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