[PATCH] fix compare symlink against readlink not data
[git/git.git] / update-cache.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #include <signal.h>
7 #include "cache.h"
8
9 /*
10 * Default to not allowing changes to the list of files. The
11 * tool doesn't actually care, but this makes it harder to add
12 * files to the revision control by mistake by doing something
13 * like "update-cache *" and suddenly having all the object
14 * files be revision controlled.
15 */
16 static int allow_add = 0, allow_remove = 0, not_new = 0;
17
18 /*
19 * update-cache --refresh may not touch anything at all, in which case
20 * writing 1.6MB of the same thing is a waste.
21 */
22 static int cache_changed = 0;
23
24 /* Three functions to allow overloaded pointer return; see linux/err.h */
25 static inline void *ERR_PTR(long error)
26 {
27 return (void *) error;
28 }
29
30 static inline long PTR_ERR(const void *ptr)
31 {
32 return (long) ptr;
33 }
34
35 static inline long IS_ERR(const void *ptr)
36 {
37 return (unsigned long)ptr > (unsigned long)-1000L;
38 }
39
40 /*
41 * This only updates the "non-critical" parts of the directory
42 * cache, ie the parts that aren't tracked by GIT, and only used
43 * to validate the cache.
44 */
45 static void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
46 {
47 ce->ce_ctime.sec = htonl(st->st_ctime);
48 ce->ce_mtime.sec = htonl(st->st_mtime);
49 #ifdef NSEC
50 ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
51 ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
52 #endif
53 ce->ce_dev = htonl(st->st_dev);
54 ce->ce_ino = htonl(st->st_ino);
55 ce->ce_uid = htonl(st->st_uid);
56 ce->ce_gid = htonl(st->st_gid);
57 ce->ce_size = htonl(st->st_size);
58 }
59
60 static int add_file_to_cache_1(char *path)
61 {
62 int size, namelen;
63 struct cache_entry *ce;
64 struct stat st;
65 int fd;
66 unsigned int len;
67 char *target;
68
69 if (lstat(path, &st) < 0) {
70 if (errno == ENOENT || errno == ENOTDIR) {
71 if (allow_remove)
72 return remove_file_from_cache(path);
73 }
74 return -1;
75 }
76 namelen = strlen(path);
77 size = cache_entry_size(namelen);
78 ce = xmalloc(size);
79 memset(ce, 0, size);
80 memcpy(ce->name, path, namelen);
81 fill_stat_cache_info(ce, &st);
82 ce->ce_mode = create_ce_mode(st.st_mode);
83 ce->ce_flags = htons(namelen);
84 switch (st.st_mode & S_IFMT) {
85 case S_IFREG:
86 fd = open(path, O_RDONLY);
87 if (fd < 0)
88 return -1;
89 if (index_fd(ce->sha1, fd, &st) < 0)
90 return -1;
91 break;
92 case S_IFLNK:
93 target = xmalloc(st.st_size+1);
94 if (readlink(path, target, st.st_size+1) != st.st_size) {
95 free(target);
96 return -1;
97 }
98 if (write_sha1_file(target, st.st_size, "blob", ce->sha1))
99 return -1;
100 free(target);
101 break;
102 default:
103 return -1;
104 }
105 if (!cache_changed) {
106 /* If we have not smudged the cache, be careful
107 * to keep it clean. Find out if we have a matching
108 * cache entry that add_cache_entry would replace with,
109 * and if it matches then do not bother calling it.
110 */
111 int pos = cache_name_pos(ce->name, namelen);
112 if ((0 <= pos) &&
113 !memcmp(active_cache[pos], ce, sizeof(*ce))) {
114 free(ce);
115 /* magic to tell add_file_to_cache that
116 * we have not updated anything.
117 */
118 return 999;
119 }
120 }
121 return add_cache_entry(ce, allow_add);
122 }
123
124 static int add_file_to_cache(char *path)
125 {
126 int ret = add_file_to_cache_1(path);
127 if (ret == 0)
128 cache_changed = 1;
129 else if (ret == 999)
130 ret = 0;
131 return ret;
132 }
133
134 static int match_data(int fd, void *buffer, unsigned long size)
135 {
136 while (size) {
137 char compare[1024];
138 int ret = read(fd, compare, sizeof(compare));
139
140 if (ret <= 0 || ret > size || memcmp(buffer, compare, ret))
141 return -1;
142 size -= ret;
143 buffer += ret;
144 }
145 return 0;
146 }
147
148 static int compare_data(struct cache_entry *ce, unsigned long expected_size)
149 {
150 int match = -1;
151 int fd = open(ce->name, O_RDONLY);
152
153 if (fd >= 0) {
154 void *buffer;
155 unsigned long size;
156 char type[10];
157
158 buffer = read_sha1_file(ce->sha1, type, &size);
159 if (buffer) {
160 if (size == expected_size && !strcmp(type, "blob"))
161 match = match_data(fd, buffer, size);
162 free(buffer);
163 }
164 close(fd);
165 }
166 return match;
167 }
168
169 static int compare_link(struct cache_entry *ce, unsigned long expected_size)
170 {
171 int match = -1;
172 char *target;
173 void *buffer;
174 unsigned long size;
175 char type[10];
176 int len;
177
178 target = xmalloc(expected_size);
179 len = readlink(ce->name, target, expected_size);
180 if (len != expected_size) {
181 free(target);
182 return -1;
183 }
184 buffer = read_sha1_file(ce->sha1, type, &size);
185 if (!buffer) {
186 free(target);
187 return -1;
188 }
189 if (size == expected_size)
190 match = memcmp(buffer, target, size);
191 free(buffer);
192 free(target);
193 return match;
194 }
195
196 /*
197 * "refresh" does not calculate a new sha1 file or bring the
198 * cache up-to-date for mode/content changes. But what it
199 * _does_ do is to "re-match" the stat information of a file
200 * with the cache, so that you can refresh the cache for a
201 * file that hasn't been changed but where the stat entry is
202 * out of date.
203 *
204 * For example, you'd want to do this after doing a "read-tree",
205 * to link up the stat cache details with the proper files.
206 */
207 static struct cache_entry *refresh_entry(struct cache_entry *ce)
208 {
209 struct stat st;
210 struct cache_entry *updated;
211 int changed, size;
212
213 if (lstat(ce->name, &st) < 0)
214 return ERR_PTR(-errno);
215
216 changed = cache_match_stat(ce, &st);
217 if (!changed)
218 return ce;
219
220 /*
221 * If the mode or type has changed, there's no point in trying
222 * to refresh the entry - it's not going to match
223 */
224 if (changed & (MODE_CHANGED | TYPE_CHANGED))
225 return ERR_PTR(-EINVAL);
226
227 switch (st.st_mode & S_IFMT) {
228 case S_IFREG:
229 if (compare_data(ce, st.st_size))
230 return ERR_PTR(-EINVAL);
231 break;
232 case S_IFLNK:
233 if (compare_link(ce, st.st_size))
234 return ERR_PTR(-EINVAL);
235 break;
236 default:
237 return ERR_PTR(-EINVAL);
238 }
239
240 cache_changed = 1;
241 size = ce_size(ce);
242 updated = xmalloc(size);
243 memcpy(updated, ce, size);
244 fill_stat_cache_info(updated, &st);
245 return updated;
246 }
247
248 static int refresh_cache(void)
249 {
250 int i;
251 int has_errors = 0;
252
253 for (i = 0; i < active_nr; i++) {
254 struct cache_entry *ce, *new;
255 ce = active_cache[i];
256 if (ce_stage(ce)) {
257 printf("%s: needs merge\n", ce->name);
258 has_errors = 1;
259 while ((i < active_nr) &&
260 ! strcmp(active_cache[i]->name, ce->name))
261 i++;
262 i--;
263 continue;
264 }
265
266 new = refresh_entry(ce);
267 if (IS_ERR(new)) {
268 if (!(not_new && PTR_ERR(new) == -ENOENT)) {
269 printf("%s: needs update\n", ce->name);
270 has_errors = 1;
271 }
272 continue;
273 }
274 active_cache[i] = new;
275 }
276 return has_errors;
277 }
278
279 /*
280 * We fundamentally don't like some paths: we don't want
281 * dot or dot-dot anywhere, and in fact, we don't even want
282 * any other dot-files (.git or anything else). They
283 * are hidden, for chist sake.
284 *
285 * Also, we don't want double slashes or slashes at the
286 * end that can make pathnames ambiguous.
287 */
288 static int verify_path(char *path)
289 {
290 char c;
291
292 goto inside;
293 for (;;) {
294 if (!c)
295 return 1;
296 if (c == '/') {
297 inside:
298 c = *path++;
299 if (c != '/' && c != '.' && c != '\0')
300 continue;
301 return 0;
302 }
303 c = *path++;
304 }
305 }
306
307 static int add_cacheinfo(char *arg1, char *arg2, char *arg3)
308 {
309 int size, len;
310 unsigned int mode;
311 unsigned char sha1[20];
312 struct cache_entry *ce;
313
314 if (sscanf(arg1, "%o", &mode) != 1)
315 return -1;
316 if (get_sha1_hex(arg2, sha1))
317 return -1;
318 if (!verify_path(arg3))
319 return -1;
320
321 cache_changed = 1;
322 len = strlen(arg3);
323 size = cache_entry_size(len);
324 ce = xmalloc(size);
325 memset(ce, 0, size);
326
327 memcpy(ce->sha1, sha1, 20);
328 memcpy(ce->name, arg3, len);
329 ce->ce_flags = htons(len);
330 ce->ce_mode = create_ce_mode(mode);
331 return add_cache_entry(ce, allow_add);
332 }
333
334 static const char *lockfile_name = NULL;
335
336 static void remove_lock_file(void)
337 {
338 if (lockfile_name)
339 unlink(lockfile_name);
340 }
341
342 static void remove_lock_file_on_signal(int signo)
343 {
344 remove_lock_file();
345 }
346
347 int main(int argc, char **argv)
348 {
349 int i, newfd, entries, has_errors = 0;
350 int allow_options = 1;
351 static char lockfile[MAXPATHLEN+1];
352 const char *indexfile = get_index_file();
353
354 snprintf(lockfile, sizeof(lockfile), "%s.lock", indexfile);
355
356 newfd = open(lockfile, O_RDWR | O_CREAT | O_EXCL, 0600);
357 if (newfd < 0)
358 die("unable to create new cachefile");
359
360 signal(SIGINT, remove_lock_file_on_signal);
361 atexit(remove_lock_file);
362 lockfile_name = lockfile;
363
364 entries = read_cache();
365 if (entries < 0)
366 die("cache corrupted");
367
368 for (i = 1 ; i < argc; i++) {
369 char *path = argv[i];
370
371 if (allow_options && *path == '-') {
372 if (!strcmp(path, "--")) {
373 allow_options = 0;
374 continue;
375 }
376 if (!strcmp(path, "--add")) {
377 allow_add = 1;
378 continue;
379 }
380 if (!strcmp(path, "--remove")) {
381 allow_remove = 1;
382 continue;
383 }
384 if (!strcmp(path, "--refresh")) {
385 has_errors |= refresh_cache();
386 continue;
387 }
388 if (!strcmp(path, "--cacheinfo")) {
389 if (i+3 >= argc || add_cacheinfo(argv[i+1], argv[i+2], argv[i+3]))
390 die("update-cache: --cacheinfo <mode> <sha1> <path>");
391 i += 3;
392 continue;
393 }
394 if (!strcmp(path, "--force-remove")) {
395 if (argc <= i + 1)
396 die("update-cache: --force-remove <path>");
397 if (remove_file_from_cache(argv[i+1]))
398 die("update-cache: --force-remove cannot remove %s", argv[i+1]);
399 i++;
400 continue;
401 }
402
403 if (!strcmp(path, "--ignore-missing")) {
404 not_new = 1;
405 continue;
406 }
407 die("unknown option %s", path);
408 }
409 if (!verify_path(path)) {
410 fprintf(stderr, "Ignoring path %s\n", argv[i]);
411 continue;
412 }
413 if (add_file_to_cache(path))
414 die("Unable to add %s to database", path);
415 }
416
417 if (!cache_changed)
418 unlink(lockfile);
419 else if (write_cache(newfd, active_cache, active_nr) ||
420 rename(lockfile, indexfile))
421 die("Unable to write new cachefile");
422
423 lockfile_name = NULL;
424 return has_errors ? 1 : 0;
425 }