tools/perf/builtin-trace.c

   1 /*
   2  * builtin-trace.c
   3  *
   4  * Builtin 'trace' command:
   5  *
   6  * Display a continuously updated trace of any workload, CPU, specific PID,
   7  * system wide, etc.  Default format is loosely strace like, but any other
   8  * event may be specified using --event.
   9  *
  10  * Copyright (C) 2012, 2013, 2014, 2015 Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
  11  *
  12  * Initially based on the 'trace' prototype by Thomas Gleixner:
  13  *
  14  * http://lwn.net/Articles/415728/ ("Announcing a new utility: 'trace'")
  15  */
  16
  17 #include "util/record.h"
  18 #include <traceevent/event-parse.h>
  19 #include <api/fs/tracing_path.h>
  20 #include <bpf/bpf.h>
  21 #include "util/bpf_map.h"
  22 #include "util/rlimit.h"
  23 #include "builtin.h"
  24 #include "util/cgroup.h"
  25 #include "util/color.h"
  26 #include "util/config.h"
  27 #include "util/debug.h"
  28 #include "util/dso.h"
  29 #include "util/env.h"
  30 #include "util/event.h"
  31 #include "util/evsel.h"
  32 #include "util/evsel_fprintf.h"
  33 #include "util/synthetic-events.h"
  34 #include "util/evlist.h"
  35 #include "util/evswitch.h"
  36 #include "util/mmap.h"
  37 #include <subcmd/pager.h>
  38 #include <subcmd/exec-cmd.h>
  39 #include "util/machine.h"
  40 #include "util/map.h"
  41 #include "util/symbol.h"
  42 #include "util/path.h"
  43 #include "util/session.h"
  44 #include "util/thread.h"
  45 #include <subcmd/parse-options.h>
  46 #include "util/strlist.h"
  47 #include "util/intlist.h"
  48 #include "util/thread_map.h"
  49 #include "util/stat.h"
  50 #include "util/tool.h"
  51 #include "util/util.h"
  52 #include "trace/beauty/beauty.h"
  53 #include "trace-event.h"
  54 #include "util/parse-events.h"
  55 #include "util/bpf-loader.h"
  56 #include "util/tracepoint.h"
  57 #include "callchain.h"
  58 #include "print_binary.h"
  59 #include "string2.h"
  60 #include "syscalltbl.h"
  61 #include "rb_resort.h"
  62 #include "../perf.h"
  63
  64 #include <errno.h>
  65 #include <inttypes.h>
  66 #include <poll.h>
  67 #include <signal.h>
  68 #include <stdlib.h>
  69 #include <string.h>
  70 #include <linux/err.h>
  71 #include <linux/filter.h>
  72 #include <linux/kernel.h>
  73 #include <linux/random.h>
  74 #include <linux/stringify.h>
  75 #include <linux/time64.h>
  76 #include <linux/zalloc.h>
  77 #include <fcntl.h>
  78 #include <sys/sysmacros.h>
  79
  80 #include <linux/ctype.h>
  81 #include <perf/mmap.h>
  82
  83 #ifndef O_CLOEXEC
  84 # define O_CLOEXEC              02000000
  85 #endif
  86
  87 #ifndef F_LINUX_SPECIFIC_BASE
  88 # define F_LINUX_SPECIFIC_BASE  1024
  89 #endif
  90
  91 /*
  92  * strtoul: Go from a string to a value, i.e. for msr: MSR_FS_BASE to 0xc0000100
  93  */
  94 struct syscall_arg_fmt {
  95         size_t     (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
  96         bool       (*strtoul)(char *bf, size_t size, struct syscall_arg *arg, u64 *val);
  97         unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
  98         void       *parm;
  99         const char *name;
 100         u16        nr_entries; // for arrays
 101         bool       show_zero;
 102 };
 103
 104 struct syscall_fmt {
 105         const char *name;
 106         const char *alias;
 107         struct {
 108                 const char *sys_enter,
 109                            *sys_exit;
 110         }          bpf_prog_name;
 111         struct syscall_arg_fmt arg[6];
 112         u8         nr_args;
 113         bool       errpid;
 114         bool       timeout;
 115         bool       hexret;
 116 };
 117
 118 struct trace {
 119         struct perf_tool        tool;
 120         struct syscalltbl       *sctbl;
 121         struct {
 122                 struct syscall  *table;
 123                 struct bpf_map  *map;
 124                 struct { // per syscall BPF_MAP_TYPE_PROG_ARRAY
 125                         struct bpf_map  *sys_enter,
 126                                         *sys_exit;
 127                 }               prog_array;
 128                 struct {
 129                         struct evsel *sys_enter,
 130                                           *sys_exit,
 131                                           *augmented;
 132                 }               events;
 133                 struct bpf_program *unaugmented_prog;
 134         } syscalls;
 135         struct {
 136                 struct bpf_map *map;
 137         } dump;
 138         struct record_opts      opts;
 139         struct evlist   *evlist;
 140         struct machine          *host;
 141         struct thread           *current;
 142         struct bpf_object       *bpf_obj;
 143         struct cgroup           *cgroup;
 144         u64                     base_time;
 145         FILE                    *output;
 146         unsigned long           nr_events;
 147         unsigned long           nr_events_printed;
 148         unsigned long           max_events;
 149         struct evswitch         evswitch;
 150         struct strlist          *ev_qualifier;
 151         struct {
 152                 size_t          nr;
 153                 int             *entries;
 154         }                       ev_qualifier_ids;
 155         struct {
 156                 size_t          nr;
 157                 pid_t           *entries;
 158                 struct bpf_map  *map;
 159         }                       filter_pids;
 160         double                  duration_filter;
 161         double                  runtime_ms;
 162         struct {
 163                 u64             vfs_getname,
 164                                 proc_getname;
 165         } stats;
 166         unsigned int            max_stack;
 167         unsigned int            min_stack;
 168         int                     raw_augmented_syscalls_args_size;
 169         bool                    raw_augmented_syscalls;
 170         bool                    fd_path_disabled;
 171         bool                    sort_events;
 172         bool                    not_ev_qualifier;
 173         bool                    live;
 174         bool                    full_time;
 175         bool                    sched;
 176         bool                    multiple_threads;
 177         bool                    summary;
 178         bool                    summary_only;
 179         bool                    errno_summary;
 180         bool                    failure_only;
 181         bool                    show_comm;
 182         bool                    print_sample;
 183         bool                    show_tool_stats;
 184         bool                    trace_syscalls;
 185         bool                    libtraceevent_print;
 186         bool                    kernel_syscallchains;
 187         s16                     args_alignment;
 188         bool                    show_tstamp;
 189         bool                    show_duration;
 190         bool                    show_zeros;
 191         bool                    show_arg_names;
 192         bool                    show_string_prefix;
 193         bool                    force;
 194         bool                    vfs_getname;
 195         int                     trace_pgfaults;
 196         char                    *perfconfig_events;
 197         struct {
 198                 struct ordered_events   data;
 199                 u64                     last;
 200         } oe;
 201 };
 202
 203 struct tp_field {
 204         int offset;
 205         union {
 206                 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
 207                 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
 208         };
 209 };
 210
 211 #define TP_UINT_FIELD(bits) \
 212 static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
 213 { \
 214         u##bits value; \
 215         memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
 216         return value;  \
 217 }
 218
 219 TP_UINT_FIELD(8);
 220 TP_UINT_FIELD(16);
 221 TP_UINT_FIELD(32);
 222 TP_UINT_FIELD(64);
 223
 224 #define TP_UINT_FIELD__SWAPPED(bits) \
 225 static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
 226 { \
 227         u##bits value; \
 228         memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
 229         return bswap_##bits(value);\
 230 }
 231
 232 TP_UINT_FIELD__SWAPPED(16);
 233 TP_UINT_FIELD__SWAPPED(32);
 234 TP_UINT_FIELD__SWAPPED(64);
 235
 236 static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)
 237 {
 238         field->offset = offset;
 239
 240         switch (size) {
 241         case 1:
 242                 field->integer = tp_field__u8;
 243                 break;
 244         case 2:
 245                 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
 246                 break;
 247         case 4:
 248                 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
 249                 break;
 250         case 8:
 251                 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
 252                 break;
 253         default:
 254                 return -1;
 255         }
 256
 257         return 0;
 258 }
 259
 260 static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)
 261 {
 262         return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);
 263 }
 264
 265 static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
 266 {
 267         return sample->raw_data + field->offset;
 268 }
 269
 270 static int __tp_field__init_ptr(struct tp_field *field, int offset)
 271 {
 272         field->offset = offset;
 273         field->pointer = tp_field__ptr;
 274         return 0;
 275 }
 276
 277 static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)
 278 {
 279         return __tp_field__init_ptr(field, format_field->offset);
 280 }
 281
 282 struct syscall_tp {
 283         struct tp_field id;
 284         union {
 285                 struct tp_field args, ret;
 286         };
 287 };
 288
 289 /*
 290  * The evsel->priv as used by 'perf trace'
 291  * sc:  for raw_syscalls:sys_{enter,exit} and syscalls:sys_{enter,exit}_SYSCALLNAME
 292  * fmt: for all the other tracepoints
 293  */
 294 struct evsel_trace {
 295         struct syscall_tp       sc;
 296         struct syscall_arg_fmt  *fmt;
 297 };
 298
 299 static struct evsel_trace *evsel_trace__new(void)
 300 {
 301         return zalloc(sizeof(struct evsel_trace));
 302 }
 303
 304 static void evsel_trace__delete(struct evsel_trace *et)
 305 {
 306         if (et == NULL)
 307                 return;
 308
 309         zfree(&et->fmt);
 310         free(et);
 311 }
 312
 313 /*
 314  * Used with raw_syscalls:sys_{enter,exit} and with the
 315  * syscalls:sys_{enter,exit}_SYSCALL tracepoints
 316  */
 317 static inline struct syscall_tp *__evsel__syscall_tp(struct evsel *evsel)
 318 {
 319         struct evsel_trace *et = evsel->priv;
 320
 321         return &et->sc;
 322 }
 323
 324 static struct syscall_tp *evsel__syscall_tp(struct evsel *evsel)
 325 {
 326         if (evsel->priv == NULL) {
 327                 evsel->priv = evsel_trace__new();
 328                 if (evsel->priv == NULL)
 329                         return NULL;
 330         }
 331
 332         return __evsel__syscall_tp(evsel);
 333 }
 334
 335 /*
 336  * Used with all the other tracepoints.
 337  */
 338 static inline struct syscall_arg_fmt *__evsel__syscall_arg_fmt(struct evsel *evsel)
 339 {
 340         struct evsel_trace *et = evsel->priv;
 341
 342         return et->fmt;
 343 }
 344
 345 static struct syscall_arg_fmt *evsel__syscall_arg_fmt(struct evsel *evsel)
 346 {
 347         struct evsel_trace *et = evsel->priv;
 348
 349         if (evsel->priv == NULL) {
 350                 et = evsel->priv = evsel_trace__new();
 351
 352                 if (et == NULL)
 353                         return NULL;
 354         }
 355
 356         if (et->fmt == NULL) {
 357                 et->fmt = calloc(evsel->tp_format->format.nr_fields, sizeof(struct syscall_arg_fmt));
 358                 if (et->fmt == NULL)
 359                         goto out_delete;
 360         }
 361
 362         return __evsel__syscall_arg_fmt(evsel);
 363
 364 out_delete:
 365         evsel_trace__delete(evsel->priv);
 366         evsel->priv = NULL;
 367         return NULL;
 368 }
 369
 370 static int evsel__init_tp_uint_field(struct evsel *evsel, struct tp_field *field, const char *name)
 371 {
 372         struct tep_format_field *format_field = evsel__field(evsel, name);
 373
 374         if (format_field == NULL)
 375                 return -1;
 376
 377         return tp_field__init_uint(field, format_field, evsel->needs_swap);
 378 }
 379
 380 #define perf_evsel__init_sc_tp_uint_field(evsel, name) \
 381         ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
 382            evsel__init_tp_uint_field(evsel, &sc->name, #name); })
 383
 384 static int evsel__init_tp_ptr_field(struct evsel *evsel, struct tp_field *field, const char *name)
 385 {
 386         struct tep_format_field *format_field = evsel__field(evsel, name);
 387
 388         if (format_field == NULL)
 389                 return -1;
 390
 391         return tp_field__init_ptr(field, format_field);
 392 }
 393
 394 #define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
 395         ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
 396            evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
 397
 398 static void evsel__delete_priv(struct evsel *evsel)
 399 {
 400         zfree(&evsel->priv);
 401         evsel__delete(evsel);
 402 }
 403
 404 static int evsel__init_syscall_tp(struct evsel *evsel)
 405 {
 406         struct syscall_tp *sc = evsel__syscall_tp(evsel);
 407
 408         if (sc != NULL) {
 409                 if (evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&
 410                     evsel__init_tp_uint_field(evsel, &sc->id, "nr"))
 411                         return -ENOENT;
 412                 return 0;
 413         }
 414
 415         return -ENOMEM;
 416 }
 417
 418 static int evsel__init_augmented_syscall_tp(struct evsel *evsel, struct evsel *tp)
 419 {
 420         struct syscall_tp *sc = evsel__syscall_tp(evsel);
 421
 422         if (sc != NULL) {
 423                 struct tep_format_field *syscall_id = evsel__field(tp, "id");
 424                 if (syscall_id == NULL)
 425                         syscall_id = evsel__field(tp, "__syscall_nr");
 426                 if (syscall_id == NULL ||
 427                     __tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))
 428                         return -EINVAL;
 429
 430                 return 0;
 431         }
 432
 433         return -ENOMEM;
 434 }
 435
 436 static int evsel__init_augmented_syscall_tp_args(struct evsel *evsel)
 437 {
 438         struct syscall_tp *sc = __evsel__syscall_tp(evsel);
 439
 440         return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));
 441 }
 442
 443 static int evsel__init_augmented_syscall_tp_ret(struct evsel *evsel)
 444 {
 445         struct syscall_tp *sc = __evsel__syscall_tp(evsel);
 446
 447         return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);
 448 }
 449
 450 static int evsel__init_raw_syscall_tp(struct evsel *evsel, void *handler)
 451 {
 452         if (evsel__syscall_tp(evsel) != NULL) {
 453                 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
 454                         return -ENOENT;
 455
 456                 evsel->handler = handler;
 457                 return 0;
 458         }
 459
 460         return -ENOMEM;
 461 }
 462
 463 static struct evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)
 464 {
 465         struct evsel *evsel = evsel__newtp("raw_syscalls", direction);
 466
 467         /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
 468         if (IS_ERR(evsel))
 469                 evsel = evsel__newtp("syscalls", direction);
 470
 471         if (IS_ERR(evsel))
 472                 return NULL;
 473
 474         if (evsel__init_raw_syscall_tp(evsel, handler))
 475                 goto out_delete;
 476
 477         return evsel;
 478
 479 out_delete:
 480         evsel__delete_priv(evsel);
 481         return NULL;
 482 }
 483
 484 #define perf_evsel__sc_tp_uint(evsel, name, sample) \
 485         ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
 486            fields->name.integer(&fields->name, sample); })
 487
 488 #define perf_evsel__sc_tp_ptr(evsel, name, sample) \
 489         ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
 490            fields->name.pointer(&fields->name, sample); })
 491
 492 size_t strarray__scnprintf_suffix(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_suffix, int val)
 493 {
 494         int idx = val - sa->offset;
 495
 496         if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
 497                 size_t printed = scnprintf(bf, size, intfmt, val);
 498                 if (show_suffix)
 499                         printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
 500                 return printed;
 501         }
 502
 503         return scnprintf(bf, size, "%s%s", sa->entries[idx], show_suffix ? sa->prefix : "");
 504 }
 505
 506 size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
 507 {
 508         int idx = val - sa->offset;
 509
 510         if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
 511                 size_t printed = scnprintf(bf, size, intfmt, val);
 512                 if (show_prefix)
 513                         printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
 514                 return printed;
 515         }
 516
 517         return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
 518 }
 519
 520 static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
 521                                                 const char *intfmt,
 522                                                 struct syscall_arg *arg)
 523 {
 524         return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);
 525 }
 526
 527 static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
 528                                               struct syscall_arg *arg)
 529 {
 530         return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
 531 }
 532
 533 #define SCA_STRARRAY syscall_arg__scnprintf_strarray
 534
 535 bool syscall_arg__strtoul_strarray(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
 536 {
 537         return strarray__strtoul(arg->parm, bf, size, ret);
 538 }
 539
 540 bool syscall_arg__strtoul_strarray_flags(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
 541 {
 542         return strarray__strtoul_flags(arg->parm, bf, size, ret);
 543 }
 544
 545 bool syscall_arg__strtoul_strarrays(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
 546 {
 547         return strarrays__strtoul(arg->parm, bf, size, ret);
 548 }
 549
 550 size_t syscall_arg__scnprintf_strarray_flags(char *bf, size_t size, struct syscall_arg *arg)
 551 {
 552         return strarray__scnprintf_flags(arg->parm, bf, size, arg->show_string_prefix, arg->val);
 553 }
 554
 555 size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
 556 {
 557         size_t printed;
 558         int i;
 559
 560         for (i = 0; i < sas->nr_entries; ++i) {
 561                 struct strarray *sa = sas->entries[i];
 562                 int idx = val - sa->offset;
 563
 564                 if (idx >= 0 && idx < sa->nr_entries) {
 565                         if (sa->entries[idx] == NULL)
 566                                 break;
 567                         return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
 568                 }
 569         }
 570
 571         printed = scnprintf(bf, size, intfmt, val);
 572         if (show_prefix)
 573                 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);
 574         return printed;
 575 }
 576
 577 bool strarray__strtoul(struct strarray *sa, char *bf, size_t size, u64 *ret)
 578 {
 579         int i;
 580
 581         for (i = 0; i < sa->nr_entries; ++i) {
 582                 if (sa->entries[i] && strncmp(sa->entries[i], bf, size) == 0 && sa->entries[i][size] == '\0') {
 583                         *ret = sa->offset + i;
 584                         return true;
 585                 }
 586         }
 587
 588         return false;
 589 }
 590
 591 bool strarray__strtoul_flags(struct strarray *sa, char *bf, size_t size, u64 *ret)
 592 {
 593         u64 val = 0;
 594         char *tok = bf, *sep, *end;
 595
 596         *ret = 0;
 597
 598         while (size != 0) {
 599                 int toklen = size;
 600
 601                 sep = memchr(tok, '|', size);
 602                 if (sep != NULL) {
 603                         size -= sep - tok + 1;
 604
 605                         end = sep - 1;
 606                         while (end > tok && isspace(*end))
 607                                 --end;
 608
 609                         toklen = end - tok + 1;
 610                 }
 611
 612                 while (isspace(*tok))
 613                         ++tok;
 614
 615                 if (isalpha(*tok) || *tok == '_') {
 616                         if (!strarray__strtoul(sa, tok, toklen, &val))
 617                                 return false;
 618                 } else {
 619                         bool is_hexa = tok[0] == 0 && (tok[1] = 'x' || tok[1] == 'X');
 620
 621                         val = strtoul(tok, NULL, is_hexa ? 16 : 0);
 622                 }
 623
 624                 *ret |= (1 << (val - 1));
 625
 626                 if (sep == NULL)
 627                         break;
 628                 tok = sep + 1;
 629         }
 630
 631         return true;
 632 }
 633
 634 bool strarrays__strtoul(struct strarrays *sas, char *bf, size_t size, u64 *ret)
 635 {
 636         int i;
 637
 638         for (i = 0; i < sas->nr_entries; ++i) {
 639                 struct strarray *sa = sas->entries[i];
 640
 641                 if (strarray__strtoul(sa, bf, size, ret))
 642                         return true;
 643         }
 644
 645         return false;
 646 }
 647
 648 size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,
 649                                         struct syscall_arg *arg)
 650 {
 651         return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);
 652 }
 653
 654 #ifndef AT_FDCWD
 655 #define AT_FDCWD        -100
 656 #endif
 657
 658 static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
 659                                            struct syscall_arg *arg)
 660 {
 661         int fd = arg->val;
 662         const char *prefix = "AT_FD";
 663
 664         if (fd == AT_FDCWD)
 665                 return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");
 666
 667         return syscall_arg__scnprintf_fd(bf, size, arg);
 668 }
 669
 670 #define SCA_FDAT syscall_arg__scnprintf_fd_at
 671
 672 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
 673                                               struct syscall_arg *arg);
 674
 675 #define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
 676
 677 size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)
 678 {
 679         return scnprintf(bf, size, "%#lx", arg->val);
 680 }
 681
 682 size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)
 683 {
 684         if (arg->val == 0)
 685                 return scnprintf(bf, size, "NULL");
 686         return syscall_arg__scnprintf_hex(bf, size, arg);
 687 }
 688
 689 size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)
 690 {
 691         return scnprintf(bf, size, "%d", arg->val);
 692 }
 693
 694 size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)
 695 {
 696         return scnprintf(bf, size, "%ld", arg->val);
 697 }
 698
 699 static size_t syscall_arg__scnprintf_char_array(char *bf, size_t size, struct syscall_arg *arg)
 700 {
 701         // XXX Hey, maybe for sched:sched_switch prev/next comm fields we can
 702         //     fill missing comms using thread__set_comm()...
 703         //     here or in a special syscall_arg__scnprintf_pid_sched_tp...
 704         return scnprintf(bf, size, "\"%-.*s\"", arg->fmt->nr_entries ?: arg->len, arg->val);
 705 }
 706
 707 #define SCA_CHAR_ARRAY syscall_arg__scnprintf_char_array
 708
 709 static const char *bpf_cmd[] = {
 710         "MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
 711         "MAP_GET_NEXT_KEY", "PROG_LOAD", "OBJ_PIN", "OBJ_GET", "PROG_ATTACH",
 712         "PROG_DETACH", "PROG_TEST_RUN", "PROG_GET_NEXT_ID", "MAP_GET_NEXT_ID",
 713         "PROG_GET_FD_BY_ID", "MAP_GET_FD_BY_ID", "OBJ_GET_INFO_BY_FD",
 714         "PROG_QUERY", "RAW_TRACEPOINT_OPEN", "BTF_LOAD", "BTF_GET_FD_BY_ID",
 715         "TASK_FD_QUERY", "MAP_LOOKUP_AND_DELETE_ELEM", "MAP_FREEZE",
 716         "BTF_GET_NEXT_ID", "MAP_LOOKUP_BATCH", "MAP_LOOKUP_AND_DELETE_BATCH",
 717         "MAP_UPDATE_BATCH", "MAP_DELETE_BATCH", "LINK_CREATE", "LINK_UPDATE",
 718         "LINK_GET_FD_BY_ID", "LINK_GET_NEXT_ID", "ENABLE_STATS", "ITER_CREATE",
 719         "LINK_DETACH", "PROG_BIND_MAP",
 720 };
 721 static DEFINE_STRARRAY(bpf_cmd, "BPF_");
 722
 723 static const char *fsmount_flags[] = {
 724         [1] = "CLOEXEC",
 725 };
 726 static DEFINE_STRARRAY(fsmount_flags, "FSMOUNT_");
 727
 728 #include "trace/beauty/generated/fsconfig_arrays.c"
 729
 730 static DEFINE_STRARRAY(fsconfig_cmds, "FSCONFIG_");
 731
 732 static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
 733 static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);
 734
 735 static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
 736 static DEFINE_STRARRAY(itimers, "ITIMER_");
 737
 738 static const char *keyctl_options[] = {
 739         "GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
 740         "SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
 741         "INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
 742         "ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
 743         "INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
 744 };
 745 static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");
 746
 747 static const char *whences[] = { "SET", "CUR", "END",
 748 #ifdef SEEK_DATA
 749 "DATA",
 750 #endif
 751 #ifdef SEEK_HOLE
 752 "HOLE",
 753 #endif
 754 };
 755 static DEFINE_STRARRAY(whences, "SEEK_");
 756
 757 static const char *fcntl_cmds[] = {
 758         "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
 759         "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",
 760         "SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",
 761         "GETOWNER_UIDS",
 762 };
 763 static DEFINE_STRARRAY(fcntl_cmds, "F_");
 764
 765 static const char *fcntl_linux_specific_cmds[] = {
 766         "SETLEASE", "GETLEASE", "NOTIFY", [5] = "CANCELLK", "DUPFD_CLOEXEC",
 767         "SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",
 768         "GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",
 769 };
 770
 771 static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);
 772
 773 static struct strarray *fcntl_cmds_arrays[] = {
 774         &strarray__fcntl_cmds,
 775         &strarray__fcntl_linux_specific_cmds,
 776 };
 777
 778 static DEFINE_STRARRAYS(fcntl_cmds_arrays);
 779
 780 static const char *rlimit_resources[] = {
 781         "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
 782         "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
 783         "RTTIME",
 784 };
 785 static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");
 786
 787 static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
 788 static DEFINE_STRARRAY(sighow, "SIG_");
 789
 790 static const char *clockid[] = {
 791         "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
 792         "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
 793         "REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
 794 };
 795 static DEFINE_STRARRAY(clockid, "CLOCK_");
 796
 797 static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
 798                                                  struct syscall_arg *arg)
 799 {
 800         bool show_prefix = arg->show_string_prefix;
 801         const char *suffix = "_OK";
 802         size_t printed = 0;
 803         int mode = arg->val;
 804
 805         if (mode == F_OK) /* 0 */
 806                 return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");
 807 #define P_MODE(n) \
 808         if (mode & n##_OK) { \
 809                 printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \
 810                 mode &= ~n##_OK; \
 811         }
 812
 813         P_MODE(R);
 814         P_MODE(W);
 815         P_MODE(X);
 816 #undef P_MODE
 817
 818         if (mode)
 819                 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
 820
 821         return printed;
 822 }
 823
 824 #define SCA_ACCMODE syscall_arg__scnprintf_access_mode
 825
 826 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
 827                                               struct syscall_arg *arg);
 828
 829 #define SCA_FILENAME syscall_arg__scnprintf_filename
 830
 831 static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
 832                                                 struct syscall_arg *arg)
 833 {
 834         bool show_prefix = arg->show_string_prefix;
 835         const char *prefix = "O_";
 836         int printed = 0, flags = arg->val;
 837
 838 #define P_FLAG(n) \
 839         if (flags & O_##n) { \
 840                 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
 841                 flags &= ~O_##n; \
 842         }
 843
 844         P_FLAG(CLOEXEC);
 845         P_FLAG(NONBLOCK);
 846 #undef P_FLAG
 847
 848         if (flags)
 849                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
 850
 851         return printed;
 852 }
 853
 854 #define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
 855
 856 #ifndef GRND_NONBLOCK
 857 #define GRND_NONBLOCK   0x0001
 858 #endif
 859 #ifndef GRND_RANDOM
 860 #define GRND_RANDOM     0x0002
 861 #endif
 862
 863 static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,
 864                                                    struct syscall_arg *arg)
 865 {
 866         bool show_prefix = arg->show_string_prefix;
 867         const char *prefix = "GRND_";
 868         int printed = 0, flags = arg->val;
 869
 870 #define P_FLAG(n) \
 871         if (flags & GRND_##n) { \
 872                 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
 873                 flags &= ~GRND_##n; \
 874         }
 875
 876         P_FLAG(RANDOM);
 877         P_FLAG(NONBLOCK);
 878 #undef P_FLAG
 879
 880         if (flags)
 881                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
 882
 883         return printed;
 884 }
 885
 886 #define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags
 887
 888 #define STRARRAY(name, array) \
 889           { .scnprintf  = SCA_STRARRAY, \
 890             .strtoul    = STUL_STRARRAY, \
 891             .parm       = &strarray__##array, }
 892
 893 #define STRARRAY_FLAGS(name, array) \
 894           { .scnprintf  = SCA_STRARRAY_FLAGS, \
 895             .strtoul    = STUL_STRARRAY_FLAGS, \
 896             .parm       = &strarray__##array, }
 897
 898 #include "trace/beauty/arch_errno_names.c"
 899 #include "trace/beauty/eventfd.c"
 900 #include "trace/beauty/futex_op.c"
 901 #include "trace/beauty/futex_val3.c"
 902 #include "trace/beauty/mmap.c"
 903 #include "trace/beauty/mode_t.c"
 904 #include "trace/beauty/msg_flags.c"
 905 #include "trace/beauty/open_flags.c"
 906 #include "trace/beauty/perf_event_open.c"
 907 #include "trace/beauty/pid.c"
 908 #include "trace/beauty/sched_policy.c"
 909 #include "trace/beauty/seccomp.c"
 910 #include "trace/beauty/signum.c"
 911 #include "trace/beauty/socket_type.c"
 912 #include "trace/beauty/waitid_options.c"
 913
 914 static struct syscall_fmt syscall_fmts[] = {
 915         { .name     = "access",
 916           .arg = { [1] = { .scnprintf = SCA_ACCMODE,  /* mode */ }, }, },
 917         { .name     = "arch_prctl",
 918           .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },
 919                    [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },
 920         { .name     = "bind",
 921           .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
 922                    [1] = { .scnprintf = SCA_SOCKADDR, /* umyaddr */ },
 923                    [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
 924         { .name     = "bpf",
 925           .arg = { [0] = STRARRAY(cmd, bpf_cmd), }, },
 926         { .name     = "brk",        .hexret = true,
 927           .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },
 928         { .name     = "clock_gettime",
 929           .arg = { [0] = STRARRAY(clk_id, clockid), }, },
 930         { .name     = "clone",      .errpid = true, .nr_args = 5,
 931           .arg = { [0] = { .name = "flags",         .scnprintf = SCA_CLONE_FLAGS, },
 932                    [1] = { .name = "child_stack",   .scnprintf = SCA_HEX, },
 933                    [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },
 934                    [3] = { .name = "child_tidptr",  .scnprintf = SCA_HEX, },
 935                    [4] = { .name = "tls",           .scnprintf = SCA_HEX, }, }, },
 936         { .name     = "close",
 937           .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },
 938         { .name     = "connect",
 939           .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
 940                    [1] = { .scnprintf = SCA_SOCKADDR, /* servaddr */ },
 941                    [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
 942         { .name     = "epoll_ctl",
 943           .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },
 944         { .name     = "eventfd2",
 945           .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },
 946         { .name     = "fchmodat",
 947           .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
 948         { .name     = "fchownat",
 949           .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
 950         { .name     = "fcntl",
 951           .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD,  /* cmd */
 952                            .strtoul   = STUL_STRARRAYS,
 953                            .parm      = &strarrays__fcntl_cmds_arrays,
 954                            .show_zero = true, },
 955                    [2] = { .scnprintf =  SCA_FCNTL_ARG, /* arg */ }, }, },
 956         { .name     = "flock",
 957           .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },
 958         { .name     = "fsconfig",
 959           .arg = { [1] = STRARRAY(cmd, fsconfig_cmds), }, },
 960         { .name     = "fsmount",
 961           .arg = { [1] = STRARRAY_FLAGS(flags, fsmount_flags),
 962                    [2] = { .scnprintf = SCA_FSMOUNT_ATTR_FLAGS, /* attr_flags */ }, }, },
 963         { .name     = "fspick",
 964           .arg = { [0] = { .scnprintf = SCA_FDAT,         /* dfd */ },
 965                    [1] = { .scnprintf = SCA_FILENAME,     /* path */ },
 966                    [2] = { .scnprintf = SCA_FSPICK_FLAGS, /* flags */ }, }, },
 967         { .name     = "fstat", .alias = "newfstat", },
 968         { .name     = "fstatat", .alias = "newfstatat", },
 969         { .name     = "futex",
 970           .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },
 971                    [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },
 972         { .name     = "futimesat",
 973           .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
 974         { .name     = "getitimer",
 975           .arg = { [0] = STRARRAY(which, itimers), }, },
 976         { .name     = "getpid",     .errpid = true, },
 977         { .name     = "getpgid",    .errpid = true, },
 978         { .name     = "getppid",    .errpid = true, },
 979         { .name     = "getrandom",
 980           .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },
 981         { .name     = "getrlimit",
 982           .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
 983         { .name     = "getsockopt",
 984           .arg = { [1] = STRARRAY(level, socket_level), }, },
 985         { .name     = "gettid",     .errpid = true, },
 986         { .name     = "ioctl",
 987           .arg = {
 988 #if defined(__i386__) || defined(__x86_64__)
 989 /*
 990  * FIXME: Make this available to all arches.
 991  */
 992                    [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },
 993                    [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
 994 #else
 995                    [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
 996 #endif
 997         { .name     = "kcmp",       .nr_args = 5,
 998           .arg = { [0] = { .name = "pid1",      .scnprintf = SCA_PID, },
 999                    [1] = { .name = "pid2",      .scnprintf = SCA_PID, },
1000                    [2] = { .name = "type",      .scnprintf = SCA_KCMP_TYPE, },
1001                    [3] = { .name = "idx1",      .scnprintf = SCA_KCMP_IDX, },
1002                    [4] = { .name = "idx2",      .scnprintf = SCA_KCMP_IDX, }, }, },
1003         { .name     = "keyctl",
1004           .arg = { [0] = STRARRAY(option, keyctl_options), }, },
1005         { .name     = "kill",
1006           .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1007         { .name     = "linkat",
1008           .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1009         { .name     = "lseek",
1010           .arg = { [2] = STRARRAY(whence, whences), }, },
1011         { .name     = "lstat", .alias = "newlstat", },
1012         { .name     = "madvise",
1013           .arg = { [0] = { .scnprintf = SCA_HEX,      /* start */ },
1014                    [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },
1015         { .name     = "mkdirat",
1016           .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1017         { .name     = "mknodat",
1018           .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1019         { .name     = "mmap",       .hexret = true,
1020 /* The standard mmap maps to old_mmap on s390x */
1021 #if defined(__s390x__)
1022         .alias = "old_mmap",
1023 #endif
1024           .arg = { [2] = { .scnprintf = SCA_MMAP_PROT,  /* prot */ },
1025                    [3] = { .scnprintf = SCA_MMAP_FLAGS, /* flags */
1026                            .strtoul   = STUL_STRARRAY_FLAGS,
1027                            .parm      = &strarray__mmap_flags, },
1028                    [5] = { .scnprintf = SCA_HEX,        /* offset */ }, }, },
1029         { .name     = "mount",
1030           .arg = { [0] = { .scnprintf = SCA_FILENAME, /* dev_name */ },
1031                    [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */
1032                            .mask_val  = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },
1033         { .name     = "move_mount",
1034           .arg = { [0] = { .scnprintf = SCA_FDAT,       /* from_dfd */ },
1035                    [1] = { .scnprintf = SCA_FILENAME, /* from_pathname */ },
1036                    [2] = { .scnprintf = SCA_FDAT,       /* to_dfd */ },
1037                    [3] = { .scnprintf = SCA_FILENAME, /* to_pathname */ },
1038                    [4] = { .scnprintf = SCA_MOVE_MOUNT_FLAGS, /* flags */ }, }, },
1039         { .name     = "mprotect",
1040           .arg = { [0] = { .scnprintf = SCA_HEX,        /* start */ },
1041                    [2] = { .scnprintf = SCA_MMAP_PROT,  /* prot */ }, }, },
1042         { .name     = "mq_unlink",
1043           .arg = { [0] = { .scnprintf = SCA_FILENAME, /* u_name */ }, }, },
1044         { .name     = "mremap",     .hexret = true,
1045           .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },
1046         { .name     = "name_to_handle_at",
1047           .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1048         { .name     = "newfstatat",
1049           .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1050         { .name     = "open",
1051           .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1052         { .name     = "open_by_handle_at",
1053           .arg = { [0] = { .scnprintf = SCA_FDAT,       /* dfd */ },
1054                    [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1055         { .name     = "openat",
1056           .arg = { [0] = { .scnprintf = SCA_FDAT,       /* dfd */ },
1057                    [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1058         { .name     = "perf_event_open",
1059           .arg = { [2] = { .scnprintf = SCA_INT,        /* cpu */ },
1060                    [3] = { .scnprintf = SCA_FD,         /* group_fd */ },
1061                    [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
1062         { .name     = "pipe2",
1063           .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },
1064         { .name     = "pkey_alloc",
1065           .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS,   /* access_rights */ }, }, },
1066         { .name     = "pkey_free",
1067           .arg = { [0] = { .scnprintf = SCA_INT,        /* key */ }, }, },
1068         { .name     = "pkey_mprotect",
1069           .arg = { [0] = { .scnprintf = SCA_HEX,        /* start */ },
1070                    [2] = { .scnprintf = SCA_MMAP_PROT,  /* prot */ },
1071                    [3] = { .scnprintf = SCA_INT,        /* pkey */ }, }, },
1072         { .name     = "poll", .timeout = true, },
1073         { .name     = "ppoll", .timeout = true, },
1074         { .name     = "prctl",
1075           .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */
1076                            .strtoul   = STUL_STRARRAY,
1077                            .parm      = &strarray__prctl_options, },
1078                    [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },
1079                    [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },
1080         { .name     = "pread", .alias = "pread64", },
1081         { .name     = "preadv", .alias = "pread", },
1082         { .name     = "prlimit64",
1083           .arg = { [1] = STRARRAY(resource, rlimit_resources), }, },
1084         { .name     = "pwrite", .alias = "pwrite64", },
1085         { .name     = "readlinkat",
1086           .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1087         { .name     = "recvfrom",
1088           .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1089         { .name     = "recvmmsg",
1090           .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1091         { .name     = "recvmsg",
1092           .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1093         { .name     = "renameat",
1094           .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
1095                    [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },
1096         { .name     = "renameat2",
1097           .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
1098                    [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },
1099                    [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },
1100         { .name     = "rt_sigaction",
1101           .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1102         { .name     = "rt_sigprocmask",
1103           .arg = { [0] = STRARRAY(how, sighow), }, },
1104         { .name     = "rt_sigqueueinfo",
1105           .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1106         { .name     = "rt_tgsigqueueinfo",
1107           .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1108         { .name     = "sched_setscheduler",
1109           .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },
1110         { .name     = "seccomp",
1111           .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP,    /* op */ },
1112                    [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },
1113         { .name     = "select", .timeout = true, },
1114         { .name     = "sendfile", .alias = "sendfile64", },
1115         { .name     = "sendmmsg",
1116           .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1117         { .name     = "sendmsg",
1118           .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1119         { .name     = "sendto",
1120           .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },
1121                    [4] = { .scnprintf = SCA_SOCKADDR, /* addr */ }, }, },
1122         { .name     = "set_tid_address", .errpid = true, },
1123         { .name     = "setitimer",
1124           .arg = { [0] = STRARRAY(which, itimers), }, },
1125         { .name     = "setrlimit",
1126           .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
1127         { .name     = "setsockopt",
1128           .arg = { [1] = STRARRAY(level, socket_level), }, },
1129         { .name     = "socket",
1130           .arg = { [0] = STRARRAY(family, socket_families),
1131                    [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1132                    [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1133         { .name     = "socketpair",
1134           .arg = { [0] = STRARRAY(family, socket_families),
1135                    [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1136                    [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1137         { .name     = "stat", .alias = "newstat", },
1138         { .name     = "statx",
1139           .arg = { [0] = { .scnprintf = SCA_FDAT,        /* fdat */ },
1140                    [2] = { .scnprintf = SCA_STATX_FLAGS, /* flags */ } ,
1141                    [3] = { .scnprintf = SCA_STATX_MASK,  /* mask */ }, }, },
1142         { .name     = "swapoff",
1143           .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
1144         { .name     = "swapon",
1145           .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
1146         { .name     = "symlinkat",
1147           .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1148         { .name     = "sync_file_range",
1149           .arg = { [3] = { .scnprintf = SCA_SYNC_FILE_RANGE_FLAGS, /* flags */ }, }, },
1150         { .name     = "tgkill",
1151           .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1152         { .name     = "tkill",
1153           .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1154         { .name     = "umount2", .alias = "umount",
1155           .arg = { [0] = { .scnprintf = SCA_FILENAME, /* name */ }, }, },
1156         { .name     = "uname", .alias = "newuname", },
1157         { .name     = "unlinkat",
1158           .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1159         { .name     = "utimensat",
1160           .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },
1161         { .name     = "wait4",      .errpid = true,
1162           .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
1163         { .name     = "waitid",     .errpid = true,
1164           .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
1165 };
1166
1167 static int syscall_fmt__cmp(const void *name, const void *fmtp)
1168 {
1169         const struct syscall_fmt *fmt = fmtp;
1170         return strcmp(name, fmt->name);
1171 }
1172
1173 static struct syscall_fmt *__syscall_fmt__find(struct syscall_fmt *fmts, const int nmemb, const char *name)
1174 {
1175         return bsearch(name, fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
1176 }
1177
1178 static struct syscall_fmt *syscall_fmt__find(const char *name)
1179 {
1180         const int nmemb = ARRAY_SIZE(syscall_fmts);
1181         return __syscall_fmt__find(syscall_fmts, nmemb, name);
1182 }
1183
1184 static struct syscall_fmt *__syscall_fmt__find_by_alias(struct syscall_fmt *fmts, const int nmemb, const char *alias)
1185 {
1186         int i;
1187
1188         for (i = 0; i < nmemb; ++i) {
1189                 if (fmts[i].alias && strcmp(fmts[i].alias, alias) == 0)
1190                         return &fmts[i];
1191         }
1192
1193         return NULL;
1194 }
1195
1196 static struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
1197 {
1198         const int nmemb = ARRAY_SIZE(syscall_fmts);
1199         return __syscall_fmt__find_by_alias(syscall_fmts, nmemb, alias);
1200 }
1201
1202 /*
1203  * is_exit: is this "exit" or "exit_group"?
1204  * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
1205  * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
1206  * nonexistent: Just a hole in the syscall table, syscall id not allocated
1207  */
1208 struct syscall {
1209         struct tep_event    *tp_format;
1210         int                 nr_args;
1211         int                 args_size;
1212         struct {
1213                 struct bpf_program *sys_enter,
1214                                    *sys_exit;
1215         }                   bpf_prog;
1216         bool                is_exit;
1217         bool                is_open;
1218         bool                nonexistent;
1219         struct tep_format_field *args;
1220         const char          *name;
1221         struct syscall_fmt  *fmt;
1222         struct syscall_arg_fmt *arg_fmt;
1223 };
1224
1225 /*
1226  * Must match what is in the BPF program:
1227  *
1228  * tools/perf/examples/bpf/augmented_raw_syscalls.c
1229  */
1230 struct bpf_map_syscall_entry {
1231         bool    enabled;
1232         u16     string_args_len[6];
1233 };
1234
1235 /*
1236  * We need to have this 'calculated' boolean because in some cases we really
1237  * don't know what is the duration of a syscall, for instance, when we start
1238  * a session and some threads are waiting for a syscall to finish, say 'poll',
1239  * in which case all we can do is to print "( ? ) for duration and for the
1240  * start timestamp.
1241  */
1242 static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
1243 {
1244         double duration = (double)t / NSEC_PER_MSEC;
1245         size_t printed = fprintf(fp, "(");
1246
1247         if (!calculated)
1248                 printed += fprintf(fp, "         ");
1249         else if (duration >= 1.0)
1250                 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1251         else if (duration >= 0.01)
1252                 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1253         else
1254                 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
1255         return printed + fprintf(fp, "): ");
1256 }
1257
1258 /**
1259  * filename.ptr: The filename char pointer that will be vfs_getname'd
1260  * filename.entry_str_pos: Where to insert the string translated from
1261  *                         filename.ptr by the vfs_getname tracepoint/kprobe.
1262  * ret_scnprintf: syscall args may set this to a different syscall return
1263  *                formatter, for instance, fcntl may return fds, file flags, etc.
1264  */
1265 struct thread_trace {
1266         u64               entry_time;
1267         bool              entry_pending;
1268         unsigned long     nr_events;
1269         unsigned long     pfmaj, pfmin;
1270         char              *entry_str;
1271         double            runtime_ms;
1272         size_t            (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
1273         struct {
1274                 unsigned long ptr;
1275                 short int     entry_str_pos;
1276                 bool          pending_open;
1277                 unsigned int  namelen;
1278                 char          *name;
1279         } filename;
1280         struct {
1281                 int           max;
1282                 struct file   *table;
1283         } files;
1284
1285         struct intlist *syscall_stats;
1286 };
1287
1288 static struct thread_trace *thread_trace__new(void)
1289 {
1290         struct thread_trace *ttrace =  zalloc(sizeof(struct thread_trace));
1291
1292         if (ttrace) {
1293                 ttrace->files.max = -1;
1294                 ttrace->syscall_stats = intlist__new(NULL);
1295         }
1296
1297         return ttrace;
1298 }
1299
1300 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
1301 {
1302         struct thread_trace *ttrace;
1303
1304         if (thread == NULL)
1305                 goto fail;
1306
1307         if (thread__priv(thread) == NULL)
1308                 thread__set_priv(thread, thread_trace__new());
1309
1310         if (thread__priv(thread) == NULL)
1311                 goto fail;
1312
1313         ttrace = thread__priv(thread);
1314         ++ttrace->nr_events;
1315
1316         return ttrace;
1317 fail:
1318         color_fprintf(fp, PERF_COLOR_RED,
1319                       "WARNING: not enough memory, dropping samples!\n");
1320         return NULL;
1321 }
1322
1323
1324 void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
1325                                     size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
1326 {
1327         struct thread_trace *ttrace = thread__priv(arg->thread);
1328
1329         ttrace->ret_scnprintf = ret_scnprintf;
1330 }
1331
1332 #define TRACE_PFMAJ             (1 << 0)
1333 #define TRACE_PFMIN             (1 << 1)
1334
1335 static const size_t trace__entry_str_size = 2048;
1336
1337 static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
1338 {
1339         if (fd < 0)
1340                 return NULL;
1341
1342         if (fd > ttrace->files.max) {
1343                 struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
1344
1345                 if (nfiles == NULL)
1346                         return NULL;
1347
1348                 if (ttrace->files.max != -1) {
1349                         memset(nfiles + ttrace->files.max + 1, 0,
1350                                (fd - ttrace->files.max) * sizeof(struct file));
1351                 } else {
1352                         memset(nfiles, 0, (fd + 1) * sizeof(struct file));
1353                 }
1354
1355                 ttrace->files.table = nfiles;
1356                 ttrace->files.max   = fd;
1357         }
1358
1359         return ttrace->files.table + fd;
1360 }
1361
1362 struct file *thread__files_entry(struct thread *thread, int fd)
1363 {
1364         return thread_trace__files_entry(thread__priv(thread), fd);
1365 }
1366
1367 static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1368 {
1369         struct thread_trace *ttrace = thread__priv(thread);
1370         struct file *file = thread_trace__files_entry(ttrace, fd);
1371
1372         if (file != NULL) {
1373                 struct stat st;
1374                 if (stat(pathname, &st) == 0)
1375                         file->dev_maj = major(st.st_rdev);
1376                 file->pathname = strdup(pathname);
1377                 if (file->pathname)
1378                         return 0;
1379         }
1380
1381         return -1;
1382 }
1383
1384 static int thread__read_fd_path(struct thread *thread, int fd)
1385 {
1386         char linkname[PATH_MAX], pathname[PATH_MAX];
1387         struct stat st;
1388         int ret;
1389
1390         if (thread->pid_ == thread->tid) {
1391                 scnprintf(linkname, sizeof(linkname),
1392                           "/proc/%d/fd/%d", thread->pid_, fd);
1393         } else {
1394                 scnprintf(linkname, sizeof(linkname),
1395                           "/proc/%d/task/%d/fd/%d", thread->pid_, thread->tid, fd);
1396         }
1397
1398         if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1399                 return -1;
1400
1401         ret = readlink(linkname, pathname, sizeof(pathname));
1402
1403         if (ret < 0 || ret > st.st_size)
1404                 return -1;
1405
1406         pathname[ret] = '\0';
1407         return trace__set_fd_pathname(thread, fd, pathname);
1408 }
1409
1410 static const char *thread__fd_path(struct thread *thread, int fd,
1411                                    struct trace *trace)
1412 {
1413         struct thread_trace *ttrace = thread__priv(thread);
1414
1415         if (ttrace == NULL || trace->fd_path_disabled)
1416                 return NULL;
1417
1418         if (fd < 0)
1419                 return NULL;
1420
1421         if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
1422                 if (!trace->live)
1423                         return NULL;
1424                 ++trace->stats.proc_getname;
1425                 if (thread__read_fd_path(thread, fd))
1426                         return NULL;
1427         }
1428
1429         return ttrace->files.table[fd].pathname;
1430 }
1431
1432 size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
1433 {
1434         int fd = arg->val;
1435         size_t printed = scnprintf(bf, size, "%d", fd);
1436         const char *path = thread__fd_path(arg->thread, fd, arg->trace);
1437
1438         if (path)
1439                 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1440
1441         return printed;
1442 }
1443
1444 size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1445 {
1446         size_t printed = scnprintf(bf, size, "%d", fd);
1447         struct thread *thread = machine__find_thread(trace->host, pid, pid);
1448
1449         if (thread) {
1450                 const char *path = thread__fd_path(thread, fd, trace);
1451
1452                 if (path)
1453                         printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1454
1455                 thread__put(thread);
1456         }
1457
1458         return printed;
1459 }
1460
1461 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1462                                               struct syscall_arg *arg)
1463 {
1464         int fd = arg->val;
1465         size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
1466         struct thread_trace *ttrace = thread__priv(arg->thread);
1467
1468         if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1469                 zfree(&ttrace->files.table[fd].pathname);
1470
1471         return printed;
1472 }
1473
1474 static void thread__set_filename_pos(struct thread *thread, const char *bf,
1475                                      unsigned long ptr)
1476 {
1477         struct thread_trace *ttrace = thread__priv(thread);
1478
1479         ttrace->filename.ptr = ptr;
1480         ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1481 }
1482
1483 static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1484 {
1485         struct augmented_arg *augmented_arg = arg->augmented.args;
1486         size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1487         /*
1488          * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls
1489          * we would have two strings, each prefixed by its size.
1490          */
1491         int consumed = sizeof(*augmented_arg) + augmented_arg->size;
1492
1493         arg->augmented.args = ((void *)arg->augmented.args) + consumed;
1494         arg->augmented.size -= consumed;
1495
1496         return printed;
1497 }
1498
1499 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1500                                               struct syscall_arg *arg)
1501 {
1502         unsigned long ptr = arg->val;
1503
1504         if (arg->augmented.args)
1505                 return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1506
1507         if (!arg->trace->vfs_getname)
1508                 return scnprintf(bf, size, "%#x", ptr);
1509
1510         thread__set_filename_pos(arg->thread, bf, ptr);
1511         return 0;
1512 }
1513
1514 static bool trace__filter_duration(struct trace *trace, double t)
1515 {
1516         return t < (trace->duration_filter * NSEC_PER_MSEC);
1517 }
1518
1519 static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1520 {
1521         double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1522
1523         return fprintf(fp, "%10.3f ", ts);
1524 }
1525
1526 /*
1527  * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1528  * using ttrace->entry_time for a thread that receives a sys_exit without
1529  * first having received a sys_enter ("poll" issued before tracing session
1530  * starts, lost sys_enter exit due to ring buffer overflow).
1531  */
1532 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1533 {
1534         if (tstamp > 0)
1535                 return __trace__fprintf_tstamp(trace, tstamp, fp);
1536
1537         return fprintf(fp, "         ? ");
1538 }
1539
1540 static pid_t workload_pid = -1;
1541 static bool done = false;
1542 static bool interrupted = false;
1543
1544 static void sighandler_interrupt(int sig __maybe_unused)
1545 {
1546         done = interrupted = true;
1547 }
1548
1549 static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
1550                             void *context __maybe_unused)
1551 {
1552         if (info->si_pid == workload_pid)
1553                 done = true;
1554 }
1555
1556 static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
1557 {
1558         size_t printed = 0;
1559
1560         if (trace->multiple_threads) {
1561                 if (trace->show_comm)
1562                         printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
1563                 printed += fprintf(fp, "%d ", thread->tid);
1564         }
1565
1566         return printed;
1567 }
1568
1569 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1570                                         u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1571 {
1572         size_t printed = 0;
1573
1574         if (trace->show_tstamp)
1575                 printed = trace__fprintf_tstamp(trace, tstamp, fp);
1576         if (trace->show_duration)
1577                 printed += fprintf_duration(duration, duration_calculated, fp);
1578         return printed + trace__fprintf_comm_tid(trace, thread, fp);
1579 }
1580
1581 static int trace__process_event(struct trace *trace, struct machine *machine,
1582                                 union perf_event *event, struct perf_sample *sample)
1583 {
1584         int ret = 0;
1585
1586         switch (event->header.type) {
1587         case PERF_RECORD_LOST:
1588                 color_fprintf(trace->output, PERF_COLOR_RED,
1589                               "LOST %" PRIu64 " events!\n", event->lost.lost);
1590                 ret = machine__process_lost_event(machine, event, sample);
1591                 break;
1592         default:
1593                 ret = machine__process_event(machine, event, sample);
1594                 break;
1595         }
1596
1597         return ret;
1598 }
1599
1600 static int trace__tool_process(struct perf_tool *tool,
1601                                union perf_event *event,
1602                                struct perf_sample *sample,
1603                                struct machine *machine)
1604 {
1605         struct trace *trace = container_of(tool, struct trace, tool);
1606         return trace__process_event(trace, machine, event, sample);
1607 }
1608
1609 static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1610 {
1611         struct machine *machine = vmachine;
1612
1613         if (machine->kptr_restrict_warned)
1614                 return NULL;
1615
1616         if (symbol_conf.kptr_restrict) {
1617                 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
1618                            "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
1619                            "Kernel samples will not be resolved.\n");
1620                 machine->kptr_restrict_warned = true;
1621                 return NULL;
1622         }
1623
1624         return machine__resolve_kernel_addr(vmachine, addrp, modp);
1625 }
1626
1627 static int trace__symbols_init(struct trace *trace, struct evlist *evlist)
1628 {
1629         int err = symbol__init(NULL);
1630
1631         if (err)
1632                 return err;
1633
1634         trace->host = machine__new_host();
1635         if (trace->host == NULL)
1636                 return -ENOMEM;
1637
1638         err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1639         if (err < 0)
1640                 goto out;
1641
1642         err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1643                                             evlist->core.threads, trace__tool_process,
1644                                             true, false, 1);
1645 out:
1646         if (err)
1647                 symbol__exit();
1648
1649         return err;
1650 }
1651
1652 static void trace__symbols__exit(struct trace *trace)
1653 {
1654         machine__exit(trace->host);
1655         trace->host = NULL;
1656
1657         symbol__exit();
1658 }
1659
1660 static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
1661 {
1662         int idx;
1663
1664         if (nr_args == 6 && sc->fmt && sc->fmt->nr_args != 0)
1665                 nr_args = sc->fmt->nr_args;
1666
1667         sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
1668         if (sc->arg_fmt == NULL)
1669                 return -1;
1670
1671         for (idx = 0; idx < nr_args; ++idx) {
1672                 if (sc->fmt)
1673                         sc->arg_fmt[idx] = sc->fmt->arg[idx];
1674         }
1675
1676         sc->nr_args = nr_args;
1677         return 0;
1678 }
1679
1680 static struct syscall_arg_fmt syscall_arg_fmts__by_name[] = {
1681         { .name = "msr",        .scnprintf = SCA_X86_MSR,         .strtoul = STUL_X86_MSR,         },
1682         { .name = "vector",     .scnprintf = SCA_X86_IRQ_VECTORS, .strtoul = STUL_X86_IRQ_VECTORS, },
1683 };
1684
1685 static int syscall_arg_fmt__cmp(const void *name, const void *fmtp)
1686 {
1687        const struct syscall_arg_fmt *fmt = fmtp;
1688        return strcmp(name, fmt->name);
1689 }
1690
1691 static struct syscall_arg_fmt *
1692 __syscall_arg_fmt__find_by_name(struct syscall_arg_fmt *fmts, const int nmemb, const char *name)
1693 {
1694        return bsearch(name, fmts, nmemb, sizeof(struct syscall_arg_fmt), syscall_arg_fmt__cmp);
1695 }
1696
1697 static struct syscall_arg_fmt *syscall_arg_fmt__find_by_name(const char *name)
1698 {
1699        const int nmemb = ARRAY_SIZE(syscall_arg_fmts__by_name);
1700        return __syscall_arg_fmt__find_by_name(syscall_arg_fmts__by_name, nmemb, name);
1701 }
1702
1703 static struct tep_format_field *
1704 syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field)
1705 {
1706         struct tep_format_field *last_field = NULL;
1707         int len;
1708
1709         for (; field; field = field->next, ++arg) {
1710                 last_field = field;
1711
1712                 if (arg->scnprintf)
1713                         continue;
1714
1715                 len = strlen(field->name);
1716
1717                 if (strcmp(field->type, "const char *") == 0 &&
1718                     ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||
1719                      strstr(field->name, "path") != NULL))
1720                         arg->scnprintf = SCA_FILENAME;
1721                 else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
1722                         arg->scnprintf = SCA_PTR;
1723                 else if (strcmp(field->type, "pid_t") == 0)
1724                         arg->scnprintf = SCA_PID;
1725                 else if (strcmp(field->type, "umode_t") == 0)
1726                         arg->scnprintf = SCA_MODE_T;
1727                 else if ((field->flags & TEP_FIELD_IS_ARRAY) && strstr(field->type, "char")) {
1728                         arg->scnprintf = SCA_CHAR_ARRAY;
1729                         arg->nr_entries = field->arraylen;
1730                 } else if ((strcmp(field->type, "int") == 0 ||
1731                           strcmp(field->type, "unsigned int") == 0 ||
1732                           strcmp(field->type, "long") == 0) &&
1733                          len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {
1734                         /*
1735                          * /sys/kernel/tracing/events/syscalls/sys_enter*
1736                          * egrep 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
1737                          * 65 int
1738                          * 23 unsigned int
1739                          * 7 unsigned long
1740                          */
1741                         arg->scnprintf = SCA_FD;
1742                } else {
1743                         struct syscall_arg_fmt *fmt = syscall_arg_fmt__find_by_name(field->name);
1744
1745                         if (fmt) {
1746                                 arg->scnprintf = fmt->scnprintf;
1747                                 arg->strtoul   = fmt->strtoul;
1748                         }
1749                 }
1750         }
1751
1752         return last_field;
1753 }
1754
1755 static int syscall__set_arg_fmts(struct syscall *sc)
1756 {
1757         struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args);
1758
1759         if (last_field)
1760                 sc->args_size = last_field->offset + last_field->size;
1761
1762         return 0;
1763 }
1764
1765 static int trace__read_syscall_info(struct trace *trace, int id)
1766 {
1767         char tp_name[128];
1768         struct syscall *sc;
1769         const char *name = syscalltbl__name(trace->sctbl, id);
1770
1771 #ifdef HAVE_SYSCALL_TABLE_SUPPORT
1772         if (trace->syscalls.table == NULL) {
1773                 trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));
1774                 if (trace->syscalls.table == NULL)
1775                         return -ENOMEM;
1776         }
1777 #else
1778         if (id > trace->sctbl->syscalls.max_id || (id == 0 && trace->syscalls.table == NULL)) {
1779                 // When using libaudit we don't know beforehand what is the max syscall id
1780                 struct syscall *table = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1781
1782                 if (table == NULL)
1783                         return -ENOMEM;
1784
1785                 // Need to memset from offset 0 and +1 members if brand new
1786                 if (trace->syscalls.table == NULL)
1787                         memset(table, 0, (id + 1) * sizeof(*sc));
1788                 else
1789                         memset(table + trace->sctbl->syscalls.max_id + 1, 0, (id - trace->sctbl->syscalls.max_id) * sizeof(*sc));
1790
1791                 trace->syscalls.table         = table;
1792                 trace->sctbl->syscalls.max_id = id;
1793         }
1794 #endif
1795         sc = trace->syscalls.table + id;
1796         if (sc->nonexistent)
1797                 return 0;
1798
1799         if (name == NULL) {
1800                 sc->nonexistent = true;
1801                 return 0;
1802         }
1803
1804         sc->name = name;
1805         sc->fmt  = syscall_fmt__find(sc->name);
1806
1807         snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
1808         sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1809
1810         if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
1811                 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
1812                 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1813         }
1814
1815         if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ? 6 : sc->tp_format->format.nr_fields))
1816                 return -ENOMEM;
1817
1818         if (IS_ERR(sc->tp_format))
1819                 return PTR_ERR(sc->tp_format);
1820
1821         sc->args = sc->tp_format->format.fields;
1822         /*
1823          * We need to check and discard the first variable '__syscall_nr'
1824          * or 'nr' that mean the syscall number. It is needless here.
1825          * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1826          */
1827         if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
1828                 sc->args = sc->args->next;
1829                 --sc->nr_args;
1830         }
1831
1832         sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1833         sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
1834
1835         return syscall__set_arg_fmts(sc);
1836 }
1837
1838 static int evsel__init_tp_arg_scnprintf(struct evsel *evsel)
1839 {
1840         struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
1841
1842         if (fmt != NULL) {
1843                 syscall_arg_fmt__init_array(fmt, evsel->tp_format->format.fields);
1844                 return 0;
1845         }
1846
1847         return -ENOMEM;
1848 }
1849
1850 static int intcmp(const void *a, const void *b)
1851 {
1852         const int *one = a, *another = b;
1853
1854         return *one - *another;
1855 }
1856
1857 static int trace__validate_ev_qualifier(struct trace *trace)
1858 {
1859         int err = 0;
1860         bool printed_invalid_prefix = false;
1861         struct str_node *pos;
1862         size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);
1863
1864         trace->ev_qualifier_ids.entries = malloc(nr_allocated *
1865                                                  sizeof(trace->ev_qualifier_ids.entries[0]));
1866
1867         if (trace->ev_qualifier_ids.entries == NULL) {
1868                 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1869                        trace->output);
1870                 err = -EINVAL;
1871                 goto out;
1872         }
1873
1874         strlist__for_each_entry(pos, trace->ev_qualifier) {
1875                 const char *sc = pos->s;
1876                 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
1877
1878                 if (id < 0) {
1879                         id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1880                         if (id >= 0)
1881                                 goto matches;
1882
1883                         if (!printed_invalid_prefix) {
1884                                 pr_debug("Skipping unknown syscalls: ");
1885                                 printed_invalid_prefix = true;
1886                         } else {
1887                                 pr_debug(", ");
1888                         }
1889
1890                         pr_debug("%s", sc);
1891                         continue;
1892                 }
1893 matches:
1894                 trace->ev_qualifier_ids.entries[nr_used++] = id;
1895                 if (match_next == -1)
1896                         continue;
1897
1898                 while (1) {
1899                         id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1900                         if (id < 0)
1901                                 break;
1902                         if (nr_allocated == nr_used) {
1903                                 void *entries;
1904
1905                                 nr_allocated += 8;
1906                                 entries = realloc(trace->ev_qualifier_ids.entries,
1907                                                   nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1908                                 if (entries == NULL) {
1909                                         err = -ENOMEM;
1910                                         fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1911                                         goto out_free;
1912                                 }
1913                                 trace->ev_qualifier_ids.entries = entries;
1914                         }
1915                         trace->ev_qualifier_ids.entries[nr_used++] = id;
1916                 }
1917         }
1918
1919         trace->ev_qualifier_ids.nr = nr_used;
1920         qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);
1921 out:
1922         if (printed_invalid_prefix)
1923                 pr_debug("\n");
1924         return err;
1925 out_free:
1926         zfree(&trace->ev_qualifier_ids.entries);
1927         trace->ev_qualifier_ids.nr = 0;
1928         goto out;
1929 }
1930
1931 static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
1932 {
1933         bool in_ev_qualifier;
1934
1935         if (trace->ev_qualifier_ids.nr == 0)
1936                 return true;
1937
1938         in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
1939                                   trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
1940
1941         if (in_ev_qualifier)
1942                return !trace->not_ev_qualifier;
1943
1944         return trace->not_ev_qualifier;
1945 }
1946
1947 /*
1948  * args is to be interpreted as a series of longs but we need to handle
1949  * 8-byte unaligned accesses. args points to raw_data within the event
1950  * and raw_data is guaranteed to be 8-byte unaligned because it is
1951  * preceded by raw_size which is a u32. So we need to copy args to a temp
1952  * variable to read it. Most notably this avoids extended load instructions
1953  * on unaligned addresses
1954  */
1955 unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
1956 {
1957         unsigned long val;
1958         unsigned char *p = arg->args + sizeof(unsigned long) * idx;
1959
1960         memcpy(&val, p, sizeof(val));
1961         return val;
1962 }
1963
1964 static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
1965                                       struct syscall_arg *arg)
1966 {
1967         if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
1968                 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
1969
1970         return scnprintf(bf, size, "arg%d: ", arg->idx);
1971 }
1972
1973 /*
1974  * Check if the value is in fact zero, i.e. mask whatever needs masking, such
1975  * as mount 'flags' argument that needs ignoring some magic flag, see comment
1976  * in tools/perf/trace/beauty/mount_flags.c
1977  */
1978 static unsigned long syscall_arg_fmt__mask_val(struct syscall_arg_fmt *fmt, struct syscall_arg *arg, unsigned long val)
1979 {
1980         if (fmt && fmt->mask_val)
1981                 return fmt->mask_val(arg, val);
1982
1983         return val;
1984 }
1985
1986 static size_t syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt *fmt, char *bf, size_t size,
1987                                              struct syscall_arg *arg, unsigned long val)
1988 {
1989         if (fmt && fmt->scnprintf) {
1990                 arg->val = val;
1991                 if (fmt->parm)
1992                         arg->parm = fmt->parm;
1993                 return fmt->scnprintf(bf, size, arg);
1994         }
1995         return scnprintf(bf, size, "%ld", val);
1996 }
1997
1998 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
1999                                       unsigned char *args, void *augmented_args, int augmented_args_size,
2000                                       struct trace *trace, struct thread *thread)
2001 {
2002         size_t printed = 0;
2003         unsigned long val;
2004         u8 bit = 1;
2005         struct syscall_arg arg = {
2006                 .args   = args,
2007                 .augmented = {
2008                         .size = augmented_args_size,
2009                         .args = augmented_args,
2010                 },
2011                 .idx    = 0,
2012                 .mask   = 0,
2013                 .trace  = trace,
2014                 .thread = thread,
2015                 .show_string_prefix = trace->show_string_prefix,
2016         };
2017         struct thread_trace *ttrace = thread__priv(thread);
2018
2019         /*
2020          * Things like fcntl will set this in its 'cmd' formatter to pick the
2021          * right formatter for the return value (an fd? file flags?), which is
2022          * not needed for syscalls that always return a given type, say an fd.
2023          */
2024         ttrace->ret_scnprintf = NULL;
2025
2026         if (sc->args != NULL) {
2027                 struct tep_format_field *field;
2028
2029                 for (field = sc->args; field;
2030                      field = field->next, ++arg.idx, bit <<= 1) {
2031                         if (arg.mask & bit)
2032                                 continue;
2033
2034                         arg.fmt = &sc->arg_fmt[arg.idx];
2035                         val = syscall_arg__val(&arg, arg.idx);
2036                         /*
2037                          * Some syscall args need some mask, most don't and
2038                          * return val untouched.
2039                          */
2040                         val = syscall_arg_fmt__mask_val(&sc->arg_fmt[arg.idx], &arg, val);
2041
2042                         /*
2043                          * Suppress this argument if its value is zero and
2044                          * and we don't have a string associated in an
2045                          * strarray for it.
2046                          */
2047                         if (val == 0 &&
2048                             !trace->show_zeros &&
2049                             !(sc->arg_fmt &&
2050                               (sc->arg_fmt[arg.idx].show_zero ||
2051                                sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY ||
2052                                sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) &&
2053                               sc->arg_fmt[arg.idx].parm))
2054                                 continue;
2055
2056                         printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2057
2058                         if (trace->show_arg_names)
2059                                 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2060
2061                         printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx],
2062                                                                   bf + printed, size - printed, &arg, val);
2063                 }
2064         } else if (IS_ERR(sc->tp_format)) {
2065                 /*
2066                  * If we managed to read the tracepoint /format file, then we
2067                  * may end up not having any args, like with gettid(), so only
2068                  * print the raw args when we didn't manage to read it.
2069                  */
2070                 while (arg.idx < sc->nr_args) {
2071                         if (arg.mask & bit)
2072                                 goto next_arg;
2073                         val = syscall_arg__val(&arg, arg.idx);
2074                         if (printed)
2075                                 printed += scnprintf(bf + printed, size - printed, ", ");
2076                         printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
2077                         printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], bf + printed, size - printed, &arg, val);
2078 next_arg:
2079                         ++arg.idx;
2080                         bit <<= 1;
2081                 }
2082         }
2083
2084         return printed;
2085 }
2086
2087 typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,
2088                                   union perf_event *event,
2089                                   struct perf_sample *sample);
2090
2091 static struct syscall *trace__syscall_info(struct trace *trace,
2092                                            struct evsel *evsel, int id)
2093 {
2094         int err = 0;
2095
2096         if (id < 0) {
2097
2098                 /*
2099                  * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
2100                  * before that, leaving at a higher verbosity level till that is
2101                  * explained. Reproduced with plain ftrace with:
2102                  *
2103                  * echo 1 > /t/events/raw_syscalls/sys_exit/enable
2104                  * grep "NR -1 " /t/trace_pipe
2105                  *
2106                  * After generating some load on the machine.
2107                  */
2108                 if (verbose > 1) {
2109                         static u64 n;
2110                         fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
2111                                 id, evsel__name(evsel), ++n);
2112                 }
2113                 return NULL;
2114         }
2115
2116         err = -EINVAL;
2117
2118 #ifdef HAVE_SYSCALL_TABLE_SUPPORT
2119         if (id > trace->sctbl->syscalls.max_id) {
2120 #else
2121         if (id >= trace->sctbl->syscalls.max_id) {
2122                 /*
2123                  * With libaudit we don't know beforehand what is the max_id,
2124                  * so we let trace__read_syscall_info() figure that out as we
2125                  * go on reading syscalls.
2126                  */
2127                 err = trace__read_syscall_info(trace, id);
2128                 if (err)
2129 #endif
2130                 goto out_cant_read;
2131         }
2132
2133         if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&
2134             (err = trace__read_syscall_info(trace, id)) != 0)
2135                 goto out_cant_read;
2136
2137         if (trace->syscalls.table[id].name == NULL) {
2138                 if (trace->syscalls.table[id].nonexistent)
2139                         return NULL;
2140                 goto out_cant_read;
2141         }
2142
2143         return &trace->syscalls.table[id];
2144
2145 out_cant_read:
2146         if (verbose > 0) {
2147                 char sbuf[STRERR_BUFSIZE];
2148                 fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));
2149                 if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)
2150                         fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
2151                 fputs(" information\n", trace->output);
2152         }
2153         return NULL;
2154 }
2155
2156 struct syscall_stats {
2157         struct stats stats;
2158         u64          nr_failures;
2159         int          max_errno;
2160         u32          *errnos;
2161 };
2162
2163 static void thread__update_stats(struct thread *thread, struct thread_trace *ttrace,
2164                                  int id, struct perf_sample *sample, long err, bool errno_summary)
2165 {
2166         struct int_node *inode;
2167         struct syscall_stats *stats;
2168         u64 duration = 0;
2169
2170         inode = intlist__findnew(ttrace->syscall_stats, id);
2171         if (inode == NULL)
2172                 return;
2173
2174         stats = inode->priv;
2175         if (stats == NULL) {
2176                 stats = malloc(sizeof(*stats));
2177                 if (stats == NULL)
2178                         return;
2179
2180                 stats->nr_failures = 0;
2181                 stats->max_errno   = 0;
2182                 stats->errnos      = NULL;
2183                 init_stats(&stats->stats);
2184                 inode->priv = stats;
2185         }
2186
2187         if (ttrace->entry_time && sample->time > ttrace->entry_time)
2188                 duration = sample->time - ttrace->entry_time;
2189
2190         update_stats(&stats->stats, duration);
2191
2192         if (err < 0) {
2193                 ++stats->nr_failures;
2194
2195                 if (!errno_summary)
2196                         return;
2197
2198                 err = -err;
2199                 if (err > stats->max_errno) {
2200                         u32 *new_errnos = realloc(stats->errnos, err * sizeof(u32));
2201
2202                         if (new_errnos) {
2203                                 memset(new_errnos + stats->max_errno, 0, (err - stats->max_errno) * sizeof(u32));
2204                         } else {
2205                                 pr_debug("Not enough memory for errno stats for thread \"%s\"(%d/%d), results will be incomplete\n",
2206                                          thread__comm_str(thread), thread->pid_, thread->tid);
2207                                 return;
2208                         }
2209
2210                         stats->errnos = new_errnos;
2211                         stats->max_errno = err;
2212                 }
2213
2214                 ++stats->errnos[err - 1];
2215         }
2216 }
2217
2218 static int trace__printf_interrupted_entry(struct trace *trace)
2219 {
2220         struct thread_trace *ttrace;
2221         size_t printed;
2222         int len;
2223
2224         if (trace->failure_only || trace->current == NULL)
2225                 return 0;
2226
2227         ttrace = thread__priv(trace->current);
2228
2229         if (!ttrace->entry_pending)
2230                 return 0;
2231
2232         printed  = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
2233         printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
2234
2235         if (len < trace->args_alignment - 4)
2236                 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
2237
2238         printed += fprintf(trace->output, " ...\n");
2239
2240         ttrace->entry_pending = false;
2241         ++trace->nr_events_printed;
2242
2243         return printed;
2244 }
2245
2246 static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,
2247                                  struct perf_sample *sample, struct thread *thread)
2248 {
2249         int printed = 0;
2250
2251         if (trace->print_sample) {
2252                 double ts = (double)sample->time / NSEC_PER_MSEC;
2253
2254                 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
2255                                    evsel__name(evsel), ts,
2256                                    thread__comm_str(thread),
2257                                    sample->pid, sample->tid, sample->cpu);
2258         }
2259
2260         return printed;
2261 }
2262
2263 static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
2264 {
2265         void *augmented_args = NULL;
2266         /*
2267          * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
2268          * and there we get all 6 syscall args plus the tracepoint common fields
2269          * that gets calculated at the start and the syscall_nr (another long).
2270          * So we check if that is the case and if so don't look after the
2271          * sc->args_size but always after the full raw_syscalls:sys_enter payload,
2272          * which is fixed.
2273          *
2274          * We'll revisit this later to pass s->args_size to the BPF augmenter
2275          * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
2276          * copies only what we need for each syscall, like what happens when we
2277          * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
2278          * traffic to just what is needed for each syscall.
2279          */
2280         int args_size = raw_augmented_args_size ?: sc->args_size;
2281
2282         *augmented_args_size = sample->raw_size - args_size;
2283         if (*augmented_args_size > 0)
2284                 augmented_args = sample->raw_data + args_size;
2285
2286         return augmented_args;
2287 }
2288
2289 static void syscall__exit(struct syscall *sc)
2290 {
2291         if (!sc)
2292                 return;
2293
2294         free(sc->arg_fmt);
2295 }
2296
2297 static int trace__sys_enter(struct trace *trace, struct evsel *evsel,
2298                             union perf_event *event __maybe_unused,
2299                             struct perf_sample *sample)
2300 {
2301         char *msg;
2302         void *args;
2303         int printed = 0;
2304         struct thread *thread;
2305         int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2306         int augmented_args_size = 0;
2307         void *augmented_args = NULL;
2308         struct syscall *sc = trace__syscall_info(trace, evsel, id);
2309         struct thread_trace *ttrace;
2310
2311         if (sc == NULL)
2312                 return -1;
2313
2314         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2315         ttrace = thread__trace(thread, trace->output);
2316         if (ttrace == NULL)
2317                 goto out_put;
2318
2319         trace__fprintf_sample(trace, evsel, sample, thread);
2320
2321         args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2322
2323         if (ttrace->entry_str == NULL) {
2324                 ttrace->entry_str = malloc(trace__entry_str_size);
2325                 if (!ttrace->entry_str)
2326                         goto out_put;
2327         }
2328
2329         if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
2330                 trace__printf_interrupted_entry(trace);
2331         /*
2332          * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
2333          * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
2334          * this breaks syscall__augmented_args() check for augmented args, as we calculate
2335          * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
2336          * so when handling, say the openat syscall, we end up getting 6 args for the
2337          * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
2338          * thinking that the extra 2 u64 args are the augmented filename, so just check
2339          * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
2340          */
2341         if (evsel != trace->syscalls.events.sys_enter)
2342                 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2343         ttrace->entry_time = sample->time;
2344         msg = ttrace->entry_str;
2345         printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
2346
2347         printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
2348                                            args, augmented_args, augmented_args_size, trace, thread);
2349
2350         if (sc->is_exit) {
2351                 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
2352                         int alignment = 0;
2353
2354                         trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
2355                         printed = fprintf(trace->output, "%s)", ttrace->entry_str);
2356                         if (trace->args_alignment > printed)
2357                                 alignment = trace->args_alignment - printed;
2358                         fprintf(trace->output, "%*s= ?\n", alignment, " ");
2359                 }
2360         } else {
2361                 ttrace->entry_pending = true;
2362                 /* See trace__vfs_getname & trace__sys_exit */
2363                 ttrace->filename.pending_open = false;
2364         }
2365
2366         if (trace->current != thread) {
2367                 thread__put(trace->current);
2368                 trace->current = thread__get(thread);
2369         }
2370         err = 0;
2371 out_put:
2372         thread__put(thread);
2373         return err;
2374 }
2375
2376 static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,
2377                                     struct perf_sample *sample)
2378 {
2379         struct thread_trace *ttrace;
2380         struct thread *thread;
2381         int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2382         struct syscall *sc = trace__syscall_info(trace, evsel, id);
2383         char msg[1024];
2384         void *args, *augmented_args = NULL;
2385         int augmented_args_size;
2386
2387         if (sc == NULL)
2388                 return -1;
2389
2390         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2391         ttrace = thread__trace(thread, trace->output);
2392         /*
2393          * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
2394          * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
2395          */
2396         if (ttrace == NULL)
2397                 goto out_put;
2398
2399         args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2400         augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2401         syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
2402         fprintf(trace->output, "%s", msg);
2403         err = 0;
2404 out_put:
2405         thread__put(thread);
2406         return err;
2407 }
2408
2409 static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,
2410                                     struct perf_sample *sample,
2411                                     struct callchain_cursor *cursor)
2412 {
2413         struct addr_location al;
2414         int max_stack = evsel->core.attr.sample_max_stack ?
2415                         evsel->core.attr.sample_max_stack :
2416                         trace->max_stack;
2417         int err;
2418
2419         if (machine__resolve(trace->host, &al, sample) < 0)
2420                 return -1;
2421
2422         err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
2423         addr_location__put(&al);
2424         return err;
2425 }
2426
2427 static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
2428 {
2429         /* TODO: user-configurable print_opts */
2430         const unsigned int print_opts = EVSEL__PRINT_SYM |
2431                                         EVSEL__PRINT_DSO |
2432                                         EVSEL__PRINT_UNKNOWN_AS_ADDR;
2433
2434         return sample__fprintf_callchain(sample, 38, print_opts, &callchain_cursor, symbol_conf.bt_stop_list, trace->output);
2435 }
2436
2437 static const char *errno_to_name(struct evsel *evsel, int err)
2438 {
2439         struct perf_env *env = evsel__env(evsel);
2440         const char *arch_name = perf_env__arch(env);
2441
2442         return arch_syscalls__strerrno(arch_name, err);
2443 }
2444
2445 static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
2446                            union perf_event *event __maybe_unused,
2447                            struct perf_sample *sample)
2448 {
2449         long ret;
2450         u64 duration = 0;
2451         bool duration_calculated = false;
2452         struct thread *thread;
2453         int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
2454         int alignment = trace->args_alignment;
2455         struct syscall *sc = trace__syscall_info(trace, evsel, id);
2456         struct thread_trace *ttrace;
2457
2458         if (sc == NULL)
2459                 return -1;
2460
2461         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2462         ttrace = thread__trace(thread, trace->output);
2463         if (ttrace == NULL)
2464                 goto out_put;
2465
2466         trace__fprintf_sample(trace, evsel, sample, thread);
2467
2468         ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
2469
2470         if (trace->summary)
2471                 thread__update_stats(thread, ttrace, id, sample, ret, trace->errno_summary);
2472
2473         if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
2474                 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2475                 ttrace->filename.pending_open = false;
2476                 ++trace->stats.vfs_getname;
2477         }
2478
2479         if (ttrace->entry_time) {
2480                 duration = sample->time - ttrace->entry_time;
2481                 if (trace__filter_duration(trace, duration))
2482                         goto out;
2483                 duration_calculated = true;
2484         } else if (trace->duration_filter)
2485                 goto out;
2486
2487         if (sample->callchain) {
2488                 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2489                 if (callchain_ret == 0) {
2490                         if (callchain_cursor.nr < trace->min_stack)
2491                                 goto out;
2492                         callchain_ret = 1;
2493                 }
2494         }
2495
2496         if (trace->summary_only || (ret >= 0 && trace->failure_only))
2497                 goto out;
2498
2499         trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
2500
2501         if (ttrace->entry_pending) {
2502                 printed = fprintf(trace->output, "%s", ttrace->entry_str);
2503         } else {
2504                 printed += fprintf(trace->output, " ... [");
2505                 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
2506                 printed += 9;
2507                 printed += fprintf(trace->output, "]: %s()", sc->name);
2508         }
2509
2510         printed++; /* the closing ')' */
2511
2512         if (alignment > printed)
2513                 alignment -= printed;
2514         else
2515                 alignment = 0;
2516
2517         fprintf(trace->output, ")%*s= ", alignment, " ");
2518
2519         if (sc->fmt == NULL) {
2520                 if (ret < 0)
2521                         goto errno_print;
2522 signed_print:
2523                 fprintf(trace->output, "%ld", ret);
2524         } else if (ret < 0) {
2525 errno_print: {
2526                 char bf[STRERR_BUFSIZE];
2527                 const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
2528                            *e = errno_to_name(evsel, -ret);
2529
2530                 fprintf(trace->output, "-1 %s (%s)", e, emsg);
2531         }
2532         } else if (ret == 0 && sc->fmt->timeout)
2533                 fprintf(trace->output, "0 (Timeout)");
2534         else if (ttrace->ret_scnprintf) {
2535                 char bf[1024];
2536                 struct syscall_arg arg = {
2537                         .val    = ret,
2538                         .thread = thread,
2539                         .trace  = trace,
2540                 };
2541                 ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
2542                 ttrace->ret_scnprintf = NULL;
2543                 fprintf(trace->output, "%s", bf);
2544         } else if (sc->fmt->hexret)
2545                 fprintf(trace->output, "%#lx", ret);
2546         else if (sc->fmt->errpid) {
2547                 struct thread *child = machine__find_thread(trace->host, ret, ret);
2548
2549                 if (child != NULL) {
2550                         fprintf(trace->output, "%ld", ret);
2551                         if (child->comm_set)
2552                                 fprintf(trace->output, " (%s)", thread__comm_str(child));
2553                         thread__put(child);
2554                 }
2555         } else
2556                 goto signed_print;
2557
2558         fputc('\n', trace->output);
2559
2560         /*
2561          * We only consider an 'event' for the sake of --max-events a non-filtered
2562          * sys_enter + sys_exit and other tracepoint events.
2563          */
2564         if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2565                 interrupted = true;
2566
2567         if (callchain_ret > 0)
2568                 trace__fprintf_callchain(trace, sample);
2569         else if (callchain_ret < 0)
2570                 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2571 out:
2572         ttrace->entry_pending = false;
2573         err = 0;
2574 out_put:
2575         thread__put(thread);
2576         return err;
2577 }
2578
2579 static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,
2580                               union perf_event *event __maybe_unused,
2581                               struct perf_sample *sample)
2582 {
2583         struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2584         struct thread_trace *ttrace;
2585         size_t filename_len, entry_str_len, to_move;
2586         ssize_t remaining_space;
2587         char *pos;
2588         const char *filename = evsel__rawptr(evsel, sample, "pathname");
2589
2590         if (!thread)
2591                 goto out;
2592
2593         ttrace = thread__priv(thread);
2594         if (!ttrace)
2595                 goto out_put;
2596
2597         filename_len = strlen(filename);
2598         if (filename_len == 0)
2599                 goto out_put;
2600
2601         if (ttrace->filename.namelen < filename_len) {
2602                 char *f = realloc(ttrace->filename.name, filename_len + 1);
2603
2604                 if (f == NULL)
2605                         goto out_put;
2606
2607                 ttrace->filename.namelen = filename_len;
2608                 ttrace->filename.name = f;
2609         }
2610
2611         strcpy(ttrace->filename.name, filename);
2612         ttrace->filename.pending_open = true;
2613
2614         if (!ttrace->filename.ptr)
2615                 goto out_put;
2616
2617         entry_str_len = strlen(ttrace->entry_str);
2618         remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2619         if (remaining_space <= 0)
2620                 goto out_put;
2621
2622         if (filename_len > (size_t)remaining_space) {
2623                 filename += filename_len - remaining_space;
2624                 filename_len = remaining_space;
2625         }
2626
2627         to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2628         pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2629         memmove(pos + filename_len, pos, to_move);
2630         memcpy(pos, filename, filename_len);
2631
2632         ttrace->filename.ptr = 0;
2633         ttrace->filename.entry_str_pos = 0;
2634 out_put:
2635         thread__put(thread);
2636 out:
2637         return 0;
2638 }
2639
2640 static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,
2641                                      union perf_event *event __maybe_unused,
2642                                      struct perf_sample *sample)
2643 {
2644         u64 runtime = evsel__intval(evsel, sample, "runtime");
2645         double runtime_ms = (double)runtime / NSEC_PER_MSEC;
2646         struct thread *thread = machine__findnew_thread(trace->host,
2647                                                         sample->pid,
2648                                                         sample->tid);
2649         struct thread_trace *ttrace = thread__trace(thread, trace->output);
2650
2651         if (ttrace == NULL)
2652                 goto out_dump;
2653
2654         ttrace->runtime_ms += runtime_ms;
2655         trace->runtime_ms += runtime_ms;
2656 out_put:
2657         thread__put(thread);
2658         return 0;
2659
2660 out_dump:
2661         fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
2662                evsel->name,
2663                evsel__strval(evsel, sample, "comm"),
2664                (pid_t)evsel__intval(evsel, sample, "pid"),
2665                runtime,
2666                evsel__intval(evsel, sample, "vruntime"));
2667         goto out_put;
2668 }
2669
2670 static int bpf_output__printer(enum binary_printer_ops op,
2671                                unsigned int val, void *extra __maybe_unused, FILE *fp)
2672 {
2673         unsigned char ch = (unsigned char)val;
2674
2675         switch (op) {
2676         case BINARY_PRINT_CHAR_DATA:
2677                 return fprintf(fp, "%c", isprint(ch) ? ch : '.');
2678         case BINARY_PRINT_DATA_BEGIN:
2679         case BINARY_PRINT_LINE_BEGIN:
2680         case BINARY_PRINT_ADDR:
2681         case BINARY_PRINT_NUM_DATA:
2682         case BINARY_PRINT_NUM_PAD:
2683         case BINARY_PRINT_SEP:
2684         case BINARY_PRINT_CHAR_PAD:
2685         case BINARY_PRINT_LINE_END:
2686         case BINARY_PRINT_DATA_END:
2687         default:
2688                 break;
2689         }
2690
2691         return 0;
2692 }
2693
2694 static void bpf_output__fprintf(struct trace *trace,
2695                                 struct perf_sample *sample)
2696 {
2697         binary__fprintf(sample->raw_data, sample->raw_size, 8,
2698                         bpf_output__printer, NULL, trace->output);
2699         ++trace->nr_events_printed;
2700 }
2701
2702 static size_t trace__fprintf_tp_fields(struct trace *trace, struct evsel *evsel, struct perf_sample *sample,
2703                                        struct thread *thread, void *augmented_args, int augmented_args_size)
2704 {
2705         char bf[2048];
2706         size_t size = sizeof(bf);
2707         struct tep_format_field *field = evsel->tp_format->format.fields;
2708         struct syscall_arg_fmt *arg = __evsel__syscall_arg_fmt(evsel);
2709         size_t printed = 0;
2710         unsigned long val;
2711         u8 bit = 1;
2712         struct syscall_arg syscall_arg = {
2713                 .augmented = {
2714                         .size = augmented_args_size,
2715                         .args = augmented_args,
2716                 },
2717                 .idx    = 0,
2718                 .mask   = 0,
2719                 .trace  = trace,
2720                 .thread = thread,
2721                 .show_string_prefix = trace->show_string_prefix,
2722         };
2723
2724         for (; field && arg; field = field->next, ++syscall_arg.idx, bit <<= 1, ++arg) {
2725                 if (syscall_arg.mask & bit)
2726                         continue;
2727
2728                 syscall_arg.len = 0;
2729                 syscall_arg.fmt = arg;
2730                 if (field->flags & TEP_FIELD_IS_ARRAY) {
2731                         int offset = field->offset;
2732
2733                         if (field->flags & TEP_FIELD_IS_DYNAMIC) {
2734                                 offset = format_field__intval(field, sample, evsel->needs_swap);
2735                                 syscall_arg.len = offset >> 16;
2736                                 offset &= 0xffff;
2737                                 if (field->flags & TEP_FIELD_IS_RELATIVE)
2738                                         offset += field->offset + field->size;
2739                         }
2740
2741                         val = (uintptr_t)(sample->raw_data + offset);
2742                 } else
2743                         val = format_field__intval(field, sample, evsel->needs_swap);
2744                 /*
2745                  * Some syscall args need some mask, most don't and
2746                  * return val untouched.
2747                  */
2748                 val = syscall_arg_fmt__mask_val(arg, &syscall_arg, val);
2749
2750                 /*
2751                  * Suppress this argument if its value is zero and
2752                  * and we don't have a string associated in an
2753                  * strarray for it.
2754                  */
2755                 if (val == 0 &&
2756                     !trace->show_zeros &&
2757                     !((arg->show_zero ||
2758                        arg->scnprintf == SCA_STRARRAY ||
2759                        arg->scnprintf == SCA_STRARRAYS) &&
2760                       arg->parm))
2761                         continue;
2762
2763                 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2764
2765                 /*
2766                  * XXX Perhaps we should have a show_tp_arg_names,
2767                  * leaving show_arg_names just for syscalls?
2768                  */
2769                 if (1 || trace->show_arg_names)
2770                         printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2771
2772                 printed += syscall_arg_fmt__scnprintf_val(arg, bf + printed, size - printed, &syscall_arg, val);
2773         }
2774
2775         return printed + fprintf(trace->output, "%s", bf);
2776 }
2777
2778 static int trace__event_handler(struct trace *trace, struct evsel *evsel,
2779                                 union perf_event *event __maybe_unused,
2780                                 struct perf_sample *sample)
2781 {
2782         struct thread *thread;
2783         int callchain_ret = 0;
2784         /*
2785          * Check if we called perf_evsel__disable(evsel) due to, for instance,
2786          * this event's max_events having been hit and this is an entry coming
2787          * from the ring buffer that we should discard, since the max events
2788          * have already been considered/printed.
2789          */
2790         if (evsel->disabled)
2791                 return 0;
2792
2793         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2794
2795         if (sample->callchain) {
2796                 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2797                 if (callchain_ret == 0) {
2798                         if (callchain_cursor.nr < trace->min_stack)
2799                                 goto out;
2800                         callchain_ret = 1;
2801                 }
2802         }
2803
2804         trace__printf_interrupted_entry(trace);
2805         trace__fprintf_tstamp(trace, sample->time, trace->output);
2806
2807         if (trace->trace_syscalls && trace->show_duration)
2808                 fprintf(trace->output, "(         ): ");
2809
2810         if (thread)
2811                 trace__fprintf_comm_tid(trace, thread, trace->output);
2812
2813         if (evsel == trace->syscalls.events.augmented) {
2814                 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2815                 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2816
2817                 if (sc) {
2818                         fprintf(trace->output, "%s(", sc->name);
2819                         trace__fprintf_sys_enter(trace, evsel, sample);
2820                         fputc(')', trace->output);
2821                         goto newline;
2822                 }
2823
2824                 /*
2825                  * XXX: Not having the associated syscall info or not finding/adding
2826                  *      the thread should never happen, but if it does...
2827                  *      fall thru and print it as a bpf_output event.
2828                  */
2829         }
2830
2831         fprintf(trace->output, "%s(", evsel->name);
2832
2833         if (evsel__is_bpf_output(evsel)) {
2834                 bpf_output__fprintf(trace, sample);
2835         } else if (evsel->tp_format) {
2836                 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2837                     trace__fprintf_sys_enter(trace, evsel, sample)) {
2838                         if (trace->libtraceevent_print) {
2839                                 event_format__fprintf(evsel->tp_format, sample->cpu,
2840                                                       sample->raw_data, sample->raw_size,
2841                                                       trace->output);
2842                         } else {
2843                                 trace__fprintf_tp_fields(trace, evsel, sample, thread, NULL, 0);
2844                         }
2845                 }
2846         }
2847
2848 newline:
2849         fprintf(trace->output, ")\n");
2850
2851         if (callchain_ret > 0)
2852                 trace__fprintf_callchain(trace, sample);
2853         else if (callchain_ret < 0)
2854                 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2855
2856         ++trace->nr_events_printed;
2857
2858         if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2859                 evsel__disable(evsel);
2860                 evsel__close(evsel);
2861         }
2862 out:
2863         thread__put(thread);
2864         return 0;
2865 }
2866
2867 static void print_location(FILE *f, struct perf_sample *sample,
2868                            struct addr_location *al,
2869                            bool print_dso, bool print_sym)
2870 {
2871
2872         if ((verbose > 0 || print_dso) && al->map)
2873                 fprintf(f, "%s@", al->map->dso->long_name);
2874
2875         if ((verbose > 0 || print_sym) && al->sym)
2876                 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
2877                         al->addr - al->sym->start);
2878         else if (al->map)
2879                 fprintf(f, "0x%" PRIx64, al->addr);
2880         else
2881                 fprintf(f, "0x%" PRIx64, sample->addr);
2882 }
2883
2884 static int trace__pgfault(struct trace *trace,
2885                           struct evsel *evsel,
2886                           union perf_event *event __maybe_unused,
2887                           struct perf_sample *sample)
2888 {
2889         struct thread *thread;
2890         struct addr_location al;
2891         char map_type = 'd';
2892         struct thread_trace *ttrace;
2893         int err = -1;
2894         int callchain_ret = 0;
2895
2896         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2897
2898         if (sample->callchain) {
2899                 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2900                 if (callchain_ret == 0) {
2901                         if (callchain_cursor.nr < trace->min_stack)
2902                                 goto out_put;
2903                         callchain_ret = 1;
2904                 }
2905         }
2906
2907         ttrace = thread__trace(thread, trace->output);
2908         if (ttrace == NULL)
2909                 goto out_put;
2910
2911         if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
2912                 ttrace->pfmaj++;
2913         else
2914                 ttrace->pfmin++;
2915
2916         if (trace->summary_only)
2917                 goto out;
2918
2919         thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
2920
2921         trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
2922
2923         fprintf(trace->output, "%sfault [",
2924                 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
2925                 "maj" : "min");
2926
2927         print_location(trace->output, sample, &al, false, true);
2928
2929         fprintf(trace->output, "] => ");
2930
2931         thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2932
2933         if (!al.map) {
2934                 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2935
2936                 if (al.map)
2937                         map_type = 'x';
2938                 else
2939                         map_type = '?';
2940         }
2941
2942         print_location(trace->output, sample, &al, true, false);
2943
2944         fprintf(trace->output, " (%c%c)\n", map_type, al.level);
2945
2946         if (callchain_ret > 0)
2947                 trace__fprintf_callchain(trace, sample);
2948         else if (callchain_ret < 0)
2949                 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2950
2951         ++trace->nr_events_printed;
2952 out:
2953         err = 0;
2954 out_put:
2955         thread__put(thread);
2956         return err;
2957 }
2958
2959 static void trace__set_base_time(struct trace *trace,
2960                                  struct evsel *evsel,
2961                                  struct perf_sample *sample)
2962 {
2963         /*
2964          * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
2965          * and don't use sample->time unconditionally, we may end up having
2966          * some other event in the future without PERF_SAMPLE_TIME for good
2967          * reason, i.e. we may not be interested in its timestamps, just in
2968          * it taking place, picking some piece of information when it
2969          * appears in our event stream (vfs_getname comes to mind).
2970          */
2971         if (trace->base_time == 0 && !trace->full_time &&
2972             (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))
2973                 trace->base_time = sample->time;
2974 }
2975
2976 static int trace__process_sample(struct perf_tool *tool,
2977                                  union perf_event *event,
2978                                  struct perf_sample *sample,
2979                                  struct evsel *evsel,
2980                                  struct machine *machine __maybe_unused)
2981 {
2982         struct trace *trace = container_of(tool, struct trace, tool);
2983         struct thread *thread;
2984         int err = 0;
2985
2986         tracepoint_handler handler = evsel->handler;
2987
2988         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2989         if (thread && thread__is_filtered(thread))
2990                 goto out;
2991
2992         trace__set_base_time(trace, evsel, sample);
2993
2994         if (handler) {
2995                 ++trace->nr_events;
2996                 handler(trace, evsel, event, sample);
2997         }
2998 out:
2999         thread__put(thread);
3000         return err;
3001 }
3002
3003 static int trace__record(struct trace *trace, int argc, const char **argv)
3004 {
3005         unsigned int rec_argc, i, j;
3006         const char **rec_argv;
3007         const char * const record_args[] = {
3008                 "record",
3009                 "-R",
3010                 "-m", "1024",
3011                 "-c", "1",
3012         };
3013         pid_t pid = getpid();
3014         char *filter = asprintf__tp_filter_pids(1, &pid);
3015         const char * const sc_args[] = { "-e", };
3016         unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
3017         const char * const majpf_args[] = { "-e", "major-faults" };
3018         unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
3019         const char * const minpf_args[] = { "-e", "minor-faults" };
3020         unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
3021         int err = -1;
3022
3023         /* +3 is for the event string below and the pid filter */
3024         rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 3 +
3025                 majpf_args_nr + minpf_args_nr + argc;
3026         rec_argv = calloc(rec_argc + 1, sizeof(char *));
3027
3028         if (rec_argv == NULL || filter == NULL)
3029                 goto out_free;
3030
3031         j = 0;
3032         for (i = 0; i < ARRAY_SIZE(record_args); i++)
3033                 rec_argv[j++] = record_args[i];
3034
3035         if (trace->trace_syscalls) {
3036                 for (i = 0; i < sc_args_nr; i++)
3037                         rec_argv[j++] = sc_args[i];
3038
3039                 /* event string may be different for older kernels - e.g., RHEL6 */
3040                 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
3041                         rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
3042                 else if (is_valid_tracepoint("syscalls:sys_enter"))
3043                         rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
3044                 else {
3045                         pr_err("Neither raw_syscalls nor syscalls events exist.\n");
3046                         goto out_free;
3047                 }
3048         }
3049
3050         rec_argv[j++] = "--filter";
3051         rec_argv[j++] = filter;
3052
3053         if (trace->trace_pgfaults & TRACE_PFMAJ)
3054                 for (i = 0; i < majpf_args_nr; i++)
3055                         rec_argv[j++] = majpf_args[i];
3056
3057         if (trace->trace_pgfaults & TRACE_PFMIN)
3058                 for (i = 0; i < minpf_args_nr; i++)
3059                         rec_argv[j++] = minpf_args[i];
3060
3061         for (i = 0; i < (unsigned int)argc; i++)
3062                 rec_argv[j++] = argv[i];
3063
3064         err = cmd_record(j, rec_argv);
3065 out_free:
3066         free(filter);
3067         free(rec_argv);
3068         return err;
3069 }
3070
3071 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
3072
3073 static bool evlist__add_vfs_getname(struct evlist *evlist)
3074 {
3075         bool found = false;
3076         struct evsel *evsel, *tmp;
3077         struct parse_events_error err;
3078         int ret;
3079
3080         parse_events_error__init(&err);
3081         ret = parse_events(evlist, "probe:vfs_getname*", &err);
3082         parse_events_error__exit(&err);
3083         if (ret)
3084                 return false;
3085
3086         evlist__for_each_entry_safe(evlist, evsel, tmp) {
3087                 if (!strstarts(evsel__name(evsel), "probe:vfs_getname"))
3088                         continue;
3089
3090                 if (evsel__field(evsel, "pathname")) {
3091                         evsel->handler = trace__vfs_getname;
3092                         found = true;
3093                         continue;
3094                 }
3095
3096                 list_del_init(&evsel->core.node);
3097                 evsel->evlist = NULL;
3098                 evsel__delete(evsel);
3099         }
3100
3101         return found;
3102 }
3103
3104 static struct evsel *evsel__new_pgfault(u64 config)
3105 {
3106         struct evsel *evsel;
3107         struct perf_event_attr attr = {
3108                 .type = PERF_TYPE_SOFTWARE,
3109                 .mmap_data = 1,
3110         };
3111
3112         attr.config = config;
3113         attr.sample_period = 1;
3114
3115         event_attr_init(&attr);
3116
3117         evsel = evsel__new(&attr);
3118         if (evsel)
3119                 evsel->handler = trace__pgfault;
3120
3121         return evsel;
3122 }
3123
3124 static void evlist__free_syscall_tp_fields(struct evlist *evlist)
3125 {
3126         struct evsel *evsel;
3127
3128         evlist__for_each_entry(evlist, evsel) {
3129                 struct evsel_trace *et = evsel->priv;
3130
3131                 if (!et || !evsel->tp_format || strcmp(evsel->tp_format->system, "syscalls"))
3132                         continue;
3133
3134                 free(et->fmt);
3135                 free(et);
3136         }
3137 }
3138
3139 static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
3140 {
3141         const u32 type = event->header.type;
3142         struct evsel *evsel;
3143
3144         if (type != PERF_RECORD_SAMPLE) {
3145                 trace__process_event(trace, trace->host, event, sample);
3146                 return;
3147         }
3148
3149         evsel = evlist__id2evsel(trace->evlist, sample->id);
3150         if (evsel == NULL) {
3151                 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
3152                 return;
3153         }
3154
3155         if (evswitch__discard(&trace->evswitch, evsel))
3156                 return;
3157
3158         trace__set_base_time(trace, evsel, sample);
3159
3160         if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&
3161             sample->raw_data == NULL) {
3162                 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
3163                        evsel__name(evsel), sample->tid,
3164                        sample->cpu, sample->raw_size);
3165         } else {
3166                 tracepoint_handler handler = evsel->handler;
3167                 handler(trace, evsel, event, sample);
3168         }
3169
3170         if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
3171                 interrupted = true;
3172 }
3173
3174 static int trace__add_syscall_newtp(struct trace *trace)
3175 {
3176         int ret = -1;
3177         struct evlist *evlist = trace->evlist;
3178         struct evsel *sys_enter, *sys_exit;
3179
3180         sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
3181         if (sys_enter == NULL)
3182                 goto out;
3183
3184         if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
3185                 goto out_delete_sys_enter;
3186
3187         sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
3188         if (sys_exit == NULL)
3189                 goto out_delete_sys_enter;
3190
3191         if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
3192                 goto out_delete_sys_exit;
3193
3194         evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
3195         evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
3196
3197         evlist__add(evlist, sys_enter);
3198         evlist__add(evlist, sys_exit);
3199
3200         if (callchain_param.enabled && !trace->kernel_syscallchains) {
3201                 /*
3202                  * We're interested only in the user space callchain
3203                  * leading to the syscall, allow overriding that for
3204                  * debugging reasons using --kernel_syscall_callchains
3205                  */
3206                 sys_exit->core.attr.exclude_callchain_kernel = 1;
3207         }
3208
3209         trace->syscalls.events.sys_enter = sys_enter;
3210         trace->syscalls.events.sys_exit  = sys_exit;
3211
3212         ret = 0;
3213 out:
3214         return ret;
3215
3216 out_delete_sys_exit:
3217         evsel__delete_priv(sys_exit);
3218 out_delete_sys_enter:
3219         evsel__delete_priv(sys_enter);
3220         goto out;
3221 }
3222
3223 static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
3224 {
3225         int err = -1;
3226         struct evsel *sys_exit;
3227         char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
3228                                                 trace->ev_qualifier_ids.nr,
3229                                                 trace->ev_qualifier_ids.entries);
3230
3231         if (filter == NULL)
3232                 goto out_enomem;
3233
3234         if (!evsel__append_tp_filter(trace->syscalls.events.sys_enter, filter)) {
3235                 sys_exit = trace->syscalls.events.sys_exit;
3236                 err = evsel__append_tp_filter(sys_exit, filter);
3237         }
3238
3239         free(filter);
3240 out:
3241         return err;
3242 out_enomem:
3243         errno = ENOMEM;
3244         goto out;
3245 }
3246
3247 #ifdef HAVE_LIBBPF_SUPPORT
3248 static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace, const char *name)
3249 {
3250         if (trace->bpf_obj == NULL)
3251                 return NULL;
3252
3253         return bpf_object__find_map_by_name(trace->bpf_obj, name);
3254 }
3255
3256 static void trace__set_bpf_map_filtered_pids(struct trace *trace)
3257 {
3258         trace->filter_pids.map = trace__find_bpf_map_by_name(trace, "pids_filtered");
3259 }
3260
3261 static void trace__set_bpf_map_syscalls(struct trace *trace)
3262 {
3263         trace->syscalls.map = trace__find_bpf_map_by_name(trace, "syscalls");
3264         trace->syscalls.prog_array.sys_enter = trace__find_bpf_map_by_name(trace, "syscalls_sys_enter");
3265         trace->syscalls.prog_array.sys_exit  = trace__find_bpf_map_by_name(trace, "syscalls_sys_exit");
3266 }
3267
3268 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)
3269 {
3270         struct bpf_program *pos, *prog = NULL;
3271         const char *sec_name;
3272
3273         if (trace->bpf_obj == NULL)
3274                 return NULL;
3275
3276         bpf_object__for_each_program(pos, trace->bpf_obj) {
3277                 sec_name = bpf_program__section_name(pos);
3278                 if (sec_name && !strcmp(sec_name, name)) {
3279                         prog = pos;
3280                         break;
3281                 }
3282         }
3283
3284         return prog;
3285 }
3286
3287 static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,
3288                                                         const char *prog_name, const char *type)
3289 {
3290         struct bpf_program *prog;
3291
3292         if (prog_name == NULL) {
3293                 char default_prog_name[256];
3294                 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->name);
3295                 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3296                 if (prog != NULL)
3297                         goto out_found;
3298                 if (sc->fmt && sc->fmt->alias) {
3299                         scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->fmt->alias);
3300                         prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3301                         if (prog != NULL)
3302                                 goto out_found;
3303                 }
3304                 goto out_unaugmented;
3305         }
3306
3307         prog = trace__find_bpf_program_by_title(trace, prog_name);
3308
3309         if (prog != NULL) {
3310 out_found:
3311                 return prog;
3312         }
3313
3314         pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",
3315                  prog_name, type, sc->name);
3316 out_unaugmented:
3317         return trace->syscalls.unaugmented_prog;
3318 }
3319
3320 static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
3321 {
3322         struct syscall *sc = trace__syscall_info(trace, NULL, id);
3323
3324         if (sc == NULL)
3325                 return;
3326
3327         sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3328         sc->bpf_prog.sys_exit  = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_exit  : NULL,  "exit");
3329 }
3330
3331 static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
3332 {
3333         struct syscall *sc = trace__syscall_info(trace, NULL, id);
3334         return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->syscalls.unaugmented_prog);
3335 }
3336
3337 static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
3338 {
3339         struct syscall *sc = trace__syscall_info(trace, NULL, id);
3340         return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->syscalls.unaugmented_prog);
3341 }
3342
3343 static void trace__init_bpf_map_syscall_args(struct trace *trace, int id, struct bpf_map_syscall_entry *entry)
3344 {
3345         struct syscall *sc = trace__syscall_info(trace, NULL, id);
3346         int arg = 0;
3347
3348         if (sc == NULL)
3349                 goto out;
3350
3351         for (; arg < sc->nr_args; ++arg) {
3352                 entry->string_args_len[arg] = 0;
3353                 if (sc->arg_fmt[arg].scnprintf == SCA_FILENAME) {
3354                         /* Should be set like strace -s strsize */
3355                         entry->string_args_len[arg] = PATH_MAX;
3356                 }
3357         }
3358 out:
3359         for (; arg < 6; ++arg)
3360                 entry->string_args_len[arg] = 0;
3361 }
3362 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace)
3363 {
3364         int fd = bpf_map__fd(trace->syscalls.map);
3365         struct bpf_map_syscall_entry value = {
3366                 .enabled = !trace->not_ev_qualifier,
3367         };
3368         int err = 0;
3369         size_t i;
3370
3371         for (i = 0; i < trace->ev_qualifier_ids.nr; ++i) {
3372                 int key = trace->ev_qualifier_ids.entries[i];
3373
3374                 if (value.enabled) {
3375                         trace__init_bpf_map_syscall_args(trace, key, &value);
3376                         trace__init_syscall_bpf_progs(trace, key);
3377                 }
3378
3379                 err = bpf_map_update_elem(fd, &key, &value, BPF_EXIST);
3380                 if (err)
3381                         break;
3382         }
3383
3384         return err;
3385 }
3386
3387 static int __trace__init_syscalls_bpf_map(struct trace *trace, bool enabled)
3388 {
3389         int fd = bpf_map__fd(trace->syscalls.map);
3390         struct bpf_map_syscall_entry value = {
3391                 .enabled = enabled,
3392         };
3393         int err = 0, key;
3394
3395         for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3396                 if (enabled)
3397                         trace__init_bpf_map_syscall_args(trace, key, &value);
3398
3399                 err = bpf_map_update_elem(fd, &key, &value, BPF_ANY);
3400                 if (err)
3401                         break;
3402         }
3403
3404         return err;
3405 }
3406
3407 static int trace__init_syscalls_bpf_map(struct trace *trace)
3408 {
3409         bool enabled = true;
3410
3411         if (trace->ev_qualifier_ids.nr)
3412                 enabled = trace->not_ev_qualifier;
3413
3414         return __trace__init_syscalls_bpf_map(trace, enabled);
3415 }
3416
3417 static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)
3418 {
3419         struct tep_format_field *field, *candidate_field;
3420         int id;
3421
3422         /*
3423          * We're only interested in syscalls that have a pointer:
3424          */
3425         for (field = sc->args; field; field = field->next) {
3426                 if (field->flags & TEP_FIELD_IS_POINTER)
3427                         goto try_to_find_pair;
3428         }
3429
3430         return NULL;
3431
3432 try_to_find_pair:
3433         for (id = 0; id < trace->sctbl->syscalls.nr_entries; ++id) {
3434                 struct syscall *pair = trace__syscall_info(trace, NULL, id);
3435                 struct bpf_program *pair_prog;
3436                 bool is_candidate = false;
3437
3438                 if (pair == NULL || pair == sc ||
3439                     pair->bpf_prog.sys_enter == trace->syscalls.unaugmented_prog)
3440                         continue;
3441
3442                 for (field = sc->args, candidate_field = pair->args;
3443                      field && candidate_field; field = field->next, candidate_field = candidate_field->next) {
3444                         bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,
3445                              candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;
3446
3447                         if (is_pointer) {
3448                                if (!candidate_is_pointer) {
3449                                         // The candidate just doesn't copies our pointer arg, might copy other pointers we want.
3450                                         continue;
3451                                }
3452                         } else {
3453                                 if (candidate_is_pointer) {
3454                                         // The candidate might copy a pointer we don't have, skip it.
3455                                         goto next_candidate;
3456                                 }
3457                                 continue;
3458                         }
3459
3460                         if (strcmp(field->type, candidate_field->type))
3461                                 goto next_candidate;
3462
3463                         is_candidate = true;
3464                 }
3465
3466                 if (!is_candidate)
3467                         goto next_candidate;
3468
3469                 /*
3470                  * Check if the tentative pair syscall augmenter has more pointers, if it has,
3471                  * then it may be collecting that and we then can't use it, as it would collect
3472                  * more than what is common to the two syscalls.
3473                  */
3474                 if (candidate_field) {
3475                         for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)
3476                                 if (candidate_field->flags & TEP_FIELD_IS_POINTER)
3477                                         goto next_candidate;
3478                 }
3479
3480                 pair_prog = pair->bpf_prog.sys_enter;
3481                 /*
3482                  * If the pair isn't enabled, then its bpf_prog.sys_enter will not
3483                  * have been searched for, so search it here and if it returns the
3484                  * unaugmented one, then ignore it, otherwise we'll reuse that BPF
3485                  * program for a filtered syscall on a non-filtered one.
3486                  *
3487                  * For instance, we have "!syscalls:sys_enter_renameat" and that is
3488                  * useful for "renameat2".
3489                  */
3490                 if (pair_prog == NULL) {
3491                         pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3492                         if (pair_prog == trace->syscalls.unaugmented_prog)
3493                                 goto next_candidate;
3494                 }
3495
3496                 pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);
3497                 return pair_prog;
3498         next_candidate:
3499                 continue;
3500         }
3501
3502         return NULL;
3503 }
3504
3505 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
3506 {
3507         int map_enter_fd = bpf_map__fd(trace->syscalls.prog_array.sys_enter),
3508             map_exit_fd  = bpf_map__fd(trace->syscalls.prog_array.sys_exit);
3509         int err = 0, key;
3510
3511         for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3512                 int prog_fd;
3513
3514                 if (!trace__syscall_enabled(trace, key))
3515                         continue;
3516
3517                 trace__init_syscall_bpf_progs(trace, key);
3518
3519                 // It'll get at least the "!raw_syscalls:unaugmented"
3520                 prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
3521                 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3522                 if (err)
3523                         break;
3524                 prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
3525                 err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
3526                 if (err)
3527                         break;
3528         }
3529
3530         /*
3531          * Now lets do a second pass looking for enabled syscalls without
3532          * an augmenter that have a signature that is a superset of another
3533          * syscall with an augmenter so that we can auto-reuse it.
3534          *
3535          * I.e. if we have an augmenter for the "open" syscall that has
3536          * this signature:
3537          *
3538          *   int open(const char *pathname, int flags, mode_t mode);
3539          *
3540          * I.e. that will collect just the first string argument, then we
3541          * can reuse it for the 'creat' syscall, that has this signature:
3542          *
3543          *   int creat(const char *pathname, mode_t mode);
3544          *
3545          * and for:
3546          *
3547          *   int stat(const char *pathname, struct stat *statbuf);
3548          *   int lstat(const char *pathname, struct stat *statbuf);
3549          *
3550          * Because the 'open' augmenter will collect the first arg as a string,
3551          * and leave alone all the other args, which already helps with
3552          * beautifying 'stat' and 'lstat''s pathname arg.
3553          *
3554          * Then, in time, when 'stat' gets an augmenter that collects both
3555          * first and second arg (this one on the raw_syscalls:sys_exit prog
3556          * array tail call, then that one will be used.
3557          */
3558         for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3559                 struct syscall *sc = trace__syscall_info(trace, NULL, key);
3560                 struct bpf_program *pair_prog;
3561                 int prog_fd;
3562
3563                 if (sc == NULL || sc->bpf_prog.sys_enter == NULL)
3564                         continue;
3565
3566                 /*
3567                  * For now we're just reusing the sys_enter prog, and if it
3568                  * already has an augmenter, we don't need to find one.
3569                  */
3570                 if (sc->bpf_prog.sys_enter != trace->syscalls.unaugmented_prog)
3571                         continue;
3572
3573                 /*
3574                  * Look at all the other syscalls for one that has a signature
3575                  * that is close enough that we can share:
3576                  */
3577                 pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);
3578                 if (pair_prog == NULL)
3579                         continue;
3580
3581                 sc->bpf_prog.sys_enter = pair_prog;
3582
3583                 /*
3584                  * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter
3585                  * with the fd for the program we're reusing:
3586                  */
3587                 prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);
3588                 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3589                 if (err)
3590                         break;
3591         }
3592
3593
3594         return err;
3595 }
3596
3597 static void trace__delete_augmented_syscalls(struct trace *trace)
3598 {
3599         struct evsel *evsel, *tmp;
3600
3601         evlist__remove(trace->evlist, trace->syscalls.events.augmented);
3602         evsel__delete(trace->syscalls.events.augmented);
3603         trace->syscalls.events.augmented = NULL;
3604
3605         evlist__for_each_entry_safe(trace->evlist, tmp, evsel) {
3606                 if (evsel->bpf_obj == trace->bpf_obj) {
3607                         evlist__remove(trace->evlist, evsel);
3608                         evsel__delete(evsel);
3609                 }
3610
3611         }
3612
3613         bpf_object__close(trace->bpf_obj);
3614         trace->bpf_obj = NULL;
3615 }
3616 #else // HAVE_LIBBPF_SUPPORT
3617 static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace __maybe_unused,
3618                                                    const char *name __maybe_unused)
3619 {
3620         return NULL;
3621 }
3622
3623 static void trace__set_bpf_map_filtered_pids(struct trace *trace __maybe_unused)
3624 {
3625 }
3626
3627 static void trace__set_bpf_map_syscalls(struct trace *trace __maybe_unused)
3628 {
3629 }
3630
3631 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace __maybe_unused)
3632 {
3633         return 0;
3634 }
3635
3636 static int trace__init_syscalls_bpf_map(struct trace *trace __maybe_unused)
3637 {
3638         return 0;
3639 }
3640
3641 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace __maybe_unused,
3642                                                             const char *name __maybe_unused)
3643 {
3644         return NULL;
3645 }
3646
3647 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace __maybe_unused)
3648 {
3649         return 0;
3650 }
3651
3652 static void trace__delete_augmented_syscalls(struct trace *trace __maybe_unused)
3653 {
3654 }
3655 #endif // HAVE_LIBBPF_SUPPORT
3656
3657 static bool trace__only_augmented_syscalls_evsels(struct trace *trace)
3658 {
3659         struct evsel *evsel;
3660
3661         evlist__for_each_entry(trace->evlist, evsel) {
3662                 if (evsel == trace->syscalls.events.augmented ||
3663                     evsel->bpf_obj == trace->bpf_obj)
3664                         continue;
3665
3666                 return false;
3667         }
3668
3669         return true;
3670 }
3671
3672 static int trace__set_ev_qualifier_filter(struct trace *trace)
3673 {
3674         if (trace->syscalls.map)
3675                 return trace__set_ev_qualifier_bpf_filter(trace);
3676         if (trace->syscalls.events.sys_enter)
3677                 return trace__set_ev_qualifier_tp_filter(trace);
3678         return 0;
3679 }
3680
3681 static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
3682                                     size_t npids __maybe_unused, pid_t *pids __maybe_unused)
3683 {
3684         int err = 0;
3685 #ifdef HAVE_LIBBPF_SUPPORT
3686         bool value = true;
3687         int map_fd = bpf_map__fd(map);
3688         size_t i;
3689
3690         for (i = 0; i < npids; ++i) {
3691                 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
3692                 if (err)
3693                         break;
3694         }
3695 #endif
3696         return err;
3697 }
3698
3699 static int trace__set_filter_loop_pids(struct trace *trace)
3700 {
3701         unsigned int nr = 1, err;
3702         pid_t pids[32] = {
3703                 getpid(),
3704         };
3705         struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
3706
3707         while (thread && nr < ARRAY_SIZE(pids)) {
3708                 struct thread *parent = machine__find_thread(trace->host, thread->ppid, thread->ppid);
3709
3710                 if (parent == NULL)
3711                         break;
3712
3713                 if (!strcmp(thread__comm_str(parent), "sshd") ||
3714                     strstarts(thread__comm_str(parent), "gnome-terminal")) {
3715                         pids[nr++] = parent->tid;
3716                         break;
3717                 }
3718                 thread = parent;
3719         }
3720
3721         err = evlist__append_tp_filter_pids(trace->evlist, nr, pids);
3722         if (!err && trace->filter_pids.map)
3723                 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
3724
3725         return err;
3726 }
3727
3728 static int trace__set_filter_pids(struct trace *trace)
3729 {
3730         int err = 0;
3731         /*
3732          * Better not use !target__has_task() here because we need to cover the
3733          * case where no threads were specified in the command line, but a
3734          * workload was, and in that case we will fill in the thread_map when
3735          * we fork the workload in evlist__prepare_workload.
3736          */
3737         if (trace->filter_pids.nr > 0) {
3738                 err = evlist__append_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
3739                                                     trace->filter_pids.entries);
3740                 if (!err && trace->filter_pids.map) {
3741                         err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
3742                                                        trace->filter_pids.entries);
3743                 }
3744         } else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {
3745                 err = trace__set_filter_loop_pids(trace);
3746         }
3747
3748         return err;
3749 }
3750
3751 static int __trace__deliver_event(struct trace *trace, union perf_event *event)
3752 {
3753         struct evlist *evlist = trace->evlist;
3754         struct perf_sample sample;
3755         int err = evlist__parse_sample(evlist, event, &sample);
3756
3757         if (err)
3758                 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
3759         else
3760                 trace__handle_event(trace, event, &sample);
3761
3762         return 0;
3763 }
3764
3765 static int __trace__flush_events(struct trace *trace)
3766 {
3767         u64 first = ordered_events__first_time(&trace->oe.data);
3768         u64 flush = trace->oe.last - NSEC_PER_SEC;
3769
3770         /* Is there some thing to flush.. */
3771         if (first && first < flush)
3772                 return ordered_events__flush_time(&trace->oe.data, flush);
3773
3774         return 0;
3775 }
3776
3777 static int trace__flush_events(struct trace *trace)
3778 {
3779         return !trace->sort_events ? 0 : __trace__flush_events(trace);
3780 }
3781
3782 static int trace__deliver_event(struct trace *trace, union perf_event *event)
3783 {
3784         int err;
3785
3786         if (!trace->sort_events)
3787                 return __trace__deliver_event(trace, event);
3788
3789         err = evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
3790         if (err && err != -1)
3791                 return err;
3792
3793         err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0, NULL);
3794         if (err)
3795                 return err;
3796
3797         return trace__flush_events(trace);
3798 }
3799
3800 static int ordered_events__deliver_event(struct ordered_events *oe,
3801                                          struct ordered_event *event)
3802 {
3803         struct trace *trace = container_of(oe, struct trace, oe.data);
3804
3805         return __trace__deliver_event(trace, event->event);
3806 }
3807
3808 static struct syscall_arg_fmt *evsel__find_syscall_arg_fmt_by_name(struct evsel *evsel, char *arg)
3809 {
3810         struct tep_format_field *field;
3811         struct syscall_arg_fmt *fmt = __evsel__syscall_arg_fmt(evsel);
3812
3813         if (evsel->tp_format == NULL || fmt == NULL)
3814                 return NULL;
3815
3816         for (field = evsel->tp_format->format.fields; field; field = field->next, ++fmt)
3817                 if (strcmp(field->name, arg) == 0)
3818                         return fmt;
3819
3820         return NULL;
3821 }
3822
3823 static int trace__expand_filter(struct trace *trace __maybe_unused, struct evsel *evsel)
3824 {
3825         char *tok, *left = evsel->filter, *new_filter = evsel->filter;
3826
3827         while ((tok = strpbrk(left, "=<>!")) != NULL) {
3828                 char *right = tok + 1, *right_end;
3829
3830                 if (*right == '=')
3831                         ++right;
3832
3833                 while (isspace(*right))
3834                         ++right;
3835
3836                 if (*right == '\0')
3837                         break;
3838
3839                 while (!isalpha(*left))
3840                         if (++left == tok) {
3841                                 /*
3842                                  * Bail out, can't find the name of the argument that is being
3843                                  * used in the filter, let it try to set this filter, will fail later.
3844                                  */
3845                                 return 0;
3846                         }
3847
3848                 right_end = right + 1;
3849                 while (isalnum(*right_end) || *right_end == '_' || *right_end == '|')
3850                         ++right_end;
3851
3852                 if (isalpha(*right)) {
3853                         struct syscall_arg_fmt *fmt;
3854                         int left_size = tok - left,
3855                             right_size = right_end - right;
3856                         char arg[128];
3857
3858                         while (isspace(left[left_size - 1]))
3859                                 --left_size;
3860
3861                         scnprintf(arg, sizeof(arg), "%.*s", left_size, left);
3862
3863                         fmt = evsel__find_syscall_arg_fmt_by_name(evsel, arg);
3864                         if (fmt == NULL) {
3865                                 pr_err("\"%s\" not found in \"%s\", can't set filter \"%s\"\n",
3866                                        arg, evsel->name, evsel->filter);
3867                                 return -1;
3868                         }
3869
3870                         pr_debug2("trying to expand \"%s\" \"%.*s\" \"%.*s\" -> ",
3871                                  arg, (int)(right - tok), tok, right_size, right);
3872
3873                         if (fmt->strtoul) {
3874                                 u64 val;
3875                                 struct syscall_arg syscall_arg = {
3876                                         .parm = fmt->parm,
3877                                 };
3878
3879                                 if (fmt->strtoul(right, right_size, &syscall_arg, &val)) {
3880                                         char *n, expansion[19];
3881                                         int expansion_lenght = scnprintf(expansion, sizeof(expansion), "%#" PRIx64, val);
3882                                         int expansion_offset = right - new_filter;
3883
3884                                         pr_debug("%s", expansion);
3885
3886                                         if (asprintf(&n, "%.*s%s%s", expansion_offset, new_filter, expansion, right_end) < 0) {
3887                                                 pr_debug(" out of memory!\n");
3888                                                 free(new_filter);
3889                                                 return -1;
3890                                         }
3891                                         if (new_filter != evsel->filter)
3892                                                 free(new_filter);
3893                                         left = n + expansion_offset + expansion_lenght;
3894                                         new_filter = n;
3895                                 } else {
3896                                         pr_err("\"%.*s\" not found for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3897                                                right_size, right, arg, evsel->name, evsel->filter);
3898                                         return -1;
3899                                 }
3900                         } else {
3901                                 pr_err("No resolver (strtoul) for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3902                                        arg, evsel->name, evsel->filter);
3903                                 return -1;
3904                         }
3905
3906                         pr_debug("\n");
3907                 } else {
3908                         left = right_end;
3909                 }
3910         }
3911
3912         if (new_filter != evsel->filter) {
3913                 pr_debug("New filter for %s: %s\n", evsel->name, new_filter);
3914                 evsel__set_filter(evsel, new_filter);
3915                 free(new_filter);
3916         }
3917
3918         return 0;
3919 }
3920
3921 static int trace__expand_filters(struct trace *trace, struct evsel **err_evsel)
3922 {
3923         struct evlist *evlist = trace->evlist;
3924         struct evsel *evsel;
3925
3926         evlist__for_each_entry(evlist, evsel) {
3927                 if (evsel->filter == NULL)
3928                         continue;
3929
3930                 if (trace__expand_filter(trace, evsel)) {
3931                         *err_evsel = evsel;
3932                         return -1;
3933                 }
3934         }
3935
3936         return 0;
3937 }
3938
3939 static int trace__run(struct trace *trace, int argc, const char **argv)
3940 {
3941         struct evlist *evlist = trace->evlist;
3942         struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
3943         int err = -1, i;
3944         unsigned long before;
3945         const bool forks = argc > 0;
3946         bool draining = false;
3947
3948         trace->live = true;
3949
3950         if (!trace->raw_augmented_syscalls) {
3951                 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
3952                         goto out_error_raw_syscalls;
3953
3954                 if (trace->trace_syscalls)
3955                         trace->vfs_getname = evlist__add_vfs_getname(evlist);
3956         }
3957
3958         if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
3959                 pgfault_maj = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
3960                 if (pgfault_maj == NULL)
3961                         goto out_error_mem;
3962                 evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
3963                 evlist__add(evlist, pgfault_maj);
3964         }
3965
3966         if ((trace->trace_pgfaults & TRACE_PFMIN)) {
3967                 pgfault_min = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
3968                 if (pgfault_min == NULL)
3969                         goto out_error_mem;
3970                 evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
3971                 evlist__add(evlist, pgfault_min);
3972         }
3973
3974         /* Enable ignoring missing threads when -u/-p option is defined. */
3975         trace->opts.ignore_missing_thread = trace->opts.target.uid != UINT_MAX || trace->opts.target.pid;
3976
3977         if (trace->sched &&
3978             evlist__add_newtp(evlist, "sched", "sched_stat_runtime", trace__sched_stat_runtime))
3979                 goto out_error_sched_stat_runtime;
3980         /*
3981          * If a global cgroup was set, apply it to all the events without an
3982          * explicit cgroup. I.e.:
3983          *
3984          *      trace -G A -e sched:*switch
3985          *
3986          * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
3987          * _and_ sched:sched_switch to the 'A' cgroup, while:
3988          *
3989          * trace -e sched:*switch -G A
3990          *
3991          * will only set the sched:sched_switch event to the 'A' cgroup, all the
3992          * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
3993          * a cgroup (on the root cgroup, sys wide, etc).
3994          *
3995          * Multiple cgroups:
3996          *
3997          * trace -G A -e sched:*switch -G B
3998          *
3999          * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
4000          * to the 'B' cgroup.
4001          *
4002          * evlist__set_default_cgroup() grabs a reference of the passed cgroup
4003          * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
4004          */
4005         if (trace->cgroup)
4006                 evlist__set_default_cgroup(trace->evlist, trace->cgroup);
4007
4008         err = evlist__create_maps(evlist, &trace->opts.target);
4009         if (err < 0) {
4010                 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
4011                 goto out_delete_evlist;
4012         }
4013
4014         err = trace__symbols_init(trace, evlist);
4015         if (err < 0) {
4016                 fprintf(trace->output, "Problems initializing symbol libraries!\n");
4017                 goto out_delete_evlist;
4018         }
4019
4020         evlist__config(evlist, &trace->opts, &callchain_param);
4021
4022         if (forks) {
4023                 err = evlist__prepare_workload(evlist, &trace->opts.target, argv, false, NULL);
4024                 if (err < 0) {
4025                         fprintf(trace->output, "Couldn't run the workload!\n");
4026                         goto out_delete_evlist;
4027                 }
4028                 workload_pid = evlist->workload.pid;
4029         }
4030
4031         err = evlist__open(evlist);
4032         if (err < 0)
4033                 goto out_error_open;
4034
4035         err = bpf__apply_obj_config();
4036         if (err) {
4037                 char errbuf[BUFSIZ];
4038
4039                 bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf));
4040                 pr_err("ERROR: Apply config to BPF failed: %s\n",
4041                          errbuf);
4042                 goto out_error_open;
4043         }
4044
4045         err = trace__set_filter_pids(trace);
4046         if (err < 0)
4047                 goto out_error_mem;
4048
4049         if (trace->syscalls.map)
4050                 trace__init_syscalls_bpf_map(trace);
4051
4052         if (trace->syscalls.prog_array.sys_enter)
4053                 trace__init_syscalls_bpf_prog_array_maps(trace);
4054
4055         if (trace->ev_qualifier_ids.nr > 0) {
4056                 err = trace__set_ev_qualifier_filter(trace);
4057                 if (err < 0)
4058                         goto out_errno;
4059
4060                 if (trace->syscalls.events.sys_exit) {
4061                         pr_debug("event qualifier tracepoint filter: %s\n",
4062                                  trace->syscalls.events.sys_exit->filter);
4063                 }
4064         }
4065
4066         /*
4067          * If the "close" syscall is not traced, then we will not have the
4068          * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the
4069          * fd->pathname table and were ending up showing the last value set by
4070          * syscalls opening a pathname and associating it with a descriptor or
4071          * reading it from /proc/pid/fd/ in cases where that doesn't make
4072          * sense.
4073          *
4074          *  So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is
4075          *  not in use.
4076          */
4077         trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));
4078
4079         err = trace__expand_filters(trace, &evsel);
4080         if (err)
4081                 goto out_delete_evlist;
4082         err = evlist__apply_filters(evlist, &evsel);
4083         if (err < 0)
4084                 goto out_error_apply_filters;
4085
4086         if (trace->dump.map)
4087                 bpf_map__fprintf(trace->dump.map, trace->output);
4088
4089         err = evlist__mmap(evlist, trace->opts.mmap_pages);
4090         if (err < 0)
4091                 goto out_error_mmap;
4092
4093         if (!target__none(&trace->opts.target) && !trace->opts.initial_delay)
4094                 evlist__enable(evlist);
4095
4096         if (forks)
4097                 evlist__start_workload(evlist);
4098
4099         if (trace->opts.initial_delay) {
4100                 usleep(trace->opts.initial_delay * 1000);
4101                 evlist__enable(evlist);
4102         }
4103
4104         trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||
4105                                   evlist->core.threads->nr > 1 ||
4106                                   evlist__first(evlist)->core.attr.inherit;
4107
4108         /*
4109          * Now that we already used evsel->core.attr to ask the kernel to setup the
4110          * events, lets reuse evsel->core.attr.sample_max_stack as the limit in
4111          * trace__resolve_callchain(), allowing per-event max-stack settings
4112          * to override an explicitly set --max-stack global setting.
4113          */
4114         evlist__for_each_entry(evlist, evsel) {
4115                 if (evsel__has_callchain(evsel) &&
4116                     evsel->core.attr.sample_max_stack == 0)
4117                         evsel->core.attr.sample_max_stack = trace->max_stack;
4118         }
4119 again:
4120         before = trace->nr_events;
4121
4122         for (i = 0; i < evlist->core.nr_mmaps; i++) {
4123                 union perf_event *event;
4124                 struct mmap *md;
4125
4126                 md = &evlist->mmap[i];
4127                 if (perf_mmap__read_init(&md->core) < 0)
4128                         continue;
4129
4130                 while ((event = perf_mmap__read_event(&md->core)) != NULL) {
4131                         ++trace->nr_events;
4132
4133                         err = trace__deliver_event(trace, event);
4134                         if (err)
4135                                 goto out_disable;
4136
4137                         perf_mmap__consume(&md->core);
4138
4139                         if (interrupted)
4140                                 goto out_disable;
4141
4142                         if (done && !draining) {
4143                                 evlist__disable(evlist);
4144                                 draining = true;
4145                         }
4146                 }
4147                 perf_mmap__read_done(&md->core);
4148         }
4149
4150         if (trace->nr_events == before) {
4151                 int timeout = done ? 100 : -1;
4152
4153                 if (!draining && evlist__poll(evlist, timeout) > 0) {
4154                         if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
4155                                 draining = true;
4156
4157                         goto again;
4158                 } else {
4159                         if (trace__flush_events(trace))
4160                                 goto out_disable;
4161                 }
4162         } else {
4163                 goto again;
4164         }
4165
4166 out_disable:
4167         thread__zput(trace->current);
4168
4169         evlist__disable(evlist);
4170
4171         if (trace->sort_events)
4172                 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
4173
4174         if (!err) {
4175                 if (trace->summary)
4176                         trace__fprintf_thread_summary(trace, trace->output);
4177
4178                 if (trace->show_tool_stats) {
4179                         fprintf(trace->output, "Stats:\n "
4180                                                " vfs_getname : %" PRIu64 "\n"
4181                                                " proc_getname: %" PRIu64 "\n",
4182                                 trace->stats.vfs_getname,
4183                                 trace->stats.proc_getname);
4184                 }
4185         }
4186
4187 out_delete_evlist:
4188         trace__symbols__exit(trace);
4189         evlist__free_syscall_tp_fields(evlist);
4190         evlist__delete(evlist);
4191         cgroup__put(trace->cgroup);
4192         trace->evlist = NULL;
4193         trace->live = false;
4194         return err;
4195 {
4196         char errbuf[BUFSIZ];
4197
4198 out_error_sched_stat_runtime:
4199         tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
4200         goto out_error;
4201
4202 out_error_raw_syscalls:
4203         tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
4204         goto out_error;
4205
4206 out_error_mmap:
4207         evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
4208         goto out_error;
4209
4210 out_error_open:
4211         evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
4212
4213 out_error:
4214         fprintf(trace->output, "%s\n", errbuf);
4215         goto out_delete_evlist;
4216
4217 out_error_apply_filters:
4218         fprintf(trace->output,
4219                 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
4220                 evsel->filter, evsel__name(evsel), errno,
4221                 str_error_r(errno, errbuf, sizeof(errbuf)));
4222         goto out_delete_evlist;
4223 }
4224 out_error_mem:
4225         fprintf(trace->output, "Not enough memory to run!\n");
4226         goto out_delete_evlist;
4227
4228 out_errno:
4229         fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
4230         goto out_delete_evlist;
4231 }
4232
4233 static int trace__replay(struct trace *trace)
4234 {
4235         const struct evsel_str_handler handlers[] = {
4236                 { "probe:vfs_getname",       trace__vfs_getname, },
4237         };
4238         struct perf_data data = {
4239                 .path  = input_name,
4240                 .mode  = PERF_DATA_MODE_READ,
4241                 .force = trace->force,
4242         };
4243         struct perf_session *session;
4244         struct evsel *evsel;
4245         int err = -1;
4246
4247         trace->tool.sample        = trace__process_sample;
4248         trace->tool.mmap          = perf_event__process_mmap;
4249         trace->tool.mmap2         = perf_event__process_mmap2;
4250         trace->tool.comm          = perf_event__process_comm;
4251         trace->tool.exit          = perf_event__process_exit;
4252         trace->tool.fork          = perf_event__process_fork;
4253         trace->tool.attr          = perf_event__process_attr;
4254         trace->tool.tracing_data  = perf_event__process_tracing_data;
4255         trace->tool.build_id      = perf_event__process_build_id;
4256         trace->tool.namespaces    = perf_event__process_namespaces;
4257
4258         trace->tool.ordered_events = true;
4259         trace->tool.ordering_requires_timestamps = true;
4260
4261         /* add tid to output */
4262         trace->multiple_threads = true;
4263
4264         session = perf_session__new(&data, &trace->tool);
4265         if (IS_ERR(session))
4266                 return PTR_ERR(session);
4267
4268         if (trace->opts.target.pid)
4269                 symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
4270
4271         if (trace->opts.target.tid)
4272                 symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
4273
4274         if (symbol__init(&session->header.env) < 0)
4275                 goto out;
4276
4277         trace->host = &session->machines.host;
4278
4279         err = perf_session__set_tracepoints_handlers(session, handlers);
4280         if (err)
4281                 goto out;
4282
4283         evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_enter");
4284         trace->syscalls.events.sys_enter = evsel;
4285         /* older kernels have syscalls tp versus raw_syscalls */
4286         if (evsel == NULL)
4287                 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_enter");
4288
4289         if (evsel &&
4290             (evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
4291             perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
4292                 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
4293                 goto out;
4294         }
4295
4296         evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_exit");
4297         trace->syscalls.events.sys_exit = evsel;
4298         if (evsel == NULL)
4299                 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_exit");
4300         if (evsel &&
4301             (evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
4302             perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
4303                 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
4304                 goto out;
4305         }
4306
4307         evlist__for_each_entry(session->evlist, evsel) {
4308                 if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&
4309                     (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
4310                      evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
4311                      evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))
4312                         evsel->handler = trace__pgfault;
4313         }
4314
4315         setup_pager();
4316
4317         err = perf_session__process_events(session);
4318         if (err)
4319                 pr_err("Failed to process events, error %d", err);
4320
4321         else if (trace->summary)
4322                 trace__fprintf_thread_summary(trace, trace->output);
4323
4324 out:
4325         perf_session__delete(session);
4326
4327         return err;
4328 }
4329
4330 static size_t trace__fprintf_threads_header(FILE *fp)
4331 {
4332         size_t printed;
4333
4334         printed  = fprintf(fp, "\n Summary of events:\n\n");
4335
4336         return printed;
4337 }
4338
4339 DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
4340         struct syscall_stats *stats;
4341         double               msecs;
4342         int                  syscall;
4343 )
4344 {
4345         struct int_node *source = rb_entry(nd, struct int_node, rb_node);
4346         struct syscall_stats *stats = source->priv;
4347
4348         entry->syscall = source->i;
4349         entry->stats   = stats;
4350         entry->msecs   = stats ? (u64)stats->stats.n * (avg_stats(&stats->stats) / NSEC_PER_MSEC) : 0;
4351 }
4352
4353 static size_t thread__dump_stats(struct thread_trace *ttrace,
4354                                  struct trace *trace, FILE *fp)
4355 {
4356         size_t printed = 0;
4357         struct syscall *sc;
4358         struct rb_node *nd;
4359         DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
4360
4361         if (syscall_stats == NULL)
4362                 return 0;
4363
4364         printed += fprintf(fp, "\n");
4365
4366         printed += fprintf(fp, "   syscall            calls  errors  total       min       avg       max       stddev\n");
4367         printed += fprintf(fp, "                                     (msec)    (msec)    (msec)    (msec)        (%%)\n");
4368         printed += fprintf(fp, "   --------------- --------  ------ -------- --------- --------- ---------     ------\n");
4369
4370         resort_rb__for_each_entry(nd, syscall_stats) {
4371                 struct syscall_stats *stats = syscall_stats_entry->stats;
4372                 if (stats) {
4373                         double min = (double)(stats->stats.min) / NSEC_PER_MSEC;
4374                         double max = (double)(stats->stats.max) / NSEC_PER_MSEC;
4375                         double avg = avg_stats(&stats->stats);
4376                         double pct;
4377                         u64 n = (u64)stats->stats.n;
4378
4379                         pct = avg ? 100.0 * stddev_stats(&stats->stats) / avg : 0.0;
4380                         avg /= NSEC_PER_MSEC;
4381
4382                         sc = &trace->syscalls.table[syscall_stats_entry->syscall];
4383                         printed += fprintf(fp, "   %-15s", sc->name);
4384                         printed += fprintf(fp, " %8" PRIu64 " %6" PRIu64 " %9.3f %9.3f %9.3f",
4385                                            n, stats->nr_failures, syscall_stats_entry->msecs, min, avg);
4386                         printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
4387
4388                         if (trace->errno_summary && stats->nr_failures) {
4389                                 const char *arch_name = perf_env__arch(trace->host->env);
4390                                 int e;
4391
4392                                 for (e = 0; e < stats->max_errno; ++e) {
4393                                         if (stats->errnos[e] != 0)
4394                                                 fprintf(fp, "\t\t\t\t%s: %d\n", arch_syscalls__strerrno(arch_name, e + 1), stats->errnos[e]);
4395                                 }
4396                         }
4397                 }
4398         }
4399
4400         resort_rb__delete(syscall_stats);
4401         printed += fprintf(fp, "\n\n");
4402
4403         return printed;
4404 }
4405
4406 static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
4407 {
4408         size_t printed = 0;
4409         struct thread_trace *ttrace = thread__priv(thread);
4410         double ratio;
4411
4412         if (ttrace == NULL)
4413                 return 0;
4414
4415         ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
4416
4417         printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
4418         printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
4419         printed += fprintf(fp, "%.1f%%", ratio);
4420         if (ttrace->pfmaj)
4421                 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
4422         if (ttrace->pfmin)
4423                 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
4424         if (trace->sched)
4425                 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
4426         else if (fputc('\n', fp) != EOF)
4427                 ++printed;
4428
4429         printed += thread__dump_stats(ttrace, trace, fp);
4430
4431         return printed;
4432 }
4433
4434 static unsigned long thread__nr_events(struct thread_trace *ttrace)
4435 {
4436         return ttrace ? ttrace->nr_events : 0;
4437 }
4438
4439 DEFINE_RESORT_RB(threads, (thread__nr_events(a->thread->priv) < thread__nr_events(b->thread->priv)),
4440         struct thread *thread;
4441 )
4442 {
4443         entry->thread = rb_entry(nd, struct thread, rb_node);
4444 }
4445
4446 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
4447 {
4448         size_t printed = trace__fprintf_threads_header(fp);
4449         struct rb_node *nd;
4450         int i;
4451
4452         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
4453                 DECLARE_RESORT_RB_MACHINE_THREADS(threads, trace->host, i);
4454
4455                 if (threads == NULL) {
4456                         fprintf(fp, "%s", "Error sorting output by nr_events!\n");
4457                         return 0;
4458                 }
4459
4460                 resort_rb__for_each_entry(nd, threads)
4461                         printed += trace__fprintf_thread(fp, threads_entry->thread, trace);
4462
4463                 resort_rb__delete(threads);
4464         }
4465         return printed;
4466 }
4467
4468 static int trace__set_duration(const struct option *opt, const char *str,
4469                                int unset __maybe_unused)
4470 {
4471         struct trace *trace = opt->value;
4472
4473         trace->duration_filter = atof(str);
4474         return 0;
4475 }
4476
4477 static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
4478                                               int unset __maybe_unused)
4479 {
4480         int ret = -1;
4481         size_t i;
4482         struct trace *trace = opt->value;
4483         /*
4484          * FIXME: introduce a intarray class, plain parse csv and create a
4485          * { int nr, int entries[] } struct...
4486          */
4487         struct intlist *list = intlist__new(str);
4488
4489         if (list == NULL)
4490                 return -1;
4491
4492         i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
4493         trace->filter_pids.entries = calloc(i, sizeof(pid_t));
4494
4495         if (trace->filter_pids.entries == NULL)
4496                 goto out;
4497
4498         trace->filter_pids.entries[0] = getpid();
4499
4500         for (i = 1; i < trace->filter_pids.nr; ++i)
4501                 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
4502
4503         intlist__delete(list);
4504         ret = 0;
4505 out:
4506         return ret;
4507 }
4508
4509 static int trace__open_output(struct trace *trace, const char *filename)
4510 {
4511         struct stat st;
4512
4513         if (!stat(filename, &st) && st.st_size) {
4514                 char oldname[PATH_MAX];
4515
4516                 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
4517                 unlink(oldname);
4518                 rename(filename, oldname);
4519         }
4520
4521         trace->output = fopen(filename, "w");
4522
4523         return trace->output == NULL ? -errno : 0;
4524 }
4525
4526 static int parse_pagefaults(const struct option *opt, const char *str,
4527                             int unset __maybe_unused)
4528 {
4529         int *trace_pgfaults = opt->value;
4530
4531         if (strcmp(str, "all") == 0)
4532                 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
4533         else if (strcmp(str, "maj") == 0)
4534                 *trace_pgfaults |= TRACE_PFMAJ;
4535         else if (strcmp(str, "min") == 0)
4536                 *trace_pgfaults |= TRACE_PFMIN;
4537         else
4538                 return -1;
4539
4540         return 0;
4541 }
4542
4543 static void evlist__set_default_evsel_handler(struct evlist *evlist, void *handler)
4544 {
4545         struct evsel *evsel;
4546
4547         evlist__for_each_entry(evlist, evsel) {
4548                 if (evsel->handler == NULL)
4549                         evsel->handler = handler;
4550         }
4551 }
4552
4553 static void evsel__set_syscall_arg_fmt(struct evsel *evsel, const char *name)
4554 {
4555         struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
4556
4557         if (fmt) {
4558                 struct syscall_fmt *scfmt = syscall_fmt__find(name);
4559
4560                 if (scfmt) {
4561                         int skip = 0;
4562
4563                         if (strcmp(evsel->tp_format->format.fields->name, "__syscall_nr") == 0 ||
4564                             strcmp(evsel->tp_format->format.fields->name, "nr") == 0)
4565                                 ++skip;
4566
4567                         memcpy(fmt + skip, scfmt->arg, (evsel->tp_format->format.nr_fields - skip) * sizeof(*fmt));
4568                 }
4569         }
4570 }
4571
4572 static int evlist__set_syscall_tp_fields(struct evlist *evlist)
4573 {
4574         struct evsel *evsel;
4575
4576         evlist__for_each_entry(evlist, evsel) {
4577                 if (evsel->priv || !evsel->tp_format)
4578                         continue;
4579
4580                 if (strcmp(evsel->tp_format->system, "syscalls")) {
4581                         evsel__init_tp_arg_scnprintf(evsel);
4582                         continue;
4583                 }
4584
4585                 if (evsel__init_syscall_tp(evsel))
4586                         return -1;
4587
4588                 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
4589                         struct syscall_tp *sc = __evsel__syscall_tp(evsel);
4590
4591                         if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
4592                                 return -1;
4593
4594                         evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_enter_") - 1);
4595                 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
4596                         struct syscall_tp *sc = __evsel__syscall_tp(evsel);
4597
4598                         if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
4599                                 return -1;
4600
4601                         evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_exit_") - 1);
4602                 }
4603         }
4604
4605         return 0;
4606 }
4607
4608 /*
4609  * XXX: Hackish, just splitting the combined -e+--event (syscalls
4610  * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
4611  * existing facilities unchanged (trace->ev_qualifier + parse_options()).
4612  *
4613  * It'd be better to introduce a parse_options() variant that would return a
4614  * list with the terms it didn't match to an event...
4615  */
4616 static int trace__parse_events_option(const struct option *opt, const char *str,
4617                                       int unset __maybe_unused)
4618 {
4619         struct trace *trace = (struct trace *)opt->value;
4620         const char *s = str;
4621         char *sep = NULL, *lists[2] = { NULL, NULL, };
4622         int len = strlen(str) + 1, err = -1, list, idx;
4623         char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
4624         char group_name[PATH_MAX];
4625         struct syscall_fmt *fmt;
4626
4627         if (strace_groups_dir == NULL)
4628                 return -1;
4629
4630         if (*s == '!') {
4631                 ++s;
4632                 trace->not_ev_qualifier = true;
4633         }
4634
4635         while (1) {
4636                 if ((sep = strchr(s, ',')) != NULL)
4637                         *sep = '\0';
4638
4639                 list = 0;
4640                 if (syscalltbl__id(trace->sctbl, s) >= 0 ||
4641                     syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
4642                         list = 1;
4643                         goto do_concat;
4644                 }
4645
4646                 fmt = syscall_fmt__find_by_alias(s);
4647                 if (fmt != NULL) {
4648                         list = 1;
4649                         s = fmt->name;
4650                 } else {
4651                         path__join(group_name, sizeof(group_name), strace_groups_dir, s);
4652                         if (access(group_name, R_OK) == 0)
4653                                 list = 1;
4654                 }
4655 do_concat:
4656                 if (lists[list]) {
4657                         sprintf(lists[list] + strlen(lists[list]), ",%s", s);
4658                 } else {
4659                         lists[list] = malloc(len);
4660                         if (lists[list] == NULL)
4661                                 goto out;
4662                         strcpy(lists[list], s);
4663                 }
4664
4665                 if (!sep)
4666                         break;
4667
4668                 *sep = ',';
4669                 s = sep + 1;
4670         }
4671
4672         if (lists[1] != NULL) {
4673                 struct strlist_config slist_config = {
4674                         .dirname = strace_groups_dir,
4675                 };
4676
4677                 trace->ev_qualifier = strlist__new(lists[1], &slist_config);
4678                 if (trace->ev_qualifier == NULL) {
4679                         fputs("Not enough memory to parse event qualifier", trace->output);
4680                         goto out;
4681                 }
4682
4683                 if (trace__validate_ev_qualifier(trace))
4684                         goto out;
4685                 trace->trace_syscalls = true;
4686         }
4687
4688         err = 0;
4689
4690         if (lists[0]) {
4691                 struct option o = {
4692                         .value = &trace->evlist,
4693                 };
4694                 err = parse_events_option(&o, lists[0], 0);
4695         }
4696 out:
4697         free(strace_groups_dir);
4698         free(lists[0]);
4699         free(lists[1]);
4700         if (sep)
4701                 *sep = ',';
4702
4703         return err;
4704 }
4705
4706 static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
4707 {
4708         struct trace *trace = opt->value;
4709
4710         if (!list_empty(&trace->evlist->core.entries)) {
4711                 struct option o = {
4712                         .value = &trace->evlist,
4713                 };
4714                 return parse_cgroups(&o, str, unset);
4715         }
4716         trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
4717
4718         return 0;
4719 }
4720
4721 static int trace__config(const char *var, const char *value, void *arg)
4722 {
4723         struct trace *trace = arg;
4724         int err = 0;
4725
4726         if (!strcmp(var, "trace.add_events")) {
4727                 trace->perfconfig_events = strdup(value);
4728                 if (trace->perfconfig_events == NULL) {
4729                         pr_err("Not enough memory for %s\n", "trace.add_events");
4730                         return -1;
4731                 }
4732         } else if (!strcmp(var, "trace.show_timestamp")) {
4733                 trace->show_tstamp = perf_config_bool(var, value);
4734         } else if (!strcmp(var, "trace.show_duration")) {
4735                 trace->show_duration = perf_config_bool(var, value);
4736         } else if (!strcmp(var, "trace.show_arg_names")) {
4737                 trace->show_arg_names = perf_config_bool(var, value);
4738                 if (!trace->show_arg_names)
4739                         trace->show_zeros = true;
4740         } else if (!strcmp(var, "trace.show_zeros")) {
4741                 bool new_show_zeros = perf_config_bool(var, value);
4742                 if (!trace->show_arg_names && !new_show_zeros) {
4743                         pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
4744                         goto out;
4745                 }
4746                 trace->show_zeros = new_show_zeros;
4747         } else if (!strcmp(var, "trace.show_prefix")) {
4748                 trace->show_string_prefix = perf_config_bool(var, value);
4749         } else if (!strcmp(var, "trace.no_inherit")) {
4750                 trace->opts.no_inherit = perf_config_bool(var, value);
4751         } else if (!strcmp(var, "trace.args_alignment")) {
4752                 int args_alignment = 0;
4753                 if (perf_config_int(&args_alignment, var, value) == 0)
4754                         trace->args_alignment = args_alignment;
4755         } else if (!strcmp(var, "trace.tracepoint_beautifiers")) {
4756                 if (strcasecmp(value, "libtraceevent") == 0)
4757                         trace->libtraceevent_print = true;
4758                 else if (strcasecmp(value, "libbeauty") == 0)
4759                         trace->libtraceevent_print = false;
4760         }
4761 out:
4762         return err;
4763 }
4764
4765 static void trace__exit(struct trace *trace)
4766 {
4767         int i;
4768
4769         strlist__delete(trace->ev_qualifier);
4770         free(trace->ev_qualifier_ids.entries);
4771         if (trace->syscalls.table) {
4772                 for (i = 0; i <= trace->sctbl->syscalls.max_id; i++)
4773                         syscall__exit(&trace->syscalls.table[i]);
4774                 free(trace->syscalls.table);
4775         }
4776         syscalltbl__delete(trace->sctbl);
4777         zfree(&trace->perfconfig_events);
4778 }
4779
4780 int cmd_trace(int argc, const char **argv)
4781 {
4782         const char *trace_usage[] = {
4783                 "perf trace [<options>] [<command>]",
4784                 "perf trace [<options>] -- <command> [<options>]",
4785                 "perf trace record [<options>] [<command>]",
4786                 "perf trace record [<options>] -- <command> [<options>]",
4787                 NULL
4788         };
4789         struct trace trace = {
4790                 .opts = {
4791                         .target = {
4792                                 .uid       = UINT_MAX,
4793                                 .uses_mmap = true,
4794                         },
4795                         .user_freq     = UINT_MAX,
4796                         .user_interval = ULLONG_MAX,
4797                         .no_buffering  = true,
4798                         .mmap_pages    = UINT_MAX,
4799                 },
4800                 .output = stderr,
4801                 .show_comm = true,
4802                 .show_tstamp = true,
4803                 .show_duration = true,
4804                 .show_arg_names = true,
4805                 .args_alignment = 70,
4806                 .trace_syscalls = false,
4807                 .kernel_syscallchains = false,
4808                 .max_stack = UINT_MAX,
4809                 .max_events = ULONG_MAX,
4810         };
4811         const char *map_dump_str = NULL;
4812         const char *output_name = NULL;
4813         const struct option trace_options[] = {
4814         OPT_CALLBACK('e', "event", &trace, "event",
4815                      "event/syscall selector. use 'perf list' to list available events",
4816                      trace__parse_events_option),
4817         OPT_CALLBACK(0, "filter", &trace.evlist, "filter",
4818                      "event filter", parse_filter),
4819         OPT_BOOLEAN(0, "comm", &trace.show_comm,
4820                     "show the thread COMM next to its id"),
4821         OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
4822         OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
4823                      trace__parse_events_option),
4824         OPT_STRING('o', "output", &output_name, "file", "output file name"),
4825         OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
4826         OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
4827                     "trace events on existing process id"),
4828         OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
4829                     "trace events on existing thread id"),
4830         OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
4831                      "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
4832         OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
4833                     "system-wide collection from all CPUs"),
4834         OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
4835                     "list of cpus to monitor"),
4836         OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
4837                     "child tasks do not inherit counters"),
4838         OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
4839                      "number of mmap data pages", evlist__parse_mmap_pages),
4840         OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
4841                    "user to profile"),
4842         OPT_CALLBACK(0, "duration", &trace, "float",
4843                      "show only events with duration > N.M ms",
4844                      trace__set_duration),
4845 #ifdef HAVE_LIBBPF_SUPPORT
4846         OPT_STRING(0, "map-dump", &map_dump_str, "BPF map", "BPF map to periodically dump"),
4847 #endif
4848         OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
4849         OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4850         OPT_BOOLEAN('T', "time", &trace.full_time,
4851                     "Show full timestamp, not time relative to first start"),
4852         OPT_BOOLEAN(0, "failure", &trace.failure_only,
4853                     "Show only syscalls that failed"),
4854         OPT_BOOLEAN('s', "summary", &trace.summary_only,
4855                     "Show only syscall summary with statistics"),
4856         OPT_BOOLEAN('S', "with-summary", &trace.summary,
4857                     "Show all syscalls and summary with statistics"),
4858         OPT_BOOLEAN(0, "errno-summary", &trace.errno_summary,
4859                     "Show errno stats per syscall, use with -s or -S"),
4860         OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
4861                      "Trace pagefaults", parse_pagefaults, "maj"),
4862         OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
4863         OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
4864         OPT_CALLBACK(0, "call-graph", &trace.opts,
4865                      "record_mode[,record_size]", record_callchain_help,
4866                      &record_parse_callchain_opt),
4867         OPT_BOOLEAN(0, "libtraceevent_print", &trace.libtraceevent_print,
4868                     "Use libtraceevent to print the tracepoint arguments."),
4869         OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
4870                     "Show the kernel callchains on the syscall exit path"),
4871         OPT_ULONG(0, "max-events", &trace.max_events,
4872                 "Set the maximum number of events to print, exit after that is reached. "),
4873         OPT_UINTEGER(0, "min-stack", &trace.min_stack,
4874                      "Set the minimum stack depth when parsing the callchain, "
4875                      "anything below the specified depth will be ignored."),
4876         OPT_UINTEGER(0, "max-stack", &trace.max_stack,
4877                      "Set the maximum stack depth when parsing the callchain, "
4878                      "anything beyond the specified depth will be ignored. "
4879                      "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
4880         OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
4881                         "Sort batch of events before processing, use if getting out of order events"),
4882         OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
4883                         "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
4884         OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
4885                         "per thread proc mmap processing timeout in ms"),
4886         OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
4887                      trace__parse_cgroups),
4888         OPT_INTEGER('D', "delay", &trace.opts.initial_delay,
4889                      "ms to wait before starting measurement after program "
4890                      "start"),
4891         OPTS_EVSWITCH(&trace.evswitch),
4892         OPT_END()
4893         };
4894         bool __maybe_unused max_stack_user_set = true;
4895         bool mmap_pages_user_set = true;
4896         struct evsel *evsel;
4897         const char * const trace_subcommands[] = { "record", NULL };
4898         int err = -1;
4899         char bf[BUFSIZ];
4900         struct sigaction sigchld_act;
4901
4902         signal(SIGSEGV, sighandler_dump_stack);
4903         signal(SIGFPE, sighandler_dump_stack);
4904         signal(SIGINT, sighandler_interrupt);
4905
4906         memset(&sigchld_act, 0, sizeof(sigchld_act));
4907         sigchld_act.sa_flags = SA_SIGINFO;
4908         sigchld_act.sa_sigaction = sighandler_chld;
4909         sigaction(SIGCHLD, &sigchld_act, NULL);
4910
4911         trace.evlist = evlist__new();
4912         trace.sctbl = syscalltbl__new();
4913
4914         if (trace.evlist == NULL || trace.sctbl == NULL) {
4915                 pr_err("Not enough memory to run!\n");
4916                 err = -ENOMEM;
4917                 goto out;
4918         }
4919
4920         /*
4921          * Parsing .perfconfig may entail creating a BPF event, that may need
4922          * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting
4923          * is too small. This affects just this process, not touching the
4924          * global setting. If it fails we'll get something in 'perf trace -v'
4925          * to help diagnose the problem.
4926          */
4927         rlimit__bump_memlock();
4928
4929         err = perf_config(trace__config, &trace);
4930         if (err)
4931                 goto out;
4932
4933         argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
4934                                  trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
4935
4936         /*
4937          * Here we already passed thru trace__parse_events_option() and it has
4938          * already figured out if -e syscall_name, if not but if --event
4939          * foo:bar was used, the user is interested _just_ in those, say,
4940          * tracepoint events, not in the strace-like syscall-name-based mode.
4941          *
4942          * This is important because we need to check if strace-like mode is
4943          * needed to decided if we should filter out the eBPF
4944          * __augmented_syscalls__ code, if it is in the mix, say, via
4945          * .perfconfig trace.add_events, and filter those out.
4946          */
4947         if (!trace.trace_syscalls && !trace.trace_pgfaults &&
4948             trace.evlist->core.nr_entries == 0 /* Was --events used? */) {
4949                 trace.trace_syscalls = true;
4950         }
4951         /*
4952          * Now that we have --verbose figured out, lets see if we need to parse
4953          * events from .perfconfig, so that if those events fail parsing, say some
4954          * BPF program fails, then we'll be able to use --verbose to see what went
4955          * wrong in more detail.
4956          */
4957         if (trace.perfconfig_events != NULL) {
4958                 struct parse_events_error parse_err;
4959
4960                 parse_events_error__init(&parse_err);
4961                 err = parse_events(trace.evlist, trace.perfconfig_events, &parse_err);
4962                 if (err)
4963                         parse_events_error__print(&parse_err, trace.perfconfig_events);
4964                 parse_events_error__exit(&parse_err);
4965                 if (err)
4966                         goto out;
4967         }
4968
4969         if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
4970                 usage_with_options_msg(trace_usage, trace_options,
4971                                        "cgroup monitoring only available in system-wide mode");
4972         }
4973
4974         evsel = bpf__setup_output_event(trace.evlist, "__augmented_syscalls__");
4975         if (IS_ERR(evsel)) {
4976                 bpf__strerror_setup_output_event(trace.evlist, PTR_ERR(evsel), bf, sizeof(bf));
4977                 pr_err("ERROR: Setup trace syscalls enter failed: %s\n", bf);
4978                 goto out;
4979         }
4980
4981         if (evsel) {
4982                 trace.syscalls.events.augmented = evsel;
4983
4984                 evsel = evlist__find_tracepoint_by_name(trace.evlist, "raw_syscalls:sys_enter");
4985                 if (evsel == NULL) {
4986                         pr_err("ERROR: raw_syscalls:sys_enter not found in the augmented BPF object\n");
4987                         goto out;
4988                 }
4989
4990                 if (evsel->bpf_obj == NULL) {
4991                         pr_err("ERROR: raw_syscalls:sys_enter not associated to a BPF object\n");
4992                         goto out;
4993                 }
4994
4995                 trace.bpf_obj = evsel->bpf_obj;
4996
4997                 /*
4998                  * If we have _just_ the augmenter event but don't have a
4999                  * explicit --syscalls, then assume we want all strace-like
5000                  * syscalls:
5001                  */
5002                 if (!trace.trace_syscalls && trace__only_augmented_syscalls_evsels(&trace))
5003                         trace.trace_syscalls = true;
5004                 /*
5005                  * So, if we have a syscall augmenter, but trace_syscalls, aka
5006                  * strace-like syscall tracing is not set, then we need to trow
5007                  * away the augmenter, i.e. all the events that were created
5008                  * from that BPF object file.
5009                  *
5010                  * This is more to fix the current .perfconfig trace.add_events
5011                  * style of setting up the strace-like eBPF based syscall point
5012                  * payload augmenter.
5013                  *
5014                  * All this complexity will be avoided by adding an alternative
5015                  * to trace.add_events in the form of
5016                  * trace.bpf_augmented_syscalls, that will be only parsed if we
5017                  * need it.
5018                  *
5019                  * .perfconfig trace.add_events is still useful if we want, for
5020                  * instance, have msr_write.msr in some .perfconfig profile based
5021                  * 'perf trace --config determinism.profile' mode, where for some
5022                  * particular goal/workload type we want a set of events and
5023                  * output mode (with timings, etc) instead of having to add
5024                  * all via the command line.
5025                  *
5026                  * Also --config to specify an alternate .perfconfig file needs
5027                  * to be implemented.
5028                  */
5029                 if (!trace.trace_syscalls) {
5030                         trace__delete_augmented_syscalls(&trace);
5031                 } else {
5032                         trace__set_bpf_map_filtered_pids(&trace);
5033                         trace__set_bpf_map_syscalls(&trace);
5034                         trace.syscalls.unaugmented_prog = trace__find_bpf_program_by_title(&trace, "!raw_syscalls:unaugmented");
5035                 }
5036         }
5037
5038         err = bpf__setup_stdout(trace.evlist);
5039         if (err) {
5040                 bpf__strerror_setup_stdout(trace.evlist, err, bf, sizeof(bf));
5041                 pr_err("ERROR: Setup BPF stdout failed: %s\n", bf);
5042                 goto out;
5043         }
5044
5045         err = -1;
5046
5047         if (map_dump_str) {
5048                 trace.dump.map = trace__find_bpf_map_by_name(&trace, map_dump_str);
5049                 if (trace.dump.map == NULL) {
5050                         pr_err("ERROR: BPF map \"%s\" not found\n", map_dump_str);
5051                         goto out;
5052                 }
5053         }
5054
5055         if (trace.trace_pgfaults) {
5056                 trace.opts.sample_address = true;
5057                 trace.opts.sample_time = true;
5058         }
5059
5060         if (trace.opts.mmap_pages == UINT_MAX)
5061                 mmap_pages_user_set = false;
5062
5063         if (trace.max_stack == UINT_MAX) {
5064                 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
5065                 max_stack_user_set = false;
5066         }
5067
5068 #ifdef HAVE_DWARF_UNWIND_SUPPORT
5069         if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
5070                 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
5071         }
5072 #endif
5073
5074         if (callchain_param.enabled) {
5075                 if (!mmap_pages_user_set && geteuid() == 0)
5076                         trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
5077
5078                 symbol_conf.use_callchain = true;
5079         }
5080
5081         if (trace.evlist->core.nr_entries > 0) {
5082                 evlist__set_default_evsel_handler(trace.evlist, trace__event_handler);
5083                 if (evlist__set_syscall_tp_fields(trace.evlist)) {
5084                         perror("failed to set syscalls:* tracepoint fields");
5085                         goto out;
5086                 }
5087         }
5088
5089         if (trace.sort_events) {
5090                 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
5091                 ordered_events__set_copy_on_queue(&trace.oe.data, true);
5092         }
5093
5094         /*
5095          * If we are augmenting syscalls, then combine what we put in the
5096          * __augmented_syscalls__ BPF map with what is in the
5097          * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
5098          * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
5099          *
5100          * We'll switch to look at two BPF maps, one for sys_enter and the
5101          * other for sys_exit when we start augmenting the sys_exit paths with
5102          * buffers that are being copied from kernel to userspace, think 'read'
5103          * syscall.
5104          */
5105         if (trace.syscalls.events.augmented) {
5106                 evlist__for_each_entry(trace.evlist, evsel) {
5107                         bool raw_syscalls_sys_exit = strcmp(evsel__name(evsel), "raw_syscalls:sys_exit") == 0;
5108
5109                         if (raw_syscalls_sys_exit) {
5110                                 trace.raw_augmented_syscalls = true;
5111                                 goto init_augmented_syscall_tp;
5112                         }
5113
5114                         if (trace.syscalls.events.augmented->priv == NULL &&
5115                             strstr(evsel__name(evsel), "syscalls:sys_enter")) {
5116                                 struct evsel *augmented = trace.syscalls.events.augmented;
5117                                 if (evsel__init_augmented_syscall_tp(augmented, evsel) ||
5118                                     evsel__init_augmented_syscall_tp_args(augmented))
5119                                         goto out;
5120                                 /*
5121                                  * Augmented is __augmented_syscalls__ BPF_OUTPUT event
5122                                  * Above we made sure we can get from the payload the tp fields
5123                                  * that we get from syscalls:sys_enter tracefs format file.
5124                                  */
5125                                 augmented->handler = trace__sys_enter;
5126                                 /*
5127                                  * Now we do the same for the *syscalls:sys_enter event so that
5128                                  * if we handle it directly, i.e. if the BPF prog returns 0 so
5129                                  * as not to filter it, then we'll handle it just like we would
5130                                  * for the BPF_OUTPUT one:
5131                                  */
5132                                 if (evsel__init_augmented_syscall_tp(evsel, evsel) ||
5133                                     evsel__init_augmented_syscall_tp_args(evsel))
5134                                         goto out;
5135                                 evsel->handler = trace__sys_enter;
5136                         }
5137
5138                         if (strstarts(evsel__name(evsel), "syscalls:sys_exit_")) {
5139                                 struct syscall_tp *sc;
5140 init_augmented_syscall_tp:
5141                                 if (evsel__init_augmented_syscall_tp(evsel, evsel))
5142                                         goto out;
5143                                 sc = __evsel__syscall_tp(evsel);
5144                                 /*
5145                                  * For now with BPF raw_augmented we hook into
5146                                  * raw_syscalls:sys_enter and there we get all
5147                                  * 6 syscall args plus the tracepoint common
5148                                  * fields and the syscall_nr (another long).
5149                                  * So we check if that is the case and if so
5150                                  * don't look after the sc->args_size but
5151                                  * always after the full raw_syscalls:sys_enter
5152                                  * payload, which is fixed.
5153                                  *
5154                                  * We'll revisit this later to pass
5155                                  * s->args_size to the BPF augmenter (now
5156                                  * tools/perf/examples/bpf/augmented_raw_syscalls.c,
5157                                  * so that it copies only what we need for each
5158                                  * syscall, like what happens when we use
5159                                  * syscalls:sys_enter_NAME, so that we reduce
5160                                  * the kernel/userspace traffic to just what is
5161                                  * needed for each syscall.
5162                                  */
5163                                 if (trace.raw_augmented_syscalls)
5164                                         trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
5165                                 evsel__init_augmented_syscall_tp_ret(evsel);
5166                                 evsel->handler = trace__sys_exit;
5167                         }
5168                 }
5169         }
5170
5171         if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
5172                 return trace__record(&trace, argc-1, &argv[1]);
5173
5174         /* Using just --errno-summary will trigger --summary */
5175         if (trace.errno_summary && !trace.summary && !trace.summary_only)
5176                 trace.summary_only = true;
5177
5178         /* summary_only implies summary option, but don't overwrite summary if set */
5179         if (trace.summary_only)
5180                 trace.summary = trace.summary_only;
5181
5182         if (output_name != NULL) {
5183                 err = trace__open_output(&trace, output_name);
5184                 if (err < 0) {
5185                         perror("failed to create output file");
5186                         goto out;
5187                 }
5188         }
5189
5190         err = evswitch__init(&trace.evswitch, trace.evlist, stderr);
5191         if (err)
5192                 goto out_close;
5193
5194         err = target__validate(&trace.opts.target);
5195         if (err) {
5196                 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
5197                 fprintf(trace.output, "%s", bf);
5198                 goto out_close;
5199         }
5200
5201         err = target__parse_uid(&trace.opts.target);
5202         if (err) {
5203                 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
5204                 fprintf(trace.output, "%s", bf);
5205                 goto out_close;
5206         }
5207
5208         if (!argc && target__none(&trace.opts.target))
5209                 trace.opts.target.system_wide = true;
5210
5211         if (input_name)
5212                 err = trace__replay(&trace);
5213         else
5214                 err = trace__run(&trace, argc, argv);
5215
5216 out_close:
5217         if (output_name != NULL)
5218                 fclose(trace.output);
5219 out:
5220         trace__exit(&trace);
5221         return err;
5222 }