xref: /freebsd/sys/kern/subr_trap.c (revision 74ca7bf1d4c7173d5575ba168bc4b5f6d181ff5a)
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/pioctl.h>
63 #include <sys/ptrace.h>
64 #include <sys/racct.h>
65 #include <sys/resourcevar.h>
66 #include <sys/sched.h>
67 #include <sys/signalvar.h>
68 #include <sys/syscall.h>
69 #include <sys/syscallsubr.h>
70 #include <sys/sysent.h>
71 #include <sys/systm.h>
72 #include <sys/vmmeter.h>
73 #ifdef KTRACE
74 #include <sys/uio.h>
75 #include <sys/ktrace.h>
76 #endif
77 #include <security/audit/audit.h>
78 
79 #include <machine/cpu.h>
80 
81 #ifdef VIMAGE
82 #include <net/vnet.h>
83 #endif
84 
85 #ifdef	HWPMC_HOOKS
86 #include <sys/pmckern.h>
87 #endif
88 
89 #include <security/mac/mac_framework.h>
90 
91 void (*softdep_ast_cleanup)(struct thread *);
92 
93 /*
94  * Define the code needed before returning to user mode, for trap and
95  * syscall.
96  */
97 void
98 userret(struct thread *td, struct trapframe *frame)
99 {
100 	struct proc *p = td->td_proc;
101 
102 	CTR3(KTR_SYSC, "userret: thread %p (pid %d, %s)", td, p->p_pid,
103             td->td_name);
104 	KASSERT((p->p_flag & P_WEXIT) == 0,
105 	    ("Exiting process returns to usermode"));
106 #ifdef DIAGNOSTIC
107 	/*
108 	 * Check that we called signotify() enough.  For
109 	 * multi-threaded processes, where signal distribution might
110 	 * change due to other threads changing sigmask, the check is
111 	 * racy and cannot be performed reliably.
112 	 * If current process is vfork child, indicated by P_PPWAIT, then
113 	 * issignal() ignores stops, so we block the check to avoid
114 	 * classifying pending signals.
115 	 */
116 	if (p->p_numthreads == 1) {
117 		PROC_LOCK(p);
118 		thread_lock(td);
119 		if ((p->p_flag & P_PPWAIT) == 0) {
120 			KASSERT(!SIGPENDING(td) || (td->td_flags &
121 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING)) ==
122 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING),
123 			    ("failed to set signal flags for ast p %p "
124 			    "td %p fl %x", p, td, td->td_flags));
125 		}
126 		thread_unlock(td);
127 		PROC_UNLOCK(p);
128 	}
129 #endif
130 #ifdef KTRACE
131 	KTRUSERRET(td);
132 #endif
133 	td_softdep_cleanup(td);
134 	MPASS(td->td_su == NULL);
135 
136 	/*
137 	 * If this thread tickled GEOM, we need to wait for the giggling to
138 	 * stop before we return to userland
139 	 */
140 	if (td->td_pflags & TDP_GEOM)
141 		g_waitidle();
142 
143 	/*
144 	 * Charge system time if profiling.
145 	 */
146 	if (p->p_flag & P_PROFIL)
147 		addupc_task(td, TRAPF_PC(frame), td->td_pticks * psratio);
148 	/*
149 	 * Let the scheduler adjust our priority etc.
150 	 */
151 	sched_userret(td);
152 
153 	/*
154 	 * Check for misbehavior.
155 	 *
156 	 * In case there is a callchain tracing ongoing because of
157 	 * hwpmc(4), skip the scheduler pinning check.
158 	 * hwpmc(4) subsystem, infact, will collect callchain informations
159 	 * at ast() checkpoint, which is past userret().
160 	 */
161 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
162 	KASSERT(td->td_critnest == 0,
163 	    ("userret: Returning in a critical section"));
164 	KASSERT(td->td_locks == 0,
165 	    ("userret: Returning with %d locks held", td->td_locks));
166 	KASSERT(td->td_rw_rlocks == 0,
167 	    ("userret: Returning with %d rwlocks held in read mode",
168 	    td->td_rw_rlocks));
169 	KASSERT((td->td_pflags & TDP_NOFAULTING) == 0,
170 	    ("userret: Returning with pagefaults disabled"));
171 	KASSERT(td->td_no_sleeping == 0,
172 	    ("userret: Returning with sleep disabled"));
173 	KASSERT(td->td_pinned == 0 || (td->td_pflags & TDP_CALLCHAIN) != 0,
174 	    ("userret: Returning with with pinned thread"));
175 	KASSERT(td->td_vp_reserv == 0,
176 	    ("userret: Returning while holding vnode reservation"));
177 	KASSERT((td->td_flags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
178 	    ("userret: Returning with stop signals deferred"));
179 	KASSERT(td->td_su == NULL,
180 	    ("userret: Returning with SU cleanup request not handled"));
181 	KASSERT(td->td_vslock_sz == 0,
182 	    ("userret: Returning with vslock-wired space"));
183 #ifdef VIMAGE
184 	/* Unfortunately td_vnet_lpush needs VNET_DEBUG. */
185 	VNET_ASSERT(curvnet == NULL,
186 	    ("%s: Returning on td %p (pid %d, %s) with vnet %p set in %s",
187 	    __func__, td, p->p_pid, td->td_name, curvnet,
188 	    (td->td_vnet_lpush != NULL) ? td->td_vnet_lpush : "N/A"));
189 #endif
190 #ifdef RACCT
191 	if (racct_enable && p->p_throttled != 0) {
192 		PROC_LOCK(p);
193 		while (p->p_throttled != 0) {
194 			msleep(p->p_racct, &p->p_mtx, 0, "racct",
195 			    p->p_throttled < 0 ? 0 : p->p_throttled);
196 			if (p->p_throttled > 0)
197 				p->p_throttled = 0;
198 		}
199 		PROC_UNLOCK(p);
200 	}
201 #endif
202 }
203 
204 /*
205  * Process an asynchronous software trap.
206  * This is relatively easy.
207  * This function will return with preemption disabled.
208  */
209 void
210 ast(struct trapframe *framep)
211 {
212 	struct thread *td;
213 	struct proc *p;
214 	int flags;
215 	int sig;
216 
217 	td = curthread;
218 	p = td->td_proc;
219 
220 	CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td, p->p_pid,
221             p->p_comm);
222 	KASSERT(TRAPF_USERMODE(framep), ("ast in kernel mode"));
223 	WITNESS_WARN(WARN_PANIC, NULL, "Returning to user mode");
224 	mtx_assert(&Giant, MA_NOTOWNED);
225 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
226 	td->td_frame = framep;
227 	td->td_pticks = 0;
228 
229 	/*
230 	 * This updates the td_flag's for the checks below in one
231 	 * "atomic" operation with turning off the astpending flag.
232 	 * If another AST is triggered while we are handling the
233 	 * AST's saved in flags, the astpending flag will be set and
234 	 * ast() will be called again.
235 	 */
236 	thread_lock(td);
237 	flags = td->td_flags;
238 	td->td_flags &= ~(TDF_ASTPENDING | TDF_NEEDSIGCHK | TDF_NEEDSUSPCHK |
239 	    TDF_NEEDRESCHED | TDF_ALRMPEND | TDF_PROFPEND | TDF_MACPEND);
240 	thread_unlock(td);
241 	VM_CNT_INC(v_trap);
242 
243 	if (td->td_cowgen != p->p_cowgen)
244 		thread_cow_update(td);
245 	if (td->td_pflags & TDP_OWEUPC && p->p_flag & P_PROFIL) {
246 		addupc_task(td, td->td_profil_addr, td->td_profil_ticks);
247 		td->td_profil_ticks = 0;
248 		td->td_pflags &= ~TDP_OWEUPC;
249 	}
250 #ifdef HWPMC_HOOKS
251 	/* Handle Software PMC callchain capture. */
252 	if (PMC_IS_PENDING_CALLCHAIN(td))
253 		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_USER_CALLCHAIN_SOFT, (void *) framep);
254 #endif
255 	if (flags & TDF_ALRMPEND) {
256 		PROC_LOCK(p);
257 		kern_psignal(p, SIGVTALRM);
258 		PROC_UNLOCK(p);
259 	}
260 	if (flags & TDF_PROFPEND) {
261 		PROC_LOCK(p);
262 		kern_psignal(p, SIGPROF);
263 		PROC_UNLOCK(p);
264 	}
265 #ifdef MAC
266 	if (flags & TDF_MACPEND)
267 		mac_thread_userret(td);
268 #endif
269 	if (flags & TDF_NEEDRESCHED) {
270 #ifdef KTRACE
271 		if (KTRPOINT(td, KTR_CSW))
272 			ktrcsw(1, 1, __func__);
273 #endif
274 		thread_lock(td);
275 		sched_prio(td, td->td_user_pri);
276 		mi_switch(SW_INVOL | SWT_NEEDRESCHED, NULL);
277 		thread_unlock(td);
278 #ifdef KTRACE
279 		if (KTRPOINT(td, KTR_CSW))
280 			ktrcsw(0, 1, __func__);
281 #endif
282 	}
283 
284 #ifdef DIAGNOSTIC
285 	if (p->p_numthreads == 1 && (flags & TDF_NEEDSIGCHK) == 0) {
286 		PROC_LOCK(p);
287 		thread_lock(td);
288 		/*
289 		 * Note that TDF_NEEDSIGCHK should be re-read from
290 		 * td_flags, since signal might have been delivered
291 		 * after we cleared td_flags above.  This is one of
292 		 * the reason for looping check for AST condition.
293 		 * See comment in userret() about P_PPWAIT.
294 		 */
295 		if ((p->p_flag & P_PPWAIT) == 0) {
296 			KASSERT(!SIGPENDING(td) || (td->td_flags &
297 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING)) ==
298 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING),
299 			    ("failed2 to set signal flags for ast p %p td %p "
300 			    "fl %x %x", p, td, flags, td->td_flags));
301 		}
302 		thread_unlock(td);
303 		PROC_UNLOCK(p);
304 	}
305 #endif
306 
307 	/*
308 	 * Check for signals. Unlocked reads of p_pendingcnt or
309 	 * p_siglist might cause process-directed signal to be handled
310 	 * later.
311 	 */
312 	if (flags & TDF_NEEDSIGCHK || p->p_pendingcnt > 0 ||
313 	    !SIGISEMPTY(p->p_siglist)) {
314 		PROC_LOCK(p);
315 		mtx_lock(&p->p_sigacts->ps_mtx);
316 		while ((sig = cursig(td)) != 0) {
317 			KASSERT(sig >= 0, ("sig %d", sig));
318 			postsig(sig);
319 		}
320 		mtx_unlock(&p->p_sigacts->ps_mtx);
321 		PROC_UNLOCK(p);
322 	}
323 	/*
324 	 * We need to check to see if we have to exit or wait due to a
325 	 * single threading requirement or some other STOP condition.
326 	 */
327 	if (flags & TDF_NEEDSUSPCHK) {
328 		PROC_LOCK(p);
329 		thread_suspend_check(0);
330 		PROC_UNLOCK(p);
331 	}
332 
333 	if (td->td_pflags & TDP_OLDMASK) {
334 		td->td_pflags &= ~TDP_OLDMASK;
335 		kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
336 	}
337 
338 	userret(td, framep);
339 }
340 
341 const char *
342 syscallname(struct proc *p, u_int code)
343 {
344 	static const char unknown[] = "unknown";
345 	struct sysentvec *sv;
346 
347 	sv = p->p_sysent;
348 	if (sv->sv_syscallnames == NULL || code >= sv->sv_size)
349 		return (unknown);
350 	return (sv->sv_syscallnames[code]);
351 }
352