xref: /linux-6.15/include/linux/trace_events.h (revision bcb7bdcc)
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 /* Used to find the offset and length of dynamic fields in trace events */
63 struct trace_dynamic_info {
64 #ifdef CONFIG_CPU_BIG_ENDIAN
65 	u16	len;
66 	u16	offset;
67 #else
68 	u16	offset;
69 	u16	len;
70 #endif
71 } __packed;
72 
73 /*
74  * The trace entry - the most basic unit of tracing. This is what
75  * is printed in the end as a single line in the trace output, such as:
76  *
77  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
78  */
79 struct trace_entry {
80 	unsigned short		type;
81 	unsigned char		flags;
82 	unsigned char		preempt_count;
83 	int			pid;
84 };
85 
86 #define TRACE_EVENT_TYPE_MAX						\
87 	((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
88 
89 /*
90  * Trace iterator - used by printout routines who present trace
91  * results to users and which routines might sleep, etc:
92  */
93 struct trace_iterator {
94 	struct trace_array	*tr;
95 	struct tracer		*trace;
96 	struct array_buffer	*array_buffer;
97 	void			*private;
98 	int			cpu_file;
99 	struct mutex		mutex;
100 	struct ring_buffer_iter	**buffer_iter;
101 	unsigned long		iter_flags;
102 	void			*temp;	/* temp holder */
103 	unsigned int		temp_size;
104 	char			*fmt;	/* modified format holder */
105 	unsigned int		fmt_size;
106 	atomic_t		wait_index;
107 
108 	/* trace_seq for __print_flags() and __print_symbolic() etc. */
109 	struct trace_seq	tmp_seq;
110 
111 	cpumask_var_t		started;
112 
113 	/* Set when the file is closed to prevent new waiters */
114 	bool			closed;
115 
116 	/* it's true when current open file is snapshot */
117 	bool			snapshot;
118 
119 	/* The below is zeroed out in pipe_read */
120 	struct trace_seq	seq;
121 	struct trace_entry	*ent;
122 	unsigned long		lost_events;
123 	int			leftover;
124 	int			ent_size;
125 	int			cpu;
126 	u64			ts;
127 
128 	loff_t			pos;
129 	long			idx;
130 
131 	/* All new field here will be zeroed out in pipe_read */
132 };
133 
134 enum trace_iter_flags {
135 	TRACE_FILE_LAT_FMT	= 1,
136 	TRACE_FILE_ANNOTATE	= 2,
137 	TRACE_FILE_TIME_IN_NS	= 4,
138 };
139 
140 
141 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
142 				      int flags, struct trace_event *event);
143 
144 struct trace_event_functions {
145 	trace_print_func	trace;
146 	trace_print_func	raw;
147 	trace_print_func	hex;
148 	trace_print_func	binary;
149 };
150 
151 struct trace_event {
152 	struct hlist_node		node;
153 	int				type;
154 	struct trace_event_functions	*funcs;
155 };
156 
157 extern int register_trace_event(struct trace_event *event);
158 extern int unregister_trace_event(struct trace_event *event);
159 
160 /* Return values for print_line callback */
161 enum print_line_t {
162 	TRACE_TYPE_PARTIAL_LINE	= 0,	/* Retry after flushing the seq */
163 	TRACE_TYPE_HANDLED	= 1,
164 	TRACE_TYPE_UNHANDLED	= 2,	/* Relay to other output functions */
165 	TRACE_TYPE_NO_CONSUME	= 3	/* Handled but ask to not consume */
166 };
167 
168 enum print_line_t trace_handle_return(struct trace_seq *s);
169 
170 static inline void tracing_generic_entry_update(struct trace_entry *entry,
171 						unsigned short type,
172 						unsigned int trace_ctx)
173 {
174 	entry->preempt_count		= trace_ctx & 0xff;
175 	entry->pid			= current->pid;
176 	entry->type			= type;
177 	entry->flags =			trace_ctx >> 16;
178 }
179 
180 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status);
181 
182 enum trace_flag_type {
183 	TRACE_FLAG_IRQS_OFF		= 0x01,
184 	TRACE_FLAG_IRQS_NOSUPPORT	= 0x02,
185 	TRACE_FLAG_NEED_RESCHED		= 0x04,
186 	TRACE_FLAG_HARDIRQ		= 0x08,
187 	TRACE_FLAG_SOFTIRQ		= 0x10,
188 	TRACE_FLAG_PREEMPT_RESCHED	= 0x20,
189 	TRACE_FLAG_NMI			= 0x40,
190 	TRACE_FLAG_BH_OFF		= 0x80,
191 };
192 
193 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
194 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
195 {
196 	unsigned int irq_status = irqs_disabled_flags(irqflags) ?
197 		TRACE_FLAG_IRQS_OFF : 0;
198 	return tracing_gen_ctx_irq_test(irq_status);
199 }
200 static inline unsigned int tracing_gen_ctx(void)
201 {
202 	unsigned long irqflags;
203 
204 	local_save_flags(irqflags);
205 	return tracing_gen_ctx_flags(irqflags);
206 }
207 #else
208 
209 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
210 {
211 	return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
212 }
213 static inline unsigned int tracing_gen_ctx(void)
214 {
215 	return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
216 }
217 #endif
218 
219 static inline unsigned int tracing_gen_ctx_dec(void)
220 {
221 	unsigned int trace_ctx;
222 
223 	trace_ctx = tracing_gen_ctx();
224 	/*
225 	 * Subtract one from the preemption counter if preemption is enabled,
226 	 * see trace_event_buffer_reserve()for details.
227 	 */
228 	if (IS_ENABLED(CONFIG_PREEMPTION))
229 		trace_ctx--;
230 	return trace_ctx;
231 }
232 
233 struct trace_event_file;
234 
235 struct ring_buffer_event *
236 trace_event_buffer_lock_reserve(struct trace_buffer **current_buffer,
237 				struct trace_event_file *trace_file,
238 				int type, unsigned long len,
239 				unsigned int trace_ctx);
240 
241 #define TRACE_RECORD_CMDLINE	BIT(0)
242 #define TRACE_RECORD_TGID	BIT(1)
243 
244 void tracing_record_taskinfo(struct task_struct *task, int flags);
245 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
246 					  struct task_struct *next, int flags);
247 
248 void tracing_record_cmdline(struct task_struct *task);
249 void tracing_record_tgid(struct task_struct *task);
250 
251 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...)
252 	 __printf(3, 4);
253 
254 struct event_filter;
255 
256 enum trace_reg {
257 	TRACE_REG_REGISTER,
258 	TRACE_REG_UNREGISTER,
259 #ifdef CONFIG_PERF_EVENTS
260 	TRACE_REG_PERF_REGISTER,
261 	TRACE_REG_PERF_UNREGISTER,
262 	TRACE_REG_PERF_OPEN,
263 	TRACE_REG_PERF_CLOSE,
264 	/*
265 	 * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
266 	 * custom action was taken and the default action is not to be
267 	 * performed.
268 	 */
269 	TRACE_REG_PERF_ADD,
270 	TRACE_REG_PERF_DEL,
271 #endif
272 };
273 
274 struct trace_event_call;
275 
276 #define TRACE_FUNCTION_TYPE ((const char *)~0UL)
277 
278 struct trace_event_fields {
279 	const char *type;
280 	union {
281 		struct {
282 			const char *name;
283 			const int  size;
284 			const int  align;
285 			const int  is_signed;
286 			const int  filter_type;
287 			const int  len;
288 		};
289 		int (*define_fields)(struct trace_event_call *);
290 	};
291 };
292 
293 struct trace_event_class {
294 	const char		*system;
295 	void			*probe;
296 #ifdef CONFIG_PERF_EVENTS
297 	void			*perf_probe;
298 #endif
299 	int			(*reg)(struct trace_event_call *event,
300 				       enum trace_reg type, void *data);
301 	struct trace_event_fields *fields_array;
302 	struct list_head	*(*get_fields)(struct trace_event_call *);
303 	struct list_head	fields;
304 	int			(*raw_init)(struct trace_event_call *);
305 };
306 
307 extern int trace_event_reg(struct trace_event_call *event,
308 			    enum trace_reg type, void *data);
309 
310 struct trace_event_buffer {
311 	struct trace_buffer		*buffer;
312 	struct ring_buffer_event	*event;
313 	struct trace_event_file		*trace_file;
314 	void				*entry;
315 	unsigned int			trace_ctx;
316 	struct pt_regs			*regs;
317 };
318 
319 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
320 				  struct trace_event_file *trace_file,
321 				  unsigned long len);
322 
323 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
324 
325 enum {
326 	TRACE_EVENT_FL_FILTERED_BIT,
327 	TRACE_EVENT_FL_CAP_ANY_BIT,
328 	TRACE_EVENT_FL_NO_SET_FILTER_BIT,
329 	TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
330 	TRACE_EVENT_FL_TRACEPOINT_BIT,
331 	TRACE_EVENT_FL_DYNAMIC_BIT,
332 	TRACE_EVENT_FL_KPROBE_BIT,
333 	TRACE_EVENT_FL_UPROBE_BIT,
334 	TRACE_EVENT_FL_EPROBE_BIT,
335 	TRACE_EVENT_FL_FPROBE_BIT,
336 	TRACE_EVENT_FL_CUSTOM_BIT,
337 };
338 
339 /*
340  * Event flags:
341  *  FILTERED	  - The event has a filter attached
342  *  CAP_ANY	  - Any user can enable for perf
343  *  NO_SET_FILTER - Set when filter has error and is to be ignored
344  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
345  *  TRACEPOINT    - Event is a tracepoint
346  *  DYNAMIC       - Event is a dynamic event (created at run time)
347  *  KPROBE        - Event is a kprobe
348  *  UPROBE        - Event is a uprobe
349  *  EPROBE        - Event is an event probe
350  *  FPROBE        - Event is an function probe
351  *  CUSTOM        - Event is a custom event (to be attached to an exsiting tracepoint)
352  *                   This is set when the custom event has not been attached
353  *                   to a tracepoint yet, then it is cleared when it is.
354  */
355 enum {
356 	TRACE_EVENT_FL_FILTERED		= (1 << TRACE_EVENT_FL_FILTERED_BIT),
357 	TRACE_EVENT_FL_CAP_ANY		= (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
358 	TRACE_EVENT_FL_NO_SET_FILTER	= (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
359 	TRACE_EVENT_FL_IGNORE_ENABLE	= (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
360 	TRACE_EVENT_FL_TRACEPOINT	= (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
361 	TRACE_EVENT_FL_DYNAMIC		= (1 << TRACE_EVENT_FL_DYNAMIC_BIT),
362 	TRACE_EVENT_FL_KPROBE		= (1 << TRACE_EVENT_FL_KPROBE_BIT),
363 	TRACE_EVENT_FL_UPROBE		= (1 << TRACE_EVENT_FL_UPROBE_BIT),
364 	TRACE_EVENT_FL_EPROBE		= (1 << TRACE_EVENT_FL_EPROBE_BIT),
365 	TRACE_EVENT_FL_FPROBE		= (1 << TRACE_EVENT_FL_FPROBE_BIT),
366 	TRACE_EVENT_FL_CUSTOM		= (1 << TRACE_EVENT_FL_CUSTOM_BIT),
367 };
368 
369 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
370 
371 struct trace_event_call {
372 	struct list_head	list;
373 	struct trace_event_class *class;
374 	union {
375 		char			*name;
376 		/* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
377 		struct tracepoint	*tp;
378 	};
379 	struct trace_event	event;
380 	char			*print_fmt;
381 	struct event_filter	*filter;
382 	/*
383 	 * Static events can disappear with modules,
384 	 * where as dynamic ones need their own ref count.
385 	 */
386 	union {
387 		void				*module;
388 		atomic_t			refcnt;
389 	};
390 	void			*data;
391 
392 	/* See the TRACE_EVENT_FL_* flags above */
393 	int			flags; /* static flags of different events */
394 
395 #ifdef CONFIG_PERF_EVENTS
396 	int				perf_refcount;
397 	struct hlist_head __percpu	*perf_events;
398 	struct bpf_prog_array __rcu	*prog_array;
399 
400 	int	(*perf_perm)(struct trace_event_call *,
401 			     struct perf_event *);
402 #endif
403 };
404 
405 #ifdef CONFIG_DYNAMIC_EVENTS
406 bool trace_event_dyn_try_get_ref(struct trace_event_call *call);
407 void trace_event_dyn_put_ref(struct trace_event_call *call);
408 bool trace_event_dyn_busy(struct trace_event_call *call);
409 #else
410 static inline bool trace_event_dyn_try_get_ref(struct trace_event_call *call)
411 {
412 	/* Without DYNAMIC_EVENTS configured, nothing should be calling this */
413 	return false;
414 }
415 static inline void trace_event_dyn_put_ref(struct trace_event_call *call)
416 {
417 }
418 static inline bool trace_event_dyn_busy(struct trace_event_call *call)
419 {
420 	/* Nothing should call this without DYNAIMIC_EVENTS configured. */
421 	return true;
422 }
423 #endif
424 
425 static inline bool trace_event_try_get_ref(struct trace_event_call *call)
426 {
427 	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
428 		return trace_event_dyn_try_get_ref(call);
429 	else
430 		return try_module_get(call->module);
431 }
432 
433 static inline void trace_event_put_ref(struct trace_event_call *call)
434 {
435 	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
436 		trace_event_dyn_put_ref(call);
437 	else
438 		module_put(call->module);
439 }
440 
441 #ifdef CONFIG_PERF_EVENTS
442 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
443 {
444 	/*
445 	 * This inline function checks whether call->prog_array
446 	 * is valid or not. The function is called in various places,
447 	 * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
448 	 *
449 	 * If this function returns true, and later call->prog_array
450 	 * becomes false inside rcu_read_lock/unlock region,
451 	 * we bail out then. If this function return false,
452 	 * there is a risk that we might miss a few events if the checking
453 	 * were delayed until inside rcu_read_lock/unlock region and
454 	 * call->prog_array happened to become non-NULL then.
455 	 *
456 	 * Here, READ_ONCE() is used instead of rcu_access_pointer().
457 	 * rcu_access_pointer() requires the actual definition of
458 	 * "struct bpf_prog_array" while READ_ONCE() only needs
459 	 * a declaration of the same type.
460 	 */
461 	return !!READ_ONCE(call->prog_array);
462 }
463 #endif
464 
465 static inline const char *
466 trace_event_name(struct trace_event_call *call)
467 {
468 	if (call->flags & TRACE_EVENT_FL_CUSTOM)
469 		return call->name;
470 	else if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
471 		return call->tp ? call->tp->name : NULL;
472 	else
473 		return call->name;
474 }
475 
476 static inline struct list_head *
477 trace_get_fields(struct trace_event_call *event_call)
478 {
479 	if (!event_call->class->get_fields)
480 		return &event_call->class->fields;
481 	return event_call->class->get_fields(event_call);
482 }
483 
484 struct trace_subsystem_dir;
485 
486 enum {
487 	EVENT_FILE_FL_ENABLED_BIT,
488 	EVENT_FILE_FL_RECORDED_CMD_BIT,
489 	EVENT_FILE_FL_RECORDED_TGID_BIT,
490 	EVENT_FILE_FL_FILTERED_BIT,
491 	EVENT_FILE_FL_NO_SET_FILTER_BIT,
492 	EVENT_FILE_FL_SOFT_MODE_BIT,
493 	EVENT_FILE_FL_SOFT_DISABLED_BIT,
494 	EVENT_FILE_FL_TRIGGER_MODE_BIT,
495 	EVENT_FILE_FL_TRIGGER_COND_BIT,
496 	EVENT_FILE_FL_PID_FILTER_BIT,
497 	EVENT_FILE_FL_WAS_ENABLED_BIT,
498 	EVENT_FILE_FL_FREED_BIT,
499 };
500 
501 extern struct trace_event_file *trace_get_event_file(const char *instance,
502 						     const char *system,
503 						     const char *event);
504 extern void trace_put_event_file(struct trace_event_file *file);
505 
506 #define MAX_DYNEVENT_CMD_LEN	(2048)
507 
508 enum dynevent_type {
509 	DYNEVENT_TYPE_SYNTH = 1,
510 	DYNEVENT_TYPE_KPROBE,
511 	DYNEVENT_TYPE_NONE,
512 };
513 
514 struct dynevent_cmd;
515 
516 typedef int (*dynevent_create_fn_t)(struct dynevent_cmd *cmd);
517 
518 struct dynevent_cmd {
519 	struct seq_buf		seq;
520 	const char		*event_name;
521 	unsigned int		n_fields;
522 	enum dynevent_type	type;
523 	dynevent_create_fn_t	run_command;
524 	void			*private_data;
525 };
526 
527 extern int dynevent_create(struct dynevent_cmd *cmd);
528 
529 extern int synth_event_delete(const char *name);
530 
531 extern void synth_event_cmd_init(struct dynevent_cmd *cmd,
532 				 char *buf, int maxlen);
533 
534 extern int __synth_event_gen_cmd_start(struct dynevent_cmd *cmd,
535 				       const char *name,
536 				       struct module *mod, ...);
537 
538 #define synth_event_gen_cmd_start(cmd, name, mod, ...)	\
539 	__synth_event_gen_cmd_start(cmd, name, mod, ## __VA_ARGS__, NULL)
540 
541 struct synth_field_desc {
542 	const char *type;
543 	const char *name;
544 };
545 
546 extern int synth_event_gen_cmd_array_start(struct dynevent_cmd *cmd,
547 					   const char *name,
548 					   struct module *mod,
549 					   struct synth_field_desc *fields,
550 					   unsigned int n_fields);
551 extern int synth_event_create(const char *name,
552 			      struct synth_field_desc *fields,
553 			      unsigned int n_fields, struct module *mod);
554 
555 extern int synth_event_add_field(struct dynevent_cmd *cmd,
556 				 const char *type,
557 				 const char *name);
558 extern int synth_event_add_field_str(struct dynevent_cmd *cmd,
559 				     const char *type_name);
560 extern int synth_event_add_fields(struct dynevent_cmd *cmd,
561 				  struct synth_field_desc *fields,
562 				  unsigned int n_fields);
563 
564 #define synth_event_gen_cmd_end(cmd)	\
565 	dynevent_create(cmd)
566 
567 struct synth_event;
568 
569 struct synth_event_trace_state {
570 	struct trace_event_buffer fbuffer;
571 	struct synth_trace_event *entry;
572 	struct trace_buffer *buffer;
573 	struct synth_event *event;
574 	unsigned int cur_field;
575 	unsigned int n_u64;
576 	bool disabled;
577 	bool add_next;
578 	bool add_name;
579 };
580 
581 extern int synth_event_trace(struct trace_event_file *file,
582 			     unsigned int n_vals, ...);
583 extern int synth_event_trace_array(struct trace_event_file *file, u64 *vals,
584 				   unsigned int n_vals);
585 extern int synth_event_trace_start(struct trace_event_file *file,
586 				   struct synth_event_trace_state *trace_state);
587 extern int synth_event_add_next_val(u64 val,
588 				    struct synth_event_trace_state *trace_state);
589 extern int synth_event_add_val(const char *field_name, u64 val,
590 			       struct synth_event_trace_state *trace_state);
591 extern int synth_event_trace_end(struct synth_event_trace_state *trace_state);
592 
593 extern int kprobe_event_delete(const char *name);
594 
595 extern void kprobe_event_cmd_init(struct dynevent_cmd *cmd,
596 				  char *buf, int maxlen);
597 
598 #define kprobe_event_gen_cmd_start(cmd, name, loc, ...)			\
599 	__kprobe_event_gen_cmd_start(cmd, false, name, loc, ## __VA_ARGS__, NULL)
600 
601 #define kretprobe_event_gen_cmd_start(cmd, name, loc, ...)		\
602 	__kprobe_event_gen_cmd_start(cmd, true, name, loc, ## __VA_ARGS__, NULL)
603 
604 extern int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd,
605 					bool kretprobe,
606 					const char *name,
607 					const char *loc, ...);
608 
609 #define kprobe_event_add_fields(cmd, ...)	\
610 	__kprobe_event_add_fields(cmd, ## __VA_ARGS__, NULL)
611 
612 #define kprobe_event_add_field(cmd, field)	\
613 	__kprobe_event_add_fields(cmd, field, NULL)
614 
615 extern int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...);
616 
617 #define kprobe_event_gen_cmd_end(cmd)		\
618 	dynevent_create(cmd)
619 
620 #define kretprobe_event_gen_cmd_end(cmd)	\
621 	dynevent_create(cmd)
622 
623 /*
624  * Event file flags:
625  *  ENABLED	  - The event is enabled
626  *  RECORDED_CMD  - The comms should be recorded at sched_switch
627  *  RECORDED_TGID - The tgids should be recorded at sched_switch
628  *  FILTERED	  - The event has a filter attached
629  *  NO_SET_FILTER - Set when filter has error and is to be ignored
630  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
631  *  SOFT_DISABLED - When set, do not trace the event (even though its
632  *                   tracepoint may be enabled)
633  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
634  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
635  *  PID_FILTER    - When set, the event is filtered based on pid
636  *  WAS_ENABLED   - Set when enabled to know to clear trace on module removal
637  *  FREED         - File descriptor is freed, all fields should be considered invalid
638  */
639 enum {
640 	EVENT_FILE_FL_ENABLED		= (1 << EVENT_FILE_FL_ENABLED_BIT),
641 	EVENT_FILE_FL_RECORDED_CMD	= (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
642 	EVENT_FILE_FL_RECORDED_TGID	= (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
643 	EVENT_FILE_FL_FILTERED		= (1 << EVENT_FILE_FL_FILTERED_BIT),
644 	EVENT_FILE_FL_NO_SET_FILTER	= (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
645 	EVENT_FILE_FL_SOFT_MODE		= (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
646 	EVENT_FILE_FL_SOFT_DISABLED	= (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
647 	EVENT_FILE_FL_TRIGGER_MODE	= (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
648 	EVENT_FILE_FL_TRIGGER_COND	= (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
649 	EVENT_FILE_FL_PID_FILTER	= (1 << EVENT_FILE_FL_PID_FILTER_BIT),
650 	EVENT_FILE_FL_WAS_ENABLED	= (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
651 	EVENT_FILE_FL_FREED		= (1 << EVENT_FILE_FL_FREED_BIT),
652 };
653 
654 struct trace_event_file {
655 	struct list_head		list;
656 	struct trace_event_call		*event_call;
657 	struct event_filter __rcu	*filter;
658 	struct eventfs_inode		*ei;
659 	struct trace_array		*tr;
660 	struct trace_subsystem_dir	*system;
661 	struct list_head		triggers;
662 
663 	/*
664 	 * 32 bit flags:
665 	 *   bit 0:		enabled
666 	 *   bit 1:		enabled cmd record
667 	 *   bit 2:		enable/disable with the soft disable bit
668 	 *   bit 3:		soft disabled
669 	 *   bit 4:		trigger enabled
670 	 *
671 	 * Note: The bits must be set atomically to prevent races
672 	 * from other writers. Reads of flags do not need to be in
673 	 * sync as they occur in critical sections. But the way flags
674 	 * is currently used, these changes do not affect the code
675 	 * except that when a change is made, it may have a slight
676 	 * delay in propagating the changes to other CPUs due to
677 	 * caching and such. Which is mostly OK ;-)
678 	 */
679 	unsigned long		flags;
680 	atomic_t		ref;	/* ref count for opened files */
681 	atomic_t		sm_ref;	/* soft-mode reference counter */
682 	atomic_t		tm_ref;	/* trigger-mode reference counter */
683 };
684 
685 #define __TRACE_EVENT_FLAGS(name, value)				\
686 	static int __init trace_init_flags_##name(void)			\
687 	{								\
688 		event_##name.flags |= value;				\
689 		return 0;						\
690 	}								\
691 	early_initcall(trace_init_flags_##name);
692 
693 #define __TRACE_EVENT_PERF_PERM(name, expr...)				\
694 	static int perf_perm_##name(struct trace_event_call *tp_event, \
695 				    struct perf_event *p_event)		\
696 	{								\
697 		return ({ expr; });					\
698 	}								\
699 	static int __init trace_init_perf_perm_##name(void)		\
700 	{								\
701 		event_##name.perf_perm = &perf_perm_##name;		\
702 		return 0;						\
703 	}								\
704 	early_initcall(trace_init_perf_perm_##name);
705 
706 #define PERF_MAX_TRACE_SIZE	8192
707 
708 #define MAX_FILTER_STR_VAL	256U	/* Should handle KSYM_SYMBOL_LEN */
709 
710 enum event_trigger_type {
711 	ETT_NONE		= (0),
712 	ETT_TRACE_ONOFF		= (1 << 0),
713 	ETT_SNAPSHOT		= (1 << 1),
714 	ETT_STACKTRACE		= (1 << 2),
715 	ETT_EVENT_ENABLE	= (1 << 3),
716 	ETT_EVENT_HIST		= (1 << 4),
717 	ETT_HIST_ENABLE		= (1 << 5),
718 	ETT_EVENT_EPROBE	= (1 << 6),
719 };
720 
721 extern int filter_match_preds(struct event_filter *filter, void *rec);
722 
723 extern enum event_trigger_type
724 event_triggers_call(struct trace_event_file *file,
725 		    struct trace_buffer *buffer, void *rec,
726 		    struct ring_buffer_event *event);
727 extern void
728 event_triggers_post_call(struct trace_event_file *file,
729 			 enum event_trigger_type tt);
730 
731 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
732 
733 bool __trace_trigger_soft_disabled(struct trace_event_file *file);
734 
735 /**
736  * trace_trigger_soft_disabled - do triggers and test if soft disabled
737  * @file: The file pointer of the event to test
738  *
739  * If any triggers without filters are attached to this event, they
740  * will be called here. If the event is soft disabled and has no
741  * triggers that require testing the fields, it will return true,
742  * otherwise false.
743  */
744 static __always_inline bool
745 trace_trigger_soft_disabled(struct trace_event_file *file)
746 {
747 	unsigned long eflags = file->flags;
748 
749 	if (likely(!(eflags & (EVENT_FILE_FL_TRIGGER_MODE |
750 			       EVENT_FILE_FL_SOFT_DISABLED |
751 			       EVENT_FILE_FL_PID_FILTER))))
752 		return false;
753 
754 	if (likely(eflags & EVENT_FILE_FL_TRIGGER_COND))
755 		return false;
756 
757 	return __trace_trigger_soft_disabled(file);
758 }
759 
760 #ifdef CONFIG_BPF_EVENTS
761 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
762 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
763 void perf_event_detach_bpf_prog(struct perf_event *event);
764 int perf_event_query_prog_array(struct perf_event *event, void __user *info);
765 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
766 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
767 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
768 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
769 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
770 			    u32 *fd_type, const char **buf,
771 			    u64 *probe_offset, u64 *probe_addr,
772 			    unsigned long *missed);
773 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
774 int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
775 #else
776 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
777 {
778 	return 1;
779 }
780 
781 static inline int
782 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie)
783 {
784 	return -EOPNOTSUPP;
785 }
786 
787 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
788 
789 static inline int
790 perf_event_query_prog_array(struct perf_event *event, void __user *info)
791 {
792 	return -EOPNOTSUPP;
793 }
794 static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
795 {
796 	return -EOPNOTSUPP;
797 }
798 static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
799 {
800 	return -EOPNOTSUPP;
801 }
802 static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
803 {
804 	return NULL;
805 }
806 static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
807 {
808 }
809 static inline int bpf_get_perf_event_info(const struct perf_event *event,
810 					  u32 *prog_id, u32 *fd_type,
811 					  const char **buf, u64 *probe_offset,
812 					  u64 *probe_addr, unsigned long *missed)
813 {
814 	return -EOPNOTSUPP;
815 }
816 static inline int
817 bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
818 {
819 	return -EOPNOTSUPP;
820 }
821 static inline int
822 bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
823 {
824 	return -EOPNOTSUPP;
825 }
826 #endif
827 
828 enum {
829 	FILTER_OTHER = 0,
830 	FILTER_STATIC_STRING,
831 	FILTER_DYN_STRING,
832 	FILTER_RDYN_STRING,
833 	FILTER_PTR_STRING,
834 	FILTER_TRACE_FN,
835 	FILTER_CPUMASK,
836 	FILTER_COMM,
837 	FILTER_CPU,
838 	FILTER_STACKTRACE,
839 };
840 
841 extern int trace_event_raw_init(struct trace_event_call *call);
842 extern int trace_define_field(struct trace_event_call *call, const char *type,
843 			      const char *name, int offset, int size,
844 			      int is_signed, int filter_type);
845 extern int trace_add_event_call(struct trace_event_call *call);
846 extern int trace_remove_event_call(struct trace_event_call *call);
847 extern int trace_event_get_offsets(struct trace_event_call *call);
848 
849 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
850 int trace_set_clr_event(const char *system, const char *event, int set);
851 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
852 		const char *event, bool enable);
853 /*
854  * The double __builtin_constant_p is because gcc will give us an error
855  * if we try to allocate the static variable to fmt if it is not a
856  * constant. Even with the outer if statement optimizing out.
857  */
858 #define event_trace_printk(ip, fmt, args...)				\
859 do {									\
860 	__trace_printk_check_format(fmt, ##args);			\
861 	tracing_record_cmdline(current);				\
862 	if (__builtin_constant_p(fmt)) {				\
863 		static const char *trace_printk_fmt			\
864 		  __section("__trace_printk_fmt") =			\
865 			__builtin_constant_p(fmt) ? fmt : NULL;		\
866 									\
867 		__trace_bprintk(ip, trace_printk_fmt, ##args);		\
868 	} else								\
869 		__trace_printk(ip, fmt, ##args);			\
870 } while (0)
871 
872 #ifdef CONFIG_PERF_EVENTS
873 struct perf_event;
874 
875 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
876 DECLARE_PER_CPU(int, bpf_kprobe_override);
877 
878 extern int  perf_trace_init(struct perf_event *event);
879 extern void perf_trace_destroy(struct perf_event *event);
880 extern int  perf_trace_add(struct perf_event *event, int flags);
881 extern void perf_trace_del(struct perf_event *event, int flags);
882 #ifdef CONFIG_KPROBE_EVENTS
883 extern int  perf_kprobe_init(struct perf_event *event, bool is_retprobe);
884 extern void perf_kprobe_destroy(struct perf_event *event);
885 extern int bpf_get_kprobe_info(const struct perf_event *event,
886 			       u32 *fd_type, const char **symbol,
887 			       u64 *probe_offset, u64 *probe_addr,
888 			       unsigned long *missed,
889 			       bool perf_type_tracepoint);
890 #endif
891 #ifdef CONFIG_UPROBE_EVENTS
892 extern int  perf_uprobe_init(struct perf_event *event,
893 			     unsigned long ref_ctr_offset, bool is_retprobe);
894 extern void perf_uprobe_destroy(struct perf_event *event);
895 extern int bpf_get_uprobe_info(const struct perf_event *event,
896 			       u32 *fd_type, const char **filename,
897 			       u64 *probe_offset, u64 *probe_addr,
898 			       bool perf_type_tracepoint);
899 #endif
900 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
901 				     char *filter_str);
902 extern void ftrace_profile_free_filter(struct perf_event *event);
903 void perf_trace_buf_update(void *record, u16 type);
904 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
905 
906 int perf_event_set_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
907 void perf_event_free_bpf_prog(struct perf_event *event);
908 
909 void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
910 void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
911 void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
912 		    u64 arg3);
913 void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
914 		    u64 arg3, u64 arg4);
915 void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
916 		    u64 arg3, u64 arg4, u64 arg5);
917 void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
918 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6);
919 void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
920 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
921 void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
922 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
923 		    u64 arg8);
924 void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
925 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
926 		    u64 arg8, u64 arg9);
927 void bpf_trace_run10(struct bpf_prog *prog, u64 arg1, u64 arg2,
928 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
929 		     u64 arg8, u64 arg9, u64 arg10);
930 void bpf_trace_run11(struct bpf_prog *prog, u64 arg1, u64 arg2,
931 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
932 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11);
933 void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
934 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
935 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
936 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
937 			       struct trace_event_call *call, u64 count,
938 			       struct pt_regs *regs, struct hlist_head *head,
939 			       struct task_struct *task);
940 
941 static inline void
942 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
943 		       u64 count, struct pt_regs *regs, void *head,
944 		       struct task_struct *task)
945 {
946 	perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
947 }
948 
949 #endif
950 
951 #define TRACE_EVENT_STR_MAX	512
952 
953 /*
954  * gcc warns that you can not use a va_list in an inlined
955  * function. But lets me make it into a macro :-/
956  */
957 #define __trace_event_vstr_len(fmt, va)			\
958 ({							\
959 	va_list __ap;					\
960 	int __ret;					\
961 							\
962 	va_copy(__ap, *(va));				\
963 	__ret = vsnprintf(NULL, 0, fmt, __ap) + 1;	\
964 	va_end(__ap);					\
965 							\
966 	min(__ret, TRACE_EVENT_STR_MAX);		\
967 })
968 
969 #endif /* _LINUX_TRACE_EVENT_H */
970 
971 /*
972  * Note: we keep the TRACE_CUSTOM_EVENT outside the include file ifdef protection.
973  *  This is due to the way trace custom events work. If a file includes two
974  *  trace event headers under one "CREATE_CUSTOM_TRACE_EVENTS" the first include
975  *  will override the TRACE_CUSTOM_EVENT and break the second include.
976  */
977 
978 #ifndef TRACE_CUSTOM_EVENT
979 
980 #define DECLARE_CUSTOM_EVENT_CLASS(name, proto, args, tstruct, assign, print)
981 #define DEFINE_CUSTOM_EVENT(template, name, proto, args)
982 #define TRACE_CUSTOM_EVENT(name, proto, args, struct, assign, print)
983 
984 #endif /* ifdef TRACE_CUSTOM_EVENT (see note above) */
985