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_symbols_seq_u64(struct trace_seq *p,
  27                                         unsigned long long val,
  28                                         const struct trace_print_flags_u64
  29                                                                  *symbol_array);
  30 #endif
  31
  32 const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
  33                                     unsigned int bitmask_size);
  34
  35 const char *trace_print_hex_seq(struct trace_seq *p,
  36                                 const unsigned char *buf, int len);
  37
  38 const char *trace_print_array_seq(struct trace_seq *p,
  39                                    const void *buf, int count,
  40                                    size_t el_size);
  41
  42 struct trace_iterator;
  43 struct trace_event;
  44
  45 int trace_raw_output_prep(struct trace_iterator *iter,
  46                           struct trace_event *event);
  47
  48 /*
  49  * The trace entry - the most basic unit of tracing. This is what
  50  * is printed in the end as a single line in the trace output, such as:
  51  *
  52  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
  53  */
  54 struct trace_entry {
  55         unsigned short          type;
  56         unsigned char           flags;
  57         unsigned char           preempt_count;
  58         int                     pid;
  59 };
  60
  61 #define TRACE_EVENT_TYPE_MAX                                            \
  62         ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
  63
  64 /*
  65  * Trace iterator - used by printout routines who present trace
  66  * results to users and which routines might sleep, etc:
  67  */
  68 struct trace_iterator {
  69         struct trace_array      *tr;
  70         struct tracer           *trace;
  71         struct trace_buffer     *trace_buffer;
  72         void                    *private;
  73         int                     cpu_file;
  74         struct mutex            mutex;
  75         struct ring_buffer_iter **buffer_iter;
  76         unsigned long           iter_flags;
  77
  78         /* trace_seq for __print_flags() and __print_symbolic() etc. */
  79         struct trace_seq        tmp_seq;
  80
  81         cpumask_var_t           started;
  82
  83         /* it's true when current open file is snapshot */
  84         bool                    snapshot;
  85
  86         /* The below is zeroed out in pipe_read */
  87         struct trace_seq        seq;
  88         struct trace_entry      *ent;
  89         unsigned long           lost_events;
  90         int                     leftover;
  91         int                     ent_size;
  92         int                     cpu;
  93         u64                     ts;
  94
  95         loff_t                  pos;
  96         long                    idx;
  97
  98         /* All new field here will be zeroed out in pipe_read */
  99 };
 100
 101 enum trace_iter_flags {
 102         TRACE_FILE_LAT_FMT      = 1,
 103         TRACE_FILE_ANNOTATE     = 2,
 104         TRACE_FILE_TIME_IN_NS   = 4,
 105 };
 106
 107
 108 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
 109                                       int flags, struct trace_event *event);
 110
 111 struct trace_event_functions {
 112         trace_print_func        trace;
 113         trace_print_func        raw;
 114         trace_print_func        hex;
 115         trace_print_func        binary;
 116 };
 117
 118 struct trace_event {
 119         struct hlist_node               node;
 120         struct list_head                list;
 121         int                             type;
 122         struct trace_event_functions    *funcs;
 123 };
 124
 125 extern int register_trace_event(struct trace_event *event);
 126 extern int unregister_trace_event(struct trace_event *event);
 127
 128 /* Return values for print_line callback */
 129 enum print_line_t {
 130         TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
 131         TRACE_TYPE_HANDLED      = 1,
 132         TRACE_TYPE_UNHANDLED    = 2,    /* Relay to other output functions */
 133         TRACE_TYPE_NO_CONSUME   = 3     /* Handled but ask to not consume */
 134 };
 135
 136 /*
 137  * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
 138  * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
 139  * simplifies those functions and keeps them in sync.
 140  */
 141 static inline enum print_line_t trace_handle_return(struct trace_seq *s)
 142 {
 143         return trace_seq_has_overflowed(s) ?
 144                 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
 145 }
 146
 147 void tracing_generic_entry_update(struct trace_entry *entry,
 148                                   unsigned long flags,
 149                                   int pc);
 150 struct trace_event_file;
 151
 152 struct ring_buffer_event *
 153 trace_event_buffer_lock_reserve(struct ring_buffer **current_buffer,
 154                                 struct trace_event_file *trace_file,
 155                                 int type, unsigned long len,
 156                                 unsigned long flags, int pc);
 157
 158 void tracing_record_cmdline(struct task_struct *tsk);
 159
 160 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
 161
 162 struct event_filter;
 163
 164 enum trace_reg {
 165         TRACE_REG_REGISTER,
 166         TRACE_REG_UNREGISTER,
 167 #ifdef CONFIG_PERF_EVENTS
 168         TRACE_REG_PERF_REGISTER,
 169         TRACE_REG_PERF_UNREGISTER,
 170         TRACE_REG_PERF_OPEN,
 171         TRACE_REG_PERF_CLOSE,
 172         TRACE_REG_PERF_ADD,
 173         TRACE_REG_PERF_DEL,
 174 #endif
 175 };
 176
 177 struct trace_event_call;
 178
 179 struct trace_event_class {
 180         const char              *system;
 181         void                    *probe;
 182 #ifdef CONFIG_PERF_EVENTS
 183         void                    *perf_probe;
 184 #endif
 185         int                     (*reg)(struct trace_event_call *event,
 186                                        enum trace_reg type, void *data);
 187         int                     (*define_fields)(struct trace_event_call *);
 188         struct list_head        *(*get_fields)(struct trace_event_call *);
 189         struct list_head        fields;
 190         int                     (*raw_init)(struct trace_event_call *);
 191 };
 192
 193 extern int trace_event_reg(struct trace_event_call *event,
 194                             enum trace_reg type, void *data);
 195
 196 struct trace_event_buffer {
 197         struct ring_buffer              *buffer;
 198         struct ring_buffer_event        *event;
 199         struct trace_event_file         *trace_file;
 200         void                            *entry;
 201         unsigned long                   flags;
 202         int                             pc;
 203 };
 204
 205 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 206                                   struct trace_event_file *trace_file,
 207                                   unsigned long len);
 208
 209 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
 210
 211 enum {
 212         TRACE_EVENT_FL_FILTERED_BIT,
 213         TRACE_EVENT_FL_CAP_ANY_BIT,
 214         TRACE_EVENT_FL_NO_SET_FILTER_BIT,
 215         TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
 216         TRACE_EVENT_FL_WAS_ENABLED_BIT,
 217         TRACE_EVENT_FL_TRACEPOINT_BIT,
 218         TRACE_EVENT_FL_KPROBE_BIT,
 219         TRACE_EVENT_FL_UPROBE_BIT,
 220         TRACE_EVENT_FL_MAP_BIT,
 221 };
 222
 223 /*
 224  * Event flags:
 225  *  FILTERED      - The event has a filter attached
 226  *  CAP_ANY       - Any user can enable for perf
 227  *  NO_SET_FILTER - Set when filter has error and is to be ignored
 228  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
 229  *  WAS_ENABLED   - Set and stays set when an event was ever enabled
 230  *                    (used for module unloading, if a module event is enabled,
 231  *                     it is best to clear the buffers that used it).
 232  *  TRACEPOINT    - Event is a tracepoint
 233  *  KPROBE        - Event is a kprobe
 234  *  UPROBE        - Event is a uprobe
 235  *  MAP           - Event maps to a tracepoint as an alias
 236  */
 237 enum {
 238         TRACE_EVENT_FL_FILTERED         = (1 << TRACE_EVENT_FL_FILTERED_BIT),
 239         TRACE_EVENT_FL_CAP_ANY          = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
 240         TRACE_EVENT_FL_NO_SET_FILTER    = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
 241         TRACE_EVENT_FL_IGNORE_ENABLE    = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
 242         TRACE_EVENT_FL_WAS_ENABLED      = (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT),
 243         TRACE_EVENT_FL_TRACEPOINT       = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
 244         TRACE_EVENT_FL_KPROBE           = (1 << TRACE_EVENT_FL_KPROBE_BIT),
 245         TRACE_EVENT_FL_UPROBE           = (1 << TRACE_EVENT_FL_UPROBE_BIT),
 246         TRACE_EVENT_FL_MAP              = (1 << TRACE_EVENT_FL_MAP_BIT),
 247 };
 248
 249 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
 250
 251 struct trace_event_map {
 252         struct tracepoint       *tp;
 253         char                    *name;
 254 };
 255
 256 struct trace_event_call {
 257         struct list_head        list;
 258         struct trace_event_class *class;
 259         union {
 260                 char                    *name;
 261                 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
 262                 struct tracepoint       *tp;
 263                 /* Set TRACE_EVENT_FL_MAP flag when using "map" instead */
 264                 struct trace_event_map  *map;
 265         };
 266         struct trace_event      event;
 267         char                    *print_fmt;
 268         struct event_filter     *filter;
 269         void                    *mod;
 270         void                    *data;
 271         /*
 272          *   bit 0:             filter_active
 273          *   bit 1:             allow trace by non root (cap any)
 274          *   bit 2:             failed to apply filter
 275          *   bit 3:             trace internal event (do not enable)
 276          *   bit 4:             Event was enabled by module
 277          *   bit 5:             use call filter rather than file filter
 278          *   bit 6:             Event is a tracepoint
 279          */
 280         int                     flags; /* static flags of different events */
 281
 282 #ifdef CONFIG_PERF_EVENTS
 283         int                             perf_refcount;
 284         struct hlist_head __percpu      *perf_events;
 285         struct bpf_prog                 *prog;
 286
 287         int     (*perf_perm)(struct trace_event_call *,
 288                              struct perf_event *);
 289 #endif
 290 };
 291
 292 static inline const char *
 293 trace_event_name(struct trace_event_call *call)
 294 {
 295         if (call->flags & TRACE_EVENT_FL_MAP)
 296                 return call->map->name;
 297         else if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
 298                 return call->tp ? call->tp->name : NULL;
 299         else
 300                 return call->name;
 301 }
 302
 303 struct trace_array;
 304 struct trace_subsystem_dir;
 305
 306 enum {
 307         EVENT_FILE_FL_ENABLED_BIT,
 308         EVENT_FILE_FL_RECORDED_CMD_BIT,
 309         EVENT_FILE_FL_FILTERED_BIT,
 310         EVENT_FILE_FL_NO_SET_FILTER_BIT,
 311         EVENT_FILE_FL_SOFT_MODE_BIT,
 312         EVENT_FILE_FL_SOFT_DISABLED_BIT,
 313         EVENT_FILE_FL_TRIGGER_MODE_BIT,
 314         EVENT_FILE_FL_TRIGGER_COND_BIT,
 315         EVENT_FILE_FL_PID_FILTER_BIT,
 316 };
 317
 318 /*
 319  * Event file flags:
 320  *  ENABLED       - The event is enabled
 321  *  RECORDED_CMD  - The comms should be recorded at sched_switch
 322  *  FILTERED      - The event has a filter attached
 323  *  NO_SET_FILTER - Set when filter has error and is to be ignored
 324  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
 325  *  SOFT_DISABLED - When set, do not trace the event (even though its
 326  *                   tracepoint may be enabled)
 327  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
 328  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
 329  *  PID_FILTER    - When set, the event is filtered based on pid
 330  */
 331 enum {
 332         EVENT_FILE_FL_ENABLED           = (1 << EVENT_FILE_FL_ENABLED_BIT),
 333         EVENT_FILE_FL_RECORDED_CMD      = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
 334         EVENT_FILE_FL_FILTERED          = (1 << EVENT_FILE_FL_FILTERED_BIT),
 335         EVENT_FILE_FL_NO_SET_FILTER     = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
 336         EVENT_FILE_FL_SOFT_MODE         = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
 337         EVENT_FILE_FL_SOFT_DISABLED     = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
 338         EVENT_FILE_FL_TRIGGER_MODE      = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
 339         EVENT_FILE_FL_TRIGGER_COND      = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
 340         EVENT_FILE_FL_PID_FILTER        = (1 << EVENT_FILE_FL_PID_FILTER_BIT),
 341 };
 342
 343 struct trace_event_file {
 344         struct list_head                list;
 345         struct trace_event_call         *event_call;
 346         struct event_filter             *filter;
 347         struct dentry                   *dir;
 348         struct trace_array              *tr;
 349         struct trace_subsystem_dir      *system;
 350         struct list_head                triggers;
 351
 352         /*
 353          * 32 bit flags:
 354          *   bit 0:             enabled
 355          *   bit 1:             enabled cmd record
 356          *   bit 2:             enable/disable with the soft disable bit
 357          *   bit 3:             soft disabled
 358          *   bit 4:             trigger enabled
 359          *
 360          * Note: The bits must be set atomically to prevent races
 361          * from other writers. Reads of flags do not need to be in
 362          * sync as they occur in critical sections. But the way flags
 363          * is currently used, these changes do not affect the code
 364          * except that when a change is made, it may have a slight
 365          * delay in propagating the changes to other CPUs due to
 366          * caching and such. Which is mostly OK ;-)
 367          */
 368         unsigned long           flags;
 369         atomic_t                sm_ref; /* soft-mode reference counter */
 370         atomic_t                tm_ref; /* trigger-mode reference counter */
 371 };
 372
 373 #define __TRACE_EVENT_FLAGS(name, value)                                \
 374         static int __init trace_init_flags_##name(void)                 \
 375         {                                                               \
 376                 event_##name.flags |= value;                            \
 377                 return 0;                                               \
 378         }                                                               \
 379         early_initcall(trace_init_flags_##name);
 380
 381 #define __TRACE_EVENT_PERF_PERM(name, expr...)                          \
 382         static int perf_perm_##name(struct trace_event_call *tp_event, \
 383                                     struct perf_event *p_event)         \
 384         {                                                               \
 385                 return ({ expr; });                                     \
 386         }                                                               \
 387         static int __init trace_init_perf_perm_##name(void)             \
 388         {                                                               \
 389                 event_##name.perf_perm = &perf_perm_##name;             \
 390                 return 0;                                               \
 391         }                                                               \
 392         early_initcall(trace_init_perf_perm_##name);
 393
 394 #define PERF_MAX_TRACE_SIZE     2048
 395
 396 #define MAX_FILTER_STR_VAL      256     /* Should handle KSYM_SYMBOL_LEN */
 397
 398 enum event_trigger_type {
 399         ETT_NONE                = (0),
 400         ETT_TRACE_ONOFF         = (1 << 0),
 401         ETT_SNAPSHOT            = (1 << 1),
 402         ETT_STACKTRACE          = (1 << 2),
 403         ETT_EVENT_ENABLE        = (1 << 3),
 404         ETT_EVENT_HIST          = (1 << 4),
 405         ETT_HIST_ENABLE         = (1 << 5),
 406 };
 407
 408 extern int filter_match_preds(struct event_filter *filter, void *rec);
 409
 410 extern enum event_trigger_type event_triggers_call(struct trace_event_file *file,
 411                                                    void *rec);
 412 extern void event_triggers_post_call(struct trace_event_file *file,
 413                                      enum event_trigger_type tt,
 414                                      void *rec);
 415
 416 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
 417
 418 /**
 419  * trace_trigger_soft_disabled - do triggers and test if soft disabled
 420  * @file: The file pointer of the event to test
 421  *
 422  * If any triggers without filters are attached to this event, they
 423  * will be called here. If the event is soft disabled and has no
 424  * triggers that require testing the fields, it will return true,
 425  * otherwise false.
 426  */
 427 static inline bool
 428 trace_trigger_soft_disabled(struct trace_event_file *file)
 429 {
 430         unsigned long eflags = file->flags;
 431
 432         if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
 433                 if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
 434                         event_triggers_call(file, NULL);
 435                 if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
 436                         return true;
 437                 if (eflags & EVENT_FILE_FL_PID_FILTER)
 438                         return trace_event_ignore_this_pid(file);
 439         }
 440         return false;
 441 }
 442
 443 #ifdef CONFIG_BPF_EVENTS
 444 unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx);
 445 #else
 446 static inline unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx)
 447 {
 448         return 1;
 449 }
 450 #endif
 451
 452 enum {
 453         FILTER_OTHER = 0,
 454         FILTER_STATIC_STRING,
 455         FILTER_DYN_STRING,
 456         FILTER_PTR_STRING,
 457         FILTER_TRACE_FN,
 458         FILTER_COMM,
 459         FILTER_CPU,
 460 };
 461
 462 extern int trace_event_raw_init(struct trace_event_call *call);
 463 extern int trace_define_field(struct trace_event_call *call, const char *type,
 464                               const char *name, int offset, int size,
 465                               int is_signed, int filter_type);
 466 extern int trace_add_event_call(struct trace_event_call *call);
 467 extern int trace_remove_event_call(struct trace_event_call *call);
 468 extern int trace_event_get_offsets(struct trace_event_call *call);
 469
 470 #define is_signed_type(type)    (((type)(-1)) < (type)1)
 471
 472 int trace_set_clr_event(const char *system, const char *event, int set);
 473
 474 /*
 475  * The double __builtin_constant_p is because gcc will give us an error
 476  * if we try to allocate the static variable to fmt if it is not a
 477  * constant. Even with the outer if statement optimizing out.
 478  */
 479 #define event_trace_printk(ip, fmt, args...)                            \
 480 do {                                                                    \
 481         __trace_printk_check_format(fmt, ##args);                       \
 482         tracing_record_cmdline(current);                                \
 483         if (__builtin_constant_p(fmt)) {                                \
 484                 static const char *trace_printk_fmt                     \
 485                   __attribute__((section("__trace_printk_fmt"))) =      \
 486                         __builtin_constant_p(fmt) ? fmt : NULL;         \
 487                                                                         \
 488                 __trace_bprintk(ip, trace_printk_fmt, ##args);          \
 489         } else                                                          \
 490                 __trace_printk(ip, fmt, ##args);                        \
 491 } while (0)
 492
 493 #ifdef CONFIG_PERF_EVENTS
 494 struct perf_event;
 495
 496 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
 497
 498 extern int  perf_trace_init(struct perf_event *event);
 499 extern void perf_trace_destroy(struct perf_event *event);
 500 extern int  perf_trace_add(struct perf_event *event, int flags);
 501 extern void perf_trace_del(struct perf_event *event, int flags);
 502 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
 503                                      char *filter_str);
 504 extern void ftrace_profile_free_filter(struct perf_event *event);
 505 void perf_trace_buf_update(void *record, u16 type);
 506 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
 507
 508 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
 509                                struct trace_event_call *call, u64 count,
 510                                struct pt_regs *regs, struct hlist_head *head,
 511                                struct task_struct *task);
 512
 513 static inline void
 514 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
 515                        u64 count, struct pt_regs *regs, void *head,
 516                        struct task_struct *task)
 517 {
 518         perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
 519 }
 520 #endif
 521
 522 #endif /* _LINUX_TRACE_EVENT_H */