]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
f4f15635 LP |
2 | /*** |
3 | This file is part of systemd. | |
4 | ||
5 | Copyright 2010 Lennart Poettering | |
f4f15635 LP |
6 | ***/ |
7 | ||
11c3a366 TA |
8 | #include <errno.h> |
9 | #include <stddef.h> | |
10 | #include <stdio.h> | |
11 | #include <stdlib.h> | |
12 | #include <string.h> | |
13 | #include <sys/stat.h> | |
655f2da0 | 14 | #include <linux/magic.h> |
11c3a366 TA |
15 | #include <time.h> |
16 | #include <unistd.h> | |
17 | ||
b5efdb8a | 18 | #include "alloc-util.h" |
f4f15635 LP |
19 | #include "dirent-util.h" |
20 | #include "fd-util.h" | |
21 | #include "fileio.h" | |
22 | #include "fs-util.h" | |
11c3a366 TA |
23 | #include "log.h" |
24 | #include "macro.h" | |
25 | #include "missing.h" | |
93cc7779 TA |
26 | #include "mkdir.h" |
27 | #include "parse-util.h" | |
28 | #include "path-util.h" | |
dccca82b | 29 | #include "process-util.h" |
34a8f081 | 30 | #include "stat-util.h" |
430fbf8e | 31 | #include "stdio-util.h" |
f4f15635 LP |
32 | #include "string-util.h" |
33 | #include "strv.h" | |
93cc7779 | 34 | #include "time-util.h" |
ee104e11 | 35 | #include "user-util.h" |
f4f15635 LP |
36 | #include "util.h" |
37 | ||
38 | int unlink_noerrno(const char *path) { | |
39 | PROTECT_ERRNO; | |
40 | int r; | |
41 | ||
42 | r = unlink(path); | |
43 | if (r < 0) | |
44 | return -errno; | |
45 | ||
46 | return 0; | |
47 | } | |
48 | ||
49 | int rmdir_parents(const char *path, const char *stop) { | |
50 | size_t l; | |
51 | int r = 0; | |
52 | ||
53 | assert(path); | |
54 | assert(stop); | |
55 | ||
56 | l = strlen(path); | |
57 | ||
58 | /* Skip trailing slashes */ | |
59 | while (l > 0 && path[l-1] == '/') | |
60 | l--; | |
61 | ||
62 | while (l > 0) { | |
63 | char *t; | |
64 | ||
65 | /* Skip last component */ | |
66 | while (l > 0 && path[l-1] != '/') | |
67 | l--; | |
68 | ||
69 | /* Skip trailing slashes */ | |
70 | while (l > 0 && path[l-1] == '/') | |
71 | l--; | |
72 | ||
73 | if (l <= 0) | |
74 | break; | |
75 | ||
76 | t = strndup(path, l); | |
77 | if (!t) | |
78 | return -ENOMEM; | |
79 | ||
80 | if (path_startswith(stop, t)) { | |
81 | free(t); | |
82 | return 0; | |
83 | } | |
84 | ||
85 | r = rmdir(t); | |
86 | free(t); | |
87 | ||
88 | if (r < 0) | |
89 | if (errno != ENOENT) | |
90 | return -errno; | |
91 | } | |
92 | ||
93 | return 0; | |
94 | } | |
95 | ||
f4f15635 LP |
96 | int rename_noreplace(int olddirfd, const char *oldpath, int newdirfd, const char *newpath) { |
97 | struct stat buf; | |
98 | int ret; | |
99 | ||
100 | ret = renameat2(olddirfd, oldpath, newdirfd, newpath, RENAME_NOREPLACE); | |
101 | if (ret >= 0) | |
102 | return 0; | |
103 | ||
104 | /* renameat2() exists since Linux 3.15, btrfs added support for it later. | |
105 | * If it is not implemented, fallback to another method. */ | |
106 | if (!IN_SET(errno, EINVAL, ENOSYS)) | |
107 | return -errno; | |
108 | ||
109 | /* The link()/unlink() fallback does not work on directories. But | |
110 | * renameat() without RENAME_NOREPLACE gives the same semantics on | |
111 | * directories, except when newpath is an *empty* directory. This is | |
112 | * good enough. */ | |
113 | ret = fstatat(olddirfd, oldpath, &buf, AT_SYMLINK_NOFOLLOW); | |
114 | if (ret >= 0 && S_ISDIR(buf.st_mode)) { | |
115 | ret = renameat(olddirfd, oldpath, newdirfd, newpath); | |
116 | return ret >= 0 ? 0 : -errno; | |
117 | } | |
118 | ||
119 | /* If it is not a directory, use the link()/unlink() fallback. */ | |
120 | ret = linkat(olddirfd, oldpath, newdirfd, newpath, 0); | |
121 | if (ret < 0) | |
122 | return -errno; | |
123 | ||
124 | ret = unlinkat(olddirfd, oldpath, 0); | |
125 | if (ret < 0) { | |
126 | /* backup errno before the following unlinkat() alters it */ | |
127 | ret = errno; | |
128 | (void) unlinkat(newdirfd, newpath, 0); | |
129 | errno = ret; | |
130 | return -errno; | |
131 | } | |
132 | ||
133 | return 0; | |
134 | } | |
135 | ||
136 | int readlinkat_malloc(int fd, const char *p, char **ret) { | |
137 | size_t l = 100; | |
138 | int r; | |
139 | ||
140 | assert(p); | |
141 | assert(ret); | |
142 | ||
143 | for (;;) { | |
144 | char *c; | |
145 | ssize_t n; | |
146 | ||
147 | c = new(char, l); | |
148 | if (!c) | |
149 | return -ENOMEM; | |
150 | ||
151 | n = readlinkat(fd, p, c, l-1); | |
152 | if (n < 0) { | |
153 | r = -errno; | |
154 | free(c); | |
155 | return r; | |
156 | } | |
157 | ||
158 | if ((size_t) n < l-1) { | |
159 | c[n] = 0; | |
160 | *ret = c; | |
161 | return 0; | |
162 | } | |
163 | ||
164 | free(c); | |
165 | l *= 2; | |
166 | } | |
167 | } | |
168 | ||
169 | int readlink_malloc(const char *p, char **ret) { | |
170 | return readlinkat_malloc(AT_FDCWD, p, ret); | |
171 | } | |
172 | ||
173 | int readlink_value(const char *p, char **ret) { | |
174 | _cleanup_free_ char *link = NULL; | |
175 | char *value; | |
176 | int r; | |
177 | ||
178 | r = readlink_malloc(p, &link); | |
179 | if (r < 0) | |
180 | return r; | |
181 | ||
182 | value = basename(link); | |
183 | if (!value) | |
184 | return -ENOENT; | |
185 | ||
186 | value = strdup(value); | |
187 | if (!value) | |
188 | return -ENOMEM; | |
189 | ||
190 | *ret = value; | |
191 | ||
192 | return 0; | |
193 | } | |
194 | ||
195 | int readlink_and_make_absolute(const char *p, char **r) { | |
196 | _cleanup_free_ char *target = NULL; | |
197 | char *k; | |
198 | int j; | |
199 | ||
200 | assert(p); | |
201 | assert(r); | |
202 | ||
203 | j = readlink_malloc(p, &target); | |
204 | if (j < 0) | |
205 | return j; | |
206 | ||
207 | k = file_in_same_dir(p, target); | |
208 | if (!k) | |
209 | return -ENOMEM; | |
210 | ||
211 | *r = k; | |
212 | return 0; | |
213 | } | |
214 | ||
f4f15635 LP |
215 | int chmod_and_chown(const char *path, mode_t mode, uid_t uid, gid_t gid) { |
216 | assert(path); | |
217 | ||
218 | /* Under the assumption that we are running privileged we | |
219 | * first change the access mode and only then hand out | |
220 | * ownership to avoid a window where access is too open. */ | |
221 | ||
222 | if (mode != MODE_INVALID) | |
223 | if (chmod(path, mode) < 0) | |
224 | return -errno; | |
225 | ||
226 | if (uid != UID_INVALID || gid != GID_INVALID) | |
227 | if (chown(path, uid, gid) < 0) | |
228 | return -errno; | |
229 | ||
230 | return 0; | |
231 | } | |
232 | ||
f4f15635 LP |
233 | int fchmod_umask(int fd, mode_t m) { |
234 | mode_t u; | |
235 | int r; | |
236 | ||
237 | u = umask(0777); | |
238 | r = fchmod(fd, m & (~u)) < 0 ? -errno : 0; | |
239 | umask(u); | |
240 | ||
241 | return r; | |
242 | } | |
243 | ||
244 | int fd_warn_permissions(const char *path, int fd) { | |
245 | struct stat st; | |
246 | ||
247 | if (fstat(fd, &st) < 0) | |
248 | return -errno; | |
249 | ||
250 | if (st.st_mode & 0111) | |
251 | log_warning("Configuration file %s is marked executable. Please remove executable permission bits. Proceeding anyway.", path); | |
252 | ||
253 | if (st.st_mode & 0002) | |
254 | log_warning("Configuration file %s is marked world-writable. Please remove world writability permission bits. Proceeding anyway.", path); | |
255 | ||
df0ff127 | 256 | if (getpid_cached() == 1 && (st.st_mode & 0044) != 0044) |
f4f15635 LP |
257 | log_warning("Configuration file %s is marked world-inaccessible. This has no effect as configuration data is accessible via APIs without restrictions. Proceeding anyway.", path); |
258 | ||
259 | return 0; | |
260 | } | |
261 | ||
262 | int touch_file(const char *path, bool parents, usec_t stamp, uid_t uid, gid_t gid, mode_t mode) { | |
9e3fa6e8 LP |
263 | char fdpath[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(int)]; |
264 | _cleanup_close_ int fd = -1; | |
265 | int r, ret = 0; | |
f4f15635 LP |
266 | |
267 | assert(path); | |
268 | ||
9e3fa6e8 LP |
269 | /* Note that touch_file() does not follow symlinks: if invoked on an existing symlink, then it is the symlink |
270 | * itself which is updated, not its target | |
271 | * | |
272 | * Returns the first error we encounter, but tries to apply as much as possible. */ | |
f4f15635 | 273 | |
9e3fa6e8 LP |
274 | if (parents) |
275 | (void) mkdir_parents(path, 0755); | |
276 | ||
277 | /* Initially, we try to open the node with O_PATH, so that we get a reference to the node. This is useful in | |
278 | * case the path refers to an existing device or socket node, as we can open it successfully in all cases, and | |
279 | * won't trigger any driver magic or so. */ | |
280 | fd = open(path, O_PATH|O_CLOEXEC|O_NOFOLLOW); | |
281 | if (fd < 0) { | |
282 | if (errno != ENOENT) | |
f4f15635 | 283 | return -errno; |
f4f15635 | 284 | |
9e3fa6e8 LP |
285 | /* if the node doesn't exist yet, we create it, but with O_EXCL, so that we only create a regular file |
286 | * here, and nothing else */ | |
287 | fd = open(path, O_WRONLY|O_CREAT|O_EXCL|O_CLOEXEC, IN_SET(mode, 0, MODE_INVALID) ? 0644 : mode); | |
288 | if (fd < 0) | |
f4f15635 LP |
289 | return -errno; |
290 | } | |
291 | ||
9e3fa6e8 LP |
292 | /* Let's make a path from the fd, and operate on that. With this logic, we can adjust the access mode, |
293 | * ownership and time of the file node in all cases, even if the fd refers to an O_PATH object — which is | |
294 | * something fchown(), fchmod(), futimensat() don't allow. */ | |
295 | xsprintf(fdpath, "/proc/self/fd/%i", fd); | |
296 | ||
297 | if (mode != MODE_INVALID) | |
298 | if (chmod(fdpath, mode) < 0) | |
299 | ret = -errno; | |
300 | ||
301 | if (uid_is_valid(uid) || gid_is_valid(gid)) | |
302 | if (chown(fdpath, uid, gid) < 0 && ret >= 0) | |
303 | ret = -errno; | |
304 | ||
f4f15635 LP |
305 | if (stamp != USEC_INFINITY) { |
306 | struct timespec ts[2]; | |
307 | ||
308 | timespec_store(&ts[0], stamp); | |
309 | ts[1] = ts[0]; | |
9e3fa6e8 | 310 | r = utimensat(AT_FDCWD, fdpath, ts, 0); |
f4f15635 | 311 | } else |
9e3fa6e8 LP |
312 | r = utimensat(AT_FDCWD, fdpath, NULL, 0); |
313 | if (r < 0 && ret >= 0) | |
f4f15635 LP |
314 | return -errno; |
315 | ||
9e3fa6e8 | 316 | return ret; |
f4f15635 LP |
317 | } |
318 | ||
319 | int touch(const char *path) { | |
ee735086 | 320 | return touch_file(path, false, USEC_INFINITY, UID_INVALID, GID_INVALID, MODE_INVALID); |
f4f15635 LP |
321 | } |
322 | ||
323 | int symlink_idempotent(const char *from, const char *to) { | |
f4f15635 LP |
324 | int r; |
325 | ||
326 | assert(from); | |
327 | assert(to); | |
328 | ||
329 | if (symlink(from, to) < 0) { | |
77b79723 LP |
330 | _cleanup_free_ char *p = NULL; |
331 | ||
f4f15635 LP |
332 | if (errno != EEXIST) |
333 | return -errno; | |
334 | ||
335 | r = readlink_malloc(to, &p); | |
77b79723 LP |
336 | if (r == -EINVAL) /* Not a symlink? In that case return the original error we encountered: -EEXIST */ |
337 | return -EEXIST; | |
338 | if (r < 0) /* Any other error? In that case propagate it as is */ | |
f4f15635 LP |
339 | return r; |
340 | ||
77b79723 LP |
341 | if (!streq(p, from)) /* Not the symlink we want it to be? In that case, propagate the original -EEXIST */ |
342 | return -EEXIST; | |
f4f15635 LP |
343 | } |
344 | ||
345 | return 0; | |
346 | } | |
347 | ||
348 | int symlink_atomic(const char *from, const char *to) { | |
349 | _cleanup_free_ char *t = NULL; | |
350 | int r; | |
351 | ||
352 | assert(from); | |
353 | assert(to); | |
354 | ||
355 | r = tempfn_random(to, NULL, &t); | |
356 | if (r < 0) | |
357 | return r; | |
358 | ||
359 | if (symlink(from, t) < 0) | |
360 | return -errno; | |
361 | ||
362 | if (rename(t, to) < 0) { | |
363 | unlink_noerrno(t); | |
364 | return -errno; | |
365 | } | |
366 | ||
367 | return 0; | |
368 | } | |
369 | ||
370 | int mknod_atomic(const char *path, mode_t mode, dev_t dev) { | |
371 | _cleanup_free_ char *t = NULL; | |
372 | int r; | |
373 | ||
374 | assert(path); | |
375 | ||
376 | r = tempfn_random(path, NULL, &t); | |
377 | if (r < 0) | |
378 | return r; | |
379 | ||
380 | if (mknod(t, mode, dev) < 0) | |
381 | return -errno; | |
382 | ||
383 | if (rename(t, path) < 0) { | |
384 | unlink_noerrno(t); | |
385 | return -errno; | |
386 | } | |
387 | ||
388 | return 0; | |
389 | } | |
390 | ||
391 | int mkfifo_atomic(const char *path, mode_t mode) { | |
392 | _cleanup_free_ char *t = NULL; | |
393 | int r; | |
394 | ||
395 | assert(path); | |
396 | ||
397 | r = tempfn_random(path, NULL, &t); | |
398 | if (r < 0) | |
399 | return r; | |
400 | ||
401 | if (mkfifo(t, mode) < 0) | |
402 | return -errno; | |
403 | ||
404 | if (rename(t, path) < 0) { | |
405 | unlink_noerrno(t); | |
406 | return -errno; | |
407 | } | |
408 | ||
409 | return 0; | |
410 | } | |
411 | ||
412 | int get_files_in_directory(const char *path, char ***list) { | |
413 | _cleanup_closedir_ DIR *d = NULL; | |
8fb3f009 | 414 | struct dirent *de; |
f4f15635 LP |
415 | size_t bufsize = 0, n = 0; |
416 | _cleanup_strv_free_ char **l = NULL; | |
417 | ||
418 | assert(path); | |
419 | ||
420 | /* Returns all files in a directory in *list, and the number | |
421 | * of files as return value. If list is NULL returns only the | |
422 | * number. */ | |
423 | ||
424 | d = opendir(path); | |
425 | if (!d) | |
426 | return -errno; | |
427 | ||
8fb3f009 | 428 | FOREACH_DIRENT_ALL(de, d, return -errno) { |
f4f15635 LP |
429 | dirent_ensure_type(d, de); |
430 | ||
431 | if (!dirent_is_file(de)) | |
432 | continue; | |
433 | ||
434 | if (list) { | |
435 | /* one extra slot is needed for the terminating NULL */ | |
436 | if (!GREEDY_REALLOC(l, bufsize, n + 2)) | |
437 | return -ENOMEM; | |
438 | ||
439 | l[n] = strdup(de->d_name); | |
440 | if (!l[n]) | |
441 | return -ENOMEM; | |
442 | ||
443 | l[++n] = NULL; | |
444 | } else | |
445 | n++; | |
446 | } | |
447 | ||
ae2a15bc LP |
448 | if (list) |
449 | *list = TAKE_PTR(l); | |
f4f15635 LP |
450 | |
451 | return n; | |
452 | } | |
430fbf8e | 453 | |
992e8f22 LP |
454 | static int getenv_tmp_dir(const char **ret_path) { |
455 | const char *n; | |
456 | int r, ret = 0; | |
34a8f081 | 457 | |
992e8f22 | 458 | assert(ret_path); |
34a8f081 | 459 | |
992e8f22 LP |
460 | /* We use the same order of environment variables python uses in tempfile.gettempdir(): |
461 | * https://docs.python.org/3/library/tempfile.html#tempfile.gettempdir */ | |
462 | FOREACH_STRING(n, "TMPDIR", "TEMP", "TMP") { | |
463 | const char *e; | |
464 | ||
465 | e = secure_getenv(n); | |
466 | if (!e) | |
467 | continue; | |
468 | if (!path_is_absolute(e)) { | |
469 | r = -ENOTDIR; | |
470 | goto next; | |
471 | } | |
99be45a4 | 472 | if (!path_is_normalized(e)) { |
992e8f22 LP |
473 | r = -EPERM; |
474 | goto next; | |
475 | } | |
476 | ||
477 | r = is_dir(e, true); | |
478 | if (r < 0) | |
479 | goto next; | |
480 | if (r == 0) { | |
481 | r = -ENOTDIR; | |
482 | goto next; | |
483 | } | |
484 | ||
485 | *ret_path = e; | |
486 | return 1; | |
487 | ||
488 | next: | |
489 | /* Remember first error, to make this more debuggable */ | |
490 | if (ret >= 0) | |
491 | ret = r; | |
34a8f081 OW |
492 | } |
493 | ||
992e8f22 LP |
494 | if (ret < 0) |
495 | return ret; | |
34a8f081 | 496 | |
992e8f22 LP |
497 | *ret_path = NULL; |
498 | return ret; | |
499 | } | |
34a8f081 | 500 | |
992e8f22 LP |
501 | static int tmp_dir_internal(const char *def, const char **ret) { |
502 | const char *e; | |
503 | int r, k; | |
504 | ||
505 | assert(def); | |
506 | assert(ret); | |
507 | ||
508 | r = getenv_tmp_dir(&e); | |
509 | if (r > 0) { | |
510 | *ret = e; | |
511 | return 0; | |
512 | } | |
513 | ||
514 | k = is_dir(def, true); | |
515 | if (k == 0) | |
516 | k = -ENOTDIR; | |
517 | if (k < 0) | |
518 | return r < 0 ? r : k; | |
519 | ||
520 | *ret = def; | |
34a8f081 OW |
521 | return 0; |
522 | } | |
523 | ||
992e8f22 LP |
524 | int var_tmp_dir(const char **ret) { |
525 | ||
526 | /* Returns the location for "larger" temporary files, that is backed by physical storage if available, and thus | |
527 | * even might survive a boot: /var/tmp. If $TMPDIR (or related environment variables) are set, its value is | |
528 | * returned preferably however. Note that both this function and tmp_dir() below are affected by $TMPDIR, | |
529 | * making it a variable that overrides all temporary file storage locations. */ | |
530 | ||
531 | return tmp_dir_internal("/var/tmp", ret); | |
532 | } | |
533 | ||
534 | int tmp_dir(const char **ret) { | |
535 | ||
536 | /* Similar to var_tmp_dir() above, but returns the location for "smaller" temporary files, which is usually | |
537 | * backed by an in-memory file system: /tmp. */ | |
538 | ||
539 | return tmp_dir_internal("/tmp", ret); | |
540 | } | |
541 | ||
af229d7a ZJS |
542 | int unlink_or_warn(const char *filename) { |
543 | if (unlink(filename) < 0 && errno != ENOENT) | |
544 | /* If the file doesn't exist and the fs simply was read-only (in which | |
545 | * case unlink() returns EROFS even if the file doesn't exist), don't | |
546 | * complain */ | |
547 | if (errno != EROFS || access(filename, F_OK) >= 0) | |
548 | return log_error_errno(errno, "Failed to remove \"%s\": %m", filename); | |
549 | ||
550 | return 0; | |
551 | } | |
552 | ||
430fbf8e | 553 | int inotify_add_watch_fd(int fd, int what, uint32_t mask) { |
fbd0b64f | 554 | char path[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(int) + 1]; |
430fbf8e LP |
555 | int r; |
556 | ||
557 | /* This is like inotify_add_watch(), except that the file to watch is not referenced by a path, but by an fd */ | |
558 | xsprintf(path, "/proc/self/fd/%i", what); | |
559 | ||
560 | r = inotify_add_watch(fd, path, mask); | |
561 | if (r < 0) | |
562 | return -errno; | |
563 | ||
564 | return r; | |
565 | } | |
d944dc95 | 566 | |
f14f1806 LP |
567 | static bool safe_transition(const struct stat *a, const struct stat *b) { |
568 | /* Returns true if the transition from a to b is safe, i.e. that we never transition from unprivileged to | |
569 | * privileged files or directories. Why bother? So that unprivileged code can't symlink to privileged files | |
570 | * making us believe we read something safe even though it isn't safe in the specific context we open it in. */ | |
571 | ||
572 | if (a->st_uid == 0) /* Transitioning from privileged to unprivileged is always fine */ | |
573 | return true; | |
574 | ||
575 | return a->st_uid == b->st_uid; /* Otherwise we need to stay within the same UID */ | |
576 | } | |
577 | ||
c4f4fce7 | 578 | int chase_symlinks(const char *path, const char *original_root, unsigned flags, char **ret) { |
d944dc95 LP |
579 | _cleanup_free_ char *buffer = NULL, *done = NULL, *root = NULL; |
580 | _cleanup_close_ int fd = -1; | |
581 | unsigned max_follow = 32; /* how many symlinks to follow before giving up and returning ELOOP */ | |
f14f1806 | 582 | struct stat previous_stat; |
a9fb0867 | 583 | bool exists = true; |
d944dc95 LP |
584 | char *todo; |
585 | int r; | |
586 | ||
587 | assert(path); | |
588 | ||
1ed34d75 LP |
589 | /* Either the file may be missing, or we return an fd to the final object, but both make no sense */ |
590 | if ((flags & (CHASE_NONEXISTENT|CHASE_OPEN)) == (CHASE_NONEXISTENT|CHASE_OPEN)) | |
591 | return -EINVAL; | |
592 | ||
a49424af LP |
593 | if (isempty(path)) |
594 | return -EINVAL; | |
595 | ||
d944dc95 LP |
596 | /* This is a lot like canonicalize_file_name(), but takes an additional "root" parameter, that allows following |
597 | * symlinks relative to a root directory, instead of the root of the host. | |
598 | * | |
fc4b68e5 | 599 | * Note that "root" primarily matters if we encounter an absolute symlink. It is also used when following |
c4f4fce7 LP |
600 | * relative symlinks to ensure they cannot be used to "escape" the root directory. The path parameter passed is |
601 | * assumed to be already prefixed by it, except if the CHASE_PREFIX_ROOT flag is set, in which case it is first | |
602 | * prefixed accordingly. | |
d944dc95 LP |
603 | * |
604 | * Algorithmically this operates on two path buffers: "done" are the components of the path we already | |
605 | * processed and resolved symlinks, "." and ".." of. "todo" are the components of the path we still need to | |
606 | * process. On each iteration, we move one component from "todo" to "done", processing it's special meaning | |
607 | * each time. The "todo" path always starts with at least one slash, the "done" path always ends in no | |
608 | * slash. We always keep an O_PATH fd to the component we are currently processing, thus keeping lookup races | |
fc4b68e5 LP |
609 | * at a minimum. |
610 | * | |
611 | * Suggested usage: whenever you want to canonicalize a path, use this function. Pass the absolute path you got | |
612 | * as-is: fully qualified and relative to your host's root. Optionally, specify the root parameter to tell this | |
613 | * function what to do when encountering a symlink with an absolute path as directory: prefix it by the | |
46e92680 | 614 | * specified path. */ |
d944dc95 | 615 | |
22bc57c5 | 616 | /* A root directory of "/" or "" is identical to none */ |
57ea45e1 | 617 | if (empty_or_root(original_root)) |
22bc57c5 | 618 | original_root = NULL; |
b1bfb848 | 619 | |
244d2f07 LP |
620 | if (!original_root && !ret && (flags & (CHASE_NONEXISTENT|CHASE_NO_AUTOFS|CHASE_SAFE|CHASE_OPEN)) == CHASE_OPEN) { |
621 | /* Shortcut the CHASE_OPEN case if the caller isn't interested in the actual path and has no root set | |
622 | * and doesn't care about any of the other special features we provide either. */ | |
623 | r = open(path, O_PATH|O_CLOEXEC); | |
624 | if (r < 0) | |
625 | return -errno; | |
626 | ||
627 | return r; | |
628 | } | |
629 | ||
c4f4fce7 LP |
630 | if (original_root) { |
631 | r = path_make_absolute_cwd(original_root, &root); | |
d944dc95 LP |
632 | if (r < 0) |
633 | return r; | |
c4f4fce7 | 634 | |
382a5078 LP |
635 | if (flags & CHASE_PREFIX_ROOT) { |
636 | ||
637 | /* We don't support relative paths in combination with a root directory */ | |
638 | if (!path_is_absolute(path)) | |
639 | return -EINVAL; | |
640 | ||
c4f4fce7 | 641 | path = prefix_roota(root, path); |
382a5078 | 642 | } |
d944dc95 LP |
643 | } |
644 | ||
c4f4fce7 LP |
645 | r = path_make_absolute_cwd(path, &buffer); |
646 | if (r < 0) | |
647 | return r; | |
648 | ||
d944dc95 LP |
649 | fd = open("/", O_CLOEXEC|O_NOFOLLOW|O_PATH); |
650 | if (fd < 0) | |
651 | return -errno; | |
652 | ||
f14f1806 LP |
653 | if (flags & CHASE_SAFE) { |
654 | if (fstat(fd, &previous_stat) < 0) | |
655 | return -errno; | |
656 | } | |
657 | ||
d944dc95 LP |
658 | todo = buffer; |
659 | for (;;) { | |
660 | _cleanup_free_ char *first = NULL; | |
661 | _cleanup_close_ int child = -1; | |
662 | struct stat st; | |
663 | size_t n, m; | |
664 | ||
665 | /* Determine length of first component in the path */ | |
666 | n = strspn(todo, "/"); /* The slashes */ | |
667 | m = n + strcspn(todo + n, "/"); /* The entire length of the component */ | |
668 | ||
669 | /* Extract the first component. */ | |
670 | first = strndup(todo, m); | |
671 | if (!first) | |
672 | return -ENOMEM; | |
673 | ||
674 | todo += m; | |
675 | ||
b12d25a8 ZJS |
676 | /* Empty? Then we reached the end. */ |
677 | if (isempty(first)) | |
678 | break; | |
679 | ||
d944dc95 | 680 | /* Just a single slash? Then we reached the end. */ |
b12d25a8 ZJS |
681 | if (path_equal(first, "/")) { |
682 | /* Preserve the trailing slash */ | |
62570f6f LP |
683 | |
684 | if (flags & CHASE_TRAIL_SLASH) | |
685 | if (!strextend(&done, "/", NULL)) | |
686 | return -ENOMEM; | |
b12d25a8 | 687 | |
d944dc95 | 688 | break; |
b12d25a8 | 689 | } |
d944dc95 LP |
690 | |
691 | /* Just a dot? Then let's eat this up. */ | |
692 | if (path_equal(first, "/.")) | |
693 | continue; | |
694 | ||
695 | /* Two dots? Then chop off the last bit of what we already found out. */ | |
696 | if (path_equal(first, "/..")) { | |
697 | _cleanup_free_ char *parent = NULL; | |
2b6d2dda | 698 | _cleanup_close_ int fd_parent = -1; |
d944dc95 | 699 | |
a4eaf3cf LP |
700 | /* If we already are at the top, then going up will not change anything. This is in-line with |
701 | * how the kernel handles this. */ | |
57ea45e1 | 702 | if (empty_or_root(done)) |
a4eaf3cf | 703 | continue; |
d944dc95 LP |
704 | |
705 | parent = dirname_malloc(done); | |
706 | if (!parent) | |
707 | return -ENOMEM; | |
708 | ||
a4eaf3cf | 709 | /* Don't allow this to leave the root dir. */ |
d944dc95 LP |
710 | if (root && |
711 | path_startswith(done, root) && | |
712 | !path_startswith(parent, root)) | |
a4eaf3cf | 713 | continue; |
d944dc95 | 714 | |
3b319885 | 715 | free_and_replace(done, parent); |
d944dc95 LP |
716 | |
717 | fd_parent = openat(fd, "..", O_CLOEXEC|O_NOFOLLOW|O_PATH); | |
718 | if (fd_parent < 0) | |
719 | return -errno; | |
720 | ||
f14f1806 LP |
721 | if (flags & CHASE_SAFE) { |
722 | if (fstat(fd_parent, &st) < 0) | |
723 | return -errno; | |
724 | ||
725 | if (!safe_transition(&previous_stat, &st)) | |
726 | return -EPERM; | |
727 | ||
728 | previous_stat = st; | |
729 | } | |
730 | ||
d944dc95 | 731 | safe_close(fd); |
c10d6bdb | 732 | fd = TAKE_FD(fd_parent); |
d944dc95 LP |
733 | |
734 | continue; | |
735 | } | |
736 | ||
737 | /* Otherwise let's see what this is. */ | |
738 | child = openat(fd, first + n, O_CLOEXEC|O_NOFOLLOW|O_PATH); | |
a9fb0867 LP |
739 | if (child < 0) { |
740 | ||
741 | if (errno == ENOENT && | |
cb638b5e | 742 | (flags & CHASE_NONEXISTENT) && |
99be45a4 | 743 | (isempty(todo) || path_is_normalized(todo))) { |
a9fb0867 | 744 | |
cb638b5e | 745 | /* If CHASE_NONEXISTENT is set, and the path does not exist, then that's OK, return |
a9fb0867 LP |
746 | * what we got so far. But don't allow this if the remaining path contains "../ or "./" |
747 | * or something else weird. */ | |
748 | ||
a1904a46 YW |
749 | /* If done is "/", as first also contains slash at the head, then remove this redundant slash. */ |
750 | if (streq_ptr(done, "/")) | |
751 | *done = '\0'; | |
752 | ||
a9fb0867 LP |
753 | if (!strextend(&done, first, todo, NULL)) |
754 | return -ENOMEM; | |
755 | ||
756 | exists = false; | |
757 | break; | |
758 | } | |
759 | ||
d944dc95 | 760 | return -errno; |
a9fb0867 | 761 | } |
d944dc95 LP |
762 | |
763 | if (fstat(child, &st) < 0) | |
764 | return -errno; | |
f14f1806 LP |
765 | if ((flags & CHASE_SAFE) && |
766 | !safe_transition(&previous_stat, &st)) | |
767 | return -EPERM; | |
768 | ||
769 | previous_stat = st; | |
770 | ||
655f2da0 | 771 | if ((flags & CHASE_NO_AUTOFS) && |
a66fee2e | 772 | fd_is_fs_type(child, AUTOFS_SUPER_MAGIC) > 0) |
655f2da0 | 773 | return -EREMOTE; |
d944dc95 LP |
774 | |
775 | if (S_ISLNK(st.st_mode)) { | |
877777d7 CCW |
776 | char *joined; |
777 | ||
d944dc95 LP |
778 | _cleanup_free_ char *destination = NULL; |
779 | ||
780 | /* This is a symlink, in this case read the destination. But let's make sure we don't follow | |
781 | * symlinks without bounds. */ | |
782 | if (--max_follow <= 0) | |
783 | return -ELOOP; | |
784 | ||
785 | r = readlinkat_malloc(fd, first + n, &destination); | |
786 | if (r < 0) | |
787 | return r; | |
788 | if (isempty(destination)) | |
789 | return -EINVAL; | |
790 | ||
791 | if (path_is_absolute(destination)) { | |
792 | ||
793 | /* An absolute destination. Start the loop from the beginning, but use the root | |
794 | * directory as base. */ | |
795 | ||
796 | safe_close(fd); | |
797 | fd = open(root ?: "/", O_CLOEXEC|O_NOFOLLOW|O_PATH); | |
798 | if (fd < 0) | |
799 | return -errno; | |
800 | ||
f14f1806 LP |
801 | if (flags & CHASE_SAFE) { |
802 | if (fstat(fd, &st) < 0) | |
803 | return -errno; | |
804 | ||
805 | if (!safe_transition(&previous_stat, &st)) | |
806 | return -EPERM; | |
807 | ||
808 | previous_stat = st; | |
809 | } | |
810 | ||
b539437a YW |
811 | free(done); |
812 | ||
d944dc95 LP |
813 | /* Note that we do not revalidate the root, we take it as is. */ |
814 | if (isempty(root)) | |
815 | done = NULL; | |
816 | else { | |
817 | done = strdup(root); | |
818 | if (!done) | |
819 | return -ENOMEM; | |
820 | } | |
821 | ||
8c4a8ea2 LP |
822 | /* Prefix what's left to do with what we just read, and start the loop again, but |
823 | * remain in the current directory. */ | |
824 | joined = strjoin(destination, todo); | |
825 | } else | |
826 | joined = strjoin("/", destination, todo); | |
877777d7 CCW |
827 | if (!joined) |
828 | return -ENOMEM; | |
d944dc95 | 829 | |
877777d7 CCW |
830 | free(buffer); |
831 | todo = buffer = joined; | |
d944dc95 LP |
832 | |
833 | continue; | |
834 | } | |
835 | ||
836 | /* If this is not a symlink, then let's just add the name we read to what we already verified. */ | |
ae2a15bc LP |
837 | if (!done) |
838 | done = TAKE_PTR(first); | |
839 | else { | |
a1904a46 YW |
840 | /* If done is "/", as first also contains slash at the head, then remove this redundant slash. */ |
841 | if (streq(done, "/")) | |
842 | *done = '\0'; | |
843 | ||
d944dc95 LP |
844 | if (!strextend(&done, first, NULL)) |
845 | return -ENOMEM; | |
846 | } | |
847 | ||
848 | /* And iterate again, but go one directory further down. */ | |
849 | safe_close(fd); | |
c10d6bdb | 850 | fd = TAKE_FD(child); |
d944dc95 LP |
851 | } |
852 | ||
853 | if (!done) { | |
854 | /* Special case, turn the empty string into "/", to indicate the root directory. */ | |
855 | done = strdup("/"); | |
856 | if (!done) | |
857 | return -ENOMEM; | |
858 | } | |
859 | ||
ae2a15bc LP |
860 | if (ret) |
861 | *ret = TAKE_PTR(done); | |
d944dc95 | 862 | |
1ed34d75 | 863 | if (flags & CHASE_OPEN) { |
1ed34d75 LP |
864 | /* Return the O_PATH fd we currently are looking to the caller. It can translate it to a proper fd by |
865 | * opening /proc/self/fd/xyz. */ | |
866 | ||
867 | assert(fd >= 0); | |
c10d6bdb | 868 | return TAKE_FD(fd); |
1ed34d75 LP |
869 | } |
870 | ||
a9fb0867 | 871 | return exists; |
d944dc95 | 872 | } |
57a4359e | 873 | |
21c692e9 LP |
874 | int chase_symlinks_and_open( |
875 | const char *path, | |
876 | const char *root, | |
877 | unsigned chase_flags, | |
878 | int open_flags, | |
879 | char **ret_path) { | |
880 | ||
881 | _cleanup_close_ int path_fd = -1; | |
882 | _cleanup_free_ char *p = NULL; | |
883 | int r; | |
884 | ||
885 | if (chase_flags & CHASE_NONEXISTENT) | |
886 | return -EINVAL; | |
887 | ||
57ea45e1 | 888 | if (empty_or_root(root) && !ret_path && (chase_flags & (CHASE_NO_AUTOFS|CHASE_SAFE)) == 0) { |
21c692e9 LP |
889 | /* Shortcut this call if none of the special features of this call are requested */ |
890 | r = open(path, open_flags); | |
891 | if (r < 0) | |
892 | return -errno; | |
893 | ||
894 | return r; | |
895 | } | |
896 | ||
897 | path_fd = chase_symlinks(path, root, chase_flags|CHASE_OPEN, ret_path ? &p : NULL); | |
898 | if (path_fd < 0) | |
899 | return path_fd; | |
900 | ||
901 | r = fd_reopen(path_fd, open_flags); | |
902 | if (r < 0) | |
903 | return r; | |
904 | ||
905 | if (ret_path) | |
906 | *ret_path = TAKE_PTR(p); | |
907 | ||
908 | return r; | |
909 | } | |
910 | ||
911 | int chase_symlinks_and_opendir( | |
912 | const char *path, | |
913 | const char *root, | |
914 | unsigned chase_flags, | |
915 | char **ret_path, | |
916 | DIR **ret_dir) { | |
917 | ||
918 | char procfs_path[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(int)]; | |
919 | _cleanup_close_ int path_fd = -1; | |
920 | _cleanup_free_ char *p = NULL; | |
921 | DIR *d; | |
922 | ||
923 | if (!ret_dir) | |
924 | return -EINVAL; | |
925 | if (chase_flags & CHASE_NONEXISTENT) | |
926 | return -EINVAL; | |
927 | ||
57ea45e1 | 928 | if (empty_or_root(root) && !ret_path && (chase_flags & (CHASE_NO_AUTOFS|CHASE_SAFE)) == 0) { |
21c692e9 LP |
929 | /* Shortcut this call if none of the special features of this call are requested */ |
930 | d = opendir(path); | |
931 | if (!d) | |
932 | return -errno; | |
933 | ||
934 | *ret_dir = d; | |
935 | return 0; | |
936 | } | |
937 | ||
938 | path_fd = chase_symlinks(path, root, chase_flags|CHASE_OPEN, ret_path ? &p : NULL); | |
939 | if (path_fd < 0) | |
940 | return path_fd; | |
941 | ||
942 | xsprintf(procfs_path, "/proc/self/fd/%i", path_fd); | |
943 | d = opendir(procfs_path); | |
944 | if (!d) | |
945 | return -errno; | |
946 | ||
947 | if (ret_path) | |
948 | *ret_path = TAKE_PTR(p); | |
949 | ||
950 | *ret_dir = d; | |
951 | return 0; | |
952 | } | |
953 | ||
d2bcd0ba LP |
954 | int chase_symlinks_and_stat( |
955 | const char *path, | |
956 | const char *root, | |
957 | unsigned chase_flags, | |
958 | char **ret_path, | |
959 | struct stat *ret_stat) { | |
960 | ||
961 | _cleanup_close_ int path_fd = -1; | |
962 | _cleanup_free_ char *p = NULL; | |
963 | ||
964 | assert(path); | |
965 | assert(ret_stat); | |
966 | ||
967 | if (chase_flags & CHASE_NONEXISTENT) | |
968 | return -EINVAL; | |
969 | ||
970 | if (empty_or_root(root) && !ret_path && (chase_flags & (CHASE_NO_AUTOFS|CHASE_SAFE)) == 0) { | |
971 | /* Shortcut this call if none of the special features of this call are requested */ | |
972 | if (stat(path, ret_stat) < 0) | |
973 | return -errno; | |
974 | ||
975 | return 1; | |
976 | } | |
977 | ||
978 | path_fd = chase_symlinks(path, root, chase_flags|CHASE_OPEN, ret_path ? &p : NULL); | |
979 | if (path_fd < 0) | |
980 | return path_fd; | |
981 | ||
982 | if (fstat(path_fd, ret_stat) < 0) | |
983 | return -errno; | |
984 | ||
985 | if (ret_path) | |
986 | *ret_path = TAKE_PTR(p); | |
987 | ||
988 | if (chase_flags & CHASE_OPEN) | |
989 | return TAKE_FD(path_fd); | |
990 | ||
991 | return 1; | |
992 | } | |
993 | ||
57a4359e | 994 | int access_fd(int fd, int mode) { |
fbd0b64f | 995 | char p[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(fd) + 1]; |
57a4359e LP |
996 | int r; |
997 | ||
998 | /* Like access() but operates on an already open fd */ | |
999 | ||
1000 | xsprintf(p, "/proc/self/fd/%i", fd); | |
57a4359e LP |
1001 | r = access(p, mode); |
1002 | if (r < 0) | |
21c692e9 | 1003 | return -errno; |
57a4359e LP |
1004 | |
1005 | return r; | |
1006 | } | |
43767d9d LP |
1007 | |
1008 | int unlinkat_deallocate(int fd, const char *name, int flags) { | |
1009 | _cleanup_close_ int truncate_fd = -1; | |
1010 | struct stat st; | |
1011 | off_t l, bs; | |
1012 | ||
1013 | /* Operates like unlinkat() but also deallocates the file contents if it is a regular file and there's no other | |
1014 | * link to it. This is useful to ensure that other processes that might have the file open for reading won't be | |
1015 | * able to keep the data pinned on disk forever. This call is particular useful whenever we execute clean-up | |
1016 | * jobs ("vacuuming"), where we want to make sure the data is really gone and the disk space released and | |
1017 | * returned to the free pool. | |
1018 | * | |
1019 | * Deallocation is preferably done by FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE (👊) if supported, which means | |
1020 | * the file won't change size. That's a good thing since we shouldn't needlessly trigger SIGBUS in other | |
1021 | * programs that have mmap()ed the file. (The assumption here is that changing file contents to all zeroes | |
1022 | * underneath those programs is the better choice than simply triggering SIGBUS in them which truncation does.) | |
1023 | * However if hole punching is not implemented in the kernel or file system we'll fall back to normal file | |
1024 | * truncation (🔪), as our goal of deallocating the data space trumps our goal of being nice to readers (💐). | |
1025 | * | |
1026 | * Note that we attempt deallocation, but failure to succeed with that is not considered fatal, as long as the | |
1027 | * primary job – to delete the file – is accomplished. */ | |
1028 | ||
1029 | if ((flags & AT_REMOVEDIR) == 0) { | |
1030 | truncate_fd = openat(fd, name, O_WRONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW|O_NONBLOCK); | |
1031 | if (truncate_fd < 0) { | |
1032 | ||
1033 | /* If this failed because the file doesn't exist propagate the error right-away. Also, | |
1034 | * AT_REMOVEDIR wasn't set, and we tried to open the file for writing, which means EISDIR is | |
1035 | * returned when this is a directory but we are not supposed to delete those, hence propagate | |
1036 | * the error right-away too. */ | |
1037 | if (IN_SET(errno, ENOENT, EISDIR)) | |
1038 | return -errno; | |
1039 | ||
1040 | if (errno != ELOOP) /* don't complain if this is a symlink */ | |
1041 | log_debug_errno(errno, "Failed to open file '%s' for deallocation, ignoring: %m", name); | |
1042 | } | |
1043 | } | |
1044 | ||
1045 | if (unlinkat(fd, name, flags) < 0) | |
1046 | return -errno; | |
1047 | ||
1048 | if (truncate_fd < 0) /* Don't have a file handle, can't do more ☹️ */ | |
1049 | return 0; | |
1050 | ||
1051 | if (fstat(truncate_fd, &st) < 0) { | |
1052 | log_debug_errno(errno, "Failed to stat file '%s' for deallocation, ignoring.", name); | |
1053 | return 0; | |
1054 | } | |
1055 | ||
1056 | if (!S_ISREG(st.st_mode) || st.st_blocks == 0 || st.st_nlink > 0) | |
1057 | return 0; | |
1058 | ||
1059 | /* If this is a regular file, it actually took up space on disk and there are no other links it's time to | |
1060 | * punch-hole/truncate this to release the disk space. */ | |
1061 | ||
1062 | bs = MAX(st.st_blksize, 512); | |
1063 | l = DIV_ROUND_UP(st.st_size, bs) * bs; /* Round up to next block size */ | |
1064 | ||
1065 | if (fallocate(truncate_fd, FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE, 0, l) >= 0) | |
1066 | return 0; /* Successfully punched a hole! 😊 */ | |
1067 | ||
1068 | /* Fall back to truncation */ | |
1069 | if (ftruncate(truncate_fd, 0) < 0) { | |
1070 | log_debug_errno(errno, "Failed to truncate file to 0, ignoring: %m"); | |
1071 | return 0; | |
1072 | } | |
1073 | ||
1074 | return 0; | |
1075 | } | |
11b29a96 LP |
1076 | |
1077 | int fsync_directory_of_file(int fd) { | |
1078 | _cleanup_free_ char *path = NULL, *dn = NULL; | |
1079 | _cleanup_close_ int dfd = -1; | |
1080 | int r; | |
1081 | ||
1082 | r = fd_verify_regular(fd); | |
1083 | if (r < 0) | |
1084 | return r; | |
1085 | ||
1086 | r = fd_get_path(fd, &path); | |
3ceae1bc ZJS |
1087 | if (r < 0) { |
1088 | log_debug("Failed to query /proc/self/fd/%d%s: %m", | |
1089 | fd, | |
1090 | r == -EOPNOTSUPP ? ", ignoring" : ""); | |
1091 | ||
1092 | if (r == -EOPNOTSUPP) | |
1093 | /* If /proc is not available, we're most likely running in some | |
1094 | * chroot environment, and syncing the directory is not very | |
1095 | * important in that case. Let's just silently do nothing. */ | |
1096 | return 0; | |
1097 | ||
11b29a96 | 1098 | return r; |
3ceae1bc | 1099 | } |
11b29a96 LP |
1100 | |
1101 | if (!path_is_absolute(path)) | |
1102 | return -EINVAL; | |
1103 | ||
1104 | dn = dirname_malloc(path); | |
1105 | if (!dn) | |
1106 | return -ENOMEM; | |
1107 | ||
1108 | dfd = open(dn, O_RDONLY|O_CLOEXEC|O_DIRECTORY); | |
1109 | if (dfd < 0) | |
1110 | return -errno; | |
1111 | ||
1112 | if (fsync(dfd) < 0) | |
1113 | return -errno; | |
1114 | ||
1115 | return 0; | |
1116 | } |