xref: /netbsd/sys/kern/kern_syscall.c (revision 00e1384f)
1 /*	$NetBSD: kern_syscall.c,v 1.21 2020/08/31 19:51:30 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software developed for The NetBSD Foundation
8  * by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: kern_syscall.c,v 1.21 2020/08/31 19:51:30 christos Exp $");
34 
35 #ifdef _KERNEL_OPT
36 #include "opt_modular.h"
37 #include "opt_syscall_debug.h"
38 #include "opt_ktrace.h"
39 #include "opt_ptrace.h"
40 #include "opt_dtrace.h"
41 #endif
42 
43 /* XXX To get syscall prototypes. */
44 #define SYSVSHM
45 #define SYSVSEM
46 #define SYSVMSG
47 
48 #include <sys/param.h>
49 #include <sys/module.h>
50 #include <sys/sched.h>
51 #include <sys/syscall.h>
52 #include <sys/syscallargs.h>
53 #include <sys/syscallvar.h>
54 #include <sys/systm.h>
55 #include <sys/xcall.h>
56 #include <sys/ktrace.h>
57 #include <sys/ptrace.h>
58 
59 int
sys_nomodule(struct lwp * l,const void * v,register_t * retval)60 sys_nomodule(struct lwp *l, const void *v, register_t *retval)
61 {
62 #ifdef MODULAR
63 
64 	const struct sysent *sy;
65 	const struct emul *em;
66 	const struct sc_autoload *auto_list;
67 	u_int code;
68 
69 	/*
70 	 * Restart the syscall if we interrupted a module unload that
71 	 * failed.  Acquiring kernconfig_lock delays us until any unload
72 	 * has been completed or rolled back.
73 	 */
74 	kernconfig_lock();
75 	sy = l->l_sysent;
76 	if (sy->sy_call != sys_nomodule) {
77 		kernconfig_unlock();
78 		return ERESTART;
79 	}
80 	/*
81 	 * Try to autoload a module to satisfy the request.  If it
82 	 * works, retry the request.
83 	 */
84 	em = l->l_proc->p_emul;
85 	code = sy - em->e_sysent;
86 
87 	if ((auto_list = em->e_sc_autoload) != NULL)
88 		for (; auto_list->al_code > 0; auto_list++) {
89 			if (auto_list->al_code != code) {
90 				continue;
91 			}
92 			if (module_autoload(auto_list->al_module,
93 			    MODULE_CLASS_ANY) != 0 ||
94 			    sy->sy_call == sys_nomodule) {
95 			    	break;
96 			}
97 			kernconfig_unlock();
98 			return ERESTART;
99 		}
100 	kernconfig_unlock();
101 #endif	/* MODULAR */
102 
103 	return sys_nosys(l, v, retval);
104 }
105 
106 int
syscall_establish(const struct emul * em,const struct syscall_package * sp)107 syscall_establish(const struct emul *em, const struct syscall_package *sp)
108 {
109 	struct sysent *sy;
110 	int i;
111 
112 	KASSERT(kernconfig_is_held());
113 
114 	if (em == NULL) {
115 		em = &emul_netbsd;
116 	}
117 	sy = em->e_sysent;
118 
119 	/*
120 	 * Ensure that all preconditions are valid, since this is
121 	 * an all or nothing deal.  Once a system call is entered,
122 	 * it can become busy and we could be unable to remove it
123 	 * on error.
124 	 */
125 	for (i = 0; sp[i].sp_call != NULL; i++) {
126 		if (sp[i].sp_code >= SYS_NSYSENT)
127 			return EINVAL;
128 		if (sy[sp[i].sp_code].sy_call != sys_nomodule &&
129 		    sy[sp[i].sp_code].sy_call != sys_nosys) {
130 #ifdef DIAGNOSTIC
131 			printf("syscall %d is busy\n", sp[i].sp_code);
132 #endif
133 			return EBUSY;
134 		}
135 	}
136 	/* Everything looks good, patch them in. */
137 	for (i = 0; sp[i].sp_call != NULL; i++) {
138 		sy[sp[i].sp_code].sy_call = sp[i].sp_call;
139 	}
140 
141 	return 0;
142 }
143 
144 int
syscall_disestablish(const struct emul * em,const struct syscall_package * sp)145 syscall_disestablish(const struct emul *em, const struct syscall_package *sp)
146 {
147 	struct sysent *sy;
148 	const uint32_t *sb;
149 	lwp_t *l;
150 	int i;
151 
152 	KASSERT(kernconfig_is_held());
153 
154 	if (em == NULL) {
155 		em = &emul_netbsd;
156 	}
157 	sy = em->e_sysent;
158 	sb = em->e_nomodbits;
159 
160 	/*
161 	 * First, patch the system calls to sys_nomodule or sys_nosys
162 	 * to gate further activity.
163 	 */
164 	for (i = 0; sp[i].sp_call != NULL; i++) {
165 		KASSERT(sy[sp[i].sp_code].sy_call == sp[i].sp_call);
166 		sy[sp[i].sp_code].sy_call =
167 		    sb[sp[i].sp_code / 32] & (1 << (sp[i].sp_code % 32)) ?
168 		      sys_nomodule : sys_nosys;
169 	}
170 
171 	/*
172 	 * Run a cross call to cycle through all CPUs.  This does two
173 	 * things: lock activity provides a barrier and makes our update
174 	 * of sy_call visible to all CPUs, and upon return we can be sure
175 	 * that we see pertinent values of l_sysent posted by remote CPUs.
176 	 */
177 	xc_barrier(0);
178 
179 	/*
180 	 * Now it's safe to check l_sysent.  Run through all LWPs and see
181 	 * if anyone is still using the system call.
182 	 */
183 	for (i = 0; sp[i].sp_call != NULL; i++) {
184 		mutex_enter(&proc_lock);
185 		LIST_FOREACH(l, &alllwp, l_list) {
186 			if (l->l_sysent == &sy[sp[i].sp_code]) {
187 				break;
188 			}
189 		}
190 		mutex_exit(&proc_lock);
191 		if (l == NULL) {
192 			continue;
193 		}
194 		/*
195 		 * We lose: one or more calls are still in use.  Put back
196 		 * the old entrypoints and act like nothing happened.
197 		 * When we drop kernconfig_lock, any system calls held in
198 		 * sys_nomodule() will be restarted.
199 		 */
200 		for (i = 0; sp[i].sp_call != NULL; i++) {
201 			sy[sp[i].sp_code].sy_call = sp[i].sp_call;
202 		}
203 		return EBUSY;
204 	}
205 
206 	return 0;
207 }
208 
209 /*
210  * Return true if system call tracing is enabled for the specified process.
211  */
212 bool
trace_is_enabled(struct proc * p)213 trace_is_enabled(struct proc *p)
214 {
215 #ifdef SYSCALL_DEBUG
216 	return (true);
217 #endif
218 #ifdef KTRACE
219 	if (ISSET(p->p_traceflag, (KTRFAC_SYSCALL | KTRFAC_SYSRET)))
220 		return (true);
221 #endif
222 #ifdef PTRACE
223 	if (ISSET(p->p_slflag, PSL_SYSCALL))
224 		return (true);
225 #endif
226 
227 	return (false);
228 }
229 
230 /*
231  * Start trace of particular system call. If process is being traced,
232  * this routine is called by MD syscall dispatch code just before
233  * a system call is actually executed.
234  */
235 int
trace_enter(register_t code,const struct sysent * sy,const void * args)236 trace_enter(register_t code, const struct sysent *sy, const void *args)
237 {
238 	int error = 0;
239 #if defined(PTRACE) || defined(KDTRACE_HOOKS)
240 	struct proc *p = curlwp->l_proc;
241 #endif
242 
243 #ifdef KDTRACE_HOOKS
244 	if (sy->sy_entry) {
245 		struct emul *e = p->p_emul;
246 		if (e->e_dtrace_syscall)
247 			(*e->e_dtrace_syscall)(sy->sy_entry, code, sy, args,
248 			    NULL, 0);
249 	}
250 #endif
251 
252 #ifdef SYSCALL_DEBUG
253 	scdebug_call(code, args);
254 #endif /* SYSCALL_DEBUG */
255 
256 	ktrsyscall(code, args, sy->sy_narg);
257 
258 #ifdef PTRACE
259 	if ((p->p_slflag & (PSL_SYSCALL|PSL_TRACED)) ==
260 	    (PSL_SYSCALL|PSL_TRACED)) {
261 		proc_stoptrace(TRAP_SCE, code, args, NULL, 0);
262 		if (curlwp->l_proc->p_slflag & PSL_SYSCALLEMU) {
263 			/* tracer will emulate syscall for us */
264 			error = EJUSTRETURN;
265 		}
266 	}
267 #endif
268 	return error;
269 }
270 
271 /*
272  * End trace of particular system call. If process is being traced,
273  * this routine is called by MD syscall dispatch code just after
274  * a system call finishes.
275  * MD caller guarantees the passed 'code' is within the supported
276  * system call number range for emulation the process runs under.
277  */
278 void
trace_exit(register_t code,const struct sysent * sy,const void * args,register_t rval[],int error)279 trace_exit(register_t code, const struct sysent *sy, const void *args,
280     register_t rval[], int error)
281 {
282 #if defined(PTRACE) || defined(KDTRACE_HOOKS)
283 	struct proc *p = curlwp->l_proc;
284 #endif
285 
286 #ifdef KDTRACE_HOOKS
287 	if (sy->sy_return) {
288 		struct emul *e = p->p_emul;
289 		if (e->e_dtrace_syscall)
290 			(*p->p_emul->e_dtrace_syscall)(sy->sy_return, code, sy,
291 			    args, rval, error);
292 	}
293 #endif
294 
295 #ifdef SYSCALL_DEBUG
296 	scdebug_ret(code, error, rval);
297 #endif /* SYSCALL_DEBUG */
298 
299 	ktrsysret(code, error, rval);
300 
301 #ifdef PTRACE
302 	if ((p->p_slflag & (PSL_SYSCALL|PSL_TRACED|PSL_SYSCALLEMU)) ==
303 	    (PSL_SYSCALL|PSL_TRACED)) {
304 		proc_stoptrace(TRAP_SCX, code, args, rval, error);
305 	}
306 	CLR(p->p_slflag, PSL_SYSCALLEMU);
307 #endif
308 }
309