]>
Commit | Line | Data |
---|---|---|
fd37ccaf MT |
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 | ||
32d5f21d | 21 | #include <errno.h> |
980b15af | 22 | #include <linux/capability.h> |
d853213d | 23 | #include <linux/fcntl.h> |
0bd84dc1 | 24 | #include <linux/sched.h> |
d853213d | 25 | #include <linux/wait.h> |
0bd84dc1 MT |
26 | #include <sched.h> |
27 | #include <signal.h> | |
32d5f21d | 28 | #include <stdlib.h> |
0bd84dc1 | 29 | #include <syscall.h> |
980b15af | 30 | #include <sys/capability.h> |
616f1fca | 31 | #include <sys/epoll.h> |
43dc0e16 | 32 | #include <sys/eventfd.h> |
90d92b5c | 33 | #include <sys/personality.h> |
980b15af | 34 | #include <sys/prctl.h> |
cf440db8 | 35 | #include <sys/resource.h> |
0bd84dc1 MT |
36 | #include <sys/types.h> |
37 | #include <sys/wait.h> | |
32d5f21d | 38 | |
739d5b57 MT |
39 | // libseccomp |
40 | #include <seccomp.h> | |
41 | ||
ae5201c5 MT |
42 | // libuuid |
43 | #include <uuid.h> | |
44 | ||
90d92b5c | 45 | #include <pakfire/arch.h> |
e3ddb498 | 46 | #include <pakfire/cgroup.h> |
fd37ccaf | 47 | #include <pakfire/jail.h> |
4f59c39b MT |
48 | #include <pakfire/logging.h> |
49 | #include <pakfire/mount.h> | |
fd37ccaf | 50 | #include <pakfire/pakfire.h> |
6ce56f90 | 51 | #include <pakfire/private.h> |
4896e62c | 52 | #include <pakfire/pwd.h> |
d973a13d | 53 | #include <pakfire/string.h> |
32d5f21d MT |
54 | #include <pakfire/util.h> |
55 | ||
616f1fca MT |
56 | #define BUFFER_SIZE 1024 * 64 |
57 | #define ENVIRON_SIZE 128 | |
58 | #define EPOLL_MAX_EVENTS 2 | |
cc6e2264 | 59 | #define MAX_MOUNTPOINTS 8 |
fd37ccaf | 60 | |
d5bc8fe0 MT |
61 | // The default environment that will be set for every command |
62 | static const struct environ { | |
63 | const char* key; | |
64 | const char* val; | |
65 | } ENV[] = { | |
66 | { "LANG", "en_US.utf-8" }, | |
67 | { "TERM", "vt100" }, | |
68 | { NULL, NULL }, | |
69 | }; | |
70 | ||
cc6e2264 MT |
71 | struct pakfire_jail_mountpoint { |
72 | char source[PATH_MAX]; | |
73 | char target[PATH_MAX]; | |
74 | int flags; | |
75 | }; | |
76 | ||
fd37ccaf MT |
77 | struct pakfire_jail { |
78 | struct pakfire* pakfire; | |
79 | int nrefs; | |
32d5f21d | 80 | |
ae5201c5 MT |
81 | // A unique ID for each jail |
82 | uuid_t uuid; | |
83 | char __uuid[UUID_STR_LEN]; | |
84 | ||
d639929b MT |
85 | // Flags |
86 | int flags; | |
87 | ||
cf440db8 MT |
88 | // Resource Limits |
89 | int nice; | |
90 | ||
15503538 MT |
91 | // CGroup |
92 | struct pakfire_cgroup* cgroup; | |
93 | ||
32d5f21d MT |
94 | // Environment |
95 | char* env[ENVIRON_SIZE]; | |
616f1fca MT |
96 | |
97 | // Logging | |
98 | pakfire_jail_log_callback log_callback; | |
99 | void* log_data; | |
cc6e2264 MT |
100 | |
101 | // Mountpoints | |
102 | struct pakfire_jail_mountpoint mountpoints[MAX_MOUNTPOINTS]; | |
103 | unsigned int num_mountpoints; | |
616f1fca MT |
104 | }; |
105 | ||
106 | struct pakfire_log_buffer { | |
107 | char data[BUFFER_SIZE]; | |
108 | size_t used; | |
109 | }; | |
110 | ||
111 | struct pakfire_jail_exec { | |
112 | // PID (of the child) | |
113 | pid_t pid; | |
d853213d | 114 | int pidfd; |
616f1fca | 115 | |
d853213d MT |
116 | // Process status (from waitid) |
117 | siginfo_t status; | |
616f1fca | 118 | |
f7d240a7 MT |
119 | // FD to notify the client that the parent has finished initialization |
120 | int completed_fd; | |
121 | ||
616f1fca | 122 | // Log pipes |
e33387d3 | 123 | struct pakfire_jail_pipes { |
616f1fca MT |
124 | int stdout[2]; |
125 | int stderr[2]; | |
e33387d3 MT |
126 | |
127 | // Logging | |
128 | int log_INFO[2]; | |
129 | int log_ERROR[2]; | |
130 | int log_DEBUG[2]; | |
616f1fca MT |
131 | } pipes; |
132 | ||
133 | // Log buffers | |
e33387d3 | 134 | struct pakfire_jail_buffers { |
616f1fca MT |
135 | struct pakfire_log_buffer stdout; |
136 | struct pakfire_log_buffer stderr; | |
e33387d3 MT |
137 | |
138 | // Logging | |
139 | struct pakfire_log_buffer log_INFO; | |
140 | struct pakfire_log_buffer log_ERROR; | |
141 | struct pakfire_log_buffer log_DEBUG; | |
616f1fca | 142 | } buffers; |
aca565fc MT |
143 | |
144 | struct pakfire_cgroup* cgroup; | |
6b7cf275 | 145 | struct pakfire_cgroup_stats cgroup_stats; |
fd37ccaf MT |
146 | }; |
147 | ||
0bd84dc1 MT |
148 | static int clone3(struct clone_args* args, size_t size) { |
149 | return syscall(__NR_clone3, args, size); | |
150 | } | |
151 | ||
d5bc8fe0 MT |
152 | static void pakfire_jail_free(struct pakfire_jail* jail) { |
153 | DEBUG(jail->pakfire, "Freeing jail at %p\n", jail); | |
154 | ||
155 | // Free environment | |
156 | for (unsigned int i = 0; jail->env[i]; i++) | |
157 | free(jail->env[i]); | |
158 | ||
d34b1e00 MT |
159 | if (jail->cgroup) |
160 | pakfire_cgroup_unref(jail->cgroup); | |
161 | ||
d5bc8fe0 MT |
162 | pakfire_unref(jail->pakfire); |
163 | free(jail); | |
164 | } | |
165 | ||
e33387d3 MT |
166 | /* |
167 | Passes any log messages on to the default pakfire log callback | |
168 | */ | |
fed41508 MT |
169 | static int pakfire_jail_default_log_callback(struct pakfire* pakfire, void* data, |
170 | int priority, const char* line, size_t length) { | |
171 | switch (priority) { | |
172 | case LOG_INFO: | |
173 | INFO(pakfire, "%s", line); | |
174 | break; | |
175 | ||
176 | case LOG_ERR: | |
177 | ERROR(pakfire, "%s", line); | |
178 | break; | |
e33387d3 MT |
179 | |
180 | #ifdef ENABLE_DEBUG | |
181 | case LOG_DEBUG: | |
182 | DEBUG(pakfire, "%s", line); | |
183 | break; | |
184 | #endif | |
fed41508 MT |
185 | } |
186 | ||
187 | return 0; | |
188 | } | |
189 | ||
00ba1d9a MT |
190 | static int pakfire_jail_setup_interactive_env(struct pakfire_jail* jail) { |
191 | // Set PS1 | |
192 | int r = pakfire_jail_set_env(jail, "PS1", "pakfire-jail \\w> "); | |
193 | if (r) | |
194 | return r; | |
195 | ||
196 | // Copy TERM | |
197 | char* TERM = secure_getenv("TERM"); | |
198 | if (TERM) { | |
199 | r = pakfire_jail_set_env(jail, "TERM", TERM); | |
200 | if (r) | |
201 | return r; | |
202 | } | |
203 | ||
204 | // Copy LANG | |
205 | char* LANG = secure_getenv("LANG"); | |
206 | if (LANG) { | |
207 | r = pakfire_jail_set_env(jail, "LANG", LANG); | |
208 | if (r) | |
209 | return r; | |
210 | } | |
211 | ||
212 | return 0; | |
213 | } | |
214 | ||
6ce56f90 MT |
215 | PAKFIRE_EXPORT int pakfire_jail_create(struct pakfire_jail** jail, |
216 | struct pakfire* pakfire, int flags) { | |
d5bc8fe0 MT |
217 | int r; |
218 | ||
219 | // Allocate a new jail | |
fd37ccaf MT |
220 | struct pakfire_jail* j = calloc(1, sizeof(*j)); |
221 | if (!j) | |
222 | return 1; | |
223 | ||
224 | // Reference Pakfire | |
225 | j->pakfire = pakfire_ref(pakfire); | |
226 | ||
227 | // Initialize reference counter | |
228 | j->nrefs = 1; | |
229 | ||
d639929b MT |
230 | // Store flags |
231 | j->flags = flags; | |
232 | ||
ae5201c5 MT |
233 | // Generate a random UUID |
234 | uuid_generate_random(j->uuid); | |
235 | ||
84bd7655 MT |
236 | DEBUG(j->pakfire, "Allocated new jail at %p\n", j); |
237 | ||
fed41508 MT |
238 | // Set default log callback |
239 | r = pakfire_jail_set_log_callback(j, pakfire_jail_default_log_callback, NULL); | |
240 | if (r) | |
241 | goto ERROR; | |
242 | ||
d5bc8fe0 MT |
243 | // Set default environment |
244 | for (const struct environ* e = ENV; e->key; e++) { | |
245 | r = pakfire_jail_set_env(j, e->key, e->val); | |
246 | if (r) | |
247 | goto ERROR; | |
248 | } | |
249 | ||
fd37ccaf MT |
250 | // Done |
251 | *jail = j; | |
252 | return 0; | |
84bd7655 | 253 | |
d5bc8fe0 MT |
254 | ERROR: |
255 | pakfire_jail_free(j); | |
32d5f21d | 256 | |
d5bc8fe0 | 257 | return r; |
fd37ccaf MT |
258 | } |
259 | ||
6ce56f90 | 260 | PAKFIRE_EXPORT struct pakfire_jail* pakfire_jail_ref(struct pakfire_jail* jail) { |
fd37ccaf MT |
261 | ++jail->nrefs; |
262 | ||
263 | return jail; | |
264 | } | |
265 | ||
6ce56f90 | 266 | PAKFIRE_EXPORT struct pakfire_jail* pakfire_jail_unref(struct pakfire_jail* jail) { |
fd37ccaf MT |
267 | if (--jail->nrefs > 0) |
268 | return jail; | |
269 | ||
270 | pakfire_jail_free(jail); | |
271 | return NULL; | |
272 | } | |
32d5f21d | 273 | |
ae5201c5 | 274 | static const char* pakfire_jail_uuid(struct pakfire_jail* jail) { |
8b2cc8ae | 275 | if (!*jail->__uuid) |
ae5201c5 MT |
276 | uuid_unparse_lower(jail->uuid, jail->__uuid); |
277 | ||
278 | return jail->__uuid; | |
279 | } | |
280 | ||
cf440db8 MT |
281 | // Resource Limits |
282 | ||
283 | PAKFIRE_EXPORT int pakfire_jail_nice(struct pakfire_jail* jail, int nice) { | |
284 | // Check if nice level is in range | |
285 | if (nice < -19 || nice > 20) { | |
286 | errno = EINVAL; | |
287 | return 1; | |
288 | } | |
289 | ||
290 | // Store nice level | |
291 | jail->nice = nice; | |
292 | ||
293 | return 0; | |
294 | } | |
295 | ||
15503538 MT |
296 | int pakfire_jail_set_cgroup(struct pakfire_jail* jail, struct pakfire_cgroup* cgroup) { |
297 | // Free any previous cgroup | |
298 | if (jail->cgroup) { | |
299 | pakfire_cgroup_unref(jail->cgroup); | |
300 | jail->cgroup = NULL; | |
301 | } | |
302 | ||
303 | // Set any new cgroup | |
304 | if (cgroup) { | |
305 | DEBUG(jail->pakfire, "Setting cgroup %p\n", cgroup); | |
306 | ||
307 | jail->cgroup = pakfire_cgroup_ref(cgroup); | |
308 | } | |
309 | ||
310 | // Done | |
311 | return 0; | |
312 | } | |
313 | ||
32d5f21d MT |
314 | // Environment |
315 | ||
316 | // Returns the length of the environment | |
317 | static unsigned int pakfire_jail_env_length(struct pakfire_jail* jail) { | |
318 | unsigned int i = 0; | |
319 | ||
320 | // Count everything in the environment | |
321 | for (char** e = jail->env; *e; e++) | |
322 | i++; | |
323 | ||
324 | return i; | |
325 | } | |
326 | ||
327 | // Finds an existing environment variable and returns its index or -1 if not found | |
328 | static int pakfire_jail_find_env(struct pakfire_jail* jail, const char* key) { | |
329 | if (!key) { | |
330 | errno = EINVAL; | |
331 | return -1; | |
332 | } | |
333 | ||
334 | char buffer[strlen(key) + 2]; | |
335 | pakfire_string_format(buffer, "%s=", key); | |
336 | ||
337 | for (unsigned int i = 0; jail->env[i]; i++) { | |
338 | if (pakfire_string_startswith(jail->env[i], buffer)) | |
339 | return i; | |
340 | } | |
341 | ||
342 | // Nothing found | |
343 | return -1; | |
344 | } | |
345 | ||
346 | // Returns the value of an environment variable or NULL | |
6ce56f90 MT |
347 | PAKFIRE_EXPORT const char* pakfire_jail_get_env(struct pakfire_jail* jail, |
348 | const char* key) { | |
32d5f21d MT |
349 | int i = pakfire_jail_find_env(jail, key); |
350 | if (i < 0) | |
351 | return NULL; | |
352 | ||
353 | return jail->env[i] + strlen(key) + 1; | |
354 | } | |
355 | ||
356 | // Sets an environment variable | |
6ce56f90 MT |
357 | PAKFIRE_EXPORT int pakfire_jail_set_env(struct pakfire_jail* jail, |
358 | const char* key, const char* value) { | |
32d5f21d MT |
359 | // Find the index where to write this value to |
360 | int i = pakfire_jail_find_env(jail, key); | |
361 | if (i < 0) | |
362 | i = pakfire_jail_env_length(jail); | |
363 | ||
364 | // Return -ENOSPC when the environment is full | |
365 | if (i >= ENVIRON_SIZE) { | |
366 | errno = ENOSPC; | |
367 | return -1; | |
368 | } | |
369 | ||
370 | // Free any previous value | |
371 | if (jail->env[i]) | |
372 | free(jail->env[i]); | |
373 | ||
374 | // Format and set environment variable | |
375 | asprintf(&jail->env[i], "%s=%s", key, value); | |
376 | ||
377 | DEBUG(jail->pakfire, "Set environment variable: %s\n", jail->env[i]); | |
378 | ||
379 | return 0; | |
380 | } | |
9f50bf71 | 381 | |
939025e7 | 382 | // Imports an environment |
6ce56f90 | 383 | PAKFIRE_EXPORT int pakfire_jail_import_env(struct pakfire_jail* jail, const char* env[]) { |
939025e7 MT |
384 | if (!env) |
385 | return 0; | |
386 | ||
387 | char* key; | |
388 | char* val; | |
389 | int r; | |
390 | ||
391 | // Copy environment variables | |
392 | for (unsigned int i = 0; env[i]; i++) { | |
393 | r = pakfire_string_partition(env[i], "=", &key, &val); | |
394 | if (r) | |
395 | continue; | |
396 | ||
397 | // Set value | |
398 | r = pakfire_jail_set_env(jail, key, val); | |
399 | ||
400 | if (key) | |
401 | free(key); | |
402 | if (val) | |
403 | free(val); | |
404 | ||
405 | // Break on error | |
406 | if (r) | |
407 | return r; | |
408 | } | |
409 | ||
410 | return 0; | |
411 | } | |
412 | ||
616f1fca MT |
413 | // Logging |
414 | ||
6ce56f90 | 415 | PAKFIRE_EXPORT int pakfire_jail_set_log_callback(struct pakfire_jail* jail, |
dc3e6dd4 MT |
416 | pakfire_jail_log_callback callback, void* data) { |
417 | jail->log_callback = callback; | |
418 | jail->log_data = data; | |
419 | ||
420 | return 0; | |
421 | } | |
422 | ||
e33387d3 MT |
423 | /* |
424 | This function replaces any logging in the child process. | |
425 | ||
426 | All log messages will be sent to the parent process through their respective pipes. | |
427 | */ | |
428 | static void pakfire_jail_log(void* data, int priority, const char* file, | |
429 | int line, const char* fn, const char* format, va_list args) { | |
430 | struct pakfire_jail_pipes* pipes = (struct pakfire_jail_pipes*)data; | |
431 | int fd; | |
432 | ||
433 | switch (priority) { | |
434 | case LOG_INFO: | |
435 | fd = pipes->log_INFO[1]; | |
436 | break; | |
437 | ||
438 | case LOG_ERR: | |
439 | fd = pipes->log_ERROR[1]; | |
440 | break; | |
441 | ||
442 | #ifdef ENABLE_DEBUG | |
443 | case LOG_DEBUG: | |
444 | fd = pipes->log_DEBUG[1]; | |
445 | break; | |
446 | #endif /* ENABLE_DEBUG */ | |
447 | ||
448 | // Ignore any messages of an unknown priority | |
449 | default: | |
450 | return; | |
451 | } | |
452 | ||
453 | // Send the log message | |
454 | if (fd) | |
455 | vdprintf(fd, format, args); | |
456 | } | |
457 | ||
616f1fca MT |
458 | static int pakfire_jail_log_buffer_is_full(const struct pakfire_log_buffer* buffer) { |
459 | return (sizeof(buffer->data) == buffer->used); | |
460 | } | |
461 | ||
462 | /* | |
463 | This function reads as much data as it can from the file descriptor. | |
464 | If it finds a whole line in it, it will send it to the logger and repeat the process. | |
465 | If not newline character is found, it will try to read more data until it finds one. | |
466 | */ | |
467 | static int pakfire_jail_handle_log(struct pakfire_jail* jail, | |
e33387d3 MT |
468 | struct pakfire_jail_exec* ctx, int priority, int fd, |
469 | struct pakfire_log_buffer* buffer, pakfire_jail_log_callback callback, void* data) { | |
616f1fca MT |
470 | char line[BUFFER_SIZE + 1]; |
471 | ||
472 | // Fill up buffer from fd | |
473 | if (buffer->used < sizeof(buffer->data)) { | |
474 | ssize_t bytes_read = read(fd, buffer->data + buffer->used, | |
475 | sizeof(buffer->data) - buffer->used); | |
476 | ||
477 | // Handle errors | |
478 | if (bytes_read < 0) { | |
479 | ERROR(jail->pakfire, "Could not read from fd %d: %m\n", fd); | |
480 | return -1; | |
481 | } | |
482 | ||
483 | // Update buffer size | |
484 | buffer->used += bytes_read; | |
485 | } | |
486 | ||
487 | // See if we have any lines that we can write | |
488 | while (buffer->used) { | |
489 | // Search for the end of the first line | |
490 | char* eol = memchr(buffer->data, '\n', buffer->used); | |
491 | ||
492 | // No newline found | |
493 | if (!eol) { | |
494 | // If the buffer is full, we send the content to the logger and try again | |
495 | // This should not happen in practise | |
496 | if (pakfire_jail_log_buffer_is_full(buffer)) { | |
497 | DEBUG(jail->pakfire, "Logging buffer is full. Sending all content\n"); | |
498 | ||
499 | eol = buffer->data + sizeof(buffer->data) - 1; | |
500 | ||
501 | // Otherwise we might have only read parts of the output | |
502 | } else | |
503 | break; | |
504 | } | |
505 | ||
506 | // Find the length of the string | |
507 | size_t length = eol - buffer->data + 1; | |
508 | ||
509 | // Copy the line into the buffer | |
510 | memcpy(line, buffer->data, length); | |
511 | ||
512 | // Terminate the string | |
513 | line[length] = '\0'; | |
514 | ||
515 | // Log the line | |
e33387d3 MT |
516 | if (callback) { |
517 | int r = callback(jail->pakfire, data, priority, line, length); | |
616f1fca MT |
518 | if (r) { |
519 | ERROR(jail->pakfire, "The logging callback returned an error: %d\n", r); | |
520 | return r; | |
521 | } | |
522 | } | |
523 | ||
524 | // Remove line from buffer | |
525 | memmove(buffer->data, buffer->data + length, buffer->used - length); | |
526 | buffer->used -= length; | |
527 | } | |
528 | ||
529 | return 0; | |
530 | } | |
531 | ||
195fe455 MT |
532 | static int pakfire_jail_setup_pipe(struct pakfire_jail* jail, int (*fds)[2], const int flags) { |
533 | int r = pipe2(*fds, flags); | |
534 | if (r < 0) { | |
535 | ERROR(jail->pakfire, "Could not setup pipe: %m\n"); | |
536 | return 1; | |
537 | } | |
538 | ||
539 | return 0; | |
540 | } | |
541 | ||
542 | static void pakfire_jail_close_pipe(struct pakfire_jail* jail, int fds[2]) { | |
543 | for (unsigned int i = 0; i < 2; i++) | |
544 | if (fds[i]) | |
545 | close(fds[i]); | |
546 | } | |
547 | ||
e33387d3 MT |
548 | /* |
549 | This is a convenience function to fetch the reading end of a pipe and | |
550 | closes the write end. | |
551 | */ | |
552 | static int pakfire_jail_get_pipe(struct pakfire_jail* jail, int (*fds)[2]) { | |
553 | // Give the variables easier names to avoid confusion | |
554 | int* fd_read = &(*fds)[0]; | |
555 | int* fd_write = &(*fds)[1]; | |
556 | ||
557 | // Close the write end of the pipe | |
558 | if (*fd_write) { | |
559 | close(*fd_write); | |
560 | *fd_write = 0; | |
561 | } | |
562 | ||
563 | // Return the read end | |
564 | return *fd_read; | |
565 | } | |
566 | ||
d853213d | 567 | static int pakfire_jail_wait(struct pakfire_jail* jail, struct pakfire_jail_exec* ctx) { |
616f1fca MT |
568 | int epollfd = -1; |
569 | struct epoll_event ev; | |
570 | struct epoll_event events[EPOLL_MAX_EVENTS]; | |
571 | int r = 0; | |
572 | ||
573 | // Fetch file descriptors from context | |
e33387d3 MT |
574 | const int stdout = pakfire_jail_get_pipe(jail, &ctx->pipes.stdout); |
575 | const int stderr = pakfire_jail_get_pipe(jail, &ctx->pipes.stderr); | |
d853213d | 576 | const int pidfd = ctx->pidfd; |
616f1fca | 577 | |
e33387d3 MT |
578 | // Logging |
579 | const int log_INFO = pakfire_jail_get_pipe(jail, &ctx->pipes.log_INFO); | |
580 | const int log_ERROR = pakfire_jail_get_pipe(jail, &ctx->pipes.log_ERROR); | |
581 | const int log_DEBUG = pakfire_jail_get_pipe(jail, &ctx->pipes.log_DEBUG); | |
7ebfb7cb MT |
582 | |
583 | // Make a list of all file descriptors we are interested in | |
d853213d | 584 | int fds[] = { |
e33387d3 | 585 | stdout, stderr, pidfd, log_INFO, log_ERROR, log_DEBUG, |
616f1fca MT |
586 | }; |
587 | ||
588 | // Setup epoll | |
589 | epollfd = epoll_create1(0); | |
590 | if (epollfd < 0) { | |
591 | ERROR(jail->pakfire, "Could not initialize epoll(): %m\n"); | |
592 | r = 1; | |
d853213d | 593 | goto ERROR; |
616f1fca MT |
594 | } |
595 | ||
e068b964 | 596 | ev.events = EPOLLIN|EPOLLHUP; |
616f1fca MT |
597 | |
598 | // Turn file descriptors into non-blocking mode and add them to epoll() | |
3aad8c0d | 599 | for (unsigned int i = 0; i < sizeof(fds) / sizeof(*fds); i++) { |
616f1fca MT |
600 | int fd = fds[i]; |
601 | ||
d853213d MT |
602 | // Skip fds which were not initialized |
603 | if (fd <= 0) | |
604 | continue; | |
605 | ||
616f1fca MT |
606 | ev.data.fd = fd; |
607 | ||
608 | if (epoll_ctl(epollfd, EPOLL_CTL_ADD, fd, &ev) < 0) { | |
609 | ERROR(jail->pakfire, "Could not add file descriptor %d to epoll(): %m\n", fd); | |
610 | r = 1; | |
d853213d | 611 | goto ERROR; |
616f1fca MT |
612 | } |
613 | } | |
614 | ||
615 | int ended = 0; | |
616 | ||
617 | // Loop for as long as the process is alive | |
618 | while (!ended) { | |
616f1fca MT |
619 | int num = epoll_wait(epollfd, events, EPOLL_MAX_EVENTS, -1); |
620 | if (num < 1) { | |
621 | // Ignore if epoll_wait() has been interrupted | |
622 | if (errno == EINTR) | |
623 | continue; | |
624 | ||
625 | ERROR(jail->pakfire, "epoll_wait() failed: %m\n"); | |
626 | r = 1; | |
627 | ||
d853213d | 628 | goto ERROR; |
616f1fca MT |
629 | } |
630 | ||
616f1fca | 631 | for (int i = 0; i < num; i++) { |
e068b964 | 632 | int e = events[i].events; |
616f1fca MT |
633 | int fd = events[i].data.fd; |
634 | ||
e33387d3 MT |
635 | struct pakfire_log_buffer* buffer = NULL; |
636 | pakfire_jail_log_callback callback = NULL; | |
637 | void* data = NULL; | |
638 | int priority; | |
639 | ||
e068b964 MT |
640 | // Check if there is any data to be read |
641 | if (e & EPOLLIN) { | |
642 | // Handle any changes to the PIDFD | |
643 | if (fd == pidfd) { | |
644 | // Call waidid() and store the result | |
645 | r = waitid(P_PIDFD, ctx->pidfd, &ctx->status, WEXITED); | |
646 | if (r) { | |
647 | ERROR(jail->pakfire, "waitid() failed: %m\n"); | |
648 | goto ERROR; | |
649 | } | |
d853213d | 650 | |
e068b964 MT |
651 | // Mark that we have ended so that we will process the remaining |
652 | // events from epoll() now, but won't restart the outer loop. | |
653 | ended = 1; | |
654 | continue; | |
d853213d | 655 | |
e068b964 MT |
656 | // Handle logging messages |
657 | } else if (fd == log_INFO) { | |
658 | buffer = &ctx->buffers.log_INFO; | |
659 | priority = LOG_INFO; | |
e33387d3 | 660 | |
e068b964 | 661 | callback = pakfire_jail_default_log_callback; |
e33387d3 | 662 | |
e068b964 MT |
663 | } else if (fd == log_ERROR) { |
664 | buffer = &ctx->buffers.log_ERROR; | |
665 | priority = LOG_ERR; | |
e33387d3 | 666 | |
e068b964 | 667 | callback = pakfire_jail_default_log_callback; |
e33387d3 | 668 | |
e068b964 MT |
669 | } else if (fd == log_DEBUG) { |
670 | buffer = &ctx->buffers.log_DEBUG; | |
671 | priority = LOG_DEBUG; | |
e33387d3 | 672 | |
e068b964 | 673 | callback = pakfire_jail_default_log_callback; |
e33387d3 | 674 | |
e068b964 MT |
675 | // Handle anything from the log pipes |
676 | } else if (fd == stdout) { | |
677 | buffer = &ctx->buffers.stdout; | |
678 | priority = LOG_INFO; | |
616f1fca | 679 | |
e068b964 MT |
680 | callback = jail->log_callback; |
681 | data = jail->log_data; | |
e33387d3 | 682 | |
e068b964 MT |
683 | } else if (fd == stderr) { |
684 | buffer = &ctx->buffers.stderr; | |
685 | priority = LOG_ERR; | |
616f1fca | 686 | |
e068b964 MT |
687 | callback = jail->log_callback; |
688 | data = jail->log_data; | |
e33387d3 | 689 | |
e068b964 MT |
690 | } else { |
691 | DEBUG(jail->pakfire, "Received invalid file descriptor %d\n", fd); | |
692 | continue; | |
693 | } | |
694 | ||
695 | // Handle log event | |
696 | r = pakfire_jail_handle_log(jail, ctx, priority, fd, buffer, callback, data); | |
697 | if (r) | |
698 | goto ERROR; | |
616f1fca MT |
699 | } |
700 | ||
e068b964 MT |
701 | // Check if any file descriptors have been closed |
702 | if (e & EPOLLHUP) { | |
703 | // Remove the file descriptor | |
704 | r = epoll_ctl(epollfd, EPOLL_CTL_DEL, fd, NULL); | |
705 | if (r) { | |
706 | ERROR(jail->pakfire, "Could not remove closed file-descriptor %d: %m\n", fd); | |
707 | goto ERROR; | |
708 | } | |
709 | } | |
616f1fca MT |
710 | } |
711 | } | |
712 | ||
d853213d | 713 | ERROR: |
616f1fca MT |
714 | if (epollfd > 0) |
715 | close(epollfd); | |
716 | ||
717 | return r; | |
718 | } | |
719 | ||
0de6bb30 MT |
720 | static int pakfire_jail_capture_stdout(struct pakfire* pakfire, void* data, int priority, |
721 | const char* line, size_t length) { | |
12b9b39f MT |
722 | char** output = (char**)data; |
723 | int r; | |
0de6bb30 | 724 | |
12b9b39f | 725 | // Append everything from stdout to a buffer |
0de6bb30 | 726 | if (priority == LOG_INFO) { |
12b9b39f MT |
727 | r = asprintf(output, "%s%s", (output && *output) ? *output : "", line); |
728 | if (r < 0) | |
0de6bb30 | 729 | return 1; |
0de6bb30 MT |
730 | return 0; |
731 | } | |
732 | ||
733 | // Send everything else to the default logger | |
734 | return pakfire_jail_default_log_callback(pakfire, NULL, priority, line, length); | |
735 | } | |
736 | ||
980b15af MT |
737 | // Capabilities |
738 | ||
739 | static int pakfire_jail_drop_capabilities(struct pakfire_jail* jail) { | |
740 | const int capabilities[] = { | |
741 | // Deny access to the kernel's audit system | |
742 | CAP_AUDIT_CONTROL, | |
743 | CAP_AUDIT_READ, | |
744 | CAP_AUDIT_WRITE, | |
745 | ||
746 | // Deny suspending block devices | |
747 | CAP_BLOCK_SUSPEND, | |
748 | ||
749 | // Deny any stuff with BPF | |
750 | CAP_BPF, | |
751 | ||
752 | // Deny checkpoint restore | |
753 | CAP_CHECKPOINT_RESTORE, | |
754 | ||
755 | // Deny opening files by inode number (open_by_handle_at) | |
756 | CAP_DAC_READ_SEARCH, | |
757 | ||
758 | // Deny setting SUID bits | |
759 | CAP_FSETID, | |
760 | ||
761 | // Deny locking more memory | |
762 | CAP_IPC_LOCK, | |
763 | ||
764 | // Deny modifying any Apparmor/SELinux/SMACK configuration | |
765 | CAP_MAC_ADMIN, | |
766 | CAP_MAC_OVERRIDE, | |
767 | ||
768 | // Deny creating any special devices | |
769 | CAP_MKNOD, | |
770 | ||
771 | // Deny setting any capabilities | |
772 | CAP_SETFCAP, | |
773 | ||
774 | // Deny reading from syslog | |
775 | CAP_SYSLOG, | |
776 | ||
777 | // Deny any admin actions (mount, sethostname, ...) | |
778 | CAP_SYS_ADMIN, | |
779 | ||
780 | // Deny rebooting the system | |
781 | CAP_SYS_BOOT, | |
782 | ||
783 | // Deny loading kernel modules | |
784 | CAP_SYS_MODULE, | |
785 | ||
786 | // Deny setting nice level | |
787 | CAP_SYS_NICE, | |
788 | ||
789 | // Deny access to /proc/kcore, /dev/mem, /dev/kmem | |
790 | CAP_SYS_RAWIO, | |
791 | ||
792 | // Deny circumventing any resource limits | |
793 | CAP_SYS_RESOURCE, | |
794 | ||
795 | // Deny setting the system time | |
796 | CAP_SYS_TIME, | |
797 | ||
798 | // Deny playing with suspend | |
799 | CAP_WAKE_ALARM, | |
800 | ||
801 | 0, | |
802 | }; | |
803 | ||
804 | DEBUG(jail->pakfire, "Dropping capabilities...\n"); | |
805 | ||
806 | size_t num_caps = 0; | |
807 | int r; | |
808 | ||
809 | // Drop any capabilities | |
810 | for (const int* cap = capabilities; *cap; cap++) { | |
811 | r = prctl(PR_CAPBSET_DROP, *cap, 0, 0, 0); | |
812 | if (r) { | |
813 | ERROR(jail->pakfire, "Could not drop capability %d: %m\n", *cap); | |
814 | return r; | |
815 | } | |
816 | ||
817 | num_caps++; | |
818 | } | |
819 | ||
820 | // Fetch any capabilities | |
821 | cap_t caps = cap_get_proc(); | |
822 | if (!caps) { | |
823 | ERROR(jail->pakfire, "Could not read capabilities: %m\n"); | |
824 | return 1; | |
825 | } | |
826 | ||
827 | /* | |
828 | Set inheritable capabilities | |
829 | ||
830 | This ensures that no processes will be able to gain any of the listed | |
831 | capabilities again. | |
832 | */ | |
833 | r = cap_set_flag(caps, CAP_INHERITABLE, num_caps, capabilities, CAP_CLEAR); | |
834 | if (r) { | |
835 | ERROR(jail->pakfire, "cap_set_flag() failed: %m\n"); | |
836 | goto ERROR; | |
837 | } | |
838 | ||
839 | // Restore capabilities | |
840 | r = cap_set_proc(caps); | |
841 | if (r) { | |
842 | ERROR(jail->pakfire, "Could not restore capabilities: %m\n"); | |
843 | goto ERROR; | |
844 | } | |
845 | ||
846 | ERROR: | |
847 | if (caps) | |
848 | cap_free(caps); | |
849 | ||
850 | return r; | |
851 | } | |
852 | ||
739d5b57 MT |
853 | // Syscall Filter |
854 | ||
855 | static int pakfire_jail_limit_syscalls(struct pakfire_jail* jail) { | |
856 | const int syscalls[] = { | |
857 | // The kernel's keyring isn't namespaced | |
858 | SCMP_SYS(keyctl), | |
859 | SCMP_SYS(add_key), | |
860 | SCMP_SYS(request_key), | |
861 | ||
862 | // Disable userfaultfd | |
863 | SCMP_SYS(userfaultfd), | |
864 | ||
865 | // Disable perf which could leak a lot of information about the host | |
866 | SCMP_SYS(perf_event_open), | |
867 | ||
868 | 0, | |
869 | }; | |
870 | int r = 1; | |
871 | ||
872 | DEBUG(jail->pakfire, "Applying syscall filter...\n"); | |
873 | ||
874 | // Setup a syscall filter which allows everything by default | |
875 | scmp_filter_ctx ctx = seccomp_init(SCMP_ACT_ALLOW); | |
876 | if (!ctx) { | |
877 | ERROR(jail->pakfire, "Could not setup seccomp filter: %m\n"); | |
878 | goto ERROR; | |
879 | } | |
880 | ||
881 | // All all syscalls | |
882 | for (const int* syscall = syscalls; *syscall; syscall++) { | |
883 | r = seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), *syscall, 0); | |
884 | if (r) { | |
885 | ERROR(jail->pakfire, "Could not configure syscall %d: %m\n", *syscall); | |
886 | goto ERROR; | |
887 | } | |
888 | } | |
889 | ||
890 | // Load syscall filter into the kernel | |
891 | r = seccomp_load(ctx); | |
892 | if (r) { | |
893 | ERROR(jail->pakfire, "Could not load syscall filter into the kernel: %m\n"); | |
894 | goto ERROR; | |
895 | } | |
896 | ||
897 | ERROR: | |
898 | if (ctx) | |
899 | seccomp_release(ctx); | |
900 | ||
901 | return r; | |
902 | } | |
903 | ||
cc6e2264 MT |
904 | // Mountpoints |
905 | ||
061223f7 | 906 | PAKFIRE_EXPORT int pakfire_jail_bind(struct pakfire_jail* jail, |
cc6e2264 MT |
907 | const char* source, const char* target, int flags) { |
908 | struct pakfire_jail_mountpoint* mp = NULL; | |
909 | int r; | |
910 | ||
911 | // Check if there is any space left | |
912 | if (jail->num_mountpoints >= MAX_MOUNTPOINTS) { | |
913 | errno = ENOSPC; | |
914 | return 1; | |
915 | } | |
916 | ||
917 | // Check for valid inputs | |
918 | if (!source || !target) { | |
919 | errno = EINVAL; | |
920 | return 1; | |
921 | } | |
922 | ||
923 | // Select the next free slot | |
924 | mp = &jail->mountpoints[jail->num_mountpoints]; | |
925 | ||
926 | // Copy source | |
927 | r = pakfire_string_set(mp->source, source); | |
a60955af | 928 | if (r) { |
cc6e2264 | 929 | ERROR(jail->pakfire, "Could not copy source: %m\n"); |
a60955af | 930 | return r; |
cc6e2264 MT |
931 | } |
932 | ||
933 | // Copy target | |
934 | r = pakfire_string_set(mp->target, target); | |
a60955af | 935 | if (r) { |
cc6e2264 | 936 | ERROR(jail->pakfire, "Could not copy target: %m\n"); |
a60955af | 937 | return r; |
cc6e2264 MT |
938 | } |
939 | ||
940 | // Copy flags | |
941 | mp->flags = flags; | |
942 | ||
943 | // Increment counter | |
944 | jail->num_mountpoints++; | |
945 | ||
946 | return 0; | |
947 | } | |
948 | ||
949 | /* | |
950 | Mounts everything that we require in the new namespace | |
951 | */ | |
952 | static int pakfire_jail_mount(struct pakfire_jail* jail) { | |
953 | struct pakfire_jail_mountpoint* mp = NULL; | |
954 | int r; | |
955 | ||
956 | // Mount all default stuff | |
957 | r = pakfire_mount_all(jail->pakfire); | |
958 | if (r) | |
959 | return r; | |
960 | ||
961 | // Mount all custom stuff | |
962 | for (unsigned int i = 0; i < jail->num_mountpoints; i++) { | |
963 | // Fetch mountpoint | |
964 | mp = &jail->mountpoints[i]; | |
965 | ||
966 | // Mount it | |
967 | r = pakfire_bind(jail->pakfire, mp->source, mp->target, mp->flags); | |
968 | if (r) | |
969 | return r; | |
970 | } | |
971 | ||
972 | // Log all mountpoints | |
973 | pakfire_mount_list(jail->pakfire); | |
974 | ||
975 | return 0; | |
976 | } | |
977 | ||
679ee2fa MT |
978 | // UID/GID Mapping |
979 | ||
980 | static int pakfire_jail_write_uidgid_mapping(struct pakfire_jail* jail, | |
a1ff2863 | 981 | const char* path, const struct pakfire_subid* subid) { |
1d4b54ef MT |
982 | return pakfire_file_write(jail->pakfire, path, 0, 0, 0, |
983 | "%d %u %lu\n", 0, subid->id, subid->length); | |
679ee2fa MT |
984 | } |
985 | ||
986 | static int pakfire_jail_setup_uid_mapping(struct pakfire_jail* jail, pid_t pid) { | |
987 | char path[PATH_MAX]; | |
988 | int r; | |
989 | ||
4896e62c MT |
990 | // Skip mapping anything when running on / |
991 | if (pakfire_on_root(jail->pakfire)) | |
992 | return 0; | |
0f7f068b | 993 | |
4896e62c | 994 | // Fetch SUBUID |
a1ff2863 | 995 | const struct pakfire_subid* subuid = pakfire_subuid(jail->pakfire); |
4896e62c MT |
996 | if (!subuid) |
997 | return 1; | |
679ee2fa MT |
998 | |
999 | // Make path | |
1000 | r = pakfire_string_format(path, "/proc/%d/uid_map", pid); | |
a60955af MT |
1001 | if (r) |
1002 | return r; | |
679ee2fa | 1003 | |
4896e62c | 1004 | DEBUG(jail->pakfire, "Mapping UID range (%u - %lu)\n", |
a1ff2863 | 1005 | subuid->id, subuid->id + subuid->length); |
679ee2fa | 1006 | |
a1ff2863 | 1007 | return pakfire_jail_write_uidgid_mapping(jail, path, subuid); |
679ee2fa MT |
1008 | } |
1009 | ||
1010 | static int pakfire_jail_setup_gid_mapping(struct pakfire_jail* jail, pid_t pid) { | |
1011 | char path[PATH_MAX]; | |
1012 | int r; | |
1013 | ||
4896e62c MT |
1014 | // Skip mapping anything when running on / |
1015 | if (pakfire_on_root(jail->pakfire)) | |
1016 | return 0; | |
0f7f068b | 1017 | |
4896e62c | 1018 | // Fetch SUBGID |
a1ff2863 | 1019 | const struct pakfire_subid* subgid = pakfire_subgid(jail->pakfire); |
4896e62c MT |
1020 | if (!subgid) |
1021 | return 1; | |
679ee2fa MT |
1022 | |
1023 | // Make path | |
1024 | r = pakfire_string_format(path, "/proc/%d/gid_map", pid); | |
a60955af MT |
1025 | if (r) |
1026 | return r; | |
679ee2fa | 1027 | |
4896e62c | 1028 | DEBUG(jail->pakfire, "Mapping GID range (%u - %lu)\n", |
a1ff2863 | 1029 | subgid->id, subgid->id + subgid->length); |
679ee2fa | 1030 | |
a1ff2863 | 1031 | return pakfire_jail_write_uidgid_mapping(jail, path, subgid); |
679ee2fa MT |
1032 | } |
1033 | ||
78d7488a MT |
1034 | static int pakfire_jail_setgroups(struct pakfire_jail* jail, pid_t pid) { |
1035 | char path[PATH_MAX]; | |
1036 | int r = 1; | |
1037 | ||
1038 | // Make path | |
1039 | r = pakfire_string_format(path, "/proc/%d/setgroups", pid); | |
a60955af MT |
1040 | if (r) |
1041 | return r; | |
78d7488a MT |
1042 | |
1043 | // Open file for writing | |
1044 | FILE* f = fopen(path, "w"); | |
1045 | if (!f) { | |
1046 | ERROR(jail->pakfire, "Could not open %s for writing: %m\n", path); | |
1047 | goto ERROR; | |
1048 | } | |
1049 | ||
1050 | // Write content | |
1051 | int bytes_written = fprintf(f, "deny\n"); | |
1052 | if (bytes_written <= 0) { | |
1053 | ERROR(jail->pakfire, "Could not write to %s: %m\n", path); | |
1054 | goto ERROR; | |
1055 | } | |
1056 | ||
1057 | r = fclose(f); | |
1058 | f = NULL; | |
1059 | if (r) { | |
1060 | ERROR(jail->pakfire, "Could not close %s: %m\n", path); | |
1061 | goto ERROR; | |
1062 | } | |
1063 | ||
1064 | ERROR: | |
1065 | if (f) | |
1066 | fclose(f); | |
1067 | ||
1068 | return r; | |
1069 | } | |
1070 | ||
43dc0e16 | 1071 | static int pakfire_jail_send_signal(struct pakfire_jail* jail, int fd) { |
743f449e MT |
1072 | const uint64_t val = 1; |
1073 | int r = 0; | |
43dc0e16 MT |
1074 | |
1075 | DEBUG(jail->pakfire, "Sending signal...\n"); | |
1076 | ||
743f449e MT |
1077 | // Write to the file descriptor |
1078 | ssize_t bytes_written = write(fd, &val, sizeof(val)); | |
1079 | if (bytes_written < 0 || (size_t)bytes_written < sizeof(val)) { | |
1080 | ERROR(jail->pakfire, "Could not send signal: %m\n"); | |
1081 | r = 1; | |
1082 | } | |
1083 | ||
1084 | // Close the file descriptor | |
43dc0e16 MT |
1085 | close(fd); |
1086 | ||
743f449e | 1087 | return r; |
43dc0e16 MT |
1088 | } |
1089 | ||
1090 | static int pakfire_jail_wait_for_signal(struct pakfire_jail* jail, int fd) { | |
743f449e MT |
1091 | uint64_t val = 0; |
1092 | int r = 0; | |
43dc0e16 MT |
1093 | |
1094 | DEBUG(jail->pakfire, "Waiting for signal...\n"); | |
1095 | ||
743f449e MT |
1096 | ssize_t bytes_read = read(fd, &val, sizeof(val)); |
1097 | if (bytes_read < 0 || (size_t)bytes_read < sizeof(val)) { | |
1098 | ERROR(jail->pakfire, "Error waiting for signal: %m\n"); | |
1099 | r = 1; | |
1100 | } | |
1101 | ||
1102 | // Close the file descriptor | |
43dc0e16 MT |
1103 | close(fd); |
1104 | ||
743f449e | 1105 | return r; |
43dc0e16 MT |
1106 | } |
1107 | ||
679ee2fa MT |
1108 | /* |
1109 | Performs the initialisation that needs to happen in the parent part | |
1110 | */ | |
f7d240a7 | 1111 | static int pakfire_jail_parent(struct pakfire_jail* jail, struct pakfire_jail_exec* ctx) { |
679ee2fa MT |
1112 | int r; |
1113 | ||
d3d3d7b7 MT |
1114 | // Write "deny" to /proc/PID/setgroups |
1115 | r = pakfire_jail_setgroups(jail, ctx->pid); | |
679ee2fa MT |
1116 | if (r) |
1117 | return r; | |
1118 | ||
d3d3d7b7 MT |
1119 | // Setup UID mapping |
1120 | r = pakfire_jail_setup_uid_mapping(jail, ctx->pid); | |
78d7488a MT |
1121 | if (r) |
1122 | return r; | |
1123 | ||
679ee2fa | 1124 | // Setup GID mapping |
616f1fca | 1125 | r = pakfire_jail_setup_gid_mapping(jail, ctx->pid); |
679ee2fa MT |
1126 | if (r) |
1127 | return r; | |
1128 | ||
43dc0e16 MT |
1129 | // Parent has finished initialisation |
1130 | DEBUG(jail->pakfire, "Parent has finished initialization\n"); | |
1131 | ||
1132 | // Send signal to client | |
f7d240a7 | 1133 | r = pakfire_jail_send_signal(jail, ctx->completed_fd); |
43dc0e16 MT |
1134 | if (r) |
1135 | return r; | |
1136 | ||
679ee2fa MT |
1137 | return 0; |
1138 | } | |
1139 | ||
616f1fca | 1140 | static int pakfire_jail_child(struct pakfire_jail* jail, struct pakfire_jail_exec* ctx, |
f7d240a7 | 1141 | const char* argv[]) { |
43dc0e16 MT |
1142 | int r; |
1143 | ||
e33387d3 MT |
1144 | // Redirect any logging to our log pipe |
1145 | pakfire_set_log_callback(jail->pakfire, pakfire_jail_log, &ctx->pipes); | |
0bd84dc1 | 1146 | |
2a7b5e00 MT |
1147 | // Die with parent |
1148 | r = prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0); | |
1149 | if (r) { | |
1150 | ERROR(jail->pakfire, "Could not configure to die with parent: %m\n"); | |
1151 | return 126; | |
1152 | } | |
1153 | ||
cf440db8 MT |
1154 | // Fetch my own PID |
1155 | pid_t pid = getpid(); | |
1156 | ||
1157 | DEBUG(jail->pakfire, "Launched child process in jail with PID %d\n", pid); | |
0bd84dc1 | 1158 | |
b324c1f7 MT |
1159 | // Log argv |
1160 | for (unsigned int i = 0; argv[i]; i++) | |
1161 | DEBUG(jail->pakfire, " argv[%d] = %s\n", i, argv[i]); | |
1162 | ||
43dc0e16 | 1163 | // Wait for the parent to finish initialization |
f7d240a7 | 1164 | r = pakfire_jail_wait_for_signal(jail, ctx->completed_fd); |
43dc0e16 MT |
1165 | if (r) |
1166 | return r; | |
1167 | ||
4f59c39b MT |
1168 | // Perform further initialization |
1169 | ||
1170 | // Fetch UID/GID | |
1171 | uid_t uid = getuid(); | |
1172 | gid_t gid = getgid(); | |
1173 | ||
1174 | // Fetch EUID/EGID | |
1175 | uid_t euid = geteuid(); | |
1176 | gid_t egid = getegid(); | |
1177 | ||
1178 | DEBUG(jail->pakfire, " UID: %d (effective %d)\n", uid, euid); | |
1179 | DEBUG(jail->pakfire, " GID: %d (effective %d)\n", gid, egid); | |
1180 | ||
1181 | // Check if we are (effectively running as root) | |
4f719e21 | 1182 | if (uid || gid || euid || egid) { |
4f59c39b MT |
1183 | ERROR(jail->pakfire, "Child process is not running as root\n"); |
1184 | return 126; | |
1185 | } | |
1186 | ||
1187 | const char* root = pakfire_get_path(jail->pakfire); | |
1188 | const char* arch = pakfire_get_arch(jail->pakfire); | |
1189 | ||
1190 | // Change root (unless root is /) | |
1191 | if (!pakfire_on_root(jail->pakfire)) { | |
1192 | // Mount everything | |
cc6e2264 | 1193 | r = pakfire_jail_mount(jail); |
4f59c39b MT |
1194 | if (r) |
1195 | return r; | |
1196 | ||
4f59c39b MT |
1197 | // Call chroot() |
1198 | r = chroot(root); | |
1199 | if (r) { | |
1200 | ERROR(jail->pakfire, "chroot() to %s failed: %m\n", root); | |
1201 | return 1; | |
1202 | } | |
1203 | ||
1204 | // Change directory to / | |
1205 | r = chdir("/"); | |
1206 | if (r) { | |
1207 | ERROR(jail->pakfire, "chdir() after chroot() failed: %m\n"); | |
1208 | return 1; | |
1209 | } | |
1210 | } | |
1211 | ||
90d92b5c MT |
1212 | // Set personality |
1213 | unsigned long persona = pakfire_arch_personality(arch); | |
1214 | if (persona) { | |
1215 | r = personality(persona); | |
1216 | if (r < 0) { | |
1217 | ERROR(jail->pakfire, "Could not set personality (%x)\n", (unsigned int)persona); | |
1218 | return 1; | |
1219 | } | |
1220 | } | |
1221 | ||
cf440db8 MT |
1222 | // Set nice level |
1223 | if (jail->nice) { | |
1224 | DEBUG(jail->pakfire, "Setting nice level to %d\n", jail->nice); | |
1225 | ||
1226 | r = setpriority(PRIO_PROCESS, pid, jail->nice); | |
1227 | if (r) { | |
1228 | ERROR(jail->pakfire, "Could not set nice level: %m\n"); | |
1229 | return 1; | |
1230 | } | |
1231 | } | |
1232 | ||
e33387d3 MT |
1233 | // Close other end of log pipes |
1234 | close(ctx->pipes.log_INFO[0]); | |
1235 | close(ctx->pipes.log_ERROR[0]); | |
1236 | #ifdef ENABLE_DEBUG | |
1237 | close(ctx->pipes.log_DEBUG[0]); | |
1238 | #endif /* ENABLE_DEBUG */ | |
1239 | ||
7ebfb7cb MT |
1240 | // Connect standard output and error |
1241 | if (ctx->pipes.stdout[1] && ctx->pipes.stderr[1]) { | |
1242 | r = dup2(ctx->pipes.stdout[1], STDOUT_FILENO); | |
1243 | if (r < 0) { | |
1244 | ERROR(jail->pakfire, "Could not connect fd %d to stdout: %m\n", | |
1245 | ctx->pipes.stdout[1]); | |
1246 | ||
1247 | return 1; | |
1248 | } | |
1249 | ||
1250 | r = dup2(ctx->pipes.stderr[1], STDERR_FILENO); | |
1251 | if (r < 0) { | |
1252 | ERROR(jail->pakfire, "Could not connect fd %d to stderr: %m\n", | |
1253 | ctx->pipes.stderr[1]); | |
1254 | ||
1255 | return 1; | |
1256 | } | |
1257 | ||
195fe455 MT |
1258 | // Close the pipe (as we have moved the original file descriptors) |
1259 | pakfire_jail_close_pipe(jail, ctx->pipes.stdout); | |
1260 | pakfire_jail_close_pipe(jail, ctx->pipes.stderr); | |
7ebfb7cb MT |
1261 | } |
1262 | ||
007bc66c MT |
1263 | // Reset open file limit (http://0pointer.net/blog/file-descriptor-limits.html) |
1264 | r = pakfire_rlimit_reset_nofile(jail->pakfire); | |
1265 | if (r) | |
1266 | return r; | |
1267 | ||
980b15af MT |
1268 | // Drop capabilities |
1269 | r = pakfire_jail_drop_capabilities(jail); | |
1270 | if (r) | |
1271 | return r; | |
1272 | ||
739d5b57 MT |
1273 | // Filter syscalls |
1274 | r = pakfire_jail_limit_syscalls(jail); | |
1275 | if (r) | |
1276 | return r; | |
1277 | ||
b3498aeb MT |
1278 | // exec() command |
1279 | r = execvpe(argv[0], (char**)argv, jail->env); | |
1280 | if (r < 0) | |
1281 | ERROR(jail->pakfire, "Could not execve(): %m\n"); | |
1282 | ||
1283 | // Translate errno into regular exit code | |
1284 | switch (errno) { | |
1285 | case ENOENT: | |
1286 | r = 127; | |
1287 | break; | |
1288 | ||
1289 | default: | |
1290 | r = 1; | |
1291 | } | |
1292 | ||
1293 | // We should not get here | |
1294 | return r; | |
0bd84dc1 MT |
1295 | } |
1296 | ||
9f50bf71 | 1297 | // Run a command in the jail |
db4f234f MT |
1298 | static int __pakfire_jail_exec(struct pakfire_jail* jail, const char* argv[], |
1299 | const int interactive) { | |
4f59c39b | 1300 | int exit = -1; |
0bd84dc1 MT |
1301 | int r; |
1302 | ||
b3498aeb MT |
1303 | // Check if argv is valid |
1304 | if (!argv || !argv[0]) { | |
1305 | errno = EINVAL; | |
1306 | return -1; | |
1307 | } | |
1308 | ||
616f1fca MT |
1309 | // Initialize context for this call |
1310 | struct pakfire_jail_exec ctx = { | |
1311 | .pipes = { | |
e33387d3 MT |
1312 | .stdout = { 0, 0 }, |
1313 | .stderr = { 0, 0 }, | |
616f1fca | 1314 | }, |
616f1fca MT |
1315 | }; |
1316 | ||
0bd84dc1 MT |
1317 | DEBUG(jail->pakfire, "Executing jail...\n"); |
1318 | ||
43dc0e16 MT |
1319 | /* |
1320 | Setup a file descriptor which can be used to notify the client that the parent | |
1321 | has completed configuration. | |
1322 | */ | |
f7d240a7 MT |
1323 | ctx.completed_fd = eventfd(0, EFD_CLOEXEC); |
1324 | if (ctx.completed_fd < 0) { | |
43dc0e16 MT |
1325 | ERROR(jail->pakfire, "eventfd() failed: %m\n"); |
1326 | return -1; | |
1327 | } | |
1328 | ||
616f1fca | 1329 | // Create pipes to communicate with child process if we are not running interactively |
58963c75 | 1330 | if (!interactive) { |
616f1fca | 1331 | // stdout |
e33387d3 MT |
1332 | r = pakfire_jail_setup_pipe(jail, &ctx.pipes.stdout, 0); |
1333 | if (r) | |
616f1fca | 1334 | goto ERROR; |
616f1fca MT |
1335 | |
1336 | // stderr | |
e33387d3 MT |
1337 | r = pakfire_jail_setup_pipe(jail, &ctx.pipes.stderr, 0); |
1338 | if (r) | |
616f1fca | 1339 | goto ERROR; |
616f1fca MT |
1340 | } |
1341 | ||
e33387d3 MT |
1342 | // Setup pipes for logging |
1343 | // INFO | |
1344 | r = pakfire_jail_setup_pipe(jail, &ctx.pipes.log_INFO, O_CLOEXEC); | |
1345 | if (r) | |
1346 | goto ERROR; | |
1347 | ||
1348 | // ERROR | |
1349 | r = pakfire_jail_setup_pipe(jail, &ctx.pipes.log_ERROR, O_CLOEXEC); | |
1350 | if (r) | |
1351 | goto ERROR; | |
1352 | ||
1353 | #ifdef ENABLE_DEBUG | |
1354 | // DEBUG | |
1355 | r = pakfire_jail_setup_pipe(jail, &ctx.pipes.log_DEBUG, O_CLOEXEC); | |
1356 | if (r) | |
1357 | goto ERROR; | |
1358 | #endif /* ENABLE_DEBUG */ | |
1359 | ||
0bd84dc1 MT |
1360 | // Configure child process |
1361 | struct clone_args args = { | |
1362 | .flags = | |
1363 | CLONE_NEWCGROUP | | |
1364 | CLONE_NEWIPC | | |
1365 | CLONE_NEWNS | | |
1366 | CLONE_NEWPID | | |
1367 | CLONE_NEWUSER | | |
d853213d | 1368 | CLONE_NEWUTS | |
02fd4f8b | 1369 | CLONE_PIDFD, |
0bd84dc1 | 1370 | .exit_signal = SIGCHLD, |
d853213d | 1371 | .pidfd = (long long unsigned int)&ctx.pidfd, |
0bd84dc1 MT |
1372 | }; |
1373 | ||
aca565fc | 1374 | // Launch the process in a cgroup that is a leaf of the configured cgroup |
02fd4f8b MT |
1375 | if (jail->cgroup) { |
1376 | args.flags |= CLONE_INTO_CGROUP; | |
1377 | ||
ae5201c5 MT |
1378 | // Fetch our UUID |
1379 | const char* uuid = pakfire_jail_uuid(jail); | |
aca565fc MT |
1380 | |
1381 | // Create a temporary cgroup | |
ae5201c5 | 1382 | r = pakfire_cgroup_child(&ctx.cgroup, jail->cgroup, uuid, 0); |
aca565fc MT |
1383 | if (r) { |
1384 | ERROR(jail->pakfire, "Could not create cgroup for jail: %m\n"); | |
1385 | goto ERROR; | |
1386 | } | |
1387 | ||
02fd4f8b | 1388 | // Clone into this cgroup |
aca565fc | 1389 | args.cgroup = pakfire_cgroup_fd(ctx.cgroup); |
02fd4f8b MT |
1390 | } |
1391 | ||
0bd84dc1 | 1392 | // Fork this process |
616f1fca MT |
1393 | ctx.pid = clone3(&args, sizeof(args)); |
1394 | if (ctx.pid < 0) { | |
0bd84dc1 MT |
1395 | ERROR(jail->pakfire, "Could not clone: %m\n"); |
1396 | return -1; | |
1397 | ||
1398 | // Child process | |
616f1fca | 1399 | } else if (ctx.pid == 0) { |
f7d240a7 | 1400 | r = pakfire_jail_child(jail, &ctx, argv); |
0bd84dc1 MT |
1401 | _exit(r); |
1402 | } | |
1403 | ||
679ee2fa | 1404 | // Parent process |
f7d240a7 | 1405 | r = pakfire_jail_parent(jail, &ctx); |
679ee2fa MT |
1406 | if (r) |
1407 | goto ERROR; | |
1408 | ||
616f1fca | 1409 | DEBUG(jail->pakfire, "Waiting for PID %d to finish its work\n", ctx.pid); |
0bd84dc1 | 1410 | |
616f1fca | 1411 | // Read output of the child process |
d853213d MT |
1412 | r = pakfire_jail_wait(jail, &ctx); |
1413 | if (r) | |
1414 | goto ERROR; | |
0bd84dc1 | 1415 | |
d853213d MT |
1416 | // Handle exit status |
1417 | switch (ctx.status.si_code) { | |
1418 | case CLD_EXITED: | |
1419 | DEBUG(jail->pakfire, "The child process exited with code %d\n", | |
1420 | ctx.status.si_status); | |
616f1fca | 1421 | |
d853213d MT |
1422 | // Pass exit code |
1423 | exit = ctx.status.si_status; | |
1424 | break; | |
0bd84dc1 | 1425 | |
d853213d MT |
1426 | case CLD_KILLED: |
1427 | case CLD_DUMPED: | |
1428 | ERROR(jail->pakfire, "The child process was killed\n"); | |
1429 | break; | |
0bd84dc1 | 1430 | |
d853213d MT |
1431 | // Log anything else |
1432 | default: | |
1433 | ERROR(jail->pakfire, "Unknown child exit code: %d\n", ctx.status.si_code); | |
1434 | break; | |
0bd84dc1 MT |
1435 | } |
1436 | ||
679ee2fa | 1437 | ERROR: |
aca565fc MT |
1438 | // Destroy the temporary cgroup (if any) |
1439 | if (ctx.cgroup) { | |
6b7cf275 MT |
1440 | // Read cgroup stats |
1441 | r = pakfire_cgroup_stat(ctx.cgroup, &ctx.cgroup_stats); | |
1442 | if (r) { | |
1443 | ERROR(jail->pakfire, "Could not read cgroup stats: %m\n"); | |
1444 | } else { | |
1445 | pakfire_cgroup_stat_dump(ctx.cgroup, &ctx.cgroup_stats); | |
1446 | } | |
1447 | ||
aca565fc MT |
1448 | pakfire_cgroup_destroy(ctx.cgroup); |
1449 | pakfire_cgroup_unref(ctx.cgroup); | |
1450 | } | |
1451 | ||
616f1fca | 1452 | // Close any file descriptors |
e33387d3 MT |
1453 | pakfire_jail_close_pipe(jail, ctx.pipes.stdout); |
1454 | pakfire_jail_close_pipe(jail, ctx.pipes.stderr); | |
d853213d MT |
1455 | if (ctx.pidfd) |
1456 | close(ctx.pidfd); | |
e33387d3 MT |
1457 | pakfire_jail_close_pipe(jail, ctx.pipes.log_INFO); |
1458 | pakfire_jail_close_pipe(jail, ctx.pipes.log_ERROR); | |
1459 | pakfire_jail_close_pipe(jail, ctx.pipes.log_DEBUG); | |
616f1fca | 1460 | |
4f59c39b | 1461 | return exit; |
9f50bf71 | 1462 | } |
a45ed6b0 | 1463 | |
6ce56f90 | 1464 | PAKFIRE_EXPORT int pakfire_jail_exec(struct pakfire_jail* jail, |
12b9b39f | 1465 | const char* argv[], char** output) { |
0de6bb30 MT |
1466 | int r; |
1467 | ||
1468 | // Store logging callback | |
1469 | pakfire_jail_log_callback log_callback = jail->log_callback; | |
1470 | void* log_data = jail->log_data; | |
1471 | ||
1472 | // Capture output if requested by user | |
1473 | if (output) | |
1474 | pakfire_jail_set_log_callback(jail, pakfire_jail_capture_stdout, output); | |
1475 | ||
1476 | // Run exec() | |
db4f234f | 1477 | r = __pakfire_jail_exec(jail, argv, 0); |
0de6bb30 MT |
1478 | |
1479 | // Restore log callback | |
1480 | pakfire_jail_set_log_callback(jail, log_callback, log_data); | |
1481 | ||
1482 | return r; | |
1483 | } | |
1484 | ||
db4f234f MT |
1485 | static int pakfire_jail_exec_interactive( |
1486 | struct pakfire_jail* jail, const char* argv[]) { | |
1487 | int r; | |
1488 | ||
1489 | // Setup interactive stuff | |
1490 | r = pakfire_jail_setup_interactive_env(jail); | |
1491 | if (r) | |
1492 | return r; | |
1493 | ||
1494 | return __pakfire_jail_exec(jail, argv, 1); | |
1495 | } | |
1496 | ||
6ce56f90 | 1497 | PAKFIRE_EXPORT int pakfire_jail_exec_script(struct pakfire_jail* jail, |
12b9b39f | 1498 | const char* script, const size_t size, const char* args[], char** output) { |
a45ed6b0 MT |
1499 | char path[PATH_MAX]; |
1500 | const char** argv = NULL; | |
1501 | int r; | |
1502 | ||
1503 | const char* root = pakfire_get_path(jail->pakfire); | |
1504 | ||
1505 | // Write the scriptlet to disk | |
1506 | r = pakfire_path_join(path, root, "pakfire-script.XXXXXX"); | |
56796f84 | 1507 | if (r) |
a45ed6b0 MT |
1508 | goto ERROR; |
1509 | ||
1510 | // Open a temporary file | |
1511 | int fd = mkstemp(path); | |
1512 | if (fd < 0) { | |
1513 | ERROR(jail->pakfire, "Could not open a temporary file: %m\n"); | |
1514 | r = 1; | |
1515 | goto ERROR; | |
1516 | } | |
1517 | ||
1518 | DEBUG(jail->pakfire, "Writing script to %s:\n%.*s\n", path, (int)size, script); | |
1519 | ||
1520 | // Write data | |
1521 | ssize_t bytes_written = write(fd, script, size); | |
1522 | if (bytes_written < (ssize_t)size) { | |
1523 | ERROR(jail->pakfire, "Could not write script to file %s: %m\n", path); | |
1524 | r = 1; | |
1525 | goto ERROR; | |
1526 | } | |
1527 | ||
1528 | // Make the script executable | |
1529 | r = fchmod(fd, S_IRUSR|S_IWUSR|S_IXUSR); | |
1530 | if (r) { | |
1531 | ERROR(jail->pakfire, "Could not set executable permissions on %s: %m\n", path); | |
1532 | goto ERROR; | |
1533 | } | |
1534 | ||
1535 | // Close file | |
1536 | r = close(fd); | |
1537 | if (r) { | |
1538 | ERROR(jail->pakfire, "Could not close script file %s: %m\n", path); | |
1539 | r = 1; | |
1540 | goto ERROR; | |
1541 | } | |
1542 | ||
1543 | // Count how many arguments were passed | |
1544 | unsigned int argc = 1; | |
1545 | if (args) { | |
1546 | for (const char** arg = args; *arg; arg++) | |
1547 | argc++; | |
1548 | } | |
1549 | ||
1550 | argv = calloc(argc + 1, sizeof(*argv)); | |
1551 | if (!argv) { | |
1552 | ERROR(jail->pakfire, "Could not allocate argv: %m\n"); | |
1553 | goto ERROR; | |
1554 | } | |
1555 | ||
1556 | // Set command | |
1557 | argv[0] = (root) ? pakfire_path_relpath(root, path) : path; | |
1558 | ||
1559 | // Copy args | |
1560 | for (unsigned int i = 1; i < argc; i++) | |
1561 | argv[i] = args[i-1]; | |
1562 | ||
1563 | // Run the script | |
0de6bb30 | 1564 | r = pakfire_jail_exec(jail, argv, output); |
a45ed6b0 MT |
1565 | |
1566 | ERROR: | |
1567 | if (argv) | |
1568 | free(argv); | |
1569 | ||
1570 | // Remove script from disk | |
1571 | if (*path) | |
1572 | unlink(path); | |
1573 | ||
1574 | return r; | |
1575 | } | |
82df3c77 MT |
1576 | |
1577 | /* | |
1578 | A convenience function that creates a new jail, runs the given command and destroys | |
1579 | the jail again. | |
1580 | */ | |
12b9b39f | 1581 | int pakfire_jail_run(struct pakfire* pakfire, const char* argv[], int flags, char** output) { |
82df3c77 MT |
1582 | struct pakfire_jail* jail = NULL; |
1583 | int r; | |
1584 | ||
1585 | // Create a new jail | |
1586 | r = pakfire_jail_create(&jail, pakfire, flags); | |
1587 | if (r) | |
1588 | goto ERROR; | |
1589 | ||
1590 | // Execute the command | |
0de6bb30 | 1591 | r = pakfire_jail_exec(jail, argv, output); |
82df3c77 MT |
1592 | |
1593 | ERROR: | |
1594 | if (jail) | |
1595 | pakfire_jail_unref(jail); | |
1596 | ||
1597 | return r; | |
1598 | } | |
4f688bd8 MT |
1599 | |
1600 | int pakfire_jail_run_script(struct pakfire* pakfire, | |
12b9b39f | 1601 | const char* script, const size_t length, const char* argv[], int flags, char** output) { |
4f688bd8 MT |
1602 | struct pakfire_jail* jail = NULL; |
1603 | int r; | |
1604 | ||
1605 | // Create a new jail | |
1606 | r = pakfire_jail_create(&jail, pakfire, flags); | |
1607 | if (r) | |
1608 | goto ERROR; | |
1609 | ||
1610 | // Execute the command | |
0de6bb30 | 1611 | r = pakfire_jail_exec_script(jail, script, length, argv, output); |
4f688bd8 MT |
1612 | |
1613 | ERROR: | |
1614 | if (jail) | |
1615 | pakfire_jail_unref(jail); | |
1616 | ||
1617 | return r; | |
1618 | } | |
e43489f7 | 1619 | |
5f6e42a2 | 1620 | int pakfire_jail_shell(struct pakfire_jail* jail) { |
e43489f7 MT |
1621 | const char* argv[] = { |
1622 | "/bin/bash", "--login", NULL, | |
1623 | }; | |
1624 | ||
1625 | // Execute /bin/bash | |
db4f234f | 1626 | return pakfire_jail_exec_interactive(jail, argv); |
e43489f7 MT |
1627 | } |
1628 | ||
1629 | int pakfire_jail_ldconfig(struct pakfire* pakfire) { | |
1630 | char path[PATH_MAX]; | |
1631 | ||
1632 | const char* ldconfig = "/sbin/ldconfig"; | |
1633 | ||
1634 | // Check if ldconfig exists before calling it to avoid overhead | |
77e26129 MT |
1635 | int r = pakfire_path(pakfire, path, "%s", ldconfig); |
1636 | if (r) | |
1637 | return r; | |
e43489f7 MT |
1638 | |
1639 | // Check if ldconfig is executable | |
1640 | r = access(path, X_OK); | |
1641 | if (r) { | |
1642 | DEBUG(pakfire, "%s is not executable. Skipping...\n", ldconfig); | |
1643 | return 0; | |
1644 | } | |
1645 | ||
1646 | const char* argv[] = { | |
1647 | ldconfig, NULL, | |
1648 | }; | |
1649 | ||
1650 | // Run ldconfig | |
1651 | return pakfire_jail_run(pakfire, argv, 0, NULL); | |
1652 | } |