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
24/*
25 * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
26 * Use is subject to license terms.
27 */
28
29#include <sys/param.h>
30#include <sys/systm.h>
31#include <sys/conf.h>
32#include <sys/cpuvar.h>
33#include <sys/dtrace.h>
34#include <sys/fcntl.h>
35#include <sys/filio.h>
36#include <sys/kdb.h>
37#include <sys/kernel.h>
38#include <sys/kmem.h>
39#include <sys/kthread.h>
40#include <sys/limits.h>
41#include <sys/linker.h>
42#include <sys/lock.h>
43#include <sys/malloc.h>
44#include <sys/module.h>
45#include <sys/mutex.h>
46#include <sys/poll.h>
47#include <sys/proc.h>
48#include <sys/selinfo.h>
49#include <sys/smp.h>
50#include <sys/sysent.h>
51#include <sys/sysproto.h>
52#include <sys/uio.h>
53#include <sys/unistd.h>
54
55#include <cddl/dev/dtrace/dtrace_cddl.h>
56
57#include <machine/stdarg.h>
58
59#ifdef LINUX_SYSTRACE
60#if defined(__amd64__)
61#include <amd64/linux/linux.h>
62#include <amd64/linux/linux_proto.h>
63#include <amd64/linux/linux_syscalls.c>
64#include <amd64/linux/linux_systrace_args.c>
65#elif defined(__i386__)
66#include <i386/linux/linux.h>
67#include <i386/linux/linux_proto.h>
68#include <i386/linux/linux_syscalls.c>
69#include <i386/linux/linux_systrace_args.c>
70#else
71#error Only i386 and amd64 are supported.
72#endif
73#define	MODNAME		"linux"
74extern struct sysent linux_sysent[];
75#define	MAXSYSCALL	LINUX_SYS_MAXSYSCALL
76#define	SYSCALLNAMES	linux_syscallnames
77#define	SYSENT		linux_sysent
78#elif defined(LINUX32_SYSTRACE)
79#if defined(__amd64__)
80#include <amd64/linux32/linux.h>
81#include <amd64/linux32/linux32_proto.h>
82#include <amd64/linux32/linux32_syscalls.c>
83#include <amd64/linux32/linux32_systrace_args.c>
84#else
85#error Only amd64 is supported.
86#endif
87#define	MODNAME		"linux32"
88extern struct sysent linux32_sysent[];
89#define	MAXSYSCALL	LINUX32_SYS_MAXSYSCALL
90#define	SYSCALLNAMES	linux32_syscallnames
91#define	SYSENT		linux32_sysent
92#elif defined(FREEBSD32_SYSTRACE)
93/*
94 * The syscall arguments are processed into a DTrace argument array
95 * using a generated function. See sys/tools/makesyscalls.lua.
96 */
97#include <compat/freebsd32/freebsd32_proto.h>
98#include <compat/freebsd32/freebsd32_util.h>
99#include <compat/freebsd32/freebsd32_syscall.h>
100#include <compat/freebsd32/freebsd32_systrace_args.c>
101extern const char *freebsd32_syscallnames[];
102#define	MODNAME		"freebsd32"
103#define	MAXSYSCALL	FREEBSD32_SYS_MAXSYSCALL
104#define	SYSCALLNAMES	freebsd32_syscallnames
105#define	SYSENT		freebsd32_sysent
106#else
107/*
108 * The syscall arguments are processed into a DTrace argument array
109 * using a generated function. See sys/tools/makesyscalls.lua.
110 */
111#include <sys/syscall.h>
112#include <kern/systrace_args.c>
113#define	MODNAME		"freebsd"
114#define	MAXSYSCALL	SYS_MAXSYSCALL
115#define	SYSCALLNAMES	syscallnames
116#define	SYSENT		sysent
117#define	NATIVE_ABI
118#endif
119
120#define	PROVNAME	"syscall"
121#define	DEVNAME	        "dtrace/systrace/" MODNAME
122
123#define	SYSTRACE_ARTIFICIAL_FRAMES	1
124
125#define	SYSTRACE_SHIFT			16
126#define	SYSTRACE_ISENTRY(x)		((int)(x) >> SYSTRACE_SHIFT)
127#define	SYSTRACE_SYSNUM(x)		((int)(x) & ((1 << SYSTRACE_SHIFT) - 1))
128#define	SYSTRACE_ENTRY(id)		((1 << SYSTRACE_SHIFT) | (id))
129#define	SYSTRACE_RETURN(id)		(id)
130
131#if ((1 << SYSTRACE_SHIFT) <= MAXSYSCALL)
132#error 1 << SYSTRACE_SHIFT must exceed number of system calls
133#endif
134
135static int systrace_enabled_count;
136
137static void	systrace_load(void *);
138static void	systrace_unload(void *);
139
140static void	systrace_getargdesc(void *, dtrace_id_t, void *,
141		    dtrace_argdesc_t *);
142static uint64_t	systrace_getargval(void *, dtrace_id_t, void *, int, int);
143static void	systrace_provide(void *, dtrace_probedesc_t *);
144static void	systrace_destroy(void *, dtrace_id_t, void *);
145static void	systrace_enable(void *, dtrace_id_t, void *);
146static void	systrace_disable(void *, dtrace_id_t, void *);
147
148static union {
149	const char	**p_constnames;
150	char		**pp_syscallnames;
151} uglyhack = { SYSCALLNAMES };
152
153static dtrace_pattr_t systrace_attr = {
154{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
155{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
156{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA },
157{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
158{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA },
159};
160
161static dtrace_pops_t systrace_pops = {
162	.dtps_provide =		systrace_provide,
163	.dtps_provide_module =	NULL,
164	.dtps_enable =		systrace_enable,
165	.dtps_disable =		systrace_disable,
166	.dtps_suspend =		NULL,
167	.dtps_resume =		NULL,
168	.dtps_getargdesc =	systrace_getargdesc,
169	.dtps_getargval =	systrace_getargval,
170	.dtps_usermode =	NULL,
171	.dtps_destroy =		systrace_destroy
172};
173
174static dtrace_provider_id_t	systrace_id;
175
176#ifdef NATIVE_ABI
177/*
178 * Probe callback function.
179 *
180 * Note: This function is called for _all_ syscalls, regardless of which sysent
181 *       array the syscall comes from. It could be a standard syscall or a
182 *       compat syscall from something like Linux.
183 */
184static void
185systrace_probe(struct syscall_args *sa, enum systrace_probe_t type, int retval)
186{
187	uint64_t uargs[nitems(sa->args)];
188	dtrace_id_t id;
189	int n_args, sysnum;
190
191	sysnum = sa->code;
192	memset(uargs, 0, sizeof(uargs));
193
194	if (type == SYSTRACE_ENTRY) {
195		if ((id = sa->callp->sy_entry) == DTRACE_IDNONE)
196			return;
197
198		if (sa->callp->sy_systrace_args_func != NULL)
199			/*
200			 * Convert the syscall parameters using the registered
201			 * function.
202			 */
203			(*sa->callp->sy_systrace_args_func)(sysnum, sa->args,
204			    uargs, &n_args);
205		else
206			/*
207			 * Use the built-in system call argument conversion
208			 * function to translate the syscall structure fields
209			 * into the array of 64-bit values that DTrace expects.
210			 */
211			systrace_args(sysnum, sa->args, uargs, &n_args);
212		/*
213		 * Save probe arguments now so that we can retrieve them if
214		 * the getargval method is called from further down the stack.
215		 */
216		curthread->t_dtrace_systrace_args = uargs;
217	} else {
218		if ((id = sa->callp->sy_return) == DTRACE_IDNONE)
219			return;
220
221		curthread->t_dtrace_systrace_args = NULL;
222		/* Set arg0 and arg1 as the return value of this syscall. */
223		uargs[0] = uargs[1] = retval;
224	}
225
226	/* Process the probe using the converted argments. */
227	dtrace_probe(id, uargs[0], uargs[1], uargs[2], uargs[3], uargs[4]);
228}
229#endif
230
231static void
232systrace_getargdesc(void *arg, dtrace_id_t id, void *parg,
233    dtrace_argdesc_t *desc)
234{
235	int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
236
237	if (SYSTRACE_ISENTRY((uintptr_t)parg))
238		systrace_entry_setargdesc(sysnum, desc->dtargd_ndx,
239		    desc->dtargd_native, sizeof(desc->dtargd_native));
240	else
241		systrace_return_setargdesc(sysnum, desc->dtargd_ndx,
242		    desc->dtargd_native, sizeof(desc->dtargd_native));
243
244	if (desc->dtargd_native[0] == '\0')
245		desc->dtargd_ndx = DTRACE_ARGNONE;
246}
247
248static uint64_t
249systrace_getargval(void *arg __unused, dtrace_id_t id __unused,
250    void *parg __unused, int argno, int aframes __unused)
251{
252	uint64_t *uargs;
253
254	uargs = curthread->t_dtrace_systrace_args;
255	if (uargs == NULL)
256		/* This is a return probe. */
257		return (0);
258	if (argno >= nitems(((struct syscall_args *)NULL)->args))
259		return (0);
260	return (uargs[argno]);
261}
262
263static void
264systrace_provide(void *arg, dtrace_probedesc_t *desc)
265{
266	int i;
267
268	if (desc != NULL)
269		return;
270
271	for (i = 0; i < MAXSYSCALL; i++) {
272		if (dtrace_probe_lookup(systrace_id, MODNAME,
273		    uglyhack.pp_syscallnames[i], "entry") != 0)
274			continue;
275
276		(void)dtrace_probe_create(systrace_id, MODNAME,
277		    uglyhack.pp_syscallnames[i], "entry",
278		    SYSTRACE_ARTIFICIAL_FRAMES,
279		    (void *)((uintptr_t)SYSTRACE_ENTRY(i)));
280		(void)dtrace_probe_create(systrace_id, MODNAME,
281		    uglyhack.pp_syscallnames[i], "return",
282		    SYSTRACE_ARTIFICIAL_FRAMES,
283		    (void *)((uintptr_t)SYSTRACE_RETURN(i)));
284	}
285}
286
287static void
288systrace_destroy(void *arg, dtrace_id_t id, void *parg)
289{
290#ifdef SYSTRACE_DEBUG
291	int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
292
293	/*
294	 * There's nothing to do here but assert that we have actually been
295	 * disabled.
296	 */
297	if (SYSTRACE_ISENTRY((uintptr_t)parg)) {
298		ASSERT(sysent[sysnum].sy_entry == DTRACE_IDNONE);
299	} else {
300		ASSERT(sysent[sysnum].sy_return == DTRACE_IDNONE);
301	}
302#endif
303}
304
305static void
306systrace_enable(void *arg, dtrace_id_t id, void *parg)
307{
308	int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
309
310	SYSENT[sysnum].sy_systrace_args_func = systrace_args;
311
312	if (SYSTRACE_ISENTRY((uintptr_t)parg))
313		SYSENT[sysnum].sy_entry = id;
314	else
315		SYSENT[sysnum].sy_return = id;
316	systrace_enabled_count++;
317	if (systrace_enabled_count == 1)
318		systrace_enabled = true;
319}
320
321static void
322systrace_disable(void *arg, dtrace_id_t id, void *parg)
323{
324	int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
325
326	SYSENT[sysnum].sy_systrace_args_func = NULL;
327	SYSENT[sysnum].sy_entry = DTRACE_IDNONE;
328	SYSENT[sysnum].sy_return = DTRACE_IDNONE;
329	systrace_enabled_count--;
330	if (systrace_enabled_count == 0)
331		systrace_enabled = false;
332}
333
334static void
335systrace_load(void *dummy __unused)
336{
337
338	if (dtrace_register(PROVNAME, &systrace_attr, DTRACE_PRIV_USER, NULL,
339	    &systrace_pops, NULL, &systrace_id) != 0)
340		return;
341
342#ifdef NATIVE_ABI
343	systrace_probe_func = systrace_probe;
344#endif
345}
346
347static void
348systrace_unload(void *dummy __unused)
349{
350
351#ifdef NATIVE_ABI
352	systrace_probe_func = NULL;
353#endif
354
355	if (dtrace_unregister(systrace_id) != 0)
356		return;
357}
358
359static int
360systrace_modevent(module_t mod __unused, int type, void *data __unused)
361{
362	int error;
363
364	error = 0;
365	switch (type) {
366	case MOD_LOAD:
367		break;
368
369	case MOD_UNLOAD:
370		break;
371
372	case MOD_SHUTDOWN:
373		break;
374
375	default:
376		error = EOPNOTSUPP;
377		break;
378
379	}
380	return (error);
381}
382
383SYSINIT(systrace_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY,
384    systrace_load, NULL);
385SYSUNINIT(systrace_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY,
386    systrace_unload, NULL);
387
388#ifdef LINUX_SYSTRACE
389DEV_MODULE(systrace_linux, systrace_modevent, NULL);
390MODULE_VERSION(systrace_linux, 1);
391#ifdef __amd64__
392MODULE_DEPEND(systrace_linux, linux64, 1, 1, 1);
393#else
394MODULE_DEPEND(systrace_linux, linux, 1, 1, 1);
395#endif
396MODULE_DEPEND(systrace_linux, dtrace, 1, 1, 1);
397MODULE_DEPEND(systrace_linux, opensolaris, 1, 1, 1);
398#elif defined(LINUX32_SYSTRACE)
399DEV_MODULE(systrace_linux32, systrace_modevent, NULL);
400MODULE_VERSION(systrace_linux32, 1);
401MODULE_DEPEND(systrace_linux32, linux, 1, 1, 1);
402MODULE_DEPEND(systrace_linux32, dtrace, 1, 1, 1);
403MODULE_DEPEND(systrace_linux32, opensolaris, 1, 1, 1);
404#elif defined(FREEBSD32_SYSTRACE)
405DEV_MODULE(systrace_freebsd32, systrace_modevent, NULL);
406MODULE_VERSION(systrace_freebsd32, 1);
407MODULE_DEPEND(systrace_freebsd32, dtrace, 1, 1, 1);
408MODULE_DEPEND(systrace_freebsd32, opensolaris, 1, 1, 1);
409#else
410DEV_MODULE(systrace, systrace_modevent, NULL);
411MODULE_VERSION(systrace, 1);
412MODULE_DEPEND(systrace, dtrace, 1, 1, 1);
413MODULE_DEPEND(systrace, opensolaris, 1, 1, 1);
414#endif
415