tests: write commit-graph with Bloom filters
[git.git] / entry.c
blob00b49033668160574c35555293c0c755cbafaeaa
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"
10 static void create_directories(const char *path, int path_len,
11 const struct checkout *state)
13 char *buf = xmallocz(path_len);
14 int len = 0;
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;
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.
32 if (has_dirs_only_path(buf, len, state->base_dir_len))
33 continue; /* ok, it is already a directory. */
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.
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);
48 free(buf);
51 static void remove_subtree(struct strbuf *path)
53 DIR *dir = opendir(path->buf);
54 struct dirent *de;
55 int origlen = path->len;
57 if (!dir)
58 die_errno("cannot opendir '%s'", path->buf);
59 while ((de = readdir(dir)) != NULL) {
60 struct stat st;
62 if (is_dot_or_dotdot(de->d_name))
63 continue;
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);
75 closedir(dir);
76 if (rmdir(path->buf))
77 die_errno("cannot rmdir '%s'", path->buf);
80 static int create_file(const char *path, unsigned int mode)
82 mode = (mode & 0100) ? 0777 : 0666;
83 return open(path, O_WRONLY | O_CREAT | O_EXCL, mode);
86 static void *read_blob_entry(const struct cache_entry *ce, unsigned long *size)
88 enum object_type type;
89 void *blob_data = read_object_file(&ce->oid, &type, size);
91 if (blob_data) {
92 if (type == OBJ_BLOB)
93 return blob_data;
94 free(blob_data);
96 return NULL;
99 static int open_output_fd(char *path, const struct cache_entry *ce, int to_tempfile)
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);
111 static int fstat_output(int fd, const struct checkout *state, struct stat *st)
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;
119 return 0;
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)
127 int result = 0;
128 int fd;
130 fd = open_output_fd(path, ce, to_tempfile);
131 if (fd < 0)
132 return -1;
134 result |= stream_blob_to_fd(fd, &ce->oid, filter, 1);
135 *fstat_done = fstat_output(fd, state, statbuf);
136 result |= close(fd);
138 if (result)
139 unlink(path);
140 return result;
143 void enable_delayed_checkout(struct checkout *state)
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);
153 static int remove_available_paths(struct string_list_item *item, void *cb_data)
155 struct string_list *available_paths = cb_data;
156 struct string_list_item *available;
158 available = string_list_lookup(available_paths, item->string);
159 if (available)
160 available->util = (void *)item->string;
161 return !available;
164 int finish_delayed_checkout(struct checkout *state, int *nr_checkouts)
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;
173 if (!state->delayed_checkout)
174 return errs;
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);
184 if (!async_query_available_blobs(filter->string, &available_paths)) {
185 /* Filter reported an error */
186 errs = 1;
187 filter->string = "";
188 continue;
190 if (available_paths.nr <= 0) {
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).
197 filter->string = "";
198 continue;
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.
206 filter_string_list(&dco->paths, 0,
207 &remove_available_paths, &available_paths);
209 for_each_string_list_item(path, &available_paths) {
210 struct cache_entry* ce;
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;
220 * Do not ask the filter for available blobs,
221 * again, as the filter is likely buggy.
223 filter->string = "";
224 continue;
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;
236 string_list_remove_empty_items(&dco->filters, 0);
238 stop_progress(&progress);
239 string_list_clear(&dco->filters, 0);
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);
246 string_list_clear(&dco->paths, 0);
248 free(dco);
249 state->delayed_checkout = NULL;
251 return errs;
254 static int write_entry(struct cache_entry *ce,
255 char *path, const struct checkout *state, int to_tempfile)
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 struct checkout_metadata meta;
269 clone_checkout_metadata(&meta, &state->meta, &ce->oid);
271 if (ce_mode_s_ifmt == S_IFREG) {
272 struct stream_filter *filter = get_stream_filter(state->istate, ce->name,
273 &ce->oid);
274 if (filter &&
275 !streaming_write_entry(ce, path, filter,
276 state, to_tempfile,
277 &fstat_done, &st))
278 goto finish;
281 switch (ce_mode_s_ifmt) {
282 case S_IFLNK:
283 new_blob = read_blob_entry(ce, &size);
284 if (!new_blob)
285 return error("unable to read sha1 file of %s (%s)",
286 path, oid_to_hex(&ce->oid));
289 * We can't make a real symlink; write out a regular file entry
290 * with the symlink destination as its contents.
292 if (!has_symlinks || to_tempfile)
293 goto write_file_entry;
295 ret = symlink(new_blob, path);
296 free(new_blob);
297 if (ret)
298 return error_errno("unable to create symlink %s", path);
299 break;
301 case S_IFREG:
303 * We do not send the blob in case of a retry, so do not
304 * bother reading it at all.
306 if (dco && dco->state == CE_RETRY) {
307 new_blob = NULL;
308 size = 0;
309 } else {
310 new_blob = read_blob_entry(ce, &size);
311 if (!new_blob)
312 return error("unable to read sha1 file of %s (%s)",
313 path, oid_to_hex(&ce->oid));
317 * Convert from git internal format to working tree format
319 if (dco && dco->state != CE_NO_DELAY) {
320 ret = async_convert_to_working_tree(state->istate, ce->name, new_blob,
321 size, &buf, &meta, dco);
322 if (ret && string_list_has_string(&dco->paths, ce->name)) {
323 free(new_blob);
324 goto delayed;
326 } else
327 ret = convert_to_working_tree(state->istate, ce->name, new_blob, size, &buf, &meta);
329 if (ret) {
330 free(new_blob);
331 new_blob = strbuf_detach(&buf, &newsize);
332 size = newsize;
335 * No "else" here as errors from convert are OK at this
336 * point. If the error would have been fatal (e.g.
337 * filter is required), then we would have died already.
340 write_file_entry:
341 fd = open_output_fd(path, ce, to_tempfile);
342 if (fd < 0) {
343 free(new_blob);
344 return error_errno("unable to create file %s", path);
347 wrote = write_in_full(fd, new_blob, size);
348 if (!to_tempfile)
349 fstat_done = fstat_output(fd, state, &st);
350 close(fd);
351 free(new_blob);
352 if (wrote < 0)
353 return error("unable to write file %s", path);
354 break;
356 case S_IFGITLINK:
357 if (to_tempfile)
358 return error("cannot create temporary submodule %s", path);
359 if (mkdir(path, 0777) < 0)
360 return error("cannot create submodule directory %s", path);
361 sub = submodule_from_ce(ce);
362 if (sub)
363 return submodule_move_head(ce->name,
364 NULL, oid_to_hex(&ce->oid),
365 state->force ? SUBMODULE_MOVE_HEAD_FORCE : 0);
366 break;
368 default:
369 return error("unknown file mode for %s in index", path);
372 finish:
373 if (state->refresh_cache) {
374 assert(state->istate);
375 if (!fstat_done)
376 if (lstat(ce->name, &st) < 0)
377 return error_errno("unable to stat just-written file %s",
378 ce->name);
379 fill_stat_cache_info(state->istate, ce, &st);
380 ce->ce_flags |= CE_UPDATE_IN_BASE;
381 mark_fsmonitor_invalid(state->istate, ce);
382 state->istate->cache_changed |= CE_ENTRY_CHANGED;
384 delayed:
385 return 0;
389 * This is like 'lstat()', except it refuses to follow symlinks
390 * in the path, after skipping "skiplen".
392 static int check_path(const char *path, int len, struct stat *st, int skiplen)
394 const char *slash = path + len;
396 while (path < slash && *slash != '/')
397 slash--;
398 if (!has_dirs_only_path(path, slash - path, skiplen)) {
399 errno = ENOENT;
400 return -1;
402 return lstat(path, st);
405 static void mark_colliding_entries(const struct checkout *state,
406 struct cache_entry *ce, struct stat *st)
408 int i, trust_ino = check_stat;
410 #if defined(GIT_WINDOWS_NATIVE) || defined(__CYGWIN__)
411 trust_ino = 0;
412 #endif
414 ce->ce_flags |= CE_MATCHED;
416 for (i = 0; i < state->istate->cache_nr; i++) {
417 struct cache_entry *dup = state->istate->cache[i];
419 if (dup == ce)
420 break;
422 if (dup->ce_flags & (CE_MATCHED | CE_VALID | CE_SKIP_WORKTREE))
423 continue;
425 if ((trust_ino && !match_stat_data(&dup->ce_stat_data, st)) ||
426 (!trust_ino && !fspathcmp(ce->name, dup->name))) {
427 dup->ce_flags |= CE_MATCHED;
428 break;
434 * Write the contents from ce out to the working tree.
436 * When topath[] is not NULL, instead of writing to the working tree
437 * file named by ce, a temporary file is created by this function and
438 * its name is returned in topath[], which must be able to hold at
439 * least TEMPORARY_FILENAME_LENGTH bytes long.
441 int checkout_entry(struct cache_entry *ce, const struct checkout *state,
442 char *topath, int *nr_checkouts)
444 static struct strbuf path = STRBUF_INIT;
445 struct stat st;
447 if (ce->ce_flags & CE_WT_REMOVE) {
448 if (topath)
450 * No content and thus no path to create, so we have
451 * no pathname to return.
453 BUG("Can't remove entry to a path");
454 unlink_entry(ce);
455 return 0;
458 if (topath)
459 return write_entry(ce, topath, state, 1);
461 strbuf_reset(&path);
462 strbuf_add(&path, state->base_dir, state->base_dir_len);
463 strbuf_add(&path, ce->name, ce_namelen(ce));
465 if (!check_path(path.buf, path.len, &st, state->base_dir_len)) {
466 const struct submodule *sub;
467 unsigned changed = ie_match_stat(state->istate, ce, &st,
468 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
470 * Needs to be checked before !changed returns early,
471 * as the possibly empty directory was not changed
473 sub = submodule_from_ce(ce);
474 if (sub) {
475 int err;
476 if (!is_submodule_populated_gently(ce->name, &err)) {
477 struct stat sb;
478 if (lstat(ce->name, &sb))
479 die(_("could not stat file '%s'"), ce->name);
480 if (!(st.st_mode & S_IFDIR))
481 unlink_or_warn(ce->name);
483 return submodule_move_head(ce->name,
484 NULL, oid_to_hex(&ce->oid), 0);
485 } else
486 return submodule_move_head(ce->name,
487 "HEAD", oid_to_hex(&ce->oid),
488 state->force ? SUBMODULE_MOVE_HEAD_FORCE : 0);
491 if (!changed)
492 return 0;
493 if (!state->force) {
494 if (!state->quiet)
495 fprintf(stderr,
496 "%s already exists, no checkout\n",
497 path.buf);
498 return -1;
501 if (state->clone)
502 mark_colliding_entries(state, ce, &st);
505 * We unlink the old file, to get the new one with the
506 * right permissions (including umask, which is nasty
507 * to emulate by hand - much easier to let the system
508 * just do the right thing)
510 if (S_ISDIR(st.st_mode)) {
511 /* If it is a gitlink, leave it alone! */
512 if (S_ISGITLINK(ce->ce_mode))
513 return 0;
514 if (!state->force)
515 return error("%s is a directory", path.buf);
516 remove_subtree(&path);
517 } else if (unlink(path.buf))
518 return error_errno("unable to unlink old '%s'", path.buf);
519 } else if (state->not_new)
520 return 0;
522 create_directories(path.buf, path.len, state);
523 if (nr_checkouts)
524 (*nr_checkouts)++;
525 return write_entry(ce, path.buf, state, 0);
528 void unlink_entry(const struct cache_entry *ce)
530 const struct submodule *sub = submodule_from_ce(ce);
531 if (sub) {
532 /* state.force is set at the caller. */
533 submodule_move_head(ce->name, "HEAD", NULL,
534 SUBMODULE_MOVE_HEAD_FORCE);
536 if (!check_leading_path(ce->name, ce_namelen(ce)))
537 return;
538 if (remove_or_warn(ce->ce_mode, ce->name))
539 return;
540 schedule_dir_for_removal(ce->name, ce_namelen(ce));