xref: /freebsd/sys/kern/kern_priv.c (revision 96fbe51956c9df8bdb8317413b1c487b14e4ee68)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2006 nCircle Network Security, Inc.
5  * Copyright (c) 2009 Robert N. M. Watson
6  * Copyright (c) 2020 Mariusz Zaborski <oshogbo@FreeBSD.org>
7  * All rights reserved.
8  *
9  * This software was developed by Robert N. M. Watson for the TrustedBSD
10  * Project under contract to nCircle Network Security, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR, NCIRCLE NETWORK SECURITY,
25  * INC., OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
27  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
28  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <sys/param.h>
38 #include <sys/jail.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/mutex.h>
42 #include <sys/sx.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/sdt.h>
46 #include <sys/sysctl.h>
47 #include <sys/systm.h>
48 
49 #include <security/mac/mac_framework.h>
50 
51 /*
52  * `suser_enabled' (which can be set by the security.bsd.suser_enabled
53  * sysctl) determines whether the system 'super-user' policy is in effect.  If
54  * it is nonzero, an effective uid of 0 connotes special privilege,
55  * overriding many mandatory and discretionary protections.  If it is zero,
56  * uid 0 is offered no special privilege in the kernel security policy.
57  * Setting it to zero may seriously impact the functionality of many existing
58  * userland programs, and should not be done without careful consideration of
59  * the consequences.
60  */
61 
62 static bool
63 suser_enabled(struct ucred *cred)
64 {
65 
66 	return (prison_allow(cred, PR_ALLOW_SUSER) ? true : false);
67 }
68 
69 static void inline
70 prison_suser_set(struct prison *pr, int enabled)
71 {
72 
73 	if (enabled) {
74 		pr->pr_allow |= PR_ALLOW_SUSER;
75 	} else {
76 		pr->pr_allow &= ~PR_ALLOW_SUSER;
77 	}
78 }
79 
80 static int
81 sysctl_kern_suser_enabled(SYSCTL_HANDLER_ARGS)
82 {
83 	struct prison *pr, *cpr;
84 	struct ucred *cred;
85 	int descend, error, enabled;
86 
87 	cred = req->td->td_ucred;
88 	enabled = suser_enabled(cred);
89 
90 	error = sysctl_handle_int(oidp, &enabled, 0, req);
91 	if (error || !req->newptr)
92 		return (error);
93 
94 	pr = cred->cr_prison;
95 	sx_slock(&allprison_lock);
96 	mtx_lock(&pr->pr_mtx);
97 
98 	prison_suser_set(pr, enabled);
99 	if (!enabled) {
100 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend) {
101 			prison_suser_set(cpr, 0);
102 		}
103 	}
104 	mtx_unlock(&pr->pr_mtx);
105 	sx_sunlock(&allprison_lock);
106 	return (0);
107 }
108 
109 SYSCTL_PROC(_security_bsd, OID_AUTO, suser_enabled, CTLTYPE_INT |
110     CTLFLAG_RWTUN | CTLFLAG_PRISON | CTLFLAG_MPSAFE, 0, 0,
111     &sysctl_kern_suser_enabled, "I", "Processes with uid 0 have privilege");
112 
113 static int	unprivileged_mlock = 1;
114 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_mlock, CTLFLAG_RWTUN,
115     &unprivileged_mlock, 0, "Allow non-root users to call mlock(2)");
116 
117 static int	unprivileged_read_msgbuf = 1;
118 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_read_msgbuf,
119     CTLFLAG_RW, &unprivileged_read_msgbuf, 0,
120     "Unprivileged processes may read the kernel message buffer");
121 
122 SDT_PROVIDER_DEFINE(priv);
123 SDT_PROBE_DEFINE1(priv, kernel, priv_check, priv__ok, "int");
124 SDT_PROBE_DEFINE1(priv, kernel, priv_check, priv__err, "int");
125 
126 static __always_inline int
127 priv_check_cred_pre(struct ucred *cred, int priv)
128 {
129 	int error;
130 
131 #ifdef MAC
132 	error = mac_priv_check(cred, priv);
133 #else
134 	error = 0;
135 #endif
136 	return (error);
137 }
138 
139 static __always_inline int
140 priv_check_cred_post(struct ucred *cred, int priv, int error, bool handled)
141 {
142 
143 	if (__predict_true(handled))
144 		goto out;
145 	/*
146 	 * Now check with MAC, if enabled, to see if a policy module grants
147 	 * privilege.
148 	 */
149 #ifdef MAC
150 	if (mac_priv_grant(cred, priv) == 0) {
151 		error = 0;
152 		goto out;
153 	}
154 #endif
155 
156 	/*
157 	 * The default is deny, so if no policies have granted it, reject
158 	 * with a privilege error here.
159 	 */
160 	error = EPERM;
161 out:
162 	if (SDT_PROBES_ENABLED()) {
163 		if (error)
164 			SDT_PROBE1(priv, kernel, priv_check, priv__err, priv);
165 		else
166 			SDT_PROBE1(priv, kernel, priv_check, priv__ok, priv);
167 	}
168 	return (error);
169 }
170 
171 /*
172  * Check a credential for privilege.  Lots of good reasons to deny privilege;
173  * only a few to grant it.
174  */
175 int
176 priv_check_cred(struct ucred *cred, int priv)
177 {
178 	int error;
179 
180 	KASSERT(PRIV_VALID(priv), ("priv_check_cred: invalid privilege %d",
181 	    priv));
182 
183 	switch (priv) {
184 	case PRIV_VFS_LOOKUP:
185 		return (priv_check_cred_vfs_lookup(cred));
186 	case PRIV_VFS_GENERATION:
187 		return (priv_check_cred_vfs_generation(cred));
188 	}
189 
190 	/*
191 	 * We first evaluate policies that may deny the granting of
192 	 * privilege unilaterally.
193 	 */
194 	error = priv_check_cred_pre(cred, priv);
195 	if (error)
196 		goto out;
197 
198 	/*
199 	 * Jail policy will restrict certain privileges that may otherwise be
200 	 * be granted.
201 	 */
202 	error = prison_priv_check(cred, priv);
203 	if (error)
204 		goto out;
205 
206 	if (unprivileged_mlock) {
207 		/*
208 		 * Allow unprivileged users to call mlock(2)/munlock(2) and
209 		 * mlockall(2)/munlockall(2).
210 		 */
211 		switch (priv) {
212 		case PRIV_VM_MLOCK:
213 		case PRIV_VM_MUNLOCK:
214 			error = 0;
215 			goto out;
216 		}
217 	}
218 
219 	if (unprivileged_read_msgbuf) {
220 		/*
221 		 * Allow an unprivileged user to read the kernel message
222 		 * buffer.
223 		 */
224 		if (priv == PRIV_MSGBUF) {
225 			error = 0;
226 			goto out;
227 		}
228 	}
229 
230 	/*
231 	 * Having determined if privilege is restricted by various policies,
232 	 * now determine if privilege is granted.  At this point, any policy
233 	 * may grant privilege.  For now, we allow short-circuit boolean
234 	 * evaluation, so may not call all policies.  Perhaps we should.
235 	 *
236 	 * Superuser policy grants privilege based on the effective (or in
237 	 * the case of specific privileges, real) uid being 0.  We allow the
238 	 * superuser policy to be globally disabled, although this is
239 	 * currenty of limited utility.
240 	 */
241 	if (suser_enabled(cred)) {
242 		switch (priv) {
243 		case PRIV_MAXFILES:
244 		case PRIV_MAXPROC:
245 		case PRIV_PROC_LIMIT:
246 			if (cred->cr_ruid == 0) {
247 				error = 0;
248 				goto out;
249 			}
250 			break;
251 		case PRIV_VFS_READ_DIR:
252 			/*
253 			 * Allow PRIV_VFS_READ_DIR for root if we're not in a
254 			 * jail, otherwise deny unless a MAC policy grants it.
255 			 */
256 			if (jailed(cred))
257 				break;
258 			/* FALLTHROUGH */
259 		default:
260 			if (cred->cr_uid == 0) {
261 				error = 0;
262 				goto out;
263 			}
264 			break;
265 		}
266 	}
267 
268 	/*
269 	 * Writes to kernel/physical memory are a typical root-only operation,
270 	 * but non-root users are expected to be able to read it (provided they
271 	 * have permission to access /dev/[k]mem).
272 	 */
273 	if (priv == PRIV_KMEM_READ) {
274 		error = 0;
275 		goto out;
276 	}
277 
278 	/*
279 	 * Allow unprivileged process debugging on a per-jail basis.
280 	 * Do this here instead of prison_priv_check(), so it can also
281 	 * apply to prison0.
282 	 */
283 	if (priv == PRIV_DEBUG_UNPRIV) {
284 		if (prison_allow(cred, PR_ALLOW_UNPRIV_DEBUG)) {
285 			error = 0;
286 			goto out;
287 		}
288 	}
289 
290 	return (priv_check_cred_post(cred, priv, error, false));
291 out:
292 	return (priv_check_cred_post(cred, priv, error, true));
293 }
294 
295 int
296 priv_check(struct thread *td, int priv)
297 {
298 
299 	KASSERT(td == curthread, ("priv_check: td != curthread"));
300 
301 	return (priv_check_cred(td->td_ucred, priv));
302 }
303 
304 static int __noinline
305 priv_check_cred_vfs_lookup_slow(struct ucred *cred)
306 {
307 	int error;
308 
309 	error = priv_check_cred_pre(cred, PRIV_VFS_LOOKUP);
310 	if (error)
311 		goto out;
312 
313 	if (cred->cr_uid == 0 && suser_enabled(cred)) {
314 		error = 0;
315 		goto out;
316 	}
317 
318 	return (priv_check_cred_post(cred, PRIV_VFS_LOOKUP, error, false));
319 out:
320 	return (priv_check_cred_post(cred, PRIV_VFS_LOOKUP, error, true));
321 
322 }
323 
324 int
325 priv_check_cred_vfs_lookup(struct ucred *cred)
326 {
327 	int error;
328 
329 	if (__predict_false(mac_priv_check_fp_flag ||
330 	    mac_priv_grant_fp_flag || SDT_PROBES_ENABLED()))
331 		return (priv_check_cred_vfs_lookup_slow(cred));
332 
333 	error = EPERM;
334 	if (cred->cr_uid == 0 && suser_enabled(cred))
335 		error = 0;
336 	return (error);
337 }
338 
339 int
340 priv_check_cred_vfs_lookup_nomac(struct ucred *cred)
341 {
342 	int error;
343 
344 	if (__predict_false(mac_priv_check_fp_flag ||
345 	    mac_priv_grant_fp_flag || SDT_PROBES_ENABLED()))
346 		return (EAGAIN);
347 
348 	error = EPERM;
349 	if (cred->cr_uid == 0 && suser_enabled(cred))
350 		error = 0;
351 	return (error);
352 }
353 
354 static int __noinline
355 priv_check_cred_vfs_generation_slow(struct ucred *cred)
356 {
357 	int error;
358 
359 	error = priv_check_cred_pre(cred, PRIV_VFS_GENERATION);
360 	if (error)
361 		goto out;
362 
363 	if (jailed(cred)) {
364 		error = EPERM;
365 		goto out;
366 	}
367 
368 	if (cred->cr_uid == 0 && suser_enabled(cred)) {
369 		error = 0;
370 		goto out;
371 	}
372 
373 	return (priv_check_cred_post(cred, PRIV_VFS_GENERATION, error, false));
374 out:
375 	return (priv_check_cred_post(cred, PRIV_VFS_GENERATION, error, true));
376 
377 }
378 
379 int
380 priv_check_cred_vfs_generation(struct ucred *cred)
381 {
382 	int error;
383 
384 	if (__predict_false(mac_priv_check_fp_flag ||
385 	    mac_priv_grant_fp_flag || SDT_PROBES_ENABLED()))
386 		return (priv_check_cred_vfs_generation_slow(cred));
387 
388 	error = EPERM;
389 	if (!jailed(cred) && cred->cr_uid == 0 && suser_enabled(cred))
390 		error = 0;
391 	return (error);
392 }
393