index: make the index file format extensible.
[git.git] / update-index.c
1 /*
2  * GIT - The information manager from hell
3  *
4  * Copyright (C) Linus Torvalds, 2005
5  */
6 #include "cache.h"
7 #include "strbuf.h"
8 #include "quote.h"
9 #include "cache-tree.h"
10
11 /*
12  * Default to not allowing changes to the list of files. The
13  * tool doesn't actually care, but this makes it harder to add
14  * files to the revision control by mistake by doing something
15  * like "git-update-index *" and suddenly having all the object
16  * files be revision controlled.
17  */
18 static int allow_add;
19 static int allow_remove;
20 static int allow_replace;
21 static int allow_unmerged; /* --refresh needing merge is not error */
22 static int not_new; /* --refresh not having working tree files is not error */
23 static int quiet; /* --refresh needing update is not error */
24 static int info_only;
25 static int force_remove;
26 static int verbose;
27 static int mark_valid_only = 0;
28 #define MARK_VALID 1
29 #define UNMARK_VALID 2
30
31
32 /* Three functions to allow overloaded pointer return; see linux/err.h */
33 static inline void *ERR_PTR(long error)
34 {
35         return (void *) error;
36 }
37
38 static inline long PTR_ERR(const void *ptr)
39 {
40         return (long) ptr;
41 }
42
43 static inline long IS_ERR(const void *ptr)
44 {
45         return (unsigned long)ptr > (unsigned long)-1000L;
46 }
47
48 static void report(const char *fmt, ...)
49 {
50         va_list vp;
51
52         if (!verbose)
53                 return;
54
55         va_start(vp, fmt);
56         vprintf(fmt, vp);
57         putchar('\n');
58         va_end(vp);
59 }
60
61 static int mark_valid(const char *path)
62 {
63         int namelen = strlen(path);
64         int pos = cache_name_pos(path, namelen);
65         if (0 <= pos) {
66                 switch (mark_valid_only) {
67                 case MARK_VALID:
68                         active_cache[pos]->ce_flags |= htons(CE_VALID);
69                         break;
70                 case UNMARK_VALID:
71                         active_cache[pos]->ce_flags &= ~htons(CE_VALID);
72                         break;
73                 }
74                 cache_tree_invalidate_path(active_cache_tree, path);
75                 active_cache_changed = 1;
76                 return 0;
77         }
78         return -1;
79 }
80
81 static int add_file_to_cache(const char *path)
82 {
83         int size, namelen, option, status;
84         struct cache_entry *ce;
85         struct stat st;
86
87         status = lstat(path, &st);
88
89         /* We probably want to do this in remove_file_from_cache() and
90          * add_cache_entry() instead...
91          */
92         cache_tree_invalidate_path(active_cache_tree, path);
93
94         if (status < 0 || S_ISDIR(st.st_mode)) {
95                 /* When we used to have "path" and now we want to add
96                  * "path/file", we need a way to remove "path" before
97                  * being able to add "path/file".  However,
98                  * "git-update-index --remove path" would not work.
99                  * --force-remove can be used but this is more user
100                  * friendly, especially since we can do the opposite
101                  * case just fine without --force-remove.
102                  */
103                 if (status == 0 || (errno == ENOENT || errno == ENOTDIR)) {
104                         if (allow_remove) {
105                                 if (remove_file_from_cache(path))
106                                         return error("%s: cannot remove from the index",
107                                                      path);
108                                 else
109                                         return 0;
110                         } else if (status < 0) {
111                                 return error("%s: does not exist and --remove not passed",
112                                              path);
113                         }
114                 }
115                 if (0 == status)
116                         return error("%s: is a directory - add files inside instead",
117                                      path);
118                 else
119                         return error("lstat(\"%s\"): %s", path,
120                                      strerror(errno));
121         }
122
123         namelen = strlen(path);
124         size = cache_entry_size(namelen);
125         ce = xcalloc(1, size);
126         memcpy(ce->name, path, namelen);
127         ce->ce_flags = htons(namelen);
128         fill_stat_cache_info(ce, &st);
129
130         ce->ce_mode = create_ce_mode(st.st_mode);
131         if (!trust_executable_bit) {
132                 /* If there is an existing entry, pick the mode bits
133                  * from it.
134                  */
135                 int pos = cache_name_pos(path, namelen);
136                 if (0 <= pos)
137                         ce->ce_mode = active_cache[pos]->ce_mode;
138         }
139
140         if (index_path(ce->sha1, path, &st, !info_only))
141                 return -1;
142         option = allow_add ? ADD_CACHE_OK_TO_ADD : 0;
143         option |= allow_replace ? ADD_CACHE_OK_TO_REPLACE : 0;
144         if (add_cache_entry(ce, option))
145                 return error("%s: cannot add to the index - missing --add option?",
146                              path);
147         return 0;
148 }
149
150 /*
151  * "refresh" does not calculate a new sha1 file or bring the
152  * cache up-to-date for mode/content changes. But what it
153  * _does_ do is to "re-match" the stat information of a file
154  * with the cache, so that you can refresh the cache for a
155  * file that hasn't been changed but where the stat entry is
156  * out of date.
157  *
158  * For example, you'd want to do this after doing a "git-read-tree",
159  * to link up the stat cache details with the proper files.
160  */
161 static struct cache_entry *refresh_entry(struct cache_entry *ce, int really)
162 {
163         struct stat st;
164         struct cache_entry *updated;
165         int changed, size;
166
167         if (lstat(ce->name, &st) < 0)
168                 return ERR_PTR(-errno);
169
170         changed = ce_match_stat(ce, &st, really);
171         if (!changed) {
172                 if (really && assume_unchanged &&
173                     !(ce->ce_flags & htons(CE_VALID)))
174                         ; /* mark this one VALID again */
175                 else
176                         return NULL;
177         }
178
179         if (ce_modified(ce, &st, really))
180                 return ERR_PTR(-EINVAL);
181
182         size = ce_size(ce);
183         updated = xmalloc(size);
184         memcpy(updated, ce, size);
185         fill_stat_cache_info(updated, &st);
186
187         /* In this case, if really is not set, we should leave
188          * CE_VALID bit alone.  Otherwise, paths marked with
189          * --no-assume-unchanged (i.e. things to be edited) will
190          * reacquire CE_VALID bit automatically, which is not
191          * really what we want.
192          */
193         if (!really && assume_unchanged && !(ce->ce_flags & htons(CE_VALID)))
194                 updated->ce_flags &= ~htons(CE_VALID);
195
196         return updated;
197 }
198
199 static int refresh_cache(int really)
200 {
201         int i;
202         int has_errors = 0;
203
204         for (i = 0; i < active_nr; i++) {
205                 struct cache_entry *ce, *new;
206                 ce = active_cache[i];
207                 if (ce_stage(ce)) {
208                         while ((i < active_nr) &&
209                                ! strcmp(active_cache[i]->name, ce->name))
210                                 i++;
211                         i--;
212                         if (allow_unmerged)
213                                 continue;
214                         printf("%s: needs merge\n", ce->name);
215                         has_errors = 1;
216                         continue;
217                 }
218
219                 new = refresh_entry(ce, really);
220                 if (!new)
221                         continue;
222                 if (IS_ERR(new)) {
223                         if (not_new && PTR_ERR(new) == -ENOENT)
224                                 continue;
225                         if (really && PTR_ERR(new) == -EINVAL) {
226                                 /* If we are doing --really-refresh that
227                                  * means the index is not valid anymore.
228                                  */
229                                 ce->ce_flags &= ~htons(CE_VALID);
230                                 active_cache_changed = 1;
231                         }
232                         if (quiet)
233                                 continue;
234                         printf("%s: needs update\n", ce->name);
235                         has_errors = 1;
236                         continue;
237                 }
238                 active_cache_changed = 1;
239                 /* You can NOT just free active_cache[i] here, since it
240                  * might not be necessarily malloc()ed but can also come
241                  * from mmap(). */
242                 active_cache[i] = new;
243         }
244         return has_errors;
245 }
246
247 /*
248  * We fundamentally don't like some paths: we don't want
249  * dot or dot-dot anywhere, and for obvious reasons don't
250  * want to recurse into ".git" either.
251  *
252  * Also, we don't want double slashes or slashes at the
253  * end that can make pathnames ambiguous.
254  */
255 static int verify_dotfile(const char *rest)
256 {
257         /*
258          * The first character was '.', but that
259          * has already been discarded, we now test
260          * the rest.
261          */
262         switch (*rest) {
263         /* "." is not allowed */
264         case '\0': case '/':
265                 return 0;
266
267         /*
268          * ".git" followed by  NUL or slash is bad. This
269          * shares the path end test with the ".." case.
270          */
271         case 'g':
272                 if (rest[1] != 'i')
273                         break;
274                 if (rest[2] != 't')
275                         break;
276                 rest += 2;
277         /* fallthrough */
278         case '.':
279                 if (rest[1] == '\0' || rest[1] == '/')
280                         return 0;
281         }
282         return 1;
283 }
284
285 static int verify_path(const char *path)
286 {
287         char c;
288
289         goto inside;
290         for (;;) {
291                 if (!c)
292                         return 1;
293                 if (c == '/') {
294 inside:
295                         c = *path++;
296                         switch (c) {
297                         default:
298                                 continue;
299                         case '/': case '\0':
300                                 break;
301                         case '.':
302                                 if (verify_dotfile(path))
303                                         continue;
304                         }
305                         return 0;
306                 }
307                 c = *path++;
308         }
309 }
310
311 static int add_cacheinfo(unsigned int mode, const unsigned char *sha1,
312                          const char *path, int stage)
313 {
314         int size, len, option;
315         struct cache_entry *ce;
316
317         if (!verify_path(path))
318                 return -1;
319
320         len = strlen(path);
321         size = cache_entry_size(len);
322         ce = xcalloc(1, size);
323
324         memcpy(ce->sha1, sha1, 20);
325         memcpy(ce->name, path, len);
326         ce->ce_flags = create_ce_flags(len, stage);
327         ce->ce_mode = create_ce_mode(mode);
328         if (assume_unchanged)
329                 ce->ce_flags |= htons(CE_VALID);
330         option = allow_add ? ADD_CACHE_OK_TO_ADD : 0;
331         option |= allow_replace ? ADD_CACHE_OK_TO_REPLACE : 0;
332         if (add_cache_entry(ce, option))
333                 return error("%s: cannot add to the index - missing --add option?",
334                              path);
335         report("add '%s'", path);
336         cache_tree_invalidate_path(active_cache_tree, path);
337         return 0;
338 }
339
340 static int chmod_path(int flip, const char *path)
341 {
342         int pos;
343         struct cache_entry *ce;
344         unsigned int mode;
345
346         pos = cache_name_pos(path, strlen(path));
347         if (pos < 0)
348                 return -1;
349         ce = active_cache[pos];
350         mode = ntohl(ce->ce_mode);
351         if (!S_ISREG(mode))
352                 return -1;
353         switch (flip) {
354         case '+':
355                 ce->ce_mode |= htonl(0111); break;
356         case '-':
357                 ce->ce_mode &= htonl(~0111); break;
358         default:
359                 return -1;
360         }
361         cache_tree_invalidate_path(active_cache_tree, path);
362         active_cache_changed = 1;
363         return 0;
364 }
365
366 static struct cache_file cache_file;
367
368 static void update_one(const char *path, const char *prefix, int prefix_length)
369 {
370         const char *p = prefix_path(prefix, prefix_length, path);
371         if (!verify_path(p)) {
372                 fprintf(stderr, "Ignoring path %s\n", path);
373                 return;
374         }
375         if (mark_valid_only) {
376                 if (mark_valid(p))
377                         die("Unable to mark file %s", path);
378                 return;
379         }
380         cache_tree_invalidate_path(active_cache_tree, path);
381
382         if (force_remove) {
383                 if (remove_file_from_cache(p))
384                         die("git-update-index: unable to remove %s", path);
385                 report("remove '%s'", path);
386                 return;
387         }
388         if (add_file_to_cache(p))
389                 die("Unable to process file %s", path);
390         report("add '%s'", path);
391 }
392
393 static void read_index_info(int line_termination)
394 {
395         struct strbuf buf;
396         strbuf_init(&buf);
397         while (1) {
398                 char *ptr, *tab;
399                 char *path_name;
400                 unsigned char sha1[20];
401                 unsigned int mode;
402                 int stage;
403
404                 /* This reads lines formatted in one of three formats:
405                  *
406                  * (1) mode         SP sha1          TAB path
407                  * The first format is what "git-apply --index-info"
408                  * reports, and used to reconstruct a partial tree
409                  * that is used for phony merge base tree when falling
410                  * back on 3-way merge.
411                  *
412                  * (2) mode SP type SP sha1          TAB path
413                  * The second format is to stuff git-ls-tree output
414                  * into the index file.
415                  * 
416                  * (3) mode         SP sha1 SP stage TAB path
417                  * This format is to put higher order stages into the
418                  * index file and matches git-ls-files --stage output.
419                  */
420                 read_line(&buf, stdin, line_termination);
421                 if (buf.eof)
422                         break;
423
424                 mode = strtoul(buf.buf, &ptr, 8);
425                 if (ptr == buf.buf || *ptr != ' ')
426                         goto bad_line;
427
428                 tab = strchr(ptr, '\t');
429                 if (!tab || tab - ptr < 41)
430                         goto bad_line;
431
432                 if (tab[-2] == ' ' && '0' <= tab[-1] && tab[-1] <= '3') {
433                         stage = tab[-1] - '0';
434                         ptr = tab + 1; /* point at the head of path */
435                         tab = tab - 2; /* point at tail of sha1 */
436                 }
437                 else {
438                         stage = 0;
439                         ptr = tab + 1; /* point at the head of path */
440                 }
441
442                 if (get_sha1_hex(tab - 40, sha1) || tab[-41] != ' ')
443                         goto bad_line;
444
445                 if (line_termination && ptr[0] == '"')
446                         path_name = unquote_c_style(ptr, NULL);
447                 else
448                         path_name = ptr;
449
450                 if (!verify_path(path_name)) {
451                         fprintf(stderr, "Ignoring path %s\n", path_name);
452                         if (path_name != ptr)
453                                 free(path_name);
454                         continue;
455                 }
456                 cache_tree_invalidate_path(active_cache_tree, path_name);
457
458                 if (!mode) {
459                         /* mode == 0 means there is no such path -- remove */
460                         if (remove_file_from_cache(path_name))
461                                 die("git-update-index: unable to remove %s",
462                                     ptr);
463                 }
464                 else {
465                         /* mode ' ' sha1 '\t' name
466                          * ptr[-1] points at tab,
467                          * ptr[-41] is at the beginning of sha1
468                          */
469                         ptr[-42] = ptr[-1] = 0;
470                         if (add_cacheinfo(mode, sha1, path_name, stage))
471                                 die("git-update-index: unable to update %s",
472                                     path_name);
473                 }
474                 if (path_name != ptr)
475                         free(path_name);
476                 continue;
477
478         bad_line:
479                 die("malformed index info %s", buf.buf);
480         }
481 }
482
483 static const char update_index_usage[] =
484 "git-update-index [-q] [--add] [--replace] [--remove] [--unmerged] [--refresh] [--cacheinfo] [--chmod=(+|-)x] [--info-only] [--force-remove] [--stdin] [--index-info] [--ignore-missing] [-z] [--verbose] [--] <file>...";
485
486 int main(int argc, const char **argv)
487 {
488         int i, newfd, entries, has_errors = 0, line_termination = '\n';
489         int allow_options = 1;
490         int read_from_stdin = 0;
491         const char *prefix = setup_git_directory();
492         int prefix_length = prefix ? strlen(prefix) : 0;
493
494         git_config(git_default_config);
495
496         newfd = hold_index_file_for_update(&cache_file, get_index_file());
497         if (newfd < 0)
498                 die("unable to create new cachefile");
499
500         entries = read_cache();
501         if (entries < 0)
502                 die("cache corrupted");
503
504         for (i = 1 ; i < argc; i++) {
505                 const char *path = argv[i];
506
507                 if (allow_options && *path == '-') {
508                         if (!strcmp(path, "--")) {
509                                 allow_options = 0;
510                                 continue;
511                         }
512                         if (!strcmp(path, "-q")) {
513                                 quiet = 1;
514                                 continue;
515                         }
516                         if (!strcmp(path, "--add")) {
517                                 allow_add = 1;
518                                 continue;
519                         }
520                         if (!strcmp(path, "--replace")) {
521                                 allow_replace = 1;
522                                 continue;
523                         }
524                         if (!strcmp(path, "--remove")) {
525                                 allow_remove = 1;
526                                 continue;
527                         }
528                         if (!strcmp(path, "--unmerged")) {
529                                 allow_unmerged = 1;
530                                 continue;
531                         }
532                         if (!strcmp(path, "--refresh")) {
533                                 has_errors |= refresh_cache(0);
534                                 continue;
535                         }
536                         if (!strcmp(path, "--really-refresh")) {
537                                 has_errors |= refresh_cache(1);
538                                 continue;
539                         }
540                         if (!strcmp(path, "--cacheinfo")) {
541                                 unsigned char sha1[20];
542                                 unsigned int mode;
543
544                                 if (i+3 >= argc)
545                                         die("git-update-index: --cacheinfo <mode> <sha1> <path>");
546
547                                 if ((sscanf(argv[i+1], "%o", &mode) != 1) ||
548                                     get_sha1_hex(argv[i+2], sha1) ||
549                                     add_cacheinfo(mode, sha1, argv[i+3], 0))
550                                         die("git-update-index: --cacheinfo"
551                                             " cannot add %s", argv[i+3]);
552                                 i += 3;
553                                 continue;
554                         }
555                         if (!strcmp(path, "--chmod=-x") ||
556                             !strcmp(path, "--chmod=+x")) {
557                                 if (argc <= i+1)
558                                         die("git-update-index: %s <path>", path);
559                                 if (chmod_path(path[8], argv[++i]))
560                                         die("git-update-index: %s cannot chmod %s", path, argv[i]);
561                                 continue;
562                         }
563                         if (!strcmp(path, "--assume-unchanged")) {
564                                 mark_valid_only = MARK_VALID;
565                                 continue;
566                         }
567                         if (!strcmp(path, "--no-assume-unchanged")) {
568                                 mark_valid_only = UNMARK_VALID;
569                                 continue;
570                         }
571                         if (!strcmp(path, "--info-only")) {
572                                 info_only = 1;
573                                 continue;
574                         }
575                         if (!strcmp(path, "--force-remove")) {
576                                 force_remove = 1;
577                                 continue;
578                         }
579                         if (!strcmp(path, "-z")) {
580                                 line_termination = 0;
581                                 continue;
582                         }
583                         if (!strcmp(path, "--stdin")) {
584                                 if (i != argc - 1)
585                                         die("--stdin must be at the end");
586                                 read_from_stdin = 1;
587                                 break;
588                         }
589                         if (!strcmp(path, "--index-info")) {
590                                 if (i != argc - 1)
591                                         die("--index-info must be at the end");
592                                 allow_add = allow_replace = allow_remove = 1;
593                                 read_index_info(line_termination);
594                                 break;
595                         }
596                         if (!strcmp(path, "--ignore-missing")) {
597                                 not_new = 1;
598                                 continue;
599                         }
600                         if (!strcmp(path, "--verbose")) {
601                                 verbose = 1;
602                                 continue;
603                         }
604                         if (!strcmp(path, "-h") || !strcmp(path, "--help"))
605                                 usage(update_index_usage);
606                         die("unknown option %s", path);
607                 }
608                 update_one(path, prefix, prefix_length);
609         }
610         if (read_from_stdin) {
611                 struct strbuf buf;
612                 strbuf_init(&buf);
613                 while (1) {
614                         char *path_name;
615                         read_line(&buf, stdin, line_termination);
616                         if (buf.eof)
617                                 break;
618                         if (line_termination && buf.buf[0] == '"')
619                                 path_name = unquote_c_style(buf.buf, NULL);
620                         else
621                                 path_name = buf.buf;
622                         update_one(path_name, prefix, prefix_length);
623                         if (path_name != buf.buf)
624                                 free(path_name);
625                 }
626         }
627         if (active_cache_changed) {
628                 if (write_cache(newfd, active_cache, active_nr) ||
629                     commit_index_file(&cache_file))
630                         die("Unable to write new cachefile");
631         }
632
633         return has_errors ? 1 : 0;
634 }