1 /* SPDX-License-Identifier: GPL-2.0 */
2
3 #ifndef _LINUX_TRACE_EVENT_H
4 #define _LINUX_TRACE_EVENT_H
5
6 #include <linux/ring_buffer.h>
7 #include <linux/trace_seq.h>
8 #include <linux/percpu.h>
9 #include <linux/hardirq.h>
10 #include <linux/perf_event.h>
11 #include <linux/tracepoint.h>
12
13 struct trace_array;
14 struct array_buffer;
15 struct tracer;
16 struct dentry;
17 struct bpf_prog;
18 union bpf_attr;
19
20 const char *trace_print_flags_seq(struct trace_seq *p, const char *delim,
21 unsigned long flags,
22 const struct trace_print_flags *flag_array);
23
24 const char *trace_print_symbols_seq(struct trace_seq *p, unsigned long val,
25 const struct trace_print_flags *symbol_array);
26
27 #if BITS_PER_LONG == 32
28 const char *trace_print_flags_seq_u64(struct trace_seq *p, const char *delim,
29 unsigned long long flags,
30 const struct trace_print_flags_u64 *flag_array);
31
32 const char *trace_print_symbols_seq_u64(struct trace_seq *p,
33 unsigned long long val,
34 const struct trace_print_flags_u64
35 *symbol_array);
36 #endif
37
38 const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
39 unsigned int bitmask_size);
40
41 const char *trace_print_hex_seq(struct trace_seq *p,
42 const unsigned char *buf, int len,
43 bool concatenate);
44
45 const char *trace_print_array_seq(struct trace_seq *p,
46 const void *buf, int count,
47 size_t el_size);
48
49 const char *
50 trace_print_hex_dump_seq(struct trace_seq *p, const char *prefix_str,
51 int prefix_type, int rowsize, int groupsize,
52 const void *buf, size_t len, bool ascii);
53
54 struct trace_iterator;
55 struct trace_event;
56
57 int trace_raw_output_prep(struct trace_iterator *iter,
58 struct trace_event *event);
59 extern __printf(2, 3)
60 void trace_event_printf(struct trace_iterator *iter, const char *fmt, ...);
61
62 /*
63 * The trace entry - the most basic unit of tracing. This is what
64 * is printed in the end as a single line in the trace output, such as:
65 *
66 * bash-15816 [01] 235.197585: idle_cpu <- irq_enter
67 */
68 struct trace_entry {
69 unsigned short type;
70 unsigned char flags;
71 unsigned char preempt_count;
72 int pid;
73 };
74
75 #define TRACE_EVENT_TYPE_MAX \
76 ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
77
78 /*
79 * Trace iterator - used by printout routines who present trace
80 * results to users and which routines might sleep, etc:
81 */
82 struct trace_iterator {
83 struct trace_array *tr;
84 struct tracer *trace;
85 struct array_buffer *array_buffer;
86 void *private;
87 int cpu_file;
88 struct mutex mutex;
89 struct ring_buffer_iter **buffer_iter;
90 unsigned long iter_flags;
91 void *temp; /* temp holder */
92 unsigned int temp_size;
93 char *fmt; /* modified format holder */
94 unsigned int fmt_size;
95 long wait_index;
96
97 /* trace_seq for __print_flags() and __print_symbolic() etc. */
98 struct trace_seq tmp_seq;
99
100 cpumask_var_t started;
101
102 /* it's true when current open file is snapshot */
103 bool snapshot;
104
105 /* The below is zeroed out in pipe_read */
106 struct trace_seq seq;
107 struct trace_entry *ent;
108 unsigned long lost_events;
109 int leftover;
110 int ent_size;
111 int cpu;
112 u64 ts;
113
114 loff_t pos;
115 long idx;
116
117 /* All new field here will be zeroed out in pipe_read */
118 };
119
120 enum trace_iter_flags {
121 TRACE_FILE_LAT_FMT = 1,
122 TRACE_FILE_ANNOTATE = 2,
123 TRACE_FILE_TIME_IN_NS = 4,
124 };
125
126
127 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
128 int flags, struct trace_event *event);
129
130 struct trace_event_functions {
131 trace_print_func trace;
132 trace_print_func raw;
133 trace_print_func hex;
134 trace_print_func binary;
135 };
136
137 struct trace_event {
138 struct hlist_node node;
139 int type;
140 struct trace_event_functions *funcs;
141 };
142
143 extern int register_trace_event(struct trace_event *event);
144 extern int unregister_trace_event(struct trace_event *event);
145
146 /* Return values for print_line callback */
147 enum print_line_t {
148 TRACE_TYPE_PARTIAL_LINE = 0, /* Retry after flushing the seq */
149 TRACE_TYPE_HANDLED = 1,
150 TRACE_TYPE_UNHANDLED = 2, /* Relay to other output functions */
151 TRACE_TYPE_NO_CONSUME = 3 /* Handled but ask to not consume */
152 };
153
154 enum print_line_t trace_handle_return(struct trace_seq *s);
155
tracing_generic_entry_update(struct trace_entry * entry,unsigned short type,unsigned int trace_ctx)156 static inline void tracing_generic_entry_update(struct trace_entry *entry,
157 unsigned short type,
158 unsigned int trace_ctx)
159 {
160 entry->preempt_count = trace_ctx & 0xff;
161 entry->pid = current->pid;
162 entry->type = type;
163 entry->flags = trace_ctx >> 16;
164 }
165
166 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status);
167
168 enum trace_flag_type {
169 TRACE_FLAG_IRQS_OFF = 0x01,
170 TRACE_FLAG_IRQS_NOSUPPORT = 0x02,
171 TRACE_FLAG_NEED_RESCHED = 0x04,
172 TRACE_FLAG_HARDIRQ = 0x08,
173 TRACE_FLAG_SOFTIRQ = 0x10,
174 TRACE_FLAG_PREEMPT_RESCHED = 0x20,
175 TRACE_FLAG_NMI = 0x40,
176 TRACE_FLAG_BH_OFF = 0x80,
177 };
178
179 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
tracing_gen_ctx_flags(unsigned long irqflags)180 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
181 {
182 unsigned int irq_status = irqs_disabled_flags(irqflags) ?
183 TRACE_FLAG_IRQS_OFF : 0;
184 return tracing_gen_ctx_irq_test(irq_status);
185 }
tracing_gen_ctx(void)186 static inline unsigned int tracing_gen_ctx(void)
187 {
188 unsigned long irqflags;
189
190 local_save_flags(irqflags);
191 return tracing_gen_ctx_flags(irqflags);
192 }
193 #else
194
tracing_gen_ctx_flags(unsigned long irqflags)195 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
196 {
197 return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
198 }
tracing_gen_ctx(void)199 static inline unsigned int tracing_gen_ctx(void)
200 {
201 return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
202 }
203 #endif
204
tracing_gen_ctx_dec(void)205 static inline unsigned int tracing_gen_ctx_dec(void)
206 {
207 unsigned int trace_ctx;
208
209 trace_ctx = tracing_gen_ctx();
210 /*
211 * Subtract one from the preemption counter if preemption is enabled,
212 * see trace_event_buffer_reserve()for details.
213 */
214 if (IS_ENABLED(CONFIG_PREEMPTION))
215 trace_ctx--;
216 return trace_ctx;
217 }
218
219 struct trace_event_file;
220
221 struct ring_buffer_event *
222 trace_event_buffer_lock_reserve(struct trace_buffer **current_buffer,
223 struct trace_event_file *trace_file,
224 int type, unsigned long len,
225 unsigned int trace_ctx);
226
227 #define TRACE_RECORD_CMDLINE BIT(0)
228 #define TRACE_RECORD_TGID BIT(1)
229
230 void tracing_record_taskinfo(struct task_struct *task, int flags);
231 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
232 struct task_struct *next, int flags);
233
234 void tracing_record_cmdline(struct task_struct *task);
235 void tracing_record_tgid(struct task_struct *task);
236
237 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...)
238 __printf(3, 4);
239
240 struct event_filter;
241
242 enum trace_reg {
243 TRACE_REG_REGISTER,
244 TRACE_REG_UNREGISTER,
245 #ifdef CONFIG_PERF_EVENTS
246 TRACE_REG_PERF_REGISTER,
247 TRACE_REG_PERF_UNREGISTER,
248 TRACE_REG_PERF_OPEN,
249 TRACE_REG_PERF_CLOSE,
250 /*
251 * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
252 * custom action was taken and the default action is not to be
253 * performed.
254 */
255 TRACE_REG_PERF_ADD,
256 TRACE_REG_PERF_DEL,
257 #endif
258 };
259
260 struct trace_event_call;
261
262 #define TRACE_FUNCTION_TYPE ((const char *)~0UL)
263
264 struct trace_event_fields {
265 const char *type;
266 union {
267 struct {
268 const char *name;
269 const int size;
270 const int align;
271 const int is_signed;
272 const int filter_type;
273 const int len;
274 };
275 int (*define_fields)(struct trace_event_call *);
276 };
277 };
278
279 struct trace_event_class {
280 const char *system;
281 void *probe;
282 #ifdef CONFIG_PERF_EVENTS
283 void *perf_probe;
284 #endif
285 int (*reg)(struct trace_event_call *event,
286 enum trace_reg type, void *data);
287 struct trace_event_fields *fields_array;
288 struct list_head *(*get_fields)(struct trace_event_call *);
289 struct list_head fields;
290 int (*raw_init)(struct trace_event_call *);
291 };
292
293 extern int trace_event_reg(struct trace_event_call *event,
294 enum trace_reg type, void *data);
295
296 struct trace_event_buffer {
297 struct trace_buffer *buffer;
298 struct ring_buffer_event *event;
299 struct trace_event_file *trace_file;
300 void *entry;
301 unsigned int trace_ctx;
302 struct pt_regs *regs;
303 };
304
305 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
306 struct trace_event_file *trace_file,
307 unsigned long len);
308
309 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
310
311 enum {
312 TRACE_EVENT_FL_FILTERED_BIT,
313 TRACE_EVENT_FL_CAP_ANY_BIT,
314 TRACE_EVENT_FL_NO_SET_FILTER_BIT,
315 TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
316 TRACE_EVENT_FL_TRACEPOINT_BIT,
317 TRACE_EVENT_FL_DYNAMIC_BIT,
318 TRACE_EVENT_FL_KPROBE_BIT,
319 TRACE_EVENT_FL_UPROBE_BIT,
320 TRACE_EVENT_FL_EPROBE_BIT,
321 TRACE_EVENT_FL_CUSTOM_BIT,
322 };
323
324 /*
325 * Event flags:
326 * FILTERED - The event has a filter attached
327 * CAP_ANY - Any user can enable for perf
328 * NO_SET_FILTER - Set when filter has error and is to be ignored
329 * IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
330 * TRACEPOINT - Event is a tracepoint
331 * DYNAMIC - Event is a dynamic event (created at run time)
332 * KPROBE - Event is a kprobe
333 * UPROBE - Event is a uprobe
334 * EPROBE - Event is an event probe
335 * CUSTOM - Event is a custom event (to be attached to an exsiting tracepoint)
336 * This is set when the custom event has not been attached
337 * to a tracepoint yet, then it is cleared when it is.
338 */
339 enum {
340 TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
341 TRACE_EVENT_FL_CAP_ANY = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
342 TRACE_EVENT_FL_NO_SET_FILTER = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
343 TRACE_EVENT_FL_IGNORE_ENABLE = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
344 TRACE_EVENT_FL_TRACEPOINT = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
345 TRACE_EVENT_FL_DYNAMIC = (1 << TRACE_EVENT_FL_DYNAMIC_BIT),
346 TRACE_EVENT_FL_KPROBE = (1 << TRACE_EVENT_FL_KPROBE_BIT),
347 TRACE_EVENT_FL_UPROBE = (1 << TRACE_EVENT_FL_UPROBE_BIT),
348 TRACE_EVENT_FL_EPROBE = (1 << TRACE_EVENT_FL_EPROBE_BIT),
349 TRACE_EVENT_FL_CUSTOM = (1 << TRACE_EVENT_FL_CUSTOM_BIT),
350 };
351
352 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
353
354 struct trace_event_call {
355 struct list_head list;
356 struct trace_event_class *class;
357 union {
358 char *name;
359 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
360 struct tracepoint *tp;
361 };
362 struct trace_event event;
363 char *print_fmt;
364 struct event_filter *filter;
365 /*
366 * Static events can disappear with modules,
367 * where as dynamic ones need their own ref count.
368 */
369 union {
370 void *module;
371 atomic_t refcnt;
372 };
373 void *data;
374
375 /* See the TRACE_EVENT_FL_* flags above */
376 int flags; /* static flags of different events */
377
378 #ifdef CONFIG_PERF_EVENTS
379 int perf_refcount;
380 struct hlist_head __percpu *perf_events;
381 struct bpf_prog_array __rcu *prog_array;
382
383 int (*perf_perm)(struct trace_event_call *,
384 struct perf_event *);
385 #endif
386 };
387
388 #ifdef CONFIG_DYNAMIC_EVENTS
389 bool trace_event_dyn_try_get_ref(struct trace_event_call *call);
390 void trace_event_dyn_put_ref(struct trace_event_call *call);
391 bool trace_event_dyn_busy(struct trace_event_call *call);
392 #else
trace_event_dyn_try_get_ref(struct trace_event_call * call)393 static inline bool trace_event_dyn_try_get_ref(struct trace_event_call *call)
394 {
395 /* Without DYNAMIC_EVENTS configured, nothing should be calling this */
396 return false;
397 }
trace_event_dyn_put_ref(struct trace_event_call * call)398 static inline void trace_event_dyn_put_ref(struct trace_event_call *call)
399 {
400 }
trace_event_dyn_busy(struct trace_event_call * call)401 static inline bool trace_event_dyn_busy(struct trace_event_call *call)
402 {
403 /* Nothing should call this without DYNAIMIC_EVENTS configured. */
404 return true;
405 }
406 #endif
407
trace_event_try_get_ref(struct trace_event_call * call)408 static inline bool trace_event_try_get_ref(struct trace_event_call *call)
409 {
410 if (call->flags & TRACE_EVENT_FL_DYNAMIC)
411 return trace_event_dyn_try_get_ref(call);
412 else
413 return try_module_get(call->module);
414 }
415
trace_event_put_ref(struct trace_event_call * call)416 static inline void trace_event_put_ref(struct trace_event_call *call)
417 {
418 if (call->flags & TRACE_EVENT_FL_DYNAMIC)
419 trace_event_dyn_put_ref(call);
420 else
421 module_put(call->module);
422 }
423
424 #ifdef CONFIG_PERF_EVENTS
bpf_prog_array_valid(struct trace_event_call * call)425 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
426 {
427 /*
428 * This inline function checks whether call->prog_array
429 * is valid or not. The function is called in various places,
430 * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
431 *
432 * If this function returns true, and later call->prog_array
433 * becomes false inside rcu_read_lock/unlock region,
434 * we bail out then. If this function return false,
435 * there is a risk that we might miss a few events if the checking
436 * were delayed until inside rcu_read_lock/unlock region and
437 * call->prog_array happened to become non-NULL then.
438 *
439 * Here, READ_ONCE() is used instead of rcu_access_pointer().
440 * rcu_access_pointer() requires the actual definition of
441 * "struct bpf_prog_array" while READ_ONCE() only needs
442 * a declaration of the same type.
443 */
444 return !!READ_ONCE(call->prog_array);
445 }
446 #endif
447
448 static inline const char *
trace_event_name(struct trace_event_call * call)449 trace_event_name(struct trace_event_call *call)
450 {
451 if (call->flags & TRACE_EVENT_FL_CUSTOM)
452 return call->name;
453 else if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
454 return call->tp ? call->tp->name : NULL;
455 else
456 return call->name;
457 }
458
459 static inline struct list_head *
trace_get_fields(struct trace_event_call * event_call)460 trace_get_fields(struct trace_event_call *event_call)
461 {
462 if (!event_call->class->get_fields)
463 return &event_call->class->fields;
464 return event_call->class->get_fields(event_call);
465 }
466
467 struct trace_subsystem_dir;
468
469 enum {
470 EVENT_FILE_FL_ENABLED_BIT,
471 EVENT_FILE_FL_RECORDED_CMD_BIT,
472 EVENT_FILE_FL_RECORDED_TGID_BIT,
473 EVENT_FILE_FL_FILTERED_BIT,
474 EVENT_FILE_FL_NO_SET_FILTER_BIT,
475 EVENT_FILE_FL_SOFT_MODE_BIT,
476 EVENT_FILE_FL_SOFT_DISABLED_BIT,
477 EVENT_FILE_FL_TRIGGER_MODE_BIT,
478 EVENT_FILE_FL_TRIGGER_COND_BIT,
479 EVENT_FILE_FL_PID_FILTER_BIT,
480 EVENT_FILE_FL_WAS_ENABLED_BIT,
481 };
482
483 extern struct trace_event_file *trace_get_event_file(const char *instance,
484 const char *system,
485 const char *event);
486 extern void trace_put_event_file(struct trace_event_file *file);
487
488 #define MAX_DYNEVENT_CMD_LEN (2048)
489
490 enum dynevent_type {
491 DYNEVENT_TYPE_SYNTH = 1,
492 DYNEVENT_TYPE_KPROBE,
493 DYNEVENT_TYPE_NONE,
494 };
495
496 struct dynevent_cmd;
497
498 typedef int (*dynevent_create_fn_t)(struct dynevent_cmd *cmd);
499
500 struct dynevent_cmd {
501 struct seq_buf seq;
502 const char *event_name;
503 unsigned int n_fields;
504 enum dynevent_type type;
505 dynevent_create_fn_t run_command;
506 void *private_data;
507 };
508
509 extern int dynevent_create(struct dynevent_cmd *cmd);
510
511 extern int synth_event_delete(const char *name);
512
513 extern void synth_event_cmd_init(struct dynevent_cmd *cmd,
514 char *buf, int maxlen);
515
516 extern int __synth_event_gen_cmd_start(struct dynevent_cmd *cmd,
517 const char *name,
518 struct module *mod, ...);
519
520 #define synth_event_gen_cmd_start(cmd, name, mod, ...) \
521 __synth_event_gen_cmd_start(cmd, name, mod, ## __VA_ARGS__, NULL)
522
523 struct synth_field_desc {
524 const char *type;
525 const char *name;
526 };
527
528 extern int synth_event_gen_cmd_array_start(struct dynevent_cmd *cmd,
529 const char *name,
530 struct module *mod,
531 struct synth_field_desc *fields,
532 unsigned int n_fields);
533 extern int synth_event_create(const char *name,
534 struct synth_field_desc *fields,
535 unsigned int n_fields, struct module *mod);
536
537 extern int synth_event_add_field(struct dynevent_cmd *cmd,
538 const char *type,
539 const char *name);
540 extern int synth_event_add_field_str(struct dynevent_cmd *cmd,
541 const char *type_name);
542 extern int synth_event_add_fields(struct dynevent_cmd *cmd,
543 struct synth_field_desc *fields,
544 unsigned int n_fields);
545
546 #define synth_event_gen_cmd_end(cmd) \
547 dynevent_create(cmd)
548
549 struct synth_event;
550
551 struct synth_event_trace_state {
552 struct trace_event_buffer fbuffer;
553 struct synth_trace_event *entry;
554 struct trace_buffer *buffer;
555 struct synth_event *event;
556 unsigned int cur_field;
557 unsigned int n_u64;
558 bool disabled;
559 bool add_next;
560 bool add_name;
561 };
562
563 extern int synth_event_trace(struct trace_event_file *file,
564 unsigned int n_vals, ...);
565 extern int synth_event_trace_array(struct trace_event_file *file, u64 *vals,
566 unsigned int n_vals);
567 extern int synth_event_trace_start(struct trace_event_file *file,
568 struct synth_event_trace_state *trace_state);
569 extern int synth_event_add_next_val(u64 val,
570 struct synth_event_trace_state *trace_state);
571 extern int synth_event_add_val(const char *field_name, u64 val,
572 struct synth_event_trace_state *trace_state);
573 extern int synth_event_trace_end(struct synth_event_trace_state *trace_state);
574
575 extern int kprobe_event_delete(const char *name);
576
577 extern void kprobe_event_cmd_init(struct dynevent_cmd *cmd,
578 char *buf, int maxlen);
579
580 #define kprobe_event_gen_cmd_start(cmd, name, loc, ...) \
581 __kprobe_event_gen_cmd_start(cmd, false, name, loc, ## __VA_ARGS__, NULL)
582
583 #define kretprobe_event_gen_cmd_start(cmd, name, loc, ...) \
584 __kprobe_event_gen_cmd_start(cmd, true, name, loc, ## __VA_ARGS__, NULL)
585
586 extern int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd,
587 bool kretprobe,
588 const char *name,
589 const char *loc, ...);
590
591 #define kprobe_event_add_fields(cmd, ...) \
592 __kprobe_event_add_fields(cmd, ## __VA_ARGS__, NULL)
593
594 #define kprobe_event_add_field(cmd, field) \
595 __kprobe_event_add_fields(cmd, field, NULL)
596
597 extern int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...);
598
599 #define kprobe_event_gen_cmd_end(cmd) \
600 dynevent_create(cmd)
601
602 #define kretprobe_event_gen_cmd_end(cmd) \
603 dynevent_create(cmd)
604
605 /*
606 * Event file flags:
607 * ENABLED - The event is enabled
608 * RECORDED_CMD - The comms should be recorded at sched_switch
609 * RECORDED_TGID - The tgids should be recorded at sched_switch
610 * FILTERED - The event has a filter attached
611 * NO_SET_FILTER - Set when filter has error and is to be ignored
612 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
613 * SOFT_DISABLED - When set, do not trace the event (even though its
614 * tracepoint may be enabled)
615 * TRIGGER_MODE - When set, invoke the triggers associated with the event
616 * TRIGGER_COND - When set, one or more triggers has an associated filter
617 * PID_FILTER - When set, the event is filtered based on pid
618 * WAS_ENABLED - Set when enabled to know to clear trace on module removal
619 */
620 enum {
621 EVENT_FILE_FL_ENABLED = (1 << EVENT_FILE_FL_ENABLED_BIT),
622 EVENT_FILE_FL_RECORDED_CMD = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
623 EVENT_FILE_FL_RECORDED_TGID = (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
624 EVENT_FILE_FL_FILTERED = (1 << EVENT_FILE_FL_FILTERED_BIT),
625 EVENT_FILE_FL_NO_SET_FILTER = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
626 EVENT_FILE_FL_SOFT_MODE = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
627 EVENT_FILE_FL_SOFT_DISABLED = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
628 EVENT_FILE_FL_TRIGGER_MODE = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
629 EVENT_FILE_FL_TRIGGER_COND = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
630 EVENT_FILE_FL_PID_FILTER = (1 << EVENT_FILE_FL_PID_FILTER_BIT),
631 EVENT_FILE_FL_WAS_ENABLED = (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
632 };
633
634 struct trace_event_file {
635 struct list_head list;
636 struct trace_event_call *event_call;
637 struct event_filter __rcu *filter;
638 struct dentry *dir;
639 struct trace_array *tr;
640 struct trace_subsystem_dir *system;
641 struct list_head triggers;
642
643 /*
644 * 32 bit flags:
645 * bit 0: enabled
646 * bit 1: enabled cmd record
647 * bit 2: enable/disable with the soft disable bit
648 * bit 3: soft disabled
649 * bit 4: trigger enabled
650 *
651 * Note: The bits must be set atomically to prevent races
652 * from other writers. Reads of flags do not need to be in
653 * sync as they occur in critical sections. But the way flags
654 * is currently used, these changes do not affect the code
655 * except that when a change is made, it may have a slight
656 * delay in propagating the changes to other CPUs due to
657 * caching and such. Which is mostly OK ;-)
658 */
659 unsigned long flags;
660 atomic_t sm_ref; /* soft-mode reference counter */
661 atomic_t tm_ref; /* trigger-mode reference counter */
662 };
663
664 #define __TRACE_EVENT_FLAGS(name, value) \
665 static int __init trace_init_flags_##name(void) \
666 { \
667 event_##name.flags |= value; \
668 return 0; \
669 } \
670 early_initcall(trace_init_flags_##name);
671
672 #define __TRACE_EVENT_PERF_PERM(name, expr...) \
673 static int perf_perm_##name(struct trace_event_call *tp_event, \
674 struct perf_event *p_event) \
675 { \
676 return ({ expr; }); \
677 } \
678 static int __init trace_init_perf_perm_##name(void) \
679 { \
680 event_##name.perf_perm = &perf_perm_##name; \
681 return 0; \
682 } \
683 early_initcall(trace_init_perf_perm_##name);
684
685 #define PERF_MAX_TRACE_SIZE 8192
686
687 #define MAX_FILTER_STR_VAL 256U /* Should handle KSYM_SYMBOL_LEN */
688
689 enum event_trigger_type {
690 ETT_NONE = (0),
691 ETT_TRACE_ONOFF = (1 << 0),
692 ETT_SNAPSHOT = (1 << 1),
693 ETT_STACKTRACE = (1 << 2),
694 ETT_EVENT_ENABLE = (1 << 3),
695 ETT_EVENT_HIST = (1 << 4),
696 ETT_HIST_ENABLE = (1 << 5),
697 ETT_EVENT_EPROBE = (1 << 6),
698 };
699
700 extern int filter_match_preds(struct event_filter *filter, void *rec);
701
702 extern enum event_trigger_type
703 event_triggers_call(struct trace_event_file *file,
704 struct trace_buffer *buffer, void *rec,
705 struct ring_buffer_event *event);
706 extern void
707 event_triggers_post_call(struct trace_event_file *file,
708 enum event_trigger_type tt);
709
710 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
711
712 bool __trace_trigger_soft_disabled(struct trace_event_file *file);
713
714 /**
715 * trace_trigger_soft_disabled - do triggers and test if soft disabled
716 * @file: The file pointer of the event to test
717 *
718 * If any triggers without filters are attached to this event, they
719 * will be called here. If the event is soft disabled and has no
720 * triggers that require testing the fields, it will return true,
721 * otherwise false.
722 */
723 static __always_inline bool
trace_trigger_soft_disabled(struct trace_event_file * file)724 trace_trigger_soft_disabled(struct trace_event_file *file)
725 {
726 unsigned long eflags = file->flags;
727
728 if (likely(!(eflags & (EVENT_FILE_FL_TRIGGER_MODE |
729 EVENT_FILE_FL_SOFT_DISABLED |
730 EVENT_FILE_FL_PID_FILTER))))
731 return false;
732
733 if (likely(eflags & EVENT_FILE_FL_TRIGGER_COND))
734 return false;
735
736 return __trace_trigger_soft_disabled(file);
737 }
738
739 #ifdef CONFIG_BPF_EVENTS
740 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
741 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
742 void perf_event_detach_bpf_prog(struct perf_event *event);
743 int perf_event_query_prog_array(struct perf_event *event, void __user *info);
744 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
745 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
746 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
747 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
748 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
749 u32 *fd_type, const char **buf,
750 u64 *probe_offset, u64 *probe_addr);
751 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
752 #else
trace_call_bpf(struct trace_event_call * call,void * ctx)753 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
754 {
755 return 1;
756 }
757
758 static inline int
perf_event_attach_bpf_prog(struct perf_event * event,struct bpf_prog * prog,u64 bpf_cookie)759 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie)
760 {
761 return -EOPNOTSUPP;
762 }
763
perf_event_detach_bpf_prog(struct perf_event * event)764 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
765
766 static inline int
perf_event_query_prog_array(struct perf_event * event,void __user * info)767 perf_event_query_prog_array(struct perf_event *event, void __user *info)
768 {
769 return -EOPNOTSUPP;
770 }
bpf_probe_register(struct bpf_raw_event_map * btp,struct bpf_prog * p)771 static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
772 {
773 return -EOPNOTSUPP;
774 }
bpf_probe_unregister(struct bpf_raw_event_map * btp,struct bpf_prog * p)775 static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
776 {
777 return -EOPNOTSUPP;
778 }
bpf_get_raw_tracepoint(const char * name)779 static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
780 {
781 return NULL;
782 }
bpf_put_raw_tracepoint(struct bpf_raw_event_map * btp)783 static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
784 {
785 }
bpf_get_perf_event_info(const struct perf_event * event,u32 * prog_id,u32 * fd_type,const char ** buf,u64 * probe_offset,u64 * probe_addr)786 static inline int bpf_get_perf_event_info(const struct perf_event *event,
787 u32 *prog_id, u32 *fd_type,
788 const char **buf, u64 *probe_offset,
789 u64 *probe_addr)
790 {
791 return -EOPNOTSUPP;
792 }
793 static inline int
bpf_kprobe_multi_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)794 bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
795 {
796 return -EOPNOTSUPP;
797 }
798 #endif
799
800 enum {
801 FILTER_OTHER = 0,
802 FILTER_STATIC_STRING,
803 FILTER_DYN_STRING,
804 FILTER_RDYN_STRING,
805 FILTER_PTR_STRING,
806 FILTER_TRACE_FN,
807 FILTER_COMM,
808 FILTER_CPU,
809 };
810
811 extern int trace_event_raw_init(struct trace_event_call *call);
812 extern int trace_define_field(struct trace_event_call *call, const char *type,
813 const char *name, int offset, int size,
814 int is_signed, int filter_type);
815 extern int trace_add_event_call(struct trace_event_call *call);
816 extern int trace_remove_event_call(struct trace_event_call *call);
817 extern int trace_event_get_offsets(struct trace_event_call *call);
818
819 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
820 int trace_set_clr_event(const char *system, const char *event, int set);
821 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
822 const char *event, bool enable);
823 /*
824 * The double __builtin_constant_p is because gcc will give us an error
825 * if we try to allocate the static variable to fmt if it is not a
826 * constant. Even with the outer if statement optimizing out.
827 */
828 #define event_trace_printk(ip, fmt, args...) \
829 do { \
830 __trace_printk_check_format(fmt, ##args); \
831 tracing_record_cmdline(current); \
832 if (__builtin_constant_p(fmt)) { \
833 static const char *trace_printk_fmt \
834 __section("__trace_printk_fmt") = \
835 __builtin_constant_p(fmt) ? fmt : NULL; \
836 \
837 __trace_bprintk(ip, trace_printk_fmt, ##args); \
838 } else \
839 __trace_printk(ip, fmt, ##args); \
840 } while (0)
841
842 #ifdef CONFIG_PERF_EVENTS
843 struct perf_event;
844
845 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
846 DECLARE_PER_CPU(int, bpf_kprobe_override);
847
848 extern int perf_trace_init(struct perf_event *event);
849 extern void perf_trace_destroy(struct perf_event *event);
850 extern int perf_trace_add(struct perf_event *event, int flags);
851 extern void perf_trace_del(struct perf_event *event, int flags);
852 #ifdef CONFIG_KPROBE_EVENTS
853 extern int perf_kprobe_init(struct perf_event *event, bool is_retprobe);
854 extern void perf_kprobe_destroy(struct perf_event *event);
855 extern int bpf_get_kprobe_info(const struct perf_event *event,
856 u32 *fd_type, const char **symbol,
857 u64 *probe_offset, u64 *probe_addr,
858 bool perf_type_tracepoint);
859 #endif
860 #ifdef CONFIG_UPROBE_EVENTS
861 extern int perf_uprobe_init(struct perf_event *event,
862 unsigned long ref_ctr_offset, bool is_retprobe);
863 extern void perf_uprobe_destroy(struct perf_event *event);
864 extern int bpf_get_uprobe_info(const struct perf_event *event,
865 u32 *fd_type, const char **filename,
866 u64 *probe_offset, bool perf_type_tracepoint);
867 #endif
868 extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
869 char *filter_str);
870 extern void ftrace_profile_free_filter(struct perf_event *event);
871 void perf_trace_buf_update(void *record, u16 type);
872 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
873
874 int perf_event_set_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
875 void perf_event_free_bpf_prog(struct perf_event *event);
876
877 void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
878 void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
879 void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
880 u64 arg3);
881 void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
882 u64 arg3, u64 arg4);
883 void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
884 u64 arg3, u64 arg4, u64 arg5);
885 void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
886 u64 arg3, u64 arg4, u64 arg5, u64 arg6);
887 void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
888 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
889 void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
890 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
891 u64 arg8);
892 void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
893 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
894 u64 arg8, u64 arg9);
895 void bpf_trace_run10(struct bpf_prog *prog, u64 arg1, u64 arg2,
896 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
897 u64 arg8, u64 arg9, u64 arg10);
898 void bpf_trace_run11(struct bpf_prog *prog, u64 arg1, u64 arg2,
899 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
900 u64 arg8, u64 arg9, u64 arg10, u64 arg11);
901 void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
902 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
903 u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
904 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
905 struct trace_event_call *call, u64 count,
906 struct pt_regs *regs, struct hlist_head *head,
907 struct task_struct *task);
908
909 static inline void
perf_trace_buf_submit(void * raw_data,int size,int rctx,u16 type,u64 count,struct pt_regs * regs,void * head,struct task_struct * task)910 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
911 u64 count, struct pt_regs *regs, void *head,
912 struct task_struct *task)
913 {
914 perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
915 }
916
917 #endif
918
919 #define TRACE_EVENT_STR_MAX 512
920
921 /*
922 * gcc warns that you can not use a va_list in an inlined
923 * function. But lets me make it into a macro :-/
924 */
925 #define __trace_event_vstr_len(fmt, va) \
926 ({ \
927 va_list __ap; \
928 int __ret; \
929 \
930 va_copy(__ap, *(va)); \
931 __ret = vsnprintf(NULL, 0, fmt, __ap) + 1; \
932 va_end(__ap); \
933 \
934 min(__ret, TRACE_EVENT_STR_MAX); \
935 })
936
937 #endif /* _LINUX_TRACE_EVENT_H */
938
939 /*
940 * Note: we keep the TRACE_CUSTOM_EVENT outside the include file ifdef protection.
941 * This is due to the way trace custom events work. If a file includes two
942 * trace event headers under one "CREATE_CUSTOM_TRACE_EVENTS" the first include
943 * will override the TRACE_CUSTOM_EVENT and break the second include.
944 */
945
946 #ifndef TRACE_CUSTOM_EVENT
947
948 #define DECLARE_CUSTOM_EVENT_CLASS(name, proto, args, tstruct, assign, print)
949 #define DEFINE_CUSTOM_EVENT(template, name, proto, args)
950 #define TRACE_CUSTOM_EVENT(name, proto, args, struct, assign, print)
951
952 #endif /* ifdef TRACE_CUSTOM_EVENT (see note above) */
953