1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * event tracer 4 * 5 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com> 6 * 7 * - Added format output of fields of the trace point. 8 * This was based off of work by Tom Zanussi <tzanussi@gmail.com>. 9 * 10 */ 11 12 #define pr_fmt(fmt) fmt 13 14 #include <linux/workqueue.h> 15 #include <linux/security.h> 16 #include <linux/spinlock.h> 17 #include <linux/kthread.h> 18 #include <linux/tracefs.h> 19 #include <linux/uaccess.h> 20 #include <linux/module.h> 21 #include <linux/ctype.h> 22 #include <linux/sort.h> 23 #include <linux/slab.h> 24 #include <linux/delay.h> 25 26 #include <trace/events/sched.h> 27 #include <trace/syscall.h> 28 29 #include <asm/setup.h> 30 31 #include "trace_output.h" 32 33 #undef TRACE_SYSTEM 34 #define TRACE_SYSTEM "TRACE_SYSTEM" 35 36 DEFINE_MUTEX(event_mutex); 37 38 LIST_HEAD(ftrace_events); 39 static LIST_HEAD(ftrace_generic_fields); 40 static LIST_HEAD(ftrace_common_fields); 41 static bool eventdir_initialized; 42 43 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO) 44 45 static struct kmem_cache *field_cachep; 46 static struct kmem_cache *file_cachep; 47 48 static inline int system_refcount(struct event_subsystem *system) 49 { 50 return system->ref_count; 51 } 52 53 static int system_refcount_inc(struct event_subsystem *system) 54 { 55 return system->ref_count++; 56 } 57 58 static int system_refcount_dec(struct event_subsystem *system) 59 { 60 return --system->ref_count; 61 } 62 63 /* Double loops, do not use break, only goto's work */ 64 #define do_for_each_event_file(tr, file) \ 65 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \ 66 list_for_each_entry(file, &tr->events, list) 67 68 #define do_for_each_event_file_safe(tr, file) \ 69 list_for_each_entry(tr, &ftrace_trace_arrays, list) { \ 70 struct trace_event_file *___n; \ 71 list_for_each_entry_safe(file, ___n, &tr->events, list) 72 73 #define while_for_each_event_file() \ 74 } 75 76 static struct ftrace_event_field * 77 __find_event_field(struct list_head *head, char *name) 78 { 79 struct ftrace_event_field *field; 80 81 list_for_each_entry(field, head, link) { 82 if (!strcmp(field->name, name)) 83 return field; 84 } 85 86 return NULL; 87 } 88 89 struct ftrace_event_field * 90 trace_find_event_field(struct trace_event_call *call, char *name) 91 { 92 struct ftrace_event_field *field; 93 struct list_head *head; 94 95 head = trace_get_fields(call); 96 field = __find_event_field(head, name); 97 if (field) 98 return field; 99 100 field = __find_event_field(&ftrace_generic_fields, name); 101 if (field) 102 return field; 103 104 return __find_event_field(&ftrace_common_fields, name); 105 } 106 107 static int __trace_define_field(struct list_head *head, const char *type, 108 const char *name, int offset, int size, 109 int is_signed, int filter_type) 110 { 111 struct ftrace_event_field *field; 112 113 field = kmem_cache_alloc(field_cachep, GFP_TRACE); 114 if (!field) 115 return -ENOMEM; 116 117 field->name = name; 118 field->type = type; 119 120 if (filter_type == FILTER_OTHER) 121 field->filter_type = filter_assign_type(type); 122 else 123 field->filter_type = filter_type; 124 125 field->offset = offset; 126 field->size = size; 127 field->is_signed = is_signed; 128 129 list_add(&field->link, head); 130 131 return 0; 132 } 133 134 int trace_define_field(struct trace_event_call *call, const char *type, 135 const char *name, int offset, int size, int is_signed, 136 int filter_type) 137 { 138 struct list_head *head; 139 140 if (WARN_ON(!call->class)) 141 return 0; 142 143 head = trace_get_fields(call); 144 return __trace_define_field(head, type, name, offset, size, 145 is_signed, filter_type); 146 } 147 EXPORT_SYMBOL_GPL(trace_define_field); 148 149 #define __generic_field(type, item, filter_type) \ 150 ret = __trace_define_field(&ftrace_generic_fields, #type, \ 151 #item, 0, 0, is_signed_type(type), \ 152 filter_type); \ 153 if (ret) \ 154 return ret; 155 156 #define __common_field(type, item) \ 157 ret = __trace_define_field(&ftrace_common_fields, #type, \ 158 "common_" #item, \ 159 offsetof(typeof(ent), item), \ 160 sizeof(ent.item), \ 161 is_signed_type(type), FILTER_OTHER); \ 162 if (ret) \ 163 return ret; 164 165 static int trace_define_generic_fields(void) 166 { 167 int ret; 168 169 __generic_field(int, CPU, FILTER_CPU); 170 __generic_field(int, cpu, FILTER_CPU); 171 __generic_field(char *, COMM, FILTER_COMM); 172 __generic_field(char *, comm, FILTER_COMM); 173 174 return ret; 175 } 176 177 static int trace_define_common_fields(void) 178 { 179 int ret; 180 struct trace_entry ent; 181 182 __common_field(unsigned short, type); 183 __common_field(unsigned char, flags); 184 __common_field(unsigned char, preempt_count); 185 __common_field(int, pid); 186 187 return ret; 188 } 189 190 static void trace_destroy_fields(struct trace_event_call *call) 191 { 192 struct ftrace_event_field *field, *next; 193 struct list_head *head; 194 195 head = trace_get_fields(call); 196 list_for_each_entry_safe(field, next, head, link) { 197 list_del(&field->link); 198 kmem_cache_free(field_cachep, field); 199 } 200 } 201 202 /* 203 * run-time version of trace_event_get_offsets_<call>() that returns the last 204 * accessible offset of trace fields excluding __dynamic_array bytes 205 */ 206 int trace_event_get_offsets(struct trace_event_call *call) 207 { 208 struct ftrace_event_field *tail; 209 struct list_head *head; 210 211 head = trace_get_fields(call); 212 /* 213 * head->next points to the last field with the largest offset, 214 * since it was added last by trace_define_field() 215 */ 216 tail = list_first_entry(head, struct ftrace_event_field, link); 217 return tail->offset + tail->size; 218 } 219 220 int trace_event_raw_init(struct trace_event_call *call) 221 { 222 int id; 223 224 id = register_trace_event(&call->event); 225 if (!id) 226 return -ENODEV; 227 228 return 0; 229 } 230 EXPORT_SYMBOL_GPL(trace_event_raw_init); 231 232 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file) 233 { 234 struct trace_array *tr = trace_file->tr; 235 struct trace_array_cpu *data; 236 struct trace_pid_list *no_pid_list; 237 struct trace_pid_list *pid_list; 238 239 pid_list = rcu_dereference_raw(tr->filtered_pids); 240 no_pid_list = rcu_dereference_raw(tr->filtered_no_pids); 241 242 if (!pid_list && !no_pid_list) 243 return false; 244 245 data = this_cpu_ptr(tr->array_buffer.data); 246 247 return data->ignore_pid; 248 } 249 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid); 250 251 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer, 252 struct trace_event_file *trace_file, 253 unsigned long len) 254 { 255 struct trace_event_call *event_call = trace_file->event_call; 256 257 if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) && 258 trace_event_ignore_this_pid(trace_file)) 259 return NULL; 260 261 /* 262 * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables 263 * preemption (adding one to the preempt_count). Since we are 264 * interested in the preempt_count at the time the tracepoint was 265 * hit, we need to subtract one to offset the increment. 266 */ 267 fbuffer->trace_ctx = tracing_gen_ctx_dec(); 268 fbuffer->trace_file = trace_file; 269 270 fbuffer->event = 271 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file, 272 event_call->event.type, len, 273 fbuffer->trace_ctx); 274 if (!fbuffer->event) 275 return NULL; 276 277 fbuffer->regs = NULL; 278 fbuffer->entry = ring_buffer_event_data(fbuffer->event); 279 return fbuffer->entry; 280 } 281 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve); 282 283 int trace_event_reg(struct trace_event_call *call, 284 enum trace_reg type, void *data) 285 { 286 struct trace_event_file *file = data; 287 288 WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT)); 289 switch (type) { 290 case TRACE_REG_REGISTER: 291 return tracepoint_probe_register(call->tp, 292 call->class->probe, 293 file); 294 case TRACE_REG_UNREGISTER: 295 tracepoint_probe_unregister(call->tp, 296 call->class->probe, 297 file); 298 return 0; 299 300 #ifdef CONFIG_PERF_EVENTS 301 case TRACE_REG_PERF_REGISTER: 302 return tracepoint_probe_register(call->tp, 303 call->class->perf_probe, 304 call); 305 case TRACE_REG_PERF_UNREGISTER: 306 tracepoint_probe_unregister(call->tp, 307 call->class->perf_probe, 308 call); 309 return 0; 310 case TRACE_REG_PERF_OPEN: 311 case TRACE_REG_PERF_CLOSE: 312 case TRACE_REG_PERF_ADD: 313 case TRACE_REG_PERF_DEL: 314 return 0; 315 #endif 316 } 317 return 0; 318 } 319 EXPORT_SYMBOL_GPL(trace_event_reg); 320 321 void trace_event_enable_cmd_record(bool enable) 322 { 323 struct trace_event_file *file; 324 struct trace_array *tr; 325 326 lockdep_assert_held(&event_mutex); 327 328 do_for_each_event_file(tr, file) { 329 330 if (!(file->flags & EVENT_FILE_FL_ENABLED)) 331 continue; 332 333 if (enable) { 334 tracing_start_cmdline_record(); 335 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags); 336 } else { 337 tracing_stop_cmdline_record(); 338 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags); 339 } 340 } while_for_each_event_file(); 341 } 342 343 void trace_event_enable_tgid_record(bool enable) 344 { 345 struct trace_event_file *file; 346 struct trace_array *tr; 347 348 lockdep_assert_held(&event_mutex); 349 350 do_for_each_event_file(tr, file) { 351 if (!(file->flags & EVENT_FILE_FL_ENABLED)) 352 continue; 353 354 if (enable) { 355 tracing_start_tgid_record(); 356 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags); 357 } else { 358 tracing_stop_tgid_record(); 359 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, 360 &file->flags); 361 } 362 } while_for_each_event_file(); 363 } 364 365 static int __ftrace_event_enable_disable(struct trace_event_file *file, 366 int enable, int soft_disable) 367 { 368 struct trace_event_call *call = file->event_call; 369 struct trace_array *tr = file->tr; 370 unsigned long file_flags = file->flags; 371 int ret = 0; 372 int disable; 373 374 switch (enable) { 375 case 0: 376 /* 377 * When soft_disable is set and enable is cleared, the sm_ref 378 * reference counter is decremented. If it reaches 0, we want 379 * to clear the SOFT_DISABLED flag but leave the event in the 380 * state that it was. That is, if the event was enabled and 381 * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED 382 * is set we do not want the event to be enabled before we 383 * clear the bit. 384 * 385 * When soft_disable is not set but the SOFT_MODE flag is, 386 * we do nothing. Do not disable the tracepoint, otherwise 387 * "soft enable"s (clearing the SOFT_DISABLED bit) wont work. 388 */ 389 if (soft_disable) { 390 if (atomic_dec_return(&file->sm_ref) > 0) 391 break; 392 disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED; 393 clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags); 394 } else 395 disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE); 396 397 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) { 398 clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags); 399 if (file->flags & EVENT_FILE_FL_RECORDED_CMD) { 400 tracing_stop_cmdline_record(); 401 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags); 402 } 403 404 if (file->flags & EVENT_FILE_FL_RECORDED_TGID) { 405 tracing_stop_tgid_record(); 406 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags); 407 } 408 409 call->class->reg(call, TRACE_REG_UNREGISTER, file); 410 } 411 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */ 412 if (file->flags & EVENT_FILE_FL_SOFT_MODE) 413 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags); 414 else 415 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags); 416 break; 417 case 1: 418 /* 419 * When soft_disable is set and enable is set, we want to 420 * register the tracepoint for the event, but leave the event 421 * as is. That means, if the event was already enabled, we do 422 * nothing (but set SOFT_MODE). If the event is disabled, we 423 * set SOFT_DISABLED before enabling the event tracepoint, so 424 * it still seems to be disabled. 425 */ 426 if (!soft_disable) 427 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags); 428 else { 429 if (atomic_inc_return(&file->sm_ref) > 1) 430 break; 431 set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags); 432 } 433 434 if (!(file->flags & EVENT_FILE_FL_ENABLED)) { 435 bool cmd = false, tgid = false; 436 437 /* Keep the event disabled, when going to SOFT_MODE. */ 438 if (soft_disable) 439 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags); 440 441 if (tr->trace_flags & TRACE_ITER_RECORD_CMD) { 442 cmd = true; 443 tracing_start_cmdline_record(); 444 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags); 445 } 446 447 if (tr->trace_flags & TRACE_ITER_RECORD_TGID) { 448 tgid = true; 449 tracing_start_tgid_record(); 450 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags); 451 } 452 453 ret = call->class->reg(call, TRACE_REG_REGISTER, file); 454 if (ret) { 455 if (cmd) 456 tracing_stop_cmdline_record(); 457 if (tgid) 458 tracing_stop_tgid_record(); 459 pr_info("event trace: Could not enable event " 460 "%s\n", trace_event_name(call)); 461 break; 462 } 463 set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags); 464 465 /* WAS_ENABLED gets set but never cleared. */ 466 set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags); 467 } 468 break; 469 } 470 471 /* Enable or disable use of trace_buffered_event */ 472 if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) != 473 (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) { 474 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED) 475 trace_buffered_event_enable(); 476 else 477 trace_buffered_event_disable(); 478 } 479 480 return ret; 481 } 482 483 int trace_event_enable_disable(struct trace_event_file *file, 484 int enable, int soft_disable) 485 { 486 return __ftrace_event_enable_disable(file, enable, soft_disable); 487 } 488 489 static int ftrace_event_enable_disable(struct trace_event_file *file, 490 int enable) 491 { 492 return __ftrace_event_enable_disable(file, enable, 0); 493 } 494 495 static void ftrace_clear_events(struct trace_array *tr) 496 { 497 struct trace_event_file *file; 498 499 mutex_lock(&event_mutex); 500 list_for_each_entry(file, &tr->events, list) { 501 ftrace_event_enable_disable(file, 0); 502 } 503 mutex_unlock(&event_mutex); 504 } 505 506 static void 507 event_filter_pid_sched_process_exit(void *data, struct task_struct *task) 508 { 509 struct trace_pid_list *pid_list; 510 struct trace_array *tr = data; 511 512 pid_list = rcu_dereference_raw(tr->filtered_pids); 513 trace_filter_add_remove_task(pid_list, NULL, task); 514 515 pid_list = rcu_dereference_raw(tr->filtered_no_pids); 516 trace_filter_add_remove_task(pid_list, NULL, task); 517 } 518 519 static void 520 event_filter_pid_sched_process_fork(void *data, 521 struct task_struct *self, 522 struct task_struct *task) 523 { 524 struct trace_pid_list *pid_list; 525 struct trace_array *tr = data; 526 527 pid_list = rcu_dereference_sched(tr->filtered_pids); 528 trace_filter_add_remove_task(pid_list, self, task); 529 530 pid_list = rcu_dereference_sched(tr->filtered_no_pids); 531 trace_filter_add_remove_task(pid_list, self, task); 532 } 533 534 void trace_event_follow_fork(struct trace_array *tr, bool enable) 535 { 536 if (enable) { 537 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork, 538 tr, INT_MIN); 539 register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit, 540 tr, INT_MAX); 541 } else { 542 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork, 543 tr); 544 unregister_trace_sched_process_free(event_filter_pid_sched_process_exit, 545 tr); 546 } 547 } 548 549 static void 550 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt, 551 struct task_struct *prev, struct task_struct *next) 552 { 553 struct trace_array *tr = data; 554 struct trace_pid_list *no_pid_list; 555 struct trace_pid_list *pid_list; 556 bool ret; 557 558 pid_list = rcu_dereference_sched(tr->filtered_pids); 559 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids); 560 561 /* 562 * Sched switch is funny, as we only want to ignore it 563 * in the notrace case if both prev and next should be ignored. 564 */ 565 ret = trace_ignore_this_task(NULL, no_pid_list, prev) && 566 trace_ignore_this_task(NULL, no_pid_list, next); 567 568 this_cpu_write(tr->array_buffer.data->ignore_pid, ret || 569 (trace_ignore_this_task(pid_list, NULL, prev) && 570 trace_ignore_this_task(pid_list, NULL, next))); 571 } 572 573 static void 574 event_filter_pid_sched_switch_probe_post(void *data, bool preempt, 575 struct task_struct *prev, struct task_struct *next) 576 { 577 struct trace_array *tr = data; 578 struct trace_pid_list *no_pid_list; 579 struct trace_pid_list *pid_list; 580 581 pid_list = rcu_dereference_sched(tr->filtered_pids); 582 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids); 583 584 this_cpu_write(tr->array_buffer.data->ignore_pid, 585 trace_ignore_this_task(pid_list, no_pid_list, next)); 586 } 587 588 static void 589 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task) 590 { 591 struct trace_array *tr = data; 592 struct trace_pid_list *no_pid_list; 593 struct trace_pid_list *pid_list; 594 595 /* Nothing to do if we are already tracing */ 596 if (!this_cpu_read(tr->array_buffer.data->ignore_pid)) 597 return; 598 599 pid_list = rcu_dereference_sched(tr->filtered_pids); 600 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids); 601 602 this_cpu_write(tr->array_buffer.data->ignore_pid, 603 trace_ignore_this_task(pid_list, no_pid_list, task)); 604 } 605 606 static void 607 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task) 608 { 609 struct trace_array *tr = data; 610 struct trace_pid_list *no_pid_list; 611 struct trace_pid_list *pid_list; 612 613 /* Nothing to do if we are not tracing */ 614 if (this_cpu_read(tr->array_buffer.data->ignore_pid)) 615 return; 616 617 pid_list = rcu_dereference_sched(tr->filtered_pids); 618 no_pid_list = rcu_dereference_sched(tr->filtered_no_pids); 619 620 /* Set tracing if current is enabled */ 621 this_cpu_write(tr->array_buffer.data->ignore_pid, 622 trace_ignore_this_task(pid_list, no_pid_list, current)); 623 } 624 625 static void unregister_pid_events(struct trace_array *tr) 626 { 627 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr); 628 unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr); 629 630 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr); 631 unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr); 632 633 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr); 634 unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr); 635 636 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr); 637 unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr); 638 } 639 640 static void __ftrace_clear_event_pids(struct trace_array *tr, int type) 641 { 642 struct trace_pid_list *pid_list; 643 struct trace_pid_list *no_pid_list; 644 struct trace_event_file *file; 645 int cpu; 646 647 pid_list = rcu_dereference_protected(tr->filtered_pids, 648 lockdep_is_held(&event_mutex)); 649 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids, 650 lockdep_is_held(&event_mutex)); 651 652 /* Make sure there's something to do */ 653 if (!pid_type_enabled(type, pid_list, no_pid_list)) 654 return; 655 656 if (!still_need_pid_events(type, pid_list, no_pid_list)) { 657 unregister_pid_events(tr); 658 659 list_for_each_entry(file, &tr->events, list) { 660 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags); 661 } 662 663 for_each_possible_cpu(cpu) 664 per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false; 665 } 666 667 if (type & TRACE_PIDS) 668 rcu_assign_pointer(tr->filtered_pids, NULL); 669 670 if (type & TRACE_NO_PIDS) 671 rcu_assign_pointer(tr->filtered_no_pids, NULL); 672 673 /* Wait till all users are no longer using pid filtering */ 674 tracepoint_synchronize_unregister(); 675 676 if ((type & TRACE_PIDS) && pid_list) 677 trace_free_pid_list(pid_list); 678 679 if ((type & TRACE_NO_PIDS) && no_pid_list) 680 trace_free_pid_list(no_pid_list); 681 } 682 683 static void ftrace_clear_event_pids(struct trace_array *tr, int type) 684 { 685 mutex_lock(&event_mutex); 686 __ftrace_clear_event_pids(tr, type); 687 mutex_unlock(&event_mutex); 688 } 689 690 static void __put_system(struct event_subsystem *system) 691 { 692 struct event_filter *filter = system->filter; 693 694 WARN_ON_ONCE(system_refcount(system) == 0); 695 if (system_refcount_dec(system)) 696 return; 697 698 list_del(&system->list); 699 700 if (filter) { 701 kfree(filter->filter_string); 702 kfree(filter); 703 } 704 kfree_const(system->name); 705 kfree(system); 706 } 707 708 static void __get_system(struct event_subsystem *system) 709 { 710 WARN_ON_ONCE(system_refcount(system) == 0); 711 system_refcount_inc(system); 712 } 713 714 static void __get_system_dir(struct trace_subsystem_dir *dir) 715 { 716 WARN_ON_ONCE(dir->ref_count == 0); 717 dir->ref_count++; 718 __get_system(dir->subsystem); 719 } 720 721 static void __put_system_dir(struct trace_subsystem_dir *dir) 722 { 723 WARN_ON_ONCE(dir->ref_count == 0); 724 /* If the subsystem is about to be freed, the dir must be too */ 725 WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1); 726 727 __put_system(dir->subsystem); 728 if (!--dir->ref_count) 729 kfree(dir); 730 } 731 732 static void put_system(struct trace_subsystem_dir *dir) 733 { 734 mutex_lock(&event_mutex); 735 __put_system_dir(dir); 736 mutex_unlock(&event_mutex); 737 } 738 739 static void remove_subsystem(struct trace_subsystem_dir *dir) 740 { 741 if (!dir) 742 return; 743 744 if (!--dir->nr_events) { 745 tracefs_remove(dir->entry); 746 list_del(&dir->list); 747 __put_system_dir(dir); 748 } 749 } 750 751 static void remove_event_file_dir(struct trace_event_file *file) 752 { 753 struct dentry *dir = file->dir; 754 struct dentry *child; 755 756 if (dir) { 757 spin_lock(&dir->d_lock); /* probably unneeded */ 758 list_for_each_entry(child, &dir->d_subdirs, d_child) { 759 if (d_really_is_positive(child)) /* probably unneeded */ 760 d_inode(child)->i_private = NULL; 761 } 762 spin_unlock(&dir->d_lock); 763 764 tracefs_remove(dir); 765 } 766 767 list_del(&file->list); 768 remove_subsystem(file->system); 769 free_event_filter(file->filter); 770 kmem_cache_free(file_cachep, file); 771 } 772 773 /* 774 * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events. 775 */ 776 static int 777 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match, 778 const char *sub, const char *event, int set) 779 { 780 struct trace_event_file *file; 781 struct trace_event_call *call; 782 const char *name; 783 int ret = -EINVAL; 784 int eret = 0; 785 786 list_for_each_entry(file, &tr->events, list) { 787 788 call = file->event_call; 789 name = trace_event_name(call); 790 791 if (!name || !call->class || !call->class->reg) 792 continue; 793 794 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) 795 continue; 796 797 if (match && 798 strcmp(match, name) != 0 && 799 strcmp(match, call->class->system) != 0) 800 continue; 801 802 if (sub && strcmp(sub, call->class->system) != 0) 803 continue; 804 805 if (event && strcmp(event, name) != 0) 806 continue; 807 808 ret = ftrace_event_enable_disable(file, set); 809 810 /* 811 * Save the first error and return that. Some events 812 * may still have been enabled, but let the user 813 * know that something went wrong. 814 */ 815 if (ret && !eret) 816 eret = ret; 817 818 ret = eret; 819 } 820 821 return ret; 822 } 823 824 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match, 825 const char *sub, const char *event, int set) 826 { 827 int ret; 828 829 mutex_lock(&event_mutex); 830 ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set); 831 mutex_unlock(&event_mutex); 832 833 return ret; 834 } 835 836 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set) 837 { 838 char *event = NULL, *sub = NULL, *match; 839 int ret; 840 841 if (!tr) 842 return -ENOENT; 843 /* 844 * The buf format can be <subsystem>:<event-name> 845 * *:<event-name> means any event by that name. 846 * :<event-name> is the same. 847 * 848 * <subsystem>:* means all events in that subsystem 849 * <subsystem>: means the same. 850 * 851 * <name> (no ':') means all events in a subsystem with 852 * the name <name> or any event that matches <name> 853 */ 854 855 match = strsep(&buf, ":"); 856 if (buf) { 857 sub = match; 858 event = buf; 859 match = NULL; 860 861 if (!strlen(sub) || strcmp(sub, "*") == 0) 862 sub = NULL; 863 if (!strlen(event) || strcmp(event, "*") == 0) 864 event = NULL; 865 } 866 867 ret = __ftrace_set_clr_event(tr, match, sub, event, set); 868 869 /* Put back the colon to allow this to be called again */ 870 if (buf) 871 *(buf - 1) = ':'; 872 873 return ret; 874 } 875 876 /** 877 * trace_set_clr_event - enable or disable an event 878 * @system: system name to match (NULL for any system) 879 * @event: event name to match (NULL for all events, within system) 880 * @set: 1 to enable, 0 to disable 881 * 882 * This is a way for other parts of the kernel to enable or disable 883 * event recording. 884 * 885 * Returns 0 on success, -EINVAL if the parameters do not match any 886 * registered events. 887 */ 888 int trace_set_clr_event(const char *system, const char *event, int set) 889 { 890 struct trace_array *tr = top_trace_array(); 891 892 if (!tr) 893 return -ENODEV; 894 895 return __ftrace_set_clr_event(tr, NULL, system, event, set); 896 } 897 EXPORT_SYMBOL_GPL(trace_set_clr_event); 898 899 /** 900 * trace_array_set_clr_event - enable or disable an event for a trace array. 901 * @tr: concerned trace array. 902 * @system: system name to match (NULL for any system) 903 * @event: event name to match (NULL for all events, within system) 904 * @enable: true to enable, false to disable 905 * 906 * This is a way for other parts of the kernel to enable or disable 907 * event recording. 908 * 909 * Returns 0 on success, -EINVAL if the parameters do not match any 910 * registered events. 911 */ 912 int trace_array_set_clr_event(struct trace_array *tr, const char *system, 913 const char *event, bool enable) 914 { 915 int set; 916 917 if (!tr) 918 return -ENOENT; 919 920 set = (enable == true) ? 1 : 0; 921 return __ftrace_set_clr_event(tr, NULL, system, event, set); 922 } 923 EXPORT_SYMBOL_GPL(trace_array_set_clr_event); 924 925 /* 128 should be much more than enough */ 926 #define EVENT_BUF_SIZE 127 927 928 static ssize_t 929 ftrace_event_write(struct file *file, const char __user *ubuf, 930 size_t cnt, loff_t *ppos) 931 { 932 struct trace_parser parser; 933 struct seq_file *m = file->private_data; 934 struct trace_array *tr = m->private; 935 ssize_t read, ret; 936 937 if (!cnt) 938 return 0; 939 940 ret = tracing_update_buffers(); 941 if (ret < 0) 942 return ret; 943 944 if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1)) 945 return -ENOMEM; 946 947 read = trace_get_user(&parser, ubuf, cnt, ppos); 948 949 if (read >= 0 && trace_parser_loaded((&parser))) { 950 int set = 1; 951 952 if (*parser.buffer == '!') 953 set = 0; 954 955 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set); 956 if (ret) 957 goto out_put; 958 } 959 960 ret = read; 961 962 out_put: 963 trace_parser_put(&parser); 964 965 return ret; 966 } 967 968 static void * 969 t_next(struct seq_file *m, void *v, loff_t *pos) 970 { 971 struct trace_event_file *file = v; 972 struct trace_event_call *call; 973 struct trace_array *tr = m->private; 974 975 (*pos)++; 976 977 list_for_each_entry_continue(file, &tr->events, list) { 978 call = file->event_call; 979 /* 980 * The ftrace subsystem is for showing formats only. 981 * They can not be enabled or disabled via the event files. 982 */ 983 if (call->class && call->class->reg && 984 !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) 985 return file; 986 } 987 988 return NULL; 989 } 990 991 static void *t_start(struct seq_file *m, loff_t *pos) 992 { 993 struct trace_event_file *file; 994 struct trace_array *tr = m->private; 995 loff_t l; 996 997 mutex_lock(&event_mutex); 998 999 file = list_entry(&tr->events, struct trace_event_file, list); 1000 for (l = 0; l <= *pos; ) { 1001 file = t_next(m, file, &l); 1002 if (!file) 1003 break; 1004 } 1005 return file; 1006 } 1007 1008 static void * 1009 s_next(struct seq_file *m, void *v, loff_t *pos) 1010 { 1011 struct trace_event_file *file = v; 1012 struct trace_array *tr = m->private; 1013 1014 (*pos)++; 1015 1016 list_for_each_entry_continue(file, &tr->events, list) { 1017 if (file->flags & EVENT_FILE_FL_ENABLED) 1018 return file; 1019 } 1020 1021 return NULL; 1022 } 1023 1024 static void *s_start(struct seq_file *m, loff_t *pos) 1025 { 1026 struct trace_event_file *file; 1027 struct trace_array *tr = m->private; 1028 loff_t l; 1029 1030 mutex_lock(&event_mutex); 1031 1032 file = list_entry(&tr->events, struct trace_event_file, list); 1033 for (l = 0; l <= *pos; ) { 1034 file = s_next(m, file, &l); 1035 if (!file) 1036 break; 1037 } 1038 return file; 1039 } 1040 1041 static int t_show(struct seq_file *m, void *v) 1042 { 1043 struct trace_event_file *file = v; 1044 struct trace_event_call *call = file->event_call; 1045 1046 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) 1047 seq_printf(m, "%s:", call->class->system); 1048 seq_printf(m, "%s\n", trace_event_name(call)); 1049 1050 return 0; 1051 } 1052 1053 static void t_stop(struct seq_file *m, void *p) 1054 { 1055 mutex_unlock(&event_mutex); 1056 } 1057 1058 static void * 1059 __next(struct seq_file *m, void *v, loff_t *pos, int type) 1060 { 1061 struct trace_array *tr = m->private; 1062 struct trace_pid_list *pid_list; 1063 1064 if (type == TRACE_PIDS) 1065 pid_list = rcu_dereference_sched(tr->filtered_pids); 1066 else 1067 pid_list = rcu_dereference_sched(tr->filtered_no_pids); 1068 1069 return trace_pid_next(pid_list, v, pos); 1070 } 1071 1072 static void * 1073 p_next(struct seq_file *m, void *v, loff_t *pos) 1074 { 1075 return __next(m, v, pos, TRACE_PIDS); 1076 } 1077 1078 static void * 1079 np_next(struct seq_file *m, void *v, loff_t *pos) 1080 { 1081 return __next(m, v, pos, TRACE_NO_PIDS); 1082 } 1083 1084 static void *__start(struct seq_file *m, loff_t *pos, int type) 1085 __acquires(RCU) 1086 { 1087 struct trace_pid_list *pid_list; 1088 struct trace_array *tr = m->private; 1089 1090 /* 1091 * Grab the mutex, to keep calls to p_next() having the same 1092 * tr->filtered_pids as p_start() has. 1093 * If we just passed the tr->filtered_pids around, then RCU would 1094 * have been enough, but doing that makes things more complex. 1095 */ 1096 mutex_lock(&event_mutex); 1097 rcu_read_lock_sched(); 1098 1099 if (type == TRACE_PIDS) 1100 pid_list = rcu_dereference_sched(tr->filtered_pids); 1101 else 1102 pid_list = rcu_dereference_sched(tr->filtered_no_pids); 1103 1104 if (!pid_list) 1105 return NULL; 1106 1107 return trace_pid_start(pid_list, pos); 1108 } 1109 1110 static void *p_start(struct seq_file *m, loff_t *pos) 1111 __acquires(RCU) 1112 { 1113 return __start(m, pos, TRACE_PIDS); 1114 } 1115 1116 static void *np_start(struct seq_file *m, loff_t *pos) 1117 __acquires(RCU) 1118 { 1119 return __start(m, pos, TRACE_NO_PIDS); 1120 } 1121 1122 static void p_stop(struct seq_file *m, void *p) 1123 __releases(RCU) 1124 { 1125 rcu_read_unlock_sched(); 1126 mutex_unlock(&event_mutex); 1127 } 1128 1129 static ssize_t 1130 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt, 1131 loff_t *ppos) 1132 { 1133 struct trace_event_file *file; 1134 unsigned long flags; 1135 char buf[4] = "0"; 1136 1137 mutex_lock(&event_mutex); 1138 file = event_file_data(filp); 1139 if (likely(file)) 1140 flags = file->flags; 1141 mutex_unlock(&event_mutex); 1142 1143 if (!file) 1144 return -ENODEV; 1145 1146 if (flags & EVENT_FILE_FL_ENABLED && 1147 !(flags & EVENT_FILE_FL_SOFT_DISABLED)) 1148 strcpy(buf, "1"); 1149 1150 if (flags & EVENT_FILE_FL_SOFT_DISABLED || 1151 flags & EVENT_FILE_FL_SOFT_MODE) 1152 strcat(buf, "*"); 1153 1154 strcat(buf, "\n"); 1155 1156 return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf)); 1157 } 1158 1159 static ssize_t 1160 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt, 1161 loff_t *ppos) 1162 { 1163 struct trace_event_file *file; 1164 unsigned long val; 1165 int ret; 1166 1167 ret = kstrtoul_from_user(ubuf, cnt, 10, &val); 1168 if (ret) 1169 return ret; 1170 1171 ret = tracing_update_buffers(); 1172 if (ret < 0) 1173 return ret; 1174 1175 switch (val) { 1176 case 0: 1177 case 1: 1178 ret = -ENODEV; 1179 mutex_lock(&event_mutex); 1180 file = event_file_data(filp); 1181 if (likely(file)) 1182 ret = ftrace_event_enable_disable(file, val); 1183 mutex_unlock(&event_mutex); 1184 break; 1185 1186 default: 1187 return -EINVAL; 1188 } 1189 1190 *ppos += cnt; 1191 1192 return ret ? ret : cnt; 1193 } 1194 1195 static ssize_t 1196 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt, 1197 loff_t *ppos) 1198 { 1199 const char set_to_char[4] = { '?', '0', '1', 'X' }; 1200 struct trace_subsystem_dir *dir = filp->private_data; 1201 struct event_subsystem *system = dir->subsystem; 1202 struct trace_event_call *call; 1203 struct trace_event_file *file; 1204 struct trace_array *tr = dir->tr; 1205 char buf[2]; 1206 int set = 0; 1207 int ret; 1208 1209 mutex_lock(&event_mutex); 1210 list_for_each_entry(file, &tr->events, list) { 1211 call = file->event_call; 1212 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) || 1213 !trace_event_name(call) || !call->class || !call->class->reg) 1214 continue; 1215 1216 if (system && strcmp(call->class->system, system->name) != 0) 1217 continue; 1218 1219 /* 1220 * We need to find out if all the events are set 1221 * or if all events or cleared, or if we have 1222 * a mixture. 1223 */ 1224 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED)); 1225 1226 /* 1227 * If we have a mixture, no need to look further. 1228 */ 1229 if (set == 3) 1230 break; 1231 } 1232 mutex_unlock(&event_mutex); 1233 1234 buf[0] = set_to_char[set]; 1235 buf[1] = '\n'; 1236 1237 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2); 1238 1239 return ret; 1240 } 1241 1242 static ssize_t 1243 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt, 1244 loff_t *ppos) 1245 { 1246 struct trace_subsystem_dir *dir = filp->private_data; 1247 struct event_subsystem *system = dir->subsystem; 1248 const char *name = NULL; 1249 unsigned long val; 1250 ssize_t ret; 1251 1252 ret = kstrtoul_from_user(ubuf, cnt, 10, &val); 1253 if (ret) 1254 return ret; 1255 1256 ret = tracing_update_buffers(); 1257 if (ret < 0) 1258 return ret; 1259 1260 if (val != 0 && val != 1) 1261 return -EINVAL; 1262 1263 /* 1264 * Opening of "enable" adds a ref count to system, 1265 * so the name is safe to use. 1266 */ 1267 if (system) 1268 name = system->name; 1269 1270 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val); 1271 if (ret) 1272 goto out; 1273 1274 ret = cnt; 1275 1276 out: 1277 *ppos += cnt; 1278 1279 return ret; 1280 } 1281 1282 enum { 1283 FORMAT_HEADER = 1, 1284 FORMAT_FIELD_SEPERATOR = 2, 1285 FORMAT_PRINTFMT = 3, 1286 }; 1287 1288 static void *f_next(struct seq_file *m, void *v, loff_t *pos) 1289 { 1290 struct trace_event_call *call = event_file_data(m->private); 1291 struct list_head *common_head = &ftrace_common_fields; 1292 struct list_head *head = trace_get_fields(call); 1293 struct list_head *node = v; 1294 1295 (*pos)++; 1296 1297 switch ((unsigned long)v) { 1298 case FORMAT_HEADER: 1299 node = common_head; 1300 break; 1301 1302 case FORMAT_FIELD_SEPERATOR: 1303 node = head; 1304 break; 1305 1306 case FORMAT_PRINTFMT: 1307 /* all done */ 1308 return NULL; 1309 } 1310 1311 node = node->prev; 1312 if (node == common_head) 1313 return (void *)FORMAT_FIELD_SEPERATOR; 1314 else if (node == head) 1315 return (void *)FORMAT_PRINTFMT; 1316 else 1317 return node; 1318 } 1319 1320 static int f_show(struct seq_file *m, void *v) 1321 { 1322 struct trace_event_call *call = event_file_data(m->private); 1323 struct ftrace_event_field *field; 1324 const char *array_descriptor; 1325 1326 switch ((unsigned long)v) { 1327 case FORMAT_HEADER: 1328 seq_printf(m, "name: %s\n", trace_event_name(call)); 1329 seq_printf(m, "ID: %d\n", call->event.type); 1330 seq_puts(m, "format:\n"); 1331 return 0; 1332 1333 case FORMAT_FIELD_SEPERATOR: 1334 seq_putc(m, '\n'); 1335 return 0; 1336 1337 case FORMAT_PRINTFMT: 1338 seq_printf(m, "\nprint fmt: %s\n", 1339 call->print_fmt); 1340 return 0; 1341 } 1342 1343 field = list_entry(v, struct ftrace_event_field, link); 1344 /* 1345 * Smartly shows the array type(except dynamic array). 1346 * Normal: 1347 * field:TYPE VAR 1348 * If TYPE := TYPE[LEN], it is shown: 1349 * field:TYPE VAR[LEN] 1350 */ 1351 array_descriptor = strchr(field->type, '['); 1352 1353 if (str_has_prefix(field->type, "__data_loc")) 1354 array_descriptor = NULL; 1355 1356 if (!array_descriptor) 1357 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n", 1358 field->type, field->name, field->offset, 1359 field->size, !!field->is_signed); 1360 else 1361 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n", 1362 (int)(array_descriptor - field->type), 1363 field->type, field->name, 1364 array_descriptor, field->offset, 1365 field->size, !!field->is_signed); 1366 1367 return 0; 1368 } 1369 1370 static void *f_start(struct seq_file *m, loff_t *pos) 1371 { 1372 void *p = (void *)FORMAT_HEADER; 1373 loff_t l = 0; 1374 1375 /* ->stop() is called even if ->start() fails */ 1376 mutex_lock(&event_mutex); 1377 if (!event_file_data(m->private)) 1378 return ERR_PTR(-ENODEV); 1379 1380 while (l < *pos && p) 1381 p = f_next(m, p, &l); 1382 1383 return p; 1384 } 1385 1386 static void f_stop(struct seq_file *m, void *p) 1387 { 1388 mutex_unlock(&event_mutex); 1389 } 1390 1391 static const struct seq_operations trace_format_seq_ops = { 1392 .start = f_start, 1393 .next = f_next, 1394 .stop = f_stop, 1395 .show = f_show, 1396 }; 1397 1398 static int trace_format_open(struct inode *inode, struct file *file) 1399 { 1400 struct seq_file *m; 1401 int ret; 1402 1403 /* Do we want to hide event format files on tracefs lockdown? */ 1404 1405 ret = seq_open(file, &trace_format_seq_ops); 1406 if (ret < 0) 1407 return ret; 1408 1409 m = file->private_data; 1410 m->private = file; 1411 1412 return 0; 1413 } 1414 1415 static ssize_t 1416 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos) 1417 { 1418 int id = (long)event_file_data(filp); 1419 char buf[32]; 1420 int len; 1421 1422 if (unlikely(!id)) 1423 return -ENODEV; 1424 1425 len = sprintf(buf, "%d\n", id); 1426 1427 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len); 1428 } 1429 1430 static ssize_t 1431 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt, 1432 loff_t *ppos) 1433 { 1434 struct trace_event_file *file; 1435 struct trace_seq *s; 1436 int r = -ENODEV; 1437 1438 if (*ppos) 1439 return 0; 1440 1441 s = kmalloc(sizeof(*s), GFP_KERNEL); 1442 1443 if (!s) 1444 return -ENOMEM; 1445 1446 trace_seq_init(s); 1447 1448 mutex_lock(&event_mutex); 1449 file = event_file_data(filp); 1450 if (file) 1451 print_event_filter(file, s); 1452 mutex_unlock(&event_mutex); 1453 1454 if (file) 1455 r = simple_read_from_buffer(ubuf, cnt, ppos, 1456 s->buffer, trace_seq_used(s)); 1457 1458 kfree(s); 1459 1460 return r; 1461 } 1462 1463 static ssize_t 1464 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt, 1465 loff_t *ppos) 1466 { 1467 struct trace_event_file *file; 1468 char *buf; 1469 int err = -ENODEV; 1470 1471 if (cnt >= PAGE_SIZE) 1472 return -EINVAL; 1473 1474 buf = memdup_user_nul(ubuf, cnt); 1475 if (IS_ERR(buf)) 1476 return PTR_ERR(buf); 1477 1478 mutex_lock(&event_mutex); 1479 file = event_file_data(filp); 1480 if (file) 1481 err = apply_event_filter(file, buf); 1482 mutex_unlock(&event_mutex); 1483 1484 kfree(buf); 1485 if (err < 0) 1486 return err; 1487 1488 *ppos += cnt; 1489 1490 return cnt; 1491 } 1492 1493 static LIST_HEAD(event_subsystems); 1494 1495 static int subsystem_open(struct inode *inode, struct file *filp) 1496 { 1497 struct event_subsystem *system = NULL; 1498 struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */ 1499 struct trace_array *tr; 1500 int ret; 1501 1502 if (tracing_is_disabled()) 1503 return -ENODEV; 1504 1505 /* Make sure the system still exists */ 1506 mutex_lock(&event_mutex); 1507 mutex_lock(&trace_types_lock); 1508 list_for_each_entry(tr, &ftrace_trace_arrays, list) { 1509 list_for_each_entry(dir, &tr->systems, list) { 1510 if (dir == inode->i_private) { 1511 /* Don't open systems with no events */ 1512 if (dir->nr_events) { 1513 __get_system_dir(dir); 1514 system = dir->subsystem; 1515 } 1516 goto exit_loop; 1517 } 1518 } 1519 } 1520 exit_loop: 1521 mutex_unlock(&trace_types_lock); 1522 mutex_unlock(&event_mutex); 1523 1524 if (!system) 1525 return -ENODEV; 1526 1527 /* Some versions of gcc think dir can be uninitialized here */ 1528 WARN_ON(!dir); 1529 1530 /* Still need to increment the ref count of the system */ 1531 if (trace_array_get(tr) < 0) { 1532 put_system(dir); 1533 return -ENODEV; 1534 } 1535 1536 ret = tracing_open_generic(inode, filp); 1537 if (ret < 0) { 1538 trace_array_put(tr); 1539 put_system(dir); 1540 } 1541 1542 return ret; 1543 } 1544 1545 static int system_tr_open(struct inode *inode, struct file *filp) 1546 { 1547 struct trace_subsystem_dir *dir; 1548 struct trace_array *tr = inode->i_private; 1549 int ret; 1550 1551 /* Make a temporary dir that has no system but points to tr */ 1552 dir = kzalloc(sizeof(*dir), GFP_KERNEL); 1553 if (!dir) 1554 return -ENOMEM; 1555 1556 ret = tracing_open_generic_tr(inode, filp); 1557 if (ret < 0) { 1558 kfree(dir); 1559 return ret; 1560 } 1561 dir->tr = tr; 1562 filp->private_data = dir; 1563 1564 return 0; 1565 } 1566 1567 static int subsystem_release(struct inode *inode, struct file *file) 1568 { 1569 struct trace_subsystem_dir *dir = file->private_data; 1570 1571 trace_array_put(dir->tr); 1572 1573 /* 1574 * If dir->subsystem is NULL, then this is a temporary 1575 * descriptor that was made for a trace_array to enable 1576 * all subsystems. 1577 */ 1578 if (dir->subsystem) 1579 put_system(dir); 1580 else 1581 kfree(dir); 1582 1583 return 0; 1584 } 1585 1586 static ssize_t 1587 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt, 1588 loff_t *ppos) 1589 { 1590 struct trace_subsystem_dir *dir = filp->private_data; 1591 struct event_subsystem *system = dir->subsystem; 1592 struct trace_seq *s; 1593 int r; 1594 1595 if (*ppos) 1596 return 0; 1597 1598 s = kmalloc(sizeof(*s), GFP_KERNEL); 1599 if (!s) 1600 return -ENOMEM; 1601 1602 trace_seq_init(s); 1603 1604 print_subsystem_event_filter(system, s); 1605 r = simple_read_from_buffer(ubuf, cnt, ppos, 1606 s->buffer, trace_seq_used(s)); 1607 1608 kfree(s); 1609 1610 return r; 1611 } 1612 1613 static ssize_t 1614 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt, 1615 loff_t *ppos) 1616 { 1617 struct trace_subsystem_dir *dir = filp->private_data; 1618 char *buf; 1619 int err; 1620 1621 if (cnt >= PAGE_SIZE) 1622 return -EINVAL; 1623 1624 buf = memdup_user_nul(ubuf, cnt); 1625 if (IS_ERR(buf)) 1626 return PTR_ERR(buf); 1627 1628 err = apply_subsystem_event_filter(dir, buf); 1629 kfree(buf); 1630 if (err < 0) 1631 return err; 1632 1633 *ppos += cnt; 1634 1635 return cnt; 1636 } 1637 1638 static ssize_t 1639 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos) 1640 { 1641 int (*func)(struct trace_seq *s) = filp->private_data; 1642 struct trace_seq *s; 1643 int r; 1644 1645 if (*ppos) 1646 return 0; 1647 1648 s = kmalloc(sizeof(*s), GFP_KERNEL); 1649 if (!s) 1650 return -ENOMEM; 1651 1652 trace_seq_init(s); 1653 1654 func(s); 1655 r = simple_read_from_buffer(ubuf, cnt, ppos, 1656 s->buffer, trace_seq_used(s)); 1657 1658 kfree(s); 1659 1660 return r; 1661 } 1662 1663 static void ignore_task_cpu(void *data) 1664 { 1665 struct trace_array *tr = data; 1666 struct trace_pid_list *pid_list; 1667 struct trace_pid_list *no_pid_list; 1668 1669 /* 1670 * This function is called by on_each_cpu() while the 1671 * event_mutex is held. 1672 */ 1673 pid_list = rcu_dereference_protected(tr->filtered_pids, 1674 mutex_is_locked(&event_mutex)); 1675 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids, 1676 mutex_is_locked(&event_mutex)); 1677 1678 this_cpu_write(tr->array_buffer.data->ignore_pid, 1679 trace_ignore_this_task(pid_list, no_pid_list, current)); 1680 } 1681 1682 static void register_pid_events(struct trace_array *tr) 1683 { 1684 /* 1685 * Register a probe that is called before all other probes 1686 * to set ignore_pid if next or prev do not match. 1687 * Register a probe this is called after all other probes 1688 * to only keep ignore_pid set if next pid matches. 1689 */ 1690 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre, 1691 tr, INT_MAX); 1692 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post, 1693 tr, 0); 1694 1695 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, 1696 tr, INT_MAX); 1697 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, 1698 tr, 0); 1699 1700 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, 1701 tr, INT_MAX); 1702 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, 1703 tr, 0); 1704 1705 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre, 1706 tr, INT_MAX); 1707 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post, 1708 tr, 0); 1709 } 1710 1711 static ssize_t 1712 event_pid_write(struct file *filp, const char __user *ubuf, 1713 size_t cnt, loff_t *ppos, int type) 1714 { 1715 struct seq_file *m = filp->private_data; 1716 struct trace_array *tr = m->private; 1717 struct trace_pid_list *filtered_pids = NULL; 1718 struct trace_pid_list *other_pids = NULL; 1719 struct trace_pid_list *pid_list; 1720 struct trace_event_file *file; 1721 ssize_t ret; 1722 1723 if (!cnt) 1724 return 0; 1725 1726 ret = tracing_update_buffers(); 1727 if (ret < 0) 1728 return ret; 1729 1730 mutex_lock(&event_mutex); 1731 1732 if (type == TRACE_PIDS) { 1733 filtered_pids = rcu_dereference_protected(tr->filtered_pids, 1734 lockdep_is_held(&event_mutex)); 1735 other_pids = rcu_dereference_protected(tr->filtered_no_pids, 1736 lockdep_is_held(&event_mutex)); 1737 } else { 1738 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids, 1739 lockdep_is_held(&event_mutex)); 1740 other_pids = rcu_dereference_protected(tr->filtered_pids, 1741 lockdep_is_held(&event_mutex)); 1742 } 1743 1744 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt); 1745 if (ret < 0) 1746 goto out; 1747 1748 if (type == TRACE_PIDS) 1749 rcu_assign_pointer(tr->filtered_pids, pid_list); 1750 else 1751 rcu_assign_pointer(tr->filtered_no_pids, pid_list); 1752 1753 list_for_each_entry(file, &tr->events, list) { 1754 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags); 1755 } 1756 1757 if (filtered_pids) { 1758 tracepoint_synchronize_unregister(); 1759 trace_free_pid_list(filtered_pids); 1760 } else if (pid_list && !other_pids) { 1761 register_pid_events(tr); 1762 } 1763 1764 /* 1765 * Ignoring of pids is done at task switch. But we have to 1766 * check for those tasks that are currently running. 1767 * Always do this in case a pid was appended or removed. 1768 */ 1769 on_each_cpu(ignore_task_cpu, tr, 1); 1770 1771 out: 1772 mutex_unlock(&event_mutex); 1773 1774 if (ret > 0) 1775 *ppos += ret; 1776 1777 return ret; 1778 } 1779 1780 static ssize_t 1781 ftrace_event_pid_write(struct file *filp, const char __user *ubuf, 1782 size_t cnt, loff_t *ppos) 1783 { 1784 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS); 1785 } 1786 1787 static ssize_t 1788 ftrace_event_npid_write(struct file *filp, const char __user *ubuf, 1789 size_t cnt, loff_t *ppos) 1790 { 1791 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS); 1792 } 1793 1794 static int ftrace_event_avail_open(struct inode *inode, struct file *file); 1795 static int ftrace_event_set_open(struct inode *inode, struct file *file); 1796 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file); 1797 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file); 1798 static int ftrace_event_release(struct inode *inode, struct file *file); 1799 1800 static const struct seq_operations show_event_seq_ops = { 1801 .start = t_start, 1802 .next = t_next, 1803 .show = t_show, 1804 .stop = t_stop, 1805 }; 1806 1807 static const struct seq_operations show_set_event_seq_ops = { 1808 .start = s_start, 1809 .next = s_next, 1810 .show = t_show, 1811 .stop = t_stop, 1812 }; 1813 1814 static const struct seq_operations show_set_pid_seq_ops = { 1815 .start = p_start, 1816 .next = p_next, 1817 .show = trace_pid_show, 1818 .stop = p_stop, 1819 }; 1820 1821 static const struct seq_operations show_set_no_pid_seq_ops = { 1822 .start = np_start, 1823 .next = np_next, 1824 .show = trace_pid_show, 1825 .stop = p_stop, 1826 }; 1827 1828 static const struct file_operations ftrace_avail_fops = { 1829 .open = ftrace_event_avail_open, 1830 .read = seq_read, 1831 .llseek = seq_lseek, 1832 .release = seq_release, 1833 }; 1834 1835 static const struct file_operations ftrace_set_event_fops = { 1836 .open = ftrace_event_set_open, 1837 .read = seq_read, 1838 .write = ftrace_event_write, 1839 .llseek = seq_lseek, 1840 .release = ftrace_event_release, 1841 }; 1842 1843 static const struct file_operations ftrace_set_event_pid_fops = { 1844 .open = ftrace_event_set_pid_open, 1845 .read = seq_read, 1846 .write = ftrace_event_pid_write, 1847 .llseek = seq_lseek, 1848 .release = ftrace_event_release, 1849 }; 1850 1851 static const struct file_operations ftrace_set_event_notrace_pid_fops = { 1852 .open = ftrace_event_set_npid_open, 1853 .read = seq_read, 1854 .write = ftrace_event_npid_write, 1855 .llseek = seq_lseek, 1856 .release = ftrace_event_release, 1857 }; 1858 1859 static const struct file_operations ftrace_enable_fops = { 1860 .open = tracing_open_generic, 1861 .read = event_enable_read, 1862 .write = event_enable_write, 1863 .llseek = default_llseek, 1864 }; 1865 1866 static const struct file_operations ftrace_event_format_fops = { 1867 .open = trace_format_open, 1868 .read = seq_read, 1869 .llseek = seq_lseek, 1870 .release = seq_release, 1871 }; 1872 1873 static const struct file_operations ftrace_event_id_fops = { 1874 .read = event_id_read, 1875 .llseek = default_llseek, 1876 }; 1877 1878 static const struct file_operations ftrace_event_filter_fops = { 1879 .open = tracing_open_generic, 1880 .read = event_filter_read, 1881 .write = event_filter_write, 1882 .llseek = default_llseek, 1883 }; 1884 1885 static const struct file_operations ftrace_subsystem_filter_fops = { 1886 .open = subsystem_open, 1887 .read = subsystem_filter_read, 1888 .write = subsystem_filter_write, 1889 .llseek = default_llseek, 1890 .release = subsystem_release, 1891 }; 1892 1893 static const struct file_operations ftrace_system_enable_fops = { 1894 .open = subsystem_open, 1895 .read = system_enable_read, 1896 .write = system_enable_write, 1897 .llseek = default_llseek, 1898 .release = subsystem_release, 1899 }; 1900 1901 static const struct file_operations ftrace_tr_enable_fops = { 1902 .open = system_tr_open, 1903 .read = system_enable_read, 1904 .write = system_enable_write, 1905 .llseek = default_llseek, 1906 .release = subsystem_release, 1907 }; 1908 1909 static const struct file_operations ftrace_show_header_fops = { 1910 .open = tracing_open_generic, 1911 .read = show_header, 1912 .llseek = default_llseek, 1913 }; 1914 1915 static int 1916 ftrace_event_open(struct inode *inode, struct file *file, 1917 const struct seq_operations *seq_ops) 1918 { 1919 struct seq_file *m; 1920 int ret; 1921 1922 ret = security_locked_down(LOCKDOWN_TRACEFS); 1923 if (ret) 1924 return ret; 1925 1926 ret = seq_open(file, seq_ops); 1927 if (ret < 0) 1928 return ret; 1929 m = file->private_data; 1930 /* copy tr over to seq ops */ 1931 m->private = inode->i_private; 1932 1933 return ret; 1934 } 1935 1936 static int ftrace_event_release(struct inode *inode, struct file *file) 1937 { 1938 struct trace_array *tr = inode->i_private; 1939 1940 trace_array_put(tr); 1941 1942 return seq_release(inode, file); 1943 } 1944 1945 static int 1946 ftrace_event_avail_open(struct inode *inode, struct file *file) 1947 { 1948 const struct seq_operations *seq_ops = &show_event_seq_ops; 1949 1950 /* Checks for tracefs lockdown */ 1951 return ftrace_event_open(inode, file, seq_ops); 1952 } 1953 1954 static int 1955 ftrace_event_set_open(struct inode *inode, struct file *file) 1956 { 1957 const struct seq_operations *seq_ops = &show_set_event_seq_ops; 1958 struct trace_array *tr = inode->i_private; 1959 int ret; 1960 1961 ret = tracing_check_open_get_tr(tr); 1962 if (ret) 1963 return ret; 1964 1965 if ((file->f_mode & FMODE_WRITE) && 1966 (file->f_flags & O_TRUNC)) 1967 ftrace_clear_events(tr); 1968 1969 ret = ftrace_event_open(inode, file, seq_ops); 1970 if (ret < 0) 1971 trace_array_put(tr); 1972 return ret; 1973 } 1974 1975 static int 1976 ftrace_event_set_pid_open(struct inode *inode, struct file *file) 1977 { 1978 const struct seq_operations *seq_ops = &show_set_pid_seq_ops; 1979 struct trace_array *tr = inode->i_private; 1980 int ret; 1981 1982 ret = tracing_check_open_get_tr(tr); 1983 if (ret) 1984 return ret; 1985 1986 if ((file->f_mode & FMODE_WRITE) && 1987 (file->f_flags & O_TRUNC)) 1988 ftrace_clear_event_pids(tr, TRACE_PIDS); 1989 1990 ret = ftrace_event_open(inode, file, seq_ops); 1991 if (ret < 0) 1992 trace_array_put(tr); 1993 return ret; 1994 } 1995 1996 static int 1997 ftrace_event_set_npid_open(struct inode *inode, struct file *file) 1998 { 1999 const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops; 2000 struct trace_array *tr = inode->i_private; 2001 int ret; 2002 2003 ret = tracing_check_open_get_tr(tr); 2004 if (ret) 2005 return ret; 2006 2007 if ((file->f_mode & FMODE_WRITE) && 2008 (file->f_flags & O_TRUNC)) 2009 ftrace_clear_event_pids(tr, TRACE_NO_PIDS); 2010 2011 ret = ftrace_event_open(inode, file, seq_ops); 2012 if (ret < 0) 2013 trace_array_put(tr); 2014 return ret; 2015 } 2016 2017 static struct event_subsystem * 2018 create_new_subsystem(const char *name) 2019 { 2020 struct event_subsystem *system; 2021 2022 /* need to create new entry */ 2023 system = kmalloc(sizeof(*system), GFP_KERNEL); 2024 if (!system) 2025 return NULL; 2026 2027 system->ref_count = 1; 2028 2029 /* Only allocate if dynamic (kprobes and modules) */ 2030 system->name = kstrdup_const(name, GFP_KERNEL); 2031 if (!system->name) 2032 goto out_free; 2033 2034 system->filter = NULL; 2035 2036 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL); 2037 if (!system->filter) 2038 goto out_free; 2039 2040 list_add(&system->list, &event_subsystems); 2041 2042 return system; 2043 2044 out_free: 2045 kfree_const(system->name); 2046 kfree(system); 2047 return NULL; 2048 } 2049 2050 static struct dentry * 2051 event_subsystem_dir(struct trace_array *tr, const char *name, 2052 struct trace_event_file *file, struct dentry *parent) 2053 { 2054 struct trace_subsystem_dir *dir; 2055 struct event_subsystem *system; 2056 struct dentry *entry; 2057 2058 /* First see if we did not already create this dir */ 2059 list_for_each_entry(dir, &tr->systems, list) { 2060 system = dir->subsystem; 2061 if (strcmp(system->name, name) == 0) { 2062 dir->nr_events++; 2063 file->system = dir; 2064 return dir->entry; 2065 } 2066 } 2067 2068 /* Now see if the system itself exists. */ 2069 list_for_each_entry(system, &event_subsystems, list) { 2070 if (strcmp(system->name, name) == 0) 2071 break; 2072 } 2073 /* Reset system variable when not found */ 2074 if (&system->list == &event_subsystems) 2075 system = NULL; 2076 2077 dir = kmalloc(sizeof(*dir), GFP_KERNEL); 2078 if (!dir) 2079 goto out_fail; 2080 2081 if (!system) { 2082 system = create_new_subsystem(name); 2083 if (!system) 2084 goto out_free; 2085 } else 2086 __get_system(system); 2087 2088 dir->entry = tracefs_create_dir(name, parent); 2089 if (!dir->entry) { 2090 pr_warn("Failed to create system directory %s\n", name); 2091 __put_system(system); 2092 goto out_free; 2093 } 2094 2095 dir->tr = tr; 2096 dir->ref_count = 1; 2097 dir->nr_events = 1; 2098 dir->subsystem = system; 2099 file->system = dir; 2100 2101 /* the ftrace system is special, do not create enable or filter files */ 2102 if (strcmp(name, "ftrace") != 0) { 2103 2104 entry = tracefs_create_file("filter", 0644, dir->entry, dir, 2105 &ftrace_subsystem_filter_fops); 2106 if (!entry) { 2107 kfree(system->filter); 2108 system->filter = NULL; 2109 pr_warn("Could not create tracefs '%s/filter' entry\n", name); 2110 } 2111 2112 trace_create_file("enable", 0644, dir->entry, dir, 2113 &ftrace_system_enable_fops); 2114 } 2115 2116 list_add(&dir->list, &tr->systems); 2117 2118 return dir->entry; 2119 2120 out_free: 2121 kfree(dir); 2122 out_fail: 2123 /* Only print this message if failed on memory allocation */ 2124 if (!dir || !system) 2125 pr_warn("No memory to create event subsystem %s\n", name); 2126 return NULL; 2127 } 2128 2129 static int 2130 event_define_fields(struct trace_event_call *call) 2131 { 2132 struct list_head *head; 2133 int ret = 0; 2134 2135 /* 2136 * Other events may have the same class. Only update 2137 * the fields if they are not already defined. 2138 */ 2139 head = trace_get_fields(call); 2140 if (list_empty(head)) { 2141 struct trace_event_fields *field = call->class->fields_array; 2142 unsigned int offset = sizeof(struct trace_entry); 2143 2144 for (; field->type; field++) { 2145 if (field->type == TRACE_FUNCTION_TYPE) { 2146 field->define_fields(call); 2147 break; 2148 } 2149 2150 offset = ALIGN(offset, field->align); 2151 ret = trace_define_field(call, field->type, field->name, 2152 offset, field->size, 2153 field->is_signed, field->filter_type); 2154 if (WARN_ON_ONCE(ret)) { 2155 pr_err("error code is %d\n", ret); 2156 break; 2157 } 2158 2159 offset += field->size; 2160 } 2161 } 2162 2163 return ret; 2164 } 2165 2166 static int 2167 event_create_dir(struct dentry *parent, struct trace_event_file *file) 2168 { 2169 struct trace_event_call *call = file->event_call; 2170 struct trace_array *tr = file->tr; 2171 struct dentry *d_events; 2172 const char *name; 2173 int ret; 2174 2175 /* 2176 * If the trace point header did not define TRACE_SYSTEM 2177 * then the system would be called "TRACE_SYSTEM". 2178 */ 2179 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) { 2180 d_events = event_subsystem_dir(tr, call->class->system, file, parent); 2181 if (!d_events) 2182 return -ENOMEM; 2183 } else 2184 d_events = parent; 2185 2186 name = trace_event_name(call); 2187 file->dir = tracefs_create_dir(name, d_events); 2188 if (!file->dir) { 2189 pr_warn("Could not create tracefs '%s' directory\n", name); 2190 return -1; 2191 } 2192 2193 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) 2194 trace_create_file("enable", 0644, file->dir, file, 2195 &ftrace_enable_fops); 2196 2197 #ifdef CONFIG_PERF_EVENTS 2198 if (call->event.type && call->class->reg) 2199 trace_create_file("id", 0444, file->dir, 2200 (void *)(long)call->event.type, 2201 &ftrace_event_id_fops); 2202 #endif 2203 2204 ret = event_define_fields(call); 2205 if (ret < 0) { 2206 pr_warn("Could not initialize trace point events/%s\n", name); 2207 return ret; 2208 } 2209 2210 /* 2211 * Only event directories that can be enabled should have 2212 * triggers or filters. 2213 */ 2214 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) { 2215 trace_create_file("filter", 0644, file->dir, file, 2216 &ftrace_event_filter_fops); 2217 2218 trace_create_file("trigger", 0644, file->dir, file, 2219 &event_trigger_fops); 2220 } 2221 2222 #ifdef CONFIG_HIST_TRIGGERS 2223 trace_create_file("hist", 0444, file->dir, file, 2224 &event_hist_fops); 2225 #endif 2226 #ifdef CONFIG_HIST_TRIGGERS_DEBUG 2227 trace_create_file("hist_debug", 0444, file->dir, file, 2228 &event_hist_debug_fops); 2229 #endif 2230 trace_create_file("format", 0444, file->dir, call, 2231 &ftrace_event_format_fops); 2232 2233 #ifdef CONFIG_TRACE_EVENT_INJECT 2234 if (call->event.type && call->class->reg) 2235 trace_create_file("inject", 0200, file->dir, file, 2236 &event_inject_fops); 2237 #endif 2238 2239 return 0; 2240 } 2241 2242 static void remove_event_from_tracers(struct trace_event_call *call) 2243 { 2244 struct trace_event_file *file; 2245 struct trace_array *tr; 2246 2247 do_for_each_event_file_safe(tr, file) { 2248 if (file->event_call != call) 2249 continue; 2250 2251 remove_event_file_dir(file); 2252 /* 2253 * The do_for_each_event_file_safe() is 2254 * a double loop. After finding the call for this 2255 * trace_array, we use break to jump to the next 2256 * trace_array. 2257 */ 2258 break; 2259 } while_for_each_event_file(); 2260 } 2261 2262 static void event_remove(struct trace_event_call *call) 2263 { 2264 struct trace_array *tr; 2265 struct trace_event_file *file; 2266 2267 do_for_each_event_file(tr, file) { 2268 if (file->event_call != call) 2269 continue; 2270 2271 if (file->flags & EVENT_FILE_FL_WAS_ENABLED) 2272 tr->clear_trace = true; 2273 2274 ftrace_event_enable_disable(file, 0); 2275 /* 2276 * The do_for_each_event_file() is 2277 * a double loop. After finding the call for this 2278 * trace_array, we use break to jump to the next 2279 * trace_array. 2280 */ 2281 break; 2282 } while_for_each_event_file(); 2283 2284 if (call->event.funcs) 2285 __unregister_trace_event(&call->event); 2286 remove_event_from_tracers(call); 2287 list_del(&call->list); 2288 } 2289 2290 static int event_init(struct trace_event_call *call) 2291 { 2292 int ret = 0; 2293 const char *name; 2294 2295 name = trace_event_name(call); 2296 if (WARN_ON(!name)) 2297 return -EINVAL; 2298 2299 if (call->class->raw_init) { 2300 ret = call->class->raw_init(call); 2301 if (ret < 0 && ret != -ENOSYS) 2302 pr_warn("Could not initialize trace events/%s\n", name); 2303 } 2304 2305 return ret; 2306 } 2307 2308 static int 2309 __register_event(struct trace_event_call *call, struct module *mod) 2310 { 2311 int ret; 2312 2313 ret = event_init(call); 2314 if (ret < 0) 2315 return ret; 2316 2317 list_add(&call->list, &ftrace_events); 2318 call->mod = mod; 2319 2320 return 0; 2321 } 2322 2323 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len) 2324 { 2325 int rlen; 2326 int elen; 2327 2328 /* Find the length of the eval value as a string */ 2329 elen = snprintf(ptr, 0, "%ld", map->eval_value); 2330 /* Make sure there's enough room to replace the string with the value */ 2331 if (len < elen) 2332 return NULL; 2333 2334 snprintf(ptr, elen + 1, "%ld", map->eval_value); 2335 2336 /* Get the rest of the string of ptr */ 2337 rlen = strlen(ptr + len); 2338 memmove(ptr + elen, ptr + len, rlen); 2339 /* Make sure we end the new string */ 2340 ptr[elen + rlen] = 0; 2341 2342 return ptr + elen; 2343 } 2344 2345 static void update_event_printk(struct trace_event_call *call, 2346 struct trace_eval_map *map) 2347 { 2348 char *ptr; 2349 int quote = 0; 2350 int len = strlen(map->eval_string); 2351 2352 for (ptr = call->print_fmt; *ptr; ptr++) { 2353 if (*ptr == '\\') { 2354 ptr++; 2355 /* paranoid */ 2356 if (!*ptr) 2357 break; 2358 continue; 2359 } 2360 if (*ptr == '"') { 2361 quote ^= 1; 2362 continue; 2363 } 2364 if (quote) 2365 continue; 2366 if (isdigit(*ptr)) { 2367 /* skip numbers */ 2368 do { 2369 ptr++; 2370 /* Check for alpha chars like ULL */ 2371 } while (isalnum(*ptr)); 2372 if (!*ptr) 2373 break; 2374 /* 2375 * A number must have some kind of delimiter after 2376 * it, and we can ignore that too. 2377 */ 2378 continue; 2379 } 2380 if (isalpha(*ptr) || *ptr == '_') { 2381 if (strncmp(map->eval_string, ptr, len) == 0 && 2382 !isalnum(ptr[len]) && ptr[len] != '_') { 2383 ptr = eval_replace(ptr, map, len); 2384 /* enum/sizeof string smaller than value */ 2385 if (WARN_ON_ONCE(!ptr)) 2386 return; 2387 /* 2388 * No need to decrement here, as eval_replace() 2389 * returns the pointer to the character passed 2390 * the eval, and two evals can not be placed 2391 * back to back without something in between. 2392 * We can skip that something in between. 2393 */ 2394 continue; 2395 } 2396 skip_more: 2397 do { 2398 ptr++; 2399 } while (isalnum(*ptr) || *ptr == '_'); 2400 if (!*ptr) 2401 break; 2402 /* 2403 * If what comes after this variable is a '.' or 2404 * '->' then we can continue to ignore that string. 2405 */ 2406 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) { 2407 ptr += *ptr == '.' ? 1 : 2; 2408 if (!*ptr) 2409 break; 2410 goto skip_more; 2411 } 2412 /* 2413 * Once again, we can skip the delimiter that came 2414 * after the string. 2415 */ 2416 continue; 2417 } 2418 } 2419 } 2420 2421 void trace_event_eval_update(struct trace_eval_map **map, int len) 2422 { 2423 struct trace_event_call *call, *p; 2424 const char *last_system = NULL; 2425 bool first = false; 2426 int last_i; 2427 int i; 2428 2429 down_write(&trace_event_sem); 2430 list_for_each_entry_safe(call, p, &ftrace_events, list) { 2431 /* events are usually grouped together with systems */ 2432 if (!last_system || call->class->system != last_system) { 2433 first = true; 2434 last_i = 0; 2435 last_system = call->class->system; 2436 } 2437 2438 /* 2439 * Since calls are grouped by systems, the likelyhood that the 2440 * next call in the iteration belongs to the same system as the 2441 * previous call is high. As an optimization, we skip searching 2442 * for a map[] that matches the call's system if the last call 2443 * was from the same system. That's what last_i is for. If the 2444 * call has the same system as the previous call, then last_i 2445 * will be the index of the first map[] that has a matching 2446 * system. 2447 */ 2448 for (i = last_i; i < len; i++) { 2449 if (call->class->system == map[i]->system) { 2450 /* Save the first system if need be */ 2451 if (first) { 2452 last_i = i; 2453 first = false; 2454 } 2455 update_event_printk(call, map[i]); 2456 } 2457 } 2458 } 2459 up_write(&trace_event_sem); 2460 } 2461 2462 static struct trace_event_file * 2463 trace_create_new_event(struct trace_event_call *call, 2464 struct trace_array *tr) 2465 { 2466 struct trace_event_file *file; 2467 2468 file = kmem_cache_alloc(file_cachep, GFP_TRACE); 2469 if (!file) 2470 return NULL; 2471 2472 file->event_call = call; 2473 file->tr = tr; 2474 atomic_set(&file->sm_ref, 0); 2475 atomic_set(&file->tm_ref, 0); 2476 INIT_LIST_HEAD(&file->triggers); 2477 list_add(&file->list, &tr->events); 2478 2479 return file; 2480 } 2481 2482 /* Add an event to a trace directory */ 2483 static int 2484 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr) 2485 { 2486 struct trace_event_file *file; 2487 2488 file = trace_create_new_event(call, tr); 2489 if (!file) 2490 return -ENOMEM; 2491 2492 if (eventdir_initialized) 2493 return event_create_dir(tr->event_dir, file); 2494 else 2495 return event_define_fields(call); 2496 } 2497 2498 /* 2499 * Just create a decriptor for early init. A descriptor is required 2500 * for enabling events at boot. We want to enable events before 2501 * the filesystem is initialized. 2502 */ 2503 static int 2504 __trace_early_add_new_event(struct trace_event_call *call, 2505 struct trace_array *tr) 2506 { 2507 struct trace_event_file *file; 2508 2509 file = trace_create_new_event(call, tr); 2510 if (!file) 2511 return -ENOMEM; 2512 2513 return event_define_fields(call); 2514 } 2515 2516 struct ftrace_module_file_ops; 2517 static void __add_event_to_tracers(struct trace_event_call *call); 2518 2519 /* Add an additional event_call dynamically */ 2520 int trace_add_event_call(struct trace_event_call *call) 2521 { 2522 int ret; 2523 lockdep_assert_held(&event_mutex); 2524 2525 mutex_lock(&trace_types_lock); 2526 2527 ret = __register_event(call, NULL); 2528 if (ret >= 0) 2529 __add_event_to_tracers(call); 2530 2531 mutex_unlock(&trace_types_lock); 2532 return ret; 2533 } 2534 2535 /* 2536 * Must be called under locking of trace_types_lock, event_mutex and 2537 * trace_event_sem. 2538 */ 2539 static void __trace_remove_event_call(struct trace_event_call *call) 2540 { 2541 event_remove(call); 2542 trace_destroy_fields(call); 2543 free_event_filter(call->filter); 2544 call->filter = NULL; 2545 } 2546 2547 static int probe_remove_event_call(struct trace_event_call *call) 2548 { 2549 struct trace_array *tr; 2550 struct trace_event_file *file; 2551 2552 #ifdef CONFIG_PERF_EVENTS 2553 if (call->perf_refcount) 2554 return -EBUSY; 2555 #endif 2556 do_for_each_event_file(tr, file) { 2557 if (file->event_call != call) 2558 continue; 2559 /* 2560 * We can't rely on ftrace_event_enable_disable(enable => 0) 2561 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress 2562 * TRACE_REG_UNREGISTER. 2563 */ 2564 if (file->flags & EVENT_FILE_FL_ENABLED) 2565 return -EBUSY; 2566 /* 2567 * The do_for_each_event_file_safe() is 2568 * a double loop. After finding the call for this 2569 * trace_array, we use break to jump to the next 2570 * trace_array. 2571 */ 2572 break; 2573 } while_for_each_event_file(); 2574 2575 __trace_remove_event_call(call); 2576 2577 return 0; 2578 } 2579 2580 /* Remove an event_call */ 2581 int trace_remove_event_call(struct trace_event_call *call) 2582 { 2583 int ret; 2584 2585 lockdep_assert_held(&event_mutex); 2586 2587 mutex_lock(&trace_types_lock); 2588 down_write(&trace_event_sem); 2589 ret = probe_remove_event_call(call); 2590 up_write(&trace_event_sem); 2591 mutex_unlock(&trace_types_lock); 2592 2593 return ret; 2594 } 2595 2596 #define for_each_event(event, start, end) \ 2597 for (event = start; \ 2598 (unsigned long)event < (unsigned long)end; \ 2599 event++) 2600 2601 #ifdef CONFIG_MODULES 2602 2603 static void trace_module_add_events(struct module *mod) 2604 { 2605 struct trace_event_call **call, **start, **end; 2606 2607 if (!mod->num_trace_events) 2608 return; 2609 2610 /* Don't add infrastructure for mods without tracepoints */ 2611 if (trace_module_has_bad_taint(mod)) { 2612 pr_err("%s: module has bad taint, not creating trace events\n", 2613 mod->name); 2614 return; 2615 } 2616 2617 start = mod->trace_events; 2618 end = mod->trace_events + mod->num_trace_events; 2619 2620 for_each_event(call, start, end) { 2621 __register_event(*call, mod); 2622 __add_event_to_tracers(*call); 2623 } 2624 } 2625 2626 static void trace_module_remove_events(struct module *mod) 2627 { 2628 struct trace_event_call *call, *p; 2629 2630 down_write(&trace_event_sem); 2631 list_for_each_entry_safe(call, p, &ftrace_events, list) { 2632 if (call->mod == mod) 2633 __trace_remove_event_call(call); 2634 } 2635 up_write(&trace_event_sem); 2636 2637 /* 2638 * It is safest to reset the ring buffer if the module being unloaded 2639 * registered any events that were used. The only worry is if 2640 * a new module gets loaded, and takes on the same id as the events 2641 * of this module. When printing out the buffer, traced events left 2642 * over from this module may be passed to the new module events and 2643 * unexpected results may occur. 2644 */ 2645 tracing_reset_all_online_cpus(); 2646 } 2647 2648 static int trace_module_notify(struct notifier_block *self, 2649 unsigned long val, void *data) 2650 { 2651 struct module *mod = data; 2652 2653 mutex_lock(&event_mutex); 2654 mutex_lock(&trace_types_lock); 2655 switch (val) { 2656 case MODULE_STATE_COMING: 2657 trace_module_add_events(mod); 2658 break; 2659 case MODULE_STATE_GOING: 2660 trace_module_remove_events(mod); 2661 break; 2662 } 2663 mutex_unlock(&trace_types_lock); 2664 mutex_unlock(&event_mutex); 2665 2666 return NOTIFY_OK; 2667 } 2668 2669 static struct notifier_block trace_module_nb = { 2670 .notifier_call = trace_module_notify, 2671 .priority = 1, /* higher than trace.c module notify */ 2672 }; 2673 #endif /* CONFIG_MODULES */ 2674 2675 /* Create a new event directory structure for a trace directory. */ 2676 static void 2677 __trace_add_event_dirs(struct trace_array *tr) 2678 { 2679 struct trace_event_call *call; 2680 int ret; 2681 2682 list_for_each_entry(call, &ftrace_events, list) { 2683 ret = __trace_add_new_event(call, tr); 2684 if (ret < 0) 2685 pr_warn("Could not create directory for event %s\n", 2686 trace_event_name(call)); 2687 } 2688 } 2689 2690 /* Returns any file that matches the system and event */ 2691 struct trace_event_file * 2692 __find_event_file(struct trace_array *tr, const char *system, const char *event) 2693 { 2694 struct trace_event_file *file; 2695 struct trace_event_call *call; 2696 const char *name; 2697 2698 list_for_each_entry(file, &tr->events, list) { 2699 2700 call = file->event_call; 2701 name = trace_event_name(call); 2702 2703 if (!name || !call->class) 2704 continue; 2705 2706 if (strcmp(event, name) == 0 && 2707 strcmp(system, call->class->system) == 0) 2708 return file; 2709 } 2710 return NULL; 2711 } 2712 2713 /* Returns valid trace event files that match system and event */ 2714 struct trace_event_file * 2715 find_event_file(struct trace_array *tr, const char *system, const char *event) 2716 { 2717 struct trace_event_file *file; 2718 2719 file = __find_event_file(tr, system, event); 2720 if (!file || !file->event_call->class->reg || 2721 file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) 2722 return NULL; 2723 2724 return file; 2725 } 2726 2727 /** 2728 * trace_get_event_file - Find and return a trace event file 2729 * @instance: The name of the trace instance containing the event 2730 * @system: The name of the system containing the event 2731 * @event: The name of the event 2732 * 2733 * Return a trace event file given the trace instance name, trace 2734 * system, and trace event name. If the instance name is NULL, it 2735 * refers to the top-level trace array. 2736 * 2737 * This function will look it up and return it if found, after calling 2738 * trace_array_get() to prevent the instance from going away, and 2739 * increment the event's module refcount to prevent it from being 2740 * removed. 2741 * 2742 * To release the file, call trace_put_event_file(), which will call 2743 * trace_array_put() and decrement the event's module refcount. 2744 * 2745 * Return: The trace event on success, ERR_PTR otherwise. 2746 */ 2747 struct trace_event_file *trace_get_event_file(const char *instance, 2748 const char *system, 2749 const char *event) 2750 { 2751 struct trace_array *tr = top_trace_array(); 2752 struct trace_event_file *file = NULL; 2753 int ret = -EINVAL; 2754 2755 if (instance) { 2756 tr = trace_array_find_get(instance); 2757 if (!tr) 2758 return ERR_PTR(-ENOENT); 2759 } else { 2760 ret = trace_array_get(tr); 2761 if (ret) 2762 return ERR_PTR(ret); 2763 } 2764 2765 mutex_lock(&event_mutex); 2766 2767 file = find_event_file(tr, system, event); 2768 if (!file) { 2769 trace_array_put(tr); 2770 ret = -EINVAL; 2771 goto out; 2772 } 2773 2774 /* Don't let event modules unload while in use */ 2775 ret = try_module_get(file->event_call->mod); 2776 if (!ret) { 2777 trace_array_put(tr); 2778 ret = -EBUSY; 2779 goto out; 2780 } 2781 2782 ret = 0; 2783 out: 2784 mutex_unlock(&event_mutex); 2785 2786 if (ret) 2787 file = ERR_PTR(ret); 2788 2789 return file; 2790 } 2791 EXPORT_SYMBOL_GPL(trace_get_event_file); 2792 2793 /** 2794 * trace_put_event_file - Release a file from trace_get_event_file() 2795 * @file: The trace event file 2796 * 2797 * If a file was retrieved using trace_get_event_file(), this should 2798 * be called when it's no longer needed. It will cancel the previous 2799 * trace_array_get() called by that function, and decrement the 2800 * event's module refcount. 2801 */ 2802 void trace_put_event_file(struct trace_event_file *file) 2803 { 2804 mutex_lock(&event_mutex); 2805 module_put(file->event_call->mod); 2806 mutex_unlock(&event_mutex); 2807 2808 trace_array_put(file->tr); 2809 } 2810 EXPORT_SYMBOL_GPL(trace_put_event_file); 2811 2812 #ifdef CONFIG_DYNAMIC_FTRACE 2813 2814 /* Avoid typos */ 2815 #define ENABLE_EVENT_STR "enable_event" 2816 #define DISABLE_EVENT_STR "disable_event" 2817 2818 struct event_probe_data { 2819 struct trace_event_file *file; 2820 unsigned long count; 2821 int ref; 2822 bool enable; 2823 }; 2824 2825 static void update_event_probe(struct event_probe_data *data) 2826 { 2827 if (data->enable) 2828 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags); 2829 else 2830 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags); 2831 } 2832 2833 static void 2834 event_enable_probe(unsigned long ip, unsigned long parent_ip, 2835 struct trace_array *tr, struct ftrace_probe_ops *ops, 2836 void *data) 2837 { 2838 struct ftrace_func_mapper *mapper = data; 2839 struct event_probe_data *edata; 2840 void **pdata; 2841 2842 pdata = ftrace_func_mapper_find_ip(mapper, ip); 2843 if (!pdata || !*pdata) 2844 return; 2845 2846 edata = *pdata; 2847 update_event_probe(edata); 2848 } 2849 2850 static void 2851 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, 2852 struct trace_array *tr, struct ftrace_probe_ops *ops, 2853 void *data) 2854 { 2855 struct ftrace_func_mapper *mapper = data; 2856 struct event_probe_data *edata; 2857 void **pdata; 2858 2859 pdata = ftrace_func_mapper_find_ip(mapper, ip); 2860 if (!pdata || !*pdata) 2861 return; 2862 2863 edata = *pdata; 2864 2865 if (!edata->count) 2866 return; 2867 2868 /* Skip if the event is in a state we want to switch to */ 2869 if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED)) 2870 return; 2871 2872 if (edata->count != -1) 2873 (edata->count)--; 2874 2875 update_event_probe(edata); 2876 } 2877 2878 static int 2879 event_enable_print(struct seq_file *m, unsigned long ip, 2880 struct ftrace_probe_ops *ops, void *data) 2881 { 2882 struct ftrace_func_mapper *mapper = data; 2883 struct event_probe_data *edata; 2884 void **pdata; 2885 2886 pdata = ftrace_func_mapper_find_ip(mapper, ip); 2887 2888 if (WARN_ON_ONCE(!pdata || !*pdata)) 2889 return 0; 2890 2891 edata = *pdata; 2892 2893 seq_printf(m, "%ps:", (void *)ip); 2894 2895 seq_printf(m, "%s:%s:%s", 2896 edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR, 2897 edata->file->event_call->class->system, 2898 trace_event_name(edata->file->event_call)); 2899 2900 if (edata->count == -1) 2901 seq_puts(m, ":unlimited\n"); 2902 else 2903 seq_printf(m, ":count=%ld\n", edata->count); 2904 2905 return 0; 2906 } 2907 2908 static int 2909 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr, 2910 unsigned long ip, void *init_data, void **data) 2911 { 2912 struct ftrace_func_mapper *mapper = *data; 2913 struct event_probe_data *edata = init_data; 2914 int ret; 2915 2916 if (!mapper) { 2917 mapper = allocate_ftrace_func_mapper(); 2918 if (!mapper) 2919 return -ENODEV; 2920 *data = mapper; 2921 } 2922 2923 ret = ftrace_func_mapper_add_ip(mapper, ip, edata); 2924 if (ret < 0) 2925 return ret; 2926 2927 edata->ref++; 2928 2929 return 0; 2930 } 2931 2932 static int free_probe_data(void *data) 2933 { 2934 struct event_probe_data *edata = data; 2935 2936 edata->ref--; 2937 if (!edata->ref) { 2938 /* Remove the SOFT_MODE flag */ 2939 __ftrace_event_enable_disable(edata->file, 0, 1); 2940 module_put(edata->file->event_call->mod); 2941 kfree(edata); 2942 } 2943 return 0; 2944 } 2945 2946 static void 2947 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr, 2948 unsigned long ip, void *data) 2949 { 2950 struct ftrace_func_mapper *mapper = data; 2951 struct event_probe_data *edata; 2952 2953 if (!ip) { 2954 if (!mapper) 2955 return; 2956 free_ftrace_func_mapper(mapper, free_probe_data); 2957 return; 2958 } 2959 2960 edata = ftrace_func_mapper_remove_ip(mapper, ip); 2961 2962 if (WARN_ON_ONCE(!edata)) 2963 return; 2964 2965 if (WARN_ON_ONCE(edata->ref <= 0)) 2966 return; 2967 2968 free_probe_data(edata); 2969 } 2970 2971 static struct ftrace_probe_ops event_enable_probe_ops = { 2972 .func = event_enable_probe, 2973 .print = event_enable_print, 2974 .init = event_enable_init, 2975 .free = event_enable_free, 2976 }; 2977 2978 static struct ftrace_probe_ops event_enable_count_probe_ops = { 2979 .func = event_enable_count_probe, 2980 .print = event_enable_print, 2981 .init = event_enable_init, 2982 .free = event_enable_free, 2983 }; 2984 2985 static struct ftrace_probe_ops event_disable_probe_ops = { 2986 .func = event_enable_probe, 2987 .print = event_enable_print, 2988 .init = event_enable_init, 2989 .free = event_enable_free, 2990 }; 2991 2992 static struct ftrace_probe_ops event_disable_count_probe_ops = { 2993 .func = event_enable_count_probe, 2994 .print = event_enable_print, 2995 .init = event_enable_init, 2996 .free = event_enable_free, 2997 }; 2998 2999 static int 3000 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash, 3001 char *glob, char *cmd, char *param, int enabled) 3002 { 3003 struct trace_event_file *file; 3004 struct ftrace_probe_ops *ops; 3005 struct event_probe_data *data; 3006 const char *system; 3007 const char *event; 3008 char *number; 3009 bool enable; 3010 int ret; 3011 3012 if (!tr) 3013 return -ENODEV; 3014 3015 /* hash funcs only work with set_ftrace_filter */ 3016 if (!enabled || !param) 3017 return -EINVAL; 3018 3019 system = strsep(¶m, ":"); 3020 if (!param) 3021 return -EINVAL; 3022 3023 event = strsep(¶m, ":"); 3024 3025 mutex_lock(&event_mutex); 3026 3027 ret = -EINVAL; 3028 file = find_event_file(tr, system, event); 3029 if (!file) 3030 goto out; 3031 3032 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0; 3033 3034 if (enable) 3035 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops; 3036 else 3037 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops; 3038 3039 if (glob[0] == '!') { 3040 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops); 3041 goto out; 3042 } 3043 3044 ret = -ENOMEM; 3045 3046 data = kzalloc(sizeof(*data), GFP_KERNEL); 3047 if (!data) 3048 goto out; 3049 3050 data->enable = enable; 3051 data->count = -1; 3052 data->file = file; 3053 3054 if (!param) 3055 goto out_reg; 3056 3057 number = strsep(¶m, ":"); 3058 3059 ret = -EINVAL; 3060 if (!strlen(number)) 3061 goto out_free; 3062 3063 /* 3064 * We use the callback data field (which is a pointer) 3065 * as our counter. 3066 */ 3067 ret = kstrtoul(number, 0, &data->count); 3068 if (ret) 3069 goto out_free; 3070 3071 out_reg: 3072 /* Don't let event modules unload while probe registered */ 3073 ret = try_module_get(file->event_call->mod); 3074 if (!ret) { 3075 ret = -EBUSY; 3076 goto out_free; 3077 } 3078 3079 ret = __ftrace_event_enable_disable(file, 1, 1); 3080 if (ret < 0) 3081 goto out_put; 3082 3083 ret = register_ftrace_function_probe(glob, tr, ops, data); 3084 /* 3085 * The above returns on success the # of functions enabled, 3086 * but if it didn't find any functions it returns zero. 3087 * Consider no functions a failure too. 3088 */ 3089 if (!ret) { 3090 ret = -ENOENT; 3091 goto out_disable; 3092 } else if (ret < 0) 3093 goto out_disable; 3094 /* Just return zero, not the number of enabled functions */ 3095 ret = 0; 3096 out: 3097 mutex_unlock(&event_mutex); 3098 return ret; 3099 3100 out_disable: 3101 __ftrace_event_enable_disable(file, 0, 1); 3102 out_put: 3103 module_put(file->event_call->mod); 3104 out_free: 3105 kfree(data); 3106 goto out; 3107 } 3108 3109 static struct ftrace_func_command event_enable_cmd = { 3110 .name = ENABLE_EVENT_STR, 3111 .func = event_enable_func, 3112 }; 3113 3114 static struct ftrace_func_command event_disable_cmd = { 3115 .name = DISABLE_EVENT_STR, 3116 .func = event_enable_func, 3117 }; 3118 3119 static __init int register_event_cmds(void) 3120 { 3121 int ret; 3122 3123 ret = register_ftrace_command(&event_enable_cmd); 3124 if (WARN_ON(ret < 0)) 3125 return ret; 3126 ret = register_ftrace_command(&event_disable_cmd); 3127 if (WARN_ON(ret < 0)) 3128 unregister_ftrace_command(&event_enable_cmd); 3129 return ret; 3130 } 3131 #else 3132 static inline int register_event_cmds(void) { return 0; } 3133 #endif /* CONFIG_DYNAMIC_FTRACE */ 3134 3135 /* 3136 * The top level array and trace arrays created by boot-time tracing 3137 * have already had its trace_event_file descriptors created in order 3138 * to allow for early events to be recorded. 3139 * This function is called after the tracefs has been initialized, 3140 * and we now have to create the files associated to the events. 3141 */ 3142 static void __trace_early_add_event_dirs(struct trace_array *tr) 3143 { 3144 struct trace_event_file *file; 3145 int ret; 3146 3147 3148 list_for_each_entry(file, &tr->events, list) { 3149 ret = event_create_dir(tr->event_dir, file); 3150 if (ret < 0) 3151 pr_warn("Could not create directory for event %s\n", 3152 trace_event_name(file->event_call)); 3153 } 3154 } 3155 3156 /* 3157 * For early boot up, the top trace array and the trace arrays created 3158 * by boot-time tracing require to have a list of events that can be 3159 * enabled. This must be done before the filesystem is set up in order 3160 * to allow events to be traced early. 3161 */ 3162 void __trace_early_add_events(struct trace_array *tr) 3163 { 3164 struct trace_event_call *call; 3165 int ret; 3166 3167 list_for_each_entry(call, &ftrace_events, list) { 3168 /* Early boot up should not have any modules loaded */ 3169 if (WARN_ON_ONCE(call->mod)) 3170 continue; 3171 3172 ret = __trace_early_add_new_event(call, tr); 3173 if (ret < 0) 3174 pr_warn("Could not create early event %s\n", 3175 trace_event_name(call)); 3176 } 3177 } 3178 3179 /* Remove the event directory structure for a trace directory. */ 3180 static void 3181 __trace_remove_event_dirs(struct trace_array *tr) 3182 { 3183 struct trace_event_file *file, *next; 3184 3185 list_for_each_entry_safe(file, next, &tr->events, list) 3186 remove_event_file_dir(file); 3187 } 3188 3189 static void __add_event_to_tracers(struct trace_event_call *call) 3190 { 3191 struct trace_array *tr; 3192 3193 list_for_each_entry(tr, &ftrace_trace_arrays, list) 3194 __trace_add_new_event(call, tr); 3195 } 3196 3197 extern struct trace_event_call *__start_ftrace_events[]; 3198 extern struct trace_event_call *__stop_ftrace_events[]; 3199 3200 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata; 3201 3202 static __init int setup_trace_event(char *str) 3203 { 3204 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE); 3205 ring_buffer_expanded = true; 3206 disable_tracing_selftest("running event tracing"); 3207 3208 return 1; 3209 } 3210 __setup("trace_event=", setup_trace_event); 3211 3212 /* Expects to have event_mutex held when called */ 3213 static int 3214 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr) 3215 { 3216 struct dentry *d_events; 3217 struct dentry *entry; 3218 3219 entry = tracefs_create_file("set_event", 0644, parent, 3220 tr, &ftrace_set_event_fops); 3221 if (!entry) { 3222 pr_warn("Could not create tracefs 'set_event' entry\n"); 3223 return -ENOMEM; 3224 } 3225 3226 d_events = tracefs_create_dir("events", parent); 3227 if (!d_events) { 3228 pr_warn("Could not create tracefs 'events' directory\n"); 3229 return -ENOMEM; 3230 } 3231 3232 entry = trace_create_file("enable", 0644, d_events, 3233 tr, &ftrace_tr_enable_fops); 3234 if (!entry) { 3235 pr_warn("Could not create tracefs 'enable' entry\n"); 3236 return -ENOMEM; 3237 } 3238 3239 /* There are not as crucial, just warn if they are not created */ 3240 3241 entry = tracefs_create_file("set_event_pid", 0644, parent, 3242 tr, &ftrace_set_event_pid_fops); 3243 if (!entry) 3244 pr_warn("Could not create tracefs 'set_event_pid' entry\n"); 3245 3246 entry = tracefs_create_file("set_event_notrace_pid", 0644, parent, 3247 tr, &ftrace_set_event_notrace_pid_fops); 3248 if (!entry) 3249 pr_warn("Could not create tracefs 'set_event_notrace_pid' entry\n"); 3250 3251 /* ring buffer internal formats */ 3252 entry = trace_create_file("header_page", 0444, d_events, 3253 ring_buffer_print_page_header, 3254 &ftrace_show_header_fops); 3255 if (!entry) 3256 pr_warn("Could not create tracefs 'header_page' entry\n"); 3257 3258 entry = trace_create_file("header_event", 0444, d_events, 3259 ring_buffer_print_entry_header, 3260 &ftrace_show_header_fops); 3261 if (!entry) 3262 pr_warn("Could not create tracefs 'header_event' entry\n"); 3263 3264 tr->event_dir = d_events; 3265 3266 return 0; 3267 } 3268 3269 /** 3270 * event_trace_add_tracer - add a instance of a trace_array to events 3271 * @parent: The parent dentry to place the files/directories for events in 3272 * @tr: The trace array associated with these events 3273 * 3274 * When a new instance is created, it needs to set up its events 3275 * directory, as well as other files associated with events. It also 3276 * creates the event hierarchy in the @parent/events directory. 3277 * 3278 * Returns 0 on success. 3279 * 3280 * Must be called with event_mutex held. 3281 */ 3282 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr) 3283 { 3284 int ret; 3285 3286 lockdep_assert_held(&event_mutex); 3287 3288 ret = create_event_toplevel_files(parent, tr); 3289 if (ret) 3290 goto out; 3291 3292 down_write(&trace_event_sem); 3293 /* If tr already has the event list, it is initialized in early boot. */ 3294 if (unlikely(!list_empty(&tr->events))) 3295 __trace_early_add_event_dirs(tr); 3296 else 3297 __trace_add_event_dirs(tr); 3298 up_write(&trace_event_sem); 3299 3300 out: 3301 return ret; 3302 } 3303 3304 /* 3305 * The top trace array already had its file descriptors created. 3306 * Now the files themselves need to be created. 3307 */ 3308 static __init int 3309 early_event_add_tracer(struct dentry *parent, struct trace_array *tr) 3310 { 3311 int ret; 3312 3313 mutex_lock(&event_mutex); 3314 3315 ret = create_event_toplevel_files(parent, tr); 3316 if (ret) 3317 goto out_unlock; 3318 3319 down_write(&trace_event_sem); 3320 __trace_early_add_event_dirs(tr); 3321 up_write(&trace_event_sem); 3322 3323 out_unlock: 3324 mutex_unlock(&event_mutex); 3325 3326 return ret; 3327 } 3328 3329 /* Must be called with event_mutex held */ 3330 int event_trace_del_tracer(struct trace_array *tr) 3331 { 3332 lockdep_assert_held(&event_mutex); 3333 3334 /* Disable any event triggers and associated soft-disabled events */ 3335 clear_event_triggers(tr); 3336 3337 /* Clear the pid list */ 3338 __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS); 3339 3340 /* Disable any running events */ 3341 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0); 3342 3343 /* Make sure no more events are being executed */ 3344 tracepoint_synchronize_unregister(); 3345 3346 down_write(&trace_event_sem); 3347 __trace_remove_event_dirs(tr); 3348 tracefs_remove(tr->event_dir); 3349 up_write(&trace_event_sem); 3350 3351 tr->event_dir = NULL; 3352 3353 return 0; 3354 } 3355 3356 static __init int event_trace_memsetup(void) 3357 { 3358 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC); 3359 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC); 3360 return 0; 3361 } 3362 3363 static __init void 3364 early_enable_events(struct trace_array *tr, bool disable_first) 3365 { 3366 char *buf = bootup_event_buf; 3367 char *token; 3368 int ret; 3369 3370 while (true) { 3371 token = strsep(&buf, ","); 3372 3373 if (!token) 3374 break; 3375 3376 if (*token) { 3377 /* Restarting syscalls requires that we stop them first */ 3378 if (disable_first) 3379 ftrace_set_clr_event(tr, token, 0); 3380 3381 ret = ftrace_set_clr_event(tr, token, 1); 3382 if (ret) 3383 pr_warn("Failed to enable trace event: %s\n", token); 3384 } 3385 3386 /* Put back the comma to allow this to be called again */ 3387 if (buf) 3388 *(buf - 1) = ','; 3389 } 3390 } 3391 3392 static __init int event_trace_enable(void) 3393 { 3394 struct trace_array *tr = top_trace_array(); 3395 struct trace_event_call **iter, *call; 3396 int ret; 3397 3398 if (!tr) 3399 return -ENODEV; 3400 3401 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) { 3402 3403 call = *iter; 3404 ret = event_init(call); 3405 if (!ret) 3406 list_add(&call->list, &ftrace_events); 3407 } 3408 3409 /* 3410 * We need the top trace array to have a working set of trace 3411 * points at early init, before the debug files and directories 3412 * are created. Create the file entries now, and attach them 3413 * to the actual file dentries later. 3414 */ 3415 __trace_early_add_events(tr); 3416 3417 early_enable_events(tr, false); 3418 3419 trace_printk_start_comm(); 3420 3421 register_event_cmds(); 3422 3423 register_trigger_cmds(); 3424 3425 return 0; 3426 } 3427 3428 /* 3429 * event_trace_enable() is called from trace_event_init() first to 3430 * initialize events and perhaps start any events that are on the 3431 * command line. Unfortunately, there are some events that will not 3432 * start this early, like the system call tracepoints that need 3433 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But 3434 * event_trace_enable() is called before pid 1 starts, and this flag 3435 * is never set, making the syscall tracepoint never get reached, but 3436 * the event is enabled regardless (and not doing anything). 3437 */ 3438 static __init int event_trace_enable_again(void) 3439 { 3440 struct trace_array *tr; 3441 3442 tr = top_trace_array(); 3443 if (!tr) 3444 return -ENODEV; 3445 3446 early_enable_events(tr, true); 3447 3448 return 0; 3449 } 3450 3451 early_initcall(event_trace_enable_again); 3452 3453 /* Init fields which doesn't related to the tracefs */ 3454 static __init int event_trace_init_fields(void) 3455 { 3456 if (trace_define_generic_fields()) 3457 pr_warn("tracing: Failed to allocated generic fields"); 3458 3459 if (trace_define_common_fields()) 3460 pr_warn("tracing: Failed to allocate common fields"); 3461 3462 return 0; 3463 } 3464 3465 __init int event_trace_init(void) 3466 { 3467 struct trace_array *tr; 3468 struct dentry *entry; 3469 int ret; 3470 3471 tr = top_trace_array(); 3472 if (!tr) 3473 return -ENODEV; 3474 3475 entry = tracefs_create_file("available_events", 0444, NULL, 3476 tr, &ftrace_avail_fops); 3477 if (!entry) 3478 pr_warn("Could not create tracefs 'available_events' entry\n"); 3479 3480 ret = early_event_add_tracer(NULL, tr); 3481 if (ret) 3482 return ret; 3483 3484 #ifdef CONFIG_MODULES 3485 ret = register_module_notifier(&trace_module_nb); 3486 if (ret) 3487 pr_warn("Failed to register trace events module notifier\n"); 3488 #endif 3489 3490 eventdir_initialized = true; 3491 3492 return 0; 3493 } 3494 3495 void __init trace_event_init(void) 3496 { 3497 event_trace_memsetup(); 3498 init_ftrace_syscalls(); 3499 event_trace_enable(); 3500 event_trace_init_fields(); 3501 } 3502 3503 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST 3504 3505 static DEFINE_SPINLOCK(test_spinlock); 3506 static DEFINE_SPINLOCK(test_spinlock_irq); 3507 static DEFINE_MUTEX(test_mutex); 3508 3509 static __init void test_work(struct work_struct *dummy) 3510 { 3511 spin_lock(&test_spinlock); 3512 spin_lock_irq(&test_spinlock_irq); 3513 udelay(1); 3514 spin_unlock_irq(&test_spinlock_irq); 3515 spin_unlock(&test_spinlock); 3516 3517 mutex_lock(&test_mutex); 3518 msleep(1); 3519 mutex_unlock(&test_mutex); 3520 } 3521 3522 static __init int event_test_thread(void *unused) 3523 { 3524 void *test_malloc; 3525 3526 test_malloc = kmalloc(1234, GFP_KERNEL); 3527 if (!test_malloc) 3528 pr_info("failed to kmalloc\n"); 3529 3530 schedule_on_each_cpu(test_work); 3531 3532 kfree(test_malloc); 3533 3534 set_current_state(TASK_INTERRUPTIBLE); 3535 while (!kthread_should_stop()) { 3536 schedule(); 3537 set_current_state(TASK_INTERRUPTIBLE); 3538 } 3539 __set_current_state(TASK_RUNNING); 3540 3541 return 0; 3542 } 3543 3544 /* 3545 * Do various things that may trigger events. 3546 */ 3547 static __init void event_test_stuff(void) 3548 { 3549 struct task_struct *test_thread; 3550 3551 test_thread = kthread_run(event_test_thread, NULL, "test-events"); 3552 msleep(1); 3553 kthread_stop(test_thread); 3554 } 3555 3556 /* 3557 * For every trace event defined, we will test each trace point separately, 3558 * and then by groups, and finally all trace points. 3559 */ 3560 static __init void event_trace_self_tests(void) 3561 { 3562 struct trace_subsystem_dir *dir; 3563 struct trace_event_file *file; 3564 struct trace_event_call *call; 3565 struct event_subsystem *system; 3566 struct trace_array *tr; 3567 int ret; 3568 3569 tr = top_trace_array(); 3570 if (!tr) 3571 return; 3572 3573 pr_info("Running tests on trace events:\n"); 3574 3575 list_for_each_entry(file, &tr->events, list) { 3576 3577 call = file->event_call; 3578 3579 /* Only test those that have a probe */ 3580 if (!call->class || !call->class->probe) 3581 continue; 3582 3583 /* 3584 * Testing syscall events here is pretty useless, but 3585 * we still do it if configured. But this is time consuming. 3586 * What we really need is a user thread to perform the 3587 * syscalls as we test. 3588 */ 3589 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS 3590 if (call->class->system && 3591 strcmp(call->class->system, "syscalls") == 0) 3592 continue; 3593 #endif 3594 3595 pr_info("Testing event %s: ", trace_event_name(call)); 3596 3597 /* 3598 * If an event is already enabled, someone is using 3599 * it and the self test should not be on. 3600 */ 3601 if (file->flags & EVENT_FILE_FL_ENABLED) { 3602 pr_warn("Enabled event during self test!\n"); 3603 WARN_ON_ONCE(1); 3604 continue; 3605 } 3606 3607 ftrace_event_enable_disable(file, 1); 3608 event_test_stuff(); 3609 ftrace_event_enable_disable(file, 0); 3610 3611 pr_cont("OK\n"); 3612 } 3613 3614 /* Now test at the sub system level */ 3615 3616 pr_info("Running tests on trace event systems:\n"); 3617 3618 list_for_each_entry(dir, &tr->systems, list) { 3619 3620 system = dir->subsystem; 3621 3622 /* the ftrace system is special, skip it */ 3623 if (strcmp(system->name, "ftrace") == 0) 3624 continue; 3625 3626 pr_info("Testing event system %s: ", system->name); 3627 3628 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1); 3629 if (WARN_ON_ONCE(ret)) { 3630 pr_warn("error enabling system %s\n", 3631 system->name); 3632 continue; 3633 } 3634 3635 event_test_stuff(); 3636 3637 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0); 3638 if (WARN_ON_ONCE(ret)) { 3639 pr_warn("error disabling system %s\n", 3640 system->name); 3641 continue; 3642 } 3643 3644 pr_cont("OK\n"); 3645 } 3646 3647 /* Test with all events enabled */ 3648 3649 pr_info("Running tests on all trace events:\n"); 3650 pr_info("Testing all events: "); 3651 3652 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1); 3653 if (WARN_ON_ONCE(ret)) { 3654 pr_warn("error enabling all events\n"); 3655 return; 3656 } 3657 3658 event_test_stuff(); 3659 3660 /* reset sysname */ 3661 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0); 3662 if (WARN_ON_ONCE(ret)) { 3663 pr_warn("error disabling all events\n"); 3664 return; 3665 } 3666 3667 pr_cont("OK\n"); 3668 } 3669 3670 #ifdef CONFIG_FUNCTION_TRACER 3671 3672 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable); 3673 3674 static struct trace_event_file event_trace_file __initdata; 3675 3676 static void __init 3677 function_test_events_call(unsigned long ip, unsigned long parent_ip, 3678 struct ftrace_ops *op, struct ftrace_regs *regs) 3679 { 3680 struct trace_buffer *buffer; 3681 struct ring_buffer_event *event; 3682 struct ftrace_entry *entry; 3683 unsigned int trace_ctx; 3684 long disabled; 3685 int cpu; 3686 3687 trace_ctx = tracing_gen_ctx(); 3688 preempt_disable_notrace(); 3689 cpu = raw_smp_processor_id(); 3690 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu)); 3691 3692 if (disabled != 1) 3693 goto out; 3694 3695 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file, 3696 TRACE_FN, sizeof(*entry), 3697 trace_ctx); 3698 if (!event) 3699 goto out; 3700 entry = ring_buffer_event_data(event); 3701 entry->ip = ip; 3702 entry->parent_ip = parent_ip; 3703 3704 event_trigger_unlock_commit(&event_trace_file, buffer, event, 3705 entry, trace_ctx); 3706 out: 3707 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu)); 3708 preempt_enable_notrace(); 3709 } 3710 3711 static struct ftrace_ops trace_ops __initdata = 3712 { 3713 .func = function_test_events_call, 3714 }; 3715 3716 static __init void event_trace_self_test_with_function(void) 3717 { 3718 int ret; 3719 3720 event_trace_file.tr = top_trace_array(); 3721 if (WARN_ON(!event_trace_file.tr)) 3722 return; 3723 3724 ret = register_ftrace_function(&trace_ops); 3725 if (WARN_ON(ret < 0)) { 3726 pr_info("Failed to enable function tracer for event tests\n"); 3727 return; 3728 } 3729 pr_info("Running tests again, along with the function tracer\n"); 3730 event_trace_self_tests(); 3731 unregister_ftrace_function(&trace_ops); 3732 } 3733 #else 3734 static __init void event_trace_self_test_with_function(void) 3735 { 3736 } 3737 #endif 3738 3739 static __init int event_trace_self_tests_init(void) 3740 { 3741 if (!tracing_selftest_disabled) { 3742 event_trace_self_tests(); 3743 event_trace_self_test_with_function(); 3744 } 3745 3746 return 0; 3747 } 3748 3749 late_initcall(event_trace_self_tests_init); 3750 3751 #endif 3752