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 #include <linux.h> 63 #include <linux_syscall.h> 64 #include <linux_proto.h> 65 #include <linux_syscallnames.c> 66 #include <linux_systrace.c> 67 extern struct sysent linux_sysent[]; 68 #define DEVNAME "dtrace/linsystrace" 69 #define PROVNAME "linsyscall" 70 #define MAXSYSCALL LINUX_SYS_MAXSYSCALL 71 #define SYSCALLNAMES linux_syscallnames 72 #define SYSENT linux_sysent 73 #else 74 /* 75 * The syscall arguments are processed into a DTrace argument array 76 * using a generated function. See sys/kern/makesyscalls.sh. 77 */ 78 #include <sys/syscall.h> 79 #include <kern/systrace_args.c> 80 extern const char *syscallnames[]; 81 #define DEVNAME "dtrace/systrace" 82 #define PROVNAME "syscall" 83 #define MAXSYSCALL SYS_MAXSYSCALL 84 #define SYSCALLNAMES syscallnames 85 #define SYSENT sysent 86 #endif 87 88 #define SYSTRACE_ARTIFICIAL_FRAMES 1 89 90 #define SYSTRACE_SHIFT 16 91 #define SYSTRACE_ISENTRY(x) ((int)(x) >> SYSTRACE_SHIFT) 92 #define SYSTRACE_SYSNUM(x) ((int)(x) & ((1 << SYSTRACE_SHIFT) - 1)) 93 #define SYSTRACE_ENTRY(id) ((1 << SYSTRACE_SHIFT) | (id)) 94 #define SYSTRACE_RETURN(id) (id) 95 96 #if ((1 << SYSTRACE_SHIFT) <= MAXSYSCALL) 97 #error 1 << SYSTRACE_SHIFT must exceed number of system calls 98 #endif 99 100 static d_open_t systrace_open; 101 static int systrace_unload(void); 102 static void systrace_getargdesc(void *, dtrace_id_t, void *, dtrace_argdesc_t *); 103 static void systrace_provide(void *, dtrace_probedesc_t *); 104 static void systrace_destroy(void *, dtrace_id_t, void *); 105 static void systrace_enable(void *, dtrace_id_t, void *); 106 static void systrace_disable(void *, dtrace_id_t, void *); 107 static void systrace_load(void *); 108 109 static struct cdevsw systrace_cdevsw = { 110 .d_version = D_VERSION, 111 .d_open = systrace_open, 112 #ifdef LINUX_SYSTRACE 113 .d_name = "linsystrace", 114 #else 115 .d_name = "systrace", 116 #endif 117 }; 118 119 static union { 120 const char **p_constnames; 121 char **pp_syscallnames; 122 } uglyhack = { SYSCALLNAMES }; 123 124 static dtrace_pattr_t systrace_attr = { 125 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 126 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, 127 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, 128 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 129 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, 130 }; 131 132 static dtrace_pops_t systrace_pops = { 133 systrace_provide, 134 NULL, 135 systrace_enable, 136 systrace_disable, 137 NULL, 138 NULL, 139 systrace_getargdesc, 140 NULL, 141 NULL, 142 systrace_destroy 143 }; 144 145 static struct cdev *systrace_cdev; 146 static dtrace_provider_id_t systrace_id; 147 148 #if !defined(LINUX_SYSTRACE) 149 /* 150 * Probe callback function. 151 * 152 * Note: This function is called for _all_ syscalls, regardless of which sysent 153 * array the syscall comes from. It could be a standard syscall or a 154 * compat syscall from something like Linux. 155 */ 156 static void 157 systrace_probe(u_int32_t id, int sysnum, struct sysent *sysent, void *params) 158 { 159 int n_args = 0; 160 u_int64_t uargs[8]; 161 162 /* 163 * Check if this syscall has an argument conversion function 164 * registered. 165 */ 166 if (sysent->sy_systrace_args_func != NULL) 167 /* 168 * Convert the syscall parameters using the registered 169 * function. 170 */ 171 (*sysent->sy_systrace_args_func)(sysnum, params, uargs, &n_args); 172 else 173 /* 174 * Use the built-in system call argument conversion 175 * function to translate the syscall structure fields 176 * into the array of 64-bit values that DTrace 177 * expects. 178 */ 179 systrace_args(sysnum, params, uargs, &n_args); 180 181 /* Process the probe using the converted argments. */ 182 dtrace_probe(id, uargs[0], uargs[1], uargs[2], uargs[3], uargs[4]); 183 } 184 #endif 185 186 static void 187 systrace_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc) 188 { 189 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 190 191 systrace_setargdesc(sysnum, desc->dtargd_ndx, desc->dtargd_native, 192 sizeof(desc->dtargd_native)); 193 194 if (desc->dtargd_native[0] == '\0') 195 desc->dtargd_ndx = DTRACE_ARGNONE; 196 197 return; 198 } 199 200 static void 201 systrace_provide(void *arg, dtrace_probedesc_t *desc) 202 { 203 int i; 204 205 if (desc != NULL) 206 return; 207 208 for (i = 0; i < MAXSYSCALL; i++) { 209 if (dtrace_probe_lookup(systrace_id, NULL, 210 uglyhack.pp_syscallnames[i], "entry") != 0) 211 continue; 212 213 (void) dtrace_probe_create(systrace_id, NULL, uglyhack.pp_syscallnames[i], 214 "entry", SYSTRACE_ARTIFICIAL_FRAMES, 215 (void *)((uintptr_t)SYSTRACE_ENTRY(i))); 216 (void) dtrace_probe_create(systrace_id, NULL, uglyhack.pp_syscallnames[i], 217 "return", SYSTRACE_ARTIFICIAL_FRAMES, 218 (void *)((uintptr_t)SYSTRACE_RETURN(i))); 219 } 220 } 221 222 static void 223 systrace_destroy(void *arg, dtrace_id_t id, void *parg) 224 { 225 #ifdef DEBUG 226 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 227 228 /* 229 * There's nothing to do here but assert that we have actually been 230 * disabled. 231 */ 232 if (SYSTRACE_ISENTRY((uintptr_t)parg)) { 233 ASSERT(sysent[sysnum].sy_entry == 0); 234 } else { 235 ASSERT(sysent[sysnum].sy_return == 0); 236 } 237 #endif 238 } 239 240 static void 241 systrace_enable(void *arg, dtrace_id_t id, void *parg) 242 { 243 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 244 245 if (SYSENT[sysnum].sy_systrace_args_func == NULL) 246 SYSENT[sysnum].sy_systrace_args_func = systrace_args; 247 248 if (SYSTRACE_ISENTRY((uintptr_t)parg)) 249 SYSENT[sysnum].sy_entry = id; 250 else 251 SYSENT[sysnum].sy_return = id; 252 } 253 254 static void 255 systrace_disable(void *arg, dtrace_id_t id, void *parg) 256 { 257 int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); 258 259 SYSENT[sysnum].sy_entry = 0; 260 SYSENT[sysnum].sy_return = 0; 261 } 262 263 static void 264 systrace_load(void *dummy) 265 { 266 /* Create the /dev/dtrace/systrace entry. */ 267 systrace_cdev = make_dev(&systrace_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, 268 DEVNAME); 269 270 if (dtrace_register(PROVNAME, &systrace_attr, DTRACE_PRIV_USER, 271 NULL, &systrace_pops, NULL, &systrace_id) != 0) 272 return; 273 274 #if !defined(LINUX_SYSTRACE) 275 systrace_probe_func = systrace_probe; 276 #endif 277 } 278 279 280 static int 281 systrace_unload() 282 { 283 int error = 0; 284 285 if ((error = dtrace_unregister(systrace_id)) != 0) 286 return (error); 287 288 #if !defined(LINUX_SYSTRACE) 289 systrace_probe_func = NULL; 290 #endif 291 292 destroy_dev(systrace_cdev); 293 294 return (error); 295 } 296 297 static int 298 systrace_modevent(module_t mod __unused, int type, void *data __unused) 299 { 300 int error = 0; 301 302 switch (type) { 303 case MOD_LOAD: 304 break; 305 306 case MOD_UNLOAD: 307 break; 308 309 case MOD_SHUTDOWN: 310 break; 311 312 default: 313 error = EOPNOTSUPP; 314 break; 315 316 } 317 return (error); 318 } 319 320 static int 321 systrace_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused) 322 { 323 return (0); 324 } 325 326 SYSINIT(systrace_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, systrace_load, NULL); 327 SYSUNINIT(systrace_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, systrace_unload, NULL); 328 329 #ifdef LINUX_SYSTRACE 330 DEV_MODULE(linsystrace, systrace_modevent, NULL); 331 MODULE_VERSION(linsystrace, 1); 332 MODULE_DEPEND(linsystrace, linux, 1, 1, 1); 333 MODULE_DEPEND(linsystrace, systrace, 1, 1, 1); 334 MODULE_DEPEND(linsystrace, dtrace, 1, 1, 1); 335 MODULE_DEPEND(linsystrace, opensolaris, 1, 1, 1); 336 #else 337 DEV_MODULE(systrace, systrace_modevent, NULL); 338 MODULE_VERSION(systrace, 1); 339 MODULE_DEPEND(systrace, dtrace, 1, 1, 1); 340 MODULE_DEPEND(systrace, opensolaris, 1, 1, 1); 341 #endif 342