff King5b086402013-07-12 02:34:57 -04002451 int rtype;
Johannes Schindelinf0df4ed2006-11-28 00:18:55 +01002452
Nguyễn Thái Ngọc Duyc4d99862011-02-05 21:03:02 +07002453 co = find_cached_object(sha1);
2454 if (co) {
Jeff King5b086402013-07-12 02:34:57 -04002455 if (oi->typep)
2456 *(oi->typep) = co->type;
Junio C Hamano9a490592011-05-12 15:51:38 -07002457 if (oi->sizep)
2458 *(oi->sizep) = co->size;
Jeff King161f00e2013-07-07 06:04:00 -04002459 if (oi->disk_sizep)
2460 *(oi->disk_sizep) = 0;
Junio C Hamano9a490592011-05-12 15:51:38 -07002461 oi->whence = OI_CACHED;
Jeff King5b086402013-07-12 02:34:57 -04002462 return 0;
Nguyễn Thái Ngọc Duyc4d99862011-02-05 21:03:02 +07002463 }
2464
Junio C Hamanocd673c12009-02-27 23:15:53 -08002465 if (!find_pack_entry(sha1, &e)) {
Steven Grimmddd63e62008-08-05 13:08:41 -07002466 /* Most likely it's a loose object. */
Jeff King23c339c2013-07-12 02:37:53 -04002467 if (!sha1_loose_object_info(sha1, oi)) {
Junio C Hamano9a490592011-05-12 15:51:38 -07002468 oi->whence = OI_LOOSE;
Jeff King5b086402013-07-12 02:34:57 -04002469 return 0;
Junio C Hamano9a490592011-05-12 15:51:38 -07002470 }
Steven Grimmddd63e62008-08-05 13:08:41 -07002471
2472 /* Not a loose object; someone else may have just packed it. */
Johannes Schindelinf0df4ed2006-11-28 00:18:55 +01002473 reprepare_packed_git();
Junio C Hamanocd673c12009-02-27 23:15:53 -08002474 if (!find_pack_entry(sha1, &e))
Jeff King052fe5e2013-07-12 02:30:48 -04002475 return -1;
Johannes Schindelinf0df4ed2006-11-28 00:18:55 +01002476 }
Nicolas Pitre3d77d872008-10-29 19:02:47 -04002477
Jeff King23c339c2013-07-12 02:37:53 -04002478 rtype = packed_object_info(e.p, e.offset, oi);
Jeff King412916e2013-07-12 02:32:25 -04002479 if (rtype < 0) {
Nicolas Pitre3d77d872008-10-29 19:02:47 -04002480 mark_bad_packed_object(e.p, sha1);
Jeff King412916e2013-07-12 02:32:25 -04002481 return sha1_object_info_extended(sha1, oi);
Junio C Hamano5266d362011-05-13 13:20:43 -07002482 } else if (in_delta_base_cache(e.p, e.offset)) {
2483 oi->whence = OI_DBCACHED;
Junio C Hamano9a490592011-05-12 15:51:38 -07002484 } else {
2485 oi->whence = OI_PACKED;
2486 oi->u.packed.offset = e.offset;
2487 oi->u.packed.pack = e.p;
2488 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2489 rtype == OBJ_OFS_DELTA);
Nicolas Pitre3d77d872008-10-29 19:02:47 -04002490 }
2491
Jeff King5b086402013-07-12 02:34:57 -04002492 return 0;
Johannes Schindelinf0df4ed2006-11-28 00:18:55 +01002493}
2494
Christian Couder3fc0dca2013-10-27 00:34:30 +02002495/* returns enum object_type or negative */
Junio C Hamano9a490592011-05-12 15:51:38 -07002496int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2497{
Jeff King5b086402013-07-12 02:34:57 -04002498 enum object_type type;
Ramsay Jonesd099b712013-07-18 21:25:50 +01002499 struct object_info oi = {NULL};
Junio C Hamano9a490592011-05-12 15:51:38 -07002500
Jeff King5b086402013-07-12 02:34:57 -04002501 oi.typep = &type;
Junio C Hamano9a490592011-05-12 15:51:38 -07002502 oi.sizep = sizep;
Jeff King5b086402013-07-12 02:34:57 -04002503 if (sha1_object_info_extended(sha1, &oi) < 0)
2504 return -1;
2505 return type;
Junio C Hamano9a490592011-05-12 15:51:38 -07002506}
2507
Nicolas Pitre21666f12007-02-26 14:55:59 -05002508static void *read_packed_sha1(const unsigned char *sha1,
2509 enum object_type *type, unsigned long *size)
Junio C Hamano1f688552005-06-27 03:35:33 -07002510{
2511 struct pack_entry e;
Nicolas Pitre8eca0b42008-06-23 21:23:39 -04002512 void *data;
Junio C Hamano1f688552005-06-27 03:35:33 -07002513
Junio C Hamanocd673c12009-02-27 23:15:53 -08002514 if (!find_pack_entry(sha1, &e))
Junio C Hamano1f688552005-06-27 03:35:33 -07002515 return NULL;
Nicolas Pitre8eca0b42008-06-23 21:23:39 -04002516 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2517 if (!data) {
2518 /*
2519 * We're probably in deep shit, but let's try to fetch
2520 * the required object anyway from another pack or loose.
2521 * This should happen only in the presence of a corrupted
2522 * pack, and is better than failing outright.
2523 */
2524 error("failed to read object %s at offset %"PRIuMAX" from %s",
2525 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2526 mark_bad_packed_object(e.p, sha1);
Nicolas Pitreac939102008-07-14 21:46:48 -04002527 data = read_object(sha1, type, size);
Nicolas Pitre8eca0b42008-06-23 21:23:39 -04002528 }
2529 return data;
Junio C Hamano1f688552005-06-27 03:35:33 -07002530}
2531
Nicolas Pitre21666f12007-02-26 14:55:59 -05002532int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2533 unsigned char *sha1)
Junio C Hamanod66b37b2007-02-04 21:42:38 -08002534{
2535 struct cached_object *co;
2536
Nicolas Pitre21666f12007-02-26 14:55:59 -05002537 hash_sha1_file(buf, len, typename(type), sha1);
Junio C Hamanod66b37b2007-02-04 21:42:38 -08002538 if (has_sha1_file(sha1) || find_cached_object(sha1))
2539 return 0;
2540 if (cached_object_alloc <= cached_object_nr) {
2541 cached_object_alloc = alloc_nr(cached_object_alloc);
2542 cached_objects = xrealloc(cached_objects,
2543 sizeof(*cached_objects) *
2544 cached_object_alloc);
2545 }
2546 co = &cached_objects[cached_object_nr++];
2547 co->size = len;
Nicolas Pitre21666f12007-02-26 14:55:59 -05002548 co->type = type;
Junio C Hamanoefa13f72007-02-15 17:02:06 -08002549 co->buf = xmalloc(len);
2550 memcpy(co->buf, buf, len);
Junio C Hamanod66b37b2007-02-04 21:42:38 -08002551 hashcpy(co->sha1, sha1);
2552 return 0;
2553}
2554
Christian Couderc2c5b272009-01-12 18:42:24 +01002555static void *read_object(const unsigned char *sha1, enum object_type *type,
2556 unsigned long *size)
Linus Torvalds0fcfd162005-04-18 13:04:43 -07002557{
2558 unsigned long mapsize;
2559 void *map, *buf;
Junio C Hamanod66b37b2007-02-04 21:42:38 -08002560 struct cached_object *co;
2561
2562 co = find_cached_object(sha1);
2563 if (co) {
Nicolas Pitre21666f12007-02-26 14:55:59 -05002564 *type = co->type;
Junio C Hamanod66b37b2007-02-04 21:42:38 -08002565 *size = co->size;
Pierre Habouzit182af832007-09-16 00:32:36 +02002566 return xmemdupz(co->buf, co->size);
Junio C Hamanod66b37b2007-02-04 21:42:38 -08002567 }
Linus Torvalds0fcfd162005-04-18 13:04:43 -07002568
Peter Eriksen8276c002007-01-22 21:29:45 +01002569 buf = read_packed_sha1(sha1, type, size);
2570 if (buf)
2571 return buf;
Junio C Hamanobb6b8e42006-07-17 15:04:47 -07002572 map = map_sha1_file(sha1, &mapsize);
Linus Torvalds0fcfd162005-04-18 13:04:43 -07002573 if (map) {
Junio C Hamano7efbff72007-03-05 00:21:37 -08002574 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
Linus Torvalds0fcfd162005-04-18 13:04:43 -07002575 munmap(map, mapsize);
2576 return buf;
2577 }
Jeff King637cdd92006-06-02 11:32:23 -04002578 reprepare_packed_git();
Peter Eriksen8276c002007-01-22 21:29:45 +01002579 return read_packed_sha1(sha1, type, size);
Linus Torvalds0fcfd162005-04-18 13:04:43 -07002580}
2581
Junio C Hamanob6c4cec2010-10-28 11:13:06 -07002582/*
2583 * This function dies on corrupt objects; the callers who want to
2584 * deal with them should arrange to call read_object() and give error
2585 * messages themselves.
2586 */
Junio C Hamano5bf29b92011-05-15 12:54:54 -07002587void *read_sha1_file_extended(const unsigned char *sha1,
2588 enum object_type *type,
2589 unsigned long *size,
2590 unsigned flag)
Nicolas Pitreac939102008-07-14 21:46:48 -04002591{
Junio C Hamano3ba7a062010-10-28 11:13:06 -07002592 void *data;
Ævar Arnfjörð Bjarmasone8b15e62010-06-10 12:47:01 +00002593 char *path;
Junio C Hamanob6c4cec2010-10-28 11:13:06 -07002594 const struct packed_git *p;
Junio C Hamano5bf29b92011-05-15 12:54:54 -07002595 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2596 ? lookup_replace_object(sha1) : sha1;
Junio C Hamanob6c4cec2010-10-28 11:13:06 -07002597
Junio C Hamano3ba7a062010-10-28 11:13:06 -07002598 errno = 0;
2599 data = read_object(repl, type, size);
Junio C Hamano4bbf5a22011-05-15 12:54:52 -07002600 if (data)
Junio C Hamanob6c4cec2010-10-28 11:13:06 -07002601 return data;
Christian Couder68095572009-01-23 10:06:53 +01002602
Björn Steinbrink25f3af32011-01-20 21:12:20 +01002603 if (errno && errno != ENOENT)
Junio C Hamano3ba7a062010-10-28 11:13:06 -07002604 die_errno("failed to read object %s", sha1_to_hex(sha1));
2605
Christian Couder68095572009-01-23 10:06:53 +01002606 /* die if we replaced an object with one that does not exist */
Junio C Hamanob6c4cec2010-10-28 11:13:06 -07002607 if (repl != sha1)
Christian Couder68095572009-01-23 10:06:53 +01002608 die("replacement %s not found for %s",
2609 sha1_to_hex(repl), sha1_to_hex(sha1));
2610
Junio C Hamanob6c4cec2010-10-28 11:13:06 -07002611 if (has_loose_object(repl)) {
2612 path = sha1_file_name(sha1);
2613 die("loose object %s (stored in %s) is corrupt",
2614 sha1_to_hex(repl), path);
Ævar Arnfjörð Bjarmasone8b15e62010-06-10 12:47:01 +00002615 }
Christian Couder68095572009-01-23 10:06:53 +01002616
Junio C Hamanob6c4cec2010-10-28 11:13:06 -07002617 if ((p = has_packed_and_bad(repl)) != NULL)
2618 die("packed object %s (stored in %s) is corrupt",
2619 sha1_to_hex(repl), p->pack_name);
Christian Couderf5552ae2009-01-23 10:07:01 +01002620
Junio C Hamanob6c4cec2010-10-28 11:13:06 -07002621 return NULL;
Nicolas Pitreac939102008-07-14 21:46:48 -04002622}
2623
Junio C Hamano40469ee2005-04-28 16:42:27 -07002624void *read_object_with_reference(const unsigned char *sha1,
Nicolas Pitre21666f12007-02-26 14:55:59 -05002625 const char *required_type_name,
Junio C Hamano40469ee2005-04-28 16:42:27 -07002626 unsigned long *size,
2627 unsigned char *actual_sha1_return)
Junio C Hamanof4913f92005-04-20 18:06:49 -07002628{
Nicolas Pitre21666f12007-02-26 14:55:59 -05002629 enum object_type type, required_type;
Junio C Hamanof4913f92005-04-20 18:06:49 -07002630 void *buffer;
2631 unsigned long isize;
Junio C Hamano40469ee2005-04-28 16:42:27 -07002632 unsigned char actual_sha1[20];
Junio C Hamanof4913f92005-04-20 18:06:49 -07002633
Nicolas Pitre21666f12007-02-26 14:55:59 -05002634 required_type = type_from_string(required_type_name);
Shawn Pearcee7024962006-08-23 02:49:00 -04002635 hashcpy(actual_sha1, sha1);
Junio C Hamano40469ee2005-04-28 16:42:27 -07002636 while (1) {
2637 int ref_length = -1;
2638 const char *ref_type = NULL;
Junio C Hamanof4913f92005-04-20 18:06:49 -07002639
Nicolas Pitre21666f12007-02-26 14:55:59 -05002640 buffer = read_sha1_file(actual_sha1, &type, &isize);
Junio C Hamano40469ee2005-04-28 16:42:27 -07002641 if (!buffer)
2642 return NULL;
Nicolas Pitre21666f12007-02-26 14:55:59 -05002643 if (type == required_type) {
Junio C Hamano40469ee2005-04-28 16:42:27 -07002644 *size = isize;
2645 if (actual_sha1_return)
Shawn Pearcee7024962006-08-23 02:49:00 -04002646 hashcpy(actual_sha1_return, actual_sha1);
Junio C Hamano40469ee2005-04-28 16:42:27 -07002647 return buffer;
2648 }
2649 /* Handle references */
Nicolas Pitre21666f12007-02-26 14:55:59 -05002650 else if (type == OBJ_COMMIT)
Junio C Hamano40469ee2005-04-28 16:42:27 -07002651 ref_type = "tree ";
Nicolas Pitre21666f12007-02-26 14:55:59 -05002652 else if (type == OBJ_TAG)
Junio C Hamano40469ee2005-04-28 16:42:27 -07002653 ref_type = "object ";
2654 else {
2655 free(buffer);
2656 return NULL;
2657 }
2658 ref_length = strlen(ref_type);
2659
Martin Koegler50974ec2008-02-18 21:47:52 +01002660 if (ref_length + 40 > isize ||
2661 memcmp(buffer, ref_type, ref_length) ||
Florian Forster1d7f1712006-06-18 17:18:09 +02002662 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
Junio C Hamano40469ee2005-04-28 16:42:27 -07002663 free(buffer);
2664 return NULL;
2665 }
Sergey Vlasov1cf58e72005-08-08 22:44:43 +04002666 free(buffer);
Junio C Hamano40469ee2005-04-28 16:42:27 -07002667 /* Now we have the ID of the referred-to object in
2668 * actual_sha1. Check again. */
Junio C Hamanof4913f92005-04-20 18:06:49 -07002669 }
Junio C Hamanof4913f92005-04-20 18:06:49 -07002670}
2671
Nicolas Pitrece9fbf12007-03-20 16:02:09 -04002672static void write_sha1_file_prepare(const void *buf, unsigned long len,
Rene Scharfe972a9152006-10-15 14:02:03 +02002673 const char *type, unsigned char *sha1,
Nicolas Pitred65a16f2007-02-26 14:55:55 -05002674 char *hdr, int *hdrlen)
Junio C Hamanod410c0f2005-06-27 19:03:13 -07002675{
Nicolas Pitre9126f002008-10-01 14:05:20 -04002676 git_SHA_CTX c;
Junio C Hamanod410c0f2005-06-27 19:03:13 -07002677
2678 /* Generate the header */
Nicolas Pitred65a16f2007-02-26 14:55:55 -05002679 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
Junio C Hamanod410c0f2005-06-27 19:03:13 -07002680
2681 /* Sha1.. */
Nicolas Pitre9126f002008-10-01 14:05:20 -04002682 git_SHA1_Init(&c);
2683 git_SHA1_Update(&c, hdr, *hdrlen);
2684 git_SHA1_Update(&c, buf, len);
2685 git_SHA1_Final(sha1, &c);
Junio C Hamanod410c0f2005-06-27 19:03:13 -07002686}
2687
Linus Torvalds230f1322005-10-08 15:54:01 -07002688/*
Junio C Hamano5a688fe2009-03-25 16:19:36 -07002689 * Move the just written object into its final resting place.
2690 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2691 * "moving" is only a part of what it does, when no patch between
2692 * master to pu changes the call sites of this function.
Linus Torvalds230f1322005-10-08 15:54:01 -07002693 */
Junio C Hamano839837b2006-09-01 00:17:47 -07002694int move_temp_to_file(const char *tmpfile, const char *filename)
Linus Torvalds230f1322005-10-08 15:54:01 -07002695{
Thomas Raste32c0a92008-09-19 00:24:46 +02002696 int ret = 0;
Junio C Hamano5a688fe2009-03-25 16:19:36 -07002697
Johannes Schindelin348df162009-04-28 00:32:25 +02002698 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
Johannes Schindelinbe66a6c2009-04-25 11:57:14 +02002699 goto try_rename;
2700 else if (link(tmpfile, filename))
Thomas Raste32c0a92008-09-19 00:24:46 +02002701 ret = errno;
Linus Torvalds7ebb6fc2005-10-26 10:27:36 -07002702
2703 /*
2704 * Coda hack - coda doesn't like cross-directory links,
2705 * so we fall back to a rename, which will mean that it
2706 * won't be able to check collisions, but that's not a
2707 * big deal.
2708 *
2709 * The same holds for FAT formatted media.
2710 *
Junio C Hamano3be1f182009-03-27 23:14:39 -07002711 * When this succeeds, we just return. We have nothing
Linus Torvalds7ebb6fc2005-10-26 10:27:36 -07002712 * left to unlink.
2713 */
2714 if (ret && ret != EEXIST) {
Johannes Schindelinbe66a6c2009-04-25 11:57:14 +02002715 try_rename:
Linus Torvalds7ebb6fc2005-10-26 10:27:36 -07002716 if (!rename(tmpfile, filename))
Junio C Hamano3be1f182009-03-27 23:14:39 -07002717 goto out;
Johannes Schindelin9e48b382005-10-26 01:41:20 +02002718 ret = errno;
Linus Torvalds230f1322005-10-08 15:54:01 -07002719 }
Alex Riesen691f1a22009-04-29 23:22:56 +02002720 unlink_or_warn(tmpfile);
Linus Torvalds230f1322005-10-08 15:54:01 -07002721 if (ret) {
2722 if (ret != EEXIST) {
Pete Wyckoff82247e92012-04-29 20:28:45 -04002723 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
Linus Torvalds230f1322005-10-08 15:54:01 -07002724 }
2725 /* FIXME!!! Collision check here ? */
2726 }
2727
Junio C Hamano3be1f182009-03-27 23:14:39 -07002728out:
Matthieu Moy5256b002010-02-22 23:32:16 +01002729 if (adjust_shared_perm(filename))
Junio C Hamano5a688fe2009-03-25 16:19:36 -07002730 return error("unable to set permission to '%s'", filename);
Linus Torvalds230f1322005-10-08 15:54:01 -07002731 return 0;
2732}
2733
Linus Torvalds4d548152006-05-24 08:30:54 -07002734static int write_buffer(int fd, const void *buf, size_t len)
2735{
Linus Torvaldsd34cf192007-01-11 20:23:00 -08002736 if (write_in_full(fd, buf, len) < 0)
Andy Whitcroft93822c22007-01-08 15:58:23 +00002737 return error("file write error (%s)", strerror(errno));
Linus Torvalds4d548152006-05-24 08:30:54 -07002738 return 0;
2739}
2740
Nicolas Pitrece9fbf12007-03-20 16:02:09 -04002741int hash_sha1_file(const void *buf, unsigned long len, const char *type,
Rene Scharfeabdc3fc2006-10-14 12:45:36 +02002742 unsigned char *sha1)
2743{
Nicolas Pitred65a16f2007-02-26 14:55:55 -05002744 char hdr[32];
Rene Scharfeabdc3fc2006-10-14 12:45:36 +02002745 int hdrlen;
2746 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2747 return 0;
2748}
2749
Linus Torvaldse9039dd2008-06-10 18:47:18 -07002750/* Finalize a file on disk, and close it. */
2751static void close_sha1_file(int fd)
2752{
Linus Torvaldsaafe9fb2008-06-18 15:18:44 -07002753 if (fsync_object_files)
2754 fsync_or_die(fd, "sha1 file");
Linus Torvaldse9039dd2008-06-10 18:47:18 -07002755 if (close(fd) != 0)
Thomas Rastd824cbb2009-06-27 17:58:46 +02002756 die_errno("error when closing sha1 file");
Linus Torvaldse9039dd2008-06-10 18:47:18 -07002757}
2758
Linus Torvalds5723fe72008-06-14 10:50:12 -07002759/* Size of directory component, including the ending '/' */
2760static inline int directory_size(const char *filename)
2761{
2762 const char *s = strrchr(filename, '/');
2763 if (!s)
2764 return 0;
2765 return s - filename + 1;
2766}
2767
2768/*
2769 * This creates a temporary file in the same directory as the final
2770 * 'filename'
2771 *
2772 * We want to avoid cross-directory filename renames, because those
2773 * can have problems on various filesystems (FAT, NFS, Coda).
2774 */
2775static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2776{
2777 int fd, dirlen = directory_size(filename);
2778
2779 if (dirlen + 20 > bufsiz) {
2780 errno = ENAMETOOLONG;
2781 return -1;
2782 }
2783 memcpy(buffer, filename, dirlen);
2784 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
Matthieu Moy5256b002010-02-22 23:32:16 +01002785 fd = git_mkstemp_mode(buffer, 0444);
Joey Hesscbacbf42008-11-20 13:56:28 -05002786 if (fd < 0 && dirlen && errno == ENOENT) {
Linus Torvalds5723fe72008-06-14 10:50:12 -07002787 /* Make sure the directory exists */
Patrick Higgins6ff6af62008-06-23 15:33:41 -06002788 memcpy(buffer, filename, dirlen);
Linus Torvalds5723fe72008-06-14 10:50:12 -07002789 buffer[dirlen-1] = 0;
Junio C Hamano3bfaf012008-06-16 22:02:12 -07002790 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
Linus Torvalds5723fe72008-06-14 10:50:12 -07002791 return -1;
2792
2793 /* Try again */
2794 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
Matthieu Moy5256b002010-02-22 23:32:16 +01002795 fd = git_mkstemp_mode(buffer, 0444);
Linus Torvalds5723fe72008-06-14 10:50:12 -07002796 }
2797 return fd;
2798}
2799
Nicolas Pitrebbac7312008-05-14 01:32:48 -04002800static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
Jeff Kingc00e6572010-04-01 20:03:18 -04002801 const void *buf, unsigned long len, time_t mtime)
Linus Torvalds0fcfd162005-04-18 13:04:43 -07002802{
Jeff King915308b2009-01-29 00:56:34 -05002803 int fd, ret;
Nicolas Pitre9892beb2010-02-20 23:27:31 -05002804 unsigned char compressed[4096];
Junio C Hamanoef49a7a2011-06-10 11:52:15 -07002805 git_zstream stream;
Nicolas Pitre748af442010-02-21 15:48:06 -05002806 git_SHA_CTX c;
2807 unsigned char parano_sha1[20];
Linus Torvalds706bc532005-04-20 09:28:05 -07002808 char *filename;
Ævar Arnfjörð Bjarmasonab1900a2011-12-21 01:18:21 +00002809 static char tmp_file[PATH_MAX];
Linus Torvaldsa44c9a52005-04-25 10:19:53 -07002810
Rene Scharfe972a9152006-10-15 14:02:03 +02002811 filename = sha1_file_name(sha1);
Ævar Arnfjörð Bjarmasonab1900a2011-12-21 01:18:21 +00002812 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
Linus Torvaldsaac17942005-05-03 11:46:16 -07002813 if (fd < 0) {
Sam Vilain35243572008-11-14 20:19:34 +13002814 if (errno == EACCES)
Pete Wyckoff82247e92012-04-29 20:28:45 -04002815 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
Petr Baudis916d0812006-11-09 13:52:05 +01002816 else
Pete Wyckoff5eaeda72012-04-29 20:28:46 -04002817 return error("unable to create temporary file: %s", strerror(errno));
Linus Torvaldsaac17942005-05-03 11:46:16 -07002818 }
2819
Linus Torvalds0fcfd162005-04-18 13:04:43 -07002820 /* Set it up */
2821 memset(&stream, 0, sizeof(stream));
Junio C Hamano55bb5c92011-06-10 10:55:10 -07002822 git_deflate_init(&stream, zlib_compression_level);
Linus Torvalds0fcfd162005-04-18 13:04:43 -07002823 stream.next_out = compressed;
Nicolas Pitre9892beb2010-02-20 23:27:31 -05002824 stream.avail_out = sizeof(compressed);
Nicolas Pitre748af442010-02-21 15:48:06 -05002825 git_SHA1_Init(&c);
Linus Torvaldsa44c9a52005-04-25 10:19:53 -07002826
2827 /* First header.. */
Nicolas Pitred65a16f2007-02-26 14:55:55 -05002828 stream.next_in = (unsigned char *)hdr;
Linus Torvaldsa44c9a52005-04-25 10:19:53 -07002829 stream.avail_in = hdrlen;
Junio C Hamano55bb5c92011-06-10 10:55:10 -07002830 while (git_deflate(&stream, 0) == Z_OK)
2831 ; /* nothing */
Nicolas Pitre748af442010-02-21 15:48:06 -05002832 git_SHA1_Update(&c, hdr, hdrlen);
Linus Torvaldsa44c9a52005-04-25 10:19:53 -07002833
2834 /* Then the data itself.. */
Jeff Kingc00e6572010-04-01 20:03:18 -04002835 stream.next_in = (void *)buf;
Linus Torvaldsa44c9a52005-04-25 10:19:53 -07002836 stream.avail_in = len;
Nicolas Pitre9892beb2010-02-20 23:27:31 -05002837 do {
Nicolas Pitre748af442010-02-21 15:48:06 -05002838 unsigned char *in0 = stream.next_in;
Junio C Hamano55bb5c92011-06-10 10:55:10 -07002839 ret = git_deflate(&stream, Z_FINISH);
Nicolas Pitre748af442010-02-21 15:48:06 -05002840 git_SHA1_Update(&c, in0, stream.next_in - in0);
Nicolas Pitre9892beb2010-02-20 23:27:31 -05002841 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2842 die("unable to write sha1 file");
2843 stream.next_out = compressed;
2844 stream.avail_out = sizeof(compressed);
2845 } while (ret == Z_OK);
2846
Linus Torvaldsac54c272007-03-20 11:38:34 -07002847 if (ret != Z_STREAM_END)
2848 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
Junio C Hamano55bb5c92011-06-10 10:55:10 -07002849 ret = git_deflate_end_gently(&stream);
Linus Torvaldsac54c272007-03-20 11:38:34 -07002850 if (ret != Z_OK)
2851 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
Nicolas Pitre748af442010-02-21 15:48:06 -05002852 git_SHA1_Final(parano_sha1, &c);
2853 if (hashcmp(sha1, parano_sha1) != 0)
2854 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
Linus Torvaldsac54c272007-03-20 11:38:34 -07002855
Linus Torvaldse9039dd2008-06-10 18:47:18 -07002856 close_sha1_file(fd);
Linus Torvalds0fcfd162005-04-18 13:04:43 -07002857
Nicolas Pitrebbac7312008-05-14 01:32:48 -04002858 if (mtime) {
2859 struct utimbuf utb;
2860 utb.actime = mtime;
2861 utb.modtime = mtime;
Ævar Arnfjörð Bjarmasonab1900a2011-12-21 01:18:21 +00002862 if (utime(tmp_file, &utb) < 0)
Nicolas Pitrebbac7312008-05-14 01:32:48 -04002863 warning("failed utime() on %s: %s",
Ævar Arnfjörð Bjarmasonab1900a2011-12-21 01:18:21 +00002864 tmp_file, strerror(errno));
Nicolas Pitrebbac7312008-05-14 01:32:48 -04002865 }
2866
Ævar Arnfjörð Bjarmasonab1900a2011-12-21 01:18:21 +00002867 return move_temp_to_file(tmp_file, filename);
Linus Torvalds0fcfd162005-04-18 13:04:43 -07002868}
Daniel Barkalow8237b182005-04-23 18:47:23 -07002869
Jeff Kingc00e6572010-04-01 20:03:18 -04002870int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
Nicolas Pitrebbac7312008-05-14 01:32:48 -04002871{
2872 unsigned char sha1[20];
2873 char hdr[32];
2874 int hdrlen;
2875
2876 /* Normally if we have it in the pack then we do not bother writing
2877 * it out into .git/objects/??/?{38} file.
2878 */
2879 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2880 if (returnsha1)
2881 hashcpy(returnsha1, sha1);
2882 if (has_sha1_file(sha1))
2883 return 0;
2884 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2885}
2886
2887int force_object_loose(const unsigned char *sha1, time_t mtime)
2888{
Nicolas Pitrebbac7312008-05-14 01:32:48 -04002889 void *buf;
2890 unsigned long len;
2891 enum object_type type;
2892 char hdr[32];
2893 int hdrlen;
Björn Steinbrink1fb23e62008-10-18 02:37:31 +02002894 int ret;
Nicolas Pitrebbac7312008-05-14 01:32:48 -04002895
Linus Torvaldsc529d752008-06-14 11:43:01 -07002896 if (has_loose_object(sha1))
Nicolas Pitrebbac7312008-05-14 01:32:48 -04002897 return 0;
2898 buf = read_packed_sha1(sha1, &type, &len);
2899 if (!buf)
2900 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2901 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
Björn Steinbrink1fb23e62008-10-18 02:37:31 +02002902 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2903 free(buf);
2904
2905 return ret;
Nicolas Pitrebbac7312008-05-14 01:32:48 -04002906}
2907
barkalow@iabervon.orgbf592c52005-07-31 20:53:44 -04002908int has_pack_index(const unsigned char *sha1)
2909{
2910 struct stat st;
2911 if (stat(sha1_pack_index_name(sha1), &st))
2912 return 0;
2913 return 1;
2914}
2915
Junio C Hamanocd673c12009-02-27 23:15:53 -08002916int has_sha1_pack(const unsigned char *sha1)
Linus Torvaldsdade09c2005-07-03 13:06:36 -07002917{
2918 struct pack_entry e;
Junio C Hamanocd673c12009-02-27 23:15:53 -08002919 return find_pack_entry(sha1, &e);
2920}
2921
Daniel Barkalow8237b182005-04-23 18:47:23 -07002922int has_sha1_file(const unsigned char *sha1)
2923{
Junio C Hamano1f688552005-06-27 03:35:33 -07002924 struct pack_entry e;
2925
Junio C Hamanocd673c12009-02-27 23:15:53 -08002926 if (find_pack_entry(sha1, &e))
Junio C Hamano1f688552005-06-27 03:35:33 -07002927 return 1;
Jeff King45e8a742013-08-30 15:14:13 -04002928 if (has_loose_object(sha1))
2929 return 1;
2930 reprepare_packed_git();
2931 return find_pack_entry(sha1, &e);
Daniel Barkalow8237b182005-04-23 18:47:23 -07002932}
Junio C Hamano74400e72005-05-01 23:45:49 -07002933
Nguyễn Thái Ngọc Duyc879daa2011-02-05 17:52:21 +07002934static void check_tree(const void *buf, size_t size)
2935{
2936 struct tree_desc desc;
2937 struct name_entry entry;
2938
2939 init_tree_desc(&desc, buf, size);
2940 while (tree_entry(&desc, &entry))
2941 /* do nothing
2942 * tree_entry() will die() on malformed entries */
2943 ;
2944}
2945
2946static void check_commit(const void *buf, size_t size)
2947{
2948 struct commit c;
2949 memset(&c, 0, sizeof(c));
2950 if (parse_commit_buffer(&c, buf, size))
2951 die("corrupt commit");
2952}
2953
2954static void check_tag(const void *buf, size_t size)
2955{
2956 struct tag t;
2957 memset(&t, 0, sizeof(t));
2958 if (parse_tag_buffer(&t, buf, size))
2959 die("corrupt tag");
2960}
2961
Dmitry Potapov43df4f82008-08-03 08:39:16 +04002962static int index_mem(unsigned char *sha1, void *buf, size_t size,
Junio C Hamanoc4ce46f2011-05-08 01:47:33 -07002963 enum object_type type,
2964 const char *path, unsigned flags)
Björn Engelmanne7332f92006-05-23 20:19:04 +02002965{
Linus Torvalds6c510be2007-02-13 11:07:23 -08002966 int ret, re_allocated = 0;
Junio C Hamanoc4ce46f2011-05-08 01:47:33 -07002967 int write_object = flags & HASH_WRITE_OBJECT;
Junio C Hamano74400e72005-05-01 23:45:49 -07002968
Bryan Larsen7672db22005-07-08 16:51:55 -07002969 if (!type)
Junio C Hamanoedaec3f2007-02-28 11:45:56 -08002970 type = OBJ_BLOB;
Linus Torvalds6c510be2007-02-13 11:07:23 -08002971
2972 /*
2973 * Convert blobs to git internal format
2974 */
Dmitry Potapov43df4f82008-08-03 08:39:16 +04002975 if ((type == OBJ_BLOB) && path) {
Brandon Caseyf285a2d2008-10-09 14:12:12 -05002976 struct strbuf nbuf = STRBUF_INIT;
Steffen Prohaska21e5ad52008-02-06 12:25:58 +01002977 if (convert_to_git(path, buf, size, &nbuf,
Ramkumar Ramachandra5e12e782011-11-15 22:29:39 +05302978 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
Pierre Habouzitb315c5c2007-09-27 12:58:23 +02002979 buf = strbuf_detach(&nbuf, &size);
Linus Torvalds6c510be2007-02-13 11:07:23 -08002980 re_allocated = 1;
2981 }
2982 }
Junio C Hamanoc4ce46f2011-05-08 01:47:33 -07002983 if (flags & HASH_FORMAT_CHECK) {
Nguyễn Thái Ngọc Duyc879daa2011-02-05 17:52:21 +07002984 if (type == OBJ_TREE)
2985 check_tree(buf, size);
2986 if (type == OBJ_COMMIT)
2987 check_commit(buf, size);
2988 if (type == OBJ_TAG)
2989 check_tag(buf, size);
2990 }
Linus Torvalds6c510be2007-02-13 11:07:23 -08002991
Bryan Larsen7672db22005-07-08 16:51:55 -07002992 if (write_object)
Junio C Hamanoedaec3f2007-02-28 11:45:56 -08002993 ret = write_sha1_file(buf, size, typename(type), sha1);
Rene Scharfeabdc3fc2006-10-14 12:45:36 +02002994 else
Junio C Hamanoedaec3f2007-02-28 11:45:56 -08002995 ret = hash_sha1_file(buf, size, typename(type), sha1);
Dmitry Potapov43df4f82008-08-03 08:39:16 +04002996 if (re_allocated)
Linus Torvalds6c510be2007-02-13 11:07:23 -08002997 free(buf);
Dmitry Potapov43df4f82008-08-03 08:39:16 +04002998 return ret;
2999}
3000
Junio C Hamano7b41e1e2011-05-08 01:47:34 -07003001static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3002 const char *path, unsigned flags)
3003{
3004 struct strbuf sbuf = STRBUF_INIT;
3005 int ret;
3006
3007 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3008 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3009 else
3010 ret = -1;
3011 strbuf_release(&sbuf);
3012 return ret;
3013}
3014
Dmitry Potapovea68b0c2010-02-21 09:32:19 +03003015#define SMALL_FILE_SIZE (32*1024)
3016
Junio C Hamano7b41e1e2011-05-08 01:47:34 -07003017static int index_core(unsigned char *sha1, int fd, size_t size,
3018 enum object_type type, const char *path,
3019 unsigned flags)
Dmitry Potapov43df4f82008-08-03 08:39:16 +04003020{
3021 int ret;
Dmitry Potapov43df4f82008-08-03 08:39:16 +04003022
Junio C Hamano7b41e1e2011-05-08 01:47:34 -07003023 if (!size) {
Junio C Hamanoc4ce46f2011-05-08 01:47:33 -07003024 ret = index_mem(sha1, NULL, size, type, path, flags);
Dmitry Potapovea68b0c2010-02-21 09:32:19 +03003025 } else if (size <= SMALL_FILE_SIZE) {
3026 char *buf = xmalloc(size);
3027 if (size == read_in_full(fd, buf, size))
Junio C Hamanoc4ce46f2011-05-08 01:47:33 -07003028 ret = index_mem(sha1, buf, size, type, path, flags);
Dmitry Potapovea68b0c2010-02-21 09:32:19 +03003029 else
3030 ret = error("short read %s", strerror(errno));
3031 free(buf);
Dmitry Potapov08bda202010-05-11 01:38:17 +04003032 } else {
Dmitry Potapov43df4f82008-08-03 08:39:16 +04003033 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
Junio C Hamanoc4ce46f2011-05-08 01:47:33 -07003034 ret = index_mem(sha1, buf, size, type, path, flags);
Linus Torvaldsaac17942005-05-03 11:46:16 -07003035 munmap(buf, size);
Dmitry Potapov08bda202010-05-11 01:38:17 +04003036 }
Junio C Hamano7b41e1e2011-05-08 01:47:34 -07003037 return ret;
3038}
3039
Junio C Hamano4dd1fbc2011-05-08 01:47:35 -07003040/*
Junio C Hamano568508e2011-10-28 14:48:40 -07003041 * This creates one packfile per large blob unless bulk-checkin
3042 * machinery is "plugged".
Junio C Hamano4dd1fbc2011-05-08 01:47:35 -07003043 *
3044 * This also bypasses the usual "convert-to-git" dance, and that is on
3045 * purpose. We could write a streaming version of the converting
3046 * functions and insert that before feeding the data to fast-import
Jeff King4f22b102012-02-24 17:10:17 -05003047 * (or equivalent in-core API described above). However, that is
3048 * somewhat complicated, as we do not know the size of the filter
3049 * result, which we need to know beforehand when writing a git object.
3050 * Since the primary motivation for trying to stream from the working
3051 * tree file and to avoid mmaping it in core is to deal with large
3052 * binary blobs, they generally do not want to get any conversion, and
3053 * callers should avoid this code path when filters are requested.
Junio C Hamano4dd1fbc2011-05-08 01:47:35 -07003054 */
3055static int index_stream(unsigned char *sha1, int fd, size_t size,
3056 enum object_type type, const char *path,
3057 unsigned flags)
3058{
Junio C Hamano568508e2011-10-28 14:48:40 -07003059 return index_bulk_checkin(sha1, fd, size, type, path, flags);
Junio C Hamano4dd1fbc2011-05-08 01:47:35 -07003060}
3061
Junio C Hamano7b41e1e2011-05-08 01:47:34 -07003062int index_fd(unsigned char *sha1, int fd, struct stat *st,
3063 enum object_type type, const char *path, unsigned flags)
3064{
3065 int ret;
3066 size_t size = xsize_t(st->st_size);
3067
3068 if (!S_ISREG(st->st_mode))
3069 ret = index_pipe(sha1, fd, type, path, flags);
Jeff King4f22b102012-02-24 17:10:17 -05003070 else if (size <= big_file_threshold || type != OBJ_BLOB ||
3071 (path && would_convert_to_git(path, NULL, 0, 0)))
Junio C Hamano7b41e1e2011-05-08 01:47:34 -07003072 ret = index_core(sha1, fd, size, type, path, flags);
Junio C Hamano4dd1fbc2011-05-08 01:47:35 -07003073 else
3074 ret = index_stream(sha1, fd, size, type, path, flags);
Dmitry Potapov43df4f82008-08-03 08:39:16 +04003075 close(fd);
Linus Torvaldsaac17942005-05-03 11:46:16 -07003076 return ret;
Junio C Hamano74400e72005-05-01 23:45:49 -07003077}
Junio C Hamanoec1fcc12005-10-07 03:42:00 -07003078
Junio C Hamanoc4ce46f2011-05-08 01:47:33 -07003079int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
Junio C Hamanoec1fcc12005-10-07 03:42:00 -07003080{
3081 int fd;
Linus Torvaldsb760d3a2008-12-17 09:51:53 -08003082 struct strbuf sb = STRBUF_INIT;
Junio C Hamanoec1fcc12005-10-07 03:42:00 -07003083
3084 switch (st->st_mode & S_IFMT) {
3085 case S_IFREG:
3086 fd = open(path, O_RDONLY);
3087 if (fd < 0)
3088 return error("open(\"%s\"): %s", path,
3089 strerror(errno));
Junio C Hamanoc4ce46f2011-05-08 01:47:33 -07003090 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
Junio C Hamanoec1fcc12005-10-07 03:42:00 -07003091 return error("%s: failed to insert into database",
3092 path);
3093 break;
3094 case S_IFLNK:
Linus Torvaldsb760d3a2008-12-17 09:51:53 -08003095 if (strbuf_readlink(&sb, path, st->st_size)) {
Junio C Hamanoec1fcc12005-10-07 03:42:00 -07003096 char *errstr = strerror(errno);
Junio C Hamanoec1fcc12005-10-07 03:42:00 -07003097 return error("readlink(\"%s\"): %s", path,
3098 errstr);
3099 }
Junio C Hamanoc4ce46f2011-05-08 01:47:33 -07003100 if (!(flags & HASH_WRITE_OBJECT))
Linus Torvaldsb760d3a2008-12-17 09:51:53 -08003101 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3102 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
Junio C Hamanoec1fcc12005-10-07 03:42:00 -07003103 return error("%s: failed to insert into database",
3104 path);
Linus Torvaldsb760d3a2008-12-17 09:51:53 -08003105 strbuf_release(&sb);
Junio C Hamanoec1fcc12005-10-07 03:42:00 -07003106 break;
Linus Torvaldsf35a6d32007-04-09 21:20:29 -07003107 case S_IFDIR:
3108 return resolve_gitlink_ref(path, "HEAD", sha1);
Junio C Hamanoec1fcc12005-10-07 03:42:00 -07003109 default:
3110 return error("%s: unsupported file type", path);
3111 }
3112 return 0;
3113}
Junio C Hamanoa69e5422007-01-22 21:55:18 -08003114
3115int read_pack_header(int fd, struct pack_header *header)
3116{
Heikki Orsilac697ad12008-05-03 16:27:26 +03003117 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3118 /* "eof before pack header was fully read" */
3119 return PH_ERROR_EOF;
3120
Junio C Hamanoa69e5422007-01-22 21:55:18 -08003121 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3122 /* "protocol error (pack signature mismatch detected)" */
3123 return PH_ERROR_PACK_SIGNATURE;
3124 if (!pack_version_ok(header->hdr_version))
3125 /* "protocol error (pack version unsupported)" */
3126 return PH_ERROR_PROTOCOL;
3127 return 0;
3128}
Jeff King40d52ff2010-04-01 20:05:23 -04003129
3130void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3131{
3132 enum object_type type = sha1_object_info(sha1, NULL);
3133 if (type < 0)
3134 die("%s is not a valid object", sha1_to_hex(sha1));
3135 if (type != expect)
3136 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3137 typename(expect));
3138}