xref: /linux-6.15/include/linux/trace_events.h (revision 8564c315)
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 	int				type;
140 	struct trace_event_functions	*funcs;
141 };
142 
143 extern int register_trace_event(struct trace_event *event);
144 extern int unregister_trace_event(struct trace_event *event);
145 
146 /* Return values for print_line callback */
147 enum print_line_t {
148 	TRACE_TYPE_PARTIAL_LINE	= 0,	/* Retry after flushing the seq */
149 	TRACE_TYPE_HANDLED	= 1,
150 	TRACE_TYPE_UNHANDLED	= 2,	/* Relay to other output functions */
151 	TRACE_TYPE_NO_CONSUME	= 3	/* Handled but ask to not consume */
152 };
153 
154 enum print_line_t trace_handle_return(struct trace_seq *s);
155 
156 static inline void tracing_generic_entry_update(struct trace_entry *entry,
157 						unsigned short type,
158 						unsigned int trace_ctx)
159 {
160 	entry->preempt_count		= trace_ctx & 0xff;
161 	entry->pid			= current->pid;
162 	entry->type			= type;
163 	entry->flags =			trace_ctx >> 16;
164 }
165 
166 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status);
167 
168 enum trace_flag_type {
169 	TRACE_FLAG_IRQS_OFF		= 0x01,
170 	TRACE_FLAG_IRQS_NOSUPPORT	= 0x02,
171 	TRACE_FLAG_NEED_RESCHED		= 0x04,
172 	TRACE_FLAG_HARDIRQ		= 0x08,
173 	TRACE_FLAG_SOFTIRQ		= 0x10,
174 	TRACE_FLAG_PREEMPT_RESCHED	= 0x20,
175 	TRACE_FLAG_NMI			= 0x40,
176 	TRACE_FLAG_BH_OFF		= 0x80,
177 };
178 
179 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
180 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
181 {
182 	unsigned int irq_status = irqs_disabled_flags(irqflags) ?
183 		TRACE_FLAG_IRQS_OFF : 0;
184 	return tracing_gen_ctx_irq_test(irq_status);
185 }
186 static inline unsigned int tracing_gen_ctx(void)
187 {
188 	unsigned long irqflags;
189 
190 	local_save_flags(irqflags);
191 	return tracing_gen_ctx_flags(irqflags);
192 }
193 #else
194 
195 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
196 {
197 	return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
198 }
199 static inline unsigned int tracing_gen_ctx(void)
200 {
201 	return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
202 }
203 #endif
204 
205 static inline unsigned int tracing_gen_ctx_dec(void)
206 {
207 	unsigned int trace_ctx;
208 
209 	trace_ctx = tracing_gen_ctx();
210 	/*
211 	 * Subtract one from the preemption counter if preemption is enabled,
212 	 * see trace_event_buffer_reserve()for details.
213 	 */
214 	if (IS_ENABLED(CONFIG_PREEMPTION))
215 		trace_ctx--;
216 	return trace_ctx;
217 }
218 
219 struct trace_event_file;
220 
221 struct ring_buffer_event *
222 trace_event_buffer_lock_reserve(struct trace_buffer **current_buffer,
223 				struct trace_event_file *trace_file,
224 				int type, unsigned long len,
225 				unsigned int trace_ctx);
226 
227 #define TRACE_RECORD_CMDLINE	BIT(0)
228 #define TRACE_RECORD_TGID	BIT(1)
229 
230 void tracing_record_taskinfo(struct task_struct *task, int flags);
231 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
232 					  struct task_struct *next, int flags);
233 
234 void tracing_record_cmdline(struct task_struct *task);
235 void tracing_record_tgid(struct task_struct *task);
236 
237 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...)
238 	 __printf(3, 4);
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 			const int  len;
274 		};
275 		int (*define_fields)(struct trace_event_call *);
276 	};
277 };
278 
279 struct trace_event_class {
280 	const char		*system;
281 	void			*probe;
282 #ifdef CONFIG_PERF_EVENTS
283 	void			*perf_probe;
284 #endif
285 	int			(*reg)(struct trace_event_call *event,
286 				       enum trace_reg type, void *data);
287 	struct trace_event_fields *fields_array;
288 	struct list_head	*(*get_fields)(struct trace_event_call *);
289 	struct list_head	fields;
290 	int			(*raw_init)(struct trace_event_call *);
291 };
292 
293 extern int trace_event_reg(struct trace_event_call *event,
294 			    enum trace_reg type, void *data);
295 
296 struct trace_event_buffer {
297 	struct trace_buffer		*buffer;
298 	struct ring_buffer_event	*event;
299 	struct trace_event_file		*trace_file;
300 	void				*entry;
301 	unsigned int			trace_ctx;
302 	struct pt_regs			*regs;
303 };
304 
305 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
306 				  struct trace_event_file *trace_file,
307 				  unsigned long len);
308 
309 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
310 
311 enum {
312 	TRACE_EVENT_FL_FILTERED_BIT,
313 	TRACE_EVENT_FL_CAP_ANY_BIT,
314 	TRACE_EVENT_FL_NO_SET_FILTER_BIT,
315 	TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
316 	TRACE_EVENT_FL_TRACEPOINT_BIT,
317 	TRACE_EVENT_FL_DYNAMIC_BIT,
318 	TRACE_EVENT_FL_KPROBE_BIT,
319 	TRACE_EVENT_FL_UPROBE_BIT,
320 	TRACE_EVENT_FL_EPROBE_BIT,
321 	TRACE_EVENT_FL_FPROBE_BIT,
322 	TRACE_EVENT_FL_CUSTOM_BIT,
323 };
324 
325 /*
326  * Event flags:
327  *  FILTERED	  - The event has a filter attached
328  *  CAP_ANY	  - Any user can enable for perf
329  *  NO_SET_FILTER - Set when filter has error and is to be ignored
330  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
331  *  TRACEPOINT    - Event is a tracepoint
332  *  DYNAMIC       - Event is a dynamic event (created at run time)
333  *  KPROBE        - Event is a kprobe
334  *  UPROBE        - Event is a uprobe
335  *  EPROBE        - Event is an event probe
336  *  FPROBE        - Event is an function probe
337  *  CUSTOM        - Event is a custom event (to be attached to an exsiting tracepoint)
338  *                   This is set when the custom event has not been attached
339  *                   to a tracepoint yet, then it is cleared when it is.
340  */
341 enum {
342 	TRACE_EVENT_FL_FILTERED		= (1 << TRACE_EVENT_FL_FILTERED_BIT),
343 	TRACE_EVENT_FL_CAP_ANY		= (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
344 	TRACE_EVENT_FL_NO_SET_FILTER	= (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
345 	TRACE_EVENT_FL_IGNORE_ENABLE	= (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
346 	TRACE_EVENT_FL_TRACEPOINT	= (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
347 	TRACE_EVENT_FL_DYNAMIC		= (1 << TRACE_EVENT_FL_DYNAMIC_BIT),
348 	TRACE_EVENT_FL_KPROBE		= (1 << TRACE_EVENT_FL_KPROBE_BIT),
349 	TRACE_EVENT_FL_UPROBE		= (1 << TRACE_EVENT_FL_UPROBE_BIT),
350 	TRACE_EVENT_FL_EPROBE		= (1 << TRACE_EVENT_FL_EPROBE_BIT),
351 	TRACE_EVENT_FL_FPROBE		= (1 << TRACE_EVENT_FL_FPROBE_BIT),
352 	TRACE_EVENT_FL_CUSTOM		= (1 << TRACE_EVENT_FL_CUSTOM_BIT),
353 };
354 
355 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
356 
357 struct trace_event_call {
358 	struct list_head	list;
359 	struct trace_event_class *class;
360 	union {
361 		char			*name;
362 		/* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
363 		struct tracepoint	*tp;
364 	};
365 	struct trace_event	event;
366 	char			*print_fmt;
367 	struct event_filter	*filter;
368 	/*
369 	 * Static events can disappear with modules,
370 	 * where as dynamic ones need their own ref count.
371 	 */
372 	union {
373 		void				*module;
374 		atomic_t			refcnt;
375 	};
376 	void			*data;
377 
378 	/* See the TRACE_EVENT_FL_* flags above */
379 	int			flags; /* static flags of different events */
380 
381 #ifdef CONFIG_PERF_EVENTS
382 	int				perf_refcount;
383 	struct hlist_head __percpu	*perf_events;
384 	struct bpf_prog_array __rcu	*prog_array;
385 
386 	int	(*perf_perm)(struct trace_event_call *,
387 			     struct perf_event *);
388 #endif
389 };
390 
391 #ifdef CONFIG_DYNAMIC_EVENTS
392 bool trace_event_dyn_try_get_ref(struct trace_event_call *call);
393 void trace_event_dyn_put_ref(struct trace_event_call *call);
394 bool trace_event_dyn_busy(struct trace_event_call *call);
395 #else
396 static inline bool trace_event_dyn_try_get_ref(struct trace_event_call *call)
397 {
398 	/* Without DYNAMIC_EVENTS configured, nothing should be calling this */
399 	return false;
400 }
401 static inline void trace_event_dyn_put_ref(struct trace_event_call *call)
402 {
403 }
404 static inline bool trace_event_dyn_busy(struct trace_event_call *call)
405 {
406 	/* Nothing should call this without DYNAIMIC_EVENTS configured. */
407 	return true;
408 }
409 #endif
410 
411 static inline bool trace_event_try_get_ref(struct trace_event_call *call)
412 {
413 	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
414 		return trace_event_dyn_try_get_ref(call);
415 	else
416 		return try_module_get(call->module);
417 }
418 
419 static inline void trace_event_put_ref(struct trace_event_call *call)
420 {
421 	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
422 		trace_event_dyn_put_ref(call);
423 	else
424 		module_put(call->module);
425 }
426 
427 #ifdef CONFIG_PERF_EVENTS
428 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
429 {
430 	/*
431 	 * This inline function checks whether call->prog_array
432 	 * is valid or not. The function is called in various places,
433 	 * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
434 	 *
435 	 * If this function returns true, and later call->prog_array
436 	 * becomes false inside rcu_read_lock/unlock region,
437 	 * we bail out then. If this function return false,
438 	 * there is a risk that we might miss a few events if the checking
439 	 * were delayed until inside rcu_read_lock/unlock region and
440 	 * call->prog_array happened to become non-NULL then.
441 	 *
442 	 * Here, READ_ONCE() is used instead of rcu_access_pointer().
443 	 * rcu_access_pointer() requires the actual definition of
444 	 * "struct bpf_prog_array" while READ_ONCE() only needs
445 	 * a declaration of the same type.
446 	 */
447 	return !!READ_ONCE(call->prog_array);
448 }
449 #endif
450 
451 static inline const char *
452 trace_event_name(struct trace_event_call *call)
453 {
454 	if (call->flags & TRACE_EVENT_FL_CUSTOM)
455 		return call->name;
456 	else if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
457 		return call->tp ? call->tp->name : NULL;
458 	else
459 		return call->name;
460 }
461 
462 static inline struct list_head *
463 trace_get_fields(struct trace_event_call *event_call)
464 {
465 	if (!event_call->class->get_fields)
466 		return &event_call->class->fields;
467 	return event_call->class->get_fields(event_call);
468 }
469 
470 struct trace_subsystem_dir;
471 
472 enum {
473 	EVENT_FILE_FL_ENABLED_BIT,
474 	EVENT_FILE_FL_RECORDED_CMD_BIT,
475 	EVENT_FILE_FL_RECORDED_TGID_BIT,
476 	EVENT_FILE_FL_FILTERED_BIT,
477 	EVENT_FILE_FL_NO_SET_FILTER_BIT,
478 	EVENT_FILE_FL_SOFT_MODE_BIT,
479 	EVENT_FILE_FL_SOFT_DISABLED_BIT,
480 	EVENT_FILE_FL_TRIGGER_MODE_BIT,
481 	EVENT_FILE_FL_TRIGGER_COND_BIT,
482 	EVENT_FILE_FL_PID_FILTER_BIT,
483 	EVENT_FILE_FL_WAS_ENABLED_BIT,
484 };
485 
486 extern struct trace_event_file *trace_get_event_file(const char *instance,
487 						     const char *system,
488 						     const char *event);
489 extern void trace_put_event_file(struct trace_event_file *file);
490 
491 #define MAX_DYNEVENT_CMD_LEN	(2048)
492 
493 enum dynevent_type {
494 	DYNEVENT_TYPE_SYNTH = 1,
495 	DYNEVENT_TYPE_KPROBE,
496 	DYNEVENT_TYPE_NONE,
497 };
498 
499 struct dynevent_cmd;
500 
501 typedef int (*dynevent_create_fn_t)(struct dynevent_cmd *cmd);
502 
503 struct dynevent_cmd {
504 	struct seq_buf		seq;
505 	const char		*event_name;
506 	unsigned int		n_fields;
507 	enum dynevent_type	type;
508 	dynevent_create_fn_t	run_command;
509 	void			*private_data;
510 };
511 
512 extern int dynevent_create(struct dynevent_cmd *cmd);
513 
514 extern int synth_event_delete(const char *name);
515 
516 extern void synth_event_cmd_init(struct dynevent_cmd *cmd,
517 				 char *buf, int maxlen);
518 
519 extern int __synth_event_gen_cmd_start(struct dynevent_cmd *cmd,
520 				       const char *name,
521 				       struct module *mod, ...);
522 
523 #define synth_event_gen_cmd_start(cmd, name, mod, ...)	\
524 	__synth_event_gen_cmd_start(cmd, name, mod, ## __VA_ARGS__, NULL)
525 
526 struct synth_field_desc {
527 	const char *type;
528 	const char *name;
529 };
530 
531 extern int synth_event_gen_cmd_array_start(struct dynevent_cmd *cmd,
532 					   const char *name,
533 					   struct module *mod,
534 					   struct synth_field_desc *fields,
535 					   unsigned int n_fields);
536 extern int synth_event_create(const char *name,
537 			      struct synth_field_desc *fields,
538 			      unsigned int n_fields, struct module *mod);
539 
540 extern int synth_event_add_field(struct dynevent_cmd *cmd,
541 				 const char *type,
542 				 const char *name);
543 extern int synth_event_add_field_str(struct dynevent_cmd *cmd,
544 				     const char *type_name);
545 extern int synth_event_add_fields(struct dynevent_cmd *cmd,
546 				  struct synth_field_desc *fields,
547 				  unsigned int n_fields);
548 
549 #define synth_event_gen_cmd_end(cmd)	\
550 	dynevent_create(cmd)
551 
552 struct synth_event;
553 
554 struct synth_event_trace_state {
555 	struct trace_event_buffer fbuffer;
556 	struct synth_trace_event *entry;
557 	struct trace_buffer *buffer;
558 	struct synth_event *event;
559 	unsigned int cur_field;
560 	unsigned int n_u64;
561 	bool disabled;
562 	bool add_next;
563 	bool add_name;
564 };
565 
566 extern int synth_event_trace(struct trace_event_file *file,
567 			     unsigned int n_vals, ...);
568 extern int synth_event_trace_array(struct trace_event_file *file, u64 *vals,
569 				   unsigned int n_vals);
570 extern int synth_event_trace_start(struct trace_event_file *file,
571 				   struct synth_event_trace_state *trace_state);
572 extern int synth_event_add_next_val(u64 val,
573 				    struct synth_event_trace_state *trace_state);
574 extern int synth_event_add_val(const char *field_name, u64 val,
575 			       struct synth_event_trace_state *trace_state);
576 extern int synth_event_trace_end(struct synth_event_trace_state *trace_state);
577 
578 extern int kprobe_event_delete(const char *name);
579 
580 extern void kprobe_event_cmd_init(struct dynevent_cmd *cmd,
581 				  char *buf, int maxlen);
582 
583 #define kprobe_event_gen_cmd_start(cmd, name, loc, ...)			\
584 	__kprobe_event_gen_cmd_start(cmd, false, name, loc, ## __VA_ARGS__, NULL)
585 
586 #define kretprobe_event_gen_cmd_start(cmd, name, loc, ...)		\
587 	__kprobe_event_gen_cmd_start(cmd, true, name, loc, ## __VA_ARGS__, NULL)
588 
589 extern int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd,
590 					bool kretprobe,
591 					const char *name,
592 					const char *loc, ...);
593 
594 #define kprobe_event_add_fields(cmd, ...)	\
595 	__kprobe_event_add_fields(cmd, ## __VA_ARGS__, NULL)
596 
597 #define kprobe_event_add_field(cmd, field)	\
598 	__kprobe_event_add_fields(cmd, field, NULL)
599 
600 extern int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...);
601 
602 #define kprobe_event_gen_cmd_end(cmd)		\
603 	dynevent_create(cmd)
604 
605 #define kretprobe_event_gen_cmd_end(cmd)	\
606 	dynevent_create(cmd)
607 
608 /*
609  * Event file flags:
610  *  ENABLED	  - The event is enabled
611  *  RECORDED_CMD  - The comms should be recorded at sched_switch
612  *  RECORDED_TGID - The tgids should be recorded at sched_switch
613  *  FILTERED	  - The event has a filter attached
614  *  NO_SET_FILTER - Set when filter has error and is to be ignored
615  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
616  *  SOFT_DISABLED - When set, do not trace the event (even though its
617  *                   tracepoint may be enabled)
618  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
619  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
620  *  PID_FILTER    - When set, the event is filtered based on pid
621  *  WAS_ENABLED   - Set when enabled to know to clear trace on module removal
622  */
623 enum {
624 	EVENT_FILE_FL_ENABLED		= (1 << EVENT_FILE_FL_ENABLED_BIT),
625 	EVENT_FILE_FL_RECORDED_CMD	= (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
626 	EVENT_FILE_FL_RECORDED_TGID	= (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
627 	EVENT_FILE_FL_FILTERED		= (1 << EVENT_FILE_FL_FILTERED_BIT),
628 	EVENT_FILE_FL_NO_SET_FILTER	= (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
629 	EVENT_FILE_FL_SOFT_MODE		= (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
630 	EVENT_FILE_FL_SOFT_DISABLED	= (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
631 	EVENT_FILE_FL_TRIGGER_MODE	= (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
632 	EVENT_FILE_FL_TRIGGER_COND	= (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
633 	EVENT_FILE_FL_PID_FILTER	= (1 << EVENT_FILE_FL_PID_FILTER_BIT),
634 	EVENT_FILE_FL_WAS_ENABLED	= (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
635 };
636 
637 struct trace_event_file {
638 	struct list_head		list;
639 	struct trace_event_call		*event_call;
640 	struct event_filter __rcu	*filter;
641 	struct dentry			*dir;
642 	struct trace_array		*tr;
643 	struct trace_subsystem_dir	*system;
644 	struct list_head		triggers;
645 
646 	/*
647 	 * 32 bit flags:
648 	 *   bit 0:		enabled
649 	 *   bit 1:		enabled cmd record
650 	 *   bit 2:		enable/disable with the soft disable bit
651 	 *   bit 3:		soft disabled
652 	 *   bit 4:		trigger enabled
653 	 *
654 	 * Note: The bits must be set atomically to prevent races
655 	 * from other writers. Reads of flags do not need to be in
656 	 * sync as they occur in critical sections. But the way flags
657 	 * is currently used, these changes do not affect the code
658 	 * except that when a change is made, it may have a slight
659 	 * delay in propagating the changes to other CPUs due to
660 	 * caching and such. Which is mostly OK ;-)
661 	 */
662 	unsigned long		flags;
663 	atomic_t		sm_ref;	/* soft-mode reference counter */
664 	atomic_t		tm_ref;	/* trigger-mode reference counter */
665 };
666 
667 #define __TRACE_EVENT_FLAGS(name, value)				\
668 	static int __init trace_init_flags_##name(void)			\
669 	{								\
670 		event_##name.flags |= value;				\
671 		return 0;						\
672 	}								\
673 	early_initcall(trace_init_flags_##name);
674 
675 #define __TRACE_EVENT_PERF_PERM(name, expr...)				\
676 	static int perf_perm_##name(struct trace_event_call *tp_event, \
677 				    struct perf_event *p_event)		\
678 	{								\
679 		return ({ expr; });					\
680 	}								\
681 	static int __init trace_init_perf_perm_##name(void)		\
682 	{								\
683 		event_##name.perf_perm = &perf_perm_##name;		\
684 		return 0;						\
685 	}								\
686 	early_initcall(trace_init_perf_perm_##name);
687 
688 #define PERF_MAX_TRACE_SIZE	8192
689 
690 #define MAX_FILTER_STR_VAL	256U	/* Should handle KSYM_SYMBOL_LEN */
691 
692 enum event_trigger_type {
693 	ETT_NONE		= (0),
694 	ETT_TRACE_ONOFF		= (1 << 0),
695 	ETT_SNAPSHOT		= (1 << 1),
696 	ETT_STACKTRACE		= (1 << 2),
697 	ETT_EVENT_ENABLE	= (1 << 3),
698 	ETT_EVENT_HIST		= (1 << 4),
699 	ETT_HIST_ENABLE		= (1 << 5),
700 	ETT_EVENT_EPROBE	= (1 << 6),
701 };
702 
703 extern int filter_match_preds(struct event_filter *filter, void *rec);
704 
705 extern enum event_trigger_type
706 event_triggers_call(struct trace_event_file *file,
707 		    struct trace_buffer *buffer, void *rec,
708 		    struct ring_buffer_event *event);
709 extern void
710 event_triggers_post_call(struct trace_event_file *file,
711 			 enum event_trigger_type tt);
712 
713 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
714 
715 bool __trace_trigger_soft_disabled(struct trace_event_file *file);
716 
717 /**
718  * trace_trigger_soft_disabled - do triggers and test if soft disabled
719  * @file: The file pointer of the event to test
720  *
721  * If any triggers without filters are attached to this event, they
722  * will be called here. If the event is soft disabled and has no
723  * triggers that require testing the fields, it will return true,
724  * otherwise false.
725  */
726 static __always_inline bool
727 trace_trigger_soft_disabled(struct trace_event_file *file)
728 {
729 	unsigned long eflags = file->flags;
730 
731 	if (likely(!(eflags & (EVENT_FILE_FL_TRIGGER_MODE |
732 			       EVENT_FILE_FL_SOFT_DISABLED |
733 			       EVENT_FILE_FL_PID_FILTER))))
734 		return false;
735 
736 	if (likely(eflags & EVENT_FILE_FL_TRIGGER_COND))
737 		return false;
738 
739 	return __trace_trigger_soft_disabled(file);
740 }
741 
742 #ifdef CONFIG_BPF_EVENTS
743 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
744 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
745 void perf_event_detach_bpf_prog(struct perf_event *event);
746 int perf_event_query_prog_array(struct perf_event *event, void __user *info);
747 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
748 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
749 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
750 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
751 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
752 			    u32 *fd_type, const char **buf,
753 			    u64 *probe_offset, u64 *probe_addr);
754 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
755 #else
756 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
757 {
758 	return 1;
759 }
760 
761 static inline int
762 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie)
763 {
764 	return -EOPNOTSUPP;
765 }
766 
767 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
768 
769 static inline int
770 perf_event_query_prog_array(struct perf_event *event, void __user *info)
771 {
772 	return -EOPNOTSUPP;
773 }
774 static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
775 {
776 	return -EOPNOTSUPP;
777 }
778 static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
779 {
780 	return -EOPNOTSUPP;
781 }
782 static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
783 {
784 	return NULL;
785 }
786 static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
787 {
788 }
789 static inline int bpf_get_perf_event_info(const struct perf_event *event,
790 					  u32 *prog_id, u32 *fd_type,
791 					  const char **buf, u64 *probe_offset,
792 					  u64 *probe_addr)
793 {
794 	return -EOPNOTSUPP;
795 }
796 static inline int
797 bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
798 {
799 	return -EOPNOTSUPP;
800 }
801 #endif
802 
803 enum {
804 	FILTER_OTHER = 0,
805 	FILTER_STATIC_STRING,
806 	FILTER_DYN_STRING,
807 	FILTER_RDYN_STRING,
808 	FILTER_PTR_STRING,
809 	FILTER_TRACE_FN,
810 	FILTER_COMM,
811 	FILTER_CPU,
812 	FILTER_STACKTRACE,
813 };
814 
815 extern int trace_event_raw_init(struct trace_event_call *call);
816 extern int trace_define_field(struct trace_event_call *call, const char *type,
817 			      const char *name, int offset, int size,
818 			      int is_signed, int filter_type);
819 extern int trace_add_event_call(struct trace_event_call *call);
820 extern int trace_remove_event_call(struct trace_event_call *call);
821 extern int trace_event_get_offsets(struct trace_event_call *call);
822 
823 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
824 int trace_set_clr_event(const char *system, const char *event, int set);
825 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
826 		const char *event, bool enable);
827 /*
828  * The double __builtin_constant_p is because gcc will give us an error
829  * if we try to allocate the static variable to fmt if it is not a
830  * constant. Even with the outer if statement optimizing out.
831  */
832 #define event_trace_printk(ip, fmt, args...)				\
833 do {									\
834 	__trace_printk_check_format(fmt, ##args);			\
835 	tracing_record_cmdline(current);				\
836 	if (__builtin_constant_p(fmt)) {				\
837 		static const char *trace_printk_fmt			\
838 		  __section("__trace_printk_fmt") =			\
839 			__builtin_constant_p(fmt) ? fmt : NULL;		\
840 									\
841 		__trace_bprintk(ip, trace_printk_fmt, ##args);		\
842 	} else								\
843 		__trace_printk(ip, fmt, ##args);			\
844 } while (0)
845 
846 #ifdef CONFIG_PERF_EVENTS
847 struct perf_event;
848 
849 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
850 DECLARE_PER_CPU(int, bpf_kprobe_override);
851 
852 extern int  perf_trace_init(struct perf_event *event);
853 extern void perf_trace_destroy(struct perf_event *event);
854 extern int  perf_trace_add(struct perf_event *event, int flags);
855 extern void perf_trace_del(struct perf_event *event, int flags);
856 #ifdef CONFIG_KPROBE_EVENTS
857 extern int  perf_kprobe_init(struct perf_event *event, bool is_retprobe);
858 extern void perf_kprobe_destroy(struct perf_event *event);
859 extern int bpf_get_kprobe_info(const struct perf_event *event,
860 			       u32 *fd_type, const char **symbol,
861 			       u64 *probe_offset, u64 *probe_addr,
862 			       bool perf_type_tracepoint);
863 #endif
864 #ifdef CONFIG_UPROBE_EVENTS
865 extern int  perf_uprobe_init(struct perf_event *event,
866 			     unsigned long ref_ctr_offset, bool is_retprobe);
867 extern void perf_uprobe_destroy(struct perf_event *event);
868 extern int bpf_get_uprobe_info(const struct perf_event *event,
869 			       u32 *fd_type, const char **filename,
870 			       u64 *probe_offset, bool perf_type_tracepoint);
871 #endif
872 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
873 				     char *filter_str);
874 extern void ftrace_profile_free_filter(struct perf_event *event);
875 void perf_trace_buf_update(void *record, u16 type);
876 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
877 
878 int perf_event_set_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
879 void perf_event_free_bpf_prog(struct perf_event *event);
880 
881 void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
882 void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
883 void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
884 		    u64 arg3);
885 void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
886 		    u64 arg3, u64 arg4);
887 void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
888 		    u64 arg3, u64 arg4, u64 arg5);
889 void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
890 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6);
891 void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
892 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
893 void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
894 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
895 		    u64 arg8);
896 void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
897 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
898 		    u64 arg8, u64 arg9);
899 void bpf_trace_run10(struct bpf_prog *prog, u64 arg1, u64 arg2,
900 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
901 		     u64 arg8, u64 arg9, u64 arg10);
902 void bpf_trace_run11(struct bpf_prog *prog, u64 arg1, u64 arg2,
903 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
904 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11);
905 void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
906 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
907 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
908 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
909 			       struct trace_event_call *call, u64 count,
910 			       struct pt_regs *regs, struct hlist_head *head,
911 			       struct task_struct *task);
912 
913 static inline void
914 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
915 		       u64 count, struct pt_regs *regs, void *head,
916 		       struct task_struct *task)
917 {
918 	perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
919 }
920 
921 #endif
922 
923 #define TRACE_EVENT_STR_MAX	512
924 
925 /*
926  * gcc warns that you can not use a va_list in an inlined
927  * function. But lets me make it into a macro :-/
928  */
929 #define __trace_event_vstr_len(fmt, va)			\
930 ({							\
931 	va_list __ap;					\
932 	int __ret;					\
933 							\
934 	va_copy(__ap, *(va));				\
935 	__ret = vsnprintf(NULL, 0, fmt, __ap) + 1;	\
936 	va_end(__ap);					\
937 							\
938 	min(__ret, TRACE_EVENT_STR_MAX);		\
939 })
940 
941 #endif /* _LINUX_TRACE_EVENT_H */
942 
943 /*
944  * Note: we keep the TRACE_CUSTOM_EVENT outside the include file ifdef protection.
945  *  This is due to the way trace custom events work. If a file includes two
946  *  trace event headers under one "CREATE_CUSTOM_TRACE_EVENTS" the first include
947  *  will override the TRACE_CUSTOM_EVENT and break the second include.
948  */
949 
950 #ifndef TRACE_CUSTOM_EVENT
951 
952 #define DECLARE_CUSTOM_EVENT_CLASS(name, proto, args, tstruct, assign, print)
953 #define DEFINE_CUSTOM_EVENT(template, name, proto, args)
954 #define TRACE_CUSTOM_EVENT(name, proto, args, struct, assign, print)
955 
956 #endif /* ifdef TRACE_CUSTOM_EVENT (see note above) */
957