xref: /freebsd/sys/kern/subr_trap.c (revision 2008043f386721d58158e37e0d7e50df8095942d)
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, 2022 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 #include "opt_hwpmc_hooks.h"
48 
49 #include <sys/param.h>
50 #include <sys/kernel.h>
51 #include <sys/limits.h>
52 #include <sys/lock.h>
53 #include <sys/msan.h>
54 #include <sys/mutex.h>
55 #include <sys/proc.h>
56 #include <sys/ktr.h>
57 #include <sys/resourcevar.h>
58 #include <sys/sched.h>
59 #include <sys/syscall.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysent.h>
62 #include <sys/systm.h>
63 #include <sys/vmmeter.h>
64 
65 #include <machine/cpu.h>
66 
67 #ifdef VIMAGE
68 #include <net/vnet.h>
69 #endif
70 
71 #ifdef	HWPMC_HOOKS
72 #include <sys/pmckern.h>
73 #endif
74 
75 #ifdef EPOCH_TRACE
76 #include <sys/epoch.h>
77 #endif
78 
79 /*
80  * Define the code needed before returning to user mode, for trap and
81  * syscall.
82  */
83 void
84 userret(struct thread *td, struct trapframe *frame)
85 {
86 	struct proc *p = td->td_proc;
87 
88 	CTR3(KTR_SYSC, "userret: thread %p (pid %d, %s)", td, p->p_pid,
89             td->td_name);
90 	KASSERT((p->p_flag & P_WEXIT) == 0,
91 	    ("Exiting process returns to usermode"));
92 #ifdef DIAGNOSTIC
93 	/*
94 	 * Check that we called signotify() enough.  For
95 	 * multi-threaded processes, where signal distribution might
96 	 * change due to other threads changing sigmask, the check is
97 	 * racy and cannot be performed reliably.
98 	 * If current process is vfork child, indicated by P_PPWAIT, then
99 	 * issignal() ignores stops, so we block the check to avoid
100 	 * classifying pending signals.
101 	 */
102 	if (p->p_numthreads == 1) {
103 		PROC_LOCK(p);
104 		thread_lock(td);
105 		if ((p->p_flag & P_PPWAIT) == 0 &&
106 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0 &&
107 		    SIGPENDING(td) && !td_ast_pending(td, TDA_AST) &&
108 		    !td_ast_pending(td, TDA_SIG)) {
109 			thread_unlock(td);
110 			panic(
111 			    "failed to set signal flags for ast p %p "
112 			    "td %p td_ast %#x fl %#x",
113 			    p, td, td->td_ast, td->td_flags);
114 		}
115 		thread_unlock(td);
116 		PROC_UNLOCK(p);
117 	}
118 #endif
119 
120 	/*
121 	 * Charge system time if profiling.
122 	 */
123 	if (__predict_false(p->p_flag & P_PROFIL))
124 		addupc_task(td, TRAPF_PC(frame), td->td_pticks * psratio);
125 
126 #ifdef HWPMC_HOOKS
127 	if (PMC_THREAD_HAS_SAMPLES(td))
128 		PMC_CALL_HOOK(td, PMC_FN_THR_USERRET, NULL);
129 #endif
130 #ifdef TCPHPTS
131 	/*
132 	 * @gallatin is adament that this needs to go here, I
133 	 * am not so sure. Running hpts is a lot like
134 	 * a lro_flush() that happens while a user process
135 	 * is running. But he may know best so I will go
136 	 * with his view of accounting. :-)
137 	 */
138 	tcp_run_hpts();
139 #endif
140 	/*
141 	 * Let the scheduler adjust our priority etc.
142 	 */
143 	sched_userret(td);
144 
145 	/*
146 	 * Check for misbehavior.
147 	 *
148 	 * In case there is a callchain tracing ongoing because of
149 	 * hwpmc(4), skip the scheduler pinning check.
150 	 * hwpmc(4) subsystem, infact, will collect callchain informations
151 	 * at ast() checkpoint, which is past userret().
152 	 */
153 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
154 	KASSERT(td->td_critnest == 0,
155 	    ("userret: Returning in a critical section"));
156 	KASSERT(td->td_locks == 0,
157 	    ("userret: Returning with %d locks held", td->td_locks));
158 	KASSERT(td->td_rw_rlocks == 0,
159 	    ("userret: Returning with %d rwlocks held in read mode",
160 	    td->td_rw_rlocks));
161 	KASSERT(td->td_sx_slocks == 0,
162 	    ("userret: Returning with %d sx locks held in shared mode",
163 	    td->td_sx_slocks));
164 	KASSERT(td->td_lk_slocks == 0,
165 	    ("userret: Returning with %d lockmanager locks held in shared mode",
166 	    td->td_lk_slocks));
167 	KASSERT((td->td_pflags & TDP_NOFAULTING) == 0,
168 	    ("userret: Returning with pagefaults disabled"));
169 	if (__predict_false(!THREAD_CAN_SLEEP())) {
170 #ifdef EPOCH_TRACE
171 		epoch_trace_list(curthread);
172 #endif
173 		KASSERT(0, ("userret: Returning with sleep disabled"));
174 	}
175 	KASSERT(td->td_pinned == 0 || (td->td_pflags & TDP_CALLCHAIN) != 0,
176 	    ("userret: Returning with pinned thread"));
177 	KASSERT(td->td_vp_reserved == NULL,
178 	    ("userret: Returning with preallocated vnode"));
179 	KASSERT((td->td_flags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
180 	    ("userret: Returning with stop signals deferred"));
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 }
191 
192 static void
193 ast_prep(struct thread *td, int tda __unused)
194 {
195 	VM_CNT_INC(v_trap);
196 	td->td_pticks = 0;
197 	if (td->td_cowgen != atomic_load_int(&td->td_proc->p_cowgen))
198 		thread_cow_update(td);
199 
200 }
201 
202 struct ast_entry {
203 	int	ae_flags;
204 	int	ae_tdp;
205 	void	(*ae_f)(struct thread *td, int ast);
206 };
207 
208 _Static_assert(TDAI(TDA_MAX) <= UINT_MAX, "Too many ASTs");
209 
210 static struct ast_entry ast_entries[TDA_MAX] __read_mostly = {
211 	[TDA_AST] = { .ae_f = ast_prep, .ae_flags = ASTR_UNCOND},
212 };
213 
214 void
215 ast_register(int ast, int flags, int tdp,
216     void (*f)(struct thread *, int asts))
217 {
218 	struct ast_entry *ae;
219 
220 	MPASS(ast < TDA_MAX);
221 	MPASS((flags & ASTR_TDP) == 0 || ((flags & ASTR_ASTF_REQUIRED) != 0
222 	    && __bitcount(tdp) == 1));
223 	ae = &ast_entries[ast];
224 	MPASS(ae->ae_f == NULL);
225 	ae->ae_flags = flags;
226 	ae->ae_tdp = tdp;
227 	atomic_interrupt_fence();
228 	ae->ae_f = f;
229 }
230 
231 /*
232  * XXXKIB Note that the deregistration of an AST handler does not
233  * drain threads possibly executing it, which affects unloadable
234  * modules.  The issue is either handled by the subsystem using
235  * handlers, or simply ignored.  Fixing the problem is considered not
236  * worth the overhead.
237  */
238 void
239 ast_deregister(int ast)
240 {
241 	struct ast_entry *ae;
242 
243 	MPASS(ast < TDA_MAX);
244 	ae = &ast_entries[ast];
245 	MPASS(ae->ae_f != NULL);
246 	ae->ae_f = NULL;
247 	atomic_interrupt_fence();
248 	ae->ae_flags = 0;
249 	ae->ae_tdp = 0;
250 }
251 
252 void
253 ast_sched_locked(struct thread *td, int tda)
254 {
255 	THREAD_LOCK_ASSERT(td, MA_OWNED);
256 	MPASS(tda < TDA_MAX);
257 
258 	td->td_ast |= TDAI(tda);
259 }
260 
261 void
262 ast_unsched_locked(struct thread *td, int tda)
263 {
264 	THREAD_LOCK_ASSERT(td, MA_OWNED);
265 	MPASS(tda < TDA_MAX);
266 
267 	td->td_ast &= ~TDAI(tda);
268 }
269 
270 void
271 ast_sched(struct thread *td, int tda)
272 {
273 	thread_lock(td);
274 	ast_sched_locked(td, tda);
275 	thread_unlock(td);
276 }
277 
278 void
279 ast_sched_mask(struct thread *td, int ast)
280 {
281 	thread_lock(td);
282 	td->td_ast |= ast;
283 	thread_unlock(td);
284 }
285 
286 static bool
287 ast_handler_calc_tdp_run(struct thread *td, const struct ast_entry *ae)
288 {
289 	return ((ae->ae_flags & ASTR_TDP) == 0 ||
290 	    (td->td_pflags & ae->ae_tdp) != 0);
291 }
292 
293 /*
294  * Process an asynchronous software trap.
295  */
296 static void
297 ast_handler(struct thread *td, struct trapframe *framep, bool dtor)
298 {
299 	struct ast_entry *ae;
300 	void (*f)(struct thread *td, int asts);
301 	int a, td_ast;
302 	bool run;
303 
304 	if (framep != NULL) {
305 		kmsan_mark(framep, sizeof(*framep), KMSAN_STATE_INITED);
306 		td->td_frame = framep;
307 	}
308 
309 	if (__predict_true(!dtor)) {
310 		WITNESS_WARN(WARN_PANIC, NULL, "Returning to user mode");
311 		mtx_assert(&Giant, MA_NOTOWNED);
312 		THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
313 
314 		/*
315 		 * This updates the td_ast for the checks below in one
316 		 * atomic operation with turning off all scheduled AST's.
317 		 * If another AST is triggered while we are handling the
318 		 * AST's saved in td_ast, the td_ast is again non-zero and
319 		 * ast() will be called again.
320 		 */
321 		thread_lock(td);
322 		td_ast = td->td_ast;
323 		td->td_ast = 0;
324 		thread_unlock(td);
325 	} else {
326 		/*
327 		 * The td thread's td_lock is not guaranteed to exist,
328 		 * the thread might be not initialized enough when it's
329 		 * destructor is called.  It is safe to read and
330 		 * update td_ast without locking since the thread is
331 		 * not runnable or visible to other threads.
332 		 */
333 		td_ast = td->td_ast;
334 		td->td_ast = 0;
335 	}
336 
337 	CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td, td->td_proc->p_pid,
338             td->td_proc->p_comm);
339 	KASSERT(framep == NULL || TRAPF_USERMODE(framep),
340 	    ("ast in kernel mode"));
341 
342 	for (a = 0; a < nitems(ast_entries); a++) {
343 		ae = &ast_entries[a];
344 		f = ae->ae_f;
345 		if (f == NULL)
346 			continue;
347 		atomic_interrupt_fence();
348 
349 		run = false;
350 		if (__predict_false(framep == NULL)) {
351 			if ((ae->ae_flags & ASTR_KCLEAR) != 0)
352 				run = ast_handler_calc_tdp_run(td, ae);
353 		} else {
354 			if ((ae->ae_flags & ASTR_UNCOND) != 0)
355 				run = true;
356 			else if ((ae->ae_flags & ASTR_ASTF_REQUIRED) != 0 &&
357 			    (td_ast & TDAI(a)) != 0)
358 				run = ast_handler_calc_tdp_run(td, ae);
359 		}
360 		if (run)
361 			f(td, td_ast);
362 	}
363 }
364 
365 void
366 ast(struct trapframe *framep)
367 {
368 	struct thread *td;
369 
370 	td = curthread;
371 	ast_handler(td, framep, false);
372 	userret(td, framep);
373 }
374 
375 void
376 ast_kclear(struct thread *td)
377 {
378 	ast_handler(td, NULL, td != curthread);
379 }
380 
381 const char *
382 syscallname(struct proc *p, u_int code)
383 {
384 	static const char unknown[] = "unknown";
385 	struct sysentvec *sv;
386 
387 	sv = p->p_sysent;
388 	if (sv->sv_syscallnames == NULL || code >= sv->sv_size)
389 		return (unknown);
390 	return (sv->sv_syscallnames[code]);
391 }
392