Loading...
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
13struct trace_array;
14struct array_buffer;
15struct tracer;
16struct dentry;
17struct bpf_prog;
18
19const 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
23const 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
27const 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
31const 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
37const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
38 unsigned int bitmask_size);
39
40const char *trace_print_hex_seq(struct trace_seq *p,
41 const unsigned char *buf, int len,
42 bool concatenate);
43
44const char *trace_print_array_seq(struct trace_seq *p,
45 const void *buf, int count,
46 size_t el_size);
47
48const char *
49trace_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
53struct trace_iterator;
54struct trace_event;
55
56int trace_raw_output_prep(struct trace_iterator *iter,
57 struct trace_event *event);
58extern __printf(2, 3)
59void 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 */
67struct 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 */
81struct 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
95 /* trace_seq for __print_flags() and __print_symbolic() etc. */
96 struct trace_seq tmp_seq;
97
98 cpumask_var_t started;
99
100 /* it's true when current open file is snapshot */
101 bool snapshot;
102
103 /* The below is zeroed out in pipe_read */
104 struct trace_seq seq;
105 struct trace_entry *ent;
106 unsigned long lost_events;
107 int leftover;
108 int ent_size;
109 int cpu;
110 u64 ts;
111
112 loff_t pos;
113 long idx;
114
115 /* All new field here will be zeroed out in pipe_read */
116};
117
118enum trace_iter_flags {
119 TRACE_FILE_LAT_FMT = 1,
120 TRACE_FILE_ANNOTATE = 2,
121 TRACE_FILE_TIME_IN_NS = 4,
122};
123
124
125typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
126 int flags, struct trace_event *event);
127
128struct trace_event_functions {
129 trace_print_func trace;
130 trace_print_func raw;
131 trace_print_func hex;
132 trace_print_func binary;
133};
134
135struct trace_event {
136 struct hlist_node node;
137 struct list_head list;
138 int type;
139 struct trace_event_functions *funcs;
140};
141
142extern int register_trace_event(struct trace_event *event);
143extern int unregister_trace_event(struct trace_event *event);
144
145/* Return values for print_line callback */
146enum print_line_t {
147 TRACE_TYPE_PARTIAL_LINE = 0, /* Retry after flushing the seq */
148 TRACE_TYPE_HANDLED = 1,
149 TRACE_TYPE_UNHANDLED = 2, /* Relay to other output functions */
150 TRACE_TYPE_NO_CONSUME = 3 /* Handled but ask to not consume */
151};
152
153enum print_line_t trace_handle_return(struct trace_seq *s);
154
155static inline void tracing_generic_entry_update(struct trace_entry *entry,
156 unsigned short type,
157 unsigned int trace_ctx)
158{
159 entry->preempt_count = trace_ctx & 0xff;
160 entry->pid = current->pid;
161 entry->type = type;
162 entry->flags = trace_ctx >> 16;
163}
164
165unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status);
166
167enum trace_flag_type {
168 TRACE_FLAG_IRQS_OFF = 0x01,
169 TRACE_FLAG_IRQS_NOSUPPORT = 0x02,
170 TRACE_FLAG_NEED_RESCHED = 0x04,
171 TRACE_FLAG_HARDIRQ = 0x08,
172 TRACE_FLAG_SOFTIRQ = 0x10,
173 TRACE_FLAG_PREEMPT_RESCHED = 0x20,
174 TRACE_FLAG_NMI = 0x40,
175};
176
177#ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
178static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
179{
180 unsigned int irq_status = irqs_disabled_flags(irqflags) ?
181 TRACE_FLAG_IRQS_OFF : 0;
182 return tracing_gen_ctx_irq_test(irq_status);
183}
184static inline unsigned int tracing_gen_ctx(void)
185{
186 unsigned long irqflags;
187
188 local_save_flags(irqflags);
189 return tracing_gen_ctx_flags(irqflags);
190}
191#else
192
193static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
194{
195 return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
196}
197static inline unsigned int tracing_gen_ctx(void)
198{
199 return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
200}
201#endif
202
203static inline unsigned int tracing_gen_ctx_dec(void)
204{
205 unsigned int trace_ctx;
206
207 trace_ctx = tracing_gen_ctx();
208 /*
209 * Subtract one from the preemption counter if preemption is enabled,
210 * see trace_event_buffer_reserve()for details.
211 */
212 if (IS_ENABLED(CONFIG_PREEMPTION))
213 trace_ctx--;
214 return trace_ctx;
215}
216
217struct trace_event_file;
218
219struct ring_buffer_event *
220trace_event_buffer_lock_reserve(struct trace_buffer **current_buffer,
221 struct trace_event_file *trace_file,
222 int type, unsigned long len,
223 unsigned int trace_ctx);
224
225#define TRACE_RECORD_CMDLINE BIT(0)
226#define TRACE_RECORD_TGID BIT(1)
227
228void tracing_record_taskinfo(struct task_struct *task, int flags);
229void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
230 struct task_struct *next, int flags);
231
232void tracing_record_cmdline(struct task_struct *task);
233void tracing_record_tgid(struct task_struct *task);
234
235int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
236
237struct event_filter;
238
239enum trace_reg {
240 TRACE_REG_REGISTER,
241 TRACE_REG_UNREGISTER,
242#ifdef CONFIG_PERF_EVENTS
243 TRACE_REG_PERF_REGISTER,
244 TRACE_REG_PERF_UNREGISTER,
245 TRACE_REG_PERF_OPEN,
246 TRACE_REG_PERF_CLOSE,
247 /*
248 * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
249 * custom action was taken and the default action is not to be
250 * performed.
251 */
252 TRACE_REG_PERF_ADD,
253 TRACE_REG_PERF_DEL,
254#endif
255};
256
257struct trace_event_call;
258
259#define TRACE_FUNCTION_TYPE ((const char *)~0UL)
260
261struct trace_event_fields {
262 const char *type;
263 union {
264 struct {
265 const char *name;
266 const int size;
267 const int align;
268 const int is_signed;
269 const int filter_type;
270 };
271 int (*define_fields)(struct trace_event_call *);
272 };
273};
274
275struct trace_event_class {
276 const char *system;
277 void *probe;
278#ifdef CONFIG_PERF_EVENTS
279 void *perf_probe;
280#endif
281 int (*reg)(struct trace_event_call *event,
282 enum trace_reg type, void *data);
283 struct trace_event_fields *fields_array;
284 struct list_head *(*get_fields)(struct trace_event_call *);
285 struct list_head fields;
286 int (*raw_init)(struct trace_event_call *);
287};
288
289extern int trace_event_reg(struct trace_event_call *event,
290 enum trace_reg type, void *data);
291
292struct trace_event_buffer {
293 struct trace_buffer *buffer;
294 struct ring_buffer_event *event;
295 struct trace_event_file *trace_file;
296 void *entry;
297 unsigned int trace_ctx;
298 struct pt_regs *regs;
299};
300
301void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
302 struct trace_event_file *trace_file,
303 unsigned long len);
304
305void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
306
307enum {
308 TRACE_EVENT_FL_FILTERED_BIT,
309 TRACE_EVENT_FL_CAP_ANY_BIT,
310 TRACE_EVENT_FL_NO_SET_FILTER_BIT,
311 TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
312 TRACE_EVENT_FL_TRACEPOINT_BIT,
313 TRACE_EVENT_FL_KPROBE_BIT,
314 TRACE_EVENT_FL_UPROBE_BIT,
315};
316
317/*
318 * Event flags:
319 * FILTERED - The event has a filter attached
320 * CAP_ANY - Any user can enable for perf
321 * NO_SET_FILTER - Set when filter has error and is to be ignored
322 * IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
323 * TRACEPOINT - Event is a tracepoint
324 * KPROBE - Event is a kprobe
325 * UPROBE - Event is a uprobe
326 */
327enum {
328 TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
329 TRACE_EVENT_FL_CAP_ANY = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
330 TRACE_EVENT_FL_NO_SET_FILTER = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
331 TRACE_EVENT_FL_IGNORE_ENABLE = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
332 TRACE_EVENT_FL_TRACEPOINT = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
333 TRACE_EVENT_FL_KPROBE = (1 << TRACE_EVENT_FL_KPROBE_BIT),
334 TRACE_EVENT_FL_UPROBE = (1 << TRACE_EVENT_FL_UPROBE_BIT),
335};
336
337#define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
338
339struct trace_event_call {
340 struct list_head list;
341 struct trace_event_class *class;
342 union {
343 char *name;
344 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
345 struct tracepoint *tp;
346 };
347 struct trace_event event;
348 char *print_fmt;
349 struct event_filter *filter;
350 void *mod;
351 void *data;
352
353 /* See the TRACE_EVENT_FL_* flags above */
354 int flags; /* static flags of different events */
355
356#ifdef CONFIG_PERF_EVENTS
357 int perf_refcount;
358 struct hlist_head __percpu *perf_events;
359 struct bpf_prog_array __rcu *prog_array;
360
361 int (*perf_perm)(struct trace_event_call *,
362 struct perf_event *);
363#endif
364};
365
366#ifdef CONFIG_PERF_EVENTS
367static inline bool bpf_prog_array_valid(struct trace_event_call *call)
368{
369 /*
370 * This inline function checks whether call->prog_array
371 * is valid or not. The function is called in various places,
372 * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
373 *
374 * If this function returns true, and later call->prog_array
375 * becomes false inside rcu_read_lock/unlock region,
376 * we bail out then. If this function return false,
377 * there is a risk that we might miss a few events if the checking
378 * were delayed until inside rcu_read_lock/unlock region and
379 * call->prog_array happened to become non-NULL then.
380 *
381 * Here, READ_ONCE() is used instead of rcu_access_pointer().
382 * rcu_access_pointer() requires the actual definition of
383 * "struct bpf_prog_array" while READ_ONCE() only needs
384 * a declaration of the same type.
385 */
386 return !!READ_ONCE(call->prog_array);
387}
388#endif
389
390static inline const char *
391trace_event_name(struct trace_event_call *call)
392{
393 if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
394 return call->tp ? call->tp->name : NULL;
395 else
396 return call->name;
397}
398
399static inline struct list_head *
400trace_get_fields(struct trace_event_call *event_call)
401{
402 if (!event_call->class->get_fields)
403 return &event_call->class->fields;
404 return event_call->class->get_fields(event_call);
405}
406
407struct trace_subsystem_dir;
408
409enum {
410 EVENT_FILE_FL_ENABLED_BIT,
411 EVENT_FILE_FL_RECORDED_CMD_BIT,
412 EVENT_FILE_FL_RECORDED_TGID_BIT,
413 EVENT_FILE_FL_FILTERED_BIT,
414 EVENT_FILE_FL_NO_SET_FILTER_BIT,
415 EVENT_FILE_FL_SOFT_MODE_BIT,
416 EVENT_FILE_FL_SOFT_DISABLED_BIT,
417 EVENT_FILE_FL_TRIGGER_MODE_BIT,
418 EVENT_FILE_FL_TRIGGER_COND_BIT,
419 EVENT_FILE_FL_PID_FILTER_BIT,
420 EVENT_FILE_FL_WAS_ENABLED_BIT,
421};
422
423extern struct trace_event_file *trace_get_event_file(const char *instance,
424 const char *system,
425 const char *event);
426extern void trace_put_event_file(struct trace_event_file *file);
427
428#define MAX_DYNEVENT_CMD_LEN (2048)
429
430enum dynevent_type {
431 DYNEVENT_TYPE_SYNTH = 1,
432 DYNEVENT_TYPE_KPROBE,
433 DYNEVENT_TYPE_NONE,
434};
435
436struct dynevent_cmd;
437
438typedef int (*dynevent_create_fn_t)(struct dynevent_cmd *cmd);
439
440struct dynevent_cmd {
441 struct seq_buf seq;
442 const char *event_name;
443 unsigned int n_fields;
444 enum dynevent_type type;
445 dynevent_create_fn_t run_command;
446 void *private_data;
447};
448
449extern int dynevent_create(struct dynevent_cmd *cmd);
450
451extern int synth_event_delete(const char *name);
452
453extern void synth_event_cmd_init(struct dynevent_cmd *cmd,
454 char *buf, int maxlen);
455
456extern int __synth_event_gen_cmd_start(struct dynevent_cmd *cmd,
457 const char *name,
458 struct module *mod, ...);
459
460#define synth_event_gen_cmd_start(cmd, name, mod, ...) \
461 __synth_event_gen_cmd_start(cmd, name, mod, ## __VA_ARGS__, NULL)
462
463struct synth_field_desc {
464 const char *type;
465 const char *name;
466};
467
468extern int synth_event_gen_cmd_array_start(struct dynevent_cmd *cmd,
469 const char *name,
470 struct module *mod,
471 struct synth_field_desc *fields,
472 unsigned int n_fields);
473extern int synth_event_create(const char *name,
474 struct synth_field_desc *fields,
475 unsigned int n_fields, struct module *mod);
476
477extern int synth_event_add_field(struct dynevent_cmd *cmd,
478 const char *type,
479 const char *name);
480extern int synth_event_add_field_str(struct dynevent_cmd *cmd,
481 const char *type_name);
482extern int synth_event_add_fields(struct dynevent_cmd *cmd,
483 struct synth_field_desc *fields,
484 unsigned int n_fields);
485
486#define synth_event_gen_cmd_end(cmd) \
487 dynevent_create(cmd)
488
489struct synth_event;
490
491struct synth_event_trace_state {
492 struct trace_event_buffer fbuffer;
493 struct synth_trace_event *entry;
494 struct trace_buffer *buffer;
495 struct synth_event *event;
496 unsigned int cur_field;
497 unsigned int n_u64;
498 bool disabled;
499 bool add_next;
500 bool add_name;
501};
502
503extern int synth_event_trace(struct trace_event_file *file,
504 unsigned int n_vals, ...);
505extern int synth_event_trace_array(struct trace_event_file *file, u64 *vals,
506 unsigned int n_vals);
507extern int synth_event_trace_start(struct trace_event_file *file,
508 struct synth_event_trace_state *trace_state);
509extern int synth_event_add_next_val(u64 val,
510 struct synth_event_trace_state *trace_state);
511extern int synth_event_add_val(const char *field_name, u64 val,
512 struct synth_event_trace_state *trace_state);
513extern int synth_event_trace_end(struct synth_event_trace_state *trace_state);
514
515extern int kprobe_event_delete(const char *name);
516
517extern void kprobe_event_cmd_init(struct dynevent_cmd *cmd,
518 char *buf, int maxlen);
519
520#define kprobe_event_gen_cmd_start(cmd, name, loc, ...) \
521 __kprobe_event_gen_cmd_start(cmd, false, name, loc, ## __VA_ARGS__, NULL)
522
523#define kretprobe_event_gen_cmd_start(cmd, name, loc, ...) \
524 __kprobe_event_gen_cmd_start(cmd, true, name, loc, ## __VA_ARGS__, NULL)
525
526extern int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd,
527 bool kretprobe,
528 const char *name,
529 const char *loc, ...);
530
531#define kprobe_event_add_fields(cmd, ...) \
532 __kprobe_event_add_fields(cmd, ## __VA_ARGS__, NULL)
533
534#define kprobe_event_add_field(cmd, field) \
535 __kprobe_event_add_fields(cmd, field, NULL)
536
537extern int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...);
538
539#define kprobe_event_gen_cmd_end(cmd) \
540 dynevent_create(cmd)
541
542#define kretprobe_event_gen_cmd_end(cmd) \
543 dynevent_create(cmd)
544
545/*
546 * Event file flags:
547 * ENABLED - The event is enabled
548 * RECORDED_CMD - The comms should be recorded at sched_switch
549 * RECORDED_TGID - The tgids should be recorded at sched_switch
550 * FILTERED - The event has a filter attached
551 * NO_SET_FILTER - Set when filter has error and is to be ignored
552 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
553 * SOFT_DISABLED - When set, do not trace the event (even though its
554 * tracepoint may be enabled)
555 * TRIGGER_MODE - When set, invoke the triggers associated with the event
556 * TRIGGER_COND - When set, one or more triggers has an associated filter
557 * PID_FILTER - When set, the event is filtered based on pid
558 * WAS_ENABLED - Set when enabled to know to clear trace on module removal
559 */
560enum {
561 EVENT_FILE_FL_ENABLED = (1 << EVENT_FILE_FL_ENABLED_BIT),
562 EVENT_FILE_FL_RECORDED_CMD = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
563 EVENT_FILE_FL_RECORDED_TGID = (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
564 EVENT_FILE_FL_FILTERED = (1 << EVENT_FILE_FL_FILTERED_BIT),
565 EVENT_FILE_FL_NO_SET_FILTER = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
566 EVENT_FILE_FL_SOFT_MODE = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
567 EVENT_FILE_FL_SOFT_DISABLED = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
568 EVENT_FILE_FL_TRIGGER_MODE = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
569 EVENT_FILE_FL_TRIGGER_COND = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
570 EVENT_FILE_FL_PID_FILTER = (1 << EVENT_FILE_FL_PID_FILTER_BIT),
571 EVENT_FILE_FL_WAS_ENABLED = (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
572};
573
574struct trace_event_file {
575 struct list_head list;
576 struct trace_event_call *event_call;
577 struct event_filter __rcu *filter;
578 struct dentry *dir;
579 struct trace_array *tr;
580 struct trace_subsystem_dir *system;
581 struct list_head triggers;
582
583 /*
584 * 32 bit flags:
585 * bit 0: enabled
586 * bit 1: enabled cmd record
587 * bit 2: enable/disable with the soft disable bit
588 * bit 3: soft disabled
589 * bit 4: trigger enabled
590 *
591 * Note: The bits must be set atomically to prevent races
592 * from other writers. Reads of flags do not need to be in
593 * sync as they occur in critical sections. But the way flags
594 * is currently used, these changes do not affect the code
595 * except that when a change is made, it may have a slight
596 * delay in propagating the changes to other CPUs due to
597 * caching and such. Which is mostly OK ;-)
598 */
599 unsigned long flags;
600 atomic_t sm_ref; /* soft-mode reference counter */
601 atomic_t tm_ref; /* trigger-mode reference counter */
602};
603
604#define __TRACE_EVENT_FLAGS(name, value) \
605 static int __init trace_init_flags_##name(void) \
606 { \
607 event_##name.flags |= value; \
608 return 0; \
609 } \
610 early_initcall(trace_init_flags_##name);
611
612#define __TRACE_EVENT_PERF_PERM(name, expr...) \
613 static int perf_perm_##name(struct trace_event_call *tp_event, \
614 struct perf_event *p_event) \
615 { \
616 return ({ expr; }); \
617 } \
618 static int __init trace_init_perf_perm_##name(void) \
619 { \
620 event_##name.perf_perm = &perf_perm_##name; \
621 return 0; \
622 } \
623 early_initcall(trace_init_perf_perm_##name);
624
625#define PERF_MAX_TRACE_SIZE 2048
626
627#define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */
628
629enum event_trigger_type {
630 ETT_NONE = (0),
631 ETT_TRACE_ONOFF = (1 << 0),
632 ETT_SNAPSHOT = (1 << 1),
633 ETT_STACKTRACE = (1 << 2),
634 ETT_EVENT_ENABLE = (1 << 3),
635 ETT_EVENT_HIST = (1 << 4),
636 ETT_HIST_ENABLE = (1 << 5),
637};
638
639extern int filter_match_preds(struct event_filter *filter, void *rec);
640
641extern enum event_trigger_type
642event_triggers_call(struct trace_event_file *file,
643 struct trace_buffer *buffer, void *rec,
644 struct ring_buffer_event *event);
645extern void
646event_triggers_post_call(struct trace_event_file *file,
647 enum event_trigger_type tt);
648
649bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
650
651/**
652 * trace_trigger_soft_disabled - do triggers and test if soft disabled
653 * @file: The file pointer of the event to test
654 *
655 * If any triggers without filters are attached to this event, they
656 * will be called here. If the event is soft disabled and has no
657 * triggers that require testing the fields, it will return true,
658 * otherwise false.
659 */
660static inline bool
661trace_trigger_soft_disabled(struct trace_event_file *file)
662{
663 unsigned long eflags = file->flags;
664
665 if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
666 if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
667 event_triggers_call(file, NULL, NULL, NULL);
668 if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
669 return true;
670 if (eflags & EVENT_FILE_FL_PID_FILTER)
671 return trace_event_ignore_this_pid(file);
672 }
673 return false;
674}
675
676#ifdef CONFIG_BPF_EVENTS
677unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
678int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog);
679void perf_event_detach_bpf_prog(struct perf_event *event);
680int perf_event_query_prog_array(struct perf_event *event, void __user *info);
681int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
682int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
683struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
684void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
685int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
686 u32 *fd_type, const char **buf,
687 u64 *probe_offset, u64 *probe_addr);
688#else
689static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
690{
691 return 1;
692}
693
694static inline int
695perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog)
696{
697 return -EOPNOTSUPP;
698}
699
700static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
701
702static inline int
703perf_event_query_prog_array(struct perf_event *event, void __user *info)
704{
705 return -EOPNOTSUPP;
706}
707static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
708{
709 return -EOPNOTSUPP;
710}
711static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
712{
713 return -EOPNOTSUPP;
714}
715static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
716{
717 return NULL;
718}
719static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
720{
721}
722static inline int bpf_get_perf_event_info(const struct perf_event *event,
723 u32 *prog_id, u32 *fd_type,
724 const char **buf, u64 *probe_offset,
725 u64 *probe_addr)
726{
727 return -EOPNOTSUPP;
728}
729#endif
730
731enum {
732 FILTER_OTHER = 0,
733 FILTER_STATIC_STRING,
734 FILTER_DYN_STRING,
735 FILTER_PTR_STRING,
736 FILTER_TRACE_FN,
737 FILTER_COMM,
738 FILTER_CPU,
739};
740
741extern int trace_event_raw_init(struct trace_event_call *call);
742extern int trace_define_field(struct trace_event_call *call, const char *type,
743 const char *name, int offset, int size,
744 int is_signed, int filter_type);
745extern int trace_add_event_call(struct trace_event_call *call);
746extern int trace_remove_event_call(struct trace_event_call *call);
747extern int trace_event_get_offsets(struct trace_event_call *call);
748
749#define is_signed_type(type) (((type)(-1)) < (type)1)
750
751int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
752int trace_set_clr_event(const char *system, const char *event, int set);
753int trace_array_set_clr_event(struct trace_array *tr, const char *system,
754 const char *event, bool enable);
755/*
756 * The double __builtin_constant_p is because gcc will give us an error
757 * if we try to allocate the static variable to fmt if it is not a
758 * constant. Even with the outer if statement optimizing out.
759 */
760#define event_trace_printk(ip, fmt, args...) \
761do { \
762 __trace_printk_check_format(fmt, ##args); \
763 tracing_record_cmdline(current); \
764 if (__builtin_constant_p(fmt)) { \
765 static const char *trace_printk_fmt \
766 __section("__trace_printk_fmt") = \
767 __builtin_constant_p(fmt) ? fmt : NULL; \
768 \
769 __trace_bprintk(ip, trace_printk_fmt, ##args); \
770 } else \
771 __trace_printk(ip, fmt, ##args); \
772} while (0)
773
774#ifdef CONFIG_PERF_EVENTS
775struct perf_event;
776
777DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
778DECLARE_PER_CPU(int, bpf_kprobe_override);
779
780extern int perf_trace_init(struct perf_event *event);
781extern void perf_trace_destroy(struct perf_event *event);
782extern int perf_trace_add(struct perf_event *event, int flags);
783extern void perf_trace_del(struct perf_event *event, int flags);
784#ifdef CONFIG_KPROBE_EVENTS
785extern int perf_kprobe_init(struct perf_event *event, bool is_retprobe);
786extern void perf_kprobe_destroy(struct perf_event *event);
787extern int bpf_get_kprobe_info(const struct perf_event *event,
788 u32 *fd_type, const char **symbol,
789 u64 *probe_offset, u64 *probe_addr,
790 bool perf_type_tracepoint);
791#endif
792#ifdef CONFIG_UPROBE_EVENTS
793extern int perf_uprobe_init(struct perf_event *event,
794 unsigned long ref_ctr_offset, bool is_retprobe);
795extern void perf_uprobe_destroy(struct perf_event *event);
796extern int bpf_get_uprobe_info(const struct perf_event *event,
797 u32 *fd_type, const char **filename,
798 u64 *probe_offset, bool perf_type_tracepoint);
799#endif
800extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
801 char *filter_str);
802extern void ftrace_profile_free_filter(struct perf_event *event);
803void perf_trace_buf_update(void *record, u16 type);
804void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
805
806void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
807void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
808void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
809 u64 arg3);
810void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
811 u64 arg3, u64 arg4);
812void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
813 u64 arg3, u64 arg4, u64 arg5);
814void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
815 u64 arg3, u64 arg4, u64 arg5, u64 arg6);
816void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
817 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
818void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
819 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
820 u64 arg8);
821void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
822 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
823 u64 arg8, u64 arg9);
824void bpf_trace_run10(struct bpf_prog *prog, u64 arg1, u64 arg2,
825 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
826 u64 arg8, u64 arg9, u64 arg10);
827void bpf_trace_run11(struct bpf_prog *prog, u64 arg1, u64 arg2,
828 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
829 u64 arg8, u64 arg9, u64 arg10, u64 arg11);
830void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
831 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
832 u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
833void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
834 struct trace_event_call *call, u64 count,
835 struct pt_regs *regs, struct hlist_head *head,
836 struct task_struct *task);
837
838static inline void
839perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
840 u64 count, struct pt_regs *regs, void *head,
841 struct task_struct *task)
842{
843 perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
844}
845
846#endif
847
848#endif /* _LINUX_TRACE_EVENT_H */
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
13struct trace_array;
14struct array_buffer;
15struct tracer;
16struct dentry;
17struct bpf_prog;
18union bpf_attr;
19
20/* Used for event string fields when they are NULL */
21#define EVENT_NULL_STR "(null)"
22
23const char *trace_print_flags_seq(struct trace_seq *p, const char *delim,
24 unsigned long flags,
25 const struct trace_print_flags *flag_array);
26
27const char *trace_print_symbols_seq(struct trace_seq *p, unsigned long val,
28 const struct trace_print_flags *symbol_array);
29
30#if BITS_PER_LONG == 32
31const char *trace_print_flags_seq_u64(struct trace_seq *p, const char *delim,
32 unsigned long long flags,
33 const struct trace_print_flags_u64 *flag_array);
34
35const char *trace_print_symbols_seq_u64(struct trace_seq *p,
36 unsigned long long val,
37 const struct trace_print_flags_u64
38 *symbol_array);
39#endif
40
41const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
42 unsigned int bitmask_size);
43
44const char *trace_print_hex_seq(struct trace_seq *p,
45 const unsigned char *buf, int len,
46 bool concatenate);
47
48const char *trace_print_array_seq(struct trace_seq *p,
49 const void *buf, int count,
50 size_t el_size);
51
52const char *
53trace_print_hex_dump_seq(struct trace_seq *p, const char *prefix_str,
54 int prefix_type, int rowsize, int groupsize,
55 const void *buf, size_t len, bool ascii);
56
57struct trace_iterator;
58struct trace_event;
59
60int trace_raw_output_prep(struct trace_iterator *iter,
61 struct trace_event *event);
62extern __printf(2, 3)
63void trace_event_printf(struct trace_iterator *iter, const char *fmt, ...);
64
65/* Used to find the offset and length of dynamic fields in trace events */
66struct trace_dynamic_info {
67#ifdef CONFIG_CPU_BIG_ENDIAN
68 u16 len;
69 u16 offset;
70#else
71 u16 offset;
72 u16 len;
73#endif
74} __packed;
75
76/*
77 * The trace entry - the most basic unit of tracing. This is what
78 * is printed in the end as a single line in the trace output, such as:
79 *
80 * bash-15816 [01] 235.197585: idle_cpu <- irq_enter
81 */
82struct trace_entry {
83 unsigned short type;
84 unsigned char flags;
85 unsigned char preempt_count;
86 int pid;
87};
88
89#define TRACE_EVENT_TYPE_MAX \
90 ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
91
92/*
93 * Trace iterator - used by printout routines who present trace
94 * results to users and which routines might sleep, etc:
95 */
96struct trace_iterator {
97 struct trace_array *tr;
98 struct tracer *trace;
99 struct array_buffer *array_buffer;
100 void *private;
101 int cpu_file;
102 struct mutex mutex;
103 struct ring_buffer_iter **buffer_iter;
104 unsigned long iter_flags;
105 void *temp; /* temp holder */
106 unsigned int temp_size;
107 char *fmt; /* modified format holder */
108 unsigned int fmt_size;
109 atomic_t wait_index;
110
111 /* trace_seq for __print_flags() and __print_symbolic() etc. */
112 struct trace_seq tmp_seq;
113
114 cpumask_var_t started;
115
116 /* Set when the file is closed to prevent new waiters */
117 bool closed;
118
119 /* it's true when current open file is snapshot */
120 bool snapshot;
121
122 /* The below is zeroed out in pipe_read */
123 struct trace_seq seq;
124 struct trace_entry *ent;
125 unsigned long lost_events;
126 int leftover;
127 int ent_size;
128 int cpu;
129 u64 ts;
130
131 loff_t pos;
132 long idx;
133
134 /* All new field here will be zeroed out in pipe_read */
135};
136
137enum trace_iter_flags {
138 TRACE_FILE_LAT_FMT = 1,
139 TRACE_FILE_ANNOTATE = 2,
140 TRACE_FILE_TIME_IN_NS = 4,
141};
142
143
144typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
145 int flags, struct trace_event *event);
146
147struct trace_event_functions {
148 trace_print_func trace;
149 trace_print_func raw;
150 trace_print_func hex;
151 trace_print_func binary;
152};
153
154struct trace_event {
155 struct hlist_node node;
156 int type;
157 struct trace_event_functions *funcs;
158};
159
160extern int register_trace_event(struct trace_event *event);
161extern int unregister_trace_event(struct trace_event *event);
162
163/* Return values for print_line callback */
164enum print_line_t {
165 TRACE_TYPE_PARTIAL_LINE = 0, /* Retry after flushing the seq */
166 TRACE_TYPE_HANDLED = 1,
167 TRACE_TYPE_UNHANDLED = 2, /* Relay to other output functions */
168 TRACE_TYPE_NO_CONSUME = 3 /* Handled but ask to not consume */
169};
170
171enum print_line_t trace_handle_return(struct trace_seq *s);
172
173static inline void tracing_generic_entry_update(struct trace_entry *entry,
174 unsigned short type,
175 unsigned int trace_ctx)
176{
177 entry->preempt_count = trace_ctx & 0xff;
178 entry->pid = current->pid;
179 entry->type = type;
180 entry->flags = trace_ctx >> 16;
181}
182
183unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status);
184
185enum trace_flag_type {
186 TRACE_FLAG_IRQS_OFF = 0x01,
187 TRACE_FLAG_IRQS_NOSUPPORT = 0x02,
188 TRACE_FLAG_NEED_RESCHED = 0x04,
189 TRACE_FLAG_HARDIRQ = 0x08,
190 TRACE_FLAG_SOFTIRQ = 0x10,
191 TRACE_FLAG_PREEMPT_RESCHED = 0x20,
192 TRACE_FLAG_NMI = 0x40,
193 TRACE_FLAG_BH_OFF = 0x80,
194};
195
196#ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
197static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
198{
199 unsigned int irq_status = irqs_disabled_flags(irqflags) ?
200 TRACE_FLAG_IRQS_OFF : 0;
201 return tracing_gen_ctx_irq_test(irq_status);
202}
203static inline unsigned int tracing_gen_ctx(void)
204{
205 unsigned long irqflags;
206
207 local_save_flags(irqflags);
208 return tracing_gen_ctx_flags(irqflags);
209}
210#else
211
212static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
213{
214 return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
215}
216static inline unsigned int tracing_gen_ctx(void)
217{
218 return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
219}
220#endif
221
222static inline unsigned int tracing_gen_ctx_dec(void)
223{
224 unsigned int trace_ctx;
225
226 trace_ctx = tracing_gen_ctx();
227 /*
228 * Subtract one from the preemption counter if preemption is enabled,
229 * see trace_event_buffer_reserve()for details.
230 */
231 if (IS_ENABLED(CONFIG_PREEMPTION))
232 trace_ctx--;
233 return trace_ctx;
234}
235
236struct trace_event_file;
237
238struct ring_buffer_event *
239trace_event_buffer_lock_reserve(struct trace_buffer **current_buffer,
240 struct trace_event_file *trace_file,
241 int type, unsigned long len,
242 unsigned int trace_ctx);
243
244#define TRACE_RECORD_CMDLINE BIT(0)
245#define TRACE_RECORD_TGID BIT(1)
246
247void tracing_record_taskinfo(struct task_struct *task, int flags);
248void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
249 struct task_struct *next, int flags);
250
251void tracing_record_cmdline(struct task_struct *task);
252void tracing_record_tgid(struct task_struct *task);
253
254int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...)
255 __printf(3, 4);
256
257struct event_filter;
258
259enum trace_reg {
260 TRACE_REG_REGISTER,
261 TRACE_REG_UNREGISTER,
262#ifdef CONFIG_PERF_EVENTS
263 TRACE_REG_PERF_REGISTER,
264 TRACE_REG_PERF_UNREGISTER,
265 TRACE_REG_PERF_OPEN,
266 TRACE_REG_PERF_CLOSE,
267 /*
268 * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
269 * custom action was taken and the default action is not to be
270 * performed.
271 */
272 TRACE_REG_PERF_ADD,
273 TRACE_REG_PERF_DEL,
274#endif
275};
276
277struct trace_event_call;
278
279#define TRACE_FUNCTION_TYPE ((const char *)~0UL)
280
281struct trace_event_fields {
282 const char *type;
283 union {
284 struct {
285 const char *name;
286 const int size;
287 const int align;
288 const int is_signed;
289 const int filter_type;
290 const int len;
291 };
292 int (*define_fields)(struct trace_event_call *);
293 };
294};
295
296struct trace_event_class {
297 const char *system;
298 void *probe;
299#ifdef CONFIG_PERF_EVENTS
300 void *perf_probe;
301#endif
302 int (*reg)(struct trace_event_call *event,
303 enum trace_reg type, void *data);
304 struct trace_event_fields *fields_array;
305 struct list_head *(*get_fields)(struct trace_event_call *);
306 struct list_head fields;
307 int (*raw_init)(struct trace_event_call *);
308};
309
310extern int trace_event_reg(struct trace_event_call *event,
311 enum trace_reg type, void *data);
312
313struct trace_event_buffer {
314 struct trace_buffer *buffer;
315 struct ring_buffer_event *event;
316 struct trace_event_file *trace_file;
317 void *entry;
318 unsigned int trace_ctx;
319 struct pt_regs *regs;
320};
321
322void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
323 struct trace_event_file *trace_file,
324 unsigned long len);
325
326void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
327
328enum {
329 TRACE_EVENT_FL_FILTERED_BIT,
330 TRACE_EVENT_FL_CAP_ANY_BIT,
331 TRACE_EVENT_FL_NO_SET_FILTER_BIT,
332 TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
333 TRACE_EVENT_FL_TRACEPOINT_BIT,
334 TRACE_EVENT_FL_DYNAMIC_BIT,
335 TRACE_EVENT_FL_KPROBE_BIT,
336 TRACE_EVENT_FL_UPROBE_BIT,
337 TRACE_EVENT_FL_EPROBE_BIT,
338 TRACE_EVENT_FL_FPROBE_BIT,
339 TRACE_EVENT_FL_CUSTOM_BIT,
340};
341
342/*
343 * Event flags:
344 * FILTERED - The event has a filter attached
345 * CAP_ANY - Any user can enable for perf
346 * NO_SET_FILTER - Set when filter has error and is to be ignored
347 * IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
348 * TRACEPOINT - Event is a tracepoint
349 * DYNAMIC - Event is a dynamic event (created at run time)
350 * KPROBE - Event is a kprobe
351 * UPROBE - Event is a uprobe
352 * EPROBE - Event is an event probe
353 * FPROBE - Event is an function probe
354 * CUSTOM - Event is a custom event (to be attached to an exsiting tracepoint)
355 * This is set when the custom event has not been attached
356 * to a tracepoint yet, then it is cleared when it is.
357 */
358enum {
359 TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
360 TRACE_EVENT_FL_CAP_ANY = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
361 TRACE_EVENT_FL_NO_SET_FILTER = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
362 TRACE_EVENT_FL_IGNORE_ENABLE = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
363 TRACE_EVENT_FL_TRACEPOINT = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
364 TRACE_EVENT_FL_DYNAMIC = (1 << TRACE_EVENT_FL_DYNAMIC_BIT),
365 TRACE_EVENT_FL_KPROBE = (1 << TRACE_EVENT_FL_KPROBE_BIT),
366 TRACE_EVENT_FL_UPROBE = (1 << TRACE_EVENT_FL_UPROBE_BIT),
367 TRACE_EVENT_FL_EPROBE = (1 << TRACE_EVENT_FL_EPROBE_BIT),
368 TRACE_EVENT_FL_FPROBE = (1 << TRACE_EVENT_FL_FPROBE_BIT),
369 TRACE_EVENT_FL_CUSTOM = (1 << TRACE_EVENT_FL_CUSTOM_BIT),
370};
371
372#define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
373
374struct trace_event_call {
375 struct list_head list;
376 struct trace_event_class *class;
377 union {
378 char *name;
379 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
380 struct tracepoint *tp;
381 };
382 struct trace_event event;
383 char *print_fmt;
384 struct event_filter *filter;
385 /*
386 * Static events can disappear with modules,
387 * where as dynamic ones need their own ref count.
388 */
389 union {
390 void *module;
391 atomic_t refcnt;
392 };
393 void *data;
394
395 /* See the TRACE_EVENT_FL_* flags above */
396 int flags; /* static flags of different events */
397
398#ifdef CONFIG_PERF_EVENTS
399 int perf_refcount;
400 struct hlist_head __percpu *perf_events;
401 struct bpf_prog_array __rcu *prog_array;
402
403 int (*perf_perm)(struct trace_event_call *,
404 struct perf_event *);
405#endif
406};
407
408#ifdef CONFIG_DYNAMIC_EVENTS
409bool trace_event_dyn_try_get_ref(struct trace_event_call *call);
410void trace_event_dyn_put_ref(struct trace_event_call *call);
411bool trace_event_dyn_busy(struct trace_event_call *call);
412#else
413static inline bool trace_event_dyn_try_get_ref(struct trace_event_call *call)
414{
415 /* Without DYNAMIC_EVENTS configured, nothing should be calling this */
416 return false;
417}
418static inline void trace_event_dyn_put_ref(struct trace_event_call *call)
419{
420}
421static inline bool trace_event_dyn_busy(struct trace_event_call *call)
422{
423 /* Nothing should call this without DYNAIMIC_EVENTS configured. */
424 return true;
425}
426#endif
427
428static inline bool trace_event_try_get_ref(struct trace_event_call *call)
429{
430 if (call->flags & TRACE_EVENT_FL_DYNAMIC)
431 return trace_event_dyn_try_get_ref(call);
432 else
433 return try_module_get(call->module);
434}
435
436static inline void trace_event_put_ref(struct trace_event_call *call)
437{
438 if (call->flags & TRACE_EVENT_FL_DYNAMIC)
439 trace_event_dyn_put_ref(call);
440 else
441 module_put(call->module);
442}
443
444#ifdef CONFIG_PERF_EVENTS
445static inline bool bpf_prog_array_valid(struct trace_event_call *call)
446{
447 /*
448 * This inline function checks whether call->prog_array
449 * is valid or not. The function is called in various places,
450 * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
451 *
452 * If this function returns true, and later call->prog_array
453 * becomes false inside rcu_read_lock/unlock region,
454 * we bail out then. If this function return false,
455 * there is a risk that we might miss a few events if the checking
456 * were delayed until inside rcu_read_lock/unlock region and
457 * call->prog_array happened to become non-NULL then.
458 *
459 * Here, READ_ONCE() is used instead of rcu_access_pointer().
460 * rcu_access_pointer() requires the actual definition of
461 * "struct bpf_prog_array" while READ_ONCE() only needs
462 * a declaration of the same type.
463 */
464 return !!READ_ONCE(call->prog_array);
465}
466#endif
467
468static inline const char *
469trace_event_name(struct trace_event_call *call)
470{
471 if (call->flags & TRACE_EVENT_FL_CUSTOM)
472 return call->name;
473 else if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
474 return call->tp ? call->tp->name : NULL;
475 else
476 return call->name;
477}
478
479static inline struct list_head *
480trace_get_fields(struct trace_event_call *event_call)
481{
482 if (!event_call->class->get_fields)
483 return &event_call->class->fields;
484 return event_call->class->get_fields(event_call);
485}
486
487struct trace_subsystem_dir;
488
489enum {
490 EVENT_FILE_FL_ENABLED_BIT,
491 EVENT_FILE_FL_RECORDED_CMD_BIT,
492 EVENT_FILE_FL_RECORDED_TGID_BIT,
493 EVENT_FILE_FL_FILTERED_BIT,
494 EVENT_FILE_FL_NO_SET_FILTER_BIT,
495 EVENT_FILE_FL_SOFT_MODE_BIT,
496 EVENT_FILE_FL_SOFT_DISABLED_BIT,
497 EVENT_FILE_FL_TRIGGER_MODE_BIT,
498 EVENT_FILE_FL_TRIGGER_COND_BIT,
499 EVENT_FILE_FL_PID_FILTER_BIT,
500 EVENT_FILE_FL_WAS_ENABLED_BIT,
501 EVENT_FILE_FL_FREED_BIT,
502};
503
504extern struct trace_event_file *trace_get_event_file(const char *instance,
505 const char *system,
506 const char *event);
507extern void trace_put_event_file(struct trace_event_file *file);
508
509#define MAX_DYNEVENT_CMD_LEN (2048)
510
511enum dynevent_type {
512 DYNEVENT_TYPE_SYNTH = 1,
513 DYNEVENT_TYPE_KPROBE,
514 DYNEVENT_TYPE_NONE,
515};
516
517struct dynevent_cmd;
518
519typedef int (*dynevent_create_fn_t)(struct dynevent_cmd *cmd);
520
521struct dynevent_cmd {
522 struct seq_buf seq;
523 const char *event_name;
524 unsigned int n_fields;
525 enum dynevent_type type;
526 dynevent_create_fn_t run_command;
527 void *private_data;
528};
529
530extern int dynevent_create(struct dynevent_cmd *cmd);
531
532extern int synth_event_delete(const char *name);
533
534extern void synth_event_cmd_init(struct dynevent_cmd *cmd,
535 char *buf, int maxlen);
536
537extern int __synth_event_gen_cmd_start(struct dynevent_cmd *cmd,
538 const char *name,
539 struct module *mod, ...);
540
541#define synth_event_gen_cmd_start(cmd, name, mod, ...) \
542 __synth_event_gen_cmd_start(cmd, name, mod, ## __VA_ARGS__, NULL)
543
544struct synth_field_desc {
545 const char *type;
546 const char *name;
547};
548
549extern int synth_event_gen_cmd_array_start(struct dynevent_cmd *cmd,
550 const char *name,
551 struct module *mod,
552 struct synth_field_desc *fields,
553 unsigned int n_fields);
554extern int synth_event_create(const char *name,
555 struct synth_field_desc *fields,
556 unsigned int n_fields, struct module *mod);
557
558extern int synth_event_add_field(struct dynevent_cmd *cmd,
559 const char *type,
560 const char *name);
561extern int synth_event_add_field_str(struct dynevent_cmd *cmd,
562 const char *type_name);
563extern int synth_event_add_fields(struct dynevent_cmd *cmd,
564 struct synth_field_desc *fields,
565 unsigned int n_fields);
566
567#define synth_event_gen_cmd_end(cmd) \
568 dynevent_create(cmd)
569
570struct synth_event;
571
572struct synth_event_trace_state {
573 struct trace_event_buffer fbuffer;
574 struct synth_trace_event *entry;
575 struct trace_buffer *buffer;
576 struct synth_event *event;
577 unsigned int cur_field;
578 unsigned int n_u64;
579 bool disabled;
580 bool add_next;
581 bool add_name;
582};
583
584extern int synth_event_trace(struct trace_event_file *file,
585 unsigned int n_vals, ...);
586extern int synth_event_trace_array(struct trace_event_file *file, u64 *vals,
587 unsigned int n_vals);
588extern int synth_event_trace_start(struct trace_event_file *file,
589 struct synth_event_trace_state *trace_state);
590extern int synth_event_add_next_val(u64 val,
591 struct synth_event_trace_state *trace_state);
592extern int synth_event_add_val(const char *field_name, u64 val,
593 struct synth_event_trace_state *trace_state);
594extern int synth_event_trace_end(struct synth_event_trace_state *trace_state);
595
596extern int kprobe_event_delete(const char *name);
597
598extern void kprobe_event_cmd_init(struct dynevent_cmd *cmd,
599 char *buf, int maxlen);
600
601#define kprobe_event_gen_cmd_start(cmd, name, loc, ...) \
602 __kprobe_event_gen_cmd_start(cmd, false, name, loc, ## __VA_ARGS__, NULL)
603
604#define kretprobe_event_gen_cmd_start(cmd, name, loc, ...) \
605 __kprobe_event_gen_cmd_start(cmd, true, name, loc, ## __VA_ARGS__, NULL)
606
607extern int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd,
608 bool kretprobe,
609 const char *name,
610 const char *loc, ...);
611
612#define kprobe_event_add_fields(cmd, ...) \
613 __kprobe_event_add_fields(cmd, ## __VA_ARGS__, NULL)
614
615#define kprobe_event_add_field(cmd, field) \
616 __kprobe_event_add_fields(cmd, field, NULL)
617
618extern int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...);
619
620#define kprobe_event_gen_cmd_end(cmd) \
621 dynevent_create(cmd)
622
623#define kretprobe_event_gen_cmd_end(cmd) \
624 dynevent_create(cmd)
625
626/*
627 * Event file flags:
628 * ENABLED - The event is enabled
629 * RECORDED_CMD - The comms should be recorded at sched_switch
630 * RECORDED_TGID - The tgids should be recorded at sched_switch
631 * FILTERED - The event has a filter attached
632 * NO_SET_FILTER - Set when filter has error and is to be ignored
633 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
634 * SOFT_DISABLED - When set, do not trace the event (even though its
635 * tracepoint may be enabled)
636 * TRIGGER_MODE - When set, invoke the triggers associated with the event
637 * TRIGGER_COND - When set, one or more triggers has an associated filter
638 * PID_FILTER - When set, the event is filtered based on pid
639 * WAS_ENABLED - Set when enabled to know to clear trace on module removal
640 * FREED - File descriptor is freed, all fields should be considered invalid
641 */
642enum {
643 EVENT_FILE_FL_ENABLED = (1 << EVENT_FILE_FL_ENABLED_BIT),
644 EVENT_FILE_FL_RECORDED_CMD = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
645 EVENT_FILE_FL_RECORDED_TGID = (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
646 EVENT_FILE_FL_FILTERED = (1 << EVENT_FILE_FL_FILTERED_BIT),
647 EVENT_FILE_FL_NO_SET_FILTER = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
648 EVENT_FILE_FL_SOFT_MODE = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
649 EVENT_FILE_FL_SOFT_DISABLED = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
650 EVENT_FILE_FL_TRIGGER_MODE = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
651 EVENT_FILE_FL_TRIGGER_COND = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
652 EVENT_FILE_FL_PID_FILTER = (1 << EVENT_FILE_FL_PID_FILTER_BIT),
653 EVENT_FILE_FL_WAS_ENABLED = (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
654 EVENT_FILE_FL_FREED = (1 << EVENT_FILE_FL_FREED_BIT),
655};
656
657struct trace_event_file {
658 struct list_head list;
659 struct trace_event_call *event_call;
660 struct event_filter __rcu *filter;
661 struct eventfs_inode *ei;
662 struct trace_array *tr;
663 struct trace_subsystem_dir *system;
664 struct list_head triggers;
665
666 /*
667 * 32 bit flags:
668 * bit 0: enabled
669 * bit 1: enabled cmd record
670 * bit 2: enable/disable with the soft disable bit
671 * bit 3: soft disabled
672 * bit 4: trigger enabled
673 *
674 * Note: The bits must be set atomically to prevent races
675 * from other writers. Reads of flags do not need to be in
676 * sync as they occur in critical sections. But the way flags
677 * is currently used, these changes do not affect the code
678 * except that when a change is made, it may have a slight
679 * delay in propagating the changes to other CPUs due to
680 * caching and such. Which is mostly OK ;-)
681 */
682 unsigned long flags;
683 atomic_t ref; /* ref count for opened files */
684 atomic_t sm_ref; /* soft-mode reference counter */
685 atomic_t tm_ref; /* trigger-mode reference counter */
686};
687
688#define __TRACE_EVENT_FLAGS(name, value) \
689 static int __init trace_init_flags_##name(void) \
690 { \
691 event_##name.flags |= value; \
692 return 0; \
693 } \
694 early_initcall(trace_init_flags_##name);
695
696#define __TRACE_EVENT_PERF_PERM(name, expr...) \
697 static int perf_perm_##name(struct trace_event_call *tp_event, \
698 struct perf_event *p_event) \
699 { \
700 return ({ expr; }); \
701 } \
702 static int __init trace_init_perf_perm_##name(void) \
703 { \
704 event_##name.perf_perm = &perf_perm_##name; \
705 return 0; \
706 } \
707 early_initcall(trace_init_perf_perm_##name);
708
709#define PERF_MAX_TRACE_SIZE 8192
710
711#define MAX_FILTER_STR_VAL 256U /* Should handle KSYM_SYMBOL_LEN */
712
713enum event_trigger_type {
714 ETT_NONE = (0),
715 ETT_TRACE_ONOFF = (1 << 0),
716 ETT_SNAPSHOT = (1 << 1),
717 ETT_STACKTRACE = (1 << 2),
718 ETT_EVENT_ENABLE = (1 << 3),
719 ETT_EVENT_HIST = (1 << 4),
720 ETT_HIST_ENABLE = (1 << 5),
721 ETT_EVENT_EPROBE = (1 << 6),
722};
723
724extern int filter_match_preds(struct event_filter *filter, void *rec);
725
726extern enum event_trigger_type
727event_triggers_call(struct trace_event_file *file,
728 struct trace_buffer *buffer, void *rec,
729 struct ring_buffer_event *event);
730extern void
731event_triggers_post_call(struct trace_event_file *file,
732 enum event_trigger_type tt);
733
734bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
735
736bool __trace_trigger_soft_disabled(struct trace_event_file *file);
737
738/**
739 * trace_trigger_soft_disabled - do triggers and test if soft disabled
740 * @file: The file pointer of the event to test
741 *
742 * If any triggers without filters are attached to this event, they
743 * will be called here. If the event is soft disabled and has no
744 * triggers that require testing the fields, it will return true,
745 * otherwise false.
746 */
747static __always_inline bool
748trace_trigger_soft_disabled(struct trace_event_file *file)
749{
750 unsigned long eflags = file->flags;
751
752 if (likely(!(eflags & (EVENT_FILE_FL_TRIGGER_MODE |
753 EVENT_FILE_FL_SOFT_DISABLED |
754 EVENT_FILE_FL_PID_FILTER))))
755 return false;
756
757 if (likely(eflags & EVENT_FILE_FL_TRIGGER_COND))
758 return false;
759
760 return __trace_trigger_soft_disabled(file);
761}
762
763#ifdef CONFIG_BPF_EVENTS
764unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
765int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
766void perf_event_detach_bpf_prog(struct perf_event *event);
767int perf_event_query_prog_array(struct perf_event *event, void __user *info);
768int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
769int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
770struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
771void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
772int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
773 u32 *fd_type, const char **buf,
774 u64 *probe_offset, u64 *probe_addr,
775 unsigned long *missed);
776int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
777int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
778#else
779static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
780{
781 return 1;
782}
783
784static inline int
785perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie)
786{
787 return -EOPNOTSUPP;
788}
789
790static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
791
792static inline int
793perf_event_query_prog_array(struct perf_event *event, void __user *info)
794{
795 return -EOPNOTSUPP;
796}
797static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
798{
799 return -EOPNOTSUPP;
800}
801static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
802{
803 return -EOPNOTSUPP;
804}
805static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
806{
807 return NULL;
808}
809static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
810{
811}
812static inline int bpf_get_perf_event_info(const struct perf_event *event,
813 u32 *prog_id, u32 *fd_type,
814 const char **buf, u64 *probe_offset,
815 u64 *probe_addr, unsigned long *missed)
816{
817 return -EOPNOTSUPP;
818}
819static inline int
820bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
821{
822 return -EOPNOTSUPP;
823}
824static inline int
825bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
826{
827 return -EOPNOTSUPP;
828}
829#endif
830
831enum {
832 FILTER_OTHER = 0,
833 FILTER_STATIC_STRING,
834 FILTER_DYN_STRING,
835 FILTER_RDYN_STRING,
836 FILTER_PTR_STRING,
837 FILTER_TRACE_FN,
838 FILTER_CPUMASK,
839 FILTER_COMM,
840 FILTER_CPU,
841 FILTER_STACKTRACE,
842};
843
844extern int trace_event_raw_init(struct trace_event_call *call);
845extern int trace_define_field(struct trace_event_call *call, const char *type,
846 const char *name, int offset, int size,
847 int is_signed, int filter_type);
848extern int trace_add_event_call(struct trace_event_call *call);
849extern int trace_remove_event_call(struct trace_event_call *call);
850extern int trace_event_get_offsets(struct trace_event_call *call);
851
852int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
853int trace_set_clr_event(const char *system, const char *event, int set);
854int trace_array_set_clr_event(struct trace_array *tr, const char *system,
855 const char *event, bool enable);
856/*
857 * The double __builtin_constant_p is because gcc will give us an error
858 * if we try to allocate the static variable to fmt if it is not a
859 * constant. Even with the outer if statement optimizing out.
860 */
861#define event_trace_printk(ip, fmt, args...) \
862do { \
863 __trace_printk_check_format(fmt, ##args); \
864 tracing_record_cmdline(current); \
865 if (__builtin_constant_p(fmt)) { \
866 static const char *trace_printk_fmt \
867 __section("__trace_printk_fmt") = \
868 __builtin_constant_p(fmt) ? fmt : NULL; \
869 \
870 __trace_bprintk(ip, trace_printk_fmt, ##args); \
871 } else \
872 __trace_printk(ip, fmt, ##args); \
873} while (0)
874
875#ifdef CONFIG_PERF_EVENTS
876struct perf_event;
877
878DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
879DECLARE_PER_CPU(int, bpf_kprobe_override);
880
881extern int perf_trace_init(struct perf_event *event);
882extern void perf_trace_destroy(struct perf_event *event);
883extern int perf_trace_add(struct perf_event *event, int flags);
884extern void perf_trace_del(struct perf_event *event, int flags);
885#ifdef CONFIG_KPROBE_EVENTS
886extern int perf_kprobe_init(struct perf_event *event, bool is_retprobe);
887extern void perf_kprobe_destroy(struct perf_event *event);
888extern int bpf_get_kprobe_info(const struct perf_event *event,
889 u32 *fd_type, const char **symbol,
890 u64 *probe_offset, u64 *probe_addr,
891 unsigned long *missed,
892 bool perf_type_tracepoint);
893#endif
894#ifdef CONFIG_UPROBE_EVENTS
895extern int perf_uprobe_init(struct perf_event *event,
896 unsigned long ref_ctr_offset, bool is_retprobe);
897extern void perf_uprobe_destroy(struct perf_event *event);
898extern int bpf_get_uprobe_info(const struct perf_event *event,
899 u32 *fd_type, const char **filename,
900 u64 *probe_offset, u64 *probe_addr,
901 bool perf_type_tracepoint);
902#endif
903extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
904 char *filter_str);
905extern void ftrace_profile_free_filter(struct perf_event *event);
906void perf_trace_buf_update(void *record, u16 type);
907void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
908
909int perf_event_set_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
910void perf_event_free_bpf_prog(struct perf_event *event);
911
912void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
913void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
914void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
915 u64 arg3);
916void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
917 u64 arg3, u64 arg4);
918void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
919 u64 arg3, u64 arg4, u64 arg5);
920void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
921 u64 arg3, u64 arg4, u64 arg5, u64 arg6);
922void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
923 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
924void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
925 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
926 u64 arg8);
927void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
928 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
929 u64 arg8, u64 arg9);
930void bpf_trace_run10(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);
933void bpf_trace_run11(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);
936void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
937 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
938 u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
939void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
940 struct trace_event_call *call, u64 count,
941 struct pt_regs *regs, struct hlist_head *head,
942 struct task_struct *task);
943
944static inline void
945perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
946 u64 count, struct pt_regs *regs, void *head,
947 struct task_struct *task)
948{
949 perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
950}
951
952#endif
953
954#define TRACE_EVENT_STR_MAX 512
955
956/*
957 * gcc warns that you can not use a va_list in an inlined
958 * function. But lets me make it into a macro :-/
959 */
960#define __trace_event_vstr_len(fmt, va) \
961({ \
962 va_list __ap; \
963 int __ret; \
964 \
965 va_copy(__ap, *(va)); \
966 __ret = vsnprintf(NULL, 0, fmt, __ap) + 1; \
967 va_end(__ap); \
968 \
969 min(__ret, TRACE_EVENT_STR_MAX); \
970})
971
972#endif /* _LINUX_TRACE_EVENT_H */
973
974/*
975 * Note: we keep the TRACE_CUSTOM_EVENT outside the include file ifdef protection.
976 * This is due to the way trace custom events work. If a file includes two
977 * trace event headers under one "CREATE_CUSTOM_TRACE_EVENTS" the first include
978 * will override the TRACE_CUSTOM_EVENT and break the second include.
979 */
980
981#ifndef TRACE_CUSTOM_EVENT
982
983#define DECLARE_CUSTOM_EVENT_CLASS(name, proto, args, tstruct, assign, print)
984#define DEFINE_CUSTOM_EVENT(template, name, proto, args)
985#define TRACE_CUSTOM_EVENT(name, proto, args, struct, assign, print)
986
987#endif /* ifdef TRACE_CUSTOM_EVENT (see note above) */