xref: /freebsd/sys/kern/sys_capability.c (revision 79b015331632b33f22bc8d8fef136c2672f2241f) 
1 /*-
2  * Copyright (c) 2008-2011 Robert N. M. Watson
3  * Copyright (c) 2010-2011 Jonathan Anderson
4  * Copyright (c) 2012 FreeBSD Foundation
5  * All rights reserved.
6  *
7  * This software was developed at the University of Cambridge Computer
8  * Laboratory with support from a grant from Google, Inc.
9  *
10  * Portions of this software were developed by Pawel Jakub Dawidek under
11  * sponsorship from the FreeBSD Foundation.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 /*
36  * FreeBSD kernel capability facility.
37  *
38  * Two kernel features are implemented here: capability mode, a sandboxed mode
39  * of execution for processes, and capabilities, a refinement on file
40  * descriptors that allows fine-grained control over operations on the file
41  * descriptor.  Collectively, these allow processes to run in the style of a
42  * historic "capability system" in which they can use only resources
43  * explicitly delegated to them.  This model is enforced by restricting access
44  * to global namespaces in capability mode.
45  *
46  * Capabilities wrap other file descriptor types, binding them to a constant
47  * rights mask set when the capability is created.  New capabilities may be
48  * derived from existing capabilities, but only if they have the same or a
49  * strict subset of the rights on the original capability.
50  *
51  * System calls permitted in capability mode are defined in capabilities.conf;
52  * calls must be carefully audited for safety to ensure that they don't allow
53  * escape from a sandbox.  Some calls permit only a subset of operations in
54  * capability mode -- for example, shm_open(2) is limited to creating
55  * anonymous, rather than named, POSIX shared memory objects.
56  */
57 
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD$");
60 
61 #include "opt_capsicum.h"
62 #include "opt_ktrace.h"
63 
64 #include <sys/param.h>
65 #include <sys/capsicum.h>
66 #include <sys/file.h>
67 #include <sys/filedesc.h>
68 #include <sys/kernel.h>
69 #include <sys/limits.h>
70 #include <sys/lock.h>
71 #include <sys/mutex.h>
72 #include <sys/proc.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysproto.h>
75 #include <sys/sysctl.h>
76 #include <sys/systm.h>
77 #include <sys/ucred.h>
78 #include <sys/uio.h>
79 #include <sys/ktrace.h>
80 
81 #include <security/audit/audit.h>
82 
83 #include <vm/uma.h>
84 #include <vm/vm.h>
85 
86 #ifdef CAPABILITY_MODE
87 
88 FEATURE(security_capability_mode, "Capsicum Capability Mode");
89 
90 /*
91  * System call to enter capability mode for the process.
92  */
93 int
94 sys_cap_enter(struct thread *td, struct cap_enter_args *uap)
95 {
96 	struct ucred *newcred, *oldcred;
97 	struct proc *p;
98 
99 	if (IN_CAPABILITY_MODE(td))
100 		return (0);
101 
102 	newcred = crget();
103 	p = td->td_proc;
104 	PROC_LOCK(p);
105 	oldcred = crcopysafe(p, newcred);
106 	newcred->cr_flags |= CRED_FLAG_CAPMODE;
107 	proc_set_cred(p, newcred);
108 	PROC_UNLOCK(p);
109 	crfree(oldcred);
110 	return (0);
111 }
112 
113 /*
114  * System call to query whether the process is in capability mode.
115  */
116 int
117 sys_cap_getmode(struct thread *td, struct cap_getmode_args *uap)
118 {
119 	u_int i;
120 
121 	i = IN_CAPABILITY_MODE(td) ? 1 : 0;
122 	return (copyout(&i, uap->modep, sizeof(i)));
123 }
124 
125 #else /* !CAPABILITY_MODE */
126 
127 int
128 sys_cap_enter(struct thread *td, struct cap_enter_args *uap)
129 {
130 
131 	return (ENOSYS);
132 }
133 
134 int
135 sys_cap_getmode(struct thread *td, struct cap_getmode_args *uap)
136 {
137 
138 	return (ENOSYS);
139 }
140 
141 #endif /* CAPABILITY_MODE */
142 
143 #ifdef CAPABILITIES
144 
145 FEATURE(security_capabilities, "Capsicum Capabilities");
146 
147 MALLOC_DECLARE(M_FILECAPS);
148 
149 static inline int
150 _cap_check(const cap_rights_t *havep, const cap_rights_t *needp,
151     enum ktr_cap_fail_type type)
152 {
153 	int i;
154 
155 	for (i = 0; i < nitems(havep->cr_rights); i++) {
156 		if (!cap_rights_contains(havep, needp)) {
157 #ifdef KTRACE
158 			if (KTRPOINT(curthread, KTR_CAPFAIL))
159 				ktrcapfail(type, needp, havep);
160 #endif
161 			return (ENOTCAPABLE);
162 		}
163 	}
164 	return (0);
165 }
166 
167 /*
168  * Test whether a capability grants the requested rights.
169  */
170 int
171 cap_check(const cap_rights_t *havep, const cap_rights_t *needp)
172 {
173 
174 	return (_cap_check(havep, needp, CAPFAIL_NOTCAPABLE));
175 }
176 
177 /*
178  * Convert capability rights into VM access flags.
179  */
180 u_char
181 cap_rights_to_vmprot(cap_rights_t *havep)
182 {
183 	u_char maxprot;
184 
185 	maxprot = VM_PROT_NONE;
186 	if (cap_rights_is_set(havep, CAP_MMAP_R))
187 		maxprot |= VM_PROT_READ;
188 	if (cap_rights_is_set(havep, CAP_MMAP_W))
189 		maxprot |= VM_PROT_WRITE;
190 	if (cap_rights_is_set(havep, CAP_MMAP_X))
191 		maxprot |= VM_PROT_EXECUTE;
192 
193 	return (maxprot);
194 }
195 
196 /*
197  * Extract rights from a capability for monitoring purposes -- not for use in
198  * any other way, as we want to keep all capability permission evaluation in
199  * this one file.
200  */
201 
202 cap_rights_t *
203 cap_rights_fde(struct filedescent *fde)
204 {
205 
206 	return (&fde->fde_rights);
207 }
208 
209 cap_rights_t *
210 cap_rights(struct filedesc *fdp, int fd)
211 {
212 
213 	return (cap_rights_fde(&fdp->fd_ofiles[fd]));
214 }
215 
216 int
217 kern_cap_rights_limit(struct thread *td, int fd, cap_rights_t *rights)
218 {
219 	struct filedesc *fdp;
220 	int error;
221 
222 	fdp = td->td_proc->p_fd;
223 	FILEDESC_XLOCK(fdp);
224 	if (fget_locked(fdp, fd) == NULL) {
225 		FILEDESC_XUNLOCK(fdp);
226 		return (EBADF);
227 	}
228 	error = _cap_check(cap_rights(fdp, fd), rights, CAPFAIL_INCREASE);
229 	if (error == 0) {
230 		fdp->fd_ofiles[fd].fde_rights = *rights;
231 		if (!cap_rights_is_set(rights, CAP_IOCTL)) {
232 			free(fdp->fd_ofiles[fd].fde_ioctls, M_FILECAPS);
233 			fdp->fd_ofiles[fd].fde_ioctls = NULL;
234 			fdp->fd_ofiles[fd].fde_nioctls = 0;
235 		}
236 		if (!cap_rights_is_set(rights, CAP_FCNTL))
237 			fdp->fd_ofiles[fd].fde_fcntls = 0;
238 	}
239 	FILEDESC_XUNLOCK(fdp);
240 	return (error);
241 }
242 
243 /*
244  * System call to limit rights of the given capability.
245  */
246 int
247 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
248 {
249 	cap_rights_t rights;
250 	int error, version;
251 
252 	cap_rights_init(&rights);
253 
254 	error = copyin(uap->rightsp, &rights, sizeof(rights.cr_rights[0]));
255 	if (error != 0)
256 		return (error);
257 	version = CAPVER(&rights);
258 	if (version != CAP_RIGHTS_VERSION_00)
259 		return (EINVAL);
260 
261 	error = copyin(uap->rightsp, &rights,
262 	    sizeof(rights.cr_rights[0]) * CAPARSIZE(&rights));
263 	if (error != 0)
264 		return (error);
265 	/* Check for race. */
266 	if (CAPVER(&rights) != version)
267 		return (EINVAL);
268 
269 	if (!cap_rights_is_valid(&rights))
270 		return (EINVAL);
271 
272 	if (version != CAP_RIGHTS_VERSION) {
273 		rights.cr_rights[0] &= ~(0x3ULL << 62);
274 		rights.cr_rights[0] |= ((uint64_t)CAP_RIGHTS_VERSION << 62);
275 	}
276 #ifdef KTRACE
277 	if (KTRPOINT(td, KTR_STRUCT))
278 		ktrcaprights(&rights);
279 #endif
280 
281 	AUDIT_ARG_FD(uap->fd);
282 	AUDIT_ARG_RIGHTS(&rights);
283 	return (kern_cap_rights_limit(td, uap->fd, &rights));
284 }
285 
286 /*
287  * System call to query the rights mask associated with a capability.
288  */
289 int
290 sys___cap_rights_get(struct thread *td, struct __cap_rights_get_args *uap)
291 {
292 	struct filedesc *fdp;
293 	cap_rights_t rights;
294 	int error, fd, i, n;
295 
296 	if (uap->version != CAP_RIGHTS_VERSION_00)
297 		return (EINVAL);
298 
299 	fd = uap->fd;
300 
301 	AUDIT_ARG_FD(fd);
302 
303 	fdp = td->td_proc->p_fd;
304 	FILEDESC_SLOCK(fdp);
305 	if (fget_locked(fdp, fd) == NULL) {
306 		FILEDESC_SUNLOCK(fdp);
307 		return (EBADF);
308 	}
309 	rights = *cap_rights(fdp, fd);
310 	FILEDESC_SUNLOCK(fdp);
311 	n = uap->version + 2;
312 	if (uap->version != CAPVER(&rights)) {
313 		/*
314 		 * For older versions we need to check if the descriptor
315 		 * doesn't contain rights not understood by the caller.
316 		 * If it does, we have to return an error.
317 		 */
318 		for (i = n; i < CAPARSIZE(&rights); i++) {
319 			if ((rights.cr_rights[i] & ~(0x7FULL << 57)) != 0)
320 				return (EINVAL);
321 		}
322 	}
323 	error = copyout(&rights, uap->rightsp, sizeof(rights.cr_rights[0]) * n);
324 #ifdef KTRACE
325 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
326 		ktrcaprights(&rights);
327 #endif
328 	return (error);
329 }
330 
331 /*
332  * Test whether a capability grants the given ioctl command.
333  * If descriptor doesn't have CAP_IOCTL, then ioctls list is empty and
334  * ENOTCAPABLE will be returned.
335  */
336 int
337 cap_ioctl_check(struct filedesc *fdp, int fd, u_long cmd)
338 {
339 	u_long *cmds;
340 	ssize_t ncmds;
341 	long i;
342 
343 	FILEDESC_LOCK_ASSERT(fdp);
344 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
345 	    ("%s: invalid fd=%d", __func__, fd));
346 
347 	ncmds = fdp->fd_ofiles[fd].fde_nioctls;
348 	if (ncmds == -1)
349 		return (0);
350 
351 	cmds = fdp->fd_ofiles[fd].fde_ioctls;
352 	for (i = 0; i < ncmds; i++) {
353 		if (cmds[i] == cmd)
354 			return (0);
355 	}
356 
357 	return (ENOTCAPABLE);
358 }
359 
360 /*
361  * Check if the current ioctls list can be replaced by the new one.
362  */
363 static int
364 cap_ioctl_limit_check(struct filedesc *fdp, int fd, const u_long *cmds,
365     size_t ncmds)
366 {
367 	u_long *ocmds;
368 	ssize_t oncmds;
369 	u_long i;
370 	long j;
371 
372 	oncmds = fdp->fd_ofiles[fd].fde_nioctls;
373 	if (oncmds == -1)
374 		return (0);
375 	if (oncmds < (ssize_t)ncmds)
376 		return (ENOTCAPABLE);
377 
378 	ocmds = fdp->fd_ofiles[fd].fde_ioctls;
379 	for (i = 0; i < ncmds; i++) {
380 		for (j = 0; j < oncmds; j++) {
381 			if (cmds[i] == ocmds[j])
382 				break;
383 		}
384 		if (j == oncmds)
385 			return (ENOTCAPABLE);
386 	}
387 
388 	return (0);
389 }
390 
391 int
392 kern_cap_ioctls_limit(struct thread *td, int fd, u_long *cmds, size_t ncmds)
393 {
394 	struct filedesc *fdp;
395 	u_long *ocmds;
396 	int error;
397 
398 	AUDIT_ARG_FD(fd);
399 
400 	fdp = td->td_proc->p_fd;
401 	FILEDESC_XLOCK(fdp);
402 
403 	if (fget_locked(fdp, fd) == NULL) {
404 		error = EBADF;
405 		goto out;
406 	}
407 
408 	error = cap_ioctl_limit_check(fdp, fd, cmds, ncmds);
409 	if (error != 0)
410 		goto out;
411 
412 	ocmds = fdp->fd_ofiles[fd].fde_ioctls;
413 	fdp->fd_ofiles[fd].fde_ioctls = cmds;
414 	fdp->fd_ofiles[fd].fde_nioctls = ncmds;
415 
416 	cmds = ocmds;
417 	error = 0;
418 out:
419 	FILEDESC_XUNLOCK(fdp);
420 	free(cmds, M_FILECAPS);
421 	return (error);
422 }
423 
424 int
425 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
426 {
427 	u_long *cmds;
428 	size_t ncmds;
429 	int error;
430 
431 	ncmds = uap->ncmds;
432 
433 	if (ncmds > 256)	/* XXX: Is 256 sane? */
434 		return (EINVAL);
435 
436 	if (ncmds == 0) {
437 		cmds = NULL;
438 	} else {
439 		cmds = malloc(sizeof(cmds[0]) * ncmds, M_FILECAPS, M_WAITOK);
440 		error = copyin(uap->cmds, cmds, sizeof(cmds[0]) * ncmds);
441 		if (error != 0) {
442 			free(cmds, M_FILECAPS);
443 			return (error);
444 		}
445 	}
446 
447 	return (kern_cap_ioctls_limit(td, uap->fd, cmds, ncmds));
448 }
449 
450 int
451 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
452 {
453 	struct filedesc *fdp;
454 	struct filedescent *fdep;
455 	u_long *cmds;
456 	size_t maxcmds;
457 	int error, fd;
458 
459 	fd = uap->fd;
460 	cmds = uap->cmds;
461 	maxcmds = uap->maxcmds;
462 
463 	AUDIT_ARG_FD(fd);
464 
465 	fdp = td->td_proc->p_fd;
466 	FILEDESC_SLOCK(fdp);
467 
468 	if (fget_locked(fdp, fd) == NULL) {
469 		error = EBADF;
470 		goto out;
471 	}
472 
473 	/*
474 	 * If all ioctls are allowed (fde_nioctls == -1 && fde_ioctls == NULL)
475 	 * the only sane thing we can do is to not populate the given array and
476 	 * return CAP_IOCTLS_ALL.
477 	 */
478 
479 	fdep = &fdp->fd_ofiles[fd];
480 	if (cmds != NULL && fdep->fde_ioctls != NULL) {
481 		error = copyout(fdep->fde_ioctls, cmds,
482 		    sizeof(cmds[0]) * MIN(fdep->fde_nioctls, maxcmds));
483 		if (error != 0)
484 			goto out;
485 	}
486 	if (fdep->fde_nioctls == -1)
487 		td->td_retval[0] = CAP_IOCTLS_ALL;
488 	else
489 		td->td_retval[0] = fdep->fde_nioctls;
490 
491 	error = 0;
492 out:
493 	FILEDESC_SUNLOCK(fdp);
494 	return (error);
495 }
496 
497 /*
498  * Test whether a capability grants the given fcntl command.
499  */
500 int
501 cap_fcntl_check_fde(struct filedescent *fde, int cmd)
502 {
503 	uint32_t fcntlcap;
504 
505 	fcntlcap = (1 << cmd);
506 	KASSERT((CAP_FCNTL_ALL & fcntlcap) != 0,
507 	    ("Unsupported fcntl=%d.", cmd));
508 
509 	if ((fde->fde_fcntls & fcntlcap) != 0)
510 		return (0);
511 
512 	return (ENOTCAPABLE);
513 }
514 
515 int
516 cap_fcntl_check(struct filedesc *fdp, int fd, int cmd)
517 {
518 
519 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
520 	    ("%s: invalid fd=%d", __func__, fd));
521 
522 	return (cap_fcntl_check_fde(&fdp->fd_ofiles[fd], cmd));
523 }
524 
525 int
526 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
527 {
528 	struct filedesc *fdp;
529 	uint32_t fcntlrights;
530 	int fd;
531 
532 	fd = uap->fd;
533 	fcntlrights = uap->fcntlrights;
534 
535 	AUDIT_ARG_FD(fd);
536 	AUDIT_ARG_FCNTL_RIGHTS(fcntlrights);
537 
538 	if ((fcntlrights & ~CAP_FCNTL_ALL) != 0)
539 		return (EINVAL);
540 
541 	fdp = td->td_proc->p_fd;
542 	FILEDESC_XLOCK(fdp);
543 
544 	if (fget_locked(fdp, fd) == NULL) {
545 		FILEDESC_XUNLOCK(fdp);
546 		return (EBADF);
547 	}
548 
549 	if ((fcntlrights & ~fdp->fd_ofiles[fd].fde_fcntls) != 0) {
550 		FILEDESC_XUNLOCK(fdp);
551 		return (ENOTCAPABLE);
552 	}
553 
554 	fdp->fd_ofiles[fd].fde_fcntls = fcntlrights;
555 	FILEDESC_XUNLOCK(fdp);
556 
557 	return (0);
558 }
559 
560 int
561 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
562 {
563 	struct filedesc *fdp;
564 	uint32_t rights;
565 	int fd;
566 
567 	fd = uap->fd;
568 
569 	AUDIT_ARG_FD(fd);
570 
571 	fdp = td->td_proc->p_fd;
572 	FILEDESC_SLOCK(fdp);
573 	if (fget_locked(fdp, fd) == NULL) {
574 		FILEDESC_SUNLOCK(fdp);
575 		return (EBADF);
576 	}
577 	rights = fdp->fd_ofiles[fd].fde_fcntls;
578 	FILEDESC_SUNLOCK(fdp);
579 
580 	return (copyout(&rights, uap->fcntlrightsp, sizeof(rights)));
581 }
582 
583 #else /* !CAPABILITIES */
584 
585 /*
586  * Stub Capability functions for when options CAPABILITIES isn't compiled
587  * into the kernel.
588  */
589 
590 int
591 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
592 {
593 
594 	return (ENOSYS);
595 }
596 
597 int
598 sys___cap_rights_get(struct thread *td, struct __cap_rights_get_args *uap)
599 {
600 
601 	return (ENOSYS);
602 }
603 
604 int
605 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
606 {
607 
608 	return (ENOSYS);
609 }
610 
611 int
612 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
613 {
614 
615 	return (ENOSYS);
616 }
617 
618 int
619 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
620 {
621 
622 	return (ENOSYS);
623 }
624 
625 int
626 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
627 {
628 
629 	return (ENOSYS);
630 }
631 
632 #endif /* CAPABILITIES */
633