]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
f4f15635 | 2 | |
11c3a366 TA |
3 | #include <errno.h> |
4 | #include <stddef.h> | |
5 | #include <stdio.h> | |
6 | #include <stdlib.h> | |
7 | #include <string.h> | |
8 | #include <sys/stat.h> | |
655f2da0 | 9 | #include <linux/magic.h> |
11c3a366 TA |
10 | #include <time.h> |
11 | #include <unistd.h> | |
12 | ||
b5efdb8a | 13 | #include "alloc-util.h" |
f4f15635 LP |
14 | #include "dirent-util.h" |
15 | #include "fd-util.h" | |
f4f15635 | 16 | #include "fs-util.h" |
11c3a366 TA |
17 | #include "log.h" |
18 | #include "macro.h" | |
19 | #include "missing.h" | |
93cc7779 TA |
20 | #include "mkdir.h" |
21 | #include "parse-util.h" | |
22 | #include "path-util.h" | |
dccca82b | 23 | #include "process-util.h" |
34a8f081 | 24 | #include "stat-util.h" |
430fbf8e | 25 | #include "stdio-util.h" |
f4f15635 LP |
26 | #include "string-util.h" |
27 | #include "strv.h" | |
93cc7779 | 28 | #include "time-util.h" |
e4de7287 | 29 | #include "tmpfile-util.h" |
ee104e11 | 30 | #include "user-util.h" |
f4f15635 LP |
31 | #include "util.h" |
32 | ||
33 | int unlink_noerrno(const char *path) { | |
34 | PROTECT_ERRNO; | |
35 | int r; | |
36 | ||
37 | r = unlink(path); | |
38 | if (r < 0) | |
39 | return -errno; | |
40 | ||
41 | return 0; | |
42 | } | |
43 | ||
44 | int rmdir_parents(const char *path, const char *stop) { | |
45 | size_t l; | |
46 | int r = 0; | |
47 | ||
48 | assert(path); | |
49 | assert(stop); | |
50 | ||
51 | l = strlen(path); | |
52 | ||
53 | /* Skip trailing slashes */ | |
54 | while (l > 0 && path[l-1] == '/') | |
55 | l--; | |
56 | ||
57 | while (l > 0) { | |
58 | char *t; | |
59 | ||
60 | /* Skip last component */ | |
61 | while (l > 0 && path[l-1] != '/') | |
62 | l--; | |
63 | ||
64 | /* Skip trailing slashes */ | |
65 | while (l > 0 && path[l-1] == '/') | |
66 | l--; | |
67 | ||
68 | if (l <= 0) | |
69 | break; | |
70 | ||
71 | t = strndup(path, l); | |
72 | if (!t) | |
73 | return -ENOMEM; | |
74 | ||
75 | if (path_startswith(stop, t)) { | |
76 | free(t); | |
77 | return 0; | |
78 | } | |
79 | ||
80 | r = rmdir(t); | |
81 | free(t); | |
82 | ||
83 | if (r < 0) | |
84 | if (errno != ENOENT) | |
85 | return -errno; | |
86 | } | |
87 | ||
88 | return 0; | |
89 | } | |
90 | ||
f4f15635 | 91 | int rename_noreplace(int olddirfd, const char *oldpath, int newdirfd, const char *newpath) { |
2f15b625 | 92 | int r; |
f4f15635 | 93 | |
2f15b625 LP |
94 | /* Try the ideal approach first */ |
95 | if (renameat2(olddirfd, oldpath, newdirfd, newpath, RENAME_NOREPLACE) >= 0) | |
f4f15635 LP |
96 | return 0; |
97 | ||
2f15b625 LP |
98 | /* renameat2() exists since Linux 3.15, btrfs and FAT added support for it later. If it is not implemented, |
99 | * fall back to a different method. */ | |
100 | if (!IN_SET(errno, EINVAL, ENOSYS, ENOTTY)) | |
f4f15635 LP |
101 | return -errno; |
102 | ||
2f15b625 LP |
103 | /* Let's try to use linkat()+unlinkat() as fallback. This doesn't work on directories and on some file systems |
104 | * that do not support hard links (such as FAT, most prominently), but for files it's pretty close to what we | |
105 | * want — though not atomic (i.e. for a short period both the new and the old filename will exist). */ | |
106 | if (linkat(olddirfd, oldpath, newdirfd, newpath, 0) >= 0) { | |
107 | ||
108 | if (unlinkat(olddirfd, oldpath, 0) < 0) { | |
109 | r = -errno; /* Backup errno before the following unlinkat() alters it */ | |
110 | (void) unlinkat(newdirfd, newpath, 0); | |
111 | return r; | |
112 | } | |
113 | ||
114 | return 0; | |
f4f15635 LP |
115 | } |
116 | ||
2f15b625 | 117 | if (!IN_SET(errno, EINVAL, ENOSYS, ENOTTY, EPERM)) /* FAT returns EPERM on link()… */ |
f4f15635 LP |
118 | return -errno; |
119 | ||
2f15b625 LP |
120 | /* OK, neither RENAME_NOREPLACE nor linkat()+unlinkat() worked. Let's then fallback to the racy TOCTOU |
121 | * vulnerable accessat(F_OK) check followed by classic, replacing renameat(), we have nothing better. */ | |
122 | ||
123 | if (faccessat(newdirfd, newpath, F_OK, AT_SYMLINK_NOFOLLOW) >= 0) | |
124 | return -EEXIST; | |
125 | if (errno != ENOENT) | |
126 | return -errno; | |
127 | ||
128 | if (renameat(olddirfd, oldpath, newdirfd, newpath) < 0) | |
f4f15635 | 129 | return -errno; |
f4f15635 LP |
130 | |
131 | return 0; | |
132 | } | |
133 | ||
134 | int readlinkat_malloc(int fd, const char *p, char **ret) { | |
8e060ec2 | 135 | size_t l = FILENAME_MAX+1; |
f4f15635 LP |
136 | int r; |
137 | ||
138 | assert(p); | |
139 | assert(ret); | |
140 | ||
141 | for (;;) { | |
142 | char *c; | |
143 | ssize_t n; | |
144 | ||
145 | c = new(char, l); | |
146 | if (!c) | |
147 | return -ENOMEM; | |
148 | ||
149 | n = readlinkat(fd, p, c, l-1); | |
150 | if (n < 0) { | |
151 | r = -errno; | |
152 | free(c); | |
153 | return r; | |
154 | } | |
155 | ||
156 | if ((size_t) n < l-1) { | |
157 | c[n] = 0; | |
158 | *ret = c; | |
159 | return 0; | |
160 | } | |
161 | ||
162 | free(c); | |
163 | l *= 2; | |
164 | } | |
165 | } | |
166 | ||
167 | int readlink_malloc(const char *p, char **ret) { | |
168 | return readlinkat_malloc(AT_FDCWD, p, ret); | |
169 | } | |
170 | ||
171 | int readlink_value(const char *p, char **ret) { | |
172 | _cleanup_free_ char *link = NULL; | |
173 | char *value; | |
174 | int r; | |
175 | ||
176 | r = readlink_malloc(p, &link); | |
177 | if (r < 0) | |
178 | return r; | |
179 | ||
180 | value = basename(link); | |
181 | if (!value) | |
182 | return -ENOENT; | |
183 | ||
184 | value = strdup(value); | |
185 | if (!value) | |
186 | return -ENOMEM; | |
187 | ||
188 | *ret = value; | |
189 | ||
190 | return 0; | |
191 | } | |
192 | ||
193 | int readlink_and_make_absolute(const char *p, char **r) { | |
194 | _cleanup_free_ char *target = NULL; | |
195 | char *k; | |
196 | int j; | |
197 | ||
198 | assert(p); | |
199 | assert(r); | |
200 | ||
201 | j = readlink_malloc(p, &target); | |
202 | if (j < 0) | |
203 | return j; | |
204 | ||
205 | k = file_in_same_dir(p, target); | |
206 | if (!k) | |
207 | return -ENOMEM; | |
208 | ||
209 | *r = k; | |
210 | return 0; | |
211 | } | |
212 | ||
f4f15635 | 213 | int chmod_and_chown(const char *path, mode_t mode, uid_t uid, gid_t gid) { |
de321f52 LP |
214 | char fd_path[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(int) + 1]; |
215 | _cleanup_close_ int fd = -1; | |
f4f15635 LP |
216 | assert(path); |
217 | ||
de321f52 | 218 | /* Under the assumption that we are running privileged we first change the access mode and only then hand out |
f4f15635 LP |
219 | * ownership to avoid a window where access is too open. */ |
220 | ||
de321f52 LP |
221 | fd = open(path, O_PATH|O_CLOEXEC|O_NOFOLLOW); /* Let's acquire an O_PATH fd, as precaution to change mode/owner |
222 | * on the same file */ | |
223 | if (fd < 0) | |
224 | return -errno; | |
225 | ||
226 | xsprintf(fd_path, "/proc/self/fd/%i", fd); | |
227 | ||
228 | if (mode != MODE_INVALID) { | |
229 | ||
230 | if ((mode & S_IFMT) != 0) { | |
231 | struct stat st; | |
232 | ||
233 | if (stat(fd_path, &st) < 0) | |
234 | return -errno; | |
235 | ||
236 | if ((mode & S_IFMT) != (st.st_mode & S_IFMT)) | |
237 | return -EINVAL; | |
238 | } | |
239 | ||
240 | if (chmod(fd_path, mode & 07777) < 0) | |
f4f15635 | 241 | return -errno; |
de321f52 | 242 | } |
f4f15635 LP |
243 | |
244 | if (uid != UID_INVALID || gid != GID_INVALID) | |
de321f52 | 245 | if (chown(fd_path, uid, gid) < 0) |
f4f15635 | 246 | return -errno; |
b8da477e YW |
247 | |
248 | return 0; | |
249 | } | |
250 | ||
251 | int fchmod_and_chown(int fd, mode_t mode, uid_t uid, gid_t gid) { | |
de321f52 | 252 | /* Under the assumption that we are running privileged we first change the access mode and only then hand out |
b8da477e YW |
253 | * ownership to avoid a window where access is too open. */ |
254 | ||
de321f52 LP |
255 | if (mode != MODE_INVALID) { |
256 | ||
257 | if ((mode & S_IFMT) != 0) { | |
258 | struct stat st; | |
259 | ||
260 | if (fstat(fd, &st) < 0) | |
261 | return -errno; | |
262 | ||
263 | if ((mode & S_IFMT) != (st.st_mode & S_IFMT)) | |
264 | return -EINVAL; | |
265 | } | |
266 | ||
267 | if (fchmod(fd, mode & 0777) < 0) | |
b8da477e | 268 | return -errno; |
de321f52 | 269 | } |
b8da477e YW |
270 | |
271 | if (uid != UID_INVALID || gid != GID_INVALID) | |
272 | if (fchown(fd, uid, gid) < 0) | |
273 | return -errno; | |
f4f15635 LP |
274 | |
275 | return 0; | |
276 | } | |
277 | ||
f4f15635 LP |
278 | int fchmod_umask(int fd, mode_t m) { |
279 | mode_t u; | |
280 | int r; | |
281 | ||
282 | u = umask(0777); | |
283 | r = fchmod(fd, m & (~u)) < 0 ? -errno : 0; | |
284 | umask(u); | |
285 | ||
286 | return r; | |
287 | } | |
288 | ||
4dfaa528 | 289 | int fchmod_opath(int fd, mode_t m) { |
22dd8d35 | 290 | char procfs_path[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(int)]; |
4dfaa528 FB |
291 | |
292 | /* This function operates also on fd that might have been opened with | |
293 | * O_PATH. Indeed fchmodat() doesn't have the AT_EMPTY_PATH flag like | |
294 | * fchownat() does. */ | |
295 | ||
296 | xsprintf(procfs_path, "/proc/self/fd/%i", fd); | |
4dfaa528 FB |
297 | if (chmod(procfs_path, m) < 0) |
298 | return -errno; | |
299 | ||
300 | return 0; | |
301 | } | |
302 | ||
f4f15635 LP |
303 | int fd_warn_permissions(const char *path, int fd) { |
304 | struct stat st; | |
305 | ||
306 | if (fstat(fd, &st) < 0) | |
307 | return -errno; | |
308 | ||
309 | if (st.st_mode & 0111) | |
310 | log_warning("Configuration file %s is marked executable. Please remove executable permission bits. Proceeding anyway.", path); | |
311 | ||
312 | if (st.st_mode & 0002) | |
313 | log_warning("Configuration file %s is marked world-writable. Please remove world writability permission bits. Proceeding anyway.", path); | |
314 | ||
df0ff127 | 315 | if (getpid_cached() == 1 && (st.st_mode & 0044) != 0044) |
f4f15635 LP |
316 | 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); |
317 | ||
318 | return 0; | |
319 | } | |
320 | ||
321 | int touch_file(const char *path, bool parents, usec_t stamp, uid_t uid, gid_t gid, mode_t mode) { | |
9e3fa6e8 LP |
322 | char fdpath[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(int)]; |
323 | _cleanup_close_ int fd = -1; | |
324 | int r, ret = 0; | |
f4f15635 LP |
325 | |
326 | assert(path); | |
327 | ||
9e3fa6e8 LP |
328 | /* Note that touch_file() does not follow symlinks: if invoked on an existing symlink, then it is the symlink |
329 | * itself which is updated, not its target | |
330 | * | |
331 | * Returns the first error we encounter, but tries to apply as much as possible. */ | |
f4f15635 | 332 | |
9e3fa6e8 LP |
333 | if (parents) |
334 | (void) mkdir_parents(path, 0755); | |
335 | ||
336 | /* Initially, we try to open the node with O_PATH, so that we get a reference to the node. This is useful in | |
337 | * case the path refers to an existing device or socket node, as we can open it successfully in all cases, and | |
338 | * won't trigger any driver magic or so. */ | |
339 | fd = open(path, O_PATH|O_CLOEXEC|O_NOFOLLOW); | |
340 | if (fd < 0) { | |
341 | if (errno != ENOENT) | |
f4f15635 | 342 | return -errno; |
f4f15635 | 343 | |
9e3fa6e8 LP |
344 | /* if the node doesn't exist yet, we create it, but with O_EXCL, so that we only create a regular file |
345 | * here, and nothing else */ | |
346 | fd = open(path, O_WRONLY|O_CREAT|O_EXCL|O_CLOEXEC, IN_SET(mode, 0, MODE_INVALID) ? 0644 : mode); | |
347 | if (fd < 0) | |
f4f15635 LP |
348 | return -errno; |
349 | } | |
350 | ||
9e3fa6e8 LP |
351 | /* Let's make a path from the fd, and operate on that. With this logic, we can adjust the access mode, |
352 | * ownership and time of the file node in all cases, even if the fd refers to an O_PATH object — which is | |
353 | * something fchown(), fchmod(), futimensat() don't allow. */ | |
354 | xsprintf(fdpath, "/proc/self/fd/%i", fd); | |
355 | ||
356 | if (mode != MODE_INVALID) | |
357 | if (chmod(fdpath, mode) < 0) | |
358 | ret = -errno; | |
359 | ||
360 | if (uid_is_valid(uid) || gid_is_valid(gid)) | |
361 | if (chown(fdpath, uid, gid) < 0 && ret >= 0) | |
362 | ret = -errno; | |
363 | ||
f4f15635 LP |
364 | if (stamp != USEC_INFINITY) { |
365 | struct timespec ts[2]; | |
366 | ||
367 | timespec_store(&ts[0], stamp); | |
368 | ts[1] = ts[0]; | |
9e3fa6e8 | 369 | r = utimensat(AT_FDCWD, fdpath, ts, 0); |
f4f15635 | 370 | } else |
9e3fa6e8 LP |
371 | r = utimensat(AT_FDCWD, fdpath, NULL, 0); |
372 | if (r < 0 && ret >= 0) | |
f4f15635 LP |
373 | return -errno; |
374 | ||
9e3fa6e8 | 375 | return ret; |
f4f15635 LP |
376 | } |
377 | ||
378 | int touch(const char *path) { | |
ee735086 | 379 | return touch_file(path, false, USEC_INFINITY, UID_INVALID, GID_INVALID, MODE_INVALID); |
f4f15635 LP |
380 | } |
381 | ||
6c9c51e5 YW |
382 | int symlink_idempotent(const char *from, const char *to, bool make_relative) { |
383 | _cleanup_free_ char *relpath = NULL; | |
f4f15635 LP |
384 | int r; |
385 | ||
386 | assert(from); | |
387 | assert(to); | |
388 | ||
6c9c51e5 YW |
389 | if (make_relative) { |
390 | _cleanup_free_ char *parent = NULL; | |
391 | ||
392 | parent = dirname_malloc(to); | |
393 | if (!parent) | |
394 | return -ENOMEM; | |
395 | ||
396 | r = path_make_relative(parent, from, &relpath); | |
397 | if (r < 0) | |
398 | return r; | |
399 | ||
400 | from = relpath; | |
401 | } | |
402 | ||
f4f15635 | 403 | if (symlink(from, to) < 0) { |
77b79723 LP |
404 | _cleanup_free_ char *p = NULL; |
405 | ||
f4f15635 LP |
406 | if (errno != EEXIST) |
407 | return -errno; | |
408 | ||
409 | r = readlink_malloc(to, &p); | |
77b79723 LP |
410 | if (r == -EINVAL) /* Not a symlink? In that case return the original error we encountered: -EEXIST */ |
411 | return -EEXIST; | |
412 | if (r < 0) /* Any other error? In that case propagate it as is */ | |
f4f15635 LP |
413 | return r; |
414 | ||
77b79723 LP |
415 | if (!streq(p, from)) /* Not the symlink we want it to be? In that case, propagate the original -EEXIST */ |
416 | return -EEXIST; | |
f4f15635 LP |
417 | } |
418 | ||
419 | return 0; | |
420 | } | |
421 | ||
422 | int symlink_atomic(const char *from, const char *to) { | |
423 | _cleanup_free_ char *t = NULL; | |
424 | int r; | |
425 | ||
426 | assert(from); | |
427 | assert(to); | |
428 | ||
429 | r = tempfn_random(to, NULL, &t); | |
430 | if (r < 0) | |
431 | return r; | |
432 | ||
433 | if (symlink(from, t) < 0) | |
434 | return -errno; | |
435 | ||
436 | if (rename(t, to) < 0) { | |
437 | unlink_noerrno(t); | |
438 | return -errno; | |
439 | } | |
440 | ||
441 | return 0; | |
442 | } | |
443 | ||
444 | int mknod_atomic(const char *path, mode_t mode, dev_t dev) { | |
445 | _cleanup_free_ char *t = NULL; | |
446 | int r; | |
447 | ||
448 | assert(path); | |
449 | ||
450 | r = tempfn_random(path, NULL, &t); | |
451 | if (r < 0) | |
452 | return r; | |
453 | ||
454 | if (mknod(t, mode, dev) < 0) | |
455 | return -errno; | |
456 | ||
457 | if (rename(t, path) < 0) { | |
458 | unlink_noerrno(t); | |
459 | return -errno; | |
460 | } | |
461 | ||
462 | return 0; | |
463 | } | |
464 | ||
465 | int mkfifo_atomic(const char *path, mode_t mode) { | |
466 | _cleanup_free_ char *t = NULL; | |
467 | int r; | |
468 | ||
469 | assert(path); | |
470 | ||
471 | r = tempfn_random(path, NULL, &t); | |
472 | if (r < 0) | |
473 | return r; | |
474 | ||
475 | if (mkfifo(t, mode) < 0) | |
476 | return -errno; | |
477 | ||
478 | if (rename(t, path) < 0) { | |
4fe3828c FB |
479 | unlink_noerrno(t); |
480 | return -errno; | |
481 | } | |
482 | ||
483 | return 0; | |
484 | } | |
485 | ||
486 | int mkfifoat_atomic(int dirfd, const char *path, mode_t mode) { | |
487 | _cleanup_free_ char *t = NULL; | |
488 | int r; | |
489 | ||
490 | assert(path); | |
491 | ||
492 | if (path_is_absolute(path)) | |
493 | return mkfifo_atomic(path, mode); | |
494 | ||
495 | /* We're only interested in the (random) filename. */ | |
496 | r = tempfn_random_child("", NULL, &t); | |
497 | if (r < 0) | |
498 | return r; | |
499 | ||
500 | if (mkfifoat(dirfd, t, mode) < 0) | |
501 | return -errno; | |
502 | ||
503 | if (renameat(dirfd, t, dirfd, path) < 0) { | |
f4f15635 LP |
504 | unlink_noerrno(t); |
505 | return -errno; | |
506 | } | |
507 | ||
508 | return 0; | |
509 | } | |
510 | ||
511 | int get_files_in_directory(const char *path, char ***list) { | |
512 | _cleanup_closedir_ DIR *d = NULL; | |
8fb3f009 | 513 | struct dirent *de; |
f4f15635 LP |
514 | size_t bufsize = 0, n = 0; |
515 | _cleanup_strv_free_ char **l = NULL; | |
516 | ||
517 | assert(path); | |
518 | ||
519 | /* Returns all files in a directory in *list, and the number | |
520 | * of files as return value. If list is NULL returns only the | |
521 | * number. */ | |
522 | ||
523 | d = opendir(path); | |
524 | if (!d) | |
525 | return -errno; | |
526 | ||
8fb3f009 | 527 | FOREACH_DIRENT_ALL(de, d, return -errno) { |
f4f15635 LP |
528 | dirent_ensure_type(d, de); |
529 | ||
530 | if (!dirent_is_file(de)) | |
531 | continue; | |
532 | ||
533 | if (list) { | |
534 | /* one extra slot is needed for the terminating NULL */ | |
535 | if (!GREEDY_REALLOC(l, bufsize, n + 2)) | |
536 | return -ENOMEM; | |
537 | ||
538 | l[n] = strdup(de->d_name); | |
539 | if (!l[n]) | |
540 | return -ENOMEM; | |
541 | ||
542 | l[++n] = NULL; | |
543 | } else | |
544 | n++; | |
545 | } | |
546 | ||
ae2a15bc LP |
547 | if (list) |
548 | *list = TAKE_PTR(l); | |
f4f15635 LP |
549 | |
550 | return n; | |
551 | } | |
430fbf8e | 552 | |
992e8f22 LP |
553 | static int getenv_tmp_dir(const char **ret_path) { |
554 | const char *n; | |
555 | int r, ret = 0; | |
34a8f081 | 556 | |
992e8f22 | 557 | assert(ret_path); |
34a8f081 | 558 | |
992e8f22 LP |
559 | /* We use the same order of environment variables python uses in tempfile.gettempdir(): |
560 | * https://docs.python.org/3/library/tempfile.html#tempfile.gettempdir */ | |
561 | FOREACH_STRING(n, "TMPDIR", "TEMP", "TMP") { | |
562 | const char *e; | |
563 | ||
564 | e = secure_getenv(n); | |
565 | if (!e) | |
566 | continue; | |
567 | if (!path_is_absolute(e)) { | |
568 | r = -ENOTDIR; | |
569 | goto next; | |
570 | } | |
99be45a4 | 571 | if (!path_is_normalized(e)) { |
992e8f22 LP |
572 | r = -EPERM; |
573 | goto next; | |
574 | } | |
575 | ||
576 | r = is_dir(e, true); | |
577 | if (r < 0) | |
578 | goto next; | |
579 | if (r == 0) { | |
580 | r = -ENOTDIR; | |
581 | goto next; | |
582 | } | |
583 | ||
584 | *ret_path = e; | |
585 | return 1; | |
586 | ||
587 | next: | |
588 | /* Remember first error, to make this more debuggable */ | |
589 | if (ret >= 0) | |
590 | ret = r; | |
34a8f081 OW |
591 | } |
592 | ||
992e8f22 LP |
593 | if (ret < 0) |
594 | return ret; | |
34a8f081 | 595 | |
992e8f22 LP |
596 | *ret_path = NULL; |
597 | return ret; | |
598 | } | |
34a8f081 | 599 | |
992e8f22 LP |
600 | static int tmp_dir_internal(const char *def, const char **ret) { |
601 | const char *e; | |
602 | int r, k; | |
603 | ||
604 | assert(def); | |
605 | assert(ret); | |
606 | ||
607 | r = getenv_tmp_dir(&e); | |
608 | if (r > 0) { | |
609 | *ret = e; | |
610 | return 0; | |
611 | } | |
612 | ||
613 | k = is_dir(def, true); | |
614 | if (k == 0) | |
615 | k = -ENOTDIR; | |
616 | if (k < 0) | |
617 | return r < 0 ? r : k; | |
618 | ||
619 | *ret = def; | |
34a8f081 OW |
620 | return 0; |
621 | } | |
622 | ||
992e8f22 LP |
623 | int var_tmp_dir(const char **ret) { |
624 | ||
625 | /* Returns the location for "larger" temporary files, that is backed by physical storage if available, and thus | |
626 | * even might survive a boot: /var/tmp. If $TMPDIR (or related environment variables) are set, its value is | |
627 | * returned preferably however. Note that both this function and tmp_dir() below are affected by $TMPDIR, | |
628 | * making it a variable that overrides all temporary file storage locations. */ | |
629 | ||
630 | return tmp_dir_internal("/var/tmp", ret); | |
631 | } | |
632 | ||
633 | int tmp_dir(const char **ret) { | |
634 | ||
635 | /* Similar to var_tmp_dir() above, but returns the location for "smaller" temporary files, which is usually | |
636 | * backed by an in-memory file system: /tmp. */ | |
637 | ||
638 | return tmp_dir_internal("/tmp", ret); | |
639 | } | |
640 | ||
af229d7a ZJS |
641 | int unlink_or_warn(const char *filename) { |
642 | if (unlink(filename) < 0 && errno != ENOENT) | |
643 | /* If the file doesn't exist and the fs simply was read-only (in which | |
644 | * case unlink() returns EROFS even if the file doesn't exist), don't | |
645 | * complain */ | |
646 | if (errno != EROFS || access(filename, F_OK) >= 0) | |
647 | return log_error_errno(errno, "Failed to remove \"%s\": %m", filename); | |
648 | ||
649 | return 0; | |
650 | } | |
651 | ||
430fbf8e | 652 | int inotify_add_watch_fd(int fd, int what, uint32_t mask) { |
fbd0b64f | 653 | char path[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(int) + 1]; |
430fbf8e LP |
654 | int r; |
655 | ||
656 | /* This is like inotify_add_watch(), except that the file to watch is not referenced by a path, but by an fd */ | |
657 | xsprintf(path, "/proc/self/fd/%i", what); | |
658 | ||
659 | r = inotify_add_watch(fd, path, mask); | |
660 | if (r < 0) | |
661 | return -errno; | |
662 | ||
663 | return r; | |
664 | } | |
d944dc95 | 665 | |
f14f1806 LP |
666 | static bool safe_transition(const struct stat *a, const struct stat *b) { |
667 | /* Returns true if the transition from a to b is safe, i.e. that we never transition from unprivileged to | |
668 | * privileged files or directories. Why bother? So that unprivileged code can't symlink to privileged files | |
669 | * making us believe we read something safe even though it isn't safe in the specific context we open it in. */ | |
670 | ||
671 | if (a->st_uid == 0) /* Transitioning from privileged to unprivileged is always fine */ | |
672 | return true; | |
673 | ||
674 | return a->st_uid == b->st_uid; /* Otherwise we need to stay within the same UID */ | |
675 | } | |
676 | ||
c4f4fce7 | 677 | int chase_symlinks(const char *path, const char *original_root, unsigned flags, char **ret) { |
d944dc95 LP |
678 | _cleanup_free_ char *buffer = NULL, *done = NULL, *root = NULL; |
679 | _cleanup_close_ int fd = -1; | |
f10f4215 | 680 | unsigned max_follow = CHASE_SYMLINKS_MAX; /* how many symlinks to follow before giving up and returning ELOOP */ |
f14f1806 | 681 | struct stat previous_stat; |
a9fb0867 | 682 | bool exists = true; |
d944dc95 LP |
683 | char *todo; |
684 | int r; | |
685 | ||
686 | assert(path); | |
687 | ||
1ed34d75 | 688 | /* Either the file may be missing, or we return an fd to the final object, but both make no sense */ |
d94a24ca | 689 | if (FLAGS_SET(flags, CHASE_NONEXISTENT | CHASE_OPEN)) |
1ed34d75 LP |
690 | return -EINVAL; |
691 | ||
d94a24ca | 692 | if (FLAGS_SET(flags, CHASE_STEP | CHASE_OPEN)) |
49eb3659 LP |
693 | return -EINVAL; |
694 | ||
a49424af LP |
695 | if (isempty(path)) |
696 | return -EINVAL; | |
697 | ||
d944dc95 LP |
698 | /* This is a lot like canonicalize_file_name(), but takes an additional "root" parameter, that allows following |
699 | * symlinks relative to a root directory, instead of the root of the host. | |
700 | * | |
fc4b68e5 | 701 | * Note that "root" primarily matters if we encounter an absolute symlink. It is also used when following |
c4f4fce7 LP |
702 | * relative symlinks to ensure they cannot be used to "escape" the root directory. The path parameter passed is |
703 | * assumed to be already prefixed by it, except if the CHASE_PREFIX_ROOT flag is set, in which case it is first | |
704 | * prefixed accordingly. | |
d944dc95 LP |
705 | * |
706 | * Algorithmically this operates on two path buffers: "done" are the components of the path we already | |
707 | * processed and resolved symlinks, "." and ".." of. "todo" are the components of the path we still need to | |
708 | * process. On each iteration, we move one component from "todo" to "done", processing it's special meaning | |
709 | * each time. The "todo" path always starts with at least one slash, the "done" path always ends in no | |
710 | * slash. We always keep an O_PATH fd to the component we are currently processing, thus keeping lookup races | |
fc4b68e5 LP |
711 | * at a minimum. |
712 | * | |
713 | * Suggested usage: whenever you want to canonicalize a path, use this function. Pass the absolute path you got | |
714 | * as-is: fully qualified and relative to your host's root. Optionally, specify the root parameter to tell this | |
715 | * function what to do when encountering a symlink with an absolute path as directory: prefix it by the | |
49eb3659 LP |
716 | * specified path. |
717 | * | |
718 | * There are three ways to invoke this function: | |
719 | * | |
720 | * 1. Without CHASE_STEP or CHASE_OPEN: in this case the path is resolved and the normalized path is returned | |
721 | * in `ret`. The return value is < 0 on error. If CHASE_NONEXISTENT is also set 0 is returned if the file | |
722 | * doesn't exist, > 0 otherwise. If CHASE_NONEXISTENT is not set >= 0 is returned if the destination was | |
723 | * found, -ENOENT if it doesn't. | |
724 | * | |
725 | * 2. With CHASE_OPEN: in this case the destination is opened after chasing it as O_PATH and this file | |
726 | * descriptor is returned as return value. This is useful to open files relative to some root | |
727 | * directory. Note that the returned O_PATH file descriptors must be converted into a regular one (using | |
728 | * fd_reopen() or such) before it can be used for reading/writing. CHASE_OPEN may not be combined with | |
729 | * CHASE_NONEXISTENT. | |
730 | * | |
731 | * 3. With CHASE_STEP: in this case only a single step of the normalization is executed, i.e. only the first | |
732 | * symlink or ".." component of the path is resolved, and the resulting path is returned. This is useful if | |
733 | * a caller wants to trace the a path through the file system verbosely. Returns < 0 on error, > 0 if the | |
734 | * path is fully normalized, and == 0 for each normalization step. This may be combined with | |
735 | * CHASE_NONEXISTENT, in which case 1 is returned when a component is not found. | |
736 | * | |
737 | * */ | |
d944dc95 | 738 | |
22bc57c5 | 739 | /* A root directory of "/" or "" is identical to none */ |
57ea45e1 | 740 | if (empty_or_root(original_root)) |
22bc57c5 | 741 | original_root = NULL; |
b1bfb848 | 742 | |
49eb3659 | 743 | if (!original_root && !ret && (flags & (CHASE_NONEXISTENT|CHASE_NO_AUTOFS|CHASE_SAFE|CHASE_OPEN|CHASE_STEP)) == CHASE_OPEN) { |
244d2f07 LP |
744 | /* Shortcut the CHASE_OPEN case if the caller isn't interested in the actual path and has no root set |
745 | * and doesn't care about any of the other special features we provide either. */ | |
1f56e4ce | 746 | r = open(path, O_PATH|O_CLOEXEC|((flags & CHASE_NOFOLLOW) ? O_NOFOLLOW : 0)); |
244d2f07 LP |
747 | if (r < 0) |
748 | return -errno; | |
749 | ||
750 | return r; | |
751 | } | |
752 | ||
c4f4fce7 LP |
753 | if (original_root) { |
754 | r = path_make_absolute_cwd(original_root, &root); | |
d944dc95 LP |
755 | if (r < 0) |
756 | return r; | |
c4f4fce7 | 757 | |
382a5078 LP |
758 | if (flags & CHASE_PREFIX_ROOT) { |
759 | ||
760 | /* We don't support relative paths in combination with a root directory */ | |
761 | if (!path_is_absolute(path)) | |
762 | return -EINVAL; | |
763 | ||
c4f4fce7 | 764 | path = prefix_roota(root, path); |
382a5078 | 765 | } |
d944dc95 LP |
766 | } |
767 | ||
c4f4fce7 LP |
768 | r = path_make_absolute_cwd(path, &buffer); |
769 | if (r < 0) | |
770 | return r; | |
771 | ||
d944dc95 LP |
772 | fd = open("/", O_CLOEXEC|O_NOFOLLOW|O_PATH); |
773 | if (fd < 0) | |
774 | return -errno; | |
775 | ||
f14f1806 LP |
776 | if (flags & CHASE_SAFE) { |
777 | if (fstat(fd, &previous_stat) < 0) | |
778 | return -errno; | |
779 | } | |
780 | ||
d944dc95 LP |
781 | todo = buffer; |
782 | for (;;) { | |
783 | _cleanup_free_ char *first = NULL; | |
784 | _cleanup_close_ int child = -1; | |
785 | struct stat st; | |
786 | size_t n, m; | |
787 | ||
788 | /* Determine length of first component in the path */ | |
789 | n = strspn(todo, "/"); /* The slashes */ | |
790 | m = n + strcspn(todo + n, "/"); /* The entire length of the component */ | |
791 | ||
792 | /* Extract the first component. */ | |
793 | first = strndup(todo, m); | |
794 | if (!first) | |
795 | return -ENOMEM; | |
796 | ||
797 | todo += m; | |
798 | ||
b12d25a8 ZJS |
799 | /* Empty? Then we reached the end. */ |
800 | if (isempty(first)) | |
801 | break; | |
802 | ||
d944dc95 | 803 | /* Just a single slash? Then we reached the end. */ |
b12d25a8 ZJS |
804 | if (path_equal(first, "/")) { |
805 | /* Preserve the trailing slash */ | |
62570f6f LP |
806 | |
807 | if (flags & CHASE_TRAIL_SLASH) | |
808 | if (!strextend(&done, "/", NULL)) | |
809 | return -ENOMEM; | |
b12d25a8 | 810 | |
d944dc95 | 811 | break; |
b12d25a8 | 812 | } |
d944dc95 LP |
813 | |
814 | /* Just a dot? Then let's eat this up. */ | |
815 | if (path_equal(first, "/.")) | |
816 | continue; | |
817 | ||
818 | /* Two dots? Then chop off the last bit of what we already found out. */ | |
819 | if (path_equal(first, "/..")) { | |
820 | _cleanup_free_ char *parent = NULL; | |
2b6d2dda | 821 | _cleanup_close_ int fd_parent = -1; |
d944dc95 | 822 | |
a4eaf3cf LP |
823 | /* If we already are at the top, then going up will not change anything. This is in-line with |
824 | * how the kernel handles this. */ | |
57ea45e1 | 825 | if (empty_or_root(done)) |
a4eaf3cf | 826 | continue; |
d944dc95 LP |
827 | |
828 | parent = dirname_malloc(done); | |
829 | if (!parent) | |
830 | return -ENOMEM; | |
831 | ||
a4eaf3cf | 832 | /* Don't allow this to leave the root dir. */ |
d944dc95 LP |
833 | if (root && |
834 | path_startswith(done, root) && | |
835 | !path_startswith(parent, root)) | |
a4eaf3cf | 836 | continue; |
d944dc95 | 837 | |
3b319885 | 838 | free_and_replace(done, parent); |
d944dc95 | 839 | |
49eb3659 LP |
840 | if (flags & CHASE_STEP) |
841 | goto chased_one; | |
842 | ||
d944dc95 LP |
843 | fd_parent = openat(fd, "..", O_CLOEXEC|O_NOFOLLOW|O_PATH); |
844 | if (fd_parent < 0) | |
845 | return -errno; | |
846 | ||
f14f1806 LP |
847 | if (flags & CHASE_SAFE) { |
848 | if (fstat(fd_parent, &st) < 0) | |
849 | return -errno; | |
850 | ||
851 | if (!safe_transition(&previous_stat, &st)) | |
852 | return -EPERM; | |
853 | ||
854 | previous_stat = st; | |
855 | } | |
856 | ||
d944dc95 | 857 | safe_close(fd); |
c10d6bdb | 858 | fd = TAKE_FD(fd_parent); |
d944dc95 LP |
859 | |
860 | continue; | |
861 | } | |
862 | ||
863 | /* Otherwise let's see what this is. */ | |
864 | child = openat(fd, first + n, O_CLOEXEC|O_NOFOLLOW|O_PATH); | |
a9fb0867 LP |
865 | if (child < 0) { |
866 | ||
867 | if (errno == ENOENT && | |
cb638b5e | 868 | (flags & CHASE_NONEXISTENT) && |
99be45a4 | 869 | (isempty(todo) || path_is_normalized(todo))) { |
a9fb0867 | 870 | |
cb638b5e | 871 | /* If CHASE_NONEXISTENT is set, and the path does not exist, then that's OK, return |
a9fb0867 LP |
872 | * what we got so far. But don't allow this if the remaining path contains "../ or "./" |
873 | * or something else weird. */ | |
874 | ||
a1904a46 YW |
875 | /* If done is "/", as first also contains slash at the head, then remove this redundant slash. */ |
876 | if (streq_ptr(done, "/")) | |
877 | *done = '\0'; | |
878 | ||
a9fb0867 LP |
879 | if (!strextend(&done, first, todo, NULL)) |
880 | return -ENOMEM; | |
881 | ||
882 | exists = false; | |
883 | break; | |
884 | } | |
885 | ||
d944dc95 | 886 | return -errno; |
a9fb0867 | 887 | } |
d944dc95 LP |
888 | |
889 | if (fstat(child, &st) < 0) | |
890 | return -errno; | |
f14f1806 LP |
891 | if ((flags & CHASE_SAFE) && |
892 | !safe_transition(&previous_stat, &st)) | |
893 | return -EPERM; | |
894 | ||
895 | previous_stat = st; | |
896 | ||
655f2da0 | 897 | if ((flags & CHASE_NO_AUTOFS) && |
a66fee2e | 898 | fd_is_fs_type(child, AUTOFS_SUPER_MAGIC) > 0) |
655f2da0 | 899 | return -EREMOTE; |
d944dc95 | 900 | |
1f56e4ce | 901 | if (S_ISLNK(st.st_mode) && !((flags & CHASE_NOFOLLOW) && isempty(todo))) { |
877777d7 CCW |
902 | char *joined; |
903 | ||
d944dc95 LP |
904 | _cleanup_free_ char *destination = NULL; |
905 | ||
906 | /* This is a symlink, in this case read the destination. But let's make sure we don't follow | |
907 | * symlinks without bounds. */ | |
908 | if (--max_follow <= 0) | |
909 | return -ELOOP; | |
910 | ||
911 | r = readlinkat_malloc(fd, first + n, &destination); | |
912 | if (r < 0) | |
913 | return r; | |
914 | if (isempty(destination)) | |
915 | return -EINVAL; | |
916 | ||
917 | if (path_is_absolute(destination)) { | |
918 | ||
919 | /* An absolute destination. Start the loop from the beginning, but use the root | |
920 | * directory as base. */ | |
921 | ||
922 | safe_close(fd); | |
923 | fd = open(root ?: "/", O_CLOEXEC|O_NOFOLLOW|O_PATH); | |
924 | if (fd < 0) | |
925 | return -errno; | |
926 | ||
f14f1806 LP |
927 | if (flags & CHASE_SAFE) { |
928 | if (fstat(fd, &st) < 0) | |
929 | return -errno; | |
930 | ||
931 | if (!safe_transition(&previous_stat, &st)) | |
932 | return -EPERM; | |
933 | ||
934 | previous_stat = st; | |
935 | } | |
936 | ||
b539437a YW |
937 | free(done); |
938 | ||
d944dc95 LP |
939 | /* Note that we do not revalidate the root, we take it as is. */ |
940 | if (isempty(root)) | |
941 | done = NULL; | |
942 | else { | |
943 | done = strdup(root); | |
944 | if (!done) | |
945 | return -ENOMEM; | |
946 | } | |
947 | ||
8c4a8ea2 LP |
948 | /* Prefix what's left to do with what we just read, and start the loop again, but |
949 | * remain in the current directory. */ | |
950 | joined = strjoin(destination, todo); | |
951 | } else | |
952 | joined = strjoin("/", destination, todo); | |
877777d7 CCW |
953 | if (!joined) |
954 | return -ENOMEM; | |
d944dc95 | 955 | |
877777d7 CCW |
956 | free(buffer); |
957 | todo = buffer = joined; | |
d944dc95 | 958 | |
49eb3659 LP |
959 | if (flags & CHASE_STEP) |
960 | goto chased_one; | |
961 | ||
d944dc95 LP |
962 | continue; |
963 | } | |
964 | ||
965 | /* If this is not a symlink, then let's just add the name we read to what we already verified. */ | |
ae2a15bc LP |
966 | if (!done) |
967 | done = TAKE_PTR(first); | |
968 | else { | |
a1904a46 YW |
969 | /* If done is "/", as first also contains slash at the head, then remove this redundant slash. */ |
970 | if (streq(done, "/")) | |
971 | *done = '\0'; | |
972 | ||
d944dc95 LP |
973 | if (!strextend(&done, first, NULL)) |
974 | return -ENOMEM; | |
975 | } | |
976 | ||
977 | /* And iterate again, but go one directory further down. */ | |
978 | safe_close(fd); | |
c10d6bdb | 979 | fd = TAKE_FD(child); |
d944dc95 LP |
980 | } |
981 | ||
982 | if (!done) { | |
983 | /* Special case, turn the empty string into "/", to indicate the root directory. */ | |
984 | done = strdup("/"); | |
985 | if (!done) | |
986 | return -ENOMEM; | |
987 | } | |
988 | ||
ae2a15bc LP |
989 | if (ret) |
990 | *ret = TAKE_PTR(done); | |
d944dc95 | 991 | |
1ed34d75 | 992 | if (flags & CHASE_OPEN) { |
1ed34d75 LP |
993 | /* Return the O_PATH fd we currently are looking to the caller. It can translate it to a proper fd by |
994 | * opening /proc/self/fd/xyz. */ | |
995 | ||
996 | assert(fd >= 0); | |
c10d6bdb | 997 | return TAKE_FD(fd); |
1ed34d75 LP |
998 | } |
999 | ||
49eb3659 LP |
1000 | if (flags & CHASE_STEP) |
1001 | return 1; | |
1002 | ||
a9fb0867 | 1003 | return exists; |
49eb3659 LP |
1004 | |
1005 | chased_one: | |
49eb3659 LP |
1006 | if (ret) { |
1007 | char *c; | |
1008 | ||
027cc9c9 ZJS |
1009 | c = strjoin(strempty(done), todo); |
1010 | if (!c) | |
1011 | return -ENOMEM; | |
49eb3659 LP |
1012 | |
1013 | *ret = c; | |
1014 | } | |
1015 | ||
1016 | return 0; | |
d944dc95 | 1017 | } |
57a4359e | 1018 | |
21c692e9 LP |
1019 | int chase_symlinks_and_open( |
1020 | const char *path, | |
1021 | const char *root, | |
1022 | unsigned chase_flags, | |
1023 | int open_flags, | |
1024 | char **ret_path) { | |
1025 | ||
1026 | _cleanup_close_ int path_fd = -1; | |
1027 | _cleanup_free_ char *p = NULL; | |
1028 | int r; | |
1029 | ||
1030 | if (chase_flags & CHASE_NONEXISTENT) | |
1031 | return -EINVAL; | |
1032 | ||
57ea45e1 | 1033 | if (empty_or_root(root) && !ret_path && (chase_flags & (CHASE_NO_AUTOFS|CHASE_SAFE)) == 0) { |
21c692e9 LP |
1034 | /* Shortcut this call if none of the special features of this call are requested */ |
1035 | r = open(path, open_flags); | |
1036 | if (r < 0) | |
1037 | return -errno; | |
1038 | ||
1039 | return r; | |
1040 | } | |
1041 | ||
1042 | path_fd = chase_symlinks(path, root, chase_flags|CHASE_OPEN, ret_path ? &p : NULL); | |
1043 | if (path_fd < 0) | |
1044 | return path_fd; | |
1045 | ||
1046 | r = fd_reopen(path_fd, open_flags); | |
1047 | if (r < 0) | |
1048 | return r; | |
1049 | ||
1050 | if (ret_path) | |
1051 | *ret_path = TAKE_PTR(p); | |
1052 | ||
1053 | return r; | |
1054 | } | |
1055 | ||
1056 | int chase_symlinks_and_opendir( | |
1057 | const char *path, | |
1058 | const char *root, | |
1059 | unsigned chase_flags, | |
1060 | char **ret_path, | |
1061 | DIR **ret_dir) { | |
1062 | ||
1063 | char procfs_path[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(int)]; | |
1064 | _cleanup_close_ int path_fd = -1; | |
1065 | _cleanup_free_ char *p = NULL; | |
1066 | DIR *d; | |
1067 | ||
1068 | if (!ret_dir) | |
1069 | return -EINVAL; | |
1070 | if (chase_flags & CHASE_NONEXISTENT) | |
1071 | return -EINVAL; | |
1072 | ||
57ea45e1 | 1073 | if (empty_or_root(root) && !ret_path && (chase_flags & (CHASE_NO_AUTOFS|CHASE_SAFE)) == 0) { |
21c692e9 LP |
1074 | /* Shortcut this call if none of the special features of this call are requested */ |
1075 | d = opendir(path); | |
1076 | if (!d) | |
1077 | return -errno; | |
1078 | ||
1079 | *ret_dir = d; | |
1080 | return 0; | |
1081 | } | |
1082 | ||
1083 | path_fd = chase_symlinks(path, root, chase_flags|CHASE_OPEN, ret_path ? &p : NULL); | |
1084 | if (path_fd < 0) | |
1085 | return path_fd; | |
1086 | ||
1087 | xsprintf(procfs_path, "/proc/self/fd/%i", path_fd); | |
1088 | d = opendir(procfs_path); | |
1089 | if (!d) | |
1090 | return -errno; | |
1091 | ||
1092 | if (ret_path) | |
1093 | *ret_path = TAKE_PTR(p); | |
1094 | ||
1095 | *ret_dir = d; | |
1096 | return 0; | |
1097 | } | |
1098 | ||
d2bcd0ba LP |
1099 | int chase_symlinks_and_stat( |
1100 | const char *path, | |
1101 | const char *root, | |
1102 | unsigned chase_flags, | |
1103 | char **ret_path, | |
1104 | struct stat *ret_stat) { | |
1105 | ||
1106 | _cleanup_close_ int path_fd = -1; | |
1107 | _cleanup_free_ char *p = NULL; | |
1108 | ||
1109 | assert(path); | |
1110 | assert(ret_stat); | |
1111 | ||
1112 | if (chase_flags & CHASE_NONEXISTENT) | |
1113 | return -EINVAL; | |
1114 | ||
1115 | if (empty_or_root(root) && !ret_path && (chase_flags & (CHASE_NO_AUTOFS|CHASE_SAFE)) == 0) { | |
1116 | /* Shortcut this call if none of the special features of this call are requested */ | |
1117 | if (stat(path, ret_stat) < 0) | |
1118 | return -errno; | |
1119 | ||
1120 | return 1; | |
1121 | } | |
1122 | ||
1123 | path_fd = chase_symlinks(path, root, chase_flags|CHASE_OPEN, ret_path ? &p : NULL); | |
1124 | if (path_fd < 0) | |
1125 | return path_fd; | |
1126 | ||
1127 | if (fstat(path_fd, ret_stat) < 0) | |
1128 | return -errno; | |
1129 | ||
1130 | if (ret_path) | |
1131 | *ret_path = TAKE_PTR(p); | |
1132 | ||
1133 | if (chase_flags & CHASE_OPEN) | |
1134 | return TAKE_FD(path_fd); | |
1135 | ||
1136 | return 1; | |
1137 | } | |
1138 | ||
57a4359e | 1139 | int access_fd(int fd, int mode) { |
fbd0b64f | 1140 | char p[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(fd) + 1]; |
57a4359e LP |
1141 | int r; |
1142 | ||
1143 | /* Like access() but operates on an already open fd */ | |
1144 | ||
1145 | xsprintf(p, "/proc/self/fd/%i", fd); | |
57a4359e LP |
1146 | r = access(p, mode); |
1147 | if (r < 0) | |
21c692e9 | 1148 | return -errno; |
57a4359e LP |
1149 | |
1150 | return r; | |
1151 | } | |
43767d9d | 1152 | |
627d2bac ZJS |
1153 | void unlink_tempfilep(char (*p)[]) { |
1154 | /* If the file is created with mkstemp(), it will (almost always) | |
1155 | * change the suffix. Treat this as a sign that the file was | |
1156 | * successfully created. We ignore both the rare case where the | |
1157 | * original suffix is used and unlink failures. */ | |
1158 | if (!endswith(*p, ".XXXXXX")) | |
69821560 | 1159 | (void) unlink_noerrno(*p); |
627d2bac ZJS |
1160 | } |
1161 | ||
43767d9d LP |
1162 | int unlinkat_deallocate(int fd, const char *name, int flags) { |
1163 | _cleanup_close_ int truncate_fd = -1; | |
1164 | struct stat st; | |
1165 | off_t l, bs; | |
1166 | ||
1167 | /* Operates like unlinkat() but also deallocates the file contents if it is a regular file and there's no other | |
1168 | * link to it. This is useful to ensure that other processes that might have the file open for reading won't be | |
1169 | * able to keep the data pinned on disk forever. This call is particular useful whenever we execute clean-up | |
1170 | * jobs ("vacuuming"), where we want to make sure the data is really gone and the disk space released and | |
1171 | * returned to the free pool. | |
1172 | * | |
1173 | * Deallocation is preferably done by FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE (👊) if supported, which means | |
1174 | * the file won't change size. That's a good thing since we shouldn't needlessly trigger SIGBUS in other | |
1175 | * programs that have mmap()ed the file. (The assumption here is that changing file contents to all zeroes | |
1176 | * underneath those programs is the better choice than simply triggering SIGBUS in them which truncation does.) | |
1177 | * However if hole punching is not implemented in the kernel or file system we'll fall back to normal file | |
1178 | * truncation (🔪), as our goal of deallocating the data space trumps our goal of being nice to readers (💐). | |
1179 | * | |
1180 | * Note that we attempt deallocation, but failure to succeed with that is not considered fatal, as long as the | |
1181 | * primary job – to delete the file – is accomplished. */ | |
1182 | ||
1183 | if ((flags & AT_REMOVEDIR) == 0) { | |
1184 | truncate_fd = openat(fd, name, O_WRONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW|O_NONBLOCK); | |
1185 | if (truncate_fd < 0) { | |
1186 | ||
1187 | /* If this failed because the file doesn't exist propagate the error right-away. Also, | |
1188 | * AT_REMOVEDIR wasn't set, and we tried to open the file for writing, which means EISDIR is | |
1189 | * returned when this is a directory but we are not supposed to delete those, hence propagate | |
1190 | * the error right-away too. */ | |
1191 | if (IN_SET(errno, ENOENT, EISDIR)) | |
1192 | return -errno; | |
1193 | ||
1194 | if (errno != ELOOP) /* don't complain if this is a symlink */ | |
1195 | log_debug_errno(errno, "Failed to open file '%s' for deallocation, ignoring: %m", name); | |
1196 | } | |
1197 | } | |
1198 | ||
1199 | if (unlinkat(fd, name, flags) < 0) | |
1200 | return -errno; | |
1201 | ||
1202 | if (truncate_fd < 0) /* Don't have a file handle, can't do more ☹️ */ | |
1203 | return 0; | |
1204 | ||
1205 | if (fstat(truncate_fd, &st) < 0) { | |
011723a4 | 1206 | log_debug_errno(errno, "Failed to stat file '%s' for deallocation, ignoring: %m", name); |
43767d9d LP |
1207 | return 0; |
1208 | } | |
1209 | ||
1210 | if (!S_ISREG(st.st_mode) || st.st_blocks == 0 || st.st_nlink > 0) | |
1211 | return 0; | |
1212 | ||
1213 | /* If this is a regular file, it actually took up space on disk and there are no other links it's time to | |
1214 | * punch-hole/truncate this to release the disk space. */ | |
1215 | ||
1216 | bs = MAX(st.st_blksize, 512); | |
1217 | l = DIV_ROUND_UP(st.st_size, bs) * bs; /* Round up to next block size */ | |
1218 | ||
1219 | if (fallocate(truncate_fd, FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE, 0, l) >= 0) | |
1220 | return 0; /* Successfully punched a hole! 😊 */ | |
1221 | ||
1222 | /* Fall back to truncation */ | |
1223 | if (ftruncate(truncate_fd, 0) < 0) { | |
1224 | log_debug_errno(errno, "Failed to truncate file to 0, ignoring: %m"); | |
1225 | return 0; | |
1226 | } | |
1227 | ||
1228 | return 0; | |
1229 | } | |
11b29a96 LP |
1230 | |
1231 | int fsync_directory_of_file(int fd) { | |
0c462ea4 | 1232 | _cleanup_free_ char *path = NULL; |
11b29a96 LP |
1233 | _cleanup_close_ int dfd = -1; |
1234 | int r; | |
1235 | ||
1236 | r = fd_verify_regular(fd); | |
1237 | if (r < 0) | |
1238 | return r; | |
1239 | ||
1240 | r = fd_get_path(fd, &path); | |
3ceae1bc | 1241 | if (r < 0) { |
b8b846d7 LP |
1242 | log_debug_errno(r, "Failed to query /proc/self/fd/%d%s: %m", |
1243 | fd, | |
1244 | r == -EOPNOTSUPP ? ", ignoring" : ""); | |
3ceae1bc ZJS |
1245 | |
1246 | if (r == -EOPNOTSUPP) | |
1247 | /* If /proc is not available, we're most likely running in some | |
1248 | * chroot environment, and syncing the directory is not very | |
1249 | * important in that case. Let's just silently do nothing. */ | |
1250 | return 0; | |
1251 | ||
11b29a96 | 1252 | return r; |
3ceae1bc | 1253 | } |
11b29a96 LP |
1254 | |
1255 | if (!path_is_absolute(path)) | |
1256 | return -EINVAL; | |
1257 | ||
0c462ea4 | 1258 | dfd = open_parent(path, O_CLOEXEC, 0); |
11b29a96 | 1259 | if (dfd < 0) |
0c462ea4 | 1260 | return dfd; |
11b29a96 LP |
1261 | |
1262 | if (fsync(dfd) < 0) | |
1263 | return -errno; | |
1264 | ||
1265 | return 0; | |
1266 | } | |
ef8becfa | 1267 | |
36695e88 LP |
1268 | int fsync_path_at(int at_fd, const char *path) { |
1269 | _cleanup_close_ int opened_fd = -1; | |
1270 | int fd; | |
1271 | ||
1272 | if (isempty(path)) { | |
1273 | if (at_fd == AT_FDCWD) { | |
1274 | opened_fd = open(".", O_RDONLY|O_DIRECTORY|O_CLOEXEC); | |
1275 | if (opened_fd < 0) | |
1276 | return -errno; | |
1277 | ||
1278 | fd = opened_fd; | |
1279 | } else | |
1280 | fd = at_fd; | |
1281 | } else { | |
1282 | ||
1283 | opened_fd = openat(at_fd, path, O_RDONLY|O_CLOEXEC); | |
1284 | if (opened_fd < 0) | |
1285 | return -errno; | |
1286 | ||
1287 | fd = opened_fd; | |
1288 | } | |
1289 | ||
1290 | if (fsync(fd) < 0) | |
1291 | return -errno; | |
1292 | ||
1293 | return 0; | |
1294 | } | |
1295 | ||
ef8becfa LP |
1296 | int open_parent(const char *path, int flags, mode_t mode) { |
1297 | _cleanup_free_ char *parent = NULL; | |
1298 | int fd; | |
1299 | ||
1300 | if (isempty(path)) | |
1301 | return -EINVAL; | |
1302 | if (path_equal(path, "/")) /* requesting the parent of the root dir is fishy, let's prohibit that */ | |
1303 | return -EINVAL; | |
1304 | ||
1305 | parent = dirname_malloc(path); | |
1306 | if (!parent) | |
1307 | return -ENOMEM; | |
1308 | ||
1309 | /* Let's insist on O_DIRECTORY since the parent of a file or directory is a directory. Except if we open an | |
1310 | * O_TMPFILE file, because in that case we are actually create a regular file below the parent directory. */ | |
1311 | ||
1312 | if ((flags & O_PATH) == O_PATH) | |
1313 | flags |= O_DIRECTORY; | |
1314 | else if ((flags & O_TMPFILE) != O_TMPFILE) | |
1315 | flags |= O_DIRECTORY|O_RDONLY; | |
1316 | ||
1317 | fd = open(parent, flags, mode); | |
1318 | if (fd < 0) | |
1319 | return -errno; | |
1320 | ||
1321 | return fd; | |
1322 | } |