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/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include <sys/param.h> 33 #include <sys/systm.h> 34 #include <sys/conf.h> 35 #include <sys/cpuvar.h> 36 #include <sys/dtrace.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/sysent.h> 54 #include <sys/sysproto.h> 55 #include <sys/uio.h> 56 #include <sys/unistd.h> 57 58 #include <cddl/dev/dtrace/dtrace_cddl.h> 59 60 #include <machine/stdarg.h> 61 62 #ifdef LINUX_SYSTRACE 63 #if defined(__amd64__) 64 #include <amd64/linux/linux.h> 65 #include <amd64/linux/linux_proto.h> 66 #include <amd64/linux/linux_syscalls.c> 67 #include <amd64/linux/linux_systrace_args.c> 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 #else 74 #error Only i386 and amd64 are supported. 75 #endif 76 #define MODNAME "linux" 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(LINUX32_SYSTRACE) 82 #if defined(__amd64__) 83 #include <amd64/linux32/linux.h> 84 #include <amd64/linux32/linux32_proto.h> 85 #include <amd64/linux32/linux32_syscalls.c> 86 #include <amd64/linux32/linux32_systrace_args.c> 87 #else 88 #error Only amd64 is supported. 89 #endif 90 #define MODNAME "linux32" 91 extern struct sysent linux32_sysent[]; 92 #define MAXSYSCALL LINUX32_SYS_MAXSYSCALL 93 #define SYSCALLNAMES linux32_syscallnames 94 #define SYSENT linux32_sysent 95 #elif defined(FREEBSD32_SYSTRACE) 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 <compat/freebsd32/freebsd32_proto.h> 101 #include <compat/freebsd32/freebsd32_util.h> 102 #include <compat/freebsd32/freebsd32_syscall.h> 103 #include <compat/freebsd32/freebsd32_systrace_args.c> 104 extern const char *freebsd32_syscallnames[]; 105 #define MODNAME "freebsd32" 106 #define MAXSYSCALL FREEBSD32_SYS_MAXSYSCALL 107 #define SYSCALLNAMES freebsd32_syscallnames 108 #define SYSENT freebsd32_sysent 109 #else 110 /* 111 * The syscall arguments are processed into a DTrace argument array 112 * using a generated function. See sys/kern/makesyscalls.sh. 113 */ 114 #include <sys/syscall.h> 115 #include <kern/systrace_args.c> 116 #define MODNAME "freebsd" 117 #define MAXSYSCALL SYS_MAXSYSCALL 118 #define SYSCALLNAMES syscallnames 119 #define SYSENT sysent 120 #define NATIVE_ABI 121 #endif 122 123 #define PROVNAME "syscall" 124 #define DEVNAME "dtrace/systrace/" MODNAME 125 126 #define SYSTRACE_ARTIFICIAL_FRAMES 1 127 128 #define SYSTRACE_SHIFT 16 129 #define SYSTRACE_ISENTRY(x) ((int)(x) >> SYSTRACE_SHIFT) 130 #define SYSTRACE_SYSNUM(x) ((int)(x) & ((1 << SYSTRACE_SHIFT) - 1)) 131 #define SYSTRACE_ENTRY(id) ((1 << SYSTRACE_SHIFT) | (id)) 132 #define SYSTRACE_RETURN(id) (id) 133 134 #if ((1 << SYSTRACE_SHIFT) <= MAXSYSCALL) 135 #error 1 << SYSTRACE_SHIFT must exceed number of system calls 136 #endif 137 138 static int systrace_enabled_count; 139 140 static void systrace_load(void *); 141 static void systrace_unload(void *); 142 143 static void systrace_getargdesc(void *, dtrace_id_t, void *, 144 dtrace_argdesc_t *); 145 static uint64_t systrace_getargval(void *, dtrace_id_t, void *, int, int); 146 static void systrace_provide(void *, dtrace_probedesc_t *); 147 static void systrace_destroy(void *, dtrace_id_t, void *); 148 static void systrace_enable(void *, dtrace_id_t, void *); 149 static void systrace_disable(void *, dtrace_id_t, void *); 150 151 static union { 152 const char **p_constnames; 153 char **pp_syscallnames; 154 } uglyhack = { SYSCALLNAMES }; 155 156 static dtrace_pattr_t systrace_attr = { 157 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 158 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, 159 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, 160 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 161 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, 162 }; 163 164 static dtrace_pops_t systrace_pops = { 165 .dtps_provide = systrace_provide, 166 .dtps_provide_module = NULL, 167 .dtps_enable = systrace_enable, 168 .dtps_disable = systrace_disable, 169 .dtps_suspend = NULL, 170 .dtps_resume = NULL, 171 .dtps_getargdesc = systrace_getargdesc, 172 .dtps_getargval = systrace_getargval, 173 .dtps_usermode = NULL, 174 .dtps_destroy = systrace_destroy 175 }; 176 177 static dtrace_provider_id_t systrace_id; 178 179 #ifdef NATIVE_ABI 180 /* 181 * Probe callback function. 182 * 183 * Note: This function is called for _all_ syscalls, regardless of which sysent 184 * array the syscall comes from. It could be a standard syscall or a 185 * compat syscall from something like Linux. 186 */ 187 static void 188 systrace_probe(struct syscall_args *sa, enum systrace_probe_t type, int retval) 189 { 190 uint64_t uargs[nitems(sa->args)]; 191 dtrace_id_t id; 192 int n_args, sysnum; 193 194 sysnum = sa->code; 195 memset(uargs, 0, sizeof(uargs)); 196 197 if (type == SYSTRACE_ENTRY) { 198 if ((id = sa->callp->sy_entry) == DTRACE_IDNONE) 199 return; 200 201 if (sa->callp->sy_systrace_args_func != NULL) 202 /* 203 * Convert the syscall parameters using the registered 204 * function. 205 */ 206 (*sa->callp->sy_systrace_args_func)(sysnum, sa->args, 207 uargs, &n_args); 208 else 209 /* 210 * Use the built-in system call argument conversion 211 * function to translate the syscall structure fields 212 * into the array of 64-bit values that DTrace expects. 213 */ 214 systrace_args(sysnum, sa->args, uargs, &n_args); 215 /* 216 * Save probe arguments now so that we can retrieve them if 217 * the getargval method is called from further down the stack. 218 */ 219 curthread->t_dtrace_systrace_args = uargs; 220 } else { 221 if ((id = sa->callp->sy_return) == DTRACE_IDNONE) 222 return; 223 224 curthread->t_dtrace_systrace_args = NULL; 225 /* Set arg0 and arg1 as the return value of this syscall. */ 226 uargs[0] = uargs[1] = retval; 227 } 228 229 /* Process the probe using the converted argments. */ 230 dtrace_probe(id, uargs[0], uargs[1], uargs[2], uargs[3], uargs[4]); 231 } 232 #endif 233 234 static void 235 systrace_getargdesc(void *arg, dtrace_id_t id, void *parg, 236 dtrace_argdesc_t *desc) 237 { 238 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 239 240 if (SYSTRACE_ISENTRY((uintptr_t)parg)) 241 systrace_entry_setargdesc(sysnum, desc->dtargd_ndx, 242 desc->dtargd_native, sizeof(desc->dtargd_native)); 243 else 244 systrace_return_setargdesc(sysnum, desc->dtargd_ndx, 245 desc->dtargd_native, sizeof(desc->dtargd_native)); 246 247 if (desc->dtargd_native[0] == '\0') 248 desc->dtargd_ndx = DTRACE_ARGNONE; 249 } 250 251 static uint64_t 252 systrace_getargval(void *arg __unused, dtrace_id_t id __unused, 253 void *parg __unused, int argno, int aframes __unused) 254 { 255 uint64_t *uargs; 256 257 uargs = curthread->t_dtrace_systrace_args; 258 if (uargs == NULL) 259 /* This is a return probe. */ 260 return (0); 261 if (argno >= nitems(((struct syscall_args *)NULL)->args)) 262 return (0); 263 return (uargs[argno]); 264 } 265 266 static void 267 systrace_provide(void *arg, dtrace_probedesc_t *desc) 268 { 269 int i; 270 271 if (desc != NULL) 272 return; 273 274 for (i = 0; i < MAXSYSCALL; i++) { 275 if (dtrace_probe_lookup(systrace_id, MODNAME, 276 uglyhack.pp_syscallnames[i], "entry") != 0) 277 continue; 278 279 (void)dtrace_probe_create(systrace_id, MODNAME, 280 uglyhack.pp_syscallnames[i], "entry", 281 SYSTRACE_ARTIFICIAL_FRAMES, 282 (void *)((uintptr_t)SYSTRACE_ENTRY(i))); 283 (void)dtrace_probe_create(systrace_id, MODNAME, 284 uglyhack.pp_syscallnames[i], "return", 285 SYSTRACE_ARTIFICIAL_FRAMES, 286 (void *)((uintptr_t)SYSTRACE_RETURN(i))); 287 } 288 } 289 290 static void 291 systrace_destroy(void *arg, dtrace_id_t id, void *parg) 292 { 293 #ifdef DEBUG 294 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 295 296 /* 297 * There's nothing to do here but assert that we have actually been 298 * disabled. 299 */ 300 if (SYSTRACE_ISENTRY((uintptr_t)parg)) { 301 ASSERT(sysent[sysnum].sy_entry == 0); 302 } else { 303 ASSERT(sysent[sysnum].sy_return == 0); 304 } 305 #endif 306 } 307 308 static void 309 systrace_enable(void *arg, dtrace_id_t id, void *parg) 310 { 311 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 312 313 if (SYSENT[sysnum].sy_systrace_args_func == NULL) 314 SYSENT[sysnum].sy_systrace_args_func = systrace_args; 315 316 if (SYSTRACE_ISENTRY((uintptr_t)parg)) 317 SYSENT[sysnum].sy_entry = id; 318 else 319 SYSENT[sysnum].sy_return = id; 320 systrace_enabled_count++; 321 if (systrace_enabled_count == 1) 322 systrace_enabled = true; 323 } 324 325 static void 326 systrace_disable(void *arg, dtrace_id_t id, void *parg) 327 { 328 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 329 330 SYSENT[sysnum].sy_entry = 0; 331 SYSENT[sysnum].sy_return = 0; 332 systrace_enabled_count--; 333 if (systrace_enabled_count == 0) 334 systrace_enabled = false; 335 } 336 337 static void 338 systrace_load(void *dummy __unused) 339 { 340 341 if (dtrace_register(PROVNAME, &systrace_attr, DTRACE_PRIV_USER, NULL, 342 &systrace_pops, NULL, &systrace_id) != 0) 343 return; 344 345 #ifdef NATIVE_ABI 346 systrace_probe_func = systrace_probe; 347 #endif 348 } 349 350 static void 351 systrace_unload(void *dummy __unused) 352 { 353 354 #ifdef NATIVE_ABI 355 systrace_probe_func = NULL; 356 #endif 357 358 if (dtrace_unregister(systrace_id) != 0) 359 return; 360 } 361 362 static int 363 systrace_modevent(module_t mod __unused, int type, void *data __unused) 364 { 365 int error; 366 367 error = 0; 368 switch (type) { 369 case MOD_LOAD: 370 break; 371 372 case MOD_UNLOAD: 373 break; 374 375 case MOD_SHUTDOWN: 376 break; 377 378 default: 379 error = EOPNOTSUPP; 380 break; 381 382 } 383 return (error); 384 } 385 386 SYSINIT(systrace_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, 387 systrace_load, NULL); 388 SYSUNINIT(systrace_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, 389 systrace_unload, NULL); 390 391 #ifdef LINUX_SYSTRACE 392 DEV_MODULE(systrace_linux, systrace_modevent, NULL); 393 MODULE_VERSION(systrace_linux, 1); 394 #ifdef __amd64__ 395 MODULE_DEPEND(systrace_linux, linux64, 1, 1, 1); 396 #else 397 MODULE_DEPEND(systrace_linux, linux, 1, 1, 1); 398 #endif 399 MODULE_DEPEND(systrace_linux, dtrace, 1, 1, 1); 400 MODULE_DEPEND(systrace_linux, opensolaris, 1, 1, 1); 401 #elif defined(LINUX32_SYSTRACE) 402 DEV_MODULE(systrace_linux32, systrace_modevent, NULL); 403 MODULE_VERSION(systrace_linux32, 1); 404 MODULE_DEPEND(systrace_linux32, linux, 1, 1, 1); 405 MODULE_DEPEND(systrace_linux32, dtrace, 1, 1, 1); 406 MODULE_DEPEND(systrace_linux32, opensolaris, 1, 1, 1); 407 #elif defined(FREEBSD32_SYSTRACE) 408 DEV_MODULE(systrace_freebsd32, systrace_modevent, NULL); 409 MODULE_VERSION(systrace_freebsd32, 1); 410 MODULE_DEPEND(systrace_freebsd32, dtrace, 1, 1, 1); 411 MODULE_DEPEND(systrace_freebsd32, opensolaris, 1, 1, 1); 412 #else 413 DEV_MODULE(systrace, systrace_modevent, NULL); 414 MODULE_VERSION(systrace, 1); 415 MODULE_DEPEND(systrace, dtrace, 1, 1, 1); 416 MODULE_DEPEND(systrace, opensolaris, 1, 1, 1); 417 #endif 418