xref: /freebsd/sys/kern/subr_syscall.c (revision d9a42747950146bf03cda7f6e25d219253f8a57a)
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("$FreeBSD$");
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
58 syscallenter(struct thread *td)
59 {
60 	struct proc *p;
61 	struct syscall_args *sa;
62 	struct sysent *se;
63 	int error, traced;
64 	bool sy_thr_static;
65 
66 	VM_CNT_INC(v_syscall);
67 	p = td->td_proc;
68 	sa = &td->td_sa;
69 
70 	td->td_pticks = 0;
71 	if (__predict_false(td->td_cowgen != atomic_load_int(&p->p_cowgen)))
72 		thread_cow_update(td);
73 	traced = (p->p_flag & P_TRACED) != 0;
74 	if (__predict_false(traced || td->td_dbgflags & TDB_USERWR)) {
75 		PROC_LOCK(p);
76 		MPASS((td->td_dbgflags & TDB_BOUNDARY) == 0);
77 		td->td_dbgflags &= ~TDB_USERWR;
78 		if (traced)
79 			td->td_dbgflags |= TDB_SCE;
80 		PROC_UNLOCK(p);
81 	}
82 	error = (p->p_sysent->sv_fetch_syscall_args)(td);
83 	se = sa->callp;
84 #ifdef KTRACE
85 	if (KTRPOINT(td, KTR_SYSCALL))
86 		ktrsyscall(sa->code, se->sy_narg, sa->args);
87 #endif
88 	KTR_START4(KTR_SYSC, "syscall", syscallname(p, sa->code),
89 	    (uintptr_t)td, "pid:%d", td->td_proc->p_pid, "arg0:%p", sa->args[0],
90 	    "arg1:%p", sa->args[1], "arg2:%p", sa->args[2]);
91 
92 	if (__predict_false(error != 0)) {
93 		td->td_errno = error;
94 		goto retval;
95 	}
96 
97 	if (__predict_false(traced)) {
98 		PROC_LOCK(p);
99 		if (p->p_ptevents & PTRACE_SCE)
100 			ptracestop((td), SIGTRAP, NULL);
101 		PROC_UNLOCK(p);
102 
103 		if ((td->td_dbgflags & TDB_USERWR) != 0) {
104 			/*
105 			 * Reread syscall number and arguments if debugger
106 			 * modified registers or memory.
107 			 */
108 			error = (p->p_sysent->sv_fetch_syscall_args)(td);
109 			se = sa->callp;
110 #ifdef KTRACE
111 			if (KTRPOINT(td, KTR_SYSCALL))
112 				ktrsyscall(sa->code, se->sy_narg, sa->args);
113 #endif
114 			if (error != 0) {
115 				td->td_errno = error;
116 				goto retval;
117 			}
118 		}
119 	}
120 
121 #ifdef CAPABILITY_MODE
122 	/*
123 	 * In capability mode, we only allow access to system calls
124 	 * flagged with SYF_CAPENABLED.
125 	 */
126 	if (__predict_false(IN_CAPABILITY_MODE(td) &&
127 	    (se->sy_flags & SYF_CAPENABLED) == 0)) {
128 		td->td_errno = error = ECAPMODE;
129 		goto retval;
130 	}
131 #endif
132 
133 	/*
134 	 * Fetch fast sigblock value at the time of syscall entry to
135 	 * handle sleepqueue primitives which might call cursig().
136 	 */
137 	if (__predict_false(sigfastblock_fetch_always))
138 		(void)sigfastblock_fetch(td);
139 
140 	/* Let system calls set td_errno directly. */
141 	KASSERT((td->td_pflags & TDP_NERRNO) == 0,
142 	    ("%s: TDP_NERRNO set", __func__));
143 
144 	sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
145 
146 	if (__predict_false(SYSTRACE_ENABLED() ||
147 	    AUDIT_SYSCALL_ENTER(sa->code, td) ||
148 	    !sy_thr_static)) {
149 		if (!sy_thr_static) {
150 			error = syscall_thread_enter(td, se);
151 			if (error != 0) {
152 				td->td_errno = error;
153 				goto retval;
154 			}
155 		}
156 
157 #ifdef KDTRACE_HOOKS
158 		/* Give the syscall:::entry DTrace probe a chance to fire. */
159 		if (__predict_false(se->sy_entry != 0))
160 			(*systrace_probe_func)(sa, SYSTRACE_ENTRY, 0);
161 #endif
162 		error = (se->sy_call)(td, sa->args);
163 		/* Save the latest error return value. */
164 		if (__predict_false((td->td_pflags & TDP_NERRNO) != 0))
165 			td->td_pflags &= ~TDP_NERRNO;
166 		else
167 			td->td_errno = error;
168 
169 		/*
170 		 * Note that some syscall implementations (e.g., sys_execve)
171 		 * will commit the audit record just before their final return.
172 		 * These were done under the assumption that nothing of interest
173 		 * would happen between their return and here, where we would
174 		 * normally commit the audit record.  These assumptions will
175 		 * need to be revisited should any substantial logic be added
176 		 * above.
177 		 */
178 		AUDIT_SYSCALL_EXIT(error, td);
179 
180 #ifdef KDTRACE_HOOKS
181 		/* Give the syscall:::return DTrace probe a chance to fire. */
182 		if (__predict_false(se->sy_return != 0))
183 			(*systrace_probe_func)(sa, SYSTRACE_RETURN,
184 			    error ? -1 : td->td_retval[0]);
185 #endif
186 
187 		if (!sy_thr_static)
188 			syscall_thread_exit(td, se);
189 	} else {
190 		error = (se->sy_call)(td, sa->args);
191 		/* Save the latest error return value. */
192 		if (__predict_false((td->td_pflags & TDP_NERRNO) != 0))
193 			td->td_pflags &= ~TDP_NERRNO;
194 		else
195 			td->td_errno = error;
196 	}
197 
198  retval:
199 	KTR_STOP4(KTR_SYSC, "syscall", syscallname(p, sa->code),
200 	    (uintptr_t)td, "pid:%d", td->td_proc->p_pid, "error:%d", error,
201 	    "retval0:%#lx", td->td_retval[0], "retval1:%#lx",
202 	    td->td_retval[1]);
203 	if (__predict_false(traced)) {
204 		PROC_LOCK(p);
205 		td->td_dbgflags &= ~(TDB_SCE | TDB_BOUNDARY);
206 		PROC_UNLOCK(p);
207 	}
208 	(p->p_sysent->sv_set_syscall_retval)(td, error);
209 }
210 
211 static inline void
212 syscallret(struct thread *td)
213 {
214 	struct proc *p;
215 	struct syscall_args *sa;
216 	ksiginfo_t ksi;
217 	int traced;
218 
219 	KASSERT(td->td_errno != ERELOOKUP,
220 	    ("ERELOOKUP not consumed syscall %d", td->td_sa.code));
221 
222 	p = td->td_proc;
223 	sa = &td->td_sa;
224 	if (__predict_false(td->td_errno == ENOTCAPABLE ||
225 	    td->td_errno == ECAPMODE)) {
226 		if ((trap_enotcap ||
227 		    (p->p_flag2 & P2_TRAPCAP) != 0) && IN_CAPABILITY_MODE(td)) {
228 			ksiginfo_init_trap(&ksi);
229 			ksi.ksi_signo = SIGTRAP;
230 			ksi.ksi_errno = td->td_errno;
231 			ksi.ksi_code = TRAP_CAP;
232 			ksi.ksi_info.si_syscall = sa->original_code;
233 			trapsignal(td, &ksi);
234 		}
235 	}
236 
237 	/*
238 	 * Handle reschedule and other end-of-syscall issues
239 	 */
240 	userret(td, td->td_frame);
241 
242 #ifdef KTRACE
243 	if (KTRPOINT(td, KTR_SYSRET)) {
244 		ktrsysret(sa->code, td->td_errno, td->td_retval[0]);
245 	}
246 #endif
247 
248 	traced = 0;
249 	if (__predict_false(p->p_flag & P_TRACED)) {
250 		traced = 1;
251 		PROC_LOCK(p);
252 		td->td_dbgflags |= TDB_SCX;
253 		PROC_UNLOCK(p);
254 	}
255 	if (__predict_false(traced ||
256 	    (td->td_dbgflags & (TDB_EXEC | TDB_FORK)) != 0)) {
257 		PROC_LOCK(p);
258 		/*
259 		 * Linux debuggers expect an additional stop for exec,
260 		 * between the usual syscall entry and exit.  Raise
261 		 * the exec event now and then clear TDB_EXEC so that
262 		 * the next stop is reported as a syscall exit by
263 		 * linux_ptrace_status().
264 		 *
265 		 * We are accessing p->p_pptr without any additional
266 		 * locks here: it cannot change while p is kept locked;
267 		 * while the debugger could in theory change its ABI
268 		 * while tracing another process, the outcome of such
269 		 * a race wouln't be deterministic anyway.
270 		 */
271 		if (traced && (td->td_dbgflags & TDB_EXEC) != 0 &&
272 		    SV_PROC_ABI(p->p_pptr) == SV_ABI_LINUX) {
273 			ptracestop(td, SIGTRAP, NULL);
274 			td->td_dbgflags &= ~TDB_EXEC;
275 		}
276 		/*
277 		 * If tracing the execed process, trap to the debugger
278 		 * so that breakpoints can be set before the program
279 		 * executes.  If debugger requested tracing of syscall
280 		 * returns, do it now too.
281 		 */
282 		if (traced &&
283 		    ((td->td_dbgflags & (TDB_FORK | TDB_EXEC)) != 0 ||
284 		    (p->p_ptevents & PTRACE_SCX) != 0)) {
285 			MPASS((td->td_dbgflags & TDB_BOUNDARY) == 0);
286 			td->td_dbgflags |= TDB_BOUNDARY;
287 			ptracestop(td, SIGTRAP, NULL);
288 		}
289 		td->td_dbgflags &= ~(TDB_SCX | TDB_EXEC | TDB_FORK |
290 		    TDB_BOUNDARY);
291 		PROC_UNLOCK(p);
292 	}
293 }
294