| 1 | /* |
| 2 | * Copyright (c) 2005-2006 Apple Computer, Inc. All rights reserved. |
| 3 | * |
| 4 | * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ |
| 5 | * |
| 6 | * This file contains Original Code and/or Modifications of Original Code |
| 7 | * as defined in and that are subject to the Apple Public Source License |
| 8 | * Version 2.0 (the 'License'). You may not use this file except in |
| 9 | * compliance with the License. The rights granted to you under the License |
| 10 | * may not be used to create, or enable the creation or redistribution of, |
| 11 | * unlawful or unlicensed copies of an Apple operating system, or to |
| 12 | * circumvent, violate, or enable the circumvention or violation of, any |
| 13 | * terms of an Apple operating system software license agreement. |
| 14 | * |
| 15 | * Please obtain a copy of the License at |
| 16 | * http://www.opensource.apple.com/apsl/ and read it before using this file. |
| 17 | * |
| 18 | * The Original Code and all software distributed under the License are |
| 19 | * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER |
| 20 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
| 21 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, |
| 22 | * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. |
| 23 | * Please see the License for the specific language governing rights and |
| 24 | * limitations under the License. |
| 25 | * |
| 26 | * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ |
| 27 | */ |
| 28 | |
| 29 | |
| 30 | /* |
| 31 | * APPLE NOTE: This file is compiled even if dtrace is unconfig'd. A symbol |
| 32 | * from this file (_dtrace_register_anon_DOF) always needs to be exported for |
| 33 | * an external kext to link against. |
| 34 | */ |
| 35 | |
| 36 | #if CONFIG_DTRACE |
| 37 | |
| 38 | #define MACH__POSIX_C_SOURCE_PRIVATE 1 /* pulls in suitable savearea from mach/ppc/thread_status.h */ |
| 39 | #include <kern/thread.h> |
| 40 | #include <mach/thread_status.h> |
| 41 | |
| 42 | #include <stdarg.h> |
| 43 | #include <string.h> |
| 44 | #include <sys/malloc.h> |
| 45 | #include <sys/time.h> |
| 46 | #include <sys/proc.h> |
| 47 | #include <sys/proc_internal.h> |
| 48 | #include <sys/kauth.h> |
| 49 | #include <sys/user.h> |
| 50 | #include <sys/systm.h> |
| 51 | #include <sys/dtrace.h> |
| 52 | #include <sys/dtrace_impl.h> |
| 53 | #include <libkern/OSAtomic.h> |
| 54 | #include <libkern/OSKextLibPrivate.h> |
| 55 | #include <kern/kern_types.h> |
| 56 | #include <kern/timer_call.h> |
| 57 | #include <kern/thread_call.h> |
| 58 | #include <kern/task.h> |
| 59 | #include <kern/sched_prim.h> |
| 60 | #include <kern/queue.h> |
| 61 | #include <miscfs/devfs/devfs.h> |
| 62 | #include <kern/kalloc.h> |
| 63 | |
| 64 | #include <mach/vm_param.h> |
| 65 | #include <mach/mach_vm.h> |
| 66 | #include <mach/task.h> |
| 67 | #include <vm/pmap.h> |
| 68 | #include <vm/vm_map.h> /* All the bits we care about are guarded by MACH_KERNEL_PRIVATE :-( */ |
| 69 | |
| 70 | /* |
| 71 | * pid/proc |
| 72 | */ |
| 73 | /* Solaris proc_t is the struct. Darwin's proc_t is a pointer to it. */ |
| 74 | #define proc_t struct proc /* Steer clear of the Darwin typedef for proc_t */ |
| 75 | |
| 76 | void |
| 77 | dtrace_sprlock(proc_t *p) |
| 78 | { |
| 79 | lck_mtx_assert(&p->p_mlock, LCK_MTX_ASSERT_NOTOWNED); |
| 80 | lck_mtx_lock(&p->p_dtrace_sprlock); |
| 81 | } |
| 82 | |
| 83 | void |
| 84 | dtrace_sprunlock(proc_t *p) |
| 85 | { |
| 86 | lck_mtx_unlock(&p->p_dtrace_sprlock); |
| 87 | |
| 88 | } |
| 89 | |
| 90 | /* Not called from probe context */ |
| 91 | proc_t * |
| 92 | sprlock(pid_t pid) |
| 93 | { |
| 94 | proc_t* p; |
| 95 | |
| 96 | if ((p = proc_find(pid)) == PROC_NULL) { |
| 97 | return PROC_NULL; |
| 98 | } |
| 99 | |
| 100 | task_suspend_internal(p->task); |
| 101 | |
| 102 | dtrace_sprlock(p); |
| 103 | |
| 104 | proc_lock(p); |
| 105 | |
| 106 | return p; |
| 107 | } |
| 108 | |
| 109 | /* Not called from probe context */ |
| 110 | void |
| 111 | sprunlock(proc_t *p) |
| 112 | { |
| 113 | if (p != PROC_NULL) { |
| 114 | proc_unlock(p); |
| 115 | |
| 116 | dtrace_sprunlock(p); |
| 117 | |
| 118 | task_resume_internal(p->task); |
| 119 | |
| 120 | proc_rele(p); |
| 121 | } |
| 122 | } |
| 123 | |
| 124 | /* |
| 125 | * uread/uwrite |
| 126 | */ |
| 127 | |
| 128 | // These are not exported from vm_map.h. |
| 129 | extern kern_return_t vm_map_read_user(vm_map_t map, vm_map_address_t src_addr, void *dst_p, vm_size_t size); |
| 130 | extern kern_return_t vm_map_write_user(vm_map_t map, void *src_p, vm_map_address_t dst_addr, vm_size_t size); |
| 131 | |
| 132 | /* Not called from probe context */ |
| 133 | int |
| 134 | uread(proc_t *p, void *buf, user_size_t len, user_addr_t a) |
| 135 | { |
| 136 | kern_return_t ret; |
| 137 | |
| 138 | ASSERT(p != PROC_NULL); |
| 139 | ASSERT(p->task != NULL); |
| 140 | |
| 141 | task_t task = p->task; |
| 142 | |
| 143 | /* |
| 144 | * Grab a reference to the task vm_map_t to make sure |
| 145 | * the map isn't pulled out from under us. |
| 146 | * |
| 147 | * Because the proc_lock is not held at all times on all code |
| 148 | * paths leading here, it is possible for the proc to have |
| 149 | * exited. If the map is null, fail. |
| 150 | */ |
| 151 | vm_map_t map = get_task_map_reference(task); |
| 152 | if (map) { |
| 153 | ret = vm_map_read_user( map, (vm_map_address_t)a, buf, (vm_size_t)len); |
| 154 | vm_map_deallocate(map); |
| 155 | } else |
| 156 | ret = KERN_TERMINATED; |
| 157 | |
| 158 | return (int)ret; |
| 159 | } |
| 160 | |
| 161 | |
| 162 | /* Not called from probe context */ |
| 163 | int |
| 164 | uwrite(proc_t *p, void *buf, user_size_t len, user_addr_t a) |
| 165 | { |
| 166 | kern_return_t ret; |
| 167 | |
| 168 | ASSERT(p != NULL); |
| 169 | ASSERT(p->task != NULL); |
| 170 | |
| 171 | task_t task = p->task; |
| 172 | |
| 173 | /* |
| 174 | * Grab a reference to the task vm_map_t to make sure |
| 175 | * the map isn't pulled out from under us. |
| 176 | * |
| 177 | * Because the proc_lock is not held at all times on all code |
| 178 | * paths leading here, it is possible for the proc to have |
| 179 | * exited. If the map is null, fail. |
| 180 | */ |
| 181 | vm_map_t map = get_task_map_reference(task); |
| 182 | if (map) { |
| 183 | /* Find the memory permissions. */ |
| 184 | uint32_t nestingDepth=999999; |
| 185 | vm_region_submap_short_info_data_64_t info; |
| 186 | mach_msg_type_number_t count = VM_REGION_SUBMAP_SHORT_INFO_COUNT_64; |
| 187 | mach_vm_address_t address = (mach_vm_address_t)a; |
| 188 | mach_vm_size_t sizeOfRegion = (mach_vm_size_t)len; |
| 189 | |
| 190 | ret = mach_vm_region_recurse(map, &address, &sizeOfRegion, &nestingDepth, (vm_region_recurse_info_t)&info, &count); |
| 191 | if (ret != KERN_SUCCESS) |
| 192 | goto done; |
| 193 | |
| 194 | vm_prot_t reprotect; |
| 195 | |
| 196 | if (!(info.protection & VM_PROT_WRITE)) { |
| 197 | /* Save the original protection values for restoration later */ |
| 198 | reprotect = info.protection; |
| 199 | |
| 200 | if (info.max_protection & VM_PROT_WRITE) { |
| 201 | /* The memory is not currently writable, but can be made writable. */ |
| 202 | ret = mach_vm_protect (map, (mach_vm_offset_t)a, (mach_vm_size_t)len, 0, (reprotect & ~VM_PROT_EXECUTE) | VM_PROT_WRITE); |
| 203 | } else { |
| 204 | /* |
| 205 | * The memory is not currently writable, and cannot be made writable. We need to COW this memory. |
| 206 | * |
| 207 | * Strange, we can't just say "reprotect | VM_PROT_COPY", that fails. |
| 208 | */ |
| 209 | ret = mach_vm_protect (map, (mach_vm_offset_t)a, (mach_vm_size_t)len, 0, VM_PROT_COPY | VM_PROT_READ | VM_PROT_WRITE); |
| 210 | } |
| 211 | |
| 212 | if (ret != KERN_SUCCESS) |
| 213 | goto done; |
| 214 | |
| 215 | } else { |
| 216 | /* The memory was already writable. */ |
| 217 | reprotect = VM_PROT_NONE; |
| 218 | } |
| 219 | |
| 220 | ret = vm_map_write_user( map, |
| 221 | buf, |
| 222 | (vm_map_address_t)a, |
| 223 | (vm_size_t)len); |
| 224 | |
| 225 | dtrace_flush_caches(); |
| 226 | |
| 227 | if (ret != KERN_SUCCESS) |
| 228 | goto done; |
| 229 | |
| 230 | if (reprotect != VM_PROT_NONE) { |
| 231 | ASSERT(reprotect & VM_PROT_EXECUTE); |
| 232 | ret = mach_vm_protect (map, (mach_vm_offset_t)a, (mach_vm_size_t)len, 0, reprotect); |
| 233 | } |
| 234 | |
| 235 | done: |
| 236 | vm_map_deallocate(map); |
| 237 | } else |
| 238 | ret = KERN_TERMINATED; |
| 239 | |
| 240 | return (int)ret; |
| 241 | } |
| 242 | |
| 243 | /* |
| 244 | * cpuvar |
| 245 | */ |
| 246 | lck_mtx_t cpu_lock; |
| 247 | lck_mtx_t cyc_lock; |
| 248 | lck_mtx_t mod_lock; |
| 249 | |
| 250 | dtrace_cpu_t *cpu_list; |
| 251 | cpu_core_t *cpu_core; /* XXX TLB lockdown? */ |
| 252 | |
| 253 | /* |
| 254 | * cred_t |
| 255 | */ |
| 256 | |
| 257 | /* |
| 258 | * dtrace_CRED() can be called from probe context. We cannot simply call kauth_cred_get() since |
| 259 | * that function may try to resolve a lazy credential binding, which entails taking the proc_lock. |
| 260 | */ |
| 261 | cred_t * |
| 262 | dtrace_CRED(void) |
| 263 | { |
| 264 | struct uthread *uthread = get_bsdthread_info(current_thread()); |
| 265 | |
| 266 | if (uthread == NULL) |
| 267 | return NULL; |
| 268 | else |
| 269 | return uthread->uu_ucred; /* May return NOCRED which is defined to be 0 */ |
| 270 | } |
| 271 | |
| 272 | #define HAS_ALLPRIVS(cr) priv_isfullset(&CR_OEPRIV(cr)) |
| 273 | #define HAS_PRIVILEGE(cr, pr) ((pr) == PRIV_ALL ? \ |
| 274 | HAS_ALLPRIVS(cr) : \ |
| 275 | PRIV_ISASSERT(&CR_OEPRIV(cr), pr)) |
| 276 | |
| 277 | int PRIV_POLICY_CHOICE(void* cred, int priv, int all) |
| 278 | { |
| 279 | #pragma unused(priv, all) |
| 280 | return kauth_cred_issuser(cred); /* XXX TODO: How is this different from PRIV_POLICY_ONLY? */ |
| 281 | } |
| 282 | |
| 283 | int |
| 284 | PRIV_POLICY_ONLY(void *cr, int priv, int boolean) |
| 285 | { |
| 286 | #pragma unused(priv, boolean) |
| 287 | return kauth_cred_issuser(cr); /* XXX TODO: HAS_PRIVILEGE(cr, priv); */ |
| 288 | } |
| 289 | |
| 290 | uid_t |
| 291 | crgetuid(const cred_t *cr) { cred_t copy_cr = *cr; return kauth_cred_getuid(©_cr); } |
| 292 | |
| 293 | /* |
| 294 | * "cyclic" |
| 295 | */ |
| 296 | |
| 297 | typedef struct wrap_timer_call { |
| 298 | /* node attributes */ |
| 299 | cyc_handler_t hdlr; |
| 300 | cyc_time_t when; |
| 301 | uint64_t deadline; |
| 302 | int cpuid; |
| 303 | boolean_t suspended; |
| 304 | struct timer_call call; |
| 305 | |
| 306 | /* next item in the linked list */ |
| 307 | LIST_ENTRY(wrap_timer_call) entries; |
| 308 | } wrap_timer_call_t; |
| 309 | |
| 310 | #define WAKEUP_REAPER 0x7FFFFFFFFFFFFFFFLL |
| 311 | #define NEARLY_FOREVER 0x7FFFFFFFFFFFFFFELL |
| 312 | |
| 313 | |
| 314 | typedef struct cyc_list { |
| 315 | cyc_omni_handler_t cyl_omni; |
| 316 | wrap_timer_call_t cyl_wrap_by_cpus[]; |
| 317 | #if __arm__ && (__BIGGEST_ALIGNMENT__ > 4) |
| 318 | } __attribute__ ((aligned (8))) cyc_list_t; |
| 319 | #else |
| 320 | } cyc_list_t; |
| 321 | #endif |
| 322 | |
| 323 | /* CPU going online/offline notifications */ |
| 324 | void (*dtrace_cpu_state_changed_hook)(int, boolean_t) = NULL; |
| 325 | void dtrace_cpu_state_changed(int, boolean_t); |
| 326 | |
| 327 | void |
| 328 | dtrace_install_cpu_hooks(void) { |
| 329 | dtrace_cpu_state_changed_hook = dtrace_cpu_state_changed; |
| 330 | } |
| 331 | |
| 332 | void |
| 333 | dtrace_cpu_state_changed(int cpuid, boolean_t is_running) { |
| 334 | #pragma unused(cpuid) |
| 335 | wrap_timer_call_t *wrapTC = NULL; |
| 336 | boolean_t suspend = (is_running ? FALSE : TRUE); |
| 337 | dtrace_icookie_t s; |
| 338 | |
| 339 | /* Ensure that we're not going to leave the CPU */ |
| 340 | s = dtrace_interrupt_disable(); |
| 341 | assert(cpuid == cpu_number()); |
| 342 | |
| 343 | LIST_FOREACH(wrapTC, &(cpu_list[cpu_number()].cpu_cyc_list), entries) { |
| 344 | assert(wrapTC->cpuid == cpu_number()); |
| 345 | if (suspend) { |
| 346 | assert(!wrapTC->suspended); |
| 347 | /* If this fails, we'll panic anyway, so let's do this now. */ |
| 348 | if (!timer_call_cancel(&wrapTC->call)) |
| 349 | panic("timer_call_set_suspend() failed to cancel a timer call" ); |
| 350 | wrapTC->suspended = TRUE; |
| 351 | } else { |
| 352 | /* Rearm the timer, but ensure it was suspended first. */ |
| 353 | assert(wrapTC->suspended); |
| 354 | clock_deadline_for_periodic_event(wrapTC->when.cyt_interval, mach_absolute_time(), |
| 355 | &wrapTC->deadline); |
| 356 | timer_call_enter1(&wrapTC->call, (void*) wrapTC, wrapTC->deadline, |
| 357 | TIMER_CALL_SYS_CRITICAL | TIMER_CALL_LOCAL); |
| 358 | wrapTC->suspended = FALSE; |
| 359 | } |
| 360 | |
| 361 | } |
| 362 | |
| 363 | /* Restore the previous interrupt state. */ |
| 364 | dtrace_interrupt_enable(s); |
| 365 | } |
| 366 | |
| 367 | static void |
| 368 | _timer_call_apply_cyclic( void *ignore, void *vTChdl ) |
| 369 | { |
| 370 | #pragma unused(ignore) |
| 371 | wrap_timer_call_t *wrapTC = (wrap_timer_call_t *)vTChdl; |
| 372 | |
| 373 | (*(wrapTC->hdlr.cyh_func))( wrapTC->hdlr.cyh_arg ); |
| 374 | |
| 375 | clock_deadline_for_periodic_event( wrapTC->when.cyt_interval, mach_absolute_time(), &(wrapTC->deadline) ); |
| 376 | timer_call_enter1( &(wrapTC->call), (void *)wrapTC, wrapTC->deadline, TIMER_CALL_SYS_CRITICAL | TIMER_CALL_LOCAL ); |
| 377 | } |
| 378 | |
| 379 | static cyclic_id_t |
| 380 | timer_call_add_cyclic(wrap_timer_call_t *wrapTC, cyc_handler_t *handler, cyc_time_t *when) |
| 381 | { |
| 382 | uint64_t now; |
| 383 | dtrace_icookie_t s; |
| 384 | |
| 385 | timer_call_setup( &(wrapTC->call), _timer_call_apply_cyclic, NULL ); |
| 386 | wrapTC->hdlr = *handler; |
| 387 | wrapTC->when = *when; |
| 388 | |
| 389 | nanoseconds_to_absolutetime( wrapTC->when.cyt_interval, (uint64_t *)&wrapTC->when.cyt_interval ); |
| 390 | |
| 391 | now = mach_absolute_time(); |
| 392 | wrapTC->deadline = now; |
| 393 | |
| 394 | clock_deadline_for_periodic_event( wrapTC->when.cyt_interval, now, &(wrapTC->deadline) ); |
| 395 | |
| 396 | /* Insert the timer to the list of the running timers on this CPU, and start it. */ |
| 397 | s = dtrace_interrupt_disable(); |
| 398 | wrapTC->cpuid = cpu_number(); |
| 399 | LIST_INSERT_HEAD(&cpu_list[wrapTC->cpuid].cpu_cyc_list, wrapTC, entries); |
| 400 | timer_call_enter1(&wrapTC->call, (void*) wrapTC, wrapTC->deadline, |
| 401 | TIMER_CALL_SYS_CRITICAL | TIMER_CALL_LOCAL); |
| 402 | wrapTC->suspended = FALSE; |
| 403 | dtrace_interrupt_enable(s); |
| 404 | |
| 405 | return (cyclic_id_t)wrapTC; |
| 406 | } |
| 407 | |
| 408 | /* |
| 409 | * Executed on the CPU the timer is running on. |
| 410 | */ |
| 411 | static void |
| 412 | timer_call_remove_cyclic(wrap_timer_call_t *wrapTC) |
| 413 | { |
| 414 | assert(wrapTC); |
| 415 | assert(cpu_number() == wrapTC->cpuid); |
| 416 | |
| 417 | if (!timer_call_cancel(&wrapTC->call)) |
| 418 | panic("timer_call_remove_cyclic() failed to cancel a timer call" ); |
| 419 | |
| 420 | LIST_REMOVE(wrapTC, entries); |
| 421 | } |
| 422 | |
| 423 | static void * |
| 424 | timer_call_get_cyclic_arg(wrap_timer_call_t *wrapTC) |
| 425 | { |
| 426 | return (wrapTC ? wrapTC->hdlr.cyh_arg : NULL); |
| 427 | } |
| 428 | |
| 429 | cyclic_id_t |
| 430 | cyclic_timer_add(cyc_handler_t *handler, cyc_time_t *when) |
| 431 | { |
| 432 | wrap_timer_call_t *wrapTC = _MALLOC(sizeof(wrap_timer_call_t), M_TEMP, M_ZERO | M_WAITOK); |
| 433 | if (NULL == wrapTC) |
| 434 | return CYCLIC_NONE; |
| 435 | else |
| 436 | return timer_call_add_cyclic( wrapTC, handler, when ); |
| 437 | } |
| 438 | |
| 439 | void |
| 440 | cyclic_timer_remove(cyclic_id_t cyclic) |
| 441 | { |
| 442 | ASSERT( cyclic != CYCLIC_NONE ); |
| 443 | |
| 444 | /* Removing a timer call must be done on the CPU the timer is running on. */ |
| 445 | wrap_timer_call_t *wrapTC = (wrap_timer_call_t *) cyclic; |
| 446 | dtrace_xcall(wrapTC->cpuid, (dtrace_xcall_t) timer_call_remove_cyclic, (void*) cyclic); |
| 447 | |
| 448 | _FREE((void *)cyclic, M_TEMP); |
| 449 | } |
| 450 | |
| 451 | static void |
| 452 | _cyclic_add_omni(cyc_list_t *cyc_list) |
| 453 | { |
| 454 | cyc_time_t cT; |
| 455 | cyc_handler_t cH; |
| 456 | cyc_omni_handler_t *omni = &cyc_list->cyl_omni; |
| 457 | |
| 458 | (omni->cyo_online)(omni->cyo_arg, CPU, &cH, &cT); |
| 459 | |
| 460 | wrap_timer_call_t *wrapTC = &cyc_list->cyl_wrap_by_cpus[cpu_number()]; |
| 461 | timer_call_add_cyclic(wrapTC, &cH, &cT); |
| 462 | } |
| 463 | |
| 464 | cyclic_id_list_t |
| 465 | cyclic_add_omni(cyc_omni_handler_t *omni) |
| 466 | { |
| 467 | cyc_list_t *cyc_list = |
| 468 | _MALLOC(sizeof(cyc_list_t) + NCPU * sizeof(wrap_timer_call_t), M_TEMP, M_ZERO | M_WAITOK); |
| 469 | |
| 470 | if (NULL == cyc_list) |
| 471 | return NULL; |
| 472 | |
| 473 | cyc_list->cyl_omni = *omni; |
| 474 | |
| 475 | dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)_cyclic_add_omni, (void *)cyc_list); |
| 476 | |
| 477 | return (cyclic_id_list_t)cyc_list; |
| 478 | } |
| 479 | |
| 480 | static void |
| 481 | _cyclic_remove_omni(cyc_list_t *cyc_list) |
| 482 | { |
| 483 | cyc_omni_handler_t *omni = &cyc_list->cyl_omni; |
| 484 | void *oarg; |
| 485 | wrap_timer_call_t *wrapTC; |
| 486 | |
| 487 | /* |
| 488 | * If the processor was offline when dtrace started, we did not allocate |
| 489 | * a cyclic timer for this CPU. |
| 490 | */ |
| 491 | if ((wrapTC = &cyc_list->cyl_wrap_by_cpus[cpu_number()]) != NULL) { |
| 492 | oarg = timer_call_get_cyclic_arg(wrapTC); |
| 493 | timer_call_remove_cyclic(wrapTC); |
| 494 | (omni->cyo_offline)(omni->cyo_arg, CPU, oarg); |
| 495 | } |
| 496 | } |
| 497 | |
| 498 | void |
| 499 | cyclic_remove_omni(cyclic_id_list_t cyc_list) |
| 500 | { |
| 501 | ASSERT(cyc_list != NULL); |
| 502 | |
| 503 | dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)_cyclic_remove_omni, (void *)cyc_list); |
| 504 | _FREE(cyc_list, M_TEMP); |
| 505 | } |
| 506 | |
| 507 | typedef struct wrap_thread_call { |
| 508 | thread_call_t TChdl; |
| 509 | cyc_handler_t hdlr; |
| 510 | cyc_time_t when; |
| 511 | uint64_t deadline; |
| 512 | } wrap_thread_call_t; |
| 513 | |
| 514 | /* |
| 515 | * _cyclic_apply will run on some thread under kernel_task. That's OK for the |
| 516 | * cleaner and the deadman, but too distant in time and place for the profile provider. |
| 517 | */ |
| 518 | static void |
| 519 | _cyclic_apply( void *ignore, void *vTChdl ) |
| 520 | { |
| 521 | #pragma unused(ignore) |
| 522 | wrap_thread_call_t *wrapTC = (wrap_thread_call_t *)vTChdl; |
| 523 | |
| 524 | (*(wrapTC->hdlr.cyh_func))( wrapTC->hdlr.cyh_arg ); |
| 525 | |
| 526 | clock_deadline_for_periodic_event( wrapTC->when.cyt_interval, mach_absolute_time(), &(wrapTC->deadline) ); |
| 527 | (void)thread_call_enter1_delayed( wrapTC->TChdl, (void *)wrapTC, wrapTC->deadline ); |
| 528 | |
| 529 | /* Did cyclic_remove request a wakeup call when this thread call was re-armed? */ |
| 530 | if (wrapTC->when.cyt_interval == WAKEUP_REAPER) |
| 531 | thread_wakeup((event_t)wrapTC); |
| 532 | } |
| 533 | |
| 534 | cyclic_id_t |
| 535 | cyclic_add(cyc_handler_t *handler, cyc_time_t *when) |
| 536 | { |
| 537 | uint64_t now; |
| 538 | |
| 539 | wrap_thread_call_t *wrapTC = _MALLOC(sizeof(wrap_thread_call_t), M_TEMP, M_ZERO | M_WAITOK); |
| 540 | if (NULL == wrapTC) |
| 541 | return CYCLIC_NONE; |
| 542 | |
| 543 | wrapTC->TChdl = thread_call_allocate( _cyclic_apply, NULL ); |
| 544 | wrapTC->hdlr = *handler; |
| 545 | wrapTC->when = *when; |
| 546 | |
| 547 | ASSERT(when->cyt_when == 0); |
| 548 | ASSERT(when->cyt_interval < WAKEUP_REAPER); |
| 549 | |
| 550 | nanoseconds_to_absolutetime(wrapTC->when.cyt_interval, (uint64_t *)&wrapTC->when.cyt_interval); |
| 551 | |
| 552 | now = mach_absolute_time(); |
| 553 | wrapTC->deadline = now; |
| 554 | |
| 555 | clock_deadline_for_periodic_event( wrapTC->when.cyt_interval, now, &(wrapTC->deadline) ); |
| 556 | (void)thread_call_enter1_delayed( wrapTC->TChdl, (void *)wrapTC, wrapTC->deadline ); |
| 557 | |
| 558 | return (cyclic_id_t)wrapTC; |
| 559 | } |
| 560 | |
| 561 | static void |
| 562 | noop_cyh_func(void * ignore) |
| 563 | { |
| 564 | #pragma unused(ignore) |
| 565 | } |
| 566 | |
| 567 | void |
| 568 | cyclic_remove(cyclic_id_t cyclic) |
| 569 | { |
| 570 | wrap_thread_call_t *wrapTC = (wrap_thread_call_t *)cyclic; |
| 571 | |
| 572 | ASSERT(cyclic != CYCLIC_NONE); |
| 573 | |
| 574 | while (!thread_call_cancel(wrapTC->TChdl)) { |
| 575 | int ret = assert_wait(wrapTC, THREAD_UNINT); |
| 576 | ASSERT(ret == THREAD_WAITING); |
| 577 | |
| 578 | wrapTC->when.cyt_interval = WAKEUP_REAPER; |
| 579 | |
| 580 | ret = thread_block(THREAD_CONTINUE_NULL); |
| 581 | ASSERT(ret == THREAD_AWAKENED); |
| 582 | } |
| 583 | |
| 584 | if (thread_call_free(wrapTC->TChdl)) |
| 585 | _FREE(wrapTC, M_TEMP); |
| 586 | else { |
| 587 | /* Gut this cyclic and move on ... */ |
| 588 | wrapTC->hdlr.cyh_func = noop_cyh_func; |
| 589 | wrapTC->when.cyt_interval = NEARLY_FOREVER; |
| 590 | } |
| 591 | } |
| 592 | |
| 593 | kern_return_t _dtrace_register_anon_DOF(char *, uchar_t *, uint_t); |
| 594 | |
| 595 | kern_return_t |
| 596 | _dtrace_register_anon_DOF(char *name, uchar_t *data, uint_t nelements) |
| 597 | { |
| 598 | #pragma unused(name, data, nelements) |
| 599 | return KERN_FAILURE; |
| 600 | } |
| 601 | |
| 602 | int |
| 603 | ddi_driver_major(dev_info_t *devi) { return (int)major(CAST_DOWN_EXPLICIT(int,devi)); } |
| 604 | |
| 605 | int |
| 606 | ddi_create_minor_node(dev_info_t *dip, const char *name, int spec_type, |
| 607 | minor_t minor_num, const char *node_type, int flag) |
| 608 | { |
| 609 | #pragma unused(spec_type,node_type,flag) |
| 610 | dev_t dev = makedev( ddi_driver_major(dip), minor_num ); |
| 611 | |
| 612 | if (NULL == devfs_make_node( dev, DEVFS_CHAR, UID_ROOT, GID_WHEEL, 0666, name, 0 )) |
| 613 | return DDI_FAILURE; |
| 614 | else |
| 615 | return DDI_SUCCESS; |
| 616 | } |
| 617 | |
| 618 | void |
| 619 | ddi_remove_minor_node(dev_info_t *dip, char *name) |
| 620 | { |
| 621 | #pragma unused(dip,name) |
| 622 | /* XXX called from dtrace_detach, so NOTREACHED for now. */ |
| 623 | } |
| 624 | |
| 625 | major_t |
| 626 | getemajor( dev_t d ) |
| 627 | { |
| 628 | return (major_t) major(d); |
| 629 | } |
| 630 | |
| 631 | minor_t |
| 632 | getminor ( dev_t d ) |
| 633 | { |
| 634 | return (minor_t) minor(d); |
| 635 | } |
| 636 | |
| 637 | extern void Debugger(const char*); |
| 638 | |
| 639 | void |
| 640 | debug_enter(char *c) { Debugger(c); } |
| 641 | |
| 642 | /* |
| 643 | * kmem |
| 644 | */ |
| 645 | |
| 646 | void * |
| 647 | dt_kmem_alloc_site(size_t size, int kmflag, vm_allocation_site_t *site) |
| 648 | { |
| 649 | #pragma unused(kmflag) |
| 650 | |
| 651 | /* |
| 652 | * We ignore the M_NOWAIT bit in kmflag (all of kmflag, in fact). |
| 653 | * Requests larger than 8K with M_NOWAIT fail in kalloc_canblock. |
| 654 | */ |
| 655 | vm_size_t vsize = size; |
| 656 | return kalloc_canblock(&vsize, TRUE, site); |
| 657 | } |
| 658 | |
| 659 | void * |
| 660 | dt_kmem_zalloc_site(size_t size, int kmflag, vm_allocation_site_t *site) |
| 661 | { |
| 662 | #pragma unused(kmflag) |
| 663 | |
| 664 | /* |
| 665 | * We ignore the M_NOWAIT bit in kmflag (all of kmflag, in fact). |
| 666 | * Requests larger than 8K with M_NOWAIT fail in kalloc_canblock. |
| 667 | */ |
| 668 | vm_size_t vsize = size; |
| 669 | void* buf = kalloc_canblock(&vsize, TRUE, site); |
| 670 | |
| 671 | if(!buf) |
| 672 | return NULL; |
| 673 | |
| 674 | bzero(buf, size); |
| 675 | |
| 676 | return buf; |
| 677 | } |
| 678 | |
| 679 | void |
| 680 | dt_kmem_free(void *buf, size_t size) |
| 681 | { |
| 682 | #pragma unused(size) |
| 683 | /* |
| 684 | * DTrace relies on this, its doing a lot of NULL frees. |
| 685 | * A null free causes the debug builds to panic. |
| 686 | */ |
| 687 | if (buf == NULL) return; |
| 688 | |
| 689 | ASSERT(size > 0); |
| 690 | |
| 691 | kfree(buf, size); |
| 692 | } |
| 693 | |
| 694 | |
| 695 | |
| 696 | /* |
| 697 | * aligned dt_kmem allocator |
| 698 | * align should be a power of two |
| 699 | */ |
| 700 | |
| 701 | void* |
| 702 | dt_kmem_alloc_aligned_site(size_t size, size_t align, int kmflag, vm_allocation_site_t *site) |
| 703 | { |
| 704 | void *mem, **addr_to_free; |
| 705 | intptr_t mem_aligned; |
| 706 | size_t *size_to_free, hdr_size; |
| 707 | |
| 708 | /* Must be a power of two. */ |
| 709 | assert(align != 0); |
| 710 | assert((align & (align - 1)) == 0); |
| 711 | |
| 712 | /* |
| 713 | * We are going to add a header to the allocation. It contains |
| 714 | * the address to free and the total size of the buffer. |
| 715 | */ |
| 716 | hdr_size = sizeof(size_t) + sizeof(void*); |
| 717 | mem = dt_kmem_alloc_site(size + align + hdr_size, kmflag, site); |
| 718 | if (mem == NULL) |
| 719 | return NULL; |
| 720 | |
| 721 | mem_aligned = (intptr_t) (((intptr_t) mem + align + hdr_size) & ~(align - 1)); |
| 722 | |
| 723 | /* Write the address to free in the header. */ |
| 724 | addr_to_free = (void**) (mem_aligned - sizeof(void*)); |
| 725 | *addr_to_free = mem; |
| 726 | |
| 727 | /* Write the size to free in the header. */ |
| 728 | size_to_free = (size_t*) (mem_aligned - hdr_size); |
| 729 | *size_to_free = size + align + hdr_size; |
| 730 | |
| 731 | return (void*) mem_aligned; |
| 732 | } |
| 733 | |
| 734 | void* |
| 735 | dt_kmem_zalloc_aligned_site(size_t size, size_t align, int kmflag, vm_allocation_site_t *s) |
| 736 | { |
| 737 | void* buf; |
| 738 | |
| 739 | buf = dt_kmem_alloc_aligned_site(size, align, kmflag, s); |
| 740 | |
| 741 | if(!buf) |
| 742 | return NULL; |
| 743 | |
| 744 | bzero(buf, size); |
| 745 | |
| 746 | return buf; |
| 747 | } |
| 748 | |
| 749 | void |
| 750 | dt_kmem_free_aligned(void* buf, size_t size) |
| 751 | { |
| 752 | #pragma unused(size) |
| 753 | intptr_t ptr = (intptr_t) buf; |
| 754 | void **addr_to_free = (void**) (ptr - sizeof(void*)); |
| 755 | size_t *size_to_free = (size_t*) (ptr - (sizeof(size_t) + sizeof(void*))); |
| 756 | |
| 757 | if (buf == NULL) |
| 758 | return; |
| 759 | |
| 760 | dt_kmem_free(*addr_to_free, *size_to_free); |
| 761 | } |
| 762 | |
| 763 | /* |
| 764 | * dtrace wants to manage just a single block: dtrace_state_percpu_t * NCPU, and |
| 765 | * doesn't specify constructor, destructor, or reclaim methods. |
| 766 | * At present, it always zeroes the block it obtains from kmem_cache_alloc(). |
| 767 | * We'll manage this constricted use of kmem_cache with ordinary _MALLOC and _FREE. |
| 768 | */ |
| 769 | kmem_cache_t * |
| 770 | kmem_cache_create( |
| 771 | const char *name, /* descriptive name for this cache */ |
| 772 | size_t bufsize, /* size of the objects it manages */ |
| 773 | size_t align, /* required object alignment */ |
| 774 | int (*constructor)(void *, void *, int), /* object constructor */ |
| 775 | void (*destructor)(void *, void *), /* object destructor */ |
| 776 | void (*reclaim)(void *), /* memory reclaim callback */ |
| 777 | void *private, /* pass-thru arg for constr/destr/reclaim */ |
| 778 | vmem_t *vmp, /* vmem source for slab allocation */ |
| 779 | int cflags) /* cache creation flags */ |
| 780 | { |
| 781 | #pragma unused(name,align,constructor,destructor,reclaim,private,vmp,cflags) |
| 782 | return (kmem_cache_t *)bufsize; /* A cookie that tracks the single object size. */ |
| 783 | } |
| 784 | |
| 785 | void * |
| 786 | kmem_cache_alloc(kmem_cache_t *cp, int kmflag) |
| 787 | { |
| 788 | #pragma unused(kmflag) |
| 789 | size_t bufsize = (size_t)cp; |
| 790 | return (void *)_MALLOC(bufsize, M_TEMP, M_WAITOK); |
| 791 | } |
| 792 | |
| 793 | void |
| 794 | kmem_cache_free(kmem_cache_t *cp, void *buf) |
| 795 | { |
| 796 | #pragma unused(cp) |
| 797 | _FREE(buf, M_TEMP); |
| 798 | } |
| 799 | |
| 800 | void |
| 801 | kmem_cache_destroy(kmem_cache_t *cp) |
| 802 | { |
| 803 | #pragma unused(cp) |
| 804 | } |
| 805 | |
| 806 | /* |
| 807 | * vmem (Solaris "slab" allocator) used by DTrace solely to hand out resource ids |
| 808 | */ |
| 809 | typedef unsigned int u_daddr_t; |
| 810 | #include "blist.h" |
| 811 | |
| 812 | /* By passing around blist *handles*, the underlying blist can be resized as needed. */ |
| 813 | struct blist_hdl { |
| 814 | blist_t blist; |
| 815 | }; |
| 816 | |
| 817 | vmem_t * |
| 818 | vmem_create(const char *name, void *base, size_t size, size_t quantum, void *ignore5, |
| 819 | void *ignore6, vmem_t *source, size_t qcache_max, int vmflag) |
| 820 | { |
| 821 | #pragma unused(name,quantum,ignore5,ignore6,source,qcache_max,vmflag) |
| 822 | blist_t bl; |
| 823 | struct blist_hdl *p = _MALLOC(sizeof(struct blist_hdl), M_TEMP, M_WAITOK); |
| 824 | |
| 825 | ASSERT(quantum == 1); |
| 826 | ASSERT(NULL == ignore5); |
| 827 | ASSERT(NULL == ignore6); |
| 828 | ASSERT(NULL == source); |
| 829 | ASSERT(0 == qcache_max); |
| 830 | ASSERT(vmflag & VMC_IDENTIFIER); |
| 831 | |
| 832 | size = MIN(128, size); /* Clamp to 128 initially, since the underlying data structure is pre-allocated */ |
| 833 | |
| 834 | p->blist = bl = blist_create( size ); |
| 835 | blist_free(bl, 0, size); |
| 836 | if (base) blist_alloc( bl, (daddr_t)(uintptr_t)base ); /* Chomp off initial ID(s) */ |
| 837 | |
| 838 | return (vmem_t *)p; |
| 839 | } |
| 840 | |
| 841 | void * |
| 842 | vmem_alloc(vmem_t *vmp, size_t size, int vmflag) |
| 843 | { |
| 844 | #pragma unused(vmflag) |
| 845 | struct blist_hdl *q = (struct blist_hdl *)vmp; |
| 846 | blist_t bl = q->blist; |
| 847 | daddr_t p; |
| 848 | |
| 849 | p = blist_alloc(bl, (daddr_t)size); |
| 850 | |
| 851 | if ((daddr_t)-1 == p) { |
| 852 | blist_resize(&bl, (bl->bl_blocks) << 1, 1); |
| 853 | q->blist = bl; |
| 854 | p = blist_alloc(bl, (daddr_t)size); |
| 855 | if ((daddr_t)-1 == p) |
| 856 | panic("vmem_alloc: failure after blist_resize!" ); |
| 857 | } |
| 858 | |
| 859 | return (void *)(uintptr_t)p; |
| 860 | } |
| 861 | |
| 862 | void |
| 863 | vmem_free(vmem_t *vmp, void *vaddr, size_t size) |
| 864 | { |
| 865 | struct blist_hdl *p = (struct blist_hdl *)vmp; |
| 866 | |
| 867 | blist_free( p->blist, (daddr_t)(uintptr_t)vaddr, (daddr_t)size ); |
| 868 | } |
| 869 | |
| 870 | void |
| 871 | vmem_destroy(vmem_t *vmp) |
| 872 | { |
| 873 | struct blist_hdl *p = (struct blist_hdl *)vmp; |
| 874 | |
| 875 | blist_destroy( p->blist ); |
| 876 | _FREE( p, sizeof(struct blist_hdl) ); |
| 877 | } |
| 878 | |
| 879 | /* |
| 880 | * Timing |
| 881 | */ |
| 882 | |
| 883 | /* |
| 884 | * dtrace_gethrestime() provides the "walltimestamp", a value that is anchored at |
| 885 | * January 1, 1970. Because it can be called from probe context, it must take no locks. |
| 886 | */ |
| 887 | |
| 888 | hrtime_t |
| 889 | dtrace_gethrestime(void) |
| 890 | { |
| 891 | clock_sec_t secs; |
| 892 | clock_nsec_t nanosecs; |
| 893 | uint64_t secs64, ns64; |
| 894 | |
| 895 | clock_get_calendar_nanotime_nowait(&secs, &nanosecs); |
| 896 | secs64 = (uint64_t)secs; |
| 897 | ns64 = (uint64_t)nanosecs; |
| 898 | |
| 899 | ns64 = ns64 + (secs64 * 1000000000LL); |
| 900 | return ns64; |
| 901 | } |
| 902 | |
| 903 | /* |
| 904 | * dtrace_gethrtime() provides high-resolution timestamps with machine-dependent origin. |
| 905 | * Hence its primary use is to specify intervals. |
| 906 | */ |
| 907 | |
| 908 | hrtime_t |
| 909 | dtrace_abs_to_nano(uint64_t elapsed) |
| 910 | { |
| 911 | static mach_timebase_info_data_t sTimebaseInfo = { 0, 0 }; |
| 912 | |
| 913 | /* |
| 914 | * If this is the first time we've run, get the timebase. |
| 915 | * We can use denom == 0 to indicate that sTimebaseInfo is |
| 916 | * uninitialised because it makes no sense to have a zero |
| 917 | * denominator in a fraction. |
| 918 | */ |
| 919 | |
| 920 | if ( sTimebaseInfo.denom == 0 ) { |
| 921 | (void) clock_timebase_info(&sTimebaseInfo); |
| 922 | } |
| 923 | |
| 924 | /* |
| 925 | * Convert to nanoseconds. |
| 926 | * return (elapsed * (uint64_t)sTimebaseInfo.numer)/(uint64_t)sTimebaseInfo.denom; |
| 927 | * |
| 928 | * Provided the final result is representable in 64 bits the following maneuver will |
| 929 | * deliver that result without intermediate overflow. |
| 930 | */ |
| 931 | if (sTimebaseInfo.denom == sTimebaseInfo.numer) |
| 932 | return elapsed; |
| 933 | else if (sTimebaseInfo.denom == 1) |
| 934 | return elapsed * (uint64_t)sTimebaseInfo.numer; |
| 935 | else { |
| 936 | /* Decompose elapsed = eta32 * 2^32 + eps32: */ |
| 937 | uint64_t eta32 = elapsed >> 32; |
| 938 | uint64_t eps32 = elapsed & 0x00000000ffffffffLL; |
| 939 | |
| 940 | uint32_t numer = sTimebaseInfo.numer, denom = sTimebaseInfo.denom; |
| 941 | |
| 942 | /* Form product of elapsed64 (decomposed) and numer: */ |
| 943 | uint64_t mu64 = numer * eta32; |
| 944 | uint64_t lambda64 = numer * eps32; |
| 945 | |
| 946 | /* Divide the constituents by denom: */ |
| 947 | uint64_t q32 = mu64/denom; |
| 948 | uint64_t r32 = mu64 - (q32 * denom); /* mu64 % denom */ |
| 949 | |
| 950 | return (q32 << 32) + ((r32 << 32) + lambda64)/denom; |
| 951 | } |
| 952 | } |
| 953 | |
| 954 | hrtime_t |
| 955 | dtrace_gethrtime(void) |
| 956 | { |
| 957 | static uint64_t start = 0; |
| 958 | |
| 959 | if (start == 0) |
| 960 | start = mach_absolute_time(); |
| 961 | |
| 962 | return dtrace_abs_to_nano(mach_absolute_time() - start); |
| 963 | } |
| 964 | |
| 965 | /* |
| 966 | * Atomicity and synchronization |
| 967 | */ |
| 968 | uint32_t |
| 969 | dtrace_cas32(uint32_t *target, uint32_t cmp, uint32_t new) |
| 970 | { |
| 971 | if (OSCompareAndSwap( (UInt32)cmp, (UInt32)new, (volatile UInt32 *)target )) |
| 972 | return cmp; |
| 973 | else |
| 974 | return ~cmp; /* Must return something *other* than cmp */ |
| 975 | } |
| 976 | |
| 977 | void * |
| 978 | dtrace_casptr(void *target, void *cmp, void *new) |
| 979 | { |
| 980 | if (OSCompareAndSwapPtr( cmp, new, (void**)target )) |
| 981 | return cmp; |
| 982 | else |
| 983 | return (void *)(~(uintptr_t)cmp); /* Must return something *other* than cmp */ |
| 984 | } |
| 985 | |
| 986 | /* |
| 987 | * Interrupt manipulation |
| 988 | */ |
| 989 | dtrace_icookie_t |
| 990 | dtrace_interrupt_disable(void) |
| 991 | { |
| 992 | return (dtrace_icookie_t)ml_set_interrupts_enabled(FALSE); |
| 993 | } |
| 994 | |
| 995 | void |
| 996 | dtrace_interrupt_enable(dtrace_icookie_t reenable) |
| 997 | { |
| 998 | (void)ml_set_interrupts_enabled((boolean_t)reenable); |
| 999 | } |
| 1000 | |
| 1001 | /* |
| 1002 | * MP coordination |
| 1003 | */ |
| 1004 | static void |
| 1005 | dtrace_sync_func(void) {} |
| 1006 | |
| 1007 | /* |
| 1008 | * dtrace_sync() is not called from probe context. |
| 1009 | */ |
| 1010 | void |
| 1011 | dtrace_sync(void) |
| 1012 | { |
| 1013 | dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_sync_func, NULL); |
| 1014 | } |
| 1015 | |
| 1016 | /* |
| 1017 | * The dtrace_copyin/out/instr and dtrace_fuword* routines can be called from probe context. |
| 1018 | */ |
| 1019 | |
| 1020 | extern kern_return_t dtrace_copyio_preflight(addr64_t); |
| 1021 | extern kern_return_t dtrace_copyio_postflight(addr64_t); |
| 1022 | |
| 1023 | static int |
| 1024 | dtrace_copycheck(user_addr_t uaddr, uintptr_t kaddr, size_t size) |
| 1025 | { |
| 1026 | #pragma unused(kaddr) |
| 1027 | |
| 1028 | vm_offset_t recover = dtrace_set_thread_recover( current_thread(), 0 ); /* Snare any extant recovery point. */ |
| 1029 | dtrace_set_thread_recover( current_thread(), recover ); /* Put it back. We *must not* re-enter and overwrite. */ |
| 1030 | |
| 1031 | ASSERT(kaddr + size >= kaddr); |
| 1032 | |
| 1033 | if ( uaddr + size < uaddr || /* Avoid address wrap. */ |
| 1034 | KERN_FAILURE == dtrace_copyio_preflight(uaddr)) /* Machine specific setup/constraints. */ |
| 1035 | { |
| 1036 | DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); |
| 1037 | cpu_core[CPU->cpu_id].cpuc_dtrace_illval = uaddr; |
| 1038 | return (0); |
| 1039 | } |
| 1040 | return (1); |
| 1041 | } |
| 1042 | |
| 1043 | void |
| 1044 | dtrace_copyin(user_addr_t src, uintptr_t dst, size_t len, volatile uint16_t *flags) |
| 1045 | { |
| 1046 | #pragma unused(flags) |
| 1047 | |
| 1048 | if (dtrace_copycheck( src, dst, len )) { |
| 1049 | if (copyin((const user_addr_t)src, (char *)dst, (vm_size_t)len)) { |
| 1050 | DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); |
| 1051 | cpu_core[CPU->cpu_id].cpuc_dtrace_illval = src; |
| 1052 | } |
| 1053 | dtrace_copyio_postflight(src); |
| 1054 | } |
| 1055 | } |
| 1056 | |
| 1057 | void |
| 1058 | dtrace_copyinstr(user_addr_t src, uintptr_t dst, size_t len, volatile uint16_t *flags) |
| 1059 | { |
| 1060 | #pragma unused(flags) |
| 1061 | |
| 1062 | size_t actual; |
| 1063 | |
| 1064 | if (dtrace_copycheck( src, dst, len )) { |
| 1065 | /* copyin as many as 'len' bytes. */ |
| 1066 | int error = copyinstr((const user_addr_t)src, (char *)dst, (vm_size_t)len, &actual); |
| 1067 | |
| 1068 | /* |
| 1069 | * ENAMETOOLONG is returned when 'len' bytes have been copied in but the NUL terminator was |
| 1070 | * not encountered. That does not require raising CPU_DTRACE_BADADDR, and we press on. |
| 1071 | * Note that we do *not* stuff a NUL terminator when returning ENAMETOOLONG, that's left |
| 1072 | * to the caller. |
| 1073 | */ |
| 1074 | if (error && error != ENAMETOOLONG) { |
| 1075 | DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); |
| 1076 | cpu_core[CPU->cpu_id].cpuc_dtrace_illval = src; |
| 1077 | } |
| 1078 | dtrace_copyio_postflight(src); |
| 1079 | } |
| 1080 | } |
| 1081 | |
| 1082 | void |
| 1083 | dtrace_copyout(uintptr_t src, user_addr_t dst, size_t len, volatile uint16_t *flags) |
| 1084 | { |
| 1085 | #pragma unused(flags) |
| 1086 | |
| 1087 | if (dtrace_copycheck( dst, src, len )) { |
| 1088 | if (copyout((const void *)src, dst, (vm_size_t)len)) { |
| 1089 | DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); |
| 1090 | cpu_core[CPU->cpu_id].cpuc_dtrace_illval = dst; |
| 1091 | } |
| 1092 | dtrace_copyio_postflight(dst); |
| 1093 | } |
| 1094 | } |
| 1095 | |
| 1096 | void |
| 1097 | dtrace_copyoutstr(uintptr_t src, user_addr_t dst, size_t len, volatile uint16_t *flags) |
| 1098 | { |
| 1099 | #pragma unused(flags) |
| 1100 | |
| 1101 | size_t actual; |
| 1102 | |
| 1103 | if (dtrace_copycheck( dst, src, len )) { |
| 1104 | |
| 1105 | /* |
| 1106 | * ENAMETOOLONG is returned when 'len' bytes have been copied out but the NUL terminator was |
| 1107 | * not encountered. We raise CPU_DTRACE_BADADDR in that case. |
| 1108 | * Note that we do *not* stuff a NUL terminator when returning ENAMETOOLONG, that's left |
| 1109 | * to the caller. |
| 1110 | */ |
| 1111 | if (copyoutstr((const void *)src, dst, (size_t)len, &actual)) { |
| 1112 | DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); |
| 1113 | cpu_core[CPU->cpu_id].cpuc_dtrace_illval = dst; |
| 1114 | } |
| 1115 | dtrace_copyio_postflight(dst); |
| 1116 | } |
| 1117 | } |
| 1118 | |
| 1119 | extern const int copysize_limit_panic; |
| 1120 | |
| 1121 | int dtrace_copy_maxsize(void) |
| 1122 | { |
| 1123 | return copysize_limit_panic; |
| 1124 | } |
| 1125 | |
| 1126 | |
| 1127 | int |
| 1128 | dtrace_buffer_copyout(const void *kaddr, user_addr_t uaddr, vm_size_t nbytes) |
| 1129 | { |
| 1130 | int maxsize = dtrace_copy_maxsize(); |
| 1131 | /* |
| 1132 | * Partition the copyout in copysize_limit_panic-sized chunks |
| 1133 | */ |
| 1134 | while (nbytes >= (vm_size_t)maxsize) { |
| 1135 | if (copyout(kaddr, uaddr, maxsize) != 0) |
| 1136 | return (EFAULT); |
| 1137 | |
| 1138 | nbytes -= maxsize; |
| 1139 | uaddr += maxsize; |
| 1140 | kaddr += maxsize; |
| 1141 | } |
| 1142 | if (nbytes > 0) { |
| 1143 | if (copyout(kaddr, uaddr, nbytes) != 0) |
| 1144 | return (EFAULT); |
| 1145 | } |
| 1146 | |
| 1147 | return (0); |
| 1148 | } |
| 1149 | |
| 1150 | uint8_t |
| 1151 | dtrace_fuword8(user_addr_t uaddr) |
| 1152 | { |
| 1153 | uint8_t ret = 0; |
| 1154 | |
| 1155 | DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); |
| 1156 | if (dtrace_copycheck( uaddr, (uintptr_t)&ret, sizeof(ret))) { |
| 1157 | if (copyin((const user_addr_t)uaddr, (char *)&ret, sizeof(ret))) { |
| 1158 | DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); |
| 1159 | cpu_core[CPU->cpu_id].cpuc_dtrace_illval = uaddr; |
| 1160 | } |
| 1161 | dtrace_copyio_postflight(uaddr); |
| 1162 | } |
| 1163 | DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); |
| 1164 | |
| 1165 | return(ret); |
| 1166 | } |
| 1167 | |
| 1168 | uint16_t |
| 1169 | dtrace_fuword16(user_addr_t uaddr) |
| 1170 | { |
| 1171 | uint16_t ret = 0; |
| 1172 | |
| 1173 | DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); |
| 1174 | if (dtrace_copycheck( uaddr, (uintptr_t)&ret, sizeof(ret))) { |
| 1175 | if (copyin((const user_addr_t)uaddr, (char *)&ret, sizeof(ret))) { |
| 1176 | DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); |
| 1177 | cpu_core[CPU->cpu_id].cpuc_dtrace_illval = uaddr; |
| 1178 | } |
| 1179 | dtrace_copyio_postflight(uaddr); |
| 1180 | } |
| 1181 | DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); |
| 1182 | |
| 1183 | return(ret); |
| 1184 | } |
| 1185 | |
| 1186 | uint32_t |
| 1187 | dtrace_fuword32(user_addr_t uaddr) |
| 1188 | { |
| 1189 | uint32_t ret = 0; |
| 1190 | |
| 1191 | DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); |
| 1192 | if (dtrace_copycheck( uaddr, (uintptr_t)&ret, sizeof(ret))) { |
| 1193 | if (copyin((const user_addr_t)uaddr, (char *)&ret, sizeof(ret))) { |
| 1194 | DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); |
| 1195 | cpu_core[CPU->cpu_id].cpuc_dtrace_illval = uaddr; |
| 1196 | } |
| 1197 | dtrace_copyio_postflight(uaddr); |
| 1198 | } |
| 1199 | DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); |
| 1200 | |
| 1201 | return(ret); |
| 1202 | } |
| 1203 | |
| 1204 | uint64_t |
| 1205 | dtrace_fuword64(user_addr_t uaddr) |
| 1206 | { |
| 1207 | uint64_t ret = 0; |
| 1208 | |
| 1209 | DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); |
| 1210 | if (dtrace_copycheck( uaddr, (uintptr_t)&ret, sizeof(ret))) { |
| 1211 | if (copyin((const user_addr_t)uaddr, (char *)&ret, sizeof(ret))) { |
| 1212 | DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); |
| 1213 | cpu_core[CPU->cpu_id].cpuc_dtrace_illval = uaddr; |
| 1214 | } |
| 1215 | dtrace_copyio_postflight(uaddr); |
| 1216 | } |
| 1217 | DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); |
| 1218 | |
| 1219 | return(ret); |
| 1220 | } |
| 1221 | |
| 1222 | /* |
| 1223 | * Emulation of Solaris fuword / suword |
| 1224 | * Called from the fasttrap provider, so the use of copyin/out requires fewer safegaurds. |
| 1225 | */ |
| 1226 | |
| 1227 | int |
| 1228 | fuword8(user_addr_t uaddr, uint8_t *value) |
| 1229 | { |
| 1230 | if (copyin((const user_addr_t)uaddr, (char *)value, sizeof(uint8_t)) != 0) { |
| 1231 | return -1; |
| 1232 | } |
| 1233 | |
| 1234 | return 0; |
| 1235 | } |
| 1236 | |
| 1237 | int |
| 1238 | fuword16(user_addr_t uaddr, uint16_t *value) |
| 1239 | { |
| 1240 | if (copyin((const user_addr_t)uaddr, (char *)value, sizeof(uint16_t)) != 0) { |
| 1241 | return -1; |
| 1242 | } |
| 1243 | |
| 1244 | return 0; |
| 1245 | } |
| 1246 | |
| 1247 | int |
| 1248 | fuword32(user_addr_t uaddr, uint32_t *value) |
| 1249 | { |
| 1250 | if (copyin((const user_addr_t)uaddr, (char *)value, sizeof(uint32_t)) != 0) { |
| 1251 | return -1; |
| 1252 | } |
| 1253 | |
| 1254 | return 0; |
| 1255 | } |
| 1256 | |
| 1257 | int |
| 1258 | fuword64(user_addr_t uaddr, uint64_t *value) |
| 1259 | { |
| 1260 | if (copyin((const user_addr_t)uaddr, (char *)value, sizeof(uint64_t)) != 0) { |
| 1261 | return -1; |
| 1262 | } |
| 1263 | |
| 1264 | return 0; |
| 1265 | } |
| 1266 | |
| 1267 | void |
| 1268 | fuword32_noerr(user_addr_t uaddr, uint32_t *value) |
| 1269 | { |
| 1270 | if (copyin((const user_addr_t)uaddr, (char *)value, sizeof(uint32_t))) { |
| 1271 | *value = 0; |
| 1272 | } |
| 1273 | } |
| 1274 | |
| 1275 | void |
| 1276 | fuword64_noerr(user_addr_t uaddr, uint64_t *value) |
| 1277 | { |
| 1278 | if (copyin((const user_addr_t)uaddr, (char *)value, sizeof(uint64_t))) { |
| 1279 | *value = 0; |
| 1280 | } |
| 1281 | } |
| 1282 | |
| 1283 | int |
| 1284 | suword64(user_addr_t addr, uint64_t value) |
| 1285 | { |
| 1286 | if (copyout((const void *)&value, addr, sizeof(value)) != 0) { |
| 1287 | return -1; |
| 1288 | } |
| 1289 | |
| 1290 | return 0; |
| 1291 | } |
| 1292 | |
| 1293 | int |
| 1294 | suword32(user_addr_t addr, uint32_t value) |
| 1295 | { |
| 1296 | if (copyout((const void *)&value, addr, sizeof(value)) != 0) { |
| 1297 | return -1; |
| 1298 | } |
| 1299 | |
| 1300 | return 0; |
| 1301 | } |
| 1302 | |
| 1303 | /* |
| 1304 | * Miscellaneous |
| 1305 | */ |
| 1306 | extern boolean_t dtrace_tally_fault(user_addr_t); |
| 1307 | |
| 1308 | boolean_t |
| 1309 | dtrace_tally_fault(user_addr_t uaddr) |
| 1310 | { |
| 1311 | DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); |
| 1312 | cpu_core[CPU->cpu_id].cpuc_dtrace_illval = uaddr; |
| 1313 | return( DTRACE_CPUFLAG_ISSET(CPU_DTRACE_NOFAULT) ? TRUE : FALSE ); |
| 1314 | } |
| 1315 | |
| 1316 | #define TOTTY 0x02 |
| 1317 | extern int prf(const char *, va_list, int, struct tty *); /* bsd/kern/subr_prf.h */ |
| 1318 | |
| 1319 | int |
| 1320 | vuprintf(const char *format, va_list ap) |
| 1321 | { |
| 1322 | return prf(format, ap, TOTTY, NULL); |
| 1323 | } |
| 1324 | |
| 1325 | /* Not called from probe context */ |
| 1326 | void cmn_err( int level, const char *format, ... ) |
| 1327 | { |
| 1328 | #pragma unused(level) |
| 1329 | va_list alist; |
| 1330 | |
| 1331 | va_start(alist, format); |
| 1332 | vuprintf(format, alist); |
| 1333 | va_end(alist); |
| 1334 | uprintf("\n" ); |
| 1335 | } |
| 1336 | |
| 1337 | /* |
| 1338 | * History: |
| 1339 | * 2002-01-24 gvdl Initial implementation of strstr |
| 1340 | */ |
| 1341 | |
| 1342 | __private_extern__ const char * |
| 1343 | strstr(const char *in, const char *str) |
| 1344 | { |
| 1345 | char c; |
| 1346 | size_t len; |
| 1347 | if (!in || !str) |
| 1348 | return in; |
| 1349 | |
| 1350 | c = *str++; |
| 1351 | if (!c) |
| 1352 | return (const char *) in; // Trivial empty string case |
| 1353 | |
| 1354 | len = strlen(str); |
| 1355 | do { |
| 1356 | char sc; |
| 1357 | |
| 1358 | do { |
| 1359 | sc = *in++; |
| 1360 | if (!sc) |
| 1361 | return (char *) 0; |
| 1362 | } while (sc != c); |
| 1363 | } while (strncmp(in, str, len) != 0); |
| 1364 | |
| 1365 | return (const char *) (in - 1); |
| 1366 | } |
| 1367 | |
| 1368 | const void* |
| 1369 | bsearch(const void *key, const void *base0, size_t nmemb, size_t size, int (*compar)(const void *, const void *)) |
| 1370 | { |
| 1371 | const char *base = base0; |
| 1372 | size_t lim; |
| 1373 | int cmp; |
| 1374 | const void *p; |
| 1375 | for (lim = nmemb; lim != 0; lim >>= 1) { |
| 1376 | p = base + (lim >> 1) * size; |
| 1377 | cmp = (*compar)(key, p); |
| 1378 | if (cmp == 0) |
| 1379 | return p; |
| 1380 | if (cmp > 0) { /* key > p: move right */ |
| 1381 | base = (const char *)p + size; |
| 1382 | lim--; |
| 1383 | } /* else move left */ |
| 1384 | } |
| 1385 | return (NULL); |
| 1386 | } |
| 1387 | |
| 1388 | /* |
| 1389 | * Runtime and ABI |
| 1390 | */ |
| 1391 | uintptr_t |
| 1392 | dtrace_caller(int ignore) |
| 1393 | { |
| 1394 | #pragma unused(ignore) |
| 1395 | return -1; /* Just as in Solaris dtrace_asm.s */ |
| 1396 | } |
| 1397 | |
| 1398 | int |
| 1399 | dtrace_getstackdepth(int aframes) |
| 1400 | { |
| 1401 | struct frame *fp = (struct frame *)__builtin_frame_address(0); |
| 1402 | struct frame *nextfp, *minfp, *stacktop; |
| 1403 | int depth = 0; |
| 1404 | int on_intr; |
| 1405 | |
| 1406 | if ((on_intr = CPU_ON_INTR(CPU)) != 0) |
| 1407 | stacktop = (struct frame *)dtrace_get_cpu_int_stack_top(); |
| 1408 | else |
| 1409 | stacktop = (struct frame *)(dtrace_get_kernel_stack(current_thread()) + kernel_stack_size); |
| 1410 | |
| 1411 | minfp = fp; |
| 1412 | |
| 1413 | aframes++; |
| 1414 | |
| 1415 | for (;;) { |
| 1416 | depth++; |
| 1417 | |
| 1418 | nextfp = *(struct frame **)fp; |
| 1419 | |
| 1420 | if (nextfp <= minfp || nextfp >= stacktop) { |
| 1421 | if (on_intr) { |
| 1422 | /* |
| 1423 | * Hop from interrupt stack to thread stack. |
| 1424 | */ |
| 1425 | vm_offset_t kstack_base = dtrace_get_kernel_stack(current_thread()); |
| 1426 | |
| 1427 | minfp = (struct frame *)kstack_base; |
| 1428 | stacktop = (struct frame *)(kstack_base + kernel_stack_size); |
| 1429 | |
| 1430 | on_intr = 0; |
| 1431 | continue; |
| 1432 | } |
| 1433 | break; |
| 1434 | } |
| 1435 | |
| 1436 | fp = nextfp; |
| 1437 | minfp = fp; |
| 1438 | } |
| 1439 | |
| 1440 | if (depth <= aframes) |
| 1441 | return (0); |
| 1442 | |
| 1443 | return (depth - aframes); |
| 1444 | } |
| 1445 | |
| 1446 | int |
| 1447 | dtrace_addr_in_module(void* addr, struct modctl *ctl) |
| 1448 | { |
| 1449 | return OSKextKextForAddress(addr) == (void*)ctl->mod_address; |
| 1450 | } |
| 1451 | |
| 1452 | /* |
| 1453 | * Unconsidered |
| 1454 | */ |
| 1455 | void |
| 1456 | dtrace_vtime_enable(void) {} |
| 1457 | |
| 1458 | void |
| 1459 | dtrace_vtime_disable(void) {} |
| 1460 | |
| 1461 | #else /* else ! CONFIG_DTRACE */ |
| 1462 | |
| 1463 | #include <sys/types.h> |
| 1464 | #include <mach/vm_types.h> |
| 1465 | #include <mach/kmod.h> |
| 1466 | |
| 1467 | /* |
| 1468 | * This exists to prevent build errors when dtrace is unconfigured. |
| 1469 | */ |
| 1470 | |
| 1471 | kern_return_t _dtrace_register_anon_DOF(char *, unsigned char *, uint32_t); |
| 1472 | |
| 1473 | kern_return_t _dtrace_register_anon_DOF(char *arg1, unsigned char *arg2, uint32_t arg3) { |
| 1474 | #pragma unused(arg1, arg2, arg3) |
| 1475 | |
| 1476 | return KERN_FAILURE; |
| 1477 | } |
| 1478 | |
| 1479 | #endif /* CONFIG_DTRACE */ |
| 1480 | |