]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
3ffd4af2 LP |
2 | /*** |
3 | This file is part of systemd. | |
4 | ||
5 | Copyright 2010 Lennart Poettering | |
6 | ||
7 | systemd is free software; you can redistribute it and/or modify it | |
8 | under the terms of the GNU Lesser General Public License as published by | |
9 | the Free Software Foundation; either version 2.1 of the License, or | |
10 | (at your option) any later version. | |
11 | ||
12 | systemd is distributed in the hope that it will be useful, but | |
13 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 | Lesser General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU Lesser General Public License | |
18 | along with systemd; If not, see <http://www.gnu.org/licenses/>. | |
19 | ***/ | |
20 | ||
11c3a366 TA |
21 | #include <errno.h> |
22 | #include <fcntl.h> | |
23 | #include <sys/resource.h> | |
24 | #include <sys/socket.h> | |
25 | #include <sys/stat.h> | |
26 | #include <unistd.h> | |
27 | ||
8fb3f009 | 28 | #include "dirent-util.h" |
3ffd4af2 | 29 | #include "fd-util.h" |
a548e14d | 30 | #include "fileio.h" |
4aeb20f5 | 31 | #include "fs-util.h" |
11c3a366 | 32 | #include "macro.h" |
a548e14d | 33 | #include "memfd-util.h" |
11c3a366 | 34 | #include "missing.h" |
93cc7779 | 35 | #include "parse-util.h" |
11c3a366 | 36 | #include "path-util.h" |
df0ff127 | 37 | #include "process-util.h" |
93cc7779 | 38 | #include "socket-util.h" |
4aeb20f5 | 39 | #include "stdio-util.h" |
3ffd4af2 LP |
40 | #include "util.h" |
41 | ||
42 | int close_nointr(int fd) { | |
43 | assert(fd >= 0); | |
44 | ||
45 | if (close(fd) >= 0) | |
46 | return 0; | |
47 | ||
48 | /* | |
49 | * Just ignore EINTR; a retry loop is the wrong thing to do on | |
50 | * Linux. | |
51 | * | |
52 | * http://lkml.indiana.edu/hypermail/linux/kernel/0509.1/0877.html | |
53 | * https://bugzilla.gnome.org/show_bug.cgi?id=682819 | |
54 | * http://utcc.utoronto.ca/~cks/space/blog/unix/CloseEINTR | |
55 | * https://sites.google.com/site/michaelsafyan/software-engineering/checkforeintrwheninvokingclosethinkagain | |
56 | */ | |
57 | if (errno == EINTR) | |
58 | return 0; | |
59 | ||
60 | return -errno; | |
61 | } | |
62 | ||
63 | int safe_close(int fd) { | |
64 | ||
65 | /* | |
66 | * Like close_nointr() but cannot fail. Guarantees errno is | |
67 | * unchanged. Is a NOP with negative fds passed, and returns | |
68 | * -1, so that it can be used in this syntax: | |
69 | * | |
70 | * fd = safe_close(fd); | |
71 | */ | |
72 | ||
73 | if (fd >= 0) { | |
74 | PROTECT_ERRNO; | |
75 | ||
76 | /* The kernel might return pretty much any error code | |
77 | * via close(), but the fd will be closed anyway. The | |
78 | * only condition we want to check for here is whether | |
79 | * the fd was invalid at all... */ | |
80 | ||
81 | assert_se(close_nointr(fd) != -EBADF); | |
82 | } | |
83 | ||
84 | return -1; | |
85 | } | |
86 | ||
87 | void safe_close_pair(int p[]) { | |
88 | assert(p); | |
89 | ||
90 | if (p[0] == p[1]) { | |
91 | /* Special case pairs which use the same fd in both | |
92 | * directions... */ | |
93 | p[0] = p[1] = safe_close(p[0]); | |
94 | return; | |
95 | } | |
96 | ||
97 | p[0] = safe_close(p[0]); | |
98 | p[1] = safe_close(p[1]); | |
99 | } | |
100 | ||
101 | void close_many(const int fds[], unsigned n_fd) { | |
102 | unsigned i; | |
103 | ||
104 | assert(fds || n_fd <= 0); | |
105 | ||
106 | for (i = 0; i < n_fd; i++) | |
107 | safe_close(fds[i]); | |
108 | } | |
109 | ||
110 | int fclose_nointr(FILE *f) { | |
111 | assert(f); | |
112 | ||
113 | /* Same as close_nointr(), but for fclose() */ | |
114 | ||
115 | if (fclose(f) == 0) | |
116 | return 0; | |
117 | ||
118 | if (errno == EINTR) | |
119 | return 0; | |
120 | ||
121 | return -errno; | |
122 | } | |
123 | ||
124 | FILE* safe_fclose(FILE *f) { | |
125 | ||
126 | /* Same as safe_close(), but for fclose() */ | |
127 | ||
128 | if (f) { | |
129 | PROTECT_ERRNO; | |
130 | ||
131 | assert_se(fclose_nointr(f) != EBADF); | |
132 | } | |
133 | ||
134 | return NULL; | |
135 | } | |
136 | ||
137 | DIR* safe_closedir(DIR *d) { | |
138 | ||
139 | if (d) { | |
140 | PROTECT_ERRNO; | |
141 | ||
142 | assert_se(closedir(d) >= 0 || errno != EBADF); | |
143 | } | |
144 | ||
145 | return NULL; | |
146 | } | |
147 | ||
148 | int fd_nonblock(int fd, bool nonblock) { | |
149 | int flags, nflags; | |
150 | ||
151 | assert(fd >= 0); | |
152 | ||
153 | flags = fcntl(fd, F_GETFL, 0); | |
154 | if (flags < 0) | |
155 | return -errno; | |
156 | ||
157 | if (nonblock) | |
158 | nflags = flags | O_NONBLOCK; | |
159 | else | |
160 | nflags = flags & ~O_NONBLOCK; | |
161 | ||
162 | if (nflags == flags) | |
163 | return 0; | |
164 | ||
165 | if (fcntl(fd, F_SETFL, nflags) < 0) | |
166 | return -errno; | |
167 | ||
168 | return 0; | |
169 | } | |
170 | ||
171 | int fd_cloexec(int fd, bool cloexec) { | |
172 | int flags, nflags; | |
173 | ||
174 | assert(fd >= 0); | |
175 | ||
176 | flags = fcntl(fd, F_GETFD, 0); | |
177 | if (flags < 0) | |
178 | return -errno; | |
179 | ||
180 | if (cloexec) | |
181 | nflags = flags | FD_CLOEXEC; | |
182 | else | |
183 | nflags = flags & ~FD_CLOEXEC; | |
184 | ||
185 | if (nflags == flags) | |
186 | return 0; | |
187 | ||
188 | if (fcntl(fd, F_SETFD, nflags) < 0) | |
189 | return -errno; | |
190 | ||
191 | return 0; | |
192 | } | |
193 | ||
3b9a1d87 | 194 | void stdio_unset_cloexec(void) { |
61ccf772 LP |
195 | (void) fd_cloexec(STDIN_FILENO, false); |
196 | (void) fd_cloexec(STDOUT_FILENO, false); | |
197 | (void) fd_cloexec(STDERR_FILENO, false); | |
3b9a1d87 AK |
198 | } |
199 | ||
3ffd4af2 LP |
200 | _pure_ static bool fd_in_set(int fd, const int fdset[], unsigned n_fdset) { |
201 | unsigned i; | |
202 | ||
203 | assert(n_fdset == 0 || fdset); | |
204 | ||
205 | for (i = 0; i < n_fdset; i++) | |
206 | if (fdset[i] == fd) | |
207 | return true; | |
208 | ||
209 | return false; | |
210 | } | |
211 | ||
212 | int close_all_fds(const int except[], unsigned n_except) { | |
213 | _cleanup_closedir_ DIR *d = NULL; | |
214 | struct dirent *de; | |
215 | int r = 0; | |
216 | ||
217 | assert(n_except == 0 || except); | |
218 | ||
219 | d = opendir("/proc/self/fd"); | |
220 | if (!d) { | |
221 | int fd; | |
222 | struct rlimit rl; | |
223 | ||
224 | /* When /proc isn't available (for example in chroots) | |
225 | * the fallback is brute forcing through the fd | |
226 | * table */ | |
227 | ||
228 | assert_se(getrlimit(RLIMIT_NOFILE, &rl) >= 0); | |
229 | for (fd = 3; fd < (int) rl.rlim_max; fd ++) { | |
230 | ||
231 | if (fd_in_set(fd, except, n_except)) | |
232 | continue; | |
233 | ||
234 | if (close_nointr(fd) < 0) | |
235 | if (errno != EBADF && r == 0) | |
236 | r = -errno; | |
237 | } | |
238 | ||
239 | return r; | |
240 | } | |
241 | ||
8fb3f009 | 242 | FOREACH_DIRENT(de, d, return -errno) { |
3ffd4af2 LP |
243 | int fd = -1; |
244 | ||
3ffd4af2 LP |
245 | if (safe_atoi(de->d_name, &fd) < 0) |
246 | /* Let's better ignore this, just in case */ | |
247 | continue; | |
248 | ||
249 | if (fd < 3) | |
250 | continue; | |
251 | ||
252 | if (fd == dirfd(d)) | |
253 | continue; | |
254 | ||
255 | if (fd_in_set(fd, except, n_except)) | |
256 | continue; | |
257 | ||
258 | if (close_nointr(fd) < 0) { | |
259 | /* Valgrind has its own FD and doesn't want to have it closed */ | |
260 | if (errno != EBADF && r == 0) | |
261 | r = -errno; | |
262 | } | |
263 | } | |
264 | ||
265 | return r; | |
266 | } | |
267 | ||
268 | int same_fd(int a, int b) { | |
269 | struct stat sta, stb; | |
270 | pid_t pid; | |
271 | int r, fa, fb; | |
272 | ||
273 | assert(a >= 0); | |
274 | assert(b >= 0); | |
275 | ||
276 | /* Compares two file descriptors. Note that semantics are | |
277 | * quite different depending on whether we have kcmp() or we | |
278 | * don't. If we have kcmp() this will only return true for | |
279 | * dup()ed file descriptors, but not otherwise. If we don't | |
280 | * have kcmp() this will also return true for two fds of the same | |
281 | * file, created by separate open() calls. Since we use this | |
282 | * call mostly for filtering out duplicates in the fd store | |
283 | * this difference hopefully doesn't matter too much. */ | |
284 | ||
285 | if (a == b) | |
286 | return true; | |
287 | ||
288 | /* Try to use kcmp() if we have it. */ | |
df0ff127 | 289 | pid = getpid_cached(); |
3ffd4af2 LP |
290 | r = kcmp(pid, pid, KCMP_FILE, a, b); |
291 | if (r == 0) | |
292 | return true; | |
293 | if (r > 0) | |
294 | return false; | |
295 | if (errno != ENOSYS) | |
296 | return -errno; | |
297 | ||
298 | /* We don't have kcmp(), use fstat() instead. */ | |
299 | if (fstat(a, &sta) < 0) | |
300 | return -errno; | |
301 | ||
302 | if (fstat(b, &stb) < 0) | |
303 | return -errno; | |
304 | ||
305 | if ((sta.st_mode & S_IFMT) != (stb.st_mode & S_IFMT)) | |
306 | return false; | |
307 | ||
308 | /* We consider all device fds different, since two device fds | |
309 | * might refer to quite different device contexts even though | |
310 | * they share the same inode and backing dev_t. */ | |
311 | ||
312 | if (S_ISCHR(sta.st_mode) || S_ISBLK(sta.st_mode)) | |
313 | return false; | |
314 | ||
315 | if (sta.st_dev != stb.st_dev || sta.st_ino != stb.st_ino) | |
316 | return false; | |
317 | ||
318 | /* The fds refer to the same inode on disk, let's also check | |
319 | * if they have the same fd flags. This is useful to | |
320 | * distinguish the read and write side of a pipe created with | |
321 | * pipe(). */ | |
322 | fa = fcntl(a, F_GETFL); | |
323 | if (fa < 0) | |
324 | return -errno; | |
325 | ||
326 | fb = fcntl(b, F_GETFL); | |
327 | if (fb < 0) | |
328 | return -errno; | |
329 | ||
330 | return fa == fb; | |
331 | } | |
332 | ||
333 | void cmsg_close_all(struct msghdr *mh) { | |
334 | struct cmsghdr *cmsg; | |
335 | ||
336 | assert(mh); | |
337 | ||
338 | CMSG_FOREACH(cmsg, mh) | |
339 | if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) | |
340 | close_many((int*) CMSG_DATA(cmsg), (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int)); | |
341 | } | |
4fee3975 LP |
342 | |
343 | bool fdname_is_valid(const char *s) { | |
344 | const char *p; | |
345 | ||
346 | /* Validates a name for $LISTEN_FDNAMES. We basically allow | |
347 | * everything ASCII that's not a control character. Also, as | |
348 | * special exception the ":" character is not allowed, as we | |
349 | * use that as field separator in $LISTEN_FDNAMES. | |
350 | * | |
351 | * Note that the empty string is explicitly allowed | |
352 | * here. However, we limit the length of the names to 255 | |
353 | * characters. */ | |
354 | ||
355 | if (!s) | |
356 | return false; | |
357 | ||
358 | for (p = s; *p; p++) { | |
359 | if (*p < ' ') | |
360 | return false; | |
361 | if (*p >= 127) | |
362 | return false; | |
363 | if (*p == ':') | |
364 | return false; | |
365 | } | |
366 | ||
367 | return p - s < 256; | |
368 | } | |
4aeb20f5 LP |
369 | |
370 | int fd_get_path(int fd, char **ret) { | |
dbcb4a90 | 371 | char procfs_path[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(int)]; |
a0fe2a2d | 372 | int r; |
4aeb20f5 LP |
373 | |
374 | xsprintf(procfs_path, "/proc/self/fd/%i", fd); | |
375 | ||
a0fe2a2d LP |
376 | r = readlink_malloc(procfs_path, ret); |
377 | ||
378 | if (r == -ENOENT) /* If the file doesn't exist the fd is invalid */ | |
379 | return -EBADF; | |
380 | ||
381 | return r; | |
4aeb20f5 | 382 | } |
046a82c1 LP |
383 | |
384 | int move_fd(int from, int to, int cloexec) { | |
385 | int r; | |
386 | ||
387 | /* Move fd 'from' to 'to', make sure FD_CLOEXEC remains equal if requested, and release the old fd. If | |
388 | * 'cloexec' is passed as -1, the original FD_CLOEXEC is inherited for the new fd. If it is 0, it is turned | |
389 | * off, if it is > 0 it is turned on. */ | |
390 | ||
391 | if (from < 0) | |
392 | return -EBADF; | |
393 | if (to < 0) | |
394 | return -EBADF; | |
395 | ||
396 | if (from == to) { | |
397 | ||
398 | if (cloexec >= 0) { | |
399 | r = fd_cloexec(to, cloexec); | |
400 | if (r < 0) | |
401 | return r; | |
402 | } | |
403 | ||
404 | return to; | |
405 | } | |
406 | ||
407 | if (cloexec < 0) { | |
408 | int fl; | |
409 | ||
410 | fl = fcntl(from, F_GETFD, 0); | |
411 | if (fl < 0) | |
412 | return -errno; | |
413 | ||
414 | cloexec = !!(fl & FD_CLOEXEC); | |
415 | } | |
416 | ||
417 | r = dup3(from, to, cloexec ? O_CLOEXEC : 0); | |
418 | if (r < 0) | |
419 | return -errno; | |
420 | ||
421 | assert(r == to); | |
422 | ||
423 | safe_close(from); | |
424 | ||
425 | return to; | |
426 | } | |
a548e14d LP |
427 | |
428 | int acquire_data_fd(const void *data, size_t size, unsigned flags) { | |
429 | ||
fbd0b64f | 430 | char procfs_path[STRLEN("/proc/self/fd/") + DECIMAL_STR_MAX(int)]; |
a548e14d LP |
431 | _cleanup_close_pair_ int pipefds[2] = { -1, -1 }; |
432 | char pattern[] = "/dev/shm/data-fd-XXXXXX"; | |
433 | _cleanup_close_ int fd = -1; | |
434 | int isz = 0, r; | |
435 | ssize_t n; | |
436 | off_t f; | |
437 | ||
438 | assert(data || size == 0); | |
439 | ||
440 | /* Acquire a read-only file descriptor that when read from returns the specified data. This is much more | |
441 | * complex than I wish it was. But here's why: | |
442 | * | |
443 | * a) First we try to use memfds. They are the best option, as we can seal them nicely to make them | |
444 | * read-only. Unfortunately they require kernel 3.17, and – at the time of writing – we still support 3.14. | |
445 | * | |
446 | * b) Then, we try classic pipes. They are the second best options, as we can close the writing side, retaining | |
447 | * a nicely read-only fd in the reading side. However, they are by default quite small, and unprivileged | |
448 | * clients can only bump their size to a system-wide limit, which might be quite low. | |
449 | * | |
450 | * c) Then, we try an O_TMPFILE file in /dev/shm (that dir is the only suitable one known to exist from | |
451 | * earliest boot on). To make it read-only we open the fd a second time with O_RDONLY via | |
452 | * /proc/self/<fd>. Unfortunately O_TMPFILE is not available on older kernels on tmpfs. | |
453 | * | |
454 | * d) Finally, we try creating a regular file in /dev/shm, which we then delete. | |
455 | * | |
456 | * It sucks a bit that depending on the situation we return very different objects here, but that's Linux I | |
457 | * figure. */ | |
458 | ||
459 | if (size == 0 && ((flags & ACQUIRE_NO_DEV_NULL) == 0)) { | |
460 | /* As a special case, return /dev/null if we have been called for an empty data block */ | |
461 | r = open("/dev/null", O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
462 | if (r < 0) | |
463 | return -errno; | |
464 | ||
465 | return r; | |
466 | } | |
467 | ||
468 | if ((flags & ACQUIRE_NO_MEMFD) == 0) { | |
469 | fd = memfd_new("data-fd"); | |
470 | if (fd < 0) | |
471 | goto try_pipe; | |
472 | ||
473 | n = write(fd, data, size); | |
474 | if (n < 0) | |
475 | return -errno; | |
476 | if ((size_t) n != size) | |
477 | return -EIO; | |
478 | ||
479 | f = lseek(fd, 0, SEEK_SET); | |
480 | if (f != 0) | |
481 | return -errno; | |
482 | ||
483 | r = memfd_set_sealed(fd); | |
484 | if (r < 0) | |
485 | return r; | |
486 | ||
487 | r = fd; | |
488 | fd = -1; | |
489 | ||
490 | return r; | |
491 | } | |
492 | ||
493 | try_pipe: | |
494 | if ((flags & ACQUIRE_NO_PIPE) == 0) { | |
495 | if (pipe2(pipefds, O_CLOEXEC|O_NONBLOCK) < 0) | |
496 | return -errno; | |
497 | ||
498 | isz = fcntl(pipefds[1], F_GETPIPE_SZ, 0); | |
499 | if (isz < 0) | |
500 | return -errno; | |
501 | ||
502 | if ((size_t) isz < size) { | |
503 | isz = (int) size; | |
504 | if (isz < 0 || (size_t) isz != size) | |
505 | return -E2BIG; | |
506 | ||
507 | /* Try to bump the pipe size */ | |
508 | (void) fcntl(pipefds[1], F_SETPIPE_SZ, isz); | |
509 | ||
510 | /* See if that worked */ | |
511 | isz = fcntl(pipefds[1], F_GETPIPE_SZ, 0); | |
512 | if (isz < 0) | |
513 | return -errno; | |
514 | ||
515 | if ((size_t) isz < size) | |
516 | goto try_dev_shm; | |
517 | } | |
518 | ||
519 | n = write(pipefds[1], data, size); | |
520 | if (n < 0) | |
521 | return -errno; | |
522 | if ((size_t) n != size) | |
523 | return -EIO; | |
524 | ||
525 | (void) fd_nonblock(pipefds[0], false); | |
526 | ||
527 | r = pipefds[0]; | |
528 | pipefds[0] = -1; | |
529 | ||
530 | return r; | |
531 | } | |
532 | ||
533 | try_dev_shm: | |
534 | if ((flags & ACQUIRE_NO_TMPFILE) == 0) { | |
535 | fd = open("/dev/shm", O_RDWR|O_TMPFILE|O_CLOEXEC, 0500); | |
536 | if (fd < 0) | |
537 | goto try_dev_shm_without_o_tmpfile; | |
538 | ||
539 | n = write(fd, data, size); | |
540 | if (n < 0) | |
541 | return -errno; | |
542 | if ((size_t) n != size) | |
543 | return -EIO; | |
544 | ||
545 | /* Let's reopen the thing, in order to get an O_RDONLY fd for the original O_RDWR one */ | |
546 | xsprintf(procfs_path, "/proc/self/fd/%i", fd); | |
547 | r = open(procfs_path, O_RDONLY|O_CLOEXEC); | |
548 | if (r < 0) | |
549 | return -errno; | |
550 | ||
551 | return r; | |
552 | } | |
553 | ||
554 | try_dev_shm_without_o_tmpfile: | |
555 | if ((flags & ACQUIRE_NO_REGULAR) == 0) { | |
556 | fd = mkostemp_safe(pattern); | |
557 | if (fd < 0) | |
558 | return fd; | |
559 | ||
560 | n = write(fd, data, size); | |
561 | if (n < 0) { | |
562 | r = -errno; | |
563 | goto unlink_and_return; | |
564 | } | |
565 | if ((size_t) n != size) { | |
566 | r = -EIO; | |
567 | goto unlink_and_return; | |
568 | } | |
569 | ||
570 | /* Let's reopen the thing, in order to get an O_RDONLY fd for the original O_RDWR one */ | |
571 | r = open(pattern, O_RDONLY|O_CLOEXEC); | |
572 | if (r < 0) | |
573 | r = -errno; | |
574 | ||
575 | unlink_and_return: | |
576 | (void) unlink(pattern); | |
577 | return r; | |
578 | } | |
579 | ||
580 | return -EOPNOTSUPP; | |
581 | } |