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 };
 221
 222 /*
 223  * Event flags:
 224  *  FILTERED      - The event has a filter attached
 225  *  CAP_ANY       - Any user can enable for perf
 226  *  NO_SET_FILTER - Set when filter has error and is to be ignored
 227  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
 228  *  WAS_ENABLED   - Set and stays set when an event was ever enabled
 229  *                    (used for module unloading, if a module event is enabled,
 230  *                     it is best to clear the buffers that used it).
 231  *  TRACEPOINT    - Event is a tracepoint
 232  *  KPROBE        - Event is a kprobe
 233  *  UPROBE        - Event is a uprobe
 234  */
 235 enum {
 236         TRACE_EVENT_FL_FILTERED         = (1 << TRACE_EVENT_FL_FILTERED_BIT),
 237         TRACE_EVENT_FL_CAP_ANY          = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
 238         TRACE_EVENT_FL_NO_SET_FILTER    = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
 239         TRACE_EVENT_FL_IGNORE_ENABLE    = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
 240         TRACE_EVENT_FL_WAS_ENABLED      = (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT),
 241         TRACE_EVENT_FL_TRACEPOINT       = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
 242         TRACE_EVENT_FL_KPROBE           = (1 << TRACE_EVENT_FL_KPROBE_BIT),
 243         TRACE_EVENT_FL_UPROBE           = (1 << TRACE_EVENT_FL_UPROBE_BIT),
 244 };
 245
 246 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
 247
 248 struct trace_event_call {
 249         struct list_head        list;
 250         struct trace_event_class *class;
 251         union {
 252                 char                    *name;
 253                 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
 254                 struct tracepoint       *tp;
 255         };
 256         struct trace_event      event;
 257         char                    *print_fmt;
 258         struct event_filter     *filter;
 259         void                    *mod;
 260         void                    *data;
 261         /*
 262          *   bit 0:             filter_active
 263          *   bit 1:             allow trace by non root (cap any)
 264          *   bit 2:             failed to apply filter
 265          *   bit 3:             trace internal event (do not enable)
 266          *   bit 4:             Event was enabled by module
 267          *   bit 5:             use call filter rather than file filter
 268          *   bit 6:             Event is a tracepoint
 269          */
 270         int                     flags; /* static flags of different events */
 271
 272 #ifdef CONFIG_PERF_EVENTS
 273         int                             perf_refcount;
 274         struct hlist_head __percpu      *perf_events;
 275         struct bpf_prog                 *prog;
 276
 277         int     (*perf_perm)(struct trace_event_call *,
 278                              struct perf_event *);
 279 #endif
 280 };
 281
 282 static inline const char *
 283 trace_event_name(struct trace_event_call *call)
 284 {
 285         if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
 286                 return call->tp ? call->tp->name : NULL;
 287         else
 288                 return call->name;
 289 }
 290
 291 struct trace_array;
 292 struct trace_subsystem_dir;
 293
 294 enum {
 295         EVENT_FILE_FL_ENABLED_BIT,
 296         EVENT_FILE_FL_RECORDED_CMD_BIT,
 297         EVENT_FILE_FL_FILTERED_BIT,
 298         EVENT_FILE_FL_NO_SET_FILTER_BIT,
 299         EVENT_FILE_FL_SOFT_MODE_BIT,
 300         EVENT_FILE_FL_SOFT_DISABLED_BIT,
 301         EVENT_FILE_FL_TRIGGER_MODE_BIT,
 302         EVENT_FILE_FL_TRIGGER_COND_BIT,
 303         EVENT_FILE_FL_PID_FILTER_BIT,
 304 };
 305
 306 /*
 307  * Event file flags:
 308  *  ENABLED       - The event is enabled
 309  *  RECORDED_CMD  - The comms should be recorded at sched_switch
 310  *  FILTERED      - The event has a filter attached
 311  *  NO_SET_FILTER - Set when filter has error and is to be ignored
 312  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
 313  *  SOFT_DISABLED - When set, do not trace the event (even though its
 314  *                   tracepoint may be enabled)
 315  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
 316  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
 317  *  PID_FILTER    - When set, the event is filtered based on pid
 318  */
 319 enum {
 320         EVENT_FILE_FL_ENABLED           = (1 << EVENT_FILE_FL_ENABLED_BIT),
 321         EVENT_FILE_FL_RECORDED_CMD      = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
 322         EVENT_FILE_FL_FILTERED          = (1 << EVENT_FILE_FL_FILTERED_BIT),
 323         EVENT_FILE_FL_NO_SET_FILTER     = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
 324         EVENT_FILE_FL_SOFT_MODE         = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
 325         EVENT_FILE_FL_SOFT_DISABLED     = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
 326         EVENT_FILE_FL_TRIGGER_MODE      = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
 327         EVENT_FILE_FL_TRIGGER_COND      = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
 328         EVENT_FILE_FL_PID_FILTER        = (1 << EVENT_FILE_FL_PID_FILTER_BIT),
 329 };
 330
 331 struct trace_event_file {
 332         struct list_head                list;
 333         struct trace_event_call         *event_call;
 334         struct event_filter             *filter;
 335         struct dentry                   *dir;
 336         struct trace_array              *tr;
 337         struct trace_subsystem_dir      *system;
 338         struct list_head                triggers;
 339
 340         /*
 341          * 32 bit flags:
 342          *   bit 0:             enabled
 343          *   bit 1:             enabled cmd record
 344          *   bit 2:             enable/disable with the soft disable bit
 345          *   bit 3:             soft disabled
 346          *   bit 4:             trigger enabled
 347          *
 348          * Note: The bits must be set atomically to prevent races
 349          * from other writers. Reads of flags do not need to be in
 350          * sync as they occur in critical sections. But the way flags
 351          * is currently used, these changes do not affect the code
 352          * except that when a change is made, it may have a slight
 353          * delay in propagating the changes to other CPUs due to
 354          * caching and such. Which is mostly OK ;-)
 355          */
 356         unsigned long           flags;
 357         atomic_t                sm_ref; /* soft-mode reference counter */
 358         atomic_t                tm_ref; /* trigger-mode reference counter */
 359 };
 360
 361 #define __TRACE_EVENT_FLAGS(name, value)                                \
 362         static int __init trace_init_flags_##name(void)                 \
 363         {                                                               \
 364                 event_##name.flags |= value;                            \
 365                 return 0;                                               \
 366         }                                                               \
 367         early_initcall(trace_init_flags_##name);
 368
 369 #define __TRACE_EVENT_PERF_PERM(name, expr...)                          \
 370         static int perf_perm_##name(struct trace_event_call *tp_event, \
 371                                     struct perf_event *p_event)         \
 372         {                                                               \
 373                 return ({ expr; });                                     \
 374         }                                                               \
 375         static int __init trace_init_perf_perm_##name(void)             \
 376         {                                                               \
 377                 event_##name.perf_perm = &perf_perm_##name;             \
 378                 return 0;                                               \
 379         }                                                               \
 380         early_initcall(trace_init_perf_perm_##name);
 381
 382 #define PERF_MAX_TRACE_SIZE     2048
 383
 384 #define MAX_FILTER_STR_VAL      256     /* Should handle KSYM_SYMBOL_LEN */
 385
 386 enum event_trigger_type {
 387         ETT_NONE                = (0),
 388         ETT_TRACE_ONOFF         = (1 << 0),
 389         ETT_SNAPSHOT            = (1 << 1),
 390         ETT_STACKTRACE          = (1 << 2),
 391         ETT_EVENT_ENABLE        = (1 << 3),
 392         ETT_EVENT_HIST          = (1 << 4),
 393         ETT_HIST_ENABLE         = (1 << 5),
 394 };
 395
 396 extern int filter_match_preds(struct event_filter *filter, void *rec);
 397
 398 extern enum event_trigger_type event_triggers_call(struct trace_event_file *file,
 399                                                    void *rec);
 400 extern void event_triggers_post_call(struct trace_event_file *file,
 401                                      enum event_trigger_type tt,
 402                                      void *rec);
 403
 404 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
 405
 406 /**
 407  * trace_trigger_soft_disabled - do triggers and test if soft disabled
 408  * @file: The file pointer of the event to test
 409  *
 410  * If any triggers without filters are attached to this event, they
 411  * will be called here. If the event is soft disabled and has no
 412  * triggers that require testing the fields, it will return true,
 413  * otherwise false.
 414  */
 415 static inline bool
 416 trace_trigger_soft_disabled(struct trace_event_file *file)
 417 {
 418         unsigned long eflags = file->flags;
 419
 420         if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
 421                 if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
 422                         event_triggers_call(file, NULL);
 423                 if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
 424                         return true;
 425                 if (eflags & EVENT_FILE_FL_PID_FILTER)
 426                         return trace_event_ignore_this_pid(file);
 427         }
 428         return false;
 429 }
 430
 431 #ifdef CONFIG_BPF_EVENTS
 432 unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx);
 433 #else
 434 static inline unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx)
 435 {
 436         return 1;
 437 }
 438 #endif
 439
 440 enum {
 441         FILTER_OTHER = 0,
 442         FILTER_STATIC_STRING,
 443         FILTER_DYN_STRING,
 444         FILTER_PTR_STRING,
 445         FILTER_TRACE_FN,
 446         FILTER_COMM,
 447         FILTER_CPU,
 448 };
 449
 450 extern int trace_event_raw_init(struct trace_event_call *call);
 451 extern int trace_define_field(struct trace_event_call *call, const char *type,
 452                               const char *name, int offset, int size,
 453                               int is_signed, int filter_type);
 454 extern int trace_add_event_call(struct trace_event_call *call);
 455 extern int trace_remove_event_call(struct trace_event_call *call);
 456 extern int trace_event_get_offsets(struct trace_event_call *call);
 457
 458 #define is_signed_type(type)    (((type)(-1)) < (type)1)
 459
 460 int trace_set_clr_event(const char *system, const char *event, int set);
 461
 462 /*
 463  * The double __builtin_constant_p is because gcc will give us an error
 464  * if we try to allocate the static variable to fmt if it is not a
 465  * constant. Even with the outer if statement optimizing out.
 466  */
 467 #define event_trace_printk(ip, fmt, args...)                            \
 468 do {                                                                    \
 469         __trace_printk_check_format(fmt, ##args);                       \
 470         tracing_record_cmdline(current);                                \
 471         if (__builtin_constant_p(fmt)) {                                \
 472                 static const char *trace_printk_fmt                     \
 473                   __attribute__((section("__trace_printk_fmt"))) =      \
 474                         __builtin_constant_p(fmt) ? fmt : NULL;         \
 475                                                                         \
 476                 __trace_bprintk(ip, trace_printk_fmt, ##args);          \
 477         } else                                                          \
 478                 __trace_printk(ip, fmt, ##args);                        \
 479 } while (0)
 480
 481 #ifdef CONFIG_PERF_EVENTS
 482 struct perf_event;
 483
 484 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
 485
 486 extern int  perf_trace_init(struct perf_event *event);
 487 extern void perf_trace_destroy(struct perf_event *event);
 488 extern int  perf_trace_add(struct perf_event *event, int flags);
 489 extern void perf_trace_del(struct perf_event *event, int flags);
 490 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
 491                                      char *filter_str);
 492 extern void ftrace_profile_free_filter(struct perf_event *event);
 493 void perf_trace_buf_update(void *record, u16 type);
 494 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
 495
 496 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
 497                                struct trace_event_call *call, u64 count,
 498                                struct pt_regs *regs, struct hlist_head *head,
 499                                struct task_struct *task);
 500
 501 static inline void
 502 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
 503                        u64 count, struct pt_regs *regs, void *head,
 504                        struct task_struct *task)
 505 {
 506         perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
 507 }
 508 #endif
 509
 510 #endif /* _LINUX_TRACE_EVENT_H */