include/linux/trace_events.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2
   3 #ifndef _LINUX_TRACE_EVENT_H
   4 #define _LINUX_TRACE_EVENT_H
   5
   6 #include <linux/ring_buffer.h>
   7 #include <linux/trace_seq.h>
   8 #include <linux/percpu.h>
   9 #include <linux/hardirq.h>
  10 #include <linux/perf_event.h>
  11 #include <linux/tracepoint.h>
  12
  13 struct trace_array;
  14 struct trace_buffer;
  15 struct tracer;
  16 struct dentry;
  17 struct bpf_prog;
  18
  19 const char *trace_print_flags_seq(struct trace_seq *p, const char *delim,
  20                                   unsigned long flags,
  21                                   const struct trace_print_flags *flag_array);
  22
  23 const char *trace_print_symbols_seq(struct trace_seq *p, unsigned long val,
  24                                     const struct trace_print_flags *symbol_array);
  25
  26 #if BITS_PER_LONG == 32
  27 const char *trace_print_flags_seq_u64(struct trace_seq *p, const char *delim,
  28                       unsigned long long flags,
  29                       const struct trace_print_flags_u64 *flag_array);
  30
  31 const char *trace_print_symbols_seq_u64(struct trace_seq *p,
  32                                         unsigned long long val,
  33                                         const struct trace_print_flags_u64
  34                                                                  *symbol_array);
  35 #endif
  36
  37 const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
  38                                     unsigned int bitmask_size);
  39
  40 const char *trace_print_hex_seq(struct trace_seq *p,
  41                                 const unsigned char *buf, int len,
  42                                 bool concatenate);
  43
  44 const char *trace_print_array_seq(struct trace_seq *p,
  45                                    const void *buf, int count,
  46                                    size_t el_size);
  47
  48 struct trace_iterator;
  49 struct trace_event;
  50
  51 int trace_raw_output_prep(struct trace_iterator *iter,
  52                           struct trace_event *event);
  53
  54 /*
  55  * The trace entry - the most basic unit of tracing. This is what
  56  * is printed in the end as a single line in the trace output, such as:
  57  *
  58  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
  59  */
  60 struct trace_entry {
  61         unsigned short          type;
  62         unsigned char           flags;
  63         unsigned char           preempt_count;
  64         int                     pid;
  65 };
  66
  67 #define TRACE_EVENT_TYPE_MAX                                            \
  68         ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
  69
  70 /*
  71  * Trace iterator - used by printout routines who present trace
  72  * results to users and which routines might sleep, etc:
  73  */
  74 struct trace_iterator {
  75         struct trace_array      *tr;
  76         struct tracer           *trace;
  77         struct trace_buffer     *trace_buffer;
  78         void                    *private;
  79         int                     cpu_file;
  80         struct mutex            mutex;
  81         struct ring_buffer_iter **buffer_iter;
  82         unsigned long           iter_flags;
  83
  84         /* trace_seq for __print_flags() and __print_symbolic() etc. */
  85         struct trace_seq        tmp_seq;
  86
  87         cpumask_var_t           started;
  88
  89         /* it's true when current open file is snapshot */
  90         bool                    snapshot;
  91
  92         /* The below is zeroed out in pipe_read */
  93         struct trace_seq        seq;
  94         struct trace_entry      *ent;
  95         unsigned long           lost_events;
  96         int                     leftover;
  97         int                     ent_size;
  98         int                     cpu;
  99         u64                     ts;
 100
 101         loff_t                  pos;
 102         long                    idx;
 103
 104         /* All new field here will be zeroed out in pipe_read */
 105 };
 106
 107 enum trace_iter_flags {
 108         TRACE_FILE_LAT_FMT      = 1,
 109         TRACE_FILE_ANNOTATE     = 2,
 110         TRACE_FILE_TIME_IN_NS   = 4,
 111 };
 112
 113
 114 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
 115                                       int flags, struct trace_event *event);
 116
 117 struct trace_event_functions {
 118         trace_print_func        trace;
 119         trace_print_func        raw;
 120         trace_print_func        hex;
 121         trace_print_func        binary;
 122 };
 123
 124 struct trace_event {
 125         struct hlist_node               node;
 126         struct list_head                list;
 127         int                             type;
 128         struct trace_event_functions    *funcs;
 129 };
 130
 131 extern int register_trace_event(struct trace_event *event);
 132 extern int unregister_trace_event(struct trace_event *event);
 133
 134 /* Return values for print_line callback */
 135 enum print_line_t {
 136         TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
 137         TRACE_TYPE_HANDLED      = 1,
 138         TRACE_TYPE_UNHANDLED    = 2,    /* Relay to other output functions */
 139         TRACE_TYPE_NO_CONSUME   = 3     /* Handled but ask to not consume */
 140 };
 141
 142 enum print_line_t trace_handle_return(struct trace_seq *s);
 143
 144 void tracing_generic_entry_update(struct trace_entry *entry,
 145                                   unsigned long flags,
 146                                   int pc);
 147 struct trace_event_file;
 148
 149 struct ring_buffer_event *
 150 trace_event_buffer_lock_reserve(struct ring_buffer **current_buffer,
 151                                 struct trace_event_file *trace_file,
 152                                 int type, unsigned long len,
 153                                 unsigned long flags, int pc);
 154
 155 #define TRACE_RECORD_CMDLINE    BIT(0)
 156 #define TRACE_RECORD_TGID       BIT(1)
 157
 158 void tracing_record_taskinfo(struct task_struct *task, int flags);
 159 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
 160                                           struct task_struct *next, int flags);
 161
 162 void tracing_record_cmdline(struct task_struct *task);
 163 void tracing_record_tgid(struct task_struct *task);
 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_TRACEPOINT_BIT,
 222         TRACE_EVENT_FL_KPROBE_BIT,
 223         TRACE_EVENT_FL_UPROBE_BIT,
 224 };
 225
 226 /*
 227  * Event flags:
 228  *  FILTERED      - The event has a filter attached
 229  *  CAP_ANY       - Any user can enable for perf
 230  *  NO_SET_FILTER - Set when filter has error and is to be ignored
 231  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
 232  *  TRACEPOINT    - Event is a tracepoint
 233  *  KPROBE        - Event is a kprobe
 234  *  UPROBE        - Event is a uprobe
 235  */
 236 enum {
 237         TRACE_EVENT_FL_FILTERED         = (1 << TRACE_EVENT_FL_FILTERED_BIT),
 238         TRACE_EVENT_FL_CAP_ANY          = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
 239         TRACE_EVENT_FL_NO_SET_FILTER    = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
 240         TRACE_EVENT_FL_IGNORE_ENABLE    = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_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_array __rcu     *prog_array;
 276
 277         int     (*perf_perm)(struct trace_event_call *,
 278                              struct perf_event *);
 279 #endif
 280 };
 281
 282 #ifdef CONFIG_PERF_EVENTS
 283 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
 284 {
 285         /*
 286          * This inline function checks whether call->prog_array
 287          * is valid or not. The function is called in various places,
 288          * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
 289          *
 290          * If this function returns true, and later call->prog_array
 291          * becomes false inside rcu_read_lock/unlock region,
 292          * we bail out then. If this function return false,
 293          * there is a risk that we might miss a few events if the checking
 294          * were delayed until inside rcu_read_lock/unlock region and
 295          * call->prog_array happened to become non-NULL then.
 296          *
 297          * Here, READ_ONCE() is used instead of rcu_access_pointer().
 298          * rcu_access_pointer() requires the actual definition of
 299          * "struct bpf_prog_array" while READ_ONCE() only needs
 300          * a declaration of the same type.
 301          */
 302         return !!READ_ONCE(call->prog_array);
 303 }
 304 #endif
 305
 306 static inline const char *
 307 trace_event_name(struct trace_event_call *call)
 308 {
 309         if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
 310                 return call->tp ? call->tp->name : NULL;
 311         else
 312                 return call->name;
 313 }
 314
 315 struct trace_array;
 316 struct trace_subsystem_dir;
 317
 318 enum {
 319         EVENT_FILE_FL_ENABLED_BIT,
 320         EVENT_FILE_FL_RECORDED_CMD_BIT,
 321         EVENT_FILE_FL_RECORDED_TGID_BIT,
 322         EVENT_FILE_FL_FILTERED_BIT,
 323         EVENT_FILE_FL_NO_SET_FILTER_BIT,
 324         EVENT_FILE_FL_SOFT_MODE_BIT,
 325         EVENT_FILE_FL_SOFT_DISABLED_BIT,
 326         EVENT_FILE_FL_TRIGGER_MODE_BIT,
 327         EVENT_FILE_FL_TRIGGER_COND_BIT,
 328         EVENT_FILE_FL_PID_FILTER_BIT,
 329         EVENT_FILE_FL_WAS_ENABLED_BIT,
 330 };
 331
 332 /*
 333  * Event file flags:
 334  *  ENABLED       - The event is enabled
 335  *  RECORDED_CMD  - The comms should be recorded at sched_switch
 336  *  RECORDED_TGID - The tgids should be recorded at sched_switch
 337  *  FILTERED      - The event has a filter attached
 338  *  NO_SET_FILTER - Set when filter has error and is to be ignored
 339  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
 340  *  SOFT_DISABLED - When set, do not trace the event (even though its
 341  *                   tracepoint may be enabled)
 342  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
 343  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
 344  *  PID_FILTER    - When set, the event is filtered based on pid
 345  *  WAS_ENABLED   - Set when enabled to know to clear trace on module removal
 346  */
 347 enum {
 348         EVENT_FILE_FL_ENABLED           = (1 << EVENT_FILE_FL_ENABLED_BIT),
 349         EVENT_FILE_FL_RECORDED_CMD      = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
 350         EVENT_FILE_FL_RECORDED_TGID     = (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
 351         EVENT_FILE_FL_FILTERED          = (1 << EVENT_FILE_FL_FILTERED_BIT),
 352         EVENT_FILE_FL_NO_SET_FILTER     = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
 353         EVENT_FILE_FL_SOFT_MODE         = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
 354         EVENT_FILE_FL_SOFT_DISABLED     = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
 355         EVENT_FILE_FL_TRIGGER_MODE      = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
 356         EVENT_FILE_FL_TRIGGER_COND      = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
 357         EVENT_FILE_FL_PID_FILTER        = (1 << EVENT_FILE_FL_PID_FILTER_BIT),
 358         EVENT_FILE_FL_WAS_ENABLED       = (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
 359 };
 360
 361 struct trace_event_file {
 362         struct list_head                list;
 363         struct trace_event_call         *event_call;
 364         struct event_filter __rcu       *filter;
 365         struct dentry                   *dir;
 366         struct trace_array              *tr;
 367         struct trace_subsystem_dir      *system;
 368         struct list_head                triggers;
 369
 370         /*
 371          * 32 bit flags:
 372          *   bit 0:             enabled
 373          *   bit 1:             enabled cmd record
 374          *   bit 2:             enable/disable with the soft disable bit
 375          *   bit 3:             soft disabled
 376          *   bit 4:             trigger enabled
 377          *
 378          * Note: The bits must be set atomically to prevent races
 379          * from other writers. Reads of flags do not need to be in
 380          * sync as they occur in critical sections. But the way flags
 381          * is currently used, these changes do not affect the code
 382          * except that when a change is made, it may have a slight
 383          * delay in propagating the changes to other CPUs due to
 384          * caching and such. Which is mostly OK ;-)
 385          */
 386         unsigned long           flags;
 387         atomic_t                sm_ref; /* soft-mode reference counter */
 388         atomic_t                tm_ref; /* trigger-mode reference counter */
 389 };
 390
 391 #define __TRACE_EVENT_FLAGS(name, value)                                \
 392         static int __init trace_init_flags_##name(void)                 \
 393         {                                                               \
 394                 event_##name.flags |= value;                            \
 395                 return 0;                                               \
 396         }                                                               \
 397         early_initcall(trace_init_flags_##name);
 398
 399 #define __TRACE_EVENT_PERF_PERM(name, expr...)                          \
 400         static int perf_perm_##name(struct trace_event_call *tp_event, \
 401                                     struct perf_event *p_event)         \
 402         {                                                               \
 403                 return ({ expr; });                                     \
 404         }                                                               \
 405         static int __init trace_init_perf_perm_##name(void)             \
 406         {                                                               \
 407                 event_##name.perf_perm = &perf_perm_##name;             \
 408                 return 0;                                               \
 409         }                                                               \
 410         early_initcall(trace_init_perf_perm_##name);
 411
 412 #define PERF_MAX_TRACE_SIZE     2048
 413
 414 #define MAX_FILTER_STR_VAL      256     /* Should handle KSYM_SYMBOL_LEN */
 415
 416 enum event_trigger_type {
 417         ETT_NONE                = (0),
 418         ETT_TRACE_ONOFF         = (1 << 0),
 419         ETT_SNAPSHOT            = (1 << 1),
 420         ETT_STACKTRACE          = (1 << 2),
 421         ETT_EVENT_ENABLE        = (1 << 3),
 422         ETT_EVENT_HIST          = (1 << 4),
 423         ETT_HIST_ENABLE         = (1 << 5),
 424 };
 425
 426 extern int filter_match_preds(struct event_filter *filter, void *rec);
 427
 428 extern enum event_trigger_type event_triggers_call(struct trace_event_file *file,
 429                                                    void *rec);
 430 extern void event_triggers_post_call(struct trace_event_file *file,
 431                                      enum event_trigger_type tt,
 432                                      void *rec);
 433
 434 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
 435
 436 /**
 437  * trace_trigger_soft_disabled - do triggers and test if soft disabled
 438  * @file: The file pointer of the event to test
 439  *
 440  * If any triggers without filters are attached to this event, they
 441  * will be called here. If the event is soft disabled and has no
 442  * triggers that require testing the fields, it will return true,
 443  * otherwise false.
 444  */
 445 static inline bool
 446 trace_trigger_soft_disabled(struct trace_event_file *file)
 447 {
 448         unsigned long eflags = file->flags;
 449
 450         if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
 451                 if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
 452                         event_triggers_call(file, NULL);
 453                 if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
 454                         return true;
 455                 if (eflags & EVENT_FILE_FL_PID_FILTER)
 456                         return trace_event_ignore_this_pid(file);
 457         }
 458         return false;
 459 }
 460
 461 #ifdef CONFIG_BPF_EVENTS
 462 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
 463 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog);
 464 void perf_event_detach_bpf_prog(struct perf_event *event);
 465 #else
 466 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
 467 {
 468         return 1;
 469 }
 470
 471 static inline int
 472 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog)
 473 {
 474         return -EOPNOTSUPP;
 475 }
 476
 477 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
 478
 479 #endif
 480
 481 enum {
 482         FILTER_OTHER = 0,
 483         FILTER_STATIC_STRING,
 484         FILTER_DYN_STRING,
 485         FILTER_PTR_STRING,
 486         FILTER_TRACE_FN,
 487         FILTER_COMM,
 488         FILTER_CPU,
 489 };
 490
 491 extern int trace_event_raw_init(struct trace_event_call *call);
 492 extern int trace_define_field(struct trace_event_call *call, const char *type,
 493                               const char *name, int offset, int size,
 494                               int is_signed, int filter_type);
 495 extern int trace_add_event_call(struct trace_event_call *call);
 496 extern int trace_remove_event_call(struct trace_event_call *call);
 497 extern int trace_event_get_offsets(struct trace_event_call *call);
 498
 499 #define is_signed_type(type)    (((type)(-1)) < (type)1)
 500
 501 int trace_set_clr_event(const char *system, const char *event, int set);
 502
 503 /*
 504  * The double __builtin_constant_p is because gcc will give us an error
 505  * if we try to allocate the static variable to fmt if it is not a
 506  * constant. Even with the outer if statement optimizing out.
 507  */
 508 #define event_trace_printk(ip, fmt, args...)                            \
 509 do {                                                                    \
 510         __trace_printk_check_format(fmt, ##args);                       \
 511         tracing_record_cmdline(current);                                \
 512         if (__builtin_constant_p(fmt)) {                                \
 513                 static const char *trace_printk_fmt                     \
 514                   __attribute__((section("__trace_printk_fmt"))) =      \
 515                         __builtin_constant_p(fmt) ? fmt : NULL;         \
 516                                                                         \
 517                 __trace_bprintk(ip, trace_printk_fmt, ##args);          \
 518         } else                                                          \
 519                 __trace_printk(ip, fmt, ##args);                        \
 520 } while (0)
 521
 522 #ifdef CONFIG_PERF_EVENTS
 523 struct perf_event;
 524
 525 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
 526
 527 extern int  perf_trace_init(struct perf_event *event);
 528 extern void perf_trace_destroy(struct perf_event *event);
 529 extern int  perf_trace_add(struct perf_event *event, int flags);
 530 extern void perf_trace_del(struct perf_event *event, int flags);
 531 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
 532                                      char *filter_str);
 533 extern void ftrace_profile_free_filter(struct perf_event *event);
 534 void perf_trace_buf_update(void *record, u16 type);
 535 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
 536
 537 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
 538                                struct trace_event_call *call, u64 count,
 539                                struct pt_regs *regs, struct hlist_head *head,
 540                                struct task_struct *task);
 541
 542 static inline void
 543 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
 544                        u64 count, struct pt_regs *regs, void *head,
 545                        struct task_struct *task, struct perf_event *event)
 546 {
 547         perf_tp_event(type, count, raw_data, size, regs, head, rctx, task, event);
 548 }
 549
 550 #endif
 551
 552 #endif /* _LINUX_TRACE_EVENT_H */