xref: /netbsd-src/sys/kern/kern_syscall.c (revision aceb213538ec08a74028e213127af18aa17bf1cf)
1 /*	$NetBSD: kern_syscall.c,v 1.8 2013/12/09 16:49:43 pooka 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 1.8 2013/12/09 16:49:43 pooka 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 #endif
41 
42 /* XXX To get syscall prototypes. */
43 #define SYSVSHM
44 #define SYSVSEM
45 #define SYSVMSG
46 
47 #include <sys/param.h>
48 #include <sys/module.h>
49 #include <sys/sched.h>
50 #include <sys/syscall.h>
51 #include <sys/syscallargs.h>
52 #include <sys/syscallvar.h>
53 #include <sys/systm.h>
54 #include <sys/xcall.h>
55 #include <sys/ktrace.h>
56 #include <sys/ptrace.h>
57 
58 int
59 sys_nomodule(struct lwp *l, const void *v, register_t *retval)
60 {
61 #ifdef MODULAR
62 	static struct {
63 		u_int		al_code;
64 		const char	*al_module;
65 	} const autoload[] = {
66 	    { SYS_aio_cancel, "aio" },
67 	    { SYS_aio_error, "aio" },
68 	    { SYS_aio_fsync, "aio" },
69 	    { SYS_aio_read, "aio" },
70 	    { SYS_aio_return, "aio" },
71 	    { SYS___aio_suspend50, "aio" },
72 	    { SYS_aio_write, "aio" },
73 	    { SYS_lio_listio, "aio" },
74 	    { SYS_mq_open, "mqueue" },
75 	    { SYS_mq_close, "mqueue" },
76 	    { SYS_mq_unlink, "mqueue" },
77 	    { SYS_mq_getattr, "mqueue" },
78 	    { SYS_mq_setattr, "mqueue" },
79 	    { SYS_mq_notify, "mqueue" },
80 	    { SYS_mq_send, "mqueue" },
81 	    { SYS_mq_receive, "mqueue" },
82 	    { SYS___mq_timedsend50, "mqueue" },
83 	    { SYS___mq_timedreceive50, "mqueue" },
84 	    { SYS_compat_43_fstat43, "compat" },
85 	    { SYS_compat_43_lstat43, "compat" },
86 	    { SYS_compat_43_oaccept, "compat" },
87 	    { SYS_compat_43_ocreat, "compat" },
88 	    { SYS_compat_43_oftruncate, "compat" },
89 	    { SYS_compat_43_ogetdirentries, "compat" },
90 	    { SYS_compat_43_ogetdtablesize, "compat" },
91 	    { SYS_compat_43_ogethostid, "compat" },
92 	    { SYS_compat_43_ogethostname, "compat" },
93 	    { SYS_compat_43_ogetkerninfo, "compat" },
94 	    { SYS_compat_43_ogetpagesize, "compat" },
95 	    { SYS_compat_43_ogetpeername, "compat" },
96 	    { SYS_compat_43_ogetrlimit, "compat" },
97 	    { SYS_compat_43_ogetsockname, "compat" },
98 	    { SYS_compat_43_okillpg, "compat" },
99 	    { SYS_compat_43_olseek, "compat" },
100 	    { SYS_compat_43_ommap, "compat" },
101 	    { SYS_compat_43_oquota, "compat" },
102 	    { SYS_compat_43_orecv, "compat" },
103 	    { SYS_compat_43_orecvfrom, "compat" },
104 	    { SYS_compat_43_orecvmsg, "compat" },
105 	    { SYS_compat_43_osend, "compat" },
106 	    { SYS_compat_43_osendmsg, "compat" },
107 	    { SYS_compat_43_osethostid, "compat" },
108 	    { SYS_compat_43_osethostname, "compat" },
109 	    { SYS_compat_43_osetrlimit, "compat" },
110 	    { SYS_compat_43_osigblock, "compat" },
111 	    { SYS_compat_43_osigsetmask, "compat" },
112 	    { SYS_compat_43_osigstack, "compat" },
113 	    { SYS_compat_43_osigvec, "compat" },
114 	    { SYS_compat_43_otruncate, "compat" },
115 	    { SYS_compat_43_owait, "compat" },
116 	    { SYS_compat_43_stat43, "compat" },
117 	    { SYS_compat_09_ogetdomainname, "compat" },
118 	    { SYS_compat_09_osetdomainname, "compat" },
119 	    { SYS_compat_09_ouname, "compat" },
120 #ifndef _LP64
121 	    { SYS_compat_10_omsgsys, "compat" },
122 	    { SYS_compat_10_osemsys, "compat" },
123 	    { SYS_compat_10_oshmsys, "compat" },
124 #endif
125 	    { SYS_compat_12_fstat12, "compat" },
126 	    { SYS_compat_12_getdirentries, "compat" },
127 	    { SYS_compat_12_lstat12, "compat" },
128 	    { SYS_compat_12_msync, "compat" },
129 	    { SYS_compat_12_oreboot, "compat" },
130 	    { SYS_compat_12_oswapon, "compat" },
131 	    { SYS_compat_12_stat12, "compat" },
132 	    { SYS_compat_13_sigaction13, "compat" },
133 	    { SYS_compat_13_sigaltstack13, "compat" },
134 	    { SYS_compat_13_sigpending13, "compat" },
135 	    { SYS_compat_13_sigprocmask13, "compat" },
136 	    { SYS_compat_13_sigreturn13, "compat" },
137 	    { SYS_compat_13_sigsuspend13, "compat" },
138 	    { SYS_compat_14___semctl, "compat" },
139 	    { SYS_compat_14_msgctl, "compat" },
140 	    { SYS_compat_14_shmctl, "compat" },
141 	    { SYS_compat_16___sigaction14, "compat" },
142 	    { SYS_compat_16___sigreturn14, "compat" },
143 	    { SYS_compat_20_fhstatfs, "compat" },
144 	    { SYS_compat_20_fstatfs, "compat" },
145 	    { SYS_compat_20_getfsstat, "compat" },
146 	    { SYS_compat_20_statfs, "compat" },
147 	    { SYS_compat_30___fhstat30, "compat" },
148 	    { SYS_compat_30___fstat13, "compat" },
149 	    { SYS_compat_30___lstat13, "compat" },
150 	    { SYS_compat_30___stat13, "compat" },
151 	    { SYS_compat_30_fhopen, "compat" },
152 	    { SYS_compat_30_fhstat, "compat" },
153 	    { SYS_compat_30_fhstatvfs1, "compat" },
154 	    { SYS_compat_30_getdents, "compat" },
155 	    { SYS_compat_30_getfh, "compat" },
156 	    { SYS_compat_30_socket, "compat" },
157 	    { SYS_compat_40_mount, "compat" },
158 	    { SYS_compat_50_wait4, "compat" },
159 	    { SYS_compat_50_mknod, "compat" },
160 	    { SYS_compat_50_setitimer, "compat" },
161 	    { SYS_compat_50_getitimer, "compat" },
162 	    { SYS_compat_50_select, "compat" },
163 	    { SYS_compat_50_gettimeofday, "compat" },
164 	    { SYS_compat_50_getrusage, "compat" },
165 	    { SYS_compat_50_settimeofday, "compat" },
166 	    { SYS_compat_50_utimes, "compat" },
167 	    { SYS_compat_50_adjtime, "compat" },
168 	    { SYS_compat_50_lfs_segwait, "compat" },
169 	    { SYS_compat_50_futimes, "compat" },
170 	    { SYS_compat_50_clock_gettime, "compat" },
171 	    { SYS_compat_50_clock_settime, "compat" },
172 	    { SYS_compat_50_clock_getres, "compat" },
173 	    { SYS_compat_50_timer_settime, "compat" },
174 	    { SYS_compat_50_timer_gettime, "compat" },
175 	    { SYS_compat_50_nanosleep, "compat" },
176 	    { SYS_compat_50___sigtimedwait, "compat" },
177 	    { SYS_compat_50_mq_timedsend, "compat" },
178 	    { SYS_compat_50_mq_timedreceive, "compat" },
179 	    { SYS_compat_50_lutimes, "compat" },
180 	    { SYS_compat_50_____semctl13, "compat" },
181 	    { SYS_compat_50___msgctl13, "compat" },
182 	    { SYS_compat_50___shmctl13, "compat" },
183 	    { SYS_compat_50__lwp_park, "compat" },
184 	    { SYS_compat_50_kevent, "compat" },
185 	    { SYS_compat_50_pselect, "compat" },
186 	    { SYS_compat_50_pollts, "compat" },
187 	    { SYS_compat_50___stat30, "compat" },
188 	    { SYS_compat_50___fstat30, "compat" },
189 	    { SYS_compat_50___lstat30, "compat" },
190 	    { SYS_compat_50___ntp_gettime30, "compat" },
191 	    { SYS_compat_50___fhstat40, "compat" },
192 	    { SYS_compat_50_aio_suspend, "compat" },
193 	    { SYS__ksem_init, "ksem" },
194 	    { SYS__ksem_open, "ksem" },
195 	    { SYS__ksem_unlink, "ksem" },
196 	    { SYS__ksem_close, "ksem" },
197 	    { SYS__ksem_post, "ksem" },
198 	    { SYS__ksem_wait, "ksem" },
199 	    { SYS__ksem_trywait, "ksem" },
200 	    { SYS__ksem_getvalue, "ksem" },
201 	    { SYS__ksem_destroy, "ksem" },
202 	    { SYS__ksem_timedwait, "ksem" },
203 	    { SYS_nfssvc, "nfsserver" },
204 	    { SYS_afssys, "openafs" },
205 	};
206 	const struct sysent *sy;
207 	const struct emul *em;
208 	int code, i;
209 
210 	/*
211 	 * Restart the syscall if we interrupted a module unload that
212 	 * failed.  Acquiring kernconfig_lock delays us until any unload
213 	 * has been completed or rolled back.
214 	 */
215 	kernconfig_lock();
216 	sy = l->l_sysent;
217 	if (sy->sy_call != sys_nomodule) {
218 		kernconfig_unlock();
219 		return ERESTART;
220 	}
221 	/*
222 	 * Try to autoload a module to satisfy the request.  If it
223 	 * works, retry the request.
224 	 */
225 	em = l->l_proc->p_emul;
226 	if (em == &emul_netbsd) {
227 		code = sy - em->e_sysent;
228 		for (i = 0; i < __arraycount(autoload); i++) {
229 			if (autoload[i].al_code != code) {
230 				continue;
231 			}
232 			if (module_autoload(autoload[i].al_module,
233 			    MODULE_CLASS_ANY) != 0 ||
234 			    sy->sy_call == sys_nomodule) {
235 			    	break;
236 			}
237 			kernconfig_unlock();
238 			return ERESTART;
239 		}
240 	}
241 	kernconfig_unlock();
242 #endif	/* MODULAR */
243 
244 	return sys_nosys(l, v, retval);
245 }
246 
247 int
248 syscall_establish(const struct emul *em, const struct syscall_package *sp)
249 {
250 	struct sysent *sy;
251 	int i;
252 
253 	KASSERT(kernconfig_is_held());
254 
255 	if (em == NULL) {
256 		em = &emul_netbsd;
257 	}
258 	sy = em->e_sysent;
259 
260 	/*
261 	 * Ensure that all preconditions are valid, since this is
262 	 * an all or nothing deal.  Once a system call is entered,
263 	 * it can become busy and we could be unable to remove it
264 	 * on error.
265 	 */
266 	for (i = 0; sp[i].sp_call != NULL; i++) {
267 		if (sy[sp[i].sp_code].sy_call != sys_nomodule) {
268 #ifdef DIAGNOSTIC
269 			printf("syscall %d is busy\n", sp[i].sp_code);
270 #endif
271 			return EBUSY;
272 		}
273 	}
274 	/* Everything looks good, patch them in. */
275 	for (i = 0; sp[i].sp_call != NULL; i++) {
276 		sy[sp[i].sp_code].sy_call = sp[i].sp_call;
277 	}
278 
279 	return 0;
280 }
281 
282 int
283 syscall_disestablish(const struct emul *em, const struct syscall_package *sp)
284 {
285 	struct sysent *sy;
286 	uint64_t where;
287 	lwp_t *l;
288 	int i;
289 
290 	KASSERT(kernconfig_is_held());
291 
292 	if (em == NULL) {
293 		em = &emul_netbsd;
294 	}
295 	sy = em->e_sysent;
296 
297 	/*
298 	 * First, patch the system calls to sys_nomodule to gate further
299 	 * activity.
300 	 */
301 	for (i = 0; sp[i].sp_call != NULL; i++) {
302 		KASSERT(sy[sp[i].sp_code].sy_call == sp[i].sp_call);
303 		sy[sp[i].sp_code].sy_call = sys_nomodule;
304 	}
305 
306 	/*
307 	 * Run a cross call to cycle through all CPUs.  This does two
308 	 * things: lock activity provides a barrier and makes our update
309 	 * of sy_call visible to all CPUs, and upon return we can be sure
310 	 * that we see pertinent values of l_sysent posted by remote CPUs.
311 	 */
312 	where = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
313 	xc_wait(where);
314 
315 	/*
316 	 * Now it's safe to check l_sysent.  Run through all LWPs and see
317 	 * if anyone is still using the system call.
318 	 */
319 	for (i = 0; sp[i].sp_call != NULL; i++) {
320 		mutex_enter(proc_lock);
321 		LIST_FOREACH(l, &alllwp, l_list) {
322 			if (l->l_sysent == &sy[sp[i].sp_code]) {
323 				break;
324 			}
325 		}
326 		mutex_exit(proc_lock);
327 		if (l == NULL) {
328 			continue;
329 		}
330 		/*
331 		 * We lose: one or more calls are still in use.  Put back
332 		 * the old entrypoints and act like nothing happened.
333 		 * When we drop kernconfig_lock, any system calls held in
334 		 * sys_nomodule() will be restarted.
335 		 */
336 		for (i = 0; sp[i].sp_call != NULL; i++) {
337 			sy[sp[i].sp_code].sy_call = sp[i].sp_call;
338 		}
339 		return EBUSY;
340 	}
341 
342 	return 0;
343 }
344 
345 /*
346  * Return true if system call tracing is enabled for the specified process.
347  */
348 bool
349 trace_is_enabled(struct proc *p)
350 {
351 #ifdef SYSCALL_DEBUG
352 	return (true);
353 #endif
354 #ifdef KTRACE
355 	if (ISSET(p->p_traceflag, (KTRFAC_SYSCALL | KTRFAC_SYSRET)))
356 		return (true);
357 #endif
358 #ifdef PTRACE
359 	if (ISSET(p->p_slflag, PSL_SYSCALL))
360 		return (true);
361 #endif
362 
363 	return (false);
364 }
365 
366 /*
367  * Start trace of particular system call. If process is being traced,
368  * this routine is called by MD syscall dispatch code just before
369  * a system call is actually executed.
370  */
371 int
372 trace_enter(register_t code, const register_t *args, int narg)
373 {
374 	int error = 0;
375 
376 #ifdef SYSCALL_DEBUG
377 	scdebug_call(code, args);
378 #endif /* SYSCALL_DEBUG */
379 
380 	ktrsyscall(code, args, narg);
381 
382 #ifdef PTRACE
383 	if ((curlwp->l_proc->p_slflag & (PSL_SYSCALL|PSL_TRACED)) ==
384 	    (PSL_SYSCALL|PSL_TRACED)) {
385 		process_stoptrace();
386 		if (curlwp->l_proc->p_slflag & PSL_SYSCALLEMU) {
387 			/* tracer will emulate syscall for us */
388 			error = EJUSTRETURN;
389 		}
390 	}
391 #endif
392 	return error;
393 }
394 
395 /*
396  * End trace of particular system call. If process is being traced,
397  * this routine is called by MD syscall dispatch code just after
398  * a system call finishes.
399  * MD caller guarantees the passed 'code' is within the supported
400  * system call number range for emulation the process runs under.
401  */
402 void
403 trace_exit(register_t code, register_t rval[], int error)
404 {
405 #ifdef PTRACE
406 	struct proc *p = curlwp->l_proc;
407 #endif
408 
409 #ifdef SYSCALL_DEBUG
410 	scdebug_ret(code, error, rval);
411 #endif /* SYSCALL_DEBUG */
412 
413 	ktrsysret(code, error, rval);
414 
415 #ifdef PTRACE
416 	if ((p->p_slflag & (PSL_SYSCALL|PSL_TRACED|PSL_SYSCALLEMU)) ==
417 	    (PSL_SYSCALL|PSL_TRACED))
418 		process_stoptrace();
419 	CLR(p->p_slflag, PSL_SYSCALLEMU);
420 #endif
421 }
422