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 * $FreeBSD$ 24 * 25 */ 26 27 /* 28 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 29 * Use is subject to license terms. 30 */ 31 32 #include <sys/cdefs.h> 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/conf.h> 36 #include <sys/cpuvar.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/sysproto.h> 54 #include <sys/sysent.h> 55 #include <sys/uio.h> 56 #include <sys/unistd.h> 57 #include <machine/stdarg.h> 58 59 #include <sys/dtrace.h> 60 61 #ifdef LINUX_SYSTRACE 62 #if defined(__amd64__) 63 #include <amd64/linux/linux.h> 64 #include <amd64/linux/linux_proto.h> 65 #include <amd64/linux/linux_syscalls.c> 66 #include <amd64/linux/linux_systrace_args.c> 67 #elif defined(__i386__) 68 #include <i386/linux/linux.h> 69 #include <i386/linux/linux_proto.h> 70 #include <i386/linux/linux_syscalls.c> 71 #include <i386/linux/linux_systrace_args.c> 72 #else 73 #error Only i386 and amd64 are supported. 74 #endif 75 #define MODNAME "linux" 76 extern struct sysent linux_sysent[]; 77 #define MAXSYSCALL LINUX_SYS_MAXSYSCALL 78 #define SYSCALLNAMES linux_syscallnames 79 #define SYSENT linux_sysent 80 #elif defined(LINUX32_SYSTRACE) 81 #if defined(__amd64__) 82 #include <amd64/linux32/linux.h> 83 #include <amd64/linux32/linux32_proto.h> 84 #include <amd64/linux32/linux32_syscalls.c> 85 #include <amd64/linux32/linux32_systrace_args.c> 86 #else 87 #error Only amd64 is supported. 88 #endif 89 #define MODNAME "linux32" 90 extern struct sysent linux32_sysent[]; 91 #define MAXSYSCALL LINUX32_SYS_MAXSYSCALL 92 #define SYSCALLNAMES linux32_syscallnames 93 #define SYSENT linux32_sysent 94 #elif defined(FREEBSD32_SYSTRACE) 95 /* 96 * The syscall arguments are processed into a DTrace argument array 97 * using a generated function. See sys/kern/makesyscalls.sh. 98 */ 99 #include <compat/freebsd32/freebsd32_proto.h> 100 #include <compat/freebsd32/freebsd32_util.h> 101 #include <compat/freebsd32/freebsd32_syscall.h> 102 #include <compat/freebsd32/freebsd32_systrace_args.c> 103 extern const char *freebsd32_syscallnames[]; 104 #define MODNAME "freebsd32" 105 #define MAXSYSCALL FREEBSD32_SYS_MAXSYSCALL 106 #define SYSCALLNAMES freebsd32_syscallnames 107 #define SYSENT freebsd32_sysent 108 #else 109 /* 110 * The syscall arguments are processed into a DTrace argument array 111 * using a generated function. See sys/kern/makesyscalls.sh. 112 */ 113 #include <sys/syscall.h> 114 #include <kern/systrace_args.c> 115 #define MODNAME "freebsd" 116 #define MAXSYSCALL SYS_MAXSYSCALL 117 #define SYSCALLNAMES syscallnames 118 #define SYSENT sysent 119 #define NATIVE_ABI 120 #endif 121 122 #define PROVNAME "syscall" 123 #define DEVNAME "dtrace/systrace/" MODNAME 124 125 #define SYSTRACE_ARTIFICIAL_FRAMES 1 126 127 #define SYSTRACE_SHIFT 16 128 #define SYSTRACE_ISENTRY(x) ((int)(x) >> SYSTRACE_SHIFT) 129 #define SYSTRACE_SYSNUM(x) ((int)(x) & ((1 << SYSTRACE_SHIFT) - 1)) 130 #define SYSTRACE_ENTRY(id) ((1 << SYSTRACE_SHIFT) | (id)) 131 #define SYSTRACE_RETURN(id) (id) 132 133 #if ((1 << SYSTRACE_SHIFT) <= MAXSYSCALL) 134 #error 1 << SYSTRACE_SHIFT must exceed number of system calls 135 #endif 136 137 static d_open_t systrace_open; 138 static int systrace_unload(void); 139 static void systrace_getargdesc(void *, dtrace_id_t, void *, dtrace_argdesc_t *); 140 static void systrace_provide(void *, dtrace_probedesc_t *); 141 static void systrace_destroy(void *, dtrace_id_t, void *); 142 static void systrace_enable(void *, dtrace_id_t, void *); 143 static void systrace_disable(void *, dtrace_id_t, void *); 144 static void systrace_load(void *); 145 146 static struct cdevsw systrace_cdevsw = { 147 .d_version = D_VERSION, 148 .d_open = systrace_open, 149 #ifndef NATIVE_ABI 150 .d_name = "systrace_" MODNAME, 151 #else 152 .d_name = "systrace", 153 #endif 154 }; 155 156 static union { 157 const char **p_constnames; 158 char **pp_syscallnames; 159 } uglyhack = { SYSCALLNAMES }; 160 161 static dtrace_pattr_t systrace_attr = { 162 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 163 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, 164 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, 165 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 166 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, 167 }; 168 169 static dtrace_pops_t systrace_pops = { 170 systrace_provide, 171 NULL, 172 systrace_enable, 173 systrace_disable, 174 NULL, 175 NULL, 176 systrace_getargdesc, 177 NULL, 178 NULL, 179 systrace_destroy 180 }; 181 182 static struct cdev *systrace_cdev; 183 static dtrace_provider_id_t systrace_id; 184 185 typedef void (*systrace_dtrace_probe_t)(dtrace_id_t, uintptr_t, uintptr_t, 186 uintptr_t, uintptr_t, uintptr_t, uintptr_t, uintptr_t, uintptr_t); 187 188 #ifdef NATIVE_ABI 189 /* 190 * Probe callback function. 191 * 192 * Note: This function is called for _all_ syscalls, regardless of which sysent 193 * array the syscall comes from. It could be a standard syscall or a 194 * compat syscall from something like Linux. 195 */ 196 static void 197 systrace_probe(u_int32_t id, int sysnum, struct sysent *sysent, void *params, 198 int ret) 199 { 200 systrace_dtrace_probe_t probe; 201 int n_args = 0; 202 u_int64_t uargs[8]; 203 204 memset(uargs, 0, sizeof(uargs)); 205 /* 206 * Check if this syscall has an argument conversion function 207 * registered. 208 */ 209 if (params && sysent->sy_systrace_args_func != NULL) { 210 /* 211 * Convert the syscall parameters using the registered 212 * function. 213 */ 214 (*sysent->sy_systrace_args_func)(sysnum, params, uargs, &n_args); 215 } else if (params) { 216 /* 217 * Use the built-in system call argument conversion 218 * function to translate the syscall structure fields 219 * into the array of 64-bit values that DTrace 220 * expects. 221 */ 222 systrace_args(sysnum, params, uargs, &n_args); 223 } else { 224 /* 225 * Since params is NULL, this is a 'return' probe. 226 * Set arg0 and arg1 as the return value of this syscall. 227 */ 228 uargs[0] = uargs[1] = ret; 229 } 230 231 /* Process the probe using the converted argments. */ 232 probe = (systrace_dtrace_probe_t)dtrace_probe; 233 probe(id, uargs[0], uargs[1], uargs[2], uargs[3], uargs[4], uargs[5], 234 uargs[6], uargs[7]); 235 } 236 237 #endif 238 239 static void 240 systrace_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc) 241 { 242 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 243 244 if (SYSTRACE_ISENTRY((uintptr_t)parg)) 245 systrace_entry_setargdesc(sysnum, desc->dtargd_ndx, 246 desc->dtargd_native, sizeof(desc->dtargd_native)); 247 else 248 systrace_return_setargdesc(sysnum, desc->dtargd_ndx, 249 desc->dtargd_native, sizeof(desc->dtargd_native)); 250 251 if (desc->dtargd_native[0] == '\0') 252 desc->dtargd_ndx = DTRACE_ARGNONE; 253 254 return; 255 } 256 257 static void 258 systrace_provide(void *arg, dtrace_probedesc_t *desc) 259 { 260 int i; 261 262 if (desc != NULL) 263 return; 264 265 for (i = 0; i < MAXSYSCALL; i++) { 266 if (dtrace_probe_lookup(systrace_id, MODNAME, 267 uglyhack.pp_syscallnames[i], "entry") != 0) 268 continue; 269 270 (void) dtrace_probe_create(systrace_id, MODNAME, uglyhack.pp_syscallnames[i], 271 "entry", SYSTRACE_ARTIFICIAL_FRAMES, 272 (void *)((uintptr_t)SYSTRACE_ENTRY(i))); 273 (void) dtrace_probe_create(systrace_id, MODNAME, uglyhack.pp_syscallnames[i], 274 "return", SYSTRACE_ARTIFICIAL_FRAMES, 275 (void *)((uintptr_t)SYSTRACE_RETURN(i))); 276 } 277 } 278 279 static void 280 systrace_destroy(void *arg, dtrace_id_t id, void *parg) 281 { 282 #ifdef DEBUG 283 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 284 285 /* 286 * There's nothing to do here but assert that we have actually been 287 * disabled. 288 */ 289 if (SYSTRACE_ISENTRY((uintptr_t)parg)) { 290 ASSERT(sysent[sysnum].sy_entry == 0); 291 } else { 292 ASSERT(sysent[sysnum].sy_return == 0); 293 } 294 #endif 295 } 296 297 static void 298 systrace_enable(void *arg, dtrace_id_t id, void *parg) 299 { 300 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 301 302 if (SYSENT[sysnum].sy_systrace_args_func == NULL) 303 SYSENT[sysnum].sy_systrace_args_func = systrace_args; 304 305 if (SYSTRACE_ISENTRY((uintptr_t)parg)) 306 SYSENT[sysnum].sy_entry = id; 307 else 308 SYSENT[sysnum].sy_return = id; 309 } 310 311 static void 312 systrace_disable(void *arg, dtrace_id_t id, void *parg) 313 { 314 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 315 316 SYSENT[sysnum].sy_entry = 0; 317 SYSENT[sysnum].sy_return = 0; 318 } 319 320 static void 321 systrace_load(void *dummy) 322 { 323 /* Create the /dev/dtrace/systrace entry. */ 324 systrace_cdev = make_dev(&systrace_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, 325 DEVNAME); 326 327 if (dtrace_register(PROVNAME, &systrace_attr, DTRACE_PRIV_USER, 328 NULL, &systrace_pops, NULL, &systrace_id) != 0) 329 return; 330 331 #ifdef NATIVE_ABI 332 systrace_probe_func = systrace_probe; 333 #endif 334 } 335 336 337 static int 338 systrace_unload() 339 { 340 int error = 0; 341 342 if ((error = dtrace_unregister(systrace_id)) != 0) 343 return (error); 344 345 #ifdef NATIVE_ABI 346 systrace_probe_func = NULL; 347 #endif 348 349 destroy_dev(systrace_cdev); 350 351 return (error); 352 } 353 354 static int 355 systrace_modevent(module_t mod __unused, int type, void *data __unused) 356 { 357 int error = 0; 358 359 switch (type) { 360 case MOD_LOAD: 361 break; 362 363 case MOD_UNLOAD: 364 break; 365 366 case MOD_SHUTDOWN: 367 break; 368 369 default: 370 error = EOPNOTSUPP; 371 break; 372 373 } 374 return (error); 375 } 376 377 static int 378 systrace_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused) 379 { 380 return (0); 381 } 382 383 SYSINIT(systrace_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, systrace_load, NULL); 384 SYSUNINIT(systrace_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, systrace_unload, NULL); 385 386 #ifdef LINUX_SYSTRACE 387 DEV_MODULE(systrace_linux, systrace_modevent, NULL); 388 MODULE_VERSION(systrace_linux, 1); 389 #ifdef __amd64__ 390 MODULE_DEPEND(systrace_linux, linux64, 1, 1, 1); 391 #else 392 MODULE_DEPEND(systrace_linux, linux, 1, 1, 1); 393 #endif 394 MODULE_DEPEND(systrace_linux, dtrace, 1, 1, 1); 395 MODULE_DEPEND(systrace_linux, opensolaris, 1, 1, 1); 396 #elif defined(LINUX32_SYSTRACE) 397 DEV_MODULE(systrace_linux32, systrace_modevent, NULL); 398 MODULE_VERSION(systrace_linux32, 1); 399 MODULE_DEPEND(systrace_linux32, linux, 1, 1, 1); 400 MODULE_DEPEND(systrace_linux32, dtrace, 1, 1, 1); 401 MODULE_DEPEND(systrace_linux32, opensolaris, 1, 1, 1); 402 #elif defined(FREEBSD32_SYSTRACE) 403 DEV_MODULE(systrace_freebsd32, systrace_modevent, NULL); 404 MODULE_VERSION(systrace_freebsd32, 1); 405 MODULE_DEPEND(systrace_freebsd32, dtrace, 1, 1, 1); 406 MODULE_DEPEND(systrace_freebsd32, opensolaris, 1, 1, 1); 407 #else 408 DEV_MODULE(systrace, systrace_modevent, NULL); 409 MODULE_VERSION(systrace, 1); 410 MODULE_DEPEND(systrace, dtrace, 1, 1, 1); 411 MODULE_DEPEND(systrace, opensolaris, 1, 1, 1); 412 #endif 413