Merge branch 'fm/user-manual-use-preface'
[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 #include "entry.h"
10
11 static void create_directories(const char *path, int path_len,
12                                const struct checkout *state)
13 {
14         char *buf = xmallocz(path_len);
15         int len = 0;
16
17         while (len < path_len) {
18                 do {
19                         buf[len] = path[len];
20                         len++;
21                 } while (len < path_len && path[len] != '/');
22                 if (len >= path_len)
23                         break;
24                 buf[len] = 0;
25
26                 /*
27                  * For 'checkout-index --prefix=<dir>', <dir> is
28                  * allowed to be a symlink to an existing directory,
29                  * and we set 'state->base_dir_len' below, such that
30                  * we test the path components of the prefix with the
31                  * stat() function instead of the lstat() function.
32                  */
33                 if (has_dirs_only_path(buf, len, state->base_dir_len))
34                         continue; /* ok, it is already a directory. */
35
36                 /*
37                  * If this mkdir() would fail, it could be that there
38                  * is already a symlink or something else exists
39                  * there, therefore we then try to unlink it and try
40                  * one more time to create the directory.
41                  */
42                 if (mkdir(buf, 0777)) {
43                         if (errno == EEXIST && state->force &&
44                             !unlink_or_warn(buf) && !mkdir(buf, 0777))
45                                 continue;
46                         die_errno("cannot create directory at '%s'", buf);
47                 }
48         }
49         free(buf);
50 }
51
52 static void remove_subtree(struct strbuf *path)
53 {
54         DIR *dir = opendir(path->buf);
55         struct dirent *de;
56         int origlen = path->len;
57
58         if (!dir)
59                 die_errno("cannot opendir '%s'", path->buf);
60         while ((de = readdir(dir)) != NULL) {
61                 struct stat st;
62
63                 if (is_dot_or_dotdot(de->d_name))
64                         continue;
65
66                 strbuf_addch(path, '/');
67                 strbuf_addstr(path, de->d_name);
68                 if (lstat(path->buf, &st))
69                         die_errno("cannot lstat '%s'", path->buf);
70                 if (S_ISDIR(st.st_mode))
71                         remove_subtree(path);
72                 else if (unlink(path->buf))
73                         die_errno("cannot unlink '%s'", path->buf);
74                 strbuf_setlen(path, origlen);
75         }
76         closedir(dir);
77         if (rmdir(path->buf))
78                 die_errno("cannot rmdir '%s'", path->buf);
79 }
80
81 static int create_file(const char *path, unsigned int mode)
82 {
83         mode = (mode & 0100) ? 0777 : 0666;
84         return open(path, O_WRONLY | O_CREAT | O_EXCL, mode);
85 }
86
87 void *read_blob_entry(const struct cache_entry *ce, unsigned long *size)
88 {
89         enum object_type type;
90         void *blob_data = read_object_file(&ce->oid, &type, size);
91
92         if (blob_data) {
93                 if (type == OBJ_BLOB)
94                         return blob_data;
95                 free(blob_data);
96         }
97         return NULL;
98 }
99
100 static int open_output_fd(char *path, const struct cache_entry *ce, int to_tempfile)
101 {
102         int symlink = (ce->ce_mode & S_IFMT) != S_IFREG;
103         if (to_tempfile) {
104                 xsnprintf(path, TEMPORARY_FILENAME_LENGTH, "%s",
105                           symlink ? ".merge_link_XXXXXX" : ".merge_file_XXXXXX");
106                 return mkstemp(path);
107         } else {
108                 return create_file(path, !symlink ? ce->ce_mode : 0666);
109         }
110 }
111
112 int fstat_checkout_output(int fd, const struct checkout *state, struct stat *st)
113 {
114         /* use fstat() only when path == ce->name */
115         if (fstat_is_reliable() &&
116             state->refresh_cache && !state->base_dir_len) {
117                 return !fstat(fd, st);
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_checkout_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, int *nr_checkouts)
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, nr_checkouts);
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 void update_ce_after_write(const struct checkout *state, struct cache_entry *ce,
255                            struct stat *st)
256 {
257         if (state->refresh_cache) {
258                 assert(state->istate);
259                 fill_stat_cache_info(state->istate, ce, st);
260                 ce->ce_flags |= CE_UPDATE_IN_BASE;
261                 mark_fsmonitor_invalid(state->istate, ce);
262                 state->istate->cache_changed |= CE_ENTRY_CHANGED;
263         }
264 }
265
266 /* Note: ca is used (and required) iff the entry refers to a regular file. */
267 static int write_entry(struct cache_entry *ce, char *path, struct conv_attrs *ca,
268                        const struct checkout *state, int to_tempfile)
269 {
270         unsigned int ce_mode_s_ifmt = ce->ce_mode & S_IFMT;
271         struct delayed_checkout *dco = state->delayed_checkout;
272         int fd, ret, fstat_done = 0;
273         char *new_blob;
274         struct strbuf buf = STRBUF_INIT;
275         unsigned long size;
276         ssize_t wrote;
277         size_t newsize = 0;
278         struct stat st;
279         const struct submodule *sub;
280         struct checkout_metadata meta;
281
282         clone_checkout_metadata(&meta, &state->meta, &ce->oid);
283
284         if (ce_mode_s_ifmt == S_IFREG) {
285                 struct stream_filter *filter = get_stream_filter_ca(ca, &ce->oid);
286                 if (filter &&
287                     !streaming_write_entry(ce, path, filter,
288                                            state, to_tempfile,
289                                            &fstat_done, &st))
290                         goto finish;
291         }
292
293         switch (ce_mode_s_ifmt) {
294         case S_IFLNK:
295                 new_blob = read_blob_entry(ce, &size);
296                 if (!new_blob)
297                         return error("unable to read sha1 file of %s (%s)",
298                                      ce->name, oid_to_hex(&ce->oid));
299
300                 /*
301                  * We can't make a real symlink; write out a regular file entry
302                  * with the symlink destination as its contents.
303                  */
304                 if (!has_symlinks || to_tempfile)
305                         goto write_file_entry;
306
307                 ret = symlink(new_blob, path);
308                 free(new_blob);
309                 if (ret)
310                         return error_errno("unable to create symlink %s", path);
311                 break;
312
313         case S_IFREG:
314                 /*
315                  * We do not send the blob in case of a retry, so do not
316                  * bother reading it at all.
317                  */
318                 if (dco && dco->state == CE_RETRY) {
319                         new_blob = NULL;
320                         size = 0;
321                 } else {
322                         new_blob = read_blob_entry(ce, &size);
323                         if (!new_blob)
324                                 return error("unable to read sha1 file of %s (%s)",
325                                              ce->name, oid_to_hex(&ce->oid));
326                 }
327
328                 /*
329                  * Convert from git internal format to working tree format
330                  */
331                 if (dco && dco->state != CE_NO_DELAY) {
332                         ret = async_convert_to_working_tree_ca(ca, ce->name,
333                                                                new_blob, size,
334                                                                &buf, &meta, dco);
335                         if (ret && string_list_has_string(&dco->paths, ce->name)) {
336                                 free(new_blob);
337                                 goto delayed;
338                         }
339                 } else {
340                         ret = convert_to_working_tree_ca(ca, ce->name, new_blob,
341                                                          size, &buf, &meta);
342                 }
343
344                 if (ret) {
345                         free(new_blob);
346                         new_blob = strbuf_detach(&buf, &newsize);
347                         size = newsize;
348                 }
349                 /*
350                  * No "else" here as errors from convert are OK at this
351                  * point. If the error would have been fatal (e.g.
352                  * filter is required), then we would have died already.
353                  */
354
355         write_file_entry:
356                 fd = open_output_fd(path, ce, to_tempfile);
357                 if (fd < 0) {
358                         free(new_blob);
359                         return error_errno("unable to create file %s", path);
360                 }
361
362                 wrote = write_in_full(fd, new_blob, size);
363                 if (!to_tempfile)
364                         fstat_done = fstat_checkout_output(fd, state, &st);
365                 close(fd);
366                 free(new_blob);
367                 if (wrote < 0)
368                         return error("unable to write file %s", path);
369                 break;
370
371         case S_IFGITLINK:
372                 if (to_tempfile)
373                         return error("cannot create temporary submodule %s", ce->name);
374                 if (mkdir(path, 0777) < 0)
375                         return error("cannot create submodule directory %s", path);
376                 sub = submodule_from_ce(ce);
377                 if (sub)
378                         return submodule_move_head(ce->name,
379                                 NULL, oid_to_hex(&ce->oid),
380                                 state->force ? SUBMODULE_MOVE_HEAD_FORCE : 0);
381                 break;
382
383         default:
384                 return error("unknown file mode for %s in index", ce->name);
385         }
386
387 finish:
388         if (state->refresh_cache) {
389                 if (!fstat_done && lstat(ce->name, &st) < 0)
390                         return error_errno("unable to stat just-written file %s",
391                                            ce->name);
392                 update_ce_after_write(state, ce , &st);
393         }
394 delayed:
395         return 0;
396 }
397
398 /*
399  * This is like 'lstat()', except it refuses to follow symlinks
400  * in the path, after skipping "skiplen".
401  */
402 static int check_path(const char *path, int len, struct stat *st, int skiplen)
403 {
404         const char *slash = path + len;
405
406         while (path < slash && *slash != '/')
407                 slash--;
408         if (!has_dirs_only_path(path, slash - path, skiplen)) {
409                 errno = ENOENT;
410                 return -1;
411         }
412         return lstat(path, st);
413 }
414
415 static void mark_colliding_entries(const struct checkout *state,
416                                    struct cache_entry *ce, struct stat *st)
417 {
418         int i, trust_ino = check_stat;
419
420 #if defined(GIT_WINDOWS_NATIVE) || defined(__CYGWIN__)
421         trust_ino = 0;
422 #endif
423
424         ce->ce_flags |= CE_MATCHED;
425
426         for (i = 0; i < state->istate->cache_nr; i++) {
427                 struct cache_entry *dup = state->istate->cache[i];
428
429                 if (dup == ce)
430                         break;
431
432                 if (dup->ce_flags & (CE_MATCHED | CE_VALID | CE_SKIP_WORKTREE))
433                         continue;
434
435                 if ((trust_ino && !match_stat_data(&dup->ce_stat_data, st)) ||
436                     (!trust_ino && !fspathcmp(ce->name, dup->name))) {
437                         dup->ce_flags |= CE_MATCHED;
438                         break;
439                 }
440         }
441 }
442
443 int checkout_entry_ca(struct cache_entry *ce, struct conv_attrs *ca,
444                       const struct checkout *state, char *topath,
445                       int *nr_checkouts)
446 {
447         static struct strbuf path = STRBUF_INIT;
448         struct stat st;
449         struct conv_attrs ca_buf;
450
451         if (ce->ce_flags & CE_WT_REMOVE) {
452                 if (topath)
453                         /*
454                          * No content and thus no path to create, so we have
455                          * no pathname to return.
456                          */
457                         BUG("Can't remove entry to a path");
458                 unlink_entry(ce);
459                 return 0;
460         }
461
462         if (topath) {
463                 if (S_ISREG(ce->ce_mode) && !ca) {
464                         convert_attrs(state->istate, &ca_buf, ce->name);
465                         ca = &ca_buf;
466                 }
467                 return write_entry(ce, topath, ca, state, 1);
468         }
469
470         strbuf_reset(&path);
471         strbuf_add(&path, state->base_dir, state->base_dir_len);
472         strbuf_add(&path, ce->name, ce_namelen(ce));
473
474         if (!check_path(path.buf, path.len, &st, state->base_dir_len)) {
475                 const struct submodule *sub;
476                 unsigned changed = ie_match_stat(state->istate, ce, &st,
477                                                  CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
478                 /*
479                  * Needs to be checked before !changed returns early,
480                  * as the possibly empty directory was not changed
481                  */
482                 sub = submodule_from_ce(ce);
483                 if (sub) {
484                         int err;
485                         if (!is_submodule_populated_gently(ce->name, &err)) {
486                                 struct stat sb;
487                                 if (lstat(ce->name, &sb))
488                                         die(_("could not stat file '%s'"), ce->name);
489                                 if (!(st.st_mode & S_IFDIR))
490                                         unlink_or_warn(ce->name);
491
492                                 return submodule_move_head(ce->name,
493                                         NULL, oid_to_hex(&ce->oid), 0);
494                         } else
495                                 return submodule_move_head(ce->name,
496                                         "HEAD", oid_to_hex(&ce->oid),
497                                         state->force ? SUBMODULE_MOVE_HEAD_FORCE : 0);
498                 }
499
500                 if (!changed)
501                         return 0;
502                 if (!state->force) {
503                         if (!state->quiet)
504                                 fprintf(stderr,
505                                         "%s already exists, no checkout\n",
506                                         path.buf);
507                         return -1;
508                 }
509
510                 if (state->clone)
511                         mark_colliding_entries(state, ce, &st);
512
513                 /*
514                  * We unlink the old file, to get the new one with the
515                  * right permissions (including umask, which is nasty
516                  * to emulate by hand - much easier to let the system
517                  * just do the right thing)
518                  */
519                 if (S_ISDIR(st.st_mode)) {
520                         /* If it is a gitlink, leave it alone! */
521                         if (S_ISGITLINK(ce->ce_mode))
522                                 return 0;
523                         remove_subtree(&path);
524                 } else if (unlink(path.buf))
525                         return error_errno("unable to unlink old '%s'", path.buf);
526         } else if (state->not_new)
527                 return 0;
528
529         create_directories(path.buf, path.len, state);
530
531         if (nr_checkouts)
532                 (*nr_checkouts)++;
533
534         if (S_ISREG(ce->ce_mode) && !ca) {
535                 convert_attrs(state->istate, &ca_buf, ce->name);
536                 ca = &ca_buf;
537         }
538
539         return write_entry(ce, path.buf, ca, state, 0);
540 }
541
542 void unlink_entry(const struct cache_entry *ce)
543 {
544         const struct submodule *sub = submodule_from_ce(ce);
545         if (sub) {
546                 /* state.force is set at the caller. */
547                 submodule_move_head(ce->name, "HEAD", NULL,
548                                     SUBMODULE_MOVE_HEAD_FORCE);
549         }
550         if (check_leading_path(ce->name, ce_namelen(ce), 1) >= 0)
551                 return;
552         if (remove_or_warn(ce->ce_mode, ce->name))
553                 return;
554         schedule_dir_for_removal(ce->name, ce_namelen(ce));
555 }