1 | /* |
2 | * Copyright (c) 2013-2017 Apple 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 | #include <mach/mach_types.h> |
30 | #include <mach/vm_param.h> |
31 | #include <mach/mach_vm.h> |
32 | #include <mach/clock_types.h> |
33 | #include <sys/errno.h> |
34 | #include <sys/stackshot.h> |
35 | #ifdef IMPORTANCE_INHERITANCE |
36 | #include <ipc/ipc_importance.h> |
37 | #endif |
38 | #include <sys/appleapiopts.h> |
39 | #include <kern/debug.h> |
40 | #include <kern/block_hint.h> |
41 | #include <uuid/uuid.h> |
42 | |
43 | #include <kdp/kdp_dyld.h> |
44 | #include <kdp/kdp_en_debugger.h> |
45 | |
46 | #include <libsa/types.h> |
47 | #include <libkern/version.h> |
48 | #include <libkern/section_keywords.h> |
49 | |
50 | #include <string.h> /* bcopy */ |
51 | |
52 | #include <kern/coalition.h> |
53 | #include <kern/processor.h> |
54 | #include <kern/thread.h> |
55 | #include <kern/thread_group.h> |
56 | #include <kern/task.h> |
57 | #include <kern/telemetry.h> |
58 | #include <kern/clock.h> |
59 | #include <kern/policy_internal.h> |
60 | #include <vm/vm_map.h> |
61 | #include <vm/vm_kern.h> |
62 | #include <vm/vm_pageout.h> |
63 | #include <vm/vm_fault.h> |
64 | #include <vm/vm_shared_region.h> |
65 | #include <libkern/OSKextLibPrivate.h> |
66 | |
67 | #if defined(__x86_64__) |
68 | #include <i386/mp.h> |
69 | #include <i386/cpu_threads.h> |
70 | #endif |
71 | |
72 | #if CONFIG_EMBEDDED |
73 | #include <pexpert/pexpert.h> /* For gPanicBase/gPanicBase */ |
74 | #endif |
75 | |
76 | #if MONOTONIC |
77 | #include <kern/monotonic.h> |
78 | #endif /* MONOTONIC */ |
79 | |
80 | #include <san/kasan.h> |
81 | |
82 | extern unsigned int not_in_kdp; |
83 | |
84 | |
85 | /* indicate to the compiler that some accesses are unaligned */ |
86 | typedef uint64_t unaligned_u64 __attribute__((aligned(1))); |
87 | |
88 | extern addr64_t kdp_vtophys(pmap_t pmap, addr64_t va); |
89 | |
90 | int kdp_snapshot = 0; |
91 | static kern_return_t stack_snapshot_ret = 0; |
92 | static uint32_t stack_snapshot_bytes_traced = 0; |
93 | |
94 | static kcdata_descriptor_t stackshot_kcdata_p = NULL; |
95 | static void *stack_snapshot_buf; |
96 | static uint32_t stack_snapshot_bufsize; |
97 | int stack_snapshot_pid; |
98 | static uint32_t stack_snapshot_flags; |
99 | static uint64_t stack_snapshot_delta_since_timestamp; |
100 | static boolean_t panic_stackshot; |
101 | |
102 | static boolean_t stack_enable_faulting = FALSE; |
103 | static struct stackshot_fault_stats fault_stats; |
104 | |
105 | static unaligned_u64 * stackshot_duration_outer; |
106 | static uint64_t stackshot_microsecs; |
107 | |
108 | void * kernel_stackshot_buf = NULL; /* Pointer to buffer for stackshots triggered from the kernel and retrieved later */ |
109 | int kernel_stackshot_buf_size = 0; |
110 | |
111 | void * stackshot_snapbuf = NULL; /* Used by stack_snapshot2 (to be removed) */ |
112 | |
113 | __private_extern__ void stackshot_init( void ); |
114 | static boolean_t memory_iszero(void *addr, size_t size); |
115 | #if CONFIG_TELEMETRY |
116 | kern_return_t stack_microstackshot(user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t flags, int32_t *retval); |
117 | #endif |
118 | uint32_t get_stackshot_estsize(uint32_t prev_size_hint); |
119 | kern_return_t kern_stack_snapshot_internal(int stackshot_config_version, void *stackshot_config, |
120 | size_t stackshot_config_size, boolean_t stackshot_from_user); |
121 | kern_return_t do_stackshot(void *); |
122 | void kdp_snapshot_preflight(int pid, void * tracebuf, uint32_t tracebuf_size, uint32_t flags, kcdata_descriptor_t data_p, uint64_t since_timestamp); |
123 | boolean_t stackshot_thread_is_idle_worker_unsafe(thread_t thread); |
124 | static int kdp_stackshot_kcdata_format(int pid, uint32_t trace_flags, uint32_t *pBytesTraced); |
125 | uint32_t kdp_stack_snapshot_bytes_traced(void); |
126 | static void kdp_mem_and_io_snapshot(struct mem_and_io_snapshot *memio_snap); |
127 | static boolean_t kdp_copyin(vm_map_t map, uint64_t uaddr, void *dest, size_t size, boolean_t try_fault, uint32_t *kdp_fault_result); |
128 | static boolean_t kdp_copyin_word(task_t task, uint64_t addr, uint64_t *result, boolean_t try_fault, uint32_t *kdp_fault_results); |
129 | static uint64_t proc_was_throttled_from_task(task_t task); |
130 | static void stackshot_thread_wait_owner_info(thread_t thread, thread_waitinfo_t * waitinfo); |
131 | static int stackshot_thread_has_valid_waitinfo(thread_t thread); |
132 | |
133 | #if CONFIG_COALITIONS |
134 | static void stackshot_coalition_jetsam_count(void *arg, int i, coalition_t coal); |
135 | static void stackshot_coalition_jetsam_snapshot(void *arg, int i, coalition_t coal); |
136 | #endif /* CONFIG_COALITIONS */ |
137 | |
138 | |
139 | extern uint32_t workqueue_get_pwq_state_kdp(void *proc); |
140 | |
141 | extern int proc_pid(void *p); |
142 | extern uint64_t proc_uniqueid(void *p); |
143 | extern uint64_t proc_was_throttled(void *p); |
144 | extern uint64_t proc_did_throttle(void *p); |
145 | extern int proc_exiting(void *p); |
146 | extern int proc_in_teardown(void *p); |
147 | static uint64_t proc_did_throttle_from_task(task_t task); |
148 | extern void proc_name_kdp(task_t task, char * buf, int size); |
149 | extern int proc_threadname_kdp(void * uth, char * buf, size_t size); |
150 | extern void proc_starttime_kdp(void * p, uint64_t * tv_sec, uint64_t * tv_usec, uint64_t * abstime); |
151 | extern int memorystatus_get_pressure_status_kdp(void); |
152 | extern void memorystatus_proc_flags_unsafe(void * v, boolean_t *is_dirty, boolean_t *is_dirty_tracked, boolean_t *allow_idle_exit); |
153 | |
154 | extern int count_busy_buffers(void); /* must track with declaration in bsd/sys/buf_internal.h */ |
155 | extern void bcopy_phys(addr64_t, addr64_t, vm_size_t); |
156 | |
157 | #if CONFIG_TELEMETRY |
158 | extern kern_return_t stack_microstackshot(user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t flags, int32_t *retval); |
159 | #endif /* CONFIG_TELEMETRY */ |
160 | |
161 | extern kern_return_t kern_stack_snapshot_with_reason(char* reason); |
162 | extern kern_return_t kern_stack_snapshot_internal(int stackshot_config_version, void *stackshot_config, size_t stackshot_config_size, boolean_t stackshot_from_user); |
163 | |
164 | /* |
165 | * Validates that the given address is both a valid page and has |
166 | * default caching attributes for the current map. Returns |
167 | * 0 if the address is invalid, and a kernel virtual address for |
168 | * the given address if it is valid. |
169 | */ |
170 | vm_offset_t machine_trace_thread_get_kva(vm_offset_t cur_target_addr, vm_map_t map, uint32_t *thread_trace_flags); |
171 | |
172 | #define KDP_FAULT_RESULT_PAGED_OUT 0x1 /* some data was unable to be retrieved */ |
173 | #define KDP_FAULT_RESULT_TRIED_FAULT 0x2 /* tried to fault in data */ |
174 | #define KDP_FAULT_RESULT_FAULTED_IN 0x4 /* successfully faulted in data */ |
175 | |
176 | /* |
177 | * Looks up the physical translation for the given address in the target map, attempting |
178 | * to fault data in if requested and it is not resident. Populates thread_trace_flags if requested |
179 | * as well. |
180 | */ |
181 | vm_offset_t kdp_find_phys(vm_map_t map, vm_offset_t target_addr, boolean_t try_fault, uint32_t *kdp_fault_results); |
182 | |
183 | static size_t stackshot_strlcpy(char *dst, const char *src, size_t maxlen); |
184 | static void stackshot_memcpy(void *dst, const void *src, size_t len); |
185 | |
186 | /* Clears caching information used by the above validation routine |
187 | * (in case the current map has been changed or cleared). |
188 | */ |
189 | void machine_trace_thread_clear_validation_cache(void); |
190 | |
191 | #define MAX_FRAMES 1000 |
192 | #define MAX_LOADINFOS 500 |
193 | #define TASK_IMP_WALK_LIMIT 20 |
194 | |
195 | typedef struct thread_snapshot *thread_snapshot_t; |
196 | typedef struct task_snapshot *task_snapshot_t; |
197 | |
198 | #if CONFIG_KDP_INTERACTIVE_DEBUGGING |
199 | extern kdp_send_t kdp_en_send_pkt; |
200 | #endif |
201 | |
202 | /* |
203 | * Globals to support machine_trace_thread_get_kva. |
204 | */ |
205 | static vm_offset_t prev_target_page = 0; |
206 | static vm_offset_t prev_target_kva = 0; |
207 | static boolean_t validate_next_addr = TRUE; |
208 | |
209 | /* |
210 | * Stackshot locking and other defines. |
211 | */ |
212 | static lck_grp_t *stackshot_subsys_lck_grp; |
213 | static lck_grp_attr_t *stackshot_subsys_lck_grp_attr; |
214 | static lck_attr_t *stackshot_subsys_lck_attr; |
215 | static lck_mtx_t stackshot_subsys_mutex; |
216 | |
217 | #define STACKSHOT_SUBSYS_LOCK() lck_mtx_lock(&stackshot_subsys_mutex) |
218 | #define STACKSHOT_SUBSYS_TRY_LOCK() lck_mtx_try_lock(&stackshot_subsys_mutex) |
219 | #define STACKSHOT_SUBSYS_UNLOCK() lck_mtx_unlock(&stackshot_subsys_mutex) |
220 | |
221 | #define SANE_BOOTPROFILE_TRACEBUF_SIZE (64 * 1024 * 1024) |
222 | #define SANE_TRACEBUF_SIZE (8 * 1024 * 1024) |
223 | |
224 | SECURITY_READ_ONLY_LATE(static uint32_t) max_tracebuf_size = SANE_TRACEBUF_SIZE; |
225 | |
226 | /* |
227 | * We currently set a ceiling of 3 milliseconds spent in the kdp fault path |
228 | * for non-panic stackshots where faulting is requested. |
229 | */ |
230 | #define KDP_FAULT_PATH_MAX_TIME_PER_STACKSHOT_NSECS (3 * NSEC_PER_MSEC) |
231 | |
232 | #define STACKSHOT_SUPP_SIZE (16 * 1024) /* Minimum stackshot size */ |
233 | #define TASK_UUID_AVG_SIZE (16 * sizeof(uuid_t)) /* Average space consumed by UUIDs/task */ |
234 | |
235 | /* |
236 | * Initialize the mutex governing access to the stack snapshot subsystem |
237 | * and other stackshot related bits. |
238 | */ |
239 | __private_extern__ void |
240 | stackshot_init( void ) |
241 | { |
242 | mach_timebase_info_data_t timebase; |
243 | |
244 | stackshot_subsys_lck_grp_attr = lck_grp_attr_alloc_init(); |
245 | |
246 | stackshot_subsys_lck_grp = lck_grp_alloc_init("stackshot_subsys_lock" , stackshot_subsys_lck_grp_attr); |
247 | |
248 | stackshot_subsys_lck_attr = lck_attr_alloc_init(); |
249 | |
250 | lck_mtx_init(&stackshot_subsys_mutex, stackshot_subsys_lck_grp, stackshot_subsys_lck_attr); |
251 | |
252 | clock_timebase_info(&timebase); |
253 | fault_stats.sfs_system_max_fault_time = ((KDP_FAULT_PATH_MAX_TIME_PER_STACKSHOT_NSECS * timebase.denom)/ timebase.numer); |
254 | |
255 | PE_parse_boot_argn("stackshot_maxsz" , &max_tracebuf_size, sizeof(max_tracebuf_size)); |
256 | } |
257 | |
258 | /* |
259 | * Method for grabbing timer values safely, in the sense that no infinite loop will occur |
260 | * Certain flavors of the timer_grab function, which would seem to be the thing to use, |
261 | * can loop infinitely if called while the timer is in the process of being updated. |
262 | * Unfortunately, it is (rarely) possible to get inconsistent top and bottom halves of |
263 | * the timer using this method. This seems insoluble, since stackshot runs in a context |
264 | * where the timer might be half-updated, and has no way of yielding control just long |
265 | * enough to finish the update. |
266 | */ |
267 | |
268 | static uint64_t safe_grab_timer_value(struct timer *t) |
269 | { |
270 | #if defined(__LP64__) |
271 | return t->all_bits; |
272 | #else |
273 | uint64_t time = t->high_bits; /* endian independent grab */ |
274 | time = (time << 32) | t->low_bits; |
275 | return time; |
276 | #endif |
277 | } |
278 | |
279 | /* |
280 | * Called with interrupts disabled after stackshot context has been |
281 | * initialized. Updates stack_snapshot_ret. |
282 | */ |
283 | static kern_return_t |
284 | stackshot_trap() |
285 | { |
286 | kern_return_t rv; |
287 | |
288 | #if defined(__x86_64__) |
289 | /* |
290 | * Since mp_rendezvous and stackshot both attempt to capture cpus then perform an |
291 | * operation, it's essential to apply mutual exclusion to the other when one |
292 | * mechanism is in operation, lest there be a deadlock as the mechanisms race to |
293 | * capture CPUs. |
294 | * |
295 | * Further, we assert that invoking stackshot from mp_rendezvous*() is not |
296 | * allowed, so we check to ensure there there is no rendezvous in progress before |
297 | * trying to grab the lock (if there is, a deadlock will occur when we try to |
298 | * grab the lock). This is accomplished by setting cpu_rendezvous_in_progress to |
299 | * TRUE in the mp rendezvous action function. If stackshot_trap() is called by |
300 | * a subordinate of the call chain within the mp rendezvous action, this flag will |
301 | * be set and can be used to detect the inevitable deadlock that would occur |
302 | * if this thread tried to grab the rendezvous lock. |
303 | */ |
304 | |
305 | if (current_cpu_datap()->cpu_rendezvous_in_progress == TRUE) { |
306 | panic("Calling stackshot from a rendezvous is not allowed!" ); |
307 | } |
308 | |
309 | mp_rendezvous_lock(); |
310 | #endif |
311 | |
312 | rv = DebuggerTrapWithState(DBOP_STACKSHOT, NULL, NULL, NULL, 0, NULL, FALSE, 0); |
313 | |
314 | #if defined(__x86_64__) |
315 | mp_rendezvous_unlock(); |
316 | #endif |
317 | return (rv); |
318 | } |
319 | |
320 | |
321 | kern_return_t |
322 | stack_snapshot_from_kernel(int pid, void *buf, uint32_t size, uint32_t flags, uint64_t delta_since_timestamp, unsigned *bytes_traced) |
323 | { |
324 | kern_return_t error = KERN_SUCCESS; |
325 | boolean_t istate; |
326 | |
327 | #if DEVELOPMENT || DEBUG |
328 | if (kern_feature_override(KF_STACKSHOT_OVRD) == TRUE) { |
329 | error = KERN_NOT_SUPPORTED; |
330 | goto out; |
331 | } |
332 | #endif |
333 | if ((buf == NULL) || (size <= 0) || (bytes_traced == NULL)) { |
334 | return KERN_INVALID_ARGUMENT; |
335 | } |
336 | |
337 | /* cap in individual stackshot to max_tracebuf_size */ |
338 | if (size > max_tracebuf_size) { |
339 | size = max_tracebuf_size; |
340 | } |
341 | |
342 | /* Serialize tracing */ |
343 | if (flags & STACKSHOT_TRYLOCK) { |
344 | if (!STACKSHOT_SUBSYS_TRY_LOCK()) { |
345 | return KERN_LOCK_OWNED; |
346 | } |
347 | } else { |
348 | STACKSHOT_SUBSYS_LOCK(); |
349 | } |
350 | |
351 | struct kcdata_descriptor kcdata; |
352 | uint32_t hdr_tag = (flags & STACKSHOT_COLLECT_DELTA_SNAPSHOT) ? |
353 | KCDATA_BUFFER_BEGIN_DELTA_STACKSHOT : KCDATA_BUFFER_BEGIN_STACKSHOT; |
354 | |
355 | error = kcdata_memory_static_init(&kcdata, (mach_vm_address_t)buf, hdr_tag, size, |
356 | KCFLAG_USE_MEMCOPY | KCFLAG_NO_AUTO_ENDBUFFER); |
357 | if (error) { |
358 | goto out; |
359 | } |
360 | |
361 | istate = ml_set_interrupts_enabled(FALSE); |
362 | |
363 | /* Preload trace parameters*/ |
364 | kdp_snapshot_preflight(pid, buf, size, flags, &kcdata, delta_since_timestamp); |
365 | |
366 | /* |
367 | * Trap to the debugger to obtain a coherent stack snapshot; this populates |
368 | * the trace buffer |
369 | */ |
370 | error = stackshot_trap(); |
371 | |
372 | ml_set_interrupts_enabled(istate); |
373 | |
374 | *bytes_traced = kdp_stack_snapshot_bytes_traced(); |
375 | |
376 | out: |
377 | stackshot_kcdata_p = NULL; |
378 | STACKSHOT_SUBSYS_UNLOCK(); |
379 | return error; |
380 | } |
381 | |
382 | #if CONFIG_TELEMETRY |
383 | kern_return_t |
384 | stack_microstackshot(user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t flags, int32_t *retval) |
385 | { |
386 | int error = KERN_SUCCESS; |
387 | uint32_t bytes_traced = 0; |
388 | |
389 | *retval = -1; |
390 | |
391 | /* |
392 | * Control related operations |
393 | */ |
394 | if (flags & STACKSHOT_GLOBAL_MICROSTACKSHOT_ENABLE) { |
395 | telemetry_global_ctl(1); |
396 | *retval = 0; |
397 | goto exit; |
398 | } else if (flags & STACKSHOT_GLOBAL_MICROSTACKSHOT_DISABLE) { |
399 | telemetry_global_ctl(0); |
400 | *retval = 0; |
401 | goto exit; |
402 | } |
403 | |
404 | /* |
405 | * Data related operations |
406 | */ |
407 | *retval = -1; |
408 | |
409 | if ((((void*)tracebuf) == NULL) || (tracebuf_size == 0)) { |
410 | error = KERN_INVALID_ARGUMENT; |
411 | goto exit; |
412 | } |
413 | |
414 | STACKSHOT_SUBSYS_LOCK(); |
415 | |
416 | if (flags & STACKSHOT_GET_MICROSTACKSHOT) { |
417 | if (tracebuf_size > max_tracebuf_size) { |
418 | error = KERN_INVALID_ARGUMENT; |
419 | goto unlock_exit; |
420 | } |
421 | |
422 | bytes_traced = tracebuf_size; |
423 | error = telemetry_gather(tracebuf, &bytes_traced, |
424 | (flags & STACKSHOT_SET_MICROSTACKSHOT_MARK) ? TRUE : FALSE); |
425 | *retval = (int)bytes_traced; |
426 | goto unlock_exit; |
427 | } |
428 | |
429 | if (flags & STACKSHOT_GET_BOOT_PROFILE) { |
430 | |
431 | if (tracebuf_size > SANE_BOOTPROFILE_TRACEBUF_SIZE) { |
432 | error = KERN_INVALID_ARGUMENT; |
433 | goto unlock_exit; |
434 | } |
435 | |
436 | bytes_traced = tracebuf_size; |
437 | error = bootprofile_gather(tracebuf, &bytes_traced); |
438 | *retval = (int)bytes_traced; |
439 | } |
440 | |
441 | unlock_exit: |
442 | STACKSHOT_SUBSYS_UNLOCK(); |
443 | exit: |
444 | return error; |
445 | } |
446 | #endif /* CONFIG_TELEMETRY */ |
447 | |
448 | /* |
449 | * Return the estimated size of a stackshot based on the |
450 | * number of currently running threads and tasks. |
451 | */ |
452 | uint32_t |
453 | get_stackshot_estsize(uint32_t prev_size_hint) |
454 | { |
455 | vm_size_t thread_total; |
456 | vm_size_t task_total; |
457 | uint32_t estimated_size; |
458 | |
459 | thread_total = (threads_count * sizeof(struct thread_snapshot)); |
460 | task_total = (tasks_count * (sizeof(struct task_snapshot) + TASK_UUID_AVG_SIZE)); |
461 | |
462 | estimated_size = (uint32_t) VM_MAP_ROUND_PAGE((thread_total + task_total + STACKSHOT_SUPP_SIZE), PAGE_MASK); |
463 | if (estimated_size < prev_size_hint) { |
464 | estimated_size = (uint32_t) VM_MAP_ROUND_PAGE(prev_size_hint, PAGE_MASK); |
465 | } |
466 | |
467 | return estimated_size; |
468 | } |
469 | |
470 | /* |
471 | * stackshot_remap_buffer: Utility function to remap bytes_traced bytes starting at stackshotbuf |
472 | * into the current task's user space and subsequently copy out the address |
473 | * at which the buffer has been mapped in user space to out_buffer_addr. |
474 | * |
475 | * Inputs: stackshotbuf - pointer to the original buffer in the kernel's address space |
476 | * bytes_traced - length of the buffer to remap starting from stackshotbuf |
477 | * out_buffer_addr - pointer to placeholder where newly mapped buffer will be mapped. |
478 | * out_size_addr - pointer to be filled in with the size of the buffer |
479 | * |
480 | * Outputs: ENOSPC if there is not enough free space in the task's address space to remap the buffer |
481 | * EINVAL for all other errors returned by task_remap_buffer/mach_vm_remap |
482 | * an error from copyout |
483 | */ |
484 | static kern_return_t |
485 | stackshot_remap_buffer(void *stackshotbuf, uint32_t bytes_traced, uint64_t out_buffer_addr, uint64_t out_size_addr) |
486 | { |
487 | int error = 0; |
488 | mach_vm_offset_t stackshotbuf_user_addr = (mach_vm_offset_t)NULL; |
489 | vm_prot_t cur_prot, max_prot; |
490 | |
491 | error = mach_vm_remap_kernel(get_task_map(current_task()), &stackshotbuf_user_addr, bytes_traced, 0, |
492 | VM_FLAGS_ANYWHERE, VM_KERN_MEMORY_NONE, kernel_map, (mach_vm_offset_t)stackshotbuf, FALSE, &cur_prot, &max_prot, VM_INHERIT_DEFAULT); |
493 | /* |
494 | * If the call to mach_vm_remap fails, we return the appropriate converted error |
495 | */ |
496 | if (error == KERN_SUCCESS) { |
497 | /* |
498 | * If we fail to copy out the address or size of the new buffer, we remove the buffer mapping that |
499 | * we just made in the task's user space. |
500 | */ |
501 | error = copyout(CAST_DOWN(void *, &stackshotbuf_user_addr), (user_addr_t)out_buffer_addr, sizeof(stackshotbuf_user_addr)); |
502 | if (error != KERN_SUCCESS) { |
503 | mach_vm_deallocate(get_task_map(current_task()), stackshotbuf_user_addr, (mach_vm_size_t)bytes_traced); |
504 | return error; |
505 | } |
506 | error = copyout(&bytes_traced, (user_addr_t)out_size_addr, sizeof(bytes_traced)); |
507 | if (error != KERN_SUCCESS) { |
508 | mach_vm_deallocate(get_task_map(current_task()), stackshotbuf_user_addr, (mach_vm_size_t)bytes_traced); |
509 | return error; |
510 | } |
511 | } |
512 | return error; |
513 | } |
514 | |
515 | kern_return_t |
516 | kern_stack_snapshot_internal(int stackshot_config_version, void *stackshot_config, size_t stackshot_config_size, boolean_t stackshot_from_user) |
517 | { |
518 | int error = 0; |
519 | boolean_t prev_interrupt_state; |
520 | uint32_t bytes_traced = 0; |
521 | uint32_t stackshotbuf_size = 0; |
522 | void * stackshotbuf = NULL; |
523 | kcdata_descriptor_t kcdata_p = NULL; |
524 | |
525 | void * buf_to_free = NULL; |
526 | int size_to_free = 0; |
527 | |
528 | /* Parsed arguments */ |
529 | uint64_t out_buffer_addr; |
530 | uint64_t out_size_addr; |
531 | int pid = -1; |
532 | uint32_t flags; |
533 | uint64_t since_timestamp; |
534 | uint32_t size_hint = 0; |
535 | |
536 | if(stackshot_config == NULL) { |
537 | return KERN_INVALID_ARGUMENT; |
538 | } |
539 | #if DEVELOPMENT || DEBUG |
540 | /* TBD: ask stackshot clients to avoid issuing stackshots in this |
541 | * configuration in lieu of the kernel feature override. |
542 | */ |
543 | if (kern_feature_override(KF_STACKSHOT_OVRD) == TRUE) { |
544 | return KERN_NOT_SUPPORTED; |
545 | } |
546 | #endif |
547 | |
548 | switch (stackshot_config_version) { |
549 | case STACKSHOT_CONFIG_TYPE: |
550 | if (stackshot_config_size != sizeof(stackshot_config_t)) { |
551 | return KERN_INVALID_ARGUMENT; |
552 | } |
553 | stackshot_config_t *config = (stackshot_config_t *) stackshot_config; |
554 | out_buffer_addr = config->sc_out_buffer_addr; |
555 | out_size_addr = config->sc_out_size_addr; |
556 | pid = config->sc_pid; |
557 | flags = config->sc_flags; |
558 | since_timestamp = config->sc_delta_timestamp; |
559 | if (config->sc_size <= max_tracebuf_size) { |
560 | size_hint = config->sc_size; |
561 | } |
562 | break; |
563 | default: |
564 | return KERN_NOT_SUPPORTED; |
565 | } |
566 | |
567 | /* |
568 | * Currently saving a kernel buffer and trylock are only supported from the |
569 | * internal/KEXT API. |
570 | */ |
571 | if (stackshot_from_user) { |
572 | if (flags & (STACKSHOT_TRYLOCK | STACKSHOT_SAVE_IN_KERNEL_BUFFER | STACKSHOT_FROM_PANIC)) { |
573 | return KERN_NO_ACCESS; |
574 | } |
575 | } else { |
576 | if (!(flags & STACKSHOT_SAVE_IN_KERNEL_BUFFER)) { |
577 | return KERN_NOT_SUPPORTED; |
578 | } |
579 | } |
580 | |
581 | if (!((flags & STACKSHOT_KCDATA_FORMAT) || (flags & STACKSHOT_RETRIEVE_EXISTING_BUFFER))) { |
582 | return KERN_NOT_SUPPORTED; |
583 | } |
584 | |
585 | /* |
586 | * If we're not saving the buffer in the kernel pointer, we need a place to copy into. |
587 | */ |
588 | if ((!out_buffer_addr || !out_size_addr) && !(flags & STACKSHOT_SAVE_IN_KERNEL_BUFFER)) { |
589 | return KERN_INVALID_ARGUMENT; |
590 | } |
591 | |
592 | if (since_timestamp != 0 && ((flags & STACKSHOT_COLLECT_DELTA_SNAPSHOT) == 0)) { |
593 | return KERN_INVALID_ARGUMENT; |
594 | } |
595 | |
596 | #if MONOTONIC |
597 | if (!mt_core_supported) { |
598 | flags &= ~STACKSHOT_INSTRS_CYCLES; |
599 | } |
600 | #else /* MONOTONIC */ |
601 | flags &= ~STACKSHOT_INSTRS_CYCLES; |
602 | #endif /* !MONOTONIC */ |
603 | |
604 | STACKSHOT_SUBSYS_LOCK(); |
605 | |
606 | if (flags & STACKSHOT_SAVE_IN_KERNEL_BUFFER) { |
607 | /* |
608 | * Don't overwrite an existing stackshot |
609 | */ |
610 | if (kernel_stackshot_buf != NULL) { |
611 | error = KERN_MEMORY_PRESENT; |
612 | goto error_exit; |
613 | } |
614 | } else if (flags & STACKSHOT_RETRIEVE_EXISTING_BUFFER) { |
615 | if ((kernel_stackshot_buf == NULL) || (kernel_stackshot_buf_size <= 0)) { |
616 | error = KERN_NOT_IN_SET; |
617 | goto error_exit; |
618 | } |
619 | error = stackshot_remap_buffer(kernel_stackshot_buf, kernel_stackshot_buf_size, |
620 | out_buffer_addr, out_size_addr); |
621 | /* |
622 | * If we successfully remapped the buffer into the user's address space, we |
623 | * set buf_to_free and size_to_free so the prior kernel mapping will be removed |
624 | * and then clear the kernel stackshot pointer and associated size. |
625 | */ |
626 | if (error == KERN_SUCCESS) { |
627 | buf_to_free = kernel_stackshot_buf; |
628 | size_to_free = (int) VM_MAP_ROUND_PAGE(kernel_stackshot_buf_size, PAGE_MASK); |
629 | kernel_stackshot_buf = NULL; |
630 | kernel_stackshot_buf_size = 0; |
631 | } |
632 | |
633 | goto error_exit; |
634 | } |
635 | |
636 | if (flags & STACKSHOT_GET_BOOT_PROFILE) { |
637 | void *bootprofile = NULL; |
638 | uint32_t len = 0; |
639 | #if CONFIG_TELEMETRY |
640 | bootprofile_get(&bootprofile, &len); |
641 | #endif |
642 | if (!bootprofile || !len) { |
643 | error = KERN_NOT_IN_SET; |
644 | goto error_exit; |
645 | } |
646 | error = stackshot_remap_buffer(bootprofile, len, out_buffer_addr, out_size_addr); |
647 | goto error_exit; |
648 | } |
649 | |
650 | stackshotbuf_size = get_stackshot_estsize(size_hint); |
651 | |
652 | for (; stackshotbuf_size <= max_tracebuf_size; stackshotbuf_size <<= 1) { |
653 | if (kmem_alloc(kernel_map, (vm_offset_t *)&stackshotbuf, stackshotbuf_size, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) { |
654 | error = KERN_RESOURCE_SHORTAGE; |
655 | goto error_exit; |
656 | } |
657 | |
658 | |
659 | uint32_t hdr_tag = (flags & STACKSHOT_COLLECT_DELTA_SNAPSHOT) ? KCDATA_BUFFER_BEGIN_DELTA_STACKSHOT : KCDATA_BUFFER_BEGIN_STACKSHOT; |
660 | kcdata_p = kcdata_memory_alloc_init((mach_vm_address_t)stackshotbuf, hdr_tag, stackshotbuf_size, |
661 | KCFLAG_USE_MEMCOPY | KCFLAG_NO_AUTO_ENDBUFFER); |
662 | |
663 | stackshot_duration_outer = NULL; |
664 | uint64_t time_start = mach_absolute_time(); |
665 | |
666 | /* |
667 | * Disable interrupts and save the current interrupt state. |
668 | */ |
669 | prev_interrupt_state = ml_set_interrupts_enabled(FALSE); |
670 | |
671 | /* |
672 | * Load stackshot parameters. |
673 | */ |
674 | kdp_snapshot_preflight(pid, stackshotbuf, stackshotbuf_size, flags, kcdata_p, since_timestamp); |
675 | |
676 | error = stackshot_trap(); |
677 | |
678 | ml_set_interrupts_enabled(prev_interrupt_state); |
679 | |
680 | /* record the duration that interupts were disabled */ |
681 | |
682 | uint64_t time_end = mach_absolute_time(); |
683 | if (stackshot_duration_outer) { |
684 | *stackshot_duration_outer = time_end - time_start; |
685 | } |
686 | |
687 | if (error != KERN_SUCCESS) { |
688 | if (kcdata_p != NULL) { |
689 | kcdata_memory_destroy(kcdata_p); |
690 | kcdata_p = NULL; |
691 | stackshot_kcdata_p = NULL; |
692 | } |
693 | kmem_free(kernel_map, (vm_offset_t)stackshotbuf, stackshotbuf_size); |
694 | stackshotbuf = NULL; |
695 | if (error == KERN_INSUFFICIENT_BUFFER_SIZE) { |
696 | /* |
697 | * If we didn't allocate a big enough buffer, deallocate and try again. |
698 | */ |
699 | continue; |
700 | } else { |
701 | goto error_exit; |
702 | } |
703 | } |
704 | |
705 | bytes_traced = kdp_stack_snapshot_bytes_traced(); |
706 | |
707 | if (bytes_traced <= 0) { |
708 | error = KERN_ABORTED; |
709 | goto error_exit; |
710 | } |
711 | |
712 | assert(bytes_traced <= stackshotbuf_size); |
713 | if (!(flags & STACKSHOT_SAVE_IN_KERNEL_BUFFER)) { |
714 | error = stackshot_remap_buffer(stackshotbuf, bytes_traced, out_buffer_addr, out_size_addr); |
715 | goto error_exit; |
716 | } |
717 | |
718 | /* |
719 | * Save the stackshot in the kernel buffer. |
720 | */ |
721 | kernel_stackshot_buf = stackshotbuf; |
722 | kernel_stackshot_buf_size = bytes_traced; |
723 | /* |
724 | * Figure out if we didn't use all the pages in the buffer. If so, we set buf_to_free to the beginning of |
725 | * the next page after the end of the stackshot in the buffer so that the kmem_free clips the buffer and |
726 | * update size_to_free for kmem_free accordingly. |
727 | */ |
728 | size_to_free = stackshotbuf_size - (int) VM_MAP_ROUND_PAGE(bytes_traced, PAGE_MASK); |
729 | |
730 | assert(size_to_free >= 0); |
731 | |
732 | if (size_to_free != 0) { |
733 | buf_to_free = (void *)((uint64_t)stackshotbuf + stackshotbuf_size - size_to_free); |
734 | } |
735 | |
736 | stackshotbuf = NULL; |
737 | stackshotbuf_size = 0; |
738 | goto error_exit; |
739 | } |
740 | |
741 | if (stackshotbuf_size > max_tracebuf_size) { |
742 | error = KERN_RESOURCE_SHORTAGE; |
743 | } |
744 | |
745 | error_exit: |
746 | if (kcdata_p != NULL) { |
747 | kcdata_memory_destroy(kcdata_p); |
748 | kcdata_p = NULL; |
749 | stackshot_kcdata_p = NULL; |
750 | } |
751 | |
752 | if (stackshotbuf != NULL) { |
753 | kmem_free(kernel_map, (vm_offset_t)stackshotbuf, stackshotbuf_size); |
754 | } |
755 | if (buf_to_free != NULL) { |
756 | kmem_free(kernel_map, (vm_offset_t)buf_to_free, size_to_free); |
757 | } |
758 | STACKSHOT_SUBSYS_UNLOCK(); |
759 | return error; |
760 | } |
761 | |
762 | /* |
763 | * Cache stack snapshot parameters in preparation for a trace. |
764 | */ |
765 | void |
766 | kdp_snapshot_preflight(int pid, void * tracebuf, uint32_t tracebuf_size, uint32_t flags, |
767 | kcdata_descriptor_t data_p, uint64_t since_timestamp) |
768 | { |
769 | uint64_t microsecs = 0, secs = 0; |
770 | clock_get_calendar_microtime((clock_sec_t *)&secs, (clock_usec_t *)µsecs); |
771 | |
772 | stackshot_microsecs = microsecs + (secs * USEC_PER_SEC); |
773 | stack_snapshot_pid = pid; |
774 | stack_snapshot_buf = tracebuf; |
775 | stack_snapshot_bufsize = tracebuf_size; |
776 | stack_snapshot_flags = flags; |
777 | stack_snapshot_delta_since_timestamp = since_timestamp; |
778 | |
779 | panic_stackshot = ((flags & STACKSHOT_FROM_PANIC) != 0); |
780 | |
781 | assert(data_p != NULL); |
782 | assert(stackshot_kcdata_p == NULL); |
783 | stackshot_kcdata_p = data_p; |
784 | |
785 | stack_snapshot_bytes_traced = 0; |
786 | } |
787 | |
788 | void |
789 | panic_stackshot_reset_state() |
790 | { |
791 | stackshot_kcdata_p = NULL; |
792 | } |
793 | |
794 | boolean_t |
795 | stackshot_active() |
796 | { |
797 | return (stackshot_kcdata_p != NULL); |
798 | } |
799 | |
800 | uint32_t |
801 | kdp_stack_snapshot_bytes_traced(void) |
802 | { |
803 | return stack_snapshot_bytes_traced; |
804 | } |
805 | |
806 | static boolean_t memory_iszero(void *addr, size_t size) |
807 | { |
808 | char *data = (char *)addr; |
809 | for (size_t i = 0; i < size; i++){ |
810 | if (data[i] != 0) |
811 | return FALSE; |
812 | } |
813 | return TRUE; |
814 | } |
815 | |
816 | #define kcd_end_address(kcd) ((void *)((uint64_t)((kcd)->kcd_addr_begin) + kcdata_memory_get_used_bytes((kcd)))) |
817 | #define kcd_max_address(kcd) ((void *)((kcd)->kcd_addr_begin + (kcd)->kcd_length)) |
818 | /* |
819 | * Use of the kcd_exit_on_error(action) macro requires a local |
820 | * 'kern_return_t error' variable and 'error_exit' label. |
821 | */ |
822 | #define kcd_exit_on_error(action) \ |
823 | do { \ |
824 | if (KERN_SUCCESS != (error = (action))) { \ |
825 | if (error == KERN_RESOURCE_SHORTAGE) { \ |
826 | error = KERN_INSUFFICIENT_BUFFER_SIZE; \ |
827 | } \ |
828 | goto error_exit; \ |
829 | } \ |
830 | } while (0); /* end kcd_exit_on_error */ |
831 | |
832 | static uint64_t |
833 | kcdata_get_task_ss_flags(task_t task) |
834 | { |
835 | uint64_t ss_flags = 0; |
836 | boolean_t task_64bit_addr = task_has_64Bit_addr(task); |
837 | |
838 | if (task_64bit_addr) |
839 | ss_flags |= kUser64_p; |
840 | if (!task->active || task_is_a_corpse(task) || proc_exiting(task->bsd_info)) |
841 | ss_flags |= kTerminatedSnapshot; |
842 | if (task->pidsuspended) |
843 | ss_flags |= kPidSuspended; |
844 | if (task->frozen) |
845 | ss_flags |= kFrozen; |
846 | if (task->effective_policy.tep_darwinbg == 1) |
847 | ss_flags |= kTaskDarwinBG; |
848 | if (task->requested_policy.trp_role == TASK_FOREGROUND_APPLICATION) |
849 | ss_flags |= kTaskIsForeground; |
850 | if (task->requested_policy.trp_boosted == 1) |
851 | ss_flags |= kTaskIsBoosted; |
852 | if (task->effective_policy.tep_sup_active == 1) |
853 | ss_flags |= kTaskIsSuppressed; |
854 | #if CONFIG_MEMORYSTATUS |
855 | |
856 | boolean_t dirty = FALSE, dirty_tracked = FALSE, allow_idle_exit = FALSE; |
857 | memorystatus_proc_flags_unsafe(task->bsd_info, &dirty, &dirty_tracked, &allow_idle_exit); |
858 | if (dirty) |
859 | ss_flags |= kTaskIsDirty; |
860 | if (dirty_tracked) |
861 | ss_flags |= kTaskIsDirtyTracked; |
862 | if (allow_idle_exit) |
863 | ss_flags |= kTaskAllowIdleExit; |
864 | |
865 | #endif |
866 | if (task->effective_policy.tep_tal_engaged) |
867 | ss_flags |= kTaskTALEngaged; |
868 | |
869 | ss_flags |= (0x7 & workqueue_get_pwq_state_kdp(task->bsd_info)) << 17; |
870 | |
871 | #if IMPORTANCE_INHERITANCE |
872 | if (task->task_imp_base) { |
873 | if (task->task_imp_base->iit_donor) |
874 | ss_flags |= kTaskIsImpDonor; |
875 | if (task->task_imp_base->iit_live_donor) |
876 | ss_flags |= kTaskIsLiveImpDonor; |
877 | } |
878 | #endif |
879 | return ss_flags; |
880 | } |
881 | |
882 | static kern_return_t |
883 | kcdata_record_shared_cache_info(kcdata_descriptor_t kcd, task_t task, unaligned_u64 *task_snap_ss_flags) |
884 | { |
885 | kern_return_t error = KERN_SUCCESS; |
886 | mach_vm_address_t out_addr = 0; |
887 | |
888 | uint64_t shared_cache_slide = 0; |
889 | uint64_t shared_cache_base_address = 0; |
890 | uint32_t kdp_fault_results = 0; |
891 | |
892 | assert(task_snap_ss_flags != NULL); |
893 | |
894 | if (task->shared_region && ml_validate_nofault((vm_offset_t)task->shared_region, sizeof(struct vm_shared_region))) { |
895 | struct vm_shared_region *sr = task->shared_region; |
896 | shared_cache_base_address = sr->sr_base_address + sr->sr_first_mapping; |
897 | } else { |
898 | *task_snap_ss_flags |= kTaskSharedRegionInfoUnavailable; |
899 | goto error_exit; |
900 | } |
901 | |
902 | /* We haven't copied in the shared region UUID yet as part of setup */ |
903 | if (!shared_cache_base_address || !task->shared_region->sr_uuid_copied) { |
904 | goto error_exit; |
905 | } |
906 | |
907 | /* |
908 | * No refcounting here, but we are in debugger |
909 | * context, so that should be safe. |
910 | */ |
911 | shared_cache_slide = task->shared_region->sr_slide_info.slide; |
912 | |
913 | if (task->shared_region == init_task_shared_region) { |
914 | /* skip adding shared cache info -- it's the same as the system level one */ |
915 | goto error_exit; |
916 | } |
917 | |
918 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_SHAREDCACHE_LOADINFO, sizeof(struct dyld_uuid_info_64_v2), &out_addr)); |
919 | struct dyld_uuid_info_64_v2 *shared_cache_data = (struct dyld_uuid_info_64_v2 *)out_addr; |
920 | shared_cache_data->imageLoadAddress = shared_cache_slide; |
921 | stackshot_memcpy(shared_cache_data->imageUUID, task->shared_region->sr_uuid, sizeof(task->shared_region->sr_uuid)); |
922 | shared_cache_data->imageSlidBaseAddress = shared_cache_base_address; |
923 | |
924 | error_exit: |
925 | if (kdp_fault_results & KDP_FAULT_RESULT_PAGED_OUT) { |
926 | *task_snap_ss_flags |= kTaskUUIDInfoMissing; |
927 | } |
928 | |
929 | if (kdp_fault_results & KDP_FAULT_RESULT_TRIED_FAULT) { |
930 | *task_snap_ss_flags |= kTaskUUIDInfoTriedFault; |
931 | } |
932 | |
933 | if (kdp_fault_results & KDP_FAULT_RESULT_FAULTED_IN) { |
934 | *task_snap_ss_flags |= kTaskUUIDInfoFaultedIn; |
935 | } |
936 | |
937 | return error; |
938 | } |
939 | |
940 | static kern_return_t |
941 | kcdata_record_uuid_info(kcdata_descriptor_t kcd, task_t task, uint32_t trace_flags, boolean_t have_pmap, unaligned_u64 *task_snap_ss_flags) |
942 | { |
943 | boolean_t save_loadinfo_p = ((trace_flags & STACKSHOT_SAVE_LOADINFO) != 0); |
944 | boolean_t save_kextloadinfo_p = ((trace_flags & STACKSHOT_SAVE_KEXT_LOADINFO) != 0); |
945 | boolean_t should_fault = (trace_flags & STACKSHOT_ENABLE_UUID_FAULTING); |
946 | |
947 | kern_return_t error = KERN_SUCCESS; |
948 | mach_vm_address_t out_addr = 0; |
949 | |
950 | uint32_t uuid_info_count = 0; |
951 | mach_vm_address_t uuid_info_addr = 0; |
952 | uint64_t uuid_info_timestamp = 0; |
953 | uint32_t kdp_fault_results = 0; |
954 | |
955 | assert(task_snap_ss_flags != NULL); |
956 | |
957 | int task_pid = pid_from_task(task); |
958 | boolean_t task_64bit_addr = task_has_64Bit_addr(task); |
959 | |
960 | if (save_loadinfo_p && have_pmap && task->active && task_pid > 0) { |
961 | /* Read the dyld_all_image_infos struct from the task memory to get UUID array count and location */ |
962 | if (task_64bit_addr) { |
963 | struct user64_dyld_all_image_infos task_image_infos; |
964 | if (kdp_copyin(task->map, task->all_image_info_addr, &task_image_infos, |
965 | sizeof(struct user64_dyld_all_image_infos), should_fault, &kdp_fault_results)) { |
966 | uuid_info_count = (uint32_t)task_image_infos.uuidArrayCount; |
967 | uuid_info_addr = task_image_infos.uuidArray; |
968 | if (task_image_infos.version >= DYLD_ALL_IMAGE_INFOS_TIMESTAMP_MINIMUM_VERSION) { |
969 | uuid_info_timestamp = task_image_infos.timestamp; |
970 | } |
971 | } |
972 | } else { |
973 | struct user32_dyld_all_image_infos task_image_infos; |
974 | if (kdp_copyin(task->map, task->all_image_info_addr, &task_image_infos, |
975 | sizeof(struct user32_dyld_all_image_infos), should_fault, &kdp_fault_results)) { |
976 | uuid_info_count = task_image_infos.uuidArrayCount; |
977 | uuid_info_addr = task_image_infos.uuidArray; |
978 | if (task_image_infos.version >= DYLD_ALL_IMAGE_INFOS_TIMESTAMP_MINIMUM_VERSION) { |
979 | uuid_info_timestamp = task_image_infos.timestamp; |
980 | } |
981 | } |
982 | } |
983 | |
984 | /* |
985 | * If we get a NULL uuid_info_addr (which can happen when we catch dyld in the middle of updating |
986 | * this data structure), we zero the uuid_info_count so that we won't even try to save load info |
987 | * for this task. |
988 | */ |
989 | if (!uuid_info_addr) { |
990 | uuid_info_count = 0; |
991 | } |
992 | } |
993 | |
994 | if (have_pmap && task_pid == 0) { |
995 | if (save_kextloadinfo_p && ml_validate_nofault((vm_offset_t)(gLoadedKextSummaries), sizeof(OSKextLoadedKextSummaryHeader))) { |
996 | uuid_info_count = gLoadedKextSummaries->numSummaries + 1; /* include main kernel UUID */ |
997 | } else { |
998 | uuid_info_count = 1; /* include kernelcache UUID (embedded) or kernel UUID (desktop) */ |
999 | } |
1000 | } |
1001 | |
1002 | if (task_pid > 0 && uuid_info_count > 0 && uuid_info_count < MAX_LOADINFOS) { |
1003 | uint32_t uuid_info_size = (uint32_t)(task_64bit_addr ? sizeof(struct user64_dyld_uuid_info) : sizeof(struct user32_dyld_uuid_info)); |
1004 | uint32_t uuid_info_array_size = uuid_info_count * uuid_info_size; |
1005 | |
1006 | kcd_exit_on_error(kcdata_get_memory_addr_for_array(kcd, (task_64bit_addr ? KCDATA_TYPE_LIBRARY_LOADINFO64 : KCDATA_TYPE_LIBRARY_LOADINFO), |
1007 | uuid_info_size, uuid_info_count, &out_addr)); |
1008 | |
1009 | /* Copy in the UUID info array |
1010 | * It may be nonresident, in which case just fix up nloadinfos to 0 in the task_snap |
1011 | */ |
1012 | if (have_pmap && !kdp_copyin(task->map, uuid_info_addr, (void *)out_addr, uuid_info_array_size, should_fault, &kdp_fault_results)) { |
1013 | bzero((void *)out_addr, uuid_info_array_size); |
1014 | } |
1015 | |
1016 | } else if (task_pid == 0 && uuid_info_count > 0 && uuid_info_count < MAX_LOADINFOS) { |
1017 | uintptr_t image_load_address; |
1018 | |
1019 | do { |
1020 | |
1021 | #if CONFIG_EMBEDDED |
1022 | if (kernelcache_uuid_valid && !save_kextloadinfo_p) { |
1023 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_KERNELCACHE_LOADINFO, sizeof(struct dyld_uuid_info_64), &out_addr)); |
1024 | struct dyld_uuid_info_64 *kc_uuid = (struct dyld_uuid_info_64 *)out_addr; |
1025 | kc_uuid->imageLoadAddress = VM_MIN_KERNEL_AND_KEXT_ADDRESS; |
1026 | stackshot_memcpy(&kc_uuid->imageUUID, &kernelcache_uuid, sizeof(uuid_t)); |
1027 | break; |
1028 | } |
1029 | #endif /* CONFIG_EMBEDDED */ |
1030 | |
1031 | if (!kernel_uuid || !ml_validate_nofault((vm_offset_t)kernel_uuid, sizeof(uuid_t))) { |
1032 | /* Kernel UUID not found or inaccessible */ |
1033 | break; |
1034 | } |
1035 | |
1036 | kcd_exit_on_error(kcdata_get_memory_addr_for_array( |
1037 | kcd, (sizeof(kernel_uuid_info) == sizeof(struct user64_dyld_uuid_info)) ? KCDATA_TYPE_LIBRARY_LOADINFO64 |
1038 | : KCDATA_TYPE_LIBRARY_LOADINFO, |
1039 | sizeof(kernel_uuid_info), uuid_info_count, &out_addr)); |
1040 | kernel_uuid_info *uuid_info_array = (kernel_uuid_info *)out_addr; |
1041 | image_load_address = (uintptr_t)VM_KERNEL_UNSLIDE(vm_kernel_stext); |
1042 | uuid_info_array[0].imageLoadAddress = image_load_address; |
1043 | stackshot_memcpy(&uuid_info_array[0].imageUUID, kernel_uuid, sizeof(uuid_t)); |
1044 | |
1045 | if (save_kextloadinfo_p && |
1046 | ml_validate_nofault((vm_offset_t)(gLoadedKextSummaries), sizeof(OSKextLoadedKextSummaryHeader)) && |
1047 | ml_validate_nofault((vm_offset_t)(&gLoadedKextSummaries->summaries[0]), |
1048 | gLoadedKextSummaries->entry_size * gLoadedKextSummaries->numSummaries)) { |
1049 | uint32_t kexti; |
1050 | for (kexti=0 ; kexti < gLoadedKextSummaries->numSummaries; kexti++) { |
1051 | image_load_address = (uintptr_t)VM_KERNEL_UNSLIDE(gLoadedKextSummaries->summaries[kexti].address); |
1052 | uuid_info_array[kexti + 1].imageLoadAddress = image_load_address; |
1053 | stackshot_memcpy(&uuid_info_array[kexti + 1].imageUUID, &gLoadedKextSummaries->summaries[kexti].uuid, sizeof(uuid_t)); |
1054 | } |
1055 | } |
1056 | } while(0); |
1057 | } |
1058 | |
1059 | error_exit: |
1060 | if (kdp_fault_results & KDP_FAULT_RESULT_PAGED_OUT) { |
1061 | *task_snap_ss_flags |= kTaskUUIDInfoMissing; |
1062 | } |
1063 | |
1064 | if (kdp_fault_results & KDP_FAULT_RESULT_TRIED_FAULT) { |
1065 | *task_snap_ss_flags |= kTaskUUIDInfoTriedFault; |
1066 | } |
1067 | |
1068 | if (kdp_fault_results & KDP_FAULT_RESULT_FAULTED_IN) { |
1069 | *task_snap_ss_flags |= kTaskUUIDInfoFaultedIn; |
1070 | } |
1071 | |
1072 | return error; |
1073 | } |
1074 | |
1075 | static kern_return_t |
1076 | kcdata_record_task_iostats(kcdata_descriptor_t kcd, task_t task) |
1077 | { |
1078 | kern_return_t error = KERN_SUCCESS; |
1079 | mach_vm_address_t out_addr = 0; |
1080 | |
1081 | /* I/O Statistics if any counters are non zero */ |
1082 | assert(IO_NUM_PRIORITIES == STACKSHOT_IO_NUM_PRIORITIES); |
1083 | if (task->task_io_stats && !memory_iszero(task->task_io_stats, sizeof(struct io_stat_info))) { |
1084 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_IOSTATS, sizeof(struct io_stats_snapshot), &out_addr)); |
1085 | struct io_stats_snapshot *_iostat = (struct io_stats_snapshot *)out_addr; |
1086 | _iostat->ss_disk_reads_count = task->task_io_stats->disk_reads.count; |
1087 | _iostat->ss_disk_reads_size = task->task_io_stats->disk_reads.size; |
1088 | _iostat->ss_disk_writes_count = (task->task_io_stats->total_io.count - task->task_io_stats->disk_reads.count); |
1089 | _iostat->ss_disk_writes_size = (task->task_io_stats->total_io.size - task->task_io_stats->disk_reads.size); |
1090 | _iostat->ss_paging_count = task->task_io_stats->paging.count; |
1091 | _iostat->ss_paging_size = task->task_io_stats->paging.size; |
1092 | _iostat->ss_non_paging_count = (task->task_io_stats->total_io.count - task->task_io_stats->paging.count); |
1093 | _iostat->ss_non_paging_size = (task->task_io_stats->total_io.size - task->task_io_stats->paging.size); |
1094 | _iostat->ss_metadata_count = task->task_io_stats->metadata.count; |
1095 | _iostat->ss_metadata_size = task->task_io_stats->metadata.size; |
1096 | _iostat->ss_data_count = (task->task_io_stats->total_io.count - task->task_io_stats->metadata.count); |
1097 | _iostat->ss_data_size = (task->task_io_stats->total_io.size - task->task_io_stats->metadata.size); |
1098 | for(int i = 0; i < IO_NUM_PRIORITIES; i++) { |
1099 | _iostat->ss_io_priority_count[i] = task->task_io_stats->io_priority[i].count; |
1100 | _iostat->ss_io_priority_size[i] = task->task_io_stats->io_priority[i].size; |
1101 | } |
1102 | } |
1103 | |
1104 | error_exit: |
1105 | return error; |
1106 | } |
1107 | |
1108 | #if MONOTONIC |
1109 | static kern_return_t |
1110 | kcdata_record_task_instrs_cycles(kcdata_descriptor_t kcd, task_t task) |
1111 | { |
1112 | uint64_t instrs = 0, cycles = 0; |
1113 | mt_stackshot_task(task, &instrs, &cycles); |
1114 | |
1115 | kern_return_t error = KERN_SUCCESS; |
1116 | mach_vm_address_t out_addr = 0; |
1117 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_INSTRS_CYCLES, sizeof(struct instrs_cycles_snapshot), &out_addr)); |
1118 | struct instrs_cycles_snapshot *instrs_cycles = (struct instrs_cycles_snapshot *)out_addr; |
1119 | instrs_cycles->ics_instructions = instrs; |
1120 | instrs_cycles->ics_cycles = cycles; |
1121 | |
1122 | error_exit: |
1123 | return error; |
1124 | } |
1125 | #endif /* MONOTONIC */ |
1126 | |
1127 | static kern_return_t |
1128 | kcdata_record_task_snapshot(kcdata_descriptor_t kcd, task_t task, uint32_t trace_flags, boolean_t have_pmap, unaligned_u64 **task_snap_ss_flags) |
1129 | { |
1130 | boolean_t collect_delta_stackshot = ((trace_flags & STACKSHOT_COLLECT_DELTA_SNAPSHOT) != 0); |
1131 | boolean_t collect_iostats = !collect_delta_stackshot && !(trace_flags & STACKSHOT_NO_IO_STATS); |
1132 | #if MONOTONIC |
1133 | boolean_t collect_instrs_cycles = ((trace_flags & STACKSHOT_INSTRS_CYCLES) != 0); |
1134 | #endif /* MONOTONIC */ |
1135 | #if __arm__ || __arm64__ |
1136 | boolean_t collect_asid = ((trace_flags & STACKSHOT_ASID) != 0); |
1137 | #endif |
1138 | boolean_t collect_pagetables = ((trace_flags & STACKSHOT_PAGE_TABLES) != 0); |
1139 | |
1140 | |
1141 | kern_return_t error = KERN_SUCCESS; |
1142 | mach_vm_address_t out_addr = 0; |
1143 | struct task_snapshot_v2 * cur_tsnap = NULL; |
1144 | |
1145 | assert(task_snap_ss_flags != NULL); |
1146 | |
1147 | int task_pid = pid_from_task(task); |
1148 | uint64_t task_uniqueid = get_task_uniqueid(task); |
1149 | uint64_t proc_starttime_secs = 0; |
1150 | |
1151 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_TASK_SNAPSHOT, sizeof(struct task_snapshot_v2), &out_addr)); |
1152 | cur_tsnap = (struct task_snapshot_v2 *)out_addr; |
1153 | bzero(cur_tsnap, sizeof(*cur_tsnap)); |
1154 | |
1155 | cur_tsnap->ts_unique_pid = task_uniqueid; |
1156 | cur_tsnap->ts_ss_flags = kcdata_get_task_ss_flags(task); |
1157 | *task_snap_ss_flags = (unaligned_u64 *)&cur_tsnap->ts_ss_flags; |
1158 | cur_tsnap->ts_user_time_in_terminated_threads = task->total_user_time; |
1159 | cur_tsnap->ts_system_time_in_terminated_threads = task->total_system_time; |
1160 | |
1161 | proc_starttime_kdp(task->bsd_info, &proc_starttime_secs, NULL, NULL); |
1162 | cur_tsnap->ts_p_start_sec = proc_starttime_secs; |
1163 | cur_tsnap->ts_task_size = have_pmap ? get_task_phys_footprint(task) : 0; |
1164 | cur_tsnap->ts_max_resident_size = get_task_resident_max(task); |
1165 | cur_tsnap->ts_was_throttled = (uint32_t) proc_was_throttled_from_task(task); |
1166 | cur_tsnap->ts_did_throttle = (uint32_t) proc_did_throttle_from_task(task); |
1167 | |
1168 | cur_tsnap->ts_suspend_count = task->suspend_count; |
1169 | cur_tsnap->ts_faults = task->faults; |
1170 | cur_tsnap->ts_pageins = task->pageins; |
1171 | cur_tsnap->ts_cow_faults = task->cow_faults; |
1172 | cur_tsnap->ts_latency_qos = (task->effective_policy.tep_latency_qos == LATENCY_QOS_TIER_UNSPECIFIED) ? |
1173 | LATENCY_QOS_TIER_UNSPECIFIED : ((0xFF << 16) | task->effective_policy.tep_latency_qos); |
1174 | cur_tsnap->ts_pid = task_pid; |
1175 | |
1176 | #if __arm__ || __arm64__ |
1177 | if (collect_asid && have_pmap) { |
1178 | uint32_t asid = task->map->pmap->asid; |
1179 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_ASID, sizeof(uint32_t), &out_addr)); |
1180 | stackshot_memcpy((void*)out_addr, &asid, sizeof(asid)); |
1181 | } |
1182 | #endif |
1183 | if (collect_pagetables && have_pmap) { |
1184 | #if INTERRUPT_MASKED_DEBUG |
1185 | // pagetable dumps can be large; reset the interrupt timeout to avoid a panic |
1186 | ml_spin_debug_clear_self(); |
1187 | #endif |
1188 | size_t bytes_dumped = pmap_dump_page_tables(task->map->pmap, kcd_end_address(kcd), kcd_max_address(kcd)); |
1189 | if (bytes_dumped == 0) { |
1190 | error = KERN_INSUFFICIENT_BUFFER_SIZE; |
1191 | goto error_exit; |
1192 | } else if (bytes_dumped == (size_t)-1) { |
1193 | error = KERN_NOT_SUPPORTED; |
1194 | goto error_exit; |
1195 | } else { |
1196 | kcd_exit_on_error(kcdata_get_memory_addr_for_array(kcd, STACKSHOT_KCTYPE_PAGE_TABLES, |
1197 | sizeof(uint64_t), (uint32_t)(bytes_dumped / sizeof(uint64_t)), &out_addr)); |
1198 | } |
1199 | } |
1200 | |
1201 | /* Add the BSD process identifiers */ |
1202 | if (task_pid != -1 && task->bsd_info != NULL) { |
1203 | proc_name_kdp(task, cur_tsnap->ts_p_comm, sizeof(cur_tsnap->ts_p_comm)); |
1204 | #if CONFIG_COALITIONS |
1205 | if ((trace_flags & STACKSHOT_SAVE_JETSAM_COALITIONS) && (task->coalition[COALITION_TYPE_JETSAM] != NULL)) { |
1206 | uint64_t jetsam_coal_id = coalition_id(task->coalition[COALITION_TYPE_JETSAM]); |
1207 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_JETSAM_COALITION, sizeof(jetsam_coal_id), &out_addr)); |
1208 | stackshot_memcpy((void*)out_addr, &jetsam_coal_id, sizeof(jetsam_coal_id)); |
1209 | } |
1210 | #endif /* CONFIG_COALITIONS */ |
1211 | } |
1212 | else { |
1213 | cur_tsnap->ts_p_comm[0] = '\0'; |
1214 | #if IMPORTANCE_INHERITANCE && (DEVELOPMENT || DEBUG) |
1215 | if (task->task_imp_base != NULL) { |
1216 | stackshot_strlcpy(cur_tsnap->ts_p_comm, &task->task_imp_base->iit_procname[0], |
1217 | MIN((int)sizeof(task->task_imp_base->iit_procname), (int)sizeof(cur_tsnap->ts_p_comm))); |
1218 | } |
1219 | #endif /* IMPORTANCE_INHERITANCE && (DEVELOPMENT || DEBUG) */ |
1220 | } |
1221 | |
1222 | if (collect_iostats) { |
1223 | kcd_exit_on_error(kcdata_record_task_iostats(kcd, task)); |
1224 | } |
1225 | |
1226 | #if MONOTONIC |
1227 | if (collect_instrs_cycles) { |
1228 | kcd_exit_on_error(kcdata_record_task_instrs_cycles(kcd, task)); |
1229 | } |
1230 | #endif /* MONOTONIC */ |
1231 | |
1232 | error_exit: |
1233 | return error; |
1234 | } |
1235 | |
1236 | static kern_return_t |
1237 | kcdata_record_task_delta_snapshot(kcdata_descriptor_t kcd, task_t task, uint32_t trace_flags, boolean_t have_pmap, unaligned_u64 **task_snap_ss_flags) |
1238 | { |
1239 | #if !MONOTONIC |
1240 | #pragma unused(trace_flags) |
1241 | #endif /* !MONOTONIC */ |
1242 | kern_return_t error = KERN_SUCCESS; |
1243 | struct task_delta_snapshot_v2 * cur_tsnap = NULL; |
1244 | mach_vm_address_t out_addr = 0; |
1245 | (void) trace_flags; |
1246 | #if __arm__ || __arm64__ |
1247 | boolean_t collect_asid = ((trace_flags & STACKSHOT_ASID) != 0); |
1248 | #endif |
1249 | #if MONOTONIC |
1250 | boolean_t collect_instrs_cycles = ((trace_flags & STACKSHOT_INSTRS_CYCLES) != 0); |
1251 | #endif /* MONOTONIC */ |
1252 | |
1253 | uint64_t task_uniqueid = get_task_uniqueid(task); |
1254 | assert(task_snap_ss_flags != NULL); |
1255 | |
1256 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_TASK_DELTA_SNAPSHOT, sizeof(struct task_delta_snapshot_v2), &out_addr)); |
1257 | |
1258 | cur_tsnap = (struct task_delta_snapshot_v2 *)out_addr; |
1259 | |
1260 | cur_tsnap->tds_unique_pid = task_uniqueid; |
1261 | cur_tsnap->tds_ss_flags = kcdata_get_task_ss_flags(task); |
1262 | *task_snap_ss_flags = (unaligned_u64 *)&cur_tsnap->tds_ss_flags; |
1263 | |
1264 | cur_tsnap->tds_user_time_in_terminated_threads = task->total_user_time; |
1265 | cur_tsnap->tds_system_time_in_terminated_threads = task->total_system_time; |
1266 | |
1267 | cur_tsnap->tds_task_size = have_pmap ? get_task_phys_footprint(task) : 0; |
1268 | |
1269 | cur_tsnap->tds_max_resident_size = get_task_resident_max(task); |
1270 | cur_tsnap->tds_suspend_count = task->suspend_count; |
1271 | cur_tsnap->tds_faults = task->faults; |
1272 | cur_tsnap->tds_pageins = task->pageins; |
1273 | cur_tsnap->tds_cow_faults = task->cow_faults; |
1274 | cur_tsnap->tds_was_throttled = (uint32_t)proc_was_throttled_from_task(task); |
1275 | cur_tsnap->tds_did_throttle = (uint32_t)proc_did_throttle_from_task(task); |
1276 | cur_tsnap->tds_latency_qos = (task-> effective_policy.tep_latency_qos == LATENCY_QOS_TIER_UNSPECIFIED) |
1277 | ? LATENCY_QOS_TIER_UNSPECIFIED |
1278 | : ((0xFF << 16) | task-> effective_policy.tep_latency_qos); |
1279 | |
1280 | #if __arm__ || __arm64__ |
1281 | if (collect_asid && have_pmap) { |
1282 | uint32_t asid = task->map->pmap->asid; |
1283 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_ASID, sizeof(uint32_t), &out_addr)); |
1284 | stackshot_memcpy((void*)out_addr, &asid, sizeof(asid)); |
1285 | } |
1286 | #endif |
1287 | |
1288 | #if MONOTONIC |
1289 | if (collect_instrs_cycles) { |
1290 | kcd_exit_on_error(kcdata_record_task_instrs_cycles(kcd, task)); |
1291 | } |
1292 | #endif /* MONOTONIC */ |
1293 | |
1294 | error_exit: |
1295 | return error; |
1296 | } |
1297 | |
1298 | static kern_return_t |
1299 | kcdata_record_thread_iostats(kcdata_descriptor_t kcd, thread_t thread) |
1300 | { |
1301 | kern_return_t error = KERN_SUCCESS; |
1302 | mach_vm_address_t out_addr = 0; |
1303 | |
1304 | /* I/O Statistics */ |
1305 | assert(IO_NUM_PRIORITIES == STACKSHOT_IO_NUM_PRIORITIES); |
1306 | if (thread->thread_io_stats && !memory_iszero(thread->thread_io_stats, sizeof(struct io_stat_info))) { |
1307 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_IOSTATS, sizeof(struct io_stats_snapshot), &out_addr)); |
1308 | struct io_stats_snapshot *_iostat = (struct io_stats_snapshot *)out_addr; |
1309 | _iostat->ss_disk_reads_count = thread->thread_io_stats->disk_reads.count; |
1310 | _iostat->ss_disk_reads_size = thread->thread_io_stats->disk_reads.size; |
1311 | _iostat->ss_disk_writes_count = (thread->thread_io_stats->total_io.count - thread->thread_io_stats->disk_reads.count); |
1312 | _iostat->ss_disk_writes_size = (thread->thread_io_stats->total_io.size - thread->thread_io_stats->disk_reads.size); |
1313 | _iostat->ss_paging_count = thread->thread_io_stats->paging.count; |
1314 | _iostat->ss_paging_size = thread->thread_io_stats->paging.size; |
1315 | _iostat->ss_non_paging_count = (thread->thread_io_stats->total_io.count - thread->thread_io_stats->paging.count); |
1316 | _iostat->ss_non_paging_size = (thread->thread_io_stats->total_io.size - thread->thread_io_stats->paging.size); |
1317 | _iostat->ss_metadata_count = thread->thread_io_stats->metadata.count; |
1318 | _iostat->ss_metadata_size = thread->thread_io_stats->metadata.size; |
1319 | _iostat->ss_data_count = (thread->thread_io_stats->total_io.count - thread->thread_io_stats->metadata.count); |
1320 | _iostat->ss_data_size = (thread->thread_io_stats->total_io.size - thread->thread_io_stats->metadata.size); |
1321 | for(int i = 0; i < IO_NUM_PRIORITIES; i++) { |
1322 | _iostat->ss_io_priority_count[i] = thread->thread_io_stats->io_priority[i].count; |
1323 | _iostat->ss_io_priority_size[i] = thread->thread_io_stats->io_priority[i].size; |
1324 | } |
1325 | } |
1326 | |
1327 | error_exit: |
1328 | return error; |
1329 | } |
1330 | |
1331 | static kern_return_t |
1332 | kcdata_record_thread_snapshot( |
1333 | kcdata_descriptor_t kcd, thread_t thread, task_t task, uint32_t trace_flags, boolean_t have_pmap, boolean_t thread_on_core) |
1334 | { |
1335 | boolean_t dispatch_p = ((trace_flags & STACKSHOT_GET_DQ) != 0); |
1336 | boolean_t active_kthreads_only_p = ((trace_flags & STACKSHOT_ACTIVE_KERNEL_THREADS_ONLY) != 0); |
1337 | boolean_t trace_fp_p = false; |
1338 | boolean_t collect_delta_stackshot = ((trace_flags & STACKSHOT_COLLECT_DELTA_SNAPSHOT) != 0); |
1339 | boolean_t collect_iostats = !collect_delta_stackshot && !(trace_flags & STACKSHOT_NO_IO_STATS); |
1340 | #if MONOTONIC |
1341 | boolean_t collect_instrs_cycles = ((trace_flags & STACKSHOT_INSTRS_CYCLES) != 0); |
1342 | #endif /* MONOTONIC */ |
1343 | |
1344 | kern_return_t error = KERN_SUCCESS; |
1345 | mach_vm_address_t out_addr = 0; |
1346 | int saved_count = 0; |
1347 | |
1348 | struct thread_snapshot_v4 * cur_thread_snap = NULL; |
1349 | char cur_thread_name[STACKSHOT_MAX_THREAD_NAME_SIZE]; |
1350 | uint64_t tval = 0; |
1351 | const boolean_t is_64bit_data = task_has_64Bit_data(task); |
1352 | |
1353 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_THREAD_SNAPSHOT, sizeof(struct thread_snapshot_v4), &out_addr)); |
1354 | cur_thread_snap = (struct thread_snapshot_v4 *)out_addr; |
1355 | |
1356 | /* Populate the thread snapshot header */ |
1357 | cur_thread_snap->ths_thread_id = thread_tid(thread); |
1358 | cur_thread_snap->ths_wait_event = VM_KERNEL_UNSLIDE_OR_PERM(thread->wait_event); |
1359 | cur_thread_snap->ths_continuation = VM_KERNEL_UNSLIDE(thread->continuation); |
1360 | cur_thread_snap->ths_total_syscalls = thread->syscalls_mach + thread->syscalls_unix; |
1361 | |
1362 | if (IPC_VOUCHER_NULL != thread->ith_voucher) |
1363 | cur_thread_snap->ths_voucher_identifier = VM_KERNEL_ADDRPERM(thread->ith_voucher); |
1364 | else |
1365 | cur_thread_snap->ths_voucher_identifier = 0; |
1366 | |
1367 | cur_thread_snap->ths_dqserialnum = 0; |
1368 | if (dispatch_p && (task != kernel_task) && (task->active) && have_pmap) { |
1369 | uint64_t dqkeyaddr = thread_dispatchqaddr(thread); |
1370 | if (dqkeyaddr != 0) { |
1371 | uint64_t dqaddr = 0; |
1372 | boolean_t copyin_ok = kdp_copyin_word(task, dqkeyaddr, &dqaddr, FALSE, NULL); |
1373 | if (copyin_ok && dqaddr != 0) { |
1374 | uint64_t dqserialnumaddr = dqaddr + get_task_dispatchqueue_serialno_offset(task); |
1375 | uint64_t dqserialnum = 0; |
1376 | copyin_ok = kdp_copyin_word(task, dqserialnumaddr, &dqserialnum, FALSE, NULL); |
1377 | if (copyin_ok) { |
1378 | cur_thread_snap->ths_ss_flags |= kHasDispatchSerial; |
1379 | cur_thread_snap->ths_dqserialnum = dqserialnum; |
1380 | } |
1381 | } |
1382 | } |
1383 | } |
1384 | |
1385 | tval = safe_grab_timer_value(&thread->user_timer); |
1386 | cur_thread_snap->ths_user_time = tval; |
1387 | tval = safe_grab_timer_value(&thread->system_timer); |
1388 | |
1389 | if (thread->precise_user_kernel_time) { |
1390 | cur_thread_snap->ths_sys_time = tval; |
1391 | } else { |
1392 | cur_thread_snap->ths_user_time += tval; |
1393 | cur_thread_snap->ths_sys_time = 0; |
1394 | } |
1395 | |
1396 | cur_thread_snap->ths_ss_flags = 0; |
1397 | if (thread->thread_tag & THREAD_TAG_MAINTHREAD) |
1398 | cur_thread_snap->ths_ss_flags |= kThreadMain; |
1399 | if (thread->effective_policy.thep_darwinbg) |
1400 | cur_thread_snap->ths_ss_flags |= kThreadDarwinBG; |
1401 | if (proc_get_effective_thread_policy(thread, TASK_POLICY_PASSIVE_IO)) |
1402 | cur_thread_snap->ths_ss_flags |= kThreadIOPassive; |
1403 | if (thread->suspend_count > 0) |
1404 | cur_thread_snap->ths_ss_flags |= kThreadSuspended; |
1405 | if (thread->options & TH_OPT_GLOBAL_FORCED_IDLE) |
1406 | cur_thread_snap->ths_ss_flags |= kGlobalForcedIdle; |
1407 | if (thread_on_core) |
1408 | cur_thread_snap->ths_ss_flags |= kThreadOnCore; |
1409 | if (stackshot_thread_is_idle_worker_unsafe(thread)) |
1410 | cur_thread_snap->ths_ss_flags |= kThreadIdleWorker; |
1411 | |
1412 | /* make sure state flags defined in kcdata.h still match internal flags */ |
1413 | static_assert(SS_TH_WAIT == TH_WAIT); |
1414 | static_assert(SS_TH_SUSP == TH_SUSP); |
1415 | static_assert(SS_TH_RUN == TH_RUN); |
1416 | static_assert(SS_TH_UNINT == TH_UNINT); |
1417 | static_assert(SS_TH_TERMINATE == TH_TERMINATE); |
1418 | static_assert(SS_TH_TERMINATE2 == TH_TERMINATE2); |
1419 | static_assert(SS_TH_IDLE == TH_IDLE); |
1420 | |
1421 | cur_thread_snap->ths_last_run_time = thread->last_run_time; |
1422 | cur_thread_snap->ths_last_made_runnable_time = thread->last_made_runnable_time; |
1423 | cur_thread_snap->ths_state = thread->state; |
1424 | cur_thread_snap->ths_sched_flags = thread->sched_flags; |
1425 | cur_thread_snap->ths_base_priority = thread->base_pri; |
1426 | cur_thread_snap->ths_sched_priority = thread->sched_pri; |
1427 | cur_thread_snap->ths_eqos = thread->effective_policy.thep_qos; |
1428 | cur_thread_snap->ths_rqos = thread->requested_policy.thrp_qos; |
1429 | cur_thread_snap->ths_rqos_override = MAX(thread->requested_policy.thrp_qos_override, |
1430 | thread->requested_policy.thrp_qos_workq_override); |
1431 | cur_thread_snap->ths_io_tier = proc_get_effective_thread_policy(thread, TASK_POLICY_IO); |
1432 | cur_thread_snap->ths_thread_t = VM_KERNEL_UNSLIDE_OR_PERM(thread); |
1433 | |
1434 | static_assert(sizeof(thread->effective_policy) == sizeof(uint64_t)); |
1435 | static_assert(sizeof(thread->requested_policy) == sizeof(uint64_t)); |
1436 | cur_thread_snap->ths_requested_policy = *(unaligned_u64 *) &thread->requested_policy; |
1437 | cur_thread_snap->ths_effective_policy = *(unaligned_u64 *) &thread->effective_policy; |
1438 | |
1439 | /* if there is thread name then add to buffer */ |
1440 | cur_thread_name[0] = '\0'; |
1441 | proc_threadname_kdp(thread->uthread, cur_thread_name, STACKSHOT_MAX_THREAD_NAME_SIZE); |
1442 | if (strnlen(cur_thread_name, STACKSHOT_MAX_THREAD_NAME_SIZE) > 0) { |
1443 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_THREAD_NAME, sizeof(cur_thread_name), &out_addr)); |
1444 | stackshot_memcpy((void *)out_addr, (void *)cur_thread_name, sizeof(cur_thread_name)); |
1445 | } |
1446 | |
1447 | /* record system, user, and runnable times */ |
1448 | time_value_t user_time, system_time, runnable_time; |
1449 | thread_read_times(thread, &user_time, &system_time, &runnable_time); |
1450 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_CPU_TIMES, sizeof(struct stackshot_cpu_times_v2), &out_addr)); |
1451 | struct stackshot_cpu_times_v2 *stackshot_cpu_times = (struct stackshot_cpu_times_v2 *)out_addr; |
1452 | *stackshot_cpu_times = (struct stackshot_cpu_times_v2){ |
1453 | .user_usec = (uint64_t)user_time.seconds * USEC_PER_SEC + user_time.microseconds, |
1454 | .system_usec = (uint64_t)system_time.seconds * USEC_PER_SEC + system_time.microseconds, |
1455 | .runnable_usec = (uint64_t)runnable_time.seconds * USEC_PER_SEC + runnable_time.microseconds, |
1456 | }; |
1457 | |
1458 | /* Trace user stack, if any */ |
1459 | if (!active_kthreads_only_p && task->active && thread->task->map != kernel_map) { |
1460 | uint32_t thread_snapshot_flags = 0; |
1461 | |
1462 | /* Uses 64-bit machine state? */ |
1463 | if (is_64bit_data) { |
1464 | uint64_t sp = 0; |
1465 | out_addr = (mach_vm_address_t)kcd_end_address(kcd); |
1466 | saved_count = machine_trace_thread64(thread, (char *)out_addr, (char *)kcd_max_address(kcd), MAX_FRAMES, TRUE, |
1467 | trace_fp_p, &thread_snapshot_flags, &sp); |
1468 | if (saved_count > 0) { |
1469 | int frame_size = trace_fp_p ? sizeof(struct stack_snapshot_frame64) : sizeof(uint64_t); |
1470 | kcd_exit_on_error(kcdata_get_memory_addr_for_array(kcd, trace_fp_p ? STACKSHOT_KCTYPE_USER_STACKFRAME64 |
1471 | : STACKSHOT_KCTYPE_USER_STACKLR64, |
1472 | frame_size, saved_count / frame_size, &out_addr)); |
1473 | cur_thread_snap->ths_ss_flags |= kUser64_p; |
1474 | } |
1475 | #if __x86_64__ |
1476 | if (sp) { |
1477 | // I'm using 8 here and not sizeof(stack_contents) because this |
1478 | // code would not work if you just made stack_contents bigger. |
1479 | vm_offset_t kern_virt_addr = machine_trace_thread_get_kva(sp, thread->task->map, &thread_snapshot_flags); |
1480 | if (kern_virt_addr && (kern_virt_addr % 8) == 0) { |
1481 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_USER_STACKTOP, sizeof(struct stack_snapshot_stacktop), &out_addr)); |
1482 | struct stack_snapshot_stacktop *stacktop = (struct stack_snapshot_stacktop *)out_addr; |
1483 | stacktop->sp = sp; |
1484 | memcpy(stacktop->stack_contents, (void*) kern_virt_addr, 8); |
1485 | } |
1486 | } |
1487 | #endif |
1488 | } else { |
1489 | out_addr = (mach_vm_address_t)kcd_end_address(kcd); |
1490 | saved_count = machine_trace_thread(thread, (char *)out_addr, (char *)kcd_max_address(kcd), MAX_FRAMES, TRUE, trace_fp_p, |
1491 | &thread_snapshot_flags); |
1492 | if (saved_count > 0) { |
1493 | int frame_size = trace_fp_p ? sizeof(struct stack_snapshot_frame32) : sizeof(uint32_t); |
1494 | kcd_exit_on_error(kcdata_get_memory_addr_for_array(kcd, trace_fp_p ? STACKSHOT_KCTYPE_USER_STACKFRAME |
1495 | : STACKSHOT_KCTYPE_USER_STACKLR, |
1496 | frame_size, saved_count / frame_size, &out_addr)); |
1497 | } |
1498 | } |
1499 | |
1500 | if (thread_snapshot_flags != 0) { |
1501 | cur_thread_snap->ths_ss_flags |= thread_snapshot_flags; |
1502 | } |
1503 | } |
1504 | |
1505 | /* Call through to the machine specific trace routines |
1506 | * Frames are added past the snapshot header. |
1507 | */ |
1508 | if (thread->kernel_stack != 0) { |
1509 | uint32_t thread_snapshot_flags = 0; |
1510 | #if defined(__LP64__) |
1511 | out_addr = (mach_vm_address_t)kcd_end_address(kcd); |
1512 | saved_count = machine_trace_thread64(thread, (char *)out_addr, (char *)kcd_max_address(kcd), MAX_FRAMES, FALSE, trace_fp_p, |
1513 | &thread_snapshot_flags, NULL); |
1514 | if (saved_count > 0) { |
1515 | int frame_size = trace_fp_p ? sizeof(struct stack_snapshot_frame64) : sizeof(uint64_t); |
1516 | cur_thread_snap->ths_ss_flags |= kKernel64_p; |
1517 | kcd_exit_on_error(kcdata_get_memory_addr_for_array(kcd, trace_fp_p ? STACKSHOT_KCTYPE_KERN_STACKFRAME64 |
1518 | : STACKSHOT_KCTYPE_KERN_STACKLR64, |
1519 | frame_size, saved_count / frame_size, &out_addr)); |
1520 | } |
1521 | #else |
1522 | out_addr = (mach_vm_address_t)kcd_end_address(kcd); |
1523 | saved_count = machine_trace_thread(thread, (char *)out_addr, (char *)kcd_max_address(kcd), MAX_FRAMES, FALSE, trace_fp_p, |
1524 | &thread_snapshot_flags); |
1525 | if (saved_count > 0) { |
1526 | int frame_size = trace_fp_p ? sizeof(struct stack_snapshot_frame32) : sizeof(uint32_t); |
1527 | kcd_exit_on_error( |
1528 | kcdata_get_memory_addr_for_array(kcd, trace_fp_p ? STACKSHOT_KCTYPE_KERN_STACKFRAME : STACKSHOT_KCTYPE_KERN_STACKLR, |
1529 | frame_size, saved_count / frame_size, &out_addr)); |
1530 | } |
1531 | #endif |
1532 | if (thread_snapshot_flags != 0) { |
1533 | cur_thread_snap->ths_ss_flags |= thread_snapshot_flags; |
1534 | } |
1535 | } |
1536 | |
1537 | |
1538 | if (collect_iostats) { |
1539 | kcd_exit_on_error(kcdata_record_thread_iostats(kcd, thread)); |
1540 | } |
1541 | |
1542 | #if MONOTONIC |
1543 | if (collect_instrs_cycles) { |
1544 | uint64_t instrs = 0, cycles = 0; |
1545 | mt_stackshot_thread(thread, &instrs, &cycles); |
1546 | |
1547 | kcd_exit_on_error(kcdata_get_memory_addr(kcd, STACKSHOT_KCTYPE_INSTRS_CYCLES, sizeof(struct instrs_cycles_snapshot), &out_addr)); |
1548 | struct instrs_cycles_snapshot *instrs_cycles = (struct instrs_cycles_snapshot *)out_addr; |
1549 | instrs_cycles->ics_instructions = instrs; |
1550 | instrs_cycles->ics_cycles = cycles; |
1551 | } |
1552 | #endif /* MONOTONIC */ |
1553 | |
1554 | error_exit: |
1555 | return error; |
1556 | } |
1557 | |
1558 | static int |
1559 | kcdata_record_thread_delta_snapshot(struct thread_delta_snapshot_v3 * cur_thread_snap, thread_t thread, boolean_t thread_on_core) |
1560 | { |
1561 | cur_thread_snap->tds_thread_id = thread_tid(thread); |
1562 | if (IPC_VOUCHER_NULL != thread->ith_voucher) |
1563 | cur_thread_snap->tds_voucher_identifier = VM_KERNEL_ADDRPERM(thread->ith_voucher); |
1564 | else |
1565 | cur_thread_snap->tds_voucher_identifier = 0; |
1566 | |
1567 | cur_thread_snap->tds_ss_flags = 0; |
1568 | if (thread->effective_policy.thep_darwinbg) |
1569 | cur_thread_snap->tds_ss_flags |= kThreadDarwinBG; |
1570 | if (proc_get_effective_thread_policy(thread, TASK_POLICY_PASSIVE_IO)) |
1571 | cur_thread_snap->tds_ss_flags |= kThreadIOPassive; |
1572 | if (thread->suspend_count > 0) |
1573 | cur_thread_snap->tds_ss_flags |= kThreadSuspended; |
1574 | if (thread->options & TH_OPT_GLOBAL_FORCED_IDLE) |
1575 | cur_thread_snap->tds_ss_flags |= kGlobalForcedIdle; |
1576 | if (thread_on_core) |
1577 | cur_thread_snap->tds_ss_flags |= kThreadOnCore; |
1578 | if (stackshot_thread_is_idle_worker_unsafe(thread)) |
1579 | cur_thread_snap->tds_ss_flags |= kThreadIdleWorker; |
1580 | |
1581 | cur_thread_snap->tds_last_made_runnable_time = thread->last_made_runnable_time; |
1582 | cur_thread_snap->tds_state = thread->state; |
1583 | cur_thread_snap->tds_sched_flags = thread->sched_flags; |
1584 | cur_thread_snap->tds_base_priority = thread->base_pri; |
1585 | cur_thread_snap->tds_sched_priority = thread->sched_pri; |
1586 | cur_thread_snap->tds_eqos = thread->effective_policy.thep_qos; |
1587 | cur_thread_snap->tds_rqos = thread->requested_policy.thrp_qos; |
1588 | cur_thread_snap->tds_rqos_override = MAX(thread->requested_policy.thrp_qos_override, |
1589 | thread->requested_policy.thrp_qos_workq_override); |
1590 | cur_thread_snap->tds_io_tier = proc_get_effective_thread_policy(thread, TASK_POLICY_IO); |
1591 | |
1592 | static_assert(sizeof(thread->effective_policy) == sizeof(uint64_t)); |
1593 | static_assert(sizeof(thread->requested_policy) == sizeof(uint64_t)); |
1594 | cur_thread_snap->tds_requested_policy = *(unaligned_u64 *) &thread->requested_policy; |
1595 | cur_thread_snap->tds_effective_policy = *(unaligned_u64 *) &thread->effective_policy; |
1596 | |
1597 | return 0; |
1598 | } |
1599 | |
1600 | /* |
1601 | * Why 12? 12 strikes a decent balance between allocating a large array on |
1602 | * the stack and having large kcdata item overheads for recording nonrunable |
1603 | * tasks. |
1604 | */ |
1605 | #define UNIQUEIDSPERFLUSH 12 |
1606 | |
1607 | struct saved_uniqueids { |
1608 | uint64_t ids[UNIQUEIDSPERFLUSH]; |
1609 | unsigned count; |
1610 | }; |
1611 | |
1612 | enum thread_classification { |
1613 | tc_full_snapshot, /* take a full snapshot */ |
1614 | tc_delta_snapshot, /* take a delta snapshot */ |
1615 | }; |
1616 | |
1617 | static enum thread_classification |
1618 | classify_thread(thread_t thread, boolean_t * thread_on_core_p, uint32_t trace_flags) |
1619 | { |
1620 | boolean_t collect_delta_stackshot = ((trace_flags & STACKSHOT_COLLECT_DELTA_SNAPSHOT) != 0); |
1621 | |
1622 | processor_t last_processor = thread->last_processor; |
1623 | |
1624 | boolean_t thread_on_core = |
1625 | (last_processor != PROCESSOR_NULL && last_processor->state == PROCESSOR_RUNNING && last_processor->active_thread == thread); |
1626 | |
1627 | *thread_on_core_p = thread_on_core; |
1628 | |
1629 | /* Capture the full thread snapshot if this is not a delta stackshot or if the thread has run subsequent to the |
1630 | * previous full stackshot */ |
1631 | if (!collect_delta_stackshot || thread_on_core || (thread->last_run_time > stack_snapshot_delta_since_timestamp)) { |
1632 | return tc_full_snapshot; |
1633 | } else { |
1634 | return tc_delta_snapshot; |
1635 | } |
1636 | } |
1637 | |
1638 | struct stackshot_context |
1639 | { |
1640 | int pid; |
1641 | uint32_t trace_flags; |
1642 | }; |
1643 | |
1644 | static kern_return_t |
1645 | kdp_stackshot_record_task(struct stackshot_context *ctx, task_t task) |
1646 | { |
1647 | boolean_t active_kthreads_only_p = ((ctx->trace_flags & STACKSHOT_ACTIVE_KERNEL_THREADS_ONLY) != 0); |
1648 | boolean_t save_donating_pids_p = ((ctx->trace_flags & STACKSHOT_SAVE_IMP_DONATION_PIDS) != 0); |
1649 | boolean_t collect_delta_stackshot = ((ctx->trace_flags & STACKSHOT_COLLECT_DELTA_SNAPSHOT) != 0); |
1650 | boolean_t save_owner_info = ((ctx->trace_flags & STACKSHOT_THREAD_WAITINFO) != 0); |
1651 | |
1652 | |
1653 | kern_return_t error = KERN_SUCCESS; |
1654 | mach_vm_address_t out_addr = 0; |
1655 | int saved_count = 0; |
1656 | |
1657 | int task_pid = 0; |
1658 | uint64_t task_uniqueid = 0; |
1659 | int num_delta_thread_snapshots = 0; |
1660 | int num_nonrunnable_threads = 0; |
1661 | int num_waitinfo_threads = 0; |
1662 | |
1663 | uint64_t task_start_abstime = 0; |
1664 | boolean_t task_delta_stackshot = FALSE; |
1665 | boolean_t have_map = FALSE, have_pmap = FALSE; |
1666 | boolean_t some_thread_ran = FALSE; |
1667 | unaligned_u64 *task_snap_ss_flags = NULL; |
1668 | |
1669 | if ((task == NULL) || !ml_validate_nofault((vm_offset_t)task, sizeof(struct task))) { |
1670 | error = KERN_FAILURE; |
1671 | goto error_exit; |
1672 | } |
1673 | |
1674 | have_map = (task->map != NULL) && (ml_validate_nofault((vm_offset_t)(task->map), sizeof(struct _vm_map))); |
1675 | have_pmap = have_map && (task->map->pmap != NULL) && (ml_validate_nofault((vm_offset_t)(task->map->pmap), sizeof(struct pmap))); |
1676 | |
1677 | task_pid = pid_from_task(task); |
1678 | task_uniqueid = get_task_uniqueid(task); |
1679 | |
1680 | if (!task->active || task_is_a_corpse(task)) { |
1681 | /* |
1682 | * Not interested in terminated tasks without threads, and |
1683 | * at the moment, stackshot can't handle a task without a name. |
1684 | */ |
1685 | if (queue_empty(&task->threads) || task_pid == -1) { |
1686 | return KERN_SUCCESS; |
1687 | } |
1688 | } |
1689 | |
1690 | if (collect_delta_stackshot) { |
1691 | proc_starttime_kdp(task->bsd_info, NULL, NULL, &task_start_abstime); |
1692 | } |
1693 | |
1694 | /* Trace everything, unless a process was specified */ |
1695 | if ((ctx->pid == -1) || (ctx->pid == task_pid)) { |
1696 | |
1697 | /* add task snapshot marker */ |
1698 | kcd_exit_on_error(kcdata_add_container_marker(stackshot_kcdata_p, KCDATA_TYPE_CONTAINER_BEGIN, |
1699 | STACKSHOT_KCCONTAINER_TASK, task_uniqueid)); |
1700 | |
1701 | if (!collect_delta_stackshot || (task_start_abstime == 0) || |
1702 | (task_start_abstime > stack_snapshot_delta_since_timestamp)) { |
1703 | kcd_exit_on_error(kcdata_record_task_snapshot(stackshot_kcdata_p, task, ctx->trace_flags, have_pmap, &task_snap_ss_flags)); |
1704 | } else { |
1705 | task_delta_stackshot = TRUE; |
1706 | kcd_exit_on_error(kcdata_record_task_delta_snapshot(stackshot_kcdata_p, task, ctx->trace_flags, have_pmap, &task_snap_ss_flags)); |
1707 | } |
1708 | |
1709 | /* Iterate over task threads */ |
1710 | thread_t thread = THREAD_NULL; |
1711 | queue_iterate(&task->threads, thread, thread_t, task_threads) |
1712 | { |
1713 | uint64_t thread_uniqueid; |
1714 | |
1715 | if ((thread == NULL) || !ml_validate_nofault((vm_offset_t)thread, sizeof(struct thread))) { |
1716 | error = KERN_FAILURE; |
1717 | goto error_exit; |
1718 | } |
1719 | |
1720 | if (active_kthreads_only_p && thread->kernel_stack == 0) |
1721 | continue; |
1722 | |
1723 | thread_uniqueid = thread_tid(thread); |
1724 | |
1725 | boolean_t thread_on_core; |
1726 | enum thread_classification thread_classification = classify_thread(thread, &thread_on_core, ctx->trace_flags); |
1727 | |
1728 | switch (thread_classification) { |
1729 | case tc_full_snapshot: |
1730 | /* add thread marker */ |
1731 | kcd_exit_on_error(kcdata_add_container_marker(stackshot_kcdata_p, KCDATA_TYPE_CONTAINER_BEGIN, |
1732 | STACKSHOT_KCCONTAINER_THREAD, thread_uniqueid)); |
1733 | kcd_exit_on_error( |
1734 | kcdata_record_thread_snapshot(stackshot_kcdata_p, thread, task, ctx->trace_flags, have_pmap, thread_on_core)); |
1735 | |
1736 | /* mark end of thread snapshot data */ |
1737 | kcd_exit_on_error(kcdata_add_container_marker(stackshot_kcdata_p, KCDATA_TYPE_CONTAINER_END, |
1738 | STACKSHOT_KCCONTAINER_THREAD, thread_uniqueid)); |
1739 | |
1740 | some_thread_ran = TRUE; |
1741 | break; |
1742 | |
1743 | case tc_delta_snapshot: |
1744 | num_delta_thread_snapshots++; |
1745 | break; |
1746 | } |
1747 | |
1748 | /* We want to report owner information regardless of whether a thread |
1749 | * has changed since the last delta, whether it's a normal stackshot, |
1750 | * or whether it's nonrunnable */ |
1751 | if (save_owner_info && stackshot_thread_has_valid_waitinfo(thread)) |
1752 | num_waitinfo_threads++; |
1753 | } |
1754 | |
1755 | struct thread_delta_snapshot_v3 * delta_snapshots = NULL; |
1756 | int current_delta_snapshot_index = 0; |
1757 | |
1758 | if (num_delta_thread_snapshots > 0) { |
1759 | kcd_exit_on_error(kcdata_get_memory_addr_for_array(stackshot_kcdata_p, STACKSHOT_KCTYPE_THREAD_DELTA_SNAPSHOT, |
1760 | sizeof(struct thread_delta_snapshot_v3), |
1761 | num_delta_thread_snapshots, &out_addr)); |
1762 | delta_snapshots = (struct thread_delta_snapshot_v3 *)out_addr; |
1763 | } |
1764 | |
1765 | uint64_t * nonrunnable_tids = NULL; |
1766 | |
1767 | if (num_nonrunnable_threads > 0) { |
1768 | kcd_exit_on_error(kcdata_get_memory_addr_for_array(stackshot_kcdata_p, STACKSHOT_KCTYPE_NONRUNNABLE_TIDS, |
1769 | sizeof(uint64_t), num_nonrunnable_threads, &out_addr)); |
1770 | nonrunnable_tids = (uint64_t *)out_addr; |
1771 | } |
1772 | |
1773 | thread_waitinfo_t *thread_waitinfo = NULL; |
1774 | int current_waitinfo_index = 0; |
1775 | |
1776 | if (num_waitinfo_threads > 0) { |
1777 | kcd_exit_on_error(kcdata_get_memory_addr_for_array(stackshot_kcdata_p, STACKSHOT_KCTYPE_THREAD_WAITINFO, |
1778 | sizeof(thread_waitinfo_t), num_waitinfo_threads, &out_addr)); |
1779 | thread_waitinfo = (thread_waitinfo_t *)out_addr; |
1780 | } |
1781 | |
1782 | if (num_delta_thread_snapshots > 0 || num_nonrunnable_threads > 0 || num_waitinfo_threads > 0) { |
1783 | queue_iterate(&task->threads, thread, thread_t, task_threads) |
1784 | { |
1785 | if (active_kthreads_only_p && thread->kernel_stack == 0) |
1786 | continue; |
1787 | |
1788 | /* If we want owner info, we should capture it regardless of its classification */ |
1789 | if (save_owner_info && stackshot_thread_has_valid_waitinfo(thread)) { |
1790 | stackshot_thread_wait_owner_info( |
1791 | thread, |
1792 | &thread_waitinfo[current_waitinfo_index++]); |
1793 | } |
1794 | |
1795 | boolean_t thread_on_core; |
1796 | enum thread_classification thread_classification = classify_thread(thread, &thread_on_core, ctx->trace_flags); |
1797 | |
1798 | switch (thread_classification) { |
1799 | case tc_full_snapshot: |
1800 | /* full thread snapshot captured above */ |
1801 | continue; |
1802 | |
1803 | case tc_delta_snapshot: |
1804 | kcd_exit_on_error(kcdata_record_thread_delta_snapshot(&delta_snapshots[current_delta_snapshot_index++], |
1805 | thread, thread_on_core)); |
1806 | break; |
1807 | } |
1808 | } |
1809 | |
1810 | #if DEBUG || DEVELOPMENT |
1811 | if (current_delta_snapshot_index != num_delta_thread_snapshots) { |
1812 | panic("delta thread snapshot count mismatch while capturing snapshots for task %p. expected %d, found %d" , task, |
1813 | num_delta_thread_snapshots, current_delta_snapshot_index); |
1814 | } |
1815 | if (current_waitinfo_index != num_waitinfo_threads) { |
1816 | panic("thread wait info count mismatch while capturing snapshots for task %p. expected %d, found %d" , task, |
1817 | num_waitinfo_threads, current_waitinfo_index); |
1818 | } |
1819 | #endif |
1820 | } |
1821 | |
1822 | #if IMPORTANCE_INHERITANCE |
1823 | if (save_donating_pids_p) { |
1824 | kcd_exit_on_error( |
1825 | ((((mach_vm_address_t)kcd_end_address(stackshot_kcdata_p) + (TASK_IMP_WALK_LIMIT * sizeof(int32_t))) < |
1826 | (mach_vm_address_t)kcd_max_address(stackshot_kcdata_p)) |
1827 | ? KERN_SUCCESS |
1828 | : KERN_RESOURCE_SHORTAGE)); |
1829 | saved_count = task_importance_list_pids(task, TASK_IMP_LIST_DONATING_PIDS, |
1830 | (void *)kcd_end_address(stackshot_kcdata_p), TASK_IMP_WALK_LIMIT); |
1831 | if (saved_count > 0) |
1832 | kcd_exit_on_error(kcdata_get_memory_addr_for_array(stackshot_kcdata_p, STACKSHOT_KCTYPE_DONATING_PIDS, |
1833 | sizeof(int32_t), saved_count, &out_addr)); |
1834 | } |
1835 | #endif |
1836 | |
1837 | if (!collect_delta_stackshot || (num_delta_thread_snapshots != task->thread_count) || !task_delta_stackshot) { |
1838 | /* |
1839 | * Collect shared cache info and UUID info in these scenarios |
1840 | * 1) a full stackshot |
1841 | * 2) a delta stackshot where the task started after the previous full stackshot OR |
1842 | * any thread from the task has run since the previous full stackshot |
1843 | */ |
1844 | |
1845 | kcd_exit_on_error(kcdata_record_shared_cache_info(stackshot_kcdata_p, task, task_snap_ss_flags)); |
1846 | kcd_exit_on_error(kcdata_record_uuid_info(stackshot_kcdata_p, task, ctx->trace_flags, have_pmap, task_snap_ss_flags)); |
1847 | } |
1848 | /* mark end of task snapshot data */ |
1849 | kcd_exit_on_error(kcdata_add_container_marker(stackshot_kcdata_p, KCDATA_TYPE_CONTAINER_END, STACKSHOT_KCCONTAINER_TASK, |
1850 | task_uniqueid)); |
1851 | } |
1852 | |
1853 | error_exit: |
1854 | return error; |
1855 | } |
1856 | |
1857 | |
1858 | static kern_return_t |
1859 | kdp_stackshot_kcdata_format(int pid, uint32_t trace_flags, uint32_t * pBytesTraced) |
1860 | { |
1861 | kern_return_t error = KERN_SUCCESS; |
1862 | mach_vm_address_t out_addr = 0; |
1863 | uint64_t abs_time = 0, abs_time_end = 0; |
1864 | uint64_t *abs_time_addr = NULL; |
1865 | uint64_t system_state_flags = 0; |
1866 | task_t task = TASK_NULL; |
1867 | mach_timebase_info_data_t timebase = {0, 0}; |
1868 | uint32_t length_to_copy = 0, tmp32 = 0; |
1869 | abs_time = mach_absolute_time(); |
1870 | |
1871 | /* process the flags */ |
1872 | boolean_t collect_delta_stackshot = ((trace_flags & STACKSHOT_COLLECT_DELTA_SNAPSHOT) != 0); |
1873 | boolean_t use_fault_path = ((trace_flags & (STACKSHOT_ENABLE_UUID_FAULTING | STACKSHOT_ENABLE_BT_FAULTING)) != 0); |
1874 | stack_enable_faulting = (trace_flags & (STACKSHOT_ENABLE_BT_FAULTING)); |
1875 | |
1876 | #if CONFIG_EMBEDDED |
1877 | /* KEXTs can't be described by just a base address on embedded */ |
1878 | trace_flags &= ~(STACKSHOT_SAVE_KEXT_LOADINFO); |
1879 | #endif |
1880 | |
1881 | struct stackshot_context ctx = {}; |
1882 | ctx.trace_flags = trace_flags; |
1883 | ctx.pid = pid; |
1884 | |
1885 | if (use_fault_path) { |
1886 | fault_stats.sfs_pages_faulted_in = 0; |
1887 | fault_stats.sfs_time_spent_faulting = 0; |
1888 | fault_stats.sfs_stopped_faulting = (uint8_t) FALSE; |
1889 | } |
1890 | |
1891 | if (sizeof(void *) == 8) |
1892 | system_state_flags |= kKernel64_p; |
1893 | |
1894 | if (stackshot_kcdata_p == NULL || pBytesTraced == NULL) { |
1895 | error = KERN_INVALID_ARGUMENT; |
1896 | goto error_exit; |
1897 | } |
1898 | |
1899 | /* setup mach_absolute_time and timebase info -- copy out in some cases and needed to convert since_timestamp to seconds for proc start time */ |
1900 | clock_timebase_info(&timebase); |
1901 | |
1902 | /* begin saving data into the buffer */ |
1903 | *pBytesTraced = 0; |
1904 | kcd_exit_on_error(kcdata_add_uint32_with_description(stackshot_kcdata_p, trace_flags, "stackshot_in_flags" )); |
1905 | kcd_exit_on_error(kcdata_add_uint32_with_description(stackshot_kcdata_p, (uint32_t)pid, "stackshot_in_pid" )); |
1906 | kcd_exit_on_error(kcdata_add_uint64_with_description(stackshot_kcdata_p, system_state_flags, "system_state_flags" )); |
1907 | |
1908 | #if CONFIG_JETSAM |
1909 | tmp32 = memorystatus_get_pressure_status_kdp(); |
1910 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, STACKSHOT_KCTYPE_JETSAM_LEVEL, sizeof(uint32_t), &out_addr)); |
1911 | stackshot_memcpy((void *)out_addr, &tmp32, sizeof(tmp32)); |
1912 | #endif |
1913 | |
1914 | if (!collect_delta_stackshot) { |
1915 | tmp32 = THREAD_POLICY_INTERNAL_STRUCT_VERSION; |
1916 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, STACKSHOT_KCTYPE_THREAD_POLICY_VERSION, sizeof(uint32_t), &out_addr)); |
1917 | stackshot_memcpy((void *)out_addr, &tmp32, sizeof(tmp32)); |
1918 | |
1919 | tmp32 = PAGE_SIZE; |
1920 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, STACKSHOT_KCTYPE_KERN_PAGE_SIZE, sizeof(uint32_t), &out_addr)); |
1921 | stackshot_memcpy((void *)out_addr, &tmp32, sizeof(tmp32)); |
1922 | |
1923 | /* save boot-args and osversion string */ |
1924 | length_to_copy = MIN((uint32_t)(strlen(version) + 1), OSVERSIZE); |
1925 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, STACKSHOT_KCTYPE_OSVERSION, length_to_copy, &out_addr)); |
1926 | stackshot_strlcpy((char*)out_addr, &version[0], length_to_copy); |
1927 | |
1928 | length_to_copy = MIN((uint32_t)(strlen(PE_boot_args()) + 1), OSVERSIZE); |
1929 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, STACKSHOT_KCTYPE_BOOTARGS, length_to_copy, &out_addr)); |
1930 | stackshot_strlcpy((char*)out_addr, PE_boot_args(), length_to_copy); |
1931 | |
1932 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, KCDATA_TYPE_TIMEBASE, sizeof(timebase), &out_addr)); |
1933 | stackshot_memcpy((void *)out_addr, &timebase, sizeof(timebase)); |
1934 | } else { |
1935 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, STACKSHOT_KCTYPE_DELTA_SINCE_TIMESTAMP, sizeof(uint64_t), &out_addr)); |
1936 | stackshot_memcpy((void*)out_addr, &stack_snapshot_delta_since_timestamp, sizeof(stack_snapshot_delta_since_timestamp)); |
1937 | } |
1938 | |
1939 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, KCDATA_TYPE_MACH_ABSOLUTE_TIME, sizeof(uint64_t), &out_addr)); |
1940 | abs_time_addr = (uint64_t *)out_addr; |
1941 | stackshot_memcpy((void *)abs_time_addr, &abs_time, sizeof(uint64_t)); |
1942 | |
1943 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, KCDATA_TYPE_USECS_SINCE_EPOCH, sizeof(uint64_t), &out_addr)); |
1944 | stackshot_memcpy((void *)out_addr, &stackshot_microsecs, sizeof(uint64_t)); |
1945 | |
1946 | /* record system level shared cache load info (if available) */ |
1947 | if (!collect_delta_stackshot && init_task_shared_region && |
1948 | ml_validate_nofault((vm_offset_t)init_task_shared_region, sizeof(struct vm_shared_region))) { |
1949 | struct dyld_uuid_info_64_v2 *sys_shared_cache_info = NULL; |
1950 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, STACKSHOT_KCTYPE_SHAREDCACHE_LOADINFO, |
1951 | sizeof(struct dyld_uuid_info_64_v2), &out_addr)); |
1952 | sys_shared_cache_info = (struct dyld_uuid_info_64_v2 *)out_addr; |
1953 | |
1954 | stackshot_memcpy(sys_shared_cache_info->imageUUID, &init_task_shared_region->sr_uuid, sizeof(init_task_shared_region->sr_uuid)); |
1955 | sys_shared_cache_info->imageLoadAddress = init_task_shared_region->sr_slide_info.slide; |
1956 | sys_shared_cache_info->imageSlidBaseAddress = init_task_shared_region->sr_slide_info.slide + init_task_shared_region->sr_base_address; |
1957 | |
1958 | if (trace_flags & STACKSHOT_COLLECT_SHAREDCACHE_LAYOUT) { |
1959 | /* |
1960 | * Include a map of the system shared cache layout if it has been populated |
1961 | * (which is only when the system is using a custom shared cache). |
1962 | */ |
1963 | if (init_task_shared_region->sr_images && ml_validate_nofault((vm_offset_t)init_task_shared_region->sr_images, |
1964 | (init_task_shared_region->sr_images_count * sizeof(struct dyld_uuid_info_64)))) { |
1965 | assert(init_task_shared_region->sr_images_count != 0); |
1966 | kcd_exit_on_error(kcdata_get_memory_addr_for_array(stackshot_kcdata_p, STACKSHOT_KCTYPE_SYS_SHAREDCACHE_LAYOUT, |
1967 | sizeof(struct dyld_uuid_info_64), |
1968 | init_task_shared_region->sr_images_count, &out_addr)); |
1969 | stackshot_memcpy((void*)out_addr, init_task_shared_region->sr_images, |
1970 | (init_task_shared_region->sr_images_count * sizeof(struct dyld_uuid_info_64))); |
1971 | } |
1972 | } |
1973 | } |
1974 | |
1975 | /* Add requested information first */ |
1976 | if (trace_flags & STACKSHOT_GET_GLOBAL_MEM_STATS) { |
1977 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, STACKSHOT_KCTYPE_GLOBAL_MEM_STATS, sizeof(struct mem_and_io_snapshot), &out_addr)); |
1978 | kdp_mem_and_io_snapshot((struct mem_and_io_snapshot *)out_addr); |
1979 | } |
1980 | |
1981 | #if CONFIG_COALITIONS |
1982 | int num_coalitions = 0; |
1983 | struct jetsam_coalition_snapshot *coalitions = NULL; |
1984 | /* Iterate over coalitions */ |
1985 | if (trace_flags & STACKSHOT_SAVE_JETSAM_COALITIONS) { |
1986 | if (coalition_iterate_stackshot(stackshot_coalition_jetsam_count, &num_coalitions, COALITION_TYPE_JETSAM) != KERN_SUCCESS) { |
1987 | trace_flags &= ~(STACKSHOT_SAVE_JETSAM_COALITIONS); |
1988 | } |
1989 | } |
1990 | if (trace_flags & STACKSHOT_SAVE_JETSAM_COALITIONS) { |
1991 | if (num_coalitions > 0) { |
1992 | kcd_exit_on_error(kcdata_get_memory_addr_for_array(stackshot_kcdata_p, STACKSHOT_KCTYPE_JETSAM_COALITION_SNAPSHOT, sizeof(struct jetsam_coalition_snapshot), num_coalitions, &out_addr)); |
1993 | coalitions = (struct jetsam_coalition_snapshot*)out_addr; |
1994 | } |
1995 | |
1996 | if (coalition_iterate_stackshot(stackshot_coalition_jetsam_snapshot, coalitions, COALITION_TYPE_JETSAM) != KERN_SUCCESS) { |
1997 | error = KERN_FAILURE; |
1998 | goto error_exit; |
1999 | } |
2000 | |
2001 | } |
2002 | #else |
2003 | trace_flags &= ~(STACKSHOT_SAVE_JETSAM_COALITIONS); |
2004 | #endif /* CONFIG_COALITIONS */ |
2005 | |
2006 | trace_flags &= ~(STACKSHOT_THREAD_GROUP); |
2007 | |
2008 | |
2009 | /* Iterate over tasks */ |
2010 | queue_iterate(&tasks, task, task_t, tasks) |
2011 | { |
2012 | error = kdp_stackshot_record_task(&ctx, task); |
2013 | if (error) |
2014 | goto error_exit; |
2015 | } |
2016 | /* |
2017 | * Iterate over the tasks in the terminated tasks list. We only inspect |
2018 | * tasks that have a valid bsd_info pointer where P_LPEXIT is NOT set. |
2019 | * We're only interested in tasks that have remaining threads (which |
2020 | * could be involved in a deadlock, etc), and the last thread that tears |
2021 | * itself down during exit sets P_LPEXIT during proc_exit(). |
2022 | */ |
2023 | queue_iterate(&terminated_tasks, task, task_t, tasks) |
2024 | { |
2025 | if (task->bsd_info && !proc_in_teardown(task->bsd_info)) { |
2026 | error = kdp_stackshot_record_task(&ctx, task); |
2027 | if (error) |
2028 | goto error_exit; |
2029 | } |
2030 | } |
2031 | |
2032 | if (use_fault_path) { |
2033 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, STACKSHOT_KCTYPE_STACKSHOT_FAULT_STATS, |
2034 | sizeof(struct stackshot_fault_stats), &out_addr)); |
2035 | stackshot_memcpy((void*)out_addr, &fault_stats, sizeof(struct stackshot_fault_stats)); |
2036 | } |
2037 | |
2038 | /* update timestamp of the stackshot */ |
2039 | abs_time_end = mach_absolute_time(); |
2040 | #if DEVELOPMENT || DEBUG |
2041 | kcd_exit_on_error(kcdata_get_memory_addr(stackshot_kcdata_p, STACKSHOT_KCTYPE_STACKSHOT_DURATION, |
2042 | sizeof(struct stackshot_duration), &out_addr)); |
2043 | struct stackshot_duration * stackshot_duration = (struct stackshot_duration *)out_addr; |
2044 | stackshot_duration->stackshot_duration = (abs_time_end - abs_time); |
2045 | stackshot_duration->stackshot_duration_outer = 0; |
2046 | stackshot_duration_outer = (unaligned_u64 *)&stackshot_duration->stackshot_duration_outer; |
2047 | #endif |
2048 | stackshot_memcpy((void *)abs_time_addr, &abs_time_end, sizeof(uint64_t)); |
2049 | |
2050 | kcd_exit_on_error(kcdata_add_uint32_with_description(stackshot_kcdata_p, trace_flags, "stackshot_out_flags" )); |
2051 | |
2052 | kcd_exit_on_error(kcdata_write_buffer_end(stackshot_kcdata_p)); |
2053 | |
2054 | /* === END of populating stackshot data === */ |
2055 | |
2056 | *pBytesTraced = (uint32_t) kcdata_memory_get_used_bytes(stackshot_kcdata_p); |
2057 | error_exit: |
2058 | |
2059 | #if INTERRUPT_MASKED_DEBUG |
2060 | if (!panic_stackshot) { |
2061 | /* |
2062 | * Try to catch instances where stackshot takes too long BEFORE returning from |
2063 | * the debugger |
2064 | */ |
2065 | ml_check_interrupts_disabled_duration(current_thread()); |
2066 | } |
2067 | #endif |
2068 | |
2069 | stack_enable_faulting = FALSE; |
2070 | |
2071 | return error; |
2072 | } |
2073 | |
2074 | static uint64_t |
2075 | proc_was_throttled_from_task(task_t task) |
2076 | { |
2077 | uint64_t was_throttled = 0; |
2078 | |
2079 | if (task->bsd_info) |
2080 | was_throttled = proc_was_throttled(task->bsd_info); |
2081 | |
2082 | return was_throttled; |
2083 | } |
2084 | |
2085 | static uint64_t |
2086 | proc_did_throttle_from_task(task_t task) |
2087 | { |
2088 | uint64_t did_throttle = 0; |
2089 | |
2090 | if (task->bsd_info) |
2091 | did_throttle = proc_did_throttle(task->bsd_info); |
2092 | |
2093 | return did_throttle; |
2094 | } |
2095 | |
2096 | static void |
2097 | kdp_mem_and_io_snapshot(struct mem_and_io_snapshot *memio_snap) |
2098 | { |
2099 | unsigned int pages_reclaimed; |
2100 | unsigned int pages_wanted; |
2101 | kern_return_t kErr; |
2102 | |
2103 | processor_t processor; |
2104 | vm_statistics64_t stat; |
2105 | vm_statistics64_data_t host_vm_stat; |
2106 | |
2107 | processor = processor_list; |
2108 | stat = &PROCESSOR_DATA(processor, vm_stat); |
2109 | host_vm_stat = *stat; |
2110 | |
2111 | if (processor_count > 1) { |
2112 | /* |
2113 | * processor_list may be in the process of changing as we are |
2114 | * attempting a stackshot. Ordinarily it will be lock protected, |
2115 | * but it is not safe to lock in the context of the debugger. |
2116 | * Fortunately we never remove elements from the processor list, |
2117 | * and only add to to the end of the list, so we SHOULD be able |
2118 | * to walk it. If we ever want to truly tear down processors, |
2119 | * this will have to change. |
2120 | */ |
2121 | while ((processor = processor->processor_list) != NULL) { |
2122 | stat = &PROCESSOR_DATA(processor, vm_stat); |
2123 | host_vm_stat.compressions += stat->compressions; |
2124 | host_vm_stat.decompressions += stat->decompressions; |
2125 | } |
2126 | } |
2127 | |
2128 | memio_snap->snapshot_magic = STACKSHOT_MEM_AND_IO_SNAPSHOT_MAGIC; |
2129 | memio_snap->free_pages = vm_page_free_count; |
2130 | memio_snap->active_pages = vm_page_active_count; |
2131 | memio_snap->inactive_pages = vm_page_inactive_count; |
2132 | memio_snap->purgeable_pages = vm_page_purgeable_count; |
2133 | memio_snap->wired_pages = vm_page_wire_count; |
2134 | memio_snap->speculative_pages = vm_page_speculative_count; |
2135 | memio_snap->throttled_pages = vm_page_throttled_count; |
2136 | memio_snap->busy_buffer_count = count_busy_buffers(); |
2137 | memio_snap->filebacked_pages = vm_page_pageable_external_count; |
2138 | memio_snap->compressions = (uint32_t)host_vm_stat.compressions; |
2139 | memio_snap->decompressions = (uint32_t)host_vm_stat.decompressions; |
2140 | memio_snap->compressor_size = VM_PAGE_COMPRESSOR_COUNT; |
2141 | kErr = mach_vm_pressure_monitor(FALSE, VM_PRESSURE_TIME_WINDOW, &pages_reclaimed, &pages_wanted); |
2142 | |
2143 | if ( ! kErr ) { |
2144 | memio_snap->pages_wanted = (uint32_t)pages_wanted; |
2145 | memio_snap->pages_reclaimed = (uint32_t)pages_reclaimed; |
2146 | memio_snap->pages_wanted_reclaimed_valid = 1; |
2147 | } else { |
2148 | memio_snap->pages_wanted = 0; |
2149 | memio_snap->pages_reclaimed = 0; |
2150 | memio_snap->pages_wanted_reclaimed_valid = 0; |
2151 | } |
2152 | } |
2153 | |
2154 | void |
2155 | stackshot_memcpy(void *dst, const void *src, size_t len) |
2156 | { |
2157 | #if CONFIG_EMBEDDED |
2158 | if (panic_stackshot) { |
2159 | uint8_t *dest_bytes = (uint8_t *)dst; |
2160 | const uint8_t *src_bytes = (const uint8_t *)src; |
2161 | for (size_t i = 0; i < len; i++) { |
2162 | dest_bytes[i] = src_bytes[i]; |
2163 | } |
2164 | } else |
2165 | #endif |
2166 | memcpy(dst, src, len); |
2167 | } |
2168 | |
2169 | size_t |
2170 | stackshot_strlcpy(char *dst, const char *src, size_t maxlen) |
2171 | { |
2172 | const size_t srclen = strlen(src); |
2173 | |
2174 | if (srclen < maxlen) { |
2175 | stackshot_memcpy(dst, src, srclen+1); |
2176 | } else if (maxlen != 0) { |
2177 | stackshot_memcpy(dst, src, maxlen-1); |
2178 | dst[maxlen-1] = '\0'; |
2179 | } |
2180 | |
2181 | return srclen; |
2182 | } |
2183 | |
2184 | |
2185 | /* |
2186 | * Returns the physical address of the specified map:target address, |
2187 | * using the kdp fault path if requested and the page is not resident. |
2188 | */ |
2189 | vm_offset_t |
2190 | kdp_find_phys(vm_map_t map, vm_offset_t target_addr, boolean_t try_fault, uint32_t *kdp_fault_results) |
2191 | { |
2192 | vm_offset_t cur_phys_addr; |
2193 | unsigned cur_wimg_bits; |
2194 | uint64_t fault_start_time = 0; |
2195 | |
2196 | if (map == VM_MAP_NULL) { |
2197 | return 0; |
2198 | } |
2199 | |
2200 | cur_phys_addr = kdp_vtophys(map->pmap, target_addr); |
2201 | if (!pmap_valid_page((ppnum_t) atop(cur_phys_addr))) { |
2202 | if (!try_fault || fault_stats.sfs_stopped_faulting) { |
2203 | if (kdp_fault_results) |
2204 | *kdp_fault_results |= KDP_FAULT_RESULT_PAGED_OUT; |
2205 | |
2206 | return 0; |
2207 | } |
2208 | |
2209 | /* |
2210 | * The pmap doesn't have a valid page so we start at the top level |
2211 | * vm map and try a lightweight fault. Update fault path usage stats. |
2212 | */ |
2213 | fault_start_time = mach_absolute_time(); |
2214 | cur_phys_addr = kdp_lightweight_fault(map, (target_addr & ~PAGE_MASK)); |
2215 | fault_stats.sfs_time_spent_faulting += (mach_absolute_time() - fault_start_time); |
2216 | |
2217 | if ((fault_stats.sfs_time_spent_faulting >= fault_stats.sfs_system_max_fault_time) && !panic_stackshot) { |
2218 | fault_stats.sfs_stopped_faulting = (uint8_t) TRUE; |
2219 | } |
2220 | |
2221 | cur_phys_addr += (target_addr & PAGE_MASK); |
2222 | |
2223 | if (!pmap_valid_page((ppnum_t) atop(cur_phys_addr))) { |
2224 | if (kdp_fault_results) |
2225 | *kdp_fault_results |= (KDP_FAULT_RESULT_TRIED_FAULT | KDP_FAULT_RESULT_PAGED_OUT); |
2226 | |
2227 | return 0; |
2228 | } |
2229 | |
2230 | if (kdp_fault_results) |
2231 | *kdp_fault_results |= KDP_FAULT_RESULT_FAULTED_IN; |
2232 | |
2233 | fault_stats.sfs_pages_faulted_in++; |
2234 | } else { |
2235 | /* |
2236 | * This check is done in kdp_lightweight_fault for the fault path. |
2237 | */ |
2238 | cur_wimg_bits = pmap_cache_attributes((ppnum_t) atop(cur_phys_addr)); |
2239 | |
2240 | if ((cur_wimg_bits & VM_WIMG_MASK) != VM_WIMG_DEFAULT) { |
2241 | return 0; |
2242 | } |
2243 | } |
2244 | |
2245 | return cur_phys_addr; |
2246 | } |
2247 | |
2248 | boolean_t |
2249 | kdp_copyin_word( |
2250 | task_t task, uint64_t addr, uint64_t *result, boolean_t try_fault, uint32_t *kdp_fault_results) |
2251 | { |
2252 | if (task_has_64Bit_data(task)) { |
2253 | return kdp_copyin(task->map, addr, result, sizeof(uint64_t), try_fault, kdp_fault_results); |
2254 | } else { |
2255 | uint32_t buf; |
2256 | boolean_t r = kdp_copyin(task->map, addr, &buf, sizeof(uint32_t), try_fault, kdp_fault_results); |
2257 | *result = buf; |
2258 | return r; |
2259 | } |
2260 | } |
2261 | |
2262 | boolean_t |
2263 | kdp_copyin(vm_map_t map, uint64_t uaddr, void *dest, size_t size, boolean_t try_fault, uint32_t *kdp_fault_results) |
2264 | { |
2265 | size_t rem = size; |
2266 | char *kvaddr = dest; |
2267 | |
2268 | #if CONFIG_EMBEDDED |
2269 | /* Identify if destination buffer is in panic storage area */ |
2270 | if (panic_stackshot && ((vm_offset_t)dest >= gPanicBase) && ((vm_offset_t)dest < (gPanicBase + gPanicSize))) { |
2271 | if (((vm_offset_t)dest + size) > (gPanicBase + gPanicSize)) { |
2272 | return FALSE; |
2273 | } |
2274 | } |
2275 | #endif |
2276 | |
2277 | while (rem) { |
2278 | uint64_t phys_src = kdp_find_phys(map, uaddr, try_fault, kdp_fault_results); |
2279 | uint64_t phys_dest = kvtophys((vm_offset_t)kvaddr); |
2280 | uint64_t src_rem = PAGE_SIZE - (phys_src & PAGE_MASK); |
2281 | uint64_t dst_rem = PAGE_SIZE - (phys_dest & PAGE_MASK); |
2282 | size_t cur_size = (uint32_t) MIN(src_rem, dst_rem); |
2283 | cur_size = MIN(cur_size, rem); |
2284 | |
2285 | if (phys_src && phys_dest) { |
2286 | #if CONFIG_EMBEDDED |
2287 | /* |
2288 | * On embedded the panic buffer is mapped as device memory and doesn't allow |
2289 | * unaligned accesses. To prevent these, we copy over bytes individually here. |
2290 | */ |
2291 | if (panic_stackshot) |
2292 | stackshot_memcpy(kvaddr, (const void *)phystokv(phys_src), cur_size); |
2293 | else |
2294 | #endif /* CONFIG_EMBEDDED */ |
2295 | bcopy_phys(phys_src, phys_dest, cur_size); |
2296 | } else { |
2297 | break; |
2298 | } |
2299 | |
2300 | uaddr += cur_size; |
2301 | kvaddr += cur_size; |
2302 | rem -= cur_size; |
2303 | } |
2304 | |
2305 | return (rem == 0); |
2306 | } |
2307 | |
2308 | kern_return_t |
2309 | do_stackshot(void *context) |
2310 | { |
2311 | #pragma unused(context) |
2312 | kdp_snapshot++; |
2313 | |
2314 | stack_snapshot_ret = kdp_stackshot_kcdata_format(stack_snapshot_pid, |
2315 | stack_snapshot_flags, |
2316 | &stack_snapshot_bytes_traced); |
2317 | |
2318 | kdp_snapshot--; |
2319 | return stack_snapshot_ret; |
2320 | } |
2321 | |
2322 | /* |
2323 | * A fantastical routine that tries to be fast about returning |
2324 | * translations. Caches the last page we found a translation |
2325 | * for, so that we can be quick about multiple queries to the |
2326 | * same page. It turns out this is exactly the workflow |
2327 | * machine_trace_thread and its relatives tend to throw at us. |
2328 | * |
2329 | * Please zero the nasty global this uses after a bulk lookup; |
2330 | * this isn't safe across a switch of the map or changes |
2331 | * to a pmap. |
2332 | * |
2333 | * This also means that if zero is a valid KVA, we are |
2334 | * screwed. Sucks to be us. Fortunately, this should never |
2335 | * happen. |
2336 | */ |
2337 | vm_offset_t |
2338 | machine_trace_thread_get_kva(vm_offset_t cur_target_addr, vm_map_t map, uint32_t *thread_trace_flags) |
2339 | { |
2340 | vm_offset_t cur_target_page; |
2341 | vm_offset_t cur_phys_addr; |
2342 | vm_offset_t kern_virt_target_addr; |
2343 | uint32_t kdp_fault_results = 0; |
2344 | |
2345 | cur_target_page = atop(cur_target_addr); |
2346 | |
2347 | if ((cur_target_page != prev_target_page) || validate_next_addr) { |
2348 | |
2349 | /* |
2350 | * Alright; it wasn't our previous page. So |
2351 | * we must validate that there is a page |
2352 | * table entry for this address under the |
2353 | * current pmap, and that it has default |
2354 | * cache attributes (otherwise it may not be |
2355 | * safe to access it). |
2356 | */ |
2357 | cur_phys_addr = kdp_find_phys(map, cur_target_addr, stack_enable_faulting, &kdp_fault_results); |
2358 | if (thread_trace_flags) { |
2359 | if (kdp_fault_results & KDP_FAULT_RESULT_PAGED_OUT) { |
2360 | *thread_trace_flags |= kThreadTruncatedBT; |
2361 | } |
2362 | |
2363 | if (kdp_fault_results & KDP_FAULT_RESULT_TRIED_FAULT) { |
2364 | *thread_trace_flags |= kThreadTriedFaultBT; |
2365 | } |
2366 | |
2367 | if (kdp_fault_results & KDP_FAULT_RESULT_FAULTED_IN) { |
2368 | *thread_trace_flags |= kThreadFaultedBT; |
2369 | } |
2370 | } |
2371 | |
2372 | if (cur_phys_addr == 0) { |
2373 | return 0; |
2374 | } |
2375 | #if __x86_64__ |
2376 | kern_virt_target_addr = (vm_offset_t) PHYSMAP_PTOV(cur_phys_addr); |
2377 | #elif __arm__ || __arm64__ |
2378 | kern_virt_target_addr = phystokv(cur_phys_addr); |
2379 | #else |
2380 | #error Oh come on... we should really unify the physical -> kernel virtual interface |
2381 | #endif |
2382 | prev_target_page = cur_target_page; |
2383 | prev_target_kva = (kern_virt_target_addr & ~PAGE_MASK); |
2384 | validate_next_addr = FALSE; |
2385 | } else { |
2386 | /* We found a translation, so stash this page */ |
2387 | kern_virt_target_addr = prev_target_kva + (cur_target_addr & PAGE_MASK); |
2388 | } |
2389 | |
2390 | #if KASAN |
2391 | kasan_notify_address(kern_virt_target_addr, sizeof(uint64_t)); |
2392 | #endif |
2393 | return kern_virt_target_addr; |
2394 | } |
2395 | |
2396 | void |
2397 | machine_trace_thread_clear_validation_cache(void) |
2398 | { |
2399 | validate_next_addr = TRUE; |
2400 | } |
2401 | |
2402 | boolean_t |
2403 | stackshot_thread_is_idle_worker_unsafe(thread_t thread) |
2404 | { |
2405 | /* When the pthread kext puts a worker thread to sleep, it will |
2406 | * set kThreadWaitParkedWorkQueue in the block_hint of the thread |
2407 | * struct. See parkit() in kern/kern_support.c in libpthread. |
2408 | */ |
2409 | return (thread->state & TH_WAIT) && |
2410 | (thread->block_hint == kThreadWaitParkedWorkQueue); |
2411 | } |
2412 | |
2413 | #if CONFIG_COALITIONS |
2414 | static void |
2415 | stackshot_coalition_jetsam_count(void *arg, int i, coalition_t coal) |
2416 | { |
2417 | #pragma unused(i, coal) |
2418 | unsigned int *coalition_count = (unsigned int*)arg; |
2419 | (*coalition_count)++; |
2420 | } |
2421 | |
2422 | static void |
2423 | stackshot_coalition_jetsam_snapshot(void *arg, int i, coalition_t coal) |
2424 | { |
2425 | if (coalition_type(coal) != COALITION_TYPE_JETSAM) |
2426 | return; |
2427 | |
2428 | struct jetsam_coalition_snapshot *coalitions = (struct jetsam_coalition_snapshot*)arg; |
2429 | struct jetsam_coalition_snapshot *jcs = &coalitions[i]; |
2430 | task_t leader = TASK_NULL; |
2431 | jcs->jcs_id = coalition_id(coal); |
2432 | jcs->jcs_flags = 0; |
2433 | |
2434 | if (coalition_term_requested(coal)) |
2435 | jcs->jcs_flags |= kCoalitionTermRequested; |
2436 | if (coalition_is_terminated(coal)) |
2437 | jcs->jcs_flags |= kCoalitionTerminated; |
2438 | if (coalition_is_reaped(coal)) |
2439 | jcs->jcs_flags |= kCoalitionReaped; |
2440 | if (coalition_is_privileged(coal)) |
2441 | jcs->jcs_flags |= kCoalitionPrivileged; |
2442 | |
2443 | |
2444 | leader = kdp_coalition_get_leader(coal); |
2445 | if (leader) |
2446 | jcs->jcs_leader_task_uniqueid = get_task_uniqueid(leader); |
2447 | else |
2448 | jcs->jcs_leader_task_uniqueid = 0; |
2449 | } |
2450 | #endif /* CONFIG_COALITIONS */ |
2451 | |
2452 | |
2453 | /* Determine if a thread has waitinfo that stackshot can provide */ |
2454 | static int |
2455 | stackshot_thread_has_valid_waitinfo(thread_t thread) |
2456 | { |
2457 | if (!(thread->state & TH_WAIT)) |
2458 | return 0; |
2459 | |
2460 | switch (thread->block_hint) { |
2461 | // If set to None or is a parked work queue, ignore it |
2462 | case kThreadWaitParkedWorkQueue: |
2463 | case kThreadWaitNone: |
2464 | return 0; |
2465 | // There is a short window where the pthread kext removes a thread |
2466 | // from its ksyn wait queue before waking the thread up |
2467 | case kThreadWaitPThreadMutex: |
2468 | case kThreadWaitPThreadRWLockRead: |
2469 | case kThreadWaitPThreadRWLockWrite: |
2470 | case kThreadWaitPThreadCondVar: |
2471 | return (kdp_pthread_get_thread_kwq(thread) != NULL); |
2472 | // All other cases are valid block hints if in a wait state |
2473 | default: |
2474 | return 1; |
2475 | } |
2476 | } |
2477 | |
2478 | static void |
2479 | stackshot_thread_wait_owner_info(thread_t thread, thread_waitinfo_t *waitinfo) |
2480 | { |
2481 | waitinfo->waiter = thread_tid(thread); |
2482 | waitinfo->wait_type = thread->block_hint; |
2483 | switch (waitinfo->wait_type) { |
2484 | case kThreadWaitKernelMutex: |
2485 | kdp_lck_mtx_find_owner(thread->waitq, thread->wait_event, waitinfo); |
2486 | break; |
2487 | case kThreadWaitPortReceive: |
2488 | kdp_mqueue_recv_find_owner(thread->waitq, thread->wait_event, waitinfo); |
2489 | break; |
2490 | case kThreadWaitPortSend: |
2491 | kdp_mqueue_send_find_owner(thread->waitq, thread->wait_event, waitinfo); |
2492 | break; |
2493 | case kThreadWaitSemaphore: |
2494 | kdp_sema_find_owner(thread->waitq, thread->wait_event, waitinfo); |
2495 | break; |
2496 | case kThreadWaitUserLock: |
2497 | kdp_ulock_find_owner(thread->waitq, thread->wait_event, waitinfo); |
2498 | break; |
2499 | case kThreadWaitKernelRWLockRead: |
2500 | case kThreadWaitKernelRWLockWrite: |
2501 | case kThreadWaitKernelRWLockUpgrade: |
2502 | kdp_rwlck_find_owner(thread->waitq, thread->wait_event, waitinfo); |
2503 | break; |
2504 | case kThreadWaitPThreadMutex: |
2505 | case kThreadWaitPThreadRWLockRead: |
2506 | case kThreadWaitPThreadRWLockWrite: |
2507 | case kThreadWaitPThreadCondVar: |
2508 | kdp_pthread_find_owner(thread, waitinfo); |
2509 | break; |
2510 | case kThreadWaitWorkloopSyncWait: |
2511 | kdp_workloop_sync_wait_find_owner(thread, thread->wait_event, waitinfo); |
2512 | break; |
2513 | case kThreadWaitOnProcess: |
2514 | kdp_wait4_find_process(thread, thread->wait_event, waitinfo); |
2515 | break; |
2516 | default: |
2517 | waitinfo->owner = 0; |
2518 | waitinfo->context = 0; |
2519 | break; |
2520 | } |
2521 | } |
2522 | |
2523 | |