]>
Commit | Line | Data |
---|---|---|
a7334b09 LP |
1 | /*** |
2 | This file is part of systemd. | |
3 | ||
4 | Copyright 2010 Lennart Poettering | |
5 | ||
6 | systemd is free software; you can redistribute it and/or modify it | |
5430f7f2 LP |
7 | under the terms of the GNU Lesser General Public License as published by |
8 | the Free Software Foundation; either version 2.1 of the License, or | |
a7334b09 LP |
9 | (at your option) any later version. |
10 | ||
11 | systemd is distributed in the hope that it will be useful, but | |
12 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
5430f7f2 | 14 | Lesser General Public License for more details. |
a7334b09 | 15 | |
5430f7f2 | 16 | You should have received a copy of the GNU Lesser General Public License |
a7334b09 LP |
17 | along with systemd; If not, see <http://www.gnu.org/licenses/>. |
18 | ***/ | |
19 | ||
034c6ed7 LP |
20 | #include <errno.h> |
21 | #include <fcntl.h> | |
8dd4c05b LP |
22 | #include <glob.h> |
23 | #include <grp.h> | |
24 | #include <poll.h> | |
309bff19 | 25 | #include <signal.h> |
8dd4c05b | 26 | #include <string.h> |
19c0b0b9 | 27 | #include <sys/capability.h> |
d251207d | 28 | #include <sys/eventfd.h> |
f3e43635 | 29 | #include <sys/mman.h> |
8dd4c05b | 30 | #include <sys/personality.h> |
94f04347 | 31 | #include <sys/prctl.h> |
d2ffa389 | 32 | #include <sys/shm.h> |
8dd4c05b | 33 | #include <sys/socket.h> |
451a074f | 34 | #include <sys/stat.h> |
d2ffa389 | 35 | #include <sys/types.h> |
8dd4c05b LP |
36 | #include <sys/un.h> |
37 | #include <unistd.h> | |
023a4f67 | 38 | #include <utmpx.h> |
5cb5a6ff | 39 | |
5b6319dc LP |
40 | #ifdef HAVE_PAM |
41 | #include <security/pam_appl.h> | |
42 | #endif | |
43 | ||
7b52a628 MS |
44 | #ifdef HAVE_SELINUX |
45 | #include <selinux/selinux.h> | |
46 | #endif | |
47 | ||
17df7223 LP |
48 | #ifdef HAVE_SECCOMP |
49 | #include <seccomp.h> | |
50 | #endif | |
51 | ||
eef65bf3 MS |
52 | #ifdef HAVE_APPARMOR |
53 | #include <sys/apparmor.h> | |
54 | #endif | |
55 | ||
24882e06 | 56 | #include "sd-messages.h" |
8dd4c05b LP |
57 | |
58 | #include "af-list.h" | |
b5efdb8a | 59 | #include "alloc-util.h" |
3ffd4af2 LP |
60 | #ifdef HAVE_APPARMOR |
61 | #include "apparmor-util.h" | |
62 | #endif | |
8dd4c05b LP |
63 | #include "async.h" |
64 | #include "barrier.h" | |
8dd4c05b | 65 | #include "cap-list.h" |
430f0182 | 66 | #include "capability-util.h" |
f6a6225e | 67 | #include "def.h" |
4d1a6904 | 68 | #include "env-util.h" |
17df7223 | 69 | #include "errno-list.h" |
3ffd4af2 | 70 | #include "execute.h" |
8dd4c05b | 71 | #include "exit-status.h" |
3ffd4af2 | 72 | #include "fd-util.h" |
8dd4c05b | 73 | #include "fileio.h" |
f97b34a6 | 74 | #include "format-util.h" |
f4f15635 | 75 | #include "fs-util.h" |
7d50b32a | 76 | #include "glob-util.h" |
c004493c | 77 | #include "io-util.h" |
8dd4c05b LP |
78 | #include "ioprio.h" |
79 | #include "log.h" | |
80 | #include "macro.h" | |
81 | #include "missing.h" | |
82 | #include "mkdir.h" | |
83 | #include "namespace.h" | |
6bedfcbb | 84 | #include "parse-util.h" |
8dd4c05b | 85 | #include "path-util.h" |
0b452006 | 86 | #include "process-util.h" |
78f22b97 | 87 | #include "rlimit-util.h" |
8dd4c05b | 88 | #include "rm-rf.h" |
3ffd4af2 LP |
89 | #ifdef HAVE_SECCOMP |
90 | #include "seccomp-util.h" | |
91 | #endif | |
8dd4c05b LP |
92 | #include "securebits.h" |
93 | #include "selinux-util.h" | |
24882e06 | 94 | #include "signal-util.h" |
8dd4c05b | 95 | #include "smack-util.h" |
fd63e712 | 96 | #include "special.h" |
8b43440b | 97 | #include "string-table.h" |
07630cea | 98 | #include "string-util.h" |
8dd4c05b | 99 | #include "strv.h" |
7ccbd1ae | 100 | #include "syslog-util.h" |
8dd4c05b LP |
101 | #include "terminal-util.h" |
102 | #include "unit.h" | |
b1d4f8e1 | 103 | #include "user-util.h" |
8dd4c05b LP |
104 | #include "util.h" |
105 | #include "utmp-wtmp.h" | |
5cb5a6ff | 106 | |
e056b01d | 107 | #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC) |
31a7eb86 | 108 | #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC) |
e6a26745 | 109 | |
02a51aba LP |
110 | /* This assumes there is a 'tty' group */ |
111 | #define TTY_MODE 0620 | |
112 | ||
531dca78 LP |
113 | #define SNDBUF_SIZE (8*1024*1024) |
114 | ||
034c6ed7 LP |
115 | static int shift_fds(int fds[], unsigned n_fds) { |
116 | int start, restart_from; | |
117 | ||
118 | if (n_fds <= 0) | |
119 | return 0; | |
120 | ||
a0d40ac5 LP |
121 | /* Modifies the fds array! (sorts it) */ |
122 | ||
034c6ed7 LP |
123 | assert(fds); |
124 | ||
125 | start = 0; | |
126 | for (;;) { | |
127 | int i; | |
128 | ||
129 | restart_from = -1; | |
130 | ||
131 | for (i = start; i < (int) n_fds; i++) { | |
132 | int nfd; | |
133 | ||
134 | /* Already at right index? */ | |
135 | if (fds[i] == i+3) | |
136 | continue; | |
137 | ||
3cc2aff1 LP |
138 | nfd = fcntl(fds[i], F_DUPFD, i + 3); |
139 | if (nfd < 0) | |
034c6ed7 LP |
140 | return -errno; |
141 | ||
03e334a1 | 142 | safe_close(fds[i]); |
034c6ed7 LP |
143 | fds[i] = nfd; |
144 | ||
145 | /* Hmm, the fd we wanted isn't free? Then | |
ee33e53a | 146 | * let's remember that and try again from here */ |
034c6ed7 LP |
147 | if (nfd != i+3 && restart_from < 0) |
148 | restart_from = i; | |
149 | } | |
150 | ||
151 | if (restart_from < 0) | |
152 | break; | |
153 | ||
154 | start = restart_from; | |
155 | } | |
156 | ||
157 | return 0; | |
158 | } | |
159 | ||
c2748801 | 160 | static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) { |
47a71eed | 161 | unsigned i; |
e2c76839 | 162 | int r; |
47a71eed LP |
163 | |
164 | if (n_fds <= 0) | |
165 | return 0; | |
166 | ||
167 | assert(fds); | |
168 | ||
451a074f | 169 | /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */ |
47a71eed LP |
170 | |
171 | for (i = 0; i < n_fds; i++) { | |
47a71eed | 172 | |
3cc2aff1 LP |
173 | r = fd_nonblock(fds[i], nonblock); |
174 | if (r < 0) | |
e2c76839 | 175 | return r; |
47a71eed | 176 | |
451a074f LP |
177 | /* We unconditionally drop FD_CLOEXEC from the fds, |
178 | * since after all we want to pass these fds to our | |
179 | * children */ | |
47a71eed | 180 | |
3cc2aff1 LP |
181 | r = fd_cloexec(fds[i], false); |
182 | if (r < 0) | |
e2c76839 | 183 | return r; |
47a71eed LP |
184 | } |
185 | ||
186 | return 0; | |
187 | } | |
188 | ||
1e22b5cd | 189 | static const char *exec_context_tty_path(const ExecContext *context) { |
80876c20 LP |
190 | assert(context); |
191 | ||
1e22b5cd LP |
192 | if (context->stdio_as_fds) |
193 | return NULL; | |
194 | ||
80876c20 LP |
195 | if (context->tty_path) |
196 | return context->tty_path; | |
197 | ||
198 | return "/dev/console"; | |
199 | } | |
200 | ||
1e22b5cd LP |
201 | static void exec_context_tty_reset(const ExecContext *context, const ExecParameters *p) { |
202 | const char *path; | |
203 | ||
6ea832a2 LP |
204 | assert(context); |
205 | ||
1e22b5cd | 206 | path = exec_context_tty_path(context); |
6ea832a2 | 207 | |
1e22b5cd LP |
208 | if (context->tty_vhangup) { |
209 | if (p && p->stdin_fd >= 0) | |
210 | (void) terminal_vhangup_fd(p->stdin_fd); | |
211 | else if (path) | |
212 | (void) terminal_vhangup(path); | |
213 | } | |
6ea832a2 | 214 | |
1e22b5cd LP |
215 | if (context->tty_reset) { |
216 | if (p && p->stdin_fd >= 0) | |
217 | (void) reset_terminal_fd(p->stdin_fd, true); | |
218 | else if (path) | |
219 | (void) reset_terminal(path); | |
220 | } | |
221 | ||
222 | if (context->tty_vt_disallocate && path) | |
223 | (void) vt_disallocate(path); | |
6ea832a2 LP |
224 | } |
225 | ||
6af760f3 LP |
226 | static bool is_terminal_input(ExecInput i) { |
227 | return IN_SET(i, | |
228 | EXEC_INPUT_TTY, | |
229 | EXEC_INPUT_TTY_FORCE, | |
230 | EXEC_INPUT_TTY_FAIL); | |
231 | } | |
232 | ||
3a1286b6 | 233 | static bool is_terminal_output(ExecOutput o) { |
6af760f3 LP |
234 | return IN_SET(o, |
235 | EXEC_OUTPUT_TTY, | |
236 | EXEC_OUTPUT_SYSLOG_AND_CONSOLE, | |
237 | EXEC_OUTPUT_KMSG_AND_CONSOLE, | |
238 | EXEC_OUTPUT_JOURNAL_AND_CONSOLE); | |
239 | } | |
240 | ||
241 | static bool exec_context_needs_term(const ExecContext *c) { | |
242 | assert(c); | |
243 | ||
244 | /* Return true if the execution context suggests we should set $TERM to something useful. */ | |
245 | ||
246 | if (is_terminal_input(c->std_input)) | |
247 | return true; | |
248 | ||
249 | if (is_terminal_output(c->std_output)) | |
250 | return true; | |
251 | ||
252 | if (is_terminal_output(c->std_error)) | |
253 | return true; | |
254 | ||
255 | return !!c->tty_path; | |
3a1286b6 MS |
256 | } |
257 | ||
80876c20 LP |
258 | static int open_null_as(int flags, int nfd) { |
259 | int fd, r; | |
071830ff | 260 | |
80876c20 | 261 | assert(nfd >= 0); |
071830ff | 262 | |
613b411c LP |
263 | fd = open("/dev/null", flags|O_NOCTTY); |
264 | if (fd < 0) | |
071830ff LP |
265 | return -errno; |
266 | ||
80876c20 LP |
267 | if (fd != nfd) { |
268 | r = dup2(fd, nfd) < 0 ? -errno : nfd; | |
03e334a1 | 269 | safe_close(fd); |
80876c20 LP |
270 | } else |
271 | r = nfd; | |
071830ff | 272 | |
80876c20 | 273 | return r; |
071830ff LP |
274 | } |
275 | ||
524daa8c | 276 | static int connect_journal_socket(int fd, uid_t uid, gid_t gid) { |
b92bea5d ZJS |
277 | union sockaddr_union sa = { |
278 | .un.sun_family = AF_UNIX, | |
279 | .un.sun_path = "/run/systemd/journal/stdout", | |
280 | }; | |
524daa8c ZJS |
281 | uid_t olduid = UID_INVALID; |
282 | gid_t oldgid = GID_INVALID; | |
283 | int r; | |
284 | ||
285 | if (gid != GID_INVALID) { | |
286 | oldgid = getgid(); | |
287 | ||
288 | r = setegid(gid); | |
289 | if (r < 0) | |
290 | return -errno; | |
291 | } | |
292 | ||
293 | if (uid != UID_INVALID) { | |
294 | olduid = getuid(); | |
295 | ||
296 | r = seteuid(uid); | |
297 | if (r < 0) { | |
298 | r = -errno; | |
299 | goto restore_gid; | |
300 | } | |
301 | } | |
302 | ||
fc2fffe7 | 303 | r = connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)); |
524daa8c ZJS |
304 | if (r < 0) |
305 | r = -errno; | |
306 | ||
307 | /* If we fail to restore the uid or gid, things will likely | |
308 | fail later on. This should only happen if an LSM interferes. */ | |
309 | ||
310 | if (uid != UID_INVALID) | |
311 | (void) seteuid(olduid); | |
312 | ||
313 | restore_gid: | |
314 | if (gid != GID_INVALID) | |
315 | (void) setegid(oldgid); | |
316 | ||
317 | return r; | |
318 | } | |
319 | ||
fd1f9c89 | 320 | static int connect_logger_as( |
7a1ab780 | 321 | Unit *unit, |
fd1f9c89 LP |
322 | const ExecContext *context, |
323 | ExecOutput output, | |
324 | const char *ident, | |
fd1f9c89 LP |
325 | int nfd, |
326 | uid_t uid, | |
327 | gid_t gid) { | |
328 | ||
524daa8c | 329 | int fd, r; |
071830ff LP |
330 | |
331 | assert(context); | |
80876c20 LP |
332 | assert(output < _EXEC_OUTPUT_MAX); |
333 | assert(ident); | |
334 | assert(nfd >= 0); | |
071830ff | 335 | |
54fe0cdb LP |
336 | fd = socket(AF_UNIX, SOCK_STREAM, 0); |
337 | if (fd < 0) | |
80876c20 | 338 | return -errno; |
071830ff | 339 | |
524daa8c ZJS |
340 | r = connect_journal_socket(fd, uid, gid); |
341 | if (r < 0) | |
342 | return r; | |
071830ff | 343 | |
80876c20 | 344 | if (shutdown(fd, SHUT_RD) < 0) { |
03e334a1 | 345 | safe_close(fd); |
80876c20 LP |
346 | return -errno; |
347 | } | |
071830ff | 348 | |
fd1f9c89 | 349 | (void) fd_inc_sndbuf(fd, SNDBUF_SIZE); |
531dca78 | 350 | |
80876c20 | 351 | dprintf(fd, |
62bca2c6 | 352 | "%s\n" |
80876c20 LP |
353 | "%s\n" |
354 | "%i\n" | |
54fe0cdb LP |
355 | "%i\n" |
356 | "%i\n" | |
357 | "%i\n" | |
4f4a1dbf | 358 | "%i\n", |
4f4a1dbf | 359 | context->syslog_identifier ? context->syslog_identifier : ident, |
7a1ab780 | 360 | unit->id, |
54fe0cdb LP |
361 | context->syslog_priority, |
362 | !!context->syslog_level_prefix, | |
363 | output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE, | |
364 | output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE, | |
3a1286b6 | 365 | is_terminal_output(output)); |
80876c20 | 366 | |
fd1f9c89 LP |
367 | if (fd == nfd) |
368 | return nfd; | |
369 | ||
370 | r = dup2(fd, nfd) < 0 ? -errno : nfd; | |
371 | safe_close(fd); | |
071830ff | 372 | |
80876c20 LP |
373 | return r; |
374 | } | |
375 | static int open_terminal_as(const char *path, mode_t mode, int nfd) { | |
376 | int fd, r; | |
071830ff | 377 | |
80876c20 LP |
378 | assert(path); |
379 | assert(nfd >= 0); | |
071830ff | 380 | |
3cc2aff1 LP |
381 | fd = open_terminal(path, mode | O_NOCTTY); |
382 | if (fd < 0) | |
80876c20 | 383 | return fd; |
071830ff | 384 | |
80876c20 LP |
385 | if (fd != nfd) { |
386 | r = dup2(fd, nfd) < 0 ? -errno : nfd; | |
03e334a1 | 387 | safe_close(fd); |
80876c20 LP |
388 | } else |
389 | r = nfd; | |
071830ff | 390 | |
80876c20 LP |
391 | return r; |
392 | } | |
071830ff | 393 | |
1e3ad081 LP |
394 | static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) { |
395 | ||
396 | if (is_terminal_input(std_input) && !apply_tty_stdin) | |
397 | return EXEC_INPUT_NULL; | |
071830ff | 398 | |
03fd9c49 | 399 | if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0) |
4f2d528d LP |
400 | return EXEC_INPUT_NULL; |
401 | ||
03fd9c49 | 402 | return std_input; |
4f2d528d LP |
403 | } |
404 | ||
03fd9c49 | 405 | static int fixup_output(ExecOutput std_output, int socket_fd) { |
4f2d528d | 406 | |
03fd9c49 | 407 | if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0) |
4f2d528d LP |
408 | return EXEC_OUTPUT_INHERIT; |
409 | ||
03fd9c49 | 410 | return std_output; |
4f2d528d LP |
411 | } |
412 | ||
a34ceba6 LP |
413 | static int setup_input( |
414 | const ExecContext *context, | |
415 | const ExecParameters *params, | |
52c239d7 LB |
416 | int socket_fd, |
417 | int named_iofds[3]) { | |
a34ceba6 | 418 | |
4f2d528d LP |
419 | ExecInput i; |
420 | ||
421 | assert(context); | |
a34ceba6 LP |
422 | assert(params); |
423 | ||
424 | if (params->stdin_fd >= 0) { | |
425 | if (dup2(params->stdin_fd, STDIN_FILENO) < 0) | |
426 | return -errno; | |
427 | ||
428 | /* Try to make this the controlling tty, if it is a tty, and reset it */ | |
429 | (void) ioctl(STDIN_FILENO, TIOCSCTTY, context->std_input == EXEC_INPUT_TTY_FORCE); | |
430 | (void) reset_terminal_fd(STDIN_FILENO, true); | |
431 | ||
432 | return STDIN_FILENO; | |
433 | } | |
4f2d528d | 434 | |
c39f1ce2 | 435 | i = fixup_input(context->std_input, socket_fd, params->flags & EXEC_APPLY_TTY_STDIN); |
4f2d528d LP |
436 | |
437 | switch (i) { | |
071830ff | 438 | |
80876c20 LP |
439 | case EXEC_INPUT_NULL: |
440 | return open_null_as(O_RDONLY, STDIN_FILENO); | |
441 | ||
442 | case EXEC_INPUT_TTY: | |
443 | case EXEC_INPUT_TTY_FORCE: | |
444 | case EXEC_INPUT_TTY_FAIL: { | |
445 | int fd, r; | |
071830ff | 446 | |
1e22b5cd | 447 | fd = acquire_terminal(exec_context_tty_path(context), |
970edce6 ZJS |
448 | i == EXEC_INPUT_TTY_FAIL, |
449 | i == EXEC_INPUT_TTY_FORCE, | |
450 | false, | |
3a43da28 | 451 | USEC_INFINITY); |
970edce6 | 452 | if (fd < 0) |
80876c20 LP |
453 | return fd; |
454 | ||
455 | if (fd != STDIN_FILENO) { | |
456 | r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO; | |
03e334a1 | 457 | safe_close(fd); |
80876c20 LP |
458 | } else |
459 | r = STDIN_FILENO; | |
460 | ||
461 | return r; | |
462 | } | |
463 | ||
4f2d528d LP |
464 | case EXEC_INPUT_SOCKET: |
465 | return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO; | |
466 | ||
52c239d7 LB |
467 | case EXEC_INPUT_NAMED_FD: |
468 | (void) fd_nonblock(named_iofds[STDIN_FILENO], false); | |
469 | return dup2(named_iofds[STDIN_FILENO], STDIN_FILENO) < 0 ? -errno : STDIN_FILENO; | |
470 | ||
80876c20 LP |
471 | default: |
472 | assert_not_reached("Unknown input type"); | |
473 | } | |
474 | } | |
475 | ||
a34ceba6 LP |
476 | static int setup_output( |
477 | Unit *unit, | |
478 | const ExecContext *context, | |
479 | const ExecParameters *params, | |
480 | int fileno, | |
481 | int socket_fd, | |
52c239d7 | 482 | int named_iofds[3], |
a34ceba6 | 483 | const char *ident, |
7bce046b LP |
484 | uid_t uid, |
485 | gid_t gid, | |
486 | dev_t *journal_stream_dev, | |
487 | ino_t *journal_stream_ino) { | |
a34ceba6 | 488 | |
4f2d528d LP |
489 | ExecOutput o; |
490 | ExecInput i; | |
47c1d80d | 491 | int r; |
4f2d528d | 492 | |
f2341e0a | 493 | assert(unit); |
80876c20 | 494 | assert(context); |
a34ceba6 | 495 | assert(params); |
80876c20 | 496 | assert(ident); |
7bce046b LP |
497 | assert(journal_stream_dev); |
498 | assert(journal_stream_ino); | |
80876c20 | 499 | |
a34ceba6 LP |
500 | if (fileno == STDOUT_FILENO && params->stdout_fd >= 0) { |
501 | ||
502 | if (dup2(params->stdout_fd, STDOUT_FILENO) < 0) | |
503 | return -errno; | |
504 | ||
505 | return STDOUT_FILENO; | |
506 | } | |
507 | ||
508 | if (fileno == STDERR_FILENO && params->stderr_fd >= 0) { | |
509 | if (dup2(params->stderr_fd, STDERR_FILENO) < 0) | |
510 | return -errno; | |
511 | ||
512 | return STDERR_FILENO; | |
513 | } | |
514 | ||
c39f1ce2 | 515 | i = fixup_input(context->std_input, socket_fd, params->flags & EXEC_APPLY_TTY_STDIN); |
03fd9c49 | 516 | o = fixup_output(context->std_output, socket_fd); |
4f2d528d | 517 | |
eb17e935 MS |
518 | if (fileno == STDERR_FILENO) { |
519 | ExecOutput e; | |
520 | e = fixup_output(context->std_error, socket_fd); | |
80876c20 | 521 | |
eb17e935 MS |
522 | /* This expects the input and output are already set up */ |
523 | ||
524 | /* Don't change the stderr file descriptor if we inherit all | |
525 | * the way and are not on a tty */ | |
526 | if (e == EXEC_OUTPUT_INHERIT && | |
527 | o == EXEC_OUTPUT_INHERIT && | |
528 | i == EXEC_INPUT_NULL && | |
529 | !is_terminal_input(context->std_input) && | |
530 | getppid () != 1) | |
531 | return fileno; | |
532 | ||
533 | /* Duplicate from stdout if possible */ | |
52c239d7 | 534 | if ((e == o && e != EXEC_OUTPUT_NAMED_FD) || e == EXEC_OUTPUT_INHERIT) |
eb17e935 | 535 | return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno; |
071830ff | 536 | |
eb17e935 | 537 | o = e; |
80876c20 | 538 | |
eb17e935 | 539 | } else if (o == EXEC_OUTPUT_INHERIT) { |
21d21ea4 LP |
540 | /* If input got downgraded, inherit the original value */ |
541 | if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input)) | |
1e22b5cd | 542 | return open_terminal_as(exec_context_tty_path(context), O_WRONLY, fileno); |
21d21ea4 | 543 | |
acb591e4 | 544 | /* If the input is connected to anything that's not a /dev/null, inherit that... */ |
ff876e28 | 545 | if (i != EXEC_INPUT_NULL) |
eb17e935 | 546 | return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno; |
071830ff | 547 | |
acb591e4 LP |
548 | /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */ |
549 | if (getppid() != 1) | |
eb17e935 | 550 | return fileno; |
94f04347 | 551 | |
eb17e935 MS |
552 | /* We need to open /dev/null here anew, to get the right access mode. */ |
553 | return open_null_as(O_WRONLY, fileno); | |
071830ff | 554 | } |
94f04347 | 555 | |
eb17e935 | 556 | switch (o) { |
80876c20 LP |
557 | |
558 | case EXEC_OUTPUT_NULL: | |
eb17e935 | 559 | return open_null_as(O_WRONLY, fileno); |
80876c20 LP |
560 | |
561 | case EXEC_OUTPUT_TTY: | |
4f2d528d | 562 | if (is_terminal_input(i)) |
eb17e935 | 563 | return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno; |
80876c20 LP |
564 | |
565 | /* We don't reset the terminal if this is just about output */ | |
1e22b5cd | 566 | return open_terminal_as(exec_context_tty_path(context), O_WRONLY, fileno); |
80876c20 LP |
567 | |
568 | case EXEC_OUTPUT_SYSLOG: | |
28dbc1e8 | 569 | case EXEC_OUTPUT_SYSLOG_AND_CONSOLE: |
9a6bca7a | 570 | case EXEC_OUTPUT_KMSG: |
28dbc1e8 | 571 | case EXEC_OUTPUT_KMSG_AND_CONSOLE: |
706343f4 LP |
572 | case EXEC_OUTPUT_JOURNAL: |
573 | case EXEC_OUTPUT_JOURNAL_AND_CONSOLE: | |
7a1ab780 | 574 | r = connect_logger_as(unit, context, o, ident, fileno, uid, gid); |
47c1d80d | 575 | if (r < 0) { |
f2341e0a | 576 | log_unit_error_errno(unit, r, "Failed to connect %s to the journal socket, ignoring: %m", fileno == STDOUT_FILENO ? "stdout" : "stderr"); |
eb17e935 | 577 | r = open_null_as(O_WRONLY, fileno); |
7bce046b LP |
578 | } else { |
579 | struct stat st; | |
580 | ||
581 | /* If we connected this fd to the journal via a stream, patch the device/inode into the passed | |
582 | * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits | |
583 | * services to detect whether they are connected to the journal or not. */ | |
584 | ||
585 | if (fstat(fileno, &st) >= 0) { | |
586 | *journal_stream_dev = st.st_dev; | |
587 | *journal_stream_ino = st.st_ino; | |
588 | } | |
47c1d80d MS |
589 | } |
590 | return r; | |
4f2d528d LP |
591 | |
592 | case EXEC_OUTPUT_SOCKET: | |
593 | assert(socket_fd >= 0); | |
eb17e935 | 594 | return dup2(socket_fd, fileno) < 0 ? -errno : fileno; |
94f04347 | 595 | |
52c239d7 LB |
596 | case EXEC_OUTPUT_NAMED_FD: |
597 | (void) fd_nonblock(named_iofds[fileno], false); | |
598 | return dup2(named_iofds[fileno], fileno) < 0 ? -errno : fileno; | |
599 | ||
94f04347 | 600 | default: |
80876c20 | 601 | assert_not_reached("Unknown error type"); |
94f04347 | 602 | } |
071830ff LP |
603 | } |
604 | ||
02a51aba LP |
605 | static int chown_terminal(int fd, uid_t uid) { |
606 | struct stat st; | |
607 | ||
608 | assert(fd >= 0); | |
02a51aba | 609 | |
1ff74fb6 LP |
610 | /* Before we chown/chmod the TTY, let's ensure this is actually a tty */ |
611 | if (isatty(fd) < 1) | |
612 | return 0; | |
613 | ||
02a51aba | 614 | /* This might fail. What matters are the results. */ |
bab45044 LP |
615 | (void) fchown(fd, uid, -1); |
616 | (void) fchmod(fd, TTY_MODE); | |
02a51aba LP |
617 | |
618 | if (fstat(fd, &st) < 0) | |
619 | return -errno; | |
620 | ||
d8b4e2e9 | 621 | if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE) |
02a51aba LP |
622 | return -EPERM; |
623 | ||
624 | return 0; | |
625 | } | |
626 | ||
3d18b167 LP |
627 | static int setup_confirm_stdio(int *_saved_stdin, int *_saved_stdout) { |
628 | _cleanup_close_ int fd = -1, saved_stdin = -1, saved_stdout = -1; | |
629 | int r; | |
80876c20 | 630 | |
80876c20 LP |
631 | assert(_saved_stdin); |
632 | assert(_saved_stdout); | |
633 | ||
af6da548 LP |
634 | saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3); |
635 | if (saved_stdin < 0) | |
636 | return -errno; | |
80876c20 | 637 | |
af6da548 | 638 | saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3); |
3d18b167 LP |
639 | if (saved_stdout < 0) |
640 | return -errno; | |
80876c20 | 641 | |
af6da548 LP |
642 | fd = acquire_terminal( |
643 | "/dev/console", | |
644 | false, | |
645 | false, | |
646 | false, | |
647 | DEFAULT_CONFIRM_USEC); | |
3d18b167 LP |
648 | if (fd < 0) |
649 | return fd; | |
80876c20 | 650 | |
af6da548 LP |
651 | r = chown_terminal(fd, getuid()); |
652 | if (r < 0) | |
3d18b167 | 653 | return r; |
02a51aba | 654 | |
3d18b167 LP |
655 | r = reset_terminal_fd(fd, true); |
656 | if (r < 0) | |
657 | return r; | |
80876c20 | 658 | |
3d18b167 LP |
659 | if (dup2(fd, STDIN_FILENO) < 0) |
660 | return -errno; | |
661 | ||
662 | if (dup2(fd, STDOUT_FILENO) < 0) | |
663 | return -errno; | |
80876c20 LP |
664 | |
665 | if (fd >= 2) | |
03e334a1 | 666 | safe_close(fd); |
3d18b167 | 667 | fd = -1; |
80876c20 LP |
668 | |
669 | *_saved_stdin = saved_stdin; | |
670 | *_saved_stdout = saved_stdout; | |
671 | ||
3d18b167 | 672 | saved_stdin = saved_stdout = -1; |
80876c20 | 673 | |
3d18b167 | 674 | return 0; |
80876c20 LP |
675 | } |
676 | ||
44b601bc | 677 | _printf_(1, 2) static int write_confirm_message(const char *format, ...) { |
03e334a1 | 678 | _cleanup_close_ int fd = -1; |
af6da548 | 679 | va_list ap; |
80876c20 | 680 | |
af6da548 | 681 | assert(format); |
80876c20 | 682 | |
af6da548 LP |
683 | fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC); |
684 | if (fd < 0) | |
685 | return fd; | |
80876c20 | 686 | |
af6da548 LP |
687 | va_start(ap, format); |
688 | vdprintf(fd, format, ap); | |
689 | va_end(ap); | |
80876c20 | 690 | |
af6da548 LP |
691 | return 0; |
692 | } | |
80876c20 | 693 | |
3d18b167 | 694 | static int restore_confirm_stdio(int *saved_stdin, int *saved_stdout) { |
af6da548 | 695 | int r = 0; |
80876c20 | 696 | |
af6da548 LP |
697 | assert(saved_stdin); |
698 | assert(saved_stdout); | |
699 | ||
700 | release_terminal(); | |
701 | ||
702 | if (*saved_stdin >= 0) | |
80876c20 | 703 | if (dup2(*saved_stdin, STDIN_FILENO) < 0) |
af6da548 | 704 | r = -errno; |
80876c20 | 705 | |
af6da548 | 706 | if (*saved_stdout >= 0) |
80876c20 | 707 | if (dup2(*saved_stdout, STDOUT_FILENO) < 0) |
af6da548 | 708 | r = -errno; |
80876c20 | 709 | |
3d18b167 LP |
710 | *saved_stdin = safe_close(*saved_stdin); |
711 | *saved_stdout = safe_close(*saved_stdout); | |
af6da548 LP |
712 | |
713 | return r; | |
714 | } | |
715 | ||
716 | static int ask_for_confirmation(char *response, char **argv) { | |
717 | int saved_stdout = -1, saved_stdin = -1, r; | |
e1d75803 | 718 | _cleanup_free_ char *line = NULL; |
af6da548 LP |
719 | |
720 | r = setup_confirm_stdio(&saved_stdin, &saved_stdout); | |
721 | if (r < 0) | |
722 | return r; | |
723 | ||
724 | line = exec_command_line(argv); | |
725 | if (!line) | |
726 | return -ENOMEM; | |
727 | ||
418b9be5 | 728 | r = ask_char(response, "yns", "Execute %s? [Yes, No, Skip] ", line); |
af6da548 LP |
729 | |
730 | restore_confirm_stdio(&saved_stdin, &saved_stdout); | |
731 | ||
732 | return r; | |
80876c20 LP |
733 | } |
734 | ||
4d885bd3 DH |
735 | static int get_fixed_user(const ExecContext *c, const char **user, |
736 | uid_t *uid, gid_t *gid, | |
737 | const char **home, const char **shell) { | |
81a2b7ce | 738 | int r; |
4d885bd3 | 739 | const char *name; |
81a2b7ce | 740 | |
4d885bd3 | 741 | assert(c); |
81a2b7ce | 742 | |
4d885bd3 DH |
743 | if (!c->user) |
744 | return 0; | |
81a2b7ce | 745 | |
4d885bd3 DH |
746 | /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway |
747 | * (i.e. are "/" or "/bin/nologin"). */ | |
81a2b7ce | 748 | |
4d885bd3 DH |
749 | name = c->user; |
750 | r = get_user_creds_clean(&name, uid, gid, home, shell); | |
751 | if (r < 0) | |
752 | return r; | |
81a2b7ce | 753 | |
4d885bd3 DH |
754 | *user = name; |
755 | return 0; | |
756 | } | |
757 | ||
758 | static int get_fixed_group(const ExecContext *c, const char **group, gid_t *gid) { | |
759 | int r; | |
760 | const char *name; | |
761 | ||
762 | assert(c); | |
763 | ||
764 | if (!c->group) | |
765 | return 0; | |
766 | ||
767 | name = c->group; | |
768 | r = get_group_creds(&name, gid); | |
769 | if (r < 0) | |
770 | return r; | |
771 | ||
772 | *group = name; | |
773 | return 0; | |
774 | } | |
775 | ||
cdc5d5c5 DH |
776 | static int get_supplementary_groups(const ExecContext *c, const char *user, |
777 | const char *group, gid_t gid, | |
778 | gid_t **supplementary_gids, int *ngids) { | |
4d885bd3 DH |
779 | char **i; |
780 | int r, k = 0; | |
781 | int ngroups_max; | |
782 | bool keep_groups = false; | |
783 | gid_t *groups = NULL; | |
784 | _cleanup_free_ gid_t *l_gids = NULL; | |
785 | ||
786 | assert(c); | |
787 | ||
bbeea271 DH |
788 | /* |
789 | * If user is given, then lookup GID and supplementary groups list. | |
790 | * We avoid NSS lookups for gid=0. Also we have to initialize groups | |
cdc5d5c5 DH |
791 | * here and as early as possible so we keep the list of supplementary |
792 | * groups of the caller. | |
bbeea271 DH |
793 | */ |
794 | if (user && gid_is_valid(gid) && gid != 0) { | |
795 | /* First step, initialize groups from /etc/groups */ | |
796 | if (initgroups(user, gid) < 0) | |
797 | return -errno; | |
798 | ||
799 | keep_groups = true; | |
800 | } | |
801 | ||
4d885bd3 DH |
802 | if (!c->supplementary_groups) |
803 | return 0; | |
804 | ||
366ddd25 DH |
805 | /* |
806 | * If SupplementaryGroups= was passed then NGROUPS_MAX has to | |
807 | * be positive, otherwise fail. | |
808 | */ | |
809 | errno = 0; | |
810 | ngroups_max = (int) sysconf(_SC_NGROUPS_MAX); | |
811 | if (ngroups_max <= 0) { | |
812 | if (errno > 0) | |
813 | return -errno; | |
814 | else | |
815 | return -EOPNOTSUPP; /* For all other values */ | |
816 | } | |
817 | ||
4d885bd3 DH |
818 | l_gids = new(gid_t, ngroups_max); |
819 | if (!l_gids) | |
820 | return -ENOMEM; | |
81a2b7ce | 821 | |
4d885bd3 DH |
822 | if (keep_groups) { |
823 | /* | |
824 | * Lookup the list of groups that the user belongs to, we | |
825 | * avoid NSS lookups here too for gid=0. | |
826 | */ | |
827 | k = ngroups_max; | |
828 | if (getgrouplist(user, gid, l_gids, &k) < 0) | |
829 | return -EINVAL; | |
830 | } else | |
831 | k = 0; | |
81a2b7ce | 832 | |
4d885bd3 DH |
833 | STRV_FOREACH(i, c->supplementary_groups) { |
834 | const char *g; | |
81a2b7ce | 835 | |
4d885bd3 DH |
836 | if (k >= ngroups_max) |
837 | return -E2BIG; | |
81a2b7ce | 838 | |
4d885bd3 DH |
839 | g = *i; |
840 | r = get_group_creds(&g, l_gids+k); | |
841 | if (r < 0) | |
842 | return r; | |
81a2b7ce | 843 | |
4d885bd3 DH |
844 | k++; |
845 | } | |
81a2b7ce | 846 | |
4d885bd3 DH |
847 | /* |
848 | * Sets ngids to zero to drop all supplementary groups, happens | |
849 | * when we are under root and SupplementaryGroups= is empty. | |
850 | */ | |
851 | if (k == 0) { | |
852 | *ngids = 0; | |
853 | return 0; | |
854 | } | |
81a2b7ce | 855 | |
4d885bd3 DH |
856 | /* Otherwise get the final list of supplementary groups */ |
857 | groups = memdup(l_gids, sizeof(gid_t) * k); | |
858 | if (!groups) | |
859 | return -ENOMEM; | |
860 | ||
861 | *supplementary_gids = groups; | |
862 | *ngids = k; | |
863 | ||
864 | groups = NULL; | |
865 | ||
866 | return 0; | |
867 | } | |
868 | ||
869 | static int enforce_groups(const ExecContext *context, gid_t gid, | |
870 | gid_t *supplementary_gids, int ngids) { | |
871 | int r; | |
872 | ||
873 | assert(context); | |
874 | ||
875 | /* Handle SupplementaryGroups= even if it is empty */ | |
876 | if (context->supplementary_groups) { | |
877 | r = maybe_setgroups(ngids, supplementary_gids); | |
878 | if (r < 0) | |
97f0e76f | 879 | return r; |
4d885bd3 | 880 | } |
81a2b7ce | 881 | |
4d885bd3 DH |
882 | if (gid_is_valid(gid)) { |
883 | /* Then set our gids */ | |
884 | if (setresgid(gid, gid, gid) < 0) | |
885 | return -errno; | |
81a2b7ce LP |
886 | } |
887 | ||
888 | return 0; | |
889 | } | |
890 | ||
891 | static int enforce_user(const ExecContext *context, uid_t uid) { | |
81a2b7ce LP |
892 | assert(context); |
893 | ||
4d885bd3 DH |
894 | if (!uid_is_valid(uid)) |
895 | return 0; | |
896 | ||
479050b3 | 897 | /* Sets (but doesn't look up) the uid and make sure we keep the |
81a2b7ce LP |
898 | * capabilities while doing so. */ |
899 | ||
479050b3 | 900 | if (context->capability_ambient_set != 0) { |
81a2b7ce LP |
901 | |
902 | /* First step: If we need to keep capabilities but | |
903 | * drop privileges we need to make sure we keep our | |
cbb21cca | 904 | * caps, while we drop privileges. */ |
693ced48 | 905 | if (uid != 0) { |
cbb21cca | 906 | int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS; |
693ced48 LP |
907 | |
908 | if (prctl(PR_GET_SECUREBITS) != sb) | |
909 | if (prctl(PR_SET_SECUREBITS, sb) < 0) | |
910 | return -errno; | |
911 | } | |
81a2b7ce LP |
912 | } |
913 | ||
479050b3 | 914 | /* Second step: actually set the uids */ |
81a2b7ce LP |
915 | if (setresuid(uid, uid, uid) < 0) |
916 | return -errno; | |
917 | ||
918 | /* At this point we should have all necessary capabilities but | |
919 | are otherwise a normal user. However, the caps might got | |
920 | corrupted due to the setresuid() so we need clean them up | |
921 | later. This is done outside of this call. */ | |
922 | ||
923 | return 0; | |
924 | } | |
925 | ||
5b6319dc LP |
926 | #ifdef HAVE_PAM |
927 | ||
928 | static int null_conv( | |
929 | int num_msg, | |
930 | const struct pam_message **msg, | |
931 | struct pam_response **resp, | |
932 | void *appdata_ptr) { | |
933 | ||
934 | /* We don't support conversations */ | |
935 | ||
936 | return PAM_CONV_ERR; | |
937 | } | |
938 | ||
cefc33ae LP |
939 | #endif |
940 | ||
5b6319dc LP |
941 | static int setup_pam( |
942 | const char *name, | |
943 | const char *user, | |
940c5210 | 944 | uid_t uid, |
2d6fce8d | 945 | gid_t gid, |
5b6319dc | 946 | const char *tty, |
2065ca69 | 947 | char ***env, |
5b6319dc LP |
948 | int fds[], unsigned n_fds) { |
949 | ||
cefc33ae LP |
950 | #ifdef HAVE_PAM |
951 | ||
5b6319dc LP |
952 | static const struct pam_conv conv = { |
953 | .conv = null_conv, | |
954 | .appdata_ptr = NULL | |
955 | }; | |
956 | ||
2d7c6aa2 | 957 | _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL; |
5b6319dc | 958 | pam_handle_t *handle = NULL; |
d6e5f3ad | 959 | sigset_t old_ss; |
7bb70b6e | 960 | int pam_code = PAM_SUCCESS, r; |
84eada2f | 961 | char **nv, **e = NULL; |
5b6319dc LP |
962 | bool close_session = false; |
963 | pid_t pam_pid = 0, parent_pid; | |
970edce6 | 964 | int flags = 0; |
5b6319dc LP |
965 | |
966 | assert(name); | |
967 | assert(user); | |
2065ca69 | 968 | assert(env); |
5b6319dc LP |
969 | |
970 | /* We set up PAM in the parent process, then fork. The child | |
35b8ca3a | 971 | * will then stay around until killed via PR_GET_PDEATHSIG or |
5b6319dc LP |
972 | * systemd via the cgroup logic. It will then remove the PAM |
973 | * session again. The parent process will exec() the actual | |
974 | * daemon. We do things this way to ensure that the main PID | |
975 | * of the daemon is the one we initially fork()ed. */ | |
976 | ||
7bb70b6e LP |
977 | r = barrier_create(&barrier); |
978 | if (r < 0) | |
2d7c6aa2 DH |
979 | goto fail; |
980 | ||
553d2243 | 981 | if (log_get_max_level() < LOG_DEBUG) |
970edce6 ZJS |
982 | flags |= PAM_SILENT; |
983 | ||
f546241b ZJS |
984 | pam_code = pam_start(name, user, &conv, &handle); |
985 | if (pam_code != PAM_SUCCESS) { | |
5b6319dc LP |
986 | handle = NULL; |
987 | goto fail; | |
988 | } | |
989 | ||
f546241b ZJS |
990 | if (tty) { |
991 | pam_code = pam_set_item(handle, PAM_TTY, tty); | |
992 | if (pam_code != PAM_SUCCESS) | |
5b6319dc | 993 | goto fail; |
f546241b | 994 | } |
5b6319dc | 995 | |
84eada2f JW |
996 | STRV_FOREACH(nv, *env) { |
997 | pam_code = pam_putenv(handle, *nv); | |
2065ca69 JW |
998 | if (pam_code != PAM_SUCCESS) |
999 | goto fail; | |
1000 | } | |
1001 | ||
970edce6 | 1002 | pam_code = pam_acct_mgmt(handle, flags); |
f546241b | 1003 | if (pam_code != PAM_SUCCESS) |
5b6319dc LP |
1004 | goto fail; |
1005 | ||
970edce6 | 1006 | pam_code = pam_open_session(handle, flags); |
f546241b | 1007 | if (pam_code != PAM_SUCCESS) |
5b6319dc LP |
1008 | goto fail; |
1009 | ||
1010 | close_session = true; | |
1011 | ||
f546241b ZJS |
1012 | e = pam_getenvlist(handle); |
1013 | if (!e) { | |
5b6319dc LP |
1014 | pam_code = PAM_BUF_ERR; |
1015 | goto fail; | |
1016 | } | |
1017 | ||
1018 | /* Block SIGTERM, so that we know that it won't get lost in | |
1019 | * the child */ | |
ce30c8dc | 1020 | |
72c0a2c2 | 1021 | assert_se(sigprocmask_many(SIG_BLOCK, &old_ss, SIGTERM, -1) >= 0); |
5b6319dc LP |
1022 | |
1023 | parent_pid = getpid(); | |
1024 | ||
f546241b | 1025 | pam_pid = fork(); |
7bb70b6e LP |
1026 | if (pam_pid < 0) { |
1027 | r = -errno; | |
5b6319dc | 1028 | goto fail; |
7bb70b6e | 1029 | } |
5b6319dc LP |
1030 | |
1031 | if (pam_pid == 0) { | |
7bb70b6e | 1032 | int sig, ret = EXIT_PAM; |
5b6319dc LP |
1033 | |
1034 | /* The child's job is to reset the PAM session on | |
1035 | * termination */ | |
2d7c6aa2 | 1036 | barrier_set_role(&barrier, BARRIER_CHILD); |
5b6319dc LP |
1037 | |
1038 | /* This string must fit in 10 chars (i.e. the length | |
5d6b1584 LP |
1039 | * of "/sbin/init"), to look pretty in /bin/ps */ |
1040 | rename_process("(sd-pam)"); | |
5b6319dc LP |
1041 | |
1042 | /* Make sure we don't keep open the passed fds in this | |
1043 | child. We assume that otherwise only those fds are | |
1044 | open here that have been opened by PAM. */ | |
1045 | close_many(fds, n_fds); | |
1046 | ||
940c5210 AK |
1047 | /* Drop privileges - we don't need any to pam_close_session |
1048 | * and this will make PR_SET_PDEATHSIG work in most cases. | |
1049 | * If this fails, ignore the error - but expect sd-pam threads | |
1050 | * to fail to exit normally */ | |
2d6fce8d | 1051 | |
97f0e76f LP |
1052 | r = maybe_setgroups(0, NULL); |
1053 | if (r < 0) | |
1054 | log_warning_errno(r, "Failed to setgroups() in sd-pam: %m"); | |
2d6fce8d LP |
1055 | if (setresgid(gid, gid, gid) < 0) |
1056 | log_warning_errno(errno, "Failed to setresgid() in sd-pam: %m"); | |
940c5210 | 1057 | if (setresuid(uid, uid, uid) < 0) |
2d6fce8d | 1058 | log_warning_errno(errno, "Failed to setresuid() in sd-pam: %m"); |
940c5210 | 1059 | |
ce30c8dc LP |
1060 | (void) ignore_signals(SIGPIPE, -1); |
1061 | ||
940c5210 AK |
1062 | /* Wait until our parent died. This will only work if |
1063 | * the above setresuid() succeeds, otherwise the kernel | |
1064 | * will not allow unprivileged parents kill their privileged | |
1065 | * children this way. We rely on the control groups kill logic | |
5b6319dc LP |
1066 | * to do the rest for us. */ |
1067 | if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) | |
1068 | goto child_finish; | |
1069 | ||
2d7c6aa2 DH |
1070 | /* Tell the parent that our setup is done. This is especially |
1071 | * important regarding dropping privileges. Otherwise, unit | |
1072 | * setup might race against our setresuid(2) call. */ | |
1073 | barrier_place(&barrier); | |
1074 | ||
5b6319dc LP |
1075 | /* Check if our parent process might already have |
1076 | * died? */ | |
1077 | if (getppid() == parent_pid) { | |
d6e5f3ad DM |
1078 | sigset_t ss; |
1079 | ||
1080 | assert_se(sigemptyset(&ss) >= 0); | |
1081 | assert_se(sigaddset(&ss, SIGTERM) >= 0); | |
1082 | ||
3dead8d9 LP |
1083 | for (;;) { |
1084 | if (sigwait(&ss, &sig) < 0) { | |
1085 | if (errno == EINTR) | |
1086 | continue; | |
1087 | ||
1088 | goto child_finish; | |
1089 | } | |
5b6319dc | 1090 | |
3dead8d9 LP |
1091 | assert(sig == SIGTERM); |
1092 | break; | |
1093 | } | |
5b6319dc LP |
1094 | } |
1095 | ||
3dead8d9 | 1096 | /* If our parent died we'll end the session */ |
f546241b | 1097 | if (getppid() != parent_pid) { |
970edce6 | 1098 | pam_code = pam_close_session(handle, flags); |
f546241b | 1099 | if (pam_code != PAM_SUCCESS) |
5b6319dc | 1100 | goto child_finish; |
f546241b | 1101 | } |
5b6319dc | 1102 | |
7bb70b6e | 1103 | ret = 0; |
5b6319dc LP |
1104 | |
1105 | child_finish: | |
970edce6 | 1106 | pam_end(handle, pam_code | flags); |
7bb70b6e | 1107 | _exit(ret); |
5b6319dc LP |
1108 | } |
1109 | ||
2d7c6aa2 DH |
1110 | barrier_set_role(&barrier, BARRIER_PARENT); |
1111 | ||
5b6319dc LP |
1112 | /* If the child was forked off successfully it will do all the |
1113 | * cleanups, so forget about the handle here. */ | |
1114 | handle = NULL; | |
1115 | ||
3b8bddde | 1116 | /* Unblock SIGTERM again in the parent */ |
72c0a2c2 | 1117 | assert_se(sigprocmask(SIG_SETMASK, &old_ss, NULL) >= 0); |
5b6319dc LP |
1118 | |
1119 | /* We close the log explicitly here, since the PAM modules | |
1120 | * might have opened it, but we don't want this fd around. */ | |
1121 | closelog(); | |
1122 | ||
2d7c6aa2 DH |
1123 | /* Synchronously wait for the child to initialize. We don't care for |
1124 | * errors as we cannot recover. However, warn loudly if it happens. */ | |
1125 | if (!barrier_place_and_sync(&barrier)) | |
1126 | log_error("PAM initialization failed"); | |
1127 | ||
2065ca69 JW |
1128 | strv_free(*env); |
1129 | *env = e; | |
aa87e624 | 1130 | |
5b6319dc LP |
1131 | return 0; |
1132 | ||
1133 | fail: | |
970edce6 ZJS |
1134 | if (pam_code != PAM_SUCCESS) { |
1135 | log_error("PAM failed: %s", pam_strerror(handle, pam_code)); | |
7bb70b6e LP |
1136 | r = -EPERM; /* PAM errors do not map to errno */ |
1137 | } else | |
1138 | log_error_errno(r, "PAM failed: %m"); | |
9ba35398 | 1139 | |
5b6319dc LP |
1140 | if (handle) { |
1141 | if (close_session) | |
970edce6 | 1142 | pam_code = pam_close_session(handle, flags); |
5b6319dc | 1143 | |
970edce6 | 1144 | pam_end(handle, pam_code | flags); |
5b6319dc LP |
1145 | } |
1146 | ||
1147 | strv_free(e); | |
5b6319dc LP |
1148 | closelog(); |
1149 | ||
7bb70b6e | 1150 | return r; |
cefc33ae LP |
1151 | #else |
1152 | return 0; | |
5b6319dc | 1153 | #endif |
cefc33ae | 1154 | } |
5b6319dc | 1155 | |
5d6b1584 LP |
1156 | static void rename_process_from_path(const char *path) { |
1157 | char process_name[11]; | |
1158 | const char *p; | |
1159 | size_t l; | |
1160 | ||
1161 | /* This resulting string must fit in 10 chars (i.e. the length | |
1162 | * of "/sbin/init") to look pretty in /bin/ps */ | |
1163 | ||
2b6bf07d | 1164 | p = basename(path); |
5d6b1584 LP |
1165 | if (isempty(p)) { |
1166 | rename_process("(...)"); | |
1167 | return; | |
1168 | } | |
1169 | ||
1170 | l = strlen(p); | |
1171 | if (l > 8) { | |
1172 | /* The end of the process name is usually more | |
1173 | * interesting, since the first bit might just be | |
1174 | * "systemd-" */ | |
1175 | p = p + l - 8; | |
1176 | l = 8; | |
1177 | } | |
1178 | ||
1179 | process_name[0] = '('; | |
1180 | memcpy(process_name+1, p, l); | |
1181 | process_name[1+l] = ')'; | |
1182 | process_name[1+l+1] = 0; | |
1183 | ||
1184 | rename_process(process_name); | |
1185 | } | |
1186 | ||
c0467cf3 | 1187 | #ifdef HAVE_SECCOMP |
17df7223 | 1188 | |
83f12b27 | 1189 | static bool skip_seccomp_unavailable(const Unit* u, const char* msg) { |
f673b62d LP |
1190 | |
1191 | if (is_seccomp_available()) | |
1192 | return false; | |
1193 | ||
1194 | log_open(); | |
1195 | log_unit_debug(u, "SECCOMP features not detected in the kernel, skipping %s", msg); | |
1196 | log_close(); | |
1197 | return true; | |
83f12b27 FS |
1198 | } |
1199 | ||
1200 | static int apply_seccomp(const Unit* u, const ExecContext *c) { | |
17df7223 | 1201 | uint32_t negative_action, action; |
8d7b0c8f | 1202 | scmp_filter_ctx seccomp; |
c0467cf3 RC |
1203 | Iterator i; |
1204 | void *id; | |
17df7223 | 1205 | int r; |
8351ceae | 1206 | |
c0467cf3 | 1207 | assert(c); |
8351ceae | 1208 | |
83f12b27 FS |
1209 | if (skip_seccomp_unavailable(u, "syscall filtering")) |
1210 | return 0; | |
1211 | ||
17df7223 LP |
1212 | negative_action = c->syscall_errno == 0 ? SCMP_ACT_KILL : SCMP_ACT_ERRNO(c->syscall_errno); |
1213 | ||
1214 | seccomp = seccomp_init(c->syscall_whitelist ? negative_action : SCMP_ACT_ALLOW); | |
1215 | if (!seccomp) | |
1216 | return -ENOMEM; | |
8351ceae | 1217 | |
e9642be2 LP |
1218 | if (c->syscall_archs) { |
1219 | ||
1220 | SET_FOREACH(id, c->syscall_archs, i) { | |
1221 | r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1); | |
1222 | if (r == -EEXIST) | |
1223 | continue; | |
7c66bae2 LP |
1224 | if (r < 0) |
1225 | goto finish; | |
e9642be2 | 1226 | } |
e9642be2 | 1227 | |
7c66bae2 | 1228 | } else { |
e9642be2 | 1229 | r = seccomp_add_secondary_archs(seccomp); |
7c66bae2 LP |
1230 | if (r < 0) |
1231 | goto finish; | |
57183d11 | 1232 | } |
8351ceae | 1233 | |
57183d11 | 1234 | action = c->syscall_whitelist ? SCMP_ACT_ALLOW : negative_action; |
17df7223 LP |
1235 | SET_FOREACH(id, c->syscall_filter, i) { |
1236 | r = seccomp_rule_add(seccomp, action, PTR_TO_INT(id) - 1, 0); | |
7c66bae2 LP |
1237 | if (r < 0) |
1238 | goto finish; | |
c0467cf3 | 1239 | } |
8351ceae | 1240 | |
7c66bae2 LP |
1241 | r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0); |
1242 | if (r < 0) | |
1243 | goto finish; | |
1244 | ||
17df7223 | 1245 | r = seccomp_load(seccomp); |
7c66bae2 LP |
1246 | |
1247 | finish: | |
17df7223 | 1248 | seccomp_release(seccomp); |
4298d0b5 LP |
1249 | return r; |
1250 | } | |
1251 | ||
83f12b27 | 1252 | static int apply_address_families(const Unit* u, const ExecContext *c) { |
8d7b0c8f | 1253 | scmp_filter_ctx seccomp; |
4298d0b5 LP |
1254 | Iterator i; |
1255 | int r; | |
1256 | ||
1257 | assert(c); | |
1258 | ||
83f12b27 FS |
1259 | if (skip_seccomp_unavailable(u, "RestrictAddressFamilies=")) |
1260 | return 0; | |
1261 | ||
8d7b0c8f | 1262 | r = seccomp_init_conservative(&seccomp, SCMP_ACT_ALLOW); |
4298d0b5 | 1263 | if (r < 0) |
8d7b0c8f | 1264 | return r; |
4298d0b5 LP |
1265 | |
1266 | if (c->address_families_whitelist) { | |
1267 | int af, first = 0, last = 0; | |
1268 | void *afp; | |
1269 | ||
1270 | /* If this is a whitelist, we first block the address | |
1271 | * families that are out of range and then everything | |
1272 | * that is not in the set. First, we find the lowest | |
1273 | * and highest address family in the set. */ | |
1274 | ||
1275 | SET_FOREACH(afp, c->address_families, i) { | |
1276 | af = PTR_TO_INT(afp); | |
17df7223 | 1277 | |
4298d0b5 LP |
1278 | if (af <= 0 || af >= af_max()) |
1279 | continue; | |
1280 | ||
1281 | if (first == 0 || af < first) | |
1282 | first = af; | |
1283 | ||
1284 | if (last == 0 || af > last) | |
1285 | last = af; | |
1286 | } | |
1287 | ||
1288 | assert((first == 0) == (last == 0)); | |
1289 | ||
1290 | if (first == 0) { | |
1291 | ||
1292 | /* No entries in the valid range, block everything */ | |
1293 | r = seccomp_rule_add( | |
1294 | seccomp, | |
1295 | SCMP_ACT_ERRNO(EPROTONOSUPPORT), | |
1296 | SCMP_SYS(socket), | |
1297 | 0); | |
1298 | if (r < 0) | |
1299 | goto finish; | |
1300 | ||
1301 | } else { | |
1302 | ||
1303 | /* Block everything below the first entry */ | |
1304 | r = seccomp_rule_add( | |
1305 | seccomp, | |
1306 | SCMP_ACT_ERRNO(EPROTONOSUPPORT), | |
1307 | SCMP_SYS(socket), | |
1308 | 1, | |
1309 | SCMP_A0(SCMP_CMP_LT, first)); | |
1310 | if (r < 0) | |
1311 | goto finish; | |
1312 | ||
1313 | /* Block everything above the last entry */ | |
1314 | r = seccomp_rule_add( | |
1315 | seccomp, | |
1316 | SCMP_ACT_ERRNO(EPROTONOSUPPORT), | |
1317 | SCMP_SYS(socket), | |
1318 | 1, | |
1319 | SCMP_A0(SCMP_CMP_GT, last)); | |
1320 | if (r < 0) | |
1321 | goto finish; | |
1322 | ||
1323 | /* Block everything between the first and last | |
1324 | * entry */ | |
1325 | for (af = 1; af < af_max(); af++) { | |
1326 | ||
1327 | if (set_contains(c->address_families, INT_TO_PTR(af))) | |
1328 | continue; | |
1329 | ||
1330 | r = seccomp_rule_add( | |
1331 | seccomp, | |
1332 | SCMP_ACT_ERRNO(EPROTONOSUPPORT), | |
1333 | SCMP_SYS(socket), | |
1334 | 1, | |
1335 | SCMP_A0(SCMP_CMP_EQ, af)); | |
1336 | if (r < 0) | |
1337 | goto finish; | |
1338 | } | |
1339 | } | |
1340 | ||
1341 | } else { | |
1342 | void *af; | |
1343 | ||
1344 | /* If this is a blacklist, then generate one rule for | |
1345 | * each address family that are then combined in OR | |
1346 | * checks. */ | |
1347 | ||
1348 | SET_FOREACH(af, c->address_families, i) { | |
1349 | ||
1350 | r = seccomp_rule_add( | |
1351 | seccomp, | |
1352 | SCMP_ACT_ERRNO(EPROTONOSUPPORT), | |
1353 | SCMP_SYS(socket), | |
1354 | 1, | |
1355 | SCMP_A0(SCMP_CMP_EQ, PTR_TO_INT(af))); | |
1356 | if (r < 0) | |
1357 | goto finish; | |
1358 | } | |
1359 | } | |
1360 | ||
4298d0b5 LP |
1361 | r = seccomp_load(seccomp); |
1362 | ||
1363 | finish: | |
1364 | seccomp_release(seccomp); | |
17df7223 | 1365 | return r; |
8351ceae | 1366 | } |
4298d0b5 | 1367 | |
83f12b27 | 1368 | static int apply_memory_deny_write_execute(const Unit* u, const ExecContext *c) { |
8d7b0c8f | 1369 | scmp_filter_ctx seccomp; |
f3e43635 TM |
1370 | int r; |
1371 | ||
1372 | assert(c); | |
1373 | ||
83f12b27 FS |
1374 | if (skip_seccomp_unavailable(u, "MemoryDenyWriteExecute=")) |
1375 | return 0; | |
1376 | ||
8d7b0c8f | 1377 | r = seccomp_init_conservative(&seccomp, SCMP_ACT_ALLOW); |
72246c2a | 1378 | if (r < 0) |
8d7b0c8f | 1379 | return r; |
72246c2a | 1380 | |
f3e43635 TM |
1381 | r = seccomp_rule_add( |
1382 | seccomp, | |
abd84d4d | 1383 | SCMP_ACT_ERRNO(EPERM), |
f3e43635 TM |
1384 | SCMP_SYS(mmap), |
1385 | 1, | |
1386 | SCMP_A2(SCMP_CMP_MASKED_EQ, PROT_EXEC|PROT_WRITE, PROT_EXEC|PROT_WRITE)); | |
1387 | if (r < 0) | |
1388 | goto finish; | |
1389 | ||
1390 | r = seccomp_rule_add( | |
1391 | seccomp, | |
abd84d4d | 1392 | SCMP_ACT_ERRNO(EPERM), |
f3e43635 TM |
1393 | SCMP_SYS(mprotect), |
1394 | 1, | |
1395 | SCMP_A2(SCMP_CMP_MASKED_EQ, PROT_EXEC, PROT_EXEC)); | |
1396 | if (r < 0) | |
1397 | goto finish; | |
1398 | ||
d2ffa389 TM |
1399 | r = seccomp_rule_add( |
1400 | seccomp, | |
1401 | SCMP_ACT_ERRNO(EPERM), | |
1402 | SCMP_SYS(shmat), | |
1403 | 1, | |
1404 | SCMP_A2(SCMP_CMP_MASKED_EQ, SHM_EXEC, SHM_EXEC)); | |
1405 | if (r < 0) | |
1406 | goto finish; | |
1407 | ||
f3e43635 TM |
1408 | r = seccomp_load(seccomp); |
1409 | ||
1410 | finish: | |
1411 | seccomp_release(seccomp); | |
1412 | return r; | |
1413 | } | |
1414 | ||
83f12b27 | 1415 | static int apply_restrict_realtime(const Unit* u, const ExecContext *c) { |
f4170c67 LP |
1416 | static const int permitted_policies[] = { |
1417 | SCHED_OTHER, | |
1418 | SCHED_BATCH, | |
1419 | SCHED_IDLE, | |
1420 | }; | |
1421 | ||
8d7b0c8f | 1422 | scmp_filter_ctx seccomp; |
f4170c67 LP |
1423 | unsigned i; |
1424 | int r, p, max_policy = 0; | |
1425 | ||
1426 | assert(c); | |
1427 | ||
83f12b27 FS |
1428 | if (skip_seccomp_unavailable(u, "RestrictRealtime=")) |
1429 | return 0; | |
1430 | ||
8d7b0c8f | 1431 | r = seccomp_init_conservative(&seccomp, SCMP_ACT_ALLOW); |
72246c2a | 1432 | if (r < 0) |
8d7b0c8f | 1433 | return r; |
72246c2a | 1434 | |
f4170c67 LP |
1435 | /* Determine the highest policy constant we want to allow */ |
1436 | for (i = 0; i < ELEMENTSOF(permitted_policies); i++) | |
1437 | if (permitted_policies[i] > max_policy) | |
1438 | max_policy = permitted_policies[i]; | |
1439 | ||
1440 | /* Go through all policies with lower values than that, and block them -- unless they appear in the | |
1441 | * whitelist. */ | |
1442 | for (p = 0; p < max_policy; p++) { | |
1443 | bool good = false; | |
1444 | ||
1445 | /* Check if this is in the whitelist. */ | |
1446 | for (i = 0; i < ELEMENTSOF(permitted_policies); i++) | |
1447 | if (permitted_policies[i] == p) { | |
1448 | good = true; | |
1449 | break; | |
1450 | } | |
1451 | ||
1452 | if (good) | |
1453 | continue; | |
1454 | ||
1455 | /* Deny this policy */ | |
1456 | r = seccomp_rule_add( | |
1457 | seccomp, | |
1458 | SCMP_ACT_ERRNO(EPERM), | |
1459 | SCMP_SYS(sched_setscheduler), | |
1460 | 1, | |
1461 | SCMP_A1(SCMP_CMP_EQ, p)); | |
1462 | if (r < 0) | |
1463 | goto finish; | |
1464 | } | |
1465 | ||
1466 | /* Blacklist all other policies, i.e. the ones with higher values. Note that all comparisons are unsigned here, | |
1467 | * hence no need no check for < 0 values. */ | |
1468 | r = seccomp_rule_add( | |
1469 | seccomp, | |
1470 | SCMP_ACT_ERRNO(EPERM), | |
1471 | SCMP_SYS(sched_setscheduler), | |
1472 | 1, | |
1473 | SCMP_A1(SCMP_CMP_GT, max_policy)); | |
1474 | if (r < 0) | |
1475 | goto finish; | |
1476 | ||
f4170c67 LP |
1477 | r = seccomp_load(seccomp); |
1478 | ||
1479 | finish: | |
1480 | seccomp_release(seccomp); | |
1481 | return r; | |
1482 | } | |
1483 | ||
59e856c7 | 1484 | static int apply_protect_sysctl(const Unit *u, const ExecContext *c) { |
8d7b0c8f | 1485 | scmp_filter_ctx seccomp; |
59eeb84b LP |
1486 | int r; |
1487 | ||
1488 | assert(c); | |
1489 | ||
1490 | /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but | |
1491 | * let's protect even those systems where this is left on in the kernel. */ | |
1492 | ||
1493 | if (skip_seccomp_unavailable(u, "ProtectKernelTunables=")) | |
1494 | return 0; | |
1495 | ||
8d7b0c8f | 1496 | r = seccomp_init_conservative(&seccomp, SCMP_ACT_ALLOW); |
59eeb84b | 1497 | if (r < 0) |
8d7b0c8f | 1498 | return r; |
59eeb84b LP |
1499 | |
1500 | r = seccomp_rule_add( | |
1501 | seccomp, | |
1502 | SCMP_ACT_ERRNO(EPERM), | |
1503 | SCMP_SYS(_sysctl), | |
1504 | 0); | |
1505 | if (r < 0) | |
1506 | goto finish; | |
1507 | ||
59eeb84b LP |
1508 | r = seccomp_load(seccomp); |
1509 | ||
1510 | finish: | |
1511 | seccomp_release(seccomp); | |
1512 | return r; | |
1513 | } | |
1514 | ||
59e856c7 | 1515 | static int apply_protect_kernel_modules(const Unit *u, const ExecContext *c) { |
502d704e DH |
1516 | assert(c); |
1517 | ||
25a8d8a0 | 1518 | /* Turn off module syscalls on ProtectKernelModules=yes */ |
502d704e DH |
1519 | |
1520 | if (skip_seccomp_unavailable(u, "ProtectKernelModules=")) | |
1521 | return 0; | |
1522 | ||
a3be2849 | 1523 | return seccomp_load_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_MODULE, SCMP_ACT_ERRNO(EPERM)); |
502d704e DH |
1524 | } |
1525 | ||
59e856c7 | 1526 | static int apply_private_devices(const Unit *u, const ExecContext *c) { |
ba128bb8 LP |
1527 | assert(c); |
1528 | ||
8f81a5f6 | 1529 | /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */ |
ba128bb8 LP |
1530 | |
1531 | if (skip_seccomp_unavailable(u, "PrivateDevices=")) | |
1532 | return 0; | |
1533 | ||
a3be2849 | 1534 | return seccomp_load_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_RAW_IO, SCMP_ACT_ERRNO(EPERM)); |
ba128bb8 LP |
1535 | } |
1536 | ||
add00535 LP |
1537 | static int apply_restrict_namespaces(Unit *u, const ExecContext *c) { |
1538 | assert(c); | |
1539 | ||
1540 | if (!exec_context_restrict_namespaces_set(c)) | |
1541 | return 0; | |
1542 | ||
1543 | if (skip_seccomp_unavailable(u, "RestrictNamespaces=")) | |
1544 | return 0; | |
1545 | ||
1546 | return seccomp_restrict_namespaces(c->restrict_namespaces); | |
1547 | } | |
1548 | ||
c0467cf3 | 1549 | #endif |
8351ceae | 1550 | |
31a7eb86 ZJS |
1551 | static void do_idle_pipe_dance(int idle_pipe[4]) { |
1552 | assert(idle_pipe); | |
1553 | ||
54eb2300 LP |
1554 | idle_pipe[1] = safe_close(idle_pipe[1]); |
1555 | idle_pipe[2] = safe_close(idle_pipe[2]); | |
31a7eb86 ZJS |
1556 | |
1557 | if (idle_pipe[0] >= 0) { | |
1558 | int r; | |
1559 | ||
1560 | r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC); | |
1561 | ||
1562 | if (idle_pipe[3] >= 0 && r == 0 /* timeout */) { | |
c7cc737f LP |
1563 | ssize_t n; |
1564 | ||
31a7eb86 | 1565 | /* Signal systemd that we are bored and want to continue. */ |
c7cc737f LP |
1566 | n = write(idle_pipe[3], "x", 1); |
1567 | if (n > 0) | |
cd972d69 ZJS |
1568 | /* Wait for systemd to react to the signal above. */ |
1569 | fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC); | |
31a7eb86 ZJS |
1570 | } |
1571 | ||
54eb2300 | 1572 | idle_pipe[0] = safe_close(idle_pipe[0]); |
31a7eb86 ZJS |
1573 | |
1574 | } | |
1575 | ||
54eb2300 | 1576 | idle_pipe[3] = safe_close(idle_pipe[3]); |
31a7eb86 ZJS |
1577 | } |
1578 | ||
7cae38c4 | 1579 | static int build_environment( |
fd63e712 | 1580 | Unit *u, |
9fa95f85 | 1581 | const ExecContext *c, |
1e22b5cd | 1582 | const ExecParameters *p, |
7cae38c4 LP |
1583 | unsigned n_fds, |
1584 | const char *home, | |
1585 | const char *username, | |
1586 | const char *shell, | |
7bce046b LP |
1587 | dev_t journal_stream_dev, |
1588 | ino_t journal_stream_ino, | |
7cae38c4 LP |
1589 | char ***ret) { |
1590 | ||
1591 | _cleanup_strv_free_ char **our_env = NULL; | |
1592 | unsigned n_env = 0; | |
1593 | char *x; | |
1594 | ||
4b58153d | 1595 | assert(u); |
7cae38c4 LP |
1596 | assert(c); |
1597 | assert(ret); | |
1598 | ||
4b58153d | 1599 | our_env = new0(char*, 14); |
7cae38c4 LP |
1600 | if (!our_env) |
1601 | return -ENOMEM; | |
1602 | ||
1603 | if (n_fds > 0) { | |
8dd4c05b LP |
1604 | _cleanup_free_ char *joined = NULL; |
1605 | ||
ccd06097 | 1606 | if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid()) < 0) |
7cae38c4 LP |
1607 | return -ENOMEM; |
1608 | our_env[n_env++] = x; | |
1609 | ||
1610 | if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0) | |
1611 | return -ENOMEM; | |
1612 | our_env[n_env++] = x; | |
8dd4c05b | 1613 | |
1e22b5cd | 1614 | joined = strv_join(p->fd_names, ":"); |
8dd4c05b LP |
1615 | if (!joined) |
1616 | return -ENOMEM; | |
1617 | ||
605405c6 | 1618 | x = strjoin("LISTEN_FDNAMES=", joined); |
8dd4c05b LP |
1619 | if (!x) |
1620 | return -ENOMEM; | |
1621 | our_env[n_env++] = x; | |
7cae38c4 LP |
1622 | } |
1623 | ||
b08af3b1 | 1624 | if ((p->flags & EXEC_SET_WATCHDOG) && p->watchdog_usec > 0) { |
ccd06097 | 1625 | if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid()) < 0) |
09812eb7 LP |
1626 | return -ENOMEM; |
1627 | our_env[n_env++] = x; | |
1628 | ||
1e22b5cd | 1629 | if (asprintf(&x, "WATCHDOG_USEC="USEC_FMT, p->watchdog_usec) < 0) |
09812eb7 LP |
1630 | return -ENOMEM; |
1631 | our_env[n_env++] = x; | |
1632 | } | |
1633 | ||
fd63e712 LP |
1634 | /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic |
1635 | * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but | |
1636 | * check the database directly. */ | |
1637 | if (unit_has_name(u, SPECIAL_DBUS_SERVICE)) { | |
1638 | x = strdup("SYSTEMD_NSS_BYPASS_BUS=1"); | |
1639 | if (!x) | |
1640 | return -ENOMEM; | |
1641 | our_env[n_env++] = x; | |
1642 | } | |
1643 | ||
7cae38c4 LP |
1644 | if (home) { |
1645 | x = strappend("HOME=", home); | |
1646 | if (!x) | |
1647 | return -ENOMEM; | |
1648 | our_env[n_env++] = x; | |
1649 | } | |
1650 | ||
1651 | if (username) { | |
1652 | x = strappend("LOGNAME=", username); | |
1653 | if (!x) | |
1654 | return -ENOMEM; | |
1655 | our_env[n_env++] = x; | |
1656 | ||
1657 | x = strappend("USER=", username); | |
1658 | if (!x) | |
1659 | return -ENOMEM; | |
1660 | our_env[n_env++] = x; | |
1661 | } | |
1662 | ||
1663 | if (shell) { | |
1664 | x = strappend("SHELL=", shell); | |
1665 | if (!x) | |
1666 | return -ENOMEM; | |
1667 | our_env[n_env++] = x; | |
1668 | } | |
1669 | ||
4b58153d LP |
1670 | if (!sd_id128_is_null(u->invocation_id)) { |
1671 | if (asprintf(&x, "INVOCATION_ID=" SD_ID128_FORMAT_STR, SD_ID128_FORMAT_VAL(u->invocation_id)) < 0) | |
1672 | return -ENOMEM; | |
1673 | ||
1674 | our_env[n_env++] = x; | |
1675 | } | |
1676 | ||
6af760f3 LP |
1677 | if (exec_context_needs_term(c)) { |
1678 | const char *tty_path, *term = NULL; | |
1679 | ||
1680 | tty_path = exec_context_tty_path(c); | |
1681 | ||
1682 | /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit | |
1683 | * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager | |
1684 | * passes to PID 1 ends up all the way in the console login shown. */ | |
1685 | ||
1686 | if (path_equal(tty_path, "/dev/console") && getppid() == 1) | |
1687 | term = getenv("TERM"); | |
1688 | if (!term) | |
1689 | term = default_term_for_tty(tty_path); | |
7cae38c4 | 1690 | |
6af760f3 | 1691 | x = strappend("TERM=", term); |
7cae38c4 LP |
1692 | if (!x) |
1693 | return -ENOMEM; | |
1694 | our_env[n_env++] = x; | |
1695 | } | |
1696 | ||
7bce046b LP |
1697 | if (journal_stream_dev != 0 && journal_stream_ino != 0) { |
1698 | if (asprintf(&x, "JOURNAL_STREAM=" DEV_FMT ":" INO_FMT, journal_stream_dev, journal_stream_ino) < 0) | |
1699 | return -ENOMEM; | |
1700 | ||
1701 | our_env[n_env++] = x; | |
1702 | } | |
1703 | ||
7cae38c4 | 1704 | our_env[n_env++] = NULL; |
7bce046b | 1705 | assert(n_env <= 12); |
7cae38c4 LP |
1706 | |
1707 | *ret = our_env; | |
1708 | our_env = NULL; | |
1709 | ||
1710 | return 0; | |
1711 | } | |
1712 | ||
b4c14404 FB |
1713 | static int build_pass_environment(const ExecContext *c, char ***ret) { |
1714 | _cleanup_strv_free_ char **pass_env = NULL; | |
1715 | size_t n_env = 0, n_bufsize = 0; | |
1716 | char **i; | |
1717 | ||
1718 | STRV_FOREACH(i, c->pass_environment) { | |
1719 | _cleanup_free_ char *x = NULL; | |
1720 | char *v; | |
1721 | ||
1722 | v = getenv(*i); | |
1723 | if (!v) | |
1724 | continue; | |
605405c6 | 1725 | x = strjoin(*i, "=", v); |
b4c14404 FB |
1726 | if (!x) |
1727 | return -ENOMEM; | |
1728 | if (!GREEDY_REALLOC(pass_env, n_bufsize, n_env + 2)) | |
1729 | return -ENOMEM; | |
1730 | pass_env[n_env++] = x; | |
1731 | pass_env[n_env] = NULL; | |
1732 | x = NULL; | |
1733 | } | |
1734 | ||
1735 | *ret = pass_env; | |
1736 | pass_env = NULL; | |
1737 | ||
1738 | return 0; | |
1739 | } | |
1740 | ||
8b44a3d2 LP |
1741 | static bool exec_needs_mount_namespace( |
1742 | const ExecContext *context, | |
1743 | const ExecParameters *params, | |
1744 | ExecRuntime *runtime) { | |
1745 | ||
1746 | assert(context); | |
1747 | assert(params); | |
1748 | ||
2a624c36 AP |
1749 | if (!strv_isempty(context->read_write_paths) || |
1750 | !strv_isempty(context->read_only_paths) || | |
1751 | !strv_isempty(context->inaccessible_paths)) | |
8b44a3d2 LP |
1752 | return true; |
1753 | ||
1754 | if (context->mount_flags != 0) | |
1755 | return true; | |
1756 | ||
1757 | if (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) | |
1758 | return true; | |
1759 | ||
8b44a3d2 LP |
1760 | if (context->private_devices || |
1761 | context->protect_system != PROTECT_SYSTEM_NO || | |
59eeb84b LP |
1762 | context->protect_home != PROTECT_HOME_NO || |
1763 | context->protect_kernel_tunables || | |
c575770b | 1764 | context->protect_kernel_modules || |
59eeb84b | 1765 | context->protect_control_groups) |
8b44a3d2 LP |
1766 | return true; |
1767 | ||
1768 | return false; | |
1769 | } | |
1770 | ||
d251207d LP |
1771 | static int setup_private_users(uid_t uid, gid_t gid) { |
1772 | _cleanup_free_ char *uid_map = NULL, *gid_map = NULL; | |
1773 | _cleanup_close_pair_ int errno_pipe[2] = { -1, -1 }; | |
1774 | _cleanup_close_ int unshare_ready_fd = -1; | |
1775 | _cleanup_(sigkill_waitp) pid_t pid = 0; | |
1776 | uint64_t c = 1; | |
1777 | siginfo_t si; | |
1778 | ssize_t n; | |
1779 | int r; | |
1780 | ||
1781 | /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to | |
1782 | * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which | |
1783 | * we however lack after opening the user namespace. To work around this we fork() a temporary child process, | |
1784 | * which waits for the parent to create the new user namespace while staying in the original namespace. The | |
1785 | * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and | |
1786 | * continues execution normally. */ | |
1787 | ||
1788 | if (uid != 0 && uid_is_valid(uid)) | |
1789 | asprintf(&uid_map, | |
1790 | "0 0 1\n" /* Map root → root */ | |
1791 | UID_FMT " " UID_FMT " 1\n", /* Map $UID → $UID */ | |
e0f3720e | 1792 | uid, uid); |
d251207d | 1793 | else |
e0f3720e | 1794 | uid_map = strdup("0 0 1\n"); /* The case where the above is the same */ |
d251207d LP |
1795 | if (!uid_map) |
1796 | return -ENOMEM; | |
1797 | ||
1798 | if (gid != 0 && gid_is_valid(gid)) | |
1799 | asprintf(&gid_map, | |
1800 | "0 0 1\n" /* Map root → root */ | |
1801 | GID_FMT " " GID_FMT " 1\n", /* Map $GID → $GID */ | |
1802 | gid, gid); | |
1803 | else | |
1804 | gid_map = strdup("0 0 1\n"); /* The case where the above is the same */ | |
1805 | if (!gid_map) | |
1806 | return -ENOMEM; | |
1807 | ||
1808 | /* Create a communication channel so that the parent can tell the child when it finished creating the user | |
1809 | * namespace. */ | |
1810 | unshare_ready_fd = eventfd(0, EFD_CLOEXEC); | |
1811 | if (unshare_ready_fd < 0) | |
1812 | return -errno; | |
1813 | ||
1814 | /* Create a communication channel so that the child can tell the parent a proper error code in case it | |
1815 | * failed. */ | |
1816 | if (pipe2(errno_pipe, O_CLOEXEC) < 0) | |
1817 | return -errno; | |
1818 | ||
1819 | pid = fork(); | |
1820 | if (pid < 0) | |
1821 | return -errno; | |
1822 | ||
1823 | if (pid == 0) { | |
1824 | _cleanup_close_ int fd = -1; | |
1825 | const char *a; | |
1826 | pid_t ppid; | |
1827 | ||
1828 | /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from | |
1829 | * here, after the parent opened its own user namespace. */ | |
1830 | ||
1831 | ppid = getppid(); | |
1832 | errno_pipe[0] = safe_close(errno_pipe[0]); | |
1833 | ||
1834 | /* Wait until the parent unshared the user namespace */ | |
1835 | if (read(unshare_ready_fd, &c, sizeof(c)) < 0) { | |
1836 | r = -errno; | |
1837 | goto child_fail; | |
1838 | } | |
1839 | ||
1840 | /* Disable the setgroups() system call in the child user namespace, for good. */ | |
1841 | a = procfs_file_alloca(ppid, "setgroups"); | |
1842 | fd = open(a, O_WRONLY|O_CLOEXEC); | |
1843 | if (fd < 0) { | |
1844 | if (errno != ENOENT) { | |
1845 | r = -errno; | |
1846 | goto child_fail; | |
1847 | } | |
1848 | ||
1849 | /* If the file is missing the kernel is too old, let's continue anyway. */ | |
1850 | } else { | |
1851 | if (write(fd, "deny\n", 5) < 0) { | |
1852 | r = -errno; | |
1853 | goto child_fail; | |
1854 | } | |
1855 | ||
1856 | fd = safe_close(fd); | |
1857 | } | |
1858 | ||
1859 | /* First write the GID map */ | |
1860 | a = procfs_file_alloca(ppid, "gid_map"); | |
1861 | fd = open(a, O_WRONLY|O_CLOEXEC); | |
1862 | if (fd < 0) { | |
1863 | r = -errno; | |
1864 | goto child_fail; | |
1865 | } | |
1866 | if (write(fd, gid_map, strlen(gid_map)) < 0) { | |
1867 | r = -errno; | |
1868 | goto child_fail; | |
1869 | } | |
1870 | fd = safe_close(fd); | |
1871 | ||
1872 | /* The write the UID map */ | |
1873 | a = procfs_file_alloca(ppid, "uid_map"); | |
1874 | fd = open(a, O_WRONLY|O_CLOEXEC); | |
1875 | if (fd < 0) { | |
1876 | r = -errno; | |
1877 | goto child_fail; | |
1878 | } | |
1879 | if (write(fd, uid_map, strlen(uid_map)) < 0) { | |
1880 | r = -errno; | |
1881 | goto child_fail; | |
1882 | } | |
1883 | ||
1884 | _exit(EXIT_SUCCESS); | |
1885 | ||
1886 | child_fail: | |
1887 | (void) write(errno_pipe[1], &r, sizeof(r)); | |
1888 | _exit(EXIT_FAILURE); | |
1889 | } | |
1890 | ||
1891 | errno_pipe[1] = safe_close(errno_pipe[1]); | |
1892 | ||
1893 | if (unshare(CLONE_NEWUSER) < 0) | |
1894 | return -errno; | |
1895 | ||
1896 | /* Let the child know that the namespace is ready now */ | |
1897 | if (write(unshare_ready_fd, &c, sizeof(c)) < 0) | |
1898 | return -errno; | |
1899 | ||
1900 | /* Try to read an error code from the child */ | |
1901 | n = read(errno_pipe[0], &r, sizeof(r)); | |
1902 | if (n < 0) | |
1903 | return -errno; | |
1904 | if (n == sizeof(r)) { /* an error code was sent to us */ | |
1905 | if (r < 0) | |
1906 | return r; | |
1907 | return -EIO; | |
1908 | } | |
1909 | if (n != 0) /* on success we should have read 0 bytes */ | |
1910 | return -EIO; | |
1911 | ||
1912 | r = wait_for_terminate(pid, &si); | |
1913 | if (r < 0) | |
1914 | return r; | |
1915 | pid = 0; | |
1916 | ||
1917 | /* If something strange happened with the child, let's consider this fatal, too */ | |
1918 | if (si.si_code != CLD_EXITED || si.si_status != 0) | |
1919 | return -EIO; | |
1920 | ||
1921 | return 0; | |
1922 | } | |
1923 | ||
07689d5d LP |
1924 | static int setup_runtime_directory( |
1925 | const ExecContext *context, | |
1926 | const ExecParameters *params, | |
1927 | uid_t uid, | |
1928 | gid_t gid) { | |
1929 | ||
1930 | char **rt; | |
1931 | int r; | |
1932 | ||
1933 | assert(context); | |
1934 | assert(params); | |
1935 | ||
1936 | STRV_FOREACH(rt, context->runtime_directory) { | |
1937 | _cleanup_free_ char *p; | |
1938 | ||
605405c6 | 1939 | p = strjoin(params->runtime_prefix, "/", *rt); |
07689d5d LP |
1940 | if (!p) |
1941 | return -ENOMEM; | |
1942 | ||
1943 | r = mkdir_p_label(p, context->runtime_directory_mode); | |
1944 | if (r < 0) | |
1945 | return r; | |
1946 | ||
1947 | r = chmod_and_chown(p, context->runtime_directory_mode, uid, gid); | |
1948 | if (r < 0) | |
1949 | return r; | |
1950 | } | |
1951 | ||
1952 | return 0; | |
1953 | } | |
1954 | ||
cefc33ae LP |
1955 | static int setup_smack( |
1956 | const ExecContext *context, | |
1957 | const ExecCommand *command) { | |
1958 | ||
1959 | #ifdef HAVE_SMACK | |
1960 | int r; | |
1961 | ||
1962 | assert(context); | |
1963 | assert(command); | |
1964 | ||
1965 | if (!mac_smack_use()) | |
1966 | return 0; | |
1967 | ||
1968 | if (context->smack_process_label) { | |
1969 | r = mac_smack_apply_pid(0, context->smack_process_label); | |
1970 | if (r < 0) | |
1971 | return r; | |
1972 | } | |
1973 | #ifdef SMACK_DEFAULT_PROCESS_LABEL | |
1974 | else { | |
1975 | _cleanup_free_ char *exec_label = NULL; | |
1976 | ||
1977 | r = mac_smack_read(command->path, SMACK_ATTR_EXEC, &exec_label); | |
1978 | if (r < 0 && r != -ENODATA && r != -EOPNOTSUPP) | |
1979 | return r; | |
1980 | ||
1981 | r = mac_smack_apply_pid(0, exec_label ? : SMACK_DEFAULT_PROCESS_LABEL); | |
1982 | if (r < 0) | |
1983 | return r; | |
1984 | } | |
1985 | #endif | |
1986 | #endif | |
1987 | ||
1988 | return 0; | |
1989 | } | |
1990 | ||
3fbe8dbe LP |
1991 | static int compile_read_write_paths( |
1992 | const ExecContext *context, | |
1993 | const ExecParameters *params, | |
1994 | char ***ret) { | |
1995 | ||
1996 | _cleanup_strv_free_ char **l = NULL; | |
1997 | char **rt; | |
1998 | ||
1999 | /* Compile the list of writable paths. This is the combination of the explicitly configured paths, plus all | |
2000 | * runtime directories. */ | |
2001 | ||
2002 | if (strv_isempty(context->read_write_paths) && | |
2003 | strv_isempty(context->runtime_directory)) { | |
2004 | *ret = NULL; /* NOP if neither is set */ | |
2005 | return 0; | |
2006 | } | |
2007 | ||
2008 | l = strv_copy(context->read_write_paths); | |
2009 | if (!l) | |
2010 | return -ENOMEM; | |
2011 | ||
2012 | STRV_FOREACH(rt, context->runtime_directory) { | |
2013 | char *s; | |
2014 | ||
605405c6 | 2015 | s = strjoin(params->runtime_prefix, "/", *rt); |
3fbe8dbe LP |
2016 | if (!s) |
2017 | return -ENOMEM; | |
2018 | ||
2019 | if (strv_consume(&l, s) < 0) | |
2020 | return -ENOMEM; | |
2021 | } | |
2022 | ||
2023 | *ret = l; | |
2024 | l = NULL; | |
2025 | ||
2026 | return 0; | |
2027 | } | |
2028 | ||
93c6bb51 DH |
2029 | static int apply_mount_namespace(Unit *u, const ExecContext *context, |
2030 | const ExecParameters *params, | |
2031 | ExecRuntime *runtime) { | |
2032 | int r; | |
2033 | _cleanup_free_ char **rw = NULL; | |
2034 | char *tmp = NULL, *var = NULL; | |
2035 | const char *root_dir = NULL; | |
2036 | NameSpaceInfo ns_info = { | |
af964954 | 2037 | .ignore_protect_paths = false, |
93c6bb51 DH |
2038 | .private_dev = context->private_devices, |
2039 | .protect_control_groups = context->protect_control_groups, | |
2040 | .protect_kernel_tunables = context->protect_kernel_tunables, | |
2041 | .protect_kernel_modules = context->protect_kernel_modules, | |
2042 | }; | |
2043 | ||
2b3c1b9e DH |
2044 | assert(context); |
2045 | ||
93c6bb51 DH |
2046 | /* The runtime struct only contains the parent of the private /tmp, |
2047 | * which is non-accessible to world users. Inside of it there's a /tmp | |
2048 | * that is sticky, and that's the one we want to use here. */ | |
2049 | ||
2050 | if (context->private_tmp && runtime) { | |
2051 | if (runtime->tmp_dir) | |
2052 | tmp = strjoina(runtime->tmp_dir, "/tmp"); | |
2053 | if (runtime->var_tmp_dir) | |
2054 | var = strjoina(runtime->var_tmp_dir, "/tmp"); | |
2055 | } | |
2056 | ||
2057 | r = compile_read_write_paths(context, params, &rw); | |
2058 | if (r < 0) | |
2059 | return r; | |
2060 | ||
2061 | if (params->flags & EXEC_APPLY_CHROOT) | |
2062 | root_dir = context->root_directory; | |
2063 | ||
af964954 DH |
2064 | /* |
2065 | * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed | |
2066 | * sandbox info, otherwise enforce it, don't ignore protected paths and | |
2067 | * fail if we are enable to apply the sandbox inside the mount namespace. | |
2068 | */ | |
2069 | if (!context->dynamic_user && root_dir) | |
2070 | ns_info.ignore_protect_paths = true; | |
2071 | ||
93c6bb51 DH |
2072 | r = setup_namespace(root_dir, &ns_info, rw, |
2073 | context->read_only_paths, | |
2074 | context->inaccessible_paths, | |
2075 | tmp, | |
2076 | var, | |
2077 | context->protect_home, | |
2078 | context->protect_system, | |
2079 | context->mount_flags); | |
2080 | ||
2081 | /* If we couldn't set up the namespace this is probably due to a | |
2082 | * missing capability. In this case, silently proceeed. */ | |
2083 | if (IN_SET(r, -EPERM, -EACCES)) { | |
2084 | log_open(); | |
2085 | log_unit_debug_errno(u, r, "Failed to set up namespace, assuming containerized execution, ignoring: %m"); | |
2086 | log_close(); | |
2087 | r = 0; | |
2088 | } | |
2089 | ||
2090 | return r; | |
2091 | } | |
2092 | ||
e7f1e7c6 DH |
2093 | static int apply_working_directory(const ExecContext *context, |
2094 | const ExecParameters *params, | |
2b3c1b9e | 2095 | const char *home, |
e7f1e7c6 | 2096 | const bool needs_mount_ns) { |
2b3c1b9e DH |
2097 | const char *d; |
2098 | const char *wd; | |
2099 | ||
2100 | assert(context); | |
2101 | ||
2102 | if (context->working_directory_home) | |
2103 | wd = home; | |
2104 | else if (context->working_directory) | |
2105 | wd = context->working_directory; | |
2106 | else | |
2107 | wd = "/"; | |
e7f1e7c6 DH |
2108 | |
2109 | if (params->flags & EXEC_APPLY_CHROOT) { | |
2110 | if (!needs_mount_ns && context->root_directory) | |
2111 | if (chroot(context->root_directory) < 0) | |
2112 | return -errno; | |
2113 | ||
2b3c1b9e DH |
2114 | d = wd; |
2115 | } else | |
2116 | d = strjoina(strempty(context->root_directory), "/", strempty(wd)); | |
e7f1e7c6 | 2117 | |
2b3c1b9e DH |
2118 | if (chdir(d) < 0 && !context->working_directory_missing_ok) |
2119 | return -errno; | |
e7f1e7c6 DH |
2120 | |
2121 | return 0; | |
2122 | } | |
2123 | ||
29206d46 LP |
2124 | static void append_socket_pair(int *array, unsigned *n, int pair[2]) { |
2125 | assert(array); | |
2126 | assert(n); | |
2127 | ||
2128 | if (!pair) | |
2129 | return; | |
2130 | ||
2131 | if (pair[0] >= 0) | |
2132 | array[(*n)++] = pair[0]; | |
2133 | if (pair[1] >= 0) | |
2134 | array[(*n)++] = pair[1]; | |
2135 | } | |
2136 | ||
a34ceba6 LP |
2137 | static int close_remaining_fds( |
2138 | const ExecParameters *params, | |
2139 | ExecRuntime *runtime, | |
29206d46 | 2140 | DynamicCreds *dcreds, |
00d9ef85 | 2141 | int user_lookup_fd, |
a34ceba6 LP |
2142 | int socket_fd, |
2143 | int *fds, unsigned n_fds) { | |
2144 | ||
2145 | unsigned n_dont_close = 0; | |
00d9ef85 | 2146 | int dont_close[n_fds + 12]; |
a34ceba6 LP |
2147 | |
2148 | assert(params); | |
2149 | ||
2150 | if (params->stdin_fd >= 0) | |
2151 | dont_close[n_dont_close++] = params->stdin_fd; | |
2152 | if (params->stdout_fd >= 0) | |
2153 | dont_close[n_dont_close++] = params->stdout_fd; | |
2154 | if (params->stderr_fd >= 0) | |
2155 | dont_close[n_dont_close++] = params->stderr_fd; | |
2156 | ||
2157 | if (socket_fd >= 0) | |
2158 | dont_close[n_dont_close++] = socket_fd; | |
2159 | if (n_fds > 0) { | |
2160 | memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds); | |
2161 | n_dont_close += n_fds; | |
2162 | } | |
2163 | ||
29206d46 LP |
2164 | if (runtime) |
2165 | append_socket_pair(dont_close, &n_dont_close, runtime->netns_storage_socket); | |
2166 | ||
2167 | if (dcreds) { | |
2168 | if (dcreds->user) | |
2169 | append_socket_pair(dont_close, &n_dont_close, dcreds->user->storage_socket); | |
2170 | if (dcreds->group) | |
2171 | append_socket_pair(dont_close, &n_dont_close, dcreds->group->storage_socket); | |
a34ceba6 LP |
2172 | } |
2173 | ||
00d9ef85 LP |
2174 | if (user_lookup_fd >= 0) |
2175 | dont_close[n_dont_close++] = user_lookup_fd; | |
2176 | ||
a34ceba6 LP |
2177 | return close_all_fds(dont_close, n_dont_close); |
2178 | } | |
2179 | ||
59eeb84b LP |
2180 | static bool context_has_address_families(const ExecContext *c) { |
2181 | assert(c); | |
2182 | ||
2183 | return c->address_families_whitelist || | |
2184 | !set_isempty(c->address_families); | |
2185 | } | |
2186 | ||
2187 | static bool context_has_syscall_filters(const ExecContext *c) { | |
2188 | assert(c); | |
2189 | ||
2190 | return c->syscall_whitelist || | |
2191 | !set_isempty(c->syscall_filter) || | |
2192 | !set_isempty(c->syscall_archs); | |
2193 | } | |
2194 | ||
2195 | static bool context_has_no_new_privileges(const ExecContext *c) { | |
2196 | assert(c); | |
2197 | ||
2198 | if (c->no_new_privileges) | |
2199 | return true; | |
2200 | ||
2201 | if (have_effective_cap(CAP_SYS_ADMIN)) /* if we are privileged, we don't need NNP */ | |
2202 | return false; | |
2203 | ||
2204 | return context_has_address_families(c) || /* we need NNP if we have any form of seccomp and are unprivileged */ | |
2205 | c->memory_deny_write_execute || | |
2206 | c->restrict_realtime || | |
add00535 | 2207 | exec_context_restrict_namespaces_set(c) || |
59eeb84b | 2208 | c->protect_kernel_tunables || |
4084e8fc DH |
2209 | c->protect_kernel_modules || |
2210 | c->private_devices || | |
59eeb84b LP |
2211 | context_has_syscall_filters(c); |
2212 | } | |
2213 | ||
00d9ef85 LP |
2214 | static int send_user_lookup( |
2215 | Unit *unit, | |
2216 | int user_lookup_fd, | |
2217 | uid_t uid, | |
2218 | gid_t gid) { | |
2219 | ||
2220 | assert(unit); | |
2221 | ||
2222 | /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID | |
2223 | * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was | |
2224 | * specified. */ | |
2225 | ||
2226 | if (user_lookup_fd < 0) | |
2227 | return 0; | |
2228 | ||
2229 | if (!uid_is_valid(uid) && !gid_is_valid(gid)) | |
2230 | return 0; | |
2231 | ||
2232 | if (writev(user_lookup_fd, | |
2233 | (struct iovec[]) { | |
2234 | { .iov_base = &uid, .iov_len = sizeof(uid) }, | |
2235 | { .iov_base = &gid, .iov_len = sizeof(gid) }, | |
2236 | { .iov_base = unit->id, .iov_len = strlen(unit->id) }}, 3) < 0) | |
2237 | return -errno; | |
2238 | ||
2239 | return 0; | |
2240 | } | |
2241 | ||
ff0af2a1 | 2242 | static int exec_child( |
f2341e0a | 2243 | Unit *unit, |
ff0af2a1 LP |
2244 | ExecCommand *command, |
2245 | const ExecContext *context, | |
2246 | const ExecParameters *params, | |
2247 | ExecRuntime *runtime, | |
29206d46 | 2248 | DynamicCreds *dcreds, |
ff0af2a1 LP |
2249 | char **argv, |
2250 | int socket_fd, | |
52c239d7 | 2251 | int named_iofds[3], |
ff0af2a1 LP |
2252 | int *fds, unsigned n_fds, |
2253 | char **files_env, | |
00d9ef85 | 2254 | int user_lookup_fd, |
ff0af2a1 | 2255 | int *exit_status) { |
d35fbf6b | 2256 | |
2065ca69 | 2257 | _cleanup_strv_free_ char **our_env = NULL, **pass_env = NULL, **accum_env = NULL, **final_argv = NULL; |
9008e1ac | 2258 | _cleanup_free_ char *mac_selinux_context_net = NULL; |
4d885bd3 DH |
2259 | _cleanup_free_ gid_t *supplementary_gids = NULL; |
2260 | const char *username = NULL, *groupname = NULL; | |
2b3c1b9e | 2261 | const char *home = NULL, *shell = NULL; |
7bce046b LP |
2262 | dev_t journal_stream_dev = 0; |
2263 | ino_t journal_stream_ino = 0; | |
2264 | bool needs_mount_namespace; | |
fed1e721 LP |
2265 | uid_t uid = UID_INVALID; |
2266 | gid_t gid = GID_INVALID; | |
4d885bd3 | 2267 | int i, r, ngids = 0; |
034c6ed7 | 2268 | |
f2341e0a | 2269 | assert(unit); |
5cb5a6ff LP |
2270 | assert(command); |
2271 | assert(context); | |
d35fbf6b | 2272 | assert(params); |
ff0af2a1 | 2273 | assert(exit_status); |
d35fbf6b DM |
2274 | |
2275 | rename_process_from_path(command->path); | |
2276 | ||
2277 | /* We reset exactly these signals, since they are the | |
2278 | * only ones we set to SIG_IGN in the main daemon. All | |
2279 | * others we leave untouched because we set them to | |
2280 | * SIG_DFL or a valid handler initially, both of which | |
2281 | * will be demoted to SIG_DFL. */ | |
ce30c8dc LP |
2282 | (void) default_signals(SIGNALS_CRASH_HANDLER, |
2283 | SIGNALS_IGNORE, -1); | |
d35fbf6b DM |
2284 | |
2285 | if (context->ignore_sigpipe) | |
ce30c8dc | 2286 | (void) ignore_signals(SIGPIPE, -1); |
d35fbf6b | 2287 | |
ff0af2a1 LP |
2288 | r = reset_signal_mask(); |
2289 | if (r < 0) { | |
2290 | *exit_status = EXIT_SIGNAL_MASK; | |
2291 | return r; | |
d35fbf6b | 2292 | } |
034c6ed7 | 2293 | |
d35fbf6b DM |
2294 | if (params->idle_pipe) |
2295 | do_idle_pipe_dance(params->idle_pipe); | |
4f2d528d | 2296 | |
d35fbf6b DM |
2297 | /* Close sockets very early to make sure we don't |
2298 | * block init reexecution because it cannot bind its | |
2299 | * sockets */ | |
ff0af2a1 | 2300 | |
d35fbf6b | 2301 | log_forget_fds(); |
4f2d528d | 2302 | |
00d9ef85 | 2303 | r = close_remaining_fds(params, runtime, dcreds, user_lookup_fd, socket_fd, fds, n_fds); |
ff0af2a1 LP |
2304 | if (r < 0) { |
2305 | *exit_status = EXIT_FDS; | |
2306 | return r; | |
8c7be95e LP |
2307 | } |
2308 | ||
d35fbf6b DM |
2309 | if (!context->same_pgrp) |
2310 | if (setsid() < 0) { | |
ff0af2a1 | 2311 | *exit_status = EXIT_SETSID; |
d35fbf6b DM |
2312 | return -errno; |
2313 | } | |
9e2f7c11 | 2314 | |
1e22b5cd | 2315 | exec_context_tty_reset(context, params); |
d35fbf6b | 2316 | |
c39f1ce2 | 2317 | if (params->flags & EXEC_CONFIRM_SPAWN) { |
d35fbf6b DM |
2318 | char response; |
2319 | ||
ff0af2a1 LP |
2320 | r = ask_for_confirmation(&response, argv); |
2321 | if (r == -ETIMEDOUT) | |
d35fbf6b | 2322 | write_confirm_message("Confirmation question timed out, assuming positive response.\n"); |
ff0af2a1 LP |
2323 | else if (r < 0) |
2324 | write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-r)); | |
d35fbf6b DM |
2325 | else if (response == 's') { |
2326 | write_confirm_message("Skipping execution.\n"); | |
ff0af2a1 | 2327 | *exit_status = EXIT_CONFIRM; |
d35fbf6b DM |
2328 | return -ECANCELED; |
2329 | } else if (response == 'n') { | |
2330 | write_confirm_message("Failing execution.\n"); | |
ff0af2a1 | 2331 | *exit_status = 0; |
d35fbf6b DM |
2332 | return 0; |
2333 | } | |
2334 | } | |
1a63a750 | 2335 | |
29206d46 LP |
2336 | if (context->dynamic_user && dcreds) { |
2337 | ||
409093fe LP |
2338 | /* Make sure we bypass our own NSS module for any NSS checks */ |
2339 | if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) { | |
2340 | *exit_status = EXIT_USER; | |
2341 | return -errno; | |
2342 | } | |
2343 | ||
29206d46 | 2344 | r = dynamic_creds_realize(dcreds, &uid, &gid); |
ff0af2a1 LP |
2345 | if (r < 0) { |
2346 | *exit_status = EXIT_USER; | |
2347 | return r; | |
524daa8c | 2348 | } |
524daa8c | 2349 | |
92b25bca | 2350 | if (!uid_is_valid(uid) || !gid_is_valid(gid)) { |
29206d46 LP |
2351 | *exit_status = EXIT_USER; |
2352 | return -ESRCH; | |
2353 | } | |
5bc7452b | 2354 | |
29206d46 LP |
2355 | if (dcreds->user) |
2356 | username = dcreds->user->name; | |
2357 | ||
2358 | } else { | |
4d885bd3 DH |
2359 | r = get_fixed_user(context, &username, &uid, &gid, &home, &shell); |
2360 | if (r < 0) { | |
2361 | *exit_status = EXIT_USER; | |
2362 | return r; | |
5bc7452b | 2363 | } |
5bc7452b | 2364 | |
4d885bd3 DH |
2365 | r = get_fixed_group(context, &groupname, &gid); |
2366 | if (r < 0) { | |
2367 | *exit_status = EXIT_GROUP; | |
2368 | return r; | |
2369 | } | |
cdc5d5c5 | 2370 | } |
29206d46 | 2371 | |
cdc5d5c5 DH |
2372 | /* Initialize user supplementary groups and get SupplementaryGroups= ones */ |
2373 | r = get_supplementary_groups(context, username, groupname, gid, | |
2374 | &supplementary_gids, &ngids); | |
2375 | if (r < 0) { | |
2376 | *exit_status = EXIT_GROUP; | |
2377 | return r; | |
29206d46 | 2378 | } |
5bc7452b | 2379 | |
00d9ef85 LP |
2380 | r = send_user_lookup(unit, user_lookup_fd, uid, gid); |
2381 | if (r < 0) { | |
2382 | *exit_status = EXIT_USER; | |
2383 | return r; | |
2384 | } | |
2385 | ||
2386 | user_lookup_fd = safe_close(user_lookup_fd); | |
2387 | ||
d35fbf6b DM |
2388 | /* If a socket is connected to STDIN/STDOUT/STDERR, we |
2389 | * must sure to drop O_NONBLOCK */ | |
2390 | if (socket_fd >= 0) | |
a34ceba6 | 2391 | (void) fd_nonblock(socket_fd, false); |
acbb0225 | 2392 | |
52c239d7 | 2393 | r = setup_input(context, params, socket_fd, named_iofds); |
ff0af2a1 LP |
2394 | if (r < 0) { |
2395 | *exit_status = EXIT_STDIN; | |
2396 | return r; | |
d35fbf6b | 2397 | } |
034c6ed7 | 2398 | |
52c239d7 | 2399 | r = setup_output(unit, context, params, STDOUT_FILENO, socket_fd, named_iofds, basename(command->path), uid, gid, &journal_stream_dev, &journal_stream_ino); |
ff0af2a1 LP |
2400 | if (r < 0) { |
2401 | *exit_status = EXIT_STDOUT; | |
2402 | return r; | |
d35fbf6b DM |
2403 | } |
2404 | ||
52c239d7 | 2405 | r = setup_output(unit, context, params, STDERR_FILENO, socket_fd, named_iofds, basename(command->path), uid, gid, &journal_stream_dev, &journal_stream_ino); |
ff0af2a1 LP |
2406 | if (r < 0) { |
2407 | *exit_status = EXIT_STDERR; | |
2408 | return r; | |
d35fbf6b DM |
2409 | } |
2410 | ||
2411 | if (params->cgroup_path) { | |
ff0af2a1 LP |
2412 | r = cg_attach_everywhere(params->cgroup_supported, params->cgroup_path, 0, NULL, NULL); |
2413 | if (r < 0) { | |
2414 | *exit_status = EXIT_CGROUP; | |
2415 | return r; | |
309bff19 | 2416 | } |
d35fbf6b | 2417 | } |
309bff19 | 2418 | |
d35fbf6b | 2419 | if (context->oom_score_adjust_set) { |
d5243d62 | 2420 | char t[DECIMAL_STR_MAX(context->oom_score_adjust)]; |
f2b68789 | 2421 | |
d5243d62 LP |
2422 | /* When we can't make this change due to EPERM, then |
2423 | * let's silently skip over it. User namespaces | |
2424 | * prohibit write access to this file, and we | |
2425 | * shouldn't trip up over that. */ | |
613b411c | 2426 | |
d5243d62 | 2427 | sprintf(t, "%i", context->oom_score_adjust); |
ad118bda | 2428 | r = write_string_file("/proc/self/oom_score_adj", t, 0); |
6cb7fa17 | 2429 | if (r == -EPERM || r == -EACCES) { |
ff0af2a1 | 2430 | log_open(); |
f2341e0a | 2431 | log_unit_debug_errno(unit, r, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m"); |
ff0af2a1 LP |
2432 | log_close(); |
2433 | } else if (r < 0) { | |
2434 | *exit_status = EXIT_OOM_ADJUST; | |
d35fbf6b | 2435 | return -errno; |
613b411c | 2436 | } |
d35fbf6b DM |
2437 | } |
2438 | ||
2439 | if (context->nice_set) | |
2440 | if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) { | |
ff0af2a1 | 2441 | *exit_status = EXIT_NICE; |
d35fbf6b | 2442 | return -errno; |
613b411c LP |
2443 | } |
2444 | ||
d35fbf6b DM |
2445 | if (context->cpu_sched_set) { |
2446 | struct sched_param param = { | |
2447 | .sched_priority = context->cpu_sched_priority, | |
2448 | }; | |
2449 | ||
ff0af2a1 LP |
2450 | r = sched_setscheduler(0, |
2451 | context->cpu_sched_policy | | |
2452 | (context->cpu_sched_reset_on_fork ? | |
2453 | SCHED_RESET_ON_FORK : 0), | |
2454 | ¶m); | |
2455 | if (r < 0) { | |
2456 | *exit_status = EXIT_SETSCHEDULER; | |
d35fbf6b | 2457 | return -errno; |
fc9b2a84 | 2458 | } |
d35fbf6b | 2459 | } |
fc9b2a84 | 2460 | |
d35fbf6b DM |
2461 | if (context->cpuset) |
2462 | if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) { | |
ff0af2a1 | 2463 | *exit_status = EXIT_CPUAFFINITY; |
d35fbf6b | 2464 | return -errno; |
034c6ed7 LP |
2465 | } |
2466 | ||
d35fbf6b DM |
2467 | if (context->ioprio_set) |
2468 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) { | |
ff0af2a1 | 2469 | *exit_status = EXIT_IOPRIO; |
d35fbf6b DM |
2470 | return -errno; |
2471 | } | |
da726a4d | 2472 | |
d35fbf6b DM |
2473 | if (context->timer_slack_nsec != NSEC_INFINITY) |
2474 | if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) { | |
ff0af2a1 | 2475 | *exit_status = EXIT_TIMERSLACK; |
d35fbf6b | 2476 | return -errno; |
4c2630eb | 2477 | } |
9eba9da4 | 2478 | |
050f7277 | 2479 | if (context->personality != PERSONALITY_INVALID) |
d35fbf6b | 2480 | if (personality(context->personality) < 0) { |
ff0af2a1 | 2481 | *exit_status = EXIT_PERSONALITY; |
d35fbf6b | 2482 | return -errno; |
4c2630eb | 2483 | } |
94f04347 | 2484 | |
d35fbf6b | 2485 | if (context->utmp_id) |
023a4f67 LP |
2486 | utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path, |
2487 | context->utmp_mode == EXEC_UTMP_INIT ? INIT_PROCESS : | |
2488 | context->utmp_mode == EXEC_UTMP_LOGIN ? LOGIN_PROCESS : | |
2489 | USER_PROCESS, | |
2490 | username ? "root" : context->user); | |
d35fbf6b | 2491 | |
e0d2adfd | 2492 | if (context->user) { |
ff0af2a1 LP |
2493 | r = chown_terminal(STDIN_FILENO, uid); |
2494 | if (r < 0) { | |
2495 | *exit_status = EXIT_STDIN; | |
2496 | return r; | |
071830ff | 2497 | } |
d35fbf6b | 2498 | } |
8e274523 | 2499 | |
a931ad47 LP |
2500 | /* If delegation is enabled we'll pass ownership of the cgroup |
2501 | * (but only in systemd's own controller hierarchy!) to the | |
2502 | * user of the new process. */ | |
2503 | if (params->cgroup_path && context->user && params->cgroup_delegate) { | |
ff0af2a1 LP |
2504 | r = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0644, uid, gid); |
2505 | if (r < 0) { | |
2506 | *exit_status = EXIT_CGROUP; | |
2507 | return r; | |
d35fbf6b | 2508 | } |
034c6ed7 | 2509 | |
034c6ed7 | 2510 | |
ff0af2a1 LP |
2511 | r = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0755, uid, gid); |
2512 | if (r < 0) { | |
2513 | *exit_status = EXIT_CGROUP; | |
2514 | return r; | |
034c6ed7 | 2515 | } |
d35fbf6b | 2516 | } |
034c6ed7 | 2517 | |
d35fbf6b | 2518 | if (!strv_isempty(context->runtime_directory) && params->runtime_prefix) { |
07689d5d LP |
2519 | r = setup_runtime_directory(context, params, uid, gid); |
2520 | if (r < 0) { | |
2521 | *exit_status = EXIT_RUNTIME_DIRECTORY; | |
2522 | return r; | |
d35fbf6b DM |
2523 | } |
2524 | } | |
94f04347 | 2525 | |
7bce046b | 2526 | r = build_environment( |
fd63e712 | 2527 | unit, |
7bce046b LP |
2528 | context, |
2529 | params, | |
2530 | n_fds, | |
2531 | home, | |
2532 | username, | |
2533 | shell, | |
2534 | journal_stream_dev, | |
2535 | journal_stream_ino, | |
2536 | &our_env); | |
2065ca69 JW |
2537 | if (r < 0) { |
2538 | *exit_status = EXIT_MEMORY; | |
2539 | return r; | |
2540 | } | |
2541 | ||
2542 | r = build_pass_environment(context, &pass_env); | |
2543 | if (r < 0) { | |
2544 | *exit_status = EXIT_MEMORY; | |
2545 | return r; | |
2546 | } | |
2547 | ||
2548 | accum_env = strv_env_merge(5, | |
2549 | params->environment, | |
2550 | our_env, | |
2551 | pass_env, | |
2552 | context->environment, | |
2553 | files_env, | |
2554 | NULL); | |
2555 | if (!accum_env) { | |
2556 | *exit_status = EXIT_MEMORY; | |
2557 | return -ENOMEM; | |
2558 | } | |
1280503b | 2559 | accum_env = strv_env_clean(accum_env); |
2065ca69 | 2560 | |
096424d1 | 2561 | (void) umask(context->umask); |
b213e1c1 | 2562 | |
c39f1ce2 | 2563 | if ((params->flags & EXEC_APPLY_PERMISSIONS) && !command->privileged) { |
b213e1c1 | 2564 | if (context->pam_name && username) { |
2d6fce8d | 2565 | r = setup_pam(context->pam_name, username, uid, gid, context->tty_path, &accum_env, fds, n_fds); |
b213e1c1 SW |
2566 | if (r < 0) { |
2567 | *exit_status = EXIT_PAM; | |
2568 | return r; | |
2569 | } | |
d35fbf6b | 2570 | } |
b213e1c1 | 2571 | } |
ac45f971 | 2572 | |
d35fbf6b | 2573 | if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) { |
ff0af2a1 LP |
2574 | r = setup_netns(runtime->netns_storage_socket); |
2575 | if (r < 0) { | |
2576 | *exit_status = EXIT_NETWORK; | |
2577 | return r; | |
d35fbf6b DM |
2578 | } |
2579 | } | |
169c1bda | 2580 | |
ee818b89 | 2581 | needs_mount_namespace = exec_needs_mount_namespace(context, params, runtime); |
ee818b89 | 2582 | if (needs_mount_namespace) { |
93c6bb51 | 2583 | r = apply_mount_namespace(unit, context, params, runtime); |
3fbe8dbe LP |
2584 | if (r < 0) { |
2585 | *exit_status = EXIT_NAMESPACE; | |
2586 | return r; | |
2587 | } | |
d35fbf6b | 2588 | } |
81a2b7ce | 2589 | |
50b3dfb9 DH |
2590 | /* Apply just after mount namespace setup */ |
2591 | r = apply_working_directory(context, params, home, needs_mount_namespace); | |
2592 | if (r < 0) { | |
2593 | *exit_status = EXIT_CHROOT; | |
2594 | return r; | |
2595 | } | |
2596 | ||
bbeea271 | 2597 | /* Drop groups as early as possbile */ |
096424d1 | 2598 | if ((params->flags & EXEC_APPLY_PERMISSIONS) && !command->privileged) { |
4d885bd3 | 2599 | r = enforce_groups(context, gid, supplementary_gids, ngids); |
096424d1 LP |
2600 | if (r < 0) { |
2601 | *exit_status = EXIT_GROUP; | |
2602 | return r; | |
2603 | } | |
2604 | } | |
2605 | ||
9008e1ac | 2606 | #ifdef HAVE_SELINUX |
c39f1ce2 LP |
2607 | if ((params->flags & EXEC_APPLY_PERMISSIONS) && |
2608 | mac_selinux_use() && | |
2609 | params->selinux_context_net && | |
2610 | socket_fd >= 0 && | |
2611 | !command->privileged) { | |
2612 | ||
ff0af2a1 LP |
2613 | r = mac_selinux_get_child_mls_label(socket_fd, command->path, context->selinux_context, &mac_selinux_context_net); |
2614 | if (r < 0) { | |
2615 | *exit_status = EXIT_SELINUX_CONTEXT; | |
2616 | return r; | |
9008e1ac MS |
2617 | } |
2618 | } | |
2619 | #endif | |
2620 | ||
d87a2ef7 | 2621 | if ((params->flags & EXEC_APPLY_PERMISSIONS) && context->private_users) { |
d251207d LP |
2622 | r = setup_private_users(uid, gid); |
2623 | if (r < 0) { | |
2624 | *exit_status = EXIT_USER; | |
2625 | return r; | |
2626 | } | |
2627 | } | |
2628 | ||
d35fbf6b DM |
2629 | /* We repeat the fd closing here, to make sure that |
2630 | * nothing is leaked from the PAM modules. Note that | |
2631 | * we are more aggressive this time since socket_fd | |
e44da745 DM |
2632 | * and the netns fds we don't need anymore. The custom |
2633 | * endpoint fd was needed to upload the policy and can | |
2634 | * now be closed as well. */ | |
ff0af2a1 LP |
2635 | r = close_all_fds(fds, n_fds); |
2636 | if (r >= 0) | |
2637 | r = shift_fds(fds, n_fds); | |
2638 | if (r >= 0) | |
2639 | r = flags_fds(fds, n_fds, context->non_blocking); | |
2640 | if (r < 0) { | |
2641 | *exit_status = EXIT_FDS; | |
2642 | return r; | |
d35fbf6b | 2643 | } |
e66cf1a3 | 2644 | |
c39f1ce2 | 2645 | if ((params->flags & EXEC_APPLY_PERMISSIONS) && !command->privileged) { |
e66cf1a3 | 2646 | |
755d4b67 IP |
2647 | int secure_bits = context->secure_bits; |
2648 | ||
d35fbf6b | 2649 | for (i = 0; i < _RLIMIT_MAX; i++) { |
03857c43 | 2650 | |
d35fbf6b DM |
2651 | if (!context->rlimit[i]) |
2652 | continue; | |
2653 | ||
03857c43 LP |
2654 | r = setrlimit_closest(i, context->rlimit[i]); |
2655 | if (r < 0) { | |
ff0af2a1 | 2656 | *exit_status = EXIT_LIMITS; |
03857c43 | 2657 | return r; |
e66cf1a3 LP |
2658 | } |
2659 | } | |
2660 | ||
f4170c67 LP |
2661 | /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */ |
2662 | if (context->restrict_realtime && !context->rlimit[RLIMIT_RTPRIO]) { | |
2663 | if (setrlimit(RLIMIT_RTPRIO, &RLIMIT_MAKE_CONST(0)) < 0) { | |
2664 | *exit_status = EXIT_LIMITS; | |
2665 | return -errno; | |
2666 | } | |
2667 | } | |
2668 | ||
a103496c IP |
2669 | if (!cap_test_all(context->capability_bounding_set)) { |
2670 | r = capability_bounding_set_drop(context->capability_bounding_set, false); | |
ff0af2a1 LP |
2671 | if (r < 0) { |
2672 | *exit_status = EXIT_CAPABILITIES; | |
2673 | return r; | |
3b8bddde | 2674 | } |
4c2630eb | 2675 | } |
3b8bddde | 2676 | |
755d4b67 IP |
2677 | /* This is done before enforce_user, but ambient set |
2678 | * does not survive over setresuid() if keep_caps is not set. */ | |
2679 | if (context->capability_ambient_set != 0) { | |
2680 | r = capability_ambient_set_apply(context->capability_ambient_set, true); | |
2681 | if (r < 0) { | |
2682 | *exit_status = EXIT_CAPABILITIES; | |
2683 | return r; | |
2684 | } | |
755d4b67 IP |
2685 | } |
2686 | ||
d35fbf6b | 2687 | if (context->user) { |
ff0af2a1 LP |
2688 | r = enforce_user(context, uid); |
2689 | if (r < 0) { | |
2690 | *exit_status = EXIT_USER; | |
2691 | return r; | |
5b6319dc | 2692 | } |
755d4b67 IP |
2693 | if (context->capability_ambient_set != 0) { |
2694 | ||
2695 | /* Fix the ambient capabilities after user change. */ | |
2696 | r = capability_ambient_set_apply(context->capability_ambient_set, false); | |
2697 | if (r < 0) { | |
2698 | *exit_status = EXIT_CAPABILITIES; | |
2699 | return r; | |
2700 | } | |
2701 | ||
2702 | /* If we were asked to change user and ambient capabilities | |
2703 | * were requested, we had to add keep-caps to the securebits | |
2704 | * so that we would maintain the inherited capability set | |
2705 | * through the setresuid(). Make sure that the bit is added | |
2706 | * also to the context secure_bits so that we don't try to | |
2707 | * drop the bit away next. */ | |
2708 | ||
7f508f2c | 2709 | secure_bits |= 1<<SECURE_KEEP_CAPS; |
755d4b67 | 2710 | } |
5b6319dc | 2711 | } |
d35fbf6b | 2712 | |
5cd9cd35 LP |
2713 | /* Apply the MAC contexts late, but before seccomp syscall filtering, as those should really be last to |
2714 | * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires | |
2715 | * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls | |
2716 | * are restricted. */ | |
2717 | ||
2718 | #ifdef HAVE_SELINUX | |
2719 | if (mac_selinux_use()) { | |
2720 | char *exec_context = mac_selinux_context_net ?: context->selinux_context; | |
2721 | ||
2722 | if (exec_context) { | |
2723 | r = setexeccon(exec_context); | |
2724 | if (r < 0) { | |
2725 | *exit_status = EXIT_SELINUX_CONTEXT; | |
2726 | return r; | |
2727 | } | |
2728 | } | |
2729 | } | |
2730 | #endif | |
2731 | ||
2732 | r = setup_smack(context, command); | |
2733 | if (r < 0) { | |
2734 | *exit_status = EXIT_SMACK_PROCESS_LABEL; | |
2735 | return r; | |
2736 | } | |
2737 | ||
2738 | #ifdef HAVE_APPARMOR | |
2739 | if (context->apparmor_profile && mac_apparmor_use()) { | |
2740 | r = aa_change_onexec(context->apparmor_profile); | |
2741 | if (r < 0 && !context->apparmor_profile_ignore) { | |
2742 | *exit_status = EXIT_APPARMOR_PROFILE; | |
2743 | return -errno; | |
2744 | } | |
2745 | } | |
2746 | #endif | |
2747 | ||
d35fbf6b DM |
2748 | /* PR_GET_SECUREBITS is not privileged, while |
2749 | * PR_SET_SECUREBITS is. So to suppress | |
2750 | * potential EPERMs we'll try not to call | |
2751 | * PR_SET_SECUREBITS unless necessary. */ | |
755d4b67 IP |
2752 | if (prctl(PR_GET_SECUREBITS) != secure_bits) |
2753 | if (prctl(PR_SET_SECUREBITS, secure_bits) < 0) { | |
ff0af2a1 | 2754 | *exit_status = EXIT_SECUREBITS; |
d35fbf6b | 2755 | return -errno; |
ff01d048 | 2756 | } |
5b6319dc | 2757 | |
59eeb84b | 2758 | if (context_has_no_new_privileges(context)) |
d35fbf6b | 2759 | if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { |
ff0af2a1 | 2760 | *exit_status = EXIT_NO_NEW_PRIVILEGES; |
d35fbf6b DM |
2761 | return -errno; |
2762 | } | |
2763 | ||
2764 | #ifdef HAVE_SECCOMP | |
59eeb84b | 2765 | if (context_has_address_families(context)) { |
83f12b27 | 2766 | r = apply_address_families(unit, context); |
ff0af2a1 LP |
2767 | if (r < 0) { |
2768 | *exit_status = EXIT_ADDRESS_FAMILIES; | |
2769 | return r; | |
4c2630eb MS |
2770 | } |
2771 | } | |
04aa0cb9 | 2772 | |
f3e43635 | 2773 | if (context->memory_deny_write_execute) { |
83f12b27 | 2774 | r = apply_memory_deny_write_execute(unit, context); |
f3e43635 TM |
2775 | if (r < 0) { |
2776 | *exit_status = EXIT_SECCOMP; | |
2777 | return r; | |
2778 | } | |
2779 | } | |
f4170c67 LP |
2780 | |
2781 | if (context->restrict_realtime) { | |
83f12b27 | 2782 | r = apply_restrict_realtime(unit, context); |
f4170c67 LP |
2783 | if (r < 0) { |
2784 | *exit_status = EXIT_SECCOMP; | |
2785 | return r; | |
2786 | } | |
2787 | } | |
2788 | ||
add00535 LP |
2789 | r = apply_restrict_namespaces(unit, context); |
2790 | if (r < 0) { | |
2791 | *exit_status = EXIT_SECCOMP; | |
2792 | return r; | |
2793 | } | |
2794 | ||
59eeb84b LP |
2795 | if (context->protect_kernel_tunables) { |
2796 | r = apply_protect_sysctl(unit, context); | |
2797 | if (r < 0) { | |
2798 | *exit_status = EXIT_SECCOMP; | |
2799 | return r; | |
502d704e DH |
2800 | } |
2801 | } | |
2802 | ||
2803 | if (context->protect_kernel_modules) { | |
2804 | r = apply_protect_kernel_modules(unit, context); | |
2805 | if (r < 0) { | |
2806 | *exit_status = EXIT_SECCOMP; | |
2807 | return r; | |
59eeb84b LP |
2808 | } |
2809 | } | |
2810 | ||
ba128bb8 LP |
2811 | if (context->private_devices) { |
2812 | r = apply_private_devices(unit, context); | |
2813 | if (r < 0) { | |
2814 | *exit_status = EXIT_SECCOMP; | |
2815 | return r; | |
2816 | } | |
2817 | } | |
2818 | ||
5cd9cd35 LP |
2819 | /* This really should remain the last step before the execve(), to make sure our own code is unaffected |
2820 | * by the filter as little as possible. */ | |
59eeb84b | 2821 | if (context_has_syscall_filters(context)) { |
83f12b27 | 2822 | r = apply_seccomp(unit, context); |
ff0af2a1 LP |
2823 | if (r < 0) { |
2824 | *exit_status = EXIT_SECCOMP; | |
2825 | return r; | |
81a2b7ce | 2826 | } |
d35fbf6b DM |
2827 | } |
2828 | #endif | |
d35fbf6b | 2829 | } |
034c6ed7 | 2830 | |
2065ca69 | 2831 | final_argv = replace_env_argv(argv, accum_env); |
d35fbf6b | 2832 | if (!final_argv) { |
ff0af2a1 | 2833 | *exit_status = EXIT_MEMORY; |
d35fbf6b DM |
2834 | return -ENOMEM; |
2835 | } | |
034c6ed7 | 2836 | |
553d2243 | 2837 | if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) { |
d35fbf6b | 2838 | _cleanup_free_ char *line; |
81a2b7ce | 2839 | |
d35fbf6b DM |
2840 | line = exec_command_line(final_argv); |
2841 | if (line) { | |
2842 | log_open(); | |
f2341e0a LP |
2843 | log_struct(LOG_DEBUG, |
2844 | LOG_UNIT_ID(unit), | |
2845 | "EXECUTABLE=%s", command->path, | |
2846 | LOG_UNIT_MESSAGE(unit, "Executing: %s", line), | |
2847 | NULL); | |
d35fbf6b DM |
2848 | log_close(); |
2849 | } | |
2850 | } | |
dd305ec9 | 2851 | |
2065ca69 | 2852 | execve(command->path, final_argv, accum_env); |
ff0af2a1 | 2853 | *exit_status = EXIT_EXEC; |
d35fbf6b DM |
2854 | return -errno; |
2855 | } | |
81a2b7ce | 2856 | |
f2341e0a LP |
2857 | int exec_spawn(Unit *unit, |
2858 | ExecCommand *command, | |
d35fbf6b DM |
2859 | const ExecContext *context, |
2860 | const ExecParameters *params, | |
2861 | ExecRuntime *runtime, | |
29206d46 | 2862 | DynamicCreds *dcreds, |
d35fbf6b | 2863 | pid_t *ret) { |
8351ceae | 2864 | |
d35fbf6b DM |
2865 | _cleanup_strv_free_ char **files_env = NULL; |
2866 | int *fds = NULL; unsigned n_fds = 0; | |
ff0af2a1 LP |
2867 | _cleanup_free_ char *line = NULL; |
2868 | int socket_fd, r; | |
52c239d7 | 2869 | int named_iofds[3] = { -1, -1, -1 }; |
ff0af2a1 | 2870 | char **argv; |
d35fbf6b | 2871 | pid_t pid; |
8351ceae | 2872 | |
f2341e0a | 2873 | assert(unit); |
d35fbf6b DM |
2874 | assert(command); |
2875 | assert(context); | |
2876 | assert(ret); | |
2877 | assert(params); | |
2878 | assert(params->fds || params->n_fds <= 0); | |
4298d0b5 | 2879 | |
d35fbf6b DM |
2880 | if (context->std_input == EXEC_INPUT_SOCKET || |
2881 | context->std_output == EXEC_OUTPUT_SOCKET || | |
2882 | context->std_error == EXEC_OUTPUT_SOCKET) { | |
17df7223 | 2883 | |
ff0af2a1 | 2884 | if (params->n_fds != 1) { |
f2341e0a | 2885 | log_unit_error(unit, "Got more than one socket."); |
d35fbf6b | 2886 | return -EINVAL; |
ff0af2a1 | 2887 | } |
eef65bf3 | 2888 | |
d35fbf6b DM |
2889 | socket_fd = params->fds[0]; |
2890 | } else { | |
2891 | socket_fd = -1; | |
2892 | fds = params->fds; | |
2893 | n_fds = params->n_fds; | |
2894 | } | |
94f04347 | 2895 | |
52c239d7 LB |
2896 | r = exec_context_named_iofds(unit, context, params, named_iofds); |
2897 | if (r < 0) | |
2898 | return log_unit_error_errno(unit, r, "Failed to load a named file descriptor: %m"); | |
2899 | ||
f2341e0a | 2900 | r = exec_context_load_environment(unit, context, &files_env); |
ff0af2a1 | 2901 | if (r < 0) |
f2341e0a | 2902 | return log_unit_error_errno(unit, r, "Failed to load environment files: %m"); |
034c6ed7 | 2903 | |
d35fbf6b | 2904 | argv = params->argv ?: command->argv; |
d35fbf6b DM |
2905 | line = exec_command_line(argv); |
2906 | if (!line) | |
2907 | return log_oom(); | |
fab56fc5 | 2908 | |
f2341e0a LP |
2909 | log_struct(LOG_DEBUG, |
2910 | LOG_UNIT_ID(unit), | |
2911 | LOG_UNIT_MESSAGE(unit, "About to execute: %s", line), | |
2912 | "EXECUTABLE=%s", command->path, | |
2913 | NULL); | |
d35fbf6b DM |
2914 | pid = fork(); |
2915 | if (pid < 0) | |
74129a12 | 2916 | return log_unit_error_errno(unit, errno, "Failed to fork: %m"); |
d35fbf6b DM |
2917 | |
2918 | if (pid == 0) { | |
ff0af2a1 LP |
2919 | int exit_status; |
2920 | ||
f2341e0a LP |
2921 | r = exec_child(unit, |
2922 | command, | |
ff0af2a1 LP |
2923 | context, |
2924 | params, | |
2925 | runtime, | |
29206d46 | 2926 | dcreds, |
ff0af2a1 LP |
2927 | argv, |
2928 | socket_fd, | |
52c239d7 | 2929 | named_iofds, |
ff0af2a1 LP |
2930 | fds, n_fds, |
2931 | files_env, | |
00d9ef85 | 2932 | unit->manager->user_lookup_fds[1], |
ff0af2a1 LP |
2933 | &exit_status); |
2934 | if (r < 0) { | |
4c2630eb | 2935 | log_open(); |
f2341e0a LP |
2936 | log_struct_errno(LOG_ERR, r, |
2937 | LOG_MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED), | |
2938 | LOG_UNIT_ID(unit), | |
2939 | LOG_UNIT_MESSAGE(unit, "Failed at step %s spawning %s: %m", | |
2940 | exit_status_to_string(exit_status, EXIT_STATUS_SYSTEMD), | |
2941 | command->path), | |
2942 | "EXECUTABLE=%s", command->path, | |
2943 | NULL); | |
4c2630eb MS |
2944 | } |
2945 | ||
ff0af2a1 | 2946 | _exit(exit_status); |
034c6ed7 LP |
2947 | } |
2948 | ||
f2341e0a | 2949 | log_unit_debug(unit, "Forked %s as "PID_FMT, command->path, pid); |
23635a85 | 2950 | |
80876c20 LP |
2951 | /* We add the new process to the cgroup both in the child (so |
2952 | * that we can be sure that no user code is ever executed | |
2953 | * outside of the cgroup) and in the parent (so that we can be | |
2954 | * sure that when we kill the cgroup the process will be | |
2955 | * killed too). */ | |
d35fbf6b | 2956 | if (params->cgroup_path) |
dd305ec9 | 2957 | (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, pid); |
2da3263a | 2958 | |
b58b4116 | 2959 | exec_status_start(&command->exec_status, pid); |
9fb86720 | 2960 | |
034c6ed7 | 2961 | *ret = pid; |
5cb5a6ff LP |
2962 | return 0; |
2963 | } | |
2964 | ||
034c6ed7 LP |
2965 | void exec_context_init(ExecContext *c) { |
2966 | assert(c); | |
2967 | ||
4c12626c | 2968 | c->umask = 0022; |
9eba9da4 | 2969 | c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0); |
94f04347 | 2970 | c->cpu_sched_policy = SCHED_OTHER; |
071830ff | 2971 | c->syslog_priority = LOG_DAEMON|LOG_INFO; |
74922904 | 2972 | c->syslog_level_prefix = true; |
353e12c2 | 2973 | c->ignore_sigpipe = true; |
3a43da28 | 2974 | c->timer_slack_nsec = NSEC_INFINITY; |
050f7277 | 2975 | c->personality = PERSONALITY_INVALID; |
e66cf1a3 | 2976 | c->runtime_directory_mode = 0755; |
a103496c | 2977 | c->capability_bounding_set = CAP_ALL; |
add00535 | 2978 | c->restrict_namespaces = NAMESPACE_FLAGS_ALL; |
034c6ed7 LP |
2979 | } |
2980 | ||
613b411c | 2981 | void exec_context_done(ExecContext *c) { |
5cb5a6ff LP |
2982 | unsigned l; |
2983 | ||
2984 | assert(c); | |
2985 | ||
6796073e LP |
2986 | c->environment = strv_free(c->environment); |
2987 | c->environment_files = strv_free(c->environment_files); | |
b4c14404 | 2988 | c->pass_environment = strv_free(c->pass_environment); |
8c7be95e | 2989 | |
1f6b4113 | 2990 | for (l = 0; l < ELEMENTSOF(c->rlimit); l++) |
a1e58e8e | 2991 | c->rlimit[l] = mfree(c->rlimit[l]); |
034c6ed7 | 2992 | |
52c239d7 LB |
2993 | for (l = 0; l < 3; l++) |
2994 | c->stdio_fdname[l] = mfree(c->stdio_fdname[l]); | |
2995 | ||
a1e58e8e LP |
2996 | c->working_directory = mfree(c->working_directory); |
2997 | c->root_directory = mfree(c->root_directory); | |
2998 | c->tty_path = mfree(c->tty_path); | |
2999 | c->syslog_identifier = mfree(c->syslog_identifier); | |
3000 | c->user = mfree(c->user); | |
3001 | c->group = mfree(c->group); | |
034c6ed7 | 3002 | |
6796073e | 3003 | c->supplementary_groups = strv_free(c->supplementary_groups); |
94f04347 | 3004 | |
a1e58e8e | 3005 | c->pam_name = mfree(c->pam_name); |
5b6319dc | 3006 | |
2a624c36 AP |
3007 | c->read_only_paths = strv_free(c->read_only_paths); |
3008 | c->read_write_paths = strv_free(c->read_write_paths); | |
3009 | c->inaccessible_paths = strv_free(c->inaccessible_paths); | |
82c121a4 LP |
3010 | |
3011 | if (c->cpuset) | |
3012 | CPU_FREE(c->cpuset); | |
86a3475b | 3013 | |
a1e58e8e LP |
3014 | c->utmp_id = mfree(c->utmp_id); |
3015 | c->selinux_context = mfree(c->selinux_context); | |
3016 | c->apparmor_profile = mfree(c->apparmor_profile); | |
eef65bf3 | 3017 | |
525d3cc7 LP |
3018 | c->syscall_filter = set_free(c->syscall_filter); |
3019 | c->syscall_archs = set_free(c->syscall_archs); | |
3020 | c->address_families = set_free(c->address_families); | |
e66cf1a3 | 3021 | |
6796073e | 3022 | c->runtime_directory = strv_free(c->runtime_directory); |
e66cf1a3 LP |
3023 | } |
3024 | ||
3025 | int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) { | |
3026 | char **i; | |
3027 | ||
3028 | assert(c); | |
3029 | ||
3030 | if (!runtime_prefix) | |
3031 | return 0; | |
3032 | ||
3033 | STRV_FOREACH(i, c->runtime_directory) { | |
3034 | _cleanup_free_ char *p; | |
3035 | ||
605405c6 | 3036 | p = strjoin(runtime_prefix, "/", *i); |
e66cf1a3 LP |
3037 | if (!p) |
3038 | return -ENOMEM; | |
3039 | ||
3040 | /* We execute this synchronously, since we need to be | |
3041 | * sure this is gone when we start the service | |
3042 | * next. */ | |
c6878637 | 3043 | (void) rm_rf(p, REMOVE_ROOT); |
e66cf1a3 LP |
3044 | } |
3045 | ||
3046 | return 0; | |
5cb5a6ff LP |
3047 | } |
3048 | ||
43d0fcbd LP |
3049 | void exec_command_done(ExecCommand *c) { |
3050 | assert(c); | |
3051 | ||
a1e58e8e | 3052 | c->path = mfree(c->path); |
43d0fcbd | 3053 | |
6796073e | 3054 | c->argv = strv_free(c->argv); |
43d0fcbd LP |
3055 | } |
3056 | ||
3057 | void exec_command_done_array(ExecCommand *c, unsigned n) { | |
3058 | unsigned i; | |
3059 | ||
3060 | for (i = 0; i < n; i++) | |
3061 | exec_command_done(c+i); | |
3062 | } | |
3063 | ||
f1acf85a | 3064 | ExecCommand* exec_command_free_list(ExecCommand *c) { |
5cb5a6ff LP |
3065 | ExecCommand *i; |
3066 | ||
3067 | while ((i = c)) { | |
71fda00f | 3068 | LIST_REMOVE(command, c, i); |
43d0fcbd | 3069 | exec_command_done(i); |
5cb5a6ff LP |
3070 | free(i); |
3071 | } | |
f1acf85a ZJS |
3072 | |
3073 | return NULL; | |
5cb5a6ff LP |
3074 | } |
3075 | ||
034c6ed7 LP |
3076 | void exec_command_free_array(ExecCommand **c, unsigned n) { |
3077 | unsigned i; | |
3078 | ||
f1acf85a ZJS |
3079 | for (i = 0; i < n; i++) |
3080 | c[i] = exec_command_free_list(c[i]); | |
034c6ed7 LP |
3081 | } |
3082 | ||
039f0e70 | 3083 | typedef struct InvalidEnvInfo { |
f2341e0a | 3084 | Unit *unit; |
039f0e70 LP |
3085 | const char *path; |
3086 | } InvalidEnvInfo; | |
3087 | ||
3088 | static void invalid_env(const char *p, void *userdata) { | |
3089 | InvalidEnvInfo *info = userdata; | |
3090 | ||
f2341e0a | 3091 | log_unit_error(info->unit, "Ignoring invalid environment assignment '%s': %s", p, info->path); |
039f0e70 LP |
3092 | } |
3093 | ||
52c239d7 LB |
3094 | const char* exec_context_fdname(const ExecContext *c, int fd_index) { |
3095 | assert(c); | |
3096 | ||
3097 | switch (fd_index) { | |
3098 | case STDIN_FILENO: | |
3099 | if (c->std_input != EXEC_INPUT_NAMED_FD) | |
3100 | return NULL; | |
3101 | return c->stdio_fdname[STDIN_FILENO] ?: "stdin"; | |
3102 | case STDOUT_FILENO: | |
3103 | if (c->std_output != EXEC_OUTPUT_NAMED_FD) | |
3104 | return NULL; | |
3105 | return c->stdio_fdname[STDOUT_FILENO] ?: "stdout"; | |
3106 | case STDERR_FILENO: | |
3107 | if (c->std_error != EXEC_OUTPUT_NAMED_FD) | |
3108 | return NULL; | |
3109 | return c->stdio_fdname[STDERR_FILENO] ?: "stderr"; | |
3110 | default: | |
3111 | return NULL; | |
3112 | } | |
3113 | } | |
3114 | ||
3115 | int exec_context_named_iofds(Unit *unit, const ExecContext *c, const ExecParameters *p, int named_iofds[3]) { | |
3116 | unsigned i, targets; | |
3117 | const char *stdio_fdname[3]; | |
3118 | ||
3119 | assert(c); | |
3120 | assert(p); | |
3121 | ||
3122 | targets = (c->std_input == EXEC_INPUT_NAMED_FD) + | |
3123 | (c->std_output == EXEC_OUTPUT_NAMED_FD) + | |
3124 | (c->std_error == EXEC_OUTPUT_NAMED_FD); | |
3125 | ||
3126 | for (i = 0; i < 3; i++) | |
3127 | stdio_fdname[i] = exec_context_fdname(c, i); | |
3128 | ||
3129 | for (i = 0; i < p->n_fds && targets > 0; i++) | |
3130 | if (named_iofds[STDIN_FILENO] < 0 && c->std_input == EXEC_INPUT_NAMED_FD && stdio_fdname[STDIN_FILENO] && streq(p->fd_names[i], stdio_fdname[STDIN_FILENO])) { | |
3131 | named_iofds[STDIN_FILENO] = p->fds[i]; | |
3132 | targets--; | |
3133 | } else if (named_iofds[STDOUT_FILENO] < 0 && c->std_output == EXEC_OUTPUT_NAMED_FD && stdio_fdname[STDOUT_FILENO] && streq(p->fd_names[i], stdio_fdname[STDOUT_FILENO])) { | |
3134 | named_iofds[STDOUT_FILENO] = p->fds[i]; | |
3135 | targets--; | |
3136 | } else if (named_iofds[STDERR_FILENO] < 0 && c->std_error == EXEC_OUTPUT_NAMED_FD && stdio_fdname[STDERR_FILENO] && streq(p->fd_names[i], stdio_fdname[STDERR_FILENO])) { | |
3137 | named_iofds[STDERR_FILENO] = p->fds[i]; | |
3138 | targets--; | |
3139 | } | |
3140 | ||
3141 | return (targets == 0 ? 0 : -ENOENT); | |
3142 | } | |
3143 | ||
f2341e0a | 3144 | int exec_context_load_environment(Unit *unit, const ExecContext *c, char ***l) { |
8c7be95e LP |
3145 | char **i, **r = NULL; |
3146 | ||
3147 | assert(c); | |
3148 | assert(l); | |
3149 | ||
3150 | STRV_FOREACH(i, c->environment_files) { | |
3151 | char *fn; | |
3152 | int k; | |
3153 | bool ignore = false; | |
3154 | char **p; | |
7fd1b19b | 3155 | _cleanup_globfree_ glob_t pglob = {}; |
2bef10ab | 3156 | int count, n; |
8c7be95e LP |
3157 | |
3158 | fn = *i; | |
3159 | ||
3160 | if (fn[0] == '-') { | |
3161 | ignore = true; | |
313cefa1 | 3162 | fn++; |
8c7be95e LP |
3163 | } |
3164 | ||
3165 | if (!path_is_absolute(fn)) { | |
8c7be95e LP |
3166 | if (ignore) |
3167 | continue; | |
3168 | ||
3169 | strv_free(r); | |
3170 | return -EINVAL; | |
3171 | } | |
3172 | ||
2bef10ab | 3173 | /* Filename supports globbing, take all matching files */ |
2bef10ab PL |
3174 | errno = 0; |
3175 | if (glob(fn, 0, NULL, &pglob) != 0) { | |
2bef10ab PL |
3176 | if (ignore) |
3177 | continue; | |
8c7be95e | 3178 | |
2bef10ab | 3179 | strv_free(r); |
f5e5c28f | 3180 | return errno > 0 ? -errno : -EINVAL; |
2bef10ab PL |
3181 | } |
3182 | count = pglob.gl_pathc; | |
3183 | if (count == 0) { | |
8c7be95e LP |
3184 | if (ignore) |
3185 | continue; | |
3186 | ||
3187 | strv_free(r); | |
2bef10ab | 3188 | return -EINVAL; |
8c7be95e | 3189 | } |
2bef10ab | 3190 | for (n = 0; n < count; n++) { |
717603e3 | 3191 | k = load_env_file(NULL, pglob.gl_pathv[n], NULL, &p); |
2bef10ab PL |
3192 | if (k < 0) { |
3193 | if (ignore) | |
3194 | continue; | |
8c7be95e | 3195 | |
2bef10ab | 3196 | strv_free(r); |
2bef10ab | 3197 | return k; |
e9c1ea9d | 3198 | } |
ebc05a09 | 3199 | /* Log invalid environment variables with filename */ |
039f0e70 LP |
3200 | if (p) { |
3201 | InvalidEnvInfo info = { | |
f2341e0a | 3202 | .unit = unit, |
039f0e70 LP |
3203 | .path = pglob.gl_pathv[n] |
3204 | }; | |
3205 | ||
3206 | p = strv_env_clean_with_callback(p, invalid_env, &info); | |
3207 | } | |
8c7be95e | 3208 | |
2bef10ab PL |
3209 | if (r == NULL) |
3210 | r = p; | |
3211 | else { | |
3212 | char **m; | |
8c7be95e | 3213 | |
2bef10ab PL |
3214 | m = strv_env_merge(2, r, p); |
3215 | strv_free(r); | |
3216 | strv_free(p); | |
c84a9488 | 3217 | if (!m) |
2bef10ab | 3218 | return -ENOMEM; |
2bef10ab PL |
3219 | |
3220 | r = m; | |
3221 | } | |
8c7be95e LP |
3222 | } |
3223 | } | |
3224 | ||
3225 | *l = r; | |
3226 | ||
3227 | return 0; | |
3228 | } | |
3229 | ||
6ac8fdc9 | 3230 | static bool tty_may_match_dev_console(const char *tty) { |
e1d75803 | 3231 | _cleanup_free_ char *active = NULL; |
7d6884b6 | 3232 | char *console; |
6ac8fdc9 | 3233 | |
1e22b5cd LP |
3234 | if (!tty) |
3235 | return true; | |
3236 | ||
6ac8fdc9 MS |
3237 | if (startswith(tty, "/dev/")) |
3238 | tty += 5; | |
3239 | ||
3240 | /* trivial identity? */ | |
3241 | if (streq(tty, "console")) | |
3242 | return true; | |
3243 | ||
3244 | console = resolve_dev_console(&active); | |
3245 | /* if we could not resolve, assume it may */ | |
3246 | if (!console) | |
3247 | return true; | |
3248 | ||
3249 | /* "tty0" means the active VC, so it may be the same sometimes */ | |
e1d75803 | 3250 | return streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty)); |
6ac8fdc9 MS |
3251 | } |
3252 | ||
3253 | bool exec_context_may_touch_console(ExecContext *ec) { | |
1e22b5cd LP |
3254 | |
3255 | return (ec->tty_reset || | |
3256 | ec->tty_vhangup || | |
3257 | ec->tty_vt_disallocate || | |
6ac8fdc9 MS |
3258 | is_terminal_input(ec->std_input) || |
3259 | is_terminal_output(ec->std_output) || | |
3260 | is_terminal_output(ec->std_error)) && | |
1e22b5cd | 3261 | tty_may_match_dev_console(exec_context_tty_path(ec)); |
6ac8fdc9 MS |
3262 | } |
3263 | ||
15ae422b LP |
3264 | static void strv_fprintf(FILE *f, char **l) { |
3265 | char **g; | |
3266 | ||
3267 | assert(f); | |
3268 | ||
3269 | STRV_FOREACH(g, l) | |
3270 | fprintf(f, " %s", *g); | |
3271 | } | |
3272 | ||
5cb5a6ff | 3273 | void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { |
c2bbd90b | 3274 | char **e, **d; |
94f04347 | 3275 | unsigned i; |
add00535 | 3276 | int r; |
9eba9da4 | 3277 | |
5cb5a6ff LP |
3278 | assert(c); |
3279 | assert(f); | |
3280 | ||
4ad49000 | 3281 | prefix = strempty(prefix); |
5cb5a6ff LP |
3282 | |
3283 | fprintf(f, | |
94f04347 LP |
3284 | "%sUMask: %04o\n" |
3285 | "%sWorkingDirectory: %s\n" | |
451a074f | 3286 | "%sRootDirectory: %s\n" |
15ae422b | 3287 | "%sNonBlocking: %s\n" |
64747e2d | 3288 | "%sPrivateTmp: %s\n" |
7f112f50 | 3289 | "%sPrivateDevices: %s\n" |
59eeb84b | 3290 | "%sProtectKernelTunables: %s\n" |
e66a2f65 | 3291 | "%sProtectKernelModules: %s\n" |
59eeb84b | 3292 | "%sProtectControlGroups: %s\n" |
d251207d LP |
3293 | "%sPrivateNetwork: %s\n" |
3294 | "%sPrivateUsers: %s\n" | |
1b8689f9 LP |
3295 | "%sProtectHome: %s\n" |
3296 | "%sProtectSystem: %s\n" | |
f3e43635 | 3297 | "%sIgnoreSIGPIPE: %s\n" |
f4170c67 LP |
3298 | "%sMemoryDenyWriteExecute: %s\n" |
3299 | "%sRestrictRealtime: %s\n", | |
5cb5a6ff | 3300 | prefix, c->umask, |
9eba9da4 | 3301 | prefix, c->working_directory ? c->working_directory : "/", |
451a074f | 3302 | prefix, c->root_directory ? c->root_directory : "/", |
15ae422b | 3303 | prefix, yes_no(c->non_blocking), |
64747e2d | 3304 | prefix, yes_no(c->private_tmp), |
7f112f50 | 3305 | prefix, yes_no(c->private_devices), |
59eeb84b | 3306 | prefix, yes_no(c->protect_kernel_tunables), |
e66a2f65 | 3307 | prefix, yes_no(c->protect_kernel_modules), |
59eeb84b | 3308 | prefix, yes_no(c->protect_control_groups), |
d251207d LP |
3309 | prefix, yes_no(c->private_network), |
3310 | prefix, yes_no(c->private_users), | |
1b8689f9 LP |
3311 | prefix, protect_home_to_string(c->protect_home), |
3312 | prefix, protect_system_to_string(c->protect_system), | |
f3e43635 | 3313 | prefix, yes_no(c->ignore_sigpipe), |
f4170c67 LP |
3314 | prefix, yes_no(c->memory_deny_write_execute), |
3315 | prefix, yes_no(c->restrict_realtime)); | |
fb33a393 | 3316 | |
8c7be95e LP |
3317 | STRV_FOREACH(e, c->environment) |
3318 | fprintf(f, "%sEnvironment: %s\n", prefix, *e); | |
3319 | ||
3320 | STRV_FOREACH(e, c->environment_files) | |
3321 | fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e); | |
94f04347 | 3322 | |
b4c14404 FB |
3323 | STRV_FOREACH(e, c->pass_environment) |
3324 | fprintf(f, "%sPassEnvironment: %s\n", prefix, *e); | |
3325 | ||
c2bbd90b EV |
3326 | fprintf(f, "%sRuntimeDirectoryMode: %04o\n", prefix, c->runtime_directory_mode); |
3327 | ||
3328 | STRV_FOREACH(d, c->runtime_directory) | |
3329 | fprintf(f, "%sRuntimeDirectory: %s\n", prefix, *d); | |
3330 | ||
fb33a393 LP |
3331 | if (c->nice_set) |
3332 | fprintf(f, | |
3333 | "%sNice: %i\n", | |
3334 | prefix, c->nice); | |
3335 | ||
dd6c17b1 | 3336 | if (c->oom_score_adjust_set) |
fb33a393 | 3337 | fprintf(f, |
dd6c17b1 LP |
3338 | "%sOOMScoreAdjust: %i\n", |
3339 | prefix, c->oom_score_adjust); | |
9eba9da4 | 3340 | |
94f04347 | 3341 | for (i = 0; i < RLIM_NLIMITS; i++) |
3c11da9d EV |
3342 | if (c->rlimit[i]) { |
3343 | fprintf(f, "%s%s: " RLIM_FMT "\n", | |
3344 | prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max); | |
3345 | fprintf(f, "%s%sSoft: " RLIM_FMT "\n", | |
3346 | prefix, rlimit_to_string(i), c->rlimit[i]->rlim_cur); | |
3347 | } | |
94f04347 | 3348 | |
f8b69d1d | 3349 | if (c->ioprio_set) { |
1756a011 | 3350 | _cleanup_free_ char *class_str = NULL; |
f8b69d1d | 3351 | |
1756a011 | 3352 | ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str); |
9eba9da4 LP |
3353 | fprintf(f, |
3354 | "%sIOSchedulingClass: %s\n" | |
3355 | "%sIOPriority: %i\n", | |
f8b69d1d | 3356 | prefix, strna(class_str), |
9eba9da4 | 3357 | prefix, (int) IOPRIO_PRIO_DATA(c->ioprio)); |
f8b69d1d | 3358 | } |
94f04347 | 3359 | |
f8b69d1d | 3360 | if (c->cpu_sched_set) { |
1756a011 | 3361 | _cleanup_free_ char *policy_str = NULL; |
f8b69d1d | 3362 | |
1756a011 | 3363 | sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str); |
94f04347 LP |
3364 | fprintf(f, |
3365 | "%sCPUSchedulingPolicy: %s\n" | |
38b48754 LP |
3366 | "%sCPUSchedulingPriority: %i\n" |
3367 | "%sCPUSchedulingResetOnFork: %s\n", | |
f8b69d1d | 3368 | prefix, strna(policy_str), |
38b48754 LP |
3369 | prefix, c->cpu_sched_priority, |
3370 | prefix, yes_no(c->cpu_sched_reset_on_fork)); | |
b929bf04 | 3371 | } |
94f04347 | 3372 | |
82c121a4 | 3373 | if (c->cpuset) { |
94f04347 | 3374 | fprintf(f, "%sCPUAffinity:", prefix); |
82c121a4 LP |
3375 | for (i = 0; i < c->cpuset_ncpus; i++) |
3376 | if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset)) | |
43a99a7a | 3377 | fprintf(f, " %u", i); |
94f04347 LP |
3378 | fputs("\n", f); |
3379 | } | |
3380 | ||
3a43da28 | 3381 | if (c->timer_slack_nsec != NSEC_INFINITY) |
ccd06097 | 3382 | fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec); |
94f04347 LP |
3383 | |
3384 | fprintf(f, | |
80876c20 LP |
3385 | "%sStandardInput: %s\n" |
3386 | "%sStandardOutput: %s\n" | |
3387 | "%sStandardError: %s\n", | |
3388 | prefix, exec_input_to_string(c->std_input), | |
3389 | prefix, exec_output_to_string(c->std_output), | |
3390 | prefix, exec_output_to_string(c->std_error)); | |
3391 | ||
3392 | if (c->tty_path) | |
3393 | fprintf(f, | |
6ea832a2 LP |
3394 | "%sTTYPath: %s\n" |
3395 | "%sTTYReset: %s\n" | |
3396 | "%sTTYVHangup: %s\n" | |
3397 | "%sTTYVTDisallocate: %s\n", | |
3398 | prefix, c->tty_path, | |
3399 | prefix, yes_no(c->tty_reset), | |
3400 | prefix, yes_no(c->tty_vhangup), | |
3401 | prefix, yes_no(c->tty_vt_disallocate)); | |
94f04347 | 3402 | |
5ce70e5b ZJS |
3403 | if (c->std_output == EXEC_OUTPUT_SYSLOG || |
3404 | c->std_output == EXEC_OUTPUT_KMSG || | |
3405 | c->std_output == EXEC_OUTPUT_JOURNAL || | |
3406 | c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || | |
3407 | c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE || | |
3408 | c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE || | |
3409 | c->std_error == EXEC_OUTPUT_SYSLOG || | |
3410 | c->std_error == EXEC_OUTPUT_KMSG || | |
3411 | c->std_error == EXEC_OUTPUT_JOURNAL || | |
3412 | c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || | |
3413 | c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || | |
3414 | c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) { | |
f8b69d1d | 3415 | |
5ce70e5b | 3416 | _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL; |
f8b69d1d | 3417 | |
5ce70e5b ZJS |
3418 | log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str); |
3419 | log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str); | |
f8b69d1d | 3420 | |
94f04347 LP |
3421 | fprintf(f, |
3422 | "%sSyslogFacility: %s\n" | |
3423 | "%sSyslogLevel: %s\n", | |
f8b69d1d MS |
3424 | prefix, strna(fac_str), |
3425 | prefix, strna(lvl_str)); | |
f8b69d1d | 3426 | } |
94f04347 | 3427 | |
94f04347 LP |
3428 | if (c->secure_bits) |
3429 | fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n", | |
3430 | prefix, | |
cbb21cca ZJS |
3431 | (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "", |
3432 | (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "", | |
3433 | (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "", | |
3434 | (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "", | |
3435 | (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "", | |
3436 | (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : ""); | |
94f04347 | 3437 | |
a103496c | 3438 | if (c->capability_bounding_set != CAP_ALL) { |
ae556c21 | 3439 | unsigned long l; |
260abb78 | 3440 | fprintf(f, "%sCapabilityBoundingSet:", prefix); |
94f04347 | 3441 | |
64685e0c | 3442 | for (l = 0; l <= cap_last_cap(); l++) |
a103496c | 3443 | if (c->capability_bounding_set & (UINT64_C(1) << l)) |
2822da4f | 3444 | fprintf(f, " %s", strna(capability_to_name(l))); |
94f04347 LP |
3445 | |
3446 | fputs("\n", f); | |
755d4b67 IP |
3447 | } |
3448 | ||
3449 | if (c->capability_ambient_set != 0) { | |
3450 | unsigned long l; | |
3451 | fprintf(f, "%sAmbientCapabilities:", prefix); | |
3452 | ||
3453 | for (l = 0; l <= cap_last_cap(); l++) | |
3454 | if (c->capability_ambient_set & (UINT64_C(1) << l)) | |
3455 | fprintf(f, " %s", strna(capability_to_name(l))); | |
3456 | ||
3457 | fputs("\n", f); | |
94f04347 LP |
3458 | } |
3459 | ||
3460 | if (c->user) | |
f2d3769a | 3461 | fprintf(f, "%sUser: %s\n", prefix, c->user); |
94f04347 | 3462 | if (c->group) |
f2d3769a | 3463 | fprintf(f, "%sGroup: %s\n", prefix, c->group); |
94f04347 | 3464 | |
29206d46 LP |
3465 | fprintf(f, "%sDynamicUser: %s\n", prefix, yes_no(c->dynamic_user)); |
3466 | ||
15ae422b | 3467 | if (strv_length(c->supplementary_groups) > 0) { |
94f04347 | 3468 | fprintf(f, "%sSupplementaryGroups:", prefix); |
15ae422b LP |
3469 | strv_fprintf(f, c->supplementary_groups); |
3470 | fputs("\n", f); | |
3471 | } | |
94f04347 | 3472 | |
5b6319dc | 3473 | if (c->pam_name) |
f2d3769a | 3474 | fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name); |
5b6319dc | 3475 | |
2a624c36 AP |
3476 | if (strv_length(c->read_write_paths) > 0) { |
3477 | fprintf(f, "%sReadWritePaths:", prefix); | |
3478 | strv_fprintf(f, c->read_write_paths); | |
15ae422b LP |
3479 | fputs("\n", f); |
3480 | } | |
3481 | ||
2a624c36 AP |
3482 | if (strv_length(c->read_only_paths) > 0) { |
3483 | fprintf(f, "%sReadOnlyPaths:", prefix); | |
3484 | strv_fprintf(f, c->read_only_paths); | |
15ae422b LP |
3485 | fputs("\n", f); |
3486 | } | |
94f04347 | 3487 | |
2a624c36 AP |
3488 | if (strv_length(c->inaccessible_paths) > 0) { |
3489 | fprintf(f, "%sInaccessiblePaths:", prefix); | |
3490 | strv_fprintf(f, c->inaccessible_paths); | |
94f04347 LP |
3491 | fputs("\n", f); |
3492 | } | |
2e22afe9 | 3493 | |
169c1bda LP |
3494 | if (c->utmp_id) |
3495 | fprintf(f, | |
3496 | "%sUtmpIdentifier: %s\n", | |
3497 | prefix, c->utmp_id); | |
7b52a628 MS |
3498 | |
3499 | if (c->selinux_context) | |
3500 | fprintf(f, | |
5f8640fb LP |
3501 | "%sSELinuxContext: %s%s\n", |
3502 | prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context); | |
17df7223 | 3503 | |
050f7277 | 3504 | if (c->personality != PERSONALITY_INVALID) |
ac45f971 LP |
3505 | fprintf(f, |
3506 | "%sPersonality: %s\n", | |
3507 | prefix, strna(personality_to_string(c->personality))); | |
3508 | ||
17df7223 | 3509 | if (c->syscall_filter) { |
351a19b1 | 3510 | #ifdef HAVE_SECCOMP |
17df7223 LP |
3511 | Iterator j; |
3512 | void *id; | |
3513 | bool first = true; | |
351a19b1 | 3514 | #endif |
17df7223 LP |
3515 | |
3516 | fprintf(f, | |
57183d11 | 3517 | "%sSystemCallFilter: ", |
17df7223 LP |
3518 | prefix); |
3519 | ||
3520 | if (!c->syscall_whitelist) | |
3521 | fputc('~', f); | |
3522 | ||
351a19b1 | 3523 | #ifdef HAVE_SECCOMP |
17df7223 LP |
3524 | SET_FOREACH(id, c->syscall_filter, j) { |
3525 | _cleanup_free_ char *name = NULL; | |
3526 | ||
3527 | if (first) | |
3528 | first = false; | |
3529 | else | |
3530 | fputc(' ', f); | |
3531 | ||
57183d11 | 3532 | name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1); |
17df7223 LP |
3533 | fputs(strna(name), f); |
3534 | } | |
351a19b1 | 3535 | #endif |
17df7223 LP |
3536 | |
3537 | fputc('\n', f); | |
3538 | } | |
3539 | ||
57183d11 LP |
3540 | if (c->syscall_archs) { |
3541 | #ifdef HAVE_SECCOMP | |
3542 | Iterator j; | |
3543 | void *id; | |
3544 | #endif | |
3545 | ||
3546 | fprintf(f, | |
3547 | "%sSystemCallArchitectures:", | |
3548 | prefix); | |
3549 | ||
3550 | #ifdef HAVE_SECCOMP | |
3551 | SET_FOREACH(id, c->syscall_archs, j) | |
3552 | fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1))); | |
3553 | #endif | |
3554 | fputc('\n', f); | |
3555 | } | |
3556 | ||
add00535 LP |
3557 | if (exec_context_restrict_namespaces_set(c)) { |
3558 | _cleanup_free_ char *s = NULL; | |
3559 | ||
3560 | r = namespace_flag_to_string_many(c->restrict_namespaces, &s); | |
3561 | if (r >= 0) | |
3562 | fprintf(f, "%sRestrictNamespaces: %s\n", | |
3563 | prefix, s); | |
3564 | } | |
3565 | ||
b3267152 | 3566 | if (c->syscall_errno > 0) |
17df7223 LP |
3567 | fprintf(f, |
3568 | "%sSystemCallErrorNumber: %s\n", | |
3569 | prefix, strna(errno_to_name(c->syscall_errno))); | |
eef65bf3 MS |
3570 | |
3571 | if (c->apparmor_profile) | |
3572 | fprintf(f, | |
3573 | "%sAppArmorProfile: %s%s\n", | |
3574 | prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile); | |
5cb5a6ff LP |
3575 | } |
3576 | ||
a931ad47 LP |
3577 | bool exec_context_maintains_privileges(ExecContext *c) { |
3578 | assert(c); | |
3579 | ||
61233823 | 3580 | /* Returns true if the process forked off would run under |
a931ad47 LP |
3581 | * an unchanged UID or as root. */ |
3582 | ||
3583 | if (!c->user) | |
3584 | return true; | |
3585 | ||
3586 | if (streq(c->user, "root") || streq(c->user, "0")) | |
3587 | return true; | |
3588 | ||
3589 | return false; | |
3590 | } | |
3591 | ||
b58b4116 | 3592 | void exec_status_start(ExecStatus *s, pid_t pid) { |
034c6ed7 | 3593 | assert(s); |
5cb5a6ff | 3594 | |
b58b4116 LP |
3595 | zero(*s); |
3596 | s->pid = pid; | |
3597 | dual_timestamp_get(&s->start_timestamp); | |
3598 | } | |
3599 | ||
6ea832a2 | 3600 | void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) { |
b58b4116 LP |
3601 | assert(s); |
3602 | ||
0b1f4ae6 | 3603 | if (s->pid && s->pid != pid) |
b58b4116 LP |
3604 | zero(*s); |
3605 | ||
034c6ed7 | 3606 | s->pid = pid; |
63983207 | 3607 | dual_timestamp_get(&s->exit_timestamp); |
9fb86720 | 3608 | |
034c6ed7 LP |
3609 | s->code = code; |
3610 | s->status = status; | |
169c1bda | 3611 | |
6ea832a2 LP |
3612 | if (context) { |
3613 | if (context->utmp_id) | |
3614 | utmp_put_dead_process(context->utmp_id, pid, code, status); | |
3615 | ||
1e22b5cd | 3616 | exec_context_tty_reset(context, NULL); |
6ea832a2 | 3617 | } |
9fb86720 LP |
3618 | } |
3619 | ||
3620 | void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) { | |
3621 | char buf[FORMAT_TIMESTAMP_MAX]; | |
3622 | ||
3623 | assert(s); | |
3624 | assert(f); | |
3625 | ||
9fb86720 LP |
3626 | if (s->pid <= 0) |
3627 | return; | |
3628 | ||
4c940960 LP |
3629 | prefix = strempty(prefix); |
3630 | ||
9fb86720 | 3631 | fprintf(f, |
ccd06097 ZJS |
3632 | "%sPID: "PID_FMT"\n", |
3633 | prefix, s->pid); | |
9fb86720 | 3634 | |
af9d16e1 | 3635 | if (dual_timestamp_is_set(&s->start_timestamp)) |
9fb86720 LP |
3636 | fprintf(f, |
3637 | "%sStart Timestamp: %s\n", | |
63983207 | 3638 | prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime)); |
9fb86720 | 3639 | |
af9d16e1 | 3640 | if (dual_timestamp_is_set(&s->exit_timestamp)) |
9fb86720 LP |
3641 | fprintf(f, |
3642 | "%sExit Timestamp: %s\n" | |
3643 | "%sExit Code: %s\n" | |
3644 | "%sExit Status: %i\n", | |
63983207 | 3645 | prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime), |
9fb86720 LP |
3646 | prefix, sigchld_code_to_string(s->code), |
3647 | prefix, s->status); | |
5cb5a6ff | 3648 | } |
44d8db9e | 3649 | |
9e2f7c11 | 3650 | char *exec_command_line(char **argv) { |
44d8db9e LP |
3651 | size_t k; |
3652 | char *n, *p, **a; | |
3653 | bool first = true; | |
3654 | ||
9e2f7c11 | 3655 | assert(argv); |
44d8db9e | 3656 | |
9164977d | 3657 | k = 1; |
9e2f7c11 | 3658 | STRV_FOREACH(a, argv) |
44d8db9e LP |
3659 | k += strlen(*a)+3; |
3660 | ||
5cd9cd35 LP |
3661 | n = new(char, k); |
3662 | if (!n) | |
44d8db9e LP |
3663 | return NULL; |
3664 | ||
3665 | p = n; | |
9e2f7c11 | 3666 | STRV_FOREACH(a, argv) { |
44d8db9e LP |
3667 | |
3668 | if (!first) | |
3669 | *(p++) = ' '; | |
3670 | else | |
3671 | first = false; | |
3672 | ||
3673 | if (strpbrk(*a, WHITESPACE)) { | |
3674 | *(p++) = '\''; | |
3675 | p = stpcpy(p, *a); | |
3676 | *(p++) = '\''; | |
3677 | } else | |
3678 | p = stpcpy(p, *a); | |
3679 | ||
3680 | } | |
3681 | ||
9164977d LP |
3682 | *p = 0; |
3683 | ||
44d8db9e LP |
3684 | /* FIXME: this doesn't really handle arguments that have |
3685 | * spaces and ticks in them */ | |
3686 | ||
3687 | return n; | |
3688 | } | |
3689 | ||
3690 | void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) { | |
e1d75803 | 3691 | _cleanup_free_ char *cmd = NULL; |
4c940960 | 3692 | const char *prefix2; |
44d8db9e LP |
3693 | |
3694 | assert(c); | |
3695 | assert(f); | |
3696 | ||
4c940960 | 3697 | prefix = strempty(prefix); |
63c372cb | 3698 | prefix2 = strjoina(prefix, "\t"); |
44d8db9e | 3699 | |
9e2f7c11 | 3700 | cmd = exec_command_line(c->argv); |
44d8db9e LP |
3701 | fprintf(f, |
3702 | "%sCommand Line: %s\n", | |
3703 | prefix, cmd ? cmd : strerror(ENOMEM)); | |
3704 | ||
9fb86720 | 3705 | exec_status_dump(&c->exec_status, f, prefix2); |
44d8db9e LP |
3706 | } |
3707 | ||
3708 | void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) { | |
3709 | assert(f); | |
3710 | ||
4c940960 | 3711 | prefix = strempty(prefix); |
44d8db9e LP |
3712 | |
3713 | LIST_FOREACH(command, c, c) | |
3714 | exec_command_dump(c, f, prefix); | |
3715 | } | |
94f04347 | 3716 | |
a6a80b4f LP |
3717 | void exec_command_append_list(ExecCommand **l, ExecCommand *e) { |
3718 | ExecCommand *end; | |
3719 | ||
3720 | assert(l); | |
3721 | assert(e); | |
3722 | ||
3723 | if (*l) { | |
35b8ca3a | 3724 | /* It's kind of important, that we keep the order here */ |
71fda00f LP |
3725 | LIST_FIND_TAIL(command, *l, end); |
3726 | LIST_INSERT_AFTER(command, *l, end, e); | |
a6a80b4f LP |
3727 | } else |
3728 | *l = e; | |
3729 | } | |
3730 | ||
26fd040d LP |
3731 | int exec_command_set(ExecCommand *c, const char *path, ...) { |
3732 | va_list ap; | |
3733 | char **l, *p; | |
3734 | ||
3735 | assert(c); | |
3736 | assert(path); | |
3737 | ||
3738 | va_start(ap, path); | |
3739 | l = strv_new_ap(path, ap); | |
3740 | va_end(ap); | |
3741 | ||
3742 | if (!l) | |
3743 | return -ENOMEM; | |
3744 | ||
250a918d LP |
3745 | p = strdup(path); |
3746 | if (!p) { | |
26fd040d LP |
3747 | strv_free(l); |
3748 | return -ENOMEM; | |
3749 | } | |
3750 | ||
3751 | free(c->path); | |
3752 | c->path = p; | |
3753 | ||
3754 | strv_free(c->argv); | |
3755 | c->argv = l; | |
3756 | ||
3757 | return 0; | |
3758 | } | |
3759 | ||
86b23b07 | 3760 | int exec_command_append(ExecCommand *c, const char *path, ...) { |
e63ff941 | 3761 | _cleanup_strv_free_ char **l = NULL; |
86b23b07 | 3762 | va_list ap; |
86b23b07 JS |
3763 | int r; |
3764 | ||
3765 | assert(c); | |
3766 | assert(path); | |
3767 | ||
3768 | va_start(ap, path); | |
3769 | l = strv_new_ap(path, ap); | |
3770 | va_end(ap); | |
3771 | ||
3772 | if (!l) | |
3773 | return -ENOMEM; | |
3774 | ||
e287086b | 3775 | r = strv_extend_strv(&c->argv, l, false); |
e63ff941 | 3776 | if (r < 0) |
86b23b07 | 3777 | return r; |
86b23b07 JS |
3778 | |
3779 | return 0; | |
3780 | } | |
3781 | ||
3782 | ||
613b411c LP |
3783 | static int exec_runtime_allocate(ExecRuntime **rt) { |
3784 | ||
3785 | if (*rt) | |
3786 | return 0; | |
3787 | ||
3788 | *rt = new0(ExecRuntime, 1); | |
f146f5e1 | 3789 | if (!*rt) |
613b411c LP |
3790 | return -ENOMEM; |
3791 | ||
3792 | (*rt)->n_ref = 1; | |
3793 | (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1; | |
3794 | ||
3795 | return 0; | |
3796 | } | |
3797 | ||
3798 | int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) { | |
3799 | int r; | |
3800 | ||
3801 | assert(rt); | |
3802 | assert(c); | |
3803 | assert(id); | |
3804 | ||
3805 | if (*rt) | |
3806 | return 1; | |
3807 | ||
3808 | if (!c->private_network && !c->private_tmp) | |
3809 | return 0; | |
3810 | ||
3811 | r = exec_runtime_allocate(rt); | |
3812 | if (r < 0) | |
3813 | return r; | |
3814 | ||
3815 | if (c->private_network && (*rt)->netns_storage_socket[0] < 0) { | |
33df919d | 3816 | if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, (*rt)->netns_storage_socket) < 0) |
613b411c LP |
3817 | return -errno; |
3818 | } | |
3819 | ||
3820 | if (c->private_tmp && !(*rt)->tmp_dir) { | |
3821 | r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir); | |
3822 | if (r < 0) | |
3823 | return r; | |
3824 | } | |
3825 | ||
3826 | return 1; | |
3827 | } | |
3828 | ||
3829 | ExecRuntime *exec_runtime_ref(ExecRuntime *r) { | |
3830 | assert(r); | |
3831 | assert(r->n_ref > 0); | |
3832 | ||
3833 | r->n_ref++; | |
3834 | return r; | |
3835 | } | |
3836 | ||
3837 | ExecRuntime *exec_runtime_unref(ExecRuntime *r) { | |
3838 | ||
3839 | if (!r) | |
3840 | return NULL; | |
3841 | ||
3842 | assert(r->n_ref > 0); | |
3843 | ||
3844 | r->n_ref--; | |
f2341e0a LP |
3845 | if (r->n_ref > 0) |
3846 | return NULL; | |
3847 | ||
3848 | free(r->tmp_dir); | |
3849 | free(r->var_tmp_dir); | |
3850 | safe_close_pair(r->netns_storage_socket); | |
6b430fdb | 3851 | return mfree(r); |
613b411c LP |
3852 | } |
3853 | ||
f2341e0a | 3854 | int exec_runtime_serialize(Unit *u, ExecRuntime *rt, FILE *f, FDSet *fds) { |
613b411c LP |
3855 | assert(u); |
3856 | assert(f); | |
3857 | assert(fds); | |
3858 | ||
3859 | if (!rt) | |
3860 | return 0; | |
3861 | ||
3862 | if (rt->tmp_dir) | |
3863 | unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir); | |
3864 | ||
3865 | if (rt->var_tmp_dir) | |
3866 | unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir); | |
3867 | ||
3868 | if (rt->netns_storage_socket[0] >= 0) { | |
3869 | int copy; | |
3870 | ||
3871 | copy = fdset_put_dup(fds, rt->netns_storage_socket[0]); | |
3872 | if (copy < 0) | |
3873 | return copy; | |
3874 | ||
3875 | unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy); | |
3876 | } | |
3877 | ||
3878 | if (rt->netns_storage_socket[1] >= 0) { | |
3879 | int copy; | |
3880 | ||
3881 | copy = fdset_put_dup(fds, rt->netns_storage_socket[1]); | |
3882 | if (copy < 0) | |
3883 | return copy; | |
3884 | ||
3885 | unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy); | |
3886 | } | |
3887 | ||
3888 | return 0; | |
3889 | } | |
3890 | ||
f2341e0a | 3891 | int exec_runtime_deserialize_item(Unit *u, ExecRuntime **rt, const char *key, const char *value, FDSet *fds) { |
613b411c LP |
3892 | int r; |
3893 | ||
3894 | assert(rt); | |
3895 | assert(key); | |
3896 | assert(value); | |
3897 | ||
3898 | if (streq(key, "tmp-dir")) { | |
3899 | char *copy; | |
3900 | ||
3901 | r = exec_runtime_allocate(rt); | |
3902 | if (r < 0) | |
f2341e0a | 3903 | return log_oom(); |
613b411c LP |
3904 | |
3905 | copy = strdup(value); | |
3906 | if (!copy) | |
3907 | return log_oom(); | |
3908 | ||
3909 | free((*rt)->tmp_dir); | |
3910 | (*rt)->tmp_dir = copy; | |
3911 | ||
3912 | } else if (streq(key, "var-tmp-dir")) { | |
3913 | char *copy; | |
3914 | ||
3915 | r = exec_runtime_allocate(rt); | |
3916 | if (r < 0) | |
f2341e0a | 3917 | return log_oom(); |
613b411c LP |
3918 | |
3919 | copy = strdup(value); | |
3920 | if (!copy) | |
3921 | return log_oom(); | |
3922 | ||
3923 | free((*rt)->var_tmp_dir); | |
3924 | (*rt)->var_tmp_dir = copy; | |
3925 | ||
3926 | } else if (streq(key, "netns-socket-0")) { | |
3927 | int fd; | |
3928 | ||
3929 | r = exec_runtime_allocate(rt); | |
3930 | if (r < 0) | |
f2341e0a | 3931 | return log_oom(); |
613b411c LP |
3932 | |
3933 | if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd)) | |
f2341e0a | 3934 | log_unit_debug(u, "Failed to parse netns socket value: %s", value); |
613b411c | 3935 | else { |
03e334a1 | 3936 | safe_close((*rt)->netns_storage_socket[0]); |
613b411c LP |
3937 | (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd); |
3938 | } | |
3939 | } else if (streq(key, "netns-socket-1")) { | |
3940 | int fd; | |
3941 | ||
3942 | r = exec_runtime_allocate(rt); | |
3943 | if (r < 0) | |
f2341e0a | 3944 | return log_oom(); |
613b411c LP |
3945 | |
3946 | if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd)) | |
f2341e0a | 3947 | log_unit_debug(u, "Failed to parse netns socket value: %s", value); |
613b411c | 3948 | else { |
03e334a1 | 3949 | safe_close((*rt)->netns_storage_socket[1]); |
613b411c LP |
3950 | (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd); |
3951 | } | |
3952 | } else | |
3953 | return 0; | |
3954 | ||
3955 | return 1; | |
3956 | } | |
3957 | ||
3958 | static void *remove_tmpdir_thread(void *p) { | |
3959 | _cleanup_free_ char *path = p; | |
3960 | ||
c6878637 | 3961 | (void) rm_rf(path, REMOVE_ROOT|REMOVE_PHYSICAL); |
613b411c LP |
3962 | return NULL; |
3963 | } | |
3964 | ||
3965 | void exec_runtime_destroy(ExecRuntime *rt) { | |
98b47d54 LP |
3966 | int r; |
3967 | ||
613b411c LP |
3968 | if (!rt) |
3969 | return; | |
3970 | ||
3971 | /* If there are multiple users of this, let's leave the stuff around */ | |
3972 | if (rt->n_ref > 1) | |
3973 | return; | |
3974 | ||
3975 | if (rt->tmp_dir) { | |
3976 | log_debug("Spawning thread to nuke %s", rt->tmp_dir); | |
98b47d54 LP |
3977 | |
3978 | r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir); | |
3979 | if (r < 0) { | |
da927ba9 | 3980 | log_warning_errno(r, "Failed to nuke %s: %m", rt->tmp_dir); |
98b47d54 LP |
3981 | free(rt->tmp_dir); |
3982 | } | |
3983 | ||
613b411c LP |
3984 | rt->tmp_dir = NULL; |
3985 | } | |
3986 | ||
3987 | if (rt->var_tmp_dir) { | |
3988 | log_debug("Spawning thread to nuke %s", rt->var_tmp_dir); | |
98b47d54 LP |
3989 | |
3990 | r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir); | |
3991 | if (r < 0) { | |
da927ba9 | 3992 | log_warning_errno(r, "Failed to nuke %s: %m", rt->var_tmp_dir); |
98b47d54 LP |
3993 | free(rt->var_tmp_dir); |
3994 | } | |
3995 | ||
613b411c LP |
3996 | rt->var_tmp_dir = NULL; |
3997 | } | |
3998 | ||
3d94f76c | 3999 | safe_close_pair(rt->netns_storage_socket); |
613b411c LP |
4000 | } |
4001 | ||
80876c20 LP |
4002 | static const char* const exec_input_table[_EXEC_INPUT_MAX] = { |
4003 | [EXEC_INPUT_NULL] = "null", | |
4004 | [EXEC_INPUT_TTY] = "tty", | |
4005 | [EXEC_INPUT_TTY_FORCE] = "tty-force", | |
4f2d528d | 4006 | [EXEC_INPUT_TTY_FAIL] = "tty-fail", |
52c239d7 LB |
4007 | [EXEC_INPUT_SOCKET] = "socket", |
4008 | [EXEC_INPUT_NAMED_FD] = "fd", | |
80876c20 LP |
4009 | }; |
4010 | ||
8a0867d6 LP |
4011 | DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput); |
4012 | ||
94f04347 | 4013 | static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = { |
80876c20 | 4014 | [EXEC_OUTPUT_INHERIT] = "inherit", |
94f04347 | 4015 | [EXEC_OUTPUT_NULL] = "null", |
80876c20 | 4016 | [EXEC_OUTPUT_TTY] = "tty", |
94f04347 | 4017 | [EXEC_OUTPUT_SYSLOG] = "syslog", |
28dbc1e8 | 4018 | [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console", |
9a6bca7a | 4019 | [EXEC_OUTPUT_KMSG] = "kmsg", |
28dbc1e8 | 4020 | [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console", |
706343f4 LP |
4021 | [EXEC_OUTPUT_JOURNAL] = "journal", |
4022 | [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console", | |
52c239d7 LB |
4023 | [EXEC_OUTPUT_SOCKET] = "socket", |
4024 | [EXEC_OUTPUT_NAMED_FD] = "fd", | |
94f04347 LP |
4025 | }; |
4026 | ||
4027 | DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput); | |
023a4f67 LP |
4028 | |
4029 | static const char* const exec_utmp_mode_table[_EXEC_UTMP_MODE_MAX] = { | |
4030 | [EXEC_UTMP_INIT] = "init", | |
4031 | [EXEC_UTMP_LOGIN] = "login", | |
4032 | [EXEC_UTMP_USER] = "user", | |
4033 | }; | |
4034 | ||
4035 | DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode, ExecUtmpMode); |