1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause
3  *
4  * Copyright (C) 1994, David Greenman
5  * Copyright (c) 1990, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  * Copyright (C) 2010 Konstantin Belousov <kib@freebsd.org>
8  *
9  * This code is derived from software contributed to Berkeley by
10  * the University of Utah, and William Jolitz.
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  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by the University of
23  *	California, Berkeley and its contributors.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	from: @(#)trap.c	7.4 (Berkeley) 5/13/91
41  */
42 
43 #include "opt_capsicum.h"
44 #include "opt_ktrace.h"
45 
46 __FBSDID("$FreeBSD: stable/12/sys/kern/subr_syscall.c 367080 2020-10-27 13:13:04Z kevans $");
47 
48 #include <sys/capsicum.h>
49 #include <sys/ktr.h>
50 #include <sys/vmmeter.h>
51 #ifdef KTRACE
52 #include <sys/uio.h>
53 #include <sys/ktrace.h>
54 #endif
55 #include <security/audit/audit.h>
56 
57 static inline void
syscallenter(struct thread * td)58 syscallenter(struct thread *td)
59 {
60 	struct proc *p;
61 	struct syscall_args *sa;
62 	int error, traced;
63 
64 	VM_CNT_INC(v_syscall);
65 	p = td->td_proc;
66 	sa = &td->td_sa;
67 
68 	td->td_pticks = 0;
69 	if (td->td_cowgen != p->p_cowgen)
70 		thread_cow_update(td);
71 	traced = (p->p_flag & P_TRACED) != 0;
72 	if (traced || td->td_dbgflags & TDB_USERWR) {
73 		PROC_LOCK(p);
74 		td->td_dbgflags &= ~TDB_USERWR;
75 		if (traced)
76 			td->td_dbgflags |= TDB_SCE;
77 		PROC_UNLOCK(p);
78 	}
79 	error = (p->p_sysent->sv_fetch_syscall_args)(td);
80 #ifdef KTRACE
81 	if (KTRPOINT(td, KTR_SYSCALL))
82 		ktrsyscall(sa->code, sa->narg, sa->args);
83 #endif
84 	KTR_START4(KTR_SYSC, "syscall", syscallname(p, sa->code),
85 	    (uintptr_t)td, "pid:%d", td->td_proc->p_pid, "arg0:%p", sa->args[0],
86 	    "arg1:%p", sa->args[1], "arg2:%p", sa->args[2]);
87 
88 	if (error != 0) {
89 		td->td_errno = error;
90 		goto retval;
91 	}
92 
93 	STOPEVENT(p, S_SCE, sa->narg);
94 	if ((p->p_flag & P_TRACED) != 0) {
95 		PROC_LOCK(p);
96 		if (p->p_ptevents & PTRACE_SCE)
97 			ptracestop((td), SIGTRAP, NULL);
98 		PROC_UNLOCK(p);
99 	}
100 	if ((td->td_dbgflags & TDB_USERWR) != 0) {
101 		/*
102 		 * Reread syscall number and arguments if debugger
103 		 * modified registers or memory.
104 		 */
105 		error = (p->p_sysent->sv_fetch_syscall_args)(td);
106 #ifdef KTRACE
107 		if (KTRPOINT(td, KTR_SYSCALL))
108 			ktrsyscall(sa->code, sa->narg, sa->args);
109 #endif
110 		if (error != 0) {
111 			td->td_errno = error;
112 			goto retval;
113 		}
114 	}
115 
116 #ifdef CAPABILITY_MODE
117 	/*
118 	 * In capability mode, we only allow access to system calls
119 	 * flagged with SYF_CAPENABLED.
120 	 */
121 	if (IN_CAPABILITY_MODE(td) &&
122 	    !(sa->callp->sy_flags & SYF_CAPENABLED)) {
123 		td->td_errno = error = ECAPMODE;
124 		goto retval;
125 	}
126 #endif
127 
128 	error = syscall_thread_enter(td, sa->callp);
129 	if (error != 0) {
130 		td->td_errno = error;
131 		goto retval;
132 	}
133 
134 #ifdef KDTRACE_HOOKS
135 	/* Give the syscall:::entry DTrace probe a chance to fire. */
136 	if (__predict_false(systrace_enabled && sa->callp->sy_entry != 0))
137 		(*systrace_probe_func)(sa, SYSTRACE_ENTRY, 0);
138 #endif
139 
140 	/* Let system calls set td_errno directly. */
141 	td->td_pflags &= ~TDP_NERRNO;
142 
143 	AUDIT_SYSCALL_ENTER(sa->code, td);
144 	error = (sa->callp->sy_call)(td, sa->args);
145 
146 	/*
147 	 * Note that some syscall implementations (e.g., sys_execve)
148 	 * will commit the audit record just before their final return.
149 	 * These were done under the assumption that nothing of interest
150 	 * would happen between their return and here, where we would
151 	 * normally commit the audit record.  These assumptions will
152 	 * need to be revisited should any substantial logic be added
153 	 * above.
154 	 */
155 	AUDIT_SYSCALL_EXIT(error, td);
156 
157 	/* Save the latest error return value. */
158 	if ((td->td_pflags & TDP_NERRNO) == 0)
159 		td->td_errno = error;
160 
161 #ifdef KDTRACE_HOOKS
162 	/* Give the syscall:::return DTrace probe a chance to fire. */
163 	if (__predict_false(systrace_enabled && sa->callp->sy_return != 0))
164 		(*systrace_probe_func)(sa, SYSTRACE_RETURN,
165 		    error ? -1 : td->td_retval[0]);
166 #endif
167 	syscall_thread_exit(td, sa->callp);
168 
169  retval:
170 	KTR_STOP4(KTR_SYSC, "syscall", syscallname(p, sa->code),
171 	    (uintptr_t)td, "pid:%d", td->td_proc->p_pid, "error:%d", error,
172 	    "retval0:%#lx", td->td_retval[0], "retval1:%#lx",
173 	    td->td_retval[1]);
174 	if (traced) {
175 		PROC_LOCK(p);
176 		td->td_dbgflags &= ~TDB_SCE;
177 		PROC_UNLOCK(p);
178 	}
179 	(p->p_sysent->sv_set_syscall_retval)(td, error);
180 }
181 
182 static inline void
syscallret(struct thread * td)183 syscallret(struct thread *td)
184 {
185 	struct proc *p, *p2;
186 	struct syscall_args *sa;
187 	ksiginfo_t ksi;
188 	int traced;
189 
190 	KASSERT((td->td_pflags & TDP_FORKING) == 0,
191 	    ("fork() did not clear TDP_FORKING upon completion"));
192 
193 	p = td->td_proc;
194 	sa = &td->td_sa;
195 	if ((trap_enotcap || (p->p_flag2 & P2_TRAPCAP) != 0) &&
196 	    IN_CAPABILITY_MODE(td)) {
197 		if (td->td_errno == ENOTCAPABLE || td->td_errno == ECAPMODE) {
198 			ksiginfo_init_trap(&ksi);
199 			ksi.ksi_signo = SIGTRAP;
200 			ksi.ksi_errno = td->td_errno;
201 			ksi.ksi_code = TRAP_CAP;
202 			trapsignal(td, &ksi);
203 		}
204 	}
205 
206 	/*
207 	 * Handle reschedule and other end-of-syscall issues
208 	 */
209 	userret(td, td->td_frame);
210 
211 #ifdef KTRACE
212 	if (KTRPOINT(td, KTR_SYSRET)) {
213 		ktrsysret(sa->code, td->td_errno, td->td_retval[0]);
214 	}
215 #endif
216 
217 	if (p->p_flag & P_TRACED) {
218 		traced = 1;
219 		PROC_LOCK(p);
220 		td->td_dbgflags |= TDB_SCX;
221 		PROC_UNLOCK(p);
222 	} else
223 		traced = 0;
224 	/*
225 	 * This works because errno is findable through the
226 	 * register set.  If we ever support an emulation where this
227 	 * is not the case, this code will need to be revisited.
228 	 */
229 	STOPEVENT(p, S_SCX, sa->code);
230 	if (traced || (td->td_dbgflags & (TDB_EXEC | TDB_FORK)) != 0) {
231 		PROC_LOCK(p);
232 		/*
233 		 * If tracing the execed process, trap to the debugger
234 		 * so that breakpoints can be set before the program
235 		 * executes.  If debugger requested tracing of syscall
236 		 * returns, do it now too.
237 		 */
238 		if (traced &&
239 		    ((td->td_dbgflags & (TDB_FORK | TDB_EXEC)) != 0 ||
240 		    (p->p_ptevents & PTRACE_SCX) != 0))
241 			ptracestop(td, SIGTRAP, NULL);
242 		td->td_dbgflags &= ~(TDB_SCX | TDB_EXEC | TDB_FORK);
243 		PROC_UNLOCK(p);
244 	}
245 
246 	if (td->td_pflags & TDP_RFPPWAIT) {
247 		/*
248 		 * Preserve synchronization semantics of vfork.  If
249 		 * waiting for child to exec or exit, fork set
250 		 * P_PPWAIT on child, and there we sleep on our proc
251 		 * (in case of exit).
252 		 *
253 		 * Do it after the ptracestop() above is finished, to
254 		 * not block our debugger until child execs or exits
255 		 * to finish vfork wait.
256 		 */
257 		td->td_pflags &= ~TDP_RFPPWAIT;
258 		p2 = td->td_rfppwait_p;
259 again:
260 		PROC_LOCK(p2);
261 		while (p2->p_flag & P_PPWAIT) {
262 			PROC_LOCK(p);
263 			if (thread_suspend_check_needed()) {
264 				PROC_UNLOCK(p2);
265 				thread_suspend_check(0);
266 				PROC_UNLOCK(p);
267 				goto again;
268 			} else {
269 				PROC_UNLOCK(p);
270 			}
271 			cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
272 		}
273 		PROC_UNLOCK(p2);
274 
275 		if (td->td_dbgflags & TDB_VFORK) {
276 			PROC_LOCK(p);
277 			if (p->p_ptevents & PTRACE_VFORK)
278 				ptracestop(td, SIGTRAP, NULL);
279 			td->td_dbgflags &= ~TDB_VFORK;
280 			PROC_UNLOCK(p);
281 		}
282 	}
283 }
284