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 (!trace_event_name(call) || !call->class || !call->class->reg) 1213 continue; 1214 1215 if (system && strcmp(call->class->system, system->name) != 0) 1216 continue; 1217 1218 /* 1219 * We need to find out if all the events are set 1220 * or if all events or cleared, or if we have 1221 * a mixture. 1222 */ 1223 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED)); 1224 1225 /* 1226 * If we have a mixture, no need to look further. 1227 */ 1228 if (set == 3) 1229 break; 1230 } 1231 mutex_unlock(&event_mutex); 1232 1233 buf[0] = set_to_char[set]; 1234 buf[1] = '\n'; 1235 1236 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2); 1237 1238 return ret; 1239 } 1240 1241 static ssize_t 1242 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt, 1243 loff_t *ppos) 1244 { 1245 struct trace_subsystem_dir *dir = filp->private_data; 1246 struct event_subsystem *system = dir->subsystem; 1247 const char *name = NULL; 1248 unsigned long val; 1249 ssize_t ret; 1250 1251 ret = kstrtoul_from_user(ubuf, cnt, 10, &val); 1252 if (ret) 1253 return ret; 1254 1255 ret = tracing_update_buffers(); 1256 if (ret < 0) 1257 return ret; 1258 1259 if (val != 0 && val != 1) 1260 return -EINVAL; 1261 1262 /* 1263 * Opening of "enable" adds a ref count to system, 1264 * so the name is safe to use. 1265 */ 1266 if (system) 1267 name = system->name; 1268 1269 ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val); 1270 if (ret) 1271 goto out; 1272 1273 ret = cnt; 1274 1275 out: 1276 *ppos += cnt; 1277 1278 return ret; 1279 } 1280 1281 enum { 1282 FORMAT_HEADER = 1, 1283 FORMAT_FIELD_SEPERATOR = 2, 1284 FORMAT_PRINTFMT = 3, 1285 }; 1286 1287 static void *f_next(struct seq_file *m, void *v, loff_t *pos) 1288 { 1289 struct trace_event_call *call = event_file_data(m->private); 1290 struct list_head *common_head = &ftrace_common_fields; 1291 struct list_head *head = trace_get_fields(call); 1292 struct list_head *node = v; 1293 1294 (*pos)++; 1295 1296 switch ((unsigned long)v) { 1297 case FORMAT_HEADER: 1298 node = common_head; 1299 break; 1300 1301 case FORMAT_FIELD_SEPERATOR: 1302 node = head; 1303 break; 1304 1305 case FORMAT_PRINTFMT: 1306 /* all done */ 1307 return NULL; 1308 } 1309 1310 node = node->prev; 1311 if (node == common_head) 1312 return (void *)FORMAT_FIELD_SEPERATOR; 1313 else if (node == head) 1314 return (void *)FORMAT_PRINTFMT; 1315 else 1316 return node; 1317 } 1318 1319 static int f_show(struct seq_file *m, void *v) 1320 { 1321 struct trace_event_call *call = event_file_data(m->private); 1322 struct ftrace_event_field *field; 1323 const char *array_descriptor; 1324 1325 switch ((unsigned long)v) { 1326 case FORMAT_HEADER: 1327 seq_printf(m, "name: %s\n", trace_event_name(call)); 1328 seq_printf(m, "ID: %d\n", call->event.type); 1329 seq_puts(m, "format:\n"); 1330 return 0; 1331 1332 case FORMAT_FIELD_SEPERATOR: 1333 seq_putc(m, '\n'); 1334 return 0; 1335 1336 case FORMAT_PRINTFMT: 1337 seq_printf(m, "\nprint fmt: %s\n", 1338 call->print_fmt); 1339 return 0; 1340 } 1341 1342 field = list_entry(v, struct ftrace_event_field, link); 1343 /* 1344 * Smartly shows the array type(except dynamic array). 1345 * Normal: 1346 * field:TYPE VAR 1347 * If TYPE := TYPE[LEN], it is shown: 1348 * field:TYPE VAR[LEN] 1349 */ 1350 array_descriptor = strchr(field->type, '['); 1351 1352 if (str_has_prefix(field->type, "__data_loc")) 1353 array_descriptor = NULL; 1354 1355 if (!array_descriptor) 1356 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n", 1357 field->type, field->name, field->offset, 1358 field->size, !!field->is_signed); 1359 else 1360 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n", 1361 (int)(array_descriptor - field->type), 1362 field->type, field->name, 1363 array_descriptor, field->offset, 1364 field->size, !!field->is_signed); 1365 1366 return 0; 1367 } 1368 1369 static void *f_start(struct seq_file *m, loff_t *pos) 1370 { 1371 void *p = (void *)FORMAT_HEADER; 1372 loff_t l = 0; 1373 1374 /* ->stop() is called even if ->start() fails */ 1375 mutex_lock(&event_mutex); 1376 if (!event_file_data(m->private)) 1377 return ERR_PTR(-ENODEV); 1378 1379 while (l < *pos && p) 1380 p = f_next(m, p, &l); 1381 1382 return p; 1383 } 1384 1385 static void f_stop(struct seq_file *m, void *p) 1386 { 1387 mutex_unlock(&event_mutex); 1388 } 1389 1390 static const struct seq_operations trace_format_seq_ops = { 1391 .start = f_start, 1392 .next = f_next, 1393 .stop = f_stop, 1394 .show = f_show, 1395 }; 1396 1397 static int trace_format_open(struct inode *inode, struct file *file) 1398 { 1399 struct seq_file *m; 1400 int ret; 1401 1402 /* Do we want to hide event format files on tracefs lockdown? */ 1403 1404 ret = seq_open(file, &trace_format_seq_ops); 1405 if (ret < 0) 1406 return ret; 1407 1408 m = file->private_data; 1409 m->private = file; 1410 1411 return 0; 1412 } 1413 1414 static ssize_t 1415 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos) 1416 { 1417 int id = (long)event_file_data(filp); 1418 char buf[32]; 1419 int len; 1420 1421 if (unlikely(!id)) 1422 return -ENODEV; 1423 1424 len = sprintf(buf, "%d\n", id); 1425 1426 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len); 1427 } 1428 1429 static ssize_t 1430 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt, 1431 loff_t *ppos) 1432 { 1433 struct trace_event_file *file; 1434 struct trace_seq *s; 1435 int r = -ENODEV; 1436 1437 if (*ppos) 1438 return 0; 1439 1440 s = kmalloc(sizeof(*s), GFP_KERNEL); 1441 1442 if (!s) 1443 return -ENOMEM; 1444 1445 trace_seq_init(s); 1446 1447 mutex_lock(&event_mutex); 1448 file = event_file_data(filp); 1449 if (file) 1450 print_event_filter(file, s); 1451 mutex_unlock(&event_mutex); 1452 1453 if (file) 1454 r = simple_read_from_buffer(ubuf, cnt, ppos, 1455 s->buffer, trace_seq_used(s)); 1456 1457 kfree(s); 1458 1459 return r; 1460 } 1461 1462 static ssize_t 1463 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt, 1464 loff_t *ppos) 1465 { 1466 struct trace_event_file *file; 1467 char *buf; 1468 int err = -ENODEV; 1469 1470 if (cnt >= PAGE_SIZE) 1471 return -EINVAL; 1472 1473 buf = memdup_user_nul(ubuf, cnt); 1474 if (IS_ERR(buf)) 1475 return PTR_ERR(buf); 1476 1477 mutex_lock(&event_mutex); 1478 file = event_file_data(filp); 1479 if (file) 1480 err = apply_event_filter(file, buf); 1481 mutex_unlock(&event_mutex); 1482 1483 kfree(buf); 1484 if (err < 0) 1485 return err; 1486 1487 *ppos += cnt; 1488 1489 return cnt; 1490 } 1491 1492 static LIST_HEAD(event_subsystems); 1493 1494 static int subsystem_open(struct inode *inode, struct file *filp) 1495 { 1496 struct event_subsystem *system = NULL; 1497 struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */ 1498 struct trace_array *tr; 1499 int ret; 1500 1501 if (tracing_is_disabled()) 1502 return -ENODEV; 1503 1504 /* Make sure the system still exists */ 1505 mutex_lock(&event_mutex); 1506 mutex_lock(&trace_types_lock); 1507 list_for_each_entry(tr, &ftrace_trace_arrays, list) { 1508 list_for_each_entry(dir, &tr->systems, list) { 1509 if (dir == inode->i_private) { 1510 /* Don't open systems with no events */ 1511 if (dir->nr_events) { 1512 __get_system_dir(dir); 1513 system = dir->subsystem; 1514 } 1515 goto exit_loop; 1516 } 1517 } 1518 } 1519 exit_loop: 1520 mutex_unlock(&trace_types_lock); 1521 mutex_unlock(&event_mutex); 1522 1523 if (!system) 1524 return -ENODEV; 1525 1526 /* Some versions of gcc think dir can be uninitialized here */ 1527 WARN_ON(!dir); 1528 1529 /* Still need to increment the ref count of the system */ 1530 if (trace_array_get(tr) < 0) { 1531 put_system(dir); 1532 return -ENODEV; 1533 } 1534 1535 ret = tracing_open_generic(inode, filp); 1536 if (ret < 0) { 1537 trace_array_put(tr); 1538 put_system(dir); 1539 } 1540 1541 return ret; 1542 } 1543 1544 static int system_tr_open(struct inode *inode, struct file *filp) 1545 { 1546 struct trace_subsystem_dir *dir; 1547 struct trace_array *tr = inode->i_private; 1548 int ret; 1549 1550 /* Make a temporary dir that has no system but points to tr */ 1551 dir = kzalloc(sizeof(*dir), GFP_KERNEL); 1552 if (!dir) 1553 return -ENOMEM; 1554 1555 ret = tracing_open_generic_tr(inode, filp); 1556 if (ret < 0) { 1557 kfree(dir); 1558 return ret; 1559 } 1560 dir->tr = tr; 1561 filp->private_data = dir; 1562 1563 return 0; 1564 } 1565 1566 static int subsystem_release(struct inode *inode, struct file *file) 1567 { 1568 struct trace_subsystem_dir *dir = file->private_data; 1569 1570 trace_array_put(dir->tr); 1571 1572 /* 1573 * If dir->subsystem is NULL, then this is a temporary 1574 * descriptor that was made for a trace_array to enable 1575 * all subsystems. 1576 */ 1577 if (dir->subsystem) 1578 put_system(dir); 1579 else 1580 kfree(dir); 1581 1582 return 0; 1583 } 1584 1585 static ssize_t 1586 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt, 1587 loff_t *ppos) 1588 { 1589 struct trace_subsystem_dir *dir = filp->private_data; 1590 struct event_subsystem *system = dir->subsystem; 1591 struct trace_seq *s; 1592 int r; 1593 1594 if (*ppos) 1595 return 0; 1596 1597 s = kmalloc(sizeof(*s), GFP_KERNEL); 1598 if (!s) 1599 return -ENOMEM; 1600 1601 trace_seq_init(s); 1602 1603 print_subsystem_event_filter(system, s); 1604 r = simple_read_from_buffer(ubuf, cnt, ppos, 1605 s->buffer, trace_seq_used(s)); 1606 1607 kfree(s); 1608 1609 return r; 1610 } 1611 1612 static ssize_t 1613 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt, 1614 loff_t *ppos) 1615 { 1616 struct trace_subsystem_dir *dir = filp->private_data; 1617 char *buf; 1618 int err; 1619 1620 if (cnt >= PAGE_SIZE) 1621 return -EINVAL; 1622 1623 buf = memdup_user_nul(ubuf, cnt); 1624 if (IS_ERR(buf)) 1625 return PTR_ERR(buf); 1626 1627 err = apply_subsystem_event_filter(dir, buf); 1628 kfree(buf); 1629 if (err < 0) 1630 return err; 1631 1632 *ppos += cnt; 1633 1634 return cnt; 1635 } 1636 1637 static ssize_t 1638 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos) 1639 { 1640 int (*func)(struct trace_seq *s) = filp->private_data; 1641 struct trace_seq *s; 1642 int r; 1643 1644 if (*ppos) 1645 return 0; 1646 1647 s = kmalloc(sizeof(*s), GFP_KERNEL); 1648 if (!s) 1649 return -ENOMEM; 1650 1651 trace_seq_init(s); 1652 1653 func(s); 1654 r = simple_read_from_buffer(ubuf, cnt, ppos, 1655 s->buffer, trace_seq_used(s)); 1656 1657 kfree(s); 1658 1659 return r; 1660 } 1661 1662 static void ignore_task_cpu(void *data) 1663 { 1664 struct trace_array *tr = data; 1665 struct trace_pid_list *pid_list; 1666 struct trace_pid_list *no_pid_list; 1667 1668 /* 1669 * This function is called by on_each_cpu() while the 1670 * event_mutex is held. 1671 */ 1672 pid_list = rcu_dereference_protected(tr->filtered_pids, 1673 mutex_is_locked(&event_mutex)); 1674 no_pid_list = rcu_dereference_protected(tr->filtered_no_pids, 1675 mutex_is_locked(&event_mutex)); 1676 1677 this_cpu_write(tr->array_buffer.data->ignore_pid, 1678 trace_ignore_this_task(pid_list, no_pid_list, current)); 1679 } 1680 1681 static void register_pid_events(struct trace_array *tr) 1682 { 1683 /* 1684 * Register a probe that is called before all other probes 1685 * to set ignore_pid if next or prev do not match. 1686 * Register a probe this is called after all other probes 1687 * to only keep ignore_pid set if next pid matches. 1688 */ 1689 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre, 1690 tr, INT_MAX); 1691 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post, 1692 tr, 0); 1693 1694 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, 1695 tr, INT_MAX); 1696 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, 1697 tr, 0); 1698 1699 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, 1700 tr, INT_MAX); 1701 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, 1702 tr, 0); 1703 1704 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre, 1705 tr, INT_MAX); 1706 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post, 1707 tr, 0); 1708 } 1709 1710 static ssize_t 1711 event_pid_write(struct file *filp, const char __user *ubuf, 1712 size_t cnt, loff_t *ppos, int type) 1713 { 1714 struct seq_file *m = filp->private_data; 1715 struct trace_array *tr = m->private; 1716 struct trace_pid_list *filtered_pids = NULL; 1717 struct trace_pid_list *other_pids = NULL; 1718 struct trace_pid_list *pid_list; 1719 struct trace_event_file *file; 1720 ssize_t ret; 1721 1722 if (!cnt) 1723 return 0; 1724 1725 ret = tracing_update_buffers(); 1726 if (ret < 0) 1727 return ret; 1728 1729 mutex_lock(&event_mutex); 1730 1731 if (type == TRACE_PIDS) { 1732 filtered_pids = rcu_dereference_protected(tr->filtered_pids, 1733 lockdep_is_held(&event_mutex)); 1734 other_pids = rcu_dereference_protected(tr->filtered_no_pids, 1735 lockdep_is_held(&event_mutex)); 1736 } else { 1737 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids, 1738 lockdep_is_held(&event_mutex)); 1739 other_pids = rcu_dereference_protected(tr->filtered_pids, 1740 lockdep_is_held(&event_mutex)); 1741 } 1742 1743 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt); 1744 if (ret < 0) 1745 goto out; 1746 1747 if (type == TRACE_PIDS) 1748 rcu_assign_pointer(tr->filtered_pids, pid_list); 1749 else 1750 rcu_assign_pointer(tr->filtered_no_pids, pid_list); 1751 1752 list_for_each_entry(file, &tr->events, list) { 1753 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags); 1754 } 1755 1756 if (filtered_pids) { 1757 tracepoint_synchronize_unregister(); 1758 trace_free_pid_list(filtered_pids); 1759 } else if (pid_list && !other_pids) { 1760 register_pid_events(tr); 1761 } 1762 1763 /* 1764 * Ignoring of pids is done at task switch. But we have to 1765 * check for those tasks that are currently running. 1766 * Always do this in case a pid was appended or removed. 1767 */ 1768 on_each_cpu(ignore_task_cpu, tr, 1); 1769 1770 out: 1771 mutex_unlock(&event_mutex); 1772 1773 if (ret > 0) 1774 *ppos += ret; 1775 1776 return ret; 1777 } 1778 1779 static ssize_t 1780 ftrace_event_pid_write(struct file *filp, const char __user *ubuf, 1781 size_t cnt, loff_t *ppos) 1782 { 1783 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS); 1784 } 1785 1786 static ssize_t 1787 ftrace_event_npid_write(struct file *filp, const char __user *ubuf, 1788 size_t cnt, loff_t *ppos) 1789 { 1790 return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS); 1791 } 1792 1793 static int ftrace_event_avail_open(struct inode *inode, struct file *file); 1794 static int ftrace_event_set_open(struct inode *inode, struct file *file); 1795 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file); 1796 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file); 1797 static int ftrace_event_release(struct inode *inode, struct file *file); 1798 1799 static const struct seq_operations show_event_seq_ops = { 1800 .start = t_start, 1801 .next = t_next, 1802 .show = t_show, 1803 .stop = t_stop, 1804 }; 1805 1806 static const struct seq_operations show_set_event_seq_ops = { 1807 .start = s_start, 1808 .next = s_next, 1809 .show = t_show, 1810 .stop = t_stop, 1811 }; 1812 1813 static const struct seq_operations show_set_pid_seq_ops = { 1814 .start = p_start, 1815 .next = p_next, 1816 .show = trace_pid_show, 1817 .stop = p_stop, 1818 }; 1819 1820 static const struct seq_operations show_set_no_pid_seq_ops = { 1821 .start = np_start, 1822 .next = np_next, 1823 .show = trace_pid_show, 1824 .stop = p_stop, 1825 }; 1826 1827 static const struct file_operations ftrace_avail_fops = { 1828 .open = ftrace_event_avail_open, 1829 .read = seq_read, 1830 .llseek = seq_lseek, 1831 .release = seq_release, 1832 }; 1833 1834 static const struct file_operations ftrace_set_event_fops = { 1835 .open = ftrace_event_set_open, 1836 .read = seq_read, 1837 .write = ftrace_event_write, 1838 .llseek = seq_lseek, 1839 .release = ftrace_event_release, 1840 }; 1841 1842 static const struct file_operations ftrace_set_event_pid_fops = { 1843 .open = ftrace_event_set_pid_open, 1844 .read = seq_read, 1845 .write = ftrace_event_pid_write, 1846 .llseek = seq_lseek, 1847 .release = ftrace_event_release, 1848 }; 1849 1850 static const struct file_operations ftrace_set_event_notrace_pid_fops = { 1851 .open = ftrace_event_set_npid_open, 1852 .read = seq_read, 1853 .write = ftrace_event_npid_write, 1854 .llseek = seq_lseek, 1855 .release = ftrace_event_release, 1856 }; 1857 1858 static const struct file_operations ftrace_enable_fops = { 1859 .open = tracing_open_generic, 1860 .read = event_enable_read, 1861 .write = event_enable_write, 1862 .llseek = default_llseek, 1863 }; 1864 1865 static const struct file_operations ftrace_event_format_fops = { 1866 .open = trace_format_open, 1867 .read = seq_read, 1868 .llseek = seq_lseek, 1869 .release = seq_release, 1870 }; 1871 1872 static const struct file_operations ftrace_event_id_fops = { 1873 .read = event_id_read, 1874 .llseek = default_llseek, 1875 }; 1876 1877 static const struct file_operations ftrace_event_filter_fops = { 1878 .open = tracing_open_generic, 1879 .read = event_filter_read, 1880 .write = event_filter_write, 1881 .llseek = default_llseek, 1882 }; 1883 1884 static const struct file_operations ftrace_subsystem_filter_fops = { 1885 .open = subsystem_open, 1886 .read = subsystem_filter_read, 1887 .write = subsystem_filter_write, 1888 .llseek = default_llseek, 1889 .release = subsystem_release, 1890 }; 1891 1892 static const struct file_operations ftrace_system_enable_fops = { 1893 .open = subsystem_open, 1894 .read = system_enable_read, 1895 .write = system_enable_write, 1896 .llseek = default_llseek, 1897 .release = subsystem_release, 1898 }; 1899 1900 static const struct file_operations ftrace_tr_enable_fops = { 1901 .open = system_tr_open, 1902 .read = system_enable_read, 1903 .write = system_enable_write, 1904 .llseek = default_llseek, 1905 .release = subsystem_release, 1906 }; 1907 1908 static const struct file_operations ftrace_show_header_fops = { 1909 .open = tracing_open_generic, 1910 .read = show_header, 1911 .llseek = default_llseek, 1912 }; 1913 1914 static int 1915 ftrace_event_open(struct inode *inode, struct file *file, 1916 const struct seq_operations *seq_ops) 1917 { 1918 struct seq_file *m; 1919 int ret; 1920 1921 ret = security_locked_down(LOCKDOWN_TRACEFS); 1922 if (ret) 1923 return ret; 1924 1925 ret = seq_open(file, seq_ops); 1926 if (ret < 0) 1927 return ret; 1928 m = file->private_data; 1929 /* copy tr over to seq ops */ 1930 m->private = inode->i_private; 1931 1932 return ret; 1933 } 1934 1935 static int ftrace_event_release(struct inode *inode, struct file *file) 1936 { 1937 struct trace_array *tr = inode->i_private; 1938 1939 trace_array_put(tr); 1940 1941 return seq_release(inode, file); 1942 } 1943 1944 static int 1945 ftrace_event_avail_open(struct inode *inode, struct file *file) 1946 { 1947 const struct seq_operations *seq_ops = &show_event_seq_ops; 1948 1949 /* Checks for tracefs lockdown */ 1950 return ftrace_event_open(inode, file, seq_ops); 1951 } 1952 1953 static int 1954 ftrace_event_set_open(struct inode *inode, struct file *file) 1955 { 1956 const struct seq_operations *seq_ops = &show_set_event_seq_ops; 1957 struct trace_array *tr = inode->i_private; 1958 int ret; 1959 1960 ret = tracing_check_open_get_tr(tr); 1961 if (ret) 1962 return ret; 1963 1964 if ((file->f_mode & FMODE_WRITE) && 1965 (file->f_flags & O_TRUNC)) 1966 ftrace_clear_events(tr); 1967 1968 ret = ftrace_event_open(inode, file, seq_ops); 1969 if (ret < 0) 1970 trace_array_put(tr); 1971 return ret; 1972 } 1973 1974 static int 1975 ftrace_event_set_pid_open(struct inode *inode, struct file *file) 1976 { 1977 const struct seq_operations *seq_ops = &show_set_pid_seq_ops; 1978 struct trace_array *tr = inode->i_private; 1979 int ret; 1980 1981 ret = tracing_check_open_get_tr(tr); 1982 if (ret) 1983 return ret; 1984 1985 if ((file->f_mode & FMODE_WRITE) && 1986 (file->f_flags & O_TRUNC)) 1987 ftrace_clear_event_pids(tr, TRACE_PIDS); 1988 1989 ret = ftrace_event_open(inode, file, seq_ops); 1990 if (ret < 0) 1991 trace_array_put(tr); 1992 return ret; 1993 } 1994 1995 static int 1996 ftrace_event_set_npid_open(struct inode *inode, struct file *file) 1997 { 1998 const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops; 1999 struct trace_array *tr = inode->i_private; 2000 int ret; 2001 2002 ret = tracing_check_open_get_tr(tr); 2003 if (ret) 2004 return ret; 2005 2006 if ((file->f_mode & FMODE_WRITE) && 2007 (file->f_flags & O_TRUNC)) 2008 ftrace_clear_event_pids(tr, TRACE_NO_PIDS); 2009 2010 ret = ftrace_event_open(inode, file, seq_ops); 2011 if (ret < 0) 2012 trace_array_put(tr); 2013 return ret; 2014 } 2015 2016 static struct event_subsystem * 2017 create_new_subsystem(const char *name) 2018 { 2019 struct event_subsystem *system; 2020 2021 /* need to create new entry */ 2022 system = kmalloc(sizeof(*system), GFP_KERNEL); 2023 if (!system) 2024 return NULL; 2025 2026 system->ref_count = 1; 2027 2028 /* Only allocate if dynamic (kprobes and modules) */ 2029 system->name = kstrdup_const(name, GFP_KERNEL); 2030 if (!system->name) 2031 goto out_free; 2032 2033 system->filter = NULL; 2034 2035 system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL); 2036 if (!system->filter) 2037 goto out_free; 2038 2039 list_add(&system->list, &event_subsystems); 2040 2041 return system; 2042 2043 out_free: 2044 kfree_const(system->name); 2045 kfree(system); 2046 return NULL; 2047 } 2048 2049 static struct dentry * 2050 event_subsystem_dir(struct trace_array *tr, const char *name, 2051 struct trace_event_file *file, struct dentry *parent) 2052 { 2053 struct trace_subsystem_dir *dir; 2054 struct event_subsystem *system; 2055 struct dentry *entry; 2056 2057 /* First see if we did not already create this dir */ 2058 list_for_each_entry(dir, &tr->systems, list) { 2059 system = dir->subsystem; 2060 if (strcmp(system->name, name) == 0) { 2061 dir->nr_events++; 2062 file->system = dir; 2063 return dir->entry; 2064 } 2065 } 2066 2067 /* Now see if the system itself exists. */ 2068 list_for_each_entry(system, &event_subsystems, list) { 2069 if (strcmp(system->name, name) == 0) 2070 break; 2071 } 2072 /* Reset system variable when not found */ 2073 if (&system->list == &event_subsystems) 2074 system = NULL; 2075 2076 dir = kmalloc(sizeof(*dir), GFP_KERNEL); 2077 if (!dir) 2078 goto out_fail; 2079 2080 if (!system) { 2081 system = create_new_subsystem(name); 2082 if (!system) 2083 goto out_free; 2084 } else 2085 __get_system(system); 2086 2087 dir->entry = tracefs_create_dir(name, parent); 2088 if (!dir->entry) { 2089 pr_warn("Failed to create system directory %s\n", name); 2090 __put_system(system); 2091 goto out_free; 2092 } 2093 2094 dir->tr = tr; 2095 dir->ref_count = 1; 2096 dir->nr_events = 1; 2097 dir->subsystem = system; 2098 file->system = dir; 2099 2100 /* the ftrace system is special, do not create enable or filter files */ 2101 if (strcmp(name, "ftrace") != 0) { 2102 2103 entry = tracefs_create_file("filter", 0644, dir->entry, dir, 2104 &ftrace_subsystem_filter_fops); 2105 if (!entry) { 2106 kfree(system->filter); 2107 system->filter = NULL; 2108 pr_warn("Could not create tracefs '%s/filter' entry\n", name); 2109 } 2110 2111 trace_create_file("enable", 0644, dir->entry, dir, 2112 &ftrace_system_enable_fops); 2113 } 2114 2115 list_add(&dir->list, &tr->systems); 2116 2117 return dir->entry; 2118 2119 out_free: 2120 kfree(dir); 2121 out_fail: 2122 /* Only print this message if failed on memory allocation */ 2123 if (!dir || !system) 2124 pr_warn("No memory to create event subsystem %s\n", name); 2125 return NULL; 2126 } 2127 2128 static int 2129 event_define_fields(struct trace_event_call *call) 2130 { 2131 struct list_head *head; 2132 int ret = 0; 2133 2134 /* 2135 * Other events may have the same class. Only update 2136 * the fields if they are not already defined. 2137 */ 2138 head = trace_get_fields(call); 2139 if (list_empty(head)) { 2140 struct trace_event_fields *field = call->class->fields_array; 2141 unsigned int offset = sizeof(struct trace_entry); 2142 2143 for (; field->type; field++) { 2144 if (field->type == TRACE_FUNCTION_TYPE) { 2145 field->define_fields(call); 2146 break; 2147 } 2148 2149 offset = ALIGN(offset, field->align); 2150 ret = trace_define_field(call, field->type, field->name, 2151 offset, field->size, 2152 field->is_signed, field->filter_type); 2153 if (WARN_ON_ONCE(ret)) { 2154 pr_err("error code is %d\n", ret); 2155 break; 2156 } 2157 2158 offset += field->size; 2159 } 2160 } 2161 2162 return ret; 2163 } 2164 2165 static int 2166 event_create_dir(struct dentry *parent, struct trace_event_file *file) 2167 { 2168 struct trace_event_call *call = file->event_call; 2169 struct trace_array *tr = file->tr; 2170 struct dentry *d_events; 2171 const char *name; 2172 int ret; 2173 2174 /* 2175 * If the trace point header did not define TRACE_SYSTEM 2176 * then the system would be called "TRACE_SYSTEM". 2177 */ 2178 if (strcmp(call->class->system, TRACE_SYSTEM) != 0) { 2179 d_events = event_subsystem_dir(tr, call->class->system, file, parent); 2180 if (!d_events) 2181 return -ENOMEM; 2182 } else 2183 d_events = parent; 2184 2185 name = trace_event_name(call); 2186 file->dir = tracefs_create_dir(name, d_events); 2187 if (!file->dir) { 2188 pr_warn("Could not create tracefs '%s' directory\n", name); 2189 return -1; 2190 } 2191 2192 if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) 2193 trace_create_file("enable", 0644, file->dir, file, 2194 &ftrace_enable_fops); 2195 2196 #ifdef CONFIG_PERF_EVENTS 2197 if (call->event.type && call->class->reg) 2198 trace_create_file("id", 0444, file->dir, 2199 (void *)(long)call->event.type, 2200 &ftrace_event_id_fops); 2201 #endif 2202 2203 ret = event_define_fields(call); 2204 if (ret < 0) { 2205 pr_warn("Could not initialize trace point events/%s\n", name); 2206 return ret; 2207 } 2208 2209 /* 2210 * Only event directories that can be enabled should have 2211 * triggers or filters. 2212 */ 2213 if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) { 2214 trace_create_file("filter", 0644, file->dir, file, 2215 &ftrace_event_filter_fops); 2216 2217 trace_create_file("trigger", 0644, file->dir, file, 2218 &event_trigger_fops); 2219 } 2220 2221 #ifdef CONFIG_HIST_TRIGGERS 2222 trace_create_file("hist", 0444, file->dir, file, 2223 &event_hist_fops); 2224 #endif 2225 #ifdef CONFIG_HIST_TRIGGERS_DEBUG 2226 trace_create_file("hist_debug", 0444, file->dir, file, 2227 &event_hist_debug_fops); 2228 #endif 2229 trace_create_file("format", 0444, file->dir, call, 2230 &ftrace_event_format_fops); 2231 2232 #ifdef CONFIG_TRACE_EVENT_INJECT 2233 if (call->event.type && call->class->reg) 2234 trace_create_file("inject", 0200, file->dir, file, 2235 &event_inject_fops); 2236 #endif 2237 2238 return 0; 2239 } 2240 2241 static void remove_event_from_tracers(struct trace_event_call *call) 2242 { 2243 struct trace_event_file *file; 2244 struct trace_array *tr; 2245 2246 do_for_each_event_file_safe(tr, file) { 2247 if (file->event_call != call) 2248 continue; 2249 2250 remove_event_file_dir(file); 2251 /* 2252 * The do_for_each_event_file_safe() is 2253 * a double loop. After finding the call for this 2254 * trace_array, we use break to jump to the next 2255 * trace_array. 2256 */ 2257 break; 2258 } while_for_each_event_file(); 2259 } 2260 2261 static void event_remove(struct trace_event_call *call) 2262 { 2263 struct trace_array *tr; 2264 struct trace_event_file *file; 2265 2266 do_for_each_event_file(tr, file) { 2267 if (file->event_call != call) 2268 continue; 2269 2270 if (file->flags & EVENT_FILE_FL_WAS_ENABLED) 2271 tr->clear_trace = true; 2272 2273 ftrace_event_enable_disable(file, 0); 2274 /* 2275 * The do_for_each_event_file() is 2276 * a double loop. After finding the call for this 2277 * trace_array, we use break to jump to the next 2278 * trace_array. 2279 */ 2280 break; 2281 } while_for_each_event_file(); 2282 2283 if (call->event.funcs) 2284 __unregister_trace_event(&call->event); 2285 remove_event_from_tracers(call); 2286 list_del(&call->list); 2287 } 2288 2289 static int event_init(struct trace_event_call *call) 2290 { 2291 int ret = 0; 2292 const char *name; 2293 2294 name = trace_event_name(call); 2295 if (WARN_ON(!name)) 2296 return -EINVAL; 2297 2298 if (call->class->raw_init) { 2299 ret = call->class->raw_init(call); 2300 if (ret < 0 && ret != -ENOSYS) 2301 pr_warn("Could not initialize trace events/%s\n", name); 2302 } 2303 2304 return ret; 2305 } 2306 2307 static int 2308 __register_event(struct trace_event_call *call, struct module *mod) 2309 { 2310 int ret; 2311 2312 ret = event_init(call); 2313 if (ret < 0) 2314 return ret; 2315 2316 list_add(&call->list, &ftrace_events); 2317 call->mod = mod; 2318 2319 return 0; 2320 } 2321 2322 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len) 2323 { 2324 int rlen; 2325 int elen; 2326 2327 /* Find the length of the eval value as a string */ 2328 elen = snprintf(ptr, 0, "%ld", map->eval_value); 2329 /* Make sure there's enough room to replace the string with the value */ 2330 if (len < elen) 2331 return NULL; 2332 2333 snprintf(ptr, elen + 1, "%ld", map->eval_value); 2334 2335 /* Get the rest of the string of ptr */ 2336 rlen = strlen(ptr + len); 2337 memmove(ptr + elen, ptr + len, rlen); 2338 /* Make sure we end the new string */ 2339 ptr[elen + rlen] = 0; 2340 2341 return ptr + elen; 2342 } 2343 2344 static void update_event_printk(struct trace_event_call *call, 2345 struct trace_eval_map *map) 2346 { 2347 char *ptr; 2348 int quote = 0; 2349 int len = strlen(map->eval_string); 2350 2351 for (ptr = call->print_fmt; *ptr; ptr++) { 2352 if (*ptr == '\\') { 2353 ptr++; 2354 /* paranoid */ 2355 if (!*ptr) 2356 break; 2357 continue; 2358 } 2359 if (*ptr == '"') { 2360 quote ^= 1; 2361 continue; 2362 } 2363 if (quote) 2364 continue; 2365 if (isdigit(*ptr)) { 2366 /* skip numbers */ 2367 do { 2368 ptr++; 2369 /* Check for alpha chars like ULL */ 2370 } while (isalnum(*ptr)); 2371 if (!*ptr) 2372 break; 2373 /* 2374 * A number must have some kind of delimiter after 2375 * it, and we can ignore that too. 2376 */ 2377 continue; 2378 } 2379 if (isalpha(*ptr) || *ptr == '_') { 2380 if (strncmp(map->eval_string, ptr, len) == 0 && 2381 !isalnum(ptr[len]) && ptr[len] != '_') { 2382 ptr = eval_replace(ptr, map, len); 2383 /* enum/sizeof string smaller than value */ 2384 if (WARN_ON_ONCE(!ptr)) 2385 return; 2386 /* 2387 * No need to decrement here, as eval_replace() 2388 * returns the pointer to the character passed 2389 * the eval, and two evals can not be placed 2390 * back to back without something in between. 2391 * We can skip that something in between. 2392 */ 2393 continue; 2394 } 2395 skip_more: 2396 do { 2397 ptr++; 2398 } while (isalnum(*ptr) || *ptr == '_'); 2399 if (!*ptr) 2400 break; 2401 /* 2402 * If what comes after this variable is a '.' or 2403 * '->' then we can continue to ignore that string. 2404 */ 2405 if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) { 2406 ptr += *ptr == '.' ? 1 : 2; 2407 if (!*ptr) 2408 break; 2409 goto skip_more; 2410 } 2411 /* 2412 * Once again, we can skip the delimiter that came 2413 * after the string. 2414 */ 2415 continue; 2416 } 2417 } 2418 } 2419 2420 void trace_event_eval_update(struct trace_eval_map **map, int len) 2421 { 2422 struct trace_event_call *call, *p; 2423 const char *last_system = NULL; 2424 bool first = false; 2425 int last_i; 2426 int i; 2427 2428 down_write(&trace_event_sem); 2429 list_for_each_entry_safe(call, p, &ftrace_events, list) { 2430 /* events are usually grouped together with systems */ 2431 if (!last_system || call->class->system != last_system) { 2432 first = true; 2433 last_i = 0; 2434 last_system = call->class->system; 2435 } 2436 2437 /* 2438 * Since calls are grouped by systems, the likelyhood that the 2439 * next call in the iteration belongs to the same system as the 2440 * previous call is high. As an optimization, we skip searching 2441 * for a map[] that matches the call's system if the last call 2442 * was from the same system. That's what last_i is for. If the 2443 * call has the same system as the previous call, then last_i 2444 * will be the index of the first map[] that has a matching 2445 * system. 2446 */ 2447 for (i = last_i; i < len; i++) { 2448 if (call->class->system == map[i]->system) { 2449 /* Save the first system if need be */ 2450 if (first) { 2451 last_i = i; 2452 first = false; 2453 } 2454 update_event_printk(call, map[i]); 2455 } 2456 } 2457 } 2458 up_write(&trace_event_sem); 2459 } 2460 2461 static struct trace_event_file * 2462 trace_create_new_event(struct trace_event_call *call, 2463 struct trace_array *tr) 2464 { 2465 struct trace_event_file *file; 2466 2467 file = kmem_cache_alloc(file_cachep, GFP_TRACE); 2468 if (!file) 2469 return NULL; 2470 2471 file->event_call = call; 2472 file->tr = tr; 2473 atomic_set(&file->sm_ref, 0); 2474 atomic_set(&file->tm_ref, 0); 2475 INIT_LIST_HEAD(&file->triggers); 2476 list_add(&file->list, &tr->events); 2477 2478 return file; 2479 } 2480 2481 /* Add an event to a trace directory */ 2482 static int 2483 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr) 2484 { 2485 struct trace_event_file *file; 2486 2487 file = trace_create_new_event(call, tr); 2488 if (!file) 2489 return -ENOMEM; 2490 2491 if (eventdir_initialized) 2492 return event_create_dir(tr->event_dir, file); 2493 else 2494 return event_define_fields(call); 2495 } 2496 2497 /* 2498 * Just create a decriptor for early init. A descriptor is required 2499 * for enabling events at boot. We want to enable events before 2500 * the filesystem is initialized. 2501 */ 2502 static int 2503 __trace_early_add_new_event(struct trace_event_call *call, 2504 struct trace_array *tr) 2505 { 2506 struct trace_event_file *file; 2507 2508 file = trace_create_new_event(call, tr); 2509 if (!file) 2510 return -ENOMEM; 2511 2512 return event_define_fields(call); 2513 } 2514 2515 struct ftrace_module_file_ops; 2516 static void __add_event_to_tracers(struct trace_event_call *call); 2517 2518 /* Add an additional event_call dynamically */ 2519 int trace_add_event_call(struct trace_event_call *call) 2520 { 2521 int ret; 2522 lockdep_assert_held(&event_mutex); 2523 2524 mutex_lock(&trace_types_lock); 2525 2526 ret = __register_event(call, NULL); 2527 if (ret >= 0) 2528 __add_event_to_tracers(call); 2529 2530 mutex_unlock(&trace_types_lock); 2531 return ret; 2532 } 2533 2534 /* 2535 * Must be called under locking of trace_types_lock, event_mutex and 2536 * trace_event_sem. 2537 */ 2538 static void __trace_remove_event_call(struct trace_event_call *call) 2539 { 2540 event_remove(call); 2541 trace_destroy_fields(call); 2542 free_event_filter(call->filter); 2543 call->filter = NULL; 2544 } 2545 2546 static int probe_remove_event_call(struct trace_event_call *call) 2547 { 2548 struct trace_array *tr; 2549 struct trace_event_file *file; 2550 2551 #ifdef CONFIG_PERF_EVENTS 2552 if (call->perf_refcount) 2553 return -EBUSY; 2554 #endif 2555 do_for_each_event_file(tr, file) { 2556 if (file->event_call != call) 2557 continue; 2558 /* 2559 * We can't rely on ftrace_event_enable_disable(enable => 0) 2560 * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress 2561 * TRACE_REG_UNREGISTER. 2562 */ 2563 if (file->flags & EVENT_FILE_FL_ENABLED) 2564 return -EBUSY; 2565 /* 2566 * The do_for_each_event_file_safe() is 2567 * a double loop. After finding the call for this 2568 * trace_array, we use break to jump to the next 2569 * trace_array. 2570 */ 2571 break; 2572 } while_for_each_event_file(); 2573 2574 __trace_remove_event_call(call); 2575 2576 return 0; 2577 } 2578 2579 /* Remove an event_call */ 2580 int trace_remove_event_call(struct trace_event_call *call) 2581 { 2582 int ret; 2583 2584 lockdep_assert_held(&event_mutex); 2585 2586 mutex_lock(&trace_types_lock); 2587 down_write(&trace_event_sem); 2588 ret = probe_remove_event_call(call); 2589 up_write(&trace_event_sem); 2590 mutex_unlock(&trace_types_lock); 2591 2592 return ret; 2593 } 2594 2595 #define for_each_event(event, start, end) \ 2596 for (event = start; \ 2597 (unsigned long)event < (unsigned long)end; \ 2598 event++) 2599 2600 #ifdef CONFIG_MODULES 2601 2602 static void trace_module_add_events(struct module *mod) 2603 { 2604 struct trace_event_call **call, **start, **end; 2605 2606 if (!mod->num_trace_events) 2607 return; 2608 2609 /* Don't add infrastructure for mods without tracepoints */ 2610 if (trace_module_has_bad_taint(mod)) { 2611 pr_err("%s: module has bad taint, not creating trace events\n", 2612 mod->name); 2613 return; 2614 } 2615 2616 start = mod->trace_events; 2617 end = mod->trace_events + mod->num_trace_events; 2618 2619 for_each_event(call, start, end) { 2620 __register_event(*call, mod); 2621 __add_event_to_tracers(*call); 2622 } 2623 } 2624 2625 static void trace_module_remove_events(struct module *mod) 2626 { 2627 struct trace_event_call *call, *p; 2628 2629 down_write(&trace_event_sem); 2630 list_for_each_entry_safe(call, p, &ftrace_events, list) { 2631 if (call->mod == mod) 2632 __trace_remove_event_call(call); 2633 } 2634 up_write(&trace_event_sem); 2635 2636 /* 2637 * It is safest to reset the ring buffer if the module being unloaded 2638 * registered any events that were used. The only worry is if 2639 * a new module gets loaded, and takes on the same id as the events 2640 * of this module. When printing out the buffer, traced events left 2641 * over from this module may be passed to the new module events and 2642 * unexpected results may occur. 2643 */ 2644 tracing_reset_all_online_cpus(); 2645 } 2646 2647 static int trace_module_notify(struct notifier_block *self, 2648 unsigned long val, void *data) 2649 { 2650 struct module *mod = data; 2651 2652 mutex_lock(&event_mutex); 2653 mutex_lock(&trace_types_lock); 2654 switch (val) { 2655 case MODULE_STATE_COMING: 2656 trace_module_add_events(mod); 2657 break; 2658 case MODULE_STATE_GOING: 2659 trace_module_remove_events(mod); 2660 break; 2661 } 2662 mutex_unlock(&trace_types_lock); 2663 mutex_unlock(&event_mutex); 2664 2665 return NOTIFY_OK; 2666 } 2667 2668 static struct notifier_block trace_module_nb = { 2669 .notifier_call = trace_module_notify, 2670 .priority = 1, /* higher than trace.c module notify */ 2671 }; 2672 #endif /* CONFIG_MODULES */ 2673 2674 /* Create a new event directory structure for a trace directory. */ 2675 static void 2676 __trace_add_event_dirs(struct trace_array *tr) 2677 { 2678 struct trace_event_call *call; 2679 int ret; 2680 2681 list_for_each_entry(call, &ftrace_events, list) { 2682 ret = __trace_add_new_event(call, tr); 2683 if (ret < 0) 2684 pr_warn("Could not create directory for event %s\n", 2685 trace_event_name(call)); 2686 } 2687 } 2688 2689 /* Returns any file that matches the system and event */ 2690 struct trace_event_file * 2691 __find_event_file(struct trace_array *tr, const char *system, const char *event) 2692 { 2693 struct trace_event_file *file; 2694 struct trace_event_call *call; 2695 const char *name; 2696 2697 list_for_each_entry(file, &tr->events, list) { 2698 2699 call = file->event_call; 2700 name = trace_event_name(call); 2701 2702 if (!name || !call->class) 2703 continue; 2704 2705 if (strcmp(event, name) == 0 && 2706 strcmp(system, call->class->system) == 0) 2707 return file; 2708 } 2709 return NULL; 2710 } 2711 2712 /* Returns valid trace event files that match system and event */ 2713 struct trace_event_file * 2714 find_event_file(struct trace_array *tr, const char *system, const char *event) 2715 { 2716 struct trace_event_file *file; 2717 2718 file = __find_event_file(tr, system, event); 2719 if (!file || !file->event_call->class->reg || 2720 file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) 2721 return NULL; 2722 2723 return file; 2724 } 2725 2726 /** 2727 * trace_get_event_file - Find and return a trace event file 2728 * @instance: The name of the trace instance containing the event 2729 * @system: The name of the system containing the event 2730 * @event: The name of the event 2731 * 2732 * Return a trace event file given the trace instance name, trace 2733 * system, and trace event name. If the instance name is NULL, it 2734 * refers to the top-level trace array. 2735 * 2736 * This function will look it up and return it if found, after calling 2737 * trace_array_get() to prevent the instance from going away, and 2738 * increment the event's module refcount to prevent it from being 2739 * removed. 2740 * 2741 * To release the file, call trace_put_event_file(), which will call 2742 * trace_array_put() and decrement the event's module refcount. 2743 * 2744 * Return: The trace event on success, ERR_PTR otherwise. 2745 */ 2746 struct trace_event_file *trace_get_event_file(const char *instance, 2747 const char *system, 2748 const char *event) 2749 { 2750 struct trace_array *tr = top_trace_array(); 2751 struct trace_event_file *file = NULL; 2752 int ret = -EINVAL; 2753 2754 if (instance) { 2755 tr = trace_array_find_get(instance); 2756 if (!tr) 2757 return ERR_PTR(-ENOENT); 2758 } else { 2759 ret = trace_array_get(tr); 2760 if (ret) 2761 return ERR_PTR(ret); 2762 } 2763 2764 mutex_lock(&event_mutex); 2765 2766 file = find_event_file(tr, system, event); 2767 if (!file) { 2768 trace_array_put(tr); 2769 ret = -EINVAL; 2770 goto out; 2771 } 2772 2773 /* Don't let event modules unload while in use */ 2774 ret = try_module_get(file->event_call->mod); 2775 if (!ret) { 2776 trace_array_put(tr); 2777 ret = -EBUSY; 2778 goto out; 2779 } 2780 2781 ret = 0; 2782 out: 2783 mutex_unlock(&event_mutex); 2784 2785 if (ret) 2786 file = ERR_PTR(ret); 2787 2788 return file; 2789 } 2790 EXPORT_SYMBOL_GPL(trace_get_event_file); 2791 2792 /** 2793 * trace_put_event_file - Release a file from trace_get_event_file() 2794 * @file: The trace event file 2795 * 2796 * If a file was retrieved using trace_get_event_file(), this should 2797 * be called when it's no longer needed. It will cancel the previous 2798 * trace_array_get() called by that function, and decrement the 2799 * event's module refcount. 2800 */ 2801 void trace_put_event_file(struct trace_event_file *file) 2802 { 2803 mutex_lock(&event_mutex); 2804 module_put(file->event_call->mod); 2805 mutex_unlock(&event_mutex); 2806 2807 trace_array_put(file->tr); 2808 } 2809 EXPORT_SYMBOL_GPL(trace_put_event_file); 2810 2811 #ifdef CONFIG_DYNAMIC_FTRACE 2812 2813 /* Avoid typos */ 2814 #define ENABLE_EVENT_STR "enable_event" 2815 #define DISABLE_EVENT_STR "disable_event" 2816 2817 struct event_probe_data { 2818 struct trace_event_file *file; 2819 unsigned long count; 2820 int ref; 2821 bool enable; 2822 }; 2823 2824 static void update_event_probe(struct event_probe_data *data) 2825 { 2826 if (data->enable) 2827 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags); 2828 else 2829 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags); 2830 } 2831 2832 static void 2833 event_enable_probe(unsigned long ip, unsigned long parent_ip, 2834 struct trace_array *tr, struct ftrace_probe_ops *ops, 2835 void *data) 2836 { 2837 struct ftrace_func_mapper *mapper = data; 2838 struct event_probe_data *edata; 2839 void **pdata; 2840 2841 pdata = ftrace_func_mapper_find_ip(mapper, ip); 2842 if (!pdata || !*pdata) 2843 return; 2844 2845 edata = *pdata; 2846 update_event_probe(edata); 2847 } 2848 2849 static void 2850 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, 2851 struct trace_array *tr, struct ftrace_probe_ops *ops, 2852 void *data) 2853 { 2854 struct ftrace_func_mapper *mapper = data; 2855 struct event_probe_data *edata; 2856 void **pdata; 2857 2858 pdata = ftrace_func_mapper_find_ip(mapper, ip); 2859 if (!pdata || !*pdata) 2860 return; 2861 2862 edata = *pdata; 2863 2864 if (!edata->count) 2865 return; 2866 2867 /* Skip if the event is in a state we want to switch to */ 2868 if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED)) 2869 return; 2870 2871 if (edata->count != -1) 2872 (edata->count)--; 2873 2874 update_event_probe(edata); 2875 } 2876 2877 static int 2878 event_enable_print(struct seq_file *m, unsigned long ip, 2879 struct ftrace_probe_ops *ops, void *data) 2880 { 2881 struct ftrace_func_mapper *mapper = data; 2882 struct event_probe_data *edata; 2883 void **pdata; 2884 2885 pdata = ftrace_func_mapper_find_ip(mapper, ip); 2886 2887 if (WARN_ON_ONCE(!pdata || !*pdata)) 2888 return 0; 2889 2890 edata = *pdata; 2891 2892 seq_printf(m, "%ps:", (void *)ip); 2893 2894 seq_printf(m, "%s:%s:%s", 2895 edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR, 2896 edata->file->event_call->class->system, 2897 trace_event_name(edata->file->event_call)); 2898 2899 if (edata->count == -1) 2900 seq_puts(m, ":unlimited\n"); 2901 else 2902 seq_printf(m, ":count=%ld\n", edata->count); 2903 2904 return 0; 2905 } 2906 2907 static int 2908 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr, 2909 unsigned long ip, void *init_data, void **data) 2910 { 2911 struct ftrace_func_mapper *mapper = *data; 2912 struct event_probe_data *edata = init_data; 2913 int ret; 2914 2915 if (!mapper) { 2916 mapper = allocate_ftrace_func_mapper(); 2917 if (!mapper) 2918 return -ENODEV; 2919 *data = mapper; 2920 } 2921 2922 ret = ftrace_func_mapper_add_ip(mapper, ip, edata); 2923 if (ret < 0) 2924 return ret; 2925 2926 edata->ref++; 2927 2928 return 0; 2929 } 2930 2931 static int free_probe_data(void *data) 2932 { 2933 struct event_probe_data *edata = data; 2934 2935 edata->ref--; 2936 if (!edata->ref) { 2937 /* Remove the SOFT_MODE flag */ 2938 __ftrace_event_enable_disable(edata->file, 0, 1); 2939 module_put(edata->file->event_call->mod); 2940 kfree(edata); 2941 } 2942 return 0; 2943 } 2944 2945 static void 2946 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr, 2947 unsigned long ip, void *data) 2948 { 2949 struct ftrace_func_mapper *mapper = data; 2950 struct event_probe_data *edata; 2951 2952 if (!ip) { 2953 if (!mapper) 2954 return; 2955 free_ftrace_func_mapper(mapper, free_probe_data); 2956 return; 2957 } 2958 2959 edata = ftrace_func_mapper_remove_ip(mapper, ip); 2960 2961 if (WARN_ON_ONCE(!edata)) 2962 return; 2963 2964 if (WARN_ON_ONCE(edata->ref <= 0)) 2965 return; 2966 2967 free_probe_data(edata); 2968 } 2969 2970 static struct ftrace_probe_ops event_enable_probe_ops = { 2971 .func = event_enable_probe, 2972 .print = event_enable_print, 2973 .init = event_enable_init, 2974 .free = event_enable_free, 2975 }; 2976 2977 static struct ftrace_probe_ops event_enable_count_probe_ops = { 2978 .func = event_enable_count_probe, 2979 .print = event_enable_print, 2980 .init = event_enable_init, 2981 .free = event_enable_free, 2982 }; 2983 2984 static struct ftrace_probe_ops event_disable_probe_ops = { 2985 .func = event_enable_probe, 2986 .print = event_enable_print, 2987 .init = event_enable_init, 2988 .free = event_enable_free, 2989 }; 2990 2991 static struct ftrace_probe_ops event_disable_count_probe_ops = { 2992 .func = event_enable_count_probe, 2993 .print = event_enable_print, 2994 .init = event_enable_init, 2995 .free = event_enable_free, 2996 }; 2997 2998 static int 2999 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash, 3000 char *glob, char *cmd, char *param, int enabled) 3001 { 3002 struct trace_event_file *file; 3003 struct ftrace_probe_ops *ops; 3004 struct event_probe_data *data; 3005 const char *system; 3006 const char *event; 3007 char *number; 3008 bool enable; 3009 int ret; 3010 3011 if (!tr) 3012 return -ENODEV; 3013 3014 /* hash funcs only work with set_ftrace_filter */ 3015 if (!enabled || !param) 3016 return -EINVAL; 3017 3018 system = strsep(¶m, ":"); 3019 if (!param) 3020 return -EINVAL; 3021 3022 event = strsep(¶m, ":"); 3023 3024 mutex_lock(&event_mutex); 3025 3026 ret = -EINVAL; 3027 file = find_event_file(tr, system, event); 3028 if (!file) 3029 goto out; 3030 3031 enable = strcmp(cmd, ENABLE_EVENT_STR) == 0; 3032 3033 if (enable) 3034 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops; 3035 else 3036 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops; 3037 3038 if (glob[0] == '!') { 3039 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops); 3040 goto out; 3041 } 3042 3043 ret = -ENOMEM; 3044 3045 data = kzalloc(sizeof(*data), GFP_KERNEL); 3046 if (!data) 3047 goto out; 3048 3049 data->enable = enable; 3050 data->count = -1; 3051 data->file = file; 3052 3053 if (!param) 3054 goto out_reg; 3055 3056 number = strsep(¶m, ":"); 3057 3058 ret = -EINVAL; 3059 if (!strlen(number)) 3060 goto out_free; 3061 3062 /* 3063 * We use the callback data field (which is a pointer) 3064 * as our counter. 3065 */ 3066 ret = kstrtoul(number, 0, &data->count); 3067 if (ret) 3068 goto out_free; 3069 3070 out_reg: 3071 /* Don't let event modules unload while probe registered */ 3072 ret = try_module_get(file->event_call->mod); 3073 if (!ret) { 3074 ret = -EBUSY; 3075 goto out_free; 3076 } 3077 3078 ret = __ftrace_event_enable_disable(file, 1, 1); 3079 if (ret < 0) 3080 goto out_put; 3081 3082 ret = register_ftrace_function_probe(glob, tr, ops, data); 3083 /* 3084 * The above returns on success the # of functions enabled, 3085 * but if it didn't find any functions it returns zero. 3086 * Consider no functions a failure too. 3087 */ 3088 if (!ret) { 3089 ret = -ENOENT; 3090 goto out_disable; 3091 } else if (ret < 0) 3092 goto out_disable; 3093 /* Just return zero, not the number of enabled functions */ 3094 ret = 0; 3095 out: 3096 mutex_unlock(&event_mutex); 3097 return ret; 3098 3099 out_disable: 3100 __ftrace_event_enable_disable(file, 0, 1); 3101 out_put: 3102 module_put(file->event_call->mod); 3103 out_free: 3104 kfree(data); 3105 goto out; 3106 } 3107 3108 static struct ftrace_func_command event_enable_cmd = { 3109 .name = ENABLE_EVENT_STR, 3110 .func = event_enable_func, 3111 }; 3112 3113 static struct ftrace_func_command event_disable_cmd = { 3114 .name = DISABLE_EVENT_STR, 3115 .func = event_enable_func, 3116 }; 3117 3118 static __init int register_event_cmds(void) 3119 { 3120 int ret; 3121 3122 ret = register_ftrace_command(&event_enable_cmd); 3123 if (WARN_ON(ret < 0)) 3124 return ret; 3125 ret = register_ftrace_command(&event_disable_cmd); 3126 if (WARN_ON(ret < 0)) 3127 unregister_ftrace_command(&event_enable_cmd); 3128 return ret; 3129 } 3130 #else 3131 static inline int register_event_cmds(void) { return 0; } 3132 #endif /* CONFIG_DYNAMIC_FTRACE */ 3133 3134 /* 3135 * The top level array and trace arrays created by boot-time tracing 3136 * have already had its trace_event_file descriptors created in order 3137 * to allow for early events to be recorded. 3138 * This function is called after the tracefs has been initialized, 3139 * and we now have to create the files associated to the events. 3140 */ 3141 static void __trace_early_add_event_dirs(struct trace_array *tr) 3142 { 3143 struct trace_event_file *file; 3144 int ret; 3145 3146 3147 list_for_each_entry(file, &tr->events, list) { 3148 ret = event_create_dir(tr->event_dir, file); 3149 if (ret < 0) 3150 pr_warn("Could not create directory for event %s\n", 3151 trace_event_name(file->event_call)); 3152 } 3153 } 3154 3155 /* 3156 * For early boot up, the top trace array and the trace arrays created 3157 * by boot-time tracing require to have a list of events that can be 3158 * enabled. This must be done before the filesystem is set up in order 3159 * to allow events to be traced early. 3160 */ 3161 void __trace_early_add_events(struct trace_array *tr) 3162 { 3163 struct trace_event_call *call; 3164 int ret; 3165 3166 list_for_each_entry(call, &ftrace_events, list) { 3167 /* Early boot up should not have any modules loaded */ 3168 if (WARN_ON_ONCE(call->mod)) 3169 continue; 3170 3171 ret = __trace_early_add_new_event(call, tr); 3172 if (ret < 0) 3173 pr_warn("Could not create early event %s\n", 3174 trace_event_name(call)); 3175 } 3176 } 3177 3178 /* Remove the event directory structure for a trace directory. */ 3179 static void 3180 __trace_remove_event_dirs(struct trace_array *tr) 3181 { 3182 struct trace_event_file *file, *next; 3183 3184 list_for_each_entry_safe(file, next, &tr->events, list) 3185 remove_event_file_dir(file); 3186 } 3187 3188 static void __add_event_to_tracers(struct trace_event_call *call) 3189 { 3190 struct trace_array *tr; 3191 3192 list_for_each_entry(tr, &ftrace_trace_arrays, list) 3193 __trace_add_new_event(call, tr); 3194 } 3195 3196 extern struct trace_event_call *__start_ftrace_events[]; 3197 extern struct trace_event_call *__stop_ftrace_events[]; 3198 3199 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata; 3200 3201 static __init int setup_trace_event(char *str) 3202 { 3203 strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE); 3204 ring_buffer_expanded = true; 3205 disable_tracing_selftest("running event tracing"); 3206 3207 return 1; 3208 } 3209 __setup("trace_event=", setup_trace_event); 3210 3211 /* Expects to have event_mutex held when called */ 3212 static int 3213 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr) 3214 { 3215 struct dentry *d_events; 3216 struct dentry *entry; 3217 3218 entry = tracefs_create_file("set_event", 0644, parent, 3219 tr, &ftrace_set_event_fops); 3220 if (!entry) { 3221 pr_warn("Could not create tracefs 'set_event' entry\n"); 3222 return -ENOMEM; 3223 } 3224 3225 d_events = tracefs_create_dir("events", parent); 3226 if (!d_events) { 3227 pr_warn("Could not create tracefs 'events' directory\n"); 3228 return -ENOMEM; 3229 } 3230 3231 entry = trace_create_file("enable", 0644, d_events, 3232 tr, &ftrace_tr_enable_fops); 3233 if (!entry) { 3234 pr_warn("Could not create tracefs 'enable' entry\n"); 3235 return -ENOMEM; 3236 } 3237 3238 /* There are not as crucial, just warn if they are not created */ 3239 3240 entry = tracefs_create_file("set_event_pid", 0644, parent, 3241 tr, &ftrace_set_event_pid_fops); 3242 if (!entry) 3243 pr_warn("Could not create tracefs 'set_event_pid' entry\n"); 3244 3245 entry = tracefs_create_file("set_event_notrace_pid", 0644, parent, 3246 tr, &ftrace_set_event_notrace_pid_fops); 3247 if (!entry) 3248 pr_warn("Could not create tracefs 'set_event_notrace_pid' entry\n"); 3249 3250 /* ring buffer internal formats */ 3251 entry = trace_create_file("header_page", 0444, d_events, 3252 ring_buffer_print_page_header, 3253 &ftrace_show_header_fops); 3254 if (!entry) 3255 pr_warn("Could not create tracefs 'header_page' entry\n"); 3256 3257 entry = trace_create_file("header_event", 0444, d_events, 3258 ring_buffer_print_entry_header, 3259 &ftrace_show_header_fops); 3260 if (!entry) 3261 pr_warn("Could not create tracefs 'header_event' entry\n"); 3262 3263 tr->event_dir = d_events; 3264 3265 return 0; 3266 } 3267 3268 /** 3269 * event_trace_add_tracer - add a instance of a trace_array to events 3270 * @parent: The parent dentry to place the files/directories for events in 3271 * @tr: The trace array associated with these events 3272 * 3273 * When a new instance is created, it needs to set up its events 3274 * directory, as well as other files associated with events. It also 3275 * creates the event hierarchy in the @parent/events directory. 3276 * 3277 * Returns 0 on success. 3278 * 3279 * Must be called with event_mutex held. 3280 */ 3281 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr) 3282 { 3283 int ret; 3284 3285 lockdep_assert_held(&event_mutex); 3286 3287 ret = create_event_toplevel_files(parent, tr); 3288 if (ret) 3289 goto out; 3290 3291 down_write(&trace_event_sem); 3292 /* If tr already has the event list, it is initialized in early boot. */ 3293 if (unlikely(!list_empty(&tr->events))) 3294 __trace_early_add_event_dirs(tr); 3295 else 3296 __trace_add_event_dirs(tr); 3297 up_write(&trace_event_sem); 3298 3299 out: 3300 return ret; 3301 } 3302 3303 /* 3304 * The top trace array already had its file descriptors created. 3305 * Now the files themselves need to be created. 3306 */ 3307 static __init int 3308 early_event_add_tracer(struct dentry *parent, struct trace_array *tr) 3309 { 3310 int ret; 3311 3312 mutex_lock(&event_mutex); 3313 3314 ret = create_event_toplevel_files(parent, tr); 3315 if (ret) 3316 goto out_unlock; 3317 3318 down_write(&trace_event_sem); 3319 __trace_early_add_event_dirs(tr); 3320 up_write(&trace_event_sem); 3321 3322 out_unlock: 3323 mutex_unlock(&event_mutex); 3324 3325 return ret; 3326 } 3327 3328 /* Must be called with event_mutex held */ 3329 int event_trace_del_tracer(struct trace_array *tr) 3330 { 3331 lockdep_assert_held(&event_mutex); 3332 3333 /* Disable any event triggers and associated soft-disabled events */ 3334 clear_event_triggers(tr); 3335 3336 /* Clear the pid list */ 3337 __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS); 3338 3339 /* Disable any running events */ 3340 __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0); 3341 3342 /* Make sure no more events are being executed */ 3343 tracepoint_synchronize_unregister(); 3344 3345 down_write(&trace_event_sem); 3346 __trace_remove_event_dirs(tr); 3347 tracefs_remove(tr->event_dir); 3348 up_write(&trace_event_sem); 3349 3350 tr->event_dir = NULL; 3351 3352 return 0; 3353 } 3354 3355 static __init int event_trace_memsetup(void) 3356 { 3357 field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC); 3358 file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC); 3359 return 0; 3360 } 3361 3362 static __init void 3363 early_enable_events(struct trace_array *tr, bool disable_first) 3364 { 3365 char *buf = bootup_event_buf; 3366 char *token; 3367 int ret; 3368 3369 while (true) { 3370 token = strsep(&buf, ","); 3371 3372 if (!token) 3373 break; 3374 3375 if (*token) { 3376 /* Restarting syscalls requires that we stop them first */ 3377 if (disable_first) 3378 ftrace_set_clr_event(tr, token, 0); 3379 3380 ret = ftrace_set_clr_event(tr, token, 1); 3381 if (ret) 3382 pr_warn("Failed to enable trace event: %s\n", token); 3383 } 3384 3385 /* Put back the comma to allow this to be called again */ 3386 if (buf) 3387 *(buf - 1) = ','; 3388 } 3389 } 3390 3391 static __init int event_trace_enable(void) 3392 { 3393 struct trace_array *tr = top_trace_array(); 3394 struct trace_event_call **iter, *call; 3395 int ret; 3396 3397 if (!tr) 3398 return -ENODEV; 3399 3400 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) { 3401 3402 call = *iter; 3403 ret = event_init(call); 3404 if (!ret) 3405 list_add(&call->list, &ftrace_events); 3406 } 3407 3408 /* 3409 * We need the top trace array to have a working set of trace 3410 * points at early init, before the debug files and directories 3411 * are created. Create the file entries now, and attach them 3412 * to the actual file dentries later. 3413 */ 3414 __trace_early_add_events(tr); 3415 3416 early_enable_events(tr, false); 3417 3418 trace_printk_start_comm(); 3419 3420 register_event_cmds(); 3421 3422 register_trigger_cmds(); 3423 3424 return 0; 3425 } 3426 3427 /* 3428 * event_trace_enable() is called from trace_event_init() first to 3429 * initialize events and perhaps start any events that are on the 3430 * command line. Unfortunately, there are some events that will not 3431 * start this early, like the system call tracepoints that need 3432 * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But 3433 * event_trace_enable() is called before pid 1 starts, and this flag 3434 * is never set, making the syscall tracepoint never get reached, but 3435 * the event is enabled regardless (and not doing anything). 3436 */ 3437 static __init int event_trace_enable_again(void) 3438 { 3439 struct trace_array *tr; 3440 3441 tr = top_trace_array(); 3442 if (!tr) 3443 return -ENODEV; 3444 3445 early_enable_events(tr, true); 3446 3447 return 0; 3448 } 3449 3450 early_initcall(event_trace_enable_again); 3451 3452 /* Init fields which doesn't related to the tracefs */ 3453 static __init int event_trace_init_fields(void) 3454 { 3455 if (trace_define_generic_fields()) 3456 pr_warn("tracing: Failed to allocated generic fields"); 3457 3458 if (trace_define_common_fields()) 3459 pr_warn("tracing: Failed to allocate common fields"); 3460 3461 return 0; 3462 } 3463 3464 __init int event_trace_init(void) 3465 { 3466 struct trace_array *tr; 3467 struct dentry *entry; 3468 int ret; 3469 3470 tr = top_trace_array(); 3471 if (!tr) 3472 return -ENODEV; 3473 3474 entry = tracefs_create_file("available_events", 0444, NULL, 3475 tr, &ftrace_avail_fops); 3476 if (!entry) 3477 pr_warn("Could not create tracefs 'available_events' entry\n"); 3478 3479 ret = early_event_add_tracer(NULL, tr); 3480 if (ret) 3481 return ret; 3482 3483 #ifdef CONFIG_MODULES 3484 ret = register_module_notifier(&trace_module_nb); 3485 if (ret) 3486 pr_warn("Failed to register trace events module notifier\n"); 3487 #endif 3488 3489 eventdir_initialized = true; 3490 3491 return 0; 3492 } 3493 3494 void __init trace_event_init(void) 3495 { 3496 event_trace_memsetup(); 3497 init_ftrace_syscalls(); 3498 event_trace_enable(); 3499 event_trace_init_fields(); 3500 } 3501 3502 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST 3503 3504 static DEFINE_SPINLOCK(test_spinlock); 3505 static DEFINE_SPINLOCK(test_spinlock_irq); 3506 static DEFINE_MUTEX(test_mutex); 3507 3508 static __init void test_work(struct work_struct *dummy) 3509 { 3510 spin_lock(&test_spinlock); 3511 spin_lock_irq(&test_spinlock_irq); 3512 udelay(1); 3513 spin_unlock_irq(&test_spinlock_irq); 3514 spin_unlock(&test_spinlock); 3515 3516 mutex_lock(&test_mutex); 3517 msleep(1); 3518 mutex_unlock(&test_mutex); 3519 } 3520 3521 static __init int event_test_thread(void *unused) 3522 { 3523 void *test_malloc; 3524 3525 test_malloc = kmalloc(1234, GFP_KERNEL); 3526 if (!test_malloc) 3527 pr_info("failed to kmalloc\n"); 3528 3529 schedule_on_each_cpu(test_work); 3530 3531 kfree(test_malloc); 3532 3533 set_current_state(TASK_INTERRUPTIBLE); 3534 while (!kthread_should_stop()) { 3535 schedule(); 3536 set_current_state(TASK_INTERRUPTIBLE); 3537 } 3538 __set_current_state(TASK_RUNNING); 3539 3540 return 0; 3541 } 3542 3543 /* 3544 * Do various things that may trigger events. 3545 */ 3546 static __init void event_test_stuff(void) 3547 { 3548 struct task_struct *test_thread; 3549 3550 test_thread = kthread_run(event_test_thread, NULL, "test-events"); 3551 msleep(1); 3552 kthread_stop(test_thread); 3553 } 3554 3555 /* 3556 * For every trace event defined, we will test each trace point separately, 3557 * and then by groups, and finally all trace points. 3558 */ 3559 static __init void event_trace_self_tests(void) 3560 { 3561 struct trace_subsystem_dir *dir; 3562 struct trace_event_file *file; 3563 struct trace_event_call *call; 3564 struct event_subsystem *system; 3565 struct trace_array *tr; 3566 int ret; 3567 3568 tr = top_trace_array(); 3569 if (!tr) 3570 return; 3571 3572 pr_info("Running tests on trace events:\n"); 3573 3574 list_for_each_entry(file, &tr->events, list) { 3575 3576 call = file->event_call; 3577 3578 /* Only test those that have a probe */ 3579 if (!call->class || !call->class->probe) 3580 continue; 3581 3582 /* 3583 * Testing syscall events here is pretty useless, but 3584 * we still do it if configured. But this is time consuming. 3585 * What we really need is a user thread to perform the 3586 * syscalls as we test. 3587 */ 3588 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS 3589 if (call->class->system && 3590 strcmp(call->class->system, "syscalls") == 0) 3591 continue; 3592 #endif 3593 3594 pr_info("Testing event %s: ", trace_event_name(call)); 3595 3596 /* 3597 * If an event is already enabled, someone is using 3598 * it and the self test should not be on. 3599 */ 3600 if (file->flags & EVENT_FILE_FL_ENABLED) { 3601 pr_warn("Enabled event during self test!\n"); 3602 WARN_ON_ONCE(1); 3603 continue; 3604 } 3605 3606 ftrace_event_enable_disable(file, 1); 3607 event_test_stuff(); 3608 ftrace_event_enable_disable(file, 0); 3609 3610 pr_cont("OK\n"); 3611 } 3612 3613 /* Now test at the sub system level */ 3614 3615 pr_info("Running tests on trace event systems:\n"); 3616 3617 list_for_each_entry(dir, &tr->systems, list) { 3618 3619 system = dir->subsystem; 3620 3621 /* the ftrace system is special, skip it */ 3622 if (strcmp(system->name, "ftrace") == 0) 3623 continue; 3624 3625 pr_info("Testing event system %s: ", system->name); 3626 3627 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1); 3628 if (WARN_ON_ONCE(ret)) { 3629 pr_warn("error enabling system %s\n", 3630 system->name); 3631 continue; 3632 } 3633 3634 event_test_stuff(); 3635 3636 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0); 3637 if (WARN_ON_ONCE(ret)) { 3638 pr_warn("error disabling system %s\n", 3639 system->name); 3640 continue; 3641 } 3642 3643 pr_cont("OK\n"); 3644 } 3645 3646 /* Test with all events enabled */ 3647 3648 pr_info("Running tests on all trace events:\n"); 3649 pr_info("Testing all events: "); 3650 3651 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1); 3652 if (WARN_ON_ONCE(ret)) { 3653 pr_warn("error enabling all events\n"); 3654 return; 3655 } 3656 3657 event_test_stuff(); 3658 3659 /* reset sysname */ 3660 ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0); 3661 if (WARN_ON_ONCE(ret)) { 3662 pr_warn("error disabling all events\n"); 3663 return; 3664 } 3665 3666 pr_cont("OK\n"); 3667 } 3668 3669 #ifdef CONFIG_FUNCTION_TRACER 3670 3671 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable); 3672 3673 static struct trace_event_file event_trace_file __initdata; 3674 3675 static void __init 3676 function_test_events_call(unsigned long ip, unsigned long parent_ip, 3677 struct ftrace_ops *op, struct ftrace_regs *regs) 3678 { 3679 struct trace_buffer *buffer; 3680 struct ring_buffer_event *event; 3681 struct ftrace_entry *entry; 3682 unsigned int trace_ctx; 3683 long disabled; 3684 int cpu; 3685 3686 trace_ctx = tracing_gen_ctx(); 3687 preempt_disable_notrace(); 3688 cpu = raw_smp_processor_id(); 3689 disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu)); 3690 3691 if (disabled != 1) 3692 goto out; 3693 3694 event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file, 3695 TRACE_FN, sizeof(*entry), 3696 trace_ctx); 3697 if (!event) 3698 goto out; 3699 entry = ring_buffer_event_data(event); 3700 entry->ip = ip; 3701 entry->parent_ip = parent_ip; 3702 3703 event_trigger_unlock_commit(&event_trace_file, buffer, event, 3704 entry, trace_ctx); 3705 out: 3706 atomic_dec(&per_cpu(ftrace_test_event_disable, cpu)); 3707 preempt_enable_notrace(); 3708 } 3709 3710 static struct ftrace_ops trace_ops __initdata = 3711 { 3712 .func = function_test_events_call, 3713 }; 3714 3715 static __init void event_trace_self_test_with_function(void) 3716 { 3717 int ret; 3718 3719 event_trace_file.tr = top_trace_array(); 3720 if (WARN_ON(!event_trace_file.tr)) 3721 return; 3722 3723 ret = register_ftrace_function(&trace_ops); 3724 if (WARN_ON(ret < 0)) { 3725 pr_info("Failed to enable function tracer for event tests\n"); 3726 return; 3727 } 3728 pr_info("Running tests again, along with the function tracer\n"); 3729 event_trace_self_tests(); 3730 unregister_ftrace_function(&trace_ops); 3731 } 3732 #else 3733 static __init void event_trace_self_test_with_function(void) 3734 { 3735 } 3736 #endif 3737 3738 static __init int event_trace_self_tests_init(void) 3739 { 3740 if (!tracing_selftest_disabled) { 3741 event_trace_self_tests(); 3742 event_trace_self_test_with_function(); 3743 } 3744 3745 return 0; 3746 } 3747 3748 late_initcall(event_trace_self_tests_init); 3749 3750 #endif 3751