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         /*
 178          * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
 179          * custom action was taken and the default action is not to be
 180          * performed.
 181          */
 182         TRACE_REG_PERF_ADD,
 183         TRACE_REG_PERF_DEL,
 184 #endif
 185 };
 186
 187 struct trace_event_call;
 188
 189 struct trace_event_class {
 190         const char              *system;
 191         void                    *probe;
 192 #ifdef CONFIG_PERF_EVENTS
 193         void                    *perf_probe;
 194 #endif
 195         int                     (*reg)(struct trace_event_call *event,
 196                                        enum trace_reg type, void *data);
 197         int                     (*define_fields)(struct trace_event_call *);
 198         struct list_head        *(*get_fields)(struct trace_event_call *);
 199         struct list_head        fields;
 200         int                     (*raw_init)(struct trace_event_call *);
 201 };
 202
 203 extern int trace_event_reg(struct trace_event_call *event,
 204                             enum trace_reg type, void *data);
 205
 206 struct trace_event_buffer {
 207         struct ring_buffer              *buffer;
 208         struct ring_buffer_event        *event;
 209         struct trace_event_file         *trace_file;
 210         void                            *entry;
 211         unsigned long                   flags;
 212         int                             pc;
 213 };
 214
 215 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 216                                   struct trace_event_file *trace_file,
 217                                   unsigned long len);
 218
 219 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
 220
 221 enum {
 222         TRACE_EVENT_FL_FILTERED_BIT,
 223         TRACE_EVENT_FL_CAP_ANY_BIT,
 224         TRACE_EVENT_FL_NO_SET_FILTER_BIT,
 225         TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
 226         TRACE_EVENT_FL_TRACEPOINT_BIT,
 227         TRACE_EVENT_FL_KPROBE_BIT,
 228         TRACE_EVENT_FL_UPROBE_BIT,
 229 };
 230
 231 /*
 232  * Event flags:
 233  *  FILTERED      - The event has a filter attached
 234  *  CAP_ANY       - Any user can enable for perf
 235  *  NO_SET_FILTER - Set when filter has error and is to be ignored
 236  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
 237  *  TRACEPOINT    - Event is a tracepoint
 238  *  KPROBE        - Event is a kprobe
 239  *  UPROBE        - Event is a uprobe
 240  */
 241 enum {
 242         TRACE_EVENT_FL_FILTERED         = (1 << TRACE_EVENT_FL_FILTERED_BIT),
 243         TRACE_EVENT_FL_CAP_ANY          = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
 244         TRACE_EVENT_FL_NO_SET_FILTER    = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
 245         TRACE_EVENT_FL_IGNORE_ENABLE    = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
 246         TRACE_EVENT_FL_TRACEPOINT       = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
 247         TRACE_EVENT_FL_KPROBE           = (1 << TRACE_EVENT_FL_KPROBE_BIT),
 248         TRACE_EVENT_FL_UPROBE           = (1 << TRACE_EVENT_FL_UPROBE_BIT),
 249 };
 250
 251 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
 252
 253 struct trace_event_call {
 254         struct list_head        list;
 255         struct trace_event_class *class;
 256         union {
 257                 char                    *name;
 258                 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
 259                 struct tracepoint       *tp;
 260         };
 261         struct trace_event      event;
 262         char                    *print_fmt;
 263         struct event_filter     *filter;
 264         void                    *mod;
 265         void                    *data;
 266         /*
 267          *   bit 0:             filter_active
 268          *   bit 1:             allow trace by non root (cap any)
 269          *   bit 2:             failed to apply filter
 270          *   bit 3:             trace internal event (do not enable)
 271          *   bit 4:             Event was enabled by module
 272          *   bit 5:             use call filter rather than file filter
 273          *   bit 6:             Event is a tracepoint
 274          */
 275         int                     flags; /* static flags of different events */
 276
 277 #ifdef CONFIG_PERF_EVENTS
 278         int                             perf_refcount;
 279         struct hlist_head __percpu      *perf_events;
 280         struct bpf_prog_array __rcu     *prog_array;
 281
 282         int     (*perf_perm)(struct trace_event_call *,
 283                              struct perf_event *);
 284 #endif
 285 };
 286
 287 #ifdef CONFIG_PERF_EVENTS
 288 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
 289 {
 290         /*
 291          * This inline function checks whether call->prog_array
 292          * is valid or not. The function is called in various places,
 293          * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
 294          *
 295          * If this function returns true, and later call->prog_array
 296          * becomes false inside rcu_read_lock/unlock region,
 297          * we bail out then. If this function return false,
 298          * there is a risk that we might miss a few events if the checking
 299          * were delayed until inside rcu_read_lock/unlock region and
 300          * call->prog_array happened to become non-NULL then.
 301          *
 302          * Here, READ_ONCE() is used instead of rcu_access_pointer().
 303          * rcu_access_pointer() requires the actual definition of
 304          * "struct bpf_prog_array" while READ_ONCE() only needs
 305          * a declaration of the same type.
 306          */
 307         return !!READ_ONCE(call->prog_array);
 308 }
 309 #endif
 310
 311 static inline const char *
 312 trace_event_name(struct trace_event_call *call)
 313 {
 314         if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
 315                 return call->tp ? call->tp->name : NULL;
 316         else
 317                 return call->name;
 318 }
 319
 320 struct trace_array;
 321 struct trace_subsystem_dir;
 322
 323 enum {
 324         EVENT_FILE_FL_ENABLED_BIT,
 325         EVENT_FILE_FL_RECORDED_CMD_BIT,
 326         EVENT_FILE_FL_RECORDED_TGID_BIT,
 327         EVENT_FILE_FL_FILTERED_BIT,
 328         EVENT_FILE_FL_NO_SET_FILTER_BIT,
 329         EVENT_FILE_FL_SOFT_MODE_BIT,
 330         EVENT_FILE_FL_SOFT_DISABLED_BIT,
 331         EVENT_FILE_FL_TRIGGER_MODE_BIT,
 332         EVENT_FILE_FL_TRIGGER_COND_BIT,
 333         EVENT_FILE_FL_PID_FILTER_BIT,
 334         EVENT_FILE_FL_WAS_ENABLED_BIT,
 335 };
 336
 337 /*
 338  * Event file flags:
 339  *  ENABLED       - The event is enabled
 340  *  RECORDED_CMD  - The comms should be recorded at sched_switch
 341  *  RECORDED_TGID - The tgids should be recorded at sched_switch
 342  *  FILTERED      - The event has a filter attached
 343  *  NO_SET_FILTER - Set when filter has error and is to be ignored
 344  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
 345  *  SOFT_DISABLED - When set, do not trace the event (even though its
 346  *                   tracepoint may be enabled)
 347  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
 348  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
 349  *  PID_FILTER    - When set, the event is filtered based on pid
 350  *  WAS_ENABLED   - Set when enabled to know to clear trace on module removal
 351  */
 352 enum {
 353         EVENT_FILE_FL_ENABLED           = (1 << EVENT_FILE_FL_ENABLED_BIT),
 354         EVENT_FILE_FL_RECORDED_CMD      = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
 355         EVENT_FILE_FL_RECORDED_TGID     = (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
 356         EVENT_FILE_FL_FILTERED          = (1 << EVENT_FILE_FL_FILTERED_BIT),
 357         EVENT_FILE_FL_NO_SET_FILTER     = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
 358         EVENT_FILE_FL_SOFT_MODE         = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
 359         EVENT_FILE_FL_SOFT_DISABLED     = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
 360         EVENT_FILE_FL_TRIGGER_MODE      = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
 361         EVENT_FILE_FL_TRIGGER_COND      = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
 362         EVENT_FILE_FL_PID_FILTER        = (1 << EVENT_FILE_FL_PID_FILTER_BIT),
 363         EVENT_FILE_FL_WAS_ENABLED       = (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
 364 };
 365
 366 struct trace_event_file {
 367         struct list_head                list;
 368         struct trace_event_call         *event_call;
 369         struct event_filter __rcu       *filter;
 370         struct dentry                   *dir;
 371         struct trace_array              *tr;
 372         struct trace_subsystem_dir      *system;
 373         struct list_head                triggers;
 374
 375         /*
 376          * 32 bit flags:
 377          *   bit 0:             enabled
 378          *   bit 1:             enabled cmd record
 379          *   bit 2:             enable/disable with the soft disable bit
 380          *   bit 3:             soft disabled
 381          *   bit 4:             trigger enabled
 382          *
 383          * Note: The bits must be set atomically to prevent races
 384          * from other writers. Reads of flags do not need to be in
 385          * sync as they occur in critical sections. But the way flags
 386          * is currently used, these changes do not affect the code
 387          * except that when a change is made, it may have a slight
 388          * delay in propagating the changes to other CPUs due to
 389          * caching and such. Which is mostly OK ;-)
 390          */
 391         unsigned long           flags;
 392         atomic_t                sm_ref; /* soft-mode reference counter */
 393         atomic_t                tm_ref; /* trigger-mode reference counter */
 394 };
 395
 396 #define __TRACE_EVENT_FLAGS(name, value)                                \
 397         static int __init trace_init_flags_##name(void)                 \
 398         {                                                               \
 399                 event_##name.flags |= value;                            \
 400                 return 0;                                               \
 401         }                                                               \
 402         early_initcall(trace_init_flags_##name);
 403
 404 #define __TRACE_EVENT_PERF_PERM(name, expr...)                          \
 405         static int perf_perm_##name(struct trace_event_call *tp_event, \
 406                                     struct perf_event *p_event)         \
 407         {                                                               \
 408                 return ({ expr; });                                     \
 409         }                                                               \
 410         static int __init trace_init_perf_perm_##name(void)             \
 411         {                                                               \
 412                 event_##name.perf_perm = &perf_perm_##name;             \
 413                 return 0;                                               \
 414         }                                                               \
 415         early_initcall(trace_init_perf_perm_##name);
 416
 417 #define PERF_MAX_TRACE_SIZE     2048
 418
 419 #define MAX_FILTER_STR_VAL      256     /* Should handle KSYM_SYMBOL_LEN */
 420
 421 enum event_trigger_type {
 422         ETT_NONE                = (0),
 423         ETT_TRACE_ONOFF         = (1 << 0),
 424         ETT_SNAPSHOT            = (1 << 1),
 425         ETT_STACKTRACE          = (1 << 2),
 426         ETT_EVENT_ENABLE        = (1 << 3),
 427         ETT_EVENT_HIST          = (1 << 4),
 428         ETT_HIST_ENABLE         = (1 << 5),
 429 };
 430
 431 extern int filter_match_preds(struct event_filter *filter, void *rec);
 432
 433 extern enum event_trigger_type event_triggers_call(struct trace_event_file *file,
 434                                                    void *rec);
 435 extern void event_triggers_post_call(struct trace_event_file *file,
 436                                      enum event_trigger_type tt,
 437                                      void *rec);
 438
 439 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
 440
 441 /**
 442  * trace_trigger_soft_disabled - do triggers and test if soft disabled
 443  * @file: The file pointer of the event to test
 444  *
 445  * If any triggers without filters are attached to this event, they
 446  * will be called here. If the event is soft disabled and has no
 447  * triggers that require testing the fields, it will return true,
 448  * otherwise false.
 449  */
 450 static inline bool
 451 trace_trigger_soft_disabled(struct trace_event_file *file)
 452 {
 453         unsigned long eflags = file->flags;
 454
 455         if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
 456                 if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
 457                         event_triggers_call(file, NULL);
 458                 if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
 459                         return true;
 460                 if (eflags & EVENT_FILE_FL_PID_FILTER)
 461                         return trace_event_ignore_this_pid(file);
 462         }
 463         return false;
 464 }
 465
 466 #ifdef CONFIG_BPF_EVENTS
 467 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
 468 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog);
 469 void perf_event_detach_bpf_prog(struct perf_event *event);
 470 int perf_event_query_prog_array(struct perf_event *event, void __user *info);
 471 #else
 472 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
 473 {
 474         return 1;
 475 }
 476
 477 static inline int
 478 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog)
 479 {
 480         return -EOPNOTSUPP;
 481 }
 482
 483 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
 484
 485 static inline int
 486 perf_event_query_prog_array(struct perf_event *event, void __user *info)
 487 {
 488         return -EOPNOTSUPP;
 489 }
 490 #endif
 491
 492 enum {
 493         FILTER_OTHER = 0,
 494         FILTER_STATIC_STRING,
 495         FILTER_DYN_STRING,
 496         FILTER_PTR_STRING,
 497         FILTER_TRACE_FN,
 498         FILTER_COMM,
 499         FILTER_CPU,
 500 };
 501
 502 extern int trace_event_raw_init(struct trace_event_call *call);
 503 extern int trace_define_field(struct trace_event_call *call, const char *type,
 504                               const char *name, int offset, int size,
 505                               int is_signed, int filter_type);
 506 extern int trace_add_event_call(struct trace_event_call *call);
 507 extern int trace_remove_event_call(struct trace_event_call *call);
 508 extern int trace_event_get_offsets(struct trace_event_call *call);
 509
 510 #define is_signed_type(type)    (((type)(-1)) < (type)1)
 511
 512 int trace_set_clr_event(const char *system, const char *event, int set);
 513
 514 /*
 515  * The double __builtin_constant_p is because gcc will give us an error
 516  * if we try to allocate the static variable to fmt if it is not a
 517  * constant. Even with the outer if statement optimizing out.
 518  */
 519 #define event_trace_printk(ip, fmt, args...)                            \
 520 do {                                                                    \
 521         __trace_printk_check_format(fmt, ##args);                       \
 522         tracing_record_cmdline(current);                                \
 523         if (__builtin_constant_p(fmt)) {                                \
 524                 static const char *trace_printk_fmt                     \
 525                   __attribute__((section("__trace_printk_fmt"))) =      \
 526                         __builtin_constant_p(fmt) ? fmt : NULL;         \
 527                                                                         \
 528                 __trace_bprintk(ip, trace_printk_fmt, ##args);          \
 529         } else                                                          \
 530                 __trace_printk(ip, fmt, ##args);                        \
 531 } while (0)
 532
 533 #ifdef CONFIG_PERF_EVENTS
 534 struct perf_event;
 535
 536 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
 537 DECLARE_PER_CPU(int, bpf_kprobe_override);
 538
 539 extern int  perf_trace_init(struct perf_event *event);
 540 extern void perf_trace_destroy(struct perf_event *event);
 541 extern int  perf_trace_add(struct perf_event *event, int flags);
 542 extern void perf_trace_del(struct perf_event *event, int flags);
 543 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
 544                                      char *filter_str);
 545 extern void ftrace_profile_free_filter(struct perf_event *event);
 546 void perf_trace_buf_update(void *record, u16 type);
 547 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
 548
 549 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
 550                                struct trace_event_call *call, u64 count,
 551                                struct pt_regs *regs, struct hlist_head *head,
 552                                struct task_struct *task);
 553
 554 static inline void
 555 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
 556                        u64 count, struct pt_regs *regs, void *head,
 557                        struct task_struct *task)
 558 {
 559         perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
 560 }
 561
 562 #endif
 563
 564 #endif /* _LINUX_TRACE_EVENT_H */