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 array_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 const char *
  49 trace_print_hex_dump_seq(struct trace_seq *p, const char *prefix_str,
  50                          int prefix_type, int rowsize, int groupsize,
  51                          const void *buf, size_t len, bool ascii);
  52
  53 struct trace_iterator;
  54 struct trace_event;
  55
  56 int trace_raw_output_prep(struct trace_iterator *iter,
  57                           struct trace_event *event);
  58 extern __printf(2, 3)
  59 void trace_event_printf(struct trace_iterator *iter, const char *fmt, ...);
  60
  61 /*
  62  * The trace entry - the most basic unit of tracing. This is what
  63  * is printed in the end as a single line in the trace output, such as:
  64  *
  65  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
  66  */
  67 struct trace_entry {
  68         unsigned short          type;
  69         unsigned char           flags;
  70         unsigned char           preempt_count;
  71         int                     pid;
  72 };
  73
  74 #define TRACE_EVENT_TYPE_MAX                                            \
  75         ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
  76
  77 /*
  78  * Trace iterator - used by printout routines who present trace
  79  * results to users and which routines might sleep, etc:
  80  */
  81 struct trace_iterator {
  82         struct trace_array      *tr;
  83         struct tracer           *trace;
  84         struct array_buffer     *array_buffer;
  85         void                    *private;
  86         int                     cpu_file;
  87         struct mutex            mutex;
  88         struct ring_buffer_iter **buffer_iter;
  89         unsigned long           iter_flags;
  90         void                    *temp;  /* temp holder */
  91         unsigned int            temp_size;
  92         char                    *fmt;   /* modified format holder */
  93         unsigned int            fmt_size;
  94
  95         /* trace_seq for __print_flags() and __print_symbolic() etc. */
  96         struct trace_seq        tmp_seq;
  97
  98         cpumask_var_t           started;
  99
 100         /* it's true when current open file is snapshot */
 101         bool                    snapshot;
 102
 103         /* The below is zeroed out in pipe_read */
 104         struct trace_seq        seq;
 105         struct trace_entry      *ent;
 106         unsigned long           lost_events;
 107         int                     leftover;
 108         int                     ent_size;
 109         int                     cpu;
 110         u64                     ts;
 111
 112         loff_t                  pos;
 113         long                    idx;
 114
 115         /* All new field here will be zeroed out in pipe_read */
 116 };
 117
 118 enum trace_iter_flags {
 119         TRACE_FILE_LAT_FMT      = 1,
 120         TRACE_FILE_ANNOTATE     = 2,
 121         TRACE_FILE_TIME_IN_NS   = 4,
 122 };
 123
 124
 125 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
 126                                       int flags, struct trace_event *event);
 127
 128 struct trace_event_functions {
 129         trace_print_func        trace;
 130         trace_print_func        raw;
 131         trace_print_func        hex;
 132         trace_print_func        binary;
 133 };
 134
 135 struct trace_event {
 136         struct hlist_node               node;
 137         struct list_head                list;
 138         int                             type;
 139         struct trace_event_functions    *funcs;
 140 };
 141
 142 extern int register_trace_event(struct trace_event *event);
 143 extern int unregister_trace_event(struct trace_event *event);
 144
 145 /* Return values for print_line callback */
 146 enum print_line_t {
 147         TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
 148         TRACE_TYPE_HANDLED      = 1,
 149         TRACE_TYPE_UNHANDLED    = 2,    /* Relay to other output functions */
 150         TRACE_TYPE_NO_CONSUME   = 3     /* Handled but ask to not consume */
 151 };
 152
 153 enum print_line_t trace_handle_return(struct trace_seq *s);
 154
 155 static inline void tracing_generic_entry_update(struct trace_entry *entry,
 156                                                 unsigned short type,
 157                                                 unsigned int trace_ctx)
 158 {
 159         entry->preempt_count            = trace_ctx & 0xff;
 160         entry->pid                      = current->pid;
 161         entry->type                     = type;
 162         entry->flags =                  trace_ctx >> 16;
 163 }
 164
 165 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status);
 166
 167 enum trace_flag_type {
 168         TRACE_FLAG_IRQS_OFF             = 0x01,
 169         TRACE_FLAG_IRQS_NOSUPPORT       = 0x02,
 170         TRACE_FLAG_NEED_RESCHED         = 0x04,
 171         TRACE_FLAG_HARDIRQ              = 0x08,
 172         TRACE_FLAG_SOFTIRQ              = 0x10,
 173         TRACE_FLAG_PREEMPT_RESCHED      = 0x20,
 174         TRACE_FLAG_NMI                  = 0x40,
 175 };
 176
 177 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
 178 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
 179 {
 180         unsigned int irq_status = irqs_disabled_flags(irqflags) ?
 181                 TRACE_FLAG_IRQS_OFF : 0;
 182         return tracing_gen_ctx_irq_test(irq_status);
 183 }
 184 static inline unsigned int tracing_gen_ctx(void)
 185 {
 186         unsigned long irqflags;
 187
 188         local_save_flags(irqflags);
 189         return tracing_gen_ctx_flags(irqflags);
 190 }
 191 #else
 192
 193 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
 194 {
 195         return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
 196 }
 197 static inline unsigned int tracing_gen_ctx(void)
 198 {
 199         return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
 200 }
 201 #endif
 202
 203 static inline unsigned int tracing_gen_ctx_dec(void)
 204 {
 205         unsigned int trace_ctx;
 206
 207         trace_ctx = tracing_gen_ctx();
 208         /*
 209          * Subtract one from the preemption counter if preemption is enabled,
 210          * see trace_event_buffer_reserve()for details.
 211          */
 212         if (IS_ENABLED(CONFIG_PREEMPTION))
 213                 trace_ctx--;
 214         return trace_ctx;
 215 }
 216
 217 struct trace_event_file;
 218
 219 struct ring_buffer_event *
 220 trace_event_buffer_lock_reserve(struct trace_buffer **current_buffer,
 221                                 struct trace_event_file *trace_file,
 222                                 int type, unsigned long len,
 223                                 unsigned int trace_ctx);
 224
 225 #define TRACE_RECORD_CMDLINE    BIT(0)
 226 #define TRACE_RECORD_TGID       BIT(1)
 227
 228 void tracing_record_taskinfo(struct task_struct *task, int flags);
 229 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
 230                                           struct task_struct *next, int flags);
 231
 232 void tracing_record_cmdline(struct task_struct *task);
 233 void tracing_record_tgid(struct task_struct *task);
 234
 235 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
 236
 237 struct event_filter;
 238
 239 enum trace_reg {
 240         TRACE_REG_REGISTER,
 241         TRACE_REG_UNREGISTER,
 242 #ifdef CONFIG_PERF_EVENTS
 243         TRACE_REG_PERF_REGISTER,
 244         TRACE_REG_PERF_UNREGISTER,
 245         TRACE_REG_PERF_OPEN,
 246         TRACE_REG_PERF_CLOSE,
 247         /*
 248          * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
 249          * custom action was taken and the default action is not to be
 250          * performed.
 251          */
 252         TRACE_REG_PERF_ADD,
 253         TRACE_REG_PERF_DEL,
 254 #endif
 255 };
 256
 257 struct trace_event_call;
 258
 259 #define TRACE_FUNCTION_TYPE ((const char *)~0UL)
 260
 261 struct trace_event_fields {
 262         const char *type;
 263         union {
 264                 struct {
 265                         const char *name;
 266                         const int  size;
 267                         const int  align;
 268                         const int  is_signed;
 269                         const int  filter_type;
 270                 };
 271                 int (*define_fields)(struct trace_event_call *);
 272         };
 273 };
 274
 275 struct trace_event_class {
 276         const char              *system;
 277         void                    *probe;
 278 #ifdef CONFIG_PERF_EVENTS
 279         void                    *perf_probe;
 280 #endif
 281         int                     (*reg)(struct trace_event_call *event,
 282                                        enum trace_reg type, void *data);
 283         struct trace_event_fields *fields_array;
 284         struct list_head        *(*get_fields)(struct trace_event_call *);
 285         struct list_head        fields;
 286         int                     (*raw_init)(struct trace_event_call *);
 287 };
 288
 289 extern int trace_event_reg(struct trace_event_call *event,
 290                             enum trace_reg type, void *data);
 291
 292 struct trace_event_buffer {
 293         struct trace_buffer             *buffer;
 294         struct ring_buffer_event        *event;
 295         struct trace_event_file         *trace_file;
 296         void                            *entry;
 297         unsigned int                    trace_ctx;
 298         struct pt_regs                  *regs;
 299 };
 300
 301 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 302                                   struct trace_event_file *trace_file,
 303                                   unsigned long len);
 304
 305 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
 306
 307 enum {
 308         TRACE_EVENT_FL_FILTERED_BIT,
 309         TRACE_EVENT_FL_CAP_ANY_BIT,
 310         TRACE_EVENT_FL_NO_SET_FILTER_BIT,
 311         TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
 312         TRACE_EVENT_FL_TRACEPOINT_BIT,
 313         TRACE_EVENT_FL_DYNAMIC_BIT,
 314         TRACE_EVENT_FL_KPROBE_BIT,
 315         TRACE_EVENT_FL_UPROBE_BIT,
 316         TRACE_EVENT_FL_EPROBE_BIT,
 317 };
 318
 319 /*
 320  * Event flags:
 321  *  FILTERED      - The event has a filter attached
 322  *  CAP_ANY       - Any user can enable for perf
 323  *  NO_SET_FILTER - Set when filter has error and is to be ignored
 324  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
 325  *  TRACEPOINT    - Event is a tracepoint
 326  *  DYNAMIC       - Event is a dynamic event (created at run time)
 327  *  KPROBE        - Event is a kprobe
 328  *  UPROBE        - Event is a uprobe
 329  *  EPROBE        - Event is an event probe
 330  */
 331 enum {
 332         TRACE_EVENT_FL_FILTERED         = (1 << TRACE_EVENT_FL_FILTERED_BIT),
 333         TRACE_EVENT_FL_CAP_ANY          = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
 334         TRACE_EVENT_FL_NO_SET_FILTER    = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
 335         TRACE_EVENT_FL_IGNORE_ENABLE    = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
 336         TRACE_EVENT_FL_TRACEPOINT       = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
 337         TRACE_EVENT_FL_DYNAMIC          = (1 << TRACE_EVENT_FL_DYNAMIC_BIT),
 338         TRACE_EVENT_FL_KPROBE           = (1 << TRACE_EVENT_FL_KPROBE_BIT),
 339         TRACE_EVENT_FL_UPROBE           = (1 << TRACE_EVENT_FL_UPROBE_BIT),
 340         TRACE_EVENT_FL_EPROBE           = (1 << TRACE_EVENT_FL_EPROBE_BIT),
 341 };
 342
 343 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
 344
 345 struct trace_event_call {
 346         struct list_head        list;
 347         struct trace_event_class *class;
 348         union {
 349                 char                    *name;
 350                 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
 351                 struct tracepoint       *tp;
 352         };
 353         struct trace_event      event;
 354         char                    *print_fmt;
 355         struct event_filter     *filter;
 356         /*
 357          * Static events can disappear with modules,
 358          * where as dynamic ones need their own ref count.
 359          */
 360         union {
 361                 void                            *module;
 362                 atomic_t                        refcnt;
 363         };
 364         void                    *data;
 365
 366         /* See the TRACE_EVENT_FL_* flags above */
 367         int                     flags; /* static flags of different events */
 368
 369 #ifdef CONFIG_PERF_EVENTS
 370         int                             perf_refcount;
 371         struct hlist_head __percpu      *perf_events;
 372         struct bpf_prog_array __rcu     *prog_array;
 373
 374         int     (*perf_perm)(struct trace_event_call *,
 375                              struct perf_event *);
 376 #endif
 377 };
 378
 379 #ifdef CONFIG_DYNAMIC_EVENTS
 380 bool trace_event_dyn_try_get_ref(struct trace_event_call *call);
 381 void trace_event_dyn_put_ref(struct trace_event_call *call);
 382 bool trace_event_dyn_busy(struct trace_event_call *call);
 383 #else
 384 static inline bool trace_event_dyn_try_get_ref(struct trace_event_call *call)
 385 {
 386         /* Without DYNAMIC_EVENTS configured, nothing should be calling this */
 387         return false;
 388 }
 389 static inline void trace_event_dyn_put_ref(struct trace_event_call *call)
 390 {
 391 }
 392 static inline bool trace_event_dyn_busy(struct trace_event_call *call)
 393 {
 394         /* Nothing should call this without DYNAIMIC_EVENTS configured. */
 395         return true;
 396 }
 397 #endif
 398
 399 static inline bool trace_event_try_get_ref(struct trace_event_call *call)
 400 {
 401         if (call->flags & TRACE_EVENT_FL_DYNAMIC)
 402                 return trace_event_dyn_try_get_ref(call);
 403         else
 404                 return try_module_get(call->module);
 405 }
 406
 407 static inline void trace_event_put_ref(struct trace_event_call *call)
 408 {
 409         if (call->flags & TRACE_EVENT_FL_DYNAMIC)
 410                 trace_event_dyn_put_ref(call);
 411         else
 412                 module_put(call->module);
 413 }
 414
 415 #ifdef CONFIG_PERF_EVENTS
 416 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
 417 {
 418         /*
 419          * This inline function checks whether call->prog_array
 420          * is valid or not. The function is called in various places,
 421          * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
 422          *
 423          * If this function returns true, and later call->prog_array
 424          * becomes false inside rcu_read_lock/unlock region,
 425          * we bail out then. If this function return false,
 426          * there is a risk that we might miss a few events if the checking
 427          * were delayed until inside rcu_read_lock/unlock region and
 428          * call->prog_array happened to become non-NULL then.
 429          *
 430          * Here, READ_ONCE() is used instead of rcu_access_pointer().
 431          * rcu_access_pointer() requires the actual definition of
 432          * "struct bpf_prog_array" while READ_ONCE() only needs
 433          * a declaration of the same type.
 434          */
 435         return !!READ_ONCE(call->prog_array);
 436 }
 437 #endif
 438
 439 static inline const char *
 440 trace_event_name(struct trace_event_call *call)
 441 {
 442         if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
 443                 return call->tp ? call->tp->name : NULL;
 444         else
 445                 return call->name;
 446 }
 447
 448 static inline struct list_head *
 449 trace_get_fields(struct trace_event_call *event_call)
 450 {
 451         if (!event_call->class->get_fields)
 452                 return &event_call->class->fields;
 453         return event_call->class->get_fields(event_call);
 454 }
 455
 456 struct trace_subsystem_dir;
 457
 458 enum {
 459         EVENT_FILE_FL_ENABLED_BIT,
 460         EVENT_FILE_FL_RECORDED_CMD_BIT,
 461         EVENT_FILE_FL_RECORDED_TGID_BIT,
 462         EVENT_FILE_FL_FILTERED_BIT,
 463         EVENT_FILE_FL_NO_SET_FILTER_BIT,
 464         EVENT_FILE_FL_SOFT_MODE_BIT,
 465         EVENT_FILE_FL_SOFT_DISABLED_BIT,
 466         EVENT_FILE_FL_TRIGGER_MODE_BIT,
 467         EVENT_FILE_FL_TRIGGER_COND_BIT,
 468         EVENT_FILE_FL_PID_FILTER_BIT,
 469         EVENT_FILE_FL_WAS_ENABLED_BIT,
 470 };
 471
 472 extern struct trace_event_file *trace_get_event_file(const char *instance,
 473                                                      const char *system,
 474                                                      const char *event);
 475 extern void trace_put_event_file(struct trace_event_file *file);
 476
 477 #define MAX_DYNEVENT_CMD_LEN    (2048)
 478
 479 enum dynevent_type {
 480         DYNEVENT_TYPE_SYNTH = 1,
 481         DYNEVENT_TYPE_KPROBE,
 482         DYNEVENT_TYPE_NONE,
 483 };
 484
 485 struct dynevent_cmd;
 486
 487 typedef int (*dynevent_create_fn_t)(struct dynevent_cmd *cmd);
 488
 489 struct dynevent_cmd {
 490         struct seq_buf          seq;
 491         const char              *event_name;
 492         unsigned int            n_fields;
 493         enum dynevent_type      type;
 494         dynevent_create_fn_t    run_command;
 495         void                    *private_data;
 496 };
 497
 498 extern int dynevent_create(struct dynevent_cmd *cmd);
 499
 500 extern int synth_event_delete(const char *name);
 501
 502 extern void synth_event_cmd_init(struct dynevent_cmd *cmd,
 503                                  char *buf, int maxlen);
 504
 505 extern int __synth_event_gen_cmd_start(struct dynevent_cmd *cmd,
 506                                        const char *name,
 507                                        struct module *mod, ...);
 508
 509 #define synth_event_gen_cmd_start(cmd, name, mod, ...)  \
 510         __synth_event_gen_cmd_start(cmd, name, mod, ## __VA_ARGS__, NULL)
 511
 512 struct synth_field_desc {
 513         const char *type;
 514         const char *name;
 515 };
 516
 517 extern int synth_event_gen_cmd_array_start(struct dynevent_cmd *cmd,
 518                                            const char *name,
 519                                            struct module *mod,
 520                                            struct synth_field_desc *fields,
 521                                            unsigned int n_fields);
 522 extern int synth_event_create(const char *name,
 523                               struct synth_field_desc *fields,
 524                               unsigned int n_fields, struct module *mod);
 525
 526 extern int synth_event_add_field(struct dynevent_cmd *cmd,
 527                                  const char *type,
 528                                  const char *name);
 529 extern int synth_event_add_field_str(struct dynevent_cmd *cmd,
 530                                      const char *type_name);
 531 extern int synth_event_add_fields(struct dynevent_cmd *cmd,
 532                                   struct synth_field_desc *fields,
 533                                   unsigned int n_fields);
 534
 535 #define synth_event_gen_cmd_end(cmd)    \
 536         dynevent_create(cmd)
 537
 538 struct synth_event;
 539
 540 struct synth_event_trace_state {
 541         struct trace_event_buffer fbuffer;
 542         struct synth_trace_event *entry;
 543         struct trace_buffer *buffer;
 544         struct synth_event *event;
 545         unsigned int cur_field;
 546         unsigned int n_u64;
 547         bool disabled;
 548         bool add_next;
 549         bool add_name;
 550 };
 551
 552 extern int synth_event_trace(struct trace_event_file *file,
 553                              unsigned int n_vals, ...);
 554 extern int synth_event_trace_array(struct trace_event_file *file, u64 *vals,
 555                                    unsigned int n_vals);
 556 extern int synth_event_trace_start(struct trace_event_file *file,
 557                                    struct synth_event_trace_state *trace_state);
 558 extern int synth_event_add_next_val(u64 val,
 559                                     struct synth_event_trace_state *trace_state);
 560 extern int synth_event_add_val(const char *field_name, u64 val,
 561                                struct synth_event_trace_state *trace_state);
 562 extern int synth_event_trace_end(struct synth_event_trace_state *trace_state);
 563
 564 extern int kprobe_event_delete(const char *name);
 565
 566 extern void kprobe_event_cmd_init(struct dynevent_cmd *cmd,
 567                                   char *buf, int maxlen);
 568
 569 #define kprobe_event_gen_cmd_start(cmd, name, loc, ...)                 \
 570         __kprobe_event_gen_cmd_start(cmd, false, name, loc, ## __VA_ARGS__, NULL)
 571
 572 #define kretprobe_event_gen_cmd_start(cmd, name, loc, ...)              \
 573         __kprobe_event_gen_cmd_start(cmd, true, name, loc, ## __VA_ARGS__, NULL)
 574
 575 extern int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd,
 576                                         bool kretprobe,
 577                                         const char *name,
 578                                         const char *loc, ...);
 579
 580 #define kprobe_event_add_fields(cmd, ...)       \
 581         __kprobe_event_add_fields(cmd, ## __VA_ARGS__, NULL)
 582
 583 #define kprobe_event_add_field(cmd, field)      \
 584         __kprobe_event_add_fields(cmd, field, NULL)
 585
 586 extern int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...);
 587
 588 #define kprobe_event_gen_cmd_end(cmd)           \
 589         dynevent_create(cmd)
 590
 591 #define kretprobe_event_gen_cmd_end(cmd)        \
 592         dynevent_create(cmd)
 593
 594 /*
 595  * Event file flags:
 596  *  ENABLED       - The event is enabled
 597  *  RECORDED_CMD  - The comms should be recorded at sched_switch
 598  *  RECORDED_TGID - The tgids should be recorded at sched_switch
 599  *  FILTERED      - The event has a filter attached
 600  *  NO_SET_FILTER - Set when filter has error and is to be ignored
 601  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
 602  *  SOFT_DISABLED - When set, do not trace the event (even though its
 603  *                   tracepoint may be enabled)
 604  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
 605  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
 606  *  PID_FILTER    - When set, the event is filtered based on pid
 607  *  WAS_ENABLED   - Set when enabled to know to clear trace on module removal
 608  */
 609 enum {
 610         EVENT_FILE_FL_ENABLED           = (1 << EVENT_FILE_FL_ENABLED_BIT),
 611         EVENT_FILE_FL_RECORDED_CMD      = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
 612         EVENT_FILE_FL_RECORDED_TGID     = (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
 613         EVENT_FILE_FL_FILTERED          = (1 << EVENT_FILE_FL_FILTERED_BIT),
 614         EVENT_FILE_FL_NO_SET_FILTER     = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
 615         EVENT_FILE_FL_SOFT_MODE         = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
 616         EVENT_FILE_FL_SOFT_DISABLED     = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
 617         EVENT_FILE_FL_TRIGGER_MODE      = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
 618         EVENT_FILE_FL_TRIGGER_COND      = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
 619         EVENT_FILE_FL_PID_FILTER        = (1 << EVENT_FILE_FL_PID_FILTER_BIT),
 620         EVENT_FILE_FL_WAS_ENABLED       = (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
 621 };
 622
 623 struct trace_event_file {
 624         struct list_head                list;
 625         struct trace_event_call         *event_call;
 626         struct event_filter __rcu       *filter;
 627         struct dentry                   *dir;
 628         struct trace_array              *tr;
 629         struct trace_subsystem_dir      *system;
 630         struct list_head                triggers;
 631
 632         /*
 633          * 32 bit flags:
 634          *   bit 0:             enabled
 635          *   bit 1:             enabled cmd record
 636          *   bit 2:             enable/disable with the soft disable bit
 637          *   bit 3:             soft disabled
 638          *   bit 4:             trigger enabled
 639          *
 640          * Note: The bits must be set atomically to prevent races
 641          * from other writers. Reads of flags do not need to be in
 642          * sync as they occur in critical sections. But the way flags
 643          * is currently used, these changes do not affect the code
 644          * except that when a change is made, it may have a slight
 645          * delay in propagating the changes to other CPUs due to
 646          * caching and such. Which is mostly OK ;-)
 647          */
 648         unsigned long           flags;
 649         atomic_t                sm_ref; /* soft-mode reference counter */
 650         atomic_t                tm_ref; /* trigger-mode reference counter */
 651 };
 652
 653 #define __TRACE_EVENT_FLAGS(name, value)                                \
 654         static int __init trace_init_flags_##name(void)                 \
 655         {                                                               \
 656                 event_##name.flags |= value;                            \
 657                 return 0;                                               \
 658         }                                                               \
 659         early_initcall(trace_init_flags_##name);
 660
 661 #define __TRACE_EVENT_PERF_PERM(name, expr...)                          \
 662         static int perf_perm_##name(struct trace_event_call *tp_event, \
 663                                     struct perf_event *p_event)         \
 664         {                                                               \
 665                 return ({ expr; });                                     \
 666         }                                                               \
 667         static int __init trace_init_perf_perm_##name(void)             \
 668         {                                                               \
 669                 event_##name.perf_perm = &perf_perm_##name;             \
 670                 return 0;                                               \
 671         }                                                               \
 672         early_initcall(trace_init_perf_perm_##name);
 673
 674 #define PERF_MAX_TRACE_SIZE     2048
 675
 676 #define MAX_FILTER_STR_VAL      256U    /* Should handle KSYM_SYMBOL_LEN */
 677
 678 enum event_trigger_type {
 679         ETT_NONE                = (0),
 680         ETT_TRACE_ONOFF         = (1 << 0),
 681         ETT_SNAPSHOT            = (1 << 1),
 682         ETT_STACKTRACE          = (1 << 2),
 683         ETT_EVENT_ENABLE        = (1 << 3),
 684         ETT_EVENT_HIST          = (1 << 4),
 685         ETT_HIST_ENABLE         = (1 << 5),
 686         ETT_EVENT_EPROBE        = (1 << 6),
 687 };
 688
 689 extern int filter_match_preds(struct event_filter *filter, void *rec);
 690
 691 extern enum event_trigger_type
 692 event_triggers_call(struct trace_event_file *file,
 693                     struct trace_buffer *buffer, void *rec,
 694                     struct ring_buffer_event *event);
 695 extern void
 696 event_triggers_post_call(struct trace_event_file *file,
 697                          enum event_trigger_type tt);
 698
 699 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
 700
 701 /**
 702  * trace_trigger_soft_disabled - do triggers and test if soft disabled
 703  * @file: The file pointer of the event to test
 704  *
 705  * If any triggers without filters are attached to this event, they
 706  * will be called here. If the event is soft disabled and has no
 707  * triggers that require testing the fields, it will return true,
 708  * otherwise false.
 709  */
 710 static inline bool
 711 trace_trigger_soft_disabled(struct trace_event_file *file)
 712 {
 713         unsigned long eflags = file->flags;
 714
 715         if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
 716                 if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
 717                         event_triggers_call(file, NULL, NULL, NULL);
 718                 if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
 719                         return true;
 720                 if (eflags & EVENT_FILE_FL_PID_FILTER)
 721                         return trace_event_ignore_this_pid(file);
 722         }
 723         return false;
 724 }
 725
 726 #ifdef CONFIG_BPF_EVENTS
 727 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
 728 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
 729 void perf_event_detach_bpf_prog(struct perf_event *event);
 730 int perf_event_query_prog_array(struct perf_event *event, void __user *info);
 731 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
 732 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
 733 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
 734 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
 735 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
 736                             u32 *fd_type, const char **buf,
 737                             u64 *probe_offset, u64 *probe_addr);
 738 #else
 739 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
 740 {
 741         return 1;
 742 }
 743
 744 static inline int
 745 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie)
 746 {
 747         return -EOPNOTSUPP;
 748 }
 749
 750 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
 751
 752 static inline int
 753 perf_event_query_prog_array(struct perf_event *event, void __user *info)
 754 {
 755         return -EOPNOTSUPP;
 756 }
 757 static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
 758 {
 759         return -EOPNOTSUPP;
 760 }
 761 static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
 762 {
 763         return -EOPNOTSUPP;
 764 }
 765 static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
 766 {
 767         return NULL;
 768 }
 769 static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
 770 {
 771 }
 772 static inline int bpf_get_perf_event_info(const struct perf_event *event,
 773                                           u32 *prog_id, u32 *fd_type,
 774                                           const char **buf, u64 *probe_offset,
 775                                           u64 *probe_addr)
 776 {
 777         return -EOPNOTSUPP;
 778 }
 779 #endif
 780
 781 enum {
 782         FILTER_OTHER = 0,
 783         FILTER_STATIC_STRING,
 784         FILTER_DYN_STRING,
 785         FILTER_PTR_STRING,
 786         FILTER_TRACE_FN,
 787         FILTER_COMM,
 788         FILTER_CPU,
 789 };
 790
 791 extern int trace_event_raw_init(struct trace_event_call *call);
 792 extern int trace_define_field(struct trace_event_call *call, const char *type,
 793                               const char *name, int offset, int size,
 794                               int is_signed, int filter_type);
 795 extern int trace_add_event_call(struct trace_event_call *call);
 796 extern int trace_remove_event_call(struct trace_event_call *call);
 797 extern int trace_event_get_offsets(struct trace_event_call *call);
 798
 799 #define is_signed_type(type)    (((type)(-1)) < (type)1)
 800
 801 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
 802 int trace_set_clr_event(const char *system, const char *event, int set);
 803 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
 804                 const char *event, bool enable);
 805 /*
 806  * The double __builtin_constant_p is because gcc will give us an error
 807  * if we try to allocate the static variable to fmt if it is not a
 808  * constant. Even with the outer if statement optimizing out.
 809  */
 810 #define event_trace_printk(ip, fmt, args...)                            \
 811 do {                                                                    \
 812         __trace_printk_check_format(fmt, ##args);                       \
 813         tracing_record_cmdline(current);                                \
 814         if (__builtin_constant_p(fmt)) {                                \
 815                 static const char *trace_printk_fmt                     \
 816                   __section("__trace_printk_fmt") =                     \
 817                         __builtin_constant_p(fmt) ? fmt : NULL;         \
 818                                                                         \
 819                 __trace_bprintk(ip, trace_printk_fmt, ##args);          \
 820         } else                                                          \
 821                 __trace_printk(ip, fmt, ##args);                        \
 822 } while (0)
 823
 824 #ifdef CONFIG_PERF_EVENTS
 825 struct perf_event;
 826
 827 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
 828 DECLARE_PER_CPU(int, bpf_kprobe_override);
 829
 830 extern int  perf_trace_init(struct perf_event *event);
 831 extern void perf_trace_destroy(struct perf_event *event);
 832 extern int  perf_trace_add(struct perf_event *event, int flags);
 833 extern void perf_trace_del(struct perf_event *event, int flags);
 834 #ifdef CONFIG_KPROBE_EVENTS
 835 extern int  perf_kprobe_init(struct perf_event *event, bool is_retprobe);
 836 extern void perf_kprobe_destroy(struct perf_event *event);
 837 extern int bpf_get_kprobe_info(const struct perf_event *event,
 838                                u32 *fd_type, const char **symbol,
 839                                u64 *probe_offset, u64 *probe_addr,
 840                                bool perf_type_tracepoint);
 841 #endif
 842 #ifdef CONFIG_UPROBE_EVENTS
 843 extern int  perf_uprobe_init(struct perf_event *event,
 844                              unsigned long ref_ctr_offset, bool is_retprobe);
 845 extern void perf_uprobe_destroy(struct perf_event *event);
 846 extern int bpf_get_uprobe_info(const struct perf_event *event,
 847                                u32 *fd_type, const char **filename,
 848                                u64 *probe_offset, bool perf_type_tracepoint);
 849 #endif
 850 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
 851                                      char *filter_str);
 852 extern void ftrace_profile_free_filter(struct perf_event *event);
 853 void perf_trace_buf_update(void *record, u16 type);
 854 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
 855
 856 int perf_event_set_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
 857 void perf_event_free_bpf_prog(struct perf_event *event);
 858
 859 void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
 860 void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
 861 void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
 862                     u64 arg3);
 863 void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
 864                     u64 arg3, u64 arg4);
 865 void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
 866                     u64 arg3, u64 arg4, u64 arg5);
 867 void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
 868                     u64 arg3, u64 arg4, u64 arg5, u64 arg6);
 869 void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
 870                     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
 871 void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
 872                     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
 873                     u64 arg8);
 874 void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
 875                     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
 876                     u64 arg8, u64 arg9);
 877 void bpf_trace_run10(struct bpf_prog *prog, u64 arg1, u64 arg2,
 878                      u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
 879                      u64 arg8, u64 arg9, u64 arg10);
 880 void bpf_trace_run11(struct bpf_prog *prog, u64 arg1, u64 arg2,
 881                      u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
 882                      u64 arg8, u64 arg9, u64 arg10, u64 arg11);
 883 void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
 884                      u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
 885                      u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
 886 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
 887                                struct trace_event_call *call, u64 count,
 888                                struct pt_regs *regs, struct hlist_head *head,
 889                                struct task_struct *task);
 890
 891 static inline void
 892 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
 893                        u64 count, struct pt_regs *regs, void *head,
 894                        struct task_struct *task)
 895 {
 896         perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
 897 }
 898
 899 #endif
 900
 901 #endif /* _LINUX_TRACE_EVENT_H */