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