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