]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
a7334b09 | 2 | |
034c6ed7 LP |
3 | #include <errno.h> |
4 | #include <fcntl.h> | |
8dd4c05b | 5 | #include <poll.h> |
d251207d | 6 | #include <sys/eventfd.h> |
ac8db36c | 7 | #include <sys/file.h> |
f5947a5e | 8 | #include <sys/ioctl.h> |
f3e43635 | 9 | #include <sys/mman.h> |
bb0c0d6f | 10 | #include <sys/mount.h> |
8dd4c05b | 11 | #include <sys/personality.h> |
94f04347 | 12 | #include <sys/prctl.h> |
d2ffa389 | 13 | #include <sys/shm.h> |
d2ffa389 | 14 | #include <sys/types.h> |
8dd4c05b LP |
15 | #include <sys/un.h> |
16 | #include <unistd.h> | |
023a4f67 | 17 | #include <utmpx.h> |
5cb5a6ff | 18 | |
9c0c6701 DDM |
19 | #include <linux/fs.h> /* Must be included after <sys/mount.h> */ |
20 | ||
349cc4a5 | 21 | #if HAVE_PAM |
5b6319dc LP |
22 | #include <security/pam_appl.h> |
23 | #endif | |
24 | ||
349cc4a5 | 25 | #if HAVE_SELINUX |
7b52a628 MS |
26 | #include <selinux/selinux.h> |
27 | #endif | |
28 | ||
349cc4a5 | 29 | #if HAVE_SECCOMP |
17df7223 LP |
30 | #include <seccomp.h> |
31 | #endif | |
32 | ||
349cc4a5 | 33 | #if HAVE_APPARMOR |
eef65bf3 MS |
34 | #include <sys/apparmor.h> |
35 | #endif | |
36 | ||
24882e06 | 37 | #include "sd-messages.h" |
8dd4c05b | 38 | |
bb0c0d6f | 39 | #include "acl-util.h" |
8dd4c05b | 40 | #include "af-list.h" |
b5efdb8a | 41 | #include "alloc-util.h" |
349cc4a5 | 42 | #if HAVE_APPARMOR |
3ffd4af2 LP |
43 | #include "apparmor-util.h" |
44 | #endif | |
ee617a4e | 45 | #include "argv-util.h" |
8dd4c05b LP |
46 | #include "async.h" |
47 | #include "barrier.h" | |
b1994387 | 48 | #include "bpf-lsm.h" |
9c0c6701 | 49 | #include "btrfs-util.h" |
8dd4c05b | 50 | #include "cap-list.h" |
430f0182 | 51 | #include "capability-util.h" |
9c0c6701 | 52 | #include "chattr-util.h" |
fdb3deca | 53 | #include "cgroup-setup.h" |
f461a28d | 54 | #include "chase.h" |
bb0c0d6f | 55 | #include "chown-recursive.h" |
28db6fbf | 56 | #include "constants.h" |
da681e1b | 57 | #include "cpu-set-util.h" |
43144be4 | 58 | #include "creds-util.h" |
6a818c3c | 59 | #include "data-fd-util.h" |
686d13b9 | 60 | #include "env-file.h" |
4d1a6904 | 61 | #include "env-util.h" |
17df7223 | 62 | #include "errno-list.h" |
8a62620e | 63 | #include "escape.h" |
3ffd4af2 | 64 | #include "execute.h" |
8dd4c05b | 65 | #include "exit-status.h" |
3ffd4af2 | 66 | #include "fd-util.h" |
bb0c0d6f | 67 | #include "fileio.h" |
f97b34a6 | 68 | #include "format-util.h" |
7d50b32a | 69 | #include "glob-util.h" |
0389f4fa | 70 | #include "hexdecoct.h" |
c004493c | 71 | #include "io-util.h" |
032b3afb | 72 | #include "ioprio-util.h" |
0690160e | 73 | #include "label-util.h" |
9c0c6701 | 74 | #include "lock-util.h" |
8dd4c05b LP |
75 | #include "log.h" |
76 | #include "macro.h" | |
e8a565cb | 77 | #include "manager.h" |
2a341bb9 | 78 | #include "manager-dump.h" |
0a970718 | 79 | #include "memory-util.h" |
f5947a5e | 80 | #include "missing_fs.h" |
5bead76e | 81 | #include "missing_ioprio.h" |
7a114ed4 | 82 | #include "missing_prctl.h" |
35cd0ba5 | 83 | #include "mkdir-label.h" |
21935150 | 84 | #include "mount-util.h" |
bb0c0d6f | 85 | #include "mountpoint-util.h" |
8dd4c05b | 86 | #include "namespace.h" |
6bedfcbb | 87 | #include "parse-util.h" |
8dd4c05b | 88 | #include "path-util.h" |
4d62ee55 | 89 | #include "proc-cmdline.h" |
0b452006 | 90 | #include "process-util.h" |
6bb00842 | 91 | #include "psi-util.h" |
d3dcf4e3 | 92 | #include "random-util.h" |
3989bdc1 | 93 | #include "recurse-dir.h" |
78f22b97 | 94 | #include "rlimit-util.h" |
8dd4c05b | 95 | #include "rm-rf.h" |
349cc4a5 | 96 | #if HAVE_SECCOMP |
3ffd4af2 LP |
97 | #include "seccomp-util.h" |
98 | #endif | |
07d46372 | 99 | #include "securebits-util.h" |
8dd4c05b | 100 | #include "selinux-util.h" |
24882e06 | 101 | #include "signal-util.h" |
8dd4c05b | 102 | #include "smack-util.h" |
57b7a260 | 103 | #include "socket-util.h" |
a2ab603c | 104 | #include "sort-util.h" |
fd63e712 | 105 | #include "special.h" |
949befd3 | 106 | #include "stat-util.h" |
8b43440b | 107 | #include "string-table.h" |
07630cea | 108 | #include "string-util.h" |
8dd4c05b | 109 | #include "strv.h" |
7ccbd1ae | 110 | #include "syslog-util.h" |
8dd4c05b | 111 | #include "terminal-util.h" |
bb0c0d6f | 112 | #include "tmpfile-util.h" |
566b7d23 | 113 | #include "umask-util.h" |
2d3b784d | 114 | #include "unit-serialize.h" |
b1d4f8e1 | 115 | #include "user-util.h" |
8dd4c05b | 116 | #include "utmp-wtmp.h" |
5cb5a6ff | 117 | |
e056b01d | 118 | #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC) |
31a7eb86 | 119 | #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC) |
e6a26745 | 120 | |
531dca78 LP |
121 | #define SNDBUF_SIZE (8*1024*1024) |
122 | ||
da6053d0 | 123 | static int shift_fds(int fds[], size_t n_fds) { |
034c6ed7 LP |
124 | if (n_fds <= 0) |
125 | return 0; | |
126 | ||
a0d40ac5 LP |
127 | /* Modifies the fds array! (sorts it) */ |
128 | ||
034c6ed7 LP |
129 | assert(fds); |
130 | ||
5b10116e ZJS |
131 | for (int start = 0;;) { |
132 | int restart_from = -1; | |
034c6ed7 | 133 | |
5b10116e | 134 | for (int i = start; i < (int) n_fds; i++) { |
034c6ed7 LP |
135 | int nfd; |
136 | ||
137 | /* Already at right index? */ | |
138 | if (fds[i] == i+3) | |
139 | continue; | |
140 | ||
3cc2aff1 LP |
141 | nfd = fcntl(fds[i], F_DUPFD, i + 3); |
142 | if (nfd < 0) | |
034c6ed7 LP |
143 | return -errno; |
144 | ||
03e334a1 | 145 | safe_close(fds[i]); |
034c6ed7 LP |
146 | fds[i] = nfd; |
147 | ||
148 | /* Hmm, the fd we wanted isn't free? Then | |
ee33e53a | 149 | * let's remember that and try again from here */ |
034c6ed7 LP |
150 | if (nfd != i+3 && restart_from < 0) |
151 | restart_from = i; | |
152 | } | |
153 | ||
154 | if (restart_from < 0) | |
155 | break; | |
156 | ||
157 | start = restart_from; | |
158 | } | |
159 | ||
160 | return 0; | |
161 | } | |
162 | ||
cd48e23f RP |
163 | static int flags_fds( |
164 | const int fds[], | |
165 | size_t n_socket_fds, | |
166 | size_t n_fds, | |
167 | bool nonblock) { | |
168 | ||
e2c76839 | 169 | int r; |
47a71eed LP |
170 | |
171 | if (n_fds <= 0) | |
172 | return 0; | |
173 | ||
174 | assert(fds); | |
175 | ||
9b141911 FB |
176 | /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags. |
177 | * O_NONBLOCK only applies to socket activation though. */ | |
47a71eed | 178 | |
5b10116e | 179 | for (size_t i = 0; i < n_fds; i++) { |
47a71eed | 180 | |
9b141911 FB |
181 | if (i < n_socket_fds) { |
182 | r = fd_nonblock(fds[i], nonblock); | |
183 | if (r < 0) | |
184 | return r; | |
185 | } | |
47a71eed | 186 | |
451a074f LP |
187 | /* We unconditionally drop FD_CLOEXEC from the fds, |
188 | * since after all we want to pass these fds to our | |
189 | * children */ | |
47a71eed | 190 | |
3cc2aff1 LP |
191 | r = fd_cloexec(fds[i], false); |
192 | if (r < 0) | |
e2c76839 | 193 | return r; |
47a71eed LP |
194 | } |
195 | ||
196 | return 0; | |
197 | } | |
198 | ||
1e22b5cd | 199 | static const char *exec_context_tty_path(const ExecContext *context) { |
80876c20 LP |
200 | assert(context); |
201 | ||
1e22b5cd LP |
202 | if (context->stdio_as_fds) |
203 | return NULL; | |
204 | ||
80876c20 LP |
205 | if (context->tty_path) |
206 | return context->tty_path; | |
207 | ||
208 | return "/dev/console"; | |
209 | } | |
210 | ||
4d62ee55 | 211 | static int exec_context_tty_size(const ExecContext *context, unsigned *ret_rows, unsigned *ret_cols) { |
4d62ee55 DDM |
212 | unsigned rows, cols; |
213 | const char *tty; | |
4d62ee55 DDM |
214 | |
215 | assert(context); | |
216 | assert(ret_rows); | |
217 | assert(ret_cols); | |
218 | ||
219 | rows = context->tty_rows; | |
220 | cols = context->tty_cols; | |
221 | ||
222 | tty = exec_context_tty_path(context); | |
29f5a5ae DDM |
223 | if (tty) |
224 | (void) proc_cmdline_tty_size(tty, rows == UINT_MAX ? &rows : NULL, cols == UINT_MAX ? &cols : NULL); | |
4d62ee55 DDM |
225 | |
226 | *ret_rows = rows; | |
227 | *ret_cols = cols; | |
228 | ||
229 | return 0; | |
230 | } | |
231 | ||
1e22b5cd | 232 | static void exec_context_tty_reset(const ExecContext *context, const ExecParameters *p) { |
a0043bfa ZJS |
233 | _cleanup_close_ int fd = -EBADF; |
234 | const char *path = exec_context_tty_path(ASSERT_PTR(context)); | |
1e22b5cd | 235 | |
a0043bfa ZJS |
236 | /* Take a lock around the device for the duration of the setup that we do here. |
237 | * systemd-vconsole-setup.service also takes the lock to avoid being interrupted. | |
238 | * We open a new fd that will be closed automatically, and operate on it for convenience. | |
239 | */ | |
6ea832a2 | 240 | |
a0043bfa ZJS |
241 | if (p && p->stdin_fd >= 0) { |
242 | fd = xopenat_lock(p->stdin_fd, NULL, | |
243 | O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY, 0, 0, LOCK_BSD, LOCK_EX); | |
244 | if (fd < 0) | |
245 | return; | |
246 | } else if (path) { | |
247 | fd = open_terminal(path, O_RDWR|O_NOCTTY|O_CLOEXEC|O_NONBLOCK); | |
248 | if (fd < 0) | |
249 | return; | |
6ea832a2 | 250 | |
a0043bfa ZJS |
251 | if (lock_generic(fd, LOCK_BSD, LOCK_EX) < 0) |
252 | return; | |
253 | } else | |
254 | return; /* nothing to do */ | |
6ea832a2 | 255 | |
a0043bfa ZJS |
256 | if (context->tty_vhangup) |
257 | (void) terminal_vhangup_fd(fd); | |
258 | ||
259 | if (context->tty_reset) | |
260 | (void) reset_terminal_fd(fd, true); | |
1e22b5cd | 261 | |
4d62ee55 DDM |
262 | if (p && p->stdin_fd >= 0) { |
263 | unsigned rows = context->tty_rows, cols = context->tty_cols; | |
264 | ||
265 | (void) exec_context_tty_size(context, &rows, &cols); | |
266 | (void) terminal_set_size_fd(p->stdin_fd, path, rows, cols); | |
267 | } | |
51462135 | 268 | |
1e22b5cd LP |
269 | if (context->tty_vt_disallocate && path) |
270 | (void) vt_disallocate(path); | |
6ea832a2 LP |
271 | } |
272 | ||
6af760f3 LP |
273 | static bool is_terminal_input(ExecInput i) { |
274 | return IN_SET(i, | |
275 | EXEC_INPUT_TTY, | |
276 | EXEC_INPUT_TTY_FORCE, | |
277 | EXEC_INPUT_TTY_FAIL); | |
278 | } | |
279 | ||
3a1286b6 | 280 | static bool is_terminal_output(ExecOutput o) { |
6af760f3 LP |
281 | return IN_SET(o, |
282 | EXEC_OUTPUT_TTY, | |
6af760f3 LP |
283 | EXEC_OUTPUT_KMSG_AND_CONSOLE, |
284 | EXEC_OUTPUT_JOURNAL_AND_CONSOLE); | |
285 | } | |
286 | ||
aac8c0c3 LP |
287 | static bool is_kmsg_output(ExecOutput o) { |
288 | return IN_SET(o, | |
289 | EXEC_OUTPUT_KMSG, | |
290 | EXEC_OUTPUT_KMSG_AND_CONSOLE); | |
291 | } | |
292 | ||
6af760f3 LP |
293 | static bool exec_context_needs_term(const ExecContext *c) { |
294 | assert(c); | |
295 | ||
296 | /* Return true if the execution context suggests we should set $TERM to something useful. */ | |
297 | ||
298 | if (is_terminal_input(c->std_input)) | |
299 | return true; | |
300 | ||
301 | if (is_terminal_output(c->std_output)) | |
302 | return true; | |
303 | ||
304 | if (is_terminal_output(c->std_error)) | |
305 | return true; | |
306 | ||
307 | return !!c->tty_path; | |
3a1286b6 MS |
308 | } |
309 | ||
80876c20 | 310 | static int open_null_as(int flags, int nfd) { |
046a82c1 | 311 | int fd; |
071830ff | 312 | |
80876c20 | 313 | assert(nfd >= 0); |
071830ff | 314 | |
613b411c LP |
315 | fd = open("/dev/null", flags|O_NOCTTY); |
316 | if (fd < 0) | |
071830ff LP |
317 | return -errno; |
318 | ||
046a82c1 | 319 | return move_fd(fd, nfd, false); |
071830ff LP |
320 | } |
321 | ||
91dd5f7c LP |
322 | static int connect_journal_socket( |
323 | int fd, | |
324 | const char *log_namespace, | |
325 | uid_t uid, | |
326 | gid_t gid) { | |
327 | ||
524daa8c ZJS |
328 | uid_t olduid = UID_INVALID; |
329 | gid_t oldgid = GID_INVALID; | |
91dd5f7c | 330 | const char *j; |
524daa8c ZJS |
331 | int r; |
332 | ||
91dd5f7c LP |
333 | j = log_namespace ? |
334 | strjoina("/run/systemd/journal.", log_namespace, "/stdout") : | |
335 | "/run/systemd/journal/stdout"; | |
91dd5f7c | 336 | |
cad93f29 | 337 | if (gid_is_valid(gid)) { |
524daa8c ZJS |
338 | oldgid = getgid(); |
339 | ||
92a17af9 | 340 | if (setegid(gid) < 0) |
524daa8c ZJS |
341 | return -errno; |
342 | } | |
343 | ||
cad93f29 | 344 | if (uid_is_valid(uid)) { |
524daa8c ZJS |
345 | olduid = getuid(); |
346 | ||
92a17af9 | 347 | if (seteuid(uid) < 0) { |
524daa8c ZJS |
348 | r = -errno; |
349 | goto restore_gid; | |
350 | } | |
351 | } | |
352 | ||
1861986a | 353 | r = connect_unix_path(fd, AT_FDCWD, j); |
524daa8c | 354 | |
1861986a LP |
355 | /* If we fail to restore the uid or gid, things will likely fail later on. This should only happen if |
356 | an LSM interferes. */ | |
524daa8c | 357 | |
cad93f29 | 358 | if (uid_is_valid(uid)) |
524daa8c ZJS |
359 | (void) seteuid(olduid); |
360 | ||
361 | restore_gid: | |
cad93f29 | 362 | if (gid_is_valid(gid)) |
524daa8c ZJS |
363 | (void) setegid(oldgid); |
364 | ||
365 | return r; | |
366 | } | |
367 | ||
fd1f9c89 | 368 | static int connect_logger_as( |
34cf6c43 | 369 | const Unit *unit, |
fd1f9c89 | 370 | const ExecContext *context, |
af635cf3 | 371 | const ExecParameters *params, |
fd1f9c89 LP |
372 | ExecOutput output, |
373 | const char *ident, | |
fd1f9c89 LP |
374 | int nfd, |
375 | uid_t uid, | |
376 | gid_t gid) { | |
377 | ||
254d1313 | 378 | _cleanup_close_ int fd = -EBADF; |
2ac1ff68 | 379 | int r; |
071830ff LP |
380 | |
381 | assert(context); | |
af635cf3 | 382 | assert(params); |
80876c20 LP |
383 | assert(output < _EXEC_OUTPUT_MAX); |
384 | assert(ident); | |
385 | assert(nfd >= 0); | |
071830ff | 386 | |
54fe0cdb LP |
387 | fd = socket(AF_UNIX, SOCK_STREAM, 0); |
388 | if (fd < 0) | |
80876c20 | 389 | return -errno; |
071830ff | 390 | |
91dd5f7c | 391 | r = connect_journal_socket(fd, context->log_namespace, uid, gid); |
524daa8c ZJS |
392 | if (r < 0) |
393 | return r; | |
071830ff | 394 | |
2ac1ff68 | 395 | if (shutdown(fd, SHUT_RD) < 0) |
80876c20 | 396 | return -errno; |
071830ff | 397 | |
fd1f9c89 | 398 | (void) fd_inc_sndbuf(fd, SNDBUF_SIZE); |
531dca78 | 399 | |
2ac1ff68 | 400 | if (dprintf(fd, |
62bca2c6 | 401 | "%s\n" |
80876c20 LP |
402 | "%s\n" |
403 | "%i\n" | |
54fe0cdb LP |
404 | "%i\n" |
405 | "%i\n" | |
406 | "%i\n" | |
4f4a1dbf | 407 | "%i\n", |
c867611e | 408 | context->syslog_identifier ?: ident, |
af635cf3 | 409 | params->flags & EXEC_PASS_LOG_UNIT ? unit->id : "", |
54fe0cdb LP |
410 | context->syslog_priority, |
411 | !!context->syslog_level_prefix, | |
f3dc6af2 | 412 | false, |
aac8c0c3 | 413 | is_kmsg_output(output), |
2ac1ff68 EV |
414 | is_terminal_output(output)) < 0) |
415 | return -errno; | |
80876c20 | 416 | |
2ac1ff68 | 417 | return move_fd(TAKE_FD(fd), nfd, false); |
80876c20 | 418 | } |
2ac1ff68 | 419 | |
3a274a21 | 420 | static int open_terminal_as(const char *path, int flags, int nfd) { |
046a82c1 | 421 | int fd; |
071830ff | 422 | |
80876c20 LP |
423 | assert(path); |
424 | assert(nfd >= 0); | |
fd1f9c89 | 425 | |
3a274a21 | 426 | fd = open_terminal(path, flags | O_NOCTTY); |
3cc2aff1 | 427 | if (fd < 0) |
80876c20 | 428 | return fd; |
071830ff | 429 | |
046a82c1 | 430 | return move_fd(fd, nfd, false); |
80876c20 | 431 | } |
071830ff | 432 | |
2038c3f5 | 433 | static int acquire_path(const char *path, int flags, mode_t mode) { |
254d1313 | 434 | _cleanup_close_ int fd = -EBADF; |
86fca584 | 435 | int r; |
071830ff | 436 | |
80876c20 | 437 | assert(path); |
071830ff | 438 | |
2038c3f5 LP |
439 | if (IN_SET(flags & O_ACCMODE, O_WRONLY, O_RDWR)) |
440 | flags |= O_CREAT; | |
441 | ||
442 | fd = open(path, flags|O_NOCTTY, mode); | |
443 | if (fd >= 0) | |
15a3e96f | 444 | return TAKE_FD(fd); |
071830ff | 445 | |
2038c3f5 LP |
446 | if (errno != ENXIO) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */ |
447 | return -errno; | |
2038c3f5 LP |
448 | |
449 | /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */ | |
450 | ||
451 | fd = socket(AF_UNIX, SOCK_STREAM, 0); | |
452 | if (fd < 0) | |
453 | return -errno; | |
454 | ||
1861986a LP |
455 | r = connect_unix_path(fd, AT_FDCWD, path); |
456 | if (IN_SET(r, -ENOTSOCK, -EINVAL)) | |
457 | /* Propagate initial error if we get ENOTSOCK or EINVAL, i.e. we have indication that this | |
458 | * wasn't an AF_UNIX socket after all */ | |
459 | return -ENXIO; | |
460 | if (r < 0) | |
461 | return r; | |
071830ff | 462 | |
2038c3f5 LP |
463 | if ((flags & O_ACCMODE) == O_RDONLY) |
464 | r = shutdown(fd, SHUT_WR); | |
465 | else if ((flags & O_ACCMODE) == O_WRONLY) | |
466 | r = shutdown(fd, SHUT_RD); | |
467 | else | |
86fca584 | 468 | r = 0; |
15a3e96f | 469 | if (r < 0) |
2038c3f5 | 470 | return -errno; |
2038c3f5 | 471 | |
15a3e96f | 472 | return TAKE_FD(fd); |
80876c20 | 473 | } |
071830ff | 474 | |
08f3be7a LP |
475 | static int fixup_input( |
476 | const ExecContext *context, | |
477 | int socket_fd, | |
478 | bool apply_tty_stdin) { | |
479 | ||
480 | ExecInput std_input; | |
481 | ||
482 | assert(context); | |
483 | ||
484 | std_input = context->std_input; | |
1e3ad081 LP |
485 | |
486 | if (is_terminal_input(std_input) && !apply_tty_stdin) | |
487 | return EXEC_INPUT_NULL; | |
071830ff | 488 | |
03fd9c49 | 489 | if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0) |
4f2d528d LP |
490 | return EXEC_INPUT_NULL; |
491 | ||
08f3be7a LP |
492 | if (std_input == EXEC_INPUT_DATA && context->stdin_data_size == 0) |
493 | return EXEC_INPUT_NULL; | |
494 | ||
03fd9c49 | 495 | return std_input; |
4f2d528d LP |
496 | } |
497 | ||
7966a916 | 498 | static int fixup_output(ExecOutput output, int socket_fd) { |
4f2d528d | 499 | |
7966a916 | 500 | if (output == EXEC_OUTPUT_SOCKET && socket_fd < 0) |
4f2d528d LP |
501 | return EXEC_OUTPUT_INHERIT; |
502 | ||
7966a916 | 503 | return output; |
4f2d528d LP |
504 | } |
505 | ||
a34ceba6 LP |
506 | static int setup_input( |
507 | const ExecContext *context, | |
508 | const ExecParameters *params, | |
52c239d7 | 509 | int socket_fd, |
2caa38e9 | 510 | const int named_iofds[static 3]) { |
a34ceba6 | 511 | |
4f2d528d | 512 | ExecInput i; |
51462135 | 513 | int r; |
4f2d528d LP |
514 | |
515 | assert(context); | |
a34ceba6 | 516 | assert(params); |
2caa38e9 | 517 | assert(named_iofds); |
a34ceba6 LP |
518 | |
519 | if (params->stdin_fd >= 0) { | |
520 | if (dup2(params->stdin_fd, STDIN_FILENO) < 0) | |
521 | return -errno; | |
522 | ||
523 | /* Try to make this the controlling tty, if it is a tty, and reset it */ | |
1fb0682e | 524 | if (isatty(STDIN_FILENO)) { |
4d62ee55 DDM |
525 | unsigned rows = context->tty_rows, cols = context->tty_cols; |
526 | ||
527 | (void) exec_context_tty_size(context, &rows, &cols); | |
1fb0682e LP |
528 | (void) ioctl(STDIN_FILENO, TIOCSCTTY, context->std_input == EXEC_INPUT_TTY_FORCE); |
529 | (void) reset_terminal_fd(STDIN_FILENO, true); | |
4d62ee55 | 530 | (void) terminal_set_size_fd(STDIN_FILENO, NULL, rows, cols); |
1fb0682e | 531 | } |
a34ceba6 LP |
532 | |
533 | return STDIN_FILENO; | |
534 | } | |
4f2d528d | 535 | |
08f3be7a | 536 | i = fixup_input(context, socket_fd, params->flags & EXEC_APPLY_TTY_STDIN); |
4f2d528d LP |
537 | |
538 | switch (i) { | |
071830ff | 539 | |
80876c20 LP |
540 | case EXEC_INPUT_NULL: |
541 | return open_null_as(O_RDONLY, STDIN_FILENO); | |
542 | ||
543 | case EXEC_INPUT_TTY: | |
544 | case EXEC_INPUT_TTY_FORCE: | |
545 | case EXEC_INPUT_TTY_FAIL: { | |
4d62ee55 | 546 | unsigned rows, cols; |
046a82c1 | 547 | int fd; |
071830ff | 548 | |
1e22b5cd | 549 | fd = acquire_terminal(exec_context_tty_path(context), |
8854d795 LP |
550 | i == EXEC_INPUT_TTY_FAIL ? ACQUIRE_TERMINAL_TRY : |
551 | i == EXEC_INPUT_TTY_FORCE ? ACQUIRE_TERMINAL_FORCE : | |
552 | ACQUIRE_TERMINAL_WAIT, | |
3a43da28 | 553 | USEC_INFINITY); |
970edce6 | 554 | if (fd < 0) |
80876c20 LP |
555 | return fd; |
556 | ||
4d62ee55 DDM |
557 | r = exec_context_tty_size(context, &rows, &cols); |
558 | if (r < 0) | |
559 | return r; | |
560 | ||
561 | r = terminal_set_size_fd(fd, exec_context_tty_path(context), rows, cols); | |
51462135 DDM |
562 | if (r < 0) |
563 | return r; | |
564 | ||
046a82c1 | 565 | return move_fd(fd, STDIN_FILENO, false); |
80876c20 LP |
566 | } |
567 | ||
4f2d528d | 568 | case EXEC_INPUT_SOCKET: |
e75a9ed1 LP |
569 | assert(socket_fd >= 0); |
570 | ||
7c248223 | 571 | return RET_NERRNO(dup2(socket_fd, STDIN_FILENO)); |
4f2d528d | 572 | |
52c239d7 | 573 | case EXEC_INPUT_NAMED_FD: |
e75a9ed1 LP |
574 | assert(named_iofds[STDIN_FILENO] >= 0); |
575 | ||
52c239d7 | 576 | (void) fd_nonblock(named_iofds[STDIN_FILENO], false); |
7c248223 | 577 | return RET_NERRNO(dup2(named_iofds[STDIN_FILENO], STDIN_FILENO)); |
52c239d7 | 578 | |
08f3be7a LP |
579 | case EXEC_INPUT_DATA: { |
580 | int fd; | |
581 | ||
582 | fd = acquire_data_fd(context->stdin_data, context->stdin_data_size, 0); | |
583 | if (fd < 0) | |
584 | return fd; | |
585 | ||
586 | return move_fd(fd, STDIN_FILENO, false); | |
587 | } | |
588 | ||
2038c3f5 LP |
589 | case EXEC_INPUT_FILE: { |
590 | bool rw; | |
591 | int fd; | |
592 | ||
593 | assert(context->stdio_file[STDIN_FILENO]); | |
594 | ||
595 | rw = (context->std_output == EXEC_OUTPUT_FILE && streq_ptr(context->stdio_file[STDIN_FILENO], context->stdio_file[STDOUT_FILENO])) || | |
596 | (context->std_error == EXEC_OUTPUT_FILE && streq_ptr(context->stdio_file[STDIN_FILENO], context->stdio_file[STDERR_FILENO])); | |
597 | ||
598 | fd = acquire_path(context->stdio_file[STDIN_FILENO], rw ? O_RDWR : O_RDONLY, 0666 & ~context->umask); | |
599 | if (fd < 0) | |
600 | return fd; | |
601 | ||
602 | return move_fd(fd, STDIN_FILENO, false); | |
603 | } | |
604 | ||
80876c20 | 605 | default: |
04499a70 | 606 | assert_not_reached(); |
80876c20 LP |
607 | } |
608 | } | |
609 | ||
41fc585a LP |
610 | static bool can_inherit_stderr_from_stdout( |
611 | const ExecContext *context, | |
612 | ExecOutput o, | |
613 | ExecOutput e) { | |
614 | ||
615 | assert(context); | |
616 | ||
617 | /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the | |
618 | * stderr fd */ | |
619 | ||
620 | if (e == EXEC_OUTPUT_INHERIT) | |
621 | return true; | |
622 | if (e != o) | |
623 | return false; | |
624 | ||
625 | if (e == EXEC_OUTPUT_NAMED_FD) | |
626 | return streq_ptr(context->stdio_fdname[STDOUT_FILENO], context->stdio_fdname[STDERR_FILENO]); | |
627 | ||
8d7dab1f | 628 | if (IN_SET(e, EXEC_OUTPUT_FILE, EXEC_OUTPUT_FILE_APPEND, EXEC_OUTPUT_FILE_TRUNCATE)) |
41fc585a LP |
629 | return streq_ptr(context->stdio_file[STDOUT_FILENO], context->stdio_file[STDERR_FILENO]); |
630 | ||
631 | return true; | |
632 | } | |
633 | ||
a34ceba6 | 634 | static int setup_output( |
34cf6c43 | 635 | const Unit *unit, |
a34ceba6 LP |
636 | const ExecContext *context, |
637 | const ExecParameters *params, | |
638 | int fileno, | |
639 | int socket_fd, | |
2caa38e9 | 640 | const int named_iofds[static 3], |
a34ceba6 | 641 | const char *ident, |
7bce046b LP |
642 | uid_t uid, |
643 | gid_t gid, | |
644 | dev_t *journal_stream_dev, | |
645 | ino_t *journal_stream_ino) { | |
a34ceba6 | 646 | |
4f2d528d LP |
647 | ExecOutput o; |
648 | ExecInput i; | |
47c1d80d | 649 | int r; |
4f2d528d | 650 | |
f2341e0a | 651 | assert(unit); |
80876c20 | 652 | assert(context); |
a34ceba6 | 653 | assert(params); |
80876c20 | 654 | assert(ident); |
7bce046b LP |
655 | assert(journal_stream_dev); |
656 | assert(journal_stream_ino); | |
80876c20 | 657 | |
a34ceba6 LP |
658 | if (fileno == STDOUT_FILENO && params->stdout_fd >= 0) { |
659 | ||
660 | if (dup2(params->stdout_fd, STDOUT_FILENO) < 0) | |
661 | return -errno; | |
662 | ||
663 | return STDOUT_FILENO; | |
664 | } | |
665 | ||
666 | if (fileno == STDERR_FILENO && params->stderr_fd >= 0) { | |
667 | if (dup2(params->stderr_fd, STDERR_FILENO) < 0) | |
668 | return -errno; | |
669 | ||
670 | return STDERR_FILENO; | |
671 | } | |
672 | ||
08f3be7a | 673 | i = fixup_input(context, socket_fd, params->flags & EXEC_APPLY_TTY_STDIN); |
03fd9c49 | 674 | o = fixup_output(context->std_output, socket_fd); |
4f2d528d | 675 | |
eb17e935 MS |
676 | if (fileno == STDERR_FILENO) { |
677 | ExecOutput e; | |
678 | e = fixup_output(context->std_error, socket_fd); | |
80876c20 | 679 | |
eb17e935 MS |
680 | /* This expects the input and output are already set up */ |
681 | ||
682 | /* Don't change the stderr file descriptor if we inherit all | |
683 | * the way and are not on a tty */ | |
684 | if (e == EXEC_OUTPUT_INHERIT && | |
685 | o == EXEC_OUTPUT_INHERIT && | |
686 | i == EXEC_INPUT_NULL && | |
687 | !is_terminal_input(context->std_input) && | |
7966a916 | 688 | getppid() != 1) |
eb17e935 MS |
689 | return fileno; |
690 | ||
691 | /* Duplicate from stdout if possible */ | |
41fc585a | 692 | if (can_inherit_stderr_from_stdout(context, o, e)) |
7c248223 | 693 | return RET_NERRNO(dup2(STDOUT_FILENO, fileno)); |
071830ff | 694 | |
eb17e935 | 695 | o = e; |
80876c20 | 696 | |
eb17e935 | 697 | } else if (o == EXEC_OUTPUT_INHERIT) { |
21d21ea4 LP |
698 | /* If input got downgraded, inherit the original value */ |
699 | if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input)) | |
1e22b5cd | 700 | return open_terminal_as(exec_context_tty_path(context), O_WRONLY, fileno); |
21d21ea4 | 701 | |
08f3be7a LP |
702 | /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */ |
703 | if (!IN_SET(i, EXEC_INPUT_NULL, EXEC_INPUT_DATA)) | |
7c248223 | 704 | return RET_NERRNO(dup2(STDIN_FILENO, fileno)); |
071830ff | 705 | |
acb591e4 LP |
706 | /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */ |
707 | if (getppid() != 1) | |
eb17e935 | 708 | return fileno; |
94f04347 | 709 | |
eb17e935 MS |
710 | /* We need to open /dev/null here anew, to get the right access mode. */ |
711 | return open_null_as(O_WRONLY, fileno); | |
071830ff | 712 | } |
94f04347 | 713 | |
eb17e935 | 714 | switch (o) { |
80876c20 LP |
715 | |
716 | case EXEC_OUTPUT_NULL: | |
eb17e935 | 717 | return open_null_as(O_WRONLY, fileno); |
80876c20 LP |
718 | |
719 | case EXEC_OUTPUT_TTY: | |
4f2d528d | 720 | if (is_terminal_input(i)) |
7c248223 | 721 | return RET_NERRNO(dup2(STDIN_FILENO, fileno)); |
80876c20 LP |
722 | |
723 | /* We don't reset the terminal if this is just about output */ | |
1e22b5cd | 724 | return open_terminal_as(exec_context_tty_path(context), O_WRONLY, fileno); |
80876c20 | 725 | |
9a6bca7a | 726 | case EXEC_OUTPUT_KMSG: |
28dbc1e8 | 727 | case EXEC_OUTPUT_KMSG_AND_CONSOLE: |
706343f4 LP |
728 | case EXEC_OUTPUT_JOURNAL: |
729 | case EXEC_OUTPUT_JOURNAL_AND_CONSOLE: | |
af635cf3 | 730 | r = connect_logger_as(unit, context, params, o, ident, fileno, uid, gid); |
47c1d80d | 731 | if (r < 0) { |
7966a916 ZJS |
732 | log_unit_warning_errno(unit, r, "Failed to connect %s to the journal socket, ignoring: %m", |
733 | fileno == STDOUT_FILENO ? "stdout" : "stderr"); | |
eb17e935 | 734 | r = open_null_as(O_WRONLY, fileno); |
7bce046b LP |
735 | } else { |
736 | struct stat st; | |
737 | ||
738 | /* If we connected this fd to the journal via a stream, patch the device/inode into the passed | |
739 | * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits | |
ab2116b1 LP |
740 | * services to detect whether they are connected to the journal or not. |
741 | * | |
742 | * If both stdout and stderr are connected to a stream then let's make sure to store the data | |
743 | * about STDERR as that's usually the best way to do logging. */ | |
7bce046b | 744 | |
ab2116b1 LP |
745 | if (fstat(fileno, &st) >= 0 && |
746 | (*journal_stream_ino == 0 || fileno == STDERR_FILENO)) { | |
7bce046b LP |
747 | *journal_stream_dev = st.st_dev; |
748 | *journal_stream_ino = st.st_ino; | |
749 | } | |
47c1d80d MS |
750 | } |
751 | return r; | |
4f2d528d LP |
752 | |
753 | case EXEC_OUTPUT_SOCKET: | |
754 | assert(socket_fd >= 0); | |
e75a9ed1 | 755 | |
7c248223 | 756 | return RET_NERRNO(dup2(socket_fd, fileno)); |
94f04347 | 757 | |
52c239d7 | 758 | case EXEC_OUTPUT_NAMED_FD: |
e75a9ed1 LP |
759 | assert(named_iofds[fileno] >= 0); |
760 | ||
52c239d7 | 761 | (void) fd_nonblock(named_iofds[fileno], false); |
7c248223 | 762 | return RET_NERRNO(dup2(named_iofds[fileno], fileno)); |
52c239d7 | 763 | |
566b7d23 | 764 | case EXEC_OUTPUT_FILE: |
8d7dab1f LW |
765 | case EXEC_OUTPUT_FILE_APPEND: |
766 | case EXEC_OUTPUT_FILE_TRUNCATE: { | |
2038c3f5 | 767 | bool rw; |
566b7d23 | 768 | int fd, flags; |
2038c3f5 LP |
769 | |
770 | assert(context->stdio_file[fileno]); | |
771 | ||
772 | rw = context->std_input == EXEC_INPUT_FILE && | |
773 | streq_ptr(context->stdio_file[fileno], context->stdio_file[STDIN_FILENO]); | |
774 | ||
775 | if (rw) | |
7c248223 | 776 | return RET_NERRNO(dup2(STDIN_FILENO, fileno)); |
2038c3f5 | 777 | |
566b7d23 ZD |
778 | flags = O_WRONLY; |
779 | if (o == EXEC_OUTPUT_FILE_APPEND) | |
780 | flags |= O_APPEND; | |
8d7dab1f LW |
781 | else if (o == EXEC_OUTPUT_FILE_TRUNCATE) |
782 | flags |= O_TRUNC; | |
566b7d23 ZD |
783 | |
784 | fd = acquire_path(context->stdio_file[fileno], flags, 0666 & ~context->umask); | |
2038c3f5 LP |
785 | if (fd < 0) |
786 | return fd; | |
787 | ||
566b7d23 | 788 | return move_fd(fd, fileno, 0); |
2038c3f5 LP |
789 | } |
790 | ||
94f04347 | 791 | default: |
04499a70 | 792 | assert_not_reached(); |
94f04347 | 793 | } |
071830ff LP |
794 | } |
795 | ||
02a51aba | 796 | static int chown_terminal(int fd, uid_t uid) { |
4b3b5bc7 | 797 | int r; |
02a51aba LP |
798 | |
799 | assert(fd >= 0); | |
02a51aba | 800 | |
1ff74fb6 | 801 | /* Before we chown/chmod the TTY, let's ensure this is actually a tty */ |
4b3b5bc7 LP |
802 | if (isatty(fd) < 1) { |
803 | if (IN_SET(errno, EINVAL, ENOTTY)) | |
804 | return 0; /* not a tty */ | |
1ff74fb6 | 805 | |
02a51aba | 806 | return -errno; |
4b3b5bc7 | 807 | } |
02a51aba | 808 | |
4b3b5bc7 | 809 | /* This might fail. What matters are the results. */ |
f2df231f | 810 | r = fchmod_and_chown(fd, TTY_MODE, uid, GID_INVALID); |
4b3b5bc7 LP |
811 | if (r < 0) |
812 | return r; | |
02a51aba | 813 | |
4b3b5bc7 | 814 | return 1; |
02a51aba LP |
815 | } |
816 | ||
aedec452 | 817 | static int setup_confirm_stdio( |
51462135 | 818 | const ExecContext *context, |
aedec452 LP |
819 | const char *vc, |
820 | int *ret_saved_stdin, | |
821 | int *ret_saved_stdout) { | |
822 | ||
254d1313 | 823 | _cleanup_close_ int fd = -EBADF, saved_stdin = -EBADF, saved_stdout = -EBADF; |
4d62ee55 | 824 | unsigned rows, cols; |
3d18b167 | 825 | int r; |
80876c20 | 826 | |
aedec452 LP |
827 | assert(ret_saved_stdin); |
828 | assert(ret_saved_stdout); | |
80876c20 | 829 | |
af6da548 LP |
830 | saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3); |
831 | if (saved_stdin < 0) | |
832 | return -errno; | |
80876c20 | 833 | |
af6da548 | 834 | saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3); |
3d18b167 LP |
835 | if (saved_stdout < 0) |
836 | return -errno; | |
80876c20 | 837 | |
8854d795 | 838 | fd = acquire_terminal(vc, ACQUIRE_TERMINAL_WAIT, DEFAULT_CONFIRM_USEC); |
3d18b167 LP |
839 | if (fd < 0) |
840 | return fd; | |
80876c20 | 841 | |
af6da548 LP |
842 | r = chown_terminal(fd, getuid()); |
843 | if (r < 0) | |
3d18b167 | 844 | return r; |
02a51aba | 845 | |
3d18b167 LP |
846 | r = reset_terminal_fd(fd, true); |
847 | if (r < 0) | |
848 | return r; | |
80876c20 | 849 | |
4d62ee55 DDM |
850 | r = exec_context_tty_size(context, &rows, &cols); |
851 | if (r < 0) | |
852 | return r; | |
853 | ||
854 | r = terminal_set_size_fd(fd, vc, rows, cols); | |
51462135 DDM |
855 | if (r < 0) |
856 | return r; | |
857 | ||
aedec452 LP |
858 | r = rearrange_stdio(fd, fd, STDERR_FILENO); /* Invalidates 'fd' also on failure */ |
859 | TAKE_FD(fd); | |
2b33ab09 LP |
860 | if (r < 0) |
861 | return r; | |
80876c20 | 862 | |
aedec452 LP |
863 | *ret_saved_stdin = TAKE_FD(saved_stdin); |
864 | *ret_saved_stdout = TAKE_FD(saved_stdout); | |
3d18b167 | 865 | return 0; |
80876c20 LP |
866 | } |
867 | ||
63d77c92 | 868 | static void write_confirm_error_fd(int err, int fd, const Unit *u) { |
3b20f877 FB |
869 | assert(err < 0); |
870 | ||
871 | if (err == -ETIMEDOUT) | |
63d77c92 | 872 | dprintf(fd, "Confirmation question timed out for %s, assuming positive response.\n", u->id); |
3b20f877 FB |
873 | else { |
874 | errno = -err; | |
63d77c92 | 875 | dprintf(fd, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u->id); |
3b20f877 FB |
876 | } |
877 | } | |
878 | ||
63d77c92 | 879 | static void write_confirm_error(int err, const char *vc, const Unit *u) { |
254d1313 | 880 | _cleanup_close_ int fd = -EBADF; |
80876c20 | 881 | |
3b20f877 | 882 | assert(vc); |
80876c20 | 883 | |
7d5ceb64 | 884 | fd = open_terminal(vc, O_WRONLY|O_NOCTTY|O_CLOEXEC); |
af6da548 | 885 | if (fd < 0) |
3b20f877 | 886 | return; |
80876c20 | 887 | |
63d77c92 | 888 | write_confirm_error_fd(err, fd, u); |
af6da548 | 889 | } |
80876c20 | 890 | |
3d18b167 | 891 | static int restore_confirm_stdio(int *saved_stdin, int *saved_stdout) { |
af6da548 | 892 | int r = 0; |
80876c20 | 893 | |
af6da548 LP |
894 | assert(saved_stdin); |
895 | assert(saved_stdout); | |
896 | ||
897 | release_terminal(); | |
898 | ||
899 | if (*saved_stdin >= 0) | |
80876c20 | 900 | if (dup2(*saved_stdin, STDIN_FILENO) < 0) |
af6da548 | 901 | r = -errno; |
80876c20 | 902 | |
af6da548 | 903 | if (*saved_stdout >= 0) |
80876c20 | 904 | if (dup2(*saved_stdout, STDOUT_FILENO) < 0) |
af6da548 | 905 | r = -errno; |
80876c20 | 906 | |
3d18b167 LP |
907 | *saved_stdin = safe_close(*saved_stdin); |
908 | *saved_stdout = safe_close(*saved_stdout); | |
af6da548 LP |
909 | |
910 | return r; | |
911 | } | |
912 | ||
3b20f877 FB |
913 | enum { |
914 | CONFIRM_PRETEND_FAILURE = -1, | |
915 | CONFIRM_PRETEND_SUCCESS = 0, | |
916 | CONFIRM_EXECUTE = 1, | |
917 | }; | |
918 | ||
51462135 | 919 | static int ask_for_confirmation(const ExecContext *context, const char *vc, Unit *u, const char *cmdline) { |
af6da548 | 920 | int saved_stdout = -1, saved_stdin = -1, r; |
2bcd3c26 | 921 | _cleanup_free_ char *e = NULL; |
3b20f877 | 922 | char c; |
af6da548 | 923 | |
3b20f877 | 924 | /* For any internal errors, assume a positive response. */ |
51462135 | 925 | r = setup_confirm_stdio(context, vc, &saved_stdin, &saved_stdout); |
3b20f877 | 926 | if (r < 0) { |
63d77c92 | 927 | write_confirm_error(r, vc, u); |
3b20f877 FB |
928 | return CONFIRM_EXECUTE; |
929 | } | |
af6da548 | 930 | |
b0eb2944 FB |
931 | /* confirm_spawn might have been disabled while we were sleeping. */ |
932 | if (manager_is_confirm_spawn_disabled(u->manager)) { | |
933 | r = 1; | |
934 | goto restore_stdio; | |
935 | } | |
af6da548 | 936 | |
2bcd3c26 FB |
937 | e = ellipsize(cmdline, 60, 100); |
938 | if (!e) { | |
939 | log_oom(); | |
940 | r = CONFIRM_EXECUTE; | |
941 | goto restore_stdio; | |
942 | } | |
af6da548 | 943 | |
d172b175 | 944 | for (;;) { |
539622bd | 945 | r = ask_char(&c, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e); |
d172b175 | 946 | if (r < 0) { |
63d77c92 | 947 | write_confirm_error_fd(r, STDOUT_FILENO, u); |
d172b175 FB |
948 | r = CONFIRM_EXECUTE; |
949 | goto restore_stdio; | |
950 | } | |
af6da548 | 951 | |
d172b175 | 952 | switch (c) { |
b0eb2944 FB |
953 | case 'c': |
954 | printf("Resuming normal execution.\n"); | |
955 | manager_disable_confirm_spawn(); | |
956 | r = 1; | |
957 | break; | |
dd6f9ac0 FB |
958 | case 'D': |
959 | unit_dump(u, stdout, " "); | |
960 | continue; /* ask again */ | |
d172b175 FB |
961 | case 'f': |
962 | printf("Failing execution.\n"); | |
963 | r = CONFIRM_PRETEND_FAILURE; | |
964 | break; | |
965 | case 'h': | |
b0eb2944 FB |
966 | printf(" c - continue, proceed without asking anymore\n" |
967 | " D - dump, show the state of the unit\n" | |
dd6f9ac0 | 968 | " f - fail, don't execute the command and pretend it failed\n" |
d172b175 | 969 | " h - help\n" |
eedf223a | 970 | " i - info, show a short summary of the unit\n" |
56fde33a | 971 | " j - jobs, show jobs that are in progress\n" |
d172b175 FB |
972 | " s - skip, don't execute the command and pretend it succeeded\n" |
973 | " y - yes, execute the command\n"); | |
dd6f9ac0 | 974 | continue; /* ask again */ |
eedf223a FB |
975 | case 'i': |
976 | printf(" Description: %s\n" | |
977 | " Unit: %s\n" | |
978 | " Command: %s\n", | |
979 | u->id, u->description, cmdline); | |
980 | continue; /* ask again */ | |
56fde33a | 981 | case 'j': |
d1d8786c | 982 | manager_dump_jobs(u->manager, stdout, /* patterns= */ NULL, " "); |
56fde33a | 983 | continue; /* ask again */ |
539622bd FB |
984 | case 'n': |
985 | /* 'n' was removed in favor of 'f'. */ | |
986 | printf("Didn't understand 'n', did you mean 'f'?\n"); | |
987 | continue; /* ask again */ | |
d172b175 FB |
988 | case 's': |
989 | printf("Skipping execution.\n"); | |
990 | r = CONFIRM_PRETEND_SUCCESS; | |
991 | break; | |
992 | case 'y': | |
993 | r = CONFIRM_EXECUTE; | |
994 | break; | |
995 | default: | |
04499a70 | 996 | assert_not_reached(); |
d172b175 | 997 | } |
3b20f877 | 998 | break; |
3b20f877 | 999 | } |
af6da548 | 1000 | |
3b20f877 | 1001 | restore_stdio: |
af6da548 | 1002 | restore_confirm_stdio(&saved_stdin, &saved_stdout); |
af6da548 | 1003 | return r; |
80876c20 LP |
1004 | } |
1005 | ||
4d885bd3 DH |
1006 | static int get_fixed_user(const ExecContext *c, const char **user, |
1007 | uid_t *uid, gid_t *gid, | |
1008 | const char **home, const char **shell) { | |
81a2b7ce | 1009 | int r; |
4d885bd3 | 1010 | const char *name; |
81a2b7ce | 1011 | |
4d885bd3 | 1012 | assert(c); |
81a2b7ce | 1013 | |
23deef88 LP |
1014 | if (!c->user) |
1015 | return 0; | |
1016 | ||
4d885bd3 DH |
1017 | /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway |
1018 | * (i.e. are "/" or "/bin/nologin"). */ | |
81a2b7ce | 1019 | |
23deef88 | 1020 | name = c->user; |
fafff8f1 | 1021 | r = get_user_creds(&name, uid, gid, home, shell, USER_CREDS_CLEAN); |
4d885bd3 DH |
1022 | if (r < 0) |
1023 | return r; | |
81a2b7ce | 1024 | |
4d885bd3 DH |
1025 | *user = name; |
1026 | return 0; | |
1027 | } | |
1028 | ||
1029 | static int get_fixed_group(const ExecContext *c, const char **group, gid_t *gid) { | |
1030 | int r; | |
1031 | const char *name; | |
1032 | ||
1033 | assert(c); | |
1034 | ||
1035 | if (!c->group) | |
1036 | return 0; | |
1037 | ||
1038 | name = c->group; | |
fafff8f1 | 1039 | r = get_group_creds(&name, gid, 0); |
4d885bd3 DH |
1040 | if (r < 0) |
1041 | return r; | |
1042 | ||
1043 | *group = name; | |
1044 | return 0; | |
1045 | } | |
1046 | ||
cdc5d5c5 DH |
1047 | static int get_supplementary_groups(const ExecContext *c, const char *user, |
1048 | const char *group, gid_t gid, | |
1049 | gid_t **supplementary_gids, int *ngids) { | |
4d885bd3 DH |
1050 | int r, k = 0; |
1051 | int ngroups_max; | |
1052 | bool keep_groups = false; | |
1053 | gid_t *groups = NULL; | |
1054 | _cleanup_free_ gid_t *l_gids = NULL; | |
1055 | ||
1056 | assert(c); | |
1057 | ||
bbeea271 DH |
1058 | /* |
1059 | * If user is given, then lookup GID and supplementary groups list. | |
1060 | * We avoid NSS lookups for gid=0. Also we have to initialize groups | |
cdc5d5c5 DH |
1061 | * here and as early as possible so we keep the list of supplementary |
1062 | * groups of the caller. | |
bbeea271 DH |
1063 | */ |
1064 | if (user && gid_is_valid(gid) && gid != 0) { | |
1065 | /* First step, initialize groups from /etc/groups */ | |
1066 | if (initgroups(user, gid) < 0) | |
1067 | return -errno; | |
1068 | ||
1069 | keep_groups = true; | |
1070 | } | |
1071 | ||
ac6e8be6 | 1072 | if (strv_isempty(c->supplementary_groups)) |
4d885bd3 DH |
1073 | return 0; |
1074 | ||
366ddd25 DH |
1075 | /* |
1076 | * If SupplementaryGroups= was passed then NGROUPS_MAX has to | |
1077 | * be positive, otherwise fail. | |
1078 | */ | |
1079 | errno = 0; | |
1080 | ngroups_max = (int) sysconf(_SC_NGROUPS_MAX); | |
66855de7 LP |
1081 | if (ngroups_max <= 0) |
1082 | return errno_or_else(EOPNOTSUPP); | |
366ddd25 | 1083 | |
4d885bd3 DH |
1084 | l_gids = new(gid_t, ngroups_max); |
1085 | if (!l_gids) | |
1086 | return -ENOMEM; | |
81a2b7ce | 1087 | |
4d885bd3 DH |
1088 | if (keep_groups) { |
1089 | /* | |
1090 | * Lookup the list of groups that the user belongs to, we | |
1091 | * avoid NSS lookups here too for gid=0. | |
1092 | */ | |
1093 | k = ngroups_max; | |
1094 | if (getgrouplist(user, gid, l_gids, &k) < 0) | |
1095 | return -EINVAL; | |
1096 | } else | |
1097 | k = 0; | |
81a2b7ce | 1098 | |
4d885bd3 DH |
1099 | STRV_FOREACH(i, c->supplementary_groups) { |
1100 | const char *g; | |
81a2b7ce | 1101 | |
4d885bd3 DH |
1102 | if (k >= ngroups_max) |
1103 | return -E2BIG; | |
81a2b7ce | 1104 | |
4d885bd3 | 1105 | g = *i; |
fafff8f1 | 1106 | r = get_group_creds(&g, l_gids+k, 0); |
4d885bd3 DH |
1107 | if (r < 0) |
1108 | return r; | |
81a2b7ce | 1109 | |
4d885bd3 DH |
1110 | k++; |
1111 | } | |
81a2b7ce | 1112 | |
4d885bd3 DH |
1113 | /* |
1114 | * Sets ngids to zero to drop all supplementary groups, happens | |
1115 | * when we are under root and SupplementaryGroups= is empty. | |
1116 | */ | |
1117 | if (k == 0) { | |
1118 | *ngids = 0; | |
1119 | return 0; | |
1120 | } | |
81a2b7ce | 1121 | |
4d885bd3 DH |
1122 | /* Otherwise get the final list of supplementary groups */ |
1123 | groups = memdup(l_gids, sizeof(gid_t) * k); | |
1124 | if (!groups) | |
1125 | return -ENOMEM; | |
1126 | ||
1127 | *supplementary_gids = groups; | |
1128 | *ngids = k; | |
1129 | ||
1130 | groups = NULL; | |
1131 | ||
1132 | return 0; | |
1133 | } | |
1134 | ||
34cf6c43 | 1135 | static int enforce_groups(gid_t gid, const gid_t *supplementary_gids, int ngids) { |
4d885bd3 DH |
1136 | int r; |
1137 | ||
709dbeac YW |
1138 | /* Handle SupplementaryGroups= if it is not empty */ |
1139 | if (ngids > 0) { | |
4d885bd3 DH |
1140 | r = maybe_setgroups(ngids, supplementary_gids); |
1141 | if (r < 0) | |
97f0e76f | 1142 | return r; |
4d885bd3 | 1143 | } |
81a2b7ce | 1144 | |
4d885bd3 DH |
1145 | if (gid_is_valid(gid)) { |
1146 | /* Then set our gids */ | |
1147 | if (setresgid(gid, gid, gid) < 0) | |
1148 | return -errno; | |
81a2b7ce LP |
1149 | } |
1150 | ||
1151 | return 0; | |
1152 | } | |
1153 | ||
a954b249 LP |
1154 | static int set_securebits(unsigned bits, unsigned mask) { |
1155 | unsigned applied; | |
1156 | int current; | |
1157 | ||
dbdc4098 TK |
1158 | current = prctl(PR_GET_SECUREBITS); |
1159 | if (current < 0) | |
1160 | return -errno; | |
a954b249 | 1161 | |
dbdc4098 | 1162 | /* Clear all securebits defined in mask and set bits */ |
a954b249 LP |
1163 | applied = ((unsigned) current & ~mask) | bits; |
1164 | if ((unsigned) current == applied) | |
dbdc4098 | 1165 | return 0; |
a954b249 | 1166 | |
dbdc4098 TK |
1167 | if (prctl(PR_SET_SECUREBITS, applied) < 0) |
1168 | return -errno; | |
a954b249 | 1169 | |
dbdc4098 TK |
1170 | return 1; |
1171 | } | |
1172 | ||
638fd8cc LP |
1173 | static int enforce_user( |
1174 | const ExecContext *context, | |
1175 | uid_t uid, | |
1176 | uint64_t capability_ambient_set) { | |
81a2b7ce | 1177 | assert(context); |
dbdc4098 | 1178 | int r; |
81a2b7ce | 1179 | |
4d885bd3 DH |
1180 | if (!uid_is_valid(uid)) |
1181 | return 0; | |
1182 | ||
a954b249 LP |
1183 | /* Sets (but doesn't look up) the UIS and makes sure we keep the capabilities while doing so. For |
1184 | * setting secure bits the capability CAP_SETPCAP is required, so we also need keep-caps in this | |
1185 | * case. */ | |
81a2b7ce | 1186 | |
638fd8cc | 1187 | if ((capability_ambient_set != 0 || context->secure_bits != 0) && uid != 0) { |
81a2b7ce | 1188 | |
a954b249 LP |
1189 | /* First step: If we need to keep capabilities but drop privileges we need to make sure we |
1190 | * keep our caps, while we drop privileges. Add KEEP_CAPS to the securebits */ | |
1191 | r = set_securebits(1U << SECURE_KEEP_CAPS, 0); | |
1192 | if (r < 0) | |
1193 | return r; | |
81a2b7ce LP |
1194 | } |
1195 | ||
479050b3 | 1196 | /* Second step: actually set the uids */ |
81a2b7ce LP |
1197 | if (setresuid(uid, uid, uid) < 0) |
1198 | return -errno; | |
1199 | ||
a954b249 LP |
1200 | /* At this point we should have all necessary capabilities but are otherwise a normal user. However, |
1201 | * the caps might got corrupted due to the setresuid() so we need clean them up later. This is done | |
1202 | * outside of this call. */ | |
81a2b7ce LP |
1203 | return 0; |
1204 | } | |
1205 | ||
349cc4a5 | 1206 | #if HAVE_PAM |
5b6319dc LP |
1207 | |
1208 | static int null_conv( | |
1209 | int num_msg, | |
1210 | const struct pam_message **msg, | |
1211 | struct pam_response **resp, | |
1212 | void *appdata_ptr) { | |
1213 | ||
1214 | /* We don't support conversations */ | |
1215 | ||
1216 | return PAM_CONV_ERR; | |
1217 | } | |
1218 | ||
cefc33ae LP |
1219 | #endif |
1220 | ||
5b6319dc LP |
1221 | static int setup_pam( |
1222 | const char *name, | |
1223 | const char *user, | |
940c5210 | 1224 | uid_t uid, |
2d6fce8d | 1225 | gid_t gid, |
5b6319dc | 1226 | const char *tty, |
421bb42d | 1227 | char ***env, /* updated on success */ |
5b8d1f6b | 1228 | const int fds[], size_t n_fds) { |
5b6319dc | 1229 | |
349cc4a5 | 1230 | #if HAVE_PAM |
cefc33ae | 1231 | |
5b6319dc LP |
1232 | static const struct pam_conv conv = { |
1233 | .conv = null_conv, | |
1234 | .appdata_ptr = NULL | |
1235 | }; | |
1236 | ||
2d7c6aa2 | 1237 | _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL; |
46e5bbab | 1238 | _cleanup_strv_free_ char **e = NULL; |
5b6319dc | 1239 | pam_handle_t *handle = NULL; |
d6e5f3ad | 1240 | sigset_t old_ss; |
7bb70b6e | 1241 | int pam_code = PAM_SUCCESS, r; |
5b6319dc LP |
1242 | bool close_session = false; |
1243 | pid_t pam_pid = 0, parent_pid; | |
970edce6 | 1244 | int flags = 0; |
5b6319dc LP |
1245 | |
1246 | assert(name); | |
1247 | assert(user); | |
2065ca69 | 1248 | assert(env); |
5b6319dc LP |
1249 | |
1250 | /* We set up PAM in the parent process, then fork. The child | |
35b8ca3a | 1251 | * will then stay around until killed via PR_GET_PDEATHSIG or |
5b6319dc LP |
1252 | * systemd via the cgroup logic. It will then remove the PAM |
1253 | * session again. The parent process will exec() the actual | |
1254 | * daemon. We do things this way to ensure that the main PID | |
1255 | * of the daemon is the one we initially fork()ed. */ | |
1256 | ||
7bb70b6e LP |
1257 | r = barrier_create(&barrier); |
1258 | if (r < 0) | |
2d7c6aa2 DH |
1259 | goto fail; |
1260 | ||
553d2243 | 1261 | if (log_get_max_level() < LOG_DEBUG) |
970edce6 ZJS |
1262 | flags |= PAM_SILENT; |
1263 | ||
f546241b ZJS |
1264 | pam_code = pam_start(name, user, &conv, &handle); |
1265 | if (pam_code != PAM_SUCCESS) { | |
5b6319dc LP |
1266 | handle = NULL; |
1267 | goto fail; | |
1268 | } | |
1269 | ||
3cd24c1a LP |
1270 | if (!tty) { |
1271 | _cleanup_free_ char *q = NULL; | |
1272 | ||
1273 | /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure | |
1274 | * out if that's the case, and read the TTY off it. */ | |
1275 | ||
1276 | if (getttyname_malloc(STDIN_FILENO, &q) >= 0) | |
1277 | tty = strjoina("/dev/", q); | |
1278 | } | |
1279 | ||
513cf7da MS |
1280 | if (tty) { |
1281 | pam_code = pam_set_item(handle, PAM_TTY, tty); | |
1282 | if (pam_code != PAM_SUCCESS) | |
1283 | goto fail; | |
1284 | } | |
5b6319dc | 1285 | |
84eada2f JW |
1286 | STRV_FOREACH(nv, *env) { |
1287 | pam_code = pam_putenv(handle, *nv); | |
2065ca69 JW |
1288 | if (pam_code != PAM_SUCCESS) |
1289 | goto fail; | |
1290 | } | |
1291 | ||
970edce6 | 1292 | pam_code = pam_acct_mgmt(handle, flags); |
f546241b | 1293 | if (pam_code != PAM_SUCCESS) |
5b6319dc LP |
1294 | goto fail; |
1295 | ||
3bb39ea9 DG |
1296 | pam_code = pam_setcred(handle, PAM_ESTABLISH_CRED | flags); |
1297 | if (pam_code != PAM_SUCCESS) | |
46d7c6af | 1298 | log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle, pam_code)); |
3bb39ea9 | 1299 | |
970edce6 | 1300 | pam_code = pam_open_session(handle, flags); |
f546241b | 1301 | if (pam_code != PAM_SUCCESS) |
5b6319dc LP |
1302 | goto fail; |
1303 | ||
1304 | close_session = true; | |
1305 | ||
f546241b ZJS |
1306 | e = pam_getenvlist(handle); |
1307 | if (!e) { | |
5b6319dc LP |
1308 | pam_code = PAM_BUF_ERR; |
1309 | goto fail; | |
1310 | } | |
1311 | ||
cafc5ca1 | 1312 | /* Block SIGTERM, so that we know that it won't get lost in the child */ |
ce30c8dc | 1313 | |
72c0a2c2 | 1314 | assert_se(sigprocmask_many(SIG_BLOCK, &old_ss, SIGTERM, -1) >= 0); |
5b6319dc | 1315 | |
df0ff127 | 1316 | parent_pid = getpid_cached(); |
5b6319dc | 1317 | |
4c253ed1 LP |
1318 | r = safe_fork("(sd-pam)", 0, &pam_pid); |
1319 | if (r < 0) | |
5b6319dc | 1320 | goto fail; |
4c253ed1 | 1321 | if (r == 0) { |
7bb70b6e | 1322 | int sig, ret = EXIT_PAM; |
5b6319dc | 1323 | |
cafc5ca1 | 1324 | /* The child's job is to reset the PAM session on termination */ |
2d7c6aa2 | 1325 | barrier_set_role(&barrier, BARRIER_CHILD); |
5b6319dc | 1326 | |
1da37e58 ZJS |
1327 | /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only |
1328 | * those fds are open here that have been opened by PAM. */ | |
4c253ed1 | 1329 | (void) close_many(fds, n_fds); |
5b6319dc | 1330 | |
cafc5ca1 LP |
1331 | /* Drop privileges - we don't need any to pam_close_session and this will make |
1332 | * PR_SET_PDEATHSIG work in most cases. If this fails, ignore the error - but expect sd-pam | |
1333 | * threads to fail to exit normally */ | |
2d6fce8d | 1334 | |
97f0e76f LP |
1335 | r = maybe_setgroups(0, NULL); |
1336 | if (r < 0) | |
1337 | log_warning_errno(r, "Failed to setgroups() in sd-pam: %m"); | |
2d6fce8d LP |
1338 | if (setresgid(gid, gid, gid) < 0) |
1339 | log_warning_errno(errno, "Failed to setresgid() in sd-pam: %m"); | |
940c5210 | 1340 | if (setresuid(uid, uid, uid) < 0) |
2d6fce8d | 1341 | log_warning_errno(errno, "Failed to setresuid() in sd-pam: %m"); |
940c5210 | 1342 | |
9c274488 | 1343 | (void) ignore_signals(SIGPIPE); |
ce30c8dc | 1344 | |
cafc5ca1 LP |
1345 | /* Wait until our parent died. This will only work if the above setresuid() succeeds, |
1346 | * otherwise the kernel will not allow unprivileged parents kill their privileged children | |
1347 | * this way. We rely on the control groups kill logic to do the rest for us. */ | |
5b6319dc LP |
1348 | if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) |
1349 | goto child_finish; | |
1350 | ||
cafc5ca1 LP |
1351 | /* Tell the parent that our setup is done. This is especially important regarding dropping |
1352 | * privileges. Otherwise, unit setup might race against our setresuid(2) call. | |
643f4706 | 1353 | * |
cafc5ca1 | 1354 | * If the parent aborted, we'll detect this below, hence ignore return failure here. */ |
643f4706 | 1355 | (void) barrier_place(&barrier); |
2d7c6aa2 | 1356 | |
643f4706 | 1357 | /* Check if our parent process might already have died? */ |
5b6319dc | 1358 | if (getppid() == parent_pid) { |
d6e5f3ad DM |
1359 | sigset_t ss; |
1360 | ||
1361 | assert_se(sigemptyset(&ss) >= 0); | |
1362 | assert_se(sigaddset(&ss, SIGTERM) >= 0); | |
1363 | ||
3dead8d9 LP |
1364 | for (;;) { |
1365 | if (sigwait(&ss, &sig) < 0) { | |
1366 | if (errno == EINTR) | |
1367 | continue; | |
1368 | ||
1369 | goto child_finish; | |
1370 | } | |
5b6319dc | 1371 | |
3dead8d9 LP |
1372 | assert(sig == SIGTERM); |
1373 | break; | |
1374 | } | |
5b6319dc LP |
1375 | } |
1376 | ||
3bb39ea9 DG |
1377 | pam_code = pam_setcred(handle, PAM_DELETE_CRED | flags); |
1378 | if (pam_code != PAM_SUCCESS) | |
1379 | goto child_finish; | |
1380 | ||
3dead8d9 | 1381 | /* If our parent died we'll end the session */ |
f546241b | 1382 | if (getppid() != parent_pid) { |
970edce6 | 1383 | pam_code = pam_close_session(handle, flags); |
f546241b | 1384 | if (pam_code != PAM_SUCCESS) |
5b6319dc | 1385 | goto child_finish; |
f546241b | 1386 | } |
5b6319dc | 1387 | |
7bb70b6e | 1388 | ret = 0; |
5b6319dc LP |
1389 | |
1390 | child_finish: | |
7feb2b57 LP |
1391 | /* NB: pam_end() when called in child processes should set PAM_DATA_SILENT to let the module |
1392 | * know about this. See pam_end(3) */ | |
1393 | (void) pam_end(handle, pam_code | flags | PAM_DATA_SILENT); | |
7bb70b6e | 1394 | _exit(ret); |
5b6319dc LP |
1395 | } |
1396 | ||
2d7c6aa2 DH |
1397 | barrier_set_role(&barrier, BARRIER_PARENT); |
1398 | ||
cafc5ca1 LP |
1399 | /* If the child was forked off successfully it will do all the cleanups, so forget about the handle |
1400 | * here. */ | |
5b6319dc LP |
1401 | handle = NULL; |
1402 | ||
3b8bddde | 1403 | /* Unblock SIGTERM again in the parent */ |
72c0a2c2 | 1404 | assert_se(sigprocmask(SIG_SETMASK, &old_ss, NULL) >= 0); |
5b6319dc | 1405 | |
cafc5ca1 LP |
1406 | /* We close the log explicitly here, since the PAM modules might have opened it, but we don't want |
1407 | * this fd around. */ | |
5b6319dc LP |
1408 | closelog(); |
1409 | ||
cafc5ca1 LP |
1410 | /* Synchronously wait for the child to initialize. We don't care for errors as we cannot |
1411 | * recover. However, warn loudly if it happens. */ | |
2d7c6aa2 DH |
1412 | if (!barrier_place_and_sync(&barrier)) |
1413 | log_error("PAM initialization failed"); | |
1414 | ||
130d3d22 | 1415 | return strv_free_and_replace(*env, e); |
5b6319dc LP |
1416 | |
1417 | fail: | |
970edce6 ZJS |
1418 | if (pam_code != PAM_SUCCESS) { |
1419 | log_error("PAM failed: %s", pam_strerror(handle, pam_code)); | |
7bb70b6e LP |
1420 | r = -EPERM; /* PAM errors do not map to errno */ |
1421 | } else | |
1422 | log_error_errno(r, "PAM failed: %m"); | |
9ba35398 | 1423 | |
5b6319dc LP |
1424 | if (handle) { |
1425 | if (close_session) | |
970edce6 | 1426 | pam_code = pam_close_session(handle, flags); |
5b6319dc | 1427 | |
7feb2b57 | 1428 | (void) pam_end(handle, pam_code | flags); |
5b6319dc LP |
1429 | } |
1430 | ||
5b6319dc | 1431 | closelog(); |
7bb70b6e | 1432 | return r; |
cefc33ae LP |
1433 | #else |
1434 | return 0; | |
5b6319dc | 1435 | #endif |
cefc33ae | 1436 | } |
5b6319dc | 1437 | |
5d6b1584 | 1438 | static void rename_process_from_path(const char *path) { |
a99626c1 | 1439 | _cleanup_free_ char *buf = NULL; |
5d6b1584 | 1440 | const char *p; |
5d6b1584 | 1441 | |
a99626c1 LP |
1442 | assert(path); |
1443 | ||
1444 | /* This resulting string must fit in 10 chars (i.e. the length of "/sbin/init") to look pretty in | |
1445 | * /bin/ps */ | |
5d6b1584 | 1446 | |
a99626c1 | 1447 | if (path_extract_filename(path, &buf) < 0) { |
5d6b1584 LP |
1448 | rename_process("(...)"); |
1449 | return; | |
1450 | } | |
1451 | ||
a99626c1 | 1452 | size_t l = strlen(buf); |
5d6b1584 | 1453 | if (l > 8) { |
a99626c1 | 1454 | /* The end of the process name is usually more interesting, since the first bit might just be |
5d6b1584 | 1455 | * "systemd-" */ |
a99626c1 | 1456 | p = buf + l - 8; |
5d6b1584 | 1457 | l = 8; |
a99626c1 LP |
1458 | } else |
1459 | p = buf; | |
5d6b1584 | 1460 | |
a99626c1 | 1461 | char process_name[11]; |
5d6b1584 LP |
1462 | process_name[0] = '('; |
1463 | memcpy(process_name+1, p, l); | |
1464 | process_name[1+l] = ')'; | |
1465 | process_name[1+l+1] = 0; | |
1466 | ||
1467 | rename_process(process_name); | |
1468 | } | |
1469 | ||
469830d1 LP |
1470 | static bool context_has_address_families(const ExecContext *c) { |
1471 | assert(c); | |
1472 | ||
6b000af4 | 1473 | return c->address_families_allow_list || |
469830d1 LP |
1474 | !set_isempty(c->address_families); |
1475 | } | |
1476 | ||
1477 | static bool context_has_syscall_filters(const ExecContext *c) { | |
1478 | assert(c); | |
1479 | ||
6b000af4 | 1480 | return c->syscall_allow_list || |
8cfa775f | 1481 | !hashmap_isempty(c->syscall_filter); |
469830d1 LP |
1482 | } |
1483 | ||
9df2cdd8 TM |
1484 | static bool context_has_syscall_logs(const ExecContext *c) { |
1485 | assert(c); | |
1486 | ||
1487 | return c->syscall_log_allow_list || | |
1488 | !hashmap_isempty(c->syscall_log); | |
1489 | } | |
1490 | ||
469830d1 LP |
1491 | static bool context_has_no_new_privileges(const ExecContext *c) { |
1492 | assert(c); | |
1493 | ||
1494 | if (c->no_new_privileges) | |
1495 | return true; | |
1496 | ||
26c45a6c | 1497 | if (have_effective_cap(CAP_SYS_ADMIN) > 0) /* if we are privileged, we don't need NNP */ |
469830d1 LP |
1498 | return false; |
1499 | ||
1500 | /* We need NNP if we have any form of seccomp and are unprivileged */ | |
0538d2a8 | 1501 | return c->lock_personality || |
469830d1 | 1502 | c->memory_deny_write_execute || |
0538d2a8 | 1503 | c->private_devices || |
fc64760d | 1504 | c->protect_clock || |
0538d2a8 | 1505 | c->protect_hostname || |
469830d1 LP |
1506 | c->protect_kernel_tunables || |
1507 | c->protect_kernel_modules || | |
84703040 | 1508 | c->protect_kernel_logs || |
0538d2a8 YW |
1509 | context_has_address_families(c) || |
1510 | exec_context_restrict_namespaces_set(c) || | |
1511 | c->restrict_realtime || | |
1512 | c->restrict_suid_sgid || | |
78e864e5 | 1513 | !set_isempty(c->syscall_archs) || |
0538d2a8 YW |
1514 | context_has_syscall_filters(c) || |
1515 | context_has_syscall_logs(c); | |
469830d1 LP |
1516 | } |
1517 | ||
ce18c396 | 1518 | bool exec_context_has_credentials(const ExecContext *context) { |
bb0c0d6f LP |
1519 | |
1520 | assert(context); | |
1521 | ||
1522 | return !hashmap_isempty(context->set_credentials) || | |
bbfb25f4 DDM |
1523 | !hashmap_isempty(context->load_credentials) || |
1524 | !set_isempty(context->import_credentials); | |
bb0c0d6f LP |
1525 | } |
1526 | ||
349cc4a5 | 1527 | #if HAVE_SECCOMP |
17df7223 | 1528 | |
83f12b27 | 1529 | static bool skip_seccomp_unavailable(const Unit* u, const char* msg) { |
f673b62d LP |
1530 | |
1531 | if (is_seccomp_available()) | |
1532 | return false; | |
1533 | ||
f673b62d | 1534 | log_unit_debug(u, "SECCOMP features not detected in the kernel, skipping %s", msg); |
f673b62d | 1535 | return true; |
83f12b27 FS |
1536 | } |
1537 | ||
165a31c0 | 1538 | static int apply_syscall_filter(const Unit* u, const ExecContext *c, bool needs_ambient_hack) { |
469830d1 | 1539 | uint32_t negative_action, default_action, action; |
165a31c0 | 1540 | int r; |
8351ceae | 1541 | |
469830d1 | 1542 | assert(u); |
c0467cf3 | 1543 | assert(c); |
8351ceae | 1544 | |
469830d1 | 1545 | if (!context_has_syscall_filters(c)) |
83f12b27 FS |
1546 | return 0; |
1547 | ||
469830d1 LP |
1548 | if (skip_seccomp_unavailable(u, "SystemCallFilter=")) |
1549 | return 0; | |
e9642be2 | 1550 | |
005bfaf1 | 1551 | negative_action = c->syscall_errno == SECCOMP_ERROR_NUMBER_KILL ? scmp_act_kill_process() : SCMP_ACT_ERRNO(c->syscall_errno); |
e9642be2 | 1552 | |
6b000af4 | 1553 | if (c->syscall_allow_list) { |
469830d1 LP |
1554 | default_action = negative_action; |
1555 | action = SCMP_ACT_ALLOW; | |
7c66bae2 | 1556 | } else { |
469830d1 LP |
1557 | default_action = SCMP_ACT_ALLOW; |
1558 | action = negative_action; | |
57183d11 | 1559 | } |
8351ceae | 1560 | |
165a31c0 | 1561 | if (needs_ambient_hack) { |
6b000af4 | 1562 | r = seccomp_filter_set_add(c->syscall_filter, c->syscall_allow_list, syscall_filter_sets + SYSCALL_FILTER_SET_SETUID); |
165a31c0 LP |
1563 | if (r < 0) |
1564 | return r; | |
1565 | } | |
1566 | ||
b54f36c6 | 1567 | return seccomp_load_syscall_filter_set_raw(default_action, c->syscall_filter, action, false); |
4298d0b5 LP |
1568 | } |
1569 | ||
9df2cdd8 TM |
1570 | static int apply_syscall_log(const Unit* u, const ExecContext *c) { |
1571 | #ifdef SCMP_ACT_LOG | |
1572 | uint32_t default_action, action; | |
1573 | #endif | |
1574 | ||
1575 | assert(u); | |
1576 | assert(c); | |
1577 | ||
1578 | if (!context_has_syscall_logs(c)) | |
1579 | return 0; | |
1580 | ||
1581 | #ifdef SCMP_ACT_LOG | |
1582 | if (skip_seccomp_unavailable(u, "SystemCallLog=")) | |
1583 | return 0; | |
1584 | ||
1585 | if (c->syscall_log_allow_list) { | |
1586 | /* Log nothing but the ones listed */ | |
1587 | default_action = SCMP_ACT_ALLOW; | |
1588 | action = SCMP_ACT_LOG; | |
1589 | } else { | |
1590 | /* Log everything but the ones listed */ | |
1591 | default_action = SCMP_ACT_LOG; | |
1592 | action = SCMP_ACT_ALLOW; | |
1593 | } | |
1594 | ||
1595 | return seccomp_load_syscall_filter_set_raw(default_action, c->syscall_log, action, false); | |
1596 | #else | |
1597 | /* old libseccomp */ | |
1598 | log_unit_debug(u, "SECCOMP feature SCMP_ACT_LOG not available, skipping SystemCallLog="); | |
1599 | return 0; | |
1600 | #endif | |
1601 | } | |
1602 | ||
469830d1 LP |
1603 | static int apply_syscall_archs(const Unit *u, const ExecContext *c) { |
1604 | assert(u); | |
4298d0b5 LP |
1605 | assert(c); |
1606 | ||
469830d1 | 1607 | if (set_isempty(c->syscall_archs)) |
83f12b27 FS |
1608 | return 0; |
1609 | ||
469830d1 LP |
1610 | if (skip_seccomp_unavailable(u, "SystemCallArchitectures=")) |
1611 | return 0; | |
4298d0b5 | 1612 | |
469830d1 LP |
1613 | return seccomp_restrict_archs(c->syscall_archs); |
1614 | } | |
4298d0b5 | 1615 | |
469830d1 LP |
1616 | static int apply_address_families(const Unit* u, const ExecContext *c) { |
1617 | assert(u); | |
1618 | assert(c); | |
4298d0b5 | 1619 | |
469830d1 LP |
1620 | if (!context_has_address_families(c)) |
1621 | return 0; | |
4298d0b5 | 1622 | |
469830d1 LP |
1623 | if (skip_seccomp_unavailable(u, "RestrictAddressFamilies=")) |
1624 | return 0; | |
4298d0b5 | 1625 | |
6b000af4 | 1626 | return seccomp_restrict_address_families(c->address_families, c->address_families_allow_list); |
8351ceae | 1627 | } |
4298d0b5 | 1628 | |
83f12b27 | 1629 | static int apply_memory_deny_write_execute(const Unit* u, const ExecContext *c) { |
7a114ed4 TM |
1630 | int r; |
1631 | ||
469830d1 | 1632 | assert(u); |
f3e43635 TM |
1633 | assert(c); |
1634 | ||
469830d1 | 1635 | if (!c->memory_deny_write_execute) |
83f12b27 FS |
1636 | return 0; |
1637 | ||
7a114ed4 TM |
1638 | /* use prctl() if kernel supports it (6.3) */ |
1639 | r = prctl(PR_SET_MDWE, PR_MDWE_REFUSE_EXEC_GAIN, 0, 0, 0); | |
1640 | if (r == 0) { | |
1641 | log_unit_debug(u, "Enabled MemoryDenyWriteExecute= with PR_SET_MDWE"); | |
1642 | return 0; | |
1643 | } | |
1644 | if (r < 0 && errno != EINVAL) | |
1645 | return log_unit_debug_errno(u, errno, "Failed to enable MemoryDenyWriteExecute= with PR_SET_MDWE: %m"); | |
1646 | /* else use seccomp */ | |
1647 | log_unit_debug(u, "Kernel doesn't support PR_SET_MDWE: falling back to seccomp"); | |
1648 | ||
469830d1 LP |
1649 | if (skip_seccomp_unavailable(u, "MemoryDenyWriteExecute=")) |
1650 | return 0; | |
f3e43635 | 1651 | |
469830d1 | 1652 | return seccomp_memory_deny_write_execute(); |
f3e43635 TM |
1653 | } |
1654 | ||
83f12b27 | 1655 | static int apply_restrict_realtime(const Unit* u, const ExecContext *c) { |
469830d1 | 1656 | assert(u); |
f4170c67 LP |
1657 | assert(c); |
1658 | ||
469830d1 | 1659 | if (!c->restrict_realtime) |
83f12b27 FS |
1660 | return 0; |
1661 | ||
469830d1 LP |
1662 | if (skip_seccomp_unavailable(u, "RestrictRealtime=")) |
1663 | return 0; | |
f4170c67 | 1664 | |
469830d1 | 1665 | return seccomp_restrict_realtime(); |
f4170c67 LP |
1666 | } |
1667 | ||
f69567cb LP |
1668 | static int apply_restrict_suid_sgid(const Unit* u, const ExecContext *c) { |
1669 | assert(u); | |
1670 | assert(c); | |
1671 | ||
1672 | if (!c->restrict_suid_sgid) | |
1673 | return 0; | |
1674 | ||
1675 | if (skip_seccomp_unavailable(u, "RestrictSUIDSGID=")) | |
1676 | return 0; | |
1677 | ||
1678 | return seccomp_restrict_suid_sgid(); | |
1679 | } | |
1680 | ||
59e856c7 | 1681 | static int apply_protect_sysctl(const Unit *u, const ExecContext *c) { |
469830d1 | 1682 | assert(u); |
59eeb84b LP |
1683 | assert(c); |
1684 | ||
1685 | /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but | |
1686 | * let's protect even those systems where this is left on in the kernel. */ | |
1687 | ||
469830d1 | 1688 | if (!c->protect_kernel_tunables) |
59eeb84b LP |
1689 | return 0; |
1690 | ||
469830d1 LP |
1691 | if (skip_seccomp_unavailable(u, "ProtectKernelTunables=")) |
1692 | return 0; | |
59eeb84b | 1693 | |
469830d1 | 1694 | return seccomp_protect_sysctl(); |
59eeb84b LP |
1695 | } |
1696 | ||
59e856c7 | 1697 | static int apply_protect_kernel_modules(const Unit *u, const ExecContext *c) { |
469830d1 | 1698 | assert(u); |
502d704e DH |
1699 | assert(c); |
1700 | ||
25a8d8a0 | 1701 | /* Turn off module syscalls on ProtectKernelModules=yes */ |
502d704e | 1702 | |
469830d1 LP |
1703 | if (!c->protect_kernel_modules) |
1704 | return 0; | |
1705 | ||
502d704e DH |
1706 | if (skip_seccomp_unavailable(u, "ProtectKernelModules=")) |
1707 | return 0; | |
1708 | ||
b54f36c6 | 1709 | return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_MODULE, SCMP_ACT_ERRNO(EPERM), false); |
502d704e DH |
1710 | } |
1711 | ||
84703040 KK |
1712 | static int apply_protect_kernel_logs(const Unit *u, const ExecContext *c) { |
1713 | assert(u); | |
1714 | assert(c); | |
1715 | ||
1716 | if (!c->protect_kernel_logs) | |
1717 | return 0; | |
1718 | ||
1719 | if (skip_seccomp_unavailable(u, "ProtectKernelLogs=")) | |
1720 | return 0; | |
1721 | ||
1722 | return seccomp_protect_syslog(); | |
1723 | } | |
1724 | ||
daf8f72b | 1725 | static int apply_protect_clock(const Unit *u, const ExecContext *c) { |
fc64760d KK |
1726 | assert(u); |
1727 | assert(c); | |
1728 | ||
1729 | if (!c->protect_clock) | |
1730 | return 0; | |
1731 | ||
1732 | if (skip_seccomp_unavailable(u, "ProtectClock=")) | |
1733 | return 0; | |
1734 | ||
1735 | return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_CLOCK, SCMP_ACT_ERRNO(EPERM), false); | |
1736 | } | |
1737 | ||
59e856c7 | 1738 | static int apply_private_devices(const Unit *u, const ExecContext *c) { |
469830d1 | 1739 | assert(u); |
ba128bb8 LP |
1740 | assert(c); |
1741 | ||
8f81a5f6 | 1742 | /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */ |
ba128bb8 | 1743 | |
469830d1 LP |
1744 | if (!c->private_devices) |
1745 | return 0; | |
1746 | ||
ba128bb8 LP |
1747 | if (skip_seccomp_unavailable(u, "PrivateDevices=")) |
1748 | return 0; | |
1749 | ||
b54f36c6 | 1750 | return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_RAW_IO, SCMP_ACT_ERRNO(EPERM), false); |
ba128bb8 LP |
1751 | } |
1752 | ||
34cf6c43 | 1753 | static int apply_restrict_namespaces(const Unit *u, const ExecContext *c) { |
469830d1 | 1754 | assert(u); |
add00535 LP |
1755 | assert(c); |
1756 | ||
1757 | if (!exec_context_restrict_namespaces_set(c)) | |
1758 | return 0; | |
1759 | ||
1760 | if (skip_seccomp_unavailable(u, "RestrictNamespaces=")) | |
1761 | return 0; | |
1762 | ||
1763 | return seccomp_restrict_namespaces(c->restrict_namespaces); | |
1764 | } | |
1765 | ||
78e864e5 | 1766 | static int apply_lock_personality(const Unit* u, const ExecContext *c) { |
e8132d63 LP |
1767 | unsigned long personality; |
1768 | int r; | |
78e864e5 TM |
1769 | |
1770 | assert(u); | |
1771 | assert(c); | |
1772 | ||
1773 | if (!c->lock_personality) | |
1774 | return 0; | |
1775 | ||
1776 | if (skip_seccomp_unavailable(u, "LockPersonality=")) | |
1777 | return 0; | |
1778 | ||
e8132d63 LP |
1779 | personality = c->personality; |
1780 | ||
1781 | /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */ | |
1782 | if (personality == PERSONALITY_INVALID) { | |
1783 | ||
1784 | r = opinionated_personality(&personality); | |
1785 | if (r < 0) | |
1786 | return r; | |
1787 | } | |
78e864e5 TM |
1788 | |
1789 | return seccomp_lock_personality(personality); | |
1790 | } | |
1791 | ||
c0467cf3 | 1792 | #endif |
8351ceae | 1793 | |
7a8288f6 | 1794 | #if HAVE_LIBBPF |
7a8288f6 DM |
1795 | static int apply_restrict_filesystems(Unit *u, const ExecContext *c) { |
1796 | assert(u); | |
1797 | assert(c); | |
1798 | ||
1799 | if (!exec_context_restrict_filesystems_set(c)) | |
1800 | return 0; | |
1801 | ||
46004616 ZJS |
1802 | if (!u->manager->restrict_fs) { |
1803 | /* LSM BPF is unsupported or lsm_bpf_setup failed */ | |
1804 | log_unit_debug(u, "LSM BPF not supported, skipping RestrictFileSystems="); | |
7a8288f6 | 1805 | return 0; |
46004616 | 1806 | } |
7a8288f6 DM |
1807 | |
1808 | return lsm_bpf_unit_restrict_filesystems(u, c->restrict_filesystems, c->restrict_filesystems_allow_list); | |
1809 | } | |
1810 | #endif | |
1811 | ||
daf8f72b | 1812 | static int apply_protect_hostname(const Unit *u, const ExecContext *c, int *ret_exit_status) { |
daf8f72b LP |
1813 | assert(u); |
1814 | assert(c); | |
1815 | ||
1816 | if (!c->protect_hostname) | |
1817 | return 0; | |
1818 | ||
1819 | if (ns_type_supported(NAMESPACE_UTS)) { | |
1820 | if (unshare(CLONE_NEWUTS) < 0) { | |
1821 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) { | |
1822 | *ret_exit_status = EXIT_NAMESPACE; | |
1823 | return log_unit_error_errno(u, errno, "Failed to set up UTS namespacing: %m"); | |
1824 | } | |
1825 | ||
1826 | log_unit_warning(u, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup."); | |
1827 | } | |
1828 | } else | |
1829 | log_unit_warning(u, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup."); | |
1830 | ||
1831 | #if HAVE_SECCOMP | |
8f3e342f ZJS |
1832 | int r; |
1833 | ||
daf8f72b LP |
1834 | if (skip_seccomp_unavailable(u, "ProtectHostname=")) |
1835 | return 0; | |
1836 | ||
1837 | r = seccomp_protect_hostname(); | |
1838 | if (r < 0) { | |
1839 | *ret_exit_status = EXIT_SECCOMP; | |
1840 | return log_unit_error_errno(u, r, "Failed to apply hostname restrictions: %m"); | |
1841 | } | |
1842 | #endif | |
1843 | ||
1844 | return 0; | |
1845 | } | |
1846 | ||
3042bbeb | 1847 | static void do_idle_pipe_dance(int idle_pipe[static 4]) { |
31a7eb86 ZJS |
1848 | assert(idle_pipe); |
1849 | ||
54eb2300 LP |
1850 | idle_pipe[1] = safe_close(idle_pipe[1]); |
1851 | idle_pipe[2] = safe_close(idle_pipe[2]); | |
31a7eb86 ZJS |
1852 | |
1853 | if (idle_pipe[0] >= 0) { | |
1854 | int r; | |
1855 | ||
1856 | r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC); | |
1857 | ||
1858 | if (idle_pipe[3] >= 0 && r == 0 /* timeout */) { | |
c7cc737f LP |
1859 | ssize_t n; |
1860 | ||
31a7eb86 | 1861 | /* Signal systemd that we are bored and want to continue. */ |
c7cc737f LP |
1862 | n = write(idle_pipe[3], "x", 1); |
1863 | if (n > 0) | |
cd972d69 | 1864 | /* Wait for systemd to react to the signal above. */ |
54756dce | 1865 | (void) fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC); |
31a7eb86 ZJS |
1866 | } |
1867 | ||
54eb2300 | 1868 | idle_pipe[0] = safe_close(idle_pipe[0]); |
31a7eb86 ZJS |
1869 | |
1870 | } | |
1871 | ||
54eb2300 | 1872 | idle_pipe[3] = safe_close(idle_pipe[3]); |
31a7eb86 ZJS |
1873 | } |
1874 | ||
fb2042dd YW |
1875 | static const char *exec_directory_env_name_to_string(ExecDirectoryType t); |
1876 | ||
7cae38c4 | 1877 | static int build_environment( |
34cf6c43 | 1878 | const Unit *u, |
9fa95f85 | 1879 | const ExecContext *c, |
1e22b5cd | 1880 | const ExecParameters *p, |
6bb00842 | 1881 | const CGroupContext *cgroup_context, |
da6053d0 | 1882 | size_t n_fds, |
cd48e23f | 1883 | char **fdnames, |
7cae38c4 LP |
1884 | const char *home, |
1885 | const char *username, | |
1886 | const char *shell, | |
7bce046b LP |
1887 | dev_t journal_stream_dev, |
1888 | ino_t journal_stream_ino, | |
6bb00842 | 1889 | const char *memory_pressure_path, |
7cae38c4 LP |
1890 | char ***ret) { |
1891 | ||
1892 | _cleanup_strv_free_ char **our_env = NULL; | |
da6053d0 | 1893 | size_t n_env = 0; |
7cae38c4 | 1894 | char *x; |
4d62ee55 | 1895 | int r; |
7cae38c4 | 1896 | |
4b58153d | 1897 | assert(u); |
7cae38c4 | 1898 | assert(c); |
7c1cb6f1 | 1899 | assert(p); |
7cae38c4 LP |
1900 | assert(ret); |
1901 | ||
6bb00842 | 1902 | #define N_ENV_VARS 19 |
8d5bb13d | 1903 | our_env = new0(char*, N_ENV_VARS + _EXEC_DIRECTORY_TYPE_MAX); |
7cae38c4 LP |
1904 | if (!our_env) |
1905 | return -ENOMEM; | |
1906 | ||
1907 | if (n_fds > 0) { | |
8dd4c05b LP |
1908 | _cleanup_free_ char *joined = NULL; |
1909 | ||
df0ff127 | 1910 | if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid_cached()) < 0) |
7cae38c4 LP |
1911 | return -ENOMEM; |
1912 | our_env[n_env++] = x; | |
1913 | ||
da6053d0 | 1914 | if (asprintf(&x, "LISTEN_FDS=%zu", n_fds) < 0) |
7cae38c4 LP |
1915 | return -ENOMEM; |
1916 | our_env[n_env++] = x; | |
8dd4c05b | 1917 | |
cd48e23f | 1918 | joined = strv_join(fdnames, ":"); |
8dd4c05b LP |
1919 | if (!joined) |
1920 | return -ENOMEM; | |
1921 | ||
605405c6 | 1922 | x = strjoin("LISTEN_FDNAMES=", joined); |
8dd4c05b LP |
1923 | if (!x) |
1924 | return -ENOMEM; | |
1925 | our_env[n_env++] = x; | |
7cae38c4 LP |
1926 | } |
1927 | ||
b08af3b1 | 1928 | if ((p->flags & EXEC_SET_WATCHDOG) && p->watchdog_usec > 0) { |
df0ff127 | 1929 | if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid_cached()) < 0) |
09812eb7 LP |
1930 | return -ENOMEM; |
1931 | our_env[n_env++] = x; | |
1932 | ||
1e22b5cd | 1933 | if (asprintf(&x, "WATCHDOG_USEC="USEC_FMT, p->watchdog_usec) < 0) |
09812eb7 LP |
1934 | return -ENOMEM; |
1935 | our_env[n_env++] = x; | |
1936 | } | |
1937 | ||
de90700f LP |
1938 | /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use blocking |
1939 | * Varlink calls back to us for look up dynamic users in PID 1. Break the deadlock between D-Bus and | |
1940 | * PID 1 by disabling use of PID1' NSS interface for looking up dynamic users. */ | |
1941 | if (p->flags & EXEC_NSS_DYNAMIC_BYPASS) { | |
1942 | x = strdup("SYSTEMD_NSS_DYNAMIC_BYPASS=1"); | |
fd63e712 LP |
1943 | if (!x) |
1944 | return -ENOMEM; | |
1945 | our_env[n_env++] = x; | |
1946 | } | |
1947 | ||
7cae38c4 | 1948 | if (home) { |
b910cc72 | 1949 | x = strjoin("HOME=", home); |
7cae38c4 LP |
1950 | if (!x) |
1951 | return -ENOMEM; | |
7bbead1d | 1952 | |
4ff361cc | 1953 | path_simplify(x + 5); |
7cae38c4 LP |
1954 | our_env[n_env++] = x; |
1955 | } | |
1956 | ||
1957 | if (username) { | |
b910cc72 | 1958 | x = strjoin("LOGNAME=", username); |
7cae38c4 LP |
1959 | if (!x) |
1960 | return -ENOMEM; | |
1961 | our_env[n_env++] = x; | |
1962 | ||
b910cc72 | 1963 | x = strjoin("USER=", username); |
7cae38c4 LP |
1964 | if (!x) |
1965 | return -ENOMEM; | |
1966 | our_env[n_env++] = x; | |
1967 | } | |
1968 | ||
1969 | if (shell) { | |
b910cc72 | 1970 | x = strjoin("SHELL=", shell); |
7cae38c4 LP |
1971 | if (!x) |
1972 | return -ENOMEM; | |
7bbead1d | 1973 | |
4ff361cc | 1974 | path_simplify(x + 6); |
7cae38c4 LP |
1975 | our_env[n_env++] = x; |
1976 | } | |
1977 | ||
4b58153d LP |
1978 | if (!sd_id128_is_null(u->invocation_id)) { |
1979 | if (asprintf(&x, "INVOCATION_ID=" SD_ID128_FORMAT_STR, SD_ID128_FORMAT_VAL(u->invocation_id)) < 0) | |
1980 | return -ENOMEM; | |
1981 | ||
1982 | our_env[n_env++] = x; | |
1983 | } | |
1984 | ||
6af760f3 | 1985 | if (exec_context_needs_term(c)) { |
4d62ee55 | 1986 | _cleanup_free_ char *cmdline = NULL; |
6af760f3 LP |
1987 | const char *tty_path, *term = NULL; |
1988 | ||
1989 | tty_path = exec_context_tty_path(c); | |
1990 | ||
e8cf09b2 LP |
1991 | /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try |
1992 | * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the | |
1993 | * container manager passes to PID 1 ends up all the way in the console login shown. */ | |
6af760f3 | 1994 | |
e8cf09b2 | 1995 | if (path_equal_ptr(tty_path, "/dev/console") && getppid() == 1) |
6af760f3 | 1996 | term = getenv("TERM"); |
4d62ee55 DDM |
1997 | else if (tty_path && in_charset(skip_dev_prefix(tty_path), ALPHANUMERICAL)) { |
1998 | _cleanup_free_ char *key = NULL; | |
1999 | ||
2000 | key = strjoin("systemd.tty.term.", skip_dev_prefix(tty_path)); | |
2001 | if (!key) | |
2002 | return -ENOMEM; | |
2003 | ||
2004 | r = proc_cmdline_get_key(key, 0, &cmdline); | |
2005 | if (r < 0) | |
2006 | log_debug_errno(r, "Failed to read %s from kernel cmdline, ignoring: %m", key); | |
2007 | else if (r > 0) | |
2008 | term = cmdline; | |
2009 | } | |
e8cf09b2 | 2010 | |
6af760f3 LP |
2011 | if (!term) |
2012 | term = default_term_for_tty(tty_path); | |
7cae38c4 | 2013 | |
b910cc72 | 2014 | x = strjoin("TERM=", term); |
7cae38c4 LP |
2015 | if (!x) |
2016 | return -ENOMEM; | |
2017 | our_env[n_env++] = x; | |
2018 | } | |
2019 | ||
7bce046b LP |
2020 | if (journal_stream_dev != 0 && journal_stream_ino != 0) { |
2021 | if (asprintf(&x, "JOURNAL_STREAM=" DEV_FMT ":" INO_FMT, journal_stream_dev, journal_stream_ino) < 0) | |
2022 | return -ENOMEM; | |
2023 | ||
2024 | our_env[n_env++] = x; | |
2025 | } | |
2026 | ||
91dd5f7c LP |
2027 | if (c->log_namespace) { |
2028 | x = strjoin("LOG_NAMESPACE=", c->log_namespace); | |
2029 | if (!x) | |
2030 | return -ENOMEM; | |
2031 | ||
2032 | our_env[n_env++] = x; | |
2033 | } | |
2034 | ||
5b10116e | 2035 | for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) { |
211a3d87 | 2036 | _cleanup_free_ char *joined = NULL; |
fb2042dd YW |
2037 | const char *n; |
2038 | ||
2039 | if (!p->prefix[t]) | |
2040 | continue; | |
2041 | ||
211a3d87 | 2042 | if (c->directories[t].n_items == 0) |
fb2042dd YW |
2043 | continue; |
2044 | ||
2045 | n = exec_directory_env_name_to_string(t); | |
2046 | if (!n) | |
2047 | continue; | |
2048 | ||
211a3d87 LB |
2049 | for (size_t i = 0; i < c->directories[t].n_items; i++) { |
2050 | _cleanup_free_ char *prefixed = NULL; | |
fb2042dd | 2051 | |
211a3d87 LB |
2052 | prefixed = path_join(p->prefix[t], c->directories[t].items[i].path); |
2053 | if (!prefixed) | |
2054 | return -ENOMEM; | |
2055 | ||
2056 | if (!strextend_with_separator(&joined, ":", prefixed)) | |
2057 | return -ENOMEM; | |
2058 | } | |
fb2042dd YW |
2059 | |
2060 | x = strjoin(n, "=", joined); | |
2061 | if (!x) | |
2062 | return -ENOMEM; | |
2063 | ||
2064 | our_env[n_env++] = x; | |
2065 | } | |
2066 | ||
bb0c0d6f LP |
2067 | if (exec_context_has_credentials(c) && p->prefix[EXEC_DIRECTORY_RUNTIME]) { |
2068 | x = strjoin("CREDENTIALS_DIRECTORY=", p->prefix[EXEC_DIRECTORY_RUNTIME], "/credentials/", u->id); | |
2069 | if (!x) | |
2070 | return -ENOMEM; | |
2071 | ||
2072 | our_env[n_env++] = x; | |
2073 | } | |
2074 | ||
dc4e2940 YW |
2075 | if (asprintf(&x, "SYSTEMD_EXEC_PID=" PID_FMT, getpid_cached()) < 0) |
2076 | return -ENOMEM; | |
2077 | ||
2078 | our_env[n_env++] = x; | |
2079 | ||
6bb00842 LP |
2080 | if (memory_pressure_path) { |
2081 | x = strjoin("MEMORY_PRESSURE_WATCH=", memory_pressure_path); | |
2082 | if (!x) | |
2083 | return -ENOMEM; | |
2084 | ||
2085 | our_env[n_env++] = x; | |
2086 | ||
2087 | if (cgroup_context && !path_equal(memory_pressure_path, "/dev/null")) { | |
2088 | _cleanup_free_ char *b = NULL, *e = NULL; | |
2089 | ||
2090 | if (asprintf(&b, "%s " USEC_FMT " " USEC_FMT, | |
2091 | MEMORY_PRESSURE_DEFAULT_TYPE, | |
2092 | cgroup_context->memory_pressure_threshold_usec == USEC_INFINITY ? MEMORY_PRESSURE_DEFAULT_THRESHOLD_USEC : | |
2093 | CLAMP(cgroup_context->memory_pressure_threshold_usec, 1U, MEMORY_PRESSURE_DEFAULT_WINDOW_USEC), | |
2094 | MEMORY_PRESSURE_DEFAULT_WINDOW_USEC) < 0) | |
2095 | return -ENOMEM; | |
2096 | ||
2097 | if (base64mem(b, strlen(b) + 1, &e) < 0) | |
2098 | return -ENOMEM; | |
2099 | ||
2100 | x = strjoin("MEMORY_PRESSURE_WRITE=", e); | |
2101 | if (!x) | |
2102 | return -ENOMEM; | |
2103 | ||
2104 | our_env[n_env++] = x; | |
2105 | } | |
2106 | } | |
2107 | ||
2108 | assert(n_env < N_ENV_VARS + _EXEC_DIRECTORY_TYPE_MAX); | |
8d5bb13d | 2109 | #undef N_ENV_VARS |
7cae38c4 | 2110 | |
ae2a15bc | 2111 | *ret = TAKE_PTR(our_env); |
7cae38c4 LP |
2112 | |
2113 | return 0; | |
2114 | } | |
2115 | ||
b4c14404 FB |
2116 | static int build_pass_environment(const ExecContext *c, char ***ret) { |
2117 | _cleanup_strv_free_ char **pass_env = NULL; | |
319a4f4b | 2118 | size_t n_env = 0; |
b4c14404 FB |
2119 | |
2120 | STRV_FOREACH(i, c->pass_environment) { | |
2121 | _cleanup_free_ char *x = NULL; | |
2122 | char *v; | |
2123 | ||
2124 | v = getenv(*i); | |
2125 | if (!v) | |
2126 | continue; | |
605405c6 | 2127 | x = strjoin(*i, "=", v); |
b4c14404 FB |
2128 | if (!x) |
2129 | return -ENOMEM; | |
00819cc1 | 2130 | |
319a4f4b | 2131 | if (!GREEDY_REALLOC(pass_env, n_env + 2)) |
b4c14404 | 2132 | return -ENOMEM; |
00819cc1 | 2133 | |
1cc6c93a | 2134 | pass_env[n_env++] = TAKE_PTR(x); |
b4c14404 | 2135 | pass_env[n_env] = NULL; |
b4c14404 FB |
2136 | } |
2137 | ||
ae2a15bc | 2138 | *ret = TAKE_PTR(pass_env); |
b4c14404 FB |
2139 | |
2140 | return 0; | |
2141 | } | |
2142 | ||
fbbb9697 YW |
2143 | bool exec_needs_network_namespace(const ExecContext *context) { |
2144 | assert(context); | |
2145 | ||
2146 | return context->private_network || context->network_namespace_path; | |
2147 | } | |
2148 | ||
9c0c6701 DDM |
2149 | static bool exec_needs_ephemeral(const ExecContext *context) { |
2150 | return (context->root_image || context->root_directory) && context->root_ephemeral; | |
2151 | } | |
2152 | ||
fde36d25 YW |
2153 | static bool exec_needs_ipc_namespace(const ExecContext *context) { |
2154 | assert(context); | |
2155 | ||
2156 | return context->private_ipc || context->ipc_namespace_path; | |
2157 | } | |
2158 | ||
5e8deb94 | 2159 | bool exec_needs_mount_namespace( |
8b44a3d2 LP |
2160 | const ExecContext *context, |
2161 | const ExecParameters *params, | |
28135da3 | 2162 | const ExecRuntime *runtime) { |
8b44a3d2 LP |
2163 | |
2164 | assert(context); | |
8b44a3d2 | 2165 | |
915e6d16 LP |
2166 | if (context->root_image) |
2167 | return true; | |
2168 | ||
2a624c36 AP |
2169 | if (!strv_isempty(context->read_write_paths) || |
2170 | !strv_isempty(context->read_only_paths) || | |
ddc155b2 TM |
2171 | !strv_isempty(context->inaccessible_paths) || |
2172 | !strv_isempty(context->exec_paths) || | |
2173 | !strv_isempty(context->no_exec_paths)) | |
8b44a3d2 LP |
2174 | return true; |
2175 | ||
42b1d8e0 | 2176 | if (context->n_bind_mounts > 0) |
d2d6c096 LP |
2177 | return true; |
2178 | ||
2abd4e38 YW |
2179 | if (context->n_temporary_filesystems > 0) |
2180 | return true; | |
2181 | ||
b3d13314 LB |
2182 | if (context->n_mount_images > 0) |
2183 | return true; | |
2184 | ||
93f59701 LB |
2185 | if (context->n_extension_images > 0) |
2186 | return true; | |
2187 | ||
a07b9926 LB |
2188 | if (!strv_isempty(context->extension_directories)) |
2189 | return true; | |
2190 | ||
874cdcbc | 2191 | if (!IN_SET(context->mount_propagation_flag, 0, MS_SHARED)) |
8b44a3d2 LP |
2192 | return true; |
2193 | ||
28135da3 | 2194 | if (context->private_tmp && runtime && runtime->shared && (runtime->shared->tmp_dir || runtime->shared->var_tmp_dir)) |
8b44a3d2 LP |
2195 | return true; |
2196 | ||
8b44a3d2 | 2197 | if (context->private_devices || |
24002121 | 2198 | context->private_mounts > 0 || |
c2da3bf2 | 2199 | (context->private_mounts < 0 && exec_needs_network_namespace(context)) || |
8b44a3d2 | 2200 | context->protect_system != PROTECT_SYSTEM_NO || |
59eeb84b LP |
2201 | context->protect_home != PROTECT_HOME_NO || |
2202 | context->protect_kernel_tunables || | |
c575770b | 2203 | context->protect_kernel_modules || |
94a7b275 | 2204 | context->protect_kernel_logs || |
4e399953 LP |
2205 | context->protect_control_groups || |
2206 | context->protect_proc != PROTECT_PROC_DEFAULT || | |
80271a44 | 2207 | context->proc_subset != PROC_SUBSET_ALL || |
fde36d25 | 2208 | exec_needs_ipc_namespace(context)) |
8b44a3d2 LP |
2209 | return true; |
2210 | ||
37c56f89 | 2211 | if (context->root_directory) { |
5e98086d | 2212 | if (exec_context_get_effective_mount_apivfs(context)) |
37c56f89 YW |
2213 | return true; |
2214 | ||
5b10116e | 2215 | for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) { |
5e8deb94 | 2216 | if (params && !params->prefix[t]) |
37c56f89 YW |
2217 | continue; |
2218 | ||
211a3d87 | 2219 | if (context->directories[t].n_items > 0) |
37c56f89 YW |
2220 | return true; |
2221 | } | |
2222 | } | |
5d997827 | 2223 | |
42b1d8e0 | 2224 | if (context->dynamic_user && |
211a3d87 LB |
2225 | (context->directories[EXEC_DIRECTORY_STATE].n_items > 0 || |
2226 | context->directories[EXEC_DIRECTORY_CACHE].n_items > 0 || | |
2227 | context->directories[EXEC_DIRECTORY_LOGS].n_items > 0)) | |
42b1d8e0 YW |
2228 | return true; |
2229 | ||
91dd5f7c LP |
2230 | if (context->log_namespace) |
2231 | return true; | |
2232 | ||
8b44a3d2 LP |
2233 | return false; |
2234 | } | |
2235 | ||
5749f855 | 2236 | static int setup_private_users(uid_t ouid, gid_t ogid, uid_t uid, gid_t gid) { |
d251207d | 2237 | _cleanup_free_ char *uid_map = NULL, *gid_map = NULL; |
19ee48a6 | 2238 | _cleanup_close_pair_ int errno_pipe[2] = PIPE_EBADF; |
254d1313 | 2239 | _cleanup_close_ int unshare_ready_fd = -EBADF; |
d251207d LP |
2240 | _cleanup_(sigkill_waitp) pid_t pid = 0; |
2241 | uint64_t c = 1; | |
d251207d LP |
2242 | ssize_t n; |
2243 | int r; | |
2244 | ||
5749f855 AZ |
2245 | /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e. |
2246 | * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to | |
d251207d LP |
2247 | * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which |
2248 | * we however lack after opening the user namespace. To work around this we fork() a temporary child process, | |
2249 | * which waits for the parent to create the new user namespace while staying in the original namespace. The | |
2250 | * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and | |
5749f855 AZ |
2251 | * continues execution normally. |
2252 | * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it | |
2253 | * does not need CAP_SETUID to write the single line mapping to itself. */ | |
d251207d | 2254 | |
5749f855 | 2255 | /* Can only set up multiple mappings with CAP_SETUID. */ |
26c45a6c | 2256 | if (have_effective_cap(CAP_SETUID) > 0 && uid != ouid && uid_is_valid(uid)) |
587ab01b | 2257 | r = asprintf(&uid_map, |
5749f855 | 2258 | UID_FMT " " UID_FMT " 1\n" /* Map $OUID → $OUID */ |
587ab01b | 2259 | UID_FMT " " UID_FMT " 1\n", /* Map $UID → $UID */ |
5749f855 AZ |
2260 | ouid, ouid, uid, uid); |
2261 | else | |
2262 | r = asprintf(&uid_map, | |
2263 | UID_FMT " " UID_FMT " 1\n", /* Map $OUID → $OUID */ | |
2264 | ouid, ouid); | |
d251207d | 2265 | |
5749f855 AZ |
2266 | if (r < 0) |
2267 | return -ENOMEM; | |
2268 | ||
2269 | /* Can only set up multiple mappings with CAP_SETGID. */ | |
26c45a6c | 2270 | if (have_effective_cap(CAP_SETGID) > 0 && gid != ogid && gid_is_valid(gid)) |
587ab01b | 2271 | r = asprintf(&gid_map, |
5749f855 | 2272 | GID_FMT " " GID_FMT " 1\n" /* Map $OGID → $OGID */ |
587ab01b | 2273 | GID_FMT " " GID_FMT " 1\n", /* Map $GID → $GID */ |
5749f855 AZ |
2274 | ogid, ogid, gid, gid); |
2275 | else | |
2276 | r = asprintf(&gid_map, | |
2277 | GID_FMT " " GID_FMT " 1\n", /* Map $OGID -> $OGID */ | |
2278 | ogid, ogid); | |
2279 | ||
2280 | if (r < 0) | |
2281 | return -ENOMEM; | |
d251207d LP |
2282 | |
2283 | /* Create a communication channel so that the parent can tell the child when it finished creating the user | |
2284 | * namespace. */ | |
2285 | unshare_ready_fd = eventfd(0, EFD_CLOEXEC); | |
2286 | if (unshare_ready_fd < 0) | |
2287 | return -errno; | |
2288 | ||
2289 | /* Create a communication channel so that the child can tell the parent a proper error code in case it | |
2290 | * failed. */ | |
2291 | if (pipe2(errno_pipe, O_CLOEXEC) < 0) | |
2292 | return -errno; | |
2293 | ||
4c253ed1 LP |
2294 | r = safe_fork("(sd-userns)", FORK_RESET_SIGNALS|FORK_DEATHSIG, &pid); |
2295 | if (r < 0) | |
2296 | return r; | |
2297 | if (r == 0) { | |
254d1313 | 2298 | _cleanup_close_ int fd = -EBADF; |
d251207d LP |
2299 | const char *a; |
2300 | pid_t ppid; | |
2301 | ||
2302 | /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from | |
2303 | * here, after the parent opened its own user namespace. */ | |
2304 | ||
2305 | ppid = getppid(); | |
2306 | errno_pipe[0] = safe_close(errno_pipe[0]); | |
2307 | ||
2308 | /* Wait until the parent unshared the user namespace */ | |
2309 | if (read(unshare_ready_fd, &c, sizeof(c)) < 0) { | |
2310 | r = -errno; | |
2311 | goto child_fail; | |
2312 | } | |
2313 | ||
2314 | /* Disable the setgroups() system call in the child user namespace, for good. */ | |
2315 | a = procfs_file_alloca(ppid, "setgroups"); | |
2316 | fd = open(a, O_WRONLY|O_CLOEXEC); | |
2317 | if (fd < 0) { | |
2318 | if (errno != ENOENT) { | |
2319 | r = -errno; | |
2320 | goto child_fail; | |
2321 | } | |
2322 | ||
2323 | /* If the file is missing the kernel is too old, let's continue anyway. */ | |
2324 | } else { | |
2325 | if (write(fd, "deny\n", 5) < 0) { | |
2326 | r = -errno; | |
2327 | goto child_fail; | |
2328 | } | |
2329 | ||
2330 | fd = safe_close(fd); | |
2331 | } | |
2332 | ||
2333 | /* First write the GID map */ | |
2334 | a = procfs_file_alloca(ppid, "gid_map"); | |
2335 | fd = open(a, O_WRONLY|O_CLOEXEC); | |
2336 | if (fd < 0) { | |
2337 | r = -errno; | |
2338 | goto child_fail; | |
2339 | } | |
2340 | if (write(fd, gid_map, strlen(gid_map)) < 0) { | |
2341 | r = -errno; | |
2342 | goto child_fail; | |
2343 | } | |
2344 | fd = safe_close(fd); | |
2345 | ||
2346 | /* The write the UID map */ | |
2347 | a = procfs_file_alloca(ppid, "uid_map"); | |
2348 | fd = open(a, O_WRONLY|O_CLOEXEC); | |
2349 | if (fd < 0) { | |
2350 | r = -errno; | |
2351 | goto child_fail; | |
2352 | } | |
2353 | if (write(fd, uid_map, strlen(uid_map)) < 0) { | |
2354 | r = -errno; | |
2355 | goto child_fail; | |
2356 | } | |
2357 | ||
2358 | _exit(EXIT_SUCCESS); | |
2359 | ||
2360 | child_fail: | |
2361 | (void) write(errno_pipe[1], &r, sizeof(r)); | |
2362 | _exit(EXIT_FAILURE); | |
2363 | } | |
2364 | ||
2365 | errno_pipe[1] = safe_close(errno_pipe[1]); | |
2366 | ||
2367 | if (unshare(CLONE_NEWUSER) < 0) | |
2368 | return -errno; | |
2369 | ||
2370 | /* Let the child know that the namespace is ready now */ | |
2371 | if (write(unshare_ready_fd, &c, sizeof(c)) < 0) | |
2372 | return -errno; | |
2373 | ||
2374 | /* Try to read an error code from the child */ | |
2375 | n = read(errno_pipe[0], &r, sizeof(r)); | |
2376 | if (n < 0) | |
2377 | return -errno; | |
2378 | if (n == sizeof(r)) { /* an error code was sent to us */ | |
2379 | if (r < 0) | |
2380 | return r; | |
2381 | return -EIO; | |
2382 | } | |
2383 | if (n != 0) /* on success we should have read 0 bytes */ | |
2384 | return -EIO; | |
2385 | ||
8f03de53 | 2386 | r = wait_for_terminate_and_check("(sd-userns)", TAKE_PID(pid), 0); |
d251207d LP |
2387 | if (r < 0) |
2388 | return r; | |
2e87a1fd | 2389 | if (r != EXIT_SUCCESS) /* If something strange happened with the child, let's consider this fatal, too */ |
d251207d LP |
2390 | return -EIO; |
2391 | ||
2392 | return 0; | |
2393 | } | |
2394 | ||
494d0247 | 2395 | static bool exec_directory_is_private(const ExecContext *context, ExecDirectoryType type) { |
e43911a7 LP |
2396 | assert(context); |
2397 | ||
494d0247 YW |
2398 | if (!context->dynamic_user) |
2399 | return false; | |
2400 | ||
2401 | if (type == EXEC_DIRECTORY_CONFIGURATION) | |
2402 | return false; | |
2403 | ||
2404 | if (type == EXEC_DIRECTORY_RUNTIME && context->runtime_directory_preserve_mode == EXEC_PRESERVE_NO) | |
2405 | return false; | |
2406 | ||
2407 | return true; | |
2408 | } | |
2409 | ||
211a3d87 LB |
2410 | static int create_many_symlinks(const char *root, const char *source, char **symlinks) { |
2411 | _cleanup_free_ char *src_abs = NULL; | |
211a3d87 LB |
2412 | int r; |
2413 | ||
2414 | assert(source); | |
2415 | ||
2416 | src_abs = path_join(root, source); | |
2417 | if (!src_abs) | |
2418 | return -ENOMEM; | |
2419 | ||
2420 | STRV_FOREACH(dst, symlinks) { | |
2421 | _cleanup_free_ char *dst_abs = NULL; | |
2422 | ||
2423 | dst_abs = path_join(root, *dst); | |
2424 | if (!dst_abs) | |
2425 | return -ENOMEM; | |
2426 | ||
2427 | r = mkdir_parents_label(dst_abs, 0755); | |
2428 | if (r < 0) | |
2429 | return r; | |
2430 | ||
2431 | r = symlink_idempotent(src_abs, dst_abs, true); | |
2432 | if (r < 0) | |
2433 | return r; | |
2434 | } | |
2435 | ||
2436 | return 0; | |
2437 | } | |
2438 | ||
3536f49e | 2439 | static int setup_exec_directory( |
59dd2bbb | 2440 | Unit *u, |
07689d5d LP |
2441 | const ExecContext *context, |
2442 | const ExecParameters *params, | |
2443 | uid_t uid, | |
3536f49e | 2444 | gid_t gid, |
3536f49e | 2445 | ExecDirectoryType type, |
211a3d87 | 2446 | bool needs_mount_namespace, |
3536f49e | 2447 | int *exit_status) { |
07689d5d | 2448 | |
72fd1768 | 2449 | static const int exit_status_table[_EXEC_DIRECTORY_TYPE_MAX] = { |
3536f49e YW |
2450 | [EXEC_DIRECTORY_RUNTIME] = EXIT_RUNTIME_DIRECTORY, |
2451 | [EXEC_DIRECTORY_STATE] = EXIT_STATE_DIRECTORY, | |
2452 | [EXEC_DIRECTORY_CACHE] = EXIT_CACHE_DIRECTORY, | |
2453 | [EXEC_DIRECTORY_LOGS] = EXIT_LOGS_DIRECTORY, | |
2454 | [EXEC_DIRECTORY_CONFIGURATION] = EXIT_CONFIGURATION_DIRECTORY, | |
2455 | }; | |
07689d5d LP |
2456 | int r; |
2457 | ||
2458 | assert(context); | |
2459 | assert(params); | |
72fd1768 | 2460 | assert(type >= 0 && type < _EXEC_DIRECTORY_TYPE_MAX); |
3536f49e | 2461 | assert(exit_status); |
07689d5d | 2462 | |
3536f49e YW |
2463 | if (!params->prefix[type]) |
2464 | return 0; | |
2465 | ||
8679efde | 2466 | if (params->flags & EXEC_CHOWN_DIRECTORIES) { |
3536f49e YW |
2467 | if (!uid_is_valid(uid)) |
2468 | uid = 0; | |
2469 | if (!gid_is_valid(gid)) | |
2470 | gid = 0; | |
2471 | } | |
2472 | ||
211a3d87 | 2473 | for (size_t i = 0; i < context->directories[type].n_items; i++) { |
6c47cd7d | 2474 | _cleanup_free_ char *p = NULL, *pp = NULL; |
07689d5d | 2475 | |
211a3d87 | 2476 | p = path_join(params->prefix[type], context->directories[type].items[i].path); |
3536f49e YW |
2477 | if (!p) { |
2478 | r = -ENOMEM; | |
2479 | goto fail; | |
2480 | } | |
07689d5d | 2481 | |
23a7448e YW |
2482 | r = mkdir_parents_label(p, 0755); |
2483 | if (r < 0) | |
3536f49e | 2484 | goto fail; |
23a7448e | 2485 | |
f9c91932 LP |
2486 | if (IN_SET(type, EXEC_DIRECTORY_STATE, EXEC_DIRECTORY_LOGS) && params->runtime_scope == RUNTIME_SCOPE_USER) { |
2487 | ||
2488 | /* If we are in user mode, and a configuration directory exists but a state directory | |
2489 | * doesn't exist, then we likely are upgrading from an older systemd version that | |
2490 | * didn't know the more recent addition to the xdg-basedir spec: the $XDG_STATE_HOME | |
2491 | * directory. In older systemd versions EXEC_DIRECTORY_STATE was aliased to | |
2492 | * EXEC_DIRECTORY_CONFIGURATION, with the advent of $XDG_STATE_HOME is is now | |
627cdcc7 | 2493 | * separated. If a service has both dirs configured but only the configuration dir |
f9c91932 LP |
2494 | * exists and the state dir does not, we assume we are looking at an update |
2495 | * situation. Hence, create a compatibility symlink, so that all expectations are | |
2496 | * met. | |
2497 | * | |
2498 | * (We also do something similar with the log directory, which still doesn't exist in | |
2499 | * the xdg basedir spec. We'll make it a subdir of the state dir.) */ | |
2500 | ||
2501 | /* this assumes the state dir is always created before the configuration dir */ | |
2502 | assert_cc(EXEC_DIRECTORY_STATE < EXEC_DIRECTORY_LOGS); | |
2503 | assert_cc(EXEC_DIRECTORY_LOGS < EXEC_DIRECTORY_CONFIGURATION); | |
2504 | ||
2505 | r = laccess(p, F_OK); | |
2506 | if (r == -ENOENT) { | |
2507 | _cleanup_free_ char *q = NULL; | |
2508 | ||
2509 | /* OK, we know that the state dir does not exist. Let's see if the dir exists | |
2510 | * under the configuration hierarchy. */ | |
2511 | ||
2512 | if (type == EXEC_DIRECTORY_STATE) | |
2513 | q = path_join(params->prefix[EXEC_DIRECTORY_CONFIGURATION], context->directories[type].items[i].path); | |
2514 | else if (type == EXEC_DIRECTORY_LOGS) | |
2515 | q = path_join(params->prefix[EXEC_DIRECTORY_CONFIGURATION], "log", context->directories[type].items[i].path); | |
2516 | else | |
2517 | assert_not_reached(); | |
2518 | if (!q) { | |
2519 | r = -ENOMEM; | |
2520 | goto fail; | |
2521 | } | |
2522 | ||
2523 | r = laccess(q, F_OK); | |
2524 | if (r >= 0) { | |
2525 | /* It does exist! This hence looks like an update. Symlink the | |
2526 | * configuration directory into the state directory. */ | |
2527 | ||
2528 | r = symlink_idempotent(q, p, /* make_relative= */ true); | |
2529 | if (r < 0) | |
2530 | goto fail; | |
2531 | ||
59dd2bbb | 2532 | log_unit_notice(u, "Unit state directory %s missing but matching configuration directory %s exists, assuming update from systemd 253 or older, creating compatibility symlink.", p, q); |
f9c91932 LP |
2533 | continue; |
2534 | } else if (r != -ENOENT) | |
59dd2bbb | 2535 | log_unit_warning_errno(u, r, "Unable to detect whether unit configuration directory '%s' exists, assuming not: %m", q); |
f9c91932 LP |
2536 | |
2537 | } else if (r < 0) | |
59dd2bbb | 2538 | log_unit_warning_errno(u, r, "Unable to detect whether unit state directory '%s' is missing, assuming it is: %m", p); |
f9c91932 LP |
2539 | } |
2540 | ||
494d0247 | 2541 | if (exec_directory_is_private(context, type)) { |
3f5b1508 LP |
2542 | /* So, here's one extra complication when dealing with DynamicUser=1 units. In that |
2543 | * case we want to avoid leaving a directory around fully accessible that is owned by | |
2544 | * a dynamic user whose UID is later on reused. To lock this down we use the same | |
2545 | * trick used by container managers to prohibit host users to get access to files of | |
2546 | * the same UID in containers: we place everything inside a directory that has an | |
2547 | * access mode of 0700 and is owned root:root, so that it acts as security boundary | |
2548 | * for unprivileged host code. We then use fs namespacing to make this directory | |
2549 | * permeable for the service itself. | |
6c47cd7d | 2550 | * |
3f5b1508 LP |
2551 | * Specifically: for a service which wants a special directory "foo/" we first create |
2552 | * a directory "private/" with access mode 0700 owned by root:root. Then we place | |
2553 | * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to | |
2554 | * "private/foo". This way, privileged host users can access "foo/" as usual, but | |
2555 | * unprivileged host users can't look into it. Inside of the namespace of the unit | |
2556 | * "private/" is replaced by a more liberally accessible tmpfs, into which the host's | |
2557 | * "private/foo/" is mounted under the same name, thus disabling the access boundary | |
2558 | * for the service and making sure it only gets access to the dirs it needs but no | |
2559 | * others. Tricky? Yes, absolutely, but it works! | |
6c47cd7d | 2560 | * |
3f5b1508 LP |
2561 | * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not |
2562 | * to be owned by the service itself. | |
2563 | * | |
2564 | * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used | |
2565 | * for sharing files or sockets with other services. */ | |
6c47cd7d | 2566 | |
4ede9802 LP |
2567 | pp = path_join(params->prefix[type], "private"); |
2568 | if (!pp) { | |
6c47cd7d LP |
2569 | r = -ENOMEM; |
2570 | goto fail; | |
2571 | } | |
2572 | ||
2573 | /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */ | |
4ede9802 | 2574 | r = mkdir_safe_label(pp, 0700, 0, 0, MKDIR_WARN_MODE); |
6c47cd7d LP |
2575 | if (r < 0) |
2576 | goto fail; | |
2577 | ||
211a3d87 | 2578 | if (!path_extend(&pp, context->directories[type].items[i].path)) { |
6c47cd7d LP |
2579 | r = -ENOMEM; |
2580 | goto fail; | |
2581 | } | |
2582 | ||
2583 | /* Create all directories between the configured directory and this private root, and mark them 0755 */ | |
2584 | r = mkdir_parents_label(pp, 0755); | |
2585 | if (r < 0) | |
2586 | goto fail; | |
2587 | ||
949befd3 | 2588 | if (is_dir(p, false) > 0 && |
b93d24e0 | 2589 | (laccess(pp, F_OK) == -ENOENT)) { |
949befd3 LP |
2590 | |
2591 | /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move | |
2592 | * it over. Most likely the service has been upgraded from one that didn't use | |
2593 | * DynamicUser=1, to one that does. */ | |
2594 | ||
59dd2bbb LP |
2595 | log_unit_info(u, "Found pre-existing public %s= directory %s, migrating to %s.\n" |
2596 | "Apparently, service previously had DynamicUser= turned off, and has now turned it on.", | |
2597 | exec_directory_type_to_string(type), p, pp); | |
cf52c45d | 2598 | |
db58f5de LP |
2599 | r = RET_NERRNO(rename(p, pp)); |
2600 | if (r < 0) | |
949befd3 | 2601 | goto fail; |
949befd3 LP |
2602 | } else { |
2603 | /* Otherwise, create the actual directory for the service */ | |
2604 | ||
2605 | r = mkdir_label(pp, context->directories[type].mode); | |
2606 | if (r < 0 && r != -EEXIST) | |
2607 | goto fail; | |
2608 | } | |
6c47cd7d | 2609 | |
a2ab603c YW |
2610 | if (!context->directories[type].items[i].only_create) { |
2611 | /* And link it up from the original place. | |
2612 | * Notes | |
2613 | * 1) If a mount namespace is going to be used, then this symlink remains on | |
2614 | * the host, and a new one for the child namespace will be created later. | |
2615 | * 2) It is not necessary to create this symlink when one of its parent | |
2616 | * directories is specified and already created. E.g. | |
2617 | * StateDirectory=foo foo/bar | |
2618 | * In that case, the inode points to pp and p for "foo/bar" are the same: | |
2619 | * pp = "/var/lib/private/foo/bar" | |
2620 | * p = "/var/lib/foo/bar" | |
2621 | * and, /var/lib/foo is a symlink to /var/lib/private/foo. So, not only | |
2622 | * we do not need to create the symlink, but we cannot create the symlink. | |
2623 | * See issue #24783. */ | |
2624 | r = symlink_idempotent(pp, p, true); | |
2625 | if (r < 0) | |
2626 | goto fail; | |
2627 | } | |
6c47cd7d | 2628 | |
6c47cd7d | 2629 | } else { |
5c6d40d1 LP |
2630 | _cleanup_free_ char *target = NULL; |
2631 | ||
2632 | if (type != EXEC_DIRECTORY_CONFIGURATION && | |
2633 | readlink_and_make_absolute(p, &target) >= 0) { | |
578dc69f | 2634 | _cleanup_free_ char *q = NULL, *q_resolved = NULL, *target_resolved = NULL; |
5c6d40d1 LP |
2635 | |
2636 | /* This already exists and is a symlink? Interesting. Maybe it's one created | |
2193f17c LP |
2637 | * by DynamicUser=1 (see above)? |
2638 | * | |
2639 | * We do this for all directory types except for ConfigurationDirectory=, | |
2640 | * since they all support the private/ symlink logic at least in some | |
2641 | * configurations, see above. */ | |
5c6d40d1 | 2642 | |
f461a28d | 2643 | r = chase(target, NULL, 0, &target_resolved, NULL); |
578dc69f YW |
2644 | if (r < 0) |
2645 | goto fail; | |
2646 | ||
211a3d87 | 2647 | q = path_join(params->prefix[type], "private", context->directories[type].items[i].path); |
5c6d40d1 LP |
2648 | if (!q) { |
2649 | r = -ENOMEM; | |
2650 | goto fail; | |
2651 | } | |
2652 | ||
578dc69f | 2653 | /* /var/lib or friends may be symlinks. So, let's chase them also. */ |
f461a28d | 2654 | r = chase(q, NULL, CHASE_NONEXISTENT, &q_resolved, NULL); |
578dc69f YW |
2655 | if (r < 0) |
2656 | goto fail; | |
2657 | ||
2658 | if (path_equal(q_resolved, target_resolved)) { | |
5c6d40d1 LP |
2659 | |
2660 | /* Hmm, apparently DynamicUser= was once turned on for this service, | |
2661 | * but is no longer. Let's move the directory back up. */ | |
2662 | ||
59dd2bbb LP |
2663 | log_unit_info(u, "Found pre-existing private %s= directory %s, migrating to %s.\n" |
2664 | "Apparently, service previously had DynamicUser= turned on, and has now turned it off.", | |
2665 | exec_directory_type_to_string(type), q, p); | |
cf52c45d | 2666 | |
db58f5de LP |
2667 | r = RET_NERRNO(unlink(p)); |
2668 | if (r < 0) | |
5c6d40d1 | 2669 | goto fail; |
5c6d40d1 | 2670 | |
db58f5de LP |
2671 | r = RET_NERRNO(rename(q, p)); |
2672 | if (r < 0) | |
5c6d40d1 | 2673 | goto fail; |
5c6d40d1 LP |
2674 | } |
2675 | } | |
2676 | ||
6c47cd7d | 2677 | r = mkdir_label(p, context->directories[type].mode); |
d484580c | 2678 | if (r < 0) { |
d484580c LP |
2679 | if (r != -EEXIST) |
2680 | goto fail; | |
2681 | ||
206e9864 LP |
2682 | if (type == EXEC_DIRECTORY_CONFIGURATION) { |
2683 | struct stat st; | |
2684 | ||
2685 | /* Don't change the owner/access mode of the configuration directory, | |
2686 | * as in the common case it is not written to by a service, and shall | |
2687 | * not be writable. */ | |
2688 | ||
db58f5de LP |
2689 | r = RET_NERRNO(stat(p, &st)); |
2690 | if (r < 0) | |
206e9864 | 2691 | goto fail; |
206e9864 LP |
2692 | |
2693 | /* Still complain if the access mode doesn't match */ | |
2694 | if (((st.st_mode ^ context->directories[type].mode) & 07777) != 0) | |
59dd2bbb LP |
2695 | log_unit_warning(u, "%s \'%s\' already exists but the mode is different. " |
2696 | "(File system: %o %sMode: %o)", | |
2697 | exec_directory_type_to_string(type), context->directories[type].items[i].path, | |
2698 | st.st_mode & 07777, exec_directory_type_to_string(type), context->directories[type].mode & 07777); | |
206e9864 | 2699 | |
6cff72eb | 2700 | continue; |
206e9864 | 2701 | } |
6cff72eb | 2702 | } |
a1164ae3 | 2703 | } |
07689d5d | 2704 | |
206e9864 | 2705 | /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't |
5238e957 | 2706 | * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the |
206e9864 LP |
2707 | * current UID/GID ownership.) */ |
2708 | r = chmod_and_chown(pp ?: p, context->directories[type].mode, UID_INVALID, GID_INVALID); | |
2709 | if (r < 0) | |
2710 | goto fail; | |
c71b2eb7 | 2711 | |
f5bb36dc LP |
2712 | /* Skip the rest (which deals with ownership) in user mode, since ownership changes are not |
2713 | * available to user code anyway */ | |
2714 | if (params->runtime_scope != RUNTIME_SCOPE_SYSTEM) | |
2715 | continue; | |
2716 | ||
607b358e LP |
2717 | /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we |
2718 | * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID | |
7802194a | 2719 | * assignments to exist. */ |
d5602c16 | 2720 | r = path_chown_recursive(pp ?: p, uid, gid, context->dynamic_user ? 01777 : 07777, AT_SYMLINK_FOLLOW); |
07689d5d | 2721 | if (r < 0) |
3536f49e | 2722 | goto fail; |
07689d5d LP |
2723 | } |
2724 | ||
211a3d87 LB |
2725 | /* If we are not going to run in a namespace, set up the symlinks - otherwise |
2726 | * they are set up later, to allow configuring empty var/run/etc. */ | |
2727 | if (!needs_mount_namespace) | |
2728 | for (size_t i = 0; i < context->directories[type].n_items; i++) { | |
2729 | r = create_many_symlinks(params->prefix[type], | |
2730 | context->directories[type].items[i].path, | |
2731 | context->directories[type].items[i].symlinks); | |
2732 | if (r < 0) | |
2733 | goto fail; | |
2734 | } | |
2735 | ||
07689d5d | 2736 | return 0; |
3536f49e YW |
2737 | |
2738 | fail: | |
2739 | *exit_status = exit_status_table[type]; | |
3536f49e | 2740 | return r; |
07689d5d LP |
2741 | } |
2742 | ||
bb0c0d6f LP |
2743 | static int write_credential( |
2744 | int dfd, | |
2745 | const char *id, | |
2746 | const void *data, | |
2747 | size_t size, | |
2748 | uid_t uid, | |
3a78b0e9 | 2749 | gid_t gid, |
bb0c0d6f LP |
2750 | bool ownership_ok) { |
2751 | ||
2752 | _cleanup_(unlink_and_freep) char *tmp = NULL; | |
254d1313 | 2753 | _cleanup_close_ int fd = -EBADF; |
bb0c0d6f LP |
2754 | int r; |
2755 | ||
2756 | r = tempfn_random_child("", "cred", &tmp); | |
2757 | if (r < 0) | |
2758 | return r; | |
2759 | ||
2760 | fd = openat(dfd, tmp, O_CREAT|O_RDWR|O_CLOEXEC|O_EXCL|O_NOFOLLOW|O_NOCTTY, 0600); | |
2761 | if (fd < 0) { | |
2762 | tmp = mfree(tmp); | |
2763 | return -errno; | |
2764 | } | |
2765 | ||
43144be4 | 2766 | r = loop_write(fd, data, size, /* do_poll = */ false); |
bb0c0d6f LP |
2767 | if (r < 0) |
2768 | return r; | |
2769 | ||
2770 | if (fchmod(fd, 0400) < 0) /* Take away "w" bit */ | |
2771 | return -errno; | |
2772 | ||
2773 | if (uid_is_valid(uid) && uid != getuid()) { | |
567aeb58 | 2774 | r = fd_add_uid_acl_permission(fd, uid, ACL_READ); |
bb0c0d6f LP |
2775 | if (r < 0) { |
2776 | if (!ERRNO_IS_NOT_SUPPORTED(r) && !ERRNO_IS_PRIVILEGE(r)) | |
2777 | return r; | |
2778 | ||
2779 | if (!ownership_ok) /* Ideally we use ACLs, since we can neatly express what we want | |
2780 | * to express: that the user gets read access and nothing | |
2781 | * else. But if the backing fs can't support that (e.g. ramfs) | |
2782 | * then we can use file ownership instead. But that's only safe if | |
2783 | * we can then re-mount the whole thing read-only, so that the | |
2784 | * user can no longer chmod() the file to gain write access. */ | |
2785 | return r; | |
2786 | ||
3a78b0e9 | 2787 | if (fchown(fd, uid, gid) < 0) |
bb0c0d6f LP |
2788 | return -errno; |
2789 | } | |
2790 | } | |
2791 | ||
2792 | if (renameat(dfd, tmp, dfd, id) < 0) | |
2793 | return -errno; | |
2794 | ||
2795 | tmp = mfree(tmp); | |
2796 | return 0; | |
2797 | } | |
2798 | ||
96df2bd8 DDM |
2799 | typedef enum CredentialSearchPath { |
2800 | CREDENTIAL_SEARCH_PATH_TRUSTED, | |
2801 | CREDENTIAL_SEARCH_PATH_ENCRYPTED, | |
2802 | CREDENTIAL_SEARCH_PATH_ALL, | |
2803 | _CREDENTIAL_SEARCH_PATH_MAX, | |
2804 | _CREDENTIAL_SEARCH_PATH_INVALID = -EINVAL, | |
2805 | } CredentialSearchPath; | |
2806 | ||
2807 | static char **credential_search_path(const ExecParameters *params, CredentialSearchPath path) { | |
2ad591a3 LP |
2808 | |
2809 | _cleanup_strv_free_ char **l = NULL; | |
2810 | ||
2811 | assert(params); | |
96df2bd8 | 2812 | assert(path >= 0 && path < _CREDENTIAL_SEARCH_PATH_MAX); |
2ad591a3 | 2813 | |
96df2bd8 DDM |
2814 | /* Assemble a search path to find credentials in. For non-encrypted credentials, We'll look in |
2815 | * /etc/credstore/ (and similar directories in /usr/lib/ + /run/). If we're looking for encrypted | |
2816 | * credentials, we'll look in /etc/credstore.encrypted/ (and similar dirs). */ | |
2ad591a3 | 2817 | |
96df2bd8 | 2818 | if (IN_SET(path, CREDENTIAL_SEARCH_PATH_ENCRYPTED, CREDENTIAL_SEARCH_PATH_ALL)) { |
2ad591a3 LP |
2819 | if (strv_extend(&l, params->received_encrypted_credentials_directory) < 0) |
2820 | return NULL; | |
2821 | ||
2822 | if (strv_extend_strv(&l, CONF_PATHS_STRV("credstore.encrypted"), /* filter_duplicates= */ true) < 0) | |
2823 | return NULL; | |
2824 | } | |
2825 | ||
96df2bd8 DDM |
2826 | if (IN_SET(path, CREDENTIAL_SEARCH_PATH_TRUSTED, CREDENTIAL_SEARCH_PATH_ALL)) { |
2827 | if (params->received_credentials_directory) | |
2828 | if (strv_extend(&l, params->received_credentials_directory) < 0) | |
2829 | return NULL; | |
2ad591a3 | 2830 | |
96df2bd8 DDM |
2831 | if (strv_extend_strv(&l, CONF_PATHS_STRV("credstore"), /* filter_duplicates= */ true) < 0) |
2832 | return NULL; | |
2833 | } | |
2ad591a3 LP |
2834 | |
2835 | if (DEBUG_LOGGING) { | |
2836 | _cleanup_free_ char *t = strv_join(l, ":"); | |
2837 | ||
00078fb3 | 2838 | log_debug("Credential search path is: %s", strempty(t)); |
2ad591a3 LP |
2839 | } |
2840 | ||
2841 | return TAKE_PTR(l); | |
2842 | } | |
2843 | ||
bbfb25f4 DDM |
2844 | static int maybe_decrypt_and_write_credential( |
2845 | int dir_fd, | |
2846 | const char *id, | |
2847 | bool encrypted, | |
2848 | uid_t uid, | |
3a78b0e9 | 2849 | gid_t gid, |
bbfb25f4 DDM |
2850 | bool ownership_ok, |
2851 | const char *data, | |
2852 | size_t size, | |
2853 | uint64_t *left) { | |
2854 | ||
2855 | _cleanup_free_ void *plaintext = NULL; | |
2856 | size_t add; | |
2857 | int r; | |
2858 | ||
2859 | if (encrypted) { | |
2860 | size_t plaintext_size = 0; | |
2861 | ||
2862 | r = decrypt_credential_and_warn(id, now(CLOCK_REALTIME), NULL, NULL, data, size, | |
2863 | &plaintext, &plaintext_size); | |
2864 | if (r < 0) | |
2865 | return r; | |
2866 | ||
2867 | data = plaintext; | |
2868 | size = plaintext_size; | |
2869 | } | |
2870 | ||
2871 | add = strlen(id) + size; | |
2872 | if (add > *left) | |
2873 | return -E2BIG; | |
2874 | ||
3a78b0e9 | 2875 | r = write_credential(dir_fd, id, data, size, uid, gid, ownership_ok); |
bbfb25f4 DDM |
2876 | if (r < 0) |
2877 | return log_debug_errno(r, "Failed to write credential '%s': %m", id); | |
2878 | ||
2879 | *left -= add; | |
2880 | return 0; | |
2881 | } | |
2882 | ||
2883 | static int load_credential_glob( | |
2884 | const char *path, | |
2885 | bool encrypted, | |
2886 | char **search_path, | |
2887 | ReadFullFileFlags flags, | |
2888 | int write_dfd, | |
2889 | uid_t uid, | |
3a78b0e9 | 2890 | gid_t gid, |
bbfb25f4 DDM |
2891 | bool ownership_ok, |
2892 | uint64_t *left) { | |
2893 | ||
2894 | int r; | |
2895 | ||
2896 | STRV_FOREACH(d, search_path) { | |
2897 | _cleanup_globfree_ glob_t pglob = {}; | |
2898 | _cleanup_free_ char *j = NULL; | |
2899 | ||
2900 | j = path_join(*d, path); | |
2901 | if (!j) | |
2902 | return -ENOMEM; | |
2903 | ||
2904 | r = safe_glob(j, 0, &pglob); | |
2905 | if (r == -ENOENT) | |
2906 | continue; | |
2907 | if (r < 0) | |
2908 | return r; | |
2909 | ||
fcc06682 | 2910 | for (size_t n = 0; n < pglob.gl_pathc; n++) { |
bbfb25f4 DDM |
2911 | _cleanup_free_ char *fn = NULL; |
2912 | _cleanup_(erase_and_freep) char *data = NULL; | |
2913 | size_t size; | |
2914 | ||
2915 | /* path is absolute, hence pass AT_FDCWD as nop dir fd here */ | |
2916 | r = read_full_file_full( | |
2917 | AT_FDCWD, | |
2918 | pglob.gl_pathv[n], | |
2919 | UINT64_MAX, | |
2920 | encrypted ? CREDENTIAL_ENCRYPTED_SIZE_MAX : CREDENTIAL_SIZE_MAX, | |
2921 | flags, | |
2922 | NULL, | |
2923 | &data, &size); | |
2924 | if (r < 0) | |
2925 | return log_debug_errno(r, "Failed to read credential '%s': %m", | |
2926 | pglob.gl_pathv[n]); | |
2927 | ||
2928 | r = path_extract_filename(pglob.gl_pathv[n], &fn); | |
2929 | if (r < 0) | |
2930 | return log_debug_errno(r, "Failed to extract filename from '%s': %m", | |
2931 | pglob.gl_pathv[n]); | |
2932 | ||
2933 | r = maybe_decrypt_and_write_credential( | |
2934 | write_dfd, | |
2935 | fn, | |
2936 | encrypted, | |
2937 | uid, | |
3a78b0e9 | 2938 | gid, |
bbfb25f4 DDM |
2939 | ownership_ok, |
2940 | data, size, | |
2941 | left); | |
2942 | if (r == -EEXIST) | |
2943 | continue; | |
2944 | if (r < 0) | |
2945 | return r; | |
2946 | } | |
2947 | } | |
2948 | ||
2949 | return 0; | |
2950 | } | |
2951 | ||
3989bdc1 AB |
2952 | static int load_credential( |
2953 | const ExecContext *context, | |
2954 | const ExecParameters *params, | |
10b44e1d LP |
2955 | const char *id, |
2956 | const char *path, | |
2957 | bool encrypted, | |
3989bdc1 AB |
2958 | const char *unit, |
2959 | int read_dfd, | |
2960 | int write_dfd, | |
2961 | uid_t uid, | |
3a78b0e9 | 2962 | gid_t gid, |
3989bdc1 AB |
2963 | bool ownership_ok, |
2964 | uint64_t *left) { | |
2965 | ||
3989bdc1 | 2966 | ReadFullFileFlags flags = READ_FULL_FILE_SECURE|READ_FULL_FILE_FAIL_WHEN_LARGER; |
2ad591a3 | 2967 | _cleanup_strv_free_ char **search_path = NULL; |
3989bdc1 | 2968 | _cleanup_(erase_and_freep) char *data = NULL; |
2ad591a3 LP |
2969 | _cleanup_free_ char *bindname = NULL; |
2970 | const char *source = NULL; | |
3989bdc1 | 2971 | bool missing_ok = true; |
bbfb25f4 | 2972 | size_t size, maxsz; |
3989bdc1 AB |
2973 | int r; |
2974 | ||
10b44e1d LP |
2975 | assert(context); |
2976 | assert(params); | |
2977 | assert(id); | |
2978 | assert(path); | |
2979 | assert(unit); | |
661e4251 | 2980 | assert(read_dfd >= 0 || read_dfd == AT_FDCWD); |
10b44e1d LP |
2981 | assert(write_dfd >= 0); |
2982 | assert(left); | |
2983 | ||
2ad591a3 LP |
2984 | if (read_dfd >= 0) { |
2985 | /* If a directory fd is specified, then read the file directly from that dir. In this case we | |
2986 | * won't do AF_UNIX stuff (we simply don't want to recursively iterate down a tree of AF_UNIX | |
2987 | * IPC sockets). It's OK if a file vanishes here in the time we enumerate it and intend to | |
2988 | * open it. */ | |
2989 | ||
2990 | if (!filename_is_valid(path)) /* safety check */ | |
2991 | return -EINVAL; | |
2992 | ||
2993 | missing_ok = true; | |
10b44e1d | 2994 | source = path; |
2ad591a3 LP |
2995 | |
2996 | } else if (path_is_absolute(path)) { | |
2997 | /* If this is an absolute path, read the data directly from it, and support AF_UNIX | |
2998 | * sockets */ | |
2999 | ||
3000 | if (!path_is_valid(path)) /* safety check */ | |
3001 | return -EINVAL; | |
3002 | ||
3989bdc1 AB |
3003 | flags |= READ_FULL_FILE_CONNECT_SOCKET; |
3004 | ||
3005 | /* Pass some minimal info about the unit and the credential name we are looking to acquire | |
3006 | * via the source socket address in case we read off an AF_UNIX socket. */ | |
10b44e1d | 3007 | if (asprintf(&bindname, "@%" PRIx64"/unit/%s/%s", random_u64(), unit, id) < 0) |
3989bdc1 AB |
3008 | return -ENOMEM; |
3009 | ||
3010 | missing_ok = false; | |
2ad591a3 | 3011 | source = path; |
3989bdc1 | 3012 | |
2ad591a3 LP |
3013 | } else if (credential_name_valid(path)) { |
3014 | /* If this is a relative path, take it as credential name relative to the credentials | |
3015 | * directory we received ourselves. We don't support the AF_UNIX stuff in this mode, since we | |
3016 | * are operating on a credential store, i.e. this is guaranteed to be regular files. */ | |
3017 | ||
96df2bd8 | 3018 | search_path = credential_search_path(params, CREDENTIAL_SEARCH_PATH_ALL); |
2ad591a3 | 3019 | if (!search_path) |
3989bdc1 AB |
3020 | return -ENOMEM; |
3021 | ||
2ad591a3 | 3022 | missing_ok = true; |
3989bdc1 AB |
3023 | } else |
3024 | source = NULL; | |
3025 | ||
2ad591a3 LP |
3026 | if (encrypted) |
3027 | flags |= READ_FULL_FILE_UNBASE64; | |
3028 | ||
3029 | maxsz = encrypted ? CREDENTIAL_ENCRYPTED_SIZE_MAX : CREDENTIAL_SIZE_MAX; | |
3030 | ||
3031 | if (search_path) { | |
3032 | STRV_FOREACH(d, search_path) { | |
3033 | _cleanup_free_ char *j = NULL; | |
3034 | ||
3035 | j = path_join(*d, path); | |
3036 | if (!j) | |
3037 | return -ENOMEM; | |
3038 | ||
3039 | r = read_full_file_full( | |
3040 | AT_FDCWD, j, /* path is absolute, hence pass AT_FDCWD as nop dir fd here */ | |
3041 | UINT64_MAX, | |
3042 | maxsz, | |
3043 | flags, | |
3044 | NULL, | |
3045 | &data, &size); | |
3046 | if (r != -ENOENT) | |
3047 | break; | |
3048 | } | |
3049 | } else if (source) | |
3989bdc1 AB |
3050 | r = read_full_file_full( |
3051 | read_dfd, source, | |
3052 | UINT64_MAX, | |
2ad591a3 LP |
3053 | maxsz, |
3054 | flags, | |
3989bdc1 AB |
3055 | bindname, |
3056 | &data, &size); | |
3057 | else | |
3058 | r = -ENOENT; | |
3059 | ||
10b44e1d | 3060 | if (r == -ENOENT && (missing_ok || hashmap_contains(context->set_credentials, id))) { |
3989bdc1 AB |
3061 | /* Make a missing inherited credential non-fatal, let's just continue. After all apps |
3062 | * will get clear errors if we don't pass such a missing credential on as they | |
3063 | * themselves will get ENOENT when trying to read them, which should not be much | |
3064 | * worse than when we handle the error here and make it fatal. | |
3065 | * | |
3066 | * Also, if the source file doesn't exist, but a fallback is set via SetCredentials= | |
3067 | * we are fine, too. */ | |
10b44e1d | 3068 | log_debug_errno(r, "Couldn't read inherited credential '%s', skipping: %m", path); |
3989bdc1 AB |
3069 | return 0; |
3070 | } | |
3071 | if (r < 0) | |
10b44e1d | 3072 | return log_debug_errno(r, "Failed to read credential '%s': %m", path); |
3989bdc1 | 3073 | |
3a78b0e9 | 3074 | return maybe_decrypt_and_write_credential(write_dfd, id, encrypted, uid, gid, ownership_ok, data, size, left); |
3989bdc1 AB |
3075 | } |
3076 | ||
3077 | struct load_cred_args { | |
3989bdc1 AB |
3078 | const ExecContext *context; |
3079 | const ExecParameters *params; | |
461345a1 | 3080 | bool encrypted; |
3989bdc1 AB |
3081 | const char *unit; |
3082 | int dfd; | |
3083 | uid_t uid; | |
3a78b0e9 | 3084 | gid_t gid; |
3989bdc1 AB |
3085 | bool ownership_ok; |
3086 | uint64_t *left; | |
3087 | }; | |
3088 | ||
3089 | static int load_cred_recurse_dir_cb( | |
3090 | RecurseDirEvent event, | |
3091 | const char *path, | |
3092 | int dir_fd, | |
3093 | int inode_fd, | |
3094 | const struct dirent *de, | |
3095 | const struct statx *sx, | |
3096 | void *userdata) { | |
3097 | ||
6394e5cd | 3098 | struct load_cred_args *args = ASSERT_PTR(userdata); |
11348386 | 3099 | _cleanup_free_ char *sub_id = NULL; |
3989bdc1 AB |
3100 | int r; |
3101 | ||
3102 | if (event != RECURSE_DIR_ENTRY) | |
3103 | return RECURSE_DIR_CONTINUE; | |
3104 | ||
3105 | if (!IN_SET(de->d_type, DT_REG, DT_SOCK)) | |
3106 | return RECURSE_DIR_CONTINUE; | |
3107 | ||
11348386 | 3108 | sub_id = strreplace(path, "/", "_"); |
3989bdc1 AB |
3109 | if (!sub_id) |
3110 | return -ENOMEM; | |
3111 | ||
3112 | if (!credential_name_valid(sub_id)) | |
1451435c | 3113 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Credential would get ID %s, which is not valid, refusing", sub_id); |
3989bdc1 | 3114 | |
5bec447a | 3115 | if (faccessat(args->dfd, sub_id, F_OK, AT_SYMLINK_NOFOLLOW) >= 0) { |
3989bdc1 AB |
3116 | log_debug("Skipping credential with duplicated ID %s at %s", sub_id, path); |
3117 | return RECURSE_DIR_CONTINUE; | |
3118 | } | |
5bec447a LP |
3119 | if (errno != ENOENT) |
3120 | return log_debug_errno(errno, "Failed to test if credential %s exists: %m", sub_id); | |
3989bdc1 | 3121 | |
10b44e1d LP |
3122 | r = load_credential( |
3123 | args->context, | |
3124 | args->params, | |
3125 | sub_id, | |
3126 | de->d_name, | |
461345a1 | 3127 | args->encrypted, |
10b44e1d LP |
3128 | args->unit, |
3129 | dir_fd, | |
3130 | args->dfd, | |
3131 | args->uid, | |
3a78b0e9 | 3132 | args->gid, |
10b44e1d LP |
3133 | args->ownership_ok, |
3134 | args->left); | |
3989bdc1 AB |
3135 | if (r < 0) |
3136 | return r; | |
3137 | ||
3138 | return RECURSE_DIR_CONTINUE; | |
3139 | } | |
3140 | ||
bb0c0d6f LP |
3141 | static int acquire_credentials( |
3142 | const ExecContext *context, | |
3143 | const ExecParameters *params, | |
d3dcf4e3 | 3144 | const char *unit, |
bb0c0d6f LP |
3145 | const char *p, |
3146 | uid_t uid, | |
3a78b0e9 | 3147 | gid_t gid, |
bb0c0d6f LP |
3148 | bool ownership_ok) { |
3149 | ||
43144be4 | 3150 | uint64_t left = CREDENTIALS_TOTAL_SIZE_MAX; |
254d1313 | 3151 | _cleanup_close_ int dfd = -EBADF; |
bbfb25f4 | 3152 | const char *ic; |
43144be4 | 3153 | ExecLoadCredential *lc; |
bb0c0d6f | 3154 | ExecSetCredential *sc; |
bb0c0d6f LP |
3155 | int r; |
3156 | ||
3157 | assert(context); | |
3158 | assert(p); | |
3159 | ||
3160 | dfd = open(p, O_DIRECTORY|O_CLOEXEC); | |
3161 | if (dfd < 0) | |
3162 | return -errno; | |
3163 | ||
f76ce81b LP |
3164 | r = fd_acl_make_writable(dfd); /* Add the "w" bit, if we are reusing an already set up credentials dir where it was unset */ |
3165 | if (r < 0) | |
3166 | return r; | |
3167 | ||
43144be4 LP |
3168 | /* First, load credentials off disk (or acquire via AF_UNIX socket) */ |
3169 | HASHMAP_FOREACH(lc, context->load_credentials) { | |
254d1313 | 3170 | _cleanup_close_ int sub_fd = -EBADF; |
d3dcf4e3 | 3171 | |
f344f7fd LP |
3172 | /* If this is an absolute path, then try to open it as a directory. If that works, then we'll |
3173 | * recurse into it. If it is an absolute path but it isn't a directory, then we'll open it as | |
3174 | * a regular file. Finally, if it's a relative path we will use it as a credential name to | |
3175 | * propagate a credential passed to us from further up. */ | |
43144be4 | 3176 | |
f344f7fd LP |
3177 | if (path_is_absolute(lc->path)) { |
3178 | sub_fd = open(lc->path, O_DIRECTORY|O_CLOEXEC|O_RDONLY); | |
1d68a2e1 LP |
3179 | if (sub_fd < 0 && !IN_SET(errno, |
3180 | ENOTDIR, /* Not a directory */ | |
3181 | ENOENT)) /* Doesn't exist? */ | |
3182 | return log_debug_errno(errno, "Failed to open '%s': %m", lc->path); | |
f344f7fd | 3183 | } |
43144be4 | 3184 | |
61c5a49e | 3185 | if (sub_fd < 0) |
f344f7fd | 3186 | /* Regular file (incl. a credential passed in from higher up) */ |
10b44e1d LP |
3187 | r = load_credential( |
3188 | context, | |
3189 | params, | |
3190 | lc->id, | |
3191 | lc->path, | |
3192 | lc->encrypted, | |
3193 | unit, | |
661e4251 | 3194 | AT_FDCWD, |
10b44e1d LP |
3195 | dfd, |
3196 | uid, | |
3a78b0e9 | 3197 | gid, |
10b44e1d LP |
3198 | ownership_ok, |
3199 | &left); | |
61c5a49e | 3200 | else |
10b44e1d | 3201 | /* Directory */ |
3989bdc1 AB |
3202 | r = recurse_dir( |
3203 | sub_fd, | |
11348386 | 3204 | /* path= */ lc->id, /* recurse_dir() will suffix the subdir paths from here to the top-level id */ |
3989bdc1 AB |
3205 | /* statx_mask= */ 0, |
3206 | /* n_depth_max= */ UINT_MAX, | |
9883cbb2 | 3207 | RECURSE_DIR_SORT|RECURSE_DIR_IGNORE_DOT|RECURSE_DIR_ENSURE_TYPE, |
3989bdc1 AB |
3208 | load_cred_recurse_dir_cb, |
3209 | &(struct load_cred_args) { | |
3989bdc1 AB |
3210 | .context = context, |
3211 | .params = params, | |
461345a1 | 3212 | .encrypted = lc->encrypted, |
3989bdc1 AB |
3213 | .unit = unit, |
3214 | .dfd = dfd, | |
3215 | .uid = uid, | |
3a78b0e9 | 3216 | .gid = gid, |
3989bdc1 AB |
3217 | .ownership_ok = ownership_ok, |
3218 | .left = &left, | |
3219 | }); | |
61c5a49e LP |
3220 | if (r < 0) |
3221 | return r; | |
bb0c0d6f LP |
3222 | } |
3223 | ||
bbfb25f4 DDM |
3224 | /* Next, look for system credentials and credentials in the credentials store. Note that these do not |
3225 | * override any credentials found earlier. */ | |
3226 | SET_FOREACH(ic, context->import_credentials) { | |
3227 | _cleanup_free_ char **search_path = NULL; | |
3228 | ||
3229 | search_path = credential_search_path(params, CREDENTIAL_SEARCH_PATH_TRUSTED); | |
3230 | if (!search_path) | |
3231 | return -ENOMEM; | |
3232 | ||
3233 | r = load_credential_glob( | |
3234 | ic, | |
3235 | /* encrypted = */ false, | |
3236 | search_path, | |
3237 | READ_FULL_FILE_SECURE|READ_FULL_FILE_FAIL_WHEN_LARGER, | |
3238 | dfd, | |
3239 | uid, | |
3a78b0e9 | 3240 | gid, |
bbfb25f4 DDM |
3241 | ownership_ok, |
3242 | &left); | |
3243 | if (r < 0) | |
3244 | return r; | |
3245 | ||
3246 | search_path = strv_free(search_path); | |
3247 | search_path = credential_search_path(params, CREDENTIAL_SEARCH_PATH_ENCRYPTED); | |
3248 | if (!search_path) | |
3249 | return -ENOMEM; | |
3250 | ||
3251 | r = load_credential_glob( | |
3252 | ic, | |
3253 | /* encrypted = */ true, | |
3254 | search_path, | |
3255 | READ_FULL_FILE_SECURE|READ_FULL_FILE_FAIL_WHEN_LARGER|READ_FULL_FILE_UNBASE64, | |
3256 | dfd, | |
3257 | uid, | |
3a78b0e9 | 3258 | gid, |
bbfb25f4 DDM |
3259 | ownership_ok, |
3260 | &left); | |
3261 | if (r < 0) | |
3262 | return r; | |
3263 | } | |
3264 | ||
3265 | /* Finally, we add in literally specified credentials. If the credentials already exist, we'll not | |
3266 | * add them, so that they can act as a "default" if the same credential is specified multiple times. */ | |
43144be4 LP |
3267 | HASHMAP_FOREACH(sc, context->set_credentials) { |
3268 | _cleanup_(erase_and_freep) void *plaintext = NULL; | |
3269 | const char *data; | |
3270 | size_t size, add; | |
3271 | ||
9e6e9d61 LP |
3272 | /* Note that we check ahead of time here instead of relying on O_EXCL|O_CREAT later to return |
3273 | * EEXIST if the credential already exists. That's because the TPM2-based decryption is kinda | |
3274 | * slow and involved, hence it's nice to be able to skip that if the credential already | |
3275 | * exists anyway. */ | |
43144be4 LP |
3276 | if (faccessat(dfd, sc->id, F_OK, AT_SYMLINK_NOFOLLOW) >= 0) |
3277 | continue; | |
3278 | if (errno != ENOENT) | |
3279 | return log_debug_errno(errno, "Failed to test if credential %s exists: %m", sc->id); | |
3280 | ||
3281 | if (sc->encrypted) { | |
6a0779cb | 3282 | r = decrypt_credential_and_warn(sc->id, now(CLOCK_REALTIME), NULL, NULL, sc->data, sc->size, &plaintext, &size); |
43144be4 LP |
3283 | if (r < 0) |
3284 | return r; | |
3285 | ||
3286 | data = plaintext; | |
3287 | } else { | |
3288 | data = sc->data; | |
3289 | size = sc->size; | |
3290 | } | |
3291 | ||
3292 | add = strlen(sc->id) + size; | |
3293 | if (add > left) | |
3294 | return -E2BIG; | |
3295 | ||
3a78b0e9 | 3296 | r = write_credential(dfd, sc->id, data, size, uid, gid, ownership_ok); |
43144be4 LP |
3297 | if (r < 0) |
3298 | return r; | |
3299 | ||
43144be4 LP |
3300 | left -= add; |
3301 | } | |
3302 | ||
f76ce81b LP |
3303 | r = fd_acl_make_read_only(dfd); /* Now take away the "w" bit */ |
3304 | if (r < 0) | |
3305 | return r; | |
bb0c0d6f LP |
3306 | |
3307 | /* After we created all keys with the right perms, also make sure the credential store as a whole is | |
3308 | * accessible */ | |
3309 | ||
3310 | if (uid_is_valid(uid) && uid != getuid()) { | |
567aeb58 | 3311 | r = fd_add_uid_acl_permission(dfd, uid, ACL_READ | ACL_EXECUTE); |
bb0c0d6f LP |
3312 | if (r < 0) { |
3313 | if (!ERRNO_IS_NOT_SUPPORTED(r) && !ERRNO_IS_PRIVILEGE(r)) | |
3314 | return r; | |
3315 | ||
3316 | if (!ownership_ok) | |
3317 | return r; | |
3318 | ||
3a78b0e9 | 3319 | if (fchown(dfd, uid, gid) < 0) |
bb0c0d6f LP |
3320 | return -errno; |
3321 | } | |
3322 | } | |
3323 | ||
3324 | return 0; | |
3325 | } | |
3326 | ||
3327 | static int setup_credentials_internal( | |
3328 | const ExecContext *context, | |
3329 | const ExecParameters *params, | |
d3dcf4e3 | 3330 | const char *unit, |
bb0c0d6f LP |
3331 | const char *final, /* This is where the credential store shall eventually end up at */ |
3332 | const char *workspace, /* This is where we can prepare it before moving it to the final place */ | |
3333 | bool reuse_workspace, /* Whether to reuse any existing workspace mount if it already is a mount */ | |
3334 | bool must_mount, /* Whether to require that we mount something, it's not OK to use the plain directory fall back */ | |
3a78b0e9 DDM |
3335 | uid_t uid, |
3336 | gid_t gid) { | |
bb0c0d6f LP |
3337 | |
3338 | int r, workspace_mounted; /* negative if we don't know yet whether we have/can mount something; true | |
3339 | * if we mounted something; false if we definitely can't mount anything */ | |
3340 | bool final_mounted; | |
3341 | const char *where; | |
3342 | ||
3343 | assert(context); | |
3344 | assert(final); | |
3345 | assert(workspace); | |
3346 | ||
3347 | if (reuse_workspace) { | |
3348 | r = path_is_mount_point(workspace, NULL, 0); | |
3349 | if (r < 0) | |
3350 | return r; | |
3351 | if (r > 0) | |
3352 | workspace_mounted = true; /* If this is already a mount, and we are supposed to reuse it, let's keep this in mind */ | |
3353 | else | |
3354 | workspace_mounted = -1; /* We need to figure out if we can mount something to the workspace */ | |
3355 | } else | |
3356 | workspace_mounted = -1; /* ditto */ | |
3357 | ||
3358 | r = path_is_mount_point(final, NULL, 0); | |
3359 | if (r < 0) | |
3360 | return r; | |
3361 | if (r > 0) { | |
3362 | /* If the final place already has something mounted, we use that. If the workspace also has | |
3363 | * something mounted we assume it's actually the same mount (but with MS_RDONLY | |
3364 | * different). */ | |
3365 | final_mounted = true; | |
3366 | ||
3367 | if (workspace_mounted < 0) { | |
f0353cf2 | 3368 | /* If the final place is mounted, but the workspace isn't, then let's bind mount |
bb0c0d6f LP |
3369 | * the final version to the workspace, and make it writable, so that we can make |
3370 | * changes */ | |
3371 | ||
21935150 LP |
3372 | r = mount_nofollow_verbose(LOG_DEBUG, final, workspace, NULL, MS_BIND|MS_REC, NULL); |
3373 | if (r < 0) | |
3374 | return r; | |
bb0c0d6f | 3375 | |
1155f44f | 3376 | r = mount_nofollow_verbose(LOG_DEBUG, NULL, workspace, NULL, MS_BIND|MS_REMOUNT|credentials_fs_mount_flags(/* ro= */ false), NULL); |
21935150 LP |
3377 | if (r < 0) |
3378 | return r; | |
bb0c0d6f LP |
3379 | |
3380 | workspace_mounted = true; | |
3381 | } | |
3382 | } else | |
3383 | final_mounted = false; | |
3384 | ||
3385 | if (workspace_mounted < 0) { | |
3386 | /* Nothing is mounted on the workspace yet, let's try to mount something now */ | |
bb0c0d6f | 3387 | |
1155f44f LP |
3388 | r = mount_credentials_fs(workspace, CREDENTIALS_TOTAL_SIZE_MAX, /* ro= */ false); |
3389 | if (r < 0) { | |
3390 | /* If that didn't work, try to make a bind mount from the final to the workspace, so that we can make it writable there. */ | |
3391 | r = mount_nofollow_verbose(LOG_DEBUG, final, workspace, NULL, MS_BIND|MS_REC, NULL); | |
3392 | if (r < 0) { | |
3393 | if (!ERRNO_IS_PRIVILEGE(r)) /* Propagate anything that isn't a permission problem */ | |
3394 | return r; | |
bb0c0d6f | 3395 | |
1155f44f LP |
3396 | if (must_mount) /* If we it's not OK to use the plain directory |
3397 | * fallback, propagate all errors too */ | |
3398 | return r; | |
bb0c0d6f | 3399 | |
1155f44f LP |
3400 | /* If we lack privileges to bind mount stuff, then let's gracefully |
3401 | * proceed for compat with container envs, and just use the final dir | |
3402 | * as is. */ | |
bb0c0d6f | 3403 | |
1155f44f | 3404 | workspace_mounted = false; |
bb0c0d6f | 3405 | } else { |
bb0c0d6f | 3406 | /* Make the new bind mount writable (i.e. drop MS_RDONLY) */ |
1155f44f | 3407 | r = mount_nofollow_verbose(LOG_DEBUG, NULL, workspace, NULL, MS_BIND|MS_REMOUNT|credentials_fs_mount_flags(/* ro= */ false), NULL); |
21935150 LP |
3408 | if (r < 0) |
3409 | return r; | |
bb0c0d6f LP |
3410 | |
3411 | workspace_mounted = true; | |
bb0c0d6f | 3412 | } |
1155f44f LP |
3413 | } else |
3414 | workspace_mounted = true; | |
bb0c0d6f LP |
3415 | } |
3416 | ||
3417 | assert(!must_mount || workspace_mounted > 0); | |
3418 | where = workspace_mounted ? workspace : final; | |
3419 | ||
03bc11d1 | 3420 | (void) label_fix_full(AT_FDCWD, where, final, 0); |
e3a0a862 | 3421 | |
3a78b0e9 | 3422 | r = acquire_credentials(context, params, unit, where, uid, gid, workspace_mounted); |
bb0c0d6f LP |
3423 | if (r < 0) |
3424 | return r; | |
3425 | ||
3426 | if (workspace_mounted) { | |
21dd1de6 | 3427 | bool install; |
bb0c0d6f | 3428 | |
21dd1de6 LP |
3429 | /* Determine if we should actually install the prepared mount in the final location by bind |
3430 | * mounting it there. We do so only if the mount is not established there already, and if the | |
3431 | * mount is actually non-empty (i.e. carries at least one credential). Not that in the best | |
fa18faf1 | 3432 | * case we are doing all this in a mount namespace, thus no one else will see that we |
21dd1de6 | 3433 | * allocated a file system we are getting rid of again here. */ |
21935150 | 3434 | if (final_mounted) |
21dd1de6 LP |
3435 | install = false; /* already installed */ |
3436 | else { | |
3437 | r = dir_is_empty(where, /* ignore_hidden_or_backup= */ false); | |
3438 | if (r < 0) | |
3439 | return r; | |
3440 | ||
3441 | install = r == 0; /* install only if non-empty */ | |
3442 | } | |
3443 | ||
3444 | if (install) { | |
3445 | /* Make workspace read-only now, so that any bind mount we make from it defaults to read-only too */ | |
1155f44f | 3446 | r = mount_nofollow_verbose(LOG_DEBUG, NULL, workspace, NULL, MS_BIND|MS_REMOUNT|credentials_fs_mount_flags(/* ro= */ true), NULL); |
21dd1de6 LP |
3447 | if (r < 0) |
3448 | return r; | |
3449 | ||
3450 | /* And mount it to the final place, read-only */ | |
21935150 | 3451 | r = mount_nofollow_verbose(LOG_DEBUG, workspace, final, NULL, MS_MOVE, NULL); |
21dd1de6 LP |
3452 | } else |
3453 | /* Otherwise get rid of it */ | |
3454 | r = umount_verbose(LOG_DEBUG, workspace, MNT_DETACH|UMOUNT_NOFOLLOW); | |
21935150 LP |
3455 | if (r < 0) |
3456 | return r; | |
bb0c0d6f LP |
3457 | } else { |
3458 | _cleanup_free_ char *parent = NULL; | |
3459 | ||
3460 | /* If we do not have our own mount put used the plain directory fallback, then we need to | |
3461 | * open access to the top-level credential directory and the per-service directory now */ | |
3462 | ||
45519d13 LP |
3463 | r = path_extract_directory(final, &parent); |
3464 | if (r < 0) | |
3465 | return r; | |
bb0c0d6f LP |
3466 | if (chmod(parent, 0755) < 0) |
3467 | return -errno; | |
3468 | } | |
3469 | ||
3470 | return 0; | |
3471 | } | |
3472 | ||
3473 | static int setup_credentials( | |
3474 | const ExecContext *context, | |
3475 | const ExecParameters *params, | |
3476 | const char *unit, | |
3a78b0e9 DDM |
3477 | uid_t uid, |
3478 | gid_t gid) { | |
bb0c0d6f LP |
3479 | |
3480 | _cleanup_free_ char *p = NULL, *q = NULL; | |
bb0c0d6f LP |
3481 | int r; |
3482 | ||
3483 | assert(context); | |
3484 | assert(params); | |
3485 | ||
3486 | if (!exec_context_has_credentials(context)) | |
3487 | return 0; | |
3488 | ||
3489 | if (!params->prefix[EXEC_DIRECTORY_RUNTIME]) | |
3490 | return -EINVAL; | |
3491 | ||
3492 | /* This where we'll place stuff when we are done; this main credentials directory is world-readable, | |
3493 | * and the subdir we mount over with a read-only file system readable by the service's user */ | |
3494 | q = path_join(params->prefix[EXEC_DIRECTORY_RUNTIME], "credentials"); | |
3495 | if (!q) | |
3496 | return -ENOMEM; | |
3497 | ||
3498 | r = mkdir_label(q, 0755); /* top-level dir: world readable/searchable */ | |
3499 | if (r < 0 && r != -EEXIST) | |
3500 | return r; | |
3501 | ||
3502 | p = path_join(q, unit); | |
3503 | if (!p) | |
3504 | return -ENOMEM; | |
3505 | ||
3506 | r = mkdir_label(p, 0700); /* per-unit dir: private to user */ | |
3507 | if (r < 0 && r != -EEXIST) | |
3508 | return r; | |
3509 | ||
3510 | r = safe_fork("(sd-mkdcreds)", FORK_DEATHSIG|FORK_WAIT|FORK_NEW_MOUNTNS, NULL); | |
3511 | if (r < 0) { | |
3512 | _cleanup_free_ char *t = NULL, *u = NULL; | |
3513 | ||
3514 | /* If this is not a privilege or support issue then propagate the error */ | |
3515 | if (!ERRNO_IS_NOT_SUPPORTED(r) && !ERRNO_IS_PRIVILEGE(r)) | |
3516 | return r; | |
3517 | ||
3518 | /* Temporary workspace, that remains inaccessible all the time. We prepare stuff there before moving | |
3519 | * it into place, so that users can't access half-initialized credential stores. */ | |
3520 | t = path_join(params->prefix[EXEC_DIRECTORY_RUNTIME], "systemd/temporary-credentials"); | |
3521 | if (!t) | |
3522 | return -ENOMEM; | |
3523 | ||
3524 | /* We can't set up a mount namespace. In that case operate on a fixed, inaccessible per-unit | |
3525 | * directory outside of /run/credentials/ first, and then move it over to /run/credentials/ | |
3526 | * after it is fully set up */ | |
3527 | u = path_join(t, unit); | |
3528 | if (!u) | |
3529 | return -ENOMEM; | |
3530 | ||
3531 | FOREACH_STRING(i, t, u) { | |
3532 | r = mkdir_label(i, 0700); | |
3533 | if (r < 0 && r != -EEXIST) | |
3534 | return r; | |
3535 | } | |
3536 | ||
3537 | r = setup_credentials_internal( | |
3538 | context, | |
3539 | params, | |
d3dcf4e3 | 3540 | unit, |
bb0c0d6f LP |
3541 | p, /* final mount point */ |
3542 | u, /* temporary workspace to overmount */ | |
3543 | true, /* reuse the workspace if it is already a mount */ | |
3544 | false, /* it's OK to fall back to a plain directory if we can't mount anything */ | |
3a78b0e9 DDM |
3545 | uid, |
3546 | gid); | |
bb0c0d6f LP |
3547 | |
3548 | (void) rmdir(u); /* remove the workspace again if we can. */ | |
3549 | ||
3550 | if (r < 0) | |
3551 | return r; | |
3552 | ||
3553 | } else if (r == 0) { | |
3554 | ||
3555 | /* We managed to set up a mount namespace, and are now in a child. That's great. In this case | |
3556 | * we can use the same directory for all cases, after turning off propagation. Question | |
3557 | * though is: where do we turn off propagation exactly, and where do we place the workspace | |
3558 | * directory? We need some place that is guaranteed to be a mount point in the host, and | |
3559 | * which is guaranteed to have a subdir we can mount over. /run/ is not suitable for this, | |
3560 | * since we ultimately want to move the resulting file system there, i.e. we need propagation | |
3561 | * for /run/ eventually. We could use our own /run/systemd/bind mount on itself, but that | |
3562 | * would be visible in the host mount table all the time, which we want to avoid. Hence, what | |
3563 | * we do here instead we use /dev/ and /dev/shm/ for our purposes. We know for sure that | |
3564 | * /dev/ is a mount point and we now for sure that /dev/shm/ exists. Hence we can turn off | |
3565 | * propagation on the former, and then overmount the latter. | |
3566 | * | |
3567 | * Yes it's nasty playing games with /dev/ and /dev/shm/ like this, since it does not exist | |
3568 | * for this purpose, but there are few other candidates that work equally well for us, and | |
3569 | * given that the we do this in a privately namespaced short-lived single-threaded process | |
7802194a | 3570 | * that no one else sees this should be OK to do. */ |
bb0c0d6f | 3571 | |
21935150 LP |
3572 | r = mount_nofollow_verbose(LOG_DEBUG, NULL, "/dev", NULL, MS_SLAVE|MS_REC, NULL); /* Turn off propagation from our namespace to host */ |
3573 | if (r < 0) | |
bb0c0d6f LP |
3574 | goto child_fail; |
3575 | ||
3576 | r = setup_credentials_internal( | |
3577 | context, | |
3578 | params, | |
d3dcf4e3 | 3579 | unit, |
bb0c0d6f LP |
3580 | p, /* final mount point */ |
3581 | "/dev/shm", /* temporary workspace to overmount */ | |
3582 | false, /* do not reuse /dev/shm if it is already a mount, under no circumstances */ | |
3583 | true, /* insist that something is mounted, do not allow fallback to plain directory */ | |
3a78b0e9 DDM |
3584 | uid, |
3585 | gid); | |
bb0c0d6f LP |
3586 | if (r < 0) |
3587 | goto child_fail; | |
3588 | ||
3589 | _exit(EXIT_SUCCESS); | |
3590 | ||
3591 | child_fail: | |
3592 | _exit(EXIT_FAILURE); | |
3593 | } | |
3594 | ||
3aaa3763 LP |
3595 | /* If the credentials dir is empty and not a mount point, then there's no point in having it. Let's |
3596 | * try to remove it. This matters in particular if we created the dir as mount point but then didn't | |
3597 | * actually end up mounting anything on it. In that case we'd rather have ENOENT than EACCESS being | |
3598 | * seen by users when trying access this inode. */ | |
3599 | (void) rmdir(p); | |
bb0c0d6f LP |
3600 | return 0; |
3601 | } | |
3602 | ||
92b423b9 | 3603 | #if ENABLE_SMACK |
cefc33ae | 3604 | static int setup_smack( |
aa5ae971 | 3605 | const Manager *manager, |
cefc33ae | 3606 | const ExecContext *context, |
b83d5050 | 3607 | int executable_fd) { |
cefc33ae LP |
3608 | int r; |
3609 | ||
3610 | assert(context); | |
b83d5050 | 3611 | assert(executable_fd >= 0); |
cefc33ae | 3612 | |
cefc33ae LP |
3613 | if (context->smack_process_label) { |
3614 | r = mac_smack_apply_pid(0, context->smack_process_label); | |
3615 | if (r < 0) | |
3616 | return r; | |
aa5ae971 | 3617 | } else if (manager->default_smack_process_label) { |
cefc33ae LP |
3618 | _cleanup_free_ char *exec_label = NULL; |
3619 | ||
b83d5050 | 3620 | r = mac_smack_read_fd(executable_fd, SMACK_ATTR_EXEC, &exec_label); |
00675c36 | 3621 | if (r < 0 && !ERRNO_IS_XATTR_ABSENT(r)) |
cefc33ae LP |
3622 | return r; |
3623 | ||
1da3cb81 | 3624 | r = mac_smack_apply_pid(0, exec_label ?: manager->default_smack_process_label); |
cefc33ae LP |
3625 | if (r < 0) |
3626 | return r; | |
3627 | } | |
cefc33ae LP |
3628 | |
3629 | return 0; | |
3630 | } | |
92b423b9 | 3631 | #endif |
cefc33ae | 3632 | |
6c47cd7d LP |
3633 | static int compile_bind_mounts( |
3634 | const ExecContext *context, | |
3635 | const ExecParameters *params, | |
3636 | BindMount **ret_bind_mounts, | |
da6053d0 | 3637 | size_t *ret_n_bind_mounts, |
6c47cd7d LP |
3638 | char ***ret_empty_directories) { |
3639 | ||
3640 | _cleanup_strv_free_ char **empty_directories = NULL; | |
ed8267c7 | 3641 | BindMount *bind_mounts = NULL; |
5b10116e | 3642 | size_t n, h = 0; |
6c47cd7d LP |
3643 | int r; |
3644 | ||
3645 | assert(context); | |
3646 | assert(params); | |
3647 | assert(ret_bind_mounts); | |
3648 | assert(ret_n_bind_mounts); | |
3649 | assert(ret_empty_directories); | |
3650 | ||
ed8267c7 DT |
3651 | CLEANUP_ARRAY(bind_mounts, h, bind_mount_free_many); |
3652 | ||
6c47cd7d | 3653 | n = context->n_bind_mounts; |
5b10116e | 3654 | for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) { |
6c47cd7d LP |
3655 | if (!params->prefix[t]) |
3656 | continue; | |
3657 | ||
a2ab603c YW |
3658 | for (size_t i = 0; i < context->directories[t].n_items; i++) |
3659 | n += !context->directories[t].items[i].only_create; | |
6c47cd7d LP |
3660 | } |
3661 | ||
3662 | if (n <= 0) { | |
3663 | *ret_bind_mounts = NULL; | |
3664 | *ret_n_bind_mounts = 0; | |
3665 | *ret_empty_directories = NULL; | |
3666 | return 0; | |
3667 | } | |
3668 | ||
3669 | bind_mounts = new(BindMount, n); | |
3670 | if (!bind_mounts) | |
3671 | return -ENOMEM; | |
3672 | ||
5b10116e | 3673 | for (size_t i = 0; i < context->n_bind_mounts; i++) { |
6c47cd7d | 3674 | BindMount *item = context->bind_mounts + i; |
93404d34 | 3675 | _cleanup_free_ char *s = NULL, *d = NULL; |
6c47cd7d LP |
3676 | |
3677 | s = strdup(item->source); | |
ed8267c7 DT |
3678 | if (!s) |
3679 | return -ENOMEM; | |
6c47cd7d LP |
3680 | |
3681 | d = strdup(item->destination); | |
93404d34 | 3682 | if (!d) |
ed8267c7 | 3683 | return -ENOMEM; |
6c47cd7d LP |
3684 | |
3685 | bind_mounts[h++] = (BindMount) { | |
93404d34 DT |
3686 | .source = TAKE_PTR(s), |
3687 | .destination = TAKE_PTR(d), | |
6c47cd7d LP |
3688 | .read_only = item->read_only, |
3689 | .recursive = item->recursive, | |
3690 | .ignore_enoent = item->ignore_enoent, | |
3691 | }; | |
3692 | } | |
3693 | ||
5b10116e | 3694 | for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) { |
6c47cd7d LP |
3695 | if (!params->prefix[t]) |
3696 | continue; | |
3697 | ||
211a3d87 | 3698 | if (context->directories[t].n_items == 0) |
6c47cd7d LP |
3699 | continue; |
3700 | ||
494d0247 | 3701 | if (exec_directory_is_private(context, t) && |
74e12520 | 3702 | !exec_context_with_rootfs(context)) { |
6c47cd7d LP |
3703 | char *private_root; |
3704 | ||
3705 | /* So this is for a dynamic user, and we need to make sure the process can access its own | |
3706 | * directory. For that we overmount the usually inaccessible "private" subdirectory with a | |
3707 | * tmpfs that makes it accessible and is empty except for the submounts we do this for. */ | |
3708 | ||
657ee2d8 | 3709 | private_root = path_join(params->prefix[t], "private"); |
ed8267c7 DT |
3710 | if (!private_root) |
3711 | return -ENOMEM; | |
6c47cd7d LP |
3712 | |
3713 | r = strv_consume(&empty_directories, private_root); | |
a635a7ae | 3714 | if (r < 0) |
ed8267c7 | 3715 | return r; |
6c47cd7d LP |
3716 | } |
3717 | ||
211a3d87 | 3718 | for (size_t i = 0; i < context->directories[t].n_items; i++) { |
93404d34 | 3719 | _cleanup_free_ char *s = NULL, *d = NULL; |
6c47cd7d | 3720 | |
a2ab603c YW |
3721 | /* When one of the parent directories is in the list, we cannot create the symlink |
3722 | * for the child directory. See also the comments in setup_exec_directory(). */ | |
3723 | if (context->directories[t].items[i].only_create) | |
3724 | continue; | |
3725 | ||
494d0247 | 3726 | if (exec_directory_is_private(context, t)) |
211a3d87 | 3727 | s = path_join(params->prefix[t], "private", context->directories[t].items[i].path); |
6c47cd7d | 3728 | else |
211a3d87 | 3729 | s = path_join(params->prefix[t], context->directories[t].items[i].path); |
ed8267c7 DT |
3730 | if (!s) |
3731 | return -ENOMEM; | |
6c47cd7d | 3732 | |
494d0247 | 3733 | if (exec_directory_is_private(context, t) && |
74e12520 | 3734 | exec_context_with_rootfs(context)) |
5609f688 YW |
3735 | /* When RootDirectory= or RootImage= are set, then the symbolic link to the private |
3736 | * directory is not created on the root directory. So, let's bind-mount the directory | |
3737 | * on the 'non-private' place. */ | |
211a3d87 | 3738 | d = path_join(params->prefix[t], context->directories[t].items[i].path); |
5609f688 YW |
3739 | else |
3740 | d = strdup(s); | |
93404d34 | 3741 | if (!d) |
ed8267c7 | 3742 | return -ENOMEM; |
6c47cd7d LP |
3743 | |
3744 | bind_mounts[h++] = (BindMount) { | |
93404d34 DT |
3745 | .source = TAKE_PTR(s), |
3746 | .destination = TAKE_PTR(d), | |
6c47cd7d | 3747 | .read_only = false, |
9ce4e4b0 | 3748 | .nosuid = context->dynamic_user, /* don't allow suid/sgid when DynamicUser= is on */ |
6c47cd7d LP |
3749 | .recursive = true, |
3750 | .ignore_enoent = false, | |
3751 | }; | |
3752 | } | |
3753 | } | |
3754 | ||
3755 | assert(h == n); | |
3756 | ||
ed8267c7 | 3757 | *ret_bind_mounts = TAKE_PTR(bind_mounts); |
6c47cd7d | 3758 | *ret_n_bind_mounts = n; |
ae2a15bc | 3759 | *ret_empty_directories = TAKE_PTR(empty_directories); |
6c47cd7d LP |
3760 | |
3761 | return (int) n; | |
6c47cd7d LP |
3762 | } |
3763 | ||
df61e79a LB |
3764 | /* ret_symlinks will contain a list of pairs src:dest that describes |
3765 | * the symlinks to create later on. For example, the symlinks needed | |
3766 | * to safely give private directories to DynamicUser=1 users. */ | |
3767 | static int compile_symlinks( | |
3768 | const ExecContext *context, | |
3769 | const ExecParameters *params, | |
3770 | char ***ret_symlinks) { | |
3771 | ||
3772 | _cleanup_strv_free_ char **symlinks = NULL; | |
3773 | int r; | |
3774 | ||
3775 | assert(context); | |
3776 | assert(params); | |
3777 | assert(ret_symlinks); | |
3778 | ||
3779 | for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) { | |
211a3d87 LB |
3780 | for (size_t i = 0; i < context->directories[dt].n_items; i++) { |
3781 | _cleanup_free_ char *private_path = NULL, *path = NULL; | |
df61e79a | 3782 | |
211a3d87 LB |
3783 | STRV_FOREACH(symlink, context->directories[dt].items[i].symlinks) { |
3784 | _cleanup_free_ char *src_abs = NULL, *dst_abs = NULL; | |
df61e79a | 3785 | |
211a3d87 LB |
3786 | src_abs = path_join(params->prefix[dt], context->directories[dt].items[i].path); |
3787 | dst_abs = path_join(params->prefix[dt], *symlink); | |
3788 | if (!src_abs || !dst_abs) | |
3789 | return -ENOMEM; | |
df61e79a | 3790 | |
211a3d87 LB |
3791 | r = strv_consume_pair(&symlinks, TAKE_PTR(src_abs), TAKE_PTR(dst_abs)); |
3792 | if (r < 0) | |
3793 | return r; | |
3794 | } | |
3795 | ||
a2ab603c YW |
3796 | if (!exec_directory_is_private(context, dt) || |
3797 | exec_context_with_rootfs(context) || | |
3798 | context->directories[dt].items[i].only_create) | |
211a3d87 LB |
3799 | continue; |
3800 | ||
3801 | private_path = path_join(params->prefix[dt], "private", context->directories[dt].items[i].path); | |
df61e79a LB |
3802 | if (!private_path) |
3803 | return -ENOMEM; | |
3804 | ||
211a3d87 | 3805 | path = path_join(params->prefix[dt], context->directories[dt].items[i].path); |
df61e79a LB |
3806 | if (!path) |
3807 | return -ENOMEM; | |
3808 | ||
3809 | r = strv_consume_pair(&symlinks, TAKE_PTR(private_path), TAKE_PTR(path)); | |
3810 | if (r < 0) | |
3811 | return r; | |
3812 | } | |
3813 | } | |
3814 | ||
3815 | *ret_symlinks = TAKE_PTR(symlinks); | |
3816 | ||
3817 | return 0; | |
3818 | } | |
3819 | ||
4e677599 LP |
3820 | static bool insist_on_sandboxing( |
3821 | const ExecContext *context, | |
3822 | const char *root_dir, | |
3823 | const char *root_image, | |
3824 | const BindMount *bind_mounts, | |
3825 | size_t n_bind_mounts) { | |
3826 | ||
4e677599 LP |
3827 | assert(context); |
3828 | assert(n_bind_mounts == 0 || bind_mounts); | |
3829 | ||
3830 | /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that | |
86b52a39 | 3831 | * would alter the view on the file system beyond making things read-only or invisible, i.e. would |
4e677599 LP |
3832 | * rearrange stuff in a way we cannot ignore gracefully. */ |
3833 | ||
3834 | if (context->n_temporary_filesystems > 0) | |
3835 | return true; | |
3836 | ||
3837 | if (root_dir || root_image) | |
3838 | return true; | |
3839 | ||
b3d13314 LB |
3840 | if (context->n_mount_images > 0) |
3841 | return true; | |
3842 | ||
4e677599 LP |
3843 | if (context->dynamic_user) |
3844 | return true; | |
3845 | ||
4355c04f LB |
3846 | if (context->n_extension_images > 0 || !strv_isempty(context->extension_directories)) |
3847 | return true; | |
3848 | ||
4e677599 LP |
3849 | /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes |
3850 | * essential. */ | |
5b10116e | 3851 | for (size_t i = 0; i < n_bind_mounts; i++) |
4e677599 LP |
3852 | if (!path_equal(bind_mounts[i].source, bind_mounts[i].destination)) |
3853 | return true; | |
3854 | ||
91dd5f7c LP |
3855 | if (context->log_namespace) |
3856 | return true; | |
3857 | ||
4e677599 LP |
3858 | return false; |
3859 | } | |
3860 | ||
9c0c6701 DDM |
3861 | static int setup_ephemeral(const ExecContext *context, ExecRuntime *runtime) { |
3862 | _cleanup_close_ int fd = -EBADF; | |
3863 | int r; | |
3864 | ||
3865 | if (!runtime || !runtime->ephemeral_copy) | |
3866 | return 0; | |
3867 | ||
3868 | r = posix_lock(runtime->ephemeral_storage_socket[0], LOCK_EX); | |
3869 | if (r < 0) | |
3870 | return log_debug_errno(r, "Failed to lock ephemeral storage socket: %m"); | |
3871 | ||
3872 | CLEANUP_POSIX_UNLOCK(runtime->ephemeral_storage_socket[0]); | |
3873 | ||
3874 | fd = receive_one_fd(runtime->ephemeral_storage_socket[0], MSG_PEEK|MSG_DONTWAIT); | |
3875 | if (fd >= 0) | |
3876 | /* We got an fd! That means ephemeral has already been set up, so nothing to do here. */ | |
3877 | return 0; | |
3878 | ||
3879 | if (fd != -EAGAIN) | |
3880 | return log_debug_errno(fd, "Failed to receive file descriptor queued on ephemeral storage socket: %m"); | |
3881 | ||
3882 | log_debug("Making ephemeral snapshot of %s to %s", | |
3883 | context->root_image ?: context->root_directory, runtime->ephemeral_copy); | |
3884 | ||
3885 | if (context->root_image) | |
3886 | fd = copy_file(context->root_image, runtime->ephemeral_copy, O_EXCL, 0600, | |
3887 | COPY_LOCK_BSD|COPY_REFLINK|COPY_CRTIME); | |
3888 | else | |
3889 | fd = btrfs_subvol_snapshot_at(AT_FDCWD, context->root_directory, | |
3890 | AT_FDCWD, runtime->ephemeral_copy, | |
3891 | BTRFS_SNAPSHOT_FALLBACK_COPY | | |
3892 | BTRFS_SNAPSHOT_FALLBACK_DIRECTORY | | |
3893 | BTRFS_SNAPSHOT_RECURSIVE | | |
3894 | BTRFS_SNAPSHOT_LOCK_BSD); | |
3895 | if (fd < 0) | |
3896 | return log_debug_errno(fd, "Failed to snapshot %s to %s: %m", | |
3897 | context->root_image ?: context->root_directory, runtime->ephemeral_copy); | |
3898 | ||
3899 | if (context->root_image) { | |
3900 | /* A root image might be subject to lots of random writes so let's try to disable COW on it | |
3901 | * which tends to not perform well in combination with lots of random writes. | |
3902 | * | |
3903 | * Note: btrfs actually isn't impressed by us setting the flag after making the reflink'ed | |
3904 | * copy, but we at least want to make the intention clear. | |
3905 | */ | |
3906 | r = chattr_fd(fd, FS_NOCOW_FL, FS_NOCOW_FL, NULL); | |
3907 | if (r < 0) | |
3908 | log_debug_errno(fd, "Failed to disable copy-on-write for %s, ignoring: %m", runtime->ephemeral_copy); | |
3909 | } | |
3910 | ||
3911 | r = send_one_fd(runtime->ephemeral_storage_socket[1], fd, MSG_DONTWAIT); | |
3912 | if (r < 0) | |
3913 | return log_debug_errno(r, "Failed to queue file descriptor on ephemeral storage socket: %m"); | |
3914 | ||
3915 | return 1; | |
3916 | } | |
3917 | ||
66130f0a DDM |
3918 | static int verity_settings_prepare( |
3919 | VeritySettings *verity, | |
3920 | const char *root_image, | |
3921 | const void *root_hash, | |
3922 | size_t root_hash_size, | |
3923 | const char *root_hash_path, | |
3924 | const void *root_hash_sig, | |
3925 | size_t root_hash_sig_size, | |
3926 | const char *root_hash_sig_path, | |
3927 | const char *verity_data_path) { | |
3928 | ||
3929 | int r; | |
3930 | ||
3931 | assert(verity); | |
3932 | ||
3933 | if (root_hash) { | |
3934 | void *d; | |
3935 | ||
3936 | d = memdup(root_hash, root_hash_size); | |
3937 | if (!d) | |
3938 | return -ENOMEM; | |
3939 | ||
3940 | free_and_replace(verity->root_hash, d); | |
3941 | verity->root_hash_size = root_hash_size; | |
3942 | verity->designator = PARTITION_ROOT; | |
3943 | } | |
3944 | ||
3945 | if (root_hash_sig) { | |
3946 | void *d; | |
3947 | ||
3948 | d = memdup(root_hash_sig, root_hash_sig_size); | |
3949 | if (!d) | |
3950 | return -ENOMEM; | |
3951 | ||
3952 | free_and_replace(verity->root_hash_sig, d); | |
3953 | verity->root_hash_sig_size = root_hash_sig_size; | |
3954 | verity->designator = PARTITION_ROOT; | |
3955 | } | |
3956 | ||
3957 | if (verity_data_path) { | |
3958 | r = free_and_strdup(&verity->data_path, verity_data_path); | |
3959 | if (r < 0) | |
3960 | return r; | |
3961 | } | |
3962 | ||
3963 | r = verity_settings_load( | |
3964 | verity, | |
3965 | root_image, | |
3966 | root_hash_path, | |
3967 | root_hash_sig_path); | |
3968 | if (r < 0) | |
3969 | return log_debug_errno(r, "Failed to load root hash: %m"); | |
3970 | ||
3971 | return 0; | |
3972 | } | |
3973 | ||
6818c54c | 3974 | static int apply_mount_namespace( |
34cf6c43 | 3975 | const Unit *u, |
9f71ba8d | 3976 | ExecCommandFlags command_flags, |
6818c54c LP |
3977 | const ExecContext *context, |
3978 | const ExecParameters *params, | |
9c0c6701 | 3979 | ExecRuntime *runtime, |
d4b6ec98 | 3980 | const char *memory_pressure_path, |
7cc5ef5f | 3981 | char **error_path) { |
6818c54c | 3982 | |
66130f0a | 3983 | _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT; |
d4b6ec98 LB |
3984 | _cleanup_strv_free_ char **empty_directories = NULL, **symlinks = NULL, |
3985 | **read_write_paths_cleanup = NULL; | |
24759d8f | 3986 | _cleanup_free_ char *creds_path = NULL, *incoming_dir = NULL, *propagate_dir = NULL, |
3f37a825 | 3987 | *extension_dir = NULL, *host_os_release = NULL; |
66130f0a | 3988 | const char *root_dir = NULL, *root_image = NULL, *tmp_dir = NULL, *var_tmp_dir = NULL; |
d4b6ec98 | 3989 | char **read_write_paths; |
228af36f | 3990 | NamespaceInfo ns_info; |
165a31c0 | 3991 | bool needs_sandboxing; |
6c47cd7d | 3992 | BindMount *bind_mounts = NULL; |
da6053d0 | 3993 | size_t n_bind_mounts = 0; |
6818c54c | 3994 | int r; |
93c6bb51 | 3995 | |
2b3c1b9e DH |
3996 | assert(context); |
3997 | ||
29933daf DT |
3998 | CLEANUP_ARRAY(bind_mounts, n_bind_mounts, bind_mount_free_many); |
3999 | ||
915e6d16 | 4000 | if (params->flags & EXEC_APPLY_CHROOT) { |
9c0c6701 DDM |
4001 | r = setup_ephemeral(context, runtime); |
4002 | if (r < 0) | |
4003 | return r; | |
915e6d16 | 4004 | |
9c0c6701 DDM |
4005 | if (context->root_image) |
4006 | root_image = (runtime ? runtime->ephemeral_copy : NULL) ?: context->root_image; | |
4007 | else | |
4008 | root_dir = (runtime ? runtime->ephemeral_copy : NULL) ?: context->root_directory; | |
915e6d16 | 4009 | } |
93c6bb51 | 4010 | |
6c47cd7d LP |
4011 | r = compile_bind_mounts(context, params, &bind_mounts, &n_bind_mounts, &empty_directories); |
4012 | if (r < 0) | |
4013 | return r; | |
4014 | ||
211a3d87 | 4015 | /* Symlinks for exec dirs are set up after other mounts, before they are made read-only. */ |
df61e79a LB |
4016 | r = compile_symlinks(context, params, &symlinks); |
4017 | if (r < 0) | |
29933daf | 4018 | return r; |
df61e79a | 4019 | |
d4b6ec98 LB |
4020 | /* We need to make the pressure path writable even if /sys/fs/cgroups is made read-only, as the |
4021 | * service will need to write to it in order to start the notifications. */ | |
4022 | if (context->protect_control_groups && memory_pressure_path && !streq(memory_pressure_path, "/dev/null")) { | |
4023 | read_write_paths_cleanup = strv_copy(context->read_write_paths); | |
29933daf DT |
4024 | if (!read_write_paths_cleanup) |
4025 | return -ENOMEM; | |
d4b6ec98 LB |
4026 | |
4027 | r = strv_extend(&read_write_paths_cleanup, memory_pressure_path); | |
4028 | if (r < 0) | |
29933daf | 4029 | return r; |
d4b6ec98 LB |
4030 | |
4031 | read_write_paths = read_write_paths_cleanup; | |
4032 | } else | |
4033 | read_write_paths = context->read_write_paths; | |
4034 | ||
9f71ba8d | 4035 | needs_sandboxing = (params->flags & EXEC_APPLY_SANDBOXING) && !(command_flags & EXEC_COMMAND_FULLY_PRIVILEGED); |
ecf63c91 NJ |
4036 | if (needs_sandboxing) { |
4037 | /* The runtime struct only contains the parent of the private /tmp, | |
4038 | * which is non-accessible to world users. Inside of it there's a /tmp | |
56a13a49 ZJS |
4039 | * that is sticky, and that's the one we want to use here. |
4040 | * This does not apply when we are using /run/systemd/empty as fallback. */ | |
ecf63c91 | 4041 | |
28135da3 DDM |
4042 | if (context->private_tmp && runtime && runtime->shared) { |
4043 | if (streq_ptr(runtime->shared->tmp_dir, RUN_SYSTEMD_EMPTY)) | |
4044 | tmp_dir = runtime->shared->tmp_dir; | |
4045 | else if (runtime->shared->tmp_dir) | |
4046 | tmp_dir = strjoina(runtime->shared->tmp_dir, "/tmp"); | |
56a13a49 | 4047 | |
28135da3 DDM |
4048 | if (streq_ptr(runtime->shared->var_tmp_dir, RUN_SYSTEMD_EMPTY)) |
4049 | var_tmp_dir = runtime->shared->var_tmp_dir; | |
4050 | else if (runtime->shared->var_tmp_dir) | |
4051 | var_tmp_dir = strjoina(runtime->shared->var_tmp_dir, "/tmp"); | |
ecf63c91 NJ |
4052 | } |
4053 | ||
b5a33299 YW |
4054 | ns_info = (NamespaceInfo) { |
4055 | .ignore_protect_paths = false, | |
4056 | .private_dev = context->private_devices, | |
4057 | .protect_control_groups = context->protect_control_groups, | |
4058 | .protect_kernel_tunables = context->protect_kernel_tunables, | |
4059 | .protect_kernel_modules = context->protect_kernel_modules, | |
94a7b275 | 4060 | .protect_kernel_logs = context->protect_kernel_logs, |
aecd5ac6 | 4061 | .protect_hostname = context->protect_hostname, |
5e98086d | 4062 | .mount_apivfs = exec_context_get_effective_mount_apivfs(context), |
52b3d652 LP |
4063 | .protect_home = context->protect_home, |
4064 | .protect_system = context->protect_system, | |
4e399953 LP |
4065 | .protect_proc = context->protect_proc, |
4066 | .proc_subset = context->proc_subset, | |
c2da3bf2 | 4067 | .private_network = exec_needs_network_namespace(context), |
fde36d25 | 4068 | .private_ipc = exec_needs_ipc_namespace(context), |
6720e356 | 4069 | /* If NNP is on, we can turn on MS_NOSUID, since it won't have any effect anymore. */ |
5181630f | 4070 | .mount_nosuid = context->no_new_privileges && !mac_selinux_use(), |
b5a33299 | 4071 | }; |
ecf63c91 | 4072 | } else if (!context->dynamic_user && root_dir) |
228af36f LP |
4073 | /* |
4074 | * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed | |
4075 | * sandbox info, otherwise enforce it, don't ignore protected paths and | |
4076 | * fail if we are enable to apply the sandbox inside the mount namespace. | |
4077 | */ | |
4078 | ns_info = (NamespaceInfo) { | |
4079 | .ignore_protect_paths = true, | |
4080 | }; | |
4081 | else | |
4082 | ns_info = (NamespaceInfo) {}; | |
b5a33299 | 4083 | |
874cdcbc | 4084 | if (context->mount_propagation_flag == MS_SHARED) |
37ed15d7 FB |
4085 | log_unit_debug(u, "shared mount propagation hidden by other fs namespacing unit settings: ignoring"); |
4086 | ||
a631cbfa LP |
4087 | if (exec_context_has_credentials(context) && |
4088 | params->prefix[EXEC_DIRECTORY_RUNTIME] && | |
4089 | FLAGS_SET(params->flags, EXEC_WRITE_CREDENTIALS)) { | |
bbb4e7f3 | 4090 | creds_path = path_join(params->prefix[EXEC_DIRECTORY_RUNTIME], "credentials", u->id); |
29933daf DT |
4091 | if (!creds_path) |
4092 | return -ENOMEM; | |
bbb4e7f3 LP |
4093 | } |
4094 | ||
170d978b | 4095 | if (params->runtime_scope == RUNTIME_SCOPE_SYSTEM) { |
5e8deb94 | 4096 | propagate_dir = path_join("/run/systemd/propagate/", u->id); |
29933daf DT |
4097 | if (!propagate_dir) |
4098 | return -ENOMEM; | |
f2550b98 | 4099 | |
5e8deb94 | 4100 | incoming_dir = strdup("/run/systemd/incoming"); |
29933daf DT |
4101 | if (!incoming_dir) |
4102 | return -ENOMEM; | |
24759d8f LB |
4103 | |
4104 | extension_dir = strdup("/run/systemd/unit-extensions"); | |
29933daf DT |
4105 | if (!extension_dir) |
4106 | return -ENOMEM; | |
3f37a825 LB |
4107 | |
4108 | /* If running under a different root filesystem, propagate the host's os-release. We make a | |
4109 | * copy rather than just bind mounting it, so that it can be updated on soft-reboot. */ | |
4110 | if (root_dir || root_image) { | |
4111 | host_os_release = strdup("/run/systemd/propagate/os-release"); | |
4112 | if (!host_os_release) | |
4113 | return -ENOMEM; | |
4114 | } | |
170d978b LP |
4115 | } else { |
4116 | assert(params->runtime_scope == RUNTIME_SCOPE_USER); | |
4117 | ||
29933daf DT |
4118 | if (asprintf(&extension_dir, "/run/user/" UID_FMT "/systemd/unit-extensions", geteuid()) < 0) |
4119 | return -ENOMEM; | |
3f37a825 LB |
4120 | |
4121 | if (root_dir || root_image) { | |
4122 | if (asprintf(&host_os_release, "/run/user/" UID_FMT "/systemd/propagate/os-release", geteuid()) < 0) | |
4123 | return -ENOMEM; | |
4124 | } | |
170d978b | 4125 | } |
5e8deb94 | 4126 | |
66130f0a DDM |
4127 | if (root_image) { |
4128 | r = verity_settings_prepare( | |
4129 | &verity, | |
4130 | root_image, | |
4131 | context->root_hash, context->root_hash_size, context->root_hash_path, | |
4132 | context->root_hash_sig, context->root_hash_sig_size, context->root_hash_sig_path, | |
4133 | context->root_verity); | |
4134 | if (r < 0) | |
4135 | return r; | |
4136 | } | |
4137 | ||
84be0c71 LP |
4138 | r = setup_namespace( |
4139 | root_dir, | |
4140 | root_image, | |
4141 | context->root_image_options, | |
4142 | context->root_image_policy ?: &image_policy_service, | |
4143 | &ns_info, | |
4144 | read_write_paths, | |
4145 | needs_sandboxing ? context->read_only_paths : NULL, | |
4146 | needs_sandboxing ? context->inaccessible_paths : NULL, | |
4147 | needs_sandboxing ? context->exec_paths : NULL, | |
4148 | needs_sandboxing ? context->no_exec_paths : NULL, | |
4149 | empty_directories, | |
4150 | symlinks, | |
4151 | bind_mounts, | |
4152 | n_bind_mounts, | |
4153 | context->temporary_filesystems, | |
4154 | context->n_temporary_filesystems, | |
4155 | context->mount_images, | |
4156 | context->n_mount_images, | |
4157 | context->mount_image_policy ?: &image_policy_service, | |
4158 | tmp_dir, | |
4159 | var_tmp_dir, | |
4160 | creds_path, | |
4161 | context->log_namespace, | |
4162 | context->mount_propagation_flag, | |
66130f0a | 4163 | &verity, |
84be0c71 LP |
4164 | context->extension_images, |
4165 | context->n_extension_images, | |
4166 | context->extension_image_policy ?: &image_policy_sysext, | |
4167 | context->extension_directories, | |
4168 | propagate_dir, | |
4169 | incoming_dir, | |
4170 | extension_dir, | |
4171 | root_dir || root_image ? params->notify_socket : NULL, | |
3f37a825 | 4172 | host_os_release, |
84be0c71 | 4173 | error_path); |
93c6bb51 | 4174 | |
1beab8b0 | 4175 | /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports |
5238e957 | 4176 | * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively |
1beab8b0 LP |
4177 | * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a |
4178 | * completely different execution environment. */ | |
aca835ed | 4179 | if (r == -ENOANO) { |
4e677599 LP |
4180 | if (insist_on_sandboxing( |
4181 | context, | |
4182 | root_dir, root_image, | |
4183 | bind_mounts, | |
29933daf DT |
4184 | n_bind_mounts)) |
4185 | return log_unit_debug_errno(u, | |
4186 | SYNTHETIC_ERRNO(EOPNOTSUPP), | |
4187 | "Failed to set up namespace, and refusing to continue since " | |
4188 | "the selected namespacing options alter mount environment non-trivially.\n" | |
4189 | "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s", | |
4190 | n_bind_mounts, | |
4191 | context->n_temporary_filesystems, | |
4192 | yes_no(root_dir), | |
4193 | yes_no(root_image), | |
4194 | yes_no(context->dynamic_user)); | |
4195 | ||
4196 | log_unit_debug(u, "Failed to set up namespace, assuming containerized execution and ignoring."); | |
4197 | return 0; | |
93c6bb51 DH |
4198 | } |
4199 | ||
4200 | return r; | |
4201 | } | |
4202 | ||
915e6d16 LP |
4203 | static int apply_working_directory( |
4204 | const ExecContext *context, | |
4205 | const ExecParameters *params, | |
9c0c6701 | 4206 | ExecRuntime *runtime, |
915e6d16 | 4207 | const char *home, |
376fecf6 | 4208 | int *exit_status) { |
915e6d16 | 4209 | |
6732edab | 4210 | const char *d, *wd; |
2b3c1b9e DH |
4211 | |
4212 | assert(context); | |
376fecf6 | 4213 | assert(exit_status); |
2b3c1b9e | 4214 | |
6732edab LP |
4215 | if (context->working_directory_home) { |
4216 | ||
376fecf6 LP |
4217 | if (!home) { |
4218 | *exit_status = EXIT_CHDIR; | |
6732edab | 4219 | return -ENXIO; |
376fecf6 | 4220 | } |
6732edab | 4221 | |
2b3c1b9e | 4222 | wd = home; |
6732edab | 4223 | |
14eb3285 LP |
4224 | } else |
4225 | wd = empty_to_root(context->working_directory); | |
e7f1e7c6 | 4226 | |
fa97f630 | 4227 | if (params->flags & EXEC_APPLY_CHROOT) |
2b3c1b9e | 4228 | d = wd; |
fa97f630 | 4229 | else |
9c0c6701 | 4230 | d = prefix_roota((runtime ? runtime->ephemeral_copy : NULL) ?: context->root_directory, wd); |
e7f1e7c6 | 4231 | |
376fecf6 LP |
4232 | if (chdir(d) < 0 && !context->working_directory_missing_ok) { |
4233 | *exit_status = EXIT_CHDIR; | |
2b3c1b9e | 4234 | return -errno; |
376fecf6 | 4235 | } |
e7f1e7c6 DH |
4236 | |
4237 | return 0; | |
4238 | } | |
4239 | ||
fa97f630 JB |
4240 | static int apply_root_directory( |
4241 | const ExecContext *context, | |
4242 | const ExecParameters *params, | |
9c0c6701 | 4243 | ExecRuntime *runtime, |
fa97f630 JB |
4244 | const bool needs_mount_ns, |
4245 | int *exit_status) { | |
4246 | ||
4247 | assert(context); | |
4248 | assert(exit_status); | |
4249 | ||
5b10116e | 4250 | if (params->flags & EXEC_APPLY_CHROOT) |
fa97f630 | 4251 | if (!needs_mount_ns && context->root_directory) |
9c0c6701 | 4252 | if (chroot((runtime ? runtime->ephemeral_copy : NULL) ?: context->root_directory) < 0) { |
fa97f630 JB |
4253 | *exit_status = EXIT_CHROOT; |
4254 | return -errno; | |
4255 | } | |
fa97f630 JB |
4256 | |
4257 | return 0; | |
4258 | } | |
4259 | ||
b1edf445 | 4260 | static int setup_keyring( |
34cf6c43 | 4261 | const Unit *u, |
b1edf445 LP |
4262 | const ExecContext *context, |
4263 | const ExecParameters *p, | |
4264 | uid_t uid, gid_t gid) { | |
4265 | ||
74dd6b51 | 4266 | key_serial_t keyring; |
e64c2d0b DJL |
4267 | int r = 0; |
4268 | uid_t saved_uid; | |
4269 | gid_t saved_gid; | |
74dd6b51 LP |
4270 | |
4271 | assert(u); | |
b1edf445 | 4272 | assert(context); |
74dd6b51 LP |
4273 | assert(p); |
4274 | ||
4275 | /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that | |
4276 | * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond | |
4277 | * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be | |
4278 | * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in | |
4279 | * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where | |
4280 | * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */ | |
4281 | ||
b1edf445 LP |
4282 | if (context->keyring_mode == EXEC_KEYRING_INHERIT) |
4283 | return 0; | |
4284 | ||
e64c2d0b DJL |
4285 | /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up |
4286 | * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel | |
4287 | * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user | |
4288 | * & group is just as nasty as acquiring a reference to the user keyring. */ | |
4289 | ||
4290 | saved_uid = getuid(); | |
4291 | saved_gid = getgid(); | |
4292 | ||
4293 | if (gid_is_valid(gid) && gid != saved_gid) { | |
4294 | if (setregid(gid, -1) < 0) | |
4295 | return log_unit_error_errno(u, errno, "Failed to change GID for user keyring: %m"); | |
4296 | } | |
4297 | ||
4298 | if (uid_is_valid(uid) && uid != saved_uid) { | |
4299 | if (setreuid(uid, -1) < 0) { | |
4300 | r = log_unit_error_errno(u, errno, "Failed to change UID for user keyring: %m"); | |
4301 | goto out; | |
4302 | } | |
4303 | } | |
4304 | ||
74dd6b51 LP |
4305 | keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0); |
4306 | if (keyring == -1) { | |
4307 | if (errno == ENOSYS) | |
8002fb97 | 4308 | log_unit_debug_errno(u, errno, "Kernel keyring not supported, ignoring."); |
065b4774 | 4309 | else if (ERRNO_IS_PRIVILEGE(errno)) |
8002fb97 | 4310 | log_unit_debug_errno(u, errno, "Kernel keyring access prohibited, ignoring."); |
74dd6b51 | 4311 | else if (errno == EDQUOT) |
8002fb97 | 4312 | log_unit_debug_errno(u, errno, "Out of kernel keyrings to allocate, ignoring."); |
74dd6b51 | 4313 | else |
e64c2d0b | 4314 | r = log_unit_error_errno(u, errno, "Setting up kernel keyring failed: %m"); |
74dd6b51 | 4315 | |
e64c2d0b | 4316 | goto out; |
74dd6b51 LP |
4317 | } |
4318 | ||
e64c2d0b DJL |
4319 | /* When requested link the user keyring into the session keyring. */ |
4320 | if (context->keyring_mode == EXEC_KEYRING_SHARED) { | |
4321 | ||
4322 | if (keyctl(KEYCTL_LINK, | |
4323 | KEY_SPEC_USER_KEYRING, | |
4324 | KEY_SPEC_SESSION_KEYRING, 0, 0) < 0) { | |
4325 | r = log_unit_error_errno(u, errno, "Failed to link user keyring into session keyring: %m"); | |
4326 | goto out; | |
4327 | } | |
4328 | } | |
4329 | ||
4330 | /* Restore uid/gid back */ | |
4331 | if (uid_is_valid(uid) && uid != saved_uid) { | |
4332 | if (setreuid(saved_uid, -1) < 0) { | |
4333 | r = log_unit_error_errno(u, errno, "Failed to change UID back for user keyring: %m"); | |
4334 | goto out; | |
4335 | } | |
4336 | } | |
4337 | ||
4338 | if (gid_is_valid(gid) && gid != saved_gid) { | |
4339 | if (setregid(saved_gid, -1) < 0) | |
4340 | return log_unit_error_errno(u, errno, "Failed to change GID back for user keyring: %m"); | |
4341 | } | |
4342 | ||
4343 | /* Populate they keyring with the invocation ID by default, as original saved_uid. */ | |
b3415f5d LP |
4344 | if (!sd_id128_is_null(u->invocation_id)) { |
4345 | key_serial_t key; | |
4346 | ||
4347 | key = add_key("user", "invocation_id", &u->invocation_id, sizeof(u->invocation_id), KEY_SPEC_SESSION_KEYRING); | |
4348 | if (key == -1) | |
8002fb97 | 4349 | log_unit_debug_errno(u, errno, "Failed to add invocation ID to keyring, ignoring: %m"); |
b3415f5d LP |
4350 | else { |
4351 | if (keyctl(KEYCTL_SETPERM, key, | |
4352 | KEY_POS_VIEW|KEY_POS_READ|KEY_POS_SEARCH| | |
4353 | KEY_USR_VIEW|KEY_USR_READ|KEY_USR_SEARCH, 0, 0) < 0) | |
e64c2d0b | 4354 | r = log_unit_error_errno(u, errno, "Failed to restrict invocation ID permission: %m"); |
b3415f5d LP |
4355 | } |
4356 | } | |
4357 | ||
e64c2d0b | 4358 | out: |
37b22b3b | 4359 | /* Revert back uid & gid for the last time, and exit */ |
e64c2d0b DJL |
4360 | /* no extra logging, as only the first already reported error matters */ |
4361 | if (getuid() != saved_uid) | |
4362 | (void) setreuid(saved_uid, -1); | |
b1edf445 | 4363 | |
e64c2d0b DJL |
4364 | if (getgid() != saved_gid) |
4365 | (void) setregid(saved_gid, -1); | |
b1edf445 | 4366 | |
e64c2d0b | 4367 | return r; |
74dd6b51 LP |
4368 | } |
4369 | ||
3042bbeb | 4370 | static void append_socket_pair(int *array, size_t *n, const int pair[static 2]) { |
29206d46 LP |
4371 | assert(array); |
4372 | assert(n); | |
2caa38e9 | 4373 | assert(pair); |
29206d46 LP |
4374 | |
4375 | if (pair[0] >= 0) | |
4376 | array[(*n)++] = pair[0]; | |
4377 | if (pair[1] >= 0) | |
4378 | array[(*n)++] = pair[1]; | |
4379 | } | |
4380 | ||
a34ceba6 LP |
4381 | static int close_remaining_fds( |
4382 | const ExecParameters *params, | |
28135da3 | 4383 | const ExecRuntime *runtime, |
00d9ef85 | 4384 | int user_lookup_fd, |
a34ceba6 | 4385 | int socket_fd, |
5b8d1f6b | 4386 | const int *fds, size_t n_fds) { |
a34ceba6 | 4387 | |
da6053d0 | 4388 | size_t n_dont_close = 0; |
9c0c6701 | 4389 | int dont_close[n_fds + 14]; |
a34ceba6 LP |
4390 | |
4391 | assert(params); | |
4392 | ||
4393 | if (params->stdin_fd >= 0) | |
4394 | dont_close[n_dont_close++] = params->stdin_fd; | |
4395 | if (params->stdout_fd >= 0) | |
4396 | dont_close[n_dont_close++] = params->stdout_fd; | |
4397 | if (params->stderr_fd >= 0) | |
4398 | dont_close[n_dont_close++] = params->stderr_fd; | |
4399 | ||
4400 | if (socket_fd >= 0) | |
4401 | dont_close[n_dont_close++] = socket_fd; | |
4402 | if (n_fds > 0) { | |
4403 | memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds); | |
4404 | n_dont_close += n_fds; | |
4405 | } | |
4406 | ||
9c0c6701 DDM |
4407 | if (runtime) |
4408 | append_socket_pair(dont_close, &n_dont_close, runtime->ephemeral_storage_socket); | |
4409 | ||
28135da3 DDM |
4410 | if (runtime && runtime->shared) { |
4411 | append_socket_pair(dont_close, &n_dont_close, runtime->shared->netns_storage_socket); | |
4412 | append_socket_pair(dont_close, &n_dont_close, runtime->shared->ipcns_storage_socket); | |
a70581ff | 4413 | } |
29206d46 | 4414 | |
15220772 DDM |
4415 | if (runtime && runtime->dynamic_creds) { |
4416 | if (runtime->dynamic_creds->user) | |
4417 | append_socket_pair(dont_close, &n_dont_close, runtime->dynamic_creds->user->storage_socket); | |
4418 | if (runtime->dynamic_creds->group) | |
4419 | append_socket_pair(dont_close, &n_dont_close, runtime->dynamic_creds->group->storage_socket); | |
a34ceba6 LP |
4420 | } |
4421 | ||
00d9ef85 LP |
4422 | if (user_lookup_fd >= 0) |
4423 | dont_close[n_dont_close++] = user_lookup_fd; | |
4424 | ||
a34ceba6 LP |
4425 | return close_all_fds(dont_close, n_dont_close); |
4426 | } | |
4427 | ||
00d9ef85 LP |
4428 | static int send_user_lookup( |
4429 | Unit *unit, | |
4430 | int user_lookup_fd, | |
4431 | uid_t uid, | |
4432 | gid_t gid) { | |
4433 | ||
4434 | assert(unit); | |
4435 | ||
4436 | /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID | |
4437 | * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was | |
4438 | * specified. */ | |
4439 | ||
4440 | if (user_lookup_fd < 0) | |
4441 | return 0; | |
4442 | ||
4443 | if (!uid_is_valid(uid) && !gid_is_valid(gid)) | |
4444 | return 0; | |
4445 | ||
4446 | if (writev(user_lookup_fd, | |
4447 | (struct iovec[]) { | |
ce16d177 YW |
4448 | IOVEC_MAKE(&uid, sizeof(uid)), |
4449 | IOVEC_MAKE(&gid, sizeof(gid)), | |
4450 | IOVEC_MAKE_STRING(unit->id) }, 3) < 0) | |
00d9ef85 LP |
4451 | return -errno; |
4452 | ||
4453 | return 0; | |
4454 | } | |
4455 | ||
6732edab LP |
4456 | static int acquire_home(const ExecContext *c, uid_t uid, const char** home, char **buf) { |
4457 | int r; | |
4458 | ||
4459 | assert(c); | |
4460 | assert(home); | |
4461 | assert(buf); | |
4462 | ||
4463 | /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */ | |
4464 | ||
4465 | if (*home) | |
4466 | return 0; | |
4467 | ||
4468 | if (!c->working_directory_home) | |
4469 | return 0; | |
4470 | ||
6732edab LP |
4471 | r = get_home_dir(buf); |
4472 | if (r < 0) | |
4473 | return r; | |
4474 | ||
4475 | *home = *buf; | |
4476 | return 1; | |
4477 | } | |
4478 | ||
da50b85a LP |
4479 | static int compile_suggested_paths(const ExecContext *c, const ExecParameters *p, char ***ret) { |
4480 | _cleanup_strv_free_ char ** list = NULL; | |
da50b85a LP |
4481 | int r; |
4482 | ||
4483 | assert(c); | |
4484 | assert(p); | |
4485 | assert(ret); | |
4486 | ||
4487 | assert(c->dynamic_user); | |
4488 | ||
4489 | /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for | |
4490 | * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special | |
4491 | * directories. */ | |
4492 | ||
5b10116e | 4493 | for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) { |
da50b85a LP |
4494 | if (t == EXEC_DIRECTORY_CONFIGURATION) |
4495 | continue; | |
4496 | ||
4497 | if (!p->prefix[t]) | |
4498 | continue; | |
4499 | ||
211a3d87 | 4500 | for (size_t i = 0; i < c->directories[t].n_items; i++) { |
da50b85a LP |
4501 | char *e; |
4502 | ||
494d0247 | 4503 | if (exec_directory_is_private(c, t)) |
211a3d87 | 4504 | e = path_join(p->prefix[t], "private", c->directories[t].items[i].path); |
494d0247 | 4505 | else |
211a3d87 | 4506 | e = path_join(p->prefix[t], c->directories[t].items[i].path); |
da50b85a LP |
4507 | if (!e) |
4508 | return -ENOMEM; | |
4509 | ||
4510 | r = strv_consume(&list, e); | |
4511 | if (r < 0) | |
4512 | return r; | |
4513 | } | |
4514 | } | |
4515 | ||
ae2a15bc | 4516 | *ret = TAKE_PTR(list); |
da50b85a LP |
4517 | |
4518 | return 0; | |
4519 | } | |
4520 | ||
a8b993dc LP |
4521 | static int exec_parameters_get_cgroup_path( |
4522 | const ExecParameters *params, | |
4523 | const CGroupContext *c, | |
4524 | char **ret) { | |
4525 | ||
4526 | const char *subgroup = NULL; | |
78f93209 LP |
4527 | char *p; |
4528 | ||
4529 | assert(params); | |
4530 | assert(ret); | |
4531 | ||
4532 | if (!params->cgroup_path) | |
4533 | return -EINVAL; | |
4534 | ||
4535 | /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated | |
4536 | * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control | |
4537 | * processes started after the main unit's process in the unit's main cgroup because it is now an inner one, | |
4538 | * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process, | |
4539 | * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=, | |
4540 | * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre= | |
4541 | * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP | |
4542 | * flag, which is only passed for the former statements, not for the latter. */ | |
4543 | ||
a8b993dc LP |
4544 | if (FLAGS_SET(params->flags, EXEC_CGROUP_DELEGATE) && (FLAGS_SET(params->flags, EXEC_CONTROL_CGROUP) || c->delegate_subgroup)) { |
4545 | if (FLAGS_SET(params->flags, EXEC_IS_CONTROL)) | |
4546 | subgroup = ".control"; | |
4547 | else | |
4548 | subgroup = c->delegate_subgroup; | |
4549 | } | |
4550 | ||
4551 | if (subgroup) | |
4552 | p = path_join(params->cgroup_path, subgroup); | |
78f93209 LP |
4553 | else |
4554 | p = strdup(params->cgroup_path); | |
4555 | if (!p) | |
4556 | return -ENOMEM; | |
4557 | ||
4558 | *ret = p; | |
a8b993dc | 4559 | return !!subgroup; |
78f93209 LP |
4560 | } |
4561 | ||
e2b2fb7f MS |
4562 | static int exec_context_cpu_affinity_from_numa(const ExecContext *c, CPUSet *ret) { |
4563 | _cleanup_(cpu_set_reset) CPUSet s = {}; | |
4564 | int r; | |
4565 | ||
4566 | assert(c); | |
4567 | assert(ret); | |
4568 | ||
4569 | if (!c->numa_policy.nodes.set) { | |
4570 | log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring"); | |
4571 | return 0; | |
4572 | } | |
4573 | ||
4574 | r = numa_to_cpu_set(&c->numa_policy, &s); | |
4575 | if (r < 0) | |
4576 | return r; | |
4577 | ||
4578 | cpu_set_reset(ret); | |
4579 | ||
4580 | return cpu_set_add_all(ret, &s); | |
4581 | } | |
4582 | ||
4583 | bool exec_context_get_cpu_affinity_from_numa(const ExecContext *c) { | |
4584 | assert(c); | |
4585 | ||
4586 | return c->cpu_affinity_from_numa; | |
4587 | } | |
4588 | ||
1da37e58 ZJS |
4589 | static int add_shifted_fd(int *fds, size_t fds_size, size_t *n_fds, int fd, int *ret_fd) { |
4590 | int r; | |
4591 | ||
4592 | assert(fds); | |
4593 | assert(n_fds); | |
4594 | assert(*n_fds < fds_size); | |
4595 | assert(ret_fd); | |
4596 | ||
4597 | if (fd < 0) { | |
254d1313 | 4598 | *ret_fd = -EBADF; |
1da37e58 ZJS |
4599 | return 0; |
4600 | } | |
4601 | ||
4602 | if (fd < 3 + (int) *n_fds) { | |
4603 | /* Let's move the fd up, so that it's outside of the fd range we will use to store | |
4604 | * the fds we pass to the process (or which are closed only during execve). */ | |
4605 | ||
4606 | r = fcntl(fd, F_DUPFD_CLOEXEC, 3 + (int) *n_fds); | |
4607 | if (r < 0) | |
4608 | return -errno; | |
4609 | ||
ee3455cf | 4610 | close_and_replace(fd, r); |
1da37e58 ZJS |
4611 | } |
4612 | ||
4613 | *ret_fd = fds[*n_fds] = fd; | |
4614 | (*n_fds) ++; | |
4615 | return 1; | |
4616 | } | |
4617 | ||
cd48e23f RP |
4618 | static int connect_unix_harder(Unit *u, const OpenFile *of, int ofd) { |
4619 | union sockaddr_union addr = { | |
4620 | .un.sun_family = AF_UNIX, | |
4621 | }; | |
4622 | socklen_t sa_len; | |
4623 | static const int socket_types[] = { SOCK_DGRAM, SOCK_STREAM, SOCK_SEQPACKET }; | |
4624 | int r; | |
4625 | ||
4626 | assert(u); | |
4627 | assert(of); | |
4628 | assert(ofd >= 0); | |
4629 | ||
4630 | r = sockaddr_un_set_path(&addr.un, FORMAT_PROC_FD_PATH(ofd)); | |
4631 | if (r < 0) | |
4632 | return log_unit_error_errno(u, r, "Failed to set sockaddr for %s: %m", of->path); | |
4633 | ||
4634 | sa_len = r; | |
4635 | ||
4636 | for (size_t i = 0; i < ELEMENTSOF(socket_types); i++) { | |
4637 | _cleanup_close_ int fd = -EBADF; | |
4638 | ||
4639 | fd = socket(AF_UNIX, socket_types[i] | SOCK_CLOEXEC, 0); | |
4640 | if (fd < 0) | |
4641 | return log_unit_error_errno(u, errno, "Failed to create socket for %s: %m", of->path); | |
4642 | ||
4643 | r = RET_NERRNO(connect(fd, &addr.sa, sa_len)); | |
4644 | if (r == -EPROTOTYPE) | |
4645 | continue; | |
4646 | if (r < 0) | |
4647 | return log_unit_error_errno(u, r, "Failed to connect socket for %s: %m", of->path); | |
4648 | ||
4649 | return TAKE_FD(fd); | |
4650 | } | |
4651 | ||
4652 | return log_unit_error_errno(u, SYNTHETIC_ERRNO(EPROTOTYPE), "Failed to connect socket for \"%s\".", of->path); | |
4653 | } | |
4654 | ||
4655 | static int get_open_file_fd(Unit *u, const OpenFile *of) { | |
4656 | struct stat st; | |
4657 | _cleanup_close_ int fd = -EBADF, ofd = -EBADF; | |
4658 | ||
4659 | assert(u); | |
4660 | assert(of); | |
4661 | ||
4662 | ofd = open(of->path, O_PATH | O_CLOEXEC); | |
4663 | if (ofd < 0) | |
dcebb015 DDM |
4664 | return log_unit_error_errno(u, errno, "Could not open \"%s\": %m", of->path); |
4665 | ||
cd48e23f | 4666 | if (fstat(ofd, &st) < 0) |
dcebb015 | 4667 | return log_unit_error_errno(u, errno, "Failed to stat %s: %m", of->path); |
cd48e23f RP |
4668 | |
4669 | if (S_ISSOCK(st.st_mode)) { | |
4670 | fd = connect_unix_harder(u, of, ofd); | |
4671 | if (fd < 0) | |
4672 | return fd; | |
4673 | ||
4674 | if (FLAGS_SET(of->flags, OPENFILE_READ_ONLY) && shutdown(fd, SHUT_WR) < 0) | |
dcebb015 DDM |
4675 | return log_unit_error_errno(u, errno, "Failed to shutdown send for socket %s: %m", |
4676 | of->path); | |
cd48e23f RP |
4677 | |
4678 | log_unit_debug(u, "socket %s opened (fd=%d)", of->path, fd); | |
4679 | } else { | |
4680 | int flags = FLAGS_SET(of->flags, OPENFILE_READ_ONLY) ? O_RDONLY : O_RDWR; | |
4681 | if (FLAGS_SET(of->flags, OPENFILE_APPEND)) | |
4682 | flags |= O_APPEND; | |
4683 | else if (FLAGS_SET(of->flags, OPENFILE_TRUNCATE)) | |
4684 | flags |= O_TRUNC; | |
4685 | ||
4686 | fd = fd_reopen(ofd, flags | O_CLOEXEC); | |
4687 | if (fd < 0) | |
4688 | return log_unit_error_errno(u, fd, "Failed to open file %s: %m", of->path); | |
4689 | ||
4690 | log_unit_debug(u, "file %s opened (fd=%d)", of->path, fd); | |
4691 | } | |
4692 | ||
4693 | return TAKE_FD(fd); | |
4694 | } | |
4695 | ||
4696 | static int collect_open_file_fds( | |
4697 | Unit *u, | |
4698 | OpenFile* open_files, | |
4699 | int **fds, | |
4700 | char ***fdnames, | |
4701 | size_t *n_fds) { | |
4702 | int r; | |
4703 | ||
4704 | assert(u); | |
4705 | assert(fds); | |
4706 | assert(fdnames); | |
4707 | assert(n_fds); | |
4708 | ||
4709 | LIST_FOREACH(open_files, of, open_files) { | |
4710 | _cleanup_close_ int fd = -EBADF; | |
4711 | ||
4712 | fd = get_open_file_fd(u, of); | |
4713 | if (fd < 0) { | |
4714 | if (FLAGS_SET(of->flags, OPENFILE_GRACEFUL)) { | |
4715 | log_unit_debug_errno(u, fd, "Failed to get OpenFile= file descriptor for %s, ignoring: %m", of->path); | |
4716 | continue; | |
4717 | } | |
4718 | ||
4719 | return fd; | |
4720 | } | |
4721 | ||
4722 | if (!GREEDY_REALLOC(*fds, *n_fds + 1)) | |
4723 | return -ENOMEM; | |
4724 | ||
4725 | r = strv_extend(fdnames, of->fdname); | |
4726 | if (r < 0) | |
4727 | return r; | |
4728 | ||
4729 | (*fds)[*n_fds] = TAKE_FD(fd); | |
4730 | ||
4731 | (*n_fds)++; | |
4732 | } | |
4733 | ||
4734 | return 0; | |
4735 | } | |
4736 | ||
3ff67ec4 ZJS |
4737 | static void log_command_line(Unit *unit, const char *msg, const char *executable, char **argv) { |
4738 | assert(unit); | |
4739 | assert(msg); | |
4740 | assert(executable); | |
4741 | ||
4742 | if (!DEBUG_LOGGING) | |
4743 | return; | |
4744 | ||
4745 | _cleanup_free_ char *cmdline = quote_command_line(argv, SHELL_ESCAPE_EMPTY); | |
4746 | ||
4747 | log_unit_struct(unit, LOG_DEBUG, | |
4748 | "EXECUTABLE=%s", executable, | |
4749 | LOG_UNIT_MESSAGE(unit, "%s: %s", msg, strnull(cmdline)), | |
4750 | LOG_UNIT_INVOCATION_ID(unit)); | |
4751 | } | |
4752 | ||
170d978b LP |
4753 | static bool exec_context_need_unprivileged_private_users( |
4754 | const ExecContext *context, | |
4755 | const ExecParameters *params) { | |
4756 | ||
6ef721cb | 4757 | assert(context); |
170d978b | 4758 | assert(params); |
6ef721cb LB |
4759 | |
4760 | /* These options require PrivateUsers= when used in user units, as we need to be in a user namespace | |
4761 | * to have permission to enable them when not running as root. If we have effective CAP_SYS_ADMIN | |
4762 | * (system manager) then we have privileges and don't need this. */ | |
170d978b | 4763 | if (params->runtime_scope != RUNTIME_SCOPE_USER) |
6ef721cb LB |
4764 | return false; |
4765 | ||
4766 | return context->private_users || | |
4767 | context->private_tmp || | |
4768 | context->private_devices || | |
4769 | context->private_network || | |
4770 | context->network_namespace_path || | |
4771 | context->private_ipc || | |
4772 | context->ipc_namespace_path || | |
adeff822 | 4773 | context->private_mounts > 0 || |
6ef721cb LB |
4774 | context->mount_apivfs || |
4775 | context->n_bind_mounts > 0 || | |
4776 | context->n_temporary_filesystems > 0 || | |
4777 | context->root_directory || | |
4778 | !strv_isempty(context->extension_directories) || | |
4779 | context->protect_system != PROTECT_SYSTEM_NO || | |
4780 | context->protect_home != PROTECT_HOME_NO || | |
4781 | context->protect_kernel_tunables || | |
4782 | context->protect_kernel_modules || | |
4783 | context->protect_kernel_logs || | |
4784 | context->protect_control_groups || | |
4785 | context->protect_clock || | |
4786 | context->protect_hostname || | |
4787 | !strv_isempty(context->read_write_paths) || | |
4788 | !strv_isempty(context->read_only_paths) || | |
4789 | !strv_isempty(context->inaccessible_paths) || | |
4790 | !strv_isempty(context->exec_paths) || | |
4791 | !strv_isempty(context->no_exec_paths); | |
4792 | } | |
4793 | ||
ff0af2a1 | 4794 | static int exec_child( |
f2341e0a | 4795 | Unit *unit, |
34cf6c43 | 4796 | const ExecCommand *command, |
ff0af2a1 LP |
4797 | const ExecContext *context, |
4798 | const ExecParameters *params, | |
28135da3 | 4799 | ExecRuntime *runtime, |
6bb00842 | 4800 | const CGroupContext *cgroup_context, |
ff0af2a1 | 4801 | int socket_fd, |
2caa38e9 | 4802 | const int named_iofds[static 3], |
cd48e23f | 4803 | int *params_fds, |
da6053d0 | 4804 | size_t n_socket_fds, |
25b583d7 | 4805 | size_t n_storage_fds, |
ff0af2a1 | 4806 | char **files_env, |
00d9ef85 | 4807 | int user_lookup_fd, |
12145637 | 4808 | int *exit_status) { |
d35fbf6b | 4809 | |
8c35c10d | 4810 | _cleanup_strv_free_ char **our_env = NULL, **pass_env = NULL, **joined_exec_search_path = NULL, **accum_env = NULL, **replaced_argv = NULL; |
1da37e58 | 4811 | int r, ngids = 0, exec_fd; |
4d885bd3 DH |
4812 | _cleanup_free_ gid_t *supplementary_gids = NULL; |
4813 | const char *username = NULL, *groupname = NULL; | |
6bb00842 | 4814 | _cleanup_free_ char *home_buffer = NULL, *memory_pressure_path = NULL; |
2b3c1b9e | 4815 | const char *home = NULL, *shell = NULL; |
7ca69792 | 4816 | char **final_argv = NULL; |
7bce046b LP |
4817 | dev_t journal_stream_dev = 0; |
4818 | ino_t journal_stream_ino = 0; | |
5749f855 | 4819 | bool userns_set_up = false; |
165a31c0 LP |
4820 | bool needs_sandboxing, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */ |
4821 | needs_setuid, /* Do we need to do the actual setresuid()/setresgid() calls? */ | |
4822 | needs_mount_namespace, /* Do we need to set up a mount namespace for this kernel? */ | |
4823 | needs_ambient_hack; /* Do we need to apply the ambient capabilities hack? */ | |
349cc4a5 | 4824 | #if HAVE_SELINUX |
7f59dd35 | 4825 | _cleanup_free_ char *mac_selinux_context_net = NULL; |
43b1f709 | 4826 | bool use_selinux = false; |
ecfbc84f | 4827 | #endif |
f9fa32f0 | 4828 | #if ENABLE_SMACK |
43b1f709 | 4829 | bool use_smack = false; |
ecfbc84f | 4830 | #endif |
349cc4a5 | 4831 | #if HAVE_APPARMOR |
43b1f709 | 4832 | bool use_apparmor = false; |
ecfbc84f | 4833 | #endif |
5749f855 AZ |
4834 | uid_t saved_uid = getuid(); |
4835 | gid_t saved_gid = getgid(); | |
fed1e721 LP |
4836 | uid_t uid = UID_INVALID; |
4837 | gid_t gid = GID_INVALID; | |
1da37e58 ZJS |
4838 | size_t n_fds = n_socket_fds + n_storage_fds, /* fds to pass to the child */ |
4839 | n_keep_fds; /* total number of fds not to close */ | |
165a31c0 | 4840 | int secure_bits; |
afb11bf1 DG |
4841 | _cleanup_free_ gid_t *gids_after_pam = NULL; |
4842 | int ngids_after_pam = 0; | |
cd48e23f RP |
4843 | _cleanup_free_ int *fds = NULL; |
4844 | _cleanup_strv_free_ char **fdnames = NULL; | |
034c6ed7 | 4845 | |
f2341e0a | 4846 | assert(unit); |
5cb5a6ff LP |
4847 | assert(command); |
4848 | assert(context); | |
d35fbf6b | 4849 | assert(params); |
ff0af2a1 | 4850 | assert(exit_status); |
d35fbf6b | 4851 | |
69339ae9 LP |
4852 | /* Explicitly test for CVE-2021-4034 inspired invocations */ |
4853 | assert(command->path); | |
4854 | assert(!strv_isempty(command->argv)); | |
4855 | ||
d35fbf6b DM |
4856 | rename_process_from_path(command->path); |
4857 | ||
9c274488 LP |
4858 | /* We reset exactly these signals, since they are the only ones we set to SIG_IGN in the main |
4859 | * daemon. All others we leave untouched because we set them to SIG_DFL or a valid handler initially, | |
4860 | * both of which will be demoted to SIG_DFL. */ | |
ce30c8dc | 4861 | (void) default_signals(SIGNALS_CRASH_HANDLER, |
9c274488 | 4862 | SIGNALS_IGNORE); |
d35fbf6b DM |
4863 | |
4864 | if (context->ignore_sigpipe) | |
9c274488 | 4865 | (void) ignore_signals(SIGPIPE); |
d35fbf6b | 4866 | |
ff0af2a1 LP |
4867 | r = reset_signal_mask(); |
4868 | if (r < 0) { | |
4869 | *exit_status = EXIT_SIGNAL_MASK; | |
12145637 | 4870 | return log_unit_error_errno(unit, r, "Failed to set process signal mask: %m"); |
d35fbf6b | 4871 | } |
034c6ed7 | 4872 | |
d35fbf6b DM |
4873 | if (params->idle_pipe) |
4874 | do_idle_pipe_dance(params->idle_pipe); | |
4f2d528d | 4875 | |
2c027c62 LP |
4876 | /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its |
4877 | * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have | |
4878 | * any fds open we don't really want open during the transition. In order to make logging work, we switch the | |
4879 | * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */ | |
ff0af2a1 | 4880 | |
d35fbf6b | 4881 | log_forget_fds(); |
2c027c62 | 4882 | log_set_open_when_needed(true); |
a3b00f91 | 4883 | log_settle_target(); |
4f2d528d | 4884 | |
40a80078 LP |
4885 | /* In case anything used libc syslog(), close this here, too */ |
4886 | closelog(); | |
4887 | ||
cd48e23f RP |
4888 | fds = newdup(int, params_fds, n_fds); |
4889 | if (!fds) { | |
4890 | *exit_status = EXIT_MEMORY; | |
4891 | return log_oom(); | |
4892 | } | |
4893 | ||
4894 | fdnames = strv_copy((char**) params->fd_names); | |
4895 | if (!fdnames) { | |
4896 | *exit_status = EXIT_MEMORY; | |
4897 | return log_oom(); | |
4898 | } | |
4899 | ||
4900 | r = collect_open_file_fds(unit, params->open_files, &fds, &fdnames, &n_fds); | |
4901 | if (r < 0) { | |
4902 | *exit_status = EXIT_FDS; | |
4903 | return log_unit_error_errno(unit, r, "Failed to get OpenFile= file descriptors: %m"); | |
4904 | } | |
4905 | ||
b1994387 | 4906 | int keep_fds[n_fds + 3]; |
1da37e58 ZJS |
4907 | memcpy_safe(keep_fds, fds, n_fds * sizeof(int)); |
4908 | n_keep_fds = n_fds; | |
4909 | ||
4910 | r = add_shifted_fd(keep_fds, ELEMENTSOF(keep_fds), &n_keep_fds, params->exec_fd, &exec_fd); | |
4911 | if (r < 0) { | |
4912 | *exit_status = EXIT_FDS; | |
4913 | return log_unit_error_errno(unit, r, "Failed to shift fd and set FD_CLOEXEC: %m"); | |
4914 | } | |
4915 | ||
b1994387 | 4916 | #if HAVE_LIBBPF |
46004616 ZJS |
4917 | if (unit->manager->restrict_fs) { |
4918 | int bpf_map_fd = lsm_bpf_map_restrict_fs_fd(unit); | |
b1994387 ILG |
4919 | if (bpf_map_fd < 0) { |
4920 | *exit_status = EXIT_FDS; | |
46004616 | 4921 | return log_unit_error_errno(unit, bpf_map_fd, "Failed to get restrict filesystems BPF map fd: %m"); |
b1994387 ILG |
4922 | } |
4923 | ||
4924 | r = add_shifted_fd(keep_fds, ELEMENTSOF(keep_fds), &n_keep_fds, bpf_map_fd, &bpf_map_fd); | |
4925 | if (r < 0) { | |
4926 | *exit_status = EXIT_FDS; | |
4927 | return log_unit_error_errno(unit, r, "Failed to shift fd and set FD_CLOEXEC: %m"); | |
4928 | } | |
4929 | } | |
4930 | #endif | |
4931 | ||
15220772 | 4932 | r = close_remaining_fds(params, runtime, user_lookup_fd, socket_fd, keep_fds, n_keep_fds); |
ff0af2a1 LP |
4933 | if (r < 0) { |
4934 | *exit_status = EXIT_FDS; | |
12145637 | 4935 | return log_unit_error_errno(unit, r, "Failed to close unwanted file descriptors: %m"); |
8c7be95e LP |
4936 | } |
4937 | ||
0af07108 ZJS |
4938 | if (!context->same_pgrp && |
4939 | setsid() < 0) { | |
4940 | *exit_status = EXIT_SETSID; | |
4941 | return log_unit_error_errno(unit, errno, "Failed to create new process session: %m"); | |
4942 | } | |
9e2f7c11 | 4943 | |
1e22b5cd | 4944 | exec_context_tty_reset(context, params); |
d35fbf6b | 4945 | |
c891efaf | 4946 | if (unit_shall_confirm_spawn(unit)) { |
3b20f877 FB |
4947 | _cleanup_free_ char *cmdline = NULL; |
4948 | ||
4ef15008 | 4949 | cmdline = quote_command_line(command->argv, SHELL_ESCAPE_EMPTY); |
3b20f877 | 4950 | if (!cmdline) { |
0460aa5c | 4951 | *exit_status = EXIT_MEMORY; |
12145637 | 4952 | return log_oom(); |
3b20f877 | 4953 | } |
d35fbf6b | 4954 | |
4ef15008 | 4955 | r = ask_for_confirmation(context, params->confirm_spawn, unit, cmdline); |
3b20f877 FB |
4956 | if (r != CONFIRM_EXECUTE) { |
4957 | if (r == CONFIRM_PRETEND_SUCCESS) { | |
4958 | *exit_status = EXIT_SUCCESS; | |
4959 | return 0; | |
4960 | } | |
ff0af2a1 | 4961 | *exit_status = EXIT_CONFIRM; |
0af07108 ZJS |
4962 | return log_unit_error_errno(unit, SYNTHETIC_ERRNO(ECANCELED), |
4963 | "Execution cancelled by the user"); | |
d35fbf6b DM |
4964 | } |
4965 | } | |
1a63a750 | 4966 | |
d521916d LP |
4967 | /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is |
4968 | * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note | |
4969 | * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS | |
4970 | * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they | |
4971 | * might internally call into other NSS modules that are involved in hostname resolution, we never know. */ | |
4972 | if (setenv("SYSTEMD_ACTIVATION_UNIT", unit->id, true) != 0 || | |
170d978b | 4973 | setenv("SYSTEMD_ACTIVATION_SCOPE", runtime_scope_to_string(params->runtime_scope), true) != 0) { |
d521916d LP |
4974 | *exit_status = EXIT_MEMORY; |
4975 | return log_unit_error_errno(unit, errno, "Failed to update environment: %m"); | |
4976 | } | |
4977 | ||
15220772 | 4978 | if (context->dynamic_user && runtime && runtime->dynamic_creds) { |
da50b85a | 4979 | _cleanup_strv_free_ char **suggested_paths = NULL; |
29206d46 | 4980 | |
d521916d | 4981 | /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS |
7802194a | 4982 | * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here. */ |
409093fe LP |
4983 | if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) { |
4984 | *exit_status = EXIT_USER; | |
12145637 | 4985 | return log_unit_error_errno(unit, errno, "Failed to update environment: %m"); |
409093fe LP |
4986 | } |
4987 | ||
da50b85a LP |
4988 | r = compile_suggested_paths(context, params, &suggested_paths); |
4989 | if (r < 0) { | |
4990 | *exit_status = EXIT_MEMORY; | |
4991 | return log_oom(); | |
4992 | } | |
4993 | ||
15220772 | 4994 | r = dynamic_creds_realize(runtime->dynamic_creds, suggested_paths, &uid, &gid); |
ff0af2a1 LP |
4995 | if (r < 0) { |
4996 | *exit_status = EXIT_USER; | |
d85ff944 YW |
4997 | if (r == -EILSEQ) |
4998 | return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EOPNOTSUPP), | |
4999 | "Failed to update dynamic user credentials: User or group with specified name already exists."); | |
12145637 | 5000 | return log_unit_error_errno(unit, r, "Failed to update dynamic user credentials: %m"); |
524daa8c | 5001 | } |
524daa8c | 5002 | |
70dd455c | 5003 | if (!uid_is_valid(uid)) { |
29206d46 | 5004 | *exit_status = EXIT_USER; |
d85ff944 | 5005 | return log_unit_error_errno(unit, SYNTHETIC_ERRNO(ESRCH), "UID validation failed for \""UID_FMT"\"", uid); |
70dd455c ZJS |
5006 | } |
5007 | ||
5008 | if (!gid_is_valid(gid)) { | |
5009 | *exit_status = EXIT_USER; | |
d85ff944 | 5010 | return log_unit_error_errno(unit, SYNTHETIC_ERRNO(ESRCH), "GID validation failed for \""GID_FMT"\"", gid); |
29206d46 | 5011 | } |
5bc7452b | 5012 | |
15220772 DDM |
5013 | if (runtime->dynamic_creds->user) |
5014 | username = runtime->dynamic_creds->user->name; | |
29206d46 LP |
5015 | |
5016 | } else { | |
4d885bd3 DH |
5017 | r = get_fixed_user(context, &username, &uid, &gid, &home, &shell); |
5018 | if (r < 0) { | |
5019 | *exit_status = EXIT_USER; | |
12145637 | 5020 | return log_unit_error_errno(unit, r, "Failed to determine user credentials: %m"); |
5bc7452b | 5021 | } |
5bc7452b | 5022 | |
4d885bd3 DH |
5023 | r = get_fixed_group(context, &groupname, &gid); |
5024 | if (r < 0) { | |
5025 | *exit_status = EXIT_GROUP; | |
12145637 | 5026 | return log_unit_error_errno(unit, r, "Failed to determine group credentials: %m"); |
4d885bd3 | 5027 | } |
cdc5d5c5 | 5028 | } |
29206d46 | 5029 | |
cdc5d5c5 DH |
5030 | /* Initialize user supplementary groups and get SupplementaryGroups= ones */ |
5031 | r = get_supplementary_groups(context, username, groupname, gid, | |
5032 | &supplementary_gids, &ngids); | |
5033 | if (r < 0) { | |
5034 | *exit_status = EXIT_GROUP; | |
12145637 | 5035 | return log_unit_error_errno(unit, r, "Failed to determine supplementary groups: %m"); |
29206d46 | 5036 | } |
5bc7452b | 5037 | |
00d9ef85 LP |
5038 | r = send_user_lookup(unit, user_lookup_fd, uid, gid); |
5039 | if (r < 0) { | |
5040 | *exit_status = EXIT_USER; | |
12145637 | 5041 | return log_unit_error_errno(unit, r, "Failed to send user credentials to PID1: %m"); |
00d9ef85 LP |
5042 | } |
5043 | ||
5044 | user_lookup_fd = safe_close(user_lookup_fd); | |
5045 | ||
6732edab LP |
5046 | r = acquire_home(context, uid, &home, &home_buffer); |
5047 | if (r < 0) { | |
5048 | *exit_status = EXIT_CHDIR; | |
12145637 | 5049 | return log_unit_error_errno(unit, r, "Failed to determine $HOME for user: %m"); |
6732edab LP |
5050 | } |
5051 | ||
4a055e5a | 5052 | /* If a socket is connected to STDIN/STDOUT/STDERR, we must drop O_NONBLOCK */ |
d35fbf6b | 5053 | if (socket_fd >= 0) |
a34ceba6 | 5054 | (void) fd_nonblock(socket_fd, false); |
acbb0225 | 5055 | |
4c70a4a7 MS |
5056 | /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields. |
5057 | * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */ | |
5058 | if (params->cgroup_path) { | |
5059 | _cleanup_free_ char *p = NULL; | |
5060 | ||
a8b993dc | 5061 | r = exec_parameters_get_cgroup_path(params, cgroup_context, &p); |
4c70a4a7 MS |
5062 | if (r < 0) { |
5063 | *exit_status = EXIT_CGROUP; | |
5064 | return log_unit_error_errno(unit, r, "Failed to acquire cgroup path: %m"); | |
5065 | } | |
5066 | ||
5067 | r = cg_attach_everywhere(params->cgroup_supported, p, 0, NULL, NULL); | |
702cf08f YW |
5068 | if (r == -EUCLEAN) { |
5069 | *exit_status = EXIT_CGROUP; | |
5070 | return log_unit_error_errno(unit, r, "Failed to attach process to cgroup %s " | |
5071 | "because the cgroup or one of its parents or " | |
5072 | "siblings is in the threaded mode: %m", p); | |
5073 | } | |
4c70a4a7 MS |
5074 | if (r < 0) { |
5075 | *exit_status = EXIT_CGROUP; | |
5076 | return log_unit_error_errno(unit, r, "Failed to attach to cgroup %s: %m", p); | |
5077 | } | |
5078 | } | |
5079 | ||
28135da3 DDM |
5080 | if (context->network_namespace_path && runtime && runtime->shared && runtime->shared->netns_storage_socket[0] >= 0) { |
5081 | r = open_shareable_ns_path(runtime->shared->netns_storage_socket, context->network_namespace_path, CLONE_NEWNET); | |
a8d08f39 LP |
5082 | if (r < 0) { |
5083 | *exit_status = EXIT_NETWORK; | |
5084 | return log_unit_error_errno(unit, r, "Failed to open network namespace path %s: %m", context->network_namespace_path); | |
5085 | } | |
5086 | } | |
5087 | ||
28135da3 DDM |
5088 | if (context->ipc_namespace_path && runtime && runtime->shared && runtime->shared->ipcns_storage_socket[0] >= 0) { |
5089 | r = open_shareable_ns_path(runtime->shared->ipcns_storage_socket, context->ipc_namespace_path, CLONE_NEWIPC); | |
a70581ff XR |
5090 | if (r < 0) { |
5091 | *exit_status = EXIT_NAMESPACE; | |
5092 | return log_unit_error_errno(unit, r, "Failed to open IPC namespace path %s: %m", context->ipc_namespace_path); | |
5093 | } | |
5094 | } | |
5095 | ||
52c239d7 | 5096 | r = setup_input(context, params, socket_fd, named_iofds); |
ff0af2a1 LP |
5097 | if (r < 0) { |
5098 | *exit_status = EXIT_STDIN; | |
12145637 | 5099 | return log_unit_error_errno(unit, r, "Failed to set up standard input: %m"); |
d35fbf6b | 5100 | } |
034c6ed7 | 5101 | |
52c239d7 | 5102 | 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 |
5103 | if (r < 0) { |
5104 | *exit_status = EXIT_STDOUT; | |
12145637 | 5105 | return log_unit_error_errno(unit, r, "Failed to set up standard output: %m"); |
d35fbf6b DM |
5106 | } |
5107 | ||
52c239d7 | 5108 | 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 |
5109 | if (r < 0) { |
5110 | *exit_status = EXIT_STDERR; | |
12145637 | 5111 | return log_unit_error_errno(unit, r, "Failed to set up standard error output: %m"); |
d35fbf6b DM |
5112 | } |
5113 | ||
d35fbf6b | 5114 | if (context->oom_score_adjust_set) { |
bb44fd07 ZJS |
5115 | /* When we can't make this change due to EPERM, then let's silently skip over it. User |
5116 | * namespaces prohibit write access to this file, and we shouldn't trip up over that. */ | |
9f8168eb | 5117 | r = set_oom_score_adjust(context->oom_score_adjust); |
bb44fd07 ZJS |
5118 | if (ERRNO_IS_NEG_PRIVILEGE(r)) |
5119 | log_unit_debug_errno(unit, r, | |
5120 | "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m"); | |
5121 | else if (r < 0) { | |
5122 | *exit_status = EXIT_OOM_ADJUST; | |
5123 | return log_unit_error_errno(unit, r, "Failed to adjust OOM setting: %m"); | |
613b411c | 5124 | } |
d35fbf6b DM |
5125 | } |
5126 | ||
ad21e542 ZJS |
5127 | if (context->coredump_filter_set) { |
5128 | r = set_coredump_filter(context->coredump_filter); | |
bb44fd07 ZJS |
5129 | if (ERRNO_IS_NEG_PRIVILEGE(r)) |
5130 | log_unit_debug_errno(unit, r, "Failed to adjust coredump_filter, ignoring: %m"); | |
5131 | else if (r < 0) | |
5132 | return log_unit_error_errno(unit, r, "Failed to adjust coredump_filter: %m"); | |
ad21e542 ZJS |
5133 | } |
5134 | ||
39090201 DJL |
5135 | if (context->nice_set) { |
5136 | r = setpriority_closest(context->nice); | |
5137 | if (r < 0) | |
5138 | return log_unit_error_errno(unit, r, "Failed to set up process scheduling priority (nice level): %m"); | |
5139 | } | |
613b411c | 5140 | |
d35fbf6b DM |
5141 | if (context->cpu_sched_set) { |
5142 | struct sched_param param = { | |
5143 | .sched_priority = context->cpu_sched_priority, | |
5144 | }; | |
5145 | ||
ff0af2a1 LP |
5146 | r = sched_setscheduler(0, |
5147 | context->cpu_sched_policy | | |
5148 | (context->cpu_sched_reset_on_fork ? | |
5149 | SCHED_RESET_ON_FORK : 0), | |
5150 | ¶m); | |
5151 | if (r < 0) { | |
5152 | *exit_status = EXIT_SETSCHEDULER; | |
12145637 | 5153 | return log_unit_error_errno(unit, errno, "Failed to set up CPU scheduling: %m"); |
fc9b2a84 | 5154 | } |
d35fbf6b | 5155 | } |
fc9b2a84 | 5156 | |
e2b2fb7f MS |
5157 | if (context->cpu_affinity_from_numa || context->cpu_set.set) { |
5158 | _cleanup_(cpu_set_reset) CPUSet converted_cpu_set = {}; | |
5159 | const CPUSet *cpu_set; | |
5160 | ||
5161 | if (context->cpu_affinity_from_numa) { | |
5162 | r = exec_context_cpu_affinity_from_numa(context, &converted_cpu_set); | |
5163 | if (r < 0) { | |
5164 | *exit_status = EXIT_CPUAFFINITY; | |
5165 | return log_unit_error_errno(unit, r, "Failed to derive CPU affinity mask from NUMA mask: %m"); | |
5166 | } | |
5167 | ||
5168 | cpu_set = &converted_cpu_set; | |
5169 | } else | |
5170 | cpu_set = &context->cpu_set; | |
5171 | ||
5172 | if (sched_setaffinity(0, cpu_set->allocated, cpu_set->set) < 0) { | |
ff0af2a1 | 5173 | *exit_status = EXIT_CPUAFFINITY; |
12145637 | 5174 | return log_unit_error_errno(unit, errno, "Failed to set up CPU affinity: %m"); |
034c6ed7 | 5175 | } |
e2b2fb7f | 5176 | } |
034c6ed7 | 5177 | |
b070c7c0 MS |
5178 | if (mpol_is_valid(numa_policy_get_type(&context->numa_policy))) { |
5179 | r = apply_numa_policy(&context->numa_policy); | |
bb44fd07 ZJS |
5180 | if (ERRNO_IS_NEG_NOT_SUPPORTED(r)) |
5181 | log_unit_debug_errno(unit, r, "NUMA support not available, ignoring."); | |
5182 | else if (r < 0) { | |
5183 | *exit_status = EXIT_NUMA_POLICY; | |
5184 | return log_unit_error_errno(unit, r, "Failed to set NUMA memory policy: %m"); | |
b070c7c0 MS |
5185 | } |
5186 | } | |
5187 | ||
d35fbf6b DM |
5188 | if (context->ioprio_set) |
5189 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) { | |
ff0af2a1 | 5190 | *exit_status = EXIT_IOPRIO; |
12145637 | 5191 | return log_unit_error_errno(unit, errno, "Failed to set up IO scheduling priority: %m"); |
d35fbf6b | 5192 | } |
da726a4d | 5193 | |
d35fbf6b DM |
5194 | if (context->timer_slack_nsec != NSEC_INFINITY) |
5195 | if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) { | |
ff0af2a1 | 5196 | *exit_status = EXIT_TIMERSLACK; |
12145637 | 5197 | return log_unit_error_errno(unit, errno, "Failed to set up timer slack: %m"); |
4c2630eb | 5198 | } |
9eba9da4 | 5199 | |
21022b9d LP |
5200 | if (context->personality != PERSONALITY_INVALID) { |
5201 | r = safe_personality(context->personality); | |
5202 | if (r < 0) { | |
ff0af2a1 | 5203 | *exit_status = EXIT_PERSONALITY; |
12145637 | 5204 | return log_unit_error_errno(unit, r, "Failed to set up execution domain (personality): %m"); |
4c2630eb | 5205 | } |
21022b9d | 5206 | } |
94f04347 | 5207 | |
33331d11 VB |
5208 | if (context->utmp_id) { |
5209 | const char *line = context->tty_path ? | |
5210 | (path_startswith(context->tty_path, "/dev/") ?: context->tty_path) : | |
5211 | NULL; | |
df0ff127 | 5212 | utmp_put_init_process(context->utmp_id, getpid_cached(), getsid(0), |
33331d11 | 5213 | line, |
023a4f67 LP |
5214 | context->utmp_mode == EXEC_UTMP_INIT ? INIT_PROCESS : |
5215 | context->utmp_mode == EXEC_UTMP_LOGIN ? LOGIN_PROCESS : | |
5216 | USER_PROCESS, | |
6a93917d | 5217 | username); |
33331d11 | 5218 | } |
d35fbf6b | 5219 | |
08f67696 | 5220 | if (uid_is_valid(uid)) { |
ff0af2a1 LP |
5221 | r = chown_terminal(STDIN_FILENO, uid); |
5222 | if (r < 0) { | |
5223 | *exit_status = EXIT_STDIN; | |
12145637 | 5224 | return log_unit_error_errno(unit, r, "Failed to change ownership of terminal: %m"); |
071830ff | 5225 | } |
d35fbf6b | 5226 | } |
8e274523 | 5227 | |
6bb00842 LP |
5228 | if (params->cgroup_path) { |
5229 | /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1 | |
5230 | * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not | |
5231 | * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only | |
5232 | * touch a single hierarchy too. */ | |
5233 | ||
5234 | if (params->flags & EXEC_CGROUP_DELEGATE) { | |
a8b993dc LP |
5235 | _cleanup_free_ char *p = NULL; |
5236 | ||
6bb00842 LP |
5237 | r = cg_set_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, uid, gid); |
5238 | if (r < 0) { | |
5239 | *exit_status = EXIT_CGROUP; | |
5240 | return log_unit_error_errno(unit, r, "Failed to adjust control group access: %m"); | |
5241 | } | |
a8b993dc LP |
5242 | |
5243 | r = exec_parameters_get_cgroup_path(params, cgroup_context, &p); | |
5244 | if (r < 0) { | |
5245 | *exit_status = EXIT_CGROUP; | |
5246 | return log_unit_error_errno(unit, r, "Failed to acquire cgroup path: %m"); | |
5247 | } | |
5248 | if (r > 0) { | |
bcd9b981 | 5249 | r = cg_set_access_recursive(SYSTEMD_CGROUP_CONTROLLER, p, uid, gid); |
a8b993dc LP |
5250 | if (r < 0) { |
5251 | *exit_status = EXIT_CGROUP; | |
5252 | return log_unit_error_errno(unit, r, "Failed to adjust control subgroup access: %m"); | |
5253 | } | |
5254 | } | |
6bb00842 LP |
5255 | } |
5256 | ||
5257 | if (cgroup_context && cg_unified() > 0 && is_pressure_supported() > 0) { | |
5258 | if (cgroup_context_want_memory_pressure(cgroup_context)) { | |
5259 | r = cg_get_path("memory", params->cgroup_path, "memory.pressure", &memory_pressure_path); | |
5260 | if (r < 0) { | |
5261 | *exit_status = EXIT_MEMORY; | |
5262 | return log_oom(); | |
5263 | } | |
5264 | ||
5265 | r = chmod_and_chown(memory_pressure_path, 0644, uid, gid); | |
5266 | if (r < 0) { | |
5267 | log_unit_full_errno(unit, r == -ENOENT || ERRNO_IS_PRIVILEGE(r) ? LOG_DEBUG : LOG_WARNING, r, | |
5268 | "Failed to adjust ownership of '%s', ignoring: %m", memory_pressure_path); | |
5269 | memory_pressure_path = mfree(memory_pressure_path); | |
5270 | } | |
5271 | } else if (cgroup_context->memory_pressure_watch == CGROUP_PRESSURE_WATCH_OFF) { | |
5272 | memory_pressure_path = strdup("/dev/null"); /* /dev/null is explicit indicator for turning of memory pressure watch */ | |
5273 | if (!memory_pressure_path) { | |
5274 | *exit_status = EXIT_MEMORY; | |
5275 | return log_oom(); | |
5276 | } | |
5277 | } | |
034c6ed7 | 5278 | } |
d35fbf6b | 5279 | } |
034c6ed7 | 5280 | |
211a3d87 LB |
5281 | needs_mount_namespace = exec_needs_mount_namespace(context, params, runtime); |
5282 | ||
5b10116e | 5283 | for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) { |
59dd2bbb | 5284 | r = setup_exec_directory(unit, context, params, uid, gid, dt, needs_mount_namespace, exit_status); |
12145637 LP |
5285 | if (r < 0) |
5286 | return log_unit_error_errno(unit, r, "Failed to set up special execution directory in %s: %m", params->prefix[dt]); | |
d35fbf6b | 5287 | } |
94f04347 | 5288 | |
bb0c0d6f | 5289 | if (FLAGS_SET(params->flags, EXEC_WRITE_CREDENTIALS)) { |
3a78b0e9 | 5290 | r = setup_credentials(context, params, unit->id, uid, gid); |
bb0c0d6f LP |
5291 | if (r < 0) { |
5292 | *exit_status = EXIT_CREDENTIALS; | |
5293 | return log_unit_error_errno(unit, r, "Failed to set up credentials: %m"); | |
5294 | } | |
5295 | } | |
5296 | ||
7bce046b | 5297 | r = build_environment( |
fd63e712 | 5298 | unit, |
7bce046b LP |
5299 | context, |
5300 | params, | |
6bb00842 | 5301 | cgroup_context, |
7bce046b | 5302 | n_fds, |
cd48e23f | 5303 | fdnames, |
7bce046b LP |
5304 | home, |
5305 | username, | |
5306 | shell, | |
5307 | journal_stream_dev, | |
5308 | journal_stream_ino, | |
6bb00842 | 5309 | memory_pressure_path, |
7bce046b | 5310 | &our_env); |
2065ca69 JW |
5311 | if (r < 0) { |
5312 | *exit_status = EXIT_MEMORY; | |
12145637 | 5313 | return log_oom(); |
2065ca69 JW |
5314 | } |
5315 | ||
5316 | r = build_pass_environment(context, &pass_env); | |
5317 | if (r < 0) { | |
5318 | *exit_status = EXIT_MEMORY; | |
12145637 | 5319 | return log_oom(); |
2065ca69 JW |
5320 | } |
5321 | ||
adf769b0 ZJS |
5322 | /* The $PATH variable is set to the default path in params->environment. However, this is overridden |
5323 | * if user-specified fields have $PATH set. The intention is to also override $PATH if the unit does | |
5324 | * not specify PATH but the unit has ExecSearchPath. */ | |
8c35c10d | 5325 | if (!strv_isempty(context->exec_search_path)) { |
5326 | _cleanup_free_ char *joined = NULL; | |
5327 | ||
5328 | joined = strv_join(context->exec_search_path, ":"); | |
5329 | if (!joined) { | |
5330 | *exit_status = EXIT_MEMORY; | |
5331 | return log_oom(); | |
5332 | } | |
5333 | ||
5334 | r = strv_env_assign(&joined_exec_search_path, "PATH", joined); | |
5335 | if (r < 0) { | |
5336 | *exit_status = EXIT_MEMORY; | |
5337 | return log_oom(); | |
5338 | } | |
5339 | } | |
5340 | ||
4ab3d29f | 5341 | accum_env = strv_env_merge(params->environment, |
2065ca69 | 5342 | our_env, |
8c35c10d | 5343 | joined_exec_search_path, |
2065ca69 JW |
5344 | pass_env, |
5345 | context->environment, | |
44e5d006 | 5346 | files_env); |
2065ca69 JW |
5347 | if (!accum_env) { |
5348 | *exit_status = EXIT_MEMORY; | |
12145637 | 5349 | return log_oom(); |
2065ca69 | 5350 | } |
1280503b | 5351 | accum_env = strv_env_clean(accum_env); |
2065ca69 | 5352 | |
096424d1 | 5353 | (void) umask(context->umask); |
b213e1c1 | 5354 | |
b1edf445 | 5355 | r = setup_keyring(unit, context, params, uid, gid); |
74dd6b51 LP |
5356 | if (r < 0) { |
5357 | *exit_status = EXIT_KEYRING; | |
12145637 | 5358 | return log_unit_error_errno(unit, r, "Failed to set up kernel keyring: %m"); |
74dd6b51 LP |
5359 | } |
5360 | ||
adf769b0 ZJS |
5361 | /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted |
5362 | * from it. */ | |
1703fa41 | 5363 | needs_sandboxing = (params->flags & EXEC_APPLY_SANDBOXING) && !(command->flags & EXEC_COMMAND_FULLY_PRIVILEGED); |
7f18ef0a | 5364 | |
adf769b0 ZJS |
5365 | /* We need the ambient capability hack, if the caller asked us to apply it and the command is marked |
5366 | * for it, and the kernel doesn't actually support ambient caps. */ | |
165a31c0 | 5367 | needs_ambient_hack = (params->flags & EXEC_APPLY_SANDBOXING) && (command->flags & EXEC_COMMAND_AMBIENT_MAGIC) && !ambient_capabilities_supported(); |
7f18ef0a | 5368 | |
adf769b0 ZJS |
5369 | /* We need setresuid() if the caller asked us to apply sandboxing and the command isn't explicitly |
5370 | * excepted from either whole sandboxing or just setresuid() itself, and the ambient hack is not | |
5371 | * desired. */ | |
165a31c0 LP |
5372 | if (needs_ambient_hack) |
5373 | needs_setuid = false; | |
5374 | else | |
5375 | needs_setuid = (params->flags & EXEC_APPLY_SANDBOXING) && !(command->flags & (EXEC_COMMAND_FULLY_PRIVILEGED|EXEC_COMMAND_NO_SETUID)); | |
5376 | ||
638fd8cc LP |
5377 | uint64_t capability_ambient_set = context->capability_ambient_set; |
5378 | ||
165a31c0 | 5379 | if (needs_sandboxing) { |
adf769b0 ZJS |
5380 | /* MAC enablement checks need to be done before a new mount ns is created, as they rely on |
5381 | * /sys being present. The actual MAC context application will happen later, as late as | |
5382 | * possible, to avoid impacting our own code paths. */ | |
7f18ef0a | 5383 | |
349cc4a5 | 5384 | #if HAVE_SELINUX |
43b1f709 | 5385 | use_selinux = mac_selinux_use(); |
7f18ef0a | 5386 | #endif |
f9fa32f0 | 5387 | #if ENABLE_SMACK |
43b1f709 | 5388 | use_smack = mac_smack_use(); |
7f18ef0a | 5389 | #endif |
349cc4a5 | 5390 | #if HAVE_APPARMOR |
43b1f709 | 5391 | use_apparmor = mac_apparmor_use(); |
7f18ef0a | 5392 | #endif |
165a31c0 | 5393 | } |
7f18ef0a | 5394 | |
ce932d2d LP |
5395 | if (needs_sandboxing) { |
5396 | int which_failed; | |
5397 | ||
5398 | /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what | |
5399 | * is set here. (See below.) */ | |
5400 | ||
5401 | r = setrlimit_closest_all((const struct rlimit* const *) context->rlimit, &which_failed); | |
5402 | if (r < 0) { | |
5403 | *exit_status = EXIT_LIMITS; | |
5404 | return log_unit_error_errno(unit, r, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed)); | |
5405 | } | |
5406 | } | |
5407 | ||
0af07108 | 5408 | if (needs_setuid && context->pam_name && username) { |
ce932d2d LP |
5409 | /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits |
5410 | * wins here. (See above.) */ | |
5411 | ||
1da37e58 | 5412 | /* All fds passed in the fds array will be closed in the pam child process. */ |
0af07108 ZJS |
5413 | r = setup_pam(context->pam_name, username, uid, gid, context->tty_path, &accum_env, fds, n_fds); |
5414 | if (r < 0) { | |
5415 | *exit_status = EXIT_PAM; | |
5416 | return log_unit_error_errno(unit, r, "Failed to set up PAM session: %m"); | |
165a31c0 | 5417 | } |
ac45f971 | 5418 | |
638fd8cc LP |
5419 | if (ambient_capabilities_supported()) { |
5420 | uint64_t ambient_after_pam; | |
5421 | ||
5422 | /* PAM modules might have set some ambient caps. Query them here and merge them into | |
5423 | * the caps we want to set in the end, so that we don't end up unsetting them. */ | |
5424 | r = capability_get_ambient(&ambient_after_pam); | |
5425 | if (r < 0) { | |
5426 | *exit_status = EXIT_CAPABILITIES; | |
5427 | return log_unit_error_errno(unit, r, "Failed to query ambient caps: %m"); | |
5428 | } | |
5429 | ||
5430 | capability_ambient_set |= ambient_after_pam; | |
5431 | } | |
5432 | ||
0af07108 ZJS |
5433 | ngids_after_pam = getgroups_alloc(&gids_after_pam); |
5434 | if (ngids_after_pam < 0) { | |
5435 | *exit_status = EXIT_MEMORY; | |
5436 | return log_unit_error_errno(unit, ngids_after_pam, "Failed to obtain groups after setting up PAM: %m"); | |
5749f855 | 5437 | } |
b213e1c1 | 5438 | } |
5749f855 | 5439 | |
170d978b | 5440 | if (needs_sandboxing && exec_context_need_unprivileged_private_users(context, params)) { |
5749f855 AZ |
5441 | /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces. |
5442 | * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to | |
5443 | * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */ | |
0af07108 | 5444 | |
0af07108 | 5445 | r = setup_private_users(saved_uid, saved_gid, uid, gid); |
6ef721cb LB |
5446 | /* If it was requested explicitly and we can't set it up, fail early. Otherwise, continue and let |
5447 | * the actual requested operations fail (or silently continue). */ | |
5448 | if (r < 0 && context->private_users) { | |
0af07108 ZJS |
5449 | *exit_status = EXIT_USER; |
5450 | return log_unit_error_errno(unit, r, "Failed to set up user namespacing for unprivileged user: %m"); | |
5749f855 | 5451 | } |
6ef721cb LB |
5452 | if (r < 0) |
5453 | log_unit_info_errno(unit, r, "Failed to set up user namespacing for unprivileged user, ignoring: %m"); | |
5454 | else | |
5455 | userns_set_up = true; | |
5749f855 AZ |
5456 | } |
5457 | ||
28135da3 | 5458 | if (exec_needs_network_namespace(context) && runtime && runtime->shared && runtime->shared->netns_storage_socket[0] >= 0) { |
a8d08f39 | 5459 | |
5a3627e5 LP |
5460 | /* Try to enable network namespacing if network namespacing is available and we have |
5461 | * CAP_NET_ADMIN. We need CAP_NET_ADMIN to be able to configure the loopback device in the | |
5462 | * new network namespace. And if we don't have that, then we could only create a network | |
5463 | * namespace without the ability to set up "lo". Hence gracefully skip things then. */ | |
5464 | if (ns_type_supported(NAMESPACE_NET) && have_effective_cap(CAP_NET_ADMIN) > 0) { | |
28135da3 | 5465 | r = setup_shareable_ns(runtime->shared->netns_storage_socket, CLONE_NEWNET); |
bb44fd07 ZJS |
5466 | if (ERRNO_IS_NEG_PRIVILEGE(r)) |
5467 | log_unit_notice_errno(unit, r, | |
5468 | "PrivateNetwork=yes is configured, but network namespace setup not permitted, proceeding without: %m"); | |
5469 | else if (r < 0) { | |
5470 | *exit_status = EXIT_NETWORK; | |
5471 | return log_unit_error_errno(unit, r, "Failed to set up network namespacing: %m"); | |
6e2d7c4f | 5472 | } |
a8d08f39 LP |
5473 | } else if (context->network_namespace_path) { |
5474 | *exit_status = EXIT_NETWORK; | |
ee00d1e9 ZJS |
5475 | return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EOPNOTSUPP), |
5476 | "NetworkNamespacePath= is not supported, refusing."); | |
6e2d7c4f | 5477 | } else |
5a3627e5 | 5478 | log_unit_notice(unit, "PrivateNetwork=yes is configured, but the kernel does not support or we lack privileges for network namespace, proceeding without."); |
d35fbf6b | 5479 | } |
169c1bda | 5480 | |
28135da3 | 5481 | if (exec_needs_ipc_namespace(context) && runtime && runtime->shared && runtime->shared->ipcns_storage_socket[0] >= 0) { |
a70581ff XR |
5482 | |
5483 | if (ns_type_supported(NAMESPACE_IPC)) { | |
28135da3 | 5484 | r = setup_shareable_ns(runtime->shared->ipcns_storage_socket, CLONE_NEWIPC); |
a70581ff XR |
5485 | if (r == -EPERM) |
5486 | log_unit_warning_errno(unit, r, | |
5487 | "PrivateIPC=yes is configured, but IPC namespace setup failed, ignoring: %m"); | |
5488 | else if (r < 0) { | |
5489 | *exit_status = EXIT_NAMESPACE; | |
5490 | return log_unit_error_errno(unit, r, "Failed to set up IPC namespacing: %m"); | |
5491 | } | |
5492 | } else if (context->ipc_namespace_path) { | |
5493 | *exit_status = EXIT_NAMESPACE; | |
5494 | return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EOPNOTSUPP), | |
5495 | "IPCNamespacePath= is not supported, refusing."); | |
5496 | } else | |
5497 | log_unit_warning(unit, "PrivateIPC=yes is configured, but the kernel does not support IPC namespaces, ignoring."); | |
5498 | } | |
5499 | ||
ee818b89 | 5500 | if (needs_mount_namespace) { |
7cc5ef5f ZJS |
5501 | _cleanup_free_ char *error_path = NULL; |
5502 | ||
d4b6ec98 | 5503 | r = apply_mount_namespace(unit, command->flags, context, params, runtime, memory_pressure_path, &error_path); |
3fbe8dbe LP |
5504 | if (r < 0) { |
5505 | *exit_status = EXIT_NAMESPACE; | |
7cc5ef5f ZJS |
5506 | return log_unit_error_errno(unit, r, "Failed to set up mount namespacing%s%s: %m", |
5507 | error_path ? ": " : "", strempty(error_path)); | |
3fbe8dbe | 5508 | } |
d35fbf6b | 5509 | } |
81a2b7ce | 5510 | |
daf8f72b LP |
5511 | if (needs_sandboxing) { |
5512 | r = apply_protect_hostname(unit, context, exit_status); | |
5513 | if (r < 0) | |
5514 | return r; | |
aecd5ac6 TM |
5515 | } |
5516 | ||
85614c6e SR |
5517 | if (context->memory_ksm >= 0) |
5518 | if (prctl(PR_SET_MEMORY_MERGE, context->memory_ksm) < 0) { | |
5519 | if (ERRNO_IS_NOT_SUPPORTED(errno)) | |
5520 | log_unit_debug_errno(unit, errno, "KSM support not available, ignoring."); | |
5521 | else { | |
5522 | *exit_status = EXIT_KSM; | |
5523 | return log_unit_error_errno(unit, errno, "Failed to set KSM: %m"); | |
5524 | } | |
5525 | } | |
5526 | ||
5749f855 AZ |
5527 | /* Drop groups as early as possible. |
5528 | * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root. | |
5529 | * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */ | |
165a31c0 | 5530 | if (needs_setuid) { |
afb11bf1 DG |
5531 | _cleanup_free_ gid_t *gids_to_enforce = NULL; |
5532 | int ngids_to_enforce = 0; | |
5533 | ||
5534 | ngids_to_enforce = merge_gid_lists(supplementary_gids, | |
5535 | ngids, | |
5536 | gids_after_pam, | |
5537 | ngids_after_pam, | |
5538 | &gids_to_enforce); | |
5539 | if (ngids_to_enforce < 0) { | |
5540 | *exit_status = EXIT_MEMORY; | |
5541 | return log_unit_error_errno(unit, | |
5542 | ngids_to_enforce, | |
5543 | "Failed to merge group lists. Group membership might be incorrect: %m"); | |
5544 | } | |
5545 | ||
5546 | r = enforce_groups(gid, gids_to_enforce, ngids_to_enforce); | |
096424d1 LP |
5547 | if (r < 0) { |
5548 | *exit_status = EXIT_GROUP; | |
12145637 | 5549 | return log_unit_error_errno(unit, r, "Changing group credentials failed: %m"); |
096424d1 | 5550 | } |
165a31c0 | 5551 | } |
096424d1 | 5552 | |
5749f855 AZ |
5553 | /* If the user namespace was not set up above, try to do it now. |
5554 | * It's preferred to set up the user namespace later (after all other namespaces) so as not to be | |
d09df6b9 | 5555 | * restricted by rules pertaining to combining user namespaces with other namespaces (e.g. in the |
5749f855 AZ |
5556 | * case of mount namespaces being less privileged when the mount point list is copied from a |
5557 | * different user namespace). */ | |
9008e1ac | 5558 | |
5749f855 AZ |
5559 | if (needs_sandboxing && context->private_users && !userns_set_up) { |
5560 | r = setup_private_users(saved_uid, saved_gid, uid, gid); | |
5561 | if (r < 0) { | |
5562 | *exit_status = EXIT_USER; | |
5563 | return log_unit_error_errno(unit, r, "Failed to set up user namespacing: %m"); | |
d251207d LP |
5564 | } |
5565 | } | |
5566 | ||
9f71ba8d ZJS |
5567 | /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we |
5568 | * shall execute. */ | |
5569 | ||
5570 | _cleanup_free_ char *executable = NULL; | |
254d1313 | 5571 | _cleanup_close_ int executable_fd = -EBADF; |
8c35c10d | 5572 | r = find_executable_full(command->path, /* root= */ NULL, context->exec_search_path, false, &executable, &executable_fd); |
9f71ba8d ZJS |
5573 | if (r < 0) { |
5574 | if (r != -ENOMEM && (command->flags & EXEC_COMMAND_IGNORE_FAILURE)) { | |
c2503e35 RH |
5575 | log_unit_struct_errno(unit, LOG_INFO, r, |
5576 | "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR, | |
5577 | LOG_UNIT_INVOCATION_ID(unit), | |
5578 | LOG_UNIT_MESSAGE(unit, "Executable %s missing, skipping: %m", | |
5579 | command->path), | |
5580 | "EXECUTABLE=%s", command->path); | |
9f71ba8d ZJS |
5581 | return 0; |
5582 | } | |
5583 | ||
5584 | *exit_status = EXIT_EXEC; | |
c2503e35 RH |
5585 | |
5586 | return log_unit_struct_errno(unit, LOG_INFO, r, | |
5587 | "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR, | |
5588 | LOG_UNIT_INVOCATION_ID(unit), | |
5589 | LOG_UNIT_MESSAGE(unit, "Failed to locate executable %s: %m", | |
5590 | command->path), | |
5591 | "EXECUTABLE=%s", command->path); | |
9f71ba8d ZJS |
5592 | } |
5593 | ||
b83d5050 ZJS |
5594 | r = add_shifted_fd(keep_fds, ELEMENTSOF(keep_fds), &n_keep_fds, executable_fd, &executable_fd); |
5595 | if (r < 0) { | |
5596 | *exit_status = EXIT_FDS; | |
5597 | return log_unit_error_errno(unit, r, "Failed to shift fd and set FD_CLOEXEC: %m"); | |
5598 | } | |
5599 | ||
9f71ba8d | 5600 | #if HAVE_SELINUX |
49590d67 | 5601 | if (needs_sandboxing && use_selinux && params->selinux_context_net) { |
254d1313 | 5602 | int fd = -EBADF; |
49590d67 MS |
5603 | |
5604 | if (socket_fd >= 0) | |
5605 | fd = socket_fd; | |
5606 | else if (params->n_socket_fds == 1) | |
5607 | /* If stdin is not connected to a socket but we are triggered by exactly one socket unit then we | |
5608 | * use context from that fd to compute the label. */ | |
5609 | fd = params->fds[0]; | |
5610 | ||
5611 | if (fd >= 0) { | |
5612 | r = mac_selinux_get_child_mls_label(fd, executable, context->selinux_context, &mac_selinux_context_net); | |
006d1864 TM |
5613 | if (r < 0) { |
5614 | if (!context->selinux_context_ignore) { | |
5615 | *exit_status = EXIT_SELINUX_CONTEXT; | |
5616 | return log_unit_error_errno(unit, r, "Failed to determine SELinux context: %m"); | |
5617 | } | |
5618 | log_unit_debug_errno(unit, r, "Failed to determine SELinux context, ignoring: %m"); | |
49590d67 | 5619 | } |
9f71ba8d ZJS |
5620 | } |
5621 | } | |
5622 | #endif | |
5623 | ||
4a055e5a ZJS |
5624 | /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that |
5625 | * we are more aggressive this time, since we don't need socket_fd and the netns and ipcns fds any | |
5626 | * more. We do keep exec_fd however, if we have it, since we need to keep it open until the final | |
5627 | * execve(). */ | |
5686391b | 5628 | |
1da37e58 | 5629 | r = close_all_fds(keep_fds, n_keep_fds); |
ff0af2a1 LP |
5630 | if (r >= 0) |
5631 | r = shift_fds(fds, n_fds); | |
5632 | if (r >= 0) | |
cd48e23f | 5633 | r = flags_fds(fds, n_socket_fds, n_fds, context->non_blocking); |
ff0af2a1 LP |
5634 | if (r < 0) { |
5635 | *exit_status = EXIT_FDS; | |
12145637 | 5636 | return log_unit_error_errno(unit, r, "Failed to adjust passed file descriptors: %m"); |
d35fbf6b | 5637 | } |
e66cf1a3 | 5638 | |
5686391b LP |
5639 | /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off |
5640 | * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined, | |
5641 | * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we | |
5642 | * came this far. */ | |
5643 | ||
165a31c0 | 5644 | secure_bits = context->secure_bits; |
e66cf1a3 | 5645 | |
165a31c0 LP |
5646 | if (needs_sandboxing) { |
5647 | uint64_t bset; | |
e66cf1a3 | 5648 | |
4a055e5a ZJS |
5649 | /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. |
5650 | * (Note this is placed after the general resource limit initialization, see above, in order | |
5651 | * to take precedence.) */ | |
f4170c67 LP |
5652 | if (context->restrict_realtime && !context->rlimit[RLIMIT_RTPRIO]) { |
5653 | if (setrlimit(RLIMIT_RTPRIO, &RLIMIT_MAKE_CONST(0)) < 0) { | |
5654 | *exit_status = EXIT_LIMITS; | |
12145637 | 5655 | return log_unit_error_errno(unit, errno, "Failed to adjust RLIMIT_RTPRIO resource limit: %m"); |
f4170c67 LP |
5656 | } |
5657 | } | |
5658 | ||
37ac2744 JB |
5659 | #if ENABLE_SMACK |
5660 | /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the | |
5661 | * process. This is the latest place before dropping capabilities. Other MAC context are set later. */ | |
5662 | if (use_smack) { | |
aa5ae971 | 5663 | r = setup_smack(unit->manager, context, executable_fd); |
29ff6247 | 5664 | if (r < 0 && !context->smack_process_label_ignore) { |
37ac2744 JB |
5665 | *exit_status = EXIT_SMACK_PROCESS_LABEL; |
5666 | return log_unit_error_errno(unit, r, "Failed to set SMACK process label: %m"); | |
5667 | } | |
5668 | } | |
5669 | #endif | |
5670 | ||
165a31c0 LP |
5671 | bset = context->capability_bounding_set; |
5672 | /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for | |
5673 | * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own, | |
5674 | * instead of us doing that */ | |
5675 | if (needs_ambient_hack) | |
5676 | bset |= (UINT64_C(1) << CAP_SETPCAP) | | |
5677 | (UINT64_C(1) << CAP_SETUID) | | |
5678 | (UINT64_C(1) << CAP_SETGID); | |
5679 | ||
5680 | if (!cap_test_all(bset)) { | |
638fd8cc | 5681 | r = capability_bounding_set_drop(bset, /* right_now= */ false); |
ff0af2a1 LP |
5682 | if (r < 0) { |
5683 | *exit_status = EXIT_CAPABILITIES; | |
12145637 | 5684 | return log_unit_error_errno(unit, r, "Failed to drop capabilities: %m"); |
3b8bddde | 5685 | } |
4c2630eb | 5686 | } |
3b8bddde | 5687 | |
16fcb191 TK |
5688 | /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with |
5689 | * keep-caps set. | |
a954b249 LP |
5690 | * |
5691 | * To be able to raise the ambient capabilities after setresuid() they have to be added to | |
5692 | * the inherited set and keep caps has to be set (done in enforce_user()). After setresuid() | |
5693 | * the ambient capabilities can be raised as they are present in the permitted and | |
5694 | * inhertiable set. However it is possible that someone wants to set ambient capabilities | |
5695 | * without changing the user, so we also set the ambient capabilities here. | |
5696 | * | |
5697 | * The requested ambient capabilities are raised in the inheritable set if the second | |
5698 | * argument is true. */ | |
943800f4 | 5699 | if (!needs_ambient_hack) { |
638fd8cc | 5700 | r = capability_ambient_set_apply(capability_ambient_set, /* also_inherit= */ true); |
755d4b67 IP |
5701 | if (r < 0) { |
5702 | *exit_status = EXIT_CAPABILITIES; | |
12145637 | 5703 | return log_unit_error_errno(unit, r, "Failed to apply ambient capabilities (before UID change): %m"); |
755d4b67 | 5704 | } |
755d4b67 | 5705 | } |
165a31c0 | 5706 | } |
755d4b67 | 5707 | |
fa97f630 | 5708 | /* chroot to root directory first, before we lose the ability to chroot */ |
9c0c6701 | 5709 | r = apply_root_directory(context, params, runtime, needs_mount_namespace, exit_status); |
fa97f630 JB |
5710 | if (r < 0) |
5711 | return log_unit_error_errno(unit, r, "Chrooting to the requested root directory failed: %m"); | |
5712 | ||
165a31c0 | 5713 | if (needs_setuid) { |
08f67696 | 5714 | if (uid_is_valid(uid)) { |
638fd8cc | 5715 | r = enforce_user(context, uid, capability_ambient_set); |
ff0af2a1 LP |
5716 | if (r < 0) { |
5717 | *exit_status = EXIT_USER; | |
12145637 | 5718 | return log_unit_error_errno(unit, r, "Failed to change UID to " UID_FMT ": %m", uid); |
5b6319dc | 5719 | } |
165a31c0 | 5720 | |
638fd8cc | 5721 | if (!needs_ambient_hack && capability_ambient_set != 0) { |
755d4b67 | 5722 | |
16fcb191 | 5723 | /* Raise the ambient capabilities after user change. */ |
638fd8cc | 5724 | r = capability_ambient_set_apply(capability_ambient_set, /* also_inherit= */ false); |
755d4b67 IP |
5725 | if (r < 0) { |
5726 | *exit_status = EXIT_CAPABILITIES; | |
12145637 | 5727 | return log_unit_error_errno(unit, r, "Failed to apply ambient capabilities (after UID change): %m"); |
755d4b67 | 5728 | } |
755d4b67 | 5729 | } |
5b6319dc | 5730 | } |
165a31c0 | 5731 | } |
d35fbf6b | 5732 | |
56ef8db9 JB |
5733 | /* Apply working directory here, because the working directory might be on NFS and only the user running |
5734 | * this service might have the correct privilege to change to the working directory */ | |
9c0c6701 | 5735 | r = apply_working_directory(context, params, runtime, home, exit_status); |
56ef8db9 JB |
5736 | if (r < 0) |
5737 | return log_unit_error_errno(unit, r, "Changing to the requested working directory failed: %m"); | |
5738 | ||
165a31c0 | 5739 | if (needs_sandboxing) { |
37ac2744 | 5740 | /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to |
5cd9cd35 LP |
5741 | * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires |
5742 | * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls | |
5743 | * are restricted. */ | |
5744 | ||
349cc4a5 | 5745 | #if HAVE_SELINUX |
43b1f709 | 5746 | if (use_selinux) { |
5cd9cd35 LP |
5747 | char *exec_context = mac_selinux_context_net ?: context->selinux_context; |
5748 | ||
5749 | if (exec_context) { | |
5750 | r = setexeccon(exec_context); | |
006d1864 TM |
5751 | if (r < 0) { |
5752 | if (!context->selinux_context_ignore) { | |
5753 | *exit_status = EXIT_SELINUX_CONTEXT; | |
5754 | return log_unit_error_errno(unit, r, "Failed to change SELinux context to %s: %m", exec_context); | |
5755 | } | |
5756 | log_unit_debug_errno(unit, r, "Failed to change SELinux context to %s, ignoring: %m", exec_context); | |
5cd9cd35 LP |
5757 | } |
5758 | } | |
5759 | } | |
5760 | #endif | |
5761 | ||
349cc4a5 | 5762 | #if HAVE_APPARMOR |
43b1f709 | 5763 | if (use_apparmor && context->apparmor_profile) { |
5cd9cd35 LP |
5764 | r = aa_change_onexec(context->apparmor_profile); |
5765 | if (r < 0 && !context->apparmor_profile_ignore) { | |
5766 | *exit_status = EXIT_APPARMOR_PROFILE; | |
12145637 | 5767 | return log_unit_error_errno(unit, errno, "Failed to prepare AppArmor profile change to %s: %m", context->apparmor_profile); |
5cd9cd35 LP |
5768 | } |
5769 | } | |
5770 | #endif | |
5771 | ||
a954b249 LP |
5772 | /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential |
5773 | * EPERMs we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits | |
5774 | * requires CAP_SETPCAP. */ | |
dbdc4098 | 5775 | if (prctl(PR_GET_SECUREBITS) != secure_bits) { |
69e3234d | 5776 | /* CAP_SETPCAP is required to set securebits. This capability is raised into the |
dbdc4098 | 5777 | * effective set here. |
a954b249 LP |
5778 | * |
5779 | * The effective set is overwritten during execve() with the following values: | |
5780 | * | |
dbdc4098 | 5781 | * - ambient set (for non-root processes) |
a954b249 | 5782 | * |
dbdc4098 TK |
5783 | * - (inheritable | bounding) set for root processes) |
5784 | * | |
5785 | * Hence there is no security impact to raise it in the effective set before execve | |
5786 | */ | |
a954b249 | 5787 | r = capability_gain_cap_setpcap(/* return_caps= */ NULL); |
dbdc4098 TK |
5788 | if (r < 0) { |
5789 | *exit_status = EXIT_CAPABILITIES; | |
5790 | return log_unit_error_errno(unit, r, "Failed to gain CAP_SETPCAP for setting secure bits"); | |
5791 | } | |
755d4b67 | 5792 | if (prctl(PR_SET_SECUREBITS, secure_bits) < 0) { |
ff0af2a1 | 5793 | *exit_status = EXIT_SECUREBITS; |
12145637 | 5794 | return log_unit_error_errno(unit, errno, "Failed to set process secure bits: %m"); |
ff01d048 | 5795 | } |
dbdc4098 | 5796 | } |
5b6319dc | 5797 | |
59eeb84b | 5798 | if (context_has_no_new_privileges(context)) |
d35fbf6b | 5799 | if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { |
ff0af2a1 | 5800 | *exit_status = EXIT_NO_NEW_PRIVILEGES; |
12145637 | 5801 | return log_unit_error_errno(unit, errno, "Failed to disable new privileges: %m"); |
d35fbf6b DM |
5802 | } |
5803 | ||
349cc4a5 | 5804 | #if HAVE_SECCOMP |
469830d1 LP |
5805 | r = apply_address_families(unit, context); |
5806 | if (r < 0) { | |
5807 | *exit_status = EXIT_ADDRESS_FAMILIES; | |
12145637 | 5808 | return log_unit_error_errno(unit, r, "Failed to restrict address families: %m"); |
4c2630eb | 5809 | } |
04aa0cb9 | 5810 | |
469830d1 LP |
5811 | r = apply_memory_deny_write_execute(unit, context); |
5812 | if (r < 0) { | |
5813 | *exit_status = EXIT_SECCOMP; | |
12145637 | 5814 | return log_unit_error_errno(unit, r, "Failed to disable writing to executable memory: %m"); |
f3e43635 | 5815 | } |
f4170c67 | 5816 | |
469830d1 LP |
5817 | r = apply_restrict_realtime(unit, context); |
5818 | if (r < 0) { | |
5819 | *exit_status = EXIT_SECCOMP; | |
12145637 | 5820 | return log_unit_error_errno(unit, r, "Failed to apply realtime restrictions: %m"); |
f4170c67 LP |
5821 | } |
5822 | ||
f69567cb LP |
5823 | r = apply_restrict_suid_sgid(unit, context); |
5824 | if (r < 0) { | |
5825 | *exit_status = EXIT_SECCOMP; | |
5826 | return log_unit_error_errno(unit, r, "Failed to apply SUID/SGID restrictions: %m"); | |
5827 | } | |
5828 | ||
add00535 LP |
5829 | r = apply_restrict_namespaces(unit, context); |
5830 | if (r < 0) { | |
5831 | *exit_status = EXIT_SECCOMP; | |
12145637 | 5832 | return log_unit_error_errno(unit, r, "Failed to apply namespace restrictions: %m"); |
add00535 LP |
5833 | } |
5834 | ||
469830d1 LP |
5835 | r = apply_protect_sysctl(unit, context); |
5836 | if (r < 0) { | |
5837 | *exit_status = EXIT_SECCOMP; | |
12145637 | 5838 | return log_unit_error_errno(unit, r, "Failed to apply sysctl restrictions: %m"); |
502d704e DH |
5839 | } |
5840 | ||
469830d1 LP |
5841 | r = apply_protect_kernel_modules(unit, context); |
5842 | if (r < 0) { | |
5843 | *exit_status = EXIT_SECCOMP; | |
12145637 | 5844 | return log_unit_error_errno(unit, r, "Failed to apply module loading restrictions: %m"); |
59eeb84b LP |
5845 | } |
5846 | ||
84703040 KK |
5847 | r = apply_protect_kernel_logs(unit, context); |
5848 | if (r < 0) { | |
5849 | *exit_status = EXIT_SECCOMP; | |
5850 | return log_unit_error_errno(unit, r, "Failed to apply kernel log restrictions: %m"); | |
5851 | } | |
5852 | ||
fc64760d KK |
5853 | r = apply_protect_clock(unit, context); |
5854 | if (r < 0) { | |
5855 | *exit_status = EXIT_SECCOMP; | |
5856 | return log_unit_error_errno(unit, r, "Failed to apply clock restrictions: %m"); | |
5857 | } | |
5858 | ||
469830d1 LP |
5859 | r = apply_private_devices(unit, context); |
5860 | if (r < 0) { | |
5861 | *exit_status = EXIT_SECCOMP; | |
12145637 | 5862 | return log_unit_error_errno(unit, r, "Failed to set up private devices: %m"); |
469830d1 LP |
5863 | } |
5864 | ||
5865 | r = apply_syscall_archs(unit, context); | |
5866 | if (r < 0) { | |
5867 | *exit_status = EXIT_SECCOMP; | |
12145637 | 5868 | return log_unit_error_errno(unit, r, "Failed to apply syscall architecture restrictions: %m"); |
ba128bb8 LP |
5869 | } |
5870 | ||
78e864e5 TM |
5871 | r = apply_lock_personality(unit, context); |
5872 | if (r < 0) { | |
5873 | *exit_status = EXIT_SECCOMP; | |
12145637 | 5874 | return log_unit_error_errno(unit, r, "Failed to lock personalities: %m"); |
78e864e5 TM |
5875 | } |
5876 | ||
9df2cdd8 TM |
5877 | r = apply_syscall_log(unit, context); |
5878 | if (r < 0) { | |
5879 | *exit_status = EXIT_SECCOMP; | |
5880 | return log_unit_error_errno(unit, r, "Failed to apply system call log filters: %m"); | |
5881 | } | |
5882 | ||
5cd9cd35 LP |
5883 | /* This really should remain the last step before the execve(), to make sure our own code is unaffected |
5884 | * by the filter as little as possible. */ | |
165a31c0 | 5885 | r = apply_syscall_filter(unit, context, needs_ambient_hack); |
469830d1 LP |
5886 | if (r < 0) { |
5887 | *exit_status = EXIT_SECCOMP; | |
12145637 | 5888 | return log_unit_error_errno(unit, r, "Failed to apply system call filters: %m"); |
d35fbf6b DM |
5889 | } |
5890 | #endif | |
b1994387 ILG |
5891 | |
5892 | #if HAVE_LIBBPF | |
5893 | r = apply_restrict_filesystems(unit, context); | |
5894 | if (r < 0) { | |
5895 | *exit_status = EXIT_BPF; | |
5896 | return log_unit_error_errno(unit, r, "Failed to restrict filesystems: %m"); | |
5897 | } | |
5898 | #endif | |
5899 | ||
d35fbf6b | 5900 | } |
034c6ed7 | 5901 | |
00819cc1 LP |
5902 | if (!strv_isempty(context->unset_environment)) { |
5903 | char **ee = NULL; | |
5904 | ||
5905 | ee = strv_env_delete(accum_env, 1, context->unset_environment); | |
5906 | if (!ee) { | |
5907 | *exit_status = EXIT_MEMORY; | |
12145637 | 5908 | return log_oom(); |
00819cc1 LP |
5909 | } |
5910 | ||
130d3d22 | 5911 | strv_free_and_replace(accum_env, ee); |
00819cc1 LP |
5912 | } |
5913 | ||
7ca69792 | 5914 | if (!FLAGS_SET(command->flags, EXEC_COMMAND_NO_ENV_EXPAND)) { |
f331434d LP |
5915 | _cleanup_strv_free_ char **unset_variables = NULL, **bad_variables = NULL; |
5916 | ||
5917 | r = replace_env_argv(command->argv, accum_env, &replaced_argv, &unset_variables, &bad_variables); | |
5918 | if (r < 0) { | |
7ca69792 | 5919 | *exit_status = EXIT_MEMORY; |
f331434d | 5920 | return log_unit_error_errno(unit, r, "Failed to replace environment variables: %m"); |
7ca69792 AZ |
5921 | } |
5922 | final_argv = replaced_argv; | |
f331434d LP |
5923 | |
5924 | if (!strv_isempty(unset_variables)) { | |
5925 | _cleanup_free_ char *ju = strv_join(unset_variables, ", "); | |
5926 | log_unit_warning(unit, "Referenced but unset environment variable evaluates to an empty string: %s", strna(ju)); | |
5927 | } | |
5928 | ||
5929 | if (!strv_isempty(bad_variables)) { | |
5930 | _cleanup_free_ char *jb = strv_join(bad_variables, ", "); | |
5931 | log_unit_warning(unit, "Invalid environment variable name evaluates to an empty string: %s", strna(jb));; | |
5932 | } | |
7ca69792 AZ |
5933 | } else |
5934 | final_argv = command->argv; | |
034c6ed7 | 5935 | |
3ff67ec4 | 5936 | log_command_line(unit, "Executing", executable, final_argv); |
dd305ec9 | 5937 | |
5686391b LP |
5938 | if (exec_fd >= 0) { |
5939 | uint8_t hot = 1; | |
5940 | ||
5941 | /* We have finished with all our initializations. Let's now let the manager know that. From this point | |
5942 | * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */ | |
5943 | ||
5944 | if (write(exec_fd, &hot, sizeof(hot)) < 0) { | |
5945 | *exit_status = EXIT_EXEC; | |
5946 | return log_unit_error_errno(unit, errno, "Failed to enable exec_fd: %m"); | |
5947 | } | |
5948 | } | |
5949 | ||
a6d9111c | 5950 | r = fexecve_or_execve(executable_fd, executable, final_argv, accum_env); |
5686391b LP |
5951 | |
5952 | if (exec_fd >= 0) { | |
5953 | uint8_t hot = 0; | |
5954 | ||
5955 | /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager | |
5956 | * that POLLHUP on it no longer means execve() succeeded. */ | |
5957 | ||
5958 | if (write(exec_fd, &hot, sizeof(hot)) < 0) { | |
5959 | *exit_status = EXIT_EXEC; | |
5960 | return log_unit_error_errno(unit, errno, "Failed to disable exec_fd: %m"); | |
5961 | } | |
5962 | } | |
12145637 | 5963 | |
ff0af2a1 | 5964 | *exit_status = EXIT_EXEC; |
9f71ba8d | 5965 | return log_unit_error_errno(unit, r, "Failed to execute %s: %m", executable); |
d35fbf6b | 5966 | } |
81a2b7ce | 5967 | |
34cf6c43 | 5968 | static int exec_context_load_environment(const Unit *unit, const ExecContext *c, char ***l); |
2caa38e9 | 5969 | static int exec_context_named_iofds(const ExecContext *c, const ExecParameters *p, int named_iofds[static 3]); |
34cf6c43 | 5970 | |
f2341e0a LP |
5971 | int exec_spawn(Unit *unit, |
5972 | ExecCommand *command, | |
d35fbf6b DM |
5973 | const ExecContext *context, |
5974 | const ExecParameters *params, | |
28135da3 | 5975 | ExecRuntime *runtime, |
6bb00842 | 5976 | const CGroupContext *cgroup_context, |
d35fbf6b | 5977 | pid_t *ret) { |
8351ceae | 5978 | |
ee39ca20 | 5979 | int socket_fd, r, named_iofds[3] = { -1, -1, -1 }, *fds = NULL; |
78f93209 | 5980 | _cleanup_free_ char *subcgroup_path = NULL; |
d35fbf6b | 5981 | _cleanup_strv_free_ char **files_env = NULL; |
da6053d0 | 5982 | size_t n_storage_fds = 0, n_socket_fds = 0; |
d35fbf6b | 5983 | pid_t pid; |
8351ceae | 5984 | |
f2341e0a | 5985 | assert(unit); |
d35fbf6b DM |
5986 | assert(command); |
5987 | assert(context); | |
5988 | assert(ret); | |
5989 | assert(params); | |
25b583d7 | 5990 | assert(params->fds || (params->n_socket_fds + params->n_storage_fds <= 0)); |
4298d0b5 | 5991 | |
4b2af439 DDM |
5992 | LOG_CONTEXT_PUSH_UNIT(unit); |
5993 | ||
d35fbf6b DM |
5994 | if (context->std_input == EXEC_INPUT_SOCKET || |
5995 | context->std_output == EXEC_OUTPUT_SOCKET || | |
5996 | context->std_error == EXEC_OUTPUT_SOCKET) { | |
17df7223 | 5997 | |
d85ff944 YW |
5998 | if (params->n_socket_fds > 1) |
5999 | return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EINVAL), "Got more than one socket."); | |
eef65bf3 | 6000 | |
d85ff944 YW |
6001 | if (params->n_socket_fds == 0) |
6002 | return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EINVAL), "Got no socket."); | |
488ab41c | 6003 | |
d35fbf6b DM |
6004 | socket_fd = params->fds[0]; |
6005 | } else { | |
254d1313 | 6006 | socket_fd = -EBADF; |
d35fbf6b | 6007 | fds = params->fds; |
9b141911 | 6008 | n_socket_fds = params->n_socket_fds; |
25b583d7 | 6009 | n_storage_fds = params->n_storage_fds; |
d35fbf6b | 6010 | } |
94f04347 | 6011 | |
34cf6c43 | 6012 | r = exec_context_named_iofds(context, params, named_iofds); |
52c239d7 LB |
6013 | if (r < 0) |
6014 | return log_unit_error_errno(unit, r, "Failed to load a named file descriptor: %m"); | |
6015 | ||
f2341e0a | 6016 | r = exec_context_load_environment(unit, context, &files_env); |
ff0af2a1 | 6017 | if (r < 0) |
f2341e0a | 6018 | return log_unit_error_errno(unit, r, "Failed to load environment files: %m"); |
034c6ed7 | 6019 | |
9f71ba8d ZJS |
6020 | /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db |
6021 | and, until the next SELinux policy changes, we save further reloads in future children. */ | |
2df2152c CG |
6022 | mac_selinux_maybe_reload(); |
6023 | ||
3ff67ec4 ZJS |
6024 | /* We won't know the real executable path until we create the mount namespace in the child, but we |
6025 | want to log from the parent, so we use the possibly inaccurate path here. */ | |
6026 | log_command_line(unit, "About to execute", command->path, command->argv); | |
12145637 | 6027 | |
78f93209 | 6028 | if (params->cgroup_path) { |
a8b993dc | 6029 | r = exec_parameters_get_cgroup_path(params, cgroup_context, &subcgroup_path); |
78f93209 LP |
6030 | if (r < 0) |
6031 | return log_unit_error_errno(unit, r, "Failed to acquire subcgroup path: %m"); | |
18c1e481 LP |
6032 | if (r > 0) { |
6033 | /* If there's a subcgroup, then let's create it here now (the main cgroup was already | |
6034 | * realized by the unit logic) */ | |
6035 | ||
78f93209 LP |
6036 | r = cg_create(SYSTEMD_CGROUP_CONTROLLER, subcgroup_path); |
6037 | if (r < 0) | |
a8b993dc | 6038 | return log_unit_error_errno(unit, r, "Failed to create subcgroup '%s': %m", subcgroup_path); |
78f93209 LP |
6039 | } |
6040 | } | |
6041 | ||
d35fbf6b DM |
6042 | pid = fork(); |
6043 | if (pid < 0) | |
74129a12 | 6044 | return log_unit_error_errno(unit, errno, "Failed to fork: %m"); |
d35fbf6b DM |
6045 | |
6046 | if (pid == 0) { | |
12145637 | 6047 | int exit_status = EXIT_SUCCESS; |
ff0af2a1 | 6048 | |
f2341e0a LP |
6049 | r = exec_child(unit, |
6050 | command, | |
ff0af2a1 LP |
6051 | context, |
6052 | params, | |
6053 | runtime, | |
6bb00842 | 6054 | cgroup_context, |
ff0af2a1 | 6055 | socket_fd, |
52c239d7 | 6056 | named_iofds, |
4c47affc | 6057 | fds, |
9b141911 | 6058 | n_socket_fds, |
25b583d7 | 6059 | n_storage_fds, |
ff0af2a1 | 6060 | files_env, |
00d9ef85 | 6061 | unit->manager->user_lookup_fds[1], |
12145637 LP |
6062 | &exit_status); |
6063 | ||
e1714f02 ZJS |
6064 | if (r < 0) { |
6065 | const char *status = | |
6066 | exit_status_to_string(exit_status, | |
e04ed6db | 6067 | EXIT_STATUS_LIBC | EXIT_STATUS_SYSTEMD); |
e1714f02 | 6068 | |
c2503e35 RH |
6069 | log_unit_struct_errno(unit, LOG_ERR, r, |
6070 | "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR, | |
6071 | LOG_UNIT_INVOCATION_ID(unit), | |
6072 | LOG_UNIT_MESSAGE(unit, "Failed at step %s spawning %s: %m", | |
6073 | status, command->path), | |
6074 | "EXECUTABLE=%s", command->path); | |
e1714f02 | 6075 | } |
4c2630eb | 6076 | |
ff0af2a1 | 6077 | _exit(exit_status); |
034c6ed7 LP |
6078 | } |
6079 | ||
f2341e0a | 6080 | log_unit_debug(unit, "Forked %s as "PID_FMT, command->path, pid); |
23635a85 | 6081 | |
78f93209 LP |
6082 | /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever |
6083 | * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the | |
6084 | * process will be killed too). */ | |
6085 | if (subcgroup_path) | |
6086 | (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER, subcgroup_path, pid); | |
2da3263a | 6087 | |
b58b4116 | 6088 | exec_status_start(&command->exec_status, pid); |
9fb86720 | 6089 | |
034c6ed7 | 6090 | *ret = pid; |
5cb5a6ff LP |
6091 | return 0; |
6092 | } | |
6093 | ||
034c6ed7 LP |
6094 | void exec_context_init(ExecContext *c) { |
6095 | assert(c); | |
6096 | ||
4c12626c | 6097 | c->umask = 0022; |
0692548c | 6098 | c->ioprio = IOPRIO_DEFAULT_CLASS_AND_PRIO; |
94f04347 | 6099 | c->cpu_sched_policy = SCHED_OTHER; |
071830ff | 6100 | c->syslog_priority = LOG_DAEMON|LOG_INFO; |
74922904 | 6101 | c->syslog_level_prefix = true; |
353e12c2 | 6102 | c->ignore_sigpipe = true; |
3a43da28 | 6103 | c->timer_slack_nsec = NSEC_INFINITY; |
050f7277 | 6104 | c->personality = PERSONALITY_INVALID; |
5b10116e ZJS |
6105 | for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) |
6106 | c->directories[t].mode = 0755; | |
12213aed | 6107 | c->timeout_clean_usec = USEC_INFINITY; |
3fd5190b | 6108 | c->capability_bounding_set = CAP_MASK_UNSET; |
aa9d574d YW |
6109 | assert_cc(NAMESPACE_FLAGS_INITIAL != NAMESPACE_FLAGS_ALL); |
6110 | c->restrict_namespaces = NAMESPACE_FLAGS_INITIAL; | |
d3070fbd | 6111 | c->log_level_max = -1; |
005bfaf1 TM |
6112 | #if HAVE_SECCOMP |
6113 | c->syscall_errno = SECCOMP_ERROR_NUMBER_KILL; | |
6114 | #endif | |
51462135 DDM |
6115 | c->tty_rows = UINT_MAX; |
6116 | c->tty_cols = UINT_MAX; | |
b070c7c0 | 6117 | numa_policy_reset(&c->numa_policy); |
24002121 | 6118 | c->private_mounts = -1; |
85614c6e | 6119 | c->memory_ksm = -1; |
034c6ed7 LP |
6120 | } |
6121 | ||
613b411c | 6122 | void exec_context_done(ExecContext *c) { |
5cb5a6ff LP |
6123 | assert(c); |
6124 | ||
6796073e LP |
6125 | c->environment = strv_free(c->environment); |
6126 | c->environment_files = strv_free(c->environment_files); | |
b4c14404 | 6127 | c->pass_environment = strv_free(c->pass_environment); |
00819cc1 | 6128 | c->unset_environment = strv_free(c->unset_environment); |
8c7be95e | 6129 | |
31ce987c | 6130 | rlimit_free_all(c->rlimit); |
034c6ed7 | 6131 | |
5b10116e | 6132 | for (size_t l = 0; l < 3; l++) { |
52c239d7 | 6133 | c->stdio_fdname[l] = mfree(c->stdio_fdname[l]); |
2038c3f5 LP |
6134 | c->stdio_file[l] = mfree(c->stdio_file[l]); |
6135 | } | |
52c239d7 | 6136 | |
a1e58e8e LP |
6137 | c->working_directory = mfree(c->working_directory); |
6138 | c->root_directory = mfree(c->root_directory); | |
915e6d16 | 6139 | c->root_image = mfree(c->root_image); |
18d73705 | 6140 | c->root_image_options = mount_options_free_all(c->root_image_options); |
0389f4fa LB |
6141 | c->root_hash = mfree(c->root_hash); |
6142 | c->root_hash_size = 0; | |
6143 | c->root_hash_path = mfree(c->root_hash_path); | |
d4d55b0d LB |
6144 | c->root_hash_sig = mfree(c->root_hash_sig); |
6145 | c->root_hash_sig_size = 0; | |
6146 | c->root_hash_sig_path = mfree(c->root_hash_sig_path); | |
0389f4fa | 6147 | c->root_verity = mfree(c->root_verity); |
93f59701 | 6148 | c->extension_images = mount_image_free_many(c->extension_images, &c->n_extension_images); |
a07b9926 | 6149 | c->extension_directories = strv_free(c->extension_directories); |
a1e58e8e LP |
6150 | c->tty_path = mfree(c->tty_path); |
6151 | c->syslog_identifier = mfree(c->syslog_identifier); | |
6152 | c->user = mfree(c->user); | |
6153 | c->group = mfree(c->group); | |
034c6ed7 | 6154 | |
6796073e | 6155 | c->supplementary_groups = strv_free(c->supplementary_groups); |
94f04347 | 6156 | |
a1e58e8e | 6157 | c->pam_name = mfree(c->pam_name); |
5b6319dc | 6158 | |
2a624c36 AP |
6159 | c->read_only_paths = strv_free(c->read_only_paths); |
6160 | c->read_write_paths = strv_free(c->read_write_paths); | |
6161 | c->inaccessible_paths = strv_free(c->inaccessible_paths); | |
ddc155b2 TM |
6162 | c->exec_paths = strv_free(c->exec_paths); |
6163 | c->no_exec_paths = strv_free(c->no_exec_paths); | |
8c35c10d | 6164 | c->exec_search_path = strv_free(c->exec_search_path); |
82c121a4 | 6165 | |
d2d6c096 | 6166 | bind_mount_free_many(c->bind_mounts, c->n_bind_mounts); |
8e06d57c YW |
6167 | c->bind_mounts = NULL; |
6168 | c->n_bind_mounts = 0; | |
2abd4e38 YW |
6169 | temporary_filesystem_free_many(c->temporary_filesystems, c->n_temporary_filesystems); |
6170 | c->temporary_filesystems = NULL; | |
6171 | c->n_temporary_filesystems = 0; | |
b3d13314 | 6172 | c->mount_images = mount_image_free_many(c->mount_images, &c->n_mount_images); |
d2d6c096 | 6173 | |
0985c7c4 | 6174 | cpu_set_reset(&c->cpu_set); |
b070c7c0 | 6175 | numa_policy_reset(&c->numa_policy); |
86a3475b | 6176 | |
a1e58e8e LP |
6177 | c->utmp_id = mfree(c->utmp_id); |
6178 | c->selinux_context = mfree(c->selinux_context); | |
6179 | c->apparmor_profile = mfree(c->apparmor_profile); | |
5b8e1b77 | 6180 | c->smack_process_label = mfree(c->smack_process_label); |
eef65bf3 | 6181 | |
9b412709 | 6182 | c->restrict_filesystems = set_free_free(c->restrict_filesystems); |
b1994387 | 6183 | |
8cfa775f | 6184 | c->syscall_filter = hashmap_free(c->syscall_filter); |
525d3cc7 LP |
6185 | c->syscall_archs = set_free(c->syscall_archs); |
6186 | c->address_families = set_free(c->address_families); | |
e66cf1a3 | 6187 | |
5b10116e | 6188 | for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) |
211a3d87 | 6189 | exec_directory_done(&c->directories[t]); |
d3070fbd LP |
6190 | |
6191 | c->log_level_max = -1; | |
6192 | ||
6193 | exec_context_free_log_extra_fields(c); | |
9b412709 FS |
6194 | c->log_filter_allowed_patterns = set_free_free(c->log_filter_allowed_patterns); |
6195 | c->log_filter_denied_patterns = set_free_free(c->log_filter_denied_patterns); | |
08f3be7a | 6196 | |
5ac1530e ZJS |
6197 | c->log_ratelimit_interval_usec = 0; |
6198 | c->log_ratelimit_burst = 0; | |
90fc172e | 6199 | |
08f3be7a LP |
6200 | c->stdin_data = mfree(c->stdin_data); |
6201 | c->stdin_data_size = 0; | |
a8d08f39 LP |
6202 | |
6203 | c->network_namespace_path = mfree(c->network_namespace_path); | |
71d1e583 | 6204 | c->ipc_namespace_path = mfree(c->ipc_namespace_path); |
91dd5f7c LP |
6205 | |
6206 | c->log_namespace = mfree(c->log_namespace); | |
bb0c0d6f | 6207 | |
43144be4 | 6208 | c->load_credentials = hashmap_free(c->load_credentials); |
bb0c0d6f | 6209 | c->set_credentials = hashmap_free(c->set_credentials); |
9b412709 | 6210 | c->import_credentials = set_free_free(c->import_credentials); |
84be0c71 LP |
6211 | |
6212 | c->root_image_policy = image_policy_free(c->root_image_policy); | |
6213 | c->mount_image_policy = image_policy_free(c->mount_image_policy); | |
6214 | c->extension_image_policy = image_policy_free(c->extension_image_policy); | |
e66cf1a3 LP |
6215 | } |
6216 | ||
34cf6c43 | 6217 | int exec_context_destroy_runtime_directory(const ExecContext *c, const char *runtime_prefix) { |
e66cf1a3 LP |
6218 | assert(c); |
6219 | ||
6220 | if (!runtime_prefix) | |
6221 | return 0; | |
6222 | ||
211a3d87 | 6223 | for (size_t i = 0; i < c->directories[EXEC_DIRECTORY_RUNTIME].n_items; i++) { |
c2b2df60 | 6224 | _cleanup_free_ char *p = NULL; |
e66cf1a3 | 6225 | |
494d0247 | 6226 | if (exec_directory_is_private(c, EXEC_DIRECTORY_RUNTIME)) |
211a3d87 | 6227 | p = path_join(runtime_prefix, "private", c->directories[EXEC_DIRECTORY_RUNTIME].items[i].path); |
494d0247 | 6228 | else |
211a3d87 | 6229 | p = path_join(runtime_prefix, c->directories[EXEC_DIRECTORY_RUNTIME].items[i].path); |
e66cf1a3 LP |
6230 | if (!p) |
6231 | return -ENOMEM; | |
6232 | ||
7bc4bf4a LP |
6233 | /* We execute this synchronously, since we need to be sure this is gone when we start the |
6234 | * service next. */ | |
c6878637 | 6235 | (void) rm_rf(p, REMOVE_ROOT); |
211a3d87 | 6236 | |
211a3d87 LB |
6237 | STRV_FOREACH(symlink, c->directories[EXEC_DIRECTORY_RUNTIME].items[i].symlinks) { |
6238 | _cleanup_free_ char *symlink_abs = NULL; | |
6239 | ||
6240 | if (exec_directory_is_private(c, EXEC_DIRECTORY_RUNTIME)) | |
6241 | symlink_abs = path_join(runtime_prefix, "private", *symlink); | |
6242 | else | |
6243 | symlink_abs = path_join(runtime_prefix, *symlink); | |
6244 | if (!symlink_abs) | |
6245 | return -ENOMEM; | |
6246 | ||
6247 | (void) unlink(symlink_abs); | |
6248 | } | |
e66cf1a3 LP |
6249 | } |
6250 | ||
6251 | return 0; | |
5cb5a6ff LP |
6252 | } |
6253 | ||
bb0c0d6f LP |
6254 | int exec_context_destroy_credentials(const ExecContext *c, const char *runtime_prefix, const char *unit) { |
6255 | _cleanup_free_ char *p = NULL; | |
6256 | ||
6257 | assert(c); | |
6258 | ||
6259 | if (!runtime_prefix || !unit) | |
6260 | return 0; | |
6261 | ||
6262 | p = path_join(runtime_prefix, "credentials", unit); | |
6263 | if (!p) | |
6264 | return -ENOMEM; | |
6265 | ||
6266 | /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to | |
6267 | * unmount it, and afterwards remove the mount point */ | |
6268 | (void) umount2(p, MNT_DETACH|UMOUNT_NOFOLLOW); | |
6269 | (void) rm_rf(p, REMOVE_ROOT|REMOVE_CHMOD); | |
6270 | ||
6271 | return 0; | |
6272 | } | |
6273 | ||
b9f976fb MK |
6274 | int exec_context_destroy_mount_ns_dir(Unit *u) { |
6275 | _cleanup_free_ char *p = NULL; | |
6276 | ||
6277 | if (!u || !MANAGER_IS_SYSTEM(u->manager)) | |
6278 | return 0; | |
6279 | ||
6280 | p = path_join("/run/systemd/propagate/", u->id); | |
6281 | if (!p) | |
6282 | return -ENOMEM; | |
6283 | ||
6284 | /* This is only filled transiently (see mount_in_namespace()), should be empty or even non-existent*/ | |
6285 | if (rmdir(p) < 0 && errno != ENOENT) | |
6286 | log_unit_debug_errno(u, errno, "Unable to remove propagation dir '%s', ignoring: %m", p); | |
6287 | ||
6288 | return 0; | |
6289 | } | |
6290 | ||
34cf6c43 | 6291 | static void exec_command_done(ExecCommand *c) { |
43d0fcbd LP |
6292 | assert(c); |
6293 | ||
a1e58e8e | 6294 | c->path = mfree(c->path); |
6796073e | 6295 | c->argv = strv_free(c->argv); |
43d0fcbd LP |
6296 | } |
6297 | ||
da6053d0 | 6298 | void exec_command_done_array(ExecCommand *c, size_t n) { |
fe96c0f8 | 6299 | for (size_t i = 0; i < n; i++) |
43d0fcbd LP |
6300 | exec_command_done(c+i); |
6301 | } | |
6302 | ||
f1acf85a | 6303 | ExecCommand* exec_command_free_list(ExecCommand *c) { |
5cb5a6ff LP |
6304 | ExecCommand *i; |
6305 | ||
6306 | while ((i = c)) { | |
71fda00f | 6307 | LIST_REMOVE(command, c, i); |
43d0fcbd | 6308 | exec_command_done(i); |
5cb5a6ff LP |
6309 | free(i); |
6310 | } | |
f1acf85a ZJS |
6311 | |
6312 | return NULL; | |
5cb5a6ff LP |
6313 | } |
6314 | ||
da6053d0 | 6315 | void exec_command_free_array(ExecCommand **c, size_t n) { |
5b10116e | 6316 | for (size_t i = 0; i < n; i++) |
f1acf85a | 6317 | c[i] = exec_command_free_list(c[i]); |
034c6ed7 LP |
6318 | } |
6319 | ||
6a1d4d9f | 6320 | void exec_command_reset_status_array(ExecCommand *c, size_t n) { |
5b10116e | 6321 | for (size_t i = 0; i < n; i++) |
6a1d4d9f LP |
6322 | exec_status_reset(&c[i].exec_status); |
6323 | } | |
6324 | ||
6325 | void exec_command_reset_status_list_array(ExecCommand **c, size_t n) { | |
03677889 | 6326 | for (size_t i = 0; i < n; i++) |
6a1d4d9f LP |
6327 | LIST_FOREACH(command, z, c[i]) |
6328 | exec_status_reset(&z->exec_status); | |
6a1d4d9f LP |
6329 | } |
6330 | ||
039f0e70 | 6331 | typedef struct InvalidEnvInfo { |
34cf6c43 | 6332 | const Unit *unit; |
039f0e70 LP |
6333 | const char *path; |
6334 | } InvalidEnvInfo; | |
6335 | ||
6336 | static void invalid_env(const char *p, void *userdata) { | |
6337 | InvalidEnvInfo *info = userdata; | |
6338 | ||
f2341e0a | 6339 | log_unit_error(info->unit, "Ignoring invalid environment assignment '%s': %s", p, info->path); |
039f0e70 LP |
6340 | } |
6341 | ||
52c239d7 LB |
6342 | const char* exec_context_fdname(const ExecContext *c, int fd_index) { |
6343 | assert(c); | |
6344 | ||
6345 | switch (fd_index) { | |
5073ff6b | 6346 | |
52c239d7 LB |
6347 | case STDIN_FILENO: |
6348 | if (c->std_input != EXEC_INPUT_NAMED_FD) | |
6349 | return NULL; | |
5073ff6b | 6350 | |
52c239d7 | 6351 | return c->stdio_fdname[STDIN_FILENO] ?: "stdin"; |
5073ff6b | 6352 | |
52c239d7 LB |
6353 | case STDOUT_FILENO: |
6354 | if (c->std_output != EXEC_OUTPUT_NAMED_FD) | |
6355 | return NULL; | |
5073ff6b | 6356 | |
52c239d7 | 6357 | return c->stdio_fdname[STDOUT_FILENO] ?: "stdout"; |
5073ff6b | 6358 | |
52c239d7 LB |
6359 | case STDERR_FILENO: |
6360 | if (c->std_error != EXEC_OUTPUT_NAMED_FD) | |
6361 | return NULL; | |
5073ff6b | 6362 | |
52c239d7 | 6363 | return c->stdio_fdname[STDERR_FILENO] ?: "stderr"; |
5073ff6b | 6364 | |
52c239d7 LB |
6365 | default: |
6366 | return NULL; | |
6367 | } | |
6368 | } | |
6369 | ||
2caa38e9 LP |
6370 | static int exec_context_named_iofds( |
6371 | const ExecContext *c, | |
6372 | const ExecParameters *p, | |
6373 | int named_iofds[static 3]) { | |
6374 | ||
5b10116e | 6375 | size_t targets; |
56fbd561 | 6376 | const char* stdio_fdname[3]; |
da6053d0 | 6377 | size_t n_fds; |
52c239d7 LB |
6378 | |
6379 | assert(c); | |
6380 | assert(p); | |
2caa38e9 | 6381 | assert(named_iofds); |
52c239d7 LB |
6382 | |
6383 | targets = (c->std_input == EXEC_INPUT_NAMED_FD) + | |
6384 | (c->std_output == EXEC_OUTPUT_NAMED_FD) + | |
6385 | (c->std_error == EXEC_OUTPUT_NAMED_FD); | |
6386 | ||
5b10116e | 6387 | for (size_t i = 0; i < 3; i++) |
52c239d7 LB |
6388 | stdio_fdname[i] = exec_context_fdname(c, i); |
6389 | ||
4c47affc FB |
6390 | n_fds = p->n_storage_fds + p->n_socket_fds; |
6391 | ||
5b10116e | 6392 | for (size_t i = 0; i < n_fds && targets > 0; i++) |
56fbd561 ZJS |
6393 | if (named_iofds[STDIN_FILENO] < 0 && |
6394 | c->std_input == EXEC_INPUT_NAMED_FD && | |
6395 | stdio_fdname[STDIN_FILENO] && | |
6396 | streq(p->fd_names[i], stdio_fdname[STDIN_FILENO])) { | |
6397 | ||
52c239d7 LB |
6398 | named_iofds[STDIN_FILENO] = p->fds[i]; |
6399 | targets--; | |
56fbd561 ZJS |
6400 | |
6401 | } else if (named_iofds[STDOUT_FILENO] < 0 && | |
6402 | c->std_output == EXEC_OUTPUT_NAMED_FD && | |
6403 | stdio_fdname[STDOUT_FILENO] && | |
6404 | streq(p->fd_names[i], stdio_fdname[STDOUT_FILENO])) { | |
6405 | ||
52c239d7 LB |
6406 | named_iofds[STDOUT_FILENO] = p->fds[i]; |
6407 | targets--; | |
56fbd561 ZJS |
6408 | |
6409 | } else if (named_iofds[STDERR_FILENO] < 0 && | |
6410 | c->std_error == EXEC_OUTPUT_NAMED_FD && | |
6411 | stdio_fdname[STDERR_FILENO] && | |
6412 | streq(p->fd_names[i], stdio_fdname[STDERR_FILENO])) { | |
6413 | ||
52c239d7 LB |
6414 | named_iofds[STDERR_FILENO] = p->fds[i]; |
6415 | targets--; | |
6416 | } | |
6417 | ||
56fbd561 | 6418 | return targets == 0 ? 0 : -ENOENT; |
52c239d7 LB |
6419 | } |
6420 | ||
398a5009 ZJS |
6421 | static int exec_context_load_environment(const Unit *unit, const ExecContext *c, char ***ret) { |
6422 | _cleanup_strv_free_ char **v = NULL; | |
398a5009 | 6423 | int r; |
8c7be95e LP |
6424 | |
6425 | assert(c); | |
398a5009 | 6426 | assert(ret); |
8c7be95e LP |
6427 | |
6428 | STRV_FOREACH(i, c->environment_files) { | |
7fd1b19b | 6429 | _cleanup_globfree_ glob_t pglob = {}; |
398a5009 ZJS |
6430 | bool ignore = false; |
6431 | char *fn = *i; | |
8c7be95e LP |
6432 | |
6433 | if (fn[0] == '-') { | |
6434 | ignore = true; | |
313cefa1 | 6435 | fn++; |
8c7be95e LP |
6436 | } |
6437 | ||
6438 | if (!path_is_absolute(fn)) { | |
8c7be95e LP |
6439 | if (ignore) |
6440 | continue; | |
8c7be95e LP |
6441 | return -EINVAL; |
6442 | } | |
6443 | ||
2bef10ab | 6444 | /* Filename supports globbing, take all matching files */ |
398a5009 ZJS |
6445 | r = safe_glob(fn, 0, &pglob); |
6446 | if (r < 0) { | |
2bef10ab PL |
6447 | if (ignore) |
6448 | continue; | |
398a5009 | 6449 | return r; |
2bef10ab | 6450 | } |
8c7be95e | 6451 | |
d8c92e8b ZJS |
6452 | /* When we don't match anything, -ENOENT should be returned */ |
6453 | assert(pglob.gl_pathc > 0); | |
6454 | ||
fcc06682 | 6455 | for (size_t n = 0; n < pglob.gl_pathc; n++) { |
398a5009 ZJS |
6456 | _cleanup_strv_free_ char **p = NULL; |
6457 | ||
6458 | r = load_env_file(NULL, pglob.gl_pathv[n], &p); | |
6459 | if (r < 0) { | |
2bef10ab PL |
6460 | if (ignore) |
6461 | continue; | |
398a5009 | 6462 | return r; |
e9c1ea9d | 6463 | } |
398a5009 | 6464 | |
ebc05a09 | 6465 | /* Log invalid environment variables with filename */ |
039f0e70 LP |
6466 | if (p) { |
6467 | InvalidEnvInfo info = { | |
f2341e0a | 6468 | .unit = unit, |
039f0e70 LP |
6469 | .path = pglob.gl_pathv[n] |
6470 | }; | |
6471 | ||
6472 | p = strv_env_clean_with_callback(p, invalid_env, &info); | |
6473 | } | |
8c7be95e | 6474 | |
398a5009 ZJS |
6475 | if (!v) |
6476 | v = TAKE_PTR(p); | |
2bef10ab | 6477 | else { |
398a5009 | 6478 | char **m = strv_env_merge(v, p); |
c84a9488 | 6479 | if (!m) |
2bef10ab | 6480 | return -ENOMEM; |
2bef10ab | 6481 | |
398a5009 | 6482 | strv_free_and_replace(v, m); |
2bef10ab | 6483 | } |
8c7be95e LP |
6484 | } |
6485 | } | |
6486 | ||
398a5009 | 6487 | *ret = TAKE_PTR(v); |
8c7be95e LP |
6488 | |
6489 | return 0; | |
6490 | } | |
6491 | ||
6ac8fdc9 | 6492 | static bool tty_may_match_dev_console(const char *tty) { |
7b912648 | 6493 | _cleanup_free_ char *resolved = NULL; |
6ac8fdc9 | 6494 | |
1e22b5cd LP |
6495 | if (!tty) |
6496 | return true; | |
6497 | ||
a119ec7c | 6498 | tty = skip_dev_prefix(tty); |
6ac8fdc9 MS |
6499 | |
6500 | /* trivial identity? */ | |
6501 | if (streq(tty, "console")) | |
6502 | return true; | |
6503 | ||
7b912648 LP |
6504 | if (resolve_dev_console(&resolved) < 0) |
6505 | return true; /* if we could not resolve, assume it may */ | |
6ac8fdc9 MS |
6506 | |
6507 | /* "tty0" means the active VC, so it may be the same sometimes */ | |
955f1c85 | 6508 | return path_equal(resolved, tty) || (streq(resolved, "tty0") && tty_is_vc(tty)); |
6ac8fdc9 MS |
6509 | } |
6510 | ||
6c0ae739 LP |
6511 | static bool exec_context_may_touch_tty(const ExecContext *ec) { |
6512 | assert(ec); | |
1e22b5cd | 6513 | |
6c0ae739 | 6514 | return ec->tty_reset || |
1e22b5cd LP |
6515 | ec->tty_vhangup || |
6516 | ec->tty_vt_disallocate || | |
6ac8fdc9 MS |
6517 | is_terminal_input(ec->std_input) || |
6518 | is_terminal_output(ec->std_output) || | |
6c0ae739 LP |
6519 | is_terminal_output(ec->std_error); |
6520 | } | |
6521 | ||
6522 | bool exec_context_may_touch_console(const ExecContext *ec) { | |
6523 | ||
6524 | return exec_context_may_touch_tty(ec) && | |
1e22b5cd | 6525 | tty_may_match_dev_console(exec_context_tty_path(ec)); |
6ac8fdc9 MS |
6526 | } |
6527 | ||
15ae422b | 6528 | static void strv_fprintf(FILE *f, char **l) { |
15ae422b LP |
6529 | assert(f); |
6530 | ||
6531 | STRV_FOREACH(g, l) | |
6532 | fprintf(f, " %s", *g); | |
6533 | } | |
6534 | ||
ddc155b2 TM |
6535 | static void strv_dump(FILE* f, const char *prefix, const char *name, char **strv) { |
6536 | assert(f); | |
6537 | assert(prefix); | |
6538 | assert(name); | |
6539 | ||
6540 | if (!strv_isempty(strv)) { | |
a7bd1656 | 6541 | fprintf(f, "%s%s:", prefix, name); |
ddc155b2 TM |
6542 | strv_fprintf(f, strv); |
6543 | fputs("\n", f); | |
6544 | } | |
6545 | } | |
6546 | ||
34cf6c43 | 6547 | void exec_context_dump(const ExecContext *c, FILE* f, const char *prefix) { |
add00535 | 6548 | int r; |
9eba9da4 | 6549 | |
5cb5a6ff LP |
6550 | assert(c); |
6551 | assert(f); | |
6552 | ||
4ad49000 | 6553 | prefix = strempty(prefix); |
5cb5a6ff LP |
6554 | |
6555 | fprintf(f, | |
94f04347 LP |
6556 | "%sUMask: %04o\n" |
6557 | "%sWorkingDirectory: %s\n" | |
451a074f | 6558 | "%sRootDirectory: %s\n" |
9c0c6701 | 6559 | "%sRootEphemeral: %s\n" |
15ae422b | 6560 | "%sNonBlocking: %s\n" |
64747e2d | 6561 | "%sPrivateTmp: %s\n" |
7f112f50 | 6562 | "%sPrivateDevices: %s\n" |
59eeb84b | 6563 | "%sProtectKernelTunables: %s\n" |
e66a2f65 | 6564 | "%sProtectKernelModules: %s\n" |
84703040 | 6565 | "%sProtectKernelLogs: %s\n" |
fc64760d | 6566 | "%sProtectClock: %s\n" |
59eeb84b | 6567 | "%sProtectControlGroups: %s\n" |
d251207d LP |
6568 | "%sPrivateNetwork: %s\n" |
6569 | "%sPrivateUsers: %s\n" | |
1b8689f9 LP |
6570 | "%sProtectHome: %s\n" |
6571 | "%sProtectSystem: %s\n" | |
5d997827 | 6572 | "%sMountAPIVFS: %s\n" |
f3e43635 | 6573 | "%sIgnoreSIGPIPE: %s\n" |
f4170c67 | 6574 | "%sMemoryDenyWriteExecute: %s\n" |
b1edf445 | 6575 | "%sRestrictRealtime: %s\n" |
f69567cb | 6576 | "%sRestrictSUIDSGID: %s\n" |
aecd5ac6 | 6577 | "%sKeyringMode: %s\n" |
4e399953 LP |
6578 | "%sProtectHostname: %s\n" |
6579 | "%sProtectProc: %s\n" | |
6580 | "%sProcSubset: %s\n", | |
5cb5a6ff | 6581 | prefix, c->umask, |
14eb3285 LP |
6582 | prefix, empty_to_root(c->working_directory), |
6583 | prefix, empty_to_root(c->root_directory), | |
9c0c6701 | 6584 | prefix, yes_no(c->root_ephemeral), |
15ae422b | 6585 | prefix, yes_no(c->non_blocking), |
64747e2d | 6586 | prefix, yes_no(c->private_tmp), |
7f112f50 | 6587 | prefix, yes_no(c->private_devices), |
59eeb84b | 6588 | prefix, yes_no(c->protect_kernel_tunables), |
e66a2f65 | 6589 | prefix, yes_no(c->protect_kernel_modules), |
84703040 | 6590 | prefix, yes_no(c->protect_kernel_logs), |
fc64760d | 6591 | prefix, yes_no(c->protect_clock), |
59eeb84b | 6592 | prefix, yes_no(c->protect_control_groups), |
d251207d LP |
6593 | prefix, yes_no(c->private_network), |
6594 | prefix, yes_no(c->private_users), | |
1b8689f9 LP |
6595 | prefix, protect_home_to_string(c->protect_home), |
6596 | prefix, protect_system_to_string(c->protect_system), | |
5e98086d | 6597 | prefix, yes_no(exec_context_get_effective_mount_apivfs(c)), |
f3e43635 | 6598 | prefix, yes_no(c->ignore_sigpipe), |
f4170c67 | 6599 | prefix, yes_no(c->memory_deny_write_execute), |
b1edf445 | 6600 | prefix, yes_no(c->restrict_realtime), |
f69567cb | 6601 | prefix, yes_no(c->restrict_suid_sgid), |
aecd5ac6 | 6602 | prefix, exec_keyring_mode_to_string(c->keyring_mode), |
4e399953 LP |
6603 | prefix, yes_no(c->protect_hostname), |
6604 | prefix, protect_proc_to_string(c->protect_proc), | |
6605 | prefix, proc_subset_to_string(c->proc_subset)); | |
fb33a393 | 6606 | |
915e6d16 LP |
6607 | if (c->root_image) |
6608 | fprintf(f, "%sRootImage: %s\n", prefix, c->root_image); | |
6609 | ||
18d73705 | 6610 | if (c->root_image_options) { |
18d73705 LB |
6611 | fprintf(f, "%sRootImageOptions:", prefix); |
6612 | LIST_FOREACH(mount_options, o, c->root_image_options) | |
6613 | if (!isempty(o->options)) | |
9ece6444 LB |
6614 | fprintf(f, " %s:%s", |
6615 | partition_designator_to_string(o->partition_designator), | |
6616 | o->options); | |
18d73705 LB |
6617 | fprintf(f, "\n"); |
6618 | } | |
6619 | ||
0389f4fa LB |
6620 | if (c->root_hash) { |
6621 | _cleanup_free_ char *encoded = NULL; | |
6622 | encoded = hexmem(c->root_hash, c->root_hash_size); | |
6623 | if (encoded) | |
6624 | fprintf(f, "%sRootHash: %s\n", prefix, encoded); | |
6625 | } | |
6626 | ||
6627 | if (c->root_hash_path) | |
6628 | fprintf(f, "%sRootHash: %s\n", prefix, c->root_hash_path); | |
6629 | ||
d4d55b0d LB |
6630 | if (c->root_hash_sig) { |
6631 | _cleanup_free_ char *encoded = NULL; | |
6632 | ssize_t len; | |
6633 | len = base64mem(c->root_hash_sig, c->root_hash_sig_size, &encoded); | |
6634 | if (len) | |
6635 | fprintf(f, "%sRootHashSignature: base64:%s\n", prefix, encoded); | |
6636 | } | |
6637 | ||
6638 | if (c->root_hash_sig_path) | |
6639 | fprintf(f, "%sRootHashSignature: %s\n", prefix, c->root_hash_sig_path); | |
6640 | ||
0389f4fa LB |
6641 | if (c->root_verity) |
6642 | fprintf(f, "%sRootVerity: %s\n", prefix, c->root_verity); | |
6643 | ||
8c7be95e LP |
6644 | STRV_FOREACH(e, c->environment) |
6645 | fprintf(f, "%sEnvironment: %s\n", prefix, *e); | |
6646 | ||
6647 | STRV_FOREACH(e, c->environment_files) | |
6648 | fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e); | |
94f04347 | 6649 | |
b4c14404 FB |
6650 | STRV_FOREACH(e, c->pass_environment) |
6651 | fprintf(f, "%sPassEnvironment: %s\n", prefix, *e); | |
6652 | ||
00819cc1 LP |
6653 | STRV_FOREACH(e, c->unset_environment) |
6654 | fprintf(f, "%sUnsetEnvironment: %s\n", prefix, *e); | |
6655 | ||
53f47dfc YW |
6656 | fprintf(f, "%sRuntimeDirectoryPreserve: %s\n", prefix, exec_preserve_mode_to_string(c->runtime_directory_preserve_mode)); |
6657 | ||
5b10116e | 6658 | for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) { |
3536f49e YW |
6659 | fprintf(f, "%s%sMode: %04o\n", prefix, exec_directory_type_to_string(dt), c->directories[dt].mode); |
6660 | ||
211a3d87 LB |
6661 | for (size_t i = 0; i < c->directories[dt].n_items; i++) { |
6662 | fprintf(f, "%s%s: %s\n", prefix, exec_directory_type_to_string(dt), c->directories[dt].items[i].path); | |
6663 | ||
6664 | STRV_FOREACH(d, c->directories[dt].items[i].symlinks) | |
6665 | fprintf(f, "%s%s: %s:%s\n", prefix, exec_directory_type_symlink_to_string(dt), c->directories[dt].items[i].path, *d); | |
6666 | } | |
3536f49e | 6667 | } |
c2bbd90b | 6668 | |
5291f26d | 6669 | fprintf(f, "%sTimeoutCleanSec: %s\n", prefix, FORMAT_TIMESPAN(c->timeout_clean_usec, USEC_PER_SEC)); |
12213aed | 6670 | |
fb33a393 | 6671 | if (c->nice_set) |
5291f26d | 6672 | fprintf(f, "%sNice: %i\n", prefix, c->nice); |
fb33a393 | 6673 | |
dd6c17b1 | 6674 | if (c->oom_score_adjust_set) |
5291f26d | 6675 | fprintf(f, "%sOOMScoreAdjust: %i\n", prefix, c->oom_score_adjust); |
9eba9da4 | 6676 | |
ad21e542 | 6677 | if (c->coredump_filter_set) |
5291f26d | 6678 | fprintf(f, "%sCoredumpFilter: 0x%"PRIx64"\n", prefix, c->coredump_filter); |
ad21e542 | 6679 | |
5b10116e | 6680 | for (unsigned i = 0; i < RLIM_NLIMITS; i++) |
3c11da9d | 6681 | if (c->rlimit[i]) { |
4c3a2b84 | 6682 | fprintf(f, "%sLimit%s: " RLIM_FMT "\n", |
3c11da9d | 6683 | prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max); |
4c3a2b84 | 6684 | fprintf(f, "%sLimit%sSoft: " RLIM_FMT "\n", |
3c11da9d EV |
6685 | prefix, rlimit_to_string(i), c->rlimit[i]->rlim_cur); |
6686 | } | |
94f04347 | 6687 | |
f8b69d1d | 6688 | if (c->ioprio_set) { |
1756a011 | 6689 | _cleanup_free_ char *class_str = NULL; |
f8b69d1d | 6690 | |
5bead76e | 6691 | r = ioprio_class_to_string_alloc(ioprio_prio_class(c->ioprio), &class_str); |
837df140 YW |
6692 | if (r >= 0) |
6693 | fprintf(f, "%sIOSchedulingClass: %s\n", prefix, class_str); | |
6694 | ||
5bead76e | 6695 | fprintf(f, "%sIOPriority: %d\n", prefix, ioprio_prio_data(c->ioprio)); |
f8b69d1d | 6696 | } |
94f04347 | 6697 | |
f8b69d1d | 6698 | if (c->cpu_sched_set) { |
1756a011 | 6699 | _cleanup_free_ char *policy_str = NULL; |
f8b69d1d | 6700 | |
837df140 YW |
6701 | r = sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str); |
6702 | if (r >= 0) | |
6703 | fprintf(f, "%sCPUSchedulingPolicy: %s\n", prefix, policy_str); | |
6704 | ||
94f04347 | 6705 | fprintf(f, |
38b48754 LP |
6706 | "%sCPUSchedulingPriority: %i\n" |
6707 | "%sCPUSchedulingResetOnFork: %s\n", | |
38b48754 LP |
6708 | prefix, c->cpu_sched_priority, |
6709 | prefix, yes_no(c->cpu_sched_reset_on_fork)); | |
b929bf04 | 6710 | } |
94f04347 | 6711 | |
0985c7c4 | 6712 | if (c->cpu_set.set) { |
e7fca352 MS |
6713 | _cleanup_free_ char *affinity = NULL; |
6714 | ||
6715 | affinity = cpu_set_to_range_string(&c->cpu_set); | |
6716 | fprintf(f, "%sCPUAffinity: %s\n", prefix, affinity); | |
94f04347 LP |
6717 | } |
6718 | ||
b070c7c0 MS |
6719 | if (mpol_is_valid(numa_policy_get_type(&c->numa_policy))) { |
6720 | _cleanup_free_ char *nodes = NULL; | |
6721 | ||
6722 | nodes = cpu_set_to_range_string(&c->numa_policy.nodes); | |
6723 | fprintf(f, "%sNUMAPolicy: %s\n", prefix, mpol_to_string(numa_policy_get_type(&c->numa_policy))); | |
6724 | fprintf(f, "%sNUMAMask: %s\n", prefix, strnull(nodes)); | |
6725 | } | |
6726 | ||
3a43da28 | 6727 | if (c->timer_slack_nsec != NSEC_INFINITY) |
ccd06097 | 6728 | fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec); |
94f04347 LP |
6729 | |
6730 | fprintf(f, | |
80876c20 LP |
6731 | "%sStandardInput: %s\n" |
6732 | "%sStandardOutput: %s\n" | |
6733 | "%sStandardError: %s\n", | |
6734 | prefix, exec_input_to_string(c->std_input), | |
6735 | prefix, exec_output_to_string(c->std_output), | |
6736 | prefix, exec_output_to_string(c->std_error)); | |
6737 | ||
befc4a80 LP |
6738 | if (c->std_input == EXEC_INPUT_NAMED_FD) |
6739 | fprintf(f, "%sStandardInputFileDescriptorName: %s\n", prefix, c->stdio_fdname[STDIN_FILENO]); | |
6740 | if (c->std_output == EXEC_OUTPUT_NAMED_FD) | |
6741 | fprintf(f, "%sStandardOutputFileDescriptorName: %s\n", prefix, c->stdio_fdname[STDOUT_FILENO]); | |
6742 | if (c->std_error == EXEC_OUTPUT_NAMED_FD) | |
6743 | fprintf(f, "%sStandardErrorFileDescriptorName: %s\n", prefix, c->stdio_fdname[STDERR_FILENO]); | |
6744 | ||
6745 | if (c->std_input == EXEC_INPUT_FILE) | |
6746 | fprintf(f, "%sStandardInputFile: %s\n", prefix, c->stdio_file[STDIN_FILENO]); | |
6747 | if (c->std_output == EXEC_OUTPUT_FILE) | |
6748 | fprintf(f, "%sStandardOutputFile: %s\n", prefix, c->stdio_file[STDOUT_FILENO]); | |
566b7d23 ZD |
6749 | if (c->std_output == EXEC_OUTPUT_FILE_APPEND) |
6750 | fprintf(f, "%sStandardOutputFileToAppend: %s\n", prefix, c->stdio_file[STDOUT_FILENO]); | |
8d7dab1f LW |
6751 | if (c->std_output == EXEC_OUTPUT_FILE_TRUNCATE) |
6752 | fprintf(f, "%sStandardOutputFileToTruncate: %s\n", prefix, c->stdio_file[STDOUT_FILENO]); | |
befc4a80 LP |
6753 | if (c->std_error == EXEC_OUTPUT_FILE) |
6754 | fprintf(f, "%sStandardErrorFile: %s\n", prefix, c->stdio_file[STDERR_FILENO]); | |
566b7d23 ZD |
6755 | if (c->std_error == EXEC_OUTPUT_FILE_APPEND) |
6756 | fprintf(f, "%sStandardErrorFileToAppend: %s\n", prefix, c->stdio_file[STDERR_FILENO]); | |
8d7dab1f LW |
6757 | if (c->std_error == EXEC_OUTPUT_FILE_TRUNCATE) |
6758 | fprintf(f, "%sStandardErrorFileToTruncate: %s\n", prefix, c->stdio_file[STDERR_FILENO]); | |
befc4a80 | 6759 | |
80876c20 LP |
6760 | if (c->tty_path) |
6761 | fprintf(f, | |
6ea832a2 LP |
6762 | "%sTTYPath: %s\n" |
6763 | "%sTTYReset: %s\n" | |
6764 | "%sTTYVHangup: %s\n" | |
51462135 DDM |
6765 | "%sTTYVTDisallocate: %s\n" |
6766 | "%sTTYRows: %u\n" | |
6767 | "%sTTYColumns: %u\n", | |
6ea832a2 LP |
6768 | prefix, c->tty_path, |
6769 | prefix, yes_no(c->tty_reset), | |
6770 | prefix, yes_no(c->tty_vhangup), | |
51462135 DDM |
6771 | prefix, yes_no(c->tty_vt_disallocate), |
6772 | prefix, c->tty_rows, | |
6773 | prefix, c->tty_cols); | |
94f04347 | 6774 | |
9f6444eb | 6775 | if (IN_SET(c->std_output, |
9f6444eb LP |
6776 | EXEC_OUTPUT_KMSG, |
6777 | EXEC_OUTPUT_JOURNAL, | |
9f6444eb LP |
6778 | EXEC_OUTPUT_KMSG_AND_CONSOLE, |
6779 | EXEC_OUTPUT_JOURNAL_AND_CONSOLE) || | |
6780 | IN_SET(c->std_error, | |
9f6444eb LP |
6781 | EXEC_OUTPUT_KMSG, |
6782 | EXEC_OUTPUT_JOURNAL, | |
9f6444eb LP |
6783 | EXEC_OUTPUT_KMSG_AND_CONSOLE, |
6784 | EXEC_OUTPUT_JOURNAL_AND_CONSOLE)) { | |
f8b69d1d | 6785 | |
5ce70e5b | 6786 | _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL; |
f8b69d1d | 6787 | |
837df140 YW |
6788 | r = log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str); |
6789 | if (r >= 0) | |
6790 | fprintf(f, "%sSyslogFacility: %s\n", prefix, fac_str); | |
f8b69d1d | 6791 | |
837df140 YW |
6792 | r = log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str); |
6793 | if (r >= 0) | |
6794 | fprintf(f, "%sSyslogLevel: %s\n", prefix, lvl_str); | |
f8b69d1d | 6795 | } |
94f04347 | 6796 | |
d3070fbd LP |
6797 | if (c->log_level_max >= 0) { |
6798 | _cleanup_free_ char *t = NULL; | |
6799 | ||
6800 | (void) log_level_to_string_alloc(c->log_level_max, &t); | |
6801 | ||
6802 | fprintf(f, "%sLogLevelMax: %s\n", prefix, strna(t)); | |
6803 | } | |
6804 | ||
5291f26d | 6805 | if (c->log_ratelimit_interval_usec > 0) |
90fc172e AZ |
6806 | fprintf(f, |
6807 | "%sLogRateLimitIntervalSec: %s\n", | |
5291f26d | 6808 | prefix, FORMAT_TIMESPAN(c->log_ratelimit_interval_usec, USEC_PER_SEC)); |
90fc172e | 6809 | |
5ac1530e ZJS |
6810 | if (c->log_ratelimit_burst > 0) |
6811 | fprintf(f, "%sLogRateLimitBurst: %u\n", prefix, c->log_ratelimit_burst); | |
90fc172e | 6812 | |
523ea123 QD |
6813 | if (!set_isempty(c->log_filter_allowed_patterns) || !set_isempty(c->log_filter_denied_patterns)) { |
6814 | fprintf(f, "%sLogFilterPatterns:", prefix); | |
6815 | ||
6816 | char *pattern; | |
6817 | SET_FOREACH(pattern, c->log_filter_allowed_patterns) | |
6818 | fprintf(f, " %s", pattern); | |
6819 | SET_FOREACH(pattern, c->log_filter_denied_patterns) | |
6820 | fprintf(f, " ~%s", pattern); | |
6821 | fputc('\n', f); | |
6822 | } | |
6823 | ||
5b10116e ZJS |
6824 | for (size_t j = 0; j < c->n_log_extra_fields; j++) { |
6825 | fprintf(f, "%sLogExtraFields: ", prefix); | |
6826 | fwrite(c->log_extra_fields[j].iov_base, | |
6827 | 1, c->log_extra_fields[j].iov_len, | |
6828 | f); | |
6829 | fputc('\n', f); | |
d3070fbd LP |
6830 | } |
6831 | ||
91dd5f7c LP |
6832 | if (c->log_namespace) |
6833 | fprintf(f, "%sLogNamespace: %s\n", prefix, c->log_namespace); | |
6834 | ||
07d46372 YW |
6835 | if (c->secure_bits) { |
6836 | _cleanup_free_ char *str = NULL; | |
6837 | ||
6838 | r = secure_bits_to_string_alloc(c->secure_bits, &str); | |
6839 | if (r >= 0) | |
6840 | fprintf(f, "%sSecure Bits: %s\n", prefix, str); | |
6841 | } | |
94f04347 | 6842 | |
3fd5190b | 6843 | if (c->capability_bounding_set != CAP_MASK_UNSET) { |
dd1f5bd0 | 6844 | _cleanup_free_ char *str = NULL; |
94f04347 | 6845 | |
8142d735 | 6846 | r = capability_set_to_string(c->capability_bounding_set, &str); |
dd1f5bd0 YW |
6847 | if (r >= 0) |
6848 | fprintf(f, "%sCapabilityBoundingSet: %s\n", prefix, str); | |
755d4b67 IP |
6849 | } |
6850 | ||
6851 | if (c->capability_ambient_set != 0) { | |
dd1f5bd0 | 6852 | _cleanup_free_ char *str = NULL; |
755d4b67 | 6853 | |
8142d735 | 6854 | r = capability_set_to_string(c->capability_ambient_set, &str); |
dd1f5bd0 YW |
6855 | if (r >= 0) |
6856 | fprintf(f, "%sAmbientCapabilities: %s\n", prefix, str); | |
94f04347 LP |
6857 | } |
6858 | ||
6859 | if (c->user) | |
f2d3769a | 6860 | fprintf(f, "%sUser: %s\n", prefix, c->user); |
94f04347 | 6861 | if (c->group) |
f2d3769a | 6862 | fprintf(f, "%sGroup: %s\n", prefix, c->group); |
94f04347 | 6863 | |
29206d46 LP |
6864 | fprintf(f, "%sDynamicUser: %s\n", prefix, yes_no(c->dynamic_user)); |
6865 | ||
ddc155b2 | 6866 | strv_dump(f, prefix, "SupplementaryGroups", c->supplementary_groups); |
94f04347 | 6867 | |
5b6319dc | 6868 | if (c->pam_name) |
f2d3769a | 6869 | fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name); |
5b6319dc | 6870 | |
ddc155b2 TM |
6871 | strv_dump(f, prefix, "ReadWritePaths", c->read_write_paths); |
6872 | strv_dump(f, prefix, "ReadOnlyPaths", c->read_only_paths); | |
6873 | strv_dump(f, prefix, "InaccessiblePaths", c->inaccessible_paths); | |
6874 | strv_dump(f, prefix, "ExecPaths", c->exec_paths); | |
6875 | strv_dump(f, prefix, "NoExecPaths", c->no_exec_paths); | |
8c35c10d | 6876 | strv_dump(f, prefix, "ExecSearchPath", c->exec_search_path); |
2e22afe9 | 6877 | |
5b10116e ZJS |
6878 | for (size_t i = 0; i < c->n_bind_mounts; i++) |
6879 | fprintf(f, "%s%s: %s%s:%s:%s\n", prefix, | |
6880 | c->bind_mounts[i].read_only ? "BindReadOnlyPaths" : "BindPaths", | |
6881 | c->bind_mounts[i].ignore_enoent ? "-": "", | |
6882 | c->bind_mounts[i].source, | |
6883 | c->bind_mounts[i].destination, | |
6884 | c->bind_mounts[i].recursive ? "rbind" : "norbind"); | |
d2d6c096 | 6885 | |
5b10116e ZJS |
6886 | for (size_t i = 0; i < c->n_temporary_filesystems; i++) { |
6887 | const TemporaryFileSystem *t = c->temporary_filesystems + i; | |
2abd4e38 | 6888 | |
5b10116e ZJS |
6889 | fprintf(f, "%sTemporaryFileSystem: %s%s%s\n", prefix, |
6890 | t->path, | |
6891 | isempty(t->options) ? "" : ":", | |
6892 | strempty(t->options)); | |
6893 | } | |
2abd4e38 | 6894 | |
169c1bda LP |
6895 | if (c->utmp_id) |
6896 | fprintf(f, | |
6897 | "%sUtmpIdentifier: %s\n", | |
6898 | prefix, c->utmp_id); | |
7b52a628 MS |
6899 | |
6900 | if (c->selinux_context) | |
6901 | fprintf(f, | |
5f8640fb LP |
6902 | "%sSELinuxContext: %s%s\n", |
6903 | prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context); | |
17df7223 | 6904 | |
80c21aea WC |
6905 | if (c->apparmor_profile) |
6906 | fprintf(f, | |
6907 | "%sAppArmorProfile: %s%s\n", | |
6908 | prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile); | |
6909 | ||
6910 | if (c->smack_process_label) | |
6911 | fprintf(f, | |
6912 | "%sSmackProcessLabel: %s%s\n", | |
6913 | prefix, c->smack_process_label_ignore ? "-" : "", c->smack_process_label); | |
6914 | ||
050f7277 | 6915 | if (c->personality != PERSONALITY_INVALID) |
ac45f971 LP |
6916 | fprintf(f, |
6917 | "%sPersonality: %s\n", | |
6918 | prefix, strna(personality_to_string(c->personality))); | |
6919 | ||
78e864e5 TM |
6920 | fprintf(f, |
6921 | "%sLockPersonality: %s\n", | |
6922 | prefix, yes_no(c->lock_personality)); | |
6923 | ||
17df7223 | 6924 | if (c->syscall_filter) { |
17df7223 | 6925 | fprintf(f, |
57183d11 | 6926 | "%sSystemCallFilter: ", |
17df7223 LP |
6927 | prefix); |
6928 | ||
6b000af4 | 6929 | if (!c->syscall_allow_list) |
17df7223 LP |
6930 | fputc('~', f); |
6931 | ||
349cc4a5 | 6932 | #if HAVE_SECCOMP |
d5a99b7c JJ |
6933 | void *id, *val; |
6934 | bool first = true; | |
90e74a66 | 6935 | HASHMAP_FOREACH_KEY(val, id, c->syscall_filter) { |
17df7223 | 6936 | _cleanup_free_ char *name = NULL; |
8cfa775f YW |
6937 | const char *errno_name = NULL; |
6938 | int num = PTR_TO_INT(val); | |
17df7223 LP |
6939 | |
6940 | if (first) | |
6941 | first = false; | |
6942 | else | |
6943 | fputc(' ', f); | |
6944 | ||
57183d11 | 6945 | name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1); |
17df7223 | 6946 | fputs(strna(name), f); |
8cfa775f YW |
6947 | |
6948 | if (num >= 0) { | |
005bfaf1 | 6949 | errno_name = seccomp_errno_or_action_to_string(num); |
8cfa775f YW |
6950 | if (errno_name) |
6951 | fprintf(f, ":%s", errno_name); | |
6952 | else | |
6953 | fprintf(f, ":%d", num); | |
6954 | } | |
17df7223 | 6955 | } |
351a19b1 | 6956 | #endif |
17df7223 LP |
6957 | |
6958 | fputc('\n', f); | |
6959 | } | |
6960 | ||
57183d11 | 6961 | if (c->syscall_archs) { |
57183d11 LP |
6962 | fprintf(f, |
6963 | "%sSystemCallArchitectures:", | |
6964 | prefix); | |
6965 | ||
349cc4a5 | 6966 | #if HAVE_SECCOMP |
d5a99b7c | 6967 | void *id; |
90e74a66 | 6968 | SET_FOREACH(id, c->syscall_archs) |
57183d11 LP |
6969 | fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1))); |
6970 | #endif | |
6971 | fputc('\n', f); | |
6972 | } | |
6973 | ||
add00535 LP |
6974 | if (exec_context_restrict_namespaces_set(c)) { |
6975 | _cleanup_free_ char *s = NULL; | |
6976 | ||
86c2a9f1 | 6977 | r = namespace_flags_to_string(c->restrict_namespaces, &s); |
add00535 LP |
6978 | if (r >= 0) |
6979 | fprintf(f, "%sRestrictNamespaces: %s\n", | |
dd0395b5 | 6980 | prefix, strna(s)); |
add00535 LP |
6981 | } |
6982 | ||
b1994387 | 6983 | #if HAVE_LIBBPF |
8fe84dc8 YW |
6984 | if (exec_context_restrict_filesystems_set(c)) { |
6985 | char *fs; | |
6986 | SET_FOREACH(fs, c->restrict_filesystems) | |
6987 | fprintf(f, "%sRestrictFileSystems: %s\n", prefix, fs); | |
6988 | } | |
b1994387 ILG |
6989 | #endif |
6990 | ||
a8d08f39 LP |
6991 | if (c->network_namespace_path) |
6992 | fprintf(f, | |
6993 | "%sNetworkNamespacePath: %s\n", | |
6994 | prefix, c->network_namespace_path); | |
6995 | ||
3df90f24 | 6996 | if (c->syscall_errno > 0) { |
3df90f24 YW |
6997 | fprintf(f, "%sSystemCallErrorNumber: ", prefix); |
6998 | ||
005bfaf1 | 6999 | #if HAVE_SECCOMP |
d5a99b7c | 7000 | const char *errno_name = seccomp_errno_or_action_to_string(c->syscall_errno); |
3df90f24 | 7001 | if (errno_name) |
005bfaf1 | 7002 | fputs(errno_name, f); |
3df90f24 | 7003 | else |
005bfaf1 TM |
7004 | fprintf(f, "%d", c->syscall_errno); |
7005 | #endif | |
7006 | fputc('\n', f); | |
3df90f24 | 7007 | } |
b3d13314 | 7008 | |
5b10116e | 7009 | for (size_t i = 0; i < c->n_mount_images; i++) { |
79e20ceb | 7010 | fprintf(f, "%sMountImages: %s%s:%s", prefix, |
b3d13314 LB |
7011 | c->mount_images[i].ignore_enoent ? "-": "", |
7012 | c->mount_images[i].source, | |
79e20ceb | 7013 | c->mount_images[i].destination); |
427353f6 | 7014 | LIST_FOREACH(mount_options, o, c->mount_images[i].mount_options) |
79e20ceb | 7015 | fprintf(f, ":%s:%s", |
427353f6 | 7016 | partition_designator_to_string(o->partition_designator), |
79e20ceb | 7017 | strempty(o->options)); |
427353f6 LB |
7018 | fprintf(f, "\n"); |
7019 | } | |
93f59701 LB |
7020 | |
7021 | for (size_t i = 0; i < c->n_extension_images; i++) { | |
93f59701 LB |
7022 | fprintf(f, "%sExtensionImages: %s%s", prefix, |
7023 | c->extension_images[i].ignore_enoent ? "-": "", | |
7024 | c->extension_images[i].source); | |
7025 | LIST_FOREACH(mount_options, o, c->extension_images[i].mount_options) | |
7026 | fprintf(f, ":%s:%s", | |
7027 | partition_designator_to_string(o->partition_designator), | |
7028 | strempty(o->options)); | |
7029 | fprintf(f, "\n"); | |
7030 | } | |
a07b9926 LB |
7031 | |
7032 | strv_dump(f, prefix, "ExtensionDirectories", c->extension_directories); | |
5cb5a6ff LP |
7033 | } |
7034 | ||
34cf6c43 | 7035 | bool exec_context_maintains_privileges(const ExecContext *c) { |
a931ad47 LP |
7036 | assert(c); |
7037 | ||
61233823 | 7038 | /* Returns true if the process forked off would run under |
a931ad47 LP |
7039 | * an unchanged UID or as root. */ |
7040 | ||
7041 | if (!c->user) | |
7042 | return true; | |
7043 | ||
7044 | if (streq(c->user, "root") || streq(c->user, "0")) | |
7045 | return true; | |
7046 | ||
7047 | return false; | |
7048 | } | |
7049 | ||
34cf6c43 | 7050 | int exec_context_get_effective_ioprio(const ExecContext *c) { |
7f452159 LP |
7051 | int p; |
7052 | ||
7053 | assert(c); | |
7054 | ||
7055 | if (c->ioprio_set) | |
7056 | return c->ioprio; | |
7057 | ||
7058 | p = ioprio_get(IOPRIO_WHO_PROCESS, 0); | |
7059 | if (p < 0) | |
0692548c | 7060 | return IOPRIO_DEFAULT_CLASS_AND_PRIO; |
7f452159 | 7061 | |
8b330d7d | 7062 | return ioprio_normalize(p); |
7f452159 LP |
7063 | } |
7064 | ||
5e98086d ZJS |
7065 | bool exec_context_get_effective_mount_apivfs(const ExecContext *c) { |
7066 | assert(c); | |
7067 | ||
61198784 | 7068 | /* Explicit setting wins */ |
5e98086d ZJS |
7069 | if (c->mount_apivfs_set) |
7070 | return c->mount_apivfs; | |
7071 | ||
61198784 | 7072 | /* Default to "yes" if root directory or image are specified */ |
74e12520 | 7073 | if (exec_context_with_rootfs(c)) |
61198784 ZJS |
7074 | return true; |
7075 | ||
5e98086d ZJS |
7076 | return false; |
7077 | } | |
7078 | ||
d3070fbd | 7079 | void exec_context_free_log_extra_fields(ExecContext *c) { |
d3070fbd LP |
7080 | assert(c); |
7081 | ||
5b10116e | 7082 | for (size_t l = 0; l < c->n_log_extra_fields; l++) |
d3070fbd LP |
7083 | free(c->log_extra_fields[l].iov_base); |
7084 | c->log_extra_fields = mfree(c->log_extra_fields); | |
7085 | c->n_log_extra_fields = 0; | |
7086 | } | |
7087 | ||
6f765baf | 7088 | void exec_context_revert_tty(ExecContext *c) { |
254d1313 | 7089 | _cleanup_close_ int fd = -EBADF; |
0ba976e8 LP |
7090 | const char *path; |
7091 | struct stat st; | |
6f765baf LP |
7092 | int r; |
7093 | ||
7094 | assert(c); | |
7095 | ||
7096 | /* First, reset the TTY (possibly kicking everybody else from the TTY) */ | |
7097 | exec_context_tty_reset(c, NULL); | |
7098 | ||
7099 | /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path | |
7100 | * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed | |
7101 | * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */ | |
0ba976e8 LP |
7102 | if (!exec_context_may_touch_tty(c)) |
7103 | return; | |
6f765baf | 7104 | |
0ba976e8 LP |
7105 | path = exec_context_tty_path(c); |
7106 | if (!path) | |
7107 | return; | |
6f765baf | 7108 | |
0ba976e8 LP |
7109 | fd = open(path, O_PATH|O_CLOEXEC); |
7110 | if (fd < 0) | |
7111 | return (void) log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, | |
7112 | "Failed to open TTY inode of '%s' to adjust ownership/access mode, ignoring: %m", | |
7113 | path); | |
7114 | ||
7115 | if (fstat(fd, &st) < 0) | |
7116 | return (void) log_warning_errno(errno, "Failed to stat TTY '%s', ignoring: %m", path); | |
7117 | ||
7118 | /* Let's add a superficial check that we only do this for stuff that looks like a TTY. We only check | |
7119 | * if things are a character device, since a proper check either means we'd have to open the TTY and | |
7120 | * use isatty(), but we'd rather not do that since opening TTYs comes with all kinds of side-effects | |
7121 | * and is slow. Or we'd have to hardcode dev_t major information, which we'd rather avoid. Why bother | |
7122 | * with this at all? → https://github.com/systemd/systemd/issues/19213 */ | |
7123 | if (!S_ISCHR(st.st_mode)) | |
7124 | return log_warning("Configured TTY '%s' is not actually a character device, ignoring.", path); | |
7125 | ||
7126 | r = fchmod_and_chown(fd, TTY_MODE, 0, TTY_GID); | |
7127 | if (r < 0) | |
7128 | log_warning_errno(r, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path); | |
6f765baf LP |
7129 | } |
7130 | ||
4c2f5842 LP |
7131 | int exec_context_get_clean_directories( |
7132 | ExecContext *c, | |
7133 | char **prefix, | |
7134 | ExecCleanMask mask, | |
7135 | char ***ret) { | |
7136 | ||
7137 | _cleanup_strv_free_ char **l = NULL; | |
4c2f5842 LP |
7138 | int r; |
7139 | ||
7140 | assert(c); | |
7141 | assert(prefix); | |
7142 | assert(ret); | |
7143 | ||
5b10116e | 7144 | for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) { |
4c2f5842 LP |
7145 | if (!FLAGS_SET(mask, 1U << t)) |
7146 | continue; | |
7147 | ||
7148 | if (!prefix[t]) | |
7149 | continue; | |
7150 | ||
211a3d87 | 7151 | for (size_t i = 0; i < c->directories[t].n_items; i++) { |
4c2f5842 LP |
7152 | char *j; |
7153 | ||
211a3d87 | 7154 | j = path_join(prefix[t], c->directories[t].items[i].path); |
4c2f5842 LP |
7155 | if (!j) |
7156 | return -ENOMEM; | |
7157 | ||
7158 | r = strv_consume(&l, j); | |
7159 | if (r < 0) | |
7160 | return r; | |
7f622a19 YW |
7161 | |
7162 | /* Also remove private directories unconditionally. */ | |
7163 | if (t != EXEC_DIRECTORY_CONFIGURATION) { | |
211a3d87 LB |
7164 | j = path_join(prefix[t], "private", c->directories[t].items[i].path); |
7165 | if (!j) | |
7166 | return -ENOMEM; | |
7167 | ||
7168 | r = strv_consume(&l, j); | |
7169 | if (r < 0) | |
7170 | return r; | |
7171 | } | |
7172 | ||
211a3d87 LB |
7173 | STRV_FOREACH(symlink, c->directories[t].items[i].symlinks) { |
7174 | j = path_join(prefix[t], *symlink); | |
7f622a19 YW |
7175 | if (!j) |
7176 | return -ENOMEM; | |
7177 | ||
7178 | r = strv_consume(&l, j); | |
7179 | if (r < 0) | |
7180 | return r; | |
7181 | } | |
4c2f5842 LP |
7182 | } |
7183 | } | |
7184 | ||
7185 | *ret = TAKE_PTR(l); | |
7186 | return 0; | |
7187 | } | |
7188 | ||
7189 | int exec_context_get_clean_mask(ExecContext *c, ExecCleanMask *ret) { | |
7190 | ExecCleanMask mask = 0; | |
7191 | ||
7192 | assert(c); | |
7193 | assert(ret); | |
7194 | ||
7195 | for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) | |
211a3d87 | 7196 | if (c->directories[t].n_items > 0) |
4c2f5842 LP |
7197 | mask |= 1U << t; |
7198 | ||
7199 | *ret = mask; | |
7200 | return 0; | |
7201 | } | |
7202 | ||
50a4217b LP |
7203 | bool exec_context_has_encrypted_credentials(ExecContext *c) { |
7204 | ExecLoadCredential *load_cred; | |
7205 | ExecSetCredential *set_cred; | |
7206 | ||
7207 | assert(c); | |
7208 | ||
7209 | HASHMAP_FOREACH(load_cred, c->load_credentials) | |
7210 | if (load_cred->encrypted) | |
7211 | return true; | |
7212 | ||
7213 | HASHMAP_FOREACH(set_cred, c->set_credentials) | |
7214 | if (set_cred->encrypted) | |
7215 | return true; | |
7216 | ||
7217 | return false; | |
7218 | } | |
7219 | ||
b58b4116 | 7220 | void exec_status_start(ExecStatus *s, pid_t pid) { |
034c6ed7 | 7221 | assert(s); |
5cb5a6ff | 7222 | |
2ed26ed0 LP |
7223 | *s = (ExecStatus) { |
7224 | .pid = pid, | |
7225 | }; | |
7226 | ||
b58b4116 LP |
7227 | dual_timestamp_get(&s->start_timestamp); |
7228 | } | |
7229 | ||
34cf6c43 | 7230 | void exec_status_exit(ExecStatus *s, const ExecContext *context, pid_t pid, int code, int status) { |
b58b4116 LP |
7231 | assert(s); |
7232 | ||
d46b79bb | 7233 | if (s->pid != pid) |
2ed26ed0 LP |
7234 | *s = (ExecStatus) { |
7235 | .pid = pid, | |
7236 | }; | |
b58b4116 | 7237 | |
63983207 | 7238 | dual_timestamp_get(&s->exit_timestamp); |
9fb86720 | 7239 | |
034c6ed7 LP |
7240 | s->code = code; |
7241 | s->status = status; | |
169c1bda | 7242 | |
6f765baf LP |
7243 | if (context && context->utmp_id) |
7244 | (void) utmp_put_dead_process(context->utmp_id, pid, code, status); | |
9fb86720 LP |
7245 | } |
7246 | ||
6a1d4d9f LP |
7247 | void exec_status_reset(ExecStatus *s) { |
7248 | assert(s); | |
7249 | ||
7250 | *s = (ExecStatus) {}; | |
7251 | } | |
7252 | ||
34cf6c43 | 7253 | void exec_status_dump(const ExecStatus *s, FILE *f, const char *prefix) { |
9fb86720 LP |
7254 | assert(s); |
7255 | assert(f); | |
7256 | ||
9fb86720 LP |
7257 | if (s->pid <= 0) |
7258 | return; | |
7259 | ||
4c940960 LP |
7260 | prefix = strempty(prefix); |
7261 | ||
9fb86720 | 7262 | fprintf(f, |
ccd06097 ZJS |
7263 | "%sPID: "PID_FMT"\n", |
7264 | prefix, s->pid); | |
9fb86720 | 7265 | |
af9d16e1 | 7266 | if (dual_timestamp_is_set(&s->start_timestamp)) |
9fb86720 LP |
7267 | fprintf(f, |
7268 | "%sStart Timestamp: %s\n", | |
04f5c018 | 7269 | prefix, FORMAT_TIMESTAMP(s->start_timestamp.realtime)); |
9fb86720 | 7270 | |
af9d16e1 | 7271 | if (dual_timestamp_is_set(&s->exit_timestamp)) |
9fb86720 LP |
7272 | fprintf(f, |
7273 | "%sExit Timestamp: %s\n" | |
7274 | "%sExit Code: %s\n" | |
7275 | "%sExit Status: %i\n", | |
04f5c018 | 7276 | prefix, FORMAT_TIMESTAMP(s->exit_timestamp.realtime), |
9fb86720 LP |
7277 | prefix, sigchld_code_to_string(s->code), |
7278 | prefix, s->status); | |
5cb5a6ff | 7279 | } |
44d8db9e | 7280 | |
34cf6c43 | 7281 | static void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) { |
e1d75803 | 7282 | _cleanup_free_ char *cmd = NULL; |
4c940960 | 7283 | const char *prefix2; |
44d8db9e LP |
7284 | |
7285 | assert(c); | |
7286 | assert(f); | |
7287 | ||
4c940960 | 7288 | prefix = strempty(prefix); |
63c372cb | 7289 | prefix2 = strjoina(prefix, "\t"); |
44d8db9e | 7290 | |
4ef15008 | 7291 | cmd = quote_command_line(c->argv, SHELL_ESCAPE_EMPTY); |
38553034 | 7292 | |
44d8db9e LP |
7293 | fprintf(f, |
7294 | "%sCommand Line: %s\n", | |
38553034 | 7295 | prefix, strnull(cmd)); |
44d8db9e | 7296 | |
9fb86720 | 7297 | exec_status_dump(&c->exec_status, f, prefix2); |
44d8db9e LP |
7298 | } |
7299 | ||
7300 | void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) { | |
7301 | assert(f); | |
7302 | ||
4c940960 | 7303 | prefix = strempty(prefix); |
44d8db9e | 7304 | |
03677889 YW |
7305 | LIST_FOREACH(command, i, c) |
7306 | exec_command_dump(i, f, prefix); | |
44d8db9e | 7307 | } |
94f04347 | 7308 | |
a6a80b4f LP |
7309 | void exec_command_append_list(ExecCommand **l, ExecCommand *e) { |
7310 | ExecCommand *end; | |
7311 | ||
7312 | assert(l); | |
7313 | assert(e); | |
7314 | ||
7315 | if (*l) { | |
35b8ca3a | 7316 | /* It's kind of important, that we keep the order here */ |
cc232fa0 | 7317 | end = LIST_FIND_TAIL(command, *l); |
71fda00f | 7318 | LIST_INSERT_AFTER(command, *l, end, e); |
a6a80b4f | 7319 | } else |
3ff67ec4 | 7320 | *l = e; |
a6a80b4f LP |
7321 | } |
7322 | ||
26fd040d LP |
7323 | int exec_command_set(ExecCommand *c, const char *path, ...) { |
7324 | va_list ap; | |
7325 | char **l, *p; | |
7326 | ||
7327 | assert(c); | |
7328 | assert(path); | |
7329 | ||
7330 | va_start(ap, path); | |
7331 | l = strv_new_ap(path, ap); | |
7332 | va_end(ap); | |
7333 | ||
7334 | if (!l) | |
7335 | return -ENOMEM; | |
7336 | ||
250a918d LP |
7337 | p = strdup(path); |
7338 | if (!p) { | |
26fd040d LP |
7339 | strv_free(l); |
7340 | return -ENOMEM; | |
7341 | } | |
7342 | ||
6897dfe8 | 7343 | free_and_replace(c->path, p); |
26fd040d | 7344 | |
130d3d22 | 7345 | return strv_free_and_replace(c->argv, l); |
26fd040d LP |
7346 | } |
7347 | ||
86b23b07 | 7348 | int exec_command_append(ExecCommand *c, const char *path, ...) { |
e63ff941 | 7349 | _cleanup_strv_free_ char **l = NULL; |
86b23b07 | 7350 | va_list ap; |
86b23b07 JS |
7351 | int r; |
7352 | ||
7353 | assert(c); | |
7354 | assert(path); | |
7355 | ||
7356 | va_start(ap, path); | |
7357 | l = strv_new_ap(path, ap); | |
7358 | va_end(ap); | |
7359 | ||
7360 | if (!l) | |
7361 | return -ENOMEM; | |
7362 | ||
e287086b | 7363 | r = strv_extend_strv(&c->argv, l, false); |
e63ff941 | 7364 | if (r < 0) |
86b23b07 | 7365 | return r; |
86b23b07 JS |
7366 | |
7367 | return 0; | |
7368 | } | |
7369 | ||
437f3e35 LP |
7370 | static char *destroy_tree(char *path) { |
7371 | if (!path) | |
7372 | return NULL; | |
9c0c6701 | 7373 | |
437f3e35 LP |
7374 | if (!path_equal(path, RUN_SYSTEMD_EMPTY)) { |
7375 | log_debug("Spawning process to nuke '%s'", path); | |
9c0c6701 | 7376 | |
437f3e35 LP |
7377 | (void) asynchronous_rm_rf(path, REMOVE_ROOT|REMOVE_SUBVOLUME|REMOVE_PHYSICAL); |
7378 | } | |
9c0c6701 | 7379 | |
437f3e35 | 7380 | return mfree(path); |
9c0c6701 DDM |
7381 | } |
7382 | ||
e52a696a | 7383 | static ExecSharedRuntime* exec_shared_runtime_free(ExecSharedRuntime *rt) { |
e8a565cb YW |
7384 | if (!rt) |
7385 | return NULL; | |
7386 | ||
7387 | if (rt->manager) | |
e76506b7 | 7388 | (void) hashmap_remove(rt->manager->exec_shared_runtime_by_id, rt->id); |
e8a565cb | 7389 | |
e52a696a DDM |
7390 | rt->id = mfree(rt->id); |
7391 | rt->tmp_dir = mfree(rt->tmp_dir); | |
7392 | rt->var_tmp_dir = mfree(rt->var_tmp_dir); | |
7393 | safe_close_pair(rt->netns_storage_socket); | |
7394 | safe_close_pair(rt->ipcns_storage_socket); | |
7395 | return mfree(rt); | |
7396 | } | |
7397 | ||
7398 | DEFINE_TRIVIAL_UNREF_FUNC(ExecSharedRuntime, exec_shared_runtime, exec_shared_runtime_free); | |
7399 | DEFINE_TRIVIAL_CLEANUP_FUNC(ExecSharedRuntime*, exec_shared_runtime_free); | |
7400 | ||
7401 | ExecSharedRuntime* exec_shared_runtime_destroy(ExecSharedRuntime *rt) { | |
e52a696a DDM |
7402 | if (!rt) |
7403 | return NULL; | |
7404 | ||
7405 | assert(rt->n_ref > 0); | |
7406 | rt->n_ref--; | |
7407 | ||
7408 | if (rt->n_ref > 0) | |
7409 | return NULL; | |
56a13a49 | 7410 | |
437f3e35 LP |
7411 | rt->tmp_dir = destroy_tree(rt->tmp_dir); |
7412 | rt->var_tmp_dir = destroy_tree(rt->var_tmp_dir); | |
e8a565cb | 7413 | |
e52a696a | 7414 | return exec_shared_runtime_free(rt); |
e8a565cb YW |
7415 | } |
7416 | ||
e76506b7 | 7417 | static int exec_shared_runtime_allocate(ExecSharedRuntime **ret, const char *id) { |
56a13a49 | 7418 | _cleanup_free_ char *id_copy = NULL; |
e76506b7 | 7419 | ExecSharedRuntime *n; |
613b411c | 7420 | |
8e8009dc | 7421 | assert(ret); |
613b411c | 7422 | |
56a13a49 ZJS |
7423 | id_copy = strdup(id); |
7424 | if (!id_copy) | |
7425 | return -ENOMEM; | |
7426 | ||
e76506b7 | 7427 | n = new(ExecSharedRuntime, 1); |
8e8009dc | 7428 | if (!n) |
613b411c LP |
7429 | return -ENOMEM; |
7430 | ||
e76506b7 | 7431 | *n = (ExecSharedRuntime) { |
56a13a49 | 7432 | .id = TAKE_PTR(id_copy), |
19ee48a6 YW |
7433 | .netns_storage_socket = PIPE_EBADF, |
7434 | .ipcns_storage_socket = PIPE_EBADF, | |
8e8009dc LP |
7435 | }; |
7436 | ||
7437 | *ret = n; | |
613b411c LP |
7438 | return 0; |
7439 | } | |
7440 | ||
e76506b7 | 7441 | static int exec_shared_runtime_add( |
e8a565cb YW |
7442 | Manager *m, |
7443 | const char *id, | |
56a13a49 ZJS |
7444 | char **tmp_dir, |
7445 | char **var_tmp_dir, | |
7446 | int netns_storage_socket[2], | |
a70581ff | 7447 | int ipcns_storage_socket[2], |
e76506b7 | 7448 | ExecSharedRuntime **ret) { |
e8a565cb | 7449 | |
e76506b7 | 7450 | _cleanup_(exec_shared_runtime_freep) ExecSharedRuntime *rt = NULL; |
613b411c LP |
7451 | int r; |
7452 | ||
e8a565cb | 7453 | assert(m); |
613b411c LP |
7454 | assert(id); |
7455 | ||
a70581ff | 7456 | /* tmp_dir, var_tmp_dir, {net,ipc}ns_storage_socket fds are donated on success */ |
56a13a49 | 7457 | |
e76506b7 | 7458 | r = exec_shared_runtime_allocate(&rt, id); |
613b411c LP |
7459 | if (r < 0) |
7460 | return r; | |
7461 | ||
e76506b7 | 7462 | r = hashmap_ensure_put(&m->exec_shared_runtime_by_id, &string_hash_ops, rt->id, rt); |
56a13a49 ZJS |
7463 | if (r < 0) |
7464 | return r; | |
e8a565cb | 7465 | |
56a13a49 ZJS |
7466 | assert(!!rt->tmp_dir == !!rt->var_tmp_dir); /* We require both to be set together */ |
7467 | rt->tmp_dir = TAKE_PTR(*tmp_dir); | |
7468 | rt->var_tmp_dir = TAKE_PTR(*var_tmp_dir); | |
e8a565cb YW |
7469 | |
7470 | if (netns_storage_socket) { | |
56a13a49 ZJS |
7471 | rt->netns_storage_socket[0] = TAKE_FD(netns_storage_socket[0]); |
7472 | rt->netns_storage_socket[1] = TAKE_FD(netns_storage_socket[1]); | |
613b411c LP |
7473 | } |
7474 | ||
a70581ff XR |
7475 | if (ipcns_storage_socket) { |
7476 | rt->ipcns_storage_socket[0] = TAKE_FD(ipcns_storage_socket[0]); | |
7477 | rt->ipcns_storage_socket[1] = TAKE_FD(ipcns_storage_socket[1]); | |
7478 | } | |
7479 | ||
e8a565cb YW |
7480 | rt->manager = m; |
7481 | ||
7482 | if (ret) | |
7483 | *ret = rt; | |
e76506b7 | 7484 | /* do not remove created ExecSharedRuntime object when the operation succeeds. */ |
56a13a49 | 7485 | TAKE_PTR(rt); |
e8a565cb YW |
7486 | return 0; |
7487 | } | |
7488 | ||
e76506b7 | 7489 | static int exec_shared_runtime_make( |
74aaf59b LP |
7490 | Manager *m, |
7491 | const ExecContext *c, | |
7492 | const char *id, | |
e76506b7 | 7493 | ExecSharedRuntime **ret) { |
74aaf59b | 7494 | |
56a13a49 | 7495 | _cleanup_(namespace_cleanup_tmpdirp) char *tmp_dir = NULL, *var_tmp_dir = NULL; |
19ee48a6 | 7496 | _cleanup_close_pair_ int netns_storage_socket[2] = PIPE_EBADF, ipcns_storage_socket[2] = PIPE_EBADF; |
e8a565cb YW |
7497 | int r; |
7498 | ||
7499 | assert(m); | |
7500 | assert(c); | |
7501 | assert(id); | |
7502 | ||
e76506b7 | 7503 | /* It is not necessary to create ExecSharedRuntime object. */ |
fde36d25 | 7504 | if (!exec_needs_network_namespace(c) && !exec_needs_ipc_namespace(c) && !c->private_tmp) { |
74aaf59b | 7505 | *ret = NULL; |
e8a565cb | 7506 | return 0; |
74aaf59b | 7507 | } |
e8a565cb | 7508 | |
efa2f3a1 TM |
7509 | if (c->private_tmp && |
7510 | !(prefixed_path_strv_contains(c->inaccessible_paths, "/tmp") && | |
7511 | (prefixed_path_strv_contains(c->inaccessible_paths, "/var/tmp") || | |
7512 | prefixed_path_strv_contains(c->inaccessible_paths, "/var")))) { | |
e8a565cb | 7513 | r = setup_tmp_dirs(id, &tmp_dir, &var_tmp_dir); |
613b411c LP |
7514 | if (r < 0) |
7515 | return r; | |
7516 | } | |
7517 | ||
fbbb9697 | 7518 | if (exec_needs_network_namespace(c)) { |
e8a565cb YW |
7519 | if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, netns_storage_socket) < 0) |
7520 | return -errno; | |
7521 | } | |
7522 | ||
fde36d25 | 7523 | if (exec_needs_ipc_namespace(c)) { |
a70581ff XR |
7524 | if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, ipcns_storage_socket) < 0) |
7525 | return -errno; | |
7526 | } | |
7527 | ||
e76506b7 | 7528 | r = exec_shared_runtime_add(m, id, &tmp_dir, &var_tmp_dir, netns_storage_socket, ipcns_storage_socket, ret); |
e8a565cb YW |
7529 | if (r < 0) |
7530 | return r; | |
7531 | ||
613b411c LP |
7532 | return 1; |
7533 | } | |
7534 | ||
e76506b7 DDM |
7535 | int exec_shared_runtime_acquire(Manager *m, const ExecContext *c, const char *id, bool create, ExecSharedRuntime **ret) { |
7536 | ExecSharedRuntime *rt; | |
e8a565cb | 7537 | int r; |
613b411c | 7538 | |
e8a565cb YW |
7539 | assert(m); |
7540 | assert(id); | |
7541 | assert(ret); | |
7542 | ||
e76506b7 | 7543 | rt = hashmap_get(m->exec_shared_runtime_by_id, id); |
e8a565cb | 7544 | if (rt) |
e76506b7 | 7545 | /* We already have an ExecSharedRuntime object, let's increase the ref count and reuse it */ |
e8a565cb YW |
7546 | goto ref; |
7547 | ||
74aaf59b LP |
7548 | if (!create) { |
7549 | *ret = NULL; | |
e8a565cb | 7550 | return 0; |
74aaf59b | 7551 | } |
e8a565cb YW |
7552 | |
7553 | /* If not found, then create a new object. */ | |
e76506b7 | 7554 | r = exec_shared_runtime_make(m, c, id, &rt); |
74aaf59b | 7555 | if (r < 0) |
e8a565cb | 7556 | return r; |
74aaf59b | 7557 | if (r == 0) { |
e76506b7 | 7558 | /* When r == 0, it is not necessary to create ExecSharedRuntime object. */ |
74aaf59b LP |
7559 | *ret = NULL; |
7560 | return 0; | |
7561 | } | |
613b411c | 7562 | |
e8a565cb YW |
7563 | ref: |
7564 | /* increment reference counter. */ | |
7565 | rt->n_ref++; | |
7566 | *ret = rt; | |
7567 | return 1; | |
7568 | } | |
613b411c | 7569 | |
e76506b7 DDM |
7570 | int exec_shared_runtime_serialize(const Manager *m, FILE *f, FDSet *fds) { |
7571 | ExecSharedRuntime *rt; | |
e8a565cb YW |
7572 | |
7573 | assert(m); | |
613b411c LP |
7574 | assert(f); |
7575 | assert(fds); | |
7576 | ||
e76506b7 | 7577 | HASHMAP_FOREACH(rt, m->exec_shared_runtime_by_id) { |
e8a565cb | 7578 | fprintf(f, "exec-runtime=%s", rt->id); |
613b411c | 7579 | |
e8a565cb YW |
7580 | if (rt->tmp_dir) |
7581 | fprintf(f, " tmp-dir=%s", rt->tmp_dir); | |
613b411c | 7582 | |
e8a565cb YW |
7583 | if (rt->var_tmp_dir) |
7584 | fprintf(f, " var-tmp-dir=%s", rt->var_tmp_dir); | |
613b411c | 7585 | |
e8a565cb YW |
7586 | if (rt->netns_storage_socket[0] >= 0) { |
7587 | int copy; | |
613b411c | 7588 | |
e8a565cb YW |
7589 | copy = fdset_put_dup(fds, rt->netns_storage_socket[0]); |
7590 | if (copy < 0) | |
7591 | return copy; | |
613b411c | 7592 | |
e8a565cb YW |
7593 | fprintf(f, " netns-socket-0=%i", copy); |
7594 | } | |
613b411c | 7595 | |
e8a565cb YW |
7596 | if (rt->netns_storage_socket[1] >= 0) { |
7597 | int copy; | |
613b411c | 7598 | |
e8a565cb YW |
7599 | copy = fdset_put_dup(fds, rt->netns_storage_socket[1]); |
7600 | if (copy < 0) | |
7601 | return copy; | |
613b411c | 7602 | |
e8a565cb YW |
7603 | fprintf(f, " netns-socket-1=%i", copy); |
7604 | } | |
7605 | ||
a70581ff XR |
7606 | if (rt->ipcns_storage_socket[0] >= 0) { |
7607 | int copy; | |
7608 | ||
7609 | copy = fdset_put_dup(fds, rt->ipcns_storage_socket[0]); | |
7610 | if (copy < 0) | |
7611 | return copy; | |
7612 | ||
7613 | fprintf(f, " ipcns-socket-0=%i", copy); | |
7614 | } | |
7615 | ||
7616 | if (rt->ipcns_storage_socket[1] >= 0) { | |
7617 | int copy; | |
7618 | ||
7619 | copy = fdset_put_dup(fds, rt->ipcns_storage_socket[1]); | |
7620 | if (copy < 0) | |
7621 | return copy; | |
7622 | ||
7623 | fprintf(f, " ipcns-socket-1=%i", copy); | |
7624 | } | |
7625 | ||
e8a565cb | 7626 | fputc('\n', f); |
613b411c LP |
7627 | } |
7628 | ||
7629 | return 0; | |
7630 | } | |
7631 | ||
e76506b7 DDM |
7632 | int exec_shared_runtime_deserialize_compat(Unit *u, const char *key, const char *value, FDSet *fds) { |
7633 | _cleanup_(exec_shared_runtime_freep) ExecSharedRuntime *rt_create = NULL; | |
7634 | ExecSharedRuntime *rt; | |
613b411c LP |
7635 | int r; |
7636 | ||
e8a565cb YW |
7637 | /* This is for the migration from old (v237 or earlier) deserialization text. |
7638 | * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=. | |
e76506b7 | 7639 | * Even if the ExecSharedRuntime object originally created by the other unit, we cannot judge |
e8a565cb YW |
7640 | * so or not from the serialized text, then we always creates a new object owned by this. */ |
7641 | ||
7642 | assert(u); | |
613b411c LP |
7643 | assert(key); |
7644 | assert(value); | |
7645 | ||
e76506b7 | 7646 | /* Manager manages ExecSharedRuntime objects by the unit id. |
e8a565cb YW |
7647 | * So, we omit the serialized text when the unit does not have id (yet?)... */ |
7648 | if (isempty(u->id)) { | |
7649 | log_unit_debug(u, "Invocation ID not found. Dropping runtime parameter."); | |
7650 | return 0; | |
7651 | } | |
613b411c | 7652 | |
e76506b7 | 7653 | if (hashmap_ensure_allocated(&u->manager->exec_shared_runtime_by_id, &string_hash_ops) < 0) |
cbc165d1 | 7654 | return log_oom(); |
e8a565cb | 7655 | |
e76506b7 | 7656 | rt = hashmap_get(u->manager->exec_shared_runtime_by_id, u->id); |
e8a565cb | 7657 | if (!rt) { |
e76506b7 | 7658 | if (exec_shared_runtime_allocate(&rt_create, u->id) < 0) |
f2341e0a | 7659 | return log_oom(); |
613b411c | 7660 | |
e8a565cb YW |
7661 | rt = rt_create; |
7662 | } | |
7663 | ||
7664 | if (streq(key, "tmp-dir")) { | |
cbc165d1 ZJS |
7665 | if (free_and_strdup_warn(&rt->tmp_dir, value) < 0) |
7666 | return -ENOMEM; | |
613b411c LP |
7667 | |
7668 | } else if (streq(key, "var-tmp-dir")) { | |
cbc165d1 ZJS |
7669 | if (free_and_strdup_warn(&rt->var_tmp_dir, value) < 0) |
7670 | return -ENOMEM; | |
613b411c LP |
7671 | |
7672 | } else if (streq(key, "netns-socket-0")) { | |
7673 | int fd; | |
7674 | ||
e652663a | 7675 | if ((fd = parse_fd(value)) < 0 || !fdset_contains(fds, fd)) { |
f2341e0a | 7676 | log_unit_debug(u, "Failed to parse netns socket value: %s", value); |
e8a565cb | 7677 | return 0; |
613b411c | 7678 | } |
e8a565cb YW |
7679 | |
7680 | safe_close(rt->netns_storage_socket[0]); | |
7681 | rt->netns_storage_socket[0] = fdset_remove(fds, fd); | |
7682 | ||
613b411c LP |
7683 | } else if (streq(key, "netns-socket-1")) { |
7684 | int fd; | |
7685 | ||
e652663a | 7686 | if ((fd = parse_fd(value)) < 0 || !fdset_contains(fds, fd)) { |
f2341e0a | 7687 | log_unit_debug(u, "Failed to parse netns socket value: %s", value); |
e8a565cb | 7688 | return 0; |
613b411c | 7689 | } |
e8a565cb YW |
7690 | |
7691 | safe_close(rt->netns_storage_socket[1]); | |
7692 | rt->netns_storage_socket[1] = fdset_remove(fds, fd); | |
a70581ff | 7693 | |
613b411c LP |
7694 | } else |
7695 | return 0; | |
7696 | ||
e76506b7 | 7697 | /* If the object is newly created, then put it to the hashmap which manages ExecSharedRuntime objects. */ |
e8a565cb | 7698 | if (rt_create) { |
e76506b7 | 7699 | r = hashmap_put(u->manager->exec_shared_runtime_by_id, rt_create->id, rt_create); |
e8a565cb | 7700 | if (r < 0) { |
3fe91079 | 7701 | log_unit_debug_errno(u, r, "Failed to put runtime parameter to manager's storage: %m"); |
e8a565cb YW |
7702 | return 0; |
7703 | } | |
613b411c | 7704 | |
e8a565cb | 7705 | rt_create->manager = u->manager; |
613b411c | 7706 | |
e8a565cb | 7707 | /* Avoid cleanup */ |
56a13a49 | 7708 | TAKE_PTR(rt_create); |
e8a565cb | 7709 | } |
98b47d54 | 7710 | |
e8a565cb YW |
7711 | return 1; |
7712 | } | |
613b411c | 7713 | |
e76506b7 | 7714 | int exec_shared_runtime_deserialize_one(Manager *m, const char *value, FDSet *fds) { |
56a13a49 ZJS |
7715 | _cleanup_free_ char *tmp_dir = NULL, *var_tmp_dir = NULL; |
7716 | char *id = NULL; | |
a70581ff | 7717 | int r, netns_fdpair[] = {-1, -1}, ipcns_fdpair[] = {-1, -1}; |
99534007 | 7718 | const char *p, *v = ASSERT_PTR(value); |
e8a565cb | 7719 | size_t n; |
613b411c | 7720 | |
e8a565cb | 7721 | assert(m); |
e8a565cb | 7722 | assert(fds); |
98b47d54 | 7723 | |
e8a565cb | 7724 | n = strcspn(v, " "); |
2f82562b | 7725 | id = strndupa_safe(v, n); |
e8a565cb YW |
7726 | if (v[n] != ' ') |
7727 | goto finalize; | |
7728 | p = v + n + 1; | |
7729 | ||
7730 | v = startswith(p, "tmp-dir="); | |
7731 | if (v) { | |
7732 | n = strcspn(v, " "); | |
56a13a49 ZJS |
7733 | tmp_dir = strndup(v, n); |
7734 | if (!tmp_dir) | |
7735 | return log_oom(); | |
e8a565cb YW |
7736 | if (v[n] != ' ') |
7737 | goto finalize; | |
7738 | p = v + n + 1; | |
7739 | } | |
7740 | ||
7741 | v = startswith(p, "var-tmp-dir="); | |
7742 | if (v) { | |
7743 | n = strcspn(v, " "); | |
56a13a49 ZJS |
7744 | var_tmp_dir = strndup(v, n); |
7745 | if (!var_tmp_dir) | |
7746 | return log_oom(); | |
e8a565cb YW |
7747 | if (v[n] != ' ') |
7748 | goto finalize; | |
7749 | p = v + n + 1; | |
7750 | } | |
7751 | ||
7752 | v = startswith(p, "netns-socket-0="); | |
7753 | if (v) { | |
7754 | char *buf; | |
7755 | ||
7756 | n = strcspn(v, " "); | |
2f82562b | 7757 | buf = strndupa_safe(v, n); |
c413bb28 | 7758 | |
e652663a DT |
7759 | netns_fdpair[0] = parse_fd(buf); |
7760 | if (netns_fdpair[0] < 0) | |
7761 | return log_debug_errno(netns_fdpair[0], "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf); | |
a70581ff | 7762 | if (!fdset_contains(fds, netns_fdpair[0])) |
c413bb28 | 7763 | return log_debug_errno(SYNTHETIC_ERRNO(EBADF), |
a70581ff XR |
7764 | "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", netns_fdpair[0]); |
7765 | netns_fdpair[0] = fdset_remove(fds, netns_fdpair[0]); | |
e8a565cb YW |
7766 | if (v[n] != ' ') |
7767 | goto finalize; | |
7768 | p = v + n + 1; | |
613b411c LP |
7769 | } |
7770 | ||
e8a565cb YW |
7771 | v = startswith(p, "netns-socket-1="); |
7772 | if (v) { | |
7773 | char *buf; | |
98b47d54 | 7774 | |
e8a565cb | 7775 | n = strcspn(v, " "); |
2f82562b | 7776 | buf = strndupa_safe(v, n); |
a70581ff | 7777 | |
e652663a DT |
7778 | netns_fdpair[1] = parse_fd(buf); |
7779 | if (netns_fdpair[1] < 0) | |
7780 | return log_debug_errno(netns_fdpair[1], "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf); | |
a70581ff XR |
7781 | if (!fdset_contains(fds, netns_fdpair[1])) |
7782 | return log_debug_errno(SYNTHETIC_ERRNO(EBADF), | |
7783 | "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", netns_fdpair[1]); | |
7784 | netns_fdpair[1] = fdset_remove(fds, netns_fdpair[1]); | |
7785 | if (v[n] != ' ') | |
7786 | goto finalize; | |
7787 | p = v + n + 1; | |
7788 | } | |
7789 | ||
7790 | v = startswith(p, "ipcns-socket-0="); | |
7791 | if (v) { | |
7792 | char *buf; | |
7793 | ||
7794 | n = strcspn(v, " "); | |
2f82562b | 7795 | buf = strndupa_safe(v, n); |
a70581ff | 7796 | |
e652663a DT |
7797 | ipcns_fdpair[0] = parse_fd(buf); |
7798 | if (ipcns_fdpair[0] < 0) | |
7799 | return log_debug_errno(ipcns_fdpair[0], "Unable to parse exec-runtime specification ipcns-socket-0=%s: %m", buf); | |
a70581ff XR |
7800 | if (!fdset_contains(fds, ipcns_fdpair[0])) |
7801 | return log_debug_errno(SYNTHETIC_ERRNO(EBADF), | |
7802 | "exec-runtime specification ipcns-socket-0= refers to unknown fd %d: %m", ipcns_fdpair[0]); | |
7803 | ipcns_fdpair[0] = fdset_remove(fds, ipcns_fdpair[0]); | |
7804 | if (v[n] != ' ') | |
7805 | goto finalize; | |
7806 | p = v + n + 1; | |
7807 | } | |
7808 | ||
7809 | v = startswith(p, "ipcns-socket-1="); | |
7810 | if (v) { | |
7811 | char *buf; | |
7812 | ||
7813 | n = strcspn(v, " "); | |
2f82562b | 7814 | buf = strndupa_safe(v, n); |
a70581ff | 7815 | |
e652663a DT |
7816 | ipcns_fdpair[1] = parse_fd(buf); |
7817 | if (ipcns_fdpair[1] < 0) | |
7818 | return log_debug_errno(ipcns_fdpair[1], "Unable to parse exec-runtime specification ipcns-socket-1=%s: %m", buf); | |
a70581ff | 7819 | if (!fdset_contains(fds, ipcns_fdpair[1])) |
c413bb28 | 7820 | return log_debug_errno(SYNTHETIC_ERRNO(EBADF), |
a70581ff XR |
7821 | "exec-runtime specification ipcns-socket-1= refers to unknown fd %d: %m", ipcns_fdpair[1]); |
7822 | ipcns_fdpair[1] = fdset_remove(fds, ipcns_fdpair[1]); | |
e8a565cb | 7823 | } |
98b47d54 | 7824 | |
e8a565cb | 7825 | finalize: |
e76506b7 | 7826 | r = exec_shared_runtime_add(m, id, &tmp_dir, &var_tmp_dir, netns_fdpair, ipcns_fdpair, NULL); |
7d853ca6 | 7827 | if (r < 0) |
56a13a49 ZJS |
7828 | return log_debug_errno(r, "Failed to add exec-runtime: %m"); |
7829 | return 0; | |
e8a565cb | 7830 | } |
613b411c | 7831 | |
e76506b7 DDM |
7832 | void exec_shared_runtime_vacuum(Manager *m) { |
7833 | ExecSharedRuntime *rt; | |
e8a565cb YW |
7834 | |
7835 | assert(m); | |
7836 | ||
e76506b7 | 7837 | /* Free unreferenced ExecSharedRuntime objects. This is used after manager deserialization process. */ |
e8a565cb | 7838 | |
e76506b7 | 7839 | HASHMAP_FOREACH(rt, m->exec_shared_runtime_by_id) { |
e8a565cb YW |
7840 | if (rt->n_ref > 0) |
7841 | continue; | |
7842 | ||
e52a696a | 7843 | (void) exec_shared_runtime_free(rt); |
e8a565cb | 7844 | } |
613b411c LP |
7845 | } |
7846 | ||
9c0c6701 DDM |
7847 | int exec_runtime_make( |
7848 | const Unit *unit, | |
7849 | const ExecContext *context, | |
7850 | ExecSharedRuntime *shared, | |
7851 | DynamicCreds *creds, | |
7852 | ExecRuntime **ret) { | |
7853 | _cleanup_close_pair_ int ephemeral_storage_socket[2] = PIPE_EBADF; | |
7854 | _cleanup_free_ char *ephemeral = NULL; | |
28135da3 | 7855 | _cleanup_(exec_runtime_freep) ExecRuntime *rt = NULL; |
9c0c6701 | 7856 | int r; |
28135da3 | 7857 | |
9c0c6701 DDM |
7858 | assert(unit); |
7859 | assert(context); | |
28135da3 DDM |
7860 | assert(ret); |
7861 | ||
9c0c6701 | 7862 | if (!shared && !creds && !exec_needs_ephemeral(context)) { |
28135da3 DDM |
7863 | *ret = NULL; |
7864 | return 0; | |
7865 | } | |
7866 | ||
9c0c6701 DDM |
7867 | if (exec_needs_ephemeral(context)) { |
7868 | r = mkdir_p("/var/lib/systemd/ephemeral-trees", 0755); | |
7869 | if (r < 0) | |
7870 | return r; | |
7871 | ||
7872 | r = tempfn_random_child("/var/lib/systemd/ephemeral-trees", unit->id, &ephemeral); | |
7873 | if (r < 0) | |
7874 | return r; | |
7875 | ||
7876 | if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, ephemeral_storage_socket) < 0) | |
7877 | return -errno; | |
7878 | } | |
7879 | ||
28135da3 DDM |
7880 | rt = new(ExecRuntime, 1); |
7881 | if (!rt) | |
7882 | return -ENOMEM; | |
7883 | ||
7884 | *rt = (ExecRuntime) { | |
7885 | .shared = shared, | |
15220772 | 7886 | .dynamic_creds = creds, |
9c0c6701 DDM |
7887 | .ephemeral_copy = TAKE_PTR(ephemeral), |
7888 | .ephemeral_storage_socket[0] = TAKE_FD(ephemeral_storage_socket[0]), | |
7889 | .ephemeral_storage_socket[1] = TAKE_FD(ephemeral_storage_socket[1]), | |
28135da3 DDM |
7890 | }; |
7891 | ||
7892 | *ret = TAKE_PTR(rt); | |
7893 | return 1; | |
7894 | } | |
7895 | ||
7896 | ExecRuntime* exec_runtime_free(ExecRuntime *rt) { | |
7897 | if (!rt) | |
7898 | return NULL; | |
7899 | ||
7900 | exec_shared_runtime_unref(rt->shared); | |
15220772 | 7901 | dynamic_creds_unref(rt->dynamic_creds); |
9c0c6701 | 7902 | |
437f3e35 | 7903 | rt->ephemeral_copy = destroy_tree(rt->ephemeral_copy); |
9c0c6701 | 7904 | |
9c0c6701 | 7905 | safe_close_pair(rt->ephemeral_storage_socket); |
28135da3 DDM |
7906 | return mfree(rt); |
7907 | } | |
7908 | ||
7909 | ExecRuntime* exec_runtime_destroy(ExecRuntime *rt) { | |
7910 | if (!rt) | |
7911 | return NULL; | |
7912 | ||
7913 | rt->shared = exec_shared_runtime_destroy(rt->shared); | |
15220772 | 7914 | rt->dynamic_creds = dynamic_creds_destroy(rt->dynamic_creds); |
28135da3 DDM |
7915 | return exec_runtime_free(rt); |
7916 | } | |
7917 | ||
b9c04eaf YW |
7918 | void exec_params_clear(ExecParameters *p) { |
7919 | if (!p) | |
7920 | return; | |
7921 | ||
c3f8a065 LP |
7922 | p->environment = strv_free(p->environment); |
7923 | p->fd_names = strv_free(p->fd_names); | |
7924 | p->fds = mfree(p->fds); | |
7925 | p->exec_fd = safe_close(p->exec_fd); | |
b9c04eaf YW |
7926 | } |
7927 | ||
bb0c0d6f LP |
7928 | ExecSetCredential *exec_set_credential_free(ExecSetCredential *sc) { |
7929 | if (!sc) | |
7930 | return NULL; | |
7931 | ||
7932 | free(sc->id); | |
7933 | free(sc->data); | |
7934 | return mfree(sc); | |
7935 | } | |
7936 | ||
43144be4 LP |
7937 | ExecLoadCredential *exec_load_credential_free(ExecLoadCredential *lc) { |
7938 | if (!lc) | |
7939 | return NULL; | |
7940 | ||
7941 | free(lc->id); | |
7942 | free(lc->path); | |
7943 | return mfree(lc); | |
7944 | } | |
7945 | ||
211a3d87 LB |
7946 | void exec_directory_done(ExecDirectory *d) { |
7947 | if (!d) | |
7948 | return; | |
7949 | ||
7950 | for (size_t i = 0; i < d->n_items; i++) { | |
7951 | free(d->items[i].path); | |
7952 | strv_free(d->items[i].symlinks); | |
7953 | } | |
7954 | ||
7955 | d->items = mfree(d->items); | |
7956 | d->n_items = 0; | |
7957 | d->mode = 0755; | |
7958 | } | |
7959 | ||
564e5c98 YW |
7960 | static ExecDirectoryItem *exec_directory_find(ExecDirectory *d, const char *path) { |
7961 | assert(d); | |
7962 | assert(path); | |
7963 | ||
7964 | for (size_t i = 0; i < d->n_items; i++) | |
7965 | if (path_equal(d->items[i].path, path)) | |
7966 | return &d->items[i]; | |
7967 | ||
7968 | return NULL; | |
7969 | } | |
7970 | ||
7971 | int exec_directory_add(ExecDirectory *d, const char *path, const char *symlink) { | |
211a3d87 LB |
7972 | _cleanup_strv_free_ char **s = NULL; |
7973 | _cleanup_free_ char *p = NULL; | |
564e5c98 YW |
7974 | ExecDirectoryItem *existing; |
7975 | int r; | |
211a3d87 LB |
7976 | |
7977 | assert(d); | |
211a3d87 LB |
7978 | assert(path); |
7979 | ||
564e5c98 YW |
7980 | existing = exec_directory_find(d, path); |
7981 | if (existing) { | |
7982 | r = strv_extend(&existing->symlinks, symlink); | |
7983 | if (r < 0) | |
7984 | return r; | |
7985 | ||
7986 | return 0; /* existing item is updated */ | |
7987 | } | |
7988 | ||
211a3d87 LB |
7989 | p = strdup(path); |
7990 | if (!p) | |
7991 | return -ENOMEM; | |
7992 | ||
564e5c98 YW |
7993 | if (symlink) { |
7994 | s = strv_new(symlink); | |
211a3d87 LB |
7995 | if (!s) |
7996 | return -ENOMEM; | |
7997 | } | |
7998 | ||
564e5c98 | 7999 | if (!GREEDY_REALLOC(d->items, d->n_items + 1)) |
211a3d87 LB |
8000 | return -ENOMEM; |
8001 | ||
564e5c98 | 8002 | d->items[d->n_items++] = (ExecDirectoryItem) { |
211a3d87 LB |
8003 | .path = TAKE_PTR(p), |
8004 | .symlinks = TAKE_PTR(s), | |
8005 | }; | |
8006 | ||
564e5c98 | 8007 | return 1; /* new item is added */ |
211a3d87 LB |
8008 | } |
8009 | ||
a2ab603c YW |
8010 | static int exec_directory_item_compare_func(const ExecDirectoryItem *a, const ExecDirectoryItem *b) { |
8011 | assert(a); | |
8012 | assert(b); | |
8013 | ||
8014 | return path_compare(a->path, b->path); | |
8015 | } | |
8016 | ||
8017 | void exec_directory_sort(ExecDirectory *d) { | |
8018 | assert(d); | |
8019 | ||
8020 | /* Sort the exec directories to make always parent directories processed at first in | |
8021 | * setup_exec_directory(), e.g., even if StateDirectory=foo/bar foo, we need to create foo at first, | |
8022 | * then foo/bar. Also, set .only_create flag if one of the parent directories is contained in the | |
8023 | * list. See also comments in setup_exec_directory() and issue #24783. */ | |
8024 | ||
8025 | if (d->n_items <= 1) | |
8026 | return; | |
8027 | ||
8028 | typesafe_qsort(d->items, d->n_items, exec_directory_item_compare_func); | |
8029 | ||
8030 | for (size_t i = 1; i < d->n_items; i++) | |
8031 | for (size_t j = 0; j < i; j++) | |
8032 | if (path_startswith(d->items[i].path, d->items[j].path)) { | |
8033 | d->items[i].only_create = true; | |
8034 | break; | |
8035 | } | |
211a3d87 LB |
8036 | } |
8037 | ||
4fb8f1e8 LP |
8038 | ExecCleanMask exec_clean_mask_from_string(const char *s) { |
8039 | ExecDirectoryType t; | |
8040 | ||
8041 | assert(s); | |
8042 | ||
8043 | if (streq(s, "all")) | |
8044 | return EXEC_CLEAN_ALL; | |
8045 | if (streq(s, "fdstore")) | |
8046 | return EXEC_CLEAN_FDSTORE; | |
8047 | ||
8048 | t = exec_resource_type_from_string(s); | |
8049 | if (t < 0) | |
8050 | return (ExecCleanMask) t; | |
8051 | ||
8052 | return 1U << t; | |
8053 | } | |
8054 | ||
bb0c0d6f | 8055 | DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops, char, string_hash_func, string_compare_func, ExecSetCredential, exec_set_credential_free); |
43144be4 | 8056 | DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_load_credential_hash_ops, char, string_hash_func, string_compare_func, ExecLoadCredential, exec_load_credential_free); |
bb0c0d6f | 8057 | |
80876c20 LP |
8058 | static const char* const exec_input_table[_EXEC_INPUT_MAX] = { |
8059 | [EXEC_INPUT_NULL] = "null", | |
8060 | [EXEC_INPUT_TTY] = "tty", | |
8061 | [EXEC_INPUT_TTY_FORCE] = "tty-force", | |
4f2d528d | 8062 | [EXEC_INPUT_TTY_FAIL] = "tty-fail", |
52c239d7 LB |
8063 | [EXEC_INPUT_SOCKET] = "socket", |
8064 | [EXEC_INPUT_NAMED_FD] = "fd", | |
08f3be7a | 8065 | [EXEC_INPUT_DATA] = "data", |
2038c3f5 | 8066 | [EXEC_INPUT_FILE] = "file", |
80876c20 LP |
8067 | }; |
8068 | ||
8a0867d6 LP |
8069 | DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput); |
8070 | ||
94f04347 | 8071 | static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = { |
80876c20 | 8072 | [EXEC_OUTPUT_INHERIT] = "inherit", |
94f04347 | 8073 | [EXEC_OUTPUT_NULL] = "null", |
80876c20 | 8074 | [EXEC_OUTPUT_TTY] = "tty", |
9a6bca7a | 8075 | [EXEC_OUTPUT_KMSG] = "kmsg", |
28dbc1e8 | 8076 | [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console", |
706343f4 LP |
8077 | [EXEC_OUTPUT_JOURNAL] = "journal", |
8078 | [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console", | |
52c239d7 LB |
8079 | [EXEC_OUTPUT_SOCKET] = "socket", |
8080 | [EXEC_OUTPUT_NAMED_FD] = "fd", | |
2038c3f5 | 8081 | [EXEC_OUTPUT_FILE] = "file", |
566b7d23 | 8082 | [EXEC_OUTPUT_FILE_APPEND] = "append", |
8d7dab1f | 8083 | [EXEC_OUTPUT_FILE_TRUNCATE] = "truncate", |
94f04347 LP |
8084 | }; |
8085 | ||
8086 | DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput); | |
023a4f67 LP |
8087 | |
8088 | static const char* const exec_utmp_mode_table[_EXEC_UTMP_MODE_MAX] = { | |
8089 | [EXEC_UTMP_INIT] = "init", | |
8090 | [EXEC_UTMP_LOGIN] = "login", | |
8091 | [EXEC_UTMP_USER] = "user", | |
8092 | }; | |
8093 | ||
8094 | DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode, ExecUtmpMode); | |
53f47dfc YW |
8095 | |
8096 | static const char* const exec_preserve_mode_table[_EXEC_PRESERVE_MODE_MAX] = { | |
8097 | [EXEC_PRESERVE_NO] = "no", | |
8098 | [EXEC_PRESERVE_YES] = "yes", | |
8099 | [EXEC_PRESERVE_RESTART] = "restart", | |
8100 | }; | |
8101 | ||
8102 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode, ExecPreserveMode, EXEC_PRESERVE_YES); | |
3536f49e | 8103 | |
6b7b2ed9 | 8104 | /* This table maps ExecDirectoryType to the setting it is configured with in the unit */ |
72fd1768 | 8105 | static const char* const exec_directory_type_table[_EXEC_DIRECTORY_TYPE_MAX] = { |
3536f49e YW |
8106 | [EXEC_DIRECTORY_RUNTIME] = "RuntimeDirectory", |
8107 | [EXEC_DIRECTORY_STATE] = "StateDirectory", | |
8108 | [EXEC_DIRECTORY_CACHE] = "CacheDirectory", | |
8109 | [EXEC_DIRECTORY_LOGS] = "LogsDirectory", | |
8110 | [EXEC_DIRECTORY_CONFIGURATION] = "ConfigurationDirectory", | |
8111 | }; | |
8112 | ||
8113 | DEFINE_STRING_TABLE_LOOKUP(exec_directory_type, ExecDirectoryType); | |
b1edf445 | 8114 | |
211a3d87 LB |
8115 | /* This table maps ExecDirectoryType to the symlink setting it is configured with in the unit */ |
8116 | static const char* const exec_directory_type_symlink_table[_EXEC_DIRECTORY_TYPE_MAX] = { | |
8117 | [EXEC_DIRECTORY_RUNTIME] = "RuntimeDirectorySymlink", | |
8118 | [EXEC_DIRECTORY_STATE] = "StateDirectorySymlink", | |
8119 | [EXEC_DIRECTORY_CACHE] = "CacheDirectorySymlink", | |
8120 | [EXEC_DIRECTORY_LOGS] = "LogsDirectorySymlink", | |
8121 | [EXEC_DIRECTORY_CONFIGURATION] = "ConfigurationDirectorySymlink", | |
8122 | }; | |
8123 | ||
8124 | DEFINE_STRING_TABLE_LOOKUP(exec_directory_type_symlink, ExecDirectoryType); | |
8125 | ||
6b7b2ed9 LP |
8126 | /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This |
8127 | * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit | |
8128 | * directories, specifically .timer units with their timestamp touch file. */ | |
8129 | static const char* const exec_resource_type_table[_EXEC_DIRECTORY_TYPE_MAX] = { | |
8130 | [EXEC_DIRECTORY_RUNTIME] = "runtime", | |
8131 | [EXEC_DIRECTORY_STATE] = "state", | |
8132 | [EXEC_DIRECTORY_CACHE] = "cache", | |
8133 | [EXEC_DIRECTORY_LOGS] = "logs", | |
8134 | [EXEC_DIRECTORY_CONFIGURATION] = "configuration", | |
8135 | }; | |
8136 | ||
8137 | DEFINE_STRING_TABLE_LOOKUP(exec_resource_type, ExecDirectoryType); | |
8138 | ||
8139 | /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to | |
8140 | * the service payload in. */ | |
fb2042dd YW |
8141 | static const char* const exec_directory_env_name_table[_EXEC_DIRECTORY_TYPE_MAX] = { |
8142 | [EXEC_DIRECTORY_RUNTIME] = "RUNTIME_DIRECTORY", | |
8143 | [EXEC_DIRECTORY_STATE] = "STATE_DIRECTORY", | |
8144 | [EXEC_DIRECTORY_CACHE] = "CACHE_DIRECTORY", | |
8145 | [EXEC_DIRECTORY_LOGS] = "LOGS_DIRECTORY", | |
8146 | [EXEC_DIRECTORY_CONFIGURATION] = "CONFIGURATION_DIRECTORY", | |
8147 | }; | |
8148 | ||
8149 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name, ExecDirectoryType); | |
8150 | ||
b1edf445 LP |
8151 | static const char* const exec_keyring_mode_table[_EXEC_KEYRING_MODE_MAX] = { |
8152 | [EXEC_KEYRING_INHERIT] = "inherit", | |
8153 | [EXEC_KEYRING_PRIVATE] = "private", | |
8154 | [EXEC_KEYRING_SHARED] = "shared", | |
8155 | }; | |
8156 | ||
8157 | DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode, ExecKeyringMode); |