sparse-index: add guard to ensure full index
[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                 return !fstat(fd, st);
117         }
118         return 0;
119 }
120
121 static int streaming_write_entry(const struct cache_entry *ce, char *path,
122                                  struct stream_filter *filter,
123                                  const struct checkout *state, int to_tempfile,
124                                  int *fstat_done, struct stat *statbuf)
125 {
126         int result = 0;
127         int fd;
128
129         fd = open_output_fd(path, ce, to_tempfile);
130         if (fd < 0)
131                 return -1;
132
133         result |= stream_blob_to_fd(fd, &ce->oid, filter, 1);
134         *fstat_done = fstat_output(fd, state, statbuf);
135         result |= close(fd);
136
137         if (result)
138                 unlink(path);
139         return result;
140 }
141
142 void enable_delayed_checkout(struct checkout *state)
143 {
144         if (!state->delayed_checkout) {
145                 state->delayed_checkout = xmalloc(sizeof(*state->delayed_checkout));
146                 state->delayed_checkout->state = CE_CAN_DELAY;
147                 string_list_init(&state->delayed_checkout->filters, 0);
148                 string_list_init(&state->delayed_checkout->paths, 0);
149         }
150 }
151
152 static int remove_available_paths(struct string_list_item *item, void *cb_data)
153 {
154         struct string_list *available_paths = cb_data;
155         struct string_list_item *available;
156
157         available = string_list_lookup(available_paths, item->string);
158         if (available)
159                 available->util = (void *)item->string;
160         return !available;
161 }
162
163 int finish_delayed_checkout(struct checkout *state, int *nr_checkouts)
164 {
165         int errs = 0;
166         unsigned delayed_object_count;
167         off_t filtered_bytes = 0;
168         struct string_list_item *filter, *path;
169         struct progress *progress;
170         struct delayed_checkout *dco = state->delayed_checkout;
171
172         if (!state->delayed_checkout)
173                 return errs;
174
175         dco->state = CE_RETRY;
176         delayed_object_count = dco->paths.nr;
177         progress = start_delayed_progress(_("Filtering content"), delayed_object_count);
178         while (dco->filters.nr > 0) {
179                 for_each_string_list_item(filter, &dco->filters) {
180                         struct string_list available_paths = STRING_LIST_INIT_NODUP;
181                         display_progress(progress, delayed_object_count - dco->paths.nr);
182
183                         if (!async_query_available_blobs(filter->string, &available_paths)) {
184                                 /* Filter reported an error */
185                                 errs = 1;
186                                 filter->string = "";
187                                 continue;
188                         }
189                         if (available_paths.nr <= 0) {
190                                 /*
191                                  * Filter responded with no entries. That means
192                                  * the filter is done and we can remove the
193                                  * filter from the list (see
194                                  * "string_list_remove_empty_items" call below).
195                                  */
196                                 filter->string = "";
197                                 continue;
198                         }
199
200                         /*
201                          * In dco->paths we store a list of all delayed paths.
202                          * The filter just send us a list of available paths.
203                          * Remove them from the list.
204                          */
205                         filter_string_list(&dco->paths, 0,
206                                 &remove_available_paths, &available_paths);
207
208                         for_each_string_list_item(path, &available_paths) {
209                                 struct cache_entry* ce;
210
211                                 if (!path->util) {
212                                         error("external filter '%s' signaled that '%s' "
213                                               "is now available although it has not been "
214                                               "delayed earlier",
215                                               filter->string, path->string);
216                                         errs |= 1;
217
218                                         /*
219                                          * Do not ask the filter for available blobs,
220                                          * again, as the filter is likely buggy.
221                                          */
222                                         filter->string = "";
223                                         continue;
224                                 }
225                                 ce = index_file_exists(state->istate, path->string,
226                                                        strlen(path->string), 0);
227                                 if (ce) {
228                                         errs |= checkout_entry(ce, state, NULL, nr_checkouts);
229                                         filtered_bytes += ce->ce_stat_data.sd_size;
230                                         display_throughput(progress, filtered_bytes);
231                                 } else
232                                         errs = 1;
233                         }
234                 }
235                 string_list_remove_empty_items(&dco->filters, 0);
236         }
237         stop_progress(&progress);
238         string_list_clear(&dco->filters, 0);
239
240         /* At this point we should not have any delayed paths anymore. */
241         errs |= dco->paths.nr;
242         for_each_string_list_item(path, &dco->paths) {
243                 error("'%s' was not filtered properly", path->string);
244         }
245         string_list_clear(&dco->paths, 0);
246
247         free(dco);
248         state->delayed_checkout = NULL;
249
250         return errs;
251 }
252
253 static int write_entry(struct cache_entry *ce,
254                        char *path, const struct checkout *state, int to_tempfile)
255 {
256         unsigned int ce_mode_s_ifmt = ce->ce_mode & S_IFMT;
257         struct delayed_checkout *dco = state->delayed_checkout;
258         int fd, ret, fstat_done = 0;
259         char *new_blob;
260         struct strbuf buf = STRBUF_INIT;
261         unsigned long size;
262         ssize_t wrote;
263         size_t newsize = 0;
264         struct stat st;
265         const struct submodule *sub;
266         struct checkout_metadata meta;
267
268         clone_checkout_metadata(&meta, &state->meta, &ce->oid);
269
270         if (ce_mode_s_ifmt == S_IFREG) {
271                 struct stream_filter *filter = get_stream_filter(state->istate, ce->name,
272                                                                  &ce->oid);
273                 if (filter &&
274                     !streaming_write_entry(ce, path, filter,
275                                            state, to_tempfile,
276                                            &fstat_done, &st))
277                         goto finish;
278         }
279
280         switch (ce_mode_s_ifmt) {
281         case S_IFLNK:
282                 new_blob = read_blob_entry(ce, &size);
283                 if (!new_blob)
284                         return error("unable to read sha1 file of %s (%s)",
285                                      ce->name, oid_to_hex(&ce->oid));
286
287                 /*
288                  * We can't make a real symlink; write out a regular file entry
289                  * with the symlink destination as its contents.
290                  */
291                 if (!has_symlinks || to_tempfile)
292                         goto write_file_entry;
293
294                 ret = symlink(new_blob, path);
295                 free(new_blob);
296                 if (ret)
297                         return error_errno("unable to create symlink %s", path);
298                 break;
299
300         case S_IFREG:
301                 /*
302                  * We do not send the blob in case of a retry, so do not
303                  * bother reading it at all.
304                  */
305                 if (dco && dco->state == CE_RETRY) {
306                         new_blob = NULL;
307                         size = 0;
308                 } else {
309                         new_blob = read_blob_entry(ce, &size);
310                         if (!new_blob)
311                                 return error("unable to read sha1 file of %s (%s)",
312                                              ce->name, oid_to_hex(&ce->oid));
313                 }
314
315                 /*
316                  * Convert from git internal format to working tree format
317                  */
318                 if (dco && dco->state != CE_NO_DELAY) {
319                         ret = async_convert_to_working_tree(state->istate, ce->name, new_blob,
320                                                             size, &buf, &meta, dco);
321                         if (ret && string_list_has_string(&dco->paths, ce->name)) {
322                                 free(new_blob);
323                                 goto delayed;
324                         }
325                 } else
326                         ret = convert_to_working_tree(state->istate, ce->name, new_blob, size, &buf, &meta);
327
328                 if (ret) {
329                         free(new_blob);
330                         new_blob = strbuf_detach(&buf, &newsize);
331                         size = newsize;
332                 }
333                 /*
334                  * No "else" here as errors from convert are OK at this
335                  * point. If the error would have been fatal (e.g.
336                  * filter is required), then we would have died already.
337                  */
338
339         write_file_entry:
340                 fd = open_output_fd(path, ce, to_tempfile);
341                 if (fd < 0) {
342                         free(new_blob);
343                         return error_errno("unable to create file %s", path);
344                 }
345
346                 wrote = write_in_full(fd, new_blob, size);
347                 if (!to_tempfile)
348                         fstat_done = fstat_output(fd, state, &st);
349                 close(fd);
350                 free(new_blob);
351                 if (wrote < 0)
352                         return error("unable to write file %s", path);
353                 break;
354
355         case S_IFGITLINK:
356                 if (to_tempfile)
357                         return error("cannot create temporary submodule %s", ce->name);
358                 if (mkdir(path, 0777) < 0)
359                         return error("cannot create submodule directory %s", path);
360                 sub = submodule_from_ce(ce);
361                 if (sub)
362                         return submodule_move_head(ce->name,
363                                 NULL, oid_to_hex(&ce->oid),
364                                 state->force ? SUBMODULE_MOVE_HEAD_FORCE : 0);
365                 break;
366
367         default:
368                 return error("unknown file mode for %s in index", ce->name);
369         }
370
371 finish:
372         if (state->refresh_cache) {
373                 assert(state->istate);
374                 if (!fstat_done)
375                         if (lstat(ce->name, &st) < 0)
376                                 return error_errno("unable to stat just-written file %s",
377                                                    ce->name);
378                 fill_stat_cache_info(state->istate, ce, &st);
379                 ce->ce_flags |= CE_UPDATE_IN_BASE;
380                 mark_fsmonitor_invalid(state->istate, ce);
381                 state->istate->cache_changed |= CE_ENTRY_CHANGED;
382         }
383 delayed:
384         return 0;
385 }
386
387 /*
388  * This is like 'lstat()', except it refuses to follow symlinks
389  * in the path, after skipping "skiplen".
390  */
391 static int check_path(const char *path, int len, struct stat *st, int skiplen)
392 {
393         const char *slash = path + len;
394
395         while (path < slash && *slash != '/')
396                 slash--;
397         if (!has_dirs_only_path(path, slash - path, skiplen)) {
398                 errno = ENOENT;
399                 return -1;
400         }
401         return lstat(path, st);
402 }
403
404 static void mark_colliding_entries(const struct checkout *state,
405                                    struct cache_entry *ce, struct stat *st)
406 {
407         int i, trust_ino = check_stat;
408
409 #if defined(GIT_WINDOWS_NATIVE) || defined(__CYGWIN__)
410         trust_ino = 0;
411 #endif
412
413         ce->ce_flags |= CE_MATCHED;
414
415         for (i = 0; i < state->istate->cache_nr; i++) {
416                 struct cache_entry *dup = state->istate->cache[i];
417
418                 if (dup == ce)
419                         break;
420
421                 if (dup->ce_flags & (CE_MATCHED | CE_VALID | CE_SKIP_WORKTREE))
422                         continue;
423
424                 if ((trust_ino && !match_stat_data(&dup->ce_stat_data, st)) ||
425                     (!trust_ino && !fspathcmp(ce->name, dup->name))) {
426                         dup->ce_flags |= CE_MATCHED;
427                         break;
428                 }
429         }
430 }
431
432 /*
433  * Write the contents from ce out to the working tree.
434  *
435  * When topath[] is not NULL, instead of writing to the working tree
436  * file named by ce, a temporary file is created by this function and
437  * its name is returned in topath[], which must be able to hold at
438  * least TEMPORARY_FILENAME_LENGTH bytes long.
439  */
440 int checkout_entry(struct cache_entry *ce, const struct checkout *state,
441                    char *topath, int *nr_checkouts)
442 {
443         static struct strbuf path = STRBUF_INIT;
444         struct stat st;
445
446         if (ce->ce_flags & CE_WT_REMOVE) {
447                 if (topath)
448                         /*
449                          * No content and thus no path to create, so we have
450                          * no pathname to return.
451                          */
452                         BUG("Can't remove entry to a path");
453                 unlink_entry(ce);
454                 return 0;
455         }
456
457         if (topath)
458                 return write_entry(ce, topath, state, 1);
459
460         strbuf_reset(&path);
461         strbuf_add(&path, state->base_dir, state->base_dir_len);
462         strbuf_add(&path, ce->name, ce_namelen(ce));
463
464         if (!check_path(path.buf, path.len, &st, state->base_dir_len)) {
465                 const struct submodule *sub;
466                 unsigned changed = ie_match_stat(state->istate, ce, &st,
467                                                  CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
468                 /*
469                  * Needs to be checked before !changed returns early,
470                  * as the possibly empty directory was not changed
471                  */
472                 sub = submodule_from_ce(ce);
473                 if (sub) {
474                         int err;
475                         if (!is_submodule_populated_gently(ce->name, &err)) {
476                                 struct stat sb;
477                                 if (lstat(ce->name, &sb))
478                                         die(_("could not stat file '%s'"), ce->name);
479                                 if (!(st.st_mode & S_IFDIR))
480                                         unlink_or_warn(ce->name);
481
482                                 return submodule_move_head(ce->name,
483                                         NULL, oid_to_hex(&ce->oid), 0);
484                         } else
485                                 return submodule_move_head(ce->name,
486                                         "HEAD", oid_to_hex(&ce->oid),
487                                         state->force ? SUBMODULE_MOVE_HEAD_FORCE : 0);
488                 }
489
490                 if (!changed)
491                         return 0;
492                 if (!state->force) {
493                         if (!state->quiet)
494                                 fprintf(stderr,
495                                         "%s already exists, no checkout\n",
496                                         path.buf);
497                         return -1;
498                 }
499
500                 if (state->clone)
501                         mark_colliding_entries(state, ce, &st);
502
503                 /*
504                  * We unlink the old file, to get the new one with the
505                  * right permissions (including umask, which is nasty
506                  * to emulate by hand - much easier to let the system
507                  * just do the right thing)
508                  */
509                 if (S_ISDIR(st.st_mode)) {
510                         /* If it is a gitlink, leave it alone! */
511                         if (S_ISGITLINK(ce->ce_mode))
512                                 return 0;
513                         remove_subtree(&path);
514                 } else if (unlink(path.buf))
515                         return error_errno("unable to unlink old '%s'", path.buf);
516         } else if (state->not_new)
517                 return 0;
518
519         create_directories(path.buf, path.len, state);
520         if (nr_checkouts)
521                 (*nr_checkouts)++;
522         return write_entry(ce, path.buf, state, 0);
523 }
524
525 void unlink_entry(const struct cache_entry *ce)
526 {
527         const struct submodule *sub = submodule_from_ce(ce);
528         if (sub) {
529                 /* state.force is set at the caller. */
530                 submodule_move_head(ce->name, "HEAD", NULL,
531                                     SUBMODULE_MOVE_HEAD_FORCE);
532         }
533         if (!check_leading_path(ce->name, ce_namelen(ce)))
534                 return;
535         if (remove_or_warn(ce->ce_mode, ce->name))
536                 return;
537         schedule_dir_for_removal(ce->name, ce_namelen(ce));
538 }