xref: /freebsd/sys/kern/subr_trap.c (revision dd41de95a84d979615a2ef11df6850622bf6184e)
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) 2007 The FreeBSD Foundation
8  *
9  * This code is derived from software contributed to Berkeley by
10  * the University of Utah, and William Jolitz.
11  *
12  * Portions of this software were developed by A. Joseph Koshy under
13  * sponsorship from the FreeBSD Foundation and Google, Inc.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  * 3. All advertising materials mentioning features or use of this software
24  *    must display the following acknowledgement:
25  *	This product includes software developed by the University of
26  *	California, Berkeley and its contributors.
27  * 4. Neither the name of the University nor the names of its contributors
28  *    may be used to endorse or promote products derived from this software
29  *    without specific prior written permission.
30  *
31  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
32  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
33  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
34  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
35  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
36  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
37  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
38  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
39  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
40  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41  * SUCH DAMAGE.
42  *
43  *	from: @(#)trap.c	7.4 (Berkeley) 5/13/91
44  */
45 
46 #include <sys/cdefs.h>
47 __FBSDID("$FreeBSD$");
48 
49 #include "opt_hwpmc_hooks.h"
50 #include "opt_ktrace.h"
51 #include "opt_sched.h"
52 
53 #include <sys/param.h>
54 #include <sys/bus.h>
55 #include <sys/capsicum.h>
56 #include <sys/kernel.h>
57 #include <sys/lock.h>
58 #include <sys/mutex.h>
59 #include <sys/pmckern.h>
60 #include <sys/proc.h>
61 #include <sys/ktr.h>
62 #include <sys/ptrace.h>
63 #include <sys/racct.h>
64 #include <sys/resourcevar.h>
65 #include <sys/sched.h>
66 #include <sys/signalvar.h>
67 #include <sys/syscall.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/sysent.h>
70 #include <sys/systm.h>
71 #include <sys/vmmeter.h>
72 #ifdef KTRACE
73 #include <sys/uio.h>
74 #include <sys/ktrace.h>
75 #endif
76 #include <security/audit/audit.h>
77 
78 #include <machine/cpu.h>
79 
80 #ifdef VIMAGE
81 #include <net/vnet.h>
82 #endif
83 
84 #ifdef	HWPMC_HOOKS
85 #include <sys/pmckern.h>
86 #endif
87 
88 #include <security/mac/mac_framework.h>
89 
90 void (*softdep_ast_cleanup)(struct thread *);
91 
92 /*
93  * Define the code needed before returning to user mode, for trap and
94  * syscall.
95  */
96 void
97 userret(struct thread *td, struct trapframe *frame)
98 {
99 	struct proc *p = td->td_proc;
100 
101 	CTR3(KTR_SYSC, "userret: thread %p (pid %d, %s)", td, p->p_pid,
102             td->td_name);
103 	KASSERT((p->p_flag & P_WEXIT) == 0,
104 	    ("Exiting process returns to usermode"));
105 #ifdef DIAGNOSTIC
106 	/*
107 	 * Check that we called signotify() enough.  For
108 	 * multi-threaded processes, where signal distribution might
109 	 * change due to other threads changing sigmask, the check is
110 	 * racy and cannot be performed reliably.
111 	 * If current process is vfork child, indicated by P_PPWAIT, then
112 	 * issignal() ignores stops, so we block the check to avoid
113 	 * classifying pending signals.
114 	 */
115 	if (p->p_numthreads == 1) {
116 		PROC_LOCK(p);
117 		thread_lock(td);
118 		if ((p->p_flag & P_PPWAIT) == 0 &&
119 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
120 			if (SIGPENDING(td) && (td->td_flags &
121 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING)) !=
122 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING)) {
123 				thread_unlock(td);
124 				panic(
125 	"failed to set signal flags for ast p %p td %p fl %x",
126 				    p, td, td->td_flags);
127 			}
128 		}
129 		thread_unlock(td);
130 		PROC_UNLOCK(p);
131 	}
132 #endif
133 
134 	/*
135 	 * Charge system time if profiling.
136 	 */
137 	if (__predict_false(p->p_flag & P_PROFIL))
138 		addupc_task(td, TRAPF_PC(frame), td->td_pticks * psratio);
139 
140 #ifdef HWPMC_HOOKS
141 	if (PMC_THREAD_HAS_SAMPLES(td))
142 		PMC_CALL_HOOK(td, PMC_FN_THR_USERRET, NULL);
143 #endif
144 	/*
145 	 * Let the scheduler adjust our priority etc.
146 	 */
147 	sched_userret(td);
148 
149 	/*
150 	 * Check for misbehavior.
151 	 *
152 	 * In case there is a callchain tracing ongoing because of
153 	 * hwpmc(4), skip the scheduler pinning check.
154 	 * hwpmc(4) subsystem, infact, will collect callchain informations
155 	 * at ast() checkpoint, which is past userret().
156 	 */
157 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
158 	KASSERT(td->td_critnest == 0,
159 	    ("userret: Returning in a critical section"));
160 	KASSERT(td->td_locks == 0,
161 	    ("userret: Returning with %d locks held", td->td_locks));
162 	KASSERT(td->td_rw_rlocks == 0,
163 	    ("userret: Returning with %d rwlocks held in read mode",
164 	    td->td_rw_rlocks));
165 	KASSERT(td->td_sx_slocks == 0,
166 	    ("userret: Returning with %d sx locks held in shared mode",
167 	    td->td_sx_slocks));
168 	KASSERT(td->td_lk_slocks == 0,
169 	    ("userret: Returning with %d lockmanager locks held in shared mode",
170 	    td->td_lk_slocks));
171 	KASSERT((td->td_pflags & TDP_NOFAULTING) == 0,
172 	    ("userret: Returning with pagefaults disabled"));
173 	if (__predict_false(!THREAD_CAN_SLEEP())) {
174 #ifdef EPOCH_TRACE
175 		epoch_trace_list(curthread);
176 #endif
177 		KASSERT(0, ("userret: Returning with sleep disabled"));
178 	}
179 	KASSERT(td->td_pinned == 0 || (td->td_pflags & TDP_CALLCHAIN) != 0,
180 	    ("userret: Returning with with pinned thread"));
181 	KASSERT(td->td_vp_reserved == NULL,
182 	    ("userret: Returning with preallocated vnode"));
183 	KASSERT((td->td_flags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
184 	    ("userret: Returning with stop signals deferred"));
185 	KASSERT(td->td_vslock_sz == 0,
186 	    ("userret: Returning with vslock-wired space"));
187 #ifdef VIMAGE
188 	/* Unfortunately td_vnet_lpush needs VNET_DEBUG. */
189 	VNET_ASSERT(curvnet == NULL,
190 	    ("%s: Returning on td %p (pid %d, %s) with vnet %p set in %s",
191 	    __func__, td, p->p_pid, td->td_name, curvnet,
192 	    (td->td_vnet_lpush != NULL) ? td->td_vnet_lpush : "N/A"));
193 #endif
194 }
195 
196 /*
197  * Process an asynchronous software trap.
198  * This is relatively easy.
199  * This function will return with preemption disabled.
200  */
201 void
202 ast(struct trapframe *framep)
203 {
204 	struct thread *td;
205 	struct proc *p;
206 	int flags, sig;
207 	bool resched_sigs;
208 
209 	td = curthread;
210 	p = td->td_proc;
211 
212 	CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td, p->p_pid,
213             p->p_comm);
214 	KASSERT(TRAPF_USERMODE(framep), ("ast in kernel mode"));
215 	WITNESS_WARN(WARN_PANIC, NULL, "Returning to user mode");
216 	mtx_assert(&Giant, MA_NOTOWNED);
217 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
218 	td->td_frame = framep;
219 	td->td_pticks = 0;
220 
221 	/*
222 	 * This updates the td_flag's for the checks below in one
223 	 * "atomic" operation with turning off the astpending flag.
224 	 * If another AST is triggered while we are handling the
225 	 * AST's saved in flags, the astpending flag will be set and
226 	 * ast() will be called again.
227 	 */
228 	thread_lock(td);
229 	flags = td->td_flags;
230 	td->td_flags &= ~(TDF_ASTPENDING | TDF_NEEDSIGCHK | TDF_NEEDSUSPCHK |
231 	    TDF_NEEDRESCHED | TDF_ALRMPEND | TDF_PROFPEND | TDF_MACPEND);
232 	thread_unlock(td);
233 	VM_CNT_INC(v_trap);
234 
235 	if (td->td_cowgen != p->p_cowgen)
236 		thread_cow_update(td);
237 	if (td->td_pflags & TDP_OWEUPC && p->p_flag & P_PROFIL) {
238 		addupc_task(td, td->td_profil_addr, td->td_profil_ticks);
239 		td->td_profil_ticks = 0;
240 		td->td_pflags &= ~TDP_OWEUPC;
241 	}
242 #ifdef HWPMC_HOOKS
243 	/* Handle Software PMC callchain capture. */
244 	if (PMC_IS_PENDING_CALLCHAIN(td))
245 		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_USER_CALLCHAIN_SOFT, (void *) framep);
246 #endif
247 	if (flags & TDF_ALRMPEND) {
248 		PROC_LOCK(p);
249 		kern_psignal(p, SIGVTALRM);
250 		PROC_UNLOCK(p);
251 	}
252 	if (flags & TDF_PROFPEND) {
253 		PROC_LOCK(p);
254 		kern_psignal(p, SIGPROF);
255 		PROC_UNLOCK(p);
256 	}
257 #ifdef MAC
258 	if (flags & TDF_MACPEND)
259 		mac_thread_userret(td);
260 #endif
261 	if (flags & TDF_NEEDRESCHED) {
262 #ifdef KTRACE
263 		if (KTRPOINT(td, KTR_CSW))
264 			ktrcsw(1, 1, __func__);
265 #endif
266 		thread_lock(td);
267 		sched_prio(td, td->td_user_pri);
268 		mi_switch(SW_INVOL | SWT_NEEDRESCHED);
269 #ifdef KTRACE
270 		if (KTRPOINT(td, KTR_CSW))
271 			ktrcsw(0, 1, __func__);
272 #endif
273 	}
274 
275 	td_softdep_cleanup(td);
276 	MPASS(td->td_su == NULL);
277 
278 	/*
279 	 * If this thread tickled GEOM, we need to wait for the giggling to
280 	 * stop before we return to userland
281 	 */
282 	if (__predict_false(td->td_pflags & TDP_GEOM))
283 		g_waitidle();
284 
285 #ifdef DIAGNOSTIC
286 	if (p->p_numthreads == 1 && (flags & TDF_NEEDSIGCHK) == 0) {
287 		PROC_LOCK(p);
288 		thread_lock(td);
289 		/*
290 		 * Note that TDF_NEEDSIGCHK should be re-read from
291 		 * td_flags, since signal might have been delivered
292 		 * after we cleared td_flags above.  This is one of
293 		 * the reason for looping check for AST condition.
294 		 * See comment in userret() about P_PPWAIT.
295 		 */
296 		if ((p->p_flag & P_PPWAIT) == 0 &&
297 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
298 			if (SIGPENDING(td) && (td->td_flags &
299 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING)) !=
300 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING)) {
301 				thread_unlock(td); /* fix dumps */
302 				panic(
303 	"failed2 to set signal flags for ast p %p td %p fl %x %x",
304 				    p, td, flags, td->td_flags);
305 			}
306 		}
307 		thread_unlock(td);
308 		PROC_UNLOCK(p);
309 	}
310 #endif
311 
312 	/*
313 	 * Check for signals. Unlocked reads of p_pendingcnt or
314 	 * p_siglist might cause process-directed signal to be handled
315 	 * later.
316 	 */
317 	if (flags & TDF_NEEDSIGCHK || p->p_pendingcnt > 0 ||
318 	    !SIGISEMPTY(p->p_siglist)) {
319 		sigfastblock_fetch(td);
320 		PROC_LOCK(p);
321 		mtx_lock(&p->p_sigacts->ps_mtx);
322 		while ((sig = cursig(td)) != 0) {
323 			KASSERT(sig >= 0, ("sig %d", sig));
324 			postsig(sig);
325 		}
326 		mtx_unlock(&p->p_sigacts->ps_mtx);
327 		PROC_UNLOCK(p);
328 		resched_sigs = true;
329 	} else {
330 		resched_sigs = false;
331 	}
332 
333 	/*
334 	 * Handle deferred update of the fast sigblock value, after
335 	 * the postsig() loop was performed.
336 	 */
337 	sigfastblock_setpend(td, resched_sigs);
338 
339 #ifdef KTRACE
340 	KTRUSERRET(td);
341 #endif
342 
343 	/*
344 	 * We need to check to see if we have to exit or wait due to a
345 	 * single threading requirement or some other STOP condition.
346 	 */
347 	if (flags & TDF_NEEDSUSPCHK) {
348 		PROC_LOCK(p);
349 		thread_suspend_check(0);
350 		PROC_UNLOCK(p);
351 	}
352 
353 	if (td->td_pflags & TDP_OLDMASK) {
354 		td->td_pflags &= ~TDP_OLDMASK;
355 		kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
356 	}
357 
358 #ifdef RACCT
359 	if (__predict_false(racct_enable && p->p_throttled != 0))
360 		racct_proc_throttled(p);
361 #endif
362 
363 	userret(td, framep);
364 }
365 
366 const char *
367 syscallname(struct proc *p, u_int code)
368 {
369 	static const char unknown[] = "unknown";
370 	struct sysentvec *sv;
371 
372 	sv = p->p_sysent;
373 	if (sv->sv_syscallnames == NULL || code >= sv->sv_size)
374 		return (unknown);
375 	return (sv->sv_syscallnames[code]);
376 }
377