xref: /linux/include/linux/trace_events.h (revision c6fbb759)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 
3 #ifndef _LINUX_TRACE_EVENT_H
4 #define _LINUX_TRACE_EVENT_H
5 
6 #include <linux/ring_buffer.h>
7 #include <linux/trace_seq.h>
8 #include <linux/percpu.h>
9 #include <linux/hardirq.h>
10 #include <linux/perf_event.h>
11 #include <linux/tracepoint.h>
12 
13 struct trace_array;
14 struct array_buffer;
15 struct tracer;
16 struct dentry;
17 struct bpf_prog;
18 union bpf_attr;
19 
20 const char *trace_print_flags_seq(struct trace_seq *p, const char *delim,
21 				  unsigned long flags,
22 				  const struct trace_print_flags *flag_array);
23 
24 const char *trace_print_symbols_seq(struct trace_seq *p, unsigned long val,
25 				    const struct trace_print_flags *symbol_array);
26 
27 #if BITS_PER_LONG == 32
28 const char *trace_print_flags_seq_u64(struct trace_seq *p, const char *delim,
29 		      unsigned long long flags,
30 		      const struct trace_print_flags_u64 *flag_array);
31 
32 const char *trace_print_symbols_seq_u64(struct trace_seq *p,
33 					unsigned long long val,
34 					const struct trace_print_flags_u64
35 								 *symbol_array);
36 #endif
37 
38 const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
39 				    unsigned int bitmask_size);
40 
41 const char *trace_print_hex_seq(struct trace_seq *p,
42 				const unsigned char *buf, int len,
43 				bool concatenate);
44 
45 const char *trace_print_array_seq(struct trace_seq *p,
46 				   const void *buf, int count,
47 				   size_t el_size);
48 
49 const char *
50 trace_print_hex_dump_seq(struct trace_seq *p, const char *prefix_str,
51 			 int prefix_type, int rowsize, int groupsize,
52 			 const void *buf, size_t len, bool ascii);
53 
54 struct trace_iterator;
55 struct trace_event;
56 
57 int trace_raw_output_prep(struct trace_iterator *iter,
58 			  struct trace_event *event);
59 extern __printf(2, 3)
60 void trace_event_printf(struct trace_iterator *iter, const char *fmt, ...);
61 
62 /*
63  * The trace entry - the most basic unit of tracing. This is what
64  * is printed in the end as a single line in the trace output, such as:
65  *
66  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
67  */
68 struct trace_entry {
69 	unsigned short		type;
70 	unsigned char		flags;
71 	unsigned char		preempt_count;
72 	int			pid;
73 };
74 
75 #define TRACE_EVENT_TYPE_MAX						\
76 	((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
77 
78 /*
79  * Trace iterator - used by printout routines who present trace
80  * results to users and which routines might sleep, etc:
81  */
82 struct trace_iterator {
83 	struct trace_array	*tr;
84 	struct tracer		*trace;
85 	struct array_buffer	*array_buffer;
86 	void			*private;
87 	int			cpu_file;
88 	struct mutex		mutex;
89 	struct ring_buffer_iter	**buffer_iter;
90 	unsigned long		iter_flags;
91 	void			*temp;	/* temp holder */
92 	unsigned int		temp_size;
93 	char			*fmt;	/* modified format holder */
94 	unsigned int		fmt_size;
95 	long			wait_index;
96 
97 	/* trace_seq for __print_flags() and __print_symbolic() etc. */
98 	struct trace_seq	tmp_seq;
99 
100 	cpumask_var_t		started;
101 
102 	/* it's true when current open file is snapshot */
103 	bool			snapshot;
104 
105 	/* The below is zeroed out in pipe_read */
106 	struct trace_seq	seq;
107 	struct trace_entry	*ent;
108 	unsigned long		lost_events;
109 	int			leftover;
110 	int			ent_size;
111 	int			cpu;
112 	u64			ts;
113 
114 	loff_t			pos;
115 	long			idx;
116 
117 	/* All new field here will be zeroed out in pipe_read */
118 };
119 
120 enum trace_iter_flags {
121 	TRACE_FILE_LAT_FMT	= 1,
122 	TRACE_FILE_ANNOTATE	= 2,
123 	TRACE_FILE_TIME_IN_NS	= 4,
124 };
125 
126 
127 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
128 				      int flags, struct trace_event *event);
129 
130 struct trace_event_functions {
131 	trace_print_func	trace;
132 	trace_print_func	raw;
133 	trace_print_func	hex;
134 	trace_print_func	binary;
135 };
136 
137 struct trace_event {
138 	struct hlist_node		node;
139 	struct list_head		list;
140 	int				type;
141 	struct trace_event_functions	*funcs;
142 };
143 
144 extern int register_trace_event(struct trace_event *event);
145 extern int unregister_trace_event(struct trace_event *event);
146 
147 /* Return values for print_line callback */
148 enum print_line_t {
149 	TRACE_TYPE_PARTIAL_LINE	= 0,	/* Retry after flushing the seq */
150 	TRACE_TYPE_HANDLED	= 1,
151 	TRACE_TYPE_UNHANDLED	= 2,	/* Relay to other output functions */
152 	TRACE_TYPE_NO_CONSUME	= 3	/* Handled but ask to not consume */
153 };
154 
155 enum print_line_t trace_handle_return(struct trace_seq *s);
156 
157 static inline void tracing_generic_entry_update(struct trace_entry *entry,
158 						unsigned short type,
159 						unsigned int trace_ctx)
160 {
161 	entry->preempt_count		= trace_ctx & 0xff;
162 	entry->pid			= current->pid;
163 	entry->type			= type;
164 	entry->flags =			trace_ctx >> 16;
165 }
166 
167 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status);
168 
169 enum trace_flag_type {
170 	TRACE_FLAG_IRQS_OFF		= 0x01,
171 	TRACE_FLAG_IRQS_NOSUPPORT	= 0x02,
172 	TRACE_FLAG_NEED_RESCHED		= 0x04,
173 	TRACE_FLAG_HARDIRQ		= 0x08,
174 	TRACE_FLAG_SOFTIRQ		= 0x10,
175 	TRACE_FLAG_PREEMPT_RESCHED	= 0x20,
176 	TRACE_FLAG_NMI			= 0x40,
177 	TRACE_FLAG_BH_OFF		= 0x80,
178 };
179 
180 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
181 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
182 {
183 	unsigned int irq_status = irqs_disabled_flags(irqflags) ?
184 		TRACE_FLAG_IRQS_OFF : 0;
185 	return tracing_gen_ctx_irq_test(irq_status);
186 }
187 static inline unsigned int tracing_gen_ctx(void)
188 {
189 	unsigned long irqflags;
190 
191 	local_save_flags(irqflags);
192 	return tracing_gen_ctx_flags(irqflags);
193 }
194 #else
195 
196 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
197 {
198 	return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
199 }
200 static inline unsigned int tracing_gen_ctx(void)
201 {
202 	return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
203 }
204 #endif
205 
206 static inline unsigned int tracing_gen_ctx_dec(void)
207 {
208 	unsigned int trace_ctx;
209 
210 	trace_ctx = tracing_gen_ctx();
211 	/*
212 	 * Subtract one from the preemption counter if preemption is enabled,
213 	 * see trace_event_buffer_reserve()for details.
214 	 */
215 	if (IS_ENABLED(CONFIG_PREEMPTION))
216 		trace_ctx--;
217 	return trace_ctx;
218 }
219 
220 struct trace_event_file;
221 
222 struct ring_buffer_event *
223 trace_event_buffer_lock_reserve(struct trace_buffer **current_buffer,
224 				struct trace_event_file *trace_file,
225 				int type, unsigned long len,
226 				unsigned int trace_ctx);
227 
228 #define TRACE_RECORD_CMDLINE	BIT(0)
229 #define TRACE_RECORD_TGID	BIT(1)
230 
231 void tracing_record_taskinfo(struct task_struct *task, int flags);
232 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
233 					  struct task_struct *next, int flags);
234 
235 void tracing_record_cmdline(struct task_struct *task);
236 void tracing_record_tgid(struct task_struct *task);
237 
238 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
239 
240 struct event_filter;
241 
242 enum trace_reg {
243 	TRACE_REG_REGISTER,
244 	TRACE_REG_UNREGISTER,
245 #ifdef CONFIG_PERF_EVENTS
246 	TRACE_REG_PERF_REGISTER,
247 	TRACE_REG_PERF_UNREGISTER,
248 	TRACE_REG_PERF_OPEN,
249 	TRACE_REG_PERF_CLOSE,
250 	/*
251 	 * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
252 	 * custom action was taken and the default action is not to be
253 	 * performed.
254 	 */
255 	TRACE_REG_PERF_ADD,
256 	TRACE_REG_PERF_DEL,
257 #endif
258 };
259 
260 struct trace_event_call;
261 
262 #define TRACE_FUNCTION_TYPE ((const char *)~0UL)
263 
264 struct trace_event_fields {
265 	const char *type;
266 	union {
267 		struct {
268 			const char *name;
269 			const int  size;
270 			const int  align;
271 			const int  is_signed;
272 			const int  filter_type;
273 		};
274 		int (*define_fields)(struct trace_event_call *);
275 	};
276 };
277 
278 struct trace_event_class {
279 	const char		*system;
280 	void			*probe;
281 #ifdef CONFIG_PERF_EVENTS
282 	void			*perf_probe;
283 #endif
284 	int			(*reg)(struct trace_event_call *event,
285 				       enum trace_reg type, void *data);
286 	struct trace_event_fields *fields_array;
287 	struct list_head	*(*get_fields)(struct trace_event_call *);
288 	struct list_head	fields;
289 	int			(*raw_init)(struct trace_event_call *);
290 };
291 
292 extern int trace_event_reg(struct trace_event_call *event,
293 			    enum trace_reg type, void *data);
294 
295 struct trace_event_buffer {
296 	struct trace_buffer		*buffer;
297 	struct ring_buffer_event	*event;
298 	struct trace_event_file		*trace_file;
299 	void				*entry;
300 	unsigned int			trace_ctx;
301 	struct pt_regs			*regs;
302 };
303 
304 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
305 				  struct trace_event_file *trace_file,
306 				  unsigned long len);
307 
308 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
309 
310 enum {
311 	TRACE_EVENT_FL_FILTERED_BIT,
312 	TRACE_EVENT_FL_CAP_ANY_BIT,
313 	TRACE_EVENT_FL_NO_SET_FILTER_BIT,
314 	TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
315 	TRACE_EVENT_FL_TRACEPOINT_BIT,
316 	TRACE_EVENT_FL_DYNAMIC_BIT,
317 	TRACE_EVENT_FL_KPROBE_BIT,
318 	TRACE_EVENT_FL_UPROBE_BIT,
319 	TRACE_EVENT_FL_EPROBE_BIT,
320 	TRACE_EVENT_FL_CUSTOM_BIT,
321 };
322 
323 /*
324  * Event flags:
325  *  FILTERED	  - The event has a filter attached
326  *  CAP_ANY	  - Any user can enable for perf
327  *  NO_SET_FILTER - Set when filter has error and is to be ignored
328  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
329  *  TRACEPOINT    - Event is a tracepoint
330  *  DYNAMIC       - Event is a dynamic event (created at run time)
331  *  KPROBE        - Event is a kprobe
332  *  UPROBE        - Event is a uprobe
333  *  EPROBE        - Event is an event probe
334  *  CUSTOM        - Event is a custom event (to be attached to an exsiting tracepoint)
335  *                   This is set when the custom event has not been attached
336  *                   to a tracepoint yet, then it is cleared when it is.
337  */
338 enum {
339 	TRACE_EVENT_FL_FILTERED		= (1 << TRACE_EVENT_FL_FILTERED_BIT),
340 	TRACE_EVENT_FL_CAP_ANY		= (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
341 	TRACE_EVENT_FL_NO_SET_FILTER	= (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
342 	TRACE_EVENT_FL_IGNORE_ENABLE	= (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
343 	TRACE_EVENT_FL_TRACEPOINT	= (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
344 	TRACE_EVENT_FL_DYNAMIC		= (1 << TRACE_EVENT_FL_DYNAMIC_BIT),
345 	TRACE_EVENT_FL_KPROBE		= (1 << TRACE_EVENT_FL_KPROBE_BIT),
346 	TRACE_EVENT_FL_UPROBE		= (1 << TRACE_EVENT_FL_UPROBE_BIT),
347 	TRACE_EVENT_FL_EPROBE		= (1 << TRACE_EVENT_FL_EPROBE_BIT),
348 	TRACE_EVENT_FL_CUSTOM		= (1 << TRACE_EVENT_FL_CUSTOM_BIT),
349 };
350 
351 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
352 
353 struct trace_event_call {
354 	struct list_head	list;
355 	struct trace_event_class *class;
356 	union {
357 		char			*name;
358 		/* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
359 		struct tracepoint	*tp;
360 	};
361 	struct trace_event	event;
362 	char			*print_fmt;
363 	struct event_filter	*filter;
364 	/*
365 	 * Static events can disappear with modules,
366 	 * where as dynamic ones need their own ref count.
367 	 */
368 	union {
369 		void				*module;
370 		atomic_t			refcnt;
371 	};
372 	void			*data;
373 
374 	/* See the TRACE_EVENT_FL_* flags above */
375 	int			flags; /* static flags of different events */
376 
377 #ifdef CONFIG_PERF_EVENTS
378 	int				perf_refcount;
379 	struct hlist_head __percpu	*perf_events;
380 	struct bpf_prog_array __rcu	*prog_array;
381 
382 	int	(*perf_perm)(struct trace_event_call *,
383 			     struct perf_event *);
384 #endif
385 };
386 
387 #ifdef CONFIG_DYNAMIC_EVENTS
388 bool trace_event_dyn_try_get_ref(struct trace_event_call *call);
389 void trace_event_dyn_put_ref(struct trace_event_call *call);
390 bool trace_event_dyn_busy(struct trace_event_call *call);
391 #else
392 static inline bool trace_event_dyn_try_get_ref(struct trace_event_call *call)
393 {
394 	/* Without DYNAMIC_EVENTS configured, nothing should be calling this */
395 	return false;
396 }
397 static inline void trace_event_dyn_put_ref(struct trace_event_call *call)
398 {
399 }
400 static inline bool trace_event_dyn_busy(struct trace_event_call *call)
401 {
402 	/* Nothing should call this without DYNAIMIC_EVENTS configured. */
403 	return true;
404 }
405 #endif
406 
407 static inline bool trace_event_try_get_ref(struct trace_event_call *call)
408 {
409 	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
410 		return trace_event_dyn_try_get_ref(call);
411 	else
412 		return try_module_get(call->module);
413 }
414 
415 static inline void trace_event_put_ref(struct trace_event_call *call)
416 {
417 	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
418 		trace_event_dyn_put_ref(call);
419 	else
420 		module_put(call->module);
421 }
422 
423 #ifdef CONFIG_PERF_EVENTS
424 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
425 {
426 	/*
427 	 * This inline function checks whether call->prog_array
428 	 * is valid or not. The function is called in various places,
429 	 * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
430 	 *
431 	 * If this function returns true, and later call->prog_array
432 	 * becomes false inside rcu_read_lock/unlock region,
433 	 * we bail out then. If this function return false,
434 	 * there is a risk that we might miss a few events if the checking
435 	 * were delayed until inside rcu_read_lock/unlock region and
436 	 * call->prog_array happened to become non-NULL then.
437 	 *
438 	 * Here, READ_ONCE() is used instead of rcu_access_pointer().
439 	 * rcu_access_pointer() requires the actual definition of
440 	 * "struct bpf_prog_array" while READ_ONCE() only needs
441 	 * a declaration of the same type.
442 	 */
443 	return !!READ_ONCE(call->prog_array);
444 }
445 #endif
446 
447 static inline const char *
448 trace_event_name(struct trace_event_call *call)
449 {
450 	if (call->flags & TRACE_EVENT_FL_CUSTOM)
451 		return call->name;
452 	else if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
453 		return call->tp ? call->tp->name : NULL;
454 	else
455 		return call->name;
456 }
457 
458 static inline struct list_head *
459 trace_get_fields(struct trace_event_call *event_call)
460 {
461 	if (!event_call->class->get_fields)
462 		return &event_call->class->fields;
463 	return event_call->class->get_fields(event_call);
464 }
465 
466 struct trace_subsystem_dir;
467 
468 enum {
469 	EVENT_FILE_FL_ENABLED_BIT,
470 	EVENT_FILE_FL_RECORDED_CMD_BIT,
471 	EVENT_FILE_FL_RECORDED_TGID_BIT,
472 	EVENT_FILE_FL_FILTERED_BIT,
473 	EVENT_FILE_FL_NO_SET_FILTER_BIT,
474 	EVENT_FILE_FL_SOFT_MODE_BIT,
475 	EVENT_FILE_FL_SOFT_DISABLED_BIT,
476 	EVENT_FILE_FL_TRIGGER_MODE_BIT,
477 	EVENT_FILE_FL_TRIGGER_COND_BIT,
478 	EVENT_FILE_FL_PID_FILTER_BIT,
479 	EVENT_FILE_FL_WAS_ENABLED_BIT,
480 };
481 
482 extern struct trace_event_file *trace_get_event_file(const char *instance,
483 						     const char *system,
484 						     const char *event);
485 extern void trace_put_event_file(struct trace_event_file *file);
486 
487 #define MAX_DYNEVENT_CMD_LEN	(2048)
488 
489 enum dynevent_type {
490 	DYNEVENT_TYPE_SYNTH = 1,
491 	DYNEVENT_TYPE_KPROBE,
492 	DYNEVENT_TYPE_NONE,
493 };
494 
495 struct dynevent_cmd;
496 
497 typedef int (*dynevent_create_fn_t)(struct dynevent_cmd *cmd);
498 
499 struct dynevent_cmd {
500 	struct seq_buf		seq;
501 	const char		*event_name;
502 	unsigned int		n_fields;
503 	enum dynevent_type	type;
504 	dynevent_create_fn_t	run_command;
505 	void			*private_data;
506 };
507 
508 extern int dynevent_create(struct dynevent_cmd *cmd);
509 
510 extern int synth_event_delete(const char *name);
511 
512 extern void synth_event_cmd_init(struct dynevent_cmd *cmd,
513 				 char *buf, int maxlen);
514 
515 extern int __synth_event_gen_cmd_start(struct dynevent_cmd *cmd,
516 				       const char *name,
517 				       struct module *mod, ...);
518 
519 #define synth_event_gen_cmd_start(cmd, name, mod, ...)	\
520 	__synth_event_gen_cmd_start(cmd, name, mod, ## __VA_ARGS__, NULL)
521 
522 struct synth_field_desc {
523 	const char *type;
524 	const char *name;
525 };
526 
527 extern int synth_event_gen_cmd_array_start(struct dynevent_cmd *cmd,
528 					   const char *name,
529 					   struct module *mod,
530 					   struct synth_field_desc *fields,
531 					   unsigned int n_fields);
532 extern int synth_event_create(const char *name,
533 			      struct synth_field_desc *fields,
534 			      unsigned int n_fields, struct module *mod);
535 
536 extern int synth_event_add_field(struct dynevent_cmd *cmd,
537 				 const char *type,
538 				 const char *name);
539 extern int synth_event_add_field_str(struct dynevent_cmd *cmd,
540 				     const char *type_name);
541 extern int synth_event_add_fields(struct dynevent_cmd *cmd,
542 				  struct synth_field_desc *fields,
543 				  unsigned int n_fields);
544 
545 #define synth_event_gen_cmd_end(cmd)	\
546 	dynevent_create(cmd)
547 
548 struct synth_event;
549 
550 struct synth_event_trace_state {
551 	struct trace_event_buffer fbuffer;
552 	struct synth_trace_event *entry;
553 	struct trace_buffer *buffer;
554 	struct synth_event *event;
555 	unsigned int cur_field;
556 	unsigned int n_u64;
557 	bool disabled;
558 	bool add_next;
559 	bool add_name;
560 };
561 
562 extern int synth_event_trace(struct trace_event_file *file,
563 			     unsigned int n_vals, ...);
564 extern int synth_event_trace_array(struct trace_event_file *file, u64 *vals,
565 				   unsigned int n_vals);
566 extern int synth_event_trace_start(struct trace_event_file *file,
567 				   struct synth_event_trace_state *trace_state);
568 extern int synth_event_add_next_val(u64 val,
569 				    struct synth_event_trace_state *trace_state);
570 extern int synth_event_add_val(const char *field_name, u64 val,
571 			       struct synth_event_trace_state *trace_state);
572 extern int synth_event_trace_end(struct synth_event_trace_state *trace_state);
573 
574 extern int kprobe_event_delete(const char *name);
575 
576 extern void kprobe_event_cmd_init(struct dynevent_cmd *cmd,
577 				  char *buf, int maxlen);
578 
579 #define kprobe_event_gen_cmd_start(cmd, name, loc, ...)			\
580 	__kprobe_event_gen_cmd_start(cmd, false, name, loc, ## __VA_ARGS__, NULL)
581 
582 #define kretprobe_event_gen_cmd_start(cmd, name, loc, ...)		\
583 	__kprobe_event_gen_cmd_start(cmd, true, name, loc, ## __VA_ARGS__, NULL)
584 
585 extern int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd,
586 					bool kretprobe,
587 					const char *name,
588 					const char *loc, ...);
589 
590 #define kprobe_event_add_fields(cmd, ...)	\
591 	__kprobe_event_add_fields(cmd, ## __VA_ARGS__, NULL)
592 
593 #define kprobe_event_add_field(cmd, field)	\
594 	__kprobe_event_add_fields(cmd, field, NULL)
595 
596 extern int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...);
597 
598 #define kprobe_event_gen_cmd_end(cmd)		\
599 	dynevent_create(cmd)
600 
601 #define kretprobe_event_gen_cmd_end(cmd)	\
602 	dynevent_create(cmd)
603 
604 /*
605  * Event file flags:
606  *  ENABLED	  - The event is enabled
607  *  RECORDED_CMD  - The comms should be recorded at sched_switch
608  *  RECORDED_TGID - The tgids should be recorded at sched_switch
609  *  FILTERED	  - The event has a filter attached
610  *  NO_SET_FILTER - Set when filter has error and is to be ignored
611  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
612  *  SOFT_DISABLED - When set, do not trace the event (even though its
613  *                   tracepoint may be enabled)
614  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
615  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
616  *  PID_FILTER    - When set, the event is filtered based on pid
617  *  WAS_ENABLED   - Set when enabled to know to clear trace on module removal
618  */
619 enum {
620 	EVENT_FILE_FL_ENABLED		= (1 << EVENT_FILE_FL_ENABLED_BIT),
621 	EVENT_FILE_FL_RECORDED_CMD	= (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
622 	EVENT_FILE_FL_RECORDED_TGID	= (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
623 	EVENT_FILE_FL_FILTERED		= (1 << EVENT_FILE_FL_FILTERED_BIT),
624 	EVENT_FILE_FL_NO_SET_FILTER	= (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
625 	EVENT_FILE_FL_SOFT_MODE		= (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
626 	EVENT_FILE_FL_SOFT_DISABLED	= (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
627 	EVENT_FILE_FL_TRIGGER_MODE	= (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
628 	EVENT_FILE_FL_TRIGGER_COND	= (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
629 	EVENT_FILE_FL_PID_FILTER	= (1 << EVENT_FILE_FL_PID_FILTER_BIT),
630 	EVENT_FILE_FL_WAS_ENABLED	= (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
631 };
632 
633 struct trace_event_file {
634 	struct list_head		list;
635 	struct trace_event_call		*event_call;
636 	struct event_filter __rcu	*filter;
637 	struct dentry			*dir;
638 	struct trace_array		*tr;
639 	struct trace_subsystem_dir	*system;
640 	struct list_head		triggers;
641 
642 	/*
643 	 * 32 bit flags:
644 	 *   bit 0:		enabled
645 	 *   bit 1:		enabled cmd record
646 	 *   bit 2:		enable/disable with the soft disable bit
647 	 *   bit 3:		soft disabled
648 	 *   bit 4:		trigger enabled
649 	 *
650 	 * Note: The bits must be set atomically to prevent races
651 	 * from other writers. Reads of flags do not need to be in
652 	 * sync as they occur in critical sections. But the way flags
653 	 * is currently used, these changes do not affect the code
654 	 * except that when a change is made, it may have a slight
655 	 * delay in propagating the changes to other CPUs due to
656 	 * caching and such. Which is mostly OK ;-)
657 	 */
658 	unsigned long		flags;
659 	atomic_t		sm_ref;	/* soft-mode reference counter */
660 	atomic_t		tm_ref;	/* trigger-mode reference counter */
661 };
662 
663 #define __TRACE_EVENT_FLAGS(name, value)				\
664 	static int __init trace_init_flags_##name(void)			\
665 	{								\
666 		event_##name.flags |= value;				\
667 		return 0;						\
668 	}								\
669 	early_initcall(trace_init_flags_##name);
670 
671 #define __TRACE_EVENT_PERF_PERM(name, expr...)				\
672 	static int perf_perm_##name(struct trace_event_call *tp_event, \
673 				    struct perf_event *p_event)		\
674 	{								\
675 		return ({ expr; });					\
676 	}								\
677 	static int __init trace_init_perf_perm_##name(void)		\
678 	{								\
679 		event_##name.perf_perm = &perf_perm_##name;		\
680 		return 0;						\
681 	}								\
682 	early_initcall(trace_init_perf_perm_##name);
683 
684 #define PERF_MAX_TRACE_SIZE	8192
685 
686 #define MAX_FILTER_STR_VAL	256U	/* Should handle KSYM_SYMBOL_LEN */
687 
688 enum event_trigger_type {
689 	ETT_NONE		= (0),
690 	ETT_TRACE_ONOFF		= (1 << 0),
691 	ETT_SNAPSHOT		= (1 << 1),
692 	ETT_STACKTRACE		= (1 << 2),
693 	ETT_EVENT_ENABLE	= (1 << 3),
694 	ETT_EVENT_HIST		= (1 << 4),
695 	ETT_HIST_ENABLE		= (1 << 5),
696 	ETT_EVENT_EPROBE	= (1 << 6),
697 };
698 
699 extern int filter_match_preds(struct event_filter *filter, void *rec);
700 
701 extern enum event_trigger_type
702 event_triggers_call(struct trace_event_file *file,
703 		    struct trace_buffer *buffer, void *rec,
704 		    struct ring_buffer_event *event);
705 extern void
706 event_triggers_post_call(struct trace_event_file *file,
707 			 enum event_trigger_type tt);
708 
709 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
710 
711 bool __trace_trigger_soft_disabled(struct trace_event_file *file);
712 
713 /**
714  * trace_trigger_soft_disabled - do triggers and test if soft disabled
715  * @file: The file pointer of the event to test
716  *
717  * If any triggers without filters are attached to this event, they
718  * will be called here. If the event is soft disabled and has no
719  * triggers that require testing the fields, it will return true,
720  * otherwise false.
721  */
722 static __always_inline bool
723 trace_trigger_soft_disabled(struct trace_event_file *file)
724 {
725 	unsigned long eflags = file->flags;
726 
727 	if (likely(!(eflags & (EVENT_FILE_FL_TRIGGER_MODE |
728 			       EVENT_FILE_FL_SOFT_DISABLED |
729 			       EVENT_FILE_FL_PID_FILTER))))
730 		return false;
731 
732 	if (likely(eflags & EVENT_FILE_FL_TRIGGER_COND))
733 		return false;
734 
735 	return __trace_trigger_soft_disabled(file);
736 }
737 
738 #ifdef CONFIG_BPF_EVENTS
739 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
740 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
741 void perf_event_detach_bpf_prog(struct perf_event *event);
742 int perf_event_query_prog_array(struct perf_event *event, void __user *info);
743 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
744 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
745 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
746 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
747 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
748 			    u32 *fd_type, const char **buf,
749 			    u64 *probe_offset, u64 *probe_addr);
750 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
751 #else
752 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
753 {
754 	return 1;
755 }
756 
757 static inline int
758 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie)
759 {
760 	return -EOPNOTSUPP;
761 }
762 
763 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
764 
765 static inline int
766 perf_event_query_prog_array(struct perf_event *event, void __user *info)
767 {
768 	return -EOPNOTSUPP;
769 }
770 static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
771 {
772 	return -EOPNOTSUPP;
773 }
774 static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
775 {
776 	return -EOPNOTSUPP;
777 }
778 static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
779 {
780 	return NULL;
781 }
782 static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
783 {
784 }
785 static inline int bpf_get_perf_event_info(const struct perf_event *event,
786 					  u32 *prog_id, u32 *fd_type,
787 					  const char **buf, u64 *probe_offset,
788 					  u64 *probe_addr)
789 {
790 	return -EOPNOTSUPP;
791 }
792 static inline int
793 bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
794 {
795 	return -EOPNOTSUPP;
796 }
797 #endif
798 
799 enum {
800 	FILTER_OTHER = 0,
801 	FILTER_STATIC_STRING,
802 	FILTER_DYN_STRING,
803 	FILTER_RDYN_STRING,
804 	FILTER_PTR_STRING,
805 	FILTER_TRACE_FN,
806 	FILTER_COMM,
807 	FILTER_CPU,
808 };
809 
810 extern int trace_event_raw_init(struct trace_event_call *call);
811 extern int trace_define_field(struct trace_event_call *call, const char *type,
812 			      const char *name, int offset, int size,
813 			      int is_signed, int filter_type);
814 extern int trace_add_event_call(struct trace_event_call *call);
815 extern int trace_remove_event_call(struct trace_event_call *call);
816 extern int trace_event_get_offsets(struct trace_event_call *call);
817 
818 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
819 int trace_set_clr_event(const char *system, const char *event, int set);
820 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
821 		const char *event, bool enable);
822 /*
823  * The double __builtin_constant_p is because gcc will give us an error
824  * if we try to allocate the static variable to fmt if it is not a
825  * constant. Even with the outer if statement optimizing out.
826  */
827 #define event_trace_printk(ip, fmt, args...)				\
828 do {									\
829 	__trace_printk_check_format(fmt, ##args);			\
830 	tracing_record_cmdline(current);				\
831 	if (__builtin_constant_p(fmt)) {				\
832 		static const char *trace_printk_fmt			\
833 		  __section("__trace_printk_fmt") =			\
834 			__builtin_constant_p(fmt) ? fmt : NULL;		\
835 									\
836 		__trace_bprintk(ip, trace_printk_fmt, ##args);		\
837 	} else								\
838 		__trace_printk(ip, fmt, ##args);			\
839 } while (0)
840 
841 #ifdef CONFIG_PERF_EVENTS
842 struct perf_event;
843 
844 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
845 DECLARE_PER_CPU(int, bpf_kprobe_override);
846 
847 extern int  perf_trace_init(struct perf_event *event);
848 extern void perf_trace_destroy(struct perf_event *event);
849 extern int  perf_trace_add(struct perf_event *event, int flags);
850 extern void perf_trace_del(struct perf_event *event, int flags);
851 #ifdef CONFIG_KPROBE_EVENTS
852 extern int  perf_kprobe_init(struct perf_event *event, bool is_retprobe);
853 extern void perf_kprobe_destroy(struct perf_event *event);
854 extern int bpf_get_kprobe_info(const struct perf_event *event,
855 			       u32 *fd_type, const char **symbol,
856 			       u64 *probe_offset, u64 *probe_addr,
857 			       bool perf_type_tracepoint);
858 #endif
859 #ifdef CONFIG_UPROBE_EVENTS
860 extern int  perf_uprobe_init(struct perf_event *event,
861 			     unsigned long ref_ctr_offset, bool is_retprobe);
862 extern void perf_uprobe_destroy(struct perf_event *event);
863 extern int bpf_get_uprobe_info(const struct perf_event *event,
864 			       u32 *fd_type, const char **filename,
865 			       u64 *probe_offset, bool perf_type_tracepoint);
866 #endif
867 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
868 				     char *filter_str);
869 extern void ftrace_profile_free_filter(struct perf_event *event);
870 void perf_trace_buf_update(void *record, u16 type);
871 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
872 
873 int perf_event_set_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
874 void perf_event_free_bpf_prog(struct perf_event *event);
875 
876 void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
877 void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
878 void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
879 		    u64 arg3);
880 void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
881 		    u64 arg3, u64 arg4);
882 void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
883 		    u64 arg3, u64 arg4, u64 arg5);
884 void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
885 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6);
886 void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
887 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
888 void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
889 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
890 		    u64 arg8);
891 void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
892 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
893 		    u64 arg8, u64 arg9);
894 void bpf_trace_run10(struct bpf_prog *prog, u64 arg1, u64 arg2,
895 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
896 		     u64 arg8, u64 arg9, u64 arg10);
897 void bpf_trace_run11(struct bpf_prog *prog, u64 arg1, u64 arg2,
898 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
899 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11);
900 void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
901 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
902 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
903 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
904 			       struct trace_event_call *call, u64 count,
905 			       struct pt_regs *regs, struct hlist_head *head,
906 			       struct task_struct *task);
907 
908 static inline void
909 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
910 		       u64 count, struct pt_regs *regs, void *head,
911 		       struct task_struct *task)
912 {
913 	perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
914 }
915 
916 #endif
917 
918 #define TRACE_EVENT_STR_MAX	512
919 
920 /*
921  * gcc warns that you can not use a va_list in an inlined
922  * function. But lets me make it into a macro :-/
923  */
924 #define __trace_event_vstr_len(fmt, va)			\
925 ({							\
926 	va_list __ap;					\
927 	int __ret;					\
928 							\
929 	va_copy(__ap, *(va));				\
930 	__ret = vsnprintf(NULL, 0, fmt, __ap) + 1;	\
931 	va_end(__ap);					\
932 							\
933 	min(__ret, TRACE_EVENT_STR_MAX);		\
934 })
935 
936 #endif /* _LINUX_TRACE_EVENT_H */
937 
938 /*
939  * Note: we keep the TRACE_CUSTOM_EVENT outside the include file ifdef protection.
940  *  This is due to the way trace custom events work. If a file includes two
941  *  trace event headers under one "CREATE_CUSTOM_TRACE_EVENTS" the first include
942  *  will override the TRACE_CUSTOM_EVENT and break the second include.
943  */
944 
945 #ifndef TRACE_CUSTOM_EVENT
946 
947 #define DECLARE_CUSTOM_EVENT_CLASS(name, proto, args, tstruct, assign, print)
948 #define DEFINE_CUSTOM_EVENT(template, name, proto, args)
949 #define TRACE_CUSTOM_EVENT(name, proto, args, struct, assign, print)
950 
951 #endif /* ifdef TRACE_CUSTOM_EVENT (see note above) */
952