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.
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.
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]
21 * Portions Copyright 2006-2008 John Birrell jb@freebsd.org
28 * Copyright 2006 Sun Microsystems, Inc. All rights reserved.
29 * Use is subject to license terms.
32 #include <sys/cdefs.h>
33 #include <sys/param.h>
34 #include <sys/systm.h>
36 #include <sys/cpuvar.h>
37 #include <sys/fcntl.h>
38 #include <sys/filio.h>
40 #include <sys/kernel.h>
42 #include <sys/kthread.h>
43 #include <sys/limits.h>
44 #include <sys/linker.h>
46 #include <sys/malloc.h>
47 #include <sys/module.h>
48 #include <sys/mutex.h>
51 #include <sys/selinfo.h>
53 #include <sys/sysproto.h>
54 #include <sys/sysent.h>
56 #include <sys/unistd.h>
57 #include <machine/stdarg.h>
59 #include <sys/dtrace.h>
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"
75 #error Only i386 and amd64 are supported.
77 extern 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)
83 * The syscall arguments are processed into a DTrace argument array
84 * using a generated function. See sys/kern/makesyscalls.sh.
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>
90 extern const char *freebsd32_syscallnames[];
91 #define MODNAME "freebsd32"
92 #define MAXSYSCALL FREEBSD32_SYS_MAXSYSCALL
93 #define SYSCALLNAMES freebsd32_syscallnames
94 #define SYSENT freebsd32_sysent
97 * The syscall arguments are processed into a DTrace argument array
98 * using a generated function. See sys/kern/makesyscalls.sh.
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
108 #define PROVNAME "syscall"
109 #define DEVNAME "dtrace/systrace/" MODNAME
111 #define SYSTRACE_ARTIFICIAL_FRAMES 1
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)
119 #if ((1 << SYSTRACE_SHIFT) <= MAXSYSCALL)
120 #error 1 << SYSTRACE_SHIFT must exceed number of system calls
123 static d_open_t systrace_open;
124 static int systrace_unload(void);
125 static void systrace_getargdesc(void *, dtrace_id_t, void *, dtrace_argdesc_t *);
126 static void systrace_provide(void *, dtrace_probedesc_t *);
127 static void systrace_destroy(void *, dtrace_id_t, void *);
128 static void systrace_enable(void *, dtrace_id_t, void *);
129 static void systrace_disable(void *, dtrace_id_t, void *);
130 static void systrace_load(void *);
132 static struct cdevsw systrace_cdevsw = {
133 .d_version = D_VERSION,
134 .d_open = systrace_open,
135 #ifdef LINUX_SYSTRACE
136 .d_name = "systrace_" MODNAME,
138 .d_name = "systrace",
143 const char **p_constnames;
144 char **pp_syscallnames;
145 } uglyhack = { SYSCALLNAMES };
147 static 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 },
155 static dtrace_pops_t systrace_pops = {
168 static struct cdev *systrace_cdev;
169 static dtrace_provider_id_t systrace_id;
171 #if !defined(LINUX_SYSTRACE)
173 * Probe callback function.
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.
180 systrace_probe(u_int32_t id, int sysnum, struct sysent *sysent, void *params,
186 memset(uargs, 0, sizeof(uargs));
188 * Check if this syscall has an argument conversion function
191 if (params && sysent->sy_systrace_args_func != NULL) {
193 * Convert the syscall parameters using the registered
196 (*sysent->sy_systrace_args_func)(sysnum, params, uargs, &n_args);
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
204 systrace_args(sysnum, params, uargs, &n_args);
207 * Since params is NULL, this is a 'return' probe.
208 * Set arg0 and arg1 as the return value of this syscall.
210 uargs[0] = uargs[1] = ret;
213 /* Process the probe using the converted argments. */
214 dtrace_probe(id, uargs[0], uargs[1], uargs[2], uargs[3], uargs[4]);
219 systrace_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc)
221 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
223 systrace_setargdesc(sysnum, desc->dtargd_ndx, desc->dtargd_native,
224 sizeof(desc->dtargd_native));
226 if (desc->dtargd_native[0] == '\0')
227 desc->dtargd_ndx = DTRACE_ARGNONE;
233 systrace_provide(void *arg, dtrace_probedesc_t *desc)
240 for (i = 0; i < MAXSYSCALL; i++) {
241 if (dtrace_probe_lookup(systrace_id, MODNAME,
242 uglyhack.pp_syscallnames[i], "entry") != 0)
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)));
255 systrace_destroy(void *arg, dtrace_id_t id, void *parg)
258 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
261 * There's nothing to do here but assert that we have actually been
264 if (SYSTRACE_ISENTRY((uintptr_t)parg)) {
265 ASSERT(sysent[sysnum].sy_entry == 0);
267 ASSERT(sysent[sysnum].sy_return == 0);
273 systrace_enable(void *arg, dtrace_id_t id, void *parg)
275 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
277 if (SYSENT[sysnum].sy_systrace_args_func == NULL)
278 SYSENT[sysnum].sy_systrace_args_func = systrace_args;
280 if (SYSTRACE_ISENTRY((uintptr_t)parg))
281 SYSENT[sysnum].sy_entry = id;
283 SYSENT[sysnum].sy_return = id;
287 systrace_disable(void *arg, dtrace_id_t id, void *parg)
289 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg);
291 SYSENT[sysnum].sy_entry = 0;
292 SYSENT[sysnum].sy_return = 0;
296 systrace_load(void *dummy)
298 /* Create the /dev/dtrace/systrace entry. */
299 systrace_cdev = make_dev(&systrace_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600,
302 if (dtrace_register(PROVNAME, &systrace_attr, DTRACE_PRIV_USER,
303 NULL, &systrace_pops, NULL, &systrace_id) != 0)
306 #if !defined(LINUX_SYSTRACE)
307 systrace_probe_func = systrace_probe;
317 if ((error = dtrace_unregister(systrace_id)) != 0)
320 #if !defined(LINUX_SYSTRACE)
321 systrace_probe_func = NULL;
324 destroy_dev(systrace_cdev);
330 systrace_modevent(module_t mod __unused, int type, void *data __unused)
353 systrace_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused)
358 SYSINIT(systrace_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, systrace_load, NULL);
359 SYSUNINIT(systrace_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, systrace_unload, NULL);
361 #ifdef LINUX_SYSTRACE
362 DEV_MODULE(systrace_linux32, systrace_modevent, NULL);
363 MODULE_VERSION(systrace_linux32, 1);
364 MODULE_DEPEND(systrace_linux32, linux, 1, 1, 1);
365 MODULE_DEPEND(systrace_linux32, dtrace, 1, 1, 1);
366 MODULE_DEPEND(systrace_linux32, opensolaris, 1, 1, 1);
367 #elif defined(FREEBSD32_SYSTRACE)
368 DEV_MODULE(systrace_freebsd32, systrace_modevent, NULL);
369 MODULE_VERSION(systrace_freebsd32, 1);
370 MODULE_DEPEND(systrace_freebsd32, dtrace, 1, 1, 1);
371 MODULE_DEPEND(systrace_freebsd32, opensolaris, 1, 1, 1);
373 DEV_MODULE(systrace, systrace_modevent, NULL);
374 MODULE_VERSION(systrace, 1);
375 MODULE_DEPEND(systrace, dtrace, 1, 1, 1);
376 MODULE_DEPEND(systrace, opensolaris, 1, 1, 1);