git-commit.txt: better description what it does
[git] / entry.c
1 #include "cache.h"
2 #include "blob.h"
3 #include "object-store.h"
4 #include "dir.h"
5 #include "streaming.h"
6 #include "submodule.h"
7 #include "progress.h"
8 #include "fsmonitor.h"
9
10 static void create_directories(const char *path, int path_len,
11                                const struct checkout *state)
12 {
13         char *buf = xmallocz(path_len);
14         int len = 0;
15
16         while (len < path_len) {
17                 do {
18                         buf[len] = path[len];
19                         len++;
20                 } while (len < path_len && path[len] != '/');
21                 if (len >= path_len)
22                         break;
23                 buf[len] = 0;
24
25                 /*
26                  * For 'checkout-index --prefix=<dir>', <dir> is
27                  * allowed to be a symlink to an existing directory,
28                  * and we set 'state->base_dir_len' below, such that
29                  * we test the path components of the prefix with the
30                  * stat() function instead of the lstat() function.
31                  */
32                 if (has_dirs_only_path(buf, len, state->base_dir_len))
33                         continue; /* ok, it is already a directory. */
34
35                 /*
36                  * If this mkdir() would fail, it could be that there
37                  * is already a symlink or something else exists
38                  * there, therefore we then try to unlink it and try
39                  * one more time to create the directory.
40                  */
41                 if (mkdir(buf, 0777)) {
42                         if (errno == EEXIST && state->force &&
43                             !unlink_or_warn(buf) && !mkdir(buf, 0777))
44                                 continue;
45                         die_errno("cannot create directory at '%s'", buf);
46                 }
47         }
48         free(buf);
49 }
50
51 static void remove_subtree(struct strbuf *path)
52 {
53         DIR *dir = opendir(path->buf);
54         struct dirent *de;
55         int origlen = path->len;
56
57         if (!dir)
58                 die_errno("cannot opendir '%s'", path->buf);
59         while ((de = readdir(dir)) != NULL) {
60                 struct stat st;
61
62                 if (is_dot_or_dotdot(de->d_name))
63                         continue;
64
65                 strbuf_addch(path, '/');
66                 strbuf_addstr(path, de->d_name);
67                 if (lstat(path->buf, &st))
68                         die_errno("cannot lstat '%s'", path->buf);
69                 if (S_ISDIR(st.st_mode))
70                         remove_subtree(path);
71                 else if (unlink(path->buf))
72                         die_errno("cannot unlink '%s'", path->buf);
73                 strbuf_setlen(path, origlen);
74         }
75         closedir(dir);
76         if (rmdir(path->buf))
77                 die_errno("cannot rmdir '%s'", path->buf);
78 }
79
80 static int create_file(const char *path, unsigned int mode)
81 {
82         mode = (mode & 0100) ? 0777 : 0666;
83         return open(path, O_WRONLY | O_CREAT | O_EXCL, mode);
84 }
85
86 static void *read_blob_entry(const struct cache_entry *ce, unsigned long *size)
87 {
88         enum object_type type;
89         void *blob_data = read_object_file(&ce->oid, &type, size);
90
91         if (blob_data) {
92                 if (type == OBJ_BLOB)
93                         return blob_data;
94                 free(blob_data);
95         }
96         return NULL;
97 }
98
99 static int open_output_fd(char *path, const struct cache_entry *ce, int to_tempfile)
100 {
101         int symlink = (ce->ce_mode & S_IFMT) != S_IFREG;
102         if (to_tempfile) {
103                 xsnprintf(path, TEMPORARY_FILENAME_LENGTH, "%s",
104                           symlink ? ".merge_link_XXXXXX" : ".merge_file_XXXXXX");
105                 return mkstemp(path);
106         } else {
107                 return create_file(path, !symlink ? ce->ce_mode : 0666);
108         }
109 }
110
111 static int fstat_output(int fd, const struct checkout *state, struct stat *st)
112 {
113         /* use fstat() only when path == ce->name */
114         if (fstat_is_reliable() &&
115             state->refresh_cache && !state->base_dir_len) {
116                 fstat(fd, st);
117                 return 1;
118         }
119         return 0;
120 }
121
122 static int streaming_write_entry(const struct cache_entry *ce, char *path,
123                                  struct stream_filter *filter,
124                                  const struct checkout *state, int to_tempfile,
125                                  int *fstat_done, struct stat *statbuf)
126 {
127         int result = 0;
128         int fd;
129
130         fd = open_output_fd(path, ce, to_tempfile);
131         if (fd < 0)
132                 return -1;
133
134         result |= stream_blob_to_fd(fd, &ce->oid, filter, 1);
135         *fstat_done = fstat_output(fd, state, statbuf);
136         result |= close(fd);
137
138         if (result)
139                 unlink(path);
140         return result;
141 }
142
143 void enable_delayed_checkout(struct checkout *state)
144 {
145         if (!state->delayed_checkout) {
146                 state->delayed_checkout = xmalloc(sizeof(*state->delayed_checkout));
147                 state->delayed_checkout->state = CE_CAN_DELAY;
148                 string_list_init(&state->delayed_checkout->filters, 0);
149                 string_list_init(&state->delayed_checkout->paths, 0);
150         }
151 }
152
153 static int remove_available_paths(struct string_list_item *item, void *cb_data)
154 {
155         struct string_list *available_paths = cb_data;
156         struct string_list_item *available;
157
158         available = string_list_lookup(available_paths, item->string);
159         if (available)
160                 available->util = (void *)item->string;
161         return !available;
162 }
163
164 int finish_delayed_checkout(struct checkout *state)
165 {
166         int errs = 0;
167         unsigned delayed_object_count;
168         off_t filtered_bytes = 0;
169         struct string_list_item *filter, *path;
170         struct progress *progress;
171         struct delayed_checkout *dco = state->delayed_checkout;
172
173         if (!state->delayed_checkout)
174                 return errs;
175
176         dco->state = CE_RETRY;
177         delayed_object_count = dco->paths.nr;
178         progress = start_delayed_progress(_("Filtering content"), delayed_object_count);
179         while (dco->filters.nr > 0) {
180                 for_each_string_list_item(filter, &dco->filters) {
181                         struct string_list available_paths = STRING_LIST_INIT_NODUP;
182                         display_progress(progress, delayed_object_count - dco->paths.nr);
183
184                         if (!async_query_available_blobs(filter->string, &available_paths)) {
185                                 /* Filter reported an error */
186                                 errs = 1;
187                                 filter->string = "";
188                                 continue;
189                         }
190                         if (available_paths.nr <= 0) {
191                                 /*
192                                  * Filter responded with no entries. That means
193                                  * the filter is done and we can remove the
194                                  * filter from the list (see
195                                  * "string_list_remove_empty_items" call below).
196                                  */
197                                 filter->string = "";
198                                 continue;
199                         }
200
201                         /*
202                          * In dco->paths we store a list of all delayed paths.
203                          * The filter just send us a list of available paths.
204                          * Remove them from the list.
205                          */
206                         filter_string_list(&dco->paths, 0,
207                                 &remove_available_paths, &available_paths);
208
209                         for_each_string_list_item(path, &available_paths) {
210                                 struct cache_entry* ce;
211
212                                 if (!path->util) {
213                                         error("external filter '%s' signaled that '%s' "
214                                               "is now available although it has not been "
215                                               "delayed earlier",
216                                               filter->string, path->string);
217                                         errs |= 1;
218
219                                         /*
220                                          * Do not ask the filter for available blobs,
221                                          * again, as the filter is likely buggy.
222                                          */
223                                         filter->string = "";
224                                         continue;
225                                 }
226                                 ce = index_file_exists(state->istate, path->string,
227                                                        strlen(path->string), 0);
228                                 if (ce) {
229                                         errs |= checkout_entry(ce, state, NULL);
230                                         filtered_bytes += ce->ce_stat_data.sd_size;
231                                         display_throughput(progress, filtered_bytes);
232                                 } else
233                                         errs = 1;
234                         }
235                 }
236                 string_list_remove_empty_items(&dco->filters, 0);
237         }
238         stop_progress(&progress);
239         string_list_clear(&dco->filters, 0);
240
241         /* At this point we should not have any delayed paths anymore. */
242         errs |= dco->paths.nr;
243         for_each_string_list_item(path, &dco->paths) {
244                 error("'%s' was not filtered properly", path->string);
245         }
246         string_list_clear(&dco->paths, 0);
247
248         free(dco);
249         state->delayed_checkout = NULL;
250
251         return errs;
252 }
253
254 static int write_entry(struct cache_entry *ce,
255                        char *path, const struct checkout *state, int to_tempfile)
256 {
257         unsigned int ce_mode_s_ifmt = ce->ce_mode & S_IFMT;
258         struct delayed_checkout *dco = state->delayed_checkout;
259         int fd, ret, fstat_done = 0;
260         char *new_blob;
261         struct strbuf buf = STRBUF_INIT;
262         unsigned long size;
263         ssize_t wrote;
264         size_t newsize = 0;
265         struct stat st;
266         const struct submodule *sub;
267
268         if (ce_mode_s_ifmt == S_IFREG) {
269                 struct stream_filter *filter = get_stream_filter(state->istate, ce->name,
270                                                                  &ce->oid);
271                 if (filter &&
272                     !streaming_write_entry(ce, path, filter,
273                                            state, to_tempfile,
274                                            &fstat_done, &st))
275                         goto finish;
276         }
277
278         switch (ce_mode_s_ifmt) {
279         case S_IFLNK:
280                 new_blob = read_blob_entry(ce, &size);
281                 if (!new_blob)
282                         return error("unable to read sha1 file of %s (%s)",
283                                      path, oid_to_hex(&ce->oid));
284
285                 /*
286                  * We can't make a real symlink; write out a regular file entry
287                  * with the symlink destination as its contents.
288                  */
289                 if (!has_symlinks || to_tempfile)
290                         goto write_file_entry;
291
292                 ret = symlink(new_blob, path);
293                 free(new_blob);
294                 if (ret)
295                         return error_errno("unable to create symlink %s", path);
296                 break;
297
298         case S_IFREG:
299                 /*
300                  * We do not send the blob in case of a retry, so do not
301                  * bother reading it at all.
302                  */
303                 if (dco && dco->state == CE_RETRY) {
304                         new_blob = NULL;
305                         size = 0;
306                 } else {
307                         new_blob = read_blob_entry(ce, &size);
308                         if (!new_blob)
309                                 return error("unable to read sha1 file of %s (%s)",
310                                              path, oid_to_hex(&ce->oid));
311                 }
312
313                 /*
314                  * Convert from git internal format to working tree format
315                  */
316                 if (dco && dco->state != CE_NO_DELAY) {
317                         ret = async_convert_to_working_tree(state->istate, ce->name, new_blob,
318                                                             size, &buf, dco);
319                         if (ret && string_list_has_string(&dco->paths, ce->name)) {
320                                 free(new_blob);
321                                 goto delayed;
322                         }
323                 } else
324                         ret = convert_to_working_tree(state->istate, ce->name, new_blob, size, &buf);
325
326                 if (ret) {
327                         free(new_blob);
328                         new_blob = strbuf_detach(&buf, &newsize);
329                         size = newsize;
330                 }
331                 /*
332                  * No "else" here as errors from convert are OK at this
333                  * point. If the error would have been fatal (e.g.
334                  * filter is required), then we would have died already.
335                  */
336
337         write_file_entry:
338                 fd = open_output_fd(path, ce, to_tempfile);
339                 if (fd < 0) {
340                         free(new_blob);
341                         return error_errno("unable to create file %s", path);
342                 }
343
344                 wrote = write_in_full(fd, new_blob, size);
345                 if (!to_tempfile)
346                         fstat_done = fstat_output(fd, state, &st);
347                 close(fd);
348                 free(new_blob);
349                 if (wrote < 0)
350                         return error("unable to write file %s", path);
351                 break;
352
353         case S_IFGITLINK:
354                 if (to_tempfile)
355                         return error("cannot create temporary submodule %s", path);
356                 if (mkdir(path, 0777) < 0)
357                         return error("cannot create submodule directory %s", path);
358                 sub = submodule_from_ce(ce);
359                 if (sub)
360                         return submodule_move_head(ce->name,
361                                 NULL, oid_to_hex(&ce->oid),
362                                 state->force ? SUBMODULE_MOVE_HEAD_FORCE : 0);
363                 break;
364
365         default:
366                 return error("unknown file mode for %s in index", path);
367         }
368
369 finish:
370         if (state->refresh_cache) {
371                 assert(state->istate);
372                 if (!fstat_done)
373                         if (lstat(ce->name, &st) < 0)
374                                 return error_errno("unable to stat just-written file %s",
375                                                    ce->name);
376                 fill_stat_cache_info(ce, &st);
377                 ce->ce_flags |= CE_UPDATE_IN_BASE;
378                 mark_fsmonitor_invalid(state->istate, ce);
379                 state->istate->cache_changed |= CE_ENTRY_CHANGED;
380         }
381 delayed:
382         return 0;
383 }
384
385 /*
386  * This is like 'lstat()', except it refuses to follow symlinks
387  * in the path, after skipping "skiplen".
388  */
389 static int check_path(const char *path, int len, struct stat *st, int skiplen)
390 {
391         const char *slash = path + len;
392
393         while (path < slash && *slash != '/')
394                 slash--;
395         if (!has_dirs_only_path(path, slash - path, skiplen)) {
396                 errno = ENOENT;
397                 return -1;
398         }
399         return lstat(path, st);
400 }
401
402 static void mark_colliding_entries(const struct checkout *state,
403                                    struct cache_entry *ce, struct stat *st)
404 {
405         int i, trust_ino = check_stat;
406
407 #if defined(GIT_WINDOWS_NATIVE) || defined(__CYGWIN__)
408         trust_ino = 0;
409 #endif
410
411         ce->ce_flags |= CE_MATCHED;
412
413         for (i = 0; i < state->istate->cache_nr; i++) {
414                 struct cache_entry *dup = state->istate->cache[i];
415
416                 if (dup == ce)
417                         break;
418
419                 if (dup->ce_flags & (CE_MATCHED | CE_VALID | CE_SKIP_WORKTREE))
420                         continue;
421
422                 if ((trust_ino && !match_stat_data(&dup->ce_stat_data, st)) ||
423                     (!trust_ino && !fspathcmp(ce->name, dup->name))) {
424                         dup->ce_flags |= CE_MATCHED;
425                         break;
426                 }
427         }
428 }
429
430 /*
431  * Write the contents from ce out to the working tree.
432  *
433  * When topath[] is not NULL, instead of writing to the working tree
434  * file named by ce, a temporary file is created by this function and
435  * its name is returned in topath[], which must be able to hold at
436  * least TEMPORARY_FILENAME_LENGTH bytes long.
437  */
438 int checkout_entry(struct cache_entry *ce,
439                    const struct checkout *state, char *topath)
440 {
441         static struct strbuf path = STRBUF_INIT;
442         struct stat st;
443
444         if (topath)
445                 return write_entry(ce, topath, state, 1);
446
447         strbuf_reset(&path);
448         strbuf_add(&path, state->base_dir, state->base_dir_len);
449         strbuf_add(&path, ce->name, ce_namelen(ce));
450
451         if (!check_path(path.buf, path.len, &st, state->base_dir_len)) {
452                 const struct submodule *sub;
453                 unsigned changed = ie_match_stat(state->istate, ce, &st,
454                                                  CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
455                 /*
456                  * Needs to be checked before !changed returns early,
457                  * as the possibly empty directory was not changed
458                  */
459                 sub = submodule_from_ce(ce);
460                 if (sub) {
461                         int err;
462                         if (!is_submodule_populated_gently(ce->name, &err)) {
463                                 struct stat sb;
464                                 if (lstat(ce->name, &sb))
465                                         die(_("could not stat file '%s'"), ce->name);
466                                 if (!(st.st_mode & S_IFDIR))
467                                         unlink_or_warn(ce->name);
468
469                                 return submodule_move_head(ce->name,
470                                         NULL, oid_to_hex(&ce->oid), 0);
471                         } else
472                                 return submodule_move_head(ce->name,
473                                         "HEAD", oid_to_hex(&ce->oid),
474                                         state->force ? SUBMODULE_MOVE_HEAD_FORCE : 0);
475                 }
476
477                 if (!changed)
478                         return 0;
479                 if (!state->force) {
480                         if (!state->quiet)
481                                 fprintf(stderr,
482                                         "%s already exists, no checkout\n",
483                                         path.buf);
484                         return -1;
485                 }
486
487                 if (state->clone)
488                         mark_colliding_entries(state, ce, &st);
489
490                 /*
491                  * We unlink the old file, to get the new one with the
492                  * right permissions (including umask, which is nasty
493                  * to emulate by hand - much easier to let the system
494                  * just do the right thing)
495                  */
496                 if (S_ISDIR(st.st_mode)) {
497                         /* If it is a gitlink, leave it alone! */
498                         if (S_ISGITLINK(ce->ce_mode))
499                                 return 0;
500                         if (!state->force)
501                                 return error("%s is a directory", path.buf);
502                         remove_subtree(&path);
503                 } else if (unlink(path.buf))
504                         return error_errno("unable to unlink old '%s'", path.buf);
505         } else if (state->not_new)
506                 return 0;
507
508         create_directories(path.buf, path.len, state);
509         return write_entry(ce, path.buf, state, 0);
510 }