xref: /freebsd/sys/kern/subr_trap.c (revision 1db64f89363c97858961c4df0b7d02f3223723cf)
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 
44 #include <sys/cdefs.h>
45 #include "opt_hwpmc_hooks.h"
46 
47 #include <sys/param.h>
48 #include <sys/kernel.h>
49 #include <sys/limits.h>
50 #include <sys/lock.h>
51 #include <sys/msan.h>
52 #include <sys/mutex.h>
53 #include <sys/proc.h>
54 #include <sys/ktr.h>
55 #include <sys/resourcevar.h>
56 #include <sys/sched.h>
57 #include <sys/syscall.h>
58 #include <sys/syscallsubr.h>
59 #include <sys/sysent.h>
60 #include <sys/systm.h>
61 #include <sys/vmmeter.h>
62 
63 #include <machine/cpu.h>
64 
65 #ifdef VIMAGE
66 #include <net/vnet.h>
67 #endif
68 
69 #ifdef	HWPMC_HOOKS
70 #include <sys/pmckern.h>
71 #endif
72 
73 #ifdef EPOCH_TRACE
74 #include <sys/epoch.h>
75 #endif
76 
77 /*
78  * Define the code needed before returning to user mode, for trap and
79  * syscall.
80  */
81 void
82 userret(struct thread *td, struct trapframe *frame)
83 {
84 	struct proc *p = td->td_proc;
85 
86 	CTR3(KTR_SYSC, "userret: thread %p (pid %d, %s)", td, p->p_pid,
87             td->td_name);
88 	KASSERT((p->p_flag & P_WEXIT) == 0,
89 	    ("Exiting process returns to usermode"));
90 #ifdef DIAGNOSTIC
91 	/*
92 	 * Check that we called signotify() enough.  For
93 	 * multi-threaded processes, where signal distribution might
94 	 * change due to other threads changing sigmask, the check is
95 	 * racy and cannot be performed reliably.
96 	 * If current process is vfork child, indicated by P_PPWAIT, then
97 	 * issignal() ignores stops, so we block the check to avoid
98 	 * classifying pending signals.
99 	 */
100 	if (p->p_numthreads == 1) {
101 		PROC_LOCK(p);
102 		thread_lock(td);
103 		if ((p->p_flag & P_PPWAIT) == 0 &&
104 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0 &&
105 		    SIGPENDING(td) && !td_ast_pending(td, TDA_AST) &&
106 		    !td_ast_pending(td, TDA_SIG)) {
107 			thread_unlock(td);
108 			panic(
109 			    "failed to set signal flags for ast p %p "
110 			    "td %p td_ast %#x fl %#x",
111 			    p, td, td->td_ast, td->td_flags);
112 		}
113 		thread_unlock(td);
114 		PROC_UNLOCK(p);
115 	}
116 #endif
117 
118 	/*
119 	 * Charge system time if profiling.
120 	 */
121 	if (__predict_false(p->p_flag & P_PROFIL))
122 		addupc_task(td, TRAPF_PC(frame), td->td_pticks * psratio);
123 
124 #ifdef HWPMC_HOOKS
125 	if (PMC_THREAD_HAS_SAMPLES(td))
126 		PMC_CALL_HOOK(td, PMC_FN_THR_USERRET, NULL);
127 #endif
128 #ifdef TCPHPTS
129 	/*
130 	 * @gallatin is adament that this needs to go here, I
131 	 * am not so sure. Running hpts is a lot like
132 	 * a lro_flush() that happens while a user process
133 	 * is running. But he may know best so I will go
134 	 * with his view of accounting. :-)
135 	 */
136 	tcp_run_hpts();
137 #endif
138 	/*
139 	 * Let the scheduler adjust our priority etc.
140 	 */
141 	sched_userret(td);
142 
143 	/*
144 	 * Check for misbehavior.
145 	 *
146 	 * In case there is a callchain tracing ongoing because of
147 	 * hwpmc(4), skip the scheduler pinning check.
148 	 * hwpmc(4) subsystem, infact, will collect callchain informations
149 	 * at ast() checkpoint, which is past userret().
150 	 */
151 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
152 	KASSERT(td->td_critnest == 0,
153 	    ("userret: Returning in a critical section"));
154 	KASSERT(td->td_locks == 0,
155 	    ("userret: Returning with %d locks held", td->td_locks));
156 	KASSERT(td->td_rw_rlocks == 0,
157 	    ("userret: Returning with %d rwlocks held in read mode",
158 	    td->td_rw_rlocks));
159 	KASSERT(td->td_sx_slocks == 0,
160 	    ("userret: Returning with %d sx locks held in shared mode",
161 	    td->td_sx_slocks));
162 	KASSERT(td->td_lk_slocks == 0,
163 	    ("userret: Returning with %d lockmanager locks held in shared mode",
164 	    td->td_lk_slocks));
165 	KASSERT((td->td_pflags & TDP_NOFAULTING) == 0,
166 	    ("userret: Returning with pagefaults disabled"));
167 	if (__predict_false(!THREAD_CAN_SLEEP())) {
168 #ifdef EPOCH_TRACE
169 		epoch_trace_list(curthread);
170 #endif
171 		KASSERT(0, ("userret: Returning with sleep disabled"));
172 	}
173 	KASSERT(td->td_pinned == 0 || (td->td_pflags & TDP_CALLCHAIN) != 0,
174 	    ("userret: Returning with pinned thread"));
175 	KASSERT(td->td_vp_reserved == NULL,
176 	    ("userret: Returning with preallocated vnode"));
177 	KASSERT((td->td_flags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
178 	    ("userret: Returning with stop signals deferred"));
179 	KASSERT(td->td_vslock_sz == 0,
180 	    ("userret: Returning with vslock-wired space"));
181 #ifdef VIMAGE
182 	/* Unfortunately td_vnet_lpush needs VNET_DEBUG. */
183 	VNET_ASSERT(curvnet == NULL,
184 	    ("%s: Returning on td %p (pid %d, %s) with vnet %p set in %s",
185 	    __func__, td, p->p_pid, td->td_name, curvnet,
186 	    (td->td_vnet_lpush != NULL) ? td->td_vnet_lpush : "N/A"));
187 #endif
188 }
189 
190 static void
191 ast_prep(struct thread *td, int tda __unused)
192 {
193 	VM_CNT_INC(v_trap);
194 	td->td_pticks = 0;
195 	if (td->td_cowgen != atomic_load_int(&td->td_proc->p_cowgen))
196 		thread_cow_update(td);
197 
198 }
199 
200 struct ast_entry {
201 	int	ae_flags;
202 	int	ae_tdp;
203 	void	(*ae_f)(struct thread *td, int ast);
204 };
205 
206 _Static_assert(TDAI(TDA_MAX) <= UINT_MAX, "Too many ASTs");
207 
208 static struct ast_entry ast_entries[TDA_MAX] __read_mostly = {
209 	[TDA_AST] = { .ae_f = ast_prep, .ae_flags = ASTR_UNCOND},
210 };
211 
212 void
213 ast_register(int ast, int flags, int tdp,
214     void (*f)(struct thread *, int asts))
215 {
216 	struct ast_entry *ae;
217 
218 	MPASS(ast < TDA_MAX);
219 	MPASS((flags & ASTR_TDP) == 0 || ((flags & ASTR_ASTF_REQUIRED) != 0
220 	    && __bitcount(tdp) == 1));
221 	ae = &ast_entries[ast];
222 	MPASS(ae->ae_f == NULL);
223 	ae->ae_flags = flags;
224 	ae->ae_tdp = tdp;
225 	atomic_interrupt_fence();
226 	ae->ae_f = f;
227 }
228 
229 /*
230  * XXXKIB Note that the deregistration of an AST handler does not
231  * drain threads possibly executing it, which affects unloadable
232  * modules.  The issue is either handled by the subsystem using
233  * handlers, or simply ignored.  Fixing the problem is considered not
234  * worth the overhead.
235  */
236 void
237 ast_deregister(int ast)
238 {
239 	struct ast_entry *ae;
240 
241 	MPASS(ast < TDA_MAX);
242 	ae = &ast_entries[ast];
243 	MPASS(ae->ae_f != NULL);
244 	ae->ae_f = NULL;
245 	atomic_interrupt_fence();
246 	ae->ae_flags = 0;
247 	ae->ae_tdp = 0;
248 }
249 
250 void
251 ast_sched_locked(struct thread *td, int tda)
252 {
253 	THREAD_LOCK_ASSERT(td, MA_OWNED);
254 	MPASS(tda < TDA_MAX);
255 
256 	td->td_ast |= TDAI(tda);
257 }
258 
259 void
260 ast_unsched_locked(struct thread *td, int tda)
261 {
262 	THREAD_LOCK_ASSERT(td, MA_OWNED);
263 	MPASS(tda < TDA_MAX);
264 
265 	td->td_ast &= ~TDAI(tda);
266 }
267 
268 void
269 ast_sched(struct thread *td, int tda)
270 {
271 	thread_lock(td);
272 	ast_sched_locked(td, tda);
273 	thread_unlock(td);
274 }
275 
276 void
277 ast_sched_mask(struct thread *td, int ast)
278 {
279 	thread_lock(td);
280 	td->td_ast |= ast;
281 	thread_unlock(td);
282 }
283 
284 static bool
285 ast_handler_calc_tdp_run(struct thread *td, const struct ast_entry *ae)
286 {
287 	return ((ae->ae_flags & ASTR_TDP) == 0 ||
288 	    (td->td_pflags & ae->ae_tdp) != 0);
289 }
290 
291 /*
292  * Process an asynchronous software trap.
293  */
294 static void
295 ast_handler(struct thread *td, struct trapframe *framep, bool dtor)
296 {
297 	struct ast_entry *ae;
298 	void (*f)(struct thread *td, int asts);
299 	int a, td_ast;
300 	bool run;
301 
302 	if (framep != NULL) {
303 		kmsan_mark(framep, sizeof(*framep), KMSAN_STATE_INITED);
304 		td->td_frame = framep;
305 	}
306 
307 	if (__predict_true(!dtor)) {
308 		WITNESS_WARN(WARN_PANIC, NULL, "Returning to user mode");
309 		mtx_assert(&Giant, MA_NOTOWNED);
310 		THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
311 
312 		/*
313 		 * This updates the td_ast for the checks below in one
314 		 * atomic operation with turning off all scheduled AST's.
315 		 * If another AST is triggered while we are handling the
316 		 * AST's saved in td_ast, the td_ast is again non-zero and
317 		 * ast() will be called again.
318 		 */
319 		thread_lock(td);
320 		td_ast = td->td_ast;
321 		td->td_ast = 0;
322 		thread_unlock(td);
323 	} else {
324 		/*
325 		 * The td thread's td_lock is not guaranteed to exist,
326 		 * the thread might be not initialized enough when it's
327 		 * destructor is called.  It is safe to read and
328 		 * update td_ast without locking since the thread is
329 		 * not runnable or visible to other threads.
330 		 */
331 		td_ast = td->td_ast;
332 		td->td_ast = 0;
333 	}
334 
335 	CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td, td->td_proc->p_pid,
336             td->td_proc->p_comm);
337 	KASSERT(framep == NULL || TRAPF_USERMODE(framep),
338 	    ("ast in kernel mode"));
339 
340 	for (a = 0; a < nitems(ast_entries); a++) {
341 		ae = &ast_entries[a];
342 		f = ae->ae_f;
343 		if (f == NULL)
344 			continue;
345 		atomic_interrupt_fence();
346 
347 		run = false;
348 		if (__predict_false(framep == NULL)) {
349 			if ((ae->ae_flags & ASTR_KCLEAR) != 0)
350 				run = ast_handler_calc_tdp_run(td, ae);
351 		} else {
352 			if ((ae->ae_flags & ASTR_UNCOND) != 0)
353 				run = true;
354 			else if ((ae->ae_flags & ASTR_ASTF_REQUIRED) != 0 &&
355 			    (td_ast & TDAI(a)) != 0)
356 				run = ast_handler_calc_tdp_run(td, ae);
357 		}
358 		if (run)
359 			f(td, td_ast);
360 	}
361 }
362 
363 void
364 ast(struct trapframe *framep)
365 {
366 	struct thread *td;
367 
368 	td = curthread;
369 	ast_handler(td, framep, false);
370 	userret(td, framep);
371 }
372 
373 void
374 ast_kclear(struct thread *td)
375 {
376 	ast_handler(td, NULL, td != curthread);
377 }
378 
379 const char *
380 syscallname(struct proc *p, u_int code)
381 {
382 	static const char unknown[] = "unknown";
383 	struct sysentvec *sv;
384 
385 	sv = p->p_sysent;
386 	if (sv->sv_syscallnames == NULL || code >= sv->sv_size)
387 		return (unknown);
388 	return (sv->sv_syscallnames[code]);
389 }
390