include/linux/trace_events.h

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