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