Blob


1 /*
2 * Copyright (c) 2021 Omar Polo <op@omarpolo.com>
3 *
4 * Permission to use, copy, modify, and distribute this software for any
5 * purpose with or without fee is hereby granted, provided that the above
6 * copyright notice and this permission notice appear in all copies.
7 *
8 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
9 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
10 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
11 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
12 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
13 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
14 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
15 */
17 #include "gmid.h"
19 #if DISABLE_SANDBOX
21 #warning "Sandbox disabled! Please report issues upstream instead of disabling the sandbox."
23 void
24 sandbox_server_process(void)
25 {
26 return;
27 }
29 void
30 sandbox_executor_process(void)
31 {
32 log_notice(NULL, "Sandbox disabled! "
33 "Please report issues upstream instead of disabling the sandbox.");
34 }
36 void
37 sandbox_logger_process(void)
38 {
39 return;
40 }
42 #elif defined(__FreeBSD__)
44 #include <sys/capsicum.h>
46 void
47 sandbox_server_process(void)
48 {
49 if (cap_enter() == -1)
50 fatal("cap_enter");
51 }
53 void
54 sandbox_executor_process(void)
55 {
56 /*
57 * We cannot capsicum the executor process because it needs to
58 * fork(2)+execve(2) cgi scripts
59 */
60 return;
61 }
63 void
64 sandbox_logger_process(void)
65 {
66 if (cap_enter() == -1)
67 fatal("cap_enter");
68 }
70 #elif defined(__linux__)
72 #include <sys/prctl.h>
73 #include <sys/syscall.h>
74 #include <sys/syscall.h>
75 #include <sys/types.h>
77 #include <linux/audit.h>
78 #include <linux/filter.h>
79 #include <linux/seccomp.h>
81 #include <errno.h>
82 #include <fcntl.h>
83 #include <stddef.h>
84 #include <stdio.h>
85 #include <string.h>
87 #if HAVE_LANDLOCK
88 # include "landlock_shim.h"
89 #endif
91 /* uncomment to enable debugging. ONLY FOR DEVELOPMENT */
92 /* #define SC_DEBUG */
94 #ifdef SC_DEBUG
95 # define SC_FAIL SECCOMP_RET_TRAP
96 #else
97 # define SC_FAIL SECCOMP_RET_KILL
98 #endif
100 #if (BYTE_ORDER == LITTLE_ENDIAN)
101 # define SC_ARG_LO 0
102 # define SC_ARG_HI sizeof(uint32_t)
103 #elif (BYTE_ORDER == BIG_ENDIAN)
104 # define SC_ARG_LO sizeof(uint32_t)
105 # define SC_ARG_HI 0
106 #else
107 # error "Uknown endian"
108 #endif
110 /* make the filter more readable */
111 #define SC_ALLOW(nr) \
112 BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_##nr, 0, 1), \
113 BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW)
115 /*
116 * SC_ALLOW_ARG and the SECCOMP_AUDIT_ARCH below are courtesy of
117 * https://roy.marples.name/git/dhcpcd/blob/HEAD:/src/privsep-linux.c
118 */
119 #define SC_ALLOW_ARG(_nr, _arg, _val) \
120 BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, (_nr), 0, 6), \
121 BPF_STMT(BPF_LD + BPF_W + BPF_ABS, \
122 offsetof(struct seccomp_data, args[(_arg)]) + SC_ARG_LO), \
123 BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, \
124 ((_val) & 0xffffffff), 0, 3), \
125 BPF_STMT(BPF_LD + BPF_W + BPF_ABS, \
126 offsetof(struct seccomp_data, args[(_arg)]) + SC_ARG_HI), \
127 BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, \
128 (((uint32_t)((uint64_t)(_val) >> 32)) & 0xffffffff), 0, 1), \
129 BPF_STMT(BPF_RET + BPF_K, SECCOMP_RET_ALLOW), \
130 BPF_STMT(BPF_LD + BPF_W + BPF_ABS, \
131 offsetof(struct seccomp_data, nr))
133 /*
134 * I personally find this quite nutty. Why can a system header not
135 * define a default for this?
136 */
137 #if defined(__i386__)
138 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_I386
139 #elif defined(__x86_64__)
140 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_X86_64
141 #elif defined(__arc__)
142 # if defined(__A7__)
143 # if (BYTE_ORDER == LITTLE_ENDIAN)
144 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_ARCOMPACT
145 # else
146 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_ARCOMPACTBE
147 # endif
148 # elif defined(__HS__)
149 # if (BYTE_ORDER == LITTLE_ENDIAN)
150 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_ARCV2
151 # else
152 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_ARCV2BE
153 # endif
154 # else
155 # error "Platform does not support seccomp filter yet"
156 # endif
157 #elif defined(__arm__)
158 # ifndef EM_ARM
159 # define EM_ARM 40
160 # endif
161 # if (BYTE_ORDER == LITTLE_ENDIAN)
162 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_ARM
163 # else
164 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_ARMEB
165 # endif
166 #elif defined(__aarch64__)
167 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_AARCH64
168 #elif defined(__alpha__)
169 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_ALPHA
170 #elif defined(__hppa__)
171 # if defined(__LP64__)
172 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_PARISC64
173 # else
174 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_PARISC
175 # endif
176 #elif defined(__ia64__)
177 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_IA64
178 #elif defined(__microblaze__)
179 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_MICROBLAZE
180 #elif defined(__m68k__)
181 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_M68K
182 #elif defined(__mips__)
183 # if defined(__MIPSEL__)
184 # if defined(__LP64__)
185 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_MIPSEL64
186 # else
187 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_MIPSEL
188 # endif
189 # elif defined(__LP64__)
190 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_MIPS64
191 # else
192 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_MIPS
193 # endif
194 #elif defined(__nds32__)
195 # if (BYTE_ORDER == LITTLE_ENDIAN)
196 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_NDS32
197 #else
198 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_NDS32BE
199 #endif
200 #elif defined(__nios2__)
201 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_NIOS2
202 #elif defined(__or1k__)
203 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_OPENRISC
204 #elif defined(__powerpc64__)
205 # if (BYTE_ORDER == LITTLE_ENDIAN)
206 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_PPC64LE
207 # else
208 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_PPC64
209 # endif
210 #elif defined(__powerpc__)
211 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_PPC
212 #elif defined(__riscv)
213 # if defined(__LP64__)
214 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_RISCV64
215 # else
216 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_RISCV32
217 # endif
218 #elif defined(__s390x__)
219 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_S390X
220 #elif defined(__s390__)
221 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_S390
222 #elif defined(__sh__)
223 # if defined(__LP64__)
224 # if (BYTE_ORDER == LITTLE_ENDIAN)
225 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_SHEL64
226 # else
227 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_SH64
228 # endif
229 # else
230 # if (BYTE_ORDER == LITTLE_ENDIAN)
231 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_SHEL
232 # else
233 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_SH
234 # endif
235 # endif
236 #elif defined(__sparc__)
237 # if defined(__arch64__)
238 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_SPARC64
239 # else
240 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_SPARC
241 # endif
242 #elif defined(__xtensa__)
243 # define SECCOMP_AUDIT_ARCH AUDIT_ARCH_XTENSA
244 #else
245 # error "Platform does not support seccomp filter yet"
246 #endif
248 static struct sock_filter filter[] = {
249 /* load the *current* architecture */
250 BPF_STMT(BPF_LD | BPF_W | BPF_ABS,
251 (offsetof(struct seccomp_data, arch))),
252 /* ensure it's the same that we've been compiled on */
253 BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K,
254 SECCOMP_AUDIT_ARCH, 1, 0),
255 /* if not, kill the program */
256 BPF_STMT(BPF_RET | BPF_K, SC_FAIL),
258 /* load the syscall number */
259 BPF_STMT(BPF_LD | BPF_W | BPF_ABS,
260 (offsetof(struct seccomp_data, nr))),
262 #ifdef __NR_accept
263 SC_ALLOW(accept),
264 #endif
265 #ifdef __NR_accept4
266 SC_ALLOW(accept4),
267 #endif
268 #ifdef __NR_brk
269 SC_ALLOW(brk),
270 #endif
271 #ifdef __NR_clock_gettime
272 SC_ALLOW(clock_gettime),
273 #endif
274 #if defined(__x86_64__) && defined(__ILP32__) && defined(__X32_SYSCALL_BIT)
275 SECCOMP_ALLOW(__NR_clock_gettime & ~__X32_SYSCALL_BIT),
276 #endif
277 #ifdef __NR_clock_gettime64
278 SC_ALLOW(clock_gettime64),
279 #endif
280 #ifdef __NR_close
281 SC_ALLOW(close),
282 #endif
283 #ifdef __NR_epoll_ctl
284 SC_ALLOW(epoll_ctl),
285 #endif
286 #ifdef __NR_epoll_pwait
287 SC_ALLOW(epoll_pwait),
288 #endif
289 #ifdef __NR_epoll_wait
290 SC_ALLOW(epoll_wait),
291 #endif
292 #ifdef __NR_exit
293 SC_ALLOW(exit),
294 #endif
295 #ifdef __NR_exit_group
296 SC_ALLOW(exit_group),
297 #endif
298 #ifdef __NR_fcntl
299 SC_ALLOW(fcntl),
300 #endif
301 #ifdef __NR_fcntl64
302 SC_ALLOW(fcntl64),
303 #endif
304 #ifdef __NR_fstat
305 SC_ALLOW(fstat),
306 #endif
307 #ifdef __NR_fstat64
308 SC_ALLOW(fstat64),
309 #endif
310 #ifdef __NR_getdents64
311 SC_ALLOW(getdents64),
312 #endif
313 #ifdef __NR_getpid
314 SC_ALLOW(getpid),
315 #endif
316 #ifdef __NR_getrandom
317 SC_ALLOW(getrandom),
318 #endif
319 #ifdef __NR_gettimeofday
320 SC_ALLOW(gettimeofday),
321 #endif
322 #ifdef __NR_ioctl
323 /* allow ioctl only on fd 1, glibc doing stuff? */
324 SC_ALLOW_ARG(__NR_ioctl, 0, 1),
325 #endif
326 #ifdef __NR_lseek
327 SC_ALLOW(lseek),
328 #endif
329 #ifdef __NR_madvise
330 SC_ALLOW(madvise),
331 #endif
332 #ifdef __NR_mmap
333 SC_ALLOW(mmap),
334 #endif
335 #ifdef __NR_mmap2
336 SC_ALLOW(mmap2),
337 #endif
338 #ifdef __NR_munmap
339 SC_ALLOW(munmap),
340 #endif
341 #ifdef __NR_newfstatat
342 SC_ALLOW(newfstatat),
343 #endif
344 #ifdef __NR_oldfstat
345 SC_ALLOW(oldfstat),
346 #endif
347 #ifdef __NR_openat
348 SC_ALLOW(openat),
349 #endif
350 #ifdef __NR_prlimit64
351 SC_ALLOW(prlimit64),
352 #endif
353 #ifdef __NR_read
354 SC_ALLOW(read),
355 #endif
356 #ifdef __NR_recvmsg
357 SC_ALLOW(recvmsg),
358 #endif
359 #ifdef __NR_readv
360 SC_ALLOW(readv),
361 #endif
362 #ifdef __NR_rt_sigaction
363 SC_ALLOW(rt_sigaction),
364 #endif
365 #ifdef __NR_rt_sigreturn
366 SC_ALLOW(rt_sigreturn),
367 #endif
368 #ifdef __NR_sendmsg
369 SC_ALLOW(sendmsg),
370 #endif
371 #ifdef __NR_statx
372 SC_ALLOW(statx),
373 #endif
374 #ifdef __NR_write
375 SC_ALLOW(write),
376 #endif
377 #ifdef __NR_writev
378 SC_ALLOW(writev),
379 #endif
381 /* disallow everything else */
382 BPF_STMT(BPF_RET | BPF_K, SC_FAIL),
383 };
385 #ifdef SC_DEBUG
387 #include <signal.h>
388 #include <unistd.h>
390 static void
391 sandbox_seccomp_violation(int signum, siginfo_t *info, void *ctx)
393 (void)signum;
394 (void)ctx;
396 fprintf(stderr, "%s: unexpected system call (arch:0x%x,syscall:%d @ %p)\n",
397 __func__, info->si_arch, info->si_syscall, info->si_call_addr);
398 _exit(1);
401 static void
402 sandbox_seccomp_catch_sigsys(void)
404 struct sigaction act;
405 sigset_t mask;
407 memset(&act, 0, sizeof(act));
408 sigemptyset(&mask);
409 sigaddset(&mask, SIGSYS);
411 act.sa_sigaction = &sandbox_seccomp_violation;
412 act.sa_flags = SA_SIGINFO;
413 if (sigaction(SIGSYS, &act, NULL) == -1)
414 fatal("%s: sigaction(SIGSYS): %s",
415 __func__, strerror(errno));
417 if (sigprocmask(SIG_UNBLOCK, &mask, NULL) == -1)
418 fatal("%s: sigprocmask(SIGSYS): %s\n",
419 __func__, strerror(errno));
421 #endif /* SC_DEBUG */
423 #if HAVE_LANDLOCK
424 static int
425 server_landlock(void)
427 int fd, err;
428 struct vhost *h;
429 struct location *l;
431 /*
432 * These are all the actions that we want to either allow or
433 * disallow. Things like LANDLOCK_ACCESS_FS_EXECUTE are
434 * omitted because are already handled by seccomp.
435 */
436 struct landlock_ruleset_attr ruleset_attr = {
437 .handled_access_fs = LANDLOCK_ACCESS_FS_WRITE_FILE |
438 LANDLOCK_ACCESS_FS_READ_FILE |
439 LANDLOCK_ACCESS_FS_READ_DIR |
440 LANDLOCK_ACCESS_FS_MAKE_CHAR |
441 LANDLOCK_ACCESS_FS_MAKE_DIR |
442 LANDLOCK_ACCESS_FS_MAKE_REG |
443 LANDLOCK_ACCESS_FS_MAKE_SOCK |
444 LANDLOCK_ACCESS_FS_MAKE_FIFO |
445 LANDLOCK_ACCESS_FS_MAKE_BLOCK |
446 LANDLOCK_ACCESS_FS_MAKE_SYM,
447 };
449 /*
450 * These are all the actions allowed for the root directories
451 * of the vhosts. All the other rules mentioned in
452 * ruleset_attr and omitted here are implicitly disallowed.
453 */
454 struct landlock_path_beneath_attr path_beneath = {
455 .allowed_access = LANDLOCK_ACCESS_FS_READ_FILE |
456 LANDLOCK_ACCESS_FS_READ_DIR,
457 };
459 fd = landlock_create_ruleset(&ruleset_attr, sizeof(ruleset_attr), 0);
460 if (fd == -1) {
461 switch (errno) {
462 case ENOSYS:
463 fatal("%s: failed to create ruleset. "
464 "Landlock doesn't seem to be supported by the "
465 "current kernel.", __func__);
466 case EOPNOTSUPP:
467 log_warn(NULL, "%s: failed to create ruleset. "
468 "Landlock seems to be currently disabled; "
469 "continuing without it.", __func__);
470 return -1;
471 default:
472 fatal("%s: failed to create ruleset: %s",
473 __func__, strerror(errno));
477 TAILQ_FOREACH(h, &hosts, vhosts) {
478 TAILQ_FOREACH(l, &h->locations, locations) {
479 if (l->dir == NULL)
480 continue;
482 path_beneath.parent_fd = open(l->dir, O_PATH);
483 if (path_beneath.parent_fd == -1)
484 fatal("%s: can't open %s for landlock: %s",
485 __func__, l->dir, strerror(errno));
487 err = landlock_add_rule(fd, LANDLOCK_RULE_PATH_BENEATH,
488 &path_beneath, 0);
489 if (err)
490 fatal("%s: landlock_add_rule(%s) failed: %s",
491 __func__, l->dir, strerror(errno));
493 close(path_beneath.parent_fd);
497 return fd;
499 #endif
501 void
502 sandbox_server_process(void)
504 int fd;
505 struct sock_fprog prog = {
506 .len = (unsigned short) (sizeof(filter) / sizeof(filter[0])),
507 .filter = filter,
508 };
510 #ifdef SC_DEBUG
511 sandbox_seccomp_catch_sigsys();
512 #endif
514 #if HAVE_LANDLOCK
515 log_warn(NULL, "loading landlock...");
516 fd = server_landlock();
517 #else
518 (void)fd; /* avoid unused var warning */
519 #endif
521 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) == -1)
522 fatal("%s: prctl(PR_SET_NO_NEW_PRIVS): %s",
523 __func__, strerror(errno));
525 #if HAVE_LANDLOCK
526 if (fd != -1) {
527 if (landlock_restrict_self(fd, 0))
528 fatal("%s: landlock_restrict_self: %s",
529 __func__, strerror(errno));
530 close(fd);
532 #endif
534 if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog) == -1)
535 fatal("%s: prctl(PR_SET_SECCOMP): %s\n",
536 __func__, strerror(errno));
539 void
540 sandbox_executor_process(void)
542 /*
543 * We cannot use seccomp for the executor process because we
544 * don't know what the child will do. Also, our filter will
545 * be inherited so the child cannot set its own seccomp
546 * policy.
547 */
548 return;
551 void
552 sandbox_logger_process(void)
554 /*
555 * To be honest, here we could use a seccomp policy to only
556 * allow writev(2) and memory allocations.
557 */
558 return;
561 #elif defined(__OpenBSD__)
563 #include <unistd.h>
565 void
566 sandbox_server_process(void)
568 struct vhost *h;
569 struct location *l;
571 TAILQ_FOREACH(h, &hosts, vhosts) {
572 TAILQ_FOREACH(l, &h->locations, locations) {
573 if (l->dir == NULL)
574 continue;
576 if (unveil(l->dir, "r") == -1)
577 fatal("unveil %s for domain %s",
578 l->dir,
579 h->domain);
583 if (pledge("stdio recvfd rpath inet", NULL) == -1)
584 fatal("pledge");
587 void
588 sandbox_executor_process(void)
590 struct vhost *h;
591 struct location *l;
592 struct fcgi *f;
593 size_t i;
595 TAILQ_FOREACH(h, &hosts, vhosts) {
596 TAILQ_FOREACH(l, &h->locations, locations) {
597 if (l->dir == NULL)
598 continue;
600 /* r so we can chdir into the directory */
601 if (unveil(l->dir, "rx") == -1)
602 fatal("unveil %s for domain %s",
603 l->dir, h->domain);
607 for (i = 0; i < FCGI_MAX; i++) {
608 f = &fcgi[i];
609 if (f->path != NULL) {
610 if (unveil(f->path, "rw") == -1)
611 fatal("unveil %s", f->path);
614 if (f->prog != NULL) {
615 if (unveil(f->prog, "rx") == -1)
616 fatal("unveil %s", f->prog);
620 /*
621 * rpath: to chdir into the correct directory
622 * proc exec: CGI
623 * dns inet unix: FastCGI
624 */
625 if (pledge("stdio rpath sendfd proc exec dns inet unix", NULL))
626 err(1, "pledge");
629 void
630 sandbox_logger_process(void)
632 if (pledge("stdio recvfd", NULL) == -1)
633 err(1, "pledge");
636 #else
638 #warning "No sandbox method known for this OS"
640 void
641 sandbox_server_process(void)
643 return;
646 void
647 sandbox_executor_process(void)
649 log_notice(NULL, "no sandbox method known for this OS");
652 void
653 sandbox_logger_process(void)
655 return;
658 #endif