]>
Commit | Line | Data |
---|---|---|
1 | /*############################################################################# | |
2 | # # | |
3 | # Pakfire - The IPFire package management system # | |
4 | # Copyright (C) 2022 Pakfire development team # | |
5 | # # | |
6 | # This program is free software: you can redistribute it and/or modify # | |
7 | # it under the terms of the GNU General Public License as published by # | |
8 | # the Free Software Foundation, either version 3 of the License, or # | |
9 | # (at your option) any later version. # | |
10 | # # | |
11 | # This program is distributed in the hope that it will be useful, # | |
12 | # but WITHOUT ANY WARRANTY; without even the implied warranty of # | |
13 | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # | |
14 | # GNU General Public License for more details. # | |
15 | # # | |
16 | # You should have received a copy of the GNU General Public License # | |
17 | # along with this program. If not, see <http://www.gnu.org/licenses/>. # | |
18 | # # | |
19 | #############################################################################*/ | |
20 | ||
21 | #include <errno.h> | |
22 | #include <fcntl.h> | |
23 | #include <linux/capability.h> | |
24 | #include <linux/sched.h> | |
25 | #include <sys/wait.h> | |
26 | #include <linux/wait.h> | |
27 | #include <sched.h> | |
28 | #include <signal.h> | |
29 | #include <stdlib.h> | |
30 | #include <syscall.h> | |
31 | #include <sys/capability.h> | |
32 | #include <sys/epoll.h> | |
33 | #include <sys/eventfd.h> | |
34 | #include <sys/mount.h> | |
35 | #include <sys/personality.h> | |
36 | #include <sys/prctl.h> | |
37 | #include <sys/resource.h> | |
38 | #include <sys/signalfd.h> | |
39 | #include <sys/timerfd.h> | |
40 | #include <sys/types.h> | |
41 | #include <sys/wait.h> | |
42 | ||
43 | // libnl3 | |
44 | #include <net/if.h> | |
45 | #include <netlink/route/link.h> | |
46 | ||
47 | // libseccomp | |
48 | #include <seccomp.h> | |
49 | ||
50 | // libuuid | |
51 | #include <uuid.h> | |
52 | ||
53 | #include <pakfire/arch.h> | |
54 | #include <pakfire/cgroup.h> | |
55 | #include <pakfire/jail.h> | |
56 | #include <pakfire/logging.h> | |
57 | #include <pakfire/mount.h> | |
58 | #include <pakfire/pakfire.h> | |
59 | #include <pakfire/path.h> | |
60 | #include <pakfire/private.h> | |
61 | #include <pakfire/pwd.h> | |
62 | #include <pakfire/string.h> | |
63 | #include <pakfire/util.h> | |
64 | ||
65 | #define BUFFER_SIZE 1024 * 64 | |
66 | #define ENVIRON_SIZE 128 | |
67 | #define EPOLL_MAX_EVENTS 2 | |
68 | #define MAX_MOUNTPOINTS 8 | |
69 | ||
70 | // The default environment that will be set for every command | |
71 | static const struct environ { | |
72 | const char* key; | |
73 | const char* val; | |
74 | } ENV[] = { | |
75 | { "HOME", "/root" }, | |
76 | { "LANG", "C.utf-8" }, | |
77 | { "PATH", "/usr/local/sbin:/usr/sbin:/sbin:/usr/local/bin:/usr/bin:/bin", }, | |
78 | { "TERM", "vt100" }, | |
79 | ||
80 | // Tell everything that it is running inside a Pakfire container | |
81 | { "container", "pakfire" }, | |
82 | { NULL, NULL }, | |
83 | }; | |
84 | ||
85 | struct pakfire_jail_mountpoint { | |
86 | char source[PATH_MAX]; | |
87 | char target[PATH_MAX]; | |
88 | int flags; | |
89 | }; | |
90 | ||
91 | struct pakfire_jail { | |
92 | struct pakfire_ctx* ctx; | |
93 | struct pakfire* pakfire; | |
94 | int nrefs; | |
95 | ||
96 | // A unique ID for each jail | |
97 | uuid_t uuid; | |
98 | char __uuid[UUID_STR_LEN]; | |
99 | ||
100 | // Resource Limits | |
101 | int nice; | |
102 | ||
103 | // Timeout | |
104 | struct itimerspec timeout; | |
105 | ||
106 | // CGroup | |
107 | struct pakfire_cgroup* cgroup; | |
108 | ||
109 | // Environment | |
110 | char* env[ENVIRON_SIZE]; | |
111 | ||
112 | // Mountpoints | |
113 | struct pakfire_jail_mountpoint mountpoints[MAX_MOUNTPOINTS]; | |
114 | unsigned int num_mountpoints; | |
115 | ||
116 | // Callbacks | |
117 | struct pakfire_jail_callbacks { | |
118 | // Log | |
119 | pakfire_jail_log_callback log; | |
120 | void* log_data; | |
121 | } callbacks; | |
122 | }; | |
123 | ||
124 | struct pakfire_log_buffer { | |
125 | char data[BUFFER_SIZE]; | |
126 | size_t used; | |
127 | }; | |
128 | ||
129 | struct pakfire_jail_exec { | |
130 | int flags; | |
131 | ||
132 | // PID (of the child) | |
133 | pid_t pid; | |
134 | int pidfd; | |
135 | ||
136 | // Process status (from waitid) | |
137 | siginfo_t status; | |
138 | ||
139 | // FD to notify the client that the parent has finished initialization | |
140 | int completed_fd; | |
141 | ||
142 | // Log pipes | |
143 | struct pakfire_jail_pipes { | |
144 | int stdin[2]; | |
145 | int stdout[2]; | |
146 | int stderr[2]; | |
147 | ||
148 | // Logging | |
149 | int log_INFO[2]; | |
150 | int log_ERROR[2]; | |
151 | #ifdef ENABLE_DEBUG | |
152 | int log_DEBUG[2]; | |
153 | #endif /* ENABLE_DEBUG */ | |
154 | } pipes; | |
155 | ||
156 | // Communicate | |
157 | struct pakfire_jail_communicate { | |
158 | pakfire_jail_communicate_in in; | |
159 | pakfire_jail_communicate_out out; | |
160 | void* data; | |
161 | } communicate; | |
162 | ||
163 | // Log buffers | |
164 | struct pakfire_jail_buffers { | |
165 | struct pakfire_log_buffer stdout; | |
166 | struct pakfire_log_buffer stderr; | |
167 | ||
168 | // Logging | |
169 | struct pakfire_log_buffer log_INFO; | |
170 | struct pakfire_log_buffer log_ERROR; | |
171 | #ifdef ENABLE_DEBUG | |
172 | struct pakfire_log_buffer log_DEBUG; | |
173 | #endif /* ENABLE_DEBUG */ | |
174 | } buffers; | |
175 | ||
176 | struct pakfire_cgroup* cgroup; | |
177 | struct pakfire_cgroup_stats cgroup_stats; | |
178 | }; | |
179 | ||
180 | static int clone3(struct clone_args* args, size_t size) { | |
181 | return syscall(__NR_clone3, args, size); | |
182 | } | |
183 | ||
184 | static int pidfd_send_signal(int pidfd, int sig, siginfo_t* info, unsigned int flags) { | |
185 | return syscall(SYS_pidfd_send_signal, pidfd, sig, info, flags); | |
186 | } | |
187 | ||
188 | static int pivot_root(const char* new_root, const char* old_root) { | |
189 | return syscall(SYS_pivot_root, new_root, old_root); | |
190 | } | |
191 | ||
192 | static int pakfire_jail_exec_has_flag( | |
193 | const struct pakfire_jail_exec* ctx, const enum pakfire_jail_exec_flags flag) { | |
194 | return ctx->flags & flag; | |
195 | } | |
196 | ||
197 | static void pakfire_jail_free(struct pakfire_jail* jail) { | |
198 | DEBUG(jail->pakfire, "Freeing jail at %p\n", jail); | |
199 | ||
200 | // Free environment | |
201 | for (unsigned int i = 0; jail->env[i]; i++) | |
202 | free(jail->env[i]); | |
203 | ||
204 | if (jail->cgroup) | |
205 | pakfire_cgroup_unref(jail->cgroup); | |
206 | if (jail->pakfire) | |
207 | pakfire_unref(jail->pakfire); | |
208 | if (jail->ctx) | |
209 | pakfire_ctx_unref(jail->ctx); | |
210 | free(jail); | |
211 | } | |
212 | ||
213 | /* | |
214 | Passes any log messages on to the default pakfire log callback | |
215 | */ | |
216 | static int pakfire_jail_default_log_callback(struct pakfire* pakfire, void* data, | |
217 | int priority, const char* line, size_t length) { | |
218 | switch (priority) { | |
219 | case LOG_INFO: | |
220 | INFO(pakfire, "%s", line); | |
221 | break; | |
222 | ||
223 | case LOG_ERR: | |
224 | ERROR(pakfire, "%s", line); | |
225 | break; | |
226 | ||
227 | #ifdef ENABLE_DEBUG | |
228 | case LOG_DEBUG: | |
229 | DEBUG(pakfire, "%s", line); | |
230 | break; | |
231 | #endif | |
232 | } | |
233 | ||
234 | return 0; | |
235 | } | |
236 | ||
237 | static const char* pakfire_jail_uuid(struct pakfire_jail* jail) { | |
238 | if (!*jail->__uuid) | |
239 | uuid_unparse_lower(jail->uuid, jail->__uuid); | |
240 | ||
241 | return jail->__uuid; | |
242 | } | |
243 | ||
244 | static int pakfire_jail_setup_interactive_env(struct pakfire_jail* jail) { | |
245 | // Set PS1 | |
246 | int r = pakfire_jail_set_env(jail, "PS1", "pakfire-jail \\w> "); | |
247 | if (r) | |
248 | return r; | |
249 | ||
250 | // Copy TERM | |
251 | char* TERM = secure_getenv("TERM"); | |
252 | if (TERM) { | |
253 | r = pakfire_jail_set_env(jail, "TERM", TERM); | |
254 | if (r) | |
255 | return r; | |
256 | } | |
257 | ||
258 | // Copy LANG | |
259 | char* LANG = secure_getenv("LANG"); | |
260 | if (LANG) { | |
261 | r = pakfire_jail_set_env(jail, "LANG", LANG); | |
262 | if (r) | |
263 | return r; | |
264 | } | |
265 | ||
266 | return 0; | |
267 | } | |
268 | ||
269 | PAKFIRE_EXPORT int pakfire_jail_create(struct pakfire_jail** jail, struct pakfire* pakfire) { | |
270 | int r; | |
271 | ||
272 | const char* arch = pakfire_get_effective_arch(pakfire); | |
273 | ||
274 | // Allocate a new jail | |
275 | struct pakfire_jail* j = calloc(1, sizeof(*j)); | |
276 | if (!j) | |
277 | return 1; | |
278 | ||
279 | // Reference context | |
280 | j->ctx = pakfire_ctx(pakfire); | |
281 | ||
282 | // Reference Pakfire | |
283 | j->pakfire = pakfire_ref(pakfire); | |
284 | ||
285 | // Initialize reference counter | |
286 | j->nrefs = 1; | |
287 | ||
288 | // Generate a random UUID | |
289 | uuid_generate_random(j->uuid); | |
290 | ||
291 | DEBUG(j->pakfire, "Allocated new jail at %p\n", j); | |
292 | ||
293 | // Set the default logging callback | |
294 | pakfire_jail_set_log_callback(j, pakfire_jail_default_log_callback, NULL); | |
295 | ||
296 | // Set default environment | |
297 | for (const struct environ* e = ENV; e->key; e++) { | |
298 | r = pakfire_jail_set_env(j, e->key, e->val); | |
299 | if (r) | |
300 | goto ERROR; | |
301 | } | |
302 | ||
303 | // Enable all CPU features that CPU has to offer | |
304 | if (!pakfire_arch_is_supported_by_host(arch)) { | |
305 | r = pakfire_jail_set_env(j, "QEMU_CPU", "max"); | |
306 | if (r) | |
307 | goto ERROR; | |
308 | } | |
309 | ||
310 | // Set container UUID | |
311 | r = pakfire_jail_set_env(j, "container_uuid", pakfire_jail_uuid(j)); | |
312 | if (r) | |
313 | goto ERROR; | |
314 | ||
315 | // Disable systemctl to talk to systemd | |
316 | if (!pakfire_on_root(j->pakfire)) { | |
317 | r = pakfire_jail_set_env(j, "SYSTEMD_OFFLINE", "1"); | |
318 | if (r) | |
319 | goto ERROR; | |
320 | } | |
321 | ||
322 | // Done | |
323 | *jail = j; | |
324 | return 0; | |
325 | ||
326 | ERROR: | |
327 | pakfire_jail_free(j); | |
328 | ||
329 | return r; | |
330 | } | |
331 | ||
332 | PAKFIRE_EXPORT struct pakfire_jail* pakfire_jail_ref(struct pakfire_jail* jail) { | |
333 | ++jail->nrefs; | |
334 | ||
335 | return jail; | |
336 | } | |
337 | ||
338 | PAKFIRE_EXPORT struct pakfire_jail* pakfire_jail_unref(struct pakfire_jail* jail) { | |
339 | if (--jail->nrefs > 0) | |
340 | return jail; | |
341 | ||
342 | pakfire_jail_free(jail); | |
343 | return NULL; | |
344 | } | |
345 | ||
346 | // Logging Callback | |
347 | ||
348 | PAKFIRE_EXPORT void pakfire_jail_set_log_callback(struct pakfire_jail* jail, | |
349 | pakfire_jail_log_callback callback, void* data) { | |
350 | jail->callbacks.log = callback; | |
351 | jail->callbacks.log_data = data; | |
352 | } | |
353 | ||
354 | // Resource Limits | |
355 | ||
356 | PAKFIRE_EXPORT int pakfire_jail_nice(struct pakfire_jail* jail, int nice) { | |
357 | // Check if nice level is in range | |
358 | if (nice < -19 || nice > 20) { | |
359 | errno = EINVAL; | |
360 | return 1; | |
361 | } | |
362 | ||
363 | // Store nice level | |
364 | jail->nice = nice; | |
365 | ||
366 | return 0; | |
367 | } | |
368 | ||
369 | int pakfire_jail_set_cgroup(struct pakfire_jail* jail, struct pakfire_cgroup* cgroup) { | |
370 | // Free any previous cgroup | |
371 | if (jail->cgroup) { | |
372 | pakfire_cgroup_unref(jail->cgroup); | |
373 | jail->cgroup = NULL; | |
374 | } | |
375 | ||
376 | // Set any new cgroup | |
377 | if (cgroup) { | |
378 | DEBUG(jail->pakfire, "Setting cgroup %p\n", cgroup); | |
379 | ||
380 | jail->cgroup = pakfire_cgroup_ref(cgroup); | |
381 | } | |
382 | ||
383 | // Done | |
384 | return 0; | |
385 | } | |
386 | ||
387 | // Environment | |
388 | ||
389 | // Returns the length of the environment | |
390 | static unsigned int pakfire_jail_env_length(struct pakfire_jail* jail) { | |
391 | unsigned int i = 0; | |
392 | ||
393 | // Count everything in the environment | |
394 | for (char** e = jail->env; *e; e++) | |
395 | i++; | |
396 | ||
397 | return i; | |
398 | } | |
399 | ||
400 | // Finds an existing environment variable and returns its index or -1 if not found | |
401 | static int pakfire_jail_find_env(struct pakfire_jail* jail, const char* key) { | |
402 | if (!key) { | |
403 | errno = EINVAL; | |
404 | return -1; | |
405 | } | |
406 | ||
407 | const size_t length = strlen(key); | |
408 | ||
409 | for (unsigned int i = 0; jail->env[i]; i++) { | |
410 | if ((pakfire_string_startswith(jail->env[i], key) | |
411 | && *(jail->env[i] + length) == '=')) { | |
412 | return i; | |
413 | } | |
414 | } | |
415 | ||
416 | // Nothing found | |
417 | return -1; | |
418 | } | |
419 | ||
420 | // Returns the value of an environment variable or NULL | |
421 | PAKFIRE_EXPORT const char* pakfire_jail_get_env(struct pakfire_jail* jail, | |
422 | const char* key) { | |
423 | int i = pakfire_jail_find_env(jail, key); | |
424 | if (i < 0) | |
425 | return NULL; | |
426 | ||
427 | return jail->env[i] + strlen(key) + 1; | |
428 | } | |
429 | ||
430 | // Sets an environment variable | |
431 | PAKFIRE_EXPORT int pakfire_jail_set_env(struct pakfire_jail* jail, | |
432 | const char* key, const char* value) { | |
433 | // Find the index where to write this value to | |
434 | int i = pakfire_jail_find_env(jail, key); | |
435 | if (i < 0) | |
436 | i = pakfire_jail_env_length(jail); | |
437 | ||
438 | // Return -ENOSPC when the environment is full | |
439 | if (i >= ENVIRON_SIZE) { | |
440 | errno = ENOSPC; | |
441 | return -1; | |
442 | } | |
443 | ||
444 | // Free any previous value | |
445 | if (jail->env[i]) | |
446 | free(jail->env[i]); | |
447 | ||
448 | // Format and set environment variable | |
449 | asprintf(&jail->env[i], "%s=%s", key, value); | |
450 | ||
451 | DEBUG(jail->pakfire, "Set environment variable: %s\n", jail->env[i]); | |
452 | ||
453 | return 0; | |
454 | } | |
455 | ||
456 | // Imports an environment | |
457 | PAKFIRE_EXPORT int pakfire_jail_import_env(struct pakfire_jail* jail, const char* env[]) { | |
458 | if (!env) | |
459 | return 0; | |
460 | ||
461 | char* key; | |
462 | char* val; | |
463 | int r; | |
464 | ||
465 | // Copy environment variables | |
466 | for (unsigned int i = 0; env[i]; i++) { | |
467 | r = pakfire_string_partition(env[i], "=", &key, &val); | |
468 | if (r) | |
469 | continue; | |
470 | ||
471 | // Set value | |
472 | r = pakfire_jail_set_env(jail, key, val); | |
473 | ||
474 | if (key) | |
475 | free(key); | |
476 | if (val) | |
477 | free(val); | |
478 | ||
479 | // Break on error | |
480 | if (r) | |
481 | return r; | |
482 | } | |
483 | ||
484 | return 0; | |
485 | } | |
486 | ||
487 | // Timeout | |
488 | ||
489 | PAKFIRE_EXPORT int pakfire_jail_set_timeout( | |
490 | struct pakfire_jail* jail, unsigned int timeout) { | |
491 | // Store value | |
492 | jail->timeout.it_value.tv_sec = timeout; | |
493 | ||
494 | if (timeout > 0) | |
495 | DEBUG(jail->pakfire, "Timeout set to %u second(s)\n", timeout); | |
496 | else | |
497 | DEBUG(jail->pakfire, "Timeout disabled\n"); | |
498 | ||
499 | return 0; | |
500 | } | |
501 | ||
502 | static int pakfire_jail_create_timer(struct pakfire_jail* jail) { | |
503 | int r; | |
504 | ||
505 | // Nothing to do if no timeout has been set | |
506 | if (!jail->timeout.it_value.tv_sec) | |
507 | return -1; | |
508 | ||
509 | // Create a new timer | |
510 | const int fd = timerfd_create(CLOCK_MONOTONIC, 0); | |
511 | if (fd < 0) { | |
512 | ERROR(jail->pakfire, "Could not create timer: %m\n"); | |
513 | goto ERROR; | |
514 | } | |
515 | ||
516 | // Arm timer | |
517 | r = timerfd_settime(fd, 0, &jail->timeout, NULL); | |
518 | if (r) { | |
519 | ERROR(jail->pakfire, "Could not arm timer: %m\n"); | |
520 | goto ERROR; | |
521 | } | |
522 | ||
523 | return fd; | |
524 | ||
525 | ERROR: | |
526 | if (fd >= 0) | |
527 | close(fd); | |
528 | ||
529 | return -1; | |
530 | } | |
531 | ||
532 | // Signals | |
533 | ||
534 | static int pakfire_jail_handle_signals(struct pakfire_jail* jail) { | |
535 | sigset_t mask; | |
536 | int r; | |
537 | ||
538 | sigemptyset(&mask); | |
539 | sigaddset(&mask, SIGINT); | |
540 | ||
541 | // Block signals | |
542 | r = sigprocmask(SIG_BLOCK, &mask, NULL); | |
543 | if (r < 0) { | |
544 | ERROR(jail->pakfire, "Failed to block signals: %m\n"); | |
545 | return r; | |
546 | } | |
547 | ||
548 | // Create a file descriptor | |
549 | r = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC); | |
550 | if (r < 0) { | |
551 | ERROR(jail->pakfire, "Failed to create signalfd: %m\n"); | |
552 | return r; | |
553 | } | |
554 | ||
555 | return r; | |
556 | } | |
557 | ||
558 | /* | |
559 | This function replaces any logging in the child process. | |
560 | ||
561 | All log messages will be sent to the parent process through their respective pipes. | |
562 | */ | |
563 | static void pakfire_jail_log_redirect(void* data, int priority, const char* file, | |
564 | int line, const char* fn, const char* format, va_list args) { | |
565 | struct pakfire_jail_pipes* pipes = (struct pakfire_jail_pipes*)data; | |
566 | int fd; | |
567 | ||
568 | switch (priority) { | |
569 | case LOG_INFO: | |
570 | fd = pipes->log_INFO[1]; | |
571 | break; | |
572 | ||
573 | case LOG_ERR: | |
574 | fd = pipes->log_ERROR[1]; | |
575 | break; | |
576 | ||
577 | #ifdef ENABLE_DEBUG | |
578 | case LOG_DEBUG: | |
579 | fd = pipes->log_DEBUG[1]; | |
580 | break; | |
581 | #endif /* ENABLE_DEBUG */ | |
582 | ||
583 | // Ignore any messages of an unknown priority | |
584 | default: | |
585 | return; | |
586 | } | |
587 | ||
588 | // Send the log message | |
589 | if (fd >= 0) | |
590 | vdprintf(fd, format, args); | |
591 | } | |
592 | ||
593 | static int pakfire_jail_log_buffer_is_full(const struct pakfire_log_buffer* buffer) { | |
594 | return (sizeof(buffer->data) == buffer->used); | |
595 | } | |
596 | ||
597 | /* | |
598 | This function reads as much data as it can from the file descriptor. | |
599 | If it finds a whole line in it, it will send it to the logger and repeat the process. | |
600 | If not newline character is found, it will try to read more data until it finds one. | |
601 | */ | |
602 | static int pakfire_jail_handle_log(struct pakfire_jail* jail, | |
603 | struct pakfire_jail_exec* ctx, int priority, int fd, | |
604 | struct pakfire_log_buffer* buffer, pakfire_jail_communicate_out callback, void* data) { | |
605 | char line[BUFFER_SIZE + 1]; | |
606 | ||
607 | // Fill up buffer from fd | |
608 | if (buffer->used < sizeof(buffer->data)) { | |
609 | ssize_t bytes_read = read(fd, buffer->data + buffer->used, | |
610 | sizeof(buffer->data) - buffer->used); | |
611 | ||
612 | // Handle errors | |
613 | if (bytes_read < 0) { | |
614 | ERROR(jail->pakfire, "Could not read from fd %d: %m\n", fd); | |
615 | return -1; | |
616 | } | |
617 | ||
618 | // Update buffer size | |
619 | buffer->used += bytes_read; | |
620 | } | |
621 | ||
622 | // See if we have any lines that we can write | |
623 | while (buffer->used) { | |
624 | // Search for the end of the first line | |
625 | char* eol = memchr(buffer->data, '\n', buffer->used); | |
626 | ||
627 | // No newline found | |
628 | if (!eol) { | |
629 | // If the buffer is full, we send the content to the logger and try again | |
630 | // This should not happen in practise | |
631 | if (pakfire_jail_log_buffer_is_full(buffer)) { | |
632 | DEBUG(jail->pakfire, "Logging buffer is full. Sending all content\n"); | |
633 | ||
634 | eol = buffer->data + sizeof(buffer->data) - 1; | |
635 | ||
636 | // Otherwise we might have only read parts of the output | |
637 | } else | |
638 | break; | |
639 | } | |
640 | ||
641 | // Find the length of the string | |
642 | size_t length = eol - buffer->data + 1; | |
643 | ||
644 | // Copy the line into the buffer | |
645 | memcpy(line, buffer->data, length); | |
646 | ||
647 | // Terminate the string | |
648 | line[length] = '\0'; | |
649 | ||
650 | // Log the line | |
651 | if (callback) { | |
652 | int r = callback(jail->pakfire, data, priority, line, length); | |
653 | if (r) { | |
654 | ERROR(jail->pakfire, "The logging callback returned an error: %d\n", r); | |
655 | return r; | |
656 | } | |
657 | } | |
658 | ||
659 | // Remove line from buffer | |
660 | memmove(buffer->data, buffer->data + length, buffer->used - length); | |
661 | buffer->used -= length; | |
662 | } | |
663 | ||
664 | return 0; | |
665 | } | |
666 | ||
667 | static int pakfire_jail_stream_stdin(struct pakfire_jail* jail, | |
668 | struct pakfire_jail_exec* ctx, const int fd) { | |
669 | int r; | |
670 | ||
671 | // Nothing to do if there is no stdin callback set | |
672 | if (!ctx->communicate.in) { | |
673 | DEBUG(jail->pakfire, "Callback for standard input is not set\n"); | |
674 | return 0; | |
675 | } | |
676 | ||
677 | // Skip if the writing pipe has already been closed | |
678 | if (!ctx->pipes.stdin[1]) | |
679 | return 0; | |
680 | ||
681 | DEBUG(jail->pakfire, "Streaming standard input...\n"); | |
682 | ||
683 | // Calling the callback | |
684 | r = ctx->communicate.in(jail->pakfire, ctx->communicate.data, fd); | |
685 | ||
686 | DEBUG(jail->pakfire, "Standard input callback finished: %d\n", r); | |
687 | ||
688 | // The callback signaled that it has written everything | |
689 | if (r == EOF) { | |
690 | DEBUG(jail->pakfire, "Closing standard input pipe\n"); | |
691 | ||
692 | // Close the file-descriptor | |
693 | close(fd); | |
694 | ||
695 | // Reset the file-descriptor so it won't be closed again later | |
696 | ctx->pipes.stdin[1] = -1; | |
697 | ||
698 | // Report success | |
699 | r = 0; | |
700 | } | |
701 | ||
702 | return r; | |
703 | } | |
704 | ||
705 | static int pakfire_jail_setup_pipe(struct pakfire_jail* jail, int (*fds)[2], const int flags) { | |
706 | int r = pipe2(*fds, flags); | |
707 | if (r < 0) { | |
708 | ERROR(jail->pakfire, "Could not setup pipe: %m\n"); | |
709 | return 1; | |
710 | } | |
711 | ||
712 | return 0; | |
713 | } | |
714 | ||
715 | static void pakfire_jail_close_pipe(struct pakfire_jail* jail, int fds[2]) { | |
716 | for (unsigned int i = 0; i < 2; i++) | |
717 | if (fds[i] >= 0) | |
718 | close(fds[i]); | |
719 | } | |
720 | ||
721 | /* | |
722 | This is a convenience function to fetch the reading end of a pipe and | |
723 | closes the write end. | |
724 | */ | |
725 | static int pakfire_jail_get_pipe_to_read(struct pakfire_jail* jail, int (*fds)[2]) { | |
726 | // Give the variables easier names to avoid confusion | |
727 | int* fd_read = &(*fds)[0]; | |
728 | int* fd_write = &(*fds)[1]; | |
729 | ||
730 | // Close the write end of the pipe | |
731 | if (*fd_write >= 0) { | |
732 | close(*fd_write); | |
733 | *fd_write = -1; | |
734 | } | |
735 | ||
736 | // Return the read end | |
737 | if (*fd_read >= 0) | |
738 | return *fd_read; | |
739 | ||
740 | return -1; | |
741 | } | |
742 | ||
743 | static int pakfire_jail_get_pipe_to_write(struct pakfire_jail* jail, int (*fds)[2]) { | |
744 | // Give the variables easier names to avoid confusion | |
745 | int* fd_read = &(*fds)[0]; | |
746 | int* fd_write = &(*fds)[1]; | |
747 | ||
748 | // Close the read end of the pipe | |
749 | if (*fd_read >= 0) { | |
750 | close(*fd_read); | |
751 | *fd_read = -1; | |
752 | } | |
753 | ||
754 | // Return the write end | |
755 | if (*fd_write >= 0) | |
756 | return *fd_write; | |
757 | ||
758 | return -1; | |
759 | } | |
760 | ||
761 | static int pakfire_jail_log(struct pakfire* pakfire, void* data, int priority, | |
762 | const char* line, const size_t length) { | |
763 | // Pass everything to the parent logger | |
764 | pakfire_log_condition(pakfire, priority, 0, "%.*s", (int)length, line); | |
765 | ||
766 | return 0; | |
767 | } | |
768 | ||
769 | static int pakfire_jail_wait(struct pakfire_jail* jail, struct pakfire_jail_exec* ctx) { | |
770 | int epollfd = -1; | |
771 | struct epoll_event ev; | |
772 | struct epoll_event events[EPOLL_MAX_EVENTS]; | |
773 | struct signalfd_siginfo siginfo; | |
774 | char garbage[8]; | |
775 | int r = 0; | |
776 | ||
777 | // Fetch file descriptors from context | |
778 | const int stdin = pakfire_jail_get_pipe_to_write(jail, &ctx->pipes.stdin); | |
779 | const int stdout = pakfire_jail_get_pipe_to_read(jail, &ctx->pipes.stdout); | |
780 | const int stderr = pakfire_jail_get_pipe_to_read(jail, &ctx->pipes.stderr); | |
781 | const int pidfd = ctx->pidfd; | |
782 | ||
783 | // Timer | |
784 | const int timerfd = pakfire_jail_create_timer(jail); | |
785 | ||
786 | // Logging | |
787 | const int log_INFO = pakfire_jail_get_pipe_to_read(jail, &ctx->pipes.log_INFO); | |
788 | const int log_ERROR = pakfire_jail_get_pipe_to_read(jail, &ctx->pipes.log_ERROR); | |
789 | #ifdef ENABLE_DEBUG | |
790 | const int log_DEBUG = pakfire_jail_get_pipe_to_read(jail, &ctx->pipes.log_DEBUG); | |
791 | #endif /* ENABLE_DEBUG */ | |
792 | ||
793 | // Signals | |
794 | const int signalfd = pakfire_jail_handle_signals(jail); | |
795 | ||
796 | // Make a list of all file descriptors we are interested in | |
797 | const int fds[] = { | |
798 | stdin, | |
799 | stdout, | |
800 | stderr, | |
801 | pidfd, | |
802 | timerfd, | |
803 | signalfd, | |
804 | log_INFO, | |
805 | log_ERROR, | |
806 | #ifdef ENABLE_DEBUG | |
807 | log_DEBUG, | |
808 | #endif /* ENABLE_DEBUG */ | |
809 | }; | |
810 | ||
811 | // Setup epoll | |
812 | epollfd = epoll_create1(0); | |
813 | if (epollfd < 0) { | |
814 | ERROR(jail->pakfire, "Could not initialize epoll(): %m\n"); | |
815 | r = 1; | |
816 | goto ERROR; | |
817 | } | |
818 | ||
819 | // Turn file descriptors into non-blocking mode and add them to epoll() | |
820 | for (unsigned int i = 0; i < sizeof(fds) / sizeof(*fds); i++) { | |
821 | int fd = fds[i]; | |
822 | ||
823 | // Skip fds which were not initialized | |
824 | if (fd < 0) | |
825 | continue; | |
826 | ||
827 | ev.events = EPOLLHUP; | |
828 | ||
829 | if (fd == stdin) | |
830 | ev.events |= EPOLLOUT; | |
831 | else | |
832 | ev.events |= EPOLLIN; | |
833 | ||
834 | // Read flags | |
835 | int flags = fcntl(fd, F_GETFL, 0); | |
836 | ||
837 | // Set modified flags | |
838 | if (fcntl(fd, F_SETFL, flags|O_NONBLOCK) < 0) { | |
839 | ERROR(jail->pakfire, | |
840 | "Could not set file descriptor %d into non-blocking mode: %m\n", fd); | |
841 | r = 1; | |
842 | goto ERROR; | |
843 | } | |
844 | ||
845 | ev.data.fd = fd; | |
846 | ||
847 | if (epoll_ctl(epollfd, EPOLL_CTL_ADD, fd, &ev) < 0) { | |
848 | ERROR(jail->pakfire, "Could not add file descriptor %d to epoll(): %m\n", fd); | |
849 | r = 1; | |
850 | goto ERROR; | |
851 | } | |
852 | } | |
853 | ||
854 | int ended = 0; | |
855 | ||
856 | // Loop for as long as the process is alive | |
857 | while (!ended) { | |
858 | int num = epoll_wait(epollfd, events, EPOLL_MAX_EVENTS, -1); | |
859 | if (num < 1) { | |
860 | // Ignore if epoll_wait() has been interrupted | |
861 | if (errno == EINTR) | |
862 | continue; | |
863 | ||
864 | ERROR(jail->pakfire, "epoll_wait() failed: %m\n"); | |
865 | r = 1; | |
866 | ||
867 | goto ERROR; | |
868 | } | |
869 | ||
870 | for (int i = 0; i < num; i++) { | |
871 | int e = events[i].events; | |
872 | int fd = events[i].data.fd; | |
873 | ||
874 | struct pakfire_log_buffer* buffer = NULL; | |
875 | pakfire_jail_communicate_out callback = NULL; | |
876 | void* data = NULL; | |
877 | int priority; | |
878 | ||
879 | // Check if there is any data to be read | |
880 | if (e & EPOLLIN) { | |
881 | // Handle any changes to the PIDFD | |
882 | if (fd == pidfd) { | |
883 | // Call waidid() and store the result | |
884 | r = waitid(P_PIDFD, ctx->pidfd, &ctx->status, WEXITED); | |
885 | if (r) { | |
886 | ERROR(jail->pakfire, "waitid() failed: %m\n"); | |
887 | goto ERROR; | |
888 | } | |
889 | ||
890 | // Mark that we have ended so that we will process the remaining | |
891 | // events from epoll() now, but won't restart the outer loop. | |
892 | ended = 1; | |
893 | continue; | |
894 | ||
895 | // Handle timer events | |
896 | } else if (fd == timerfd) { | |
897 | DEBUG(jail->pakfire, "Timer event received\n"); | |
898 | ||
899 | // Disarm the timer | |
900 | r = read(timerfd, garbage, sizeof(garbage)); | |
901 | if (r < 1) { | |
902 | ERROR(jail->pakfire, "Could not disarm timer: %m\n"); | |
903 | r = 1; | |
904 | goto ERROR; | |
905 | } | |
906 | ||
907 | // Terminate the process if it hasn't already ended | |
908 | if (!ended) { | |
909 | DEBUG(jail->pakfire, "Terminating process...\n"); | |
910 | ||
911 | // Send SIGTERM to the process | |
912 | r = pidfd_send_signal(pidfd, SIGKILL, NULL, 0); | |
913 | if (r) { | |
914 | ERROR(jail->pakfire, "Could not kill process: %m\n"); | |
915 | goto ERROR; | |
916 | } | |
917 | } | |
918 | ||
919 | // There is nothing else to do | |
920 | continue; | |
921 | ||
922 | // Handle signals | |
923 | } else if (fd == signalfd) { | |
924 | // Read the signal | |
925 | r = read(signalfd, &siginfo, sizeof(siginfo)); | |
926 | if (r < 1) { | |
927 | ERROR(jail->pakfire, "Could not read signal: %m\n"); | |
928 | goto ERROR; | |
929 | } | |
930 | ||
931 | DEBUG(jail->pakfire, "Received signal %u\n", siginfo.ssi_signo); | |
932 | ||
933 | // Handle signals | |
934 | switch (siginfo.ssi_signo) { | |
935 | // Pass SIGINT down to the child process | |
936 | case SIGINT: | |
937 | r = pidfd_send_signal(pidfd, siginfo.ssi_signo, NULL, 0); | |
938 | if (r) { | |
939 | ERROR(jail->pakfire, "Could not send signal to process: %m\n"); | |
940 | goto ERROR; | |
941 | } | |
942 | break; | |
943 | ||
944 | default: | |
945 | ERROR(jail->pakfire, "Received unhandled signal %u\n", | |
946 | siginfo.ssi_signo); | |
947 | break; | |
948 | } | |
949 | ||
950 | // Don't fall through to log processing | |
951 | continue; | |
952 | ||
953 | // Handle logging messages | |
954 | } else if (fd == log_INFO) { | |
955 | buffer = &ctx->buffers.log_INFO; | |
956 | priority = LOG_INFO; | |
957 | ||
958 | callback = pakfire_jail_log; | |
959 | ||
960 | } else if (fd == log_ERROR) { | |
961 | buffer = &ctx->buffers.log_ERROR; | |
962 | priority = LOG_ERR; | |
963 | ||
964 | callback = pakfire_jail_log; | |
965 | ||
966 | #ifdef ENABLE_DEBUG | |
967 | } else if (fd == log_DEBUG) { | |
968 | buffer = &ctx->buffers.log_DEBUG; | |
969 | priority = LOG_DEBUG; | |
970 | ||
971 | callback = pakfire_jail_log; | |
972 | #endif /* ENABLE_DEBUG */ | |
973 | ||
974 | // Handle anything from the log pipes | |
975 | } else if (fd == stdout) { | |
976 | buffer = &ctx->buffers.stdout; | |
977 | priority = LOG_INFO; | |
978 | ||
979 | // Send any output to the default logger if no callback is set | |
980 | if (ctx->communicate.out) { | |
981 | callback = ctx->communicate.out; | |
982 | data = ctx->communicate.data; | |
983 | } else { | |
984 | callback = jail->callbacks.log; | |
985 | data = jail->callbacks.log_data; | |
986 | } | |
987 | ||
988 | } else if (fd == stderr) { | |
989 | buffer = &ctx->buffers.stderr; | |
990 | priority = LOG_ERR; | |
991 | ||
992 | // Send any output to the default logger if no callback is set | |
993 | if (ctx->communicate.out) { | |
994 | callback = ctx->communicate.out; | |
995 | data = ctx->communicate.data; | |
996 | } else { | |
997 | callback = jail->callbacks.log; | |
998 | data = jail->callbacks.log_data; | |
999 | } | |
1000 | ||
1001 | } else { | |
1002 | DEBUG(jail->pakfire, "Received invalid file descriptor %d\n", fd); | |
1003 | continue; | |
1004 | } | |
1005 | ||
1006 | // Handle log event | |
1007 | r = pakfire_jail_handle_log(jail, ctx, priority, fd, buffer, callback, data); | |
1008 | if (r) | |
1009 | goto ERROR; | |
1010 | } | |
1011 | ||
1012 | if (e & EPOLLOUT) { | |
1013 | // Handle standard input | |
1014 | if (fd == stdin) { | |
1015 | r = pakfire_jail_stream_stdin(jail, ctx, fd); | |
1016 | if (r) { | |
1017 | switch (errno) { | |
1018 | // Ignore if we filled up the buffer | |
1019 | case EAGAIN: | |
1020 | break; | |
1021 | ||
1022 | default: | |
1023 | ERROR(jail->pakfire, "Could not write to stdin: %m\n"); | |
1024 | goto ERROR; | |
1025 | } | |
1026 | } | |
1027 | } | |
1028 | } | |
1029 | ||
1030 | // Check if any file descriptors have been closed | |
1031 | if (e & EPOLLHUP) { | |
1032 | // Remove the file descriptor | |
1033 | r = epoll_ctl(epollfd, EPOLL_CTL_DEL, fd, NULL); | |
1034 | if (r) { | |
1035 | ERROR(jail->pakfire, "Could not remove closed file-descriptor %d: %m\n", fd); | |
1036 | goto ERROR; | |
1037 | } | |
1038 | } | |
1039 | } | |
1040 | } | |
1041 | ||
1042 | ERROR: | |
1043 | if (epollfd >= 0) | |
1044 | close(epollfd); | |
1045 | if (timerfd >= 0) | |
1046 | close(timerfd); | |
1047 | if (signalfd >= 0) | |
1048 | close(signalfd); | |
1049 | ||
1050 | return r; | |
1051 | } | |
1052 | ||
1053 | int pakfire_jail_capture_stdout(struct pakfire* pakfire, void* data, | |
1054 | int priority, const char* line, size_t length) { | |
1055 | char** output = (char**)data; | |
1056 | int r; | |
1057 | ||
1058 | // Append everything from stdout to a buffer | |
1059 | if (output && priority == LOG_INFO) { | |
1060 | r = asprintf(output, "%s%s", (output && *output) ? *output : "", line); | |
1061 | if (r < 0) | |
1062 | return 1; | |
1063 | return 0; | |
1064 | } | |
1065 | ||
1066 | // Send everything else to the default logger | |
1067 | return pakfire_jail_default_log_callback(pakfire, NULL, priority, line, length); | |
1068 | } | |
1069 | ||
1070 | // Capabilities | |
1071 | ||
1072 | // Logs all capabilities of the current process | |
1073 | static int pakfire_jail_show_capabilities(struct pakfire_jail* jail) { | |
1074 | cap_t caps = NULL; | |
1075 | char* name = NULL; | |
1076 | cap_flag_value_t value_e; | |
1077 | cap_flag_value_t value_i; | |
1078 | cap_flag_value_t value_p; | |
1079 | int r; | |
1080 | ||
1081 | // Fetch PID | |
1082 | pid_t pid = getpid(); | |
1083 | ||
1084 | // Fetch all capabilities | |
1085 | caps = cap_get_proc(); | |
1086 | if (!caps) { | |
1087 | ERROR(jail->pakfire, "Could not fetch capabilities: %m\n"); | |
1088 | r = 1; | |
1089 | goto ERROR; | |
1090 | } | |
1091 | ||
1092 | DEBUG(jail->pakfire, "Capabilities of PID %d:\n", pid); | |
1093 | ||
1094 | // Iterate over all capabilities | |
1095 | for (unsigned int cap = 0; cap_valid(cap); cap++) { | |
1096 | name = cap_to_name(cap); | |
1097 | ||
1098 | // Fetch effective value | |
1099 | r = cap_get_flag(caps, cap, CAP_EFFECTIVE, &value_e); | |
1100 | if (r) | |
1101 | goto ERROR; | |
1102 | ||
1103 | // Fetch inheritable value | |
1104 | r = cap_get_flag(caps, cap, CAP_INHERITABLE, &value_i); | |
1105 | if (r) | |
1106 | goto ERROR; | |
1107 | ||
1108 | // Fetch permitted value | |
1109 | r = cap_get_flag(caps, cap, CAP_PERMITTED, &value_p); | |
1110 | if (r) | |
1111 | goto ERROR; | |
1112 | ||
1113 | DEBUG(jail->pakfire, | |
1114 | " %-24s : %c%c%c\n", | |
1115 | name, | |
1116 | (value_e == CAP_SET) ? 'e' : '-', | |
1117 | (value_i == CAP_SET) ? 'i' : '-', | |
1118 | (value_p == CAP_SET) ? 'p' : '-' | |
1119 | ); | |
1120 | ||
1121 | // Free name | |
1122 | cap_free(name); | |
1123 | name = NULL; | |
1124 | } | |
1125 | ||
1126 | // Success | |
1127 | r = 0; | |
1128 | ||
1129 | ERROR: | |
1130 | if (name) | |
1131 | cap_free(name); | |
1132 | if (caps) | |
1133 | cap_free(caps); | |
1134 | ||
1135 | return r; | |
1136 | } | |
1137 | ||
1138 | static int pakfire_jail_set_capabilities(struct pakfire_jail* jail) { | |
1139 | cap_t caps = NULL; | |
1140 | char* name = NULL; | |
1141 | int r; | |
1142 | ||
1143 | // Fetch capabilities | |
1144 | caps = cap_get_proc(); | |
1145 | if (!caps) { | |
1146 | ERROR(jail->pakfire, "Could not read capabilities: %m\n"); | |
1147 | r = 1; | |
1148 | goto ERROR; | |
1149 | } | |
1150 | ||
1151 | // Walk through all capabilities | |
1152 | for (cap_value_t cap = 0; cap_valid(cap); cap++) { | |
1153 | cap_value_t _caps[] = { cap }; | |
1154 | ||
1155 | // Fetch the name of the capability | |
1156 | name = cap_to_name(cap); | |
1157 | ||
1158 | r = cap_set_flag(caps, CAP_EFFECTIVE, 1, _caps, CAP_SET); | |
1159 | if (r) { | |
1160 | ERROR(jail->pakfire, "Could not set %s: %m\n", name); | |
1161 | goto ERROR; | |
1162 | } | |
1163 | ||
1164 | r = cap_set_flag(caps, CAP_INHERITABLE, 1, _caps, CAP_SET); | |
1165 | if (r) { | |
1166 | ERROR(jail->pakfire, "Could not set %s: %m\n", name); | |
1167 | goto ERROR; | |
1168 | } | |
1169 | ||
1170 | r = cap_set_flag(caps, CAP_PERMITTED, 1, _caps, CAP_SET); | |
1171 | if (r) { | |
1172 | ERROR(jail->pakfire, "Could not set %s: %m\n", name); | |
1173 | goto ERROR; | |
1174 | } | |
1175 | ||
1176 | // Free name | |
1177 | cap_free(name); | |
1178 | name = NULL; | |
1179 | } | |
1180 | ||
1181 | // Restore all capabilities | |
1182 | r = cap_set_proc(caps); | |
1183 | if (r) { | |
1184 | ERROR(jail->pakfire, "Restoring capabilities failed: %m\n"); | |
1185 | goto ERROR; | |
1186 | } | |
1187 | ||
1188 | // Add all capabilities to the ambient set | |
1189 | for (unsigned int cap = 0; cap_valid(cap); cap++) { | |
1190 | name = cap_to_name(cap); | |
1191 | ||
1192 | // Raise the capability | |
1193 | r = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_RAISE, cap, 0, 0); | |
1194 | if (r) { | |
1195 | ERROR(jail->pakfire, "Could not set ambient capability %s: %m\n", name); | |
1196 | goto ERROR; | |
1197 | } | |
1198 | ||
1199 | // Free name | |
1200 | cap_free(name); | |
1201 | name = NULL; | |
1202 | } | |
1203 | ||
1204 | // Success | |
1205 | r = 0; | |
1206 | ||
1207 | ERROR: | |
1208 | if (name) | |
1209 | cap_free(name); | |
1210 | if (caps) | |
1211 | cap_free(caps); | |
1212 | ||
1213 | return r; | |
1214 | } | |
1215 | ||
1216 | // Syscall Filter | |
1217 | ||
1218 | static int pakfire_jail_limit_syscalls(struct pakfire_jail* jail) { | |
1219 | const int syscalls[] = { | |
1220 | // The kernel's keyring isn't namespaced | |
1221 | SCMP_SYS(keyctl), | |
1222 | SCMP_SYS(add_key), | |
1223 | SCMP_SYS(request_key), | |
1224 | ||
1225 | // Disable userfaultfd | |
1226 | SCMP_SYS(userfaultfd), | |
1227 | ||
1228 | // Disable perf which could leak a lot of information about the host | |
1229 | SCMP_SYS(perf_event_open), | |
1230 | ||
1231 | 0, | |
1232 | }; | |
1233 | int r = 1; | |
1234 | ||
1235 | DEBUG(jail->pakfire, "Applying syscall filter...\n"); | |
1236 | ||
1237 | // Setup a syscall filter which allows everything by default | |
1238 | scmp_filter_ctx ctx = seccomp_init(SCMP_ACT_ALLOW); | |
1239 | if (!ctx) { | |
1240 | ERROR(jail->pakfire, "Could not setup seccomp filter: %m\n"); | |
1241 | goto ERROR; | |
1242 | } | |
1243 | ||
1244 | // All all syscalls | |
1245 | for (const int* syscall = syscalls; *syscall; syscall++) { | |
1246 | r = seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), *syscall, 0); | |
1247 | if (r) { | |
1248 | ERROR(jail->pakfire, "Could not configure syscall %d: %m\n", *syscall); | |
1249 | goto ERROR; | |
1250 | } | |
1251 | } | |
1252 | ||
1253 | // Load syscall filter into the kernel | |
1254 | r = seccomp_load(ctx); | |
1255 | if (r) { | |
1256 | ERROR(jail->pakfire, "Could not load syscall filter into the kernel: %m\n"); | |
1257 | goto ERROR; | |
1258 | } | |
1259 | ||
1260 | ERROR: | |
1261 | if (ctx) | |
1262 | seccomp_release(ctx); | |
1263 | ||
1264 | return r; | |
1265 | } | |
1266 | ||
1267 | // Mountpoints | |
1268 | ||
1269 | PAKFIRE_EXPORT int pakfire_jail_bind(struct pakfire_jail* jail, | |
1270 | const char* source, const char* target, int flags) { | |
1271 | struct pakfire_jail_mountpoint* mp = NULL; | |
1272 | int r; | |
1273 | ||
1274 | // Check if there is any space left | |
1275 | if (jail->num_mountpoints >= MAX_MOUNTPOINTS) { | |
1276 | errno = ENOSPC; | |
1277 | return 1; | |
1278 | } | |
1279 | ||
1280 | // Check for valid inputs | |
1281 | if (!source || !target) { | |
1282 | errno = EINVAL; | |
1283 | return 1; | |
1284 | } | |
1285 | ||
1286 | // Select the next free slot | |
1287 | mp = &jail->mountpoints[jail->num_mountpoints]; | |
1288 | ||
1289 | // Copy source | |
1290 | r = pakfire_string_set(mp->source, source); | |
1291 | if (r) { | |
1292 | ERROR(jail->pakfire, "Could not copy source: %m\n"); | |
1293 | return r; | |
1294 | } | |
1295 | ||
1296 | // Copy target | |
1297 | r = pakfire_string_set(mp->target, target); | |
1298 | if (r) { | |
1299 | ERROR(jail->pakfire, "Could not copy target: %m\n"); | |
1300 | return r; | |
1301 | } | |
1302 | ||
1303 | // Copy flags | |
1304 | mp->flags = flags; | |
1305 | ||
1306 | // Increment counter | |
1307 | jail->num_mountpoints++; | |
1308 | ||
1309 | return 0; | |
1310 | } | |
1311 | ||
1312 | static int pakfire_jail_mount_networking(struct pakfire_jail* jail) { | |
1313 | int r; | |
1314 | ||
1315 | const char* paths[] = { | |
1316 | "/etc/hosts", | |
1317 | "/etc/resolv.conf", | |
1318 | NULL, | |
1319 | }; | |
1320 | ||
1321 | // Bind-mount all paths read-only | |
1322 | for (const char** path = paths; *path; path++) { | |
1323 | r = pakfire_bind(jail->pakfire, *path, NULL, MS_RDONLY); | |
1324 | if (r) { | |
1325 | switch (errno) { | |
1326 | // Ignore if we don't have permission | |
1327 | case EPERM: | |
1328 | continue; | |
1329 | ||
1330 | default: | |
1331 | break; | |
1332 | } | |
1333 | return r; | |
1334 | } | |
1335 | } | |
1336 | ||
1337 | return 0; | |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | Mounts everything that we require in the new namespace | |
1342 | */ | |
1343 | static int pakfire_jail_mount(struct pakfire_jail* jail, struct pakfire_jail_exec* ctx) { | |
1344 | struct pakfire_jail_mountpoint* mp = NULL; | |
1345 | int flags = 0; | |
1346 | int r; | |
1347 | ||
1348 | // Enable loop devices | |
1349 | if (pakfire_jail_exec_has_flag(ctx, PAKFIRE_JAIL_HAS_LOOP_DEVICES)) | |
1350 | flags |= PAKFIRE_MOUNT_LOOP_DEVICES; | |
1351 | ||
1352 | // Mount all default stuff | |
1353 | r = pakfire_mount_all(jail->pakfire, flags); | |
1354 | if (r) | |
1355 | return r; | |
1356 | ||
1357 | // Mount networking stuff | |
1358 | if (pakfire_jail_exec_has_flag(ctx, PAKFIRE_JAIL_HAS_NETWORKING)) { | |
1359 | r = pakfire_jail_mount_networking(jail); | |
1360 | if (r) | |
1361 | return r; | |
1362 | } | |
1363 | ||
1364 | // Mount all custom stuff | |
1365 | for (unsigned int i = 0; i < jail->num_mountpoints; i++) { | |
1366 | // Fetch mountpoint | |
1367 | mp = &jail->mountpoints[i]; | |
1368 | ||
1369 | // Mount it | |
1370 | r = pakfire_bind(jail->pakfire, mp->source, mp->target, mp->flags); | |
1371 | if (r) | |
1372 | return r; | |
1373 | } | |
1374 | ||
1375 | // Log all mountpoints | |
1376 | pakfire_mount_list(jail->pakfire); | |
1377 | ||
1378 | return 0; | |
1379 | } | |
1380 | ||
1381 | // Networking | |
1382 | ||
1383 | static int pakfire_jail_setup_loopback(struct pakfire_jail* jail) { | |
1384 | struct nl_sock* nl = NULL; | |
1385 | struct nl_cache* cache = NULL; | |
1386 | struct rtnl_link* link = NULL; | |
1387 | struct rtnl_link* change = NULL; | |
1388 | int r; | |
1389 | ||
1390 | DEBUG(jail->pakfire, "Setting up loopback...\n"); | |
1391 | ||
1392 | // Allocate a netlink socket | |
1393 | nl = nl_socket_alloc(); | |
1394 | if (!nl) { | |
1395 | ERROR(jail->pakfire, "Could not allocate a netlink socket: %m\n"); | |
1396 | r = 1; | |
1397 | goto ERROR; | |
1398 | } | |
1399 | ||
1400 | // Connect the socket | |
1401 | r = nl_connect(nl, NETLINK_ROUTE); | |
1402 | if (r) { | |
1403 | ERROR(jail->pakfire, "Could not connect netlink socket: %s\n", nl_geterror(r)); | |
1404 | goto ERROR; | |
1405 | } | |
1406 | ||
1407 | // Allocate the netlink cache | |
1408 | r = rtnl_link_alloc_cache(nl, AF_UNSPEC, &cache); | |
1409 | if (r < 0) { | |
1410 | ERROR(jail->pakfire, "Unable to allocate netlink cache: %s\n", nl_geterror(r)); | |
1411 | goto ERROR; | |
1412 | } | |
1413 | ||
1414 | // Fetch loopback interface | |
1415 | link = rtnl_link_get_by_name(cache, "lo"); | |
1416 | if (!link) { | |
1417 | ERROR(jail->pakfire, "Could not find lo interface. Ignoring.\n"); | |
1418 | r = 0; | |
1419 | goto ERROR; | |
1420 | } | |
1421 | ||
1422 | // Allocate a new link | |
1423 | change = rtnl_link_alloc(); | |
1424 | if (!change) { | |
1425 | ERROR(jail->pakfire, "Could not allocate change link\n"); | |
1426 | r = 1; | |
1427 | goto ERROR; | |
1428 | } | |
1429 | ||
1430 | // Set the link to UP | |
1431 | rtnl_link_set_flags(change, IFF_UP); | |
1432 | ||
1433 | // Apply any changes | |
1434 | r = rtnl_link_change(nl, link, change, 0); | |
1435 | if (r) { | |
1436 | ERROR(jail->pakfire, "Unable to activate loopback: %s\n", nl_geterror(r)); | |
1437 | goto ERROR; | |
1438 | } | |
1439 | ||
1440 | // Success | |
1441 | r = 0; | |
1442 | ||
1443 | ERROR: | |
1444 | if (nl) | |
1445 | nl_socket_free(nl); | |
1446 | ||
1447 | return r; | |
1448 | } | |
1449 | ||
1450 | // UID/GID Mapping | |
1451 | ||
1452 | static int pakfire_jail_setup_uid_mapping(struct pakfire_jail* jail, pid_t pid) { | |
1453 | char path[PATH_MAX]; | |
1454 | int r; | |
1455 | ||
1456 | // Skip mapping anything when running on / | |
1457 | if (pakfire_on_root(jail->pakfire)) | |
1458 | return 0; | |
1459 | ||
1460 | // Make path | |
1461 | r = pakfire_string_format(path, "/proc/%d/uid_map", pid); | |
1462 | if (r) | |
1463 | return r; | |
1464 | ||
1465 | // Fetch UID | |
1466 | const uid_t uid = pakfire_uid(jail->pakfire); | |
1467 | ||
1468 | // Fetch SUBUID | |
1469 | const struct pakfire_subid* subuid = pakfire_subuid(jail->pakfire); | |
1470 | if (!subuid) | |
1471 | return 1; | |
1472 | ||
1473 | /* When running as root, we will map the entire range. | |
1474 | ||
1475 | When running as a non-privileged user, we will map the root user inside the jail | |
1476 | to the user's UID outside of the jail, and we will map the rest starting from one. | |
1477 | */ | |
1478 | ||
1479 | // Running as root | |
1480 | if (uid == 0) { | |
1481 | r = pakfire_file_write(jail->pakfire, path, 0, 0, 0, | |
1482 | "0 %lu %lu\n", subuid->id, subuid->length); | |
1483 | } else { | |
1484 | r = pakfire_file_write(jail->pakfire, path, 0, 0, 0, | |
1485 | "0 %lu 1\n1 %lu %lu\n", uid, subuid->id, subuid->length); | |
1486 | } | |
1487 | ||
1488 | if (r) { | |
1489 | ERROR(jail->pakfire, "Could not map UIDs: %m\n"); | |
1490 | return r; | |
1491 | } | |
1492 | ||
1493 | return r; | |
1494 | } | |
1495 | ||
1496 | static int pakfire_jail_setup_gid_mapping(struct pakfire_jail* jail, pid_t pid) { | |
1497 | char path[PATH_MAX]; | |
1498 | int r; | |
1499 | ||
1500 | // Skip mapping anything when running on / | |
1501 | if (pakfire_on_root(jail->pakfire)) | |
1502 | return 0; | |
1503 | ||
1504 | // Fetch GID | |
1505 | const gid_t gid = pakfire_gid(jail->pakfire); | |
1506 | ||
1507 | // Fetch SUBGID | |
1508 | const struct pakfire_subid* subgid = pakfire_subgid(jail->pakfire); | |
1509 | if (!subgid) | |
1510 | return 1; | |
1511 | ||
1512 | // Make path | |
1513 | r = pakfire_string_format(path, "/proc/%d/gid_map", pid); | |
1514 | if (r) | |
1515 | return r; | |
1516 | ||
1517 | // Running as root | |
1518 | if (gid == 0) { | |
1519 | r = pakfire_file_write(jail->pakfire, path, 0, 0, 0, | |
1520 | "0 %lu %lu\n", subgid->id, subgid->length); | |
1521 | } else { | |
1522 | r = pakfire_file_write(jail->pakfire, path, 0, 0, 0, | |
1523 | "0 %lu 1\n%1 %lu %lu\n", gid, subgid->id, subgid->length); | |
1524 | } | |
1525 | ||
1526 | if (r) { | |
1527 | ERROR(jail->pakfire, "Could not map GIDs: %m\n"); | |
1528 | return r; | |
1529 | } | |
1530 | ||
1531 | return r; | |
1532 | } | |
1533 | ||
1534 | static int pakfire_jail_setgroups(struct pakfire_jail* jail, pid_t pid) { | |
1535 | char path[PATH_MAX]; | |
1536 | int r = 1; | |
1537 | ||
1538 | // Make path | |
1539 | r = pakfire_string_format(path, "/proc/%d/setgroups", pid); | |
1540 | if (r) | |
1541 | return r; | |
1542 | ||
1543 | // Open file for writing | |
1544 | FILE* f = fopen(path, "w"); | |
1545 | if (!f) { | |
1546 | ERROR(jail->pakfire, "Could not open %s for writing: %m\n", path); | |
1547 | goto ERROR; | |
1548 | } | |
1549 | ||
1550 | // Write content | |
1551 | int bytes_written = fprintf(f, "deny\n"); | |
1552 | if (bytes_written <= 0) { | |
1553 | ERROR(jail->pakfire, "Could not write to %s: %m\n", path); | |
1554 | goto ERROR; | |
1555 | } | |
1556 | ||
1557 | r = fclose(f); | |
1558 | f = NULL; | |
1559 | if (r) { | |
1560 | ERROR(jail->pakfire, "Could not close %s: %m\n", path); | |
1561 | goto ERROR; | |
1562 | } | |
1563 | ||
1564 | ERROR: | |
1565 | if (f) | |
1566 | fclose(f); | |
1567 | ||
1568 | return r; | |
1569 | } | |
1570 | ||
1571 | static int pakfire_jail_send_signal(struct pakfire_jail* jail, int fd) { | |
1572 | const uint64_t val = 1; | |
1573 | int r = 0; | |
1574 | ||
1575 | DEBUG(jail->pakfire, "Sending signal...\n"); | |
1576 | ||
1577 | // Write to the file descriptor | |
1578 | ssize_t bytes_written = write(fd, &val, sizeof(val)); | |
1579 | if (bytes_written < 0 || (size_t)bytes_written < sizeof(val)) { | |
1580 | ERROR(jail->pakfire, "Could not send signal: %m\n"); | |
1581 | r = 1; | |
1582 | } | |
1583 | ||
1584 | // Close the file descriptor | |
1585 | close(fd); | |
1586 | ||
1587 | return r; | |
1588 | } | |
1589 | ||
1590 | static int pakfire_jail_wait_for_signal(struct pakfire_jail* jail, int fd) { | |
1591 | uint64_t val = 0; | |
1592 | int r = 0; | |
1593 | ||
1594 | DEBUG(jail->pakfire, "Waiting for signal...\n"); | |
1595 | ||
1596 | ssize_t bytes_read = read(fd, &val, sizeof(val)); | |
1597 | if (bytes_read < 0 || (size_t)bytes_read < sizeof(val)) { | |
1598 | ERROR(jail->pakfire, "Error waiting for signal: %m\n"); | |
1599 | r = 1; | |
1600 | } | |
1601 | ||
1602 | // Close the file descriptor | |
1603 | close(fd); | |
1604 | ||
1605 | return r; | |
1606 | } | |
1607 | ||
1608 | /* | |
1609 | Performs the initialisation that needs to happen in the parent part | |
1610 | */ | |
1611 | static int pakfire_jail_parent(struct pakfire_jail* jail, struct pakfire_jail_exec* ctx) { | |
1612 | int r; | |
1613 | ||
1614 | // Setup UID mapping | |
1615 | r = pakfire_jail_setup_uid_mapping(jail, ctx->pid); | |
1616 | if (r) | |
1617 | return r; | |
1618 | ||
1619 | // Write "deny" to /proc/PID/setgroups | |
1620 | r = pakfire_jail_setgroups(jail, ctx->pid); | |
1621 | if (r) | |
1622 | return r; | |
1623 | ||
1624 | // Setup GID mapping | |
1625 | r = pakfire_jail_setup_gid_mapping(jail, ctx->pid); | |
1626 | if (r) | |
1627 | return r; | |
1628 | ||
1629 | // Parent has finished initialisation | |
1630 | DEBUG(jail->pakfire, "Parent has finished initialization\n"); | |
1631 | ||
1632 | // Send signal to client | |
1633 | r = pakfire_jail_send_signal(jail, ctx->completed_fd); | |
1634 | if (r) | |
1635 | return r; | |
1636 | ||
1637 | return 0; | |
1638 | } | |
1639 | ||
1640 | static int pakfire_jail_switch_root(struct pakfire_jail* jail, const char* root) { | |
1641 | int r; | |
1642 | ||
1643 | // Change to the new root | |
1644 | r = chdir(root); | |
1645 | if (r) { | |
1646 | ERROR(jail->pakfire, "chdir(%s) failed: %m\n", root); | |
1647 | return r; | |
1648 | } | |
1649 | ||
1650 | // Switch Root! | |
1651 | r = pivot_root(".", "."); | |
1652 | if (r) { | |
1653 | ERROR(jail->pakfire, "Failed changing into the new root directory %s: %m\n", root); | |
1654 | return r; | |
1655 | } | |
1656 | ||
1657 | // Umount the old root | |
1658 | r = umount2(".", MNT_DETACH); | |
1659 | if (r) { | |
1660 | ERROR(jail->pakfire, "Could not umount the old root filesystem: %m\n"); | |
1661 | return r; | |
1662 | } | |
1663 | ||
1664 | return 0; | |
1665 | } | |
1666 | ||
1667 | static int pakfire_jail_child(struct pakfire_jail* jail, struct pakfire_jail_exec* ctx, | |
1668 | const char* argv[]) { | |
1669 | int r; | |
1670 | ||
1671 | // Redirect any logging to our log pipe | |
1672 | pakfire_ctx_set_log_callback(jail->ctx, pakfire_jail_log_redirect, &ctx->pipes); | |
1673 | ||
1674 | // Fetch my own PID | |
1675 | pid_t pid = getpid(); | |
1676 | ||
1677 | DEBUG(jail->pakfire, "Launched child process in jail with PID %d\n", pid); | |
1678 | ||
1679 | // Wait for the parent to finish initialization | |
1680 | r = pakfire_jail_wait_for_signal(jail, ctx->completed_fd); | |
1681 | if (r) | |
1682 | return r; | |
1683 | ||
1684 | // Die with parent | |
1685 | r = prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0); | |
1686 | if (r) { | |
1687 | ERROR(jail->pakfire, "Could not configure to die with parent: %m\n"); | |
1688 | return 126; | |
1689 | } | |
1690 | ||
1691 | // Make this process dumpable | |
1692 | r = prctl (PR_SET_DUMPABLE, 1, 0, 0, 0); | |
1693 | if (r) { | |
1694 | ERROR(jail->pakfire, "Could not make the process dumpable: %m\n"); | |
1695 | return 126; | |
1696 | } | |
1697 | ||
1698 | // Don't drop any capabilities on setuid() | |
1699 | r = prctl(PR_SET_KEEPCAPS, 1); | |
1700 | if (r) { | |
1701 | ERROR(jail->pakfire, "Could not set PR_SET_KEEPCAPS: %m\n"); | |
1702 | return 126; | |
1703 | } | |
1704 | ||
1705 | // Fetch UID/GID | |
1706 | uid_t uid = getuid(); | |
1707 | gid_t gid = getgid(); | |
1708 | ||
1709 | // Fetch EUID/EGID | |
1710 | uid_t euid = geteuid(); | |
1711 | gid_t egid = getegid(); | |
1712 | ||
1713 | DEBUG(jail->pakfire, " UID: %u (effective %u)\n", uid, euid); | |
1714 | DEBUG(jail->pakfire, " GID: %u (effective %u)\n", gid, egid); | |
1715 | ||
1716 | // Check if we are (effectively running as root) | |
1717 | if (uid || gid || euid || egid) { | |
1718 | ERROR(jail->pakfire, "Child process is not running as root\n"); | |
1719 | return 126; | |
1720 | } | |
1721 | ||
1722 | const char* root = pakfire_get_path(jail->pakfire); | |
1723 | const char* arch = pakfire_get_effective_arch(jail->pakfire); | |
1724 | ||
1725 | // Change mount propagation to slave to receive anything from the parent namespace | |
1726 | r = pakfire_mount_change_propagation(jail->ctx, "/", MS_SLAVE); | |
1727 | if (r) | |
1728 | return r; | |
1729 | ||
1730 | // Make root a mountpoint in the new mount namespace | |
1731 | r = pakfire_mount_make_mounpoint(jail->pakfire, root); | |
1732 | if (r) | |
1733 | return r; | |
1734 | ||
1735 | // Change mount propagation to private | |
1736 | r = pakfire_mount_change_propagation(jail->ctx, root, MS_PRIVATE); | |
1737 | if (r) | |
1738 | return r; | |
1739 | ||
1740 | // Change root (unless root is /) | |
1741 | if (!pakfire_on_root(jail->pakfire)) { | |
1742 | // Mount everything | |
1743 | r = pakfire_jail_mount(jail, ctx); | |
1744 | if (r) | |
1745 | return r; | |
1746 | ||
1747 | // chroot() | |
1748 | r = pakfire_jail_switch_root(jail, root); | |
1749 | if (r) | |
1750 | return r; | |
1751 | } | |
1752 | ||
1753 | // Set personality | |
1754 | unsigned long persona = pakfire_arch_personality(arch); | |
1755 | if (persona) { | |
1756 | r = personality(persona); | |
1757 | if (r < 0) { | |
1758 | ERROR(jail->pakfire, "Could not set personality (%x)\n", (unsigned int)persona); | |
1759 | return 1; | |
1760 | } | |
1761 | } | |
1762 | ||
1763 | // Setup networking | |
1764 | if (!pakfire_jail_exec_has_flag(ctx, PAKFIRE_JAIL_HAS_NETWORKING)) { | |
1765 | r = pakfire_jail_setup_loopback(jail); | |
1766 | if (r) | |
1767 | return 1; | |
1768 | } | |
1769 | ||
1770 | // Set nice level | |
1771 | if (jail->nice) { | |
1772 | DEBUG(jail->pakfire, "Setting nice level to %d\n", jail->nice); | |
1773 | ||
1774 | r = setpriority(PRIO_PROCESS, pid, jail->nice); | |
1775 | if (r) { | |
1776 | ERROR(jail->pakfire, "Could not set nice level: %m\n"); | |
1777 | return 1; | |
1778 | } | |
1779 | } | |
1780 | ||
1781 | // Close other end of log pipes | |
1782 | close(ctx->pipes.log_INFO[0]); | |
1783 | close(ctx->pipes.log_ERROR[0]); | |
1784 | #ifdef ENABLE_DEBUG | |
1785 | close(ctx->pipes.log_DEBUG[0]); | |
1786 | #endif /* ENABLE_DEBUG */ | |
1787 | ||
1788 | // Connect standard input | |
1789 | if (ctx->pipes.stdin[0] >= 0) { | |
1790 | r = dup2(ctx->pipes.stdin[0], STDIN_FILENO); | |
1791 | if (r < 0) { | |
1792 | ERROR(jail->pakfire, "Could not connect fd %d to stdin: %m\n", | |
1793 | ctx->pipes.stdin[0]); | |
1794 | ||
1795 | return 1; | |
1796 | } | |
1797 | } | |
1798 | ||
1799 | // Connect standard output and error | |
1800 | if (ctx->pipes.stdout[1] >= 0 && ctx->pipes.stderr[1] >= 0) { | |
1801 | r = dup2(ctx->pipes.stdout[1], STDOUT_FILENO); | |
1802 | if (r < 0) { | |
1803 | ERROR(jail->pakfire, "Could not connect fd %d to stdout: %m\n", | |
1804 | ctx->pipes.stdout[1]); | |
1805 | ||
1806 | return 1; | |
1807 | } | |
1808 | ||
1809 | r = dup2(ctx->pipes.stderr[1], STDERR_FILENO); | |
1810 | if (r < 0) { | |
1811 | ERROR(jail->pakfire, "Could not connect fd %d to stderr: %m\n", | |
1812 | ctx->pipes.stderr[1]); | |
1813 | ||
1814 | return 1; | |
1815 | } | |
1816 | ||
1817 | // Close the pipe (as we have moved the original file descriptors) | |
1818 | pakfire_jail_close_pipe(jail, ctx->pipes.stdin); | |
1819 | pakfire_jail_close_pipe(jail, ctx->pipes.stdout); | |
1820 | pakfire_jail_close_pipe(jail, ctx->pipes.stderr); | |
1821 | } | |
1822 | ||
1823 | // Reset open file limit (http://0pointer.net/blog/file-descriptor-limits.html) | |
1824 | r = pakfire_rlimit_reset_nofile(jail->pakfire); | |
1825 | if (r) | |
1826 | return r; | |
1827 | ||
1828 | // Set capabilities | |
1829 | r = pakfire_jail_set_capabilities(jail); | |
1830 | if (r) | |
1831 | return r; | |
1832 | ||
1833 | // Show capabilities | |
1834 | r = pakfire_jail_show_capabilities(jail); | |
1835 | if (r) | |
1836 | return r; | |
1837 | ||
1838 | // Filter syscalls | |
1839 | r = pakfire_jail_limit_syscalls(jail); | |
1840 | if (r) | |
1841 | return r; | |
1842 | ||
1843 | DEBUG(jail->pakfire, "Child process initialization done\n"); | |
1844 | DEBUG(jail->pakfire, "Launching command:\n"); | |
1845 | ||
1846 | // Log argv | |
1847 | for (unsigned int i = 0; argv[i]; i++) | |
1848 | DEBUG(jail->pakfire, " argv[%u] = %s\n", i, argv[i]); | |
1849 | ||
1850 | // exec() command | |
1851 | r = execvpe(argv[0], (char**)argv, jail->env); | |
1852 | if (r < 0) { | |
1853 | // Translate errno into regular exit code | |
1854 | switch (errno) { | |
1855 | case ENOENT: | |
1856 | // Ignore if the command doesn't exist | |
1857 | if (ctx->flags & PAKFIRE_JAIL_NOENT_OK) | |
1858 | r = 0; | |
1859 | else | |
1860 | r = 127; | |
1861 | ||
1862 | break; | |
1863 | ||
1864 | default: | |
1865 | r = 1; | |
1866 | } | |
1867 | ||
1868 | ERROR(jail->pakfire, "Could not execve(%s): %m\n", argv[0]); | |
1869 | } | |
1870 | ||
1871 | // We should not get here | |
1872 | return r; | |
1873 | } | |
1874 | ||
1875 | // Run a command in the jail | |
1876 | static int __pakfire_jail_exec(struct pakfire_jail* jail, const char* argv[], | |
1877 | const int interactive, | |
1878 | pakfire_jail_communicate_in communicate_in, | |
1879 | pakfire_jail_communicate_out communicate_out, | |
1880 | void* data, int flags) { | |
1881 | int exit = -1; | |
1882 | int r; | |
1883 | ||
1884 | // Check if argv is valid | |
1885 | if (!argv || !argv[0]) { | |
1886 | errno = EINVAL; | |
1887 | return -1; | |
1888 | } | |
1889 | ||
1890 | // Initialize context for this call | |
1891 | struct pakfire_jail_exec ctx = { | |
1892 | .flags = flags, | |
1893 | ||
1894 | .pipes = { | |
1895 | .stdin = { -1, -1 }, | |
1896 | .stdout = { -1, -1 }, | |
1897 | .stderr = { -1, -1 }, | |
1898 | .log_INFO = { -1, -1 }, | |
1899 | .log_ERROR = { -1, -1 }, | |
1900 | #ifdef ENABLE_DEBUG | |
1901 | .log_DEBUG = { -1, -1 }, | |
1902 | #endif /* ENABLE_DEBUG */ | |
1903 | }, | |
1904 | ||
1905 | .communicate = { | |
1906 | .in = communicate_in, | |
1907 | .out = communicate_out, | |
1908 | .data = data, | |
1909 | }, | |
1910 | ||
1911 | .pidfd = -1, | |
1912 | }; | |
1913 | ||
1914 | DEBUG(jail->pakfire, "Executing jail...\n"); | |
1915 | ||
1916 | // Enable networking in interactive mode | |
1917 | if (interactive) | |
1918 | ctx.flags |= PAKFIRE_JAIL_HAS_NETWORKING; | |
1919 | ||
1920 | /* | |
1921 | Setup a file descriptor which can be used to notify the client that the parent | |
1922 | has completed configuration. | |
1923 | */ | |
1924 | ctx.completed_fd = eventfd(0, EFD_CLOEXEC); | |
1925 | if (ctx.completed_fd < 0) { | |
1926 | ERROR(jail->pakfire, "eventfd() failed: %m\n"); | |
1927 | return -1; | |
1928 | } | |
1929 | ||
1930 | // Create pipes to communicate with child process if we are not running interactively | |
1931 | if (!interactive) { | |
1932 | // stdin (only if callback is set) | |
1933 | if (ctx.communicate.in) { | |
1934 | r = pakfire_jail_setup_pipe(jail, &ctx.pipes.stdin, 0); | |
1935 | if (r) | |
1936 | goto ERROR; | |
1937 | } | |
1938 | ||
1939 | // stdout | |
1940 | r = pakfire_jail_setup_pipe(jail, &ctx.pipes.stdout, 0); | |
1941 | if (r) | |
1942 | goto ERROR; | |
1943 | ||
1944 | // stderr | |
1945 | r = pakfire_jail_setup_pipe(jail, &ctx.pipes.stderr, 0); | |
1946 | if (r) | |
1947 | goto ERROR; | |
1948 | } | |
1949 | ||
1950 | // Setup pipes for logging | |
1951 | // INFO | |
1952 | r = pakfire_jail_setup_pipe(jail, &ctx.pipes.log_INFO, O_CLOEXEC); | |
1953 | if (r) | |
1954 | goto ERROR; | |
1955 | ||
1956 | // ERROR | |
1957 | r = pakfire_jail_setup_pipe(jail, &ctx.pipes.log_ERROR, O_CLOEXEC); | |
1958 | if (r) | |
1959 | goto ERROR; | |
1960 | ||
1961 | #ifdef ENABLE_DEBUG | |
1962 | // DEBUG | |
1963 | r = pakfire_jail_setup_pipe(jail, &ctx.pipes.log_DEBUG, O_CLOEXEC); | |
1964 | if (r) | |
1965 | goto ERROR; | |
1966 | #endif /* ENABLE_DEBUG */ | |
1967 | ||
1968 | // Configure child process | |
1969 | struct clone_args args = { | |
1970 | .flags = | |
1971 | CLONE_NEWCGROUP | | |
1972 | CLONE_NEWIPC | | |
1973 | CLONE_NEWNS | | |
1974 | CLONE_NEWPID | | |
1975 | CLONE_NEWTIME | | |
1976 | CLONE_NEWUSER | | |
1977 | CLONE_NEWUTS | | |
1978 | CLONE_PIDFD, | |
1979 | .exit_signal = SIGCHLD, | |
1980 | .pidfd = (long long unsigned int)&ctx.pidfd, | |
1981 | }; | |
1982 | ||
1983 | // Launch the process in a cgroup that is a leaf of the configured cgroup | |
1984 | if (jail->cgroup) { | |
1985 | args.flags |= CLONE_INTO_CGROUP; | |
1986 | ||
1987 | // Fetch our UUID | |
1988 | const char* uuid = pakfire_jail_uuid(jail); | |
1989 | ||
1990 | // Create a temporary cgroup | |
1991 | r = pakfire_cgroup_child(&ctx.cgroup, jail->cgroup, uuid, 0); | |
1992 | if (r) { | |
1993 | ERROR(jail->pakfire, "Could not create cgroup for jail: %m\n"); | |
1994 | goto ERROR; | |
1995 | } | |
1996 | ||
1997 | // Clone into this cgroup | |
1998 | args.cgroup = pakfire_cgroup_fd(ctx.cgroup); | |
1999 | } | |
2000 | ||
2001 | // Setup networking | |
2002 | if (!pakfire_jail_exec_has_flag(&ctx, PAKFIRE_JAIL_HAS_NETWORKING)) { | |
2003 | args.flags |= CLONE_NEWNET; | |
2004 | } | |
2005 | ||
2006 | // Fork this process | |
2007 | ctx.pid = clone3(&args, sizeof(args)); | |
2008 | if (ctx.pid < 0) { | |
2009 | ERROR(jail->pakfire, "Could not clone: %m\n"); | |
2010 | return -1; | |
2011 | ||
2012 | // Child process | |
2013 | } else if (ctx.pid == 0) { | |
2014 | r = pakfire_jail_child(jail, &ctx, argv); | |
2015 | _exit(r); | |
2016 | } | |
2017 | ||
2018 | // Parent process | |
2019 | r = pakfire_jail_parent(jail, &ctx); | |
2020 | if (r) | |
2021 | goto ERROR; | |
2022 | ||
2023 | DEBUG(jail->pakfire, "Waiting for PID %d to finish its work\n", ctx.pid); | |
2024 | ||
2025 | // Read output of the child process | |
2026 | r = pakfire_jail_wait(jail, &ctx); | |
2027 | if (r) | |
2028 | goto ERROR; | |
2029 | ||
2030 | // Handle exit status | |
2031 | switch (ctx.status.si_code) { | |
2032 | case CLD_EXITED: | |
2033 | DEBUG(jail->pakfire, "The child process exited with code %d\n", | |
2034 | ctx.status.si_status); | |
2035 | ||
2036 | // Pass exit code | |
2037 | exit = ctx.status.si_status; | |
2038 | break; | |
2039 | ||
2040 | case CLD_KILLED: | |
2041 | ERROR(jail->pakfire, "The child process was killed\n"); | |
2042 | exit = 139; | |
2043 | break; | |
2044 | ||
2045 | case CLD_DUMPED: | |
2046 | ERROR(jail->pakfire, "The child process terminated abnormally\n"); | |
2047 | break; | |
2048 | ||
2049 | // Log anything else | |
2050 | default: | |
2051 | ERROR(jail->pakfire, "Unknown child exit code: %d\n", ctx.status.si_code); | |
2052 | break; | |
2053 | } | |
2054 | ||
2055 | ERROR: | |
2056 | // Destroy the temporary cgroup (if any) | |
2057 | if (ctx.cgroup) { | |
2058 | // Read cgroup stats | |
2059 | r = pakfire_cgroup_stat(ctx.cgroup, &ctx.cgroup_stats); | |
2060 | if (r) { | |
2061 | ERROR(jail->pakfire, "Could not read cgroup stats: %m\n"); | |
2062 | } else { | |
2063 | pakfire_cgroup_stat_dump(ctx.cgroup, &ctx.cgroup_stats); | |
2064 | } | |
2065 | ||
2066 | pakfire_cgroup_destroy(ctx.cgroup); | |
2067 | pakfire_cgroup_unref(ctx.cgroup); | |
2068 | } | |
2069 | ||
2070 | // Close any file descriptors | |
2071 | pakfire_jail_close_pipe(jail, ctx.pipes.stdin); | |
2072 | pakfire_jail_close_pipe(jail, ctx.pipes.stdout); | |
2073 | pakfire_jail_close_pipe(jail, ctx.pipes.stderr); | |
2074 | if (ctx.pidfd >= 0) | |
2075 | close(ctx.pidfd); | |
2076 | pakfire_jail_close_pipe(jail, ctx.pipes.log_INFO); | |
2077 | pakfire_jail_close_pipe(jail, ctx.pipes.log_ERROR); | |
2078 | #ifdef ENABLE_DEBUG | |
2079 | pakfire_jail_close_pipe(jail, ctx.pipes.log_DEBUG); | |
2080 | #endif /* ENABLE_DEBUG */ | |
2081 | ||
2082 | return exit; | |
2083 | } | |
2084 | ||
2085 | PAKFIRE_EXPORT int pakfire_jail_exec( | |
2086 | struct pakfire_jail* jail, | |
2087 | const char* argv[], | |
2088 | pakfire_jail_communicate_in callback_in, | |
2089 | pakfire_jail_communicate_out callback_out, | |
2090 | void* data, int flags) { | |
2091 | return __pakfire_jail_exec(jail, argv, 0, callback_in, callback_out, data, flags); | |
2092 | } | |
2093 | ||
2094 | static int pakfire_jail_exec_interactive( | |
2095 | struct pakfire_jail* jail, const char* argv[], int flags) { | |
2096 | int r; | |
2097 | ||
2098 | // Setup interactive stuff | |
2099 | r = pakfire_jail_setup_interactive_env(jail); | |
2100 | if (r) | |
2101 | return r; | |
2102 | ||
2103 | return __pakfire_jail_exec(jail, argv, 1, NULL, NULL, NULL, flags); | |
2104 | } | |
2105 | ||
2106 | int pakfire_jail_exec_script(struct pakfire_jail* jail, | |
2107 | const char* script, | |
2108 | const size_t size, | |
2109 | const char* args[], | |
2110 | pakfire_jail_communicate_in callback_in, | |
2111 | pakfire_jail_communicate_out callback_out, | |
2112 | void* data) { | |
2113 | char path[PATH_MAX]; | |
2114 | const char** argv = NULL; | |
2115 | FILE* f = NULL; | |
2116 | int r; | |
2117 | ||
2118 | const char* root = pakfire_get_path(jail->pakfire); | |
2119 | ||
2120 | // Write the scriptlet to disk | |
2121 | r = pakfire_path_append(path, root, PAKFIRE_TMP_DIR "/pakfire-script.XXXXXX"); | |
2122 | if (r) | |
2123 | goto ERROR; | |
2124 | ||
2125 | // Create a temporary file | |
2126 | f = pakfire_mktemp(path, 0700); | |
2127 | if (!f) { | |
2128 | ERROR(jail->pakfire, "Could not create temporary file: %m\n"); | |
2129 | goto ERROR; | |
2130 | } | |
2131 | ||
2132 | DEBUG(jail->pakfire, "Writing script to %s:\n%.*s\n", path, (int)size, script); | |
2133 | ||
2134 | // Write data | |
2135 | r = fprintf(f, "%s", script); | |
2136 | if (r < 0) { | |
2137 | ERROR(jail->pakfire, "Could not write script to file %s: %m\n", path); | |
2138 | goto ERROR; | |
2139 | } | |
2140 | ||
2141 | // Close file | |
2142 | r = fclose(f); | |
2143 | if (r) { | |
2144 | ERROR(jail->pakfire, "Could not close script file %s: %m\n", path); | |
2145 | goto ERROR; | |
2146 | } | |
2147 | ||
2148 | f = NULL; | |
2149 | ||
2150 | // Count how many arguments were passed | |
2151 | unsigned int argc = 1; | |
2152 | if (args) { | |
2153 | for (const char** arg = args; *arg; arg++) | |
2154 | argc++; | |
2155 | } | |
2156 | ||
2157 | argv = calloc(argc + 1, sizeof(*argv)); | |
2158 | if (!argv) { | |
2159 | ERROR(jail->pakfire, "Could not allocate argv: %m\n"); | |
2160 | goto ERROR; | |
2161 | } | |
2162 | ||
2163 | // Set command | |
2164 | argv[0] = (root) ? pakfire_path_relpath(root, path) : path; | |
2165 | ||
2166 | // Copy args | |
2167 | for (unsigned int i = 1; i < argc; i++) | |
2168 | argv[i] = args[i-1]; | |
2169 | ||
2170 | // Run the script | |
2171 | r = pakfire_jail_exec(jail, argv, callback_in, callback_out, data, 0); | |
2172 | ||
2173 | ERROR: | |
2174 | if (argv) | |
2175 | free(argv); | |
2176 | if (f) | |
2177 | fclose(f); | |
2178 | ||
2179 | // Remove script from disk | |
2180 | if (*path) | |
2181 | unlink(path); | |
2182 | ||
2183 | return r; | |
2184 | } | |
2185 | ||
2186 | /* | |
2187 | A convenience function that creates a new jail, runs the given command and destroys | |
2188 | the jail again. | |
2189 | */ | |
2190 | int pakfire_jail_run(struct pakfire* pakfire, const char* argv[], int flags, char** output) { | |
2191 | struct pakfire_jail* jail = NULL; | |
2192 | int r; | |
2193 | ||
2194 | // Create a new jail | |
2195 | r = pakfire_jail_create(&jail, pakfire); | |
2196 | if (r) | |
2197 | goto ERROR; | |
2198 | ||
2199 | // Execute the command | |
2200 | r = pakfire_jail_exec(jail, argv, NULL, pakfire_jail_capture_stdout, output, 0); | |
2201 | ||
2202 | ERROR: | |
2203 | if (jail) | |
2204 | pakfire_jail_unref(jail); | |
2205 | ||
2206 | return r; | |
2207 | } | |
2208 | ||
2209 | int pakfire_jail_run_script(struct pakfire* pakfire, | |
2210 | const char* script, const size_t length, const char* argv[], int flags) { | |
2211 | struct pakfire_jail* jail = NULL; | |
2212 | int r; | |
2213 | ||
2214 | // Create a new jail | |
2215 | r = pakfire_jail_create(&jail, pakfire); | |
2216 | if (r) | |
2217 | goto ERROR; | |
2218 | ||
2219 | // Execute the command | |
2220 | r = pakfire_jail_exec_script(jail, script, length, argv, NULL, NULL, NULL); | |
2221 | ||
2222 | ERROR: | |
2223 | if (jail) | |
2224 | pakfire_jail_unref(jail); | |
2225 | ||
2226 | return r; | |
2227 | } | |
2228 | ||
2229 | int pakfire_jail_shell(struct pakfire_jail* jail) { | |
2230 | int r; | |
2231 | ||
2232 | const char* argv[] = { | |
2233 | "/bin/bash", "--login", NULL, | |
2234 | }; | |
2235 | ||
2236 | // Execute /bin/bash | |
2237 | r = pakfire_jail_exec_interactive(jail, argv, 0); | |
2238 | ||
2239 | // Raise any errors | |
2240 | if (r < 0) | |
2241 | return r; | |
2242 | ||
2243 | // Ignore any return codes from the shell | |
2244 | return 0; | |
2245 | } | |
2246 | ||
2247 | static int pakfire_jail_run_if_possible(struct pakfire* pakfire, const char** argv) { | |
2248 | char path[PATH_MAX]; | |
2249 | int r; | |
2250 | ||
2251 | r = pakfire_path(pakfire, path, "%s", *argv); | |
2252 | if (r) | |
2253 | return r; | |
2254 | ||
2255 | // Check if the file is executable | |
2256 | r = access(path, X_OK); | |
2257 | if (r) { | |
2258 | DEBUG(pakfire, "%s is not executable. Skipping...\n", *argv); | |
2259 | return 0; | |
2260 | } | |
2261 | ||
2262 | return pakfire_jail_run(pakfire, argv, 0, NULL); | |
2263 | } | |
2264 | ||
2265 | int pakfire_jail_ldconfig(struct pakfire* pakfire) { | |
2266 | const char* argv[] = { | |
2267 | "/sbin/ldconfig", | |
2268 | NULL, | |
2269 | }; | |
2270 | ||
2271 | return pakfire_jail_run_if_possible(pakfire, argv); | |
2272 | } | |
2273 | ||
2274 | int pakfire_jail_run_systemd_tmpfiles(struct pakfire* pakfire) { | |
2275 | const char* argv[] = { | |
2276 | "/usr/bin/systemd-tmpfiles", | |
2277 | "--create", | |
2278 | NULL, | |
2279 | }; | |
2280 | ||
2281 | return pakfire_jail_run_if_possible(pakfire, argv); | |
2282 | } |