xref: /freebsd/sys/kern/subr_syscall.c (revision 734e82fe33aa764367791a7d603b383996c6b40b)
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 #include <sys/capsicum.h>
46 #include <sys/ktr.h>
47 #include <sys/vmmeter.h>
48 #ifdef KTRACE
49 #include <sys/uio.h>
50 #include <sys/ktrace.h>
51 #endif
52 #include <security/audit/audit.h>
53 
54 static inline void
55 syscallenter(struct thread *td)
56 {
57 	struct proc *p;
58 	struct syscall_args *sa;
59 	struct sysent *se;
60 	int error, traced;
61 	bool sy_thr_static;
62 
63 	VM_CNT_INC(v_syscall);
64 	p = td->td_proc;
65 	sa = &td->td_sa;
66 
67 	td->td_pticks = 0;
68 	if (__predict_false(td->td_cowgen != atomic_load_int(&p->p_cowgen)))
69 		thread_cow_update(td);
70 	traced = (p->p_flag & P_TRACED) != 0;
71 	if (__predict_false(traced || td->td_dbgflags & TDB_USERWR)) {
72 		PROC_LOCK(p);
73 		MPASS((td->td_dbgflags & TDB_BOUNDARY) == 0);
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 	se = sa->callp;
81 #ifdef KTRACE
82 	if (KTRPOINT(td, KTR_SYSCALL))
83 		ktrsyscall(sa->code, se->sy_narg, sa->args);
84 #endif
85 	KTR_START4(KTR_SYSC, "syscall", syscallname(p, sa->code),
86 	    (uintptr_t)td, "pid:%d", td->td_proc->p_pid, "arg0:%p", sa->args[0],
87 	    "arg1:%p", sa->args[1], "arg2:%p", sa->args[2]);
88 
89 	if (__predict_false(error != 0)) {
90 		td->td_errno = error;
91 		goto retval;
92 	}
93 
94 	if (__predict_false(traced)) {
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 			se = sa->callp;
107 #ifdef KTRACE
108 			if (KTRPOINT(td, KTR_SYSCALL))
109 				ktrsyscall(sa->code, se->sy_narg, sa->args);
110 #endif
111 			if (error != 0) {
112 				td->td_errno = error;
113 				goto retval;
114 			}
115 		}
116 	}
117 
118 #ifdef CAPABILITY_MODE
119 	/*
120 	 * In capability mode, we only allow access to system calls
121 	 * flagged with SYF_CAPENABLED.
122 	 */
123 	if (__predict_false(IN_CAPABILITY_MODE(td) &&
124 	    (se->sy_flags & SYF_CAPENABLED) == 0)) {
125 		td->td_errno = error = ECAPMODE;
126 		goto retval;
127 	}
128 #endif
129 
130 	/*
131 	 * Fetch fast sigblock value at the time of syscall entry to
132 	 * handle sleepqueue primitives which might call cursig().
133 	 */
134 	if (__predict_false(sigfastblock_fetch_always))
135 		(void)sigfastblock_fetch(td);
136 
137 	/* Let system calls set td_errno directly. */
138 	KASSERT((td->td_pflags & TDP_NERRNO) == 0,
139 	    ("%s: TDP_NERRNO set", __func__));
140 
141 	sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
142 
143 	if (__predict_false(SYSTRACE_ENABLED() ||
144 	    AUDIT_SYSCALL_ENTER(sa->code, td) ||
145 	    !sy_thr_static)) {
146 		if (!sy_thr_static) {
147 			error = syscall_thread_enter(td, se);
148 			if (error != 0) {
149 				td->td_errno = error;
150 				goto retval;
151 			}
152 		}
153 
154 #ifdef KDTRACE_HOOKS
155 		/* Give the syscall:::entry DTrace probe a chance to fire. */
156 		if (__predict_false(se->sy_entry != 0))
157 			(*systrace_probe_func)(sa, SYSTRACE_ENTRY, 0);
158 #endif
159 		error = (se->sy_call)(td, sa->args);
160 		/* Save the latest error return value. */
161 		if (__predict_false((td->td_pflags & TDP_NERRNO) != 0))
162 			td->td_pflags &= ~TDP_NERRNO;
163 		else
164 			td->td_errno = error;
165 
166 		/*
167 		 * Note that some syscall implementations (e.g., sys_execve)
168 		 * will commit the audit record just before their final return.
169 		 * These were done under the assumption that nothing of interest
170 		 * would happen between their return and here, where we would
171 		 * normally commit the audit record.  These assumptions will
172 		 * need to be revisited should any substantial logic be added
173 		 * above.
174 		 */
175 		AUDIT_SYSCALL_EXIT(error, td);
176 
177 #ifdef KDTRACE_HOOKS
178 		/* Give the syscall:::return DTrace probe a chance to fire. */
179 		if (__predict_false(se->sy_return != 0))
180 			(*systrace_probe_func)(sa, SYSTRACE_RETURN,
181 			    error ? -1 : td->td_retval[0]);
182 #endif
183 
184 		if (!sy_thr_static)
185 			syscall_thread_exit(td, se);
186 	} else {
187 		error = (se->sy_call)(td, sa->args);
188 		/* Save the latest error return value. */
189 		if (__predict_false((td->td_pflags & TDP_NERRNO) != 0))
190 			td->td_pflags &= ~TDP_NERRNO;
191 		else
192 			td->td_errno = error;
193 	}
194 
195  retval:
196 	KTR_STOP4(KTR_SYSC, "syscall", syscallname(p, sa->code),
197 	    (uintptr_t)td, "pid:%d", td->td_proc->p_pid, "error:%d", error,
198 	    "retval0:%#lx", td->td_retval[0], "retval1:%#lx",
199 	    td->td_retval[1]);
200 	if (__predict_false(traced)) {
201 		PROC_LOCK(p);
202 		td->td_dbgflags &= ~(TDB_SCE | TDB_BOUNDARY);
203 		PROC_UNLOCK(p);
204 	}
205 	(p->p_sysent->sv_set_syscall_retval)(td, error);
206 }
207 
208 static inline void
209 syscallret(struct thread *td)
210 {
211 	struct proc *p;
212 	struct syscall_args *sa;
213 	ksiginfo_t ksi;
214 	int traced;
215 
216 	KASSERT(td->td_errno != ERELOOKUP,
217 	    ("ERELOOKUP not consumed syscall %d", td->td_sa.code));
218 
219 	p = td->td_proc;
220 	sa = &td->td_sa;
221 	if (__predict_false(td->td_errno == ENOTCAPABLE ||
222 	    td->td_errno == ECAPMODE)) {
223 		if ((trap_enotcap ||
224 		    (p->p_flag2 & P2_TRAPCAP) != 0) && IN_CAPABILITY_MODE(td)) {
225 			ksiginfo_init_trap(&ksi);
226 			ksi.ksi_signo = SIGTRAP;
227 			ksi.ksi_errno = td->td_errno;
228 			ksi.ksi_code = TRAP_CAP;
229 			ksi.ksi_info.si_syscall = sa->original_code;
230 			trapsignal(td, &ksi);
231 		}
232 	}
233 
234 	/*
235 	 * Handle reschedule and other end-of-syscall issues
236 	 */
237 	userret(td, td->td_frame);
238 
239 #ifdef KTRACE
240 	if (KTRPOINT(td, KTR_SYSRET)) {
241 		ktrsysret(sa->code, td->td_errno, td->td_retval[0]);
242 	}
243 #endif
244 
245 	traced = 0;
246 	if (__predict_false(p->p_flag & P_TRACED)) {
247 		traced = 1;
248 		PROC_LOCK(p);
249 		td->td_dbgflags |= TDB_SCX;
250 		PROC_UNLOCK(p);
251 	}
252 	if (__predict_false(traced ||
253 	    (td->td_dbgflags & (TDB_EXEC | TDB_FORK)) != 0)) {
254 		PROC_LOCK(p);
255 		/*
256 		 * Linux debuggers expect an additional stop for exec,
257 		 * between the usual syscall entry and exit.  Raise
258 		 * the exec event now and then clear TDB_EXEC so that
259 		 * the next stop is reported as a syscall exit by
260 		 * linux_ptrace_status().
261 		 *
262 		 * We are accessing p->p_pptr without any additional
263 		 * locks here: it cannot change while p is kept locked;
264 		 * while the debugger could in theory change its ABI
265 		 * while tracing another process, the outcome of such
266 		 * a race wouln't be deterministic anyway.
267 		 */
268 		if (traced && (td->td_dbgflags & TDB_EXEC) != 0 &&
269 		    SV_PROC_ABI(p->p_pptr) == SV_ABI_LINUX) {
270 			ptracestop(td, SIGTRAP, NULL);
271 			td->td_dbgflags &= ~TDB_EXEC;
272 		}
273 		/*
274 		 * If tracing the execed process, trap to the debugger
275 		 * so that breakpoints can be set before the program
276 		 * executes.  If debugger requested tracing of syscall
277 		 * returns, do it now too.
278 		 */
279 		if (traced &&
280 		    ((td->td_dbgflags & (TDB_FORK | TDB_EXEC)) != 0 ||
281 		    (p->p_ptevents & PTRACE_SCX) != 0)) {
282 			MPASS((td->td_dbgflags & TDB_BOUNDARY) == 0);
283 			td->td_dbgflags |= TDB_BOUNDARY;
284 			ptracestop(td, SIGTRAP, NULL);
285 		}
286 		td->td_dbgflags &= ~(TDB_SCX | TDB_EXEC | TDB_FORK |
287 		    TDB_BOUNDARY);
288 		PROC_UNLOCK(p);
289 	}
290 }
291