1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 *
21 * Portions Copyright 2006-2008 John Birrell jb@freebsd.org
22 *
23 * $FreeBSD$
24 *
25 */
26
27/*
28 * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
29 * Use is subject to license terms.
30 */
31
32#include <sys/cdefs.h>
33#include <sys/param.h>
34#include <sys/systm.h>
35#include <sys/conf.h>
36#include <sys/cpuvar.h>
37#include <sys/fcntl.h>
38#include <sys/filio.h>
39#include <sys/kdb.h>
40#include <sys/kernel.h>
41#include <sys/kmem.h>
42#include <sys/kthread.h>
43#include <sys/limits.h>
44#include <sys/linker.h>
45#include <sys/lock.h>
46#include <sys/malloc.h>
47#include <sys/module.h>
48#include <sys/mutex.h>
49#include <sys/poll.h>
50#include <sys/proc.h>
51#include <sys/selinfo.h>
52#include <sys/smp.h>
53#include <sys/sysproto.h>
54#include <sys/sysent.h>
55#include <sys/uio.h>
56#include <sys/unistd.h>
57#include <machine/stdarg.h>
58
59#include <sys/dtrace.h>
60
61#ifdef LINUX_SYSTRACE
62#if defined(__amd64__)
63#include <amd64/linux32/linux.h>
64#include <amd64/linux32/linux32_proto.h>
65#include <amd64/linux32/linux32_syscalls.c>
66#include <amd64/linux32/linux32_systrace_args.c>
67#define	MODNAME		"linux32"
68#elif defined(__i386__)
69#include <i386/linux/linux.h>
70#include <i386/linux/linux_proto.h>
71#include <i386/linux/linux_syscalls.c>
72#include <i386/linux/linux_systrace_args.c>
73#define	MODNAME		"linux"
74#else
75#error Only i386 and amd64 are supported.
76#endif
77extern struct sysent linux_sysent[];
78#define	MAXSYSCALL	LINUX_SYS_MAXSYSCALL
79#define	SYSCALLNAMES	linux_syscallnames
80#define	SYSENT		linux_sysent
81#elif defined(FREEBSD32_SYSTRACE)
82/*
83 * The syscall arguments are processed into a DTrace argument array
84 * using a generated function. See sys/kern/makesyscalls.sh.
85 */
86#include <compat/freebsd32/freebsd32_proto.h>
87#include <compat/freebsd32/freebsd32_util.h>
88#include <compat/freebsd32/freebsd32_syscall.h>
89#include <compat/freebsd32/freebsd32_systrace_args.c>
90extern const char *freebsd32_syscallnames[];
91#define	MODNAME		"freebsd32"
92#define	MAXSYSCALL	FREEBSD32_SYS_MAXSYSCALL
93#define	SYSCALLNAMES	freebsd32_syscallnames
94#define	SYSENT		freebsd32_sysent
95#else
96/*
97 * The syscall arguments are processed into a DTrace argument array
98 * using a generated function. See sys/kern/makesyscalls.sh.
99 */
100#include <sys/syscall.h>
101#include <kern/systrace_args.c>
102#define	MODNAME		"freebsd"
103#define	MAXSYSCALL	SYS_MAXSYSCALL
104#define	SYSCALLNAMES	syscallnames
105#define	SYSENT		sysent
106#endif
107
108#define	PROVNAME	"syscall"
109#define	DEVNAME	        "dtrace/systrace/" MODNAME
110
111#define	SYSTRACE_ARTIFICIAL_FRAMES	1
112
113#define	SYSTRACE_SHIFT			16
114#define	SYSTRACE_ISENTRY(x)		((int)(x) >> SYSTRACE_SHIFT)
115#define	SYSTRACE_SYSNUM(x)		((int)(x) & ((1 << SYSTRACE_SHIFT) - 1))
116#define	SYSTRACE_ENTRY(id)		((1 << SYSTRACE_SHIFT) | (id))
117#define	SYSTRACE_RETURN(id)		(id)
118
119#if ((1 << SYSTRACE_SHIFT) <= MAXSYSCALL)
120#error 1 << SYSTRACE_SHIFT must exceed number of system calls
121#endif
122
123static d_open_t	systrace_open;
124static int	systrace_unload(void);
125static void	systrace_getargdesc(void *, dtrace_id_t, void *, dtrace_argdesc_t *);
126static void	systrace_provide(void *, dtrace_probedesc_t *);
127static void	systrace_destroy(void *, dtrace_id_t, void *);
128static void	systrace_enable(void *, dtrace_id_t, void *);
129static void	systrace_disable(void *, dtrace_id_t, void *);
130static void	systrace_load(void *);
131
132static struct cdevsw systrace_cdevsw = {
133	.d_version	= D_VERSION,
134	.d_open		= systrace_open,
135#ifdef LINUX_SYSTRACE
136	.d_name		= "systrace_" MODNAME,
137#else
138	.d_name		= "systrace",
139#endif
140};
141
142static union	{
143	const char	**p_constnames;
144	char		**pp_syscallnames;
145} uglyhack = { SYSCALLNAMES };
146
147static dtrace_pattr_t systrace_attr = {
148{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
149{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
150{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA },
151{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
152{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA },
153};
154
155static dtrace_pops_t systrace_pops = {
156	systrace_provide,
157	NULL,
158	systrace_enable,
159	systrace_disable,
160	NULL,
161	NULL,
162	systrace_getargdesc,
163	NULL,
164	NULL,
165	systrace_destroy
166};
167
168static struct cdev		*systrace_cdev;
169static dtrace_provider_id_t	systrace_id;
170
171#if !defined(LINUX_SYSTRACE)
172/*
173 * Probe callback function.
174 *
175 * Note: This function is called for _all_ syscalls, regardless of which sysent
176 *       array the syscall comes from. It could be a standard syscall or a
177 *       compat syscall from something like Linux.
178 */
179static void
180systrace_probe(u_int32_t id, int sysnum, struct sysent *sysent, void *params,
181    int ret)
182{
183	int		n_args	= 0;
184	u_int64_t	uargs[8];
185
186	memset(uargs, 0, sizeof(uargs));
187	/*
188	 * Check if this syscall has an argument conversion function
189	 * registered.
190	 */
191	if (params && sysent->sy_systrace_args_func != NULL) {
192		/*
193		 * Convert the syscall parameters using the registered
194		 * function.
195		 */
196		(*sysent->sy_systrace_args_func)(sysnum, params, uargs, &n_args);
197	} else if (params) {
198		/*
199		 * Use the built-in system call argument conversion
200		 * function to translate the syscall structure fields
201		 * into the array of 64-bit values that DTrace
202		 * expects.
203		 */
204		systrace_args(sysnum, params, uargs, &n_args);
205	} else {
206		/*
207		 * Since params is NULL, this is a 'return' probe.
208		 * Set arg0 and arg1 as the return value of this syscall.
209		 */
210		uargs[0] = uargs[1] = ret;
211	}
212
213	/* Process the probe using the converted argments. */
214	dtrace_probe(id, uargs[0], uargs[1], uargs[2], uargs[3], uargs[4]);
215}
216#endif
217
218static void
219systrace_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc)
220{
221	int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
222
223	systrace_setargdesc(sysnum, desc->dtargd_ndx, desc->dtargd_native,
224	    sizeof(desc->dtargd_native));
225
226	if (desc->dtargd_native[0] == '\0')
227		desc->dtargd_ndx = DTRACE_ARGNONE;
228
229	return;
230}
231
232static void
233systrace_provide(void *arg, dtrace_probedesc_t *desc)
234{
235	int i;
236
237	if (desc != NULL)
238		return;
239
240	for (i = 0; i < MAXSYSCALL; i++) {
241		if (dtrace_probe_lookup(systrace_id, MODNAME,
242		    uglyhack.pp_syscallnames[i], "entry") != 0)
243			continue;
244
245		(void) dtrace_probe_create(systrace_id, MODNAME, uglyhack.pp_syscallnames[i],
246		    "entry", SYSTRACE_ARTIFICIAL_FRAMES,
247		    (void *)((uintptr_t)SYSTRACE_ENTRY(i)));
248		(void) dtrace_probe_create(systrace_id, MODNAME, uglyhack.pp_syscallnames[i],
249		    "return", SYSTRACE_ARTIFICIAL_FRAMES,
250		    (void *)((uintptr_t)SYSTRACE_RETURN(i)));
251	}
252}
253
254static void
255systrace_destroy(void *arg, dtrace_id_t id, void *parg)
256{
257#ifdef DEBUG
258	int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
259
260	/*
261	 * There's nothing to do here but assert that we have actually been
262	 * disabled.
263	 */
264	if (SYSTRACE_ISENTRY((uintptr_t)parg)) {
265		ASSERT(sysent[sysnum].sy_entry == 0);
266	} else {
267		ASSERT(sysent[sysnum].sy_return == 0);
268	}
269#endif
270}
271
272static void
273systrace_enable(void *arg, dtrace_id_t id, void *parg)
274{
275	int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
276
277	if (SYSENT[sysnum].sy_systrace_args_func == NULL)
278		SYSENT[sysnum].sy_systrace_args_func = systrace_args;
279
280	if (SYSTRACE_ISENTRY((uintptr_t)parg))
281		SYSENT[sysnum].sy_entry = id;
282	else
283		SYSENT[sysnum].sy_return = id;
284}
285
286static void
287systrace_disable(void *arg, dtrace_id_t id, void *parg)
288{
289	int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
290
291	SYSENT[sysnum].sy_entry = 0;
292	SYSENT[sysnum].sy_return = 0;
293}
294
295static void
296systrace_load(void *dummy)
297{
298	/* Create the /dev/dtrace/systrace entry. */
299	systrace_cdev = make_dev(&systrace_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600,
300	   DEVNAME);
301
302	if (dtrace_register(PROVNAME, &systrace_attr, DTRACE_PRIV_USER,
303	    NULL, &systrace_pops, NULL, &systrace_id) != 0)
304		return;
305
306#if !defined(LINUX_SYSTRACE)
307	systrace_probe_func = systrace_probe;
308#endif
309}
310
311
312static int
313systrace_unload()
314{
315	int error = 0;
316
317	if ((error = dtrace_unregister(systrace_id)) != 0)
318		return (error);
319
320#if !defined(LINUX_SYSTRACE)
321	systrace_probe_func = NULL;
322#endif
323
324	destroy_dev(systrace_cdev);
325
326	return (error);
327}
328
329static int
330systrace_modevent(module_t mod __unused, int type, void *data __unused)
331{
332	int error = 0;
333
334	switch (type) {
335	case MOD_LOAD:
336		break;
337
338	case MOD_UNLOAD:
339		break;
340
341	case MOD_SHUTDOWN:
342		break;
343
344	default:
345		error = EOPNOTSUPP;
346		break;
347
348	}
349	return (error);
350}
351
352static int
353systrace_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused)
354{
355	return (0);
356}
357
358SYSINIT(systrace_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, systrace_load, NULL);
359SYSUNINIT(systrace_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, systrace_unload, NULL);
360
361#ifdef LINUX_SYSTRACE
362DEV_MODULE(systrace_linux32, systrace_modevent, NULL);
363MODULE_VERSION(systrace_linux32, 1);
364MODULE_DEPEND(systrace_linux32, linux, 1, 1, 1);
365MODULE_DEPEND(systrace_linux32, dtrace, 1, 1, 1);
366MODULE_DEPEND(systrace_linux32, opensolaris, 1, 1, 1);
367#elif defined(FREEBSD32_SYSTRACE)
368DEV_MODULE(systrace_freebsd32, systrace_modevent, NULL);
369MODULE_VERSION(systrace_freebsd32, 1);
370MODULE_DEPEND(systrace_freebsd32, dtrace, 1, 1, 1);
371MODULE_DEPEND(systrace_freebsd32, opensolaris, 1, 1, 1);
372#else
373DEV_MODULE(systrace, systrace_modevent, NULL);
374MODULE_VERSION(systrace, 1);
375MODULE_DEPEND(systrace, dtrace, 1, 1, 1);
376MODULE_DEPEND(systrace, opensolaris, 1, 1, 1);
377#endif
378