xref: /freebsd/sys/kern/subr_trap.c (revision 5dae51da3da0cc94d17bd67b308fad304ebec7e0)
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_hwpmc_hooks.h"
48 #include "opt_ktrace.h"
49 #include "opt_sched.h"
50 
51 #include <sys/param.h>
52 #include <sys/bus.h>
53 #include <sys/capsicum.h>
54 #include <sys/kernel.h>
55 #include <sys/lock.h>
56 #include <sys/mutex.h>
57 #include <sys/pmckern.h>
58 #include <sys/proc.h>
59 #include <sys/ktr.h>
60 #include <sys/pioctl.h>
61 #include <sys/ptrace.h>
62 #include <sys/racct.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	HWPMC_HOOKS
84 #include <sys/pmckern.h>
85 #endif
86 
87 #include <security/mac/mac_framework.h>
88 
89 void (*softdep_ast_cleanup)(void);
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 	KASSERT((p->p_flag & P_WEXIT) == 0,
103 	    ("Exiting process returns to usermode"));
104 #ifdef DIAGNOSTIC
105 	/*
106 	 * Check that we called signotify() enough.  For
107 	 * multi-threaded processes, where signal distribution might
108 	 * change due to other threads changing sigmask, the check is
109 	 * racy and cannot be performed reliably.
110 	 * If current process is vfork child, indicated by P_PPWAIT, then
111 	 * issignal() ignores stops, so we block the check to avoid
112 	 * classifying pending signals.
113 	 */
114 	if (p->p_numthreads == 1) {
115 		PROC_LOCK(p);
116 		thread_lock(td);
117 		if ((p->p_flag & P_PPWAIT) == 0) {
118 			KASSERT(!SIGPENDING(td) || (td->td_flags &
119 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING)) ==
120 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING),
121 			    ("failed to set signal flags for ast p %p "
122 			    "td %p fl %x", p, td, td->td_flags));
123 		}
124 		thread_unlock(td);
125 		PROC_UNLOCK(p);
126 	}
127 #endif
128 #ifdef KTRACE
129 	KTRUSERRET(td);
130 #endif
131 	if (softdep_ast_cleanup != NULL)
132 		softdep_ast_cleanup();
133 
134 	/*
135 	 * If this thread tickled GEOM, we need to wait for the giggling to
136 	 * stop before we return to userland
137 	 */
138 	if (td->td_pflags & TDP_GEOM)
139 		g_waitidle();
140 
141 	/*
142 	 * Charge system time if profiling.
143 	 */
144 	if (p->p_flag & P_PROFIL)
145 		addupc_task(td, TRAPF_PC(frame), td->td_pticks * psratio);
146 	/*
147 	 * Let the scheduler adjust our priority etc.
148 	 */
149 	sched_userret(td);
150 
151 	/*
152 	 * Check for misbehavior.
153 	 *
154 	 * In case there is a callchain tracing ongoing because of
155 	 * hwpmc(4), skip the scheduler pinning check.
156 	 * hwpmc(4) subsystem, infact, will collect callchain informations
157 	 * at ast() checkpoint, which is past userret().
158 	 */
159 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
160 	KASSERT(td->td_critnest == 0,
161 	    ("userret: Returning in a critical section"));
162 	KASSERT(td->td_locks == 0,
163 	    ("userret: Returning with %d locks held", td->td_locks));
164 	KASSERT(td->td_rw_rlocks == 0,
165 	    ("userret: Returning with %d rwlocks held in read mode",
166 	    td->td_rw_rlocks));
167 	KASSERT((td->td_pflags & TDP_NOFAULTING) == 0,
168 	    ("userret: Returning with pagefaults disabled"));
169 	KASSERT(td->td_no_sleeping == 0,
170 	    ("userret: Returning with sleep disabled"));
171 	KASSERT(td->td_pinned == 0 || (td->td_pflags & TDP_CALLCHAIN) != 0,
172 	    ("userret: Returning with with pinned thread"));
173 	KASSERT(td->td_vp_reserv == 0,
174 	    ("userret: Returning while holding vnode reservation"));
175 	KASSERT((td->td_flags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
176 	    ("userret: Returning with stop signals deferred"));
177 	KASSERT(td->td_su == NULL,
178 	    ("userret: Returning with SU cleanup request not handled"));
179 #ifdef VIMAGE
180 	/* Unfortunately td_vnet_lpush needs VNET_DEBUG. */
181 	VNET_ASSERT(curvnet == NULL,
182 	    ("%s: Returning on td %p (pid %d, %s) with vnet %p set in %s",
183 	    __func__, td, p->p_pid, td->td_name, curvnet,
184 	    (td->td_vnet_lpush != NULL) ? td->td_vnet_lpush : "N/A"));
185 #endif
186 #ifdef RACCT
187 	if (racct_enable && p->p_throttled != 0) {
188 		PROC_LOCK(p);
189 		while (p->p_throttled != 0) {
190 			msleep(p->p_racct, &p->p_mtx, 0, "racct",
191 			    p->p_throttled < 0 ? 0 : p->p_throttled);
192 			if (p->p_throttled > 0)
193 				p->p_throttled = 0;
194 		}
195 		PROC_UNLOCK(p);
196 	}
197 #endif
198 }
199 
200 /*
201  * Process an asynchronous software trap.
202  * This is relatively easy.
203  * This function will return with preemption disabled.
204  */
205 void
206 ast(struct trapframe *framep)
207 {
208 	struct thread *td;
209 	struct proc *p;
210 	int flags;
211 	int sig;
212 
213 	td = curthread;
214 	p = td->td_proc;
215 
216 	CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td, p->p_pid,
217             p->p_comm);
218 	KASSERT(TRAPF_USERMODE(framep), ("ast in kernel mode"));
219 	WITNESS_WARN(WARN_PANIC, NULL, "Returning to user mode");
220 	mtx_assert(&Giant, MA_NOTOWNED);
221 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
222 	td->td_frame = framep;
223 	td->td_pticks = 0;
224 
225 	/*
226 	 * This updates the td_flag's for the checks below in one
227 	 * "atomic" operation with turning off the astpending flag.
228 	 * If another AST is triggered while we are handling the
229 	 * AST's saved in flags, the astpending flag will be set and
230 	 * ast() will be called again.
231 	 */
232 	thread_lock(td);
233 	flags = td->td_flags;
234 	td->td_flags &= ~(TDF_ASTPENDING | TDF_NEEDSIGCHK | TDF_NEEDSUSPCHK |
235 	    TDF_NEEDRESCHED | TDF_ALRMPEND | TDF_PROFPEND | TDF_MACPEND);
236 	thread_unlock(td);
237 	PCPU_INC(cnt.v_trap);
238 
239 	if (td->td_cowgen != p->p_cowgen)
240 		thread_cow_update(td);
241 	if (td->td_pflags & TDP_OWEUPC && p->p_flag & P_PROFIL) {
242 		addupc_task(td, td->td_profil_addr, td->td_profil_ticks);
243 		td->td_profil_ticks = 0;
244 		td->td_pflags &= ~TDP_OWEUPC;
245 	}
246 #ifdef HWPMC_HOOKS
247 	/* Handle Software PMC callchain capture. */
248 	if (PMC_IS_PENDING_CALLCHAIN(td))
249 		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_USER_CALLCHAIN_SOFT, (void *) framep);
250 #endif
251 	if (flags & TDF_ALRMPEND) {
252 		PROC_LOCK(p);
253 		kern_psignal(p, SIGVTALRM);
254 		PROC_UNLOCK(p);
255 	}
256 	if (flags & TDF_PROFPEND) {
257 		PROC_LOCK(p);
258 		kern_psignal(p, SIGPROF);
259 		PROC_UNLOCK(p);
260 	}
261 #ifdef MAC
262 	if (flags & TDF_MACPEND)
263 		mac_thread_userret(td);
264 #endif
265 	if (flags & TDF_NEEDRESCHED) {
266 #ifdef KTRACE
267 		if (KTRPOINT(td, KTR_CSW))
268 			ktrcsw(1, 1, __func__);
269 #endif
270 		thread_lock(td);
271 		sched_prio(td, td->td_user_pri);
272 		mi_switch(SW_INVOL | SWT_NEEDRESCHED, NULL);
273 		thread_unlock(td);
274 #ifdef KTRACE
275 		if (KTRPOINT(td, KTR_CSW))
276 			ktrcsw(0, 1, __func__);
277 #endif
278 	}
279 
280 #ifdef DIAGNOSTIC
281 	if (p->p_numthreads == 1 && (flags & TDF_NEEDSIGCHK) == 0) {
282 		PROC_LOCK(p);
283 		thread_lock(td);
284 		/*
285 		 * Note that TDF_NEEDSIGCHK should be re-read from
286 		 * td_flags, since signal might have been delivered
287 		 * after we cleared td_flags above.  This is one of
288 		 * the reason for looping check for AST condition.
289 		 * See comment in userret() about P_PPWAIT.
290 		 */
291 		if ((p->p_flag & P_PPWAIT) == 0) {
292 			KASSERT(!SIGPENDING(td) || (td->td_flags &
293 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING)) ==
294 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING),
295 			    ("failed2 to set signal flags for ast p %p td %p "
296 			    "fl %x %x", p, td, flags, td->td_flags));
297 		}
298 		thread_unlock(td);
299 		PROC_UNLOCK(p);
300 	}
301 #endif
302 
303 	/*
304 	 * Check for signals. Unlocked reads of p_pendingcnt or
305 	 * p_siglist might cause process-directed signal to be handled
306 	 * later.
307 	 */
308 	if (flags & TDF_NEEDSIGCHK || p->p_pendingcnt > 0 ||
309 	    !SIGISEMPTY(p->p_siglist)) {
310 		PROC_LOCK(p);
311 		mtx_lock(&p->p_sigacts->ps_mtx);
312 		while ((sig = cursig(td)) != 0) {
313 			KASSERT(sig >= 0, ("sig %d", sig));
314 			postsig(sig);
315 		}
316 		mtx_unlock(&p->p_sigacts->ps_mtx);
317 		PROC_UNLOCK(p);
318 	}
319 	/*
320 	 * We need to check to see if we have to exit or wait due to a
321 	 * single threading requirement or some other STOP condition.
322 	 */
323 	if (flags & TDF_NEEDSUSPCHK) {
324 		PROC_LOCK(p);
325 		thread_suspend_check(0);
326 		PROC_UNLOCK(p);
327 	}
328 
329 	if (td->td_pflags & TDP_OLDMASK) {
330 		td->td_pflags &= ~TDP_OLDMASK;
331 		kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
332 	}
333 
334 	userret(td, framep);
335 }
336 
337 const char *
338 syscallname(struct proc *p, u_int code)
339 {
340 	static const char unknown[] = "unknown";
341 	struct sysentvec *sv;
342 
343 	sv = p->p_sysent;
344 	if (sv->sv_syscallnames == NULL || code >= sv->sv_size)
345 		return (unknown);
346 	return (sv->sv_syscallnames[code]);
347 }
348