xref: /freebsd/sys/kern/subr_trap.c (revision 74dc6beb30dcb5c57134d6b66e14937b42f911c0)
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 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
121 			if (SIGPENDING(td) && (td->td_flags &
122 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING)) !=
123 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING)) {
124 				thread_unlock(td);
125 				panic(
126 	"failed to set signal flags for ast p %p td %p fl %x",
127 				    p, td, td->td_flags);
128 			}
129 		}
130 		thread_unlock(td);
131 		PROC_UNLOCK(p);
132 	}
133 #endif
134 #ifdef KTRACE
135 	KTRUSERRET(td);
136 #endif
137 	td_softdep_cleanup(td);
138 	MPASS(td->td_su == NULL);
139 
140 	/*
141 	 * If this thread tickled GEOM, we need to wait for the giggling to
142 	 * stop before we return to userland
143 	 */
144 	if (__predict_false(td->td_pflags & TDP_GEOM))
145 		g_waitidle();
146 
147 	/*
148 	 * Charge system time if profiling.
149 	 */
150 	if (__predict_false(p->p_flag & P_PROFIL))
151 		addupc_task(td, TRAPF_PC(frame), td->td_pticks * psratio);
152 
153 #ifdef HWPMC_HOOKS
154 	if (PMC_THREAD_HAS_SAMPLES(td))
155 		PMC_CALL_HOOK(td, PMC_FN_THR_USERRET, NULL);
156 #endif
157 	/*
158 	 * Let the scheduler adjust our priority etc.
159 	 */
160 	sched_userret(td);
161 
162 	/*
163 	 * Check for misbehavior.
164 	 *
165 	 * In case there is a callchain tracing ongoing because of
166 	 * hwpmc(4), skip the scheduler pinning check.
167 	 * hwpmc(4) subsystem, infact, will collect callchain informations
168 	 * at ast() checkpoint, which is past userret().
169 	 */
170 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
171 	KASSERT(td->td_critnest == 0,
172 	    ("userret: Returning in a critical section"));
173 	KASSERT(td->td_locks == 0,
174 	    ("userret: Returning with %d locks held", td->td_locks));
175 	KASSERT(td->td_rw_rlocks == 0,
176 	    ("userret: Returning with %d rwlocks held in read mode",
177 	    td->td_rw_rlocks));
178 	KASSERT(td->td_sx_slocks == 0,
179 	    ("userret: Returning with %d sx locks held in shared mode",
180 	    td->td_sx_slocks));
181 	KASSERT(td->td_lk_slocks == 0,
182 	    ("userret: Returning with %d lockmanager locks held in shared mode",
183 	    td->td_lk_slocks));
184 	KASSERT((td->td_pflags & TDP_NOFAULTING) == 0,
185 	    ("userret: Returning with pagefaults disabled"));
186 	if (__predict_false(!THREAD_CAN_SLEEP())) {
187 #ifdef EPOCH_TRACE
188 		epoch_trace_list(curthread);
189 #endif
190 		KASSERT(1, ("userret: Returning with sleep disabled"));
191 	}
192 	KASSERT(td->td_pinned == 0 || (td->td_pflags & TDP_CALLCHAIN) != 0,
193 	    ("userret: Returning with with pinned thread"));
194 	KASSERT(td->td_vp_reserved == NULL,
195 	    ("userret: Returning with preallocated vnode"));
196 	KASSERT((td->td_flags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
197 	    ("userret: Returning with stop signals deferred"));
198 	KASSERT(td->td_su == NULL,
199 	    ("userret: Returning with SU cleanup request not handled"));
200 	KASSERT(td->td_vslock_sz == 0,
201 	    ("userret: Returning with vslock-wired space"));
202 #ifdef VIMAGE
203 	/* Unfortunately td_vnet_lpush needs VNET_DEBUG. */
204 	VNET_ASSERT(curvnet == NULL,
205 	    ("%s: Returning on td %p (pid %d, %s) with vnet %p set in %s",
206 	    __func__, td, p->p_pid, td->td_name, curvnet,
207 	    (td->td_vnet_lpush != NULL) ? td->td_vnet_lpush : "N/A"));
208 #endif
209 #ifdef RACCT
210 	if (__predict_false(racct_enable && p->p_throttled != 0))
211 		racct_proc_throttled(p);
212 #endif
213 }
214 
215 /*
216  * Process an asynchronous software trap.
217  * This is relatively easy.
218  * This function will return with preemption disabled.
219  */
220 void
221 ast(struct trapframe *framep)
222 {
223 	struct thread *td;
224 	struct proc *p;
225 	uint32_t oldval;
226 	int flags, sig, res;
227 
228 	td = curthread;
229 	p = td->td_proc;
230 
231 	CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td, p->p_pid,
232             p->p_comm);
233 	KASSERT(TRAPF_USERMODE(framep), ("ast in kernel mode"));
234 	WITNESS_WARN(WARN_PANIC, NULL, "Returning to user mode");
235 	mtx_assert(&Giant, MA_NOTOWNED);
236 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
237 	td->td_frame = framep;
238 	td->td_pticks = 0;
239 
240 	/*
241 	 * This updates the td_flag's for the checks below in one
242 	 * "atomic" operation with turning off the astpending flag.
243 	 * If another AST is triggered while we are handling the
244 	 * AST's saved in flags, the astpending flag will be set and
245 	 * ast() will be called again.
246 	 */
247 	thread_lock(td);
248 	flags = td->td_flags;
249 	td->td_flags &= ~(TDF_ASTPENDING | TDF_NEEDSIGCHK | TDF_NEEDSUSPCHK |
250 	    TDF_NEEDRESCHED | TDF_ALRMPEND | TDF_PROFPEND | TDF_MACPEND);
251 	thread_unlock(td);
252 	VM_CNT_INC(v_trap);
253 
254 	if (td->td_cowgen != p->p_cowgen)
255 		thread_cow_update(td);
256 	if (td->td_pflags & TDP_OWEUPC && p->p_flag & P_PROFIL) {
257 		addupc_task(td, td->td_profil_addr, td->td_profil_ticks);
258 		td->td_profil_ticks = 0;
259 		td->td_pflags &= ~TDP_OWEUPC;
260 	}
261 #ifdef HWPMC_HOOKS
262 	/* Handle Software PMC callchain capture. */
263 	if (PMC_IS_PENDING_CALLCHAIN(td))
264 		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_USER_CALLCHAIN_SOFT, (void *) framep);
265 #endif
266 	if (flags & TDF_ALRMPEND) {
267 		PROC_LOCK(p);
268 		kern_psignal(p, SIGVTALRM);
269 		PROC_UNLOCK(p);
270 	}
271 	if (flags & TDF_PROFPEND) {
272 		PROC_LOCK(p);
273 		kern_psignal(p, SIGPROF);
274 		PROC_UNLOCK(p);
275 	}
276 #ifdef MAC
277 	if (flags & TDF_MACPEND)
278 		mac_thread_userret(td);
279 #endif
280 	if (flags & TDF_NEEDRESCHED) {
281 #ifdef KTRACE
282 		if (KTRPOINT(td, KTR_CSW))
283 			ktrcsw(1, 1, __func__);
284 #endif
285 		thread_lock(td);
286 		sched_prio(td, td->td_user_pri);
287 		mi_switch(SW_INVOL | SWT_NEEDRESCHED);
288 #ifdef KTRACE
289 		if (KTRPOINT(td, KTR_CSW))
290 			ktrcsw(0, 1, __func__);
291 #endif
292 	}
293 
294 #ifdef DIAGNOSTIC
295 	if (p->p_numthreads == 1 && (flags & TDF_NEEDSIGCHK) == 0) {
296 		PROC_LOCK(p);
297 		thread_lock(td);
298 		/*
299 		 * Note that TDF_NEEDSIGCHK should be re-read from
300 		 * td_flags, since signal might have been delivered
301 		 * after we cleared td_flags above.  This is one of
302 		 * the reason for looping check for AST condition.
303 		 * See comment in userret() about P_PPWAIT.
304 		 */
305 		if ((p->p_flag & P_PPWAIT) == 0 &&
306 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
307 			if (SIGPENDING(td) && (td->td_flags &
308 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING)) !=
309 			    (TDF_NEEDSIGCHK | TDF_ASTPENDING)) {
310 				thread_unlock(td); /* fix dumps */
311 				panic(
312 	"failed2 to set signal flags for ast p %p td %p fl %x %x",
313 				    p, td, flags, td->td_flags);
314 			}
315 		}
316 		thread_unlock(td);
317 		PROC_UNLOCK(p);
318 	}
319 #endif
320 
321 	/*
322 	 * Check for signals. Unlocked reads of p_pendingcnt or
323 	 * p_siglist might cause process-directed signal to be handled
324 	 * later.
325 	 */
326 	if (flags & TDF_NEEDSIGCHK || p->p_pendingcnt > 0 ||
327 	    !SIGISEMPTY(p->p_siglist)) {
328 		fetch_sigfastblock(td);
329 		PROC_LOCK(p);
330 		mtx_lock(&p->p_sigacts->ps_mtx);
331 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0 &&
332 		    td->td_sigblock_val != 0) {
333 			reschedule_signals(p, fastblock_mask,
334 			    SIGPROCMASK_PS_LOCKED | SIGPROCMASK_FASTBLK);
335 		} else {
336 			while ((sig = cursig(td)) != 0) {
337 				KASSERT(sig >= 0, ("sig %d", sig));
338 				postsig(sig);
339 			}
340 		}
341 		mtx_unlock(&p->p_sigacts->ps_mtx);
342 		PROC_UNLOCK(p);
343 	}
344 
345 	/*
346 	 * Handle deferred update of the fast sigblock value, after
347 	 * the postsig() loop was performed.
348 	 */
349 	if (td->td_pflags & TDP_SIGFASTPENDING) {
350 		td->td_pflags &= ~TDP_SIGFASTPENDING;
351 		res = fueword32(td->td_sigblock_ptr, &oldval);
352 		if (res == -1) {
353 			fetch_sigfastblock_failed(td, false);
354 		} else {
355 			for (;;) {
356 				oldval |= SIGFASTBLOCK_PEND;
357 				res = casueword32(td->td_sigblock_ptr, oldval,
358 				    &oldval, oldval | SIGFASTBLOCK_PEND);
359 				if (res == -1) {
360 					fetch_sigfastblock_failed(td, true);
361 					break;
362 				}
363 				if (res == 0) {
364 					td->td_sigblock_val = oldval &
365 					    ~SIGFASTBLOCK_FLAGS;
366 					break;
367 				}
368 				MPASS(res == 1);
369 				res = thread_check_susp(td, false);
370 				if (res != 0)
371 					break;
372 			}
373 		}
374 	}
375 
376 	/*
377 	 * We need to check to see if we have to exit or wait due to a
378 	 * single threading requirement or some other STOP condition.
379 	 */
380 	if (flags & TDF_NEEDSUSPCHK) {
381 		PROC_LOCK(p);
382 		thread_suspend_check(0);
383 		PROC_UNLOCK(p);
384 	}
385 
386 	if (td->td_pflags & TDP_OLDMASK) {
387 		td->td_pflags &= ~TDP_OLDMASK;
388 		kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
389 	}
390 
391 	userret(td, framep);
392 }
393 
394 const char *
395 syscallname(struct proc *p, u_int code)
396 {
397 	static const char unknown[] = "unknown";
398 	struct sysentvec *sv;
399 
400 	sv = p->p_sysent;
401 	if (sv->sv_syscallnames == NULL || code >= sv->sv_size)
402 		return (unknown);
403 	return (sv->sv_syscallnames[code]);
404 }
405