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 
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 	TRACE_FLAG_BH_OFF		= 0x80,
177 };
178 
179 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
tracing_gen_ctx_flags(unsigned long irqflags)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 }
tracing_gen_ctx(void)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 
tracing_gen_ctx_flags(unsigned long irqflags)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 }
tracing_gen_ctx(void)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 
tracing_gen_ctx_dec(void)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 
239 struct event_filter;
240 
241 enum trace_reg {
242 	TRACE_REG_REGISTER,
243 	TRACE_REG_UNREGISTER,
244 #ifdef CONFIG_PERF_EVENTS
245 	TRACE_REG_PERF_REGISTER,
246 	TRACE_REG_PERF_UNREGISTER,
247 	TRACE_REG_PERF_OPEN,
248 	TRACE_REG_PERF_CLOSE,
249 	/*
250 	 * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
251 	 * custom action was taken and the default action is not to be
252 	 * performed.
253 	 */
254 	TRACE_REG_PERF_ADD,
255 	TRACE_REG_PERF_DEL,
256 #endif
257 };
258 
259 struct trace_event_call;
260 
261 #define TRACE_FUNCTION_TYPE ((const char *)~0UL)
262 
263 struct trace_event_fields {
264 	const char *type;
265 	union {
266 		struct {
267 			const char *name;
268 			const int  size;
269 			const int  align;
270 			const int  is_signed;
271 			const int  filter_type;
272 		};
273 		int (*define_fields)(struct trace_event_call *);
274 	};
275 };
276 
277 struct trace_event_class {
278 	const char		*system;
279 	void			*probe;
280 #ifdef CONFIG_PERF_EVENTS
281 	void			*perf_probe;
282 #endif
283 	int			(*reg)(struct trace_event_call *event,
284 				       enum trace_reg type, void *data);
285 	struct trace_event_fields *fields_array;
286 	struct list_head	*(*get_fields)(struct trace_event_call *);
287 	struct list_head	fields;
288 	int			(*raw_init)(struct trace_event_call *);
289 };
290 
291 extern int trace_event_reg(struct trace_event_call *event,
292 			    enum trace_reg type, void *data);
293 
294 struct trace_event_buffer {
295 	struct trace_buffer		*buffer;
296 	struct ring_buffer_event	*event;
297 	struct trace_event_file		*trace_file;
298 	void				*entry;
299 	unsigned int			trace_ctx;
300 	struct pt_regs			*regs;
301 };
302 
303 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
304 				  struct trace_event_file *trace_file,
305 				  unsigned long len);
306 
307 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
308 
309 enum {
310 	TRACE_EVENT_FL_FILTERED_BIT,
311 	TRACE_EVENT_FL_CAP_ANY_BIT,
312 	TRACE_EVENT_FL_NO_SET_FILTER_BIT,
313 	TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
314 	TRACE_EVENT_FL_TRACEPOINT_BIT,
315 	TRACE_EVENT_FL_DYNAMIC_BIT,
316 	TRACE_EVENT_FL_KPROBE_BIT,
317 	TRACE_EVENT_FL_UPROBE_BIT,
318 	TRACE_EVENT_FL_EPROBE_BIT,
319 	TRACE_EVENT_FL_CUSTOM_BIT,
320 };
321 
322 /*
323  * Event flags:
324  *  FILTERED	  - The event has a filter attached
325  *  CAP_ANY	  - Any user can enable for perf
326  *  NO_SET_FILTER - Set when filter has error and is to be ignored
327  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
328  *  TRACEPOINT    - Event is a tracepoint
329  *  DYNAMIC       - Event is a dynamic event (created at run time)
330  *  KPROBE        - Event is a kprobe
331  *  UPROBE        - Event is a uprobe
332  *  EPROBE        - Event is an event probe
333  *  CUSTOM        - Event is a custom event (to be attached to an exsiting tracepoint)
334  *                   This is set when the custom event has not been attached
335  *                   to a tracepoint yet, then it is cleared when it is.
336  */
337 enum {
338 	TRACE_EVENT_FL_FILTERED		= (1 << TRACE_EVENT_FL_FILTERED_BIT),
339 	TRACE_EVENT_FL_CAP_ANY		= (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
340 	TRACE_EVENT_FL_NO_SET_FILTER	= (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
341 	TRACE_EVENT_FL_IGNORE_ENABLE	= (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
342 	TRACE_EVENT_FL_TRACEPOINT	= (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
343 	TRACE_EVENT_FL_DYNAMIC		= (1 << TRACE_EVENT_FL_DYNAMIC_BIT),
344 	TRACE_EVENT_FL_KPROBE		= (1 << TRACE_EVENT_FL_KPROBE_BIT),
345 	TRACE_EVENT_FL_UPROBE		= (1 << TRACE_EVENT_FL_UPROBE_BIT),
346 	TRACE_EVENT_FL_EPROBE		= (1 << TRACE_EVENT_FL_EPROBE_BIT),
347 	TRACE_EVENT_FL_CUSTOM		= (1 << TRACE_EVENT_FL_CUSTOM_BIT),
348 };
349 
350 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
351 
352 struct trace_event_call {
353 	struct list_head	list;
354 	struct trace_event_class *class;
355 	union {
356 		char			*name;
357 		/* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
358 		struct tracepoint	*tp;
359 	};
360 	struct trace_event	event;
361 	char			*print_fmt;
362 	struct event_filter	*filter;
363 	/*
364 	 * Static events can disappear with modules,
365 	 * where as dynamic ones need their own ref count.
366 	 */
367 	union {
368 		void				*module;
369 		atomic_t			refcnt;
370 	};
371 	void			*data;
372 
373 	/* See the TRACE_EVENT_FL_* flags above */
374 	int			flags; /* static flags of different events */
375 
376 #ifdef CONFIG_PERF_EVENTS
377 	int				perf_refcount;
378 	struct hlist_head __percpu	*perf_events;
379 	struct bpf_prog_array __rcu	*prog_array;
380 
381 	int	(*perf_perm)(struct trace_event_call *,
382 			     struct perf_event *);
383 #endif
384 };
385 
386 #ifdef CONFIG_DYNAMIC_EVENTS
387 bool trace_event_dyn_try_get_ref(struct trace_event_call *call);
388 void trace_event_dyn_put_ref(struct trace_event_call *call);
389 bool trace_event_dyn_busy(struct trace_event_call *call);
390 #else
trace_event_dyn_try_get_ref(struct trace_event_call * call)391 static inline bool trace_event_dyn_try_get_ref(struct trace_event_call *call)
392 {
393 	/* Without DYNAMIC_EVENTS configured, nothing should be calling this */
394 	return false;
395 }
trace_event_dyn_put_ref(struct trace_event_call * call)396 static inline void trace_event_dyn_put_ref(struct trace_event_call *call)
397 {
398 }
trace_event_dyn_busy(struct trace_event_call * call)399 static inline bool trace_event_dyn_busy(struct trace_event_call *call)
400 {
401 	/* Nothing should call this without DYNAIMIC_EVENTS configured. */
402 	return true;
403 }
404 #endif
405 
trace_event_try_get_ref(struct trace_event_call * call)406 static inline bool trace_event_try_get_ref(struct trace_event_call *call)
407 {
408 	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
409 		return trace_event_dyn_try_get_ref(call);
410 	else
411 		return try_module_get(call->module);
412 }
413 
trace_event_put_ref(struct trace_event_call * call)414 static inline void trace_event_put_ref(struct trace_event_call *call)
415 {
416 	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
417 		trace_event_dyn_put_ref(call);
418 	else
419 		module_put(call->module);
420 }
421 
422 #ifdef CONFIG_PERF_EVENTS
bpf_prog_array_valid(struct trace_event_call * call)423 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
424 {
425 	/*
426 	 * This inline function checks whether call->prog_array
427 	 * is valid or not. The function is called in various places,
428 	 * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
429 	 *
430 	 * If this function returns true, and later call->prog_array
431 	 * becomes false inside rcu_read_lock/unlock region,
432 	 * we bail out then. If this function return false,
433 	 * there is a risk that we might miss a few events if the checking
434 	 * were delayed until inside rcu_read_lock/unlock region and
435 	 * call->prog_array happened to become non-NULL then.
436 	 *
437 	 * Here, READ_ONCE() is used instead of rcu_access_pointer().
438 	 * rcu_access_pointer() requires the actual definition of
439 	 * "struct bpf_prog_array" while READ_ONCE() only needs
440 	 * a declaration of the same type.
441 	 */
442 	return !!READ_ONCE(call->prog_array);
443 }
444 #endif
445 
446 static inline const char *
trace_event_name(struct trace_event_call * call)447 trace_event_name(struct trace_event_call *call)
448 {
449 	if (call->flags & TRACE_EVENT_FL_CUSTOM)
450 		return call->name;
451 	else if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
452 		return call->tp ? call->tp->name : NULL;
453 	else
454 		return call->name;
455 }
456 
457 static inline struct list_head *
trace_get_fields(struct trace_event_call * event_call)458 trace_get_fields(struct trace_event_call *event_call)
459 {
460 	if (!event_call->class->get_fields)
461 		return &event_call->class->fields;
462 	return event_call->class->get_fields(event_call);
463 }
464 
465 struct trace_subsystem_dir;
466 
467 enum {
468 	EVENT_FILE_FL_ENABLED_BIT,
469 	EVENT_FILE_FL_RECORDED_CMD_BIT,
470 	EVENT_FILE_FL_RECORDED_TGID_BIT,
471 	EVENT_FILE_FL_FILTERED_BIT,
472 	EVENT_FILE_FL_NO_SET_FILTER_BIT,
473 	EVENT_FILE_FL_SOFT_MODE_BIT,
474 	EVENT_FILE_FL_SOFT_DISABLED_BIT,
475 	EVENT_FILE_FL_TRIGGER_MODE_BIT,
476 	EVENT_FILE_FL_TRIGGER_COND_BIT,
477 	EVENT_FILE_FL_PID_FILTER_BIT,
478 	EVENT_FILE_FL_WAS_ENABLED_BIT,
479 };
480 
481 extern struct trace_event_file *trace_get_event_file(const char *instance,
482 						     const char *system,
483 						     const char *event);
484 extern void trace_put_event_file(struct trace_event_file *file);
485 
486 #define MAX_DYNEVENT_CMD_LEN	(2048)
487 
488 enum dynevent_type {
489 	DYNEVENT_TYPE_SYNTH = 1,
490 	DYNEVENT_TYPE_KPROBE,
491 	DYNEVENT_TYPE_NONE,
492 };
493 
494 struct dynevent_cmd;
495 
496 typedef int (*dynevent_create_fn_t)(struct dynevent_cmd *cmd);
497 
498 struct dynevent_cmd {
499 	struct seq_buf		seq;
500 	const char		*event_name;
501 	unsigned int		n_fields;
502 	enum dynevent_type	type;
503 	dynevent_create_fn_t	run_command;
504 	void			*private_data;
505 };
506 
507 extern int dynevent_create(struct dynevent_cmd *cmd);
508 
509 extern int synth_event_delete(const char *name);
510 
511 extern void synth_event_cmd_init(struct dynevent_cmd *cmd,
512 				 char *buf, int maxlen);
513 
514 extern int __synth_event_gen_cmd_start(struct dynevent_cmd *cmd,
515 				       const char *name,
516 				       struct module *mod, ...);
517 
518 #define synth_event_gen_cmd_start(cmd, name, mod, ...)	\
519 	__synth_event_gen_cmd_start(cmd, name, mod, ## __VA_ARGS__, NULL)
520 
521 struct synth_field_desc {
522 	const char *type;
523 	const char *name;
524 };
525 
526 extern int synth_event_gen_cmd_array_start(struct dynevent_cmd *cmd,
527 					   const char *name,
528 					   struct module *mod,
529 					   struct synth_field_desc *fields,
530 					   unsigned int n_fields);
531 extern int synth_event_create(const char *name,
532 			      struct synth_field_desc *fields,
533 			      unsigned int n_fields, struct module *mod);
534 
535 extern int synth_event_add_field(struct dynevent_cmd *cmd,
536 				 const char *type,
537 				 const char *name);
538 extern int synth_event_add_field_str(struct dynevent_cmd *cmd,
539 				     const char *type_name);
540 extern int synth_event_add_fields(struct dynevent_cmd *cmd,
541 				  struct synth_field_desc *fields,
542 				  unsigned int n_fields);
543 
544 #define synth_event_gen_cmd_end(cmd)	\
545 	dynevent_create(cmd)
546 
547 struct synth_event;
548 
549 struct synth_event_trace_state {
550 	struct trace_event_buffer fbuffer;
551 	struct synth_trace_event *entry;
552 	struct trace_buffer *buffer;
553 	struct synth_event *event;
554 	unsigned int cur_field;
555 	unsigned int n_u64;
556 	bool disabled;
557 	bool add_next;
558 	bool add_name;
559 };
560 
561 extern int synth_event_trace(struct trace_event_file *file,
562 			     unsigned int n_vals, ...);
563 extern int synth_event_trace_array(struct trace_event_file *file, u64 *vals,
564 				   unsigned int n_vals);
565 extern int synth_event_trace_start(struct trace_event_file *file,
566 				   struct synth_event_trace_state *trace_state);
567 extern int synth_event_add_next_val(u64 val,
568 				    struct synth_event_trace_state *trace_state);
569 extern int synth_event_add_val(const char *field_name, u64 val,
570 			       struct synth_event_trace_state *trace_state);
571 extern int synth_event_trace_end(struct synth_event_trace_state *trace_state);
572 
573 extern int kprobe_event_delete(const char *name);
574 
575 extern void kprobe_event_cmd_init(struct dynevent_cmd *cmd,
576 				  char *buf, int maxlen);
577 
578 #define kprobe_event_gen_cmd_start(cmd, name, loc, ...)			\
579 	__kprobe_event_gen_cmd_start(cmd, false, name, loc, ## __VA_ARGS__, NULL)
580 
581 #define kretprobe_event_gen_cmd_start(cmd, name, loc, ...)		\
582 	__kprobe_event_gen_cmd_start(cmd, true, name, loc, ## __VA_ARGS__, NULL)
583 
584 extern int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd,
585 					bool kretprobe,
586 					const char *name,
587 					const char *loc, ...);
588 
589 #define kprobe_event_add_fields(cmd, ...)	\
590 	__kprobe_event_add_fields(cmd, ## __VA_ARGS__, NULL)
591 
592 #define kprobe_event_add_field(cmd, field)	\
593 	__kprobe_event_add_fields(cmd, field, NULL)
594 
595 extern int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...);
596 
597 #define kprobe_event_gen_cmd_end(cmd)		\
598 	dynevent_create(cmd)
599 
600 #define kretprobe_event_gen_cmd_end(cmd)	\
601 	dynevent_create(cmd)
602 
603 /*
604  * Event file flags:
605  *  ENABLED	  - The event is enabled
606  *  RECORDED_CMD  - The comms should be recorded at sched_switch
607  *  RECORDED_TGID - The tgids should be recorded at sched_switch
608  *  FILTERED	  - The event has a filter attached
609  *  NO_SET_FILTER - Set when filter has error and is to be ignored
610  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
611  *  SOFT_DISABLED - When set, do not trace the event (even though its
612  *                   tracepoint may be enabled)
613  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
614  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
615  *  PID_FILTER    - When set, the event is filtered based on pid
616  *  WAS_ENABLED   - Set when enabled to know to clear trace on module removal
617  */
618 enum {
619 	EVENT_FILE_FL_ENABLED		= (1 << EVENT_FILE_FL_ENABLED_BIT),
620 	EVENT_FILE_FL_RECORDED_CMD	= (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
621 	EVENT_FILE_FL_RECORDED_TGID	= (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
622 	EVENT_FILE_FL_FILTERED		= (1 << EVENT_FILE_FL_FILTERED_BIT),
623 	EVENT_FILE_FL_NO_SET_FILTER	= (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
624 	EVENT_FILE_FL_SOFT_MODE		= (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
625 	EVENT_FILE_FL_SOFT_DISABLED	= (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
626 	EVENT_FILE_FL_TRIGGER_MODE	= (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
627 	EVENT_FILE_FL_TRIGGER_COND	= (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
628 	EVENT_FILE_FL_PID_FILTER	= (1 << EVENT_FILE_FL_PID_FILTER_BIT),
629 	EVENT_FILE_FL_WAS_ENABLED	= (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
630 };
631 
632 struct trace_event_file {
633 	struct list_head		list;
634 	struct trace_event_call		*event_call;
635 	struct event_filter __rcu	*filter;
636 	struct dentry			*dir;
637 	struct trace_array		*tr;
638 	struct trace_subsystem_dir	*system;
639 	struct list_head		triggers;
640 
641 	/*
642 	 * 32 bit flags:
643 	 *   bit 0:		enabled
644 	 *   bit 1:		enabled cmd record
645 	 *   bit 2:		enable/disable with the soft disable bit
646 	 *   bit 3:		soft disabled
647 	 *   bit 4:		trigger enabled
648 	 *
649 	 * Note: The bits must be set atomically to prevent races
650 	 * from other writers. Reads of flags do not need to be in
651 	 * sync as they occur in critical sections. But the way flags
652 	 * is currently used, these changes do not affect the code
653 	 * except that when a change is made, it may have a slight
654 	 * delay in propagating the changes to other CPUs due to
655 	 * caching and such. Which is mostly OK ;-)
656 	 */
657 	unsigned long		flags;
658 	atomic_t		sm_ref;	/* soft-mode reference counter */
659 	atomic_t		tm_ref;	/* trigger-mode reference counter */
660 };
661 
662 #define __TRACE_EVENT_FLAGS(name, value)				\
663 	static int __init trace_init_flags_##name(void)			\
664 	{								\
665 		event_##name.flags |= value;				\
666 		return 0;						\
667 	}								\
668 	early_initcall(trace_init_flags_##name);
669 
670 #define __TRACE_EVENT_PERF_PERM(name, expr...)				\
671 	static int perf_perm_##name(struct trace_event_call *tp_event, \
672 				    struct perf_event *p_event)		\
673 	{								\
674 		return ({ expr; });					\
675 	}								\
676 	static int __init trace_init_perf_perm_##name(void)		\
677 	{								\
678 		event_##name.perf_perm = &perf_perm_##name;		\
679 		return 0;						\
680 	}								\
681 	early_initcall(trace_init_perf_perm_##name);
682 
683 #define PERF_MAX_TRACE_SIZE	8192
684 
685 #define MAX_FILTER_STR_VAL	256U	/* Should handle KSYM_SYMBOL_LEN */
686 
687 enum event_trigger_type {
688 	ETT_NONE		= (0),
689 	ETT_TRACE_ONOFF		= (1 << 0),
690 	ETT_SNAPSHOT		= (1 << 1),
691 	ETT_STACKTRACE		= (1 << 2),
692 	ETT_EVENT_ENABLE	= (1 << 3),
693 	ETT_EVENT_HIST		= (1 << 4),
694 	ETT_HIST_ENABLE		= (1 << 5),
695 	ETT_EVENT_EPROBE	= (1 << 6),
696 };
697 
698 extern int filter_match_preds(struct event_filter *filter, void *rec);
699 
700 extern enum event_trigger_type
701 event_triggers_call(struct trace_event_file *file,
702 		    struct trace_buffer *buffer, void *rec,
703 		    struct ring_buffer_event *event);
704 extern void
705 event_triggers_post_call(struct trace_event_file *file,
706 			 enum event_trigger_type tt);
707 
708 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
709 
710 bool __trace_trigger_soft_disabled(struct trace_event_file *file);
711 
712 /**
713  * trace_trigger_soft_disabled - do triggers and test if soft disabled
714  * @file: The file pointer of the event to test
715  *
716  * If any triggers without filters are attached to this event, they
717  * will be called here. If the event is soft disabled and has no
718  * triggers that require testing the fields, it will return true,
719  * otherwise false.
720  */
721 static __always_inline bool
trace_trigger_soft_disabled(struct trace_event_file * file)722 trace_trigger_soft_disabled(struct trace_event_file *file)
723 {
724 	unsigned long eflags = file->flags;
725 
726 	if (likely(!(eflags & (EVENT_FILE_FL_TRIGGER_MODE |
727 			       EVENT_FILE_FL_SOFT_DISABLED |
728 			       EVENT_FILE_FL_PID_FILTER))))
729 		return false;
730 
731 	if (likely(eflags & EVENT_FILE_FL_TRIGGER_COND))
732 		return false;
733 
734 	return __trace_trigger_soft_disabled(file);
735 }
736 
737 #ifdef CONFIG_BPF_EVENTS
738 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
739 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
740 void perf_event_detach_bpf_prog(struct perf_event *event);
741 int perf_event_query_prog_array(struct perf_event *event, void __user *info);
742 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
743 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
744 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
745 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
746 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
747 			    u32 *fd_type, const char **buf,
748 			    u64 *probe_offset, u64 *probe_addr);
749 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
750 #else
trace_call_bpf(struct trace_event_call * call,void * ctx)751 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
752 {
753 	return 1;
754 }
755 
756 static inline int
perf_event_attach_bpf_prog(struct perf_event * event,struct bpf_prog * prog,u64 bpf_cookie)757 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie)
758 {
759 	return -EOPNOTSUPP;
760 }
761 
perf_event_detach_bpf_prog(struct perf_event * event)762 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
763 
764 static inline int
perf_event_query_prog_array(struct perf_event * event,void __user * info)765 perf_event_query_prog_array(struct perf_event *event, void __user *info)
766 {
767 	return -EOPNOTSUPP;
768 }
bpf_probe_register(struct bpf_raw_event_map * btp,struct bpf_prog * p)769 static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
770 {
771 	return -EOPNOTSUPP;
772 }
bpf_probe_unregister(struct bpf_raw_event_map * btp,struct bpf_prog * p)773 static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
774 {
775 	return -EOPNOTSUPP;
776 }
bpf_get_raw_tracepoint(const char * name)777 static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
778 {
779 	return NULL;
780 }
bpf_put_raw_tracepoint(struct bpf_raw_event_map * btp)781 static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
782 {
783 }
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)784 static inline int bpf_get_perf_event_info(const struct perf_event *event,
785 					  u32 *prog_id, u32 *fd_type,
786 					  const char **buf, u64 *probe_offset,
787 					  u64 *probe_addr)
788 {
789 	return -EOPNOTSUPP;
790 }
791 static inline int
bpf_kprobe_multi_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)792 bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
793 {
794 	return -EOPNOTSUPP;
795 }
796 #endif
797 
798 enum {
799 	FILTER_OTHER = 0,
800 	FILTER_STATIC_STRING,
801 	FILTER_DYN_STRING,
802 	FILTER_RDYN_STRING,
803 	FILTER_PTR_STRING,
804 	FILTER_TRACE_FN,
805 	FILTER_COMM,
806 	FILTER_CPU,
807 };
808 
809 extern int trace_event_raw_init(struct trace_event_call *call);
810 extern int trace_define_field(struct trace_event_call *call, const char *type,
811 			      const char *name, int offset, int size,
812 			      int is_signed, int filter_type);
813 extern int trace_add_event_call(struct trace_event_call *call);
814 extern int trace_remove_event_call(struct trace_event_call *call);
815 extern int trace_event_get_offsets(struct trace_event_call *call);
816 
817 #define is_signed_type(type)	(((type)(-1)) < (type)1)
818 
819 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
820 int trace_set_clr_event(const char *system, const char *event, int set);
821 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
822 		const char *event, bool enable);
823 /*
824  * The double __builtin_constant_p is because gcc will give us an error
825  * if we try to allocate the static variable to fmt if it is not a
826  * constant. Even with the outer if statement optimizing out.
827  */
828 #define event_trace_printk(ip, fmt, args...)				\
829 do {									\
830 	__trace_printk_check_format(fmt, ##args);			\
831 	tracing_record_cmdline(current);				\
832 	if (__builtin_constant_p(fmt)) {				\
833 		static const char *trace_printk_fmt			\
834 		  __section("__trace_printk_fmt") =			\
835 			__builtin_constant_p(fmt) ? fmt : NULL;		\
836 									\
837 		__trace_bprintk(ip, trace_printk_fmt, ##args);		\
838 	} else								\
839 		__trace_printk(ip, fmt, ##args);			\
840 } while (0)
841 
842 #ifdef CONFIG_PERF_EVENTS
843 struct perf_event;
844 
845 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
846 DECLARE_PER_CPU(int, bpf_kprobe_override);
847 
848 extern int  perf_trace_init(struct perf_event *event);
849 extern void perf_trace_destroy(struct perf_event *event);
850 extern int  perf_trace_add(struct perf_event *event, int flags);
851 extern void perf_trace_del(struct perf_event *event, int flags);
852 #ifdef CONFIG_KPROBE_EVENTS
853 extern int  perf_kprobe_init(struct perf_event *event, bool is_retprobe);
854 extern void perf_kprobe_destroy(struct perf_event *event);
855 extern int bpf_get_kprobe_info(const struct perf_event *event,
856 			       u32 *fd_type, const char **symbol,
857 			       u64 *probe_offset, u64 *probe_addr,
858 			       bool perf_type_tracepoint);
859 #endif
860 #ifdef CONFIG_UPROBE_EVENTS
861 extern int  perf_uprobe_init(struct perf_event *event,
862 			     unsigned long ref_ctr_offset, bool is_retprobe);
863 extern void perf_uprobe_destroy(struct perf_event *event);
864 extern int bpf_get_uprobe_info(const struct perf_event *event,
865 			       u32 *fd_type, const char **filename,
866 			       u64 *probe_offset, bool perf_type_tracepoint);
867 #endif
868 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
869 				     char *filter_str);
870 extern void ftrace_profile_free_filter(struct perf_event *event);
871 void perf_trace_buf_update(void *record, u16 type);
872 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
873 
874 int perf_event_set_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
875 void perf_event_free_bpf_prog(struct perf_event *event);
876 
877 void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
878 void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
879 void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
880 		    u64 arg3);
881 void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
882 		    u64 arg3, u64 arg4);
883 void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
884 		    u64 arg3, u64 arg4, u64 arg5);
885 void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
886 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6);
887 void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
888 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
889 void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
890 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
891 		    u64 arg8);
892 void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
893 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
894 		    u64 arg8, u64 arg9);
895 void bpf_trace_run10(struct bpf_prog *prog, u64 arg1, u64 arg2,
896 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
897 		     u64 arg8, u64 arg9, u64 arg10);
898 void bpf_trace_run11(struct bpf_prog *prog, u64 arg1, u64 arg2,
899 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
900 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11);
901 void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
902 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
903 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
904 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
905 			       struct trace_event_call *call, u64 count,
906 			       struct pt_regs *regs, struct hlist_head *head,
907 			       struct task_struct *task);
908 
909 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)910 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
911 		       u64 count, struct pt_regs *regs, void *head,
912 		       struct task_struct *task)
913 {
914 	perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
915 }
916 
917 #endif
918 
919 #define TRACE_EVENT_STR_MAX	512
920 
921 /*
922  * gcc warns that you can not use a va_list in an inlined
923  * function. But lets me make it into a macro :-/
924  */
925 #define __trace_event_vstr_len(fmt, va)			\
926 ({							\
927 	va_list __ap;					\
928 	int __ret;					\
929 							\
930 	va_copy(__ap, *(va));				\
931 	__ret = vsnprintf(NULL, 0, fmt, __ap) + 1;	\
932 	va_end(__ap);					\
933 							\
934 	min(__ret, TRACE_EVENT_STR_MAX);		\
935 })
936 
937 #endif /* _LINUX_TRACE_EVENT_H */
938 
939 /*
940  * Note: we keep the TRACE_CUSTOM_EVENT outside the include file ifdef protection.
941  *  This is due to the way trace custom events work. If a file includes two
942  *  trace event headers under one "CREATE_CUSTOM_TRACE_EVENTS" the first include
943  *  will override the TRACE_CUSTOM_EVENT and break the second include.
944  */
945 
946 #ifndef TRACE_CUSTOM_EVENT
947 
948 #define DECLARE_CUSTOM_EVENT_CLASS(name, proto, args, tstruct, assign, print)
949 #define DEFINE_CUSTOM_EVENT(template, name, proto, args)
950 #define TRACE_CUSTOM_EVENT(name, proto, args, struct, assign, print)
951 
952 #endif /* ifdef TRACE_CUSTOM_EVENT (see note above) */
953