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 void systrace_load(void *); 139 static void systrace_unload(void *); 140 141 static void systrace_getargdesc(void *, dtrace_id_t, void *, 142 dtrace_argdesc_t *); 143 static uint64_t systrace_getargval(void *, dtrace_id_t, void *, int, int); 144 static void systrace_provide(void *, dtrace_probedesc_t *); 145 static void systrace_destroy(void *, dtrace_id_t, void *); 146 static void systrace_enable(void *, dtrace_id_t, void *); 147 static void systrace_disable(void *, dtrace_id_t, void *); 148 149 static union { 150 const char **p_constnames; 151 char **pp_syscallnames; 152 } uglyhack = { SYSCALLNAMES }; 153 154 static dtrace_pattr_t systrace_attr = { 155 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 156 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, 157 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, 158 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 159 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, 160 }; 161 162 static dtrace_pops_t systrace_pops = { 163 systrace_provide, 164 NULL, 165 systrace_enable, 166 systrace_disable, 167 NULL, 168 NULL, 169 systrace_getargdesc, 170 systrace_getargval, 171 NULL, 172 systrace_destroy 173 }; 174 175 static dtrace_provider_id_t systrace_id; 176 177 #ifdef NATIVE_ABI 178 /* 179 * Probe callback function. 180 * 181 * Note: This function is called for _all_ syscalls, regardless of which sysent 182 * array the syscall comes from. It could be a standard syscall or a 183 * compat syscall from something like Linux. 184 */ 185 static void 186 systrace_probe(struct syscall_args *sa, enum systrace_probe_t type, int retval) 187 { 188 uint64_t uargs[nitems(sa->args)]; 189 dtrace_id_t id; 190 int n_args, sysnum; 191 192 sysnum = sa->code; 193 memset(uargs, 0, sizeof(uargs)); 194 195 if (type == SYSTRACE_ENTRY) { 196 id = sa->callp->sy_entry; 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 id = sa->callp->sy_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 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 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 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 287 systrace_destroy(void *arg, dtrace_id_t id, void *parg) 288 { 289 #ifdef 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 == 0); 298 } else { 299 ASSERT(sysent[sysnum].sy_return == 0); 300 } 301 #endif 302 } 303 304 static void 305 systrace_enable(void *arg, dtrace_id_t id, void *parg) 306 { 307 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 308 309 if (SYSENT[sysnum].sy_systrace_args_func == NULL) 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 } 317 318 static void 319 systrace_disable(void *arg, dtrace_id_t id, void *parg) 320 { 321 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 322 323 SYSENT[sysnum].sy_entry = 0; 324 SYSENT[sysnum].sy_return = 0; 325 } 326 327 static void 328 systrace_load(void *dummy __unused) 329 { 330 331 if (dtrace_register(PROVNAME, &systrace_attr, DTRACE_PRIV_USER, NULL, 332 &systrace_pops, NULL, &systrace_id) != 0) 333 return; 334 335 #ifdef NATIVE_ABI 336 systrace_probe_func = systrace_probe; 337 #endif 338 } 339 340 static void 341 systrace_unload(void *dummy __unused) 342 { 343 344 #ifdef NATIVE_ABI 345 systrace_probe_func = NULL; 346 #endif 347 348 if (dtrace_unregister(systrace_id) != 0) 349 return; 350 } 351 352 static int 353 systrace_modevent(module_t mod __unused, int type, void *data __unused) 354 { 355 int error; 356 357 error = 0; 358 switch (type) { 359 case MOD_LOAD: 360 break; 361 362 case MOD_UNLOAD: 363 break; 364 365 case MOD_SHUTDOWN: 366 break; 367 368 default: 369 error = EOPNOTSUPP; 370 break; 371 372 } 373 return (error); 374 } 375 376 SYSINIT(systrace_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, 377 systrace_load, NULL); 378 SYSUNINIT(systrace_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, 379 systrace_unload, NULL); 380 381 #ifdef LINUX_SYSTRACE 382 DEV_MODULE(systrace_linux, systrace_modevent, NULL); 383 MODULE_VERSION(systrace_linux, 1); 384 #ifdef __amd64__ 385 MODULE_DEPEND(systrace_linux, linux64, 1, 1, 1); 386 #else 387 MODULE_DEPEND(systrace_linux, linux, 1, 1, 1); 388 #endif 389 MODULE_DEPEND(systrace_linux, dtrace, 1, 1, 1); 390 MODULE_DEPEND(systrace_linux, opensolaris, 1, 1, 1); 391 #elif defined(LINUX32_SYSTRACE) 392 DEV_MODULE(systrace_linux32, systrace_modevent, NULL); 393 MODULE_VERSION(systrace_linux32, 1); 394 MODULE_DEPEND(systrace_linux32, linux, 1, 1, 1); 395 MODULE_DEPEND(systrace_linux32, dtrace, 1, 1, 1); 396 MODULE_DEPEND(systrace_linux32, opensolaris, 1, 1, 1); 397 #elif defined(FREEBSD32_SYSTRACE) 398 DEV_MODULE(systrace_freebsd32, systrace_modevent, NULL); 399 MODULE_VERSION(systrace_freebsd32, 1); 400 MODULE_DEPEND(systrace_freebsd32, dtrace, 1, 1, 1); 401 MODULE_DEPEND(systrace_freebsd32, opensolaris, 1, 1, 1); 402 #else 403 DEV_MODULE(systrace, systrace_modevent, NULL); 404 MODULE_VERSION(systrace, 1); 405 MODULE_DEPEND(systrace, dtrace, 1, 1, 1); 406 MODULE_DEPEND(systrace, opensolaris, 1, 1, 1); 407 #endif 408