]> git.ipfire.org Git - thirdparty/git.git/blob - entry.c
unpack-trees: add basic support for parallel checkout
[thirdparty/git.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 #include "parallel-checkout.h"
11
12 static void create_directories(const char *path, int path_len,
13 const struct checkout *state)
14 {
15 char *buf = xmallocz(path_len);
16 int len = 0;
17
18 while (len < path_len) {
19 do {
20 buf[len] = path[len];
21 len++;
22 } while (len < path_len && path[len] != '/');
23 if (len >= path_len)
24 break;
25 buf[len] = 0;
26
27 /*
28 * For 'checkout-index --prefix=<dir>', <dir> is
29 * allowed to be a symlink to an existing directory,
30 * and we set 'state->base_dir_len' below, such that
31 * we test the path components of the prefix with the
32 * stat() function instead of the lstat() function.
33 */
34 if (has_dirs_only_path(buf, len, state->base_dir_len))
35 continue; /* ok, it is already a directory. */
36
37 /*
38 * If this mkdir() would fail, it could be that there
39 * is already a symlink or something else exists
40 * there, therefore we then try to unlink it and try
41 * one more time to create the directory.
42 */
43 if (mkdir(buf, 0777)) {
44 if (errno == EEXIST && state->force &&
45 !unlink_or_warn(buf) && !mkdir(buf, 0777))
46 continue;
47 die_errno("cannot create directory at '%s'", buf);
48 }
49 }
50 free(buf);
51 }
52
53 static void remove_subtree(struct strbuf *path)
54 {
55 DIR *dir = opendir(path->buf);
56 struct dirent *de;
57 int origlen = path->len;
58
59 if (!dir)
60 die_errno("cannot opendir '%s'", path->buf);
61 while ((de = readdir(dir)) != NULL) {
62 struct stat st;
63
64 if (is_dot_or_dotdot(de->d_name))
65 continue;
66
67 strbuf_addch(path, '/');
68 strbuf_addstr(path, de->d_name);
69 if (lstat(path->buf, &st))
70 die_errno("cannot lstat '%s'", path->buf);
71 if (S_ISDIR(st.st_mode))
72 remove_subtree(path);
73 else if (unlink(path->buf))
74 die_errno("cannot unlink '%s'", path->buf);
75 strbuf_setlen(path, origlen);
76 }
77 closedir(dir);
78 if (rmdir(path->buf))
79 die_errno("cannot rmdir '%s'", path->buf);
80 }
81
82 static int create_file(const char *path, unsigned int mode)
83 {
84 mode = (mode & 0100) ? 0777 : 0666;
85 return open(path, O_WRONLY | O_CREAT | O_EXCL, mode);
86 }
87
88 void *read_blob_entry(const struct cache_entry *ce, unsigned long *size)
89 {
90 enum object_type type;
91 void *blob_data = read_object_file(&ce->oid, &type, size);
92
93 if (blob_data) {
94 if (type == OBJ_BLOB)
95 return blob_data;
96 free(blob_data);
97 }
98 return NULL;
99 }
100
101 static int open_output_fd(char *path, const struct cache_entry *ce, int to_tempfile)
102 {
103 int symlink = (ce->ce_mode & S_IFMT) != S_IFREG;
104 if (to_tempfile) {
105 xsnprintf(path, TEMPORARY_FILENAME_LENGTH, "%s",
106 symlink ? ".merge_link_XXXXXX" : ".merge_file_XXXXXX");
107 return mkstemp(path);
108 } else {
109 return create_file(path, !symlink ? ce->ce_mode : 0666);
110 }
111 }
112
113 int fstat_checkout_output(int fd, const struct checkout *state, struct stat *st)
114 {
115 /* use fstat() only when path == ce->name */
116 if (fstat_is_reliable() &&
117 state->refresh_cache && !state->base_dir_len) {
118 return !fstat(fd, st);
119 }
120 return 0;
121 }
122
123 static int streaming_write_entry(const struct cache_entry *ce, char *path,
124 struct stream_filter *filter,
125 const struct checkout *state, int to_tempfile,
126 int *fstat_done, struct stat *statbuf)
127 {
128 int result = 0;
129 int fd;
130
131 fd = open_output_fd(path, ce, to_tempfile);
132 if (fd < 0)
133 return -1;
134
135 result |= stream_blob_to_fd(fd, &ce->oid, filter, 1);
136 *fstat_done = fstat_checkout_output(fd, state, statbuf);
137 result |= close(fd);
138
139 if (result)
140 unlink(path);
141 return result;
142 }
143
144 void enable_delayed_checkout(struct checkout *state)
145 {
146 if (!state->delayed_checkout) {
147 state->delayed_checkout = xmalloc(sizeof(*state->delayed_checkout));
148 state->delayed_checkout->state = CE_CAN_DELAY;
149 string_list_init(&state->delayed_checkout->filters, 0);
150 string_list_init(&state->delayed_checkout->paths, 0);
151 }
152 }
153
154 static int remove_available_paths(struct string_list_item *item, void *cb_data)
155 {
156 struct string_list *available_paths = cb_data;
157 struct string_list_item *available;
158
159 available = string_list_lookup(available_paths, item->string);
160 if (available)
161 available->util = (void *)item->string;
162 return !available;
163 }
164
165 int finish_delayed_checkout(struct checkout *state, int *nr_checkouts)
166 {
167 int errs = 0;
168 unsigned delayed_object_count;
169 off_t filtered_bytes = 0;
170 struct string_list_item *filter, *path;
171 struct progress *progress;
172 struct delayed_checkout *dco = state->delayed_checkout;
173
174 if (!state->delayed_checkout)
175 return errs;
176
177 dco->state = CE_RETRY;
178 delayed_object_count = dco->paths.nr;
179 progress = start_delayed_progress(_("Filtering content"), delayed_object_count);
180 while (dco->filters.nr > 0) {
181 for_each_string_list_item(filter, &dco->filters) {
182 struct string_list available_paths = STRING_LIST_INIT_NODUP;
183 display_progress(progress, delayed_object_count - dco->paths.nr);
184
185 if (!async_query_available_blobs(filter->string, &available_paths)) {
186 /* Filter reported an error */
187 errs = 1;
188 filter->string = "";
189 continue;
190 }
191 if (available_paths.nr <= 0) {
192 /*
193 * Filter responded with no entries. That means
194 * the filter is done and we can remove the
195 * filter from the list (see
196 * "string_list_remove_empty_items" call below).
197 */
198 filter->string = "";
199 continue;
200 }
201
202 /*
203 * In dco->paths we store a list of all delayed paths.
204 * The filter just send us a list of available paths.
205 * Remove them from the list.
206 */
207 filter_string_list(&dco->paths, 0,
208 &remove_available_paths, &available_paths);
209
210 for_each_string_list_item(path, &available_paths) {
211 struct cache_entry* ce;
212
213 if (!path->util) {
214 error("external filter '%s' signaled that '%s' "
215 "is now available although it has not been "
216 "delayed earlier",
217 filter->string, path->string);
218 errs |= 1;
219
220 /*
221 * Do not ask the filter for available blobs,
222 * again, as the filter is likely buggy.
223 */
224 filter->string = "";
225 continue;
226 }
227 ce = index_file_exists(state->istate, path->string,
228 strlen(path->string), 0);
229 if (ce) {
230 errs |= checkout_entry(ce, state, NULL, nr_checkouts);
231 filtered_bytes += ce->ce_stat_data.sd_size;
232 display_throughput(progress, filtered_bytes);
233 } else
234 errs = 1;
235 }
236 }
237 string_list_remove_empty_items(&dco->filters, 0);
238 }
239 stop_progress(&progress);
240 string_list_clear(&dco->filters, 0);
241
242 /* At this point we should not have any delayed paths anymore. */
243 errs |= dco->paths.nr;
244 for_each_string_list_item(path, &dco->paths) {
245 error("'%s' was not filtered properly", path->string);
246 }
247 string_list_clear(&dco->paths, 0);
248
249 free(dco);
250 state->delayed_checkout = NULL;
251
252 return errs;
253 }
254
255 void update_ce_after_write(const struct checkout *state, struct cache_entry *ce,
256 struct stat *st)
257 {
258 if (state->refresh_cache) {
259 assert(state->istate);
260 fill_stat_cache_info(state->istate, ce, st);
261 ce->ce_flags |= CE_UPDATE_IN_BASE;
262 mark_fsmonitor_invalid(state->istate, ce);
263 state->istate->cache_changed |= CE_ENTRY_CHANGED;
264 }
265 }
266
267 /* Note: ca is used (and required) iff the entry refers to a regular file. */
268 static int write_entry(struct cache_entry *ce, char *path, struct conv_attrs *ca,
269 const struct checkout *state, int to_tempfile)
270 {
271 unsigned int ce_mode_s_ifmt = ce->ce_mode & S_IFMT;
272 struct delayed_checkout *dco = state->delayed_checkout;
273 int fd, ret, fstat_done = 0;
274 char *new_blob;
275 struct strbuf buf = STRBUF_INIT;
276 unsigned long size;
277 ssize_t wrote;
278 size_t newsize = 0;
279 struct stat st;
280 const struct submodule *sub;
281 struct checkout_metadata meta;
282
283 clone_checkout_metadata(&meta, &state->meta, &ce->oid);
284
285 if (ce_mode_s_ifmt == S_IFREG) {
286 struct stream_filter *filter = get_stream_filter_ca(ca, &ce->oid);
287 if (filter &&
288 !streaming_write_entry(ce, path, filter,
289 state, to_tempfile,
290 &fstat_done, &st))
291 goto finish;
292 }
293
294 switch (ce_mode_s_ifmt) {
295 case S_IFLNK:
296 new_blob = read_blob_entry(ce, &size);
297 if (!new_blob)
298 return error("unable to read sha1 file of %s (%s)",
299 ce->name, oid_to_hex(&ce->oid));
300
301 /*
302 * We can't make a real symlink; write out a regular file entry
303 * with the symlink destination as its contents.
304 */
305 if (!has_symlinks || to_tempfile)
306 goto write_file_entry;
307
308 ret = symlink(new_blob, path);
309 free(new_blob);
310 if (ret)
311 return error_errno("unable to create symlink %s", path);
312 break;
313
314 case S_IFREG:
315 /*
316 * We do not send the blob in case of a retry, so do not
317 * bother reading it at all.
318 */
319 if (dco && dco->state == CE_RETRY) {
320 new_blob = NULL;
321 size = 0;
322 } else {
323 new_blob = read_blob_entry(ce, &size);
324 if (!new_blob)
325 return error("unable to read sha1 file of %s (%s)",
326 ce->name, oid_to_hex(&ce->oid));
327 }
328
329 /*
330 * Convert from git internal format to working tree format
331 */
332 if (dco && dco->state != CE_NO_DELAY) {
333 ret = async_convert_to_working_tree_ca(ca, ce->name,
334 new_blob, size,
335 &buf, &meta, dco);
336 if (ret && string_list_has_string(&dco->paths, ce->name)) {
337 free(new_blob);
338 goto delayed;
339 }
340 } else {
341 ret = convert_to_working_tree_ca(ca, ce->name, new_blob,
342 size, &buf, &meta);
343 }
344
345 if (ret) {
346 free(new_blob);
347 new_blob = strbuf_detach(&buf, &newsize);
348 size = newsize;
349 }
350 /*
351 * No "else" here as errors from convert are OK at this
352 * point. If the error would have been fatal (e.g.
353 * filter is required), then we would have died already.
354 */
355
356 write_file_entry:
357 fd = open_output_fd(path, ce, to_tempfile);
358 if (fd < 0) {
359 free(new_blob);
360 return error_errno("unable to create file %s", path);
361 }
362
363 wrote = write_in_full(fd, new_blob, size);
364 if (!to_tempfile)
365 fstat_done = fstat_checkout_output(fd, state, &st);
366 close(fd);
367 free(new_blob);
368 if (wrote < 0)
369 return error("unable to write file %s", path);
370 break;
371
372 case S_IFGITLINK:
373 if (to_tempfile)
374 return error("cannot create temporary submodule %s", ce->name);
375 if (mkdir(path, 0777) < 0)
376 return error("cannot create submodule directory %s", path);
377 sub = submodule_from_ce(ce);
378 if (sub)
379 return submodule_move_head(ce->name,
380 NULL, oid_to_hex(&ce->oid),
381 state->force ? SUBMODULE_MOVE_HEAD_FORCE : 0);
382 break;
383
384 default:
385 return error("unknown file mode for %s in index", ce->name);
386 }
387
388 finish:
389 if (state->refresh_cache) {
390 if (!fstat_done && lstat(ce->name, &st) < 0)
391 return error_errno("unable to stat just-written file %s",
392 ce->name);
393 update_ce_after_write(state, ce , &st);
394 }
395 delayed:
396 return 0;
397 }
398
399 /*
400 * This is like 'lstat()', except it refuses to follow symlinks
401 * in the path, after skipping "skiplen".
402 */
403 static int check_path(const char *path, int len, struct stat *st, int skiplen)
404 {
405 const char *slash = path + len;
406
407 while (path < slash && *slash != '/')
408 slash--;
409 if (!has_dirs_only_path(path, slash - path, skiplen)) {
410 errno = ENOENT;
411 return -1;
412 }
413 return lstat(path, st);
414 }
415
416 static void mark_colliding_entries(const struct checkout *state,
417 struct cache_entry *ce, struct stat *st)
418 {
419 int i, trust_ino = check_stat;
420
421 #if defined(GIT_WINDOWS_NATIVE) || defined(__CYGWIN__)
422 trust_ino = 0;
423 #endif
424
425 ce->ce_flags |= CE_MATCHED;
426
427 for (i = 0; i < state->istate->cache_nr; i++) {
428 struct cache_entry *dup = state->istate->cache[i];
429
430 if (dup == ce) {
431 /*
432 * Parallel checkout doesn't create the files in index
433 * order. So the other side of the collision may appear
434 * after the given cache_entry in the array.
435 */
436 if (parallel_checkout_status() == PC_RUNNING)
437 continue;
438 else
439 break;
440 }
441
442 if (dup->ce_flags & (CE_MATCHED | CE_VALID | CE_SKIP_WORKTREE))
443 continue;
444
445 if ((trust_ino && !match_stat_data(&dup->ce_stat_data, st)) ||
446 (!trust_ino && !fspathcmp(ce->name, dup->name))) {
447 dup->ce_flags |= CE_MATCHED;
448 break;
449 }
450 }
451 }
452
453 int checkout_entry_ca(struct cache_entry *ce, struct conv_attrs *ca,
454 const struct checkout *state, char *topath,
455 int *nr_checkouts)
456 {
457 static struct strbuf path = STRBUF_INIT;
458 struct stat st;
459 struct conv_attrs ca_buf;
460
461 if (ce->ce_flags & CE_WT_REMOVE) {
462 if (topath)
463 /*
464 * No content and thus no path to create, so we have
465 * no pathname to return.
466 */
467 BUG("Can't remove entry to a path");
468 unlink_entry(ce);
469 return 0;
470 }
471
472 if (topath) {
473 if (S_ISREG(ce->ce_mode) && !ca) {
474 convert_attrs(state->istate, &ca_buf, ce->name);
475 ca = &ca_buf;
476 }
477 return write_entry(ce, topath, ca, state, 1);
478 }
479
480 strbuf_reset(&path);
481 strbuf_add(&path, state->base_dir, state->base_dir_len);
482 strbuf_add(&path, ce->name, ce_namelen(ce));
483
484 if (!check_path(path.buf, path.len, &st, state->base_dir_len)) {
485 const struct submodule *sub;
486 unsigned changed = ie_match_stat(state->istate, ce, &st,
487 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
488 /*
489 * Needs to be checked before !changed returns early,
490 * as the possibly empty directory was not changed
491 */
492 sub = submodule_from_ce(ce);
493 if (sub) {
494 int err;
495 if (!is_submodule_populated_gently(ce->name, &err)) {
496 struct stat sb;
497 if (lstat(ce->name, &sb))
498 die(_("could not stat file '%s'"), ce->name);
499 if (!(st.st_mode & S_IFDIR))
500 unlink_or_warn(ce->name);
501
502 return submodule_move_head(ce->name,
503 NULL, oid_to_hex(&ce->oid), 0);
504 } else
505 return submodule_move_head(ce->name,
506 "HEAD", oid_to_hex(&ce->oid),
507 state->force ? SUBMODULE_MOVE_HEAD_FORCE : 0);
508 }
509
510 if (!changed)
511 return 0;
512 if (!state->force) {
513 if (!state->quiet)
514 fprintf(stderr,
515 "%s already exists, no checkout\n",
516 path.buf);
517 return -1;
518 }
519
520 if (state->clone)
521 mark_colliding_entries(state, ce, &st);
522
523 /*
524 * We unlink the old file, to get the new one with the
525 * right permissions (including umask, which is nasty
526 * to emulate by hand - much easier to let the system
527 * just do the right thing)
528 */
529 if (S_ISDIR(st.st_mode)) {
530 /* If it is a gitlink, leave it alone! */
531 if (S_ISGITLINK(ce->ce_mode))
532 return 0;
533 remove_subtree(&path);
534 } else if (unlink(path.buf))
535 return error_errno("unable to unlink old '%s'", path.buf);
536 } else if (state->not_new)
537 return 0;
538
539 create_directories(path.buf, path.len, state);
540
541 if (nr_checkouts)
542 (*nr_checkouts)++;
543
544 if (S_ISREG(ce->ce_mode) && !ca) {
545 convert_attrs(state->istate, &ca_buf, ce->name);
546 ca = &ca_buf;
547 }
548
549 if (!enqueue_checkout(ce, ca))
550 return 0;
551
552 return write_entry(ce, path.buf, ca, state, 0);
553 }
554
555 void unlink_entry(const struct cache_entry *ce)
556 {
557 const struct submodule *sub = submodule_from_ce(ce);
558 if (sub) {
559 /* state.force is set at the caller. */
560 submodule_move_head(ce->name, "HEAD", NULL,
561 SUBMODULE_MOVE_HEAD_FORCE);
562 }
563 if (check_leading_path(ce->name, ce_namelen(ce), 1) >= 0)
564 return;
565 if (remove_or_warn(ce->ce_mode, ce->name))
566 return;
567 schedule_dir_for_removal(ce->name, ce_namelen(ce));
568 }