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