1 /* $NetBSD: systrace.c,v 1.4 2014/01/12 17:49:30 riz Exp $ */ 2 3 /* 4 * CDDL HEADER START 5 * 6 * The contents of this file are subject to the terms of the 7 * Common Development and Distribution License (the "License"). 8 * You may not use this file except in compliance with the License. 9 * 10 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 11 * or http://www.opensolaris.org/os/licensing. 12 * See the License for the specific language governing permissions 13 * and limitations under the License. 14 * 15 * When distributing Covered Code, include this CDDL HEADER in each 16 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 17 * If applicable, add the following below this CDDL HEADER, with the 18 * fields enclosed by brackets "[]" replaced with your own identifying 19 * information: Portions Copyright [yyyy] [name of copyright owner] 20 * 21 * CDDL HEADER END 22 * 23 * Portions Copyright 2006-2008 John Birrell jb@freebsd.org 24 * 25 * $FreeBSD: src/sys/cddl/dev/systrace/systrace.c,v 1.2.2.1 2009/08/03 08:13:06 kensmith Exp $ 26 * 27 */ 28 29 /* 30 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 31 * Use is subject to license terms. 32 */ 33 34 #include <sys/cdefs.h> 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/conf.h> 38 #include <sys/cpuvar.h> 39 #include <sys/fcntl.h> 40 #include <sys/filio.h> 41 #ifdef __FreeBSD__ 42 #include <sys/kdb.h> 43 #endif 44 #include <sys/kernel.h> 45 #include <sys/kmem.h> 46 #include <sys/kthread.h> 47 #include <sys/limits.h> 48 #include <sys/linker.h> 49 #include <sys/lock.h> 50 #include <sys/malloc.h> 51 #include <sys/module.h> 52 #include <sys/mutex.h> 53 #include <sys/poll.h> 54 #include <sys/proc.h> 55 #include <sys/selinfo.h> 56 #ifdef __FreeBSD__ 57 #include <sys/smp.h> 58 #include <sys/sysproto.h> 59 #include <sys/sysent.h> 60 #endif 61 #include <sys/syscallargs.h> 62 #include <sys/uio.h> 63 #include <sys/unistd.h> 64 65 #include <sys/dtrace.h> 66 67 #ifdef LINUX_SYSTRACE 68 #include <linux.h> 69 #include <linux_syscall.h> 70 #include <linux_proto.h> 71 #include <linux_syscallnames.c> 72 #include <linux_systrace.c> 73 extern struct sysent linux_sysent[]; 74 #define DEVNAME "dtrace/linsystrace" 75 #define PROVNAME "linsyscall" 76 #define MAXSYSCALL LINUX_SYS_MAXSYSCALL 77 #define SYSCALLNAMES linux_syscallnames 78 #define SYSENT linux_sysent 79 #else 80 /* 81 * The syscall arguments are processed into a DTrace argument array 82 * using a generated function. See sys/kern/makesyscalls.sh. 83 */ 84 #include <sys/syscall.h> 85 #include <kern/systrace_args.c> 86 extern const char * const syscallnames[]; 87 #define DEVNAME "dtrace/systrace" 88 #define PROVNAME "syscall" 89 #define MAXSYSCALL SYS_MAXSYSCALL 90 #define SYSCALLNAMES syscallnames 91 #define SYSENT sysent 92 #endif 93 94 #define SYSTRACE_ARTIFICIAL_FRAMES 1 95 96 #define SYSTRACE_SHIFT 16 97 #define SYSTRACE_ISENTRY(x) ((int)(x) >> SYSTRACE_SHIFT) 98 #define SYSTRACE_SYSNUM(x) ((int)(x) & ((1 << SYSTRACE_SHIFT) - 1)) 99 #define SYSTRACE_ENTRY(id) ((1 << SYSTRACE_SHIFT) | (id)) 100 #define SYSTRACE_RETURN(id) (id) 101 102 #if ((1 << SYSTRACE_SHIFT) <= MAXSYSCALL) 103 #error 1 << SYSTRACE_SHIFT must exceed number of system calls 104 #endif 105 106 #ifdef __FreeBSD__ 107 static d_open_t systrace_open; 108 #endif 109 static int systrace_unload(void); 110 static void systrace_getargdesc(void *, dtrace_id_t, void *, dtrace_argdesc_t *); 111 static void systrace_provide(void *, const dtrace_probedesc_t *); 112 static void systrace_destroy(void *, dtrace_id_t, void *); 113 static int systrace_enable(void *, dtrace_id_t, void *); 114 static void systrace_disable(void *, dtrace_id_t, void *); 115 static void systrace_load(void *); 116 117 #ifdef __FreeBSD__ 118 static struct cdevsw systrace_cdevsw = { 119 .d_version = D_VERSION, 120 .d_open = systrace_open, 121 #ifdef LINUX_SYSTRACE 122 .d_name = "linsystrace", 123 #else 124 .d_name = "systrace", 125 #endif 126 }; 127 #endif 128 129 static union { 130 const char * const *p_constnames; 131 char **pp_syscallnames; 132 } uglyhack = { SYSCALLNAMES }; 133 134 static dtrace_pattr_t systrace_attr = { 135 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 136 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, 137 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, 138 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 139 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, 140 }; 141 142 static dtrace_pops_t systrace_pops = { 143 systrace_provide, 144 NULL, 145 systrace_enable, 146 systrace_disable, 147 NULL, 148 NULL, 149 systrace_getargdesc, 150 NULL, 151 NULL, 152 systrace_destroy 153 }; 154 155 #ifdef __FreeBSD__ 156 static struct cdev *systrace_cdev; 157 #endif 158 static dtrace_provider_id_t systrace_id; 159 160 #if !defined(LINUX_SYSTRACE) 161 /* 162 * Probe callback function. 163 * 164 * Note: This function is called for _all_ syscalls, regardless of which sysent 165 * array the syscall comes from. It could be a standard syscall or a 166 * compat syscall from something like Linux. 167 */ 168 static void 169 systrace_probe(u_int32_t id, int sysnum, struct sysent *se, void *params) 170 { 171 int n_args = 0; 172 union systrace_probe_args_un uargs[SYS_MAXSYSARGS]; 173 174 /* 175 * Check if this syscall has an argument conversion function 176 * registered. 177 */ 178 if (se->sy_systrace_args_func != NULL) 179 /* 180 * Convert the syscall parameters using the registered 181 * function. 182 */ 183 (*se->sy_systrace_args_func)(sysnum, params, uargs, &n_args); 184 else 185 /* 186 * Use the built-in system call argument conversion 187 * function to translate the syscall structure fields 188 * into the array of 64-bit values that DTrace 189 * expects. 190 */ 191 systrace_args(sysnum, params, uargs, &n_args); 192 193 /* Process the probe using the converted argments. */ 194 dtrace_probe(id, uargs[0].u, uargs[1].u, uargs[2].u, uargs[3].u, 195 uargs[4].u); 196 } 197 #endif 198 199 static void 200 systrace_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc) 201 { 202 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 203 204 systrace_setargdesc(sysnum, desc->dtargd_ndx, desc->dtargd_native, 205 sizeof(desc->dtargd_native)); 206 207 if (desc->dtargd_native[0] == '\0') 208 desc->dtargd_ndx = DTRACE_ARGNONE; 209 210 return; 211 } 212 213 static void 214 systrace_provide(void *arg, const dtrace_probedesc_t *desc) 215 { 216 int i; 217 218 if (desc != NULL) 219 return; 220 221 for (i = 0; i < MAXSYSCALL; i++) { 222 if (dtrace_probe_lookup(systrace_id, NULL, 223 uglyhack.pp_syscallnames[i], "entry") != 0) 224 continue; 225 226 (void) dtrace_probe_create(systrace_id, NULL, uglyhack.pp_syscallnames[i], 227 "entry", SYSTRACE_ARTIFICIAL_FRAMES, 228 (void *)((uintptr_t)SYSTRACE_ENTRY(i))); 229 (void) dtrace_probe_create(systrace_id, NULL, uglyhack.pp_syscallnames[i], 230 "return", SYSTRACE_ARTIFICIAL_FRAMES, 231 (void *)((uintptr_t)SYSTRACE_RETURN(i))); 232 } 233 } 234 235 static void 236 systrace_destroy(void *arg, dtrace_id_t id, void *parg) 237 { 238 #ifdef DEBUG 239 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 240 241 /* 242 * There's nothing to do here but assert that we have actually been 243 * disabled. 244 */ 245 if (SYSTRACE_ISENTRY((uintptr_t)parg)) { 246 ASSERT(sysent[sysnum].sy_entry == 0); 247 } else { 248 ASSERT(sysent[sysnum].sy_return == 0); 249 } 250 #endif 251 } 252 253 static int 254 systrace_enable(void *arg, dtrace_id_t id, void *parg) 255 { 256 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 257 258 if (SYSENT[sysnum].sy_systrace_args_func == NULL) 259 SYSENT[sysnum].sy_systrace_args_func = systrace_args; 260 261 if (SYSTRACE_ISENTRY((uintptr_t)parg)) 262 SYSENT[sysnum].sy_entry = id; 263 else 264 SYSENT[sysnum].sy_return = id; 265 return 0; 266 } 267 268 static void 269 systrace_disable(void *arg, dtrace_id_t id, void *parg) 270 { 271 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 272 273 SYSENT[sysnum].sy_entry = 0; 274 SYSENT[sysnum].sy_return = 0; 275 } 276 277 static void 278 systrace_load(void *dummy) 279 { 280 #ifdef __FreeBSD__ 281 /* Create the /dev/dtrace/systrace entry. */ 282 systrace_cdev = make_dev(&systrace_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, 283 DEVNAME); 284 #endif 285 286 if (dtrace_register(PROVNAME, &systrace_attr, DTRACE_PRIV_USER, 287 NULL, &systrace_pops, NULL, &systrace_id) != 0) 288 return; 289 290 #if !defined(LINUX_SYSTRACE) 291 systrace_probe_func = (systrace_probe_func_t)systrace_probe; 292 #endif 293 } 294 295 296 static int 297 systrace_unload() 298 { 299 int error = 0; 300 301 if ((error = dtrace_unregister(systrace_id)) != 0) 302 return (error); 303 304 #if !defined(LINUX_SYSTRACE) 305 systrace_probe_func = NULL; 306 #endif 307 308 #ifdef __FreeBSD__ 309 destroy_dev(systrace_cdev); 310 #endif 311 312 return (error); 313 } 314 315 #ifdef __FreeBSD__ 316 static int 317 systrace_modevent(module_t mod __unused, int type, void *data __unused) 318 { 319 int error = 0; 320 321 switch (type) { 322 case MOD_LOAD: 323 break; 324 325 case MOD_UNLOAD: 326 break; 327 328 case MOD_SHUTDOWN: 329 break; 330 331 default: 332 error = EOPNOTSUPP; 333 break; 334 335 } 336 return (error); 337 } 338 339 static int 340 systrace_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused) 341 { 342 return (0); 343 } 344 345 SYSINIT(systrace_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, systrace_load, NULL); 346 SYSUNINIT(systrace_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, systrace_unload, NULL); 347 348 #ifdef LINUX_SYSTRACE 349 DEV_MODULE(linsystrace, systrace_modevent, NULL); 350 MODULE_VERSION(linsystrace, 1); 351 MODULE_DEPEND(linsystrace, linux, 1, 1, 1); 352 MODULE_DEPEND(linsystrace, systrace, 1, 1, 1); 353 MODULE_DEPEND(linsystrace, dtrace, 1, 1, 1); 354 MODULE_DEPEND(linsystrace, opensolaris, 1, 1, 1); 355 #else 356 DEV_MODULE(systrace, systrace_modevent, NULL); 357 MODULE_VERSION(systrace, 1); 358 MODULE_DEPEND(systrace, dtrace, 1, 1, 1); 359 MODULE_DEPEND(systrace, opensolaris, 1, 1, 1); 360 #endif 361 #endif /* __FreeBSD__ */ 362 363 #ifdef __NetBSD__ 364 365 static int 366 systrace_modcmd(modcmd_t cmd, void *data) 367 { 368 switch (cmd) { 369 case MODULE_CMD_INIT: 370 systrace_load(NULL); 371 return 0; 372 373 case MODULE_CMD_FINI: 374 systrace_unload(); 375 return 0; 376 377 default: 378 return ENOTTY; 379 } 380 } 381 382 MODULE(MODULE_CLASS_MISC, systrace, "dtrace"); 383 384 #endif 385