xref: /freebsd/sys/kern/subr_trap.c (revision 6f9c8e5b074419423648ffb89b83fd2f257e90b7)
1 /*-
2  * Copyright (C) 1994, David Greenman
3  * Copyright (c) 1990, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  * Copyright (c) 2007 The FreeBSD Foundation
6  *
7  * This code is derived from software contributed to Berkeley by
8  * the University of Utah, and William Jolitz.
9  *
10  * Portions of this software were developed by A. Joseph Koshy under
11  * sponsorship from the FreeBSD Foundation and Google, Inc.
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  * 3. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *	This product includes software developed by the University of
24  *	California, Berkeley and its contributors.
25  * 4. Neither the name of the University nor the names of its contributors
26  *    may be used to endorse or promote products derived from this software
27  *    without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39  * SUCH DAMAGE.
40  *
41  *	from: @(#)trap.c	7.4 (Berkeley) 5/13/91
42  */
43 
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
46 
47 #include "opt_capabilities.h"
48 #include "opt_ktrace.h"
49 #include "opt_kdtrace.h"
50 #include "opt_sched.h"
51 
52 #include <sys/param.h>
53 #include <sys/bus.h>
54 #include <sys/capability.h>
55 #include <sys/kernel.h>
56 #include <sys/lock.h>
57 #include <sys/mutex.h>
58 #include <sys/pmckern.h>
59 #include <sys/proc.h>
60 #include <sys/ktr.h>
61 #include <sys/pioctl.h>
62 #include <sys/ptrace.h>
63 #include <sys/resourcevar.h>
64 #include <sys/sched.h>
65 #include <sys/signalvar.h>
66 #include <sys/syscall.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysent.h>
69 #include <sys/systm.h>
70 #include <sys/vmmeter.h>
71 #ifdef KTRACE
72 #include <sys/uio.h>
73 #include <sys/ktrace.h>
74 #endif
75 #include <security/audit/audit.h>
76 
77 #include <machine/cpu.h>
78 
79 #ifdef VIMAGE
80 #include <net/vnet.h>
81 #endif
82 
83 #ifdef XEN
84 #include <vm/vm.h>
85 #include <vm/vm_param.h>
86 #include <vm/pmap.h>
87 #endif
88 
89 #include <security/mac/mac_framework.h>
90 
91 /*
92  * Define the code needed before returning to user mode, for trap and
93  * syscall.
94  */
95 void
96 userret(struct thread *td, struct trapframe *frame)
97 {
98 	struct proc *p = td->td_proc;
99 
100 	CTR3(KTR_SYSC, "userret: thread %p (pid %d, %s)", td, p->p_pid,
101             td->td_name);
102 #if 0
103 #ifdef DIAGNOSTIC
104 	/* Check that we called signotify() enough. */
105 	PROC_LOCK(p);
106 	thread_lock(td);
107 	if (SIGPENDING(td) && ((td->td_flags & TDF_NEEDSIGCHK) == 0 ||
108 	    (td->td_flags & TDF_ASTPENDING) == 0))
109 		printf("failed to set signal flags properly for ast()\n");
110 	thread_unlock(td);
111 	PROC_UNLOCK(p);
112 #endif
113 #endif
114 #ifdef KTRACE
115 	KTRUSERRET(td);
116 #endif
117 	/*
118 	 * If this thread tickled GEOM, we need to wait for the giggling to
119 	 * stop before we return to userland
120 	 */
121 	if (td->td_pflags & TDP_GEOM)
122 		g_waitidle();
123 
124 	/*
125 	 * Charge system time if profiling.
126 	 */
127 	if (p->p_flag & P_PROFIL)
128 		addupc_task(td, TRAPF_PC(frame), td->td_pticks * psratio);
129 	/*
130 	 * Let the scheduler adjust our priority etc.
131 	 */
132 	sched_userret(td);
133 	KASSERT(td->td_locks == 0,
134 	    ("userret: Returning with %d locks held.", td->td_locks));
135 #ifdef VIMAGE
136 	/* Unfortunately td_vnet_lpush needs VNET_DEBUG. */
137 	VNET_ASSERT(curvnet == NULL,
138 	    ("%s: Returning on td %p (pid %d, %s) with vnet %p set in %s",
139 	    __func__, td, p->p_pid, td->td_name, curvnet,
140 	    (td->td_vnet_lpush != NULL) ? td->td_vnet_lpush : "N/A"));
141 #endif
142 #ifdef XEN
143 	PT_UPDATES_FLUSH();
144 #endif
145 }
146 
147 /*
148  * Process an asynchronous software trap.
149  * This is relatively easy.
150  * This function will return with preemption disabled.
151  */
152 void
153 ast(struct trapframe *framep)
154 {
155 	struct thread *td;
156 	struct proc *p;
157 	int flags;
158 	int sig;
159 
160 	td = curthread;
161 	p = td->td_proc;
162 
163 	CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td, p->p_pid,
164             p->p_comm);
165 	KASSERT(TRAPF_USERMODE(framep), ("ast in kernel mode"));
166 	WITNESS_WARN(WARN_PANIC, NULL, "Returning to user mode");
167 	mtx_assert(&Giant, MA_NOTOWNED);
168 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
169 	td->td_frame = framep;
170 	td->td_pticks = 0;
171 
172 	/*
173 	 * This updates the td_flag's for the checks below in one
174 	 * "atomic" operation with turning off the astpending flag.
175 	 * If another AST is triggered while we are handling the
176 	 * AST's saved in flags, the astpending flag will be set and
177 	 * ast() will be called again.
178 	 */
179 	thread_lock(td);
180 	flags = td->td_flags;
181 	td->td_flags &= ~(TDF_ASTPENDING | TDF_NEEDSIGCHK | TDF_NEEDSUSPCHK |
182 	    TDF_NEEDRESCHED | TDF_ALRMPEND | TDF_PROFPEND | TDF_MACPEND);
183 	thread_unlock(td);
184 	PCPU_INC(cnt.v_trap);
185 
186 	if (td->td_ucred != p->p_ucred)
187 		cred_update_thread(td);
188 	if (td->td_pflags & TDP_OWEUPC && p->p_flag & P_PROFIL) {
189 		addupc_task(td, td->td_profil_addr, td->td_profil_ticks);
190 		td->td_profil_ticks = 0;
191 		td->td_pflags &= ~TDP_OWEUPC;
192 	}
193 	if (flags & TDF_ALRMPEND) {
194 		PROC_LOCK(p);
195 		psignal(p, SIGVTALRM);
196 		PROC_UNLOCK(p);
197 	}
198 	if (flags & TDF_PROFPEND) {
199 		PROC_LOCK(p);
200 		psignal(p, SIGPROF);
201 		PROC_UNLOCK(p);
202 	}
203 #ifdef MAC
204 	if (flags & TDF_MACPEND)
205 		mac_thread_userret(td);
206 #endif
207 	if (flags & TDF_NEEDRESCHED) {
208 #ifdef KTRACE
209 		if (KTRPOINT(td, KTR_CSW))
210 			ktrcsw(1, 1);
211 #endif
212 		thread_lock(td);
213 		sched_prio(td, td->td_user_pri);
214 		mi_switch(SW_INVOL | SWT_NEEDRESCHED, NULL);
215 		thread_unlock(td);
216 #ifdef KTRACE
217 		if (KTRPOINT(td, KTR_CSW))
218 			ktrcsw(0, 1);
219 #endif
220 	}
221 
222 	/*
223 	 * Check for signals. Unlocked reads of p_pendingcnt or
224 	 * p_siglist might cause process-directed signal to be handled
225 	 * later.
226 	 */
227 	if (flags & TDF_NEEDSIGCHK || p->p_pendingcnt > 0 ||
228 	    !SIGISEMPTY(p->p_siglist)) {
229 		PROC_LOCK(p);
230 		mtx_lock(&p->p_sigacts->ps_mtx);
231 		while ((sig = cursig(td, SIG_STOP_ALLOWED)) != 0)
232 			postsig(sig);
233 		mtx_unlock(&p->p_sigacts->ps_mtx);
234 		PROC_UNLOCK(p);
235 	}
236 	/*
237 	 * We need to check to see if we have to exit or wait due to a
238 	 * single threading requirement or some other STOP condition.
239 	 */
240 	if (flags & TDF_NEEDSUSPCHK) {
241 		PROC_LOCK(p);
242 		thread_suspend_check(0);
243 		PROC_UNLOCK(p);
244 	}
245 
246 	if (td->td_pflags & TDP_OLDMASK) {
247 		td->td_pflags &= ~TDP_OLDMASK;
248 		kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
249 	}
250 
251 	userret(td, framep);
252 	mtx_assert(&Giant, MA_NOTOWNED);
253 }
254 
255 #ifdef HAVE_SYSCALL_ARGS_DEF
256 const char *
257 syscallname(struct proc *p, u_int code)
258 {
259 	static const char unknown[] = "unknown";
260 	struct sysentvec *sv;
261 
262 	sv = p->p_sysent;
263 	if (sv->sv_syscallnames == NULL || code >= sv->sv_size)
264 		return (unknown);
265 	return (sv->sv_syscallnames[code]);
266 }
267 
268 int
269 syscallenter(struct thread *td, struct syscall_args *sa)
270 {
271 	struct proc *p;
272 	int error, traced;
273 
274 	PCPU_INC(cnt.v_syscall);
275 	p = td->td_proc;
276 
277 	td->td_pticks = 0;
278 	if (td->td_ucred != p->p_ucred)
279 		cred_update_thread(td);
280 	if (p->p_flag & P_TRACED) {
281 		traced = 1;
282 		PROC_LOCK(p);
283 		td->td_dbgflags &= ~TDB_USERWR;
284 		td->td_dbgflags |= TDB_SCE;
285 		PROC_UNLOCK(p);
286 	} else
287 		traced = 0;
288 	error = (p->p_sysent->sv_fetch_syscall_args)(td, sa);
289 #ifdef KTRACE
290 	if (KTRPOINT(td, KTR_SYSCALL))
291 		ktrsyscall(sa->code, sa->narg, sa->args);
292 #endif
293 
294 	CTR6(KTR_SYSC,
295 "syscall: td=%p pid %d %s (%#lx, %#lx, %#lx)",
296 	    td, td->td_proc->p_pid, syscallname(p, sa->code),
297 	    sa->args[0], sa->args[1], sa->args[2]);
298 
299 	if (error == 0) {
300 		STOPEVENT(p, S_SCE, sa->narg);
301 		PTRACESTOP_SC(p, td, S_PT_SCE);
302 		if (td->td_dbgflags & TDB_USERWR) {
303 			/*
304 			 * Reread syscall number and arguments if
305 			 * debugger modified registers or memory.
306 			 */
307 			error = (p->p_sysent->sv_fetch_syscall_args)(td, sa);
308 #ifdef KTRACE
309 			if (KTRPOINT(td, KTR_SYSCALL))
310 				ktrsyscall(sa->code, sa->narg, sa->args);
311 #endif
312 			if (error != 0)
313 				goto retval;
314 		}
315 
316 #ifdef CAPABILITIES
317 		/*
318 		 * In capability mode, we only allow access to system calls
319 		 * flagged with SYF_CAPENABLED.
320 		 */
321 		if (IN_CAPABILITY_MODE(td) &&
322 		    !(sa->callp->sy_flags & SYF_CAPENABLED)) {
323 			error = ECAPMODE;
324 			goto retval;
325 		}
326 #endif
327 
328 		error = syscall_thread_enter(td, sa->callp);
329 		if (error != 0)
330 			goto retval;
331 
332 #ifdef KDTRACE_HOOKS
333 		/*
334 		 * If the systrace module has registered it's probe
335 		 * callback and if there is a probe active for the
336 		 * syscall 'entry', process the probe.
337 		 */
338 		if (systrace_probe_func != NULL && sa->callp->sy_entry != 0)
339 			(*systrace_probe_func)(sa->callp->sy_entry, sa->code,
340 			    sa->callp, sa->args, 0);
341 #endif
342 
343 		AUDIT_SYSCALL_ENTER(sa->code, td);
344 		error = (sa->callp->sy_call)(td, sa->args);
345 		AUDIT_SYSCALL_EXIT(error, td);
346 
347 		/* Save the latest error return value. */
348 		td->td_errno = error;
349 
350 #ifdef KDTRACE_HOOKS
351 		/*
352 		 * If the systrace module has registered it's probe
353 		 * callback and if there is a probe active for the
354 		 * syscall 'return', process the probe.
355 		 */
356 		if (systrace_probe_func != NULL && sa->callp->sy_return != 0)
357 			(*systrace_probe_func)(sa->callp->sy_return, sa->code,
358 			    sa->callp, NULL, (error) ? -1 : td->td_retval[0]);
359 #endif
360 		syscall_thread_exit(td, sa->callp);
361 		CTR4(KTR_SYSC, "syscall: p=%p error=%d return %#lx %#lx",
362 		    p, error, td->td_retval[0], td->td_retval[1]);
363 	}
364  retval:
365 	if (traced) {
366 		PROC_LOCK(p);
367 		td->td_dbgflags &= ~TDB_SCE;
368 		PROC_UNLOCK(p);
369 	}
370 	(p->p_sysent->sv_set_syscall_retval)(td, error);
371 	return (error);
372 }
373 
374 void
375 syscallret(struct thread *td, int error, struct syscall_args *sa __unused)
376 {
377 	struct proc *p;
378 	int traced;
379 
380 	p = td->td_proc;
381 
382 	/*
383 	 * Check for misbehavior.
384 	 */
385 	WITNESS_WARN(WARN_PANIC, NULL, "System call %s returning",
386 	    syscallname(p, sa->code));
387 	KASSERT(td->td_critnest == 0,
388 	    ("System call %s returning in a critical section",
389 	    syscallname(p, sa->code)));
390 	KASSERT(td->td_locks == 0,
391 	    ("System call %s returning with %d locks held",
392 	     syscallname(p, sa->code), td->td_locks));
393 
394 	/*
395 	 * Handle reschedule and other end-of-syscall issues
396 	 */
397 	userret(td, td->td_frame);
398 
399 	CTR4(KTR_SYSC, "syscall %s exit thread %p pid %d proc %s",
400 	    syscallname(p, sa->code), td, td->td_proc->p_pid, td->td_name);
401 
402 #ifdef KTRACE
403 	if (KTRPOINT(td, KTR_SYSRET))
404 		ktrsysret(sa->code, error, td->td_retval[0]);
405 #endif
406 
407 	if (p->p_flag & P_TRACED) {
408 		traced = 1;
409 		PROC_LOCK(p);
410 		td->td_dbgflags |= TDB_SCX;
411 		PROC_UNLOCK(p);
412 	} else
413 		traced = 0;
414 	/*
415 	 * This works because errno is findable through the
416 	 * register set.  If we ever support an emulation where this
417 	 * is not the case, this code will need to be revisited.
418 	 */
419 	STOPEVENT(p, S_SCX, sa->code);
420 	PTRACESTOP_SC(p, td, S_PT_SCX);
421 	if (traced || (td->td_dbgflags & (TDB_EXEC | TDB_FORK)) != 0) {
422 		PROC_LOCK(p);
423 		td->td_dbgflags &= ~(TDB_SCX | TDB_EXEC | TDB_FORK);
424 		PROC_UNLOCK(p);
425 	}
426 }
427 #endif /* HAVE_SYSCALL_ARGS_DEF */
428