xref: /linux-6.15/include/linux/trace_events.h (revision bbb03029)
1 
2 #ifndef _LINUX_TRACE_EVENT_H
3 #define _LINUX_TRACE_EVENT_H
4 
5 #include <linux/ring_buffer.h>
6 #include <linux/trace_seq.h>
7 #include <linux/percpu.h>
8 #include <linux/hardirq.h>
9 #include <linux/perf_event.h>
10 #include <linux/tracepoint.h>
11 
12 struct trace_array;
13 struct trace_buffer;
14 struct tracer;
15 struct dentry;
16 struct bpf_prog;
17 
18 const char *trace_print_flags_seq(struct trace_seq *p, const char *delim,
19 				  unsigned long flags,
20 				  const struct trace_print_flags *flag_array);
21 
22 const char *trace_print_symbols_seq(struct trace_seq *p, unsigned long val,
23 				    const struct trace_print_flags *symbol_array);
24 
25 #if BITS_PER_LONG == 32
26 const char *trace_print_flags_seq_u64(struct trace_seq *p, const char *delim,
27 		      unsigned long long flags,
28 		      const struct trace_print_flags_u64 *flag_array);
29 
30 const char *trace_print_symbols_seq_u64(struct trace_seq *p,
31 					unsigned long long val,
32 					const struct trace_print_flags_u64
33 								 *symbol_array);
34 #endif
35 
36 const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
37 				    unsigned int bitmask_size);
38 
39 const char *trace_print_hex_seq(struct trace_seq *p,
40 				const unsigned char *buf, int len,
41 				bool concatenate);
42 
43 const char *trace_print_array_seq(struct trace_seq *p,
44 				   const void *buf, int count,
45 				   size_t el_size);
46 
47 struct trace_iterator;
48 struct trace_event;
49 
50 int trace_raw_output_prep(struct trace_iterator *iter,
51 			  struct trace_event *event);
52 
53 /*
54  * The trace entry - the most basic unit of tracing. This is what
55  * is printed in the end as a single line in the trace output, such as:
56  *
57  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
58  */
59 struct trace_entry {
60 	unsigned short		type;
61 	unsigned char		flags;
62 	unsigned char		preempt_count;
63 	int			pid;
64 };
65 
66 #define TRACE_EVENT_TYPE_MAX						\
67 	((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
68 
69 /*
70  * Trace iterator - used by printout routines who present trace
71  * results to users and which routines might sleep, etc:
72  */
73 struct trace_iterator {
74 	struct trace_array	*tr;
75 	struct tracer		*trace;
76 	struct trace_buffer	*trace_buffer;
77 	void			*private;
78 	int			cpu_file;
79 	struct mutex		mutex;
80 	struct ring_buffer_iter	**buffer_iter;
81 	unsigned long		iter_flags;
82 
83 	/* trace_seq for __print_flags() and __print_symbolic() etc. */
84 	struct trace_seq	tmp_seq;
85 
86 	cpumask_var_t		started;
87 
88 	/* it's true when current open file is snapshot */
89 	bool			snapshot;
90 
91 	/* The below is zeroed out in pipe_read */
92 	struct trace_seq	seq;
93 	struct trace_entry	*ent;
94 	unsigned long		lost_events;
95 	int			leftover;
96 	int			ent_size;
97 	int			cpu;
98 	u64			ts;
99 
100 	loff_t			pos;
101 	long			idx;
102 
103 	/* All new field here will be zeroed out in pipe_read */
104 };
105 
106 enum trace_iter_flags {
107 	TRACE_FILE_LAT_FMT	= 1,
108 	TRACE_FILE_ANNOTATE	= 2,
109 	TRACE_FILE_TIME_IN_NS	= 4,
110 };
111 
112 
113 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
114 				      int flags, struct trace_event *event);
115 
116 struct trace_event_functions {
117 	trace_print_func	trace;
118 	trace_print_func	raw;
119 	trace_print_func	hex;
120 	trace_print_func	binary;
121 };
122 
123 struct trace_event {
124 	struct hlist_node		node;
125 	struct list_head		list;
126 	int				type;
127 	struct trace_event_functions	*funcs;
128 };
129 
130 extern int register_trace_event(struct trace_event *event);
131 extern int unregister_trace_event(struct trace_event *event);
132 
133 /* Return values for print_line callback */
134 enum print_line_t {
135 	TRACE_TYPE_PARTIAL_LINE	= 0,	/* Retry after flushing the seq */
136 	TRACE_TYPE_HANDLED	= 1,
137 	TRACE_TYPE_UNHANDLED	= 2,	/* Relay to other output functions */
138 	TRACE_TYPE_NO_CONSUME	= 3	/* Handled but ask to not consume */
139 };
140 
141 enum print_line_t trace_handle_return(struct trace_seq *s);
142 
143 void tracing_generic_entry_update(struct trace_entry *entry,
144 				  unsigned long flags,
145 				  int pc);
146 struct trace_event_file;
147 
148 struct ring_buffer_event *
149 trace_event_buffer_lock_reserve(struct ring_buffer **current_buffer,
150 				struct trace_event_file *trace_file,
151 				int type, unsigned long len,
152 				unsigned long flags, int pc);
153 
154 #define TRACE_RECORD_CMDLINE	BIT(0)
155 #define TRACE_RECORD_TGID	BIT(1)
156 
157 void tracing_record_taskinfo(struct task_struct *task, int flags);
158 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
159 					  struct task_struct *next, int flags);
160 
161 void tracing_record_cmdline(struct task_struct *task);
162 void tracing_record_tgid(struct task_struct *task);
163 
164 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
165 
166 struct event_filter;
167 
168 enum trace_reg {
169 	TRACE_REG_REGISTER,
170 	TRACE_REG_UNREGISTER,
171 #ifdef CONFIG_PERF_EVENTS
172 	TRACE_REG_PERF_REGISTER,
173 	TRACE_REG_PERF_UNREGISTER,
174 	TRACE_REG_PERF_OPEN,
175 	TRACE_REG_PERF_CLOSE,
176 	TRACE_REG_PERF_ADD,
177 	TRACE_REG_PERF_DEL,
178 #endif
179 };
180 
181 struct trace_event_call;
182 
183 struct trace_event_class {
184 	const char		*system;
185 	void			*probe;
186 #ifdef CONFIG_PERF_EVENTS
187 	void			*perf_probe;
188 #endif
189 	int			(*reg)(struct trace_event_call *event,
190 				       enum trace_reg type, void *data);
191 	int			(*define_fields)(struct trace_event_call *);
192 	struct list_head	*(*get_fields)(struct trace_event_call *);
193 	struct list_head	fields;
194 	int			(*raw_init)(struct trace_event_call *);
195 };
196 
197 extern int trace_event_reg(struct trace_event_call *event,
198 			    enum trace_reg type, void *data);
199 
200 struct trace_event_buffer {
201 	struct ring_buffer		*buffer;
202 	struct ring_buffer_event	*event;
203 	struct trace_event_file		*trace_file;
204 	void				*entry;
205 	unsigned long			flags;
206 	int				pc;
207 };
208 
209 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
210 				  struct trace_event_file *trace_file,
211 				  unsigned long len);
212 
213 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
214 
215 enum {
216 	TRACE_EVENT_FL_FILTERED_BIT,
217 	TRACE_EVENT_FL_CAP_ANY_BIT,
218 	TRACE_EVENT_FL_NO_SET_FILTER_BIT,
219 	TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
220 	TRACE_EVENT_FL_WAS_ENABLED_BIT,
221 	TRACE_EVENT_FL_TRACEPOINT_BIT,
222 	TRACE_EVENT_FL_KPROBE_BIT,
223 	TRACE_EVENT_FL_UPROBE_BIT,
224 };
225 
226 /*
227  * Event flags:
228  *  FILTERED	  - The event has a filter attached
229  *  CAP_ANY	  - Any user can enable for perf
230  *  NO_SET_FILTER - Set when filter has error and is to be ignored
231  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
232  *  WAS_ENABLED   - Set and stays set when an event was ever enabled
233  *                    (used for module unloading, if a module event is enabled,
234  *                     it is best to clear the buffers that used it).
235  *  TRACEPOINT    - Event is a tracepoint
236  *  KPROBE        - Event is a kprobe
237  *  UPROBE        - Event is a uprobe
238  */
239 enum {
240 	TRACE_EVENT_FL_FILTERED		= (1 << TRACE_EVENT_FL_FILTERED_BIT),
241 	TRACE_EVENT_FL_CAP_ANY		= (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
242 	TRACE_EVENT_FL_NO_SET_FILTER	= (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
243 	TRACE_EVENT_FL_IGNORE_ENABLE	= (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
244 	TRACE_EVENT_FL_WAS_ENABLED	= (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT),
245 	TRACE_EVENT_FL_TRACEPOINT	= (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
246 	TRACE_EVENT_FL_KPROBE		= (1 << TRACE_EVENT_FL_KPROBE_BIT),
247 	TRACE_EVENT_FL_UPROBE		= (1 << TRACE_EVENT_FL_UPROBE_BIT),
248 };
249 
250 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
251 
252 struct trace_event_call {
253 	struct list_head	list;
254 	struct trace_event_class *class;
255 	union {
256 		char			*name;
257 		/* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
258 		struct tracepoint	*tp;
259 	};
260 	struct trace_event	event;
261 	char			*print_fmt;
262 	struct event_filter	*filter;
263 	void			*mod;
264 	void			*data;
265 	/*
266 	 *   bit 0:		filter_active
267 	 *   bit 1:		allow trace by non root (cap any)
268 	 *   bit 2:		failed to apply filter
269 	 *   bit 3:		trace internal event (do not enable)
270 	 *   bit 4:		Event was enabled by module
271 	 *   bit 5:		use call filter rather than file filter
272 	 *   bit 6:		Event is a tracepoint
273 	 */
274 	int			flags; /* static flags of different events */
275 
276 #ifdef CONFIG_PERF_EVENTS
277 	int				perf_refcount;
278 	struct hlist_head __percpu	*perf_events;
279 	struct bpf_prog			*prog;
280 
281 	int	(*perf_perm)(struct trace_event_call *,
282 			     struct perf_event *);
283 #endif
284 };
285 
286 static inline const char *
287 trace_event_name(struct trace_event_call *call)
288 {
289 	if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
290 		return call->tp ? call->tp->name : NULL;
291 	else
292 		return call->name;
293 }
294 
295 struct trace_array;
296 struct trace_subsystem_dir;
297 
298 enum {
299 	EVENT_FILE_FL_ENABLED_BIT,
300 	EVENT_FILE_FL_RECORDED_CMD_BIT,
301 	EVENT_FILE_FL_RECORDED_TGID_BIT,
302 	EVENT_FILE_FL_FILTERED_BIT,
303 	EVENT_FILE_FL_NO_SET_FILTER_BIT,
304 	EVENT_FILE_FL_SOFT_MODE_BIT,
305 	EVENT_FILE_FL_SOFT_DISABLED_BIT,
306 	EVENT_FILE_FL_TRIGGER_MODE_BIT,
307 	EVENT_FILE_FL_TRIGGER_COND_BIT,
308 	EVENT_FILE_FL_PID_FILTER_BIT,
309 };
310 
311 /*
312  * Event file flags:
313  *  ENABLED	  - The event is enabled
314  *  RECORDED_CMD  - The comms should be recorded at sched_switch
315  *  RECORDED_TGID - The tgids should be recorded at sched_switch
316  *  FILTERED	  - The event has a filter attached
317  *  NO_SET_FILTER - Set when filter has error and is to be ignored
318  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
319  *  SOFT_DISABLED - When set, do not trace the event (even though its
320  *                   tracepoint may be enabled)
321  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
322  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
323  *  PID_FILTER    - When set, the event is filtered based on pid
324  */
325 enum {
326 	EVENT_FILE_FL_ENABLED		= (1 << EVENT_FILE_FL_ENABLED_BIT),
327 	EVENT_FILE_FL_RECORDED_CMD	= (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
328 	EVENT_FILE_FL_RECORDED_TGID	= (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
329 	EVENT_FILE_FL_FILTERED		= (1 << EVENT_FILE_FL_FILTERED_BIT),
330 	EVENT_FILE_FL_NO_SET_FILTER	= (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
331 	EVENT_FILE_FL_SOFT_MODE		= (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
332 	EVENT_FILE_FL_SOFT_DISABLED	= (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
333 	EVENT_FILE_FL_TRIGGER_MODE	= (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
334 	EVENT_FILE_FL_TRIGGER_COND	= (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
335 	EVENT_FILE_FL_PID_FILTER	= (1 << EVENT_FILE_FL_PID_FILTER_BIT),
336 };
337 
338 struct trace_event_file {
339 	struct list_head		list;
340 	struct trace_event_call		*event_call;
341 	struct event_filter __rcu	*filter;
342 	struct dentry			*dir;
343 	struct trace_array		*tr;
344 	struct trace_subsystem_dir	*system;
345 	struct list_head		triggers;
346 
347 	/*
348 	 * 32 bit flags:
349 	 *   bit 0:		enabled
350 	 *   bit 1:		enabled cmd record
351 	 *   bit 2:		enable/disable with the soft disable bit
352 	 *   bit 3:		soft disabled
353 	 *   bit 4:		trigger enabled
354 	 *
355 	 * Note: The bits must be set atomically to prevent races
356 	 * from other writers. Reads of flags do not need to be in
357 	 * sync as they occur in critical sections. But the way flags
358 	 * is currently used, these changes do not affect the code
359 	 * except that when a change is made, it may have a slight
360 	 * delay in propagating the changes to other CPUs due to
361 	 * caching and such. Which is mostly OK ;-)
362 	 */
363 	unsigned long		flags;
364 	atomic_t		sm_ref;	/* soft-mode reference counter */
365 	atomic_t		tm_ref;	/* trigger-mode reference counter */
366 };
367 
368 #define __TRACE_EVENT_FLAGS(name, value)				\
369 	static int __init trace_init_flags_##name(void)			\
370 	{								\
371 		event_##name.flags |= value;				\
372 		return 0;						\
373 	}								\
374 	early_initcall(trace_init_flags_##name);
375 
376 #define __TRACE_EVENT_PERF_PERM(name, expr...)				\
377 	static int perf_perm_##name(struct trace_event_call *tp_event, \
378 				    struct perf_event *p_event)		\
379 	{								\
380 		return ({ expr; });					\
381 	}								\
382 	static int __init trace_init_perf_perm_##name(void)		\
383 	{								\
384 		event_##name.perf_perm = &perf_perm_##name;		\
385 		return 0;						\
386 	}								\
387 	early_initcall(trace_init_perf_perm_##name);
388 
389 #define PERF_MAX_TRACE_SIZE	2048
390 
391 #define MAX_FILTER_STR_VAL	256	/* Should handle KSYM_SYMBOL_LEN */
392 
393 enum event_trigger_type {
394 	ETT_NONE		= (0),
395 	ETT_TRACE_ONOFF		= (1 << 0),
396 	ETT_SNAPSHOT		= (1 << 1),
397 	ETT_STACKTRACE		= (1 << 2),
398 	ETT_EVENT_ENABLE	= (1 << 3),
399 	ETT_EVENT_HIST		= (1 << 4),
400 	ETT_HIST_ENABLE		= (1 << 5),
401 };
402 
403 extern int filter_match_preds(struct event_filter *filter, void *rec);
404 
405 extern enum event_trigger_type event_triggers_call(struct trace_event_file *file,
406 						   void *rec);
407 extern void event_triggers_post_call(struct trace_event_file *file,
408 				     enum event_trigger_type tt,
409 				     void *rec);
410 
411 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
412 
413 /**
414  * trace_trigger_soft_disabled - do triggers and test if soft disabled
415  * @file: The file pointer of the event to test
416  *
417  * If any triggers without filters are attached to this event, they
418  * will be called here. If the event is soft disabled and has no
419  * triggers that require testing the fields, it will return true,
420  * otherwise false.
421  */
422 static inline bool
423 trace_trigger_soft_disabled(struct trace_event_file *file)
424 {
425 	unsigned long eflags = file->flags;
426 
427 	if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
428 		if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
429 			event_triggers_call(file, NULL);
430 		if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
431 			return true;
432 		if (eflags & EVENT_FILE_FL_PID_FILTER)
433 			return trace_event_ignore_this_pid(file);
434 	}
435 	return false;
436 }
437 
438 #ifdef CONFIG_BPF_EVENTS
439 unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx);
440 #else
441 static inline unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx)
442 {
443 	return 1;
444 }
445 #endif
446 
447 enum {
448 	FILTER_OTHER = 0,
449 	FILTER_STATIC_STRING,
450 	FILTER_DYN_STRING,
451 	FILTER_PTR_STRING,
452 	FILTER_TRACE_FN,
453 	FILTER_COMM,
454 	FILTER_CPU,
455 };
456 
457 extern int trace_event_raw_init(struct trace_event_call *call);
458 extern int trace_define_field(struct trace_event_call *call, const char *type,
459 			      const char *name, int offset, int size,
460 			      int is_signed, int filter_type);
461 extern int trace_add_event_call(struct trace_event_call *call);
462 extern int trace_remove_event_call(struct trace_event_call *call);
463 extern int trace_event_get_offsets(struct trace_event_call *call);
464 
465 #define is_signed_type(type)	(((type)(-1)) < (type)1)
466 
467 int trace_set_clr_event(const char *system, const char *event, int set);
468 
469 /*
470  * The double __builtin_constant_p is because gcc will give us an error
471  * if we try to allocate the static variable to fmt if it is not a
472  * constant. Even with the outer if statement optimizing out.
473  */
474 #define event_trace_printk(ip, fmt, args...)				\
475 do {									\
476 	__trace_printk_check_format(fmt, ##args);			\
477 	tracing_record_cmdline(current);				\
478 	if (__builtin_constant_p(fmt)) {				\
479 		static const char *trace_printk_fmt			\
480 		  __attribute__((section("__trace_printk_fmt"))) =	\
481 			__builtin_constant_p(fmt) ? fmt : NULL;		\
482 									\
483 		__trace_bprintk(ip, trace_printk_fmt, ##args);		\
484 	} else								\
485 		__trace_printk(ip, fmt, ##args);			\
486 } while (0)
487 
488 #ifdef CONFIG_PERF_EVENTS
489 struct perf_event;
490 
491 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
492 
493 extern int  perf_trace_init(struct perf_event *event);
494 extern void perf_trace_destroy(struct perf_event *event);
495 extern int  perf_trace_add(struct perf_event *event, int flags);
496 extern void perf_trace_del(struct perf_event *event, int flags);
497 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
498 				     char *filter_str);
499 extern void ftrace_profile_free_filter(struct perf_event *event);
500 void perf_trace_buf_update(void *record, u16 type);
501 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
502 
503 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
504 			       struct trace_event_call *call, u64 count,
505 			       struct pt_regs *regs, struct hlist_head *head,
506 			       struct task_struct *task);
507 
508 static inline void
509 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
510 		       u64 count, struct pt_regs *regs, void *head,
511 		       struct task_struct *task)
512 {
513 	perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
514 }
515 #endif
516 
517 #endif /* _LINUX_TRACE_EVENT_H */
518