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
userret(struct thread * td,struct trapframe * frame)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 /*
129 * Let the scheduler adjust our priority etc.
130 */
131 sched_userret(td);
132
133 /*
134 * Check for misbehavior.
135 *
136 * In case there is a callchain tracing ongoing because of
137 * hwpmc(4), skip the scheduler pinning check.
138 * hwpmc(4) subsystem, infact, will collect callchain informations
139 * at ast() checkpoint, which is past userret().
140 */
141 WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
142 KASSERT(td->td_critnest == 0,
143 ("userret: Returning in a critical section"));
144 KASSERT(td->td_locks == 0,
145 ("userret: Returning with %d locks held", td->td_locks));
146 KASSERT(td->td_rw_rlocks == 0,
147 ("userret: Returning with %d rwlocks held in read mode",
148 td->td_rw_rlocks));
149 KASSERT(td->td_sx_slocks == 0,
150 ("userret: Returning with %d sx locks held in shared mode",
151 td->td_sx_slocks));
152 KASSERT(td->td_lk_slocks == 0,
153 ("userret: Returning with %d lockmanager locks held in shared mode",
154 td->td_lk_slocks));
155 KASSERT((td->td_pflags & TDP_NOFAULTING) == 0,
156 ("userret: Returning with pagefaults disabled"));
157 if (__predict_false(!THREAD_CAN_SLEEP())) {
158 #ifdef EPOCH_TRACE
159 epoch_trace_list(curthread);
160 #endif
161 KASSERT(0, ("userret: Returning with sleep disabled"));
162 }
163 KASSERT(td->td_pinned == 0 || (td->td_pflags & TDP_CALLCHAIN) != 0,
164 ("userret: Returning with pinned thread"));
165 KASSERT(td->td_vp_reserved == NULL,
166 ("userret: Returning with preallocated vnode"));
167 KASSERT((td->td_flags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
168 ("userret: Returning with stop signals deferred"));
169 KASSERT(td->td_vslock_sz == 0,
170 ("userret: Returning with vslock-wired space"));
171 #ifdef VIMAGE
172 /* Unfortunately td_vnet_lpush needs VNET_DEBUG. */
173 VNET_ASSERT(curvnet == NULL,
174 ("%s: Returning on td %p (pid %d, %s) with vnet %p set in %s",
175 __func__, td, p->p_pid, td->td_name, curvnet,
176 (td->td_vnet_lpush != NULL) ? td->td_vnet_lpush : "N/A"));
177 #endif
178 }
179
180 static void
ast_prep(struct thread * td,int tda __unused)181 ast_prep(struct thread *td, int tda __unused)
182 {
183 VM_CNT_INC(v_trap);
184 td->td_pticks = 0;
185 if (td->td_cowgen != atomic_load_int(&td->td_proc->p_cowgen))
186 thread_cow_update(td);
187
188 }
189
190 struct ast_entry {
191 int ae_flags;
192 int ae_tdp;
193 void (*ae_f)(struct thread *td, int ast);
194 };
195
196 _Static_assert(TDAI(TDA_MAX) <= UINT_MAX, "Too many ASTs");
197
198 static struct ast_entry ast_entries[TDA_MAX] __read_mostly = {
199 [TDA_AST] = { .ae_f = ast_prep, .ae_flags = ASTR_UNCOND},
200 };
201
202 void
ast_register(int ast,int flags,int tdp,void (* f)(struct thread *,int asts))203 ast_register(int ast, int flags, int tdp,
204 void (*f)(struct thread *, int asts))
205 {
206 struct ast_entry *ae;
207
208 MPASS(ast < TDA_MAX);
209 MPASS((flags & ASTR_TDP) == 0 || ((flags & ASTR_ASTF_REQUIRED) != 0
210 && __bitcount(tdp) == 1));
211 ae = &ast_entries[ast];
212 MPASS(ae->ae_f == NULL);
213 ae->ae_flags = flags;
214 ae->ae_tdp = tdp;
215 atomic_interrupt_fence();
216 ae->ae_f = f;
217 }
218
219 /*
220 * XXXKIB Note that the deregistration of an AST handler does not
221 * drain threads possibly executing it, which affects unloadable
222 * modules. The issue is either handled by the subsystem using
223 * handlers, or simply ignored. Fixing the problem is considered not
224 * worth the overhead.
225 */
226 void
ast_deregister(int ast)227 ast_deregister(int ast)
228 {
229 struct ast_entry *ae;
230
231 MPASS(ast < TDA_MAX);
232 ae = &ast_entries[ast];
233 MPASS(ae->ae_f != NULL);
234 ae->ae_f = NULL;
235 atomic_interrupt_fence();
236 ae->ae_flags = 0;
237 ae->ae_tdp = 0;
238 }
239
240 void
ast_sched_locked(struct thread * td,int tda)241 ast_sched_locked(struct thread *td, int tda)
242 {
243 THREAD_LOCK_ASSERT(td, MA_OWNED);
244 MPASS(tda < TDA_MAX);
245
246 td->td_ast |= TDAI(tda);
247 }
248
249 void
ast_unsched_locked(struct thread * td,int tda)250 ast_unsched_locked(struct thread *td, int tda)
251 {
252 THREAD_LOCK_ASSERT(td, MA_OWNED);
253 MPASS(tda < TDA_MAX);
254
255 td->td_ast &= ~TDAI(tda);
256 }
257
258 void
ast_sched(struct thread * td,int tda)259 ast_sched(struct thread *td, int tda)
260 {
261 thread_lock(td);
262 ast_sched_locked(td, tda);
263 thread_unlock(td);
264 }
265
266 void
ast_sched_mask(struct thread * td,int ast)267 ast_sched_mask(struct thread *td, int ast)
268 {
269 thread_lock(td);
270 td->td_ast |= ast;
271 thread_unlock(td);
272 }
273
274 static bool
ast_handler_calc_tdp_run(struct thread * td,const struct ast_entry * ae)275 ast_handler_calc_tdp_run(struct thread *td, const struct ast_entry *ae)
276 {
277 return ((ae->ae_flags & ASTR_TDP) == 0 ||
278 (td->td_pflags & ae->ae_tdp) != 0);
279 }
280
281 /*
282 * Process an asynchronous software trap.
283 */
284 static void
ast_handler(struct thread * td,struct trapframe * framep,bool dtor)285 ast_handler(struct thread *td, struct trapframe *framep, bool dtor)
286 {
287 struct ast_entry *ae;
288 void (*f)(struct thread *td, int asts);
289 int a, td_ast;
290 bool run;
291
292 if (framep != NULL) {
293 kmsan_mark(framep, sizeof(*framep), KMSAN_STATE_INITED);
294 td->td_frame = framep;
295 }
296
297 if (__predict_true(!dtor)) {
298 WITNESS_WARN(WARN_PANIC, NULL, "Returning to user mode");
299 mtx_assert(&Giant, MA_NOTOWNED);
300 THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
301
302 /*
303 * This updates the td_ast for the checks below in one
304 * atomic operation with turning off all scheduled AST's.
305 * If another AST is triggered while we are handling the
306 * AST's saved in td_ast, the td_ast is again non-zero and
307 * ast() will be called again.
308 */
309 thread_lock(td);
310 td_ast = td->td_ast;
311 td->td_ast = 0;
312 thread_unlock(td);
313 } else {
314 /*
315 * The td thread's td_lock is not guaranteed to exist,
316 * the thread might be not initialized enough when it's
317 * destructor is called. It is safe to read and
318 * update td_ast without locking since the thread is
319 * not runnable or visible to other threads.
320 */
321 td_ast = td->td_ast;
322 td->td_ast = 0;
323 }
324
325 CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td,
326 td->td_proc == NULL ? -1 : td->td_proc->p_pid,
327 td->td_proc == NULL ? "" : td->td_proc->p_comm);
328 KASSERT(framep == NULL || TRAPF_USERMODE(framep),
329 ("ast in kernel mode"));
330
331 for (a = 0; a < nitems(ast_entries); a++) {
332 ae = &ast_entries[a];
333 f = ae->ae_f;
334 if (f == NULL)
335 continue;
336 atomic_interrupt_fence();
337
338 run = false;
339 if (__predict_false(framep == NULL)) {
340 if ((ae->ae_flags & ASTR_KCLEAR) != 0)
341 run = ast_handler_calc_tdp_run(td, ae);
342 } else {
343 if ((ae->ae_flags & ASTR_UNCOND) != 0)
344 run = true;
345 else if ((ae->ae_flags & ASTR_ASTF_REQUIRED) != 0 &&
346 (td_ast & TDAI(a)) != 0)
347 run = ast_handler_calc_tdp_run(td, ae);
348 }
349 if (run)
350 f(td, td_ast);
351 }
352 }
353
354 void
ast(struct trapframe * framep)355 ast(struct trapframe *framep)
356 {
357 struct thread *td;
358
359 td = curthread;
360 ast_handler(td, framep, false);
361 userret(td, framep);
362 }
363
364 void
ast_kclear(struct thread * td)365 ast_kclear(struct thread *td)
366 {
367 ast_handler(td, NULL, td != curthread);
368 }
369
370 const char *
syscallname(struct proc * p,u_int code)371 syscallname(struct proc *p, u_int code)
372 {
373 static const char unknown[] = "unknown";
374 struct sysentvec *sv;
375
376 sv = p->p_sysent;
377 if (sv->sv_syscallnames == NULL || code >= sv->sv_size)
378 return (unknown);
379 return (sv->sv_syscallnames[code]);
380 }
381